Viewing: llog_osd.c
// SPDX-License-Identifier: GPL-2.0
/*
* Copyright (c) 2003, 2010, Oracle and/or its affiliates. All rights reserved.
* Use is subject to license terms.
*
* Copyright (c) 2012, 2017, Intel Corporation.
*/
/*
* This file is part of Lustre, http://www.lustre.org/
*
* Low level llog routines on top of OSD API
*
* This file provides set of methods for llog operations on top of
* dt_device. It contains all supported llog_operations interfaces and
* supplimental functions.
*
* Author: Alexey Zhuravlev <alexey.zhuravlev@intel.com>
* Author: Mikhail Pershin <mike.pershin@intel.com>
*/
#define DEBUG_SUBSYSTEM S_LOG
#include <linux/delay.h>
#include <dt_object.h>
#include <llog_swab.h>
#include <lustre_fid.h>
#include <obd.h>
#include <obd_class.h>
#include "llog_internal.h"
#include "local_storage.h"
/**
* llog_osd_declare_new_object() - Implementation of the lop_declare_create
* @env: execution environment
* @los: local_storage for bottom storage device
* @o: dt_object to create
* @th: current transaction handle
*
* Implementation of the llog_operations::lop_declare_create
* This function is a wrapper over local_storage API function
* local_object_declare_create().
*
* Return:
* * %0 on successful declaration of the new object
* * %negative error if declaration was failed
*/
static int llog_osd_declare_new_object(const struct lu_env *env,
struct local_oid_storage *los,
struct dt_object *o,
struct thandle *th)
{
struct llog_thread_info *lgi = llog_info(env);
lgi->lgi_attr.la_valid = LA_MODE;
lgi->lgi_attr.la_mode = S_IFREG | S_IRUGO | S_IWUSR;
lgi->lgi_dof.dof_type = dt_mode_to_dft(S_IFREG);
return local_object_declare_create(env, los, o, &lgi->lgi_attr,
&lgi->lgi_dof, th);
}
/**
* llog_osd_create_new_object() - Implementation of the lop_create
* @env: execution environment
* @los: local_storage for bottom storage device
* @o: dt_object to create
* @th: current transaction handle
*
* Implementation of the llog_operations::lop_create
* This function is a wrapper over local_storage API function
* local_object_create().
*
* Return:
* * %0 on successful creation of the new object
* * %negative error if creation was failed
*/
static int llog_osd_create_new_object(const struct lu_env *env,
struct local_oid_storage *los,
struct dt_object *o,
struct thandle *th)
{
struct llog_thread_info *lgi = llog_info(env);
lgi->lgi_attr.la_valid = LA_MODE | LA_CTIME | LA_MTIME | LA_ATIME;
lgi->lgi_attr.la_mode = S_IFREG | S_IRUGO | S_IWUSR;
lgi->lgi_attr.la_ctime = lgi->lgi_attr.la_mtime =
lgi->lgi_attr.la_atime = ktime_get_real_seconds();
lgi->lgi_dof.dof_type = dt_mode_to_dft(S_IFREG);
return local_object_create(env, los, o, &lgi->lgi_attr,
&lgi->lgi_dof, th);
}
/**
* llog_osd_exist() - Implementation of the llog_operations::lop_exist
* @handle: llog handle of the current llog
*
* This function checks that llog exists on storage.
*
* Return:
* * %true if llog object exists and is not just destroyed
* * %false if llog doesn't exist or just destroyed
*/
static int llog_osd_exist(struct llog_handle *handle)
{
LASSERT(handle->lgh_obj);
return dt_object_exists(handle->lgh_obj) && !handle->lgh_destroyed;
}
/**
* llog_osd_pad() - Write a padding record to the llog
* @env: execution environment
* @o: dt_object to create
* @off: pointer to the padding start offset [in,out]
* @len: padding length
* @index: index of the padding record in a llog
* @th: current transaction handle
*
* This function writes a padding record to the end of llog. That may
* be needed if llog contains records of variable size, e.g. config logs
* or changelogs.
* The padding record just aligns llog to the llog chunk_size boundary if
* the current record doesn't fit in the remaining space.
*
* It allocates full length to avoid two separate writes for header and tail.
* Such 2-steps scheme needs extra protection and complex error handling.
*
* Return:
* * %0 on successful padding write
* * %negative error if write failed
*/
static int llog_osd_pad(const struct lu_env *env, struct dt_object *o,
loff_t *off, int len, int index, struct thandle *th)
{
struct llog_thread_info *lgi = llog_info(env);
struct llog_rec_hdr *rec;
struct llog_rec_tail *tail;
int rc;
ENTRY;
LASSERT(th);
LASSERT(off);
LASSERT(len >= LLOG_MIN_REC_SIZE && (len & 0x7) == 0);
OBD_ALLOC(rec, len);
if (rec == NULL)
RETURN(-ENOMEM);
rec->lrh_len = len;
rec->lrh_index = index;
rec->lrh_type = LLOG_PAD_MAGIC;
tail = llog_get_rec_tail(rec);
tail->lrt_len = len;
tail->lrt_index = index;
lgi->lgi_buf.lb_buf = rec;
lgi->lgi_buf.lb_len = len;
rc = dt_record_write(env, o, &lgi->lgi_buf, off, th);
if (rc)
CERROR("%s: error writing padding record: rc = %d\n",
o->do_lu.lo_dev->ld_obd->obd_name, rc);
OBD_FREE(rec, len);
RETURN(rc);
}
/**
* llog_osd_read_header() - Implementation of the lop_read_header
* @env: execution environment
* @handle: llog handle of the current llog
*
* Implementation of the llog_operations::lop_read_header
* This function reads the current llog header from the bottom storage
* device.
*
* Return:
* * %0 on successful header read
* * %negative error if read failed
*/
static int llog_osd_read_header(const struct lu_env *env,
struct llog_handle *handle)
{
struct llog_rec_hdr *llh_hdr;
struct dt_object *o;
struct llog_thread_info *lgi;
enum llog_flag flags;
int rc;
ENTRY;
o = handle->lgh_obj;
LASSERT(o);
lgi = llog_info(env);
dt_read_lock(env, o, 0);
rc = dt_attr_get(env, o, &lgi->lgi_attr);
if (rc)
GOTO(unlock, rc = -EIO);
LASSERT(lgi->lgi_attr.la_valid & LA_SIZE);
if (lgi->lgi_attr.la_size == 0) {
CDEBUG(D_HA, "not reading header from 0-byte log\n");
GOTO(unlock, rc = LLOG_EEMPTY);
}
flags = handle->lgh_hdr->llh_flags;
lgi->lgi_off = 0;
lgi->lgi_buf.lb_buf = handle->lgh_hdr;
lgi->lgi_buf.lb_len = handle->lgh_hdr_size;
rc = dt_read(env, o, &lgi->lgi_buf, &lgi->lgi_off);
llh_hdr = &handle->lgh_hdr->llh_hdr;
if (rc < 0) {
CERROR("%s: can't read llog "DFID" header: rc = %d\n",
o->do_lu.lo_dev->ld_obd->obd_name,
PFID(lu_object_fid(&o->do_lu)), rc);
GOTO(unlock, rc = -EIO);
}
if (rc < sizeof(*llh_hdr) || rc < LLOG_MIN_CHUNK_SIZE) {
/* consider short header as non-initialized llog */
CERROR("%s: llog "DFID" header too small: rc = %d\n",
o->do_lu.lo_dev->ld_obd->obd_name,
PFID(lu_object_fid(&o->do_lu)), rc);
/* caller flags to be initialized */
handle->lgh_hdr->llh_flags = flags;
GOTO(unlock, rc = LLOG_EEMPTY);
}
if (LLOG_REC_HDR_NEEDS_SWABBING(llh_hdr))
lustre_swab_llog_hdr(handle->lgh_hdr);
if (llh_hdr->lrh_type != LLOG_HDR_MAGIC) {
CERROR("%s: bad log %s "DFID" header magic: %#x "
"(expected %#x)\n", o->do_lu.lo_dev->ld_obd->obd_name,
handle->lgh_name ? handle->lgh_name : "",
PFID(lu_object_fid(&o->do_lu)),
llh_hdr->lrh_type, LLOG_HDR_MAGIC);
GOTO(unlock, rc = -EINVAL);
} else if (llh_hdr->lrh_len < LLOG_MIN_CHUNK_SIZE ||
llh_hdr->lrh_len > handle->lgh_hdr_size) {
CERROR("%s: incorrectly sized log %s "DFID" header: "
"%#x (expected at least %#x)\n"
"you may need to re-run lconf --write_conf.\n",
o->do_lu.lo_dev->ld_obd->obd_name,
handle->lgh_name ? handle->lgh_name : "",
PFID(lu_object_fid(&o->do_lu)),
llh_hdr->lrh_len, LLOG_MIN_CHUNK_SIZE);
GOTO(unlock, rc = -EINVAL);
} else if (LLOG_HDR_TAIL(handle->lgh_hdr)->lrt_index >
LLOG_HDR_BITMAP_SIZE(handle->lgh_hdr) ||
LLOG_HDR_TAIL(handle->lgh_hdr)->lrt_len !=
llh_hdr->lrh_len) {
CERROR("%s: incorrectly sized log %s "DFID" tailer: "
"%#x : rc = %d\n",
o->do_lu.lo_dev->ld_obd->obd_name,
handle->lgh_name ? handle->lgh_name : "",
PFID(lu_object_fid(&o->do_lu)),
LLOG_HDR_TAIL(handle->lgh_hdr)->lrt_len, -EIO);
GOTO(unlock, rc = -EINVAL);
}
handle->lgh_hdr->llh_flags |= (flags & LLOG_F_EXT_MASK);
handle->lgh_last_idx = LLOG_HDR_TAIL(handle->lgh_hdr)->lrt_index;
rc = 0;
unlock:
dt_read_unlock(env, o);
RETURN(rc);
}
/**
* llog_osd_declare_write_rec() - Implementation of the lop_declare_write
* @env: execution environment
* @loghandle: llog handle of the current llog
* @rec: llog record header. This is a real header of the full
* llog record to write. This is the beginning of buffer to write,
* the length of buffer is stored in @rec::lrh_len
* @idx: index of the llog record. If @idx == -1 then this is append case,
* otherwise @idx is the index of record to modify
* @th: current transaction handle
*
* This function declares the new record write.
