Viewing: sanity-lnet.sh
#!/usr/bin/bash
#
# Run select tests by setting ONLY, or as arguments to the script.
# Skip specific tests by setting EXCEPT.
#
set -e
ONLY=${ONLY:-"$*"}
LUSTRE=${LUSTRE:-$(cd $(dirname $0)/..; echo $PWD)}
. $LUSTRE/tests/test-framework.sh
CLEANUP=${CLEANUP:-:}
SETUP=${SETUP:-:}
PTLDEBUG="$PTLDEBUG neterror net nettrace malloc"
init_test_env "$@"
. ${CONFIG:=$LUSTRE/tests/cfg/$NAME.sh}
init_logging
ALWAYS_EXCEPT="$SANITY_LNET_EXCEPT "
always_except LU-10391 253 254
always_except LU-18751 410
[[ "$SLOW" = "no" ]] && EXCEPT_SLOW=""
[[ -z $LNETCTL ]] && skip "Need lnetctl"
restore_mounts=false
if is_mounted $MOUNT || is_mounted $MOUNT2; then
cleanupall || error "Failed cleanup prior to test execution"
restore_mounts=true
fi
cleanup_lnet() {
echo "Cleaning up LNet"
lsmod | grep -q lnet &&
$LNETCTL lnet unconfigure 2>/dev/null
unload_modules
}
if (( $MDS1_VERSION >= $(version_code 2.16.56) )); then
net_delay_add="net_delay add"
net_delay_del="net_delay del"
else
net_delay_add="net_delay_add"
net_delay_del="net_delay_del"
fi
restore_modules=false
if module_loaded lnet ; then
cleanup_lnet || error "Failed to unload modules before test execution"
restore_modules=true
fi
cleanup_testsuite() {
trap "" EXIT
# Cleanup any tmp files created by the sub tests
rm -f $TMP/sanity-lnet-*.yaml $LNET_PARAMS_FILE
cleanup_netns
cleanup_lnet
if $restore_mounts; then
setupall || error "Failed to setup Lustre after test execution"
elif $restore_modules; then
load_modules ||
error "Couldn't load modules after test execution"
fi
return 0
}
TESTNS='test_ns'
FAKE_IF="test1pg"
FAKE_IP="10.1.2.3"
FAKE_IP_ALIAS="10.1.2.31"
FAKE_IPV6="2001:db8:0:f101::1"
do_ns() {
echo "ip netns exec $TESTNS $*"
ip netns exec $TESTNS "$@"
}
setup_fakeif() {
local netns="$1"
local netns_arg netns_exec
if [[ -n $netns ]]; then
netns_arg="netns $netns"
netns_exec="ip netns exec $netns"
fi
echo "ip link add 'test1pl' type veth peer name $FAKE_IF $netns_arg"
ip link add 'test1pl' type veth peer name $FAKE_IF $netns_arg
echo "ip link set 'test1pl' up"
ip link set 'test1pl' up
echo "$netns_exec ip addr add \"${FAKE_IP}/31\" dev $FAKE_IF"
$netns_exec ip addr add "${FAKE_IP}/31" dev $FAKE_IF
echo "$netns_exec ip -6 addr add \"${FAKE_IPV6}/64\" dev $FAKE_IF"
$netns_exec ip -6 addr add "${FAKE_IPV6}/64" dev $FAKE_IF
echo "$netns_exec ip link set $FAKE_IF up"
$netns_exec ip link set $FAKE_IF up
local ip4=$($netns_exec ip -o -4 a s $FAKE_IF 2>/dev/null |
grep $FAKE_IP)
local ip6=$($netns_exec ip -o -6 a s $FAKE_IF 2>/dev/null |
grep $FAKE_IPV6)
[[ -n $ip4 && -n $ip6 ]] || error "Failed setup $FAKE_IF"
wait_update $HOSTNAME \
"$netns_exec ip -o a s $FAKE_IF | grep -c tentative" \
"0" "10"
}
cleanup_fakeif() {
ip link show test1pl >& /dev/null && ip link del test1pl || return 0
}
setup_netns() {
cleanup_netns
ip netns add $TESTNS
setup_fakeif $TESTNS
}
cleanup_netns() {
(ip netns list | grep -q $TESTNS) && ip netns del $TESTNS
cleanup_fakeif
}
configure_dlc() {
echo "Loading LNet and configuring DLC"
load_lnet || return $?
do_lnetctl lnet configure $LNET_CONFIG_OPT
}
GLOBAL_YAML_FILE=$TMP/sanity-lnet-global.yaml
define_global_yaml() {
$LNETCTL export --backup &>${GLOBAL_YAML_FILE} ||
error "Failed to export global yaml $?"
}
reinit_dlc() {
if lsmod | grep -q lnet; then
do_lnetctl lnet unconfigure ||
error "lnetctl lnet unconfigure failed $?"
do_lnetctl lnet configure $LNET_CONFIG_OPT ||
error "lnetctl lnet configure failed $?"
else
configure_dlc || error "configure_dlc failed $?"
fi
define_global_yaml
}
append_global_yaml() {
[[ ! -e ${GLOBAL_YAML_FILE} ]] &&
error "Missing global yaml at ${GLOBAL_YAML_FILE}"
cat ${GLOBAL_YAML_FILE} >> $TMP/sanity-lnet-$testnum-expected.yaml
}
create_base_yaml_file() {
append_global_yaml
}
compare_yaml_files() {
local expected="$TMP/sanity-lnet-$testnum-expected.yaml"
local actual="$TMP/sanity-lnet-$testnum-actual.yaml"
local rc=0
! [[ -e $expected ]] && echo "$expected not found" && return 1
! [[ -e $actual ]] && echo "$actual not found" && return 1
if [ verify_yaml_available ]; then
verify_compare_yaml $actual $expected || rc=$?
else
diff -upN ${actual} ${expected} || rc=$?
fi
echo "Expected:"
cat $expected
echo "Actual:"
cat $actual
return $rc
}
validate_nid() {
local nid="$1"
local net="${nid//*@/}"
local addr="${nid//@*/}"
local rc=1
local num_re='[0-9]+'
if [[ $net =~ (gni|kfi|bxi3f)[0-9]* ]]; then
[[ $addr =~ ${num_re} ]] && rc=0
elif [[ $net =~ tcp[0-9]* ]]; then
if ip_is_v4 "$addr" || ip_is_v6 "$addr"; then
rc=0
fi
elif [[ $net =~ o2ib[0-9]* ]]; then
ip_is_v4 "$addr" && rc=0
else
echo "Unrecognized net: \"$net\""
fi
if ((rc != 0)); then
echo "Invalid nid: \"$nid\""
fi
return $rc
}
validate_nids() {
local yfile=$TMP/sanity-lnet-$testnum-actual.yaml
local primary_nids=$(awk '/-\s+primary nid:/{print $NF}' $yfile |
xargs echo)
local secondary_nids=$(awk '/-\s+nid:/{print $NF}' $yfile | xargs echo)
local gateway_nids=$(awk '/gateway:/{print $NF}' $yfile | xargs echo)
local nid
for nid in $primary_nids $secondary_nids $gateway_nids; do
validate_nid "$nid" || error "Bad NID \"${nid}\""
done
return 0
}
validate_peer_nids() {
local num_peers="$1"
local nids_per_peer="$2"
local expect_p="$num_peers"
# The primary nid also shows up in the list of secondary nids
local expect_s="$(($num_peers + $(($nids_per_peer*$num_peers))))"
local actual_p=$(awk '/-\s+primary nid:/{print $NF}' $TMP/sanity-lnet-$testnum-actual.yaml | wc -l)
local actual_s=$(awk '/-\s+nid:/{print $NF}' $TMP/sanity-lnet-$testnum-actual.yaml | wc -l)
if [[ $expect_p -ne $actual_p ]]; then
compare_yaml_files
error "Expected $expect_p but found $actual_p primary nids"
elif [[ $expect_s -ne $actual_s ]]; then
compare_yaml_files
error "Expected $expect_s but found $actual_s secondary nids"
fi
validate_nids
}
validate_gateway_nids() {
local expect_gw=$(grep -c -- 'gateway:' $TMP/sanity-lnet-$testnum-expected.yaml)
local actual_gw=$(grep -c -- 'gateway:' $TMP/sanity-lnet-$testnum-actual.yaml)
echo "expect_gw: $expect_gw actual_gw: $actual_gw"
if [[ $expect_gw -ne $actual_gw ]]; then
compare_yaml_files
error "Expected $expect_gw gateways but found $actual_gw gateways"
fi
local expect_gwnids=$(awk '/gateway:/{print $NF}' $TMP/sanity-lnet-$testnum-expected.yaml |
xargs echo)
local nid
echo "expect_gwnids: $expect_gwnids"
for nid in ${expect_gwnids}; do
if ! grep -q "gateway: ${nid}" $TMP/sanity-lnet-$testnum-actual.yaml; then
compare_yaml_files
error "${nid} not configured as gateway"
fi
done
validate_nids
}
intf_has_ipv6() {
local addr=$(ip -o -6 a s "$1" | awk '{print $4}' | head -n 1 |
grep -v '^fe80::' | sed 's,/[0-9]\+$,,')
ip_is_v6 "${addr}"
}
intf_has_ipv4() {
local addr=$(ip -o -4 a s "$1" | awk '{print $4}' | head -n 1 |
sed 's,/[0-9]\+$,,')
ip_is_v4 "${addr}"
}
cleanupall -f
# Determine the local interface(s) used for LNet
load_lnet "config_on_load=1" || error "Failed to load modules"
do_lnetctl net show
ip a
INTERFACES=( $(lnet_if_list) )
if [[ -z ${INTERFACES[@]} ]]; then
error "Did not identify any LNet interfaces"
fi
# If we don't have IPv6 addresses then make sure the test suite runs in
# "IPv4 mode". If we have IPv6, but not IPv4, then make sure we run in
# "IPv6 mode". In a mixed environment we take whatever has been specified
# by the test environment configuration.
if ! intf_has_ipv6 ${INTERFACES[0]}; then
FORCE_LARGE_NID=false
LNET_CONFIG_INIT_OPT="--all"
LNET_CONFIG_OPT=""
elif ! intf_has_ipv4 ${INTERFACES[0]}; then
FORCE_LARGE_NID=true
LNET_CONFIG_INIT_OPT="--all --large"
LNET_CONFIG_OPT="-l"
fi
if $FORCE_LARGE_NID; then
always_except LU-14288 101
always_except LU-14288 103
always_except LU-17457 199
always_except LU-17457 208
always_except LU-9680 213
always_except LU-17458 220
always_except LU-19314 228
always_except LU-5960 230
always_except LU-9680 231
always_except LU-17457 255
always_except LU-19334 257
always_except LU-19363 270
always_except LU-9680 302
FAKE_NID="${FAKE_IPV6}@tcp"
else
FAKE_NID="${FAKE_IP}@tcp"
fi
build_test_filter
cleanup_lnet || error "Failed to cleanup LNet"
stack_trap 'cleanup_testsuite' EXIT
test_0() {
configure_dlc || error "Failed to configure DLC rc = $?"
define_global_yaml
reinit_dlc || return $?
do_lnetctl import < ${GLOBAL_YAML_FILE} || error "Import failed $?"
$LNETCTL export --backup > $TMP/sanity-lnet-$testnum-actual.yaml
create_base_yaml_file
compare_yaml_files || error "Configuration changed after import"
}
run_test 0 "Export empty config, import the config, compare"
compare_peer_add() {
local prim_nid="${1:+--prim_nid $1}"
local nid="${2:+--nid $2}"
local actual="$TMP/sanity-lnet-$testnum-actual.yaml"
do_lnetctl peer add ${prim_nid} ${nid} || error "peer add failed $?"
$LNETCTL export --backup > $actual || error "export failed $?"
compare_yaml_files
return $?
}
test_1() {
reinit_dlc || return $?
cat <<EOF > $TMP/sanity-lnet-$testnum-expected.yaml
peer:
- primary nid: 1.1.1.1@tcp
Multi-Rail: True
peer ni:
- nid: 1.1.1.1@tcp
EOF
append_global_yaml
compare_peer_add "1.1.1.1@tcp"
}
run_test 1 "Add peer with single nid (tcp)"
test_2() {
reinit_dlc || return $?
cat <<EOF > $TMP/sanity-lnet-$testnum-expected.yaml
peer:
- primary nid: 2.2.2.2@o2ib
Multi-Rail: True
peer ni:
- nid: 2.2.2.2@o2ib
EOF
append_global_yaml
compare_peer_add "2.2.2.2@o2ib"
}
run_test 2 "Add peer with single nid (o2ib)"
test_3() {
reinit_dlc || return $?
cat <<EOF > $TMP/sanity-lnet-$testnum-expected.yaml
peer:
- primary nid: 3.3.3.3@tcp
Multi-Rail: True
peer ni:
- nid: 3.3.3.3@tcp
- nid: 3.3.3.3@o2ib
EOF
append_global_yaml
compare_peer_add "3.3.3.3@tcp" "3.3.3.3@o2ib"
}
run_test 3 "Add peer with tcp primary o2ib secondary"
test_4() {
reinit_dlc || return $?
cat <<EOF > $TMP/sanity-lnet-$testnum-expected.yaml
peer:
- primary nid: 4.4.4.4@tcp
Multi-Rail: True
peer ni:
- nid: 4.4.4.4@tcp
- nid: 4.4.4.1@tcp
- nid: 4.4.4.2@tcp
- nid: 4.4.4.3@tcp
EOF
append_global_yaml
echo "Add peer with nidrange (tcp)"
compare_peer_add "4.4.4.4@tcp" "4.4.4.[1-3]@tcp"
echo "Add peer with nidrange that overlaps primary nid (tcp)"
compare_peer_add "4.4.4.4@tcp" "4.4.4.[1-4]@tcp"
}
run_test 4 "Add peer with nidrange (tcp)"
test_5() {
reinit_dlc || return $?
cat <<EOF > $TMP/sanity-lnet-$testnum-expected.yaml
peer:
- primary nid: 5.5.5.5@o2ib
Multi-Rail: True
peer ni:
- nid: 5.5.5.5@o2ib
- nid: 5.5.5.1@o2ib
- nid: 5.5.5.2@o2ib
- nid: 5.5.5.3@o2ib
- nid: 5.5.5.4@o2ib
EOF
append_global_yaml
echo "Add peer with nidrange (o2ib)"
compare_peer_add "5.5.5.5@o2ib" "5.5.5.[1-4]@o2ib"
echo "Add peer with nidranage that overlaps primary nid (o2ib)"
compare_peer_add "5.5.5.5@o2ib" "5.5.5.[1-4]@o2ib"
echo "Add peer with nidranage that contain , plus primary nid (o2ib)"
compare_peer_add "5.5.5.5@o2ib" "5.5.5.[1,2,3-4]@o2ib"
}
run_test 5 "Add peer with nidrange (o2ib)"
test_6() {
reinit_dlc || return $?
cat <<EOF > $TMP/sanity-lnet-$testnum-expected.yaml
peer:
- primary nid: 6.6.6.6@tcp
Multi-Rail: True
peer ni:
- nid: 6.6.6.6@tcp
- nid: 6.6.6.0@tcp
- nid: 6.6.6.2@tcp
- nid: 6.6.6.4@tcp
- nid: 6.6.7.0@tcp
- nid: 6.6.7.2@tcp
- nid: 6.6.7.4@tcp
- nid: 6.6.1.0@o2ib
- nid: 6.6.1.3@o2ib
- nid: 6.6.1.6@o2ib
- nid: 6.6.3.0@o2ib
- nid: 6.6.3.3@o2ib
- nid: 6.6.3.6@o2ib
- nid: 6@gni
- nid: 10@gni
- nid: 6@kfi
- nid: 10@kfi
- nid: 6@bxi3f
- nid: 10@bxi3f
EOF
append_global_yaml
local nid_expr="6.6.[6-7].[0-4/2]@tcp"
nid_expr+=",6.6.[1-4/2].[0-6/3]@o2ib"
nid_expr+=",[6-12/4]@gni"
nid_expr+=",[6-12/4]@kfi"
nid_expr+=",[6-12/4]@bxi3f"
compare_peer_add "6.6.6.6@tcp" "${nid_expr}"
}
run_test 6 "Add peer with multiple nidranges"
compare_peer_del() {
local prim_nid="${1:+--prim_nid $1}"
local nid="${2:+--nid $2}"
local actual="$TMP/sanity-lnet-$testnum-actual.yaml"
do_lnetctl peer del ${prim_nid} ${nid} || error "peer del failed $?"
$LNETCTL export --backup > $actual || error "export failed $?"
compare_yaml_files
return $?
}
test_7() {
reinit_dlc || return $?
create_base_yaml_file
echo "Delete peer with single nid (tcp)"
do_lnetctl peer add --prim_nid 7.7.7.7@tcp || error "Peer add failed $?"
compare_peer_del "7.7.7.7@tcp"
echo "Delete peer with single nid (o2ib)"
do_lnetctl peer add --prim_nid 7.7.7.7@o2ib || error "Peer add failed $?"
compare_peer_del "7.7.7.7@o2ib"
echo "Delete peer that has multiple nids (tcp)"
do_lnetctl peer add --prim_nid 7.7.7.7@tcp --nid 7.7.7.[8-12]@tcp ||
error "Peer add failed $?"
compare_peer_del "7.7.7.7@tcp"
echo "Delete peer that has multiple nids (o2ib)"
do_lnetctl peer add --prim_nid 7.7.7.7@o2ib --nid 7.7.7.[8-12]@o2ib ||
error "Peer add failed $?"
compare_peer_del "7.7.7.7@o2ib"
echo "Delete peer that has both tcp and o2ib nids"
do_lnetctl peer add --prim_nid 7.7.7.7@tcp \
--nid 7.7.7.[9-12]@tcp,7.7.7.[13-15]@o2ib ||
error "Peer add failed $?"
compare_peer_del "7.7.7.7@tcp"
echo "Delete peer with single nid (gni)"
do_lnetctl peer add --prim_nid 7@gni || error "Peer add failed $?"
compare_peer_del "7@gni"
echo "Delete peer that has multiple nids (gni)"
do_lnetctl peer add --prim_nid 7@gni --nid [8-12]@gni ||
error "Peer add failed $?"
compare_peer_del "7@gni"
echo "Delete peer with single nid (kfi)"
do_lnetctl peer add --prim_nid 7@kfi || error "Peer add failed $?"
compare_peer_del "7@kfi"
echo "Delete peer that has multiple nids (kfi)"
do_lnetctl peer add --prim_nid 7@kfi --nid [8-12]@kfi ||
error "Peer add failed $?"
compare_peer_del "7@kfi"
echo "Delete peer with single nid (bxi3f)"
do_lnetctl peer add --prim_nid 7@bxi3f || error "Peer add failed $?"
compare_peer_del "7@bxi3f"
echo "Delete peer that has multiple nids (bxi3f)"
do_lnetctl peer add --prim_nid 7@bxi3f --nid [8-12]@bxi3f ||
error "Peer add failed $?"
compare_peer_del "7@bxi3f"
echo "Delete peer that has tcp, o2ib, gni, kfi and bxi3f nids"
local nid_expr="[8-12]@gni"
nid_expr+=",7.7.7.[1-4]@tcp"
nid_expr+=",7.7.7.[5-9]@o2ib"
nid_expr+=",[1-5]@kfi"
nid_expr+=",[3-7]@bxi3f"
do_lnetctl peer add --prim_nid 7@gni --nid "${nid_expr}" ||
error "Peer add failed $?"
compare_peer_del "7@gni"
echo "Delete peer that contains nid equal to primary nid"
do_lnetctl peer add --prim 1.1.1.1@tcp \
--nid 1.1.1.1@tcp,2.2.2.2@tcp,3.3.3.3@tcp,4.4.4.4@tcp ||
error "Peer add failed $?"
compare_peer_del "1.1.1.1@tcp" "1.1.1.1@tcp,2.2.2.2@tcp,3.3.3.3@tcp,4.4.4.4@tcp"
}
run_test 7 "Various peer delete tests"
test_8() {
reinit_dlc || return $?
cat <<EOF > $TMP/sanity-lnet-$testnum-expected.yaml
peer:
- primary nid: 8.8.8.8@tcp
Multi-Rail: True
peer ni:
- nid: 8.8.8.8@tcp
- nid: 8.8.8.10@tcp
- nid: 8.8.8.11@tcp
- nid: 8.8.8.12@tcp
- nid: 8.8.8.14@tcp
- nid: 8.8.8.15@tcp
EOF
append_global_yaml
do_lnetctl peer add --prim_nid 8.8.8.8@tcp --nid 8.8.8.[10-15]@tcp ||
error "Peer add failed $?"
compare_peer_del "8.8.8.8@tcp" "8.8.8.13@tcp"
}
run_test 8 "Delete single secondary nid from peer (tcp)"
test_9() {
reinit_dlc || return $?
cat <<EOF > $TMP/sanity-lnet-$testnum-expected.yaml
peer:
- primary nid: 9.9.9.9@tcp
Multi-Rail: True
peer ni:
- nid: 9.9.9.9@tcp
EOF
append_global_yaml
do_lnetctl peer add --prim_nid 9.9.9.9@tcp \
--nid 9.9.9.[11-16]@tcp || error "Peer add failed $?"
compare_peer_del "9.9.9.9@tcp" "9.9.9.[11-16]@tcp"
}
run_test 9 "Delete all secondary nids from peer (tcp)"
test_10() {
reinit_dlc || return $?
cat <<EOF > $TMP/sanity-lnet-$testnum-expected.yaml
peer:
- primary nid: 10.10.10.10@tcp
Multi-Rail: True
peer ni:
- nid: 10.10.10.10@tcp
- nid: 10.10.10.12@tcp
- nid: 10.10.10.13@tcp
- nid: 10.10.10.15@tcp
- nid: 10.10.10.16@tcp
EOF
append_global_yaml
do_lnetctl peer add --prim_nid 10.10.10.10@tcp \
--nid 10.10.10.[12-16]@tcp || error "Peer add failed $?"
compare_peer_del "10.10.10.10@tcp" "10.10.10.14@tcp"
}
run_test 10 "Delete single secondary nid from peer (o2ib)"
test_11() {
reinit_dlc || return $?
cat <<EOF > $TMP/sanity-lnet-$testnum-expected.yaml
peer:
- primary nid: 11.11.11.11@tcp
Multi-Rail: True
peer ni:
- nid: 11.11.11.11@tcp
EOF
append_global_yaml
do_lnetctl peer add --prim_nid 11.11.11.11@tcp \
--nid 11.11.11.[13-17]@tcp || error "Peer add failed $?"
compare_peer_del "11.11.11.11@tcp" "11.11.11.[13-17]@tcp"
}
run_test 11 "Delete all secondary nids from peer (o2ib)"
test_12() {
reinit_dlc || return $?
cat <<EOF > $TMP/sanity-lnet-$testnum-expected.yaml
peer:
- primary nid: 12.12.12.12@o2ib
Multi-Rail: True
peer ni:
- nid: 12.12.12.12@o2ib
- nid: 13.13.13.13@o2ib
- nid: 14.13.13.13@o2ib
- nid: 14.15.13.13@o2ib
- nid: 15.17.1.5@tcp
- nid: 15.17.1.10@tcp
- nid: 15.17.1.20@tcp
EOF
append_global_yaml
do_lnetctl peer add --prim_nid 12.12.12.12@o2ib \
--nid [13-14/1].[13-15/2].13.13@o2ib,[15-16/3].[17-19/4].[1].[5-20/5]@tcp ||
error "Peer add failed $?"
compare_peer_del "12.12.12.12@o2ib" "13.15.13.13@o2ib,15.17.1.15@tcp"
}
run_test 12 "Delete a secondary nid from peer (tcp and o2ib)"
test_13() {
reinit_dlc || return $?
cat <<EOF > $TMP/sanity-lnet-$testnum-expected.yaml
peer:
- primary nid: 13.13.13.13@o2ib
Multi-Rail: True
peer ni:
- nid: 13.13.13.13@o2ib
EOF
append_global_yaml
do_lnetctl peer add --prim_nid 13.13.13.13@o2ib \
--nid [14-15].[1-2/1].[1].[100-254/10]@tcp,14.14.[254].14@o2ib ||
error "Peer add failed $?"
compare_peer_del "13.13.13.13@o2ib" \
"[14-15].[1-2/1].[1].[100-254/10]@tcp,14.14.[254].14@o2ib"
}
run_test 13 "Delete all secondary nids from peer (tcp and o2ib)"
create_nid() {
local num="$1"
local net="$2"
if [[ $net =~ (gni|kfi|bxi3f) ]]; then
echo "${num}@${net}"
else
echo "${num}.${num}.${num}.${num}@${net}"
fi
}
create_mr_peer_yaml() {
local num_peers="$1"
local secondary_nids="$2"
local net="$3"
echo "Generating peer yaml for $num_peers peers with $secondary_nids secondary nids"
echo "peer:" >> $TMP/sanity-lnet-$testnum-expected.yaml
local i
local total_nids=$((num_peers + $((num_peers * secondary_nids))))
local created=0
local nidnum=1
while [[ $created -lt $num_peers ]]; do
local primary=$(create_nid ${nidnum} ${net})
cat <<EOF >> $TMP/sanity-lnet-$testnum-expected.yaml
- primary nid: $primary
Multi-Rail: True
peer ni:
- nid: $primary
EOF
local j
local start=$((nidnum + 1))
local end=$((nidnum + $secondary_nids))
for j in $(seq ${start} ${end}); do
local nid=$(create_nid $j ${net})
echo " - nid: $nid" >> $TMP/sanity-lnet-$testnum-expected.yaml
done
nidnum=$((end + 1))
((created++))
done
}
test_14() {
reinit_dlc || return $?
echo "Create single peer, single nid, using import"
create_mr_peer_yaml 1 0 tcp
do_lnetctl import < $TMP/sanity-lnet-$testnum-expected.yaml ||
error "Import failed $?"
append_global_yaml
$LNETCTL export --backup > $TMP/sanity-lnet-$testnum-actual.yaml
compare_yaml_files
echo "Delete single peer using import --del"
do_lnetctl import --del < $TMP/sanity-lnet-$testnum-expected.yaml ||
error "Import failed $?"
rm -f $TMP/sanity-lnet-$testnum-expected.yaml
create_base_yaml_file
$LNETCTL export --backup > $TMP/sanity-lnet-$testnum-actual.yaml
compare_yaml_files
}
run_test 14 "import peer create/delete with single nid"
test_15() {
reinit_dlc || return $?
echo "Create multiple peers, single nid per peer, using import"
create_mr_peer_yaml 5 0 o2ib
# The ordering of nids for this use-case is non-deterministic, so we
# we can't just diff the expected/actual output.
do_lnetctl import < $TMP/sanity-lnet-$testnum-expected.yaml ||
error "Import failed $?"
$LNETCTL export --backup > $TMP/sanity-lnet-$testnum-actual.yaml
validate_peer_nids 5 0
echo "Delete multiple peers, single nid per peer, using import --del"
do_lnetctl import --del < $TMP/sanity-lnet-$testnum-expected.yaml ||
error "Import failed $?"
rm -f $TMP/sanity-lnet-$testnum-expected.yaml
create_base_yaml_file
$LNETCTL export --backup > $TMP/sanity-lnet-$testnum-actual.yaml
compare_yaml_files
}
run_test 15 "import multi peer create/delete with single nid per peer"
test_16() {
reinit_dlc || return $?
echo "Create single peer, multiple nids, using import"
create_mr_peer_yaml 1 5 tcp
do_lnetctl import < $TMP/sanity-lnet-$testnum-expected.yaml ||
error "Import failed $?"
$LNETCTL export --backup > $TMP/sanity-lnet-$testnum-actual.yaml
validate_peer_nids 1 5
echo "Delete single peer, multiple nids, using import --del"
do_lnetctl import --del < $TMP/sanity-lnet-$testnum-expected.yaml ||
error "Import failed $?"
rm -f $TMP/sanity-lnet-$testnum-expected.yaml
create_base_yaml_file
$LNETCTL export --backup > $TMP/sanity-lnet-$testnum-actual.yaml
compare_yaml_files
}
run_test 16 "import peer create/delete with multiple nids"
test_17() {
reinit_dlc || return $?
echo "Create multiple peers, multiple nids per peer, using import"
create_mr_peer_yaml 5 7 o2ib
do_lnetctl import < $TMP/sanity-lnet-$testnum-expected.yaml ||
error "Import failed $?"
