Viewing: sanity.sh
#!/usr/bin/bash
#
# Run select tests by setting ONLY, or as arguments to the script.
# Skip specific tests by setting EXCEPT.
#
# e.g. ONLY="22 23" or ONLY="`seq 32 39`" or EXCEPT="31"
set -e
ONLY=${ONLY:-"$*"}
# Check Grants after these tests
GRANT_CHECK_LIST="$GRANT_CHECK_LIST 42a 42b 42c 42d 42e 63a 63b 64a 64b 64c 64d 64j"
OSC=${OSC:-"osc"}
CC=${CC:-cc}
CREATETEST=${CREATETEST:-createtest}
LVERIFY=${LVERIFY:-ll_dirstripe_verify}
OPENFILE=${OPENFILE:-openfile}
OPENUNLINK=${OPENUNLINK:-openunlink}
READS=${READS:-"reads"}
SOCKETSERVER=${SOCKETSERVER:-socketserver}
SOCKETCLIENT=${SOCKETCLIENT:-socketclient}
MEMHOG=${MEMHOG:-memhog}
DIRECTIO=${DIRECTIO:-directio}
ACCEPTOR_PORT=${ACCEPTOR_PORT:-988}
DEF_STRIPE_COUNT=${DEF_STRIPE_COUNT:--1}
CHECK_GRANT=${CHECK_GRANT:-"yes"}
GRANT_CHECK_LIST=${GRANT_CHECK_LIST:-""}
TRACE=${TRACE:-""}
LUSTRE=${LUSTRE:-$(dirname $0)/..}
LUSTRE_TESTS_API_DIR=${LUSTRE_TESTS_API_DIR:-${LUSTRE}/tests/clientapi}
. $LUSTRE/tests/test-framework.sh
init_test_env "$@"
init_logging
ALWAYS_EXCEPT="$SANITY_EXCEPT "
always_except LU-9693 42a 42c
always_except LU-6493 42b
always_except LU-16515 118c 118d
always_except LU-8411 407
always_except LU-18032 119i
always_except LU-12550 216
if $SHARED_KEY; then
always_except LU-14181 64e 64f
always_except LU-17127 39o
fi
# skip nfs tests on kernels >= 4.12.0 until they are fixed
if [ $LINUX_VERSION_CODE -ge $(version_code 4.12.0) ]; then
always_except LU-12661 817
fi
# skip cgroup tests on RHEL8.1 kernels until they are fixed
if (( $LINUX_VERSION_CODE >= $(version_code 4.18.0) &&
$LINUX_VERSION_CODE < $(version_code 5.4.0) )); then
always_except LU-13063 411a
fi
# skip cgroup tests for kernels < v4.18.0
if (( $LINUX_VERSION_CODE < $(version_code 4.18.0) )); then
always_except LU-13063 411b
fi
# minutes runtime: 5 12 8 12 15 10
[[ "$SLOW" = "no" ]] && EXCEPT_SLOW="27m 60i 64b 68 71 135 136 230d 300o 842"
if [[ "$mds1_FSTYPE" == "zfs" ]]; then
# 13 (min)"
[[ "$SLOW" == "no" ]] && EXCEPT_SLOW="$EXCEPT_SLOW 51b"
fi
# Janitor is very slow for some different subtests than Autotest
if [[ "$HOSTNAME" =~ "oleg" ]]; then
# minutes runtime: 60 32 21
EXCEPT_SLOW+=" 51c 51e 834"
fi
if [[ "$ost1_FSTYPE" == "zfs" ]]; then
always_except LU-1941 130b 130c 130d 130e 130f 130g
fi
# Check if running on specific distros to skip certain subtests
if [[ "$CLIENT_OS_ID_LIKE" =~ "rhel" ]]; then
if (( $CLIENT_OS_VERSION_CODE >= $(version_code 9.3.0) )); then
# disable test_906 temporarily until rhel9.3 solves the
# failure on fio io_uring I/O engine.
always_except LU-17289 906
fi
elif [[ "$CLIENT_OS_ID_LIKE" =~ "ubuntu" ]] &&
(( $CLIENT_OS_VERSION_CODE == $(version_code 22.04) )); then
always_except LU-19124 27P
fi
if [[ "$FSTYPE" = "wbcfs" ]]; then
# Lack of lprocfs support
always_except LU-18813 0f 27A 53 66 270a
# lack of lprocfs: osd.*.nonrotational
always_except LU-18813 119e 119f 119g 119h
# No stats (similar to openZFS)
always_except LU-18813 156
# MemFS-based OSD (wbcfs) cannot recovery from a server restart
always_except LU-18813 17o 27oo 27z 27F 60a 64i 232 257
always_except LU-18813 278 280 427 801c 818 820
# Symlink/CHR/SOCK/FIFO/BLK file types do not support
always_except LU-18813 17a 17b 17e 17g 17i 17p 21 25a
always_except LU-18813 25b 26a 26c 26d 26e 26f 27ga 27Q
always_except LU-18813 28 32e 32f 32g 32h 32m 32n 32o
always_except LU-18813 32p 48a 54a 54c 54d 56l 56m 56n 56rd
always_except LU-18813 56xb 56Eb 56eg 56eh 56ei 133a 140 170b
always_except LU-18813 162a 226a
# Truncate operation is not supported yet.
always_except LU-18813 27p 27q 34a
# cross directory hardlink in DNE env
always_except LU-18813 31g 31l 31m
# FMD not expired: cannot reproduce on local testing
always_except LU-18813 36g
# Filemap is not supported yet.
always_except LU-18813 44f 130a 130b 130c 130d 130e 130i 430a
# inodes/blocks space usage accounting and statfs() is not supported
always_except LU-18813 51b 56ab 81b 220 413 418 806
# lsattr: append-only/immutable flags
always_except LU-18813 52a 52b
# xattr_list() is not implemented yet
always_except LU-18813 102a 102h 102i 102r 102t
# linkea and fid2path wrong...
always_except LU-18813 154B 154f 154g
# changelog related failures: wbcfs-target device label is not correct
always_except LU-18813 160 161c 161d 205a 65k 807 808 812
# DNE does not work well
always_except LU-18813 56 65e 65a 406
# user.job XATTR
always_except LU-18813 205h
# Exclusive open timeout
always_except LU-18813 208
# OFD access log failure
always_except LU-18813 165
# rename() operations: the source may not empty
# always_except LU-18813 214
# Data page cache has been updated during bulk write
always_except LU-18813 224d
# fid2path failure
always_except LU-18813 226d
# ladvise failure
always_except LU-18813 255
# sec related failure
always_except LU-18813 258
# DoM migration failure
always_except LU-18813 272
# Unkown reason timeout!
always_except LU-18813 275 277 311 410 414 419 831
# last_rcvd should fail
always_except LU-18813 313 314 315
# block accting is wrong...
always_except LU-18813 317
# Other timeouts
always_except LU-18813 200 350 398 399 403 404 408 432 433
# DIO locking issue?
always_except LU-18813 398a
# Layout swap is not working
always_except LU-18813 405
# Memory pressure under memcg control
always_except LU-18813 411
# rmfid in DNE and in large numbers
always_except LU-18813 421
# local testing passed but Maloo testing failed!
always_except LU-18813 27Cg 27U 422 424 425 426 428 429 434 442
# OOM failure
always_except LU-18813 430b 430c 431 814 833 850
# Expired barrier
always_except LU-18813 801a 801b
# ro is not implemented yet
always_except LU-18813 802b
# openZFS related partial page write
always_except LU-18813 810
# Quota is not supported yet...
always_except LU-18813 812b
# ldlm kunit test
always_except LU-18813 842
# fanotify does not work
always_except LU-18813 851
# MGC locks and client umount
always_except LU-18813 901
# destroy takes too much time
always_except LU-18813 903
fi
# Although every sanity.sh test has been run, we stop sooner for
# stability reasons. As we get farther, increment the STOP_AT value.
if [[ "$FSTYPE" = "wbcfs" ]]; then
export STOP_AT=${STOP_AT:-"440"}
fi
build_test_filter
FAIL_ON_ERROR=false
proc_regexp="/{proc,sys}/{fs,sys,kernel/debug}/{lustre,lnet}/"
cleanup() {
echo -n "cln.."
wait_update_facet client "pgrep ll_sa" "" 35 ||
error "ll_sa thread is still running"
pgrep ll_sa > /dev/null && { echo "There are ll_sa thread not exit!"; exit 20; }
cleanupall ${FORCE} $* || { echo "FAILed to clean up"; exit 20; }
}
setup() {
echo -n "mnt.."
load_modules
setupall || exit 10
echo "done"
}
check_swap_layouts_support()
{
$LCTL get_param -n llite.*.sbi_flags | grep -q layout ||
skip "Does not support layout lock."
}
check_swap_layout_no_dom()
{
local FOLDER=$1
local SUPP=$(lfs getstripe $FOLDER | grep "pattern: mdt" | wc -l)
[ $SUPP -eq 0 ] || skip "layout swap does not support DOM files so far"
}
check_and_setup_lustre
DIR=${DIR:-$MOUNT}
assert_DIR
MAXFREE=${MAXFREE:-$((300000 * $OSTCOUNT))}
[ -f $DIR/d52a/foo ] && chattr -a $DIR/d52a/foo
[ -f $DIR/d52b/foo ] && chattr -i $DIR/d52b/foo
rm -rf $DIR/[Rdfs][0-9]*
# $RUNAS_ID may get set incorrectly somewhere else
[ $UID -eq 0 -a $RUNAS_ID -eq 0 ] &&
error "\$RUNAS_ID set to 0, but \$UID is also 0!"
check_runas_id $RUNAS_ID $RUNAS_GID $RUNAS
if [ "${ONLY}" = "MOUNT" ] ; then
echo "Lustre is up, please go on"
exit
fi
echo "preparing for tests involving mounts"
EXT2_DEV=${EXT2_DEV:-$TMP/SANITY.LOOP}
touch $EXT2_DEV
mke2fs -j -F $EXT2_DEV 8000 > /dev/null
echo # add a newline after mke2fs.
umask 077
OLDDEBUG=$(lctl get_param -n debug 2> /dev/null)
# ensure all internal functions know we want full debug
export PTLDEBUG=all
lctl set_param debug=$PTLDEBUG 2> /dev/null || true
test_0a() {
touch $DIR/$tfile
$CHECKSTAT -t file $DIR/$tfile || error "$tfile is not a file"
rm $DIR/$tfile
$CHECKSTAT -a $DIR/$tfile || error "$tfile was not removed"
}
run_test 0a "touch; rm ====================="
test_0b() {
chmod 0755 $DIR || error "chmod 0755 $DIR failed"
$CHECKSTAT -p 0755 $DIR || error "$DIR permission is not 0755"
}
run_test 0b "chmod 0755 $DIR ============================="
test_0c() {
$LCTL get_param mdc.*.import | grep "state: FULL" ||
error "import not FULL"
$LCTL get_param mdc.*.import | grep "target: $FSNAME-MDT" ||
error "bad target"
}
run_test 0c "check import proc"
test_0d() { # LU-3397
(( $MGS_VERSION >= $(version_code v2_10_57_0-64-gb4b773466a) )) ||
skip "need MGS >= 2.10.57 for proc exports supported"
local mgs_exp="mgs.MGS.exports"
local client_uuid=$($LCTL get_param -n mgc.*.uuid)
local exp_uuid
local exp_client_version
local exp_val
local imp_val
local temp_imp=$DIR/$tfile.import
local temp_exp=$DIR/$tfile.export
# save mgc import file to $temp_imp
$LCTL get_param mgc.*.import | tee $temp_imp
# Check if client uuid is found in MGS export
for exp_uuid in $(do_facet mgs $LCTL get_param -N $mgs_exp.*.uuid); do
echo $exp_uuid
do_facet mgs $LCTL get_param -n $exp_uuid
[[ $(do_facet mgs $LCTL get_param -n $exp_uuid) == \
$client_uuid ]] && break
done
# save mgs export file to $temp_exp
do_facet mgs $LCTL get_param ${exp_uuid%.uuid}.export | tee $temp_exp
# Compare the value of field "connect_flags"
imp_val=$(grep "connect_flags" $temp_imp)
exp_val=$(grep "connect_flags" $temp_exp)
[ "$exp_val" == "$imp_val" ] ||
error "export flags '$exp_val' != import flags '$imp_val'"
# Compare client versions.
exp_client_version=$(awk '/target_version:/ { print $2 }' $temp_exp)
exp_val=$(version_code $exp_client_version)
imp_client_version=$(lustre_build_version client)
imp_val=$(version_code $imp_client_version)
(( "$exp_val" == "$imp_val" )) ||
error "exp version '$exp_client_version'($exp_val) != '$(imp_client_version)'($imp_val)"
}
run_test 0d "check export proc"
test_0e() { # LU-13417
(( $MDSCOUNT > 1 )) ||
skip "We need at least 2 MDTs for this test"
(( $MDS1_VERSION >= $(version_code 2.14.51) )) ||
skip "Need server version at least 2.14.51"
local default_lmv_count=$($LFS getdirstripe -D -c $MOUNT)
local default_lmv_index=$($LFS getdirstripe -D -i $MOUNT)
[ $default_lmv_count -eq 1 ] ||
error "$MOUNT default stripe count $default_lmv_count"
[ $default_lmv_index -eq -1 ] ||
error "$MOUNT default stripe index $default_lmv_index"
mkdir $MOUNT/$tdir.1 || error "mkdir $MOUNT/$tdir.1 failed"
mkdir $MOUNT/$tdir.2 || error "mkdir $MOUNT/$tdir.2 failed"
local mdt_index1=$($LFS getdirstripe -i $MOUNT/$tdir.1)
local mdt_index2=$($LFS getdirstripe -i $MOUNT/$tdir.2)
[ $mdt_index1 -eq $mdt_index2 ] &&
error "directories are on the same MDT $mdt_index1=$mdt_index2"
rmdir $MOUNT/$tdir.1 $MOUNT/$tdir.2
}
run_test 0e "Enable DNE MDT balancing for mkdir in the ROOT"
test_1() {
test_mkdir $DIR/$tdir
test_mkdir $DIR/$tdir/d2
mkdir $DIR/$tdir/d2 && error "we expect EEXIST, but not returned"
$CHECKSTAT -t dir $DIR/$tdir/d2 || error "$tdir/d2 is not a dir"
rmdir $DIR/$tdir/d2
rmdir $DIR/$tdir
$CHECKSTAT -a $DIR/$tdir || error "$tdir was not removed"
}
run_test 1 "mkdir; remkdir; rmdir"
test_2() {
test_mkdir $DIR/$tdir
touch $DIR/$tdir/$tfile || error "touch $tdir/$tfile failed"
$CHECKSTAT -t file $DIR/$tdir/$tfile || error "$tdir/$tfile not a file"
rm -r $DIR/$tdir
$CHECKSTAT -a $DIR/$tdir/$tfile || error "$tdir/$file is not removed"
}
run_test 2 "mkdir; touch; rmdir; check file"
test_3() {
test_mkdir $DIR/$tdir
$CHECKSTAT -t dir $DIR/$tdir || error "$tdir is not a directory"
touch $DIR/$tdir/$tfile
$CHECKSTAT -t file $DIR/$tdir/$tfile || error "$tdir/$tfile not a file"
rm -r $DIR/$tdir
$CHECKSTAT -a $DIR/$tdir || error "$tdir is not removed"
}
run_test 3 "mkdir; touch; rmdir; check dir"
# LU-4471 - failed rmdir on remote directories still removes directory on MDT0
test_4() {
test_mkdir -i 1 $DIR/$tdir
touch $DIR/$tdir/$tfile ||
error "Create file under remote directory failed"
rmdir $DIR/$tdir &&
error "Expect error removing in-use dir $DIR/$tdir"
test -d $DIR/$tdir || error "Remote directory disappeared"
rm -rf $DIR/$tdir || error "remove remote dir error"
}
run_test 4 "mkdir; touch dir/file; rmdir; checkdir (expect error)"
test_5() {
test_mkdir $DIR/$tdir
test_mkdir $DIR/$tdir/d2
chmod 0707 $DIR/$tdir/d2 || error "chmod 0707 $tdir/d2 failed"
$CHECKSTAT -t dir -p 0707 $DIR/$tdir/d2 || error "$tdir/d2 not mode 707"
$CHECKSTAT -t dir $DIR/$tdir/d2 || error "$tdir/d2 is not a directory"
}
run_test 5 "mkdir .../d5 .../d5/d2; chmod .../d5/d2"
test_6a() {
touch $DIR/$tfile || error "touch $DIR/$tfile failed"
chmod 0666 $DIR/$tfile || error "chmod 0666 $tfile failed"
$CHECKSTAT -t file -p 0666 -u \#$UID $DIR/$tfile ||
error "$tfile does not have perm 0666 or UID $UID"
$RUNAS chmod 0444 $DIR/$tfile && error "chmod $tfile worked on UID $UID"
$CHECKSTAT -t file -p 0666 -u \#$UID $DIR/$tfile ||
error "$tfile should be 0666 and owned by UID $UID"
}
run_test 6a "touch f6a; chmod f6a; $RUNAS chmod f6a (should return error) =="
test_6c() {
[ $RUNAS_ID -eq $UID ] && skip_env "RUNAS_ID = UID = $UID"
touch $DIR/$tfile
chown $RUNAS_ID $DIR/$tfile || error "chown $RUNAS_ID $file failed"
$CHECKSTAT -t file -u \#$RUNAS_ID $DIR/$tfile ||
error "$tfile should be owned by UID $RUNAS_ID"
$RUNAS chown $UID $DIR/$tfile && error "chown $UID $file succeeded"
$CHECKSTAT -t file -u \#$RUNAS_ID $DIR/$tfile ||
error "$tfile should be owned by UID $RUNAS_ID"
}
run_test 6c "touch f6c; chown f6c; $RUNAS chown f6c (should return error) =="
test_6e() {
[ $RUNAS_ID -eq $UID ] && skip_env "RUNAS_ID = UID = $UID"
touch $DIR/$tfile
chgrp $RUNAS_ID $DIR/$tfile || error "chgrp $RUNAS_ID $file failed"
$CHECKSTAT -t file -u \#$UID -g \#$RUNAS_ID $DIR/$tfile ||
error "$tfile should be owned by GID $UID"
$RUNAS chgrp $UID $DIR/$tfile && error "chgrp $UID $file succeeded"
$CHECKSTAT -t file -u \#$UID -g \#$RUNAS_ID $DIR/$tfile ||
error "$tfile should be owned by UID $UID and GID $RUNAS_ID"
}
run_test 6e "touch+chgrp $tfile; $RUNAS chgrp $tfile (should return error)"
test_6g() {
[ $RUNAS_ID -eq $UID ] && skip_env "RUNAS_ID = UID = $UID"
test_mkdir $DIR/$tdir
chmod 777 $DIR/$tdir || error "chmod 0777 $tdir failed"
$RUNAS mkdir $DIR/$tdir/d || error "mkdir $tdir/d failed"
chmod g+s $DIR/$tdir/d || error "chmod g+s $tdir/d failed"
test_mkdir $DIR/$tdir/d/subdir
$CHECKSTAT -g \#$RUNAS_GID $DIR/$tdir/d/subdir ||
error "$tdir/d/subdir should be GID $RUNAS_GID"
if [[ $MDSCOUNT -gt 1 ]]; then
# check remote dir sgid inherite
$LFS mkdir -i 0 $DIR/$tdir.local ||
error "mkdir $tdir.local failed"
chmod g+s $DIR/$tdir.local ||
error "chmod $tdir.local failed"
chgrp $RUNAS_GID $DIR/$tdir.local ||
error "chgrp $tdir.local failed"
$LFS mkdir -i 1 $DIR/$tdir.local/$tdir.remote ||
error "mkdir $tdir.remote failed"
$CHECKSTAT -g \#$RUNAS_GID $DIR/$tdir.local/$tdir.remote ||
error "$tdir.remote should be owned by $UID.$RUNAS_ID"
$CHECKSTAT -p 02755 $DIR/$tdir.local/$tdir.remote ||
error "$tdir.remote should be mode 02755"
fi
}
run_test 6g "verify new dir in sgid dir inherits group"
test_6h() { # bug 7331
[ $RUNAS_ID -eq $UID ] && skip_env "RUNAS_ID = UID = $UID"
touch $DIR/$tfile || error "touch failed"
chown $RUNAS_ID:$RUNAS_GID $DIR/$tfile || error "initial chown failed"
$RUNAS -G$RUNAS_GID chown $RUNAS_ID:0 $DIR/$tfile &&
error "chown $RUNAS_ID:0 $tfile worked as GID $RUNAS_GID"
$CHECKSTAT -t file -u \#$RUNAS_ID -g \#$RUNAS_GID $DIR/$tfile ||
error "$tdir/$tfile should be UID $RUNAS_UID GID $RUNAS_GID"
}
run_test 6h "$RUNAS chown RUNAS_ID.0 .../$tfile (should return error)"
test_6i() {
(( MDS1_VERSION >= $(version_code 2.16.51) )) ||
skip "Need MDS version at least 2.16.51"
(( $RUNAS_ID != $UID )) || skip_env "RUNAS_ID = UID = $UID"
touch $DIR/$tfile
chmod 444 $DIR/$tfile
chown $RUNAS_ID $DIR/$tfile || error "initial chown failed"
$RUNAS -u $RUNAS_ID -g $RUNAS_ID chgrp $RUNAS_ID $DIR/$tfile ||
error "chgrp $RUNAS_ID $file failed"
$CHECKSTAT -t file -u \#$RUNAS_ID -g \#$RUNAS_ID $DIR/$tfile ||
error "$tfile should be owned by GID $RUNAS_ID"
}
run_test 6i "touch+chmod+chgrp $tfile; chgrp read-only file should succeed"
test_7a() {
test_mkdir $DIR/$tdir
$MCREATE $DIR/$tdir/$tfile
chmod 0666 $DIR/$tdir/$tfile
$CHECKSTAT -t file -p 0666 $DIR/$tdir/$tfile ||
error "$tdir/$tfile should be mode 0666"
}
run_test 7a "mkdir .../d7; mcreate .../d7/f; chmod .../d7/f ===="
test_7b() {
if [ ! -d $DIR/$tdir ]; then
test_mkdir $DIR/$tdir
fi
$MCREATE $DIR/$tdir/$tfile
echo -n foo > $DIR/$tdir/$tfile
[ "$(cat $DIR/$tdir/$tfile)" = "foo" ] || error "$tdir/$tfile not 'foo'"
$CHECKSTAT -t file -s 3 $DIR/$tdir/$tfile || error "$tfile size not 3"
}
run_test 7b "mkdir .../d7; mcreate d7/f2; echo foo > d7/f2 ====="
test_8() {
test_mkdir $DIR/$tdir
touch $DIR/$tdir/$tfile
chmod 0666 $DIR/$tdir/$tfile
$CHECKSTAT -t file -p 0666 $DIR/$tdir/$tfile ||
error "$tfile mode not 0666"
}
run_test 8 "mkdir .../d8; touch .../d8/f; chmod .../d8/f ======="
test_9() {
test_mkdir $DIR/$tdir
test_mkdir $DIR/$tdir/d2
test_mkdir $DIR/$tdir/d2/d3
$CHECKSTAT -t dir $DIR/$tdir/d2/d3 || error "$tdir/d2/d3 not a dir"
}
run_test 9 "mkdir .../d9 .../d9/d2 .../d9/d2/d3 ================"
test_10() {
test_mkdir $DIR/$tdir
test_mkdir $DIR/$tdir/d2
touch $DIR/$tdir/d2/$tfile
$CHECKSTAT -t file $DIR/$tdir/d2/$tfile ||
error "$tdir/d2/$tfile not a file"
}
run_test 10 "mkdir .../d10 .../d10/d2; touch .../d10/d2/f ======"
test_11() {
test_mkdir $DIR/$tdir
test_mkdir $DIR/$tdir/d2
chmod 0666 $DIR/$tdir/d2
chmod 0705 $DIR/$tdir/d2
$CHECKSTAT -t dir -p 0705 $DIR/$tdir/d2 ||
error "$tdir/d2 mode not 0705"
}
run_test 11 "mkdir .../d11 d11/d2; chmod .../d11/d2 ============"
test_12() {
test_mkdir $DIR/$tdir
touch $DIR/$tdir/$tfile
chmod 0666 $DIR/$tdir/$tfile
chmod 0654 $DIR/$tdir/$tfile
$CHECKSTAT -t file -p 0654 $DIR/$tdir/$tfile ||
error "$tdir/d2 mode not 0654"
}
run_test 12 "touch .../d12/f; chmod .../d12/f .../d12/f ========"
test_13() {
test_mkdir $DIR/$tdir
dd if=/dev/zero of=$DIR/$tdir/$tfile count=10
> $DIR/$tdir/$tfile
$CHECKSTAT -t file -s 0 $DIR/$tdir/$tfile ||
error "$tdir/$tfile size not 0 after truncate"
}
run_test 13 "creat .../d13/f; dd .../d13/f; > .../d13/f ========"
test_14() {
test_mkdir $DIR/$tdir
touch $DIR/$tdir/$tfile
rm $DIR/$tdir/$tfile
$CHECKSTAT -a $DIR/$tdir/$tfile || error "$tdir/$tfile not removed"
}
run_test 14 "touch .../d14/f; rm .../d14/f; rm .../d14/f ======="
test_15() {
test_mkdir $DIR/$tdir
touch $DIR/$tdir/$tfile
mv $DIR/$tdir/$tfile $DIR/$tdir/${tfile}_2
$CHECKSTAT -t file $DIR/$tdir/${tfile}_2 ||
error "$tdir/${tfile_2} not a file after rename"
rm $DIR/$tdir/${tfile}_2 || error "unlink failed after rename"
}
run_test 15 "touch .../d15/f; mv .../d15/f .../d15/f2 =========="
test_16() {
test_mkdir $DIR/$tdir
touch $DIR/$tdir/$tfile
rm -rf $DIR/$tdir/$tfile
$CHECKSTAT -a $DIR/$tdir/$tfile || error "$tdir/$tfile not removed"
}
run_test 16 "touch .../d16/f; rm -rf .../d16/f"
test_17a() {
test_mkdir $DIR/$tdir
touch $DIR/$tdir/$tfile
ln -s $DIR/$tdir/$tfile $DIR/$tdir/l-exist
ls -l $DIR/$tdir
$CHECKSTAT -l $DIR/$tdir/$tfile $DIR/$tdir/l-exist ||
error "$tdir/l-exist not a symlink"
$CHECKSTAT -f -t f $DIR/$tdir/l-exist ||
error "$tdir/l-exist not referencing a file"
rm -f $DIR/$tdir/l-exist
$CHECKSTAT -a $DIR/$tdir/l-exist || error "$tdir/l-exist not removed"
}
run_test 17a "symlinks: create, remove (real)"
test_17b() {
test_mkdir $DIR/$tdir
ln -s no-such-file $DIR/$tdir/l-dangle
ls -l $DIR/$tdir
$CHECKSTAT -l no-such-file $DIR/$tdir/l-dangle ||
error "$tdir/l-dangle not referencing no-such-file"
$CHECKSTAT -fa $DIR/$tdir/l-dangle ||
error "$tdir/l-dangle not referencing non-existent file"
rm -f $DIR/$tdir/l-dangle
$CHECKSTAT -a $DIR/$tdir/l-dangle || error "$tdir/l-dangle not removed"
}
run_test 17b "symlinks: create, remove (dangling)"
test_17c() { # bug 3440 - don't save failed open RPC for replay
test_mkdir $DIR/$tdir
ln -s foo $DIR/$tdir/$tfile
cat $DIR/$tdir/$tfile && error "opened non-existent symlink" || true
}
run_test 17c "symlinks: open dangling (should return error)"
test_17d() {
test_mkdir $DIR/$tdir
ln -s foo $DIR/$tdir/$tfile
touch $DIR/$tdir/$tfile || error "creating to new symlink"
}
run_test 17d "symlinks: create dangling"
test_17e() {
test_mkdir $DIR/$tdir
local foo=$DIR/$tdir/$tfile
ln -s $foo $foo || error "create symlink failed"
ls -l $foo || error "ls -l failed"
ls $foo && error_ignore LU-18093 "ls not failed" || true
}
run_test 17e "symlinks: create recursive symlink (should return error)"
test_17f() {
test_mkdir $DIR/$tdir
ln -s 1234567890/2234567890/3234567890/4234567890 $DIR/$tdir/111
ln -s 1234567890/2234567890/3234567890/4234567890/5234567890/6234567890 $DIR/$tdir/222
ln -s 1234567890/2234567890/3234567890/4234567890/5234567890/6234567890/7234567890/8234567890 $DIR/$tdir/333
ln -s 1234567890/2234567890/3234567890/4234567890/5234567890/6234567890/7234567890/8234567890/9234567890/a234567890/b234567890 $DIR/$tdir/444
ln -s 1234567890/2234567890/3234567890/4234567890/5234567890/6234567890/7234567890/8234567890/9234567890/a234567890/b234567890/c234567890/d234567890/f234567890 $DIR/$tdir/555
ln -s 1234567890/2234567890/3234567890/4234567890/5234567890/6234567890/7234567890/8234567890/9234567890/a234567890/b234567890/c234567890/d234567890/f234567890/aaaaaaaaaa/bbbbbbbbbb/cccccccccc/dddddddddd/eeeeeeeeee/ffffffffff/ $DIR/$tdir/666
ls -l $DIR/$tdir
}
run_test 17f "symlinks: long and very long symlink name"
# str_repeat(S, N) generate a string that is string S repeated N times
str_repeat() {
local s=$1
local n=$2
local ret=''
while [ $((n -= 1)) -ge 0 ]; do
ret=$ret$s
done
echo $ret
}
# Long symlinks and LU-2241
test_17g() {
test_mkdir $DIR/$tdir
local TESTS="59 60 61 4094 4095"
# Fix for inode size boundary in 2.1.4
[ $MDS1_VERSION -lt $(version_code 2.1.4) ] &&
TESTS="4094 4095"
# Patch not applied to 2.2 or 2.3 branches
[ $MDS1_VERSION -ge $(version_code 2.2.0) ] &&
[ $MDS1_VERSION -le $(version_code 2.3.55) ] &&
TESTS="4094 4095"
for i in $TESTS; do
local SYMNAME=$(str_repeat 'x' $i)
ln -s $SYMNAME $DIR/$tdir/f$i || error "failed $i-char symlink"
readlink $DIR/$tdir/f$i || error "failed $i-char readlink"
done
}
run_test 17g "symlinks: really long symlink name and inode boundaries"
test_17h() { #bug 17378
[ $PARALLEL == "yes" ] && skip "skip parallel run"
remote_mds_nodsh && skip "remote MDS with nodsh"
local mdt_idx
test_mkdir $DIR/$tdir
mdt_idx=$($LFS getdirstripe -i $DIR/$tdir)
$LFS setstripe -c -1 $DIR/$tdir
#define OBD_FAIL_MDS_LOV_PREP_CREATE 0x141
do_facet mds$((mdt_idx + 1)) lctl set_param fail_loc=0x80000141
touch $DIR/$tdir/$tfile || true
}
run_test 17h "create objects: lov_free_memmd() doesn't lbug"
test_17i() { #bug 20018
[ $PARALLEL == "yes" ] && skip "skip parallel run"
remote_mds_nodsh && skip "remote MDS with nodsh"
local foo=$DIR/$tdir/$tfile
local mdt_idx
test_mkdir -c1 $DIR/$tdir
mdt_idx=$($LFS getdirstripe -i $DIR/$tdir)
ln -s $foo $foo || error "create symlink failed"
#define OBD_FAIL_MDS_READLINK_EPROTO 0x143
do_facet mds$((mdt_idx + 1)) lctl set_param fail_loc=0x80000143
ls -l $foo && error "error not detected"
return 0
}
run_test 17i "don't panic on short symlink (should return error)"
test_17k() { #bug 22301
[ $PARALLEL == "yes" ] && skip "skip parallel run"
[[ -z "$(which rsync 2>/dev/null)" ]] &&
skip "no rsync command"
rsync --help | grep -q xattr ||
skip_env "$(rsync --version | head -n1) does not support xattrs"
test_mkdir $DIR/$tdir
test_mkdir $DIR/$tdir.new
touch $DIR/$tdir/$tfile
ln -s $DIR/$tdir/$tfile $DIR/$tdir/$tfile.lnk
rsync -av -X $DIR/$tdir/ $DIR/$tdir.new ||
error "rsync failed with xattrs enabled"
}
run_test 17k "symlinks: rsync with xattrs enabled"
test_17l() { # LU-279
[[ -z "$(which getfattr 2>/dev/null)" ]] &&
skip "no getfattr command"
test_mkdir $DIR/$tdir
touch $DIR/$tdir/$tfile
ln -s $DIR/$tdir/$tfile $DIR/$tdir/$tfile.lnk
for path in "$DIR/$tdir" "$DIR/$tdir/$tfile" "$DIR/$tdir/$tfile.lnk"; do
# -h to not follow symlinks. -m '' to list all the xattrs.
# grep to remove first line: '# file: $path'.
for xattr in `getfattr -hm '' $path 2>/dev/null | grep -v '^#'`;
do
lgetxattr_size_check $path $xattr ||
error "lgetxattr_size_check $path $xattr failed"
done
done
}
run_test 17l "Ensure lgetxattr's returned xattr size is consistent"
# LU-1540
test_17m() {
[ $PARALLEL == "yes" ] && skip "skip parallel run"
[ "$mds1_FSTYPE" != "ldiskfs" ] && skip_env "ldiskfs only test"
remote_mds_nodsh && skip "remote MDS with nodsh"
[ $MDS1_VERSION -ge $(version_code 2.2.0) ] &&
[ $MDS1_VERSION -le $(version_code 2.2.93) ] &&
skip "MDS 2.2.0-2.2.93 do not NUL-terminate symlinks"
local short_sym="0123456789"
local wdir=$DIR/$tdir
local i
test_mkdir $wdir
long_sym=$short_sym
# create a long symlink file
for ((i = 0; i < 4; ++i)); do
long_sym=${long_sym}${long_sym}
done
echo "create 512 short and long symlink files under $wdir"
for ((i = 0; i < 256; ++i)); do
ln -sf ${long_sym}"a5a5" $wdir/long-$i
ln -sf ${short_sym}"a5a5" $wdir/short-$i
done
echo "erase them"
rm -f $wdir/*
sync
wait_delete_completed
echo "recreate the 512 symlink files with a shorter string"
for ((i = 0; i < 512; ++i)); do
# rewrite the symlink file with a shorter string
ln -sf ${long_sym} $wdir/long-$i || error "long_sym failed"
ln -sf ${short_sym} $wdir/short-$i || error "short_sym failed"
done
local mds_index=$(($($LFS getstripe -m $wdir) + 1))
echo "stop and checking mds${mds_index}:"
# e2fsck should not return error
stop mds${mds_index}
local devname=$(mdsdevname $mds_index)
run_e2fsck $(facet_active_host mds${mds_index}) $devname -n
rc=$?
start mds${mds_index} $devname $MDS_MOUNT_OPTS ||
error "start mds${mds_index} failed"
df $MOUNT > /dev/null 2>&1
[ $rc -eq 0 ] ||
error "e2fsck detected error for short/long symlink: rc=$rc"
rm -f $wdir/*
}
run_test 17m "run e2fsck against MDT which contains short/long symlink"
check_fs_consistency_17n() {
local mdt_index
local rc=0
# create/unlink in 17n only change 2 MDTs(MDT1/MDT2),
# so it only check MDT1/MDT2 instead of all of MDTs.
for mdt_index in 1 2; do
# e2fsck should not return error
stop mds${mdt_index}
local devname=$(mdsdevname $mdt_index)
run_e2fsck $(facet_active_host mds$mdt_index) $devname -n ||
rc=$((rc + $?))
start mds${mdt_index} $devname $MDS_MOUNT_OPTS ||
error "mount mds$mdt_index failed"
df $MOUNT > /dev/null 2>&1
done
return $rc
}
test_17n() {
[[ $MDSCOUNT -lt 2 ]] && skip_env "needs >= 2 MDTs"
[ $PARALLEL == "yes" ] && skip "skip parallel run"
[ "$mds1_FSTYPE" != "ldiskfs" ] && skip_env "ldiskfs only test"
remote_mds_nodsh && skip "remote MDS with nodsh"
[ $MDS1_VERSION -ge $(version_code 2.2.0) ] &&
[ $MDS1_VERSION -le $(version_code 2.2.93) ] &&
skip "MDS 2.2.0-2.2.93 do not NUL-terminate symlinks"
local i
test_mkdir $DIR/$tdir
for ((i=0; i<10; i++)); do
$LFS mkdir -i1 -c2 $DIR/$tdir/remote_dir_${i} ||
error "create remote dir error $i"
createmany -o $DIR/$tdir/remote_dir_${i}/f 10 ||
error "create files under remote dir failed $i"
done
check_fs_consistency_17n ||
error "e2fsck report error after create files under remote dir"
for ((i = 0; i < 10; i++)); do
rm -rf $DIR/$tdir/remote_dir_${i} ||
error "destroy remote dir error $i"
done
check_fs_consistency_17n ||
error "e2fsck report error after unlink files under remote dir"
[ $MDS1_VERSION -lt $(version_code 2.4.50) ] &&
skip "lustre < 2.4.50 does not support migrate mv"
for ((i = 0; i < 10; i++)); do
mkdir -p $DIR/$tdir/remote_dir_${i}
createmany -o $DIR/$tdir/remote_dir_${i}/f 10 ||
error "create files under remote dir failed $i"
$LFS migrate --mdt-index 1 $DIR/$tdir/remote_dir_${i} ||
error "migrate remote dir error $i"
done
check_fs_consistency_17n || error "e2fsck report error after migration"
for ((i = 0; i < 10; i++)); do
rm -rf $DIR/$tdir/remote_dir_${i} ||
error "destroy remote dir error $i"
done
check_fs_consistency_17n || error "e2fsck report error after unlink"
}
run_test 17n "run e2fsck against master/slave MDT which contains remote dir"
test_17o() {
remote_mds_nodsh && skip "remote MDS with nodsh"
[ $MDS1_VERSION -lt $(version_code 2.3.64) ] &&
skip "Need MDS version at least 2.3.64"
local wdir=$DIR/${tdir}o
local mdt_index
local rc=0
test_mkdir $wdir
touch $wdir/$tfile
mdt_index=$($LFS getstripe -m $wdir/$tfile)
mdt_index=$((mdt_index + 1))
cancel_lru_locks mdc
#fail mds will wait the failover finish then set
#following fail_loc to avoid interfer the recovery process.
fail mds${mdt_index}
#define OBD_FAIL_OSD_LMA_INCOMPAT 0x194
do_facet mds${mdt_index} lctl set_param fail_loc=0x194
ls -l $wdir/$tfile && rc=1
do_facet mds${mdt_index} lctl set_param fail_loc=0
[[ $rc -eq 0 ]] || error "stat file should fail"
}
run_test 17o "stat file with incompat LMA feature"
# LU-17660: "cannot overwrite directory" when creating symlink
test_17p() {
touch $DIR/$tfile
test_mkdir $DIR/$tdir
# there is a kernel bug in el9.x series (9.0 - 9.4 as we know till now)
# kernel, we need to stat the target dir to cache it first
if [[ "$CLIENT_OS_ID_LIKE" =~ "rhel" ]]; then
if (( $CLIENT_OS_VERSION_CODE >= $(version_code 9.0) &&
$CLIENT_OS_VERSION_CODE <= $(version_code 9.100) )); then
echo "stat $DIR/$tdir to cache it in el9.x"
stat $DIR/$tdir
fi
fi
strace ln -sf $DIR/$tfile $DIR/$tdir/ ||
error "Failed to create symlink $DIR/$tfile under $DIR/$tdir/"
}
run_test 17p "symlink overwrite directory error message"
test_17q() {
(( $MDS1_VERSION >= $(version_code v2_16_56-96-gd0c75da783) )) ||
skip "Need MDS >= 2.16.56.96 for symlink xattr"
local mdts=$(mdts_nodes)
ln -s foo $DIR/$tfile
setfattr -h -n trusted.test -v "$(head -200 /etc/services)" $DIR/$tfile ||
error "setfattr large xattr on symlink failed"
cancel_lru_locks mdc
sysctl -w vm.drop_caches=3
do_nodes $mdts sysctl -w vm.drop_caches=3
ls -l $DIR/$tfile || error "ls -l failed"
}
run_test 17q "set large xattr on fast symlink"
test_18() {
touch $DIR/$tfile || error "Failed to touch $DIR/$tfile: $?"
ls $DIR || error "Failed to ls $DIR: $?"
}
run_test 18 "touch .../f ; ls ... =============================="
test_19a() {
touch $DIR/$tfile
ls -l $DIR
rm $DIR/$tfile
$CHECKSTAT -a $DIR/$tfile || error "$tfile was not removed"
}
run_test 19a "touch .../f19 ; ls -l ... ; rm .../f19 ==========="
test_19b() {
ls -l $DIR/$tfile && error "ls -l $tfile failed"|| true
}
run_test 19b "ls -l .../f19 (should return error) =============="
test_19c() {
[ $RUNAS_ID -eq $UID ] &&
skip_env "RUNAS_ID = UID = $UID -- skipping"
$RUNAS touch $DIR/$tfile && error "create non-root file failed" || true
}
run_test 19c "$RUNAS touch .../f19 (should return error) =="
test_19d() {
cat $DIR/f19 && error || true
}
run_test 19d "cat .../f19 (should return error) =============="
test_20() {
touch $DIR/$tfile
rm $DIR/$tfile
touch $DIR/$tfile
rm $DIR/$tfile
touch $DIR/$tfile
rm $DIR/$tfile
$CHECKSTAT -a $DIR/$tfile || error "$tfile was not removed"
}
run_test 20 "touch .../f ; ls -l ..."
test_21() {
test_mkdir $DIR/$tdir
[ -f $DIR/$tdir/dangle ] && rm -f $DIR/$tdir/dangle
ln -s dangle $DIR/$tdir/link
echo foo >> $DIR/$tdir/link
cat $DIR/$tdir/dangle
$CHECKSTAT -t link $DIR/$tdir/link || error "$tdir/link not a link"
$CHECKSTAT -f -t file $DIR/$tdir/link ||
error "$tdir/link not linked to a file"
}
run_test 21 "write to dangling link"
test_22() {
local wdir=$DIR/$tdir
test_mkdir $wdir
chown $RUNAS_ID:$RUNAS_GID $wdir
(cd $wdir || error "cd $wdir failed";
$RUNAS tar cf - /etc/hosts /etc/sysconfig/network |
$RUNAS tar xf -)
ls -lR $wdir/etc || error "ls -lR $wdir/etc failed"
$CHECKSTAT -t dir $wdir/etc || error "checkstat -t dir failed"
$CHECKSTAT -u \#$RUNAS_ID -g \#$RUNAS_GID $wdir/etc ||
error "checkstat -u failed"
}
run_test 22 "unpack tar archive as non-root user"
# was test_23
test_23a() {
test_mkdir $DIR/$tdir
local file=$DIR/$tdir/$tfile
openfile -f O_CREAT:O_EXCL $file || error "$file create failed"
openfile -f O_CREAT:O_EXCL $file &&
error "$file recreate succeeded" || true
}
run_test 23a "O_CREAT|O_EXCL in subdir"
test_23b() { # bug 18988
test_mkdir $DIR/$tdir
local file=$DIR/$tdir/$tfile
rm -f $file
echo foo > $file || error "write filed"
echo bar >> $file || error "append filed"
$CHECKSTAT -s 8 $file || error "wrong size"
rm $file
}
run_test 23b "O_APPEND check"
# LU-9409, size with O_APPEND and tiny writes
test_23c() {
local file=$DIR/$tfile
# single dd
dd conv=notrunc oflag=append if=/dev/zero of=$file bs=8 count=800
$CHECKSTAT -s 6400 $file || error "wrong size, expected 6400"
rm -f $file
# racing tiny writes
dd conv=notrunc oflag=append if=/dev/zero of=$file bs=8 count=800 &
dd conv=notrunc oflag=append if=/dev/zero of=$file bs=8 count=800 &
wait
$CHECKSTAT -s 12800 $file || error "wrong size, expected 12800"
rm -f $file
#racing tiny & normal writes
dd conv=notrunc oflag=append if=/dev/zero of=$file bs=4096 count=4 &
dd conv=notrunc oflag=append if=/dev/zero of=$file bs=8 count=100 &
wait
$CHECKSTAT -s 17184 $file || error "wrong size, expected 17184"
rm -f $file
#racing tiny & normal writes 2, ugly numbers
dd conv=notrunc oflag=append if=/dev/zero of=$file bs=4099 count=11 &
dd conv=notrunc oflag=append if=/dev/zero of=$file bs=17 count=173 &
wait
$CHECKSTAT -s 48030 $file || error "wrong size, expected 48030"
rm -f $file
}
run_test 23c "O_APPEND size checks for tiny writes"
# LU-11069 file offset is correct after appending writes
test_23d() {
local file=$DIR/$tfile
local offset
echo CentaurHauls > $file
offset=$($MULTIOP $file oO_WRONLY:O_APPEND:w13Zp)
if ((offset != 26)); then
error "wrong offset, expected 26, got '$offset'"
fi
}
run_test 23d "file offset is correct after appending writes"
# rename sanity
test_24a() {
echo '-- same directory rename'
test_mkdir $DIR/$tdir
touch $DIR/$tdir/$tfile.1
mv $DIR/$tdir/$tfile.1 $DIR/$tdir/$tfile.2
$CHECKSTAT -t file $DIR/$tdir/$tfile.2 || error "$tfile.2 not a file"
}
run_test 24a "rename file to non-existent target"
test_24b() {
test_mkdir $DIR/$tdir
touch $DIR/$tdir/$tfile.{1,2}
mv $DIR/$tdir/$tfile.1 $DIR/$tdir/$tfile.2
$CHECKSTAT -a $DIR/$tdir/$tfile.1 || error "$tfile.1 exists"
$CHECKSTAT -t file $DIR/$tdir/$tfile.2 || error "$tfile.2 not a file"
}
run_test 24b "rename file to existing target"
test_24c() {
test_mkdir $DIR/$tdir
test_mkdir $DIR/$tdir/d$testnum.1
mv $DIR/$tdir/d$testnum.1 $DIR/$tdir/d$testnum.2
$CHECKSTAT -a $DIR/$tdir/d$testnum.1 || error "d$testnum.1 exists"
$CHECKSTAT -t dir $DIR/$tdir/d$testnum.2 || error "d$testnum.2 not dir"
}
run_test 24c "rename directory to non-existent target"
test_24d() {
test_mkdir -c1 $DIR/$tdir
test_mkdir -c1 $DIR/$tdir/d$testnum.1
test_mkdir -c1 $DIR/$tdir/d$testnum.2
mrename $DIR/$tdir/d$testnum.1 $DIR/$tdir/d$testnum.2
$CHECKSTAT -a $DIR/$tdir/d$testnum.1 || error "d$testnum.1 exists"
$CHECKSTAT -t dir $DIR/$tdir/d$testnum.2 || error "d$testnum.2 not dir"
}
run_test 24d "rename directory to existing target"
test_24e() {
echo '-- cross directory renames --'
test_mkdir $DIR/R5a
test_mkdir $DIR/R5b
touch $DIR/R5a/f
mv $DIR/R5a/f $DIR/R5b/g
$CHECKSTAT -a $DIR/R5a/f || error "$DIR/R5a/f exists"
$CHECKSTAT -t file $DIR/R5b/g || error "$DIR/R5b/g not file type"
}
run_test 24e "touch .../R5a/f; rename .../R5a/f .../R5b/g ======"
test_24f() {
test_mkdir $DIR/R6a
test_mkdir $DIR/R6b
touch $DIR/R6a/f $DIR/R6b/g
mv $DIR/R6a/f $DIR/R6b/g
$CHECKSTAT -a $DIR/R6a/f || error "$DIR/R6a/f exists"
$CHECKSTAT -t file $DIR/R6b/g || error "$DIR/R6b/g not file type"
}
run_test 24f "touch .../R6a/f R6b/g; mv .../R6a/f .../R6b/g ===="
test_24g() {
test_mkdir $DIR/R7a
test_mkdir $DIR/R7b
test_mkdir $DIR/R7a/d
mv $DIR/R7a/d $DIR/R7b/e
$CHECKSTAT -a $DIR/R7a/d || error "$DIR/R7a/d exists"
$CHECKSTAT -t dir $DIR/R7b/e || error "$DIR/R7b/e not dir type"
}
run_test 24g "mkdir .../R7{a,b}/d; mv .../R7a/d .../R7b/e ======"
test_24h() {
test_mkdir -c1 $DIR/R8a
test_mkdir -c1 $DIR/R8b
test_mkdir -c1 $DIR/R8a/d
test_mkdir -c1 $DIR/R8b/e
mrename $DIR/R8a/d $DIR/R8b/e
$CHECKSTAT -a $DIR/R8a/d || error "$DIR/R8a/d exists"
$CHECKSTAT -t dir $DIR/R8b/e || error "$DIR/R8b/e not dir type"
}
run_test 24h "mkdir .../R8{a,b}/{d,e}; rename .../R8a/d .../R8b/e"
test_24i() {
echo "-- rename error cases"
test_mkdir $DIR/R9
test_mkdir $DIR/R9/a
touch $DIR/R9/f
mrename $DIR/R9/f $DIR/R9/a
$CHECKSTAT -t file $DIR/R9/f || error "$DIR/R9/f not file type"
$CHECKSTAT -t dir $DIR/R9/a || error "$DIR/R9/a not dir type"
$CHECKSTAT -a $DIR/R9/a/f || error "$DIR/R9/a/f exists"
}
run_test 24i "rename file to dir error: touch f ; mkdir a ; rename f a"
test_24j() {
test_mkdir $DIR/R10
mrename $DIR/R10/f $DIR/R10/g
$CHECKSTAT -t dir $DIR/R10 || error "$DIR/R10 not dir type"
$CHECKSTAT -a $DIR/R10/f || error "$DIR/R10/f exists"
$CHECKSTAT -a $DIR/R10/g || error "$DIR/R10/g exists"
}
run_test 24j "source does not exist ============================"
test_24k() {
test_mkdir $DIR/R11a
test_mkdir $DIR/R11a/d
touch $DIR/R11a/f
mv $DIR/R11a/f $DIR/R11a/d
$CHECKSTAT -a $DIR/R11a/f || error "$DIR/R11a/f exists"
$CHECKSTAT -t file $DIR/R11a/d/f || error "$DIR/R11a/d/f not file type"
}
run_test 24k "touch .../R11a/f; mv .../R11a/f .../R11a/d ======="
# bug 2429 - rename foo foo foo creates invalid file
test_24l() {
f="$DIR/f24l"
$MULTIOP $f OcNs || error "rename of ${f} to itself failed"
}
run_test 24l "Renaming a file to itself ========================"
test_24m() {
f="$DIR/f24m"
$MULTIOP $f OcLN ${f}2 ${f}2 || error "link ${f}2 ${f}2 failed"
# on ext3 this does not remove either the source or target files
# though the "expected" operation would be to remove the source
$CHECKSTAT -t file ${f} || error "${f} missing"
$CHECKSTAT -t file ${f}2 || error "${f}2 missing"
}
run_test 24m "Renaming a file to a hard link to itself ========="
test_24n() {
f="$DIR/f24n"
# this stats the old file after it was renamed, so it should fail
touch ${f}
$CHECKSTAT ${f} || error "${f} missing"
mv ${f} ${f}.rename
$CHECKSTAT ${f}.rename || error "${f}.rename missing"
$CHECKSTAT -a ${f} || error "${f} exists"
}
run_test 24n "Statting the old file after renaming (Posix rename 2)"
test_24o() {
test_mkdir $DIR/$tdir
rename_many -s random -v -n 10 $DIR/$tdir
}
run_test 24o "rename of files during htree split"
test_24p() {
test_mkdir $DIR/R12a
test_mkdir $DIR/R12b
DIRINO=`ls -lid $DIR/R12a | awk '{ print $1 }'`
mrename $DIR/R12a $DIR/R12b
$CHECKSTAT -a $DIR/R12a || error "$DIR/R12a exists"
$CHECKSTAT -t dir $DIR/R12b || error "$DIR/R12b not dir type"
DIRINO2=`ls -lid $DIR/R12b | awk '{ print $1 }'`
[ "$DIRINO" = "$DIRINO2" ] || error "R12a $DIRINO != R12b $DIRINO2"
}
run_test 24p "mkdir .../R12{a,b}; rename .../R12a .../R12b"
cleanup_multiop_pause() {
trap 0
kill -USR1 $MULTIPID
}
test_24q() {
[ $PARALLEL == "yes" ] && skip "skip parallel run"
test_mkdir $DIR/R13a
test_mkdir $DIR/R13b
local DIRINO=$(ls -lid $DIR/R13a | awk '{ print $1 }')
multiop_bg_pause $DIR/R13b D_c || error "multiop failed to start"
MULTIPID=$!
trap cleanup_multiop_pause EXIT
mrename $DIR/R13a $DIR/R13b
$CHECKSTAT -a $DIR/R13a || error "R13a still exists"
$CHECKSTAT -t dir $DIR/R13b || error "R13b does not exist"
local DIRINO2=$(ls -lid $DIR/R13b | awk '{ print $1 }')
[ "$DIRINO" = "$DIRINO2" ] || error "R13a $DIRINO != R13b $DIRINO2"
cleanup_multiop_pause
wait $MULTIPID || error "multiop close failed"
}
run_test 24q "mkdir .../R13{a,b}; open R13b rename R13a R13b ==="
test_24r() { #bug 3789
test_mkdir $DIR/R14a
test_mkdir $DIR/R14a/b
mrename $DIR/R14a $DIR/R14a/b && error "rename to subdir worked!"
$CHECKSTAT -t dir $DIR/R14a || error "$DIR/R14a missing"
$CHECKSTAT -t dir $DIR/R14a/b || error "$DIR/R14a/b missing"
}
run_test 24r "mkdir .../R14a/b; rename .../R14a .../R14a/b ====="
test_24s() {
test_mkdir $DIR/R15a
test_mkdir $DIR/R15a/b
test_mkdir $DIR/R15a/b/c
mrename $DIR/R15a $DIR/R15a/b/c && error "rename to sub-subdir worked!"
$CHECKSTAT -t dir $DIR/R15a || error "$DIR/R15a missing"
$CHECKSTAT -t dir $DIR/R15a/b/c || error "$DIR/R15a/b/c missing"
}
run_test 24s "mkdir .../R15a/b/c; rename .../R15a .../R15a/b/c ="
test_24t() {
test_mkdir $DIR/R16a
test_mkdir $DIR/R16a/b
test_mkdir $DIR/R16a/b/c
mrename $DIR/R16a/b/c $DIR/R16a && error "rename to sub-subdir worked!"
$CHECKSTAT -t dir $DIR/R16a || error "$DIR/R16a missing"
$CHECKSTAT -t dir $DIR/R16a/b/c || error "$DIR/R16a/b/c missing"
}
run_test 24t "mkdir .../R16a/b/c; rename .../R16a/b/c .../R16a ="
test_24u() { # bug12192
stack_trap "rm -f $DIR/$tfile"
$MULTIOP $DIR/$tfile C2w$((2048 * 1024))c || error "multiop failed"
$CHECKSTAT -s $((2048 * 1024)) $DIR/$tfile || error "wrong file size"
}
run_test 24u "create stripe file"
simple_cleanup_common() {
local createmany=$1
local rc=0
[[ -z "$DIR" || -z "$tdir" || ! -d "$DIR/$tdir" ]] && return 0
local start=$SECONDS
[[ -n "$createmany" ]] && unlinkmany $DIR/$tdir/$tfile $createmany
rm -rf $DIR/$tdir || error "cleanup $DIR/$tdir failed"
rc=$?
wait_delete_completed
echo "cleanup time $((SECONDS - start))"
return $rc
}
max_pages_per_rpc() {
local mdtname="$(printf "MDT%04x" ${1:-0})"
$LCTL get_param -n mdc.*$mdtname*.max_pages_per_rpc
}
test_24v() {
[ $PARALLEL == "yes" ] && skip "skip parallel run"
local nrfiles=${COUNT:-100000}
local fname="$DIR/$tdir/$tfile"
# Performance issue on ZFS see LU-4072 (c.f. LU-2887)
[ "$mds1_FSTYPE" = "zfs" ] && nrfiles=${COUNT:-10000}
test_mkdir "$(dirname $fname)"
# assume MDT0000 has the fewest inodes
local stripes=$($LFS getdirstripe -c $(dirname $fname))
local free_inodes=$(($(mdt_free_inodes 0) * ${stripes/#0/1}))
[[ $free_inodes -lt $nrfiles ]] && nrfiles=$free_inodes
stack_trap "simple_cleanup_common $nrfiles"
createmany -m "$fname" $nrfiles
cancel_lru_locks mdc
lctl set_param mdc.*.stats clear
# was previously test_24D: LU-6101
# readdir() returns correct number of entries after cursor reload
local num_ls=$(ls $DIR/$tdir | wc -l)
local num_uniq=$(ls $DIR/$tdir | sort -u | wc -l)
local num_all=$(ls -a $DIR/$tdir | wc -l)
if [ $num_ls -ne $nrfiles ] || [ $num_uniq -ne $nrfiles ] ||
[ $num_all -ne $((nrfiles + 2)) ]; then
error "Expected $nrfiles files, got $num_ls " \
"($num_uniq unique $num_all .&..)"
fi
# LU-5 large readdir
# dirent_size = 32 bytes for sizeof(struct lu_dirent) +
# N bytes for name (len($nrfiles) rounded to 8 bytes) +
# 8 bytes for luda_type (4 bytes rounded to 8 bytes)
# take into account of overhead in lu_dirpage header and end mark in
# each page, plus one in rpc_num calculation.
local dirent_size=$((32 + (${#tfile} | 7) + 1 + 8))
local page_entries=$(((PAGE_SIZE - 24) / dirent_size))
local mdt_idx=$($LFS getdirstripe -i $(dirname $fname))
local rpc_pages=$(max_pages_per_rpc $mdt_idx)
local rpc_max=$((nrfiles / (page_entries * rpc_pages) + stripes))
local mds_readpage=$(calc_stats mdc.*.stats mds_readpage)
echo "readpages: $mds_readpage rpc_max: $rpc_max-2/+1"
(( $mds_readpage >= $rpc_max - 2 && $mds_readpage <= $rpc_max + 1)) ||
error "large readdir doesn't take effect: " \
"$mds_readpage should be about $rpc_max"
}
run_test 24v "list large directory (test hash collision, b=17560)"
test_24w() { # bug21506
SZ1=234852
stack_trap "rm -f $DIR/$tfile"
dd if=/dev/zero of=$DIR/$tfile bs=1M count=1 seek=4096 || return 1
dd if=/dev/zero bs=$SZ1 count=1 >> $DIR/$tfile || return 2
dd if=$DIR/$tfile of=$DIR/${tfile}_left bs=1M skip=4097 || return 3
SZ2=`ls -l $DIR/${tfile}_left | awk '{print $5}'`
[[ "$SZ1" -eq "$SZ2" ]] ||
error "Error reading at the end of the file $tfile"
}
run_test 24w "Reading a file larger than 4Gb"
test_24x() {
[[ $MDSCOUNT -lt 2 ]] && skip_env "needs >= 2 MDTs"
[ $PARALLEL == "yes" ] && skip "skip parallel run"
[[ $MDS1_VERSION -lt $(version_code 2.7.56) ]] &&
skip "Need MDS version at least 2.7.56"
local MDTIDX=1
local remote_dir=$DIR/$tdir/remote_dir
test_mkdir $DIR/$tdir
$LFS mkdir -i $MDTIDX $remote_dir ||
error "create remote directory failed"
test_mkdir $DIR/$tdir/src_dir
touch $DIR/$tdir/src_file
test_mkdir $remote_dir/tgt_dir
touch $remote_dir/tgt_file
mrename $DIR/$tdir/src_dir $remote_dir/tgt_dir ||
error "rename dir cross MDT failed!"
mrename $DIR/$tdir/src_file $remote_dir/tgt_file ||
error "rename file cross MDT failed!"
touch $DIR/$tdir/ln_file
ln $DIR/$tdir/ln_file $remote_dir/ln_name ||
error "ln file cross MDT failed"
rm -rf $DIR/$tdir || error "Can not delete directories"
}
run_test 24x "cross MDT rename/link"
test_24y() {
[[ $MDSCOUNT -lt 2 ]] && skip_env "needs >= 2 MDTs"
[ $PARALLEL == "yes" ] && skip "skip parallel run"
local remote_dir=$DIR/$tdir/remote_dir
local mdtidx=1
test_mkdir $DIR/$tdir
$LFS mkdir -i $mdtidx $remote_dir ||
error "create remote directory failed"
test_mkdir $remote_dir/src_dir
touch $remote_dir/src_file
test_mkdir $remote_dir/tgt_dir
touch $remote_dir/tgt_file
mrename $remote_dir/src_dir $remote_dir/tgt_dir ||
error "rename subdir in the same remote dir failed!"
mrename $remote_dir/src_file $remote_dir/tgt_file ||
error "rename files in the same remote dir failed!"
ln $remote_dir/tgt_file $remote_dir/tgt_file1 ||
error "link files in the same remote dir failed!"
rm -rf $DIR/$tdir || error "Can not delete directories"
}
run_test 24y "rename/link on the same dir should succeed"
test_24z() {
[[ $MDSCOUNT -lt 2 ]] && skip_env "needs >= 2 MDTs"
[[ $MDS1_VERSION -lt $(version_code 2.12.51) ]] &&
skip "Need MDS version at least 2.12.51"
local index
for index in 0 1; do
$LFS mkdir -i $index $DIR/$tdir.$index || error "mkdir failed"
touch $DIR/$tdir.0/$tfile.$index || error "touch failed"
done
mv $DIR/$tdir.0/$tfile.0 $DIR/$tdir.1 || error "mv $tfile.0 failed"
index=$($LFS getstripe -m $DIR/$tdir.1/$tfile.0)
[ $index -eq 0 ] || error "$tfile.0 is on MDT$index"
local mdts=$(mdts_nodes)
do_nodes $mdts $LCTL set_param mdt.*.enable_remote_rename=0
stack_trap "do_nodes $mdts $LCTL \
set_param mdt.*.enable_remote_rename=1" EXIT
mv $DIR/$tdir.0/$tfile.1 $DIR/$tdir.1 || error "mv $tfile.1 failed"
index=$($LFS getstripe -m $DIR/$tdir.1/$tfile.1)
[ $index -eq 1 ] || error "$tfile.1 is on MDT$index"
}
run_test 24z "cross-MDT rename is done as cp"
test_24A() { # LU-3182
local NFILES=5000
test_mkdir $DIR/$tdir
stack_trap "simple_cleanup_common $NFILES"
createmany -m $DIR/$tdir/$tfile $NFILES
local t=$(ls $DIR/$tdir | wc -l)
local u=$(ls $DIR/$tdir | sort -u | wc -l)
local v=$(ls -ai $DIR/$tdir | sort -u | wc -l)
(( $t == $NFILES && $u == $NFILES && $v == NFILES + 2 )) ||
error "Expected $NFILES files, got $t ($u unique $v .&..)"
}
run_test 24A "readdir() returns correct number of entries."
test_24B() { # LU-4805
[ $MDSCOUNT -lt 2 ] && skip_env "needs >= 2 MDTs"
local count
test_mkdir $DIR/$tdir
$LFS setdirstripe -i0 -c$MDSCOUNT $DIR/$tdir/striped_dir/ ||
error "create striped dir failed"
count=$(ls -ai $DIR/$tdir/striped_dir | wc -l)
[ $count -eq 2 ] || error "Expected 2, got $count"
touch $DIR/$tdir/striped_dir/a
count=$(ls -ai $DIR/$tdir/striped_dir | wc -l)
[ $count -eq 3 ] || error "Expected 3, got $count"
touch $DIR/$tdir/striped_dir/.f
count=$(ls -ai $DIR/$tdir/striped_dir | wc -l)
[ $count -eq 4 ] || error "Expected 4, got $count"
rm -rf $DIR/$tdir || error "Can not delete directories"
}
run_test 24B "readdir for striped dir return correct number of entries"
test_24C() {
[ $MDSCOUNT -lt 2 ] && skip_env "needs >= 2 MDTs"
mkdir $DIR/$tdir
mkdir $DIR/$tdir/d0
mkdir $DIR/$tdir/d1
$LFS setdirstripe -i0 -c$MDSCOUNT $DIR/$tdir/d0/striped_dir ||
error "create striped dir failed"
cd $DIR/$tdir/d0/striped_dir
local d0_ino=$(ls -i -l -a $DIR/$tdir | grep "d0" | awk '{print $1}')
local d1_ino=$(ls -i -l -a $DIR/$tdir | grep "d1" | awk '{print $1}')
local parent_ino=$(ls -i -l -a | grep "\.\." | awk '{print $1}')
[ "$d0_ino" = "$parent_ino" ] ||
error ".. wrong, expect $d0_ino, get $parent_ino"
mv $DIR/$tdir/d0/striped_dir $DIR/$tdir/d1/ ||
error "mv striped dir failed"
parent_ino=$(ls -i -l -a | grep "\.\." | awk '{print $1}')
[ "$d1_ino" = "$parent_ino" ] ||
error ".. wrong after mv, expect $d1_ino, get $parent_ino"
}
run_test 24C "check .. in striped dir"
test_24E() {
[[ $MDSCOUNT -lt 4 ]] && skip_env "needs >= 4 MDTs"
[ $PARALLEL == "yes" ] && skip "skip parallel run"
mkdir -p $DIR/$tdir
mkdir $DIR/$tdir/src_dir
$LFS mkdir -i 1 $DIR/$tdir/src_dir/src_child ||
error "create remote source failed"
touch $DIR/$tdir/src_dir/src_child/a
$LFS mkdir -i 2 $DIR/$tdir/tgt_dir ||
error "create remote target dir failed"
$LFS mkdir -i 3 $DIR/$tdir/tgt_dir/tgt_child ||
error "create remote target child failed"
mrename $DIR/$tdir/src_dir/src_child $DIR/$tdir/tgt_dir/tgt_child ||
error "rename dir cross MDT failed!"
find $DIR/$tdir
$CHECKSTAT -t dir $DIR/$tdir/src_dir/src_child &&
error "src_child still exists after rename"
$CHECKSTAT -t file $DIR/$tdir/tgt_dir/tgt_child/a ||
error "missing file(a) after rename"
rm -rf $DIR/$tdir || error "Can not delete directories"
}
run_test 24E "cross MDT rename/link"
test_24F () {
[ $MDSCOUNT -lt 2 ] && skip "needs >= 2 MDTs" && return 0
local repeats=1000
[ "$SLOW" = "no" ] && repeats=100
mkdir -p $DIR/$tdir
echo "$repeats repeats"
for ((i = 0; i < repeats; i++)); do
$LFS mkdir -i0 -c2 $DIR/$tdir/test || error "mkdir fails"
touch $DIR/$tdir/test/a || error "touch fails"
mkdir $DIR/$tdir/test/b || error "mkdir fails"
rm -rf $DIR/$tdir/test || error "rmdir fails"
done
true
}
run_test 24F "hash order vs readdir (LU-11330)"
test_24G () {
[ $MDSCOUNT -lt 2 ] && skip "needs >= 2 MDTs"
(( $MDS1_VERSION >= $(version_code v2_13_56-59-g9adc02c016d2) )) ||
skip "Need MDS >= v2_13_56-59-g9adc02c016d2 for symlink migration"
local ino1
local ino2
$LFS mkdir -i 0 $DIR/$tdir-0 || error "mkdir $tdir-0"
$LFS mkdir -i 1 $DIR/$tdir-1 || error "mkdir $tdir-1"
touch $DIR/$tdir-0/f1 || error "touch f1"
ln -s $DIR/$tdir-0/f1 $DIR/$tdir-0/s1 || error "ln s1"
ino1=$(stat -c%i $DIR/$tdir-0/s1)
mv $DIR/$tdir-0/s1 $DIR/$tdir-1 || error "mv s1"
ino2=$(stat -c%i $DIR/$tdir-1/s1)
[ $ino1 -ne $ino2 ] || error "s1 should be migrated"
}
run_test 24G "migrate symlink in rename"
test_24H() {
[ $MDSCOUNT -lt 2 ] && skip "needs >= 2 MDTs"
[[ $(hostname) != $(facet_active_host mds2) ]] ||
skip "MDT1 should be on another node"
test_mkdir -i 1 -c 1 $DIR/$tdir
#define OBD_FAIL_FLD_QUERY_REQ 0x1103
do_facet mds2 $LCTL set_param fail_loc=0x80001103
touch $DIR/$tdir/$tfile || error "touch failed"
}
run_test 24H "repeat FLD_QUERY rpc"
test_24I() {
(( $MDSCOUNT > 1 )) || skip "needs >= 2 MDTs"
# LMV_MAX_STRIPE_COUNT = 2000 should be verified
$LFS mkdir -C 2001 $DIR/$tdir-2001 && error "mkdir 2001-stripe worked"
$LFS mkdir -C 2000 $DIR/$tdir-2000 || error "mkdir 2000-stripe failed"
local stripes=$($LFS getdirstripe -c $DIR/$tdir-2000)
# LMV_OVERSTRIPE_COUNT_MAX = 5 limits actual stripe count
(( $stripes >= $MDSCOUNT * 5)) ||
error "mkdir stripes $stripes < $MDSCOUNT * 5"
}
run_test 24I "large striped mkdir with limit check"
test_25a() {
echo '== symlink sanity ============================================='
test_mkdir $DIR/d25
ln -s d25 $DIR/s25
touch $DIR/s25/foo ||
error "File creation in symlinked directory failed"
}
run_test 25a "create file in symlinked directory ==============="
test_25b() {
[ ! -d $DIR/d25 ] && test_25a
$CHECKSTAT -t file $DIR/s25/foo || error "$DIR/s25/foo not file type"
}
run_test 25b "lookup file in symlinked directory ==============="
test_26a() {
test_mkdir $DIR/d26
test_mkdir $DIR/d26/d26-2
ln -s d26/d26-2 $DIR/s26
touch $DIR/s26/foo || error "File creation failed"
}
run_test 26a "multiple component symlink ======================="
test_26b() {
test_mkdir -p $DIR/$tdir/d26-2
ln -s $tdir/d26-2/foo $DIR/s26-2
touch $DIR/s26-2 || error "File creation failed"
}
run_test 26b "multiple component symlink at end of lookup ======"
test_26c() {
test_mkdir $DIR/d26.2
touch $DIR/d26.2/foo
ln -s d26.2 $DIR/s26.2-1
ln -s s26.2-1 $DIR/s26.2-2
ln -s s26.2-2 $DIR/s26.2-3
chmod 0666 $DIR/s26.2-3/foo
}
run_test 26c "chain of symlinks"
# recursive symlinks (bug 439)
test_26d() {
ln -s d26-3/foo $DIR/d26-3
}
run_test 26d "create multiple component recursive symlink"
test_26e() {
[ ! -h $DIR/d26-3 ] && test_26d
rm $DIR/d26-3
}
run_test 26e "unlink multiple component recursive symlink"
# recursive symlinks (bug 7022)
test_26f() {
test_mkdir $DIR/$tdir
test_mkdir $DIR/$tdir/$tfile
cd $DIR/$tdir/$tfile || error "cd $DIR/$tdir/$tfile failed"
test_mkdir -p lndir/bar1
test_mkdir $DIR/$tdir/$tfile/$tfile
cd $tfile || error "cd $tfile failed"
ln -s .. dotdot || error "ln dotdot failed"
ln -s dotdot/lndir lndir || error "ln lndir failed"
cd $DIR/$tdir || error "cd $DIR/$tdir failed"
output=`ls $tfile/$tfile/lndir/bar1`
[ "$output" = bar1 ] && error "unexpected output"
rm -r $tfile || error "rm $tfile failed"
$CHECKSTAT -a $DIR/$tfile || error "$tfile not gone"
}
run_test 26f "rm -r of a directory which has recursive symlink"
test_27a() {
test_mkdir $DIR/$tdir
$LFS getstripe $DIR/$tdir
$LFS setstripe -c 1 $DIR/$tdir/$tfile || error "setstripe failed"
$CHECKSTAT -t file $DIR/$tdir/$tfile || error "checkstat failed"
cp /etc/hosts $DIR/$tdir/$tfile || error "Can't copy to one stripe file"
}
run_test 27a "one stripe file"
test_27b() {
[[ $OSTCOUNT -lt 2 ]] && skip_env "needs >= 2 OSTs"
test_mkdir $DIR/$tdir
$LFS setstripe -c 2 $DIR/$tdir/$tfile || error "setstripe failed"
$LFS getstripe -c $DIR/$tdir/$tfile
[ $($LFS getstripe -c $DIR/$tdir/$tfile) -eq 2 ] ||
error "two-stripe file doesn't have two stripes"
dd if=/dev/zero of=$DIR/$tdir/$tfile bs=4k count=4 || error "dd failed"
}
run_test 27b "create and write to two stripe file"
# 27c family tests specific striping, setstripe -o
test_27ca() {
[[ $OSTCOUNT -lt 2 ]] && skip_env "skipping 2-stripe test"
test_mkdir -p $DIR/$tdir
local osts="1"
$LFS setstripe -o $osts $DIR/$tdir/$tfile || error "setstripe failed"
$LFS getstripe -i $DIR/$tdir/$tfile
[ $($LFS getstripe -i $DIR/$tdir/$tfile ) -eq $osts ] ||
error "stripe not on specified OST"
dd if=/dev/zero of=$DIR/$tdir/$tfile bs=1M count=4 || error "dd failed"
}
run_test 27ca "one stripe on specified OST"
test_27cb() {
[[ $OSTCOUNT -lt 2 ]] && skip_env "skipping 2-stripe test"
test_mkdir -p $DIR/$tdir
local osts="1,0"
$LFS setstripe -o $osts $DIR/$tdir/$tfile || error "setstripe failed"
local getstripe=$($LFS getstripe $DIR/$tdir/$tfile)
echo "$getstripe"
# Strip getstripe output to a space separated list of OSTs
local getstripe_osts=$(echo "$getstripe" | sed -e '1,/obdidx/d' |\
awk '{print $1}' | tr '\n' ' ' | sed -e 's/[[:space:]]*$//')
[ "$getstripe_osts" = "${osts//,/ }" ] ||
error "stripes not on specified OSTs"
dd if=/dev/zero of=$DIR/$tdir/$tfile bs=1M count=4 || error "dd failed"
}
run_test 27cb "two stripes on specified OSTs"
test_27cc() {
[[ $OSTCOUNT -lt 2 ]] && skip_env "skipping 2-stripe test"
[[ $($LCTL get_param mdc.*.import) =~ connect_flags.*overstriping ]] ||
skip "server does not support overstriping"
test_mkdir -p $DIR/$tdir
local osts="0,0"
$LFS setstripe -o $osts $DIR/$tdir/$tfile || error "setstripe failed"
local getstripe=$($LFS getstripe $DIR/$tdir/$tfile)
echo "$getstripe"
# Strip getstripe output to a space separated list of OSTs
local getstripe_osts=$(echo "$getstripe" | sed -e '1,/obdidx/d' |\
awk '{print $1}' | tr '\n' ' ' | sed -e 's/[[:space:]]*$//')
[ "$getstripe_osts" = "${osts//,/ }" ] ||
error "stripes not on specified OSTs"
dd if=/dev/zero of=$DIR/$tdir/$tfile bs=1M count=4 || error "dd failed"
}
run_test 27cc "two stripes on the same OST"
test_27cd() {
[[ $OSTCOUNT -lt 2 ]] && skip_env "skipping 2-stripe test"
[[ $($LCTL get_param mdc.*.import) =~ connect_flags.*overstriping ]] ||
skip "server does not support overstriping"
test_mkdir -p $DIR/$tdir
local osts="0,1,1,0"
$LFS setstripe -o $osts $DIR/$tdir/$tfile || error "setstripe failed"
local getstripe=$($LFS getstripe $DIR/$tdir/$tfile)
echo "$getstripe"
# Strip getstripe output to a space separated list of OSTs
local getstripe_osts=$(echo "$getstripe" | sed -e '1,/obdidx/d' |\
awk '{print $1}' | tr '\n' ' ' | sed -e 's/[[:space:]]*$//')
[ "$getstripe_osts" = "${osts//,/ }" ] ||
error "stripes not on specified OSTs"
dd if=/dev/zero of=$DIR/$tdir/$tfile bs=1M count=4 || error "dd failed"
}
run_test 27cd "four stripes on two OSTs"
test_27ce() {
[[ $OSTCOUNT -ge $(($LOV_MAX_STRIPE_COUNT / 2)) ]] &&
skip_env "too many osts, skipping"
[[ $($LCTL get_param mdc.*.import) =~ connect_flags.*overstriping ]] ||
skip "server does not support overstriping"
# We do one more stripe than we have OSTs
[ $OSTCOUNT -lt 159 ] || large_xattr_enabled ||
skip_env "ea_inode feature disabled"
test_mkdir -p $DIR/$tdir
local osts=""
for i in $(seq 0 $OSTCOUNT);
do
osts=$osts"0"
if [ $i -ne $OSTCOUNT ]; then
osts=$osts","
fi
done
echo "$LFS setstripe -o $osts $DIR/$tdir/$tfile"
$LFS setstripe -o $osts $DIR/$tdir/$tfile || error "setstripe failed"
local getstripe=$($LFS getstripe $DIR/$tdir/$tfile)
echo "$getstripe"
# Strip getstripe output to a space separated list of OSTs
local getstripe_osts=$(echo "$getstripe" | sed -e '1,/obdidx/d' |\
awk '{print $1}' | tr '\n' ' ' | sed -e 's/[[:space:]]*$//')
[ "$getstripe_osts" = "${osts//,/ }" ] ||
error "stripes '$getstripe_osts' not on given OSTs ${osts//,/ }"
dd if=/dev/zero of=$DIR/$tdir/$tfile bs=1M count=4 || error "dd failed"
}
run_test 27ce "more stripes than OSTs with -o"
test_27cf() {
local osp_proc="osp.$FSNAME-OST0000-osc-MDT000*.active"
local pid=0
test_mkdir -p $DIR/$tdir || error "failed to mkdir $DIR/$tdir"
do_facet $SINGLEMDS "$LCTL set_param -n $osp_proc=0"
stack_trap "do_facet $SINGLEMDS $LCTL set_param -n $osp_proc=1" EXIT
wait_update_facet $SINGLEMDS "$LCTL get_param -n $osp_proc | grep 1" ||
error "failed to set $osp_proc=0"
$LFS setstripe -o 0 $DIR/$tdir/$tfile &
pid=$!
sleep 1
do_facet $SINGLEMDS "$LCTL set_param -n $osp_proc=1"
wait_update_facet $SINGLEMDS "$LCTL get_param -n $osp_proc | grep 0" ||
error "failed to set $osp_proc=1"
wait $pid
[[ $pid -ne 0 ]] ||
error "should return error due to $osp_proc=0"
}
run_test 27cf "'setstripe -o' on inactive OSTs should return error"
test_27cg() {
[[ $($LCTL get_param mdc.*.import) =~ connect_flags.*overstriping ]] ||
skip "server does not support overstriping"
[[ $mds1_FSTYPE != "ldiskfs" ]] && skip_env "ldiskfs only test"
large_xattr_enabled || skip_env "ea_inode feature disabled"
local osts="0"
for ((i=1;i<1000;i++)); do
osts+=",$((i % OSTCOUNT))"
done
local mdts=$(mdts_nodes)
local before=$(do_nodes $mdts \
"$LCTL get_param -n osd-ldiskfs.*MDT*.stats" |
awk '/many credits/{print $3}' |
calc_sum)
$LFS setstripe -o $osts $DIR/$tfile || {
$LFS df
$LFS df -i
error "setstripe failed"
}
$LFS getstripe $DIR/$tfile | grep stripe
rm -f $DIR/$tfile || error "can't unlink"
# if running in a loop, don't run out of inodes on these OSTs
(( $ONLY_REPEAT_ITER % 20 < 19 )) || {
$LFS df
$LFS df -i
wait_delete_completed
sleep 10
}
(( $ONLY_REPEAT_ITER < 100 )) || skip "too many iterations"
after=$(do_nodes $mdts \
"$LCTL get_param -n osd-ldiskfs.*MDT*.stats" |
awk '/many credits/{print $3}' |
calc_sum)
(( before == after )) ||
error "too many credits happened: $after > $before"
}
run_test 27cg "1000 shouldn't cause too many credits"
test_27d() {
test_mkdir $DIR/$tdir
$LFS setstripe -c 0 -i -1 -S 0 $DIR/$tdir/$tfile ||
error "setstripe failed"
$CHECKSTAT -t file $DIR/$tdir/$tfile || error "checkstat failed"
dd if=/dev/zero of=$DIR/$tdir/$tfile bs=4k count=4 || error "dd failed"
}
run_test 27d "create file with default settings"
test_27e() {
# LU-5839 adds check for existed layout before setting it
[[ $MDS1_VERSION -lt $(version_code 2.7.56) ]] &&
skip "Need MDS version at least 2.7.56"
test_mkdir $DIR/$tdir
$LFS setstripe -c 2 $DIR/$tdir/$tfile || error "setstripe failed"
$LFS setstripe -c 2 $DIR/$tdir/$tfile && error "setstripe worked twice"
$CHECKSTAT -t file $DIR/$tdir/$tfile || error "checkstat failed"
}
run_test 27e "setstripe existing file (should return error)"
test_27f() {
test_mkdir $DIR/$tdir
$LFS setstripe -S 100 -i 0 -c 1 $DIR/$tdir/$tfile &&
error "$LFS setstripe $DIR/$tdir/$tfile failed"
$CHECKSTAT -t file $DIR/$tdir/$tfile &&
error "$CHECKSTAT -t file $DIR/$tdir/$tfile should fail"
dd if=/dev/zero of=$DIR/$tdir/$tfile bs=4k count=4 || error "dd failed"
$LFS getstripe $DIR/$tdir/$tfile || error "$LFS getstripe failed"
}
run_test 27f "setstripe with bad stripe size (should return error)"
test_27g() {
test_mkdir $DIR/$tdir
$MCREATE $DIR/$tdir/$tfile || error "mcreate failed"
$LFS getstripe $DIR/$tdir/$tfile 2>&1 | grep "no stripe info" ||
error "$DIR/$tdir/$tfile has object"
}
run_test 27g "$LFS getstripe with no objects"
test_27ga() {
test_mkdir $DIR/$tdir
touch $DIR/$tdir/$tfile || error "touch failed"
ln -s bogus $DIR/$tdir/$tfile.2 || error "ln failed"
$LFS getstripe -m $DIR/$tdir/$tfile $DIR/$tdir/$tfile.2
local rc=$?
(( rc == 2 )) || error "getstripe did not return ENOENT"
local err_msg=$($LFS getstripe $DIR/$tdir/typo $DIR/$tdir/$tfile \
2>&1 > /dev/null)
[[ $err_msg =~ "typo" ]] ||
error "expected message with correct filename, got '$err_msg'"
}
run_test 27ga "$LFS getstripe with missing file (should return error)"
test_27i() {
test_mkdir $DIR/$tdir
touch $DIR/$tdir/$tfile || error "touch failed"
[[ $($LFS getstripe -c $DIR/$tdir/$tfile) -gt 0 ]] ||
error "missing objects"
}
run_test 27i "$LFS getstripe with some objects"
test_27j() {
test_mkdir $DIR/$tdir
$LFS setstripe -i $OSTCOUNT $DIR/$tdir/$tfile &&
error "setstripe failed" || true
}
run_test 27j "setstripe with bad stripe offset (should return error)"
test_27k() { # bug 2844
test_mkdir $DIR/$tdir
local file=$DIR/$tdir/$tfile
local ll_max_blksize=$((4 * 1024 * 1024))
$LFS setstripe -S 67108864 $file || error "setstripe failed"
local blksize=$(stat $file | awk '/IO Block:/ { print $7 }')
[ $blksize -le $ll_max_blksize ] || error "1:$blksize > $ll_max_blksize"
dd if=/dev/zero of=$file bs=4k count=1
blksize=$(stat $file | awk '/IO Block:/ { print $7 }')
[ $blksize -le $ll_max_blksize ] || error "2:$blksize > $ll_max_blksize"
}
run_test 27k "limit i_blksize for broken user apps"
test_27l() {
mcreate $DIR/$tfile || error "creating file"
$RUNAS $LFS setstripe -c 1 $DIR/$tfile &&
error "setstripe should have failed" || true
}
run_test 27l "check setstripe permissions (should return error)"
test_27m() {
[[ $OSTCOUNT -lt 2 ]] && skip_env "needs >= 2 OSTs"
[ -n "$RCLIENTS" -o -n "$MOUNT_2" ] &&
skip_env "multiple clients -- skipping"
ORIGFREE=$($LCTL get_param -n lov.$FSNAME-clilov-*.kbytesavail |
head -n1)
if [[ $ORIGFREE -gt $MAXFREE ]]; then
skip "$ORIGFREE > $MAXFREE skipping out-of-space test on OST0"
fi
stack_trap simple_cleanup_common
test_mkdir $DIR/$tdir
$LFS setstripe -i 0 -c 1 $DIR/$tdir/$tfile.1
fast_file_write $DIR/$tdir/$tfile.1 1024K $MAXFREE &&
error "write should fill OST0"
i=2
while $LFS setstripe -i 0 -c 1 $DIR/$tdir/$tfile.$i; do
i=$((i + 1))
[ $i -gt 256 ] && break
done
i=$((i + 1))
touch $DIR/$tdir/$tfile.$i
[ $($LFS getstripe $DIR/$tdir/$tfile.$i | grep -A 10 obdidx |
awk '{print $1}'| grep -w "0") ] &&
error "OST0 was full but new created file still use it"
i=$((i + 1))
touch $DIR/$tdir/$tfile.$i
[ $($LFS getstripe $DIR/$tdir/$tfile.$i | grep -A 10 obdidx |
awk '{print $1}'| grep -w "0") ] &&
error "OST0 was full but new created file still use it" || true
}
run_test 27m "create file while OST0 was full"
# OSCs keep a NOSPC flag that will be reset after ~5s (qos_maxage)
# if the OST isn't full anymore.
reset_enospc() {
local ostidx=${1:-""}
local delay
local ready
local get_prealloc
local osts
[[ -n "$ostidx" ]] && osts=$(facet_host ost$((ostidx + 1))) ||
osts=$(osts_nodes)
do_nodes $osts lctl set_param fail_loc=0
wait_delete_completed # initiate all OST_DESTROYs from MDS to OST
delay=$(do_facet $SINGLEMDS lctl get_param -n lov.*.qos_maxage |
awk '{print $1 * 2;exit;}')
get_prealloc="$LCTL get_param -n osc.*MDT*.prealloc_status |
grep -v \"^0$\""
wait_update_facet $SINGLEMDS "$get_prealloc" "" $delay
}
test_27n() {
[[ $OSTCOUNT -lt 2 ]] && skip_env "needs >= 2 OSTs"
[ $PARALLEL == "yes" ] && skip "skip parallel run"
remote_mds_nodsh && skip "remote MDS with nodsh"
remote_ost_nodsh && skip "remote OST with nodsh"
reset_enospc
rm -f $DIR/$tdir/$tfile
exhaust_precreations 0 0x80000215
$LFS setstripe -c -1 $DIR/$tdir || error "setstripe failed"
touch $DIR/$tdir/$tfile || error "touch failed"
$LFS getstripe $DIR/$tdir/$tfile
reset_enospc
}
run_test 27n "create file with some full OSTs"
test_27o() {
[[ $OSTCOUNT -lt 2 ]] && skip_env "needs >= 2 OSTs"
[ $PARALLEL == "yes" ] && skip "skip parallel run"
remote_mds_nodsh && skip "remote MDS with nodsh"
remote_ost_nodsh && skip "remote OST with nodsh"
reset_enospc
rm -f $DIR/$tdir/$tfile
exhaust_all_precreations 0x215
touch $DIR/$tdir/$tfile && error "able to create $DIR/$tdir/$tfile"
reset_enospc
rm -rf $DIR/$tdir/*
}
run_test 27o "create file with all full OSTs (should error)"
function create_and_checktime() {
local fname=$1
local loops=$2
local i
for ((i=0; i < $loops; i++)); do
local start=$SECONDS
multiop $fname-$i Oc
((SECONDS-start < TIMEOUT)) ||
error "creation took " $((SECONDS-$start)) && return 1
done
}
test_27oo() {
local mdts=$(mdts_nodes)
(( $MDS1_VERSION >= $(version_code 2.13.57) )) ||
skip "Need MDS version at least 2.13.57"
local f0=$DIR/${tfile}-0
local f1=$DIR/${tfile}-1
wait_delete_completed
# refill precreated objects
$LFS setstripe -i0 -c1 $f0
saved=$(do_facet mds1 $LCTL get_param -n lov.*0000*.qos_threshold_rr)
# force QoS allocation policy
do_nodes $mdts $LCTL set_param lov.*.qos_threshold_rr=0%
stack_trap "do_nodes $mdts $LCTL set_param \
lov.*.qos_threshold_rr=$saved" EXIT
sleep_maxage
# one OST is unavailable, but still have few objects preallocated
stop ost1
stack_trap "start ost1 $(ostdevname 1) $OST_MOUNT_OPTS; \
rm -rf $f1 $DIR/$tdir*" EXIT
for ((i=0; i < 7; i++)); do
mkdir $DIR/$tdir$i || error "can't create dir"
$LFS setstripe -c$((OSTCOUNT-1)) $DIR/$tdir$i ||
error "can't set striping"
done
for ((i=0; i < 7; i++)); do
create_and_checktime $DIR/$tdir$i/$tfile 100 &
done
wait
}
run_test 27oo "don't let few threads to reserve too many objects"
test_27p() {
[[ $OSTCOUNT -lt 2 ]] && skip_env "needs >= 2 OSTs"
[ $PARALLEL == "yes" ] && skip "skip parallel run"
remote_mds_nodsh && skip "remote MDS with nodsh"
remote_ost_nodsh && skip "remote OST with nodsh"
reset_enospc
rm -f $DIR/$tdir/$tfile
test_mkdir $DIR/$tdir || error "failed to mkdir $DIR/$tdir"
$MCREATE $DIR/$tdir/$tfile || error "mcreate failed"
$TRUNCATE $DIR/$tdir/$tfile 80000000 || error "truncate failed"
$CHECKSTAT -s 80000000 $DIR/$tdir/$tfile || error "checkstat failed"
exhaust_precreations 0 0x80000215
echo foo >> $DIR/$tdir/$tfile || error "append failed"
$CHECKSTAT -s 80000004 $DIR/$tdir/$tfile || error "checkstat failed"
$LFS getstripe $DIR/$tdir/$tfile
reset_enospc
}
run_test 27p "append to a truncated file with some full OSTs"
test_27q() {
[[ $OSTCOUNT -lt 2 ]] && skip_env "needs >= 2 OSTs"
[ $PARALLEL == "yes" ] && skip "skip parallel run"
remote_mds_nodsh && skip "remote MDS with nodsh"
remote_ost_nodsh && skip "remote OST with nodsh"
reset_enospc
rm -f $DIR/$tdir/$tfile
mkdir_on_mdt0 $DIR/$tdir
$MCREATE $DIR/$tdir/$tfile || error "mcreate $DIR/$tdir/$tfile failed"
$TRUNCATE $DIR/$tdir/$tfile 80000000 ||
error "truncate $DIR/$tdir/$tfile failed"
$CHECKSTAT -s 80000000 $DIR/$tdir/$tfile || error "checkstat failed"
exhaust_all_precreations 0x215
echo foo >> $DIR/$tdir/$tfile && error "append succeeded"
$CHECKSTAT -s 80000000 $DIR/$tdir/$tfile || error "checkstat 2 failed"
reset_enospc
}
run_test 27q "append to truncated file with all OSTs full (should error)"
test_27r() {
[[ $OSTCOUNT -lt 2 ]] && skip_env "needs >= 2 OSTs"
[ $PARALLEL == "yes" ] && skip "skip parallel run"
remote_mds_nodsh && skip "remote MDS with nodsh"
remote_ost_nodsh && skip "remote OST with nodsh"
reset_enospc
rm -f $DIR/$tdir/$tfile
exhaust_precreations 0 0x80000215
$LFS setstripe -i 0 -c 2 $DIR/$tdir/$tfile || error "setstripe failed"
reset_enospc
}
run_test 27r "stripe file with some full OSTs (shouldn't LBUG) ="
test_27s() { # bug 10725
test_mkdir $DIR/$tdir
local stripe_size=$((4096 * 1024 * 1024)) # 2^32
local stripe_count=0
[ $OSTCOUNT -eq 1 ] || stripe_count=2
$LFS setstripe -S $stripe_size -c $stripe_count $DIR/$tdir &&
error "stripe width >= 2^32 succeeded" || true
}
run_test 27s "lsm_xfersize overflow (should error) (bug 10725)"
test_27t() { # bug 10864
WDIR=$(pwd)
WLFS=$(which lfs)
cd $DIR
touch $tfile
$WLFS getstripe $tfile
cd $WDIR
}
run_test 27t "check that utils parse path correctly"
test_27u() { # bug 4900
[[ $OSTCOUNT -lt 2 ]] && skip_env "needs >= 2 OSTs"
remote_mds_nodsh && skip "remote MDS with nodsh"
local index
local mdts=$(mdts_nodes)
#define OBD_FAIL_MDS_OSC_PRECREATE 0x139
do_nodes $mdts "$LCTL set_param fail_loc=0x139"
test_mkdir -p $DIR/$tdir
stack_trap "simple_cleanup_common 1000"
createmany -o $DIR/$tdir/$tfile 1000
do_nodes $mdts "$LCTL set_param fail_loc=0"
TLOG=$TMP/$tfile.getstripe
$LFS getstripe $DIR/$tdir > $TLOG
OBJS=$(awk -vobj=0 '($1 == 0) { obj += 1 } END { print obj; }' $TLOG)
[[ $OBJS -gt 0 ]] &&
error "$OBJS objects created on OST-0. See $TLOG" ||
rm -f $TLOG
}
run_test 27u "skip object creation on OSC w/o objects"
test_27v() { # bug 4900
[[ $OSTCOUNT -lt 2 ]] && skip_env "needs >= 2 OSTs"
[ $PARALLEL == "yes" ] && skip "skip parallel run"
remote_mds_nodsh && skip "remote MDS with nodsh"
remote_ost_nodsh && skip "remote OST with nodsh"
exhaust_all_precreations 0x215
reset_enospc
$LFS setstripe -c 1 $DIR/$tdir # 1 stripe / file
touch $DIR/$tdir/$tfile
#define OBD_FAIL_TGT_DELAY_PRECREATE 0x705
# all except ost1
for (( i=1; i < OSTCOUNT; i++ )); do
do_facet ost$i lctl set_param fail_loc=0x705
done
local START=`date +%s`
createmany -o $DIR/$tdir/$tfile 32
local FINISH=`date +%s`
local TIMEOUT=`lctl get_param -n timeout`
local PROCESS=$((FINISH - START))
[ $PROCESS -ge $((TIMEOUT / 2)) ] && \
error "$FINISH - $START >= $TIMEOUT / 2"
sleep $((TIMEOUT / 2 - PROCESS))
reset_enospc
}
run_test 27v "skip object creation on slow OST"
test_27w() { # bug 10997
test_mkdir $DIR/$tdir
$LFS setstripe -S 65536 $DIR/$tdir/f0 || error "setstripe failed"
[ $($LFS getstripe -S $DIR/$tdir/f0) -ne 65536 ] &&
error "stripe size $size != 65536" || true
[ $($LFS getstripe -d $DIR/$tdir | grep -c "stripe_count") -eq 0 ] &&
error "$LFS getstripe -d $DIR/$tdir no 'stripe_count'" || true
}
run_test 27w "check $LFS setstripe -S and getstrip -d options"
test_27wa() {
[[ $OSTCOUNT -lt 2 ]] &&
skip_env "skipping multiple stripe count/offset test"
test_mkdir $DIR/$tdir
for i in $(seq 1 $OSTCOUNT); do
offset=$((i - 1))
$LFS setstripe -c $i -i $offset $DIR/$tdir/f$i ||
error "setstripe -c $i -i $offset failed"
count=$($LFS getstripe -c $DIR/$tdir/f$i)
index=$($LFS getstripe -i $DIR/$tdir/f$i)
[ $count -ne $i ] && error "stripe count $count != $i" || true
[ $index -ne $offset ] &&
error "stripe offset $index != $offset" || true
done
}
run_test 27wa "check $LFS setstripe -c -i options"
test_27x() {
remote_ost_nodsh && skip "remote OST with nodsh"
[[ $OSTCOUNT -lt 2 ]] && skip_env "needs >= 2 OSTs"
[ $PARALLEL == "yes" ] && skip "skip parallel run"
OFFSET=$(($OSTCOUNT - 1))
OSTIDX=0
local OST=$(ostname_from_index $OSTIDX)
test_mkdir $DIR/$tdir
$LFS setstripe -c 1 $DIR/$tdir # 1 stripe per file
do_facet ost$((OSTIDX + 1)) lctl set_param -n obdfilter.$OST.degraded 1
sleep_maxage
createmany -o $DIR/$tdir/$tfile $OSTCOUNT
for i in $(seq 0 $OFFSET); do
[ $($LFS getstripe $DIR/$tdir/$tfile$i | grep -A 10 obdidx |
awk '{print $1}' | grep -w "$OSTIDX") ] &&
error "OST0 was degraded but new created file still use it"
done
do_facet ost$((OSTIDX + 1)) lctl set_param -n obdfilter.$OST.degraded 0
}
run_test 27x "create files while OST0 is degraded"
test_27y() {
[[ $OSTCOUNT -lt 2 ]] && skip_env "needs >= 2 OSTs"
remote_mds_nodsh && skip "remote MDS with nodsh"
remote_ost_nodsh && skip "remote OST with nodsh"
[ $PARALLEL == "yes" ] && skip "skip parallel run"
local mdtosc=$(get_mdtosc_proc_path $SINGLEMDS $FSNAME-OST0000)
local last_id=$(do_facet $SINGLEMDS lctl get_param -n \
osp.$mdtosc.prealloc_last_id)
local next_id=$(do_facet $SINGLEMDS lctl get_param -n \
osp.$mdtosc.prealloc_next_id)
local fcount=$((last_id - next_id))
[[ $fcount -eq 0 ]] && skip "not enough space on OST0"
[[ $fcount -gt $OSTCOUNT ]] && fcount=$OSTCOUNT
local MDS_OSCS=$(do_facet $SINGLEMDS lctl dl |
awk '/[oO][sS][cC].*md[ts]/ { print $4 }')
local OST_DEACTIVE_IDX=-1
local OSC
local OSTIDX
local OST
for OSC in $MDS_OSCS; do
OST=$(osc_to_ost $OSC)
OSTIDX=$(index_from_ostuuid $OST)
if [ $OST_DEACTIVE_IDX == -1 ]; then
OST_DEACTIVE_IDX=$OSTIDX
fi
if [ $OSTIDX != $OST_DEACTIVE_IDX ]; then
echo $OSC "is Deactivated:"
do_facet $SINGLEMDS lctl --device %$OSC deactivate
fi
done
OSTIDX=$(index_from_ostuuid $OST)
test_mkdir $DIR/$tdir
$LFS setstripe -c 1 $DIR/$tdir # 1 stripe / file
for OSC in $MDS_OSCS; do
OST=$(osc_to_ost $OSC)
OSTIDX=$(index_from_ostuuid $OST)
if [ $OSTIDX == $OST_DEACTIVE_IDX ]; then
echo $OST "is degraded:"
do_facet ost$((OSTIDX+1)) lctl set_param -n \
obdfilter.$OST.degraded=1
fi
done
sleep_maxage
createmany -o $DIR/$tdir/$tfile $fcount
for OSC in $MDS_OSCS; do
OST=$(osc_to_ost $OSC)
OSTIDX=$(index_from_ostuuid $OST)
if [ $OSTIDX == $OST_DEACTIVE_IDX ]; then
echo $OST "is recovered from degraded:"
do_facet ost$((OSTIDX+1)) lctl set_param -n \
obdfilter.$OST.degraded=0
else
do_facet $SINGLEMDS lctl --device %$OSC activate
fi
done
# all osp devices get activated, hence -1 stripe count restored
local stripe_count=0
# sleep 2*lod_qos_maxage seconds waiting for lod qos to notice osp
# devices get activated.
sleep_maxage
$LFS setstripe -c -1 $DIR/$tfile
stripe_count=$($LFS getstripe -c $DIR/$tfile)
rm -f $DIR/$tfile
[ $stripe_count -ne $OSTCOUNT ] &&
error "Of $OSTCOUNT OSTs, only $stripe_count is available"
return 0
}
run_test 27y "create files while OST0 is degraded and the rest inactive"
test_27z() {
[ $PARALLEL == "yes" ] && skip "skip parallel run"
remote_ost_nodsh && skip "remote OST with nodsh"
test_mkdir $DIR/$tdir
$LFS setstripe -c 1 -i 0 -S 64k $DIR/$tdir/$tfile-1 ||
{ error "setstripe -c -1 failed"; return 1; }
# We need to send a write to every object to get parent FID info set.
# This _should_ also work for setattr, but does not currently.
# touch $DIR/$tdir/$tfile-1 ||
dd if=/dev/zero of=$DIR/$tdir/$tfile-1 bs=1M count=1 ||
{ error "dd $tfile-1 failed"; return 2; }
$LFS setstripe -c -1 -i $((OSTCOUNT - 1)) -S 1M $DIR/$tdir/$tfile-2 ||
{ error "setstripe -c -1 failed"; return 3; }
dd if=/dev/zero of=$DIR/$tdir/$tfile-2 bs=1M count=$OSTCOUNT ||
{ error "dd $tfile-2 failed"; return 4; }
# make sure write RPCs have been sent to OSTs
sync; sleep 5; sync
check_seq_oid $DIR/$tdir/$tfile-1 || return 5
check_seq_oid $DIR/$tdir/$tfile-2 || return 6
}
run_test 27z "check SEQ/OID on the MDT and OST filesystems"
test_27A() { # b=19102
[ $PARALLEL == "yes" ] && skip "skip parallel run"
save_layout_restore_at_exit $MOUNT
$LFS setstripe -c 0 -i -1 -S 0 $MOUNT
wait_update $HOSTNAME "$LFS getstripe -c $MOUNT | sed 's/ *//g'" "1" 20 ||
error "stripe count $($LFS getstripe -c $MOUNT) != 1"
local default_size=$($LFS getstripe -S $MOUNT)
local default_offset=$($LFS getstripe -i $MOUNT)
local dsize=$(do_facet $SINGLEMDS \
"$LCTL get_param -n lod.$(facet_svc $SINGLEMDS)*.stripesize")
[ $default_size -eq $dsize ] ||
error "stripe size $default_size != $dsize"
[ $default_offset -eq -1 ] ||
error "stripe offset $default_offset != -1"
}
run_test 27A "check filesystem-wide default LOV EA values"
test_27B() { # LU-2523
test_mkdir $DIR/$tdir
rm -f $DIR/$tdir/f0 $DIR/$tdir/f1
touch $DIR/$tdir/f0
# open f1 with O_LOV_DELAY_CREATE
# rename f0 onto f1
# call setstripe ioctl on open file descriptor for f1
# close
multiop $DIR/$tdir/f1 oO_RDWR:O_CREAT:O_LOV_DELAY_CREATE:nB1c \
$DIR/$tdir/f0
rm -f $DIR/$tdir/f1
# open f1 with O_LOV_DELAY_CREATE
# unlink f1
# call setstripe ioctl on open file descriptor for f1
# close
multiop $DIR/$tdir/f1 oO_RDWR:O_CREAT:O_LOV_DELAY_CREATE:uB1c
# Allow multiop to fail in imitation of NFS's busted semantics.
true
}
run_test 27B "call setstripe on open unlinked file/rename victim"
# 27C family tests full striping and overstriping
test_27Ca() { #LU-2871
[[ $OSTCOUNT -lt 2 ]] && skip_env "needs >= 2 OSTs"
declare -a ost_idx
local index
local found
local i
local j
test_mkdir $DIR/$tdir
cd $DIR/$tdir
for i in $(seq 0 $((OSTCOUNT - 1))); do
# set stripe across all OSTs starting from OST$i
$LFS setstripe -i $i -c -1 $tfile$i
# get striping information
ost_idx=($($LFS getstripe $tfile$i |
tail -n $((OSTCOUNT + 1)) | awk '{print $1}'))
echo "OST Index: ${ost_idx[*]}"
# check the layout
[ ${#ost_idx[@]} -eq $OSTCOUNT ] ||
error "${#ost_idx[@]} != $OSTCOUNT"
for index in $(seq 0 $((OSTCOUNT - 1))); do
found=0
for j in "${ost_idx[@]}"; do
if [ $index -eq $j ]; then
found=1
break
fi
done
[ $found = 1 ] ||
error "Can not find $index in ${ost_idx[*]}"
done
done
}
run_test 27Ca "check full striping across all OSTs"
test_27Cb() {
[[ $($LCTL get_param mdc.*.import) =~ connect_flags.*overstriping ]] ||
skip "server does not support overstriping"
[[ $OSTCOUNT -ge $(($LOV_MAX_STRIPE_COUNT / 2)) ]] &&
skip_env "too many osts, skipping"
test_mkdir -p $DIR/$tdir
local setcount=$(($OSTCOUNT * 2))
[ $setcount -lt 160 ] || large_xattr_enabled ||
skip_env "ea_inode feature disabled"
$LFS setstripe -C $setcount $DIR/$tdir/$tfile ||
error "setstripe failed"
local count=$($LFS getstripe -c $DIR/$tdir/$tfile)
[ $count -eq $setcount ] ||
error "stripe count $count, should be $setcount"
$LFS getstripe $DIR/$tdir/$tfile 2>&1 | grep "overstriped" ||
error "overstriped should be set in pattern"
dd if=/dev/zero of=$DIR/$tdir/$tfile bs=1M count=4 conv=notrunc ||
error "dd failed"
}
run_test 27Cb "more stripes than OSTs with -C"
test_27Cc() {
[[ $($LCTL get_param mdc.*.import) =~ connect_flags.*overstriping ]] ||
skip "server does not support overstriping"
[[ $OSTCOUNT -lt 2 ]] && skip_env "need > 1 OST"
test_mkdir -p $DIR/$tdir
local setcount=$(($OSTCOUNT - 1))
[ $setcount -lt 160 ] || large_xattr_enabled ||
skip_env "ea_inode feature disabled"
$LFS setstripe -C $setcount $DIR/$tdir/$tfile ||
error "setstripe failed"
local count=$($LFS getstripe -c $DIR/$tdir/$tfile)
[ $count -eq $setcount ] ||
error "stripe count $count, should be $setcount"
$LFS getstripe $DIR/$tdir/$tfile 2>&1 | grep "overstriped" &&
error "overstriped should not be set in pattern"
dd if=/dev/zero of=$DIR/$tdir/$tfile bs=1M count=4 conv=notrunc ||
error "dd failed"
}
run_test 27Cc "fewer stripes than OSTs does not set overstriping"
test_27Cd() {
[[ $($LCTL get_param mdc.*.import) =~ connect_flags.*overstriping ]] ||
skip "server does not support overstriping"
[[ $OSTCOUNT -lt 2 ]] && skip_env "need > 1 OST"
large_xattr_enabled || skip_env "ea_inode feature disabled"
force_new_seq_all
test_mkdir -p $DIR/$tdir
local setcount=$LOV_MAX_STRIPE_COUNT
$LFS setstripe -C $setcount $DIR/$tdir/$tfile ||
error "setstripe failed"
local count=$($LFS getstripe -c $DIR/$tdir/$tfile)
[ $count -eq $setcount ] ||
error "stripe count $count, should be $setcount"
$LFS getstripe $DIR/$tdir/$tfile 2>&1 | grep "overstriped" ||
error "overstriped should be set in pattern"
dd if=/dev/zero of=$DIR/$tdir/$tfile bs=1M count=4 conv=notrunc ||
error "dd failed"
rm -f $DIR/$tdir/$tfile || error "Delete $tfile failed"
}
run_test 27Cd "test maximum stripe count"
test_27Ce() {
[[ $($LCTL get_param mdc.*.import) =~ connect_flags.*overstriping ]] ||
skip "server does not support overstriping"
test_mkdir -p $DIR/$tdir
pool_add $TESTNAME || error "Pool creation failed"
pool_add_targets $TESTNAME 0 || error "pool_add_targets failed"
local setcount=8
$LFS setstripe -C $setcount -p "$TESTNAME" $DIR/$tdir/$tfile ||
error "setstripe failed"
local count=$($LFS getstripe -c $DIR/$tdir/$tfile)
[ $count -eq $setcount ] ||
error "stripe count $count, should be $setcount"
$LFS getstripe $DIR/$tdir/$tfile 2>&1 | grep "overstriped" ||
error "overstriped should be set in pattern"
dd if=/dev/zero of=$DIR/$tdir/$tfile bs=1M count=4 conv=notrunc ||
error "dd failed"
rm -f $DIR/$tdir/$tfile || error "Delete $tfile failed"
}
run_test 27Ce "test pool with overstriping"
test_27Cf() {
[[ $($LCTL get_param mdc.*.import) =~ connect_flags.*overstriping ]] ||
skip "server does not support overstriping"
[[ $OSTCOUNT -ge $(($LOV_MAX_STRIPE_COUNT / 2)) ]] &&
skip_env "too many osts, skipping"
test_mkdir -p $DIR/$tdir
local setcount=$(($OSTCOUNT * 2))
[ $setcount -lt 160 ] || large_xattr_enabled ||
skip_env "ea_inode feature disabled"
$LFS setstripe -C $setcount $DIR/$tdir/ ||
error "setstripe failed"
echo 1 > $DIR/$tdir/$tfile
local count=$($LFS getstripe -c $DIR/$tdir/$tfile)
[ $count -eq $setcount ] ||
error "stripe count $count, should be $setcount"
$LFS getstripe $DIR/$tdir/$tfile 2>&1 | grep "overstriped" ||
error "overstriped should be set in pattern"
dd if=/dev/zero of=$DIR/$tdir/$tfile bs=1M count=4 conv=notrunc ||
error "dd failed"
rm -f $DIR/$tdir/$tfile || error "Delete $tfile failed"
}
run_test 27Cf "test default inheritance with overstriping"
test_27Cg() {
(( MDS1_VERSION >= $(version_code v2_15_55-80-gd96b98ee6b) )) ||
skip "need MDS version at least v2_15_55-80-gd96b98ee6b for fix"
$LFS setstripe -o 0,$OSTCOUNT $DIR/$tfile
(( $? != 0 )) || error "must be an error for not existent OST#"
}
run_test 27Cg "test setstripe with wrong OST idx"
test_27Ci() {
[[ $($LCTL get_param mdc.*.import) =~ connect_flags.*overstriping ]] ||
skip "server does not support overstriping"
local tf=$DIR/$tfile
stack_trap "rm -f $DIR/$tfile"
$LFS setstripe -E1M $tf || error "create $tf failed"
$LFS setstripe -Eeof --component-add -C 100 $tf ||
error "add component failed"
$LFS getstripe -I2 $tf | awk '/lmm_pattern/ { print $2 }' |
grep "overstriped" || {
$LFS getstripe $tf
echo "lose overstriping setting"
}
sc=$($LFS getstripe -I2 --stripe-count $tf)
(( $sc == 100 )) || {
$LFS getstripe $tf
echo "lose overstriping setting"
}
stack_trap "rm -f $tf"
dd if=/dev/zero of=$tf bs=1M count=10 || error "write $tf"
sc=$($LFS getstripe -I2 --stripe-count $tf)
(( $sc == 100 )) || {
$LFS getstripe $tf
echo "lose overstriping setting after instantiation"
}
}
run_test 27Ci "add an overstriping component"
test_27Cj() {
[[ $($LCTL get_param mdc.*.import) =~ connect_flags.*overstriping ]] ||
skip "server does not support overstriping"
(( $MDS1_VERSION >= $(version_code 2.15.61.76) )) ||
skip "need MDS >= 2.15.61.76"
stack_trap "rm -f $DIR/$tfile"
#check that beyond -32 it fails
echo "setstripe -C -33 should fail"
$LFS setstripe -C -33 $DIR/$tfile && error "setstripe should fail"
# start_full_debug_logging
# number of stripes created will be 32*ost count
$LFS setstripe -C -32 $DIR/$tfile || error "create $tfile failed"
local count=$($LFS getstripe -c $DIR/$tfile)
local stripe_cnt=$(($OSTCOUNT * 32))
(( $count == $stripe_cnt )) ||
error "$DIR/$tfile stripe count $count != $stripe_cnt"
}
run_test 27Cj "overstriping with -C for max values in multiple of targets"
test_27D() {
(( $OSTCOUNT >= 2 )) || skip_env "needs >= 2 OSTs"
remote_mds_nodsh && skip "remote MDS with nodsh"
local POOL=${POOL:-testpool}
local first_ost=0
local last_ost=$(($OSTCOUNT - 1))
local ost_step=1
local ost_list=$(seq $first_ost $ost_step $last_ost)
local ost_range="$first_ost $last_ost $ost_step"
local flr_ec_ver=v2_17_52-42-gf1b7ae0cc6
test_mkdir $DIR/$tdir
pool_add $POOL || error "pool_add failed"
pool_add_targets $POOL $ost_range || error "pool_add_targets failed"
(( $MDS1_VERSION >= $(version_code 2.8.55) )) ||
SKIP27D+=" -s 29"
(( $MDS1_VERSION >= $(version_code 2.9.55) &&
$CLIENT_VERSION >= $(version_code 2.9.55) )) ||
SKIP27D+=" -s 30,31"
[[ $($LCTL get_param mdc.*.import) =~ connect_flags.*overstriping ]] &&
(( $OSTCOUNT < $LOV_MAX_STRIPE_COUNT / 2)) ||
SKIP27D+=" -s 32,33"
(( $MDS1_VERSION >= $(version_code $SEL_VER) )) ||
SKIP27D+=" -s 34"
(( $MDS1_VERSION >= $(version_code $flr_ec_ver) )) ||
SKIP27D+=" -s 52,53,54,56,57,58,63,64,65,67"
local ec_enable=$($LCTL get_param -n llite.*.enable_erasure_coding)
$LCTL set_param llite.*.enable_erasure_coding=1
stack_trap "$LCTL set_param -n \
llite.*.enable_erasure_coding=$ec_enable"
llapi_layout_test -d$DIR/$tdir -p$POOL -o$OSTCOUNT $SKIP27D ||
error "llapi_layout_test failed"
destroy_test_pools || error "destroy test pools failed"
}
run_test 27D "validate llapi_layout API"
# Verify that default_easize is increased from its initial value after
# accessing a widely striped file.
test_27E() {
[ $OSTCOUNT -lt 2 ] && skip_env "needs >= 2 OSTs"
[ $CLIENT_VERSION -lt $(version_code 2.5.57) ] &&
skip "client does not have LU-3338 fix"
# 72 bytes is the minimum space required to store striping
# information for a file striped across one OST:
# (sizeof(struct lov_user_md_v3) +
# sizeof(struct lov_user_ost_data_v1))
local min_easize=72
$LCTL set_param -n llite.*.default_easize $min_easize ||
error "lctl set_param failed"
local easize=$($LCTL get_param -n llite.*.default_easize)
[ $easize -eq $min_easize ] ||
error "failed to set default_easize"
$LFS setstripe -c $OSTCOUNT $DIR/$tfile ||
error "setstripe failed"
# In order to ensure stat() call actually talks to MDS we need to
# do something drastic to this file to shake off all lock, e.g.
# rename it (kills lookup lock forcing cache cleaning)
mv $DIR/$tfile $DIR/${tfile}-1
ls -l $DIR/${tfile}-1
rm $DIR/${tfile}-1
easize=$($LCTL get_param -n llite.*.default_easize)
[ $easize -gt $min_easize ] ||
error "default_easize not updated"
}
run_test 27E "check that default extended attribute size properly increases"
test_27F() { # LU-5346/LU-7975
[ $PARALLEL == "yes" ] && skip "skip parallel run"
[[ $OSTCOUNT -lt 2 ]] && skip "needs >= 2 OSTs"
[[ $MDS1_VERSION -lt $(version_code 2.8.51) ]] &&
skip "Need MDS version at least 2.8.51"
remote_ost_nodsh && skip "remote OST with nodsh"
test_mkdir $DIR/$tdir
rm -f $DIR/$tdir/f0
$LFS setstripe -c 2 $DIR/$tdir
# stop all OSTs to reproduce situation for LU-7975 ticket
for num in $(seq $OSTCOUNT); do
stop ost$num
done
# open/create f0 with O_LOV_DELAY_CREATE
# truncate f0 to a non-0 size
# close
multiop $DIR/$tdir/f0 oO_RDWR:O_CREAT:O_LOV_DELAY_CREATE:T1050000c
$CHECKSTAT -s 1050000 $DIR/$tdir/f0 || error "checkstat failed"
# open/write it again to force delayed layout creation
cat /etc/hosts > $DIR/$tdir/f0 &
catpid=$!
# restart OSTs
for num in $(seq $OSTCOUNT); do
start ost$num $(ostdevname $num) $OST_MOUNT_OPTS ||
error "ost$num failed to start"
done
wait $catpid || error "cat failed"
cmp /etc/hosts $DIR/$tdir/f0 || error "cmp failed"
[[ $($LFS getstripe -c $DIR/$tdir/f0) == 2 ]] ||
error "wrong stripecount"
}
run_test 27F "Client resend delayed layout creation with non-zero size"
test_27G() { #LU-10629
(( $MDS1_VERSION < $(version_code 2.11.51) )) &&
skip "Need MDS version at least 2.11.51"
remote_mds_nodsh && skip "remote MDS with nodsh"
local POOL=${POOL:-testpool}
local ostrange="0 0 1"
test_mkdir $DIR/$tdir
touch $DIR/$tdir/$tfile.nopool
pool_add $POOL || error "pool_add failed"
pool_add_targets $POOL $ostrange || error "pool_add_targets failed"
$LFS setstripe -p $POOL $DIR/$tdir
local pool=$($LFS getstripe -p $DIR/$tdir)
[ "$pool" = "$POOL" ] || error "Striping failed got '$pool' not '$POOL'"
touch $DIR/$tdir/$tfile.default
$LFS setstripe -E 1M --pool $POOL -c 1 -E eof -c 1 $DIR/$tdir/$tfile.pfl
$LFS find $DIR/$tdir -type f --pool $POOL
local found=$($LFS find $DIR/$tdir -type f --pool $POOL | wc -l)
[[ "$found" == "2" ]] ||
error "found $found != 2 files in '$DIR/$tdir' in '$POOL'"
$LFS setstripe -d $DIR/$tdir
pool=$($LFS getstripe -p -d $DIR/$tdir)
[[ "$pool" != "$POOL" ]] || error "$DIR/$tdir is still '$pool'"
}
run_test 27G "Clear OST pool from stripe"
test_27H() {
[[ $MDS1_VERSION -le $(version_code 2.11.54) ]] &&
skip "Need MDS version newer than 2.11.54"
[[ $OSTCOUNT -lt 3 ]] && skip_env "needs >= 3 OSTs"
test_mkdir $DIR/$tdir
$LFS setstripe -o 0 -o 2 $DIR/$tdir || error "setstripe failed"
touch $DIR/$tdir/$tfile
$LFS getstripe -c $DIR/$tdir/$tfile
[ $($LFS getstripe -c $DIR/$tdir/$tfile) -eq 2 ] ||
error "two-stripe file doesn't have two stripes"
dd if=/dev/zero of=$DIR/$tdir/$tfile bs=4k count=4 || error "dd failed"
$LFS getstripe -y $DIR/$tdir/$tfile
(( $($LFS getstripe -y $DIR/$tdir/$tfile |
grep -E -c "l_ost_idx: [02]$") == "2" )) ||
error "expected l_ost_idx: [02]$ not matched"
# make sure ost list has been cleared
local stripesize=$($LFS getstripe -S $DIR/$tdir)
$LFS setstripe -S $((stripesize * 4)) -i 1 \
-c $((OSTCOUNT - 1)) $DIR/$tdir || error "setstripe"
touch $DIR/$tdir/f3
$LVERIFY $DIR/$tdir $DIR/$tdir/f3 || error "lverify failed"
}
run_test 27H "Set specific OSTs stripe"
test_27I() {
[ $PARALLEL == "yes" ] && skip "skip parallel run"
[[ $OSTCOUNT -lt 2 ]] && skip_env "needs >= 2 OSTs"
[[ $MDS1_VERSION -gt $(version_code 2.12.52) ]] ||
skip "Need MDS version newer than 2.12.52"
local pool=$TESTNAME
local ostrange="1 1 1"
save_layout_restore_at_exit $MOUNT
$LFS setstripe -c 2 -i 0 $MOUNT
pool_add $pool || error "pool_add failed"
pool_add_targets $pool $ostrange ||
error "pool_add_targets failed"
test_mkdir $DIR/$tdir
$LFS setstripe -p $pool $DIR/$tdir
$MULTIOP $DIR/$tdir/$tfile Oc || error "multiop failed"
$LFS getstripe $DIR/$tdir/$tfile
}
run_test 27I "check that root dir striping does not break parent dir one"
test_27Ia() {
(( $MDS1_VERSION >= $(version_code 2.15.61.225) )) ||
skip "need MDS >= 2.15.61.255 for pool inheritance fix"
(( $OSTCOUNT >= 2 )) || skip_env "needs >= 2 OSTs"
save_layout_restore_at_exit $MOUNT
pool_add $TESTNAME || error "pool_add failed"
pool_add_targets $TESTNAME 1 || error "pool_add_targets failed"
$LFS setstripe -p $TESTNAME $MOUNT
test_mkdir $DIR/$tdir
$LFS setstripe -i0 $DIR/$tdir
$MULTIOP $DIR/$tdir/$tfile.1 Oc || error_noexit "multiop failed"
$LFS getstripe $DIR/$tdir/$tfile.1
$LFS setstripe -d $DIR/$tdir
$LFS setstripe -o 0,1 $DIR/$tdir
$MULTIOP $DIR/$tdir/$tfile.2 Oc || error_noexit "multiop failed"
$LFS getstripe $DIR/$tdir/$tfile.2
test_mkdir $DIR/$tdir/d1
$MULTIOP $DIR/$tdir/d1/$tfile.3 Oc || error_noexit "multiop failed"
$LFS getstripe $DIR/$tdir/d1/$tfile.3
$LFS setstripe -d $DIR/$tdir
$LFS setstripe -E 128G -o 0,1 -E-1 $DIR/$tdir
test_mkdir $DIR/$tdir/d2
$MULTIOP $DIR/$tdir/d2/$tfile.4 Oc || error_noexit "multiop failed"
$LFS getstripe $DIR/$tdir/d2/$tfile.4
}
run_test 27Ia "check that root dir pool is dropped with conflict parent dir settings"
test_27J() {
(( $MDS1_VERSION >= $(version_code v2_12_53-14-g6a20bdcc60) )) ||
skip "Need MDS >= 2.12.53.15 for foreign layouts"
test_mkdir $DIR/$tdir
local uuid1=$(cat /proc/sys/kernel/random/uuid)
local uuid2=$(cat /proc/sys/kernel/random/uuid)
# create foreign file (raw way)
! $LFS setstripe --flags 0xda08 $DIR/$tdir/$tfile ||
error "creating $tfile w/ hex flags w/o --foreign should fail"
! $LFS setstripe --foreign --flags foo \
--xattr ${uuid1}@${uuid2} $DIR/$tdir/$tfile ||
error "creating $tfile with '--flags foo' should fail"
! $LFS setstripe --foreign --flags 0xffffffff \
--xattr ${uuid1}@${uuid2} $DIR/$tdir/$tfile ||
error "creating $tfile w/ 0xffffffff flags should fail"
create_foreign_file -f $DIR/$tdir/$tfile -x "${uuid1}@${uuid2}" \
-t 1 -F 0xda08 || error "create_foreign_file failed"
# verify foreign file (raw way)
parse_foreign_file -f $DIR/$tdir/$tfile |
grep "lov_foreign_magic: 0x0BD70BD0" ||
error "$DIR/$tdir/$tfile: invalid LOV EA foreign magic"
parse_foreign_file -f $DIR/$tdir/$tfile | grep "lov_xattr_size: 89" ||
error "$DIR/$tdir/$tfile: invalid LOV EA foreign size"
parse_foreign_file -f $DIR/$tdir/$tfile |
grep "lov_foreign_size: 73" ||
error "$DIR/$tdir/$tfile: invalid LOV EA foreign size"
parse_foreign_file -f $DIR/$tdir/$tfile |
grep "lov_foreign_type:.*1$" ||
error "$DIR/$tdir/$tfile: invalid LOV EA foreign type"
parse_foreign_file -f $DIR/$tdir/$tfile |
grep "lov_foreign_flags: 0x0000DA08" ||
error "$DIR/$tdir/$tfile: invalid LOV EA foreign flags"
local lov=$(parse_foreign_file -f $DIR/$tdir/$tfile |
grep "lov_foreign_value: 0x" |
sed -e 's/lov_foreign_value: 0x//')
local lov2=$(echo -n "${uuid1}@${uuid2}" | od -A n -t x1 -w160)
[[ $lov = ${lov2// /} ]] ||
error "$DIR/$tdir/$tfile: invalid LOV EA foreign value"
# create foreign file (lfs + API)
$LFS setstripe --foreign=none --flags 0xda08 \
-x "${uuid1}@${uuid2}" $DIR/$tdir/${tfile}2 ||
error "$DIR/$tdir/${tfile}2: create failed"
$LFS getstripe -v $DIR/$tdir/${tfile}2 |
grep "lfm_magic:.*0x0BD70BD0" ||
error "$DIR/$tdir/${tfile}2: invalid LOV EA foreign magic"
# lfm_length is LOV EA size - sizeof(lfm_magic) - sizeof(lfm_length)
$LFS getstripe -v $DIR/$tdir/${tfile}2 | grep "lfm_length:.*73" ||
error "$DIR/$tdir/${tfile}2: invalid LOV EA foreign size"
$LFS getstripe -v $DIR/$tdir/${tfile}2 | grep "lfm_type:.*none" ||
error "$DIR/$tdir/${tfile}2: invalid LOV EA foreign type"
$LFS getstripe -v $DIR/$tdir/${tfile}2 |
grep "lfm_flags:.*0x0000DA08" ||
error "$DIR/$tdir/${tfile}2: invalid LOV EA foreign flags"
$LFS getstripe $DIR/$tdir/${tfile}2 |
grep "lfm_value:.*${uuid1}@${uuid2}" ||
error "$DIR/$tdir/${tfile}2: invalid LOV EA foreign value"
# modify striping should fail
$LFS setstripe -c 2 $DIR/$tdir/$tfile &&
error "$DIR/$tdir/$tfile: setstripe should fail"
$LFS setstripe -c 2 $DIR/$tdir/${tfile}2 &&
error "$DIR/$tdir/${tfile}2: setstripe should fail"
# skip basic ops on file with foreign LOV tests on >5.12 kernels
(( $LINUX_VERSION_CODE < $(version_code 5.12.0) )) && {
# kernel >= 5.12.0 would skip filemap_read() with 0 sized file
# R/W should fail
cat $DIR/$tdir/$tfile &&
error "$DIR/$tdir/$tfile: read should fail"
cat $DIR/$tdir/${tfile}2 &&
error "$DIR/$tdir/${tfile}2: read should fail"
cat /etc/passwd > $DIR/$tdir/$tfile &&
error "$DIR/$tdir/$tfile: write should fail"
cat /etc/passwd > $DIR/$tdir/${tfile}2 &&
error "$DIR/$tdir/${tfile}2: write should fail"
}
# chmod should work
chmod 222 $DIR/$tdir/$tfile ||
error "$DIR/$tdir/$tfile: chmod failed"
chmod 222 $DIR/$tdir/${tfile}2 ||
error "$DIR/$tdir/${tfile}2: chmod failed"
# chown should work
chown $RUNAS_ID:$RUNAS_GID $DIR/$tdir/$tfile ||
error "$DIR/$tdir/$tfile: chown failed"
chown $RUNAS_ID:$RUNAS_GID $DIR/$tdir/${tfile}2 ||
error "$DIR/$tdir/${tfile}2: chown failed"
# rename should work
mv $DIR/$tdir/$tfile $DIR/$tdir/${tfile}.new ||
error "$DIR/$tdir/$tfile: rename of foreign file has failed"
mv $DIR/$tdir/${tfile}2 $DIR/$tdir/${tfile}2.new ||
error "$DIR/$tdir/${tfile}2: rename of foreign file has failed"
#remove foreign file
rm $DIR/$tdir/${tfile}.new ||
error "$DIR/$tdir/${tfile}.new: remove foreign file has failed"
rm $DIR/$tdir/${tfile}2.new ||
error "$DIR/$tdir/${tfile}2.new: remove foreign file has failed"
}
run_test 27J "basic ops on file with foreign LOV"
test_27K() {
(( $MDS1_VERSION >= $(version_code v2_12_53-15-gfdad38781c) )) ||
skip "Need MDS >= 2.12.53.15 for foreign LVM directory"
test_mkdir $DIR/$tdir
local uuid1=$(cat /proc/sys/kernel/random/uuid)
local uuid2=$(cat /proc/sys/kernel/random/uuid)
local foreign_magic="lmv_foreign_magic:.*cd50cd0"
# create foreign dir (raw way)
! $LFS setdirstripe --flags 0xda08 $DIR/$tdir/$tdir ||
error "creating $tdir w/ hex flags w/o --foreign should fail"
! $LFS setdirstripe --foreign --flags foo \
--xattr ${uuid1}@${uuid2} $DIR/$tdir/$tdir ||
error "creating $tdir with '--flags foo' should fail"
! $LFS setdirstripe --foreign --flags 0xffffffff \
--xattr ${uuid1}@${uuid2} $DIR/$tdir/$tdir ||
error "creating $tdir w/ 0xffffffff flags should fail"
create_foreign_dir -d $DIR/$tdir/$tdir -x "${uuid1}@${uuid2}" -t 1 ||
error "create_foreign_dir FAILED"
# verify foreign dir (raw way)
parse_foreign_dir -d $DIR/$tdir/$tdir | grep "$foreign_magic" ||
error "$DIR/$tdir/$tfile: invalid LMV EA magic"
parse_foreign_dir -d $DIR/$tdir/$tdir | grep "lmv_xattr_size:.*89$" ||
error "$DIR/$tdir/$tdir: invalid LMV EA size"
parse_foreign_dir -d $DIR/$tdir/$tdir | grep "lmv_foreign_type:.*1$" ||
error "$DIR/$tdir/$tdir: invalid LMV EA type"
parse_foreign_dir -d $DIR/$tdir/$tdir |
grep -E "lmv_foreign_flags: (55813|0xda05)$" ||
error "$DIR/$tdir/$tdir: invalid LMV EA flags"
local lmv=$(parse_foreign_dir -d $DIR/$tdir/$tdir |
grep "lmv_foreign_value: 0x" |
sed 's/lmv_foreign_value: 0x//')
local lmv2=$(echo -n "${uuid1}@${uuid2}" | od -A n -t x1 -w160 |
sed 's/ //g')
[[ $lmv == $lmv2 ]] || error "$DIR/$tdir/$tdir: invalid LMV EA value"
# create foreign dir (lfs + API)
$LFS mkdir --foreign=none --xattr="${uuid1}@${uuid2}" --flags=0xda05 \
$DIR/$tdir/${tdir}2 ||
error "$DIR/$tdir/${tdir}2: create failed"
$LFS getdirstripe -v $DIR/$tdir/${tdir}2
$LFS getdirstripe -v $DIR/$tdir/${tdir}2 |
grep "lfm_magic:.*0x0CD50CD0" ||
error "$DIR/$tdir/${tdir}2: invalid LMV EA magic"
# lfm_length is LMV EA size - sizeof(lfm_magic) - sizeof(lfm_length)
# - sizeof(lfm_type) - sizeof(lfm_flags)
$LFS getdirstripe -v $DIR/$tdir/${tdir}2 | grep "lfm_length:.*73" ||
error "$DIR/$tdir/${tdir}2: invalid LMV EA size"
$LFS getdirstripe -v $DIR/$tdir/${tdir}2 | grep "lfm_type:.*none" ||
error "$DIR/$tdir/${tdir}2: invalid LMV EA type"
$LFS getdirstripe -v $DIR/$tdir/${tdir}2 |
grep "lfm_flags:.*0x0000DA05" ||
error "$DIR/$tdir/${tdir}2: invalid LMV EA flags"
$LFS getdirstripe $DIR/$tdir/${tdir}2 |
grep "lfm_value.*${uuid1}@${uuid2}" ||
error "$DIR/$tdir/${tdir}2: invalid LMV EA value"
# file create in dir should fail
touch $DIR/$tdir/$tdir/$tfile &&
error "$DIR/$tdir: file create should fail"
touch $DIR/$tdir/${tdir}2/$tfile &&
error "$DIR/${tdir}2: file create should fail"
# chmod should work
chmod 777 $DIR/$tdir/$tdir ||
error "$DIR/$tdir: chmod failed"
chmod 777 $DIR/$tdir/${tdir}2 ||
error "$DIR/${tdir}2: chmod failed"
# chown should work
chown $RUNAS_ID:$RUNAS_GID $DIR/$tdir/$tdir ||
error "$DIR/$tdir: chown failed"
chown $RUNAS_ID:$RUNAS_GID $DIR/$tdir/${tdir}2 ||
error "$DIR/${tdir}2: chown failed"
# rename should work
mv $DIR/$tdir/$tdir $DIR/$tdir/${tdir}.new ||
error "$DIR/$tdir/$tdir: rename of foreign dir has failed"
mv $DIR/$tdir/${tdir}2 $DIR/$tdir/${tdir}2.new ||
error "$DIR/$tdir/${tdir}2: rename of foreign dir has failed"
#remove foreign dir
rmdir $DIR/$tdir/${tdir}.new ||
error "$DIR/$tdir/${tdir}.new: remove of foreign dir failed"
rmdir $DIR/$tdir/${tdir}2.new ||
error "$DIR/$tdir/${tdir}2.new: remove of foreign dir failed"
}
run_test 27K "basic ops on dir with foreign LMV"
test_27Ke() {
(( $MDS1_VERSION >= $(version_code v2_17_52-77-gaec30763f9) )) ||
skip "need MDS >= 2.17.52.77 for enable_foreign_dir"
local param=mdt.$FSNAME-*.enable_foreign_dir
local param_gid=${param}_gid
local foreign_old=($(do_facet mds1 $LCTL get_param -n $param))
local foreign_gid_old=($(do_facet mds1 $LCTL get_param -n $param_gid))
local foreign_magic="lmv_foreign_magic:.*cd50cd0"
local uuid1=$(sysctl -n kernel.random.uuid)
local testdir=$DIR/$tdir
local mdts=$(mdts_nodes)
mkdir_on_mdt0 -o 0777 $testdir
stack_trap "do_nodes $mdts \
$LCTL set_param $param=$foreign_old $param_gid=$foreign_gid_old"
start_full_debug_logging
stack_trap stop_full_debug_logging
# test with enable_foreign_dir=0: it should fail for all users
do_nodes $mdts "$LCTL set_param $param=0" || error "setting $param=0"
create_foreign_dir -d $testdir/$tdir-0 -x $uuid1 -t 1 &&
error "$tdir-0 foreign_dir worked as root with $param=0"
[[ ! -e $testdir/$tdir-0 ]] || error "$tdir-0 was created for root"
$RUNAS create_foreign_dir -d $testdir/$tdir-0u -x $uuid1 -t 1 &&
error "$tdir-0u foreign_dir worked as user with $param=0"
[[ ! -e $testdir/$tdir-0u ]] || error "$tdir-0u was created for user"
# test with enable_foreign_dir=1: _gid=0, OK for root, not others
do_nodes $mdts "$LCTL set_param $param=1 $param_gid=0" ||
error "setting $param=1, $param_gid=0"
create_foreign_dir -d $testdir/$tdir-gid0 -x $uuid1 -t 1 ||
error "$tdir-gid0 foreign_dir failed for root with _gid=0"
parse_foreign_dir -d $testdir/$tdir-gid0 | grep "$foreign_magic" ||
error "$tdir-gid0: invalid LMV EA magic"
$RUNAS create_foreign_dir -d $testdir/$tdir-gid0u -x $uuid1 -t 1 &&
error "$tdir-gid0u foreign_dir worked for user GID=$RUNAS_GID"
[[ ! -e $testdir/$tdir-gid0u ]] ||
error "$tdir-gid0u created for non-root user"
# test with enable_foreign_dir=1: _gid=$RUNAS_GID, OK for root,RUNAS_GID
do_nodes $mdts "$LCTL set_param $param_gid=$RUNAS_GID" ||
error "setting $param_gid=$RUNAS_GID"
create_foreign_dir -d $testdir/$tdir-gidRG -x $uuid1 -t 1 ||
error "$tdir-gidRG foreign_dir failed for root GID=$RUNAS_GID"
parse_foreign_dir -d $testdir/$tdir-gidRG | grep "$foreign_magic" ||
error "$tdir-gidRG: invalid LMV EA magic"
$RUNAS create_foreign_dir -d $testdir/$tdir-gidRGu -x $uuid1 -t 1 ||
error "$tdir-gidRGu foreign_dir failed for user GID=$RUNAS_GID"
parse_foreign_dir -d $testdir/$tdir-gidRGu | grep "$foreign_magic" ||
error "$tdir-gidRGu: invalid LMV EA magic"
# test with wrong primary GID (no secondary groups) should fail
local runas="$RUNAS -u $TSTUSR2 -g $TSTUSR2"
$runas create_foreign_dir -d $testdir/$tdir-gidRGn -x $uuid1 -t 1 &&
error "$tdir-gidRGn foreign_dir worked with GID!=$RUNAS_GID"
[[ ! -e $testdir/$tdir-gidRGn ]] ||
error "$tdir-gidRGn created for non-root user GID!=$RUNAS_GID"
# test with wrong primary, but correct secondary GID should work
local tst_gids=$(do_facet mds1 id -G $TSTUSR | tr ' ' ',')
local new_gids="$tst_gids,$RUNAS_ID"
local runas="$RUNAS -u $TSTUSR -g $TSTUSR -G$new_gids"
stack_trap "do_nodes $mdts usermod -G $tst_gids $TSTUSR"
do_nodes $mdts usermod -a -G $RUNAS_ID $TSTUSR ||
error "can't add $TSTUSR to group $RUNAS_ID"
echo "using $TSTUSR with secondary groups $new_gids"
do_nodes $mdts "grep ':$RUNAS_ID:' /etc/group"
do_nodes $mdts "id $TSTUSR"
do_nodes $mdts "$L_GETIDENTITY -d $(id -u $TSTUSR)"
$runas create_foreign_dir -d $testdir/$tdir-gidRGs -x $uuid1 -t 1 ||
error "$tdir-gidRGs foreign_dir failed with secondary GID"
parse_foreign_dir -d $testdir/$tdir-gidRGs | grep "$foreign_magic" ||
error "$tdir-gidRGs: invalid LMV EA magic"
# test with enable_foreign_dir=1: _gid=-1, OK for everyone
do_nodes $mdts "$LCTL set_param $param_gid=-1" ||
error "setting $param_gid=-1"
create_foreign_dir -d $testdir/$tdir-gid-1 -x $uuid1 -t 1 ||
error "$tdir-gid-1 foreign_dir failed as root with _gid=-1"
parse_foreign_dir -d $testdir/$tdir-gid-1 | grep "$foreign_magic" ||
error "$tdir-gid-1: invalid LMV EA magic"
$RUNAS create_foreign_dir -d $testdir/$tdir-gid-1u -x $uuid1 -t 1 ||
error "$tdir-gid-1u foreign_dir failed as user with _gid=-1"
parse_foreign_dir -d $testdir/$tdir-gid-1u | grep "$foreign_magic"||
error "$tdir-gid-1u: invalid LMV EA magic"
}
run_test 27Ke "test enable_foreign_dir and enable_foreign_dir_gid"
test_27L() {
remote_mds_nodsh && skip "remote MDS with nodsh"
local POOL=${POOL:-$TESTNAME}
pool_add $POOL || error "pool_add failed"
lfs pool_list $MOUNT | grep -Fx "${FSNAME}.${POOL}" ||
error "pool_list does not contain ${FSNAME}.${POOL}:" \
"$(lfs pool_list $MOUNT | grep -F "${POOL}")"
}
run_test 27L "lfs pool_list gives correct pool name"
test_27M() {
(( $MDS1_VERSION >= $(version_code 2.12.57) )) ||
skip "Need MDS version >= than 2.12.57"
remote_mds_nodsh && skip "remote MDS with nodsh"
(( $OSTCOUNT > 1 )) || skip "need > 1 OST"
# Set default striping on directory
local setcount=4
local stripe_opt
local mdts=$(mdts_nodes)
# if we run against a 2.12 server which lacks overstring support
# then the connect_flag will not report overstriping, even if client
# is 2.14+
if [[ $($LCTL get_param mdc.*.connect_flags) =~ overstriping ]]; then
stripe_opt="-C $setcount"
elif (( $OSTCOUNT >= $setcount )); then
stripe_opt="-c $setcount"
else
skip "server does not support overstriping"
fi
test_mkdir $DIR/$tdir
# Validate existing append_* params and ensure restore
local pool=$(do_facet mds1 $LCTL get_param -n mdd.$FSNAME-MDT0000.append_pool)
[[ "$pool" == "" ]] || error "expected append_pool == '', got '$pool'"
stack_trap "do_nodes $mdts $LCTL set_param mdd.*.append_pool=none"
# Validate append_pool name length
(( $MDS1_VERSION >= $(version_code 2.15.61) )) &&
do_nodes $mdts $LCTL \
set_param mdd.*.append_pool="LOV_MAXPOOLNAME*" &&
error "Wrong pool name length should report error"
local orig_count=$(do_facet mds1 $LCTL get_param -n mdd.$FSNAME-MDT0000.append_stripe_count)
stack_trap "do_nodes $mdts $LCTL set_param mdd.*.append_stripe_count=$orig_count"
do_nodes $mdts $LCTL set_param mdd.*.append_stripe_count=1
$LFS setstripe $stripe_opt $DIR/$tdir
echo 1 > $DIR/$tdir/${tfile}.1
local count=$($LFS getstripe -c $DIR/$tdir/${tfile}.1)
(( $count == $setcount )) ||
error "(1) stripe count $count, should be $setcount"
local appendcount=1
echo 1 >> $DIR/$tdir/${tfile}.2_append
count=$($LFS getstripe -c $DIR/$tdir/${tfile}.2_append)
(( $count == $appendcount )) ||
error "(2)stripe count $count, should be $appendcount for append"
# Disable O_APPEND striping, verify it works
do_nodes $mdts $LCTL set_param mdd.*.append_stripe_count=0
# Should now get the default striping, which is 4
setcount=4
echo 1 >> $DIR/$tdir/${tfile}.3_append
count=$($LFS getstripe -c $DIR/$tdir/${tfile}.3_append)
(( $count == $setcount )) ||
error "(3) stripe count $count, should be $setcount"
# Try changing the stripe count for append files
do_nodes $mdts $LCTL set_param mdd.*.append_stripe_count=2
# Append striping is now 2 (directory default is still 4)
appendcount=2
echo 1 >> $DIR/$tdir/${tfile}.4_append
count=$($LFS getstripe -c $DIR/$tdir/${tfile}.4_append)
(( $count == $appendcount )) ||
error "(4) stripe count $count, should be $appendcount for append"
# Test append stripe count of -1
# Exercise LU-16872 patch with specific striping, only if MDS has fix
(( $MDS1_VERSION > $(version_code 2.15.56.46) )) &&
$LFS setstripe -o 0,$((OSTCOUNT - 1)) $DIR/$tdir &&
touch $DIR/$tdir/$tfile.specific.{1..128}
stack_trap "rm -f $DIR/$tdir/$tfile.*"
do_nodes $mdts $LCTL set_param mdd.*.append_stripe_count=-1
appendcount=$OSTCOUNT
echo 1 >> $DIR/$tdir/${tfile}.5
count=$($LFS getstripe -c $DIR/$tdir/${tfile}.5)
(( $count == $appendcount )) ||
error "(5) stripe count $count, should be $appendcount for append"
# Set append striping back to default of 1
do_nodes $mdts $LCTL set_param mdd.*.append_stripe_count=1
# Try a new default striping, PFL + DOM
$LFS setstripe -L mdt -E 1M -E -1 -c 2 $DIR/$tdir
# Create normal DOM file, DOM returns stripe count == 0
setcount=0
touch $DIR/$tdir/${tfile}.6
count=$($LFS getstripe -c $DIR/$tdir/${tfile}.6)
(( $count == $setcount )) ||
error "(6) stripe count $count, should be $setcount"
# Show
appendcount=1
echo 1 >> $DIR/$tdir/${tfile}.7_append
count=$($LFS getstripe -c $DIR/$tdir/${tfile}.7_append)
(( $count == $appendcount )) ||
error "(7) stripe count $count, should be $appendcount for append"
# Clean up DOM layout
$LFS setstripe -d $DIR/$tdir
# rest of tests don't work with FILESET
[ -n "$FILESET" ] && exit
save_layout_restore_at_exit $MOUNT
# Now test that append striping works when layout is from root
$LFS setstripe -c 2 $MOUNT
# Make a special directory for this
mkdir $DIR/${tdir}/${tdir}.2
# Verify for normal file
setcount=2
echo 1 > $DIR/${tdir}/${tdir}.2/${tfile}.8
count=$($LFS getstripe -c $DIR/$tdir/${tdir}.2/${tfile}.8)
(( $count == $setcount )) ||
error "(8) stripe count $count, should be $setcount"
appendcount=1
echo 1 >> $DIR/${tdir}/${tdir}.2/${tfile}.9_append
count=$($LFS getstripe -c $DIR/${tdir}/${tdir}.2/${tfile}.9_append)
(( $count == $appendcount )) ||
error "(9) stripe count $count, should be $appendcount for append"
# Now test O_APPEND striping with pools
pool_add $TESTNAME || error "pool creation failed"
pool_add_targets $TESTNAME 0 1 || error "Pool add targets failed"
do_nodes $mdts $LCTL set_param mdd.*.append_pool="$TESTNAME"
echo 1 >> $DIR/$tdir/${tfile}.10_append
pool=$($LFS getstripe -p $DIR/$tdir/${tfile}.10_append)
[[ "$pool" == "$TESTNAME" ]] || error "(10) incorrect pool: $pool"
# Check that count is still correct
appendcount=1
echo 1 >> $DIR/$tdir/${tfile}.11_append
count=$($LFS getstripe -c $DIR/$tdir/${tfile}.11_append)
(( $count == $appendcount )) ||
error "(11) stripe count $count, should be $appendcount for append"
# Disable O_APPEND stripe count, verify pool works separately
do_nodes $mdts $LCTL set_param mdd.*.append_stripe_count=0
echo 1 >> $DIR/$tdir/${tfile}.12_append
pool=$($LFS getstripe -p $DIR/$tdir/${tfile}.12_append)
[[ "$pool" == "$TESTNAME" ]] || error "(12) incorrect pool: $pool"
# Remove pool setting, verify it's not applied
do_nodes $mdts $LCTL set_param mdd.*.append_pool='none'
echo 1 >> $DIR/$tdir/${tfile}.13_append
pool=$($LFS getstripe -p $DIR/$tdir/${tfile}.13_append)
[[ -z "$pool" ]] || error "(13) pool found: $pool"
}
run_test 27M "test O_APPEND striping"
test_27N() {
combined_mgs_mds && skip "needs separate MGS/MDT"
pool_add $TESTNAME || error "pool_add failed"
do_facet mgs "$LCTL pool_list $FSNAME" |
grep -Fx "${FSNAME}.${TESTNAME}" ||
error "lctl pool_list on MGS failed"
}
run_test 27N "lctl pool_list on separate MGS gives correct pool name"
clean_foreign_symlink() {
trap 0
lctl set_param llite/$FSNAME-*/foreign_symlink_enable=0
for i in $DIR/$tdir/* ; do
$LFS unlink_foreign $i || true
done
}
test_27O() {
[[ $(lustre_version_code $SINGLEMDS) -le $(version_code 2.12.51) ]] &&
skip "Need MDS version newer than 2.12.51"
test_mkdir $DIR/$tdir
local uuid1=$(cat /proc/sys/kernel/random/uuid)
local uuid2=$(cat /proc/sys/kernel/random/uuid)
trap clean_foreign_symlink EXIT
# enable foreign_symlink behaviour
$LCTL set_param llite/$FSNAME-*/foreign_symlink_enable=1
# foreign symlink LOV format is a partial path by default
# create foreign file (lfs + API)
$LFS setstripe --foreign=symlink --flags 0xda05 \
-x "${uuid1}/${uuid2}" --mode 0600 $DIR/$tdir/${tfile} ||
error "$DIR/$tdir/${tfile}: create failed"
$LFS getstripe -v $DIR/$tdir/${tfile} |
grep "lfm_magic:.*0x0BD70BD0" ||
error "$DIR/$tdir/${tfile}: invalid LOV EA foreign magic"
$LFS getstripe -v $DIR/$tdir/${tfile} | grep "lfm_type:.*symlink" ||
error "$DIR/$tdir/${tfile}: invalid LOV EA foreign type"
$LFS getstripe -v $DIR/$tdir/${tfile} |
grep "lfm_flags:.*0x0000DA05" ||
error "$DIR/$tdir/${tfile}: invalid LOV EA foreign flags"
$LFS getstripe $DIR/$tdir/${tfile} |
grep "lfm_value:.*${uuid1}/${uuid2}" ||
error "$DIR/$tdir/${tfile}: invalid LOV EA foreign value"
# modify striping should fail
$LFS setstripe -c 2 $DIR/$tdir/$tfile &&
error "$DIR/$tdir/$tfile: setstripe should fail"
# R/W should fail ("/{foreign_symlink_prefix}/${uuid1}/" missing)
cat $DIR/$tdir/$tfile && error "$DIR/$tdir/$tfile: read should fail"
cat /etc/passwd > $DIR/$tdir/$tfile &&
error "$DIR/$tdir/$tfile: write should fail"
# rename should succeed
mv $DIR/$tdir/$tfile $DIR/$tdir/${tfile}.new ||
error "$DIR/$tdir/$tfile: rename has failed"
#remove foreign_symlink file should fail
rm $DIR/$tdir/${tfile}.new &&
error "$DIR/$tdir/${tfile}.new: remove of foreign_symlink file should fail"
#test fake symlink
mkdir /tmp/${uuid1} ||
error "/tmp/${uuid1}: mkdir has failed"
echo FOOFOO > /tmp/${uuid1}/${uuid2} ||
error "/tmp/${uuid1}/${uuid2}: echo has failed"
$LCTL set_param llite/$FSNAME-*/foreign_symlink_prefix=/tmp/
$CHECKSTAT -t link -l /tmp/${uuid1}/${uuid2} $DIR/$tdir/${tfile}.new ||
error "$DIR/$tdir/${tfile}.new: not seen as a symlink"
#read should succeed now
cat $DIR/$tdir/${tfile}.new | grep FOOFOO ||
error "$DIR/$tdir/${tfile}.new: symlink resolution has failed"
#write should succeed now
cat /etc/passwd > $DIR/$tdir/${tfile}.new ||
error "$DIR/$tdir/${tfile}.new: write should succeed"
diff /etc/passwd $DIR/$tdir/${tfile}.new ||
error "$DIR/$tdir/${tfile}.new: diff has failed"
diff /etc/passwd /tmp/${uuid1}/${uuid2} ||
error "/tmp/${uuid1}/${uuid2}: diff has failed"
#check that getstripe still works
$LFS getstripe $DIR/$tdir/${tfile}.new ||
error "$DIR/$tdir/${tfile}.new: getstripe should still work with foreign_symlink enabled"
# chmod should still succeed
chmod 644 $DIR/$tdir/${tfile}.new ||
error "$DIR/$tdir/${tfile}.new: chmod has failed"
# chown should still succeed
chown $RUNAS_ID:$RUNAS_GID $DIR/$tdir/${tfile}.new ||
error "$DIR/$tdir/${tfile}.new: chown has failed"
# rename should still succeed
mv $DIR/$tdir/${tfile}.new $DIR/$tdir/${tfile} ||
error "$DIR/$tdir/${tfile}.new: rename has failed"
#remove foreign_symlink file should still fail
rm $DIR/$tdir/${tfile} &&
error "$DIR/$tdir/${tfile}: remove of foreign_symlink file should fail"
#use special ioctl() to unlink foreign_symlink file
$LFS unlink_foreign $DIR/$tdir/${tfile} ||
error "$DIR/$tdir/$tfile: unlink/ioctl failed"
}
run_test 27O "basic ops on foreign file of symlink type"
test_27P() {
[[ $(lustre_version_code $SINGLEMDS) -le $(version_code 2.12.49) ]] &&
skip "Need MDS version newer than 2.12.49"
test_mkdir $DIR/$tdir
local uuid1=$(cat /proc/sys/kernel/random/uuid)
local uuid2=$(cat /proc/sys/kernel/random/uuid)
trap clean_foreign_symlink EXIT
# enable foreign_symlink behaviour
$LCTL set_param llite/$FSNAME-*/foreign_symlink_enable=1
# foreign symlink LMV format is a partial path by default
# create foreign dir (lfs + API)
$LFS mkdir --foreign=symlink --xattr="${uuid1}/${uuid2}" \
--flags=0xda05 --mode 0750 $DIR/$tdir/${tdir} ||
error "$DIR/$tdir/${tdir}: create failed"
$LFS getdirstripe -v $DIR/$tdir/${tdir}
$LFS getdirstripe -v $DIR/$tdir/${tdir} |
grep "lfm_magic:.*0x0CD50CD0" ||
error "$DIR/$tdir/${tdir}: invalid LMV EA magic"
$LFS getdirstripe -v $DIR/$tdir/${tdir} | grep "lfm_type:.*symlink" ||
error "$DIR/$tdir/${tdir}: invalid LMV EA type"
$LFS getdirstripe -v $DIR/$tdir/${tdir} |
grep "lfm_flags:.*0x0000DA05" ||
error "$DIR/$tdir/${tdir}: invalid LMV EA flags"
$LFS getdirstripe $DIR/$tdir/${tdir} |
grep "lfm_value.*${uuid1}/${uuid2}" ||
error "$DIR/$tdir/${tdir}: invalid LMV EA value"
# file create in dir should fail
# ("/{foreign_symlink_prefix}/${uuid1}/${uuid2}/" missing)
touch $DIR/$tdir/$tdir/$tfile && error "$DIR/$tdir: file create should fail"
# rename should succeed
mv $DIR/$tdir/$tdir $DIR/$tdir/${tdir}.new ||
error "$DIR/$tdir/$tdir: rename of foreign_symlink dir has failed"
#remove foreign_symlink dir should fail
rmdir $DIR/$tdir/${tdir}.new &&
error "$DIR/$tdir/${tdir}.new: remove of foreign_symlink dir should fail"
#test fake symlink
mkdir -p /tmp/${uuid1}/${uuid2} ||
error "/tmp/${uuid1}/${uuid2}: mkdir has failed"
echo FOOFOO > /tmp/${uuid1}/${uuid2}/foo ||
error "/tmp/${uuid1}/${uuid2}/foo: echo has failed"
$LCTL set_param llite/$FSNAME-*/foreign_symlink_prefix=/tmp/
$CHECKSTAT -t link -l /tmp/${uuid1}/${uuid2} $DIR/$tdir/${tdir}.new ||
error "$DIR/$tdir/${tdir}.new: not seen as a symlink"
cat $DIR/$tdir/${tdir}.new/foo | grep FOOFOO ||
error "$DIR/$tdir/${tdir}.new: symlink resolution has failed"
#check that getstripe fails now that foreign_symlink enabled
$LFS getdirstripe $DIR/$tdir/${tdir}.new ||
error "$DIR/$tdir/${tdir}.new: getdirstripe should still work with foreign_symlink enabled"
# file create in dir should work now
cp /etc/passwd $DIR/$tdir/${tdir}.new/$tfile ||
error "$DIR/$tdir/${tdir}.new/$tfile: file create should fail"
diff /etc/passwd $DIR/$tdir/${tdir}.new/$tfile ||
error "$DIR/$tdir/${tdir}.new/$tfile: diff has failed"
diff /etc/passwd /tmp/${uuid1}/${uuid2}/$tfile ||
error "/tmp/${uuid1}/${uuid2}/$tfile: diff has failed"
# chmod should still succeed
chmod 755 $DIR/$tdir/${tdir}.new ||
error "$DIR/$tdir/${tdir}.new: chmod has failed"
# chown should still succeed
chown $RUNAS_ID:$RUNAS_GID $DIR/$tdir/${tdir}.new ||
error "$DIR/$tdir/${tdir}.new: chown has failed"
# rename should still succeed
mv $DIR/$tdir/${tdir}.new $DIR/$tdir/${tdir} ||
error "$DIR/$tdir/${tdir}.new: rename of foreign_symlink dir has failed"
#remove foreign_symlink dir should still fail
rmdir $DIR/$tdir/${tdir} &&
error "$DIR/$tdir/${tdir}: remove of foreign_symlink dir should fail"
#use special ioctl() to unlink foreign_symlink file
$LFS unlink_foreign $DIR/$tdir/${tdir} ||
error "$DIR/$tdir/$tdir: unlink/ioctl failed"
#created file should still exist
[[ -f /tmp/${uuid1}/${uuid2}/$tfile ]] ||
error "/tmp/${uuid1}/${uuid2}/$tfile has been removed"
diff /etc/passwd /tmp/${uuid1}/${uuid2}/$tfile ||
error "/tmp/${uuid1}/${uuid2}/$tfile: diff has failed"
}
run_test 27P "basic ops on foreign dir of foreign_symlink type"
test_27Q() {
rm -f $TMP/$tfile $TMP/$tfile.loop $TMP/$tfile.none $TMP/$tfile.broken
stack_trap "rm -f $TMP/$tfile*"
test_mkdir $DIR/$tdir-1
test_mkdir $DIR/$tdir-2
echo 'It is what it is' > $DIR/$tdir-1/$tfile
lov_getstripe_old $DIR/$tdir-1/$tfile || error "$DIR/$tdir-1/$tfile: rc = $?"
ln -s $DIR/$tdir-1/$tfile $DIR/$tdir-2/$tfile
lov_getstripe_old $DIR/$tdir-2/$tfile || error "$DIR/$tdir-2/$tfile: rc = $?"
ln -s $DIR/$tdir-1/$tfile $TMP/$tfile
lov_getstripe_old $TMP/$tfile || error "$TMP/$tfile: rc = $?"
# Create some bad symlinks and ensure that we don't loop
# forever or something. These should return ELOOP (40) and
# ENOENT (2) but I don't want to test for that because there's
# always some weirdo architecture that needs to ruin
# everything by defining these error numbers differently.
ln -s $TMP/$tfile.loop $TMP/$tfile.loop
lov_getstripe_old $TMP/$tfile.loop && error "$TMP/$tfile.loop: rc = $?"
ln -s $TMP/$tfile.none $TMP/$tfile.broken
lov_getstripe_old $TMP/$tfile.broken && error "$TMP/$tfile.broken: rc = $?"
return 0
}
run_test 27Q "llapi_file_get_stripe() works on symlinks"
test_27R() {
(( $MDS1_VERSION >= $(version_code 2.14.55) )) ||
skip "need MDS 2.14.55 or later"
(( $OSTCOUNT >= 2 )) || skip_env "needs at least 2 OSTs"
local testdir="$DIR/$tdir"
test_mkdir -p $testdir
stack_trap "rm -rf $testdir"
$LFS setstripe -c -1 $testdir || error "setstripe failed"
local f1="$testdir/f1"
touch $f1 || error "failed to touch $f1"
local count=$($LFS getstripe -c $f1)
(( $count == $OSTCOUNT )) || error "wrong stripe count"
do_facet $SINGLEMDS $LCTL set_param lod.*.max_stripecount=-1
(( $? == 34 )) || error "setting max_stripecount to -1 should fail and return ERANGE"
local maxcount=$(($OSTCOUNT - 1))
local mdts=$(mdts_nodes)
do_nodes $mdts "$LCTL set_param lod.*.max_stripecount=$maxcount"
stack_trap "do_nodes $mdts $LCTL set_param lod.*.max_stripecount=0"
local f2="$testdir/f2"
touch $f2 || error "failed to touch $f2"
local count=$($LFS getstripe -c $f2)
(( $count == $maxcount )) || error "wrong stripe count"
}
run_test 27R "test max_stripecount limitation when stripe count is set to -1"
test_27T() {
(( $OST1_VERSION >= $(version_code 2.15.65.18) )) ||
skip "need ost1 >= v2_15_65-18-g8a865c65c691 for ENOSPC fix"
[ $(facet_host client) == $(facet_host ost1) ] &&
skip "need ost1 and client on different nodes"
# CFS_FAIL_ONCE is needed to get cfs_fail_count reset
#define OBD_FAIL_OSC_NO_GRANT 0x411
#define CFS_FAIL_SKIP 0x20000000
#define CFS_FAIL_ONCE 0x80000000
$LCTL set_param fail_loc=0xa0000411 fail_val=1
#define OBD_FAIL_OST_ENOSPC_VALID 0x255
do_facet ost1 "$LCTL set_param fail_loc=0x80000255"
$LFS setstripe -i 0 -c 1 $DIR/$tfile
# DIO does not support partial writes to a single stripe - a write to
# each stripe will fail or succeed entirely. So we disable hybrid IO
# so we can see the partial write behavior of buffered IO
local hybrid=$($LCTL get_param -n llite.*.hybrid_io)
$LCTL set_param llite.*.hybrid_io=0
stack_trap "$LCTL set_param -n llite.*.hybrid_io=$hybrid" EXIT
local pagesz=$(getconf PAGESIZE)
local pagenr=$($LCTL get_param -n osc.*-OST0000-*.max_pages_per_rpc)
$MULTIOP $DIR/$tfile oO_WRONLY:P$(((pagenr + 10) * pagesz))c ||
error "multiop failed"
(( $(stat -c '%s' $DIR/$tfile) == pagenr * pagesz )) ||
error "wrong size"
# this is to reset cfs_fail_count and to clear ar_force_sync flags
$MULTIOP $DIR/$tfile oO_WRONLY:w${pagesz}c
}
run_test 27T "no eio on close on partial write due to enosp"
test_27U() {
local dir=$DIR/$tdir
local file=$dir/$tfile
local append_pool=${TESTNAME}-append
local normal_pool=${TESTNAME}-normal
local pool
local stripe_count
local stripe_count2
local mdts=$(mdts_nodes)
(( $MDS1_VERSION >= $(version_code 2.15.51) )) ||
skip "Need MDS version at least 2.15.51 for append pool feature"
# Validate existing append_* params and ensure restore
pool=$(do_facet mds1 $LCTL get_param -n mdd.$FSNAME-MDT0000.append_pool)
[[ "$pool" == "" ]] || error "expected append_pool == '', got '$pool'"
stack_trap "do_nodes $mdts $LCTL set_param mdd.*.append_pool=none"
stripe_count=$(do_facet mds1 $LCTL get_param -n mdd.$FSNAME-MDT0000.append_stripe_count)
stack_trap "do_nodes $mdts $LCTL set_param mdd.*.append_stripe_count=$stripe_count"
do_nodes $mdts $LCTL set_param mdd.*.append_stripe_count=1
pool_add $append_pool || error "pool creation failed"
pool_add_targets $append_pool 0 1 || error "Pool add targets failed"
pool_add $normal_pool || error "pool creation failed"
pool_add_targets $normal_pool 0 1 || error "Pool add targets failed"
test_mkdir $dir
$LFS setstripe -E 1M -c 1 -p $normal_pool -E 2M -c 2 -p $normal_pool -E eof -c -1 $dir
echo XXX >> $file.1
$LFS getstripe $file.1
pool=$($LFS getstripe -p $file.1)
[[ "$pool" == "$normal_pool" ]] || error "got pool '$pool', expected '$normal_pool'"
stripe_count2=$($LFS getstripe -c $file.1)
((stripe_count2 == 1)) ||
error "got stripe_count '$stripe_count2', expected 1"
do_nodes $mdts $LCTL set_param mdd.*.append_pool=$append_pool
echo XXX >> $file.2
$LFS getstripe $file.2
pool=$($LFS getstripe -p $file.2)
[[ "$pool" == "$append_pool" ]] || error "got pool '$pool', expected '$append_pool'"
do_nodes $mdts $LCTL set_param mdd.*.append_stripe_count=2
echo XXX >> $file.3
$LFS getstripe $file.3
stripe_count2=$($LFS getstripe -c $file.3)
((stripe_count2 == 2)) || error "got stripe_count '$stripe_count2', expected 2"
}
run_test 27U "append pool and stripe count work with composite default layout"
test_27V() {
[ $PARALLEL == "yes" ] && skip "skip parallel run"
(( $OSTCOUNT >= 4 )) || skip_env "needs >= 4 OSTs"
local dir=$DIR/$tdir
local osp_param=osp.$FSNAME-OST0000-osc-MDT0000.max_create_count
local lod_param=lod.$FSNAME-MDT0000-mdtlov.qos_threshold_rr
local saved_max=$(do_facet mds1 $LCTL get_param -n $osp_param)
local saved_qos=$(do_facet mds1 $LCTL get_param -n $lod_param)
local pid
stack_trap "do_facet mds1 $LCTL set_param $osp_param=$saved_max"
do_facet mds1 $LCTL set_param $lod_param=0
stack_trap "do_facet mds1 $LCTL set_param $lod_param=$saved_qos"
$LFS setdirstripe --mdt-count=1 --mdt-index=0 $dir
stack_trap "rm -rf $dir"
# exercise race in LU-16981 with deactivating OST while creating a file
(
while true; do
do_facet mds1 $LCTL set_param $osp_param=0 > /dev/null
sleep 0.1
do_facet mds1 \
$LCTL set_param $osp_param=$saved_max > /dev/null
done
) &
pid=$!
stack_trap "kill -9 $pid"
# errors here are OK so ignore them (just don't want to crash)
$LFS setstripe -c -1 $dir/f.{1..200} 2> /dev/null
return 0
}
run_test 27V "creating widely striped file races with deactivating OST"
test_27W() {
[[ $($LCTL get_param mdc.*.import) =~ connect_flags.*overstriping ]] ||
skip "server does not support overstriping"
[ $PARALLEL == "yes" ] && skip "skip parallel run"
local stripe_count
local defcount=4
# Set back to unrestricted
stack_trap "$LCTL set_param llite.$FSNAME-*.enable_setstripe_gid=-1"
mkdir $DIR/$tdir
# Set a default layout
$LFS setstripe -C $defcount $DIR/$tdir || error "(0) root failed to set default layout"
chmod a+rw $DIR
chmod a+rw $DIR/$tdir
# Verify this works normally
$RUNAS $LFS setstripe -c 1 $DIR/$tdir/$tfile || error "(1) failed to setstripe"
rm -f $DIR/$tdir/$tfile
# Limit to CAP_SYS_RESOURCE
$LCTL set_param llite.$FSNAME-*.enable_setstripe_gid=0
# Prints a messsage, but will use the default layout for the directory
# This is so scripts which use setstripe and don't check error will not fail
$RUNAS $LFS setstripe -c 1 $DIR/$tdir/$tfile || error "(2) setstripe failed"
# Verify default layout was used on file:
stripe_count=$($LFS getstripe -c $DIR/$tdir/$tfile)
((stripe_count == defcount)) ||
error "(3) got stripe_count '$stripe_count', expected $defcount"
rm -f $DIR/$tdir/$tfile
# Allow for RUNAS group
$LCTL set_param llite.$FSNAME-*.enable_setstripe_gid=$RUNAS_GID
$RUNAS $LFS setstripe -c 1 $DIR/$tdir/$tfile || error "(4) failed to setstripe"
# Confirm we did not use the default layout
stripe_count=$($LFS getstripe -c $DIR/$tdir/$tfile)
((stripe_count == 1)) || error "(5) got stripe_count '$stripe_count', expected 1"
rm -f $DIR/$tdir/$tfile
# Set to some other GID
$LCTL set_param llite.$FSNAME-*.enable_setstripe_gid=1
# Should give warning and succeed
$RUNAS $LFS setstripe -c 1 $DIR/$tdir/$tfile || error "(6) setstripe failed"
# Confirm we used default layout
stripe_count=$($LFS getstripe -c $DIR/$tdir/$tfile)
((stripe_count == defcount)) ||
error "(7) got stripe_count '$stripe_count', expected $defcount"
}
run_test 27W "test enable_setstripe_gid"
test_27X() {
[[ $($LCTL get_param mdc.*.import) =~ connect_flags.*overstriping ]] ||
skip "server does not support overstriping"
local file=$DIR/$tfile
local count
local overstripe_count=$((OSTCOUNT * 2))
# Create a small file
dd if=/dev/zero of=$file bs=1K count=1 || error "dd failed"
# Migrate with --overstripe-count=2*OSTCOUNT
$LFS migrate --overstripe-count=$overstripe_count $file ||
error "migrate failed"
# Check that the stripe count is 2*OSTCOUNT
count=$($LFS getstripe -c $file)
[ $count -eq $overstripe_count ] ||
error "stripe count is $count, expected $overstripe_count"
# Check that the file has the overstriping flag set
$LFS getstripe $file | grep -q "overstriped" ||
error "overstriping flag not set"
# Clean up
rm -f $file
}
run_test 27X "lfs migrate honors --overstripe-count option"
# createtest also checks that device nodes are created and
# then visible correctly (#2091)
test_28() { # bug 2091
test_mkdir $DIR/d28
$CREATETEST $DIR/d28/ct || error "createtest failed"
}
run_test 28 "create/mknod/mkdir with bad file types ============"
test_29() {
[ $PARALLEL == "yes" ] && skip "skip parallel run"
[ $MDS1_VERSION -ge $(version_code 2.14.51) ] && {
disable_opencache
stack_trap "restore_opencache"
}
sync; sleep 1; sync # flush out any dirty pages from previous tests
cancel_lru_locks
test_mkdir $DIR/d29
touch $DIR/d29/foo
log 'first d29'
ls -l $DIR/d29
local locks_orig=$(total_used_locks mdc)
(( $locks_orig != 0 )) || error "No mdc lock count"
local locks_unused_orig=$(total_unused_locks mdc)
log 'second d29'
ls -l $DIR/d29
log 'done'
local locks_current=$(total_used_locks mdc)
local locks_unused_current=$(total_unused_locks mdc)
if (( $locks_current > $locks_orig )); then
$LCTL set_param -n ldlm.dump_namespaces ""
error "CURRENT: $locks_current > $locks_orig"
fi
if (( $locks_unused_current > $locks_unused_orig )); then
error "UNUSED: $locks_unused_current > $locks_unused_orig"
fi
}
run_test 29 "IT_GETATTR regression ============================"
test_30a() { # was test_30
cp $(which ls) $DIR || cp /bin/ls $DIR
$DIR/ls / || error "Can't execute binary from lustre"
rm $DIR/ls
}
run_test 30a "execute binary from Lustre (execve) =============="
test_30b() {
cp `which ls` $DIR || cp /bin/ls $DIR
chmod go+rx $DIR/ls
$RUNAS $DIR/ls / || error "Can't execute binary from lustre as non-root"
rm $DIR/ls
}
run_test 30b "execute binary from Lustre as non-root ==========="
test_30c() { # b=22376
[ $PARALLEL == "yes" ] && skip "skip parallel run"
cp $(which ls) $DIR || cp /bin/ls $DIR
chmod a-rw $DIR/ls
cancel_lru_locks mdc
cancel_lru_locks osc
$RUNAS $DIR/ls / || error "Can't execute binary from lustre"
rm -f $DIR/ls
}
run_test 30c "execute binary from Lustre without read perms ===="
test_30d() {
cp $(which dd) $DIR || error "failed to copy dd to $DIR/dd"
for i in {1..10}; do
$DIR/dd bs=1M count=128 if=/dev/zero of=$DIR/$tfile &
local PID=$!
sleep 1
$LCTL set_param ldlm.namespaces.*MDT*.lru_size=clear
wait $PID || error "executing dd from Lustre failed"
rm -f $DIR/$tfile
done
rm -f $DIR/dd
}
run_test 30d "execute binary from Lustre while clear locks"
test_31a() {
$OPENUNLINK $DIR/f31 $DIR/f31 || error "openunlink failed"
$CHECKSTAT -a $DIR/f31 || error "$DIR/f31 exists"
}
run_test 31a "open-unlink file =================================="
test_31b() {
touch $DIR/f31 || error "touch $DIR/f31 failed"
ln $DIR/f31 $DIR/f31b || error "ln failed"
$MULTIOP $DIR/f31b Ouc || error "multiop failed"
$CHECKSTAT -t file $DIR/f31 || error "$DIR/f31 not file type"
}
run_test 31b "unlink file with multiple links while open ======="
test_31c() {
touch $DIR/f31 || error "touch $DIR/f31 failed"
ln $DIR/f31 $DIR/f31c || error "ln failed"
multiop_bg_pause $DIR/f31 O_uc ||
error "multiop_bg_pause for $DIR/f31 failed"
MULTIPID=$!
$MULTIOP $DIR/f31c Ouc
kill -USR1 $MULTIPID
wait $MULTIPID
}
run_test 31c "open-unlink file with multiple links ============="
test_31d() {
opendirunlink $DIR/d31d $DIR/d31d || error "opendirunlink failed"
$CHECKSTAT -a $DIR/d31d || error "$DIR/d31d exists"
}
run_test 31d "remove of open directory ========================="
test_31e() { # bug 2904
openfilleddirunlink $DIR/d31e || error "openfilleddirunlink failed"
}
run_test 31e "remove of open non-empty directory ==============="
test_31f() { # bug 4554
[ $PARALLEL == "yes" ] && skip "skip parallel run"
set -vx
test_mkdir $DIR/d31f
$LFS setstripe -S 1048576 -c 1 $DIR/d31f
cp /etc/hosts $DIR/d31f
ls -l $DIR/d31f
$LFS getstripe $DIR/d31f/hosts
multiop_bg_pause $DIR/d31f D_c || return 1
MULTIPID=$!
rm -rv $DIR/d31f || error "first of $DIR/d31f"
test_mkdir $DIR/d31f
$LFS setstripe -S 1048576 -c 1 $DIR/d31f
cp /etc/hosts $DIR/d31f
ls -l $DIR/d31f
$LFS getstripe $DIR/d31f/hosts
multiop_bg_pause $DIR/d31f D_c || return 1
MULTIPID2=$!
kill -USR1 $MULTIPID || error "first opendir $MULTIPID not running"
wait $MULTIPID || error "first opendir $MULTIPID failed"
sleep 6
kill -USR1 $MULTIPID2 || error "second opendir $MULTIPID not running"
wait $MULTIPID2 || error "second opendir $MULTIPID2 failed"
set +vx
}
run_test 31f "remove of open directory with open-unlink file ==="
test_31g() {
echo "-- cross directory link --"
test_mkdir -c1 $DIR/${tdir}ga
test_mkdir -c1 $DIR/${tdir}gb
touch $DIR/${tdir}ga/f
ln $DIR/${tdir}ga/f $DIR/${tdir}gb/g
$CHECKSTAT -t file $DIR/${tdir}ga/f || error "source"
[ `stat -c%h $DIR/${tdir}ga/f` == '2' ] || error "source nlink"
$CHECKSTAT -t file $DIR/${tdir}gb/g || error "target"
[ `stat -c%h $DIR/${tdir}gb/g` == '2' ] || error "target nlink"
}
run_test 31g "cross directory link==============="
test_31h() {
echo "-- cross directory link --"
test_mkdir -c1 $DIR/${tdir}
test_mkdir -c1 $DIR/${tdir}/dir
touch $DIR/${tdir}/f
ln $DIR/${tdir}/f $DIR/${tdir}/dir/g
$CHECKSTAT -t file $DIR/${tdir}/f || error "source"
[ `stat -c%h $DIR/${tdir}/f` == '2' ] || error "source nlink"
$CHECKSTAT -t file $DIR/${tdir}/dir/g || error "target"
[ `stat -c%h $DIR/${tdir}/dir/g` == '2' ] || error "target nlink"
}
run_test 31h "cross directory link under child==============="
test_31i() {
echo "-- cross directory link --"
test_mkdir -c1 $DIR/$tdir
test_mkdir -c1 $DIR/$tdir/dir
touch $DIR/$tdir/dir/f
ln $DIR/$tdir/dir/f $DIR/$tdir/g
$CHECKSTAT -t file $DIR/$tdir/dir/f || error "source"
[ `stat -c%h $DIR/$tdir/dir/f` == '2' ] || error "source nlink"
$CHECKSTAT -t file $DIR/$tdir/g || error "target"
[ `stat -c%h $DIR/$tdir/g` == '2' ] || error "target nlink"
}
run_test 31i "cross directory link under parent==============="
test_31j() {
test_mkdir -c1 -p $DIR/$tdir
test_mkdir -c1 -p $DIR/$tdir/dir1
ln $DIR/$tdir/dir1 $DIR/$tdir/dir2 && error "ln for dir"
link $DIR/$tdir/dir1 $DIR/$tdir/dir3 && error "link for dir"
link $DIR/$tdir/dir1 $DIR/$tdir/dir1 && error "link to the same dir"
return 0
}
run_test 31j "link for directory"
test_31k() {
test_mkdir -c1 -p $DIR/$tdir
touch $DIR/$tdir/s
touch $DIR/$tdir/exist
link $DIR/$tdir/s $DIR/$tdir/t || error "link"
link $DIR/$tdir/s $DIR/$tdir/exist && error "link to exist file"
link $DIR/$tdir/s $DIR/$tdir/s && error "link to the same file"
link $DIR/$tdir/s $DIR/$tdir && error "link to parent dir"
link $DIR/$tdir $DIR/$tdir/s && error "link parent dir to target"
link $DIR/$tdir/not-exist $DIR/$tdir/foo && error "link non-existing to new"
link $DIR/$tdir/not-exist $DIR/$tdir/s && error "link non-existing to exist"
return 0
}
run_test 31k "link to file: the same, non-existing, dir"
test_31l() {
local ln_ver=$(ln --version | awk '/coreutils/ { print $4 }')
(( $(version_code $ln_ver) < $(version_code 8.31) )) ||
(( $(version_code $(uname -r)) >= $(version_code 5.18) )) ||
skip "need coreutils < 8.31 or kernel >= 5.18 for ln"
touch $DIR/$tfile || error "create failed"
mkdir $DIR/$tdir || error "mkdir failed"
ln $DIR/$tfile $DIR/$tdir/ || error "ln to '$tdir/' failed"
}
run_test 31l "link to file: target dir has trailing slash"
test_31m() {
mkdir $DIR/d31m
touch $DIR/d31m/s
mkdir $DIR/d31m2
touch $DIR/d31m2/exist
link $DIR/d31m/s $DIR/d31m2/t || error "link"
link $DIR/d31m/s $DIR/d31m2/exist && error "link to exist file"
link $DIR/d31m/s $DIR/d31m2 && error "link to parent dir"
link $DIR/d31m2 $DIR/d31m/s && error "link parent dir to target"
link $DIR/d31m/not-exist $DIR/d31m2/foo && error "link non-existing to new"
link $DIR/d31m/not-exist $DIR/d31m2/s && error "link non-existing to exist"
return 0
}
run_test 31m "link to file: the same, non-existing, dir"
test_31n() {
touch $DIR/$tfile || error "cannot create '$DIR/$tfile'"
nlink=$(stat --format=%h $DIR/$tfile)
[ ${nlink:--1} -eq 1 ] || error "nlink is $nlink, expected 1"
local fd=$(free_fd)
local cmd="exec $fd<$DIR/$tfile"
eval $cmd
cmd="exec $fd<&-"
trap "eval $cmd" EXIT
nlink=$(stat --dereference --format=%h /proc/self/fd/$fd)
[ ${nlink:--1} -eq 1 ] || error "nlink is $nlink, expected 1"
rm $DIR/$tfile || error "cannot remove '$DIR/$tfile'"
nlink=$(stat --dereference --format=%h /proc/self/fd/$fd)
[ ${nlink:--1} -eq 0 ] || error "nlink is $nlink, expected 0"
eval $cmd
}
run_test 31n "check link count of unlinked file"
link_one() {
local tempfile=$(mktemp $1_XXXXXX)
link $tempfile $1 2> /dev/null &&
echo "$BASHPID: link $tempfile to $1 succeeded"
unlink $tempfile
}
test_31o() { # LU-2901
test_mkdir $DIR/$tdir
for LOOP in $(seq 100); do
rm -f $DIR/$tdir/$tfile*
for THREAD in $(seq 8); do
link_one $DIR/$tdir/$tfile.$LOOP &
done
wait
local LINKS=$(ls -1 $DIR/$tdir | grep -c $tfile.$LOOP)
[[ $LINKS -gt 1 ]] && ls $DIR/$tdir &&
error "$LINKS duplicate links to $tfile.$LOOP" &&
break || true
done
}
run_test 31o "duplicate hard links with same filename"
test_31p() {
[ $MDSCOUNT -lt 2 ] && skip_env "needs >= 2 MDTs"
test_mkdir $DIR/$tdir
$LFS setdirstripe -i0 -c2 $DIR/$tdir/striped_dir
$LFS setdirstripe -D -c2 -H all_char $DIR/$tdir/striped_dir
opendirunlink $DIR/$tdir/striped_dir/test1 ||
error "open unlink test1 failed"
opendirunlink $DIR/$tdir/striped_dir/test2 ||
error "open unlink test2 failed"
$CHECKSTAT -a $DIR/$tdir/striped_dir/test1 ||
error "test1 still exists"
$CHECKSTAT -a $DIR/$tdir/striped_dir/test2 ||
error "test2 still exists"
}
run_test 31p "remove of open striped directory"
test_31q() {
[ $MDSCOUNT -lt 3 ] && skip_env "needs >= 3 MDTs"
$LFS mkdir -i 3,1 $DIR/$tdir || error "mkdir failed"
index=$($LFS getdirstripe -i $DIR/$tdir)
[ $index -eq 3 ] || error "first stripe index $index != 3"
index=$($LFS getdirstripe $DIR/$tdir | tail -1 | awk '{print $1}')
[ $index -eq 1 ] || error "second stripe index $index != 1"
# when "-c <stripe_count>" is set, the number of MDTs specified after
# "-i" should equal to the stripe count
$LFS mkdir -i 3,1 -c 3 $DIR/$tdir.2 && error "mkdir should fail" || true
}
run_test 31q "create striped directory on specific MDTs"
#LU-14949
test_31r() {
touch $DIR/$tfile.target
touch $DIR/$tfile.source
#OBD_FAIL_LLITE_OPEN_DELAY 0x1419
$LCTL set_param fail_loc=0x1419 fail_val=3
cat $DIR/$tfile.target &
CATPID=$!