* Implementation of the llog_operations::lop_declare_write
*
* Return:
* * %0 on successful declaration
* * %negative error if declaration failed
*/
static int llog_osd_declare_write_rec(const struct lu_env *env,
struct llog_handle *loghandle,
struct llog_rec_hdr *rec,
int idx, struct thandle *th)
{
struct llog_thread_info *lgi = llog_info(env);
__u32 chunk_size;
struct dt_object *o;
int rc;
ENTRY;
LASSERT(env);
LASSERT(th);
LASSERT(loghandle);
LASSERT(rec);
LASSERT(rec->lrh_len <= loghandle->lgh_ctxt->loc_chunk_size);
o = loghandle->lgh_obj;
LASSERT(o);
chunk_size = loghandle->lgh_ctxt->loc_chunk_size;
lgi->lgi_buf.lb_len = chunk_size;
lgi->lgi_buf.lb_buf = NULL;
/* each time we update header */
rc = dt_declare_record_write(env, o, &lgi->lgi_buf, 0, th);
if (rc || idx == 0) /* if error or just header */
RETURN(rc);
/**
* the pad record can be inserted so take into account double
* record size: pad and the actual record into a new block
*/
lgi->lgi_buf.lb_len = rec->lrh_len * 2;
lgi->lgi_buf.lb_buf = NULL;
/* XXX: implement declared window or multi-chunks approach */
rc = dt_declare_record_write(env, o, &lgi->lgi_buf, -1, th);
RETURN(rc);
}
/**
* llog_osd_write_rec() - Implementation of the llog_operations::lop_write
* @env: execution environment
* @loghandle: llog handle of the current llog
* @rec: llog record header. This is a real header of the full llog record to
* write. This is the bginning of buffer to write, the length of buffer is
* stored in @rec::lrh_len
* @reccookie: pointer to the cookie to return back if needed. It is used for
* further cancel of this llog record. [in,out]
* @idx: index of the llog record. If @idx == -1 then this is append case,
* otherwise @idx is the index of record to modify
* @th: current transaction handle
*
* This function writes the new record in the llog or modify the existed one.
*
* Return:
* * %0 on successful write && @reccookie == NULL
* * %1 on successful write && @reccookie != NULL
* * %negative error if write failed
*/
static int llog_osd_write_rec(const struct lu_env *env,
struct llog_handle *loghandle,
struct llog_rec_hdr *rec,
struct llog_cookie *reccookie,
int idx, struct thandle *th)
{
struct llog_thread_info *lgi = llog_info(env);
struct llog_log_hdr *llh;
int reclen = rec->lrh_len;
int index, rc;
struct llog_rec_tail *lrt;
struct dt_object *o;
__u32 chunk_size;
size_t left;
__u32 orig_last_idx;
bool pad = false;
ENTRY;
llh = loghandle->lgh_hdr;
o = loghandle->lgh_obj;
chunk_size = llh->llh_hdr.lrh_len;
CDEBUG(D_OTHER, "new record %x to "DFID"\n",
rec->lrh_type, PFID(lu_object_fid(&o->do_lu)));
if (!llog_osd_exist(loghandle))
RETURN(-ENOENT);
/* record length should not bigger than */
if (reclen > loghandle->lgh_hdr->llh_hdr.lrh_len)
RETURN(-E2BIG);
/* sanity check for fixed-records llog */
if (idx != LLOG_HEADER_IDX && (llh->llh_flags & LLOG_F_IS_FIXSIZE)) {
LASSERT(llh->llh_size != 0);
LASSERT(llh->llh_size == reclen);
}
/* return error if osp object is stale */
if (idx != LLOG_HEADER_IDX && dt_object_stale(o))
RETURN(-ESTALE);
rc = dt_attr_get(env, o, &lgi->lgi_attr);
if (rc)
RETURN(rc);
/**
* The modification case.
* If idx set then the record with that index must be modified.
* There are three cases possible:
* 1) the common case is the llog header update (idx == 0)
* 2) the llog record modification during llog process.
* This is indicated by the \a loghandle::lgh_cur_idx > 0.
* In that case the \a loghandle::lgh_cur_offset
* 3) otherwise this is assumed that llog consist of records of
* fixed size, i.e. catalog. The llog header must has llh_size
* field equal to record size. The record offset is calculated
* just by /a idx value
*
* During modification we don't need extra header update because
* the bitmap and record count are not changed. The record header
* and tail remains the same too.
*/
if (idx != LLOG_NEXT_IDX) {
/* llog can be empty only when first record is being written */
LASSERT(ergo(idx > 0, lgi->lgi_attr.la_size > 0));
if (!test_bit_le(idx, LLOG_HDR_BITMAP(llh))) {
CERROR("%s: modify unset record %u\n",
o->do_lu.lo_dev->ld_obd->obd_name, idx);
RETURN(-ENOENT);
}
if (idx != rec->lrh_index) {
CERROR("%s: modify index mismatch %d %u\n",
o->do_lu.lo_dev->ld_obd->obd_name, idx,
rec->lrh_index);
RETURN(-EFAULT);
}
if (idx == LLOG_HEADER_IDX) {
/* llog header update */
__u32 *bitmap = LLOG_HDR_BITMAP(llh);
lgi->lgi_off = 0;
/* If it does not indicate the bitmap index
* (reccookie == NULL), then it means update
* the whole update header. Otherwise only
* update header and bits needs to be updated,
* and in DNE cases, it will signaficantly
* shrink the RPC size.
* see distribute_txn_cancel_records()*/
if (reccookie == NULL) {
lgi->lgi_buf.lb_len = reclen;
lgi->lgi_buf.lb_buf = rec;
rc = dt_record_write(env, o, &lgi->lgi_buf,
&lgi->lgi_off, th);
RETURN(rc);
}
/* update the header */
lgi->lgi_buf.lb_len = llh->llh_bitmap_offset;
lgi->lgi_buf.lb_buf = llh;
rc = dt_record_write(env, o, &lgi->lgi_buf,
&lgi->lgi_off, th);
if (rc != 0)
RETURN(rc);
/* update the bitmap */
index = reccookie->lgc_index;
lgi->lgi_off = llh->llh_bitmap_offset +
(index / (sizeof(*bitmap) * 8)) *
sizeof(*bitmap);
lgi->lgi_buf.lb_len = sizeof(*bitmap);
lgi->lgi_buf.lb_buf =
&bitmap[index/(sizeof(*bitmap)*8)];
rc = dt_record_write(env, o, &lgi->lgi_buf,
&lgi->lgi_off, th);
RETURN(rc);
} else if (llh->llh_flags & LLOG_F_IS_FIXSIZE) {
lgi->lgi_off = llh->llh_hdr.lrh_len +
(idx - 1) * reclen;
} else if (reccookie != NULL && reccookie->lgc_index > 0) {
/**
* The lgc_offset can be used only if index is
* the same.
*/
if (idx != reccookie->lgc_index) {
CERROR("%s: modify index mismatch %d %d\n",
o->do_lu.lo_dev->ld_obd->obd_name, idx,
reccookie->lgc_index);
RETURN(-EFAULT);
}
lgi->lgi_off = reccookie->lgc_offset;
CDEBUG(D_OTHER, "modify record "DFID": idx:%u, "
"len:%u offset %llu\n",
PLOGID(&loghandle->lgh_id), idx,
rec->lrh_len, (long long)lgi->lgi_off);
} else {
/* This can be result of lgh_cur_idx is not set during
* llog processing or llh_size is not set to proper
* record size for fixed records llog. Therefore it is
* impossible to get record offset. */
CERROR("%s: can't get record offset, idx:%d, "
"len:%u.\n", o->do_lu.lo_dev->ld_obd->obd_name,
idx, rec->lrh_len);
RETURN(-EFAULT);
}
/* update only data, header and tail remain the same */
lgi->lgi_off += sizeof(struct llog_rec_hdr);
lgi->lgi_buf.lb_len = REC_DATA_LEN(rec);
lgi->lgi_buf.lb_buf = REC_DATA(rec);
dt_write_lock(env, o, 0);
rc = dt_record_write(env, o, &lgi->lgi_buf, &lgi->lgi_off, th);
if (rc == 0 && reccookie) {
reccookie->lgc_lgl = loghandle->lgh_id;
reccookie->lgc_index = idx;
rc = 1;
}
dt_write_unlock(env, o);
RETURN(rc);
}
/**
* The append case.
* The most common case of using llog. The new index is assigned to
* the new record, new bit is set in llog bitmap and llog count is
* incremented.
*
* Make sure that records don't cross a chunk boundary, so we can
* process them page-at-a-time if needed. If it will cross a chunk
* boundary, write in a fake (but referenced) entry to pad the chunk.
*/
/* simulate ENOSPC when new plain llog is being added to the
* catalog */
if (CFS_FAIL_CHECK(OBD_FAIL_MDS_LLOG_CREATE_FAILED2) &&
llh->llh_flags & LLOG_F_IS_CAT)
RETURN(-ENOSPC);
LASSERT(lgi->lgi_attr.la_valid & LA_SIZE);
orig_last_idx = loghandle->lgh_last_idx;
lgi->lgi_off = lgi->lgi_attr.la_size;
if (loghandle->lgh_max_size > 0 &&
lgi->lgi_off >= loghandle->lgh_max_size) {
CDEBUG(D_OTHER, "llog is getting too large (%u >= %u) at %u "
DFID"\n", (unsigned)lgi->lgi_off,
loghandle->lgh_max_size, (int)loghandle->lgh_last_idx,
PLOGID(&loghandle->lgh_id));
/* this is to signal that this llog is full */
loghandle->lgh_last_idx = llog_max_idx(loghandle);
RETURN(-ENOSPC);
}
left = chunk_size - (lgi->lgi_off & (chunk_size - 1));
/* NOTE: padding is a record, but no bit is set */
if (left != 0 && left != reclen &&
left < (reclen + LLOG_MIN_REC_SIZE)) {
index = loghandle->lgh_last_idx + 1;
rc = llog_osd_pad(env, o, &lgi->lgi_off, left, index, th);
if (rc)
RETURN(rc);
loghandle->lgh_last_idx++; /* for pad rec */
pad = true;
}
/* if it's the last idx in log file, then return -ENOSPC
* or wrap around if a catalog */
if (llog_is_full(loghandle)) {
if (llh->llh_flags & LLOG_F_IS_CAT)
loghandle->lgh_last_idx = 0;
else
RETURN(-ENOSPC);
}
down_write(&loghandle->lgh_last_sem);
/* increment the last_idx along with llh_tail index, they should
* be equal for a llog lifetime */
if (CFS_FAIL_CHECK(OBD_FAIL_LLOG_ADD_GAP) && --cfs_fail_val == 0)
loghandle->lgh_last_idx++;
loghandle->lgh_last_idx++;
index = loghandle->lgh_last_idx;
LLOG_HDR_TAIL(llh)->lrt_index = index;
/**
* NB: the caller should make sure only 1 process access
* the lgh_last_idx, e.g. append should be exclusive.