$LNETCTL export --backup > $TMP/sanity-lnet-$testnum-actual.yaml
validate_peer_nids 5 7
echo "Delete multiple peers, multiple nids per peer, using import --del"
do_lnetctl import --del < $TMP/sanity-lnet-$testnum-expected.yaml ||
error "Import failed $?"
rm -f $TMP/sanity-lnet-$testnum-expected.yaml
create_base_yaml_file
$LNETCTL export --backup > $TMP/sanity-lnet-$testnum-actual.yaml
compare_yaml_files
}
run_test 17 "import multi peer create/delete with multiple nids"
test_18a() {
reinit_dlc || return $?
cat <<EOF > $TMP/sanity-lnet-$testnum-expected.yaml
peer:
- primary nid: 1.1.1.1@tcp
Multi-Rail: True
peer ni:
- nid: 1.1.1.1@tcp
- nid: 2.2.2.2@tcp
- nid: 4.4.4.4@tcp
- nid: 3.3.3.3@o2ib
- nid: 5@gni
EOF
echo "Import peer with 5 nids"
cat $TMP/sanity-lnet-$testnum-expected.yaml
do_lnetctl import < $TMP/sanity-lnet-$testnum-expected.yaml ||
error "Import failed $?"
cat <<EOF > $TMP/sanity-lnet-$testnum-expected.yaml
peer:
- primary nid: 1.1.1.1@tcp
Multi-Rail: True
peer ni:
- nid: 2.2.2.2@tcp
- nid: 3.3.3.3@o2ib
- nid: 5@gni
EOF
echo "Delete three of the nids"
cat $TMP/sanity-lnet-$testnum-expected.yaml
do_lnetctl import --del < $TMP/sanity-lnet-$testnum-expected.yaml
cat <<EOF > $TMP/sanity-lnet-$testnum-expected.yaml
peer:
- primary nid: 1.1.1.1@tcp
Multi-Rail: True
peer ni:
- nid: 1.1.1.1@tcp
- nid: 4.4.4.4@tcp
EOF
echo "Check peer has expected nids remaining"
$LNETCTL export --backup > $TMP/sanity-lnet-$testnum-actual.yaml
append_global_yaml
compare_yaml_files
}
run_test 18a "Delete a subset of nids from a single peer using import --del"
test_18b() {
reinit_dlc || return $?
cat <<EOF > $TMP/sanity-lnet-$testnum-expected.yaml
peer:
- primary nid: 1.1.1.1@tcp
Multi-Rail: True
peer ni:
- nid: 1.1.1.1@tcp
- nid: 2.2.2.2@tcp
- nid: 4.4.4.4@tcp
- nid: 3.3.3.3@o2ib
- nid: 5@gni
- primary nid: 6.6.6.6@o2ib
Multi-Rail: True
peer ni:
- nid: 6.6.6.6@o2ib
- nid: 7.7.7.7@tcp
- nid: 8.8.8.8@tcp
- nid: 9.9.9.9@tcp
- nid: 10@gni
EOF
echo "Import two peers with 5 nids each"
cat $TMP/sanity-lnet-$testnum-expected.yaml
do_lnetctl import < $TMP/sanity-lnet-$testnum-expected.yaml ||
error "Import failed $?"
cat <<EOF > $TMP/sanity-lnet-$testnum-expected.yaml
peer:
- primary nid: 1.1.1.1@tcp
Multi-Rail: True
peer ni:
- nid: 2.2.2.2@tcp
- nid: 3.3.3.3@o2ib
- nid: 5@gni
- primary nid: 6.6.6.6@o2ib
Multi-Rail: True
peer ni:
- nid: 7.7.7.7@tcp
- nid: 8.8.8.8@tcp
- nid: 10@gni
EOF
echo "Delete three of the nids from each peer"
cat $TMP/sanity-lnet-$testnum-expected.yaml
do_lnetctl import --del < $TMP/sanity-lnet-$testnum-expected.yaml
cat <<EOF > $TMP/sanity-lnet-$testnum-expected.yaml
peer:
- primary nid: 6.6.6.6@o2ib
Multi-Rail: True
peer ni:
- nid: 6.6.6.6@o2ib
- nid: 7.7.7.7@tcp
- primary nid: 1.1.1.1@tcp
Multi-Rail: True
peer ni:
- nid: 1.1.1.1@tcp
- nid: 4.4.4.4@tcp
EOF
append_global_yaml
echo "Check peers have expected nids remaining"
$LNETCTL export --backup > $TMP/sanity-lnet-$testnum-actual.yaml
compare_yaml_files
validate_peer_nids 2 1
}
run_test 18b "Delete multiple nids from multiple peers using import --del"
test_19() {
reinit_dlc || return $?
cat <<EOF > $TMP/sanity-lnet-$testnum-expected.yaml
peer:
- primary nid: 19@gni
Multi-Rail: True
peer ni:
- nid: 19@gni
EOF
append_global_yaml
compare_peer_add "19@gni"
}
run_test 19 "Add peer with single nid (gni)"
test_20() {
reinit_dlc || return $?
cat <<EOF > $TMP/sanity-lnet-$testnum-expected.yaml
peer:
- primary nid: 20@gni
Multi-Rail: True
peer ni:
- nid: 20@gni
- nid: 20.20.20.20@tcp
- nid: 20.20.20.20@o2ib
EOF
append_global_yaml
compare_peer_add "20@gni" "20.20.20.20@tcp,20.20.20.20@o2ib"
}
run_test 20 "Add peer with gni primary and tcp, o2ib secondary"
test_21() {
reinit_dlc || return $?
cat <<EOF > $TMP/sanity-lnet-$testnum-expected.yaml
peer:
- primary nid: 21@gni
Multi-Rail: True
peer ni:
- nid: 21@gni
- nid: 22@gni
- nid: 23@gni
- nid: 24@gni
- nid: 25@gni
EOF
append_global_yaml
echo "Add peer with nidrange (gni)"
compare_peer_add "21@gni" "[22-25]@gni" || error
echo "Add peer with nidrange that overlaps primary nid (gni)"
compare_peer_add "21@gni" "[21-25]@gni"
}
run_test 21 "Add peer with nidrange (gni)"
test_22() {
reinit_dlc || return $?
cat <<EOF > $TMP/sanity-lnet-$testnum-expected.yaml
peer:
- primary nid: 22@gni
Multi-Rail: True
peer ni:
- nid: 22@gni
- nid: 24@gni
- nid: 25@gni
- nid: 27@gni
- nid: 28@gni
- nid: 29@gni
EOF
append_global_yaml
do_lnetctl peer add --prim_nid 22@gni --nid [24-29]@gni ||
error "Peer add failed $?"
compare_peer_del "22@gni" "26@gni"
}
run_test 22 "Delete single secondary nid from peer (gni)"
test_23() {
reinit_dlc || return $?
cat <<EOF > $TMP/sanity-lnet-$testnum-expected.yaml
peer:
- primary nid: 23@gni
Multi-Rail: True
peer ni:
- nid: 23@gni
EOF
append_global_yaml
do_lnetctl peer add --prim_nid 23@gni --nid [25-29]@gni ||
error "Peer add failed $?"
compare_peer_del "23@gni" "[25-29]@gni"
}
run_test 23 "Delete all secondary nids from peer (gni)"
test_24() {
reinit_dlc || return $?
cat <<EOF > $TMP/sanity-lnet-$testnum-expected.yaml
peer:
- primary nid: 24@gni
Multi-Rail: True
peer ni:
- nid: 24@gni
- nid: 11@gni
- nid: 13.13.13.13@o2ib
- nid: 14.13.13.13@o2ib
- nid: 14.15.13.13@o2ib
- nid: 15.17.1.5@tcp
- nid: 15.17.1.10@tcp
- nid: 15.17.1.20@tcp
EOF
append_global_yaml
do_lnetctl peer add --prim_nid 24@gni \
--nid [13-14/1].[13-15/2].13.13@o2ib,[15-16/3].[17-19/4].[1].[5-20/5]@tcp,[5-12/6]@gni ||
error "Peer add failed $?"
compare_peer_del "24@gni" "5@gni,13.15.13.13@o2ib,15.17.1.15@tcp"
}
run_test 24 "Delete a secondary nid from peer (tcp, o2ib and gni)"
test_25() {
reinit_dlc || return $?
cat <<EOF > $TMP/sanity-lnet-$testnum-expected.yaml
peer:
- primary nid: 25@gni
Multi-Rail: True
peer ni:
- nid: 25@gni
EOF
append_global_yaml
do_lnetctl peer add --prim_nid 25@gni \
--nid [26-27].[4-10/3].26.26@tcp,26.26.26.26@o2ib,[30-35]@gni ||
error "Peer add failed $?"
compare_peer_del "25@gni" \
"[26-27].[4-10/3].26.26@tcp,26.26.26.26@o2ib,[30-35]@gni"
}
run_test 25 "Delete all secondary nids from peer (tcp, gni and o2ib)"
test_26() {
reinit_dlc || return $?
do_lnetctl peer add --prim_nid 1.1.1.1@tcp --lock_prim ||
error "Peer add with --lock_prim option failed $?"
local peer_state=$($LNETCTL peer show -v 4 --nid 1.1.1.1@tcp |
awk '/peer state/ {print $NF}')
# This relies on the following peer state definition:
# #define LNET_PEER_LOCK_PRIMARY BIT(20)
if ((!("$peer_state" & (1 << 20)))); then
error "Peer state does not have 'locked' bit set: $peer_state"
fi
do_lnetctl peer del --prim_nid 1.1.1.1@tcp ||
error "Peer del failed $?"
$LNETCTL peer show --nid 1.1.1.1@tcp | grep -q 1.1.1.1@tcp ||
error "1.1.1.1@tcp is not listed"
do_lnetctl peer del --prim_nid 1.1.1.1@tcp --force ||
error "Peer del --force failed $?"
do_lnetctl peer show --nid 1.1.1.1@tcp &&
error "failed to delete 1.1.1.1@tcp"
return 0
}
run_test 26 "Delete peer with primary nid locked"
test_27() {
reinit_dlc || return $?
local tmp=$(mktemp)
echo "foobar:" > $tmp
do_lnetctl import $tmp 2>&1 | grep -q "invalid key" ||
error "Expected import to detect invalid key"
cat ${GLOBAL_YAML_FILE} > $tmp
grep -q discovery $tmp ||
error "Expect discovery param in global params"
sed -i 's/discovery:.*/discovery:/' $tmp ||
error "Failed to edit $tmp rc $?"
do_lnetctl import $tmp 2>&1 | grep -q "no value" ||
error "Expected import to detect missing value"
sed -i 's/discovery:/discovery: foo/' $tmp ||
error "Failed to edit $tmp rc $?"
do_lnetctl import $tmp 2>&1 | grep -q "invalid value" ||
error "Expected import to detect invalid value"
rm -f $tmp
}
run_test 27 "Import bad config should fail gracefully"
test_28() {
reinit_dlc || return $?
do_lnetctl peer add --prim_nid 1.1.1.1@tcp --ni 7.7.7.7@tcp ||
error "First peer add failed $?"
do_lnetctl peer add --prim_nid 1.1.1.2@tcp --nid 7.7.7.8@tcp ||
error "Second peer add failed $?"
count=$(do_lnetctl peer list | awk '/-\s+nid:/{print $NF}' | wc -l)
[[ $count -eq 2 ]] || error "wrong number of peers reported"
}
run_test 28 "Test peer_list"
add_net() {
local net="$1"
local if="$2"
do_lnetctl net add --net ${net} --if ${if} ||
error "Failed to add net ${net} on if ${if}"
}
test_40() {
reinit_dlc || return $?
local lock_prim
lock_prim=$(cat /sys/module/lnet/parameters/lock_prim_nid 2>/dev/null)
[[ "$lock_prim" == "1" ]] || skip "Need lock_prim_nid=1"
add_net "${NETTYPE}" "${INTERFACES[0]}" || return $?
# A reachable NID. A health monitor pings it before the filesystem
# is mounted, so a peer record for it already exists, and is up to
# date, by the time the config llog calls LNetAddPeer().
local secondary=$($LCTL list_nids | head -n 1)
# The primary NID from the config. It lives on a network with no
# local interface, so it never appears in the secondary's ping
# reply, the same as a primary NID whose interface is down.
local primary="${secondary%@*}@${NETTYPE}99"
local uuid="lnet_add_peer_uuid"
# Health monitor discovers the reachable NID first.
do_lnetctl discover ${secondary} ||
error "Failed to discover $secondary"
(($($LNETCTL peer show --nid ${secondary} |
grep -cFw ${primary}) == 0)) ||
error "$primary present before LNetAddPeer"
# Need Lustre modules for add_uuid
load_modules_local ||
error "Failed to load modules rc=$?"
# The config llog associates both NIDs with one peer via
# LNetAddPeer(). The primary NID is added first and locked.
$LCTL add_uuid ${uuid} ${primary} ||
error "add_uuid $primary failed $?"
$LCTL add_uuid ${uuid} ${secondary} ||
error "add_uuid $secondary failed $?"
# The pre-existing, up to date peer for the secondary NID must
# still be merged into the primary peer.
local show="$LNETCTL peer show --nid ${primary} 2>/dev/null"
wait_update $HOSTNAME "$show | grep -cFw ${secondary}" "1" "20"
local rc=$?
$LCTL del_uuid ${uuid} || error "del_uuid failed $?"
((rc == 0)) || error "$secondary was not merged into peer $primary"
unload_modules_local
}
run_test 40 "LNetAddPeer merges a pre-existing non-primary peer NID"
test_50() {
reinit_dlc || return $?
local param
for param in alive_router_check_interval router_ping_timeout; do
echo 1 > /sys/module/lnet/parameters/$param
done
local tyaml=$TMP/sanity-lnet-$testnum.yaml
cat <<EOF > $tyaml
routing:
enable: 1
EOF
echo "Enable routing"
do_lnetctl import $tyaml ||
error "Import failed rc = $?"
# Get default buffer sizes for later
local tiny=$($LNETCTL export --backup |
awk '/\s+tiny:/{print $NF}')
local small=$($LNETCTL export --backup |
awk '/\s+small:/{print $NF}')
local large=$($LNETCTL export --backup |
awk '/\s+large:/{print $NF}')
wait_update $HOSTNAME \
"$LNETCTL routing show | grep enable | grep -c 1" \
"1" "10"
(($? == 0)) ||
error "Routing not enabled"
cat <<EOF > $tyaml
routing:
enable: 0
EOF
echo "Disable routing"
do_lnetctl import $tyaml ||
error "Import failed rc = $?"
wait_update $HOSTNAME \
"$LNETCTL routing show | grep enable | grep -c 0" \
"1" "10"
(($? == 0)) ||
error "Routing not disabled"
local ncpt=$($LCTL get_param -n cpu_partition_table | wc -l)
tiny=$((tiny + ncpt))
small=$((small + ncpt))
large=$((large + ncpt))
cat <<EOF > $tyaml
routing:
enable: 1
buffers:
tiny: $tiny
small: $small
large: $large
EOF
echo "Enable routing w/custom buffer sizses"
do_lnetctl import $tyaml ||
error "Import failed rc = $?"
cat /sys/module/lnet/parameters/*buffers
local tiny2=$($LNETCTL export --backup |
awk '/\s+tiny:/{print $NF}')
local small2=$($LNETCTL export --backup |
awk '/\s+small:/{print $NF}')
local large2=$($LNETCTL export --backup |
awk '/\s+large:/{print $NF}')
((tiny2 == tiny)) ||
error "Expect tiny buffers $tiny found $tiny2"
((small2 == small)) ||
error "Expect small buffers $small found $small2"
((large2 == large)) ||
error "Expect large buffers $large found $large2"
}
run_test 50 "Enable/disable routing via yaml import"
test_51() {
reinit_dlc || return $?
local range=$(cat /sys/module/lnet/parameters/lnet_numa_range)
local tyaml=$TMP/sanity-lnet-$testnum.yaml
((range++))
cat <<EOF > $tyaml
numa:
range: $range
EOF
do_lnetctl import $tyaml || error "Import failed rc = $?"
local range2=$(cat /sys/module/lnet/parameters/lnet_numa_range)
((range2 == range)) ||
error "Expect lnet_numa_range $range found $range2"
}
run_test 51 "Set numa range via yaml import"
test_99a() {
reinit_dlc || return $?
echo "Invalid prim_nid - peer add"
do_lnetctl peer add --prim_nid foobar &&
error "Command should have failed"
echo "Invalid prim_nid - peer del"
do_lnetctl peer del --prim_nid foobar &&
error "Command should have failed"
echo "Delete non-existing peer"
do_lnetctl peer del --prim_nid 1.1.1.1@o2ib &&
error "Command should have failed"
echo "Don't provide mandatory argument for peer del"
do_lnetctl peer del --nid 1.1.1.1@tcp &&
error "Command should have failed"
echo "Don't provide mandatory argument for peer add"
do_lnetctl peer add --nid 1.1.1.1@tcp &&
error "Command should have failed"
echo "Don't provide mandatory arguments peer add"
do_lnetctl peer add &&
error "Command should have failed"
echo "Invalid secondary nids"
do_lnetctl peer add --prim_nid 1.1.1.1@tcp --nid foobar &&
error "Command should have failed"
echo "Exceed max nids per peer"
do_lnetctl peer add --prim_nid 1.1.1.1@tcp --nid 1.1.1.[2-255]@tcp &&
error "Command should have failed"
echo "Invalid net type"
do_lnetctl peer add --prim_nid 1@foo &&
error "Command should have failed"
echo "Invalid nid format"
local invalid_nids="1@tcp 1@o2ib 1.1.1.1@gni"
local nid
for nid in ${invalid_nids}; do
echo "Check invalid primary nid - '$nid'"
do_lnetctl peer add --prim_nid $nid &&
error "Command should have failed"
done
local invalid_strs="[2-1]@gni [a-f/x]@gni 256.256.256.256@tcp"
invalid_strs+=" 1.1.1.1.[2-5/f]@tcp 1.]2[.3.4@o2ib"
invalid_strs+="1.[2-4,[5-6],7-8].1.1@tcp foobar"
local nidstr
for nidstr in ${invalid_strs}; do
echo "Check invalid nidstring - '$nidstr'"
do_lnetctl peer add --prim_nid 1.1.1.1@tcp --nid $nidstr &&
error "Command should have failed"
done
echo "Add non-local gateway"
do_lnetctl route add --net tcp --gateway 1@gni &&
error "Command should have failed"
return 0
}
run_test 99a "Check various invalid inputs to lnetctl peer"
test_99b() {
reinit_dlc || return $?
create_base_yaml_file
cat <<EOF > $TMP/sanity-lnet-$testnum-invalid.yaml
peer:
- primary nid: 99.99.99.99@tcp
Multi-Rail: Foobar
peer ni:
- nid: 99.99.99.99@tcp
EOF
do_lnetctl import < $TMP/sanity-lnet-$testnum-invalid.yaml &&
error "import should have failed"
$LNETCTL export --backup > $TMP/sanity-lnet-$testnum-actual.yaml
compare_yaml_files
}
run_test 99b "Invalid value for Multi-Rail in yaml import"
del_net() {
local net="$1"
local if="$2"
do_lnetctl net del --net ${net} --if ${if} ||
error "Failed to del net ${net} on if ${if}"
}
compare_route_add() {
local rnet="$1"
local gw="$2"
local actual="$TMP/sanity-lnet-$testnum-actual.yaml"
do_lnetctl route add --net ${rnet} --gateway ${gw} ||
error "route add failed $?"
echo "$LNETCTL export --backup > $actual"
$LNETCTL export --backup > $actual ||
error "export failed $?"
validate_gateway_nids
return $?
}
append_net_tunables() {
local net=${1:-tcp}
$LNETCTL net show -v --net ${net} | grep -v 'dev cpt' |
awk '/^\s+tunables:$/,/^\s+CPT:/' >> $TMP/sanity-lnet-$testnum-expected.yaml
}
ROUTERS=()
declare -A ROUTER_INTERFACES
RPEERS=()
declare -A RPEER_INTERFACES
NTRB=16
init_router_test_vars() {
local routers_required="$1"
local rpeers_required="$2"
local rnodes_required
((rnodes_required=routers_required+rpeers_required))
# all remote nodes, including some that may not be used
local rnodes_all=( $(remote_nodes_list | tr ',' ' ') )
[[ -z $rnodes_all || "${#rnodes_all[@]}" -lt $rnodes_required ]] &&
skip "Need at least $rnodes_required remote nodes" \
"found \"${rnodes_all[@]}\""
ROUTERS=( "${rnodes_all[@]:0:${routers_required}}" )
RPEERS=( "${rnodes_all[@]:${routers_required}:${rpeers_required}}" )
local rnodes=$(comma_list ${ROUTERS[@]} ${RPEERS[@]})
local all_nodes=$(comma_list ${ROUTERS[@]} ${RPEERS[@]} $HOSTNAME)
do_nodes $rnodes $LUSTRE_RMMOD ||
error "failed to unload modules"
do_rpc_nodes $rnodes "load_lnet" ||
error "Failed to load LNet"
do_nodes $rnodes "$LNETCTL lnet configure $LNET_CONFIG_INIT_OPT" ||
error "Failed to configure LNet on $rnodes rc = $?"
for router in ${ROUTERS[@]}; do
ROUTER_INTERFACES[$router]=$(do_rpc_nodes --quiet \
$router lnet_if_list)
done
for rpeer in ${RPEERS[@]}; do
RPEER_INTERFACES[$rpeer]=$(do_rpc_nodes --quiet \
$rpeer lnet_if_list)
done
# test_256 needs tiny_router_buffers > (# CPTs * peer credits)
local rtr_list=$(comma_list ${ROUTERS[@]})
local max_ncpt=$(do_nodes $rtr_list "$LCTL get_param -n cpu_partition_table | wc -l" |
sort -ug | tail -n 1)
local max_pcs=$(do_nodes $rtr_list "$LNETCTL net show -v" |
awk '/peer_credits:/{print $NF}' |
sort -ug | tail -n 1)
((NTRB=max_ncpt*(max_pcs+1)))
do_nodes $all_nodes $LUSTRE_RMMOD ||
error "Failed to unload modules"
[[ ${#INTERFACES[@]} -eq 0 ]] &&
error "No interfaces configured for local host $HOSTNAME"
for router in ${!ROUTER_INTERFACES[@]}; do
[[ -z "${ROUTER_INTERFACES[$router]}" ]] &&
error "No interfaces configured for router $router"
done
for rpeer in ${!RPEER_INTERFACES[@]}; do
[[ -z "${RPEER_INTERFACES[$rpeer]}" ]] &&
error "No interfaces configured for remote peer $rpeer"
done
return 0
}
do_net_add() {
local node=$1
local net=$2
local if=$3
do_rpc_nodes $node "$LNETCTL net add --net $net --if $if" ||
error "add $net on interface $if on node $node failed rc=$?"
}
do_route_add() {
local node=$1
local net=$2
local gw=$3
do_node $node "$LNETCTL route add --net $net --gateway $gw" ||
error "route add to $net via $gw failed rc=$?"
}
declare -A ROUTER_NIDS
declare -A RPEER_NIDS
LNIDS=()
LOCAL_NET=${NETTYPE}
LOCAL_NET2=${NETTYPE}2
REMOTE_NET=${NETTYPE}1
REMOTE_NET2=${NETTYPE}3
setup_router_test() {
(( $MDS1_VERSION >= $(version_code 2.15.0) )) ||
skip "need at least 2.15.0 for load_lnet"
local routers_required=1
local rpeers_required=1
local mr_peers=false
local flag
while getopts "mp:r:" flag; do
case $flag in
m) mr_peers=true;;
p) rpeers_required="$OPTARG";;
r) routers_required="$OPTARG";;
*) ;;
esac
done
shift $((OPTIND - 1))
local mod_opts="$@"
if [[ ${#RPEER_INTERFACES[@]} -eq 0 ]]; then
init_router_test_vars $routers_required $rpeers_required ||
return $?
fi
local all_nodes=$(comma_list ${ROUTERS[@]} ${RPEERS[@]} $HOSTNAME)
do_nodes $all_nodes $LUSTRE_RMMOD ||
error "failed to unload modules"
mod_opts+=" alive_router_check_interval=5"
mod_opts+=" router_ping_timeout=5"
mod_opts+=" large_router_buffers=4"
mod_opts+=" small_router_buffers=8"
mod_opts+=" tiny_router_buffers=$NTRB"
do_rpc_nodes $all_nodes load_lnet "${mod_opts}" ||
error "Failed to load lnet"
do_nodes $all_nodes "$LNETCTL lnet configure $LNET_CONFIG_OPT" ||
error "Failed to initialize DLC"
for router in ${!ROUTER_INTERFACES[@]}; do
local router_interfaces=( ${ROUTER_INTERFACES[$router]} )
do_net_add $router $LOCAL_NET ${router_interfaces[0]} ||
return $?
do_net_add $router $REMOTE_NET ${router_interfaces[0]} ||
return $?
if $mr_peers; then
do_net_add $router $LOCAL_NET2 ${router_interfaces[0]} ||
return $?
do_net_add $router $REMOTE_NET2 ${router_interfaces[0]} ||
return $?
fi
done
for rpeer in ${!RPEER_INTERFACES[@]}; do
local rpeer_interfaces=( ${RPEER_INTERFACES[$rpeer]} )
do_net_add $rpeer $REMOTE_NET ${rpeer_interfaces[0]} ||
return $?
if $mr_peers; then
do_net_add $rpeer $REMOTE_NET2 ${rpeer_interfaces[0]} ||
return $?
fi
done
add_net $LOCAL_NET ${INTERFACES[0]} ||
return $?
if $mr_peers; then
add_net $LOCAL_NET2 ${INTERFACES[0]} ||
return $?
fi
for router in ${!ROUTER_INTERFACES[@]}; do
ROUTER_NIDS[$router]=$(do_node $router $LCTL list_nids \
2>/dev/null | xargs echo)
echo "router: $router nids: ${ROUTER_NIDS[$router]}"
done
for rpeer in ${!RPEER_INTERFACES[@]}; do
RPEER_NIDS[$rpeer]=$(do_node $rpeer $LCTL list_nids \
2>/dev/null | xargs echo)
echo "rpeer: $rpeer nids: ${RPEER_NIDS[$rpeer]}"
done
LNIDS=( $($LCTL list_nids 2>/dev/null | xargs echo) )
echo "local: $HOSTNAME nids: ${LNIDS[@]}"
}
do_route_del() {
local node=$1
local net=$2
local gw=$3
do_nodesv $node \
"output=\\\"\\\$($LNETCTL route show --net $net --gateway $gw 2>/dev/null)\\\"; \
if [[ -n \\\"\\\${output}\\\" ]]; then \
echo \\\"Delete route to $net via $gw\\\"; \
$LNETCTL route del --net $net --gateway $gw; \
else \
exit 0; \
fi"
}
cleanup_router_test() {
local all_nodes=$(comma_list $HOSTNAME ${ROUTERS[@]} ${RPEERS[@]})
for router in ${!ROUTER_NIDS[@]}; do
local router_nids=( ${ROUTER_NIDS[$router]} )
do_route_del $HOSTNAME $REMOTE_NET ${router_nids[0]} ||
error "Failed to delete $HOSTNAME -> "\
"$REMOTE_NET via ${router_nids[0]} route"
done
for router in ${!ROUTER_INTERFACES[@]}; do
local router_nids=( ${ROUTER_NIDS[$router]} )
do_route_del $rpeer $LOCAL_NET ${router_nids[1]} ||
error "Failed to delete $rpeer -> "\
"$LOCAL_NET via ${router_nids[1]} route"
done
do_nodes $all_nodes $LUSTRE_RMMOD ||
error "failed to unload modules"
return 0
}
test_100() {
define_global_yaml
setup_router_test || return $?
local router_nids=( ${ROUTER_NIDS[${ROUTERS[0]}]} )
cat <<EOF > $TMP/sanity-lnet-$testnum-expected.yaml
net:
- net type: $LOCAL_NET
local NI(s):
- interfaces:
0: ${INTERFACES[0]}
EOF
append_net_tunables $LOCAL_NET
cat <<EOF >> $TMP/sanity-lnet-$testnum-expected.yaml
route:
- net: $REMOTE_NET
gateway: ${router_nids[0]}
hop: -1
priority: 0
peer:
- primary nid: ${router_nids[0]}
Multi-Rail: False
peer ni:
- nid: ${router_nids[0]}
EOF
append_global_yaml
compare_route_add "$REMOTE_NET" "${router_nids[0]}"
cleanup_router_test
}
run_test 100 "Add route with single gw"
ARR_IF0_IP=($(ip -o -4 a s ${INTERFACES[0]} |
awk '{print $4}' | sed 's/\/.*//'))
echo "Total IP for ${INTERFACES[0]} found [${#ARR_IF0_IP[@]}]"
echo "Interface:IP are"
for i in ${ARR_IF0_IP[@]}; do
echo "${INTERFACES[0]}:$i"
done
IF0_IP=${ARR_IF0_IP[0]}
IF0_NET=$(awk -F. '{print $1"."$2"."$3}'<<<"${IF0_IP}")
IF0_HOSTNUM=$(awk -F. '{print $4}'<<<"${IF0_IP}")
if (((IF0_HOSTNUM + 5) > 254)); then
GW_HOSTNUM=1
else
GW_HOSTNUM=$((IF0_HOSTNUM + 1))
fi
GW_NID="${IF0_NET}.${GW_HOSTNUM}@${NETTYPE}"
echo "Using GW_NID:$GW_NID"
test_101() {
[[ -n $ARR_IF0_IP ]] || skip "Need IPv4 NIDs"
reinit_dlc || return $?
add_net "${NETTYPE}" "${INTERFACES[0]}"
cat <<EOF > $TMP/sanity-lnet-$testnum-expected.yaml
net:
- net type: ${NETTYPE}
local NI(s):
- interfaces:
0: ${INTERFACES[0]}
EOF
append_net_tunables ${NETTYPE}
echo "route:" >> $TMP/sanity-lnet-$testnum-expected.yaml
for i in $(seq $GW_HOSTNUM $((GW_HOSTNUM + 4))); do
cat <<EOF >> $TMP/sanity-lnet-$testnum-expected.yaml
- net: ${REMOTE_NET}
gateway: ${IF0_NET}.${i}@${NETTYPE}
hop: -1
priority: 0
EOF
done
echo "peer:" >> $TMP/sanity-lnet-$testnum-expected.yaml
for i in $(seq $GW_HOSTNUM $((GW_HOSTNUM + 4))); do
cat <<EOF >> $TMP/sanity-lnet-$testnum-expected.yaml
- primary nid: ${IF0_NET}.${i}@${NETTYPE}
Multi-Rail: False
peer ni:
- nid: ${IF0_NET}.${i}@${NETTYPE}
EOF
done
append_global_yaml
local gw="${IF0_NET}.[$GW_HOSTNUM-$((GW_HOSTNUM + 4))]@${NETTYPE}"
compare_route_add "${REMOTE_NET}" "${gw}"
}
run_test 101 "Add route with multiple gw"
compare_route_del() {
local rnet="$1"
local gw="$2"
local actual="$TMP/sanity-lnet-$testnum-actual.yaml"
do_lnetctl route del --net ${rnet} --gateway ${gw} ||
error "route del failed $?"