# Guarantee open is waiting before we get here
sleep 1
mv $DIR/$tfile.source $DIR/$tfile.target
wait $CATPID
RC=$?
if [[ $RC -ne 0 ]]; then
error "open with cat failed, rc=$RC"
fi
}
run_test 31r "open-rename(replace) race"
cleanup_test32_mount() {
local rc=0
trap 0
local loopdev=$(losetup -a | grep $EXT2_DEV | sed -ne 's/:.*$//p')
$UMOUNT $DIR/$tdir/ext2-mountpoint || rc=$?
losetup -d $loopdev || true
rm -rf $DIR/$tdir
return $rc
}
test_32a() {
[ $PARALLEL == "yes" ] && skip "skip parallel run"
echo "== more mountpoints and symlinks ================="
[ -e $DIR/$tdir ] && rm -fr $DIR/$tdir
trap cleanup_test32_mount EXIT
test_mkdir -p $DIR/$tdir/ext2-mountpoint
mount -t ext2 -o loop $EXT2_DEV $DIR/$tdir/ext2-mountpoint ||
error "mount failed for $EXT2_DEV $DIR/$tdir/ext2-mountpoint"
$CHECKSTAT -t dir $DIR/$tdir/ext2-mountpoint/.. ||
error "$DIR/$tdir/ext2-mountpoint/.. not dir type"
cleanup_test32_mount
}
run_test 32a "stat d32a/ext2-mountpoint/.. ====================="
test_32b() {
[ $PARALLEL == "yes" ] && skip "skip parallel run"
[ -e $DIR/$tdir ] && rm -fr $DIR/$tdir
trap cleanup_test32_mount EXIT
test_mkdir -p $DIR/$tdir/ext2-mountpoint
mount -t ext2 -o loop $EXT2_DEV $DIR/$tdir/ext2-mountpoint ||
error "mount failed for $EXT2_DEV $DIR/$tdir/ext2-mountpoint"
ls -al $DIR/$tdir/ext2-mountpoint/.. ||
error "Can't list $DIR/$tdir/ext2-mountpoint/.."
cleanup_test32_mount
}
run_test 32b "open d32b/ext2-mountpoint/.. ====================="
test_32c() {
[ $PARALLEL == "yes" ] && skip "skip parallel run"
[ -e $DIR/$tdir ] && rm -fr $DIR/$tdir
trap cleanup_test32_mount EXIT
test_mkdir -p $DIR/$tdir/ext2-mountpoint
mount -t ext2 -o loop $EXT2_DEV $DIR/$tdir/ext2-mountpoint ||
error "mount failed for $EXT2_DEV $DIR/$tdir/ext2-mountpoint"
test_mkdir -p $DIR/$tdir/d2/test_dir
$CHECKSTAT -t dir $DIR/$tdir/ext2-mountpoint/../d2/test_dir ||
error "$DIR/$tdir/ext2-mountpoint/../d2/test_dir not dir type"
cleanup_test32_mount
}
run_test 32c "stat d32c/ext2-mountpoint/../d2/test_dir ========="
test_32d() {
[ $PARALLEL == "yes" ] && skip "skip parallel run"
[ -e $DIR/$tdir ] && rm -fr $DIR/$tdir
trap cleanup_test32_mount EXIT
test_mkdir -p $DIR/$tdir/ext2-mountpoint
mount -t ext2 -o loop $EXT2_DEV $DIR/$tdir/ext2-mountpoint ||
error "mount failed for $EXT2_DEV $DIR/$tdir/ext2-mountpoint"
test_mkdir -p $DIR/$tdir/d2/test_dir
ls -al $DIR/$tdir/ext2-mountpoint/../d2/test_dir ||
error "Can't list $DIR/$tdir/ext2-mountpoint/../d2/test_dir"
cleanup_test32_mount
}
run_test 32d "open d32d/ext2-mountpoint/../d2/test_dir"
test_32e() {
rm -fr $DIR/$tdir
test_mkdir -p $DIR/$tdir/tmp
local tmp_dir=$DIR/$tdir/tmp
ln -s $DIR/$tdir $tmp_dir/symlink11
ln -s $tmp_dir/symlink11 $tmp_dir/../symlink01
$CHECKSTAT -t link $DIR/$tdir/tmp/symlink11 || error "symlink11 bad"
$CHECKSTAT -t link $DIR/$tdir/symlink01 || error "symlink01 bad"
}
run_test 32e "stat d32e/symlink->tmp/symlink->lustre-subdir"
test_32f() {
rm -fr $DIR/$tdir
test_mkdir -p $DIR/$tdir/tmp
local tmp_dir=$DIR/$tdir/tmp
ln -s $DIR/$tdir $tmp_dir/symlink11
ln -s $tmp_dir/symlink11 $tmp_dir/../symlink01
ls $DIR/$tdir/tmp/symlink11 || error "symlink11 bad"
ls $DIR/$tdir/symlink01 || error "symlink01 bad"
}
run_test 32f "open d32f/symlink->tmp/symlink->lustre-subdir"
test_32g() {
local tmp_dir=$DIR/$tdir/tmp
test_mkdir -p $tmp_dir
test_mkdir $DIR/${tdir}2
ln -s $DIR/${tdir}2 $tmp_dir/symlink12
ln -s $tmp_dir/symlink12 $tmp_dir/../symlink02
$CHECKSTAT -t link $tmp_dir/symlink12 || error "symlink12 not a link"
$CHECKSTAT -t link $DIR/$tdir/symlink02 || error "symlink02 not a link"
$CHECKSTAT -t dir -f $tmp_dir/symlink12 || error "symlink12 not a dir"
$CHECKSTAT -t dir -f $DIR/$tdir/symlink02 || error "symlink12 not a dir"
}
run_test 32g "stat d32g/symlink->tmp/symlink->lustre-subdir/${tdir}2"
test_32h() {
rm -fr $DIR/$tdir $DIR/${tdir}2
tmp_dir=$DIR/$tdir/tmp
test_mkdir -p $tmp_dir
test_mkdir $DIR/${tdir}2
ln -s $DIR/${tdir}2 $tmp_dir/symlink12
ln -s $tmp_dir/symlink12 $tmp_dir/../symlink02
ls $tmp_dir/symlink12 || error "listing symlink12"
ls $DIR/$tdir/symlink02 || error "listing symlink02"
}
run_test 32h "open d32h/symlink->tmp/symlink->lustre-subdir/${tdir}2"
test_32i() {
[ $PARALLEL == "yes" ] && skip "skip parallel run"
[ -e $DIR/$tdir ] && rm -fr $DIR/$tdir
trap cleanup_test32_mount EXIT
test_mkdir -p $DIR/$tdir/ext2-mountpoint
mount -t ext2 -o loop $EXT2_DEV $DIR/$tdir/ext2-mountpoint ||
error "mount failed for $EXT2_DEV $DIR/$tdir/ext2-mountpoint"
touch $DIR/$tdir/test_file
$CHECKSTAT -t file $DIR/$tdir/ext2-mountpoint/../test_file ||
error "$DIR/$tdir/ext2-mountpoint/../test_file not file type"
cleanup_test32_mount
}
run_test 32i "stat d32i/ext2-mountpoint/../test_file ==========="
test_32j() {
[ $PARALLEL == "yes" ] && skip "skip parallel run"
[ -e $DIR/$tdir ] && rm -fr $DIR/$tdir
trap cleanup_test32_mount EXIT
test_mkdir -p $DIR/$tdir/ext2-mountpoint
mount -t ext2 -o loop $EXT2_DEV $DIR/$tdir/ext2-mountpoint ||
error "mount failed for $EXT2_DEV $DIR/$tdir/ext2-mountpoint"
touch $DIR/$tdir/test_file
cat $DIR/$tdir/ext2-mountpoint/../test_file ||
error "Can't open $DIR/$tdir/ext2-mountpoint/../test_file"
cleanup_test32_mount
}
run_test 32j "open d32j/ext2-mountpoint/../test_file ==========="
test_32k() {
[ $PARALLEL == "yes" ] && skip "skip parallel run"
rm -fr $DIR/$tdir
trap cleanup_test32_mount EXIT
test_mkdir -p $DIR/$tdir/ext2-mountpoint
mount -t ext2 -o loop $EXT2_DEV $DIR/$tdir/ext2-mountpoint ||
error "mount failed for $EXT2_DEV $DIR/$tdir/ext2-mountpoint"
test_mkdir -p $DIR/$tdir/d2
touch $DIR/$tdir/d2/test_file || error "touch failed"
$CHECKSTAT -t file $DIR/$tdir/ext2-mountpoint/../d2/test_file ||
error "$DIR/$tdir/ext2-mountpoint/../d2/test_file not file type"
cleanup_test32_mount
}
run_test 32k "stat d32k/ext2-mountpoint/../d2/test_file ========"
test_32l() {
[ $PARALLEL == "yes" ] && skip "skip parallel run"
rm -fr $DIR/$tdir
trap cleanup_test32_mount EXIT
test_mkdir -p $DIR/$tdir/ext2-mountpoint
mount -t ext2 -o loop $EXT2_DEV $DIR/$tdir/ext2-mountpoint ||
error "mount failed for $EXT2_DEV $DIR/$tdir/ext2-mountpoint"
test_mkdir -p $DIR/$tdir/d2
touch $DIR/$tdir/d2/test_file || error "touch failed"
cat $DIR/$tdir/ext2-mountpoint/../d2/test_file ||
error "Can't open $DIR/$tdir/ext2-mountpoint/../d2/test_file"
cleanup_test32_mount
}
run_test 32l "open d32l/ext2-mountpoint/../d2/test_file ========"
test_32m() {
rm -fr $DIR/d32m
test_mkdir -p $DIR/d32m/tmp
TMP_DIR=$DIR/d32m/tmp
ln -s $DIR $TMP_DIR/symlink11
ln -s $TMP_DIR/symlink11 $TMP_DIR/../symlink01
$CHECKSTAT -t link $DIR/d32m/tmp/symlink11 ||
error "symlink11 not a link"
$CHECKSTAT -t link $DIR/d32m/symlink01 ||
error "symlink01 not a link"
}
run_test 32m "stat d32m/symlink->tmp/symlink->lustre-root ======"
test_32n() {
rm -fr $DIR/d32n
test_mkdir -p $DIR/d32n/tmp
TMP_DIR=$DIR/d32n/tmp
ln -s $DIR $TMP_DIR/symlink11
ln -s $TMP_DIR/symlink11 $TMP_DIR/../symlink01
ls -l $DIR/d32n/tmp/symlink11 || error "listing symlink11"
ls -l $DIR/d32n/symlink01 || error "listing symlink01"
}
run_test 32n "open d32n/symlink->tmp/symlink->lustre-root ======"
test_32o() {
touch $DIR/$tfile
test_mkdir -p $DIR/d32o/tmp
TMP_DIR=$DIR/d32o/tmp
ln -s $DIR/$tfile $TMP_DIR/symlink12
ln -s $TMP_DIR/symlink12 $TMP_DIR/../symlink02
$CHECKSTAT -t link $DIR/d32o/tmp/symlink12 ||
error "symlink12 not a link"
$CHECKSTAT -t link $DIR/d32o/symlink02 || error "symlink02 not a link"
$CHECKSTAT -t file -f $DIR/d32o/tmp/symlink12 ||
error "$DIR/d32o/tmp/symlink12 not file type"
$CHECKSTAT -t file -f $DIR/d32o/symlink02 ||
error "$DIR/d32o/symlink02 not file type"
}
run_test 32o "stat d32o/symlink->tmp/symlink->lustre-root/$tfile"
test_32p() {
log 32p_1
rm -fr $DIR/d32p
log 32p_2
rm -f $DIR/$tfile
log 32p_3
touch $DIR/$tfile
log 32p_4
test_mkdir -p $DIR/d32p/tmp
log 32p_5
TMP_DIR=$DIR/d32p/tmp
log 32p_6
ln -s $DIR/$tfile $TMP_DIR/symlink12
log 32p_7
ln -s $TMP_DIR/symlink12 $TMP_DIR/../symlink02
log 32p_8
cat $DIR/d32p/tmp/symlink12 ||
error "Can't open $DIR/d32p/tmp/symlink12"
log 32p_9
cat $DIR/d32p/symlink02 || error "Can't open $DIR/d32p/symlink02"
log 32p_10
}
run_test 32p "open d32p/symlink->tmp/symlink->lustre-root/$tfile"
test_32q() {
[ $PARALLEL == "yes" ] && skip "skip parallel run"
[ -e $DIR/$tdir ] && rm -fr $DIR/$tdir
trap cleanup_test32_mount EXIT
test_mkdir -p $DIR/$tdir/ext2-mountpoint
touch $DIR/$tdir/ext2-mountpoint/under_the_mount || error "touch failed"
mount -t ext2 -o loop $EXT2_DEV $DIR/$tdir/ext2-mountpoint ||
error "mount failed for $EXT2_DEV $DIR/$tdir/ext2-mountpoint"
ls $DIR/$tdir/ext2-mountpoint | grep "\<under_the_mount\>" && error
cleanup_test32_mount
}
run_test 32q "stat follows mountpoints in Lustre (should return error)"
test_32r() {
[ $PARALLEL == "yes" ] && skip "skip parallel run"
[ -e $DIR/$tdir ] && rm -fr $DIR/$tdir
trap cleanup_test32_mount EXIT
test_mkdir -p $DIR/$tdir/ext2-mountpoint
touch $DIR/$tdir/ext2-mountpoint/under_the_mount || error "touch failed"
mount -t ext2 -o loop $EXT2_DEV $DIR/$tdir/ext2-mountpoint ||
error "mount failed for $EXT2_DEV $DIR/$tdir/ext2-mountpoint"
ls $DIR/$tdir/ext2-mountpoint | grep -q under_the_mount && error || true
cleanup_test32_mount
}
run_test 32r "opendir follows mountpoints in Lustre (should return error)"
test_33aa() {
rm -f $DIR/$tfile
touch $DIR/$tfile
chmod 444 $DIR/$tfile
chown $RUNAS_ID $DIR/$tfile
log 33_1
$RUNAS $OPENFILE -f O_RDWR $DIR/$tfile && error || true
log 33_2
}
run_test 33aa "write file with mode 444 (should return error)"
test_33a() {
rm -fr $DIR/$tdir
test_mkdir $DIR/$tdir
chown $RUNAS_ID $DIR/$tdir
$RUNAS $OPENFILE -f O_RDWR:O_CREAT -m 0444 $DIR/$tdir/$tfile ||
error "$RUNAS create $tdir/$tfile failed"
$RUNAS $OPENFILE -f O_RDWR:O_CREAT -m 0444 $DIR/$tdir/$tfile &&
error "open RDWR" || true
}
run_test 33a "test open file(mode=0444) with O_RDWR (should return error)"
test_33b() {
rm -fr $DIR/$tdir
test_mkdir $DIR/$tdir
chown $RUNAS_ID $DIR/$tdir
$RUNAS $OPENFILE -f 1286739555 $DIR/$tdir/$tfile || true
}
run_test 33b "test open file with malformed flags (No panic)"
test_33c() {
[ $PARALLEL == "yes" ] && skip "skip parallel run"
remote_ost_nodsh && skip "remote OST with nodsh"
local ostnum
local ostname
local write_bytes
local all_zeros
all_zeros=true
test_mkdir $DIR/$tdir
# Read: 0, Write: 4, create/destroy: 2/0, stat: 1, punch: 0
sync
for ostnum in $(seq $OSTCOUNT); do
# test-framework's OST numbering is one-based, while Lustre's
# is zero-based
ostname=$(printf "$FSNAME-OST%.4x" $((ostnum - 1)))
# check if at least some write_bytes stats are counted
write_bytes=$(do_facet ost$ostnum lctl get_param -n \
obdfilter.$ostname.stats |
awk '/^write_bytes/ {print $7}' )
echo "baseline_write_bytes@ost$ostnum/$ostname=$write_bytes"
if (( ${write_bytes:-0} > 0 )); then
all_zeros=false
break
fi
done
$all_zeros || return 0
# Write four bytes
echo foo > $DIR/$tdir/bar
# Really write them
sync
# Total up write_bytes after writing. We'd better find non-zeros.
for ostnum in $(seq $OSTCOUNT); do
ostname=$(printf "$FSNAME-OST%.4x" $((ostnum - 1)))
write_bytes=$(do_facet ost$ostnum lctl get_param -n \
obdfilter/$ostname/stats |
awk '/^write_bytes/ {print $7}' )
echo "write_bytes@ost$ostnum/$ostname=$write_bytes"
if (( ${write_bytes:-0} > 0 )); then
all_zeros=false
break
fi
done
if $all_zeros; then
for ostnum in $(seq $OSTCOUNT); do
ostname=$(printf "$FSNAME-OST%.4x" $((ostnum - 1)))
echo "Check write_bytes is in obdfilter.*.stats:"
do_facet ost$ostnum lctl get_param -n \
obdfilter.$ostname.stats
done
error "OST not keeping write_bytes stats (b=22312)"
fi
}
run_test 33c "test write_bytes stats"
test_33d() {
[[ $MDSCOUNT -lt 2 ]] && skip_env "needs >= 2 MDTs"
[ $PARALLEL == "yes" ] && skip "skip parallel run"
local MDTIDX=1
local remote_dir=$DIR/$tdir/remote_dir
test_mkdir $DIR/$tdir
$LFS mkdir -i $MDTIDX $remote_dir ||
error "create remote directory failed"
touch $remote_dir/$tfile
chmod 444 $remote_dir/$tfile
chown $RUNAS_ID $remote_dir/$tfile
$RUNAS $OPENFILE -f O_RDWR $DIR/$tfile && error || true
chown $RUNAS_ID $remote_dir
$RUNAS $OPENFILE -f O_RDWR:O_CREAT -m 0444 $remote_dir/f33 ||
error "create" || true
$RUNAS $OPENFILE -f O_RDWR:O_CREAT -m 0444 $remote_dir/f33 &&
error "open RDWR" || true
$RUNAS $OPENFILE -f 1286739555 $remote_dir/f33 || true
}
run_test 33d "openfile with 444 modes and malformed flags under remote dir"
test_33e() {
[ $MDSCOUNT -lt 2 ] && skip_env "needs >= 2 MDTs"
mkdir $DIR/$tdir
$LFS setdirstripe -i0 -c2 $DIR/$tdir/striped_dir
$LFS setdirstripe -i1 -c2 $DIR/$tdir/striped_dir1
mkdir $DIR/$tdir/local_dir
local s0_mode=$(stat -c%f $DIR/$tdir/striped_dir)
local s1_mode=$(stat -c%f $DIR/$tdir/striped_dir1)
local l_mode=$(stat -c%f $DIR/$tdir/local_dir)
[ "$l_mode" = "$s0_mode" -a "$l_mode" = "$s1_mode" ] ||
error "mkdir $l_mode striped0 $s0_mode striped1 $s1_mode"
rmdir $DIR/$tdir/* || error "rmdir failed"
umask 777
$LFS setdirstripe -i0 -c2 $DIR/$tdir/striped_dir
$LFS setdirstripe -i1 -c2 $DIR/$tdir/striped_dir1
mkdir $DIR/$tdir/local_dir
s0_mode=$(stat -c%f $DIR/$tdir/striped_dir)
s1_mode=$(stat -c%f $DIR/$tdir/striped_dir1)
l_mode=$(stat -c%f $DIR/$tdir/local_dir)
[ "$l_mode" = "$s0_mode" -a "$l_mode" = "$s1_mode" ] ||
error "mkdir $l_mode striped0 $s0_mode striped1 $s1_mode 777"
rmdir $DIR/$tdir/* || error "rmdir(umask 777) failed"
umask 000
$LFS setdirstripe -i0 -c2 $DIR/$tdir/striped_dir
$LFS setdirstripe -i1 -c2 $DIR/$tdir/striped_dir1
mkdir $DIR/$tdir/local_dir
s0_mode=$(stat -c%f $DIR/$tdir/striped_dir)
s1_mode=$(stat -c%f $DIR/$tdir/striped_dir1)
l_mode=$(stat -c%f $DIR/$tdir/local_dir)
[ "$l_mode" = "$s0_mode" -a "$l_mode" = "$s1_mode" ] ||
error "mkdir $l_mode striped0 $s0_mode striped1 $s1_mode 0"
}
run_test 33e "mkdir and striped directory should have same mode"
cleanup_33f() {
trap 0
do_facet $SINGLEMDS $LCTL set_param mdt.*.enable_remote_dir_gid=0
}
test_33f() {
[ $MDSCOUNT -lt 2 ] && skip_env "needs >= 2 MDTs"
remote_mds_nodsh && skip "remote MDS with nodsh"
mkdir $DIR/$tdir
chmod go+rwx $DIR/$tdir
do_facet $SINGLEMDS $LCTL set_param mdt.*.enable_remote_dir_gid=-1
trap cleanup_33f EXIT
$RUNAS lfs mkdir -i 0 -c$MDSCOUNT $DIR/$tdir/striped_dir ||
error "cannot create striped directory"
$RUNAS touch $DIR/$tdir/striped_dir/{0..16} ||
error "cannot create files in striped directory"
$RUNAS rm $DIR/$tdir/striped_dir/{0..16} ||
error "cannot remove files in striped directory"
$RUNAS rmdir $DIR/$tdir/striped_dir ||
error "cannot remove striped directory"
cleanup_33f
}
run_test 33f "nonroot user can create, access, and remove a striped directory"
test_33g() {
mkdir -p $DIR/$tdir/dir2
local err=$($RUNAS mkdir $DIR/$tdir/dir2 2>&1)
echo $err
[[ $err =~ "exists" ]] || error "Not exists error"
}
run_test 33g "nonroot user create already existing root created file"
sub_33h() {
local hash_type=$1
local count=250
test_mkdir -c $MDSCOUNT -H $hash_type $DIR/$tdir ||
error "lfs mkdir -H $hash_type $tdir failed"
touch $DIR/$tdir/$tfile || error "touch $tfile failed"
local index=$($LFS getstripe -m $DIR/$tdir/$tfile)
local index2
local fname
for fname in $DIR/$tdir/$tfile.bak \
$DIR/$tdir/$tfile.SAV \
$DIR/$tdir/$tfile.orig \
$DIR/$tdir/$tfile~; do
touch $fname || error "touch $fname failed"
index2=$($LFS getstripe -m $fname)
(( $index == $index2 )) ||
error "$fname MDT index mismatch $index != $index2"
done
local failed=0
local patterns=(".$tfile.XXXXXX" "$tfile.XXXXXXXX")
local pattern
local stat
for pattern in ${patterns[*]}; do
echo "pattern $pattern"
fname=$DIR/$tdir/$pattern
for (( i = 0; i < $count; i++ )); do
fname=$(mktemp $DIR/$tdir/$pattern) ||
error "mktemp $DIR/$tdir/$pattern failed"
index2=$($LFS getstripe -m $fname)
(( $index == $index2 )) && continue
failed=$((failed + 1))
echo "$fname MDT index mismatch $index != $index2"
done
done
echo "$failed/$count MDT index mismatches, expect ~2-4"
(( failed < 10 )) || error "MDT index mismatch $failed/$count times"
local same=0
local expect
# verify that "crush" is still broken with all files on same MDT,
# crush2 should have about 1/MDSCOUNT files on each MDT, with margin
[[ "$hash_type" == "crush" ]] && expect=$count ||
expect=$((count / MDSCOUNT))
local tries=3
for (( try = tries - 1; try >= 0; try-- )); do
# crush2 doesn't put all-numeric suffixes on the same MDT,
# filename like $tfile.12345678 should *not* be considered temp
for pattern in ${patterns[*]}; do
local base=${pattern%%X*}
local suff=${pattern#$base}
echo "pattern $pattern"
for (( i = 0; i < $count; i++ )); do
fname=$DIR/$tdir/$base$((${suff//X/1} + i))
touch $fname || error "touch $fname failed"
index2=$($LFS getstripe -m $fname)
(( $index != $index2 )) && continue
same=$((same + 1))
done
done
stat="$((same/${#patterns[*]}))/$count"
# the number of "bad" hashes is random, as it depends on the
# random filenames generated by "mktemp". Allow some margin in
# the results.
echo "$stat matches, expect ~$expect for $1"
(( same / ${#patterns[*]} <= expect * 9 / 7 &&
same / ${#patterns[*]} > expect * 5 / 7 )) && same=0 && break
log "MDT index match $stat times, $try tries left"
(( try > 0 )) ||
error "MDT index match $stat times, after $tries tries"
same=0
done
for (( try = tries - 1; try >= 0; try-- )); do
# crush2 doesn't put suffixes with special chars on the same MDT
# filenames like $tfile.txt.1234 should *not* be considered temp
for pattern in ${patterns[*]}; do
local base=${pattern%%X*}
local suff=${pattern#$base}
pattern=$base...${suff/XXX}
echo "pattern=$pattern"
for (( i = 0; i < $count; i++ )); do
fname=$(mktemp $DIR/$tdir/$pattern) ||
error "touch $fname failed"
index2=$($LFS getstripe -m $fname)
(( $index != $index2 )) && continue
same=$((same + 1))
done
done
stat="$((same/${#patterns[*]}))/$count"
# the number of "bad" hashes is random, as it depends on the
# filenames generated by "mktemp". Allow some margin in results
echo "$stat matches, expect ~$expect for $1"
(( same / ${#patterns[*]} <= expect * 9 / 7 &&
same / ${#patterns[*]} > expect * 5 / 7 )) && break
log "MDT index match $stat times, $try tries left"
(( try > 0 )) ||
error "MDT index match $stat times, after $tries tries"
same=0
done
}
test_33h() {
(( $MDSCOUNT >= 2 )) || skip "needs >= 2 MDTs"
(( $MDS1_VERSION >= $(version_code 2.13.50) )) ||
skip "Need MDS version at least 2.13.50"
sub_33h crush
}
run_test 33h "temp file is located on the same MDT as target (crush)"
test_33hh() {
(( $MDSCOUNT >= 2 )) || skip "needs >= 2 MDTs"
echo "MDS1_VERSION=$MDS1_VERSION version_code=$(version_code 2.15.0)"
(( $MDS1_VERSION > $(version_code 2.15.0) )) ||
skip "Need MDS version at least 2.15.0 for crush2"
sub_33h crush2
}
run_test 33hh "temp file is located on the same MDT as target (crush2)"
test_33i()
{
(( MDSCOUNT < 2 )) && skip "needs >= 2 MDTs"
local FNAME=$(str_repeat 'f' 250)
test_mkdir -i 0 -c $MDSCOUNT $DIR/$tdir || error "mkdir $tdir failed"
createmany -o $DIR/$tdir/$FNAME 1000 || error "createmany failed"
local count
local total
count=$($LFS getstripe -m $DIR/$tdir/* | grep -cw 1)
local MDC=$(lctl dl | awk '/MDT0001-mdc-[^M]/ { print $4 }')
lctl --device %$MDC deactivate
stack_trap "lctl --device %$MDC activate"
ls $DIR/$tdir > /dev/null && error "ls should return an error"
total=$(\ls -l $DIR/$tdir | wc -l)
# "ls -l" will list total in the first line
total=$((total - 1))
(( total + count == 1000 )) ||
error "ls list $total files, $count files on MDT1"
}
run_test 33i "striped directory can be accessed when one MDT is down"
test_33j() {
(( $MDSCOUNT >= 2 )) || skip "needs >= 2 MDTs"
mkdir -p $DIR/$tdir/
$LFS setdirstripe -D -i0,1 $DIR/$tdir/striped_dir_a &&
error "setdirstripe -D -i0,1 incorrectly succeeded"
$LFS setdirstripe -D -i0,1 -c1 $DIR/$tdir/striped_dir_b &&
error "setdirstripe -D -i0,1 -c1 incorrectly succeeded"
$LFS setdirstripe -D -i0,1 -c3 $DIR/$tdir/striped_dir_c &&
error "setdirstripe -D -i0,1 -c3 incorrectly succeeded"
$LFS setdirstripe -i0,1 $DIR/$tdir/striped_dir_e ||
error "-D was not specified, but still failed"
}
run_test 33j "lfs setdirstripe -D -i x,y,x should fail"
TEST_34_SIZE=${TEST_34_SIZE:-2000000000000}
test_34a() {
rm -f $DIR/f34
$MCREATE $DIR/f34 || error "mcreate failed"
$LFS getstripe $DIR/f34 2>&1 | grep -q "no stripe info" ||
error "getstripe failed"
$TRUNCATE $DIR/f34 $TEST_34_SIZE || error "truncate failed"
$LFS getstripe $DIR/f34 2>&1 | grep -q "no stripe info" ||
error "getstripe failed"
$CHECKSTAT -s $TEST_34_SIZE $DIR/f34 ||
error "Size of $DIR/f34 not equal to $TEST_34_SIZE bytes"
}
run_test 34a "truncate file that has not been opened ==========="
test_34b() {
[ ! -f $DIR/f34 ] && test_34a
$CHECKSTAT -s $TEST_34_SIZE $DIR/f34 ||
error "Size of $DIR/f34 not equal to $TEST_34_SIZE bytes"
$OPENFILE -f O_RDONLY $DIR/f34
$LFS getstripe $DIR/f34 2>&1 | grep -q "no stripe info" ||
error "getstripe failed"
$CHECKSTAT -s $TEST_34_SIZE $DIR/f34 ||
error "Size of $DIR/f34 not equal to $TEST_34_SIZE bytes"
}
run_test 34b "O_RDONLY opening file doesn't create objects ====="
test_34c() {
[ ! -f $DIR/f34 ] && test_34a
$CHECKSTAT -s $TEST_34_SIZE $DIR/f34 ||
error "Size of $DIR/f34 not equal to $TEST_34_SIZE bytes"
$OPENFILE -f O_RDWR $DIR/f34
$LFS getstripe $DIR/f34 2>&1 | grep -q "no stripe info" &&
error "$LFS getstripe failed"
$CHECKSTAT -s $TEST_34_SIZE $DIR/f34 ||
error "Size of $DIR/f34 not equal to $TEST_34_SIZE bytes"
}
run_test 34c "O_RDWR opening file-with-size works =============="
test_34d() {
[ ! -f $DIR/f34 ] && test_34a
dd if=/dev/zero of=$DIR/f34 conv=notrunc bs=4k count=1 ||
error "dd failed"
$CHECKSTAT -s $TEST_34_SIZE $DIR/f34 ||
error "Size of $DIR/f34 not equal to $TEST_34_SIZE bytes"
rm $DIR/f34
}
run_test 34d "write to sparse file ============================="
test_34e() {
rm -f $DIR/f34e
$MCREATE $DIR/f34e || error "mcreate failed"
$TRUNCATE $DIR/f34e 1000 || error "truncate failed"
$CHECKSTAT -s 1000 $DIR/f34e ||
error "Size of $DIR/f34e not equal to 1000 bytes"
$OPENFILE -f O_RDWR $DIR/f34e
$CHECKSTAT -s 1000 $DIR/f34e ||
error "Size of $DIR/f34e not equal to 1000 bytes"
}
run_test 34e "create objects, some with size and some without =="
test_34f() { # bug 6242, 6243
[ $PARALLEL == "yes" ] && skip "skip parallel run"
SIZE34F=48000
rm -f $DIR/f34f
$MCREATE $DIR/f34f || error "mcreate failed"
$TRUNCATE $DIR/f34f $SIZE34F || error "truncating $DIR/f3f to $SIZE34F"
dd if=$DIR/f34f of=$TMP/f34f
$CHECKSTAT -s $SIZE34F $TMP/f34f || error "$TMP/f34f not $SIZE34F bytes"
dd if=/dev/zero of=$TMP/f34fzero bs=$SIZE34F count=1
cmp $DIR/f34f $TMP/f34fzero || error "$DIR/f34f not all zero"
cmp $TMP/f34f $TMP/f34fzero || error "$TMP/f34f not all zero"
rm $TMP/f34f $TMP/f34fzero $DIR/f34f
}
run_test 34f "read from a file with no objects until EOF ======="
test_34g() {
[ $PARALLEL == "yes" ] && skip "skip parallel run"
dd if=/dev/zero of=$DIR/$tfile bs=1 count=100 seek=$TEST_34_SIZE ||
error "dd failed"
$TRUNCATE $DIR/$tfile $((TEST_34_SIZE / 2))|| error "truncate failed"
$CHECKSTAT -s $((TEST_34_SIZE / 2)) $DIR/$tfile ||
error "Size of $DIR/$tfile not equal to $((TEST_34_SIZE / 2))"
cancel_lru_locks osc
$CHECKSTAT -s $((TEST_34_SIZE / 2)) $DIR/$tfile ||
error "wrong size after lock cancel"
$TRUNCATE $DIR/$tfile $TEST_34_SIZE || error "truncate failed"
$CHECKSTAT -s $TEST_34_SIZE $DIR/$tfile ||
error "expanding truncate failed"
cancel_lru_locks osc
$CHECKSTAT -s $TEST_34_SIZE $DIR/$tfile ||
error "wrong expanded size after lock cancel"
}
run_test 34g "truncate long file ==============================="
test_34h() {
[ $PARALLEL == "yes" ] && skip "skip parallel run"
local gid=10
local sz=1000
dd if=/dev/zero of=$DIR/$tfile bs=1M count=10 || error "dd failed"
sync # Flush the cache so that multiop below does not block on cache
# flush when getting the group lock
$MULTIOP $DIR/$tfile OG${gid}T${sz}g${gid}c &
MULTIPID=$!
# Since just timed wait is not good enough, let's do a sync write
# that way we are sure enough time for a roundtrip + processing
# passed + 2 seconds of extra margin.
dd if=/dev/zero of=$DIR/${tfile}-1 bs=$PAGE_SIZE oflag=direct count=1
rm $DIR/${tfile}-1
sleep 2
if [[ `ps h -o comm -p $MULTIPID` == "multiop" ]]; then
error "Multiop blocked on ftruncate, pid=$MULTIPID"
kill -9 $MULTIPID
fi
wait $MULTIPID
local nsz=`stat -c %s $DIR/$tfile`
[[ $nsz == $sz ]] || error "New size wrong $nsz != $sz"
}
run_test 34h "ftruncate file under grouplock should not block"
test_35a() {
cp /usr/bin/bash $DIR/f35a
chmod 444 $DIR/f35a
chown $RUNAS_ID $DIR/f35a
$RUNAS $DIR/f35a && error || true
rm $DIR/f35a
}
run_test 35a "exec file with mode 444 (should return and not leak)"
test_36a() {
rm -f $DIR/f36
utime $DIR/f36 || error "utime failed for MDS"
}
run_test 36a "MDS utime check (mknod, utime)"
test_36b() {
echo "" > $DIR/f36
utime $DIR/f36 || error "utime failed for OST"
}
run_test 36b "OST utime check (open, utime)"
test_36c() {
rm -f $DIR/d36/f36
test_mkdir $DIR/d36
chown $RUNAS_ID $DIR/d36
$RUNAS utime $DIR/d36/f36 || error "utime failed for MDS as non-root"
}
run_test 36c "non-root MDS utime check (mknod, utime)"
test_36d() {
[ ! -d $DIR/d36 ] && test_36c
echo "" > $DIR/d36/f36
$RUNAS utime $DIR/d36/f36 || error "utime failed for OST as non-root"
}
run_test 36d "non-root OST utime check (open, utime)"
test_36e() {
[ $RUNAS_ID -eq $UID ] && skip_env "RUNAS_ID = UID = $UID -- skipping"
test_mkdir $DIR/$tdir
touch $DIR/$tdir/$tfile
$RUNAS utime $DIR/$tdir/$tfile &&
error "utime worked, expected failure" || true
}
run_test 36e "utime on non-owned file (should return error)"
subr_36fh() {
local fl="$1"
local LANG_SAVE=$LANG
local LC_LANG_SAVE=$LC_LANG
export LANG=C LC_LANG=C # for date language
DATESTR="Dec 20 2000"
test_mkdir $DIR/$tdir
lctl set_param fail_loc=$fl
date; date +%s
cp /etc/hosts $DIR/$tdir/$tfile
sync & # write RPC generated with "current" inode timestamp, but delayed
sleep 1
touch --date="$DATESTR" $DIR/$tdir/$tfile # setattr timestamp in past
LS_BEFORE="`ls -l $DIR/$tdir/$tfile`" # old timestamp from client cache
cancel_lru_locks $OSC
LS_AFTER="`ls -l $DIR/$tdir/$tfile`" # timestamp from OST object
date; date +%s
[ "$LS_BEFORE" != "$LS_AFTER" ] && \
echo "BEFORE: $LS_BEFORE" && \
echo "AFTER : $LS_AFTER" && \
echo "WANT : $DATESTR" && \
error "$DIR/$tdir/$tfile timestamps changed" || true
export LANG=$LANG_SAVE LC_LANG=$LC_LANG_SAVE
}
test_36f() {
[ $PARALLEL == "yes" ] && skip "skip parallel run"
#define OBD_FAIL_OST_BRW_PAUSE_BULK 0x214
subr_36fh "0x80000214"
}
run_test 36f "utime on file racing with OST BRW write =========="
test_36g() {
remote_ost_nodsh && skip "remote OST with nodsh"
[ $PARALLEL == "yes" ] && skip "skip parallel run"
[ $MDS1_VERSION -lt $(version_code 2.12.51) ] &&
skip "Need MDS version at least 2.12.51"
local fmd_max_age
local fmd
local facet="ost1"
local tgt="obdfilter"
[[ $OSC == "mdc" ]] && tgt="mdt" && facet="mds1"
test_mkdir $DIR/$tdir
fmd_max_age=$(do_facet $facet \
"lctl get_param -n $tgt.*.tgt_fmd_seconds 2> /dev/null | \
head -n 1")
echo "FMD max age: ${fmd_max_age}s"
touch $DIR/$tdir/$tfile
fmd=$(do_facet $facet "lctl get_param -n $tgt.*.exports.*.fmd_count" |
gawk '{cnt=cnt+$1} END{print cnt}')
echo "FMD before: $fmd"
[[ $fmd == 0 ]] &&
error "FMD wasn't create by touch"
sleep $((fmd_max_age + 12))
fmd=$(do_facet $facet "lctl get_param -n $tgt.*.exports.*.fmd_count" |
gawk '{cnt=cnt+$1} END{print cnt}')
echo "FMD after: $fmd"
[[ $fmd == 0 ]] ||
error "FMD wasn't expired by ping"
}
run_test 36g "FMD cache expiry ====================="
test_36h() {
[ $PARALLEL == "yes" ] && skip "skip parallel run"
#define OBD_FAIL_OST_BRW_PAUSE_BULK2 0x227
subr_36fh "0x80000227"
}
run_test 36h "utime on file racing with OST BRW write =========="
test_36i() {
[ $MDSCOUNT -lt 2 ] && skip_env "needs >= 2 MDTs"
test_mkdir $DIR/$tdir
$LFS setdirstripe -i0 -c$MDSCOUNT $DIR/$tdir/striped_dir
local mtime=$(stat -c%Y $DIR/$tdir/striped_dir)
local new_mtime=$((mtime + 200))
#change Modify time of striped dir
touch -m -d @$new_mtime $DIR/$tdir/striped_dir ||
error "change mtime failed"
local got=$(stat -c%Y $DIR/$tdir/striped_dir)
[ "$new_mtime" = "$got" ] || error "expect $new_mtime got $got"
}
run_test 36i "change mtime on striped directory"
# test_37 - duplicate with tests 32q 32r
test_38() {
local file=$DIR/$tfile
touch $file
openfile -f O_DIRECTORY $file
local RC=$?
local ENOTDIR=20
[ $RC -eq 0 ] && error "opened file $file with O_DIRECTORY" || true
[ $RC -eq $ENOTDIR ] || error "error $RC should be ENOTDIR ($ENOTDIR)"
}
run_test 38 "open a regular file with O_DIRECTORY should return -ENOTDIR ==="
test_39a() { # was test_39
touch $DIR/$tfile
touch $DIR/${tfile}2
# ls -l $DIR/$tfile $DIR/${tfile}2
# ls -lu $DIR/$tfile $DIR/${tfile}2
# ls -lc $DIR/$tfile $DIR/${tfile}2
sleep 2
$OPENFILE -f O_CREAT:O_TRUNC:O_WRONLY $DIR/${tfile}2
if [[ ! $DIR/${tfile}2 -nt $DIR/$tfile ]]; then
echo "mtime"
ls -l --full-time $DIR/$tfile $DIR/${tfile}2
echo "atime"
ls -lu --full-time $DIR/$tfile $DIR/${tfile}2
echo "ctime"
ls -lc --full-time $DIR/$tfile $DIR/${tfile}2
error "O_TRUNC didn't change timestamps"
fi
}
run_test 39a "mtime changed on create"
test_39b() {
test_mkdir -c1 $DIR/$tdir
cp -p /etc/passwd $DIR/$tdir/fopen
cp -p /etc/passwd $DIR/$tdir/flink
cp -p /etc/passwd $DIR/$tdir/funlink
cp -p /etc/passwd $DIR/$tdir/frename
ln $DIR/$tdir/funlink $DIR/$tdir/funlink2
sleep 1
echo "aaaaaa" >> $DIR/$tdir/fopen
echo "aaaaaa" >> $DIR/$tdir/flink
echo "aaaaaa" >> $DIR/$tdir/funlink
echo "aaaaaa" >> $DIR/$tdir/frename
local open_new=$(stat -c %Y $DIR/$tdir/fopen)
local link_new=$(stat -c %Y $DIR/$tdir/flink)
local unlink_new=$(stat -c %Y $DIR/$tdir/funlink)
local rename_new=$(stat -c %Y $DIR/$tdir/frename)
cat $DIR/$tdir/fopen > /dev/null
ln $DIR/$tdir/flink $DIR/$tdir/flink2
rm -f $DIR/$tdir/funlink2
mv -f $DIR/$tdir/frename $DIR/$tdir/frename2
for (( i=0; i < 2; i++ )) ; do
local open_new2=$(stat -c %Y $DIR/$tdir/fopen)
local link_new2=$(stat -c %Y $DIR/$tdir/flink)
local unlink_new2=$(stat -c %Y $DIR/$tdir/funlink)
local rename_new2=$(stat -c %Y $DIR/$tdir/frename2)
(( $open_new2 == $open_new )) ||
error "open file reverses mtime"
(( $link_new2 == $link_new )) ||
error "link file reverses mtime"
(( $unlink_new2 == $unlink_new )) ||
error "unlink file reverses mtime"
(( $rename_new2 == $rename_new )) ||
error "rename file reverses mtime"
cancel_lru_locks $OSC
(( $i > 0 )) || echo "repeat after cancel_lru_locks"
done
}
run_test 39b "mtime change on open, link, unlink, rename"
# this should be set to past
TEST_39_MTIME=$(date -d "1 year ago" +%s)
# bug 11063
test_39c() {
touch $DIR1/$tfile
sleep 2
local mtime0=$(stat -c %Y $DIR1/$tfile)
touch -m -d @$TEST_39_MTIME $DIR1/$tfile
local mtime1=$(stat -c %Y $DIR1/$tfile)
(( $mtime1 == $TEST_39_MTIME )) ||
error "mtime not set to past: $mtime1, should be $TEST_39_MTIME"
local d1=$(date +%s)
sleep 0.5
echo hello >> $DIR1/$tfile
sleep 0.5
local d2=$(date +%s)
local mtime2=$(stat -c %Y $DIR1/$tfile)
(( $d1 <= $mtime2 && $mtime2 <= $d2 )) ||
error "mtime is not updated on write: $d1 <= $mtime2 <= $d2"
mv $DIR1/$tfile $DIR1/$tfile-1
for (( i=0; i < 2; i++ )) ; do
local mtime3=$(stat -c %Y $DIR1/$tfile-1)
(( $mtime2 == $mtime3 )) ||
error "mtime ($mtime2) changed (to $mtime3) on rename"
cancel_lru_locks $OSC
(( $i > 0 )) || echo "repeat after cancel_lru_locks"
done
}
run_test 39c "mtime change on rename"
# bug 21114
test_39d() {
[[ $PARALLEL != "yes" ]] || skip "skip parallel run"
touch $DIR1/$tfile
touch -m -d @$TEST_39_MTIME $DIR1/$tfile
for (( i=0; i < 2; i++ )) ; do
local mtime=$(stat -c %Y $DIR1/$tfile)
(( $mtime == $TEST_39_MTIME )) ||
error "mtime($mtime) is not set to $TEST_39_MTIME"
cancel_lru_locks $OSC
(( $i > 0 )) || echo "repeat after cancel_lru_locks"
done
}
run_test 39d "create, utime, stat"
# bug 21114
test_39e() {
[[ $PARALLEL != "yes" ]] || skip "skip parallel run"
touch $DIR1/$tfile
local mtime1=$(stat -c %Y $DIR1/$tfile)
touch -m -d @$TEST_39_MTIME $DIR1/$tfile
for (( i=0; i < 2; i++ )) ; do
local mtime2=$(stat -c %Y $DIR1/$tfile)
(( $mtime2 == $TEST_39_MTIME )) ||
error "mtime($mtime2) is not set to $TEST_39_MTIME"
cancel_lru_locks $OSC
(( $i > 0 )) || echo "repeat after cancel_lru_locks"
done
}
run_test 39e "create, stat, utime, stat"
# bug 21114
test_39f() {
[[ $PARALLEL != "yes" ]] || skip "skip parallel run"
touch $DIR1/$tfile
mtime1=$(stat -c %Y $DIR1/$tfile)
sleep 2
touch -m -d @$TEST_39_MTIME $DIR1/$tfile
for (( i=0; i < 2; i++ )) ; do
local mtime2=$(stat -c %Y $DIR1/$tfile)
(( $mtime2 == $TEST_39_MTIME )) ||
error "mtime($mtime2) is not set to $TEST_39_MTIME"
cancel_lru_locks $OSC
(( $i > 0 )) || echo "repeat after cancel_lru_locks"
done
}
run_test 39f "create, stat, sleep, utime, stat"
# b=11063
test_39g() {
[[ $PARALLEL != "yes" ]] || skip "skip parallel run"
echo hello >> $DIR1/$tfile
local mtime1=$(stat -c %Y $DIR1/$tfile)
sleep 2
chmod o+r $DIR1/$tfile
for (( i=0; i < 2; i++ )) ; do
local mtime2=$(stat -c %Y $DIR1/$tfile)
(( $mtime1 == $mtime2 )) || \
error "lost mtime: $mtime2, should be $mtime1"
cancel_lru_locks $OSC
(( $i > 0 )) || echo "repeat after cancel_lru_locks"
done
}
run_test 39g "write, chmod, stat"
# bug 11063
test_39h() {
[[ $PARALLEL != "yes" ]] || skip "skip parallel run"
touch $DIR1/$tfile
sleep 1
local d1=$(date +%s)
echo hello >> $DIR1/$tfile
local mtime1=$(stat -c %Y $DIR1/$tfile)
touch -m -d @$TEST_39_MTIME $DIR1/$tfile
local d2=$(date +%s)
if (( $d1 != $d2 )); then
echo "write and touch not within one second"
else
for (( i=0; i < 2; i++ )) ; do
local mtime2=$(stat -c %Y $DIR1/$tfile)
(( $mtime2 == $TEST_39_MTIME )) ||
error "lost mtime: $mtime2, should be $TEST_39_MTIME"
cancel_lru_locks $OSC
(( $i > 0 )) || echo "repeat after cancel_lru_locks"
done
fi
}
run_test 39h "write, utime within one second, stat"
test_39i() {
touch $DIR1/$tfile
sleep 1
echo hello >> $DIR1/$tfile
local mtime1=$(stat -c %Y $DIR1/$tfile)
mv $DIR1/$tfile $DIR1/$tfile-1
for (( i=0; i < 2; i++ )) ; do
local mtime2=$(stat -c %Y $DIR1/$tfile-1)
(( $mtime1 == $mtime2 )) ||
error "lost mtime: $mtime2, should be $mtime1"
cancel_lru_locks $OSC
(( $i > 0 )) || echo "repeat after cancel_lru_locks"
done
}
run_test 39i "write, rename, stat"
test_39j() {
[[ $PARALLEL != "yes" ]] || skip "skip parallel run"
start_full_debug_logging
touch $DIR1/$tfile
sleep 1
stack_trap "rm -f $DIR1/$tfile"
#define OBD_FAIL_OSC_DELAY_SETTIME 0x412
lctl set_param fail_loc=0x80000412
multiop_bg_pause $DIR1/$tfile oO_RDWR:w2097152_c ||
error "multiop failed"
local multipid=$!
local mtime1=$(stat -c %Y $DIR1/$tfile)
mv $DIR1/$tfile $DIR1/$tfile-1
kill -USR1 $multipid
wait $multipid || error "multiop close failed"
for (( i=0; i < 2; i++ )) ; do
local mtime2=$(stat -c %Y $DIR1/$tfile-1)
(( $mtime1 == $mtime2 )) ||
error "mtime lost on close: $mtime2, should be $mtime1"
cancel_lru_locks
(( $i > 0 )) || echo "repeat after cancel_lru_locks"
done
lctl set_param fail_loc=0
stop_full_debug_logging
}
run_test 39j "write, rename, close, stat"
test_39k() {
[[ $PARALLEL != "yes" ]] || skip "skip parallel run"
touch $DIR1/$tfile
sleep 1
stack_trap "rm -f $DIR1/$tfile"
multiop_bg_pause $DIR1/$tfile oO_RDWR:w2097152_c || error "multiop failed"
local multipid=$!
local mtime1=$(stat -c %Y $DIR1/$tfile)
touch -m -d @$TEST_39_MTIME $DIR1/$tfile
kill -USR1 $multipid
wait $multipid || error "multiop close failed"
for (( i=0; i < 2; i++ )) ; do
local mtime2=$(stat -c %Y $DIR1/$tfile)
(( $mtime2 == $TEST_39_MTIME )) ||
error "mtime lost on close: $mtime2, should be $TEST_39_MTIME"
cancel_lru_locks
(( $i > 0 )) || echo "repeat after cancel_lru_locks"
done
}
run_test 39k "write, utime, close, stat"
# this should be set to future
TEST_39_ATIME=$(date -d "1 year" +%s)
test_39l() {
[[ $PARALLEL != "yes" ]] || skip "skip parallel run"
remote_mds_nodsh && skip "remote MDS with nodsh"
local atime_diff=$(do_facet $SINGLEMDS \
$LCTL get_param -n mdd.*MDT0000*.atime_diff)
rm -rf $DIR/$tdir
mkdir_on_mdt0 $DIR/$tdir
# test setting directory atime to future
touch -a -d @$TEST_39_ATIME $DIR/$tdir
local atime=$(stat -c %X $DIR/$tdir)
(( $atime == $TEST_39_ATIME )) ||
error "atime is not set to future: $atime, $TEST_39_ATIME"
# test setting directory atime from future to now
local now=$(date +%s)
touch -a -d @$now $DIR/$tdir
atime=$(stat -c %X $DIR/$tdir)
(( $atime == $now )) ||
error "atime is not updated from future: $atime, $now"
do_facet $SINGLEMDS lctl set_param -n mdd.*MDT0000*.atime_diff=2
sleep 3
# test setting directory atime when now > dir atime + atime_diff
local d1=$(date +%s)
sleep 0.5
ls $DIR/$tdir
sleep 0.5
local d2=$(date +%s)
cancel_lru_locks mdc
atime=$(stat -c %X $DIR/$tdir)
(( $d1 <= $atime && $atime <= $d2 )) ||
error "atime is not updated : $atime, should be $d2"
do_facet $SINGLEMDS lctl set_param -n mdd.*MDT0000*.atime_diff=60
sleep 3
# test not setting directory atime when now < dir atime + atime_diff
ls $DIR/$tdir
cancel_lru_locks mdc
atime=$(stat -c %X $DIR/$tdir)
(( $d1 <= $atime && $atime <= $d2 )) ||
error "atime is updated to $atime, should remain $d1<atime<$d2"
do_facet $SINGLEMDS \
$LCTL set_param -n mdd.*MDT0000*.atime_diff=$atime_diff
}
run_test 39l "directory atime update"
test_39m() {
[[ $PARALLEL != "yes" ]] || skip "skip parallel run"
touch $DIR1/$tfile
sleep 2
local far_past_mtime=$(date -d "May 29 1953" +%s)
local far_past_atime=$(date -d "Dec 17 1903" +%s)
touch -m -d @$far_past_mtime $DIR1/$tfile
touch -a -d @$far_past_atime $DIR1/$tfile
for (( i=0; i < 2; i++ )) ; do
local timestamps=$(stat -c "%X %Y" $DIR1/$tfile)
[[ "$timestamps" == "$far_past_atime $far_past_mtime" ]] ||
error "atime or mtime set incorrectly"
cancel_lru_locks $OSC
(( $i > 0 )) || echo "repeat after cancel_lru_locks"
done
}
run_test 39m "test atime and mtime before 1970"
test_39n() { # LU-3832
remote_mds_nodsh && skip "remote MDS with nodsh"
local atime_diff=$(do_facet $SINGLEMDS \
$LCTL get_param -n mdd.*MDT0000*.atime_diff)
local atime0
local atime1
local atime2
do_facet $SINGLEMDS lctl set_param -n mdd.*MDT0000*.atime_diff=1
rm -rf $DIR/$tfile
dd if=/dev/zero of=$DIR/$tfile bs=4096 count=1 status=noxfer
atime0=$(stat -c %X $DIR/$tfile)
sleep 5
$MULTIOP $DIR/$tfile oO_RDONLY:O_NOATIME:r4096c
atime1=$(stat -c %X $DIR/$tfile)
sleep 5
cancel_lru_locks mdc
cancel_lru_locks osc
$MULTIOP $DIR/$tfile oO_RDONLY:O_NOATIME:r4096c
atime2=$(stat -c %X $DIR/$tfile)
do_facet $SINGLEMDS \
lctl set_param -n mdd.*MDT0000*.atime_diff=$atime_diff
(( "$atime0" == "$atime1" )) || error "atime0 $atime0 != atime1 $atime1"
(( "$atime1" == "$atime2" )) || error "atime0 $atime0 != atime1 $atime1"
}
run_test 39n "check that O_NOATIME is honored"
test_39o() {
TESTDIR=$DIR/$tdir/$tfile
[ -e $TESTDIR ] && rm -rf $TESTDIR
mkdir -p $TESTDIR
cd $TESTDIR
links1=2
ls
mkdir a b
ls
links2=$(stat -c %h .)
[ $(($links1 + 2)) != $links2 ] &&
error "wrong links count $(($links1 + 2)) != $links2"
rmdir b
links3=$(stat -c %h .)
[ $(($links1 + 1)) != $links3 ] &&
error "wrong links count $links1 != $links3"
return 0
}
run_test 39o "directory cached attributes updated after create"
test_39p() {
[[ $MDSCOUNT -lt 2 ]] && skip_env "needs >= 2 MDTs"
local MDTIDX=1
TESTDIR=$DIR/$tdir/$tdir
[ -e $TESTDIR ] && rm -rf $TESTDIR
test_mkdir -p $TESTDIR
cd $TESTDIR
links1=2
ls
test_mkdir -i $MDTIDX $TESTDIR/remote_dir1
test_mkdir -i $MDTIDX $TESTDIR/remote_dir2
ls
links2=$(stat -c %h .)
[ $(($links1 + 2)) != $links2 ] &&
error "wrong links count $(($links1 + 2)) != $links2"
rmdir remote_dir2
links3=$(stat -c %h .)
[ $(($links1 + 1)) != $links3 ] &&
error "wrong links count $links1 != $links3"
return 0
}
run_test 39p "remote directory cached attributes updated after create ========"
test_39r() {
[ $OST1_VERSION -ge $(version_code 2.13.52) ] ||
skip "no atime update on old OST"
if [ "$ost1_FSTYPE" != ldiskfs ]; then
skip_env "ldiskfs only test"
fi
local saved_adiff
local ahost=$(facet_active_host ost1)
saved_adiff=$(do_facet ost1 \
lctl get_param -n obdfilter.*OST0000.atime_diff)
stack_trap "do_facet ost1 \
lctl set_param obdfilter.*.atime_diff=$saved_adiff"
do_facet ost1 "lctl set_param obdfilter.*.atime_diff=5"
$LFS setstripe -i 0 $DIR/$tfile
dd if=/dev/zero of=$DIR/$tfile bs=4k count=1 ||
error "can't write initial file"
cancel_lru_locks osc
# exceed atime_diff and access file
sleep 10
dd if=$DIR/$tfile of=/dev/null bs=4k count=1 ||
error "can't udpate atime"
# atime_cli value is in decimal
local atime_cli=$(stat -c %X $DIR/$tfile)
echo "client atime: $atime_cli"
local ostdev=$(ostdevname 1)
local fids=($($LFS getstripe $DIR/$tfile | grep 0x))
local fid="${fids[3]}:${fids[2]}:0"
local objpath=$(ost_fid2_objpath ost1 $fid)
local cmd="debugfs -c -R \\\"stat $objpath\\\" $ostdev"
# allow atime update to be written to device
do_facet ost1 "$LCTL set_param -n osd*.*OST*.force_sync=1"
echo "OST atime: $(do_facet ost1 "$cmd" |& grep atime)"
# Give enough time for server to get updated. Until then
# the value read is defaulted to "0x00000000:00000000"
# Wait until atime read via debugfs is not equal to zero.
# Max limit to wait is 30 seconds.
wait_update_cond $ahost \
"$cmd |& awk -F'[: ]' '/atime:/ { print \\\$4 }'" \
"-gt" "0" 30 || error "atime on ost is still 0 after 30 seconds"
# atime_ost value is in hex
local atime_ost=$(do_facet ost1 "$cmd" |&
awk -F'[: ]' '/atime:/ { print $4 }')
# debugfs returns atime in 0xNNNNNNNN:00000000 format
# convert Hex to decimal before comparing
local atime_ost_dec=$((atime_ost))
# The test pass criteria is that the client time and server should
# be same (2s gap accepted). This gap could arise due to VFS updating
# the atime after the read(dd), stat and the updated time from the
# inode
(( $((atime_cli - atime_ost_dec)) <= 2 )) ||
error "atime on client $atime_cli != ost $atime_ost_dec"
}
run_test 39r "lazy atime update on OST"
test_39q() { # LU-8041
local testdir=$DIR/$tdir
mkdir -p $testdir
multiop_bg_pause $testdir D_c || error "multiop failed"
local multipid=$!
cancel_lru_locks mdc
kill -USR1 $multipid
local atime=$(stat -c %X $testdir)
[ "$atime" -ne 0 ] || error "atime is zero"
}
run_test 39q "close won't zero out atime"
test_39s() {
local atime0
local atime1
local atime2
local atime3
local atime4
umount_client $MOUNT
mount_client $MOUNT relatime
dd if=/dev/zero of=$DIR/$tfile bs=4096 count=1 status=noxfer conv=fsync
atime0=$(stat -c %X $DIR/$tfile)
# First read updates atime
sleep 1
cat $DIR/$tfile >/dev/null
atime1=$(stat -c %X $DIR/$tfile) # (atime = atime0 + 1)
# Next reads do not update atime
sleep 1
cat $DIR/$tfile >/dev/null
atime2=$(stat -c %X $DIR/$tfile) # (atime = atime0 + 1)
# If mtime is greater than atime, atime is updated
sleep 1
touch -m $DIR/$tfile # (mtime = now)
sleep 1
cat $DIR/$tfile >/dev/null # (atime is updated because atime < mtime)
atime3=$(stat -c %X $DIR/$tfile) # (atime = mtime = atime0 + 3)
# Next reads do not update atime
sleep 1
cat $DIR/$tfile >/dev/null
atime4=$(stat -c %X $DIR/$tfile)
# Remount the client to clear 'relatime' option
remount_client $MOUNT
(( atime0 < atime1 )) ||
error "atime $atime0 should be smaller than $atime1"
(( atime1 == atime2 )) ||
error "atime $atime1 was updated to $atime2"
(( atime1 < atime3 )) || error "atime1 $atime1 != atime3 $atime3"
(( atime3 == atime4 )) || error "atime3 $atime3 != atime4 $atime4"
}
run_test 39s "relatime is supported"
test_39u() {
touch $DIR/$tfile
stack_trap "rm -f $DIR/$tfile"
sleep 2
dd if=/dev/zero of=$DIR/$tfile bs=1M count=1 conv=notrunc ||
error "dd failed"
#define OBD_FAIL_LLITE_STAT_RACE1 0x1434
$LCTL set_param fail_loc=0x80001434
local mtimes=($(stat -c "%Y" $DIR/$tfile &
sleep 1; stat -c "%Y" $DIR/$tfile; wait))
(( ${mtimes[0]} == ${mtimes[1]} )) ||
error "mtime mismatch ${mtimes[0]} != ${mtimes[1]}"
}
run_test 39u "stat race"
test_40() {
dd if=/dev/zero of=$DIR/$tfile bs=4096 count=1
$RUNAS $OPENFILE -f O_WRONLY:O_TRUNC $DIR/$tfile &&
error "openfile O_WRONLY:O_TRUNC $tfile failed"
$CHECKSTAT -t file -s 4096 $DIR/$tfile ||
error "$tfile is not 4096 bytes in size"
}
run_test 40 "failed open(O_TRUNC) doesn't truncate ============="
test_41() {
# bug 1553
small_write $DIR/f41 18
}
run_test 41 "test small file write + fstat ====================="
count_ost_writes() {
lctl get_param -n ${OSC}.*.stats |
awk -vwrites=0 '/ost_write/ { writes += $2 } \
END { printf("%0.0f", writes) }'
}
# decent default
WRITEBACK_SAVE=500
DIRTY_RATIO_SAVE=40
MAX_DIRTY_RATIO=50
BG_DIRTY_RATIO_SAVE=10
MAX_BG_DIRTY_RATIO=25
start_writeback() {
sysctl -w vm.dirty_writeback_centisecs=$WRITEBACK_SAVE
sysctl -w vm.dirty_background_ratio=$BG_DIRTY_RATIO_SAVE
sysctl -w vm.dirty_ratio=$DIRTY_RATIO_SAVE
}
stop_writeback() {
# setup the trap first, so someone cannot exit the test at the
# exact wrong time and mess up a machine
stack_trap start_writeback
# save and zero /proc/sys/vm/dirty_writeback_centisecs
WRITEBACK_SAVE=$(sysctl -n vm.dirty_writeback_centisecs)
sysctl -w vm.dirty_writeback_centisecs=0
sysctl -w vm.dirty_writeback_centisecs=0
# save and increase /proc/sys/vm/dirty_ratio
DIRTY_RATIO_SAVE=$(sysctl -n vm.dirty_ratio)
sysctl -w vm.dirty_ratio=$MAX_DIRTY_RATIO
# save and increase /proc/sys/vm/dirty_background_ratio
BG_DIRTY_RATIO_SAVE=$(sysctl -n vm.dirty_background_ratio)
sysctl -w vm.dirty_background_ratio=$MAX_BG_DIRTY_RATIO
# disable sync_on_close if enabled
local save="$($LCTL get_param llite.*.sync_on_close)"
[[ ! $save =~ "=1" ]] || {
stack_trap "$LCTL set_param $save"
$LCTL set_param llite.*.sync_on_close=0
}
}
# ensure that all stripes have some grant before we test client-side cache
setup_test42() {
for ((i=1; i <= $OSTCOUNT; i++)); do
local file=$DIR/$tfile-$i
dd if=/dev/zero of=$file bs=4k count=1
rm -f $file
done
}
# Tests 42* verify that our behaviour is correct WRT caching, file closure,
# file truncation, and file removal.
test_42a() {
[ $PARALLEL == "yes" ] && skip "skip parallel run"
setup_test42
cancel_lru_locks $OSC
stop_writeback
sync; sleep 1; sync # just to be safe
BEFOREWRITES=`count_ost_writes`
lctl get_param -n osc.*[oO][sS][cC][_-]*.cur_grant_bytes | grep "[0-9]"
dd if=/dev/zero of=$DIR/f42a bs=1024 count=100
AFTERWRITES=`count_ost_writes`
[ $BEFOREWRITES -eq $AFTERWRITES ] || \
error "$BEFOREWRITES < $AFTERWRITES"
start_writeback
}
run_test 42a "ensure that we don't flush on close"
test_42b() {
[ $PARALLEL == "yes" ] && skip "skip parallel run"
setup_test42
cancel_lru_locks $OSC
stop_writeback
sync
dd if=/dev/zero of=$DIR/f42b bs=1024 count=100
BEFOREWRITES=$(count_ost_writes)
unlink $DIR/f42b || error "unlink $DIR/f42b: $?"
AFTERWRITES=$(count_ost_writes)
if [[ $BEFOREWRITES -lt $AFTERWRITES ]]; then
error "$BEFOREWRITES < $AFTERWRITES on unlink"
fi
BEFOREWRITES=$(count_ost_writes)
sync || error "sync: $?"
AFTERWRITES=$(count_ost_writes)
if [[ $BEFOREWRITES -lt $AFTERWRITES ]]; then
error "$BEFOREWRITES < $AFTERWRITES on sync"
fi
dmesg | grep 'error from obd_brw_async' && error 'error writing back'
start_writeback
return 0
}
run_test 42b "test destroy of file with cached dirty data ======"
# if these tests just want to test the effect of truncation,
# they have to be very careful. consider:
# - the first open gets a {0,EOF}PR lock
# - the first write conflicts and gets a {0, count-1}PW
# - the rest of the writes are under {count,EOF}PW
# - the open for truncate tries to match a {0,EOF}PR
# for the filesize and cancels the PWs.
# any number of fixes (don't get {0,EOF} on open, match
# composite locks, do smarter file size management) fix
# this, but for now we want these tests to verify that
# the cancellation with truncate intent works, so we
# start the file with a full-file pw lock to match against
# until the truncate.
trunc_test() {
test=$1
file=$DIR/$test
offset=$2
cancel_lru_locks $OSC
stop_writeback
# prime the file with 0,EOF PW to match
touch $file
$TRUNCATE $file 0
sync; sync
# now the real test..
dd if=/dev/zero of=$file bs=1024 count=100
BEFOREWRITES=`count_ost_writes`
$TRUNCATE $file $offset
cancel_lru_locks $OSC
AFTERWRITES=`count_ost_writes`
start_writeback
}
test_42c() {
[ $PARALLEL == "yes" ] && skip "skip parallel run"
trunc_test 42c 1024
[ $BEFOREWRITES -eq $AFTERWRITES ] &&
error "beforewrites $BEFOREWRITES == afterwrites $AFTERWRITES on truncate"
rm $file
}
run_test 42c "test partial truncate of file with cached dirty data"
test_42d() {
[ $PARALLEL == "yes" ] && skip "skip parallel run"
local olddebug="$($LCTL get_param -n debug 2> /dev/null)"
stack_trap "$LCTL set_param -n debug='$olddebug'" EXIT
$LCTL set_param debug=+cache
trunc_test 42d 0
[ $BEFOREWRITES -eq $AFTERWRITES ] ||
error "beforewrites $BEFOREWRITES != afterwrites $AFTERWRITES on truncate"
rm $file
}
run_test 42d "test complete truncate of file with cached dirty data"
test_42e() { # bug22074
[ $PARALLEL == "yes" ] && skip "skip parallel run"
local TDIR=$DIR/${tdir}e
local pages=16 # hardcoded 16 pages, don't change it.
local files=$((OSTCOUNT * 500)) # hopefully 500 files on each OST
local proc_osc0="osc.${FSNAME}-OST0000-osc-[^MDT]*"
local max_dirty_mb
local warmup_files
test_mkdir $DIR/${tdir}e
$LFS setstripe -c 1 $TDIR
createmany -o $TDIR/f $files
max_dirty_mb=$($LCTL get_param -n $proc_osc0/max_dirty_mb)
# we assume that with $OSTCOUNT files, at least one of them will
# be allocated on OST0.
warmup_files=$((OSTCOUNT * max_dirty_mb))
createmany -o $TDIR/w $warmup_files
# write a large amount of data into one file and sync, to get good
# avail_grant number from OST.
for ((i=0; i<$warmup_files; i++)); do
idx=$($LFS getstripe -i $TDIR/w$i)
[ $idx -ne 0 ] && continue
dd if=/dev/zero of=$TDIR/w$i bs="$max_dirty_mb"M count=1
break
done
[[ $i -gt $warmup_files ]] && error "OST0 is still cold"
sync
$LCTL get_param $proc_osc0/cur_dirty_bytes
$LCTL get_param $proc_osc0/cur_grant_bytes
# create as much dirty pages as we can while not to trigger the actual
# RPCs directly. but depends on the env, VFS may trigger flush during this
# period, hopefully we are good.
for ((i=0; i<$warmup_files; i++)); do
idx=$($LFS getstripe -i $TDIR/w$i)
[ $idx -ne 0 ] && continue
dd if=/dev/zero of=$TDIR/w$i bs=1M count=1 2>/dev/null
done
$LCTL get_param $proc_osc0/cur_dirty_bytes
$LCTL get_param $proc_osc0/cur_grant_bytes
# perform the real test
$LCTL set_param $proc_osc0/rpc_stats 0
for ((;i<$files; i++)); do
[ $($LFS getstripe -i $TDIR/f$i) -eq 0 ] || continue
dd if=/dev/zero of=$TDIR/f$i bs=$PAGE_SIZE count=$pages 2>/dev/null
done
sync
$LCTL get_param $proc_osc0/rpc_stats
local percent=0
local have_ppr=false
$LCTL get_param $proc_osc0/rpc_stats |
while read PPR RRPC RPCT RCUM BAR WRPC WPCT WCUM; do
# skip lines until we are at the RPC histogram data
[ "$PPR" == "pages" ] && have_ppr=true && continue
$have_ppr || continue
# we only want the percent stat for < 16 pages
[[ $(echo $PPR | tr -d ':') -ge $pages ]] && break
percent=$((percent + WPCT))
if [[ $percent -gt 15 ]]; then
error "less than 16-pages write RPCs" \
"$percent% > 15%"
break
fi
done
rm -rf $TDIR
}
run_test 42e "verify sub-RPC writes are not done synchronously"
test_43A() { # was test_43
test_mkdir $DIR/$tdir
cp -p /bin/ls $DIR/$tdir/$tfile
$MULTIOP $DIR/$tdir/$tfile Ow_c &
pid=$!
# give multiop a chance to open
sleep 1
$DIR/$tdir/$tfile && error "execute $DIR/$tdir/$tfile succeeded" || true
kill -USR1 $pid
# Wait for multiop to exit
wait $pid
}
run_test 43A "execution of file opened for write should return -ETXTBSY"
test_43a() {
test_mkdir $DIR/$tdir
cp -p $(which sleep) $DIR/$tdir/sleep || error "can't copy"
$DIR/$tdir/sleep 60 &
SLEEP_PID=$!
# Make sure exec of $tdir/sleep wins race with truncate
sleep 1
$MULTIOP $DIR/$tdir/sleep Oc && error "expected error, got success"
kill $SLEEP_PID
}
run_test 43a "open(RDWR) of file being executed should return -ETXTBSY"
test_43b() {
[ $PARALLEL == "yes" ] && skip "skip parallel run"
test_mkdir $DIR/$tdir
cp -p $(which sleep) $DIR/$tdir/sleep || error "can't copy"
$DIR/$tdir/sleep 60 &
SLEEP_PID=$!
# Make sure exec of $tdir/sleep wins race with truncate
sleep 1
$TRUNCATE $DIR/$tdir/sleep 0 && error "expected error, got success"
kill $SLEEP_PID
}
run_test 43b "truncate of file being executed should return -ETXTBSY"
test_43c() {
local testdir="$DIR/$tdir"
test_mkdir $testdir
cp $SHELL $testdir/
( cd $(dirname $SHELL) && md5sum $(basename $SHELL) ) |
( cd $testdir && md5sum -c )
}
run_test 43c "md5sum of copy into lustre"
test_44A() { # was test_44
[[ $OSTCOUNT -lt 2 ]] && skip_env "needs >= 2 OSTs"
dd if=/dev/zero of=$DIR/f1 bs=4k count=1 seek=1023
dd if=$DIR/f1 bs=4k count=1 > /dev/null
}
run_test 44A "zero length read from a sparse stripe"
test_44a() {
local nstripe=$($LFS getstripe -c -d $DIR)
[ -z "$nstripe" ] && skip "can't get stripe info"
[[ $nstripe -gt $OSTCOUNT ]] &&
skip "Wrong default stripe_count: $nstripe OSTCOUNT: $OSTCOUNT"
local stride=$($LFS getstripe -S -d $DIR)
if [[ $nstripe -eq 0 || $nstripe -eq -1 ]]; then
nstripe=$($LFS df -o $DIR | wc -l)
fi
OFFSETS="0 $((stride/2)) $((stride-1))"
for offset in $OFFSETS; do
for i in $(seq 0 $((nstripe-1))); do
local GLOBALOFFSETS=""
# size in Bytes
local size=$((((i + 2 * $nstripe )*$stride + $offset)))
local myfn=$DIR/d44a-$size
echo "--------writing $myfn at $size"
ll_sparseness_write $myfn $size ||
error "ll_sparseness_write"
GLOBALOFFSETS="$GLOBALOFFSETS $size"
ll_sparseness_verify $myfn $GLOBALOFFSETS ||
error "ll_sparseness_verify $GLOBALOFFSETS"
for j in $(seq 0 $((nstripe-1))); do
# size in Bytes
size=$((((j + $nstripe )*$stride + $offset)))
ll_sparseness_write $myfn $size ||
error "ll_sparseness_write"
GLOBALOFFSETS="$GLOBALOFFSETS $size"
done
ll_sparseness_verify $myfn $GLOBALOFFSETS ||
error "ll_sparseness_verify $GLOBALOFFSETS"
rm -f $myfn
done
done
}
run_test 44a "test sparse pwrite ==============================="
test_44b() {
(( $OST1_VERSION >= $(version_code 2.15.61.137) )) ||
skip "Need OST >= 2.15.61.137 for large object handling"
$LFS setstripe -c 1 $DIR/$tfile || error "setstripe failed"
local off=$((2**32*4096-8192))
dd if=/dev/zero of=$DIR/$tfile bs=1 count=1 seek=$off conv=notrunc ||
error "dd failed"
cancel_lru_locks osc
$CHECKSTAT -s $((2**32*4096-8192+1)) $DIR/$tfile || error "wrong size"
}
run_test 44b "write one byte at offset 0xfffffffe000"
test_44c() {
(( $OST1_VERSION >= $(version_code 2.15.61.137) )) ||
skip "Need OST >= 2.15.61.137 for large object handling"
local osc_tgt="$FSNAME-OST0000-osc-$($LFS getname -i $DIR)"
local max_object_bytes=$(import_param $osc_tgt max_object_bytes)
$LFS setstripe -c 1 $DIR/$tfile || error "setstripe failed"
dd if=/dev/zero of=$DIR/$tfile conv=notrunc bs=1 count=1 \
seek=$((max_object_bytes - 1)) || error "dd failed"
cancel_lru_locks osc
$CHECKSTAT -s $max_object_bytes $DIR/$tfile || error "wrong size"
}
run_test 44c "write 1 byte at max_object_bytes - 1 offset"
test_44d() {
(( $OST1_VERSION >= $(version_code 2.15.61.137) )) ||
skip "Need OST >= 2.15.61.137 for large object handling"
local osc_tgt="$FSNAME-OST0000-osc-$($LFS getname -i $DIR)"
local max_object_bytes=$(import_param $osc_tgt max_object_bytes)
$LFS setstripe -c 2 $DIR/$tfile || error "setstripe failed"
local stripe_size=$($LFS getstripe -S $DIR/$tfile)
local off=$((max_object_bytes & ~(stripe_size - 1)))
$TRUNCATE $DIR/$tfile $off
dd if=/dev/zero of=$DIR/$tfile bs=1 count=1 seek=$off
local rc1=$?
dd if=/dev/zero of=$DIR/$tfile oflag=append conv=notrunc bs=1 count=1 \
seek=$off
local rc2=$?
[[ $rc1 -eq 0 && $rc2 -eq 0 ]] || error "one of dd commands failed"
}
run_test 44d "if write at position fails (EFBIG), so should do append"
# write file until maximal size is reached
max_file_size() {
local file=$1
off=1
minoff=1
while true; do
echo a | dd of=$1 bs=1 count=1 conv=notrunc seek=$off status=progress \
2>/dev/null
[[ $? -ne 0 ]] && break
minoff=$off
off=$(echo "$off * 2" | bc)
done
maxoff=$off
minoff_1=$(echo $off + 1 | bc)
while [[ maxoff -ne minoff_1 ]]; do
off=$(echo "($maxoff + $minoff) / 2" | bc)
echo a | dd of=$1 bs=1 count=1 conv=notrunc seek=$off status=progress \
2>/dev/null
[[ $? -eq 0 ]] && minoff=$off || maxoff=$off
minoff_1=$(echo $off + 1 | bc)
done
stat -c %s $file
}
test_44e_write_read()
{
local ifile=$1
local ofile=$2
local stripe_count=$3
local stripe_size=$($LFS getstripe -S $ofile)
local file_size=$(max_file_size $ofile)
local write_count=$((stripe_count * stripe_size))
local offset=$((file_size - write_count))
dd if=/dev/urandom of=$ifile bs=$write_count count=1 ||
error "failed to write random data"
dd if=$ifile of=$ofile bs=$write_count count=1 oflag=seek_bytes \
seek=$offset conv=notrunc || error "dd failed"
cancel_lru_locks osc
cmp $ifile $ofile -i 0:$offset -n $write_count || error "cmp failed"
}
test_44e() {
(( $OST1_VERSION >= $(version_code 2.15.61.137) )) ||
skip "Need OST >= 2.15.61.137 for large object handling"
local TF="$(mktemp --tmpdir -u $tfile.XXXXXX)"
$LFS setstripe -S 1M -c $OSTCOUNT $DIR/$tfile ||
error "lfs setstripe -S 1M -c $OSTCOUNT failed"
test_44e_write_read $TF $DIR/$tfile $OSTCOUNT
rm -f $DIR/$tfile
rm -f $TF
$LFS setstripe -S 1M -c $OSTCOUNT -C $((OSTCOUNT * 2)) $DIR/$tfile ||
error "lfs setstripe -S 1M -c $OSTCOUNT -C [* 2] failed"
test_44e_write_read $TF $DIR/$tfile $((OSTCOUNT * 2))
rm -f $DIR/$tfile
rm -f $TF
}
run_test 44e "write and read maximal stripes"
test_44f() {
[ "$FSTYPE" != "zfs" ] ||
skip "ORI-366/LU-1941: FIEMAP unimplemented on ZFS"
stack_trap "rm -f $DIR/$tfile-*"
local i
# required space: NUMFILES_44f * 20Mb
local NUMFILES_44f=${NUMFILES_44f:-5}
[[ "$SLOW" == "yes" ]] && NUMFILES_44f=50
for (( i=0; i<NUMFILES_44f; i++ )); do
multiop $DIR/$tfile-$i \
Ow10485760Z10485760w10485760Ic ||
error "multiop failed"
done
}
run_test 44f "Check fiemap for sparse files"
test_44g() {
# max stripe size is 4G - 64K
$LFS setstripe -C 4 -S 4194240k $DIR/$tfile
dd if=/dev/zero of=$DIR/$tfile conv=notrunc bs=64k count=1 \
seek=$((65535 * 3 + 65533))
$LFS setstripe -c 2 -S 2g $DIR/$tfile-2
dd if=/dev/zero of=$DIR/$tfile-2 conv=notrunc bs=64k count=1 \
seek=65535
cancel_lru_locks osc
$CHECKSTAT -s $(((65535 * 3 + 65534) * 65536)) $DIR/$tfile ||
error "wrong $DIR/$tfile size"
$CHECKSTAT -s 4294967296 $DIR/$tfile-2 ||
error "wrong $DIR/$tfile-2 size"
}
run_test 44g "test overflow in lov_stripe_size"
dirty_osc_total() {
tot=0
for d in `lctl get_param -n ${OSC}.*.cur_dirty_bytes`; do
tot=$(($tot + $d))
done
echo $tot
}
do_dirty_record() {
before=`dirty_osc_total`
echo executing "\"$*\""
eval $*
after=`dirty_osc_total`
echo before $before, after $after
}
test_45() {
[ $PARALLEL == "yes" ] && skip "skip parallel run"
f="$DIR/f45"
# Obtain grants from OST if it supports it
echo blah > ${f}_grant
stop_writeback
sync
# cur_dirty_bytes is related and modified by the OSC cache granting
# code. It is possible for the grants to be freed during async IO
# completion resulting in 0 vs 0 comparisons for truncates and sync.
# Since it can't be reliably proven that the operation did lower the
# count, the comparisons are changed to >= (ge) rather than > (gt).
#
# N.B. This whole test seems flawed as the process doesn't hold file
# descriptors, syscalls are not explicit, and dirty data is subject to
# async changes outside of the MM subsystem. Some consideration should
# be given to removing/rewriting this test.
do_dirty_record "echo blah > $f"
(( before == after )) && error "write wasn't cached"
do_dirty_record "> $f"
(( before >= after )) || error "truncate didn't lower dirty count"
do_dirty_record "echo blah > $f"
(( before == after )) && error "write wasn't cached"
do_dirty_record "sync"
(( after == 0 )) || error "writeback didn't zero dirty count"
do_dirty_record "echo blah > $f"
(( before == after )) && error "write wasn't cached"
do_dirty_record "cancel_lru_locks osc"
(( before >= after )) ||
error "lock cancellation didn't lower dirty count"
start_writeback
}
run_test 45 "osc io page accounting"
# in a 2 stripe file (lov.sh), page 1023 maps to page 511 in its object. this
# test tickles a bug where re-dirtying a page was failing to be mapped to the
# objects offset and an assert hit when an rpc was built with 1023's mapped
# offset 511 and 511's raw 511 offset. it also found general redirtying bugs.
test_46() {
[ $PARALLEL == "yes" ] && skip "skip parallel run"
f="$DIR/f46"
stop_writeback
sync
dd if=/dev/zero of=$f bs=$PAGE_SIZE seek=511 count=1
sync
dd conv=notrunc if=/dev/zero of=$f bs=$PAGE_SIZE seek=1023 count=1
dd conv=notrunc if=/dev/zero of=$f bs=$PAGE_SIZE seek=511 count=1
sync
start_writeback
}
run_test 46 "dirtying a previously written page ================"
# test_47 is removed "Device nodes check" is moved to test_28
test_48a() { # bug 2399
[ "$mds1_FSTYPE" = "zfs" ] &&
[ $MDS1_VERSION -lt $(version_code 2.3.63) ] &&
skip "MDS prior to 2.3.63 handle ZFS dir .. incorrectly"
test_mkdir $DIR/$tdir
cd $DIR/$tdir
mv $DIR/$tdir $DIR/$tdir.new || error "move directory failed"
test_mkdir $DIR/$tdir
touch foo || error "'touch foo' failed after recreating cwd"
test_mkdir bar
touch .foo || error "'touch .foo' failed after recreating cwd"
test_mkdir .bar
ls . > /dev/null || error "'ls .' failed after recreating cwd"
ls .. > /dev/null || error "'ls ..' failed after removing cwd"
cd . || error "'cd .' failed after recreating cwd"
mkdir . && error "'mkdir .' worked after recreating cwd"
rmdir . && error "'rmdir .' worked after recreating cwd"
ln -s . baz || error "'ln -s .' failed after recreating cwd"
cd .. || error "'cd ..' failed after recreating cwd"
}
run_test 48a "Access renamed working dir (should return errors)="
test_48b() { # bug 2399
rm -rf $DIR/$tdir
test_mkdir $DIR/$tdir
cd $DIR/$tdir
rmdir $DIR/$tdir || error "remove cwd $DIR/$tdir failed"
touch foo && error "'touch foo' worked after removing cwd"
mkdir foo && error "'mkdir foo' worked after removing cwd"
touch .foo && error "'touch .foo' worked after removing cwd"
mkdir .foo && error "'mkdir .foo' worked after removing cwd"
ls . > /dev/null && error "'ls .' worked after removing cwd"
ls .. > /dev/null || error "'ls ..' failed after removing cwd"
mkdir . && error "'mkdir .' worked after removing cwd"
rmdir . && error "'rmdir .' worked after removing cwd"
ln -s . foo && error "'ln -s .' worked after removing cwd"
cd .. || echo "'cd ..' failed after removing cwd `pwd`" #bug 3517
}
run_test 48b "Access removed working dir (should return errors)="
test_48c() { # bug 2350
#lctl set_param debug=-1
#set -vx
rm -rf $DIR/$tdir
test_mkdir -p $DIR/$tdir/dir
cd $DIR/$tdir/dir
$TRACE rmdir $DIR/$tdir/dir || error "remove cwd $DIR/$tdir/dir failed"
$TRACE touch foo && error "touch foo worked after removing cwd"
$TRACE mkdir foo && error "'mkdir foo' worked after removing cwd"
touch .foo && error "touch .foo worked after removing cwd"
mkdir .foo && error "mkdir .foo worked after removing cwd"
$TRACE ls . && error "'ls .' worked after removing cwd"
$TRACE ls .. || error "'ls ..' failed after removing cwd"
$TRACE mkdir . && error "'mkdir .' worked after removing cwd"
$TRACE rmdir . && error "'rmdir .' worked after removing cwd"
$TRACE ln -s . foo && error "'ln -s .' worked after removing cwd"
$TRACE cd .. || echo "'cd ..' failed after removing cwd `pwd`" #bug 3415
}
run_test 48c "Access removed working subdir (should return errors)"
test_48d() { # bug 2350
#lctl set_param debug=-1
#set -vx
rm -rf $DIR/$tdir
test_mkdir -p $DIR/$tdir/dir
cd $DIR/$tdir/dir
$TRACE rmdir $DIR/$tdir/dir || error "remove cwd $DIR/$tdir/dir failed"
$TRACE rmdir $DIR/$tdir || error "remove parent $DIR/$tdir failed"
$TRACE touch foo && error "'touch foo' worked after removing parent"
$TRACE mkdir foo && error "mkdir foo worked after removing parent"
touch .foo && error "'touch .foo' worked after removing parent"
mkdir .foo && error "mkdir .foo worked after removing parent"
$TRACE ls . && error "'ls .' worked after removing parent"
$TRACE ls .. && error "'ls ..' worked after removing parent"
$TRACE mkdir . && error "'mkdir .' worked after removing parent"
$TRACE rmdir . && error "'rmdir .' worked after removing parent"
$TRACE ln -s . foo && error "'ln -s .' worked after removing parent"
true
}
run_test 48d "Access removed parent subdir (should return errors)"
test_48e() { # bug 4134
#lctl set_param debug=-1
#set -vx
rm -rf $DIR/$tdir
test_mkdir -p $DIR/$tdir/dir
cd $DIR/$tdir/dir
$TRACE rmdir $DIR/$tdir/dir || error "remove cwd $DIR/$tdir/dir failed"
$TRACE rmdir $DIR/$tdir || error "remove parent $DIR/$tdir failed"
$TRACE touch $DIR/$tdir || error "'touch $DIR/$tdir' failed"
$TRACE chmod +x $DIR/$tdir || error "'chmod +x $DIR/$tdir' failed"
# On a buggy kernel addition of "touch foo" after cd .. will
# produce kernel oops in lookup_hash_it
touch ../foo && error "'cd ..' worked after recreate parent"
cd $DIR
$TRACE rm $DIR/$tdir || error "rm '$DIR/$tdir' failed"
}
run_test 48e "Access to recreated parent subdir (should return errors)"
test_48f() {
[[ $MDS1_VERSION -ge $(version_code 2.13.55) ]] ||
skip "need MDS >= 2.13.55"
[[ $MDSCOUNT -ge 2 ]] || skip "needs >= 2 MDTs"
[[ "$(facet_host mds1)" != "$(facet_host mds2)" ]] ||
skip "needs different host for mdt1 mdt2"
[[ $(facet_fstype mds1) == ldiskfs ]] || skip "ldiskfs only"
$LFS mkdir -i0 $DIR/$tdir
$LFS mkdir -i 1 $DIR/$tdir/sub1 $DIR/$tdir/sub2 $DIR/$tdir/sub3
for d in sub1 sub2 sub3; do
#define OBD_FAIL_OSD_REF_DEL 0x19c
do_facet mds1 $LCTL set_param fail_loc=0x8000019c
rm -rf $DIR/$tdir/$d && error "rm $d should fail"
done
rm -d --interactive=never $DIR/$tdir || error "rm $tdir fail"
}
run_test 48f "non-zero nlink dir unlink won't LBUG()"
test_49a() { # LU-1030 - was test_49
[[ $PARALLEL == "yes" ]] && skip "skip parallel run"
remote_ost_nodsh && skip "remote OST with nodsh"
# get ost1 size - $FSNAME-OST0000
ost1_size=$($LFS df --ost=0 --output btotal)
# write 800M at maximum
(( $ost1_size < 2 )) && ost1_size=2
(( $ost1_size > 819200 )) && ost1_size=819200
$LFS setstripe -c 1 -i 0 $DIR/$tfile
dd if=/dev/zero of=$DIR/$tfile bs=4k count=$((ost1_size >> 2)) &
local dd_pid=$!
# change max_pages_per_rpc while writing the file
local osc1_mppc=osc.$(get_osc_import_name client ost1).max_pages_per_rpc
local orig_mppc=$($LCTL get_param -n $osc1_mppc)
# loop until dd process exits
while ps ax -opid | grep -wq $dd_pid; do
$LCTL set_param $osc1_mppc=$((RANDOM % 256 + 1))
sleep $((RANDOM % 5 + 1))
done
# restore original max_pages_per_rpc
$LCTL set_param $osc1_mppc=$orig_mppc
rm $DIR/$tfile || error "rm $DIR/$tfile failed"
}
run_test 49a "Change max_pages_per_rpc won't break osc extent"
test_49b() {
local tgt="$FSNAME-OST0000-osc-*"
local max_pages_per_rpc="osc.$tgt.max_pages_per_rpc"
local max_mb_per_rpc_read="osc.$tgt.max_mb_per_rpc_read"
local max_mb_per_rpc_write="osc.$tgt.max_mb_per_rpc_write"
local old_brw=$(import_param "$tgt" max_brw_size)
local old_pages=$($LCTL get_param -n "$max_mb_per_rpc_read")
stack_trap "$LCTL set_param $max_mb_per_rpc_read=$old_pages"
local old_pages=$($LCTL get_param -n "$max_mb_per_rpc_write")
stack_trap "$LCTL set_param $max_mb_per_rpc_write=$old_pages"
local expect="1"
$LCTL set_param "$max_pages_per_rpc=1M"
local actual=$($LCTL get_param -n "$max_mb_per_rpc_read")
[[ "$expect" == "$actual" ]] ||
error "wrong max_mb_per_rpc_read: $expect != $actual"
local actual=$($LCTL get_param -n "$max_mb_per_rpc_write")
[[ "$expect" == "$actual" ]] ||
error "wrong max_mb_per_rpc_write: $expect != $actual"
if [ "$ost1_FSTYPE" == ldiskfs ]; then
expect="0.500"
$LCTL set_param "$max_pages_per_rpc=512K"
actual=$($LCTL get_param -n "$max_mb_per_rpc_read")
[[ "$expect" == "$actual" ]] ||
error "wrong max_mb_per_rpc_read: $expect != $actual"
actual=$($LCTL get_param -n "$max_mb_per_rpc_write")
[[ "$expect" == "$actual" ]] ||
error "wrong max_mb_per_rpc_write: $expect != $actual"
expect="3.500"
$LCTL set_param "$max_pages_per_rpc=3.5M"
actual=$($LCTL get_param -n "$max_mb_per_rpc_read")
[[ "$expect" == "$actual" ]] ||
error "wrong max_mb_per_rpc_read: $expect != $actual"
actual=$($LCTL get_param -n "$max_mb_per_rpc_write")
[[ "$expect" == "$actual" ]] ||
error "wrong max_mb_per_rpc_write: $expect != $actual"
expect="4"
$LCTL set_param "$max_pages_per_rpc=0.00390625G"
actual=$($LCTL get_param -n "$max_mb_per_rpc_read")
[[ "$expect" == "$actual" ]] ||
error "wrong max_mb_per_rpc_read: $expect != $actual"
actual=$($LCTL get_param -n "$max_mb_per_rpc_write")
[[ "$expect" == "$actual" ]] ||
error "wrong max_mb_per_rpc_write: $expect != $actual"
fi
}
run_test 49b "verify max_mb_per_rpc_read/write after setting max_pages_per_rpc"
test_49c() {
local tgt="$FSNAME-OST0000-osc-*"
local max_pages_per_rpc="osc.$tgt.max_pages_per_rpc"
local max_mb_per_rpc_read="osc.$tgt.max_mb_per_rpc_read"
local max_mb_per_rpc_write="osc.$tgt.max_mb_per_rpc_write"
local old_brw=$(import_param "$tgt" max_brw_size)
local old_pages=$($LCTL get_param -n "$max_mb_per_rpc_read")
stack_trap "$LCTL set_param $max_mb_per_rpc_read=$old_pages"
local old_pages=$($LCTL get_param -n "$max_mb_per_rpc_write")
stack_trap "$LCTL set_param $max_mb_per_rpc_write=$old_pages"
local expect="1"
$LCTL set_param "$max_mb_per_rpc_read=1"
local actual=$($LCTL get_param -n "$max_mb_per_rpc_read")
[[ "$expect" == "$actual" ]] ||
error "wrong max_mb_per_rpc_read: $expect != $actual"
$LCTL set_param "$max_mb_per_rpc_write=1"
local actual=$($LCTL get_param -n "$max_mb_per_rpc_write")
[[ "$expect" == "$actual" ]] ||
error "wrong max_mb_per_rpc_write: $expect != $actual"
if [ "$ost1_FSTYPE" == ldiskfs ]; then
local expect_r="2.500"
local expect_w="0.500"
$LCTL set_param "$max_mb_per_rpc_read=2560K"
$LCTL set_param "$max_mb_per_rpc_write=0.5M"
actual=$($LCTL get_param -n "$max_mb_per_rpc_read")
[[ "$expect_r" == "$actual" ]] ||
error "wrong max_mb_per_rpc_read: $expect_r != $actual"
actual=$($LCTL get_param -n "$max_mb_per_rpc_write")
[[ "$expect_w" == "$actual" ]] ||
error "wrong max_mb_per_rpc_write: $expect_w != $actual"
local expect_r="0.500"
local expect_w="2.500"
$LCTL set_param "$max_mb_per_rpc_read=0.5M"
$LCTL set_param "$max_mb_per_rpc_write=2560K"
actual=$($LCTL get_param -n "$max_mb_per_rpc_read")
[[ "$expect_r" == "$actual" ]] ||
error "wrong max_mb_per_rpc_read: $expect_r != $actual"
actual=$($LCTL get_param -n "$max_mb_per_rpc_write")
[[ "$expect_w" == "$actual" ]] ||
error "wrong max_mb_per_rpc_write: $expect_w != $actual"
fi
}
run_test 49c "verify max_mb_per_rpc_read/write"
test_50() {
# bug 1485
test_mkdir $DIR/$tdir
cd $DIR/$tdir
ls /proc/$$/cwd || error "ls /proc/$$/cwd failed"
}
run_test 50 "special situations: /proc symlinks ==============="
test_51a() { # was test_51
# bug 1516 - create an empty entry right after ".." then split dir
test_mkdir -c1 $DIR/$tdir
touch $DIR/$tdir/foo
$MCREATE $DIR/$tdir/bar
rm $DIR/$tdir/foo
createmany -m $DIR/$tdir/longfile 201
FNUM=202
while [[ $(ls -sd $DIR/$tdir | awk '{ print $1 }') -eq 4 ]]; do
$MCREATE $DIR/$tdir/longfile$FNUM
FNUM=$(($FNUM + 1))
echo -n "+"
done
echo
ls -l $DIR/$tdir > /dev/null || error "ls -l $DIR/$tdir failed"
}
run_test 51a "special situations: split htree with empty entry =="
cleanup_print_lfs_df () {
$LFS df
$LFS df -i
}
test_51b() {
local dir=$DIR/$tdir
local nrdirs=$((65536 + 100))
# cleanup the directory
rm -fr $dir
mkdir_on_mdt -i $((RANDOM % MDSCOUNT)) $dir
$LFS df
$LFS df -i
local mdtname=$(mdtname_from_index $($LFS getstripe -m $dir))
local nproc=2 #$((lctl get_param -n mdc.$mdtname*.max_rpcs_in_flight/2))
local numfree=$($LCTL get_param -n mdc.$mdtname*.filesfree)
(( $numfree >= $nrdirs )) ||
skip "not enough free inodes ($numfree) on $mdtname"
# need to check free space for the directories as well
local blkfree=$($LCTL get_param -n mdc.$mdtname*.kbytesavail)
numfree=$(( blkfree / $(fs_inode_ksize) ))
# round $nrdirs to an even multiple of the number of processes
nrdirs=$((nrdirs / nproc * nproc))
(( $numfree >= $nrdirs )) || skip "not enough blocks ($numfree)"
stack_trap cleanup_print_lfs_df
# create files
local nrdirs_pp=$((nrdirs / nproc))
for ((i = 0; i < $nproc; i++)); do
stack_trap "unlinkmany -d $dir/d.$i. $nrdirs_pp || true"
createmany -d $dir/d.$i. $nrdirs_pp &
done
wait || error "failed to create $nrdirs subdirs in $mdtname:/$dir"
# really created :
nrdirs=$(find $dir -maxdepth 1 | wc -l)
# unlink all but ~100 subdirectories, then check it still works
local left=$((100 / nproc * nproc))
local delete=$((nrdirs - left))
$LFS df
$LFS df -i
# for ldiskfs the nlink count should be 1, but this is OSD specific
# and so this is listed for informational purposes only
echo "nlink before: $(stat -c %h $dir), created before: $nrdirs"
local delete_pp=$((delete / nproc))
for ((i = 0; i < $nproc; i++)); do
unlinkmany -d $dir/d.$i. $delete_pp &
done
wait || error "unlink of first $delete subdirs failed"
echo "nlink between: $(stat -c %h $dir)"
local found=$(ls -U $dir | wc -l)
(( $found == $left )) ||
error "can't find subdirs: found only $found, expected $left"
local left_pp=$((left / nproc))
for ((i = 0; i < $nproc; i++)); do
unlinkmany -d $dir/d.$i. $delete_pp $left_pp &
done
wait || error "unlink of second $left subdirs failed"
# regardless of whether the backing filesystem tracks nlink accurately
# or not, the nlink count shouldn't be more than "." and ".." here
local after=$(stat -c %h $dir)
(( $after > 2 )) && error "nlink after: $after > 2" ||
echo "nlink after: $after"
}
run_test 51b "exceed 64k subdirectory nlink limit on create, verify unlink"
test_51c() {
(( MDSCOUNT > 1 )) || skip "needs >= 2 MDTs"
local dir=$DIR/$tdir-c2
local nrdirs=$((65536 * 2 + 2000))
local nproc=$MDSCOUNT
local mdtidx
stack_trap cleanup_print_lfs_df
$LFS mkdir -c 2 -H fnv_1a_64 $dir
while read mdtidx rest; do
local mdtname=$(mdtname_from_index $mdtidx)
local numfree=$(lctl get_param -n mdc.$mdtname*.filesfree)
local blkfree=$(lctl get_param -n mdc.$mdtname*.kbytesavail)
((numfree > nrdirs/2 && blkfree/$(fs_inode_ksize) > nrdirs/2))||
skip "not enough inodes or blocks for $mdtname"
done < <($LFS getdirstripe $dir |
awk '{ if ($2 ~ /\[0x.*:0x.*:0x.*\]/) { print $1; } }' )
# round $nrdirs to an even multiple of the number of processes
nrdirs=$((nrdirs / nproc * nproc))
local nrdirs_pp=$((nrdirs / nproc))
for ((i = 0; i < $nproc; i++)); do
stack_trap "unlinkmany -d $dir/d.$i. $nrdirs_pp || true"
createmany -d $dir/d.$i. $nrdirs_pp &
done
wait || error "failed to create $nrdirs subdirs in $dir"
# nlink for the striped dir should be either 1
# (ldiskfs, nlink is overflowed at least in one stripe)
# or $ndirs + 2
local nlinks=$(stat -c %h $dir)
(( nlinks == 1 || nlinks == nrdirs + 2 )) ||
error "Wrong nlink count of $nlinks"
for ((i = 0; i < $nproc; i++)); do
unlinkmany -d $dir/d.$i. $nrdirs_pp &
done
wait || error "Removal of $nrdirs subdirs failed"
rmdir $dir && echo "removed emtpy $dir" || error "rmdir failed"
}
run_test 51c "exceed 64k subdirectory count per dir stripe, verify nlink count"
test_51d_sub() {
local stripecount=$1
local nfiles=$2
log "create files with stripecount=$stripecount"
$LFS setstripe -C $stripecount $DIR/$tdir
createmany -o $DIR/$tdir/t- $nfiles
$LFS getstripe $DIR/$tdir > $TMP/$tfile
for ((n = 0; n < $OSTCOUNT; n++)); do
objs[$n]=$(awk -vobjs=0 '($1 == '$n') { objs += 1 } \
END { printf("%0.0f", objs) }' $TMP/$tfile)
objs0[$n]=$(grep -A 1 idx $TMP/$tfile | awk -vobjs=0 \
'($1 == '$n') { objs += 1 } \
END { printf("%0.0f", objs) }')
log "OST$n has ${objs[$n]} objects, ${objs0[$n]} are index 0"
done
unlinkmany $DIR/$tdir/t- $nfiles
rm -f $TMP/$tfile
# if running in a loop, don't run out of inodes on these OSTs
(( $ONLY_REPEAT_ITER % 5 < 4 )) || {
$LFS df
$LFS df -i
wait_delete_completed
sleep 10
}
local nlast
local min=4
local max=6 # allow variance of (1 - $min/$max) = 33% by default
# For some combinations of stripecount and OSTCOUNT current code
# is not ideal, and allocates 50% fewer *first* objects to some OSTs
# than others. Rather than skipping this test entirely, check that
# and keep testing to ensure imbalance does not get worse. LU-15282
(( (OSTCOUNT == 6 && stripecount == 4) ||
(OSTCOUNT == 10 && (stripecount == 4 || stripecount == 8)) ||
(OSTCOUNT == 12 && (stripecount == 8 || stripecount == 9)))) && max=9
for ((nlast=0, n = 1; n < $OSTCOUNT; nlast=n,n++)); do
(( ${objs[$n]} > ${objs[$nlast]} * 4 / 5 )) ||
{ $LFS df && $LFS df -i &&
error "stripecount=$stripecount: " \
"OST $n has fewer objects vs. OST $nlast " \
"(${objs[$n]} < ${objs[$nlast]} x 4/5)"; }
(( ${objs[$n]} < ${objs[$nlast]} * 5 / 4 )) ||
{ $LFS df && $LFS df -i &&
error "stripecount=$stripecount: " \
"OST $n has more objects vs. OST $nlast " \
"(${objs[$n]} > ${objs[$nlast]} x 5/4)"; }
(( ${objs0[$n]} > ${objs0[$nlast]} * $min / $max )) ||
{ $LFS df && $LFS df -i &&
error "stripecount=$stripecount: " \
"OST $n has fewer #0 objects vs. OST $nlast " \
"(${objs0[$n]} < ${objs0[$nlast]} x $min/$max)"; }
(( ${objs0[$n]} < ${objs0[$nlast]} * $max / $min )) ||
{ $LFS df && $LFS df -i &&
error "stripecount=$stripecount: " \
"OST $n has more #0 objects vs. OST $nlast " \
"(${objs0[$n]} > ${objs0[$nlast]} x $max/$min)"; }
done
}
test_51d() {
[ $PARALLEL == "yes" ] && skip "skip parallel run"
[[ $OSTCOUNT -lt 3 ]] && skip_env "needs >= 3 OSTs"
local stripecount
local per_ost=100
local nfiles=$((per_ost * OSTCOUNT))
local mdts=$(comma_list $(mdts_nodes))
local param="osp.*.create_count"
local qos_old=$(do_facet mds1 \
"$LCTL get_param -n lod.$FSNAME-*.qos_threshold_rr" | head -n 1)
do_nodes $mdts \
"$LCTL set_param lod.$FSNAME-*.qos_threshold_rr=100"
stack_trap "do_nodes $mdts \
'$LCTL set_param lod.$FSNAME-*.qos_threshold_rr=${qos_old%%%}'"
test_mkdir $DIR/$tdir
local dirstripes=$(lfs getdirstripe -c $DIR/$tdir)
(( dirstripes > 0 )) || dirstripes=1
# Ensure enough OST objects precreated for tests to pass without
# running out of objects. This is an LOV r-r OST algorithm test,
# not an OST object precreation test.
local old=$(do_facet mds1 "$LCTL get_param -n $param" | head -n 1)
(( old >= nfiles )) ||
{
local create_count=$((nfiles * OSTCOUNT / dirstripes))
do_nodes $mdts "$LCTL set_param $param=$create_count"
stack_trap "do_nodes $mdts $LCTL set_param $param=$old"
# trigger precreation from all MDTs for all OSTs
for ((i = 0; i < $MDSCOUNT * 2; i++ )); do
$LFS setstripe -c -1 $DIR/$tdir/wide.$i
done
}
for ((stripecount = 3; stripecount <= $OSTCOUNT; stripecount++)); do
sleep 8 # allow object precreation to catch up
test_51d_sub $stripecount $nfiles
done
}
run_test 51d "check LOV round-robin OST object distribution"
test_51e() {
if [ "$mds1_FSTYPE" != ldiskfs ]; then
skip_env "ldiskfs only test"
fi
test_mkdir -c1 $DIR/$tdir
test_mkdir -c1 $DIR/$tdir/d0
touch $DIR/$tdir/d0/foo
createmany -l $DIR/$tdir/d0/foo $DIR/$tdir/d0/f- 65001 &&
error "file exceed 65000 nlink limit!"
unlinkmany $DIR/$tdir/d0/f- 65001
return 0
}
run_test 51e "check file nlink limit"
test_51f() {
test_mkdir $DIR/$tdir
local max=100000
local ulimit_old=$(ulimit -n)
local spare=20 # number of spare fd's for scripts/libraries, etc.
local mdt=$($LFS getstripe -m $DIR/$tdir)
local numfree=$($LFS df --mdt=$mdt --output=ifree $DIR/$tdir)
echo "MDT$mdt numfree=$numfree, max=$max"
[[ $numfree -gt $max ]] && numfree=$max || numfree=$((numfree * 7 / 8))
if [ $((numfree + spare)) -gt $ulimit_old ]; then
while ! ulimit -n $((numfree + spare)); do
numfree=$((numfree * 3 / 4))
done
echo "changed ulimit from $ulimit_old to $((numfree + spare))"
else
echo "left ulimit at $ulimit_old"
fi
createmany -o -k -t 120 $DIR/$tdir/f $numfree || {
unlinkmany $DIR/$tdir/f $numfree
error "create+open $numfree files in $DIR/$tdir failed"
}
ulimit -n $ulimit_old
# if createmany exits at 120s there will be fewer than $numfree files
unlinkmany $DIR/$tdir/f $numfree || true
}
run_test 51f "check many open files limit"
test_52a() {
[ -f $DIR/$tdir/foo ] && chattr -a $DIR/$tdir/foo
test_mkdir $DIR/$tdir
touch $DIR/$tdir/foo
chattr +a $DIR/$tdir/foo || error "chattr +a failed"
echo bar >> $DIR/$tdir/foo || error "append bar failed"
cp /etc/hosts $DIR/$tdir/foo && error "cp worked"
rm -f $DIR/$tdir/foo 2>/dev/null && error "rm worked"
link $DIR/$tdir/foo $DIR/$tdir/foo_link 2>/dev/null &&
error "link worked"
echo foo >> $DIR/$tdir/foo || error "append foo failed"
mrename $DIR/$tdir/foo $DIR/$tdir/foo_ren && error "rename worked"
lsattr $DIR/$tdir/foo | grep -E -q "^-+a[-e]+ $DIR/$tdir/foo" ||
error "lsattr"
chattr -a $DIR/$tdir/foo || error "chattr -a failed"
cp -r $DIR/$tdir $TMP/
rm -fr $DIR/$tdir $TMP/$tdir || error "cleanup rm failed"
}
run_test 52a "append-only flag test (should return errors)"
test_52b() {
[ -f $DIR/$tdir/foo ] && chattr -i $DIR/$tdir/foo
test_mkdir $DIR/$tdir
touch $DIR/$tdir/foo
chattr +i $DIR/$tdir/foo || error "chattr +i failed"
cat test > $DIR/$tdir/foo && error "cat test worked"
cp /etc/hosts $DIR/$tdir/foo && error "cp worked"
rm -f $DIR/$tdir/foo 2>/dev/null && error "rm worked"
link $DIR/$tdir/foo $DIR/$tdir/foo_link 2>/dev/null &&
error "link worked"
echo foo >> $DIR/$tdir/foo && error "echo worked"
mrename $DIR/$tdir/foo $DIR/$tdir/foo_ren && error "rename worked"
[ -f $DIR/$tdir/foo ] || error "$tdir/foo is not a file"
[ -f $DIR/$tdir/foo_ren ] && error "$tdir/foo_ren is not a file"
lsattr $DIR/$tdir/foo | grep -E -q "^-+i[-e]+ $DIR/$tdir/foo" ||
error "lsattr"
chattr -i $DIR/$tdir/foo || error "chattr failed"
rm -fr $DIR/$tdir || error "unable to remove $DIR/$tdir"
}
run_test 52b "immutable flag test (should return errors) ======="
test_53() {
[ $PARALLEL == "yes" ] && skip "skip parallel run"
remote_mds_nodsh && skip "remote MDS with nodsh"
remote_ost_nodsh && skip "remote OST with nodsh"
local param
local param_seq
local ostname
local mds_last
local mds_last_seq
local ost_last
local ost_last_seq
local ost_last_id
local ostnum
local node
local found=false
local support_last_seq=true
[[ $MDS1_VERSION -ge $(version_code 2.3.60) ]] ||
support_last_seq=false
# only test MDT0000
local mdtosc=$(get_mdtosc_proc_path $SINGLEMDS)
local value
for value in $(do_facet $SINGLEMDS \
$LCTL get_param osp.$mdtosc.prealloc_last_id) ; do
param=$(echo ${value[0]} | cut -d "=" -f1)
ostname=$(echo $param | cut -d "." -f2 | cut -d - -f 1-2)
if $support_last_seq; then
param_seq=$(echo $param |
sed -e s/prealloc_last_id/prealloc_last_seq/g)
mds_last_seq=$(do_facet $SINGLEMDS \
$LCTL get_param -n $param_seq)
fi
mds_last=$(do_facet $SINGLEMDS $LCTL get_param -n $param)
ostnum=$(index_from_ostuuid ${ostname}_UUID)
node=$(facet_active_host ost$((ostnum+1)))
param="obdfilter.$ostname.last_id"
for ost_last in $(do_node $node $LCTL get_param -n $param) ; do
echo "$ostname.last_id=$ost_last; MDS.last_id=$mds_last"
ost_last_id=$ost_last
if $support_last_seq; then
ost_last_id=$(echo $ost_last |
awk -F':' '{print $2}' |
sed -e "s/^0x//g")
ost_last_seq=$(echo $ost_last |
awk -F':' '{print $1}')
[[ $ost_last_seq = $mds_last_seq ]] || continue
fi
if [[ $ost_last_id != $mds_last ]]; then
error "$ost_last_id != $mds_last"
else
found=true
break
fi
done
done
$found || error "can not match last_seq/last_id for $mdtosc"
return 0
}
run_test 53 "verify that MDS and OSTs agree on pre-creation ===="
test_54a() {
$SOCKETSERVER $DIR/socket ||
error "$SOCKETSERVER $DIR/socket failed: $?"
$SOCKETCLIENT $DIR/socket ||
error "$SOCKETCLIENT $DIR/socket failed: $?"
unlink $DIR/socket || error "unlink $DIR/socket failed: $?"
}
run_test 54a "unix domain socket test"
test_54b() {
f="$DIR/f54b"
mknod $f c 1 3
chmod 0666 $f
dd if=/dev/zero of=$f bs=$PAGE_SIZE count=1
}
run_test 54b "char device works in lustre ======================"
find_loop_dev() {
[ -b /dev/loop/0 ] && LOOPBASE=/dev/loop/
[ -b /dev/loop0 ] && LOOPBASE=/dev/loop
[ -z "$LOOPBASE" ] && echo "/dev/loop/0 and /dev/loop0 gone?" && return
for i in $(seq 3 7); do
losetup $LOOPBASE$i > /dev/null 2>&1 && continue
LOOPDEV=$LOOPBASE$i
LOOPNUM=$i
break
done
}
cleanup_54c() {
local rc=0
loopdev="$DIR/loop54c"
trap 0
$UMOUNT $DIR/$tdir || rc=$?
losetup -d $loopdev || true
losetup -d $LOOPDEV || true
rm -rf $loopdev $DIR/$tfile $DIR/$tdir
return $rc
}
test_54c() {
[ $PARALLEL == "yes" ] && skip "skip parallel run"
loopdev="$DIR/loop54c"
find_loop_dev
[ -z "$LOOPNUM" ] && skip_env "couldn't find empty loop device"
trap cleanup_54c EXIT
mknod $loopdev b 7 $LOOPNUM
echo "make a loop file system with $DIR/$tfile on $loopdev ($LOOPNUM)."
dd if=/dev/zero of=$DIR/$tfile bs=$PAGE_SIZE seek=1024 count=1 > /dev/null
losetup $loopdev $DIR/$tfile ||
error "can't set up $loopdev for $DIR/$tfile"
mkfs.ext2 $loopdev || error "mke2fs on $loopdev"
test_mkdir $DIR/$tdir
mount -t ext2 $loopdev $DIR/$tdir ||
error "error mounting $loopdev on $DIR/$tdir"
dd if=/dev/zero of=$DIR/$tdir/tmp bs=$PAGE_SIZE count=30 ||
error "dd write"
df $DIR/$tdir
dd if=$DIR/$tdir/tmp of=/dev/zero bs=$PAGE_SIZE count=30 ||
error "dd read"
cleanup_54c
}
run_test 54c "block device works in lustre ====================="
test_54d() {
local pipe="$DIR/$tfile.pipe"
local string="aaaaaa"
mknod $pipe p
echo -n "$string" > $pipe &
local result=$(cat $pipe)
[[ "$result" == "$string" ]] || error "$result != $string"
}
run_test 54d "fifo device works in lustre ======================"
test_54e() {
f="$DIR/f54e"
string="aaaaaa"
cp -aL /dev/console $f
echo $string > $f || error "echo $string to $f failed"
}
run_test 54e "console/tty device works in lustre ======================"
test_55a() {
load_module kunit/obd_test verbose=2 || error "load_module failed"
# This must be run in iteractive mode, since attach and setup
# are stateful
eval "$LCTL <<-EOF || error 'OBD device creation failed'
attach obd_test obd_name obd_uuid
setup obd_test
EOF"
echo "Devices:"
$LCTL dl | tail -n 10
$LCTL --device "obd_name" cleanup
$LCTL --device "obd_name" detach
dmesg | tail -n 25 | grep "Lustre: OBD:.*FAIL" &&
error "OBD unit test failed"
rmmod -v obd_test ||
error "rmmod failed (may trigger a failure in a later test)"
}
run_test 55a "OBD device life cycle unit tests"
test_55b() {
local dev_count="$($LCTL dl | wc -l)"
# Set up a large number of devices, using the number
# that can be set up in about a minute (based on prior
# testing). We don't want to run this test forever.
local num_dev_to_create="$(( 24000 - $dev_count))"
load_module kunit/obd_test || error "load_module failed"
local start=$SECONDS
# This must be run in iteractive mode, since attach and setup
# are stateful
for ((i = 1; i <= num_dev_to_create; i++)); do
echo "attach obd_test obd_name_$i obd_uuid_$i"
echo "setup obd_test_$i"
done | $LCTL || error "OBD device creation failed"
echo "Load time: $((SECONDS - start))"
echo "Devices:"
$LCTL dl | tail -n 10
for ((i = 1; i <= num_dev_to_create; i++)); do
echo "--device obd_name_$i cleanup"
echo "--device obd_name_$i detach"
done | $LCTL || error "OBD device cleanup failed"
echo "Unload time: $((SECONDS - start))"
rmmod -v obd_test ||
error "rmmod failed (may trigger a failure in a later test)"
}
run_test 55b "Load and unload max OBD devices"
test_55c()
{
load_module kunit/obd_mod_rpcs_test ||
error "load_module obd_mod_rpcs_test failed"
dmesg | tail -n 25 | grep "obd_mod_rpcs_test"
if [ $(dmesg | tail -n 25 | grep "Force grant RPC"| wc -l) -gt 1 ]; then
error "Force granted more than once, failing."
else
echo "Force granted exactly once. Passed."
fi
rmmod -v obd_mod_rpcs_test ||
error "rmmod failed (may trigger a failure in a later test)"
}
run_test 55c "obd_mod_rpcs_test"
test_56a() {
local numfiles=3
local numdirs=2
local dir=$DIR/$tdir
rm -rf $dir
test_mkdir -p $dir/dir
for i in $(seq $numfiles); do
touch $dir/file$i
touch $dir/dir/file$i
done
local numcomp=$($LFS getstripe --component-count $dir)
[[ $numcomp == 0 ]] && numcomp=1
# test lfs getstripe with --recursive
local filenum=$($LFS getstripe -r $dir | grep -E -c "obdidx|l_ost_idx")
[[ $filenum -eq $((numfiles * 2)) ]] ||
error "$LFS getstripe -r: found $filenum != $((numfiles * 2))"
filenum=$($LFS getstripe $dir | grep -E -c "obdidx|l_ost_idx")
[[ $filenum -eq $numfiles ]] ||
error "$LFS getstripe $dir: found $filenum, not $numfiles"
echo "$LFS getstripe showed obdidx or l_ost_idx"
# test lfs getstripe with file instead of dir
filenum=$($LFS getstripe $dir/file1 | grep -E -c "obdidx|l_ost_idx")
[[ $filenum -eq 1 ]] ||
error "$LFS getstripe $dir/file1: found $filenum, not 1"
echo "$LFS getstripe file1 passed"
#test lfs getstripe with --verbose
filenum=$($LFS getstripe --verbose $dir | grep -c lmm_magic)
[[ $filenum -eq $((numfiles * numcomp)) ]] ||
error "$LFS getstripe --verbose $dir: "\
"got $filenum want $((numfiles * numcomp)) lmm_magic"
[[ $($LFS getstripe $dir | grep -c lmm_magic) -eq 0 ]] ||
error "$LFS getstripe $dir: showed lmm_magic"
#test lfs getstripe with -v prints lmm_fid
filenum=$($LFS getstripe -v $dir | grep -c lmm_fid)
local countfids=$(((numdirs + numfiles) * numcomp))
[[ $filenum -eq $countfids ]] ||
error "$LFS getstripe -v $dir: "\
"got $filenum want $countfids lmm_fid"
[[ $($LFS getstripe $dir | grep -c lmm_fid) -eq 0 ]] ||
error "$LFS getstripe $dir: showed lmm_fid by default"
echo "$LFS getstripe --verbose passed"
#check for FID information
local fid1=$($LFS getstripe --fid $dir/file1)
local fid2=$($LFS getstripe --verbose $dir/file1 |
awk '/lmm_fid: / { print $2; exit; }')
local fid3=$($LFS path2fid $dir/file1)
[ "$fid1" != "$fid2" ] &&
error "getstripe --fid '$fid1' != getstripe --verbose '$fid2'"
[ "$fid1" != "$fid3" ] &&
error "getstripe --fid '$fid1' != lfs path2fid '$fid3'"
echo "$LFS getstripe --fid passed"
#test lfs getstripe with --obd
$LFS getstripe --obd wrong_uuid $dir 2>&1 | grep -q "unknown obduuid" ||
error "$LFS getstripe --obd wrong_uuid: should return error"
[[ $OSTCOUNT -lt 2 ]] && skip_env "needs >= 2 OSTs"
local ostidx=1
local obduuid=$(ostuuid_from_index $ostidx)
local found=$($LFS getstripe -r --obd $obduuid $dir |
grep 'lmm_stripe_offset:' | grep -c " $ostidx\$")
filenum=$($LFS getstripe -ir $dir | grep -c "^$ostidx\$")
[[ $($LFS getstripe -id $dir) -ne $ostidx ]] ||
((filenum--))
[[ $($LFS getstripe -id $dir/dir) -ne $ostidx ]] ||
((filenum--))
[[ $found -eq $filenum ]] ||
error "$LFS getstripe --obd: found $found expect $filenum"
[[ $($LFS getstripe -r -v --obd $obduuid $dir |
sed '/^[ ]*'${ostidx}'[ ]/d' |
sed -n '/^[ ]*[0-9][0-9]*[ ]/p' | wc -l) -eq 0 ]] ||
error "$LFS getstripe --obd: should not show file on other obd"
echo "$LFS getstripe --obd passed"
}
run_test 56a "check $LFS getstripe"
test_56b() {
local dir=$DIR/$tdir
local numdirs=3
test_mkdir $dir
for i in $(seq $numdirs); do
test_mkdir $dir/dir$i
done
# test lfs getdirstripe default mode is non-recursion, which is
# different from lfs getstripe
local dircnt=$($LFS getdirstripe $dir | grep -c lmv_stripe_count)
[[ $dircnt -eq 1 ]] ||
error "$LFS getdirstripe: found $dircnt, not 1"
dircnt=$($LFS getdirstripe --recursive $dir |
grep -c lmv_stripe_count)
[[ $dircnt -eq $((numdirs + 1)) ]] ||
error "$LFS getdirstripe -r: $dircnt, != $((numdirs + 1))"
}
run_test 56b "check $LFS getdirstripe"
test_56bb() {
verify_yaml_available || skip_env "YAML verification not installed"
local output_file=$DIR/$tfile.out
$LFS getdirstripe -v -D -y $DIR 1> $output_file
cat $output_file
verify_yaml < $output_file ||
error "default layout is not valid YAML"
# check non-default 1 stripe case
test_mkdir -c 1 $DIR/$tdir
$LFS getdirstripe -v -y $DIR 1> $output_file
cat $output_file
verify_yaml < $output_file || error "dir layout not valid YAML"
}
run_test 56bb "check $LFS getdirstripe layout is YAML"
test_56bc() {
(( $MDSCOUNT >= 2 )) || skip "need >= 2 MDTs"
verify_yaml_available || skip_env "YAML verification not installed"
local src_dir=$DIR/$tdir.src
local yaml_file=$DIR/$tfile.yaml
$LFS setdirstripe -c 4 $src_dir ||
error "failed to setstripe"
$LFS getdirstripe --yaml -v $src_dir > $yaml_file ||
error "failed to getdirstripe --yaml -v"
cat $yaml_file
verify_yaml < $yaml_file ||
error "striped dir layout is not valid YAML"
local orig=$(get_dir_layout_param $src_dir)
local rest=$(cat $yaml_file | parse_layout_param)
[[ "$orig" == "$rest" ]] ||
error "failed to parse YAML layout"
}
run_test 56bc "check '$LFS getdirstripe --yaml' params are valid"
test_56bd() {
(( $MDSCOUNT >= 2 )) || skip "need >= 2 MDTs"
local dir1=$DIR/$tdir.dir1
local dir2=$DIR/$tdir.dir2
$LFS setdirstripe -c 1 $dir1 || error "failed to setstripe"
$LFS setdirstripe -c 2 $dir2 || error "failed to setstripe"
$LFS getdirstripe $dir1 | grep -q "mdtidx" &&
error "getdirstripe printed component header when it should not have"
$LFS getdirstripe $dir2 | grep -q "mdtidx" ||
error "getdirstripe did not print component header when it should have"
}
run_test 56bd "check '$LFS getdirstripe' output contains header only when needed"
test_56c() {
remote_ost_nodsh && skip "remote OST with nodsh"
local ost_idx=0
local ost_name=$(ostname_from_index $ost_idx)
local old_status=$(ost_dev_status $ost_idx $MOUNT -v)
local inst=$(lfs getname -i $DIR)
local state="osc.$ost_name-osc-$inst.statfs_state"
local old_state=$($LCTL get_param -n $state)
local p="$TMP/$TESTSUITE-$TESTNAME.parameters"
[[ "$old_status" == "$old_state" ]] ||
error "old status '$old_status' != state '$old_state'"
[[ -z "$old_status" || "$old_status" == "f" ]] ||
skip_env "OST $ost_name is in $old_status status"
local degraded="obdfilter.$ost_name.degraded"
save_lustre_params ost1 $degraded > $p
do_facet ost1 $LCTL set_param -n $degraded=1
stack_trap "restore_lustre_params < $p; rm -f $p"
if (( $OST1_VERSION >= $(version_code 2.12.55) )); then
local no_precreate="obdfilter.$ost_name.no_precreate"
save_lustre_params ost1 $no_precreate >> $p
do_facet ost1 $LCTL set_param -n $no_precreate=1
fi
if (( $OST1_VERSION >= $(version_code 2.12.57) )) &&
[[ ! "$old_status" =~ "f" ]]; then
local nonrotational="osd-*.$ost_name.nonrotational"
save_lustre_params ost1 $nonrotational >> $p
do_facet ost1 $LCTL set_param -n $nonrotational=1
fi
[[ ! $($LFS df -v $MOUNT 2>&1) =~ "inactive device" ]] ||
error "$LFS df -v showing inactive devices"
sleep_maxage
$LFS df -v $MOUNT
$LCTL get_param *.*.statfs_state
local new_status=$(ost_dev_status $ost_idx $MOUNT -v)
[[ "$new_status" =~ "D" ]] ||
error "$ost_name status is '$new_status', missing 'D'"
if (( $OST1_VERSION >= $(version_code 2.12.55) )) &&
[[ ! "$new_status" =~ "N" ]]; then
error "$ost_name status is '$new_status', missing 'N'"
fi
if (( $OST1_VERSION >= $(version_code 2.12.57) )) &&
[[ ! "$new_status" =~ "f" ]]; then
error "$ost_name status is '$new_status', missing 'f'"
fi
wait_update_facet client "$LCTL get_param -n $state" "$new_status" ||
error "new state != new_status '$new_status'"
restore_lustre_params < $p
wait_update_facet client \
"$LFS df -v --output=state --ost=$ost_idx" "$old_status" ||
{
new_status=$(ost_dev_status $ost_idx)
error "new_status '$new_status' != old_status '$old_status'"
}
wait_update_facet client "$LCTL get_param -n $state" "$old_state" ||
{
local new_state=$($LCTL get_param -n $state)
error "restored state '$new_state' != old_state '$old_state'"
}
}
run_test 56c "check 'lfs df' showing device status"
test_56ca() {
remote_ost_nodsh && skip "remote OST with nodsh"
(( $OST1_VERSION >= $(version_code 2.16.53) )) ||
skip "Need OST1 >= 2.16.53 for this test"
local o_idx=0 # Target OST which will be set to readonly
local ost_name=$(ostname_from_index $o_idx)
# Force set OST0 as readonly
do_facet ost${o_idx} $LCTL set_param -n obdfilter.$ost_name.readonly=1
sleep_maxage # Give time for sync
local new_status=$(ost_dev_status $o_idx $MOUNT -v) # Get new stats
# revert to the normal state and wait for propagation
do_facet ost${o_idx} $LCTL set_param -n obdfilter.$ost_name.readonly=0
sleep_maxage
[[ "$new_status" =~ "R" ]] ||
error "$ost_name status is '$new_status', missing 'R'"
}
run_test 56ca "'lfs df -v' correctly reports 'R' flag when OST set as Readonly"
test_56d() {
local mdts=$($LFS df -v -m $MOUNT | wc -l)
local osts=$($LFS df -v -o $MOUNT | wc -l)
$LFS df $MOUNT
# Subtract 4 for the header line and summary information
(( mdts - 4 == MDSCOUNT )) ||
error "lfs df -v -m showed $((mdts - 4)) MDTs, not $MDSCOUNT"
(( osts - 4 == OSTCOUNT )) ||
error "lfs df -v -o showed $((osts - 4)) OSTs, not $OSTCOUNT"
}
run_test 56d "'lfs df -v' prints only configured devices"
test_56e() {
err_enoent=2 # No such file or directory
err_eopnotsupp=95 # Operation not supported
enoent_mnt=/pmt1 # Invalid dentry. Path not present
notsup_mnt=/tmp # Valid dentry, but Not a lustreFS
# Check for handling of path not exists
output=$($LFS df $enoent_mnt 2>&1)
ret=$?
fs=$(echo $output | awk -F: '{print $2}' | awk '{print $3}' | tr -d \')
[[ $fs = $enoent_mnt && $ret -eq $err_enoent ]] ||
error "expect failure $err_enoent, not $ret"
# Check for handling of non-Lustre FS
output=$($LFS df $notsup_mnt)
ret=$?
fs=$(echo $output | awk '{print $1}' | awk -F: '{print $2}')
[[ $fs = $notsup_mnt && $ret -eq $err_eopnotsupp ]] ||
error "expect success $err_eopnotsupp, not $ret"
# Check for multiple LustreFS argument
output=$($LFS df $MOUNT $MOUNT $MOUNT --only-summary | wc -l)
ret=$?
[[ $output -eq 3 && $ret -eq 0 ]] ||
error "expect success 3, not $output, rc = $ret"
# Check for correct non-Lustre FS handling among multiple
# LustreFS argument
output=$($LFS df $MOUNT $notsup_mnt $MOUNT |
grep -c "filesystem_summary:"; exit ${PIPESTATUS[0]})
ret=$?
[[ $output -eq 2 && $ret -eq $err_eopnotsupp ]] ||
error "expect success 2, not $output, rc = $ret"
}
run_test 56e "'lfs df' Handle non LustreFS & multiple LustreFS"
NUMFILES=3
NUMDIRS=3
setup_56() {
local local_tdir="$1"
local local_numfiles="$2"
local local_numdirs="$3"
local dir_params="$4"
local dir_stripe_params="$5"
if [ ! -d "$local_tdir" ] ; then
test_mkdir -p $dir_stripe_params $local_tdir
[ "$dir_params" ] && $LFS setstripe $dir_params $local_tdir
for i in $(seq $local_numfiles) ; do
touch $local_tdir/file$i
done
for i in $(seq $local_numdirs) ; do
test_mkdir $dir_stripe_params $local_tdir/dir$i
for j in $(seq $local_numfiles) ; do
touch $local_tdir/dir$i/file$j
done
done
fi
}
setup_56_special() {
local local_tdir=$1
local local_numfiles=$2
local local_numdirs=$3
setup_56 $local_tdir $local_numfiles $local_numdirs
if [ ! -e "$local_tdir/loop${local_numfiles}b" ] ; then
for i in $(seq $local_numfiles) ; do
mknod $local_tdir/loop${i}b b 7 $i
mknod $local_tdir/null${i}c c 1 3
ln -s $local_tdir/file1 $local_tdir/link${i}
done
for i in $(seq $local_numdirs) ; do
mknod $local_tdir/dir$i/loop${i}b b 7 $i
mknod $local_tdir/dir$i/null${i}c c 1 3
ln -s $local_tdir/dir$i/file1 $local_tdir/dir$i/link${i}
done
fi
}
test_56g() {
local dir=$DIR/d$(basetest $testnum)g.$TESTSUITE
local expected=$(($NUMDIRS + 2))
setup_56 $dir $NUMFILES $NUMDIRS
# test lfs find with -name
for i in $(seq $NUMFILES) ; do
local nums=$($LFS find -name "*$i" $dir | wc -l)
[ $nums -eq $expected ] ||
error "lfs find -name '*$i' $dir wrong: "\
"found $nums, expected $expected"
done
}
run_test 56g "check lfs find -name"
test_56h() {
local dir=$DIR/d$(basetest $testnum)g.$TESTSUITE
local expected=$(((NUMDIRS + 1) * (NUMFILES - 1) + NUMFILES))
setup_56 $dir $NUMFILES $NUMDIRS
# test lfs find with ! -name
for i in $(seq $NUMFILES) ; do
local nums=$($LFS find ! -name "*$i" $dir | wc -l)
[ $nums -eq $expected ] ||
error "lfs find ! -name '*$i' $dir wrong: "\
"found $nums, expected $expected"
done
}
run_test 56h "check lfs find ! -name"
test_56i() {
local dir=$DIR/$tdir
test_mkdir $dir
local cmd="$LFS find -ost $(ostuuid_from_index 0 $dir) $dir"
local out=$($cmd)
[ -z "$out" ] || error "'$cmd' returned directory '$out'"
}
run_test 56i "check 'lfs find -ost UUID' skips directories"
test_56ib() {
local dir=$DIR/$tdir
local file=$dir/$tfile
local idx=$((RANDOM % $OSTCOUNT))
local find_file
local cmd
test_mkdir $dir
echo "create file $file on OST $idx"
$LFS setstripe $file -o $idx||
error "failed to create test file $file"
cmd="$LFS find --ost 0,1-$idx,$idx $dir"
echo "Command: $cmd"
find_file=$($cmd)
echo "lfs find result: $find_file"
[[ "$find_file" == "$file" ]] ||
error "file $file is NOT found"
cmd="$LFS find ! --ost $idx-$((OSTCOUNT-1)),$idx $dir"
echo "Command: $cmd"
find_file=$($cmd)
[[ -z "$find_file" ]] ||
error "should NOT find any file, but find $find_file"
}
run_test 56ib "check 'lfs find -ost INDEX_RANGE' command"
test_56j() {
local dir=$DIR/d$(basetest $testnum)g.$TESTSUITE
setup_56_special $dir $NUMFILES $NUMDIRS
local expected=$((NUMDIRS + 1))
local cmd="$LFS find -type d $dir"
local nums=$($cmd | wc -l)
[ $nums -eq $expected ] ||
error "'$cmd' wrong: found $nums, expected $expected"
}
run_test 56j "check lfs find -type d"
test_56k() {
local dir=$DIR/d$(basetest $testnum)g.$TESTSUITE
setup_56_special $dir $NUMFILES $NUMDIRS
local expected=$(((NUMDIRS + 1) * NUMFILES))
local cmd="$LFS find -type f $dir"
local nums=$($cmd | wc -l)
[ $nums -eq $expected ] ||
error "'$cmd' wrong: found $nums, expected $expected"
}
run_test 56k "check lfs find -type f"
test_56l() {
local dir=$DIR/d$(basetest $testnum)g.$TESTSUITE
setup_56_special $dir $NUMFILES $NUMDIRS
local expected=$((NUMDIRS + NUMFILES))
local cmd="$LFS find -type b $dir"
local nums=$($cmd | wc -l)
[ $nums -eq $expected ] ||
error "'$cmd' wrong: found $nums, expected $expected"
}
run_test 56l "check lfs find -type b"
test_56m() {
local dir=$DIR/d$(basetest $testnum)g.$TESTSUITE
setup_56_special $dir $NUMFILES $NUMDIRS
local expected=$((NUMDIRS + NUMFILES))
local cmd="$LFS find -type c $dir"
local nums=$($cmd | wc -l)
[ $nums -eq $expected ] ||
error "'$cmd' wrong: found $nums, expected $expected"
}
run_test 56m "check lfs find -type c"
test_56n() {
local dir=$DIR/d$(basetest $testnum)g.$TESTSUITE
setup_56_special $dir $NUMFILES $NUMDIRS
local expected=$((NUMDIRS + NUMFILES))
local cmd="$LFS find -type l $dir"
local nums=$($cmd | wc -l)
[ $nums -eq $expected ] ||
error "'$cmd' wrong: found $nums, expected $expected"
}
run_test 56n "check lfs find -type l"
test_56o() {
local dir=$DIR/$tdir
setup_56 $dir $NUMFILES $NUMDIRS
utime $dir/file1 > /dev/null || error "utime (1)"
utime $dir/file2 > /dev/null || error "utime (2)"
utime $dir/dir1 > /dev/null || error "utime (3)"
utime $dir/dir2 > /dev/null || error "utime (4)"
utime $dir/dir1/file1 > /dev/null || error "utime (5)"
dd if=/dev/zero count=1 >> $dir/dir1/file1 && sync
local expected=4
local nums=$($LFS find -mtime +0 $dir | wc -l)
[ $nums -eq $expected ] ||
error "lfs find -mtime +0 $dir: found $nums expect $expected"
expected=12
cmd="$LFS find -mtime 0 $dir"
nums=$($cmd | wc -l)
[ $nums -eq $expected ] ||
error "'$cmd' wrong: found $nums, expected $expected"
}
run_test 56o "check lfs find -mtime for old files"
test_56ob() {
local dir=$DIR/$tdir
local expected=1
local count=0
# just to make sure there is something that won't be found
test_mkdir $dir
touch $dir/$tfile.now
for age in year week day hour min; do
count=$((count + 1))
touch $dir/$tfile-a.$age $dir/$tfile-m.$age
touch --date="$count $age ago" -a $dir/$tfile-a.$age
touch --date="$count $age ago" -m $dir/$tfile-m.$age
local cmd="$LFS find $dir -mtime $count${age:0:1}"
local nums=$($cmd | wc -l)
(( $nums == $expected )) || {
$LFS find $dir -ls
error "'$cmd' wrong: found $nums, expected $expected"
}
cmd="$LFS find $dir -atime $count${age:0:1}"
nums=$($cmd | wc -l)
(( $nums == $expected )) || {
$LFS find $dir -ls
error "'$cmd' wrong: found $nums, expected $expected"
}
done
sleep 2
cmd="$LFS find $dir -ctime +1s -type f"
nums=$($cmd | wc -l)
(( $nums == $count * 2 + 1)) || {
$LFS find $dir -ls
error "'$cmd' wrong: found $nums, expected $((count * 2 + 1))"
}
}
run_test 56ob "check lfs find -atime -mtime -ctime with units"
test_newerXY_base() {
local x=$1
local y=$2
local dir=$DIR/$tdir
local ref
local negref
stack_trap "rm -rf $DIR/$tdir $DIR/$tfile*" RETURN EXIT
if [ $y == "t" ]; then
if [ $x == "b" ]; then
ref=\"$(do_facet mds1 date +"%Y-%m-%d\ %H:%M:%S")\"
else
ref=\"$(date +"%Y-%m-%d %H:%M:%S")\"
fi
else
ref=$DIR/$tfile.newer.$x$y
touch $ref || error "touch $ref failed"
fi
echo "before = $ref"
sleep 2
setup_56 $dir $NUMFILES $NUMDIRS "-i0 -c1" "-i0 -c1"
sleep 2
if [ $y == "t" ]; then
if [ $x == "b" ]; then
negref=\"$(do_facet mds1 date +"%Y-%m-%d\ %H:%M:%S")\"
else
negref=\"$(date +"%Y-%m-%d %H:%M:%S")\"
fi
else
negref=$DIR/$tfile.negnewer.$x$y
touch $negref || error "touch $negref failed"
fi
echo "after = $negref"
local cmd="$LFS find $dir -newer$x$y $ref"
local nums=$(eval $cmd | wc -l)
local expected=$(((NUMFILES + 2) * NUMDIRS + 1))
[ $nums -eq $expected ] || { ls -lauR --full-time $dir ;
error "'$cmd' wrong: found $nums newer, expected $expected" ; }
cmd="$LFS find $dir ! -newer$x$y $negref"
nums=$(eval $cmd | wc -l)
[ $nums -eq $expected ] || { ls -lauR --full-time $dir ;
error "'$cmd' wrong: found $nums older, expected $expected" ; }
cmd="$LFS find $dir -newer$x$y $ref ! -newer$x$y $negref"
nums=$(eval $cmd | wc -l)
[ $nums -eq $expected ] || { ls -lauR --full-time $dir ;
error "'$cmd' wrong: found $nums between, expected $expected"; }
}
test_56oc() {
test_newerXY_base "a" "a"
test_newerXY_base "a" "m"
test_newerXY_base "a" "c"
test_newerXY_base "m" "a"
test_newerXY_base "m" "m"
test_newerXY_base "m" "c"
test_newerXY_base "c" "a"
test_newerXY_base "c" "m"
test_newerXY_base "c" "c"
test_newerXY_base "a" "t"
test_newerXY_base "m" "t"
test_newerXY_base "c" "t"
(( $MDS1_VERSION >= $(version_code v2_13_53-145-g186b97e68a) &&
$CLIENT_VERSION >= $(version_code v2_13_53-145-g186b97e68a) )) ||
{ echo "btime needs v2_13_53-145-g186b97e68a"; return 0; }
test_newerXY_base "b" "b"
test_newerXY_base "b" "t"
}
run_test 56oc "check lfs find -newerXY work"
test_56od() {
(( $MDS1_VERSION >= $(version_code v2_13_53-145-g186b97e68a) )) ||
skip "btime unsupported on MDS < v2_13_53-145-g186b97e68a"
(( $CLIENT_VERSION >= $(version_code v2_13_53-145-g186b97e68a) )) ||
skip "btime unsupported on clients < v2_13_53-145-g186b97e68a"
local dir=$DIR/$tdir
local ref=$DIR/$tfile.ref
local negref=$DIR/$tfile.negref
mkdir $dir || error "mkdir $dir failed"
touch $dir/$tfile.n1 || error "touch $dir/$tfile.n1 failed"
touch $dir/$tfile.n2 || error "touch $dir/$tfile.n2 failed"
mkdir $dir/$tdir.n1 || error "mkdir $dir/$tdir.n1 failed"
mkdir $dir/$tdir.n2 || error "mkdir $dir/$tdir.n2 failed"
touch $ref || error "touch $ref failed"
# sleep 3 seconds at least
sleep 3
local before=$(do_facet mds1 date +%s)
local skew=$(($(date +%s) - before + 1))
if (( skew < 0 && skew > -5 )); then
sleep $((0 - skew + 1))
skew=0
fi
# Set the dir stripe params to limit files all on MDT0,
# otherwise we need to calc the max clock skew between
# the client and MDTs.
setup_56 $dir/d.btime $NUMFILES $NUMDIRS "-i0 -c1" "-i0 -c1"
sleep 2
touch $negref || error "touch $negref failed"
local cmd="$LFS find $dir -newerbb $ref ! -newerbb $negref -type f"
local nums=$($cmd | wc -l)
local expected=$(((NUMFILES + 1) * NUMDIRS))
[ $nums -eq $expected ] ||
error "'$cmd' wrong: found $nums, expected $expected"
cmd="$LFS find $dir -newerbb $ref ! -newerbb $negref -type d"
nums=$($cmd | wc -l)
expected=$((NUMFILES + 1))
[ $nums -eq $expected ] ||
error "'$cmd' wrong: found $nums, expected $expected"
[ $skew -lt 0 ] && return
local after=$(do_facet mds1 date +%s)
local age=$((after - before + 1 + skew))
cmd="$LFS find $dir -btime -${age}s -type f"
nums=$($cmd | wc -l)
expected=$(((NUMFILES + 1) * NUMDIRS))
echo "Clock skew between client and server: $skew, age:$age"
[ $nums -eq $expected ] ||
error "'$cmd' wrong: found $nums, expected $expected"
expected=$(($NUMDIRS + 1))
cmd="$LFS find $dir -btime -${age}s -type d"
nums=$($cmd | wc -l)
[ $nums -eq $expected ] ||
error "'$cmd' wrong: found $nums, expected $expected"
rm -f $ref $negref || error "Failed to remove $ref $negref"
}
run_test 56od "check lfs find -btime with units"
test_56oe() {
local dir=$DIR/$tdir
local expected=5
local unit=hour
# just to make sure there is something that won't be found
test_mkdir $dir
touch $dir/$tfile.now
for ((val = 1; val <= 10; val++)); do
touch $dir/$tfile-a.$val $dir/$tfile-m.$val
touch $dir/$tfile-c.$val
echo -n "$tfile-c.$val @ctime (Epoch):"
stat -c "%Z" $dir/$tfile-c.$val
touch --date="$val $unit ago" -a $dir/$tfile-a.$val
touch --date="$val $unit ago" -m $dir/$tfile-m.$val
(( val % 5 != 2 )) || sleep 60
done
local now=$(date +%s)
local lower_bound=$(( now - $(stat -c "%Z" $dir/$tfile-c.8) + 30 ))
local upper_bound=$(( now - $(stat -c "%Z" $dir/$tfile-c.2) - 30 ))
cmd="$LFS find $dir -ctime -${upper_bound}s -ctime +${lower_bound}s"
nums=$($cmd | grep -c c.*)
(( nums == expected )) ||
error "'$cmd' wrong: found $nums between $lower_bound and $upper_bound, expected $expected"
local cmd="$LFS find $dir -atime +7h30m -atime -2h30m"
local nums=$($cmd | wc -l)
(( $($cmd | grep -c a.7) == 1 )) ||
error "'$cmd' didn't capture file accessed 7 hours ago"
(( $($cmd | grep -c a.3) == 1 )) ||
error "'$cmd' didn't capture file accessed 3 hours ago"
(( nums == expected )) ||
error "'$cmd' wrong: found $nums, expected $expected"
cmd="$LFS find $dir -mtime -7h30m -mtime +2h30m"
nums=$($cmd | wc -l)
(( $($cmd | grep -c m.7) == 1 )) ||
error "'$cmd' didn't capture file modified 7 hours ago"
(( $($cmd | grep -c m.3) == 1 )) ||
error "'$cmd' didn't capture file modified 3 hours ago"
(( nums == expected )) ||
error "'$cmd' wrong: found $nums, expected $expected"
}
run_test 56oe "check lfs find with time range"
test_56p() {
[ $RUNAS_ID -eq $UID ] &&
skip_env "RUNAS_ID = UID = $UID -- skipping"
local dir=$DIR/$tdir
setup_56 $dir $NUMFILES $NUMDIRS
chown $RUNAS_ID $dir/file* || error "chown $DIR/${tdir}g/file$i failed"
local expected=$NUMFILES
local cmd="$LFS find -uid $RUNAS_ID $dir"
local nums=$($cmd | wc -l)
[ $nums -eq $expected ] ||
error "'$cmd' wrong: found $nums, expected $expected"
expected=$(((NUMFILES + 1) * NUMDIRS + 1))
cmd="$LFS find ! -uid $RUNAS_ID $dir"
nums=$($cmd | wc -l)
[ $nums -eq $expected ] ||
error "'$cmd' wrong: found $nums, expected $expected"
}
run_test 56p "check lfs find -uid and ! -uid"
test_56q() {
[ $RUNAS_ID -eq $UID ] &&
skip_env "RUNAS_ID = UID = $UID -- skipping"
local dir=$DIR/$tdir
setup_56 $dir $NUMFILES $NUMDIRS
chgrp $RUNAS_GID $dir/file* || error "chown $dir/file$i failed"
local expected=$NUMFILES
local cmd="$LFS find -gid $RUNAS_GID $dir"
local nums=$($cmd | wc -l)
[ $nums -eq $expected ] ||
error "'$cmd' wrong: found $nums, expected $expected"
expected=$(( ($NUMFILES+1) * $NUMDIRS + 1))
cmd="$LFS find ! -gid $RUNAS_GID $dir"
nums=$($cmd | wc -l)
[ $nums -eq $expected ] ||
error "'$cmd' wrong: found $nums, expected $expected"
}
run_test 56q "check lfs find -gid and ! -gid"
test_56r() {
local dir=$DIR/$tdir
setup_56 $dir $NUMFILES $NUMDIRS
local expected=12
local cmd="$LFS find -size 0 -type f -lazy $dir"
local nums=$($cmd | wc -l)
[ $nums -eq $expected ] ||
error "'$cmd' wrong: found $nums, expected $expected"
cmd="$LFS find -size 0 -type f $dir"
nums=$($cmd | wc -l)
[ $nums -eq $expected ] ||
error "'$cmd' wrong: found $nums, expected $expected"
expected=0
cmd="$LFS find ! -size 0 -type f -lazy $dir"
nums=$($cmd | wc -l)
[ $nums -eq $expected ] ||
error "'$cmd' wrong: found $nums, expected $expected"
cmd="$LFS find ! -size 0 -type f $dir"
nums=$($cmd | wc -l)
[ $nums -eq $expected ] ||
error "'$cmd' wrong: found $nums, expected $expected"
echo "test" > $dir/$tfile
echo "test2" > $dir/$tfile.2 && sync
expected=1
cmd="$LFS find -size 5c -type f -lazy $dir"
nums=$($cmd | wc -l)
[ $nums -eq $expected ] ||
error "'$cmd' wrong: found $nums, expected $expected"
cmd="$LFS find -size 5c -type f $dir"
nums=$($cmd | wc -l)
[ $nums -eq $expected ] ||
error "'$cmd' wrong: found $nums, expected $expected"
expected=1
cmd="$LFS find -size +5c -type f -lazy $dir"
nums=$($cmd | wc -l)
[ $nums -eq $expected ] ||
error "'$cmd' wrong: found $nums, expected $expected"
cmd="$LFS find -size +5c -type f $dir"
nums=$($cmd | wc -l)
[ $nums -eq $expected ] ||
error "'$cmd' wrong: found $nums, expected $expected"
expected=2
cmd="$LFS find -size +0 -type f -lazy $dir"
nums=$($cmd | wc -l)
[ $nums -eq $expected ] ||
error "'$cmd' wrong: found $nums, expected $expected"
cmd="$LFS find -size +0 -type f $dir"
nums=$($cmd | wc -l)
[ $nums -eq $expected ] ||
error "'$cmd' wrong: found $nums, expected $expected"
expected=2
cmd="$LFS find ! -size -5c -type f -lazy $dir"
nums=$($cmd | wc -l)
[ $nums -eq $expected ] ||
error "'$cmd' wrong: found $nums, expected $expected"
cmd="$LFS find ! -size -5c -type f $dir"
nums=$($cmd | wc -l)
[ $nums -eq $expected ] ||
error "'$cmd' wrong: found $nums, expected $expected"
expected=12
cmd="$LFS find -size -5c -type f -lazy $dir"
nums=$($cmd | wc -l)
[ $nums -eq $expected ] ||
error "'$cmd' wrong: found $nums, expected $expected"
cmd="$LFS find -size -5c -type f $dir"
nums=$($cmd | wc -l)
[ $nums -eq $expected ] ||
error "'$cmd' wrong: found $nums, expected $expected"
}
run_test 56r "check lfs find -size works"
test_56ra_sub() {
local expected=$1
local glimpses=$2
local cmd="$3"
cancel_lru_locks $OSC
local rpcs_before=$(calc_stats $OSC.*$OSC*.stats ldlm_glimpse_enqueue)
local nums=$($cmd | wc -l)
[ $nums -eq $expected ] ||
error "'$cmd' wrong: found $nums, expected $expected"
local rpcs_after=$(calc_stats $OSC.*$OSC*.stats ldlm_glimpse_enqueue)
if (( rpcs_before + glimpses != rpcs_after )); then
echo "Before: $rpcs_before After: $rpcs_after $NUMFILES"
$LCTL get_param osc.*.stats | grep ldlm_glimpse_enqueue
if [[ $glimpses == 0 ]]; then
error "'$cmd' should not send glimpse RPCs to OST"
else
error "'$cmd' should send $glimpses glimpse RPCs to OST"
fi
fi
}
test_56ra() {
[[ $MDS1_VERSION -gt $(version_code 2.12.58) ]] ||
skip "MDS < 2.12.58 doesn't return LSOM data"
local dir=$DIR/$tdir
local old_agl=$($LCTL get_param -n llite.*.statahead_agl)
[[ $OSC == "mdc" ]] && skip "statahead not needed for DoM files"
# statahead_agl may cause extra glimpse which confuses results. LU-13017
$LCTL set_param -n llite.*.statahead_agl=0
stack_trap "$LCTL set_param -n llite.*.statahead_agl=$old_agl"
setup_56 $dir $NUMFILES $NUMDIRS "-c 1"
# open and close all files to ensure LSOM is updated
cancel_lru_locks $OSC
find $dir -type f | xargs cat > /dev/null
# expect_found glimpse_rpcs command_to_run
test_56ra_sub 12 0 "$LFS find -size 0 -type f -lazy $dir"
test_56ra_sub 12 12 "$LFS find -size 0 -type f $dir"
test_56ra_sub 0 0 "$LFS find ! -size 0 -type f -lazy $dir"
test_56ra_sub 0 12 "$LFS find ! -size 0 -type f $dir"
echo "test" > $dir/$tfile
echo "test2" > $dir/$tfile.2 && sync
cancel_lru_locks $OSC
cat $dir/$tfile $dir/$tfile.2 > /dev/null
test_56ra_sub 1 0 "$LFS find -size 5c -type f -lazy $dir"
test_56ra_sub 1 14 "$LFS find -size 5c -type f $dir"
test_56ra_sub 1 0 "$LFS find -size +5c -type f -lazy $dir"
test_56ra_sub 1 14 "$LFS find -size +5c -type f $dir"
test_56ra_sub 2 0 "$LFS find -size +0 -type f -lazy $dir"
test_56ra_sub 2 14 "$LFS find -size +0 -type f $dir"
test_56ra_sub 2 0 "$LFS find ! -size -5c -type f -lazy $dir"
test_56ra_sub 2 14 "$LFS find ! -size -5c -type f $dir"
test_56ra_sub 12 0 "$LFS find -size -5c -type f -lazy $dir"
test_56ra_sub 12 14 "$LFS find -size -5c -type f $dir"
}
run_test 56ra "check lfs find -size -lazy works for data on OSTs"
test_56rb() {
local dir=$DIR/$tdir
local tmp=$TMP/$tfile.log
local mdt_idx;
test_mkdir -p $dir || error "failed to mkdir $dir"
$LFS setstripe -c 1 -i 0 $dir/$tfile ||
error "failed to setstripe $dir/$tfile"
mdt_idx=$($LFS getdirstripe -i $dir)
dd if=/dev/zero of=$dir/$tfile bs=1M count=1
stack_trap "rm -f $tmp $dir/$tfile" EXIT
$LFS find --size +100K --ost 0 $dir |& tee $tmp
! grep -q obd_uuid $tmp ||
error "failed to find --size +100K --ost 0 $dir"
$LFS find --size +100K --mdt $mdt_idx $dir |& tee $tmp
! grep -q obd_uuid $tmp ||
error "failed to find --size +100K --mdt $mdt_idx $dir"
}
run_test 56rb "check lfs find --size --ost/--mdt works"
test_56rc() {
(( MDSCOUNT >= 2 )) || skip "needs at least 2 MDTs"
local dir=$DIR/$tdir
local found
test_mkdir -c 2 -H all_char $dir || error "failed to mkdir $dir"
$LFS mkdir -c 2 --mdt-hash all_char $dir/$tdir-all{1..10}
(( $MDSCOUNT > 2 )) &&
$LFS mkdir -c 3 --mdt-hash fnv_1a_64 $dir/$tdir-fnv{1..10}
mkdir $dir/$tdir-{1..10}
touch $dir/$tfile-{1..10}
found=$($LFS find $dir --mdt-count 2 | wc -l)
expect=11
(( $found == $expect )) || error "found $found 2-stripe, expect $expect"
found=$($LFS find $dir -T +1 | wc -l)
(( $MDSCOUNT > 2 )) && expect=$((expect + 10))
(( $found == $expect )) || error "found $found 2+stripe, expect $expect"
found=$($LFS find $dir --mdt-hash all_char | wc -l)
expect=11
(( $found == $expect )) || error "found $found all_char, expect $expect"
found=$($LFS find $dir --mdt-hash fnv_1a_64 | wc -l)
(( $MDSCOUNT > 2 )) && expect=10 || expect=0
(( $found == $expect )) || error "found $found all_char, expect $expect"
}
run_test 56rc "check lfs find --mdt-count/--mdt-hash works"
test_56rd() {
local dir=$DIR/$tdir
test_mkdir $dir
rm -f $dir/*
mkfifo $dir/fifo || error "failed to create fifo file"
$LFS find $dir -t p --printf "%p %y %LP\n" ||
error "should not fail when getting projid from pipe file"
found=$($LFS find $dir -t p --printf "%y")
[[ "p" == $found ]] || error "found $found, expect p"
mknod $dir/chardev c 1 5 ||
error "failed to create character device file"
$LFS find $dir -t c --printf "%p %y %LP\n" ||
error "should not fail when getting projid from chardev file"
found=$($LFS find $dir -t c --printf "%y")
[[ "c" == $found ]] || error "found $found, expect c"
found=$($LFS find $dir ! -type d --printf "%p %y %LP\n" | wc -l)
(( found == 2 )) || error "unable to list all files"
}
run_test 56rd "check lfs find --printf special files"
test_56re()
{
local dir=$DIR/$tdir
local file=$dir/a
local test_cases=(
"|%5s|" "|%05s|" "|%0s|" "|%-5s|" "|%-0s|"
"|%5G|" "|%05G|" "|%0G|" "|%-5G|" "|%-0G|"
"|%25i|" "|%025i|" "|%0i|" "|%-25i|" "|%-0i|"
"|%5k|" "|%05k|" "|%0k|" "|%-5k|" "|%-0k|"
"|%5y|" "|%05y|" "|%0y|" "|%-5y|" "|%-0y|"
"|%5m|" "|%05m|" "|%0m|" "|%-5m|" "|%-0m|" "|%-05m|"
"|%----5s|" "|%----00005s|" "|%00005s|"
"|%020|" "|%20|" "|%-20|" "|%-020|" "|%---020|" "|%---20|"
)
test_mkdir "$dir"
touch "$file"
truncate -s 37 "$file"
for format in ${test_cases[@]}; do
local expected=$(find "$dir" -type f -printf "$format\n")
local output=$($LFS find "$dir" -type f -printf "$format\n")
[[ $output == $expected ]] ||
error "Unexpected output to \"$format\", expected '$expected' got '$output'"
done
}
run_test 56re "check lfs find -printf width format specifiers are consistant with regular find"
test_56rf()
{
local dir=$DIR/$tdir
local file=$dir/a
local test_cases
local poolname=testpool56rf
local stripe_count
local fid
test_mkdir "$dir"
pool_add $poolname || error "Could not create pool '$poolname'"
lfs setstripe -c 2 -S 4M -p $poolname "$file"
stripe_count=$(lfs getstripe -c "$file")
fid=$($LFS path2fid "$file" | sed 's/\[\(.*\)\]/\1/')
len=$((${#fid} + 5))
test_cases=(
"|%${len}LF|" "| $fid|"
"|%0${len}LF|" "| $fid|"
"|%0LF|" "|$fid|"
"|%-${len}LF|" "|$fid |"
"|%-0LF|" "|$fid|"
"|%14Lp|" "| $poolname|"
"|%014Lp|" "| $poolname|"
"|%0Lp|" "|$poolname|"
"|%-14Lp|" "|$poolname |"
"|%-0Lp|" "|$poolname|"
"|%07Lp|" "|$poolname|"
"|%-7Lp|" "|$poolname|"
"|%4Lc|" "| $stripe_count|"
"|%04Lc|" "| $stripe_count|"
"|%0Lc|" "|$stripe_count|"
"|%-4Lc|" "|$stripe_count |"
"|%-0Lc|" "|$stripe_count|"
"|%10LS|" "| 4194304|"
"|%010LS|" "| 4194304|"
"|%0LS|" "|4194304|"
"|%-10LS|" "|4194304 |"
"|%-0LS|" "|4194304|"
)
for (( i = 0; i < ${#test_cases[@]}; i += 2 )); do
local format=${test_cases[i]}
local expected=${test_cases[i + 1]}
local output=$($LFS find "$dir" -type f -printf "$format\n")
[[ $output == $expected ]] ||
error "Unexpected output to \"$format\", expected '$expected' got '$output'"
done
}
run_test 56rf "check lfs find -printf width format specifiers for lustre specific formats"
test_56s() { # LU-611 #LU-9369
(( $OSTCOUNT >= 2 )) || skip_env "need at least 2 OSTs"
local dir=$DIR/$tdir
local onestripe=$(((NUMDIRS + 1) * NUMFILES))
local count=$((OSTCOUNT > 2 ? OSTCOUNT - 1 : 2))
setup_56 $dir $NUMFILES $NUMDIRS "-c 1"
for ((i = 1; i <= $NUMDIRS; i++)); do
$LFS setstripe -c $count $dir/dir$i/$tfile
done
local expected=$NUMDIRS
local cmd="$LFS find -c $count $dir"
local nums=$($cmd | wc -l)
(( $nums == $expected )) || {
$LFS getstripe -R $dir
error "'$cmd' wrong: found $nums, expected $expected"
}
expected=$((NUMDIRS + onestripe))
cmd="$LFS find -stripe-count +0 -type f $dir"
nums=$($cmd | wc -l)
(( $nums == $expected )) || {
$LFS getstripe -R $dir
error "'$cmd' wrong: found $nums, expected $expected"
}
expected=$onestripe
cmd="$LFS find -stripe-count 1 -type f $dir"
nums=$($cmd | wc -l)
(( $nums == $expected )) || {
$LFS getstripe -R $dir
error "'$cmd' wrong: found $nums, expected $expected"
}
cmd="$LFS find -stripe-count -2 -type f $dir"
nums=$($cmd | wc -l)
(( $nums == $expected )) || {
$LFS getstripe -R $dir
error "'$cmd' wrong: found $nums, expected $expected"
}
expected=0
cmd="$LFS find -stripe-count $((count + 1)) -type f $dir"
nums=$($cmd | wc -l)
(( $nums == $expected )) || {
$LFS getstripe -R $dir
error "'$cmd' wrong: found $nums, expected $expected"
}
}
run_test 56s "check lfs find -stripe-count works"
test_56t() { # LU-611 #LU-9369
local dir=$DIR/$tdir
setup_56 $dir 0 $NUMDIRS
for i in $(seq $NUMDIRS); do
$LFS setstripe -S 8M $dir/dir$i/$tfile
done
local expected=$NUMDIRS
local cmd="$LFS find -S 8M $dir"
local nums=$($cmd | wc -l)
[ $nums -eq $expected ] || {
$LFS getstripe -R $dir
error "'$cmd' wrong: found $nums, expected $expected"
}
rm -rf $dir
setup_56 $dir $NUMFILES $NUMDIRS "--stripe-size 512k"
$LFS setstripe -S 256k $dir/$tfile.{0,1,2,3}
expected=$(((NUMDIRS + 1) * NUMFILES))
cmd="$LFS find -stripe-size 512k -type f $dir"
nums=$($cmd | wc -l)
[ $nums -eq $expected ] ||
error "'$cmd' wrong: found $nums, expected $expected"
cmd="$LFS find -stripe-size +320k -type f $dir"
nums=$($cmd | wc -l)
[ $nums -eq $expected ] ||
error "'$cmd' wrong: found $nums, expected $expected"
expected=$(((NUMDIRS + 1) * NUMFILES + 4))
cmd="$LFS find -stripe-size +200k -type f $dir"
nums=$($cmd | wc -l)
[ $nums -eq $expected ] ||
error "'$cmd' wrong: found $nums, expected $expected"
cmd="$LFS find -stripe-size -640k -type f $dir"
nums=$($cmd | wc -l)
[ $nums -eq $expected ] ||
error "'$cmd' wrong: found $nums, expected $expected"
expected=4
cmd="$LFS find -stripe-size 256k -type f $dir"
nums=$($cmd | wc -l)
[ $nums -eq $expected ] ||
error "'$cmd' wrong: found $nums, expected $expected"
cmd="$LFS find -stripe-size -320k -type f $dir"
nums=$($cmd | wc -l)
[ $nums -eq $expected ] ||
error "'$cmd' wrong: found $nums, expected $expected"
expected=0
cmd="$LFS find -stripe-size 1024k -type f $dir"
nums=$($cmd | wc -l)
[ $nums -eq $expected ] ||
error "'$cmd' wrong: found $nums, expected $expected"
}
run_test 56t "check lfs find -stripe-size works"
test_56u() { # LU-611
local dir=$DIR/$tdir
setup_56 $dir $NUMFILES $NUMDIRS "-i 0 -c 1"
if [[ $OSTCOUNT -gt 1 ]]; then
$LFS setstripe -i 1 -c 1 $dir/$tfile.{0,1,2,3}
onestripe=4
else
onestripe=0
fi
local expected=$(((NUMDIRS + 1) * NUMFILES))
local cmd="$LFS find -stripe-index 0 -type f $dir"
local nums=$($cmd | wc -l)
[ $nums -eq $expected ] ||
error "'$cmd' wrong: found $nums, expected $expected"
expected=$onestripe
cmd="$LFS find -stripe-index 1 -type f $dir"
nums=$($cmd | wc -l)
[ $nums -eq $expected ] ||
error "'$cmd' wrong: found $nums, expected $expected"
cmd="$LFS find ! -stripe-index 0 -type f $dir"
nums=$($cmd | wc -l)
[ $nums -eq $expected ] ||
error "'$cmd' wrong: found $nums, expected $expected"
expected=0
# This should produce an error and not return any files
cmd="$LFS find -stripe-index $OSTCOUNT -type f $dir"
nums=$($cmd 2>/dev/null | wc -l)
[ $nums -eq $expected ] ||
error "'$cmd' wrong: found $nums, expected $expected"
if [[ $OSTCOUNT -gt 1 ]]; then
expected=$(((NUMDIRS + 1) * NUMFILES + onestripe))
cmd="$LFS find -stripe-index 0,1 -type f $dir"
nums=$($cmd | wc -l)
[ $nums -eq $expected ] ||
error "'$cmd' wrong: found $nums, expected $expected"
fi
}
run_test 56u "check lfs find -stripe-index works"
test_56v() {
local mdt_idx=0
local dir=$DIR/$tdir
setup_56 $dir $NUMFILES $NUMDIRS
UUID=$(mdtuuid_from_index $mdt_idx $dir)
[ -z "$UUID" ] && error "mdtuuid_from_index cannot find MDT $mdt_idx"
for file in $($LFS find -m $UUID $dir); do
file_midx=$($LFS getstripe -m $file)
[ $file_midx -eq $mdt_idx ] ||
error "lfs find -m $UUID != getstripe -m $file_midx"
done
}
run_test 56v "check 'lfs find -m match with lfs getstripe -m'"
test_56wa() {
(( $OSTCOUNT >= 2 )) || skip "needs >= 2 OSTs"
[ $PARALLEL == "yes" ] && skip "skip parallel run"
local dir=$DIR/$tdir
local expected=$((OSTCOUNT - 1))
setup_56 $dir $NUMFILES $NUMDIRS "-c $expected" "-c1"
stack_trap "rm -rf $dir"
local stripe_size=$($LFS getstripe -S -d $dir) ||
error "$LFS getstripe -S -d $dir failed"
stripe_size=${stripe_size%% *}
local file_size=$((stripe_size * expected))
local file_num=$((NUMDIRS * NUMFILES + NUMFILES))
local required_space=$((file_num * file_size))
local free_space=$($LCTL get_param -n lov.$FSNAME-clilov-*.kbytesavail |
head -n1)
(( free_space >= required_space / 1024 )) ||
skip_env "need $required_space, have $free_space kbytes"
local dd_bs=65536
local dd_count=$((file_size / dd_bs))
# write data into the files
local i
local j
local file
for ((i = 1; i <= NUMFILES; i++ )); do
file=$dir/file$i
yes | dd bs=$dd_bs count=$dd_count of=$file &>/dev/null ||
error "write data into $file failed"
done
for ((i = 1; i <= NUMDIRS; i++ )); do
for ((j = 1; j <= NUMFILES; j++ )); do
file=$dir/dir$i/file$j
yes|dd bs=$dd_bs count=$dd_count of=$file &>/dev/null ||
error "write data into $file failed"
done
done
# $LFS_MIGRATE will fail if hard link migration is unsupported
if (( MDS1_VERSION > $(version_code 2.5.55) )); then
createmany -l$dir/dir1/file1 $dir/dir1/link 200 ||
error "creating links to $dir/dir1/file1 failed"
fi
# lfs migrate file
local cmd="$LFS migrate -v -c $expected $dir/file1"
echo "$cmd"
eval $cmd || error "$cmd failed"
check_stripe_count $dir/file1 $expected
if (( $MDS1_VERSION >= $(version_code 2.6.90) )); then
# lfs migrate file onto OST 0 if it is on OST 1, or onto
# OST 1 if it is on OST 0. This file is small enough to
# be on only one stripe.
file=$dir/migr_1_ost
dd bs=$dd_bs count=1 if=/dev/urandom of=$file >/dev/null 2>&1 ||
error "write data into $file failed"
local obdidx=$($LFS getstripe -i $file)
local oldmd5=$(md5sum $file)
local newobdidx=0
(( obdidx != 0 )) || newobdidx=1
cmd="$LFS migrate -i $newobdidx $file"
echo $cmd
eval $cmd || error "$cmd failed"
local realobdix=$($LFS getstripe -i $file)
local newmd5=$(md5sum $file)
(( $newobdidx == $realobdix )) ||
error "new OST is different (was=$obdidx, wanted=$newobdidx, got=$realobdix)"
[[ "$oldmd5" == "$newmd5" ]] ||
error "md5sum differ: $oldmd5, $newmd5"
fi
# lfs migrate works with lfs find
cmd="$LFS find -stripe_count $expected -type f $dir -print0 |
$LFS migrate -0 -v -C $((expected * 2))"
echo "$cmd"
eval $cmd || error "$cmd failed"
for (( i = 2; i <= NUMFILES; i++ )); do
check_stripe_count $dir/file$i $((expected * 2))
done
for (( i = 2; i <= NUMDIRS; i++ )); do
for (( j = 1; j <= NUMFILES; j++ )); do
check_stripe_count $dir/dir$i/file$j $((expected * 2))
done
done
}
run_test 56wa "check 'lfs migrate -c stripe_count' works"
test_56wb() {
local file1=$DIR/$tdir/file1
local initial_pool=$($LFS getstripe -p $DIR)
local pool_list=()
local pool=""
echo -n "Creating test dir..."
test_mkdir $DIR/$tdir &> /dev/null || error "cannot create dir"
echo "done."
echo -n "Creating test file..."
touch $file1 || error "cannot create file"
echo "done."
echo -n "Detecting existing pools..."
pool_list=($($LFS pool_list $FSNAME | grep "$FSNAME\." | cut -d. -f2))
if [ ${#pool_list[@]} -gt 0 ]; then
echo "${pool_list[@]}"
for thispool in "${pool_list[@]}"; do
if [[ -z "$initial_pool" ||
"$initial_pool" != "$thispool" ]]; then
pool="$thispool"
echo "Using existing pool '$pool'"
break
fi
done
else
echo "none detected."
fi
if [ -z "$pool" ]; then
pool=${POOL:-testpool}
[ "$initial_pool" = "$pool" ] && pool="testpool2"
echo -n "Creating pool '$pool'..."
create_pool $FSNAME.$pool ||
error "create OST pool $pool failed"
echo "done."
echo -n "Adding target to pool..."
pool_add_targets $pool 0 0 1 &> /dev/null ||
error "pool_add_targets failed"
echo "done."
fi
echo -n "Setting pool using -p option..."
$LFS migrate -p $pool $file1 ||
error "migrate failed rc = $?"
echo "done."
echo -n "Verifying test file is in pool after migrating..."
[ "$($LFS getstripe -p $file1)" = $pool ] ||
error "file was not migrated to pool $pool"
echo "done."
echo -n "Removing test file from pool '$pool'..."
# "lfs migrate $file" won't remove the file from the pool
# until some striping information is changed.
$LFS migrate -c 1 $file1 &> /dev/null ||
error "cannot remove from pool"
[[ -z "$($LFS getstripe --pool $file1)" ]] ||
error "pool still set"
echo "done."
echo -n "Setting pool using --pool option..."
$LFS migrate --pool $pool $file1 ||
error "migrate failed rc = $?"
echo "done."
# Clean up
rm -f $file1
}
run_test 56wb "check 'lfs migrate' pool support"
test_56wc() {
local file1="$DIR/$tdir/$tfile"
local md5
local parent_ssize
local parent_scount
local cur_ssize
local cur_scount
local orig_ssize
local new_scount
local cur_comp
echo -n "Creating test dir..."
test_mkdir $DIR/$tdir &> /dev/null || error "cannot create dir"
$LFS setstripe -S 1M -c 1 "$DIR/$tdir" &> /dev/null ||
error "cannot set stripe by '-S 1M -c 1'"
echo "done"
echo -n "Setting initial stripe for test file..."
$LFS setstripe -S 512K -c 1 "$file1" &> /dev/null ||
error "cannot set stripe"
cur_ssize=$($LFS getstripe -S "$file1")
(( cur_ssize == 524288 )) || error "setstripe -S $cur_ssize != 524288"
echo "done."
dd if=/dev/urandom of=$file1 bs=1M count=12 || error "dd $file1 failed"
stack_trap "rm -f $file1"
md5="$(md5sum $file1)"
# File currently set to -S 512K -c 1
# Ensure -c and -S options are rejected when --restripe is set
echo -n "Verifying incompatible options are detected..."
$LFS migrate --restripe -c 1 "$file1" &&
error "incompatible --restripe and -c options not detected"
$LFS migrate --restripe -S 1M "$file1" &&
error "incompatible --restripe and -S options not detected"
$LFS migrate --restripe -p pool "$file1" &&
error "incompatible --restripe and -p options not detected"
$LFS migrate --restripe -E eof -c 1 "$file1" &&
error "incompatible --restripe and -E options not detected"
$LFS migrate --restripe -A "$file1" &&
error "incompatible --restripe and -A options not detected"
$LFS migrate -A -c 1 "$file1" &&
error "incompatible -A and -c options not detected"
$LFS migrate -A -E eof -c 1 "$file1" &&
error "incompatible -A and -E options not detected"
echo "done."
# Ensure unrecognized options are passed through to 'lfs migrate'
echo -n "Verifying -S option is passed through to lfs migrate..."
$LFS migrate -S 1M "$file1" || error "migration failed"
cur_ssize=$($LFS getstripe -S "$file1")
(( cur_ssize == 1048576 )) || error "migrate -S $cur_ssize != 1048576"
[[ "$(md5sum $file1)" == "$md5" ]] || error "file data has changed (1)"
echo "done."
# File currently set to -S 1M -c 1
# Ensure long options are supported
echo -n "Verifying long options supported..."
$LFS migrate --non-block "$file1" ||
error "long option without argument not supported"
$LFS migrate --stripe-size 512K "$file1" ||
error "long option with argument not supported"
cur_ssize=$($LFS getstripe -S "$file1")
(( cur_ssize == 524288 )) ||
error "migrate --stripe-size $cur_ssize != 524288"
[[ "$(md5sum $file1)" == "$md5" ]] || error "file data has changed (2)"
echo "done."
# File currently set to -S 512K -c 1
if (( OSTCOUNT > 1 )); then
echo -n "Verifying explicit stripe count can be set..."
$LFS migrate -c 2 "$file1" || error "migrate failed"
cur_scount=$($LFS getstripe -c "$file1")
(( cur_scount == 2 )) || error "migrate -c $cur_scount != 2"
[[ "$(md5sum $file1)" == "$md5" ]] ||
error "file data has changed (3)"
echo "done."
fi
# File currently set to -S 512K -c 1 or -S 512K -c 2
# Ensure parent striping is used if -R is set, and no stripe
# count or size is specified
echo -n "Setting stripe for parent directory..."
$LFS setstripe -S 2M -c 1 "$DIR/$tdir" &> /dev/null ||
error "cannot set stripe '-S 2M -c 1'"
[[ "$(md5sum $file1)" == "$md5" ]] || error "file data has changed (4)"
echo "done."
echo -n "Verifying restripe option uses parent stripe settings..."
parent_ssize=$($LFS getstripe -S $DIR/$tdir 2>/dev/null)
parent_scount=$($LFS getstripe -c $DIR/$tdir 2>/dev/null)
$LFS migrate --restripe "$file1" || error "migrate failed"
cur_ssize=$($LFS getstripe -S "$file1")
(( cur_ssize == parent_ssize )) ||
error "migrate --restripe stripe_size $cur_ssize != $parent_ssize"
cur_scount=$($LFS getstripe -c "$file1")
(( cur_scount == parent_scount )) ||
error "migrate -R stripe_count $cur_scount != $parent_scount"
[[ "$(md5sum $file1)" == "$md5" ]] || error "file data has changed (5)"
echo "done."
# File currently set to -S 1M -c 1
# Ensure striping is preserved if -R is not set, and no stripe
# count or size is specified
echo -n "Verifying striping size preserved when not specified..."
orig_ssize=$($LFS getstripe -S "$file1" 2>/dev/null)
$LFS setstripe -S 2M -c 1 "$DIR/$tdir" &> /dev/null ||
error "cannot set stripe on parent directory"
$LFS migrate "$file1" || error "migrate failed"
cur_ssize=$($LFS getstripe -S "$file1")
(( cur_ssize == orig_ssize )) ||
error "migrate by default $cur_ssize != $orig_ssize"
[[ "$(md5sum $file1)" == "$md5" ]] || error "file data has changed (6)"
echo "done."
# Ensure file name properly detected when final option has no argument
echo -n "Verifying file name properly detected..."
$LFS migrate "$file1" ||
error "file name interpreted as option argument"
[[ "$(md5sum $file1)" == "$md5" ]] || error "file data has changed (7)"
echo "done."
# Ensure PFL arguments are passed through properly
echo -n "Verifying PFL options passed through..."
new_scount=$(((OSTCOUNT + 1) / 2))
$LFS migrate -E 1M -c 1 -E 16M -c $new_scount -E eof -c -1 "$file1" ||
error "migrate PFL arguments failed"
cur_comp=$($LFS getstripe --comp-count $file1)
(( cur_comp == 3 )) || error "component count '$cur_comp' != 3"
cur_scount=$($LFS getstripe --stripe-count $file1)
(( cur_scount == new_scount)) ||
error "PFL stripe count $cur_scount != $new_scount"
[[ "$(md5sum $file1)" == "$md5" ]] || error "file data has changed (8)"
echo "done."
}
run_test 56wc "check unrecognized options for lfs migrate are passed through"
test_56wd() {
(( $CLIENT_VERSION < $(version_code 2.20.53) )) ||
skip "lfs_migrate is not supported starting with Lustre 2.20.53"
(( $OSTCOUNT >= 2 )) || skip "needs >= 2 OSTs"
local file1=$DIR/$tdir/$tfile
echo -n "Creating test dir..."
test_mkdir $DIR/$tdir || error "cannot create dir"
echo "done."
echo -n "Creating test file..."
echo "$tfile" > $file1
echo "done."
# Ensure 'lfs migrate' will fail by using a non-existent option,
# and make sure rsync is not called to recover
echo -n "Make sure --no-rsync option works..."
$LFS_MIGRATE -y --no-rsync --invalid-opt $file1 2>&1 |
grep -q 'refusing to fall back to rsync' ||
error "rsync was called with --no-rsync set"
echo "done."
# Ensure rsync is called without trying 'lfs migrate' first
echo -n "Make sure --rsync option works..."
$LFS_MIGRATE -y --rsync --invalid-opt $file1 2>&1 |
grep -q 'falling back to rsync' &&
error "lfs migrate was called with --rsync set"
echo "done."
}
run_test 56wd "check lfs_migrate --rsync and --no-rsync work"
test_56we() {
local td=$DIR/$tdir
local tf=$td/$tfile
test_mkdir $td || error "cannot create $td"
touch $tf || error "cannot touch $tf"
echo -n "Make sure --non-direct|-D works..."
$LFS migrate --non-direct -v $tf 2>&1 |
grep -q "$tf" ||
error "--non-direct option cannot work correctly"
$LFS migrate -D -v $tf 2>&1 |
grep -q "$tf" ||
error "-D option cannot work correctly"
echo "done."
}
run_test 56we "check lfs migrate --non-direct|-D support"
test_56x() {
[[ $OSTCOUNT -lt 2 ]] && skip_env "needs >= 2 OSTs"
check_swap_layouts_support
local dir=$DIR/$tdir
local ref1=/etc/passwd
local file1=$dir/file1
test_mkdir $dir || error "creating dir $dir"
$LFS setstripe -c 2 $file1
cp $ref1 $file1
$LFS migrate -c 1 $file1 || error "migrate failed rc = $?"
stripe=$($LFS getstripe -c $file1)
[[ $stripe == 1 ]] || error "stripe of $file1 is $stripe != 1"
cmp $file1 $ref1 || error "content mismatch $file1 differs from $ref1"
# clean up
rm -f $file1
}
run_test 56x "lfs migration support"
test_56xB() {
local td=$DIR/$tdir
local tf1=$td/${tfile}_1
local tf2=$td/${tfile}_2
local tf3=$td/${tfile}_3
local flist=/tmp/flist.$$
(( $OSTCOUNT >= 2 )) || skip "needs >= 2 OSTs"
test_mkdir $td
dd if=/dev/urandom of=$tf1 bs=1K count=1 || error "failed to create $tf1"
dd if=/dev/urandom of=$tf2 bs=1K count=1 || error "failed to create $tf2"
dd if=/dev/urandom of=$tf3 bs=1K count=1 || error "failed to create $tf3"
stack_trap "rm -f $flist"
$LFS find $td -type f > $flist || error "failed to generate list file"
$LFS migrate -o 0 --files-from=$flist ||
error "failed to run lfs migrate with --files-from"
$LFS find $td -type f -print0 | $LFS migrate -o 1 --null ||
error "failed to run lfs migrate from pipe"
$LFS find $td -type f -print0 > $flist ||
error "failed to generate 0-ending list file"
$LFS migrate -o 0 -0 --files-from=$flist ||
error "failed to run lfs migrate with -0 --files-from"
}
run_test 56xB "lfs migrate with -0, --null, --files-from arguments"
test_56xC() {
local td=$DIR/$tdir
local tf1=$td/${tfile}_1
local tf2=$td/${tfile}_2
local tf3=$td/${tfile}_3
local fidlist=/tmp/fidlist.$$
local stripe_count
local stripe_size
(( $OSTCOUNT >= 2 )) || skip "needs >= 2 OSTs"
test_mkdir $td
dd if=/dev/urandom of=$tf1 bs=1K count=1 || error "failed to create $tf1"
dd if=/dev/urandom of=$tf2 bs=1K count=1 || error "failed to create $tf2"
$LFS setstripe --stripe-size=3M -C 7 $tf3
dd if=/dev/urandom of=$tf3 bs=1K count=1 || error "failed to create $tf3"
$LFS getstripe $tf3
stack_trap "rm -f $fidlist"
$LFS path2fid $tf1 $tf2 $tf3 | awk '{print $2}' > $fidlist ||
error "failed to get fids of $tf1 $tf2 $tf3"
echo "================ fidlist ======================="
cat $fidlist
echo "================================================"
$LFS migrate -o 0 --lustre-dir=$DIR --fid --files-from=$fidlist ||
error "migrate -o 0 failed with --fid and --lustre-dir"
mount
echo "without --lustre-dir opiton"
$LFS migrate --stripe-size=2M -o 1 --fid --files-from=$fidlist ||
error "failed to run lfs migrate with --fid argument"
$LFS getstripe $tf3
stripe_count=$($LFS getstripe --stripe-count $tf3)
stripe_size=$($LFS getstripe --stripe-size $tf3)
(( $stripe_count == 1 )) ||
error "unexpected stripe count $stripe_count"
(( $stripe_size == 2 * 1024 * 1024 )) ||
error "unexpected stripe size $stripe_size"
}
run_test 56xC "lfs migration can accept FID list file"
test_56xD() {
local td=$DIR/$tdir
local tf1=$td/${tfile}_1
local tf2=$td/${tfile}_2
local tf3=$td/${tfile}_3
local fids
local stripe_count
local stripe_size
(( $OSTCOUNT >= 2 )) || skip "needs >= 2 OSTs"
test_mkdir $td
dd if=/dev/urandom of=$tf1 bs=1K count=1 || error "failed to create $tf1"
dd if=/dev/urandom of=$tf2 bs=1K count=1 || error "failed to create $tf2"
dd if=/dev/urandom of=$tf3 bs=1K count=1 || error "failed to create $tf3"
stack_trap "rm -f $fidlist"
fids=$($LFS path2fid $tf1 $tf2 $tf3 | awk '{printf "%s ", $2}' |
tr -d '[]') ||
error "failed to get fids of $tf1 $tf2 $tf3"
echo "FIDs: $fids"
echo "with --lustre-dir option"
$LFS migrate -o 0 --lustre-dir $DIR --fid $fids||
error "failed to run lfs migrate with --fid and --lustre-dir argument"
echo "without --lustre-dir option"
$LFS migrate -o 1 -S 2M --fid $fids||
error "failed to run lfs migrate with --fid argument"
$LFS getstripe $tf3
stripe_count=$($LFS getstripe --stripe-count $tf3)
stripe_size=$($LFS getstripe --stripe-size $tf3)
(( $stripe_count == 1 )) ||
error "unexpected stripe count $stripe_count"
(( $stripe_size == 2 * 1024 * 1024 )) ||
error "unexpected stripe size $stripe_size"
}
run_test 56xD "lfs_migration can accept FIDs"
test_56xa() {
(( $OSTCOUNT >= 2 )) || skip_env "needs >= 2 OSTs"
check_swap_layouts_support
local dir=$DIR/$tdir/$testnum
test_mkdir -p $dir
local ref1=/etc/passwd
local file1=$dir/file1
$LFS setstripe -c 2 $file1
cp $ref1 $file1
$LFS migrate --block -c 1 $file1 || error "migrate failed rc = $?"
local stripe=$($LFS getstripe -c $file1)
[[ $stripe == 1 ]] || error "stripe of $file1 is $stripe != 1"
cmp $file1 $ref1 || error "content mismatch $file1 differs from $ref1"
# clean up
rm -f $file1
}
run_test 56xa "lfs migration --block support"
test_56xab() {
local pid
local dir=$DIR/$tdir/$testnum
local file1=$dir/$tfile
(( $OSTCOUNT >= 2 )) || skip_env "needs >= 2 OSTs"
check_swap_layouts_support
test_mkdir -p $dir
$LFS setstripe -c 2 $file1
dd if=/dev/urandom of=$file1 bs=1 oflag=sync & pid=$!
kill -0 $pid ||
error "failed to start dd to write to $file1 continously"
stack_trap "kill $pid"
sleep 1
$LFS migrate --block -c 1 $file1 || error "migrate failed rc = $?"
ls -l $file1
$LFS getstripe $file1
}
run_test 56xab "lfs migration --block on changing file"
check_migrate_links() {
[[ "$1" == "--rsync" ]] && local opts="--rsync -y" && shift
local dir="$1"
local file1="$dir/file1"
local begin="$2"
local count="$3"
local runas="$4"
local total_count=$(($begin + $count - 1))
local symlink_count=10
local uniq_count=10
if [ ! -f "$file1" ]; then
echo -n "creating initial file..."
$LFS setstripe -c 1 -S "512k" "$file1" ||
error "cannot setstripe initial file"
echo "done"
echo -n "creating symlinks..."
for s in $(seq 1 $symlink_count); do
ln -s "$file1" "$dir/slink$s" ||
error "cannot create symlinks"
done
echo "done"
echo -n "creating nonlinked files..."
createmany -o "$dir/uniq" 1 10 &> /dev/null ||
error "cannot create nonlinked files"
echo "done"
fi
# create hard links
if [ ! -f "$dir/file$total_count" ]; then
echo -n "creating hard links $begin:$total_count..."
createmany -l"$file1" "$dir/file" "$begin" "$count" &> \
/dev/null || error "cannot create hard links"
echo "done"
fi
echo -n "checking number of hard links listed in xattrs..."
local fid=$($LFS getstripe -F "$file1")
local paths=($($LFS fid2path "$MOUNT" "$fid" 2> /dev/null))
echo "${#paths[*]}"
if [ ${#paths[*]} -lt $total_count -a "$begin" -eq 2 ]; then
skip "hard link list has unexpected size, skipping test"
fi
if [ ${#paths[*]} -ge $total_count -a "$begin" -ne 2 ]; then
error "link names should exceed xattrs size"
fi
echo -n "migrating files..."
local migrate_out=$($runas $LFS_MIGRATE $opts -S '1m' $dir)
local rc=$?