* Otherwise it might hit the assert.
*/
LASSERT(index < LLOG_HDR_BITMAP_SIZE(llh));
rec->lrh_index = index;
lrt = llog_get_rec_tail(rec);
lrt->lrt_len = rec->lrh_len;
lrt->lrt_index = rec->lrh_index;
/* the lgh_hdr_lock protects llog header data from concurrent
* update/cancel, the llh_count and llh_bitmap are protected */
spin_lock(&loghandle->lgh_hdr_lock);
rc = __test_and_set_bit_le(index, LLOG_HDR_BITMAP(llh));
LASSERTF(!rc,
"%s: index %u already set in llog bitmap "DFID"\n",
o->do_lu.lo_dev->ld_obd->obd_name, index,
PFID(lu_object_fid(&o->do_lu)));
llh->llh_count++;
if (!(llh->llh_flags & LLOG_F_IS_FIXSIZE)) {
/* Update the minimum size of the llog record */
if (llh->llh_size == 0)
llh->llh_size = reclen;
else if (reclen < llh->llh_size)
llh->llh_size = reclen;
}
spin_unlock(&loghandle->lgh_hdr_lock);
/*
* readers (e.g. llog_osd_read_header()) must not find
* llog updated partially (bitmap/counter claims record,
* but a record hasn't been added yet) as this results
* in EIO.
*/
dt_write_lock(env, o, 0);
if (lgi->lgi_attr.la_size == 0) {
lgi->lgi_off = 0;
lgi->lgi_buf.lb_len = llh->llh_hdr.lrh_len;
lgi->lgi_buf.lb_buf = &llh->llh_hdr;
rc = dt_record_write(env, o, &lgi->lgi_buf, &lgi->lgi_off, th);
if (rc != 0)
GOTO(out_unlock, rc);
} else {
__u32 *bitmap = LLOG_HDR_BITMAP(llh);
/* Note: If this is not initialization (size == 0), then do not
* write the whole header (8k bytes), only update header/tail
* and bits needs to be updated. Because this update might be
* part of cross-MDT operation, which needs to write these
* updates into the update log(32KB limit) and also pack inside
* the RPC (1MB limit), if we write 8K for each operation, which
* will cost a lot space, and keep us adding more updates to one
* update log.*/
lgi->lgi_off = 0;
lgi->lgi_buf.lb_len = llh->llh_bitmap_offset;
lgi->lgi_buf.lb_buf = &llh->llh_hdr;
rc = dt_record_write(env, o, &lgi->lgi_buf, &lgi->lgi_off, th);
if (rc != 0)
GOTO(out_unlock, rc);
lgi->lgi_off = llh->llh_bitmap_offset +
(index / (sizeof(*bitmap) * 8)) * sizeof(*bitmap);
lgi->lgi_buf.lb_len = sizeof(*bitmap);
lgi->lgi_buf.lb_buf = &bitmap[index/(sizeof(*bitmap)*8)];
rc = dt_record_write(env, o, &lgi->lgi_buf, &lgi->lgi_off, th);
if (rc != 0)
GOTO(out_unlock, rc);
lgi->lgi_off = (unsigned long)LLOG_HDR_TAIL(llh) -
(unsigned long)llh;
lgi->lgi_buf.lb_len = sizeof(llh->llh_tail);
lgi->lgi_buf.lb_buf = LLOG_HDR_TAIL(llh);
rc = dt_record_write(env, o, &lgi->lgi_buf, &lgi->lgi_off, th);
if (rc != 0)
GOTO(out_unlock, rc);
}
if (CFS_FAIL_PRECHECK(OBD_FAIL_LLOG_PAUSE_AFTER_PAD) && pad) {
/* a window for concurrent llog reader, see LU-12577 */
CFS_FAIL_TIMEOUT(OBD_FAIL_LLOG_PAUSE_AFTER_PAD,
cfs_fail_val ?: 1);
}
out_unlock:
/* unlock here for remote object */
if (rc) {
dt_write_unlock(env, o);
GOTO(out, rc);
}
if (CFS_FAIL_PRECHECK(OBD_FAIL_LLOG_PROCESS_TIMEOUT) &&
cfs_fail_val == (unsigned int)(loghandle->lgh_id.lgl_oi.oi.oi_id &
0xFFFFFFFF)) {
CFS_RACE(OBD_FAIL_LLOG_PROCESS_TIMEOUT);
msleep(1 * MSEC_PER_SEC);
}
/* computed index can be used to determine offset for fixed-size
* records. This also allows to handle Catalog wrap around case */
if (llh->llh_flags & LLOG_F_IS_FIXSIZE) {
lgi->lgi_off = llh->llh_hdr.lrh_len + (index - 1) * reclen;
} else {
rc = dt_attr_get(env, o, &lgi->lgi_attr);
if (rc) {
dt_write_unlock(env, o);
GOTO(out, rc);
}
LASSERT(lgi->lgi_attr.la_valid & LA_SIZE);
lgi->lgi_off = max_t(__u64, lgi->lgi_attr.la_size,
lgi->lgi_off);
}
lgi->lgi_buf.lb_len = reclen;
lgi->lgi_buf.lb_buf = rec;
rc = dt_record_write(env, o, &lgi->lgi_buf, &lgi->lgi_off, th);
dt_write_unlock(env, o);
if (rc < 0)
GOTO(out, rc);
if (loghandle->lgh_max_size > 0 &&
lgi->lgi_off >= loghandle->lgh_max_size) {
CDEBUG(D_OTHER, "llog is getting too large (%u >= %u) at %u "
DFID"\n", (unsigned int)lgi->lgi_off,
loghandle->lgh_max_size, (int)loghandle->lgh_last_idx,
PLOGID(&loghandle->lgh_id));
/* this is to signal that this llog is full */
loghandle->lgh_last_idx = llog_max_idx(loghandle);
}
up_write(&loghandle->lgh_last_sem);
CDEBUG(D_HA, "added record "DFID".%u, %u off%llu\n",
PFID(lu_object_fid(&o->do_lu)), index, rec->lrh_len,
lgi->lgi_off);
if (reccookie != NULL) {
reccookie->lgc_lgl = loghandle->lgh_id;
reccookie->lgc_index = index;
if ((rec->lrh_type == MDS_UNLINK_REC) ||
(rec->lrh_type == MDS_SETATTR64_REC))
reccookie->lgc_subsys = LLOG_MDS_OST_ORIG_CTXT;
else if (rec->lrh_type == OST_SZ_REC)
reccookie->lgc_subsys = LLOG_SIZE_ORIG_CTXT;
else
reccookie->lgc_subsys = -1;
rc = 1;
}
RETURN(rc);
out:
/* cleanup llog for error case */
spin_lock(&loghandle->lgh_hdr_lock);
clear_bit_le(index, LLOG_HDR_BITMAP(llh));
llh->llh_count--;
spin_unlock(&loghandle->lgh_hdr_lock);
/* restore llog last_idx */
if (dt_object_remote(o)) {
loghandle->lgh_last_idx = orig_last_idx;
} else if (--loghandle->lgh_last_idx == 0 &&
(llh->llh_flags & LLOG_F_IS_CAT) && llh->llh_cat_idx != 0) {
/* catalog had just wrap-around case */
loghandle->lgh_last_idx = llog_max_idx(loghandle);
}
LLOG_HDR_TAIL(llh)->lrt_index = loghandle->lgh_last_idx;
up_write(&loghandle->lgh_last_sem);
RETURN(rc);
}
/*
* We can skip reading at least as many log blocks as the number of
* minimum sized log records we are skipping. If it turns out
* that we are not far enough along the log (because the
* actual records are larger than minimum size) we just skip
* some more records.
*
* Note: in llog_process_thread, it will use bitmap offset as
* the index to locate the record, which also includs some pad
* records, whose record size is very small, and it also does not
* consider pad record when recording minimum record size (otherwise
* min_record size might be too small), so in some rare cases,
* it might skip too much record for @goal, see llog_osd_next_block().
*
* When force_mini_rec is true, it means we have to use LLOG_MIN_REC_SIZE
* as the min record size to skip over, usually because in the previous
* try, it skip too much record, see loog_osd_next(prev)_block().
*/
static inline void llog_skip_over(struct llog_handle *lgh, __u64 *off,
int curr, int goal, __u32 chunk_size,
bool force_mini_rec)
{
struct llog_log_hdr *llh = lgh->lgh_hdr;
/* Goal should not bigger than the record count */
if (goal > lgh->lgh_last_idx)
goal = lgh->lgh_last_idx;
if (goal > curr) {
if (llh->llh_flags & LLOG_F_IS_FIXSIZE) {
*off = chunk_size + (goal - 1) * llh->llh_size;
} else {
__u64 min_rec_size = LLOG_MIN_REC_SIZE;
if (llh->llh_size > 0 && !force_mini_rec)
min_rec_size = llh->llh_size;
*off = *off + (goal - curr - 1) * min_rec_size;
}
}
/* always align with lower chunk boundary*/
*off &= ~(chunk_size - 1);
}
/**
* changelog_remap_rec() - Remap record to desired format
* @rec: The record to remap.
* @crf_wanted: Flags describing the desired extensions.
* @cref_want: Flags describing the desired extra extensions.
*
* Remap a record to the desired format as specified by the crf flags.
* The record must be big enough to contain the final remapped version.
* Superfluous extension fields are removed and missing ones are added
* and zeroed. The flags of the record are updated accordingly to what
* the calling llog layer can support. Only influence user land has is
* to store the NID in large NID format. The user land end user will
* recieve all fields that supported by the kernel.
*
* The jobid and rename extensions will be added to a record, to match the
* format an application expects, typically. In this case, the newly added
* fields will be zeroed.
* The Jobid field can be removed, to guarantee compatibility with older
* clients that don't expect this field in the records they process.