$LNETCTL export --backup > $actual ||
error "export failed $?"
validate_gateway_nids
}
test_102() {
define_global_yaml
setup_router_test || return $?
local router_nids=( ${ROUTER_NIDS[${ROUTERS[0]}]} )
$LNETCTL export --backup > $TMP/sanity-lnet-$testnum-expected.yaml
do_lnetctl route add --net $REMOTE_NET --gateway ${router_nids[0]} ||
error "route add failed $?"
compare_route_del "$REMOTE_NET" "${router_nids[0]}"
cleanup_router_test
}
run_test 102 "Delete route with single gw"
NUM_NID_EXPR='[103-120/4]'
test_103() {
[[ -n $ARR_IF0_IP ]] || skip "Need IPv4 NIDs"
reinit_dlc || return $?
add_net "${NETTYPE}" "${INTERFACES[0]}"
$LNETCTL export --backup > $TMP/sanity-lnet-$testnum-expected.yaml
local nid_expr
if [[ $NETTYPE =~ (tcp|o2ib)[0-9]* ]]; then
nid_expr="${IF0_NET}.[$GW_HOSTNUM-$((GW_HOSTNUM+5))/2]"
else
nid_expr="${NUM_NID_EXPR}"
fi
do_lnetctl route add --net ${NETTYPE}103 \
--gateway ${nid_expr}@${NETTYPE} ||
error "route add failed $?"
compare_route_del "${NETTYPE}103" "${nid_expr}@${NETTYPE}"
}
run_test 103 "Delete route with multiple gw"
test_104() {
local tyaml="$TMP/sanity-lnet-$testnum-expected.yaml"
reinit_dlc || return $?
# Default value is '3'
local val=$($LNETCTL global show | awk '/response_tracking/{print $NF}')
[[ $val -ne 3 ]] &&
error "Expect 3 found $val"
echo "Set < 0; Should fail"
do_lnetctl set response_tracking -1 &&
error "should have failed $?"
reinit_dlc || return $?
cat <<EOF > $tyaml
global:
response_tracking: -10
EOF
do_lnetctl import < $tyaml &&
error "should have failed $?"
echo "Check valid values; Should succeed"
local i
for ((i = 0; i < 4; i++)); do
reinit_dlc || return $?
do_lnetctl set response_tracking $i ||
error "should have succeeded $?"
$LNETCTL global show | grep -q "response_tracking: $i" ||
error "Failed to set response_tracking to $i"
reinit_dlc || return $?
cat <<EOF > $tyaml
global:
response_tracking: $i
EOF
do_lnetctl import < $tyaml ||
error "should have succeeded $?"
$LNETCTL global show | grep -q "response_tracking: $i" ||
error "Failed to set response_tracking to $i"
done
reinit_dlc || return $?
echo "Set > 3; Should fail"
do_lnetctl set response_tracking 4 &&
error "should have failed $?"
reinit_dlc || return $?
cat <<EOF > $tyaml
global:
response_tracking: 10
EOF
do_lnetctl import < $tyaml &&
error "should have failed $?"
return 0
}
run_test 104 "Set/check response_tracking param"
test_105() {
define_global_yaml
setup_router_test || return $?
local router_nids=( ${ROUTER_NIDS[${ROUTERS[0]}]} )
do_lnetctl route add --net $REMOTE_NET --gateway ${router_nids[0]} ||
error "route add failed $?"
do_lnetctl peer add --prim ${router_nids[0]} &&
error "peer add should fail"
cleanup_router_test
}
run_test 105 "Adding duplicate GW peer should fail"
test_106() {
define_global_yaml
setup_router_test || return $?
local router_nids=( ${ROUTER_NIDS[${ROUTERS[0]}]} )
do_lnetctl route add --net $REMOTE_NET --gateway ${router_nids[0]} ||
error "route add failed $?"
do_lnetctl peer del --prim ${router_nids[0]} &&
error "peer del should fail"
cleanup_router_test
}
run_test 106 "Deleting GW peer should fail"
test_107() {
[[ ${NETTYPE} == tcp* ]] || skip "Need tcp NETTYPE"
cleanup_lnet || error "Failed to unload modules before test execution"
setup_fakeif || error "Failed to add fake IF"
reinit_dlc || return $?
add_net "tcp" "${INTERFACES[0]}" || return $?
add_net "tcp" "$FAKE_IF" || return $?
del_net "tcp" "$FAKE_IF" || return $?
cleanup_fakeif
cleanup_lnet
}
run_test 107 "Deleting extra interface doesn't crash node"
test_108() {
[[ ${NETTYPE} == tcp* ]] || skip "Need tcp NETTYPE"
cleanup_lnet || error "Failed to unload modules before test execution"
setup_fakeif || error "Failed to add fake IF"
reinit_dlc || return $?
add_net "tcp" "${INTERFACES[0]}" || return $?
$LNETCTL net show > $TMP/sanity-lnet-$testnum-expected.yaml
add_net "tcp" "$FAKE_IF" || return $?
cat <<EOF >> $TMP/sanity-lnet-$testnum-expected.yaml
- nid: $FAKE_NID
status: up
interfaces:
0: ${FAKE_IF}
EOF
$LNETCTL net show > $TMP/sanity-lnet-$testnum-actual.yaml
compare_yaml_files || error "not all interfaces were setup"
cleanup_fakeif
cleanup_lnet
return 0
}
run_test 108 "Check Multi-Rail setup"
test_109() {
[[ ${NETTYPE} == tcp* ]] || skip "Need tcp NETTYPE"
cleanup_lnet || error "Failed to unload modules before test execution"
setup_fakeif || error "Failed to add fake IF"
FAKE_IF_ALIAS="${FAKE_IF}"
FAKE_IF_ALIAS+=":0"
ifconfig "$FAKE_IF_ALIAS" "$FAKE_IP_ALIAS" up ||
error "Failed to add fake IF alias"
reinit_dlc || return $?
# add interface with longer name first
add_net "tcp" "$FAKE_IF_ALIAS" || return $?
add_net "tcp" "$FAKE_IF" || return $?
del_net "tcp" "$FAKE_IF" || return $?
del_net "tcp" "$FAKE_IF_ALIAS" || return $?
# add interface with shorter name first
add_net "tcp" "$FAKE_IF" || return $?
add_net "tcp" "$FAKE_IF_ALIAS" || return $?
ifconfig "$FAKE_IF_ALIAS" "$FAKE_IP_ALIAS" down ||
error "Failed to clean up fake IF alias"
cleanup_fakeif
cleanup_lnet
}
run_test 109 "Add NI using a network interface alias (LU-16859)"
test_110() {
[[ $NETTYPE =~ (tcp|o2ib) ]] || skip "Need tcp or o2ib NETTYPE"
reinit_dlc || return $?
add_net ${NETTYPE} ${INTERFACES[0]} || return $?
local nid=$($LCTL list_nids)
reinit_dlc || return $?
do_lnetctl net add --nid $nid ||
error "Failed to add net via nid \"$nid\""
$LNETCTL net show --net $NETTYPE | grep -q "nid: $nid" ||
error "Failed to configure $nid"
do_lnetctl net del --nid $nid ||
error "Failed to del net via nid \"$nid\""
$LNETCTL net show | grep -q "nid: $nid" &&
error "$nid should have been deleted"
cleanup_lnet
}
run_test 110 "Configure NI using lnetctl net add --nid"
test_111() {
[[ $(uname -r | grep "3.10") ]] &&
skip "Unsupported on RHEL7"
setup_router_test || return $?
local i
local gw=( ${ROUTER_NIDS[${ROUTERS[0]}]} )
for i in {2..500}
do
do_lnetctl route add -n ${REMOTE_NET}${i} -g ${gw[0]} ||
error "Failed to add route rc=$?"
done
local num_routes=$($LNETCTL route show 2>/dev/null | grep -c gateway)
(( num_routes == 499 )) ||
error "Expect 499 routes but lnetctl found $num_routes"
num_routes=$($LCTL show_route 2>/dev/null | grep -c net)
(( num_routes == 499 )) ||
error "Epect 499 routes but lctl found $num_routes"
cleanup_router_test
}
run_test 111 "Test many routes"
test_112() {
cleanup_lnet || error "Failed to unload modules before test execution"
setup_fakeif || error "Failed to add fake IF"
reinit_dlc || return $?
$LNETCTL net add --net tcp1 --if ${INTERFACES[0]}
$LNETCTL net add --nid ${FAKE_IP}@tcp2
local count=$($LNETCTL net show -v 3 2>/dev/null | grep -c "lnd tunables:")
(( count == 2 )) || error "missing lnd tunables"
cleanup_lnet
cleanup_fakeif
}
run_test 112 "multiple net configurations"
do_import_test() {
reinit_dlc || return $?
cat ${1:-"$TMP/sanity-lnet-$testnum-expected.yaml"}
do_lnetctl import ${1:-"$TMP/sanity-lnet-$testnum-expected.yaml"} ||
error "Import failed with rc = $?"
$LNETCTL export --backup > $TMP/sanity-lnet-$testnum-actual.yaml
compare_yaml_files || return $?
}
test_150() {
reinit_dlc || return $?
add_net "${NETTYPE}" "${INTERFACES[0]}" || return $?
$LNETCTL export --backup > $TMP/sanity-lnet-$testnum-expected.yaml
cat <<EOF > $TMP/sanity-lnet.yaml
ip2nets:
- net-spec: ${NETTYPE}
interfaces:
0: ${INTERFACES[0]}
ip-range:
0: $(ip -o -4 a s ${INTERFACES[0]} | awk '{print $4}' | sed 's/\/.*//')
EOF
do_import_test "$TMP/sanity-lnet.yaml" || return $?
}
run_test 150 "Check import of ip2nets yaml sequence"
test_151() {
[[ $NETTYPE == tcp* ]] || skip "Need tcp nettype"
cleanup_lnet || error "Failed to unload modules before test execution"
setup_fakeif || error "Failed to add fake IF"
reinit_dlc || return $?
add_net "${NETTYPE}" "${INTERFACES[0]}" || return $?
add_net "${NETTYPE}" "${FAKE_IF}" || return $?
$LNETCTL export --backup > $TMP/sanity-lnet-$testnum-expected.yaml
local if0_ip=$(ip -o -4 a s ${INTERFACES[0]} | awk '{print $4}' |
sed 's/\/.*//')
echo "Check import of ip2nets with two matching interfaces"
cat <<EOF > $TMP/sanity-lnet-$testnum.yaml
ip2nets:
- net-spec: ${NETTYPE}
interfaces:
0: ${INTERFACES[0]}
1: ${FAKE_IF}
ip-range:
0: ${if0_ip}
1: ${FAKE_IP}
EOF
do_import_test "$TMP/sanity-lnet-$testnum.yaml" || return $?
echo "Additional rules for same net should not change config"
cat <<EOF >> $TMP/sanity-lnet-$testnum.yaml
- net-spec: ${NETTYPE}
interfaces:
0: ${INTERFACES[0]}
- net-spec: ${NETTYPE}
interfaces:
0: ${FAKE_IF}
ip-range:
0: ${FAKE_IP}
EOF
do_import_test "$TMP/sanity-lnet-$testnum.yaml" || return $?
add_net "${NETTYPE}2" "${INTERFACES[0]}"
$LNETCTL export --backup > $TMP/sanity-lnet-$testnum-expected.yaml
echo "Additional rule for second net should apply"
cat <<EOF >> $TMP/sanity-lnet-$testnum.yaml
- net-spec: ${NETTYPE}2
interfaces:
0: ${INTERFACES[0]}
EOF
do_import_test "$TMP/sanity-lnet-$testnum.yaml" || return $?
cleanup_fakeif
}
run_test 151 "Check correct application of multiple ip2nets rules"
test_152() {
[[ $NETTYPE == tcp* ]] || skip "Need tcp nettype"
cleanup_lnet || error "Failed to unload modules before test execution"
reinit_dlc || return $?
add_net "${NETTYPE}" "${INTERFACES[0]}" || return $?
$LNETCTL export --backup > $TMP/sanity-lnet-$testnum-expected.yaml
local if0_ip=$(ip -o -4 a s ${INTERFACES[0]} | awk '{print $4}' |
sed 's/\/.*//')
local i range
echo "Check import of ip2nets with wildcard IP patterns"
for i in {3..1}; do
range=$(echo "$if0_ip" | cut -d. -f1-$i)
for ((j = 3; j >= $i; j--)); do
range="${range:+$range.}*"
done
cat <<EOF > $TMP/sanity-lnet-$testnum.yaml
ip2nets:
- net-spec: ${NETTYPE}
interfaces:
0: ${INTERFACES[0]}
ip-range:
0: ${range}
EOF
do_import_test "$TMP/sanity-lnet-$testnum.yaml" || return $?
done
setup_fakeif || error "Failed to add fake IF"
echo "Check match-all expression mask later rules"
cat <<EOF > $TMP/sanity-lnet-$testnum.yaml
ip2nets:
- net-spec: ${NETTYPE}
interfaces:
0: ${INTERFACES[0]}
ip-range:
0: "*.*.*.*"
- net-spec: ${NETTYPE}
interfaces:
0: ${INTERFACES[0]}
1: ${FAKE_IF}
- net-spec: ${NETTYPE}
interfaces:
0: ${INTERFACES[0]}
1: ${FAKE_IF}
ip-range:
0: $if0_ip
1: $FAKE_IP
EOF
do_import_test "$TMP/sanity-lnet-$testnum.yaml" || return $?
cleanup_fakeif
}
run_test 152 "Check import of ip2nets with wildcard IP patterns"
create_ipv4_range() {
local stepped=${1:-false}
local if0_ip=$(ip -o -4 a s ${INTERFACES[0]} | awk '{print $4}' |
sed 's/\/.*//')
local if0_net=$(echo "$if0_ip" | cut -d. -f1-3)
local if0_host=$(echo "$if0_ip" | cut -d. -f4)
local range_start=$((if0_host - 4 < 0 ? 0 : if0_host - 4))
local range_end=$((if0_host + 4 > 255 ? 255 : if0_host + 4))
if ${stepped}; then
(((if0_host - range_start) % 2 == 0)) || range_start=${if0_host}
echo "${if0_net}.[${range_start}-${range_end}/2]"
else
echo "${if0_net}.[${range_start}-${range_end}]"
fi
}
test_153() {
[[ $NETTYPE == tcp* ]] || skip "Need tcp nettype"
cleanup_lnet || error "Failed to unload modules before test execution"
reinit_dlc || return $?
add_net "${NETTYPE}" "${INTERFACES[0]}" || return $?
$LNETCTL export --backup > $TMP/sanity-lnet-$testnum-expected.yaml
local ip_range=$(create_ipv4_range)
echo "Check import of ip2nets with IP address ranges"
cat <<EOF > $TMP/sanity-lnet-$testnum.yaml
ip2nets:
- net-spec: ${NETTYPE}
interfaces:
0: ${INTERFACES[0]}
ip-range:
0: ${ip_range}
EOF
do_import_test "$TMP/sanity-lnet-$testnum.yaml" || return $?
}
run_test 153 "Check import of ip2nets with IP address ranges"
test_154() {
[[ $NETTYPE == tcp* ]] || skip "Need tcp nettype"
cleanup_lnet || error "Failed to unload modules before test execution"
setup_fakeif || error "Failed to add fake IF"
reinit_dlc || return $?
add_net "${NETTYPE}" "${INTERFACES[0]}" || return $?
$LNETCTL export --backup > $TMP/sanity-lnet-$testnum-expected.yaml
local ip_range=$(create_ipv4_range true)
echo "Check import of ip2nets with stepped IP address ranges"
cat <<EOF > $TMP/sanity-lnet-$testnum.yaml
ip2nets:
- net-spec: ${NETTYPE}
interfaces:
0: ${INTERFACES[0]}
ip-range:
0: ${ip_range}
EOF
do_import_test "$TMP/sanity-lnet-$testnum.yaml" || return $?
cleanup_fakeif
}
run_test 154 "Check import of ip2nets with stepped IP address ranges"
test_155() {
[[ $NETTYPE == tcp* ]] || skip "Need tcp nettype"
reinit_dlc || return $?
local if0_ip=$(ip -o -4 a s ${INTERFACES[0]} | awk '{print $4}' |
sed 's/\/.*//')
echo "Check import with no matching rules (should fail)"
cat <<EOF > $TMP/sanity-lnet-$testnum.yaml
ip2nets:
- net-spec: ${NETTYPE}
interfaces:
0: ${INTERFACES[0]}
ip-range:
0: 10.20.30.40
EOF
! do_lnetctl import "$TMP/sanity-lnet-$testnum.yaml" ||
error "Import should have failed"
}
run_test 155 "Check ip2nets import failure with no matching rules"
test_156() {
[[ $NETTYPE == tcp* ]] || skip "Need tcp nettype"
cleanup_lnet || error "Failed to unload modules before test execution"
reinit_dlc || return $?
add_net "${NETTYPE}" "${INTERFACES[0]}" || return $?
$LNETCTL export --backup > $TMP/sanity-lnet-$testnum-expected.yaml
local if0_ip=$(ip -o -4 a s ${INTERFACES[0]} | awk '{print $4}' |
sed 's/\/.*//')
local if0_net3=$(echo "$if0_ip" | cut -d. -f1-3)
local if0_host=$(echo "$if0_ip" | cut -d. -f4)
local host2=$if0_host
local i
for i in {1..4}; do
((host2 + 2 < 255)) &&
host2=$((host2 + 2)) ||
host2=1
range=${range:+$range,}$host2
((i == 2)) && range=${range},$if0_host
done
echo "Check import of ip2nets with comma-separated IP list"
cat <<EOF > $TMP/sanity-lnet-$testnum.yaml
ip2nets:
- net-spec: ${NETTYPE}
interfaces:
0: ${INTERFACES[0]}
ip-range:
0: ${if0_net3}.[${range}]
EOF
do_import_test "$TMP/sanity-lnet-$testnum.yaml"
}
run_test 156 "Check import of ip2nets with comma-separated IP ranges"
test_157() {
reinit_dlc || return $?
echo "Check import failure with missing net-spec"
cat <<EOF > $TMP/sanity-lnet-$testnum.yaml
ip2nets:
- interfaces:
0: ${INTERFACES[0]}
ip-range:
0: 192.168.1.1
EOF
! do_lnetctl import "$TMP/sanity-lnet-$testnum.yaml" ||
error "Import should have failed with missing net-spec"
}
run_test 157 "Check import failure with malformed ip2nets YAML"
test_158() {
[[ $NETTYPE == tcp* ]] || skip "Need tcp nettype"
intf_has_ipv6 ${INTERFACES[0]} || skip "Interface has no IPv6"
reinit_dlc || return $?
add_net "${NETTYPE}" "${INTERFACES[0]}" || return $?
$LNETCTL export --backup > $TMP/sanity-lnet-$testnum-expected.yaml
local if0_ipv6=$(ip -o -6 a s ${INTERFACES[0]} | awk '{print $4}' |
grep -v '^fe80::' | head -n 1 | cut -d/ -f1)
echo "Check import of ip2nets with single IPv6 address"
cat <<EOF > $TMP/sanity-lnet-$testnum.yaml
ip2nets:
- net-spec: ${NETTYPE}
interfaces:
0: ${INTERFACES[0]}
ip-range:
0: ${if0_ipv6}
EOF
do_import_test "$TMP/sanity-lnet-$testnum.yaml" || return $?
}
run_test 158 "Check import of ip2nets with single IPv6 address"
test_159() {
[[ $NETTYPE == tcp* ]] || skip "Need tcp nettype"
intf_has_ipv6 ${INTERFACES[0]} || skip "Interface has no IPv6"
reinit_dlc || return $?
add_net "${NETTYPE}" "${INTERFACES[0]}" || return $?
$LNETCTL export --backup > $TMP/sanity-lnet-$testnum-expected.yaml
local if0_ipv6=$(ip -o -6 a s ${INTERFACES[0]} | awk '{print $4}' |
grep -v '^fe80::' | head -n 1 | cut -d/ -f1)
echo "Check import of ip2nets with IPv6 /128 CIDR"
cat <<EOF > $TMP/sanity-lnet-$testnum.yaml
ip2nets:
- net-spec: ${NETTYPE}
interfaces:
0: ${INTERFACES[0]}
ip-range:
0: ${if0_ipv6}/128
EOF
do_import_test "$TMP/sanity-lnet-$testnum.yaml" || return $?
}
run_test 159 "Check import of ip2nets with IPv6 /128 CIDR notation"
test_160() {
[[ $NETTYPE == tcp* ]] || skip "Need tcp nettype"
intf_has_ipv6 ${INTERFACES[0]} || skip "Interface has no IPv6"
cleanup_lnet || error "Failed to unload modules before test execution"
setup_fakeif || error "Failed to add fake IF"
reinit_dlc || return $?
add_net "${NETTYPE}" "${INTERFACES[0]}" || return $?
add_net "${NETTYPE}" "${FAKE_IF}" || return $?
$LNETCTL export --backup > $TMP/sanity-lnet-$testnum-expected.yaml
local if0_ipv6=$(ip -o -6 a s ${INTERFACES[0]} | awk '{print $4}' |
grep -v '^fe80::' | head -n 1 | cut -d/ -f1)
echo "Check import of ip2nets with two IPv6 interfaces"
cat <<EOF > $TMP/sanity-lnet-$testnum.yaml
ip2nets:
- net-spec: ${NETTYPE}
interfaces:
0: ${INTERFACES[0]}
1: ${FAKE_IF}
ip-range:
0: ${if0_ipv6}
1: ${FAKE_IPV6}
EOF
do_import_test "$TMP/sanity-lnet-$testnum.yaml" || return $?
echo "Additional rules for same net should not change config"
cat <<EOF >> $TMP/sanity-lnet-$testnum.yaml
- net-spec: ${NETTYPE}
interfaces:
0: ${INTERFACES[0]}
- net-spec: ${NETTYPE}
interfaces:
0: ${FAKE_IF}
ip-range:
0: ${FAKE_IPV6}
EOF
do_import_test "$TMP/sanity-lnet-$testnum.yaml" || return $?
add_net "${NETTYPE}2" "${INTERFACES[0]}"
$LNETCTL export --backup > $TMP/sanity-lnet-$testnum-expected.yaml
echo "Additional rule for second net should apply"
cat <<EOF >> $TMP/sanity-lnet-$testnum.yaml
- net-spec: ${NETTYPE}2
interfaces:
0: ${INTERFACES[0]}
ip-range:
0: ${if0_ipv6}
EOF
do_import_test "$TMP/sanity-lnet-$testnum.yaml" || return $?
cleanup_fakeif
}
run_test 160 "Check correct application of multiple IPv6 ip2nets rules"
test_161() {
[[ $NETTYPE == tcp* ]] || skip "Need tcp nettype"
intf_has_ipv6 ${INTERFACES[0]} || skip "Interface has no IPv6"
reinit_dlc || return $?
add_net "${NETTYPE}" "${INTERFACES[0]}" || return $?
$LNETCTL export --backup > $TMP/sanity-lnet-$testnum-expected.yaml
local if0_ipv6=$(ip -o -6 a s ${INTERFACES[0]} | awk '{print $4}' |
grep -v '^fe80::' | head -n 1)
local network_addr=$(echo "$if0_ipv6" | cut -d/ -f1)
local prefix_len=$(cut -d/ -f2<<<"$if0_ipv6")
[[ -n $network_addr && -n $prefix_len ]] ||
error "Failed to get net addr or prefix len"
echo "Check import of ip2nets with various IPv6 prefix lengths"
local i
for i in {1..4}; do
cat <<EOF > $TMP/sanity-lnet-$testnum.yaml
ip2nets:
- net-spec: ${NETTYPE}
interfaces:
0: ${INTERFACES[0]}
ip-range:
0: ${network_addr}/${prefix_len}
EOF
echo "Testing with /${prefix_len} prefix"
do_import_test "$TMP/sanity-lnet-$testnum.yaml" || return $?
prefix_len=$((prefix_len/2))
done
}
run_test 161 "Check import of ip2nets with various IPv6 prefix lengths"
test_162() {
[[ $NETTYPE == tcp* ]] || skip "Need tcp nettype"
intf_has_ipv6 ${INTERFACES[0]} || skip "Interface has no IPv6"
reinit_dlc || return $?
local if0_ipv6=$(ip -o -6 a s ${INTERFACES[0]} | awk '{print $4}' |
grep -v '^fe80::' | head -n 1 | sed 's,/[0-9]\+$,,')
echo "Check import with non-matching IPv6 address (should fail)"
cat <<EOF > $TMP/sanity-lnet-$testnum.yaml
ip2nets:
- net-spec: ${NETTYPE}
interfaces:
0: ${INTERFACES[0]}
ip-range:
0: 2001:db8:dead:beef::1
EOF
! do_lnetctl import "$TMP/sanity-lnet-$testnum.yaml" ||
error "Import should have failed with non-matching IPv6"
}
run_test 162 "Check ip2nets import failure with non-matching IPv6"
check_tunable_warning() {
dmesg | tac | sed '/LNet: Added LNI/q' | \
grep -q "All tunables of NIs of a net should be the same"
}
test_170() {
[[ ${NETTYPE} == tcp* ]] || skip "Test written for tcp"
cleanup_lnet || error "Failed to unload modules before test execution"
cleanup_netns || error "Failed to cleanup netns before test execution"
setup_fakeif || error "Failed to add fake IF"
reinit_dlc || return $?
# Disable console ratelimit
rlpath="/sys/module/libcfs/parameters/libcfs_console_ratelimit"
local rl=$(cat $rlpath)
echo 0 > $rlpath
local IMPORT="do_lnetctl import --old-api"
local expected="$TMP/sanity-lnet-$testnum-expected.yaml"
local actual="$TMP/sanity-lnet-$testnum-actual.yaml"
echo "Different CPTs"
cat <<EOF > $expected
net:
- net type: ${NETTYPE}
local NI(s):
- interfaces:
0: ${INTERFACES[0]}
CPT: "[0]"
- interfaces:
0: ${FAKE_IF}
CPT: "[1]"
EOF
$IMPORT < ${expected} || error "Import failed $?"
cat <<EOF > $expected
CPT:"[0]"
CPT:"[1]"
EOF
$LNETCTL export --backup | grep CPT: | tr -d ' ' > $actual
diff $expected $actual || error "Bad parsing"
reinit_dlc
echo "lnd_tunables not read if no tunables"
cat <<EOF > $expected
net:
- net type: ${NETTYPE}
local NI(s):
- interfaces:
0: ${INTERFACES[0]}
lnd tunables:
conns_per_peer: 5
tos: -1
EOF
$IMPORT < ${expected} || error "Import failed $?"
$LNETCTL export --backup | grep -q "conns_per_peer: 5" || error "Bad parsing"
reinit_dlc
echo "lnd_tunables cannot be different"
cat <<EOF > $expected
net:
- net type: ${NETTYPE}
local NI(s):
- interfaces:
0: ${INTERFACES[0]}
lnd tunables:
conns_per_peer: 5
tos: -1
- interfaces:
0: ${FAKE_IF}
lnd tunables:
conns_per_peer: 3
tos: -1
EOF
$IMPORT < ${expected} || error "Import failed $?"
$LNETCTL export --backup | grep -q "conns_per_peer: 5" || error "Bad parsing"
$LNETCTL export --backup | grep -q "conns_per_peer: 3" || error "Bad parsing"
reinit_dlc
echo "Wrong tunables (lnd_tunables) read from NI if first is empty"
cat <<EOF > $expected
net:
- net type: ${NETTYPE}
local NI(s):
- interfaces:
0: ${INTERFACES[0]}
- interfaces:
0: ${FAKE_IF}
tunables:
peer_timeout: 181
peer_credits: 8
peer_buffer_credits: 0
credits: 256
lnd tunables:
conns_per_peer: 3
tos: -1
EOF
$IMPORT < ${expected} || error "Import failed $?"
# conns_per_peer has default value for first ni and 3 for second
$LNETCTL export --backup | grep "conns_per_peer:" | grep -qv 3 || error "Bad parsing"
$LNETCTL export --backup | grep -q "conns_per_peer: 3" || error "Bad parsing"
# Previous ni got set tunables to default value
(( $($LNETCTL export --backup | grep -c "peer_timeout: 181") == 0 )) || \
error "Bad parsing"
check_tunable_warning || error "Missing warning"
reinit_dlc
echo "Tunables can be set directly in the net"
cat <<EOF > $expected
net:
- net type: ${NETTYPE}
local NI(s):
- interfaces:
0: ${INTERFACES[0]}
- interfaces:
0: ${FAKE_IF}
tunables:
peer_timeout: 182
peer_credits: 8
peer_buffer_credits: 0
credits: 256
lnd tunables:
conns_per_peer: 3
tos: -1
EOF
$IMPORT < ${expected} || error "Import failed $?"
(( $($LNETCTL export --backup | grep -c "peer_timeout: 182") == 2 )) ||
error "Bad parsing"
(( $($LNETCTL export --backup | grep -c "conns_per_peer: 3") == 2 )) ||
error "Bad parsing"
reinit_dlc
echo "Show warning if we try to set different tunables (not lnd_tunables) for NIs of a same net"
cat <<EOF > $expected
net:
- net type: ${NETTYPE}
local NI(s):
- interfaces:
0: ${INTERFACES[0]}
tunables:
peer_timeout: 182
peer_credits: 8
peer_buffer_credits: 0
credits: 256
- interfaces:
0: ${FAKE_IF}
tunables:
peer_timeout: 183
peer_credits: 8
peer_buffer_credits: 0
credits: 256
EOF
$IMPORT < ${expected} || error "Import failed $?"
# Tunables for the two nis will be the same, a warning will be emitted
(( $($LNETCTL export --backup | grep -c "peer_timeout: 182") == 2 )) ||
check_tunable_warning || error "Missing warning"
# Restore console_ratelimit
echo $rl > $rlpath
cleanup_lnet
}
run_test 170 "Check CPTs and tunables parsing when importing"
test_199() {
[[ ${NETTYPE} == tcp* || ${NETTYPE} == o2ib* ]] ||
skip "Need tcp or o2ib NETTYPE"
reinit_dlc || return $?
echo "Add interface ${INTERFACES[0]}"
do_lnetctl net add --net ${NETTYPE} --if ${INTERFACES[0]} ||
error "Failed to add net ${NETTYPE} with ${INTERFACES[0]}"
local ipaddress=$(ip --oneline addr show dev ${INTERFACES[0]} |
awk '/inet /{print $4}' |
sed 's:/.*::')
echo "Add IP address ${ipaddress} for interface ${INTERFACES[0]}"
do_lnetctl net add --ip2net "${NETTYPE} ${ipaddress}" || return 0
error "Failed to add net ${NETTYPE} with ${INTERFACES[0]}"
}
run_test 199 "load lnet w/o module option, configure interface twice"
test_200() {
[[ ${NETTYPE} == tcp* ]] ||
skip "Need tcp NETTYPE"
cleanup_lnet || return $?
setup_netns || error "setup_netns failed with $?"
load_lnet "networks=\"\""
ip a
ss -ltunp
do_ns ip a
do_ns ss -ltunp
lsmod | grep lnet
do_ns $LNETCTL lnet configure $LNET_CONFIG_INIT_OPT ||
error "Failed to configure LNet in non-default namespace rc = $?"