[ $rc -eq 0 ] || error "migrate failed rc = $rc"
echo "done"
# make sure all links have been properly migrated
echo -n "verifying files..."
fid=$($LFS getstripe -F "$file1") ||
error "cannot get fid for file $file1"
for i in $(seq 2 $total_count); do
local fid2=$($LFS getstripe -F $dir/file$i)
[ "$fid2" == "$fid" ] ||
error "migrated hard link has mismatched FID"
done
# make sure hard links were properly detected, and migration was
# performed only once for the entire link set; nonlinked files should
# also be migrated
local actual=$(grep -c 'done' <<< "$migrate_out")
local expected=$(($uniq_count + 1))
[ "$actual" -eq "$expected" ] ||
error "hard links individually migrated ($actual != $expected)"
# make sure the correct number of hard links are present
local hardlinks=$(stat -c '%h' "$file1")
[ $hardlinks -eq $total_count ] ||
error "num hard links $hardlinks != $total_count"
echo "done"
return 0
}
test_56xb() {
(( $CLIENT_VERSION < $(version_code 2.20.53) )) ||
skip "lfs_migrate is not supported starting with Lustre 2.20.53"
[ $MDS1_VERSION -lt $(version_code 2.10.55) ] &&
skip "Need MDS version at least 2.10.55"
local dir="$DIR/$tdir"
test_mkdir "$dir" || error "cannot create dir $dir"
echo "testing lfs migrate mode when all links fit within xattrs"
check_migrate_links "$dir" 2 99
echo "testing rsync mode when all links fit within xattrs"
check_migrate_links --rsync "$dir" 2 99
echo "testing lfs migrate mode when all links do not fit within xattrs"
check_migrate_links "$dir" 101 100
echo "testing rsync mode when all links do not fit within xattrs"
check_migrate_links --rsync "$dir" 101 100
chown -R $RUNAS_ID $dir
echo "testing non-root lfs migrate mode when not all links are in xattr"
check_migrate_links "$dir" 101 100 "$RUNAS -G0"
# clean up
rm -rf $dir
}
run_test 56xb "lfs migration hard link support"
test_56xc() {
[[ $OSTCOUNT -lt 2 ]] && skip_env "needs >= 2 OSTs"
local dir="$DIR/$tdir"
test_mkdir "$dir" || error "cannot create dir $dir"
# Test 1: ensure file < 1 GB is always migrated with 1 stripe
echo -n "Setting initial stripe for 20MB test file..."
$LFS setstripe -c 2 -i 0 "$dir/20mb" ||
error "cannot setstripe 20MB file"
echo "done"
echo -n "Sizing 20MB test file..."
$TRUNCATE "$dir/20mb" 20971520 || error "cannot create 20MB test file"
echo "done"
echo -n "Verifying small file autostripe count is 1..."
$LFS migrate -A -K 1 "$dir/20mb" ||
error "cannot migrate 20MB file"
local stripe_count=$($LFS getstripe -c "$dir/20mb") ||
error "cannot get stripe for $dir/20mb"
[ $stripe_count -eq 1 ] ||
error "unexpected stripe count $stripe_count for 20MB file"
rm -f "$dir/20mb"
echo "done"
# Test 2: File is small enough to fit within the available space on
# sqrt(size_in_gb) + 1 OSTs but is larger than 1GB. The file must
# have at least an additional 1KB for each desired stripe for test 3
echo -n "Setting stripe for 1GB test file..."
$LFS setstripe -c 1 -i 0 "$dir/1gb" || error "cannot setstripe 1GB file"
echo "done"
echo -n "Sizing 1GB test file..."
# File size is 1GB + 3KB
$TRUNCATE "$dir/1gb" 1073744896 || error "cannot create 1GB test file"
echo "done"
# need at least 512MB per OST for 1GB file to fit in 2 stripes
local avail=$($LCTL get_param -n llite.$FSNAME*.kbytesavail)
if (( avail > 524288 * OSTCOUNT )); then
echo -n "Migrating 1GB file..."
$LFS migrate -A -K 1 "$dir/1gb" ||
error "cannot migrate 1GB file"
echo "done"
echo -n "Verifying autostripe count is sqrt(n) + 1..."
stripe_count=$($LFS getstripe -c "$dir/1gb") ||
error "cannot getstripe for 1GB file"
[ $stripe_count -eq 2 ] ||
error "unexpected stripe count $stripe_count != 2"
echo "done"
fi
# Test 3: File is too large to fit within the available space on
# sqrt(n) + 1 OSTs. Simulate limited available space with -X
if [ $OSTCOUNT -ge 3 ]; then
# The required available space is calculated as
# file size (1GB + 3KB) / OST count (3).
local kb_per_ost=349526
echo -n "Migrating 1GB file with limit..."
$LFS migrate -A -K 1 -X $kb_per_ost "$dir/1gb" ||
error "cannot migrate 1GB file with limit"
echo "done"
stripe_count=$($LFS getstripe -c "$dir/1gb")
echo -n "Verifying 1GB autostripe count with limited space..."
[ "$stripe_count" -a $stripe_count -ge 3 ] ||
error "unexpected stripe count $stripe_count (min 3)"
echo "done"
fi
# clean up
rm -rf $dir
}
run_test 56xc "lfs migration autostripe"
test_56xd() {
[[ $OSTCOUNT -lt 2 ]] && skip_env "needs >= 2 OSTs"
local dir=$DIR/$tdir
local f_mgrt=$dir/$tfile.mgrt
local f_yaml=$dir/$tfile.yaml
local f_copy=$dir/$tfile.copy
local layout_yaml="-E 1M -S 512K -c 1 -E -1 -S 1M -c 2 -i 0"
local layout_copy="-c 2 -S 2M -i 1"
local yamlfile=$dir/yamlfile
local layout_before;
local layout_after;
test_mkdir "$dir" || error "cannot create dir $dir"
stack_trap "rm -rf $dir"
$LFS setstripe $layout_yaml $f_yaml ||
error "cannot setstripe $f_yaml with layout $layout_yaml"
$LFS getstripe --yaml $f_yaml > $yamlfile
$LFS setstripe $layout_copy $f_copy ||
error "cannot setstripe $f_copy with layout $layout_copy"
touch $f_mgrt
dd if=/dev/zero of=$f_mgrt bs=1M count=4
# 1. test option --yaml
$LFS migrate --yaml $yamlfile $f_mgrt ||
error "cannot migrate $f_mgrt with --yaml $yamlfile"
layout_before=$(get_layout_param $f_yaml)
layout_after=$(get_layout_param $f_mgrt)
[ "$layout_after" == "$layout_before" ] ||
error "lfs migrate --yaml: $layout_after != $layout_before"
# 2. test option --copy
$LFS migrate --copy $f_copy $f_mgrt ||
error "cannot migrate $f_mgrt with --copy $f_copy"
layout_before=$(SKIP_INDEX=yes get_layout_param $f_copy)
layout_after=$(SKIP_INDEX=yes get_layout_param $f_mgrt)
[ "$layout_after" == "$layout_before" ] ||
error "lfs migrate --copy: $layout_after != $layout_before"
}
run_test 56xd "check lfs migrate --yaml and --copy support"
test_56xe() {
(( $OSTCOUNT >= 2 )) || skip_env "needs >= 2 OSTs"
local dir=$DIR/$tdir
local f_comp=$dir/$tfile
local layout="-E 1M -S 512K -E 2M -c 2 -E 3M -E eof -c $((OSTCOUNT-1))"
local layout_before=""
local layout_after=""
test_mkdir "$dir" || error "cannot create dir $dir"
stack_trap "rm -rf $dir"
$LFS setstripe $layout $f_comp ||
error "cannot setstripe $f_comp with layout $layout"
layout_before=$(SKIP_INDEX=yes get_layout_param $f_comp)
dd if=/dev/zero of=$f_comp bs=1M count=4
# extract starting OST index of each component before migration
idxs_before=$($LFS getstripe $f_comp | awk '$2 == "0:" { print $5 }' |
tr '\n' ' ')
# migrate a comp layout file by lfs migrate
$LFS migrate $f_comp || error "cannot migrate $f_comp by lfs migrate"
layout_after=$(SKIP_INDEX=yes get_layout_param $f_comp)
# extract starting OST index of each component after migration
idxs_after=$($LFS getstripe $f_comp | awk '$2 == "0:" { print $5 }' |
tr '\n' ' ')
[ "$layout_before" == "$layout_after" ] ||
error "lfs migrate: layout $layout_before != $layout_after"
# compare starting indexes of all components before/after migration.
# this may not fail every time with a broken lfs migrate, but will fail
# often enough to notice, and will not have false positives very often
[[ "$idxs_before" != "$idxs_after" ]] ||
error "lfs migrate: indexes $idxs_before == $idxs_after"
}
run_test 56xe "migrate a composite layout file"
test_56xf() {
[[ $OSTCOUNT -ge 2 ]] || skip_env "needs >= 2 OSTs"
[[ $MDS1_VERSION -ge $(version_code 2.13.53) ]] ||
skip "Need server version at least 2.13.53"
local dir=$DIR/$tdir
local f_comp=$dir/$tfile
local layout="-E 1M -c1 -E -1 -c2"
local fid_before=""
local fid_after=""
test_mkdir "$dir" || error "cannot create dir $dir"
stack_trap "rm -rf $dir"
$LFS setstripe $layout $f_comp ||
error "cannot setstripe $f_comp with layout $layout"
fid_before=$($LFS getstripe --fid $f_comp)
dd if=/dev/zero of=$f_comp bs=1M count=4
# 1. migrate a comp layout file to a comp layout
$LFS migrate $f_comp || error "cannot migrate $f_comp by lfs migrate"
fid_after=$($LFS getstripe --fid $f_comp)
[ "$fid_before" == "$fid_after" ] ||
error "comp-to-comp migrate: $fid_before != $fid_after"
# 2. migrate a comp layout file to a plain layout
$LFS migrate -c2 $f_comp ||
error "cannot migrate $f_comp by lfs migrate"
fid_after=$($LFS getstripe --fid $f_comp)
[ "$fid_before" == "$fid_after" ] ||
error "comp-to-plain migrate: $fid_before != $fid_after"
# 3. migrate a plain layout file to a comp layout
$LFS migrate $layout $f_comp ||
error "cannot migrate $f_comp by lfs migrate"
fid_after=$($LFS getstripe --fid $f_comp)
[ "$fid_before" == "$fid_after" ] ||
error "plain-to-comp migrate: $fid_before != $fid_after"
}
run_test 56xf "FID is not lost during migration of a composite layout file"
check_file_ost_range() {
local file="$1"
shift
local range="$*"
local -a file_range
local idx
file_range=($($LFS getstripe -y "$file" |
awk '/l_ost_idx:/ { print $NF }'))
if [[ "${#file_range[@]}" = 0 ]]; then
echo "No osts found for $file"
return 1
fi
for idx in "${file_range[@]}"; do
[[ " $range " =~ " $idx " ]] ||
return 1
done
return 0
}
sub_test_56xg() {
local stripe_opt="$1"
local pool="$2"
shift 2
local pool_ostidx="$(seq $* | tr '\n' ' ')"
$LFS migrate $stripe_opt -p $pool $DIR/$tfile ||
error "Fail to migrate $tfile on $pool"
[[ "$($LFS getstripe -p $DIR/$tfile)" = "$pool" ]] ||
error "$tfile is not in pool $pool"
check_file_ost_range "$DIR/$tfile" $pool_ostidx ||
error "$tfile osts mismatch with pool $pool (osts $pool_ostidx)"
}
test_56xg() {
[[ $PARALLEL != "yes" ]] || skip "skip parallel run"
[[ $OSTCOUNT -ge 2 ]] || skip "needs >= 2 OSTs"
[[ $MDS1_VERSION -gt $(version_code 2.14.52) ]] ||
skip "Need MDS version newer than 2.14.52"
local -a pool_names=("${TESTNAME}_0" "${TESTNAME}_1" "${TESTNAME}_2")
local -a pool_ranges=("0 0" "1 1" "0 1")
# init pools
for i in "${!pool_names[@]}"; do
pool_add ${pool_names[$i]} ||
error "pool_add failed (pool: ${pool_names[$i]})"
pool_add_targets ${pool_names[$i]} ${pool_ranges[$i]} ||
error "pool_add_targets failed (pool: ${pool_names[$i]})"
done
# init the file to migrate
$LFS setstripe -c1 -i1 $DIR/$tfile ||
error "Unable to create $tfile on OST1"
stack_trap "rm -f $DIR/$tfile"
dd if=/dev/urandom of=$DIR/$tfile bs=1M count=4 status=none ||
error "Unable to write on $tfile"
echo "1. migrate $tfile on pool ${pool_names[0]}"
sub_test_56xg "-c-1" "${pool_names[0]}" ${pool_ranges[0]}
echo "2. migrate $tfile on pool ${pool_names[2]}"
sub_test_56xg "-c-1 -S2M" "${pool_names[2]}" ${pool_ranges[2]}
echo "3. migrate $tfile on pool ${pool_names[1]}"
sub_test_56xg "-n -c-1" "${pool_names[1]}" ${pool_ranges[1]}
echo "4. migrate $tfile on pool ${pool_names[2]} with default stripe parameters"
sub_test_56xg "" "${pool_names[2]}" ${pool_ranges[2]}
echo
# Clean pools
destroy_test_pools ||
error "pool_destroy failed"
}
run_test 56xg "lfs migrate pool support"
test_56xh() {
(( $OSTCOUNT >= 2 )) || skip "needs >= 2 OSTs"
local size_mb=25
local file1=$DIR/$tfile
local tmp1=$TMP/$tfile.tmp
$LFS setstripe -c 2 $file1
stack_trap "rm -f $file1 $tmp1"
dd if=/dev/urandom of=$tmp1 bs=1M count=$size_mb ||
error "error creating $tmp1"
ls -lsh $tmp1
cp $tmp1 $file1
local start=$SECONDS
$LFS migrate --stats --stats-interval=1 -W 1M -c 1 $file1 ||
error "migrate failed rc = $?"
local elapsed=$((SECONDS - start))
# with 1MB/s, elapsed should equal size_mb
(( elapsed >= size_mb * 95 / 100 )) ||
error "'lfs migrate -W' too fast ($elapsed < 0.95 * $size_mb)?"
(( elapsed <= size_mb * 120 / 100 )) ||
error_not_in_vm "'lfs migrate -W' slow ($elapsed > 1.2 * $size_mb)"
(( elapsed <= size_mb * 350 / 100 )) ||
error "'lfs migrate -W' too slow in VM ($elapsed > 3.5 * $size_mb)"
stripe=$($LFS getstripe -c $file1)
(( $stripe == 1 )) || error "stripe of $file1 is $stripe != 1"
cmp $file1 $tmp1 || error "content mismatch $file1 differs from $tmp1"
# Clean up file (since it is multiple MB)
rm -f $file1 $tmp1
}
run_test 56xh "lfs migrate bandwidth limitation support"
test_56xi() {
(( $OSTCOUNT >= 2 )) || skip "needs >= 2 OSTs"
local size_mb=5
local file1=$DIR/$tfile.1
local file2=$DIR/$tfile.2
local file3=$DIR/$tfile.3
local output_file=$DIR/$tfile.out
local tmp1=$TMP/$tfile.tmp
$LFS setstripe -c 2 $file1
$LFS setstripe -c 2 $file2
$LFS setstripe -c 2 $file3
stack_trap "rm -f $file1 $file2 $file3 $tmp1 $output_file"
dd if=/dev/urandom of=$tmp1 bs=1M count=$size_mb ||
error "error creating $tmp1"
ls -lsh $tmp1
cp $tmp1 $file1
cp $tmp1 $file2
cp $tmp1 $file3
$LFS migrate --stats --stats-interval=1 \
-c 1 $file1 $file2 $file3 1> $output_file ||
error "migrate failed rc = $?"
cat $output_file
verify_yaml_available || skip_env "YAML verification not installed"
verify_yaml < $output_file || error "rename_stats is not valid YAML"
}
run_test 56xi "lfs migrate stats support"
test_56xj() { # LU-16571 "lfs migrate -b" can cause thread starvation on OSS
(( $OSTCOUNT >= 2 )) || skip "needs >= 2 OSTs"
local file=$DIR/$tfile
local linkdir=$DIR/$tdir
test_mkdir $linkdir || error "fail to create $linkdir"
$LFS setstripe -i 0 -c 1 -S1M $file
stack_trap "rm -rf $file $linkdir"
dd if=/dev/urandom of=$file bs=1M count=10 ||
error "fail to create $file"
# Create file links
local cpts
local threads_max
local nlinks
thread_max=$(do_facet ost1 "$LCTL get_param -n ost.OSS.ost.threads_max")
cpts=$(do_facet ost1 "$LCTL get_param -n cpu_partition_table | wc -l")
(( nlinks = thread_max * 3 / 2 / cpts))
echo "create $nlinks hard links of $file"
createmany -l $file $linkdir/link $nlinks
# Parallel migrates (should not block)
local i
for ((i = 0; i < nlinks; i++)); do
echo $linkdir/link$i
done | xargs -n1 -P $nlinks $LFS migrate -c2
local stripe_count
stripe_count=$($LFS getstripe -c $file) ||
error "fail to get stripe count on $file"
((stripe_count == 2)) ||
error "fail to migrate $file (stripe_count = $stripe_count)"
}
run_test 56xj "lfs migrate -b should not cause starvation of threads on OSS"
test_56xk() {
(( $OSTCOUNT >= 2 )) || skip "needs >= 2 OSTs"
local size_mb=5
local file1=$DIR/$tfile
stack_trap "rm -f $file1"
$LFS setstripe -c 1 $file1
dd if=/dev/zero of=$file1 bs=1M count=$size_mb ||
error "error creating $file1"
$LFS mirror extend -N $file1 || error "can't mirror"
dd if=/dev/zero of=$file1 bs=4k count=1 conv=notrunc ||
error "can't dd"
$LFS getstripe $file1 | grep stale ||
error "one component must be stale"
local start=$SECONDS
$LFS mirror resync --stats --stats-interval=1 -W 1M $file1 ||
error "migrate failed rc = $?"
local elapsed=$((SECONDS - start))
$LFS getstripe $file1 | grep stale &&
error "all components must be sync"
# with 1MB/s, elapsed should equal size_mb
(( elapsed >= size_mb * 95 / 100 )) ||
error "'lfs mirror resync -W' too fast ($elapsed < 0.95 * $size_mb)?"
(( elapsed <= size_mb * 120 / 100 )) ||
error_not_in_vm "'lfs mirror resync -W' slow ($elapsed > 1.2 * $size_mb)"
(( elapsed <= size_mb * 350 / 100 )) ||
error "'lfs mirror resync -W' too slow in VM ($elapsed > 3.5 * $size_mb)"
}
run_test 56xk "lfs mirror resync bandwidth limitation support"
test_56xl() {
(( $OSTCOUNT >= 2 )) || skip "needs >= 2 OSTs"
local size_mb=5
local file1=$DIR/$tfile.1
local output_file=$DIR/$tfile.out
stack_trap "rm -f $file1"
$LFS setstripe -c 1 $file1
dd if=/dev/zero of=$file1 bs=1M count=$size_mb ||
error "error creating $file1"
$LFS mirror extend -N $file1 || error "can't mirror"
dd if=/dev/zero of=$file1 bs=4k count=1 conv=notrunc ||
error "can't dd"
$LFS getstripe $file1 | grep stale ||
error "one component must be stale"
$LFS getstripe $file1
$LFS mirror resync --stats --stats-interval=1 $file1 >$output_file ||
error "resync failed rc = $?"
$LFS getstripe $file1 | grep stale &&
error "all components must be sync"
cat $output_file
verify_yaml_available || skip_env "YAML verification not installed"
verify_yaml < $output_file || error "stats is not valid YAML"
}
run_test 56xl "lfs mirror resync stats support"
test_56y() {
[ $MDS1_VERSION -lt $(version_code 2.4.53) ] &&
skip "No HSM $(lustre_build_version $SINGLEMDS) MDS < 2.4.53"
local res=""
local dir=$DIR/$tdir
local f1=$dir/file1
local f2=$dir/file2
test_mkdir -p $dir || error "creating dir $dir"
touch $f1 || error "creating std file $f1"
$MULTIOP $f2 H2c || error "creating released file $f2"
# a directory can be raid0, so ask only for files
res=$($LFS find $dir -L raid0 -type f | wc -l)
[[ $res == 2 ]] || error "search raid0: found $res files != 2"
res=$($LFS find $dir \! -L raid0 -type f | wc -l)
[[ $res == 0 ]] || error "search !raid0: found $res files != 0"
# only files can be released, so no need to force file search
res=$($LFS find $dir -L released)
[[ $res == $f2 ]] || error "search released: found $res != $f2"
res=$($LFS find $dir -type f \! -L released)
[[ $res == $f1 ]] || error "search !released: found $res != $f1"
}
run_test 56y "lfs find -L raid0|released"
test_56z() { # LU-4824
# This checks to make sure 'lfs find' continues after errors
# There are two classes of errors that should be caught:
# - If multiple paths are provided, all should be searched even if one
# errors out
# - If errors are encountered during the search, it should not terminate
# early
local dir=$DIR/$tdir
local i
test_mkdir $dir
for i in d{0..9}; do
test_mkdir $dir/$i
touch $dir/$i/$tfile
done
$LFS find $DIR/non_existent_dir $dir &&
error "$LFS find did not return an error"
# Make a directory unsearchable. This should NOT be the last entry in
# directory order. Arbitrarily pick the 6th entry
chmod 700 $($LFS find $dir -type d | sed '6!d')
$RUNAS $LFS find $DIR/non_existent $dir
local count=$($RUNAS $LFS find $DIR/non_existent $dir | wc -l)
# The user should be able to see 10 directories and 9 files
(( count == 19 )) ||
error "$LFS find found $count != 19 entries after error"
}
run_test 56z "lfs find should continue after an error"
test_56Aa() { # was test_56aa, LU-5937
[ $MDSCOUNT -lt 2 ] && skip_env "needs >= 2 MDTs"
local dir=$DIR/$tdir
mkdir $dir
$LFS setdirstripe -c$MDSCOUNT $dir/striped_dir
createmany -o $dir/striped_dir/${tfile}- 1024
local dirs=$($LFS find --size +7k $dir/)
[ -n "$dirs" ] || error "lfs find --size wrong under striped dir"
}
run_test 56Aa "lfs find --size under striped dir"
test_56ab() { # LU-10705
test_mkdir $DIR/$tdir
dd if=/dev/urandom of=$DIR/$tdir/$tfile.1 bs=8k count=1 seek=2k
dd if=/dev/urandom of=$DIR/$tdir/$tfile.2 bs=4k count=1 seek=4k
dd if=/dev/urandom of=$DIR/$tdir/$tfile.3 bs=1M count=2 seek=16
# Flush writes to ensure valid blocks. Need to be more thorough for
# ZFS, since blocks are not allocated/returned to client immediately.
sync_all_data
wait_zfs_commit ost1 2
cancel_lru_locks osc
ls -ls $DIR/$tdir
local files=$($LFS find --size +16M $DIR/$tdir | wc -l)
[[ $files == 3 ]] || error ">16M size files $files isn't 3 as expected"
files=$($LFS find --blocks +1M $DIR/$tdir | wc -l)
[[ $files == 1 ]] || error ">1M blocks files $files isn't 1 as expected"
rm -f $DIR/$tdir/$tfile.[123]
}
run_test 56ab "lfs find --blocks"
# LU-11188
test_56aca() {
local dir="$DIR/$tdir"
local perms=(001 002 003 004 005 006 007
010 020 030 040 050 060 070
100 200 300 400 500 600 700
111 222 333 444 555 666 777)
local perm_minus=(8 8 4 8 4 4 2
8 8 4 8 4 4 2
8 8 4 8 4 4 2
4 4 2 4 2 2 1)
local perm_slash=(8 8 12 8 12 12 14
8 8 12 8 12 12 14
8 8 12 8 12 12 14
16 16 24 16 24 24 28)
test_mkdir "$dir"
for perm in ${perms[*]}; do
touch "$dir/$tfile.$perm"
chmod $perm "$dir/$tfile.$perm"
done
for ((i = 0; i < ${#perms[*]}; i++)); do
local num=$($LFS find $dir -perm ${perms[i]} | wc -l)
(( $num == 1 )) ||
error "lfs find -perm ${perms[i]}:"\
"$num != 1"
num=$($LFS find $dir -perm -${perms[i]} -type f| wc -l)
(( $num == ${perm_minus[i]} )) ||
error "lfs find -perm -${perms[i]}:"\
"$num != ${perm_minus[i]}"
num=$($LFS find $dir -perm /${perms[i]} -type f| wc -l)
(( $num == ${perm_slash[i]} )) ||
error "lfs find -perm /${perms[i]}:"\
"$num != ${perm_slash[i]}"
done
}
run_test 56aca "check lfs find -perm with octal representation"
test_56acb() {
local dir=$DIR/$tdir
# p is the permission of write and execute for user, group and other
# without the umask. It is used to test +wx.
local p=$(printf "%o" "$((0333 & ~$(umask)))")
local perms=(1000 000 2000 4000 $p 644 111 110 100 004)
local symbolic=(+t a+t u+t g+t o+t
g+s u+s o+s +s o+sr
o=r,ug+o,u+w
u+ g+ o+ a+ ugo+
u- g- o- a- ugo-
u= g= o= a= ugo=
o=r,ug+o,u+w u=r,a+u,u+w
g=r,ugo=g,u+w u+x,+X +X
u+x,u+X u+X u+x,g+X o+r,+X
u+x,go+X +wx +rwx)
test_mkdir $dir
for perm in ${perms[*]}; do
touch "$dir/$tfile.$perm"
chmod $perm "$dir/$tfile.$perm"
done
for (( i = 0; i < ${#symbolic[*]}; i++ )); do
local num=$($LFS find $dir -perm ${symbolic[i]} | wc -l)
(( $num == 1 )) ||
error "lfs find $dir -perm ${symbolic[i]}: $num != 1"
done
}
run_test 56acb "check lfs find -perm with symbolic representation"
test_56acc() {
local dir=$DIR/$tdir
local tests="17777 787 789 abcd
ug=uu ug=a ug=gu uo=ou urw
u+xg+x a=r,u+x,"
test_mkdir $dir
for err in $tests; do
if $LFS find $dir -perm $err 2>/dev/null; then
error "lfs find -perm $err: parsing should have failed"
fi
done
}
run_test 56acc "check parsing error for lfs find -perm"
test_56Ba() { # was test_56ba
[ $MDS1_VERSION -lt $(version_code 2.10.50) ] &&
skip "Need MDS version at least 2.10.50"
# Create composite files with one component
local dir=$DIR/$tdir
setup_56 $dir/1Mfiles 5 1 "-S 1M --component-end 1M"
# Create composite files with three components
setup_56 $dir/2Mfiles 5 2 "-E 2M -S 1M -E 4M -E 6M"
# LU-16904 Create plain layout files
lfs setstripe -c 1 $dir/$tfile-{1..10}
local nfiles=$($LFS find --component-end 1M --type f $dir | wc -l)
[[ $nfiles == 10 ]] ||
error "lfs find -E 1M found $nfiles != 10 files"
nfiles=$($LFS find ! -E 1M --type f $dir | wc -l)
[[ $nfiles == 25 ]] ||
error "lfs find ! -E 1M found $nfiles != 25 files"
# All files have a component that starts at 0
nfiles=$($LFS find --component-start 0 --type f $dir | wc -l)
[[ $nfiles == 35 ]] ||
error "lfs find --component-start 0 - $nfiles != 35 files"
nfiles=$($LFS find --component-start 2M --type f $dir | wc -l)
[[ $nfiles == 15 ]] ||
error "lfs find --component-start 2M - $nfiles != 15 files"
# All files created here have a componenet that does not starts at 2M
nfiles=$($LFS find ! --component-start 2M --type f $dir | wc -l)
[[ $nfiles == 35 ]] ||
error "lfs find ! --component-start 2M - $nfiles != 35 files"
# Find files with a specified number of components
local nfiles=$($LFS find --component-count 3 --type f $dir | wc -l)
[[ $nfiles == 15 ]] ||
error "lfs find --component-count 3 - $nfiles != 15 files"
# Remember non-composite files have a component count of zero
local nfiles=$($LFS find --component-count 0 --type f $dir | wc -l)
[[ $nfiles == 10 ]] ||
error "lfs find --component-count 0 - $nfiles != 10 files"
nfiles=$($LFS find ! --component-count 3 --type f $dir | wc -l)
[[ $nfiles == 20 ]] ||
error "lfs find ! --component-count 3 - $nfiles != 20 files"
# All files have a flag called "init"
local nfiles=$($LFS find --component-flags init --type f $dir | wc -l)
[[ $nfiles == 35 ]] ||
error "lfs find --component-flags init - $nfiles != 35 files"
# Multi-component files will have a component not initialized
local nfiles=$($LFS find ! --component-flags init --type f $dir | wc -l)
[[ $nfiles == 15 ]] ||
error "lfs find !--component-flags init - $nfiles != 15 files"
rm -rf $dir
}
run_test 56Ba "test lfs find --component-end, -start, -count, and -flags"
test_56Ca() { # was (also) test_56ca
[[ $MDS1_VERSION -ge $(version_code 2.10.57) ]] ||
skip "Need MDS version at least 2.10.57"
local td=$DIR/$tdir
local tf=$td/$tfile
local dir
local nfiles
local cmd
local i
local j
# create mirrored directories and mirrored files
mkdir $td || error "mkdir $td failed"
$LFS mirror create -N3 $td || error "create mirrored dir $td failed"
createmany -o $tf- 10 || error "create $tf- failed"
for i in $(seq 2); do
dir=$td/dir$i
mkdir $dir || error "mkdir $dir failed"
$LFS mirror create -N$((3 + i)) $dir ||
error "create mirrored dir $dir failed"
createmany -o $dir/$tfile- 10 ||
error "create $dir/$tfile- failed"
done
# change the states of some mirrored files
echo foo > $tf-6
for i in $(seq 2); do
dir=$td/dir$i
for j in $(seq 4 9); do
echo foo > $dir/$tfile-$j
done
done
# find mirrored files with specific mirror count
cmd="$LFS find --mirror-count 3 --type f $td"
nfiles=$($cmd | wc -l)
[[ $nfiles = 10 ]] || error "$cmd: $nfiles != 10 files"
cmd="$LFS find ! --mirror-count 3 --type f $td"
nfiles=$($cmd | wc -l)
[[ $nfiles = 20 ]] || error "$cmd: $nfiles != 20 files"
cmd="$LFS find --mirror-count +2 --type f $td"
nfiles=$($cmd | wc -l)
[[ $nfiles = 30 ]] || error "$cmd: $nfiles != 30 files"
cmd="$LFS find --mirror-count -6 --type f $td"
nfiles=$($cmd | wc -l)
[[ $nfiles = 30 ]] || error "$cmd: $nfiles != 30 files"
# find mirrored files with specific file state
cmd="$LFS find --maxdepth 1 --mirror-state=^ro --type f $td"
[[ $($cmd) = $tf-6 ]] || error "$cmd: didn't return $tf-6"
cmd="$LFS find --mirror-state=ro --type f $td"
nfiles=$($cmd | wc -l)
[[ $nfiles = 17 ]] || error "$cmd: $nfiles != 17 files"
cmd="$LFS find ! --mirror-state=ro --type f $td"
nfiles=$($cmd | wc -l)
[[ $nfiles = 13 ]] || error "$cmd: $nfiles != 13 files"
cmd="$LFS find --mirror-state=wp --type f $td"
nfiles=$($cmd | wc -l)
[[ $nfiles = 13 ]] || error "$cmd: $nfiles != 13 files"
cmd="$LFS find ! --mirror-state=sp --type f $td"
nfiles=$($cmd | wc -l)
[[ $nfiles = 30 ]] || error "$cmd: $nfiles != 30 files"
}
run_test 56Ca "check lfs find --mirror-count|-N and --mirror-state"
test_56Da() { # was test_56da, LU-14179
local path=$DIR/$tdir
test_mkdir $path
cd $path
local longdir=$(str_repeat 'a' 255)
for i in {1..15}; do
path=$path/$longdir
test_mkdir $longdir
cd $longdir
done
local len=${#path}
local lastdir=$(str_repeat 'a' $((4096 - 1 - $len - 1)))
test_mkdir $lastdir
cd $lastdir
# PATH_MAX-1
(( ${#PWD} == 4095 )) || error "bad PWD length ${#PWD}, expect 4095"
# NAME_MAX
touch $(str_repeat 'f' 255)
$LFS find $DIR/$tdir --type d |& grep "lfs find: error" &&
error "lfs find reported an error"
rm -rf $DIR/$tdir
}
run_test 56Da "test lfs find with long paths"
test_56Db() { # was test_56db
local mdts=$($LFS df -m | grep -c MDT)
local osts=$($LFS df -m | grep -c OST)
$LFS df
(( mdts == MDSCOUNT )) ||
error "lfs df -m showed $mdts MDTs, not $MDSCOUNT"
(( osts == 0 )) ||
error "lfs df -m showed $osts OSTs, not 0"
}
run_test 56Db "test 'lfs df -m' only shows MDT devices"
test_56Dc() { # was test_56dc
local mdts=$($LFS df -o | grep -c MDT)
local osts=$($LFS df -o | grep -c OST)
$LFS df
(( osts == OSTCOUNT )) ||
error "lfs df -o showed $osts OSTs, not $OSTCOUNT"
(( mdts == 0 )) ||
error "lfs df -o showed $mdts MDTs, not 0"
}
run_test 56Dc "test 'lfs df -o' only shows OST devices"
test_56Dd() { # was test_56dd
local dir=$DIR/d$(basetest $testnum)g.$TESTSUITE
setup_56 $dir $NUMFILES $NUMDIRS
local lfscount=($($LFS find $dir -mindepth 1 -maxdepth 2 | sort))
local findcount=($(find $dir -mindepth 1 -maxdepth 2 | sort))
if [[ ${#lfscount[@]} != ${#findcount[@]} ]]; then
error "lfs find returned ${#lfscount[@]} files, find returned ${#findcount[@]} files"
fi
for ((i = 0; i < ${#lfscount[@]}; i++)); do
[[ "${lfscount[$i]}" == "${findcount[$i]}" ]] &&
echo "${lfscount[$i]}" ||
error "${#lfscount[@]} != ${#findcount[@]}"
done
}
run_test 56Dd "test lfs find with mindepth argument"
test_56Ea() { #LU-10378
local path=$DIR/$tdir
local pool=$TESTNAME
# Create ost pool
pool_add $pool || error "pool_add $pool failed"
pool_add_targets $pool 0 $((OSTCOUNT - 1)) 1 ||
error "adding targets to $pool failed"
# Set default pool on directory before creating file
mkdir $path || error "mkdir $path failed"
$LFS setstripe -p $pool $path ||
error "set OST pool on $pool failed"
touch $path/$tfile || error "touch $path/$tfile failed"
# Compare basic file attributes from -printf and stat
local attr_printf=$($LFS find $path/$tfile -printf "%A@ %T@ %C@ %U %G %n")
local attr_stat=$(stat -c "%X %Y %Z %u %g %h" $path/$tfile)
[[ "${attr_printf}" == "${attr_stat}" ]] ||
error "Attrs from lfs find and stat don't match"
# Compare Lustre attributes from lfs find and lfs getstripe
local lattr_printf=$($LFS find $path/$tfile -printf "%Lc %LS %Li %Lp")
local str_cnt=$($LFS getstripe --stripe-count $path/$tfile)
local str_size=$($LFS getstripe --stripe-size $path/$tfile)
local str_idx=$($LFS getstripe --stripe-index $path/$tfile)
local fpool=$($LFS getstripe --pool $path/$tfile)
local lattr_getstr="${str_cnt} ${str_size} ${str_idx} ${fpool}"
[[ "${lattr_printf}" == "${lattr_getstr}" ]] ||
error "Attrs from lfs find and lfs getstripe don't match"
# Verify behavior for unknown escape/format sequences
local esc_printf=$($LFS find $path/$tfile -printf '\\ %% \Q %Q')
[[ "${esc_printf}" == '\ % \Q %Q' ]] ||
error "Escape/format codes don't match"
}
run_test 56Ea "test lfs find -printf option"
test_56Eaa() {
local lfs_find=($($LFS find $DIR -printf "%i %u %g %M %p\n" | sort -n))
local normal_find=($(find $DIR -printf "%i %u %g %M %p\n" | sort -n))
echo "comparing ${#normal_find[@]} elements starting with:"
echo "lfs: ${lfs_find[0]} ${lfs_find[1]} ${lfs_find[2]} ${lfs_find[3]} ${lfs_find[4]}"
echo "find: ${normal_find[0]} ${normal_find[1]} ${normal_find[2]} ${normal_find[3]} ${normal_find[4]}"
for ((i = 0; i < ${#normal_find[@]}; i++)); do
if [[ "${lfs_find[i]}" != "${normal_find[i]}" ]]; then
error "expected '${normal_find[i]}' but got '${lfs_find[i]}' from file ${lfs_find[$((i - (i % 5) + 4))]}"
fi
done
}
run_test 56Eaa "test lfs find -printf added functions"
test_56Eab() {
touch $DIR/$tfile
local lfs_ls=($($LFS find $DIR -name $tfile -ls))
local find_ls=($(find $DIR -name $tfile -ls))
# "-1" is last field, since dates are not printed exactly the same.
for ((i = -1; i < 7; i++)); do
[[ "${lfs_ls[i]}" == "${find_ls[i]}" ]] ||
error "expected '${lfs_ls[i]}' but got '${find_ls[i]}'"
done
}
run_test 56Eab "test lfs find -ls function"
test_56Eb() {
local dir=$DIR/$tdir
local subdir_1=$dir/subdir_1
test_mkdir -p $subdir_1
ln -s subdir_1 $dir/link_1
$LFS getstripe $dir | grep "^$dir/link_1$" -A1 ||
error "symlink is not followed"
$LFS getstripe --no-follow $dir |
grep "^$dir/link_1 has no stripe info$" ||
error "symlink should not have stripe info"
touch $dir/testfile
ln -s testfile $dir/file_link_2
$LFS getstripe $dir | grep "^$dir/file_link_2$" -A1 ||
error "symlink is not followed"
$LFS getstripe --no-follow $dir |
grep "^$dir/file_link_2 has no stripe info$" ||
error "symlink should not have stripe info"
}
run_test 56Eb "check lfs getstripe on symlink"
test_56Ebb() {
mkdir $DIR/$tdir
mkfifo $DIR/$tdir/$tfile-fifo
$LFS getdirstripe -r $DIR/$tdir || error "$LFS getdirstripe -r: $DIR"
}
run_test 56Ebb "check $LFS getdirstripe for FIFO file"
test_56Ec() {
[[ $OSTCOUNT -lt 2 ]] && skip_env "needs >= 2 OSTs"
local dir=$DIR/$tdir
local srcfile=$dir/srcfile
local srcyaml=$dir/srcyaml
local destfile=$dir/destfile
test_mkdir -p $dir
$LFS setstripe -i 1 $srcfile
$LFS getstripe --hex-idx --yaml $srcfile > $srcyaml
# if the setstripe yaml parsing fails for any reason, the command can
# randomly assign the correct OST index, leading to an erroneous
# success. but the chance of false success is low enough that a
# regression should still be quickly caught.
$LFS setstripe --yaml=$srcyaml $destfile
local srcindex=$($LFS getstripe -i $srcfile)
local destindex=$($LFS getstripe -i $destfile)
if [[ ! $srcindex -eq $destindex ]]; then
error "setstripe did not set OST index correctly"
fi
}
run_test 56Ec "check lfs getstripe,setstripe --hex --yaml"
# LU-15565
test_56Ed() {
local new=$DIR/$tdir/$tfile-layout-new
local old=$DIR/$tdir/$tfile-layout-old
test_mkdir $DIR/$tdir
cat << PFL_LAYOUT_NEW > $new
lcm_layout_gen: 3
lcm_mirror_count: 1
lcm_entry_count: 3
components:
- lcme_id: 1
lcme_mirror_id: 0
lcme_flags: init
lcme_extent.e_start: 0
lcme_extent.e_end: 1073741824
sub_layout:
lmm_stripe_count: 1
lmm_stripe_size: 4194304
lmm_pattern: raid0
lmm_layout_gen: 0
lmm_stripe_offset: 0
lmm_objects:
- l_ost_idx: 0
l_fid: 0x280000400:0x2:0x0
- lcme_id: 2
lcme_mirror_id: 0
lcme_flags: 0
lcme_extent.e_start: 1073741824
lcme_extent.e_end: 4294967296
sub_layout:
lmm_stripe_count: 4
lmm_stripe_size: 4194304
lmm_pattern: raid0
lmm_layout_gen: 0
lmm_stripe_offset: -1
- lcme_id: 3
lcme_mirror_id: 0
lcme_flags: 0
lcme_extent.e_start: 4294967296
lcme_extent.e_end: EOF
sub_layout:
lmm_stripe_count: 4
lmm_stripe_size: 4194304
lmm_pattern: raid0
lmm_layout_gen: 0
lmm_stripe_offset: -1
PFL_LAYOUT_NEW
# Layout changed in 2.16.0
cat << PFL_LAYOUT_OLD > $old
lcm_layout_gen: 3
lcm_mirror_count: 1
lcm_entry_count: 3
component0:
lcme_id: 1
lcme_mirror_id: 0
lcme_flags: init
lcme_extent.e_start: 0
lcme_extent.e_end: 1073741824
sub_layout:
lmm_stripe_count: 1
lmm_stripe_size: 4194304
lmm_pattern: raid0
lmm_layout_gen: 0
lmm_stripe_offset: 0
lmm_objects:
- l_ost_idx: 0
l_fid: 0x280000400:0x2:0x0
component1:
lcme_id: 2
lcme_mirror_id: 0
lcme_flags: 0
lcme_extent.e_start: 1073741824
lcme_extent.e_end: 4294967296
sub_layout:
lmm_stripe_count: 4
lmm_stripe_size: 4194304
lmm_pattern: raid0
lmm_layout_gen: 0
lmm_stripe_offset: -1
component2:
lcme_id: 3
lcme_mirror_id: 0
lcme_flags: 0
lcme_extent.e_start: 4294967296
lcme_extent.e_end: EOF
sub_layout:
lmm_stripe_count: 4
lmm_stripe_size: 4194304
lmm_pattern: raid0
lmm_layout_gen: 0
lmm_stripe_offset: -1
PFL_LAYOUT_OLD
local yaml_file=$DIR/$tdir/layout.yaml
local dir=$DIR/$tdir
local test=$dir/$tfile.test
$LFS setstripe -E 4M -c 2 -E EOF -c 4 $test ||
error "failed to setstripe"
$LFS getstripe --yaml $test > $yaml_file ||
error "failed to getstripe --yaml"
$LFS setstripe --yaml $old $dir/$tfile.new ||
error "failed to setstripe from new YAML format"
# test that the old format can still be used
$LFS setstripe --yaml $old $dir/$tfile.old ||
error "failed to setstripe from old YAML format"
local orig=$(get_layout_param $test)
local rest=$(cat $yaml_file | parse_layout_param)
[[ "$orig" == "$rest" ]] ||
error "failed to parse current YAML layout"
orig=$(get_layout_param $dir/$tfile.old)
rest=$(cat $old | parse_layout_param)
[[ "$orig" == "$rest" ]] ||
error "failed to parse old YAML layout"
orig=$(get_layout_param $dir/$tfile.new)
rest=$(cat $new | parse_layout_param)
[[ "$orig" == "$rest" ]] ||
error "failed to parse new YAML layout"
}
run_test 56Ed "verify new YAML format is valid and back-compatible"
test_56Eda() {
local dir=$DIR/$tdir
local subdir=$dir/subdir
local file1=$dir/$tfile
local file2=$dir/$tfile\2
local link=$dir/$tfile-link
local nfiles
test_mkdir -p $dir
$LFS setdirstripe -c1 $subdir
touch $file1
touch $file2
ln $file2 $link
nfiles=$($LFS find --links 1 $dir | wc -l)
(( $nfiles == 1 )) ||
error "lfs find --links expected 1 file, got $nfiles"
nfiles=$($LFS find --type f --links 2 $dir | wc -l)
(( $nfiles == 2 )) ||
error "lfs find --links expected 2 files, got $nfiles"
nfiles=$($LFS find --type d --links 2 $dir | wc -l)
(( $nfiles == 1 )) ||
error "lfs find --links expected 1 directory, got $nfiles"
}
run_test 56Eda "check lfs find --links"
test_56Edb() {
[[ $MDSCOUNT -lt 2 ]] && skip_env "needs >= 2 MDTs"
local dir=$DIR/$tdir
local stripedir=$dir/stripedir
local nfiles
test_mkdir -p $dir
$LFS setdirstripe -c2 $stripedir
$LFS getdirstripe $stripedir
nfiles=$($LFS find --type d --links 2 $stripedir | wc -l)
(( $nfiles == 1 )) ||
error "lfs find --links expected 1 directory, got $nfiles"
}
run_test 56Edb "check lfs find --links for directory striped on multiple MDTs"
test_56Ef() {
local dir=$DIR/$tdir
local dir1=$dir/d1
local dir2=$dir/d2
local nfiles
local err_msg
test_mkdir -p $dir
mkdir $dir1
mkdir $dir2
touch $dir1/f
touch $dir2/f
nfiles=$($LFS find $dir1 $dir2 ! -type d | wc -l)
(( $nfiles == 2 )) ||
error "(1) lfs find expected 2 files, got $nfiles"
nfiles=$($LFS find $dir1 $dir2 -type f | wc -l)
(( $nfiles == 2 )) ||
error "(2) lfs find expected 2 files, got $nfiles"
nfiles=$($LFS find -type f $dir1 $dir2 | wc -l)
(( $nfiles == 2 )) ||
error "(3) lfs find expected 2 files, got $nfiles"
err_msg=$($LFS find $dir1/typo $dir1/f 2>&1 > /dev/null)
[[ $err_msg =~ "No such file or directory" ]] ||
error "expected standard error message, got: '$err_msg'"
}
run_test 56Ef "lfs find with multiple paths"
test_56Eg() {
local dir=$DIR/$tdir
local found
which setfattr > /dev/null 2>&1 || skip_env "no setfattr command"
test_mkdir -p $dir
touch $dir/$tfile
ln -s $dir/$tfile $dir/$tfile.symlink
setfattr -n "trusted.test" -v "test_target" $dir/$tfile
setfattr --no-dereference -n "trusted.test" -v "test_link" \
$dir/$tfile.symlink
setfattr --no-dereference -n "trusted.common" \
$dir/{$tfile,$tfile.symlink}
found=$($LFS find -xattr "trusted.*=test_target" \
-xattr "trusted.common" $dir)
[[ "$found" == "$dir/$tfile" ]] || {
getfattr -d -m trusted.* $dir/$tfile
error "should have found '$tfile' with xattr 'trusted.test=test_target', got '$found'"
}
found=$($LFS find -xattr "trusted.*=test_link" \
-xattr "trusted.common" $dir)
[[ "$found" == "$dir/$tfile.symlink" ]] || {
getfattr --no-dereference -d -m trusted.* $dir/$tfile.symlink
error "should have found '$tfile.symlink' with xattr 'trusted.test=test_link', got '$found'"
}
rm -f $dir/*
touch $dir/$tfile.1
touch $dir/$tfile.2
setfattr -n "user.test" -v "1" $dir/$tfile.1
setfattr -n "user.test" -v "2" $dir/$tfile.2
setfattr -n "user.test2" -v "common" $dir/$tfile.{1,2}
found=$($LFS find -xattr "user.*=common" -xattr "user.test=1" $dir)
[[ "$found" == "$dir/$tfile.1" ]] || {
getfattr -d $dir/$tfile.1
error "should have found '$tfile.1' with xattr user.test=1', got '$found'"
}
found=$($LFS find -xattr "user.*=common" ! -xattr "user.test=1" $dir)
[[ "$found" == "$dir/$tfile.2" ]] || {
getfattr -d $dir/$tfile.2
error "should have found '$tfile.2' without xattr 'user.test=1', got '$found'"
}
setfattr -n "user.empty" $dir/$tfile.1
found=$($LFS find -xattr "user.empty" $dir)
[[ "$found" == "$dir/$tfile.1" ]] || {
getfattr -d $dir/$tfile.1
error "should have found '$tfile.1' with xattr 'user.empty=', got '$found'"
}
# setfattr command normally does not store terminating null byte
# when writing a string as an xattr value.
#
# In order to test matching a value string that includes a terminating
# null, explicitly encode the string "test\0" with the null terminator.
setfattr -n "user.test" -v "0x7465737400" $dir/$tfile.1
found=$($LFS find -xattr "user.test=test" $dir)
[[ "$found" == "$dir/$tfile.1" ]] || {
getfattr -d --encoding=hex $dir/$tfile.1
error "should have found '$tfile.1' with xattr 'user.test=0x7465737400', got '$found'"
}
}
run_test 56Eg "lfs find -xattr"
test_56Eh() {
local dir=$DIR/d$(basetest $testnum)g.$TESTSUITE
# Enough files to give us a statistically reliable sampling
# setup_special is complicated, but this works out to about 2100 files
setup_56_special $dir 400 $NUMDIRS
local cmd="$LFS find $dir"
local total=$($cmd | wc -l)
local n=1
local nums
echo "Running tests on $dir, with $total total results."
while (( $n < 100 )); do
nums=$($LFS find --skip=$((100 - $n)) $dir | wc -l)
(( $nums <= $(($total * ($n + 10) / 100)) &&
$nums >= $(($total * ($n - 10) / 100)) )) ||
error "--skip=$((100 - $n)): expected" \
"$(($total * $n / 100)) results with 10% error" \
"margin, got $nums"
(( n++ ))
done
}
run_test 56Eh "check lfs find --skip"
test_56Ei() {
(( $MDS1_VERSION >= $(version_code 2.15.64.110) )) ||
skip "need MDS >= v2_15_64-110-g501e5b2c8a for special projid"
local path=$DIR/$tdir
local projid=1234
local expected_count=3
# Create test dir containing:
# - regular file, with default projid
# - regular file, with unique projid
# - symbolic link, with unique projid
# - special char dev file, with unique projid
test_mkdir $path || error "mkdir $path failed"
touch $path/file0 $path/file$projid || error "touch failed"
ln -s $path/file$projid $path/link$projid
mknod $path/char$projid c 1 3
$LFS project -p $projid -s $path/file$projid
$LFS project -p $projid -s $path/link$projid
$LFS project -p $projid -s $path/char$projid
stack_trap "rm -rf $path" EXIT
$LFS project -r $path/
echo -e "Actual output:\n$($LFS find $path --printf '%LP %p\n')"
# Find all files and print their projids along with their path; count
found_count=$($LFS find $path --printf "%LP %p\n" |
grep -E "$projid $path/(link|char|file)$projid" | wc -l)
echo -e "found_count: $found_count"
[[ $found_count == $expected_count ]] ||
error "Did not find any entries with expected projid $projid"
}
run_test 56Ei "test lfs find --printf prints correct projid for special files"
test_56Ej() {
test_mkdir $DIR/$tdir.src ||
error "mkdir failed on $DIR/$tdir.src"
test_mkdir $DIR/$tdir.dest ||
error "mkdir failed on $DIR/$tdir.dest"
local f_mgrt=$DIR/$tdir.dest/$tfile.mgrt
$LFS setstripe -o 1 $DIR/$tdir.src
$LFS setstripe -o 0 $DIR/$tdir.dest
dd if=/dev/random of=$f_mgrt bs=1M count=1
$LFS migrate --non-block --copy $DIR/$tdir.src $f_mgrt ||
error "migrate remote dir error $DIR/$tdir.src $f_mgrt"
}
run_test 56Ej "lfs migration --non-block copy"
test_56Ek() {
local dir=$DIR/$tdir
local numfiles=10
local numdirs=5
# Setup test directory structure
echo "Creating test directory structure..."
mkdir -p $dir || error "Failed to create $dir"
# Create test directories and files
for ((i = 0; i < numdirs; i++)); do
mkdir -p $dir/subdir$i || error "Failed to create subdir$i"
createmany -o $dir/subdir$i/file $numfiles ||
error "Failed to create files in subdir$i"
done
# Normal find should work
echo "Verifying normal find works..."
local count=$($LFS find $dir | wc -l)
# dirs + files + root
local expected=$(($numdirs + ($numfiles * $numdirs) + 1))
[ $count -eq $expected ] ||
error "Wrong file count: found $count, expected $expected"
# Set failure injection
echo "Testing error handling..."
LLAPI_FAIL_LOC="LLAPI_FAIL_PFIND_SEM" \
$LFS find $dir 2>&1 | grep -q "Input/output error" ||
error "Expected I/O error not found with LLAPI_FAIL_LOC set"
# Cleanup
rm -rf $dir
return 0
}
run_test 56Ek "Test lfs find error handling with LLAPI_FAIL_LOC"
test_57a() {
[ $PARALLEL == "yes" ] && skip "skip parallel run"
# note test will not do anything if MDS is not local
if [ "$mds1_FSTYPE" != ldiskfs ]; then
skip_env "ldiskfs only test"
fi
remote_mds_nodsh && skip "remote MDS with nodsh"
local MNTDEV="osd*.*MDT*.mntdev"
DEV=$(do_facet $SINGLEMDS lctl get_param -n $MNTDEV)
[ -z "$DEV" ] && error "can't access $MNTDEV"
for DEV in $(do_facet $SINGLEMDS lctl get_param -n $MNTDEV); do
do_facet $SINGLEMDS $DUMPE2FS -h $DEV > $TMP/t57a.dump ||
error "can't access $DEV"
DEVISIZE=$(awk '/Inode size:/ { print $3 }' $TMP/t57a.dump)
[[ $DEVISIZE -gt 128 ]] || error "inode size $DEVISIZE"
rm $TMP/t57a.dump
done
}
run_test 57a "verify MDS filesystem created with large inodes =="
test_57b() {
[ $PARALLEL == "yes" ] && skip "skip parallel run"
if [ "$mds1_FSTYPE" != ldiskfs ]; then
skip_env "ldiskfs only test"
fi
remote_mds_nodsh && skip "remote MDS with nodsh"
local dir=$DIR/$tdir
local filecount=100
local file1=$dir/f1
local fileN=$dir/f$filecount
rm -rf $dir || error "removing $dir"
test_mkdir -c1 $dir
local mdtidx=$($LFS getstripe -m $dir)
local mdtname=MDT$(printf %04x $mdtidx)
local facet=mds$((mdtidx + 1))
echo "mcreating $filecount files"
createmany -m $dir/f 1 $filecount || error "creating files in $dir"
# verify that files do not have EAs yet
$LFS getstripe $file1 2>&1 | grep -q "no stripe" ||
error "$file1 has an EA"
$LFS getstripe $fileN 2>&1 | grep -q "no stripe" ||
error "$fileN has an EA"
sync
sleep 1
df $dir #make sure we get new statfs data
local mdsfree=$(do_facet $facet \
lctl get_param -n osd*.*$mdtname.kbytesfree)
local mdcfree=$(lctl get_param -n mdc.*$mdtname-mdc-*.kbytesfree)
local file
echo "opening files to create objects/EAs"
for file in $(seq -f $dir/f%g 1 $filecount); do
$OPENFILE -f O_RDWR $file > /dev/null 2>&1 ||
error "opening $file"
done
# verify that files have EAs now
$LFS getstripe -y $file1 | grep -q "l_ost_idx" ||
error "$file1 missing EA"
$LFS getstripe -y $fileN | grep -q "l_ost_idx" ||
error "$fileN missing EA"
sleep 1 #make sure we get new statfs data
df $dir
local mdsfree2=$(do_facet $facet \
lctl get_param -n osd*.*$mdtname.kbytesfree)
local mdcfree2=$(lctl get_param -n mdc.*$mdtname-mdc-*.kbytesfree)
if [[ $mdcfree2 -lt $((mdcfree - 16)) ]]; then
if [ "$mdsfree" != "$mdsfree2" ]; then
error "MDC before $mdcfree != after $mdcfree2"
else
echo "MDC before $mdcfree != after $mdcfree2"
echo "unable to confirm if MDS has large inodes"
fi
fi
rm -rf $dir
}
run_test 57b "default LOV EAs are stored inside large inodes ==="
test_58() {
[ $PARALLEL == "yes" ] && skip "skip parallel run"
[ -z "$(which wiretest 2>/dev/null)" ] &&
skip_env "could not find wiretest"
wiretest
}
run_test 58 "verify cross-platform wire constants =============="
test_59() {
[ $PARALLEL == "yes" ] && skip "skip parallel run"
echo "touch 130 files"
createmany -o $DIR/f59- 130
echo "rm 130 files"
unlinkmany $DIR/f59- 130
sync
# wait for commitment of removal
wait_delete_completed
}
run_test 59 "verify cancellation of llog records async ========="
TEST60_HEAD="test_60 run $RANDOM"
test_60a() {
[ $PARALLEL == "yes" ] && skip "skip parallel run"
remote_mgs_nodsh && skip "remote MGS with nodsh"
do_facet mgs "! which run-llog.sh &> /dev/null" &&
do_facet mgs "! ls run-llog.sh &> /dev/null" &&
skip_env "missing subtest run-llog.sh"
log "$TEST60_HEAD - from kernel mode"
do_facet mgs "$LCTL clear"
local saved_debug=$(do_facet mgs $LCTL get_param -n debug)
# ensure we get our records in the debug log
do_facet mgs $LCTL set_param debug=all-trace-malloc-net-other-ha-info-inode
do_facet mgs "bash run-llog.sh" || error "run-llog.sh failed"
do_facet mgs "$LCTL set_param debug='$saved_debug'"
do_facet mgs $LCTL dk > $TMP/$tfile
# LU-6388: test llog_reader
local llog_reader=$(do_facet mgs "which llog_reader 2> /dev/null")
llog_reader=${llog_reader:-$LUSTRE/utils/llog_reader}
[ -z $(do_facet mgs ls -d $llog_reader 2> /dev/null) ] &&
skip_env "missing llog_reader"
local fstype=$(facet_fstype mgs)
[ $fstype != ldiskfs -a $fstype != zfs ] &&
skip_env "Only for ldiskfs or zfs type mgs"
local mntpt=$(facet_mntpt mgs)
local mgsdev=$(mgsdevname 1)
local fid_list
local fid
local rec_list
local rec
local rec_type
local obj_file
local path
local seq
local oid
local pass=true
#get fid and record list
fid_list=($(awk '/9_sub.*record/ { print $NF }' $TMP/$tfile |
tail -n 4))
rec_list=($(awk '/9_sub.*record/ { print $((NF-3)) }' $TMP/$tfile |
tail -n 4))
#remount mgs as ldiskfs or zfs type
stop mgs || error "stop mgs failed"
mount_fstype mgs || error "remount mgs failed"
echo "fid_list=${fid_list[@]}"
for ((i = 0; i < ${#fid_list[@]}; i++)); do
fid=${fid_list[i]}
rec=${rec_list[i]}
obj_file=$mntpt/$(ost_fid2_objpath mgs $fid)
echo "obj_file is $obj_file"
do_facet mgs $llog_reader $obj_file
rec_type=$(do_facet mgs $llog_reader $obj_file | grep "type=" |
awk '{ print $3 }' | sed -e "s/^type=//g")
if [ $rec_type != $rec ]; then
echo "FAILED test_60a wrong record type $rec_type," \
"should be $rec"
pass=false
break
fi
#check obj path if record type is LLOG_LOGID_MAGIC
if [ "$rec" == "1064553b" ]; then
path=$(do_facet mgs $llog_reader $obj_file |
grep "path=" | awk '{ print $NF }' |
sed -e "s/^path=//g")
if [ $obj_file != $mntpt/$path ]; then
echo "FAILED test_60a wrong obj path" \
"$montpt/$path, should be $obj_file"
pass=false
break
fi
fi
done
rm -f $TMP/$tfile
#restart mgs before "error", otherwise it will block the next test
stop mgs || error "stop mgs failed"
start mgs $(mgsdevname) $MGS_MOUNT_OPTS || error "start mgs failed"
$pass || error "test failed, see FAILED test_60a messages for specifics"
}
run_test 60a "llog_test run from kernel module and test llog_reader"
test_60b() { # bug 6411
[ $PARALLEL == "yes" ] && skip "skip parallel run"
dmesg > $DIR/$tfile
LLOG_COUNT=$(do_facet mgs dmesg |
awk "/$TEST60_HEAD/ { marker = 1; from_marker = 0; }
/llog_[a-z]*.c:[0-9]/ {
if (marker)
from_marker++
from_begin++
}
END {
if (marker)
print from_marker
else
print from_begin
}")
(( LLOG_COUNT <= 140 )) ||
error "CDEBUG_LIMIT not limiting messages ($LLOG_COUNT)" || true
}
run_test 60b "limit repeated messages from CERROR/CWARN"
test_60c() {
[ $PARALLEL == "yes" ] && skip "skip parallel run"
echo "create 5000 files"
createmany -o $DIR/f60c- 5000
#define OBD_FAIL_MDS_LLOG_CREATE_FAILED 0x137
lctl set_param fail_loc=0x80000137
unlinkmany $DIR/f60c- 5000
lctl set_param fail_loc=0
}
run_test 60c "unlink file when mds full"
test_60d() {
[ $PARALLEL == "yes" ] && skip "skip parallel run"
SAVEPRINTK=$(lctl get_param -n printk)
# verify "lctl mark" is even working"
MESSAGE="test message ID $RANDOM $$"
$LCTL mark "$MESSAGE" || error "$LCTL mark failed"
dmesg | grep -q "$MESSAGE" || error "didn't find debug marker in log"
lctl set_param printk=0 || error "set lnet.printk failed"
lctl get_param -n printk | grep emerg || error "lnet.printk dropped emerg"
MESSAGE="new test message ID $RANDOM $$"
# Assume here that libcfs_debug_mark_buffer() uses D_WARNING
$LCTL mark "$MESSAGE" || error "$LCTL mark failed"
dmesg | grep -q "$MESSAGE" && error "D_WARNING wasn't masked" || true
lctl set_param -n printk="$SAVEPRINTK"
}
run_test 60d "test printk console message masking"
test_60e() {
[ $PARALLEL == "yes" ] && skip "skip parallel run"
remote_mds_nodsh && skip "remote MDS with nodsh"
touch $DIR/$tfile
#define OBD_FAIL_MDS_LLOG_CREATE_FAILED2 0x15b
do_facet mds1 lctl set_param fail_loc=0x15b
rm $DIR/$tfile
}
run_test 60e "no space while new llog is being created"
test_60f() {
local old_path=$($LCTL get_param -n debug_path)
stack_trap "$LCTL set_param debug_path=$old_path"
stack_trap "rm -f $TMP/$tfile*"
rm -f $TMP/$tfile* 2> /dev/null
#define OBD_FAIL_PTLRPC_DUMP_LOG 0x50e
$LCTL set_param debug_path=$TMP/$tfile fail_loc=0x8000050e
test_mkdir $DIR/$tdir
# retry in case the open is cached and not released
for (( i = 0; i < 100 && $(ls $TMP/$tfile* | wc -l) == 0; i++ )); do
echo $i > $DIR/$tdir/$tfile.$i && cat $DIR/$tdir/$tfile.$i
sleep 0.1
done
ls $TMP/$tfile*
(( $(ls $TMP/$tfile* | wc -l) > 0 )) || error "$TMP/$tfile not dumped"
}
run_test 60f "change debug_path works"
test_60g() {
(( $MDS1_VERSION >= $(version_code v2_16_56-96-gd0c75da78301) )) ||
skip "Need MDS >= v2_16_56-96-gd0c75da78301 for symlink fix"
local pid
local i
test_mkdir -c $MDSCOUNT $DIR/$tdir
(
local index=0
while true; do
$LFS setdirstripe -i $(($index % $MDSCOUNT)) \
-c $MDSCOUNT $DIR/$tdir/subdir$index \
2>/dev/null
mkdir $DIR/$tdir/subdir$index 2>/dev/null
rmdir $DIR/$tdir/subdir$index 2>/dev/null
index=$((index + 1))
done
) &
pid=$!
for i in {0..100}; do
# define OBD_FAIL_OSD_TXN_START 0x19a
local index=$((i % MDSCOUNT + 1))
do_facet mds$index $LCTL set_param fail_loc=0x8000019a \
> /dev/null
sleep 0.01
done
kill -9 $pid
for i in $(seq $MDSCOUNT); do
do_facet mds$i $LCTL set_param fail_loc=0 > /dev/null
done
mkdir $DIR/$tdir/new || error "mkdir failed"
rmdir $DIR/$tdir/new || error "rmdir failed"
do_facet mds1 $LCTL lfsck_start -M $(facet_svc mds1) -A -C \
-t namespace
for i in $(seq $MDSCOUNT); do
wait_update_facet mds$i "$LCTL get_param -n \
mdd.$(facet_svc mds$i).lfsck_namespace |
awk '/^status/ { print \\\$2 }'" "completed"
done
ls -R $DIR/$tdir
rm -rf $DIR/$tdir || error "rmdir failed"
}
run_test 60g "transaction abort won't cause MDT hung"
test_60h() {
[ $MDS1_VERSION -le $(version_code 2.12.52) ] &&
skip "Need MDS version at least 2.12.52"
[ $MDSCOUNT -lt 2 ] && skip "Need at least 2 MDTs"
local f
#define OBD_FAIL_MDS_STRIPE_CREATE 0x188
#define OBD_FAIL_MDS_STRIPE_FID 0x189
for fail_loc in 0x80000188 0x80000189; do
do_facet mds1 "$LCTL set_param fail_loc=$fail_loc"
$LFS mkdir -c $MDSCOUNT -i 0 $DIR/$tdir-$fail_loc ||
error "mkdir $dir-$fail_loc failed"
for i in {0..10}; do
# create may fail on missing stripe
echo $i > $DIR/$tdir-$fail_loc/$i
done
$LFS getdirstripe $DIR/$tdir-$fail_loc ||
error "getdirstripe $tdir-$fail_loc failed"
$LFS migrate -m 1 $DIR/$tdir-$fail_loc ||
error "migrate $tdir-$fail_loc failed"
$LFS getdirstripe $DIR/$tdir-$fail_loc ||
error "getdirstripe $tdir-$fail_loc failed"
pushd $DIR/$tdir-$fail_loc
for f in *; do
echo $f | cmp $f - || error "$f data mismatch"
done
popd
rm -rf $DIR/$tdir-$fail_loc
done
}
run_test 60h "striped directory with missing stripes can be accessed"
function t60i_load() {
mkdir $DIR/$tdir
#define OBD_FAIL_LLOG_PAUSE_AFTER_PAD 0x131c
$LCTL set_param fail_loc=0x131c fail_val=1
for ((i=0; i<5000; i++)); do
touch $DIR/$tdir/f$i
done
}
test_60i() {
changelog_register || error "changelog_register failed"
local cl_user="${CL_USERS[$SINGLEMDS]%% *}"
changelog_users $SINGLEMDS | grep -q $cl_user ||
error "User $cl_user not found in changelog_users"
changelog_chmask "ALL"
t60i_load &
local PID=$!
for((i=0; i<100; i++)); do
changelog_dump >/dev/null ||
error "can't read changelog"
done
kill $PID
wait $PID
changelog_deregister || error "changelog_deregister failed"
$LCTL set_param fail_loc=0
}
run_test 60i "llog: new record vs reader race"
test_60j() {
(( $MDS1_VERSION >= $(version_code 2.15.50) )) ||
skip "need MDS version at least 2.15.50"
[[ $PARALLEL != "yes" ]] || skip "skip parallel run"
remote_mds_nodsh && skip "remote MDS with nodsh"
[[ "$mds1_FSTYPE" == "ldiskfs" ]] || skip "ldiskfs only test"
changelog_users $SINGLEMDS | grep "^cl" &&
skip "active changelog user"
local llog_reader=$(do_facet $SINGLEMDS "which llog_reader 2> /dev/null")
[[ -z $(do_facet $SINGLEMDS ls -d $llog_reader 2> /dev/null) ]] &&
skip_env "missing llog_reader"
mkdir_on_mdt0 $DIR/$tdir
local f=$DIR/$tdir/$tfile
local mdt_dev
local tmpfile
local plain
changelog_register || error "cannot register changelog user"
# set changelog_mask to ALL
changelog_chmask "ALL"
changelog_clear
createmany -o ${f}- 100 || error "createmany failed as $RUNAS_ID"
unlinkmany ${f}- 100 || error "unlinkmany failed"
tmpfile="$(mktemp --tmpdir -u $tfile.XXXXXX)"
mdt_dev=$(facet_device $SINGLEMDS)
do_facet $SINGLEMDS sync
plain=$(do_facet $SINGLEMDS "$DEBUGFS -c -R 'dump changelog_catalog \
$tmpfile' $mdt_dev; $llog_reader $tmpfile" |
awk '{match($0,"path=([^ ]+)",a)}END{print a[1]}')
stack_trap "do_facet $SINGLEMDS rm -f $tmpfile"
# if $tmpfile is not on EXT3 filesystem for some reason
[[ ${plain:0:1} == 'O' ]] ||
skip "path $plain is not in 'O/1/d<n>/<n>' format"
size=$(do_facet $SINGLEMDS "$DEBUGFS -c -R 'dump $plain $tmpfile' \
$mdt_dev; stat -c %s $tmpfile")
echo "Truncate llog from $size to $((size - size % 8192))"
size=$((size - size % 8192))
do_facet $SINGLEMDS $TRUNCATE $tmpfile $size
errs=$(do_facet $SINGLEMDS "$llog_reader $tmpfile" |
grep -c 'in bitmap only')
(( $errs > 0 )) || error "llog_reader didn't find lost records"
size=$((size - 9000))
echo "Corrupt llog in the middle at $size"
do_facet $SINGLEMDS dd if=/dev/urandom of=$tmpfile bs=1 seek=$size \
count=333 conv=notrunc
errs=$(do_facet $SINGLEMDS "$llog_reader $tmpfile" |
grep -c 'next chunk')
(( $errs > 0 )) || error "llog_reader didn't skip bad chunk"
}
run_test 60j "llog_reader reports corruptions"
test_61a() {
[ $PARALLEL == "yes" ] && skip "skip parallel run"
f="$DIR/f61"
dd if=/dev/zero of=$f bs=$PAGE_SIZE count=1 || error "dd $f failed"
cancel_lru_locks osc
$MULTIOP $f OSMWUc || error "$MULTIOP $f failed"
sync
}
run_test 61a "mmap() writes don't make sync hang ================"
test_61b() {
mmap_mknod_test $DIR/$tfile || error "mmap_mknod_test failed"
}
run_test 61b "mmap() of unstriped file is successful"
# bug 2319 - oig_wait() interrupted causes crash because of invalid waitq.
# Though this test is irrelevant anymore, it helped to reveal some
# other grant bugs (LU-4482), let's keep it.
test_63a() { # was test_63
[ $PARALLEL == "yes" ] && skip "skip parallel run"
MAX_DIRTY_MB=$(lctl get_param -n osc.*.max_dirty_mb | head -n 1)
for i in `seq 10` ; do
dd if=/dev/zero of=$DIR/f63 bs=8k &
sleep 5
kill $!
sleep 1
done
rm -f $DIR/f63 || true
}
run_test 63a "Verify oig_wait interruption does not crash ======="
# bug 2248 - async write errors didn't return to application on sync
# bug 3677 - async write errors left page locked
test_63b() {
[ $PARALLEL == "yes" ] && skip "skip parallel run"
debugsave
lctl set_param debug=-1
# ensure we have a grant to do async writes
dd if=/dev/zero of=$DIR/$tfile bs=4k count=1
rm $DIR/$tfile
sync # sync lest earlier test intercept the fail_loc
#define OBD_FAIL_OSC_BRW_PREP_REQ 0x406
lctl set_param fail_loc=0x80000406
$MULTIOP $DIR/$tfile Owy && \
error "sync didn't return ENOMEM"
sync; sleep 2; sync # do a real sync this time to flush page
lctl get_param -n llite.*.dump_page_cache | grep locked && \
error "locked page left in cache after async error" || true
debugrestore
}
run_test 63b "async write errors should be returned to fsync ==="
test_63c() {
stack_trap "$LCTL set_param $($LCTL get_param llite.*.sync_on_close)"
$LCTL set_param llite.$FSNAME*.sync_on_close=1
local count=20
local dir=$DIR/$tdir
mkdir $dir || error "mkdir $dir failed"
# ensure only a single OST stripe per file so accounting works
$LFS setstripe -c 1 $dir || error "setstripe $dir failed"
# open+create files without writing, expect one MDT sync for each
# MDT (N*open/create)
local mdt_before=$(calc_stats mdc.*.stats mds_sync)
local ost_before=$(calc_stats osc.*.stats ost_sync)
createmany -o $dir/$tfile-empty- $count ||
error "'createmany -o $tfile-empty $count' failed"
local mdt_after=$(calc_stats mdc.*.stats mds_sync)
local ost_after=$(calc_stats osc.*.stats ost_sync)
(( $mdt_after == $mdt_before + $count )) ||
error "expect MDT create $mdt_after == $mdt_before + $count"
(( $ost_after == $ost_before )) ||
error "expect OST create $ost_after == $ost_before"
# append data to existing files, expect one OST sync for each
# MDT(open) OST(N*setattr)
mdt_before=$mdt_after
ost_before=$ost_after
for ((i = 0; i < count; i++)); do
echo "foo-$i" >> $dir/$tfile-empty-$i ||
error "write $tfile-empty-$i failed"
done
mdt_after=$(calc_stats mdc.*.stats mds_sync)
ost_after=$(calc_stats osc.*.stats ost_sync)
(( $mdt_after == $mdt_before )) ||
error "expect MDT write $mdt_after == $mdt_before"
(( $ost_after == $ost_before + $count )) ||
error "expect OST write $ost_after == $ost_before + $count"
# overwrite same file notrunc+fsync, expect only one OST sync for each
# MDT (open/create), OST (N*write+sync)
mdt_before=$mdt_after
ost_before=$ost_after
for ((i = 0; i < count; i++)); do
dd if=/dev/zero of=$dir/$tfile-1 bs=4k count=2 \
conv=notrunc,fsync status=none ||
error "dd $tfile-1 $i failed"
done
mdt_after=$(calc_stats mdc.*.stats mds_sync)
ost_after=$(calc_stats osc.*.stats ost_sync)
(( $mdt_after == $mdt_before + 1 )) ||
error "expect MDT write+sync $mdt_after == $mdt_before + 1"
(( $ost_after == $ost_before + $count )) ||
error "expect OST write+sync $ost_after == $ost_before + $count"
# extend existing file, expect one OST sync for each
# OST (write)
mdt_before=$mdt_after
ost_before=$ost_after
for ((i = 0; i < count; i++)); do
dd if=/dev/zero of=$dir/$tfile-1 bs=4k count=2 seek=$((i+1)) \
conv=notrunc status=none ||
error "dd append $tfile-1 $i failed"
done
mdt_after=$(calc_stats mdc.*.stats mds_sync)
ost_after=$(calc_stats osc.*.stats ost_sync)
(( $mdt_after == $mdt_before )) ||
error "expect MDT append $mdt_after == $mdt_before"
(( $ost_after == $ost_before + $count )) ||
error "expect OST append $ost_after == $ost_before + $count"
# list directory, expect no MDT or OST sync
# MDT (readdir,getattr), OST (getattr)
mdt_before=$mdt_after
ost_before=$ost_after
ls -l $dir || error "ls -l $dir failed"
mdt_after=$(calc_stats mdc.*.stats mds_sync)
ost_after=$(calc_stats osc.*.stats ost_sync)
(( $mdt_after == $mdt_before )) ||
error "expect MDT ls $mdt_after == $mdt_before"
(( $ost_after == $ost_before )) ||
error "expect OST ls $ost_after == $ost_before"
# truncate existing files, expect one MDT+OST sync for each
# MDT (open+truncate) OST (truncate)
mdt_before=$mdt_after
ost_before=$ost_after
for ((i = 0; i < count; i++)); do
> $dir/$tfile-empty-$i || error "truncate $tfile-empty-$i fail"
done
mdt_after=$(calc_stats mdc.*.stats mds_sync)
ost_after=$(calc_stats osc.*.stats ost_sync)
(( $mdt_after == $mdt_before + $count)) ||
error "expect MDT truncate $mdt_after == $mdt_before + $count"
(( $ost_after == $ost_before + $count )) ||
error "expect OST truncate $ost_after == $ost_before + $count"
stack_trap "set_default_debug"
set_default_debug "rpctrace+dlmtrace+inode+entry" "" 512
# write to multi-striped file, expect one MDT+OSTCOUNT OST sync
# MDT (create/setstripe+truncate), OST (8*OSTCOUNT*write)
mdt_before=$mdt_after
ost_before=$ost_after
$LFS setstripe -c $OSTCOUNT -S 1M $dir/$tfile-striped
local stripes=$($LFS getstripe -c $dir/$tfile-striped)
stack_trap "rm -f $dir/$tfile-stripes"
dd if=/dev/zero of=$dir/$tfile-striped bs=1M count=$((8 * $stripes)) ||
error "dd striped $tfile-striped failed"
mdt_after=$(calc_stats mdc.*.stats mds_sync)
ost_after=$(calc_stats osc.*.stats ost_sync)
(( $mdt_after == $mdt_before + 2 )) ||
error "expect MDT striped $mdt_after == $mdt_before + 2"
(( $ost_after >= $ost_before + $stripes &&
$ost_after < $ost_before + 2 * $stripes )) ||
error "expect OST striped $ost_after == $ost_before + $stripes"
# read file, expect no MDT or OST sync
# MDT(open), OST(2*OSTCOUNT*read)
mdt_before=$mdt_after
ost_before=$ost_after
dd if=$dir/$tfile-1 of=/dev/null bs=4k status=none ||
error "dd read $tfile-1 failed"
dd if=$dir/$tfile-striped of=/dev/null bs=1M status=none ||
error "dd read $tfile-1 failed"
mdt_after=$(calc_stats mdc.*.stats mds_sync)
ost_after=$(calc_stats osc.*.stats ost_sync)
(( $mdt_after == $mdt_before )) ||
error "expect MDT read $mdt_after == $mdt_before"
(( $ost_after == $ost_before )) ||
error "expect OST read $ost_after == $ost_before"
}
run_test 63c "test sync_on_close=1"
test_64a () {
[ $PARALLEL == "yes" ] && skip "skip parallel run"
lfs df $DIR
lctl get_param osc.*[oO][sS][cC][_-]*.cur* | grep "=[1-9]"
}
run_test 64a "verify filter grant calculations (in kernel) ====="
test_64b () {
[ $PARALLEL == "yes" ] && skip "skip parallel run"
bash oos.sh $MOUNT || error "oos.sh failed: $?"
}
run_test 64b "check out-of-space detection on client"
test_64c() {
$LCTL set_param osc.*OST0000-osc-[^mM]*.cur_grant_bytes=0
}
run_test 64c "verify grant shrink"
# this does exactly what osc_request.c:osc_announce_cached() does in
# order to calculate max amount of grants to ask from server
want_grant() {
local tgt=$1
local nrpages=$($LCTL get_param -n osc.$tgt.max_pages_per_rpc)
local rpc_in_flight=$($LCTL get_param -n osc.$tgt.max_rpcs_in_flight)
((rpc_in_flight++));
nrpages=$((nrpages * rpc_in_flight))
local dirty_max_pages=$($LCTL get_param -n osc.$tgt.max_dirty_mb)
dirty_max_pages=$((dirty_max_pages * 1024 * 1024 / PAGE_SIZE))
[[ $dirty_max_pages -gt $nrpages ]] && nrpages=$dirty_max_pages
local undirty=$((nrpages * PAGE_SIZE))
local max_extent_pages
max_extent_pages=$(import_param $tgt grant_max_extent_size)
max_extent_pages=$((max_extent_pages / PAGE_SIZE))
local nrextents=$(((nrpages + max_extent_pages - 1) / max_extent_pages))
local grant_extent_tax
grant_extent_tax=$(import_param $tgt grant_extent_tax)
undirty=$((undirty + nrextents * grant_extent_tax))
echo $undirty
}
# this is size of unit for grant allocation. It should be equal to
# what tgt_grant.c:tgt_grant_chunk() calculates
grant_chunk() {
local tgt=$1
local max_brw_size
local grant_extent_tax
max_brw_size=$(import_param $tgt max_brw_size)
grant_extent_tax=$(import_param $tgt grant_extent_tax)
echo $(((max_brw_size + grant_extent_tax) * 2))
}
test_64d() {
[ $OST1_VERSION -ge $(version_code 2.10.56) ] ||
skip "OST < 2.10.55 doesn't limit grants enough"
local tgt=$($LCTL dl | awk '/OST0000-osc-[^mM]/ { print $4 }')
[[ "$($LCTL get_param osc.${tgt}.import)" =~ "grant_param" ]] ||
skip "no grant_param connect flag"
local olddebug="$($LCTL get_param -n debug 2> /dev/null)"
$LCTL set_param -n -n debug="$OLDDEBUG" || true
stack_trap "$LCTL set_param -n debug='$olddebug'" EXIT
local max_cur_granted=$(($(want_grant $tgt) + $(grant_chunk $tgt)))
stack_trap "rm -f $DIR/$tfile && wait_delete_completed" EXIT
$LFS setstripe $DIR/$tfile -i 0 -c 1
dd if=/dev/zero of=$DIR/$tfile bs=1M count=1000 &
ddpid=$!
while kill -0 $ddpid; do
local cur_grant=$($LCTL get_param -n osc.$tgt.cur_grant_bytes)
if [[ $cur_grant -gt $max_cur_granted ]]; then
kill $ddpid
error "cur_grant $cur_grant > $max_cur_granted"
fi
sleep 1
done
}
run_test 64d "check grant limit exceed"
check_grants() {
local tgt=$1
local expected=$2
local msg=$3
local cur_grants=$($LCTL get_param -n osc.$tgt.cur_grant_bytes)
((cur_grants == expected)) ||
error "$msg: grants mismatch: $cur_grants, expected $expected"
}
round_up_p2() {
echo $((($1 + $2 - 1) & ~($2 - 1)))
}
test_64e() {
[ $PARALLEL == "yes" ] && skip "skip parallel run"
[ $OST1_VERSION -ge $(version_code 2.11.56) ] ||
skip "Need OSS version at least 2.11.56"
local olddebug="$($LCTL get_param -n debug 2> /dev/null)"
stack_trap "$LCTL set_param -n debug='$olddebug'" EXIT
$LCTL set_param debug=+cache
# Remount client to reset grant
remount_client $MOUNT || error "failed to remount client"
local osc_tgt="$FSNAME-OST0000-osc-$($LFS getname -i $DIR)"
local init_grants=$(import_param $osc_tgt initial_grant)
check_grants $osc_tgt $init_grants "init grants"
local extent_tax=$(import_param $osc_tgt grant_extent_tax)
local max_brw_size=$(import_param $osc_tgt max_brw_size)
local gbs=$(import_param $osc_tgt grant_block_size)
# write random number of bytes from max_brw_size / 4 to max_brw_size
local write_bytes=$(shuf -i $((max_brw_size / 4))-$max_brw_size -n 1)
# align for direct io
write_bytes=$(round_up_p2 $write_bytes PAGE_SIZE)
# round to grant consumption unit
local wb_round_up=$(round_up_p2 $write_bytes gbs)
local grants=$((wb_round_up + extent_tax))
$LFS setstripe -c 1 -i 0 $DIR/$tfile || error "lfs setstripe failed"
stack_trap "rm -f $DIR/$tfile"
# define OBD_FAIL_TGT_NO_GRANT 0x725
# make the server not grant more back
do_facet ost1 $LCTL set_param fail_loc=0x725
dd if=/dev/zero of=$DIR/$tfile bs=$write_bytes count=1 oflag=direct
do_facet ost1 $LCTL set_param fail_loc=0
check_grants $osc_tgt $((init_grants - grants)) "dio w/o grant alloc"
rm -f $DIR/$tfile || error "rm failed"
# Remount client to reset grant
remount_client $MOUNT || error "failed to remount client"
osc_tgt="$FSNAME-OST0000-osc-$($LFS getname -i $DIR)"
$LFS setstripe -c 1 -i 0 $DIR/$tfile || error "lfs setstripe failed"
# define OBD_FAIL_TGT_NO_GRANT 0x725
# make the server not grant more back
do_facet ost1 $LCTL set_param fail_loc=0x725
$MULTIOP $DIR/$tfile "oO_WRONLY:w${write_bytes}yc"
do_facet ost1 $LCTL set_param fail_loc=0
check_grants $osc_tgt $((init_grants - grants)) "buf io w/o grant alloc"
}
run_test 64e "check grant consumption (no grant allocation)"
test_64f() {
[ $PARALLEL == "yes" ] && skip "skip parallel run"
local olddebug="$($LCTL get_param -n debug 2> /dev/null)"
stack_trap "$LCTL set_param -n debug='$olddebug'" EXIT
$LCTL set_param debug=+cache
# Remount client to reset grant
remount_client $MOUNT || error "failed to remount client"
local osc_tgt="$FSNAME-OST0000-osc-$($LFS getname -i $DIR)"
local init_grants=$(import_param $osc_tgt initial_grant)
local extent_tax=$(import_param $osc_tgt grant_extent_tax)
local max_brw_size=$(import_param $osc_tgt max_brw_size)
local gbs=$(import_param $osc_tgt grant_block_size)
local chunk=$(grant_chunk $osc_tgt)
# write random number of bytes from max_brw_size / 4 to max_brw_size
local write_bytes=$(shuf -i $((max_brw_size / 4))-$max_brw_size -n 1)
# align for direct io
write_bytes=$(round_up_p2 $write_bytes PAGE_SIZE)
# round to grant consumption unit
local wb_round_up=$(round_up_p2 $write_bytes gbs)
local grants=$((wb_round_up + extent_tax))
$LFS setstripe -c 1 -i 0 $DIR/$tfile || error "lfs setstripe failed"
dd if=/dev/zero of=$DIR/$tfile bs=$write_bytes count=1 oflag=direct ||
error "error writing to $DIR/$tfile"
stack_trap "rm -f $DIR/$tfile"
check_grants $osc_tgt $((init_grants - grants + chunk)) \
"direct io with grant allocation"
rm -f $DIR/$tfile || error "rm failed"
# Remount client to reset grant
remount_client $MOUNT || error "failed to remount client"
osc_tgt="$FSNAME-OST0000-osc-$($LFS getname -i $DIR)"
$LFS setstripe -c 1 -i 0 $DIR/$tfile || error "lfs setstripe failed"
# Hybrid means this won't really be buffered IO, so we disable it for
# this part of the test
local hybrid=$($LCTL get_param -n llite.*.hybrid_io)
$LCTL set_param llite.*.hybrid_io=0
stack_trap "$LCTL set_param -n llite.*.hybrid_io=$hybrid" EXIT
# Testing that buffered IO consumes grant on the client
# Delay the RPC on the server so it's guaranteed to not complete even
# if the RPC is sent from the client
#define OBD_FAIL_PTLRPC_PAUSE_REQ 0x50a
$LCTL set_param fail_loc=0x50a fail_val=3
dd if=/dev/zero of=$DIR/$tfile bs=$write_bytes count=1 conv=notrunc ||
error "error writing to $DIR/$tfile with buffered IO"
check_grants $osc_tgt $((init_grants - grants)) \
"buffered io, not write rpc"
# Clear the fail loc and do a sync on the client
$LCTL set_param fail_loc=0 fail_val=0
sync
# RPC is now known to have sent
check_grants $osc_tgt $((init_grants - grants + chunk)) \
"buffered io, one RPC"
}
run_test 64f "check grant consumption (with grant allocation)"
test_64g() {
(( $MDS1_VERSION >= $(version_code 2.14.56) )) ||
skip "Need MDS version at least 2.14.56"
local mdts=$(mdts_nodes)
local old=$($LCTL get_param mdc.$FSNAME-*.grant_shrink_interval |
tr '\n' ' ')
stack_trap "$LCTL set_param $old"
# generate dirty pages and increase dirty granted on MDT
stack_trap "rm -f $DIR/$tfile-*"
for (( i = 0; i < 10; i++)); do
$LFS setstripe -E 1M -L mdt $DIR/$tfile-$i ||
error "can't set stripe"
dd if=/dev/zero of=$DIR/$tfile-$i bs=128k count=1 ||
error "can't dd"
$LFS getstripe $DIR/$tfile-$i | grep -q pattern.*mdt || {
$LFS getstripe $DIR/$tfile-$i
error "not DoM file"
}
done
# flush dirty pages
sync
# wait until grant shrink reset grant dirty on MDTs
for ((i = 0; i < 120; i++)); do
grant_dirty=$(do_nodes $mdts $LCTL get_param -n mdt.*.tot_dirty |
awk '{sum=sum+$1} END {print sum}')
vm_dirty=$(awk '/Dirty:/{print $2}' /proc/meminfo)
echo "$grant_dirty grants, $vm_dirty pages"
(( grant_dirty + vm_dirty == 0 )) && break
(( i == 3 )) && sync &&
$LCTL set_param mdc.$FSNAME-*.grant_shrink_interval=5
sleep 1
done
grant_dirty=$(do_nodes $mdts $LCTL get_param -n mdt.*.tot_dirty |
awk '{sum=sum+$1} END {print sum}')
(( grant_dirty == 0 )) || error "$grant_dirty on MDT"
}
run_test 64g "grant shrink on MDT"
test_64h() {
(( $OST1_VERSION >= $(version_code 2.14.56) )) ||
skip "need OST at least 2.14.56 to avoid grant shrink on read"
local instance=$($LFS getname -i $DIR)
local osc_tgt="$FSNAME-OST0000-osc-$instance"
local num_exps=$(do_facet ost1 \
$LCTL get_param -n obdfilter.*OST0000*.num_exports)
local max_brw_size=$(import_param $osc_tgt max_brw_size)
local avail=$($LCTL get_param -n osc.*OST0000-osc-$instance.kbytesavail)
local p="$TMP/$TESTSUITE-$TESTNAME.parameters"
# 10MiB is for file to be written, max_brw_size * 16 *
# num_exps is space reserve so that tgt_grant_shrink() decided
# to not shrink
local expect=$((max_brw_size * 16 * num_exps + 10 * 1048576))
(( avail * 1024 < expect )) &&
skip "need $expect bytes on ost1, have $(( avail * 1024 )) only"
save_lustre_params client "osc.*OST0000*.grant_shrink" > $p
save_lustre_params client "osc.*OST0000*.grant_shrink_interval" >> $p
stack_trap "restore_lustre_params < $p; rm -f $p" EXIT
$LCTL set_param osc.*OST0000*.grant_shrink=1
$LCTL set_param osc.*OST0000*.grant_shrink_interval=10
$LFS setstripe -c 1 -i 0 $DIR/$tfile
stack_trap "rm -f $DIR/$tfile"
dd if=/dev/zero of=$DIR/$tfile bs=1M count=10 oflag=sync
# drop cache so that coming read would do rpc
cancel_lru_locks osc
# shrink interval is set to 10, pause for 7 seconds so that
# grant thread did not wake up yet but coming read entered
# shrink mode for rpc (osc_should_shrink_grant())
sleep 7
declare -a cur_grant_bytes
declare -a tot_granted
cur_grant_bytes[0]=$($LCTL get_param -n osc.*OST0000*.cur_grant_bytes)
tot_granted[0]=$(do_facet ost1 \
$LCTL get_param -n obdfilter.*OST0000*.tot_granted)
dd if=$DIR/$tfile bs=4K count=1 of=/dev/null
cur_grant_bytes[1]=$($LCTL get_param -n osc.*OST0000*.cur_grant_bytes)
tot_granted[1]=$(do_facet ost1 \
$LCTL get_param -n obdfilter.*OST0000*.tot_granted)
# grant change should be equal on both sides
(( cur_grant_bytes[0] - cur_grant_bytes[1] ==
tot_granted[0] - tot_granted[1])) ||
error "grant change mismatch, " \
"server: ${tot_granted[0]} to ${tot_granted[1]}, " \
"client: ${cur_grant_bytes[0]} to ${cur_grant_bytes[1]}"
}
run_test 64h "grant shrink on read"
test_64i() {
(( $OST1_VERSION >= $(version_code 2.14.56) )) ||
skip "need OST at least 2.14.56 to avoid grant shrink on replay"
[ $PARALLEL == "yes" ] && skip "skip parallel run"
remote_ost_nodsh && skip "remote OSTs with nodsh"
$LFS setstripe -c 1 -i 0 $DIR/$tfile
stack_trap "rm -f $DIR/$tfile"
dd if=/dev/zero of=$DIR/$tfile bs=1M count=64
# lustre-ffff9fc75e850800 /mnt/lustre -> ffff9fc75e850800
local instance=$($LFS getname -i $DIR)
local osc_tgt="$FSNAME-OST0000-osc-$instance"
local cgb=$($LCTL get_param -n osc.$osc_tgt.cur_grant_bytes)
# shrink grants and simulate rpc loss
#define OBD_FAIL_PTLRPC_DROP_REQ_OPC 0x513
do_facet ost1 "$LCTL set_param fail_loc=0x80000513 fail_val=17"
$LCTL set_param osc.$osc_tgt.cur_grant_bytes=$((cgb/2))B
fail ost1
dd if=/dev/zero of=$DIR/$tfile oflag=append bs=1M count=8 conv=notrunc
local testid=$(echo $TESTNAME | tr '_' ' ')
do_facet ost1 dmesg | tac | sed "/$testid/,$ d" |
grep "GRANT, real grant" &&
error "client has more grants then it owns" || true
}
run_test 64i "shrink on reconnect"
test_64j() {
$LFS setstripe -c 1 -i 0 $DIR/$tfile
# get rid of lost grants which could be formed on previous test
$MULTIOP $DIR/$tfile oO_RDWR:O_SYNC:w4096c
#define OBD_FAIL_OST_GRANT_PREPARE 0x256
do_facet ost1 "$LCTL set_param fail_loc=0x80000256"
$MULTIOP $DIR/$tfile oO_RDWR:O_DIRECT:w4096c
}
run_test 64j "check grants on re-done rpc"
# bug 1414 - set/get directories' stripe info
test_65a() {
[ $PARALLEL == "yes" ] && skip "skip parallel run"
# LU-16904 check if the root is set as PFL layout
local numcomp=$($LFS getstripe --component-count $MOUNT)
[ $numcomp -eq 0 ] || skip "Skip test_65a for PFL layout"
test_mkdir $DIR/$tdir
touch $DIR/$tdir/f1
$LVERIFY $DIR/$tdir $DIR/$tdir/f1 || error "lverify failed"
}
run_test 65a "directory with no stripe info"
test_65b() {
[ $PARALLEL == "yes" ] && skip "skip parallel run"
test_mkdir $DIR/$tdir
local STRIPESIZE=$($LFS getstripe -S $DIR/$tdir)
$LFS setstripe -S $((STRIPESIZE * 2)) -i 0 -c 1 $DIR/$tdir ||
error "setstripe"
touch $DIR/$tdir/f2
$LVERIFY $DIR/$tdir $DIR/$tdir/f2 || error "lverify failed"
}
run_test 65b "directory setstripe -S stripe_size*2 -i 0 -c 1"
test_65c() {
[ $PARALLEL == "yes" ] && skip "skip parallel run"
[ $OSTCOUNT -lt 2 ] && skip_env "need at least 2 OSTs"
test_mkdir $DIR/$tdir
local stripesize=$($LFS getstripe -S $DIR/$tdir)
$LFS setstripe -S $((stripesize * 4)) -i 1 \
-c $((OSTCOUNT - 1)) $DIR/$tdir || error "setstripe"
touch $DIR/$tdir/f3
$LVERIFY $DIR/$tdir $DIR/$tdir/f3 || error "lverify failed"
}
run_test 65c "directory setstripe -S stripe_size*4 -i 1 -c $((OSTCOUNT-1))"
test_65d() {
[ $PARALLEL == "yes" ] && skip "skip parallel run"
test_mkdir $DIR/$tdir
local STRIPECOUNT=$($LFS getstripe -c $DIR/$tdir)
local STRIPESIZE=$($LFS getstripe -S $DIR/$tdir)
if [[ $STRIPECOUNT -le 0 ]]; then
sc=1
elif [[ $STRIPECOUNT -gt $LOV_MAX_STRIPE_COUNT ]]; then
[[ $OSTCOUNT -gt $LOV_MAX_STRIPE_COUNT ]] &&
sc=$LOV_MAX_STRIPE_COUNT || sc=$(($OSTCOUNT - 1))
else
sc=$(($STRIPECOUNT - 1))
fi
$LFS setstripe -S $STRIPESIZE -c $sc $DIR/$tdir || error "setstripe"
touch $DIR/$tdir/f4 $DIR/$tdir/f5
$LVERIFY $DIR/$tdir $DIR/$tdir/f4 $DIR/$tdir/f5 ||
error "lverify failed"
}
run_test 65d "directory setstripe -S stripe_size -c stripe_count"
test_65e() {
[ $PARALLEL == "yes" ] && skip "skip parallel run"
# LU-16904 delete layout when root is set as PFL layout
save_layout_restore_at_exit $MOUNT
$LFS setstripe -d $MOUNT || error "setstripe failed"
test_mkdir $DIR/$tdir
$LFS setstripe $DIR/$tdir || error "setstripe"
$LFS getstripe -v $DIR/$tdir | grep "Default" ||
error "no stripe info failed"
touch $DIR/$tdir/f6
$LVERIFY $DIR/$tdir $DIR/$tdir/f6 || error "lverify failed"
}
run_test 65e "directory setstripe defaults"
test_65f() {
[ $PARALLEL == "yes" ] && skip "skip parallel run"
test_mkdir $DIR/${tdir}f
$RUNAS $LFS setstripe $DIR/${tdir}f &&
error "setstripe succeeded" || true
}
run_test 65f "dir setstripe permission (should return error) ==="
test_65g() {
[ $PARALLEL == "yes" ] && skip "skip parallel run"
# LU-16904 delete layout when root is set as PFL layout
save_layout_restore_at_exit $MOUNT
$LFS setstripe -d $MOUNT || error "setstripe failed"
test_mkdir $DIR/$tdir
local STRIPESIZE=$($LFS getstripe -S $DIR/$tdir)
$LFS setstripe -S $((STRIPESIZE * 2)) -i 0 -c 1 $DIR/$tdir ||
error "setstripe -S failed"
$LFS setstripe -d $DIR/$tdir || error "setstripe -d failed"
$LFS getstripe -v $DIR/$tdir | grep "Default" ||
error "delete default stripe failed"
}
run_test 65g "directory setstripe -d"
test_65h() {
[ $PARALLEL == "yes" ] && skip "skip parallel run"
test_mkdir $DIR/$tdir
local STRIPESIZE=$($LFS getstripe -S $DIR/$tdir)
$LFS setstripe -S $((STRIPESIZE * 2)) -i 0 -c 1 $DIR/$tdir ||
error "setstripe -S failed"
test_mkdir $DIR/$tdir/dd1
[ $($LFS getstripe -c $DIR/$tdir) = $($LFS getstripe -c $DIR/$tdir/dd1) ] ||
error "stripe info inherit failed"
}
run_test 65h "directory stripe info inherit ===================="
test_65i() {
[ $PARALLEL == "yes" ] && skip "skip parallel run"
save_layout_restore_at_exit $MOUNT
# bug6367: set non-default striping on root directory
$LFS setstripe -S 65536 -c -1 $MOUNT || error "error setting stripe"
# bug12836: getstripe on -1 default directory striping
$LFS getstripe $MOUNT || error "getstripe $MOUNT failed"
# bug12836: getstripe -v on -1 default directory striping
$LFS getstripe -v $MOUNT || error "getstripe -v $MOUNT failed"
# bug12836: new find on -1 default directory striping
$LFS find -mtime -1 $MOUNT > /dev/null || error "find $MOUNT failed"
}
run_test 65i "various tests to set root directory striping"
test_65j() { # bug6367
[ $PARALLEL == "yes" ] && skip "skip parallel run"
sync; sleep 1
# if we aren't already remounting for each test, do so for this test
if [ "$I_MOUNTED" = "yes" ]; then
cleanup || error "failed to unmount"
setup
fi
save_layout_restore_at_exit $MOUNT
$LFS setstripe -d $MOUNT || error "setstripe failed"
}
run_test 65j "set default striping on root directory (bug 6367)="
cleanup_65k() {
rm -rf $DIR/$tdir
wait_delete_completed
do_facet $SINGLEMDS "lctl set_param -n \
osp.$ost*MDT0000.max_create_count=$max_count"
do_facet $SINGLEMDS "lctl set_param -n \
osp.$ost*MDT0000.create_count=$count"
do_facet $SINGLEMDS lctl --device %$INACTIVE_OSC activate
echo $INACTIVE_OSC "is Activate"
wait_osc_import_state mds ost$(( ostnum + 1 )) FULL
}
test_65k() { # bug11679
[ $PARALLEL == "yes" ] && skip "skip parallel run"
[[ $OSTCOUNT -lt 2 ]] && skip_env "needs >= 2 OSTs"
remote_mds_nodsh && skip "remote MDS with nodsh"
local disable_precreate=true
[ $MDS1_VERSION -le $(version_code 2.8.54) ] &&
disable_precreate=false
echo "Check OST status: "
local MDS_OSCS=$(do_facet $SINGLEMDS lctl dl |
awk '/[oO][sS][cC].*md[ts]/ { print $4 }')
for OSC in $MDS_OSCS; do
echo $OSC "is active"
do_facet $SINGLEMDS lctl --device %$OSC activate
done
for INACTIVE_OSC in $MDS_OSCS; do
local ost=$(osc_to_ost $INACTIVE_OSC)
local ostnum=$(do_facet $SINGLEMDS lctl get_param -n \
lov.*md*.target_obd |
awk -F: /$ost/'{ print $1 }' | head -n 1)
mkdir -p $DIR/$tdir
$LFS setstripe -i $ostnum -c 1 $DIR/$tdir
createmany -o $DIR/$tdir/$tfile.$ostnum. 1000
echo "Deactivate: " $INACTIVE_OSC
do_facet $SINGLEMDS lctl --device %$INACTIVE_OSC deactivate
local count=$(do_facet $SINGLEMDS "lctl get_param -n \
osp.$ost*MDT0000.create_count")
local max_count=$(do_facet $SINGLEMDS "lctl get_param -n \
osp.$ost*MDT0000.max_create_count")
$disable_precreate &&
do_facet $SINGLEMDS "lctl set_param -n \
osp.$ost*MDT0000.max_create_count=0"
for idx in $(seq 0 $((OSTCOUNT - 1))); do
[ -f $DIR/$tdir/$idx ] && continue
echo "$LFS setstripe -i $idx -c 1 $DIR/$tdir/$idx"
$LFS setstripe -i $idx -c 1 $DIR/$tdir/$idx ||
{ cleanup_65k;
error "setstripe $idx should succeed"; }
rm -f $DIR/$tdir/$idx || error "rm $idx failed"
done
unlinkmany $DIR/$tdir/$tfile.$ostnum. 1000
rmdir $DIR/$tdir
do_facet $SINGLEMDS "lctl set_param -n \
osp.$ost*MDT0000.max_create_count=$max_count"
do_facet $SINGLEMDS "lctl set_param -n \
osp.$ost*MDT0000.create_count=$count"
do_facet $SINGLEMDS lctl --device %$INACTIVE_OSC activate
echo $INACTIVE_OSC "is Activate"
wait_osc_import_state mds ost$(( ostnum + 1 )) FULL
done
}
run_test 65k "validate manual striping works properly with deactivated OSCs"
test_65l() { # bug 12836
[ $PARALLEL == "yes" ] && skip "skip parallel run"
test_mkdir -p $DIR/$tdir/test_dir
$LFS setstripe -c -1 $DIR/$tdir/test_dir
$LFS find -mtime -1 $DIR/$tdir >/dev/null
}
run_test 65l "lfs find on -1 stripe dir ========================"
test_65m() {
local layout=$(save_layout $MOUNT)
$RUNAS $LFS setstripe -c 2 $MOUNT && {
restore_layout $MOUNT $layout
error "setstripe should fail by non-root users"
}
true
}
run_test 65m "normal user can't set filesystem default stripe"
test_65n() {
[ -n "$FILESET" ] && skip "Not functional for FILESET set"
[[ $MDS1_VERSION -ge $(version_code 2.12.50) ]] ||
skip "Need MDS version at least 2.12.50"
[[ $PARALLEL != "yes" ]] || skip "skip parallel run"
[[ $OSTCOUNT -ge 2 ]] || skip_env "needs >= 2 OSTs"
which getfattr > /dev/null 2>&1 || skip_env "no getfattr command"
which setfattr > /dev/null 2>&1 || skip_env "no setfattr command"
save_layout_restore_at_exit $MOUNT
# new subdirectory under root directory should not inherit
# the default layout from root
# LU-16904 check if the root is set as PFL layout
local numcomp=$($LFS getstripe --component-count $MOUNT)
if [[ $numcomp -eq 0 ]]; then
local dir1=$MOUNT/$tdir-1
mkdir $dir1 || error "mkdir $dir1 failed"
! getfattr -n trusted.lov $dir1 &> /dev/null ||
error "$dir1 shouldn't have LOV EA"
fi
# delete the default layout on root directory
$LFS setstripe -d $MOUNT || error "delete root default layout failed"
local dir2=$MOUNT/$tdir-2
mkdir $dir2 || error "mkdir $dir2 failed"
! getfattr -n trusted.lov $dir2 &> /dev/null ||
error "$dir2 shouldn't have LOV EA"
# set a new striping pattern on root directory
local def_stripe_size=$($LFS getstripe -S $MOUNT)
local new_def_stripe_size=$((def_stripe_size * 2))
$LFS setstripe -S $new_def_stripe_size $MOUNT ||
error "set stripe size on $MOUNT failed"
# new file created in $dir2 should inherit the new stripe size from
# the filesystem default
local file2=$dir2/$tfile-2
touch $file2 || error "touch $file2 failed"
local file2_stripe_size=$($LFS getstripe -S $file2)
[[ $file2_stripe_size -eq $new_def_stripe_size ]] ||
{
echo "file2_stripe_size: '$file2_stripe_size'"
echo "new_def_stripe_size: '$new_def_stripe_size'"
error "$file2 didn't inherit stripe size $new_def_stripe_size"
}
local dir3=$MOUNT/$tdir-3
mkdir $dir3 || error "mkdir $dir3 failed"
# $dir3 shouldn't have LOV EA, but "lfs getstripe -d $dir3" should show
# the root layout, which is the actual default layout that will be used
# when new files are created in $dir3.
local dir3_layout=$(get_layout_param $dir3)
local root_dir_layout=$(get_layout_param $MOUNT)
[[ "$dir3_layout" = "$root_dir_layout" ]] ||
{
echo "dir3_layout: '$dir3_layout'"
echo "root_dir_layout: '$root_dir_layout'"
error "$dir3 should show the default layout from $MOUNT"
}
# set OST pool on root directory
local pool=$TESTNAME
pool_add $pool || error "add $pool failed"
pool_add_targets $pool 0 $((OSTCOUNT - 1)) 1 ||
error "add targets to $pool failed"
$LFS setstripe -p $pool $MOUNT ||
error "set OST pool on $MOUNT failed"
# new file created in $dir3 should inherit the pool from
# the filesystem default
local file3=$dir3/$tfile-3
touch $file3 || error "touch $file3 failed"
local file3_pool=$($LFS getstripe -p $file3)
[[ "$file3_pool" = "$pool" ]] ||
error "$file3 ('$file3_pool') didn't inherit OST pool '$pool'"
local dir4=$MOUNT/$tdir-4
mkdir $dir4 || error "mkdir $dir4 failed"
local dir4_layout=$(get_layout_param $dir4)
root_dir_layout=$(get_layout_param $MOUNT)
echo "$LFS getstripe -d $dir4"
$LFS getstripe -d $dir4
echo "$LFS getstripe -d $MOUNT"
$LFS getstripe -d $MOUNT
[[ "$dir4_layout" = "$root_dir_layout" ]] ||
{
echo "dir4_layout: '$dir4_layout'"
echo "root_dir_layout: '$root_dir_layout'"
error "$dir4 should show the default layout from $MOUNT"
}
# new file created in $dir4 should inherit the pool from
# the filesystem default
local file4=$dir4/$tfile-4
touch $file4 || error "touch $file4 failed"
local file4_pool=$($LFS getstripe -p $file4)
[[ "$file4_pool" = "$pool" ]] ||
error "$file4 ('$file4_pool') didn't inherit OST pool $pool"
# new subdirectory under non-root directory should inherit
# the default layout from its parent directory
$LFS setstripe -S $new_def_stripe_size -p $pool $dir4 ||
error "set directory layout on $dir4 failed"
local dir5=$dir4/$tdir-5
mkdir $dir5 || error "mkdir $dir5 failed"
dir4_layout=$(get_layout_param $dir4)
local dir5_layout=$(get_layout_param $dir5)
[[ "$dir4_layout" = "$dir5_layout" ]] ||
{
echo "dir4_layout: '$dir4_layout'"
echo "dir5_layout: '$dir5_layout'"
error "$dir5 should inherit the default layout from $dir4"
}
# though subdir under ROOT doesn't inherit default layout, but
# its sub dir/file should be created with default layout.
[[ $MDSCOUNT -ge 2 ]] || skip_env "needs >= 2 MDTs"
[[ $MDS1_VERSION -ge $(version_code 2.12.59) ]] ||
skip "Need MDS version at least 2.12.59"
local default_lmv_count=$($LFS getdirstripe -D -c $MOUNT)
local default_lmv_index=$($LFS getdirstripe -D -i $MOUNT)
local default_lmv_hash=$($LFS getdirstripe -D -H $MOUNT)
if [ $default_lmv_hash == "none" ]; then
stack_trap "$LFS setdirstripe -D -d $MOUNT" EXIT
else
stack_trap "$LFS setdirstripe -D -i $default_lmv_index \
-c $default_lmv_count -H $default_lmv_hash $MOUNT" EXIT
fi
$LFS setdirstripe -D -c 2 $MOUNT ||
error "setdirstripe -D -c 2 failed"
mkdir $MOUNT/$tdir-6 || error "mkdir $tdir-6 failed"
local lmv_count=$($LFS getdirstripe -c $MOUNT/$tdir-6)
[ $lmv_count -eq 2 ] || error "$tdir-6 stripe count $lmv_count"
(( $MDS1_VERSION >= $(version_code v2_14_57-22-g6e59408f1a) )) ||
skip "Need MDS >= 2.14.57.22 for proper pool inheritance"
# $dir4 layout includes pool
$LFS setstripe -S $((new_def_stripe_size * 2)) $dir4
[[ "$pool" = $($LFS getstripe -p -d $dir4) ]] ||
error "pool lost on setstripe"
$LFS setstripe -E -1 -S $new_def_stripe_size $dir4
[[ "$pool" = $($LFS getstripe -p -d $dir4) ]] ||
error "pool lost on compound layout setstripe"
}
run_test 65n "don't inherit default layout from root for new subdirectories"
test_65o() {
(( $MDS1_VERSION >= $(version_code 2.14.57) )) ||
skip "need MDS version at least 2.14.57"
# set OST pool on root directory
local pool=$TESTNAME
pool_add $pool || error "add $pool failed"
pool_add_targets $pool 0 $((OSTCOUNT - 1)) 1 ||
error "add targets to $pool failed"
local dir1=$MOUNT/$tdir
mkdir $dir1 || error "mkdir $dir1 failed"
# set a new striping pattern on root directory
local def_stripe_size=$($LFS getstripe -S $MOUNT)
$LFS setstripe -p $pool $dir1 ||
error "set directory layout on $dir1 failed"
# $dir1 layout includes pool
$LFS setstripe -S $((def_stripe_size * 2)) $dir1
[[ "$pool" = $($LFS getstripe -p -d $dir1) ]] ||
error "pool lost on setstripe"
$LFS setstripe -E 1M -L mdt -E -1 -c 1 $dir1
$LFS getstripe $dir1
[[ "$pool" = $($LFS getstripe -p -d $dir1) ]] ||
error "pool lost on compound layout setstripe"
$LFS setdirstripe -i 0 -c 2 $dir1/dir2 ||
error "setdirstripe failed on sub-dir with inherited pool"
$LFS getstripe $dir1/dir2
[[ "$pool" = $($LFS getstripe -p -d $dir1/dir2) ]] ||
error "pool lost on compound layout setdirstripe"
$LFS setstripe -E -1 -c 1 $dir1
$LFS getstripe -d $dir1
[[ "$pool" = $($LFS getstripe -p -d $dir1) ]] ||
error "pool lost on setstripe"
}
run_test 65o "pool inheritance for mdt component"
test_65p () { # LU-16152
local src_dir=$DIR/$tdir/src_dir
local dst_dir=$DIR/$tdir/dst_dir
local yaml_file=$DIR/$tdir/layout.yaml
local border
(( $CLIENT_VERSION >= $(version_code 2.15.51) )) ||
skip "Need at least version 2.15.51"
test_mkdir -p $src_dir
$LFS setstripe -E 2048M -c 4 -E EOF -c 8 $src_dir ||
error "failed to setstripe"
$LFS getstripe --yaml -d $src_dir > $yaml_file ||
error "failed to getstripe"
test_mkdir -p $dst_dir
$LFS setstripe --yaml $yaml_file $dst_dir ||
error "failed to setstripe with yaml file"
border=$($LFS getstripe -d $dst_dir |
awk '/lcme_extent.e_end:/ { print $2; exit; }') ||
error "failed to getstripe"
# 2048M is 0x80000000, or 2147483648
(( $border == 2147483648 )) ||
error "failed to handle huge number in yaml layout"
}
run_test 65p "setstripe with yaml file and huge number"
test_65q () { # LU-16194
local src_dir=$DIR/$tdir/src_dir
(( $CLIENT_VERSION >= $(version_code 2.15.51) )) &&
(( $MDS1_VERSION >= $(version_code 2.15.51) )) ||
skip "Need at least version 2.15.51"
test_mkdir -p $src_dir
# 8E is 0x8000 0000 0000 0000, which is negative as s64
$LFS setstripe -E 8E -c 4 -E EOF -c 8 $src_dir &&
error "should fail if extent start/end >=8E"
# EOF should work as before
$LFS setstripe -E 8M -c 4 -E EOF -c 8 $src_dir ||
error "failed to setstripe normally"
}
run_test 65q "setstripe with >=8E offset should fail"
65r_check_offsets() {
local offsets
local all_zeroes=1
local pfl=$1
offsets=$($LFS getstripe -y $pfl | awk '/stripe_offset:/ { print $2 }')
for off in ${offsets[@]}; do
((off == 0)) || all_zeroes=0
done
echo $all_zeroes
}
test_65r() { #LU-13062
(( OSTCOUNT >= 2 )) || skip_env "needs >= 2 OSTs"
local dir=$DIR/$tdir
local pfl=$DIR/$tdir/$tfile
test_mkdir -p $DIR/$tdir
# lfs setstripe can set all 0 offsets directly
$LFS setstripe -E1m -c1 -i0 -E2m -c2 -i0 -E4m -c4 -i0 -E -1 -i0 $pfl ||
error "Create file $pfl failed"
(( $(65r_check_offsets $pfl) == 1 )) ||
error "lfs setstripe directions are ignored"
# lfs setstripe --copy should reset specified offsets to default
lfs setstripe --copy=$pfl ${pfl}.copy ||
error "lfs setstripe --copy failed"
(( $(65r_check_offsets ${pfl}.copy) == 0 )) ||
error "lfs setstripe --copy keeps specific offsets"
# directory layout copy should drop specific offsets
$LFS setstripe -E1m -c1 -i0 -E2m -c2 -i0 -E4m -c4 -i0 -E -1 -i0 $dir ||
error "Failed to set default layout for $dir"
lfs setstripe --copy=$dir ${pfl}.dircopy ||
error "lfs setstripe --copy from directory failed"
(( $(65r_check_offsets ${pfl}.dircopy) == 0 )) ||
error "lfs setstripe --copy for directory keeps offsets"
}
run_test 65r "prevent all-zero offsets"
# bug 2543 - update blocks count on client
test_66() {
[ $PARALLEL == "yes" ] && skip "skip parallel run"
local COUNT=${COUNT:-8}
dd if=/dev/urandom of=$DIR/f66 bs=1k count=$COUNT
sync; sync_all_data; sync; sync_all_data
cancel_lru_locks osc
local BLOCKS=$(ls -s --block-size=1k $DIR/f66 | awk '{ print $1 }')
(( BLOCKS >= COUNT )) || error "$DIR/f66 blocks $BLOCKS < $COUNT"
}
run_test 66 "update inode blocks count on client ==============="
meminfo() {
awk '($1 == "'$1':") { print $2 }' /proc/meminfo
}
swap_used() {
swapon -s | awk '($1 == "'$1'") { print $4 }'
}
# bug5265, obdfilter oa2dentry return -ENOENT
# #define OBD_FAIL_SRV_ENOENT 0x217
test_69() {
[ $PARALLEL == "yes" ] && skip "skip parallel run"
remote_ost_nodsh && skip "remote OST with nodsh"
f="$DIR/$tfile"
$LFS setstripe -c 1 -i 0 $f
stack_trap "rm -f $f ${f}.2"
$DIRECTIO write ${f}.2 0 1 || error "directio write error"
do_facet ost1 lctl set_param fail_loc=0x217
$TRUNCATE $f 1 # vmtruncate() will ignore truncate() error.
$DIRECTIO write $f 0 2 && error "write succeeded, expect -ENOENT"
do_facet ost1 lctl set_param fail_loc=0
$DIRECTIO write $f 0 2 || error "write error"
cancel_lru_locks osc
$DIRECTIO read $f 0 1 || error "read error"
do_facet ost1 lctl set_param fail_loc=0x217
$DIRECTIO read $f 1 1 && error "read succeeded, expect -ENOENT"
do_facet ost1 lctl set_param fail_loc=0
}
run_test 69 "verify oa2dentry return -ENOENT doesn't LBUG ======"
test_70a() {
# Perform a really simple test of health write and health check
(( $OST1_VERSION >= $(version_code 2.15.59) )) ||
skip "OSTs < 2.15.59 doesn't have enable_health_write"
local orig_value="$(do_facet ost1 $LCTL get_param -n enable_health_write)"
stack_trap "do_facet ost1 $LCTL set_param enable_health_write $orig_value"
# Test with health write off
do_facet ost1 $LCTL set_param enable_health_write off ||
error "can't set enable_health_write off"
do_facet ost1 $LCTL get_param enable_health_write ||
error "can't get enable_health_write"
[[ "$(do_facet ost1 $LCTL get_param health_check)" =~ "healthy" ]] ||
error "not healthy (1)"
# Test with health write on
do_facet ost1 $LCTL set_param enable_health_write on ||
error "can't set enable_health_write on"
do_facet ost1 $LCTL get_param enable_health_write ||
error "can't get enable_health_write"
[[ "$(do_facet ost1 $LCTL get_param health_check)" =~ "healthy" ]] ||
error "not healthy (2)"
}
run_test 70a "verify health_check, health_write don't explode (on OST)"
test_71() {
test_mkdir $DIR/$tdir
$LFS setdirstripe -D -c$MDSCOUNT $DIR/$tdir
bash rundbench -C -D $DIR/$tdir 2 || error "dbench failed!"
}
run_test 71 "Running dbench on lustre (don't segment fault) ===="
test_72a() { # bug 5695 - Test that on 2.6 remove_suid works properly
[ $PARALLEL == "yes" ] && skip "skip parallel run"
[ "$RUNAS_ID" = "$UID" ] &&
skip_env "RUNAS_ID = UID = $UID -- skipping"
# Check that testing environment is properly set up. Skip if not
FAIL_ON_ERROR=false check_runas_id_ret $RUNAS_ID $RUNAS_GID $RUNAS ||
skip_env "User $RUNAS_ID does not exist - skipping"
touch $DIR/$tfile
chmod 777 $DIR/$tfile
chmod ug+s $DIR/$tfile
$RUNAS dd if=/dev/zero of=$DIR/$tfile bs=512 count=1 ||
error "$RUNAS dd $DIR/$tfile failed"
# See if we are still setuid/sgid
[ -u $DIR/$tfile ] || [ -g $DIR/$tfile ] &&
error "S/gid is not dropped on write"
# Now test that MDS is updated too
cancel_lru_locks mdc
[ -u $DIR/$tfile ] || [ -g $DIR/$tfile ] &&
error "S/gid is not dropped on MDS"
rm -f $DIR/$tfile
}
run_test 72a "Test that remove suid works properly (bug5695) ===="
test_72b() { # bug 24226 -- keep mode setting when size is not changing
local perm
[ "$RUNAS_ID" = "$UID" ] &&
skip_env "RUNAS_ID = UID = $UID -- skipping"
[ "$RUNAS_ID" -eq 0 ] &&
skip_env "RUNAS_ID = 0 -- skipping"
[ $PARALLEL == "yes" ] && skip "skip parallel run"
# Check that testing environment is properly set up. Skip if not
FAIL_ON_ERROR=false check_runas_id_ret $RUNAS_ID $RUNAS_ID $RUNAS ||
skip_env "User $RUNAS_ID does not exist - skipping"
touch $DIR/${tfile}-f{g,u}
test_mkdir $DIR/${tfile}-dg
test_mkdir $DIR/${tfile}-du
chmod 770 $DIR/${tfile}-{f,d}{g,u}
chmod g+s $DIR/${tfile}-{f,d}g
chmod u+s $DIR/${tfile}-{f,d}u
for perm in 777 2777 4777; do
$RUNAS chmod $perm $DIR/${tfile}-fg && error "S/gid file allowed improper chmod to $perm"
$RUNAS chmod $perm $DIR/${tfile}-fu && error "S/uid file allowed improper chmod to $perm"
$RUNAS chmod $perm $DIR/${tfile}-dg && error "S/gid dir allowed improper chmod to $perm"
$RUNAS chmod $perm $DIR/${tfile}-du && error "S/uid dir allowed improper chmod to $perm"
done
true
}
run_test 72b "Test that we keep mode setting if without file data changed (bug 24226)"
# bug 3462 - multiple simultaneous MDC requests
test_73() {
[ $PARALLEL == "yes" ] && skip "skip parallel run"
test_mkdir $DIR/d73-1
test_mkdir $DIR/d73-2
multiop_bg_pause $DIR/d73-1/f73-1 O_c || return 1
pid1=$!
lctl set_param fail_loc=0x80000129
$MULTIOP $DIR/d73-1/f73-2 Oc &
sleep 1
lctl set_param fail_loc=0
$MULTIOP $DIR/d73-2/f73-3 Oc &
pid3=$!
kill -USR1 $pid1
wait $pid1 || return 1
sleep 25
$CHECKSTAT -t file $DIR/d73-1/f73-1 || return 4
$CHECKSTAT -t file $DIR/d73-1/f73-2 || return 5
$CHECKSTAT -t file $DIR/d73-2/f73-3 || return 6
rm -rf $DIR/d73-*
}
run_test 73 "multiple MDC requests (should not deadlock)"
test_74a() { # bug 6149, 6184
[ $PARALLEL == "yes" ] && skip "skip parallel run"
touch $DIR/f74a
#define OBD_FAIL_LDLM_ENQUEUE_OLD_EXPORT 0x30e
#
# very important to OR with CFS_FAIL_ONCE (0x80000000) -- otherwise it
# will spin in a tight reconnection loop
$LCTL set_param fail_loc=0x8000030e
# get any lock that won't be difficult - lookup works.
ls $DIR/f74a
$LCTL set_param fail_loc=0
rm -f $DIR/f74a
true
}
run_test 74a "ldlm_enqueue freed-export error path, ls (shouldn't LBUG)"
test_74b() { # bug 13310
[ $PARALLEL == "yes" ] && skip "skip parallel run"
#define OBD_FAIL_LDLM_ENQUEUE_OLD_EXPORT 0x30e
#
# very important to OR with CFS_FAIL_ONCE (0x80000000) -- otherwise it
# will spin in a tight reconnection loop
$LCTL set_param fail_loc=0x8000030e
# get a "difficult" lock
touch $DIR/f74b
$LCTL set_param fail_loc=0
rm -f $DIR/f74b
true
}
run_test 74b "ldlm_enqueue freed-export error path, touch (shouldn't LBUG)"
test_74c() {
[ $PARALLEL == "yes" ] && skip "skip parallel run"
#define OBD_FAIL_LDLM_NEW_LOCK
$LCTL set_param fail_loc=0x319
touch $DIR/$tfile && error "touch successful"
$LCTL set_param fail_loc=0
true
}
run_test 74c "ldlm_lock_create error path, (shouldn't LBUG)"
slab_lic=/sys/kernel/slab/lustre_inode_cache
num_objects() {
[ -f $slab_lic/shrink ] && echo 1 > $slab_lic/shrink
[ -f $slab_lic/objects ] && awk '{ print $1 }' $slab_lic/objects ||
awk '/lustre_inode_cache/ { print $2; exit }' /proc/slabinfo
}
test_76a() { # Now for b=20433, added originally in b=1443
[ $PARALLEL == "yes" ] && skip "skip parallel run"
cancel_lru_locks osc
# there may be some slab objects cached per core
local cpus=$(getconf _NPROCESSORS_ONLN 2>/dev/null)
local before=$(num_objects)
local count=$((512 * cpus))
[ "$SLOW" = "no" ] && count=$((128 * cpus))
local margin=$((count / 10))
if [[ -f $slab_lic/aliases ]]; then
local aliases=$(cat $slab_lic/aliases)
(( aliases > 0 )) && margin=$((margin * aliases))
fi
echo "before slab objects: $before"
for i in $(seq $count); do
touch $DIR/$tfile
rm -f $DIR/$tfile
done
cancel_lru_locks osc
local after=$(num_objects)
echo "created: $count, after slab objects: $after"
# shared slab counts are not very accurate, allow significant margin
# the main goal is that the cache growth is not permanently > $count
while (( after > before + margin )); do
sleep 1
after=$(num_objects)
wait=$((wait + 1))
(( wait % 5 == 0 )) && echo "wait $wait seconds objects: $after"
if (( wait > 60 )); then
error "inode slab grew from $before+$margin to $after"
fi
done
}
run_test 76a "confirm clients recycle inodes properly ===="
test_76b() {
[ $PARALLEL == "yes" ] && skip "skip parallel run"
[ $CLIENT_VERSION -ge $(version_code 2.13.55) ] || skip "not supported"
local count=512
local before=$(num_objects)
for i in $(seq $count); do
mkdir $DIR/$tdir
rmdir $DIR/$tdir
done
local after=$(num_objects)
local wait=0
while (( after > before )); do
sleep 1
after=$(num_objects)
wait=$((wait + 1))
(( wait % 5 == 0 )) && echo "wait $wait seconds objects: $after"
if (( wait > 60 )); then
error "inode slab grew from $before to $after"
fi
done
echo "slab objects before: $before, after: $after"
}
run_test 76b "confirm clients recycle directory inodes properly ===="
export ORIG_CSUM=""
set_checksums()
{
# Note: in sptlrpc modes which enable its own bulk checksum, the
# original crc32_le bulk checksum will be automatically disabled,
# and the OBD_FAIL_OSC_CHECKSUM_SEND/OBD_FAIL_OSC_CHECKSUM_RECEIVE
# will be checked by sptlrpc code against sptlrpc bulk checksum.
# In this case set_checksums() will not be no-op, because sptlrpc
# bulk checksum will be enabled all through the test.
[ "$ORIG_CSUM" ] || ORIG_CSUM=`lctl get_param -n osc.*.checksums | head -n1`
lctl set_param -n osc.*.checksums $1
return 0
}
export ORIG_CSUM_TYPE="`lctl get_param -n osc.*osc-[^mM]*.checksum_type |
sed 's/.*\[\(.*\)\].*/\1/g' | head -n1`"
CKSUM_TYPES=${CKSUM_TYPES:-$(lctl get_param -n osc.*osc-[^mM]*.checksum_type |
tr -d [] | head -n1)}
set_checksum_type()
{
lctl set_param -n osc.*osc-[^mM]*.checksum_type $1
rc=$?
log "set checksum type to $1, rc = $rc"
return $rc
}
get_osc_checksum_type()
{
# arugment 1: OST name, like OST0000
ost=$1
checksum_type=$(lctl get_param -n osc.*${ost}-osc-[^mM]*.checksum_type |
sed 's/.*\[\(.*\)\].*/\1/g')
rc=$?
[ $rc -ne 0 ] && error "failed to get checksum type of $ost, rc = $rc, output = $checksum_type"
echo $checksum_type
}
F77_TMP=$TMP/f77-temp
F77SZ=8
setup_f77() {
dd if=/dev/urandom of=$F77_TMP bs=1M count=$F77SZ || \
error "error writing to $F77_TMP"
}
test_77a() { # bug 10889
[ $PARALLEL == "yes" ] && skip "skip parallel run"
$GSS && skip_env "could not run with gss"
[ ! -f $F77_TMP ] && setup_f77
set_checksums 1
dd if=$F77_TMP of=$DIR/$tfile bs=1M count=$F77SZ || error "dd error"
set_checksums 0
rm -f $DIR/$tfile
}
run_test 77a "normal checksum read/write operation"
test_77b() { # bug 10889
[ $PARALLEL == "yes" ] && skip "skip parallel run"
$GSS && skip_env "could not run with gss"
[ ! -f $F77_TMP ] && setup_f77
#define OBD_FAIL_OSC_CHECKSUM_SEND 0x409
$LCTL set_param fail_loc=0x80000409
set_checksums 1
dd if=$F77_TMP of=$DIR/$tfile bs=1M count=$F77SZ conv=sync ||
error "dd error: $?"
$LCTL set_param fail_loc=0
for algo in $CKSUM_TYPES; do
cancel_lru_locks osc
set_checksum_type $algo
#define OBD_FAIL_OSC_CHECKSUM_RECEIVE 0x408
$LCTL set_param fail_loc=0x80000408
cmp $F77_TMP $DIR/$tfile || error "file compare failed"
$LCTL set_param fail_loc=0
done
set_checksums 0
set_checksum_type $ORIG_CSUM_TYPE
rm -f $DIR/$tfile
}
run_test 77b "checksum error on client write, read"
cleanup_77c() {
trap 0
set_checksums 0
$LCTL set_param osc.*osc-[^mM]*.checksum_dump=0
$check_ost &&
do_facet ost1 $LCTL set_param obdfilter.*-OST*.checksum_dump=0
[ -n "$osc_file_prefix" ] && rm -f ${osc_file_prefix}*
$check_ost && [ -n "$ost_file_prefix" ] &&
do_facet ost1 rm -f ${ost_file_prefix}\*
}
test_77c() {
[ $PARALLEL == "yes" ] && skip "skip parallel run"
$GSS && skip_env "could not run with gss"
remote_ost_nodsh && skip "remote OST with nodsh"
local bad1
local osc_file_prefix
local osc_file
local check_ost=false
local ost_file_prefix="$TMP/lustre-log"
local ost_file
local orig_cksum
local dump_cksum
local fid
local old_ost_path=$(do_facet ost1 $LCTL get_param -n debug_path)
stack_trap "do_facet ost1 $LCTL set_param debug_path=$old_ost_path"
do_facet ost1 $LCTL set_param debug_path=${ost_file_prefix}
# ensure corruption will occur on first OSS/OST
$LFS setstripe -i 0 $DIR/$tfile
[ ! -f $F77_TMP ] && setup_f77
dd if=$F77_TMP of=$DIR/$tfile bs=1M count=$F77SZ conv=sync ||
error "dd write error: $?"
fid=$($LFS path2fid $DIR/$tfile)
if [ $OST1_VERSION -ge $(version_code 2.9.57) ]
then
check_ost=true
ost_file_prefix=${ost_file_prefix}-checksum_dump-ost-\\${fid}
else
echo "OSS do not support bulk pages dump upon error"
fi
osc_file_prefix=$($LCTL get_param -n debug_path)
osc_file_prefix=${osc_file_prefix}-checksum_dump-osc-\\${fid}
trap cleanup_77c EXIT
set_checksums 1
# enable bulk pages dump upon error on Client
$LCTL set_param osc.*osc-[^mM]*.checksum_dump=1
# enable bulk pages dump upon error on OSS
$check_ost &&
do_facet ost1 $LCTL set_param obdfilter.*-OST*.checksum_dump=1
# flush Client cache to allow next read to reach OSS
cancel_lru_locks osc
#define OBD_FAIL_OSC_CHECKSUM_RECEIVE 0x408
$LCTL set_param fail_loc=0x80000408
dd if=$DIR/$tfile of=/dev/null bs=1M || error "dd read error: $?"
$LCTL set_param fail_loc=0
rm -f $DIR/$tfile
# check cksum dump on Client
osc_file=$(ls ${osc_file_prefix}*)
[ -n "$osc_file" ] || error "no checksum dump file on Client"
# OBD_FAIL_OSC_CHECKSUM_RECEIVE corrupts with "bad1" at start of file
bad1=$(dd if=$osc_file bs=1 count=4 2>/dev/null) || error "dd error: $?"
[ $bad1 == "bad1" ] || error "unexpected corrupt pattern"
orig_cksum=$(dd if=$F77_TMP bs=1 skip=4 count=1048572 2>/dev/null |
cksum)
dump_cksum=$(dd if=$osc_file bs=1 skip=4 2>/dev/null | cksum)
[[ "$orig_cksum" == "$dump_cksum" ]] ||
error "dump content does not match on Client"
$check_ost || skip "No need to check cksum dump on OSS"
# check cksum dump on OSS
ost_file=$(do_facet ost1 ls ${ost_file_prefix}\*)
[ -n "$ost_file" ] || error "no checksum dump file on OSS"
orig_cksum=$(dd if=$F77_TMP bs=1048576 count=1 2>/dev/null | cksum)
dump_cksum=$(do_facet ost1 dd if=$ost_file 2>/dev/null \| cksum)
[[ "$orig_cksum" == "$dump_cksum" ]] ||
error "dump content does not match on OSS"
cleanup_77c
}
run_test 77c "checksum error on client read with debug"
test_77d() { # bug 10889
[ $PARALLEL == "yes" ] && skip "skip parallel run"
$GSS && skip_env "could not run with gss"
stack_trap "rm -f $DIR/$tfile"
#define OBD_FAIL_OSC_CHECKSUM_SEND 0x409
$LCTL set_param fail_loc=0x80000409
set_checksums 1
$DIRECTIO write $DIR/$tfile 0 $F77SZ $((1024 * 1024)) ||
error "direct write: rc=$?"
$LCTL set_param fail_loc=0
set_checksums 0
#define OBD_FAIL_OSC_CHECKSUM_RECEIVE 0x408
$LCTL set_param fail_loc=0x80000408
set_checksums 1
cancel_lru_locks osc
$DIRECTIO read $DIR/$tfile 0 $F77SZ $((1024 * 1024)) ||
error "direct read: rc=$?"
$LCTL set_param fail_loc=0
set_checksums 0
}
run_test 77d "checksum error on OST direct write, read"
test_77f() { # bug 10889
[[ $PARALLEL == "yes" ]] && skip "skip parallel run"
$GSS && skip_env "could not run with gss"
stack_trap "$LCTL set_param $($LCTL get_param -b osc.*.resend_count)"
$LCTL set_param osc.*.resend_count=1
set_checksums 1
stack_trap "set_checksum_type $ORIG_CSUM_TYPE"
stack_trap "rm -f $DIR/$tfile.*"
for algo in $CKSUM_TYPES; do
cancel_lru_locks osc
set_checksum_type $algo
#define OBD_FAIL_OSC_CHECKSUM_SEND 0x409
$LCTL set_param fail_loc=0x409
$DIRECTIO write $DIR/$tfile.$algo 0 $F77SZ $((1024 * 1024)) &&
error "direct write succeeded"
$LCTL set_param fail_loc=0
done
}
run_test 77f "repeat checksum error on write (expect error)"
test_77g() { # bug 10889
[ $PARALLEL == "yes" ] && skip "skip parallel run"
$GSS && skip_env "could not run with gss"
remote_ost_nodsh && skip "remote OST with nodsh"
[ ! -f $F77_TMP ] && setup_f77
local file=$DIR/$tfile
stack_trap "rm -f $file" EXIT
$LFS setstripe -c 1 -i 0 $file
#define OBD_FAIL_OST_CHECKSUM_RECEIVE 0x21a
do_facet ost1 lctl set_param fail_loc=0x8000021a
set_checksums 1
dd if=$F77_TMP of=$file bs=1M count=$F77SZ ||
error "write error: rc=$?"
do_facet ost1 lctl set_param fail_loc=0
set_checksums 0
cancel_lru_locks osc
#define OBD_FAIL_OST_CHECKSUM_SEND 0x21b
do_facet ost1 lctl set_param fail_loc=0x8000021b
set_checksums 1
cmp $F77_TMP $file || error "file compare failed"
do_facet ost1 lctl set_param fail_loc=0
set_checksums 0
}
run_test 77g "checksum error on OST write, read"
test_77k() { # LU-10906
[ $PARALLEL == "yes" ] && skip "skip parallel run"
$GSS && skip_env "could not run with gss"
local cksum_param="osc.$FSNAME*.checksums"
local get_checksum="$LCTL get_param -n $cksum_param | head -n1"
local checksum
local i
[ "$ORIG_CSUM" ] || ORIG_CSUM=$(eval $get_checksum)
stack_trap "wait_update $HOSTNAME '$get_checksum' $ORIG_CSUM || true"
stack_trap "do_facet mgs $LCTL set_param -P $cksum_param=$ORIG_CSUM"
set_checksums 1 # so wait_update below will really wait for MGS
for i in 0 1; do
do_facet mgs $LCTL set_param -P $cksum_param=$i ||
error "failed to set checksum=$i on MGS"
wait_update $HOSTNAME "$get_checksum" $i
#remount
echo "remount client, checksum should be $i"
remount_client $MOUNT || error "failed to remount client"
sleep 1 # to don't race with setting from mount for sure
checksum=$(eval $get_checksum)
[ $checksum -eq $i ] || error "checksum($checksum) != $i"
done
# remove persistent param to avoid races with checksum mountopt below
do_facet mgs $LCTL set_param -P -d $cksum_param ||
error "failed to delete checksum on MGS"
for opt in "checksum" "nochecksum"; do
#remount with mount option
echo "remount client with option $opt, checksum should be $i"
umount_client $MOUNT || error "failed to umount client"
mount_client $MOUNT "$MOUNT_OPTS,$opt" ||
error "failed to mount client with option '$opt'"
checksum=$(eval $get_checksum)
[ $checksum -eq $i ] || error "checksum($checksum) != $i"
i=$((i - 1))
done
remount_client $MOUNT || error "failed to remount client"
}
run_test 77k "enable/disable checksum correctly"
test_77l() {
[ $PARALLEL == "yes" ] && skip "skip parallel run"
$GSS && skip_env "could not run with gss"
set_checksums 1
stack_trap "set_checksums $ORIG_CSUM" EXIT
stack_trap "set_checksum_type $ORIG_CSUM_TYPE" EXIT
local old
old=$($LCTL get_param -n osc.*.idle_timeout | head -n 1)
$LCTL set_param osc.*.idle_timeout=10
stack_trap "$LCTL set_param osc.*.idle_timeout=$old" EXIT
set_checksum_type invalid && error "unexpected success of invalid checksum type"
$LFS setstripe -c 1 -i 0 $DIR/$tfile
for algo in $CKSUM_TYPES; do
set_checksum_type $algo || error "fail to set checksum type $algo"
osc_algo=$(get_osc_checksum_type OST0000)
[ "$osc_algo" != "$algo" ] && error "checksum type is $osc_algo after setting it to $algo"
# no locks, no reqs to let the connection idle
cancel_lru_locks osc
lru_resize_disable osc
wait_osc_import_state client ost1 IDLE
# ensure ost1 is connected
stat $DIR/$tfile >/dev/null || error "can't stat"
wait_osc_import_state client ost1 FULL
osc_algo=$(get_osc_checksum_type OST0000)
[ "$osc_algo" != "$algo" ] && error "checksum type changed from $algo to $osc_algo after reconnection"
done
return 0
}
run_test 77l "preferred checksum type is remembered after reconnected"
[ "$ORIG_CSUM" ] && set_checksums $ORIG_CSUM || true
rm -f $F77_TMP
unset F77_TMP
test_77m() {
(( $CLIENT_VERSION >= $(version_code 2.14.52) )) ||
skip "Need at least version 2.14.52"
local param=checksum_speed
$LCTL get_param $param || error "reading $param failed"
csum_speeds=$($LCTL get_param -n $param)
[[ "$csum_speeds" =~ "adler32" && "$csum_speeds" =~ "crc32" ]] ||
error "known checksum types are missing"
}
run_test 77m "Verify checksum_speed is correctly read"
check_filefrag_77n() {
local nr_ext=0
local starts=()
local ends=()
while read extidx a b start end rest; do
if [[ "${extidx}" =~ ^[0-9]+: ]]; then
nr_ext=$(( $nr_ext + 1 ))
starts+=( ${start%..} )
ends+=( ${end%:} )
fi
done < <( filefrag -sv $1 )
[[ $nr_ext -eq 2 ]] && [[ "${starts[-1]}" == $(( ${ends[0]} + 1 )) ]] && return 0
return 1
}
test_77n() {
local filefrag_op=$(filefrag -l 2>&1 | grep "invalid option")
[[ -z "$filefrag_op" ]] || skip_env "filefrag missing logical ordering"
[[ "$CKSUM_TYPES" =~ t10 ]] || skip "no T10 checksum support on osc"
touch $DIR/$tfile
$TRUNCATE $DIR/$tfile 0
dd if=/dev/urandom of=$DIR/$tfile bs=4k conv=notrunc count=1 seek=0
dd if=/dev/urandom of=$DIR/$tfile bs=4k conv=notrunc count=1 seek=2
check_filefrag_77n $DIR/$tfile ||
skip "$tfile blocks not contiguous around hole"
set_checksums 1
stack_trap "set_checksums $ORIG_CSUM" EXIT
stack_trap "set_checksum_type $ORIG_CSUM_TYPE" EXIT
stack_trap "rm -f $DIR/$tfile"
for algo in $CKSUM_TYPES; do
if [[ "$algo" =~ ^t10 ]]; then
set_checksum_type $algo ||
error "fail to set checksum type $algo"
dd if=$DIR/$tfile of=/dev/null bs=12k count=1 iflag=direct ||
error "fail to read $tfile with $algo"
fi
done
rm -f $DIR/$tfile
return 0
}
run_test 77n "Verify read from a hole inside contiguous blocks with T10PI"
test_77o() {
(( $MDS1_VERSION >= $(version_code 2.14.55) )) ||
skip "Need MDS version at least 2.14.55"
(( $OST1_VERSION >= $(version_code 2.14.55) )) ||
skip "Need OST version at least 2.14.55"
local ofd=obdfilter
local mdt=mdt
# print OST checksum_type
echo "$ofd.$FSNAME-*.checksum_type:"
do_nodes $(osts_nodes) "$LCTL get_param -n $ofd.$FSNAME-*.checksum_type"
# print MDT checksum_type
echo "$mdt.$FSNAME-*.checksum_type:"
do_nodes $(mdts_nodes) "$LCTL get_param -n $mdt.$FSNAME-*.checksum_type"
local o_count=$(do_nodes $(osts_nodes) \
$LCTL get_param -n $ofd.$FSNAME-*.checksum_type | wc -l)
(( $o_count == $OSTCOUNT )) ||
error "found $o_count checksums, not \$OSTCOUNT=$OSTCOUNT"
local m_count=$(do_nodes $(mdts_nodes) \
$LCTL get_param -n $mdt.$FSNAME-*.checksum_type | wc -l)
(( $m_count == $MDSCOUNT )) ||
error "found $m_count checksums, not \$MDSCOUNT=$MDSCOUNT"
}
run_test 77o "Verify checksum_type for server (mdt and ofd(obdfilter))"
cleanup_test_78() {
trap 0
rm -f $DIR/$tfile
}
test_78() { # bug 10901
[ $PARALLEL == "yes" ] && skip "skip parallel run"
remote_ost || skip_env "local OST"
NSEQ=5
F78SIZE=$(($(awk '/MemFree:/ { print $2 }' /proc/meminfo) / 1024))
echo "MemFree: $F78SIZE, Max file size: $MAXFREE"
MEMTOTAL=$(($(awk '/MemTotal:/ { print $2 }' /proc/meminfo) / 1024))
echo "MemTotal: $MEMTOTAL"
# reserve 256MB of memory for the kernel and other running processes,
# and then take 1/2 of the remaining memory for the read/write buffers.
if [ $MEMTOTAL -gt 512 ] ;then
MEMTOTAL=$(((MEMTOTAL - 256 ) / 2))
else
# for those poor memory-starved high-end clusters...
MEMTOTAL=$((MEMTOTAL / 2))
fi
echo "Mem to use for directio: $MEMTOTAL"
[[ $F78SIZE -gt $MEMTOTAL ]] && F78SIZE=$MEMTOTAL
[[ $F78SIZE -gt 512 ]] && F78SIZE=512
[[ $F78SIZE -gt $((MAXFREE / 1024)) ]] && F78SIZE=$((MAXFREE / 1024))
SMALLESTOST=$($LFS df --output=avail --no-header -o $DIR | sort -n |
head -1)
echo "Smallest OST: $SMALLESTOST"
[[ $SMALLESTOST -lt 10240 ]] &&
skip "too small OSTSIZE, useless to run large O_DIRECT test"
trap cleanup_test_78 EXIT
[[ $F78SIZE -gt $((SMALLESTOST * $OSTCOUNT / 1024 - 80)) ]] &&
F78SIZE=$((SMALLESTOST * $OSTCOUNT / 1024 - 80))
[ "$SLOW" = "no" ] && NSEQ=1 && [ $F78SIZE -gt 32 ] && F78SIZE=32
echo "File size: $F78SIZE"
$LFS setstripe -c $OSTCOUNT $DIR/$tfile || error "setstripe failed"
for i in $(seq 1 $NSEQ); do
FSIZE=$(($F78SIZE / ($NSEQ - $i + 1)))
echo directIO rdwr round $i of $NSEQ
$DIRECTIO rdwr $DIR/$tfile 0 $FSIZE 1048576||error "rdwr failed"
done
cleanup_test_78
}
run_test 78 "handle large O_DIRECT writes correctly ============"
test_79() { # bug 12743
[[ $PARALLEL != "yes" ]] || skip "skip parallel run"
wait_delete_completed
local bktotal=$(calc_osc_kbytes kbytestotal)
local bkfree=$(calc_osc_kbytes kbytesfree)
local bkavail=$(calc_osc_kbytes kbytesavail)
local string=($(df -P $MOUNT | tail -n 1))
local dftotal=${string[1]}
local dfused=${string[2]}
local dfavail=${string[3]}
local dffree=$(($dftotal - $dfused))
local allowance=$((64 * $OSTCOUNT))
$LFS df $MOUNT
echo ${string[*]}
(( dftotal >= bktotal - allowance && dftotal <= bktotal + allowance)) ||
error "df total($dftotal) mismatch OST total($bktotal)"
(( dffree >= bkfree - allowance && dffree <= bkfree + allowance )) ||
error "df free($dffree) mismatch OST free($bkfree)"
(( dfavail >= bkavail - allowance && dfavail <= bkavail + allowance)) ||
error "df avail($dfavail) mismatch OST avail($bkavail)"
$LCTL get_param *.$FSNAME-*.{maxbytes,namelen_max}
(( $MDS1_VERSION < $(version_code 2.16.52-70) )) ||
do_nodes $(mdts_nodes) \
$LCTL get_param *.$FSNAME-*.{maxbytes,namelen_max}
(( $OST1_VERSION < $(version_code 2.16.52-70) )) ||
do_nodes $(osts_nodes) \
$LCTL get_param *.$FSNAME-*.{maxbytes,namelen_max}
local dfnamelen=$(stat -f -c %l $MOUNT)
local lnamelen=($($LCTL get_param -n llite.$FSNAME-*.namelen_max))
(( $dfnamelen == $lnamelen )) ||
error "df namelen($dfnamelen) != llite namelen($lnamelen)"
}
run_test 79 "df report consistency check"
test_80() { # bug 10718
remote_ost_nodsh && skip "remote OST with nodsh"
[ $PARALLEL == "yes" ] && skip "skip parallel run"
# relax strong synchronous semantics for slow backends like ZFS
if [ "$ost1_FSTYPE" != "ldiskfs" ]; then
local soc="obdfilter.*.sync_lock_cancel"
local save=$(do_facet ost1 $LCTL get_param -n $soc | head -n1)
# "sync_on_lock_cancel" was broken by v2_11_55_0-26-g7059644e9a
# Fixed in 2.13.51~79, so we'll keep it for interop with 2.12.x
if [ -z "$save" ]; then
soc="obdfilter.*.sync_on_lock_cancel"
save=$(do_facet ost1 $LCTL get_param -n $soc | head -n1)
fi
if [ "$save" != "never" ]; then
local osts=$(osts_nodes)
do_nodes $osts $LCTL set_param $soc=never
stack_trap "do_nodes $osts $LCTL set_param $soc=$save"
fi
fi
dd if=/dev/zero of=$DIR/$tfile bs=1M count=1 seek=1M
sync; sleep 1; sync
local before=$(date +%s)
cancel_lru_locks osc
local after=$(date +%s)
local diff=$((after - before))
[ $diff -le 1 ] || error "elapsed for 1M@1T = $diff"
rm -f $DIR/$tfile
}
run_test 80 "Page eviction is equally fast at high offsets too"
test_81a() { # LU-456
[ $PARALLEL == "yes" ] && skip "skip parallel run"
remote_ost_nodsh && skip "remote OST with nodsh"
# define OBD_FAIL_OST_MAPBLK_ENOSPC 0x228
# MUST OR with the CFS_FAIL_ONCE (0x80000000)
do_facet ost1 lctl set_param fail_loc=0x80000228
# write should trigger a retry and success
$LFS setstripe -i 0 -c 1 $DIR/$tfile
$MULTIOP $DIR/$tfile oO_CREAT:O_RDWR:O_SYNC:w4096c
RC=$?
if [ $RC -ne 0 ] ; then
error "write should success, but failed for $RC"
fi
}
run_test 81a "OST should retry write when get -ENOSPC ==============="
test_81b() { # LU-456
[ $PARALLEL == "yes" ] && skip "skip parallel run"
remote_ost_nodsh && skip "remote OST with nodsh"
# define OBD_FAIL_OST_MAPBLK_ENOSPC 0x228
# Don't OR with the CFS_FAIL_ONCE (0x80000000)
do_facet ost1 lctl set_param fail_loc=0x228
# write should retry several times and return -ENOSPC finally
$LFS setstripe -i 0 -c 1 $DIR/$tfile
$MULTIOP $DIR/$tfile oO_CREAT:O_RDWR:O_SYNC:w4096c
RC=$?
ENOSPC=28
if [ $RC -ne $ENOSPC ] ; then
error "dd should fail for -ENOSPC, but succeed."
fi
}
run_test 81b "OST should return -ENOSPC when retry still fails ======="
test_99() {
[ -z "$(which cvs 2>/dev/null)" ] && skip_env "could not find cvs"
test_mkdir $DIR/$tdir.cvsroot
chown $RUNAS_ID $DIR/$tdir.cvsroot
cd $TMP
$RUNAS cvs -d $DIR/$tdir.cvsroot init || error "cvs init failed"
cd /etc/init.d
# some versions of cvs import exit(1) when asked to import links or
# files they can't read. ignore those files.
local toignore=$(find . -type l -printf '-I %f\n' -o \
! -perm /4 -printf '-I %f\n')
$RUNAS cvs -d $DIR/$tdir.cvsroot import -m "nomesg" $toignore \
$tdir.reposname vtag rtag
cd $DIR
test_mkdir $DIR/$tdir.reposname
chown $RUNAS_ID $DIR/$tdir.reposname
$RUNAS cvs -d $DIR/$tdir.cvsroot co $tdir.reposname
cd $DIR/$tdir.reposname
$RUNAS touch foo99
$RUNAS cvs add -m 'addmsg' foo99
$RUNAS cvs update
$RUNAS cvs commit -m 'nomsg' foo99
rm -fr $DIR/$tdir.cvsroot
}
run_test 99 "cvs strange file/directory operations"
test_100() {
[ $PARALLEL == "yes" ] && skip "skip parallel run"
[[ "$NETTYPE" =~ tcp ]] ||
skip_env "TCP secure port test, not useful for NETTYPE=$NETTYPE"
[[ -n "$(type -p ss)" ]] || skip_env "ss not available"
remote_ost_nodsh && skip "remote OST with nodsh"
remote_mds_nodsh && skip "remote MDS with nodsh"
remote_servers || skip "useless for local single node setup"
ss -tna | ( rc=1; while read STATE SND RCV LOCAL REMOTE STAT; do
[[ "${REMOTE/*:/}" == "$ACCEPTOR_PORT" ]] || continue
rc=0
if (( ${LOCAL/*:/} >= 1024 )); then
echo "bad: $PROT $SND $RCV $LOCAL $REMOTE $STAT"
ss -tna
error "local: ${LOCAL/*:/} > 1024 remote: ${REMOTE/*:/}"
fi
done
(( $rc == 0 )) || error "privileged port not found" )
}
run_test 100 "check local port using privileged port"
function get_named_value()
{
local tag=$1
grep -w "$tag" | sed "s/^$tag *\([0-9]*\) *.*/\1/"
}
test_101a() {
[[ $PARALLEL != "yes" ]] || skip "skip parallel run"
local s
local discard
local nreads=10000
local cache_limit=32
$LCTL set_param -n osc.*-osc*.rpc_stats=0
$LCTL set_param -n llite.*.read_ahead_stats=0
local max_cached_mb=$($LCTL get_param llite.*.max_cached_mb |
awk '/^max_cached_mb/ { print $2 }')
stack_trap "$LCTL set_param -n llite.*.max_cached_mb=$max_cached_mb"
$LCTL set_param -n llite.*.max_cached_mb=$cache_limit
#
# randomly read 10000 of 64K chunks from file 3x 32MB in size
#
echo "nreads: $nreads file size: $((cache_limit * 3))MB"
$READS -f $DIR/$tfile -s$((cache_limit * 3192 * 1024)) -b65536 -C -n$nreads -t 180
discard=0
for s in $($LCTL get_param -n llite.*.read_ahead_stats |
get_named_value 'read.but.discarded'); do
discard=$(($discard + $s))
done
$LCTL get_param osc.*-osc*.rpc_stats
$LCTL get_param llite.*.read_ahead_stats
# Discard is generally zero, but sometimes a few random reads line up
# and trigger larger readahead, which is wasted & leads to discards.
if (( $discard > $nreads )); then
error "too many discarded pages ($discard > $nreads)"
fi
rm -f $DIR/$tfile || true
}
run_test 101a "check read-ahead for random reads"
setup_test101bc() {
test_mkdir $DIR/$tdir
local ssize=$1
local FILE_LENGTH=$2
STRIPE_OFFSET=0
local FILE_SIZE_MB=$((FILE_LENGTH / ssize))
local osts=$(osts_nodes)
set_osd_param $osts '' read_cache_enable 0
set_osd_param $osts '' writethrough_cache_enable 0
trap cleanup_test101bc EXIT
# prepare the read-ahead file
$LFS setstripe -S $ssize -i $STRIPE_OFFSET -c $OSTCOUNT $DIR/$tfile
dd if=/dev/zero of=$DIR/$tfile bs=$ssize \
count=$FILE_SIZE_MB 2> /dev/null
}
cleanup_test101bc() {
trap 0
rm -rf $DIR/$tdir
rm -f $DIR/$tfile
local osts=$(osts_nodes)
set_osd_param $osts '' read_cache_enable 1
set_osd_param $osts '' writethrough_cache_enable 1
}
ra_check_101() {
local read_size=$1
local stripe_size=$2
local stride_length=$((stripe_size / read_size))
local stride_width=$((stride_length * OSTCOUNT))
local discard_limit=$(( ((stride_length - 1) * 3 / stride_width) *
(stride_width - stride_length) ))
local discard=$($LCTL get_param -n llite.*.read_ahead_stats |
get_named_value 'read.but.discarded' | calc_sum)
if [[ $discard -gt $discard_limit ]]; then
$LCTL get_param llite.*.read_ahead_stats
error "($discard limit ${discard_limit}) discarded pages with size (${read_size})"
else
echo "Read-ahead success for size ${read_size}"
fi
}
test_101b() {
[ $PARALLEL == "yes" ] && skip "skip parallel run"
[[ $OSTCOUNT -lt 2 ]] && skip_env "needs >= 2 OSTs"
local STRIPE_SIZE=1048576
local STRIDE_SIZE=$((STRIPE_SIZE*OSTCOUNT))
if [ $SLOW == "yes" ]; then
local FILE_LENGTH=$((STRIDE_SIZE * 64))
else
local FILE_LENGTH=$((STRIDE_SIZE * 8))
fi
local ITERATION=$((FILE_LENGTH / STRIDE_SIZE))
# prepare the read-ahead file
setup_test101bc $STRIPE_SIZE $FILE_LENGTH
cancel_lru_locks osc
for BIDX in 2 4 8 16 32 64 128 256
do
local BSIZE=$((BIDX*4096))
local READ_COUNT=$((STRIPE_SIZE/BSIZE))
local STRIDE_LENGTH=$((STRIDE_SIZE/BSIZE))
local OFFSET=$((STRIPE_SIZE/BSIZE*(OSTCOUNT - 1)))
$LCTL set_param -n llite.*.read_ahead_stats=0
$READS -f $DIR/$tfile -l $STRIDE_LENGTH -o $OFFSET \
-s $FILE_LENGTH -b $STRIPE_SIZE -a $READ_COUNT -n $ITERATION
cancel_lru_locks osc
ra_check_101 $BSIZE $STRIPE_SIZE $FILE_LENGTH
done
cleanup_test101bc
true
}
run_test 101b "check stride-io mode read-ahead ================="
test_101c() {
[ $PARALLEL == "yes" ] && skip "skip parallel run"
local STRIPE_SIZE=1048576
local FILE_LENGTH=$((STRIPE_SIZE*100))
local nreads=10000
local rsize=65536
local osc_rpc_stats
setup_test101bc $STRIPE_SIZE $FILE_LENGTH
cancel_lru_locks osc
$LCTL set_param osc.*.rpc_stats=0
$READS -f $DIR/$tfile -s$FILE_LENGTH -b$rsize -n$nreads -t 180
$LCTL get_param osc.*.rpc_stats
for osc_rpc_stats in $($LCTL get_param -N osc.*.rpc_stats); do
local stats=$($LCTL get_param -n $osc_rpc_stats)
local lines=$(echo "$stats" | awk 'END {print NR;}')
local size
if [ $lines -le 20 ]; then
echo "continue debug"
continue
fi
for size in 1 2 4 8; do
local rpc=$(echo "$stats" |
awk '($1 == "'$size':") {print $2; exit; }')
[ $rpc != 0 ] && ((size * PAGE_SIZE < rsize)) &&
error "Small $((size*PAGE_SIZE)) read IO $rpc!"
done
echo "$osc_rpc_stats check passed!"
done
cleanup_test101bc
true
}
run_test 101c "check stripe_size aligned read-ahead"
test_101d() {
[ $PARALLEL == "yes" ] && skip "skip parallel run"
local file=$DIR/$tfile
local sz_MB=${FILESIZE_101d:-80}
local ra_MB=${READAHEAD_MB:-40}
local free_MB=$(($(df -P $DIR | tail -n 1 | awk '{ print $4 }') / 1024))
[ $free_MB -lt $sz_MB ] &&
skip "Need free space ${sz_MB}M, have ${free_MB}M"
echo "Create test file $file size ${sz_MB}M, ${free_MB}M free"
$LFS setstripe -c -1 $file || error "setstripe failed"
dd if=/dev/zero of=$file bs=1M count=$sz_MB || error "dd failed"
echo Cancel LRU locks on lustre client to flush the client cache
cancel_lru_locks osc
echo Disable read-ahead
local old_RA=$($LCTL get_param -n llite.*.max_read_ahead_mb | head -n 1)
$LCTL set_param -n llite.*.max_read_ahead_mb=0
stack_trap "$LCTL set_param -n llite.*.max_read_ahead_mb=$old_RA" EXIT
$LCTL get_param -n llite.*.max_read_ahead_mb
echo "Reading the test file $file with read-ahead disabled"
local sz_KB=$((sz_MB * 1024 / 4))
# 10485760 bytes transferred in 0.000938 secs (11179579337 bytes/sec)
# 104857600 bytes (105 MB) copied, 0.00876352 s, 12.0 GB/s
# 83886080 bytes (84 MB, 80 MiB) copied, 16 s, 5.2 MB/s
local raOFF=$(LANG=C dd if=$file of=/dev/null bs=4k count=$sz_KB |&
sed -e '/records/d' -e 's/.* \([0-9][0-9\.]*\) *s.*/\1/')
echo "Cancel LRU locks on lustre client to flush the client cache"
cancel_lru_locks osc
echo Enable read-ahead with ${ra_MB}MB
$LCTL set_param -n llite.*.max_read_ahead_mb=$ra_MB
echo "Reading the test file $file with read-ahead enabled"
local raON=$(LANG=C dd if=$file of=/dev/null bs=4k count=$sz_KB |&
sed -e '/records/d' -e 's/.* \([0-9][0-9\.]*\) *s.*/\1/')
echo "read-ahead disabled time read '$raOFF'"
echo "read-ahead enabled time read '$raON'"
rm -f $file
wait_delete_completed
# use awk for this check instead of bash because it handles decimals
awk "{ exit !($raOFF < 0.5 || $raOFF > $raON) }" <<<"ignore_me" ||
error "readahead ${raON}s > no-readahead ${raOFF}s (${sz_MB}M)"
}
run_test 101d "file read with and without read-ahead enabled"
test_101e() {
[ $PARALLEL == "yes" ] && skip "skip parallel run"
local file=$DIR/$tfile
local size_KB=500 #KB
local count=100
local bsize=1024
local free_KB=$(df -P $DIR | tail -n 1 | awk '{ print $4 }')
local need_KB=$((count * size_KB))
[[ $free_KB -le $need_KB ]] &&
skip_env "Need free space $need_KB, have $free_KB"
echo "Creating $count ${size_KB}K test files"
for ((i = 0; i < $count; i++)); do
dd if=/dev/zero of=$file.$i bs=$bsize count=$size_KB 2>/dev/null
done
echo "Cancel LRU locks on lustre client to flush the client cache"
cancel_lru_locks $OSC
echo "Reset readahead stats"
$LCTL set_param -n llite.*.read_ahead_stats=0
for ((i = 0; i < $count; i++)); do
dd if=$file.$i of=/dev/null bs=$bsize count=$size_KB 2>/dev/null
done
$LCTL get_param llite.*.max_cached_mb
$LCTL get_param llite.*.read_ahead_stats
local miss=$($LCTL get_param -n llite.*.read_ahead_stats |
get_named_value 'misses' | calc_sum)
for ((i = 0; i < $count; i++)); do
rm -rf $file.$i 2>/dev/null
done
#10000 means 20% reads are missing in readahead
[[ $miss -lt 10000 ]] || error "misses too much for small reads"
}
run_test 101e "check read-ahead for small read(1k) for small files(500k)"
test_101f() {
which iozone || skip_env "no iozone installed"
local old_debug=$($LCTL get_param debug)
old_debug=${old_debug#*=}
$LCTL set_param debug="reada mmap"
# create a test file
iozone -i 0 -+n -r 1m -s 128m -w -f $DIR/$tfile > /dev/null 2>&1
echo Cancel LRU locks on lustre client to flush the client cache
cancel_lru_locks osc
echo Reset readahead stats
$LCTL set_param -n llite.*.read_ahead_stats=0
echo mmap read the file with small block size
iozone -i 1 -u 1 -l 1 -+n -r 32k -s 128m -B -f $DIR/$tfile \
> /dev/null 2>&1
echo checking missing pages
$LCTL get_param llite.*.read_ahead_stats
local miss=$($LCTL get_param -n llite.*.read_ahead_stats |
get_named_value 'misses' | calc_sum)
$LCTL set_param debug="$old_debug"
[ $miss -lt 3 ] || error "misses too much pages ('$miss')!"
rm -f $DIR/$tfile
}
run_test 101f "check mmap read performance"
test_101g_brw_size_test() {
local mb=$1
local pages=$((mb * 1048576 / PAGE_SIZE))
local file=$DIR/$tfile
$LCTL set_param osc.*.max_pages_per_rpc=${mb}M ||
{ error "unable to set max_pages_per_rpc=${mb}M"; return 1; }
for mp in $($LCTL get_param -n osc.*.max_pages_per_rpc); do
[ $mp -ne $pages ] && error "max_pages_per_rpc $mp != $pages" &&
return 2
done
stack_trap "rm -f $file; wait_delete_completed" EXIT
$LCTL set_param -n osc.*.rpc_stats=0
# 10 RPCs should be enough for the test
local count=10
dd if=/dev/zero of=$file bs=${mb}M count=$count ||
{ error "dd write ${mb} MB blocks failed"; return 3; }
cancel_lru_locks osc
dd of=/dev/null if=$file bs=${mb}M count=$count ||
{ error "dd write ${mb} MB blocks failed"; return 4; }
# calculate number of full-sized read and write RPCs
rpcs=($($LCTL get_param -n 'osc.*.rpc_stats' |
sed -n '/pages per rpc/,/^$/p' |
awk '/'$pages':/ { reads += $2; writes += $6 }; \
END { print reads,writes }'))
# allow one extra full-sized read RPC for async readahead
[[ ${rpcs[0]} == $count || ${rpcs[0]} == $((count + 1)) ]] ||
{ error "${rpcs[0]} != $count read RPCs"; return 5; }
[[ ${rpcs[1]} == $count ]] ||
{ error "${rpcs[1]} != $count write RPCs"; return 6; }
}
test_101g() {
remote_ost_nodsh && skip "remote OST with nodsh"
local rpcs
local facets=$(get_facets OST)
local p="$TMP/$TESTSUITE-$TESTNAME.parameters"
local brw_size="obdfilter.*.brw_size"
$LFS setstripe -i 0 -c 1 $DIR/$tfile
local orig_mb=$(do_facet ost1 $LCTL get_param -n $brw_size | head -n 1)
if (( $OST1_VERSION >= $(version_code 2.8.52) )); then
local osts=$(osts_nodes)
(( $OST1_VERSION < $(version_code 2.9.52) )) || suffix="M"
if (( $orig_mb < 16 )); then
save_lustre_params $facets "$brw_size" > $p
stack_trap "rm -f $p"
do_nodes $osts $LCTL set_param -n $brw_size=16$suffix ||
error "set 16MB RPC size failed"
echo "remount client to enable new RPC size"
remount_client $MOUNT || error "remount_client failed"
fi
test_101g_brw_size_test 16 || error "16MB RPC test failed"
# should be able to set brw_size=12, but no rpc_stats for that
test_101g_brw_size_test 8 || error "8MB RPC test failed"
fi
test_101g_brw_size_test 4 || error "4MB RPC test failed"
if (( $orig_mb < 16 )); then
restore_lustre_params < $p
remount_client $MOUNT || error "remount_client restore failed"
fi
}
run_test 101g "Big bulk(4/16 MiB) readahead"
test_101h() {
$LFS setstripe -i 0 -c 1 $DIR/$tfile
local hybrid=$($LCTL get_param -n llite.*.hybrid_io)
$LCTL set_param llite.*.hybrid_io=0
stack_trap "$LCTL set_param -n llite.*.hybrid_io=$hybrid" EXIT
dd if=/dev/zero of=$DIR/$tfile bs=1M count=70 ||
error "dd 70M file failed"
echo Cancel LRU locks on lustre client to flush the client cache
cancel_lru_locks osc
echo "Reset readahead stats"
$LCTL set_param -n llite.*.read_ahead_stats 0
echo "Read 10M of data but cross 64M bundary"
dd if=$DIR/$tfile of=/dev/null bs=10M skip=6 count=1
local miss=$($LCTL get_param -n llite.*.read_ahead_stats |
get_named_value 'misses' | calc_sum)
[ $miss -eq 1 ] || error "expected miss 1 but got $miss"
rm -f $p $DIR/$tfile
}
run_test 101h "Readahead should cover current read window"
test_101i() {
dd if=/dev/zero of=$DIR/$tfile bs=1M count=10 ||
error "dd 10M file failed"
local max_per_file_mb=$($LCTL get_param -n \
llite.*.max_read_ahead_per_file_mb 2>/dev/null)
cancel_lru_locks osc
stack_trap "$LCTL set_param llite.*.max_read_ahead_per_file_mb=$max_per_file_mb"
$LCTL set_param llite.*.max_read_ahead_per_file_mb=1 ||
error "set max_read_ahead_per_file_mb to 1 failed"
echo "Reset readahead stats"
$LCTL set_param llite.*.read_ahead_stats=0
dd if=$DIR/$tfile of=/dev/null bs=2M
$LCTL get_param llite.*.read_ahead_stats
local miss=$($LCTL get_param -n llite.*.read_ahead_stats |
awk '/misses/ { print $2 }')
[ $miss -eq 5 ] || error "expected misses 5 but got $miss"
rm -f $DIR/$tfile
}
run_test 101i "allow current readahead to exceed reservation"
test_101j() {
$LFS setstripe -i 0 -c 1 $DIR/$tfile ||
error "setstripe $DIR/$tfile failed"
local file_size=$((1048576 * 16))
local old_ra=$($LCTL get_param -n llite.*.max_read_ahead_mb | head -n 1)
stack_trap "$LCTL set_param -n llite.*.max_read_ahead_mb $old_ra" EXIT
local hybrid=$($LCTL get_param -n llite.*.hybrid_io)
$LCTL set_param llite.*.hybrid_io=0
stack_trap "$LCTL set_param -n llite.*.hybrid_io=$hybrid" EXIT
echo Disable read-ahead
$LCTL set_param -n llite.*.max_read_ahead_mb=0
dd if=/dev/zero of=$DIR/$tfile bs=1M count=$(($file_size / 1048576))
for blk in $PAGE_SIZE 1048576 $file_size; do
cancel_lru_locks osc
echo "Reset readahead stats"
$LCTL set_param -n llite.*.read_ahead_stats=0
local count=$(($file_size / $blk))
dd if=$DIR/$tfile bs=$blk count=$count of=/dev/null
local miss=$($LCTL get_param -n llite.*.read_ahead_stats |
get_named_value 'failed.to.fast.read' | calc_sum)
$LCTL get_param -n llite.*.read_ahead_stats
[ $miss -eq $count ] || error "expected $count got $miss"
done
rm -f $p $DIR/$tfile
}
run_test 101j "A complete read block should be submitted when no RA"
test_readahead_base() {
local file=$DIR/$tfile
local size=$1
local iosz
local ramax
local ranum
$LCTL set_param -n llite.*.read_ahead_stats=0
# The first page is not accounted into readahead
ramax=$(((size + PAGE_SIZE - 1) / PAGE_SIZE - 1))
iosz=$(((size + 1048575) / 1048576 * 1048576))
echo "Test readahead: size=$size ramax=$ramx iosz=$iosz"
$LCTL mark "Test readahead: size=$size ramax=$ramx iosz=$iosz"
fallocate -l $size $file || error "failed to fallocate $file"
cancel_lru_locks osc
$MULTIOP $file or${iosz}c || error "failed to read $file"
$LCTL get_param -n llite.*.read_ahead_stats
ranum=$($LCTL get_param -n llite.*.read_ahead_stats |
awk '/readahead.pages/ { print $7 }' | calc_sum)
(( $ranum <= $ramax )) ||
error "read-ahead pages is $ranum more than $ramax"
rm -rf $file || error "failed to remove $file"
}
test_101m()
{
local file=$DIR/$tfile
local ramax
local ranum
local size
local iosz
check_set_fallocate_or_skip
stack_trap "rm -f $file" EXIT
test_readahead_base 4096
# file size: 16K = 16384
test_readahead_base 16384
test_readahead_base 16385
test_readahead_base 16383
# file size: 1M + 1 = 1048576 + 1
test_readahead_base 1048577
# file size: 1M + 16K
test_readahead_base $((1048576 + 16384))
# file size: stripe_size * (stripe_count - 1) + 16K
$LFS setstripe -c -1 $file || error "failed to setstripe $file"
test_readahead_base $((1048576 * (OSTCOUNT - 1) + 16384))
# file size: stripe_size * stripe_count + 16K
$LFS setstripe -c -1 $file || error "failed to setstripe $file"
test_readahead_base $((1048576 * OSTCOUNT + 16384))
# file size: 2 * stripe_size * stripe_count + 16K
$LFS setstripe -c -1 $file || error "failed to setstripe $file"
test_readahead_base $((2 * 1048576 * OSTCOUNT + 16384))
}
run_test 101m "read ahead for small file and last stripe of the file"
setup_test102() {
test_mkdir $DIR/$tdir
chown $RUNAS_ID $DIR/$tdir
STRIPE_SIZE=65536
STRIPE_OFFSET=1
STRIPE_COUNT=$OSTCOUNT
[[ $OSTCOUNT -gt 4 ]] && STRIPE_COUNT=4
trap cleanup_test102 EXIT
cd $DIR
$1 $LFS setstripe -S $STRIPE_SIZE -i $STRIPE_OFFSET -c $STRIPE_COUNT $tdir
cd $DIR/$tdir
for num in 1 2 3 4; do
for count in $(seq 1 $STRIPE_COUNT); do
for idx in $(seq 0 $[$STRIPE_COUNT - 1]); do
local size=`expr $STRIPE_SIZE \* $num`
local file=file"$num-$idx-$count"
$1 $LFS setstripe -S $size -i $idx -c $count $file
done
done
done
cd $DIR
$1 tar cf $TMP/f102.tar $tdir --xattrs
}
cleanup_test102() {
trap 0
rm -f $TMP/f102.tar
rm -rf $DIR/d0.sanity/d102
}
test_102a() {
[ "$UID" != 0 ] && skip "must run as root"
[ -z "$(lctl get_param -n mdc.*-mdc-*.connect_flags | grep xattr)" ] &&
skip_env "must have user_xattr"
[ -z "$(which setfattr 2>/dev/null)" ] &&
skip_env "could not find setfattr"
local testfile=$DIR/$tfile
touch $testfile
echo "set/get xattr..."
setfattr -n trusted.name1 -v value1 $testfile ||
error "setfattr -n trusted.name1=value1 $testfile failed"
getfattr -n trusted.name1 $testfile 2> /dev/null |
grep "trusted.name1=.value1" ||
error "$testfile missing trusted.name1=value1"
setfattr -n user.author1 -v author1 $testfile ||
error "setfattr -n user.author1=author1 $testfile failed"
getfattr -n user.author1 $testfile 2> /dev/null |
grep "user.author1=.author1" ||
error "$testfile missing trusted.author1=author1"
echo "listxattr..."
setfattr -n trusted.name2 -v value2 $testfile ||
error "$testfile unable to set trusted.name2"
setfattr -n trusted.name3 -v value3 $testfile ||
error "$testfile unable to set trusted.name3"
[ $(getfattr -d -m "^trusted" $testfile 2> /dev/null |
grep "trusted.name" | wc -l) -eq 3 ] ||
error "$testfile missing 3 trusted.name xattrs"
setfattr -n user.author2 -v author2 $testfile ||
error "$testfile unable to set user.author2"
setfattr -n user.author3 -v author3 $testfile ||
error "$testfile unable to set user.author3"
[ $(getfattr -d -m "^user" $testfile 2> /dev/null |
grep "user.author" | wc -l) -eq 3 ] ||
error "$testfile missing 3 user.author xattrs"
echo "remove xattr..."
setfattr -x trusted.name1 $testfile ||
error "$testfile error deleting trusted.name1"
getfattr -d -m trusted $testfile 2> /dev/null | grep "trusted.name1" &&
error "$testfile did not delete trusted.name1 xattr"
setfattr -x user.author1 $testfile ||
error "$testfile error deleting user.author1"
echo "set lustre special xattr ..."
$LFS setstripe -c1 $testfile
local lovea=$(getfattr -n "trusted.lov" -e hex $testfile |
awk -F "=" '/trusted.lov/ { print $2 }' )
setfattr -n "trusted.lov" -v $lovea $testfile ||
error "$testfile doesn't ignore setting trusted.lov again"
setfattr -n "trusted.lov" -v "invalid_value" $testfile &&
error "$testfile allow setting invalid trusted.lov"
rm -f $testfile
}
run_test 102a "user xattr test =================================="
check_102b_layout() {
local layout="$*"
local testfile=$DIR/$tfile
echo "test layout '$layout'"
$LFS setstripe $layout $testfile || error "setstripe failed"
$LFS getstripe -y $testfile
echo "get/set/list trusted.lov xattr ..." # b=10930
local value=$(getfattr -n trusted.lov -e hex $testfile | grep trusted)
[[ "$value" =~ "trusted.lov" ]] ||
error "can't get trusted.lov from $testfile"
local stripe_count_orig=$($LFS getstripe -c $testfile) ||
error "getstripe failed"
$MCREATE $testfile.2 || error "mcreate $testfile.2 failed"
value=$(cut -d= -f2 <<<$value)
# LU-13168: truncated xattr should fail if short lov_user_md header
[ $CLIENT_VERSION -lt $(version_code 2.13.53) ] &&
lens="${#value}" || lens="$(seq 4 2 ${#value})"
for len in $lens; do
echo "setfattr $len $testfile.2"
setfattr -n trusted.lov -v ${value:0:$len} $testfile.2 &&
[ $len -lt 66 ] && error "short xattr len=$len worked"
done
local stripe_size=$($LFS getstripe -S $testfile.2)
local stripe_count=$($LFS getstripe -c $testfile.2)
[[ $stripe_size -eq 65536 ]] ||
error "stripe size $stripe_size != 65536"
[[ $stripe_count -eq $stripe_count_orig ]] ||
error "stripe count $stripe_count != $stripe_count_orig"
rm $testfile $testfile.2
}
test_102b() {
[ -z "$(which setfattr 2>/dev/null)" ] &&
skip_env "could not find setfattr"
[[ $OSTCOUNT -lt 2 ]] && skip_env "needs >= 2 OSTs"
# check plain layout
check_102b_layout -S 65536 -i 1 -c $OSTCOUNT
# and also check composite layout
check_102b_layout -E 1M -S 65536 -i 1 -c $OSTCOUNT -Eeof -S4M
}
run_test 102b "getfattr/setfattr for trusted.lov EAs"
test_102c() {
[ -z "$(which setfattr 2>/dev/null)" ] &&
skip_env "could not find setfattr"
[[ $OSTCOUNT -lt 2 ]] && skip_env "needs >= 2 OSTs"
# b10930: get/set/list lustre.lov xattr
echo "get/set/list lustre.lov xattr ..."
test_mkdir $DIR/$tdir
chown $RUNAS_ID $DIR/$tdir
local testfile=$DIR/$tdir/$tfile
$RUNAS $LFS setstripe -S 65536 -i 1 -c $OSTCOUNT $testfile ||
error "setstripe failed"
local STRIPECOUNT=$($RUNAS $LFS getstripe -c $testfile) ||
error "getstripe failed"
$RUNAS getfattr -d -m "^lustre" $testfile 2> /dev/null | \
grep "lustre.lov" || error "can't get lustre.lov from $testfile"
local testfile2=${testfile}2
local value=`getfattr -n lustre.lov $testfile 2> /dev/null | \
grep "lustre.lov" |sed -e 's/[^=]\+=//' `
$RUNAS $MCREATE $testfile2
$RUNAS setfattr -n lustre.lov -v $value $testfile2
local stripe_size=$($RUNAS $LFS getstripe -S $testfile2)
local stripe_count=$($RUNAS $LFS getstripe -c $testfile2)
[ $stripe_size -eq 65536 ] || error "stripe size $stripe_size != 65536"
[ $stripe_count -eq $STRIPECOUNT ] ||
error "stripe count $stripe_count != $STRIPECOUNT"
}
run_test 102c "non-root getfattr/setfattr for lustre.lov EAs ==========="
compare_stripe_info1() {
local stripe_index_all_zero=true
for num in 1 2 3 4; do
for count in $(seq 1 $STRIPE_COUNT); do
for offset in $(seq 0 $[$STRIPE_COUNT - 1]); do
local size=$((STRIPE_SIZE * num))
local file=file"$num-$offset-$count"
stripe_size=$($LFS getstripe -S $PWD/$file)
[[ $stripe_size -ne $size ]] &&
error "$file: size $stripe_size != $size"
stripe_count=$($LFS getstripe -c $PWD/$file)
# allow fewer stripes to be created, ORI-601
[[ $stripe_count -lt $(((3 * count + 3) / 4)) ]] &&
error "$file: count $stripe_count != $count"
stripe_index=$($LFS getstripe -i $PWD/$file)
[[ $stripe_index -ne 0 ]] &&
stripe_index_all_zero=false
done
done
done
$stripe_index_all_zero &&
error "all files are being extracted starting from OST index 0"
return 0
}
have_xattrs_include() {
tar --help | grep -q xattrs-include &&
echo --xattrs-include="lustre.*"
}
test_102d() {
[ $PARALLEL == "yes" ] && skip "skip parallel run"
[[ $OSTCOUNT -lt 2 ]] && skip_env "needs >= 2 OSTs"
XINC=$(have_xattrs_include)
setup_test102
tar xf $TMP/f102.tar -C $DIR/$tdir --xattrs $XINC
cd $DIR/$tdir/$tdir
compare_stripe_info1
}
run_test 102d "tar restore stripe info from tarfile,not keep osts"
test_102f() {
[ $PARALLEL == "yes" ] && skip "skip parallel run"
[[ $OSTCOUNT -lt 2 ]] && skip_env "needs >= 2 OSTs"
XINC=$(have_xattrs_include)
setup_test102
test_mkdir $DIR/$tdir.restore
cd $DIR
tar cf - --xattrs $tdir | tar xf - \
-C $DIR/$tdir.restore --xattrs $XINC
cd $DIR/$tdir.restore/$tdir
compare_stripe_info1
}
run_test 102f "tar copy files, not keep osts"
grow_xattr() {
[ -z "$(lctl get_param -n mdc.*.connect_flags | grep xattr)" ] &&
skip "must have user_xattr"
[ -z "$(which setfattr 2>/dev/null)" ] &&
skip_env "could not find setfattr"
[ -z "$(which getfattr 2>/dev/null)" ] &&
skip_env "could not find getfattr"
local xsize=${1:-1024} # in bytes
local file=$DIR/$tfile
local value="$(generate_string $xsize)"
local xbig=trusted.big
local toobig=$2
touch $file
log "save $xbig on $file"
if [ -z "$toobig" ]
then
setfattr -n $xbig -v $value $file ||
error "saving $xbig on $file failed"
else
setfattr -n $xbig -v $value $file &&
error "saving $xbig on $file succeeded"
return 0
fi
local orig=$(get_xattr_value $xbig $file)
[[ "$orig" != "$value" ]] && error "$xbig different after saving $xbig"
local xsml=trusted.sml
log "save $xsml on $file"
setfattr -n $xsml -v val $file || error "saving $xsml on $file failed"
local new=$(get_xattr_value $xbig $file)
[[ "$new" != "$orig" ]] && error "$xbig different after saving $xsml"
log "grow $xsml on $file"
setfattr -n $xsml -v "$value" $file ||
error "growing $xsml on $file failed"
new=$(get_xattr_value $xbig $file)
[[ "$new" != "$orig" ]] && error "$xbig different after growing $xsml"
log "$xbig still valid after growing $xsml"
rm -f $file
}
test_102h() { # bug 15777
grow_xattr 1024
}
run_test 102h "grow xattr from inside inode to external block"
test_102ha() {
large_xattr_enabled || skip_env "ea_inode feature disabled"
echo "setting xattr of max xattr size: $(max_xattr_size)"
grow_xattr $(max_xattr_size)
echo "setting xattr of > max xattr size: $(max_xattr_size) + 10"
echo "This should fail:"
grow_xattr $(($(max_xattr_size) + 10)) 1
}
run_test 102ha "grow xattr from inside inode to external inode"
test_102i() { # bug 17038
[ -z "$(which getfattr 2>/dev/null)" ] &&
skip "could not find getfattr"
touch $DIR/$tfile
ln -s $DIR/$tfile $DIR/${tfile}link
getfattr -n trusted.lov $DIR/$tfile ||
error "lgetxattr on $DIR/$tfile failed"
getfattr -h -n trusted.lov $DIR/${tfile}link 2>&1 |
grep -i "no such attr" ||
error "error for lgetxattr on $DIR/${tfile}link is not ENODATA"
rm -f $DIR/$tfile $DIR/${tfile}link
}
run_test 102i "lgetxattr test on symbolic link ============"
test_102j() {
[ $PARALLEL == "yes" ] && skip "skip parallel run"
[[ $OSTCOUNT -lt 2 ]] && skip_env "needs >= 2 OSTs"
XINC=$(have_xattrs_include)
setup_test102 "$RUNAS"
chown $RUNAS_ID $DIR/$tdir
$RUNAS tar xf $TMP/f102.tar -C $DIR/$tdir --xattrs $XINC
cd $DIR/$tdir/$tdir
compare_stripe_info1 "$RUNAS"
}
run_test 102j "non-root tar restore stripe info from tarfile, not keep osts ==="
test_102k() {
[ -z "$(which setfattr 2>/dev/null)" ] &&
skip "could not find setfattr"
touch $DIR/$tfile
# b22187 just check that does not crash for regular file.
setfattr -n trusted.lov $DIR/$tfile
# b22187 'setfattr -n trusted.lov' should remove LOV EA for directories
local test_kdir=$DIR/$tdir
test_mkdir $test_kdir
local default_size=$($LFS getstripe -S $test_kdir)
local default_count=$($LFS getstripe -c $test_kdir)
local default_offset=$($LFS getstripe -i $test_kdir)
$LFS setstripe -S 65536 -i 0 -c $OSTCOUNT $test_kdir ||
error 'dir setstripe failed'
setfattr -n trusted.lov $test_kdir
local stripe_size=$($LFS getstripe -S $test_kdir)
local stripe_count=$($LFS getstripe -c $test_kdir)
local stripe_offset=$($LFS getstripe -i $test_kdir)
[ $stripe_size -eq $default_size ] ||
error "stripe size $stripe_size != $default_size"
[ $stripe_count -eq $default_count ] ||
error "stripe count $stripe_count != $default_count"
[ $stripe_offset -eq $default_offset ] ||
error "stripe offset $stripe_offset != $default_offset"
rm -rf $DIR/$tfile $test_kdir
}
run_test 102k "setfattr without parameter of value shouldn't cause a crash"
test_102l() {
[ -z "$(which getfattr 2>/dev/null)" ] &&
skip "could not find getfattr"
# LU-532 trusted. xattr is invisible to non-root
local testfile=$DIR/$tfile
touch $testfile
echo "listxattr as user..."
chown $RUNAS_ID $testfile
$RUNAS getfattr -d -m '.*' $testfile 2>&1 |
grep -q "trusted" &&
error "$testfile trusted xattrs are user visible"
return 0;
}
run_test 102l "listxattr size test =================================="
test_102m() { # LU-3403 llite: error of listxattr when buffer is small
local path=$DIR/$tfile
touch $path
listxattr_size_check $path || error "listattr_size_check $path failed"
}
run_test 102m "Ensure listxattr fails on small bufffer ========"
cleanup_test102
getxattr() { # getxattr path name
# Return the base64 encoding of the value of xattr name on path.
local path=$1
local name=$2
# # getfattr --absolute-names --encoding=base64 --name=trusted.lov $path
# file: $path
# trusted.lov=0s0AvRCwEAAAAGAAAAAAAAAAAEAAACAAAAAAAQAAEAA...AAAAAAAAA=
#
# We print just 0s0AvRCwEAAAAGAAAAAAAAAAAEAAACAAAAAAAQAAEAA...AAAAAAAAA=
getfattr --absolute-names --encoding=base64 --name=$name $path |
awk -F= -v name=$name '$1 == name {
print substr($0, index($0, "=") + 1);
}'
}
test_102n() { # LU-4101 mdt: protect internal xattrs
[ -z "$(which setfattr 2>/dev/null)" ] &&
skip "could not find setfattr"
if [ $MDS1_VERSION -lt $(version_code 2.5.50) ]
then
skip "MDT < 2.5.50 allows setxattr on internal trusted xattrs"
fi
local file0=$DIR/$tfile.0
local file1=$DIR/$tfile.1
local xattr0=$TMP/$tfile.0
local xattr1=$TMP/$tfile.1
local namelist="lov lma lmv link fid version som hsm"
local name
local value
rm -rf $file0 $file1 $xattr0 $xattr1
touch $file0 $file1
# Get 'before' xattrs of $file1.
getfattr --absolute-names --dump --match=- $file1 > $xattr0
[ $MDS1_VERSION -lt $(version_code 2.8.53) ] &&
namelist+=" lfsck_namespace"
for name in $namelist; do
# Try to copy xattr from $file0 to $file1.
value=$(getxattr $file0 trusted.$name 2> /dev/null)
setfattr --name=trusted.$name --value="$value" $file1 ||
error "setxattr 'trusted.$name' failed"
# Try to set a garbage xattr.
value=0sVGhlIHF1aWNrIGJyb3duIGZveCBqdW1wcyBvdmVyIGl0c2VsZi4=
if [[ x$name == "xlov" ]]; then
setfattr --name=trusted.lov --value="$value" $file1 &&
error "setxattr invalid 'trusted.lov' success"
else
setfattr --name=trusted.$name --value="$value" $file1 ||
error "setxattr invalid 'trusted.$name' failed"
fi
# Try to remove the xattr from $file1. We don't care if this
# appears to succeed or fail, we just don't want there to be
# any changes or crashes.
setfattr --remove=$trusted.$name $file1 2> /dev/null
done
if [ $MDS1_VERSION -gt $(version_code 2.6.50) ]
then
name="lfsck_ns"
# Try to copy xattr from $file0 to $file1.
value=$(getxattr $file0 trusted.$name 2> /dev/null)
setfattr --name=trusted.$name --value="$value" $file1 ||
error "setxattr 'trusted.$name' failed"
# Try to set a garbage xattr.
value=0sVGhlIHF1aWNrIGJyb3duIGZveCBqdW1wcyBvdmVyIGl0c2VsZi4=
setfattr --name=trusted.$name --value="$value" $file1 ||
error "setxattr 'trusted.$name' failed"
# Try to remove the xattr from $file1. We don't care if this
# appears to succeed or fail, we just don't want there to be
# any changes or crashes.
setfattr --remove=$trusted.$name $file1 2> /dev/null
fi
# Get 'after' xattrs of file1.
getfattr --absolute-names --dump --match=- $file1 > $xattr1
if ! diff $xattr0 $xattr1; then
error "before and after xattrs of '$file1' differ"
fi
rm -rf $file0 $file1 $xattr0 $xattr1
return 0
}
run_test 102n "silently ignore setxattr on internal trusted xattrs"
test_102p() { # LU-4703 setxattr did not check ownership
[ $MDS1_VERSION -lt $(version_code 2.5.56) ] &&
skip "MDS needs to be at least 2.5.56"
local testfile=$DIR/$tfile
touch $testfile
echo "setfacl as user..."