*
* The following assumptions are being made:
* - CLF_RENAME will not be removed
* - CLF_JOBID will not be added without CLF_RENAME being added too
* - CLF_EXTRA_FLAGS will not be added without CLF_JOBID being added too
*/
static void changelog_remap_rec(struct changelog_rec *rec,
enum changelog_rec_flags crf_wanted,
enum changelog_rec_extra_flags cref_want)
{
char *xattr_mov = NULL;
char *omd_mov = NULL;
char *nid_mov = NULL;
char *uidgid_mov = NULL;
char *ef_mov;
char *jid_mov;
char *rnm_mov;
enum changelog_rec_extra_flags cref = CLFE_INVALID;
crf_wanted = (enum changelog_rec_flags)
(crf_wanted & CLF_SUPPORTED);
cref_want = (enum changelog_rec_extra_flags)
(cref_want & CLFE_SUPPORTED);
if ((rec->cr_flags & CLF_SUPPORTED) == crf_wanted) {
if (!(rec->cr_flags & CLF_EXTRA_FLAGS) ||
(rec->cr_flags & CLF_EXTRA_FLAGS &&
(changelog_rec_extra_flags(rec)->cr_extra_flags &
CLFE_SUPPORTED) ==
cref_want))
return;
}
/* First move the variable-length name field */
memmove((char *)rec + changelog_rec_offset(crf_wanted, cref_want),
changelog_rec_name(rec), rec->cr_namelen);
/* Locations of extensions in the remapped record */
if (rec->cr_flags & CLF_EXTRA_FLAGS) {
xattr_mov = (char *)rec +
changelog_rec_offset(
(enum changelog_rec_flags)
(crf_wanted & CLF_SUPPORTED),
(enum changelog_rec_extra_flags)
(cref_want & ~CLFE_XATTR));
omd_mov = (char *)rec +
changelog_rec_offset(
(enum changelog_rec_flags)
(crf_wanted & CLF_SUPPORTED),
(enum changelog_rec_extra_flags)
(cref_want & ~(CLFE_OPEN | CLFE_XATTR)));
nid_mov = (char *)rec +
changelog_rec_offset(
(enum changelog_rec_flags)
(crf_wanted & CLF_SUPPORTED),
(enum changelog_rec_extra_flags)
(cref_want &
~(CLFE_NID | CLFE_OPEN | CLFE_XATTR)));
uidgid_mov = (char *)rec +
changelog_rec_offset(
(enum changelog_rec_flags)
(crf_wanted & CLF_SUPPORTED),
(enum changelog_rec_extra_flags)
(cref_want & ~(CLFE_UIDGID |
CLFE_NID |
CLFE_OPEN |
CLFE_XATTR)));
cref = (enum changelog_rec_extra_flags)
changelog_rec_extra_flags(rec)->cr_extra_flags;
}
ef_mov = (char *)rec +
changelog_rec_offset(
(enum changelog_rec_flags)
(crf_wanted & ~CLF_EXTRA_FLAGS), CLFE_INVALID);
jid_mov = (char *)rec +
changelog_rec_offset((enum changelog_rec_flags)(crf_wanted &
~(CLF_EXTRA_FLAGS | CLF_JOBID)),
CLFE_INVALID);
rnm_mov = (char *)rec +
changelog_rec_offset((enum changelog_rec_flags)(crf_wanted &
~(CLF_EXTRA_FLAGS |
CLF_JOBID |
CLF_RENAME)),
CLFE_INVALID);
/* Move the extension fields to the desired positions */
if ((crf_wanted & CLF_EXTRA_FLAGS) &&
(rec->cr_flags & CLF_EXTRA_FLAGS)) {
if ((cref_want & CLFE_XATTR) && (cref & CLFE_XATTR))
memmove(xattr_mov, changelog_rec_xattr(rec),
sizeof(struct changelog_ext_xattr));
if ((cref_want & CLFE_OPEN) && (cref & CLFE_OPEN))
memmove(omd_mov, changelog_rec_openmode(rec),
sizeof(struct changelog_ext_openmode));
if ((cref_want & CLFE_NID) && (cref & CLFE_NID)) {
struct changelog_ext_nid *cen = changelog_rec_nid(rec);
if ((cref_want & CLFE_NID_BE) != (cref & CLFE_NID_BE)) {
struct lnet_nid *nid;
if (!(cref_want & CLFE_NID_BE)) {
nid = (struct lnet_nid *)cen;
if (nid_is_nid4(nid)) {
struct changelog_ext_nid *mov;
mov = (struct changelog_ext_nid *)nid_mov;
mov->cr_nid = lnet_nid_to_nid4(nid);
cref &= ~CLFE_NID_BE;
} else {
cref &= ~(CLFE_NID |
CLFE_NID_BE);
}
} else {
nid = (struct lnet_nid *)nid_mov;
lnet_nid4_to_nid(cen->cr_nid, nid);
}
changelog_rec_extra_flags(rec)->cr_extra_flags =
cref;
} else {
memmove(nid_mov, cen, sizeof(*cen));
}
}
if ((cref_want & CLFE_UIDGID) && (cref & CLFE_UIDGID))
memmove(uidgid_mov, changelog_rec_uidgid(rec),
sizeof(struct changelog_ext_uidgid));
memmove(ef_mov, changelog_rec_extra_flags(rec),
sizeof(struct changelog_ext_extra_flags));
}
if ((crf_wanted & CLF_JOBID) && (rec->cr_flags & CLF_JOBID))
memmove(jid_mov, changelog_rec_jobid(rec),
sizeof(struct changelog_ext_jobid));
if ((crf_wanted & CLF_RENAME) && (rec->cr_flags & CLF_RENAME))
memmove(rnm_mov, changelog_rec_rename(rec),
sizeof(struct changelog_ext_rename));
/* Clear newly added fields */
if (xattr_mov && (cref_want & CLFE_XATTR) &&
!(cref & CLFE_XATTR))
memset(xattr_mov, 0, sizeof(struct changelog_ext_xattr));
if (omd_mov && (cref_want & CLFE_OPEN) &&
!(cref & CLFE_OPEN))
memset(omd_mov, 0, sizeof(struct changelog_ext_openmode));
if (nid_mov && (cref_want & CLFE_NID) &&
!(cref & CLFE_NID))
memset(nid_mov, 0, sizeof(struct changelog_ext_nid));
if (uidgid_mov && (cref_want & CLFE_UIDGID) &&
!(cref & CLFE_UIDGID))
memset(uidgid_mov, 0, sizeof(struct changelog_ext_uidgid));
if ((crf_wanted & CLF_EXTRA_FLAGS) &&
!(rec->cr_flags & CLF_EXTRA_FLAGS))
memset(ef_mov, 0, sizeof(struct changelog_ext_extra_flags));
if ((crf_wanted & CLF_JOBID) && !(rec->cr_flags & CLF_JOBID))
memset(jid_mov, 0, sizeof(struct changelog_ext_jobid));
if ((crf_wanted & CLF_RENAME) && !(rec->cr_flags & CLF_RENAME))
memset(rnm_mov, 0, sizeof(struct changelog_ext_rename));
/* Update the record's flags accordingly */
rec->cr_flags = (rec->cr_flags & CLF_FLAGMASK) | crf_wanted;
if (rec->cr_flags & CLF_EXTRA_FLAGS)
changelog_rec_extra_flags(rec)->cr_extra_flags =
changelog_rec_extra_flags(rec)->cr_extra_flags |
cref_want;
}
/**
* changelog_block_trim_ext() - Remove optional fields client doesn't expect.
* @hdr: Header of the block of records to remap.[in,out]
* @last_hdr: Last header, don't read past this point.[in,out]
* @loghandle: llog handle of the current llog
*
* flags: Flags describing the fields to keep.
* extra_flags: Flags describing the extra fields to keep.
*
* This is typically in order to ensure compatibility with older clients.
* It is assumed that since we exclusively remove fields, the block will be
* big enough to handle the remapped records. It is also assumed that records
* of a block have the same format (i.e.: the same features enabled).
*/
static void changelog_block_trim_ext(struct llog_rec_hdr *hdr,
struct llog_rec_hdr *last_hdr,
struct llog_handle *loghandle)
{
enum changelog_rec_flags flags = CLF_SUPPORTED;
enum changelog_rec_extra_flags extra_flags = CLFE_SUPPORTED;
if (!(loghandle->lgh_hdr->llh_flags & LLOG_F_EXT_X_XATTR))
extra_flags &= ~CLFE_XATTR;
if (!(loghandle->lgh_hdr->llh_flags & LLOG_F_EXT_X_OMODE))
extra_flags &= ~CLFE_OPEN;
if (!(loghandle->lgh_hdr->llh_flags & LLOG_F_EXT_X_NID))
extra_flags &= ~(CLFE_NID | CLFE_NID_BE);
if (!(loghandle->lgh_hdr->llh_flags & LLOG_F_EXT_X_NID_BE)) {
if (extra_flags & CLFE_NID_BE) {
/* The large nid won't be understood */
extra_flags &= ~CLFE_NID_BE;
}
}
if (!(loghandle->lgh_hdr->llh_flags & LLOG_F_EXT_X_UIDGID))
extra_flags &= ~CLFE_UIDGID;
if (!(loghandle->lgh_hdr->llh_flags & LLOG_F_EXT_EXTRA_FLAGS))
flags &= ~CLF_EXTRA_FLAGS;
if (!(loghandle->lgh_hdr->llh_flags & LLOG_F_EXT_JOBID))
flags &= ~CLF_JOBID;
if (flags == CLF_SUPPORTED && extra_flags == CLFE_SUPPORTED)
return;
if (hdr->lrh_type != CHANGELOG_REC)
return;
do {
struct changelog_rec *rec = (struct changelog_rec *)(hdr + 1);
enum changelog_rec_extra_flags xflag = CLFE_INVALID;
if (unlikely(hdr->lrh_len == 0)) {
/* It is corruption case, we cannot know the next rec,
* jump to the last one directly to avoid dead loop. */
LCONSOLE(D_WARNING, "Hit invalid llog record: "
"idx %u, type %u, id %u\n",
hdr->lrh_index, hdr->lrh_type, hdr->lrh_id);
hdr = llog_rec_hdr_next(last_hdr);
if (unlikely(hdr == last_hdr))
LCONSOLE(D_WARNING, "The last record crashed: "
"idx %u, type %u, id %u\n",
hdr->lrh_index, hdr->lrh_type,
hdr->lrh_id);
break;
}
if (flags & CLF_EXTRA_FLAGS &&
rec->cr_flags & CLF_EXTRA_FLAGS) {
xflag = changelog_rec_extra_flags(rec)->cr_extra_flags &
extra_flags;
}
/* Fill up the changelog record with everything the kernel
* version supports.
*/
changelog_remap_rec(rec, rec->cr_flags & flags, xflag);
hdr = llog_rec_hdr_next(hdr);
/* Yield CPU to avoid soft-lockup if there are too many records
* to be handled. */
cond_resched();
} while ((char *)hdr <= (char *)last_hdr);
}
/**
* llog_osd_next_block() - Implementation of the llog_operations::lop_next_block
* @env: execution environment
* @loghandle: llog handle of the current llog
* @cur_idx: index preceeding cur_offset [in,out]
* @next_idx: target index to find
* @cur_offset: furtherst point read in the file [in,out]
* @buf: pointer to data buffer to fill
* @len: required len to read, it is usually llog chunk_size.