$LNETCTL net show --net tcp | grep -q "nid: $FAKE_NID$" ||
error "$FAKE_NID is not configured as expected"
cleanup_netns
}
run_test 200 "load lnet w/o module option, configure in a non-default namespace"
test_201() {
[[ ${NETTYPE} == tcp* ]] ||
skip "Need tcp NETTYPE"
cleanup_lnet || return $?
setup_netns || error "setup_netns failed with $?"
load_lnet "networks=tcp($FAKE_IF)"
do_ns $LNETCTL lnet configure $LNET_CONFIG_INIT_OPT ||
error "Failed to configure LNet in non-default namespace rc = $?"
$LNETCTL net show --net tcp | grep -q "nid: $FAKE_NID$" ||
error "$FAKE_NID is not configured as expected"
cleanup_netns
}
run_test 201 "load lnet using networks module options in a non-default namespace"
test_202() {
[[ ${NETTYPE} == tcp* ]] ||
skip "Need tcp NETTYPE"
(( $CLIENT_VERSION < $(version_code 2.17.53) )) ||
skip "kernel ip2nets deprecated in 2.17.53"
cleanup_lnet || return $?
setup_netns || error "setup_netns failed with $?"
load_lnet "networks=\"\" ip2nets=\"tcp0($FAKE_IF) ${FAKE_IP}\""
do_ns $LNETCTL lnet configure --all ||
error "Failed to configure LNet in non-default namespace rc = $?"
$LNETCTL net show | grep -q "nid: ${FAKE_IP}@tcp$" ||
error "$FAKE_IP@tcp is not configured as expected"
cleanup_netns
}
run_test 202 "load lnet using ip2nets in a non-default namespace"
### Add the interfaces in the target namespace
test_203() {
[[ ${NETTYPE} == tcp* ]] ||
skip "Need tcp NETTYPE"
cleanup_lnet || return $?
setup_netns || error "setup_netns failed with $?"
load_lnet
do_lnetctl lnet configure $LNET_CONFIG_OPT ||
error "Failed to configure LNet in non-default namespace rc = $?"
do_ns $LNETCTL net add --net tcp0 --if $FAKE_IF ||
error "Failed to add net in non-default namespace"
do_ns $LNETCTL net show | grep -q "nid: $FAKE_NID$" ||
error "$FAKE_NID is not configured as expected"
cleanup_netns
}
run_test 203 "add a network using an interface in the non-default namespace"
LNET_PARAMS_FILE="$TMP/$TESTSUITE.parameters"
function save_lnet_params() {
$LNETCTL global show | grep -E -v '^global:$' |
sed 's/://' > $LNET_PARAMS_FILE
}
function restore_lnet_params() {
local param value
while read param value; do
[[ $param == max_intf ]] && continue
[[ $param == lnd_timeout ]] && continue
do_lnetctl set ${param} ${value} ||
error "Failed to restore ${param} to ${value}"
done < $LNET_PARAMS_FILE
}
function set_ltt_node() {
# Achieves a desired LNet Transaction Timeout (LTT) value for a node by
# setting the LND timeout (LNDT) value for the network being used for
# tests.
local node=$1
local ltt=$2
local nettype=$3
if do_node $node $LNETCTL net set -h | grep -q -- "--lnd-timeout:"; then
# lnetctl supports setting the LNDT parameter.
local retry_count=$(do_node $node $LNETCTL global show |
awk '/retry_count/{print $NF}')
# Determine LNDT value to achieve LTT. This is taken from the
# the formula, using LNet retry count (LRC):
# LTT = LNDT(LRC + 1) + 1
# LNDT = (LTT - 1)/(LRC + 1)
local lnd_timeout=$(( (ltt - 1) / (retry_count + 1) ))
do_node $node $LNETCTL net set \
--net ${nettype} --lnd-timeout $lnd_timeout ||
error "Failed to set LND timeout on ${nettype} net"
fi
# Also set the default global LTT
do_node $node $LNETCTL set transaction_timeout $ltt ||
error "Failed to set transaction_timeout on $node"
}
function lnet_health_pre() {
save_lnet_params
# Lower transaction timeout to speed up test execution
set_ltt_node $HOSTNAME 10 $NETTYPE ||
error "Failed to set transaction timeout $?"
RETRY_PARAM=$($LNETCTL global show | awk '/retry_count/{print $NF}')
RSND_PRE=$($LNETCTL stats show | awk '/resend_count/{print $NF}')
FAILED_RSND_PRE=$($LNETCTL stats show |
awk '/failed_resends/{print $NF}')
SUCCESS_RSND_PRE=$($LNETCTL stats show |
awk '/successful_resends/{print $NF}')
LO_HVAL_PRE=$($LNETCTL net show -v 2 | awk '/health value/{print $NF}' |
xargs echo | sed 's/ /+/g' | bc -l)
RMT_HVAL_PRE=$($LNETCTL peer show --nid ${RNIDS[0]} -v 2 2>/dev/null |
awk '/health value/{print $NF}' | xargs echo |
sed 's/ /+/g' | bc -l)
# Might not have any peers so initialize to zero.
RMT_HVAL_PRE=${RMT_HVAL_PRE:-0}
return 0
}
function lnet_health_post() {
RSND_POST=$($LNETCTL stats show | awk '/resend_count/{print $NF}')
FAILED_RSND_POST=$($LNETCTL stats show |
awk '/failed_resends/{print $NF}')
SUCCESS_RSND_POST=$($LNETCTL stats show |
awk '/successful_resends/{print $NF}')
LO_HVAL_POST=$($LNETCTL net show -v 2 |
awk '/health value/{print $NF}' |
xargs echo | sed 's/ /+/g' | bc -l)
RMT_HVAL_POST=$($LNETCTL peer show --nid ${RNIDS[0]} -v 2 2>/dev/null |
awk '/health value/{print $NF}' | xargs echo |
sed 's/ /+/g' | bc -l)
# Might not have any peers so initialize to zero.
RMT_HVAL_POST=${RMT_HVAL_POST:-0}
${VERBOSE} &&
echo "Pre resends: $RSND_PRE" &&
echo "Post resends: $RSND_POST" &&
echo "Resends delta: $((RSND_POST - RSND_PRE))" &&
echo "Pre failed resends: $FAILED_RSND_PRE" &&
echo "Post failed resends: $FAILED_RSND_POST" &&
echo "Pre successful resends: $SUCCESS_RSND_PRE" &&
echo "Post successful resends: $SUCCESS_RSND_POST" &&
echo "Pre local health: $LO_HVAL_PRE" &&
echo "Post local health: $LO_HVAL_POST" &&
echo "Pre remote health: $RMT_HVAL_PRE" &&
echo "Post remote health: $RMT_HVAL_POST"
restore_lnet_params
do_lnetctl peer set --health 1000 --all ||
error "Failed to set peer health rc $?"
do_lnetctl net set --health 1000 --all ||
error "Failed to set local NI health rc $?"
return 0
}
function check_no_resends() {
echo "Check that no resends took place"
(( RSND_POST == RSND_PRE )) ||
error "Found resends: $RSND_POST != $RSND_PRE"
(( FAILED_RSND_POST == FAILED_RSND_PRE )) ||
error "Found resends: $FAILED_RSND_POST != $FAILED_RSND_PRE"
(( SUCCESS_RSND_POST == SUCCESS_RSND_PRE )) ||
error "Found resends: $SUCCESS_RSND_POST != $SUCCESS_RSND_PRE"
return 0
}
function check_resends() {
local delta=$((RSND_POST - RSND_PRE))
echo "Check that $RETRY_PARAM resends took place"
(( delta == RETRY_PARAM )) ||
error "Expected $RETRY_PARAM resends found $delta"
echo "Check for 1 failed resend"
delta=$((FAILED_RSND_POST - FAILED_RSND_PRE))
(( delta == 1 )) || error "Found $delta failed resends"
echo "Check for 0 successful resends"
delta=$((SUCCESS_RSND_POST - SUCCESS_RSND_PRE))
(( delta == 0 )) || error "Found $delta successful resends"
return 0
}
function check_successful_resends() {
local delta=$((SUCCESS_RSND_POST - SUCCESS_RSND_PRE))
echo "Check for 1 successful resend"
(( delta == 1 )) || error "Found $delta successful resends"
}
function check_no_local_health() {
echo "Check that local NI health is unchanged"
[[ $LO_HVAL_POST -ne $LO_HVAL_PRE ]] &&
error "Local health changed: $LO_HVAL_POST != $LO_HVAL_PRE"
return 0
}
function check_local_health() {
echo "Check that local NI health has been changed"
[[ $LO_HVAL_POST -eq $LO_HVAL_PRE ]] &&
error "Local health unchanged: $LO_HVAL_POST == $LO_HVAL_PRE"
return 0
}
function check_no_remote_health() {
echo "Check that remote NI health is unchanged"
[[ $RMT_HVAL_POST -ne $RMT_HVAL_PRE ]] &&
error "Remote health changed: $RMT_HVAL_POST != $RMT_HVAL_PRE"
return 0
}
function check_remote_health() {
echo "Check that remote NI health has been changed"
[[ $RMT_HVAL_POST -eq $RMT_HVAL_PRE ]] &&
error "Remote health unchanged: $RMT_HVAL_POST == $RMT_HVAL_PRE"
return 0
}
RNODE=""
RLOADED=false
NET_DEL_ARGS=""
RNIDS=( )
LNIDS=( )
setup_health_test() {
local need_mr=$1
[[ ${NETTYPE} == kfi* ]] && skip "kfi doesn't support drop rules"
local rnodes=$(remote_nodes_list)
[[ -z $rnodes ]] && skip "Need at least 1 remote node"
cleanup_lnet || error "Failed to cleanup before test execution"
# Loading modules should configure LNet with the appropriate
# test-framework configuration
load_lnet "config_on_load=1" || error "Failed to load modules"
LNIDS=( $($LCTL list_nids | xargs echo) )
RNODE=(${rnodes//,/ })
RNIDS=( $(do_node $RNODE $LCTL list_nids | xargs echo) )
if [[ -z ${RNIDS[@]} ]]; then
do_rpc_nodes $RNODE load_lnet ||
error "load_lnet failed on $RNODE rc = $?"
do_node $RNODE $LNETCTL lnet configure $LNET_CONFIG_INIT_OPT ||
error "Failed to configure lnet on $RNODE rc = $?"
RLOADED=true
RNIDS=( $(do_node $RNODE $LCTL list_nids | xargs echo) )
fi
[[ ${#LNIDS[@]} -lt 1 ]] &&
error "No NIDs configured for local host $HOSTNAME"
[[ ${#RNIDS[@]} -lt 1 ]] &&
error "No NIDs configured for remote host $RNODE"
# Ensure all peer NIs are local (i.e. non-routed config)
local rnid rnet lnid lnet
for rnid in ${RNIDS[@]}; do
rnet=${rnid##*@}
for lnid in ${LNIDS[@]}; do
lnet=${lnid##*@}
[[ ${lnet} == ${rnet} ]] &&
break
done
[[ ${lnet} != ${rnet} ]] &&
skip "Need non-routed configuration"
done
do_lnetctl discover ${RNIDS[0]} ||
error "Unable to discover ${RNIDS[0]}"
local mr=$($LNETCTL peer show --nid ${RNIDS[0]} |
awk '/Multi-Rail/{print $NF}')
if ${need_mr} && [[ $mr == False ]]; then
cleanup_health_test || return $?
skip "Need MR peer"
fi
if ( ! ${need_mr} && [[ ${#RNIDS[@]} -gt 1 ]] ) ||
( ! ${need_mr} && [[ ${#LNIDS[@]} -gt 1 ]] ); then
cleanup_health_test || return $?
skip "Need SR peer"
fi
if ${need_mr} && [[ ${#RNIDS[@]} -lt 2 ]]; then
# Add a second, reachable NID to rnode.
local net=${RNIDS[0]}
net="${net//*@/}1"
local if=$(do_rpc_nodes --quiet $RNODE lnet_if_list)
[[ -z $if ]] &&
error "Failed to determine interface for $RNODE"
do_node $RNODE "$LNETCTL lnet configure $LNET_CONFIG_OPT" ||
error "Failed to configure LNet on $RNODE rc = $?"
do_rpc_nodes $RNODE "$LNETCTL net add --net $net --if $if" ||
error "Failed to add interface to $RNODE rc = $?"
RNIDS[1]="${RNIDS[0]}1"
NET_DEL_ARGS="--net $net --if $if"
fi
if ${need_mr} && [[ ${#LNIDS[@]} -lt 2 ]]; then
local net=${LNIDS[0]}
net="${net//*@/}1"
do_lnetctl lnet configure $LNET_CONFIG_OPT ||
error "Failed to configure LNet rc = $?"
do_lnetctl net add --net $net --if ${INTERFACES[0]} ||
error "Failed to add interface rc = $?"
LNIDS[1]="${LNIDS[0]}1"
fi
$LNETCTL net show
$LNETCTL peer show -v 2 | grep -E -e nid -e health
$LCTL set_param debug=+net
return 0
}
cleanup_health_test() {
local rc=0
if [[ -n $NET_DEL_ARGS ]]; then
do_rpc_nodes $RNODE \
"$LNETCTL net del $NET_DEL_ARGS" ||
rc=$((rc + $?))
NET_DEL_ARGS=""
fi
unload_modules || rc=$((rc + $?))
if $RLOADED; then
do_rpc_nodes $RNODE unload_modules_local ||
rc=$((rc + $?))
RLOADED=false
fi
((rc == 0)) || error "Failed cleanup"
return $rc
}
add_drop_rule() {
if ((${RANDOM} % 2 == 0)); then
do_lctl net_drop add $@ ||
error "$LCTL net_drop add $@ failed rc = $?"
else
do_lnetctl fault drop add $@ ||
error "$LNETCTL fault drop add $@ failed rc = $?"
fi
}
del_drop_rule() {
if ((${RANDOM} % 2 == 0)); then
do_lctl net_drop del $@ ||
error "$LCTL net_drop del $@ failed rc = $?"
else
do_lnetctl fault drop del $@ ||
error "$LNETCTL fault drop del $@ failed rc = $?"
fi
}
add_delay_rule() {
if ((${RANDOM} % 2 == 0)); then
do_lctl $net_delay_add $@ ||
error "$LCTL net_delay add $@ failed rc = $?"
else
do_lnetctl fault delay add $@ ||
error "$LNETCTL fault delay add $@ failed rc = $?"
fi
}
del_delay_rule() {
if ((${RANDOM} % 2 == 0)); then
do_lctl $net_delay_del $@ ||
error "$LCTL net_delay del $@ failed rc = $?"
else
do_lnetctl fault delay del $@ ||
error "$LNETCTL fault delay del $@ failed rc = $?"
fi
}
add_health_test_drop_rules() {
local hstatus="-e $1"
local rate="-r ${2:-1}"
local args="-m GET $hstatus $rate"
local src dst
for src in "${LNIDS[@]}"; do
for dst in "${RNIDS[@]}" "${LNIDS[@]}"; do
add_drop_rule -s $src -d $dst ${args}
done
done
}
do_lnet_health_ping_test() {
local hstatus="$1"
echo "Simulate $hstatus"
lnet_health_pre || return $?
add_health_test_drop_rules ${hstatus}
do_lnetctl ping ${RNIDS[0]} &&
error "Should have failed"
lnet_health_post
del_drop_rule -a
return 0
}
# See lnet/lnet/lib-msg.c:lnet_health_check()
LNET_LOCAL_RESEND_STATUSES="local_interrupt local_dropped local_aborted"
LNET_LOCAL_RESEND_STATUSES+=" local_no_route local_timeout"
LNET_LOCAL_NO_RESEND_STATUSES="local_error"
test_204() {
setup_health_test false || return $?
local hstatus
for hstatus in ${LNET_LOCAL_RESEND_STATUSES} \
${LNET_LOCAL_NO_RESEND_STATUSES}; do
do_lnet_health_ping_test "${hstatus}" || return $?
check_no_resends || return $?
check_local_health || return $?
done
cleanup_health_test || return $?
return 0
}
run_test 204 "Check no health or resends for single-rail local failures"
test_205() {
(( $MDS1_VERSION >= $(version_code 2.14.58) )) ||
skip "need at least 2.14.58"
setup_health_test true || return $?
local hstatus
for hstatus in ${LNET_LOCAL_RESEND_STATUSES}; do
do_lnet_health_ping_test "${hstatus}" || return $?
check_resends || return $?
check_local_health || return $?
done
for hstatus in ${LNET_LOCAL_NO_RESEND_STATUSES}; do
do_lnet_health_ping_test "${hstatus}" || return $?
check_no_resends || return $?
check_local_health || return $?
done
cleanup_health_test || return $?
return 0
}
run_test 205 "Check health and resends for multi-rail local failures"
# See lnet/lnet/lib-msg.c:lnet_health_check()
LNET_REMOTE_RESEND_STATUSES="remote_dropped"
LNET_REMOTE_NO_RESEND_STATUSES="remote_error remote_timeout"
test_206() {
setup_health_test false || return $?
local hstatus
for hstatus in ${LNET_REMOTE_RESEND_STATUSES} \
${LNET_REMOTE_NO_RESEND_STATUSES}; do
do_lnet_health_ping_test "${hstatus}" || return $?
check_no_resends || return $?
check_no_local_health || return $?
check_no_remote_health || return $?
done
cleanup_health_test || return $?
return 0
}
run_test 206 "Check no health or resends for single-rail remote failures"
test_207() {
(( $MDS1_VERSION >= $(version_code 2.14.58) )) ||
skip "need at least 2.14.58"
setup_health_test true || return $?
local hstatus
for hstatus in ${LNET_REMOTE_RESEND_STATUSES}; do
do_lnet_health_ping_test "${hstatus}" || return $?
check_resends || return $?
check_no_local_health || return $?
check_remote_health || return $?
do_lnetctl peer set --health 1000 --all ||
error "Unable to reset health rc=$?"
done
for hstatus in ${LNET_REMOTE_NO_RESEND_STATUSES}; do
do_lnet_health_ping_test "${hstatus}" || return $?
check_no_resends || return $?
check_no_local_health || return $?
check_remote_health || return $?
do_lnetctl peer set --health 1000 --all ||
error "Unable to reset health rc=$?"
done
cleanup_health_test || return $?
return 0
}
run_test 207 "Check health and resends for multi-rail remote errors"
test_208_load_and_check_lnet() {
local ip2nets="$1"
local p_nid="$2"
local s_nid="$3"
local num_expected=1
load_lnet "networks=\"\" ip2nets=\"${ip2nets_str}\""
$LCTL net up $LNET_CONFIG_OPT ||
error "Failed to load LNet with ip2nets \"${ip2nets_str}\""
[[ -n $s_nid ]] &&
num_expected=2
declare -a nids
nids=( $($LCTL list_nids) )
[[ ${#nids[@]} -ne ${num_expected} ]] &&
error "Expect ${num_expected} NIDs found ${#nids[@]}"
[[ ${nids[0]} == ${p_nid} ]] ||
error "Expect NID \"${p_nid}\" found \"${nids[0]}\""
[[ -n $s_nid ]] && [[ ${nids[1]} != ${s_nid} ]] &&
error "Expect second NID \"${s_nid}\" found \"${nids[1]}\""
$LCTL net down &>/dev/null
cleanup_lnet
}
test_208() {
[[ ${NETTYPE} == tcp* ]] || skip "Need tcp NETTYPE"
(( $CLIENT_VERSION < $(version_code 2.17.53) )) ||
skip "kernel ip2nets deprecated in 2.17.53"
cleanup_lnet || error "Failed to unload modules before test execution"
setup_fakeif || error "Failed to add fake IF"
local if0_ip=$(ip --oneline addr show dev ${INTERFACES[0]} |
awk '/inet /{print $4}' |
sed 's:/.*::')
if0_ip=($(echo "${if0_ip[@]}" | tr ' ' '\n' | uniq | tr '\n' ' '))
local ip2nets_str="tcp(${INTERFACES[0]}) $if0_ip"
echo "Configure single NID \"$ip2nets_str\""
test_208_load_and_check_lnet "${ip2nets_str}" "${if0_ip}@tcp"
ip2nets_str="tcp(${INTERFACES[0]}) $if0_ip; tcp1($FAKE_IF) $FAKE_IP"
echo "Configure two NIDs; two NETs \"$ip2nets_str\""
test_208_load_and_check_lnet "${ip2nets_str}" "${if0_ip}@tcp" \
"${FAKE_IP}@tcp1"
ip2nets_str="tcp(${INTERFACES[0]}) $if0_ip; tcp($FAKE_IF) $FAKE_IP"
echo "Configure two NIDs; one NET \"$ip2nets_str\""
test_208_load_and_check_lnet "${ip2nets_str}" "${if0_ip}@tcp" \
"${FAKE_IP}@tcp"
local addr1=( ${if0_ip//./ } )
local addr2=( ${FAKE_IP//./ } )
local range="[${addr1[0]},${addr2[0]}]"
local i
for i in $(seq 1 3); do
range+=".[${addr1[$i]},${addr2[$i]}]"
done
ip2nets_str="tcp(${INTERFACES[0]},${FAKE_IF}) ${range}"
echo "Configured two NIDs; one NET alt syntax \"$ip2nets_str\""
test_208_load_and_check_lnet "${ip2nets_str}" "${if0_ip}@tcp" \
"${FAKE_IP}@tcp"
cleanup_fakeif
echo "alt syntax with missing IF \"$ip2nets_str\""
load_lnet "networks=\"\" ip2nets=\"${ip2nets_str}\""
echo "$LCTL net up $LNET_CONFIG_OPT should fail"
$LCTL net up $LNET_CONFIG_OPT &&
error "LNet bring up should have failed"
cleanup_lnet
}
run_test 208 "Test various kernel ip2nets configurations"
test_209() {
(( $MDS1_VERSION >= $(version_code 2.14.58) )) ||
skip "need at least 2.14.58"
setup_health_test false || return $?
echo "Simulate network_timeout w/SR config"
lnet_health_pre
add_health_test_drop_rules network_timeout
do_lnetctl discover ${RNIDS[0]} &&
error "Should have failed"
lnet_health_post
check_no_resends || return $?
check_local_health || return $?
check_no_remote_health || return $?
cleanup_health_test || return $?
setup_health_test true || return $?
echo "Simulate network_timeout w/MR config"
lnet_health_pre
add_health_test_drop_rules network_timeout
do_lnetctl discover ${RNIDS[0]} &&
error "Should have failed"
lnet_health_post
check_no_resends || return $?
check_local_health || return $?
check_remote_health || return $?
cleanup_health_test || return $?
return 0
}
run_test 209 "Check health, but not resends, for network timeout"
check_nid_in_recovq() {
local queue="$1"
local nid="$2"
local expect="$3"
local max_wait="${4:-10}"
local rc=0
(($expect == 0)) &&
echo "$queue recovery queue should be empty" ||
echo "$queue recovery queue should have $nid"
wait_update $HOSTNAME \
"$LNETCTL debug recovery $queue | \
grep -wc \"nid-0: $nid\"" \
"$expect" "$max_wait"
rc=$?
do_lnetctl debug recovery $queue
(($rc == 0)) ||
error "Expect $expect NIDs in recovery."
return 0
}
# First ping is sent at time 0.
# 2nd at 0 + 2^1 = 2
# 3rd at 2 + 2^2 = 6
# 4th at 6 + 2^3 = 14
# 5th at 14 + 2^4 = 30
# e.g. after 10 seconds we would expect 3 pings to have been sent, and the
# NI will have been enqueued for the 4th ping.
#
# If the recovery limit is 10 seconds, then after the 4th ping is sent
# we expect the peer NI to have aged out, so it will not actually be
# queued for a 5th ping.
# If max_recovery_ping_interval is set to 4 then:
# First ping is sent at time 0
# 2nd at 0 + min(2^1, 4) = 2
# 3rd at 2 + min(2^2, 4) = 6
# 4th at 6 + min(2^3, 4) = 10
# 5th at 10 + min(2^4, 4) = 14
# 6th at 14 + min(2^5, 4) = 18
# 7th at 18 + min(2^6, 4) = 22
# e.g. after 4 seconds we would expect 2 pings to have been sent, and
# after 13 seconds we would expect 4 pings to have been sent
check_ping_count() {
local queue="$1"
local nid="$2"
local expect="$3"
local max_wait="$4"
echo "Check ping counts:"
local rc=0
if [[ $queue == "ni" ]]; then
wait_update $HOSTNAME \
"$LNETCTL net show -v 2 | \
grep -e $nid -e ping_count | grep -wA1 $nid | \
awk '/ping_count/{print \\\$NF}'" "$expect" "$max_wait"
rc=$?
$LNETCTL net show -v 2 | grep -e $nid -e ping_count
elif [[ $queue == peer_ni ]]; then
wait_update $HOSTNAME \
"$LNETCTL peer show -v 2 --nid $nid | \
grep -v primary | \
grep -e $nid -e ping_count | grep -wA1 $nid | \
awk '/ping_count/{print \\\$NF}'" "$expect" "$max_wait"
rc=$?
$LNETCTL peer show -v 2 --nid $nid | grep -v primary |
grep -e $nid -e ping_count
else
error "Unrecognized queue \"$queue\""
return 1
fi
((rc == 0)) || error "Unexpected ping count"
return 0
}
test_210() {
[[ ${NETTYPE} == kfi* ]] && skip "kfi doesn't support drop rules"
reinit_dlc || return $?
add_net "${NETTYPE}" "${INTERFACES[0]}" || return $?
add_net "${NETTYPE}1" "${INTERFACES[0]}" || return $?
local prim_nid=$($LCTL list_nids | head -n 1)
do_lnetctl discover $prim_nid ||
error "failed to discover myself"
local default=$($LNETCTL global show |
awk '/recovery_limit/{print $NF}')
# Set recovery limit to 10 seconds.
do_lnetctl set recovery_limit 10 ||
error "failed to set recovery_limit"
$LCTL set_param debug=+net
add_drop_rule -s *@${NETTYPE} -d *@${NETTYPE} -m GET -r 1 -e local_error
add_drop_rule -s *@${NETTYPE}1 -d *@${NETTYPE}1 -m GET -r 1 \
-e local_error
add_drop_rule -s *@${NETTYPE} -d *@${NETTYPE} -r 1
add_drop_rule -s *@${NETTYPE}1 -d *@${NETTYPE}1 -r 1
do_lnetctl net set --health 0 --nid $prim_nid ||
error "Failed to set NI health to 0 rc $?"
check_ping_count "ni" "$prim_nid" "2" "10"
check_nid_in_recovq "-l" "$prim_nid" "1"
check_ping_count "ni" "$prim_nid" "3" "10"
check_nid_in_recovq "-l" "$prim_nid" "1"
del_drop_rule -a
reinit_dlc || return $?
add_net "${NETTYPE}" "${INTERFACES[0]}" || return $?
add_net "${NETTYPE}1" "${INTERFACES[0]}" || return $?
local prim_nid=$($LCTL list_nids | head -n 1)
do_lnetctl discover $prim_nid ||
error "failed to discover myself"
do_lnetctl set recovery_limit $default ||
error "failed to set recovery_limit"
default=$($LNETCTL global show |
awk '/max_recovery_ping_interval/{print $NF}')
do_lnetctl set max_recovery_ping_interval 4 ||
error "failed to set max_recovery_ping_interval"
$LCTL set_param debug=+net
add_drop_rule -s *@${NETTYPE} -d *@${NETTYPE} -m GET -r 1 -e local_error
add_drop_rule -s *@${NETTYPE}1 -d *@${NETTYPE}1 -m GET -r 1 \
-e local_error
add_drop_rule -s *@${NETTYPE} -d *@${NETTYPE} -r 1
add_drop_rule -s *@${NETTYPE}1 -d *@${NETTYPE}1 -r 1
do_lnetctl net set --health 0 --nid $prim_nid ||
error "Failed to set NI health to 0 rc $?"
check_ping_count "ni" "$prim_nid" "2" "10"
check_nid_in_recovq "-l" "$prim_nid" "1"
check_ping_count "ni" "$prim_nid" "4" "10"
check_nid_in_recovq "-l" "$prim_nid" "1"
del_drop_rule -a
do_lnetctl set max_recovery_ping_interval $default ||
error "failed to set max_recovery_ping_interval"
return 0
}
run_test 210 "Local NI recovery checks"
test_211() {
[[ ${NETTYPE} == kfi* ]] && skip "kfi doesn't support drop rules"
reinit_dlc || return $?
add_net "${NETTYPE}" "${INTERFACES[0]}" || return $?
add_net "${NETTYPE}1" "${INTERFACES[0]}" || return $?
local prim_nid=$($LCTL list_nids | head -n 1)
do_lnetctl discover $prim_nid ||
error "failed to discover myself"
local default=$($LNETCTL global show |
awk '/recovery_limit/{print $NF}')
# Set recovery limit to 10 seconds.
do_lnetctl set recovery_limit 10 ||
error "failed to set recovery_limit"
add_drop_rule -s *@${NETTYPE} -d *@${NETTYPE} -m GET -r 1 \
-e remote_error
add_drop_rule -s *@${NETTYPE}1 -d *@${NETTYPE}1 -m GET -r 1 \
-e remote_error
add_drop_rule -s *@${NETTYPE} -d *@${NETTYPE} -r 1
add_drop_rule -s *@${NETTYPE}1 -d *@${NETTYPE}1 -r 1
# Set health to 0 on one interface. This forces it onto the recovery
# queue.
do_lnetctl peer set --nid $prim_nid --health 0
check_nid_in_recovq "-p" "$prim_nid" "1"
# The peer should age out in 10-20 seconds
check_nid_in_recovq "-p" "$prim_nid" "0" "20"
# Ping count should reset to 0 when peer ages out
check_ping_count "peer_ni" "$prim_nid" "0"
del_drop_rule -a
# Set health to force it back onto the recovery queue. Set to 500 means
# in ~5 seconds it should be back at maximum value.
# NB: we reset the recovery limit to 0 (indefinite) so the peer NI is
# eligible again
do_lnetctl set recovery_limit 0 ||
error "failed to set recovery_limit"
do_lnetctl peer set --nid $prim_nid --health 500
check_nid_in_recovq "-p" "$prim_nid" "1"
check_nid_in_recovq "-p" "$prim_nid" "0" "20"
reinit_dlc || return $?