$RUNAS setfacl -m "u:$RUNAS_ID:rwx" $testfile
[ $? -ne 0 ] || error "setfacl by $RUNAS_ID was allowed on $testfile"
echo "setfattr as user..."
setfacl -m "u:$RUNAS_ID:---" $testfile
$RUNAS setfattr -x system.posix_acl_access $testfile
[ $? -ne 0 ] || error "setfattr by $RUNAS_ID was allowed on $testfile"
}
run_test 102p "check setxattr(2) correctly fails without permission"
test_102q() {
[ $MDS1_VERSION -lt $(version_code 2.6.92) ] &&
skip "MDS needs to be at least 2.6.92"
orphan_linkea_check $DIR/$tfile || error "orphan_linkea_check"
}
run_test 102q "flistxattr should not return trusted.link EAs for orphans"
test_102r() {
[ $MDS1_VERSION -lt $(version_code 2.6.93) ] &&
skip "MDS needs to be at least 2.6.93"
touch $DIR/$tfile || error "touch"
setfattr -n user.$(basename $tfile) $DIR/$tfile || error "setfattr"
getfattr -n user.$(basename $tfile) $DIR/$tfile || error "getfattr"
rm $DIR/$tfile || error "rm"
#normal directory
mkdir -p $DIR/$tdir || error "mkdir"
setfattr -n user.$(basename $tdir) $DIR/$tdir || error "setfattr dir"
getfattr -n user.$(basename $tdir) $DIR/$tdir || error "getfattr dir"
setfattr -x user.$(basename $tdir) $DIR/$tdir ||
error "$testfile error deleting user.author1"
getfattr -d -m user.$(basename $tdir) 2> /dev/null |
grep "user.$(basename $tdir)" &&
error "$tdir did not delete user.$(basename $tdir)"
rmdir $DIR/$tdir || error "rmdir"
#striped directory
test_mkdir $DIR/$tdir
setfattr -n user.$(basename $tdir) $DIR/$tdir || error "setfattr dir"
getfattr -n user.$(basename $tdir) $DIR/$tdir || error "getfattr dir"
setfattr -x user.$(basename $tdir) $DIR/$tdir ||
error "$testfile error deleting user.author1"
getfattr -d -m user.$(basename $tdir) 2> /dev/null |
grep "user.$(basename $tdir)" &&
error "$tdir did not delete user.$(basename $tdir)"
rmdir $DIR/$tdir || error "rm striped dir"
}
run_test 102r "set EAs with empty values"
test_102s() {
[ $MDS1_VERSION -lt $(version_code 2.11.52) ] &&
skip "MDS needs to be at least 2.11.52"
local save="$TMP/$TESTSUITE-$TESTNAME.parameters"
save_lustre_params client "llite.*.xattr_cache" > $save
for cache in 0 1; do
lctl set_param llite.*.xattr_cache=$cache
rm -f $DIR/$tfile
touch $DIR/$tfile || error "touch"
for prefix in lustre security system trusted user; do
# Note getxattr() may fail with 'Operation not
# supported' or 'No such attribute' depending
# on prefix and cache.
getfattr -n $prefix.n102s $DIR/$tfile &&
error "getxattr '$prefix.n102s' should fail (cache = $cache)"
done
done
restore_lustre_params < $save
}
run_test 102s "getting nonexistent xattrs should fail"
test_102t() {
[ $MDS1_VERSION -lt $(version_code 2.11.52) ] &&
skip "MDS needs to be at least 2.11.52"
local save="$TMP/$TESTSUITE-$TESTNAME.parameters"
save_lustre_params client "llite.*.xattr_cache" > $save
for cache in 0 1; do
lctl set_param llite.*.xattr_cache=$cache
for buf_size in 0 256; do
rm -f $DIR/$tfile
touch $DIR/$tfile || error "touch"
setfattr -n user.multiop $DIR/$tfile
$MULTIOP $DIR/$tfile oa$buf_size ||
error "cannot get zero length xattr value (buf_size = $buf_size)"
done
done
restore_lustre_params < $save
}
run_test 102t "zero length xattr values handled correctly"
run_acl_subtest()
{
local test=$LUSTRE/tests/acl/$1.test
local tmp=$(mktemp -t $1-XXXXXX).test
local bin=$2
local dmn=$3
local grp=$4
local nbd=$5
export LANG=C
local sedusers="-e s/bin/$bin/g -e s/daemon/$dmn/g"
local sedgroups="-e s/:users/:$grp/g"
[[ -z "$nbd" ]] || sedusers+=" -e s/nobody/$nbd/g"
sed $sedusers $sedgroups < $test > $tmp
stack_trap "rm -f $tmp"
[[ -s $tmp ]] || error "sed failed to create test script"
echo "performing $1 with bin='$bin' daemon='$dmn' users='$grp'..."
$LUSTRE/tests/acl/run $tmp || error "run_acl_subtest '$1' failed"
}
test_103a() {
[ "$UID" != 0 ] && skip "must run as root"
$GSS && skip_env "could not run under gss"
[[ "$(lctl get_param -n mdc.*-mdc-*.connect_flags)" =~ "acl" ]] ||
skip_env "must have acl enabled"
which setfacl || skip_env "could not find setfacl"
remote_mds_nodsh && skip "remote MDS with nodsh"
local mdts=$(mdts_nodes)
local saved=$(do_facet mds1 $LCTL get_param -n mdt.$FSNAME-MDT0000.job_xattr)
[[ -z "$saved" ]] || do_nodes $mdts $LCTL set_param mdt.*.job_xattr=NONE
stack_trap "[[ -z \"$saved\" ]] || \
do_nodes $mdts $LCTL set_param mdt.*.job_xattr=$saved" EXIT
ACLBIN=${ACLBIN:-"bin"}
ACLDMN=${ACLDMN:-"daemon"}
ACLGRP=${ACLGRP:-"users"}
ACLNBD=${ACLNBD:-"nobody"}
if ! id $ACLBIN ||
[[ "$(id -u $ACLBIN)" != "$(do_facet mds1 id -u $ACLBIN)" ]]; then
echo "bad 'bin' user '$ACLBIN', using '$USER0'"
ACLBIN=$USER0
if ! id $ACLBIN ; then
cat /etc/passwd
skip_env "can't find suitable ACL 'bin' $ACLBIN"
fi
fi
if ! id $ACLDMN || (( $(id -u $ACLDMN) < $(id -u $ACLBIN) )) ||
[[ "$(id -u $ACLDMN)" != "$(do_facet mds1 id -u $ACLDMN)" ]]; then
echo "bad 'daemon' user '$ACLDMN', using '$USER1'"
ACLDMN=$USER1
if ! id $ACLDMN ; then
cat /etc/passwd
skip_env "can't find suitable ACL 'daemon' $ACLDMN"
fi
fi
if ! getent group $ACLGRP; then
echo "missing 'users' group '$ACLGRP', using '$TSTUSR'"
ACLGRP="$TSTUSR"
if ! getent group $ACLGRP; then
echo "cannot find group '$ACLGRP', adding it"
cat /etc/group
add_group 60000 $ACLGRP
fi
fi
local bingid=$(getent group $ACLBIN | cut -d: -f 3)
local dmngid=$(getent group $ACLDMN | cut -d: -f 3)
local grpgid=$(getent group $ACLGRP | cut -d: -f 3)
if (( $bingid > $grpgid || $dmngid > $grpgid )); then
echo "group '$ACLGRP' has low gid=$grpgid, use '$TSTUSR'"
ACLGRP="$TSTUSR"
if ! getent group $ACLGRP; then
echo "cannot find group '$ACLGRP', adding it"
cat /etc/group
add_group 60000 $ACLGRP
fi
grpgid=$(getent group $ACLGRP | cut -d: -f 3)
if (( $bingid > $grpgid || $dmngid > $grpgid )); then
cat /etc/group
skip_env "$ACLGRP gid=$grpgid less than $bingid|$dmngid"
fi
fi
id -nGz "$ACLDMN" | grep -qzxF "$ACLBIN" || gpasswd -a $ACLDMN $ACLBIN ||
error "setting client group failed" # LU-5641
if ! local_mode; then
do_facet mds1 "id -nGz $ACLDMN | grep -qzxF $ACLBIN ||
gpasswd -a $ACLDMN $ACLBIN" ||
error "setting MDS group failed" # LU-5641
fi
declare -a identity_old
for ((num = 1; num <= $MDSCOUNT; num++)); do
switch_identity $num true || identity_old[$num]=$?
done
SAVE_UMASK=$(umask)
umask 0022
$LFS mkdir -c1 -i0 $DIR/$tdir
cd $DIR/$tdir
run_acl_subtest cp $ACLBIN $ACLDMN $ACLGRP
run_acl_subtest getfacl-noacl $ACLBIN $ACLDMN $ACLGRP
run_acl_subtest misc $ACLBIN $ACLDMN $ACLGRP
run_acl_subtest permissions $ACLBIN $ACLDMN $ACLGRP
# LU-1482 mdd: Setting xattr are properly checked with and without ACLs
# CentOS7- uses nobody=99, while newer distros use nobody=65534
if ! id -u $ACLNBD ||
(( $(id -u nobody) != $(do_facet mds1 id -u nobody) )); then
ACLNBD="nfsnobody"
if ! id -u $ACLNBD; then
ACLNBD=""
fi
fi
if [[ -n "$ACLNBD" ]] && ! getent group $ACLNBD; then
add_group $(id -u $ACLNBD) $ACLNBD
if ! getent group $ACLNBD; then
ACLNBD=""
fi
fi
if (( $MDS1_VERSION > $(version_code 2.8.55) )) &&
[[ -n "$ACLNBD" ]] && which setfattr; then
run_acl_subtest permissions_xattr \
$ACLBIN $ACLDMN $ACLGRP $ACLNBD
elif [[ -z "$ACLNBD" ]]; then
echo "skip 'permission_xattr' test - missing 'nobody' user/grp"
else
echo "skip 'permission_xattr' test - missing setfattr command"
fi
run_acl_subtest setfacl $ACLBIN $ACLDMN $ACLGRP
# inheritance test got from HP
cp $LUSTRE/tests/acl/make-tree . || error "cannot copy make-tree"
chmod +x make-tree || error "chmod +x failed"
run_acl_subtest inheritance $ACLBIN $ACLDMN $ACLGRP
rm -f make-tree
echo "LU-974 ignore umask when acl is enabled..."
run_acl_subtest 974 $ACLBIN $ACLDMN $ACLGRP
if [ $MDSCOUNT -ge 2 ]; then
run_acl_subtest 974_remote $ACLBIN $ACLDMN $ACLGRP
fi
echo "LU-2561 newly created file is same size as directory..."
if [ "$mds1_FSTYPE" != "zfs" ]; then
run_acl_subtest 2561 $ACLBIN $ACLDMN $ACLGRP
else
run_acl_subtest 2561_zfs $ACLBIN $ACLDMN $ACLGRP
fi
run_acl_subtest 4924 $ACLBIN $ACLDMN $ACLGRP
cd $SAVE_PWD
umask $SAVE_UMASK
for ((num = 1; num <= $MDSCOUNT; num++)); do
if [[ "${identity_old[$num]}" == 1 ]]; then
switch_identity $num false || identity_old[$num]=$?
fi
done
}
run_test 103a "acl test"
test_103b() {
declare -a pids
local U
stack_trap "rm -f $DIR/$tfile.*"
for U in {0..511}; do
{
local O=$(printf "%04o" $U)
umask $(printf "%04o" $((511 ^ $O)))
$LFS setstripe -c 1 $DIR/$tfile.s$O
local S=$(printf "%04o" 0$(stat -c%a $DIR/$tfile.s$O))
(( $S == ($O & 0666) )) ||
error "lfs setstripe $DIR/$tfile.s$O '$S' != '$O'"
$LFS setstripe -E16M -c 1 -E1G -S4M $DIR/$tfile.p$O
S=$(printf "%04o" 0$(stat -c%a $DIR/$tfile.p$O))
(( $S == ($O & 0666) )) ||
error "lfs setstripe -E $DIR/$tfile.p$O '$S' != '$O'"
$LFS setstripe -N2 -c 1 $DIR/$tfile.m$O
S=$(printf "%04o" 0$(stat -c%a $DIR/$tfile.m$O))
(( $S == ($O & 0666) )) ||
error "lfs setstripe -N2 $DIR/$tfile.m$O '$S' != '$O'"
rm -f $DIR/$tfile.[smp]$0
} &
local pid=$!
# limit the concurrently running threads to 64. LU-11878
local idx=$((U % 64))
[ -z "${pids[idx]}" ] || wait ${pids[idx]}
pids[idx]=$pid
done
wait
}
run_test 103b "umask lfs setstripe"
test_103c() {
mkdir -p $DIR/$tdir
cp -rp $DIR/$tdir $DIR/$tdir.bak
[ -n "$(getfattr -d -m. $DIR/$tdir | grep posix_acl_default)" ] &&
error "$DIR/$tdir shouldn't contain default ACL"
[ -n "$(getfattr -d -m. $DIR/$tdir.bak | grep posix_acl_default)" ] &&
error "$DIR/$tdir.bak shouldn't contain default ACL"
true
}
run_test 103c "'cp -rp' won't set empty acl"
test_103e() {
local numacl
local fileacl
local saved_debug=$($LCTL get_param -n debug)
(( $MDS1_VERSION >= $(version_code v2_14_50-1-gf3d03bc38a) )) ||
skip "MDS needs to be at least 2.14.50.1"
large_xattr_enabled || skip_env "ea_inode feature disabled"
mkdir -p $DIR/$tdir
# add big LOV EA to cause reply buffer overflow earlier
$LFS setstripe -C 1000 $DIR/$tdir
lctl set_param mdc.*-mdc*.stats=clear
$LCTL set_param debug=0
stack_trap "$LCTL set_param debug=\"$saved_debug\"" EXIT
stack_trap "$LCTL get_param mdc.*-mdc*.stats" EXIT
# add large number of default ACLs in $incr batches (8000+ for 2.13+)
local incr=100
for ((range=0; range < 7000/incr; range++)); do
for ((U = range * incr + 2; U < (range+1) * incr; U++)); do
echo "user:$U:rwx"
done | setfacl -d -M /dev/stdin $DIR/$tdir ||
error "Cannot add $U default ACLs"
done
numacl=$(getfacl $DIR/$tdir |& grep -c "default:user")
echo "$numacl default ACLs created"
stat $DIR/$tdir || error "Cannot stat directory"
# check file creation
touch $DIR/$tdir/$tfile ||
error "failed to create $tfile with $numacl default ACLs"
stat $DIR/$tdir/$tfile || error "Cannot stat file"
fileacl=$(getfacl $DIR/$tdir/$tfile |& grep -c "user:")
echo "$fileacl ACLs were inherited"
(( $fileacl == $numacl )) ||
error "Not all default ACLs were inherited: $numacl != $fileacl"
# check that new ACLs creation adds new ACLs to inherited ACLs
setfacl -m user:19000:rwx $DIR/$tdir/$tfile ||
error "Cannot set new ACL"
numacl=$((numacl + 1))
fileacl=$(getfacl $DIR/$tdir/$tfile |& grep -c "user:")
(( $fileacl == $numacl )) ||
error "failed to add new ACL: $fileacl != $numacl as expected"
# add more ACLs in $incr batches to reach their maximum at 8000+
numacl=0
for ((range = 20000/incr; range < 25/incr; range++)); do
for ((U = range * incr; U < (range+1) * incr; U++)); do
echo "user:$U:rwx"
done | setfacl -M /dev/stdin $DIR/$tdir/$tfile || break
numacl=$((numacl + incr))
done
echo "Added $numacl more ACLs to the file"
fileacl=$(getfacl $DIR/$tdir/$tfile |& grep -c "user:")
echo "Total $fileacl ACLs in file"
stat $DIR/$tdir/$tfile > /dev/null || error "Cannot stat file"
rm -f $DIR/$tdir/$tfile || error "Cannot remove file"
rmdir $DIR/$tdir || error "Cannot remove directory"
}
run_test 103e "inheritance of big amount of default ACLs"
test_103f() {
(( $MDS1_VERSION >= $(version_code 2.14.51) )) ||
skip "MDS needs to be at least 2.14.51"
large_xattr_enabled || skip_env "ea_inode feature disabled"
# enable changelog to consume more internal MDD buffers
changelog_register
mkdir -p $DIR/$tdir
# add big LOV EA
$LFS setstripe -C 1000 $DIR/$tdir
setfacl -d -m user:$U:rwx $DIR/$tdir || error "Cannot add default ACLs"
mkdir $DIR/$tdir/inherited || error "failed to create subdirectory"
rmdir $DIR/$tdir/inherited || error "Cannot remove subdirectory"
rmdir $DIR/$tdir || error "Cannot remove directory"
}
run_test 103f "changelog doesn't interfere with default ACLs buffers"
test_103g() {
which getfacl || skip "missing getfacl"
[[ "$(lctl get_param -n mdc.*-mdc-*.connect_flags)" =~ "acl" ]] ||
skip_env "must have acl enabled"
local mdc_stat_param="mdc.$FSNAME-MDT*.md_stats"
local count
local f=$DIR/$tfile
touch $f
chmod 755 $f
cancel_lru_locks mdc
$RUNAS test -r $DIR
clear_stats $mdc_stat_param ||
error "fail to clear mdc stats"
# access(2) as a non-owner triggers acl_permission_check ->
# check_acl -> get_inode_acl -> ll_get_acl_common
$RUNAS test -r $f || error "test -r $f as $RUNAS_ID failed"
count=$(calc_stats $mdc_stat_param "getxattr")
echo "getxattr ACL RPCs: $count"
# access should not send ACL getxattr (ACLs are returned by lock intent)
(( !count )) ||
error "client sent $count getxattr RPCs"
# negative cache must be invalidated when an ACL is installed
setfacl -m u:$RUNAS_ID:rwx $f || error "setfacl $f failed"
$RUNAS getfacl -n $f 2>/dev/null | grep -q "^user:$RUNAS_ID:rwx$" ||
error "ACL not visible after setfacl (negative cache stuck)"
}
run_test 103g "no MDS_GETXATTR storm for inodes without ACL_ACCESS (LU-17238)"
test_104a() {
[ $PARALLEL == "yes" ] && skip "skip parallel run"
touch $DIR/$tfile
lfs df || error "lfs df failed"
lfs df -ih || error "lfs df -ih failed"
lfs df -h $DIR || error "lfs df -h $DIR failed"
lfs df -i $DIR || error "lfs df -i $DIR failed"
lfs df $DIR/$tfile || error "lfs df $DIR/$tfile failed"
lfs df -ih $DIR/$tfile || error "lfs df -ih $DIR/$tfile failed"
local OSC=$(lctl dl | grep OST0000-osc-[^M] | awk '{ print $4 }')
lctl --device %$OSC deactivate
lfs df || error "lfs df with deactivated OSC failed"
lctl --device %$OSC activate
# wait the osc back to normal
wait_osc_import_ready client ost
lfs df || error "lfs df with reactivated OSC failed"
rm -f $DIR/$tfile
}
run_test 104a "lfs df [-ih] [path] test ========================="
test_104b() {
[ $PARALLEL == "yes" ] && skip "skip parallel run"
[ $RUNAS_ID -eq $UID ] &&
skip_env "RUNAS_ID = UID = $UID -- skipping"
denied_cnt=$(($($RUNAS $LFS check servers 2>&1 |
grep "Permission denied" | wc -l)))
if [ $denied_cnt -ne 0 ]; then
error "lfs check servers test failed"
fi
}
run_test 104b "$RUNAS lfs check servers test ===================="
#
# Verify $1 is within range of $2.
# Success when $1 is within range. That is, when $1 is >= 2% of $2 and
# $1 is <= 2% of $2. Else Fail.
#
value_in_range() {
# Strip all units (M, G, T)
actual=$(echo $1 | tr -d A-Z)
expect=$(echo $2 | tr -d A-Z)
expect_lo=$(($expect * 98 / 100)) # 2% below
expect_hi=$(($expect * 102 / 100)) # 2% above
# permit 2% drift above and below
(( $actual >= $expect_lo && $actual <= $expect_hi ))
}
test_104c() {
[ $PARALLEL == "yes" ] && skip "skip parallel run"
[ "$ost1_FSTYPE" == "zfs" ] || skip "zfs only test"
local ost_param="osd-zfs.$FSNAME-OST0000."
local mdt_param="osd-zfs.$FSNAME-MDT0000."
local ofacets=$(get_facets OST)
local mfacets=$(get_facets MDS)
local saved_ost_blocks=
local saved_mdt_blocks=
echo "Before recordsize change"
lfs_df=($($LFS df -h | grep "filesystem_summary:"))
df=($(df -h | grep "$MOUNT"$))
# For checking.
echo "lfs output : ${lfs_df[*]}"
echo "df output : ${df[*]}"
for facet in ${ofacets//,/ }; do
if [ -z $saved_ost_blocks ]; then
saved_ost_blocks=$(do_facet $facet \
lctl get_param -n $ost_param.blocksize)
echo "OST Blocksize: $saved_ost_blocks"
fi
ost=$(do_facet $facet lctl get_param -n $ost_param.mntdev)
do_facet $facet zfs set recordsize=32768 $ost
done
# BS too small. Sufficient for functional testing.
for facet in ${mfacets//,/ }; do
if [ -z $saved_mdt_blocks ]; then
saved_mdt_blocks=$(do_facet $facet \
lctl get_param -n $mdt_param.blocksize)
echo "MDT Blocksize: $saved_mdt_blocks"
fi
mdt=$(do_facet $facet lctl get_param -n $mdt_param.mntdev)
do_facet $facet zfs set recordsize=32768 $mdt
done
# Give new values chance to reflect change
sleep 2
echo "After recordsize change"
lfs_df_after=($($LFS df -h | grep "filesystem_summary:"))
df_after=($(df -h | grep "$MOUNT"$))
# For checking.
echo "lfs output : ${lfs_df_after[*]}"
echo "df output : ${df_after[*]}"
# Verify lfs df
value_in_range ${lfs_df_after[1]%.*} ${lfs_df[1]%.*} ||
error "lfs_df bytes: ${lfs_df_after[1]%.*} != ${lfs_df[1]%.*}"
value_in_range ${lfs_df_after[2]%.*} ${lfs_df[2]%.*} ||
error "lfs_df used: ${lfs_df_after[2]%.*} != ${lfs_df[2]%.*}"
value_in_range ${lfs_df_after[3]%.*} ${lfs_df[3]%.*} ||
error "lfs_df avail: ${lfs_df_after[3]%.*} != ${lfs_df[3]%.*}"
# Verify df
value_in_range ${df_after[1]%.*} ${df[1]%.*} ||
error "df bytes: ${df_after[1]%.*} != ${df[1]%.*}"
value_in_range ${df_after[2]%.*} ${df[2]%.*} ||
error "df used: ${df_after[2]%.*} != ${df[2]%.*}"
value_in_range ${df_after[3]%.*} ${df[3]%.*} ||
error "df avail: ${df_after[3]%.*} != ${df[3]%.*}"
# Restore MDT recordize back to original
for facet in ${mfacets//,/ }; do
mdt=$(do_facet $facet lctl get_param -n $mdt_param.mntdev)
do_facet $facet zfs set recordsize=$saved_mdt_blocks $mdt
done
# Restore OST recordize back to original
for facet in ${ofacets//,/ }; do
ost=$(do_facet $facet lctl get_param -n $ost_param.mntdev)
do_facet $facet zfs set recordsize=$saved_ost_blocks $ost
done
return 0
}
run_test 104c "Verify df vs lfs_df stays same after recordsize change"
test_104d() {
(( $RUNAS_ID != $UID )) ||
skip_env "RUNAS_ID = UID = $UID -- skipping"
(( $CLIENT_VERSION >= $(version_code 2.15.51) )) ||
skip "lustre version doesn't support lctl dl with non-root"
# debugfs only allows root users to access files, so the
# previous move of the "devices" file to debugfs broke
# "lctl dl" for non-root users. The LU-9680 Netlink
# interface again allows non-root users to list devices.
[ "$($RUNAS $LCTL dl | wc -l)" -ge 3 ] ||
error "lctl dl doesn't work for non root"
echo "=="
$RUNAS $LCTL dl | grep "$FSNAME-OST*"
echo "=="
$RUNAS $LCTL dl | grep "obdfilter $FSNAME-OST*"
echo "=="
ost_count="$($RUNAS $LCTL dl | grep "osc $FSNAME-OST*" | wc -l)"
[ "$ost_count" -eq $OSTCOUNT ] ||
error "lctl dl reports wrong number of OST devices"
$RUNAS $LCTL dl | grep "mdt $FSNAME-MDT*"
mdt_count="$($RUNAS $LCTL dl | grep "mdc $FSNAME-MDT*" | wc -l)"
[ "$mdt_count" -eq $MDSCOUNT ] ||
error "lctl dl reports wrong number of MDT devices"
}
run_test 104d "$RUNAS lctl dl test"
test_105a() {
# doesn't work on 2.4 kernels
touch $DIR/$tfile
if $(flock_is_enabled); then
flocks_test 1 on -f $DIR/$tfile || error "fail flock on"
else
flocks_test 1 off -f $DIR/$tfile || error "fail flock off"
fi
rm -f $DIR/$tfile
}
run_test 105a "flock when mounted without -o flock test ========"
test_105b() {
touch $DIR/$tfile
if $(flock_is_enabled); then
flocks_test 1 on -c $DIR/$tfile || error "fail flock on"
else
flocks_test 1 off -c $DIR/$tfile || error "fail flock off"
fi
rm -f $DIR/$tfile
}
run_test 105b "fcntl when mounted without -o flock test ========"
test_105c() {
touch $DIR/$tfile
if $(flock_is_enabled); then
flocks_test 1 on -l $DIR/$tfile || error "fail flock on"
else
flocks_test 1 off -l $DIR/$tfile || error "fail flock off"
fi
rm -f $DIR/$tfile
}
run_test 105c "lockf when mounted without -o flock test"
test_105d() { # bug 15924
[ $PARALLEL == "yes" ] && skip "skip parallel run"
test_mkdir $DIR/$tdir
flock_is_enabled || skip_env "mount w/o flock enabled"
#define OBD_FAIL_LDLM_CP_CB_WAIT 0x315
$LCTL set_param fail_loc=0x80000315
flocks_test 2 $DIR/$tdir
}
run_test 105d "flock race (should not freeze) ========"
test_105e() { # bug 22660 && 22040
flock_is_enabled || skip_env "mount w/o flock enabled"
touch $DIR/$tfile
flocks_test 3 $DIR/$tfile
}
run_test 105e "Two conflicting flocks from same process"
wait_end() {
echo $*
while :; do
[ -f $TMP/${tfile}_sTOP ] && return
sleep 1
done
}
test_105f() {
flock_is_enabled || skip_env "mount w/o flock enabled"
local pmax=$(ulimit -u)
local i=0
touch $DIR/$tfile
[ $pmax -gt 20 ] && pmax=20
for((i=0; i <= $pmax; i++)) {
wait_end "R4000, 5000" | flocks_test 6 $DIR/$tfile &
}
for((i=0; i <= 10; i++)) {
local locks=$(do_facet $SINGLEMDS $LCTL get_param -n \
ldlm.namespaces.mdt-${FSNAME}-MDT0000*.lock_count)
[ $locks -ge $pmax ] && break
[ $i -eq 10 ] && error "The locks cannot be added after 10 secs"
sleep 1
}
touch $TMP/${tfile}_sTOP
wait
rm -r $DIR/$tfile $TMP/${tfile}_sTOP
}
run_test 105f "Enqueue same range flocks"
test_105g() {
(( $CLIENT_VERSION >= $(version_code 2.15.64.117) )) ||
skip "Need Client >= 2.15.63.127 for ldlm dump stack"
flock_is_enabled || skip_env "mount w/o flock enabled"
mkdir $DIR/$tdir
#define OBD_FAIL_LDLM_LOCK_STACK 0x32f
do_facet client $LCTL set_param -n \
ldlm.namespaces.*.dump_stack_on_error=1
stack_trap "do_facet client $LCTL set_param -n \
ldlm.namespaces.*.dump_stack_on_error=0"
$LCTL set_param fail_loc=0x8000032f
flocks_test 2 $DIR/$tdir
dmesg | tac | sed "/$(echo $TESTNAME | tr '_' ' ')/,$ d" |
grep -q "dump_stack" || error "didn't find dump_stack"
}
run_test 105g "ldlm_lock_debug stack test"
test_105h() {
local info
touch $DIR/$tfile
stack_trap "rm $DIR/$tfile" EXIT
# enqueue 2 non-overlap locks
info=$(echo -e "W20,40\nR100,200\nT0" | flocks_test 6 $DIR/$tfile)
[ x$info == 'xW20,40;R100,200.' ] || error "flock error1-$info"
# enqueue overlap locks
info=$(echo -e "W20,400\nR100,2000\nR10,1000\nW500,1000\nT0" | flocks_test 6 $DIR/$tfile)
[ x$info == 'xR10,490;W500,1000;R1500,600.' ] || error "flock error2-$info"
# split a lock
info=$(echo -e "W2000,1000\nR2200, 200\nT0" |flocks_test 6 $DIR/$tfile)
[ x$info == 'xW2000,200;R2200,200;W2400,600.' ] || error "flock error3-$info"
# merge 2 locks
info=$(echo -e "R200,1000\nR9000,200\nW100,10000\nT0" | flocks_test 6 $DIR/$tfile)
[ x$info == 'xW100,10000.' ] || error "flock error4-$info"
# adjoining locks
info=$(echo -e "R200,100\nR300, 200\nR500,100\nT0" | flocks_test 6 $DIR/$tfile)
[ x$info == 'xR200,400.' ] || error "flock error5-$info"
# adjoining locks with diff mode
info=$(echo -e "R200,100\nW300, 200\nR500,100\nT0" | flocks_test 6 $DIR/$tfile)
[ x$info == 'xR200,100;W300,200;R500,100.' ] || error "flock error6-$info"
# split & merge
info=$(echo -e "R200,1000\nW300, 200\nR100,200\nR300,200\nT0" | flocks_test 6 $DIR/$tfile)
[ x$info == 'xR100,1100.' ] || error "flock error7-$info"
}
run_test 105h "Flock functional verify"
test_105i() {
local tmpfile=$TMP/tmpf.$$
touch $DIR/$tfile
stack_trap "rm $tmpfile $DIR/$tfile" EXIT
# enqueue 2 non-overlap locks
>$tmpfile
echo -e "R20,200\nP5\nW10,1000" | flocks_test 6 $DIR/$tfile 2>>$tmpfile &
echo -e "R20,200\nP5\nW10,1000" | flocks_test 6 $DIR/$tfile 2>>$tmpfile &
wait
# Check for EDEADLK
grep 'cannot set lock(35)' $tmpfile
}
run_test 105i "Flock deadlock verify"
test_106() { #bug 10921
test_mkdir $DIR/$tdir
$DIR/$tdir && error "exec $DIR/$tdir succeeded"
chmod 777 $DIR/$tdir || error "chmod $DIR/$tdir failed"
}
run_test 106 "attempt exec of dir followed by chown of that dir"
test_107() {
[ $PARALLEL == "yes" ] && skip "skip parallel run"
CDIR=`pwd`
local file=core
cd $DIR
rm -f $file
local save_pattern=$(sysctl -n kernel.core_pattern)
local save_uses_pid=$(sysctl -n kernel.core_uses_pid)
sysctl -w kernel.core_pattern=$file
sysctl -w kernel.core_uses_pid=0
ulimit -c unlimited
sleep 60 &
SLEEPPID=$!
sleep 1
kill -s 11 $SLEEPPID
wait $SLEEPPID
if [ -e $file ]; then
size=`stat -c%s $file`
[ $size -eq 0 ] && error "Fail to create core file $file"
else
error "Fail to create core file $file"
fi
rm -f $file
sysctl -w kernel.core_pattern=$save_pattern
sysctl -w kernel.core_uses_pid=$save_uses_pid
cd $CDIR
}
run_test 107 "Coredump on SIG"
test_110() {
test_mkdir $DIR/$tdir
test_mkdir $DIR/$tdir/$(str_repeat 'a' 255)
$LFS mkdir -c $MDSCOUNT $DIR/$tdir/$(str_repeat 'b' 256) &&
error "mkdir with 256 char should fail, but did not"
touch $DIR/$tdir/$(str_repeat 'x' 255) ||
error "create with 255 char failed"
touch $DIR/$tdir/$(str_repeat 'y' 256) &&
error "create with 256 char should fail, but did not"
ls -l $DIR/$tdir
rm -rf $DIR/$tdir
}
run_test 110 "filename length checking"
test_116a() { # was previously test_116()
[ $PARALLEL == "yes" ] && skip "skip parallel run"
[[ $OSTCOUNT -lt 2 ]] && skip_env "needs >= 2 OSTs"
remote_mds_nodsh && skip "remote MDS with nodsh"
echo -n "Free space priority "
do_facet $SINGLEMDS lctl get_param -n lo[vd].*-mdtlov.qos_prio_free |
head -n1
declare -a AVAIL
free_min_max
[ $MINV -eq 0 ] && skip "no free space in OST$MINI, skip"
[ $MINV -gt 10000000 ] && skip "too much free space in OST$MINI, skip"
stack_trap simple_cleanup_common
# Check if we need to generate uneven OSTs
test_mkdir -p $DIR/$tdir/OST${MINI}
local FILL=$((MINV / 4))
local DIFF=$((MAXV - MINV))
local DIFF2=$((DIFF * 100 / MINV))
local threshold=$(do_facet $SINGLEMDS \
lctl get_param -n *.*MDT0000-mdtlov.qos_threshold_rr | head -n1)
threshold=${threshold%%%}
echo -n "Check for uneven OSTs: "
echo -n "diff=${DIFF}KB (${DIFF2}%) must be > ${threshold}% ..."
if [[ $DIFF2 -gt $threshold ]]; then
echo "ok"
echo "Don't need to fill OST$MINI"
else
# generate uneven OSTs. Write 2% over the QOS threshold value
echo "no"
DIFF=$((threshold - DIFF2 + 2))
DIFF2=$((MINV * DIFF / 100))
echo "Fill $DIFF% remaining space in OST$MINI with ${DIFF2}KB"
$LFS setstripe -i $MINI -c 1 $DIR/$tdir/OST${MINI} ||
error "setstripe failed"
DIFF=$((DIFF2 / 2048))
fast_file_write $DIR/$tdir/OST${MINI}/$tfile 2M $DIFF
sync
sleep_maxage
free_min_max
fi
DIFF=$((MAXV - MINV))
DIFF2=$((DIFF * 100 / MINV))
echo -n "diff=$DIFF=$DIFF2% must be > $threshold% for QOS mode..."
if [ $DIFF2 -gt $threshold ]; then
echo "ok"
else
skip "QOS imbalance criteria not met"
fi
MINI1=$MINI
MINV1=$MINV
MAXI1=$MAXI
MAXV1=$MAXV
# now fill using QOS
$LFS setstripe -c 1 $DIR/$tdir
FILL=$((FILL / 200))
if [ $FILL -gt 600 ]; then
FILL=600
fi
echo "writing $FILL files to QOS-assigned OSTs"
fast_file_write $DIR/$tdir/$tfile 200k $FILL
echo "wrote $FILL 200k files"
sync
sleep_maxage
echo "Note: free space may not be updated, so measurements might be off"
free_min_max
DIFF2=$((MAXV - MINV))
echo "free space delta: orig $DIFF final $DIFF2"
[ $DIFF2 -gt $DIFF ] && echo "delta got worse!"
DIFF=$((MINV1 - ${AVAIL[$MINI1]}))
echo "Wrote ${DIFF}KB to smaller OST $MINI1"
DIFF2=$((MAXV1 - ${AVAIL[$MAXI1]}))
echo "Wrote ${DIFF2}KB to larger OST $MAXI1"
if [[ $DIFF -gt 0 ]]; then
FILL=$((DIFF2 * 100 / DIFF - 100))
echo "Wrote ${FILL}% more data to larger OST $MAXI1"
fi
# Figure out which files were written where
UUID=$(lctl get_param -n lov.${FSNAME}-clilov-*.target_obd |
awk '/'$MINI1': / {print $2; exit}')
echo $UUID
MINC=$($LFS getstripe --ost $UUID $DIR/$tdir | grep $DIR | wc -l)
echo "$MINC files created on smaller OST $MINI1"
UUID=$(lctl get_param -n lov.${FSNAME}-clilov-*.target_obd |
awk '/'$MAXI1': / {print $2; exit}')
echo $UUID
MAXC=$($LFS getstripe --ost $UUID $DIR/$tdir | grep $DIR | wc -l)
echo "$MAXC files created on larger OST $MAXI1"
if [[ $MINC -gt 0 ]]; then
FILL=$((MAXC * 100 / MINC - 100))
echo "Wrote ${FILL}% more files to larger OST $MAXI1"
fi
[[ $MAXC -gt $MINC ]] ||
error_ignore LU-9 "stripe QOS didn't balance free space"
}
run_test 116a "stripe QOS: free space balance ==================="
test_116b() { # LU-2093
[ $PARALLEL == "yes" ] && skip "skip parallel run"
remote_mds_nodsh && skip "remote MDS with nodsh"
#define OBD_FAIL_MDS_OSC_CREATE_FAIL 0x147
local old_rr=$(do_facet $SINGLEMDS lctl get_param -n \
lo[vd].$FSNAME-MDT0000-mdtlov.qos_threshold_rr | head -1)
[ -z "$old_rr" ] && skip "no QOS"
do_facet $SINGLEMDS lctl set_param \
lo[vd].$FSNAME-MDT0000-mdtlov.qos_threshold_rr=0
mkdir -p $DIR/$tdir
do_facet $SINGLEMDS lctl set_param fail_loc=0x147
createmany -o $DIR/$tdir/f- 20 || error "can't create"
do_facet $SINGLEMDS lctl set_param fail_loc=0
rm -rf $DIR/$tdir
do_facet $SINGLEMDS lctl set_param \
lo[vd].$FSNAME-MDT0000-mdtlov.qos_threshold_rr=$old_rr
}
run_test 116b "QoS shouldn't LBUG if not enough OSTs found on the 2nd pass"
test_117() # bug 10891
{
[ $PARALLEL == "yes" ] && skip "skip parallel run"
dd if=/dev/zero of=$DIR/$tfile bs=1M count=1
#define OBD_FAIL_OST_SETATTR_CREDITS 0x21e
lctl set_param fail_loc=0x21e
> $DIR/$tfile || error "truncate failed"
lctl set_param fail_loc=0
echo "Truncate succeeded."
rm -f $DIR/$tfile
}
run_test 117 "verify osd extend =========="
NO_SLOW_RESENDCOUNT=4
export OLD_RESENDCOUNT=""
set_resend_count () {
local PROC_RESENDCOUNT="osc.${FSNAME}-OST*-osc-*.resend_count"
OLD_RESENDCOUNT=$(lctl get_param -n $PROC_RESENDCOUNT | head -n1)
lctl set_param -n $PROC_RESENDCOUNT $1
echo resend_count is set to $(lctl get_param -n $PROC_RESENDCOUNT)
}
# for reduce test_118* time (b=14842)
[ "$SLOW" = "no" ] && set_resend_count $NO_SLOW_RESENDCOUNT
# Reset async IO behavior after error case
reset_async() {
FILE=$DIR/reset_async
# Ensure all OSCs are cleared
$LFS setstripe -c -1 $FILE
dd if=/dev/zero of=$FILE bs=64k count=$OSTCOUNT
sync
rm $FILE
}
test_118a() #bug 11710
{
[ $PARALLEL == "yes" ] && skip "skip parallel run"
reset_async
$MULTIOP $DIR/$tfile oO_CREAT:O_RDWR:O_SYNC:w4096c
DIRTY=$(lctl get_param -n llite.*.dump_page_cache | grep -c dirty)
WRITEBACK=$(lctl get_param -n llite.*.dump_page_cache | grep -c writeback)
if [[ $DIRTY -ne 0 || $WRITEBACK -ne 0 ]]; then
error "Dirty pages not flushed to disk, dirty=$DIRTY, writeback=$WRITEBACK"
return 1;
fi
rm -f $DIR/$tfile
}
run_test 118a "verify O_SYNC works =========="
test_118b()
{
[ $PARALLEL == "yes" ] && skip "skip parallel run"
remote_ost_nodsh && skip "remote OST with nodsh"
local osts=$(osts_nodes)
reset_async
#define OBD_FAIL_SRV_ENOENT 0x217
set_nodes_failloc $osts 0x217
$MULTIOP $DIR/$tfile oO_CREAT:O_RDWR:O_SYNC:w4096c
RC=$?
set_nodes_failloc $osts 0
DIRTY=$(lctl get_param -n llite.*.dump_page_cache | grep -c dirty)
WRITEBACK=$(lctl get_param -n llite.*.dump_page_cache |
grep -c writeback)
if [[ $RC -eq 0 ]]; then
error "Must return error due to dropped pages, rc=$RC"
return 1;
fi
if [[ $DIRTY -ne 0 || $WRITEBACK -ne 0 ]]; then
error "Dirty pages not flushed to disk, dirty=$DIRTY, writeback=$WRITEBACK"
return 1;
fi
echo "Dirty pages not leaked on ENOENT"
# Due to the above error the OSC will issue all RPCs syncronously
# until a subsequent RPC completes successfully without error.
$MULTIOP $DIR/$tfile Ow4096yc
rm -f $DIR/$tfile
return 0
}
run_test 118b "Reclaim dirty pages on fatal error =========="
test_118c()
{
[ $PARALLEL == "yes" ] && skip "skip parallel run"
# for 118c, restore the original resend count, LU-1940
[ "$SLOW" = "no" ] && [ -n "$OLD_RESENDCOUNT" ] &&
set_resend_count $OLD_RESENDCOUNT
remote_ost_nodsh && skip "remote OST with nodsh"
local osts=$(osts_nodes)
reset_async
#define OBD_FAIL_OST_EROFS 0x216
set_nodes_failloc $osts 0x216
# multiop should block due to fsync until pages are written
$MULTIOP $DIR/$tfile oO_CREAT:O_RDWR:O_SYNC:w4096c &
MULTIPID=$!
sleep 1
if [[ `ps h -o comm -p $MULTIPID` != "multiop" ]]; then
error "Multiop failed to block on fsync, pid=$MULTIPID"
fi
WRITEBACK=$(lctl get_param -n llite.*.dump_page_cache |
grep -c writeback)
if [[ $WRITEBACK -eq 0 ]]; then
error "No page in writeback, writeback=$WRITEBACK"
fi
set_nodes_failloc $osts 0
wait $MULTIPID
RC=$?
if [[ $RC -ne 0 ]]; then
error "Multiop fsync failed, rc=$RC"
fi
DIRTY=$(lctl get_param -n llite.*.dump_page_cache | grep -c dirty)
WRITEBACK=$(lctl get_param -n llite.*.dump_page_cache |
grep -c writeback)
if [[ $DIRTY -ne 0 || $WRITEBACK -ne 0 ]]; then
error "Dirty pages not flushed to disk, dirty=$DIRTY, writeback=$WRITEBACK"
fi
rm -f $DIR/$tfile
echo "Dirty pages flushed via fsync on EROFS"
return 0
}
run_test 118c "Fsync blocks on EROFS until dirty pages are flushed =========="
# continue to use small resend count to reduce test_118* time (b=14842)
[ "$SLOW" = "no" ] && set_resend_count $NO_SLOW_RESENDCOUNT
test_118d()
{
[ $PARALLEL == "yes" ] && skip "skip parallel run"
remote_ost_nodsh && skip "remote OST with nodsh"
local osts=$(osts_nodes)
reset_async
#define OBD_FAIL_OST_BRW_PAUSE_BULK
set_nodes_failloc $osts 0x214
# multiop should block due to fsync until pages are written
$MULTIOP $DIR/$tfile oO_CREAT:O_RDWR:O_SYNC:w4096c &
MULTIPID=$!
sleep 1
if [[ `ps h -o comm -p $MULTIPID` != "multiop" ]]; then
error "Multiop failed to block on fsync, pid=$MULTIPID"
fi
WRITEBACK=$(lctl get_param -n llite.*.dump_page_cache |
grep -c writeback)
if [[ $WRITEBACK -eq 0 ]]; then
error "No page in writeback, writeback=$WRITEBACK"
fi
wait $MULTIPID || error "Multiop fsync failed, rc=$?"
set_nodes_failloc $osts 0
DIRTY=$(lctl get_param -n llite.*.dump_page_cache | grep -c dirty)
WRITEBACK=$(lctl get_param -n llite.*.dump_page_cache |
grep -c writeback)
if [[ $DIRTY -ne 0 || $WRITEBACK -ne 0 ]]; then
error "Dirty pages not flushed to disk, dirty=$DIRTY, writeback=$WRITEBACK"
fi
rm -f $DIR/$tfile
echo "Dirty pages gaurenteed flushed via fsync"
return 0
}
run_test 118d "Fsync validation inject a delay of the bulk =========="
test_118f() {
[ $PARALLEL == "yes" ] && skip "skip parallel run"
reset_async
#define OBD_FAIL_OSC_BRW_PREP_REQ2 0x40a
lctl set_param fail_loc=0x8000040a
# Should simulate EINVAL error which is fatal
$MULTIOP $DIR/$tfile oO_CREAT:O_RDWR:O_SYNC:w4096c
RC=$?
if [[ $RC -eq 0 ]]; then
error "Must return error due to dropped pages, rc=$RC"
fi
lctl set_param fail_loc=0x0
LOCKED=$(lctl get_param -n llite.*.dump_page_cache | grep -c locked)
DIRTY=$(lctl get_param -n llite.*.dump_page_cache | grep -c dirty)
WRITEBACK=$(lctl get_param -n llite.*.dump_page_cache |
grep -c writeback)
if [[ $LOCKED -ne 0 ]]; then
error "Locked pages remain in cache, locked=$LOCKED"
fi
if [[ $DIRTY -ne 0 || $WRITEBACK -ne 0 ]]; then
error "Dirty pages not flushed to disk, dirty=$DIRTY, writeback=$WRITEBACK"
fi
rm -f $DIR/$tfile
echo "No pages locked after fsync"
reset_async
return 0
}
run_test 118f "Simulate unrecoverable OSC side error =========="
test_118g() {
[ $PARALLEL == "yes" ] && skip "skip parallel run"
reset_async
#define OBD_FAIL_OSC_BRW_PREP_REQ 0x406
lctl set_param fail_loc=0x406
# simulate local -ENOMEM
$MULTIOP $DIR/$tfile oO_CREAT:O_RDWR:O_SYNC:w4096c
RC=$?
lctl set_param fail_loc=0
if [[ $RC -eq 0 ]]; then
error "Must return error due to dropped pages, rc=$RC"
fi
LOCKED=$(lctl get_param -n llite.*.dump_page_cache | grep -c locked)
DIRTY=$(lctl get_param -n llite.*.dump_page_cache | grep -c dirty)
WRITEBACK=$(lctl get_param -n llite.*.dump_page_cache |
grep -c writeback)
if [[ $LOCKED -ne 0 ]]; then
error "Locked pages remain in cache, locked=$LOCKED"
fi
if [[ $DIRTY -ne 0 || $WRITEBACK -ne 0 ]]; then
error "Dirty pages not flushed to disk, dirty=$DIRTY, writeback=$WRITEBACK"
fi
rm -f $DIR/$tfile
echo "No pages locked after fsync"
reset_async
return 0
}
run_test 118g "Don't stay in wait if we got local -ENOMEM =========="
test_118h() {
[ $PARALLEL == "yes" ] && skip "skip parallel run"
remote_ost_nodsh && skip "remote OST with nodsh"
local osts=$(osts_nodes)
reset_async
#define OBD_FAIL_OST_BRW_WRITE_BULK 0x20e
set_nodes_failloc $osts 0x20e
# Should simulate ENOMEM error which is recoverable and should be handled by timeout
$MULTIOP $DIR/$tfile oO_CREAT:O_RDWR:O_SYNC:w4096c
RC=$?
set_nodes_failloc $osts 0
if [[ $RC -eq 0 ]]; then
error "Must return error due to dropped pages, rc=$RC"
fi
LOCKED=$(lctl get_param -n llite.*.dump_page_cache | grep -c locked)
DIRTY=$(lctl get_param -n llite.*.dump_page_cache | grep -c dirty)
WRITEBACK=$(lctl get_param -n llite.*.dump_page_cache |
grep -c writeback)
if [[ $LOCKED -ne 0 ]]; then
error "Locked pages remain in cache, locked=$LOCKED"
fi
if [[ $DIRTY -ne 0 || $WRITEBACK -ne 0 ]]; then
error "Dirty pages not flushed to disk, dirty=$DIRTY, writeback=$WRITEBACK"
fi
rm -f $DIR/$tfile
echo "No pages locked after fsync"
return 0
}
run_test 118h "Verify timeout in handling recoverables errors =========="
[ "$SLOW" = "no" ] && [ -n "$OLD_RESENDCOUNT" ] && set_resend_count $OLD_RESENDCOUNT
test_118i() {
[ $PARALLEL == "yes" ] && skip "skip parallel run"
remote_ost_nodsh && skip "remote OST with nodsh"
local osts=$(osts_nodes)
reset_async
#define OBD_FAIL_OST_BRW_WRITE_BULK 0x20e
set_nodes_failloc $osts 0x20e
# Should simulate ENOMEM error which is recoverable and should be handled by timeout
$MULTIOP $DIR/$tfile oO_CREAT:O_RDWR:O_SYNC:w4096c &
PID=$!
sleep 5
set_nodes_failloc $osts 0
wait $PID
RC=$?
if [[ $RC -ne 0 ]]; then
error "got error, but should be not, rc=$RC"
fi
LOCKED=$(lctl get_param -n llite.*.dump_page_cache | grep -c locked)
DIRTY=$(lctl get_param -n llite.*.dump_page_cache | grep -c dirty)
WRITEBACK=$(lctl get_param -n llite.*.dump_page_cache | grep -c writeback)
if [[ $LOCKED -ne 0 ]]; then
error "Locked pages remain in cache, locked=$LOCKED"
fi
if [[ $DIRTY -ne 0 || $WRITEBACK -ne 0 ]]; then
error "Dirty pages not flushed to disk, dirty=$DIRTY, writeback=$WRITEBACK"
fi
rm -f $DIR/$tfile
echo "No pages locked after fsync"
return 0
}
run_test 118i "Fix error before timeout in recoverable error =========="
[ "$SLOW" = "no" ] && set_resend_count 4
test_118j() {
[ $PARALLEL == "yes" ] && skip "skip parallel run"
remote_ost_nodsh && skip "remote OST with nodsh"
local osts=$(osts_nodes)
reset_async
#define OBD_FAIL_OST_BRW_WRITE_BULK2 0x220
set_nodes_failloc $osts 0x220
# return -EIO from OST
$MULTIOP $DIR/$tfile oO_CREAT:O_RDWR:O_SYNC:w4096c
RC=$?
set_nodes_failloc $osts 0x0
if [[ $RC -eq 0 ]]; then
error "Must return error due to dropped pages, rc=$RC"
fi
LOCKED=$(lctl get_param -n llite.*.dump_page_cache | grep -c locked)
DIRTY=$(lctl get_param -n llite.*.dump_page_cache | grep -c dirty)
WRITEBACK=$(lctl get_param -n llite.*.dump_page_cache | grep -c writeback)
if [[ $LOCKED -ne 0 ]]; then
error "Locked pages remain in cache, locked=$LOCKED"
fi
# in recoverable error on OST we want resend and stay until it finished
if [[ $DIRTY -ne 0 || $WRITEBACK -ne 0 ]]; then
error "Dirty pages not flushed to disk, dirty=$DIRTY, writeback=$WRITEBACK"
fi
rm -f $DIR/$tfile
echo "No pages locked after fsync"
return 0
}
run_test 118j "Simulate unrecoverable OST side error =========="
test_118k()
{
[ $PARALLEL == "yes" ] && skip "skip parallel run"
remote_ost_nodsh && skip "remote OSTs with nodsh"
local osts=$(osts_nodes)
#define OBD_FAIL_OST_BRW_WRITE_BULK 0x20e
set_nodes_failloc $osts 0x20e
test_mkdir $DIR/$tdir
for ((i=0;i<10;i++)); do
(dd if=/dev/zero of=$DIR/$tdir/$tfile-$i bs=1M count=10 || \
error "dd to $DIR/$tdir/$tfile-$i failed" )&
SLEEPPID=$!
sleep 0.500s
kill $SLEEPPID
wait $SLEEPPID
done
set_nodes_failloc $osts 0
rm -rf $DIR/$tdir
}
run_test 118k "bio alloc -ENOMEM and IO TERM handling ========="
test_118l() # LU-646
{
[ $PARALLEL == "yes" ] && skip "skip parallel run"
test_mkdir $DIR/$tdir
$MULTIOP $DIR/$tdir Dy || error "fsync dir failed"
rm -rf $DIR/$tdir
}
run_test 118l "fsync dir"
test_118m() # LU-3066
{
[ $PARALLEL == "yes" ] && skip "skip parallel run"
test_mkdir $DIR/$tdir
$MULTIOP $DIR/$tdir DY || error "fdatasync dir failed"
rm -rf $DIR/$tdir
}
run_test 118m "fdatasync dir ========="
[ "$SLOW" = "no" ] && [ -n "$OLD_RESENDCOUNT" ] && set_resend_count $OLD_RESENDCOUNT
test_118n()
{
local begin
local end
[ $PARALLEL == "yes" ] && skip "skip parallel run"
remote_ost_nodsh && skip "remote OSTs with nodsh"
local osts=$(osts_nodes)
# Sleep to avoid a cached response.
#define OBD_STATFS_CACHE_SECONDS 1
sleep 2
# Inject a 10 second delay in the OST_STATFS handler.
#define OBD_FAIL_OST_STATFS_DELAY 0x242
set_nodes_failloc $osts 0x242
begin=$SECONDS
stat --file-system $MOUNT > /dev/null
end=$SECONDS
set_nodes_failloc $osts 0
if ((end - begin > 20)); then
error "statfs took $((end - begin)) seconds, expected 10"
fi
}
run_test 118n "statfs() sends OST_STATFS requests in parallel"
# With unaligned_dio enabled there are no restrictions on dio.
unaligned_dio() {
local udio=0
$LCTL get_param osc.*.import | grep connect_flags: |
grep -q "unaligned_dio" || udio=1
return $udio
}
# With unaligned_dio enabled there are no restrictions on dio.
unaligned_dio_or_skip()
{
if ! unaligned_dio; then
skip "Need 'unaligned_dio' support"
fi
}
dio_readv_writev_support()
{
# Kernels after 3.16 work:
(( $(version_code $(uname -r)) >= $(version_code 3.16) )) &&
return 0
# Lustre with LU-17524 works:
(( $OST1_VERSION > $(version_code 2.15.61.196) )) &&
return 0
skip "need readv/writev with O_DIRECT support"
}
test_119a() # bug 11737
{
BSIZE=$((512 * 1024))
directio write $DIR/$tfile 0 1 $BSIZE
# We ask to read two blocks, which is more than a file size.
# directio will indicate an error when requested and actual
# sizes aren't equeal (a normal situation in this case) and
# print actual read amount.
NOB=`directio read $DIR/$tfile 0 2 $BSIZE | awk '/error/ {print $6}'`
if [ "$NOB" != "$BSIZE" ]; then
error "read $NOB bytes instead of $BSIZE"
fi
rm -f $DIR/$tfile
}
run_test 119a "Short directIO read must return actual read amount"
test_119b() # bug 11737
{
[[ $OSTCOUNT -lt 2 ]] && skip_env "needs >= 2 OSTs"
$LFS setstripe -c 2 $DIR/$tfile || error "setstripe failed"
dd if=/dev/zero of=$DIR/$tfile bs=1M count=1 seek=1 || error "dd failed"
sync
$MULTIOP $DIR/$tfile oO_RDONLY:O_DIRECT:r$((2048 * 1024)) ||
error "direct read failed"
rm -f $DIR/$tfile
}
run_test 119b "Sparse directIO read must return actual read amount"
test_119c() # bug 13099
{
BSIZE=1048576
directio write $DIR/$tfile 3 1 $BSIZE || error "direct write failed"
directio readhole $DIR/$tfile 0 2 $BSIZE || error "reading hole failed"
rm -f $DIR/$tfile
}
run_test 119c "Testing for direct read hitting hole"
# Note: test 119d was removed, skipping 119d for new tests to avoid polluting
# Maloo test history
test_119e()
{
unaligned_dio_or_skip
(( $OSTCOUNT >= 2 )) || skip "needs >= 2 OSTs"
local stripe_size=$((1024 * 1024)) #1 MiB
# Max i/o below is ~ 4 * stripe_size, so this gives ~5 i/os
local file_size=$((25 * stripe_size))
local bsizes
$LFS setstripe -c 2 -S $stripe_size $DIR/$tfile.1
stack_trap "rm -f $DIR/$tfile*"
# Just a bit bigger than the largest size in the test set below
dd if=/dev/urandom bs=$file_size count=1 of=$DIR/$tfile.1 ||
error "buffered i/o to create file failed"
# trivial test of unaligned DIO
dd if=$DIR/$tfile.1 bs=4095 of=$DIR/$tfile.2 count=4 \
iflag=direct oflag=direct ||
error "trivial unaligned dio failed"
# Test of disabling unaligned DIO support
$LCTL set_param llite.*.unaligned_dio=0
stack_trap "$LCTL set_param llite.*.unaligned_dio=1"
echo "testing disabling unaligned DIO - 'invalid argument' expected:"
dd if=$DIR/$tfile.1 bs=1024 of=$DIR/$tfile.2 count=4 \
iflag=direct oflag=direct &&
error "unaligned dio succeeded when disabled"
$LCTL set_param llite.*.unaligned_dio=1
# Clean up before next part of test
rm -f $DIR/$tfile.2
if zfs_or_rotational; then
# DIO on ZFS can take up to 2 seconds per IO
# rotational is better, but still slow.
# Limit testing on those media to larger sizes
if unaligned_dio; then
bsizes="$((stripe_size - PAGE_SIZE)) $stripe_size \
$((stripe_size + 1024))"
else
bsizes="$((stripe_size - PAGE_SIZE)) $stripe_size \
$((stripe_size - 1))"
fi
elif unaligned_dio; then
bsizes="$((PAGE_SIZE / 4)) $((PAGE_SIZE - 1024)) \
$((PAGE_SIZE - 1)) $PAGE_SIZE $((PAGE_SIZE + 1024)) \
$((PAGE_SIZE * 3/2)) $((PAGE_SIZE * 4)) \
$((stripe_size - 1)) $stripe_size \
$((stripe_size + 1)) $((stripe_size * 3/2)) \
$((stripe_size * 4)) $((stripe_size * 4 + 1))"
else
bsizes="$((PAGE_SIZE / 4)) $((PAGE_SIZE - 1024)) \
$((PAGE_SIZE - 1)) $PAGE_SIZE $((PAGE_SIZE + 1024)) \
$((PAGE_SIZE * 3/2)) $((PAGE_SIZE * 4))"
fi
for bs in $bsizes; do
$LFS setstripe -c 2 -S $stripe_size $DIR/$tfile.2
echo "Read/write with DIO at size $bs"
# Read and write with DIO from source to dest
dd if=$DIR/$tfile.1 bs=$bs of=$DIR/$tfile.2 \
iflag=direct oflag=direct ||
error "dio failed"
ls -la $DIR/$tfile.1 $DIR/$tfile.2
$CHECKSTAT -t file -s $file_size $DIR/$tfile.2 ||
error "size incorrect, file copy read/write bsize: $bs"
cmp --verbose $DIR/$tfile.1 $DIR/$tfile.2 ||
error "files differ, bsize $bs"
rm -f $DIR/$tfile.2
done
}
run_test 119e "Basic tests of dio read and write at various sizes"
test_119f()
{
unaligned_dio_or_skip
(( $OSTCOUNT >= 2 )) || skip "needs >= 2 OSTs"
local stripe_size=$((1024 * 1024)) #1 MiB
# Max i/o below is ~ 4 * stripe_size, so this gives ~5 i/os
local file_size=$((25 * stripe_size))
local bsizes
$LFS setstripe -c 2 -S $stripe_size $DIR/$tfile.1
stack_trap "rm -f $DIR/$tfile*"
# Just a bit bigger than the largest size in the test set below
dd if=/dev/urandom bs=$file_size count=1 of=$DIR/$tfile.1 ||
error "buffered i/o to create file failed"
if zfs_or_rotational; then
# DIO on ZFS can take up to 2 seconds per IO
# rotational is better, but still slow.
# Limit testing on those media to larger sizes
if unaligned_dio; then
bsizes="$((stripe_size - PAGE_SIZE)) $stripe_size \
$((stripe_size + 1024))"
else
bsizes="$((stripe_size - PAGE_SIZE)) $stripe_size \
$((stripe_size - 1))"
fi
elif unaligned_dio; then
bsizes="$((PAGE_SIZE / 4)) $((PAGE_SIZE - 1024)) \
$((PAGE_SIZE - 1)) $PAGE_SIZE $((PAGE_SIZE + 1024)) \
$((PAGE_SIZE * 3/2)) $((PAGE_SIZE * 4)) \
$((stripe_size - 1)) $stripe_size \
$((stripe_size + 1)) $((stripe_size * 3/2)) \
$((stripe_size * 4)) $((stripe_size * 4 + 1))"
else
bsizes="$((PAGE_SIZE / 4)) $((PAGE_SIZE - 1024)) \
$((PAGE_SIZE - 1)) $PAGE_SIZE $((PAGE_SIZE + 1024)) \
$((PAGE_SIZE * 3/2)) $((PAGE_SIZE * 4))"
fi
for bs in $bsizes; do
$LFS setstripe -c 2 -S $stripe_size $DIR/$tfile.2
# Read and write with DIO from source to dest in two
# threads - should give correct copy of file
echo "bs: $bs"
dd if=$DIR/$tfile.1 bs=$bs of=$DIR/$tfile.2 iflag=direct \
oflag=direct conv=notrunc &
pid_dio1=$!
# Note block size is different here for a more interesting race
dd if=$DIR/$tfile.1 bs=$((bs * 2)) of=$DIR/$tfile.2 \
iflag=direct oflag=direct conv=notrunc &
pid_dio2=$!
wait $pid_dio1
rc1=$?
wait $pid_dio2
rc2=$?
if (( rc1 != 0 )); then
error "dio copy 1 w/bsize $bs failed: $rc1"
fi
if (( rc2 != 0 )); then
error "dio copy 2 w/bsize $bs failed: $rc2"
fi
$CHECKSTAT -t file -s $file_size $DIR/$tfile.2 ||
error "size incorrect, file copy read/write bsize: $bs"
cmp --verbose $DIR/$tfile.1 $DIR/$tfile.2 ||
error "files differ, bsize $bs"
rm -f $DIR/$tfile.2
done
}
run_test 119f "dio vs dio race"
test_119g()
{
unaligned_dio_or_skip
(( $OSTCOUNT >= 2 )) || skip "needs >= 2 OSTs"
local stripe_size=$((1024 * 1024)) #1 MiB
# Max i/o below is ~ 4 * stripe_size, so this gives ~5 i/os
local file_size=$((25 * stripe_size))
local bsizes
$LFS setstripe -c 2 -S $stripe_size $DIR/$tfile.1
stack_trap "rm -f $DIR/$tfile*"
# Just a bit bigger than the largest size in the test set below
dd if=/dev/urandom bs=$file_size count=1 of=$DIR/$tfile.1 ||
error "buffered i/o to create file failed"
if zfs_or_rotational; then
# DIO on ZFS can take up to 2 seconds per IO
# rotational is better, but still slow.
# Limit testing on those media to larger sizes
if unaligned_dio; then
bsizes="$((stripe_size - PAGE_SIZE)) $stripe_size \
$((stripe_size + 1024))"
else
bsizes="$((stripe_size - PAGE_SIZE)) $stripe_size \
$((stripe_size - 1))"
fi
elif unaligned_dio; then
bsizes="$((PAGE_SIZE / 4)) $((PAGE_SIZE - 1024)) \
$((PAGE_SIZE - 1)) $PAGE_SIZE $((PAGE_SIZE + 1024)) \
$((PAGE_SIZE * 3/2)) $((PAGE_SIZE * 4)) \
$((stripe_size - 1)) $stripe_size \
$((stripe_size + 1)) $((stripe_size * 3/2)) \
$((stripe_size * 4)) $((stripe_size * 4 + 1))"
else
bsizes="$((PAGE_SIZE / 4)) $((PAGE_SIZE - 1024)) \
$((PAGE_SIZE - 1)) $PAGE_SIZE $((PAGE_SIZE + 1024)) \
$((PAGE_SIZE * 3/2)) $((PAGE_SIZE * 4))"
fi
for bs in $bsizes; do
$LFS setstripe -c 2 -S $stripe_size $DIR/$tfile.2
echo "bs: $bs"
dd if=$DIR/$tfile.1 bs=$bs of=$DIR/$tfile.2 iflag=direct \
oflag=direct conv=notrunc &
pid_dio1=$!
# Buffered I/O with similar but not the same block size
dd if=$DIR/$tfile.1 bs=$((bs * 2)) of=$DIR/$tfile.2 &
pid_bio2=$!
wait $pid_dio1
rc1=$?
wait $pid_bio2
rc2=$?
if (( rc1 != 0 )); then
error "dio copy 1 w/bsize $bs failed: $rc1"
fi
if (( rc2 != 0 )); then
error "buffered copy 2 w/bsize $bs failed: $rc2"
fi
$CHECKSTAT -t file -s $file_size $DIR/$tfile.2 ||
error "size incorrect"
cmp --verbose $DIR/$tfile.1 $DIR/$tfile.2 ||
error "files differ, bsize $bs"
rm -f $DIR/$tfile.2
done
}
run_test 119g "dio vs buffered I/O race"
test_119h()
{
unaligned_dio_or_skip
(( $OSTCOUNT >= 2 )) || skip "needs >= 2 OSTs"
local stripe_size=$((1024 * 1024)) #1 MiB
# Max i/o below is ~ 4 * stripe_size, so this gives ~5 i/os
local file_size=$((25 * stripe_size))
local bsizes
stack_trap "rm -f $DIR/$tfile.*"
if zfs_or_rotational; then
# DIO on ZFS can take up to 2 seconds per IO
# rotational is better, but still slow.
# Limit testing on those media to larger sizes
if unaligned_dio; then
bsizes="$((stripe_size - PAGE_SIZE)) $stripe_size \
$((stripe_size + 1024))"
else
bsizes="$((stripe_size - PAGE_SIZE)) $stripe_size \
$((stripe_size - 1))"
fi
elif unaligned_dio; then
bsizes="$((PAGE_SIZE / 4)) $((PAGE_SIZE - 1024)) \
$((PAGE_SIZE - 1)) $PAGE_SIZE $((PAGE_SIZE + 1024)) \
$((PAGE_SIZE * 3/2)) $((PAGE_SIZE * 4)) \
$((stripe_size - 1)) $stripe_size \
$((stripe_size + 1)) $((stripe_size * 3/2)) \
$((stripe_size * 4)) $((stripe_size * 4 + 1))"
else
bsizes="$((PAGE_SIZE / 4)) $((PAGE_SIZE - 1024)) \
$((PAGE_SIZE - 1)) $PAGE_SIZE $((PAGE_SIZE + 1024)) \
$((PAGE_SIZE * 3/2)) $((PAGE_SIZE * 4))"
fi
for bs in $bsizes; do
$LFS setstripe -c 2 -S $stripe_size $DIR/$tfile.1
$LFS setstripe -c 2 -S $stripe_size $DIR/$tfile.2
echo "unaligned writes of blocksize: $bs"
# Write a file with unaligned DIO and regular DIO, and compare
# them
# with 'u', multiop randomly unaligns the io from the buffer
$MULTIOP $DIR/$tfile.1 \
oO_CREAT:O_RDWR:O_DIRECT:wu${bs}wu${bs}wu${bs}wu${bs}wu${bs} ||
error "multiop memory unaligned write failed, $bs"
$MULTIOP $DIR/$tfile.2 \
oO_CREAT:O_RDWR:O_DIRECT:w${bs}w${bs}w${bs}w${bs}w${bs} ||
error "multiop memory aligned write failed, $bs"
cmp --verbose $DIR/$tfile.1 $DIR/$tfile.2 ||
error "files differ, bsize $bs"
rm -f $DIR/$tfile.*
done
$LFS setstripe -c 2 -S $stripe_size $DIR/$tfile.1
dd if=/dev/zero bs=$((stripe_size * 5)) of=$DIR/$tfile.1 count=5 ||
error "dd to create source file for read failed"
# Just a few quick tests to make sure unaligned DIO reads don't crash
for bs in $bsizes; do
echo "unaligned reads of blocksize: $bs"
# with 'u', multiop randomly unaligns the io from the buffer
$MULTIOP $DIR/$tfile.1 \
oO_CREAT:O_RDWR:O_DIRECT:ru${bs}ru${bs}ru${bs}ru${bs}ru${bs} ||
error "multiop memory unaligned read failed, $bs"
done
rm -f $DIR/$tfile*
}
run_test 119h "basic tests of memory unaligned dio"
# Unaligned AIO is disabled and may not be reenabled
# See LU-18032
# aiocp with the '-a' option makes testing memory unaligned aio trivial
test_119i()
{
unaligned_dio_or_skip
(( $OSTCOUNT >= 2 )) || skip "needs >= 2 OSTs"
which aiocp || skip_env "no aiocp installed"
skip "LU-18032: Unaligned AIO is disabled and may not be re-enabled"
local stripe_size=$((1024 * 1024)) #1 MiB
# Max i/o below is ~ 4 * stripe_size, so this gives ~5 i/os
local file_size=$((25 * stripe_size))
local bsizes
$LFS setstripe -c 2 -S $stripe_size $DIR/$tfile.1
stack_trap "rm -f $DIR/$tfile.*"
# Just a bit bigger than the largest size in the test set below
dd if=/dev/urandom bs=$file_size count=1 of=$DIR/$tfile.1 ||
error "buffered i/o to create file failed"
if zfs_or_rotational; then
# DIO on ZFS can take up to 2 seconds per IO
# rotational is better, but still slow.
# Limit testing on those media to larger sizes
if unaligned_dio; then
bsizes="$((stripe_size - PAGE_SIZE)) $stripe_size \
$((stripe_size + 1024))"
else
bsizes="$((stripe_size - PAGE_SIZE)) $stripe_size \
$((stripe_size - 1))"
fi
elif unaligned_dio; then
bsizes="$((PAGE_SIZE / 4)) $((PAGE_SIZE - 1024)) \
$((PAGE_SIZE - 1)) $PAGE_SIZE $((PAGE_SIZE + 1024)) \
$((PAGE_SIZE * 3/2)) $((PAGE_SIZE * 4)) \
$((stripe_size - 1)) $stripe_size \
$((stripe_size + 1)) $((stripe_size * 3/2)) \
$((stripe_size * 4)) $((stripe_size * 4 + 1))"
else
bsizes="$((PAGE_SIZE / 4)) $((PAGE_SIZE - 1024)) \
$((PAGE_SIZE - 1)) $PAGE_SIZE $((PAGE_SIZE + 1024)) \
$((PAGE_SIZE * 3/2)) $((PAGE_SIZE * 4))"
fi
# Do page aligned and NOT page aligned AIO
for align in 8 512 $((PAGE_SIZE)); do
# Deliberately includes a few aligned sizes
for bs in $bsizes; do
$LFS setstripe -c 2 -S $stripe_size $DIR/$tfile.2
echo "bs: $bs, align: $align, file_size $file_size"
aiocp -a $align -b $bs -s $file_size -f O_DIRECT \
$DIR/$tfile.1 $DIR/$tfile.2 ||
error "unaligned aio failed, bs: $bs, align: $align"
$CHECKSTAT -t file -s $file_size $DIR/$tfile.2 ||
error "size incorrect"
cmp --verbose $DIR/$tfile.1 $DIR/$tfile.2 ||
error "files differ"
rm -f $DIR/$tfile.2
done
done
}
run_test 119i "test unaligned aio at varying sizes"
test_119j()
{
unaligned_dio_or_skip
(( $LINUX_VERSION_CODE > $(version_code 4.5.0) )) ||
skip "needs kernel > 4.5.0 for ki_flags support"
local rpcs
local pages=1
local iosize=$((pages * PAGE_SIZE / 1024))k
dd if=/dev/urandom of=$DIR/$tfile bs=$iosize count=1 ||
error "(0) dd $iosize failed"
sync
# Warm the page cache and let any first-touch readahead RPC settle
# before resetting the stats. Otherwise an async readahead RPC from
# the first read can race the measured read below and be miscounted
# as a cache miss (seen as "(3) 1 != 0" on large-page clients).
dd if=$DIR/$tfile of=/dev/null bs=$iosize count=1 ||
error "(1) dd $iosize warmup failed"
$LCTL set_param -n osc.*.rpc_stats=0
# Read from page cache, does not generate an rpc
dd if=$DIR/$tfile of=/dev/null bs=$iosize count=1 ||
error "(2) dd $iosize failed"
$LCTL get_param osc.*.rpc_stats
rpcs=($($LCTL get_param -n 'osc.*.rpc_stats' |
sed -n '/pages per rpc/,/^$/p' |
awk '/'$pages':/ { reads += $2; writes += $6 }; \
END { print reads,writes }'))
(( ${rpcs[0]} == 0 )) || error "(3) ${rpcs[0]} != 0 read RPCs"
# Test hybrid IO read
# Force next BIO as DIO
# This forces an RPC to the server
#define OBD_FAIL_LLITE_FORCE_BIO_AS_DIO 0x1429
$LCTL set_param fail_loc=0x1429
# Drop the cache so this IO is not done by fast read. Fast read runs
# before the BIO->DIO switch, so unless the cached pages are gone it
# will serve the read and no RPC is sent ("(5) 0 != 1"). vm.drop_caches
# is best-effort and does not reliably evict pages pinned by OSC DLM
# locks (racy on large-page clients), so cancel the locks instead to
# deterministically invalidate this file's cached pages.
cancel_lru_locks osc
dd if=$DIR/$tfile of=/dev/null bs=8 count=1 || error "(4) dd failed"
$LCTL get_param osc.*.rpc_stats
rpcs=($($LCTL get_param -n 'osc.*.rpc_stats' |
sed -n '/pages per rpc/,/^$/p' |
awk '/'$pages':/ { reads += $2; writes += $6 }; \
END { print reads,writes }'))
(( ${rpcs[0]} == $pages )) || error "(5) ${rpcs[0]} != $pages read RPCs"
# Test hybrid IO write
#define OBD_FAIL_LLITE_FORCE_BIO_AS_DIO 0x1429
$LCTL set_param fail_loc=0x1429
#NB: We do not check for 0 write RPCs in the BIO case because that
# would make the test racey vs cache flushing
# but the DIO case is guaranteed to generate 1 write RPC
cancel_lru_locks osc
dd if=/dev/zero of=$DIR/$tfile bs=8 count=1 || error "(6) dd failed"
$LCTL get_param osc.*.rpc_stats
rpcs=($($LCTL get_param -n 'osc.*.rpc_stats' |
sed -n '/pages per rpc/,/^$/p' |
awk '/'$pages':/ { reads += $2; writes += $6 }; \
END { print reads,writes }'))
(( ${rpcs[1]} == $pages )) ||
error "(7) ${rpcs[1]} != $pages write RPCs"
}
run_test 119j "basic tests of hybrid IO switching"
test_119k()
{
(( $LINUX_VERSION_CODE > $(version_code 4.5.0) )) ||
skip "needs kernel > 4.5.0 for ki_flags support"
local hybrid_noswitch
local hybrid_writes
local hybrid_reads
local wthresh=$($LCTL get_param -n \
llite.*.hybrid_io_write_threshold_bytes | head -n1)
local rthresh=$($LCTL get_param -n \
llite.*.hybrid_io_read_threshold_bytes | head -n1)
# Enable hybrid IO
local hybrid=$($LCTL get_param -n llite.*.hybrid_io)
$LCTL set_param llite.*.hybrid_io=1
stack_trap "$LCTL set_param -n llite.*.hybrid_io=$hybrid" EXIT
$LCTL set_param llite.*.stats=c
# Writes below the write threshold should not switch to hybrid
dd if=/dev/zero bs=$((wthresh / 2)) count=8 of=$DIR/$tfile ||
error "(0) dd failed"
hybrid_noswitch=($($LCTL get_param -n 'llite.*.stats' |
awk '/hybrid_noswitch/ {print $2}'))
$LCTL get_param llite.*.stats
[[ $hybrid_noswitch == 8 ]] ||
error "(1) incorrect number of hybrid noswitch ($hybrid_noswitch), expected 8"
# Writes at the write threshold should switch to hybrid/DIO
dd if=/dev/zero bs=$wthresh count=8 of=$DIR/$tfile ||
error "(2) dd failed"
hybrid_writes=($($LCTL get_param -n 'llite.*.stats' |
awk '/hybrid_writesize_switch/ {print $2}'))
$LCTL get_param llite.*.stats
[[ $hybrid_writes == 8 ]] ||
error "(3) incorrect number of hybrid writes ($hybrid_writes), expected 8"
$LCTL set_param llite.*.stats=c
# Drop cached pages so reads go through the full IO path and
# reach the hybrid switch check instead of being served by
# ll_do_fast_read() from page cache.
cancel_lru_locks osc
# Reads below the read threshold should not switch to hybrid
dd if=$DIR/$tfile bs=$((rthresh / 2)) count=2 of=/dev/null ||
error "(4) dd failed"
hybrid_noswitch=($($LCTL get_param -n 'llite.*.stats' |
awk '/hybrid_noswitch/ {print $2}'))
$LCTL get_param llite.*.stats
[[ $hybrid_noswitch == 2 ]] ||
error "(5) incorrect number of hybrid noswitch ($hybrid_noswitch), expected 2"
cancel_lru_locks osc
# Reads at the read threshold should use hybrid IO
dd if=$DIR/$tfile bs=$rthresh count=2 of=/dev/null ||
error "(6) dd failed"
hybrid_reads=($($LCTL get_param -n 'llite.*.stats' |
awk '/hybrid_readsize_switch/ {print $2}'))
$LCTL get_param llite.*.stats
[[ $hybrid_reads == 2 ]] ||
error "(7) incorrect number of hybrid reads ($hybrid_reads), expected 2"
$LCTL set_param llite.*.stats=c
# disable hybrid IO (original value restored by stack_trap above)
$LCTL set_param llite.*.hybrid_io=0
# Writes at the write threshold but hybrid IO is disabled
dd if=/dev/zero bs=$wthresh count=8 of=$DIR/$tfile ||
error "(8) dd failed"
hybrid_writes=($($LCTL get_param -n 'llite.*.stats' |
awk '/hybrid_writesize_switch/ {print $2}'))
$LCTL get_param llite.*.stats
[[ -z $hybrid_writes ]] ||
error "(9) there should be no hybrid writes, but we have '$hybrid_writes'"
cancel_lru_locks osc
# Reads at the read threshold but hybrid IO is disabled
dd if=$DIR/$tfile bs=$rthresh count=2 of=/dev/null ||
error "(10) dd failed"
hybrid_reads=($($LCTL get_param -n 'llite.*.stats' |
awk '/hybrid_readsize_switch/ {print $2}'))
$LCTL get_param llite.*.stats
[[ -z $hybrid_reads ]] ||
error "(11) there should be no hybrid reads, but we have '$hybrid_reads'"
}
run_test 119k "hybrid IO counting with stats and disabling"
test_119m() {
dio_readv_writev_support
timeout 90s rwv -f $DIR/$tfile -Dw -n 3 0x80000 0x100000 0x180000 ||
error "DIO aligned writev test failed"
timeout 90s rwv -f $DIR/$tfile -Dr -v -n 2 0x180000 0x100000 ||
error "DIO aligned readv failed"
rm -f $DIR/$tfile
}
run_test 119m "Test DIO readv/writev: exercise iter duplication"
test_119n()
{
# zfs server should fail without unaligned_dio connect flag
# and report EINVAL when attempting unaligned dio.
dio_readv_writev_support
[[ "${ost1_FSTYPE}" == "zfs" ]] ||
skip "need ZFS server without unaligned_dio support"
$LCTL get_param osc.*.import | grep connect_flags: |
grep -q "unaligned_dio" &&
skip "zfs server without 'unaligned_dio' support"
error_string=$(rwv -f $DIR/$tfile -Dw -n2 1024 4096 2>&1) &&
error "Allowed unaligned dio with ZFS with unpatched server"
grep -q "Invalid argument" <<< $error_string ||
error "Expected 'Invalid argument' failure: '$error_string'."
rwv -f $DIR/$tfile -Dw -n2 65536 4096 ||
error "DIO aligned writev test failed"
error_string=$(rwv -f $DIR/$tfile -Dr -v -n2 1024 4096 2>&1) &&
error "Allowed unaligned dio with ZFS with unpatched server"
grep -q "Invalid argument" <<< $error_string ||
error "Expected 'Invalid argument' failure: '$error_string'."
rm -f $DIR/$tfile
}
run_test 119n "Test Unaligned DIO readv() and writev() with unpatched ZFS"
test_119o()
{
dio_readv_writev_support
[[ "${ost1_FSTYPE}" == "zfs" ]] ||
skip "need ldiskfs without unaligned_dio support."
$LCTL get_param osc.*.import | grep connect_flags: |
grep -q "unaligned_dio" &&
skip "need ldiskfs without 'unaligned_dio' support"
error_string=$(timeout 200s \
rwv -f $DIR/$tfile -Dw -n 3 0x7ffff 0x100001 0x180000 2>&1) &&
error "Allowed 64k unaligned dio writev"
grep -q -E 'Invalid argument|Write error:' <<< $error_string ||
error "Expected 'Invalid argument' failure: '$error_string'."
rwv -f $DIR/$tfile -Dw -n 3 0x80000 0x100000 0x180000 ||
error "DIO aligned writev test failed"
error_string=$(timeout 200s \
rwv -f $DIR/$tfile -Dr -v -n 3 0x7ffff 0x100001 0x180000 2>&1) &&
error "Allowed 64k unaligned dio readv"
grep -q -E 'Invalid argument|Read error:' <<< $error_string ||
error "Expected 'Invalid argument' failure: '$error_string'."
rm -f $DIR/$tfile
}
run_test 119o "Test Unaligned DIO readv() and writev() with unpatched servers"
test_119p()
{
# Patched servers, unaligned dio that needs interop page alignment
dio_readv_writev_support
$LCTL get_param osc.*.import | grep connect_flags: |
grep -q "unaligned_dio" ||
skip "need unaligned_dio support."
rwv -f $DIR/$tfile -Dw -n 3 0x7ffff 0x100001 0x180000 ||
error "DIO unaligned writev test failed"
rwv -f $DIR/$tfile -Dr -v -n 3 0x7ffff 0x100001 0x180000 ||
error "DIO unaligned readv failed"
rm -f $DIR/$tfile
}
run_test 119p "Test Unaligned DIO readv() and writev() with patched servers"
test_119q()
{
dio_readv_writev_support
$LCTL get_param osc.*.import | grep connect_flags: |
grep -q "unaligned_dio" ||
skip "need unaligned_dio support."
local page_size
local off0
local off1
local off2
# 4k: 0x1000, 8k: 0x2000, 32k: 0x8000 64k: 0x10000
for page_size in 0x1000 0x2000 0x8000 0x10000; do
echo "RWV interop with 64k + $page_size +/- 1"
# short by 1 byte
off0=$((page_size + 0xffff))
echo "writev: $off0 0x100001 0x100000"
timeout 90s rwv -f $DIR/$tfile -Dw -n 3 $off0 0x100001 0x100000 ||
error "DIO unaligned writev test failed: $off0"
echo "readv: 0x100001 $off0 0x100000"
timeout 90s rwv -f $DIR/$tfile -Dr -v -n 3 0x100001 $off0 0x100000 ||
error "DIO unaligned readv failed: $off0"
rm -f $DIR/$tfile
# page offset exactly
off1=$((page_size + 0x10000))
echo "writev: $off1 0x100001 0x100000"
timeout 90s rwv -f $DIR/$tfile -Dw -n 3 $off1 0x100001 0x100000 ||
error "DIO unaligned writev test failed: $off1"
echo "readv: 0x100001 $off1 0x100000"
timeout 90s rwv -f $DIR/$tfile -Dr -v -n 3 0x100001 $off1 0x100000 ||
error "DIO unaligned readv failed: $off1"
rm -f $DIR/$tfile
# page offset over by 1 byte
off2=$((page_size + 0x10001))
echo "writev: $off2 0x100001 0x100000"
timeout 90s rwv -f $DIR/$tfile -Dw -n 3 $off2 0x100001 0x100000 ||
error "DIO unaligned writev test failed: $off2"
echo "readv: 3 0x100001 $off2 0x100000"
timeout 90s rwv -f $DIR/$tfile -Dr -v -n 3 0x100001 $off2 0x100000 ||
error "DIO unaligned readv failed: $off2"
rm -f $DIR/$tfile
done
}
run_test 119q "Test patchded Unaligned DIO readv() and writev()"
test_119r() {
unaligned_dio_or_skip
# Test error handling in unaligned DIO user copy with racing threads
local file=$DIR/$tfile
$LFS setstripe -c 2 $file || error "setstripe failed"
stack_trap "rm -f $file"
# Create a file with some data
dd if=/dev/urandom of=$file bs=1M count=1 || error "dd failed"
# Set fail_loc to inject error in DIO copy
#define OBD_FAIL_LLITE_DIO_COPY_ERR 0x1437
$LCTL set_param fail_loc=0x1437
stack_trap "$LCTL set_param fail_loc=0"
# Use rwv to do unaligned DIO write at offset 1024 with size 4096
# This is unaligned because offset 1024 is not page-aligned
local output
output=$(rwv -f $file -Dw -n 1 1024 4096 2>&1) &&
error "Unaligned DIO write should have failed but succeeded"
echo "$output" | grep -q "Bad address" ||
error "Expected 'Bad address' error, got: $output"
# Clear fail_loc
$LCTL set_param fail_loc=0
# Verify normal aligned DIO works after error
dd if=/dev/zero of=$file bs=4096 count=1 oflag=direct ||
error "DIO write failed after clearing fail_loc"
}
run_test 119r "Test error handling in unaligned DIO user copy"
test_120a() {
[ $PARALLEL == "yes" ] && skip "skip parallel run"
remote_mds_nodsh && skip "remote MDS with nodsh"
test_mkdir -i0 -c1 $DIR/$tdir
$LCTL get_param -n mdc.*.connect_flags | grep -q early_lock_cancel ||
skip_env "no early lock cancel on server"
lru_resize_disable mdc
lru_resize_disable osc
cancel_lru_locks mdc
# asynchronous object destroy at MDT could cause bl ast to client
cancel_lru_locks osc
local old=$($LCTL get_param -n llite.*.dir_read_on_open)
# statahead_agl may cause extra glimpse which confuses results. LU-13017
$LCTL set_param -n llite.*.dir_read_on_open=0
stack_trap "$LCTL set_param -n llite.*.dir_read_on_open=$old" EXIT
stat $DIR/$tdir > /dev/null
can1=$(do_facet mds1 \
"$LCTL get_param -n ldlm.services.ldlm_canceld.stats" |
awk '/ldlm_cancel/ {print $2}')
blk1=$($LCTL get_param -n ldlm.services.ldlm_cbd.stats |
awk '/ldlm_bl_callback/ {print $2}')
test_mkdir -i0 -c1 $DIR/$tdir/d1
can2=$(do_facet mds1 \
"$LCTL get_param -n ldlm.services.ldlm_canceld.stats" |
awk '/ldlm_cancel/ {print $2}')
blk2=$($LCTL get_param -n ldlm.services.ldlm_cbd.stats |
awk '/ldlm_bl_callback/ {print $2}')
(( $can1 == $can2 )) || error "$((can2-can1)) cancel RPC occured."
(( $blk1 == $blk2 )) || error "$((blk2-blk1)) blocking RPC occured."
}
run_test 120a "Early Lock Cancel: mkdir test"
test_120b() {
[ $PARALLEL == "yes" ] && skip "skip parallel run"
remote_mds_nodsh && skip "remote MDS with nodsh"
test_mkdir $DIR/$tdir
$LCTL get_param -n mdc.*.connect_flags | grep -q early_lock_cancel ||
skip_env "no early lock cancel on server"
lru_resize_disable mdc
lru_resize_disable osc
cancel_lru_locks mdc
stat $DIR/$tdir > /dev/null
can1=$(do_facet $SINGLEMDS \
"$LCTL get_param -n ldlm.services.ldlm_canceld.stats" |
awk '/ldlm_cancel/ {print $2}')
blk1=$($LCTL get_param -n ldlm.services.ldlm_cbd.stats |
awk '/ldlm_bl_callback/ {print $2}')
touch $DIR/$tdir/f1
can2=$(do_facet $SINGLEMDS \
"$LCTL get_param -n ldlm.services.ldlm_canceld.stats" |
awk '/ldlm_cancel/ {print $2}')
blk2=$($LCTL get_param -n ldlm.services.ldlm_cbd.stats |
awk '/ldlm_bl_callback/ {print $2}')
(( $can1 == $can2 )) || error "$((can2-can1)) cancel RPC occured."
(( $blk1 == $blk2 )) || error "$((blk2-blk1)) blocking RPC occured."
}
run_test 120b "Early Lock Cancel: create test"
test_120c() {
[ $PARALLEL == "yes" ] && skip "skip parallel run"
remote_mds_nodsh && skip "remote MDS with nodsh"
test_mkdir -i0 -c1 $DIR/$tdir
$LCTL get_param -n mdc.*.connect_flags | grep -q early_lock_cancel ||
skip "no early lock cancel on server"
lru_resize_disable mdc
lru_resize_disable osc
test_mkdir -i0 -c1 $DIR/$tdir/d1
test_mkdir -i0 -c1 $DIR/$tdir/d2
touch $DIR/$tdir/d1/f1
cancel_lru_locks mdc
stat $DIR/$tdir/d1 $DIR/$tdir/d2 $DIR/$tdir/d1/f1 > /dev/null
can1=$(do_facet mds1 \
"$LCTL get_param -n ldlm.services.ldlm_canceld.stats" |
awk '/ldlm_cancel/ {print $2}')
blk1=$($LCTL get_param -n ldlm.services.ldlm_cbd.stats |
awk '/ldlm_bl_callback/ {print $2}')
ln $DIR/$tdir/d1/f1 $DIR/$tdir/d2/f2
can2=$(do_facet mds1 \
"$LCTL get_param -n ldlm.services.ldlm_canceld.stats" |
awk '/ldlm_cancel/ {print $2}')
blk2=$($LCTL get_param -n ldlm.services.ldlm_cbd.stats |
awk '/ldlm_bl_callback/ {print $2}')
(( $can1 == $can2 )) || error "$((can2-can1)) cancel RPC occured."
(( $blk1 == $blk2 )) || error "$((blk2-blk1)) blocking RPC occured."
}
run_test 120c "Early Lock Cancel: link test"
test_120d() {
[ $PARALLEL == "yes" ] && skip "skip parallel run"
remote_mds_nodsh && skip "remote MDS with nodsh"
test_mkdir -i0 -c1 $DIR/$tdir
$LCTL get_param -n mdc.*.connect_flags | grep -q early_lock_cancel ||
skip_env "no early lock cancel on server"
lru_resize_disable mdc
lru_resize_disable osc
touch $DIR/$tdir
cancel_lru_locks mdc
stat $DIR/$tdir > /dev/null
can1=$(do_facet mds1 \
"$LCTL get_param -n ldlm.services.ldlm_canceld.stats" |
awk '/ldlm_cancel/ {print $2}')
blk1=$($LCTL get_param -n ldlm.services.ldlm_cbd.stats |
awk '/ldlm_bl_callback/ {print $2}')
chmod a+x $DIR/$tdir
can2=$(do_facet mds1 \
"$LCTL get_param -n ldlm.services.ldlm_canceld.stats" |
awk '/ldlm_cancel/ {print $2}')
blk2=$($LCTL get_param -n ldlm.services.ldlm_cbd.stats |
awk '/ldlm_bl_callback/ {print $2}')
(( $can1 == $can2 )) || error "$((can2-can1)) cancel RPC occured."
(( $blk1 == $blk2 )) || error "$((blk2-blk1)) blocking RPC occured."
}
run_test 120d "Early Lock Cancel: setattr test"
test_120e() {
[ $PARALLEL == "yes" ] && skip "skip parallel run"
$LCTL get_param -n mdc.*.connect_flags | grep -q early_lock_cancel ||
skip_env "no early lock cancel on server"
remote_mds_nodsh && skip "remote MDS with nodsh"
local dlmtrace_set=false
test_mkdir -i0 -c1 $DIR/$tdir
lru_resize_disable mdc
lru_resize_disable osc
! $LCTL get_param debug | grep -q dlmtrace &&
$LCTL set_param debug=+dlmtrace &&
stack_trap "$LCTL set_param debug=-dlmtrace"
dd if=/dev/zero of=$DIR/$tdir/f1 count=1
cancel_lru_locks mdc
cancel_lru_locks osc
dd if=$DIR/$tdir/f1 of=/dev/null
stat $DIR/$tdir $DIR/$tdir/f1 > /dev/null
# XXX client can not do early lock cancel of OST lock
# during unlink (LU-4206), so cancel osc lock now.
sleep 2
cancel_lru_locks osc
can1=$(do_facet mds1 \
"$LCTL get_param -n ldlm.services.ldlm_canceld.stats" |
awk '/ldlm_cancel/ {print $2}')
blk1=$($LCTL get_param -n ldlm.services.ldlm_cbd.stats |
awk '/ldlm_bl_callback/ {print $2}')
unlink $DIR/$tdir/f1
sleep 5
can2=$(do_facet mds1 \
"$LCTL get_param -n ldlm.services.ldlm_canceld.stats" |
awk '/ldlm_cancel/ {print $2}')
blk2=$($LCTL get_param -n ldlm.services.ldlm_cbd.stats |
awk '/ldlm_bl_callback/ {print $2}')
(( $can1 == $can2 )) || error "$((can2 - can1)) cancel RPC occured"
(( $blk1 == $blk2 )) || error "$((blk2 - blk1)) blocking RPC occured"
}
run_test 120e "Early Lock Cancel: unlink test"
test_120f() {
[ $PARALLEL == "yes" ] && skip "skip parallel run"
$LCTL get_param -n mdc.*.connect_flags | grep -q early_lock_cancel ||
skip_env "no early lock cancel on server"
remote_mds_nodsh && skip "remote MDS with nodsh"
test_mkdir -i0 -c1 $DIR/$tdir
lru_resize_disable mdc
lru_resize_disable osc
test_mkdir -i0 -c1 $DIR/$tdir/d1
test_mkdir -i0 -c1 $DIR/$tdir/d2
dd if=/dev/zero of=$DIR/$tdir/d1/f1 count=1
dd if=/dev/zero of=$DIR/$tdir/d2/f2 count=1
cancel_lru_locks mdc
cancel_lru_locks osc
dd if=$DIR/$tdir/d1/f1 of=/dev/null
dd if=$DIR/$tdir/d2/f2 of=/dev/null
stat $DIR/$tdir/d1 $DIR/$tdir/d2 $DIR/$tdir/d1/f1 $DIR/$tdir/d2/f2 > /dev/null
# XXX client can not do early lock cancel of OST lock
# during rename (LU-4206), so cancel osc lock now.
sleep 2
cancel_lru_locks osc
can1=$(do_facet mds1 \
"$LCTL get_param -n ldlm.services.ldlm_canceld.stats" |
awk '/ldlm_cancel/ {print $2}')
blk1=$($LCTL get_param -n ldlm.services.ldlm_cbd.stats |
awk '/ldlm_bl_callback/ {print $2}')
mrename $DIR/$tdir/d1/f1 $DIR/$tdir/d2/f2
sleep 5
can2=$(do_facet mds1 \
"$LCTL get_param -n ldlm.services.ldlm_canceld.stats" |
awk '/ldlm_cancel/ {print $2}')
blk2=$($LCTL get_param -n ldlm.services.ldlm_cbd.stats |
awk '/ldlm_bl_callback/ {print $2}')
(( $can1 == $can2 )) || error "$((can2-can1)) cancel RPC occured."
(( $blk1 == $blk2 )) || error "$((blk2-blk1)) blocking RPC occured."
}
run_test 120f "Early Lock Cancel: rename test"
test_120g() {
[ $PARALLEL == "yes" ] && skip "skip parallel run"
$LCTL get_param -n mdc.*.connect_flags | grep -q early_lock_cancel ||
skip_env "no early lock cancel on server"
remote_mds_nodsh && skip "remote MDS with nodsh"
lru_resize_disable mdc
lru_resize_disable osc
count=10000
echo create $count files
test_mkdir $DIR/$tdir
cancel_lru_locks mdc
cancel_lru_locks osc
t0=$(date +%s)
can0=$(do_facet $SINGLEMDS \
"$LCTL get_param -n ldlm.services.ldlm_canceld.stats" |
awk '/ldlm_cancel/ {print $2}')
blk0=$($LCTL get_param -n ldlm.services.ldlm_cbd.stats |
awk '/ldlm_bl_callback/ {print $2}')
createmany -o $DIR/$tdir/f $count
sync
can1=$(do_facet $SINGLEMDS \
"$LCTL get_param -n ldlm.services.ldlm_canceld.stats" |
awk '/ldlm_cancel/ {print $2}')
blk1=$($LCTL get_param -n ldlm.services.ldlm_cbd.stats |
awk '/ldlm_bl_callback/ {print $2}')
t1=$(date +%s)
echo "total: $((can1-can0)) cancels, $((blk1-blk0)) blockings"
echo "rm $count files"
rm -r $DIR/$tdir
sync
can2=$(do_facet $SINGLEMDS \
"$LCTL get_param -n ldlm.services.ldlm_canceld.stats" |
awk '/ldlm_cancel/ {print $2}')
blk2=$($LCTL get_param -n ldlm.services.ldlm_cbd.stats |
awk '/ldlm_bl_callback/ {print $2}')
t2=$(date +%s)
echo "total: $count removes in $((t2-t1))"
echo "total: $((can2-can1)) cancels, $((blk2-blk1)) blockings"
wait_delete_completed
}
run_test 120g "Early Lock Cancel: performance test"
test_121() { #bug #10589
[ $PARALLEL == "yes" ] && skip "skip parallel run"
rm -rf $DIR/$tfile
writes=$(LANG=C dd if=/dev/zero of=$DIR/$tfile count=1 2>&1 | awk -F '+' '/out$/ {print $1}')
#define OBD_FAIL_LDLM_CANCEL_RACE 0x310
lctl set_param fail_loc=0x310
cancel_lru_locks osc > /dev/null
reads=$(LANG=C dd if=$DIR/$tfile of=/dev/null 2>&1 | awk -F '+' '/in$/ {print $1}')
lctl set_param fail_loc=0
[[ $reads -eq $writes ]] ||
error "read $reads blocks, must be $writes blocks"
}
run_test 121 "read cancel race ========="
test_123a_base() { # was test 123, statahead(b=11401)
local lsx="$1"
ost_set_temp_seq_width_all $DATA_SEQ_MAX_WIDTH
local slow_is=error
if ! grep -q "processor.*: 1" /proc/cpuinfo; then
log "testing UP system. Performance may be lower than expected."
slow_is=log
fi
running_in_vm && slow_is=log
lru_resize_enable
$LCTL set_param mdc.*.batch_stats=0
rm -rf $DIR/$tdir
test_mkdir $DIR/$tdir
local numfree=$(df -i -P $DIR | tail -n 1 | awk '{ print $4 }')
(( $numfree > 100000 )) && numfree=100000 || numfree=$((numfree-1000))
stack_trap "unlinkmany $DIR/$tdir/$tfile $numfree 2> /dev/null || true"
local mult=10
for ((i=100, j=0; i<=$numfree; j=$i, i=$((i * mult)) )); do
local stime=$SECONDS
createmany -o $DIR/$tdir/$tfile $j $((i - j))
local etime=$SECONDS
local delta_cr=$((etime - stime))
local max=($($LCTL get_param -n llite.*.statahead_max))
$LCTL set_param llite.*.statahead_max=0
cancel_lru_locks mdc
cancel_lru_locks osc
local stime=$SECONDS
time $lsx $DIR/$tdir | wc -l
local etime=$SECONDS
local delta=$((etime - stime))
log "'$lsx' $i files without statahead: $delta sec"
$LCTL set_param llite.*.statahead_max=$max
local swrong=$($LCTL get_param -n llite.*.statahead_stats |
awk '/statahead.wrong:/ { print $NF }')
$LCTL get_param llite.*.statahead_max
cancel_lru_locks mdc
cancel_lru_locks osc
stime=$SECONDS
time $lsx $DIR/$tdir | wc -l
etime=$SECONDS
local delta_sa=$((etime - stime))
log "'$lsx' $i files with statahead: $delta_sa sec"
$LCTL get_param -n llite.*.statahead_stats
local ewrong=$($LCTL get_param -n llite.*.statahead_stats |
awk '/statahead.wrong:/ { print $NF }')
echo "swrong=$swrong ewrong=$ewrong"
(( $swrong >= $ewrong )) ||
log "statahead was stopped, maybe too many locks held!"
(( $delta_cr <= 90 )) || [[ "$SLOW" == "yes" ]] || break
(( $delta != 0 && $delta_sa != 0 )) || continue
if (( $delta_sa*100 > $delta*105 && $delta_sa > $delta+2)); then
max=($($LCTL get_param -n llite.*.statahead_max))
$LCTL set_param llite.*.statahead_max=0
cancel_lru_locks mdc
cancel_lru_locks osc
stime=$SECONDS
time $lsx $DIR/$tdir | wc -l
etime=$SECONDS
delta=$((etime - stime))
log "'$lsx' $i files again w/o statahead: $delta sec"
$LCTL set_param llite.*.statahead_max=$max
if (( $delta_sa * 100 > $delta * 105 &&
$delta_sa > $delta + 2 )); then
$slow_is "'$lsx' $i files slower with statahead"
break
fi
fi
(( $delta <= 20 )) || break
(( $delta <= 8 )) || mult=$((50 / delta))
[[ "$SLOW" == "yes" ]] || (( $delta <= 5 )) || break
done
log "'$lsx' done"
stime=$SECONDS
find $DIR/$tdir -type f -print0 | xargs -0 -P 4 -r -n 1 unlink ||
error "failed to unlink $DIR/$tdir $i"
rm -r $DIR/$tdir
sync
etime=$SECONDS
delta=$((etime - stime))
log "unlink + rm -r $DIR/$tdir: $delta seconds"
$LCTL get_param llite.*.statahead_stats
$LCTL get_param mdc.*.batch_stats
}
test_123aa() {
[ $PARALLEL == "yes" ] && skip "skip parallel run"
test_123a_base "ls -l"
}
run_test 123aa "verify statahead work"
test_123ab() {
[ $PARALLEL == "yes" ] && skip "skip parallel run"
statx_supported || skip_env "Test must be statx() syscall supported"
test_123a_base "$STATX -l"
}
run_test 123ab "verify statahead work by using statx"
test_123ac() {
[ $PARALLEL == "yes" ] && skip "skip parallel run"
statx_supported || skip_env "Test must be statx() syscall supported"
local rpcs_before
local rpcs_after
local agl_before
local agl_after
cancel_lru_locks $OSC
rpcs_before=$(calc_stats $OSC.*$OSC*.stats ldlm_glimpse_enqueue)
agl_before=$($LCTL get_param -n llite.*.statahead_stats |
awk '/agl.total:/ { print $NF }')
test_123a_base "$STATX -c \"%n %i %A %h %u %g %W %X %Z\" -D"
test_123a_base "$STATX --cached=always -D"
agl_after=$($LCTL get_param -n llite.*.statahead_stats |
awk '/agl.total:/ { print $NF }')
(( $agl_before == $agl_after )) ||
error "Should not trigger AGL thread - $agl_before:$agl_after"
rpcs_after=$(calc_stats $OSC.*$OSC*.stats ldlm_glimpse_enqueue)
(( $rpcs_after == $rpcs_before )) ||
error "$STATX should not send glimpse RPCs to $OSC"
}
run_test 123ac "verify statahead work by using statx without glimpse RPCs"
test_batch_statahead() {
local max=$1
local batch_max=$2
local num=10000
local batch_rpcs
local unbatch_rpcs
local hit_total
echo -e "\n\n\nbatch: statahead_max=$max statahead_batch_max=$batch_max"
$LCTL set_param mdc.*.batch_stats=0
$LCTL set_param llite.*.statahead_max=$max
$LCTL set_param llite.*.statahead_batch_max=$batch_max
# Verify that batched statahead is faster than one without statahead
test_123a_base "ls -l"
stack_trap "rm -rf $DIR/$tdir" EXIT
mkdir $DIR/$tdir || error "failed to mkdir $DIR/$tdir"
createmany -o $DIR/$tdir/$tfile $num || error "failed to create files"
# unbatched statahead
$LCTL set_param llite.*.statahead_batch_max=0
$LCTL set_param llite.*.statahead_stats=clear
$LCTL set_param mdc.*.stats=clear
cancel_lru_locks mdc
cancel_lru_locks osc
echo -n "ls -l files in $tdir: "
time ls -l $DIR/$tdir | wc -l
unbatch_rpcs=$(calc_stats mdc.*.stats ldlm_ibits_enqueue)
wait_update_facet client "pgrep ll_sa" "" 35 ||
error "ll_sa thread is still running"
hit_total=$($LCTL get_param -n llite.*.statahead_stats |
awk '/hit.total:/ { print $NF }')
# hit ratio should be larger than 75% (7500).
echo "unbatched statahead RPC=$unbatch_rpcs hit count=$hit_total"
(( $hit_total > 7500 )) || {
$LCTL get_param llite.*.statahead_stats
error "unbatched statahead RPC=$unbatch_rpcs hit count=$hit_total too low"
}
# batched statahead
$LCTL set_param llite.*.statahead_batch_max=$batch_max
$LCTL set_param llite.*.statahead_stats=clear
$LCTL set_param mdc.*.batch_stats=clear
$LCTL set_param mdc.*.stats=clear
cancel_lru_locks mdc
cancel_lru_locks osc
echo -n "ls -l files in $tdir: "
time ls -l $DIR/$tdir | wc -l
batch_rpcs=$(calc_stats mdc.*.stats mds_batch)
# wait for statahead thread to quit and update statahead stats
wait_update_facet client "pgrep ll_sa" "" 35 ||
error "ll_sa thread is still running"
hit_total=$($LCTL get_param -n llite.*.statahead_stats |
awk '/hit.total:/ { print $NF }')
# hit ratio should be larger than 75% (7500).
echo "batched statahead RPC=$batch_rpc hit count=$hit_total"
(( $hit_total > 7500 )) || {
$LCTL get_param llite.*.statahead_stats
error "batched statahead RPC=$batch_rpc hit count=$hit_total too low"
}
(( $unbatch_rpcs > $batch_rpcs )) ||
error "batched statahead does not reduce RPCs ($unbatch_rpcs < $batch_rpcs)"
$LCTL get_param mdc.*.batch_stats
}
test_123ad() {
[ $PARALLEL == "yes" ] && skip "skip parallel run"
(( $MDS1_VERSION >= $(version_code 2.15.53) )) ||
skip "Need server version at least 2.15.53"
$LCTL get_param -n mdc.*.connect_flags | grep -q batch_rpc ||
skip "Server does not support batch RPC"
local max
local batch_max
max=$($LCTL get_param -n llite.*.statahead_max | head -n 1)
batch_max=$($LCTL get_param -n llite.*.statahead_batch_max | head -n 1)
stack_trap "$LCTL set_param llite.*.statahead_max=$max" EXIT
stack_trap "$LCTL set_param llite.*.statahead_batch_max=$batch_max" EXIT
test_batch_statahead 32 32
test_batch_statahead 2048 256
}
run_test 123ad "Verify batching statahead works correctly"
test_123b () { # statahead(bug 15027)
[ $PARALLEL == "yes" ] && skip "skip parallel run"
test_mkdir $DIR/$tdir
createmany -o $DIR/$tdir/$tfile-%d 1000
cancel_lru_locks mdc
cancel_lru_locks osc
#define OBD_FAIL_MDC_GETATTR_ENQUEUE 0x803
lctl set_param fail_loc=0x80000803
ls -lR $DIR/$tdir > /dev/null
log "ls done"
lctl set_param fail_loc=0x0
lctl get_param -n llite.*.statahead_stats
rm -r $DIR/$tdir
sync
}
run_test 123b "not panic with network error in statahead enqueue (bug 15027)"
test_123c() {
[[ $MDSCOUNT -lt 2 ]] && skip_env "needs >= 2 MDTs"
test_mkdir -i 0 -c 1 $DIR/$tdir.0
test_mkdir -i 1 -c 1 $DIR/$tdir.1
touch $DIR/$tdir.1/{1..3}
mv $DIR/$tdir.1/{1..3} $DIR/$tdir.0
remount_client $MOUNT
$MULTIOP $DIR/$tdir.0 Q
# let statahead to complete
ls -l $DIR/$tdir.0 > /dev/null
testid=$(echo $TESTNAME | tr '_' ' ')
dmesg | tac | sed "/$testid/,$ d" | grep "Can not initialize inode" &&
error "statahead warning" || true
}
run_test 123c "Can not initialize inode warning on DNE statahead"
test_123d() {
local num=100
local swrong
local ewrong
test_mkdir -c -1 $DIR/$tdir || error "test_mkdir $DIR/$tdir failed"
$LFS setdirstripe -D -c $MDSCOUNT $DIR/$tdir ||
error "setdirstripe $DIR/$tdir failed"
createmany -d $DIR/$tdir/$tfile $num || error "createmany $num failed"
stack_trap "unlinkmany -d $DIR/$tdir/$tfile $num"
remount_client $MOUNT
$LCTL get_param llite.*.statahead_max
$LCTL set_param llite.*.statahead_stats=0 ||
error "clear statahead_stats failed"
swrong=$(lctl get_param -n llite.*.statahead_stats |
awk '/statahead.wrong:/ { print $NF }')
ls -l $DIR/$tdir || error "ls -l $DIR/$tdir failed"
# wait for statahead thread finished to update hit/miss stats.
sleep 1
$LCTL get_param -n llite.*.statahead_stats
ewrong=$(lctl get_param -n llite.*.statahead_stats |
awk '/statahead.wrong:/ { print $NF }')
(( $swrong == $ewrong )) ||
log "statahead was stopped, maybe too many locks held!"
}
run_test 123d "Statahead on striped directories works correctly"
test_123e() {
local max
local batch_max
local dir=$DIR/$tdir
$LCTL get_param -n mdc.*.connect_flags | grep -q batch_rpc ||
skip "Server does not support batch RPC"
mkdir $dir || error "mkdir $dir failed"
$LFS setstripe -C 32 $dir || error "setstripe $dir failed"
stack_trap "rm -rf $dir"
touch $dir/$tfile.{0..1000} || error "touch 1000 files failed"
max=$($LCTL get_param -n llite.*.statahead_max | head -n 1)
batch_max=$($LCTL get_param -n llite.*.statahead_batch_max | head -n 1)
stack_trap "$LCTL set_param llite.*.statahead_max=$max" EXIT
stack_trap "$LCTL set_param llite.*.statahead_batch_max=$batch_max" EXIT
$LCTL set_param llite.*.statahead_max=2048
$LCTL set_param llite.*.statahead_batch_max=1024
ls -l $dir
$LCTL get_param mdc.*.batch_stats
$LCTL get_param llite.*.statahead_*
}
run_test 123e "statahead with large wide striping"
test_123f() {
local max
local batch_max
local dir=$DIR/$tdir
$LCTL get_param -n mdc.*.connect_flags | grep -q batch_rpc ||
skip "Server does not support batch RPC"
mkdir $dir || error "mkdir $dir failed"
$LFS setstripe -C 1000 $dir || error "setstripe $dir failed"
stack_trap "rm -rf $dir"
touch $dir/$tfile.{0..200} || error "touch 200 files failed"
max=$($LCTL get_param -n llite.*.statahead_max | head -n 1)
batch_max=$($LCTL get_param -n llite.*.statahead_batch_max | head -n 1)
$LCTL set_param llite.*.statahead_max=64
$LCTL set_param llite.*.statahead_batch_max=64
ls -l $dir
lctl get_param mdc.*.batch_stats
lctl get_param llite.*.statahead_*
$LCTL set_param llite.*.statahead_max=$max
$LCTL set_param llite.*.statahead_batch_max=$batch_max
}
run_test 123f "Retry mechanism with large wide striping files"
test_123g() {
local dir=$DIR/$tdir
local num=1000
$LCTL get_param -n mdc.*.connect_flags | grep -q batch_rpc ||
skip "Server does not support batch RPC"
mkdir $dir || error "failed to mkdir $dir"
createmany -o $dir/$tfile $num || error "failed creatmany files"
cancel_lru_locks mdc
cancel_lru_locks osc
$LCTL set_param llite.*.statahead_stats=clear
$LCTL set_param mdc.*.batch_stats=clear
aheadmany -c stat -s 0 -e $num -b $tfile -d $dir ||
error "aheadmany $dir with $tfile failed"
wait_update_facet client "pgrep ll_sa" "" 35 ||
error "ll_sa thread is still running"
$LCTL get_param -n llite.*.statahead_stats
$LCTL get_param -n mdc.*.batch_stats
local count
count=$($LCTL get_param -n llite.*.statahead_stats |
awk '/hit.total:/ {print $2}')
echo "Hit total: $count"
# Hit ratio should be >= 75%
(( $count > num * 75 / 100 )) ||
error "hit total $count is be > 75% of $num"
}
run_test 123g "Test for stat-ahead advise"
test_123h_base() {
local dir=$DIR/$tdir
local cmd="touch $dir/$tfile.{$1}"
local fcnt=$2
stack_trap "rm -rf $dir"
mkdir -p $dir || error "failed to mkdir $dir"
eval $cmd
cancel_lru_locks mdc
$LCTL set_param llite.*.statahead_stats=clear
$LCTL set_param mdc.*.batch_stats=0
$LCTL set_param llite.*.statahead_max=1024
$LCTL set_param llite.*.statahead_batch_max=1024
lctl get_param -n llite.*.statahead_stats
du -a $dir > /dev/null
echo "Wait statahead thread (ll_sa_xxx) to exit..."
wait_update_facet client "pgrep ll_sa" "" 35 ||
error "ll_sa statahead thread does not quit in 35s"
$LCTL get_param -n llite.*.statahead_stats
$LCTL get_param -n mdc.*.batch_stats
local count=$($LCTL get_param -n llite.*.statahead_stats |
awk '/fname.total:/ {print $2}')
[ $count == 1 ] || error "File name pattern statahead not trigger"
count=$($LCTL get_param -n llite.*.statahead_stats |
awk '/hit.total:/ {print $2}')
# Hit ratio should be >= 75%
(( $count > fcnt * 75 / 100 )) ||
error "hit total is too low: $count"
rm -rf $dir || error "rm -rf $dir failed"
}
test_123h() {
local max
local batch_max
local enabled
$LCTL get_param -n mdc.*.connect_flags | grep -q batch_rpc ||
skip "Server does not support batch RPC"
max=$($LCTL get_param -n llite.*.statahead_max | head -n 1)
batch_max=$($LCTL get_param -n llite.*.statahead_batch_max | head -n 1)
enabled=$($LCTL get_param -n llite.*.enable_statahead_fname | head -n 1)
stack_trap "$LCTL set_param llite.*.statahead_max=$max"
stack_trap "$LCTL set_param llite.*.statahead_batch_max=$batch_max"
stack_trap "$LCTL set_param llite.*.enable_statahead_fname=$enabled"
$LCTL set_param llite.*.enable_statahead_fname=1
echo "Scan a directory with number regularized fname"
test_123h_base "0..10000" 10000
echo "Scan a directory with zeroed padding number regularized fname"
test_123h_base "000000..010000" 10000
}
run_test 123h "Verify statahead work with the fname pattern via du"
test_123i_base() {
local fmt=$1
local iocmd=$2
local dir=$DIR/$tdir
local cmd="createmany -m $fmt"
echo "Command:"
echo "- $cmd"
echo "- $iocmd"
stack_trap "unlinkmany $fmt"
mkdir -p $dir || error "failed to mkdir $dir"
eval $cmd
cancel_lru_locks mdc
$LCTL set_param llite.*.statahead_stats=clear
$LCTL set_param mdc.*.batch_stats=0
echo "statahead_stats (Pre):"
$LCTL get_param -n llite.*.statahead_stats
eval $iocmd || error "$iocmd failed"
echo "statahead_stats (Post):"
$LCTL get_param -n llite.*.statahead_stats
$LCTL get_param -n mdc.*.batch_stats
echo "Wait the statahead thread (ll_sa_xxx) to exit ..."
wait_update_facet client "pgrep ll_sa" "" 35 ||
error "ll_sa statahead thread does not quit in 35s"
$LCTL get_param -n llite.*.statahead_stats
$LCTL get_param -n mdc.*.batch_stats
local count=$($LCTL get_param -n llite.*.statahead_stats |
awk '/fname.total:/ {print $2}')
[ $count == 1 ] || error "File name pattern statahead not trigger"
count=$($LCTL get_param -n llite.*.statahead_stats |
awk '/hit.total:/ {print $2}')
# Hit ratio should be >= 75%
(( $count > 750 )) || error "hit total is too low: $count"
}
test_123i() {
local dir=$DIR/$tdir
local cnt=1000
local max
local batch_max
local enabled
local min
$LCTL get_param -n mdc.*.connect_flags | grep -q batch_rpc ||
skip "Server does not support batch RPC"
max=$($LCTL get_param -n llite.*.statahead_max | head -n 1)
batch_max=$($LCTL get_param -n llite.*.statahead_batch_max | head -n 1)
min=$($LCTL get_param -n llite.*.statahead_min | head -n 1)
enabled=$($LCTL get_param -n llite.*.enable_statahead_fname | head -n 1)
stack_trap "$LCTL set_param llite.*.statahead_max=$max"
stack_trap "$LCTL set_param llite.*.statahead_batch_max=$batch_max"
stack_trap "$LCTL set_param llite.*.statahead_min=$min"
stack_trap "$LCTL set_param llite.*.enable_statahead_fname=$enabled"
$LCTL set_param llite.*.statahead_max=1024
$LCTL set_param llite.*.statahead_batch_max=32
$LCTL set_param llite.*.statahead_min=64
$LCTL set_param llite.*.enable_statahead_fname=1
test_123i_base "$dir/$tfile.%06d $cnt" "ls $dir/* > /dev/null"
test_123i_base "$dir/$tfile $cnt" \
"aheadmany -c stat -N -s 0 -e $cnt -b $tfile -d $dir"
}
run_test 123i "Verify statahead work with the fname indexing pattern"
test_123j() {
(( $MDSCOUNT > 1 )) || skip_env "needs >= 2 MDTs"
$LCTL get_param -n mdc.*.connect_flags | grep -q batch_rpc ||
skip "Server does not support batch RPC"
local enabled
enabled=$($LCTL get_param -n llite.*.enable_statahead_fname | head -n 1)
stack_trap "$LCTL set_param llite.*.enable_statahead_fname=$enabled"
$LCTL set_param llite.*.enable_statahead_fname=1
stack_trap "rm -rf $DIR/${tdir}.{1..11}"
mkdir $DIR/${tdir}.{1..10} ||
error "failed to mkdir $DIR/${tdir}.{1..10}"
cancel_lru_locks mdc
for i in $(seq 1 10); do
stat $DIR/${tdir}.$i || error "failed to stat $DIR/${tdir}.$i"
done
stat $DIR/${tdir}.11
$LFS mkdir -i $((MDSCOUNT - 1)) -c 2 -H all_char $DIR/${tdir}.11 ||
error "failed to mkdir $DIR/${tdir}.11"
touch $DIR/${tdir}.11/$tfile ||
error "failed to create file $DIR/${tdir}.11/$tfile"
}
run_test 123j "-ENOENT error from batched statahead be handled correctly"
test_123k() {
MDTEST=${MDTEST:=$(which mdtest 2> /dev/null || true)}
[[ -n "$MDTEST" ]] || skip_env "mdtest not found"
$LCTL get_param -n mdc.*.connect_flags | grep -q batch_rpc ||
skip "Server does not support batch RPC"
local enabled
enabled=$($LCTL get_param -n llite.*.enable_statahead_fname | head -n 1)
stack_trap "$LCTL set_param llite.*.enable_statahead_fname=$enabled"
$LCTL set_param llite.*.enable_statahead_fname=1
local np=2
mpi_run -np $np $MDTEST -C -F -n 1000 -d $DIR/$tdir
cancel_lru_locks mdc
$LCTL set_param llite.*.statahead_stats=clear
$LCTL set_param mdc.*.batch_stats=0
mpi_run -np $np $MDTEST -T -F -n 1000 -d $DIR/$tdir
#umount_client $MOUNT || error "failed to umount client"
echo "Sleep to wait statahead thread (ll_sa_xxx) to exit ..."
wait_update_facet client "pgrep ll_sa" "" 35 ||
error "ll_sa thread is still running"
$LCTL get_param -n llite.*.statahead_stats
$LCTL get_param -n mdc.*.batch_stats
ps -el | grep ll_sa
local count=$($LCTL get_param -n llite.*.statahead_stats |
awk '/fname.total:/ {print $2}')
[ $count == $np ] || error "File name pattern statahead not trigger"
count=$($LCTL get_param -n llite.*.statahead_stats |
awk '/hit.total:/ {print $2}')
# Hit ratio should be >= 75%
[ $count -gt $((np * 1000 * 75 / 100)) ] ||
error "hit total is too low: $count"
}
run_test 123k "Verify statahead work with mdtest shared stat() mode"
test_123l() {
local dir=$DIR/$tdir
local cmd="touch $dir/$tfile.{000000..000100}"
$LCTL get_param -n mdc.*.connect_flags | grep -q batch_rpc ||
skip "Server does not support batch RPC"
stack_trap "rm -rf $dir"
mkdir -p $dir || error "failed to mkdir $dir"
eval $cmd
cancel_lru_locks mdc
$LCTL set_param llite.*.statahead_stats=clear
$LCTL set_param mdc.*.batch_stats=0
local max
local batch_max
local enabled
enabled=$($LCTL get_param -n llite.*.enable_statahead_fname | head -n 1)
max=$($LCTL get_param -n llite.*.statahead_max | head -n 1)
batch_max=$($LCTL get_param -n llite.*.statahead_batch_max | head -n 1)
stack_trap "$LCTL set_param llite.*.statahead_max=$max"
stack_trap "$LCTL set_param llite.*.statahead_batch_max=$batch_max"
stack_trap "$LCTL set_param llite.*.enable_statahead_fname=$enabled"
$LCTL set_param llite.*.enable_statahead_fname=1
$LCTL set_param llite.*.statahead_max=1024
$LCTL set_param llite.*.statahead_batch_max=32
$LCTL get_param -n llite.*.statahead_stats
#define OBD_FAIL_LLITE_STATAHEAD_PAUSE 0x1433
$LCTL set_param fail_loc=0x80001433 fail_val=35
ls $dir/* > /dev/null
$LCTL get_param -n llite.*.statahead_stats
$LCTL get_param -n mdc.*.batch_stats
echo "Sleep to wait the statahead thread (ll_sa_xxx) to exit ..."
wait_update_facet client "pgrep ll_sa" "" 35 ||
error "ll_sa thread is still running"
$LCTL get_param -n llite.*.statahead_stats
}
run_test 123l "Avoid panic when revalidate a local cached entry"
test_124a() {
[ $PARALLEL == "yes" ] && skip "skip parallel run"
$LCTL get_param -n mdc.*.connect_flags | grep -q lru_resize ||
skip_env "no lru resize on server"
local NR=2000
test_mkdir $DIR/$tdir
log "create $NR files at $DIR/$tdir"
createmany -o $DIR/$tdir/f $NR ||
error "failed to create $NR files in $DIR/$tdir"
cancel_lru_locks mdc
ls -l $DIR/$tdir > /dev/null
local NSDIR=""
local LRU_SIZE=0
for VALUE in $($LCTL get_param ldlm.namespaces.*mdc-*.lru_size); do
local PARAM=$(echo ${VALUE[0]} | cut -d "=" -f1)
LRU_SIZE=$($LCTL get_param -n $PARAM)
if [[ $LRU_SIZE -gt $(default_lru_size) ]]; then
NSDIR=$(echo $PARAM | cut -d "." -f1-3)
log "NSDIR=$NSDIR"
log "NS=$(basename $NSDIR)"
break
fi
done
if [[ -z "$NSDIR" || $LRU_SIZE -lt $(default_lru_size) ]]; then
skip "Not enough cached locks created!"
fi
log "LRU=$LRU_SIZE"
local SLEEP=30
# We know that lru resize allows one client to hold $LIMIT locks
# for 10h. After that locks begin to be killed by client.
local MAX_HRS=10
local LIMIT=$($LCTL get_param -n $NSDIR.pool.limit)
log "LIMIT=$LIMIT"
if [ $LIMIT -lt $LRU_SIZE ]; then
skip "Limit is too small $LIMIT"
fi
# Make LVF so higher that sleeping for $SLEEP is enough to _start_
# killing locks. Some time was spent for creating locks. This means
# that up to the moment of sleep finish we must have killed some of
# them (10-100 locks). This depends on how fast ther were created.
# Many of them were touched in almost the same moment and thus will
# be killed in groups.
local LVF=$(($MAX_HRS * 60 * 60 / $SLEEP * $LIMIT / $LRU_SIZE * 100))
# Use $LRU_SIZE_B here to take into account real number of locks
# created in the case of CMD, LRU_SIZE_B != $NR in most of cases
local LRU_SIZE_B=$LRU_SIZE
log "LVF=$LVF"
local OLD_LVF=$($LCTL get_param -n $NSDIR.pool.lock_volume_factor)
log "OLD_LVF=$OLD_LVF"
$LCTL set_param -n $NSDIR.pool.lock_volume_factor $LVF
# Let's make sure that we really have some margin. Client checks
# cached locks every 10 sec.
SLEEP=$((SLEEP+20))
log "Sleep ${SLEEP} sec"
local SEC=0
while ((SEC<$SLEEP)); do
echo -n "..."
sleep 5
SEC=$((SEC+5))
LRU_SIZE=$($LCTL get_param -n $NSDIR/lru_size)
echo -n "$LRU_SIZE"
done
echo ""
$LCTL set_param -n $NSDIR.pool.lock_volume_factor $OLD_LVF
local LRU_SIZE_A=$($LCTL get_param -n $NSDIR.lru_size)
[[ $LRU_SIZE_B -gt $LRU_SIZE_A ]] || {
error "No locks dropped in ${SLEEP}s. LRU size: $LRU_SIZE_A"
unlinkmany $DIR/$tdir/f $NR
return
}
log "Dropped "$((LRU_SIZE_B-LRU_SIZE_A))" locks in ${SLEEP}s"
log "unlink $NR files at $DIR/$tdir"
unlinkmany $DIR/$tdir/f $NR
}
run_test 124a "lru resize ======================================="
get_max_pool_limit()
{
local limit=$($LCTL get_param \
-n ldlm.namespaces.*-MDT0000-mdc-*.pool.limit)
local max=0
for l in $limit; do
if [[ $l -gt $max ]]; then
max=$l
fi
done
echo $max
}
test_124b() {
[ $PARALLEL == "yes" ] && skip "skip parallel run"
$LCTL get_param -n mdc.*.connect_flags | grep -q lru_resize ||
skip_env "no lru resize on server"
LIMIT=$(get_max_pool_limit)
NR=$(($(default_lru_size)*20))
if [[ $NR -gt $LIMIT ]]; then
log "Limit lock number by $LIMIT locks"
NR=$LIMIT
fi
IFree=$(mdsrate_inodes_available)
if [ $IFree -lt $NR ]; then
log "Limit lock number by $IFree inodes"
NR=$IFree
fi
lru_resize_disable mdc
stack_trap "lru_resize_enable mdc" EXIT
test_mkdir -p $DIR/$tdir/disable_lru_resize
createmany -o $DIR/$tdir/disable_lru_resize/f $NR
log "doing ls -la $DIR/$tdir/disable_lru_resize 3 times"
cancel_lru_locks mdc
stime=$SECONDS
PID=""
ls -la $DIR/$tdir/disable_lru_resize > /dev/null &
PID="$PID $!"
sleep 2
ls -la $DIR/$tdir/disable_lru_resize > /dev/null &
PID="$PID $!"
sleep 2
ls -la $DIR/$tdir/disable_lru_resize > /dev/null &
PID="$PID $!"
wait $PID
etime=$SECONDS
nolruresize_delta=$((etime-stime))
log "ls -la time: $nolruresize_delta seconds"
log "lru_size = $(lctl get_param -n ldlm.namespaces.*mdc*.lru_size)"
unlinkmany $DIR/$tdir/disable_lru_resize/f $NR
lru_resize_enable mdc
test_mkdir -p $DIR/$tdir/enable_lru_resize
stack_trap "unlinkmany $DIR/$tdir/enable_lru_resize/f $NR"
createmany -o $DIR/$tdir/enable_lru_resize/f $NR
log "doing ls -la $DIR/$tdir/enable_lru_resize 3 times"
cancel_lru_locks mdc
stime=$SECONDS
PID=""
ls -la $DIR/$tdir/enable_lru_resize > /dev/null &
PID="$PID $!"
sleep 2
ls -la $DIR/$tdir/enable_lru_resize > /dev/null &
PID="$PID $!"
sleep 2
ls -la $DIR/$tdir/enable_lru_resize > /dev/null &
PID="$PID $!"
wait $PID
etime=$SECONDS
lruresize_delta=$((etime-stime))
log "ls -la time: $lruresize_delta seconds"
log "lru_size = $(lctl get_param -n ldlm.namespaces.*mdc*.lru_size)"
if [ $lruresize_delta -gt $nolruresize_delta ]; then
log "ls -la is $(((lruresize_delta - $nolruresize_delta) * 100 / $nolruresize_delta))% slower with lru resize enabled"
elif [ $nolruresize_delta -gt $lruresize_delta ]; then
log "ls -la is $(((nolruresize_delta - $lruresize_delta) * 100 / $nolruresize_delta))% faster with lru resize enabled"
else
log "lru resize performs the same with no lru resize"
fi
}
run_test 124b "lru resize (performance test) ======================="
test_124c() {
[ $PARALLEL == "yes" ] && skip "skip parallel run"
$LCTL get_param -n mdc.*.connect_flags | grep -q lru_resize ||
skip_env "no lru resize on server"
# cache ununsed locks on client
local nr=100
cancel_lru_locks mdc
test_mkdir $DIR/$tdir
createmany -o $DIR/$tdir/f $nr ||
error "failed to create $nr files in $DIR/$tdir"
ls -l $DIR/$tdir > /dev/null
local nsdir="ldlm.namespaces.*-MDT0000-mdc-*"
local unused=$($LCTL get_param -n $nsdir.lock_unused_count)
local max_age=$($LCTL get_param -n $nsdir.lru_max_age)
local recalc_p=$($LCTL get_param -n $nsdir.pool.recalc_period)
echo "unused=$unused, max_age=$max_age, recalc_p=$recalc_p"
# set lru_max_age to 1 sec
$LCTL set_param $nsdir.lru_max_age=1000 # milliseconds
echo "sleep $((recalc_p * 2)) seconds..."
sleep $((recalc_p * 2))
# There's threshold of priv_list eviction each round, hence
# priv_count here may larger than 0. And there maybe lru shrink
# invoked just after $priv_count and before $remaining which
# could cause $remaining smaller than 0
local priv_count=$($LCTL get_param -n $nsdir.lock_unused_priv_count)
local remaining=$($LCTL get_param -n $nsdir.lock_unused_count)
local normal_count=$((remaining - priv_count))
# restore lru_max_age
$LCTL set_param -n $nsdir.lru_max_age $max_age
(( normal_count <= 0 )) ||
error "$normal_count locks are not canceled"
unlinkmany $DIR/$tdir/f $nr
}
run_test 124c "LRUR cancel very aged locks"
test_124d() {
[ $PARALLEL == "yes" ] && skip "skip parallel run"
$LCTL get_param -n mdc.*.connect_flags | grep -q lru_resize ||
skip_env "no lru resize on server"
# cache unused locks on client
local nr=100
lru_resize_disable mdc
cancel_lru_locks mdc
# asynchronous object destroy at MDT could cause bl ast to client
test_mkdir $DIR/$tdir
stack_trap "unlinkmany $DIR/$tdir/f $nr" EXIT
createmany -o $DIR/$tdir/f $nr ||
error "failed to create $nr files in $DIR/$tdir"
ls -l $DIR/$tdir > /dev/null
local nsdir="ldlm.namespaces.*-MDT0000-mdc-*"
local unused=$($LCTL get_param -n $nsdir.lock_unused_count)
local max_age=$($LCTL get_param -n $nsdir.lru_max_age)
local recalc_p=$($LCTL get_param -n $nsdir.pool.recalc_period)
echo "unused=$unused, max_age=$max_age, recalc_p=$recalc_p"
# set lru_max_age to 1 sec
$LCTL set_param $nsdir.lru_max_age=1000 # milliseconds
stack_trap "$LCTL set_param -n $nsdir.lru_max_age $max_age" EXIT
echo "sleep $((recalc_p * 2)) seconds..."
sleep $((recalc_p * 2))
local priv_count=$($LCTL get_param -n $nsdir.lock_unused_priv_count)
local remaining=$($LCTL get_param -n $nsdir.lock_unused_count)
local normal_count=$((remaining - priv_count))
(( normal_count <= 0 )) ||
error "$normal_count locks are not canceled"
}
run_test 124d "cancel very aged locks if lru-resize disabled"
test_124e() {
[[ $PARALLEL != "yes" ]] || skip "skip parallel run"
local nsdir="ldlm.namespaces.*-MDT0000-mdc-*"
local max_unused=$(default_lru_size)
echo "max_unused=$max_unused"
lru_resize_disable mdc $max_unused
$LCTL set_param $nsdir.lru_priv_score_threshold=1 ||
error "fail to set lru_priv_score_threshold"
stack_trap "$LCTL set_param $nsdir.lru_priv_score_threshold=1"
# cache unused locks on client
local nr=$((max_unused * 2))
mkdir_on_mdt0 $DIR/$tdir
mkdir_on_mdt0 $DIR/$tdir/empty $DIR/$tdir/files
stack_trap "unlinkmany $DIR/$tdir/files/f $nr"
createmany -i0 -o $DIR/$tdir/files/f $nr ||
error "failed to create $nr files in $DIR/$tdir/files"
# clean cache
cancel_lru_locks mdc
sleep 5
# dir lock should be placed in priv list
for ((i = 0; i < 10; i++)); do
stat $DIR/$tdir/empty > /dev/null
done
# try to overflow lru cache
ls -l $DIR/$tdir/files > /dev/null
local unused_cnt=$($LCTL get_param -n $nsdir.lock_unused_count)
local priv_cnt=$($LCTL get_param -n $nsdir.lock_unused_priv_count)
local priv_hits=$($LCTL get_param -n $nsdir.lock_lru_priv_hits)
(( priv_cnt > 0 )) || error "priv_cnt is 0"
(( priv_hits > 0 )) || error "priv_hits is 0"
# lock should been in priv list
stat $DIR/$tdir/empty > /dev/null
sleep 2
local new_unused_cnt=$($LCTL get_param -n $nsdir.lock_unused_count)
local new_priv_cnt=$($LCTL get_param -n $nsdir.lock_unused_priv_count)
local new_priv_hits=$($LCTL get_param -n $nsdir.lock_lru_priv_hits)
(( new_unused_cnt == unused_cnt )) ||
error "new_cnt($new_unused_cnt) != unused_cnt($unused_cnt)"
(( priv_cnt == new_priv_cnt )) ||
error "priv_cnt($priv_cnt) != new_priv_cnt($new_priv_cnt)"
(( new_priv_hits > priv_hits )) ||
error "new_priv_hits($new_priv_hits) <= priv_hits($priv_hits)"
}
run_test 124e "LFRU keep priv locks from eviction"
test_124f() {
[[ $PARALLEL != "yes" ]] || skip "skip parallel run"
local nsdir="ldlm.namespaces.*-MDT0000-mdc-*"
local max_unused=$(default_lru_size)
echo "max_unused=$max_unused"
lru_resize_disable mdc $max_unused
stack_trap "$LCTL set_param $nsdir.lru_priv_score_threshold=1"
# threshold inc
local priv_thres=1
$LCTL set_param -n $nsdir.lru_priv_score_threshold=$priv_thres ||
error "fail to set lru_priv_score_threshold"
# cache unused locks on client
local nr=$((max_unused * 4))
mkdir_on_mdt0 $DIR/$tdir
mkdir_on_mdt0 $DIR/$tdir/files
stack_trap "unlinkmany $DIR/$tdir/files/f $nr"
createmany -i0 -o $DIR/$tdir/files/f $nr ||
error "failed to create $nr files in $DIR/$tdir/files"
# clean cache
cancel_lru_locks mdc
sleep 5
# try to overflow lru cache
ls -l $DIR/$tdir/files > /dev/null
local inc_priv_thres=$(
$LCTL get_param -n $nsdir.lru_priv_score_threshold
)
(( inc_priv_thres > priv_thres )) ||
error "new_thres($inc_priv_thres) <= threshold($priv_thres)"
# threshold dec
priv_thres=7
$LCTL set_param $nsdir.lru_priv_score_threshold=$priv_thres ||
error "failed to set lru_priv_score_threshold to $priv_thres"
# clean cache
cancel_lru_locks mdc
sleep 5
# try to overflow lru cache
ls -l $DIR/$tdir/files > /dev/null
local dec_priv_thres=$(
$LCTL get_param -n $nsdir.lru_priv_score_threshold
)
(( dec_priv_thres < priv_thres )) ||
error "new_thres($dec_priv_thres)>=priv_thres($priv_thres)"
}
run_test 124f "LFRU priv threshold inc/dec adjustment"
test_124g_run() {
local client_nid="$1"
local base_dir="$2/$4"
local lru_size=$3
local policy="$4"
local -n ref_enq_res=$5
local cold_nr=$lru_size
local nr=$(( lru_size / 4 ))
# clear the existing cache state before switching policy
cancel_lru_locks mdc
sleep 2
local nsdir="ldlm.namespaces.*-MDT0000-mdc-*"
$LCTL set_param $nsdir.lock_cache_policy="$policy" ||
error "failed to set lock_cache_policy to $policy"
mkdir_on_mdt0 "$base_dir"
mkdir_on_mdt0 "$base_dir"/cold
createmany -i0 -o "$base_dir"/cold/f $cold_nr ||
error "failed to create $cold_nr files in $base_dir/cold"
createmany -i0 -o "$base_dir"/f $nr ||
error "failed to create $nr files"
# record existing enqueue stats
cancel_lru_locks mdc
sleep 2
$LCTL get_param $nsdir.lock_unused_count
local nsservdir="mdt.*-MDT0000.exports.'$client_nid'.ldlm_stats"
# res format as:
# ldlm_enqueue 1022 samples [reqs]
local o_enq=$(do_facet mds1 $LCTL get_param "$nsservdir" |\
awk '/ldlm_enqueue/ {print $2}') ||
error "o_enq failed to get ldlm_enqueue count"
local prev_enq=$o_enq
do_facet mds1 $LCTL get_param "$nsservdir"
echo "start stat & ls ops..."
for ((i = 0; i < 10; i++)); do
for ((j = 0; j < nr; j++)); do
for ((cnt = 0; cnt < 10; cnt++)); do
stat "$base_dir"/f$j > /dev/null
done
done
local before_priv_cnt=$($LCTL get_param -n $nsdir.lock_unused_priv_count)
for ((c = 0; c < cold_nr; c++)); do
stat "$base_dir"/cold/f$c > /dev/null
done
# facilitate debugging output
local priv_cnt=$($LCTL get_param -n $nsdir.lock_unused_priv_count)
local temp_enq=$(do_facet mds1 $LCTL get_param "$nsservdir" |\
awk '/ldlm_enqueue/ {print $2}') ||
error "temp_enq failed to get ldlm_enqueue count"
echo "round$i done, before_priv_cnt=$before_priv_cnt, priv_cnt=$priv_cnt, inc_enq=$((temp_enq - prev_enq))"
prev_enq=$temp_enq
done
local n_enq=$(do_facet mds1 $LCTL get_param "$nsservdir" |\
awk '/ldlm_enqueue/ {print $2}') ||
error "n_enq failed to get ldlm_enqueue count"
ref_enq_res=$(( n_enq - o_enq ))
echo "base_dir=$base_dir, policy=$policy"
echo "o_enq=$o_enq, n_enq=$n_enq, ref_enq_res=$ref_enq_res"
rm -rf "$base_dir"; wait_delete_completed
}
test_124g() {
[[ $PARALLEL != "yes" ]] || skip "skip parallel run"
(( $MDS1_VERSION >= $(version_code 2.17.50.63) )) ||
skip "Need MDS with LFRU support (LU-11509, >= 2.17.50.63)"
local nsdir="ldlm.namespaces.*-MDT0000-mdc-*"
local lru_size=$(default_lru_size)
lru_resize_disable mdc $lru_size
local save="$TMP/$TESTSUITE-$TESTNAME.parameters"
save_lustre_params client "$nsdir.lock_cache_policy" > $save
save_lustre_params client "llite.*.enable_statahead_fname" >> $save
stack_trap "restore_lustre_params < $save; rm -f $save"
# disable statehead fname to avoid additional threads' execution
# which may affect the test result due to its random behavior.
$LCTL set_param llite.*.enable_statahead_fname=0
local cli_nid="0@lo"
if remote_mds; then
cli_nid=$($LCTL list_nids | grep -v "@lo" | head -1)
fi
mkdir_on_mdt0 $DIR/$tdir
local enq_priv_enabled
local enq_priv_disable
# test with lfru
test_124g_run "$cli_nid" "$DIR/$tdir" $lru_size "LFRU" enq_priv_enabled
# test with lru
test_124g_run "$cli_nid" "$DIR/$tdir" $lru_size "LRU" enq_priv_disable
echo ">> lfru_enqueue=$enq_priv_enabled, lru_enqueue=$enq_priv_disable"
(( enq_priv_enabled <= enq_priv_disable )) ||
error "lfru enqueue $enq_priv_enabled > lru $enq_priv_disable"
}
run_test 124g "LFRU performance test"
test_125() { # 13358
$LCTL get_param -n llite.*.client_type | grep -q local ||
skip "must run as local client"
$LCTL get_param -n mdc.*-mdc-*.connect_flags | grep -q acl ||
skip_env "must have acl enabled"
[ -z "$(which setfacl)" ] && skip_env "must have setfacl tool"
id $USER0 || skip_env "missing user $USER0"
test_mkdir $DIR/$tdir
$LFS setstripe -S 65536 -c -1 $DIR/$tdir || error "setstripe failed"
setfacl -R -m u:$USER0:rwx $DIR/$tdir ||
error "setfacl $DIR/$tdir failed"
ls -ld $DIR/$tdir || error "cannot access $DIR/$tdir"
}
run_test 125 "don't return EPROTO when a dir has a non-default striping and ACLs"
test_126() { # bug 12829/13455
$GSS && skip_env "must run as gss disabled"
$LCTL get_param -n llite.*.client_type | grep -q local ||
skip "must run as local client"
[ "$UID" != 0 ] && skip "must run as root, not UID $UID"
$RUNAS -u 0 -g 1 touch $DIR/$tfile || error "touch failed"
gid=`ls -n $DIR/$tfile | awk '{print $4}'`
rm -f $DIR/$tfile
[ $gid -eq "1" ] || error "gid is set to" $gid "instead of 1"
}
run_test 126 "check that the fsgid provided by the client is taken into account"
test_127a() { # bug 15521
[ $PARALLEL == "yes" ] && skip "skip parallel run"
local name count samp unit min max sum sumsq
local tmpfile=$TMP/$tfile.tmp
# enable stats header if it is disabled
$LCTL set_param enable_stats_header=1
$LFS setstripe -i 0 -c 1 $DIR/$tfile || error "setstripe failed"
echo "stats before reset"
stack_trap "rm -f $DIR/$tfile $tmpfile"
local now=$(date +%s)
$LCTL get_param osc.*.stats | tee $tmpfile
local snapshot_time=$(awk '/snapshot_time/ { print $2; exit }' $tmpfile)
local start_time=$(awk '/start_time/ { print $2; exit }' $tmpfile)
local elapsed=$(awk '/elapsed_time/ { print $2; exit }' $tmpfile)
local uptime=$(awk '{ print $1 }' /proc/uptime)
# snapshot_time should match POSIX epoch time, allow some delta for VMs
(( ${snapshot_time%\.*} >= $now - 5 &&
${snapshot_time%\.*} <= $now + 5 )) ||
error "snapshot_time=$snapshot_time != now=$now"
# elapsed _should_ be from mount, but at least less than uptime
(( ${elapsed%\.*} < ${uptime%\.*} )) ||
error "elapsed=$elapsed > uptime=$uptime"
(( ${snapshot_time%\.*} - ${start_time%\.*} >= ${elapsed%\.*} - 2 &&
${snapshot_time%\.*} - ${start_time%\.*} <= ${elapsed%\.*} + 2 )) ||
error "elapsed=$elapsed != $snapshot_time - $start_time"
$LCTL set_param osc.*.stats=0
local reset=$(date +%s)
local fsize=$((2048 * 1024))
dd if=/dev/zero of=$DIR/$tfile bs=$fsize count=1
cancel_lru_locks osc
dd if=$DIR/$tfile of=/dev/null bs=$fsize
now=$(date +%s)
$LCTL get_param osc.*0000-osc-*.stats > $tmpfile
while read name count samp unit min max sum sumsq; do
[[ "$samp" == "samples" ]] || continue
echo "got name=$name count=$count unit=$unit min=$min max=$max"
[ ! $min ] && error "Missing min value for $name proc entry"
eval $name=$count || error "Wrong proc format"
case $name in
read_bytes|write_bytes)
[[ "$unit" =~ "bytes" ]] ||
error "unit is not 'bytes': $unit"
(( $min >= 4096 )) || error "min is too small: $min"
(( $min <= $fsize )) || error "min is too big: $min"
(( $max >= 4096 )) || error "max is too small: $max"
(( $max <= $fsize )) || error "max is too big: $max"
(( $sum == $fsize )) || error "sum is wrong: $sum"
(( $sumsq >= ($fsize / 4096) * (4096 * 4096) )) ||
error "sumsquare is too small: $sumsq"
(( $sumsq <= $fsize * $fsize )) ||
error "sumsquare is too big: $sumsq"
;;
ost_read|ost_write)
[[ "$unit" =~ "usec" ]] ||
error "unit is not 'usec': $unit"
;;
*) ;;
esac
done < $tmpfile
#check that we actually got some stats
[ "$read_bytes" ] || error "Missing read_bytes stats"
[ "$write_bytes" ] || error "Missing write_bytes stats"
[ "$read_bytes" != 0 ] || error "no read done"
[ "$write_bytes" != 0 ] || error "no write done"
snapshot_time=$(awk '/snapshot_time/ { print $2; exit }' $tmpfile)
start_time=$(awk '/start_time/ { print $2; exit }' $tmpfile)
elapsed=$(awk '/elapsed_time/ { print $2; exit }' $tmpfile)
# snapshot_time should match POSIX epoch time, allow some delta for VMs
(( ${snapshot_time%\.*} >= $now - 5 &&
${snapshot_time%\.*} <= $now + 5 )) ||
error "reset snapshot_time=$snapshot_time != now=$now"
# elapsed should be from time of stats reset
(( ${elapsed%\.*} >= $now - $reset - 2 &&
${elapsed%\.*} <= $now - $reset + 2 )) ||
error "reset elapsed=$elapsed > $now - $reset"
(( ${snapshot_time%\.*} - ${start_time%\.*} >= ${elapsed%\.*} - 2 &&
${snapshot_time%\.*} - ${start_time%\.*} <= ${elapsed%\.*} + 2 )) ||
error "reset elapsed=$elapsed != $snapshot_time - $start_time"
}
run_test 127a "verify the client stats are sane"
test_127b() { # bug LU-333
[ $PARALLEL == "yes" ] && skip "skip parallel run"
local name count samp unit min max sum sumsq
echo "stats before reset"
$LCTL get_param llite.*.stats
$LCTL set_param llite.*.stats=0
# perform 2 reads and writes so MAX is different from SUM.
dd if=/dev/zero of=$DIR/$tfile bs=$PAGE_SIZE count=1
dd if=/dev/zero of=$DIR/$tfile bs=$PAGE_SIZE count=1
cancel_lru_locks osc
dd if=$DIR/$tfile of=/dev/null bs=$PAGE_SIZE count=1
dd if=$DIR/$tfile of=/dev/null bs=$PAGE_SIZE count=1
$LCTL get_param llite.*.stats | grep samples > $TMP/$tfile.tmp
stack_trap "rm -f $TMP/$tfile.tmp"
while read name count samp unit min max sum sumsq; do
echo "got name=$name count=$count unit=$unit min=$min max=$max"
eval $name=$count || error "Wrong proc format"
case $name in
read_bytes|write_bytes)
[[ "$unit" =~ "bytes" ]] ||
error "unit is not 'bytes': $unit"
(( $count == 2 )) || error "count is not 2: $count"
(( $min == $PAGE_SIZE )) ||
error "min is not $PAGE_SIZE: $min"
(( $max == $PAGE_SIZE )) ||
error "max is not $PAGE_SIZE: $max"
(( $sum == $PAGE_SIZE * 2 )) ||
error "sum is not $((PAGE_SIZE * 2)): $sum"
;;
read|write)
[[ "$unit" =~ "usec" ]] ||
error "unit is not 'usec': $unit"
;;
*) ;;
esac
done < $TMP/$tfile.tmp
#check that we actually got some stats
[ "$read_bytes" ] || error "Missing read_bytes stats"
[ "$write_bytes" ] || error "Missing write_bytes stats"
[ "$read_bytes" != 0 ] || error "no read done"
[ "$write_bytes" != 0 ] || error "no write done"
}
run_test 127b "verify the llite client stats are sane"
test_127c() { # LU-12394
[ "$OSTCOUNT" -lt "2" ] && skip_env "needs >= 2 OSTs"
local size
local bsize
local reads
local writes
local count
$LCTL set_param llite.*.extents_stats=1
stack_trap "$LCTL set_param llite.*.extents_stats=0" EXIT
# Use two stripes so there is enough space in default config
$LFS setstripe -c 2 $DIR/$tfile
# Extent stats start at 0-4K and go in power of two buckets
# LL_HIST_START = 12 --> 2^12 = 4K
# We do 3K*2^i, so 3K, 6K, 12K, 24K... hitting each bucket.
# We do not do buckets larger than 64 MiB to avoid ENOSPC issues on
# small configs
for size in 3K 6K 12K 24K 48K 96K 192K 384K 768K 1536K 3M 6M 12M 24M 48M;
do
# Write and read, 2x each, second time at a non-zero offset
dd if=/dev/zero of=$DIR/$tfile bs=$size count=1
dd if=/dev/zero of=$DIR/$tfile bs=$size count=1 seek=10
dd if=$DIR/$tfile of=/dev/null bs=$size count=1
dd if=$DIR/$tfile of=/dev/null bs=$size count=1 seek=10
rm -f $DIR/$tfile
done
$LCTL get_param llite.*.extents_stats
count=2
for bsize in 4K 8K 16K 32K 64K 128K 256K 512K 1M 2M 4M 8M 16M 32M 64M;
do
local bucket=$($LCTL get_param -n llite.*.extents_stats |
grep -m 1 $bsize)
reads=$(echo $bucket | awk '{print $5}')
writes=$(echo $bucket | awk '{print $9}')
[ "$reads" -eq $count ] ||
error "$reads reads in < $bsize bucket, expect $count"
[ "$writes" -eq $count ] ||
error "$writes writes in < $bsize bucket, expect $count"
done
# Test mmap write and read
$LCTL set_param llite.*.extents_stats=c
size=512
dd if=/dev/zero of=$DIR/$tfile bs=${size}K count=1
$MULTIOP $DIR/$tfile OSMRUc || error "$MULTIOP $DIR/$tfile failed"
$MULTIOP $DIR/$tfile OSMWUc || error "$MULTIOP $DIR/$tfile failed"
$LCTL get_param llite.*.extents_stats
count=$(((size*1024) / PAGE_SIZE))
bsize=$((2 * PAGE_SIZE / 1024))K
bucket=$($LCTL get_param -n llite.*.extents_stats |
grep -m 1 $bsize)
reads=$(echo $bucket | awk '{print $5}')
writes=$(echo $bucket | awk '{print $9}')
# mmap writes fault in the page first, creating an additonal read
[ "$reads" -eq $((2 * count)) ] ||
error "$reads reads in < $bsize bucket, expect $count"
[ "$writes" -eq $count ] ||
error "$writes writes in < $bsize bucket, expect $count"
}
run_test 127c "test llite extent stats with regular & mmap i/o"
test_127d() {
[[ $PARALLEL == "yes" ]] && skip "skip parallel run"
(( CLIENT_VERSION >= $(version_code 2.16.56.105) )) ||
skip "need client > 2.16.56.105 for osc latency stats"
local size=10
local file=$DIR/$tfile
# Clear RPC stats
$LCTL set_param osc.*.rpc_stats=clear
# Generate write I/O
dd if=/dev/urandom of=$file bs=1M count=$size conv=fsync ||
error "dd write failed"
stack_trap "rm -f $file"
# Drop caches to ensure read goes to server
cancel_lru_locks osc
# Generate read I/O
dd if=$file of=/dev/null bs=1M || error "dd read failed"
# Get RPC stats for debugging
local rpc_stats=$($LCTL get_param -n osc.*OST*.rpc_stats)
echo "RPC stats after I/O:"
echo "$rpc_stats"
# Check for read RPC latency stats
local read_latency=$(echo "$rpc_stats" |
awk '/RPC latency/,/^$/ && $1 ~ /[0-9]+:/ && $2 > 0 \
{ found = 1; exit }
END { print (found ? "1" : "0") }')
# Check for write RPC latency stats
local write_latency=$(echo "$rpc_stats" |
awk '/RPC latency/,/^$/ && $1 ~ /[0-9]+:/ && $6 > 0 \
{ found = 1; exit }
END { print (found ? "1" : "0") }')
echo "Read latency found: $read_latency"
echo "Write latency found: $write_latency"
(( read_latency == 1 )) || error "No read RPC latency stats"
(( write_latency == 1 )) || error "No write RPC latency stats"
}
run_test 127d "OSC RPC latency histograms for read and write latency"
test_127_io_latency_test() {
[[ $PARALLEL == "yes" ]] && skip "skip parallel run"
local facet=$1
local io_latency_param=$2
local dev=osc
[[ $facet == *ost* ]] && {
# disable cache to have read IO
local rcc="osd-*.$FSNAME-OST0000.read_cache_enable"
local wcc="osd-*.$FSNAME-OST0000.writethrough_cache_enable"
local rsaved=$(do_facet $facet $LCTL get_param -n $rcc)
local wsaved=$(do_facet $facet $LCTL get_param -n $wcc)
do_facet $facet $LCTL set_param $rcc=0 $wcc=0
stack_trap "do_facet $facet $LCTL set_param $rcc=$rsaved $wcc=$wsaved"
}
[[ $io_latency_param =~ "md[ct]" ]] && dev="mdc"
# avoid leftovers after preceding tests
cancel_lru_locks osc
sleep 1
# Clear RPC stats
do_facet $facet $LCTL set_param $io_latency_param=clear
[[ $facet == *ost* ]] &&
do_facet $facet $LCTL set_param osd-*.*OST*.stats=clear
# Generate I/O with different sizes to populate multiple histograms
local file=$DIR/$tdir/$tfile
local count=10
local sizes=()
local mppr=($($LCTL get_param -n $dev.*.max_pages_per_rpc))
for size in 4 64 1024; do
(( size >= $PAGE_SIZE / 1024 &&
size <= $mppr * $PAGE_SIZE / 1024 )) &&
sizes+=(${size}K)
done
$LFS mkdir -i 0 $DIR/$tdir
stack_trap "rm -f ${file}_*"
for size in ${sizes[@]}; do
echo "Generating I/O with size $size"
# Write with current size
[[ $dev == "mdc" ]] &&
$LFS setstripe -E 4M -L mdt ${file}_${size} ||
$LFS setstripe -c 1 -i 0 ${file}_${size}
dd if=/dev/urandom of=${file}_${size} bs=$size count=$count \
oflag=direct ||
error "dd write failed for size $size"
cancel_lru_locks osc
# Read with current size
dd if=${file}_${size} of=/dev/null bs=$size count=$count \
iflag=direct ||
error "dd read failed for size $size"
done
sleep 5
# Get client IO latency stats for debugging
local io_latency_stats=$(do_facet $facet $LCTL get_param $io_latency_param)
echo "$io_latency_param stats after I/O:"
echo "$io_latency_stats"
# Check that we have entries for different sizes
for size in ${sizes[@]}; do
echo "$io_latency_stats" | grep -q _$size || {
$LCTL get_param osc.*.{io_latency,rpc}_stats
[[ $facet == *ost* ]] &&
do_facet ost1 $LCTL get_param osd-*.*.{brw,io_latency}_stats &&
do_facet ost1 $LCTL get_param osd-*.*OST*.stats
error "No stats found for size $size"
}
done
# sometimes stray writes are seen, filter to expected sizes
local check_counts=$(echo "$io_latency_stats" |
grep -E "$(sed -e 's/ /K|_/g' <<<${sizes[*]})" |
awk -v count="$count" '/K: / {sum = 0
while (match($0, /[0-9]+us: ([0-9]+),/, arr)) {
sum += arr[1]
$0 = substr($0, RSTART + RLENGTH)
}
{ if (sum < count) print "Expected "count" entries, found "sum}
}')
[[ -z "$check_counts" ]] || error "$check_counts"
local has_read=$(echo "$io_latency_stats" | grep -c "rd_")
local has_write=$(echo "$io_latency_stats" | grep -c "wr_")
(( has_read >= ${#sizes[@]} )) ||
error "Expected ${#sizes[@]} read stats, found $has_read"
(( has_write >= ${#sizes[@]} )) ||
error "Expected ${#sizes[@]} write stats, found $has_write"
}
test_127e() {
test_127_io_latency_test client osc.$FSNAME-OST0000-*.io_latency_stats
}
run_test 127e "client IO latency histograms by size"
test_127f() {
(( $OST1_VERSION >= $(version_code v2_17_50-123-gc688e33f11) )) ||
skip "need OST > 2.17.50.123 for OST IO latency stats"
[[ $(facet_fstype ost1) == ldiskfs ]] || skip "ldiskfs only"
test_127_io_latency_test ost1 osd-*.$FSNAME-OST0000.io_latency_stats
}
run_test 127f "OST IO latency histograms by size"
test_128() { # bug 15212
touch $DIR/$tfile
$LFS 2>&1 <<-EOF | tee $TMP/$tfile.log
find $DIR/$tfile
find $DIR/$tfile
EOF
result=$(grep error $TMP/$tfile.log)
rm -f $DIR/$tfile $TMP/$tfile.log
[ -z "$result" ] ||
error "consecutive find's under interactive lfs failed"
}
run_test 128 "interactive lfs for 2 consecutive find's"
set_dir_limits () {
local mntdev
local canondev
local node
local ldproc=/proc/fs/ldiskfs
local facets=$(get_facets MDS)
for facet in ${facets//,/ }; do
canondev=$(ldiskfs_canon \
*.$(convert_facet2label $facet).mntdev $facet)
do_facet $facet "test -e $ldproc/$canondev/max_dir_size" ||
ldproc=/sys/fs/ldiskfs
do_facet $facet "echo $1 >$ldproc/$canondev/max_dir_size"
do_facet $facet "echo $2 >$ldproc/$canondev/warning_dir_size"
done
}
check_mds_dmesg() {
local facets=$(get_facets MDS)
for facet in ${facets//,/ }; do
do_facet $facet "dmesg | tail -3 | grep $1" && return 0
done
return 1
}
test_129() {
[ $PARALLEL == "yes" ] && skip "skip parallel run"
[[ $MDS1_VERSION -ge $(version_code 2.5.56) ]] ||
skip "Need MDS version with at least 2.5.56"
if [ "$mds1_FSTYPE" != ldiskfs ]; then
skip_env "ldiskfs only test"
fi
remote_mds_nodsh && skip "remote MDS with nodsh"
local ENOSPC=28
local has_warning=false
rm -rf $DIR/$tdir
mkdir -p $DIR/$tdir
# block size of mds1
local maxsize=$(($($LCTL get_param -n mdc.*MDT0000*.blocksize) * 8))
set_dir_limits $maxsize $((maxsize * 6 / 8))
stack_trap "set_dir_limits 0 0"
stack_trap "unlinkmany $DIR/$tdir/file_base_ 2000 || true"
local dirsize=$(stat -c%s "$DIR/$tdir")
local nfiles=0
while (( $dirsize <= $maxsize )); do
$MCREATE $DIR/$tdir/file_base_$nfiles
rc=$?
# check two errors:
# ENOSPC for ext4 max_dir_size, which has been used since
# kernel v3.6-rc1-8-gdf981d03ee, lustre v2_4_50_0-79-gaed82035c0
if (( rc == ENOSPC )); then
set_dir_limits 0 0
echo "rc=$rc returned as expected after $nfiles files"
createmany -o $DIR/$tdir/file_extra_$nfiles. 5 ||
error "create failed w/o dir size limit"
# messages may be rate limited if test is run repeatedly
check_mds_dmesg '"is approaching max"' ||
echo "warning message should be output"
check_mds_dmesg '"has reached max"' ||
echo "reached message should be output"
dirsize=$(stat -c%s "$DIR/$tdir")
[[ $dirsize -ge $maxsize ]] && return 0
error "dirsize $dirsize < $maxsize after $nfiles files"
elif (( rc != 0 )); then
break
fi
nfiles=$((nfiles + 1))
dirsize=$(stat -c%s "$DIR/$tdir")
done
error "rc=$rc, size=$dirsize/$maxsize, mdt=$MDSCOUNT, nfiles=$nfiles"
}
run_test 129 "test directory size limit ========================"
OLDIFS="$IFS"
test_130a() {
local filefrag_op=$(filefrag -l 2>&1 | grep "invalid option")
[[ -z "$filefrag_op" ]] || skip_env "filefrag missing logical ordering"
local fm_file=$DIR/$tfile
$LFS setstripe -S 65536 -c 1 $fm_file || error "setstripe on $fm_file"
stack_trap "rm -f $fm_file"
dd if=/dev/zero of=$fm_file bs=65536 count=1 ||
error "dd failed for $fm_file"
# LU-1795: test filefrag/FIEMAP once, even if unsupported on ZFS
filefrag -ves $fm_file
local rc=$?