*
* This function finds the the next llog block to return which contains
* record with required index. It is main part of llog processing.
*
* Return:
* * %0 on successful buffer read
* * %negative value on error
*/
static int llog_osd_next_block(const struct lu_env *env,
struct llog_handle *loghandle, int *cur_idx,
int next_idx, __u64 *cur_offset, void *buf,
int len)
{
struct llog_thread_info *lgi = llog_info(env);
struct dt_object *o;
struct dt_device *dt;
int rc;
__u32 chunk_size;
__u32 tail_len;
int last_idx = *cur_idx;
__u64 last_offset = *cur_offset;
bool force_mini_rec = !next_idx;
ENTRY;
LASSERT(env);
LASSERT(lgi);
chunk_size = loghandle->lgh_hdr->llh_hdr.lrh_len;
if (len == 0 || len & (chunk_size - 1))
RETURN(-EINVAL);
LASSERT(loghandle);
LASSERT(loghandle->lgh_ctxt);
if (CFS_FAIL_PRECHECK(OBD_FAIL_MDS_CHANGELOG_DEL) &&
cfs_fail_val == ((unsigned long)loghandle & 0xFFFFFFFF)) {
CFS_RACE(OBD_FAIL_MDS_CHANGELOG_DEL);
msleep(MSEC_PER_SEC >> 2);
}
o = loghandle->lgh_obj;
LASSERT(o);
dt_read_lock(env, o, 0);
if (!llog_osd_exist(loghandle))
GOTO(out, rc = -ESTALE); //object was destroyed
dt = lu2dt_dev(o->do_lu.lo_dev);
if (IS_ERR(dt))
GOTO(out, rc = PTR_ERR(dt));
rc = dt_attr_get(env, o, &lgi->lgi_attr);
if (rc)
GOTO(out, rc);
CDEBUG(D_OTHER,
"looking for log index %u (cur idx %u off %llu), size %llu\n",
next_idx, *cur_idx,
*cur_offset, lgi->lgi_attr.la_size);
while (*cur_offset < lgi->lgi_attr.la_size) {
struct llog_rec_hdr *rec, *last_rec;
struct llog_rec_tail *tail;
llog_skip_over(loghandle, cur_offset, *cur_idx,
next_idx, chunk_size, force_mini_rec);
/* read up to next llog chunk_size block */
lgi->lgi_buf.lb_len = chunk_size -
(*cur_offset & (chunk_size - 1));
lgi->lgi_buf.lb_buf = buf;
rc = dt_read(env, o, &lgi->lgi_buf, cur_offset);
if (rc < 0) {
if (rc == -EBADR) {
/* no goal is valid case */
if (!next_idx)
GOTO(out, rc);
if (!force_mini_rec)
goto retry;
}
CERROR("%s: can't read llog block from log "DFID
" offset %llu: rc = %d\n",
o->do_lu.lo_dev->ld_obd->obd_name,
PFID(lu_object_fid(&o->do_lu)), *cur_offset,
rc);
GOTO(out, rc);
}
if (rc < len) {
/* signal the end of the valid buffer to
* llog_process */
memset(buf + rc, 0, len - rc);
}
if (rc == 0) { /* end of file, nothing to do */
if (!force_mini_rec)
goto retry;
GOTO(out, rc);
}
if (rc < sizeof(*tail)) {
if (!force_mini_rec)
goto retry;
CERROR("%s: invalid llog block at log id "DFID" offset %llu\n",
o->do_lu.lo_dev->ld_obd->obd_name,
PLOGID(&loghandle->lgh_id), *cur_offset);
GOTO(out, rc = -EINVAL);
}
rec = buf;
if (LLOG_REC_HDR_NEEDS_SWABBING(rec))
lustre_swab_llog_rec(rec);
/* caller handles bad records if any */
if (llog_verify_record(loghandle, rec))
GOTO(out, rc = 0);
tail = (struct llog_rec_tail *)((char *)buf + rc -
sizeof(struct llog_rec_tail));
while ((tail->lrt_index == 0 || tail->lrt_len == 0) &&
(void *) tail > buf) {
/* looks like zeroes at the end of block */
/* searching real record, assume 4bytes align */
tail = (struct llog_rec_tail *)(((char *)tail) - 4);
};
tail_len = tail->lrt_len;
/* base on tail_len do swab */
if (tail_len > chunk_size) {
__swab32s(&tail_len);
if (tail_len > chunk_size) {
CERROR("%s: invalid llog tail at log id "DFID":%x offset %llu tail idx %u lrt len %u read_size %d\n",
o->do_lu.lo_dev->ld_obd->obd_name,
PFID(&loghandle->lgh_id.lgl_oi.oi_fid),
loghandle->lgh_id.lgl_ogen, *cur_offset,
tail->lrt_index, tail->lrt_len, rc);
/* tail is broken */
GOTO(out, rc = -EINVAL);
}
}
/* get the last record in block */
last_rec = (struct llog_rec_hdr *)((char *)tail - tail_len +
sizeof(struct llog_rec_tail));
if (LLOG_REC_HDR_NEEDS_SWABBING(last_rec))
lustre_swab_llog_rec(last_rec);
if (last_rec->lrh_index != tail->lrt_index) {
CERROR("%s: invalid llog tail at log id "DFID" offset %llu last_rec idx %u tail idx %u lrt len %u read_size %d\n",
o->do_lu.lo_dev->ld_obd->obd_name,
PLOGID(&loghandle->lgh_id), *cur_offset,
last_rec->lrh_index, tail->lrt_index,
tail->lrt_len, rc);
GOTO(out, rc = -EINVAL);
}
*cur_idx = tail->lrt_index;
/* this shouldn't happen */
if (tail->lrt_index == 0) {
CERROR("%s: invalid llog tail at log id "DFID"offset %llu bytes %d\n",
o->do_lu.lo_dev->ld_obd->obd_name,
PLOGID(&loghandle->lgh_id), *cur_offset, rc);
GOTO(out, rc = -EINVAL);
}
if (tail->lrt_index < next_idx) {
last_idx = *cur_idx;
last_offset = *cur_offset;
continue;
}
/* sanity check that the start of the new buffer is no farther
* than the record that we wanted. This shouldn't happen. */
if (next_idx && rec->lrh_index > next_idx) {
if (!force_mini_rec && next_idx > last_idx)
goto retry;
CERROR("%s: missed desired record? %u > %u\n",
o->do_lu.lo_dev->ld_obd->obd_name,
rec->lrh_index, next_idx);
GOTO(out, rc = -ENOENT);
}
/* Trim unsupported extensions for compat w/ older clients */
changelog_block_trim_ext(rec, last_rec, loghandle);
GOTO(out, rc = 0);
retry:
/* Note: because there are some pad records in the
* llog, so llog_skip_over() might skip too much
* records, let's try skip again with minimum record */
force_mini_rec = true;
*cur_offset = last_offset;
*cur_idx = last_idx;
}
/* being here means we reach end of llog but didn't find needed idx
* normally could happen while processing remote llog, return -EBADR
* to indicate access beyond end of file like dt_read() does and to
* distunguish this situation from real IO or network issues.
*/
GOTO(out, rc = -EBADR);
out:
dt_read_unlock(env, o);
return rc;
}
/**
* llog_osd_prev_block() - Implementation of the llog_operations::lop_prev_block
* @env: execution environment
* @loghandle: llog handle of the current llog
* @prev_idx: target index to find
* @buf: pointer to data buffer to fill
* @len: required len to read, it is llog_chunk_size usually.
*
* This function finds the llog block to return which contains
* record with required index but in reverse order - from end of llog
* to the beginning.
* It is main part of reverse llog processing.
*
* Return:
* * %0 on successful buffer read
* * %negative value on error
*/
static int llog_osd_prev_block(const struct lu_env *env,
struct llog_handle *loghandle,
int prev_idx, void *buf, int len)
{
loff_t cur_offset;
int cur_idx;
cur_offset = loghandle->lgh_hdr->llh_hdr.lrh_len;
cur_idx = 1;
return llog_osd_next_block(env, loghandle, &cur_idx, prev_idx,
&cur_offset, buf, len);
}
/**
* llog_osd_dir_get() - Get llog directory object.
* @env: execution environment
* @ctxt: llog context
*
* This is helper function to get llog directory object.
* It is used by named llog operations to find/insert/delete llog entry from
* llog directory.
*
* Return:
* * %dt_object of llog directory
* * %ERR_PTR of negative value on error
*/
static struct dt_object *llog_osd_dir_get(const struct lu_env *env,
struct llog_ctxt *ctxt)
{
struct dt_device *dt;
struct dt_thread_info *dti = dt_info(env);
struct dt_object *dir;
int rc;
dt = ctxt->loc_exp->exp_obd->obd_lvfs_ctxt.dt;
if (ctxt->loc_dir == NULL) {
rc = dt_root_get(env, dt, &dti->dti_fid);
if (rc)
return ERR_PTR(rc);
dir = dt_locate(env, dt, &dti->dti_fid);
if (!IS_ERR(dir) && !dt_try_as_dir(env, dir, false)) {
dt_object_put(env, dir);
return ERR_PTR(-ENOTDIR);
}
} else {
lu_object_get(&ctxt->loc_dir->do_lu);
dir = ctxt->loc_dir;
}
return dir;
}
/**
* llog_osd_open() - Implementation of the llog_operations::lop_open
* @env: execution environment
* @handle: llog handle of the current llog
* @logid: logid of llog to open (nameless llog)
* @name: name of llog to open (named llog)
* @open_param: LLOG_OPEN_NEW - new llog, may not exist
* LLOG_OPEN_EXIST - old llog, must exist
*
* This function opens the llog by its logid or by name, it may open also
* non existent llog and assing then new id to it.
* The llog_open/llog_close pair works similar to lu_object_find/put,
* the object may not exist prior open. The result of open is just dt_object
* in the llog header.
*
* Return:
* * %0 on successful open, llog_handle::lgh_obj contains dt_object of the llog.