add_net "${NETTYPE}" "${INTERFACES[0]}" || return $?
add_net "${NETTYPE}1" "${INTERFACES[0]}" || return $?
local prim_nid=$($LCTL list_nids | head -n 1)
do_lnetctl discover $prim_nid ||
error "failed to discover myself"
do_lnetctl set recovery_limit $default ||
error "failed to set recovery_limit"
default=$($LNETCTL global show |
awk '/max_recovery_ping_interval/{print $NF}')
do_lnetctl set max_recovery_ping_interval 4 ||
error "failed to set max_recovery_ping_interval"
add_drop_rule -s *@${NETTYPE} -d *@${NETTYPE} -m GET -r 1 \
-e remote_error
add_drop_rule -s *@${NETTYPE}1 -d *@${NETTYPE}1 -m GET -r 1 \
-e remote_error
add_drop_rule -s *@${NETTYPE} -d *@${NETTYPE} -r 1
add_drop_rule -s *@${NETTYPE}1 -d *@${NETTYPE}1 -r 1
# Set health to 0 on one interface. This forces it onto the recovery
# queue.
do_lnetctl peer set --nid $prim_nid --health 0
check_ping_count "peer_ni" "$prim_nid" "1" "4"
check_nid_in_recovq "-p" "$prim_nid" "1"
# After we detect the 1st ping above, the 4th ping should be sent after
# ~13 seconds
check_ping_count "peer_ni" "$prim_nid" "4" "14"
check_nid_in_recovq "-p" "$prim_nid" "1"
del_drop_rule -a
do_lnetctl set max_recovery_ping_interval $default ||
error "failed to set max_recovery_ping_interval"
return 0
}
run_test 211 "Remote NI recovery checks"
test_212() {
[[ ${NETTYPE} == kfi* ]] && skip "kfi doesn't support drop rules"
local rnodes=$(remote_nodes_list)
[[ -z $rnodes ]] && skip "Need at least 1 remote node"
cleanup_lnet || error "Failed to cleanup before test execution"
# Loading modules should configure LNet with the appropriate
# test-framework configuration
load_lnet "config_on_load=1" || error "Failed to load modules"
local my_nid=$($LCTL list_nids | head -n 1)
[[ -z $my_nid ]] &&
error "Failed to get primary NID for local host $HOSTNAME"
local rnode=(${rnodes//,/ })
local rnodenids=$(do_node $rnode $LCTL list_nids | xargs echo)
local rloaded=false
if [[ -z $rnodenids ]]; then
do_rpc_nodes $rnode load_lnet ||
error "load_lnet failed on $rnode rc = $?"
do_node $rnode $LNETCTL lnet configure $LNET_CONFIG_INIT_OPT ||
error "Failed to configure LNet on $rnode rc = $?"
rloaded=true
rnodenids=$(do_node $rnode $LCTL list_nids | xargs echo)
fi
local rnodepnid=$(awk '{print $1}' <<< $rnodenids)
[[ -z $rnodepnid ]] &&
error "Failed to get primary NID for remote host $rnode"
log "Initial discovery"
do_lnetctl discover --force $rnodepnid ||
error "Failed to discover $rnodepnid"
do_node $rnode "$LNETCTL discover --force $my_nid" ||
error "$rnode failed to discover $my_nid"
log "Fail local discover ping to set LNET_PEER_REDISCOVER flag"
add_drop_rule -s "*@$NETTYPE" -d "*@$NETTYPE" -r 1 -e local_error
do_lnetctl discover --force $rnodepnid &&
error "Discovery should have failed"
del_drop_rule -a
local nid
for nid in $rnodenids; do
# We need GET (PING) delay just long enough so we can trigger
# discovery on the remote peer
add_delay_rule -s "*@$NETTYPE" -d $nid -r 1 -m GET -l 3
add_drop_rule -s "*@$NETTYPE" -d $nid -r 1 -m GET -e local_error
# We need PUT (PUSH) delay just long enough so we can process
# the PING failure
add_delay_rule -s "*@$NETTYPE" -d $nid -r 1 -m PUT -l 6
done
log "Force $HOSTNAME to discover $rnodepnid (in background)"
# We want to get a PING sent that we know will eventually fail.
# The delay rules we added will ensure the ping is not sent until
# the PUSH is also in flight (see below), and the drop rule ensures that
# when the PING is eventually sent it will error out
do_lnetctl discover --force $rnodepnid &
local pid1=$!
# We want a discovery PUSH from rnode to put rnode back on our
# discovery queue. This should cause us to try and send a PUSH to rnode
# while the PING is still outstanding.
log "Force $rnode to discover $my_nid"
do_node $rnode $LNETCTL discover --force $my_nid
# At this point we'll have both PING_SENT and PUSH_SENT set for the
# rnode peer. Wait for the PING to error out which should terminate the
# discovery process that we backgrounded.
log "Wait for $pid1"
wait $pid1
log "Finished wait on $pid1"
# The PING send failure clears the PING_SENT flag and puts the peer back
# on the discovery queue. When discovery thread processes the peer it
# will mistakenly clear the PUSH_SENT flag (and set PUSH_FAILED).
# Discovery will then complete for this peer even though we have an
# outstanding PUSH.
# When PUSH is actually unlinked it will be forced back onto the
# discovery queue, but we no longer have a ref on the peer. When
# discovery completes again, we'll trip the ASSERT in
# lnet_destroy_peer_locked()
# Delete the delay rules to send the PUSH
del_delay_rule -a
# Delete the drop rules
del_drop_rule -a
unload_modules ||
error "Failed to unload modules"
if $rloaded; then
do_rpc_nodes $rnode unload_modules_local ||
error "Failed to unload modules on $rnode"
fi
return 0
}
run_test 212 "Check discovery refcount loss bug (LU-14627)"
test_213() {
[[ ${NETTYPE} == tcp* ]] || skip "Need tcp NETTYPE"
cleanup_lnet || error "Failed to unload modules before test execution"
setup_fakeif || error "Failed to add fake IF"
reinit_dlc || return $?
add_net "tcp" "${INTERFACES[0]}" || return $?
add_net "tcp" "$FAKE_IF" || return $?
local nid1=$(lctl list_nids | head -n 1)
local nid2=$(lctl list_nids | tail --lines 1)
[[ $(lctl which_nid $nid1 $nid2) == $nid1 ]] ||
error "Expect nid1 \"$nid1\" to be preferred"
[[ $(lctl which_nid $nid2 $nid1) == $nid2 ]] ||
error "Expect nid2 \"$nid2\" to be preferred"
return 0
}
run_test 213 "Check LNetDist calculation for multiple local NIDs"
function check_ni_status() {
local nid="$1"
local expect="$2"
local status=$($LNETCTL net show |
grep -wA 1 ${nid} |
awk '/status/{print $NF}')
echo "NI ${nid} expect status \"${expect}\" found \"${status}\""
if [[ $status != $expect ]]; then
error "Error: Expect NI status \"$expect\" for NID \"$nid\" but found \"$status\""
fi
return 0
}
test_214() {
[[ ${NETTYPE} == tcp* ]] || skip "Need tcp NETTYPE"
cleanup_lnet || error "Failed to unload modules before test execution"
setup_fakeif || error "Failed to add fake IF"
reinit_dlc || return $?
add_net "tcp" "${INTERFACES[0]}" || return $?
add_net "tcp" "$FAKE_IF" || return $?
local nid1=$(lctl list_nids | head -n 1)
local nid2=$(lctl list_nids | tail --lines 1)
check_ni_status "0@lo" up
check_ni_status "$nid1" up
check_ni_status "$nid2" up
do_lnetctl ping --source $nid2 $nid1 ||
error "$LNETCTL ping --source $nid2 $nid1 failed"
echo "Set $FAKE_IF down"
echo "ip link set dev $FAKE_IF down"
ip link set dev $FAKE_IF down
check_ni_status "0@lo" up
check_ni_status "$nid1" up
check_ni_status "$nid2" down
}
run_test 214 "Check local NI status when link is downed"
get_ni_stat() {
local nid=$1
local stat=$2
$LNETCTL net show -v 2 |
grep -E -e nid -e $stat |
grep -wA 1 $nid |
awk '/'$stat':/{print $NF}'
}
ni_stats_pre() {
local nidvar s
for nidvar in nid1 nid2; do
for stat in send_count recv_count; do
s=$(get_ni_stat ${!nidvar} $stat)
eval ${nidvar}_pre_${stat}=$s
done
done
}
ni_stats_post() {
local nidvar s
for nidvar in nid1 nid2; do
for stat in send_count recv_count; do
s=$(get_ni_stat ${!nidvar} $stat)
eval ${nidvar}_post_${stat}=$s
done
done
}
ni_stat_changed() {
local nidvar=$1
local stat=$2
local pre post
eval pre=\${${nidvar}_pre_${stat}}
eval post=\${${nidvar}_post_${stat}}
echo "${!nidvar} pre ${stat} $pre post ${stat} $post"
[[ $pre -ne $post ]]
}
test_215() {
cleanup_lnet || error "Failed to unload modules before test execution"
reinit_dlc || return $?
add_net "${NETTYPE}1" "${INTERFACES[0]}" || return $?
add_net "${NETTYPE}2" "${INTERFACES[0]}" || return $?
local nid1=$($LCTL list_nids | head -n 1)
local nid2=$($LCTL list_nids | tail --lines 1)
do_lnetctl peer add --prim $nid1 --nid $nid2 ||
error "Failed to add peer"
local npings=25
for nidvarA in nid1 nid2; do
src=${!nidvarA}
dst=${!nidvarA}
for nidvarB in nid1 nid2; do
[[ $nidvarA == $nidvarB ]] && continue
ni_stats_pre
echo "$LNETCTL ping $dst x $npings"
for i in $(seq 1 $npings); do
$LNETCTL ping $dst &>/dev/null ||
error "$LNETCTL ping $dst failed"
done
ni_stats_post
# No source specified, sends to either NID should cause
# counts to increase across both NIs
for nidvar in nid1 nid2; do
for stat in send_count recv_count; do
ni_stat_changed $nidvar $stat ||
error "$stat unchanged for ${!nidvar}"
done
done
ni_stats_pre
echo "$LNETCTL ping --source $src $dst x $npings"
for i in $(seq 1 $npings); do
$LNETCTL ping --source $src $dst &>/dev/null ||
error "$LNETCTL ping --source $src $dst failed"
done
ni_stats_post
# src nid == dest nid means stats for the _other_ NI
# should be unchanged
for nidvar in nid1 nid2; do
for stat in send_count recv_count; do
if [[ ${!nidvar} == $src ]]; then
ni_stat_changed $nidvar $stat ||
error "$stat unchanged for ${!nidvar}"
else
ni_stat_changed $nidvar $stat &&
error "$stat changed for ${!nidvar}"
fi
done
done
done
# Double number of pings for next iteration because the net
# sequence numbers will have diverged
npings=$(($npings * 2))
done
# Ping from nid1 to nid2 should fail
do_lnetctl ping --source $nid1 $nid2 &&
error "ping from $nid1 to $nid2 should fail"
# Ping from nid2 to nid1 should fail
do_lnetctl ping --source $nid2 $nid1 &&
error "ping from $nid2 to $nid1 should fail"
return 0
}
run_test 215 "Test lnetctl ping --source option"
test_216() {
[[ ${NETTYPE} == kfi* ]] && skip "kfi doesn't support drop rules"
local rc=0
reinit_dlc || return $?
add_net "${NETTYPE}" "${INTERFACES[0]}" || return $?
add_net "${NETTYPE}1" "${INTERFACES[0]}" || return $?
local nids=( $($LCTL list_nids | xargs echo) )
do_lnetctl discover ${nids[0]} ||
error "Initial discovery failed"
do_lnetctl ping --source ${nids[0]} ${nids[0]} ||
error "Initial ping failed $?"
do_lnetctl ping --source ${nids[1]} ${nids[1]} ||
error "Initial ping failed $?"
local src dst
for src in "${nids[@]}"; do
for dst in "${nids[@]}"; do
add_drop_rule -r 1 -s $src -d $dst -e network_timeout
done
done
do_lnetctl ping ${nids[0]} || rc=$?
del_drop_rule -a
[[ $rc -eq 0 ]] &&
error "expected ping to fail"
check_nid_in_recovq "-p" "${nids[0]}" "0"
check_nid_in_recovq "-l" "${nids[0]}" "1"
return 0
}
run_test 216 "Failed send to peer NI owned by local host should not trigger peer NI recovery"
test_217() {
reinit_dlc || return $?
[[ $($LNETCTL net show | grep -c nid) -ne 1 ]] &&
error "Unexpected number of NIs after initalizing DLC"
do_lnetctl discover 0@lo ||
error "Failed to discover 0@lo"
unload_modules
}
run_test 217 "Don't leak memory when discovering peer with nnis <= 1"
test_218() {
[[ ${NETTYPE} == kfi* ]] && skip "kfi doesn't support drop rules"
reinit_dlc || return $?
[[ ${#INTERFACES[@]} -lt 2 ]] &&
skip "Need two LNet interfaces"
add_net "${NETTYPE}" "${INTERFACES[0]}" || return $?
local nid1=$($LCTL list_nids | head -n 1)
do_lnetctl ping $nid1 ||
error "ping failed"
add_net "${NETTYPE}" "${INTERFACES[1]}" || return $?
local nid2=$($LCTL list_nids | tail --lines 1)
do_lnetctl ping $nid2 ||
error "ping failed"
add_drop_rule -s $nid1 -d $nid1 -e local_error -r 1
do_lnetctl ping --source $nid1 $nid1 &&
error "ping should have failed"
local health_recovered
local i
for i in $(seq 1 5); do
health_recovered=$($LNETCTL net show -v 2 |
grep -c 'health value: 1000')
if [[ $health_recovered -ne 2 ]]; then
echo "Wait 1 second for health to recover"
sleep 1
else
break
fi
done
health_recovered=$($LNETCTL net show -v 2 |
grep -c 'health value: 1000')
del_drop_rule -a
[[ $health_recovered -ne 2 ]] &&
do_lnetctl net show -v 2 | grep -E -e nid -e health &&
error "Health hasn't recovered"
return 0
}
run_test 218 "Local recovery pings should exercise all available paths"
test_219() {
reinit_dlc || return $?
add_net "${NETTYPE}" "${INTERFACES[0]}" || return $?
add_net "${NETTYPE}1" "${INTERFACES[0]}" || return $?
local nid1=$(lctl list_nids | head -n 1)
local nid2=$(lctl list_nids | tail --lines 1)
do_lnetctl ping $nid1 ||
error "Ping failed $?"
do_lnetctl ping $nid2 ||
error "Ping failed $?"
do_lnetctl discover $nid2 ||
error "Discovery failed"
$LNETCTL peer show --nid $nid1 | grep -q $nid2 ||
error "$nid2 is not listed under $nid1"
}
run_test 219 "Consolidate peer entries"
# check that all routes have the 'expect' status
# lctl, lnetctl and debugfs can all report route status, so we check all three
# to ensure they are in agreement.
check_route_aliveness() {
local node="$1"
local expect="$2"
local noerror="${3:-false}"
local lctl_status
local lnetctl_status
local debugfs_status
local chk_intvl
local timeout
chk_intvl=$(cat /sys/module/lnet/parameters/alive_router_check_interval)
timeout=$(cat /sys/module/lnet/parameters/router_ping_timeout)
# Router may delay start for chk_intvl + timeout, so wait for 2x this
local lctl_cmd="do_node $node $LCTL show_route"
local lnetctl_cmd="do_node $node $LNETCTL route show -v"
local debugfs_cmd="do_node $node $LCTL get_param -n routes"
local max_wait=$((2 * (chk_intvl + timeout)))
local waited=0
local begin=$SECONDS
local got_expected=false
echo "waiting up to $max_wait seconds for route status agreement"
while ((waited <= max_wait)); do
lctl_status=$($lctl_cmd | awk '{print $7}' | sort -u | xargs)
lnetctl_status=$($lnetctl_cmd | awk '/state/{print $NF}' |
sort -u | xargs)
debugfs_status=$($debugfs_cmd | awk '/'${NETTYPE}'/{print $4}' |
sort -u | xargs)
if [[ $lctl_status == $expect ]] &&
[[ $lnetctl_status == $expect ]] &&
[[ $debugfs_status == $expect ]]; then
got_expected=true
break
fi
if ${VERBOSE}; then
echo "Wait $((max_wait - waited))s for route '$expect'"
echo " lctl:'$lctl_status' lnetctl:'$lnetctl_status'"
echo " debugfs:'$debugfs_status'"
fi
sleep 1
waited=$((SECONDS - begin))
done
if ! $got_expected; then
echo "lctl shows:"
$lctl_cmd
echo "lnetctl shows:"
$lnetctl_cmd
echo "debugfs shows:"
$debugfs_cmd
if ${noerror}; then
return 1
fi
[[ $lctl_status == $expect ]] ||
error "Wanted '$expect' lctl found '$lctl_status'"
[[ $lnetctl_status == $expect ]] ||
error "Wanted '$expect' lnetctl found '$lnetctl_status'"
[[ $debugfs_status == $expect ]] ||
error "Wanted '$expect' debugfs found '$debugfs_status'"
fi
echo "Got '$expect' after ${waited}s"
return 0
}
check_router_ni_status() {
local router="$1"
local expected_local="$2"
local expected_remote="$3"
local actual_local
local actual_remote
local chk_intvl
local timeout
local i
chk_intvl=$(cat /sys/module/lnet/parameters/alive_router_check_interval)
timeout=$(cat /sys/module/lnet/parameters/router_ping_timeout)
actual_local=$(do_node $router "$LNETCTL net show --net $LOCAL_NET" |
awk '/status/{print $NF}')
actual_remote=$(do_node $router "$LNETCTL net show --net $REMOTE_NET" |
awk '/status/{print $NF}')
for ((i = 0; i < $((chk_intvl + timeout)); i++)); do
if [[ $actual_local == $expected_local ]] &&
[[ $actual_remote == $expected_remote ]]; then
break
fi
echo "wait 1s for NI state change"
sleep 1
actual_local=$(do_node $router \
"$LNETCTL net show --net $LOCAL_NET" |
awk '/status/{print $NF}')
actual_remote=$(do_node $router \
"$LNETCTL net show --net $REMOTE_NET" |
awk '/status/{print $NF}')
done
[[ $actual_local == $expected_local ]] ||
error "$LOCAL_NET should be $expected_local"
[[ $actual_remote == $expected_remote ]] ||
error "$REMOTE_NET should be $expected_remote"
return 0
}
config_routes() {
for router in ${!ROUTER_INTERFACES[@]}; do
do_node $router "$LNETCTL set routing 1" ||
error "Unable to enable routing on $router"
done
for router in ${!ROUTER_NIDS[@]}; do
local router_nids=( ${ROUTER_NIDS[$router]} )
do_route_add $HOSTNAME $REMOTE_NET ${router_nids[0]} ||
return $?
done
for router in ${!ROUTER_INTERFACES[@]}; do
local router_nids=( ${ROUTER_NIDS[$router]} )
for rpeer in ${!RPEER_INTERFACES[@]}; do
do_route_add $rpeer $LOCAL_NET ${router_nids[1]} ||
return $?
done
done
check_route_aliveness "$HOSTNAME" "up" ||
return $?
for rpeer in ${RPEERS[@]}; do
check_route_aliveness "$rpeer" "up" ||
return $?
done
}
do_basic_rtr_test() {
config_routes || return $?
for rpeer in ${!RPEER_NIDS[@]}; do
local rpeer_nids=( ${RPEER_NIDS[$rpeer]} )
do_lnetctl ping ${rpeer_nids[0]} ||
error "Failed to ping ${rpeer_nids[0]}"
done
for rpeer in ${RPEERS[@]}; do
do_node $rpeer "$LNETCTL ping ${LNIDS[0]}" ||
error "$rpeer failed to ping ${LNIDS[0]}"
done
return 0
}
test_220() {
setup_router_test || return $?
do_basic_rtr_test || return $?
do_rpc_nodes $HOSTNAME,${RPEERS[0]} load_module \
../lnet/selftest/lnet_selftest ||
error "Failed to load lnet-selftest module"
$LSTSH -H -t $HOSTNAME -f ${RPEERS[0]} -m rw -s 4k ||
error "lst failed"
$LSTSH -H -t $HOSTNAME -f ${RPEERS[0]} -m rw ||
error "lst failed"
cleanup_router_test || return $?
}
run_test 220 "Add routes w/default options - check aliveness"
test_221() {
setup_router_test lnet_peer_discovery_disabled=1 || return $?
do_basic_rtr_test || return $?
cleanup_router_test || return $?
}
run_test 221 "Add routes w/DD disabled - check aliveness"
# assumes 1 router, 1 peer
do_aarf_enabled_test() {
local router=${ROUTERS[0]}
local router_nids=( ${ROUTER_NIDS[$router]} )
local rpeer=${RPEERS[0]}
local rpeer_nids=( ${RPEER_NIDS[$rpeer]} )
do_node $router "$LNETCTL set routing 1" ||
error "Unable to enable routing on $ROUTER"
check_router_ni_status $router "down" "down"
do_lnetctl ping ${rpeer_nids[0]} &&
error "Ping should fail"
do_node $rpeer "$LNETCTL ping ${LNIDS[0]}" &&
error "$rpeer ping should fail"
# Adding a route should cause the router's NI on LOCAL_NET to get up
do_route_add $HOSTNAME $REMOTE_NET ${router_nids[0]} ||
return $?
check_router_ni_status $router "up" "down" ||
return $?
# But route should still be down because of avoid_asym_router_failure
check_route_aliveness "$HOSTNAME" "down" ||
return $?
do_lnetctl ping ${rpeer_nids[0]} &&
error "Ping should fail"
do_node $rpeer "$LNETCTL ping ${LNIDS[0]}" &&
error "$rpeer ping should fail"
# Adding the symmetric route should cause the remote NI to go up and
# routes to go up
do_route_add $rpeer $LOCAL_NET ${router_nids[1]} ||
return $?
check_router_ni_status $router "up" "up" ||
return $?
check_route_aliveness "$HOSTNAME" "up" ||
return $?
check_route_aliveness "$rpeer" "up" ||
return $?
do_lnetctl ping ${rpeer_nids[0]} ||
error "Failed to ping ${rpeer_nids[0]}"
do_node $rpeer "$LNETCTL ping ${LNIDS[0]}" ||
error "$rpeer failed to ping ${LNIDS[0]}"
# Stop LNet on local host
do_lnetctl lnet unconfigure ||
error "Failed to stop LNet rc=$?"
check_router_ni_status $router "down" "up" ||
return $?
check_route_aliveness "$rpeer" "down" ||
return $?
do_lnetctl ping ${rpeer_nids[0]} &&
error "Ping should fail"
do_node $rpeer "$LNETCTL ping ${LNIDS[0]}" &&
error "$rpeer ping should fail"
return 0
}
test_222() {
setup_router_test avoid_asym_router_failure=1 || return $?
do_aarf_enabled_test || return $?
cleanup_router_test || return $?
}
run_test 222 "Check avoid_asym_router_failure=1"
test_223() {
local opts="avoid_asym_router_failure=1 lnet_peer_discovery_disabled=1"
setup_router_test $opts || return $?
do_aarf_enabled_test || return $?
cleanup_router_test || return $?
}
run_test 223 "Check avoid_asym_router_failure=1 w/DD disabled"
do_aarf_disabled_test() {
local router=${ROUTERS[0]}
local router_nids=( ${ROUTER_NIDS[$router]} )
local rpeer=${RPEERS[0]}
local rpeer_nids=( ${RPEER_NIDS[$rpeer]} )
do_node $router "$LNETCTL set routing 1" ||
error "Unable to enable routing on $router"
check_router_ni_status $router "down" "down"
do_route_add $HOSTNAME $REMOTE_NET ${router_nids[0]} ||
return $?
check_router_ni_status $router "up" "down" ||
return $?
check_route_aliveness "$HOSTNAME" "up" ||
return $?
do_route_add $rpeer $LOCAL_NET ${router_nids[1]} ||
return $?
check_router_ni_status $router "up" "up" ||
return $?
check_route_aliveness "$HOSTNAME" "up" ||
return $?
check_route_aliveness "$rpeer" "up" ||
return $?
do_lnetctl ping ${rpeer_nids[0]} ||
error "Failed to ping ${rpeer_nids[0]}"
do_node $rpeer "$LNETCTL ping ${LNIDS[0]}" ||
error "$rpeer failed to ping ${LNIDS[0]}"
# Stop LNet on local host
do_lnetctl lnet unconfigure ||
error "Failed to stop LNet rc=$?"
check_router_ni_status $router "down" "up" ||
return $?
check_route_aliveness "$rpeer" "up" ||
return $?
return 0
}
test_224() {
setup_router_test avoid_asym_router_failure=0 ||
return $?
do_aarf_disabled_test ||
return $?
cleanup_router_test ||
return $?
}
run_test 224 "Check avoid_asym_router_failure=0"
test_225() {
local opts="avoid_asym_router_failure=0 lnet_peer_discovery_disabled=1"
setup_router_test $opts || return $?
do_aarf_disabled_test || return $?
cleanup_router_test ||
return $?
}
run_test 225 "Check avoid_asym_router_failure=0 w/DD disabled"
test_226() {
(( $MDS1_VERSION >= $(version_code v2_16_57-110-ged70ccb471) )) ||
skip "need MDS >= 2.16.57.110 for refcnt fix LU-17440"
setup_router_test -r 2 || return $?
do_basic_rtr_test || return $?
# ping the peer from host to make sure it works
local rpeer=${RPEERS[0]}
local rpeer_nids=( ${RPEER_NIDS[$rpeer]} )
for i in {1..4}; do
do_lnetctl ping ${rpeer_nids[0]} ||
error "Failed to ping ${rpeer_nids[0]} on try $i"
done
# remove a route from the peer
local router_nids=( ${ROUTER_NIDS[${ROUTERS[0]}]} )
do_route_del $rpeer $LOCAL_NET ${router_nids[1]}
# should attempt to use both routes due to round-robin
# failure case here is an LBUG on $rpeer
for i in {1..4}; do
do_lnetctl ping ${rpeer_nids[0]}
done
cleanup_router_test || return $?
}
run_test 226 "test missing route for 1 of 2 routers"
test_227() {
local opts="lnet_peer_discovery_disabled=1 lnet_health_sensitivity=0"
opts+=" lnet_transaction_timeout=10"
[[ $NETTYPE != kfi* ]] || skip "kfi doesn't support drop rules"
setup_router_test -p 2 $opts || return $?
do_basic_rtr_test || return $?
do_node ${RPEERS[0]} $LNETCTL lnet unconfigure ||
error "Failed to unconfigure lnet on ${RPEERS[0]}"
local rpeer_nids=( ${RPEER_NIDS[${RPEERS[0]}]} )
do_lnetctl ping ${rpeer_nids[0]} &&
error "Expected ping to fail"
do_lnetctl ping ${rpeer_nids[0]} &&
error "Expected ping to fail"
local dropped=$(do_node ${ROUTERS[0]} \
$LNETCTL peer show -v 2 --nid ${rpeer_nids[0]} |
grep -A 2 dropped_stats |
awk '/get:/{print $2}' |
xargs echo |
sed 's/ /\+/g' | bc)
((dropped > 0)) ||
error "Expected dropped > 0 found $dropped"
cleanup_router_test
}
run_test 227 "Check router peer health w/DD disabled"
test_228() {
setup_router_test -r 2 || return $?
do_basic_rtr_test || return $?
do_rpc_nodes $HOSTNAME,${RPEERS[0]} load_module \
../lnet/selftest/lnet_selftest ||
error "Failed to load lnet-selftest module"
local lstpid rc
$LSTSH -H -t $HOSTNAME -f ${RPEERS[0]} -m rw &
lstpid=$!
do_lnetctl peer set --health 500 --all ||
error "Failed to set peer NI health rc = $?"
wait $lstpid
rc=$?
((rc == 0)) ||
error "LST returned non-zero rc = $rc"
do_lnetctl peer set --health 1000 --all ||
error "Failed to set peer NI health rc = $?"
local no_route=$(dmesg | tail | grep lnet_handle_find_routed_path |
grep -c "no route")
((no_route == 0)) ||
error "Detected no route send failures"
cleanup_router_test || return $?
}
run_test 228 "Routes should stay up when health is decremented"
test_230() {
[[ ${NETTYPE} == tcp* ]] || skip "Need tcp NETTYPE"
# LU-12815
echo "Check valid values; Should succeed"
local i
local lnid
local cmd
for ((i = 4; i < 16; i+=1)); do
reinit_dlc || return $?
if ((i % 2 == 0)); then
add_net "tcp" "${INTERFACES[0]}" || return $?
do_lnetctl net set --all --conns-per-peer $i ||
error "should have succeeded $?"
else
do_lnetctl net add --net "tcp" --if ${INTERFACES[0]} --conns-per-peer $i ||
error "should have succeeded $?"
fi
$LNETCTL net show -v 1 | grep -q "conns_per_peer: $i" ||
error "failed to set conns-per-peer to $i"
lnid="$($LCTL list_nids | head -n 1)"
do_lnetctl ping "$lnid" ||
error "failed to ping myself"
# "lctl --net tcp conn_list" prints the list of active
# connections. Since we're pinging ourselves, there should be
# 2 Control connections plus 2*conns_per_peer connections
# created (one Bulk Input, one Bulk Output in each pair).
# Here's the sample output for conns_per_peer set to 1:
# 12345-1.1.1.1@tcp I[0]host01->host01:988 2626560/1061296 nonagle
# 12345-1.1.1.1@tcp O[0]host01->host01:1022 2626560/1061488 nonagle
# 12345-1.1.1.1@tcp C[0]host01->host01:988 2626560/1061296 nonagle
# 12345-1.1.1.1@tcp C[0]host01->host01:1023 2626560/1061488 nonagle
cmd="printf 'network tcp\nconn_list\n' | lctl | grep -c '$lnid'"
# Expect 2+conns_per_peer*2 connections. Wait no longer
# than 10 seconds.
wait_update $HOSTNAME "$cmd" "$((2+i*2))" 10 ||
error "expected number of tcp connections $((2+i*2))"
done
reinit_dlc || return $?
add_net "tcp" "${INTERFACES[0]}" || return $?
echo "Set > 127; Should fail"
do_lnetctl net set --all --conns-per-peer 128 &&
error "should have failed $?"
reinit_dlc || return $?
add_net "tcp" "${INTERFACES[0]}" || return $?
local default=$($LNETCTL net show -v 1 |
awk '/conns_per_peer/{print $NF}')
echo "Set < 0; Should be ignored"
do_lnetctl net set --all --conns-per-peer -1 ||
error "should have succeeded $?"