[[ "$ost1_FSTYPE" != "zfs" ]] ||
skip "LU-1941: FIEMAP unimplemented on ZFS"
(( $rc == 0 )) || error "filefrag $fm_file failed"
filefrag_op=$(filefrag -ve -k $fm_file |
sed -n '/ext:/,/found/{/ext:/d; /found/d; p}')
local lun=$($LFS getstripe -i $fm_file)
local start_blk=$(echo $filefrag_op | cut -d: -f2 | cut -d. -f1)
local tot_len=0
while read line; do
local frag_lun=$(echo $line | cut -d: -f5)
local ext_len=$(echo $line | cut -d: -f4)
(( $frag_lun == $lun )) ||
error "OST$lun was reported on OST$frag_lun"
(( tot_len += ext_len ))
done <<< "$filefrag_op"
(( start_blk == 0 )) ||
error "OST$last_lun start_blk $start_blk != 0"
(( tot_len == 64 )) ||
error "OST$last_lun $tot_len != 64"
echo "FIEMAP on single striped file succeeded"
}
run_test 130a "FIEMAP (1-stripe file)"
test_130b() {
(( $OSTCOUNT >= 2 )) || skip "needs >= 2 OSTs"
local filefrag_op=$(filefrag -l 2>&1 | grep "invalid option")
[[ -z "$filefrag_op" ]] || skip_env "filefrag missing logical ordering"
[[ "$ost1_FSTYPE" != "zfs" ]] ||
skip "LU-1941: FIEMAP unimplemented on ZFS"
local fm_file=$DIR/$tfile
$LFS setstripe -S 65536 -c $OSTCOUNT $fm_file ||
error "setstripe on $fm_file"
stack_trap "rm -f $fm_file"
local actual_stripe_count=$($LFS getstripe -c $fm_file)
dd if=/dev/zero of=$fm_file bs=1M count=$actual_stripe_count ||
error "dd failed on $fm_file"
filefrag -ves $fm_file || error "filefrag $fm_file failed"
filefrag_op=$(filefrag -ve -k $fm_file |
sed -n '/ext:/,/found/{/ext:/d; /found/d; p}')
local last_lun=$(echo $filefrag_op | cut -d: -f5 |
sed -e 's/^[ \t]*/0x/' | sed -e 's/0x0x/0x/')
local tot_len=0
local num_luns=1
while read line; do
local frag_lun=$(echo $line | cut -d: -f5 |
sed -e 's/^[ \t]*/0x/' | sed -e 's/0x0x/0x/')
local ext_len=$(echo $line | cut -d: -f4)
if (( $frag_lun != $last_lun )); then
(( tot_len == 1024 )) ||
error "OST$last_lun $tot_len != 1024"
(( num_luns += 1 ))
tot_len=0
fi
(( tot_len += ext_len ))
last_lun=$frag_lun
done <<< "$filefrag_op"
(( num_luns == $actual_stripe_count )) ||
error "OST$last_lun $num_luns != $actual_stripe_count"
(( tot_len == 1024 )) ||
error "OST$last_lun $tot_len != 1024"
echo "FIEMAP on $OSTCOUNT-stripe file succeeded"
}
run_test 130b "FIEMAP ($OSTCOUNT-stripe file)"
test_130c() {
(( $OSTCOUNT >= 2 )) || skip "needs >= 2 OSTs"
local filefrag_op=$(filefrag -l 2>&1 | grep "invalid option")
[[ -z "$filefrag_op" ]] || skip_env "filefrag missing logical ordering"
[[ "$ost1_FSTYPE" != "zfs" ]] ||
skip "LU-1941: FIEMAP unimplemented on ZFS"
local fm_file=$DIR/$tfile
$LFS setstripe -S 65536 -c 2 $fm_file || error "setstripe on $fm_file"
stack_trap "rm -f $fm_file"
dd if=/dev/zero of=$fm_file seek=1 bs=1M count=1 ||
error "dd failed on $fm_file"
filefrag -ves $fm_file || error "filefrag $fm_file failed"
filefrag_op=$(filefrag -ve -k $fm_file |
sed -n '/ext:/,/found/{/ext:/d; /found/d; p}')
local last_lun=$(echo $filefrag_op | cut -d: -f5 |
sed -e 's/^[ \t]*/0x/' | sed -e 's/0x0x/0x/')
local tot_len=0
local num_luns=1
while read line; do
local frag_lun=$(echo $line | cut -d: -f5 |
sed -e 's/^[ \t]*/0x/' | sed -e 's/0x0x/0x/')
local ext_len=$(echo $line | cut -d: -f4)
if (( $frag_lun != $last_lun )); then
local logical=$(echo $line | cut -d: -f2 | cut -d. -f1)
(( logical == 512 )) ||
error "OST$last_lun logical start $logical != 512"
(( tot_len == 512 )) ||
error "OST$last_lun $tot_len != 512"
(( num_luns += 1 ))
tot_len=0
fi
(( tot_len += ext_len ))
last_lun=$frag_lun
done <<< "$filefrag_op"
(( num_luns == 2 )) ||
error "OST$last_lun $num_luns != 2"
(( tot_len == 512 )) ||
error "OST$last_lun $tot_len != 512"
echo "FIEMAP on 2-stripe file with hole succeeded"
}
run_test 130c "FIEMAP (2-stripe file with hole)"
test_130d() {
(( $OSTCOUNT >= 3 )) || skip "needs >= 3 OSTs"
filefrag_op=$(filefrag -l 2>&1 | grep "invalid option")
[[ -z "$filefrag_op" ]] || skip_env "filefrag missing logical ordering"
[[ "$ost1_FSTYPE" != "zfs" ]] ||
skip "LU-1941: FIEMAP unimplemented on ZFS"
local fm_file=$DIR/$tfile
$LFS setstripe -S 65536 -c $OSTCOUNT $fm_file ||
error "setstripe on $fm_file"
stack_trap "rm -f $fm_file"
local actual_stripe_count=$($LFS getstripe -c $fm_file)
dd if=/dev/zero of=$fm_file bs=1M count=$actual_stripe_count ||
error "dd failed on $fm_file"
filefrag -ves $fm_file || error "filefrag $fm_file failed"
filefrag_op=$(filefrag -ve -k $fm_file |
sed -n '/ext:/,/found/{/ext:/d; /found/d; p}')
local last_lun=$(echo $filefrag_op | cut -d: -f5 |
sed -e 's/^[ \t]*/0x/' | sed -e 's/0x0x/0x/')
local tot_len=0
local num_luns=1
while read line; do
local frag_lun=$(echo $line | cut -d: -f5 |
sed -e 's/^[ \t]*/0x/' | sed -e 's/0x0x/0x/')
local ext_len=$(echo $line | cut -d: -f4)
if (( $frag_lun != $last_lun )); then
(( tot_len == 1024 )) ||
error "OST$last_lun $tot_len != 1024"
(( num_luns += 1 ))
tot_len=0
fi
(( tot_len += ext_len ))
last_lun=$frag_lun
done <<< "$filefrag_op"
(( num_luns == actual_stripe_count )) ||
error "OST$last_lun $num_luns != $actual_stripe_count"
(( tot_len == 1024 )) ||
error "OST$last_lun $tot_len != 1024"
echo "FIEMAP on N-stripe file succeeded"
}
run_test 130d "FIEMAP (N-stripe file)"
test_130e() {
(( $OSTCOUNT >= 2 )) || skip "needs >= 2 OSTs"
local filefrag_op=$(filefrag -l 2>&1 | grep "invalid option")
[[ -z "$filefrag_op" ]] || skip_env "filefrag missing logical ordering"
[[ "$ost1_FSTYPE" != "zfs" ]] ||
skip "LU-1941: FIEMAP unimplemented on ZFS"
local fm_file=$DIR/$tfile
$LFS setstripe -S 131072 -c 2 $fm_file || error "setstripe on $fm_file"
stack_trap "rm -f $fm_file"
local num_blks=512
local expected_len=$(( (num_blks / 2) * 64 ))
for ((i = 0; i < $num_blks; i++)); do
dd if=/dev/zero of=$fm_file count=1 bs=64k seek=$((2*$i)) \
conv=notrunc > /dev/null 2>&1
done
filefrag -ves $fm_file || error "filefrag $fm_file failed"
filefrag_op=$(filefrag -ve -k $fm_file |
sed -n '/ext:/,/found/{/ext:/d; /found/d; p}')
local last_lun=$(echo $filefrag_op | cut -d: -f5)
local tot_len=0
local num_luns=1
while read line; do
local frag_lun=$(echo $line | cut -d: -f5)
local ext_len=$(echo $line | cut -d: -f4)
if (( $frag_lun != $last_lun )); then
(( tot_len == $expected_len )) ||
error "OST$last_lun $tot_len != $expected_len"
(( num_luns += 1 ))
tot_len=0
fi
(( tot_len += ext_len ))
last_lun=$frag_lun
done <<< "$filefrag_op"
(( num_luns == 2 )) ||
error "OST$last_lun $num_luns != 2"
(( tot_len == $expected_len )) ||
error "OST$last_lun $tot_len != $expected_len"
echo "FIEMAP with continuation calls succeeded"
}
run_test 130e "FIEMAP (test continuation FIEMAP calls)"
test_130f() {
local filefrag_op=$(filefrag -l 2>&1 | grep "invalid option")
[[ -z "$filefrag_op" ]] || skip_env "filefrag missing logical ordering"
[[ "$ost1_FSTYPE" != "zfs" ]] ||
skip "LU-1941: FIEMAP unimplemented on ZFS"
local fm_file=$DIR/$tfile
$MULTIOP $fm_file oO_RDWR:O_CREAT:O_LOV_DELAY_CREATE:T33554432c ||
error "multiop create with lov_delay_create on $fm_file"
filefrag -ves $fm_file || error "filefrag $fm_file failed"
filefrag_extents=$(filefrag -vek $fm_file |
awk '/extents? found/ { print $2 }')
if (( $filefrag_extents != 0 )); then
error "$fm_file: filefrag_extents=$filefrag_extents != 0"
fi
rm -f $fm_file
}
run_test 130f "FIEMAP (unstriped file)"
test_130g() {
(( $MDS1_VERSION >= $(version_code 2.12.53) )) ||
skip "Need MDS version with at least 2.12.53 for overstriping"
local filefrag_op=$(filefrag -l 2>&1 | grep "invalid option")
[[ -z "$filefrag_op" ]] || skip_env "filefrag missing logical ordering"
[[ "$ost1_FSTYPE" != "zfs" ]] ||
skip "LU-1941: FIEMAP unimplemented on ZFS"
local file=$DIR/$tfile
local nr=$((OSTCOUNT * 100))
$LFS setstripe -C $nr -S1M $file ||
error "failed to setstripe -C $nr $file"
stack_trap "rm -f $file"
dd if=/dev/zero of=$file count=$nr bs=1M
sync
nr=$($LFS getstripe -c $file)
local extents=$(filefrag -v $file |
sed -n '/ext:/,/found/{/ext:/d; /found/d; p}' | wc -l)
echo "filefrag list $extents extents in file with stripecount $nr"
if (( extents < nr )); then
$LFS getstripe $file
filefrag -v $file
error "filefrag printed $extents < $nr extents"
fi
}
run_test 130g "FIEMAP (overstripe file)"
test_130h() {
(( OSTCOUNT < 2 )) && skip_env "need 2 OSTs"
$LFS setstripe -o 0,1 -S 1M $DIR/$tfile
$LFS getstripe $DIR/$tfile
dd if=/dev/zero of=$DIR/$tfile bs=1M count=2
stack_trap "rm -f $DIR/$tfile"
$LCTL set_param ldlm.namespaces.*-OST0000-osc-*.lru_size=clear
sleep 1
local before=$(date +%s)
##define OBD_FAIL_OSC_FIEMAP 0x418
$LCTL set_param fail_loc=0x80000418 fail_val=5
checkfiemap $DIR/$tfile $((2 * 1024 * 1024)) &
sleep 1
dd if=/dev/zero of=$DIR/$tfile bs=1M count=3
wait
$LCTL set_param fail_loc=0 fail_val=0
# check for client eviction
local evict=$($LCTL get_param osc.$FSNAME-OST0001-osc-f*.state |
awk -F"[ [,]" '/EVICTED ]$/ { if (mx<$5) {mx=$5;} } END { print mx }')
[ -z "$evict" ] || [[ $evict -le $before ]] || error "eviction happened"
}
run_test 130h "FIEMAP deadlock"
test_130i() {
(( $MDS1_VERSION >= $(version_code 2.15.63.195) )) ||
skip "Need MDS version at least 2.15.63.195 for DoM support"
local filefrag_op=$(filefrag -l 2>&1 | grep "invalid option")
[[ -z "$filefrag_op" ]] || skip_env "filefrag missing Lustre support"
[[ "$ost1_FSTYPE" != "zfs" ]] ||
skip "LU-1941: FIEMAP unimplemented on ZFS"
local dom_file=$DIR/$tfile
stack_trap "rm -f $dom_file"
$LFS setstripe -E 1M -L mdt -E -1 -c2 -S 131072 -o1,0 $dom_file ||
error "setstripe on $dom_file"
local page_kb=$((PAGE_SIZE / 1024))
# each component should be 1 MiB in size, 1/2 of it is written
local blks=$((1024 * 3 / page_kb / 2))
local expected=$(((blks / 3) * page_kb))
echo "BLKS: $blks"
echo "PAGE_SIZE: $PAGE_SIZE"
echo "PAGE_KB: $page_kb"
echo "EXPECTED: $expected"
for ((i = 0; i < $blks; i++)); do
dd if=/dev/zero of=$dom_file count=1 bs=${page_kb}k \
seek=$((2 * i)) conv=notrunc > /dev/null 2>&1 ||
error "dd failed to $dom_file"
done
filefrag -ves $dom_file | (head -7; echo ; tail -5)
(( ! ${PIPESTATUS[0]} )) || error "filefrag $dom_file failed"
filefrag_op=$(filefrag -ve -k $dom_file |
sed -n '/ext:/,/found/{/ext:/d; /found/d; p}')
local last_lun=$(echo $filefrag_op | cut -d: -f5)
local lun_len=0
local num_luns=1
while IFS=$'\n' read line; do
local frag_lun=$(echo $line | cut -d: -f5)
local ext_len=$(echo $line | cut -d: -f4)
if (( $frag_lun != $last_lun )); then
if (( lun_len != expected )); then
error "dev #$last_lun: $lun_len != $expected"
else
(( num_luns += 1 ))
lun_len=0
fi
fi
(( lun_len += ext_len ))
last_lun=$frag_lun
done <<< "$filefrag_op"
if (( num_luns != 3 )); then
error "num devices: $num_luns, but 3 expected"
fi
echo "LUN_LEN: $lun_len"
if (( lun_len != expected )); then
error "dev #$last_lun: $lun_len != $expected"
fi
echo "FIEMAP on DoM file succeeded"
}
run_test 130i "FIEMAP (DoM file)"
# Test for writev/readv
test_131a() {
rwv -f $DIR/$tfile -w -n 3 524288 1048576 1572864 ||
error "writev test failed"
rwv -f $DIR/$tfile -r -v -n 2 1572864 1048576 ||
error "readv failed"
rm -f $DIR/$tfile
}
run_test 131a "test iov's crossing stripe boundary for writev/readv"
test_131b() {
local fsize=$((524288 + 1048576 + 1572864))
rwv -f $DIR/$tfile -w -a -n 3 524288 1048576 1572864 &&
$CHECKSTAT -t file $DIR/$tfile -s $fsize ||
error "append writev test failed"
((fsize += 1572864 + 1048576))
rwv -f $DIR/$tfile -w -a -n 2 1572864 1048576 &&
$CHECKSTAT -t file $DIR/$tfile -s $fsize ||
error "append writev test failed"
rm -f $DIR/$tfile
}
run_test 131b "test append writev"
test_131c() {
rwv -f $DIR/$tfile -w -d -n 1 1048576 || return 0
error "NOT PASS"
}
run_test 131c "test read/write on file w/o objects"
test_131d() {
rwv -f $DIR/$tfile -w -n 1 1572864
NOB=`rwv -f $DIR/$tfile -r -n 3 524288 524288 1048576 | awk '/error/ {print $6}'`
if [ "$NOB" != 1572864 ]; then
error "Short read filed: read $NOB bytes instead of 1572864"
fi
rm -f $DIR/$tfile
}
run_test 131d "test short read"
test_131e() {
rwv -f $DIR/$tfile -w -s 1048576 -n 1 1048576
rwv -f $DIR/$tfile -r -z -s 0 -n 1 524288 || \
error "read hitting hole failed"
rm -f $DIR/$tfile
}
run_test 131e "test read hitting hole"
check_stats() {
local facet=$1
local op=$2
local want=${3:-0}
local res
# open 11 samples [usecs] 468 4793 13658 35791898
case $facet in
mds*) res=($(do_facet $facet \
$LCTL get_param mdt.$FSNAME-MDT0000.md_stats | grep "$op"))
;;
ost*) res=($(do_facet $facet \
$LCTL get_param obdfilter.$FSNAME-OST0000.stats | grep "$op"))
;;
*) error "Wrong facet '$facet'" ;;
esac
[[ -n "$res" ]] || error "counter for $op on $facet not incremented"
# if $want is zero, it means any stat increment is ok.
if (( $want > 0 )); then
local count=${res[1]}
if (( $count != $want )); then
if [[ $facet =~ "mds" ]]; then
do_nodes $(mdts_nodes) \
$LCTL get_param mdt.*.md_stats
else
do_nodes $(osts_nodes) \
$LCTL get_param obdfilter.*.stats
fi
error "The $op counter on $facet is $count, not $want"
fi
fi
}
test_133a() {
[ $PARALLEL == "yes" ] && skip "skip parallel run"
remote_ost_nodsh && skip "remote OST with nodsh"
remote_mds_nodsh && skip "remote MDS with nodsh"
do_facet $SINGLEMDS $LCTL list_param mdt.*.rename_stats ||
skip_env "MDS doesn't support rename stats"
local testdir=$DIR/${tdir}/stats_testdir
mkdir_on_mdt0 $DIR/${tdir}
# clear stats.
do_facet $SINGLEMDS $LCTL set_param mdt.*.md_stats=clear
do_facet ost1 $LCTL set_param obdfilter.*.stats=clear
# verify mdt stats first.
mkdir_on_mdt0 ${testdir} || error "mkdir_on_mdt0 failed"
check_stats $SINGLEMDS "mkdir" 1
# clear "open" from "lfs mkdir" above
do_facet $SINGLEMDS $LCTL set_param mdt.*.md_stats=clear
touch ${testdir}/${tfile} || error "touch failed"
check_stats $SINGLEMDS "open" 1
check_stats $SINGLEMDS "close" 1
(( $MDS1_VERSION >= $(version_code 2.15.62) )) && {
# open should match close
ls -lR ${testdir}
check_stats $SINGLEMDS "open" 2
check_stats $SINGLEMDS "close" 2
}
[ $MDS1_VERSION -ge $(version_code 2.8.54) ] && {
mknod ${testdir}/${tfile}-pipe p || error "mknod failed"
check_stats $SINGLEMDS "mknod" 2
}
rm -f ${testdir}/${tfile}-pipe || error "pipe remove failed"
check_stats $SINGLEMDS "unlink" 1
rm -f ${testdir}/${tfile} || error "file remove failed"
check_stats $SINGLEMDS "unlink" 2
# remove working dir and check mdt stats again.
rmdir ${testdir} || error "rmdir failed"
check_stats $SINGLEMDS "rmdir" 1
local testdir1=$DIR/${tdir}/stats_testdir1
mkdir_on_mdt0 ${testdir}
mkdir_on_mdt0 ${testdir1}
touch ${testdir1}/test1
mv ${testdir1}/test1 ${testdir} || error "file crossdir rename"
check_stats $SINGLEMDS "crossdir_rename" 1
mv ${testdir}/test1 ${testdir}/test0 || error "file samedir rename"
check_stats $SINGLEMDS "samedir_rename" 1
rm -rf $DIR/${tdir}
}
run_test 133a "Verifying MDT stats ========================================"
test_133b() {
local res
[ $PARALLEL == "yes" ] && skip "skip parallel run"
remote_ost_nodsh && skip "remote OST with nodsh"
remote_mds_nodsh && skip "remote MDS with nodsh"
local testdir=$DIR/${tdir}/stats_testdir
mkdir -p $DIR/$tdir || error "mkdir $tdir failed"
mkdir_on_mdt0 ${testdir} || error "mkdir_on_mdt0 failed"
touch ${testdir}/${tfile} || error "touch failed"
cancel_lru_locks mdc
# clear stats.
do_facet $SINGLEMDS $LCTL set_param mdt.*.md_stats=clear
do_facet ost1 $LCTL set_param obdfilter.*.stats=clear
# extra mdt stats verification.
chmod 444 ${testdir}/${tfile} || error "chmod failed"
check_stats $SINGLEMDS "setattr" 1
do_facet $SINGLEMDS $LCTL set_param mdt.*.md_stats=clear
if [ $MDS1_VERSION -ne $(version_code 2.2.0) ]
then # LU-1740
ls -l ${testdir}/${tfile} > /dev/null|| error "ls failed"
check_stats $SINGLEMDS "getattr" 1
fi
rm -rf $DIR/${tdir}
# when DNE is enabled, MDT uses STATFS RPC to ping other targets
# so the check below is not reliable
[ $MDSCOUNT -eq 1 ] || return 0
# Sleep to avoid a cached response.
#define OBD_STATFS_CACHE_SECONDS 1
sleep 2
do_facet $SINGLEMDS $LCTL set_param mdt.*.md_stats=clear
do_facet ost1 $LCTL set_param obdfilter.*.exports.*.stats=clear
$LFS df || error "lfs failed"
check_stats $SINGLEMDS "statfs" 1
# check aggregated statfs (LU-10018)
[ $MDS1_VERSION -lt $(version_code 2.11.54) ] &&
return 0
[ $CLIENT_VERSION -lt $(version_code 2.11.54) ] &&
return 0
sleep 2
do_facet $SINGLEMDS $LCTL set_param mdt.*.md_stats=clear
do_facet ost1 $LCTL set_param obdfilter.*.exports.*.stats=clear
df $DIR
check_stats $SINGLEMDS "statfs" 1
# We want to check that the client didn't send OST_STATFS to
# ost1 but the MDT also uses OST_STATFS for precreate. So some
# extra care is needed here.
if remote_mds; then
local nid=$($LCTL list_nids | head -1 | sed "s/\./\\\./g")
local param="obdfilter.$FSNAME-OST0000.exports.'$nid'.stats"
res=$(do_facet ost1 $LCTL get_param $param | grep statfs)
[ "$res" ] && error "OST got STATFS"
fi
return 0
}
run_test 133b "Verifying extra MDT stats =================================="
test_133c() {
[ $PARALLEL == "yes" ] && skip "skip parallel run"
remote_ost_nodsh && skip "remote OST with nodsh"
remote_mds_nodsh && skip "remote MDS with nodsh"
local testdir=$DIR/$tdir/stats_testdir
test_mkdir -p $testdir
# verify obdfilter stats.
$LFS setstripe -c 1 -i 0 $testdir/$tfile
sync
cancel_lru_locks osc
wait_delete_completed
# clear stats.
do_facet $SINGLEMDS $LCTL set_param mdt.*.md_stats=clear
do_facet ost1 $LCTL set_param obdfilter.*.stats=clear
dd if=/dev/zero of=$testdir/$tfile conv=notrunc bs=512k count=1 ||
error "dd failed"
sync
cancel_lru_locks osc
check_stats ost1 "write" 1
dd if=$testdir/$tfile of=/dev/null bs=1k count=1 || error "dd failed"
check_stats ost1 "read" 1
> $testdir/$tfile || error "truncate failed"
check_stats ost1 "punch" 1
rm -f $testdir/$tfile || error "file remove failed"
wait_delete_completed
check_stats ost1 "destroy" 1
rm -rf $DIR/$tdir
}
run_test 133c "Verifying OST stats ========================================"
order_2() {
local value=$1
local orig=$value
local order=1
while [ $value -ge 2 ]; do
order=$((order*2))
value=$((value/2))
done
if [ $orig -gt $order ]; then
order=$((order*2))
fi
echo $order
}
size_in_KMGT() {
local value=$1
local size=('K' 'M' 'G' 'T');
local i=0
local size_string=$value
while [ $value -ge 1024 ]; do
if [ $i -gt 3 ]; then
#T is the biggest unit we get here, if that is bigger,
#just return XXXT
size_string=${value}T
break
fi
value=$((value >> 10))
if [ $value -lt 1024 ]; then
size_string=${value}${size[$i]}
break
fi
i=$((i + 1))
done
echo $size_string
}
get_rename_size() {
local size=$1
local context=${2:-.}
local sample=$(do_facet $SINGLEMDS $LCTL \
get_param mdt.$FSNAME-MDT0000.rename_stats |
grep -A1 $context |
awk '/ '${size}'/ {print $4}' | sed -e "s/,//g")
echo $sample
}
test_133d() {
[ $PARALLEL == "yes" ] && skip "skip parallel run"
remote_ost_nodsh && skip "remote OST with nodsh"
remote_mds_nodsh && skip "remote MDS with nodsh"
do_facet $SINGLEMDS $LCTL list_param mdt.*.rename_stats ||
skip_env "MDS doesn't support rename stats"
local testdir1=$DIR/${tdir}/stats_testdir1
local testdir2=$DIR/${tdir}/stats_testdir2
mkdir -p $DIR/${tdir} || error "mkdir $tdir failed"
do_facet $SINGLEMDS $LCTL set_param mdt.*.rename_stats=clear
mkdir_on_mdt0 ${testdir1} || error "mkdir $testdir1 failed"
mkdir_on_mdt0 ${testdir2} || error "mkdir $testdir2 failed"
createmany -o $testdir1/test 512 || error "createmany failed"
# check samedir rename size
mv ${testdir1}/test0 ${testdir1}/test_0
local testdir1_size=$(ls -l $DIR/${tdir} |
awk '/stats_testdir1/ {print $5}')
local testdir2_size=$(ls -l $DIR/${tdir} |
awk '/stats_testdir2/ {print $5}')
testdir1_size=$(order_2 $testdir1_size)
testdir2_size=$(order_2 $testdir2_size)
testdir1_size=$(size_in_KMGT $testdir1_size)
testdir2_size=$(size_in_KMGT $testdir2_size)
echo "source rename dir size: ${testdir1_size}"
echo "target rename dir size: ${testdir2_size}"
local cmd="do_facet $SINGLEMDS $LCTL "
cmd+="get_param mdt.$FSNAME-MDT0000.rename_stats"
eval $cmd || error "$cmd failed"
local samedir=$($cmd | grep 'same_dir')
local same_sample=$(get_rename_size $testdir1_size)
[ -z "$samedir" ] && error "samedir_rename_size count error"
[[ $same_sample -eq 1 ]] ||
error "samedir_rename_size error $same_sample"
echo "Check same dir rename stats success"
do_facet $SINGLEMDS $LCTL set_param mdt.*.rename_stats=clear
# check crossdir rename size
mv ${testdir1}/test_0 ${testdir2}/test_0
testdir1_size=$(ls -l $DIR/${tdir} |
awk '/stats_testdir1/ {print $5}')
testdir2_size=$(ls -l $DIR/${tdir} |
awk '/stats_testdir2/ {print $5}')
testdir1_size=$(order_2 $testdir1_size)
testdir2_size=$(order_2 $testdir2_size)
testdir1_size=$(size_in_KMGT $testdir1_size)
testdir2_size=$(size_in_KMGT $testdir2_size)
echo "source rename dir size: ${testdir1_size}"
echo "target rename dir size: ${testdir2_size}"
eval $cmd || error "$cmd failed"
local crossdir=$($cmd | grep 'crossdir')
local src_sample=$(get_rename_size $testdir1_size crossdir_src)
local tgt_sample=$(get_rename_size $testdir2_size crossdir_tgt)
[ -z "$crossdir" ] && error "crossdir_rename_size count error"
[[ $src_sample -eq 1 ]] ||
error "crossdir_rename_size error $src_sample"
[[ $tgt_sample -eq 1 ]] ||
error "crossdir_rename_size error $tgt_sample"
echo "Check cross dir rename stats success"
rm -rf $DIR/${tdir}
}
run_test 133d "Verifying rename_stats ========================================"
test_133e() {
remote_mds_nodsh && skip "remote MDS with nodsh"
remote_ost_nodsh && skip "remote OST with nodsh"
[ $PARALLEL == "yes" ] && skip "skip parallel run"
local testdir=$DIR/${tdir}/stats_testdir
local ctr f0 f1 bs=32768 count=42 sum
mkdir -p ${testdir} || error "mkdir failed"
$LFS setstripe -c 1 -i 0 ${testdir}/${tfile}
for ctr in {write,read}_bytes; do
sync
cancel_lru_locks osc
do_facet ost1 $LCTL set_param -n \
"obdfilter.*.exports.clear=clear"
if [ $ctr = write_bytes ]; then
f0=/dev/zero
f1=${testdir}/${tfile}
else
f0=${testdir}/${tfile}
f1=/dev/null
fi
dd if=$f0 of=$f1 conv=notrunc bs=$bs count=$count || \
error "dd failed"
sync
cancel_lru_locks osc
sum=$(do_facet ost1 $LCTL get_param \
"obdfilter.*.exports.*.stats" |
awk -v ctr=$ctr 'BEGIN { sum = 0 }
$1 == ctr { sum += $7 }
END { printf("%0.0f", sum) }')
if ((sum != bs * count)); then
error "Bad $ctr sum, expected $((bs * count)), got $sum"
fi
done
rm -rf $DIR/${tdir}
}
run_test 133e "Verifying OST {read,write}_bytes nid stats ================="
test_133f() {
[[ $(lustre_version_code $facet) -ge $(version_code 2.7.65) ]] ||
skip "too old lustre for get_param -R ($facet_ver)"
# verifying readability.
$LCTL get_param -R '*' &> /dev/null
# Verifing writability with badarea_io.
local proc_dirs=$(eval \ls -d $proc_regexp 2>/dev/null)
local skipped_params='force_lbug|changelog_mask|daemon_file'
$LCTL list_param -FR '*' | grep '=' | tr -d = |
grep -E -v "$skipped_params" |
xargs -n 1 find $proc_dirs -name |
xargs -n 1 badarea_io ||
error "client badarea_io failed"
# remount the FS in case writes/reads /proc break the FS
cleanup || error "failed to unmount"
setup || error "failed to setup"
}
run_test 133f "Check reads/writes of client lustre proc files with bad area io"
test_133g() {
remote_mds_nodsh && skip "remote MDS with nodsh"
remote_ost_nodsh && skip "remote OST with nodsh"
local proc_dirs=$(eval \ls -d $proc_regexp 2>/dev/null)
local proc_dirs_str=$(eval echo $proc_dirs)
local skipped_params="'force_lbug|daemon_file'"
local facet
for facet in mds1 ost1; do
local facet_ver=$(lustre_version_code $facet)
if [ $facet_ver -ge $(version_code 2.7.65) ]; then
do_facet $facet "$LCTL get_param -R '*'" &> /dev/null
else
log "$facet: too old lustre for get_param -R"
fi
if [ $facet_ver -ge $(version_code 2.5.54) ]; then
do_facet $facet "$LCTL list_param -FR '*' | grep '=' |
tr -d = | grep -E -v $skipped_params |
xargs -n 1 find $proc_dirs_str -name |
xargs -n 1 badarea_io" ||
error "$facet badarea_io failed"
else
skip_noexit "$facet: too old lustre for get_param -R"
fi
done
# remount the FS in case writes/reads /proc break the FS
cleanup || error "failed to unmount"
setup || error "failed to setup"
}
run_test 133g "Check reads/writes of server lustre proc files with bad area io"
test_133h() {
remote_mds_nodsh && skip "remote MDS with nodsh"
remote_ost_nodsh && skip "remote OST with nodsh"
[[ $MDS1_VERSION -lt $(version_code 2.9.54) ]] &&
skip "Need MDS version at least 2.9.54"
local facet
for facet in client mds1 ost1; do
# Get the list of files that are missing the terminating newline
local plist=$(do_facet $facet
$LCTL list_param -FR '*' | grep '=' | tr -d =)
local ent
for ent in $plist; do
local missing=$(do_facet $facet $LCTL get_param $ent \|\
awk -v FS='\v' -v RS='\v\v' \
"'END { if(NR>0 && \\\$NF !~ /.*\\\n\$/) \
print FILENAME}'" 2>/dev/null)
[ -z $missing ] || {
do_facet $facet $LCTL get_param $ent | od -An -tx1
error "file does not end with newline: $facet-$ent"
}
done
done
}
run_test 133h "Proc files should end with newlines"
test_134a() {
remote_mds_nodsh && skip "remote MDS with nodsh"
[[ $MDS1_VERSION -lt $(version_code 2.7.54) ]] &&
skip "Need MDS version at least 2.7.54"
mkdir_on_mdt0 $DIR/$tdir || error "failed to create $DIR/$tdir"
cancel_lru_locks mdc
local nsdir="ldlm.namespaces.*-MDT0000-mdc-*"
local unused=$($LCTL get_param -n $nsdir.lock_unused_count)
[ $unused -eq 0 ] || error "$unused locks are not cleared"
local nr=1000
createmany -o $DIR/$tdir/f $nr ||
error "failed to create $nr files in $DIR/$tdir"
unused=$($LCTL get_param -n $nsdir.lock_unused_count)
#define OBD_FAIL_LDLM_WATERMARK_LOW 0x327
do_facet mds1 $LCTL set_param fail_loc=0x327
do_facet mds1 $LCTL set_param fail_val=500
touch $DIR/$tdir/m
echo "sleep 10 seconds ..."
sleep 10
local lck_cnt=$($LCTL get_param -n $nsdir.lock_unused_count)
do_facet mds1 $LCTL set_param fail_loc=0
do_facet mds1 $LCTL set_param fail_val=0
[ $lck_cnt -lt $unused ] ||
error "No locks reclaimed, before:$unused, after:$lck_cnt"
rm $DIR/$tdir/m
unlinkmany $DIR/$tdir/f $nr
}
run_test 134a "Server reclaims locks when reaching lock_reclaim_threshold"
test_134b() {
remote_mds_nodsh && skip "remote MDS with nodsh"
[[ $MDS1_VERSION -lt $(version_code 2.7.54) ]] &&
skip "Need MDS version at least 2.7.54"
mkdir_on_mdt0 $DIR/$tdir || error "failed to create $DIR/$tdir"
cancel_lru_locks mdc
local low_wm=$(do_facet mds1 $LCTL get_param -n \
ldlm.lock_reclaim_threshold_mb)
# disable reclaim temporarily
do_facet mds1 $LCTL set_param ldlm.lock_reclaim_threshold_mb=0
#define OBD_FAIL_LDLM_WATERMARK_HIGH 0x328
do_facet mds1 $LCTL set_param fail_loc=0x328
do_facet mds1 $LCTL set_param fail_val=500
$LCTL set_param debug=+trace
local nr=600
createmany -o $DIR/$tdir/f $nr &
local create_pid=$!
echo "Sleep $TIMEOUT seconds ..."
sleep $TIMEOUT
if ! ps -p $create_pid > /dev/null 2>&1; then
do_facet mds1 $LCTL set_param fail_loc=0
do_facet mds1 $LCTL set_param fail_val=0
do_facet mds1 $LCTL set_param \
ldlm.lock_reclaim_threshold_mb=${low_wm}m
error "createmany finished incorrectly!"
fi
do_facet mds1 $LCTL set_param fail_loc=0
do_facet mds1 $LCTL set_param fail_val=0
do_facet mds1 $LCTL set_param ldlm.lock_reclaim_threshold_mb=${low_wm}m
wait $create_pid || return 1
unlinkmany $DIR/$tdir/f $nr
}
run_test 134b "Server rejects lock request when reaching lock_limit_mb"
test_134c() {
remote_mds_nodsh && skip "remote MDS with nodsh"
(( $MDS1_VERSION >= $(version_code 2.17.52) )) ||
skip "Need MDS version at least 2.17.52"
local slab=/sys/kernel/slab
local lock_sz=$(do_facet mds1 \
"cat $slab/ldlm_locks/object_size 2>/dev/null")
local res_sz=$(do_facet mds1 \
"cat $slab/ldlm_resources/object_size 2>/dev/null")
local ibits_sz=$(do_facet mds1 \
"cat $slab/ldlm_ibits_node/object_size 2>/dev/null")
[[ -n "$lock_sz" && -n "$res_sz" && -n "$ibits_sz" ]] ||
skip "slab metadata for ldlm caches unavailable"
local overhead=$((lock_sz + res_sz + ibits_sz))
echo "per-lock overhead: ${overhead}B" \
"(lock=$lock_sz res=$res_sz ibits=$ibits_sz)"
local orig_low=$(do_facet mds1 $LCTL get_param -n \
ldlm.lock_reclaim_threshold_mb)
local orig_high=$(do_facet mds1 $LCTL get_param -n \
ldlm.lock_limit_mb)
# kernel enforces threshold <= limit, so raise limit first on restore
stack_trap "do_facet mds1 $LCTL set_param \
ldlm.lock_limit_mb=$orig_high \
ldlm.lock_reclaim_threshold_mb=$orig_low"
local low_mb=20
local high_mb=50
do_facet mds1 $LCTL set_param \
ldlm.lock_reclaim_threshold_mb=$low_mb \
ldlm.lock_limit_mb=$high_mb
local low=$(do_facet mds1 $LCTL get_param -n \
ldlm.lock_reclaim_threshold_count)
local high=$(do_facet mds1 $LCTL get_param -n \
ldlm.lock_limit_count)
(( low > 0 && high > 0 )) ||
error "lock counters unset (low=$low high=$high)"
# count × overhead should match the MB budget. Without the fix,
# sizeof(ldlm_lock) alone is used and count overshoots by ~1.5x,
# pushing the product well outside the tolerance band.
local low_mb_actual=$((low * overhead / 1024 / 1024))
local high_mb_actual=$((high * overhead / 1024 / 1024))
local low_pct=$((low_mb_actual * 100 / low_mb))
local high_pct=$((high_mb_actual * 100 / high_mb))
echo "reclaim_threshold: $low locks ≈ ${low_mb_actual}MB" \
"(set ${low_mb}MB, ${low_pct}%)"
echo "lock_limit: $high locks ≈ ${high_mb_actual}MB" \
"(set ${high_mb}MB, ${high_pct}%)"
(( low_pct >= 80 && low_pct <= 120 )) ||
error "reclaim_threshold accounting off: ${low_mb_actual}MB vs ${low_mb}MB"
(( high_pct >= 80 && high_pct <= 120 )) ||
error "lock_limit accounting off: ${high_mb_actual}MB vs ${high_mb}MB"
}
run_test 134c "Lock memory accounting includes associated structures"
test_135() {
remote_mds_nodsh && skip "remote MDS with nodsh"
[[ $MDS1_VERSION -lt $(version_code 2.13.50) ]] &&
skip "Need MDS version at least 2.13.50"
local fname
mkdir -p $DIR/$tdir || error "failed to create $DIR/$tdir"
#define OBD_FAIL_PLAIN_RECORDS 0x1319
#set only one record at plain llog
do_facet $SINGLEMDS $LCTL set_param fail_loc=0x1319 fail_val=1
ost_set_temp_seq_width_all $DATA_SEQ_MAX_WIDTH
#fill already existed plain llog each 64767
#wrapping whole catalog
createmany -o -u $DIR/$tdir/$tfile- $((64767 * 1))
createmany -o $DIR/$tdir/$tfile_ 64700
for (( i = 0; i < 64700; i = i + 2 ))
do
rm $DIR/$tdir/$tfile_$i &
rm $DIR/$tdir/$tfile_$((i + 1)) &
local pid=$!
wait $pid
done
#waiting osp synchronization
wait_delete_completed
}
run_test 135 "Race catalog processing"
test_136() {
remote_mds_nodsh && skip "remote MDS with nodsh"
[[ $MDS1_VERSION -lt $(version_code 2.13.50) ]] &&
skip "Need MDS version at least 2.13.50"
local fname
mkdir -p $DIR/$tdir || error "failed to create $DIR/$tdir"
$LFS setstripe -c 1 -i 0 $DIR/$tdir || error "failed to set striping"
#set only one record at plain llog
#define OBD_FAIL_CATALOG_FULL_CHECK 0x131a
do_facet $SINGLEMDS $LCTL set_param fail_loc=0x131a fail_val=1
ost_set_temp_seq_width_all $DATA_SEQ_MAX_WIDTH
# fill already existed 2 plain llogs each 64767 wrapping whole catalog,
# drop server memory periodically only on ZFS to avoid OOM during test
local items=1000
for ((created = 0; created < 64767 * 5 / 2; created += items)); do
echo "$(date +%s): create $created-$((created + items - 1))"
createmany -o -u $DIR/$tdir/$tfile- $created $items
if [[ "$mds1_FSTYPE" == "zfs" ]]; then
wait_delete_completed
do_facet mds1 "echo 1 > /proc/sys/vm/drop_caches"
fi
done
[[ "$FSTYPE" == "zfs" ]] || wait_delete_completed
createmany -o $DIR/$tdir/$tfile_ 10
sleep 25
do_facet $SINGLEMDS $LCTL set_param fail_val=3
for (( i = 0; i < 10; i = i + 3 )); do
rm $DIR/$tdir/$tfile_$i &
rm $DIR/$tdir/$tfile_$((i + 1)) &
local pid=$!
wait $pid
sleep 7
rm $DIR/$tdir/$tfile_$((i + 2)) &
done
#waiting osp synchronization
wait_delete_completed
}
run_test 136 "Race catalog processing 2"
test_140() { #bug-17379
[ $PARALLEL == "yes" ] && skip "skip parallel run"
test_mkdir $DIR/$tdir
cd $DIR/$tdir || error "Changing to $DIR/$tdir"
cp $(which stat) . || error "Copying stat to $DIR/$tdir"
# VFS limits max symlink depth to 5(4KSTACK) or 7(8KSTACK) or 8
# For kernel > 3.5, bellow only tests consecutive symlink (MAX 40)
local i=0
while i=$((i + 1)); do
test_mkdir $i
cd $i || error "Changing to $i"
ln -s ../stat stat || error "Creating stat symlink"
# Read the symlink until ELOOP present,
# not LBUGing the system is considered success,
# we didn't overrun the stack.
$OPENFILE -f O_RDONLY stat >/dev/null 2>&1; ret=$?
if [ $ret -ne 0 ]; then
if [ $ret -eq 40 ]; then
break # -ELOOP
else
error "Open stat symlink"
return
fi
fi
done
i=$((i - 1))
echo "The symlink depth = $i"
[ $i -eq 5 ] || [ $i -eq 7 ] || [ $i -eq 8 ] || [ $i -eq 40 ] ||
error "Invalid symlink depth"
# Test recursive symlink
ln -s symlink_self symlink_self
$OPENFILE -f O_RDONLY symlink_self >/dev/null 2>&1; ret=$?
echo "open symlink_self returns $ret"
[ $ret -eq 40 ] || error "recursive symlink doesn't return -ELOOP"
}
run_test 140 "Check reasonable stack depth (shouldn't LBUG) ===="
test_150a() {
[ $PARALLEL == "yes" ] && skip "skip parallel run"
local TF="$TMP/$tfile"
stack_trap "rm -f $DIR/$tfile; wait_delete_completed"
dd if=/dev/urandom of=$TF bs=6096 count=1 || error "dd failed"
cp $TF $DIR/$tfile
cancel_lru_locks $OSC
cmp $TF $DIR/$tfile || error "$TMP/$tfile $DIR/$tfile differ"
remount_client $MOUNT
df -P $MOUNT
cmp $TF $DIR/$tfile || error "$TF $DIR/$tfile differ (remount)"
$TRUNCATE $TF 6000
$TRUNCATE $DIR/$tfile 6000
cancel_lru_locks $OSC
cmp $TF $DIR/$tfile || error "$TF $DIR/$tfile differ (truncate1)"
echo "12345" >>$TF
echo "12345" >>$DIR/$tfile
cancel_lru_locks $OSC
cmp $TF $DIR/$tfile || error "$TF $DIR/$tfile differ (append1)"
echo "12345" >>$TF
echo "12345" >>$DIR/$tfile
cancel_lru_locks $OSC
cmp $TF $DIR/$tfile || error "$TF $DIR/$tfile differ (append2)"
}
run_test 150a "truncate/append tests"
test_150b() {
check_set_fallocate_or_skip
local out
touch $DIR/$tfile
stack_trap "rm -f $DIR/$tfile; wait_delete_completed"
out=$(check_fallocate $DIR/$tfile 2>&1) ||
skip_eopnotsupp "$out|check_fallocate failed"
}
run_test 150b "Verify fallocate (prealloc) functionality"
test_150bb() {
check_set_fallocate_or_skip
touch $DIR/$tfile
stack_trap "rm -f $DIR/$tfile; wait_delete_completed"
dd if=/dev/urandom of=$DIR/$tfile bs=1M count=20 || error "dd failed"
> $DIR/$tfile
fallocate -l $((1048576 * 20)) $DIR/$tfile || error "fallocate failed"
# precomputed md5sum for 20MB of zeroes
local expect="8f4e33f3dc3e414ff94e5fb6905cba8c"
local sum=($(md5sum $DIR/$tfile))
[[ "${sum[0]}" == "$expect" ]] || error "fallocate unwritten is not zero"
check_set_fallocate 1
> $DIR/$tfile
fallocate -l $((1048576 * 20)) $DIR/$tfile || error "fallocate failed"
sum=($(md5sum $DIR/$tfile))
[[ "${sum[0]}" == "$expect" ]] || error "fallocate zero is not zero"
}
run_test 150bb "Verify fallocate modes both zero space"
test_150c() {
check_set_fallocate_or_skip
local striping="-c2"
stack_trap "rm -f $DIR/$tfile; wait_delete_completed"
$LFS setstripe -c $OSTCOUNT -S1M $DIR/$tfile || error "setstripe failed"
fallocate -l ${OSTCOUNT}m $DIR/$tfile || error "fallocate failed"
local bytes=$(($(stat -c '%b * %B' $DIR/$tfile)))
local want=$((OSTCOUNT * 1048576))
# Must allocate all requested space, not more than 5% extra
(( $bytes >= $want && $bytes < $want * 105 / 100 )) ||
error "bytes $bytes is not $want"
rm -f $DIR/$tfile
echo "verify fallocate on PFL file"
[[ "x$DOM" == "xyes" ]] && striping="-L mdt"
$LFS setstripe -E1M $striping -E16M -c3 -Eeof -c 4 $DIR/$tfile ||
error "Create $DIR/$tfile failed"
fallocate -l $((1048576 * 512)) $DIR/$tfile || error "fallocate failed"
bytes=$(($(stat -c '%b * %B' $DIR/$tfile)))
want=$((512 * 1048576))
# Must allocate all requested space, not more than 5% extra
(( $bytes >= $want && $bytes < $want * 105 / 100 )) ||
error "bytes $bytes is not $want"
}
run_test 150c "Verify fallocate Size and Blocks"
test_150d() {
check_set_fallocate_or_skip
local striping="-c2"
[[ "x$DOM" == "xyes" ]] && striping="-L mdt"
stack_trap "rm -f $DIR/$tdir; wait_delete_completed"
$LFS setstripe -E1M $striping -E eof -c $OSTCOUNT -S1M $DIR/$tdir ||
error "setstripe failed"
fallocate -o 1G -l ${OSTCOUNT}m $DIR/$tdir || error "fallocate failed"
local bytes=$(($(stat -c '%b * %B' $DIR/$tdir)))
local want=$((OSTCOUNT * 1048576))
# Must allocate all requested space, not more than 5% extra
(( $bytes >= $want && $bytes < $want * 105 / 100 )) ||
error "bytes $bytes is not $want"
}
run_test 150d "Verify fallocate Size and Blocks - Non zero start"
test_150e() {
check_set_fallocate_or_skip
echo "df before:"
$LFS df
stack_trap "rm -f $DIR/$tfile; wait_delete_completed"
$LFS setstripe -c${OSTCOUNT} $DIR/$tfile ||
error "$LFS setstripe -c${OSTCOUNT} $DIR/$tfile failed"
# Find OST with Minimum Size
min_size_ost=$($LFS df --output=avail -o --no-header $DIR |
sort -un | head -1)
# Get 100MB per OST of the available space to reduce run time
# else 60% of the available space if we are running SLOW tests
if [ $SLOW == "no" ]; then
local space=$((1024 * 100 * OSTCOUNT))
else
local space=$(((min_size_ost * 60)/100 * OSTCOUNT))
fi
fallocate -l${space}k $DIR/$tfile ||
error "fallocate ${space}k $DIR/$tfile failed"
echo "'fallocate -l ${space}k $DIR/$tfile' succeeded"
# get size immediately after fallocate. This should be correctly
# updated
local size=$(stat -c '%s' $DIR/$tfile)
local used=$(( $(stat -c '%b * %B' $DIR/$tfile) / 1024))
# Sleep for a while for statfs to get updated. And not pull from cache.
sleep 2
echo "df after fallocate:"
$LFS df
(( size / 1024 == space )) || error "size $size != requested $space"
[ "$ost1_FSTYPE" != ldiskfs ] || (( used >= space )) ||
error "used $used < space $space"
rm $DIR/$tfile || error "rm failed"
sync
wait_delete_completed
echo "df after unlink:"
$LFS df
}
run_test 150e "Verify 60% of available OST space consumed by fallocate"
test_150f() {
local size
local blocks
local want_size_before=20480 # in bytes
local want_blocks_before=40 # 512 sized blocks
local want_blocks_after=24 # 512 sized blocks
local length=$(((want_blocks_before - want_blocks_after) * 512))
[[ $OST1_VERSION -ge $(version_code 2.14.0) ]] ||
skip "need at least 2.14.0 for fallocate punch"
if [ "$ost1_FSTYPE" = "zfs" ] || [ "$mds1_FSTYPE" = "zfs" ]; then
skip "LU-14160: punch mode is not implemented on OSD ZFS"
fi
check_set_fallocate_or_skip
stack_trap "rm -f $DIR/$tfile; wait_delete_completed"
[[ "x$DOM" == "xyes" ]] &&
$LFS setstripe -E1M -L mdt -E eof $DIR/$tfile
echo "Verify fallocate punch: Range within the file range"
yes 'A' | dd of=$DIR/$tfile bs=4096 count=5 ||
error "dd failed for bs 4096 and count 5"
# Call fallocate with punch range which is within the file range
out=$(fallocate -p --offset 4096 -l $length $DIR/$tfile 2>&1) ||
skip_eopnotsupp "$out|fallocate: offset 4096 and length $length"
# client must see changes immediately after fallocate
size=$(stat -c '%s' $DIR/$tfile)
blocks=$(stat -c '%b' $DIR/$tfile)
# Verify punch worked.
(( blocks == want_blocks_after )) ||
error "punch failed: blocks $blocks != $want_blocks_after"
(( size == want_size_before )) ||
error "punch failed: size $size != $want_size_before"
# Verify there is hole in file
local data_off=$(lseek_test -d 4096 $DIR/$tfile)
# precomputed md5sum
local expect="4a9a834a2db02452929c0a348273b4aa"
cksum=($(md5sum $DIR/$tfile))
[[ "${cksum[0]}" == "$expect" ]] ||
error "unexpected MD5SUM after punch: ${cksum[0]}"
# Start second sub-case for fallocate punch.
echo "Verify fallocate punch: Range overlapping and less than blocksize"
yes 'A' | dd of=$DIR/$tfile bs=4096 count=5 ||
error "dd failed for bs 4096 and count 5"
# Punch range less than block size will have no change in block count
want_blocks_after=40 # 512 sized blocks
# Punch overlaps two blocks and less than blocksize
out=$(fallocate -p --offset 4000 -l 3000 $DIR/$tfile 2>&1) ||
skip_eopnotsupp "$out|fallocate: offset 4000 length 3000"
size=$(stat -c '%s' $DIR/$tfile)
blocks=$(stat -c '%b' $DIR/$tfile)
# Verify punch worked.
(( blocks == want_blocks_after )) ||
error "punch failed: blocks $blocks != $want_blocks_after"
(( size == want_size_before )) ||
error "punch failed: size $size != $want_size_before"
# Verify if range is really zero'ed out. We expect Zeros.
# precomputed md5sum
expect="c57ec5d769c3dbe3426edc3f7d7e11d3"
cksum=($(md5sum $DIR/$tfile))
[[ "${cksum[0]}" == "$expect" ]] ||
error "unexpected MD5SUM after punch: ${cksum[0]}"
}
run_test 150f "Verify fallocate punch functionality"
test_150g() {
local space
local size
local blocks
local blocks_after
local size_after
local BS=4096 # Block size in bytes
[[ $OST1_VERSION -ge $(version_code 2.14.0) ]] ||
skip "need at least 2.14.0 for fallocate punch"
if [ "$ost1_FSTYPE" = "zfs" ] || [ "$mds1_FSTYPE" = "zfs" ]; then
skip "LU-14160: punch mode is not implemented on OSD ZFS"
fi
check_set_fallocate_or_skip
stack_trap "rm -f $DIR/$tfile; wait_delete_completed"
if [[ "x$DOM" == "xyes" ]]; then
$LFS setstripe -E2M -L mdt -E eof -c${OSTCOUNT} $DIR/$tfile ||
error "$LFS setstripe DoM + ${OSTCOUNT} OST failed"
else
$LFS setstripe -c${OSTCOUNT} $DIR/$tfile ||
error "$LFS setstripe -c${OSTCOUNT} $DIR/$tfile failed"
fi
# Get 100MB per OST of the available space to reduce run time
# else 60% of the available space if we are running SLOW tests
if [ $SLOW == "no" ]; then
space=$((1024 * 100 * OSTCOUNT))
else
# Find OST with Minimum Size
space=$($LFS df --output=avail --no-header -o $DIR |
sort -un | head -1)
echo "min size OST: $space"
space=$(((space * 60)/100 * OSTCOUNT))
fi
# space in 1k units, round to 4k blocks
local blkcount=$((space * 1024 / $BS))
echo "Verify fallocate punch: Very large Range"
fallocate -l${space}k $DIR/$tfile ||
error "fallocate ${space}k $DIR/$tfile failed"
# write 1M at the end, start and in the middle
yes 'A' | dd of=$DIR/$tfile bs=$BS count=256 ||
error "dd failed: bs $BS count 256"
yes 'A' | dd of=$DIR/$tfile bs=$BS seek=$((blkcount - 256)) count=256 ||
error "dd failed: bs $BS count 256 seek $((blkcount - 256))"
yes 'A' | dd of=$DIR/$tfile bs=$BS seek=$((blkcount / 2)) count=1024 ||
error "dd failed: bs $BS count 256 seek $((blkcount / 2))"
# Gather stats.
size=$(stat -c '%s' $DIR/$tfile)
# gather punch length.
local punch_size=$((size - (BS * 2)))
echo "punch_size = $punch_size"
echo "size - punch_size: $((size - punch_size))"
echo "size - punch_size in blocks: $(((size - punch_size)/BS))"
# Call fallocate to punch all except 2 blocks. We leave the
# first and the last block
echo "fallocate -p --offset $BS -l $punch_size $DIR/$tfile"
out=$(fallocate -p --offset $BS -l $punch_size $DIR/$tfile 2>&1) ||
skip_eopnotsupp "$out|fallocate: offset $BS length $punch_size"
size_after=$(stat -c '%s' $DIR/$tfile)
blocks_after=$(stat -c '%b' $DIR/$tfile)
# Verify punch worked.
# Size should be kept
(( size == size_after )) ||
error "punch failed: size $size != $size_after"
# two 4k data blocks to remain plus possible 1 extra extent block
(( blocks_after <= ((BS / 512) * 3) )) ||
error "too many blocks remains: $blocks_after"
# Verify that file has hole between the first and the last blocks
local hole_start=$(lseek_test -l 0 $DIR/$tfile)
local hole_end=$(lseek_test -d $BS $DIR/$tfile)
echo "Hole at [$hole_start, $hole_end)"
(( hole_start == BS )) ||
error "no hole at offset $BS after punch"
(( hole_end == BS + punch_size )) ||
error "data at offset $hole_end < $((BS + punch_size))"
}
run_test 150g "Verify fallocate punch on large range"
test_150h() {
local file=$DIR/$tfile
local size
check_set_fallocate_or_skip
statx_supported || skip_env "Test must be statx() syscall supported"
# fallocate() does not update the size information on the MDT
fallocate -l 16K $file || error "failed to fallocate $file"
cancel_lru_locks $OSC
# STATX with cached-always mode will not send glimpse RPCs to OST,
# it uses the caching attrs on the client side as much as possible.
size=$($STATX --cached=always -c %s $file)
[ $size == 16384 ] ||
error "size after fallocate() is $size, expected 16384"
}
run_test 150h "Verify extend fallocate updates the file size"
test_150ia() {
(( $MDS1_VERSION >= $(version_code 2.16.50) )) ||
skip "need MDS1 version >= 2.16.50 for falloc zero-range"
if [[ "$ost1_FSTYPE" = "zfs" || "$mds1_FSTYPE" = "zfs" ]]; then
skip "zero-range mode is not implemented on OSD ZFS"
fi
check_set_fallocate_or_skip
stack_trap "rm -f $DIR/$tfile; wait_delete_completed"
[[ "$DOM" == "yes" ]] &&
$LFS setstripe -E1M -L mdt -E eof $DIR/$tfile
local BLOCK_SIZE=65536
echo "Verify fallocate(zero): range within the file"
yes 'A' | dd of=$DIR/$tfile bs=$BLOCK_SIZE count=8 iflag=fullblock ||
error "dd failed for bs $BLOCK_SIZE and count 8"
# zero range page aligned
local offset=$((2 * BLOCK_SIZE))
local length=$((4 * BLOCK_SIZE))
out=$(fallocate -z --offset $offset -l $length $DIR/$tfile 2>&1) ||
skip_eopnotsupp "$out|falloc(zero): off $offset, len $length"
# precomputed md5sum
local expect="879e6c3bf37da2014adda326c8d5db73"
cksum=($(md5sum $DIR/$tfile))
[[ "${cksum[0]}" == "$expect" ]] ||
error "unexpected MD5SUM after zero: ${cksum[0]}"
# zero range partial page
local offset=2000
local length=1000
out=$(fallocate -z --offset $offset -l $length $DIR/$tfile 2>&1) ||
skip_eopnotsupp "$out|falloc(zero): off $offset, len $length"
expect="2ff89a843b5a656ca21036a2472a0aac"
cksum=($(md5sum $DIR/$tfile))
[[ "${cksum[0]}" == "$expect" ]] ||
error "unexpected MD5SUM after zero(partial): ${cksum[0]}"
}
run_test 150ia "Verify fallocate zero-range ZERO functionality"
test_150ib() {
(( $MDS1_VERSION >= $(version_code v2_17_50-71-g2cbe29c282) )) ||
skip "need MDS1 version >= 2.17.50 for falloc zero-range"
if [[ "$ost1_FSTYPE" = "zfs" || "$mds1_FSTYPE" = "zfs" ]]; then
skip "zero-range mode is not implemented on OSD ZFS"
fi
check_set_fallocate_or_skip
stack_trap "rm -f $DIR/$tfile; wait_delete_completed"
[[ "$DOM" == "yes" ]] &&
$LFS setstripe -E1M -L mdt -E eof $DIR/$tfile
local features=$(do_facet mds1 "$DEBUGFS -c -R stats $(mdsdevname 1)" |
grep "Filesystem features:")
echo "filesystem features: $features"
# Allow one 4K extent metadata block beyond the data blocks.
local blocks_after_zero_fill_min=$((8 * PAGE_SIZE / 512))
local blocks_after_zero_fill_max=$((8 * PAGE_SIZE / 512 + 8))
local blocks_after_extend_min=$((16 * PAGE_SIZE / 512))
local blocks_after_extend_max=$((16 * PAGE_SIZE / 512 + 8))
local expect_len=$((8 * PAGE_SIZE))
# file size [0, 32K)
echo "Verify fallocate(zero): range within the file"
yes 'A' | dd of=$DIR/$tfile bs=$PAGE_SIZE count=8 ||
error "dd failed for bs $PAGE_SIZE and count 8"
# punch across [8K,24K)
local offset=$((2 * PAGE_SIZE))
local length=$((4 * PAGE_SIZE))
out=$(fallocate -p --offset $offset -l $length $DIR/$tfile 2>&1) ||
skip_eopnotsupp "$out|falloc(zero): off $offset, len $length"
# Verify punch worked as expected
p=$(lseek_test -l 0 $DIR/$tfile)
(( p == offset )) ||
error "punch failed: hole at $p != $offset"
p=$(lseek_test -d $offset $DIR/$tfile)
(( p == offset + length )) ||
error "punch failed: data at $p != $((offset + length))"
# zero prealloc fill the hole just punched
out=$(fallocate -z --offset $offset -l $length $DIR/$tfile 2>&1) ||
skip_eopnotsupp "$out|falloc(zero): off $offset, len $length"
# Verify zero prealloc worked.
blocks=$(stat -c '%b' $DIR/$tfile)
(( blocks >= blocks_after_zero_fill_min &&
blocks <= blocks_after_zero_fill_max )) ||
error "zero prealloc failed:$blocks not in [$blocks_after_zero_fill_min,$blocks_after_zero_fill_max]"
# zero prealloc with KEEP_SIZE on
offset=$((8 * PAGE_SIZE))
length=$((8 * PAGE_SIZE))
out=$(fallocate -z -n --offset $offset -l $length $DIR/$tfile 2>&1) ||
skip_eopnotsupp "$out|falloc(zero): off $offset, len $length"
# block allocate, size remains
blocks=$(stat -c '%b' $DIR/$tfile)
(( blocks >= blocks_after_extend_min &&
blocks <= blocks_after_extend_max )) ||
error "extend failed:$blocks not in [$blocks_after_extend_min,$blocks_after_extend_max]"
lsz=$(stat -c '%s' $DIR/$tfile)
(( lsz == expect_len)) ||
error "zero extend failed(len):$lsz!=$expect_len"
}
run_test 150ib "Verify fallocate zero-range PREALLOC functionality"
test_150ic() {
(( $MDS1_VERSION >= $(version_code 2.16.54) )) ||
skip "need MDS1 version >= 2.16.54 for falloc zero-range"
if [[ "$ost1_FSTYPE" = "zfs" || "$mds1_FSTYPE" = "zfs" ]]; then
skip "zero-range mode is not implemented on OSD ZFS"
fi
[[ "$DOM" == "yes" ]] ||
skip "only check on DoM component"
check_set_fallocate_or_skip
# set a larger dom-size for test
local MB1=1048576
local mdtname=${FSNAME}-MDT0000-mdtlov
local dom_limit_saved=$(do_facet mds1 $LCTL get_param -n \
lod.$mdtname.dom_stripesize)
local dom_limit=$((256 * MB1))
do_facet mds1 $LCTL set_param -n \
lod.$mdtname.dom_stripesize=$dom_limit
stack_trap "do_facet mds1 $LCTL set_param -n \
lod.$mdtname.dom_stripesize=$dom_limit_saved"
stack_trap "rm -f $DIR/$tfile; wait_delete_completed"
echo "Verify fallocate zero-range: range extending the file"
$LFS setstripe -E 256M -L mdt -E eof $DIR/$tfile ||
error "$LFS setstripe DoM failed"
# now let's extend the range to [0, 128M), to trigger BRWs
local offset=0
local length=$((128 * MB1))
touch $DIR/$tfile
out=$(fallocate -z --offset $offset -l $length $DIR/$tfile 2>&1) ||
skip_eopnotsupp "$out|fallocate: offset $offset and len $length"
local features=$(do_facet mds1 "$DEBUGFS -c -R stats $(mdsdevname 1)" |
grep "Filesystem features:")
# Verify zero prealloc worked.
local blocks=$(stat -c '%b' $DIR/$tfile)
if [[ "$features" == *extent* ]]; then
local want_blocks=$((128 * MB1 / 512))
(( blocks >= want_blocks && blocks <= want_blocks + 16 * 8 )) ||
error "zero prealloc failed:$blocks!=$want_blocks"
else
# Given per block 4KB size, per index block could hold
# 1024 block index. 128MB needs data block 32768,
# index block 1(L2) + 32(L1)
local want_blocks=$((128 * MB1 / 512 + 33 * 8))
(( blocks == want_blocks )) ||
error "zero prealloc failed:$blocks!=$want_blocks"
fi
local expect="fde9e0818281836e4fc0edfede2b8762"
local cksum=($(md5sum $DIR/$tfile))
[[ "${cksum[0]}" == "$expect" ]] ||
error "unexpected MD5SUM after fallo(large-zero): ${cksum[0]}"
}
run_test 150ic "Verify fallocate LARGE zero PREALLOC functionality"
#LU-2902 roc_hit was not able to read all values from lproc
function roc_hit_init() {
local osts=${1:-$(osts_nodes)}
local dir=$DIR/$tdir-check
local file=$dir/$tfile
local BEFORE
local AFTER
local idx
test_mkdir $dir
#use setstripe to do a write to every ost
for ((i=0; i < OSTCOUNT; i++)); do
$LFS setstripe -c 1 -i $i $dir || error "$LFS setstripe $file failed"
dd if=/dev/urandom of=$file bs=4k count=4 2>&1 > /dev/null
idx=$(printf %04x $i)
BEFORE=$(get_osd_param $osts *OST$idx stats |
awk '$1 == "cache_access" {sum += $7}
END { printf("%0.0f", sum) }')
cancel_lru_locks osc
cat $file >/dev/null
AFTER=$(get_osd_param $osts *OST$idx stats |
awk '$1 == "cache_access" {sum += $7}
END { printf("%0.0f", sum) }')
echo BEFORE:$BEFORE AFTER:$AFTER
if ! let "AFTER - BEFORE == 4"; then
rm -rf $dir
error "roc_hit is not safe to use"
fi
rm $file
done
rm -rf $dir
}
function roc_hit() {
local osts=${1:-$(osts_nodes)}
echo $(get_osd_param $osts '' stats |
awk '$1 == "cache_hit" {sum += $7}
END { printf("%0.0f", sum) }')
}
function set_cache() {
local osts=$1
local type=$2
local param=$3
local on=1
if [ "$param" == "off" ]; then
on=0;
fi
set_osd_param $osts '' ${type}_cache_enable $on
cancel_lru_locks osc
}
test_151() {
[ $PARALLEL == "yes" ] && skip "skip parallel run"
remote_ost_nodsh && skip "remote OST with nodsh"
(( CLIENT_VERSION == OST1_VERSION )) ||
skip "LU-13081: no interop testing for OSS cache"
local CPAGES=3
local osts=$(osts_nodes)
# check whether obdfilter is cache capable at all
if ! get_osd_param $osts '' read_cache_enable >/dev/null; then
skip "not cache-capable obdfilter"
fi
# check cache is enabled on all obdfilters
if get_osd_param $osts '' read_cache_enable | grep 0; then
skip "oss cache is disabled"
fi
set_osd_param $osts '' writethrough_cache_enable 1
# check write cache is enabled on all obdfilters
if get_osd_param $osts '' writethrough_cache_enable | grep 0; then
skip "oss write cache is NOT enabled"
fi
roc_hit_init $osts
#define OBD_FAIL_OBD_NO_LRU 0x609
do_nodes $osts $LCTL set_param fail_loc=0x609
# pages should be in the case right after write
dd if=/dev/urandom of=$DIR/$tfile bs=4k count=$CPAGES ||
error "dd failed"
local BEFORE=$(roc_hit $osts)
cancel_lru_locks osc
cat $DIR/$tfile >/dev/null
local AFTER=$(roc_hit $osts)
do_nodes $osts $LCTL set_param fail_loc=0
if ! let "AFTER - BEFORE == CPAGES"; then
error "NOT IN CACHE: before: $BEFORE, after: $AFTER"
fi
cancel_lru_locks osc
# invalidates OST cache
do_nodes $osts "echo 1 > /proc/sys/vm/drop_caches"
set_osd_param $osts '' read_cache_enable 0
cat $DIR/$tfile >/dev/null
# now data shouldn't be found in the cache
BEFORE=$(roc_hit $osts)
cancel_lru_locks osc
cat $DIR/$tfile >/dev/null
AFTER=$(roc_hit $osts)
if let "AFTER - BEFORE != 0"; then
error "IN CACHE: before: $BEFORE, after: $AFTER"
fi
set_osd_param $osts '' read_cache_enable 1
rm -f $DIR/$tfile
}
run_test 151 "test cache on oss and controls ==============================="
test_152() {
[ $PARALLEL == "yes" ] && skip "skip parallel run"
local TF="$TMP/$tfile"
# simulate ENOMEM during write
#define OBD_FAIL_OST_NOMEM 0x226
lctl set_param fail_loc=0x80000226
dd if=/dev/urandom of=$TF bs=6096 count=1 || error "dd failed"
cp $TF $DIR/$tfile
sync || error "sync failed"
lctl set_param fail_loc=0
# discard client's cache
cancel_lru_locks osc
# simulate ENOMEM during read
lctl set_param fail_loc=0x80000226
cmp $TF $DIR/$tfile || error "cmp failed"
lctl set_param fail_loc=0
rm -f $TF
}
run_test 152 "test read/write with enomem ============================"
test_153() {
$MULTIOP $DIR/$tfile Ow4096Ycu || error "multiop failed"
}
run_test 153 "test if fdatasync does not crash ======================="
dot_lustre_fid_permission_check() {
local fid=$1
local ffid=$MOUNT/.lustre/fid/$fid
local test_dir=$2
echo "stat fid $fid"
stat $ffid || error "stat $ffid failed."
echo "touch fid $fid"
touch $ffid || error "touch $ffid failed."
echo "write to fid $fid"
cat /etc/hosts > $ffid || error "write $ffid failed."
echo "read fid $fid"
diff /etc/hosts $ffid || error "read $ffid failed."
echo "append write to fid $fid"
cat /etc/hosts >> $ffid || error "append write $ffid failed."
echo "rename fid $fid"
mv $ffid $test_dir/$tfile.1 &&
error "rename $ffid to $tfile.1 should fail."
touch $test_dir/$tfile.1
mv $test_dir/$tfile.1 $ffid &&
error "rename $tfile.1 to $ffid should fail."
rm -f $test_dir/$tfile.1
echo "truncate fid $fid"
$TRUNCATE $ffid 777 || error "truncate $ffid failed."
echo "link fid $fid"
ln -f $ffid $test_dir/tfile.lnk || error "link $ffid failed."
if [[ $($LCTL get_param -n mdc.*-mdc-*.connect_flags) =~ acl ]]; then
id $USER0 || skip_env "missing user $USER0"
echo "setfacl fid $fid"
setfacl -R -m u:$USER0:rwx $ffid ||
error "setfacl $ffid failed"
echo "getfacl fid $fid"
getfacl $ffid || error "getfacl $ffid failed."
fi
echo "unlink fid $fid"
unlink $MOUNT/.lustre/fid/$fid && error "unlink $ffid should fail."
echo "mknod fid $fid"
mknod $ffid c 1 3 && error "mknod $ffid should fail."
fid=[0xf00000400:0x1:0x0]
ffid=$MOUNT/.lustre/fid/$fid
echo "stat non-exist fid $fid"
stat $ffid > /dev/null && error "stat non-exist $ffid should fail."
echo "write to non-exist fid $fid"
cat /etc/hosts > $ffid && error "write non-exist $ffid should fail."
echo "link new fid $fid"
ln $test_dir/$tfile $ffid && error "link $ffid should fail."
mkdir -p $test_dir/$tdir
touch $test_dir/$tdir/$tfile
fid=$($LFS path2fid $test_dir/$tdir)
rc=$?
[ $rc -ne 0 ] &&
error "error: could not get fid for $test_dir/$dir/$tfile."
ffid=$MOUNT/.lustre/fid/$fid
echo "ls $fid"
ls $ffid || error "ls $ffid failed."
echo "touch $fid/$tfile.1"
touch $ffid/$tfile.1 || error "touch $ffid/$tfile.1 failed."
echo "touch $MOUNT/.lustre/fid/$tfile"
touch $MOUNT/.lustre/fid/$tfile && \
error "touch $MOUNT/.lustre/fid/$tfile should fail."
echo "setxattr to $MOUNT/.lustre/fid"
setfattr -n trusted.name1 -v value1 $MOUNT/.lustre/fid
echo "listxattr for $MOUNT/.lustre/fid"
getfattr -d -m "^trusted" $MOUNT/.lustre/fid
echo "delxattr from $MOUNT/.lustre/fid"
setfattr -x trusted.name1 $MOUNT/.lustre/fid
echo "touch invalid fid: $MOUNT/.lustre/fid/[0x200000400:0x2:0x3]"
touch $MOUNT/.lustre/fid/[0x200000400:0x2:0x3] &&
error "touch invalid fid should fail."
echo "touch non-normal fid: $MOUNT/.lustre/fid/[0x1:0x2:0x0]"
touch $MOUNT/.lustre/fid/[0x1:0x2:0x0] &&
error "touch non-normal fid should fail."
echo "rename $tdir to $MOUNT/.lustre/fid"
mrename $test_dir/$tdir $MOUNT/.lustre/fid &&
error "rename to $MOUNT/.lustre/fid should fail."
if [ $MDS1_VERSION -ge $(version_code 2.3.51) ]
then # LU-3547
local old_obf_mode=$(stat --format="%a" $DIR/.lustre/fid)
local new_obf_mode=777
echo "change mode of $DIR/.lustre/fid to $new_obf_mode"
chmod $new_obf_mode $DIR/.lustre/fid ||
error "chmod $new_obf_mode $DIR/.lustre/fid failed"
local obf_mode=$(stat --format=%a $DIR/.lustre/fid)
[ $obf_mode -eq $new_obf_mode ] ||
error "stat $DIR/.lustre/fid returned wrong mode $obf_mode"
echo "restore mode of $DIR/.lustre/fid to $old_obf_mode"
chmod $old_obf_mode $DIR/.lustre/fid ||
error "chmod $old_obf_mode $DIR/.lustre/fid failed"
fi
$OPENFILE -f O_LOV_DELAY_CREATE:O_CREAT $test_dir/$tfile-2
fid=$($LFS path2fid $test_dir/$tfile-2)
if [ $MDS1_VERSION -ge $(version_code 2.6.50) ]
then # LU-5424
echo "cp /etc/passwd $MOUNT/.lustre/fid/$fid"
cp /etc/passwd $MOUNT/.lustre/fid/$fid ||
error "create lov data thru .lustre failed"
fi
echo "cp /etc/passwd $test_dir/$tfile-2"
cp /etc/passwd $test_dir/$tfile-2 ||
error "copy to $test_dir/$tfile-2 failed."
echo "diff /etc/passwd $MOUNT/.lustre/fid/$fid"
diff /etc/passwd $MOUNT/.lustre/fid/$fid ||
error "diff /etc/passwd $MOUNT/.lustre/fid/$fid failed."
rm -rf $test_dir/tfile.lnk
rm -rf $test_dir/$tfile-2
}
test_154A() {
[[ $MDS1_VERSION -lt $(version_code 2.4.1) ]] &&
skip "Need MDS version at least 2.4.1"
local tf=$DIR/$tfile
touch $tf
local fid=$($LFS path2fid $tf)
[ -z "$fid" ] && error "path2fid unable to get $tf FID"
# check that we get the same pathname back
local rootpath
local found
for rootpath in "$MOUNT" "$MOUNT///" "$MOUNT/$tfile"; do
echo "$rootpath $fid"
found=$($LFS fid2path $rootpath "$fid")
[ -z "$found" ] && error "fid2path unable to get '$fid' path"
[ "$found" == "$tf" ] || error "fid2path $found != $tf"
done
# check wrong root path format
rootpath=$MOUNT"_wrong"
found=$($LFS fid2path $rootpath "$fid")
[ -z "$found" ] || error "should fail ($rootpath != $MOUNT)"
}
run_test 154A "lfs path2fid and fid2path basic checks"
test_154B() {
[[ $MDS1_VERSION -lt $(version_code 2.4.1) ]] &&
skip "Need MDS version at least 2.4.1"
mkdir -p $DIR/$tdir || error "mkdir $tdir failed"
touch $DIR/$tdir/$tfile || error "touch $DIR/$tdir/$tfile failed"
local linkea=$($LL_DECODE_LINKEA $DIR/$tdir/$tfile | grep 'pfid')
[ -z "$linkea" ] && error "decode linkea $DIR/$tdir/$tfile failed"
local name=$(echo $linkea | awk '/pfid/ {print $5}' | sed -e "s/'//g")
local PFID=$(echo $linkea | awk '/pfid/ {print $3}' | sed -e "s/,//g")
# check that we get the same pathname
echo "PFID: $PFID, name: $name"
local FOUND=$($LFS fid2path $MOUNT "$PFID")
[ -z "$FOUND" ] && error "fid2path unable to get $PFID path"
[ "$FOUND/$name" != "$DIR/$tdir/$tfile" ] &&
error "ll_decode_linkea has $FOUND/$name != $DIR/$tdir/$tfile"
rm -rf $DIR/$tdir || error "Can not delete directory $DIR/$tdir"
}
run_test 154B "verify the ll_decode_linkea tool"
test_154C() {
(( $MDS1_VERSION >= $(version_code 2.16.54) )) ||
skip "need MDS1 version >= 2.16.54 for OST FID lookup"
(( $OSTCOUNT >= 2 )) || skip_env "needs >= 2 OSTs"
local tf=$DIR/$tfile
$LFS setstripe -c2 -i0 -S1M $tf
dd if=/dev/zero of=$tf bs=1M count=8
stack_trap "rm -rf $tf; wait_delete_completed"
local stripe_info=$($LFS getstripe $tf)
# for each OST object, execute fid2path
while read -r lfid ofid; do
# keep only valid lines to parse
[[ "$lfid" == "l_fid:" ]] || continue
found=$($LFS fid2path "$MOUNT" "$ofid")
echo "ost_fid=$ofid with outpath=$found"
[[ -n "$found" ]] ||
error "fid2path failed on $ofid"
[[ "$found" == "$tf" ]] || error "fid2path $found != $tf"
done < <($LFS getstripe -y $tf)
}
run_test 154C "lfs fid2path on OST FID"
test_154a() {
[ $PARALLEL == "yes" ] && skip "skip parallel run"
[ -n "$FILESET" ] && skip "SKIP due to FILESET set"
(( $MDS1_VERSION >= $(version_code 2.2.51) )) ||
skip "Need MDS version at least 2.2.51"
[ -z "$(which setfacl)" ] && skip_env "must have setfacl tool"
cp /etc/hosts $DIR/$tfile
fid=$($LFS path2fid $DIR/$tfile)
rc=$?
[ $rc -ne 0 ] && error "error: could not get fid for $DIR/$tfile."
dot_lustre_fid_permission_check "$fid" $DIR ||
error "dot lustre permission check $fid failed"
ls -a $MOUNT | grep "\.lustre" && error ".lustre should not be listed"
rm -rf $MOUNT/.lustre && error ".lustre is not allowed to be unlinked"
touch $MOUNT/.lustre/file &&
error "creation is not allowed under .lustre"
mkdir $MOUNT/.lustre/dir &&
error "mkdir is not allowed under .lustre"
rm -rf $DIR/$tfile
}
run_test 154a "Open-by-FID"
test_154b() {
[ $PARALLEL == "yes" ] && skip "skip parallel run"
[ -n "$FILESET" ] && skip "SKIP due to FILESET set"
[ $MDSCOUNT -lt 2 ] && skip_env "needs >= 2 MDTs"
[[ $MDS1_VERSION -ge $(version_code 2.2.51) ]] ||
skip "Need MDS version at least 2.2.51"
local remote_dir=$DIR/$tdir/remote_dir
local MDTIDX=1
local rc=0
mkdir -p $DIR/$tdir
$LFS mkdir -i $MDTIDX $remote_dir ||
error "create remote directory failed"
cp /etc/hosts $remote_dir/$tfile
fid=$($LFS path2fid $remote_dir/$tfile)
rc=$?
[ $rc -ne 0 ] && error "error: could not get fid for $remote_dir/$tfile"
dot_lustre_fid_permission_check "$fid" $remote_dir ||
error "dot lustre permission check $fid failed"
rm -rf $DIR/$tdir
}
run_test 154b "Open-by-FID for remote directory"
test_154c() {
[[ $MDS1_VERSION -lt $(version_code 2.4.1) ]] &&
skip "Need MDS version at least 2.4.1"
touch $DIR/$tfile.1 $DIR/$tfile.2 $DIR/$tfile.3
local FID1=$($LFS path2fid $DIR/$tfile.1)
local FID2=$($LFS path2fid $DIR/$tfile.2)
local FID3=$($LFS path2fid $DIR/$tfile.3)
local N=1
$LFS path2fid $DIR/$tfile.[123] | while read PATHNAME FID; do
[ "$PATHNAME" = "$DIR/$tfile.$N:" ] ||
error "path2fid pathname $PATHNAME != $DIR/$tfile.$N:"
local want=FID$N
[ "$FID" = "${!want}" ] ||
error "path2fid $PATHNAME FID $FID != FID$N ${!want}"
N=$((N + 1))
done
$LFS fid2path $MOUNT "$FID1" "$FID2" "$FID3" | while read PATHNAME;
do
[ "$PATHNAME" = "$DIR/$tfile.$N" ] ||
error "fid2path pathname $PATHNAME != $DIR/$tfile.$N:"
N=$((N + 1))
done
}
run_test 154c "lfs path2fid and fid2path multiple arguments"
test_154d() {
remote_mds_nodsh && skip "remote MDS with nodsh"
[[ $MDS1_VERSION -lt $(version_code 2.5.53) ]] &&
skip "Need MDS version at least 2.5.53"
if remote_mds; then
nid=$($LCTL list_nids | sed "s/\./\\\./g")
else
nid="0@lo"
fi
local proc_ofile="mdt.*.exports.'$nid'.open_files"
local fd
local cmd
rm -f $DIR/$tfile
touch $DIR/$tfile
local fid=$($LFS path2fid $DIR/$tfile)
# Open the file
fd=$(free_fd)
cmd="exec $fd<$DIR/$tfile"
eval $cmd
local fid_list=$(do_facet $SINGLEMDS $LCTL get_param $proc_ofile)
echo "$fid_list" | grep "$fid"
rc=$?
cmd="exec $fd>/dev/null"
eval $cmd
if [ $rc -ne 0 ]; then
error "FID $fid not found in open files list $fid_list"
fi
}
run_test 154d "Verify open file fid"
test_154e()
{
[[ $MDS1_VERSION -lt $(version_code 2.6.50) ]] &&
skip "Need MDS version at least 2.6.50"
if ls -a $MOUNT | grep -q '^\.lustre$'; then
error ".lustre returned by readdir"
fi
}
run_test 154e ".lustre is not returned by readdir"
test_154ea()
{
stack_trap "rm -rf $MOUNT/[abcdeg]*" EXIT
# long specially selected names are to displace .lustre out of
# first directory block
touch $MOUNT/$(printf "a%0100x" 6)
for i in 11 15
do
touch $MOUNT/$(printf "a%0254x" $i)
done
touch $MOUNT/$(printf "b%0254x" $i)
for i in 16 6 7
do
touch $MOUNT/$(printf "c%0254x" $i)
done
for i in 4 10 8
do
touch $MOUNT/$(printf "d%0254x" $i)
done
for i in 1 14
do
touch /mnt/lustre/$(printf "e%0254x" $i)
done
for i in 13 14
do
touch /mnt/lustre/$(printf "g%0254x" $i)
done
if ls -a $MOUNT | grep -q '^\.lustre$'; then
error ".lustre returned by readdir"
fi
}
run_test 154ea ".lustre is not returned by readdir (2)"
test_154f() {
# create parent directory on a single MDT to avoid cross-MDT hardlinks
mkdir_on_mdt0 $DIR/$tdir
# test dirs inherit from its stripe
mkdir -p $DIR/$tdir/foo1 || error "mkdir error"
mkdir -p $DIR/$tdir/foo2 || error "mkdir error"
cp /etc/hosts $DIR/$tdir/foo1/$tfile
ln $DIR/$tdir/foo1/$tfile $DIR/$tdir/foo2/link
touch $DIR/f
# get fid of parents
local FID0=$($LFS path2fid $DIR/$tdir)
local FID1=$($LFS path2fid $DIR/$tdir/foo1)
local FID2=$($LFS path2fid $DIR/$tdir/foo2)
local FID3=$($LFS path2fid $DIR)
# check that path2fid --parents returns expected <parent_fid>/name
# 1) test for a directory (single parent)
local parent=$($LFS path2fid --parents $DIR/$tdir/foo1)
[ "$parent" == "$FID0/foo1" ] ||
error "expected parent: $FID0/foo1, got: $parent"
# 2) test for a file with nlink > 1 (multiple parents)
parent=$($LFS path2fid --parents $DIR/$tdir/foo1/$tfile)
echo "$parent" | grep -F "$FID1/$tfile" ||
error "$FID1/$tfile not returned in parent list"
echo "$parent" | grep -F "$FID2/link" ||
error "$FID2/link not returned in parent list"
# 3) get parent by fid
local file_fid=$($LFS path2fid $DIR/$tdir/foo1/$tfile)
parent=$($LFS path2fid --parents $MOUNT/.lustre/fid/$file_fid)
echo "$parent" | grep -F "$FID1/$tfile" ||
error "$FID1/$tfile not returned in parent list (by fid)"
echo "$parent" | grep -F "$FID2/link" ||
error "$FID2/link not returned in parent list (by fid)"
# 4) test for entry in root directory
parent=$($LFS path2fid --parents $DIR/f)
echo "$parent" | grep -F "$FID3/f" ||
error "$FID3/f not returned in parent list"
# 5) test it on root directory
[ -z "$FILESET" ] && {
# LU_ROOT_FID is wrong for fileset
[ -z "$($LFS path2fid --parents $MOUNT 2>/dev/null)" ] ||
error "$MOUNT should not have parents"
}
# enable xattr caching and check that linkea is correctly updated
local save="$TMP/$TESTSUITE-$TESTNAME.parameters"
save_lustre_params client "llite.*.xattr_cache" > $save
lctl set_param llite.*.xattr_cache 1
# 6.1) linkea update on rename
mv $DIR/$tdir/foo1/$tfile $DIR/$tdir/foo2/$tfile.moved
# get parents by fid
parent=$($LFS path2fid --parents $MOUNT/.lustre/fid/$file_fid)
# foo1 should no longer be returned in parent list
echo "$parent" | grep -F "$FID1" &&
error "$FID1 should no longer be in parent list"
# the new path should appear
echo "$parent" | grep -F "$FID2/$tfile.moved" ||
error "$FID2/$tfile.moved is not in parent list"
# 6.2) linkea update on unlink
rm -f $DIR/$tdir/foo2/link
parent=$($LFS path2fid --parents $MOUNT/.lustre/fid/$file_fid)
# foo2/link should no longer be returned in parent list
echo "$parent" | grep -F "$FID2/link" &&
error "$FID2/link should no longer be in parent list"
true
rm -f $DIR/f
restore_lustre_params < $save
rm -f $save
}
run_test 154f "get parent fids by reading link ea"
test_154g()
{
[[ $MDS1_VERSION -ge $(version_code 2.6.92) &&
$CLIENT_VERSION -gt $(version_code 2.6.99) ]] ||
skip "Need MDS version at least 2.6.92"
mkdir_on_mdt0 $DIR/$tdir
llapi_fid_test -d $DIR/$tdir
}
run_test 154g "various llapi FID tests"
test_154h()
{
(( $CLIENT_VERSION >= $(version_code 2.15.55.1) )) ||
skip "Need client at least version 2.15.55.1"
# Create an empty file
touch $DIR/$tfile
# Get FID (interactive mode) and save under $TMP/$tfile.log
$LFS 2>&1 <<-EOF | tee $TMP/$tfile.log
path2fid $DIR/$tfile
EOF
fid=$(cat $TMP/$tfile.log)
# $fid should not be empty
[[ ! -z $fid ]] || error "FID is empty"
$LFS rmfid $DIR "$fid" || error "rmfid failed for $fid"
}
run_test 154h "Verify interactive path2fid"
test_154i()
{
local long=thislongpathnameisforaverydeepsubdirthatwewanttotestagainst
local depth
local path
local max
local fid
(( $MDS1_VERSION >= $(version_code v2_15_65-55-gf5251cb7af) )) ||
skip "need MDS >= 2.15.65.55 for fid2path > PATH_MAX handling"
mkdir -p $DIR/$tdir
cd $DIR/$tdir
# create a directory tree with full path longer than PATH_MAX=4096
max=$((4096 / $(wc -c <<< $long) + 5))
for (( depth = 0; depth <= max; depth++)); do
mkdir -v $long$depth || error "mkdir $long$depth failed"
cd $long$depth
done
fid=$($LFS path2fid .) || error "path2fid failed"
path=$($LFS fid2path $MOUNT $fid) || error "fid2path failed (1)"
echo -e "Path for fid $fid is:\n$path"
path=$($LFS fid2path $(cd ..; pwd) $fid) || error "fid2path failed (2)"
echo -e "Path for fid $fid is:\n$path"
}
run_test 154i "fid2path for path longer than PATH_MAX"
test_154j()
{
(( $MDSCOUNT >= 2 )) || skip_env "needs >= 2 MDTs"
(( $MDS1_VERSION >= $(version_code v2_15_65-55-gf5251cb7af) )) ||
skip "need MDS >= 2.15.65.55 for fid2path > PATH_MAX handling"
local long=thislongpathnameisforaverydeepsubdirthatwewanttotestagainst
local depth
local path
local fid
# place $tdir on MDT0 and create a remote dir on MDT1 so that
# fid2path has to cross the MDT boundary
$LFS mkdir -i 0 $DIR/$tdir || error "mkdir MDT0 failed"
$LFS mkdir -i 1 $DIR/$tdir/remote || error "lfs mkdir -i 1 failed"
cd $DIR/$tdir/remote
# Create 66 nested dirs. Each ~60 chars, total MDT1 path ~4088.
# Combined with MDT0 path ($tdir ~12+ chars), exceeds PATH_MAX.
for (( depth = 0; depth <= 65; depth++ )); do
mkdir $long$depth || error "mkdir $long$depth failed"
cd $long$depth
done
fid=$($LFS path2fid .) || error "path2fid failed"
path=$($LFS fid2path $MOUNT $fid) || error "fid2path failed"
echo -e "Path for fid $fid is:\n$path"
}
run_test 154j "fid2path for long path crossing MDT boundary"
test_155_small_load() {
local temp=$TMP/$tfile
local file=$DIR/$tfile
dd if=/dev/urandom of=$temp bs=6096 count=1 || \
error "dd of=$temp bs=6096 count=1 failed"
cp $temp $file
cancel_lru_locks $OSC
cmp $temp $file || error "$temp $file differ"
$TRUNCATE $temp 6000
$TRUNCATE $file 6000
cmp $temp $file || error "$temp $file differ (truncate1)"
echo "12345" >>$temp
echo "12345" >>$file
cmp $temp $file || error "$temp $file differ (append1)"
echo "12345" >>$temp
echo "12345" >>$file
cmp $temp $file || error "$temp $file differ (append2)"
rm -f $temp $file
true
}
test_155_big_load() {
remote_ost_nodsh && skip "remote OST with nodsh"
local temp=$TMP/$tfile
local file=$DIR/$tfile
free_min_max
local cache_size=$(do_facet ost$((MAXI+1)) \
"awk '/cache/ {sum+=\\\$4} END {print sum}' /proc/cpuinfo")
# LU-16042: can not get the cache size on Arm64 VM here, fallback to a
# pre-set value
if [ -z "$cache_size" ]; then
cache_size=256
fi
local large_file_size=$((cache_size * 2))
echo "OSS cache size: $cache_size KB"
echo "Large file size: $large_file_size KB"
[ $MAXV -le $large_file_size ] &&
skip_env "max available OST size needs > $large_file_size KB"
$LFS setstripe $file -c 1 -i $MAXI || error "$LFS setstripe $file failed"
dd if=/dev/urandom of=$temp bs=$large_file_size count=1k ||
error "dd of=$temp bs=$large_file_size count=1k failed"
cp $temp $file
ls -lh $temp $file
cancel_lru_locks osc
cmp $temp $file || error "$temp $file differ"
rm -f $temp $file
true
}
save_writethrough() {
local facets=$(get_facets OST)
save_lustre_params $facets "osd-*.*.writethrough_cache_enable" > $1
}
test_155a() {
[ $PARALLEL == "yes" ] && skip "skip parallel run"
local p="$TMP/$TESTSUITE-$TESTNAME.parameters"
local osts=$(osts_nodes)
save_writethrough $p
set_cache $osts read on
set_cache $osts writethrough on
test_155_small_load
restore_lustre_params < $p
rm -f $p
}
run_test 155a "Verify small file correctness: read cache:on write_cache:on"
test_155b() {
[ $PARALLEL == "yes" ] && skip "skip parallel run"
local p="$TMP/$TESTSUITE-$TESTNAME.parameters"
local osts=$(osts_nodes)
save_writethrough $p
set_cache $osts read on
set_cache $osts writethrough off
test_155_small_load
restore_lustre_params < $p
rm -f $p
}
run_test 155b "Verify small file correctness: read cache:on write_cache:off"
test_155c() {
[ $PARALLEL == "yes" ] && skip "skip parallel run"
local p="$TMP/$TESTSUITE-$TESTNAME.parameters"
local osts=$(osts_nodes)
save_writethrough $p
set_cache $osts read off
set_cache $osts writethrough on
test_155_small_load
restore_lustre_params < $p
rm -f $p
}
run_test 155c "Verify small file correctness: read cache:off write_cache:on"
test_155d() {
[ $PARALLEL == "yes" ] && skip "skip parallel run"
local p="$TMP/$TESTSUITE-$TESTNAME.parameters"
local osts=$(osts_nodes)
save_writethrough $p
set_cache $osts read off
set_cache $osts writethrough off
test_155_small_load
restore_lustre_params < $p
rm -f $p
}
run_test 155d "Verify small file correctness: read cache:off write_cache:off"
test_155e() {
[ $PARALLEL == "yes" ] && skip "skip parallel run"
local p="$TMP/$TESTSUITE-$TESTNAME.parameters"
local osts=$(osts_nodes)
save_writethrough $p
set_cache $osts read on
set_cache $osts writethrough on
test_155_big_load
restore_lustre_params < $p
rm -f $p
}
run_test 155e "Verify big file correctness: read cache:on write_cache:on"
test_155f() {
[ $PARALLEL == "yes" ] && skip "skip parallel run"
local p="$TMP/$TESTSUITE-$TESTNAME.parameters"
local osts=$(osts_nodes)
save_writethrough $p
set_cache $osts read on
set_cache $osts writethrough off
test_155_big_load
restore_lustre_params < $p
rm -f $p
}
run_test 155f "Verify big file correctness: read cache:on write_cache:off"
test_155g() {
[ $PARALLEL == "yes" ] && skip "skip parallel run"
local p="$TMP/$TESTSUITE-$TESTNAME.parameters"
local osts=$(osts_nodes)
save_writethrough $p
set_cache $osts read off
set_cache $osts writethrough on
test_155_big_load
restore_lustre_params < $p
rm -f $p
}
run_test 155g "Verify big file correctness: read cache:off write_cache:on"
test_155h() {
[ $PARALLEL == "yes" ] && skip "skip parallel run"
local p="$TMP/$TESTSUITE-$TESTNAME.parameters"
local osts=$(osts_nodes)
save_writethrough $p
set_cache $osts read off
set_cache $osts writethrough off
test_155_big_load
restore_lustre_params < $p
rm -f $p
}
run_test 155h "Verify big file correctness: read cache:off write_cache:off"
test_156() {
[ $PARALLEL == "yes" ] && skip "skip parallel run"
remote_ost_nodsh && skip "remote OST with nodsh"
[ $OST1_VERSION -lt $(version_code 2.6.93) ] &&
skip "stats not implemented on old servers"
[ "$ost1_FSTYPE" = "zfs" ] &&
skip "LU-1956/LU-2261: stats not implemented on OSD ZFS"
(( CLIENT_VERSION == OST1_VERSION )) ||
skip "LU-13081: no interop testing for OSS cache"
local CPAGES=3
local BEFORE
local AFTER
local file="$DIR/$tfile"
local p="$TMP/$TESTSUITE-$TESTNAME.parameters"
local osts=$(osts_nodes)
save_writethrough $p
roc_hit_init $osts
log "Turn on read and write cache"
set_cache $osts read on
set_cache $osts writethrough on
log "Write data and read it back."
log "Read should be satisfied from the cache."
dd if=/dev/urandom of=$file bs=4k count=$CPAGES || error "dd failed"
BEFORE=$(roc_hit $osts)
cancel_lru_locks osc
cat $file >/dev/null
AFTER=$(roc_hit $osts)
if ! let "AFTER - BEFORE == CPAGES"; then
error "NOT IN CACHE (2): before: $BEFORE, after: $AFTER"
else
log "cache hits: before: $BEFORE, after: $AFTER"
fi
log "Read again; it should be satisfied from the cache."
BEFORE=$AFTER
cancel_lru_locks osc
cat $file >/dev/null
AFTER=$(roc_hit $osts)
if ! let "AFTER - BEFORE == CPAGES"; then
error "NOT IN CACHE (3): before: $BEFORE, after: $AFTER"
else
log "cache hits:: before: $BEFORE, after: $AFTER"
fi
log "Turn off the read cache and turn on the write cache"
set_cache $osts read off
set_cache $osts writethrough on
log "Read again; it should be satisfied from the cache."
BEFORE=$(roc_hit $osts)
cancel_lru_locks osc
cat $file >/dev/null
AFTER=$(roc_hit $osts)
if ! let "AFTER - BEFORE == CPAGES"; then
error "NOT IN CACHE (4): before: $BEFORE, after: $AFTER"
else
log "cache hits:: before: $BEFORE, after: $AFTER"
fi
if [ $OST1_VERSION -lt $(version_code 2.12.55) ]; then
# > 2.12.56 uses pagecache if cached
log "Read again; it should not be satisfied from the cache."
BEFORE=$AFTER
cancel_lru_locks osc
cat $file >/dev/null
AFTER=$(roc_hit $osts)
if ! let "AFTER - BEFORE == 0"; then
error "IN CACHE (5): before: $BEFORE, after: $AFTER"
else
log "cache hits:: before: $BEFORE, after: $AFTER"
fi
fi
log "Write data and read it back."
log "Read should be satisfied from the cache."
dd if=/dev/urandom of=$file bs=4k count=$CPAGES || error "dd failed"
BEFORE=$(roc_hit $osts)
cancel_lru_locks osc
cat $file >/dev/null
AFTER=$(roc_hit $osts)
if ! let "AFTER - BEFORE == CPAGES"; then
error "NOT IN CACHE (6): before: $BEFORE, after: $AFTER"
else
log "cache hits:: before: $BEFORE, after: $AFTER"
fi
if [ $OST1_VERSION -lt $(version_code 2.12.55) ]; then
# > 2.12.56 uses pagecache if cached
log "Read again; it should not be satisfied from the cache."
BEFORE=$AFTER
cancel_lru_locks osc
cat $file >/dev/null
AFTER=$(roc_hit $osts)
if ! let "AFTER - BEFORE == 0"; then
error "IN CACHE (7): before: $BEFORE, after: $AFTER"
else
log "cache hits:: before: $BEFORE, after: $AFTER"
fi
fi
log "Turn off read and write cache"
set_cache $osts read off
set_cache $osts writethrough off
log "Write data and read it back"
log "It should not be satisfied from the cache."
rm -f $file
dd if=/dev/urandom of=$file bs=4k count=$CPAGES || error "dd failed"
cancel_lru_locks osc
BEFORE=$(roc_hit $osts)
cat $file >/dev/null
AFTER=$(roc_hit $osts)
if ! let "AFTER - BEFORE == 0"; then
error_ignore bz20762 "IN CACHE (8):before:$BEFORE,after:$AFTER"
else
log "cache hits:: before: $BEFORE, after: $AFTER"
fi
log "Turn on the read cache and turn off the write cache"
set_cache $osts read on
set_cache $osts writethrough off
log "Write data and read it back"
log "It should not be satisfied from the cache."
rm -f $file
dd if=/dev/urandom of=$file bs=4k count=$CPAGES || error "dd failed"
BEFORE=$(roc_hit $osts)
cancel_lru_locks osc
cat $file >/dev/null
AFTER=$(roc_hit $osts)
if ! let "AFTER - BEFORE == 0"; then
error_ignore bz20762 "IN CACHE (9):before:$BEFORE,after:$AFTER"
else
log "cache hits:: before: $BEFORE, after: $AFTER"
fi
log "Read again; it should be satisfied from the cache."
BEFORE=$(roc_hit $osts)
cancel_lru_locks osc
cat $file >/dev/null
AFTER=$(roc_hit $osts)
if ! let "AFTER - BEFORE == CPAGES"; then
error "NOT IN CACHE (1): before: $BEFORE, after: $AFTER"
else
log "cache hits:: before: $BEFORE, after: $AFTER"
fi
restore_lustre_params < $p
rm -f $p $file
}
run_test 156 "Verification of tunables"
test_157a()
{
(( $MDS1_VERSION >= $(version_code v2_16_51-30-g84100d7f7d) )) ||
skip "Need MDS >= 2.16.52 for lustre.pin support"
mkdir_on_mdt0 $DIR/$tdir
llapi_pool_test -d $DIR/$tdir ||
error "llapi_pool_test failed"
}
run_test 157a "llapi pool pinning API tests"
test_157b()
{
[ "$UID" != 0 ] && skip "must run as root"
[ -z "$(lctl get_param -n mdc.*-mdc-*.connect_flags | grep xattr)" ] &&
skip_env "must have user_xattr"
[ -z "$(which setfattr 2>/dev/null)" ] &&
skip_env "could not find setfattr"
[ -z "$(which getfattr 2>/dev/null)" ] &&
skip_env "could not find getfattr"
(( $MDS1_VERSION >= $(version_code v2_17_53-1-g1bf8713c57) )) ||
skip "Need MDS >= 2.17.53 for lustre.pin support"
local dir=$DIR/$tdir
local file=$dir/$tfile
local subdir=$dir/subdir
local noxatrr_file=$dir/$tfile.noxattr
local pinval="[pool: pininherit]"
local parent_val file_val subdir_val noxatrr_file_val
test_mkdir $dir
touch $noxatrr_file || error "touch $noxatrr_file failed"
setfattr -n lustre.pin -v "$pinval" $dir ||
error "failed to set lustre.pin on $dir"
touch $file || error "touch $file failed"
mkdir $subdir || error "mkdir $subdir failed"
parent_val=$(getfattr --only-values -n lustre.pin $dir 2> /dev/null) ||
error "getfattr lustre.pin on parent failed"
file_val=$(getfattr --only-values -n lustre.pin $file 2> /dev/null) ||
error "getfattr lustre.pin on file failed"
subdir_val=$(getfattr --only-values -n lustre.pin $subdir 2> /dev/null) ||
error "getfattr lustre.pin on subdir failed"
noxatrr_file_val=$(getfattr --only-values -n lustre.pin $noxatrr_file 2>/dev/null)
[ "$file_val" = "$parent_val" ] ||
error "file lustre.pin '$file_val' != parent '$parent_val'"
[ "$subdir_val" = "$parent_val" ] ||
error "subdir lustre.pin '$subdir_val' != parent '$parent_val'"
[ -z "$noxatrr_file_val" ] ||
error "noxatrr_file lustre.pin '$noxatrr_file_val' != ''"
}
run_test 157b "lustre.pin inheritance on create"
test_160a() {
[ $PARALLEL == "yes" ] && skip "skip parallel run"
remote_mds_nodsh && skip "remote MDS with nodsh"
[ $MDS1_VERSION -ge $(version_code 2.2.0) ] ||
skip "Need MDS version at least 2.2.0"
changelog_register || error "changelog_register failed"
local cl_user="${CL_USERS[$SINGLEMDS]%% *}"
changelog_users $SINGLEMDS | grep -q $cl_user ||
error "User $cl_user not found in changelog_users"
mkdir_on_mdt0 $DIR/$tdir
# change something
test_mkdir -p $DIR/$tdir/pics/2008/zachy
changelog_clear 0 || error "changelog_clear failed"
touch $DIR/$tdir/pics/2008/zachy/$tfile # open 1
cp /etc/hosts $DIR/$tdir/pics/2008/zachy/pic1.jpg # open 2
mv $DIR/$tdir/pics/2008/zachy $DIR/$tdir/pics/zach
ln $DIR/$tdir/pics/zach/pic1.jpg $DIR/$tdir/pics/2008/portland.jpg
ln -s $DIR/$tdir/pics/2008/portland.jpg $DIR/$tdir/pics/desktop.jpg
rm $DIR/$tdir/pics/desktop.jpg
echo "verifying changelog mask"
changelog_chmask "-MKDIR"
changelog_chmask "-CLOSE"
test_mkdir -p $DIR/$tdir/pics/zach/sofia # not logged
echo "zzzzzz" > $DIR/$tdir/pics/zach/file # not logged
changelog_chmask "+MKDIR"
changelog_chmask "+CLOSE"
test_mkdir -p $DIR/$tdir/pics/2008/sofia # mkdir 1
echo "zzzzzz" > $DIR/$tdir/pics/zach/file # open 3
MKDIRS=$(changelog_dump | grep -c "MKDIR")
CLOSES=$(changelog_dump | grep -c "CLOSE")
[ $MKDIRS -eq 1 ] || error "MKDIR changelog mask count $MKDIRS != 1"
[ $CLOSES -eq 3 ] || error "CLOSE changelog mask count $CLOSES != 3"
# verify contents
echo "verifying target fid"
local fidc=$(changelog_extract_field "CREAT" "$tfile" "t=")
local fidf=$($LFS path2fid $DIR/$tdir/pics/zach/$tfile)
[ "$fidc" == "$fidf" ] ||
error "changelog '$tfile' fid $fidc != file fid $fidf"
echo "verifying parent fid"
# The FID returned from the Changelog may be the directory shard on
# a different MDT, and not the FID returned by path2fid on the parent.
# Instead of comparing FIDs, verify that fid2path(fidp) is correct,
# since this is what will matter when recreating this file in the tree.
local fidp=$(changelog_extract_field "CREAT" "$tfile" "p=")
local pathp=$($LFS fid2path $MOUNT "$fidp")
[ "${pathp%/}" == "$DIR/$tdir/pics/zach" ] ||
error "changelog fid2path($fidc) $pathp != $DIR/$tdir/pics/zach"
echo "getting records for $cl_user"
changelog_users $SINGLEMDS
local user_rec1=$(changelog_user_rec $SINGLEMDS $cl_user)
local nclr=3
__changelog_clear $SINGLEMDS $cl_user +$nclr ||
error "changelog_clear failed"
local user_rec2=$(changelog_user_rec $SINGLEMDS $cl_user)
echo "verifying user clear: $user_rec1 + $nclr == $user_rec2"
[ $user_rec2 == $((user_rec1 + nclr)) ] ||
error "user index expect $user_rec1 + $nclr != $user_rec2"
local min0_rec=$(changelog_users $SINGLEMDS |
awk 'min == "" || $2 < min { min = $2 }; END { print min }')
local first_rec=$($LFS changelog $(facet_svc $SINGLEMDS) |
awk '{ print $1; exit; }')
changelog_dump | tail -n 5
echo "verifying user min purge: $min0_rec + 1 == $first_rec"
[ $first_rec == $((min0_rec + 1)) ] ||
error "first index should be $min0_rec + 1 not $first_rec"
# LU-3446 changelog index reset on MDT restart
local cur_rec1=$(changelog_users $SINGLEMDS |
awk '/^current.index:/ { print $NF }')
changelog_clear 0 ||
error "clear all changelog records for $cl_user failed"
stop $SINGLEMDS || error "Fail to stop $SINGLEMDS"
start $SINGLEMDS $(mdsdevname ${SINGLEMDS//mds/}) $MDS_MOUNT_OPTS ||
error "Fail to start $SINGLEMDS"
local cur_rec2=$(changelog_users $SINGLEMDS |
awk '/^current.index:/ { print $NF }')
echo "verifying index survives MDT restart: $cur_rec1 == $cur_rec2"
[ $cur_rec1 == $cur_rec2 ] ||
error "current index should be $cur_rec1 not $cur_rec2"
echo "verifying users from this test are deregistered"
changelog_deregister || error "changelog_deregister failed"
changelog_users $SINGLEMDS | grep -q $cl_user &&
error "User '$cl_user' still in changelog_users"
# lctl get_param -n mdd.*.changelog_users
# current_index: 144
# ID index (idle seconds)
# cl3 144 (2) mask=<list>
if [ -z "$(changelog_users $SINGLEMDS | grep -v current.index)" ]; then
# this is the normal case where all users were deregistered
# make sure no new records are added when no users are present
local last_rec1=$(changelog_users $SINGLEMDS |
awk '/^current.index:/ { print $NF }')
touch $DIR/$tdir/chloe
local last_rec2=$(changelog_users $SINGLEMDS |
awk '/^current.index:/ { print $NF }')
echo "verify changelogs are off: $last_rec1 == $last_rec2"
[ $last_rec1 == $last_rec2 ] || error "changelogs not off"
else
# any changelog users must be leftovers from a previous test
changelog_users $SINGLEMDS
echo "other changelog users; can't verify off"
fi
}
run_test 160a "changelog sanity"
test_160b() { # LU-3587
[ $PARALLEL == "yes" ] && skip "skip parallel run"
remote_mds_nodsh && skip "remote MDS with nodsh"
[ $MDS1_VERSION -ge $(version_code 2.2.0) ] ||
skip "Need MDS version at least 2.2.0"
changelog_register || error "changelog_register failed"
local cl_user="${CL_USERS[$SINGLEMDS]%% *}"
changelog_users $SINGLEMDS | grep -q $cl_user ||
error "User '$cl_user' not found in changelog_users"
local longname1=$(str_repeat a 255)
local longname2=$(str_repeat b 255)
cd $DIR
echo "creating very long named file"
touch $longname1 || error "create of '$longname1' failed"
echo "renaming very long named file"
mv $longname1 $longname2
changelog_dump | grep RENME | tail -n 5
rm -f $longname2
}
run_test 160b "Verify that very long rename doesn't crash in changelog"
test_160c() {
[ $PARALLEL == "yes" ] && skip "skip parallel run"
remote_mds_nodsh && skip "remote MDS with nodsh"
[[ $MDS1_VERSION -gt $(version_code 2.5.57) ]] ||
[[ $MDS1_VERSION -gt $(version_code 2.5.1) &&
$MDS1_VERSION -lt $(version_code 2.5.50) ]] ||
skip "Need MDS version at least 2.5.58 or 2.5.2+"
local rc=0
# Registration step
changelog_register || error "changelog_register failed"
rm -rf $DIR/$tdir
mkdir -p $DIR/$tdir
$MCREATE $DIR/$tdir/foo_160c
changelog_chmask "-TRUNC"
$TRUNCATE $DIR/$tdir/foo_160c 200
changelog_chmask "+TRUNC"
$TRUNCATE $DIR/$tdir/foo_160c 199
changelog_dump | tail -n 5
local truncs=$(changelog_dump | tail -n 5 | grep -c TRUNC)
[ $truncs -eq 1 ] || error "TRUNC changelog mask count $truncs != 1"
}
run_test 160c "verify that changelog log catch the truncate event"
test_160d() {
remote_mds_nodsh && skip "remote MDS with nodsh"
[ $MDSCOUNT -lt 2 ] && skip_env "needs >= 2 MDTs"
[ $PARALLEL == "yes" ] && skip "skip parallel run"
[[ $MDS1_VERSION -ge $(version_code 2.7.60) ]] ||
skip "Need MDS version at least 2.7.60"
# Registration step
changelog_register || error "changelog_register failed"
mkdir -p $DIR/$tdir/migrate_dir
local src_mdt=$($LFS getdirstripe -m $DIR/$tdir/migrate_dir)
local tgt_mdt=1
[[ "$src_mdt" == "$tgt_mdt" ]] && tgt_mdt=0
changelog_clear 0 || error "changelog_clear failed"
$LFS migrate -m $tgt_mdt $DIR/$tdir/migrate_dir || error "migrate fails"
changelog_dump | tail -n 5
local migrates=$(changelog_dump | grep -c "MIGRT")
[ $migrates -eq 1 ] || error "MIGRATE changelog count $migrates != 1"
}
run_test 160d "verify that changelog log catch the migrate event"
test_160e() {
remote_mds_nodsh && skip "remote MDS with nodsh"
# Create a user
changelog_register || error "changelog_register failed"
local MDT0=$(facet_svc $SINGLEMDS)
local rc
# No user (expect fail)
do_facet $SINGLEMDS $LCTL --device $MDT0 changelog_deregister
rc=$?
if [ $rc -eq 0 ]; then
error "Should fail without user"
elif [ $rc -ne 4 ]; then
error "changelog_deregister failed with $rc, expect 4(CMD_HELP)"
fi
# Delete a future user (expect fail)
do_facet $SINGLEMDS $LCTL --device $MDT0 changelog_deregister "cl77"
rc=$?
if [ $rc -eq 0 ]; then
error "Deleted non-existant user cl77"
elif [ $rc -ne 2 ]; then
error "changelog_deregister failed with $rc, expect 2 (ENOENT)"
fi
# Clear to a bad index (1 billion should be safe)
$LFS changelog_clear $MDT0 "${CL_USERS[$SINGLEMDS]%% *}" 1000000000
rc=$?
if [ $rc -eq 0 ]; then
error "Successfully cleared to invalid CL index"
elif [ $rc -ne 22 ]; then
error "changelog_clear failed with $rc, expected 22 (EINVAL)"
fi
}
run_test 160e "changelog negative testing (should return errors)"
test_160f() {
remote_mds_nodsh && skip "remote MDS with nodsh" && return
(( $MDS1_VERSION >= $(version_code 2.10.56) )) ||
skip "Need MDS version at least 2.10.56"
local mdts=$(mdts_nodes)
# Create a user
changelog_register || error "first changelog_register failed"
changelog_register || error "second changelog_register failed"
local cl_users
declare -A cl_user1
declare -A cl_user2
local user_rec1
local user_rec2
local i
# generate some changelog records to accumulate on each MDT
# use all_char because created files should be evenly distributed
test_mkdir -c $MDSCOUNT -H all_char $DIR/$tdir ||
error "test_mkdir $tdir failed"
log "$(date +%s): creating first dirs"
for ((idx = 0; idx < MDSCOUNT * 2; idx++)); do
local d=$DIR/$tdir/d$idx.$((idx/MDSCOUNT))
$LFS mkdir -i $((idx%MDSCOUNT)) $d ||
error "create $d on MDT$idx failed"
done
# check changelogs have been generated
local start=$SECONDS
local idle_time=$((MDSCOUNT * 5 + 5))
local nbcl=$(changelog_dump | wc -l)
(( $nbcl != 0 )) || error "no changelogs found"
for param in "changelog_max_idle_time=$idle_time" \
"changelog_gc=1" \
"changelog_min_gc_interval=2" \
"changelog_min_free_cat_entries=3"; do
local MDT0=$(facet_svc $SINGLEMDS)
local var="${param%=*}"
local old=$(do_facet mds1 "$LCTL get_param -n mdd.$MDT0.$var")
stack_trap "do_nodes $mdts $LCTL set_param mdd.*.$var=$old" EXIT
do_nodes $mdts $LCTL set_param mdd.*.$param
done
# force cl_user2 to be idle (1st part), but also cancel the
# cl_user1 records so that it is not evicted later in the test.
local sleep1=$((idle_time / 2))
echo "$(date +%s): sleep1 $sleep1/${idle_time}s"
sleep $sleep1
# simulate changelog catalog almost full
#define OBD_FAIL_CAT_FREE_RECORDS 0x1313
do_nodes $mdts "$LCTL set_param fail_loc=0x1313 fail_val=3"
for ((i = 1; i <= $MDSCOUNT; i++)); do
cl_users=(${CL_USERS[mds$i]})
cl_user1[mds$i]="${cl_users[0]}"
cl_user2[mds$i]="${cl_users[1]}"
[ -n "${cl_user1[mds$i]}" ] ||
error "mds$i: no user registered"
[ -n "${cl_user2[mds$i]}" ] ||
error "mds$i: only ${cl_user2[mds$i]} is registered"
user_rec1=$(changelog_user_rec mds$i ${cl_user1[mds$i]})
[ -n "$user_rec1" ] ||
error "mds$i: User ${cl_user1[mds$i]} not registered"
__changelog_clear mds$i ${cl_user1[mds$i]} +2
user_rec2=$(changelog_user_rec mds$i ${cl_user1[mds$i]})
[ -n "$user_rec2" ] ||
error "mds$i: User ${cl_user1[mds$i]} not registered"
echo "mds$i: verifying user ${cl_user1[mds$i]} clear: " \
"$user_rec1 + 2 == $user_rec2"
[ $((user_rec1 + 2)) == $user_rec2 ] ||
error "mds$i: user ${cl_user1[mds$i]} index expected " \
"$user_rec1 + 2, but is $user_rec2"
user_rec2=$(changelog_user_rec mds$i ${cl_user2[mds$i]})
[ -n "$user_rec2" ] ||
error "mds$i: User ${cl_user2[mds$i]} not registered"
[ $user_rec1 == $user_rec2 ] ||
error "mds$i: user ${cl_user2[mds$i]} index expected " \
"$user_rec1, but is $user_rec2"
done
# force cl_user2 idle (2nd part) to just exceed changelog_max_idle_time
local sleep2=$((idle_time - (SECONDS - start) + 1))
echo "$(date +%s): sleep2 $sleep2/${idle_time}s"
sleep $sleep2
# Generate one more changelog to trigger GC at fail_loc for cl_user2.
# cl_user1 should be OK because it recently processed records.
echo "$(date +%s): creating $((MDSCOUNT * 2)) dirs"
for ((idx = 0; idx < MDSCOUNT * 2; idx++)); do
local d=$DIR/$tdir/d$idx.$((idx/MDSCOUNT+2))
$LFS mkdir -i $((idx%MDSCOUNT)) $d ||
error "create dir $d on MDT$idx failed"
done
# sleep again to allow GC to trigger
local sleep3=4
echo "$(date +%s): sleep3 $sleep3/${idle_time}s"
sleep $sleep3
# ensure gc thread is done
for mds in ${mdts//,/ }; do
wait_update $mds "ps -e -o comm= | grep chlg_gc_thread" "" ||
error "$mds: GC-thread not done"
done
local first_rec
for (( i = 1; i <= MDSCOUNT; i++ )); do
# check cl_user1 still registered
changelog_users mds$i | grep -q "${cl_user1[mds$i]}" ||
error "mds$i: User ${cl_user1[mds$i]} not registered"
# check cl_user2 unregistered
changelog_users mds$i | grep -q "${cl_user2[mds$i]}" &&
error "mds$i: User ${cl_user2[mds$i]} still registered"
# check changelogs are present and starting at $user_rec1 + 1
user_rec1=$(changelog_user_rec mds$i ${cl_user1[mds$i]})
[ -n "$user_rec1" ] ||
error "mds$i: User ${cl_user1[mds$i]} not registered"
first_rec=$($LFS changelog $(facet_svc mds$i) |
awk '{ print $1; exit; }')
echo "mds$i: $(date +%s) verify rec $user_rec1+1 == $first_rec"
[ $((user_rec1 + 1)) == $first_rec ] ||
error "mds$i: rec $first_rec != $user_rec1 + 1"
done
}
run_test 160f "changelog garbage collect (timestamped users)"
test_160g() {
remote_mds_nodsh && skip "remote MDS with nodsh"
(( $MDS1_VERSION >= $(version_code 2.14.55) )) ||
skip "Need MDS version at least 2.14.55"
local mdts=$(mdts_nodes)
# Create a user
changelog_register || error "first changelog_register failed"
changelog_register || error "second changelog_register failed"
local cl_users
declare -A cl_user1
declare -A cl_user2
local user_rec1
local user_rec2
local i
# generate some changelog records to accumulate on each MDT
# use all_char because created files should be evenly distributed
test_mkdir -c $MDSCOUNT -H all_char $DIR/$tdir ||
error "test_mkdir $tdir failed"
for ((idx = 0; idx < MDSCOUNT; idx++)); do
$LFS mkdir -i $idx $DIR/$tdir/d$idx.1 $DIR/$tdir/d$idx.2 ||
error "create $DIR/$tdir/d$idx.1 failed"
done
# check changelogs have been generated
local nbcl=$(changelog_dump | wc -l)
(( $nbcl > 0 )) || error "no changelogs found"
# reduce the max_idle_indexes value to make sure we exceed it
for param in "changelog_max_idle_indexes=2" \
"changelog_gc=1" \
"changelog_min_gc_interval=2"; do
local MDT0=$(facet_svc $SINGLEMDS)
local var="${param%=*}"
local old=$(do_facet mds1 "$LCTL get_param -n mdd.$MDT0.$var")
stack_trap "do_nodes $mdts $LCTL set_param mdd.*.$var=$old" EXIT
do_nodes $mdts $LCTL set_param mdd.*.$param ||
error "unable to set mdd.*.$param"
done
local start=$SECONDS
for ((i=1; i <= $MDSCOUNT; i++)); do
cl_users=(${CL_USERS[mds$i]})
cl_user1[mds$i]="${cl_users[0]}"
cl_user2[mds$i]="${cl_users[1]}"
[ -n "${cl_user1[mds$i]}" ] ||
error "mds$i: user1 is not registered"
[ -n "${cl_user2[mds$i]}" ] ||
error "mds$i: only ${cl_user1[mds$i]} is registered"
user_rec1=$(changelog_user_rec mds$i ${cl_user1[mds$i]})
[ -n "$user_rec1" ] ||
error "mds$i: user1 ${cl_user1[mds$i]} not found"
__changelog_clear mds$i ${cl_user1[mds$i]} +2
user_rec2=$(changelog_user_rec mds$i ${cl_user1[mds$i]})
[ -n "$user_rec2" ] ||
error "mds$i: user1 ${cl_user1[mds$i]} not found (2)"
echo "mds$i: verifying user1 ${cl_user1[mds$i]} clear: " \
"$user_rec1 + 2 == $user_rec2"
[ $((user_rec1 + 2)) == $user_rec2 ] ||
error "mds$i: user1 ${cl_user1[mds$i]} index " \
"expected $user_rec1 + 2, but is $user_rec2"
user_rec2=$(changelog_user_rec mds$i ${cl_user2[mds$i]})
[ -n "$user_rec2" ] ||
error "mds$i: user2 ${cl_user2[mds$i]} not found"
[ $user_rec1 == $user_rec2 ] ||
error "mds$i: user2 ${cl_user2[mds$i]} index " \
"expected $user_rec1, but is $user_rec2"
done
# ensure we are past the previous changelog_min_gc_interval set above
local sleep2=$((start + 2 - SECONDS))
(( sleep2 > 0 )) && echo "sleep $sleep2 for interval" && sleep $sleep2
# Generate one more changelog to trigger GC at fail_loc for cl_user2.
# cl_user1 should be OK because it recently processed records.
for ((idx = 0; idx < MDSCOUNT; idx++)); do
$LFS mkdir -i $idx $DIR/$tdir/d$idx.3 ||
error "create $DIR/$tdir/d$idx.3 failed"
done
# ensure gc thread is done
for mds in ${mdts//,/ }; do
wait_update $mds "ps -e -o comm= | grep chlg_gc_thread" "" 20 ||
error "$mds: GC-thread not done"
done
local first_rec
for (( i = 1; i <= MDSCOUNT; i++ )); do
# check cl_user1 still registered
changelog_users mds$i | grep -q "${cl_user1[mds$i]}" ||
error "mds$i: user1 ${cl_user1[mds$i]} not found (3)"
# check cl_user2 unregistered
changelog_users mds$i | grep -q "${cl_user2[mds$i]}" &&
error "mds$i: user2 ${cl_user2[mds$i]} is registered"
# check changelogs are present and starting at $user_rec1 + 1
user_rec1=$(changelog_user_rec mds$i ${cl_user1[mds$i]})
[ -n "$user_rec1" ] ||
error "mds$i: user1 ${cl_user1[mds$i]} not found (4)"
first_rec=$($LFS changelog $(facet_svc mds$i) |
awk '{ print $1; exit; }')
echo "mds$i: $(date +%s) verify rec $user_rec1+1 == $first_rec"
[ $((user_rec1 + 1)) == $first_rec ] ||
error "mds$i: rec $first_rec != $user_rec1 + 1"
done
}
run_test 160g "changelog garbage collect on idle records"
test_160h() {
remote_mds_nodsh && skip "remote MDS with nodsh" && return
(( $MDS1_VERSION >= $(version_code 2.10.56) )) ||
skip "Need MDS version at least 2.10.56"
local mdts=$(mdts_nodes)
# Create a user
changelog_register || error "first changelog_register failed"
changelog_register || error "second changelog_register failed"
local cl_users
declare -A cl_user1
declare -A cl_user2
local user_rec1
local user_rec2
local i
# generate some changelog records to accumulate on each MDT
# use all_char because created files should be evenly distributed
test_mkdir -c $MDSCOUNT -H all_char $DIR/$tdir ||
error "test_mkdir $tdir failed"
for ((idx = 0; idx < MDSCOUNT; idx++)); do
$LFS mkdir -i $idx $DIR/$tdir/d$idx.1 $DIR/$tdir/d$idx.2 ||
error "create $DIR/$tdir/d$idx.1 failed"
done
# check changelogs have been generated
local nbcl=$(changelog_dump | wc -l)
[[ $nbcl -eq 0 ]] && error "no changelogs found"
for param in "changelog_max_idle_time=10" \
"changelog_gc=1" \
"changelog_min_gc_interval=2"; do
local MDT0=$(facet_svc $SINGLEMDS)
local var="${param%=*}"
local old=$(do_facet mds1 "$LCTL get_param -n mdd.$MDT0.$var")
stack_trap "do_nodes $mdts $LCTL set_param mdd.*.$var=$old" EXIT
do_nodes $mdts $LCTL set_param mdd.*.$param
done
# force cl_user2 to be idle (1st part)
sleep 9
for ((i=1; i <= $MDSCOUNT; i++)); do
cl_users=(${CL_USERS[mds$i]})
cl_user1[mds$i]="${cl_users[0]}"
cl_user2[mds$i]="${cl_users[1]}"
[ -n "${cl_user1[mds$i]}" ] ||
error "mds$i: no user registered"
[ -n "${cl_user2[mds$i]}" ] ||
error "mds$i: only ${cl_user2[mds$i]} is registered"
user_rec1=$(changelog_user_rec mds$i ${cl_user1[mds$i]})
[ -n "$user_rec1" ] ||
error "mds$i: User ${cl_user1[mds$i]} not registered"
__changelog_clear mds$i ${cl_user1[mds$i]} +2
user_rec2=$(changelog_user_rec mds$i ${cl_user1[mds$i]})
[ -n "$user_rec2" ] ||
error "mds$i: User ${cl_user1[mds$i]} not registered"
echo "mds$i: verifying user ${cl_user1[mds$i]} clear: " \
"$user_rec1 + 2 == $user_rec2"
[ $((user_rec1 + 2)) == $user_rec2 ] ||
error "mds$i: user ${cl_user1[mds$i]} index expected " \
"$user_rec1 + 2, but is $user_rec2"
user_rec2=$(changelog_user_rec mds$i ${cl_user2[mds$i]})
[ -n "$user_rec2" ] ||
error "mds$i: User ${cl_user2[mds$i]} not registered"
[ $user_rec1 == $user_rec2 ] ||
error "mds$i: user ${cl_user2[mds$i]} index expected " \
"$user_rec1, but is $user_rec2"
done
# force cl_user2 to be idle (2nd part) and to reach
# changelog_max_idle_time
sleep 2
# force each GC-thread start and block then
# one per MDT/MDD, set fail_val accordingly
#define OBD_FAIL_FORCE_GC_THREAD 0x1316
do_nodes $mdts $LCTL set_param fail_loc=0x1316
# generate more changelogs to trigger fail_loc
createmany -m $DIR/$tdir/${tfile}bis $((MDSCOUNT * 2)) ||
error "create $DIR/$tdir/${tfile}bis failed"
# stop MDT to stop GC-thread, should be done in back-ground as it will
# block waiting for the thread to be released and exit
declare -A stop_pids
for ((i=1; i <= $MDSCOUNT; i++)); do
stop mds$i &
stop_pids[mds$i]=$!
done
for mds in ${mdts//,/ }; do
local facet
local nb=0
local facets=$(facets_up_on_host $mds)
for facet in ${facets//,/ }; do
if [[ $facet == mds* ]]; then
nb=$((nb + 1))
fi
done
# ensure each MDS's gc threads are still present and all in "R"
# state (OBD_FAIL_FORCE_GC_THREAD effect!)
[[ $(do_node $mds pgrep chlg_gc_thread | wc -l) -eq $nb ]] ||
error "$mds: expected $nb GC-thread"
wait_update $mds \
"ps -C chlg_gc_thread -o state --no-headers | uniq" \
"R" 20 ||
error "$mds: GC-thread not found in R-state"
# check umounts of each MDT on MDS have reached kthread_stop()
[[ $(do_node $mds pgrep umount | wc -l) -eq $nb ]] ||
error "$mds: expected $nb umount"
wait_update $mds \
"ps -C umount -o state --no-headers | uniq" "D" 20 ||
error "$mds: umount not found in D-state"
done
# release all GC-threads
do_nodes $mdts $LCTL set_param fail_loc=0
# wait for MDT stop to complete
for ((i=1; i <= $MDSCOUNT; i++)); do
wait ${stop_pids[mds$i]} || error "mds$i: stop failed"
done
# XXX
# may try to check if any orphan changelog records are present
# via ldiskfs/zfs and llog_reader...
# re-start/mount MDTs
for ((i=1; i <= $MDSCOUNT; i++)); do
start mds$i $(mdsdevname $i) $MDS_MOUNT_OPTS ||
error "Fail to start mds$i"
done
local first_rec
for ((i=1; i <= $MDSCOUNT; i++)); do
# check cl_user1 still registered
changelog_users mds$i | grep -q "${cl_user1[mds$i]}" ||
error "mds$i: User ${cl_user1[mds$i]} not registered"
# check cl_user2 unregistered
changelog_users mds$i | grep -q "${cl_user2[mds$i]}" &&
error "mds$i: User ${cl_user2[mds$i]} still registered"
# check changelogs are present and starting at $user_rec1 + 1
user_rec1=$(changelog_user_rec mds$i ${cl_user1[mds$i]})
[ -n "$user_rec1" ] ||
error "mds$i: User ${cl_user1[mds$i]} not registered"
first_rec=$($LFS changelog $(facet_svc mds$i) |
awk '{ print $1; exit; }')
echo "mds$i: verifying first index $user_rec1 + 1 == $first_rec"
[ $((user_rec1 + 1)) == $first_rec ] ||
error "mds$i: first index should be $user_rec1 + 1, " \
"but is $first_rec"
done
}
run_test 160h "changelog gc thread stop upon umount, orphan records delete " \
"during mount"
test_160i() {
local mdts=$(mdts_nodes)
changelog_register || error "first changelog_register failed"
# generate some changelog records to accumulate on each MDT
# use all_char because created files should be evenly distributed
test_mkdir -c $MDSCOUNT -H all_char $DIR/$tdir ||
error "test_mkdir $tdir failed"
for ((i = 0; i < MDSCOUNT; i++)); do
$LFS mkdir -i $i $DIR/$tdir/d$i.1 $DIR/$tdir/d$i.2 ||
error "create $DIR/$tdir/d$i.1 failed"
done
# check changelogs have been generated
local nbcl=$(changelog_dump | wc -l)
[[ $nbcl -eq 0 ]] && error "no changelogs found"
# simulate race between register and unregister
# XXX as fail_loc is set per-MDS, with DNE configs the race
# simulation will only occur for one MDT per MDS and for the
# others the normal race scenario will take place
#define CFS_FAIL_CHLOG_USER_REG_UNREG_RACE 0x1315
do_nodes $mdts $LCTL set_param fail_loc=0x10001315
do_nodes $mdts $LCTL set_param fail_val=1
# unregister 1st user
changelog_deregister &
local pid1=$!
# wait some time for deregister work to reach race rdv
sleep 2
# register 2nd user
changelog_register || error "2nd user register failed"
wait $pid1 || error "1st user deregister failed"
local i
local last_rec
declare -A LAST_REC
for ((i=1; i <= $MDSCOUNT; i++)); do
if changelog_users mds$i | grep "^cl"; then
# make sure new records are added with one user present
LAST_REC[mds$i]=$(changelog_users $SINGLEMDS |
awk '/^current.index:/ { print $NF }')
else
error "mds$i has no user registered"
fi
done
# generate more changelog records to accumulate on each MDT
createmany -m $DIR/$tdir/${tfile}bis $((MDSCOUNT * 2)) ||
error "create $DIR/$tdir/${tfile}bis failed"
for ((i=1; i <= $MDSCOUNT; i++)); do
last_rec=$(changelog_users $SINGLEMDS |
awk '/^current.index:/ { print $NF }')
echo "verify changelogs are on: $last_rec != ${LAST_REC[mds$i]}"
[ $last_rec != ${LAST_REC[mds$i]} ] ||
error "changelogs are off on mds$i"
done
}
run_test 160i "changelog user register/unregister race"
test_160j() {
remote_mds_nodsh && skip "remote MDS with nodsh"
[[ $MDS1_VERSION -lt $(version_code 2.12.56) ]] &&
skip "Need MDS version at least 2.12.56"
mount_client $MOUNT2 || error "mount_client on $MOUNT2 failed"
stack_trap "umount $MOUNT2" EXIT
changelog_register || error "first changelog_register failed"
stack_trap "changelog_deregister" EXIT
# generate some changelog
# use all_char because created files should be evenly distributed
test_mkdir -c $MDSCOUNT -H all_char $DIR/$tdir ||
error "mkdir $tdir failed"
for ((i = 0; i < MDSCOUNT; i++)); do
$LFS mkdir -i $i $DIR/$tdir/d$i.1 $DIR/$tdir/d$i.2 ||
error "create $DIR/$tdir/d$i.1 failed"
done
# open the changelog device
exec 3>/dev/changelog-$FSNAME-MDT0000
stack_trap "exec 3>&-" EXIT
exec 4</dev/changelog-$FSNAME-MDT0000
stack_trap "exec 4<&-" EXIT
# umount the first lustre mount
umount $MOUNT
stack_trap "mount_client $MOUNT" EXIT
# read changelog, which may or may not fail, but should not crash
cat <&4 >/dev/null
# clear changelog
local cl_user="${CL_USERS[$SINGLEMDS]%% *}"
changelog_users $SINGLEMDS | grep -q $cl_user ||
error "User $cl_user not found in changelog_users"
printf 'clear:'$cl_user':0' >&3
}
run_test 160j "client can be umounted while its chanangelog is being used"
test_160k() {
[ $PARALLEL == "yes" ] && skip "skip parallel run"
remote_mds_nodsh && skip "remote MDS with nodsh"
mkdir -p $DIR/$tdir/1/1
changelog_register || error "changelog_register failed"
local cl_user="${CL_USERS[$SINGLEMDS]%% *}"
changelog_users $SINGLEMDS | grep -q $cl_user ||
error "User '$cl_user' not found in changelog_users"
#define OBD_FAIL_MDS_CHANGELOG_REORDER 0x15d
do_facet mds1 $LCTL set_param fail_loc=0x8000015d fail_val=3
rmdir $DIR/$tdir/1/1 & sleep 1
mkdir $DIR/$tdir/2
touch $DIR/$tdir/2/2
rm -rf $DIR/$tdir/2
wait
sleep 4
changelog_dump | grep rmdir || error "rmdir not recorded"
}
run_test 160k "Verify that changelog records are not lost"
# Verifies that a file passed as a parameter has recently had an operation
# performed on it that has generated an MTIME changelog which contains the
# correct parent FID. As files might reside on a different MDT from the
# parent directory in DNE configurations, the FIDs are translated to paths
# before being compared, which should be identical
compare_mtime_changelog() {
local file="${1}"
local mdtidx
local mtime
local cl_fid
local pdir
local dir
mdtidx=$($LFS getstripe --mdt-index $file)
mdtidx=$(printf "%04x" $mdtidx)
# Obtain the parent FID from the MTIME changelog
mtime=$($LFS changelog $FSNAME-MDT$mdtidx | tail -n 1 | grep MTIME)
[ -z "$mtime" ] && error "MTIME changelog not recorded"
cl_fid=$(sed -e 's/.* p=//' -e 's/ .*//' <<<$mtime)
[ -z "$cl_fid" ] && error "parent FID not present"
# Verify that the path for the parent FID is the same as the path for
# the test directory
pdir=$($LFS fid2path $MOUNT "$cl_fid")
dir=$(dirname $1)
[[ "${pdir%/}" == "$dir" ]] ||
error "MTIME changelog parent FID is wrong, expected $dir, got $pdir"
}
test_160l() {
[ $PARALLEL == "yes" ] && skip "skip parallel run"
remote_mds_nodsh && skip "remote MDS with nodsh"
[[ $MDS1_VERSION -ge $(version_code 2.13.55) ]] ||
skip "Need MDS version at least 2.13.55"
local cl_user
changelog_register || error "changelog_register failed"
cl_user="${CL_USERS[$SINGLEMDS]%% *}"
changelog_users $SINGLEMDS | grep -q $cl_user ||
error "User '$cl_user' not found in changelog_users"
# Clear some types so that MTIME changelogs are generated
changelog_chmask "-CREAT"
changelog_chmask "-CLOSE"
test_mkdir $DIR/$tdir || error "failed to mkdir $DIR/$tdir"
# Test CL_MTIME during setattr
touch $DIR/$tdir/$tfile
compare_mtime_changelog $DIR/$tdir/$tfile
# Test CL_MTIME during close
$MULTIOP $DIR/$tdir/${tfile}_2 O_2w4096c || error "multiop failed"
compare_mtime_changelog $DIR/$tdir/${tfile}_2
}
run_test 160l "Verify that MTIME changelog records contain the parent FID"
test_160m() {
remote_mds_nodsh && skip "remote MDS with nodsh" && return
[[ $MDS1_VERSION -ge $(version_code 2.14.51) ]] ||
skip "Need MDS version at least 2.14.51"
local cl_users
local cl_user1
local cl_user2
local pid1
# Create a user
changelog_register || error "first changelog_register failed"
changelog_register || error "second changelog_register failed"
cl_users=(${CL_USERS[mds1]})
cl_user1="${cl_users[0]}"
cl_user2="${cl_users[1]}"
# generate some changelog records to accumulate on MDT0
test_mkdir -p -i0 -c1 $DIR/$tdir || error "test_mkdir $tdir failed"
createmany -m $DIR/$tdir/$tfile 50 ||
error "create $DIR/$tdir/$tfile failed"
unlinkmany $DIR/$tdir/$tfile 50 || error "unlinkmany failed"
rm -f $DIR/$tdir
# check changelogs have been generated
local nbcl=$(changelog_dump | wc -l)
[[ $nbcl -eq 0 ]] && error "no changelogs found"
#define OBD_FAIL_MDS_CHANGELOG_RACE 0x15f
do_facet mds1 $LCTL set_param fail_loc=0x8000015f fail_val=0
__changelog_clear mds1 $cl_user1 +10
__changelog_clear mds1 $cl_user2 0 &
pid1=$!
sleep 2
__changelog_clear mds1 $cl_user1 0 ||
error "fail to cancel record for $cl_user1"
wait $pid1
[[ $? -eq 0 ]] || error "fail to cancel record for $cl_user2"
}
run_test 160m "Changelog clear race"
test_160n() {
remote_mds_nodsh && skip "remote MDS with nodsh" && return
[[ $MDS1_VERSION -ge $(version_code 2.14.51) ]] ||
skip "Need MDS version at least 2.14.51"
local cl_users
local cl_user1
local pid1
local first_rec
local last_rec=0
# Create a user
changelog_register || error "first changelog_register failed"
cl_users=(${CL_USERS[mds1]})
cl_user1="${cl_users[0]}"
# generate some changelog records to accumulate on MDT0
test_mkdir -i0 -c1 $DIR/$tdir || error "test_mkdir $tdir failed"
first_rec=$(changelog_users $SINGLEMDS |
awk '/^current.index:/ { print $NF }')
while (( last_rec < (( first_rec + 65000)) )); do
createmany -m $DIR/$tdir/$tfile 10000 ||
error "create $DIR/$tdir/$tfile failed"
for i in $(seq 0 10000); do
mrename $DIR/$tdir/$tfile$i $DIR/$tdir/$tfile-new$i \
> /dev/null
done
unlinkmany $DIR/$tdir/$tfile-new 10000 ||
error "unlinkmany failed unlink"
last_rec=$(changelog_users $SINGLEMDS |
awk '/^current.index:/ { print $NF }')
echo last record $last_rec
(( last_rec == 0 )) && error "no changelog found"
done
#define OBD_FAIL_MDS_CHANGELOG_DEL 0x16c
do_facet mds1 $LCTL set_param fail_loc=0x8000016c fail_val=0
__changelog_clear mds1 $cl_user1 0 &
pid1=$!
sleep 2
__changelog_clear mds1 $cl_user1 0 ||
error "fail to cancel record for $cl_user1 (forground)"
wait $pid1
[[ $? -eq 0 ]] || error "fail to cancel record for $cl_user1 (background)"
}
run_test 160n "Changelog destroy race"
test_160o() {
local mdt="$(facet_svc $SINGLEMDS)"
[[ $PARALLEL != "yes" ]] || skip "skip parallel run"
remote_mds_nodsh && skip "remote MDS with nodsh"
[ $MDS1_VERSION -ge $(version_code 2.14.52) ] ||
skip "Need MDS version at least 2.14.52"
changelog_register --user test_160o -m unlnk+close+open ||
error "changelog_register failed"
do_facet $SINGLEMDS $LCTL --device $mdt \
changelog_register -u "Tt3_-#" &&
error "bad symbols in name should fail"
do_facet $SINGLEMDS $LCTL --device $mdt \
changelog_register -u test_160o &&
error "the same name registration should fail"
do_facet $SINGLEMDS $LCTL --device $mdt \
changelog_register -u test_160toolongname &&
error "too long name registration should fail"
changelog_chmask "MARK+HSM"
lctl get_param mdd.*.changelog*mask
local cl_user="${CL_USERS[$SINGLEMDS]%% *}"
changelog_users $SINGLEMDS | grep -q $cl_user ||
error "User $cl_user not found in changelog_users"
#verify username
echo $cl_user | grep -q test_160o ||
error "User $cl_user has no specific name 'test160o'"
# change something
changelog_clear 0 || error "changelog_clear failed"
# generate some changelog records to accumulate on MDT0
test_mkdir -p -i0 -c1 $DIR/$tdir || error "test_mkdir $tdir failed"
touch $DIR/$tdir/$tfile # open 1
OPENS=$(changelog_dump | grep -c "OPEN")
[[ $OPENS -eq 1 ]] || error "OPEN changelog mask count $OPENS != 1"
# must be no MKDIR it wasn't set as user mask
MKDIR=$(changelog_dump | grep -c "MKDIR")
[[ $MKDIR -eq 0 ]] || error "MKDIR changelog mask found $MKDIR > 0"
oldmask=$(do_facet $SINGLEMDS $LCTL get_param \
mdd.$mdt.changelog_current_mask -n)
# register maskless user
changelog_register || error "changelog_register failed"
# effective mask should be not changed because it is not minimal
mask=$(do_facet $SINGLEMDS $LCTL get_param \
mdd.$mdt.changelog_current_mask -n)
[[ $mask == $oldmask ]] || error "mask was changed: $mask vs $oldmask"
# set server mask to minimal value
changelog_chmask "MARK"
# check effective mask again, should be treated as DEFMASK now
mask=$(do_facet $SINGLEMDS $LCTL get_param \
mdd.$mdt.changelog_current_mask -n)
[[ $mask == *"HLINK"* ]] || error "mask is not DEFMASK as expected"
if (( $MDS1_VERSION >= $(version_code 2.15.52) )) ; then
# set server mask back to some value
changelog_chmask "CLOSE,UNLNK"
# check effective mask again, should not remain as DEFMASK
mask=$(do_facet $SINGLEMDS $LCTL get_param \
mdd.$mdt.changelog_current_mask -n)
[[ $mask != *"HLINK"* ]] || error "mask is still DEFMASK"
fi
do_facet $SINGLEMDS $LCTL --device $mdt \
changelog_deregister -u test_160o ||
error "cannot deregister by name"
}
run_test 160o "changelog user name and mask"
test_160p() {
remote_mds_nodsh && skip "remote MDS with nodsh" && return
[[ $MDS1_VERSION -ge $(version_code 2.14.51) ]] ||
skip "Need MDS version at least 2.14.51"
[[ "$mds1_FSTYPE" == "ldiskfs" ]] || skip "ldiskfs only test"
local cl_users
local cl_user1
local entry_count
# Create a user
changelog_register || error "first changelog_register failed"
cl_users=(${CL_USERS[mds1]})
cl_user1="${cl_users[0]}"
test_mkdir -p -i0 -c1 $DIR/$tdir || error "test_mkdir $tdir failed"
createmany -m $DIR/$tdir/$tfile 50 ||
error "create $DIR/$tdir/$tfile failed"
unlinkmany $DIR/$tdir/$tfile 50 || error "unlinkmany failed"
rm -rf $DIR/$tdir
# check changelogs have been generated
entry_count=$(changelog_dump | wc -l)
((entry_count != 0)) || error "no changelog entries found"
# remove changelog_users and check that orphan entries are removed
stop mds1
local dev=$(mdsdevname 1)
do_facet mds1 "$DEBUGFS -w -R 'rm changelog_users' $dev"
start mds1 $dev $MDS_MOUNT_OPTS || error "cannot start mds1"
entry_count=$(changelog_dump | wc -l)
((entry_count == 0)) ||
error "found $entry_count changelog entries, expected none"
}
run_test 160p "Changelog orphan cleanup with no users"
test_160q() {
local mdt="$(facet_svc $SINGLEMDS)"
local clu
[[ $PARALLEL != "yes" ]] || skip "skip parallel run"
remote_mds_nodsh && skip "remote MDS with nodsh"
[ $MDS1_VERSION -ge $(version_code 2.14.54) ] ||
skip "Need MDS version at least 2.14.54"
# set server mask to minimal value like server init does
changelog_chmask "MARK"
clu=$(do_facet $SINGLEMDS $LCTL --device $mdt changelog_register -n) ||
error "changelog_register failed"
# check effective mask again, should be treated as DEFMASK now
mask=$(do_facet $SINGLEMDS $LCTL get_param \
mdd.$mdt.changelog_current_mask -n)
do_facet $SINGLEMDS $LCTL --device $mdt changelog_deregister $clu ||
error "changelog_deregister failed"
[[ $mask == *"HLINK"* ]] || error "mask is not DEFMASK as expected"
}
run_test 160q "changelog effective mask is DEFMASK if not set"
test_160s() {
remote_mds_nodsh && skip "remote MDS with nodsh"
(( $MDS1_VERSION >= $(version_code 2.14.55) )) ||
skip "Need MDS version at least 2.14.55"
local mdts=$(mdts_nodes)
#define OBD_FAIL_TIME_IN_CHLOG_USER 0x1314
do_nodes $mdts $LCTL set_param fail_loc=0x1314 \
fail_val=$((24 * 3600 * 10))
# Create a user which is 10 days old
changelog_register || error "first changelog_register failed"
local cl_users
declare -A cl_user1
local i
# generate some changelog records to accumulate on each MDT
# use all_char because created files should be evenly distributed
test_mkdir -c $MDSCOUNT -H all_char $DIR/$tdir ||
error "test_mkdir $tdir failed"
for ((idx = 0; idx < MDSCOUNT; idx++)); do
$LFS mkdir -i $idx $DIR/$tdir/d$idx.1 $DIR/$tdir/d$idx.2 ||
error "create $DIR/$tdir/d$idx.1 on MDT$idx failed"
done
# check changelogs have been generated
local nbcl=$(changelog_dump | wc -l)
(( nbcl > 0 )) || error "no changelogs found"
# reduce the max_idle_indexes value to make sure we exceed it
local gc_interval=2
for param in "changelog_max_idle_indexes=2097446912" \
"changelog_max_idle_time=2592000" \
"changelog_gc=1" \
"changelog_min_gc_interval=$gc_interval"; do
local MDT0=$(facet_svc $SINGLEMDS)
local var="${param%=*}"
local old=$(do_facet mds1 "$LCTL get_param -n mdd.$MDT0.$var")
stack_trap "do_nodes $mdts $LCTL set_param mdd.*.$var=$old" EXIT
do_nodes $mdts $LCTL set_param mdd.*.$param ||
error "unable to set mdd.*.$param"
done
local start=$SECONDS
for ((i=1; i <= $MDSCOUNT; i++)); do
cl_users=(${CL_USERS[mds$i]})
cl_user1[mds$i]="${cl_users[0]}"
[[ -n "${cl_user1[mds$i]}" ]] ||
error "mds$i: no user registered"
done
#define OBD_FAIL_MDS_CHANGELOG_IDX_PUMP 0x16d
do_nodes $mdts $LCTL set_param fail_loc=0x16d fail_val=500000000
# ensure we are past the previous changelog_min_gc_interval set above
local sleep2=$((start + gc_interval + 1 - SECONDS))
(( sleep2 > 0 )) && echo "sleep $sleep2 for interval" && sleep $sleep2
# Generate one more changelog to trigger GC
for ((i = 0; i < MDSCOUNT; i++)); do
$LFS mkdir -i $i $DIR/$tdir/d$i.3 $DIR/$tdir/d$i.4 ||
error "create $DIR/$tdir/d$i.3 failed"
done
# ensure gc thread is done
for mds in ${mdts//,/ }; do
wait_update $mds "pgrep chlg_gc_thread" "" 20 ||
error "$mds: GC-thread not done"
done
do_nodes $mdts $LCTL set_param fail_loc=0
for ((i = 1; i <= MDSCOUNT; i++)); do
# check cl_user1 is purged
changelog_users mds$i | grep -q "${cl_user1[mds$i]}" &&
error "mds$i: User ${cl_user1[mds$i]} is registered"
done
return 0
}
run_test 160s "changelog garbage collect on idle records * time"
test_160t() {
remote_mds_nodsh && skip "remote MDS with nodsh"
(( $MDS1_VERSION >= $(version_code 2.15.50) )) ||
skip "Need MDS version at least 2.15.50"
local MDT0=$(facet_svc $SINGLEMDS)
local cl_users
local cl_user1
local cl_user2
local start
changelog_register --user user1 -m all ||
error "user1 failed to register"
mkdir_on_mdt0 $DIR/$tdir
# create default overstripe to maximize changelog size
$LFS setstripe -C 8 $DIR/$tdir || error "setstripe failed"
createmany -o $DIR/$tdir/u1_ 2000 || error "createmany for user1 failed"
stack_trap "unlinkmany $DIR/$tdir/u1_ 2000"
llog_size1=$(do_facet mds1 $LCTL get_param -n mdd.$MDT0.changelog_size)
# user2 consumes less records so less space
changelog_register --user user2 || error "user2 failed to register"
createmany -o $DIR/$tdir/u2_ 500 || error "createmany for user2 failed"
llog_size2=$(do_facet mds1 $LCTL get_param -n mdd.$MDT0.changelog_size)
# check changelogs have been generated
local nbcl=$(changelog_dump | wc -l)
(( nbcl > 0 )) || error "no changelogs found"
# reduce the changelog_min_gc_interval to force check
for param in "changelog_gc=1" "changelog_min_gc_interval=2"; do
local var="${param%=*}"
local old=$(do_facet mds1 "$LCTL get_param -n mdd.$MDT0.$var")
stack_trap "do_facet mds1 $LCTL set_param mdd.$MDT0.$var=$old"
do_facet mds1 $LCTL set_param mdd.$MDT0.$param ||
error "unable to set mdd.*.$param"
done
start=$SECONDS
cl_users=(${CL_USERS[mds1]})
cl_user1="${cl_users[0]}"
cl_user2="${cl_users[1]}"
[[ -n $cl_user1 ]] ||
error "mds1: user #1 isn't registered"
[[ -n $cl_user2 ]] ||
error "mds1: user #2 isn't registered"
# ensure we are past the previous changelog_min_gc_interval set above
local sleep2=$((start + 2 - SECONDS))
(( sleep2 > 0 )) && echo "sleep $sleep2 for interval" && sleep $sleep2
#define OBD_FAIL_MDS_CHANGELOG_ENOSPC 0x018c
do_facet mds1 $LCTL set_param fail_loc=0x018c \
fail_val=$(((llog_size1 + llog_size2) / 2))
# Generate more changelog to trigger GC
createmany -o $DIR/$tdir/u3_ 4 ||
error "create failed for more files"
# ensure gc thread is done
wait_update_facet mds1 "pgrep chlg_gc_thread" "" 20 ||
error "mds1: GC-thread not done"
do_facet mds1 $LCTL set_param fail_loc=0
# check cl_user1 is purged
changelog_users mds1 | grep -q "$cl_user1" &&
error "User $cl_user1 is registered"
# check cl_user2 is not purged
changelog_users mds1 | grep -q "$cl_user2" ||
error "User $cl_user2 is not registered"
}
run_test 160t "changelog garbage collect on lack of space"
test_160u() { # LU-17400
[ $PARALLEL == "yes" ] && skip "skip parallel run"
remote_mds_nodsh && skip "remote MDS with nodsh"
[ $MDS1_VERSION -ge $(version_code 2.2.0) ] ||
skip "Need MDS version at least 2.2.0"
cd $DIR || error "cd $DIR failed"
# ensure changelog has a clean view if tests are run multiple times
[ -d rename ] && rm -rf rename
changelog_register || error "changelog_register failed"
local cl_user="${CL_USERS[$SINGLEMDS]%% *}"
changelog_users $SINGLEMDS | grep -q $cl_user ||
error "User '$cl_user' not found in changelog_users"
local longname1=$(str_repeat a 255)
echo "creating simple directory tree"
mkdir -p rename/a || error "create of simple directory tree failed"
echo "creating rename/hw file"
echo "hello world" > rename/hw || error "create of rename/hw failed"
echo "creating very long named file"
touch rename/$longname1 || error "create of 'rename/$longname1' failed"
echo "move rename/hw to rename/a/a.hw"
mv rename/hw rename/a/a.hw || error "mv failed"
RENME=($(changelog_dump | grep "RENME"))
#declare -p RENME # for debugging captured value with indexes
[[ "${RENME[11]}" == "a.hw" && "${RENME[14]}" == "hw" ]] ||
error "changelog rename record type name/sname error"
}
run_test 160u "changelog rename record type name and sname strings are correct"
test_160v() {
[ $PARALLEL == "yes" ] && skip "skip parallel run"
(( CLIENTCOUNT >= 2 )) || skip "need >= 2 clients"
[[ $MDS1_VERSION -ge $(version_code 2.17.50) ]] ||
skip "Need MDS version at least 2.17.50"
local target_mtime
local mtime_after
local saved_date
test_mkdir $DIR/$tdir || error "failed to mkdir $DIR/$tdir"
touch $DIR/$tdir/$tfile
# Advance CLIENT2 clock so its touch sets ctime ahead
saved_date=$(do_node $CLIENT2 date +%s)
do_node $CLIENT2 date -s @$((saved_date + 5))
stack_trap "do_node $CLIENT2 date -s @\$(date +%s)" EXIT
do_node $CLIENT2 touch $MOUNT/$tdir/$tfile
do_node $CLIENT2 date -s @$(date +%s)
target_mtime=$(date -d "2020-01-01" +%s)
# Local client sets mtime; ctime < file's ctime from CLIENT2
touch -m -d @$target_mtime $DIR/$tdir/$tfile
cancel_lru_locks mdc
mtime_after=$(stat -c %Y $DIR/$tdir/$tfile)
[[ "$mtime_after" == "$target_mtime" ]] ||
error "mtime not updated: expected $target_mtime, got $mtime_after"
}
run_test 160v "setattr preserves mtime update despite inter-client clock skew"
test_160w() {
[[ $PARALLEL == "yes" ]] && skip "skip parallel run"
remote_mds_nodsh && skip "remote MDS with nodsh"
(( $MDS1_VERSION >= $(version_code 2.17.50) )) ||
skip "Need MDS >= 2.17.50 for changelog filtering"
local mdt="$(facet_svc $SINGLEMDS)"
local u1="u1"_$testnum
local u2="u2"_$testnum
local u1_id u2_id
local count
local logs
mkdir_on_mdt0 $DIR/$tdir || error "mkdir $tdir failed"
# Register two users with different masks
changelog_register --user $u1 -m creat+unlnk ||
error "changelog_register u1 failed"
u1_id="${CL_USERS[$SINGLEMDS]%% *}"
changelog_register --user $u2 -m mkdir+rmdir ||
error "changelog_register u2 failed"
u2_id="$(awk '{print $NF}' <<<"${CL_USERS[$SINGLEMDS]}")"
changelog_users $SINGLEMDS
changelog_clear $u1_id 0 || error "changelog_clear $u1 failed"
changelog_clear $u2_id 0 || error "changelog_clear $u2 failed"
# CREAT
touch $DIR/$tdir/f1 || error "touch f1 failed"
touch $DIR/$tdir/f2 || error "touch f2 failed"
# MKDIR
mkdir $DIR/$tdir/d1 || error "mkdir d1 failed"
mkdir $DIR/$tdir/d2 || error "mkdir d2 failed"
# UNLINK
rm $DIR/$tdir/f1 || error "rm f1 failed"
# RMDIR
rmdir $DIR/$tdir/d1 || error "rmdir d1 failed"
echo "lfs changelog --user (with different user formats)"
logs=$($LFS changelog --user $u1 $mdt | head -10)
[[ -n "$logs" ]] || error "$u1 should see changelog records"
logs=$($LFS changelog --user $u2_id $mdt 0 | head -5)
[[ -n "$logs" ]] || error "Startrec should work"
echo "lfs changelog --user (should show all registered records)"
# Show all records u1 registered
logs=$($LFS changelog --user $u1 $mdt | tee >(cat >&2))
count=$(grep -c -o -E "CREAT|UNLNK|MKDIR|RMDIR" <<< $logs)
(( $count == 3 )) ||
error "$u1 should see 3 total records, but got $count"
# No MKDIR|RMDIR for u1
count=$(echo $logs | grep -c -E "RMDIR|MKDIR")
(( $count == 0 )) ||
error "$u1 should see 0 MKDIR|RMDIR record, but got $count"
echo "lfs changelog --user --mask (with registered mask)"
logs=$($LFS changelog --user $u2_id --mask "mkdir+rmdir" $mdt |
tee >(cat >&2))
# MKDIR
count=$(grep -c -o "MKDIR" <<< $logs)
(( $count == 2 )) ||
error "$u2_id should see 2 MKDIR records, but got $count"
# RMDIR
count=$(grep -c -o "RMDIR" <<< $logs)
(( $count == 1 )) ||
error "$u2_id should see 1 RMDIR record, bug got $count"
# No CREAT|UNLNK for u2
count=$(grep -c -E "CREAT|UNLNK" <<< $logs)
(( $count == 0 )) ||
error "$u2_id should see 0 CREAT|UNLNK record, but got $count"
echo "lfs changelog --user --mask (including unregistered mask)"
logs=$($LFS changelog --user $u1 --mask creat+rmdir $mdt |
tee >(cat >&2))
# CREAT
count=$(grep -c -o "CREAT" <<< $logs)
(( $count == 2 )) ||
error "$u1 should see 2 CREAT records, but got $count"
# No RMDIR for u1
count=$(grep -c -o "RMDIR" <<< $logs)
(( $count == 0 )) ||
error "$u1 should see 0 RMDIR record, but got $count"
rm -rf $DIR/$tdir
changelog_deregister $u1 || error "cannot deregister $u1"
changelog_deregister $u2 || error "cannot deregister $u2"
}
run_test 160w "lfs changelog --user --mask"
test_161a() {
[ $PARALLEL == "yes" ] && skip "skip parallel run"
test_mkdir -c1 $DIR/$tdir
cp /etc/hosts $DIR/$tdir/$tfile
test_mkdir -c1 $DIR/$tdir/foo1
test_mkdir -c1 $DIR/$tdir/foo2
ln $DIR/$tdir/$tfile $DIR/$tdir/foo1/sofia
ln $DIR/$tdir/$tfile $DIR/$tdir/foo2/zachary
ln $DIR/$tdir/$tfile $DIR/$tdir/foo1/luna
ln $DIR/$tdir/$tfile $DIR/$tdir/foo2/thor
local FID=$($LFS path2fid $DIR/$tdir/$tfile | tr -d '[]')
if [ "$($LFS fid2path $DIR $FID | wc -l)" != "5" ]; then
$LFS fid2path $DIR $FID
error "bad link ea"
fi
# middle
rm $DIR/$tdir/foo2/zachary
# last
rm $DIR/$tdir/foo2/thor
# first
rm $DIR/$tdir/$tfile
# rename
mv $DIR/$tdir/foo1/sofia $DIR/$tdir/foo2/maggie
[ "$($LFS fid2path $FSNAME --link 1 $FID)" != "$tdir/foo2/maggie" ] &&
{ $LFS fid2path $DIR $FID; error "bad link rename"; }
rm $DIR/$tdir/foo2/maggie
# overflow the EA
local longname=$tfile.avg_len_is_thirty_two_
stack_trap "unlinkmany $DIR/$tdir/foo2/$longname 1000 || \
error_noexit 'failed to unlink many hardlinks'" EXIT
createmany -l$DIR/$tdir/foo1/luna $DIR/$tdir/foo2/$longname 1000 ||
error "failed to hardlink many files"
links=$($LFS fid2path $DIR $FID | wc -l)
echo -n "${links}/1000 links in link EA"
[[ $links -gt 60 ]] || error "expected at least 60 links in link EA"
}
run_test 161a "link ea sanity"
test_161b() {
[ $PARALLEL == "yes" ] && skip "skip parallel run"
[ $MDSCOUNT -lt 2 ] && skip_env "skipping remote directory test"
local MDTIDX=1
local remote_dir=$DIR/$tdir/remote_dir
mkdir -p $DIR/$tdir
$LFS mkdir -i $MDTIDX $remote_dir ||
error "create remote directory failed"
cp /etc/hosts $remote_dir/$tfile
mkdir -p $remote_dir/foo1
mkdir -p $remote_dir/foo2
ln $remote_dir/$tfile $remote_dir/foo1/sofia
ln $remote_dir/$tfile $remote_dir/foo2/zachary
ln $remote_dir/$tfile $remote_dir/foo1/luna
ln $remote_dir/$tfile $remote_dir/foo2/thor
local FID=$($LFS path2fid $remote_dir/$tfile | tr -d '[' |
tr -d ']')
if [ "$($LFS fid2path $DIR $FID | wc -l)" != "5" ]; then
$LFS fid2path $DIR $FID
error "bad link ea"
fi
# middle
rm $remote_dir/foo2/zachary
# last
rm $remote_dir/foo2/thor
# first
rm $remote_dir/$tfile
# rename
mv $remote_dir/foo1/sofia $remote_dir/foo2/maggie
local link_path=$($LFS fid2path $FSNAME --link 1 $FID)
if [ "$DIR/$link_path" != "$remote_dir/foo2/maggie" ]; then
$LFS fid2path $DIR $FID
error "bad link rename"
fi
rm $remote_dir/foo2/maggie
# overflow the EA
local longname=filename_avg_len_is_thirty_two_
createmany -l$remote_dir/foo1/luna $remote_dir/foo2/$longname 1000 ||
error "failed to hardlink many files"
links=$($LFS fid2path $DIR $FID | wc -l)
echo -n "${links}/1000 links in link EA"
[[ ${links} -gt 60 ]] ||
error "expected at least 60 links in link EA"
unlinkmany $remote_dir/foo2/$longname 1000 ||
error "failed to unlink many hardlinks"
}
run_test 161b "link ea sanity under remote directory"
test_161c() {
remote_mds_nodsh && skip "remote MDS with nodsh"
[ $PARALLEL == "yes" ] && skip "skip parallel run"
[[ $MDS1_VERSION -lt $(version_code 2.1.5) ]] &&
skip "Need MDS version at least 2.1.5"
# define CLF_RENAME_LAST 0x0001
# rename overwrite a target having nlink = 1 (changelog flag 0x1)
changelog_register || error "changelog_register failed"
rm -rf $DIR/$tdir
test_mkdir -i $((MDSCOUNT - 1)) $DIR/$tdir
touch $DIR/$tdir/foo_161c
touch $DIR/$tdir/bar_161c
mv -f $DIR/$tdir/foo_161c $DIR/$tdir/bar_161c
changelog_dump | grep RENME | tail -n 5
local flags=$(changelog_dump | grep "RENME.*bar_161c" | cut -f5 -d' ')
changelog_clear 0 || error "changelog_clear failed"
if [ x$flags != "x0x1" ]; then
error "flag $flags is not 0x1"
fi
echo "rename overwrite target with nlink = 1, changelog flags=$flags"
# rename overwrite a target having nlink > 1 (changelog flag 0x0)
touch $DIR/$tdir/foo_161c
touch $DIR/$tdir/bar_161c
ln $DIR/$tdir/bar_161c $DIR/$tdir/foobar_161c
mv -f $DIR/$tdir/foo_161c $DIR/$tdir/bar_161c
changelog_dump | grep RENME | tail -n 5
flags=$(changelog_dump | grep "RENME.*bar_161c" | cut -f5 -d' ')
changelog_clear 0 || error "changelog_clear failed"
if [ x$flags != "x0x0" ]; then
error "flag $flags is not 0x0"
fi
echo "rename overwrite a target having nlink > 1," \
"changelog record has flags of $flags"
# rename doesn't overwrite a target (changelog flag 0x0)
touch $DIR/$tdir/foo_161c
mv -f $DIR/$tdir/foo_161c $DIR/$tdir/foo2_161c
changelog_dump | grep RENME | tail -n 5
flags=$(changelog_dump | grep RENME | tail -1 | cut -f5 -d' ')
changelog_clear 0 || error "changelog_clear failed"
if [ x$flags != "x0x0" ]; then
error "flag $flags is not 0x0"
fi
echo "rename doesn't overwrite a target," \
"changelog record has flags of $flags"
# define CLF_UNLINK_LAST 0x0001
# unlink a file having nlink = 1 (changelog flag 0x1)
rm -f $DIR/$tdir/foo2_161c
changelog_dump | grep UNLNK | tail -n 5
flags=$(changelog_dump | grep UNLNK | tail -1 | cut -f5 -d' ')
changelog_clear 0 || error "changelog_clear failed"
if [ x$flags != "x0x1" ]; then
error "flag $flags is not 0x1"
fi
echo "unlink a file having nlink = 1," \
"changelog record has flags of $flags"
# unlink a file having nlink > 1 (changelog flag 0x0)
ln -f $DIR/$tdir/bar_161c $DIR/$tdir/foobar_161c
rm -f $DIR/$tdir/foobar_161c
changelog_dump | grep UNLNK | tail -n 5
flags=$(changelog_dump | grep UNLNK | tail -1 | cut -f5 -d' ')
changelog_clear 0 || error "changelog_clear failed"
if [ x$flags != "x0x0" ]; then
error "flag $flags is not 0x0"
fi
echo "unlink a file having nlink > 1, changelog record flags '$flags'"
(( MDS1_VERSION >= $(version_code v2_16_53-39) )) ||
{ echo "CLF_UNLINK_LAST is not set with open file (LU-18860)";
return 0; }
# unlink a file open having nlink = 1 (changelog flag 0x1)
local pid
touch $DIR/$tdir/open_161c
tail -f $DIR/$tdir/open_161c & pid=$!
sleep 0.2
unlink $DIR/$tdir/open_161c
kill $pid
changelog_dump | grep UNLNK | tail -n 5
flags=$(changelog_dump | awk '/UNLNK/ {f=$5} END {print f}')
changelog_clear 0 || error "changelog_clear failed"
[[ "$flags" == "0x1" ]] || error "flag $flags is not 0x1"
echo "unlink a open file having nlink = 1, changelog record has flags of $flags"
}
run_test 161c "check CL_RENME[UNLINK] changelog record flags"
test_161d() {
remote_mds_nodsh && skip "remote MDS with nodsh"
local pid
local fid
changelog_register || error "changelog_register failed"
# work in a standalone dir to avoid locking on $DIR/$MOUNT to
# interfer with $MOUNT/.lustre/fid/ access
mkdir $DIR/$tdir
[[ $? -eq 0 ]] || error "mkdir failed"
#define OBD_FAIL_LLITE_CREATE_NODE_PAUSE 0x140c | CFS_FAIL_ONCE
$LCTL set_param fail_loc=0x8000140c
# 5s pause
$LCTL set_param fail_val=5
# create file
echo foofoo > $DIR/$tdir/$tfile &
pid=$!
# wait for create to be delayed
sleep 2
ps -p $pid
[[ $? -eq 0 ]] || error "create should be blocked"
local tempfile="$(mktemp --tmpdir $tfile.XXXXXX)"
stack_trap "rm -f $tempfile"
fid=$(changelog_extract_field "CREAT" "$tfile" "t=")
$MULTIOP $MOUNT/.lustre/fid/$fid orp16384c >$tempfile || error "read failed"
# some delay may occur during ChangeLog publishing and file read just
# above, that could allow file write to happen finally
[[ -s $tempfile ]] && echo "file should be empty"
$LCTL set_param fail_loc=0
wait $pid
[[ $? -eq 0 ]] || error "create failed"
}
run_test 161d "create with concurrent .lustre/fid access"
check_path() {
local expected="$1"
shift
local fid="$2"
local path
path=$($LFS fid2path "$@")
local rc=$?
if [ $rc -ne 0 ]; then
error "path looked up of '$expected' failed: rc=$rc"
elif [ "$path" != "$expected" ]; then
error "path looked up '$path' instead of '$expected'"
else
echo "FID '$fid' resolves to path '$path' as expected"
fi
}
test_162a() { # was test_162
test_mkdir -p -c1 $DIR/$tdir/d2
touch $DIR/$tdir/d2/$tfile
touch $DIR/$tdir/d2/x1
touch $DIR/$tdir/d2/x2
test_mkdir -p -c1 $DIR/$tdir/d2/a/b/c
test_mkdir -p -c1 $DIR/$tdir/d2/p/q/r
# regular file
local fid=$($LFS path2fid $DIR/$tdir/d2/$tfile | tr -d '[]')
check_path "$tdir/d2/$tfile" $FSNAME "$fid" --link 0
# softlink
ln -s $DIR/$tdir/d2/$tfile $DIR/$tdir/d2/p/q/r/slink
fid=$($LFS path2fid $DIR/$tdir/d2/p/q/r/slink | tr -d '[]')
check_path "$tdir/d2/p/q/r/slink" $FSNAME "$fid" --link 0
# softlink to wrong file
ln -s /this/is/garbage $DIR/$tdir/d2/p/q/r/slink.wrong
fid=$($LFS path2fid $DIR/$tdir/d2/p/q/r/slink.wrong | tr -d '[]')
check_path "$tdir/d2/p/q/r/slink.wrong" $FSNAME "$fid" --link 0
# hardlink
ln $DIR/$tdir/d2/$tfile $DIR/$tdir/d2/p/q/r/hlink
mv $DIR/$tdir/d2/$tfile $DIR/$tdir/d2/a/b/c/new_file
fid=$($LFS path2fid $DIR/$tdir/d2/a/b/c/new_file | tr -d '[]')
# fid2path dir/fsname should both work
check_path "$tdir/d2/a/b/c/new_file" $FSNAME "$fid" --link 1
check_path "$DIR/$tdir/d2/p/q/r/hlink" $DIR "$fid" --link 0
# hardlink count: check that there are 2 links
local nlinks=$($LFS fid2path $DIR "$fid" | wc -l)
[ $nlinks -eq 2 ] || error "expect 2 links, found $nlinks"
# hardlink indexing: remove the first link
rm $DIR/$tdir/d2/p/q/r/hlink
check_path "$tdir/d2/a/b/c/new_file" $FSNAME $fid --link 0
}
run_test 162a "path lookup sanity"
test_162b() {
[ $PARALLEL == "yes" ] && skip "skip parallel run"
[ $MDSCOUNT -lt 2 ] && skip_env "needs >= 2 MDTs"
mkdir $DIR/$tdir
$LFS setdirstripe -i0 -c$MDSCOUNT -H all_char $DIR/$tdir/striped_dir ||
error "create striped dir failed"
local FID=$($LFS getdirstripe $DIR/$tdir/striped_dir |
tail -n 1 | awk '{print $2}')
stat $MOUNT/.lustre/fid/$FID && error "sub_stripe can be accessed"
touch $DIR/$tdir/striped_dir/f{0..4} || error "touch f0..4 failed"
mkdir $DIR/$tdir/striped_dir/d{0..4} || error "mkdir d0..4 failed"
# regular file
for ((i=0;i<5;i++)); do
FID=$($LFS path2fid $DIR/$tdir/striped_dir/f$i | tr -d '[]') ||
error "get fid for f$i failed"
check_path "$tdir/striped_dir/f$i" $FSNAME $FID --link 0
FID=$($LFS path2fid $DIR/$tdir/striped_dir/d$i | tr -d '[]') ||
error "get fid for d$i failed"
check_path "$tdir/striped_dir/d$i" $FSNAME $FID --link 0
done
return 0
}
run_test 162b "striped directory path lookup sanity"
# LU-4239: Verify fid2path works with paths 100 or more directories deep
test_162c() {
[[ $MDS1_VERSION -lt $(version_code 2.7.51) ]] &&
skip "Need MDS version at least 2.7.51"
local lpath=$tdir.local
local rpath=$tdir.remote
test_mkdir $DIR/$lpath
test_mkdir $DIR/$rpath
stack_trap "rm -rf $DIR/$lpath $DIR/$rpath"
for ((i = 0; i <= 101; i++)); do
lpath="$lpath/$i"
mkdir $DIR/$lpath
FID=$($LFS path2fid $DIR/$lpath | tr -d '[]') ||
error "get fid for local directory $DIR/$lpath failed"
check_path "$DIR/$lpath" $MOUNT $FID --link 0
rpath="$rpath/$i"
test_mkdir $DIR/$rpath
FID=$($LFS path2fid $DIR/$rpath | tr -d '[]') ||
error "get fid for remote directory $DIR/$rpath failed"
check_path "$DIR/$rpath" $MOUNT $FID --link 0
done
return 0
}
run_test 162c "fid2path works with paths 100 or more directories deep"
oalr_event_count() {
local event="${1}"
local trace="${2}"
awk -v name="${FSNAME}-OST0000" \
-v event="${event}" \
'$1 == "TRACE" && $2 == event && $3 == name' \
"${trace}" |
wc -l
}
oalr_expect_event_count() {
local event="${1}"
local trace="${2}"
local expect="${3}"
local count
count=$(oalr_event_count "${event}" "${trace}")
if ((count == expect)); then
return 0
fi
error_noexit "${event} event count was '${count}', expected ${expect}"
cat "${trace}" >&2
exit 1
}
cleanup_165() {
do_facet ost1 killall --quiet -KILL ofd_access_log_reader || true
stop ost1
start ost1 "$(ostdevname 1)" $OST_MOUNT_OPTS
}
setup_165() {
sync # Flush previous IOs so we can count log entries.
do_facet ost1 $LCTL set_param "obdfilter.${FSNAME}-OST0000.access_log_size=4096"
stack_trap cleanup_165 EXIT
}
test_165a() {
local trace="/tmp/${tfile}.trace"
local rc
local count
(( $OST1_VERSION >= $(version_code 2.13.54) )) ||
skip "OFD access log unsupported"
setup_165
do_facet ost1 ofd_access_log_reader --debug=- --trace=- > "${trace}" &
sleep 5
do_facet ost1 ofd_access_log_reader --list
stop ost1
do_facet ost1 killall -TERM ofd_access_log_reader
wait
rc=$?
if ((rc != 0)); then
error "ofd_access_log_reader exited with rc = '${rc}'"
fi
# Parse trace file for discovery events:
oalr_expect_event_count alr_log_add "${trace}" 1
oalr_expect_event_count alr_log_eof "${trace}" 1
oalr_expect_event_count alr_log_free "${trace}" 1
}
run_test 165a "ofd access log discovery"
test_165b() {
local trace="/tmp/${tfile}.trace"
local file="${DIR}/${tfile}"
local pfid1
local pfid2
local -a entry
local rc
local count
local size
local flags
(( $OST1_VERSION >= $(version_code 2.13.54) )) ||
skip "OFD access log unsupported"
setup_165
do_facet ost1 ofd_access_log_reader --debug=- --trace=- > "${trace}" &
sleep 5
do_facet ost1 ofd_access_log_reader --list
lfs setstripe -c 1 -i 0 "${file}"
$MULTIOP "${file}" oO_CREAT:O_DIRECT:O_WRONLY:w1048576c ||
error "cannot create '${file}'"
sleep 5
do_facet ost1 killall -TERM ofd_access_log_reader
wait
rc=$?
if ((rc != 0)); then
error "ofd_access_log_reader exited with rc = '${rc}'"
fi
oalr_expect_event_count alr_log_entry "${trace}" 1
pfid1=$($LFS path2fid "${file}")
# 1 2 3 4 5 6 7 8 9 10
# TRACE alr_log_entry OST PFID BEGIN END TIME SIZE COUNT FLAGS
entry=( - $(awk -v pfid="${pfid}" '$1 == "TRACE" && $2 == "alr_log_entry"' "${trace}" ) )
echo "entry = '${entry[*]}'" >&2
pfid2=${entry[4]}
if [[ "${pfid1}" != "${pfid2}" ]]; then
error "entry '${entry[*]}' has invalid PFID '${pfid2}', expected ${pfid1}"
fi
size=${entry[8]}
if ((size != 1048576)); then
error "entry '${entry[*]}' has invalid io size '${size}', expected 1048576"
fi
flags=${entry[10]}
if [[ "${flags}" != "w" ]]; then
error "entry '${entry[*]}' has invalid io flags '${flags}', expected 'w'"
fi
do_facet ost1 ofd_access_log_reader --debug=- --trace=- > "${trace}" &
sleep 5
$MULTIOP "${file}" oO_CREAT:O_DIRECT:O_RDONLY:r524288c ||
error "cannot read '${file}'"
sleep 5
do_facet ost1 killall -TERM ofd_access_log_reader
wait
rc=$?
if ((rc != 0)); then
error "ofd_access_log_reader exited with rc = '${rc}'"
fi
oalr_expect_event_count alr_log_entry "${trace}" 1
entry=( - $(awk -v pfid="${pfid}" '$1 == "TRACE" && $2 == "alr_log_entry"' "${trace}" ) )
echo "entry = '${entry[*]}'" >&2
pfid2=${entry[4]}
if [[ "${pfid1}" != "${pfid2}" ]]; then
error "entry '${entry[*]}' has invalid PFID '${pfid2}', expected ${pfid1}"
fi
size=${entry[8]}
if ((size != 524288)); then
error "entry '${entry[*]}' has invalid io size '${size}', 524288"
fi
flags=${entry[10]}
if [[ "${flags}" != "r" ]]; then
error "entry '${entry[*]}' has invalid io flags '${flags}', expected 'r'"
fi
}
run_test 165b "ofd access log entries are produced and consumed"
test_165c() {
local trace="/tmp/${tfile}.trace"
local file="${DIR}/${tdir}/${tfile}"
(( $OST1_VERSION >= $(version_code 2.13.54) )) ||
skip "OFD access log unsupported"
test_mkdir "${DIR}/${tdir}"
setup_165
do_facet ost1 ofd_access_log_reader --debug=- --trace=- > "${trace}" &
sleep 5
lfs setstripe -c 1 -i 0 "${DIR}/${tdir}"
# 4096 / 64 = 64. Create twice as many entries.
for ((i = 0; i < 128; i++)); do
$MULTIOP "${file}-${i}" oO_CREAT:O_WRONLY:w512c ||
error "cannot create file"
done
sync
do_facet ost1 killall -TERM ofd_access_log_reader
wait
rc=$?
if ((rc != 0)); then
error "ofd_access_log_reader exited with rc = '${rc}'"
fi
unlinkmany "${file}-%d" 128
}
run_test 165c "full ofd access logs do not block IOs"
oal_get_read_count() {
local stats="$1"
# STATS lustre-OST0001 alr_read_count 1
do_facet ost1 cat "${stats}" |
awk '$1 == "STATS" && $3 == "alr_read_count" { count = $4; }
END { print count; }'
}
oal_expect_read_count() {
local stats="$1"
local count
local expect="$2"
# Ask ofd_access_log_reader to write stats.
do_facet ost1 killall -USR1 ofd_access_log_reader
# Allow some time for things to happen.
sleep 1
count=$(oal_get_read_count "${stats}")
if ((count == expect)); then
return 0
fi
error_noexit "bad read count, got ${count}, expected ${expect}"
do_facet ost1 cat "${stats}" >&2
exit 1
}
test_165d() {
local stats="/tmp/${tfile}.stats"
local file="${DIR}/${tdir}/${tfile}"
local param="obdfilter.${FSNAME}-OST0000.access_log_mask"
(( $OST1_VERSION >= $(version_code 2.13.54) )) ||
skip "OFD access log unsupported"
test_mkdir "${DIR}/${tdir}"
setup_165
do_facet ost1 ofd_access_log_reader --stats="${stats}" &
sleep 5
lfs setstripe -c 1 -i 0 "${file}"
do_facet ost1 lctl set_param "${param}=rw"
$MULTIOP "${file}" oO_CREAT:O_DIRECT:O_WRONLY:w1048576c ||
error "cannot create '${file}'"
oal_expect_read_count "${stats}" 1
$MULTIOP "${file}" oO_CREAT:O_DIRECT:O_RDONLY:r1048576c ||
error "cannot read '${file}'"
oal_expect_read_count "${stats}" 2
do_facet ost1 lctl set_param "${param}=r"
$MULTIOP "${file}" oO_CREAT:O_DIRECT:O_WRONLY:w1048576c ||
error "cannot create '${file}'"
oal_expect_read_count "${stats}" 2
$MULTIOP "${file}" oO_CREAT:O_DIRECT:O_RDONLY:r1048576c ||
error "cannot read '${file}'"
oal_expect_read_count "${stats}" 3
do_facet ost1 lctl set_param "${param}=w"
$MULTIOP "${file}" oO_CREAT:O_DIRECT:O_WRONLY:w1048576c ||
error "cannot create '${file}'"
oal_expect_read_count "${stats}" 4
$MULTIOP "${file}" oO_CREAT:O_DIRECT:O_RDONLY:r1048576c ||
error "cannot read '${file}'"
oal_expect_read_count "${stats}" 4
do_facet ost1 lctl set_param "${param}=0"
$MULTIOP "${file}" oO_CREAT:O_DIRECT:O_WRONLY:w1048576c ||
error "cannot create '${file}'"
oal_expect_read_count "${stats}" 4
$MULTIOP "${file}" oO_CREAT:O_DIRECT:O_RDONLY:r1048576c ||
error "cannot read '${file}'"
oal_expect_read_count "${stats}" 4
do_facet ost1 killall -TERM ofd_access_log_reader
wait
rc=$?
if ((rc != 0)); then
error "ofd_access_log_reader exited with rc = '${rc}'"
fi
}
run_test 165d "ofd_access_log mask works"
test_165e() {
local stats="/tmp/${tfile}.stats"
local file0="${DIR}/${tdir}-0/${tfile}"
local file1="${DIR}/${tdir}-1/${tfile}"
(( $OST1_VERSION >= $(version_code 2.13.54) )) ||
skip "OFD access log unsupported"
[[ $MDSCOUNT -lt 2 ]] && skip_env "needs >= 2 MDTs"
test_mkdir -c 1 -i 0 "${DIR}/${tdir}-0"
test_mkdir -c 1 -i 1 "${DIR}/${tdir}-1"
lfs setstripe -c 1 -i 0 "${file0}"
lfs setstripe -c 1 -i 0 "${file1}"
setup_165
do_facet ost1 ofd_access_log_reader -I 1 --stats="${stats}" &
sleep 5
$MULTIOP "${file0}" oO_CREAT:O_WRONLY:w512c ||
error "cannot create '${file0}'"
sync
oal_expect_read_count "${stats}" 0
$MULTIOP "${file1}" oO_CREAT:O_WRONLY:w512c ||
error "cannot create '${file1}'"
sync
oal_expect_read_count "${stats}" 1
do_facet ost1 killall -TERM ofd_access_log_reader
wait
rc=$?
if ((rc != 0)); then
error "ofd_access_log_reader exited with rc = '${rc}'"
fi
}
run_test 165e "ofd_access_log MDT index filter works"
test_165f() {
local trace="/tmp/${tfile}.trace"
local rc
local count
setup_165
do_facet ost1 timeout 60 ofd_access_log_reader \
--exit-on-close --debug=- --trace=- > "${trace}" &
sleep 5
stop ost1
wait
rc=$?
if ((rc != 0)); then
error_noexit "ofd_access_log_reader exited with rc = '${rc}'"
cat "${trace}"
exit 1
fi
}
run_test 165f "ofd_access_log_reader --exit-on-close works"
test_165g() {
local count
local trace=$TMP/${tfile}.trace
local ofd_keepalive=$(do_facet ost1 ofd_access_log_reader --help |
grep keepalive)
[[ -n "$ofd_keepalive" ]] || skip "ALR keepalive message unsupported"
setup_165
do_facet ost1 timeout 60 ofd_access_log_reader \
--keepalive=3 --batch-interval=4 --debug=- > "${trace}" &
sleep 40
stop ost1
do_facet ost1 killall -TERM ofd_access_log_reader
wait
rc=$?
if ((rc != 0)); then
error "ofd_access_log_reader exited with rc = '${rc}'"
fi
count=$(grep -c "send keepalive" ${trace})
# Normal duration 40 seconds / 4 secs interval = 10 times or so.
# But sometimes the duration may be unpredictably long
(( count >= 7 )) || error "keepalive msg received $count. Expected >= 7"
echo "keepalive msg received $count"
}
run_test 165g "ofd_access_log_reader --keepalive works"
test_169() {
# do directio so as not to populate the page cache
log "creating a 10 Mb file"
$MULTIOP $DIR/$tfile oO_CREAT:O_DIRECT:O_RDWR:w$((10*1048576))c ||
error "multiop failed while creating a file"
log "starting reads"
dd if=$DIR/$tfile of=/dev/null bs=4096 &
log "truncating the file"
$MULTIOP $DIR/$tfile oO_TRUNC:c ||
error "multiop failed while truncating the file"
log "killing dd"
kill %+ || true # reads might have finished
echo "wait until dd is finished"
wait
log "removing the temporary file"
rm -rf $DIR/$tfile || error "tmp file removal failed"
}
run_test 169 "parallel read and truncate should not deadlock"
test_170a() {
[ $PARALLEL == "yes" ] && skip "skip parallel run"
$LCTL clear # bug 18514
$LCTL debug_daemon start $TMP/${tfile}_log_good
touch $DIR/$tfile
$LCTL debug_daemon stop
sed -e "s/^...../a/g" $TMP/${tfile}_log_good > $TMP/${tfile}_log_bad ||
error "sed failed to read log_good"
$LCTL debug_daemon start $TMP/${tfile}_log_good
rm -rf $DIR/$tfile
$LCTL debug_daemon stop
$LCTL df $TMP/${tfile}_log_bad > $TMP/${tfile}_log_bad.out 2>&1 ||
error "lctl df log_bad failed"
local bad_line=$(tail -n 1 $TMP/${tfile}_log_bad.out | awk '{print $9}')
local good_line1=$(tail -n 1 $TMP/${tfile}_log_bad.out | awk '{print $5}')
$LCTL df $TMP/${tfile}_log_good > $TMP/${tfile}_log_good.out 2>&1
local good_line2=$(tail -n 1 $TMP/${tfile}_log_good.out | awk '{print $5}')
[ "$bad_line" ] && [ "$good_line1" ] && [ "$good_line2" ] ||
error "bad_line good_line1 good_line2 are empty"
cat $TMP/${tfile}_log_good >> $TMP/${tfile}_logs_corrupt
cat $TMP/${tfile}_log_bad >> $TMP/${tfile}_logs_corrupt
cat $TMP/${tfile}_log_good >> $TMP/${tfile}_logs_corrupt
$LCTL df $TMP/${tfile}_logs_corrupt > $TMP/${tfile}_log_bad.out 2>&1
local bad_line_new=$(tail -n 1 $TMP/${tfile}_log_bad.out | awk '{print $9}')
local good_line_new=$(tail -n 1 $TMP/${tfile}_log_bad.out | awk '{print $5}')
[ "$bad_line_new" ] && [ "$good_line_new" ] ||
error "bad_line_new good_line_new are empty"
local expected_good=$((good_line1 + good_line2*2))
rm -f $TMP/${tfile}*
# LU-231, short malformed line may not be counted into bad lines
if [ $bad_line -ne $bad_line_new ] &&
[ $bad_line -ne $((bad_line_new - 1)) ]; then
error "expected $bad_line bad lines, but got $bad_line_new"
return 1
fi
if [ $expected_good -ne $good_line_new ]; then
error "expected $expected_good good lines, but got $good_line_new"
return 2
fi
true
}
run_test 170a "test lctl df to handle corrupted log"
test_170b() {
# enable filename encoding on one client, ost and mdt respectively
declare -a old_fl
local param="enable_fname_encoding"
local nodes_list=$HOSTNAME
local log=$TMP/sanity.$testnum
local i=0
(( $MDS1_VERSION >= $(version_code 2.16.52) )) &&
nodes_list="$nodes_list,$(facet_active_host mds1)"
(( $OST1_VERSION >= $(version_code 2.16.52) )) &&
nodes_list="$nodes_list,$(facet_active_host ost1)"
local_mode && nodes_list=$HOSTNAME
old_fl=($(do_nodes $nodes_list $LCTL get_param -n $param))
# Test with filename encoding on
set_params_nodes $nodes_list "$param=on"
for node in ${nodes_list//,/ }; do
stack_trap "set_params_nodes $node "$param=${old_fl[$i]}""
i=$((i+1))
done
do_nodes $nodes_list "$LCTL get_param -n $param" | grep 0 &&
error "Failed to enable filename encoding on all the nodes"
# Start full debug to collect the logs
start_full_debug_logging $nodes_list
do_nodes $nodes_list "$LCTL clear"
# Do some operations on $tfile
touch $DIR/$tfile || error "Failed to touch $tfile"
cp $DIR/$tfile $DIR/$testnum.cp || error "Failed to cp $tfile"
ln -s $DIR/$tfile $DIR/$testnum.ln || error "Failed to ln -s $tfile"
$LFS migrate -c $OSTCOUNT $DIR/$tfile ||
error "Failed to migrate $tfile"
$LFS mirror extend -N $DIR/$tfile ||
error "Failed to extend mirrored $tfile"
mv $DIR/$tfile $DIR/$testnum.mv || error "Failed to mv $tfile"
rm -f $DIR/$tfile $DIR/$testnum.* || error "Failed to rm $tfile"
# Do some operations on $tdir
mkdir -p $DIR/$tdir || error "Failed to mkdir $tdir"
cp -r $DIR/$tdir $DIR/$testnum.cp || error "Failed to cp $tdir"
ln -s $DIR/$tdir $DIR/$testnum.ln || error "Failed to ln -s $tdir"
$LFS migrate -d -m 0 -c $MDSCOUNT $DIR/$tdir ||
error "Failed to migrate $tdir"
mv $DIR/$tdir $DIR/$testnum.mv || error "Failed to mv $tdir"
rm -rf $DIR/$tdir $DIR/$testnum.* || error "Failed to rm $tdir"
# Stop the full debug
do_nodes $nodes_list "$LCTL dk > $log"
stack_trap "do_nodes $nodes_list rm -f $log"
stop_full_debug_logging
# Verify if $tfile and $tdir names are encoded in the logs
do_nodes $nodes_list "grep -E '$tfile|$tdir' $log" &&
error "Still found '$tfile' and '$tdir' in the logs"
return 0
}
run_test 170b "check filename encoding"
test_171() { # bug20592
[ $PARALLEL == "yes" ] && skip "skip parallel run"
#define OBD_FAIL_PTLRPC_DUMP_LOG 0x50e
$LCTL set_param fail_loc=0x50e
$LCTL set_param fail_val=3000
multiop_bg_pause $DIR/$tfile O_s || true
local MULTIPID=$!
kill -USR1 $MULTIPID
# cause log dump
sleep 3
wait $MULTIPID
if dmesg | grep "recursive fault"; then
error "caught a recursive fault"
fi
$LCTL set_param fail_loc=0
true
}
run_test 171 "test libcfs_debug_dumplog_thread stuck in do_exit() ======"
test_172() {
#define OBD_FAIL_OBD_CLEANUP 0x60e
$LCTL set_param fail_loc=0x60e
umount $MOUNT || error "umount $MOUNT failed"
stack_trap "mount_client $MOUNT"
(( $($LCTL dl | grep -E -c " osc | lov | lmv | mdc ") > 0 )) ||
error "no client OBDs are remained"
$LCTL dl | while read devno state type name foo; do
case $type in
lov|osc|lmv|mdc)
$LCTL --device $name cleanup
$LCTL --device $name detach
;;
*)
# skip server devices
;;
esac
done
if (( $($LCTL dl | grep -E -c " osc | lov | lmv | mdc ") > 0 )); then
$LCTL dl | grep -E " osc | lov | lmv | mdc "
error "some client OBDs are still remained"
fi
}
run_test 172 "manual device removal with lctl cleanup/detach ======"
# it would be good to share it with obdfilter-survey/iokit-libecho code
setup_obdecho_osc () {
local rc=0
local ost_nid=$1
local obdfilter_name=$2
echo "Creating new osc for $obdfilter_name on $ost_nid"
# make sure we can find loopback nid
$LCTL add_uuid $ost_nid $ost_nid >/dev/null 2>&1
[ $rc -eq 0 ] && { $LCTL attach osc ${obdfilter_name}_osc \
${obdfilter_name}_osc_UUID || rc=2; }
[ $rc -eq 0 ] && { $LCTL --device ${obdfilter_name}_osc setup \
${obdfilter_name}_UUID $ost_nid || rc=3; }
return $rc
}
cleanup_obdecho_osc () {
local obdfilter_name=$1
$LCTL --device ${obdfilter_name}_osc cleanup >/dev/null
$LCTL --device ${obdfilter_name}_osc detach >/dev/null
return 0
}
obdecho_test() {
local OBD=$1
local node=$2
local pages=${3:-64}
local rc=0
local id
local count=10
local obd_size=$(get_obd_size $node $OBD)
local page_size=$(get_page_size $node)
if [[ -n "$obd_size" ]]; then
local new_count=$((obd_size / (pages * page_size / 1024)))
[[ $new_count -ge $count ]] || count=$new_count
fi
do_facet $node "$LCTL attach echo_client ec ec_uuid" || rc=1
[ $rc -eq 0 ] && { do_facet $node "$LCTL --device ec setup $OBD" ||
rc=2; }
if [ $rc -eq 0 ]; then
id=$(do_facet $node "$LCTL --device ec create 1" | awk '/object id/ {print $6}')
[ ${PIPESTATUS[0]} -eq 0 -a -n "$id" ] || rc=3
fi
echo "New object id is $id"
[ $rc -eq 0 ] && { do_facet $node "$LCTL --device ec getattr $id" ||
rc=4; }
[ $rc -eq 0 ] && { do_facet $node "$LCTL --device ec " \
"test_brw $count w v $pages $id" || rc=4; }
[ $rc -eq 0 ] && { do_facet $node "$LCTL --device ec destroy $id 1" ||
rc=4; }
[ $rc -eq 0 ] || [ $rc -gt 2 ] &&
{ do_facet $node "$LCTL --device ec cleanup" || rc=5; }
[ $rc -eq 0 ] || [ $rc -gt 1 ] &&
{ do_facet $node "$LCTL --device ec detach" || rc=6; }
[ $rc -ne 0 ] && echo "obecho_create_test failed: $rc"
return $rc
}
test_180a() {
skip "obdecho on osc is no longer supported"
}
run_test 180a "test obdecho on osc"
test_180b() {
[ $PARALLEL == "yes" ] && skip "skip parallel run"
remote_ost_nodsh && skip "remote OST with nodsh"
do_rpc_nodes $(facet_active_host ost1) load_module obdecho/obdecho &&
stack_trap "do_facet ost1 rmmod obdecho" EXIT ||
error "failed to load module obdecho"
local target=$(do_facet ost1 $LCTL dl |
awk '/obdfilter/ { print $4; exit; }')
if [ -n "$target" ]; then
obdecho_test $target ost1 || error "obdecho_test failed with $?"
else
do_facet ost1 $LCTL dl
error "there is no obdfilter target on ost1"
fi
}
run_test 180b "test obdecho directly on obdfilter"
test_180c() { # LU-2598
[ $PARALLEL == "yes" ] && skip "skip parallel run"
remote_ost_nodsh && skip "remote OST with nodsh"
[[ $MDS1_VERSION -lt $(version_code 2.4.0) ]] &&
skip "Need MDS version at least 2.4.0"
do_rpc_nodes $(facet_active_host ost1) load_module obdecho/obdecho &&
stack_trap "do_facet ost1 rmmod obdecho" EXIT ||
error "failed to load module obdecho"
local target=$(do_facet ost1 $LCTL dl |
awk '/obdfilter/ { print $4; exit; }')
if [ -n "$target" ]; then
local pages=16384 # 64MB bulk I/O RPC size
obdecho_test "$target" ost1 "$pages" ||
error "obdecho_test with pages=$pages failed with $?"
else
do_facet ost1 $LCTL dl
error "there is no obdfilter target on ost1"
fi
}
run_test 180c "test huge bulk I/O size on obdfilter, don't LASSERT"
test_181() { # bug 22177
test_mkdir $DIR/$tdir
# create enough files to index the directory
createmany -o $DIR/$tdir/foobar 4000
# print attributes for debug purpose
lsattr -d .
# open dir
multiop_bg_pause $DIR/$tdir D_Sc || return 1
MULTIPID=$!
# remove the files & current working dir
unlinkmany $DIR/$tdir/foobar 4000
rmdir $DIR/$tdir
kill -USR1 $MULTIPID
wait $MULTIPID
stat $DIR/$tdir && error "open-unlinked dir was not removed!"
return 0
}
run_test 181 "Test open-unlinked dir ========================"
test_182a() {
local fcount=1000
local tcount=10
mkdir -p $DIR/$tdir || error "creating dir $DIR/$tdir"
$LCTL set_param mdc.*.rpc_stats=clear
for (( i = 0; i < $tcount; i++ )) ; do
mkdir $DIR/$tdir/$i
done
for (( i = 0; i < $tcount; i++ )) ; do
createmany -o $DIR/$tdir/$i/f- $fcount &
done
wait
for (( i = 0; i < $tcount; i++ )) ; do
unlinkmany $DIR/$tdir/$i/f- $fcount &
done
wait
$LCTL get_param mdc.*.rpc_stats
rm -rf $DIR/$tdir
}
run_test 182a "Test parallel modify metadata operations from mdc"
test_182b() {
[ $PARALLEL == "yes" ] && skip "skip parallel run" && return
[ $MDSCOUNT -lt 2 ] && skip "needs >= 2 MDTs" && return
local dcount=1000
local tcount=10
local stime
local etime
local delta
do_facet mds1 $LCTL list_param \
osp.$FSNAME-MDT*-osp-MDT*.rpc_stats ||
skip "MDS lacks parallel RPC handling"
$LFS mkdir -i 0 $DIR/$tdir || error "creating dir $DIR/$tdir"
rpc_count=$(do_facet mds1 $LCTL get_param -n \
osp.$FSNAME-MDT0001-osp-MDT0000.max_mod_rpcs_in_flight)
stime=$(date +%s)
createmany -i 0 -d $DIR/$tdir/t- $tcount
for (( i = 0; i < $tcount; i++ )) ; do
createmany -i 0 -d $DIR/$tdir/t-$i/d- 0 $dcount &
done
wait
etime=$(date +%s)
delta=$((etime - stime))
echo "Time for file creation $delta sec for $rpc_count parallel RPCs"
stime=$(date +%s)
for (( i = 0; i < $tcount; i++ )) ; do
unlinkmany -d $DIR/$tdir/$i/d- $dcount &
done
wait
etime=$(date +%s)
delta=$((etime - stime))
echo "Time for file removal $delta sec for $rpc_count parallel RPCs"
rm -rf $DIR/$tdir
$LFS mkdir -i 0 $DIR/$tdir || error "creating dir $DIR/$tdir"
do_facet mds1 $LCTL set_param osp.$FSNAME-MDT0001-osp-MDT0000.max_mod_rpcs_in_flight=1
stime=$(date +%s)
createmany -i 0 -d $DIR/$tdir/t- $tcount
for (( i = 0; i < $tcount; i++ )) ; do
createmany -i 0 -d $DIR/$tdir/t-$i/d- 0 $dcount &
done
wait
etime=$(date +%s)
delta=$((etime - stime))
echo "Time for file creation $delta sec for 1 RPC sent at a time"
stime=$(date +%s)
for (( i = 0; i < $tcount; i++ )) ; do
unlinkmany -d $DIR/$tdir/t-$i/d- $dcount &
done
wait
etime=$(date +%s)
delta=$((etime - stime))
echo "Time for file removal $delta sec for 1 RPC sent at a time"
do_facet mds1 $LCTL set_param osp.$FSNAME-MDT0001-osp-MDT0000.max_mod_rpcs_in_flight=$rpc_count
}
run_test 182b "Test parallel modify metadata operations from osp"
test_183() { # LU-2275
[ $PARALLEL == "yes" ] && skip "skip parallel run"
remote_mds_nodsh && skip "remote MDS with nodsh"
[[ $MDS1_VERSION -lt $(version_code 2.3.56) ]] &&
skip "Need MDS version at least 2.3.56"
mkdir_on_mdt0 $DIR/$tdir || error "creating dir $DIR/$tdir"
echo aaa > $DIR/$tdir/$tfile
#define OBD_FAIL_MDS_NEGATIVE_POSITIVE 0x148
do_facet $SINGLEMDS $LCTL set_param fail_loc=0x148
ls -l $DIR/$tdir && error "ls succeeded, should have failed"
cat $DIR/$tdir/$tfile && error "cat succeeded, should have failed"
do_facet $SINGLEMDS $LCTL set_param fail_loc=0
# Flush negative dentry cache
touch $DIR/$tdir/$tfile
# We are not checking for any leaked references here, they'll
# become evident next time we do cleanup with module unload.
rm -rf $DIR/$tdir
}
run_test 183 "No crash or request leak in case of strange dispositions ========"
# test suite 184 is for LU-2016, LU-2017
test_184a() {
check_swap_layouts_support
dir0=$DIR/$tdir/$testnum
test_mkdir -p -c1 $dir0
ref1=/etc/passwd
ref2=/etc/group
file1=$dir0/f1
file2=$dir0/f2
$LFS setstripe -c1 $file1
cp $ref1 $file1
$LFS setstripe -c2 $file2
cp $ref2 $file2
gen1=$($LFS getstripe -g $file1)
gen2=$($LFS getstripe -g $file2)
$LFS swap_layouts $file1 $file2 || error "swap of file layout failed"
gen=$($LFS getstripe -g $file1)
[[ $gen1 != $gen ]] ||
error "Layout generation on $file1 does not change"
gen=$($LFS getstripe -g $file2)
[[ $gen2 != $gen ]] ||
error "Layout generation on $file2 does not change"
cmp $ref1 $file2 || error "content compare failed ($ref1 != $file2)"
cmp $ref2 $file1 || error "content compare failed ($ref2 != $file1)"
lfsck_verify_pfid $file1 $file2 || error "PFID are not transferred"
}
run_test 184a "Basic layout swap"
test_184b() {
check_swap_layouts_support
dir0=$DIR/$tdir/$testnum
mkdir -p $dir0 || error "creating dir $dir0"
file1=$dir0/f1
file2=$dir0/f2
file3=$dir0/f3
dir1=$dir0/d1
dir2=$dir0/d2
mkdir $dir1 $dir2
$LFS setstripe -c1 $file1
$LFS setstripe -c2 $file2
$LFS setstripe -c1 $file3
chown $RUNAS_ID $file3
gen1=$($LFS getstripe -g $file1)
gen2=$($LFS getstripe -g $file2)
$LFS swap_layouts $dir1 $dir2 &&
error "swap of directories layouts should fail"
$LFS swap_layouts $dir1 $file1 &&
error "swap of directory and file layouts should fail"
$RUNAS $LFS swap_layouts $file1 $file2 &&
error "swap of file we cannot write should fail"
$LFS swap_layouts $file1 $file3 &&
error "swap of file with different owner should fail"
/bin/true # to clear error code
}
run_test 184b "Forbidden layout swap (will generate errors)"
test_184c() {
local cmpn_arg=$(cmp -n 2>&1 | grep "invalid option")
[ -n "$cmpn_arg" ] && skip_env "cmp does not support -n"
check_swap_layouts_support
check_swap_layout_no_dom $DIR
local dir0=$DIR/$tdir/$testnum
mkdir -p $dir0 || error "creating dir $dir0"
local ref1=$dir0/ref1
local ref2=$dir0/ref2
local file1=$dir0/file1
local file2=$dir0/file2
# create a file large enough for the concurrent test
dd if=/dev/urandom of=$ref1 bs=1M count=$((RANDOM % 50 + 20))
dd if=/dev/urandom of=$ref2 bs=1M count=$((RANDOM % 50 + 20))
echo "ref file size: ref1($(stat -c %s $ref1))," \
"ref2($(stat -c %s $ref2))"
cp $ref2 $file2
dd if=$ref1 of=$file1 bs=16k &
local DD_PID=$!