* * %negative value on error
*/
static int llog_osd_open(const struct lu_env *env, struct llog_handle *handle,
struct llog_logid *logid, char *name,
enum llog_open_param open_param)
{
struct llog_thread_info *lgi = llog_info(env);
struct llog_ctxt *ctxt = handle->lgh_ctxt;
struct dt_object *o;
struct dt_device *dt;
struct ls_device *ls;
struct local_oid_storage *los = NULL;
int rc = 0;
bool new_id = false;
ENTRY;
LASSERT(env);
LASSERT(ctxt);
LASSERT(ctxt->loc_exp);
LASSERT(ctxt->loc_exp->exp_obd);
dt = ctxt->loc_exp->exp_obd->obd_lvfs_ctxt.dt;
LASSERT(dt);
if (ctxt->loc_flags & LLOG_CTXT_FLAG_NORMAL_FID) {
struct lu_object_conf conf = { 0 };
if (logid != NULL) {
logid_to_fid(logid, &lgi->lgi_fid);
} else {
/* If logid == NULL, then it means the caller needs
* to allocate new FID (llog_cat_declare_add_rec()). */
rc = dt_fid_alloc(env, dt, &lgi->lgi_fid, NULL, NULL);
if (rc < 0)
RETURN(rc);
rc = 0;
conf.loc_flags = LOC_F_NEW;
}
o = dt_locate_at(env, dt, &lgi->lgi_fid,
dt->dd_lu_dev.ld_site->ls_top_dev, &conf);
if (IS_ERR(o))
RETURN(PTR_ERR(o));
goto after_open;
}
ls = ls_device_find_or_init(dt);
if (IS_ERR(ls))
RETURN(PTR_ERR(ls));
mutex_lock(&ls->ls_los_mutex);
los = dt_los_find(ls, name != NULL ? FID_SEQ_LLOG_NAME : FID_SEQ_LLOG);
mutex_unlock(&ls->ls_los_mutex);
LASSERT(los);
ls_device_put(env, ls);
LASSERT(handle);
if (logid != NULL) {
logid_to_fid(logid, &lgi->lgi_fid);
} else if (name) {
struct dt_object *llog_dir;
llog_dir = llog_osd_dir_get(env, ctxt);
if (IS_ERR(llog_dir))
GOTO(out, rc = PTR_ERR(llog_dir));
dt_read_lock(env, llog_dir, 0);
rc = dt_lookup_dir(env, llog_dir, name, &lgi->lgi_fid);
dt_read_unlock(env, llog_dir);
dt_object_put(env, llog_dir);
if (rc == -ENOENT && open_param == LLOG_OPEN_NEW) {
/* generate fid for new llog */
rc = local_object_fid_generate(env, los,
&lgi->lgi_fid);
new_id = true;
}
if (rc < 0)
GOTO(out, rc);
OBD_ALLOC(handle->lgh_name, strlen(name) + 1);
if (handle->lgh_name)
strcpy(handle->lgh_name, name);
else
GOTO(out, rc = -ENOMEM);
} else {
LASSERTF(open_param & LLOG_OPEN_NEW, "%#x\n", open_param);
/* generate fid for new llog */
generate:
rc = local_object_fid_generate(env, los, &lgi->lgi_fid);
if (rc < 0)
GOTO(out, rc);
new_id = true;
}
if (CFS_FAIL_PRECHECK(OBD_FAIL_MDS_LLOG_UMOUNT_RACE) &&
cfs_fail_val == 1) {
cfs_fail_val = 2;
CFS_RACE(OBD_FAIL_MDS_LLOG_UMOUNT_RACE);
msleep(MSEC_PER_SEC);
}
o = ls_locate(env, ls, &lgi->lgi_fid, NULL);
if (IS_ERR(o))
GOTO(out_name, rc = PTR_ERR(o));
if (dt_object_exists(o) && new_id) {
/* llog exists with just generated ID, e.g. some old llog file
* still is in use or is orphan, drop a warn and skip it. */
CDEBUG(D_INFO, "%s: llog exists with the same FID: "DFID
", skipping\n",
o->do_lu.lo_dev->ld_obd->obd_name,
PFID(lu_object_fid(&o->do_lu)));
dt_object_put(env, o);
/* just skip this llog ID, we shouldn't delete it because we
* don't know exactly what is its purpose and state. */
goto generate;
}
after_open:
/* No new llog is expected but doesn't exist */
if (open_param != LLOG_OPEN_NEW && !dt_object_exists(o)) {
CDEBUG(D_INFO, "%s: llog FID: "DFID" obj %p doesn`t exist\n",
o->do_lu.lo_dev->ld_obd->obd_name,
PFID(lu_object_fid(&o->do_lu)), o);
GOTO(out_put, rc = -ENOENT);
}
fid_to_logid(&lgi->lgi_fid, &handle->lgh_id);
handle->lgh_obj = o;
handle->private_data = los;
LASSERT(handle->lgh_ctxt);
RETURN(rc);
out_put:
if (ctxt->loc_flags & LLOG_CTXT_FLAG_NORMAL_FID)
/* according to llog_osd_close() */
dt_object_put_nocache(env, o);
else
dt_object_put(env, o);
out_name:
OBD_FREE(handle->lgh_name, strlen(name) + 1);
out:
if (los != NULL)
dt_los_put(los);
RETURN(rc);
}
/**
* llog_osd_get_regular_fid_dir() - Get dir for regular fid log object
* @env: execution environment
* @dto: llog object
*
* Get directory for regular fid log object, and these regular fid log
* object will be inserted under this directory, to satisfy the FS
* consistency check, e2fsck etc.
*
* Return:
* * %pointer to the directory if it is found.
* * %ERR_PTR(negative errno) if it fails.
*/
static struct dt_object *llog_osd_get_regular_fid_dir(const struct lu_env *env,
struct dt_object *dto)
{
struct llog_thread_info *lgi = llog_info(env);
struct seq_server_site *ss = dto->do_lu.lo_dev->ld_site->ld_seq_site;
struct lu_seq_range *range = &lgi->lgi_range;
struct lu_fid *dir_fid = &lgi->lgi_fid;
struct dt_object *dir;
struct dt_device *dt;
int rc;
ENTRY;
fld_range_set_any(range);
LASSERT(ss != NULL);
rc = ss->ss_server_fld->lsf_seq_lookup(env, ss->ss_server_fld,
fid_seq(lu_object_fid(&dto->do_lu)), range);
if (rc < 0)
RETURN(ERR_PTR(rc));
lu_update_log_dir_fid(dir_fid, range->lsr_index);
dt = lu2dt_dev(dto->do_lu.lo_dev);
if (IS_ERR(dt))
RETURN((struct dt_object *)dt);
dir = dt_locate(env, dt, dir_fid);
if (IS_ERR(dir))
RETURN(dir);
if (!dt_try_as_dir(env, dir, false)) {
dt_object_put(env, dir);
RETURN(ERR_PTR(-ENOTDIR));
}
RETURN(dir);
}
/**
* llog_osd_regular_fid_add_name_entry() - Add llog object with regular FID to
* name entry
* @env: execution environment
* @dto: llog object
* @th: current transaction handle
* @declare: if it is declare or execution
*
* Add llog object with regular FID to name space, and each llog
* object on each MDT will be /update_log_dir/[seq:oid:ver],
* so to satisfy the namespace consistency check, e2fsck etc.
*
* Return:
* * %0 if insertion succeeds.
* * %negative errno if insertion fails.
*/
static int
llog_osd_regular_fid_add_name_entry(const struct lu_env *env,
struct dt_object *dto,
struct thandle *th, bool declare)
{
struct llog_thread_info *lgi = llog_info(env);
const struct lu_fid *fid = lu_object_fid(&dto->do_lu);
struct dt_insert_rec *rec = &lgi->lgi_dt_rec;
struct dt_object *dir;
char *name = lgi->lgi_name;
int rc;
ENTRY;
if (!fid_is_norm(fid))
RETURN(0);
dir = llog_osd_get_regular_fid_dir(env, dto);
if (IS_ERR(dir))
RETURN(PTR_ERR(dir));
rec->rec_fid = fid;
rec->rec_type = S_IFREG;
snprintf(name, sizeof(lgi->lgi_name), DFID, PFID(fid));
dt_write_lock(env, dir, 0);
if (declare) {
rc = dt_declare_insert(env, dir, (struct dt_rec *)rec,
(struct dt_key *)name, th);
} else {
rc = dt_insert(env, dir, (struct dt_rec *)rec,
(struct dt_key *)name, th);
}
dt_write_unlock(env, dir);
dt_object_put(env, dir);
RETURN(rc);
}
/**
* llog_osd_declare_create() - Implementation of the lop_declare_create
* @env: execution environment
* @res: llog handle of the current llog
* @th: current transaction handle
*
* Implementation of the llog_operations::lop_declare_create
* This function declares the llog create. It declares also name insert
* into llog directory in case of named llog.
*
* Return:
* * %0 on successful create declaration
* * %negative value on error
*/
static int llog_osd_declare_create(const struct lu_env *env,
struct llog_handle *res, struct thandle *th)
{
struct llog_thread_info *lgi = llog_info(env);
struct dt_insert_rec *rec = &lgi->lgi_dt_rec;
struct local_oid_storage *los;
struct dt_object *o;
int rc;
ENTRY;
LASSERT(res->lgh_obj);
LASSERT(th);
/* object can be created by another thread */
o = res->lgh_obj;
if (dt_object_exists(o))
RETURN(0);
if (res->lgh_ctxt->loc_flags & LLOG_CTXT_FLAG_NORMAL_FID) {
struct llog_thread_info *lgi = llog_info(env);
lgi->lgi_attr.la_valid = LA_MODE | LA_SIZE;
lgi->lgi_attr.la_size = 0;
lgi->lgi_attr.la_mode = S_IFREG | S_IRUGO | S_IWUSR;
lgi->lgi_dof.dof_type = dt_mode_to_dft(S_IFREG);
rc = dt_declare_create(env, o, &lgi->lgi_attr, NULL,
&lgi->lgi_dof, th);
if (rc < 0)
RETURN(rc);
rc = llog_osd_regular_fid_add_name_entry(env, o, th, true);
RETURN(rc);
}
los = res->private_data;
LASSERT(los);
rc = llog_osd_declare_new_object(env, los, o, th);
if (rc)
RETURN(rc);
/* do not declare header initialization here as it's declared
* in llog_osd_declare_write_rec() which is always called */
if (res->lgh_name) {
struct dt_object *llog_dir;
llog_dir = llog_osd_dir_get(env, res->lgh_ctxt);
if (IS_ERR(llog_dir))
RETURN(PTR_ERR(llog_dir));
logid_to_fid(&res->lgh_id, &lgi->lgi_fid);
rec->rec_fid = &lgi->lgi_fid;
rec->rec_type = S_IFREG;
rc = dt_declare_insert(env, llog_dir,
(struct dt_rec *)rec,
(struct dt_key *)res->lgh_name, th);
dt_object_put(env, llog_dir);
if (rc)
CERROR("%s: can't declare named llog %s: rc = %d\n",
o->do_lu.lo_dev->ld_obd->obd_name,
res->lgh_name, rc);
}
RETURN(rc);
}
/**
* llog_osd_create() - Implementation of the llog_operations::lop_create
* @env: execution environment
* @res: llog handle of the current llog
* @th: current transaction handle
*
* This function creates the llog according with llog_handle::lgh_obj
* and llog_handle::lgh_name.