$LNETCTL net show -v 1 | grep -q "conns_per_peer: ${default}" ||
error "Did not stay at default"
reinit_dlc || return $?
echo "Add > 127; Should fail and not configure NI"
! do_lnetctl net add --net "tcp" --if ${INTERFACES[0]} \
--conns-per-peer 128 ||
error "should have failed $?"
! $LNETCTL net show --net tcp 2>/dev/null | grep -qE "nid: .*@tcp$" ||
error "tcp NI should not be configured"
reinit_dlc || return $?
echo "Add overflow value; Should fail and not configure NI"
! do_lnetctl net add --net "tcp" --if ${INTERFACES[0]} \
--conns-per-peer 1000000000000000000000000 ||
error "should have failed $?"
! $LNETCTL net show --net tcp 2>/dev/null | grep -qE "nid: .*@tcp$" ||
error "tcp NI should not be configured"
reinit_dlc || return $?
echo "Add non-numeric value; Should fail and not configure NI"
! do_lnetctl net add --net "tcp" --if ${INTERFACES[0]} \
--conns-per-peer foo ||
error "should have failed $?"
! $LNETCTL net show --net tcp 2>/dev/null | grep -qE "nid: .*@tcp$" ||
error "tcp NI should not be configured"
reinit_dlc || return $?
echo "Import > 127 conns-per-peer; Should fail and not configure NI"
local yfile=$TMP/sanity-lnet-$testnum-import.yaml
do_lnetctl net add --net "tcp" --if ${INTERFACES[0]} \
--conns-per-peer 8 ||
error "should have succeeded $?"
$LNETCTL export --backup > $yfile || error "export failed $?"
reinit_dlc || return $?
# Corrupt the exported value to one outside the valid 0-127 range
sed -i 's/conns_per_peer: 8/conns_per_peer: 200/' $yfile
! do_lnetctl import < $yfile ||
error "import of out-of-range conns-per-peer should have failed"
! $LNETCTL net show --net tcp 2>/dev/null | grep -qE "nid: .*@tcp$" ||
error "tcp NI should not be configured"
}
run_test 230 "Test setting conns-per-peer"
test_231() {
reinit_dlc || return $?
local net=${NETTYPE}231
local opts="--net $net --if ${INTERFACES[0]}"
do_lnetctl net add $opts || error "Failed to add net"
local lnd=$(basename $LNETLND)
local param_path="/sys/module/$lnd/parameters"
[[ -d $param_path ]] ||
error "Cannot find kernel params for $lnd at $param_path"
local lnd_pto=$(cat $param_path/peer_timeout)
# A value of -1 means we use the default peer timeout
#lnet/include/lnet/lib-lnet.h:#define DEFAULT_PEER_TIMEOUT 180
((lnd_pto != -1)) || lnd_pto=180
$LNETCTL export --backup > $TMP/sanity-lnet-$testnum-expected.yaml
local pto=$(awk '/^\s+peer_timeout:/{print $NF}' \
$TMP/sanity-lnet-$testnum-expected.yaml)
((pto == lnd_pto)) ||
error "Expect peer_timeout $lnd_pto but found $pto"
$LNETCTL export --backup > $TMP/sanity-lnet-$testnum-actual.yaml
compare_yaml_files ||
error "Unexpected config after net add without options"
for pto in -1 0 60 180; do
reinit_dlc || return $?
sed -i 's/peer_timeout: .*/peer_timeout: '$pto'/' \
$TMP/sanity-lnet-$testnum-expected.yaml
do_lnetctl import $TMP/sanity-lnet-$testnum-expected.yaml ||
error "Failed to import configuration"
$LNETCTL export --backup > $TMP/sanity-lnet-$testnum-actual.yaml
# Swap lnd_to for -1 in the expected output
((pto != -1)) ||
sed -i 's/peer_timeout: '$pto'/peer_timeout: '$lnd_pto'/' \
$TMP/sanity-lnet-$testnum-expected.yaml
compare_yaml_files || error "Wrong config after import"
do_lnetctl net del $opts || error "Failed to delete net $net"
do_lnetctl net add $opts --peer-timeout=$pto ||
error "Failed to add net with peer-timeout=$pto"
$LNETCTL export --backup > $TMP/sanity-lnet-$testnum-actual.yaml
compare_yaml_files || error "Wrong config after lnetctl net add"
done
return 0
}
run_test 231 "Check DLC handling of peer_timeout parameter"
test_232() {
[[ ${NETTYPE} =~ ^(tcp|o2ib) ]] ||
skip "Need tcp or o2ib NETTYPE"
local i
local modparam=-1
local net=${NETTYPE}232
reinit_dlc || return $?
if [[ ${NETTYPE} == tcp* ]];then
modparam=$(cat "/sys/module/ksocklnd/parameters/tos")
elif [[ ${NETTYPE} == o2ib* ]]; then
modparam=$(cat "/sys/module/ko2iblnd/parameters/tos")
fi
do_lnetctl net add --net $net --if ${INTERFACES[0]} ||
error "Failed to add net (no ToS)"
do_lnetctl net show --net $net -v 1 | grep -q "tos: $modparam" ||
error "Failed ToS value should inherit from module parameter $modparam"
do_lnetctl net del --net $net --if ${INTERFACES[0]} ||
error "Failed to delete net $net"
for i in -1 104 106; do
do_lnetctl net add --net $net --if ${INTERFACES[0]} --tos $i ||
error "Failed to add net (ToS: $i)"
do_lnetctl net show --net $net -v 1 | grep -q "tos: $i" ||
error "Failed to set ToS value to 104"
do_lnetctl net del --net $net --if ${INTERFACES[0]} ||
error "Failed to delete net $net"
done
! do_lnetctl net add --net $net --if ${INTERFACES[0]} --tos 300 ||
error "lnetctl should reject invalid ToS value (>255)"
}
run_test 232 "Test setting ToS value"
test_233() {
setup_health_test true || return $?
local retries=$($LNETCTL global show | awk '/retry_count:/{print $NF}')
(( retries > 0 )) || skip "Need retry_count > 0, found $retries"
local hstatus
for hstatus in ${LNET_LOCAL_RESEND_STATUSES}; do
echo "Simulate intermittent $hstatus"
lnet_health_pre || return $?
add_health_test_drop_rules ${hstatus} 2
do_lnetctl ping --source ${LNIDS[0]} ${RNIDS[0]} ||
error "ping failed with rc = $?"
$LCTL net_drop_del -a
lnet_health_post
check_successful_resends || return $?
done
for hstatus in ${LNET_REMOTE_RESEND_STATUSES}; do
echo "Simulate intermittent $hstatus"
lnet_health_pre || return $?
add_health_test_drop_rules ${hstatus} 2
do_lnetctl ping --source ${LNIDS[0]} ${RNIDS[0]} ||
error "ping failed with rc = $?"
lnet_health_post
$LCTL net_drop_del -a
check_successful_resends || return $?
done
cleanup_health_test
}
run_test 233 "Check for successful resends"
check_parameter() {
local para=$1
local value=$2
echo "check parameter ${para} value ${value}"
return $(( $(do_lnetctl net show -v | \
tee /dev/stderr | \
grep -c "^ \+${para}: ${value}$") != ${#INTERFACES[@]} ))
}
test_234() {
reinit_dlc || return $?
add_net "${NETTYPE}" "${INTERFACES[0]}" || return $?
local nid=$($LCTL list_nids | head -n 1)
# Verify NI starts with UP status and maximum health
check_ni_status "$nid" up
local health=$($LNETCTL net show -v 2 --net ${NETTYPE} |
awk '/health value/{print $NF}')
echo "Initial health value: $health"
((health == 1000)) ||
error "NI $nid should start with health value 1000, found $health"
# Prevent recovery pings from incrementing health during this test
local hs=$($LNETCTL global show |
awk '/^\s+health_sensitivity:/{print $NF}')
do_lnetctl set health_sensitivity 0 ||
error "Failed to set health_sensitivity 0"
# Manually set health to 0 - should transition status to DOWN
echo "Setting health value to 0"
do_lnetctl net set --health 0 --nid $nid ||
error "Failed to set health to 0"
# Verify status changed to DOWN
check_ni_status "$nid" down
# Verify health is actually 0
health=$($LNETCTL net show -v 2 --net ${NETTYPE} |
awk '/health value/{print $NF}')
echo "Health value after setting to 0: $health"
((health == 0)) ||
error "NI $nid health should be 0, found $health"
# Set health back to maximum - should transition status to UP
echo "Setting health value to 1000"
do_lnetctl net set --health 1000 --nid $nid ||
error "Failed to set health to 1000"
# Verify status changed back to UP
check_ni_status "$nid" up
# Verify health is at maximum
health=$($LNETCTL net show -v 2 --net ${NETTYPE} |
awk '/health value/{print $NF}')
echo "Health value after setting to 1000: $health"
((health == 1000)) ||
error "NI $nid health should be 1000, found $health"
do_lnetctl set health_sensitivity $hs ||
error "Failed to set health_sensitivity $hs"
return 0
}
run_test 234 "Verify NI status changes with manual health value transitions"
test_235() {
[[ ${NETTYPE} == kfi* ]] && skip "kfi doesn't support drop rules"
reinit_dlc || return $?
add_net "${NETTYPE}" "${INTERFACES[0]}" || return $?
add_net "${NETTYPE}1" "${INTERFACES[0]}" || return $?
local nid1=$($LCTL list_nids | head -n 1)
local nid2=$($LCTL list_nids | tail -n 1)
# Verify both NIDs start with UP status
check_ni_status "$nid1" up
check_ni_status "$nid2" up
# Add drop rule to cause local failures on nid1
echo "Adding drop rule for $nid1"
$LCTL net_drop_add -s $nid1 -d $nid1 -m GET -r 1 -e local_timeout ||
error "Failed to add drop rule"
# Manually set health to 100 to force NI into recovery
echo "Setting health to 100 for $nid1"
do_lnetctl net set --health 100 --nid $nid1 ||
error "Failed to set health to 100"
# Recovery ping should cause health to decrement to 0, and status set
# to DOWN
wait_update $HOSTNAME \
"$LNETCTL net show --net ${NETTYPE} | \
awk '/^\\\s+status:/{print \\\$NF}'" \
"down" "60"
(($? == 0)) || error "NI status did not go down"
# Verify nid2 is still UP
check_ni_status "$nid2" up
# Clear drop rules
$LCTL net_drop_del -a
# Set health 900, recovery ping should increment to max and status set
# to UP
echo "Restoring health to 900 for $nid1"
do_lnetctl net set --health 900 --nid $nid1 ||
error "Failed to set health to 900"
# Verify status is back to UP
wait_update $HOSTNAME \
"$LNETCTL net show --net ${NETTYPE} | \
awk '/^\\\s+status:/{print \\\$NF}'" \
"up" "60"
(($? == 0)) || error "NI status did not go up"
return 0
}
run_test 235 "Verify NI status DOWN when health reaches 0 after failures"
check_remote_peer_ni_status() {
local node="$1"
local nid="$2"
local expect="$3"
local rc
wait_update $node \
"$LNETCTL peer show --nid $nid | \
grep -E -A 1 '\s+nid: ${nid}$' | \
awk '/\s+state:/{print \\\$NF}'" \
"$expect" "60"
rc=$?
(($rc == 0)) ||
error "Expect peer NI state \"$expect\" for $nid on $node"
return $rc
}
test_236() {
setup_router_test || return $?
do_basic_rtr_test || return $?
# Add another net on local host and router
add_net "${NETTYPE}3" "${INTERFACES[0]}" || return $?
local router=${ROUTERS[0]}
local rtr_interfaces=( ${ROUTER_INTERFACES[$router]} )
do_net_add $router ${NETTYPE}3 ${rtr_interfaces[0]} ||
return $?
local nid1=$($LCTL list_nids | head -1)
local nid2=$($LCTL list_nids | tail -1)
check_ni_status "$nid1" up
check_ni_status "$nid2" up
# Router shows both interfaces as up
check_remote_peer_ni_status "$router" "$nid1" "up" || return $?
check_remote_peer_ni_status "$router" "$nid2" "up" || return $?
# Drop traffic on nid1
$LCTL net_drop_add -s $nid1 -d $nid1 -r 1 -e local_timeout ||
error "Failed to add drop rule"
# Set health to 0
do_lnetctl net set --health 0 --nid $nid1 ||
error "Failed to set health to 0"
# Verify nid1 status is down, nid2 up
check_ni_status "$nid1" down
check_ni_status "$nid2" up
# Verify new NI status is reflected on router
check_remote_peer_ni_status "$router" "$nid1" "down" || return $?
check_remote_peer_ni_status "$router" "$nid2" "up" || return $?
# Clear drop rules, verify status transitions back to UP
$LCTL net_drop_del -a || return $?
do_lnetctl net set --health 900 --nid $nid1 ||
error "Failed to set health to 900"
wait_update $HOSTNAME \
"$LNETCTL net show --net ${NETTYPE} | \
awk '/^\\\s+status:/{print \\\$NF}'" \
"up" "60"
(($? == 0)) || error "NI status did not go to up"
# Verify new NI status is reflected on router
check_remote_peer_ni_status "$router" "$nid1" "up" || return $?
check_remote_peer_ni_status "$router" "$nid2" "up" || return $?
cleanup_router_test
}
run_test 236 "Local NI state propagates to routers"
do_mr_forwarding_test() {
local expect_mr_forwarding="$1"
local rpeer=${RPEERS[0]}
local rpeer_nids=( ${RPEER_NIDS[$rpeer]} )
do_lnetctl discover ${rpeer_nids[0]} ||
error "Discovery failed with rc = $?"
local router=${ROUTERS[0]}
local i
#define CFS_FAIL_RTR_HEALTH_INC 0xe004
do_node $router "$LCTL set_param fail_loc=0xe004" ||
error "Failed to set fail_loc rc = $?"
do_node $router "$LNETCTL set health_sensitivity 0" ||
error "Failed to set health_sensitivity 0 rc = $?"
do_node $router "$LNETCTL peer set --health 0 --nid ${rpeer_nids[0]}" ||
error "Failed to set health 0 for ${rpeer_nids[0]} rc = $?"
local pre_sends=( $(do_node $router \
"$LNETCTL peer show -v --nid ${rpeer_nids[0]}" |
awk '/send_count/{print $NF}' | xargs echo) )
do_node $router "$LNETCTL peer show -v 2 | grep -e nid -e health"
local my_nid=$($LCTL list_nids | head -n 1)
for i in {1..50}; do
do_lnetctl ping --source ${my_nid} ${rpeer_nids[0]} ||
error "Ping failed with rc=$?"
done
local post_sends=( $(do_node $router \
"$LNETCTL peer show -v --nid ${rpeer_nids[0]}" |
awk '/send_count/{print $NF}' | xargs echo ) )
do_node $router "$LNETCTL peer show -v 2 | grep -e nid -e health"
echo "pre_sends: ${pre_sends[@]}"
echo "post_sends: ${post_sends[@]}"
do_node $router "$LCTL set_param fail_loc=0" ||
error "Failed to set fail_loc rc = $?"
if [[ ${#pre_sends[@]} != 2 ]] || [[ ${#post_sends[@]} != 2 ]]; then
error "Unexpected send counts"
fi
if $expect_mr_forwarding ; then
((post_sends[1] - pre_sends[1] >= 50)) ||
error "Expected ${post_sends[1]}-${pre_sends[1]} >= 50"
else
((post_sends[1] - pre_sends[1] < 50)) ||
error "Expected ${post_sends[1]}-${pre_sends[1]} < 50"
fi
return 0
}
test_237() {
setup_router_test -m || return $?
do_basic_rtr_test || return $?
do_mr_forwarding_test true || return $?
cleanup_router_test
}
run_test 237 "Check MR forwarding feature"
test_238() {
setup_router_test -m
do_lnetctl set discovery 0 ||
error "Failed to disable discovery rc = $?"
do_basic_rtr_test || return $?
# Router should show non-MR for me
local my_nid=$($LCTL list_nids | head -n 1)
local mr=$(do_node ${ROUTERS[0]} "$LNETCTL peer show --nid $my_nid" |
awk '/Multi-Rail:/{print $NF}')
[[ $mr == false ]] ||
error "Expect 'Multi-Rail: false', found $mr"
do_mr_forwarding_test false || return $?
cleanup_router_test
}
run_test 238 "Check MR forwarding feature skipped for non-MR source"
test_241() {
reinit_dlc || return $?
do_lnetctl net add --net ${NETTYPE} --if ${INTERFACES[0]} ||
error "Failed to add net"
do_lnetctl net set -h | grep -q -- "--lnd-timeout:" ||
skip "lnetctl net set does not support --lnd-timeout option"
# Capture existing timeout value, we'll restore to this later
local old_lnd_to=$($LNETCTL net show --net ${NETTYPE} -v |
awk '/^\s+timeout:/{print $NF}')
local expected_lnd_to=$(( old_lnd_to + 1 ))
# Set new timeout and check it shows up in tunables
do_lnetctl net set --net ${NETTYPE} --lnd-timeout ${expected_lnd_to} ||
error "Failed to set LND timeout on ${NETTYPE} net"
check_parameter "timeout" ${expected_lnd_to} ||
error "Expected LND timeout $expected_lnd_to"
# Check if setting LND timeout to zero ends up defaulting to global
# lnd_timeout value
local global_lnd_to=$($LNETCTL global show -v |
awk '/lnd_timeout:/{print $NF}')
do_lnetctl net set --net ${NETTYPE} --lnd-timeout 0 ||
"Failed to set LND timeout on ${NETTYPE} net to zero"
check_parameter "timeout" ${global_lnd_to} ||
error "Expected LND timeout $global_lnd_to"
# Restore tunable timeout to old value
do_lnetctl net set --net ${NETTYPE} --lnd-timeout ${old_lnd_to}
}
run_test 241 "Check setting LND timeout value via lnetctl updates tunables"
test_245() {
reinit_dlc || return $?
add_net ${NETTYPE} ${INTERFACES[0]} || return $?
local param count paramlist="timeout"
[[ $NETTYPE != kfi* ]] || paramlist+=" traffic_class_num"
for param in $paramlist; do
echo "Check $param present in net show -v"
count=$($LNETCTL net show -v | grep -c " \+$param:")
((count == 1)) ||
error "$param parameter is missing"
echo "Check $param absent from export -b"
count=$($LNETCTL export -b | grep -c " \+$param:")
((param == 0)) ||
error "$param parameter is present"
done
return 0
}
run_test 245 "Check read-only params do not appear in export --backup"
### Test that linux route is added for each ni
test_250() {
local skip_param
[[ ${NETTYPE} == tcp* ]] || skip "Need tcp NETTYPE"
reinit_dlc || return $?
add_net "tcp" "${INTERFACES[0]}" || return $?
skip_param=$(cat /sys/module/ksocklnd/parameters/skip_mr_route_setup)
[[ ${skip_param:-0} -ne 0 ]] &&
skip "Need skip_mr_route_setup=0 found $skip_param"
ip route show table ${INTERFACES[0]} | grep -q "${INTERFACES[0]}"
}
run_test 250 "test that linux routes are added"
test_251() {
[[ ${NETTYPE} =~ kfi* ]] ||
skip "Need kfi NETTYPE"
reinit_dlc || return $?
add_net "kfi" "${INTERFACES[0]}" || return $?
add_net "kfi1" "${INTERFACES[0]}" || return $?
add_net "kfi10" "${INTERFACES[0]}" || return $?
return 0
}
run_test 251 "Define multiple kfi networks on single interface"
test_252() {
(( $MDS1_VERSION >= $(version_code v2_14_55-10-gc807e3f33b) )) ||
skip "need MDS >= 2.14.55.10 for memory leak fix"
setup_health_test false || return $?
local rc=0
do_rpc_nodes $RNODE unload_modules_local || rc=$?
if [[ $rc -ne 0 ]]; then
cleanup_health_test || return $?
error "Failed to unload modules on $RNODE rc=$rc"
else
RLOADED=false
fi
local ts1=$(date +%s)
do_lnetctl ping --timeout 15 ${RNIDS[0]} &&
error "Expected ping ${RNIDS[0]} to fail"
local ts2=$(date +%s)
local delta=$(echo "$ts2 - $ts1" | bc)
[[ $delta -lt 15 ]] ||
error "Ping took longer than expected to fail: $delta"
cleanup_health_test
}
run_test 252 "Ping to down peer should unlink quickly"
do_expired_message_drop_test() {
local rnid lnid old_tto
local old_retry=$($LNETCTL global show |
awk '/^\s+retry_count:/{print $NF}')
# Capture default, global LNet transaction timeout (LTT). If there's an
# LND timeout (LNDT) set for $NETTYPE, the true LTT = LNDT(LRC + 1) + 1.
local old_ltt=$($LNETCTL global show |
awk '/^\s+transaction_timeout:/{print $NF}')
local old_lnd_to=$($LNETCTL net show --net $NETTYPE --verbose |
awk '/^\s+timeout:/{print $NF}')
[[ -z "$old_lnd_to" ]] ||
old_ltt=$(( old_lnd_to * (old_retry + 1) + 1 ))
do_lnetctl set retry_count 0 || error "Failed to set retry count to 0"
$LNETCTL global show
[[ -z $old_ltt ]] &&
error "Cannot determine LNet transaction timeout"
# Set new LNet transaction timeout (LTT)
local ltt=10
set_ltt_node $HOSTNAME $ltt $NETTYPE ||
error "Failed to set transaction timeout"
# We want to consume all peer credits for at least LTT seconds
local delay=$((ltt + 1))
for lnid in "${LNIDS[@]}"; do
for rnid in "${RNIDS[@]}"; do
add_delay_rule -s "${lnid}" -d "${rnid}" \
-l "${delay}" -r 1 -m GET
done
done
declare -a pcs
pcs=( $($LNETCTL peer show -v --nid "${RNIDS[0]}" |
awk '/max_ni_tx_credits:/{print $NF}' |
xargs echo) )
[[ ${#RNIDS[@]} -ne ${#pcs[@]} ]] &&
error "Expect ${#RNIDS[@]} peer credit values found ${#pcs[@]}"
local rnet lnid lnet i j
# Need to use --source for multi-rail configs to ensure we consume
# all available peer credits
for ((i = 0; i < ${#RNIDS[@]}; i++)); do
local ping_args="--timeout $((delay+2))"
rnet=${RNIDS[i]##*@}
for lnid in ${LNIDS[@]}; do
lnet=${lnid##*@}
[[ $rnet == $lnet ]] && break
done
ping_args+=" --source ${lnid} ${RNIDS[i]}"
for j in $(seq 1 "${pcs[i]}"); do
$LNETCTL ping ${ping_args} 1>/dev/null &
done
echo "Issued ${pcs[i]} pings to ${RNIDS[i]} from $lnid"
done
# This ping should be queued on peer NI tx credit
$LNETCTL ping --timeout $((delay+2)) "${RNIDS[0]}" &
sleep ${delay}
del_delay_rule -a
wait
# Messages sent from the delay list do not go through
# lnet_post_send_locked(), thus we should only have a single drop
local dropped
dropped=$($LNETCTL peer show -v 2 --nid "${RNIDS[0]}" |
grep -A 2 dropped_stats |
awk '/get:/{print $2}' |
xargs echo |
sed 's/ /\+/g' | bc)
[[ $dropped -ne 1 ]] &&
error "Expect 1 dropped GET but found $dropped"
# Restore retry_count and transaction timeout values in the order they
# were changed.
do_lnetctl set retry_count $old_retry
set_ltt_node $HOSTNAME ${old_ltt} $NETTYPE
return 0
}
test_253() {
setup_health_test false || return $?
do_expired_message_drop_test || return $?
cleanup_health_test
}
run_test 253 "Message delayed beyond deadline should be dropped (single-rail)"
test_254() {
setup_health_test true || return $?
do_expired_message_drop_test || return $?
cleanup_health_test
}
run_test 254 "Message delayed beyond deadline should be dropped (multi-rail)"
test_255() {
[[ ${NETTYPE} == tcp* ]] || skip "Need tcp NETTYPE"
(( $CLIENT_VERSION < $(version_code 2.17.53) )) ||
skip "kernel ip2nets deprecated in 2.17.53"
reinit_dlc || return $?
cleanup_lnet || return $?
local routes_str="o2ib ${IF0_NET}.[$GW_HOSTNUM-$((GW_HOSTNUM+4))]"
local network_str="${NETTYPE}(${INTERFACES[0]})"
load_lnet "networks=\"${network_str}\" routes=\"${routes_str}\"" ||
error "Failed to load LNet"
$LCTL net up $LNET_CONFIG_OPT ||
error "Failed to load LNet with networks=\"${network_str}\" routes=\"${routes_str}\""
cat <<EOF > $TMP/sanity-lnet-$testnum-expected.yaml
net:
- net type: ${NETTYPE}
local NI(s):
- interfaces:
0: ${INTERFACES[0]}
EOF
append_net_tunables tcp
echo "route:" >> $TMP/sanity-lnet-$testnum-expected.yaml
for i in $(seq $GW_HOSTNUM $((GW_HOSTNUM + 4))); do
cat <<EOF >> $TMP/sanity-lnet-$testnum-expected.yaml
- net: o2ib
gateway: ${IF0_NET}.${i}@${NETTYPE}
hop: -1
priority: 0
EOF
done
echo "peer:" >> $TMP/sanity-lnet-$testnum-expected.yaml
for i in $(seq $GW_HOSTNUM $((GW_HOSTNUM + 4))); do
cat <<EOF >> $TMP/sanity-lnet-$testnum-expected.yaml
- primary nid: ${IF0_NET}.${i}@${NETTYPE}
Multi-Rail: False
peer ni:
- nid: ${IF0_NET}.${i}@${NETTYPE}
EOF
done
append_global_yaml
$LNETCTL export --backup > $TMP/sanity-lnet-$testnum-actual.yaml ||
error "export failed $?"
validate_gateway_nids
# Since we have a complex YAML config file we can test import
reinit_dlc || return $?
do_lnetctl import < $TMP/sanity-lnet-$testnum-actual.yaml ||
error "Import failed $?"
validate_gateway_nids
}
run_test 255 "Use lnet routes param with pdsh syntax"
test_256() {
[[ $NETTYPE != kfi* ]] || skip "kfi doesn't support delay rules"
setup_router_test peer_buffer_credits=1024 || return $?
do_basic_rtr_test || return $?
local rpeer=${RPEERS[0]}
local rpeer_nids=( ${RPEER_NIDS[$rpeer]} )
local rpnid=${rpeer_nids[0]}
local router=${ROUTERS[0]}
local router_nids=( ${ROUTER_NIDS[$router]} )
local rtrpnid=${router_nids[0]}
local rtr_pc=$(do_node $router $LNETCTL peer show -v --nid $rpnid |
awk '/max_ni_tx_credits:/{print $NF}' |
xargs echo | sed 's/ /\+/g' | bc)
((rtr_pc > 0)) ||
error "$router couldn't determine peer credits for $rpnid"
local my_pc=$($LNETCTL peer show -v --nid $rtrpnid |
awk '/max_ni_tx_credits:/{print $NF}' |
xargs echo | sed 's/ /\+/g' | bc)
((my_pc > 0)) || error "couldn't determine peer credits for $rtrpnid"
if ((my_pc < rtr_pc )); then
cleanup_router_test || return $?
skip "Need local peer credits >= router's peer credits"
fi
local old_retry=$(do_node $router $LNETCTL global show |
awk '/retry_count:/{print $NF}')
[[ -n $old_retry ]] ||
error "Cannot determine LNet retry count"
# Capture default, global LNet transaction timeout (LTT). If there's an
# LND timeout (LNDT) set for the router's REMOTE_NET, the true
# LTT = LNDT(LRC + 1) + 1.
local old_ltt=$(do_node $router $LNETCTL global show |
awk '/transaction_timeout:/{print $NF}')
local old_lnd_to=$(do_node $router $LNETCTL net show --net $REMOTE_NET \
--verbose | awk '/^\s+timeout:/{print $NF}')
[[ -n $old_lnd_to ]] && old_ltt=$(( old_lnd_to * (old_retry + 1) + 1 ))
# Set router's retry_count to zero to shorten/simplify message timeout.
do_node $router $LNETCTL set retry_count 0 ||
error "Failed to set retry_count"
local ltt=10
set_ltt_node $router $ltt $REMOTE_NET ||
error "Failed to set transaction_timeout"
#define CFS_FAIL_DELAY_MSG_FORWARD 0xe002
do_node $router $LCTL set_param fail_loc=0xe002
# We want to consume all peer credits for at least transaction_timeout
# seconds
local delay=$((ltt + 1))
local rnid lnid cmd
local args="-l $delay -r 1 -m GET"
for lnid in ${LNIDS[@]}; do
for rnid in ${rpeer_nids[@]}; do
cmd="$LCTL $net_delay_add -s ${lnid} -d ${rnid} $args"
echo "$router $cmd"
do_node $router $cmd || error "Failed to add delay rule"
done
done
local idx
declare -a pids
for idx in $(seq 1 ${rtr_pc}); do
echo "$LNETCTL ping --timeout $((delay+2)) $rpnid"
$LNETCTL ping --timeout $((delay+2)) $rpnid 1>/dev/null &
pids[$idx]=$!
done
((idx++))
echo "Issued ${rtr_pc} pings to $rpnid"
# This ping should be queued on the router's peer NI tx credit queue
$LNETCTL ping --timeout $((delay+2)) $rpnid &
pids[$idx]=$!
echo "Issued last ping - sleep $delay"
sleep ${delay}
do_node $router $LCTL $net_delay_del -a
local rc=0 rcsum=0
for idx in $(seq 1 $((rtr_pc + 1))); do
wait ${pids[$idx]} || rc=$?