# Make sure dd starts to copy file, but wait at most 5 seconds
local loops=0
while [ ! -s $file1 -a $((loops++)) -lt 50 ]; do sleep 0.1; done
$LFS swap_layouts $file1 $file2
local rc=$?
wait $DD_PID
[[ $? == 0 ]] || error "concurrent write on $file1 failed"
[[ $rc == 0 ]] || error "swap of $file1 and $file2 failed"
# how many bytes copied before swapping layout
local copied=$(stat -c %s $file2)
local remaining=$(stat -c %s $ref1)
remaining=$((remaining - copied))
echo "Copied $copied bytes before swapping layout..."
cmp -n $copied $file1 $ref2 | grep differ &&
error "Content mismatch [0, $copied) of ref2 and file1"
cmp -n $copied $file2 $ref1 ||
error "Content mismatch [0, $copied) of ref1 and file2"
cmp -i $copied:$copied -n $remaining $file1 $ref1 ||
error "Content mismatch [$copied, EOF) of ref1 and file1"
# clean up
rm -f $ref1 $ref2 $file1 $file2
}
run_test 184c "Concurrent write and layout swap"
test_184d() {
check_swap_layouts_support
check_swap_layout_no_dom $DIR
[ -z "$(which getfattr 2>/dev/null)" ] &&
skip_env "no getfattr command"
local file1=$DIR/$tdir/$tfile-1
local file2=$DIR/$tdir/$tfile-2
local file3=$DIR/$tdir/$tfile-3
local lovea1
local lovea2
mkdir -p $DIR/$tdir
touch $file1 || error "create $file1 failed"
$OPENFILE -f O_CREAT:O_LOV_DELAY_CREATE $file2 ||
error "create $file2 failed"
$OPENFILE -f O_CREAT:O_LOV_DELAY_CREATE $file3 ||
error "create $file3 failed"
lovea1=$(get_layout_param $file1)
$LFS swap_layouts $file2 $file3 ||
error "swap $file2 $file3 layouts failed"
$LFS swap_layouts $file1 $file2 ||
error "swap $file1 $file2 layouts failed"
lovea2=$(get_layout_param $file2)
echo "$lovea1"
echo "$lovea2"
[ "$lovea1" == "$lovea2" ] || error "lovea $lovea1 != $lovea2"
lovea1=$(getfattr -n trusted.lov $file1 | grep ^trusted)
[[ -z "$lovea1" ]] || error "$file1 shouldn't have lovea"
}
run_test 184d "allow stripeless layouts swap"
test_184e() {
[[ $MDS1_VERSION -ge $(version_code 2.6.94) ]] ||
skip "Need MDS version at least 2.6.94"
check_swap_layouts_support
check_swap_layout_no_dom $DIR
[ -z "$(which getfattr 2>/dev/null)" ] &&
skip_env "no getfattr command"
local file1=$DIR/$tdir/$tfile-1
local file2=$DIR/$tdir/$tfile-2
local file3=$DIR/$tdir/$tfile-3
local lovea
mkdir -p $DIR/$tdir
touch $file1 || error "create $file1 failed"
$OPENFILE -f O_CREAT:O_LOV_DELAY_CREATE $file2 ||
error "create $file2 failed"
$OPENFILE -f O_CREAT:O_LOV_DELAY_CREATE $file3 ||
error "create $file3 failed"
$LFS swap_layouts $file1 $file2 ||
error "swap $file1 $file2 layouts failed"
lovea=$(getfattr -n trusted.lov $file1 | grep ^trusted)
[[ -z "$lovea" ]] || error "$file1 shouldn't have lovea"
echo 123 > $file1 || error "Should be able to write into $file1"
$LFS swap_layouts $file1 $file3 ||
error "swap $file1 $file3 layouts failed"
echo 123 > $file1 || error "Should be able to write into $file1"
rm -rf $file1 $file2 $file3
}
run_test 184e "Recreate layout after stripeless layout swaps"
test_184f() {
# Create a file with name longer than sizeof(struct stat) ==
# 144 to see if we can get chars from the file name to appear
# in the returned striping. Note that 'f' == 0x66.
local file=$(for ((i = 0; i < 200; i++)); do echo -n f; done)
mkdir -p $DIR/$tdir
mcreate $DIR/$tdir/$file
if lfs find --stripe-count 0x6666 $DIR/$tdir | grep $file; then
error "IOC_MDC_GETFILEINFO returned garbage striping"
fi
}
run_test 184f "IOC_MDC_GETFILEINFO for files with long names but no striping"
test_185() { # LU-2441
# LU-3553 - no volatile file support in old servers
[[ $MDS1_VERSION -ge $(version_code 2.3.60) ]] ||
skip "Need MDS version at least 2.3.60"
mkdir -p $DIR/$tdir || error "creating dir $DIR/$tdir"
touch $DIR/$tdir/spoo
local mtime1=$(stat -c "%Y" $DIR/$tdir)
local fid=$($MULTIOP $DIR/$tdir VFw4096c) ||
error "cannot create/write a volatile file"
$CHECKSTAT -t file $MOUNT/.lustre/fid/$fid 2>/dev/null &&
error "FID is still valid after close"
multiop_bg_pause $DIR/$tdir Vw4096_c
local multi_pid=$!
local OLD_IFS=$IFS
IFS=":"
local fidv=($fid)
IFS=$OLD_IFS
# assume that the next FID for this client is sequential, since stdout
# is unfortunately eaten by multiop_bg_pause
local n=$((${fidv[1]} + 1))
local next_fid="${fidv[0]}:$(printf "0x%x" $n):${fidv[2]}"
$CHECKSTAT -t file $MOUNT/.lustre/fid/$next_fid ||
error "FID is missing before close"
kill -USR1 $multi_pid
# 1 second delay, so if mtime change we will see it
sleep 1
local mtime2=$(stat -c "%Y" $DIR/$tdir)
[[ $mtime1 == $mtime2 ]] || error "mtime has changed"
}
run_test 185 "Volatile file support"
function create_check_volatile() {
local idx=$1
local tgt
$MULTIOP $MOUNT/.lustre/fid V${idx}Fw4096_c >&/tmp/${tfile}.fid &
local PID=$!
sleep 1
local FID=$(cat /tmp/${tfile}.fid)
[ "$FID" == "" ] && error "can't get FID for volatile"
$CHECKSTAT -t file $MOUNT/.lustre/fid/$FID || error "can't stat $FID"
tgt=$($LFS getstripe -m $MOUNT/.lustre/fid/$FID)
[ "$tgt" != "$idx" ] && error "wrong MDS $tgt, expected $idx"
kill -USR1 $PID
wait
sleep 1
cancel_lru_locks mdc # flush opencache
$CHECKSTAT -t file $MOUNT/.lustre/fid/$FID && error "can stat $FID"
return 0
}
test_185a(){
# LU-12516 - volatile creation via .lustre
(( $MDS1_VERSION >= $(version_code 2.12.55) )) ||
skip "Need MDS version at least 2.12.55"
[ -n "$FILESET" ] && skip "SKIP due to FILESET set"
create_check_volatile 0
[ $MDSCOUNT -lt 2 ] && return 0
# DNE case
create_check_volatile 1
return 0
}
run_test 185a "Volatile file creation in .lustre/fid/"
test_187a() {
remote_mds_nodsh && skip "remote MDS with nodsh"
[ $MDS1_VERSION -lt $(version_code 2.3.0) ] &&
skip "Need MDS version at least 2.3.0"
local dir0=$DIR/$tdir/$testnum
mkdir -p $dir0 || error "creating dir $dir0"
local file=$dir0/file1
dd if=/dev/urandom of=$file count=10 bs=1M conv=fsync
stack_trap "rm -f $file"
local dv1=$($LFS data_version $file)
dd if=/dev/urandom of=$file seek=10 count=1 bs=1M conv=fsync
local dv2=$($LFS data_version $file)
[[ $dv1 != $dv2 ]] ||
error "data version did not change on write $dv1 == $dv2"
}
run_test 187a "Test data version change"
test_187b() {
remote_mds_nodsh && skip "remote MDS with nodsh"
[ $MDS1_VERSION -lt $(version_code 2.3.0) ] &&
skip "Need MDS version at least 2.3.0"
local dir0=$DIR/$tdir/$testnum
mkdir -p $dir0 || error "creating dir $dir0"
declare -a DV=$($MULTIOP $dir0 Vw1000xYw1000xY | cut -f3 -d" ")
[[ ${DV[0]} != ${DV[1]} ]] ||
error "data version did not change on write"\
" ${DV[0]} == ${DV[1]}"
# clean up
rm -f $file1
}
run_test 187b "Test data version change on volatile file"
test_190a() {
local test_dir="$DIR/$tdir"
local test_file="$test_dir/$tfile"
local prj=($(get_test_project))
echo "proj - ${prj[0]} ${prj[1]}"
[[ -n "$LIBLUSTREAPI_PROJID_FILE" ]] ||
error "\$LIBLUSTREAPI_PROJID_FILE unset"
[[ -f "$LIBLUSTREAPI_PROJID_FILE" ]] ||
error "$LIBLUSTREAPI_PROJID_FILE does not exist"
mkdir -p $test_dir || error "Failed to create test directory"
touch $test_file || error "Failed to create test file"
stack_trap "rm -rf $test_dir"
# Set project using project name
$LFS project -p ${prj[0]} $test_file || {
cat $LIBLUSTREAPI_PROJID_FILE
error "Failed to set project using name ${prj[0]}"
}
local set_id=$($LFS project -d $test_file | awk '{print $1}')
[[ "$set_id" == "${prj[1]}" ]] ||
error "Project ID mismatch: expected ${prj[1]}, got $set_id"
}
run_test 190a "check lfs project -p works with project name"
test_190b() {
local test_dir="$DIR/$tdir"
local MAX_FILES=100
local fc
[[ $SLOW == "yes" ]] && MAX_FILES=50000
local prj=($(get_test_project))
echo "proj - ${prj[0]} ${prj[1]}"
mkdir -p $test_dir || error "Failed to create test directory"
stack_trap "rm -rf $test_dir"
local start=$SECONDS
local end=$((start + 300))
for ((fc = 1; fc <= MAX_FILES; fc++)); do
echo "projid$fc:$fc:$tfile.$fc" >> $LIBLUSTREAPI_PROJID_FILE
touch $test_dir/$tfile.$fc
$LFS project -p projid$fc $test_dir/$tfile.$fc || {
head -n 10 $LIBLUSTREAPI_PROJID_FILE
error "Failed to set project 'projid$fc' on $tfile.$fc"
}
(( SECONDS < end )) || break
(( fc % (MAX_FILES/10) == 0 )) &&
echo "created $fc/$MAX_FILES files/projid in $((SECONDS-start))/$((end-start))s"
done
(( fc == MAX_FILES + 1 )) && fc=$MAX_FILES
echo "created $fc/$MAX_FILES files/projid in $((SECONDS-start))/$((end-start))s"
for i in {1..5}; do
touch $test_dir/noproject_$i
done
# lfs find ! --projid files
local count_usr=$($LFS find ! --projid 0 $test_dir | wc -l)
(( count_usr == fc )) ||
error "lfs find ! --projid 0 found $count_usr files != $fc"
# lfs find --projid files
count_usr=$($LFS find --projid 0 $test_dir | wc -l)
(( count_usr == 6 )) ||
error "lfs find --projid 0 found $count_usr files != 6"
echo "lookup projid$fc timing:"
time $LFS project -p projid$fc $MOUNT
# lfs find --projid ID
start=$SECONDS
[[ "$SLOW" == "yes" ]] && end=600 || end=120
(( end += start ))
for ((i = 1; i < fc; i += fc / 5)); do
echo "start scan for projid$i/$fc at $((end - SECONDS))"
[[ -n $($LFS find --projid "projid$i" $test_dir) ]] ||
error "lfs find --projid projid$i did not find result"
((SECONDS < end)) || echo "timeout $((end - SECONDS))s" && break
done
}
run_test 190b "check lfs find --project works with project name"
test_190c() {
local test_dir="$DIR/$tdir"
local test_file="$test_dir/$tfile"
local TSTID=${TSTID:-"$(id -u $TSTUSR)"}
mkdir -p $test_dir || error "Failed to create test directory"
touch $test_file || error "Failed to create test file"
# Set project using user name
$LFS project -p u:$TSTUSR $test_file ||
error "Failed to set project using username $TSTUSR"
local set_id=$($LFS project -d $test_file | awk '{print $1}')
(( set_id == TSTID )) ||
error "Project ID mismatch: expect $TSTID, got $set_id"
$LFS project -p 0 $test_file ||
error "Failed to reset project"
local set_id=$($LFS project -d $test_file | awk '{print $1}')
(( set_id == 0 )) ||
error "Project ID mismatch: expected 0, got $set_id"
# Set project using uid
$LFS project -p u:$TSTID $test_file ||
error "Failed to set project using username $TSTUSR"
set_id=$($LFS project -d $test_file | awk '{print $1}')
(( set_id == TSTID )) ||
error "Project ID mismatch: expect $TSTID, got $set_id"
}
run_test 190c "check lfs project -p works with u:USERNAME"
test_200() {
[ $PARALLEL == "yes" ] && skip "skip parallel run"
remote_mgs_nodsh && skip "remote MGS with nodsh"
local POOL=${POOL:-cea1}
local POOL_ROOT=${POOL_ROOT:-$DIR/$tdir}
local POOL_DIR_NAME=${POOL_DIR_NAME:-dir_tst}
# Pool OST targets
local first_ost=0
local last_ost=$(($OSTCOUNT - 1))
local ost_step=2
local ost_list=$(seq $first_ost $ost_step $last_ost)
local ost_range="$first_ost $last_ost $ost_step"
local test_path=$POOL_ROOT/$POOL_DIR_NAME
local file_dir=$POOL_ROOT/file_tst
local subdir=$test_path/subdir
local rc=0
while : ; do
# former test_200a test_200b
pool_add $POOL || { rc=$? ; break; }
pool_add_targets $POOL $ost_range || { rc=$? ; break; }
# former test_200c test_200d
mkdir -p $test_path
pool_set_dir $POOL $test_path || { rc=$? ; break; }
pool_check_dir $POOL $test_path || { rc=$? ; break; }
mkdir -p $subdir
pool_check_dir $POOL $subdir || { rc=$? ; break; }
pool_dir_rel_path $POOL $POOL_DIR_NAME $POOL_ROOT \
|| { rc=$? ; break; }
# former test_200e test_200f
local files=$((OSTCOUNT*3))
pool_alloc_files $POOL $test_path $files "$ost_list" \
|| { rc=$? ; break; }
pool_create_files $POOL $file_dir $files "$ost_list" \
|| { rc=$? ; break; }
# former test_200g test_200h
pool_lfs_df $POOL || { rc=$? ; break; }
pool_file_rel_path $POOL $test_path || { rc=$? ; break; }
# former test_201a test_201b test_201c
pool_remove_first_target $POOL || { rc=$? ; break; }
local f=$test_path/$tfile
pool_remove_all_targets $POOL $f || { rc=$? ; break; }
pool_remove $POOL $f || { rc=$? ; break; }
break
done
destroy_test_pools
return $rc
}
run_test 200 "OST pools"
# usage: default_attr <count | size | offset>
default_attr() {
$LCTL get_param -n lov.$FSNAME-clilov-\*.stripe${1}
}
# usage: check_default_stripe_attr
check_default_stripe_attr() {
ACTUAL=$($LFS getstripe $* $DIR/$tdir)
case $1 in
--stripe-count|-c)
[ -n "$2" ] && EXPECTED=0 || EXPECTED=$(default_attr count);;
--stripe-size|-S)
[ -n "$2" ] && EXPECTED=0 || EXPECTED=$(default_attr size);;
--stripe-index|-i)
EXPECTED=-1;;
*)
error "unknown getstripe attr '$1'"
esac
[ $ACTUAL == $EXPECTED ] ||
error "$DIR/$tdir has $1 '$ACTUAL', not '$EXPECTED'"
}
test_204a() {
test_mkdir $DIR/$tdir
$LFS setstripe --stripe-count 0 --stripe-size 0 --stripe-index -1 $DIR/$tdir
check_default_stripe_attr --stripe-count
check_default_stripe_attr --stripe-size
check_default_stripe_attr --stripe-index
}
run_test 204a "Print default stripe attributes"
test_204b() {
test_mkdir $DIR/$tdir
$LFS setstripe --stripe-count 1 $DIR/$tdir
check_default_stripe_attr --stripe-size
check_default_stripe_attr --stripe-index
}
run_test 204b "Print default stripe size and offset"
test_204c() {
test_mkdir $DIR/$tdir
$LFS setstripe --stripe-size 65536 $DIR/$tdir
check_default_stripe_attr --stripe-count
check_default_stripe_attr --stripe-index
}
run_test 204c "Print default stripe count and offset"
test_204d() {
test_mkdir $DIR/$tdir
$LFS setstripe --stripe-index 0 $DIR/$tdir
check_default_stripe_attr --stripe-count
check_default_stripe_attr --stripe-size
}
run_test 204d "Print default stripe count and size"
test_204e() {
test_mkdir $DIR/$tdir
$LFS setstripe -d $DIR/$tdir
# LU-16904 check if root is set as PFL layout
local numcomp=$($LFS getstripe --component-count $MOUNT)
if [[ $numcomp -gt 0 ]]; then
check_default_stripe_attr --stripe-count
else
check_default_stripe_attr --stripe-count --raw
fi
check_default_stripe_attr --stripe-size --raw
check_default_stripe_attr --stripe-index --raw
}
run_test 204e "Print raw stripe attributes"
test_204f() {
test_mkdir $DIR/$tdir
$LFS setstripe --stripe-count 1 $DIR/$tdir
check_default_stripe_attr --stripe-size --raw
check_default_stripe_attr --stripe-index --raw
}
run_test 204f "Print raw stripe size and offset"
test_204g() {
test_mkdir $DIR/$tdir
$LFS setstripe --stripe-size 65536 $DIR/$tdir
check_default_stripe_attr --stripe-count --raw
check_default_stripe_attr --stripe-index --raw
}
run_test 204g "Print raw stripe count and offset"
test_204h() {
test_mkdir $DIR/$tdir
$LFS setstripe --stripe-index 0 $DIR/$tdir
check_default_stripe_attr --stripe-count --raw
check_default_stripe_attr --stripe-size --raw
}
run_test 204h "Print raw stripe count and size"
# Figure out which job scheduler is being used, if any,
# or use a fake one
if [ -n "$SLURM_JOB_ID" ]; then # SLURM
JOBENV=SLURM_JOB_ID
elif [ -n "$LSB_JOBID" ]; then # Load Sharing Facility
JOBENV=LSB_JOBID
elif [ -n "$PBS_JOBID" ]; then # PBS/Maui/Moab
JOBENV=PBS_JOBID
elif [ -n "$LOADL_STEPID" ]; then # LoadLeveller
JOBENV=LOADL_STEP_ID
elif [ -n "$JOB_ID" ]; then # Sun Grid Engine
JOBENV=JOB_ID
else
$LCTL list_param jobid_name > /dev/null 2>&1
if [ $? -eq 0 ]; then
JOBENV=nodelocal
else
JOBENV=FAKE_JOBID
fi
fi
LUSTRE_JOBID_SIZE=31 # plus NUL terminator
verify_jobstats() {
local cmd=($1)
shift
local facets="$@"
# we don't really need to clear the stats for this test to work, since each
# command has a unique jobid, but it makes debugging easier if needed.
# for facet in $facets; do
# local dev=$(convert_facet2label $facet)
# # clear old jobstats
# do_facet $facet lctl set_param *.$dev.job_stats="clear"
# done
# use a new JobID for each test, or we might see an old one
[ "$JOBENV" = "FAKE_JOBID" ] &&
FAKE_JOBID=id.$testnum.$(basename ${cmd[0]}).$RANDOM
JOBVAL=${!JOBENV:0:$LUSTRE_JOBID_SIZE}
[ "$JOBENV" = "nodelocal" ] && {
FAKE_JOBID=id.$testnum.%e.$RANDOM
$LCTL set_param jobid_name=$FAKE_JOBID
JOBVAL=${FAKE_JOBID/\%e/$(basename ${cmd[0]})}
}
log "Test: ${cmd[*]}"
log "Using JobID environment $($LCTL get_param -n jobid_var)=$JOBVAL"
if [ $JOBENV = "FAKE_JOBID" ]; then
FAKE_JOBID=$JOBVAL ${cmd[*]}
else
${cmd[*]}
fi
# all files are created on OST0000
for facet in $facets; do
local stats="*.$(convert_facet2label $facet).job_stats"
# strip out libtool wrappers for in-tree executables
if (( $(do_facet $facet lctl get_param $stats |
sed -e 's/\.lt-/./' | grep -cw $JOBVAL) != 1 )); then
do_facet $facet lctl get_param $stats
error "No jobstats for $JOBVAL found on $facet::$stats"
fi
done
}
jobstats_set() {
local new_jobenv=$1
set_persistent_param_and_check client "jobid_var" \
"$FSNAME.sys.jobid_var" $new_jobenv
}
test_205a() { # Job stats
[ $PARALLEL == "yes" ] && skip "skip parallel run"
[[ $MDS1_VERSION -ge $(version_code 2.7.1) ]] ||
skip "Need MDS version with at least 2.7.1"
remote_mgs_nodsh && skip "remote MGS with nodsh"
remote_mds_nodsh && skip "remote MDS with nodsh"
remote_ost_nodsh && skip "remote OST with nodsh"
[ -z "$(lctl get_param -n mdc.*.connect_flags | grep jobstats)" ] &&
skip "Server doesn't support jobstats"
[[ $JOBID_VAR = disable ]] && skip_env "jobstats is disabled"
local old_jobenv=$($LCTL get_param -n jobid_var)
[ $old_jobenv != $JOBENV ] && jobstats_set $JOBENV
stack_trap "jobstats_set $old_jobenv" EXIT
changelog_register
local old_jobid_name=$($LCTL get_param jobid_name)
stack_trap "$LCTL set_param $old_jobid_name" EXIT
local old_interval=$(do_facet $SINGLEMDS lctl get_param -n \
mdt.*.job_cleanup_interval | head -n 1)
local new_interval=5
do_facet $SINGLEMDS \
$LCTL set_param mdt.*.job_cleanup_interval=$new_interval
stack_trap "do_facet $SINGLEMDS \
$LCTL set_param mdt.*.job_cleanup_interval=$old_interval" EXIT
local start=$SECONDS
local cmd
# mkdir
cmd="$LFS mkdir -i 0 -c 1 $DIR/$tdir"
verify_jobstats "$cmd" "$SINGLEMDS"
# rmdir
cmd="rmdir $DIR/$tdir"
verify_jobstats "$cmd" "$SINGLEMDS"
# mkdir on secondary MDT
if [ $MDSCOUNT -gt 1 ]; then
cmd="lfs mkdir -i 1 $DIR/$tdir.remote"
verify_jobstats "$cmd" "mds2"
fi
# mknod
cmd="mknod $DIR/$tfile c 1 3"
verify_jobstats "$cmd" "$SINGLEMDS"
# unlink
cmd="rm -f $DIR/$tfile"
verify_jobstats "$cmd" "$SINGLEMDS"
# create all files on OST0000 so verify_jobstats can find OST stats
# open & close
cmd="$LFS setstripe -i 0 -c 1 $DIR/$tfile"
verify_jobstats "$cmd" "$SINGLEMDS"
# setattr
cmd="touch $DIR/$tfile"
verify_jobstats "$cmd" "$SINGLEMDS ost1"
# write
cmd="dd if=/dev/zero of=$DIR/$tfile bs=1M count=1 oflag=sync"
verify_jobstats "$cmd" "ost1"
# read
cancel_lru_locks osc
cmd="dd if=$DIR/$tfile of=/dev/null bs=1M count=1 iflag=direct"
verify_jobstats "$cmd" "ost1"
# truncate
cmd="$TRUNCATE $DIR/$tfile 0"
verify_jobstats "$cmd" "$SINGLEMDS ost1"
# rename
cmd="mv -f $DIR/$tfile $DIR/$tdir.rename"
verify_jobstats "$cmd" "$SINGLEMDS"
# jobstats expiry - sleep until old stats should be expired
local left=$((new_interval + 5 - (SECONDS - start)))
[ $left -ge 0 ] && wait_update_facet $SINGLEMDS \
"lctl get_param *.*.job_stats | grep -c 'job_id.*mkdir'" \
"0" $left
cmd="$LFS mkdir -i 0 -c 1 $DIR/$tdir.expire"
verify_jobstats "$cmd" "$SINGLEMDS"
[ $(do_facet $SINGLEMDS lctl get_param *.*.job_stats |
grep -c "job_id.*mkdir") -gt 1 ] && error "old jobstats not expired"
# Ensure that jobid are present in changelog (if supported by MDS)
if [ $MDS1_VERSION -ge $(version_code 2.6.52) ];then
changelog_dump | tail -10
jobids=$(changelog_dump | tail -9 | grep -c "j=")
[ $jobids -eq 9 ] ||
error "Wrong changelog jobid count $jobids != 9"
# LU-5862
JOBENV="disable"
jobstats_set $JOBENV
touch $DIR/$tfile
changelog_dump | grep $tfile
jobids=$(changelog_dump | grep $tfile | tail -1 | grep -c "j=")
[ $jobids -eq 0 ] ||
error "Unexpected jobids when jobid_var=$JOBENV"
fi
# test '%j' access to environment variable - if supported
if lctl set_param jobid_var=USER jobid_name="S.%j.%e.%u.%h.E"; then
JOBENV="JOBCOMPLEX"
JOBCOMPLEX="S.$USER.touch.$(id -u).$(hostname).E"
verify_jobstats "touch $DIR/$tfile" $SINGLEMDS
fi
if lctl set_param jobid_var=USER jobid_name="S.%j.%e.%u.%H.E"; then
JOBENV="JOBCOMPLEX"
JOBCOMPLEX="S.$USER.touch.$(id -u).$(hostname -s).E"
verify_jobstats "touch $DIR/$tfile" $SINGLEMDS
fi
# test '%j' access to per-session jobid - if supported
if lctl list_param jobid_this_session > /dev/null 2>&1
then
lctl set_param jobid_var=session jobid_name="S.%j.%e.%u.%h.E"
lctl set_param jobid_this_session=$USER
JOBENV="JOBCOMPLEX"
JOBCOMPLEX="S.$USER.touch.$(id -u).$(hostname).E"
verify_jobstats "touch $DIR/$tfile" $SINGLEMDS
fi
}
run_test 205a "Verify job stats"
# LU-13117, LU-13597, LU-16599
test_205b() {
(( $MDS1_VERSION >= $(version_code 2.13.54.91) )) ||
skip "Need MDS version at least 2.13.54.91"
local job_stats="mdt.*.job_stats"
local old_jobid=$(do_facet mds1 $LCTL get_param jobid_var)
do_facet mds1 $LCTL set_param $job_stats=clear
# Setting jobid_var to USER might not be supported
[[ -n "$old_jobid" ]] && stack_trap "$LCTL set_param $old_jobid"
$LCTL set_param jobid_var=USER || true
stack_trap "$LCTL set_param $($LCTL get_param jobid_name)"
$LCTL set_param jobid_name="%j.%e.%u"
env -i USERTESTJOBSTATS=foolish touch $DIR/$tfile.1
do_facet mds1 $LCTL get_param $job_stats | grep "job_id:.*foolish" &&
{ do_facet mds1 $LCTL get_param $job_stats;
error "Unexpected jobid found"; }
do_facet mds1 $LCTL get_param $job_stats | grep "open:.*min.*max.*sum"||
{ do_facet mds1 $LCTL get_param $job_stats;
error "wrong job_stats format found"; }
(( $MDS1_VERSION <= $(version_code 2.15.0) )) &&
echo "MDS does not yet escape jobid" && return 0
mkdir_on_mdt0 $DIR/$tdir
$LCTL set_param jobid_var=TEST205b
env -i TEST205b="has sp" touch $DIR/$tdir/$tfile.2
local jobid=$(do_facet mds1 $LCTL get_param $job_stats |
awk '/has\\x20sp/ {print $3}')
[[ -n "$jobid" ]] || { do_facet mds1 $LCTL get_param $job_stats;
error "jobid not escaped"; }
if (( $MDS1_VERSION >= $(version_code 2.15.53.139) )); then
# need to run such a command on mds1:
# lctl set_param mdt.$FSNAME-MDT0000.job_stats='"has\x20sp.touch.0"'
#
# there might be multiple MDTs on single mds server, so need to
# specifiy MDT0000. Or the command will fail due to other MDTs
do_facet_vp mds1 $LCTL set_param mdt.$FSNAME-MDT0000.job_stats=$jobid ||
error "cannot clear escaped jobid in job_stats";
else
echo "MDS does not support clearing escaped jobid"
fi
}
run_test 205b "Verify job stats jobid and output format"
# LU-13733
test_205c() {
$LCTL set_param llite.*.stats=0
dd if=/dev/zero of=$DIR/$tfile.1 bs=4k count=1
$LCTL get_param llite.*.stats
$LCTL get_param llite.*.stats | grep \
"write_bytes *1 samples \[bytes\] 4096 4096 4096 16777216" ||
error "wrong client stats format found"
}
run_test 205c "Verify client stats format"
test_205d() {
local file=$DIR/$tdir/$tfile
(( $MDS1_VERSION >= $(version_code 2.15.53) )) ||
skip "need lustre >= 2.15.53 for lljobstat"
(( $OST1_VERSION >= $(version_code 2.15.53) )) ||
skip "need lustre >= 2.15.53 for lljobstat"
verify_yaml_available || skip_env "YAML verification not installed"
test_mkdir -i 0 $DIR/$tdir
$LFS setstripe -E 1M -L mdt -E -1 $file || error "create file failed"
stack_trap "rm -rf $DIR/$tdir"
dd if=/dev/zero of=$file bs=1M count=10 conv=sync ||
error "failed to write data to $file"
mv $file $file.2
do_facet mds1 "$LCTL get_param -n mdt.$FSNAME-MDT0000.rename_stats"
echo -n 'verify rename_stats...'
do_facet mds1 "$LCTL get_param -n mdt.$FSNAME-MDT0000.rename_stats" |
verify_yaml || error "rename_stats is not valid YAML"
echo " OK"
echo -n 'verify mdt job_stats...'
do_facet mds1 "$LCTL get_param -n mdt.$FSNAME-MDT0000.job_stats" |
verify_yaml || error "job_stats on mds1 is not valid YAML"
echo " OK"
echo -n 'verify ost job_stats...'
do_facet ost1 "$LCTL get_param -n obdfilter.$FSNAME-OST0000.job_stats" |
verify_yaml || error "job_stats on ost1 is not valid YAML"
echo " OK"
}
run_test 205d "verify the format of some stats files"
test_205e() {
local ops_comma
local file=$DIR/$tdir/$tfile
local -a cli_params
(( $MDS1_VERSION >= $(version_code 2.15.53) )) ||
skip "need lustre >= 2.15.53 for lljobstat"
(( $OST1_VERSION >= $(version_code 2.15.53) )) ||
skip "need lustre >= 2.15.53 for lljobstat"
verify_yaml_available || skip_env "YAML verification not installed"
cli_params=( $($LCTL get_param jobid_name jobid_var) )
$LCTL set_param jobid_var=nodelocal jobid_name=205e.%e.%u
stack_trap "$LCTL set_param ${cli_params[*]}" EXIT
mkdir_on_mdt0 $DIR/$tdir || error "failed to create dir"
stack_trap "rm -rf $DIR/$tdir"
$LFS setstripe -E EOF -i 0 -c 1 $file ||
error "failed to create $file on ost1"
dd if=/dev/zero of=$file bs=1M count=10 oflag=sync ||
error "failed to write data to $file"
do_facet mds1 "$LCTL get_param *.*.job_stats"
do_facet ost1 "$LCTL get_param *.*.job_stats"
do_facet ost1 "lljobstat -n 1 -i 0 -c 1000"
do_facet ost1 "lljobstat -n 1 -i 0 -c 1000" | verify_yaml ||
error "The output of lljobstat is not an valid YAML"
# verify that job dd.0 does exist and has some ops on ost1
# typically this line is like:
# - 205e.dd.0: {ops: 20, ...}
ops_comma=$(do_facet ost1 "lljobstat -n 1 -i 0 -c 1000" |
awk '$2=="205e.dd.0:" {print $4}')
(( ${ops_comma%,} >= 10 )) ||
error "cannot find job 205e.dd.0 with ops >= 10"
}
run_test 205e "verify the output of lljobstat"
test_205f() {
verify_yaml_available || skip_env "YAML verification not installed"
# check both qos_ost_weights and qos_mdt_weights
do_facet mds1 $LCTL get_param -n lod.*.qos*weights
do_facet mds1 $LCTL get_param -n lod.*.qos*weights | verify_yaml ||
error "qos_ost_weights is not valid YAML"
}
run_test 205f "verify qos_ost_weights YAML format "
__test_205_jobstats_dump() {
local -a pids
local nbr_instance=$1
while true; do
if (( ${#pids[@]} >= nbr_instance )); then
wait ${pids[@]}
pids=()
fi
do_facet mds1 "$LCTL get_param mdt.*.job_stats > /dev/null" &
pids+=( $! )
done
}
__test_205_cleanup() {
kill $@
# Clear all job entries
do_facet mds1 "$LCTL set_param mdt.*.job_stats=clear"
}
test_205g() {
local -a mds1_params
local -a cli_params
local pids
local interval=5
mds1_params=( $(do_facet mds1 $LCTL get_param mdt.*.job_cleanup_interval) )
do_facet mds1 $LCTL set_param mdt.*.job_cleanup_interval=$interval
stack_trap "do_facet mds1 $LCTL set_param ${mds1_params[*]}" EXIT
cli_params=( $($LCTL get_param jobid_name jobid_var) )
$LCTL set_param jobid_var=TEST205G_ID jobid_name=%j.%p
stack_trap "$LCTL set_param ${cli_params[*]}" EXIT
# start jobs loop
export TEST205G_ID=205g
stack_trap "unset TEST205G_ID" EXIT
while true; do
printf $DIR/$tfile.{0001..1000} | xargs -P10 -n1 touch
done & pids="$! "
__test_205_jobstats_dump 4 & pids+="$! "
stack_trap "__test_205_cleanup $pids" EXIT INT
[[ $SLOW == "no" ]] && sleep 90 || sleep 240
}
run_test 205g "stress test for job_stats procfile"
test_205h() {
(( $MDS1_VERSION >= $(version_code 2.15.57.7) )) ||
skip "Need MDS >= v2_15_57-7-g23a2db28dc for jobid xattr"
which getfattr > /dev/null 2>&1 || skip_env "no getfattr command"
local dir=$DIR/$tdir
local f=$dir/$tfile
local f2=$dir/$tfile-2
local f3=$dir/$tfile-3
local subdir=$DIR/dir
local val
local mdts=$(mdts_nodes)
local mds_saved=$(do_facet mds1 $LCTL get_param -n mdt.$FSNAME-MDT0000.job_xattr)
local client_saved=$($LCTL get_param -n jobid_var)
stack_trap "do_nodes $mdts $LCTL set_param mdt.*.job_xattr=$mds_saved" EXIT
stack_trap "$LCTL set_param jobid_var=$client_saved" EXIT
do_nodes $mdts $LCTL set_param mdt.*.job_xattr=user.job ||
error "failed to set job_xattr parameter to user.job"
$LCTL set_param jobid_var=procname.uid ||
error "failed to set jobid_var parameter"
test_mkdir $dir
touch $f
val=$(getfattr -n user.job $f | grep user.job)
[[ $val = user.job=\"touch.0\" ]] ||
error "expected user.job=\"touch.0\", got '$val'"
mkdir $subdir
val=$(getfattr -n user.job $subdir | grep user.job)
[[ $val = user.job=\"mkdir.0\" ]] ||
error "expected user.job=\"mkdir.0\", got '$val'"
do_nodes $mdts $LCTL set_param mdt.*.job_xattr=NONE ||
error "failed to set job_xattr parameter to NONE"
touch $f2
val=$(getfattr -d $f2)
[[ -z $val ]] ||
error "expected no user xattr, got '$val'"
do_nodes $mdts $LCTL set_param mdt.*.job_xattr=trusted.job ||
error "failed to set job_xattr parameter to trusted.job"
touch $f3
val=$(getfattr -n trusted.job $f3 | grep trusted.job)
[[ $val = trusted.job=\"touch.0\" ]] ||
error "expected trusted.job=\"touch.0\", got '$val'"
}
run_test 205h "check jobid xattr is stored correctly"
test_205i() {
(( $MDS1_VERSION >= $(version_code 2.15.57.7) )) ||
skip "Need MDS >= v2_15_57-7-g23a2db28dc for jobid xattr"
local mdts=$(mdts_nodes)
local mds_saved=$(do_facet mds1 $LCTL get_param -n mdt.$FSNAME-MDT0000.job_xattr)
stack_trap "do_nodes $mdts $LCTL set_param mdt.*.job_xattr=$mds_saved" EXIT
do_nodes $mdts $LCTL set_param mdt.*.job_xattr=user.1234567 ||
error "failed to set mdt.*.job_xattr to user.1234567"
do_nodes $mdts $LCTL set_param mdt.*.job_xattr=user.12345678 &&
error "failed to reject too long job_xattr name"
do_nodes $mdts $LCTL set_param mdt.*.job_xattr=userjob &&
error "failed to reject job_xattr name in bad format"
do_nodes $mdts $LCTL set_param mdt.*.job_xattr=user.job/ &&
error "failed to reject job_xattr name with invalid character"
do_nodes $mdts "printf 'mdt.*.job_xattr=user.job\x80' |
xargs $LCTL set_param" &&
error "failed to reject job_xattr name with non-ascii character"
return 0
}
run_test 205i "check job_xattr parameter accepts and rejects values correctly"
# LU-17512
# test '?' operator
test_205k(){
(( MDS1_VERSION >= $(version_code 2.15.64.85) )) ||
skip "need MDS >= v2_15_64-85-gecdcaa3986 for jobstats fix"
# copied from test_205a
[[ $PARALLEL == "yes" ]] && skip "skip parallel run"
remote_mds_nodsh && skip "remote MDS with nodsh"
[[ "$($LCTL get_param -n mdc.*.connect_flags)" =~ jobstats ]] ||
skip "Server doesn't support jobstats"
[[ "$JOBID_VAR" == "disable" ]] && skip_env "jobstats is disabled"
local jobid_save=$($LCTL get_param jobid_var jobid_name)
stack_trap "$LCTL set_param $jobid_save"
do_facet mds1 $LCTL set_param mdt.*.job_stats=clear
$LCTL set_param jobid_var=FAKE jobid_name=%j?%H:%e:%u
cp -a /etc/hosts $DIR;
job_id_name=($(do_facet mds1 $LCTL get_param mdt.*.job_stats |
awk '/job_id:/ { print $3 }' | tr -d '"'))
[[ "$job_id_name" == "$(hostname -s):cp:0" ]] ||
error "expected $(hostname -s):cp:0, got $job_id_name instead"
do_facet mds1 $LCTL set_param mdt.*.job_stats=clear
$LCTL set_param jobid_var=USER jobid_name=%j?%H:%e:%u
cp -a /etc/hosts $DIR;
job_id_name=($(do_facet mds1 $LCTL get_param mdt.*.job_stats |
awk '/job_id:/ { print $3 }' | tr -d '"'))
[[ "$job_id_name" == "$USER:cp:0" ]] ||
error "expected $USER:cp:0 but got $job_id_name instead"
}
run_test 205k "Verify '?' operator on job stats"
test_205l() {
[[ $PARALLEL != "yes" ]] || skip "skip parallel run"
remote_mds_nodsh && skip "remote MDS with nodsh"
[[ "$($LCTL get_param -n mdc.*.connect_flags)" =~ jobstats ]] ||
skip "Server doesn't support jobstats"
[[ "$JOBID_VAR" != "disable" ]] || skip_env "jobstats is disabled"
local jobid_save=$($LCTL get_param jobid_var jobid_name)
stack_trap "$LCTL set_param $jobid_save"
local tmpdir=$(mktemp -d /tmp/jobstat-XXXXXX)
local jobs=$tmpdir/jobs.txt
local mv_save=${tmpdir}/local_mv
local mdts=$(mdts_nodes)
local mv_job
local n=1
local limit=500
[[ $SLOW == "no" ]] || limit=50000
cp -a /etc/hosts $DIR/hosts
cp $(which mv) ${mv_save}
do_nodes $mdts "$LCTL set_param jobid_var=procname_uid jobid_name=%e.%u mdt.*.job_cleanup_interval=5"
sleep 5
do_nodes $mdts "$LCTL set_param mdt.*.job_stats=clear mdt.*.job_cleanup_interval=0"
sleep 5
# Add a series of easily identifyable jobs
for ((n = 0; n < limit; n++)); do
mv_job=${tmpdir}/mv.$(printf %08d $n)
mv ${mv_save} ${mv_job}
${mv_job} $DIR/hosts $DIR/hosts.$(printf %08d $n)
${mv_job} $DIR/hosts.$(printf %08d $n) $DIR/hosts
mv ${mv_job} ${mv_save}
done
# Duplicates indicate restart issues
do_nodes $mdts \
"$LCTL get_param mdt.*.job_stats | grep job_id: | cut -d. -f2" \
> ${jobs}
local dupes=$(grep -v -e "^${RUNAS_ID}\$" -e '^0$' ${jobs} | sort |
uniq -d | wc -l)
(( ${dupes} == 0 )) ||
error "seq_write wrote ${dupes} duplicate entries."
# Unexpected jobs indicate cleanup issues
local njobs=$(grep -v -e "^${RUNAS_ID}\$" -e '^0$' ${jobs} | wc -l)
(( ${njobs} == ${limit} )) ||
error "seq_write wrote ${njobs} jobs expected ${limit}."
do_nodes $mdts "$LCTL set_param mdt.*.job_cleanup_interval=5"
sleep 5
do_nodes $mdts "$LCTL set_param mdt.*.job_stats=clear"
# On success the scrach files are not interesting
rm -fr ${tmpdir}
}
run_test 205l "Verify job stats can scale"
test_205m() {
local trunc_hostname=${HOSTNAME:0:2}
local dir=$DIR/205m/
local tests=(
"%.3e.%u" "cp /etc/hosts $dir" "cp.500"
"%.3e.%u" "touch $dir/1" "tou.500"
"%.3e.%u.%.2h" "touch $dir/2" "tou.500.${trunc_hostname}"
"%.3e" "touch $dir/3" "tou"
"%.1e.%u.%.2h" "touch $dir/4" "t.500.${trunc_hostname}"
"%.2p" "touch $dir/8" '${pid:0:2}'
)
cli_params=( $($LCTL get_param jobid_name jobid_var) )
stack_trap "$LCTL set_param ${cli_params[*]}" EXIT
stack_trap "do_facet mds1 $LCTL set_param mdt.*.job_stats=clear" EXIT
mkdir $DIR/205m
chown $RUNAS_ID $DIR/205m
chgrp $RUNAS_ID $DIR/205m
for (( i = 0; i < ${#tests[@]} ; i += 3 )); do
local jobid_name=${tests[i]}
local cmd=${tests[i+1]}
do_facet mds1 $LCTL set_param mdt.*.job_stats=clear
$LCTL set_param jobid_var=nodelocal jobid_name=${jobid_name}
runas -u 500 $cmd & pid=$!
wait $pid
eval "expected=${tests[i+2]}"
do_facet mds1 $LCTL get_param mdt.*.job_stats |
awk '/job_id:/ {print} /job_id:/ && $3 == "'$expected'"
{found=1} END {exit(!found)}' ||
error "expected ${expected}, got ${job_id_name} instead"
done
}
run_test 205m "Test width parsing of job_stats"
# LU-1480, LU-1773 and LU-1657
test_206() {
mkdir -p $DIR/$tdir
$LFS setstripe -c -1 $DIR/$tdir
#define OBD_FAIL_LOV_INIT 0x1403
$LCTL set_param fail_loc=0xa0001403
$LCTL set_param fail_val=1
touch $DIR/$tdir/$tfile || true
}
run_test 206 "fail lov_init_raid0() doesn't lbug"
test_207a() {
dd if=/dev/zero of=$DIR/$tfile bs=4k count=$((RANDOM%10+1))
local fsz=`stat -c %s $DIR/$tfile`
cancel_lru_locks mdc
# do not return layout in getattr intent
#define OBD_FAIL_MDS_NO_LL_GETATTR 0x170
$LCTL set_param fail_loc=0x170
local sz=`stat -c %s $DIR/$tfile`
[ $fsz -eq $sz ] || error "file size expected $fsz, actual $sz"
rm -rf $DIR/$tfile
}
run_test 207a "can refresh layout at glimpse"
test_207b() {
dd if=/dev/zero of=$DIR/$tfile bs=4k count=$((RANDOM%10+1))
local cksum=`md5sum $DIR/$tfile`
local fsz=`stat -c %s $DIR/$tfile`
cancel_lru_locks mdc
cancel_lru_locks osc
# do not return layout in getattr intent
#define OBD_FAIL_MDS_NO_LL_OPEN 0x171
$LCTL set_param fail_loc=0x171
# it will refresh layout after the file is opened but before read issues
echo checksum is "$cksum"
echo "$cksum" |md5sum -c --quiet || error "file differs"
rm -rf $DIR/$tfile
}
run_test 207b "can refresh layout at open"
test_208() {
# FIXME: in this test suite, only RD lease is used. This is okay
# for now as only exclusive open is supported. After generic lease
# is done, this test suite should be revised. - Jinshan
remote_mds_nodsh && skip "remote MDS with nodsh"
[[ $MDS1_VERSION -ge $(version_code 2.4.52) ]] ||
skip "Need MDS version at least 2.4.52"
echo "==== test 1: verify get lease work"
$MULTIOP $DIR/$tfile oO_CREAT:O_RDWR:eRE+eU || error "get lease error"
echo "==== test 2: verify lease can be broken by upcoming open"
$MULTIOP $DIR/$tfile oO_RDWR:eR_E-eUc &
local PID=$!
sleep 2
$MULTIOP $DIR/$tfile oO_RDWR:c
kill -USR1 $PID && wait $PID || error "break lease error"
echo "==== test 3: verify lease can't be granted if an open already exists"
$MULTIOP $DIR/$tfile oO_RDWR:_c &
local PID=$!
sleep 2
$MULTIOP $DIR/$tfile oO_RDWR:eReUc && error "apply lease should fail"
kill -USR1 $PID && wait $PID || error "open file error"
echo "==== test 4: lease can sustain over recovery"
$MULTIOP $DIR/$tfile oO_RDWR:eR_E+eUc &
PID=$!
sleep 2
fail mds1
kill -USR1 $PID && wait $PID || error "lease broken over recovery"
echo "==== test 5: lease broken can't be regained by replay"
$MULTIOP $DIR/$tfile oO_RDWR:eR_E-eUc &
PID=$!
sleep 2
# open file to break lease and then recovery
$MULTIOP $DIR/$tfile oO_RDWR:c || error "open file error"
fail mds1
kill -USR1 $PID && wait $PID || error "lease not broken over recovery"
rm -f $DIR/$tfile
}
run_test 208 "Exclusive open"
test_209() {
[ -z "$(lctl get_param -n mdc.*.connect_flags | grep disp_stripe)" ] &&
skip_env "must have disp_stripe"
touch $DIR/$tfile
sync; sleep 5; sync;
echo 3 > /proc/sys/vm/drop_caches
[ -f /sys/kernel/slab/ptlrpc_cache/shrink ] &&
echo 1 > /sys/kernel/slab/ptlrpc_cache/shrink
req_before=$(awk '/ptlrpc_cache / { print $2 }' /proc/slabinfo)
# open/close 500 times
for i in $(seq 500); do
cat $DIR/$tfile
done
echo 3 > /proc/sys/vm/drop_caches
[ -f /sys/kernel/slab/ptlrpc_cache/shrink ] &&
echo 1 > /sys/kernel/slab/ptlrpc_cache/shrink
req_after=$(awk '/ptlrpc_cache / { print $2 }' /proc/slabinfo)
echo "before: $req_before, after: $req_after"
[ $((req_after - req_before)) -ge 300 ] &&
error "open/close requests are not freed"
return 0
}
run_test 209 "read-only open/close requests should be freed promptly"
test_210() {
local pid
$MULTIOP $DIR/$tfile oO_CREAT:O_RDWR:eW_E+eUc &
pid=$!
sleep 1
$LFS getstripe $DIR/$tfile
kill -USR1 $pid
wait $pid || error "multiop failed"
$MULTIOP $DIR/$tfile oO_RDONLY:eR_E+eUc &
pid=$!
sleep 1
$LFS getstripe $DIR/$tfile
kill -USR1 $pid
wait $pid || error "multiop failed"
}
run_test 210 "lfs getstripe does not break leases"
test_212() {
size=`date +%s`
size=$((size % 8192 + 1))
dd if=/dev/urandom of=$DIR/f212 bs=1k count=$size
# check path2fid does not trigger nosplice operations
$LFS path2fid $DIR/f212
sendfile $DIR/f212 $DIR/f212.xyz || error "sendfile wrong"
rm -f $DIR/f212 $DIR/f212.xyz
}
run_test 212 "Sendfile test"
test_213() {
dd if=/dev/zero of=$DIR/$tfile bs=4k count=4
cancel_lru_locks osc
lctl set_param fail_loc=0x8000040f
# generate a read lock
cat $DIR/$tfile > /dev/null
# write to the file, it will try to cancel the above read lock.
cat /etc/hosts >> $DIR/$tfile
}
run_test 213 "OSC lock completion and cancel race don't crash - bug 18829"
test_214() { # for bug 20133
mkdir -p $DIR/$tdir/d214c || error "mkdir $DIR/$tdir/d214c failed"
for (( i=0; i < 340; i++ )) ; do
touch $DIR/$tdir/d214c/a$i
done
ls -l $DIR/$tdir || error "ls -l $DIR/d214p failed"
mv $DIR/$tdir/d214c $DIR/ || error "mv $DIR/d214p/d214c $DIR/ failed"
ls $DIR/d214c || error "ls $DIR/d214c failed"
rm -rf $DIR/$tdir || error "rm -rf $DIR/d214* failed"
rm -rf $DIR/d214* || error "rm -rf $DIR/d214* failed"
}
run_test 214 "hash-indexed directory test - bug 20133"
# having "abc" as 1st arg, creates $TMP/lnet_abc.out and $TMP/lnet_abc.sys
create_lnet_proc_files() {
lctl get_param -n $1 >$TMP/lnet_$1.sys || error "cannot read lnet.$1"
}
# counterpart of create_lnet_proc_files
remove_lnet_proc_files() {
rm -f $TMP/lnet_$1.sys
}
# uses 1st arg as trailing part of filename, 2nd arg as description for reports,
# 3rd arg as regexp for body
check_lnet_proc_stats() {
local l=$(cat "$TMP/lnet_$1" |wc -l)
[ $l = 1 ] || (cat "$TMP/lnet_$1" && error "$2 is not of 1 line: $l")
grep -E "$3" "$TMP/lnet_$1" ||
(cat "$TMP/lnet_$1" && error "$2 misformatted")
}
# uses 1st arg as trailing part of filename, 2nd arg as description for reports,
# 3rd arg as regexp for body, 4th arg as regexp for 1st line, 5th arg is
# optional and can be regexp for 2nd line (lnet.routes case)
check_lnet_proc_entry() {
local blp=2 # blp stands for 'position of 1st line of body'
[ -z "$5" ] || blp=3 # lnet.routes case
local l=$(cat "$TMP/lnet_$1" |wc -l)
# subtracting one from $blp because the body can be empty
[ "$l" -ge "$(($blp - 1))" ] || (cat "$TMP/lnet_$1" && error "$2 is too short: $l")
sed -n '1 p' "$TMP/lnet_$1" |grep -E "$4" >/dev/null ||
(cat "$TMP/lnet_$1" && error "1st line of $2 misformatted")
[ "$5" = "" ] || sed -n '2 p' "$TMP/lnet_$1" |grep -E "$5" >/dev/null ||
(cat "$TMP/lnet_$1" && error "2nd line of $2 misformatted")
# bail out if any unexpected line happened
sed -n "$blp p" "$TMP/lnet_$1" | grep -Ev "$3"
[ "$?" != 0 ] || error "$2 misformatted"
}
test_215() { # for bugs 18102, 21079, 21517
[ $PARALLEL == "yes" ] && skip "skip parallel run"
local N='(0|[1-9][0-9]*)' # non-negative numeric
local P='[1-9][0-9]*' # positive numeric
local I='(0|-?[1-9][0-9]*|NA)' # any numeric (0 | >0 | <0) or NA if no value
local NET='[a-z][a-z0-9]*' # LNet net like o2ib2
local ADDR='[0-9.]+' # LNet addr like 10.0.0.1
local ADDR6='([0-9a-f]{0,4}:){2,7}[0-9a-f]{0,4}' # IPv6 LNet addr
local NID="$ADDR@$NET" # LNet nid like 10.0.0.1@o2ib2
local NID6="$ADDR6@$NET" # LNet nid like 2601:8c1:c180::cbdd@tcp
local L1 # regexp for 1st line
local L2 # regexp for 2nd line (optional)
local BR # regexp for the rest (body)
# lnet.stats should look as 11 space-separated non-negative numerics
BR="^$N $N $N $N $N $N $N $N $N $N $N$"
create_lnet_proc_files "stats"
check_lnet_proc_stats "stats.sys" "lnet.stats" "$BR"
remove_lnet_proc_files "stats"
# lnet.routes should look like this:
# Routing disabled/enabled/starting/stopping
# net hops priority state router
# where net is a string like tcp0, hops > 0, priority >= 0,
# state is up/down,
# router is a string like 192.168.1.1@tcp2
L1="^Routing (disabled|enabled|starting|stopping)$"
L2="^net +hops +priority +state +router$"
BR="^$NET +$N +(0|1) +(up|down) +($NID|$NID6)$"
create_lnet_proc_files "routes"
check_lnet_proc_entry "routes.sys" "lnet.routes" "$BR" "$L1" "$L2"
remove_lnet_proc_files "routes"
# lnet.routers should look like this:
# ref rtr_ref alive_cnt state last_ping ping_sent deadline down_ni router
# where ref > 0, rtr_ref > 0, alive_cnt >= 0, state is up/down,
# last_ping >= 0, ping_sent is boolean (0/1), deadline and down_ni are
# numeric (0 or >0 or <0), router is a string like 192.168.1.1@tcp2
L1="^ref +rtr_ref +alive +router$"
BR="^$P +$P +(up|down) +($NID|$NID6)$"
create_lnet_proc_files "routers"
check_lnet_proc_entry "routers.sys" "lnet.routers" "$BR" "$L1"
remove_lnet_proc_files "routers"
# lnet.peers should look like this:
# nid refs state last max rtr min tx min queue
# where nid is a string like 192.168.1.1@tcp2, refs > 0,
# state is up/down/NA, max >= 0. last, rtr, min, tx, min are
# numeric (0 or >0 or <0), queue >= 0.
L1="^nid +refs +state +last +max +rtr +min +tx +min +queue$"
BR="^($NID|$NID6) +$P +(up|down|NA) +$I +$N +$I +$I +$I +$I +$N$"
create_lnet_proc_files "peers"
check_lnet_proc_entry "peers.sys" "lnet.peers" "$BR" "$L1"
remove_lnet_proc_files "peers"
# lnet.buffers should look like this:
# pages count credits min
# where pages >=0, count >=0, credits and min are numeric (0 or >0 or <0)
L1="^pages +count +credits +min$"
BR="^ +$N +$N +$I +$I$"
create_lnet_proc_files "buffers"
check_lnet_proc_entry "buffers.sys" "lnet.buffers" "$BR" "$L1"
remove_lnet_proc_files "buffers"
# lnet.nis should look like this:
# nid status alive refs peer rtr max tx min
# where nid is a string like 192.168.1.1@tcp2, status is up/down,
# alive is numeric (0 or >0 or <0), refs >= 0, peer >= 0,
# rtr >= 0, max >=0, tx and min are numeric (0 or >0 or <0).
L1="^nid +status +alive +refs +peer +rtr +max +tx +min$"
BR="^($NID|$NID6) +(up|down) +$I +$N +$N +$N +$N +$I +$I$"
create_lnet_proc_files "nis"
check_lnet_proc_entry "nis.sys" "lnet.nis" "$BR" "$L1"
remove_lnet_proc_files "nis"
# can we successfully write to lnet.stats?
lctl set_param -n stats=0 || error "cannot write to lnet.stats"
}
run_test 215 "lnet exists and has proper content - bugs 18102, 21079, 21517"
test_216() { # bug 20317
[ $PARALLEL == "yes" ] && skip "skip parallel run"
remote_ost_nodsh && skip "remote OST with nodsh"
local node
local facets=$(get_facets OST)
local p="$TMP/$TESTSUITE-$TESTNAME.parameters"
save_lustre_params $facets \
"ldlm.namespaces.filter-*.max_nolock_bytes" > $p
save_lustre_params $facets \
"ldlm.namespaces.filter-*.contended_locks" >> $p
save_lustre_params $facets \
"ldlm.namespaces.filter-*.contention_seconds" >> $p
clear_stats osc.*.osc_stats
local osts=$(osts_nodes)
# agressive lockless i/o settings
do_nodes $osts \
"lctl set_param -n ldlm.namespaces.*.max_nolock_bytes=2000000 \
ldlm.namespaces.filter-*.contended_locks=0 \
ldlm.namespaces.filter-*.contention_seconds=60"
$DIRECTIO write $DIR/$tfile 0 10 4096
$CHECKSTAT -s 40960 $DIR/$tfile
# disable lockless i/o
do_nodes $osts \
"lctl set_param -n ldlm.namespaces.filter-*.max_nolock_bytes=0 \
ldlm.namespaces.filter-*.contended_locks=32 \
ldlm.namespaces.filter-*.contention_seconds=0"
clear_stats osc.*.osc_stats
dd if=/dev/zero of=$DIR/$tfile count=0
$CHECKSTAT -s 0 $DIR/$tfile
restore_lustre_params <$p
rm -f $p
rm $DIR/$tfile
}
run_test 216 "check lockless direct write updates file size and kms correctly"
test_217() { # bug 22430
[ $PARALLEL == "yes" ] && skip "skip parallel run"
local nodes=$(nodes_list)
local node
for node in ${nodes//,/ }; do
local nid=$(host_nids_address $node $NETTYPE)
local node_ip=$(do_node $node getent ahostsv4 $node |
awk '{ print $1; exit; }')
echo "node: '$node', nid: '$nid', node_ip='$node_ip'"
# if hostname matches any NID, use hostname for better testing
if [[ -z "$nid" || "$nid" =~ "$node_ip" ]]; then
echo "lctl ping node $node@$NETTYPE"
lctl ping $node@$NETTYPE ||
error "ping $node@$NETTYPE failed rc=$?"
else # otherwise, at least test 'lctl ping' is working
echo "lctl ping nid $(h2nettype $nid)"
lctl ping $(h2nettype $nid) ||
error "ping $(h2nettype $nid) failed rc=$?"
echo "skipping $node (no hyphen detected)"
fi
done
return 0
}
run_test 217 "check lctl ping for hostnames with embedded hyphen ('-')"
test_218() {
# do directio so as not to populate the page cache
log "creating a 10 Mb file"
$MULTIOP $DIR/$tfile oO_CREAT:O_DIRECT:O_RDWR:w$((10*1048576))c ||
error "multiop failed while creating a file"
log "starting reads"
dd if=$DIR/$tfile of=/dev/null bs=4096 &
log "truncating the file"
$MULTIOP $DIR/$tfile oO_TRUNC:c ||
error "multiop failed while truncating the file"
log "killing dd"
kill %+ || true # reads might have finished
echo "wait until dd is finished"
wait
log "removing the temporary file"
rm -rf $DIR/$tfile || error "tmp file removal failed"
}
run_test 218 "parallel read and truncate should not deadlock"
test_219() {
[ $PARALLEL == "yes" ] && skip "skip parallel run"
# write one partial page
dd if=/dev/zero of=$DIR/$tfile bs=1024 count=1
# set no grant so vvp_io_commit_write will do sync write
$LCTL set_param fail_loc=0x411
# write a full page at the end of file
dd if=/dev/zero of=$DIR/$tfile bs=4096 count=1 seek=1 conv=notrunc
$LCTL set_param fail_loc=0
dd if=/dev/zero of=$DIR/$tfile bs=4096 count=1 seek=3
$LCTL set_param fail_loc=0x411
dd if=/dev/zero of=$DIR/$tfile bs=1024 count=1 seek=2 conv=notrunc
# LU-4201
dd if=/dev/zero of=$DIR/$tfile-2 bs=1024 count=1
$CHECKSTAT -s 1024 $DIR/$tfile-2 || error "checkstat wrong size"
}
run_test 219 "LU-394: Write partial won't cause uncontiguous pages vec at LND"
test_220() { #LU-325
[ $PARALLEL == "yes" ] && skip "skip parallel run"
remote_ost_nodsh && skip "remote OST with nodsh"
remote_mds_nodsh && skip "remote MDS with nodsh"
remote_mgs_nodsh && skip "remote MGS with nodsh"
local OSTIDX=0
# create on MDT0000 so the last_id and next_id are correct
mkdir_on_mdt0 $DIR/$tdir
local OST=$($LFS df --ost=$OSTIDX $DIR | awk '{ print $1 }')
OST=${OST%_UUID}
# on the mdt's osc
local mdtosc_proc1=$(get_mdtosc_proc_path $SINGLEMDS $OST)
local last_id=$(do_facet $SINGLEMDS lctl get_param -n \
osp.$mdtosc_proc1.prealloc_last_id)
local next_id=$(do_facet $SINGLEMDS lctl get_param -n \
osp.$mdtosc_proc1.prealloc_next_id)
$LFS df -i
do_facet ost$((OSTIDX + 1)) lctl set_param fail_val=-1
#define OBD_FAIL_OST_ENOINO 0x229
do_facet ost$((OSTIDX + 1)) lctl set_param fail_loc=0x229
create_pool $FSNAME.$TESTNAME || return 1
do_facet mgs $LCTL pool_add $FSNAME.$TESTNAME $OST || return 2
$LFS setstripe $DIR/$tdir -i $OSTIDX -c 1 -p $FSNAME.$TESTNAME
MDSOBJS=$((last_id - next_id))
echo "preallocated objects on MDS is $MDSOBJS" "($last_id - $next_id)"
blocks=$($LFS df $MOUNT | awk '($1 == '$OSTIDX') { print $4 }')
echo "OST still has $count kbytes free"
echo "create $MDSOBJS files @next_id..."
createmany -o $DIR/$tdir/f $MDSOBJS || return 3
local last_id2=$(do_facet mds${MDSIDX} lctl get_param -n \
osp.$mdtosc_proc1.prealloc_last_id)
local next_id2=$(do_facet mds${MDSIDX} lctl get_param -n \
osp.$mdtosc_proc1.prealloc_next_id)
echo "after creation, last_id=$last_id2, next_id=$next_id2"
$LFS df -i
echo "cleanup..."
do_facet ost$((OSTIDX + 1)) lctl set_param fail_val=0
do_facet ost$((OSTIDX + 1)) lctl set_param fail_loc=0
do_facet mgs $LCTL pool_remove $FSNAME.$TESTNAME $OST ||
error "$LCTL pool_remove $FSNAME.$TESTNAME $OST failed"
do_facet mgs $LCTL pool_destroy $FSNAME.$TESTNAME ||
error "$LCTL pool_destroy $FSNAME.$TESTNAME failed"
echo "unlink $MDSOBJS files @$next_id..."
unlinkmany $DIR/$tdir/f $MDSOBJS || error "unlinkmany failed"
}
run_test 220 "preallocated MDS objects still used if ENOSPC from OST"
test_221() {
[ $PARALLEL == "yes" ] && skip "skip parallel run"
dd if=`which date` of=$MOUNT/date oflag=sync
chmod +x $MOUNT/date
#define OBD_FAIL_LLITE_FAULT_TRUNC_RACE 0x1401
$LCTL set_param fail_loc=0x80001401
$MOUNT/date > /dev/null
rm -f $MOUNT/date
}
run_test 221 "make sure fault and truncate race to not cause OOM"
test_222a () {
[ $PARALLEL == "yes" ] && skip "skip parallel run"
rm -rf $DIR/$tdir
test_mkdir $DIR/$tdir
$LFS setstripe -c 1 -i 0 $DIR/$tdir
createmany -o $DIR/$tdir/$tfile 10
cancel_lru_locks mdc
cancel_lru_locks osc
#define OBD_FAIL_LDLM_AGL_DELAY 0x31a
$LCTL set_param fail_loc=0x31a
ls -l $DIR/$tdir > /dev/null || error "AGL for ls failed"
$LCTL set_param fail_loc=0
rm -r $DIR/$tdir
}
run_test 222a "AGL for ls should not trigger CLIO lock failure"
test_222b () {
[ $PARALLEL == "yes" ] && skip "skip parallel run"
rm -rf $DIR/$tdir
test_mkdir $DIR/$tdir
$LFS setstripe -c 1 -i 0 $DIR/$tdir
createmany -o $DIR/$tdir/$tfile 10
cancel_lru_locks mdc
cancel_lru_locks osc
#define OBD_FAIL_LDLM_AGL_DELAY 0x31a
$LCTL set_param fail_loc=0x31a
rm -r $DIR/$tdir || error "AGL for rmdir failed"
$LCTL set_param fail_loc=0
}
run_test 222b "AGL for rmdir should not trigger CLIO lock failure"
test_223 () {
[ $PARALLEL == "yes" ] && skip "skip parallel run"
rm -rf $DIR/$tdir
test_mkdir $DIR/$tdir
$LFS setstripe -c 1 -i 0 $DIR/$tdir
createmany -o $DIR/$tdir/$tfile 10
cancel_lru_locks mdc
cancel_lru_locks osc
#define OBD_FAIL_LDLM_AGL_NOLOCK 0x31b
$LCTL set_param fail_loc=0x31b
ls -l $DIR/$tdir > /dev/null || error "reenqueue failed"
$LCTL set_param fail_loc=0
rm -r $DIR/$tdir
}
run_test 223 "osc reenqueue if without AGL lock granted ======================="
test_224a() { # LU-1039, MRP-303
[ $PARALLEL == "yes" ] && skip "skip parallel run"
#define OBD_FAIL_PTLRPC_CLIENT_BULK_CB 0x508
$LCTL set_param fail_loc=0x508
dd if=/dev/zero of=$DIR/$tfile bs=1M count=1 conv=fsync
stack_trap "rm -f $DIR/$tfile"
$LCTL set_param fail_loc=0
df $DIR
}
run_test 224a "Don't panic on bulk IO failure"
test_224bd_sub() { # LU-1039, MRP-303
[ $PARALLEL == "yes" ] && skip "skip parallel run"
local timeout=$1
shift
dd if=/dev/urandom of=$TMP/$tfile bs=1M count=1
$LFS setstripe -c 1 -i 0 $DIR/$tfile
dd if=$TMP/$tfile of=$DIR/$tfile bs=1M count=1
stack_trap "rm -f $DIR/$tfile"
cancel_lru_locks osc
set_checksums 0
stack_trap "set_checksums $ORIG_CSUM" EXIT
local at_max_saved=0
# adaptive timeouts may prevent seeing the issue
if at_is_enabled; then
at_max_saved=$(at_max_get mds)
at_max_set 0 mds client
stack_trap "at_max_set $at_max_saved mds client" EXIT
fi
#define OBD_FAIL_PTLRPC_CLIENT_BULK_CB2 0x515
do_facet ost1 $LCTL set_param fail_val=$timeout fail_loc=0x80000515
dd of=$TMP/$tfile.new if=$DIR/$tfile bs=1M count=1 || "$@"
do_facet ost1 $LCTL set_param fail_loc=0
cmp $TMP/$tfile $TMP/$tfile.new || error "file contents wrong"
df $DIR
}
test_224b() {
test_224bd_sub 3 error "dd failed"
}
run_test 224b "Don't panic on bulk IO failure"
test_224c() { # LU-6441
[ $PARALLEL == "yes" ] && skip "skip parallel run"
remote_mds_nodsh && skip "remote MDS with nodsh"
local p="$TMP/$TESTSUITE-$TESTNAME.parameters"
local osts=$(osts_nodes)
save_writethrough $p
set_cache $osts writethrough on
local pages_per_rpc=$($LCTL get_param osc.*.max_pages_per_rpc)
local at_max=$($LCTL get_param -n at_max)
local timeout=$($LCTL get_param -n timeout)
local test_at="at_max"
local param_at="$FSNAME.sys.at_max"
local test_timeout="timeout"
local param_timeout="$FSNAME.sys.timeout"
$LCTL set_param -n osc.*.max_pages_per_rpc=1024
set_persistent_param_and_check client "$test_at" "$param_at" 0
set_persistent_param_and_check client "$test_timeout" "$param_timeout" 5
#define OBD_FAIL_PTLRPC_CLIENT_BULK_CB3 0x520
do_facet ost1 "$LCTL set_param fail_loc=0x520"
$LFS setstripe -c 1 -i 0 $DIR/$tfile
stack_trap "rm -f $DIR/$tfile"
dd if=/dev/zero of=$DIR/$tfile bs=8MB count=1
sync
do_facet ost1 "$LCTL set_param fail_loc=0"
set_persistent_param_and_check client "$test_at" "$param_at" $at_max
set_persistent_param_and_check client "$test_timeout" "$param_timeout" \
$timeout
$LCTL set_param -n $pages_per_rpc
restore_lustre_params < $p
rm -f $p
}
run_test 224c "Don't hang if one of md lost during large bulk RPC"
test_224d() { # LU-11169
test_224bd_sub $((TIMEOUT + 2)) error "dd failed"
}
run_test 224d "Don't corrupt data on bulk IO timeout"
MDSSURVEY=${MDSSURVEY:-$(which mds-survey 2>/dev/null || true)}
test_225a () {
[ $PARALLEL == "yes" ] && skip "skip parallel run"
if [ -z ${MDSSURVEY} ]; then
skip_env "mds-survey not found"
fi
[ $MDS1_VERSION -ge $(version_code 2.2.51) ] ||
skip "Need MDS version at least 2.2.51"
local mds=$(facet_host $SINGLEMDS)
local target=$(do_nodes $mds 'lctl dl' |
awk '{ if ($2 == "UP" && $3 == "mdt") { print $4 }}')
local cmd1="file_count=1000 thrhi=4"
local cmd2="dir_count=2 layer=mdd stripe_count=0"
local cmd3="rslt_loc=${TMP} targets=\"$mds:$target\" $MDSSURVEY"
local cmd="$cmd1 $cmd2 $cmd3"
rm -f ${TMP}/mds_survey*
echo + $cmd
eval $cmd || error "mds-survey with zero-stripe failed"
cat ${TMP}/mds_survey*
rm -f ${TMP}/mds_survey*
}
run_test 225a "Metadata survey sanity with zero-stripe"
test_225b () {
if [ -z ${MDSSURVEY} ]; then
skip_env "mds-survey not found"
fi
[ $MDS1_VERSION -ge $(version_code 2.2.51) ] ||
skip "Need MDS version at least 2.2.51"
[ $PARALLEL == "yes" ] && skip "skip parallel run"
remote_mds_nodsh && skip "remote MDS with nodsh"
if [ $($LCTL dl | grep -c osc) -eq 0 ]; then
skip_env "Need to mount OST to test"
fi
local mds=$(facet_host $SINGLEMDS)
local target=$(do_nodes $mds 'lctl dl' |
awk '{ if ($2 == "UP" && $3 == "mdt") { print $4 }}')
local cmd1="file_count=1000 thrhi=4"
local cmd2="dir_count=2 layer=mdd stripe_count=1"
local cmd3="rslt_loc=${TMP} targets=\"$mds:$target\" $MDSSURVEY"
local cmd="$cmd1 $cmd2 $cmd3"
rm -f ${TMP}/mds_survey*
echo + $cmd
eval $cmd || error "mds-survey with stripe_count failed"
cat ${TMP}/mds_survey*
rm -f ${TMP}/mds_survey*
}
run_test 225b "Metadata survey sanity with stripe_count = 1"
mcreate_path2fid () {
local mode=$1
local major=$2
local minor=$3
local name=$4
local desc=$5
local path=$DIR/$tdir/$name
local fid
local rc
local fid_path
$MCREATE --mode=$1 --major=$2 --minor=$3 $path ||
error "cannot create $desc"
fid=$($LFS path2fid $path | tr -d '[' | tr -d ']')
rc=$?
[ $rc -ne 0 ] && error "cannot get fid of a $desc"
fid_path=$($LFS fid2path $MOUNT $fid)
rc=$?
[ $rc -ne 0 ] && error "cannot get path of $desc by $DIR $path $fid"
[ "$path" == "$fid_path" ] ||
error "fid2path returned $fid_path, expected $path"
echo "pass with $path and $fid"
}
test_226a () {
rm -rf $DIR/$tdir
mkdir -p $DIR/$tdir
mcreate_path2fid 0010666 0 0 fifo "FIFO"
mcreate_path2fid 0020666 1 3 null "character special file (null)"
mcreate_path2fid 0020666 1 255 none "character special file (no device)"
mcreate_path2fid 0040666 0 0 dir "directory"
mcreate_path2fid 0060666 7 0 loop0 "block special file (loop)"
mcreate_path2fid 0100666 0 0 file "regular file"
mcreate_path2fid 0120666 0 0 link "symbolic link"
mcreate_path2fid 0140666 0 0 sock "socket"
}
run_test 226a "call path2fid and fid2path on files of all type"
test_226b () {
[ $MDSCOUNT -lt 2 ] && skip_env "needs >= 2 MDTs"
local MDTIDX=1
rm -rf $DIR/$tdir
mkdir -p $DIR/$tdir
$LFS setdirstripe -i $MDTIDX $DIR/$tdir/remote_dir ||
error "create remote directory failed"
mcreate_path2fid 0010666 0 0 "remote_dir/fifo" "FIFO"
mcreate_path2fid 0020666 1 3 "remote_dir/null" \
"character special file (null)"
mcreate_path2fid 0020666 1 255 "remote_dir/none" \
"character special file (no device)"
mcreate_path2fid 0040666 0 0 "remote_dir/dir" "directory"
mcreate_path2fid 0060666 7 0 "remote_dir/loop0" \
"block special file (loop)"
mcreate_path2fid 0100666 0 0 "remote_dir/file" "regular file"
mcreate_path2fid 0120666 0 0 "remote_dir/link" "symbolic link"
mcreate_path2fid 0140666 0 0 "remote_dir/sock" "socket"
}
run_test 226b "call path2fid and fid2path on files of all type under remote dir"
test_226c () {
[ $MDSCOUNT -lt 2 ] && skip_env "needs >= 2 MDTs"
[[ $MDS1_VERSION -ge $(version_code 2.13.55) ]] ||
skip "Need MDS version at least 2.13.55"
local submnt=/mnt/submnt
local srcfile=/etc/passwd
local dstfile=$submnt/passwd
local path
local fid
rm -rf $DIR/$tdir
rm -rf $submnt
$LFS setdirstripe -c -1 -i 1 $DIR/$tdir ||
error "create remote directory failed"
mkdir -p $submnt || error "create $submnt failed"
FILESET="$FILESET/$tdir" mount_client $submnt ||
error "mount $submnt failed"
stack_trap "umount $submnt;rm -rf $submit" EXIT
cp $srcfile $dstfile
fid=$($LFS path2fid $dstfile)
path=$($LFS fid2path $submnt "$fid")
[ "$path" = "$dstfile" ] ||
error "fid2path $submnt $fid failed ($path != $dstfile)"
}
run_test 226c "call path2fid and fid2path under remote dir with subdir mount"
test_226d () {
(( $CLIENT_VERSION >= $(version_code 2.15.57) )) ||
skip "Need client at least version 2.15.57"
# Define First test dataset
local testdirs_01=$DIR/$tdir
local testdata_01=$testdirs_01/${tdir}_01
local testresult_01=${tdir}_01
# Define Second test dataset
local testdirs_02=$DIR/$tdir/$tdir
local testdata_02=$testdirs_02/${tdir}_02
local testresult_02=${tdir}_02
# Define third test dataset (top level)
local testdata_03=$DIR/${tdir}_03
local testresult_03=${tdir}_03
# Create first test dataset
mkdir -p $testdirs_01 || error "cannot create dir $testdirs_01"
touch $testdata_01 || error "cannot create file $testdata_01"
# Create second test dataset
mkdir -p $testdirs_02 || error "cannot create dir $testdirs_02"
touch $testdata_02 || error "cannot create file $testdata_02"
# Create third test dataset
touch $testdata_03 || error "cannot create file $testdata_03"
local fid01=$($LFS getstripe -F "$testdata_01") ||
error "getstripe failed on $testdata_01"
local fid02=$($LFS getstripe -F "$testdata_02") ||
error "getstripe failed on $testdata_01"
local fid03=$($LFS getstripe -F "$testdata_03") ||
error "getstripe failed on $testdata_03"
# Verify only -n option
local out1=$($LFS fid2path -n $DIR $fid01) ||
error "fid2path failed on $fid01"
local out2=$($LFS fid2path -n $DIR $fid02) ||
error "fid2path failed on $fid02"
local out3=$($LFS fid2path -n $DIR $fid03) ||
error "fid2path failed on $fid03"
[[ "$out1" == "$testresult_01" ]] ||
error "fid2path failed: Expected $testresult_01 got $out1"
[[ "$out2" == "$testresult_02" ]] ||
error "fid2path failed: Expected $testresult_02 got $out2"
[[ "$out3" == "$testresult_03" ]] ||
error "fid2path failed: Expected $testresult_03 got $out3"
# Verify with option -fn together
out1=$($LFS fid2path -fn $DIR $fid01) ||
error "fid2path -fn failed on $fid01"
out2=$($LFS fid2path -fn $DIR $fid02) ||
error "fid2path -fn failed on $fid02"
out3=$($LFS fid2path -fn $DIR $fid03) ||
error "fid2path -fn failed on $fid03"
local tmpout=$(echo $out1 | cut -d" " -f2)
[[ "$tmpout" == "$testresult_01" ]] ||
error "fid2path -fn failed: Expected $testresult_01 got $out1"
tmpout=$(echo $out2 | cut -d" " -f2)
[[ "$tmpout" == "$testresult_02" ]] ||
error "fid2path -fn failed: Expected $testresult_02 got $out2"
tmpout=$(echo $out3 | cut -d" " -f2)
[[ "$tmpout" == "$testresult_03" ]] ||
error "fid2path -fn failed: Expected $testresult_03 got $out3"
}
run_test 226d "verify fid2path with -n and -fn option"
test_226e () {
(( $CLIENT_VERSION >= $(version_code 2.15.56) )) ||
skip "Need client at least version 2.15.56"
# Define filename with 'newline' and a space
local testfile="Test"$'\n'"file 01"
# Define link name with multiple 'newline' and a space
local linkfile="Link"$'\n'"file "$'\n'"01"
# Remove prior hard link
rm -f $DIR/"$linkfile"
# Create file
touch $DIR/"$testfile"
# Create link
ln $DIR/"$testfile" $DIR/"$linkfile"
local fid=$($LFS getstripe -F "$DIR/$testfile") ||
error "getstripe failed on $DIR/$testfile"
# Call with -0 option
local out1=$($LFS fid2path -0 $DIR $fid | xargs --null -n1 \
echo "FILE:" | grep -c "FILE:")
# With -0 option the output should be exactly 2 lines.
(( $out1 == 2 )) || error "fid2path -0 failed on $fid, $out1"
}
run_test 226e "Verify path2fid -0 option with newline and space"
# LU-1299 Executing or running ldd on a truncated executable does not
# cause an out-of-memory condition.
test_227() {
[ $PARALLEL == "yes" ] && skip "skip parallel run"
[ -z "$(which ldd)" ] && skip_env "should have ldd tool"
dd if=$(which date) of=$MOUNT/date bs=1k count=1
chmod +x $MOUNT/date
$MOUNT/date > /dev/null
ldd $MOUNT/date > /dev/null
rm -f $MOUNT/date
}
run_test 227 "running truncated executable does not cause OOM"
# LU-1512 try to reuse idle OI blocks
test_228a() {
[ $PARALLEL == "yes" ] && skip "skip parallel run"
remote_mds_nodsh && skip "remote MDS with nodsh"
[ "$mds1_FSTYPE" != "ldiskfs" ] && skip_env "ldiskfs only test"
local MDT_DEV=$(mdsdevname ${SINGLEMDS//mds/})
local myDIR=$DIR/$tdir
mkdir -p $myDIR
#define OBD_FAIL_SEQ_EXHAUST 0x1002
$LCTL set_param fail_loc=0x80001002
createmany -o $myDIR/t- 10000
$LCTL set_param fail_loc=0
# The guard is current the largest FID holder
touch $myDIR/guard
local SEQ=$($LFS path2fid $myDIR/guard | awk -F ':' '{print $1}' |
tr -d '[')
local IDX=$(($SEQ % 64))
do_facet $SINGLEMDS sync
# Make sure journal flushed.
sleep 6
local blk1=$(do_facet $SINGLEMDS \
"$DEBUGFS -c -R \\\"stat oi.16.${IDX}\\\" $MDT_DEV" |
grep Blockcount | awk '{print $4}')
# Remove old files, some OI blocks will become idle.
unlinkmany $myDIR/t- 10000
# Create new files, idle OI blocks should be reused.
createmany -o $myDIR/t- 2000
do_facet $SINGLEMDS sync
# Make sure journal flushed.
sleep 6
local blk2=$(do_facet $SINGLEMDS \
"$DEBUGFS -c -R \\\"stat oi.16.${IDX}\\\" $MDT_DEV" |
grep Blockcount | awk '{print $4}')
[ $blk1 == $blk2 ] || error "old blk1=$blk1, new blk2=$blk2, unmatched!"
}
run_test 228a "try to reuse idle OI blocks"
test_228b() {
[ $PARALLEL == "yes" ] && skip "skip parallel run"
remote_mds_nodsh && skip "remote MDS with nodsh"
[ "$mds1_FSTYPE" != "ldiskfs" ] && skip_env "ldiskfs only test"
local MDT_DEV=$(mdsdevname ${SINGLEMDS//mds/})
local myDIR=$DIR/$tdir
mkdir -p $myDIR
#define OBD_FAIL_SEQ_EXHAUST 0x1002
$LCTL set_param fail_loc=0x80001002
createmany -o $myDIR/t- 10000
$LCTL set_param fail_loc=0
# The guard is current the largest FID holder
touch $myDIR/guard
local SEQ=$($LFS path2fid $myDIR/guard | awk -F ':' '{print $1}' |
tr -d '[')
local IDX=$(($SEQ % 64))
do_facet $SINGLEMDS sync
# Make sure journal flushed.
sleep 6
local blk1=$(do_facet $SINGLEMDS \
"$DEBUGFS -c -R \\\"stat oi.16.${IDX}\\\" $MDT_DEV" |
grep Blockcount | awk '{print $4}')
# Remove old files, some OI blocks will become idle.
unlinkmany $myDIR/t- 10000
# stop the MDT
stop $SINGLEMDS || error "Fail to stop MDT."
# remount the MDT
start $SINGLEMDS $(facet_device $SINGLEMDS) $MDS_MOUNT_OPTS ||
error "Fail to start MDT."
client_up || error "Fail to df."
# Create new files, idle OI blocks should be reused.
createmany -o $myDIR/t- 2000
do_facet $SINGLEMDS sync
# Make sure journal flushed.
sleep 6
local blk2=$(do_facet $SINGLEMDS \
"$DEBUGFS -c -R \\\"stat oi.16.${IDX}\\\" $MDT_DEV" |
grep Blockcount | awk '{print $4}')
[ $blk1 == $blk2 ] || error "old blk1=$blk1, new blk2=$blk2, unmatched!"
}
run_test 228b "idle OI blocks can be reused after MDT restart"
#LU-1881
test_228c() {
[ $PARALLEL == "yes" ] && skip "skip parallel run"
remote_mds_nodsh && skip "remote MDS with nodsh"
[ "$mds1_FSTYPE" != "ldiskfs" ] && skip_env "ldiskfs only test"
local MDT_DEV=$(mdsdevname ${SINGLEMDS//mds/})
local myDIR=$DIR/$tdir
mkdir -p $myDIR
#define OBD_FAIL_SEQ_EXHAUST 0x1002
$LCTL set_param fail_loc=0x80001002
# 20000 files can guarantee there are index nodes in the OI file
createmany -o $myDIR/t- 20000
$LCTL set_param fail_loc=0
# The guard is current the largest FID holder
touch $myDIR/guard
local SEQ=$($LFS path2fid $myDIR/guard | awk -F ':' '{print $1}' |
tr -d '[')
local IDX=$(($SEQ % 64))
do_facet $SINGLEMDS sync
# Make sure journal flushed.
sleep 6
local blk1=$(do_facet $SINGLEMDS \
"$DEBUGFS -c -R \\\"stat oi.16.${IDX}\\\" $MDT_DEV" |
grep Blockcount | awk '{print $4}')
# Remove old files, some OI blocks will become idle.
unlinkmany $myDIR/t- 20000
rm -f $myDIR/guard
# The OI file should become empty now
# Create new files, idle OI blocks should be reused.
createmany -o $myDIR/t- 2000
do_facet $SINGLEMDS sync
# Make sure journal flushed.
sleep 6
local blk2=$(do_facet $SINGLEMDS \
"$DEBUGFS -c -R \\\"stat oi.16.${IDX}\\\" $MDT_DEV" |
grep Blockcount | awk '{print $4}')
[ $blk1 == $blk2 ] || error "old blk1=$blk1, new blk2=$blk2, unmatched!"
}
run_test 228c "NOT shrink the last entry in OI index node to recycle idle leaf"
test_229() { # LU-2482, LU-3448
[[ $PARALLEL == "yes" ]] && skip "skip parallel run"
(( $OSTCOUNT >= 2 )) || skip_env "needs >= 2 OSTs"
(( $MDS1_VERSION >= $(version_code 2.4.50.0-53-ga61ff5914b) )) ||
skip "need MDS >= 2.4.50.53 for HSM released file"
rm -f $DIR/$tfile
# Create a file with a released layout and stripe count 2.
$MULTIOP $DIR/$tfile H2c ||
error "failed to create file with released layout"
$LFS getstripe -v $DIR/$tfile
local pattern=$($LFS getstripe -L $DIR/$tfile)
[[ "$pattern" =~ "released" ]] || error "pattern error ($pattern)"
local stripe_count=$($LFS getstripe -c $DIR/$tfile) ||
error "getstripe"
(( $stripe_count == 2 )) || error "stripe count not 2 ($stripe_count)"
stat $DIR/$tfile || error "failed to stat released file"
chown $RUNAS_ID $DIR/$tfile ||
error "chown $RUNAS_ID $DIR/$tfile failed"
chgrp $RUNAS_ID $DIR/$tfile ||
error "chgrp $RUNAS_ID $DIR/$tfile failed"
touch $DIR/$tfile || error "touch $DIR/$tfile failed"
rm $DIR/$tfile || error "failed to remove released file"
}
run_test 229 "getstripe/stat/rm/attr changes work on released files"
test_230a() {
[ $PARALLEL == "yes" ] && skip "skip parallel run"
[ $MDSCOUNT -lt 2 ] && skip_env "needs >= 2 MDTs"
[ $MDS1_VERSION -lt $(version_code 2.11.52) ] &&
skip "Need MDS version at least 2.11.52"
local MDTIDX=1
test_mkdir $DIR/$tdir
test_mkdir -i0 -c1 $DIR/$tdir/test_230_local
local mdt_idx=$($LFS getstripe -m $DIR/$tdir/test_230_local)
[ $mdt_idx -ne 0 ] &&
error "create local directory on wrong MDT $mdt_idx"
$LFS mkdir -i $MDTIDX $DIR/$tdir/test_230 ||
error "create remote directory failed"
local mdt_idx=$($LFS getstripe -m $DIR/$tdir/test_230)
[ $mdt_idx -ne $MDTIDX ] &&
error "create remote directory on wrong MDT $mdt_idx"
createmany -o $DIR/$tdir/test_230/t- 10 ||
error "create files on remote directory failed"
mdt_idx=$($LFS getstripe -m $DIR/$tdir/test_230/t-0)
[ $mdt_idx -ne $MDTIDX ] && error "create files on wrong MDT $mdt_idx"
rm -r $DIR/$tdir || error "unlink remote directory failed"
}
run_test 230a "Create remote directory and files under the remote directory"
test_230b() {
[ $PARALLEL == "yes" ] && skip "skip parallel run"
[ $MDSCOUNT -lt 2 ] && skip_env "needs >= 2 MDTs"
[ $MDS1_VERSION -lt $(version_code 2.11.52) ] &&
skip "Need MDS version at least 2.11.52"
local MDTIDX=1
local mdt_index
local i
local file
local pid
local stripe_count
local migrate_dir=$DIR/$tdir/migrate_dir
local other_dir=$DIR/$tdir/other_dir
test_mkdir $DIR/$tdir
test_mkdir -i0 -c1 $migrate_dir
test_mkdir -i0 -c1 $other_dir
for ((i=0; i<10; i++)); do
mkdir -p $migrate_dir/dir_${i}
createmany -o $migrate_dir/dir_${i}/f 10 ||
error "create files under remote dir failed $i"
done
cp /etc/passwd $migrate_dir/$tfile
cp /etc/passwd $other_dir/$tfile
chattr +SAD $migrate_dir
chattr +SAD $migrate_dir/$tfile
local old_dir_flag=$(lsattr -a $migrate_dir | awk '/\/\.$/ {print $1}')
local old_file_flag=$(lsattr $migrate_dir/$tfile | awk '{print $1}')
local old_dir_mode=$(stat -c%f $migrate_dir)
local old_file_mode=$(stat -c%f $migrate_dir/$tfile)
mkdir -p $migrate_dir/dir_default_stripe2
$LFS setstripe -c 2 $migrate_dir/dir_default_stripe2
$LFS setstripe -c 2 $migrate_dir/${tfile}_stripe2
mkdir -p $other_dir
ln $migrate_dir/$tfile $other_dir/luna
ln $migrate_dir/$tfile $migrate_dir/sofia
ln $other_dir/$tfile $migrate_dir/david
ln -s $migrate_dir/$tfile $other_dir/zachary
ln -s $migrate_dir/$tfile $migrate_dir/${tfile}_ln
ln -s $other_dir/$tfile $migrate_dir/${tfile}_ln_other
local len
local lnktgt
# inline symlink
for len in 58 59 60; do
lnktgt=$(str_repeat 'l' $len)
touch $migrate_dir/$lnktgt
ln -s $lnktgt $migrate_dir/${len}char_ln
done
# PATH_MAX
for len in 4094 4095; do
lnktgt=$(str_repeat 'l' $len)
ln -s $lnktgt $migrate_dir/${len}char_ln
done
# NAME_MAX
for len in 254 255; do
touch $migrate_dir/$(str_repeat 'l' $len)
done
$LFS migrate -m $MDTIDX $migrate_dir ||
error "fails on migrating remote dir to MDT1"
echo "migratate to MDT1, then checking.."
for ((i = 0; i < 10; i++)); do
for file in $(find $migrate_dir/dir_${i}); do
mdt_index=$($LFS getstripe -m $file)
# broken symlink getstripe will fail
[ $mdt_index -ne $MDTIDX ] && stat -L $file &&
error "$file is not on MDT${MDTIDX}"
done
done
# the multiple link file should still in MDT0
mdt_index=$($LFS getstripe -m $migrate_dir/$tfile)
[ $mdt_index == 0 ] ||
error "$file is not on MDT${MDTIDX}"
local new_dir_flag=$(lsattr -a $migrate_dir | awk '/\/\.$/ {print $1}')
[ "$old_dir_flag" = "$new_dir_flag" ] ||
error " expect $old_dir_flag get $new_dir_flag"
local new_file_flag=$(lsattr $migrate_dir/$tfile | awk '{print $1}')
[ "$old_file_flag" = "$new_file_flag" ] ||
error " expect $old_file_flag get $new_file_flag"
local new_dir_mode=$(stat -c%f $migrate_dir)
[ "$old_dir_mode" = "$new_dir_mode" ] ||
error "expect mode $old_dir_mode get $new_dir_mode"
local new_file_mode=$(stat -c%f $migrate_dir/$tfile)
[ "$old_file_mode" = "$new_file_mode" ] ||
error "expect mode $old_file_mode get $new_file_mode"
diff /etc/passwd $migrate_dir/$tfile ||
error "$tfile different after migration"
diff /etc/passwd $other_dir/luna ||
error "luna different after migration"
diff /etc/passwd $migrate_dir/sofia ||
error "sofia different after migration"
diff /etc/passwd $migrate_dir/david ||
error "david different after migration"
diff /etc/passwd $other_dir/zachary ||
error "zachary different after migration"
diff /etc/passwd $migrate_dir/${tfile}_ln ||
error "${tfile}_ln different after migration"
diff /etc/passwd $migrate_dir/${tfile}_ln_other ||
error "${tfile}_ln_other different after migration"
stripe_count=$($LFS getstripe -c $migrate_dir/dir_default_stripe2)
[ $stripe_count = 2 ] ||
error "dir strpe_count $d != 2 after migration."
stripe_count=$($LFS getstripe -c $migrate_dir/${tfile}_stripe2)
[ $stripe_count = 2 ] ||
error "file strpe_count $d != 2 after migration."
#migrate back to MDT0
MDTIDX=0
$LFS migrate -m $MDTIDX $migrate_dir ||
error "fails on migrating remote dir to MDT0"
echo "migrate back to MDT0, checking.."
for file in $(find $migrate_dir); do
mdt_index=$($LFS getstripe -m $file)
[ $mdt_index -ne $MDTIDX ] && stat -L $file &&
error "$file is not on MDT${MDTIDX}"
done
local new_dir_flag=$(lsattr -a $migrate_dir | awk '/\/\.$/ {print $1}')
[ "$old_dir_flag" = "$new_dir_flag" ] ||
error " expect $old_dir_flag get $new_dir_flag"
local new_file_flag=$(lsattr $migrate_dir/$tfile | awk '{print $1}')
[ "$old_file_flag" = "$new_file_flag" ] ||
error " expect $old_file_flag get $new_file_flag"
local new_dir_mode=$(stat -c%f $migrate_dir)
[ "$old_dir_mode" = "$new_dir_mode" ] ||
error "expect mode $old_dir_mode get $new_dir_mode"
local new_file_mode=$(stat -c%f $migrate_dir/$tfile)
[ "$old_file_mode" = "$new_file_mode" ] ||
error "expect mode $old_file_mode get $new_file_mode"
diff /etc/passwd ${migrate_dir}/$tfile ||
error "$tfile different after migration"
diff /etc/passwd ${other_dir}/luna ||
error "luna different after migration"
diff /etc/passwd ${migrate_dir}/sofia ||
error "sofia different after migration"
diff /etc/passwd ${other_dir}/zachary ||
error "zachary different after migration"
diff /etc/passwd $migrate_dir/${tfile}_ln ||
error "${tfile}_ln different after migration"
diff /etc/passwd $migrate_dir/${tfile}_ln_other ||
error "${tfile}_ln_other different after migration"
stripe_count=$($LFS getstripe -c ${migrate_dir}/dir_default_stripe2)
[ $stripe_count = 2 ] ||
error "dir strpe_count $d != 2 after migration."
stripe_count=$($LFS getstripe -c ${migrate_dir}/${tfile}_stripe2)
[ $stripe_count = 2 ] ||
error "file strpe_count $d != 2 after migration."
rm -rf $DIR/$tdir || error "rm dir failed after migration"
}
run_test 230b "migrate directory"
test_230c() {
[ $PARALLEL == "yes" ] && skip "skip parallel run"
[ $MDSCOUNT -lt 2 ] && skip_env "needs >= 2 MDTs"
remote_mds_nodsh && skip "remote MDS with nodsh"
[ $MDS1_VERSION -lt $(version_code 2.11.52) ] &&
skip "Need MDS version at least 2.11.52"
local MDTIDX=1
local total=3
local mdt_index
local file
local migrate_dir=$DIR/$tdir/migrate_dir
#If migrating directory fails in the middle, all entries of
#the directory is still accessiable.
test_mkdir $DIR/$tdir
test_mkdir -i0 -c1 $migrate_dir
test_mkdir -i1 -c1 $DIR/$tdir/remote_dir
stat $migrate_dir
createmany -o $migrate_dir/f $total ||
error "create files under ${migrate_dir} failed"
# fail after migrating top dir, and this will fail only once, so the
# first sub file migration will fail (currently f3), others succeed.
#OBD_FAIL_MIGRATE_ENTRIES 0x1801
do_facet mds1 lctl set_param fail_loc=0x1801
local t=$(ls $migrate_dir | wc -l)
$LFS migrate --mdt-index $MDTIDX $migrate_dir &&
error "migrate should fail"
local u=$(ls $migrate_dir | wc -l)
[ "$u" == "$t" ] || error "$u != $t during migration"
# add new dir/file should succeed
mkdir $migrate_dir/dir ||
error "mkdir failed under migrating directory"
touch $migrate_dir/file ||
error "create file failed under migrating directory"
# add file with existing name should fail
for file in $migrate_dir/f*; do
stat $file > /dev/null || error "stat $file failed"
$OPENFILE -f O_CREAT:O_EXCL $file &&
error "open(O_CREAT|O_EXCL) $file should fail"
$MULTIOP $file m && error "create $file should fail"
touch $DIR/$tdir/remote_dir/$tfile ||
error "touch $tfile failed"
ln $DIR/$tdir/remote_dir/$tfile $file &&
error "link $file should fail"
mdt_index=$($LFS getstripe -m $file)
if [ $mdt_index == 0 ]; then
# file failed to migrate is not allowed to rename to
mv $DIR/$tdir/remote_dir/$tfile $file &&
error "rename to $file should fail"
else
mv $DIR/$tdir/remote_dir/$tfile $file ||
error "rename to $file failed"
fi
echo hello >> $file || error "write $file failed"
done
# resume migration with different options should fail
$LFS migrate -m 0 $migrate_dir &&
error "migrate -m 0 $migrate_dir should fail"
$LFS migrate -m $MDTIDX -c 2 $migrate_dir &&
error "migrate -c 2 $migrate_dir should fail"
# resume migration should succeed
$LFS migrate -m $MDTIDX $migrate_dir ||
error "migrate $migrate_dir failed"
echo "Finish migration, then checking.."
for file in $(find $migrate_dir); do
mdt_index=$($LFS getstripe -m $file)
[ $mdt_index == $MDTIDX ] ||
error "$file is not on MDT${MDTIDX}"
done
if (($MDS1_VERSION >= $(version_code 2.16.50) )); then
echo "Migrate a dir with an open file"
touch $migrate_dir/foo || error "failed to create file"
local foo_fid=$($LFS path2fid $migrate_dir/foo)
$MULTIOP $migrate_dir/foo o_c &
local bg_pid=$!
sleep 1
$LFS migrate -m 0 $migrate_dir || {
kill -USR1 $bg_pid
error "migrate a dir with an open file fails"
}
kill -USR1 $bg_pid
wait $bg_pid || error "multiop fails"
local foo_fid_new=$($LFS path2fid $migrate_dir/foo)
[[ "$foo_fid" == "$foo_fid_new" ]] ||
error "migrate should skip an open file $foo_fid != $foo_fid_new"
for file in $(find $migrate_dir); do
[[ $file == "$migrate_dir/foo" ]] && continue
mdt_index=$($LFS getstripe -m $file)
[[ "$mdt_index" == "0" ]] ||
error "$file is not on MDT0"
done
# repeating the migrate should leave the dir's FID the same
# and update only the skipped file's FID
local old_dir_fid=$($LFS path2fid $migrate_dir)
local old_file_fid=$($LFS path2fid $migrate_dir/file)
$LFS migrate -m 0 $migrate_dir || error "Repeating migrate fails"
local new_dir_fid=$($LFS path2fid $migrate_dir)
local new_file_fid=$($LFS path2fid $migrate_dir/file)
foo_fid_new=$($LFS path2fid $migrate_dir/foo)
[[ "$foo_fid_old" != "$foo_fid_new" ]] ||
error "Expecting the skipped file to be migrated, but its FID is the same"
[[ "$old_dir_fid" != "$new_fid_dir" ]] ||
error "The top-level dir has been migrated after repeating the same migrate cmd"
[[ "$old_file_fid" == "$new_file_fid" ]] ||
error "The migrated file has been migrated again"
fi
rm -rf $DIR/$tdir || error "rm dir failed after migration"
}
run_test 230c "check directory accessiblity if migration failed"
test_230d() {
[ $PARALLEL == "yes" ] && skip "skip parallel run"
[ $MDSCOUNT -lt 2 ] && skip_env "needs >= 2 MDTs"
[ $MDS1_VERSION -lt $(version_code 2.11.52) ] &&
skip "Need MDS version at least 2.11.52"
# LU-11235
[ "$mds1_FSTYPE" == "zfs" ] && skip "skip ZFS backend"
local migrate_dir=$DIR/$tdir/migrate_dir
local old_index
local new_index
local old_count
local new_count
local new_hash
local mdt_index
local i
local j
old_index=$((RANDOM % MDSCOUNT))
old_count=$((MDSCOUNT - old_index))
new_index=$((RANDOM % MDSCOUNT))
new_count=$((MDSCOUNT - new_index))
new_hash=1 # for all_char
[ $old_count -gt 1 ] && old_count=$((old_count - RANDOM % old_count))
[ $new_count -gt 1 ] && new_count=$((new_count - RANDOM % new_count))
test_mkdir $DIR/$tdir
test_mkdir -i $old_index -c $old_count $migrate_dir
for ((i=0; i<100; i++)); do
test_mkdir -i0 -c1 $migrate_dir/dir_${i}
createmany -o $migrate_dir/dir_${i}/f 100 ||
error "create files under remote dir failed $i"
done
echo -n "Migrate from MDT$old_index "
[ $old_count -gt 1 ] && echo -n "... MDT$((old_index + old_count - 1)) "
echo -n "to MDT$new_index"
[ $new_count -gt 1 ] && echo -n " ... MDT$((new_index + new_count - 1))"
echo
echo "$LFS migrate -m$new_index -c$new_count -H $new_hash $migrate_dir"
$LFS migrate -m $new_index -c $new_count -H $new_hash $migrate_dir ||
error "migrate remote dir error"
echo "Finish migration, then checking.."
for file in $(find $migrate_dir -maxdepth 1); do
mdt_index=$($LFS getstripe -m $file)
if [ $mdt_index -lt $new_index ] ||
[ $mdt_index -gt $((new_index + new_count - 1)) ]; then
error "$file is on MDT$mdt_index"
fi
done
rm -rf $DIR/$tdir || error "rm dir failed after migration"
}
run_test 230d "check migrate big directory"
test_230e() {
[ $PARALLEL == "yes" ] && skip "skip parallel run"
[ $MDSCOUNT -lt 2 ] && skip_env "needs >= 2 MDTs"
[ $MDS1_VERSION -lt $(version_code 2.11.52) ] &&
skip "Need MDS version at least 2.11.52"
local i
local j
local a_fid
local b_fid
mkdir_on_mdt0 $DIR/$tdir
mkdir $DIR/$tdir/migrate_dir
mkdir $DIR/$tdir/other_dir
touch $DIR/$tdir/migrate_dir/a
ln $DIR/$tdir/migrate_dir/a $DIR/$tdir/other_dir/b
ls $DIR/$tdir/other_dir
$LFS migrate -m 1 $DIR/$tdir/migrate_dir ||
error "migrate dir fails"
mdt_index=$($LFS getstripe -m $DIR/$tdir/migrate_dir)
[ $mdt_index == 1 ] || error "migrate_dir is not on MDT1"
mdt_index=$($LFS getstripe -m $DIR/$tdir/migrate_dir/a)
[ $mdt_index == 0 ] || error "a is not on MDT0"
$LFS migrate -m 1 $DIR/$tdir/other_dir ||
error "migrate dir fails"
mdt_index=$($LFS getstripe -m $DIR/$tdir/other_dir)
[ $mdt_index == 1 ] || error "other_dir is not on MDT1"
mdt_index=$($LFS getstripe -m $DIR/$tdir/migrate_dir/a)
[ $mdt_index == 1 ] || error "a is not on MDT1"
mdt_index=$($LFS getstripe -m $DIR/$tdir/other_dir/b)
[ $mdt_index == 1 ] || error "b is not on MDT1"
a_fid=$($LFS path2fid $DIR/$tdir/migrate_dir/a)
b_fid=$($LFS path2fid $DIR/$tdir/other_dir/b)
[ "$a_fid" = "$b_fid" ] || error "different fid after migration"
rm -rf $DIR/$tdir || error "rm dir failed after migration"
}
run_test 230e "migrate mulitple local link files"
test_230f() {
[ $PARALLEL == "yes" ] && skip "skip parallel run"
[ $MDSCOUNT -lt 2 ] && skip_env "needs >= 2 MDTs"
[ $MDS1_VERSION -lt $(version_code 2.11.52) ] &&
skip "Need MDS version at least 2.11.52"
local a_fid
local ln_fid
mkdir -p $DIR/$tdir
mkdir $DIR/$tdir/migrate_dir
$LFS mkdir -i1 $DIR/$tdir/other_dir
touch $DIR/$tdir/migrate_dir/a
ln $DIR/$tdir/migrate_dir/a $DIR/$tdir/other_dir/ln1
ln $DIR/$tdir/migrate_dir/a $DIR/$tdir/other_dir/ln2
ls $DIR/$tdir/other_dir
# a should be migrated to MDT1, since no other links on MDT0
$LFS migrate -m 1 $DIR/$tdir/migrate_dir ||
error "#1 migrate dir fails"
mdt_index=$($LFS getstripe -m $DIR/$tdir/migrate_dir)
[ $mdt_index == 1 ] || error "migrate_dir is not on MDT1"
mdt_index=$($LFS getstripe -m $DIR/$tdir/migrate_dir/a)
[ $mdt_index == 1 ] || error "a is not on MDT1"
# a should stay on MDT1, because it is a mulitple link file
$LFS migrate -m 0 $DIR/$tdir/migrate_dir ||
error "#2 migrate dir fails"
mdt_index=$($LFS getstripe -m $DIR/$tdir/migrate_dir/a)
[ $mdt_index == 1 ] || error "a is not on MDT1"
$LFS migrate -m 1 $DIR/$tdir/migrate_dir ||
error "#3 migrate dir fails"
a_fid=$($LFS path2fid $DIR/$tdir/migrate_dir/a)
ln_fid=$($LFS path2fid $DIR/$tdir/other_dir/ln1)
[ "$a_fid" = "$ln_fid" ] || error "different fid after migrate to MDT1"
rm -rf $DIR/$tdir/other_dir/ln1 || error "unlink ln1 fails"
rm -rf $DIR/$tdir/other_dir/ln2 || error "unlink ln2 fails"
# a should be migrated to MDT0, since no other links on MDT1
$LFS migrate -m 0 $DIR/$tdir/migrate_dir ||
error "#4 migrate dir fails"
mdt_index=$($LFS getstripe -m $DIR/$tdir/migrate_dir/a)
[ $mdt_index == 0 ] || error "a is not on MDT0"
rm -rf $DIR/$tdir || error "rm dir failed after migration"
}
run_test 230f "migrate mulitple remote link files"
test_230g() {
[ $PARALLEL == "yes" ] && skip "skip parallel run"
[ $MDSCOUNT -lt 2 ] && skip_env "needs >= 2 MDTs"
[ $MDS1_VERSION -lt $(version_code 2.11.52) ] &&
skip "Need MDS version at least 2.11.52"
mkdir -p $DIR/$tdir/migrate_dir
$LFS migrate -m 1000 $DIR/$tdir/migrate_dir &&
error "migrating dir to non-exist MDT succeeds"
true
}
run_test 230g "migrate dir to non-exist MDT"
test_230h() {
[ $PARALLEL == "yes" ] && skip "skip parallel run"
[ $MDSCOUNT -lt 2 ] && skip_env "needs >= 2 MDTs"
[ $MDS1_VERSION -lt $(version_code 2.11.52) ] &&
skip "Need MDS version at least 2.11.52"
local mdt_index
mkdir -p $DIR/$tdir/migrate_dir
$LFS migrate -m1 $DIR &&
error "migrating mountpoint1 should fail"
$LFS migrate -m1 $DIR/$tdir/.. &&
error "migrating mountpoint2 should fail"
# same as mv
$LFS migrate -m1 $DIR/$tdir/migrate_dir/.. &&
error "migrating $tdir/migrate_dir/.. should fail"
true
}
run_test 230h "migrate .. and root"
test_230i() {
[ $PARALLEL == "yes" ] && skip "skip parallel run"
[ $MDSCOUNT -lt 2 ] && skip_env "needs >= 2 MDTs"
[ $MDS1_VERSION -lt $(version_code 2.11.52) ] &&
skip "Need MDS version at least 2.11.52"
mkdir -p $DIR/$tdir/migrate_dir
$LFS migrate -m 1 $DIR/$tdir/migrate_dir/ ||
error "migration fails with a tailing slash"
$LFS migrate -m 0 $DIR/$tdir/migrate_dir// ||
error "migration fails with two tailing slashes"
}
run_test 230i "lfs migrate -m tolerates trailing slashes"
test_230j() {
[ $MDSCOUNT -lt 2 ] && skip "needs >= 2 MDTs"
[ $MDS1_VERSION -lt $(version_code 2.13.52) ] &&
skip "Need MDS version at least 2.11.52"
$LFS mkdir -i 0 -c 1 $DIR/$tdir || error "mkdir $tdir failed"
$LFS setstripe -E 1M -L mdt $DIR/$tdir/$tfile ||
error "create $tfile failed"
cat /etc/passwd > $DIR/$tdir/$tfile
$LFS migrate -m 1 $DIR/$tdir || error "migrate failed"
cmp /etc/passwd $DIR/$tdir/$tfile ||
error "DoM file mismatch after migration"
}
run_test 230j "DoM file data not changed after dir migration"
test_230k() {
[ $MDSCOUNT -lt 4 ] && skip "needs >= 4 MDTs"
(( $MDS1_VERSION > $(version_code v2_12_9-38-g80e9ebc1b1e4) )) ||
skip "Need MDS > v2_12_9-38-g80e9ebc1b1e4"
local total=20
local files_on_starting_mdt=0
$LFS mkdir -i -1 -c 2 $DIR/$tdir || error "mkdir failed"
$LFS getdirstripe $DIR/$tdir
for i in $(seq $total); do
echo $((i*i - i)) > $DIR/$tdir/$tfile.$i || error "write failed"
[[ $($LFS getstripe -m $DIR/$tdir/$tfile.$i) -eq 0 ]] &&
files_on_starting_mdt=$((files_on_starting_mdt + 1))
done
echo "$files_on_starting_mdt files on MDT0"
$LFS migrate -m 1,3 $DIR/$tdir || error "migrate -m 1,3 failed"
$LFS getdirstripe $DIR/$tdir
files_on_starting_mdt=0
for i in $(seq $total); do
$(echo $((i*i - i)) | cmp $DIR/$tdir/$tfile.$i -) ||
error "file $tfile.$i mismatch after migration"
[[ $($LFS getstripe -m $DIR/$tdir/$tfile.$i) -eq 1 ]] &&
files_on_starting_mdt=$((files_on_starting_mdt + 1))
done
echo "$files_on_starting_mdt files on MDT1 after migration"
[[ $files_on_starting_mdt -eq $total ]] && error "all files on MDT1"
$LFS migrate -m 0 -c 2 $DIR/$tdir || error "migrate -m 0 -c 2 failed"
$LFS getdirstripe $DIR/$tdir
files_on_starting_mdt=0
for i in $(seq $total); do
$(echo $((i*i - i)) | cmp $DIR/$tdir/$tfile.$i -) ||
error "file $tfile.$i mismatch after 2nd migration"
[[ $($LFS getstripe -m $DIR/$tdir/$tfile.$i) -eq 0 ]] &&
files_on_starting_mdt=$((files_on_starting_mdt + 1))
done
echo "$files_on_starting_mdt files on MDT0 after 2nd migration"
[[ $files_on_starting_mdt -eq $total ]] && error "all files on MDT0"
true
}
run_test 230k "file data not changed after dir migration"
test_230l() {
[ $MDSCOUNT -lt 2 ] && skip "needs >= 2 MDTs"
[ $MDS1_VERSION -lt $(version_code 2.11.56) ] &&
skip "Need MDS version at least 2.11.56"
$LFS mkdir -i 0 -c 1 $DIR/$tdir || error "mkdir failed"
createmany -o $DIR/$tdir/f___________________________________ 1000 ||
error "create files under remote dir failed $i"
$LFS migrate -m 1 $DIR/$tdir || error "migrate failed"
}
run_test 230l "readdir between MDTs won't crash"
test_230m() {
[ $MDSCOUNT -lt 2 ] && skip "needs >= 2 MDTs"
[ $MDS1_VERSION -lt $(version_code 2.11.56) ] &&
skip "Need MDS version at least 2.11.56"
local MDTIDX=1
local mig_dir=$DIR/$tdir/migrate_dir
local longstr="aaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaa"
local shortstr="b"
local val
echo "Creating files and dirs with xattrs"
test_mkdir $DIR/$tdir
test_mkdir -i0 -c1 $mig_dir
mkdir $mig_dir/dir
setfattr -n user.attr1 -v $longstr $mig_dir/dir ||
error "cannot set xattr attr1 on dir"
setfattr -n user.attr2 -v $shortstr $mig_dir/dir ||
error "cannot set xattr attr2 on dir"
touch $mig_dir/dir/f0
setfattr -n user.attr1 -v $longstr $mig_dir/dir/f0 ||
error "cannot set xattr attr1 on file"
setfattr -n user.attr2 -v $shortstr $mig_dir/dir/f0 ||
error "cannot set xattr attr2 on file"
sync ; sync ; echo 3 > /proc/sys/vm/drop_caches
val=$(getfattr --only-values -n user.attr1 $mig_dir/dir 2>/dev/null)
[ "$val" = $longstr ] || error "xattr attr1 not set properly on dir"
val=$(getfattr --only-values -n user.attr2 $mig_dir/dir 2>/dev/null)
[ "$val" = $shortstr ] || error "xattr attr2 not set properly on dir"
val=$(getfattr --only-values -n user.attr1 $mig_dir/dir/f0 2>/dev/null)
[ "$val" = $longstr ] || error "xattr attr1 not set properly on file"
val=$(getfattr --only-values -n user.attr2 $mig_dir/dir/f0 2>/dev/null)
[ "$val" = $shortstr ] || error "xattr attr2 not set properly on file"
echo "Migrating to MDT1"
$LFS migrate -m $MDTIDX $mig_dir ||
error "fails on migrating dir to MDT1"
sync ; sync ; echo 3 > /proc/sys/vm/drop_caches
echo "Checking xattrs"
val=$(getfattr --only-values -n user.attr1 $mig_dir/dir 2>/dev/null)
[ "$val" = $longstr ] ||
error "expecting xattr1 $longstr on dir, found $val"
val=$(getfattr --only-values -n user.attr2 $mig_dir/dir 2>/dev/null)
[ "$val" = $shortstr ] ||
error "expecting xattr2 $shortstr on dir, found $val"
val=$(getfattr --only-values -n user.attr1 $mig_dir/dir/f0 2>/dev/null)
[ "$val" = $longstr ] ||
error "expecting xattr1 $longstr on file, found $val"
val=$(getfattr --only-values -n user.attr2 $mig_dir/dir/f0 2>/dev/null)
[ "$val" = $shortstr ] ||
error "expecting xattr2 $shortstr on file, found $val"
}
run_test 230m "xattrs not changed after dir migration"
test_230n() {
[ $MDSCOUNT -lt 2 ] && skip "needs >= 2 MDTs"
[[ $MDS1_VERSION -ge $(version_code 2.13.53) ]] ||
skip "Need MDS version at least 2.13.53"
$LFS mkdir -i 0 $DIR/$tdir || error "mkdir $tdir failed"
cat /etc/hosts > $DIR/$tdir/$tfile
$LFS mirror extend -N1 $DIR/$tdir/$tfile || error "Mirroring failed"
$LFS migrate -m 1 $DIR/$tdir || error "Migration failed"
cmp /etc/hosts $DIR/$tdir/$tfile ||
error "File data mismatch after migration"
}
run_test 230n "Dir migration with mirrored file"
test_230o() {
[ $MDSCOUNT -ge 2 ] || skip "needs >= 2 MDTs"
[ $MDS1_VERSION -ge $(version_code 2.13.52) ] ||
skip "Need MDS version at least 2.13.52"
local mdts=$(mdts_nodes)
local timeout=100
local restripe_status
local delta
local i
[[ $mds1_FSTYPE == zfs ]] && timeout=300
# in case "crush" hash type is not set
do_nodes $mdts "$LCTL set_param lod.*.mdt_hash=crush"
restripe_status=$(do_facet mds1 $LCTL get_param -n \
mdt.*MDT0000.enable_dir_restripe)
do_nodes $mdts "$LCTL set_param mdt.*.enable_dir_restripe=1"
stack_trap "do_nodes $mdts $LCTL set_param \
mdt.*.enable_dir_restripe=$restripe_status"
mkdir $DIR/$tdir
createmany -m $DIR/$tdir/f 100 ||
error "create files under remote dir failed $i"
createmany -d $DIR/$tdir/d 100 ||
error "create dirs under remote dir failed $i"
for i in $(seq 2 $MDSCOUNT); do
do_nodes $mdts "$LCTL set_param mdt.*.md_stats=clear >/dev/null"
$LFS setdirstripe -c $i $DIR/$tdir ||
error "split -c $i $tdir failed"
wait_update $HOSTNAME \
"$LFS getdirstripe -H $DIR/$tdir" "crush" $timeout ||
error "dir split not finished"
delta=$(do_nodes $mdts "lctl get_param -n mdt.*MDT*.md_stats" |
awk '/migrate/ {sum += $2} END { print sum }')
echo "$delta migrated when dir split $((i - 1)) to $i stripes"
# delta is around total_files/stripe_count
(( $delta < 200 / (i - 1) + 4 )) ||
error "$delta files migrated >= $((200 / (i - 1) + 4))"
done
}
run_test 230o "dir split"
test_230p() {
(( MDSCOUNT > 1 )) || skip "needs >= 2 MDTs"
(( MDS1_VERSION >= $(version_code 2.13.52) )) ||
skip "Need MDS version at least 2.13.52"
local mdts=$(mdts_nodes)
local timeout=100
local restripe_status
local delta
local c
[[ $mds1_FSTYPE == zfs ]] && timeout=300
do_nodes $mdts "$LCTL set_param lod.*.mdt_hash=crush"
restripe_status=$(do_facet mds1 $LCTL get_param -n \
mdt.*MDT0000.enable_dir_restripe)
do_nodes $mdts "$LCTL set_param mdt.*.enable_dir_restripe=1"
stack_trap "do_nodes $mdts $LCTL set_param \
mdt.*.enable_dir_restripe=$restripe_status"
test_mkdir -c $MDSCOUNT -H crush $DIR/$tdir
createmany -m $DIR/$tdir/f 100 ||
error "create files under remote dir failed"
createmany -d $DIR/$tdir/d 100 ||
error "create dirs under remote dir failed"
for ((c = $MDSCOUNT - 1; c >= 1; c--)); do
local mdt_hash="crush"
do_nodes $mdts "$LCTL set_param mdt.*.md_stats=clear >/dev/null"
$LFS setdirstripe -c $c $DIR/$tdir ||
error "split -c $c $tdir failed"
if (( MDS1_VERSION > $(version_code 2.14.0-7-10-g3ae36f4c0a) )); then
mdt_hash="$mdt_hash,fixed"
elif (( $c == 1 )); then
mdt_hash="none"
fi
wait_update $HOSTNAME \
"$LFS getdirstripe -H $DIR/$tdir" $mdt_hash $timeout ||
error "dir merge not finished"
delta=$(do_nodes $mdts "lctl get_param -n mdt.*MDT*.md_stats" |
awk '/migrate/ {sum += $2} END { print sum }')
echo "$delta migrated when dir merge $((c + 1)) to $c stripes"
# delta is around total_files/stripe_count
(( delta < 200 / c + 4 )) ||
error "$delta files migrated >= $((200 / c + 4))"
done
(( MDS1_VERSION >= $(version_code v2_16_52-71-g103c1f560c) )) ||
skip "need MDS >= 2.16.52.71 for resplitting support"
# and restripe to -c 2
echo Splitting the dir back to 2 stripes.
$LFS setdirstripe -c 2 $DIR/$tdir || {
$LFS getdirstripe $DIR/$tdir
error "split $tdir to 2 stripes failed"
}
wait_update $HOSTNAME \
"$LFS getdirstripe -H $DIR/$tdir" "crush" $timeout ||
error "dir split not finished"
}
run_test 230p "dir merge"
test_230q() {
(( MDSCOUNT > 1)) || skip "needs >= 2 MDTs"
(( MDS1_VERSION >= $(version_code 2.13.52) )) ||
skip "Need MDS version at least 2.13.52"
local mdts=$(mdts_nodes)
local saved_threshold=$(do_facet mds1 \
$LCTL get_param -n mdt.*-MDT0000.dir_split_count)
local saved_delta=$(do_facet mds1 \
$LCTL get_param -n mdt.*-MDT0000.dir_split_delta)
local threshold=100
local delta=2
local total=0
local stripe_count=0
local stripe_index
local nr_files
local create
# test with fewer files on ZFS
[ "$mds1_FSTYPE" == "zfs" ] && threshold=40
stack_trap "do_nodes $mdts $LCTL set_param \
mdt.*.dir_split_count=$saved_threshold"
stack_trap "do_nodes $mdts $LCTL set_param \
mdt.*.dir_split_delta=$saved_delta"
stack_trap "do_nodes $mdts $LCTL set_param mdt.*.dir_restripe_nsonly=1"
do_nodes $mdts "$LCTL set_param mdt.*.enable_dir_auto_split=1"
do_nodes $mdts "$LCTL set_param mdt.*.dir_split_count=$threshold"
do_nodes $mdts "$LCTL set_param mdt.*.dir_split_delta=$delta"
do_nodes $mdts "$LCTL set_param mdt.*.dir_restripe_nsonly=0"
do_nodes $mdts "$LCTL set_param lod.*.mdt_hash=crush"
$LFS mkdir -i -1 -c 1 $DIR/$tdir || error "mkdir $tdir failed"
stripe_index=$($LFS getdirstripe -i $DIR/$tdir)
create=$((threshold * 3 / 2))
while [ $stripe_count -lt $MDSCOUNT ]; do
createmany -m $DIR/$tdir/f $total $create ||
error "create sub files failed"
stat $DIR/$tdir > /dev/null
total=$((total + create))
stripe_count=$((stripe_count + delta))
[ $stripe_count -gt $MDSCOUNT ] && stripe_count=$MDSCOUNT
wait_update $HOSTNAME \
"$LFS getdirstripe -c $DIR/$tdir" "$stripe_count" 40 ||
error "stripe count $($LFS getdirstripe -c $DIR/$tdir) != $stripe_count"
wait_update $HOSTNAME \
"$LFS getdirstripe -H $DIR/$tdir" "crush" 200 ||
error "stripe hash $($LFS getdirstripe -H $DIR/$tdir) != crush"
nr_files=$($LFS find -m 1 $DIR/$tdir | grep -c -w $stripe_index)
echo "$nr_files/$total files on MDT$stripe_index after split"
# allow 10% margin of imbalance with crush hash
(( $nr_files <= $total / $stripe_count + $create / 10)) ||
error "$nr_files files on MDT$stripe_index after split"
nr_files=$($LFS find -type f $DIR/$tdir | wc -l)
[ $nr_files -eq $total ] ||
error "total sub files $nr_files != $total"
done
(( MDS1_VERSION >= $(version_code 2.14.51) )) || return 0
echo "fixed layout directory won't auto split"
$LFS migrate -m 0 $DIR/$tdir || error "migrate $tdir failed"
wait_update $HOSTNAME "$LFS getdirstripe -H $DIR/$tdir" "crush,fixed" \
10 || error "stripe hash $($LFS getdirstripe -H $DIR/$tdir)"
wait_update $HOSTNAME "$LFS getdirstripe -c $DIR/$tdir" 1 10 ||
error "stripe count $($LFS getdirstripe -c $DIR/$tdir)"
}
run_test 230q "dir auto split"
test_230r() {
[[ $PARALLEL != "yes" ]] || skip "skip parallel run"
[[ $MDSCOUNT -ge 2 ]] || skip_env "needs >= 2 MDTs"
[[ $MDS1_VERSION -ge $(version_code 2.13.54) ]] ||
skip "Need MDS version at least 2.13.54"
# maximum amount of local locks:
# parent striped dir - 2 locks
# new stripe in parent to migrate to - 1 lock
# source and target - 2 locks
# Total 5 locks for regular file
mkdir -p $DIR/$tdir
$LFS mkdir -i1 -c2 $DIR/$tdir/dir1
touch $DIR/$tdir/dir1/eee
# create 4 hardlink for 4 more locks
# Total: 9 locks > RS_MAX_LOCKS (8)
$LFS mkdir -i1 -c1 $DIR/$tdir/dir2
$LFS mkdir -i1 -c1 $DIR/$tdir/dir3
$LFS mkdir -i1 -c1 $DIR/$tdir/dir4
$LFS mkdir -i1 -c1 $DIR/$tdir/dir5
ln $DIR/$tdir/dir1/eee $DIR/$tdir/dir2/eee
ln $DIR/$tdir/dir1/eee $DIR/$tdir/dir3/eee
ln $DIR/$tdir/dir1/eee $DIR/$tdir/dir4/eee
ln $DIR/$tdir/dir1/eee $DIR/$tdir/dir5/eee
cancel_lru_locks mdc
$LFS migrate -m1 -c1 $DIR/$tdir/dir1 ||
error "migrate dir fails"
rm -rf $DIR/$tdir || error "rm dir failed after migration"
}
run_test 230r "migrate with too many local locks"
test_230s() {
(( $MDS1_VERSION >= $(version_code 2.14.52) )) ||
skip "Need MDS version at least 2.14.52"
local mdts=$(mdts_nodes)
local restripe_status=$(do_facet mds1 $LCTL get_param -n \
mdt.*MDT0000.enable_dir_restripe)
stack_trap "do_nodes $mdts $LCTL set_param \
mdt.*.enable_dir_restripe=$restripe_status"
local st
for st in 0 1; do
do_nodes $mdts "$LCTL set_param mdt.*.enable_dir_restripe=$st"
test_mkdir $DIR/$tdir
$LFS mkdir $DIR/$tdir |& grep "File exists" ||
error "$LFS mkdir should return EEXIST if target exists"
rmdir $DIR/$tdir
done
}
run_test 230s "lfs mkdir should return -EEXIST if target exists"
test_230t()
{
(( $MDSCOUNT >= 2 )) || skip_env "needs >= 2 MDTs"
(( $MDS1_VERSION >= $(version_code 2.14.50) )) ||
skip "Need MDS version at least 2.14.50"
test_mkdir $DIR/$tdir || error "mkdir $tdir failed"
test_mkdir $DIR/$tdir/subdir || error "mkdir subdir failed"
$LFS project -p 1 -s $DIR/$tdir ||
error "set $tdir project id failed"
$LFS project -p 2 -s $DIR/$tdir/subdir ||
error "set subdir project id failed"
local pbefore="$($LFS project -d $DIR/$tdir)"
local sbefore="$($LFS project -d $DIR/$tdir/subdir)"
$LFS migrate -m 1 -c $MDSCOUNT $DIR/$tdir || error "migrate failed"
local pafter="$($LFS project -d $DIR/$tdir)"
local safter="$($LFS project -d $DIR/$tdir/subdir)"
[[ "$pbefore" == "$pafter" ]] || error "projid '$pbefore' != '$pafter'"
[[ "$sbefore" == "$safter" ]] || error "projid '$sbefore' != '$safter'"
(( $MDS1_VERSION >= $(version_code 2.15.59.107) )) ||
{ echo "Need MDS >= 2.15.59.107 for projid rename"; return 0; }
# check rename works, even if source parent projid differs (LU-17016)
test_mkdir $DIR/$tdir.2 || error "mkdir $tdir.2 failed"
local fid_before=$($LFS path2fid $DIR/$tdir/subdir)
$LFS project -p 2 -s $DIR/$tdir.2 || error "set $tdir.2 projid failed"
mrename $DIR/$tdir/subdir $DIR/$tdir.2/subdir ||
error "subdir failed rename for different source parent projid"
local fid_after=$($LFS path2fid $DIR/$tdir.2/subdir)
[[ "$fid_before" == "$fid_after" ]] ||
error "fid before '$fid_before' != after '$fid_after'"
}
run_test 230t "migrate directory with project ID set"
test_230u()
{
(( MDSCOUNT > 3 )) || skip_env "needs >= 4 MDTs"
(( MDS1_VERSION >= $(version_code 2.14.53) )) ||
skip "Need MDS version at least 2.14.53"
local count
mkdir_on_mdt0 $DIR/$tdir || error "mkdir $tdir failed"
mkdir $DIR/$tdir/sub{0..99} || error "mkdir sub failed"
$LFS migrate -m -1 $DIR/$tdir/sub{0..99} || error "migrate sub failed"
for i in $(seq 0 $((MDSCOUNT - 1))); do
count=$($LFS getstripe -m $DIR/$tdir/sub* | grep -c ^$i)
echo "$count dirs migrated to MDT$i"
done
count=$($LFS getstripe -m $DIR/$tdir/sub* | sort -u | wc -l)
(( count >= MDSCOUNT - 1 )) || error "dirs migrated to $count MDTs"
}
run_test 230u "migrate directory by QOS"
test_230v()
{
(( MDSCOUNT > 3 )) || skip_env "needs >= 4 MDTs"
(( MDS1_VERSION >= $(version_code 2.14.53) )) ||
skip "Need MDS version at least 2.14.53"
local count
mkdir $DIR/$tdir || error "mkdir $tdir failed"
mkdir $DIR/$tdir/sub{0..99} || error "mkdir sub failed"
$LFS migrate -m 0,2,1 $DIR/$tdir || error "migrate $tdir failed"
for i in $(seq 0 $((MDSCOUNT - 1))); do
count=$($LFS getstripe -m $DIR/$tdir/sub* | grep -c ^$i)
echo "$count subdirs migrated to MDT$i"
(( i == 3 )) && (( count > 0 )) &&
error "subdir shouldn't be migrated to MDT3"
done
count=$($LFS getstripe -m $DIR/$tdir/sub* | sort -u | wc -l)
(( count == 3 )) || error "dirs migrated to $count MDTs"
}
run_test 230v "subdir migrated to the MDT where its parent is located"
test_230w() {
(( MDSCOUNT > 1 )) || skip "needs >= 2 MDTs"
(( MDS1_VERSION >= $(version_code 2.15.0) )) ||
skip "Need MDS version at least 2.15.0"
mkdir -p $DIR/$tdir || error "mkdir $tdir failed"
createmany -o $DIR/$tdir/f 10 || error "create files failed"
createmany -d $DIR/$tdir/d 10 || error "create dirs failed"
$LFS migrate -m 1 -c $MDSCOUNT -d $DIR/$tdir ||
error "migrate failed"
(( $($LFS getdirstripe -c $DIR/$tdir) == MDSCOUNT )) ||
error "$tdir stripe count mismatch"
for i in $(seq 0 9); do
(( $($LFS getdirstripe -c $DIR/$tdir/d$i) == 0 )) ||
error "d$i is striped"
done
}
run_test 230w "non-recursive mode dir migration"
test_230x() {
(( MDSCOUNT > 1 )) || skip "needs >= 2 MDTs"
(( MDS1_VERSION >= $(version_code v2_15_52-83-g6aee406c84) )) ||
skip "Need MDS >= 2.15.53 for migrate space check"
mkdir -p $DIR/$tdir || error "mkdir failed"
createmany -d $DIR/$tdir/sub 100 || error "createmany failed"
local mdt_name=$(mdtname_from_index 0)
local low=$(do_facet mds2 $LCTL get_param -n \
osp.*$mdt_name-osp-MDT0001.reserved_ino_low)
local high=$(do_facet mds2 $LCTL get_param -n \
osp.*$mdt_name-osp-MDT0001.reserved_ino_high)
local ffree=$($LFS df -i $MOUNT | awk "/$mdt_name/ { print \$4 }")
local maxage=$(do_facet mds2 $LCTL get_param -n \
osp.*$mdt_name-osp-MDT0001.maxage)
stack_trap "do_facet mds2 $LCTL set_param -n \
osp.*$mdt_name-osp-MDT0001.reserved_ino_low=$low \
osp.*$mdt_name-osp-MDT0001.reserved_ino_high=$high" EXIT
stack_trap "do_facet mds2 $LCTL set_param -n \
osp.*$mdt_name-osp-MDT0001.maxage=$maxage" EXIT
do_facet mds2 $LCTL set_param -n \
osp.*$mdt_name-osp-MDT0001.reserved_ino_low=$((ffree + 1))
do_facet mds2 $LCTL set_param -n osp.*$mdt_name-osp-MDT0001.maxage=1
sleep 4
$LFS migrate -m 1 -c $MDSCOUNT $DIR/$tdir &&
error "migrate $tdir should fail"
do_facet mds2 $LCTL set_param -n \
osp.*$mdt_name-osp-MDT0001.reserved_ino_low=$low
do_facet mds2 $LCTL set_param -n \
osp.*$mdt_name-osp-MDT0001.reserved_ino_high=$high
sleep 4
$LFS migrate -m 1 -c $MDSCOUNT $DIR/$tdir ||
error "migrate failed"
(( $($LFS getdirstripe -c $DIR/$tdir) == $MDSCOUNT )) ||
error "$tdir stripe count mismatch"
}
run_test 230x "dir migration check space"
test_230y() {
(( MDSCOUNT > 1 )) || skip "needs >= 2 MDTs"
(( MDS1_VERSION >= $(version_code 2.15.55.45) )) ||
skip "Need MDS version at least 2.15.55.45"
local pid
test_mkdir -c -1 $DIR/$tdir || error "mkdir $tdir failed"
$LFS getdirstripe $DIR/$tdir
createmany -d $DIR/$tdir/d 100 || error "createmany failed"
$LFS migrate -m 1 -c 2 $DIR/$tdir &
pid=$!
sleep 1
#OBD_FAIL_MIGRATE_BAD_HASH 0x1802
do_facet mds2 lctl set_param fail_loc=0x1802
wait $pid
do_facet mds2 lctl set_param fail_loc=0
$LFS getdirstripe $DIR/$tdir
unlinkmany -d $DIR/$tdir/d 100 || error "unlinkmany failed"
rmdir $DIR/$tdir || error "rmdir $tdir failed"
}
run_test 230y "unlink dir with bad hash type"
test_230z() {
(( MDSCOUNT > 1 )) || skip "needs >= 2 MDTs"
(( MDS1_VERSION >= $(version_code 2.15.55.45) )) ||
skip "Need MDS version at least 2.15.55.45"
local pid
test_mkdir -c -1 $DIR/$tdir || error "mkdir $tdir failed"
$LFS getdirstripe $DIR/$tdir
createmany -d $DIR/$tdir/d 100 || error "createmany failed"
$LFS migrate -m 1 -c 2 -H fnv_1a_64 $DIR/$tdir &
pid=$!
sleep 1
#OBD_FAIL_MIGRATE_BAD_HASH 0x1802
do_facet mds2 lctl set_param fail_loc=0x1802
wait $pid
do_facet mds2 lctl set_param fail_loc=0
$LFS getdirstripe $DIR/$tdir
# resume migration
$LFS migrate -m 1 -c 2 -H fnv_1a_64 $DIR/$tdir ||
error "resume migration failed"
$LFS getdirstripe $DIR/$tdir
[ $($LFS getdirstripe -H $DIR/$tdir) == "fnv_1a_64,fixed" ] ||
error "migration is not finished"
}
run_test 230z "resume dir migration with bad hash type"
test_230A_check_lmm_oi() {
local FID LMM_FID
FID=$($LFS path2fid "$1")
while read LMM_FID; do
[[ "$FID" == "$LMM_FID" ]] || {
echo "FID=$FID LMM_FID=$LMM_FID"
return 1
}
done < <( $LFS getstripe -v "$1" | awk '/lmm_fid:/ {print $2 }' )
return 0
}
test_230A()
{
(( MDSCOUNT > 1 )) || skip "needs >= 2 MDTs"
(( MDS1_VERSION >= $(version_code 2.16.57) )) ||
skip "need MDS >= 2.16.57 for lmm_oi migrate update"
$LFS mkdir -i 0 -c 1 $DIR/$tdir || error "mkdir $tdir failed"
$LFS setstripe -E 10M -c 1 -E -1 -c 1 $DIR/$tdir/file || error "file creation failed"
test_230A_check_lmm_oi $DIR/$tdir/file || error "Wrong lmm oi before migrate"
$LFS migrate -m 1 $DIR/$tdir || error "migrate failed"
test_230A_check_lmm_oi $DIR/$tdir/file || error "Wrong lmm oi after migrate"
}
run_test 230A "dir migrate should update lmm_oi"
test_230B() {
(( MDSCOUNT > 1 )) || skip "needs >= 2 MDTs"
(( MDS1_VERSION >= $(version_code 2.17.51) )) ||
skip "need MDS >= 2.17.51 for lmm_oi migrate update"
test_mkdir -i 0 -c 1 $DIR/$tdir
echo "create files"
createmany -d $DIR/$tdir/d 10
(
cd $DIR/$tdir
sleep 10
echo create the second "foo" file
mkdir foo
) &
do_facet mds2 lctl set_param fail_loc=0x1801
echo "attempt to migrate"
$LFS migrate -m 1 $DIR/$tdir
echo create the first "foo" file
touch $DIR/$tdir/foo
echo wait ...
wait
num=$(ls -l $DIR/$tdir | grep -c foo)
do_facet mds2 lctl set_param fail_loc=0
$LFS getdirstripe $DIR/$tdir
$LFS path2fid $DIR/$tdir
(( num == 1 )) || error "duplicated files creation during migration"
}
run_test 230B "create duplicated entries in a migrating dir"
test_231a()
{
# For simplicity this test assumes that max_pages_per_rpc
# is the same across all OSCs
local max_pages=$($LCTL get_param -n osc.*.max_pages_per_rpc | head -n1)
local bulk_size=$((max_pages * PAGE_SIZE))
local brw_size=$(do_facet ost1 $LCTL get_param -n obdfilter.*.brw_size |
head -n 1)
mkdir -p $DIR/$tdir
$LFS setstripe -S ${brw_size}M $DIR/$tdir ||
error "failed to set stripe with -S ${brw_size}M option"
stack_trap "rm -rf $DIR/$tdir"
# clear the OSC stats
$LCTL set_param osc.*.stats=0 &>/dev/null
stop_writeback
# Client writes $bulk_size - there must be 1 rpc for $max_pages.
dd if=/dev/zero of=$DIR/$tdir/$tfile bs=$bulk_size count=1 \
oflag=direct &>/dev/null || error "dd failed"
sync; sleep 1; sync # just to be safe
local nrpcs=$($LCTL get_param osc.*.stats |awk '/ost_write/ {print $2}')
if [ x$nrpcs != "x1" ]; then
$LCTL get_param osc.*.stats
error "found $nrpcs ost_write RPCs, not 1 as expected"
fi
start_writeback
# Drop the OSC cache, otherwise we will read from it
cancel_lru_locks osc
# clear the OSC stats
$LCTL set_param osc.*.stats=0 &>/dev/null
# Client reads $bulk_size.
dd if=$DIR/$tdir/$tfile of=/dev/null bs=$bulk_size count=1 \
iflag=direct &>/dev/null || error "dd failed"
nrpcs=$($LCTL get_param osc.*.stats | awk '/ost_read/ { print $2 }')
if [ x$nrpcs != "x1" ]; then
$LCTL get_param osc.*.stats
error "found $nrpcs ost_read RPCs, not 1 as expected"
fi
}
run_test 231a "checking that reading/writing of BRW RPC size results in one RPC"
test_231b() {
mkdir -p $DIR/$tdir
stack_trap "rm -rf $DIR/$tdir"
local i
for i in {0..1023}; do
dd if=/dev/zero of=$DIR/$tdir/$tfile conv=notrunc \
seek=$((2 * i)) bs=4096 count=1 &>/dev/null ||
error "dd of=$DIR/$tdir/$tfile seek=$((2 * i)) failed"
done
sync
}
run_test 231b "must not assert on fully utilized OST request buffer"
test_232a() {
mkdir -p $DIR/$tdir
$LFS setstripe -c1 -i0 $DIR/$tdir/$tfile
#define OBD_FAIL_LDLM_OST_LVB 0x31c
do_facet ost1 $LCTL set_param fail_loc=0x31c
# ignore dd failure
dd if=/dev/zero of=$DIR/$tdir/$tfile bs=1M count=1 || true
stack_trap "rm -f $DIR/$tdir/$tfile"
do_facet ost1 $LCTL set_param fail_loc=0
umount_client $MOUNT || error "umount failed"
mount_client $MOUNT || error "mount failed"
stop ost1 || error "cannot stop ost1"
start ost1 $(ostdevname 1) $OST_MOUNT_OPTS || error "cannot start ost1"
}
run_test 232a "failed lock should not block umount"
test_232b() {
[ $MDS1_VERSION -ge $(version_code 2.10.58) ] ||
skip "Need MDS version at least 2.10.58"
mkdir -p $DIR/$tdir
$LFS setstripe -c1 -i0 $DIR/$tdir/$tfile
dd if=/dev/zero of=$DIR/$tdir/$tfile bs=1M count=1
stack_trap "rm -f $DIR/$tdir/$tfile"
sync
cancel_lru_locks osc
#define OBD_FAIL_LDLM_OST_LVB 0x31c
do_facet ost1 $LCTL set_param fail_loc=0x31c
# ignore failure
$LFS data_version $DIR/$tdir/$tfile || true
do_facet ost1 $LCTL set_param fail_loc=0
umount_client $MOUNT || error "umount failed"
mount_client $MOUNT || error "mount failed"
stop ost1 || error "cannot stop ost1"
start ost1 $(ostdevname 1) $OST_MOUNT_OPTS || error "cannot start ost1"
}
run_test 232b "failed data version lock should not block umount"
test_233a() {
[ $MDS1_VERSION -ge $(version_code 2.3.64) ] ||
skip "Need MDS version at least 2.3.64"
[ -n "$FILESET" ] && skip_env "SKIP due to FILESET set"
local fid=$($LFS path2fid $MOUNT)
stat $MOUNT/.lustre/fid/$fid > /dev/null ||
error "cannot access $MOUNT using its FID '$fid'"
}
run_test 233a "checking that OBF of the FS root succeeds"
test_233b() {
[ $MDS1_VERSION -ge $(version_code 2.5.90) ] ||
skip "Need MDS version at least 2.5.90"
[ -n "$FILESET" ] && skip_env "SKIP due to FILESET set"
local fid=$($LFS path2fid $MOUNT/.lustre)
stat $MOUNT/.lustre/fid/$fid > /dev/null ||
error "cannot access $MOUNT/.lustre using its FID '$fid'"
fid=$($LFS path2fid $MOUNT/.lustre/fid)
stat $MOUNT/.lustre/fid/$fid > /dev/null ||
error "cannot access $MOUNT/.lustre/fid using its FID '$fid'"
}
run_test 233b "checking that OBF of the FS .lustre succeeds"
test_234() {
local p="$TMP/sanityN-$TESTNAME.parameters"
save_lustre_params client "llite.*.xattr_cache" > $p
lctl set_param llite.*.xattr_cache 1 ||
skip_env "xattr cache is not supported"
mkdir -p $DIR/$tdir || error "mkdir failed"
touch $DIR/$tdir/$tfile || error "touch failed"
# OBD_FAIL_LLITE_XATTR_ENOMEM
$LCTL set_param fail_loc=0x1405
getfattr -n user.attr $DIR/$tdir/$tfile &&
error "getfattr should have failed with ENOMEM"
$LCTL set_param fail_loc=0x0
rm -rf $DIR/$tdir
restore_lustre_params < $p
rm -f $p
}
run_test 234 "xattr cache should not crash on ENOMEM"
test_235() {
[ $MDS1_VERSION -lt $(version_code 2.4.52) ] &&
skip "Need MDS version at least 2.4.52"
flock_deadlock $DIR/$tfile
local RC=$?
case $RC in
0)
;;
124) error "process hangs on a deadlock"
;;
*) error "error executing flock_deadlock $DIR/$tfile"
;;
esac
}
run_test 235 "LU-1715: flock deadlock detection does not work properly"
#LU-2935
test_236() {
check_swap_layouts_support
local ref1=/etc/passwd
local ref2=/etc/group
local file1=$DIR/$tdir/f1
local file2=$DIR/$tdir/f2
test_mkdir -c1 $DIR/$tdir
$LFS setstripe -c 1 $file1 || error "cannot setstripe on '$file1': rc = $?"
cp $ref1 $file1 || error "cp $ref1 $file1 failed: rc = $?"
$LFS setstripe -c 2 $file2 || error "cannot setstripe on '$file2': rc = $?"
cp $ref2 $file2 || error "cp $ref2 $file2 failed: rc = $?"
local fd=$(free_fd)
local cmd="exec $fd<>$file2"
eval $cmd
rm $file2
$LFS swap_layouts $file1 /proc/self/fd/${fd} ||
error "cannot swap layouts of '$file1' and /proc/self/fd/${fd}"
cmd="exec $fd>&-"
eval $cmd
cmp $ref2 $file1 || error "content compare failed ($ref2 != $file1)"
#cleanup
rm -rf $DIR/$tdir
}
run_test 236 "Layout swap on open unlinked file"
# LU-4659 linkea consistency
test_238() {
[[ $MDS1_VERSION -gt $(version_code 2.5.57) ]] ||
[[ $MDS1_VERSION -gt $(version_code 2.5.1) &&
$MDS1_VERSION -lt $(version_code 2.5.50) ]] ||
skip "Need MDS version at least 2.5.58 or 2.5.2+"
touch $DIR/$tfile
ln $DIR/$tfile $DIR/$tfile.lnk
touch $DIR/$tfile.new
mv $DIR/$tfile.new $DIR/$tfile
local fid1=$($LFS path2fid $DIR/$tfile)
local fid2=$($LFS path2fid $DIR/$tfile.lnk)
local path1=$($LFS fid2path $FSNAME "$fid1")
[ $tfile == $path1 ] || error "linkea inconsistent: $tfile $fid1 $path1"
local path2=$($LFS fid2path $FSNAME "$fid2")
[ $tfile.lnk == $path2 ] ||
error "linkea inconsistent: $tfile.lnk $fid2 $path2!"
rm -f $DIR/$tfile*
}
run_test 238 "Verify linkea consistency"
test_239A() { # was test_239
[ $MDS1_VERSION -lt $(version_code 2.5.60) ] &&
skip "Need MDS version at least 2.5.60"
local mdts=$(mdts_nodes)
mkdir -p $DIR/$tdir
createmany -o $DIR/$tdir/f- 5000
unlinkmany $DIR/$tdir/f- 5000
[ $MDS1_VERSION -gt $(version_code 2.10.4) ] &&
do_nodes $mdts "lctl set_param -n osp.*.force_sync=1"
changes=$(do_nodes $mdts "lctl get_param -n osp.*MDT*.sync_changes \
osp.*MDT*.sync_in_flight" | calc_sum)
[ "$changes" -eq 0 ] || error "$changes not synced"
}
run_test 239A "osp_sync test"
test_239a() { #LU-5297
remote_mds_nodsh && skip "remote MDS with nodsh"
touch $DIR/$tfile
#define OBD_FAIL_OSP_CHECK_INVALID_REC 0x2100
do_facet $SINGLEMDS $LCTL set_param fail_loc=0x2100
chgrp $RUNAS_GID $DIR/$tfile
wait_delete_completed
}
run_test 239a "process invalid osp sync record correctly"
test_239b() { #LU-5297
remote_mds_nodsh && skip "remote MDS with nodsh"
touch $DIR/$tfile1
#define OBD_FAIL_OSP_CHECK_ENOMEM 0x2101
do_facet $SINGLEMDS $LCTL set_param fail_loc=0x2101
chgrp $RUNAS_GID $DIR/$tfile1
wait_delete_completed
do_facet $SINGLEMDS $LCTL set_param fail_loc=0
touch $DIR/$tfile2
chgrp $RUNAS_GID $DIR/$tfile2
wait_delete_completed
}
run_test 239b "process osp sync record with ENOMEM error correctly"
test_240() {
[ $MDSCOUNT -lt 2 ] && skip_env "needs >= 2 MDTs"
remote_mds_nodsh && skip "remote MDS with nodsh"
mkdir -p $DIR/$tdir
$LFS mkdir -i 0 $DIR/$tdir/d0 ||
error "failed to mkdir $DIR/$tdir/d0 on MDT0"
$LFS mkdir -i 1 $DIR/$tdir/d0/d1 ||
error "failed to mkdir $DIR/$tdir/d0/d1 on MDT1"
umount_client $MOUNT || error "umount failed"
#define OBD_FAIL_TGT_DELAY_CONDITIONAL 0x713
do_facet mds2 lctl set_param fail_loc=0x713 fail_val=1
mount_client $MOUNT || error "failed to mount client"
echo "stat $DIR/$tdir/d0/d1, should not fail/ASSERT"
stat $DIR/$tdir/d0/d1 || error "fail to stat $DIR/$tdir/d0/d1"
}
run_test 240 "race between ldlm enqueue and the connection RPC (no ASSERT)"
test_241_bio() {
local count=$1
local bsize=$2
for LOOP in $(seq $count); do
dd if=$DIR/$tfile of=/dev/null bs=$bsize count=1 2>/dev/null
cancel_lru_locks $OSC || true
done
}
test_241_dio() {
local count=$1
local bsize=$2
for LOOP in $(seq $1); do
dd if=$DIR/$tfile of=/dev/null bs=$bsize count=1 iflag=direct \
2>/dev/null
done
}
test_241a() { # was test_241
local bsize=$PAGE_SIZE
(( bsize < 40960 )) && bsize=40960
dd if=/dev/zero of=$DIR/$tfile count=1 bs=$bsize
ls -la $DIR/$tfile
cancel_lru_locks $OSC
test_241_bio 1000 $bsize &
PID=$!
test_241_dio 1000 $bsize
wait $PID
}
run_test 241a "bio vs dio"
test_241b() {
local bsize=$PAGE_SIZE
(( bsize < 40960 )) && bsize=40960
dd if=/dev/zero of=$DIR/$tfile count=1 bs=$bsize
ls -la $DIR/$tfile
test_241_dio 1000 $bsize &
PID=$!
test_241_dio 1000 $bsize
wait $PID
}
run_test 241b "dio vs dio"
test_242() {
remote_mds_nodsh && skip "remote MDS with nodsh"
mkdir_on_mdt0 $DIR/$tdir
touch $DIR/$tdir/$tfile
#define OBD_FAIL_MDS_READPAGE_PACK 0x105
do_facet mds1 lctl set_param fail_loc=0x105
/bin/ls $DIR/$tdir && error "ls $DIR/$tdir should fail"
do_facet mds1 lctl set_param fail_loc=0
/bin/ls $DIR/$tdir || error "ls $DIR/$tdir failed"
}
run_test 242 "mdt_readpage failure should not cause directory unreadable"
test_243()
{
test_mkdir $DIR/$tdir
group_lock_test -d $DIR/$tdir || error "A group lock test failed"
}
run_test 243 "various group lock tests"
test_244a()
{
test_mkdir $DIR/$tdir
dd if=/dev/zero of=$DIR/$tdir/$tfile bs=1M count=35
sendfile_grouplock $DIR/$tdir/$tfile || \
error "sendfile+grouplock failed"
rm -rf $DIR/$tdir
}
run_test 244a "sendfile with group lock tests"
test_244b()
{
[ $PARALLEL == "yes" ] && skip "skip parallel run" && return
local threads=50
local size=$((1024*1024))
test_mkdir $DIR/$tdir
for i in $(seq 1 $threads); do
local file=$DIR/$tdir/file_$((i / 10))
$MULTIOP $file OG1234w$size_$((i % 3))w$size_$((i % 4))g1234c &
local pids[$i]=$!
done
for i in $(seq 1 $threads); do
wait ${pids[$i]}
done
}
run_test 244b "multi-threaded write with group lock"
test_245a() {
local flagname="multi_mod_rpcs"
local connect_data_name="max_mod_rpcs"
local out
# check if multiple modify RPCs flag is set
out=$($LCTL get_param mdc.$FSNAME-MDT0000-*.import |
grep "connect_flags:")
echo "$out"
echo "$out" | grep -qw $flagname
if [ $? -ne 0 ]; then
echo "connect flag $flagname is not set"
return
fi
# check if multiple modify RPCs data is set
out=$($LCTL get_param mdc.$FSNAME-MDT0000-*.import)
echo "$out"
echo "$out" | grep -qw $connect_data_name ||
error "import should have connect data $connect_data_name"
}
run_test 245a "check mdc connection flag/data: multiple modify RPCs"
test_245b() {
local flagname="multi_mod_rpcs"
local connect_data_name="max_mod_rpcs"
local out
remote_mds_nodsh && skip "remote MDS with nodsh"
[[ $MDSCOUNT -ge 2 ]] || skip "needs >= 2 MDTs"
# check if multiple modify RPCs flag is set
out=$(do_facet mds1 \
$LCTL get_param osp.$FSNAME-MDT0001-osp-MDT0000.import |
grep "connect_flags:")
echo "$out"
[[ "$out" =~ $flagname ]] || skip "connect flag $flagname is not set"
# check if multiple modify RPCs data is set
out=$(do_facet mds1 \
$LCTL get_param osp.$FSNAME-MDT0001-osp-MDT0000.import)
[[ "$out" =~ $connect_data_name ]] ||
{
echo "$out"
error "missing connect data $connect_data_name"
}
}
run_test 245b "check osp connection flag/data: multiple modify RPCs"
cleanup_247() {
local submount=$1
trap 0
umount_client $submount
rmdir $submount
}
test_247a() {
lctl get_param -n mdc.$FSNAME-MDT0000*.import |
grep -q subtree ||
skip_env "Fileset feature is not supported"
local submount=${MOUNT}_$tdir
mkdir $MOUNT/$tdir
mkdir -p $submount || error "mkdir $submount failed"
FILESET="$FILESET/$tdir" mount_client $submount ||
error "mount $submount failed"
trap "cleanup_247 $submount" EXIT
echo foo > $submount/$tfile || error "write $submount/$tfile failed"
[ $(cat $MOUNT/$tdir/$tfile) = "foo" ] ||
error "read $MOUNT/$tdir/$tfile failed"
cleanup_247 $submount
}
run_test 247a "mount subdir as fileset"
test_247b() {
lctl get_param -n mdc.$FSNAME-MDT0000*.import | grep -q subtree ||
skip_env "Fileset feature is not supported"
local submount=${MOUNT}_$tdir
rm -rf $MOUNT/$tdir
mkdir -p $submount || error "mkdir $submount failed"
FILESET="$FILESET/$tdir" mount_client $submount &&
error "mount $submount should fail"
rmdir $submount
}
run_test 247b "mount subdir that dose not exist"
test_247c() {
lctl get_param -n mdc.$FSNAME-MDT0000*.import | grep -q subtree ||
skip_env "Fileset feature is not supported"
local submount=${MOUNT}_$tdir
mkdir -p $MOUNT/$tdir/dir1
mkdir -p $submount || error "mkdir $submount failed"
trap "cleanup_247 $submount" EXIT
FILESET="$FILESET/$tdir" mount_client $submount ||
error "mount $submount failed"
local fid=$($LFS path2fid $MOUNT/)
$LFS fid2path $submount $fid && error "fid2path should fail"
cleanup_247 $submount
}
run_test 247c "running fid2path outside subdirectory root"
test_247d() {
lctl get_param -n mdc.$FSNAME-MDT0000*.import | grep -q subtree ||
skip "Fileset feature is not supported"
local submount=${MOUNT}_$tdir
mkdir -p $MOUNT/$tdir/dir1
mkdir -p $submount || error "mkdir $submount failed"
FILESET="$FILESET/$tdir" mount_client $submount ||
error "mount $submount failed"
trap "cleanup_247 $submount" EXIT
local td=$submount/dir1
local fid=$($LFS path2fid $td)
[ -z "$fid" ] && error "path2fid unable to get $td FID"
# check that we get the same pathname back
local rootpath
local found
for rootpath in "$submount" "$submount///" "$submount/dir1"; do
echo "$rootpath $fid"
found=$($LFS fid2path $rootpath "$fid")
[ -n "$found" ] || error "fid2path should succeed"
[ "$found" == "$td" ] || error "fid2path $found != $td"
done
# check wrong root path format
rootpath=$submount"_wrong"
found=$($LFS fid2path $rootpath "$fid")
[ -z "$found" ] || error "fid2path should fail ($rootpath != $submount)"
cleanup_247 $submount
}
run_test 247d "running fid2path inside subdirectory root"
# LU-8037
test_247e() {
lctl get_param -n mdc.$FSNAME-MDT0000*.import |
grep -q subtree ||
skip "Fileset feature is not supported"
local submount=${MOUNT}_$tdir
mkdir $MOUNT/$tdir
mkdir -p $submount || error "mkdir $submount failed"
FILESET="$FILESET/.." mount_client $submount &&
error "mount $submount should fail"
rmdir $submount
}
run_test 247e "mount .. as fileset"
test_247f() {
(( $MDSCOUNT > 1 )) || skip_env "needs >= 2 MDTs"
(( $MDS1_VERSION >= $(version_code 2.14.50.162) )) ||
skip "Need at least version 2.14.50.162"
lctl get_param -n mdc.$FSNAME-MDT0000*.import | grep -q subtree ||
skip "Fileset feature is not supported"
mkdir_on_mdt0 $DIR/$tdir || error "mkdir $tdir failed"
$LFS mkdir -i $((MDSCOUNT - 1)) $DIR/$tdir/remote ||
error "mkdir remote failed"
$LFS mkdir -i $((MDSCOUNT - 1)) $DIR/$tdir/remote/subdir ||
error "mkdir remote/subdir failed"
$LFS mkdir -i 0 -c $MDSCOUNT $DIR/$tdir/striped ||
error "mkdir striped failed"
mkdir $DIR/$tdir/striped/subdir || error "mkdir striped/subdir failed"
local submount=${MOUNT}_$tdir
mkdir -p $submount || error "mkdir $submount failed"
stack_trap "rmdir $submount"
local dir
local fileset=$FILESET
local mdts=$(mdts_nodes)
do_nodes $mdts "$LCTL set_param mdt.*.enable_remote_subdir_mount=1"
for dir in $tdir/remote $tdir/remote/subdir $tdir/striped \
$tdir/striped/subdir $tdir/striped/.; do
FILESET="$fileset/$dir" mount_client $submount ||
error "mount $dir failed"
umount_client $submount
done
}
run_test 247f "mount striped or remote directory as fileset"
test_subdir_mount_lock()
{
local testdir=$1
local submount=${MOUNT}_$(basename $testdir)
touch $DIR/$testdir/$tfile || error "touch $tfile failed"
mkdir -p $submount || error "mkdir $submount failed"
stack_trap "rmdir $submount"
FILESET="$FILESET/$testdir" mount_client $submount ||
error "mount $FILESET failed"
stack_trap "umount $submount"
local mdts=$(mdts_nodes)
local nrpcs
stat $submount > /dev/null || error "stat $submount failed"
cancel_lru_locks $MDC
stat $submount > /dev/null || error "stat $submount failed"
stat $submount/$tfile > /dev/null || error "stat $tfile failed"
do_nodes $mdts "$LCTL set_param mdt.*.md_stats=clear > /dev/null"
stat $submount/$tfile > /dev/null || error "stat $tfile failed"
nrpcs=$(do_nodes $mdts "lctl get_param -n mdt.*.md_stats" |
awk '/getattr/ {sum += $2} END {print sum}')
[ -z "$nrpcs" ] || error "$nrpcs extra getattr sent"
}
test_247g() {
(( $MDSCOUNT > 1 )) || skip_env "needs > 1 MDTs"
$LFS mkdir -i 0 -c 4 -H fnv_1a_64 $DIR/$tdir ||
error "mkdir $tdir failed"
test_subdir_mount_lock $tdir
}
run_test 247g "striped directory submount revalidate ROOT from cache"
test_247h() {
(( $MDSCOUNT > 1 )) || skip_env "needs > 1 MDTs"
(( $MDS1_VERSION >= $(version_code 2.15.51) )) ||
skip "Need MDS version at least 2.15.51"
$LFS mkdir -i 1 -c $MDSCOUNT $DIR/$tdir || error "mkdir $tdir failed"
test_subdir_mount_lock $tdir
mkdir_on_mdt -i 0 $DIR/$tdir/$tdir.0 || error "mkdir $tdir.0 failed"
mkdir_on_mdt -i 1 $DIR/$tdir/$tdir.0/$tdir.1 ||
error "mkdir $tdir.1 failed"
test_subdir_mount_lock $tdir/$tdir.0/$tdir.1
}
run_test 247h "remote directory submount revalidate ROOT from cache"
test_248a() {
local fast_read_sav=$($LCTL get_param -n llite.*.fast_read 2>/dev/null)
[ -z "$fast_read_sav" ] && skip "no fast read support"
# create a large file for fast read verification
dd if=/dev/zero of=$DIR/$tfile bs=1M count=128 > /dev/null 2>&1
# make sure the file is created correctly
$CHECKSTAT -s $((128*1024*1024)) $DIR/$tfile ||
{ rm -f $DIR/$tfile; skip "file creation error"; }
echo "Test 1: verify that fast read is 4 times faster on cache read"
# small read with fast read enabled
$LCTL set_param -n llite.*.fast_read=1
local t_fast=$(dd if=$DIR/$tfile of=/dev/null bs=4k 2>&1 |
grep -E -o '([[:digit:]\.\,e-]+) s' | cut -d's' -f1 |
sed -e 's/,/./' -e 's/[eE]+*/\*10\^/')
# small read with fast read disabled
$LCTL set_param -n llite.*.fast_read=0
local t_slow=$(dd if=$DIR/$tfile of=/dev/null bs=4k 2>&1 |
grep -E -o '([[:digit:]\.\,e-]+) s' | cut -d's' -f1 |
sed -e 's/,/./' -e 's/[eE]+*/\*10\^/')
# verify that fast read is 4 times faster for cache read
[ $(bc <<< "4 * $t_fast < $t_slow") -eq 1 ] ||
error_not_in_vm "fast read was not 4 times faster: " \
"$t_fast vs $t_slow"
echo "Test 2: verify the performance between big and small read"
$LCTL set_param -n llite.*.fast_read=1
# 1k non-cache read
cancel_lru_locks osc
local t_1k=$(dd if=$DIR/$tfile of=/dev/null bs=1k 2>&1 |
grep -E -o '([[:digit:]\.\,e-]+) s' | cut -d's' -f1 |
sed -e 's/,/./' -e 's/[eE]+*/\*10\^/')
# 1M non-cache read
cancel_lru_locks osc
local t_1m=$(dd if=$DIR/$tfile of=/dev/null bs=1k 2>&1 |
grep -E -o '([[:digit:]\.\,e-]+) s' | cut -d's' -f1 |
sed -e 's/,/./' -e 's/[eE]+*/\*10\^/')
# verify that big IO is not 4 times faster than small IO
[ $(bc <<< "4 * $t_1k >= $t_1m") -eq 1 ] ||
error_not_in_vm "bigger IO is way too fast: $t_1k vs $t_1m"
$LCTL set_param -n llite.*.fast_read=$fast_read_sav
rm -f $DIR/$tfile
}
run_test 248a "fast read verification"
test_248b() {
# Default short_io_bytes=16384, try both smaller and larger sizes.
# Lustre O_DIRECT read and write needs to be a multiple of PAGE_SIZE.
# 6017024 = 2^12*13*113 = 47008*128 = 11752*512 = 4096*1469 = 53248*113
echo "bs=53248 count=113 normal buffered write"
dd if=/dev/urandom of=$TMP/$tfile.0 bs=53248 count=113 ||
error "dd of initial data file failed"
stack_trap "rm -f $DIR/$tfile.[0-3] $TMP/$tfile.[0-3]" EXIT
echo "bs=47008 count=128 oflag=dsync normal write $tfile.0"
dd if=$TMP/$tfile.0 of=$DIR/$tfile.0 bs=47008 count=128 oflag=dsync ||
error "dd with sync normal writes failed"
cmp $TMP/$tfile.0 $DIR/$tfile.0 || error "compare $DIR/$tfile.0 failed"
echo "bs=11752 count=512 oflag=dsync small write $tfile.1"
dd if=$TMP/$tfile.0 of=$DIR/$tfile.1 bs=11752 count=512 oflag=dsync ||
error "dd with sync small writes failed"
cmp $TMP/$tfile.0 $DIR/$tfile.1 || error "compare $DIR/$tfile.1 failed"
cancel_lru_locks osc
# calculate the small O_DIRECT size and count for the client PAGE_SIZE
local num=$((13 * 113 / (PAGE_SIZE / 4096)))
echo "bs=$PAGE_SIZE count=$num iflag=direct small read $tfile.1"
dd if=$DIR/$tfile.1 of=$TMP/$tfile.1 bs=$PAGE_SIZE count=$num \
iflag=direct || error "dd with O_DIRECT small read failed"
# adjust bytes checked to handle larger PAGE_SIZE for ARM/PPC
cmp --bytes=$((PAGE_SIZE * num)) $TMP/$tfile.0 $TMP/$tfile.1 ||
error "compare $TMP/$tfile.1 failed"
local save=$($LCTL get_param -n osc.*OST000*.short_io_bytes | head -n 1)
stack_trap "$LCTL set_param osc.$FSNAME-*.short_io_bytes=$save" EXIT
# just to see what the maximum tunable value is, and test parsing
echo "test invalid parameter 2MB"
$LCTL set_param osc.$FSNAME-OST0000*.short_io_bytes=2M &&
error "too-large short_io_bytes allowed"
echo "test maximum parameter 512KB"
# if we can set a larger short_io_bytes, run test regardless of version
if ! $LCTL set_param osc.$FSNAME-OST0000*.short_io_bytes=512K; then
# older clients may not allow setting it this large, that's OK
[ $CLIENT_VERSION -ge $(version_code 2.13.50) ] ||
skip "Need at least client version 2.13.50"
error "medium short_io_bytes failed"
fi
$LCTL get_param osc.$FSNAME-OST0000*.short_io_bytes
size=$($LCTL get_param -n osc.$FSNAME-OST0000*.short_io_bytes)
echo "test large parameter 64KB"
$LCTL set_param osc.$FSNAME-*.short_io_bytes=65536
$LCTL get_param osc.$FSNAME-OST0000*.short_io_bytes
echo "bs=47008 count=128 oflag=dsync large write $tfile.2"
dd if=$TMP/$tfile.0 of=$DIR/$tfile.2 bs=47008 count=128 oflag=dsync ||
error "dd with sync large writes failed"
cmp $TMP/$tfile.0 $DIR/$tfile.2 || error "compare $DIR/$tfile.2 failed"
# calculate the large O_DIRECT size and count for the client PAGE_SIZE
local size=$(((4096 * 13 + PAGE_SIZE - 1) / PAGE_SIZE * PAGE_SIZE))
num=$((113 * 4096 / PAGE_SIZE))
echo "bs=$size count=$num oflag=direct large write $tfile.3"
dd if=$TMP/$tfile.0 of=$DIR/$tfile.3 bs=$size count=$num oflag=direct ||
error "dd with O_DIRECT large writes failed"
cmp --bytes=$((size * num)) $TMP/$tfile.0 $DIR/$tfile.3 ||
error "compare $DIR/$tfile.3 failed"
cancel_lru_locks osc
echo "bs=$size count=$num iflag=direct large read $tfile.2"
dd if=$DIR/$tfile.2 of=$TMP/$tfile.2 bs=$size count=$num iflag=direct ||
error "dd with O_DIRECT large read failed"
cmp --bytes=$((size * num)) $TMP/$tfile.0 $TMP/$tfile.2 ||
error "compare $TMP/$tfile.2 failed"
echo "bs=$size count=$num iflag=direct large read $tfile.3"
dd if=$DIR/$tfile.3 of=$TMP/$tfile.3 bs=$size count=$num iflag=direct ||
error "dd with O_DIRECT large read failed"
cmp --bytes=$((size * num)) $TMP/$tfile.0 $TMP/$tfile.3 ||
error "compare $TMP/$tfile.3 failed"
}
run_test 248b "test short_io read and write for both small and large sizes"
test_248c() {
$LCTL set_param llite.*.read_ahead_stats=c
# This test compares whole file readahead to non-whole file readahead
# The performance should be consistently slightly faster for whole file,
# and the bug caused whole file readahead to be 80-100x slower, but to
# account for possible test system lag, we require whole file to be
# <= 1.5xnon-whole file, and require 3 failures in a row to fail the
# test
local time1
local time2
local counter
local whole_mb=$($LCTL get_param -n llite.*.max_read_ahead_whole_mb)
stack_trap "$LCTL set_param llite.*.max_read_ahead_whole_mb=$whole_mb"
# LU-16904 Use stripe counct 1 for root before the readahead issue
# described in LU-17755 and LU-15155 get fixed
test_mkdir $DIR/$tdir || error "mkdir $tdir failed"
$LFS setstripe -c 1 $DIR/$tdir || error "setstripe $tdir failed"
for (( counter = 0; counter < 3; counter++ )); do
$LCTL set_param llite.*.max_read_ahead_whole_mb=64
rm -f $DIR/$tdir/$tfile
touch $DIR/$tdir/$tfile || error "(0) failed to create file"
# 64 MiB
$TRUNCATE $DIR/$tdir/$tfile 67108864 || error "(1) trunc failed"
time1=$(dd if=$DIR/$tdir/$tfile bs=4K of=/dev/null 2>&1 |
awk '/bytes/ { print $8 }')
echo "whole file readahead of 64 MiB took $time1 seconds"
$LCTL get_param llite.*.read_ahead_stats
$LCTL set_param llite.*.read_ahead_stats=clear
$LCTL set_param llite.*.max_read_ahead_whole_mb=8
rm -f $DIR/$tdir/$tfile
touch $DIR/$tdir/$tfile || error "(2) failed to create file"
# 64 MiB
$TRUNCATE $DIR/$tdir/$tfile 67108864 || error "(3) trunc failed"
time2=$(dd if=$DIR/$tdir/$tfile bs=4K of=/dev/null 2>&1 |
awk '/bytes/ { print $8 }')
echo "non-whole file readahead of 64 MiB took $time2 seconds"
$LCTL get_param llite.*.read_ahead_stats
# Check if time1 is not more than 1.5 times2
timecheck=$(echo "$time1 <= (1.5 * $time2)" | bc -l)
if (( $timecheck == 1 )); then
echo "Test passed on attempt $counter"
break
fi
echo "Try $counter failed: whole file readahead took: $time1, greater than non: $time2"
done
(( $counter < 3 )) ||
error "whole file readahead failed 3x, probably not just VM lag"
}
run_test 248c "verify whole file read behavior"
test_248d() {
local fast_read_sav=$($LCTL get_param -n llite.*.fast_read 2>/dev/null)
[ -z "$fast_read_sav" ] && skip "no fast read support"
stack_trap "$LCTL set_param -n llite.*.fast_read=$fast_read_sav"
# Create a small file and warm the page cache
dd if=/dev/urandom of=$DIR/$tfile bs=32K count=1 2>/dev/null ||
error "dd write failed"
stack_trap "rm -f $DIR/$tfile"
# Warm page cache with a full read
$LCTL set_param -n llite.*.fast_read=1
dd if=$DIR/$tfile of=/dev/null bs=32K 2>/dev/null ||
error "dd warmup read failed"
# Clear stats and do a burst of small reads from cache
$LCTL set_param llite.*.read_ahead_stats=0
dd if=$DIR/$tfile of=/dev/null bs=8 count=4000 2>/dev/null ||
error "dd tiny read failed"
local failed=$($LCTL get_param -n llite.*.read_ahead_stats |
awk '/failed_to_fast_read/ { print $1 }')
# All reads from a warm cache should use fast read without
# any failures. A non-zero count means pages were not
# served from cache, indicating the fast read path is
# broken or being bypassed.
[[ ${failed:-0} -eq 0 ]] ||
error "expected 0 failed_to_fast_read, got $failed"
}
run_test 248d "fast read serves tiny reads from cache without failures"
test_249() { # LU-7890
[ $MDS1_VERSION -lt $(version_code 2.8.53) ] &&
skip "Need at least version 2.8.54"
rm -f $DIR/$tfile
$LFS setstripe -c 1 $DIR/$tfile
# Offset 2T == 4k * 512M
dd if=/dev/zero of=$DIR/$tfile bs=4k count=1 seek=512M ||
error "dd to 2T offset failed"
}
run_test 249 "Write above 2T file size"
test_250() {
[ "$(facet_fstype ost$(($($LFS getstripe -i $DIR/$tfile) + 1)))" = "zfs" ] \
&& skip "no 16TB file size limit on ZFS"
[ "$(facet_fstype ost$(($($LFS getstripe -i $DIR/$tfile) + 1)))" = "wbcfs" ] \
&& skip "no 16TB file size limit on wbcfs"
$LFS setstripe -c 1 $DIR/$tfile
# ldiskfs extent file size limit is (16TB - 4KB - 1) bytes
local size=$((16 * 1024 * 1024 * 1024 * 1024 - 4096 - 1))
$TRUNCATE $DIR/$tfile $size || error "truncate $tfile to $size failed"
dd if=/dev/zero of=$DIR/$tfile bs=10 count=1 oflag=append \
conv=notrunc,fsync && error "append succeeded"
return 0
}
run_test 250 "Write above 16T limit"
test_251a() {
$LFS setstripe -c -1 -S 1048576 $DIR/$tfile
#define OBD_FAIL_LLITE_LOST_LAYOUT 0x1407
#Skip once - writing the first stripe will succeed
$LCTL set_param fail_loc=0xa0001407 fail_val=1
$MULTIOP $DIR/$tfile o:O_RDWR:w2097152c 2>&1 | grep -q "short write" &&
error "short write happened"
$LCTL set_param fail_loc=0xa0001407 fail_val=1
$MULTIOP $DIR/$tfile or2097152c 2>&1 | grep -q "short read" &&
error "short read happened"
rm -f $DIR/$tfile
}
run_test 251a "Handling short read and write correctly"
test_252() {
remote_mds_nodsh && skip "remote MDS with nodsh"
remote_ost_nodsh && skip "remote OST with nodsh"
if [ "$ost1_FSTYPE" != ldiskfs ] || [ "$mds1_FSTYPE" != ldiskfs ]; then
skip_env "ldiskfs only test"
fi
local tgt
local dev
local out
local uuid
local num
local gen
# check lr_reader on OST0000
tgt=ost1
dev=$(facet_device $tgt)
out=$(do_facet $tgt $LR_READER $dev)
[ $? -eq 0 ] || error "$LR_READER failed on target $tgt device $dev"
echo "$out"
uuid=$(echo "$out" | grep -i uuid | awk '{ print $2 }')
[ "$uuid" == "$(ostuuid_from_index 0)" ] ||
error "Invalid uuid returned by $LR_READER on target $tgt"
echo -e "uuid returned by $LR_READER is '$uuid'\n"
# check lr_reader -c on MDT0000
tgt=mds1
dev=$(facet_device $tgt)
if ! do_facet $tgt $LR_READER -h | grep -q OPTIONS; then
skip "$LR_READER does not support additional options"
fi
out=$(do_facet $tgt $LR_READER -c $dev)
[ $? -eq 0 ] || error "$LR_READER failed on target $tgt device $dev"
echo "$out"
num=$(echo "$out" | grep -c "mdtlov")
[ "$num" -eq $((MDSCOUNT - 1)) ] ||
error "Invalid number of mdtlov clients returned by $LR_READER"
echo -e "Number of mdtlov clients returned by $LR_READER is '$num'\n"
# check lr_reader -cr on MDT0000
out=$(do_facet $tgt $LR_READER -cr $dev)
[ $? -eq 0 ] || error "$LR_READER failed on target $tgt device $dev"
echo "$out"
echo "$out" | grep -q "^reply_data:$" ||
error "$LR_READER should have returned 'reply_data' section"
num=$(echo "$out" | grep -c "client_generation")
echo -e "Number of reply data returned by $LR_READER is '$num'\n"
}
run_test 252 "check lr_reader tool"
test_253() {
[ $PARALLEL == "yes" ] && skip "skip parallel run"
remote_mds_nodsh && skip "remote MDS with nodsh"
remote_mgs_nodsh && skip "remote MGS with nodsh"
check_set_fallocate_or_skip
local ostidx=0
local rc=0
local ost_name=$(ostname_from_index $ostidx)
# on the mdt's osc
local mdtosc_proc1=$(get_mdtosc_proc_path $SINGLEMDS $ost_name)
do_facet $SINGLEMDS $LCTL get_param -n \
osp.$mdtosc_proc1.reserved_mb_high ||
skip "remote MDS does not support reserved_mb_high"
rm -rf $DIR/$tdir
wait_mds_ost_sync
wait_delete_completed
mkdir $DIR/$tdir
stack_trap "rm -rf $DIR/$tdir"
pool_add $TESTNAME || error "Pool creation failed"
pool_add_targets $TESTNAME 0 || error "Pool add targets failed"
$LFS setstripe $DIR/$tdir -i $ostidx -c 1 -p $FSNAME.$TESTNAME ||
error "Setstripe failed"
dd if=/dev/zero of=$DIR/$tdir/$tfile.0 bs=1M count=10
local wms=$(ost_watermarks_get $ostidx)
ost_watermarks_set $ostidx 60 50
stack_trap "ost_watermarks_set $ostidx $wms"
local free_kb=$($LFS df $MOUNT | awk "/$ost_name/ { print \$4 }")
local size=$((free_kb * 1024))
fallocate -l $size $DIR/$tdir/fill_ost$ostidx ||
error "fallocate failed"
sleep_maxage
local oa_status=$(do_facet $SINGLEMDS $LCTL get_param -n \
osp.$mdtosc_proc1.prealloc_status)
echo "prealloc_status $oa_status"
dd if=/dev/zero of=$DIR/$tdir/$tfile.1 bs=1M count=1 &&
error "File creation should fail"
#object allocation was stopped, but we still able to append files
dd if=/dev/zero of=$DIR/$tdir/$tfile.0 bs=1M seek=6 count=5 \
oflag=append || error "Append failed"
rm -f $DIR/$tdir/$tfile.0
rm -f $DIR/$tdir/fill_ost$ostidx
wait_delete_completed
sleep_maxage
for i in $(seq 10 12); do
dd if=/dev/zero of=$DIR/$tdir/$tfile.$i bs=1M count=1 \
2>/dev/null || error "File creation failed after rm"
done
oa_status=$(do_facet $SINGLEMDS $LCTL get_param -n \
osp.$mdtosc_proc1.prealloc_status)
echo "prealloc_status $oa_status"
if (( oa_status != 0 )); then
error "Object allocation still disable after rm"
fi
}
run_test 253 "Check object allocation limit"
test_254() {
[ $PARALLEL == "yes" ] && skip "skip parallel run"
remote_mds_nodsh && skip "remote MDS with nodsh"
local mdt=$(facet_svc $SINGLEMDS)
do_facet $SINGLEMDS $LCTL get_param -n mdd.$mdt.changelog_size ||
skip "MDS does not support changelog_size"
local cl_user
changelog_register || error "changelog_register failed"
changelog_clear 0 || error "changelog_clear failed"
local size1=$(do_facet $SINGLEMDS \
$LCTL get_param -n mdd.$mdt.changelog_size)
echo "Changelog size $size1"
rm -rf $DIR/$tdir
$LFS mkdir -i 0 $DIR/$tdir
# change something
mkdir -p $DIR/$tdir/pics/2008/zachy
touch $DIR/$tdir/pics/2008/zachy/timestamp
cp /etc/hosts $DIR/$tdir/pics/2008/zachy/pic1.jpg
mv $DIR/$tdir/pics/2008/zachy $DIR/$tdir/pics/zach
ln $DIR/$tdir/pics/zach/pic1.jpg $DIR/$tdir/pics/2008/portland.jpg
ln -s $DIR/$tdir/pics/2008/portland.jpg $DIR/$tdir/pics/desktop.jpg
rm $DIR/$tdir/pics/desktop.jpg
local size2=$(do_facet $SINGLEMDS \
$LCTL get_param -n mdd.$mdt.changelog_size)
echo "Changelog size after work $size2"
(( $size2 > $size1 )) ||
error "new Changelog size=$size2 less than old size=$size1"
}
run_test 254 "Check changelog size"
ladvise_no_type()
{
local type=$1
local file=$2
lfs ladvise -a invalid $file 2>&1 | grep "Valid types" |
awk -F: '{print $2}' | grep $type > /dev/null
if [ $? -ne 0 ]; then
return 0
fi
return 1
}
ladvise_no_ioctl()
{
local file=$1
lfs ladvise -a willread $file > /dev/null 2>&1
if [ $? -eq 0 ]; then
return 1
fi
lfs ladvise -a willread $file 2>&1 |
grep "Inappropriate ioctl for device" > /dev/null
if [ $? -eq 0 ]; then
return 0
fi
return 1
}
percent() {
# don't print any decimal places to avoid bash comparison issues
bc <<< "scale=0; ($1 - $2) * 100 / $2"
}
# run a random read IO workload
# usage: random_read_iops <filename> <filesize> <iosize>
random_read_iops() {
local file=$1
local fsize=$2
local iosize=${3:-4096}
$READS -f $file -s $fsize -b $iosize -n $((fsize / iosize)) -t 60 |
sed -e '/^$/d' -e 's#.*s, ##' -e 's#MB/s##'
}
drop_file_oss_cache() {
local file="$1"
local nodes="$2"
$LFS ladvise -a dontneed $file 2>/dev/null ||
do_nodes $nodes "echo 3 > /proc/sys/vm/drop_caches"
}
ladvise_willread_performance()
{
local repeat=10
local average_origin=0
local average_cache=0
local average_ladvise=0
local osts=$(osts_nodes)
# Hybrid IO switches to DIO, which invalidates much of the caching
# So disable it for this test
local hybrid=$($LCTL get_param -n llite.*.hybrid_io)
$LCTL set_param llite.*.hybrid_io=0
stack_trap "$LCTL set_param -n llite.*.hybrid_io=$hybrid" EXIT
for ((i = 1; i <= $repeat; i++)); do
echo "Iter $i/$repeat: reading without willread hint"
cancel_lru_locks osc
drop_file_oss_cache $DIR/$tfile $osts
local speed_origin=$(random_read_iops $DIR/$tfile $size)
echo "Iter $i/$repeat: uncached speed: $speed_origin"
average_origin=$(bc <<<"$average_origin + $speed_origin")
cancel_lru_locks osc
local speed_cache=$(random_read_iops $DIR/$tfile $size)
echo "Iter $i/$repeat: OSS cache speed: $speed_cache"
average_cache=$(bc <<<"$average_cache + $speed_cache")
cancel_lru_locks osc
drop_file_oss_cache $DIR/$tfile $osts
$LFS ladvise -a willread $DIR/$tfile || error "ladvise failed"
local speed_ladvise=$(random_read_iops $DIR/$tfile $size)
echo "Iter $i/$repeat: ladvise speed: $speed_ladvise"
average_ladvise=$(bc <<<"$average_ladvise + $speed_ladvise")
done
average_origin=$(bc <<<"scale=2; $average_origin / $repeat")
average_cache=$(bc <<<"scale=2; $average_cache / $repeat")
average_ladvise=$(bc <<<"scale=2; $average_ladvise / $repeat")
speedup_cache=$(percent $average_cache $average_origin)
speedup_ladvise=$(percent $average_ladvise $average_origin)
echo "Average uncached read: $average_origin"
echo "Average speedup with OSS cached read: " \
"$average_cache = +$speedup_cache%"
echo "Average speedup with ladvise willread: " \
"$average_ladvise = +$speedup_ladvise%"
local lowest_speedup=20
if (( $speedup_cache < $lowest_speedup )); then
echo "Speedup with OSS cached read less than $lowest_speedup%,"\
" got $speedup_cache%. Skipping ladvise willread check."