*
* Return:
* * %0 on successful create
* * %negative value on error
*/
static int llog_osd_create(const struct lu_env *env, struct llog_handle *res,
struct thandle *th)
{
struct llog_thread_info *lgi = llog_info(env);
struct dt_insert_rec *rec = &lgi->lgi_dt_rec;
struct local_oid_storage *los;
struct dt_object *o;
int rc = 0;
ENTRY;
LASSERT(env);
o = res->lgh_obj;
LASSERT(o);
/* llog can be already created */
if (dt_object_exists(o))
RETURN(-EEXIST);
if (res->lgh_ctxt->loc_flags & LLOG_CTXT_FLAG_NORMAL_FID) {
struct llog_thread_info *lgi = llog_info(env);
lgi->lgi_attr.la_valid = LA_MODE | LA_SIZE | LA_TYPE |
LA_CTIME | LA_MTIME | LA_ATIME | LA_UID | LA_GID;
lgi->lgi_attr.la_size = 0;
lgi->lgi_attr.la_uid = 0;
lgi->lgi_attr.la_gid = 0;
lgi->lgi_attr.la_mode = S_IFREG | S_IRUGO | S_IWUSR;
lgi->lgi_attr.la_ctime = lgi->lgi_attr.la_mtime =
lgi->lgi_attr.la_atime = ktime_get_real_seconds();
lgi->lgi_dof.dof_type = dt_mode_to_dft(S_IFREG);
dt_write_lock(env, o, 0);
rc = dt_create(env, o, &lgi->lgi_attr, NULL,
&lgi->lgi_dof, th);
dt_write_unlock(env, o);
if (rc < 0)
RETURN(rc);
rc = llog_osd_regular_fid_add_name_entry(env, o, th, false);
RETURN(rc);
}
los = res->private_data;
LASSERT(los);
dt_write_lock(env, o, 0);
if (!dt_object_exists(o))
rc = llog_osd_create_new_object(env, los, o, th);
else
rc = -EEXIST;
dt_write_unlock(env, o);
if (rc)
RETURN(rc);
if (res->lgh_name) {
struct dt_object *llog_dir;
llog_dir = llog_osd_dir_get(env, res->lgh_ctxt);
if (IS_ERR(llog_dir))
RETURN(PTR_ERR(llog_dir));
logid_to_fid(&res->lgh_id, &lgi->lgi_fid);
rec->rec_fid = &lgi->lgi_fid;
rec->rec_type = S_IFREG;
dt_read_lock(env, llog_dir, 0);
rc = dt_insert(env, llog_dir, (struct dt_rec *)rec,
(struct dt_key *)res->lgh_name, th);
dt_read_unlock(env, llog_dir);
dt_object_put(env, llog_dir);
if (rc)
CERROR("%s: can't create named llog %s: rc = %d\n",
o->do_lu.lo_dev->ld_obd->obd_name,
res->lgh_name, rc);
}
RETURN(rc);
}
/**
* llog_osd_close() - Implementation of the llog_operations::lop_close
* @env: execution environment
* @handle: llog handle of the current llog
*
* This function closes the llog. It just put llog object and referenced
* local storage.
*
* Return:
* * %0 on successful llog close
* * %negative value on error
*/
static int llog_osd_close(const struct lu_env *env, struct llog_handle *handle)
{
struct local_oid_storage *los;
int rc = 0;
ENTRY;
LASSERT(handle->lgh_obj);
if (handle->lgh_ctxt->loc_flags & LLOG_CTXT_FLAG_NORMAL_FID) {
/* Remove the object from the cache, otherwise it may
* hold LOD being released during cleanup process */
dt_object_put_nocache(env, handle->lgh_obj);
LASSERT(handle->private_data == NULL);
RETURN(rc);
} else {
dt_object_put(env, handle->lgh_obj);
}
los = handle->private_data;
LASSERT(los);
dt_los_put(los);
OBD_FREE(handle->lgh_name, strlen(handle->lgh_name) + 1);
RETURN(rc);
}
/**
* llog_osd_regular_fid_del_name_entry() - delete llog object name entry
* @env: execution environment
* @dto: llog object
* @th: current transaction handle
* @declare: if it is declare or execution
*
* Delete llog object (with regular FID) from name space (under
* update_log_dir).
*
* Return:
* * %0 if deletion succeeds.
* * %negative errno if deletion fails.
*/
static int
llog_osd_regular_fid_del_name_entry(const struct lu_env *env,
struct dt_object *dto,
struct thandle *th, bool declare)
{
struct llog_thread_info *lgi = llog_info(env);
const struct lu_fid *fid = lu_object_fid(&dto->do_lu);
struct dt_object *dir;
char *name = lgi->lgi_name;
int rc;
ENTRY;
if (!fid_is_norm(fid))
RETURN(0);
dir = llog_osd_get_regular_fid_dir(env, dto);
if (IS_ERR(dir))
RETURN(PTR_ERR(dir));
snprintf(name, sizeof(lgi->lgi_name), DFID, PFID(fid));
dt_write_lock(env, dir, 0);
if (declare) {
rc = dt_declare_delete(env, dir, (struct dt_key *)name,
th);
} else {
rc = dt_delete(env, dir, (struct dt_key *)name, th);
}
dt_write_unlock(env, dir);
dt_object_put(env, dir);
RETURN(rc);
}
/**
* llog_osd_declare_destroy() - Implementation of the lop_declare_destroy
* @env: execution environment
* @loghandle: llog handle of the current llog
* @th: current transaction handle
*
* Implementation of the llog_operations::lop_declare_destroy
* This function declare destroys the llog and deletes also entry in the
* llog directory in case of named llog. Llog should be opened prior that.
*
* Return:
* * %0 on successful destroy
* * %negative value on error
*/
static int llog_osd_declare_destroy(const struct lu_env *env,
struct llog_handle *loghandle,
struct thandle *th)
{
struct llog_ctxt *ctxt;
struct dt_object *o, *llog_dir = NULL;
int rc;
ENTRY;
ctxt = loghandle->lgh_ctxt;
LASSERT(ctxt);
o = loghandle->lgh_obj;
LASSERT(o);
if (loghandle->lgh_name) {
llog_dir = llog_osd_dir_get(env, ctxt);
if (IS_ERR(llog_dir))
RETURN(PTR_ERR(llog_dir));
rc = dt_declare_delete(env, llog_dir,
(struct dt_key *)loghandle->lgh_name,
th);
if (rc < 0)
GOTO(out_put, rc);
}
rc = dt_declare_ref_del(env, o, th);
if (rc < 0)
GOTO(out_put, rc);
rc = dt_declare_destroy(env, o, th);
if (rc < 0)
GOTO(out_put, rc);
if (loghandle->lgh_ctxt->loc_flags & LLOG_CTXT_FLAG_NORMAL_FID) {
rc = llog_osd_regular_fid_del_name_entry(env, o, th, true);
if (rc < 0)
GOTO(out_put, rc);
}
out_put:
if (!(IS_ERR_OR_NULL(llog_dir)))
dt_object_put(env, llog_dir);
RETURN(rc);
}
/**
* llog_osd_destroy() - Implementation of the llog_operations::lop_destroy
* @env: execution environment
* @loghandle: llog handle of the current llog
* @th: current transaction handle
*
* This function destroys the llog and deletes also entry in the
* llog directory in case of named llog. Llog should be opened prior that.
* Destroy method is not part of external transaction and does everything
* inside.
*
* Return:
* * %0 on successful destroy
* * %negative value on error
*/
static int llog_osd_destroy(const struct lu_env *env,
struct llog_handle *loghandle, struct thandle *th)
{
struct llog_ctxt *ctxt;
struct dt_object *o, *llog_dir = NULL;
int rc;
ENTRY;
ctxt = loghandle->lgh_ctxt;
LASSERT(ctxt != NULL);
o = loghandle->lgh_obj;
LASSERT(o != NULL);
dt_write_lock(env, o, DT_TGT_CHILD);
if (!llog_osd_exist(loghandle))
GOTO(out_unlock, rc = 0);
if (loghandle->lgh_name) {
llog_dir = llog_osd_dir_get(env, ctxt);
if (IS_ERR(llog_dir))
GOTO(out_unlock, rc = PTR_ERR(llog_dir));
dt_read_lock(env, llog_dir, DT_TGT_PARENT);
rc = dt_delete(env, llog_dir,
(struct dt_key *)loghandle->lgh_name,
th);
dt_read_unlock(env, llog_dir);
if (rc) {
CERROR("%s: can't remove llog %s: rc = %d\n",
o->do_lu.lo_dev->ld_obd->obd_name,
loghandle->lgh_name, rc);
GOTO(out_unlock, rc);
}
}
dt_ref_del(env, o, th);
rc = dt_destroy(env, o, th);
if (rc < 0)
GOTO(out_unlock, rc);
loghandle->lgh_destroyed = true;
if (loghandle->lgh_ctxt->loc_flags & LLOG_CTXT_FLAG_NORMAL_FID) {
rc = llog_osd_regular_fid_del_name_entry(env, o, th, false);
if (rc < 0)
GOTO(out_unlock, rc);
}
out_unlock:
dt_write_unlock(env, o);
if (!(IS_ERR_OR_NULL(llog_dir)))
dt_object_put(env, llog_dir);
RETURN(rc);
}
/**
* llog_osd_setup() - Implementation of the llog_operations::lop_setup
* @env: execution environment
* @obd: obd device the llog belongs to
* @olg: the llog group, it is always zero group now.
* @ctxt_idx: the llog index, it defines the purpose of this llog.
* Every new llog type have to use own index.
* @disk_obd: the storage obd, where llog is stored.
*
* This function setup the llog on local storage.