((rc != 0)) && echo "ping pid ${pids[$idx]} returned rc=$rc"
((rcsum += rc))
done
((rcsum == 0)) || error "Detected ping failures"
# Restore old retry_count and REMOTE_NET LTT values
do_node $router $LNETCTL set retry_count ${old_retry}
set_ltt_node $router ${old_ltt} $REMOTE_NET
# Router should not drop any of the messages that have exceeded their
# deadline
local dropped=$(do_node $router $LNETCTL peer show -v 2 --nid $rpnid |
grep -A 2 dropped_stats |
awk '/get:/{print $2}' |
xargs echo |
sed 's/ /\+/g' | bc)
((dropped == 0)) ||
error "Expect 0 dropped GET but found $dropped"
cleanup_router_test
}
run_test 256 "Router should not drop messages that are past the deadline"
test_257() {
setup_router_test -r 2 -p 1 || return $?
do_basic_rtr_test || return $?
do_rpc_nodes $HOSTNAME,${RPEERS[0]} load_module \
../lnet/selftest/lnet_selftest ||
error "Failed to load lnet-selftest module"
local param
local all_nodes=$(comma_list ${ROUTERS[@]} ${RPEERS[@]} $HOSTNAME)
for param in alive_router_check_interval router_ping_timeout; do
do_nodes $all_nodes "echo 5 > /sys/module/lnet/parameters/$param"
done
$LSTSH -H -t $HOSTNAME -f ${RPEERS[0]} -m rw &
log "Wait 5s for LST to start"
sleep 5
log "Disable routing on ${ROUTERS[0]}"
do_node ${ROUTERS[0]} "$LNETCTL set routing 0" ||
error "Failed to disable routing rc = $?"
log "Wait for lst to finish"
wait
local drops=$(do_node ${ROUTERS[0]} \
"$LNETCTL net show -v; $LNETCTL peer show -v" |
awk '/drop_count:/{print $NF}' | xargs echo |
sed 's/ /\+/g' | bc)
((drops == 0)) ||
error "Detected $drops dropped messages - expect 0"
$LSTSH -H -t $HOSTNAME -f ${RPEERS[0]} -m rw &
log "Wait 5s for LST to start"
sleep 5
log "Enable routing on ${ROUTERS[0]}"
do_node ${ROUTERS[0]} "$LNETCTL set routing 1" ||
error "Failed to enable routing rc = $?"
log "Wait for lst to finish"
wait
drops=$(do_node ${ROUTERS[0]} \
"$LNETCTL net show -v; $LNETCTL peer show -v" |
awk '/drop_count:/{print $NF}' | xargs echo |
sed 's/ /\+/g' | bc)
((drops == 0)) ||
error "Detected $drops dropped messages - expect 0"
cleanup_router_test
}
run_test 257 "Test graceful router shutdown/startup"
check_sysctl() {
while IFS= read -r line; do
# Couldn't find a way to break this line
if [[ "$line" =~ \
^[[:space:]]*([a-zA-Z0-9_.-]+)[[:space:]]*=[[:space:]]*([a-zA-Z0-9_.-]+)[[:space:]]*$ \
]]; then
value=$(sysctl -n "${BASH_REMATCH[1]}" 2>/dev/null)
if [ -z "${value}" ]; then
echo "Parameter ${BASH_REMATCH[1]} not set"
return 1
fi
echo "found: ${BASH_REMATCH[1]} ${value}"
if [ "${value}" != "${BASH_REMATCH[2]}" ]; then
echo "Parameter ${BASH_REMATCH[1]}" \
" wrong value: ${value}" \
" expected: ${BASH_REMATCH[2]}"
return 2
fi
fi
done < "$1"
return 0
}
### Test that linux route is added for each ni
test_260() {
local sysctl_file="/etc/lnet-sysctl.conf"
local sysctl_conf_bak="/etc/lnet-sysctl.bak"
local sysctl_bak=$TMP/lnet-sysctl.bak
local -i max_wait=60
local -i retries=0
echo "Setting default values and create backup for check"
sysctl -w net.ipv4.neigh.default.gc_thresh1=128 > "$sysctl_bak"
sysctl -w net.ipv4.neigh.default.gc_thresh2=512 >> "$sysctl_bak"
sysctl -w net.ipv4.neigh.default.gc_thresh3=1024 >> "$sysctl_bak"
sysctl -w net.ipv6.neigh.default.gc_thresh1=128 >> "$sysctl_bak"
sysctl -w net.ipv6.neigh.default.gc_thresh2=512 >> "$sysctl_bak"
sysctl -w net.ipv6.neigh.default.gc_thresh3=1024 >> "$sysctl_bak"
echo "Check default configuration"
retries=0
until check_sysctl "${sysctl_bak}"
do
if (( retries >= max_wait )); then
error "Default sysconfig values not set"
fi
sleep 1
retries+=1
done
load_modules || error "Failed to load Modules"
sysctlstat=$(cat /sys/module/lnet/parameters/enable_sysctl_setup 2>&-)
echo "enable_sysctl_setup set to ${sysctlstat}"
echo "New configuration"
echo 1 > /sys/module/lnet/parameters/enable_sysctl_setup 2>&1
sysctlstat=$(cat /sys/module/lnet/parameters/enable_sysctl_setup 2>&-)
echo "enable_sysctl_setup set to ${sysctlstat}"
$LNETCTL setup-sysctl ||
error "setup-sysctl failed"
echo "Check new configuration"
retries=0
until check_sysctl "${sysctl_file}"; do
if (( retries >= max_wait )); then
error "New sysctl values not set"
fi
sleep 1
retries+=1
done
echo "Reset to original values"
echo 0 > /sys/module/lnet/parameters/enable_sysctl_setup 2>&1
sysctlstat=$(cat /sys/module/lnet/parameters/enable_sysctl_setup 2>&-)
echo "enable_sysctl_setup set to ${sysctlstat}"
$LNETCTL setup-sysctl ||
error "setup-sysctl failed"
echo "Check original configuration"
retries=0
until check_sysctl "${sysctl_bak}"; do
if (( retries >= max_wait )); then
echo "Original sysconfig values not set"
break
fi
sleep 1
retries+=1
done
rm -f "${sysctl_bak}"
unload_modules || error "Failed to cleanup Modules"
}
run_test 260 "test that linux sysctl parameter are set correctly"
test_265() {
reinit_dlc || return $?
local cpt
for cpt in "[1-0]" "[]"; do
cat <<EOF > $TMP/sanity-lnet-$testnum.yaml
net:
- net type: ${NETTYPE}
local NI(s):
- interfaces:
0: ${INTERFACES[0]}
CPT: "${cpt}"
EOF
! do_lnetctl import $TMP/sanity-lnet-$testnum.yaml ||
error "Import should have failed"
cat <<EOF > $TMP/sanity-lnet-$testnum.yaml
net:
- net type: ${NETTYPE}
local NI(s):
- interfaces:
0: ${INTERFACES[0]}
CPT: ${cpt}
EOF
! do_lnetctl import $TMP/sanity-lnet-$testnum.yaml ||
error "Import should have failed"
done
}
run_test 265 "Import of invalid CPT should fail"
test_266() {
reinit_dlc || return $?
add_net "${NETTYPE}" "${INTERFACES[0]}" || return $?
$LNETCTL export --backup > $TMP/sanity-lnet-$testnum-expected.yaml
local cpt=$($LNETCTL net show -v --net ${NETTYPE} |
awk '/CPT/{print $NF}')
local low=$(awk -F, '{print $1}'<<<"$cpt" | tr -d '[]"')
local high=$(awk -F, '{print $NF}'<<<"$cpt" | tr -d '[]"')
local tyaml=$TMP/sanity-lnet-$testnum.yaml
cat <<EOF > $tyaml
net:
- net type: ${NETTYPE}
local NI(s):
- interfaces:
0: ${INTERFACES[0]}
EOF
$LNETCTL net show -v -n ${NETTYPE} | awk '/^\s+tunables:$/,/^\s+CPT:/' |
grep -v -e 'dev cpt' -e 'CPT' >> $tyaml
echo " CPT: \"[$low-$high]\"" >> $tyaml
reinit_dlc || return $?
do_lnetctl import $tyaml || error "Import failed rc = $?"
$LNETCTL export --backup > $TMP/sanity-lnet-$testnum-actual.yaml
compare_yaml_files || return $?
cat <<EOF > $tyaml
net:
- net type: ${NETTYPE}
local NI(s):
- interfaces:
0: ${INTERFACES[0]}
EOF
$LNETCTL net show -v -n ${NETTYPE} | awk '/^\s+tunables:$/,/^\s+CPT:/' |
grep -v -e 'dev cpt' -e 'CPT' >> $tyaml
echo " CPT: [$low-$high]" >> $tyaml
reinit_dlc || return $?
do_lnetctl import $tyaml || error "Import failed rc = $?"
$LNETCTL export --backup > $TMP/sanity-lnet-$testnum-actual.yaml
compare_yaml_files || return $?
}
run_test 266 "Validate CPT parsing in network sequence"
test_270() {
[[ "$NETTYPE" =~ tcp ]] || skip "Need tcp NETTYPE"
cleanup_lnet || error "Failed to unload modules before test execution"
reinit_dlc || return $?
#define CFS_FAIL_TEST_PING_MD 0xe003 CFS_FAIL_SOME 0x10000000
"$LCTL" set_param fail_loc=0x1000e003 fail_val=100
"$LNETCTL" net add --net "tcp" --if "${INTERFACES[0]}"
local nid="$($LCTL list_nids | head -n 1)"
do_lnetctl ping "$nid" ||
error "$LNETCTL ping $nid with TEST_PING_MD failed"
dmesg | grep -q "found metadata reply" ||
error "Didn't find metadata reply message in log"
cleanup_lnet || error "Failed to unload modules after test execution"
}
run_test 270 "Test NID metadata"
test_280() {
cleanup_lnet || error "Failed to unload modules before test execution"
local rc=0
modinfo ksocklnd 2>/dev/null || rc=$?
((rc == 1)) || skip "Need request_module to fail"
local lnd=""
case $NETTYPE in
o2ib*) lnd=ko2iblnd;;
tcp*) lnd=ksocklnd;;
kfi*) lnd=kkfilnd;;
gni*) lnd=kgnilnd;;
esac
[[ -n $lnd ]] || skip "Unsupported NETTYPE $NETTYPE"
load_lnet || error "Failed to load lnet"
$LUSTRE_RMMOD $lnd || error "Failed to unload $lnd"
$LNETCTL lnet configure -a
$LUSTRE_RMMOD
}
run_test 280 "Don't panic when request_module fails"
test_290() {
[[ ${NETTYPE} == tcp* ]] || skip "Need tcp NETTYPE"
if (( $LINUX_VERSION_CODE < $(version_code 4.2.0) )); then
skip "Need kernel >= 4.2.0 for local net namespace"
fi
cleanup_lnet || error "Failed to unload modules before test execution"
setup_fakeif || error "Failed to add fake IF"
reinit_dlc || return $?
$LNETCTL net add --nid ${FAKE_IP}@tcp
sleep 1
accept_port=$(cat /sys/module/lnet/parameters/accept_port)
binding_found=$(ss -ltnup | awk -v ip="${FAKE_IP}" -v port="${accept_port}" '
$5 == ip ":" port { print }
')
if [[ -z "$binding_found" ]]; then
error "Explicit binding not found for ${FAKE_IP}:${accept_port}"
fi
cleanup_fakeif
}
run_test 290 "Check that lnet acceptor is using explicit address binding"
test_291() {
[[ ${NETTYPE} == tcp* ]] ||
skip "Need tcp NETTYPE"
reinit_dlc || return $?
cleanup_lnet || return $?
local lnid cmd
load_lnet "accept_port_bulk=989" || error "Failed to load LNet"
$LCTL net up $LNET_CONFIG_OPT ||
error "Failed to load LNet with accept_port_bulk=989"
$LNETCTL net show -v 1 | grep -q "conns_per_peer: 1" ||
error "failed to set conns-per-peer to 1"
lnid="$(lctl list_nids | head -n 1)"
do_lnetctl ping "$lnid" ||
error "failed to ping myself"
# "lctl --net tcp conn_list" prints the list of active
# connections. Since we're pinging ourselves, there should be
# 2 Control connections plus 2*conns_per_peer connections
# created (one Bulk Input, one Bulk Output in each pair).
# Here's the sample output for conns_per_peer set to 1:
# 12345-1.1.1.1@tcp I[0]host01->host01:988 2626560/1061296 nonagle
# 12345-1.1.1.1@tcp O[0]host01->host01:1022 2626560/1061488 nonagle
# 12345-1.1.1.1@tcp C[0]host01->host01:988 2626560/1061296 nonagle
# 12345-1.1.1.1@tcp C[0]host01->host01:1023 2626560/1061488 nonagle
cmd="printf 'network tcp\nconn_list\n' | lctl | grep -c ':989'"
# Expect one bulk connection on port 989
wait_update $HOSTNAME "$cmd" "1" 10 ||
error "expected 1 tcp connection on port 989"
cleanup_lnet || return $?
}
run_test 291 "load lnet using custom accept port for bulk tcp connections"
test_300() {
# LU-13274
local header
local out=$TMP/$tfile
local prefix=/usr/include/linux/lnet
# We use a hard coded prefix so that this test will not fail
# when run in tree.
CC=${CC:-cc}
if ! which $CC > /dev/null 2>&1; then
skip_env "$CC is not installed"
fi
cleanup_lnet || return $?
load_lnet
local cc_args="-Wall -Werror -std=c99 -c -x c /dev/null -o $out"
if ! [[ -d $prefix ]]; then
# Assume we're running in tree and fixup the include path.
prefix=$LUSTRE/../lnet/include/uapi/linux/lnet
cc_args+=" -I $LUSTRE/../lnet/include/uapi"
fi
for header in $prefix/*.h; do
if ! [[ -f "$header" ]]; then
continue
fi
echo "$CC $cc_args -include $header"
$CC $cc_args -include $header ||
error "cannot compile '$header'"
done
rm -f $out
}
run_test 300 "packaged LNet UAPI headers can be compiled"
# LU-16081 lnet: Memory leak on adding existing interface
test_301() {
reinit_dlc || return $?
do_lnetctl net add --net ${NETTYPE} --if ${INTERFACES[0]} ||
error "Failed to add net"
do_lnetctl net add --net ${NETTYPE} --if ${INTERFACES[0]} &&
error "add net should have failed"
do_lnetctl net del --net ${NETTYPE} --if ${INTERFACES[0]} ||
error "Failed to del net"
unload_modules
}
run_test 301 "Check for dynamic adds of same/wrong interface (memory leak)"
test_302() {
! [[ $NETTYPE =~ (tcp|o2ib) ]] && skip "Need tcp or o2ib NETTYPE"
reinit_dlc || return $?
add_net "${NETTYPE}" "${INTERFACES[0]}" || return $?
local nid=$($LCTL list_nids)
do_lnetctl ping ${nid} ||
error "pinging self failed $?"
do_lnetctl debug peer --nid ${nid} ||
error "failed to dump peer debug info $?"
}
run_test 302 "Check that peer debug info can be dumped"
test_303() {
[[ ${NETTYPE} == tcp* ]] || skip "Need tcp NETTYPE"
setup_health_test true || return $?
setup_fakeif || error "Failed to add fake IF"
add_net "${NETTYPE}99" "$FAKE_IF" || return $?
local nid=$($LCTL list_nids | tail --lines 1)
# Our updated config should be pushed to RNODE
local found=$(do_node $RNODE "$LNETCTL peer show --nid $nid")
[[ -z $found ]] && error "Peer not updated on $RNODE"
local prim=$($LCTL list_nids | head -n 1)
if ! grep -q -- " primary nid: $prim"<<<"${found}"; then
echo "$found"
error "Wrong primary nid"
fi
echo "Set $FAKE_IF down"
echo "ip link set dev $FAKE_IF down"
ip link set dev $FAKE_IF down
check_ni_status "$nid" down
local hval=$(do_node $RNODE "$LNETCTL peer show --nid $nid -v 2 | \
grep -e '- nid:' -e 'health value:'")
hval=$(grep -A 1 $nid<<<"$hval" | tail -n 1 | awk '{print $NF}')
(( hval < 1000 )) ||
error "Expect $hval < 1000"
return 0
}
run_test 303 "Check peer NI health after link down"
test_304() {
[[ ${NETTYPE} == tcp* ]] || skip "Need tcp NETTYPE"
cleanup_lnet || error "Failed to unload modules before test execution"
setup_fakeif || error "Failed to add fake IF"
reinit_dlc || return $?
add_net "tcp" "${INTERFACES[0]}" || return $?
add_net "tcp" "$FAKE_IF" || return $?
local nid1=$(lctl list_nids | head -n 1)
local nid2=$(lctl list_nids | tail --lines 1)
check_ni_status "$nid1" up
check_ni_status "$nid2" up
do_lnetctl peer add --prim_nid ${nid2} --lock_prim ||
error "peer add failed $?"
local locked_peer_state=($(do_lnetctl peer show -v 4 --nid ${nid2} |
awk '/peer state/{print $NF}'))
# Expect peer state bits:
# LNET_PEER_MULTI_RAIL(0) | LNET_PEER_CONFIGURED(3) |
# LNET_PEER_LOCK_PRIMARY(20)
(( $locked_peer_state != "1048585")) &&
error "Wrong peer state \"$locked_peer_state\" expected 1048585"
# Clear LNET_PEER_CONFIGURED bit and verify
do_lnetctl peer set --nid ${nid2} --state 1048577 ||
error "peer add failed $?"
locked_peer_state=($(do_lnetctl peer show -v 4 --nid ${nid2} |
awk '/peer state/{print $NF}'))
(( $locked_peer_state != "1048577")) &&
error "Wrong peer state \"$locked_peer_state\" expected 1048577"
do_lnetctl discover ${nid1} ||
error "Failed to discover peer"
# Expect nid2 and nid1 peer entries to be consolidated,
# nid2 to stay primary
cat <<EOF >> $TMP/sanity-lnet-$testnum-expected.yaml
peer:
- primary nid: ${nid2}
Multi-Rail: True
peer ni:
- nid: ${nid1}
state: NA
- nid: ${nid2}
state: NA
EOF
$LNETCTL peer show > $TMP/sanity-lnet-$testnum-actual.yaml
compare_yaml_files ||
error "Unexpected peer configuration"
locked_peer_state=($(do_lnetctl peer show -v 4 --nid ${nid2} |
awk '/peer state/{print $NF}'))
# Expect peer state bits to be added:
# LNET_PEER_DISCOVERED(4) | LNET_PEER_NIDS_UPTODATE(8)
(( $locked_peer_state != "1048849")) &&
error "Wrong peer state \"$locked_peer_state\" expected 1048849"
cleanup_lnet
}
run_test 304 "Check locked primary peer nid consolidation"
test_305() {
[[ ${NETTYPE} == tcp* ]] || skip "Need tcp NETTYPE"
reinit_dlc || return $?
add_net "${NETTYPE}" "${INTERFACES[0]}" || return $?
local nid=$($LCTL list_nids)
do_lnetctl ping ${nid} ||
error "pinging self failed $?"
[[ "${nid%@*}" == "$(hostname -i | awk '{print $1}')" ]] ||
skip "IP $(hostname -i) isn't NID $nid"
nid="$(hostname -s)@${NETTYPE}"
do_lnetctl ping $nid ||
error "pinging own hostname $nid failed $?"
}
run_test 305 "Resolve hostname before lnetctl ping"
static_config() {
local module=$1
local setting=$2
cleanup_lnet || error "Failed to cleanup LNet"
load_module ../libcfs/libcfs/libcfs ||
error "Failed to load module libcfs rc = $?"
load_module ../lnet/lnet/lnet ||
error "Failed to load module lnet rc = $?"
echo "loading ${module} ${setting} type ${NETTYPE}"
load_module "${module}" "${setting}" ||
error "Failed to load module ${module} rc = $?"
do_lnetctl lnet configure $LNET_CONFIG_INIT_OPT ||
error "lnet configure $LNET_CONFIG_INIT_OPT failed rc = $?"
return 0
}
test_310() {
local value=65
if [[ ${NETTYPE} == tcp* ]];then
static_config "../lnet/klnds/socklnd/ksocklnd" \
"sock_timeout=${value}"
elif [[ ${NETTYPE} == o2ib* ]]; then
static_config "../lnet/klnds/o2iblnd/ko2iblnd" \
"timeout=${value}"
elif [[ ${NETTYPE} == gni* ]]; then
static_config "../lnet/klnds/gnilnd/kgnilnd" \
"timeout=${value}"
else
skip "NETTYPE ${NETTYPE} not supported"
fi
check_parameter "timeout" $value
return $?
}
run_test 310 "Set timeout and verify"
test_311() {
[[ $NETTYPE == kfi* ]] ||
skip "Need kfi network type"
setupall || error "setupall failed"
mkdir -p $DIR/$tdir || error "mkdir failed"
dd if=/dev/zero of=$DIR/$tdir/$tfile bs=1M count=1 oflag=direct ||
error "dd write failed"
local osts=$(osts_nodes)
#define CFS_KFI_FAIL_WAIT_SEND_COMP1 0xF115
do_nodes $osts $LCTL set_param fail_loc=0x8000F115
dd if=$DIR/$tdir/$tfile of=/dev/null bs=1M count=1 ||
error "dd read failed"
rm -f $DIR/$tdir/$tfile
rmdir $DIR/$tdir
cleanupall || error "Failed cleanup"
}
run_test 311 "Fail bulk put in send wait completion"
test_312() {
[[ $NETTYPE == kfi* ]] ||
skip "Need kfi network type"
setupall || error "setupall failed"
mkdir -p $DIR/$tdir || error "mkdir failed"
local osts=$(osts_nodes)
#define CFS_KFI_FAIL_WAIT_SEND_COMP3 0xF117
do_nodes $osts $LCTL set_param fail_loc=0x8000F117
dd if=/dev/zero of=$DIR/$tdir/$tfile bs=1M count=1 oflag=direct ||
error "dd write failed"
local tfile2="$DIR/$tdir/testfile2"
do_nodes $osts $LCTL set_param fail_loc=0x8000F117
dd if=$DIR/$tdir/$tfile of=$tfile2 bs=1M count=1 oflag=direct ||
error "dd read failed"
rm -f $DIR/$tdir/$tfile
rm -f $tfile2
rmdir $DIR/$tdir
cleanupall || error "Failed cleanup"
}
run_test 312 "TAG_RX_OK is possible after TX_FAIL"
test_350() {
reinit_dlc || return $?
add_net ${NETTYPE} ${INTERFACES[0]} || return $?
local my_nid=$($LCTL list_nids)
cat <<EOF > $TMP/sanity-lnet-$testnum.yaml
ping:
nids:
0: $my_nid
EOF
do_lnetctl import -e $TMP/sanity-lnet-$testnum.yaml ||
error "Import failed with rc = $?"
cat <<EOF > $TMP/sanity-lnet-$testnum.yaml
discover:
nids:
0: $my_nid
EOF
do_lnetctl import -e $TMP/sanity-lnet-$testnum.yaml ||
error "Import failed with rc = $?"
cat <<EOF > $TMP/sanity-lnet-$testnum-expected.yaml
peer:
- primary nid: $my_nid
Multi-Rail: true
peer ni:
- nid: $my_nid
state: NA
EOF
$LNETCTL peer show > $TMP/sanity-lnet-$testnum-actual.yaml ||
error "export failed with rc = $?"
compare_yaml_files
}
run_test 350 "Check import --exec of ping/discover"
check_udsp_prio() {
local target_net="${1}"
local target_nid="${2}"
local expect_net="${3}"
local expect_nid="${4}"
local type="${5}"
declare -a nids
declare -a net_prios
declare -a nid_prios
nids=( $($LNETCTL ${type} show -v 5 | awk '/-\s+nid:/{print $NF}' |
xargs echo) )
net_prios=( $($LNETCTL ${type} show -v 5 |
awk '/net priority:/{print $NF}' | xargs echo) )
nid_prios=( $($LNETCTL ${type} show -v 5 |
awk '/nid priority:/{print $NF}' | xargs echo) )
(( ${#nids[@]} != ${#net_prios[@]} )) &&
error "Wrong # net prios ${#nids[@]} != ${#net_prios[@]}"
(( ${#nids[@]} != ${#nid_prios[@]} )) &&
error "Wrong # nid prios ${#nids[@]} != ${#nid_prios[@]}"
local i
for ((i = 0; i < ${#nids[@]}; i++)); do
[[ -n ${target_net} ]] &&
[[ ${nids[i]##*@} != "${target_net}" ]] &&
continue
[[ -n ${target_nid} ]] &&
[[ ${nids[i]} != "${target_nid}" ]] &&
continue
echo "${nids[i]}: net_prio ${net_prios[i]} expect ${expect_net}"
(( net_prios[i] != expect_net )) &&
error "Wrong net priority \"${net_prios[i]}\" expect ${expect_net}"
echo "${nids[i]}: nid_prio ${nid_prios[i]} expect ${expect_nid}"
(( nid_prios[i] != expect_nid )) &&
error "Wrong nid priority \"${nid_prios[i]}\" expect ${expect_nid}"
done
return 0
}
check_peer_udsp_prio() {
check_udsp_prio "${1}" "${2}" "${3}" "${4}" "peer"
}
check_net_udsp_prio() {
check_udsp_prio "${1}" "${2}" "${3}" "${4}" "net"
}
test_400() {
reinit_dlc || return $?
do_lnetctl udsp add --src tcp --priority 0 ||
error "Failed to add udsp rule"
do_lnetctl udsp del --idx 0 ||
error "Failed to del udsp rule"
unload_modules
}
run_test 400 "Check for udsp add/delete net rule without net num"
test_401() {
reinit_dlc || return $?
do_lnetctl net add --net ${NETTYPE} --if ${INTERFACES[0]} ||
error "Failed to add net"
do_lnetctl udsp add --dst ${NETTYPE} --prio 1 ||
error "Failed to add peer net priority rule"
do_lnetctl discover $($LCTL list_nids | head -n 1) ||
error "Failed to discover peer"
check_peer_udsp_prio "${NETTYPE}" "" "1" "-1"
return 0
}
run_test 401 "Discover peer after adding peer net UDSP rule"
test_402() {
reinit_dlc || return $?
do_lnetctl udsp add --dst kfi --priority 0 ||
error "Failed to add UDSP rule"
do_lnetctl peer add --prim 402@kfi ||
error "Failed to add peer"
return 0
}
run_test 402 "Destination net rule should not panic"
test_403() {
local rule_type
for rule_type in src dst; do
reinit_dlc || return $?
local nid
if [[ ${NETTYPE} == kfi* ]] || [[ ${NETTYPE} == gni* ]]; then
nid=401@kfi
else
nid=1.1.1.1@${NETTYPE}
fi
for arg in ${NETTYPE} ${nid}; do
cat <<EOF > $TMP/$TESTSUITE-$testnum-expected.yaml
udsp:
- idx: 0
${rule_type}: ${arg}
action:
priority: 1
EOF
do_lnetctl udsp add --${rule_type} ${arg} --prio 1 ||
error "udsp add failed rc=$?"
$LNETCTL udsp show > $TMP/$TESTSUITE-$testnum-actual.yaml
compare_yaml_files || return $?
do_lnetctl udsp del --idx 0 ||
error "udsp del failed rc=$?"
do_lnetctl udsp show
[[ -z $($LNETCTL udsp show) ]] ||
error "Unexpected udsp show output"
done
done
return 0
}
run_test 403 "Add and delete udsp net and nid rules"
check_udsp_priorities() {
local rule_type="$1"
if [[ $rule_type == both ]]; then
check_net_udsp_prio "$net" "$nid" "1" "-1"
check_peer_udsp_prio "$net" "$nid" "1" "-1"
elif [[ ${rule_type} == src ]]; then
check_net_udsp_prio "$net" "$nid" "1" "-1"
else
check_peer_udsp_prio "$net" "$nid" "1" "-1"
fi
}
do_udsp_test() {
local rule_type="$1"
local net="$2"
local nid="$3"
local arg arg_type
if [[ -n $net ]]; then
arg=$net
arg_type="net"
else
arg=$nid
arg_type="nid"
fi
[[ -n $net ]] && arg=$net || arg=$nid
if [[ $rule_type == both ]]; then
do_lnetctl udsp add --src $arg --prio 1 ||
error "Failed to add $rule_type $arg_type priority rule"
do_lnetctl udsp add --dst $arg --prio 1 ||
error "Failed to add $rule_type $arg_type priority rule"
else
do_lnetctl udsp add --$rule_type $arg --prio 1 ||
error "Failed to add $rule_type $arg_type priority rule"
fi
do_lnetctl discover "$($LCTL list_nids | head -n 1)" ||
error "failed to discover myself"
check_udsp_priorities "$rule_type"
$LNETCTL export --backup > $TMP/$TESTSUITE-$testnum-expected.yaml ||
error "export failed"
reinit_dlc || return $?