return 0
fi
# the test won't work on ZFS until it supports 'ladvise dontneed', but
# it is still good to run until then to exercise 'ladvise willread'
! $LFS ladvise -a dontneed $DIR/$tfile &&
[ "$ost1_FSTYPE" = "zfs" ] &&
echo "osd-zfs does not support dontneed or drop_caches" &&
return 0
lowest_speedup=$(bc <<< "scale=0; $average_cache / 2")
(( $speedup_ladvise > $lowest_speedup )) ||
error_not_in_vm "Speedup with willread is less than " \
"$lowest_speedup%, got $speedup_ladvise%"
}
test_255a() {
[ $OST1_VERSION -lt $(version_code 2.8.54) ] &&
skip "lustre < 2.8.54 does not support ladvise "
remote_ost_nodsh && skip "remote OST with nodsh"
stack_trap "rm -f $DIR/$tfile"
lfs setstripe -c -1 -i 0 $DIR/$tfile || error "$tfile failed"
ladvise_no_type willread $DIR/$tfile &&
skip "willread ladvise is not supported"
ladvise_no_ioctl $DIR/$tfile &&
skip "ladvise ioctl is not supported"
local size_mb=100
local size=$((size_mb * 1048576))
dd if=/dev/zero of=$DIR/$tfile bs=1048576 count=$size_mb ||
error "dd to $DIR/$tfile failed"
lfs ladvise -a willread $DIR/$tfile ||
error "Ladvise failed with no range argument"
lfs ladvise -a willread -s 0 $DIR/$tfile ||
error "Ladvise failed with no -l or -e argument"
lfs ladvise -a willread -e 1 $DIR/$tfile ||
error "Ladvise failed with only -e argument"
lfs ladvise -a willread -l 1 $DIR/$tfile ||
error "Ladvise failed with only -l argument"
lfs ladvise -a willread -s 2 -e 1 $DIR/$tfile &&
error "End offset should not be smaller than start offset"
lfs ladvise -a willread -s 2 -e 2 $DIR/$tfile &&
error "End offset should not be equal to start offset"
lfs ladvise -a willread -s $size -l 1 $DIR/$tfile ||
error "Ladvise failed with overflowing -s argument"
lfs ladvise -a willread -s 1 -e $((size + 1)) $DIR/$tfile ||
error "Ladvise failed with overflowing -e argument"
lfs ladvise -a willread -s 1 -l $size $DIR/$tfile ||
error "Ladvise failed with overflowing -l argument"
lfs ladvise -a willread -l 1 -e 2 $DIR/$tfile &&
error "Ladvise succeeded with conflicting -l and -e arguments"
echo "Synchronous ladvise should wait"
local delay=8
local osts=$(osts_nodes)
#define OBD_FAIL_OST_LADVISE_PAUSE 0x237
do_nodes $osts $LCTL set_param fail_val=$delay fail_loc=0x237
stack_trap "do_nodes $osts $LCTL set_param fail_loc=0"
local start_ts=$SECONDS
lfs ladvise -a willread $DIR/$tfile ||
error "Ladvise failed with no range argument"
local end_ts=$SECONDS
local inteval_ts=$((end_ts - start_ts))
if [ $inteval_ts -lt $(($delay - 1)) ]; then
error "Synchronous advice didn't wait reply"
fi
echo "Asynchronous ladvise shouldn't wait"
local start_ts=$SECONDS
lfs ladvise -a willread -b $DIR/$tfile ||
error "Ladvise failed with no range argument"
local end_ts=$SECONDS
local inteval_ts=$((end_ts - start_ts))
if [ $inteval_ts -gt $(($delay / 2)) ]; then
error "Asynchronous advice blocked"
fi
ladvise_willread_performance
}
run_test 255a "check 'lfs ladvise -a willread'"
facet_meminfo() {
local facet=$1
local info=$2
do_facet $facet "cat /proc/meminfo | grep ^${info}:" | awk '{print $2}'
}
test_255b() {
[ $OST1_VERSION -lt $(version_code 2.8.54) ] &&
skip "lustre < 2.8.54 does not support ladvise "
remote_ost_nodsh && skip "remote OST with nodsh"
stack_trap "rm -f $DIR/$tfile"
lfs setstripe -c 1 -i 0 $DIR/$tfile
ladvise_no_type dontneed $DIR/$tfile &&
skip "dontneed ladvise is not supported"
ladvise_no_ioctl $DIR/$tfile &&
skip "ladvise ioctl is not supported"
! $LFS ladvise -a dontneed $DIR/$tfile &&
[ "$ost1_FSTYPE" = "zfs" ] &&
skip "zfs-osd does not support 'ladvise dontneed'"
local size_mb=100
local size=$((size_mb * 1048576))
# In order to prevent disturbance of other processes, only check 3/4
# of the memory usage
local kibibytes=$((size_mb * 1024 * 3 / 4))
dd if=/dev/zero of=$DIR/$tfile bs=1048576 count=$size_mb ||
error "dd to $DIR/$tfile failed"
#force write to complete before dropping OST cache & checking memory
sync
local total=$(facet_meminfo ost1 MemTotal)
echo "Total memory: $total KiB"
do_facet ost1 "sync && echo 3 > /proc/sys/vm/drop_caches"
local before_read=$(facet_meminfo ost1 Cached)
echo "Cache used before read: $before_read KiB"
lfs ladvise -a willread $DIR/$tfile ||
error "Ladvise willread failed"
local after_read=$(facet_meminfo ost1 Cached)
echo "Cache used after read: $after_read KiB"
lfs ladvise -a dontneed $DIR/$tfile ||
error "Ladvise dontneed again failed"
local no_read=$(facet_meminfo ost1 Cached)
echo "Cache used after dontneed ladvise: $no_read KiB"
if [ $total -lt $((before_read + kibibytes)) ]; then
echo "Memory is too small, abort checking"
return 0
fi
if [ $((before_read + kibibytes)) -gt $after_read ]; then
error "Ladvise willread should use more memory" \
"than $kibibytes KiB"
fi
if [ $((no_read + kibibytes)) -gt $after_read ]; then
error "Ladvise dontneed should release more memory" \
"than $kibibytes KiB"
fi
}
run_test 255b "check 'lfs ladvise -a dontneed'"
test_255c() {
[ $OST1_VERSION -lt $(version_code 2.10.50) ] &&
skip "lustre < 2.10.50 does not support lockahead"
local ost1_imp=$(get_osc_import_name client ost1)
local imp_name=$($LCTL list_param osc.$ost1_imp | head -n1 |
cut -d'.' -f2)
local count
local new_count
local difference
local i
local rc
test_mkdir -p $DIR/$tdir
$LFS setstripe -i 0 -c 1 $DIR/$tdir
#test 10 returns only success/failure
i=10
lockahead_test -d $DIR/$tdir -t $i -f $tfile
rc=$?
if [ $rc -eq 255 ]; then
error "Ladvise test${i} failed, ${rc}"
fi
#test 11 counts lock enqueue requests, all others count new locks
i=11
count=$(do_facet ost1 \
$LCTL get_param -n ost.OSS.ost.stats)
count=$(echo "$count" | grep ldlm_extent_enqueue | awk '{ print $2 }')
lockahead_test -d $DIR/$tdir -t $i -f $tfile
rc=$?
if [ $rc -eq 255 ]; then
error "Ladvise test${i} failed, ${rc}"
fi
new_count=$(do_facet ost1 \
$LCTL get_param -n ost.OSS.ost.stats)
new_count=$(echo "$new_count" | grep ldlm_extent_enqueue | \
awk '{ print $2 }')
difference="$((new_count - count))"
if [ $difference -ne $rc ]; then
error "Ladvise test${i}, bad enqueue count, returned " \
"${rc}, actual ${difference}"
fi
for i in $(seq 12 21); do
# If we do not do this, we run the risk of having too many
# locks and starting lock cancellation while we are checking
# lock counts.
cancel_lru_locks osc
count=$($LCTL get_param -n \
ldlm.namespaces.$imp_name.lock_unused_count)
lockahead_test -d $DIR/$tdir -t $i -f $tfile
rc=$?
if [ $rc -eq 255 ]; then
error "Ladvise test ${i} failed, ${rc}"
fi
new_count=$($LCTL get_param -n \
ldlm.namespaces.$imp_name.lock_unused_count)
difference="$((new_count - count))"
# Test 15 output is divided by 100 to map down to valid return
if [ $i -eq 15 ]; then
rc="$((rc * 100))"
fi
if [ $difference -ne $rc ]; then
error "Ladvise test ${i}, bad lock count, returned " \
"${rc}, actual ${difference}"
fi
done
#test 22 returns only success/failure
i=22
lockahead_test -d $DIR/$tdir -t $i -f $tfile
rc=$?
if [ $rc -eq 255 ]; then
error "Ladvise test${i} failed, ${rc}"
fi
}
run_test 255c "suite of ladvise lockahead tests"
test_256() {
[ $PARALLEL == "yes" ] && skip "skip parallel run"
remote_mds_nodsh && skip "remote MDS with nodsh"
[ "$mds1_FSTYPE" != "ldiskfs" ] && skip "ldiskfs only test"
changelog_users $SINGLEMDS | grep "^cl" &&
skip "active changelog user"
local cl_user
local cat_sl
local mdt_dev
mdt_dev=$(facet_device $SINGLEMDS)
echo $mdt_dev
changelog_register || error "changelog_register failed"
rm -rf $DIR/$tdir
mkdir_on_mdt -i$(($(facet_number $SINGLEMDS) - 1)) $DIR/$tdir
changelog_clear 0 || error "changelog_clear failed"
# change something
touch $DIR/$tdir/{1..10}
# stop the MDT
stop $SINGLEMDS || error "Fail to stop MDT"
# remount the MDT
start $SINGLEMDS $(facet_device $SINGLEMDS) $MDS_MOUNT_OPTS ||
error "Fail to start MDT"
#after mount new plainllog is used
touch $DIR/$tdir/{11..19}
local tmpfile="$(mktemp --tmpdir -u $tfile.XXXXXX)"
stack_trap "rm -f $tmpfile"
cat_sl=$(do_facet $SINGLEMDS "sync; \
$DEBUGFS -c -R 'dump changelog_catalog $tmpfile' $mdt_dev; \
llog_reader $tmpfile | grep -c type=1064553b")
do_facet $SINGLEMDS llog_reader $tmpfile
[ $cat_sl != 2 ] && error "Changelog catalog has $cat_sl != 2 slots"
changelog_clear 0 || error "changelog_clear failed"
cat_sl=$(do_facet $SINGLEMDS "sync; \
$DEBUGFS -c -R 'dump changelog_catalog $tmpfile' $mdt_dev; \
llog_reader $tmpfile | grep -c type=1064553b")
if (( cat_sl == 2 )); then
error "Empty plain llog was not deleted from changelog catalog"
elif (( cat_sl != 1 )); then
error "Active plain llog shouldn't be deleted from catalog"
fi
}
run_test 256 "Check llog delete for empty and not full state"
test_257() {
remote_mds_nodsh && skip "remote MDS with nodsh"
[[ $MDS1_VERSION -lt $(version_code 2.8.55) ]] &&
skip "Need MDS version at least 2.8.55"
test_mkdir $DIR/$tdir
setfattr -n trusted.name1 -v value1 $DIR/$tdir ||
error "setfattr -n trusted.name1=value1 $DIR/$tdir failed"
stat $DIR/$tdir
#define OBD_FAIL_MDS_XATTR_REP 0x161
local mdtidx=$($LFS getstripe -m $DIR/$tdir)
local facet=mds$((mdtidx + 1))
set_nodes_failloc $(facet_active_host $facet) 0x80000161
getfattr -n trusted.name1 $DIR/$tdir 2> /dev/null
stop $facet || error "stop MDS failed"
start $facet $(mdsdevname $((mdtidx + 1))) $MDS_MOUNT_OPTS ||
error "start MDS fail"
wait_recovery_complete $facet
}
run_test 257 "xattr locks are not lost"
# Verify we take the i_mutex when security requires it
test_258a() {
#define OBD_FAIL_IMUTEX_SEC 0x141c
$LCTL set_param fail_loc=0x141c
touch $DIR/$tfile
chmod u+s $DIR/$tfile
chmod a+rwx $DIR/$tfile
$RUNAS dd if=/dev/zero of=$DIR/$tfile bs=4k count=1 oflag=append
RC=$?
if [ $RC -ne 0 ]; then
error "error, failed to take i_mutex, rc=$?"
fi
rm -f $DIR/$tfile
}
run_test 258a "verify i_mutex security behavior when suid attributes is set"
# Verify we do NOT take the i_mutex in the normal case
test_258b() {
#define OBD_FAIL_IMUTEX_NOSEC 0x141d
$LCTL set_param fail_loc=0x141d
touch $DIR/$tfile
chmod a+rwx $DIR
chmod a+rw $DIR/$tfile
$RUNAS dd if=/dev/zero of=$DIR/$tfile bs=4k count=1 oflag=append
RC=$?
if [ $RC -ne 0 ]; then
error "error, took i_mutex unnecessarily, rc=$?"
fi
rm -f $DIR/$tfile
}
run_test 258b "verify i_mutex security behavior"
test_259() {
local file=$DIR/$tfile
local before
local after
[ "$mds1_FSTYPE" != "ldiskfs" ] && skip "ldiskfs only test"
stack_trap "rm -f $file" EXIT
wait_delete_completed
before=$(do_facet ost1 "$LCTL get_param -n osd-*.*OST0000.kbytesfree")
echo "before: $before"
$LFS setstripe -i 0 -c 1 $file
dd if=/dev/zero of=$file bs=1M count=10 || error "couldn't write"
sync_all_data
after=$(do_facet ost1 "$LCTL get_param -n osd-*.*OST0000.kbytesfree")
echo "after write: $after"
#define OBD_FAIL_OSD_FAIL_AT_TRUNCATE 0x2301
do_facet ost1 $LCTL set_param fail_loc=0x2301
$TRUNCATE $file 0
after=$(do_facet ost1 "$LCTL get_param -n osd-*.*OST0000.kbytesfree")
echo "after truncate: $after"
stop ost1
do_facet ost1 $LCTL set_param fail_loc=0
start ost1 $(ostdevname 1) $OST_MOUNT_OPTS || error "cannot start ost1"
sleep 2
after=$(do_facet ost1 "$LCTL get_param -n osd-*.*OST0000.kbytesfree")
echo "after restart: $after"
[ $((after - before)) -ge $(fs_log_size ost1) ] &&
error "missing truncate?"
return 0
}
run_test 259 "crash at delayed truncate"
test_260() {
#define OBD_FAIL_MDC_CLOSE 0x806
$LCTL set_param fail_loc=0x80000806
touch $DIR/$tfile
}
run_test 260 "Check mdc_close fail"
### Data-on-MDT sanity tests ###
test_270a() {
[ $MDS1_VERSION -lt $(version_code 2.10.55) ] &&
skip "Need MDS version at least 2.10.55 for DoM"
# create DoM file
local dom=$DIR/$tdir/dom_file
local tmp=$DIR/$tdir/tmp_file
mkdir_on_mdt0 $DIR/$tdir
# basic checks for DoM component creation
$LFS setstripe -E 1024K -E 2048K -L mdt $dom 2>/dev/null &&
error "Can set MDT layout to non-first entry"
$LFS setstripe -E 1024K -L mdt -E 2048K -L mdt $dom 2>/dev/null &&
error "Can define multiple entries as MDT layout"
$LFS setstripe -E 1M -L mdt $dom || error "Can't create DoM layout"
[ $($LFS getstripe -L $dom) == "mdt" ] || error "bad pattern"
[ $($LFS getstripe -c $dom) == 0 ] || error "bad stripe count"
[ $($LFS getstripe -S $dom) == 1048576 ] || error "bad stripe size"
local mdtidx=$($LFS getstripe -m $dom)
local mdtname=MDT$(printf %04x $mdtidx)
local facet=mds$((mdtidx + 1))
local space_check=1
# previous files might released in background, but tests wants to have stable
# counters a specially in case running in loop
do_facet $facet \
"$LCTL set_param -n mdt.*.force_sync=1"
# Skip free space checks with ZFS
[ "$(facet_fstype $facet)" == "zfs" ] && space_check=0
# write
sync
local size_tmp=$((65536 * 3))
local mdtfree1=$(do_facet $facet \
lctl get_param -n osd*.*$mdtname.kbytesfree)
dd if=/dev/urandom of=$tmp bs=1024 count=$((size_tmp / 1024))
# check also direct IO along write
# IO size must be a multiple of PAGE_SIZE on all platforms (ARM=64KB)
dd if=$tmp of=$dom bs=65536 count=$((size_tmp / 65536)) oflag=direct
sync
cmp $tmp $dom || error "file data is different"
[ $(stat -c%s $dom) == $size_tmp ] ||
error "bad size after write: $(stat -c%s $dom) != $size_tmp"
if [ $space_check == 1 ]; then
local mdtfree2=$(do_facet $facet \
lctl get_param -n osd*.*$mdtname.kbytesfree)
# increase in usage from by $size_tmp
[ $(($mdtfree1 - $mdtfree2)) -ge $((size_tmp / 1024)) ] ||
error "MDT free space wrong after write: " \
"$mdtfree1 >= $mdtfree2 + $size_tmp/1024"
fi
# truncate
local size_dom=10000
$TRUNCATE $dom $size_dom
[ $(stat -c%s $dom) == $size_dom ] ||
error "bad size after truncate: $(stat -c%s $dom) != $size_dom"
if [ $space_check == 1 ]; then
mdtfree1=$(do_facet $facet \
lctl get_param -n osd*.*$mdtname.kbytesfree)
# decrease in usage from $size_tmp to new $size_dom
[ $(($mdtfree1 - $mdtfree2)) -ge \
$(((size_tmp - size_dom) / 4096 * 4)) ] ||
error "MDT free space is wrong after truncate: " \
"$mdtfree1 >= $mdtfree2 + ($size_tmp - $size_dom) / 1024"
fi
# append
cat $tmp >> $dom
sync
size_dom=$((size_dom + size_tmp))
[ $(stat -c%s $dom) == $size_dom ] ||
error "bad size after append: $(stat -c%s $dom) != $size_dom"
if [ $space_check == 1 ]; then
mdtfree2=$(do_facet $facet \
lctl get_param -n osd*.*$mdtname.kbytesfree)
# increase in usage by $size_tmp from previous
[ $(($mdtfree1 - $mdtfree2)) -ge $((size_tmp / 1024)) ] ||
error "MDT free space is wrong after append: " \
"$mdtfree1 >= $mdtfree2 + $size_tmp/1024"
fi
# delete
rm $dom
# object might released in background
do_facet $facet \
"$LCTL set_param -n mdt.*.force_sync=1"
if [ $space_check == 1 ]; then
mdtfree1=$(do_facet $facet \
lctl get_param -n osd*.*$mdtname.kbytesfree)
# decrease in usage by $size_dom from previous
[ $(($mdtfree1 - $mdtfree2)) -ge $((size_dom / 1024)) ] ||
error "MDT free space is wrong after removal: " \
"$mdtfree1 >= $mdtfree2 + $size_dom/1024"
fi
# combined striping
$LFS setstripe -E 1024K -L mdt -E EOF $dom ||
error "Can't create DoM + OST striping"
size_tmp=2031616 # must be a multiple of PAGE_SIZE=65536 on ARM
dd if=/dev/urandom of=$tmp bs=1024 count=$((size_tmp / 1024))
# check also direct IO along write
dd if=$tmp of=$dom bs=65536 count=$((size_tmp / 65536)) oflag=direct
sync
cmp $tmp $dom || error "file data is different"
[ $(stat -c%s $dom) == $size_tmp ] ||
error "bad size after write: $(stat -c%s $dom) != $size_tmp"
rm $dom $tmp
return 0
}
run_test 270a "DoM: basic functionality tests"
test_270b() {
[ $MDS1_VERSION -lt $(version_code 2.10.55) ] &&
skip "Need MDS version at least 2.10.55"
local dom=$DIR/$tdir/dom_file
local max_size=1048576
mkdir -p $DIR/$tdir
$LFS setstripe -E $max_size -L mdt $dom
# truncate over the limit
$TRUNCATE $dom $(($max_size + 1)) &&
error "successful truncate over the maximum size"
# write over the limit
dd if=/dev/zero of=$dom bs=$max_size seek=1 count=1 &&
error "successful write over the maximum size"
# append over the limit
dd if=/dev/zero of=$dom bs=$(($max_size - 3)) count=1
echo "12345" >> $dom && error "successful append over the maximum size"
rm $dom
return 0
}
run_test 270b "DoM: maximum size overflow checks for DoM-only file"
test_270c() {
[ $MDS1_VERSION -lt $(version_code 2.10.55) ] &&
skip "Need MDS version at least 2.10.55"
mkdir -p $DIR/$tdir
$LFS setstripe -E 1024K -L mdt $DIR/$tdir
# check files inherit DoM EA
touch $DIR/$tdir/first
[ $($LFS getstripe -L $DIR/$tdir/first) == "mdt" ] ||
error "bad pattern"
[ $($LFS getstripe -c $DIR/$tdir/first) == 0 ] ||
error "bad stripe count"
[ $($LFS getstripe -S $DIR/$tdir/first) == 1048576 ] ||
error "bad stripe size"
# check directory inherits DoM EA and uses it as default
mkdir $DIR/$tdir/subdir
touch $DIR/$tdir/subdir/second
[ $($LFS getstripe -L $DIR/$tdir/subdir/second) == "mdt" ] ||
error "bad pattern in sub-directory"
[ $($LFS getstripe -c $DIR/$tdir/subdir/second) == 0 ] ||
error "bad stripe count in sub-directory"
[ $($LFS getstripe -S $DIR/$tdir/subdir/second) == 1048576 ] ||
error "bad stripe size in sub-directory"
return 0
}
run_test 270c "DoM: DoM EA inheritance tests"
test_270d() {
[ $MDS1_VERSION -lt $(version_code 2.10.55) ] &&
skip "Need MDS version at least 2.10.55"
mkdir -p $DIR/$tdir
$LFS setstripe -E 1024K -L mdt $DIR/$tdir
# inherit default DoM striping
mkdir $DIR/$tdir/subdir
touch $DIR/$tdir/subdir/f1
# change default directory striping
$LFS setstripe -c 1 $DIR/$tdir/subdir
touch $DIR/$tdir/subdir/f2
[ $($LFS getstripe -c $DIR/$tdir/subdir/f2) == 1 ] ||
error "wrong default striping in file 2"
[ $($LFS getstripe -L $DIR/$tdir/subdir/f2) == "raid0" ] ||
error "bad pattern in file 2"
return 0
}
run_test 270d "DoM: change striping from DoM to RAID0"
test_270e() {
[ $MDS1_VERSION -lt $(version_code 2.10.55) ] &&
skip "Need MDS version at least 2.10.55"
mkdir -p $DIR/$tdir/dom
mkdir -p $DIR/$tdir/norm
DOMFILES=20
NORMFILES=10
$LFS setstripe -E 1M -L mdt $DIR/$tdir/dom
$LFS setstripe -i 0 -S 2M $DIR/$tdir/norm
createmany -o $DIR/$tdir/dom/dom- $DOMFILES
createmany -o $DIR/$tdir/norm/norm- $NORMFILES
# find DoM files by layout
NUM=$($LFS find -L mdt -type f $DIR/$tdir 2>/dev/null | wc -l)
[ $NUM -eq $DOMFILES ] ||
error "lfs find -L: found $NUM, expected $DOMFILES"
echo "Test 1: lfs find 20 DOM files by layout: OK"
# there should be 1 dir with default DOM striping
NUM=$($LFS find -L mdt -type d $DIR/$tdir 2>/dev/null | wc -l)
[ $NUM -eq 1 ] ||
error "lfs find -L: found $NUM, expected 1 dir"
echo "Test 2: lfs find 1 DOM dir by layout: OK"
# find DoM files by stripe size
NUM=$($LFS find -S -1200K -type f $DIR/$tdir 2>/dev/null | wc -l)
[ $NUM -eq $DOMFILES ] ||
error "lfs find -S: found $NUM, expected $DOMFILES"
echo "Test 4: lfs find 20 DOM files by stripe size: OK"
# find files by stripe offset except DoM files
NUM=$($LFS find -i 0 -type f $DIR/$tdir 2>/dev/null | wc -l)
[ $NUM -eq $NORMFILES ] ||
error "lfs find -i: found $NUM, expected $NORMFILES"
echo "Test 5: lfs find no DOM files by stripe index: OK"
return 0
}
run_test 270e "DoM: lfs find with DoM files test"
test_270f() {
[ $MDS1_VERSION -lt $(version_code 2.10.55) ] &&
skip "Need MDS version at least 2.10.55"
local mdtname=${FSNAME}-MDT0000-mdtlov
local dom=$DIR/$tdir/dom_file
local dom_limit_saved=$(do_facet mds1 $LCTL get_param -n \
lod.$mdtname.dom_stripesize)
local dom_limit=131072
do_facet mds1 $LCTL set_param -n lod.$mdtname.dom_stripesize=$dom_limit
local dom_current=$(do_facet mds1 $LCTL get_param -n \
lod.$mdtname.dom_stripesize)
[ ${dom_limit} -eq ${dom_current} ] ||
error "Cannot change per-MDT DoM stripe limit to $dom_limit"
$LFS mkdir -i 0 -c 1 $DIR/$tdir
$LFS setstripe -d $DIR/$tdir
$LFS setstripe -E $dom_limit -L mdt $DIR/$tdir ||
error "Can't set directory default striping"
# exceed maximum stripe size
$LFS setstripe -E $((dom_limit * 2)) -L mdt $dom ||
error "Can't create file with $((dom_limit * 2)) DoM stripe"
[ $($LFS getstripe -S $dom) -eq $((dom_limit * 2)) ] &&
error "Able to create DoM component size more than LOD limit"
do_facet mds1 $LCTL set_param -n lod.$mdtname.dom_stripesize=0
dom_current=$(do_facet mds1 $LCTL get_param -n \
lod.$mdtname.dom_stripesize)
[ 0 -eq ${dom_current} ] ||
error "Can't set zero DoM stripe limit"
rm $dom
# attempt to create DoM file on server with disabled DoM should
# remove DoM entry from layout and be succeed
$LFS setstripe -E $dom_limit -L mdt -E -1 $dom ||
error "Can't create DoM file (DoM is disabled)"
[ $($LFS getstripe -L $dom) == "mdt" ] &&
error "File has DoM component while DoM is disabled"
rm $dom
# attempt to create DoM file with only DoM stripe should return error
$LFS setstripe -E $dom_limit -L mdt $dom &&
error "Able to create DoM-only file while DoM is disabled"
# too low values to be aligned with smallest stripe size 64K
do_facet mds1 $LCTL set_param -n lod.$mdtname.dom_stripesize=30000
dom_current=$(do_facet mds1 $LCTL get_param -n \
lod.$mdtname.dom_stripesize)
[ 30000 -eq ${dom_current} ] &&
error "Can set too small DoM stripe limit"
# 64K is a minimal stripe size in Lustre, expect limit of that size
[ 65536 -eq ${dom_current} ] ||
error "Limit is not set to 64K but ${dom_current}"
do_facet mds1 $LCTL set_param -n lod.$mdtname.dom_stripesize=2147483648
dom_current=$(do_facet mds1 $LCTL get_param -n \
lod.$mdtname.dom_stripesize)
echo $dom_current
[ 2147483648 -eq ${dom_current} ] &&
error "Can set too large DoM stripe limit"
do_facet mds1 $LCTL set_param -n \
lod.$mdtname.dom_stripesize=$((dom_limit * 2))
$LFS setstripe -E $((dom_limit * 2)) -L mdt $dom ||
error "Can't create DoM component size after limit change"
do_facet mds1 $LCTL set_param -n \
lod.$mdtname.dom_stripesize=$((dom_limit / 2))
$LFS setstripe -E $dom_limit -L mdt ${dom}_big ||
error "Can't create DoM file after limit decrease"
[ $($LFS getstripe -S ${dom}_big) -eq $((dom_limit / 2)) ] ||
error "Can create big DoM component after limit decrease"
touch ${dom}_def ||
error "Can't create file with old default layout"
do_facet mds1 $LCTL set_param -n lod.*.dom_stripesize=$dom_limit_saved
return 0
}
run_test 270f "DoM: maximum DoM stripe size checks"
test_270g() {
[ $MDS1_VERSION -ge $(version_code 2.13.52) ] ||
skip "Need MDS version at least 2.13.52"
local dom=$DIR/$tdir/$tfile
$LFS mkdir -i 0 -c 1 $DIR/$tdir
local lodname=${FSNAME}-MDT0000-mdtlov
local save="$TMP/$TESTSUITE-$TESTNAME.parameters"
save_lustre_params mds1 "lod.${lodname}.dom_stripesize_max_kb" > $save
save_lustre_params mds1 "lod.${lodname}.dom_threshold_free_mb" >> $save
stack_trap "restore_lustre_params < $save; rm -f $save" EXIT
local dom_limit=1024
local dom_threshold="50%"
$LFS setstripe -d $DIR/$tdir
$LFS setstripe -E ${dom_limit}K -L mdt $DIR/$tdir ||
error "Can't set directory default striping"
do_facet mds1 $LCTL set_param -n \
lod.${lodname}.dom_stripesize_max_kb=$dom_limit
# set 0 threshold and create DOM file to change tunable stripesize
do_facet mds1 $LCTL set_param -n lod.${lodname}.dom_threshold_free_mb=0
$LFS setstripe -E ${dom_limit}K -L mdt -E -1 $dom ||
error "Failed to create $dom file"
# now tunable dom_cur_stripesize should reach maximum
local dom_current=$(do_facet mds1 $LCTL get_param -n \
lod.${lodname}.dom_stripesize_cur_kb)
[[ $dom_current == $dom_limit ]] ||
error "Current DOM stripesize is not maximum"
rm $dom
# set threshold for further tests
do_facet mds1 $LCTL set_param -n \
lod.${lodname}.dom_threshold_free_mb=$dom_threshold
echo "DOM threshold is $dom_threshold free space"
local dom_def
local dom_set
# Spoof bfree to exceed threshold
#define OBD_FAIL_MDS_STATFS_SPOOF 0x168
do_facet mds1 $LCTL set_param -n fail_loc=0x0168
for spfree in 40 20 0 15 30 55; do
do_facet mds1 $LCTL set_param -n fail_val=$spfree
$LFS setstripe -E ${dom_limit}K -L mdt -E -1 $dom ||
error "Failed to create $dom file"
dom_def=$(do_facet mds1 $LCTL get_param -n \
lod.${lodname}.dom_stripesize_cur_kb)
echo "Free space: ${spfree}%, default DOM stripe: ${dom_def}K"
[[ $dom_def != $dom_current ]] ||
error "Default stripe size was not changed"
if (( spfree > 0 )) ; then
dom_set=$($LFS getstripe -S $dom)
(( dom_set == dom_def * 1024 )) ||
error "DOM component size is still old"
else
[[ $($LFS getstripe -L $dom) != "mdt" ]] ||
error "DoM component is set with no free space"
fi
rm $dom
dom_current=$dom_def
done
}
run_test 270g "DoM: default DoM stripe size depends on free space"
test_270h() {
[[ $MDS1_VERSION -ge $(version_code 2.13.53) ]] ||
skip "Need MDS version at least 2.13.53"
local mdtname=${FSNAME}-MDT0000-mdtlov
local dom=$DIR/$tdir/$tfile
local save="$TMP/$TESTSUITE-$TESTNAME.parameters"
save_lustre_params mds1 "lod.*.dom_stripesize" > $save
stack_trap "restore_lustre_params < $save; rm -f $save" EXIT
$LFS mkdir -i 0 -c 1 $DIR/$tdir
$LFS setstripe -E 1M -c1 -E -1 -c2 ${dom}_1 ||
error "can't create OST file"
# mirrored file with DOM entry in the second mirror
$LFS mirror extend -N -E 1M -L mdt -E eof -c2 ${dom}_1 ||
error "can't create mirror with DoM component"
do_facet mds1 $LCTL set_param -n lod.$mdtname.dom_stripesize=0
# DOM component in the middle and has other enries in the same mirror,
# should succeed but lost DoM component
local fid1=$($LFS path2fid ${dom}_1)
$LFS setstripe --copy=$fid1 $dom ||
error "Can't create file from OST|DOM mirror layout"
# check new file has no DoM layout after all
[[ $($LFS getstripe -L $dom) != "mdt" ]] ||
error "File has DoM component while DoM is disabled"
}
run_test 270h "DoM: DoM stripe removal when disabled on server"
test_270i() {
(( $MDS1_VERSION >= $(version_code 2.14.54) )) ||
skip "Need MDS version at least 2.14.54"
mkdir $DIR/$tdir
# DoM with plain layout
$LFS setstripe -L mdt -S 128k -c -1 $DIR/$tdir &&
error "default plain layout with DoM must fail"
$LFS setstripe -L mdt -S 128k -c -1 $DIR/$tdir/$tfile &&
error "setstripe plain file layout with DoM must fail"
$LFS setstripe -E 1M -L mdt -S 128k -c -1 -E eof $DIR/$tdir &&
error "default DoM layout with bad striping must fail"
$LFS setstripe -E 1M -L mdt -S 128k -c -1 -E eof $DIR/$tdir/$tfile &&
error "setstripe to DoM layout with bad striping must fail"
return 0
}
run_test 270i "DoM: setting invalid DoM striping should fail"
test_270j() {
(( $MDS1_VERSION >= $(version_code 2.15.55.203) )) ||
skip "Need MDS version at least 2.15.55.203"
local dom=$DIR/$tdir/$tfile
local odv
local ndv
mkdir -p $DIR/$tdir
$LFS setstripe -E 1M -L mdt -E -1 -c1 $dom
stack_trap "rm -f $dom"
odv=$($LFS data_version $dom)
chmod 666 $dom
mv $dom ${dom}_moved
link ${dom}_moved $dom
setfattr -n user.attrx -v "some_attr" $dom
ndv=$($LFS data_version $dom)
(( $ndv == $odv )) ||
error "data version was changed by metadata operations"
dd if=/dev/urandom of=$dom bs=1M count=1 ||
error "failed to write data into $dom"
cancel_lru_locks mdc
ndv=$($LFS data_version $dom)
(( $ndv != $odv )) ||
error "data version wasn't changed on write"
odv=$ndv
$TRUNCATE $dom 1000 || error "failed to truncate $dom"
ndv=$($LFS data_version $dom)
(( $ndv != $odv )) ||
error "data version wasn't changed on truncate down"
odv=$ndv
$TRUNCATE $dom 25000
ndv=$($LFS data_version $dom)
(( $ndv != $odv )) ||
error "data version wasn't changed on truncate up"
# check also fallocate for ldiskfs
if [[ "$mds1_FSTYPE" == ldiskfs ]]; then
odv=$ndv
fallocate -l 1048576 $dom
ndv=$($LFS data_version $dom)
(( $ndv != $odv )) ||
error "data version wasn't changed on fallocate"
odv=$ndv
fallocate -p --offset 4096 -l 4096 $dom
ndv=$($LFS data_version $dom)
(( $ndv != $odv )) ||
error "data version wasn't changed on fallocate punch"
fi
}
run_test 270j "DoM migration: DOM file to the OST-striped file (plain)"
test_271a() {
[ $MDS1_VERSION -lt $(version_code 2.10.55) ] &&
skip "Need MDS version at least 2.10.55"
local dom=$DIR/$tdir/dom
mkdir -p $DIR/$tdir
$LFS setstripe -E 1024K -L mdt $dom
lctl set_param -n mdc.*.stats=clear
dd if=/dev/zero of=$dom bs=4096 count=1 || return 1
cat $dom > /dev/null
local reads=$(lctl get_param -n mdc.*.stats | grep -c ost_read)
[ $reads -eq 0 ] || error "Unexpected $reads READ RPCs"
ls $dom
rm -f $dom
}
run_test 271a "DoM: data is cached for read after write"
test_271b() {
[ $MDS1_VERSION -lt $(version_code 2.10.55) ] &&
skip "Need MDS version at least 2.10.55"
local dom=$DIR/$tdir/dom
mkdir -p $DIR/$tdir
$LFS setstripe -E 1024K -L mdt -E EOF $dom
lctl set_param -n mdc.*.stats=clear
dd if=/dev/zero of=$dom bs=4096 count=1 || return 1
cancel_lru_locks mdc
$CHECKSTAT -t file -s 4096 $dom || error "stat #1 fails"
# second stat to check size is cached on client
$CHECKSTAT -t file -s 4096 $dom || error "stat #2 fails"
local gls=$(lctl get_param -n mdc.*.stats | grep -c ldlm_glimpse)
[ $gls -eq 0 ] || error "Unexpected $gls glimpse RPCs"
rm -f $dom
}
run_test 271b "DoM: no glimpse RPC for stat (DoM only file)"
test_271ba() {
[ $MDS1_VERSION -lt $(version_code 2.10.55) ] &&
skip "Need MDS version at least 2.10.55"
local dom=$DIR/$tdir/dom
mkdir -p $DIR/$tdir
$LFS setstripe -E 1024K -L mdt -E EOF $dom
lctl set_param -n mdc.*.stats=clear
lctl set_param -n osc.*.stats=clear
# Hybrid switches to DIO, so does not hold the required lock to skip
# the glimpse, so we disable it here...
local hybrid=$($LCTL get_param -n llite.*.hybrid_io)
$LCTL set_param llite.*.hybrid_io=0
stack_trap "$LCTL set_param -n llite.*.hybrid_io=$hybrid" EXIT
dd if=/dev/zero of=$dom bs=2048K count=1 || return 1
cancel_lru_locks mdc
$CHECKSTAT -t file -s 2097152 $dom || error "stat"
# second stat to check size is cached on client
$CHECKSTAT -t file -s 2097152 $dom || error "stat"
local gls=$(lctl get_param -n mdc.*.stats | grep -c ldlm_glimpse)
[ $gls == 0 ] || error "Unexpected $gls glimpse RPCs"
local gls=$(lctl get_param -n osc.*.stats | grep -c ldlm_glimpse)
[ $gls == 0 ] || error "Unexpected $gls OSC glimpse RPCs"
rm -f $dom
}
run_test 271ba "DoM: no glimpse RPC for stat (combined file)"
get_mdc_stats() {
local mdtidx=$1
local param=$2
local mdt=$(mdtname_from_index $mdtidx)
if [ -z $param ]; then
$LCTL get_param -n mdc.$mdt*.stats
else
$LCTL get_param -n mdc.$mdt*.stats | awk "/$param/"'{print $2}'
fi
}
test_271c() {
(( $MDS1_VERSION >= $(version_code v2_10_55_0-74-g2f09984f3f) )) ||
skip "Need MDS >= 2.10.55.74 for combined DoM lock bits"
local dom=$DIR/$tdir/dom
mkdir -p $DIR/$tdir
$LFS setstripe -E 1024K -L mdt $DIR/$tdir ||
error "unable to set DoM layout on $DIR/$tdir"
local mdtidx=$($LFS getstripe -m $DIR/$tdir)
local facet=mds$((mdtidx + 1))
cancel_lru_locks mdc
do_facet $facet lctl set_param -n mdt.*.dom_lock=0
createmany -o $dom 1000
lctl set_param -n mdc.*.stats=clear
smalliomany -w $dom 1000 200
get_mdc_stats $mdtidx
local enq=$(get_mdc_stats $mdtidx ldlm_ibits_enqueue)
# Each file has 1 open, 1 IO enqueues, total 2000
# but now we have also +1 getxattr for security.capability, total 3000
[ $enq -ge 2000 ] || error "Too few enqueues $enq, expected > 2000"
unlinkmany $dom 1000
cancel_lru_locks mdc
do_facet $facet lctl set_param -n mdt.*.dom_lock=1
createmany -o $dom 1000
lctl set_param -n mdc.*.stats=clear
smalliomany -w $dom 1000 200
local enq_2=$(get_mdc_stats $mdtidx ldlm_ibits_enqueue)
# Expect to see reduced amount of RPCs by 1000 due to single enqueue
# for OPEN and IO lock.
[ $((enq - enq_2)) -ge 1000 ] ||
error "Too many enqueues $enq_2, expected about $((enq - 1000))"
unlinkmany $dom 1000
return 0
}
run_test 271c "DoM: IO lock at open saves enqueue RPCs"
test_271d() {
(( $MDS1_VERSION >= $(version_code v2_11_53_0-56-g13372d6c24) )) ||
skip "Need MDS version at least 2.11.53"
local dom=$DIR/$tdir/dom
local tmp=$TMP/$tfile
stack_trap "rm -f $tmp"
mkdir -p $DIR/$tdir
$LFS setstripe -E 1024K -L mdt $DIR/$tdir
local mdtidx=$($LFS getstripe --mdt-index $DIR/$tdir)
cancel_lru_locks mdc
dd if=/dev/urandom of=$tmp bs=1000 count=1
dd if=$tmp of=$dom bs=1000 count=1
cancel_lru_locks mdc
cat /etc/hosts >> $tmp
lctl set_param -n mdc.*.stats=clear
# append data to the same file it should update local page
echo "Append to the same page"
cat /etc/hosts >> $dom
local stats=($(get_mdc_stats $mdtidx 'req_wait|req_active|ost_read'))
local rw=${stats[0]}
local ra=${stats[1]}
local num=${stats[2]}
[[ -z "$num" ]] || error "$num READ RPC occured"
(( $ra == $rw )) || {
get_mdc_stats $mdtidx
echo "req_active: $ra"
echo "req_waittime: $rw"
error "$((ra - rw)) resend occured on append"
}
echo "... DONE"
# compare content
cmp $tmp $dom || error "file miscompare"
cancel_lru_locks mdc
lctl set_param -n mdc.*.stats=clear
echo "Open and read file"
cat $dom > /dev/null
stats=($(get_mdc_stats $mdtidx 'req_wait|req_active|ost_read'))
rw=${stats[0]}
ra=${stats[1]}
num=${stats[2]}
[[ -z "$num" ]] || error "$num READ RPC occured"
(( $ra == $rw )) || {
get_mdc_stats $mdtidx
echo "req_active: $ra"
echo "req_waittime: $rw"
error "$((ra - rw)) resend occured on open+read"
}
echo "... DONE"
# compare content
cmp -bl $tmp $dom || error "file miscompare"
}
run_test 271d "DoM: read on open (1K file in reply buffer)"
test_271f() {
(( $MDS1_VERSION >= $(version_code v2_11_53_0-56-g13372d6c24) )) ||
skip "Need MDS version at least 2.11.53"
local dom=$DIR/$tdir/dom
local tmp=$TMP/$tfile
stack_trap "rm -f $tmp"
mkdir -p $DIR/$tdir
$LFS setstripe -E 1024K -L mdt $DIR/$tdir
local mdtidx=$($LFS getstripe --mdt-index $DIR/$tdir)
cancel_lru_locks mdc
dd if=/dev/urandom of=$tmp bs=265000 count=1
dd if=$tmp of=$dom bs=265000 count=1
cancel_lru_locks mdc
cat /etc/hosts >> $tmp
lctl set_param -n mdc.*.stats=clear
echo "Append to the same page"
cat /etc/hosts >> $dom
local stats=($(get_mdc_stats $mdtidx 'req_wait|req_active|ost_read'))
local rw=${stats[0]}
local ra=${stats[1]}
local num=${stats[2]}
[[ -z "$num" ]] || error "$num READ RPC occured"
(( $ra == $rw )) || {
get_mdc_stats $mdtidx
echo "req_active: $ra"
echo "req_waittime: $rw"
error "$((ra - rw)) resend occured on append"
}
echo "... DONE"
# compare content
cmp $tmp $dom || error "file miscompare"
cancel_lru_locks mdc
lctl set_param -n mdc.*.stats=clear
echo "Open and read file"
cat $dom > /dev/null
stats=($(get_mdc_stats $mdtidx 'req_wait|req_active|ost_read'))
rw=${stats[0]}
ra=${stats[1]}
num=${stats[2]}
[[ -n "$num" ]] || num=0
(( $num == 1 )) || error "expect 1 READ RPC, $num occured"
(( $ra == $rw )) || {
get_mdc_stats $mdtidx
echo "req_active: $ra"
echo "req_waittime: $rw"
error "$((ra - rw)) resend occured on open+read"
}
echo "... DONE"
# compare content
cmp -bl $tmp $dom || error "file miscompare"
}
run_test 271f "DoM: read on open (200K file and read tail)"
test_271g() {
[[ $($LCTL get_param mdc.*.import) =~ async_discard ]] ||
skip "Skipping due to old client or server version"
$LFS setstripe -E 1024K -L mdt -E EOF $DIR1/$tfile
# to get layout
$CHECKSTAT -t file $DIR1/$tfile
$MULTIOP $DIR1/$tfile Ow40960_w4096c &
MULTIOP_PID=$!
sleep 1
#define OBD_FAIL_LDLM_CANCEL_BL_CB_RACE
$LCTL set_param fail_loc=0x80000314
rm $DIR1/$tfile || error "Unlink fails"
RC=$?
kill -USR1 $MULTIOP_PID && wait $MULTIOP_PID || error "multiop failure"
[ $RC -eq 0 ] || error "Failed write to stale object"
}
run_test 271g "Discard DoM data vs client flush race"
test_272a() {
[ $MDS1_VERSION -lt $(version_code 2.11.50) ] &&
skip "Need MDS version at least 2.11.50"
local dom=$DIR/$tdir/dom
mkdir -p $DIR/$tdir
$LFS setstripe -E 256K -L mdt -E -1 -c1 $dom
dd if=/dev/urandom of=$dom bs=512K count=1 ||
error "failed to write data into $dom"
local old_md5=$(md5sum $dom)
$LFS migrate -E 256K -L mdt -E -1 -c2 $dom ||
error "failed to migrate to the same DoM component"
local new_md5=$(md5sum $dom)
[ "$old_md5" == "$new_md5" ] ||
error "md5sum differ: $old_md5, $new_md5"
[ $($LFS getstripe -c $dom) -eq 2 ] ||
error "bad final stripe count: $($LFS getstripe -c $dom) != 2"
}
run_test 272a "DoM migration: new layout with the same DOM component"
test_272b() {
[ $MDS1_VERSION -lt $(version_code 2.11.50) ] &&
skip "Need MDS version at least 2.11.50"
local dom=$DIR/$tdir/dom
mkdir -p $DIR/$tdir
$LFS setstripe -E 1M -L mdt -E -1 -c1 $dom
stack_trap "rm -rf $DIR/$tdir"
local mdtidx=$($LFS getstripe -m $dom)
local mdtname=MDT$(printf %04x $mdtidx)
local facet=mds$((mdtidx + 1))
local mdtfree1=$(do_facet $facet \
lctl get_param -n osd*.*$mdtname.kbytesfree)
dd if=/dev/urandom of=$dom bs=2M count=1 ||
error "failed to write data into $dom"
local old_md5=$(md5sum $dom)
cancel_lru_locks mdc
local mdtfree1=$(do_facet $facet \
lctl get_param -n osd*.*$mdtname.kbytesfree)
$LFS migrate -c2 $dom ||
error "failed to migrate to the new composite layout"
[[ $($LFS getstripe --component-start=0 -L $dom) != 'mdt' ]] ||
error "MDT stripe was not removed"
! getfattr -n trusted.dataver $dom &> /dev/null ||
error "$dir1 shouldn't have DATAVER EA"
cancel_lru_locks mdc
local new_md5=$(md5sum $dom)
[ "$old_md5" == "$new_md5" ] ||
error "$old_md5 != $new_md5"
# Skip free space checks with ZFS
if [ "$(facet_fstype $facet)" != "zfs" ]; then
local mdtfree2=$(do_facet $facet \
lctl get_param -n osd*.*$mdtname.kbytesfree)
[ $mdtfree2 -gt $mdtfree1 ] ||
error "MDT space is not freed after migration"
fi
return 0
}
run_test 272b "DoM migration: DOM file to the OST-striped file (plain)"
test_272c() {
[ $MDS1_VERSION -lt $(version_code 2.11.50) ] &&
skip "Need MDS version at least 2.11.50"
local dom=$DIR/$tdir/$tfile
mkdir -p $DIR/$tdir
$LFS setstripe -E 1M -L mdt -E -1 -c1 $dom
stack_trap "rm -rf $DIR/$tdir"
local mdtidx=$($LFS getstripe -m $dom)
local mdtname=MDT$(printf %04x $mdtidx)
local facet=mds$((mdtidx + 1))
dd if=/dev/urandom of=$dom bs=2M count=1 oflag=direct ||
error "failed to write data into $dom"
local old_md5=$(md5sum $dom)
cancel_lru_locks mdc
local mdtfree1=$(do_facet $facet \
lctl get_param -n osd*.*$mdtname.kbytesfree)
$LFS migrate -E 2M -c1 -E -1 -c2 $dom ||
error "failed to migrate to the new composite layout"
[[ $($LFS getstripe --component-start=0 -L $dom) != 'mdt' ]] ||
error "MDT stripe was not removed"
cancel_lru_locks mdc
local new_md5=$(md5sum $dom)
[ "$old_md5" == "$new_md5" ] ||
error "$old_md5 != $new_md5"
# Skip free space checks with ZFS
if [ "$(facet_fstype $facet)" != "zfs" ]; then
local mdtfree2=$(do_facet $facet \
lctl get_param -n osd*.*$mdtname.kbytesfree)
[ $mdtfree2 -gt $mdtfree1 ] ||
error "MDS space is not freed after migration"
fi
return 0
}
run_test 272c "DoM migration: DOM file to the OST-striped file (composite)"
test_272d() {
[ $MDS1_VERSION -lt $(version_code 2.12.55) ] &&
skip "Need MDS version at least 2.12.55"
local dom=$DIR/$tdir/$tfile
mkdir -p $DIR/$tdir
$LFS setstripe -E 1M -L mdt -E -1 -c1 $dom
stack_trap "rm -f $dom"
local mdtidx=$($LFS getstripe -m $dom)
local mdtname=MDT$(printf %04x $mdtidx)
local facet=mds$((mdtidx + 1))
dd if=/dev/urandom of=$dom bs=2M count=1 oflag=direct ||
error "failed to write data into $dom"
local old_md5=$(md5sum $dom)
cancel_lru_locks mdc
local mdtfree1=$(do_facet $facet \
lctl get_param -n osd*.*$mdtname.kbytesfree)
$LFS mirror extend -N -E 2M -c1 -E -1 -c2 $dom ||
error "failed mirroring to the new composite layout"
$LFS mirror resync $dom ||
error "failed mirror resync"
$LFS mirror split --mirror-id 1 -d $dom ||
error "failed mirror split"
[ $($LFS getstripe -L $dom) != 'mdt' ] ||
error "MDT stripe was not removed"
cancel_lru_locks mdc
local new_md5=$(md5sum $dom)
[ "$old_md5" == "$new_md5" ] ||
error "$old_md5 != $new_md5"
# Skip free space checks with ZFS
if [ "$(facet_fstype $facet)" != "zfs" ]; then
local mdtfree2=$(do_facet $facet \
lctl get_param -n osd*.*$mdtname.kbytesfree)
[ $mdtfree2 -gt $mdtfree1 ] ||
error "MDS space is not freed after DOM mirror deletion"
fi
return 0
}
run_test 272d "DoM mirroring: OST-striped mirror to DOM file"
test_272e() {
[ $MDS1_VERSION -lt $(version_code 2.12.55) ] &&
skip "Need MDS version at least 2.12.55"
local dom=$DIR/$tdir/$tfile
mkdir -p $DIR/$tdir
$LFS setstripe -c 2 $dom
stack_trap "rm -f $dom"
dd if=/dev/urandom of=$dom bs=2M count=1 oflag=direct ||
error "failed to write data into $dom"
local old_md5=$(md5sum $dom)
cancel_lru_locks
$LFS mirror extend -N -E 1M -L mdt -E eof -c2 $dom ||
error "failed mirroring to the DOM layout"
$LFS mirror resync $dom ||
error "failed mirror resync"
$LFS mirror split --mirror-id 1 -d $dom ||
error "failed mirror split"
[[ $($LFS getstripe -L --component-start=0 $dom) == 'mdt' ]] ||
error "MDT stripe wasn't set"
cancel_lru_locks
local new_md5=$(md5sum $dom)
[ "$old_md5" == "$new_md5" ] ||
error "$old_md5 != $new_md5"
return 0
}
run_test 272e "DoM mirroring: DOM mirror to the OST-striped file"
test_272f() {
[ $MDS1_VERSION -lt $(version_code 2.12.55) ] &&
skip "Need MDS version at least 2.12.55"
local dom=$DIR/$tdir/$tfile
mkdir -p $DIR/$tdir
$LFS setstripe -c 2 $dom
stack_trap "rm -f $dom"
dd if=/dev/urandom of=$dom bs=2M count=1 oflag=direct ||
error "failed to write data into $dom"
local old_md5=$(md5sum $dom)
cancel_lru_locks
$LFS migrate -E 1M -L mdt -E eof -c2 -v $dom ||
error "failed migrating to the DOM file"
[[ $($LFS getstripe -L --component-start=0 $dom) == 'mdt' ]] ||
error "MDT stripe wasn't set"
cancel_lru_locks
local new_md5=$(md5sum $dom)
[ "$old_md5" != "$new_md5" ] &&
error "$old_md5 != $new_md5"
return 0
}
run_test 272f "DoM migration: OST-striped file to DOM file"
test_273a() {
[ $MDS1_VERSION -lt $(version_code 2.11.50) ] &&
skip "Need MDS version at least 2.11.50"
# Layout swap cannot be done if either file has DOM component,
# this will never be supported, migration should be used instead
local dom=$DIR/$tdir/$tfile
mkdir -p $DIR/$tdir
$LFS setstripe -c2 ${dom}_plain
$LFS setstripe -E 1M -L mdt -E -1 -c2 ${dom}_dom
$LFS swap_layouts ${dom}_plain ${dom}_dom &&
error "can swap layout with DoM component"
$LFS swap_layouts ${dom}_dom ${dom}_plain &&
error "can swap layout with DoM component"
$LFS setstripe -E 1M -c1 -E -1 -c2 ${dom}_comp
$LFS swap_layouts ${dom}_comp ${dom}_dom &&
error "can swap layout with DoM component"
$LFS swap_layouts ${dom}_dom ${dom}_comp &&
error "can swap layout with DoM component"
return 0
}
run_test 273a "DoM: layout swapping should fail with DOM"
test_273b() {
mkdir -p $DIR/$tdir
$LFS setstripe -E 1M -L mdt -E -1 -c -1 $DIR/$tdir
#define OBD_FAIL_MDS_COMMITRW_DELAY 0x16b
do_facet mds1 $LCTL set_param fail_loc=0x8000016b fail_val=2
$MULTIOP $DIR/$tdir/$tfile Ouw2097152c
}
run_test 273b "DoM: race writeback and object destroy"
test_273c() {
mkdir -p $DIR/$tdir
$LFS setstripe -E 1M -E-1 -c-1 $DIR/$tdir
#define OBD_FAIL_OFD_COMMITRW_DELAY 0x1e1
do_facet ost1 $LCTL set_param fail_loc=0x800001e1 fail_val=2
$MULTIOP $DIR/$tdir/$tfile Ouw2097152c
}
run_test 273c "race writeback and object destroy"
test_275() {
remote_ost_nodsh && skip "remote OST with nodsh"
[ $OST1_VERSION -lt $(version_code 2.10.57) ] &&
skip "Need OST version >= 2.10.57"
local file=$DIR/$tfile
local osts=$(osts_nodes)
dd if=/dev/urandom of=$file bs=1M count=2 ||
error "failed to create a file"
stack_trap "rm -f $file"
cancel_lru_locks osc
#lock 1
dd if=$file of=/dev/null bs=1M count=1 iflag=direct ||
error "failed to read a file"
#define OBD_FAIL_LDLM_PAUSE_CANCEL2 0x31f
$LCTL set_param fail_loc=0x8000031f
cancel_lru_locks osc &
sleep 1
#define OBD_FAIL_LDLM_PROLONG_PAUSE 0x32b
do_nodes $osts $LCTL set_param fail_loc=0x8000032b
#IO takes another lock, but matches the PENDING one
#and places it to the IO RPC
dd if=$file of=/dev/null bs=1M count=1 iflag=direct ||
error "failed to read a file with PENDING lock"
}
run_test 275 "Read on a canceled duplicate lock"
test_276() {
(( $OST1_VERSION > $(version_code v2_10_58_0-52-g4f40429775) )) ||
skip "need OSS > 2.10.58.52 to avoid statfs race"
remote_ost_nodsh && skip "remote OST with nodsh"
local pid
do_facet ost1 "(while true; do \
$LCTL get_param obdfilter.*.filesfree > /dev/null 2>&1; \
done) & pid=\\\$!; echo \\\$pid > $TMP/sanity_276_pid" &
pid=$!
local end=$(($SECONDS + 120))
[[ "$SLOW" == "yes" ]] && $((end += 600))
while (($SECONDS < end)); do
stop ost1
start ost1 $(ostdevname 1) $OST_MOUNT_OPTS
done
kill -9 $pid
do_facet ost1 "pid=\\\$(cat $TMP/sanity_276_pid); kill -9 \\\$pid; \
rm $TMP/sanity_276_pid"
}
run_test 276 "Race between mount and obd_statfs"
test_277() {
$LCTL set_param ldlm.namespaces.*.lru_size=0
dd if=/dev/zero of=$DIR/$tfile bs=1M count=1
local cached_mb=$($LCTL get_param llite.*.max_cached_mb |
awk '/^used_mb/ { print $2 }')
[ $cached_mb -eq 1 ] || error "expected mb 1 got $cached_mb"
dd if=/dev/zero of=$DIR/$tfile bs=1M count=1 \
oflag=direct conv=notrunc
stack_trap "rm -f $DIR/$tfile"
cached_mb=$($LCTL get_param llite.*.max_cached_mb |
awk '/^used_mb/ { print $2 }')
[ $cached_mb -eq 0 ] || error "expected mb 0 got $cached_mb"
}
run_test 277 "Direct IO shall drop page cache"
test_278() {
[ $PARALLEL == "yes" ] && skip "skip parallel run" && return
[ $MDSCOUNT -lt 2 ] && skip "needs >= 2 MDTs" && return
[[ "$(facet_host mds1)" != "$(facet_host mds2)" ]] &&
skip "needs the same host for mdt1 mdt2" && return
local pid1
local pid2
#define OBD_FAIL_OBD_STOP_MDS_RACE 0x60b
do_facet mds2 $LCTL set_param fail_loc=0x8000060c
stop mds2 &
pid2=$!
stop mds1
echo "Starting MDTs"
start mds1 $(mdsdevname 1) $MDS_MOUNT_OPTS
wait $pid2
#For the error assertion will happen. lu_env_get_key(..., &mdt_thread_key)
#will return NULL
do_facet mds2 $LCTL set_param fail_loc=0
start mds2 $(mdsdevname 2) $MDS_MOUNT_OPTS
wait_recovery_complete mds2
}
run_test 278 "Race starting MDS between MDTs stop/start"
test_280() {
[ $MGS_VERSION -lt $(version_code 2.13.52) ] &&
skip "Need MGS version at least 2.13.52"
[ $PARALLEL == "yes" ] && skip "skip parallel run"
combined_mgs_mds || skip "needs combined MGS/MDT"
umount_client $MOUNT
#define OBD_FAIL_MDS_LLOG_UMOUNT_RACE 0x15e
do_facet mgs $LCTL set_param fail_loc=0x8000015e fail_val=0
mount_client $MOUNT &
sleep 1
stop mgs || error "stop mgs failed"
#for a race mgs would crash
start mgs $(mgsdevname) $MGS_MOUNT_OPTS || error "start mgs failed"
# make sure we unmount client before remounting
wait
umount_client $MOUNT
mount_client $MOUNT || error "mount client failed"
}
run_test 280 "Race between MGS umount and client llog processing"
cleanup_test_300() {
trap 0
umask $SAVE_UMASK
}
test_striped_dir() {
local mdt_index=$1
local stripe_count=$2
local overstriping=$3
local stripe_index
local getstripe_count
mkdir -p $DIR/$tdir
SAVE_UMASK=$(umask)
trap cleanup_test_300 RETURN EXIT
if [ -z $overstriping ]; then
$LFS setdirstripe -i $mdt_index -c $stripe_count -H all_char \
-o 755 $DIR/$tdir/striped_dir ||
error "set striped dir error"
else
$LFS setdirstripe -i $mdt_index -C $stripe_count -H all_char \
-o 755 $DIR/$tdir/striped_dir ||
error "set striped dir error"
fi
local mode=$(stat -c%a $DIR/$tdir/striped_dir)
[ "$mode" = "755" ] || error "expect 755 got $mode"
$LFS getdirstripe $DIR/$tdir/striped_dir > /dev/null 2>&1 ||
error "getdirstripe failed"
getstripe_count=$($LFS getdirstripe -c $DIR/$tdir/striped_dir)
if [ "$getstripe_count" != "$stripe_count" ]; then
error "1:stripe_count is $getstripe_count, expect $stripe_count"
fi
getstripe_count=$($LFS getdirstripe -T $DIR/$tdir/striped_dir)
if [ "$getstripe_count" != "$stripe_count" ]; then
error "2:stripe_count is $getstripe_count, expect $stripe_count"
fi
stripe_index=$($LFS getdirstripe -i $DIR/$tdir/striped_dir)
if [ "$stripe_index" != "$mdt_index" ]; then
error "stripe_index is $stripe_index, expect $mdt_index"
fi
[ $(stat -c%h $DIR/$tdir/striped_dir) == '2' ] ||
error "nlink error after create striped dir"
mkdir $DIR/$tdir/striped_dir/a
mkdir $DIR/$tdir/striped_dir/b
stat $DIR/$tdir/striped_dir/a ||
error "create dir under striped dir failed"
stat $DIR/$tdir/striped_dir/b ||
error "create dir under striped dir failed"
[ $(stat -c%h $DIR/$tdir/striped_dir) == '4' ] ||
error "nlink error after mkdir"
rmdir $DIR/$tdir/striped_dir/a
[ $(stat -c%h $DIR/$tdir/striped_dir) == '3' ] ||
error "nlink error after rmdir"
rmdir $DIR/$tdir/striped_dir/b
[ $(stat -c%h $DIR/$tdir/striped_dir) == '2' ] ||
error "nlink error after rmdir"
chattr +i $DIR/$tdir/striped_dir
createmany -o $DIR/$tdir/striped_dir/f 10 &&
error "immutable flags not working under striped dir!"
chattr -i $DIR/$tdir/striped_dir
rmdir $DIR/$tdir/striped_dir ||
error "rmdir striped dir error"
cleanup_test_300
true
}
test_300a() {
[ $MDS1_VERSION -lt $(version_code 2.7.0) ] &&
skip "skipped for lustre < 2.7.0"
[ $PARALLEL == "yes" ] && skip "skip parallel run"
[ $MDSCOUNT -lt 2 ] && skip_env "needs >= 2 MDTs"
test_striped_dir 0 2 || error "failed on striped dir on MDT0"
test_striped_dir 1 2 || error "failed on striped dir on MDT0"
}
run_test 300a "basic striped dir sanity test"
test_300b() {
[ $MDS1_VERSION -lt $(version_code 2.7.0) ] &&
skip "skipped for lustre < 2.7.0"
[ $PARALLEL == "yes" ] && skip "skip parallel run"
[ $MDSCOUNT -lt 2 ] && skip_env "needs >= 2 MDTs"
local i
local mtime1
local mtime2
local mtime3
test_mkdir $DIR/$tdir || error "mkdir fail"
$LFS setdirstripe -i 0 -c 2 -H all_char $DIR/$tdir/striped_dir ||
error "set striped dir error"
for i in {0..9}; do
mtime1=$(stat -c %Y $DIR/$tdir/striped_dir)
sleep 1
touch $DIR/$tdir/striped_dir/file_$i || error "touch error $i"
mtime2=$(stat -c %Y $DIR/$tdir/striped_dir)
[ $mtime1 -eq $mtime2 ] && error "mtime unchanged after create"
sleep 1
rm -f $DIR/$tdir/striped_dir/file_$i || error "unlink error $i"
mtime3=$(stat -c %Y $DIR/$tdir/striped_dir)
[ $mtime2 -eq $mtime3 ] && error "mtime unchanged after unlink"
done
true
}
run_test 300b "check ctime/mtime for striped dir"
test_300c() {
[ $MDS1_VERSION -lt $(version_code 2.7.0) ] &&
skip "skipped for lustre < 2.7.0"
[ $PARALLEL == "yes" ] && skip "skip parallel run"
[ $MDSCOUNT -lt 2 ] && skip_env "needs >= 2 MDTs"
local file_count
mkdir_on_mdt0 $DIR/$tdir
$LFS setdirstripe -i 0 -c 2 $DIR/$tdir/striped_dir ||
error "set striped dir error"
chown $RUNAS_ID:$RUNAS_GID $DIR/$tdir/striped_dir ||
error "chown striped dir failed"
$RUNAS createmany -o $DIR/$tdir/striped_dir/f 5000 ||
error "create 5k files failed"
file_count=$(ls $DIR/$tdir/striped_dir | wc -l)
[ "$file_count" = 5000 ] || error "file count $file_count != 5000"
rm -rf $DIR/$tdir
}
run_test 300c "chown && check ls under striped directory"
test_300d() {
[ $MDS1_VERSION -lt $(version_code 2.7.0) ] &&
skip "skipped for lustre < 2.7.0"
[ $PARALLEL == "yes" ] && skip "skip parallel run"
[ $MDSCOUNT -lt 2 ] && skip_env "needs >= 2 MDTs"
local stripe_count
local file
mkdir -p $DIR/$tdir
$LFS setstripe -c 2 $DIR/$tdir
#local striped directory
$LFS setdirstripe -i 0 -c 2 -H all_char $DIR/$tdir/striped_dir ||
error "set striped dir error"
#look at the directories for debug purposes
ls -l $DIR/$tdir
$LFS getdirstripe $DIR/$tdir
ls -l $DIR/$tdir/striped_dir
$LFS getdirstripe $DIR/$tdir/striped_dir
createmany -o $DIR/$tdir/striped_dir/f 10 ||
error "create 10 files failed"
#remote striped directory
$LFS setdirstripe -i 1 -c 2 $DIR/$tdir/remote_striped_dir ||
error "set striped dir error"
#look at the directories for debug purposes
ls -l $DIR/$tdir
$LFS getdirstripe $DIR/$tdir
ls -l $DIR/$tdir/remote_striped_dir
$LFS getdirstripe $DIR/$tdir/remote_striped_dir
createmany -o $DIR/$tdir/remote_striped_dir/f 10 ||
error "create 10 files failed"
for file in $(find $DIR/$tdir); do
stripe_count=$($LFS getstripe -c $file)
[ $stripe_count -eq 2 ] ||
error "wrong stripe $stripe_count for $file"
done
rm -rf $DIR/$tdir
}
run_test 300d "check default stripe under striped directory"
test_300e() {
[ $MDS1_VERSION -lt $(version_code 2.7.55) ] &&
skip "Need MDS version at least 2.7.55"
[ $PARALLEL == "yes" ] && skip "skip parallel run"
[ $MDSCOUNT -lt 2 ] && skip_env "needs >= 2 MDTs"
local stripe_count
local file
mkdir -p $DIR/$tdir
$LFS setdirstripe -i 0 -c 2 -H all_char $DIR/$tdir/striped_dir ||
error "set striped dir error"
touch $DIR/$tdir/striped_dir/a
touch $DIR/$tdir/striped_dir/b
touch $DIR/$tdir/striped_dir/c
mkdir $DIR/$tdir/striped_dir/dir_a
mkdir $DIR/$tdir/striped_dir/dir_b
mkdir $DIR/$tdir/striped_dir/dir_c
$LFS setdirstripe -i 0 -c 2 -H all_char $DIR/$tdir/striped_dir/stp_a ||
error "set striped adir under striped dir error"
$LFS setdirstripe -i 0 -c 2 -H all_char $DIR/$tdir/striped_dir/stp_b ||
error "set striped bdir under striped dir error"
$LFS setdirstripe -i 0 -c 2 -H all_char $DIR/$tdir/striped_dir/stp_c ||
error "set striped cdir under striped dir error"
mrename $DIR/$tdir/striped_dir/dir_a $DIR/$tdir/striped_dir/dir_b ||
error "rename dir under striped dir fails"
mrename $DIR/$tdir/striped_dir/stp_a $DIR/$tdir/striped_dir/stp_b ||
error "rename dir under different stripes fails"
mrename $DIR/$tdir/striped_dir/a $DIR/$tdir/striped_dir/c ||
error "rename file under striped dir should succeed"
mrename $DIR/$tdir/striped_dir/dir_b $DIR/$tdir/striped_dir/dir_c ||
error "rename dir under striped dir should succeed"
rm -rf $DIR/$tdir
}
run_test 300e "check rename under striped directory"
test_300f() {
[ $PARALLEL == "yes" ] && skip "skip parallel run"
[ $MDSCOUNT -lt 2 ] && skip_env "needs >= 2 MDTs"
[ $MDS1_VERSION -lt $(version_code 2.7.55) ] &&
skip "Need MDS version at least 2.7.55"
local stripe_count
local file
rm -rf $DIR/$tdir
mkdir -p $DIR/$tdir
$LFS setdirstripe -i 0 -c 2 -H all_char $DIR/$tdir/striped_dir ||
error "set striped dir error"
$LFS setdirstripe -i 0 -c 2 -H all_char $DIR/$tdir/striped_dir1 ||
error "set striped dir error"
touch $DIR/$tdir/striped_dir/a
mkdir $DIR/$tdir/striped_dir/dir_a
$LFS setdirstripe -i 0 -c 2 $DIR/$tdir/striped_dir/stp_a ||
error "create striped dir under striped dir fails"
touch $DIR/$tdir/striped_dir1/b
mkdir $DIR/$tdir/striped_dir1/dir_b
$LFS setdirstripe -i 0 -c 2 $DIR/$tdir/striped_dir/stp_b ||
error "create striped dir under striped dir fails"
mrename $DIR/$tdir/striped_dir/dir_a $DIR/$tdir/striped_dir1/dir_b ||
error "rename dir under different striped dir should fail"
mrename $DIR/$tdir/striped_dir/stp_a $DIR/$tdir/striped_dir1/stp_b ||
error "rename striped dir under diff striped dir should fail"
mrename $DIR/$tdir/striped_dir/a $DIR/$tdir/striped_dir1/a ||
error "rename file under diff striped dirs fails"
rm -rf $DIR/$tdir
}
run_test 300f "check rename cross striped directory"
test_300_check_default_striped_dir()
{
local dirname=$1
local default_count=$2
local default_index=$3
local stripe_count
local stripe_index
local dir_stripe_index
local dir
echo "checking $dirname $default_count $default_index"
$LFS setdirstripe -D -c $default_count -i $default_index \
-H all_char $DIR/$tdir/$dirname ||
error "set default stripe on striped dir error"
stripe_count=$($LFS getdirstripe -D -c $DIR/$tdir/$dirname)
[ $stripe_count -eq $default_count ] ||
error "expect $default_count get $stripe_count for $dirname"
stripe_index=$($LFS getdirstripe -D -i $DIR/$tdir/$dirname)
[ $stripe_index -eq $default_index ] ||
error "expect $default_index get $stripe_index for $dirname"
mkdir $DIR/$tdir/$dirname/{test1,test2,test3,test4} ||
error "create dirs failed"
createmany -o $DIR/$tdir/$dirname/f- 10 || error "create files failed"
unlinkmany $DIR/$tdir/$dirname/f- 10 || error "unlink files failed"
for dir in $(find $DIR/$tdir/$dirname/*); do
stripe_count=$($LFS getdirstripe -c $dir)
(( $stripe_count == $default_count )) ||
(( $stripe_count == $MDSCOUNT && $default_count == -1 )) ||
(( $stripe_count == 0 )) || (( $default_count == 1 )) ||
error "stripe count $default_count != $stripe_count for $dir"
stripe_index=$($LFS getdirstripe -i $dir)
[ $default_index -eq -1 ] ||
[ $stripe_index -eq $default_index ] ||
error "$stripe_index != $default_index for $dir"
#check default stripe
stripe_count=$($LFS getdirstripe -D -c $dir)
[ $stripe_count -eq $default_count ] ||
error "default count $default_count != $stripe_count for $dir"
stripe_index=$($LFS getdirstripe -D -i $dir)
[ $stripe_index -eq $default_index ] ||
error "default index $default_index != $stripe_index for $dir"
done
rmdir $DIR/$tdir/$dirname/* || error "rmdir failed"
}
test_300g() {
[ $MDSCOUNT -lt 2 ] && skip_env "needs >= 2 MDTs"
[ $MDS1_VERSION -lt $(version_code 2.7.55) ] &&
skip "Need MDS version at least 2.7.55"
local dir
local stripe_count
local stripe_index
mkdir_on_mdt0 $DIR/$tdir
mkdir $DIR/$tdir/normal_dir
#Checking when client cache stripe index
$LFS setdirstripe -c$MDSCOUNT $DIR/$tdir/striped_dir
$LFS setdirstripe -D -i1 $DIR/$tdir/striped_dir ||
error "create striped_dir failed"
$LFS setdirstripe -i0 $DIR/$tdir/striped_dir/dir0 ||
error "create dir0 fails"
stripe_index=$($LFS getdirstripe -i $DIR/$tdir/striped_dir/dir0)
[ $stripe_index -eq 0 ] ||
error "dir0 expect index 0 got $stripe_index"
mkdir $DIR/$tdir/striped_dir/dir1 ||
error "create dir1 fails"
stripe_index=$($LFS getdirstripe -i $DIR/$tdir/striped_dir/dir1)
[ $stripe_index -eq 1 ] ||
error "dir1 expect index 1 got $stripe_index"
#check default stripe count/stripe index
test_300_check_default_striped_dir normal_dir $MDSCOUNT 1
test_300_check_default_striped_dir normal_dir 1 0
test_300_check_default_striped_dir normal_dir -1 1
test_300_check_default_striped_dir normal_dir 2 -1
#delete default stripe information
echo "delete default stripeEA"
$LFS setdirstripe -d $DIR/$tdir/normal_dir ||
error "set default stripe on striped dir error"
mkdir -p $DIR/$tdir/normal_dir/{test1,test2,test3,test4}
for dir in $(find $DIR/$tdir/normal_dir/*); do
stripe_count=$($LFS getdirstripe -c $dir)
[ $stripe_count -eq 0 ] ||
error "expect 1 get $stripe_count for $dir"
done
}
run_test 300g "check default striped directory for normal directory"
test_300h() {
[ $MDSCOUNT -lt 2 ] && skip_env "needs >= 2 MDTs"
[ $MDS1_VERSION -lt $(version_code 2.7.55) ] &&
skip "Need MDS version at least 2.7.55"
local dir
local stripe_count
mkdir $DIR/$tdir
$LFS setdirstripe -i0 -c$MDSCOUNT -H all_char $DIR/$tdir/striped_dir ||
error "set striped dir error"
test_300_check_default_striped_dir striped_dir $MDSCOUNT 1
test_300_check_default_striped_dir striped_dir 1 0
test_300_check_default_striped_dir striped_dir -1 1
test_300_check_default_striped_dir striped_dir 2 -1
#delete default stripe information
$LFS setdirstripe -d $DIR/$tdir/striped_dir ||
error "set default stripe on striped dir error"
mkdir -p $DIR/$tdir/striped_dir/{test1,test2,test3,test4}
for dir in $(find $DIR/$tdir/striped_dir/*); do
stripe_count=$($LFS getdirstripe -c $dir)
[ $stripe_count -eq 0 ] ||
error "expect 1 get $stripe_count for $dir"
done
}
run_test 300h "check default striped directory for striped directory"
test_300i() {
[[ $PARALLEL == "yes" ]] && skip "skip parallel run"
(( $MDSCOUNT >= 2 )) || skip_env "needs >= 2 MDTs"
(( $MDS1_VERSION >= $(version_code 2.7.55) )) ||
skip "Need MDS version at least 2.7.55"
local stripe_count
local file
mkdir $DIR/$tdir
$LFS setdirstripe -i 0 -c$MDSCOUNT -H all_char $DIR/$tdir/striped_dir ||
error "set striped dir error"
createmany -o $DIR/$tdir/striped_dir/f- 10 ||
error "create files under striped dir failed"
$LFS setdirstripe -i0 -c$MDSCOUNT -H all_char $DIR/$tdir/hashdir ||
error "set striped hashdir error"
$LFS setdirstripe -i0 -c$MDSCOUNT -H all_char $DIR/$tdir/hashdir/d0 ||
error "create dir0 under hash dir failed"
$LFS setdirstripe -i0 -c$MDSCOUNT -H fnv_1a_64 $DIR/$tdir/hashdir/d1 ||
error "create dir1 under hash dir failed"
$LFS setdirstripe -i0 -c$MDSCOUNT -H crush $DIR/$tdir/hashdir/d2 ||
error "create dir2 under hash dir failed"
# unfortunately, we need to umount to clear dir layout cache for now
# once we fully implement dir layout, we can drop this
umount_client $MOUNT || error "umount failed"
mount_client $MOUNT || error "mount failed"
$LFS find -H fnv_1a_64,crush $DIR/$tdir/hashdir
local dircnt=$($LFS find -H fnv_1a_64,crush $DIR/$tdir/hashdir | wc -l)
(( $dircnt == 2 )) || error "lfs find striped dir got $dircnt != 2"
if (( $MDS1_VERSION > $(version_code v2_15_50-185-g1ac4b9598a) )); then
$LFS mkdir -i0 -c$MDSCOUNT -H crush2 $DIR/$tdir/hashdir/d3 ||
error "create crush2 dir $tdir/hashdir/d3 failed"
$LFS find -H crush2 $DIR/$tdir/hashdir
dircnt=$($LFS find -H crush2 $DIR/$tdir/hashdir | wc -l)
(( $dircnt == 1 )) || error "find crush2 dir got $dircnt != 1"
# mkdir with an invalid hash type (hash=fail_val) from client
# should be replaced on MDS with a valid (default) hash type
#define OBD_FAIL_LMV_UNKNOWN_STRIPE 0x1901
$LCTL set_param fail_loc=0x1901 fail_val=99
$LFS mkdir -c2 $DIR/$tdir/hashdir/d99
local hash=$($LFS getdirstripe -H $DIR/$tdir/hashdir/d99)
local expect=$(do_facet mds1 \
$LCTL get_param -n lod.$FSNAME-MDT0000-mdtlov.mdt_hash)
[[ $hash == $expect ]] ||
error "d99 hash '$hash' != expected hash '$expect'"
fi
#set the stripe to be unknown hash type on read
#define OBD_FAIL_LMV_UNKNOWN_STRIPE 0x1901
$LCTL set_param fail_loc=0x1901 fail_val=99
for ((i = 0; i < 10; i++)); do
$CHECKSTAT -t file $DIR/$tdir/striped_dir/f-$i ||
error "stat f-$i failed"
rm $DIR/$tdir/striped_dir/f-$i || error "unlink f-$i failed"
done
touch $DIR/$tdir/striped_dir/f0 &&
error "create under striped dir with unknown hash should fail"
$LCTL set_param fail_loc=0
umount_client $MOUNT || error "umount failed"
mount_client $MOUNT || error "mount failed"
return 0
}
run_test 300i "client handle unknown hash type striped directory"
test_300j() {
[ $MDSCOUNT -lt 2 ] && skip_env "needs >= 2 MDTs"
[ $PARALLEL == "yes" ] && skip "skip parallel run"
[ $MDS1_VERSION -lt $(version_code 2.7.55) ] &&
skip "Need MDS version at least 2.7.55"
local stripe_count
local file
mkdir $DIR/$tdir
#define OBD_FAIL_SPLIT_UPDATE_REC 0x1702
$LCTL set_param fail_loc=0x1702
$LFS setdirstripe -i 0 -c$MDSCOUNT -H all_char $DIR/$tdir/striped_dir ||
error "set striped dir error"
createmany -o $DIR/$tdir/striped_dir/f- 10 ||
error "create files under striped dir failed"
$LCTL set_param fail_loc=0
rm -rf $DIR/$tdir || error "unlink striped dir fails"
return 0
}
run_test 300j "test large update record"
test_300k() {
[ $PARALLEL == "yes" ] && skip "skip parallel run"
[ $MDSCOUNT -lt 2 ] && skip_env "needs >= 2 MDTs"
[ $MDS1_VERSION -lt $(version_code 2.7.55) ] &&
skip "Need MDS version at least 2.7.55"
# this test needs a huge transaction
local kb
kb=$(do_facet $SINGLEMDS "$LCTL get_param -n \
osd*.$FSNAME-MDT0000.kbytestotal")
[ $kb -lt $((1024*1024)) ] && skip "MDT0 too small: $kb"
local stripe_count
local file
mkdir $DIR/$tdir
#define OBD_FAIL_LARGE_STRIPE 0x1703
$LCTL set_param fail_loc=0x1703
$LFS setdirstripe -i 0 -c192 $DIR/$tdir/striped_dir ||
error "set striped dir error"
$LCTL set_param fail_loc=0
$LFS getdirstripe $DIR/$tdir/striped_dir ||
error "getstripeddir fails"
rm -rf $DIR/$tdir/striped_dir ||
error "unlink striped dir fails"
return 0
}
run_test 300k "test large striped directory"
test_300l() {
[ $PARALLEL == "yes" ] && skip "skip parallel run"
[ $MDSCOUNT -lt 2 ] && skip_env "needs >= 2 MDTs"
[ $MDS1_VERSION -lt $(version_code 2.7.55) ] &&
skip "Need MDS version at least 2.7.55"
local stripe_index
test_mkdir -p $DIR/$tdir/striped_dir
chown $RUNAS_ID $DIR/$tdir/striped_dir ||
error "chown $RUNAS_ID failed"
$LFS setdirstripe -i 1 -D $DIR/$tdir/striped_dir ||
error "set default striped dir failed"
do_facet mds2 $LCTL set_param mdt.*MDT0001.enable_remote_dir_gid=-1
stack_trap "do_facet mds2 $LCTL set_param mdt.*MDT0001.enable_remote_dir_gid=0"
#define OBD_FAIL_MDS_STALE_DIR_LAYOUT 0x158
$LCTL set_param fail_loc=0x80000158
$RUNAS mkdir $DIR/$tdir/striped_dir/test_dir || error "create dir fails"
stripe_index=$($LFS getdirstripe -i $DIR/$tdir/striped_dir/test_dir)
[ $stripe_index -eq 1 ] ||
error "expect 1 get $stripe_index for $dir"
}
run_test 300l "non-root user to create dir under striped dir with stale layout"
test_300m() {
[ $PARALLEL == "yes" ] && skip "skip parallel run"
[ $MDSCOUNT -ge 2 ] && skip_env "Only for single MDT"
[ $MDS1_VERSION -lt $(version_code 2.7.55) ] &&
skip "Need MDS version at least 2.7.55"
mkdir -p $DIR/$tdir/striped_dir
$LFS setdirstripe -D -c 1 $DIR/$tdir/striped_dir ||
error "set default stripes dir error"
mkdir $DIR/$tdir/striped_dir/a || error "mkdir a fails"
stripe_count=$($LFS getdirstripe -c $DIR/$tdir/striped_dir/a)
[ $stripe_count -eq 0 ] ||
error "expect 0 get $stripe_count for a"
$LFS setdirstripe -D -c 2 $DIR/$tdir/striped_dir ||
error "set default stripes dir error"
mkdir $DIR/$tdir/striped_dir/b || error "mkdir b fails"
stripe_count=$($LFS getdirstripe -c $DIR/$tdir/striped_dir/b)
[ $stripe_count -eq 0 ] ||
error "expect 0 get $stripe_count for b"
$LFS setdirstripe -D -c1 -i2 $DIR/$tdir/striped_dir ||
error "set default stripes dir error"
mkdir $DIR/$tdir/striped_dir/c &&
error "default stripe_index is invalid, mkdir c should fails"
rm -rf $DIR/$tdir || error "rmdir fails"
}
run_test 300m "setstriped directory on single MDT FS"
test_300n() {
[[ $PARALLEL != "yes" ]] || skip "skip parallel run"
(( $MDSCOUNT >= 2 )) || skip_env "needs >= 2 MDTs"
(( $MDS1_VERSION >= $(version_code 2.7.55) )) ||
skip "Need MDS >= 2.7.55 for enable_remote_dir_gid"
remote_mds_nodsh && skip "remote MDS with nodsh"
local stripe_index
local mdts=$(mdts_nodes)
local param_gid="mdt.*.enable_remote_dir_gid"
local old_gid=($(do_facet mds1 $LCTL get_param -n $param_gid))
mkdir --mode=0777 $DIR/$tdir
stack_trap "do_nodes $mdts '$LCTL set_param $param_gid=$old_gid'"
# mkdir _gid=0 with explicit setstripe, should not work for regular user
do_nodes $mdts $LCTL set_param $param_gid=0 ||
error "set $param_gid=0 failed"
$RUNAS $LFS setdirstripe -i0 -c$MDSCOUNT \
$DIR/$tdir/striped_dir > /dev/null 2>&1 &&
error "create striped dir succeeds with remote_gid=0"
$RUNAS $LFS setdirstripe -D -i 1 -c$MDSCOUNT $DIR/$tdir/striped_dir &&
error "set default dir layout succeeds with remote_gid=0"
[[ ! -e $DIR/$tdir/striped_dir ]] || error "striped_dir was created"
# mkdir _gid=-1 with explicit setstripe, should work for any user
do_nodes $mdts $LCTL set_param $param_gid=-1
$RUNAS $LFS setdirstripe -i0 -c$MDSCOUNT $DIR/$tdir/striped_dir ||
error "create striped dir fails with remote_gid=-1"
$RUNAS $LFS setdirstripe -D -i 1 -c$MDSCOUNT $DIR/$tdir/striped_dir ||
error "set default dir layout fails with remote_gid=-1"
# mkdir _gid=0 with default setstripe, should work for any user
do_nodes $mdts $LCTL set_param $param_gid=0
$RUNAS mkdir $DIR/$tdir/striped_dir/test_dir.{1,2,3} ||
error "mkdir in default test_dir with remote_gid=0 fails"
[[ -d $DIR/$tdir/striped_dir/test_dir.1 ]] ||
error "striped_dir/test_dir.1 not created"
}
run_test 300n "non-root user to create dir under striped dir with default EA"
test_300ne()
{
[[ $PARALLEL != "yes" ]] || skip "skip parallel run"
(( $MDSCOUNT >= 2 )) || skip_env "needs >= 2 MDTs"
(( $MDS1_VERSION >= $(version_code 2.17.51) )) ||
skip "Need MDS >= 2.17.51 for supp remote_dir_gid"
remote_mds_nodsh && skip "remote MDS with nodsh"
local param_gid="mdt.*.enable_remote_dir_gid"
local old_gid=($(do_facet mds1 $LCTL get_param -n $param_gid))
local mdts=$(mdts_nodes)
echo "using primary group $RUNAS_GID"
do_nodes $mdts "grep '$RUNAS_GID:' /etc/group"
do_nodes $mdts "id $RUNAS_ID"
do_nodes $mdts "$LCTL get_param mdt.*.identity_upcall"
do_nodes $mdts "$L_GETIDENTITY -d $RUNAS_ID"
start_full_debug_logging
stack_trap stop_full_debug_logging
mkdir_on_mdt0 -o 0777 $DIR/$tdir
stack_trap "do_nodes $mdts '$LCTL set_param $param_gid=$old_gid'"
# mkdir _gid=N with explicit setstripe, should work for primary group
do_nodes $mdts $LCTL set_param $param_gid=$RUNAS_GID
$RUNAS $LFS mkdir -i 1 $DIR/$tdir/group_dir ||
error "create remote dir fails with remote_gid=$RUNAS_ID"
[[ -d $DIR/$tdir/group_dir ]] || error "$tdir/group_dir not created"
echo "using $TSTUSR with secondary groups $new_gids"
local tst_gids=$(do_facet mds1 id -G $TSTUSR | tr ' ' ',')
local new_gids="$tst_gids,$RUNAS_ID"
local runas="$RUNAS -u $TSTUSR -g $TSTUSR -G$new_gids"
# add $RUNAS_ID as a secondary group to $TSTUSR
stack_trap "do_nodes $mdts usermod -G $tst_gids $TSTUSR"
do_nodes $mdts usermod -a -G $RUNAS_ID $TSTUSR ||
error "can't add $TSTUSR to group $RUNAS_ID"
do_nodes $mdts "grep ':$RUNAS_ID:' /etc/group"
do_nodes $mdts "id $TSTUSR"
do_nodes $mdts "$L_GETIDENTITY -d $(id -u $TSTUSR)"
# mkdir _gid=N with explicit setstripe, should work for secondary group
$runas $LFS mkdir -i 1 $DIR/$tdir/secondary_dir ||
error "create remote dir fails with secondary gids=$new_gids"
[[ -d $DIR/$tdir/secondary_dir ]] || error "secondary_dir not created"
}
run_test 300ne "create remote dir with various mdt.enable_remote_gid"
test_300o() {
[ $PARALLEL == "yes" ] && skip "skip parallel run"
[ $MDSCOUNT -lt 2 ] && skip_env "needs >= 2 MDTs"
[ $MDS1_VERSION -lt $(version_code 2.7.55) ] &&
skip "Need MDS version at least 2.7.55"
local numfree1
local numfree2
mkdir -p $DIR/$tdir
numfree1=$(lctl get_param -n mdc.*MDT0000*.filesfree)
numfree2=$(lctl get_param -n mdc.*MDT0001*.filesfree)
if [ $numfree1 -lt 66000 ] || [ $numfree2 -lt 66000 ]; then
skip "not enough free inodes $numfree1 $numfree2"
fi
numfree1=$(lctl get_param -n mdc.*MDT0000-mdc-*.kbytesfree)
numfree2=$(lctl get_param -n mdc.*MDT0001-mdc-*.kbytesfree)
if [ $numfree1 -lt 300000 ] || [ $numfree2 -lt 300000 ]; then
skip "not enough free space $numfree1 $numfree2"
fi
$LFS setdirstripe -c2 $DIR/$tdir/striped_dir ||
error "setdirstripe fails"
createmany -d $DIR/$tdir/striped_dir/d 131000 ||
error "create dirs fails"
$LCTL set_param ldlm.namespaces.*mdc-*.lru_size=0
ls $DIR/$tdir/striped_dir > /dev/null ||
error "ls striped dir fails"
unlinkmany -d $DIR/$tdir/striped_dir/d 131000 ||
error "unlink big striped dir fails"
}
run_test 300o "unlink big sub stripe(> 65000 subdirs)"
test_300p() {
[ $PARALLEL == "yes" ] && skip "skip parallel run"
[ $MDSCOUNT -lt 2 ] && skip_env "needs >= 2 MDTs"
remote_mds_nodsh && skip "remote MDS with nodsh"
mkdir_on_mdt0 $DIR/$tdir
#define OBD_FAIL_OUT_ENOSPC 0x1704
do_facet mds2 lctl set_param fail_loc=0x1704
$LFS setdirstripe -i 0,1 $DIR/$tdir/bad_striped_dir &&
error "create striped directory should fail"
! [ -e $DIR/$tdir/bad_striped_dir ] || error "striped dir exists"
}
run_test 300p "create striped directory without space"
test_300q() {
[ $PARALLEL == "yes" ] && skip "skip parallel run"
[ $MDSCOUNT -lt 2 ] && skip_env "needs >= 2 MDTs"
local fd=$(free_fd)
local cmd="exec $fd<$tdir"
cd $DIR
$LFS mkdir -c $MDSCOUNT $tdir || error "create $tdir fails"
eval $cmd
cmd="exec $fd<&-"
trap "eval $cmd" EXIT
cd $tdir || error "cd $tdir fails"
rmdir ../$tdir || error "rmdir $tdir fails"
mkdir local_dir && error "create dir succeeds"
$LFS setdirstripe -i1 remote_dir && error "create remote dir succeeds"
eval $cmd
return 0
}
run_test 300q "create remote directory under orphan directory"
test_300r() {
[ $MDS1_VERSION -lt $(version_code 2.7.55) ] &&
skip "Need MDS version at least 2.7.55" && return
[ $MDSCOUNT -lt 2 ] && skip "needs >= 2 MDTs" && return
mkdir $DIR/$tdir
$LFS setdirstripe -i 0 -c -1 $DIR/$tdir/striped_dir ||
error "set striped dir error"
$LFS getdirstripe $DIR/$tdir/striped_dir ||
error "getstripeddir fails"
local stripe_count
stripe_count=$($LFS getdirstripe $DIR/$tdir/striped_dir |
awk '/lmv_stripe_count:/ { print $2 }')
[ $MDSCOUNT -ne $stripe_count ] &&
error "wrong stripe count $stripe_count expected $MDSCOUNT"
rm -rf $DIR/$tdir/striped_dir ||
error "unlink striped dir fails"
}
run_test 300r "test -1 striped directory"
test_300s_helper() {
local count=$1
local stripe_dir=$DIR/$tdir/striped_dir.$count
$LFS mkdir -c $count $stripe_dir ||
error "lfs mkdir -c error"
$LFS getdirstripe $stripe_dir ||
error "lfs getdirstripe fails"
local stripe_count
stripe_count=$($LFS getdirstripe $stripe_dir |
awk '/lmv_stripe_count:/ { print $2 }')
[ $count -ne $stripe_count ] &&
error_noexit "bad stripe count $stripe_count expected $count"
local dupe_stripes
dupe_stripes=$($LFS getdirstripe $stripe_dir |
awk '/0x/ {count[$1] += 1}; END {
for (idx in count) {
if (count[idx]>1) {
print "index " idx " count " count[idx]
}
}
}')
if [[ -n "$dupe_stripes" ]] ; then
lfs getdirstripe $stripe_dir
error_noexit "Dupe MDT above: $dupe_stripes "
fi
rm -rf $stripe_dir ||
error_noexit "unlink $stripe_dir fails"
}
test_300s() {
[ $MDS1_VERSION -lt $(version_code 2.7.55) ] &&
skip "Need MDS version at least 2.7.55" && return
[ $MDSCOUNT -lt 2 ] && skip "needs >= 2 MDTs" && return
mkdir $DIR/$tdir
for count in $(seq 2 $MDSCOUNT); do
test_300s_helper $count
done
}
run_test 300s "test lfs mkdir -c without -i"
test_300t() {
(( $MDS1_VERSION >= $(version_code 2.14.55) )) ||
skip "need MDS 2.14.55 or later"
(( $MDSCOUNT >= 2 )) || skip "needs at least 2 MDTs"
local testdir="$DIR/$tdir/striped_dir"
local dir1=$testdir/dir1
local dir2=$testdir/dir2
mkdir -p $testdir
$LFS setdirstripe -D -c -1 --max-inherit=3 $testdir ||
error "failed to set default stripe count for $testdir"
mkdir $dir1
local stripe_count=$($LFS getdirstripe -c $dir1)
(( $stripe_count == $MDSCOUNT )) || error "wrong stripe count"
local max_count=$((MDSCOUNT - 1))
local mdts=$(mdts_nodes)
do_nodes $mdts $LCTL set_param lod.*.max_mdt_stripecount=$max_count
stack_trap "do_nodes $mdts $LCTL set_param lod.*.max_mdt_stripecount=0"
mkdir $dir2
stripe_count=$($LFS getdirstripe -c $dir2)
if (( $MDSCOUNT == 2 &&
$MDS1_VERSION >= $(version_code 2.16.0) )); then
(( $stripe_count == 0 )) || error "(0) wrong stripe count"
else
(( $stripe_count == $max_count )) ||
error "(1) wrong stripe count"
fi
}
run_test 300t "test max_mdt_stripecount"
max_stripes_per_mdt=$(do_facet mds1 $LCTL get_param -n \
lod.$FSNAME-MDT0000-mdtlov.max_stripes_per_mdt || echo 0)
((max_stripes_per_mdt == 0)) ||
do_nodes $(mdts_nodes) $LCTL set_param -n \
lod.$FSNAME-MDT*.max_stripes_per_mdt=$LMV_MAX_STRIPES_PER_MDT
# 300u family tests MDT overstriping
test_300ua() {
(( MDSCOUNT > 1 )) || skip "needs >= 2 MDTs"
local setcount=$((MDSCOUNT * 2))
local expected_count
mkdir $DIR/$tdir
$LFS setdirstripe -C $setcount $DIR/$tdir/${tdir}.0 ||
error "(0) failed basic overstriped dir creation test"
local getstripe_count=$($LFS getdirstripe -c $DIR/$tdir/${tdir}.0)
# This does a basic interop test - if the MDS does not support mdt
# overstriping, we should get stripes == number of MDTs
if (( max_stripes_per_mdt == 0 )); then
expected_count=$MDSCOUNT
else
expected_count=$setcount
fi
(( getstripe_count == expected_count )) ||
error "(1) incorrect stripe count for simple overstriped dir"
rm -rf $DIR/$tdir/${tdir}.0 ||
error "(2) unable to rm overstriped dir"
# Tests after this require overstriping support
(( max_stripes_per_mdt > 0 )) || return 0
test_striped_dir 0 $setcount true ||
error "(3)failed on overstriped dir"
test_striped_dir 1 $setcount true ||
error "(4)failed on overstriped dir"
local setcount=$((MDSCOUNT * $LMV_MAX_STRIPES_PER_MDT))
test_striped_dir 0 $setcount true ||
error "(5)failed on overstriped dir"
}
run_test 300ua "basic overstriped dir sanity test"
test_300ub() {
(( max_stripes_per_mdt > 0 )) ||
skip "skipped for MDS that doesn't support metadata overstripe"
(( MDSCOUNT > 1 )) || skip "needs >= 2 MDTs"
mkdir $DIR/$tdir
echo "Testing invalid stripe count, failure expected"
local setcount=$((MDSCOUNT * 2))
$LFS setdirstripe -c $setcount $DIR/$tdir/${tdir}.0
local getstripe_count=$($LFS getdirstripe -c $DIR/$tdir/${tdir}.0)
(( getstripe_count <= MDSCOUNT )) ||
error "(0)stripe count ($setcount) > MDT count ($MDSCOUNT) succeeded with -c"
# When a user requests > LMV_MAX_STRIPES_PER_MDT, we reduce to that
setcount=$((MDSCOUNT * 2 * LMV_MAX_STRIPES_PER_MDT))
$LFS setdirstripe -C $setcount $DIR/$tdir/${tdir}.1
local maxcount=$((MDSCOUNT * LMV_MAX_STRIPES_PER_MDT))
getstripe_count=$($LFS getdirstripe -c $DIR/$tdir/${tdir}.1)
(( getstripe_count == maxcount )) ||
error "(1)stripe_count is $getstripe_count, expect $maxcount"
# Test specific striping with -i
$LFS setdirstripe -i 0,0,0,0 $DIR/$tdir/${tdir}.2
getstripe_count=$($LFS getdirstripe -c $DIR/$tdir/${tdir}.2)
(( getstripe_count == 4 )) ||
error "(2)stripe_count is $getstripe_count, expect 4"
local nonzeroindices=$($LFS getdirstripe $DIR/$tdir/${tdir}.2 | grep "\[" | \
grep -v mdtidx | awk '{print $1}' | grep -c -v 0)
[[ -n "$nonzeroindices" ]] ||
error "(3) stripes indices not all 0: $nonzeroindices"
# Test specific striping with too many stripes on one MDT
echo "Testing invalid striping, failure expected"
$LFS setdirstripe -i 0,1,0,1,0,1,0,1,0,1,0 $DIR/$tdir/${tdir}.3
$LFS getdirstripe $DIR/$tdir/${tdir}.3
getstripe_count=$($LFS getdirstripe $DIR/$tdir/${tdir}.3 | grep "\[" | \
grep -v mdtidx | awk '{print $1}' | grep -c '0')
echo "stripes on MDT0: $getstripe_count"
(( getstripe_count <= LMV_MAX_STRIPES_PER_MDT )) ||
error "(4) setstripe with too many stripes on MDT0 succeeded"
setcount=$((MDSCOUNT * 2))
$LFS setdirstripe -C $setcount -H all_char $DIR/${tdir}.4 ||
error "(5) can't setdirstripe with manually set hash function"
getstripe_count=$($LFS getdirstripe -c $DIR/${tdir}.4)
(( getstripe_count == setcount )) ||
error "(6)stripe_count is $getstripe_count, expect $setcount"
setcount=$((MDSCOUNT * 2))
mkdir $DIR/${tdir}.5
$LFS setdirstripe -C $setcount -D -H crush $DIR/${tdir}.5 ||
error "(7) can't setdirstripe with manually set hash function"
mkdir $DIR/${tdir}.5/${tdir}.6
getstripe_count=$($LFS getdirstripe -c $DIR/${tdir}.5/${tdir}.6)
(( getstripe_count == setcount )) ||
error "(8)stripe_count is $getstripe_count, expect $setcount"
}
run_test 300ub "test MDT overstriping interface & limits"
test_300uc() {
(( max_stripes_per_mdt > 0 )) ||
skip "skipped for MDS that doesn't support metadata overstripe"
(( MDSCOUNT > 1 )) || skip "needs >= 2 MDTs"
mkdir $DIR/$tdir
local setcount=$((MDSCOUNT * 2))
$LFS setdirstripe -D -C $setcount $DIR/$tdir
mkdir $DIR/$tdir/${tdir}.1
local getstripe_count=$($LFS getdirstripe -c $DIR/$tdir/${tdir}.1)
(( getstripe_count == setcount )) ||
error "(0)stripe_count is $getstripe_count, expect $setcount"
mkdir $DIR/$tdir/${tdir}.1/${tdir}.2
local getstripe_count=$($LFS getdirstripe -c \
$DIR/$tdir/${tdir}.1/${tdir}.2)
(( getstripe_count == setcount )) ||
error "(1)stripe_count is $getstripe_count, expect $setcount"
}
run_test 300uc "test MDT overstriping as default & inheritance"
test_300ud() {
(( max_stripes_per_mdt > 0 )) ||
skip "skipped for MDS that doesn't support metadata overstripe"
(( MDSCOUNT > 1 )) || skip "needs >= 2 MDTs"
local mdts=$(mdts_nodes)
local timeout=100
local restripe_status
local delta
local i
[[ $mds1_FSTYPE == zfs ]] && timeout=300
# in case "crush" hash type is not set
do_nodes $mdts "$LCTL set_param lod.*.mdt_hash=crush"
restripe_status=$(do_facet mds1 $LCTL get_param -n \
mdt.*MDT0000.enable_dir_restripe)
do_nodes $mdts "$LCTL set_param mdt.*.enable_dir_restripe=1"
stack_trap "do_nodes $mdts $LCTL set_param \
mdt.*.enable_dir_restripe=$restripe_status"
mkdir $DIR/$tdir
createmany -m $DIR/$tdir/f $((50 * MDSCOUNT)) ||
error "create files under remote dir failed $i"
createmany -d $DIR/$tdir/d $((50 * MDSCOUNT)) ||
error "create dirs under remote dir failed $i"
local setcount=$((MDSCOUNT * $LMV_MAX_STRIPES_PER_MDT))
(( setcount < 13 )) || setcount=12
for i in $(seq 2 $setcount); do
do_nodes $mdts "$LCTL set_param mdt.*.md_stats=clear >/dev/null"
$LFS setdirstripe -C $i $DIR/$tdir ||
error "split -C $i $tdir failed"
wait_update $HOSTNAME \
"$LFS getdirstripe -H $DIR/$tdir" "crush" $timeout ||
error "dir split not finished"
delta=$(do_nodes $mdts "lctl get_param -n mdt.*MDT*.md_stats" |
awk '/migrate/ {sum += $2} END { print sum }')
echo "$delta migrated when dir split $((i - 1)) to $i stripes"
# delta is around total_files/stripe_count, deviation 3%
(( delta < 100 * MDSCOUNT / i + 3 * MDSCOUNT )) ||
error "$delta files migrated >= $((100 * MDSCOUNT / i + 3 * MDSCOUNT))"
done
}
run_test 300ud "dir split"
test_300ue() {
(( max_stripes_per_mdt > 0 )) ||
skip "skipped for MDS that doesn't support metadata overstripe"
(( MDSCOUNT > 1 )) || skip "needs >= 2 MDTs"
local mdts=$(mdts_nodes)
local timeout=100
local restripe_status
local delta
local c
[[ $mds1_FSTYPE == zfs ]] && timeout=300
do_nodes $mdts "$LCTL set_param lod.*.mdt_hash=crush"
restripe_status=$(do_facet mds1 $LCTL get_param -n \
mdt.*MDT0000.enable_dir_restripe)
do_nodes $mdts "$LCTL set_param mdt.*.enable_dir_restripe=1"
stack_trap "do_nodes $mdts $LCTL set_param \
mdt.*.enable_dir_restripe=$restripe_status"
local setcount=$((MDSCOUNT * $LMV_MAX_STRIPES_PER_MDT))
(( setcount < 13 )) || setcount=12
test_mkdir -C $setcount -H crush $DIR/$tdir
createmany -m $DIR/$tdir/f $((50 * MDSCOUNT)) ||
error "create files under remote dir failed"
createmany -d $DIR/$tdir/d $((50 * MDSCOUNT)) ||
error "create dirs under remote dir failed"
for c in $(seq $((setcount - 1)) -1 1); do
do_nodes $mdts "$LCTL set_param mdt.*.md_stats=clear >/dev/null"
$LFS setdirstripe -C $c $DIR/$tdir ||
error "split -C $c $tdir failed"
wait_update $HOSTNAME \
"$LFS getdirstripe -H $DIR/$tdir" "crush,fixed" $timeout ||
error "dir merge not finished"
delta=$(do_nodes $mdts "lctl get_param -n mdt.*MDT*.md_stats" |
awk '/migrate/ {sum += $2} END { print sum }')
echo "$delta migrated when dir merge $((c + 1)) to $c stripes"
# delta is around total_files/stripe_count, deviation 3%
(( delta < 100 * MDSCOUNT / c + 3 * MDSCOUNT )) ||
error "$delta files migrated >= $((100 * MDSCOUNT / c + 3 * MDSCOUNT))"
done
}
run_test 300ue "dir merge"
test_300uf() {
(( max_stripes_per_mdt > 0 )) ||
skip "skipped for MDS that doesn't support metadata overstripe"
(( MDSCOUNT > 1 )) || skip "needs >= 2 MDTs"
# maximum amount of local locks:
# parent striped dir - 2 locks
# new stripe in parent to migrate to - 1 lock
# source and target - 2 locks
# Total 5 locks for regular file
#
# NB: Overstriping should add several extra local locks
# FIXME: Remove this once understood
#lctl set_param *debug=-1 debug_mb=10000
lctl clear
lctl mark "touch/create"
mkdir -p $DIR/$tdir
local setcount=$((MDSCOUNT * $LMV_MAX_STRIPES_PER_MDT))
local setcount=$((MDSCOUNT * 5))
$LFS mkdir -i1 -C $setcount $DIR/$tdir/dir1
touch $DIR/$tdir/dir1/eee
lctl mark "hardlinks"
# create 4 hardlink for 4 more locks
# Total: 9 locks > RS_MAX_LOCKS (8)
$LFS mkdir -i1 -c1 $DIR/$tdir/dir2
$LFS mkdir -i1 -c1 $DIR/$tdir/dir3
$LFS mkdir -i1 -c1 $DIR/$tdir/dir4
$LFS mkdir -i1 -c1 $DIR/$tdir/dir5
ln $DIR/$tdir/dir1/eee $DIR/$tdir/dir2/eee
ln $DIR/$tdir/dir1/eee $DIR/$tdir/dir3/eee
ln $DIR/$tdir/dir1/eee $DIR/$tdir/dir4/eee
ln $DIR/$tdir/dir1/eee $DIR/$tdir/dir5/eee
lctl mark "cancel lru"
cancel_lru_locks mdc
lctl mark "migrate"
$LFS migrate -m1 -c1 $DIR/$tdir/dir1 ||
error "migrate dir fails"
rm -rf $DIR/$tdir || error "rm dir failed after migration"
}
run_test 300uf "migrate with too many local locks"
test_300ug() {
(( max_stripes_per_mdt > 0 )) ||
skip "skipped for MDS that doesn't support metadata overstripe"
(( MDSCOUNT > 1 )) || skip "needs >= 2 MDTs"
mkdir -p $DIR/$tdir
local migrate_dir=$DIR/$tdir/migrate_dir
local setcount=$((MDSCOUNT * $LMV_MAX_STRIPES_PER_MDT))
local setcount2=$((setcount - 2))
$LFS setdirstripe -c 2 $migrate_dir ||
error "(0) failed to create striped directory"
$LFS migrate -m 0 -C $setcount $migrate_dir ||
error "(1)failed to migrate to overstriped directory"
local getstripe_count=$($LFS getdirstripe -c $migrate_dir)
(( getstripe_count == setcount )) ||
error "(2)stripe_count is $getstripe_count, expect $setcount"
touch $DIR/$tdir/migrate_dir/$tfile ||
error "(3)failed to create file in overstriped directory"
$LFS migrate -m 0 -C $setcount2 $migrate_dir ||
error "(4)failed to migrate overstriped directory"
# Check stripe count after migration
$LFS getdirstripe $migrate_dir
getstripe_count=$($LFS getdirstripe -c $migrate_dir)
(( getstripe_count == setcount2 )) ||
error "(5)stripe_count is $getstripe_count, expect $setcount2"
rm -rf $migrate_dir || error "(6) unable to rm overstriped dir"
}
run_test 300ug "migrate overstriped dirs"
test_300uh() {
(( max_stripes_per_mdt > 0 )) ||
skip "skipped for MDS that doesn't support metadata overstripe"
(( MDSCOUNT > 1 )) || skip "needs >= 2 MDTs"
local mdts=$(mdts_nodes)
local val=$(do_facet mds1 $LCTL get_param -n \
lod.$FSNAME-MDT0000-mdtlov.max_stripes_per_mdt)
stack_trap "do_nodes $mdts $LCTL set_param -n \
lod.$FSNAME-MDT*.max_stripes_per_mdt=$val"
# create 5 stripes will be silently clamped down
do_nodes $mdts $LCTL set_param -n \
lod.$FSNAME-MDT*.max_stripes_per_mdt=2
$LFS setdirstripe -C $((MDSCOUNT * 5)) $DIR/$tdir ||
error "mkdir $tdir failed"
$LFS getdirstripe -H $DIR/$tdir | grep overstriped ||
error "overstriped flag not found"
local count=$($LFS getdirstripe -c $DIR/$tdir)
((count == 2 * MDSCOUNT)) || error "count $count != $((2 * MDSCOUNT))"
# create 3 stripes on MDT0 should fail
$LFS setdirstripe -i 0,0,0 $DIR/$tdir/sub && error "mkdir sub succeeded"
# OVERSTRIPED flag will be cleared if not really overstriped
$LFS setdirstripe -C $MDSCOUNT $DIR/$tdir/sub ||
error "mkdir sub failed"
$LFS getdirstripe -H $DIR/$tdir/sub | grep -v overstriped ||
error "overstriped flag found"
}
run_test 300uh "overstripe tunable max_stripes_per_mdt"
test_300ui() {
(( max_stripes_per_mdt > 0 )) ||
skip "skipped for MDS that doesn't support metadata overstripe"
(( MDSCOUNT == 1 )) || skip "1 MDT only"
$LFS setdirstripe -C 2 $DIR/$tdir && error "mkdir $tdir succeeded" ||
true
}
run_test 300ui "overstripe is not supported on one MDT system"
test_300uj() {
(( MDSCOUNT > 1 )) || skip "needs >= 2 MDTs"
(( MDS1_VERSION >= $(version_code 2.16.0) )) ||
skip "need MDS >= 2.16.0 for llog timestamps"
local setcount=-2
local expected_count=$((MDSCOUNT * 2))
mkdir $DIR/$tdir
$LFS mkdir -C $setcount $DIR/$tdir/${tdir}.0 ||
error "(0) failed overstriped dir creation with -C -N test"
local getstripe_count=$($LFS getdirstripe -c $DIR/$tdir/${tdir}.0)
(( getstripe_count == expected_count )) ||
error "(1) incorrect stripe count for -C -N overstriped dir"
rm -f $DIR/$tdir/${tdir}.0
# limit testing
setcount=-6
$LFS setdirstripe -C $setcount $DIR/$tdir/${tdir}.1 2>&1 |
grep "invalid stripe count" ||
error "(0) failed overstriped dir creation limit test"
}
run_test 300uj "overstriped dir with -C -N sanity test"
(( max_stripes_per_mdt == 0 )) ||
do_nodes $(mdts_nodes) $LCTL set_param -n \
lod.$FSNAME-MDT*.max_stripes_per_mdt=$max_stripes_per_mdt
prepare_remote_file() {
mkdir $DIR/$tdir/src_dir ||
error "create remote source failed"
cp /etc/hosts $DIR/$tdir/src_dir/a ||
error "cp to remote source failed"
touch $DIR/$tdir/src_dir/a
$LFS mkdir -i 1 $DIR/$tdir/tgt_dir ||
error "create remote target dir failed"
touch $DIR/$tdir/tgt_dir/b
mrename $DIR/$tdir/src_dir/a $DIR/$tdir/tgt_dir/b ||
error "rename dir cross MDT failed!"
$CHECKSTAT -t file $DIR/$tdir/src_dir/a &&
error "src_child still exists after rename"
$CHECKSTAT -t file $DIR/$tdir/tgt_dir/b ||
error "missing file(a) after rename"
diff /etc/hosts $DIR/$tdir/tgt_dir/b ||
error "diff after rename"
}
test_310a() {
[[ $MDSCOUNT -lt 2 ]] && skip_env "needs >= 4 MDTs"
[ $PARALLEL == "yes" ] && skip "skip parallel run"
local remote_file=$DIR/$tdir/tgt_dir/b
mkdir -p $DIR/$tdir
prepare_remote_file || error "prepare remote file failed"
#open-unlink file
$OPENUNLINK $remote_file $remote_file ||
error "openunlink $remote_file failed"
$CHECKSTAT -a $remote_file || error "$remote_file exists"
}
run_test 310a "open unlink remote file"
test_310b() {
[[ $MDSCOUNT -lt 2 ]] && skip_env "needs >= 4 MDTs"
[ $PARALLEL == "yes" ] && skip "skip parallel run"
local remote_file=$DIR/$tdir/tgt_dir/b
mkdir -p $DIR/$tdir
prepare_remote_file || error "prepare remote file failed"
ln $remote_file $DIR/$tfile || error "link failed for remote file"
$MULTIOP $DIR/$tfile Ouc || error "mulitop failed"
$CHECKSTAT -t file $remote_file || error "check file failed"
}
run_test 310b "unlink remote file with multiple links while open"
test_310c() {
[ $PARALLEL == "yes" ] && skip "skip parallel run"
[[ $MDSCOUNT -lt 4 ]] && skip_env "needs >= 4 MDTs"
local remote_file=$DIR/$tdir/tgt_dir/b
mkdir -p $DIR/$tdir
prepare_remote_file || error "prepare remote file failed"
ln $remote_file $DIR/$tfile || error "link failed for remote file"
multiop_bg_pause $remote_file O_uc ||
error "mulitop failed for remote file"
MULTIPID=$!
$MULTIOP $DIR/$tfile Ouc
kill -USR1 $MULTIPID
wait $MULTIPID
}
run_test 310c "open-unlink remote file with multiple links"
#LU-4825
test_311() {
[ $PARALLEL == "yes" ] && skip "skip parallel run"
[ $OSTCOUNT -lt 2 ] && skip "needs >= 2 OSTs"
[ $MDS1_VERSION -lt $(version_code 2.8.54) ] &&
skip "lustre < 2.8.54 does not contain LU-4825 fix"
remote_mds_nodsh && skip "remote MDS with nodsh"
local old_iused=$($LFS df -i --output=iused --ost=0)
echo "old_iused=$old_iused"
local mdts=$(mdts_nodes)
mkdir -p $DIR/$tdir
$LFS setstripe -i 0 -c 1 $DIR/$tdir
createmany -o $DIR/$tdir/$tfile. 1000
# statfs data is not real time, let's just calculate it
old_iused=$((old_iused + 1000))
echo "suppose current old_iused=$old_iused"
local count=$(do_facet $SINGLEMDS "$LCTL get_param -n \
osp.*OST0000*MDT0000.create_count")
local max_count=$(do_facet $SINGLEMDS "$LCTL get_param -n \
osp.*OST0000*MDT0000.max_create_count")
echo "create_count=$count, max_create_count=$max_count"
do_nodes $mdts "$LCTL set_param -n osp.*OST0000*.max_create_count=0"
$LFS setstripe -i 0 $DIR/$tdir/$tfile || error "setstripe failed"
local index=$($LFS getstripe -i $DIR/$tdir/$tfile)
[ $index -ne 0 ] || error "$tfile stripe index is 0"
unlinkmany $DIR/$tdir/$tfile. 1000
wait_delete_completed
wait_zfs_commit $SINGLEMDS 20
do_nodes $mdts "$LCTL set_param -n \
osp.*OST0000*.max_create_count=$max_count"
[ $MDS1_VERSION -lt $(version_code 2.12.51) ] &&
do_nodes $mdts "$LCTL set_param -n \
osp.*OST0000*.create_count=$count"
do_nodes $mdts "$LCTL get_param osp.*OST0000*.create_count" |
grep "=0" && error "create_count is zero"
local new_iused
for i in $(seq 120); do
new_iused=$($LFS df --output=iused --ost=0)
echo -n "$new_iused "
# system may be too busy to destroy all objs in time, use
# a somewhat small value to not fail autotest
((old_iused - new_iused > 400)) && break
sleep 1
done
echo -e "\nwaited $i sec, old Iused $old_iused, new Iused $new_iused"
((old_iused - new_iused > 400)) ||
error "objs not destroyed after unlink"
}
run_test 311 "disable OSP precreate, and unlink should destroy objs"
zfs_get_objid()
{
local ost=$1
local tf=$2
local fid=($($LFS getstripe $tf | grep 0x))
local seq=${fid[3]#0x}
local objid=${fid[1]}
local vdevdir=$(dirname $(facet_vdevice $ost))
local cmd="$ZDB -e -p $vdevdir -ddddd $(facet_device $ost)"
local zfs_zapid=$(do_facet $ost $cmd |
grep -w "/O/$seq/d$((objid%32))" -C 5 |
awk '/Object/{getline; print $1}')
local zfs_objid=$(do_facet $ost $cmd $zfs_zapid |
awk "/$objid = /"'{printf $3}')
echo $zfs_objid
}
zfs_object_blksz() {
local ost=$1
local objid=$2
local vdevdir=$(dirname $(facet_vdevice $ost))
local cmd="$ZDB -e -p $vdevdir -dddd $(facet_device $ost)"
local blksz=$(do_facet $ost $cmd $objid |
awk '/dblk/{getline; printf $4}')
case "${blksz: -1}" in
k|K) blksz=$((${blksz:0:$((${#blksz} - 1))}*1024)) ;;
m|M) blksz=$((${blksz:0:$((${#blksz} - 1))}*1024*1024)) ;;
*) ;;
esac
echo $blksz
}
test_312() { # LU-4856
remote_ost_nodsh && skip "remote OST with nodsh"
[[ "$ost1_FSTYPE" == "zfs" ]] || skip "the test only applies to zfs"
local max_blksz=$(do_facet ost1 \
$ZFS get -p recordsize $(facet_device ost1) |
awk '!/VALUE/{print $3}')
local tf=$DIR/$tfile
$LFS setstripe -c1 $tf
local facet="ost$(($($LFS getstripe -i $tf) + 1))"
# Get ZFS object id
local zfs_objid=$(zfs_get_objid $facet $tf)
# block size change by sequential overwrite
local bs
for ((bs=$PAGE_SIZE; bs <= max_blksz; bs *= 4)) ; do
dd if=/dev/zero of=$tf bs=$bs count=1 oflag=sync conv=notrunc
local blksz=$(zfs_object_blksz $facet $zfs_objid)
[[ $blksz -eq $bs ]] || error "blksz error: $blksz, expected: $bs"
done
rm -f $tf
$LFS setstripe -c1 $tf
facet="ost$(($($LFS getstripe -i $tf) + 1))"
# block size change by sequential append write
dd if=/dev/zero of=$tf bs=$PAGE_SIZE count=1 oflag=sync conv=notrunc
zfs_objid=$(zfs_get_objid $facet $tf)
local count
for ((count = 1; count < $((max_blksz / PAGE_SIZE)); count *= 2)); do
dd if=/dev/zero of=$tf bs=$PAGE_SIZE count=$count seek=$count \
oflag=sync conv=notrunc
blksz=$(zfs_object_blksz $facet $zfs_objid)
(( $blksz == 2 * count * PAGE_SIZE )) ||
error "blksz error, actual $blksz, " \
"expected: 2 * $count * $PAGE_SIZE"
done
rm -f $tf
# random write
$LFS setstripe -c1 $tf
facet="ost$(($($LFS getstripe -i $tf) + 1))"
zfs_objid=$(zfs_get_objid $facet $tf)
dd if=/dev/zero of=$tf bs=1K count=1 oflag=sync conv=notrunc
blksz=$(zfs_object_blksz $facet $zfs_objid)
(( blksz == PAGE_SIZE )) ||
error "blksz error: $blksz, expected: $PAGE_SIZE"
dd if=/dev/zero of=$tf bs=64K count=1 oflag=sync conv=notrunc seek=128
blksz=$(zfs_object_blksz $facet $zfs_objid)
(( blksz == 65536 )) || error "blksz error: $blksz, expected: 64k"
dd if=/dev/zero of=$tf bs=1M count=1 oflag=sync conv=notrunc
blksz=$(zfs_object_blksz $facet $zfs_objid)
(( blksz == 65536 )) || error "rewrite error: $blksz, expected: 64k"
}
run_test 312 "make sure ZFS adjusts its block size by write pattern"
test_313() {
remote_ost_nodsh && skip "remote OST with nodsh"
local file=$DIR/$tfile
rm -f $file
$LFS setstripe -c 1 -i 0 $file || error "setstripe failed"
# define OBD_FAIL_TGT_RCVD_EIO 0x720
do_facet ost1 "$LCTL set_param fail_loc=0x720"
dd if=/dev/zero of=$file bs=$PAGE_SIZE oflag=direct count=1 &&
error "write should failed"
do_facet ost1 "$LCTL set_param fail_loc=0"
rm -f $file
}
run_test 313 "io should fail after last_rcvd update fail"
test_314() {
[ $OSTCOUNT -lt 2 ] && skip "needs >= 2 OSTs"
$LFS setstripe -c 2 -i 0 $DIR/$tfile || error "setstripe failed"
do_facet ost1 "$LCTL set_param fail_loc=0x720"
rm -f $DIR/$tfile
wait_delete_completed
do_facet ost1 "$LCTL set_param fail_loc=0"
}
run_test 314 "OSP shouldn't fail after last_rcvd update failure"
test_315() { # LU-618
[ -f /proc/$$/io ] || skip_env "no IO accounting in kernel"
local file=$DIR/$tfile
rm -f $file
$MULTIOP $file oO_CREAT:O_DIRECT:O_RDWR:w4063232c ||
error "multiop file write failed"
$MULTIOP $file oO_RDONLY:r4063232_c &
PID=$!
sleep 2
local rbytes=$(awk '/read_bytes/ { print $2 }' /proc/$PID/io)
kill -USR1 $PID
[ $rbytes -gt 4000000 ] || error "read is not accounted ($rbytes)"
rm -f $file
}
run_test 315 "read should be accounted"
test_316() {
(( $MDSCOUNT >= 2 )) || skip "needs >= 2 MDTs"
large_xattr_enabled || skip "ea_inode feature disabled"
mkdir_on_mdt0 $DIR/$tdir || error "mkdir $tdir failed"
mkdir $DIR/$tdir/d || error "mkdir $tdir/d failed"
chown nobody $DIR/$tdir/d || error "chown $tdir/d failed"
touch $DIR/$tdir/d/$tfile || error "touch $tdir/d/$tfile failed"
$LFS migrate -m1 $DIR/$tdir/d || error "lfs migrate -m1 failed"
}
run_test 316 "lfs migrate of file with large_xattr enabled"
test_317() {
[ $MDS1_VERSION -lt $(version_code 2.11.53) ] &&
skip "Need MDS version at least 2.11.53"
if [ "$ost1_FSTYPE" == "zfs" ]; then
skip "LU-10370: no implementation for ZFS"
fi
local trunc_sz
local grant_blk_size
grant_blk_size=$($LCTL get_param osc.$FSNAME*.import |
awk '/grant_block_size:/ { print $2; exit; }')
#
# Create File of size 5M. Truncate it to below size's and verify
# blocks count.
#
dd if=/dev/zero of=$DIR/$tfile bs=5M count=1 conv=fsync ||
error "Create file $DIR/$tfile failed"
stack_trap "rm -f $DIR/$tfile" EXIT
for trunc_sz in 2097152 4097 4000 509 0; do
$TRUNCATE $DIR/$tfile $trunc_sz ||
error "truncate $tfile to $trunc_sz failed"
local sz=$(stat --format=%s $DIR/$tfile)
local blk=$(stat --format=%b $DIR/$tfile)
local trunc_blk=$((((trunc_sz + (grant_blk_size - 1) ) /
grant_blk_size) * 8))
if [[ $blk -ne $trunc_blk ]]; then
$(which stat) $DIR/$tfile
error "Expected Block $trunc_blk got $blk for $tfile"
fi
$CHECKSTAT -s $trunc_sz $DIR/$tfile ||
error "Expected Size $trunc_sz got $sz for $tfile"
done
#
# sparse file test
# Create file with a hole and write actual 65536 bytes which aligned
# with 4K and 64K PAGE_SIZE. Block count must be 128.
#
local bs=65536
dd if=/dev/zero of=$DIR/$tfile bs=$bs count=1 seek=5 conv=fsync ||
error "Create file : $DIR/$tfile"
#
# Truncate to size $trunc_sz bytes. Strip tail blocks and leave only 8
# blocks. The block count must drop to 8.
#
trunc_sz=$(($(stat --format=%s $DIR/$tfile) -
((bs - grant_blk_size) + 1)))
$TRUNCATE $DIR/$tfile $trunc_sz ||
error "truncate $tfile to $trunc_sz failed"
local trunc_bsz=$((grant_blk_size / $(stat --format=%B $DIR/$tfile)))
sz=$(stat --format=%s $DIR/$tfile)
blk=$(stat --format=%b $DIR/$tfile)
if [[ $blk -ne $trunc_bsz ]]; then
$(which stat) $DIR/$tfile
error "Expected Block $trunc_bsz got $blk for $tfile"
fi
$CHECKSTAT -s $trunc_sz $DIR/$tfile ||
error "Expected Size $trunc_sz got $sz for $tfile"
}
run_test 317 "Verify blocks get correctly update after truncate"
test_318() {
local llite_name="llite.$($LFS getname $MOUNT | awk '{print $1}')"
local old_max_active=$($LCTL get_param -n \
${llite_name}.max_read_ahead_async_active \
2>/dev/null)
$LCTL set_param llite.*.max_read_ahead_async_active=256
local max_active=$($LCTL get_param -n \
${llite_name}.max_read_ahead_async_active \
2>/dev/null)
[ $max_active -ne 256 ] && error "expected 256 but got $max_active"
$LCTL set_param llite.*.max_read_ahead_async_active=0 ||
error "set max_read_ahead_async_active should succeed"
$LCTL set_param llite.*.max_read_ahead_async_active=512
max_active=$($LCTL get_param -n \
${llite_name}.max_read_ahead_async_active 2>/dev/null)
[ $max_active -eq 512 ] || error "expected 512 but got $max_active"
# restore @max_active
[ $old_max_active -ne 0 ] && $LCTL set_param \
llite.*.max_read_ahead_async_active=$old_max_active
local old_threshold=$($LCTL get_param -n \
${llite_name}.read_ahead_async_file_threshold_mb 2>/dev/null)
local max_per_file_mb=$($LCTL get_param -n \
${llite_name}.max_read_ahead_per_file_mb 2>/dev/null)
local invalid=$(($max_per_file_mb + 1))
$LCTL set_param \
llite.*.read_ahead_async_file_threshold_mb=$invalid\
&& error "set $invalid should fail"
local valid=$(($invalid - 1))
$LCTL set_param \
llite.*.read_ahead_async_file_threshold_mb=$valid ||
error "set $valid should succeed"
local threshold=$($LCTL get_param -n \
${llite_name}.read_ahead_async_file_threshold_mb 2>/dev/null)
[ $threshold -eq $valid ] || error \
"expect threshold $valid got $threshold"
$LCTL set_param \
llite.*.read_ahead_async_file_threshold_mb=$old_threshold
}
run_test 318 "Verify async readahead tunables"
test_319() {
(( $MDSCOUNT >= 2 )) || skip "needs >= 2 MDTs"
local before=$(date +%s)
local evict
local mdir=$DIR/$tdir
local file=$mdir/xxx
$LFS mkdir -i0 $mdir || error "mkdir $mdir fails"
touch $file
#define OBD_FAIL_LDLM_LOCAL_CANCEL_PAUSE 0x32c
$LCTL set_param fail_val=5 fail_loc=0x8000032c
$LFS migrate -m1 $mdir &
sleep 1
dd if=$file of=/dev/null
wait
evict=$($LCTL get_param mdc.$FSNAME-MDT*.state |
awk -F"[ [,]" '/EVICTED ]$/ { if (mx<$5) {mx=$5;} } END { print mx }')
[ -z "$evict" ] || [[ $evict -le $before ]] || error "eviction happened"
}
run_test 319 "lost lease lock on migrate error"
test_350() {
local mdts=$(mdts_nodes)
mkdir $DIR/$tdir || error "mkdir $DIR/$tdir failed"
stack_trap "rm -r $DIR/$tdir"
#force 1/100 of replies to take "NID mismatch" codepath
#define CFS_FAIL_MATCH_MD_NID 0xe001 CFS_FAIL_SOME 0x10000000
do_nodes $mdts $LCTL set_param fail_loc=0x1000e001 fail_val=100
while ls -lR $DIR/$tdir > /dev/null; do :; done &
stack_trap "killall -9 ls || killall -9 ls"
cp -a /etc $DIR/$tdir || error "cp failed"
}
run_test 350 "force NID mismatch path to be exercised"
test_398a() { # LU-4198
local ost1_imp=$(get_osc_import_name client ost1)
local imp_name=$($LCTL list_param osc.$ost1_imp | head -n1 |
cut -d'.' -f2)
$LFS setstripe -c 1 -i 0 $DIR/$tfile
stack_trap "rm -f $DIR/$tfile"
$LCTL set_param ldlm.namespaces.*.lru_size=clear
# request a new lock on client
dd if=/dev/zero of=$DIR/$tfile bs=1M count=1
dd if=/dev/zero of=$DIR/$tfile bs=1M count=1 oflag=direct conv=notrunc
local lock_count=$($LCTL get_param -n \
ldlm.namespaces.$imp_name.lru_size)
[[ $lock_count -eq 0 ]] || error "lock should be cancelled by direct IO"
$LCTL set_param ldlm.namespaces.$imp_name.lru_size=clear
# no lock cached, should use lockless DIO and not enqueue new lock
dd if=/dev/zero of=$DIR/$tfile bs=1M count=1 oflag=direct conv=notrunc
lock_count=$($LCTL get_param -n \
ldlm.namespaces.$imp_name.lru_size)
[[ $lock_count -eq 0 ]] || error "no lock should be held by direct IO"
$LCTL set_param ldlm.namespaces.$imp_name.lru_size=clear
# no lock cached, should use locked DIO append
dd if=/dev/zero of=$DIR/$tfile bs=1M count=1 oflag=direct oflag=append \
conv=notrunc || error "DIO append failed"
lock_count=$($LCTL get_param -n \
ldlm.namespaces.$imp_name.lru_size)
[[ $lock_count -ne 0 ]] || error "lock still must be held by DIO append"
}
run_test 398a "direct IO should cancel lock otherwise lockless"
test_398b() { # LU-4198
local before=$(date +%s)
local njobs=4
local size=48
which fio || skip_env "no fio installed"
$LFS setstripe -c -1 -S 1M $DIR/$tfile
dd if=/dev/zero of=$DIR/$tfile bs=1M count=$size
# Single page, multiple pages, stripe size, 4*stripe size
for bsize in $(( $PAGE_SIZE )) $(( 4*$PAGE_SIZE )) 1048576 4194304; do
echo "mix direct rw ${bsize} by fio with $njobs jobs..."
fio --name=rand-rw --rw=randrw --bs=$bsize --direct=1 \
--numjobs=$njobs --fallocate=none \
--iodepth=16 --allow_file_create=0 \
--size=$((size/njobs))M \
--filename=$DIR/$tfile &
bg_pid=$!
echo "mix buffer rw ${bsize} by fio with $njobs jobs..."
fio --name=rand-rw --rw=randrw --bs=$bsize \
--numjobs=$njobs --fallocate=none \
--iodepth=16 --allow_file_create=0 \
--size=$((size/njobs))M \
--filename=$DIR/$tfile || true
wait $bg_pid
done
evict=$(do_facet client $LCTL get_param \
osc.$FSNAME-OST*-osc-*/state |
awk -F"[ [,]" '/EVICTED ]$/ { if (t<$5) {t=$5;} } END { print t }')
[ -z "$evict" ] || [[ $evict -le $before ]] ||
(do_facet client $LCTL get_param \
osc.$FSNAME-OST*-osc-*/state;
error "eviction happened: $evict before:$before")
rm -f $DIR/$tfile
}
run_test 398b "DIO and buffer IO race"
test_398c() { # LU-4198
local ost1_imp="osc.$(get_osc_import_name client ost1)"
local imp_name=($($LCTL list_param $ost1_imp | cut -d'.' -f2))
which fio || skip_env "no fio installed"
saved_debug=$($LCTL get_param -n debug)
$LCTL set_param debug=0
local size=($($LCTL get_param -n $ost1_imp.kbytesavail))
(( size /= 1024 )) # by megabytes
(( size /= 2 )) # write half of the OST at most
(( size <= 40 )) || size=40 #reduce test time anyway
$LFS setstripe -c 1 $DIR/$tfile
# it seems like ldiskfs reserves more space than necessary if the
# writing blocks are not mapped, so it extends the file firstly
dd if=/dev/zero of=$DIR/$tfile bs=1M count=$size && sync
cancel_lru_locks osc
# clear and verify rpc_stats later
$LCTL set_param $ost1_imp.rpc_stats=clear
local njobs=4
echo "writing ${size}M to OST0 by fio with $njobs jobs..."
fio --name=rand-write --rw=randwrite --bs=$PAGE_SIZE --direct=1 \
--numjobs=$njobs --fallocate=none --ioengine=libaio \
--iodepth=16 --allow_file_create=0 --size=$((size/njobs))M \
--filename=$DIR/$tfile
[ $? -eq 0 ] || error "fio write error"
(( $($LCTL get_param -n ldlm.namespaces.$imp_name.lock_count) == 0 )) ||
error "Locks were requested while doing AIO"
# get the percentage of 1-page I/O
pct=$($LCTL get_param osc.${imp_name}.rpc_stats |
grep -A 1 'pages per rpc' | grep -v 'pages per rpc' |
awk '{print $7}')
(( $pct <= 50 )) || {
$LCTL get_param osc.${imp_name}.rpc_stats
error "$pct% of I/O are 1-page"
}
echo "mix rw ${size}M to OST0 by fio with $njobs jobs..."
fio --name=rand-rw --rw=randrw --bs=$PAGE_SIZE --direct=1 \
--numjobs=$njobs --fallocate=none --ioengine=libaio \
--iodepth=16 --allow_file_create=0 --size=$((size/njobs))M \
--filename=$DIR/$tfile
[ $? -eq 0 ] || error "fio mixed read write error"
echo "AIO with large block size ${size}M"
fio --name=rand-rw --rw=randrw --bs=${size}M --direct=1 \
--numjobs=1 --fallocate=none --ioengine=libaio \
--iodepth=16 --allow_file_create=0 --size=${size}M \
--filename=$DIR/$tfile
[ $? -eq 0 ] || error "fio large block size failed"
rm -f $DIR/$tfile
$LCTL set_param debug="$saved_debug"
}
run_test 398c "run fio to test AIO"
test_398d() { # LU-13846
which aiocp || skip_env "no aiocp installed"
local aio_file=$DIR/$tfile.aio
$LFS setstripe -c -1 -S 1M $DIR/$tfile $aio_file
dd if=/dev/urandom of=$DIR/$tfile bs=1M count=64
aiocp -a $PAGE_SIZE -b 64M -s 64M -f O_DIRECT $DIR/$tfile $aio_file
stack_trap "rm -f $DIR/$tfile $aio_file"
diff $DIR/$tfile $aio_file || error "file diff after aiocp"
# test memory unaligned aio
# LU-18032 - unaligned AIO is disabled
#aiocp -a 512 -b 64M -s 64M -f O_DIRECT $DIR/$tfile $aio_file ||
# error "unaligned aio failed"
#diff $DIR/$tfile $aio_file || error "file diff after aiocp"
rm -f $DIR/$tfile $aio_file
}
run_test 398d "run aiocp to verify block size > stripe size"
test_398e() {
dd if=/dev/zero of=$DIR/$tfile bs=1234 count=1
touch $DIR/$tfile.new
dd if=$DIR/$tfile of=$DIR/$tfile.new bs=1M count=1 oflag=direct
}
run_test 398e "O_Direct open cleared by fcntl doesn't cause hang"
test_398f() { # LU-14687
which aiocp || skip_env "no aiocp installed"
local aio_file=$DIR/$tfile.aio
$LFS setstripe -c -1 -S 1M $DIR/$tfile $aio_file
dd if=/dev/zero of=$DIR/$tfile bs=1M count=64
stack_trap "rm -f $DIR/$tfile $aio_file"
#define OBD_FAIL_LLITE_PAGE_ALLOC 0x1418
$LCTL set_param fail_loc=0x1418
# make sure we don't crash and fail properly
aiocp -b 64M -s 64M -f O_DIRECT $DIR/$tfile $aio_file &&
error "aio with page allocation failure succeeded"
$LCTL set_param fail_loc=0
diff $DIR/$tfile $aio_file
[[ $? != 0 ]] || error "no diff after failed aiocp"
}
run_test 398f "verify aio handles ll_direct_rw_pages errors correctly"
# NB: To get the parallel DIO behavior in LU-13798, there must be > 1
# stripe and i/o size must be > stripe size
# Old style synchronous DIO waits after submitting each chunk, resulting in a
# single RPC in flight. This test shows async DIO submission is working by
# showing multiple RPCs in flight.
test_398g() { # LU-13798
$LFS setstripe -o 0,0 -S 1M $DIR/$tfile
# We need to do some i/o first to acquire enough grant to put our RPCs
# in flight; otherwise a new connection may not have enough grant
# available
dd if=/dev/urandom of=$DIR/$tfile bs=8M count=1 oflag=direct ||
error "parallel dio failed"
stack_trap "rm -f $DIR/$tfile"
# Reduce RPC size to 1M to avoid combination in to larger RPCs
local pages_per_rpc=$($LCTL get_param osc.*-OST0000-*.max_pages_per_rpc)
$LCTL set_param osc.*-OST0000-*.max_pages_per_rpc=1M
stack_trap "$LCTL set_param -n $pages_per_rpc"
# Recreate file so it's empty
rm -f $DIR/$tfile
$LFS setstripe -o 0,0 -S 1M $DIR/$tfile
#Pause rpc completion to guarantee we see multiple rpcs in flight
#define OBD_FAIL_OST_BRW_PAUSE_BULK
do_facet ost1 $LCTL set_param fail_loc=0x214 fail_val=2
stack_trap "do_facet ost1 $LCTL set_param fail_loc=0"
# Clear rpc stats
$LCTL set_param osc.*.rpc_stats=c
dd if=/dev/urandom of=$DIR/$tfile bs=8M count=1 oflag=direct ||
error "parallel dio failed"
stack_trap "rm -f $DIR/$tfile"
$LCTL get_param osc.*-OST0000-*.rpc_stats
pct=$($LCTL get_param osc.*-OST0000-*.rpc_stats |
grep -A 8 'rpcs in flight' | grep -v 'rpcs in flight' |
grep "8:" | awk '{print $8}')
# We look at the "8 rpcs in flight" field, and verify A) it is present
# and B) it includes all RPCs. This proves we had 8 RPCs in flight,
# as expected for an 8M DIO to a file with 1M stripes.
# NB: There is occasionally a mystery extra write RPC to a different
# file. I can't identify why that's happening, so we set up a margin
# of 1 RPC here, ie, 8/9 RPCs at this size, or ~88%
[ $pct -gt 87 ] || error "we should see 8 RPCs in flight"
# Verify turning off parallel dio works as expected
# Clear rpc stats
$LCTL set_param osc.*.rpc_stats=c
$LCTL set_param llite.*.parallel_dio=0
stack_trap '$LCTL set_param llite.*.parallel_dio=1'
dd if=/dev/urandom of=$DIR/$tfile bs=8M count=1 oflag=direct ||
error "dio with parallel dio disabled failed"
# Ideally, we would see only one RPC in flight here, but there is an
# unavoidable race between i/o completion and RPC in flight counting,
# so while only 1 i/o is in flight at a time, the RPC in flight counter
# will sometimes exceed 1 (3 or 4 is not rare on VM testing).
# So instead we just verify it's always < 8.
$LCTL get_param osc.*-OST0000-*.rpc_stats
ret=$($LCTL get_param osc.*-OST0000-*.rpc_stats |
grep -A 8 'rpcs in flight' | grep -v 'rpcs in flight' |
grep '^$' -B1 | grep . | awk '{print $1}')
[ $ret != "8:" ] ||
error "we should see fewer than 8 RPCs in flight (saw $ret)"
}
run_test 398g "verify parallel dio async RPC submission"
test_398h() { # LU-13798
local dio_file=$DIR/$tfile.dio
$LFS setstripe -C 2 -S 1M $DIR/$tfile $dio_file
dd if=/dev/urandom of=$DIR/$tfile bs=8M count=8 oflag=direct
stack_trap "rm -f $DIR/$tfile $dio_file"
dd if=$DIR/$tfile of=$dio_file bs=8M count=8 iflag=direct oflag=direct ||
error "parallel dio failed"
diff $DIR/$tfile $dio_file
[[ $? == 0 ]] || error "file diff after aiocp"
}
run_test 398h "verify correctness of read & write with i/o size >> stripe size"
test_398i() { # LU-13798
local dio_file=$DIR/$tfile.dio
$LFS setstripe -C 2 -S 1M $DIR/$tfile $dio_file
dd if=/dev/urandom of=$DIR/$tfile bs=8M count=8 oflag=direct
stack_trap "rm -f $DIR/$tfile $dio_file"
#define OBD_FAIL_LLITE_PAGE_ALLOC 0x1418
$LCTL set_param fail_loc=0x1418
# make sure we don't crash and fail properly
dd if=$DIR/$tfile of=$dio_file bs=8M count=8 iflag=direct oflag=direct &&
error "parallel dio page allocation failure succeeded"
diff $DIR/$tfile $dio_file
[[ $? != 0 ]] || error "no diff after failed aiocp"
}
run_test 398i "verify parallel dio handles ll_direct_rw_pages errors correctly"
test_398j() { # LU-13798
# Stripe size > RPC size but less than i/o size tests split across
# stripes and RPCs for individual i/o op
$LFS setstripe -o 0,0 -S 4M $DIR/$tfile $DIR/$tfile.2
# Reduce RPC size to 1M to guarantee split to multiple RPCs per stripe
local pages_per_rpc=$($LCTL get_param osc.*-OST0000-*.max_pages_per_rpc)
$LCTL set_param osc.*-OST0000-*.max_pages_per_rpc=1M
stack_trap "$LCTL set_param -n $pages_per_rpc"
dd if=/dev/urandom of=$DIR/$tfile bs=8M count=8 oflag=direct ||
error "parallel dio write failed"
stack_trap "rm -f $DIR/$tfile $DIR/$tfile.2"
dd if=$DIR/$tfile of=$DIR/$tfile.2 bs=8M count=8 iflag=direct ||
error "parallel dio read failed"
diff $DIR/$tfile $DIR/$tfile.2
[[ $? == 0 ]] || error "file diff after parallel dio read"
}
run_test 398j "test parallel dio where stripe size > rpc_size"
test_398k() { # LU-13798
wait_delete_completed
wait_mds_ost_sync
# 4 stripe file; we will cause out of space on OST0
$LFS setstripe -o 0,1,0,1 -S 1M $DIR/$tfile
# Fill OST0 (if it's not too large)
ORIGFREE=$($LCTL get_param -n lov.$FSNAME-clilov-*.kbytesavail |
head -n1)
if [[ $ORIGFREE -gt $MAXFREE ]]; then
skip "$ORIGFREE > $MAXFREE skipping out-of-space test on OST0"
fi
$LFS setstripe -i 0 -c 1 $DIR/$tfile.1
dd if=/dev/zero of=$DIR/$tfile.1 bs=1024 count=$MAXFREE &&
error "dd should fill OST0"
stack_trap "rm -f $DIR/$tfile.1"
dd if=/dev/urandom of=$DIR/$tfile bs=8M count=8 oflag=direct
err=$?
ls -la $DIR/$tfile
$CHECKSTAT -t file -s 0 $DIR/$tfile ||
error "file is not 0 bytes in size"
# dd above should not succeed, but don't error until here so we can
# get debug info above
[[ $err != 0 ]] ||
error "parallel dio write with enospc succeeded"
stack_trap "rm -f $DIR/$tfile"
}
run_test 398k "test enospc on first stripe"
test_398l() { # LU-13798
wait_delete_completed
wait_mds_ost_sync
# Clean up after the test
stack_trap wait_delete_completed
stack_trap wait_mds_ost_sync
# 4 stripe file; we will cause out of space on OST0
# Note the 1M stripe size and the > 1M i/o size mean this ENOSPC
# happens on the second i/o chunk we issue
$LFS setstripe -o 1,0,1,0 -S 1M $DIR/$tfile $DIR/$tfile.2
dd if=/dev/urandom of=$DIR/$tfile bs=8M count=2 oflag=direct
stack_trap "rm -f $DIR/$tfile"
# Fill OST0 (if it's not too large)
ORIGFREE=$($LCTL get_param -n lov.$FSNAME-clilov-*.kbytesavail |
head -n1)
if [[ $ORIGFREE -gt $MAXFREE ]]; then
skip "$ORIGFREE > $MAXFREE skipping out-of-space test on OST0"
fi
$LFS setstripe -i 0 -c 1 $DIR/$tfile.1
dd if=/dev/zero of=$DIR/$tfile.1 bs=1024 count=$MAXFREE &&
error "dd should fill OST0"
stack_trap "rm -f $DIR/$tfile.1"
dd if=$DIR/$tfile of=$DIR/$tfile.2 bs=8M count=8 oflag=direct
err=$?
stack_trap "rm -f $DIR/$tfile.2"
# Check that short write completed as expected
ls -la $DIR/$tfile.2
$CHECKSTAT -t file -s 1048576 $DIR/$tfile.2 ||
error "file is not 1M in size"
# dd above should not succeed, but don't error until here so we can
# get debug info above
[[ $err != 0 ]] ||
error "parallel dio write with enospc succeeded"
# Truncate source file to same length as output file and diff them
$TRUNCATE $DIR/$tfile 1048576
diff $DIR/$tfile $DIR/$tfile.2
[[ $? == 0 ]] || error "data incorrect after short write"
}
run_test 398l "test enospc on intermediate stripe/RPC"
test_398m() { # LU-13798
$LFS setstripe -o 0,1,0,1 -S 1M $DIR/$tfile
# Set up failure on OST0, the first stripe:
#define OBD_FAIL_OST_BRW_WRITE_BULK 0x20e
#NB: Fail val is ost # + 1, because we cannot use cfs_fail_val = 0
# OST0 is on ost1, OST1 is on ost2.
# So this fail_val specifies OST0
do_facet ost1 $LCTL set_param fail_loc=0x20e fail_val=1
stack_trap "do_facet ost1 $LCTL set_param fail_loc=0"
dd if=/dev/urandom of=$DIR/$tfile bs=8M count=8 oflag=direct &&
error "parallel dio write with failure on first stripe succeeded"
stack_trap "rm -f $DIR/$tfile"
do_facet ost1 $LCTL set_param fail_loc=0 fail_val=0
# Place data in file for read
dd if=/dev/urandom of=$DIR/$tfile bs=8M count=8 oflag=direct ||
error "parallel dio write failed"
# Fail read on OST0, first stripe
#define OBD_FAIL_OST_BRW_READ_BULK 0x20f
do_facet ost1 $LCTL set_param fail_loc=0x20f fail_val=1
dd if=$DIR/$tfile of=$DIR/$tfile.2 bs=8M count=8 iflag=direct &&
error "parallel dio read with error on first stripe succeeded"
rm -f $DIR/$tfile.2
do_facet ost1 $LCTL set_param fail_loc=0 fail_val=0
# Switch to testing on OST1, second stripe
# Clear file contents, maintain striping
echo > $DIR/$tfile
# Set up failure on OST1, second stripe:
do_facet ost2 $LCTL set_param fail_loc=0x20e fail_val=2
stack_trap "do_facet ost2 $LCTL set_param fail_loc=0"
dd if=/dev/urandom of=$DIR/$tfile bs=8M count=8 oflag=direct &&
error "parallel dio write with failure on second stripe succeeded"
stack_trap "rm -f $DIR/$tfile"
do_facet ost2 $LCTL set_param fail_loc=0 fail_val=0
# Place data in file for read
dd if=/dev/urandom of=$DIR/$tfile bs=8M count=8 oflag=direct ||
error "parallel dio write failed"
# Fail read on OST1, second stripe
#define OBD_FAIL_OST_BRW_READ_BULK 0x20f
do_facet ost2 $LCTL set_param fail_loc=0x20f fail_val=2
dd if=$DIR/$tfile of=$DIR/$tfile.2 bs=8M count=8 iflag=direct &&
error "parallel dio read with error on second stripe succeeded"
rm -f $DIR/$tfile.2
do_facet ost2 $LCTL set_param fail_loc=0 fail_val=0
}
run_test 398m "test RPC failures with parallel dio"
# Parallel submission of DIO should not cause problems for append, but it's
# important to verify.
test_398n() { # LU-13798
$LFS setstripe -C 2 -S 1M $DIR/$tfile
dd if=/dev/urandom of=$DIR/$tfile bs=8M count=8 ||
error "dd to create source file failed"
stack_trap "rm -f $DIR/$tfile"
dd if=$DIR/$tfile of=$DIR/$tfile.1 bs=8M count=8 oflag=direct oflag=append ||
error "parallel dio write with failure on second stripe succeeded"
stack_trap "rm -f $DIR/$tfile $DIR/$tfile.1"
diff $DIR/$tfile $DIR/$tfile.1
[[ $? == 0 ]] || error "data incorrect after append"
}
run_test 398n "test append with parallel DIO"
test_398o() {
directio rdwr $DIR/$tfile 0 1 1 || error "bad KMS"
}
run_test 398o "right kms with DIO"
test_398p()
{
(( $OSTCOUNT >= 2 )) || skip "needs >= 2 OSTs"
which aiocp || skip_env "no aiocp installed"
local stripe_size=$((1024 * 1024)) #1 MiB
# Max i/o below is ~ 4 * stripe_size, so this gives ~5 i/os
local file_size=$((25 * stripe_size))
$LFS setstripe -c 2 -S $stripe_size $DIR/$tfile.1
stack_trap "rm -f $DIR/$tfile*"
# Just a bit bigger than the largest size in the test set below
dd if=/dev/urandom bs=$file_size count=1 of=$DIR/$tfile.1 ||
error "buffered i/o to create file failed"
for bs in $PAGE_SIZE $((PAGE_SIZE * 4)) $stripe_size \
$((stripe_size * 4)); do
$LFS setstripe -c 2 -S $stripe_size $DIR/$tfile.2
echo "bs: $bs, file_size $file_size"
aiocp -a $PAGE_SIZE -b $bs -s $file_size -f O_DIRECT \
$DIR/$tfile.1 $DIR/$tfile.2 &
pid_dio1=$!
# Buffered I/O with similar but not the same block size
dd if=$DIR/$tfile.1 bs=$((bs * 2)) of=$DIR/$tfile.2 \
conv=notrunc &
pid_bio2=$!
wait $pid_dio1
rc1=$?
wait $pid_bio2
rc2=$?
if (( rc1 != 0 )); then
error "aio copy 1 w/bsize $bs failed: $rc1"
fi
if (( rc2 != 0 )); then
error "buffered copy 2 w/bsize $bs failed: $rc2"
fi
$CHECKSTAT -t file -s $file_size $DIR/$tfile.2 ||
error "size incorrect"
cmp --verbose $DIR/$tfile.1 $DIR/$tfile.2 ||
error "files differ, bsize $bs"
rm -f $DIR/$tfile.2
done
}
run_test 398p "race aio with buffered i/o"
test_398q()
{
(( $OSTCOUNT >= 2 )) || skip "needs >= 2 OSTs"
local stripe_size=$((1024 * 1024)) #1 MiB
# Max i/o below is ~ 4 * stripe_size, so this gives ~5 i/os
local file_size=$((25 * stripe_size))
$LFS setstripe -c 2 -S $stripe_size $DIR/$tfile.1
$LFS setstripe -c 2 -S $stripe_size $DIR/$tfile.2
# Just a bit bigger than the largest size in the test set below
dd if=/dev/urandom bs=$file_size count=1 of=$DIR/$tfile.1 ||
error "buffered i/o to create file failed"
for bs in $PAGE_SIZE $((PAGE_SIZE * 4)) $stripe_size \
$((stripe_size * 4)); do
echo "bs: $bs, file_size $file_size"
dd if=$DIR/$tfile.1 bs=$((bs *2 )) of=$DIR/$tfile.2 \
conv=notrunc oflag=direct iflag=direct &
pid_dio1=$!
# Buffered I/O with similar but not the same block size
dd if=$DIR/$tfile.1 bs=$((bs * 2)) of=$DIR/$tfile.2 \
conv=notrunc &
pid_bio2=$!
wait $pid_dio1
rc1=$?
wait $pid_bio2
rc2=$?
if (( rc1 != 0 )); then
error "dio copy 1 w/bsize $bs failed: $rc1"
fi
if (( rc2 != 0 )); then
error "buffered copy 2 w/bsize $bs failed: $rc2"
fi
$CHECKSTAT -t file -s $file_size $DIR/$tfile.2 ||
error "size incorrect"
diff $DIR/$tfile.1 $DIR/$tfile.2 ||
error "files differ, bsize $bs"
done
rm -f $DIR/$tfile*
}
run_test 398q "race dio with buffered i/o"
test_398r() {
$LFS setstripe -i 0 -c 1 $DIR/$tfile || error "setstripe failed"
echo "hello, world" > $DIR/$tfile
cancel_lru_locks osc
#define OBD_FAIL_OST_BRW_READ_BULK 0x20f
do_facet ost1 $LCTL set_param fail_loc=0x20f
cat $DIR/$tfile > /dev/null && error "cat should fail"
return 0
}
run_test 398r "i/o error on file read"
test_398s() {
[[ $OSTCOUNT -ge 2 && "$ost1_HOST" = "$ost2_HOST" ]] ||
skip "remote OST"
$LFS mirror create -N -i 0 -c 1 -N -i 1 -c 1 $DIR/$tfile ||
error "mirror create failed"
echo "hello, world" > $DIR/$tfile
$LFS mirror resync $DIR/$tfile || error "mirror resync failed"
cancel_lru_locks osc
#define OBD_FAIL_OST_BRW_READ_BULK 0x20f
do_facet ost1 $LCTL set_param fail_loc=0x20f
cat $DIR/$tfile > /dev/null && error "cat should fail"
return 0
}
run_test 398s "i/o error on mirror file read"
test_398t() { # LU-19536
unaligned_dio_or_skip
$LFS setstripe -c 1 $DIR/$tfile || error "setstripe failed"
# Use a non-page-aligned block size to trigger unaligned
# DIO path. The fail_loc makes ll_allocate_dio_buffer
# fail after allocating the pages array, exercising the
# cleanup path which had a double-free bug.
#define OBD_FAIL_LLITE_DIO_BUFFER_ALLOC 0x1438
$LCTL set_param fail_loc=0x80001438
dd if=/dev/zero of=$DIR/$tfile bs=1024 count=64 oflag=direct 2>/dev/null
$LCTL set_param fail_loc=0
# write may fail or succeed (if retried), but must not crash
# verify the system is still functional
echo "sanity check" > $DIR/$tfile || error "write after fail_loc failed"
}
run_test 398t "DIO buffer alloc failure must not crash (double-free)"
test_398u() { # LU-19536
unaligned_dio_or_skip
# Force ENOMEM mid-IO to exercise the drain+retry path.
#
# OBD_FAIL_LLITE_DIO_DRAIN_RETRY (0x1439) does two things:
# 1) PRECHECK caps each sub_dio to PAGE_SIZE, so a single
# write() generates many loop iterations with in-flight
# sub_dios.
# 2) CFS_FAIL_CHECK with SKIP|ONCE triggers ENOMEM in
# ll_allocate_dio_buffer after fail_val successes.
#
# With fail_val=5, iterations 1-5 succeed (tot_bytes > 0),
# iteration 6 gets ENOMEM, and the drain+retry path fires.
$LFS setstripe -c -1 $DIR/$tfile || error "setstripe"
#define OBD_FAIL_LLITE_DIO_DRAIN_RETRY 0x1439
$LCTL set_param fail_loc=0xa0001439 fail_val=5
# bs=4608 (4096+512): not page-aligned but 512-aligned for
# O_DIRECT. write 1 at offset 0 is page-aligned (regular
# DIO), write 2 at offset 4608 is unaligned. With sub_dios
# capped to PAGE_SIZE, each unaligned write generates 2 loop
# iterations (4096+512), so 3 unaligned writes = 6 alloc
# calls. fail_val=5 skips 5, fails the 6th — at which
# point tot_bytes > 0 (one iteration succeeded in this
# call), so drain+retry fires.
dd if=/dev/zero of=$DIR/$tfile bs=$((PAGE_SIZE+512)) count=10 \
oflag=direct || error "dd with drain retry failed"
$LCTL set_param fail_loc=0
local sz=$(stat -c %s $DIR/$tfile)
(( sz == $((10 * (PAGE_SIZE+512))) )) ||
error "file size $sz != expected $((10 * 4608))"
}
run_test 398u "DIO pool ENOMEM triggers drain and retry"
test_fake_rw() {
local read_write=$1
if [ "$read_write" = "write" ]; then
local dd_cmd="dd if=/dev/zero of=$DIR/$tfile"
elif [ "$read_write" = "read" ]; then
local dd_cmd="dd of=/dev/null if=$DIR/$tfile"
else
error "argument error"
fi
# turn off debug for performance testing
local saved_debug=$($LCTL get_param -n debug)
$LCTL set_param debug=0
$LFS setstripe -c 1 -i 0 $DIR/$tfile
# get ost1 size - $FSNAME-OST0000
local ost1_avail_size=$($LFS df $DIR | awk /${ost1_svc}/'{ print $4 }')
local blocks=$((ost1_avail_size/2/1024)) # half avail space by megabytes
[ $blocks -gt 1000 ] && blocks=1000 # 1G in maximum
if [ "$read_write" = "read" ]; then
$TRUNCATE $DIR/$tfile $(expr 1048576 \* $blocks)
fi
local start_time=$(date +%s.%N)
$dd_cmd bs=1M count=$blocks oflag=sync ||
error "real dd $read_write error"
local duration=$(bc <<< "$(date +%s.%N) - $start_time")
if [ "$read_write" = "write" ]; then
rm -f $DIR/$tfile
fi
# define OBD_FAIL_OST_FAKE_RW 0x238
do_facet ost1 $LCTL set_param fail_loc=0x238
local start_time=$(date +%s.%N)
$dd_cmd bs=1M count=$blocks oflag=sync ||
error "fake dd $read_write error"
local duration_fake=$(bc <<< "$(date +%s.%N) - $start_time")
if [ "$read_write" = "write" ]; then
# verify file size
cancel_lru_locks osc
$CHECKSTAT -t file -s $((blocks * 1024 * 1024)) $DIR/$tfile ||
error "$tfile size not $blocks MB"
fi
do_facet ost1 $LCTL set_param fail_loc=0
echo "fake $read_write $duration_fake vs. normal $read_write" \
"$duration in seconds"
[ $(bc <<< "$duration_fake < $duration") -eq 1 ] ||
error_not_in_vm "fake write is slower"
$LCTL set_param -n debug="$saved_debug"
rm -f $DIR/$tfile
}
test_399a() { # LU-7655 for OST fake write
remote_ost_nodsh && skip "remote OST with nodsh"
test_fake_rw write
}
run_test 399a "fake write should not be slower than normal write"
test_399b() { # LU-8726 for OST fake read
remote_ost_nodsh && skip "remote OST with nodsh"
test_fake_rw read
}
run_test 399b "fake read should not be slower than normal read"
test_400a() { # LU-1606, was conf-sanity test_74
if ! which $CC > /dev/null 2>&1; then
skip_env "$CC is not installed"
fi
local extra_flags=''
local out=$TMP/$tfile
local prefix=/usr/include/lustre
local prog
# Oleg removes .c files in his test rig so test if any c files exist
[[ -n "$(ls -A $LUSTRE_TESTS_API_DIR)" ]] ||
skip_env "Needed .c test files are missing"
if ! [[ -d $prefix ]]; then
# Assume we're running in tree and fixup the include path.
extra_flags+=" -I$LUSTRE/../lnet/include/uapi"
extra_flags+=" -I$LUSTRE/include/uapi -I$LUSTRE/include"
extra_flags+=" -L$LUSTRE/utils/.libs"
fi
for prog in $LUSTRE_TESTS_API_DIR/*.c; do
$CC -Wall -Werror $extra_flags -o $out $prog -llustreapi ||
error "client api broken"
done
rm -f $out
}
run_test 400a "Lustre client api program can compile and link"
test_400b() { # LU-1606, LU-5011
local header
local out=$TMP/$tfile
local prefix=/usr/include/linux/lustre
# We use a hard coded prefix so that this test will not fail
# when run in tree. There are headers in lustre/include/lustre/
# that are not packaged (like lustre_idl.h) and have more
# complicated include dependencies (like config.h and lnet/types.h).
# Since this test about correct packaging we just skip them when
# they don't exist (see below) rather than try to fixup cppflags.
if ! which $CC > /dev/null 2>&1; then
skip_env "$CC is not installed"
fi
for header in $prefix/*.h; do
if ! [[ -f "$header" ]]; then
continue
fi
if [[ "$(basename $header)" == lustre_ioctl.h ]]; then
continue # lustre_ioctl.h is internal header
fi
$CC -Wall -Werror -include $header -c -x c /dev/null -o $out ||
error "cannot compile '$header'"
done
rm -f $out
}
run_test 400b "packaged headers can be compiled"
test_401a() { #LU-7437
local printf_arg=$(find -printf 2>&1 | grep "unrecognized:")
[ -n "$printf_arg" ] && skip_env "find does not support -printf"
#count the number of parameters by "list_param -R"
local params=$($LCTL list_param -R '*' 2>/dev/null | wc -l)
#count the number of parameters by listing proc files
local proc_dirs=$(eval \ls -d $proc_regexp 2>/dev/null)
echo "proc_dirs='$proc_dirs'"
[ -n "$proc_dirs" ] || error "no proc_dirs on $HOSTNAME"
local procs=$(find -L $proc_dirs -mindepth 1 -printf '%P\n' 2>/dev/null|
sort -u | wc -l)
[ $params -eq $procs ] ||
error "found $params parameters vs. $procs proc files"
# test the list_param -D option only returns directories
params=$($LCTL list_param -R -D '*' 2>/dev/null | wc -l)
#count the number of parameters by listing proc directories
procs=$(find -L $proc_dirs -mindepth 1 -type d -printf '%P\n' 2>/dev/null |
sort -u | wc -l)
[ $params -eq $procs ] ||
error "found $params parameters vs. $procs proc files '-D'"
}
run_test 401a "Verify if 'lctl list_param -R' can list parameters recursively"
test_401aa() {
$LCTL list_param -p osc.* | while read path; do
[[ -r $path ]] && echo "$path" || error "'$path' does not exist"
done
}
run_test 401aa "Verify that 'lctl list_param -p' lists the correct path names"
test_401ab() {
$LCTL list_param -pr "*" | while read path; do
[[ -r $path ]] || error "'$path' not readable"
done
$LCTL list_param -pLr "*" | while read path; do
[[ -r $path ]] || error "'$path' not readable (--no-links)"
done
}
run_test 401ab "Check that 'lctl list_param -r' lists only readable params"
test_401ac() {
$LCTL list_param -pw "*" | while read path; do
[[ -w $path ]] || error "'$path' not writable"
done
$LCTL list_param -pLw "*" | while read path; do
[[ -w $path ]] || error "'$path' not writable (--no-links)"
done
}
run_test 401ac "Check that 'lctl list_param -w' lists only writable params"
test_401ad() {
$LCTL list_param -prw "*" | while read path; do
[[ -r $path && -w $path ]] ||
error "'$path' not readable and writable"
done
$LCTL list_param -pLrw "*" | while read path; do
[[ -r $path && -w $path ]] ||
error "'$path' not readable and writable (--no-links)"
done
}
run_test 401ad "Check that 'lctl list_param -wr' is conjunctive"
test_401b() {
# jobid_var may not allow arbitrary values, so use jobid_name
# if available
if $LCTL list_param jobid_name > /dev/null 2>&1; then
local jobvarname=jobid_name tmp='testing%p'
else
local jobvarname=jobid_var tmp=testing
fi
local save=$($LCTL get_param -n $jobvarname)
$LCTL set_param foo=bar $jobvarname=$tmp bar=baz &&
error "no error returned when setting bad parameters"
local jobid_new=$($LCTL get_param -n foe $jobvarname baz)
[[ "$jobid_new" == "$tmp" ]] || error "jobid tmp $jobid_new != $tmp"
$LCTL set_param -n fog=bam $jobvarname=$save bat=fog
local jobid_old=$($LCTL get_param -n foe $jobvarname bag)
[[ "$jobid_old" == "$save" ]] || error "jobid new $jobid_old != $save"
}
run_test 401b "Verify 'lctl {get,set}_param' continue after error"
test_401c() {
# jobid_var may not allow arbitrary values, so use jobid_name
# if available
if $LCTL list_param jobid_name > /dev/null 2>&1; then
local jobvarname=jobid_name
else
local jobvarname=jobid_var
fi
local jobid_var_old=$($LCTL get_param -n $jobvarname)
local jobid_var_new
$LCTL set_param $jobvarname= &&
error "no error returned for 'set_param a='"
jobid_var_new=$($LCTL get_param -n $jobvarname)
[[ "$jobid_var_old" == "$jobid_var_new" ]] ||
error "$jobvarname was changed by setting without value"
$LCTL set_param $jobvarname &&
error "no error returned for 'set_param a'"
jobid_var_new=$($LCTL get_param -n $jobvarname)
[[ "$jobid_var_old" == "$jobid_var_new" ]] ||
error "$jobvarname was changed by setting without value"
}
run_test 401c "Verify 'lctl set_param' without value fails in either format."
test_401d() {
# jobid_var may not allow arbitrary values, so use jobid_name
# if available
if $LCTL list_param jobid_name > /dev/null 2>&1; then
local jobvarname=jobid_name new_value='foo=bar%p'
else
local jobvarname=jobid_var new_value=foo=bar
fi
local jobid_var_old=$($LCTL get_param -n $jobvarname)
local jobid_var_new
$LCTL set_param $jobvarname=$new_value ||
error "'set_param a=b' did not accept a value containing '='"
jobid_var_new=$($LCTL get_param -n $jobvarname)
[[ "$jobid_var_new" == "$new_value" ]] ||
error "'set_param a=b' failed on a value containing '='"
# Reset the $jobvarname to test the other format
$LCTL set_param $jobvarname=$jobid_var_old
jobid_var_new=$($LCTL get_param -n $jobvarname)
[[ "$jobid_var_new" == "$jobid_var_old" ]] ||
error "failed to reset $jobvarname"
$LCTL set_param $jobvarname $new_value ||
error "'set_param a b' did not accept a value containing '='"
jobid_var_new=$($LCTL get_param -n $jobvarname)
[[ "$jobid_var_new" == "$new_value" ]] ||
error "'set_param a b' failed on a value containing '='"
$LCTL set_param $jobvarname $jobid_var_old
jobid_var_new=$($LCTL get_param -n $jobvarname)
[[ "$jobid_var_new" == "$jobid_var_old" ]] ||
error "failed to reset $jobvarname"
}
run_test 401d "Verify 'lctl set_param' accepts values containing '='"
cleanup_401db() {
local saved_val=$1
echo "start cleanup... "
do_facet mgs $LCTL set_param -P at_min="$saved_val"
wait_update $HOSTNAME "$LCTL get_param -n at_min" "$saved_val"
echo "done"
}
test_401db() { #LU-9544
local param="at_min"
local saved_val=$($LCTL get_param -n $param)
local new_val=$((saved_val + 6))
stack_trap "cleanup_401db $saved_val"
do_facet mgs $LCTL set_param -P $param=$new_val ||
error "failed to set $param=$new_val"
wait_update $HOSTNAME "$LCTL get_param -n $param" $new_val
local expected=$($LCTL get_param -n $param)
do_facet mgs $LCTL set_param -P -d $param=$new_val ||
error "failed to delete $param"
echo "Wait for erroneous changes"
wait_update_cond $HOSTNAME "$LCTL get_param -n $param" != $new_val
local result=$($LCTL get_param -n $param)
! [[ "$result" =~ "=" ]] || {
echo "result:$result"
error "'lctl set_param -P -d a=b' added trailing '='"
}
[[ $result == $expected ]] || {
echo -e "result:$result\nexpected:$expected"
error "'lctl set_param -P -d a=b' changed value of a"
}
}
run_test 401db "Verify 'lctl set_param' does not add trailing '='"
test_401e() { # LU-14779
$LCTL list_param -R "ldlm.namespaces.MGC*" ||
error "lctl list_param MGC* failed"
$LCTL get_param "ldlm.namespaces.MGC*" || error "lctl get_param failed"
$LCTL get_param "ldlm.namespaces.MGC*.lru_size" ||
error "lctl get_param lru_size failed"
}
run_test 401e "verify 'lctl get_param' works with NID in parameter"
test_401fa() {
$LCTL list_param -RpL "*" | while read path; do
[[ ! -L $path ]] ||
error "list_param -RpL returned the symlink: '$path'"
done
}
run_test 401fa "check 'lctl list_param' doesn't follow symlinks with --no-links"
test_401fb() {
local modules=$(ls /sys/module/{lnet,osc,mdd,obdclass,ofd,ptlrpc,mgc,ksocklnd,mdt,osd_ldiskfs,lquota}/parameters/* | wc -l)
local params=$($LCTL list_param -M "*" | wc -l)
local module_params=$($LCTL list_param -M --module "*" | wc -l)
(( modules + params == module_params )) ||
error "expected $((modules+params)) params, got $module_params"
}
run_test 401fb "check 'lctl {get,list,set}_param' only sees modules params with --module"
test_401ga() {
local paramdir=/etc/lustre
local fsnamefile=$paramdir/mount.$FSNAME.params
local clientfile=$paramdir/mount.client.params
local param1="osc.$FSNAME-OST0000-*.max_dirty_mb"
local param2="at_max"
local value1_new value2_new
local value1_old=$($LCTL get_param -n $param1)
local value2_old=$($LCTL get_param -n $param2)
local value1_client=$(( value1_old - 1 ))
local value1_fsname=$(( value1_old - 3 ))
local value2_client=$(( value2_old - 10 ))
local value2_fsname=$(( value2_old - 13 ))
stack_trap "$LCTL set_param $param1=$value1_old"
stack_trap "$LCTL set_param $param2=$value2_old"
if [[ -e $fsnamefile ]]; then
mv $fsnamefile $fsnamefile.$TESTNAME
stack_trap "mv $fsnamefile.$TESTNAME $fsnamefile"
else
stack_trap "rm -f $fsnamefile"
fi
if [[ -e $clientfile ]]; then
mv $clientfile $clientfile.$TESTNAME
stack_trap "mv $clientfile.$TESTNAME $clientfile"
else
stack_trap "rm -f $clientfile"
fi
# do not allow bad params to be set
$LCTL set_param --client $TESTNAME=$TESTNAME &&
error "set $TESTNAME as param in $clientfile" || true
# set garbage val to be overwritten
$LCTL set_param --client $param1=$TESTNAME ||
error "failed to set $param1 in $clientfile"
$LCTL set_param --client $param1=$value1_client ||
error "failed to overwrite $param1 in $clientfile"
# nothing should be changed
$LCTL set_param --client $param1=$value1_client ||
error "error setting duplicate param $param1 in $clientfile"
$LCTL set_param --client=$FSNAME $param1=$value1_fsname ||
error "failed to set $param1 in $fsnamefile"
$LCTL set_param --client=$FSNAME $param2=$value2_fsname &&
error "set $param2 in $fsnamefile without containing $FSNAME" ||
true
$LCTL set_param --client $TESTNAME= $param2=$value2_client || true
grep -q "$param2=$value2_client" $clientfile ||
error "set_param -C did not continue after error"
remount_client $MOUNT
value1_new=$($LCTL get_param -n $param1)
value2_new=$($LCTL get_param -n $param2)
[[ "$value1_new" == "$value1_client" ]] &&
error "$param1 not $value1_fsname, got client param $value1_new"
[[ "$value1_new" == "$value1_fsname" ]] ||
error "$param1 not $FSNAME param $value1_fsname, got $value1_new"
[[ "$value2_new" == "$value2_client" ]] ||
error "$param2 not client param $value2_client, got $value2_new"
}
run_test 401ga "check 'set_param -C' sets params upon mount"
test_401gb() {
local paramdir=/etc/lustre
local fsnamefile=$paramdir/mount.$FSNAME.params
local clientfile=$paramdir/mount.client.params
local param1="osc.$FSNAME-OST0000-*.max_dirty_mb"
local param2="at_max"
local value1_new value2_new
local value1_old=$($LCTL get_param -n $param1)
local value2_old=$($LCTL get_param -n $param2)
local value1_client=$(( value1_old - 1 ))
local value1_fsname=$(( value1_old - 3 ))
local value2_client=$(( value2_old - 10 ))
stack_trap "$LCTL set_param $param1=$value1_old"
stack_trap "$LCTL set_param $param2=$value2_old"
if [[ -e $fsnamefile ]]; then
mv $fsnamefile $fsnamefile.$TESTNAME
stack_trap "mv $fsnamefile.$TESTNAME $fsnamefile"
else
stack_trap "rm $fsnamefile"
fi
if [[ -e $clientfile ]]; then
mv $clientfile $clientfile.$TESTNAME
stack_trap "mv $clientfile.$TESTNAME $clientfile"
else
stack_trap "rm $clientfile"
fi
$LCTL set_param --client $param1=$value1_client $param2=$value2_client
$LCTL set_param --client=$FSNAME $param1=$value1_fsname
$LCTL set_param -d --client $TESTNAME=$TESTNAME ||
error "error deleting bad param $TESTNAME from $clientfile"
$LCTL set_param -d --client $param2 ||
error "error deleting $param2 from $clientfile"
$LCTL set_param -d --client=$FSNAME $param1 $param2 ||
error "error deleting $param1 and $param2 from $fsnamefile"
remount_client $MOUNT
value1_new=$($LCTL get_param -n $param1)
value2_new=$($LCTL get_param -n $param2)
[[ "$value1_new" == "$value1_fsname" ]] &&
error "$param1 not $value1_client, got $FSNAME param $value1_new"
[[ "$value1_new" == "$value1_client" ]] ||
error "$param1 not client param $value1_client, got $value1_new"
[[ "$value2_new" == "$value2_old" ]] ||
error "$param2 not $value2_old, got $value2_new"
}
run_test 401gb "check 'set_param -d -C' removes client params"
test_401gc() {
local expected=$($LCTL list_param -R "*" | grep -c brw_stats)
local actual=$($LCTL find_param -N brw_stats | wc -l)
(( expected == actual )) ||
error "expected $expected, but found $actual"
expected="$($LCTL get_param checksum_speed)"
actual="$($LCTL find_param check.*speed)"
[[ "$expected" =~ "$actual" ]] ||
error "expected '$expected', but found '$actual'"
}
run_test 401gc "check 'lctl find_param' can find params using regex"
test_401gd() {
# Verify find_param only reads matching params (LU-20038)
local strace_log=$TMP/$tfile.strace
# find_param for 'memused' should not open brw_stats
strace -e trace=openat -o $strace_log \
$LCTL find_param memused > /dev/null
# Should have found at least one match
local count=$($LCTL find_param -N memused | wc -l)
(( count >= 1 )) ||
error "(0) find_param memused found no results"
# brw_stats should NOT have been opened for reading
grep -q "brw_stats.*O_RDONLY" $strace_log &&
error "(1) brw_stats was opened, but should not be"
# job_stats should NOT have been opened for reading
grep -q "job_stats.*O_RDONLY" $strace_log &&
error "(2) job_stats was opened, but should not be"
# memused SHOULD have been opened
grep -q "memused.*O_RDONLY" $strace_log ||
error "(3) memused was not opened, but should be"
rm -f $strace_log
}
run_test 401gd "check 'lctl find_param' does not read non-matching params"
test_402() {
[[ $MDS1_VERSION -ge $(version_code 2.7.66) ]] ||
[[ $MDS1_VERSION -ge $(version_code 2.7.18.4) &&
$MDS1_VERSION -lt $(version_code 2.7.50) ]] ||
[[ $MDS1_VERSION -ge $(version_code 2.7.2) &&
$MDS1_VERSION -lt $(version_code 2.7.11) ]] ||
skip "Need MDS version 2.7.2+ or 2.7.18.4+ or 2.7.66+"
remote_mds_nodsh && skip "remote MDS with nodsh"
$LFS setdirstripe -i 0 $DIR/$tdir || error "setdirstripe -i 0 failed"
#define OBD_FAIL_MDS_FLD_LOOKUP 0x15c
do_facet mds1 "lctl set_param fail_loc=0x8000015c"
touch $DIR/$tdir/$tfile && error "touch should fail with ENOENT" ||
echo "Touch failed - OK"
}
run_test 402 "Return ENOENT to lod_generate_and_set_lovea"
test_403() {
local file1=$DIR/$tfile.1
local file2=$DIR/$tfile.2
local tfile=$TMP/$tfile
rm -f $file1 $file2 $tfile
touch $file1
ln $file1 $file2
# 30 sec OBD_TIMEOUT in ll_getattr()
# right before populating st_nlink
$LCTL set_param fail_loc=0x80001409
stat -c %h $file1 > $tfile &
# create an alias, drop all locks and reclaim the dentry
< $file2
cancel_lru_locks mdc
cancel_lru_locks osc
sysctl -w vm.drop_caches=2
wait
[ $(cat $tfile) -gt 0 ] || error "wrong nlink count: $(cat $tfile)"
rm -f $tfile $file1 $file2
}
run_test 403 "i_nlink should not drop to zero due to aliasing"
test_404() { # LU-6601
[[ $MDS1_VERSION -ge $(version_code 2.8.53) ]] ||
skip "Need server version newer than 2.8.52"
remote_mds_nodsh && skip "remote MDS with nodsh"
local mosps=$(do_facet $SINGLEMDS $LCTL dl |
awk '/osp .*-osc-MDT/ { print $4}')
local osp
for osp in $mosps; do
echo "Deactivate: " $osp
do_facet $SINGLEMDS $LCTL --device %$osp deactivate
wait_update_facet $SINGLEMDS \
"$LCTL dl | awk -vp=$osp '\\\$4 == p { print \\\$2 }'" \
"IN" 10 || error "deactivate error"
echo "Activate: " $osp
do_facet $SINGLEMDS $LCTL --device %$osp activate
wait_update_facet $SINGLEMDS \
"$LCTL dl | awk -vp=$osp '\\\$4 == p { print \\\$2 }'" \
"UP" 10 || error "activate error"
done
}
run_test 404 "validate manual {de}activated works properly for OSPs"
test_405() {
[ -n "$FILESET" ] && skip "Not functional for FILESET set"
[ $MDS1_VERSION -lt $(version_code 2.6.92) ] ||
[ $CLIENT_VERSION -lt $(version_code 2.6.99) ] &&
skip "Layout swap lock is not supported"
check_swap_layouts_support
check_swap_layout_no_dom $DIR
test_mkdir $DIR/$tdir
swap_lock_test -d $DIR/$tdir ||
error "One layout swap locked test failed"
}
run_test 405 "Various layout swap lock tests"
test_406() {
[ $MDSCOUNT -lt 2 ] && skip_env "needs >= 2 MDTs"
[ $OSTCOUNT -lt 2 ] && skip_env "needs >= 2 OSTs"
[ -n "$FILESET" ] && skip "SKIP due to FILESET set"
[ $PARALLEL == "yes" ] && skip "skip parallel run"
[ $MDS1_VERSION -lt $(version_code 2.8.50) ] &&
skip "Need MDS version at least 2.8.50"
local def_stripe_size=$($LFS getstripe -S $MOUNT)
local test_pool=$TESTNAME
pool_add $test_pool || error "pool_add failed"
pool_add_targets $test_pool 0 $(($OSTCOUNT - 1)) 1 ||
error "pool_add_targets failed"
save_layout_restore_at_exit $MOUNT
# parent set default stripe count only, child will stripe from both
# parent and fs default
$LFS setstripe -c 1 -i 1 -S $((def_stripe_size * 2)) -p $test_pool $MOUNT ||
error "setstripe $MOUNT failed"
$LFS mkdir -c $MDSCOUNT $DIR/$tdir || error "mkdir $tdir failed"
$LFS setstripe -c $OSTCOUNT $DIR/$tdir || error "setstripe $tdir failed"
for i in $(seq 10); do
local f=$DIR/$tdir/$tfile.$i
touch $f || error "touch failed"
local count=$($LFS getstripe -c $f)
[ $count -eq $OSTCOUNT ] ||
error "$f stripe count $count != $OSTCOUNT"
local offset=$($LFS getstripe -i $f)
[ $offset -eq 1 ] || error "$f stripe offset $offset != 1"
local size=$($LFS getstripe -S $f)
[ $size -eq $((def_stripe_size * 2)) ] ||
error "$f stripe size $size != $((def_stripe_size * 2))"
local pool=$($LFS getstripe -p $f)
[ $pool == $test_pool ] || error "$f pool $pool != $test_pool"
done
# change fs default striping, delete parent default striping, now child
# will stripe from new fs default striping only
$LFS setstripe -c 1 -S $def_stripe_size -i 0 $MOUNT ||
error "change $MOUNT default stripe failed"
$LFS setstripe -c 0 $DIR/$tdir ||
error "delete $tdir default stripe failed"
for i in $(seq 11 20); do
local f=$DIR/$tdir/$tfile.$i
touch $f || error "touch $f failed"
local count=$($LFS getstripe -c $f)
[ $count -eq 1 ] || error "$f stripe count $count != 1"
local offset=$($LFS getstripe -i $f)
[ $offset -eq 0 ] || error "$f stripe offset $offset != 0"
local size=$($LFS getstripe -S $f)
[ $size -eq $def_stripe_size ] ||
error "$f stripe size $size != $def_stripe_size"
local pool=$($LFS getstripe -p $f)
[ $pool == $test_pool ] || error "$f pool $pool isn't set"
done
unlinkmany $DIR/$tdir/$tfile. 1 20
local f=$DIR/$tdir/$tfile
pool_remove_all_targets $test_pool $f
pool_remove $test_pool $f
}
run_test 406 "DNE support fs default striping"
test_407() {
[ $MDSCOUNT -lt 2 ] && skip_env "needs >= 2 MDTs"
[[ $MDS1_VERSION -lt $(version_code 2.8.55) ]] &&
skip "Need MDS version at least 2.8.55"
remote_mds_nodsh && skip "remote MDS with nodsh"
$LFS mkdir -i 0 -c 1 $DIR/$tdir.0 ||
error "$LFS mkdir -i 0 -c 1 $tdir.0 failed"
$LFS mkdir -i 1 -c 1 $DIR/$tdir.1 ||
error "$LFS mkdir -i 1 -c 1 $tdir.1 failed"
touch $DIR/$tdir.0/$tfile.0 || error "touch $tdir.0/$tfile.0 failed"
#define OBD_FAIL_DT_TXN_STOP 0x2019
for idx in $(seq $MDSCOUNT); do
do_facet mds$idx "lctl set_param fail_loc=0x2019"
done
$LFS mkdir -c 2 $DIR/$tdir && error "$LFS mkdir -c 2 $tdir should fail"
mv $DIR/$tdir.0/$tfile.0 $DIR/$tdir.1/$tfile.1 &&
error "mv $tdir.0/$tfile.0 $tdir.1/$tfile.1 should fail"
true
}
run_test 407 "transaction fail should cause operation fail"
test_408() {
dd if=/dev/zero of=$DIR/$tfile bs=$PAGE_SIZE count=1 oflag=direct
#define OBD_FAIL_OSC_BRW_PREP_REQ2 0x40a
lctl set_param fail_loc=0x8000040a
# let ll_prepare_partial_page() fail
dd if=/dev/zero of=$DIR/$tfile bs=2048 count=1 conv=notrunc || true
rm -f $DIR/$tfile
# create at least 100 unused inodes so that
# shrink_icache_memory(0) should not return 0
touch $DIR/$tfile-{0..100}
rm -f $DIR/$tfile-{0..100}
sync
echo 2 > /proc/sys/vm/drop_caches
}
run_test 408 "drop_caches should not hang due to page leaks"
test_409()
{
[ $MDSCOUNT -lt 2 ] && skip "needs >= 2 MDTs"
mkdir -p $DIR/$tdir || error "(0) Fail to mkdir"
$LFS mkdir -i 1 -c 2 $DIR/$tdir/foo || error "(1) Fail to mkdir"
touch $DIR/$tdir/guard || error "(2) Fail to create"
local PREFIX=$(str_repeat 'A' 128)
echo "Create 1K hard links start at $(date)"
createmany -l $DIR/$tdir/guard $DIR/$tdir/foo/${PREFIX}_ 1000 ||
error "(3) Fail to hard link"
echo "Links count should be right although linkEA overflow"
stat $DIR/$tdir/guard || error "(4) Fail to stat"
local linkcount=$(stat --format=%h $DIR/$tdir/guard)
[ $linkcount -eq 1001 ] ||
error "(5) Unexpected hard links count: $linkcount"
echo "List all links start at $(date)"
ls -l $DIR/$tdir/foo > /dev/null ||
error "(6) Fail to list $DIR/$tdir/foo"
echo "Unlink hard links start at $(date)"
unlinkmany $DIR/$tdir/foo/${PREFIX}_ 1000 ||
error "(7) Fail to unlink"
echo "Unlink hard links finished at $(date)"
}
run_test 409 "Large amount of cross-MDTs hard links on the same file"
test_410()
{
[[ $CLIENT_VERSION -lt $(version_code 2.9.59) ]] &&
skip "Need client version at least 2.9.59"
# Create a file, and stat it from the kernel
local testfile=$DIR/$tfile
touch $testfile
local run_id=$RANDOM
local my_ino=$(stat --format "%i" $testfile)
# Try to insert the module.
load_module kunit/kinode run_id=$run_id fname=$testfile ||
error "load_module failed"
# Anything but success is a test failure
dmesg | grep -q \
"lustre_kinode_$run_id: inode numbers are identical: $my_ino" ||
error "no inode match"
# Remove the test module
rmmod -v kinode ||
error "rmmod failed (may trigger a failure in a later test)"
}
run_test 410 "Test inode number returned from kernel thread"
cleanup_test411_cgroup() {
trap 0
cat $1/memory.stat
rmdir "$1"
}
test_411a() {
local cg_basedir=/sys/fs/cgroup/memory
# LU-9966
test -f "$cg_basedir/memory.kmem.limit_in_bytes" ||
skip "no setup for cgroup"
dd if=/dev/zero of=$DIR/$tfile bs=1M count=100 conv=fsync ||
error "test file creation failed"
cancel_lru_locks osc
# Create a very small memory cgroup to force a slab allocation error
local cgdir=$cg_basedir/osc_slab_alloc
mkdir $cgdir || error "cgroup mkdir '$cgdir' failed"
trap "cleanup_test411_cgroup $cgdir" EXIT
echo 2M > $cgdir/memory.kmem.limit_in_bytes
echo 1M > $cgdir/memory.limit_in_bytes
# Should not LBUG, just be killed by oom-killer
# dd will return 0 even allocation failure in some environment.
# So don't check return value
bash -c "echo \$$ > $cgdir/tasks && dd if=$DIR/$tfile of=/dev/null"
cleanup_test411_cgroup $cgdir
return 0
}
run_test 411a "Slab allocation error with cgroup does not LBUG"
test_411b() {
local cg_basedir=/sys/fs/cgroup/memory
# LU-9966
[ -e "$cg_basedir/memory.kmem.limit_in_bytes" ] ||
skip "no setup for cgroup"
local count=2
(( OSTCOUNT < 2 )) && count=$OSTCOUNT
$LFS setstripe -c $count $DIR/$tfile || error "unable to setstripe"
# To avoid testing failures on x86 and arm, we set the memcg limit
# with 1024M on both systems.
local memlimit_mb=1024
local min_size_ost=$($LFS df | awk "/$FSNAME-OST/ { print \$4 }" |
sort -un | head -1)
(( memlimit_mb * 4 * 1024 < min_size_ost * count - 100 )) ||
skip "OST space are too small: ${min_size_ost}K"
echo "count=$count/$OSTCOUNT min_size_ost=${min_size_ost}K"
$LFS df
# Create a cgroup and set memory limit
# (tfile is used as an easy way to get a recognizable cgroup name)
local cgdir=$cg_basedir/$tfile
mkdir $cgdir || error "cgroup mkdir '$cgdir' failed"
stack_trap "cleanup_test411_cgroup $cgdir" EXIT
echo $((memlimit_mb * 1024 * 1024)) > $cgdir/memory.limit_in_bytes
echo "writing first file"
# Write a file 4x the memory limit in size
bash -c "echo \$$ > $cgdir/tasks && dd if=/dev/zero of=$DIR/$tfile bs=1M count=$((memlimit_mb * 4))" ||
error "(1) failed to write successfully"
sync
cancel_lru_locks osc
rm -f $DIR/$tfile
$LFS setstripe -c $count $DIR/$tfile || error "unable to setstripe"
# Try writing at a larger block size
# NB: if block size is >= 1/2 cgroup size, we sometimes get OOM killed
# so test with 1/4 cgroup size (this seems reasonable to me - we do
# need *some* memory to do IO in)
echo "writing at larger block size"
bash -c "echo \$$ > $cgdir/tasks && dd if=/dev/zero of=$DIR/$tfile bs=64M count=$((memlimit_mb * 4 / 128))" ||
error "(3) failed to write successfully"
sync
cancel_lru_locks osc
rm -f $DIR/$tfile
$LFS setstripe -c $count $DIR/$tfile.{1..4} || error "unable to setstripe"
# Try writing multiple files at once
(( memlimit_mb * 4 * 1024 * 2 < min_size_ost * count - 100 )) ||
skip "OST space are too small: ${min_size_ost}K"
echo "writing multiple files"
bash -c "echo \$$ > $cgdir/tasks && dd if=/dev/zero of=$DIR/$tfile.1 bs=32M count=$((memlimit_mb * 4 / 64))" &
local pid1=$!