*
* Return:
* * %0 on successful llog setup
* * %negative value on error
*/
static int llog_osd_setup(const struct lu_env *env, struct obd_device *obd,
struct obd_llog_group *olg, int ctxt_idx,
struct obd_device *disk_obd)
{
struct llog_thread_info *lgi = llog_info(env);
struct llog_ctxt *ctxt;
int rc = 0;
ENTRY;
LASSERT(obd);
LASSERT(olg->olg_ctxts[ctxt_idx]);
ctxt = llog_ctxt_get(olg->olg_ctxts[ctxt_idx]);
LASSERT(ctxt);
if (disk_obd == NULL)
GOTO(out, rc = 0);
/* initialize data allowing to generate new fids,
* literally we need a sequece */
lgi->lgi_fid.f_seq = FID_SEQ_LLOG;
lgi->lgi_fid.f_oid = 1;
lgi->lgi_fid.f_ver = 0;
rc = local_oid_storage_init(env, disk_obd->obd_lvfs_ctxt.dt,
&lgi->lgi_fid,
&ctxt->loc_los_nameless);
if (rc != 0)
GOTO(out, rc);
lgi->lgi_fid.f_seq = FID_SEQ_LLOG_NAME;
lgi->lgi_fid.f_oid = 1;
lgi->lgi_fid.f_ver = 0;
rc = local_oid_storage_init(env, disk_obd->obd_lvfs_ctxt.dt,
&lgi->lgi_fid,
&ctxt->loc_los_named);
if (rc != 0) {
local_oid_storage_fini(env, ctxt->loc_los_nameless);
ctxt->loc_los_nameless = NULL;
}
GOTO(out, rc);
out:
llog_ctxt_put(ctxt);
return rc;
}
/**
* llog_osd_cleanup() - Implementation of the llog_operations::lop_cleanup
* @env: execution environment
* @ctxt: the llog context
*
* This function cleanups the llog on local storage.
*
* Return 0 always
*/
static int llog_osd_cleanup(const struct lu_env *env, struct llog_ctxt *ctxt)
{
if (ctxt->loc_los_nameless != NULL) {
local_oid_storage_fini(env, ctxt->loc_los_nameless);
ctxt->loc_los_nameless = NULL;
}
if (ctxt->loc_los_named != NULL) {
local_oid_storage_fini(env, ctxt->loc_los_named);
ctxt->loc_los_named = NULL;
}
return 0;
}
const struct llog_operations llog_osd_ops = {
.lop_next_block = llog_osd_next_block,
.lop_prev_block = llog_osd_prev_block,
.lop_read_header = llog_osd_read_header,
.lop_declare_destroy = llog_osd_declare_destroy,
.lop_destroy = llog_osd_destroy,
.lop_setup = llog_osd_setup,
.lop_cleanup = llog_osd_cleanup,
.lop_open = llog_osd_open,
.lop_exist = llog_osd_exist,
.lop_declare_create = llog_osd_declare_create,
.lop_create = llog_osd_create,
.lop_declare_write_rec = llog_osd_declare_write_rec,
.lop_write_rec = llog_osd_write_rec,
.lop_close = llog_osd_close,
};
EXPORT_SYMBOL(llog_osd_ops);
const struct llog_operations llog_common_cat_ops = {
.lop_next_block = llog_osd_next_block,
.lop_prev_block = llog_osd_prev_block,
.lop_read_header = llog_osd_read_header,
.lop_declare_destroy = llog_osd_declare_destroy,
.lop_destroy = llog_osd_destroy,
.lop_setup = llog_osd_setup,
.lop_cleanup = llog_osd_cleanup,
.lop_open = llog_osd_open,
.lop_exist = llog_osd_exist,
.lop_declare_create = llog_osd_declare_create,
.lop_create = llog_osd_create,
.lop_declare_write_rec = llog_osd_declare_write_rec,
.lop_write_rec = llog_osd_write_rec,
.lop_close = llog_osd_close,
.lop_add = llog_cat_add_rec,
.lop_declare_add = llog_cat_declare_add_rec,
};
EXPORT_SYMBOL(llog_common_cat_ops);
/**
* llog_osd_get_cat_list() - Read the special file which contains the list of
* llog catalogs IDs
* @env: execution environment
* @d: corresponding storage device
* @idx: position to start from, usually OST/MDT index
* @count: how many catalog IDs to read
* @idarray: the buffer for the data. If it is NULL then function returns just
* number of catalog IDs in the file. [out]
* @fid: LLOG_CATALOGS_OID for CATALOG object
*
* This function reads the CATALOGS file which contains the array of llog
* catalogs IDs. The main purpose of this file is to store OSP llogs indexed
* by OST/MDT number.
*
* Return:
* * %0 on successful read of catalog IDs
* * %negative value on error
* * %positive value which is number of records in the file if @idarray is NULL
*/
int llog_osd_get_cat_list(const struct lu_env *env, struct dt_device *d,
int idx, int count, struct llog_catid *idarray,
const struct lu_fid *fid)
{
struct llog_thread_info *lgi = llog_info(env);
struct dt_object *o = NULL;
struct thandle *th;
int rc, size;
ENTRY;
LASSERT(d);
size = sizeof(*idarray) * count;
lgi->lgi_off = idx * sizeof(*idarray);
lgi->lgi_fid = *fid;
o = dt_locate(env, d, &lgi->lgi_fid);
if (IS_ERR(o))
RETURN(PTR_ERR(o));
if (!dt_object_exists(o)) {
th = dt_trans_create(env, d);
if (IS_ERR(th))
GOTO(out, rc = PTR_ERR(th));
lgi->lgi_attr.la_valid = LA_MODE | LA_TYPE;
lgi->lgi_attr.la_mode = S_IFREG | S_IRUGO | S_IWUSR;
lgi->lgi_dof.dof_type = dt_mode_to_dft(S_IFREG);
th->th_wait_submit = 1;
/* Make the llog object creation synchronization, so
* it will be reliable to the reference, especially
* for remote reference */
th->th_sync = 1;
rc = dt_declare_create(env, o, &lgi->lgi_attr, NULL,
&lgi->lgi_dof, th);
if (rc)
GOTO(out_trans, rc);
rc = dt_trans_start_local(env, d, th);
if (rc)
GOTO(out_trans, rc);
dt_write_lock(env, o, 0);
if (!dt_object_exists(o))
rc = dt_create(env, o, &lgi->lgi_attr, NULL,
&lgi->lgi_dof, th);
dt_write_unlock(env, o);
out_trans:
dt_trans_stop(env, d, th);
if (rc)
GOTO(out, rc);
}
rc = dt_attr_get(env, o, &lgi->lgi_attr);
if (rc)
GOTO(out, rc);
if (!S_ISREG(lgi->lgi_attr.la_mode)) {
CERROR("%s: CATALOGS is not a regular file!: mode = %o\n",
o->do_lu.lo_dev->ld_obd->obd_name,
lgi->lgi_attr.la_mode);
GOTO(out, rc = -ENOENT);
}
CDEBUG(D_CONFIG, "cat list: disk size=%d, read=%d\n",
(int)lgi->lgi_attr.la_size, size);
/* return just number of llogs */
if (idarray == NULL) {
rc = lgi->lgi_attr.la_size / sizeof(*idarray);
GOTO(out, rc);
}
/* read for new ost index or for empty file */
memset(idarray, 0, size);
if (lgi->lgi_attr.la_size <= lgi->lgi_off)
GOTO(out, rc = 0);
if (lgi->lgi_attr.la_size < lgi->lgi_off + size)
size = lgi->lgi_attr.la_size - lgi->lgi_off;
lgi->lgi_buf.lb_buf = idarray;
lgi->lgi_buf.lb_len = size;
rc = dt_record_read(env, o, &lgi->lgi_buf, &lgi->lgi_off);
/* -EFAULT means the llog is a sparse file. This is not an error
* after arbitrary OST index is supported. */
if (rc < 0 && rc != -EFAULT) {
CERROR("%s: error reading CATALOGS: rc = %d\n",
o->do_lu.lo_dev->ld_obd->obd_name, rc);
GOTO(out, rc);
}
EXIT;
out:
dt_object_put(env, o);
RETURN(rc);
}
EXPORT_SYMBOL(llog_osd_get_cat_list);
/**
* llog_osd_put_cat_list() - Write the special file which contains the list of
* llog catalogs IDs
* @env: execution environment
* @d: corresponding storage device
* @idx: position to start from, usually OST/MDT index
* @count: how many catalog IDs to write
* @idarray: the buffer with the data to write.[out]
* @fid: LLOG_CATALOGS_OID for CATALOG object
*
* This function writes the CATALOG file which contains the array of llog
* catalogs IDs. It is used mostly to store OSP llogs indexed by OST/MDT
* number.
*
* Return:
* * %0 on successful write of catalog IDs
* * %negative value on error
*/
int llog_osd_put_cat_list(const struct lu_env *env, struct dt_device *d,
int idx, int count, struct llog_catid *idarray,
const struct lu_fid *fid)
{
struct llog_thread_info *lgi = llog_info(env);
struct dt_object *o = NULL;
struct thandle *th;
int rc, size;
if (count == 0)
RETURN(0);
LASSERT(d);
size = sizeof(*idarray) * count;
lgi->lgi_off = idx * sizeof(*idarray);
lgi->lgi_fid = *fid;
o = dt_locate(env, d, &lgi->lgi_fid);
if (IS_ERR(o))
RETURN(PTR_ERR(o));
if (!dt_object_exists(o))
GOTO(out, rc = -ENOENT);
rc = dt_attr_get(env, o, &lgi->lgi_attr);
if (rc)
GOTO(out, rc);
if (!S_ISREG(lgi->lgi_attr.la_mode)) {
CERROR("%s: CATALOGS is not a regular file!: mode = %o\n",
o->do_lu.lo_dev->ld_obd->obd_name,
lgi->lgi_attr.la_mode);
GOTO(out, rc = -ENOENT);
}
th = dt_trans_create(env, d);
if (IS_ERR(th))
GOTO(out, rc = PTR_ERR(th));
lgi->lgi_buf.lb_len = size;
lgi->lgi_buf.lb_buf = idarray;
rc = dt_declare_record_write(env, o, &lgi->lgi_buf, lgi->lgi_off, th);
if (rc)
GOTO(out_trans, rc);
/* For update log, this happens during initialization,
* see lod_sub_prep_llog(), and we need make sure catlog
* file ID is written to catlist file(committed) before
* cross-MDT operation write update records to catlog FILE,
* otherwise, during failover these update records might
* missing
*/
if (fid_is_update_log(fid))
th->th_sync = 1;
rc = dt_trans_start_local(env, d, th);
if (rc)
GOTO(out_trans, rc);
th->th_wait_submit = 1;
rc = dt_record_write(env, o, &lgi->lgi_buf, &lgi->lgi_off, th);
if (rc)
CDEBUG(D_INODE, "can't write CATALOGS at index %d: rc = %d\n",
idx, rc);
out_trans:
dt_trans_stop(env, d, th);
out:
dt_object_put(env, o);
RETURN(rc);
}
EXPORT_SYMBOL(llog_osd_put_cat_list);