$LNETCTL import $TMP/$TESTSUITE-$testnum-expected.yaml ||
error "Failed to import from backup"
$LNETCTL export --backup > $TMP/$TESTSUITE-$testnum-actual.yaml ||
error "export failed"
compare_yaml_files || error "Unexpected config after import"
do_lnetctl discover "$($LCTL list_nids | head -n 1)" ||
error "failed to discover myself"
check_udsp_priorities "$rule_type"
}
test_404() {
local rule_type nid
for rule_type in src dst both ; do
reinit_dlc || return $?
add_net "${NETTYPE}" "${INTERFACES[0]}"
add_net "${NETTYPE}2" "${INTERFACES[0]}"
nid=$($LCTL list_nids | head -n 1)
do_udsp_test ${rule_type} "${NETTYPE}" "${nid}" || return $?
reinit_dlc || return $?
add_net "${NETTYPE}" "${INTERFACES[0]}"
add_net "${NETTYPE}2" "${INTERFACES[0]}"
nid=$($LCTL list_nids | tail -n 1)
do_lnetctl discover "$($LCTL list_nids | head -n 1)" ||
error "Failed to discover myself"
do_udsp_test ${rule_type} "${NETTYPE}2" "${nid}" || return $?
done
return 0
}
run_test 404 "Check udsp src/dst net rule"
test_410() {
reinit_dlc || return $?
$LCTL mark "$LNETCTL fault reset"
echo "$LNETCTL fault reset"
$LNETCTL fault reset '' &&
error "Command should fail"
return 0
}
run_test 410 "No segfault in lnetctl fault command"
test_450() {
[[ $NETTYPE == tcp* ]] || skip "Need tcp NETTYPE"
reinit_dlc || return $?
setup_fakeif || return $?
add_net "$NETTYPE" "${INTERFACES[0]}" || return $?
add_net "$NETTYPE" "$FAKE_IF" || return $?
$LNETCTL export --backup > $TMP/sanity-lnet-$testnum-expected.yaml
cat <<EOF > $TMP/sanity-lnet-$testnum.yaml
net:
- net type: $NETTYPE
local NI(s):
- interfaces:
0: ${INTERFACES[0]}
1: ${FAKE_IF}
EOF
do_import_test "$TMP/sanity-lnet-$testnum.yaml" || return $?
do_lnetctl import --del $TMP/sanity-lnet-$testnum.yaml ||
error "Import --del failed with rc = $?"
[[ -z $($LCTL list_nids) ]] ||
error "NIDs still configured \"$($LCTL list_nids)\""
cleanup_fakeif
}
run_test 450 "Check import of multiple interfaces"
test_500() {
reinit_dlc || return $?
setup_fakeif || error "Failed to add fake IF"
add_net "tcp" "${INTERFACES[0]}"
add_net "tcp" "${FAKE_IF}"
do_lnetctl discover $($LCTL list_nids | head -n 1) ||
error "Failed to discover self"
add_delay_rule -s *@tcp -d *@tcp -r 1 -l 1 -m PUT
add_drop_rule -s *@tcp -d $($LCTL list_nids | head -n 1) -m PUT \
-e local_timeout -r 1
add_drop_rule -s *@tcp -d $($LCTL list_nids | tail -n 1) -m PUT \
-e local_timeout -r 1
ip link set $FAKE_IF down ||
error "Failed to set link down"
ip link set $FAKE_IF up ||
error "Failed to set link up"
unload_modules
}
run_test 500 "Check deadlock on ping target update"
do_peer_set_health_test() {
local test_val="$1"
shift
local nid_arg="$@"
local max_hval
if [[ ${nid_arg} == --all ]]; then
max_hval=$($LNETCTL peer show -v 2 2>/dev/null |
awk '/health value/{print $NF}' | xargs echo |
sed 's/ /+/g' | bc -l)
else
max_hval=$($LNETCTL peer show ${nid_arg} -v 2 2>/dev/null |
awk '/health value/{print $NF}' | xargs echo |
sed 's/ /+/g' | bc -l)
fi
(( max_hval >= 1000 )) && (( max_hval % 1000 == 0)) ||
error "Unexpected max health value $max_hval"
lnet_health_pre || return $?
do_lnetctl peer set --health $test_val ${nid_arg} ||
error "failed to set health value"
lnet_health_post || return $?
local hval
if [[ ${nid_arg} == --all ]]; then
hval=$($LNETCTL peer show -v 2 2>/dev/null |
awk '/health value/{print $NF}' | xargs echo |
sed 's/ /+/g' | bc -l)
else
hval=$($LNETCTL peer show ${nid_arg} -v 2 2>/dev/null |
awk '/health value/{print $NF}' | xargs echo |
sed 's/ /+/g' | bc -l)
fi
check_remote_health || return $?
echo "hval: $hval max_hval: $max_hval"
(( hval == max_hval )) || error "Failed to reset health to max"
return 0
}
test_501() {
reinit_dlc || return $?
setup_health_test false || return $?
$LCTL set_param debug=-1
add_drop_rule -s *@tcp -d *@tcp -r 1
local test_val
for test_val in 0 500; do
do_peer_set_health_test $test_val --nid ${RNIDS[0]} || return $?
do_peer_set_health_test $test_val --all || return $?
done
cleanup_health_test
}
run_test 501 "Verify lnetctl peer set --health (SR)"
test_502() {
reinit_dlc || return $?
setup_health_test true || return $?
$LCTL set_param debug=-1
add_drop_rule -s *@tcp -d *@tcp -r 1
local test_val
for test_val in 0 500; do
do_peer_set_health_test $test_val --nid ${RNIDS[0]} || return $?
do_peer_set_health_test $test_val --all || return $?
done
cleanup_health_test
}
run_test 502 "Verify lnetctl peer set --health (MR)"
test_550() {
[[ ${NETTYPE} == tcp* ]] ||
skip "Need tcp NETTYPE to fail a single local rail"
(( ${#INTERFACES[@]} >= 2 )) ||
skip "Need at least 2 local interfaces for multi-rail"
setup_router_test lnet_peer_discovery_disabled=1 || return $?
# Add a second local NI on the local net so the gateway is reachable
# via two local rails (multi-rail).
add_net $LOCAL_NET ${INTERFACES[1]} || return $?
LNIDS[1]=$($LCTL list_nids | tail -1)
# Configure this peer as MR, all others stay non-MR
local all_nodes=$(comma_list ${ROUTERS[@]} ${RPEERS[@]} $HOSTNAME)
do_nodes $all_nodes "$LNETCTL peer add --prim ${LNIDS[0]} --nid ${LNIDS[1]}" ||
error "Failed to add peer rc=$?"
config_routes || error "Failed to configure routes rc=$?"
local rpeer=${RPEERS[0]}
local rpeer_nids=( ${RPEER_NIDS[$rpeer]} )
local i
# Ping via the secondary NI so path selection favors it for rpeer
local snid=$($LCTL list_nids | tail -1)
do_lnetctl ping --source $snid ${rpeer_nids[0]} ||
error "Failed to ping from $snid to ${rpeer_nids[0]}"
check_route_aliveness "$HOSTNAME" "up" || return $?
# Drop a marker on the console so we can scope the search for a route
# up->down transition to events that happen after we fail the rail.
# lnet_set_route_aliveness() logs such transitions via CERROR, which
# reaches the console
local marker="DDN-6845-${testnum}-${RANDOM}"
log "$marker"
echo "Fail a send via ${INTERFACES[1]}"
$LCTL net_drop_add -s ${LNIDS[1]} -d "*@$REMOTE_NET" -e local_error -r 1 ||
error "Failed to add $REMOTE_NET drop rule rc=$?"
$LCTL net_drop_add -s ${LNIDS[1]} -d "*@$LOCAL_NET" -e local_error -r 1 ||
error "Failed to add $LOCAL_NET drop rule rc=$?"
echo "Set fail_loc to allow disconnect on simulated error"
#define CFS_FAIL_SOCK_CONN 0xe020
$LCTL set_param fail_loc=0x8000e020
! do_lnetctl ping ${rpeer_nids[0]} ||
error "Ping should have failed"
check_route_aliveness "$HOSTNAME" "down" true &&
error "Route went down after single local-rail failure"
local downtrans
downtrans=$( dmesg | sed -n "/$marker/,\$p" |
grep -c "has gone from up to down")
((downtrans == 0)) ||
error "route marked down on single local-rail failure $downtrans up->down transition(s)"
check_route_aliveness "$HOSTNAME" "up" ||
error "route down after single local-rail failure"
# Regression guard: confirm the route goes down when the gateway is
# unreachable from all local NIs, not just the failed one.
do_node ${ROUTERS[0]} "$LNETCTL lnet unconfigure" ||
error "Failed to unconfigure lnet on ${ROUTERS[0]}"
check_route_aliveness "$HOSTNAME" "down" ||
error "route didn't go down"
cleanup_router_test
}
run_test 550 "DD-off: keep route up when gw reachable via another local NI"
test_600() {
local actual="$TMP/sanity-lnet-$testnum-actual.yaml"
cat <<EOF > $TMP/sanity-lnet-$testnum-expected.yaml
ip2nets:
- net-spec: tcp0
interfaces:
0: eth0
1: eth1
ip-range:
0: 10.0.0.*
- net-spec: o2ib0
ip-range:
0: 192.168.[1-2].*
EOF
$LEGACY2YAML 'tcp0(eth0,eth1) 10.0.0.*; o2ib0 192.168.[1-2].*' \
> $actual || error "lnet_legacy2yaml failed with rc=$?"
compare_yaml_files || error "YAML comparison failed"
}
run_test 600 "lnet_legacy2yaml: basic ip2nets conversion"
test_601() {
local actual="$TMP/sanity-lnet-$testnum-actual.yaml"
cat <<EOF > $TMP/sanity-lnet-$testnum-expected.yaml
route:
- net: tcp0
gateway: 10.0.0.1@o2ib0
hop: 1
EOF
$LEGACY2YAML "routes='tcp0 1 10.0.0.1@o2ib0'" > $actual ||
error "lnet_legacy2yaml failed with rc=$?"
compare_yaml_files || error "YAML comparison failed"
}
run_test 601 "lnet_legacy2yaml: basic routes conversion"
test_602() {
local actual="$TMP/sanity-lnet-$testnum-actual.yaml"
cat <<EOF > $TMP/sanity-lnet-$testnum-expected.yaml
route:
- net: tcp0
gateway: 10.0.0.1@o2ib0
hop: 2
priority: 7
EOF
$LEGACY2YAML "routes='tcp0 2 10.0.0.1@o2ib0:7'" > $actual ||
error "lnet_legacy2yaml failed with rc=$?"
compare_yaml_files || error "YAML comparison failed"
}
run_test 602 "lnet_legacy2yaml: routes with priority"
test_603() {
local actual="$TMP/sanity-lnet-$testnum-actual.yaml"
cat <<EOF > $TMP/sanity-lnet-$testnum-expected.yaml
route:
- net: tcp0
gateway: 10.0.0.1@o2ib0
hop: 1
EOF
$LEGACY2YAML "routes='tcp0 10.0.0.1@o2ib0'" --default-hop \
> $actual || error "lnet_legacy2yaml failed with rc=$?"
compare_yaml_files || error "YAML comparison failed"
}
run_test 603 "lnet_legacy2yaml: default hop flag"
test_604() {
local actual="$TMP/sanity-lnet-$testnum-actual.yaml"
cat <<EOF > $TMP/sanity-lnet-$testnum-expected.yaml
route:
- net: tcp0
gateway: 10.0.0.1@o2ib0
hop: 3
EOF
$LEGACY2YAML --default-hop 3 "routes='tcp0 10.0.0.1@o2ib0'" \
> $actual || error "lnet_legacy2yaml failed with rc=$?"
compare_yaml_files || error "YAML comparison failed"
}
run_test 604 "lnet_legacy2yaml: default hop with explicit value"
test_605() {
local actual="$TMP/sanity-lnet-$testnum-actual.yaml"
cat <<EOF > $TMP/sanity-lnet-$testnum-expected.yaml
ip2nets:
- net-spec: tcp
ip-range:
0: "*.*.*.*"
EOF
$LEGACY2YAML 'tcp *.*.*.*' > $actual ||
error "lnet_legacy2yaml failed with rc=$?"
compare_yaml_files || error "YAML comparison failed"
}
run_test 605 "lnet_legacy2yaml: quote leading star in IP pattern"
test_606() {
local actual="$TMP/sanity-lnet-$testnum-actual.yaml"
local ip_file="$TMP/sanity-lnet-$testnum-ip.rules"
local routes_file="$TMP/sanity-lnet-$testnum-routes.rules"
echo 'tcp0 10.1.0.*' > "$ip_file"
echo 'tcp0 1 10.1.0.1@o2ib0:9' > "$routes_file"
cat <<EOF > $TMP/sanity-lnet-$testnum-expected.yaml
ip2nets:
- net-spec: tcp0
ip-range:
0: 10.1.0.*
route:
- net: tcp0
gateway: 10.1.0.1@o2ib0
hop: 1
priority: 9
EOF
$LEGACY2YAML -i "$ip_file" -r "$routes_file" > $actual ||
error "lnet_legacy2yaml failed with rc=$?"
compare_yaml_files || error "YAML comparison failed"
rm -f "$ip_file" "$routes_file"
}
run_test 606 "lnet_legacy2yaml: file inputs"
test_607() {
local actual="$TMP/sanity-lnet-$testnum-actual.yaml"
local opts
opts="options lnet 'ip2nets=\"tcp0 10.0.0.*\""
opts+=" routes=\"tcp0 1 10.0.0.2@o2ib0\"'"
cat <<EOF > $TMP/sanity-lnet-$testnum-expected.yaml
ip2nets:
- net-spec: tcp0
ip-range:
0: 10.0.0.*
route:
- net: tcp0
gateway: 10.0.0.2@o2ib0
hop: 1
EOF
$LEGACY2YAML "$opts" > $actual ||
error "lnet_legacy2yaml failed with rc=$?"
compare_yaml_files || error "YAML comparison failed"
}
run_test 607 \
"lnet_legacy2yaml: options line with combined ip2nets and routes"
test_608() {
local output
output=$($LEGACY2YAML 'tcp0' 2>&1)
local rc=$?
((rc == 2)) || error "Expected exit code 2, got $rc"
echo "$output" | grep -iq 'error' ||
error "Expected error message in output"
}
run_test 608 "lnet_legacy2yaml: error on invalid rule (missing IP pattern)"
test_609() {
local output
# Run with no stdin and no args - expect error
output=$(echo -n "" | $LEGACY2YAML 2>&1)
local rc=$?
((rc == 1)) || error "Expected exit code 1, got $rc"
echo "$output" | grep -iq "no ip2nets, routes, or networks data" ||
error "Expected 'no data' message in output"
}
run_test 609 "lnet_legacy2yaml: error on no input"
test_610() {
local actual="$TMP/sanity-lnet-$testnum-actual.yaml"
local ip1="$TMP/sanity-lnet-$testnum-ip1.rules"
local ip2="$TMP/sanity-lnet-$testnum-ip2.rules"
local r1="$TMP/sanity-lnet-$testnum-r1.rules"
echo 'tcp0 10.0.0.*' > "$ip1"
echo 'o2ib0 192.168.0.*' > "$ip2"
echo 'tcp0 1 10.0.0.1@o2ib0' > "$r1"
cat <<EOF > $TMP/sanity-lnet-$testnum-expected.yaml
ip2nets:
- net-spec: tcp0
ip-range:
0: 10.0.0.*
- net-spec: o2ib0
ip-range:
0: 192.168.0.*
route:
- net: tcp0
gateway: 10.0.0.1@o2ib0
hop: 1
- net: o2ib0
gateway: 192.168.0.1@tcp0
hop: 1
EOF
$LEGACY2YAML -i "$ip1" -i "$ip2" -r "$r1" \
"routes='o2ib0 1 192.168.0.1@tcp0'" > $actual ||
error "lnet_legacy2yaml failed with rc=$?"
compare_yaml_files || error "YAML comparison failed"
rm -f "$ip1" "$ip2" "$r1"
}
run_test 610 \
"lnet_legacy2yaml: multiple files and inline, check order"
test_611() {
local output
output=$($LEGACY2YAML "routes='tcp0 0 10.0.0.1@o2ib0'" 2>&1)
local rc=$?
((rc == 2)) || error "Expected exit code 2, got $rc"
echo "$output" | grep -iq "hopcount" ||
error "Expected hopcount error in output"
}
run_test 611 "lnet_legacy2yaml: invalid hopcount"
test_612() {
local output
output=$($LEGACY2YAML 'tcp0 10.0.0' 2>&1)
local rc=$?
((rc == 2)) || error "Expected exit code 2, got $rc"
echo "$output" | grep -iq "invalid.*ip pattern" ||
error "Expected invalid IP pattern error in output"
}
run_test 612 "lnet_legacy2yaml: malformed IP pattern"
test_613() {
local actual="$TMP/sanity-lnet-$testnum-actual.yaml"
cat <<EOF > $TMP/sanity-lnet-$testnum-expected.yaml
ip2nets:
- net-spec: tcp0
ip-range:
0: 10.0.0.*
1: 10.0.1.*
EOF
$LEGACY2YAML 'tcp0 10.0.0.* 10.0.1.*' > $actual ||
error "lnet_legacy2yaml failed with rc=$?"
compare_yaml_files || error "YAML comparison failed"
}
run_test 613 \
"lnet_legacy2yaml: multiple IP patterns in single rule"
test_614() {
local actual="$TMP/sanity-lnet-$testnum-actual.yaml"
cat <<EOF > $TMP/sanity-lnet-$testnum-expected.yaml
ip2nets:
- net-spec: tcp0
interfaces:
0: eth0
1: eth1
ip-range:
0: 10.0.0.*
1: 10.0.1.*
EOF
$LEGACY2YAML 'tcp0(eth0,eth1) 10.0.0.* 10.0.1.*' > $actual ||
error "lnet_legacy2yaml failed with rc=$?"
compare_yaml_files || error "YAML comparison failed"
}
run_test 614 "lnet_legacy2yaml: interfaces enumeration with multiple IP ranges"
test_615() {
local actual="$TMP/sanity-lnet-$testnum-actual.yaml"
local modprobe_file="$TMP/sanity-lnet-$testnum-modprobe.conf"
cat <<EOF > "$modprobe_file"
# Modprobe configuration file for LNet
options lnet ip2nets="tcp0(eth0) 10.1.0.*"
# Routes configuration
options lnet routes="tcp0 1 10.1.0.1@o2ib0:9"
# Line continuation test
options lnet \\
ip2nets="o2ib0 192.168.1.*"
EOF
cat <<EOF > $TMP/sanity-lnet-$testnum-expected.yaml
ip2nets:
- net-spec: tcp0
interfaces:
0: eth0
ip-range:
0: 10.1.0.*
- net-spec: o2ib0
ip-range:
0: 192.168.1.*
route:
- net: tcp0
gateway: 10.1.0.1@o2ib0
hop: 1
priority: 9
EOF
$LEGACY2YAML -m "$modprobe_file" > $actual ||
error "lnet_legacy2yaml failed with rc=$?"
compare_yaml_files || error "YAML comparison failed"
rm -f "$modprobe_file"
}
run_test 615 "lnet_legacy2yaml: modprobe configuration file input"
test_616() {
local actual="$TMP/sanity-lnet-$testnum-actual.yaml"
local modprobe_file="$TMP/sanity-lnet-$testnum-modprobe.conf"
local ip_file="$TMP/sanity-lnet-$testnum-ip.rules"
cat <<EOF > "$modprobe_file"
options lnet ip2nets="tcp0 10.0.0.*"
options lnet routes="tcp0 1 10.0.0.1@o2ib0"
EOF
echo 'o2ib0 192.168.0.*' > "$ip_file"
cat <<EOF > $TMP/sanity-lnet-$testnum-expected.yaml
ip2nets:
- net-spec: tcp0
ip-range:
0: 10.0.0.*
- net-spec: o2ib0
ip-range:
0: 192.168.0.*
route:
- net: tcp0
gateway: 10.0.0.1@o2ib0
hop: 1
EOF
$LEGACY2YAML -m "$modprobe_file" -i "$ip_file" > $actual ||
error "lnet_legacy2yaml failed with rc=$?"
compare_yaml_files || error "YAML comparison failed"
rm -f "$modprobe_file" "$ip_file"
}
run_test 616 \
"lnet_legacy2yaml: combining modprobe file with explicit ip2nets file"
test_617() {
reinit_dlc || return $?
local modprobe_file="$TMP/sanity-lnet-$testnum-modprobe.conf"
local yaml_file="$TMP/sanity-lnet-$testnum.yaml"
local export_file="$TMP/sanity-lnet-$testnum-expected.yaml"
local actual_file="$TMP/sanity-lnet-$testnum-actual.yaml"
local if0_ip
local ip_pattern
# Get IP address of the first interface and create pattern
if0_ip=$(ip -o -4 a s ${INTERFACES[0]} | awk '{print $4}' | \
sed 's/\/.*//' | head -n 1)
[[ -n $if0_ip ]] || error "Cannot determine IP for ${INTERFACES[0]}"
# Create IP pattern matching the network (e.g., 192.168.1.*)
ip_pattern=$(awk -F. '{print $1"."$2"."$3".*"}' <<< "${if0_ip}")
echo "Configuring network ${NETTYPE} on ${INTERFACES[0]}"
add_net "${NETTYPE}" "${INTERFACES[0]}" || return $?
$LNETCTL export --backup > "$export_file" ||
error "lnetctl export failed"
echo "Generating modprobe configuration"
cat <<EOF > "$modprobe_file"
options lnet ip2nets="${NETTYPE}(${INTERFACES[0]}) ${ip_pattern}"
EOF
echo "Generated modprobe configuration:"
cat "$modprobe_file"
echo "Converting modprobe config to YAML"
$LEGACY2YAML -m "$modprobe_file" > "$yaml_file" ||
error "lnet_legacy2yaml failed with rc=$?"
echo "Generated YAML configuration:"
cat "$yaml_file"
reinit_dlc || return $?
do_lnetctl import < "$yaml_file" ||
error "lnetctl import failed"
$LNETCTL export --backup > "$actual_file" ||
error "lnetctl export after import failed"
compare_yaml_files ||
error "Round-trip YAML comparison failed"
rm -f "$modprobe_file" "$yaml_file" "$export_file" "$actual_file"
}
run_test 617 \
"lnet_legacy2yaml: round-trip tunables from LNet to modprobe to YAML"
test_618() {
cat <<EOF > $TMP/sanity-lnet-$testnum-expected.yaml
route:
- net: tcp7
gateway: 10.3.3.[6-12]@tcp
hop: 8
priority: 20
EOF
local cfg="tcp7: { gateway: 10.3.3.[6-12]@tcp, priority: 20, hop: 8 }"
local actual="$TMP/sanity-lnet-$testnum-actual.yaml"
$LEGACY2YAML -r <(echo "$cfg") > $actual ||
error "lnet_legacy2yaml failed with rc=$?"
compare_yaml_files || error "YAML comparison failed"
}
run_test 618 "lnet_legacy2yaml: alternative route format"
test_619() {
local actual="$TMP/sanity-lnet-$testnum-actual.yaml"
local routes_file="$TMP/sanity-lnet-$testnum-routes.conf"
cat <<EOF > "$routes_file"
# Alternative route format examples
tcp1: { gateway: 10.1.1.2@tcp0, priority: 3 } # High priority route
# This is a comment line
tcp4: { gateway: 10.3.3.4@tcp }
tcp6: { gateway: 10.3.3.6@tcp, hop: 2, priority: 5 }
tcp7: { gateway: 10.3.3.[6-12]@tcp, priority: 20, hop: 8 }
EOF
cat <<EOF > $TMP/sanity-lnet-$testnum-expected.yaml
route:
- net: tcp1
gateway: 10.1.1.2@tcp0
priority: 3
- net: tcp4
gateway: 10.3.3.4@tcp
- net: tcp6
gateway: 10.3.3.6@tcp
hop: 2
priority: 5
- net: tcp7
gateway: 10.3.3.[6-12]@tcp
hop: 8
priority: 20
EOF
$LEGACY2YAML -r "$routes_file" > $actual ||
error "lnet_legacy2yaml failed with rc=$?"
compare_yaml_files || error "YAML comparison failed"
rm -f "$routes_file"
}
run_test 619 "lnet_legacy2yaml: multiple alternative format routes from file"
test_620() {
local actual="$TMP/sanity-lnet-$testnum-actual.yaml"
local routes_file="$TMP/sanity-lnet-$testnum-routes.conf"
cat <<EOF > "$routes_file"
# Mixed route formats
tcp1: { gateway: 10.1.1.2@tcp0, priority: 3 }
tcp0 1 10.2.2.3@tcp1
tcp6: { gateway: 10.3.3.6@tcp, hop: 2, priority: 5 }
tcp2 2 10.4.4.5@tcp0:7
EOF
cat <<EOF > $TMP/sanity-lnet-$testnum-expected.yaml
route:
- net: tcp1
gateway: 10.1.1.2@tcp0
priority: 3
- net: tcp0
gateway: 10.2.2.3@tcp1
hop: 1
- net: tcp6
gateway: 10.3.3.6@tcp
hop: 2
priority: 5
- net: tcp2
gateway: 10.4.4.5@tcp0
hop: 2
priority: 7
EOF
$LEGACY2YAML -r "$routes_file" > $actual ||
error "lnet_legacy2yaml failed with rc=$?"
compare_yaml_files || error "YAML comparison failed"
rm -f "$routes_file"
}
run_test 620 "lnet_legacy2yaml: mixed legacy and alternative route formats"
test_621() {
local cfg="tcp0: { gateway: 10.0.0.1@o2ib0, hop: 0 }"
local output
output=$($LEGACY2YAML -r <(echo "$cfg") 2>&1)
local rc=$?
((rc == 2)) || error "Expected exit code 2, got $rc"
echo "$output" | grep -iq "hopcount.*out of range" ||
error "Expected hopcount out of range error in output"
}
run_test 621 "lnet_legacy2yaml: alternative format with invalid hopcount"
test_622() {
local actual="$TMP/sanity-lnet-$testnum-actual.yaml"
local cfg="tcp5: { gateway: 10.5.5.5@tcp0 }"
cat <<EOF > $TMP/sanity-lnet-$testnum-expected.yaml
route:
- net: tcp5
gateway: 10.5.5.5@tcp0
hop: 3
EOF
$LEGACY2YAML -r <(echo "$cfg") --default-hop 3 > $actual ||
error "lnet_legacy2yaml failed with rc=$?"
compare_yaml_files || error "YAML comparison failed"
}
run_test 622 "lnet_legacy2yaml: alternative format with default-hop flag"
test_623() {
local actual="$TMP/sanity-lnet-$testnum-actual.yaml"
cat <<EOF > $TMP/sanity-lnet-$testnum-expected.yaml
net:
- net type: tcp
EOF
$LEGACY2YAML 'options lnet networks=tcp' > $actual ||
error "lnet_legacy2yaml failed with rc=$?"
compare_yaml_files || error "YAML comparison failed"
}
run_test 623 "lnet_legacy2yaml: basic networks conversion"
test_624() {
local actual="$TMP/sanity-lnet-$testnum-actual.yaml"
cat <<EOF > $TMP/sanity-lnet-$testnum-expected.yaml
net:
- net type: tcp0
local NI(s):
- interfaces:
0: eth0
EOF
$LEGACY2YAML 'options lnet networks=tcp0(eth0)' > $actual ||
error "lnet_legacy2yaml failed with rc=$?"
compare_yaml_files || error "YAML comparison failed"
}
run_test 624 "lnet_legacy2yaml: networks with interfaces"
test_625() {
local actual="$TMP/sanity-lnet-$testnum-actual.yaml"
cat <<EOF > $TMP/sanity-lnet-$testnum-expected.yaml
net:
- net type: tcp0
local NI(s):
- interfaces:
0: eth0
- net type: o2ib
local NI(s):
- interfaces:
0: ib0
EOF
$LEGACY2YAML 'options lnet networks=tcp0(eth0),o2ib(ib0)' > $actual ||
error "lnet_legacy2yaml failed with rc=$?"
compare_yaml_files || error "YAML comparison failed"
}
run_test 625 "lnet_legacy2yaml: multiple networks"
test_626() {
local actual="$TMP/sanity-lnet-$testnum-actual.yaml"
local modprobe_file="$TMP/sanity-lnet-$testnum-modprobe.conf"
cat <<EOF > "$modprobe_file"
options lnet networks=tcp0(eth0),o2ib(ib0)
options lnet ip2nets="tcp1(eth1) 192.168.1.*"
options lnet routes="tcp0 1 10.0.0.1@o2ib0"
EOF
cat <<EOF > $TMP/sanity-lnet-$testnum-expected.yaml
net:
- net type: tcp0
local NI(s):
- interfaces:
0: eth0
- net type: o2ib
local NI(s):
- interfaces:
0: ib0
ip2nets:
- net-spec: tcp1
interfaces:
0: eth1
ip-range:
0: 192.168.1.*
route:
- net: tcp0
gateway: 10.0.0.1@o2ib0
hop: 1
EOF
$LEGACY2YAML -m "$modprobe_file" > $actual ||
error "lnet_legacy2yaml failed with rc=$?"
compare_yaml_files || error "YAML comparison failed"
rm -f "$modprobe_file"
}
run_test 626 "lnet_legacy2yaml: mixed networks, ip2nets, and routes"
test_627() {
local output
output=$($LEGACY2YAML 'options lnet networks=tcp0(eth0' 2>&1)
local rc=$?
((rc == 2)) || error "Expected exit code 2, got $rc"
echo "$output" | grep -iq "Invalid network specification" ||
error "Expected invalid network specification error in output"
}
run_test 627 "lnet_legacy2yaml: invalid networks format"
test_628() {
local output
output=$($LEGACY2YAML 'tcp0(eth0) 10.0.0.* tcp1(eth1) 192.168.1.*' 2>&1)
local rc=$?
((rc == 2)) || error "Expected exit code 2, got $rc"
echo "$output" | grep -iq "Multiple network rules must be separated" ||
error "Expected missing delimiter error in output"
}
run_test 628 "lnet_legacy2yaml: missing delimiter error"
test_629() {
local output
output=$($LEGACY2YAML 'options lnet routes="tcp0 1 10.0.0.1@o2ib0 10.0.0.2@o2ib0"' 2>&1)
local rc=$?
((rc == 2)) || error "Expected exit code 2, got $rc"
echo "$output" | grep -iq "multiple gateway specifications found" ||
error "Expected multiple gateway error in output"
}
run_test 629 "lnet_legacy2yaml: multiple gateways error"
complete_test $SECONDS
cleanup_testsuite
exit_status