bash -c "echo \$$ > $cgdir/tasks && dd if=/dev/zero of=$DIR/$tfile.2 bs=32M count=$((memlimit_mb * 4 / 64))" &
local pid2=$!
bash -c "echo \$$ > $cgdir/tasks && dd if=/dev/zero of=$DIR/$tfile.3 bs=32M count=$((memlimit_mb * 4 / 64))" &
local pid3=$!
bash -c "echo \$$ > $cgdir/tasks && dd if=/dev/zero of=$DIR/$tfile.4 bs=32M count=$((memlimit_mb * 4 / 64))" &
local pid4=$!
wait $pid1
local rc1=$?
wait $pid2
local rc2=$?
wait $pid3
local rc3=$?
wait $pid4
local rc4=$?
if (( rc1 != 0)); then
error "error $rc1 writing to file from $pid1"
fi
if (( rc2 != 0)); then
error "error $rc2 writing to file from $pid2"
fi
if (( rc3 != 0)); then
error "error $rc3 writing to file from $pid3"
fi
if (( rc4 != 0)); then
error "error $rc4 writing to file from $pid4"
fi
sync
cancel_lru_locks osc
# These files can be large-ish (~1 GiB total), so delete them rather
# than leave for later cleanup
rm -f $DIR/$tfile.*
return 0
}
run_test 411b "confirm Lustre can avoid OOM with reasonable cgroups limits"
test_412() {
(( $MDSCOUNT > 1 )) || skip_env "needs >= 2 MDTs"
(( $MDS1_VERSION >= $(version_code 2.10.55) )) ||
skip "Need server version at least 2.10.55"
$LFS mkdir -i $((MDSCOUNT - 1)),$((MDSCOUNT - 2)) $DIR/$tdir ||
error "mkdir failed"
$LFS getdirstripe $DIR/$tdir
local stripe_index=$($LFS getdirstripe -i $DIR/$tdir)
[ $stripe_index -eq $((MDSCOUNT - 1)) ] ||
error "expect $((MDSCOUT - 1)) get $stripe_index"
local stripe_count=$($LFS getdirstripe -T $DIR/$tdir)
[ $stripe_count -eq 2 ] ||
error "expect 2 get $stripe_count"
(( $MDS1_VERSION >= $(version_code 2.14.55) )) || return 0
local index
local index2
# subdirs should be on the same MDT as parent
for i in $(seq 0 $((MDSCOUNT - 1))); do
$LFS mkdir -i $i $DIR/$tdir/mdt$i || error "mkdir mdt$i failed"
mkdir $DIR/$tdir/mdt$i/sub || error "mkdir sub failed"
index=$($LFS getstripe -m $DIR/$tdir/mdt$i/sub)
(( index == i )) || error "mdt$i/sub on MDT$index"
done
# stripe offset -1, ditto
for i in {1..10}; do
$LFS mkdir -i -1 $DIR/$tdir/qos$i || error "mkdir qos$i failed"
index=$($LFS getstripe -m $DIR/$tdir/qos$i)
mkdir $DIR/$tdir/qos$i/sub || error "mkdir sub failed"
index2=$($LFS getstripe -m $DIR/$tdir/qos$i/sub)
(( index == index2 )) ||
error "qos$i on MDT$index, sub on MDT$index2"
done
local testdir=$DIR/$tdir/inherit
$LFS mkdir -i 1 --max-inherit=3 $testdir || error "mkdir inherit failed"
# inherit 2 levels
for i in 1 2; do
testdir=$testdir/s$i
mkdir $testdir || error "mkdir $testdir failed"
index=$($LFS getstripe -m $testdir)
(( index == 1 )) ||
error "$testdir on MDT$index"
done
# not inherit any more
testdir=$testdir/s3
mkdir $testdir || error "mkdir $testdir failed"
getfattr -d -m dmv $testdir | grep dmv &&
error "default LMV set on $testdir" || true
}
run_test 412 "mkdir on specific MDTs"
TEST413_COUNT=${TEST413_COUNT:-200}
#
# set_maxage() is used by test_413 only.
# This is a helper function to set maxage. Does not return any value.
# Input: maxage to set
#
set_maxage() {
local lmv_qos_maxage
local lod_qos_maxage
local new_maxage=$1
local mdts=$(mdts_nodes)
lmv_qos_maxage=$($LCTL get_param -n lmv.*.qos_maxage)
$LCTL set_param lmv.*.qos_maxage=$new_maxage
stack_trap "$LCTL set_param \
lmv.*.qos_maxage=$lmv_qos_maxage > /dev/null"
lod_qos_maxage=$(do_facet mds1 $LCTL get_param -n \
lod.$FSNAME-MDT0000-mdtlov.qos_maxage | awk '{ print $1 }')
do_nodes $mdts $LCTL set_param lod.*.mdt_qos_maxage=$new_maxage
stack_trap "do_nodes $mdts $LCTL set_param lod.*.mdt_qos_maxage=$lod_qos_maxage > /dev/null"
}
generate_uneven_mdts() {
local threshold=$1
local ffree
local bavail
local max
local min
local max_index
local min_index
local tmp
local i
echo
echo "Check for uneven MDTs: "
ffree=($(lctl get_param -n mdc.*[mM][dD][cC]-[^M]*.filesfree))
bavail=($(lctl get_param -n mdc.*[mM][dD][cC]-[^M]*.kbytesavail))
bsize=$(lctl get_param -n mdc.*MDT0000*.blocksize)
max=$(((${ffree[0]} >> 8) * (${bavail[0]} * bsize >> 16)))
min=$(((${ffree[0]} >> 8) * (${bavail[0]} * bsize >> 16)))
max_index=0
min_index=0
for ((i = 1; i < ${#ffree[@]}; i++)); do
tmp=$(((${ffree[i]} >> 8) * (${bavail[i]} * bsize >> 16)))
if [ $tmp -gt $max ]; then
max=$tmp
max_index=$i
fi
if [ $tmp -lt $min ]; then
min=$tmp
min_index=$i
fi
done
(( min > 0 )) || skip "low space on MDT$min_index"
(( ${ffree[min_index]} > 0 )) ||
skip "no free files on MDT$min_index"
(( ${ffree[min_index]} < 10000000 )) ||
skip "too many free files on MDT$min_index"
# Check if we need to generate uneven MDTs
local diff=$(((max - min) * 100 / min))
local testdirp=$DIR/$tdir-fillmdt # parent fill folder
local testdir # individual folder within $testdirp
local start
local cmd
echo "using cmd $cmd"
for (( i = 0; diff < threshold; i++ )); do
testdir=${testdirp}/$i
[ -d $testdir ] && continue
(( i % 10 > 0 )) || { $LFS df; $LFS df -i; }
mkdir -p $testdirp
# generate uneven MDTs, create till $threshold% diff
echo -n "weight diff=$diff% must be > $threshold% ..."
echo "Fill MDT$min_index with $TEST413_COUNT files: loop $i"
$LFS mkdir -i $min_index $testdir ||
error "mkdir $testdir failed"
$LFS setstripe -E 1M -L mdt $testdir ||
error "setstripe $testdir failed"
start=$SECONDS
fast_file_write $testdir/f 128K $TEST413_COUNT ||
error "fast write failed"
sync; sleep 1; sync
# wait for QOS to update
(( SECONDS < start + 2 )) && sleep $((start + 2 - SECONDS))
ffree=($(lctl get_param -n mdc.*[mM][dD][cC]-*.filesfree))
bavail=($(lctl get_param -n mdc.*[mM][dD][cC]-*.kbytesavail))
max=$(((${ffree[max_index]} >> 8) *
(${bavail[max_index]} * bsize >> 16)))
min=$(((${ffree[min_index]} >> 8) *
(${bavail[min_index]} * bsize >> 16)))
(( min > 0 )) || skip "low space on MDT$min_index"
diff=$(((max - min) * 100 / min))
done
echo "MDT filesfree available: ${ffree[*]}"
echo "MDT blocks available: ${bavail[*]}"
echo "weight diff=$diff%"
}
test_qos_mkdir() {
local mkdir_cmd=$1
local stripe_count=$2
local mdts=$(mdts_nodes)
local testdir
local lmv_qos_prio_free
local lmv_qos_threshold_rr
local lod_qos_prio_free
local lod_qos_threshold_rr
local total
local count
local i
# @total is total directories created if it's testing plain
# directories, otherwise it's total stripe object count for
# striped directories test.
# remote/striped directory unlinking is slow on zfs and may
# timeout, test with fewer directories
[ "$mds1_FSTYPE" = "zfs" ] && total=120 || total=240
lmv_qos_prio_free=$($LCTL get_param -n lmv.*.qos_prio_free | head -n1)
lmv_qos_prio_free=${lmv_qos_prio_free%%%}
lmv_qos_threshold_rr=$($LCTL get_param -n lmv.*.qos_threshold_rr |
head -n1)
lmv_qos_threshold_rr=${lmv_qos_threshold_rr%%%}
stack_trap "$LCTL set_param \
lmv.*.qos_prio_free=$lmv_qos_prio_free > /dev/null"
stack_trap "$LCTL set_param \
lmv.*.qos_threshold_rr=$lmv_qos_threshold_rr > /dev/null"
lod_qos_prio_free=$(do_facet mds1 $LCTL get_param -n \
lod.$FSNAME-MDT0000-mdtlov.mdt_qos_prio_free | head -n1)
lod_qos_prio_free=${lod_qos_prio_free%%%}
lod_qos_threshold_rr=$(do_facet mds1 $LCTL get_param -n \
lod.$FSNAME-MDT0000-mdtlov.mdt_qos_threshold_rr | head -n1)
lod_qos_threshold_rr=${lod_qos_threshold_rr%%%}
stack_trap "do_nodes $mdts $LCTL set_param \
lod.*.mdt_qos_prio_free=$lod_qos_prio_free > /dev/null"
stack_trap "do_nodes $mdts $LCTL set_param \
lod.*.mdt_qos_threshold_rr=$lod_qos_threshold_rr > /dev/null"
# decrease statfs age, so that it can be updated in time
$LCTL set_param lmv.*.qos_maxage=1 > /dev/null
do_nodes $mdts $LCTL set_param lod.*.mdt_qos_maxage=1 > /dev/null
$LCTL set_param lmv.*.qos_threshold_rr=100 > /dev/null
do_nodes $mdts $LCTL set_param lod.*.mdt_qos_threshold_rr=100 > /dev/null
testdir=$DIR/$tdir-s$stripe_count/rr
local stripe_index=$($LFS getstripe -m $testdir)
local test_mkdir_rr=true
getfattr -d -m dmv -e hex $testdir | grep dmv
if (( $? == 0 && $MDS1_VERSION >= $(version_code 2.14.51) )); then
echo "defstripe: '$($LFS getdirstripe -D $testdir)'"
(( $($LFS getdirstripe -D --max-inherit-rr $testdir) == 0 )) &&
test_mkdir_rr=false
fi
echo
$test_mkdir_rr &&
echo "Mkdir (stripe_count $stripe_count) roundrobin:" ||
echo "Mkdir (stripe_count $stripe_count) on stripe $stripe_index"
stack_trap "unlinkmany -d $testdir/subdir $((total / stripe_count))"
for (( i = 0; i < total / stripe_count; i++ )); do
eval $mkdir_cmd $testdir/subdir$i ||
error "$mkdir_cmd subdir$i failed"
done
for (( i = 0; i < $MDSCOUNT; i++ )); do
count=$($LFS getdirstripe -i $testdir/* | grep -c "^$i$")
echo "$count directories created on MDT$i"
if $test_mkdir_rr; then
(( count == total / stripe_count / MDSCOUNT )) ||
error "subdirs are not evenly distributed"
elif (( i == stripe_index )); then
(( count == total / stripe_count )) ||
error "$count subdirs created on MDT$i"
else
(( count == 0 )) ||
error "$count subdirs created on MDT$i"
fi
if $test_mkdir_rr && [ $stripe_count -gt 1 ]; then
count=$($LFS getdirstripe $testdir/* |
grep -c -P "^\s+$i\t")
echo "$count stripes created on MDT$i"
# deviation should < 5% of average
delta=$((count - total / MDSCOUNT))
(( ${delta#-} <= total / MDSCOUNT / 20 )) ||
error "stripes are not evenly distributed"
fi
done
echo
echo "Check for uneven MDTs: "
local ffree
local bavail
local max
local min
local max_index
local min_index
local tmp
ffree=($(lctl get_param -n mdc.*[mM][dD][cC]-[^M]*.filesfree))
bavail=($(lctl get_param -n mdc.*[mM][dD][cC]-[^M]*.kbytesavail))
bsize=$(lctl get_param -n mdc.*MDT0000*.blocksize)
max=$(((${ffree[0]} >> 8) * (${bavail[0]} * bsize >> 16)))
min=$(((${ffree[0]} >> 8) * (${bavail[0]} * bsize >> 16)))
max_index=0
min_index=0
for ((i = 1; i < ${#ffree[@]}; i++)); do
tmp=$(((${ffree[i]} >> 8) * (${bavail[i]} * bsize >> 16)))
if [ $tmp -gt $max ]; then
max=$tmp
max_index=$i
fi
if [ $tmp -lt $min ]; then
min=$tmp
min_index=$i
fi
done
echo "stripe_count=$stripe_count min_idx=$min_index max_idx=$max_index"
(( min > 0 )) || skip "low space on MDT$min_index"
(( ${ffree[min_index]} < 10000000 )) ||
skip "too many free files on MDT$min_index"
generate_uneven_mdts 120
echo "MDT filesfree available: ${ffree[*]}"
echo "MDT blocks available: ${bavail[*]}"
echo "weight diff=$(((max - min) * 100 / min))%"
echo
echo "Mkdir (stripe_count $stripe_count) with balanced space usage:"
$LCTL set_param lmv.*.qos_threshold_rr=0 > /dev/null
$LCTL set_param lmv.*.qos_prio_free=100 > /dev/null
do_nodes $mdts $LCTL set_param lod.*.mdt_qos_threshold_rr=0 > /dev/null
do_nodes $mdts $LCTL set_param lod.*.mdt_qos_prio_free=100 > /dev/null
sleep 1
testdir=$DIR/$tdir-s$stripe_count/qos
stack_trap "unlinkmany -d $testdir/subdir $((total / stripe_count))"
for (( i = 0; i < total / stripe_count; i++ )); do
eval $mkdir_cmd $testdir/subdir$i ||
error "$mkdir_cmd subdir$i failed"
done
max=0
for (( i = 0; i < $MDSCOUNT; i++ )); do
count=$($LFS getdirstripe -i $testdir/* | grep -c "^$i$")
(( count > max )) && max=$count
echo "$count directories created on MDT$i : curmax=$max"
done
min=$($LFS getdirstripe -i $testdir/* | grep -c "^$min_index$")
# D-value should > 10% of average
(( max - min > total / stripe_count / MDSCOUNT / 10 )) ||
error "subdirs shouldn't be evenly distributed: $max - $min <= $((total / stripe_count / MDSCOUNT / 10))"
# ditto for stripes
if (( stripe_count > 1 )); then
max=0
for (( i = 0; i < $MDSCOUNT; i++ )); do
count=$($LFS getdirstripe $testdir/* |
grep -c -P "^\s+$i\t")
(( count > max )) && max=$count
echo "$count stripes created on MDT$i"
done
min=$($LFS getdirstripe $testdir/* |
grep -c -P "^\s+$min_index\t")
(( max - min > total / MDSCOUNT / 10 )) ||
error "stripes shouldn't be evenly distributed: $max - $min <= $((total / MDSCOUNT / 10))"
fi
}
most_full_mdt() {
local ffree
local bavail
local bsize
local min
local min_index
local tmp
ffree=($(lctl get_param -n mdc.*[mM][dD][cC]-[^M]*.filesfree))
bavail=($(lctl get_param -n mdc.*[mM][dD][cC]-[^M]*.kbytesavail))
bsize=$(lctl get_param -n mdc.*MDT0000*.blocksize)
min=$(((${ffree[0]} >> 8) * (${bavail[0]} * bsize >> 16)))
min_index=0
for ((i = 1; i < ${#ffree[@]}; i++)); do
tmp=$(((${ffree[i]} >> 8) * (${bavail[i]} * bsize >> 16)))
(( tmp < min )) && min=$tmp && min_index=$i
done
echo -n $min_index
}
test_413A() {
local index=0
echo "before"
do_nodes $CLIENTS $LCTL get_param lmv.*.qos_rr_index
for client in ${CLIENTS//,/ }; do
local cli_ver=$(lustre_build_version_node $client)
(( $(version_code $cli_ver) >= $(version_code 2.16.59.8) )) ||
skip "need $client $cli_ver > 2.16.59.8 for qos_rr_index"
do_node $client $LCTL set_param -n lmv.*.qos_rr_index=$index
((index++))
done
echo "after"
do_nodes $CLIENTS $LCTL get_param lmv.*.qos_rr_index
index=0
for client in ${CLIENTS//,/ }; do
got=$(do_node $client $LCTL get_param -n lmv.*.qos_rr_index)
((got == index)) ||
error "lmv.*.qos_rr_index was $got != $index"
((index++))
done
echo "all OSCs set correctly"
}
run_test 413A "get and set qos_rr_index on all clients"
test_413a() {
[ $MDSCOUNT -lt 2 ] &&
skip "We need at least 2 MDTs for this test"
[ $MDS1_VERSION -lt $(version_code 2.12.52) ] &&
skip "Need server version at least 2.12.52"
local stripe_max=$((MDSCOUNT - 1))
local stripe_count
# let caller set maxage for latest result
set_maxage 1
# fill MDT unevenly
generate_uneven_mdts 120
# test 4-stripe directory at most, otherwise it's too slow
# We are being very defensive. Although Autotest uses 4 MDTs.
# We make sure stripe_max does not go over 4.
(( stripe_max > 4 )) && stripe_max=4
# unlinking striped directory is slow on zfs, and may timeout, only test
# plain directory
[ "$mds1_FSTYPE" == "zfs" ] && stripe_max=1
for stripe_count in $(seq 1 $stripe_max); do
mkdir $DIR/$tdir-s$stripe_count || error "mkdir failed"
mkdir $DIR/$tdir-s$stripe_count/rr || error "mkdir failed"
$LFS mkdir -i $(most_full_mdt) $DIR/$tdir-s$stripe_count/qos ||
error "mkdir failed"
test_qos_mkdir "$LFS mkdir -i -1 -c $stripe_count" $stripe_count
done
}
run_test 413a "QoS mkdir with 'lfs mkdir -i -1'"
test_413b() {
[ $MDSCOUNT -lt 2 ] &&
skip "We need at least 2 MDTs for this test"
[ $MDS1_VERSION -lt $(version_code 2.12.52) ] &&
skip "Need server version at least 2.12.52"
local stripe_max=$((MDSCOUNT - 1))
local testdir
local stripe_count
# let caller set maxage for latest result
set_maxage 1
# fill MDT unevenly
generate_uneven_mdts 120
# test 4-stripe directory at most, otherwise it's too slow
# We are being very defensive. Although Autotest uses 4 MDTs.
# We make sure stripe_max does not go over 4.
(( stripe_max > 4 )) && stripe_max=4
[ "$mds1_FSTYPE" == "zfs" ] && stripe_max=1
for stripe_count in $(seq 1 $stripe_max); do
testdir=$DIR/$tdir-s$stripe_count
mkdir $testdir || error "mkdir $testdir failed"
mkdir $testdir/rr || error "mkdir rr failed"
$LFS mkdir -i $(most_full_mdt) $testdir/qos ||
error "mkdir qos failed"
$LFS setdirstripe -D -c $stripe_count --max-inherit-rr 2 \
$testdir/rr || error "setdirstripe rr failed"
$LFS setdirstripe -D -c $stripe_count $testdir/qos ||
error "setdirstripe failed"
test_qos_mkdir "mkdir" $stripe_count
done
}
run_test 413b "QoS mkdir under dir whose default LMV starting MDT offset is -1"
test_413c() {
(( $MDSCOUNT >= 2 )) ||
skip "We need at least 2 MDTs for this test"
(( $MDS1_VERSION >= $(version_code 2.14.51) )) ||
skip "Need server version at least 2.14.51"
local testdir
local inherit
local inherit_rr
local lmv_qos_maxage
local lod_qos_maxage
# let caller set maxage for latest result
lmv_qos_maxage=$($LCTL get_param -n lmv.*.qos_maxage)
$LCTL set_param lmv.*.qos_maxage=1
stack_trap "$LCTL set_param \
lmv.*.qos_maxage=$lmv_qos_maxage > /dev/null" RETURN
lod_qos_maxage=$(do_facet mds1 $LCTL get_param -n \
lod.$FSNAME-MDT0000-mdtlov.qos_maxage | awk '{ print $1 }')
do_nodes $(comma_list $(mdts_nodes)) $LCTL set_param \
lod.*.mdt_qos_maxage=1
stack_trap "do_nodes $(comma_list $(mdts_nodes)) $LCTL set_param \
lod.*.mdt_qos_maxage=$lod_qos_maxage > /dev/null" RETURN
# fill MDT unevenly
generate_uneven_mdts 120
testdir=$DIR/${tdir}-s1
mkdir $testdir || error "mkdir $testdir failed"
mkdir $testdir/rr || error "mkdir rr failed"
$LFS mkdir -i $(most_full_mdt) $testdir/qos || error "mkdir qos failed"
# default max_inherit is -1, default max_inherit_rr is 0
$LFS setdirstripe -D -c 1 $testdir/rr ||
error "setdirstripe rr failed"
$LFS setdirstripe -D -c 1 -i -1 -X 2 --max-inherit-rr 1 $testdir/qos ||
error "setdirstripe qos failed"
test_qos_mkdir "mkdir" 1
mkdir $testdir/rr/level1 || error "mkdir rr/level1 failed"
inherit=$($LFS getdirstripe -D -X $testdir/rr/level1)
(( $inherit == -1 )) || error "rr/level1 inherit $inherit != -1"
inherit_rr=$($LFS getdirstripe -D --max-inherit-rr $testdir/rr/level1)
(( $inherit_rr == 0 )) || error "rr/level1 inherit-rr $inherit_rr != 0"
mkdir $testdir/qos/level1 || error "mkdir qos/level1 failed"
inherit=$($LFS getdirstripe -D -X $testdir/qos/level1)
(( $inherit == 1 )) || error "qos/level1 inherit $inherit != 1"
inherit_rr=$($LFS getdirstripe -D --max-inherit-rr $testdir/qos/level1)
(( $inherit_rr == 0 )) || error "qos/level1 inherit-rr $inherit_rr != 0"
mkdir $testdir/qos/level1/level2 || error "mkdir level2 failed"
getfattr -d -m dmv -e hex $testdir/qos/level1/level2 | grep dmv &&
error "level2 shouldn't have default LMV" || true
}
run_test 413c "mkdir with default LMV max inherit rr"
test_413d() {
(( MDSCOUNT >= 2 )) ||
skip "We need at least 2 MDTs for this test"
(( MDS1_VERSION >= $(version_code 2.14.51) )) ||
skip "Need server version at least 2.14.51"
[ -n "$FILESET" ] && skip "Not functional for FILESET set"
local lmv_qos_threshold_rr
lmv_qos_threshold_rr=$($LCTL get_param -n lmv.*.qos_threshold_rr |
head -n1)
stack_trap "$LCTL set_param \
lmv.*.qos_threshold_rr=$lmv_qos_threshold_rr > /dev/null" EXIT
$LCTL set_param lmv.*.qos_threshold_rr=100 > /dev/null
mkdir -p $DIR/$tdir || error "mkdir $tdir failed"
getfattr -d -m dmv -e hex $DIR/$tdir | grep dmv &&
error "$tdir shouldn't have default LMV"
createmany -d $DIR/$tdir/sub $((100 * MDSCOUNT)) ||
error "mkdir sub failed"
local count=$($LFS getstripe -m $DIR/$tdir/* | grep -c ^0)
(( count == 100 )) || error "$count subdirs on MDT0"
}
run_test 413d "inherit ROOT default LMV"
test_413e() {
(( MDSCOUNT >= 2 )) ||
skip "We need at least 2 MDTs for this test"
(( MDS1_VERSION >= $(version_code 2.14.55) )) ||
skip "Need server version at least 2.14.55"
local testdir=$DIR/$tdir
local tmpfile=$TMP/temp.setdirstripe.stderr.$$
local max_inherit
local sub_max_inherit
mkdir -p $testdir || error "failed to create $testdir"
# set default max-inherit to -1 if stripe count is 0 or 1
$LFS setdirstripe -D -c 1 $testdir ||
error "failed to set default LMV"
max_inherit=$($LFS getdirstripe -D --max-inherit $testdir)
(( max_inherit == -1 )) ||
error "wrong max_inherit value $max_inherit"
# set default max_inherit to a fixed value if stripe count is not 0 or 1
$LFS setdirstripe -D -c -1 $testdir ||
error "failed to set default LMV"
max_inherit=$($LFS getdirstripe -D --max-inherit $testdir)
(( max_inherit > 0 )) ||
error "wrong max_inherit value $max_inherit"
# and the subdir will decrease the max_inherit by 1
mkdir -p $testdir/subdir-1 || error "failed to make subdir"
sub_max_inherit=$($LFS getdirstripe -D --max-inherit $testdir/subdir-1)
(( sub_max_inherit == max_inherit - 1)) ||
error "wrong max-inherit of subdir $sub_max_inherit"
# check specified --max-inherit and warning message
stack_trap "rm -f $tmpfile"
$LFS setdirstripe -D -c 2 --max-inherit=-1 $testdir 2> $tmpfile ||
error "failed to set default LMV"
max_inherit=$($LFS getdirstripe -D --max-inherit $testdir)
(( max_inherit == -1 )) ||
error "wrong max_inherit value $max_inherit"
# check the warning messages
if ! [[ $(cat $tmpfile) =~ "max-inherit=" ]]; then
error "failed to detect warning string"
fi
}
run_test 413e "check default max-inherit value"
test_fs_dmv_inherit()
{
local testdir=$DIR/$tdir
local count
local inherit
local inherit_rr
for i in 1 2; do
mkdir $testdir || error "mkdir $testdir failed"
count=$($LFS getdirstripe -D -c $testdir)
(( count == 1 )) ||
error "$testdir default LMV count mismatch $count != 1"
inherit=$($LFS getdirstripe -D -X $testdir)
(( inherit == 3 - i )) ||
error "$testdir default LMV max-inherit $inherit != $((3 - i))"
inherit_rr=$($LFS getdirstripe -D --max-inherit-rr $testdir)
(( inherit_rr == 3 - i )) ||
error "$testdir default LMV max-inherit-rr $inherit_rr != $((3 - i))"
testdir=$testdir/sub
done
mkdir $testdir || error "mkdir $testdir failed"
count=$($LFS getdirstripe -D -c $testdir)
(( count == 0 )) ||
error "$testdir default LMV count not zero: $count"
}
test_413f() {
(( MDSCOUNT >= 2 )) || skip "We need at least 2 MDTs for this test"
(( MDS1_VERSION >= $(version_code 2.14.55) )) ||
skip "Need server version at least 2.14.55"
getfattr -d -m trusted.dmv --absolute-names $DIR > $TMP/dmv.ea ||
error "dump $DIR default LMV failed"
stack_trap "setfattr --restore=$TMP/dmv.ea"
$LFS setdirstripe -D -i -1 -c 1 -X 3 --max-inherit-rr 3 $DIR ||
error "set $DIR default LMV failed"
test_fs_dmv_inherit
}
run_test 413f "lfs getdirstripe -D list ROOT default LMV if it's not set on dir"
test_413g() {
(( MDSCOUNT >= 2 )) || skip "We need at least 2 MDTs for this test"
local FILESET_orig=$FILESET
if [ -n "$FILESET" ];then
cleanup_mount $MOUNT
FILESET="" mount_client $MOUNT
fi
mkdir -p $DIR/$tdir/l2/l3/l4 || error "mkdir $tdir/l1/l2/l3 failed"
getfattr -d -m trusted.dmv --absolute-names $DIR > $TMP/dmv.ea ||
error "dump $DIR default LMV failed"
stack_trap "setfattr --restore=$TMP/dmv.ea"
$LFS setdirstripe -D -i -1 -c 1 -X 3 --max-inherit-rr 3 $DIR ||
error "set $DIR default LMV failed"
local saved_DIR=$DIR
export DIR=$MOUNT2
FILESET="/$tdir/l2/l3/l4" mount_client $MOUNT2 ||
error "mount $MOUNT2 failed"
if [ -n "$FILESET_orig" ]; then
stack_trap "umount_client $MOUNT2; export DIR=$saved_DIR;export FILESET=$FILESET_orig; mount_client $MOUNT"
else
stack_trap "umount_client $MOUNT2; export DIR=$saved_DIR"
fi
# first check filesystem-wide default LMV inheritance
test_fs_dmv_inherit || error "incorrect fs default LMV inheritance"
# then check subdirs are spread to all MDTs
createmany -d $DIR/s $((MDSCOUNT * 100)) || error "createmany failed"
local count=$($LFS getstripe -m $DIR/s* | sort -u | wc -l)
(( $count == $MDSCOUNT )) || error "dirs are spread to $count MDTs"
if [ -n "$FILESET_orig" ];then
cleanup_mount $MOUNT
FILESET=$FILESET_orig
mount_client $MOUNT
fi
}
run_test 413g "enforce ROOT default LMV on subdir mount"
test_413h() {
(( MDSCOUNT >= 2 )) ||
skip "We need at least 2 MDTs for this test"
(( MDS1_VERSION >= $(version_code 2.15.50.6) )) ||
skip "Need server version at least 2.15.50.6"
local lmv_qos_maxage=$($LCTL get_param -n lmv.*.qos_maxage)
stack_trap "$LCTL set_param \
lmv.*.qos_maxage=$lmv_qos_maxage > /dev/null"
$LCTL set_param lmv.*.qos_maxage=1
local depth=5
local rr_depth=4
local dir=$DIR/$tdir/l1/l2/l3/l4/l5
local count=$((MDSCOUNT * 20))
generate_uneven_mdts 50
mkdir -p $dir || error "mkdir $dir failed"
stack_trap "rm -rf $dir"
$LFS setdirstripe -D -c 1 -i -1 --max-inherit=$depth \
--max-inherit-rr=$rr_depth $dir
for ((d=0; d < depth + 2; d++)); do
log "dir=$dir:"
for ((sub=0; sub < count; sub++)); do
mkdir $dir/d$sub
done
$LFS getdirstripe -i $dir/d* | sort | uniq -c | sort -nr
local num=($($LFS getdirstripe -i $dir/d* | sort | uniq -c))
# subdirs within $rr_depth should be created round-robin
if (( d < rr_depth )); then
(( ${num[0]} != count )) ||
error "all objects created on MDT ${num[1]}"
fi
dir=$dir/d0
done
}
run_test 413h "don't stick to parent for round-robin dirs"
test_413i() {
[ $MDSCOUNT -lt 2 ] && skip_env "needs >= 2 MDTs"
(( MDS1_VERSION >= $(version_code 2.14.55) )) ||
skip "Need server version at least 2.14.55"
getfattr -d -m trusted.dmv --absolute-names $DIR > $TMP/dmv.ea ||
error "dump $DIR default LMV failed"
stack_trap "setfattr --restore=$TMP/dmv.ea"
local testdir=$DIR/$tdir
local def_max_rr=1
local def_max=3
local count
$LFS setdirstripe -D -i-1 -c1 --max-inherit=$def_max \
--max-inherit-rr=$def_max_rr $DIR ||
error "set $DIR default LMV failed"
for i in $(seq 2 3); do
def_max=$((def_max - 1))
(( def_max_rr == 0 )) || def_max_rr=$((def_max_rr - 1))
mkdir $testdir
# RR is decremented and keeps zeroed once exhausted
count=$($LFS getdirstripe -D --max-inherit-rr $testdir)
(( count == def_max_rr )) ||
error_noexit "$testdir: max-inherit-rr $count != $def_max_rr"
# max-inherit is decremented
count=$($LFS getdirstripe -D --max-inherit $testdir)
(( count == def_max )) ||
error_noexit "$testdir: max-inherit $count != $def_max"
testdir=$testdir/d$i
done
# d3 is the last inherited from ROOT, no inheritance anymore
# i.e. no the default layout anymore
mkdir -p $testdir/d4/d5
count=$($LFS getdirstripe -D --max-inherit $testdir)
(( count == -1 )) ||
error_noexit "$testdir: max-inherit $count != -1"
local p_count=$($LFS getdirstripe -i $testdir)
for i in $(seq 4 5); do
testdir=$testdir/d$i
# the root default layout is not applied once exhausted
count=$($LFS getdirstripe -i $testdir)
(( count == p_count )) ||
error_noexit "$testdir: stripe-offset $count != parent offset $p_count"
done
$LFS setdirstripe -i 0 $DIR/d2
count=$($LFS getdirstripe -D --max-inherit $DIR/d2)
(( count == -1 )) ||
error_noexit "$DIR/d2: max-inherit non-striped default $count != -1"
}
run_test 413i "check default layout inheritance"
test_413j()
{
(( $MDSCOUNT > 1 )) || skip_env "needs >= 2 MDTs"
local mdts=$(mdts_nodes)
mkdir -p $DIR/$tdir || error "mkdir $tdir failed"
$LFS setdirstripe -D -c2 --max-inherit=2 $DIR/$tdir ||
error "setdirstripe $tdir failed"
local value=$(getfattr -n trusted.dmv $DIR/$tdir | \
grep "trusted.dmv" |sed -e 's/[^=]\+=//')
mkdir -p $DIR/$tdir/sub || error "mkdir sub failed"
# setfattr dmv calls setdirstripe -D
setfattr -n trusted.dmv -v $value $DIR/$tdir/sub ||
error "setfattr sub failed"
local value2=$(getfattr -n trusted.dmv $DIR/$tdir/sub | \
grep "trusted.dmv" |sed -e 's/[^=]\+=//')
[ $value == $value2 ] || error "dmv mismatch"
(( MDS1_VERSION >= $(version_code 2.15.58) )) || return 0
# do not allow remove dmv by setfattr -x
do_nodes $mdts "$LCTL set_param -n mdt.*MDT*.enable_dmv_xattr=0"
setfattr -x trusted.dmv $DIR/$tdir/sub || error "setfattr sub failed"
getfattr -n trusted.dmv $DIR/$tdir/sub || error "default LMV deleted"
# allow remove dmv by setfattr -x
do_nodes $mdts "$LCTL set_param -n mdt.*MDT*.enable_dmv_xattr=1"
setfattr -x trusted.dmv $DIR/$tdir/sub || error "setfattr sub failed"
getfattr -n trusted.dmv $DIR/$tdir/sub && error "default LMV exists"
do_nodes $mdts "$LCTL set_param -n mdt.*MDT*.enable_dmv_xattr=0"
}
run_test 413j "set default LMV by setxattr"
test_413k() {
local index1
local index2
local old=$($LCTL get_param -n lmv.*.qos_exclude_prefixes)
local count=$($LCTL get_param -n lmv.*.qos_exclude_prefixes | wc -l)
local prefixes="abc:123:foo bar"
# add prefixes
stack_trap "$LCTL set_param lmv.*.qos_exclude_prefixes=\"$old\""
$LCTL set_param lmv.*.qos_exclude_prefixes="+$prefixes"
mkdir $DIR/$tdir || error "mkdir $tdir failed"
index1=$($LFS getstripe -m $DIR/$tdir)
for dname in _temporary _temporary.XXXXXX abc 123 "foo bar"; do
mkdir "$DIR/$tdir/$dname" || error "mkdir $dname failed"
index2=$($LFS getstripe -m "$DIR/$tdir/$dname")
((index1 == index2)) ||
error "$tdir on MDT$index1, $dname on MDT$index2"
done
# remove prefixes
$LCTL set_param lmv.*.qos_exclude_prefixes="-$prefixes"
# total prefixes length > PAGE_SIZE can be printed correctly
for c in {a..z}; do
prefixes=$(str_repeat $c 255)
$LCTL set_param lmv.*.qos_exclude_prefixes="+$prefixes" >/dev/null
done
local count2=$($LCTL get_param -n lmv.*.qos_exclude_prefixes | wc -l)
((count2 == count + 26)) ||
error "prefixes count $count2 != $((count + 26))"
}
run_test 413k "QoS mkdir exclude prefixes"
test_413l() {
local index1
local index2
local old=$($LCTL get_param -n lmv.*.qos_exclude_patterns)
local count=$($LCTL get_param -n lmv.*.qos_exclude_patterns | wc -l)
local patterns="????-????:[a-z][0-9]"
# add patterns
stack_trap "$LCTL set_param lmv.*.qos_exclude_patterns=\"$old\""
$LCTL set_param lmv.*.qos_exclude_patterns="+$patterns"
mkdir $DIR/$tdir || error "mkdir $tdir failed"
index1=$($LFS getstripe -m $DIR/$tdir)
for dname in abcd-1234 b9; do
mkdir "$DIR/$tdir/$dname" || error "mkdir $dname failed"
index2=$($LFS getstripe -m "$DIR/$tdir/$dname")
((index1 == index2)) ||
error "$tdir on MDT$index1, $dname on MDT$index2"
done
# remove patterns
$LCTL set_param lmv.*.qos_exclude_patterns="-$patterns"
}
run_test 413l "QoS mkdir exclude patterns"
test_413z() {
local pids=""
local subdir
local pid
for subdir in $(\ls -1 -d $DIR/d413*-fillmdt/*); do
unlinkmany $subdir/f. $TEST413_COUNT &
pids="$pids $!"
done
for pid in $pids; do
wait $pid
done
true
}
run_test 413z "413 test cleanup"
test_414() {
#define OBD_FAIL_PTLRPC_BULK_ATTACH 0x521
$LCTL set_param fail_loc=0x80000521
dd if=/dev/zero of=$DIR/$tfile bs=2M count=1 oflag=sync
rm -f $DIR/$tfile
# This error path has sometimes left inflight requests dangling, so
# test for this by remounting the client (umount will hang if there's
# a dangling request)
umount_client $MOUNT
mount_client $MOUNT
}
run_test 414 "simulate ENOMEM in ptlrpc_register_bulk()"
test_415() {
[[ $PARALLEL == "yes" ]] && skip "skip parallel run"
(( $MDS1_VERSION >= $(version_code 2.11.52) )) ||
skip "Need server version at least 2.11.52"
# LU-11102
local total=500
local max=120
# this test may be slow on ZFS
[[ "$mds1_FSTYPE" == "zfs" ]] && total=50
# though this test is designed for striped directory, let's test normal
# directory too since lock is always saved as CoS lock.
test_mkdir $DIR/$tdir || error "mkdir $tdir"
createmany -o $DIR/$tdir/$tfile. $total || error "createmany"
stack_trap "unlinkmany $DIR/$tdir/$tfile. $total || true"
# if looping with ONLY_REPEAT, wait for previous deletions to finish
wait_delete_completed_mds
# run a loop without concurrent touch to measure rename duration.
# only for test debug/robustness, NOT part of COS functional test.
local start_time=$SECONDS
for ((i = 0; i < total; i++)); do
mrename $DIR/$tdir/$tfile.$i $DIR/$tdir/$tfile-new.$i \
> /dev/null
done
local baseline=$((SECONDS - start_time))
echo "rename $total files without 'touch' took $baseline sec"
(
while true; do
touch $DIR/$tdir
done
) &
local setattr_pid=$!
# rename files back to original name so unlinkmany works
start_time=$SECONDS
for ((i = 0; i < total; i++)); do
mrename $DIR/$tdir/$tfile-new.$i $DIR/$tdir/$tfile.$i\
> /dev/null
done
local duration=$((SECONDS - start_time))
kill -9 $setattr_pid
echo "rename $total files with 'touch' took $duration sec"
(( max > 2 * baseline )) || max=$((2 * baseline + 5))
(( duration <= max )) ||
error_not_in_vm "rename took $duration > $max sec"
}
run_test 415 "lock revoke is not missing"
test_416() {
[ $MDS1_VERSION -lt $(version_code 2.11.55) ] &&
skip "Need server version at least 2.11.55"
# define OBD_FAIL_OSD_TXN_START 0x19a
do_facet mds1 lctl set_param fail_loc=0x19a
lfs mkdir -c $MDSCOUNT $DIR/$tdir
true
}
run_test 416 "transaction start failure won't cause system hung"
cleanup_417() {
local mdts=$(mdts_nodes)
trap 0
do_nodes $mdts "$LCTL set_param -n mdt.*MDT*.enable_dir_migration=1"
do_nodes $mdts "$LCTL set_param -n mdt.*MDT*.enable_remote_dir=1"
do_nodes $mdts "$LCTL set_param -n mdt.*MDT*.enable_striped_dir=1"
}
test_417() {
[ $MDSCOUNT -lt 2 ] && skip "needs >= 2 MDTs"
[[ $MDS1_VERSION -lt $(version_code 2.11.56) ]] &&
skip "Need MDS version at least 2.11.56"
trap cleanup_417 RETURN EXIT
local mdts=$(mdts_nodes)
$LFS mkdir -i 1 $DIR/$tdir.1 || error "create remote dir $tdir.1 failed"
do_nodes $mdts "$LCTL set_param -n mdt.*MDT*.enable_dir_migration=0"
$LFS migrate -m 0 $DIR/$tdir.1 &&
error "migrate dir $tdir.1 should fail"
do_nodes $mdts "$LCTL set_param -n mdt.*MDT*.enable_remote_dir=0"
$LFS mkdir -i 1 $DIR/$tdir.2 &&
error "create remote dir $tdir.2 should fail"
do_nodes $mdts "$LCTL set_param -n mdt.*MDT*.enable_striped_dir=0"
$LFS mkdir -c 2 $DIR/$tdir.3 &&
error "create striped dir $tdir.3 should fail"
true
}
run_test 417 "disable remote dir, striped dir and dir migration"
# Checks that the outputs of df [-i] and lfs df [-i] match
#
# usage: check_lfs_df <blocks | inodes> <mountpoint>
check_lfs_df() {
local dir=$2
local inodes
local df_out
local lfs_df_out
local count
local passed=false
# blocks or inodes
[ "$1" == "blocks" ] && inodes= || inodes="-i"
for count in {1..100}; do
do_nodes "$CLIENTS" \
$LCTL set_param ldlm.namespaces.*.lru_size=clear
sync; sleep 0.2
# read the lines of interest
df_out=($(df -P $inodes $dir | tail -n +2)) ||
error "df $inodes $dir | tail -n +2 failed"
lfs_df_out=($($LFS df $inodes $dir | grep summary:)) ||
error "lfs df $inodes $dir | grep summary: failed"
# skip first substrings of each output as they are different
# "<NID>:/<fsname>" for df, "filesystem_summary:" for lfs df
# compare the two outputs
passed=true
# skip "available" on MDT until LU-13997 is fixed.
#for i in {1..5}; do
for i in 1 2 4 5; do
[ "${df_out[i]}" != "${lfs_df_out[i]}" ] && passed=false
done
$passed && break
done
if ! $passed; then
df -P $inodes $dir
echo
lfs df $inodes $dir
error "df and lfs df $1 output mismatch: " \
"df ${inodes}: ${df_out[*]}, " \
"lfs df ${inodes}: ${lfs_df_out[*]}"
fi
}
test_418() {
[ $PARALLEL == "yes" ] && skip "skip parallel run"
local dir=$DIR/$tdir
local numfiles=$((RANDOM % 4096 + 2))
local numblocks=$((RANDOM % 256 + 1))
wait_delete_completed
test_mkdir $dir
# check block output
check_lfs_df blocks $dir
# check inode output
check_lfs_df inodes $dir
# create a single file and retest
echo "Creating a single file and testing"
createmany -o $dir/$tfile- 1 &>/dev/null ||
error "creating 1 file in $dir failed"
check_lfs_df blocks $dir
check_lfs_df inodes $dir
# create a random number of files
echo "Creating $((numfiles - 1)) files and testing"
createmany -o $dir/$tfile- 1 $((numfiles - 1)) &>/dev/null ||
error "creating $((numfiles - 1)) files in $dir failed"
# write a random number of blocks to the first test file
echo "Writing $numblocks 4K blocks and testing"
dd if=/dev/urandom of=$dir/${tfile}-0 bs=4K conv=fsync \
count=$numblocks &>/dev/null ||
error "dd to $dir/${tfile}-0 failed"
# retest
check_lfs_df blocks $dir
check_lfs_df inodes $dir
unlinkmany $dir/$tfile- $numfiles &>/dev/null ||
error "unlinking $numfiles files in $dir failed"
}
run_test 418 "df and lfs df outputs match"
test_419()
{
local dir=$DIR/$tdir
mkdir -p $dir
touch $dir/file
cancel_lru_locks mdc
#OBD_FAIL_LLITE_OPEN_BY_NAME 0x1410
$LCTL set_param fail_loc=0x1410
cat $dir/file
$LCTL set_param fail_loc=0
rm -rf $dir
}
run_test 419 "Verify open file by name doesn't crash kernel"
test_420()
{
[[ $MDS1_VERSION -ge $(version_code 2.12.53) ]] ||
skip "Need MDS version at least 2.12.53"
local SAVE_UMASK=$(umask)
local dir=$DIR/$tdir
local uname=$(getent passwd $RUNAS_ID | cut -d: -f1)
mkdir -p $dir
umask 0000
mkdir -m03777 $dir/testdir
ls -dn $dir/testdir
# Need to remove trailing '.' when SELinux is enabled
local dirperms=$(ls -dn $dir/testdir |
awk '{ sub(/\.$/, "", $1); print $1}')
[ $dirperms == "drwxrwsrwt" ] ||
error "incorrect perms on $dir/testdir"
su - $uname -c "PATH=$LUSTRE/tests:\$PATH; \
openfile -f O_RDONLY:O_CREAT -m 02755 $dir/testdir/testfile"
ls -n $dir/testdir/testfile
local fileperms=$(ls -n $dir/testdir/testfile |
awk '{ sub(/\.$/, "", $1); print $1}')
[ $fileperms == "-rwxr-xr-x" ] ||
error "incorrect perms on $dir/testdir/testfile"
umask $SAVE_UMASK
}
run_test 420 "clear SGID bit on non-directories for non-members"
test_421a() {
local cnt
local fid1
local fid2
[ $MDS1_VERSION -lt $(version_code 2.12.54) ] &&
skip "Need MDS version at least 2.12.54"
test_mkdir $DIR/$tdir
createmany -o $DIR/$tdir/f 3
cnt=$(ls -1 $DIR/$tdir | wc -l)
[ $cnt != 3 ] && error "unexpected #files: $cnt"
fid1=$(lfs path2fid $DIR/$tdir/f1)
fid2=$(lfs path2fid $DIR/$tdir/f2)
$LFS rmfid $DIR $fid1 $fid2 || error "rmfid failed"
stat $DIR/$tdir/f1 && error "f1 still visible on the client"
stat $DIR/$tdir/f2 && error "f2 still visible on the client"
cnt=$(ls -1 $DIR/$tdir | wc -l)
[ $cnt == 1 ] || error "unexpected #files after: $cnt"
rm -f $DIR/$tdir/f3 || error "can't remove f3"
createmany -o $DIR/$tdir/f 3
cnt=$(ls -1 $DIR/$tdir | wc -l)
[ $cnt != 3 ] && error "unexpected #files: $cnt"
fid1=$(lfs path2fid $DIR/$tdir/f1)
fid2=$(lfs path2fid $DIR/$tdir/f2)
echo "remove using fsname $FSNAME"
$LFS rmfid $FSNAME $fid1 $fid2 || error "rmfid with fsname failed"
cnt=$(ls -1 $DIR/$tdir | wc -l)
[ $cnt == 1 ] || error "unexpected #files after: $cnt"
}
run_test 421a "simple rm by fid"
test_421b() {
local cnt
local FID1
local FID2
[ $MDS1_VERSION -lt $(version_code 2.12.54) ] &&
skip "Need MDS version at least 2.12.54"
test_mkdir $DIR/$tdir
createmany -o $DIR/$tdir/f 3
multiop_bg_pause $DIR/$tdir/f1 o_c || error "multiop failed to start"
MULTIPID=$!
FID1=$(lfs path2fid $DIR/$tdir/f1)
FID2=$(lfs path2fid $DIR/$tdir/f2)
$LFS rmfid $DIR $FID1 $FID2 && error "rmfid didn't fail"
kill -USR1 $MULTIPID
wait
cnt=$(ls $DIR/$tdir | wc -l)
[ $cnt == 2 ] || error "unexpected #files after: $cnt"
}
run_test 421b "rm by fid on open file"
test_421c() {
local cnt
local FIDS
[ $MDS1_VERSION -lt $(version_code 2.12.54) ] &&
skip "Need MDS version at least 2.12.54"
test_mkdir $DIR/$tdir
createmany -o $DIR/$tdir/f 3
touch $DIR/$tdir/$tfile
createmany -l$DIR/$tdir/$tfile $DIR/$tdir/h 180
cnt=$(ls -1 $DIR/$tdir | wc -l)
[ $cnt != 184 ] && error "unexpected #files: $cnt"
FID1=$(lfs path2fid $DIR/$tdir/$tfile)
$LFS rmfid $DIR $FID1 || error "rmfid failed"
cnt=$(ls $DIR/$tdir | wc -l)
[ $cnt == 3 ] || error "unexpected #files after: $cnt"
}
run_test 421c "rm by fid against hardlinked files"
test_421d() {
local cnt
local FIDS
[ $MDS1_VERSION -lt $(version_code 2.12.54) ] &&
skip "Need MDS version at least 2.12.54"
test_mkdir $DIR/$tdir
createmany -o $DIR/$tdir/f 4097
cnt=$(ls -1 $DIR/$tdir | wc -l)
[ $cnt != 4097 ] && error "unexpected #files: $cnt"
FIDS=$(lfs path2fid $DIR/$tdir/f* | sed "s/[/][^:]*://g")
$LFS rmfid $DIR $FIDS || error "rmfid failed"
cnt=$(ls $DIR/$tdir | wc -l)
rm -rf $DIR/$tdir
[ $cnt == 0 ] || error "unexpected #files after: $cnt"
}
run_test 421d "rmfid en masse"
test_421e() {
local cnt
local FID
[ $MDSCOUNT -lt 2 ] && skip "needs >= 2 MDTs"
[ $MDS1_VERSION -lt $(version_code 2.12.54) ] &&
skip "Need MDS version at least 2.12.54"
mkdir -p $DIR/$tdir
$LFS setdirstripe -c$MDSCOUNT $DIR/$tdir/striped_dir
createmany -o $DIR/$tdir/striped_dir/f 512
cnt=$(ls -1 $DIR/$tdir/striped_dir | wc -l)
[ $cnt != 512 ] && error "unexpected #files: $cnt"
FIDS=$(lfs path2fid $DIR/$tdir/striped_dir/f* |
sed "s/[/][^:]*://g")
$LFS rmfid $DIR $FIDS || error "rmfid failed"
cnt=$(ls $DIR/$tdir/striped_dir | wc -l)
rm -rf $DIR/$tdir
[ $cnt == 0 ] || error "unexpected #files after: $cnt"
}
run_test 421e "rmfid in DNE"
test_421f() {
local cnt
local FID
[ $MDS1_VERSION -lt $(version_code 2.12.54) ] &&
skip "Need MDS version at least 2.12.54"
test_mkdir $DIR/$tdir
touch $DIR/$tdir/f
cnt=$(ls -1 $DIR/$tdir | wc -l)
[ $cnt != 1 ] && error "unexpected #files: $cnt"
FID=$(lfs path2fid $DIR/$tdir/f)
$RUNAS $LFS rmfid $DIR $FID && error "rmfid didn't fail (1)"
# rmfid should fail
cnt=$(ls -1 $DIR/$tdir | wc -l)
[ $cnt != 1 ] && error "unexpected #files after (2): $cnt"
chmod a+rw $DIR/$tdir
ls -la $DIR/$tdir
$RUNAS $LFS rmfid $DIR $FID && error "rmfid didn't fail (2)"
# rmfid should fail
cnt=$(ls -1 $DIR/$tdir | wc -l)
[ $cnt != 1 ] && error "unexpected #files after (3): $cnt"
rm -f $DIR/$tdir/f
$RUNAS touch $DIR/$tdir/f
FID=$(lfs path2fid $DIR/$tdir/f)
echo "rmfid as root"
$LFS rmfid $DIR $FID || error "rmfid as root failed"
cnt=$(ls -1 $DIR/$tdir | wc -l)
[ $cnt == 0 ] || error "unexpected #files after (4): $cnt"
rm -f $DIR/$tdir/f
$RUNAS touch $DIR/$tdir/f
cnt=$(ls -1 $DIR/$tdir | wc -l)
[ $cnt != 1 ] && error "unexpected #files (4): $cnt"
FID=$(lfs path2fid $DIR/$tdir/f)
# rmfid w/o user_fid2path mount option should fail
$RUNAS $LFS rmfid $DIR $FID && error "rmfid didn't fail(3)"
cnt=$(ls -1 $DIR/$tdir | wc -l)
[ $cnt == 1 ] || error "unexpected #files after (5): $cnt"
tmpdir=$(mktemp -d /tmp/lustre-XXXXXX)
stack_trap "rmdir $tmpdir"
mount_client $tmpdir "$MOUNT_OPTS,user_fid2path" ||
error "failed to mount client'"
stack_trap "umount_client $tmpdir"
$RUNAS $LFS rmfid $tmpdir $FID || error "rmfid failed"
# rmfid should succeed
cnt=$(ls -1 $tmpdir/$tdir | wc -l)
[ $cnt == 0 ] || error "unexpected #files after (6): $cnt"
# rmfid shouldn't allow to remove files due to dir's permission
chmod a+rwx $tmpdir/$tdir
touch $tmpdir/$tdir/f
ls -la $tmpdir/$tdir
FID=$(lfs path2fid $tmpdir/$tdir/f)
$RUNAS $LFS rmfid $tmpdir $FID && error "rmfid didn't fail"
return 0
}
run_test 421f "rmfid checks permissions"
test_421g() {
local cnt
local FIDS
[ $MDSCOUNT -lt 2 ] && skip "needs >= 2 MDTs"
[ $MDS1_VERSION -lt $(version_code 2.12.54) ] &&
skip "Need MDS version at least 2.12.54"
mkdir -p $DIR/$tdir
$LFS setdirstripe -c$MDSCOUNT $DIR/$tdir/striped_dir
createmany -o $DIR/$tdir/striped_dir/f 512
cnt=$(ls -1 $DIR/$tdir/striped_dir | wc -l)
[ $cnt != 512 ] && error "unexpected #files: $cnt"
FIDS=$(lfs path2fid $DIR/$tdir/striped_dir/f* |
sed "s/[/][^:]*://g")
rm -f $DIR/$tdir/striped_dir/f1*
cnt=$(ls -1 $DIR/$tdir/striped_dir | wc -l)
removed=$((512 - cnt))
# few files have been just removed, so we expect
# rmfid to fail on their fids
errors=$($LFS rmfid $DIR $FIDS 2>&1 | wc -l)
[ $removed != $errors ] && error "$errors != $removed"
cnt=$(ls $DIR/$tdir/striped_dir | wc -l)
rm -rf $DIR/$tdir
[ $cnt == 0 ] || error "unexpected #files after: $cnt"
}
run_test 421g "rmfid to return errors properly"
test_421h() {
local mount_other
local mount_ret
local rmfid_ret
local old_fid
local fidA
local fidB
local fidC
local fidD
(( MDS1_VERSION >= $(version_code 2.15.53) )) ||
skip "Need MDS version at least 2.15.53"
test_mkdir $DIR/$tdir
test_mkdir $DIR/$tdir/subdir
touch $DIR/$tdir/subdir/file0
old_fid=$(lfs path2fid $DIR/$tdir/subdir/file0 | sed "s/[/][^:]*://g")
echo File $DIR/$tdir/subdir/file0 FID $old_fid
rm -f $DIR/$tdir/subdir/file0
touch $DIR/$tdir/subdir/fileA
fidA=$(lfs path2fid $DIR/$tdir/subdir/fileA | sed "s/[/][^:]*://g")
echo File $DIR/$tdir/subdir/fileA FID $fidA
touch $DIR/$tdir/subdir/fileB
fidB=$(lfs path2fid $DIR/$tdir/subdir/fileB | sed "s/[/][^:]*://g")
echo File $DIR/$tdir/subdir/fileB FID $fidB
ln $DIR/$tdir/subdir/fileB $DIR/$tdir/subdir/fileB_hl
touch $DIR/$tdir/subdir/fileC
fidC=$(lfs path2fid $DIR/$tdir/subdir/fileC | sed "s/[/][^:]*://g")
echo File $DIR/$tdir/subdir/fileC FID $fidC
ln $DIR/$tdir/subdir/fileC $DIR/$tdir/fileC
touch $DIR/$tdir/fileD
fidD=$(lfs path2fid $DIR/$tdir/fileD | sed "s/[/][^:]*://g")
echo File $DIR/$tdir/fileD FID $fidD
# mount another client mount point with subdirectory mount
mount_other=${MOUNT}_other
FILESET="$FILESET/$tdir/subdir" mount_client $mount_other ${MOUNT_OPTS}
mount_ret=$?
FILESET=""
(( mount_ret == 0 )) || error "mount $mount_other failed"
echo Removing FIDs:
echo $LFS rmfid $mount_other $old_fid $fidA $fidD $fidB $fidC
$LFS rmfid $mount_other $old_fid $fidA $fidD $fidB $fidC
rmfid_ret=$?
umount_client $mount_other || error "umount $mount_other failed"
rm -rf $mount_other
(( rmfid_ret != 0 )) || error "rmfid should have failed"
# fileA should have been deleted
stat $DIR/$tdir/subdir/fileA && error "fileA not deleted"
# fileB should have been deleted
stat $DIR/$tdir/subdir/fileB && error "fileB not deleted"
# fileC should not have been deleted, fid also exists outside of fileset
stat $DIR/$tdir/subdir/fileC || error "fileC deleted"
# fileD should not have been deleted, it exists outside of fileset
stat $DIR/$tdir/fileD || error "fileD deleted"
}
run_test 421h "rmfid with fileset mount"
test_422() {
test_mkdir -i 0 -c 1 -p $DIR/$tdir/d1
test_mkdir -i 0 -c 1 -p $DIR/$tdir/d2
test_mkdir -i 0 -c 1 -p $DIR/$tdir/d3
dd if=/dev/zero of=$DIR/$tdir/d1/file1 bs=1k count=1
dd if=/dev/zero of=$DIR/$tdir/d2/file1 bs=1k count=1
local amc=$(at_max_get client)
local amo=$(at_max_get mds1)
local timeout=`lctl get_param -n timeout`
at_max_set 0 client
at_max_set 0 mds1
#define OBD_FAIL_PTLRPC_PAUSE_REQ 0x50a
do_facet mds1 $LCTL set_param fail_loc=0x8000050a \
fail_val=$(((2*timeout + 10)*1000))
touch $DIR/$tdir/d3/file &
sleep 2
#define OBD_FAIL_TGT_REPLY_DATA_RACE 0x722
do_facet mds1 $LCTL set_param fail_loc=0x80000722 \
fail_val=$((2*timeout + 5))
mv $DIR/$tdir/d1/file1 $DIR/$tdir/d1/file2 &
local pid=$!
sleep 1
kill -9 $pid
sleep $((2 * timeout))
echo kill $pid
kill -9 $pid
lctl mark touch
touch $DIR/$tdir/d2/file3
touch $DIR/$tdir/d2/file4
touch $DIR/$tdir/d2/file5
wait
at_max_set $amc client
at_max_set $amo mds1
# LU-12838 - verify the ptlrpc thread watchdog is not always throttled
do_facet mds1 "dmesg | grep 'Dumping the stack trace for debugging'" ||
error "Watchdog is always throttled"
}
run_test 422 "kill a process with RPC in progress"
stat_test() {
df -h $MOUNT &
df -h $MOUNT &
df -h $MOUNT &
df -h $MOUNT &
df -h $MOUNT &
df -h $MOUNT &
}
test_423() {
local _stats
# ensure statfs cache is expired
sleep 2;
_stats=$(stat_test | grep $MOUNT | sort -u | wc -l)
[[ ${_stats} -ne 1 ]] && error "statfs wrong"
return 0
}
run_test 423 "statfs should return a right data"
test_424() {
#define OBD_FAIL_PTLRPC_BULK_REPLY_ATTACH 0x522 | CFS_FAIL_ONCE
$LCTL set_param fail_loc=0x80000522
dd if=/dev/zero of=$DIR/$tfile bs=2M count=1 oflag=sync
rm -f $DIR/$tfile
}
run_test 424 "simulate ENOMEM in ptl_send_rpc bulk reply ME attach"
test_425() {
test_mkdir -c -1 $DIR/$tdir
$LFS setstripe -c -1 $DIR/$tdir
lru_resize_disable "" 100
sleep 5
stack_trap "unlinkmany $DIR/$tdir/$tfile $((MDSCOUNT * 125)) || true"
createmany -o -W 4096 $DIR/$tdir/$tfile $((MDSCOUNT * 125)) ||
error_noexit "Create files in $DIR/$tdir"
for oscparam in $($LCTL list_param ldlm.namespaces.*osc-[-0-9a-f]*); do
local lru_size=$($LCTL get_param -n $oscparam.lru_size)
local lock_count=$($LCTL get_param -n $oscparam.lock_count)
# allow some margin for batched lock cancellation
(( $lock_count <= $lru_size * 11 / 10)) || {
$LCTL get_param $oscparam.lru_size $oscparam.lock_count
error "osc lock count $lock_count > lru size $lru_size"
}
done
for mdcparam in $($LCTL list_param ldlm.namespaces.*mdc-*); do
local lru_size=$($LCTL get_param -n $mdcparam.lru_size)
local lock_count=$($LCTL get_param -n $mdcparam.lock_count)
# allow some margin for batched lock cancellation
(( $lock_count <= $lru_size * 11 / 10)) || {
$LCTL get_param $mdcparam.lru_size $mdcparam.lock_count
error "mdc lock count $lock_count > lru size $lru_size"
}
done
}
run_test 425 "lock count should not exceed lru size"
test_426() {
splice-test -r $DIR/$tfile
splice-test -rd $DIR/$tfile
splice-test $DIR/$tfile
splice-test -d $DIR/$tfile
}
run_test 426 "splice test on Lustre"
test_427() {
[ $MDSCOUNT -ge 2 ] || skip "needs >= 2 MDTs"
(( $MDS1_VERSION >= $(version_code 2.12.4) )) ||
skip "Need MDS version at least 2.12.4"
local log
mkdir $DIR/$tdir
mkdir $DIR/$tdir/1
mkdir $DIR/$tdir/2
test_mkdir -c $MDSCOUNT -i 1 $DIR/$tdir/1/dir
test_mkdir -c $MDSCOUNT -i 1 $DIR/$tdir/2/dir2
$LFS getdirstripe $DIR/$tdir/1/dir
#first setfattr for creating updatelog
setfattr -n user.attr0 -v "some text" $DIR/$tdir/1/dir
#define OBD_FAIL_OUT_OBJECT_MISS 0x1708
do_nodes $(mdts_nodes) "$LCTL set_param fail_loc=0x80001708"
setfattr -n user.attr1 -v "some text" $DIR/$tdir/1/dir &
setfattr -n user.attr2 -v "another attr" $DIR/$tdir/2/dir2 &
sleep 2
fail mds2
wait_recovery_complete mds2 $((2*TIMEOUT))
log=$(do_facet mds1 dmesg | tac | sed "/${TESTNAME//_/ }/,$ d")
echo $log | grep "get update log failed" &&
error "update log corruption is detected" || true
}
run_test 427 "Failed DNE2 update request shouldn't corrupt updatelog"
test_428() {
[ $PARALLEL == "yes" ] && skip "skip parallel run"
local max_cached_mb=$($LCTL get_param llite.*.max_cached_mb |
awk '/^max_cached_mb/ { print $2 }')
stack_trap "$LCTL set_param -n llite.*.max_cached_mb=$max_cached_mb"
$LCTL set_param -n llite.*.max_cached_mb=64
mkdir $DIR/$tdir
$LFS setstripe -c 1 $DIR/$tdir
eval touch $DIR/$tdir/$tfile.{1..$OSTCOUNT}
stack_trap "rm -f $DIR/$tdir/$tfile.*"
#test write
for f in $(seq 4); do
dd if=/dev/zero of=$DIR/$tdir/$tfile.$f bs=128M count=1 &
done
wait
cancel_lru_locks osc
# Test read
for f in $(seq 4); do
dd if=$DIR/$tdir/$tfile.$f of=/dev/null bs=128M count=1 &
done
wait
}
run_test 428 "large block size IO should not hang"
test_429() { # LU-7915 / LU-10948
local ll_opencache_threshold_count="llite.*.opencache_threshold_count"
local testfile=$DIR/$tfile
local mdc_rpcstats="mdc.$FSNAME-MDT0000-*.stats"
local new_flag=1
local first_rpc
local second_rpc
local third_rpc
$LCTL get_param $ll_opencache_threshold_count ||
skip "client does not have opencache parameter"
set_opencache $new_flag
stack_trap "restore_opencache"
[ $($LCTL get_param -n $ll_opencache_threshold_count) == $new_flag ] ||
error "enable opencache failed"
touch $testfile
# drop MDC DLM locks
cancel_lru_locks mdc
# clear MDC RPC stats counters
$LCTL set_param $mdc_rpcstats=clear
# According to the current implementation, we need to run 3 times
# open & close file to verify if opencache is enabled correctly.
# 1st, RPCs are sent for lookup/open and open handle is released on
# close finally.
# 2nd, RPC is sent for open, MDS_OPEN_LOCK is fetched automatically,
# so open handle won't be released thereafter.
# 3rd, No RPC is sent out.
$MULTIOP $testfile oc || error "multiop failed"
first_rpc=$(calc_stats $mdc_rpcstats ldlm_ibits_enqueue)
echo "1st: $first_rpc RPCs in flight"
$MULTIOP $testfile oc || error "multiop failed"
second_rpc=$(calc_stats $mdc_rpcstats ldlm_ibits_enqueue)
echo "2nd: $second_rpc RPCs in flight"
$MULTIOP $testfile oc || error "multiop failed"
third_rpc=$(calc_stats $mdc_rpcstats ldlm_ibits_enqueue)
echo "3rd: $third_rpc RPCs in flight"
#verify no MDC RPC is sent
[[ $second_rpc == $third_rpc ]] || error "MDC RPC is still sent"
}
run_test 429 "verify if opencache flag on client side does work"
# write data chunks in reverse order to defeat ZFS automatic blocksize
# growth during appends causing SEEK_HOLE to return unexpected offsets
dd_reverse() {
local of=$1
local bs=$2
local count=${3##count=}
local start=${4##seek=}
local end=$((start + count - 1))
local dd_opts="if=/dev/urandom $of $bs count=1"
local i
for ((i = end; i >= start; i--)); do
dd $dd_opts seek=$i oflag=direct conv=notrunc 2>&1 |
grep -v -E "copied|bytes"
if (( ${PIPESTATUS[0]} != 0 )); then
error "${of##of=}: write $start->$end @$bs failed"
return 1
fi
done
}
lseek_test_430() {
local offset
local file=$1
local i
# data at [256K, 512K)
dd_reverse of=$file bs=64K count=4 seek=4 || error "256K->512K dd fails"
# data at [2M, 3M)
dd_reverse of=$file bs=64K count=16 seek=32 || error "2M->3M dd fails"
# data at [4M, 5M)
dd_reverse of=$file bs=64K count=16 seek=64 || error "4M->5M dd fails"
echo "Data at 256K...512K, 2M...3M and 4M...5M"
# start at first component hole #1
printf "Seeking hole from 1000 ... "
offset=$(lseek_test -l 1000 $file)
echo $offset
[[ $offset == 1000 ]] || error "offset $offset != 1000"
printf "Seeking data from 1000 ... "
offset=$(lseek_test -d 1000 $file)
echo $offset
[[ $offset == 262144 ]] || error "offset $offset != 262144"
# start at first component data block
printf "Seeking hole from 300000 ... "
offset=$(lseek_test -l 300000 $file)
echo $offset
[[ $offset == 524288 ]] || error "offset $offset != 524288"
printf "Seeking data from 300000 ... "
offset=$(lseek_test -d 300000 $file)
echo $offset
[[ $offset == 300000 ]] || error "offset $offset != 300000"
# start at the first component but beyond end of object size
printf "Seeking hole from 1000000 ... "
offset=$(lseek_test -l 1000000 $file)
echo $offset
[[ $offset == 1000000 ]] || error "offset $offset != 1000000"
printf "Seeking data from 1000000 ... "
offset=$(lseek_test -d 1000000 $file)
echo $offset
[[ $offset == 2097152 ]] || error "offset $offset != 2097152"
# start at second component stripe 2 (empty file)
printf "Seeking hole from 1500000 ... "
offset=$(lseek_test -l 1500000 $file)
echo $offset
[[ $offset == 1500000 ]] || error "offset $offset != 1500000"
printf "Seeking data from 1500000 ... "
offset=$(lseek_test -d 1500000 $file)
echo $offset
[[ $offset == 2097152 ]] || error "offset $offset != 2097152"
# start at second component stripe 1 (all data)
printf "Seeking hole from 3000000 ... "
offset=$(lseek_test -l 3000000 $file)
echo $offset
[[ $offset == 3145728 ]] || error "offset $offset != 3145728"
printf "Seeking data from 3000000 ... "
offset=$(lseek_test -d 3000000 $file)
echo $offset
[[ $offset == 3000000 ]] || error "offset $offset != 3000000"
dd_reverse of=$file bs=64K count=10 seek=10 || error "2nd dd fails"
echo "Add data block at 640K...1280K"
# start at before new data block, in hole
printf "Seeking hole from 600000 ... "
offset=$(lseek_test -l 600000 $file)
echo $offset
[[ $offset == 600000 ]] || error "offset $offset != 600000"
printf "Seeking data from 600000 ... "
offset=$(lseek_test -d 600000 $file)
echo $offset
[[ $offset == 655360 ]] || error "offset $offset != 655360"
# start at the first component new data block
printf "Seeking hole from 1000000 ... "
offset=$(lseek_test -l 1000000 $file)
echo $offset
[[ $offset == 1310720 ]] || error "offset $offset != 1310720"
printf "Seeking data from 1000000 ... "
offset=$(lseek_test -d 1000000 $file)
echo $offset
[[ $offset == 1000000 ]] || error "offset $offset != 1000000"
# start at second component stripe 2, new data
printf "Seeking hole from 1200000 ... "
offset=$(lseek_test -l 1200000 $file)
echo $offset
[[ $offset == 1310720 ]] || error "offset $offset != 1310720"
printf "Seeking data from 1200000 ... "
offset=$(lseek_test -d 1200000 $file)
echo $offset
[[ $offset == 1200000 ]] || error "offset $offset != 1200000"
# start beyond file end
printf "Using offset > filesize ... "
lseek_test -l 6000000 $file && error "lseek should fail"
printf "Using offset > filesize ... "
lseek_test -d 6000000 $file && error "lseek should fail"
printf "Done\n\n"
}
test_430a() {
$LCTL get_param osc.*.import | grep -q 'connect_flags:.*seek' ||
skip "OST does not support SEEK_HOLE"
local file=$DIR/$tdir/$tfile
mkdir -p $DIR/$tdir
$LFS setstripe -E 1M -c2 -S 64K -E 10M -c2 -S 1M $file
echo "Component #1: 1M, 2 stripes 64K, component #2: EOF, 2 stripes 1M"
lseek_test_430 $file
rm $file
$LFS setstripe -c2 -S 512K $file
echo "Two stripes, stripe size 512K"
lseek_test_430 $file
rm $file
# FLR with stale mirror
$LFS setstripe -N -E 512K -c1 -S 64K -E eof -c2 -S 512K \
-N -c2 -S 1M $file
echo "Mirrored file:"
echo "Component #1: 512K, stripe 64K, component #2: EOF, 2 stripes 512K"
echo "Plain 2 stripes 1M"
lseek_test_430 $file
rm $file
$LCTL get_param mdc.*.import | grep -q 'connect_flags:.*seek' ||
skip "MDT does not support SEEK_HOLE"
$LFS setstripe -E 1M -L mdt -E eof -c2 $file
# OST stripe #1 will have continuous data at [1M, 3M)
# OST stripe #2 is empty
echo "Component #1: 1M DoM, component #2: EOF, 2 stripes 1M"
lseek_test_430 $file
rm $file
}
run_test 430a "lseek: SEEK_DATA/SEEK_HOLE basic functionality"
test_430b() {
$LCTL get_param osc.*.import | grep -q 'connect_flags:.*seek' ||
skip "OST does not support SEEK_HOLE"
local offset
local file=$DIR/$tdir/$tfile
mkdir -p $DIR/$tdir
# Empty layout lseek should fail
$MCREATE $file
# seek from 0
printf "Seeking hole from 0 ... "
lseek_test -l 0 $file && error "lseek should fail"
printf "Seeking data from 0 ... "
lseek_test -d 0 $file && error "lseek should fail"
rm $file
# 1M-hole file
$LFS setstripe -E 1M -c2 -E eof $file
$TRUNCATE $file 1048576
printf "Seeking hole from 1000000 ... "
offset=$(lseek_test -l 1000000 $file)
echo $offset
[[ $offset == 1000000 ]] || error "offset $offset != 1000000"
printf "Seeking data from 1000000 ... "
lseek_test -d 1000000 $file && error "lseek should fail"
rm $file
# full component followed by non-inited one
$LFS setstripe -E 1M -c2 -E eof $file
dd if=/dev/urandom of=$file bs=1M count=1
printf "Seeking hole from 1000000 ... "
offset=$(lseek_test -l 1000000 $file)
echo $offset
[[ $offset == 1048576 ]] || error "offset $offset != 1048576"
printf "Seeking hole from 1048576 ... "
lseek_test -l 1048576 $file && error "lseek should fail"
# init second component and truncate back
echo "123" >> $file
$TRUNCATE $file 1048576
printf "Seeking hole from 1000000 ... "
offset=$(lseek_test -l 1000000 $file)
echo $offset
[[ $offset == 1048576 ]] || error "offset $offset != 1048576"
printf "Seeking hole from 1048576 ... "
lseek_test -l 1048576 $file && error "lseek should fail"
# boundary checks for big values
dd if=/dev/urandom of=$file.10g bs=1 count=1 seek=10G
offset=$(lseek_test -d 0 $file.10g)
[[ $offset == 10737418240 ]] || error "offset $offset != 10737418240"
dd if=/dev/urandom of=$file.100g bs=1 count=1 seek=100G
offset=$(lseek_test -d 0 $file.100g)
[[ $offset == 107374182400 ]] || error "offset $offset != 107374182400"
return 0
}
run_test 430b "lseek: SEEK_DATA/SEEK_HOLE special cases"
test_430c() {
$LCTL get_param osc.*.import | grep -q 'connect_flags:.*seek' ||
skip "OST does not support SEEK_HOLE"
local file=$DIR/$tdir/$tfile
local start
mkdir -p $DIR/$tdir
stack_trap "rm -f $file $file.tmp"
dd if=/dev/urandom of=$file bs=1k count=1 seek=5M || error "dd failed"
# cp version 8.33+ prefers lseek over fiemap
local ver=$(cp --version | awk '{ print $4; exit; }')
echo "cp $ver installed"
if (( $(version_code $ver) >= $(version_code 8.33) )); then
start=$SECONDS
time cp -v $file $file.tmp || error "cp $file failed"
(( SECONDS - start < 5 )) || {
strace cp $file $file.tmp |&
grep -E "open|read|seek|FIEMAP" |
grep -A 100 $file
error "cp: too long runtime $((SECONDS - start))"
}
else
echo "cp test skipped due to $ver < 8.33"
fi
# tar version 1.29+ supports SEEK_HOLE/DATA
ver=$(tar --version | awk '{ print $4; exit; }')
echo "tar $ver installed"
if (( $(version_code $ver) >= $(version_code 1.29) )); then
start=$SECONDS
time tar cvf $file.tmp --sparse $file || error "tar $file error"
(( SECONDS - start < 5 )) || {
strace tar cf $file.tmp --sparse $file |&
grep -E "open|read|seek|FIEMAP" |
grep -A 100 $file
error "tar: too long runtime $((SECONDS - start))"
}
else
echo "tar test skipped due to $ver < 1.29"
fi
}
run_test 430c "lseek: external tools check"
test_431() { # LU-14187
local file=$DIR/$tdir/$tfile
mkdir -p $DIR/$tdir
$LFS setstripe -c 1 -i 0 $file || error "lfs setstripe failed"
dd if=/dev/urandom of=$file bs=4k count=1
dd if=/dev/urandom of=$file bs=4k count=1 seek=10 conv=notrunc
dd if=/dev/urandom of=$file bs=4k count=1 seek=12 conv=notrunc
#define OBD_FAIL_OST_RESTART_IO 0x251
do_facet ost1 "$LCTL set_param fail_loc=0x251"
$LFS setstripe -c 1 -i 0 $file.0 || error "lfs setstripe failed"
cp $file $file.0
cancel_lru_locks
sync_all_data
echo 3 > /proc/sys/vm/drop_caches
diff $file $file.0 || error "data diff"
}
run_test 431 "Restart transaction for IO"
cleanup_test_432() {
do_facet mgs $LCTL nodemap_activate 0
wait_nm_sync active
}
test_432() {
local tmpdir=$TMP/dir432
(( $MDS1_VERSION >= $(version_code 2.14.52) )) ||
skip "Need MDS version at least 2.14.52"
stack_trap cleanup_test_432 EXIT
mkdir $DIR/$tdir
mkdir $tmpdir
do_facet mgs $LCTL nodemap_activate 1
wait_nm_sync active
do_facet mgs $LCTL nodemap_modify --name default \
--property admin --value 1
do_facet mgs $LCTL nodemap_modify --name default \
--property trusted --value 1
cancel_lru_locks mdc
wait_nm_sync default admin_nodemap
wait_nm_sync default trusted_nodemap
if [ $(mv $tmpdir $DIR/$tdir/ 2>&1 |
grep -ci "Operation not permitted") -ne 0 ]; then
error "mv $tmpdir $DIR/$tdir/ hits 'Operation not permitted'"
fi
}
run_test 432 "mv dir from outside Lustre"
test_433() {
[ $PARALLEL == "yes" ] && skip "skip parallel run"
[[ -n "$($LCTL list_param llite.*.inode_cache 2>/dev/null)" ]] ||
skip "inode cache not supported"
$LCTL set_param llite.*.inode_cache=0
stack_trap "$LCTL set_param llite.*.inode_cache=1"
local count=256
local first
local before
local after
cancel_lru_locks mdc
first=$(num_objects)
test_mkdir $DIR/$tdir || error "mkdir $tdir"
createmany -m $DIR/$tdir/f $count
createmany -d $DIR/$tdir/d $count
ls -l $DIR/$tdir > /dev/null
stack_trap "rm -rf $DIR/$tdir"
before=$(num_objects)
cancel_lru_locks mdc
after=$(num_objects)
# sometimes even $before is less than 2 * count
while (( before > first + count && before - after < count )); do
sleep 1
after=$(num_objects)
wait=$((wait + 1))
(( wait % 5 == 0 )) && echo "wait $wait seconds objects: $after"
if (( wait > 60 )); then
error "inode slab grew from $first ^ $before v $after"
fi
done
echo "lustre_inode_cache grew from $first to $before, then $after after"
}
run_test 433 "ldlm lock cancel releases dentries and inodes"
test_434() {
local file
local getxattr_count
local mdc_stat_param="mdc.$FSNAME-MDT0000*.md_stats"
local p="$TMP/$TESTSUITE-$TESTNAME.parameters"
[[ $(getenforce) == "Disabled" ]] ||
skip "lsm selinux module have to be disabled for this test"
test_mkdir -i 0 -c1 $DIR/$tdir/ ||
error "fail to create $DIR/$tdir/ on MDT0000"
touch $DIR/$tdir/$tfile-{001..100}
# disable the xattr cache
save_lustre_params client "llite.*.xattr_cache" > $p
lctl set_param llite.*.xattr_cache=0
stack_trap "restore_lustre_params < $p; rm -f $p" EXIT
# clear clients mdc stats
clear_stats $mdc_stat_param ||
error "fail to clear stats on mdc MDT0000"
for file in $DIR/$tdir/$tfile-{001..100}; do
getfattr -n security.selinux $file |&
grep -q "Operation not supported" ||
error "getxattr on security.selinux should return EOPNOTSUPP"
done
getxattr_count=$(calc_stats $mdc_stat_param "getxattr")
(( getxattr_count < 100 )) ||
error "client sent $getxattr_count getxattr RPCs to the MDS"
}
run_test 434 "Client should not send RPCs for security.selinux with SElinux disabled"
test_440() {
if [[ -f $LUSTRE/scripts/bash-completion/lustre ]]; then
source $LUSTRE/scripts/bash-completion/lustre
elif [[ -f /usr/share/bash-completion/completions/lustre ]]; then
source /usr/share/bash-completion/completions/lustre
else
skip "bash completion scripts not found"
fi
local lctl_completions
local lfs_completions
lctl_completions=$(_lustre_cmds lctl)
if [[ ! $lctl_completions =~ "get_param" ]]; then
error "lctl bash completion failed"
fi
lfs_completions=$(_lustre_cmds lfs)
if [[ ! $lfs_completions =~ "setstripe" ]]; then
error "lfs bash completion failed"
fi
}
run_test 440 "bash completion for lfs, lctl"
test_442() {
local pid1
local pid2
mkdir -p $DIR/$tdir
multiop $DIR/$tdir/$tfile.1 O_w1 & pid1=$!
multiop $DIR/$tdir/$tfile.1 O_w1 & pid2=$!
sleep 1
touch $DIR/$tdir/$tfile.2
$LFS swap_layouts -n $DIR/$tdir/$tfile.1 $DIR/$tdir/$tfile.2
$LCTL set_param fail_loc=0x1430
kill -USR1 $pid1
sleep 1
kill -USR1 $pid2
wait
}
run_test 442 "truncate vs read/write should not panic"
test_460d() {
verify_yaml_available || skip_env "YAML verification not installed"
$LCTL get_param -n sptlrpc.page_pools
$LCTL get_param -n sptlrpc.page_pools | verify_yaml ||
error "The output of encrypt_page_pools is not an valid YAML"
}
run_test 460d "Check encrypt pools output"
resident_pages() {
local file=$1
vmtouch $file | awk '/Resident Pages:/ {print $3}' |
awk -F/ '{ print $1 }'
}
# The command "echo 2 > /proc/sys/vm/drop_caches" may revoke the DLM locks
# due to slab cache reclaim. Thus we should avoid to reclaim slab cache for
# DLM locks during testing since it may evict mlock()ed pages due to the
# release of the DLM extent lock.
# After the page cache shrinker is disabled, "echo 3 > /proc/sys/vm/drop_caches"
# and "echo 2 > /proc/sys/vm/drop_caches" will not scan and clear unused pages
# from the LRU list.
disable_page_cache_shrink() {
local enabled=$($LCTL get_param -n osc.*.enable_page_cache_shrink |
head -n 1)
stack_trap "$LCTL set_param osc.*.enable_page_cache_shrink=$enabled"
$LCTL set_param osc.*.enable_page_cache_shrink=0
}
enable_mlock_pages_check() {
local enabled=$($LCTL get_param -n llite.*.enable_mlock_pages)
stack_trap "$LCTL set_param llite.*.enable_mlock_pages=$enabled"
$LCTL set_param llite.*.enable_mlock_pages=1
}
test_600a() {
local file=$DIR/$tfile
local size_mb=100
local pcnt=$((size_mb * 1024 * 1024 / PAGE_SIZE))
which vmtouch || skip_env "This test needs vmtouch utility"
check_set_fallocate_or_skip
disable_page_cache_shrink
enable_mlock_pages_check
fallocate -l ${size_mb}M $file || error "failed to fallocate $file"
stack_trap "pkill -9 vmtouch || true"
vmtouch -vltdw -m 1g $file || error "failed to vmtouch $file"
local rcnt=$(resident_pages $file)
echo "before drop_caches (0):"
grep Mlocked: /proc/meminfo
$LCTL get_param llite.*.max_cached_mb
echo "drop page caches (1):"
echo 1 > /proc/sys/vm/drop_caches
grep Mlocked: /proc/meminfo
$LCTL get_param llite.*.max_cached_mb
vmtouch $file
(( $pcnt == $rcnt )) || error "resident pages are $rcnt, expected $pcnt"
local unevict_mb
$LCTL set_param llite.*.unevict_cached_mb=clear
$LCTL get_param llite.*.unevict_cached_mb
unevict_mb=$($LCTL get_param -n llite.*.unevict_cached_mb)
(( $unevict_mb == $size_mb )) ||
error "unevict_cached_mb is $unevict_mb, expected $size_mb"
$LCTL set_param $OSC.*$OSC*.osc_unevict_cached_mb=clear
$LCTL get_param $OSC.*$OSC*.osc_unevict_cached_mb
unevict_mb=$($LCTL get_param -n $OSC.*$OSC*.osc_unevict_cached_mb |
awk '{sum += $1 } END { print sum }')
(( $unevict_mb == $size_mb )) ||
error "osc_unevict_cached_mb is $unevict_mb, expected $size_mb"
# The lock revocation will evict the cached pages protected by it.
# This is desired behavior for conflict access from the remote client.
# But how to deal with the lock revocation triggered by LRU lock
# shrinking on client side, should this kind of locks that protected
# the mlocked pages be canceled in this case? Or the lock protecting
# mlock()ed pages should not put into lock LRU list?
cancel_lru_locks $OSC
echo "drop lru DLM lock:"
grep Mlocked: /proc/meminfo
$LCTL get_param llite.*.max_cached_mb
$LCTL get_param osc.*.osc_cached_mb
rcnt=$(resident_pages $file)
(( $rcnt == 0 )) || error "resident pages are $rcnt, expected zero"
unevict_mb=$($LCTL get_param -n llite.*.unevict_cached_mb)
(( $unevict_mb == 0 )) ||
error "unevict_cached_mb is $unevict_mb, expected 0"
unevict_mb=$($LCTL get_param -n $OSC.*$OSC*.osc_unevict_cached_mb |
awk '{sum += $1 } END { print sum }')
(( $unevict_mb == 0 )) ||
error "osc_unevict_cached_mb is $unevict_mb, expected $size_mb"
}
run_test 600a "basic test for mlock()ed file"
test_600b() {
local file=$DIR/$tfile
local size_mb=100
local cache_limit=64
local max_cached_mb=$($LCTL get_param llite.*.max_cached_mb |
awk '/^max_cached_mb/ { print $2 }')
which vmtouch || skip_env "This test needs vmtouch utility"
check_set_fallocate_or_skip
disable_page_cache_shrink
enable_mlock_pages_check
fallocate -l ${size_mb}M $file || error "failed to fallocate $file"
stack_trap "pkill -9 vmtouch || true"
cancel_lru_locks $OSC
$LCTL get_param llite.*.max_cached_mb
stack_trap "$LCTL set_param llite.*.max_cached_mb=$max_cached_mb"
$LCTL set_param llite.*.max_cached_mb=$cache_limit
# The required mlock()ed pages (100M) are larger than @max_cached_mb.
vmtouch -vltdw -m 1g $file || error "failed to mlock $file"
vmtouch $file
grep Mlocked: /proc/meminfo
local used_mb
local unevict_mb
echo 1 > /proc/sys/vm/drop_caches
$LCTL get_param llite.*.max_cached_mb
$LCTL set_param llite.*.unevict_cached_mb=clear
used_mb=$($LCTL get_param llite.*.max_cached_mb |
awk '/^used_mb/ { print $2 }')
unevict_mb=$($LCTL get_param -n llite.*.unevict_cached_mb)
(( $used_mb == 0 )) || error "used_mb is $used_mb, expected 0"
(( $unevict_mb >= $size_mb - 1)) || # allow a margin of 1 page
error "unevict_mb is $unevict_mb, expected $size_mb"
}
run_test 600b "mlock a file (via vmtouch) larger than max_cached_mb"
test_600c() {
local dir=$DIR/$tdir
local cache_limit=64
local max_cached_mb=$($LCTL get_param llite.*.max_cached_mb |
awk '/^max_cached_mb/ { print $2 }')
which vmtouch || skip_env "This test needs vmtouch utility"
check_set_fallocate_or_skip
disable_page_cache_shrink
enable_mlock_pages_check
stack_trap "rm -rf $dir"
stack_trap "$LCTL set_param llite.*.max_cached_mb=$max_cached_mb"
$LCTL set_param llite.*.max_cached_mb=$cache_limit
stack_trap "pkill -9 vmtouch || true"
local size=$((64 * 1048576))
local file1=$dir/$tfile.1
local file2=$dir/$tfile.2
mkdir $dir || error "failed to mkdir $dir"
fallocate -l $size $file1 || error "failed to fallocate $file1"
fallocate -l $size $file2 || error "failed to fallocate $file2"
cancel_lru_locks $OSC
vmtouch -vltdw -m 1g $file1 || error "failed to vmtouch $file1"
$LCTL get_param llite.*.max_cached_mb
$LCTL set_param llite.*.unevict_cached_mb=clear
$LCTL get_param llite.*.max_cached_mb
local cached_mb=$($LCTL get_param llite.*.max_cached_mb |
awk '/^used_mb/ { print $2 }')
(( $cached_mb == 0 )) || error "expected used_mb 0 got $cached_mb"
cached_mb=$($LCTL get_param llite.*.max_cached_mb |
awk '/^unevict_mb/ { print $2 }')
(( $cached_mb == 64 )) || error "expected unevict_mb 64 got $cached_mb"
vmtouch -vt $file2 || error "failed to vmtouch $file2"
echo 3 > /proc/sys/vm/drop_caches
dd if=$file2 of=/dev/null bs=1M count=64 ||
error "failed to reading $file2 into cache"
pkill -9 vmtouch || error "failed to kill vmtouch"
vmtouch -vt $file2 || error "failed to load $files into cache"
$LCTL get_param llite.*.max_cached_mb
echo 1 > /proc/sys/vm/drop_caches
$LCTL set_param llite.*.unevict_cached_mb=clear
cached_mb=$($LCTL get_param llite.*.max_cached_mb |
awk '/^used_mb/ { print $2 }')
(( $cached_mb == 0 )) || error "expected used_mb 0 got $cached_mb"
cached_mb=$($LCTL get_param llite.*.max_cached_mb |
awk '/^unevict_mb/ { print $2 }')
(( $cached_mb == 0 )) || error "expected unevict_mb 0 got $cached_mb"
}
run_test 600c "Test I/O when mlocked page count > @max_cached_mb"
test_600d_base() {
local mlcksz=$1
local fsz=$2
local n=$3
local dir=$DIR/$tdir
local mlckf=$dir/mlockfile
echo "mlock size: $mlcksz file size: $fsz, n: $n"
mkdir -p $dir || error "mkdir $dir failed"
fallocate -l $mlcksz $mlckf || error "failed to fallocate $mlckf"
for ((i = 0; i < $n; i++)); do
fallocate -l $fsz $dir/$tfile.$i ||
error "failed to fallocate $dir/$tfile.$i"
done
cancel_lru_locks $OSC
declare -a pids
vmtouch -vltdw -m 1G $mlckf || error "failed to mlock $mlckf"
for ((i = 0; i < $n; i++)); do
vmtouch -t -m 1g $dir/$tfile.$i &
pids[i]=$!
done
cat /proc/meminfo | grep 'Mlocked'
$LCTL get_param llite.*.max_cached_mb
echo "drop caches:"
echo 1 > /proc/sys/vm/drop_caches
$LCTL set_param llite.*.unevict_cached_mb=clear
$LCTL get_param llite.*.max_cached_mb
for ((i = 0; i < $n; i++)); do
wait ${pids[i]} || error "touch $dir/$tfile.$i failed: rc = $?"
done
cat /proc/meminfo | grep 'Mlocked:'
pkill -9 vmtouch || true
rm -rvf $dir || error "failed to rm $dir"
}
test_600d() {
local dir=$DIR/$tdir
local cache_limit=64
local max_cached_mb=$($LCTL get_param llite.*.max_cached_mb |
awk '/^max_cached_mb/ { print $2 }')
which vmtouch || skip_env "This test needs vmtouch utility"
check_set_fallocate_or_skip
disable_page_cache_shrink
enable_mlock_pages_check
stack_trap "rm -rf $dir"
stack_trap "$LCTL set_param llite.*.max_cached_mb=$max_cached_mb"
$LCTL set_param llite.*.max_cached_mb=$cache_limit
stack_trap "pkill -9 vmtouch || true"
local size=$((cache_limit * 1048576))
test_600d_base $((size - PAGE_SIZE)) 4096 16
test_600d_base $((size - 2 * PAGE_SIZE)) 16384 16
}
run_test 600d "Test I/O with limited LRU page slots (some was mlocked)"
prep_801() {
[[ $MDS1_VERSION -lt $(version_code 2.9.55) ]] ||
[[ $OST1_VERSION -lt $(version_code 2.9.55) ]] &&
skip "Need server version at least 2.9.55"
start_full_debug_logging
}
post_801() {
stop_full_debug_logging
}
barrier_stat() {
if (( $MDS1_VERSION >= $(version_code 2.16.50) )); then
barrier_stat="barrier stat"
else
barrier_stat="barrier_stat"
fi
if [ $MGS_VERSION -le $(version_code 2.10.0) ]; then
local st=$(do_facet mgs $LCTL $barrier_stat $FSNAME |
awk '/The barrier for/ { print $7 }')
echo $st
else
local st=$(do_facet mgs $LCTL $barrier_stat -s $FSNAME)
echo \'$st\'
fi
}
barrier_expired() {
local expired
if [ $MGS_VERSION -le $(version_code 2.10.0) ]; then
expired=$(do_facet mgs $LCTL barrier_stat $FSNAME |
awk '/will be expired/ { print $7 }')
else
expired=$(do_facet mgs $LCTL barrier_stat -t $FSNAME)
fi
echo $expired
}
test_801a() {
prep_801
echo "Start barrier_freeze at: $(date)"
#define OBD_FAIL_BARRIER_DELAY 0x2202
do_facet mgs $LCTL set_param fail_val=5 fail_loc=0x2202
# Do not reduce barrier time - See LU-11873
do_facet mgs $LCTL barrier_freeze $FSNAME 20 &
sleep 2
local b_status=$(barrier_stat)
echo "Got barrier status at: $(date)"
[ "$b_status" = "'freezing_p1'" ] ||
error "(1) unexpected barrier status $b_status"
do_facet mgs $LCTL set_param fail_val=0 fail_loc=0
wait
b_status=$(barrier_stat)
[ "$b_status" = "'frozen'" ] ||
error "(2) unexpected barrier status $b_status"
local expired=$(barrier_expired)
echo "sleep $((expired + 3)) seconds, then the barrier will be expired"
sleep $((expired + 3))
b_status=$(barrier_stat)
[ "$b_status" = "'expired'" ] ||
error "(3) unexpected barrier status $b_status"
# Do not reduce barrier time - See LU-11873
do_facet mgs $LCTL barrier_freeze $FSNAME 20 ||
error "(4) fail to freeze barrier"
b_status=$(barrier_stat)
[ "$b_status" = "'frozen'" ] ||
error "(5) unexpected barrier status $b_status"
echo "Start barrier_thaw at: $(date)"
#define OBD_FAIL_BARRIER_DELAY 0x2202
do_facet mgs $LCTL set_param fail_val=5 fail_loc=0x2202
do_facet mgs $LCTL barrier_thaw $FSNAME &
sleep 2
b_status=$(barrier_stat)
echo "Got barrier status at: $(date)"
[ "$b_status" = "'thawing'" ] ||
error "(6) unexpected barrier status $b_status"
do_facet mgs $LCTL set_param fail_val=0 fail_loc=0
wait
b_status=$(barrier_stat)
[ "$b_status" = "'thawed'" ] ||
error "(7) unexpected barrier status $b_status"
#define OBD_FAIL_BARRIER_FAILURE 0x2203
do_facet $SINGLEMDS $LCTL set_param fail_loc=0x2203
do_facet mgs $LCTL barrier_freeze $FSNAME
b_status=$(barrier_stat)
[ "$b_status" = "'failed'" ] ||
error "(8) unexpected barrier status $b_status"
do_facet $SINGLEMDS $LCTL set_param fail_loc=0
do_facet mgs $LCTL barrier_thaw $FSNAME
post_801
}
run_test 801a "write barrier user interfaces and stat machine"
test_801b() {
prep_801
mkdir $DIR/$tdir || error "(1) fail to mkdir"
createmany -d $DIR/$tdir/d 6 || error "(2) fail to mkdir"
touch $DIR/$tdir/d2/f10 || error "(3) fail to touch"
touch $DIR/$tdir/d3/f11 || error "(4) fail to touch"
touch $DIR/$tdir/d4/f12 || error "(5) fail to touch"
cancel_lru_locks mdc
# 180 seconds should be long enough
do_facet mgs $LCTL barrier_freeze $FSNAME 180
local b_status=$(barrier_stat)
[ "$b_status" = "'frozen'" ] ||
error "(6) unexpected barrier status $b_status"
mkdir $DIR/$tdir/d0/d10 &
mkdir_pid=$!
touch $DIR/$tdir/d1/f13 &
touch_pid=$!
ln $DIR/$tdir/d2/f10 $DIR/$tdir/d2/f14 &
ln_pid=$!
mv $DIR/$tdir/d3/f11 $DIR/$tdir/d3/f15 &
mv_pid=$!
rm -f $DIR/$tdir/d4/f12 &
rm_pid=$!
stat $DIR/$tdir/d5 || error "(7) stat should succeed"
# To guarantee taht the 'stat' is not blocked
b_status=$(barrier_stat)
[ "$b_status" = "'frozen'" ] ||
error "(8) unexpected barrier status $b_status"
# let above commands to run at background
sleep 5
ps -p $mkdir_pid || error "(9) mkdir should be blocked"
ps -p $touch_pid || error "(10) touch should be blocked"
ps -p $ln_pid || error "(11) link should be blocked"
ps -p $mv_pid || error "(12) rename should be blocked"
ps -p $rm_pid || error "(13) unlink should be blocked"
b_status=$(barrier_stat)
[ "$b_status" = "'frozen'" ] ||
error "(14) unexpected barrier status $b_status"
do_facet mgs $LCTL barrier_thaw $FSNAME
b_status=$(barrier_stat)
[ "$b_status" = "'thawed'" ] ||
error "(15) unexpected barrier status $b_status"
wait $mkdir_pid || error "(16) mkdir should succeed"
wait $touch_pid || error "(17) touch should succeed"
wait $ln_pid || error "(18) link should succeed"
wait $mv_pid || error "(19) rename should succeed"
wait $rm_pid || error "(20) unlink should succeed"
post_801
}
run_test 801b "modification will be blocked by write barrier"
test_801c() {
[[ $MDSCOUNT -lt 2 ]] && skip_env "needs >= 2 MDTs"
if (( $MDS1_VERSION >= $(version_code 2.16.50) )); then
barrier_freeze="barrier freeze"
barrier_thaw="barrier thaw"
barrier_rescan="barrier rescan"
else
barrier_freeze="barrier_freeze"
barrier_thaw="barrier_thaw"
barrier_rescan="barrier_rescan"
fi
prep_801
stop mds2 || error "(1) Fail to stop mds2"
do_facet mgs $LCTL $barrier_freeze $FSNAME 30
local b_status=$(barrier_stat)
[ "$b_status" = "'expired'" ] || [ "$b_status" = "'failed'" ] || {
do_facet mgs $LCTL $barrier_thaw $FSNAME
error "(2) unexpected barrier status $b_status"
}
do_facet mgs $LCTL $barrier_rescan $FSNAME ||
error "(3) Fail to rescan barrier bitmap"
# Do not reduce barrier time - See LU-11873
do_facet mgs $LCTL $barrier_freeze $FSNAME 20
b_status=$(barrier_stat)
[ "$b_status" = "'frozen'" ] ||
error "(4) unexpected barrier status $b_status"
do_facet mgs $LCTL $barrier_thaw $FSNAME
b_status=$(barrier_stat)
[ "$b_status" = "'thawed'" ] ||
error "(5) unexpected barrier status $b_status"
local devname=$(mdsdevname 2)
start mds2 $devname $MDS_MOUNT_OPTS || error "(6) Fail to start mds2"
do_facet mgs $LCTL $barrier_rescan $FSNAME ||
error "(7) Fail to rescan barrier bitmap"
post_801
}
run_test 801c "rescan barrier bitmap"
test_802b() {
[ $PARALLEL == "yes" ] && skip "skip parallel run"
remote_mds_nodsh && skip "remote MDS with nodsh"
do_facet $SINGLEMDS $LCTL get_param mdt.*.readonly ||
skip "readonly option not available"
$LFS mkdir -i 0 -c 1 $DIR/$tdir || error "(1) fail to mkdir"
cp $LUSTRE/tests/test-framework.sh $DIR/$tdir/ ||
error "(2) Fail to copy"
# write back all cached data before setting MDT to readonly
cancel_lru_locks
sync_all_data
do_facet $SINGLEMDS $LCTL set_param mdt.*.readonly=1
stack_trap "do_facet $SINGLEMDS $LCTL set_param mdt.*.readonly=0" EXIT
echo "Modify should be refused"
touch $DIR/$tdir/guard && error "(6) Touch should fail under ro mode"
echo "Read should be allowed"
diff $LUSTRE/tests/test-framework.sh $DIR/$tdir/test-framework.sh ||
error "(7) Read should succeed under ro mode"
# disable readonly
do_facet $SINGLEMDS $LCTL set_param mdt.*.readonly=0
}
run_test 802b "be able to set MDTs to readonly"
test_802c() {
[ $PARALLEL == "yes" ] && skip "skip parallel run"
do_facet ost1 $LCTL get_param obdfilter.*.readonly ||
skip "readonly option not available"
$LFS mkdir -i 0 -c 1 $DIR/$tdir || error "(1) fail to mkdir"
cp $LUSTRE/tests/test-framework.sh $DIR/$tdir/ ||
error "(2) Fail to copy"
# write back all cached data before setting OFD to readonly
cancel_lru_locks
sync_all_data
do_facet ost1 $LCTL set_param obdfilter.*.readonly=1
stack_trap "do_facet ost1 $LCTL set_param obdfilter.*.readonly=0" EXIT
echo "Modify should be refused"
touch $DIR/$tdir/guard && error "(6) Touch should fail under ro mode"
echo "Read should be allowed"
diff $LUSTRE/tests/test-framework.sh $DIR/$tdir/test-framework.sh ||
error "(7) Read should succeed under ro mode"
}
run_test 802c "be able to set OFDs to readonly"
test_803a() {
[[ $MDSCOUNT -lt 2 ]] && skip_env "needs >= 2 MDTs"
[ $MDS1_VERSION -lt $(version_code 2.10.54) ] &&
skip "MDS needs to be newer than 2.10.54"
mkdir_on_mdt0 $DIR/$tdir
# Create some objects on all MDTs to trigger related logs objects
for idx in $(seq $MDSCOUNT); do
$LFS mkdir -c $MDSCOUNT -i $((idx % $MDSCOUNT)) \
$DIR/$tdir/dir${idx} ||
error "Fail to create $DIR/$tdir/dir${idx}"
done
wait_delete_completed # ensure old test cleanups are finished
sleep 3
echo "before create:"
$LFS df -i $MOUNT
local before_used=$($LFS df --output=iused --mdt=0)
for i in {1..10}; do
$LFS mkdir -c 1 -i 1 $DIR/$tdir/foo$i ||
error "Fail to create $DIR/$tdir/foo$i"
done
# sync ZFS-on-MDS to refresh statfs data
wait_zfs_commit mds1
sleep 3
echo "after create:"
$LFS df -i $MOUNT
local after_used=$($LFS df --output=iused --mdt=0)
# allow for an llog to be cleaned up during the test
[ $after_used -ge $((before_used + 10 - 1)) ] ||
error "before ($before_used) + 10 > after ($after_used)"
for i in {1..10}; do
rm -rf $DIR/$tdir/foo$i ||
error "Fail to remove $DIR/$tdir/foo$i"
done
# sync ZFS-on-MDS to refresh statfs data
wait_zfs_commit mds1
wait_delete_completed
sleep 3 # avoid MDT return cached statfs
echo "after unlink:"
$LFS df -i $MOUNT
after_used=$($LFS df --output=iused --mdt=0)
# allow for an llog to be created during the test
[ $after_used -le $((before_used + 1)) ] ||
error "after ($after_used) > before ($before_used) + 1"
}
run_test 803a "verify agent object for remote object"
test_803b() {
[[ $MDSCOUNT -lt 2 ]] && skip_env "needs >= 2 MDTs"
[ $MDS1_VERSION -lt $(version_code 2.13.56) ] &&
skip "MDS needs to be newer than 2.13.56"
[ $PARALLEL == "yes" ] && skip "skip parallel run"
for i in $(seq 0 $((MDSCOUNT - 1))); do
$LFS mkdir -i $i $DIR/$tdir.$i || error "mkdir $tdir.$i"
done
local before=0
local after=0
local tmp
stat $DIR/$tdir.* >/dev/null || error "stat $tdir.*"
for i in $(seq 0 $((MDSCOUNT - 1))); do
tmp=$(do_facet mds$i $LCTL get_param mdt.*-MDT000$i.md_stats |
awk '/getattr/ { print $2 }')
before=$((before + tmp))
done
stat $DIR/$tdir.* >/dev/null || error "stat $tdir.*"
for i in $(seq 0 $((MDSCOUNT - 1))); do
tmp=$(do_facet mds$i $LCTL get_param mdt.*-MDT000$i.md_stats |
awk '/getattr/ { print $2 }')
after=$((after + tmp))
done
[ $before -eq $after ] || error "getattr count $before != $after"
}
run_test 803b "remote object can getattr from cache"
test_804() {
[[ $MDSCOUNT -lt 2 ]] && skip_env "needs >= 2 MDTs"
[ $MDS1_VERSION -lt $(version_code 2.10.54) ] &&
skip "MDS needs to be newer than 2.10.54"
[ "$mds1_FSTYPE" != "ldiskfs" ] && skip_env "ldiskfs only test"
mkdir -p $DIR/$tdir
$LFS mkdir -c 1 -i 1 $DIR/$tdir/dir0 ||
error "Fail to create $DIR/$tdir/dir0"
local fid=$($LFS path2fid $DIR/$tdir/dir0)
local dev=$(mdsdevname 2)
do_facet mds2 "$DEBUGFS -c -R 'ls /REMOTE_PARENT_DIR' $dev" |
grep ${fid} || error "NOT found agent entry for dir0"
$LFS mkdir -c $MDSCOUNT -i 0 $DIR/$tdir/dir1 ||
error "Fail to create $DIR/$tdir/dir1"
touch $DIR/$tdir/dir1/foo0 ||
error "Fail to create $DIR/$tdir/dir1/foo0"
fid=$($LFS path2fid $DIR/$tdir/dir1/foo0)
local rc=0
for idx in $(seq $MDSCOUNT); do
dev=$(mdsdevname $idx)
do_facet mds${idx} \
"$DEBUGFS -c -R 'ls /REMOTE_PARENT_DIR' $dev" |
grep ${fid} && rc=$idx
done
mv $DIR/$tdir/dir1/foo0 $DIR/$tdir/dir1/foo1 ||
error "Fail to rename foo0 to foo1"
if [ $rc -eq 0 ]; then
for idx in $(seq $MDSCOUNT); do
dev=$(mdsdevname $idx)
do_facet mds${idx} \
"$DEBUGFS -c -R 'ls /REMOTE_PARENT_DIR' $dev" |
grep ${fid} && rc=$idx
done
fi
mv $DIR/$tdir/dir1/foo1 $DIR/$tdir/dir1/foo2 ||
error "Fail to rename foo1 to foo2"
if [ $rc -eq 0 ]; then
for idx in $(seq $MDSCOUNT); do
dev=$(mdsdevname $idx)
do_facet mds${idx} \
"$DEBUGFS -c -R 'ls /REMOTE_PARENT_DIR' $dev" |
grep ${fid} && rc=$idx
done
fi
[ $rc -ne 0 ] || error "NOT found agent entry for foo"
ln $DIR/$tdir/dir1/foo2 $DIR/$tdir/dir0/guard ||
error "Fail to link to $DIR/$tdir/dir1/foo2"
mv $DIR/$tdir/dir1/foo2 $DIR/$tdir/dir1/foo0 ||
error "Fail to rename foo2 to foo0"
unlink $DIR/$tdir/dir1/foo0 ||
error "Fail to unlink $DIR/$tdir/dir1/foo0"
rm -rf $DIR/$tdir/dir0 ||
error "Fail to rm $DIR/$tdir/dir0"
for idx in $(seq $MDSCOUNT); do
rc=0
stop mds${idx}
dev=$(mdsdevname $idx)
run_e2fsck $(facet_active_host mds$idx) $dev -n ||
rc=$?
start mds${idx} $dev $MDS_MOUNT_OPTS ||
error "mount mds$idx failed"
df $MOUNT > /dev/null 2>&1
# e2fsck should not return error
[ $rc -eq 0 ] ||
error "e2fsck detected error on MDT${idx}: rc=$rc"
done
}
run_test 804 "verify agent entry for remote entry"
cleanup_805() {
do_facet $SINGLEMDS zfs set quota=$old $fsset
unlinkmany $DIR/$tdir/f- 1000000
trap 0
}
test_805() {
local zfs_version=$(do_facet mds1 cat /sys/module/zfs/version)
[ "$mds1_FSTYPE" != "zfs" ] && skip "ZFS specific test"
[ $(version_code $zfs_version) -lt $(version_code 0.7.2) ] &&
skip "netfree not implemented before 0.7"
[[ $MDS1_VERSION -ge $(version_code 2.10.57) ]] ||
skip "Need MDS version at least 2.10.57"
local fsset
local freekb
local usedkb
local old
local quota
local pref="osd-zfs.$FSNAME-MDT0000."
# limit available space on MDS dataset to meet nospace issue
# quickly. then ZFS 0.7.2 can use reserved space if asked
# properly (using netfree flag in osd_declare_destroy()
fsset=$(do_facet $SINGLEMDS lctl get_param -n $pref.mntdev)
old=$(do_facet $SINGLEMDS zfs get -H quota $fsset | \
gawk '{print $3}')
freekb=$(do_facet $SINGLEMDS lctl get_param -n $pref.kbytesfree)
usedkb=$(do_facet $SINGLEMDS lctl get_param -n $pref.kbytestotal)
let "usedkb=usedkb-freekb"
let "freekb=freekb/2"
if let "freekb > 5000"; then
let "freekb=5000"
fi
do_facet $SINGLEMDS zfs set quota=$(((usedkb+freekb)*1024)) $fsset
trap cleanup_805 EXIT
mkdir_on_mdt0 $DIR/$tdir
$LFS setstripe -E 1M -c2 -E 4M -c2 -E -1 -c2 $DIR/$tdir ||
error "Can't set PFL layout"
createmany -m $DIR/$tdir/f- 1000000 && error "ENOSPC wasn't met"
rm -rf $DIR/$tdir || error "not able to remove"
do_facet $SINGLEMDS zfs set quota=$old $fsset
trap 0
}
run_test 805 "ZFS can remove from full fs"
# Size-on-MDS test
check_lsom_data()
{
local file=$1
local expect=$(stat -c %s $file)
local msg=$2
local hybrid=$($LCTL get_param -n llite.*.hybrid_io)
check_lsom_size $1 $expect $msg
local blocks=$($LFS getsom -b $file)
expect=$(stat -c %b $file)
# ZFS does not allocate blocks immediately after writes, so the block
# size may be incorrect on ZFS with hybrid or direct IO
if [[ $hybrid != "1" && "$ost1_FSTYPE" != "zfs" ]]; then
local blocks=$($LFS getsom -b $file)
expect=$(stat -c %b $file)
[[ $blocks == $expect ]] ||
error "$msg $file expected blocks: $expect, got: $blocks"
elif [[ $hybrid == "1" && "$ost1_FSTYPE" == "zfs" ]]; then
if [[ $blocks != $expect ]]; then
echo "HIO + ZFS, block mismatch is expected"
echo "$msg $file expected blocks: $expect, got: $blocks"
fi
fi
}
check_lsom_size()
{
local size
local expect=$2
local msg=$3
cancel_lru_locks mdc
size=$($LFS getsom -s $1)
[[ $size == $expect ]] ||
error "$msg $file expected size: $expect, got: $size"
}
test_806() {
[ $MDS1_VERSION -lt $(version_code 2.11.52) ] &&
skip "Need MDS version at least 2.11.52"
local bs=1048576
$LFS setstripe -c-1 $DIR/$tfile || error "setstripe $tfile failed"
disable_opencache
stack_trap "restore_opencache"
# single-threaded write
echo "Test SOM for single-threaded write"
dd if=/dev/zero of=$DIR/$tfile bs=$bs count=1 ||
error "write $tfile failed"
check_lsom_size $DIR/$tfile $bs "(0)"
# Test SOM with DIO write (dd truncates to 0)
dd if=/dev/zero of=$DIR/$tfile bs=$bs count=1 oflag=direct ||
error "write $tfile failed"
check_lsom_size $DIR/$tfile $bs "(1)"
local num=32
local size=$(($num * $bs))
local offset=0
local i
echo "Test SOM for single client multi-threaded($num) write"
$TRUNCATE $DIR/$tfile 0
for ((i = 0; i < $num; i++)); do
$MULTIOP $DIR/$tfile Oz${offset}w${bs}c &
local pids[$i]=$!
offset=$((offset + $bs))
done
for (( i=0; i < $num; i++ )); do
wait ${pids[$i]}
done
check_lsom_size $DIR/$tfile $size "(2)"
$TRUNCATE $DIR/$tfile 0
for ((i = 0; i < $num; i++)); do
offset=$((offset - $bs))
$MULTIOP $DIR/$tfile Oz${offset}w${bs}c &
local pids[$i]=$!
done
for (( i=0; i < $num; i++ )); do
wait ${pids[$i]}
done
check_lsom_size $DIR/$tfile $size "(3)"
# multi-client writes
num=$(get_node_count ${CLIENTS//,/ })
size=$(($num * $bs))
offset=0
i=0
echo "Test SOM for multi-client ($num) writes"
$TRUNCATE $DIR/$tfile 0
for client in ${CLIENTS//,/ }; do
do_node $client $MULTIOP $DIR/$tfile Oz${offset}w${bs}c &
local pids[$i]=$!
i=$((i + 1))
offset=$((offset + $bs))
done
for (( i=0; i < $num; i++ )); do
wait ${pids[$i]}
done
check_lsom_size $DIR/$tfile $offset "(4)"
i=0
$TRUNCATE $DIR/$tfile 0
for client in ${CLIENTS//,/ }; do
offset=$((offset - $bs))
do_node $client $MULTIOP $DIR/$tfile Oz${offset}w${bs}c &
local pids[$i]=$!
i=$((i + 1))
done
for (( i=0; i < $num; i++ )); do
wait ${pids[$i]}
done
check_lsom_size $DIR/$tfile $size "(5)"
# verify SOM blocks count
echo "Verify SOM block count"
$TRUNCATE $DIR/$tfile 0
$MULTIOP $DIR/$tfile oO_TRUNC:O_RDWR:w$((bs))YSc ||
error "failed to write file $tfile with fdatasync and fstat"
check_lsom_data $DIR/$tfile "(6)"
$TRUNCATE $DIR/$tfile 0
$MULTIOP $DIR/$tfile oO_TRUNC:O_RDWR:w$((bs * 2))Yc ||
error "failed to write file $tfile with fdatasync"
check_lsom_data $DIR/$tfile "(7)"
$TRUNCATE $DIR/$tfile 0
$MULTIOP $DIR/$tfile oO_TRUNC:O_RDWR:O_SYNC:w$((bs * 3))c ||
error "failed to write file $tfile with sync IO"
check_lsom_data $DIR/$tfile "(8)"
# verify truncate
echo "Test SOM for truncate"
# use ftruncate to sync blocks on close request
$MULTIOP $DIR/$tfile oO_WRONLY:T16384c
check_lsom_size $DIR/$tfile 16384 "(9)"
check_lsom_data $DIR/$tfile "(10)"
$TRUNCATE $DIR/$tfile 1234
check_lsom_size $DIR/$tfile 1234 "(11)"
# sync blocks on the MDT
$MULTIOP $DIR/$tfile oc
check_lsom_data $DIR/$tfile "(12)"
}
run_test 806 "Verify Lazy Size on MDS"
verify_som_sync_utility() {
local utility=$1
local use_llsom_sync=false
[[ -z "$FILESET" ]] || skip "Not functional for FILESET set"
[[ $MDS1_VERSION -ge $(version_code 2.11.52) ]] ||
skip "Need MDS version at least 2.11.52"
case $utility in
llsom_sync ) use_llsom_sync=true ;;
somsync ) use_llsom_sync=false ;;
*) error "invalid utility $utility" ;;
esac
if $use_llsom_sync; then
# Registration step
changelog_register || error "changelog_register failed"
local cl_user="${CL_USERS[$SINGLEMDS]%% *}"
changelog_users $SINGLEMDS | grep -q $cl_user ||
error "User $cl_user not found in changelog_users"
fi
rm -rf $DIR/$tdir || error "rm $tdir failed"
mkdir_on_mdt0 $DIR/$tdir || error "mkdir $tdir failed"
touch $DIR/$tdir/trunc || error "touch $tdir/trunc failed"
$TRUNCATE $DIR/$tdir/trunc 1024 || error "truncate $tdir/trunc failed"
$TRUNCATE $DIR/$tdir/trunc 1048576 ||
error "truncate $tdir/trunc failed"
local bs=1048576
echo "Test SOM for single-threaded write with fsync"
dd if=/dev/zero of=$DIR/$tdir/single_dd bs=$bs count=1 ||
error "write $tfile failed"
sync;sync;sync
# multi-client wirtes
local num=$(get_node_count ${CLIENTS//,/ })
local offset=0
local -a pids
echo "Test SOM for multi-client ($num) writes"
touch $DIR/$tfile || error "touch $tfile failed"
$TRUNCATE $DIR/$tfile 0
for client in ${CLIENTS//,/ }; do
do_node $client $MULTIOP $DIR/$tfile Oz${offset}w${bs}c &
pids+=( $! )
offset=$((offset + bs))
done
wait ${pids[@]}
do_rpc_nodes "$CLIENTS" cancel_lru_locks osc
do_nodes "$CLIENTS" "sync ; sleep 5 ; sync"
if $use_llsom_sync; then
$LSOM_SYNC -u $cl_user -m $FSNAME-MDT0000 $MOUNT ||
error "$LSOM_SYNC failed"
else
$LFS somsync $DIR/$tfile ||
error "$LFS somsync $DIR/$tfile failed"
find $DIR/$tdir -type f | xargs $LFS path2fid |
awk '{print $2}' | xargs $LFS somsync --by-fid $DIR ||
error "$LFS somsync --by-fid $DIR failed"
fi
check_lsom_data $DIR/$tdir/trunc "(0)"
check_lsom_data $DIR/$tdir/single_dd "(1)"
check_lsom_data $DIR/$tfile "(2)"
rm -rf $DIR/$tdir
# Deregistration step
! $use_llsom_sync || changelog_deregister ||
error "changelog_deregister failed"
}
test_807a() {
verify_som_sync_utility llsom_sync
}
run_test 807a "verify LSOM syncing tool"
test_807b() {
verify_som_sync_utility somsync
}
run_test 807b "verify lfs somsync utility"
check_som_nologged()
{
local lines=$($LFS changelog $FSNAME-MDT0000 |
grep 'x=trusted.som' | wc -l)
[ $lines -ne 0 ] && error "trusted.som xattr is logged in Changelogs"
}
test_808() {
[ $MDS1_VERSION -lt $(version_code 2.11.55) ] &&
skip "Need MDS version at least 2.11.55"
# Registration step
changelog_register || error "changelog_register failed"
touch $DIR/$tfile || error "touch $tfile failed"
check_som_nologged
dd if=/dev/zero of=$DIR/$tfile bs=1048576 count=1 ||
error "write $tfile failed"
check_som_nologged
$TRUNCATE $DIR/$tfile 1234
check_som_nologged
$TRUNCATE $DIR/$tfile 1048576
check_som_nologged
# Deregistration step
changelog_deregister || error "changelog_deregister failed"
}
run_test 808 "Check trusted.som xattr not logged in Changelogs"
check_som_nodata()
{
$LFS getsom $1
[[ $? -eq 61 ]] || error "DoM-only file $1 has SOM xattr"
}
test_809() {
[ $MDS1_VERSION -lt $(version_code 2.11.56) ] &&
skip "Need MDS version at least 2.11.56"
$LFS setstripe -E 1M -L mdt $DIR/$tfile ||
error "failed to create DoM-only file $DIR/$tfile"
touch $DIR/$tfile || error "touch $tfile failed"
check_som_nodata $DIR/$tfile
dd if=/dev/zero of=$DIR/$tfile bs=2048 count=1 ||
error "write $tfile failed"
check_som_nodata $DIR/$tfile
$TRUNCATE $DIR/$tfile 1234
check_som_nodata $DIR/$tfile
$TRUNCATE $DIR/$tfile 4097
check_som_nodata $DIR/$file
}
run_test 809 "Verify no SOM xattr store for DoM-only files"
test_810() {
[ $PARALLEL == "yes" ] && skip "skip parallel run"
$GSS && skip_env "could not run with gss"
[[ $OST1_VERSION -gt $(version_code 2.12.58) ]] ||
skip "OST < 2.12.58 doesn't align checksum"
set_checksums 1
stack_trap "set_checksums $ORIG_CSUM" EXIT
stack_trap "set_checksum_type $ORIG_CSUM_TYPE" EXIT
local csum
local before
local after
for csum in $CKSUM_TYPES; do
#define OBD_FAIL_OSC_NO_GRANT 0x411
$LCTL set_param osc.*.checksum_type=$csum fail_loc=0x411
for i in "10240 0" "10000 0" "4000 1" "500 1"; do
eval set -- $i
dd if=/dev/urandom of=$DIR/$tfile bs=$1 count=2 seek=$2
before=$(md5sum $DIR/$tfile)
$LCTL set_param ldlm.namespaces.*osc*.lru_size=clear
after=$(md5sum $DIR/$tfile)
[ "$before" == "$after" ] ||
error "$csum: $before != $after bs=$1 seek=$2"
done
done
}
run_test 810 "partial page writes on ZFS (LU-11663)"
test_812a() {
[ $OST1_VERSION -lt $(version_code 2.12.51) ] &&
skip "OST < 2.12.51 doesn't support this fail_loc"
local old
old=$($LCTL get_param -n osc.*.idle_timeout | head -n 1)
$LCTL set_param osc.*.idle_timeout=10
stack_trap "$LCTL set_param osc.*.idle_timeout=$old" EXIT
$LFS setstripe -c 1 -i 0 $DIR/$tfile
# ensure ost1 is connected
stat $DIR/$tfile >/dev/null || error "can't stat"
wait_osc_import_state client ost1 FULL
# no locks, no reqs to let the connection idle
cancel_lru_locks osc
# delay OST_DISCONNECT on OST1 to put OSC into intermediate state
#define OBD_FAIL_OST_DISCONNECT_DELAY 0x245
do_facet ost1 "$LCTL set_param fail_loc=0x245 fail_val=8"
wait_osc_import_state client ost1 CONNECTING
do_facet ost1 "$LCTL set_param fail_loc=0 fail_val=0"
stat $DIR/$tfile >/dev/null || error "can't stat file"
}
run_test 812a "do not drop reqs generated when imp is going to idle (LU-11951)"
test_812b() { # LU-12378
[ $OST1_VERSION -lt $(version_code 2.12.51) ] &&
skip "OST < 2.12.51 doesn't support this fail_loc"
local old
old=$($LCTL get_param -n osc.*.idle_timeout | head -n 1)
$LCTL set_param osc.*.idle_timeout=10
stack_trap "$LCTL set_param osc.*.idle_timeout=$old" EXIT
$LFS setstripe -c 1 -i 0 $DIR/$tfile || error "setstripe failed"
# ensure ost1 is connected
stat $DIR/$tfile >/dev/null || error "can't stat"
wait_osc_import_state client ost1 FULL
# no locks, no reqs to let the connection idle
cancel_lru_locks osc
# delay OST_DISCONNECT on OST1 to put OSC into intermediate state
#define OBD_FAIL_OST_DISCONNECT_DELAY 0x245
do_facet ost1 "$LCTL set_param fail_loc=0x245 fail_val=8"
wait_osc_import_state client ost1 CONNECTING
do_facet ost1 "$LCTL set_param fail_loc=0 fail_val=0"
$LFS quota -u 0 $DIR/ || error "lfs quota should succeed"
wait_osc_import_state client ost1 IDLE
}
run_test 812b "do not drop no resend request for idle connect"
test_812c() {
local old
old=$($LCTL get_param -n osc.*.idle_timeout | head -n 1)
$LFS setstripe -c 1 -o 0 $DIR/$tfile
$LFS getstripe $DIR/$tfile
$LCTL set_param osc.*.idle_timeout=10
stack_trap "$LCTL set_param osc.*.idle_timeout=$old" EXIT
# ensure ost1 is connected
stat $DIR/$tfile >/dev/null || error "can't stat"
wait_osc_import_state client ost1 FULL
# no locks, no reqs to let the connection idle
cancel_lru_locks osc
#define OBD_FAIL_PTLRPC_IDLE_RACE 0x533
$LCTL set_param fail_loc=0x80000533
sleep 15
dd if=/dev/zero of=$DIR/$tfile count=1 conv=sync || error "dd failed"
}
run_test 812c "idle import vs lock enqueue race"
test_813() {
local file_heat_sav=$($LCTL get_param -n llite.*.file_heat 2>/dev/null)
[ -z "$file_heat_sav" ] && skip "no file heat support"
local readsample
local writesample
local readbyte
local writebyte
local readsample1
local writesample1
local readbyte1
local writebyte1
local period_second=$($LCTL get_param -n llite.*.heat_period_second)
local decay_pct=$($LCTL get_param -n llite.*.heat_decay_percentage)
$LCTL set_param -n llite.*.file_heat=1
echo "Turn on file heat"
echo "Period second: $period_second, Decay percentage: $decay_pct"
echo "QQQQ" > $DIR/$tfile
echo "QQQQ" > $DIR/$tfile
echo "QQQQ" > $DIR/$tfile
cat $DIR/$tfile > /dev/null
cat $DIR/$tfile > /dev/null
cat $DIR/$tfile > /dev/null
cat $DIR/$tfile > /dev/null
local out=$($LFS heat_get $DIR/$tfile)
$LFS heat_get $DIR/$tfile
readsample=$(echo "$out" | grep 'readsample' | awk '{ print $2 }')
writesample=$(echo "$out" | grep 'writesample' | awk '{ print $2 }')
readbyte=$(echo "$out" | grep 'readbyte' | awk '{ print $2 }')
writebyte=$(echo "$out" | grep 'writebyte' | awk '{ print $2 }')
[ $readsample -le 4 ] || error "read sample ($readsample) is wrong"
[ $writesample -le 3 ] || error "write sample ($writesample) is wrong"
[ $readbyte -le 20 ] || error "read bytes ($readbyte) is wrong"
[ $writebyte -le 15 ] || error "write bytes ($writebyte) is wrong"
sleep $((period_second + 3))
echo "Sleep $((period_second + 3)) seconds..."
# The recursion formula to calculate the heat of the file f is as
# follow:
# Hi+1(f) = (1-P)*Hi(f)+ P*Ci
# Where Hi is the heat value in the period between time points i*I and
# (i+1)*I; Ci is the access count in the period; the symbol P refers
# to the weight of Ci.
out=$($LFS heat_get $DIR/$tfile)
$LFS heat_get $DIR/$tfile
readsample=$(echo "$out" | grep 'readsample' | awk '{ print $2 }')
writesample=$(echo "$out" | grep 'writesample' | awk '{ print $2 }')
readbyte=$(echo "$out" | grep 'readbyte' | awk '{ print $2 }')
writebyte=$(echo "$out" | grep 'writebyte' | awk '{ print $2 }')
[ $(bc <<< "$readsample <= 4 * $decay_pct / 100") -eq 1 ] ||
error "read sample ($readsample) is wrong"
[ $(bc <<< "$writesample <= 3 * $decay_pct / 100") -eq 1 ] ||
error "write sample ($writesample) is wrong"
[ $(bc <<< "$readbyte <= 20 * $decay_pct / 100") -eq 1 ] ||
error "read bytes ($readbyte) is wrong"
[ $(bc <<< "$writebyte <= 15 * $decay_pct / 100") -eq 1 ] ||
error "write bytes ($writebyte) is wrong"
echo "QQQQ" > $DIR/$tfile
echo "QQQQ" > $DIR/$tfile
echo "QQQQ" > $DIR/$tfile
cat $DIR/$tfile > /dev/null
cat $DIR/$tfile > /dev/null
cat $DIR/$tfile > /dev/null
cat $DIR/$tfile > /dev/null
sleep $((period_second + 3))
echo "Sleep $((period_second + 3)) seconds..."
out=$($LFS heat_get $DIR/$tfile)
$LFS heat_get $DIR/$tfile
readsample1=$(echo "$out" | grep 'readsample' | awk '{ print $2 }')
writesample1=$(echo "$out" | grep 'writesample' | awk '{ print $2 }')
readbyte1=$(echo "$out" | grep 'readbyte' | awk '{ print $2 }')
writebyte1=$(echo "$out" | grep 'writebyte' | awk '{ print $2 }')
[ $(bc <<< "$readsample1 <= ($readsample * (100 - $decay_pct) + \
4 * $decay_pct) / 100") -eq 1 ] ||
error "read sample ($readsample1) is wrong"
[ $(bc <<< "$writesample1 <= ($writesample * (100 - $decay_pct) + \
3 * $decay_pct) / 100") -eq 1 ] ||
error "write sample ($writesample1) is wrong"
[ $(bc <<< "$readbyte1 <= ($readbyte * (100 - $decay_pct) + \
20 * $decay_pct) / 100") -eq 1 ] ||
error "read bytes ($readbyte1) is wrong"
[ $(bc <<< "$writebyte1 <= ($writebyte * (100 - $decay_pct) + \
15 * $decay_pct) / 100") -eq 1 ] ||
error "write bytes ($writebyte1) is wrong"
echo "Turn off file heat for the file $DIR/$tfile"
$LFS heat_set -o $DIR/$tfile
echo "QQQQ" > $DIR/$tfile
echo "QQQQ" > $DIR/$tfile
echo "QQQQ" > $DIR/$tfile
cat $DIR/$tfile > /dev/null
cat $DIR/$tfile > /dev/null
cat $DIR/$tfile > /dev/null
cat $DIR/$tfile > /dev/null
out=$($LFS heat_get $DIR/$tfile)
$LFS heat_get $DIR/$tfile
readsample=$(echo "$out" | grep 'readsample' | awk '{ print $2 }')
writesample=$(echo "$out" | grep 'writesample' | awk '{ print $2 }')
readbyte=$(echo "$out" | grep 'readbyte' | awk '{ print $2 }')
writebyte=$(echo "$out" | grep 'writebyte' | awk '{ print $2 }')
[ $readsample -eq 0 ] || error "read sample ($readsample) is wrong"
[ $writesample -eq 0 ] || error "write sample ($writesample) is wrong"
[ $readbyte -eq 0 ] || error "read bytes ($readbyte) is wrong"
[ $writebyte -eq 0 ] || error "write bytes ($writebyte) is wrong"
echo "Trun on file heat for the file $DIR/$tfile"
$LFS heat_set -O $DIR/$tfile
echo "QQQQ" > $DIR/$tfile
echo "QQQQ" > $DIR/$tfile
echo "QQQQ" > $DIR/$tfile
cat $DIR/$tfile > /dev/null
cat $DIR/$tfile > /dev/null
cat $DIR/$tfile > /dev/null
cat $DIR/$tfile > /dev/null
out=$($LFS heat_get $DIR/$tfile)
$LFS heat_get $DIR/$tfile
readsample=$(echo "$out" | grep 'readsample' | awk '{ print $2 }')
writesample=$(echo "$out" | grep 'writesample' | awk '{ print $2 }')
readbyte=$(echo "$out" | grep 'readbyte' | awk '{ print $2 }')
writebyte=$(echo "$out" | grep 'writebyte' | awk '{ print $2 }')
[ $readsample -gt 0 ] || error "read sample ($readsample) is wrong"
[ $writesample -gt 0 ] || error "write sample ($writesample) is wrong"
[ $readbyte -gt 0 ] || error "read bytes ($readbyte) is wrong"
[ $writebyte -gt 0 ] || error "write bytes ($writebyte) is wrong"
$LFS heat_set -c $DIR/$tfile
$LCTL set_param -n llite.*.file_heat=0
echo "Turn off file heat support for the Lustre filesystem"
echo "QQQQ" > $DIR/$tfile
echo "QQQQ" > $DIR/$tfile
echo "QQQQ" > $DIR/$tfile
cat $DIR/$tfile > /dev/null
cat $DIR/$tfile > /dev/null
cat $DIR/$tfile > /dev/null
cat $DIR/$tfile > /dev/null
out=$($LFS heat_get $DIR/$tfile)
$LFS heat_get $DIR/$tfile
readsample=$(echo "$out" | grep 'readsample' | awk '{ print $2 }')
writesample=$(echo "$out" | grep 'writesample' | awk '{ print $2 }')
readbyte=$(echo "$out" | grep 'readbyte' | awk '{ print $2 }')
writebyte=$(echo "$out" | grep 'writebyte' | awk '{ print $2 }')
[ $readsample -eq 0 ] || error "read sample ($readsample) is wrong"
[ $writesample -eq 0 ] || error "write sample ($writesample) is wrong"
[ $readbyte -eq 0 ] || error "read bytes ($readbyte) is wrong"
[ $writebyte -eq 0 ] || error "write bytes ($writebyte) is wrong"
$LCTL set_param -n llite.*.file_heat=$file_heat_sav
rm -f $DIR/$tfile
}
run_test 813 "File heat verfication"
test_814()
{
dd of=$DIR/$tfile seek=128 bs=1k < /dev/null
echo -n y >> $DIR/$tfile
cp --sparse=always $DIR/$tfile $DIR/${tfile}.cp || error "copy failed"
diff $DIR/$tfile $DIR/${tfile}.cp || error "files should be same"
}
run_test 814 "sparse cp works as expected (LU-12361)"
test_815()
{
writeme -b 100 $DIR/$tfile || error "write 100 bytes failed"
writeme -b 0 $DIR/$tfile || error "write 0 byte failed"
}
run_test 815 "zero byte tiny write doesn't hang (LU-12382)"
test_816() {
local ost1_imp=$(get_osc_import_name client ost1)
local imp_name=$($LCTL list_param osc.$ost1_imp | head -n1 |
cut -d'.' -f2)
local old
old=$($LCTL get_param -n osc.*.idle_timeout | head -n 1)
$LCTL set_param osc.*.idle_timeout=10
stack_trap "$LCTL set_param osc.*.idle_timeout=$old" EXIT
$LFS setstripe -c 1 -i 0 $DIR/$tfile
# ensure ost1 is connected
stat $DIR/$tfile >/dev/null || error "can't stat"
wait_osc_import_state client ost1 FULL
# no locks, no reqs to let the connection idle
cancel_lru_locks osc
lru_resize_disable osc
local before
local now
before=$($LCTL get_param -n \
ldlm.namespaces.$imp_name.lru_size)
wait_osc_import_state client ost1 IDLE
dd if=/dev/null of=$DIR/$tfile bs=1k count=1 conv=sync
now=$($LCTL get_param -n \
ldlm.namespaces.$imp_name.lru_size)
[ $before == $now ] || error "lru_size changed $before != $now"
}
run_test 816 "do not reset lru_resize on idle reconnect"
cleanup_817() {
umount $tmpdir
exportfs -u localhost:$DIR/nfsexp
rm -rf $DIR/nfsexp
}
test_817() {
systemctl restart nfs-server.service || skip "failed to restart nfsd"
mkdir -p $DIR/nfsexp
exportfs -orw,no_root_squash localhost:$DIR/nfsexp ||
error "failed to export nfs"
tmpdir=$(mktemp -d /tmp/nfs-XXXXXX)
stack_trap cleanup_817 EXIT
mount -t nfs -orw localhost:$DIR/nfsexp $tmpdir ||
error "failed to mount nfs to $tmpdir"
cp /bin/true $tmpdir
$DIR/nfsexp/true || error "failed to execute 'true' command"
}
run_test 817 "nfsd won't cache write lock for exec file"
test_818() {
test_mkdir -i0 -c1 $DIR/$tdir
$LFS setstripe -c1 -i0 $DIR/$tdir/$tfile
$LFS setstripe -c1 -i1 $DIR/$tdir/$tfile
stop $SINGLEMDS
# restore osp-syn threads
stack_trap "fail $SINGLEMDS"
# disable console ratelimit
local rl=$(do_facet mds1 $LCTL get_param -n console_ratelimit)
do_facet mds1 $LCTL set_param console_ratelimit=0
stack_trap "do_facet mds1 $LCTL set_param console_ratelimit=$rl"
#define OBD_FAIL_OSP_CANT_PROCESS_LLOG 0x2105
do_facet $SINGLEMDS lctl set_param fail_loc=0x80002105
start $SINGLEMDS $(mdsdevname ${SINGLEMDS//mds/}) $MDS_MOUNT_OPTS ||
error "start $SINGLEMDS failed"
rm -rf $DIR/$tdir
local testid=$(echo $TESTNAME | tr '_' ' ')
do_facet mds1 dmesg | tac | sed "/$testid/,$ d" |
grep "run LFSCK" || error "run LFSCK is not suggested"
}
run_test 818 "unlink with failed llog"
test_819a() {
stack_trap "rm -f $DIR/$tfile"
dd if=/dev/zero of=$DIR/$tfile bs=1M count=1
cancel_lru_locks osc
#define OBD_FAIL_OST_2BIG_NIOBUF 0x248
do_facet $SINGLEMDS lctl set_param fail_loc=0x80000248
dd if=$DIR/$tfile of=/dev/null bs=1M count=1
rm -f $TDIR/$tfile
}
run_test 819a "too big niobuf in read"
test_819b() {
stack_trap "rm -f $DIR/$tfile"
#define OBD_FAIL_OST_2BIG_NIOBUF 0x248
do_facet $SINGLEMDS lctl set_param fail_loc=0x80000248
dd if=/dev/zero of=$DIR/$tfile bs=1M count=1
cancel_lru_locks osc
sleep 1
rm -f $TDIR/$tfile
}
run_test 819b "too big niobuf in write"
function test_820_start_ost() {
sleep 5
for num in $(seq $OSTCOUNT); do
start ost$num $(ostdevname $num) $OST_MOUNT_OPTS
done
}
test_820() {
[[ $MDSCOUNT -lt 2 ]] && skip_env "needs >= 2 MDTs"
mkdir $DIR/$tdir
umount_client $MOUNT || error "umount failed"
for num in $(seq $OSTCOUNT); do
stop ost$num
done
# mount client with no active OSTs
# so that the client can't initialize max LOV EA size
# from OSC notifications
mount_client $MOUNT || error "mount failed"
# delay OST starting to keep this 0 max EA size for a while
test_820_start_ost &
# create a directory on MDS2
test_mkdir -i 1 -c1 $DIR/$tdir/mds2 ||
error "Failed to create directory"
# open intent should update default EA size
# see mdc_update_max_ea_from_body()
# notice this is the very first RPC to MDS2
out=$(cp /etc/services $DIR/$tdir/mds2 2>&1)
ret=$?
echo $out
# With SSK, this situation can lead to -EPERM being returned.
# In that case, simply retry.
if [ $ret -ne 0 ] && $SHARED_KEY; then
if echo "$out" | grep -q "not permitted"; then
cp /etc/services $DIR/$tdir/mds2
ret=$?
fi
fi
[ $ret -eq 0 ] || error "Failed to copy files to mds$n"
}
run_test 820 "update max EA from open intent"
test_823() {
local p="$TMP/$TESTSUITE-$TESTNAME.parameters"
local OST_MAX_PRECREATE=20000
(( $MDS1_VERSION >= $(version_code 2.14.56) )) ||
skip "Need MDS version at least 2.14.56"
save_lustre_params mds1 \
"osp.$FSNAME-OST*-osc-MDT0000.max_create_count" > $p
do_facet $SINGLEMDS "$LCTL set_param -n \
osp.$FSNAME-OST*MDT0000.max_create_count=0"
do_facet $SINGLEMDS "$LCTL set_param -n \
osp.$FSNAME-OST0000*MDT0000.max_create_count=$OST_MAX_PRECREATE"
stack_trap "restore_lustre_params < $p; rm $p"
do_facet $SINGLEMDS "$LCTL set_param -n \
osp.$FSNAME-OST*-osc-MDT*.create_count=100200"
local count=$(do_facet $SINGLEMDS "$LCTL get_param -n \
osp.$FSNAME-OST0000*MDT0000.create_count")
local max=$(do_facet $SINGLEMDS "$LCTL get_param -n \
osp.$FSNAME-OST0000*MDT0000.max_create_count")
local expect_count=$(((($max/2)/256) * 256))
log "setting create_count to 100200:"
log " -result- count: $count with max: $max, expecting: $expect_count"
[[ $count -eq expect_count ]] ||
error "Create count not set to max precreate."
}
run_test 823 "Setting create_count > OST_MAX_PRECREATE is lowered to maximum"
test_831() {
[[ $MDS1_VERSION -lt $(version_code 2.14.56) ]] &&
skip "Need MDS version 2.14.56"
local sync_changes=$(do_facet $SINGLEMDS \
$LCTL get_param -n osp.$FSNAME-OST0000-osc-MDT0000.sync_changes)
[ "$sync_changes" -gt 100 ] &&
skip "Sync changes $sync_changes > 100 already"
local p="$TMP/$TESTSUITE-$TESTNAME.parameters"
$LFS mkdir -i 0 $DIR/$tdir
$LFS setstripe -c 1 -i 0 $DIR/$tdir
save_lustre_params mds1 \
"osp.$FSNAME-OST*-osc-MDT0000.max_sync_changes" > $p
save_lustre_params mds1 \
"osp.$FSNAME-OST*-osc-MDT0000.max_rpcs_in_progress" >> $p
do_facet mds1 "$LCTL set_param -n \
osp.$FSNAME-OST*-osc-MDT0000.max_sync_changes=100 \
osp.$FSNAME-OST*-osc-MDT0000.max_rpcs_in_progress=128"
stack_trap "restore_lustre_params < $p" EXIT
createmany -o $DIR/$tdir/f- 1000
unlinkmany $DIR/$tdir/f- 1000 &
local UNLINK_PID=$!
while sleep 1; do
sync_changes=$(do_facet mds1 \
$LCTL get_param -n osp.$FSNAME-OST0000-osc-MDT0000.sync_changes)
# the check in the code is racy, fail the test
# if the value above the limit by 10.
[ $sync_changes -gt 110 ] && {
kill -2 $UNLINK_PID
wait
error "osp changes throttling failed, $sync_changes>110"
}
kill -0 $UNLINK_PID 2> /dev/null || break
done
wait
}
run_test 831 "throttling unlink/setattr queuing on OSP"
test_832() {
(( $MDSCOUNT >= 2 )) || skip "needs >= 2 MDTs"
(( $MDS1_VERSION >= $(version_code 2.15.52) )) ||
skip "Need MDS version 2.15.52+"
is_rmentry_supported || skip "rm_entry not supported"
mkdir_on_mdt0 $DIR/$tdir || error "mkdir $tdir failed"
mkdir $DIR/$tdir/local_dir || error "mkdir local_dir failed"
mkdir_on_mdt -i 1 $DIR/$tdir/remote_dir ||
error "mkdir remote_dir failed"
$LFS mkdir -c $MDSCOUNT $DIR/$tdir/striped_dir ||
error "mkdir striped_dir failed"
touch $DIR/$tdir/file || error "touch file failed"
$LFS rm_entry $DIR/$tdir/* || error "lfs rm_entry $tdir/* failed"
[ -z "$(ls -A $DIR/$tdir)" ] || error "$tdir not empty"
}
run_test 832 "lfs rm_entry"
test_833() {
local file=$DIR/$tfile
stack_trap "rm -f $file" EXIT
dd if=/dev/zero of=$file bs=1M count=50 || error "Write $file failed"
local wpid
local rpid
local rpid2
# Buffered I/O write
(
while [ ! -e $DIR/sanity.833.lck ]; do
dd if=/dev/zero of=$file bs=1M count=50 conv=notrunc ||
error "failed to write $file"
sleep 0.$((RANDOM % 4 + 1))
done
)&
wpid=$!
# Buffered I/O read
(
while [ ! -e $DIR/sanity.833.lck ]; do
dd if=$file of=/dev/null bs=1M count=50 ||
error "failed to read $file"
sleep 0.$((RANDOM % 4 + 1))
done
)&
rpid=$!
# Direct I/O read
(
while [ ! -e $DIR/sanity.833.lck ]; do
dd if=$file of=/dev/null bs=1M count=50 iflag=direct ||
error "failed to read $file in direct I/O mode"
sleep 0.$((RANDOM % 4 + 1))
done
)&
rpid2=$!
sleep 30
touch $DIR/sanity.833.lck
wait $wpid || error "$?: buffered write failed"
wait $rpid || error "$?: buffered read failed"
wait $rpid2 || error "$?: direct read failed"
}
run_test 833 "Mixed buffered/direct read and write should not return -EIO"
test_834() {
local file=$DIR/$tfile
dd if=/dev/zero of=$file bs=1M count=100 ||
error "failed to write $file"
cancel_lru_locks $OSC
$LCTL set_param llite.*.read_ahead_stats=clear
# Unfortunately we can not reproduce the panic with the following mmap
# read with madvise(MADV_HUGEPAGE)...
$MULTIOP $file OSMJRUc || "failed to $MULTIOP $file"
$LCTL get_param llite.*.read_ahead_stats
which fio || skip_env "no fio installed"
fio --enghelp=mmap | grep 'thp' ||
skip_env "fio mmap I/O engine does not support THP"
$TRUNCATE $file $((5 * 1024 * 1024 * 1024)) ||
error "failed to truncate $file"
# "--thp=1" indicates to use MADV_HUGEPAGE madvise for fio testing
# with mmap I/O engine.
# However, this fio test still can not reproduce the panic also...
fio --name=read_test --ioengine=mmap --filename=$file --rw=read \
--bs=1m --thp=1 || error "failed to run fio on $file"
}
run_test 834 "mmap readahead for madvise with MADV_HUGEPAGE"
test_842() {
(( $MDS1_VERSION >= $(version_code 2.15.62) )) ||
skip "Need MDS version at least 2.15.62 for ldlm_extent module"
local oss1=$(facet_host ost1)
# Try to insert the module. This will leave results in dmesg
now=$(date +%s)
log "STAMP $now" > /dev/kmsg
do_rpc_nodes $oss1 load_module kunit/ldlm_extent ||
error "$oss1 load_module ldlm_extent failed"
do_node $oss1 dmesg | sed -n -e "1,/STAMP $now/d" -e '/ldlm_extent:/p'
do_node $oss1 rmmod -v ldlm_extent ||
error "rmmod failed (may trigger a failure in a later test)"
}
run_test 842 "Measure ldlm_extent performance"
test_850() {
local dir=$DIR/$tdir
local file=$dir/$tfile
local statsfile=$dir/all_job_stats.txt
test_mkdir -p $dir || error "failed to create dir $dir"
echo "abcdefg" > $file || error "failed to create file $file"
# read job_stats in the living system
lljobstat -n 1 ||
error "failed to run lljobstat on living system"
$LCTL get_param *.*.job_stats > $statsfile
lljobstat --statsfile=$statsfile ||
error "failed to run lljobstat on file $statsfile"
}
run_test 850 "lljobstat can parse living and aggregated job_stats"
test_851() {
local dir=$DIR/$tdir
local file=$dir/f_test_851_$$
local report=/tmp/report_test_851_$$
local fanotify_prog=monitor_lustrefs
local pid
local i
test_mkdir $dir || error "failed to create dir $dir"
nice -n -10 $fanotify_prog $DIR > $report & pid=$!
kill -0 $pid || error "failed to run $fanotify_prog"
ps -q $pid -eo pid,ni,comm
stack_trap "kill $pid"
stack_trap "rm -f $report"
for i in {1..30}; do
if grep Started $report; then
break
else
echo "$fanotify_prog did not start working, wait 1 second for $i times..."
sleep 1
fi
done
((i == 30)) && error "$fanotify_prog did not start working in 30 seconds"
> $report
echo "1234567890" > $file
wait_update_cond $(facet_active_host client) "stat -c %s $report" "-gt" "0" 30 ||
error "fanotify did not report anything after 30 seconds when file is created"
grep -a -E "open.*:$file:" $report ||
error "no open event for writing $file"
grep -a -E "write.*:$file:" $report ||
error "no write event for writing $file"
grep -a -E "close.*:$file:" $report ||
error "no close event for writing $file"
> $report
cat $file
wait_update_cond $(facet_active_host client) "stat -c %s $report" "-gt" "0" 30 ||
error "fanotify did not report anything after 30 seconds when file is read"
grep -a -E "open.*:$file:" $report ||
error "no open event for reading $file"
grep -a -E "read.*:$file:" $report ||
error "no write event for reading $file"
grep -a -E "close.*:$file:" $report ||
error "no close event for reading $file"
}
run_test 851 "fanotify can monitor open/read/write/close events for lustre fs"
test_852() {
(( $MDSCOUNT >= 2 )) || skip "needs >= 2 MDTs"
(( $MDS1_VERSION >= $(version_code v2_15_62-70-g668dfb53de) )) ||
skip "Need MDS version at least 2.15.62.70 for intent mkdir"
local save="$TMP/$TESTSUITE-$TESTNAME.parameters"
save_lustre_params client "llite.*.intent_mkdir" > $save
stack_trap "restore_lustre_params < $save; rm -f $save" EXIT
$LCTL set_param llite.*.intent_mkdir=1
test_mkdir -p -c$MDSCOUNT $DIR/$tdir
if [ $MDSCOUNT -ge 2 ]; then
$LFS setdirstripe -D -c$MDSCOUNT $DIR/$tdir ||
error "set default dirstripe failed"
fi
mkdir $DIR/$tdir/tdir || error "mkdir tdir failed"
mkdir $DIR/$tdir/tdir/tfile || error "mkdir tdir/tfile failed"
touch -d "2020-08-25 15:08" $DIR/$tdir/tdir/tfile ||
error "touch time failed"
chown 0:0 $DIR/$tdir/tdir/tfile || error "chown 0:0 tdir/tfile failed"
chmod 755 $DIR/$tdir/tdir/tfile || error "chmod 755 tdir/tfile failed"
}
run_test 852 "mkdir using intent lock for striped directory"
test_853() {
local file=$DIR/$tfile
local size=$((PAGE_SIZE * 2))
dd if=/dev/zero of=$file bs=1M count=100 ||
error "failed to write $file"
cancel_lru_locks $OSC
$LCTL set_param llite.*.read_ahead_stats=clear
$MULTIOP $file or1048576c || error "failed to read $file"
$LCTL get_param llite.*.read_ahead_stats
cancel_lru_locks $OSC
$LCTL set_param llite.*.read_ahead_stats=clear
$MULTIOP $file oir1048576z1048576r${size}c ||
error "failed to read $file"
$LCTL get_param llite.*.read_ahead_stats
local ranum=$($LCTL get_param -n llite.*.read_ahead_stats |
get_named_value 'readahead.pages' | calc_sum)
(( ranum == 0 )) || error "should not trigger readahead"
local rndnum=$($LCTL get_param -n llite.*.read_ahead_stats |
get_named_value 'forceread.pages' | calc_sum)
(( rndnum == 2 )) || error "force random read: $rndnum, expected 2"
}
run_test 853 "Verify that random fadvise works as expected"
test_854() {
# Save original parameters to use in test and restore later
local ra_mb=$($LCTL get_param -n llite.*.max_read_ahead_mb) ||
error "Failed to get max_read_ahead_mb"
local ra_per_file_mb=$($LCTL get_param -n \
llite.*.max_read_ahead_per_file_mb) ||
error "Failed to get max_read_ahead_per_file_mb"
local ra_whole_mb=$($LCTL get_param -n \
llite.*.max_read_ahead_whole_mb) ||
error "Failed to get max_read_ahead_whole_mb"
# Get max_cached_mb value - extract just the numeric value
local max_cached_mb=$($LCTL get_param llite.*.max_cached_mb |
awk '/^max_cached_mb:/ { print $2 }') ||
error "Failed to get max_cached_mb"
# Get full output for max_cached_mb for debug information
$LCTL get_param llite.*.max_cached_mb ||
error "Failed to get max_cached_mb details"
stack_trap "$LCTL set_param -n llite.*.max_read_ahead_mb=$ra_mb" ||
error "Failed to set up stack_trap for max_read_ahead_mb"
stack_trap "$LCTL set_param -n \
llite.*.max_read_ahead_per_file_mb=$ra_per_file_mb" ||
error "Failed to set stack_trap for max_read_ahead_per_file_mb"
stack_trap "$LCTL set_param -n \
llite.*.max_read_ahead_whole_mb=$ra_whole_mb" ||
error "Failed to set up stack_trap for max_read_ahead_whole_mb"
stack_trap "$LCTL set_param -n \
llite.*.max_cached_mb=$max_cached_mb" ||
error "Failed to set up stack_trap for max_cached_mb"
log "Initial readahead: $ra_mb/$ra_per_file_mb/$ra_whole_mb MB"
log "Initial max_cached_mb: $max_cached_mb MB"
local total_ram_mb=$(($(facet_meminfo client MemTotal) / 1024))
log "Total RAM: $total_ram_mb MB"
# Set max_cached_mb to 75% of RAM for the test
$LCTL set_param -n llite.*.max_cached_mb=75% ||
error "Failed to set max_cached_mb to 75%"
# Get new max_cached_mb value - extract just the numeric value
local new_max_cached_mb=$($LCTL get_param llite.*.max_cached_mb |
awk '/^max_cached_mb:/ { print $2 }') ||
error "Failed to get new max_cached_mb value"
local exp_cached=$((total_ram_mb * 75 / 100))
log "max_cached_mb=75%: $new_max_cached_mb MB, expect ~$exp_cached MB"
(( new_max_cached_mb * 100 / total_ram_mb >= 74 &&
new_max_cached_mb * 100 / total_ram_mb <= 76 )) ||
error "max_cached_mb not 75% of RAM"
# Test 1: Set max_read_ahead_mb=25% of total RAM
$LCTL set_param -n llite.*.max_read_ahead_mb=25% ||
error "Failed to set max_read_ahead_mb to 25%"
local expect=$((total_ram_mb * 25 / 100))
ra_mb=$($LCTL get_param -n llite.*.max_read_ahead_mb) ||
error "Failed to get updated max_read_ahead_mb"
log "max_ra_mb=25%: got $ra_mb MB, expect ~$expect MB"
# The actual implementation may limit value differently than expected
# Accept any value that's at least 10% of RAM and not more than 30%
(( ra_mb * 100 / total_ram_mb >= 10 &&
ra_mb * 100 / total_ram_mb <= 30 )) ||
error "max_read_ahead_mb=$ra_mb MB not in range (10-30% of RAM)"
# Test 2: Set max_read_ahead_per_file_mb=5% of total RAM
$LCTL set_param -n llite.*.max_read_ahead_per_file_mb=5% ||
error "Failed to set max_read_ahead_per_file_mb to 5%"
expect=$((total_ram_mb * 5 / 100))
ra_per_file_mb=$($LCTL get_param -n \
llite.*.max_read_ahead_per_file_mb) ||
error "Failed to get updated max_read_ahead_per_file_mb"
log "max_ra_per_file_mb=5%: got $ra_per_file_mb MB, expect ~$expect"
# The actual implementation may limit value differently than expected
# Just verify that the value is reasonable (not zero and not too large)
(( ra_per_file_mb > 0 && ra_per_file_mb <= ra_mb )) ||
error "max_ra_per_file_mb=$ra_per_file_mb MB not in (0-$ra_mb MB)"
# Test 3: Set max_read_ahead_whole_mb=1% of total RAM
$LCTL set_param -n llite.*.max_read_ahead_whole_mb=1% ||
error "Failed to set max_read_ahead_whole_mb to 1%"
expect=$((total_ram_mb * 1 / 100))
ra_whole_mb=$($LCTL get_param -n llite.*.max_read_ahead_whole_mb) ||
error "Failed to get updated max_read_ahead_whole_mb"
log "max_ra_whole_mb=1%: got $ra_whole_mb MB, expect ~$expect"
# The actual implementation may limit value differently than expected
# Just verify that the value is reasonable (not too large)
(( ra_whole_mb >= 0 && ra_whole_mb <= ra_per_file_mb )) ||
error "max_ra_whole_mb=$ra_whole_mb MB not in (0-$ra_per_file_mb MB)"
# Test 4: Verify setting max_read_ahead_mb > 50% is capped at 50%
# The implementation should enforce the 50% limit by capping the value
# First try with a value just over 50%
log "Testing if max_read_ahead_mb=51% is capped at 50%"
$LCTL set_param -n llite.*.max_read_ahead_mb=51% ||
error "Failed to set max_read_ahead_mb to 51%"
# Get the actual value set
local ra_mb_51=$($LCTL get_param -n llite.*.max_read_ahead_mb) ||
error "Failed to get max_read_ahead_mb after setting to 51%"
log "max_read_ahead_mb is $ra_mb_51 MB"
# Check if the value was actually capped at 50%
local expect_50=$((total_ram_mb * 50 / 100))
if (( ra_mb_51 > expect_50 )); then
error "max_read_ahead_mb $ra_mb_51 MB > 50% of RAM ($expect_50 MB)"
fi
log "Value correctly capped at ~50% of RAM ($ra_mb_51 <= $expect_50)"
# Now try with a much larger value to be sure
log "Testing if max_read_ahead_mb=90% is capped at 50%"
$LCTL set_param -n llite.*.max_read_ahead_mb=90% ||
error "Failed to set max_read_ahead_mb to 90%"
# Get the actual value set
local ra_mb_90=$($LCTL get_param -n llite.*.max_read_ahead_mb) ||
error "Failed to get max_read_ahead_mb after setting to 90%"
log "max_read_ahead_mb is $ra_mb_90 MB"
# Check if the value was actually capped at 50%
if (( ra_mb_90 > expect_50 )); then
error "max_read_ahead_mb $ra_mb_90 MB > 50% RAM ($expect_50 MB)"
fi
log "Value correctly capped at ~50% of RAM ($ra_mb_90 <= $expect_50)"
# Test 5: Verify max_ra_per_file_mb is capped at max_ra_mb
log "Test if max_read_ahead_per_file_mb > max_read_ahead_mb is capped"
$LCTL set_param -n llite.*.max_read_ahead_per_file_mb=60% ||
error "Failed to set max_read_ahead_per_file_mb to 60%"
# Get the actual value set
local ra_per_file_mb_60=$($LCTL get_param -n \
llite.*.max_read_ahead_per_file_mb) ||
error "Fail to get max_read_ahead_per_file_mb after set to 60%"
log "max_read_ahead_per_file_mb is $ra_per_file_mb_60 MB"
# Check if the value was capped at max_ra_mb
if (( ra_per_file_mb_60 > ra_mb_90 )); then
error "max_read_ahead_per_file_mb $ra_per_file_mb_60 MB exceeds max_read_ahead_mb ($ra_mb_90 MB)"
fi
log "Value capped at max_read_ahead_mb ($ra_per_file_mb_60 <= $ra_mb_90)"
# Test 6: Verify max_ra_whole_mb is capped at max_ra_per_file_mb
log "Test if max_read_ahead_whole_mb > max_read_ahead_per_file_mb capped"
$LCTL set_param -n llite.*.max_read_ahead_whole_mb=70% ||
error "Failed to set max_read_ahead_whole_mb to 70%"
# Get the actual value set
local ra_whole_mb_70=$($LCTL get_param -n \
llite.*.max_read_ahead_whole_mb) ||
error "Failed to get max_read_ahead_whole_mb after set to 70%"
log "max_read_ahead_whole_mb is $ra_whole_mb_70 MB"
# Check if the value was capped at max_ra_per_file_mb
if (( ra_whole_mb_70 > ra_per_file_mb_60 )); then
error "max_read_ahead_whole_mb $ra_whole_mb_70 MB exceeds max_read_ahead_per_file_mb ($ra_per_file_mb_60 MB)"
fi
log "Value capped at max_read_ahead_per_file_mb ($ra_whole_mb_70 <= $ra_per_file_mb_60)"
# Verify max_cached_mb is still set correctly (we set it to 75% earlier)
new_max_cached_mb=$($LCTL get_param llite.*.max_cached_mb |
awk '/^max_cached_mb:/ { print $2 }') ||
error "Failed to get final max_cached_mb value"
log "Final max_cached_mb: $new_max_cached_mb MB"
# We set it to 75% earlier, verify it's still around that value
(( new_max_cached_mb * 100 / total_ram_mb >= 74 &&
new_max_cached_mb * 100 / total_ram_mb <= 76 )) ||
error "max_cached_mb=$new_max_cached_mb MB not close to 75% of RAM"
log "max_cached_mb percentage functionality verified successfully"
}
run_test 854 "verify llite.*.max_cached_mb setting"
test_855() {
[ $PARALLEL == "yes" ] && skip "skip parallel run"
[[ $($LCTL get_param mdc.*.import) =~ connect_flags.*readdir_open ]] ||
skip "server does not support readdir_open"
local nrfiles=1000
local fname="$DIR/$tdir/$tfile"
test_mkdir -c 1 -i 0 "$(dirname $fname)"
stack_trap "simple_cleanup_common $nrfiles" EXIT
local old=$($LCTL get_param -n llite.*.dir_read_on_open)
$LCTL set_param -n llite.*.dir_read_on_open=1
stack_trap "$LCTL set_param -n llite.*.dir_read_on_open=$old" EXIT
createmany -m "$fname" $nrfiles
cancel_lru_locks mdc
lctl set_param mdc.*.stats clear
local num_ls=$(ls -1 $DIR/$tdir | wc -l)
local mds_readpage=$(calc_stats mdc.*.stats mds_readpage)
(( $mds_readpage == 0 )) || error "readpages: $mds_readpage"
(( $nrfiles == $num_ls )) || error "incorrect reading dir"
}
run_test 855 "readdir on open validation"
test_860() {
local file=$DIR/$tfile
local size
# Create a test file with known content
echo "0123456789ABCDEF" > $file
size=$(stat -c%s $file)
echo "File size: $size bytes"
# Test Xe (SEEK_END with offset 0) - should return file size
local result=$($MULTIOP $file oXepc)
echo "Xe result: $result"
(( result == size )) ||
error "Xe failed: expected $size, got $result"
# Test Xe with positive offset - seek past EOF
result=$($MULTIOP $file oXe5pc)
echo "Xe5 result: $result"
(( result == size + 5 )) ||
error "Xe5 failed: expected $((size + 5)), got $result"
# Test Xe with negative offset - seek before EOF
result=$($MULTIOP $file oXe-5pc)
echo "Xe-5 result: $result"
(( result == size - 5 )) ||
error "Xe-5 failed: expected $((size - 5)), got $result"
# Test that Xe actually positions the file pointer correctly
# by seeking to EOF-3, writing 3 bytes, and verifying file size
# doesn't change (we're overwriting existing bytes)
local file2=$DIR/${tfile}.2
echo "0123456789ABCDEF" > $file2
local size2=$(stat -c%s $file2)
$MULTIOP $file2 OXe-3w3c
local new_size=$(stat -c%s $file2)
(( new_size == size2 )) ||
error "Xe-3 positioning failed: size changed from $size2 to $new_size"
# Test XX reserved prefix - should error
$MULTIOP $file oXXc 2>&1 | grep -q "reserved for future use" ||
error "XX should be reserved"
}
run_test 860 "verify multiop Xe (SEEK_END) command"
#
# tests that do cleanup/setup should be run at the end
#
test_900() {
[ $PARALLEL == "yes" ] && skip "skip parallel run"
local ls
#define OBD_FAIL_MGC_PAUSE_PROCESS_LOG 0x903
$LCTL set_param fail_loc=0x903
cancel_lru_locks MGC
FAIL_ON_ERROR=true cleanup
FAIL_ON_ERROR=true setup
}
run_test 900 "umount should not race with any mgc requeue thread"
# LUS-6253/LU-11185
test_901() {
local old
local count
local oldc
local newc
local olds
local news
[ $PARALLEL == "yes" ] && skip "skip parallel run"
# some get_param have a bug to handle dot in param name
cancel_lru_locks MGC
old=$(mount -t lustre | wc -l)
# 1 config+sptlrpc
# 2 params
# 3 nodemap
# 4 IR
old=$((old * 4))
oldc=0
count=0
while [ $old -ne $oldc ]; do
oldc=$($LCTL get_param -n 'ldlm.namespaces.MGC*.lock_count')
sleep 1
((count++))
if [ $count -ge $TIMEOUT ]; then
error "too large timeout"
fi
done
umount_client $MOUNT || error "umount failed"
mount_client $MOUNT || error "mount failed"
cancel_lru_locks MGC
newc=$($LCTL get_param -n 'ldlm.namespaces.MGC*.lock_count')
[ $oldc -lt $newc ] && error "mgc lock leak ($oldc != $newc)"
return 0
}
run_test 901 "don't leak a mgc lock on client umount"
# LU-13377
test_902() {
[ $CLIENT_VERSION -lt $(version_code 2.13.52) ] &&
skip "client does not have LU-13377 fix"
#define OBD_FAIL_LLITE_SHORT_COMMIT 0x1415
$LCTL set_param fail_loc=0x1415
dd if=/dev/zero of=$DIR/$tfile bs=1M count=1
cancel_lru_locks osc
rm -f $DIR/$tfile
}
run_test 902 "test short write doesn't hang lustre"
# LU-14711
test_903() {
$LFS setstripe -i 0 -c 1 $DIR/$tfile $DIR/${tfile}-2
echo "blah" > $DIR/${tfile}-2
dd if=/dev/zero of=$DIR/$tfile bs=1M count=6 conv=fsync
#define OBD_FAIL_OSC_SLOW_PAGE_EVICT 0x417
$LCTL set_param fail_loc=0x417 fail_val=20
mv $DIR/${tfile}-2 $DIR/$tfile # Destroys the big object
sleep 1 # To start the destroy
wait_destroy_complete 150 || error "Destroy taking too long"
cat $DIR/$tfile > /dev/null || error "Evicted"
}
run_test 903 "Test long page discard does not cause evictions"
test_904() {
[ "$mds1_FSTYPE" == "ldiskfs" ] || skip "ldiskfs only test"
do_facet mds1 $DEBUGFS -R features $(mdsdevname 1) |
grep -q project || skip "skip project quota not supported"
local testfile="$DIR/$tdir/$tfile"
local xattr="trusted.projid"
local projid
local mdts=$(mdts_nodes)
local saved=$(do_facet mds1 $LCTL get_param -n \
osd-ldiskfs.*MDT0000.enable_projid_xattr)
do_nodes $mdts $LCTL set_param osd-ldiskfs.*MDT*.enable_projid_xattr=0
stack_trap "do_nodes $mdts $LCTL set_param \
osd-ldiskfs.*MDT*.enable_projid_xattr=$saved"
mkdir -p $DIR/$tdir
touch $testfile
#hide projid xattr on server
$LFS project -p 1 $testfile ||
error "set $testfile project id failed"
getfattr -m - $testfile | grep $xattr &&
error "do not show trusted.projid when disabled on server"
do_nodes $mdts $LCTL set_param osd-ldiskfs.*MDT*.enable_projid_xattr=1
#should be hidden when projid is 0
$LFS project -p 0 $testfile ||
error "set $testfile project id failed"
getfattr -m - $testfile | grep $xattr &&
error "do not show trusted.projid with project ID 0"
#still can getxattr explicitly
projid=$(getfattr -n $xattr $testfile |
sed -n 's/^trusted\.projid="\(.*\)"/\1/p')
[ $projid == "0" ] ||
error "projid expected 0 not $projid"
#set the projid via setxattr
setfattr -n $xattr -v "1000" $testfile ||
error "setattr failed with $?"
projid=($($LFS project $testfile))
[ ${projid[0]} == "1000" ] ||
error "projid expected 1000 not $projid"
#check the new projid via getxattr
$LFS project -p 1001 $testfile ||
error "set $testfile project id failed"
getfattr -m - $testfile | grep $xattr ||
error "should show trusted.projid when project ID != 0"
projid=$(getfattr -n $xattr $testfile |
sed -n 's/^trusted\.projid="\(.*\)"/\1/p')
[ $projid == "1001" ] ||
error "projid expected 1001 not $projid"
#try to set invalid projid
setfattr -n $xattr -v "4294967295" $testfile &&
error "set invalid projid should fail"
#remove the xattr means setting projid to 0
setfattr -x $xattr $testfile ||
error "setfattr failed with $?"
projid=($($LFS project $testfile))
[ ${projid[0]} == "0" ] ||
error "projid expected 0 not $projid"
#should be hidden when parent has inherit flag and same projid
$LFS project -srp 1002 $DIR/$tdir ||
error "set $tdir project id failed"
getfattr -m - $testfile | grep $xattr &&
error "do not show trusted.projid with inherit flag"
#still can getxattr explicitly
projid=$(getfattr -n $xattr $testfile |
sed -n 's/^trusted\.projid="\(.*\)"/\1/p')
[ $projid == "1002" ] ||
error "projid expected 1002 not $projid"
}
run_test 904 "virtual project ID xattr"
# LU-8582
test_905() {
(( $OST1_VERSION >= $(version_code 2.15.50.220) )) ||
skip "need OST version >= 2.15.50.220 for fail_loc"
remote_ost_nodsh && skip "remote OST with nodsh"
$LFS setstripe -c -1 -i 0 $DIR/$tfile || error "setstripe failed"
$LFS ladvise -a willread $DIR/$tfile || error "ladvise does not work"
#define OBD_FAIL_OST_OPCODE 0x253
# OST_LADVISE = 21
do_facet ost1 "$LCTL set_param fail_val=21 fail_loc=0x0253"
$LFS ladvise -a willread $DIR/$tfile &&
error "unexpected success of ladvise with fault injection"
$LFS ladvise -a willread $DIR/$tfile |&
grep -q "Operation not supported"
(( $? == 0 )) || error "unexpected stderr of ladvise with fault injection"
}
run_test 905 "bad or new opcode should not stuck client"
test_906() {
grep -q io_uring_setup /proc/kallsyms ||
skip "Client OS does not support io_uring I/O engine"
io_uring_probe || skip "kernel does not support io_uring fully"
which fio || skip_env "no fio installed"
fio --enghelp | grep -q io_uring ||
skip_env "fio does not support io_uring I/O engine"
local file=$DIR/$tfile
local ioengine="io_uring"
local numjobs=2
local size=50M
fio --name=seqwrite --ioengine=$ioengine \
--bs=$PAGE_SIZE --direct=1 --numjobs=$numjobs \
--iodepth=64 --size=$size --filename=$file --rw=write ||
error "fio seqwrite $file failed"
fio --name=seqread --ioengine=$ioengine \
--bs=$PAGE_SIZE --direct=1 --numjobs=$numjobs \
--iodepth=64 --size=$size --filename=$file --rw=read ||
error "fio seqread $file failed"
rm -f $file || error "rm -f $file failed"
}
run_test 906 "Simple test for io_uring I/O engine via fio"
test_907() {
local max_pages=$($LCTL get_param -n osc.*.max_pages_per_rpc | head -n1)
# set stripe size to max rpc size
$LFS setstripe -i 0 -c 2 -S $((max_pages * PAGE_SIZE)) $DIR/$tfile
$LFS getstripe $DIR/$tfile
#define OBD_FAIL_OST_EROFS 0x216
do_facet ost1 "$LCTL set_param fail_val=3 fail_loc=0x80000216"
local bs=$((max_pages * PAGE_SIZE / 16))
# write full one stripe and one block
dd if=/dev/zero of=$DIR/$tfile bs=$bs count=17 || error "dd failed"
rm $DIR/$tfile || error "rm failed"
}
run_test 907 "write rpc error during unlink"
test_908a() {
local patch_ver=$(version_code v2_16_50-161-ge33d196b93)
(( $MGS_VERSION >= $patch_ver )) ||
skip "need MGS >= 2.16.50.161 for llog change timestamps"
[[ $(facet_fstype mgs) == ldiskfs ]] || skip "ldiskfs only test"
local format_ver=($(do_facet mgs $LCTL llog_print --raw params |
awk '/version/ { print $10 }'))
(( $(version_code $format_ver) > $patch_ver )) ||
skip "MGS configured with version $format_ver need 2.16.50.161"
local dev=$(facet_device mgs)
local cmd="debugfs -c -R 'stat CONFIGS/params' $dev"
# ctime: value from debugfs is in hex
local base_time=$(date -d "24 hours ago" +%s)
local ctime_mgs=$(do_facet mgs "$cmd" |&
awk -F'[: ]' '/ctime:/ { print $4 }')
ctime_mgs=$((ctime_mgs))
echo "ctime_mgs=$ctime_mgs, base_time=$base_time"
(( "$ctime_mgs" > "$base_time" )) ||
error "invalid ctime $ctime_mgs <= $base_time"
}
run_test 908a "llog created with valid ctime"
test_908b() {
(( MDS1_VERSION >= $(version_code v2_16_50-161-ge33d196b93) )) ||
skip "need MDS >= 2.16.50.161 for llog write timestamps"
[[ "$mds1_FSTYPE" == ldiskfs ]] || skip "ldiskfs only test"
local dev=$(mdsdevname 1)
changelog_register || error "cannot register changelog user"
# set changelog_mask to ALL
changelog_chmask "ALL"
changelog_clear
for ((i=0; i<100; i++)); do
echo "$i" > $DIR/$tfile${i}
rm $DIR/$tfile${i}
done
sleep 5
changelog_deregister || error "changelog_deregister failed"
local cmd="debugfs -c -R 'stat changelog_catalog' $dev"
# ctime: and mtime: values from debugfs are in hex
local ctime_mds=$(do_facet mds1 "$cmd" |&
awk -F'[: ]' '/ctime:/ { print $4 }')
ctime_mds=$((ctime_mds))
local mtime_mds=$(do_facet mds1 "$cmd" |&
awk -F'[: ]' '/mtime:/ { print $4 }')
mtime_mds=$((mtime_mds))
echo "ctime_mds=$ctime_mds, mtime_mds=$mtime_mds"
(( "$mtime_mds" > "$ctime_mds" )) ||
error "invalid mtime $mtime_mds <= $ctime_mds"
}
run_test 908b "changelog stores valid mtime"
test_909() {
(( $MDS1_VERSION >= $(version_code 2.16.56) )) ||
skip "Need MDS version at least 2.16.56"
(( $MDSCOUNT >= 2 )) || skip_env "needs >= 2 MDTs"
test_mkdir -i 1 $DIR/$tdir || error "mkdir $DIR/$tdir failed"
$LFS setstripe -E 1M -L mdt $DIR/$tdir/$tfile ||
error "failed to create DoM-only file $DIR/$tdir/$tfile"
mdt_index=$($LFS getstripe "$DIR/$tdir/$tfile" | \
awk 'found && /lmm_mdt_index:/ { print $2; exit }
/lmm_pattern:/ { found = ($2 == "mdt") }')
[[ "$mdt_index" == "1" ]] ||
error "invalid mdt index $mdt_index != 1"
}
run_test 909 "Verify mdt index"
test_910()
{
local run_id=$RANDOM
# Try to insert the module.
load_module kunit/ec_test run_id=$run_id ||
error "load_module failed"
# Anything but success is a test failure
dmesg | grep -q \
"lustre_ec_test_$run_id: EC test passed" ||
error "EC test failed"
# Remove the test module
rmmod -v ec_test ||
error "rmmod failed (may trigger a failure in a later test)"
}
run_test 910 "Test the erasure_coding module"
test_911()
{
local lfs_cmdlist="hsm mirror pcc"
local lctl_cmdlist="barrier llog nodemap"
local cmd
for cmd in $lfs_cmdlist; do
! $LFS $cmd --list-commands | grep -E "setstripe|find" ||
error "lfs $cmd --list-commands not limiting output"
done
(( $MGS_VERSION >= $(version_code 2.17.50) )) ||
skip "Need MGS version at least 2.17.50"
for cmd in $lctl_cmdlist; do
! do_facet mgs $LCTL $cmd --list-commands | grep -E "====" ||
error "lctl $cmd --list-commands not limiting output"
done
}
run_test 911 "Check lfs/lctl --list-commands"
test_920()
{
mount | grep lustre
mkdir -p $DIR/$tdir
#define OBD_FAIL_FID_MULTI 0x1507
do_facet $SINGLEMDS $LCTL set_param fail_loc=0x1507
touch $DIR/$tdir/$tfile || error "touch failed"
lfs path2fid $DIR/$tdir/$tfile # Verify it's a normal FID
stat $DIR/$tdir/$tfile || error "File not found after creation"
}
run_test 920 "Test multy LUFID"
complete_test $SECONDS
[ -f $EXT2_DEV ] && rm $EXT2_DEV || true
check_and_cleanup_lustre
if [ "$I_MOUNTED" != "yes" ]; then
lctl set_param debug="$OLDDEBUG" 2> /dev/null || true
fi
exit_status
