4 * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
6 * This program is free software; you can redistribute it and/or modify
7 * it under the terms of the GNU General Public License version 2 only,
8 * as published by the Free Software Foundation.
10 * This program is distributed in the hope that it will be useful, but
11 * WITHOUT ANY WARRANTY; without even the implied warranty of
12 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
13 * General Public License version 2 for more details (a copy is included
14 * in the LICENSE file that accompanied this code).
16 * You should have received a copy of the GNU General Public License
17 * version 2 along with this program; If not, see
18 * http://www.gnu.org/licenses/gpl-2.0.html
23 * Copyright (c) 2003, 2010, Oracle and/or its affiliates. All rights reserved.
24 * Use is subject to license terms.
26 * Copyright (c) 2012, 2016, Intel Corporation.
29 * This file is part of Lustre, http://www.lustre.org/
30 * Lustre is a trademark of Sun Microsystems, Inc.
32 * lustre/obdclass/llog.c
34 * OST<->MDS recovery logging infrastructure.
35 * Invariants in implementation:
36 * - we do not share logs among different OST<->MDS connections, so that
37 * if an OST or MDS fails it need only look at log(s) relevant to itself
39 * Author: Andreas Dilger <adilger@clusterfs.com>
40 * Author: Alex Zhuravlev <bzzz@whamcloud.com>
41 * Author: Mikhail Pershin <tappro@whamcloud.com>
44 #define DEBUG_SUBSYSTEM S_LOG
46 #include <linux/pid_namespace.h>
47 #include <linux/kthread.h>
48 #include <llog_swab.h>
49 #include <lustre_log.h>
50 #include <obd_class.h>
51 #include "llog_internal.h"
53 * Allocate a new log or catalog handle
54 * Used inside llog_open().
56 static struct llog_handle *llog_alloc_handle(void)
58 struct llog_handle *loghandle;
60 OBD_ALLOC_PTR(loghandle);
61 if (loghandle == NULL)
64 init_rwsem(&loghandle->lgh_lock);
65 mutex_init(&loghandle->lgh_hdr_mutex);
66 INIT_LIST_HEAD(&loghandle->u.phd.phd_entry);
67 atomic_set(&loghandle->lgh_refcount, 1);
73 * Free llog handle and header data if exists. Used in llog_close() only
75 static void llog_free_handle(struct llog_handle *loghandle)
77 LASSERT(loghandle != NULL);
79 /* failed llog_init_handle */
80 if (loghandle->lgh_hdr == NULL)
83 if (loghandle->lgh_hdr->llh_flags & LLOG_F_IS_PLAIN)
84 LASSERT(list_empty(&loghandle->u.phd.phd_entry));
85 else if (loghandle->lgh_hdr->llh_flags & LLOG_F_IS_CAT)
86 LASSERT(list_empty(&loghandle->u.chd.chd_head));
87 OBD_FREE_LARGE(loghandle->lgh_hdr, loghandle->lgh_hdr_size);
89 OBD_FREE_PTR(loghandle);
92 void llog_handle_get(struct llog_handle *loghandle)
94 atomic_inc(&loghandle->lgh_refcount);
97 void llog_handle_put(struct llog_handle *loghandle)
99 LASSERT(atomic_read(&loghandle->lgh_refcount) > 0);
100 if (atomic_dec_and_test(&loghandle->lgh_refcount))
101 llog_free_handle(loghandle);
104 static int llog_declare_destroy(const struct lu_env *env,
105 struct llog_handle *handle,
108 struct llog_operations *lop;
113 rc = llog_handle2ops(handle, &lop);
116 if (lop->lop_declare_destroy == NULL)
119 rc = lop->lop_declare_destroy(env, handle, th);
124 int llog_trans_destroy(const struct lu_env *env, struct llog_handle *handle,
127 struct llog_operations *lop;
131 rc = llog_handle2ops(handle, &lop);
134 if (lop->lop_destroy == NULL)
137 LASSERT(handle->lgh_obj != NULL);
138 if (!dt_object_exists(handle->lgh_obj))
141 rc = lop->lop_destroy(env, handle, th);
146 int llog_destroy(const struct lu_env *env, struct llog_handle *handle)
148 struct llog_operations *lop;
149 struct dt_device *dt;
155 rc = llog_handle2ops(handle, &lop);
158 if (lop->lop_destroy == NULL)
161 if (handle->lgh_obj == NULL) {
162 /* if lgh_obj == NULL, then it is from client side destroy */
163 rc = lop->lop_destroy(env, handle, NULL);
167 if (!dt_object_exists(handle->lgh_obj))
170 dt = lu2dt_dev(handle->lgh_obj->do_lu.lo_dev);
172 th = dt_trans_create(env, dt);
176 rc = llog_declare_destroy(env, handle, th);
180 rc = dt_trans_start_local(env, dt, th);
184 rc = lop->lop_destroy(env, handle, th);
187 dt_trans_stop(env, dt, th);
191 EXPORT_SYMBOL(llog_destroy);
193 /* returns negative on error; 0 if success; 1 if success & log destroyed */
194 int llog_cancel_rec(const struct lu_env *env, struct llog_handle *loghandle,
197 struct llog_thread_info *lgi = llog_info(env);
198 struct dt_device *dt;
199 struct llog_log_hdr *llh = loghandle->lgh_hdr;
203 bool subtract_count = false;
207 CDEBUG(D_RPCTRACE, "Canceling %d in log "DOSTID"\n", index,
208 POSTID(&loghandle->lgh_id.lgl_oi));
211 CERROR("Can't cancel index 0 which is header\n");
215 LASSERT(loghandle != NULL);
216 LASSERT(loghandle->lgh_ctxt != NULL);
217 LASSERT(loghandle->lgh_obj != NULL);
219 dt = lu2dt_dev(loghandle->lgh_obj->do_lu.lo_dev);
221 th = dt_trans_create(env, dt);
225 rc = llog_declare_write_rec(env, loghandle, &llh->llh_hdr, index, th);
229 if ((llh->llh_flags & LLOG_F_ZAP_WHEN_EMPTY)) {
230 rc = llog_declare_destroy(env, loghandle, th);
235 th->th_wait_submit = 1;
236 rc = dt_trans_start_local(env, dt, th);
240 down_write(&loghandle->lgh_lock);
242 mutex_lock(&loghandle->lgh_hdr_mutex);
243 if (!ext2_clear_bit(index, LLOG_HDR_BITMAP(llh))) {
244 CDEBUG(D_RPCTRACE, "Catalog index %u already clear?\n", index);
245 GOTO(out_unlock, rc);
248 loghandle->lgh_hdr->llh_count--;
249 subtract_count = true;
250 /* Pass this index to llog_osd_write_rec(), which will use the index
251 * to only update the necesary bitmap. */
252 lgi->lgi_cookie.lgc_index = index;
254 rc = llog_write_rec(env, loghandle, &llh->llh_hdr, &lgi->lgi_cookie,
255 LLOG_HEADER_IDX, th);
257 GOTO(out_unlock, rc);
259 if ((llh->llh_flags & LLOG_F_ZAP_WHEN_EMPTY) &&
260 (llh->llh_count == 1) &&
261 ((loghandle->lgh_last_idx == LLOG_HDR_BITMAP_SIZE(llh) - 1) ||
262 (loghandle->u.phd.phd_cat_handle != NULL &&
263 loghandle->u.phd.phd_cat_handle->u.chd.chd_current_log !=
265 /* never try to destroy it again */
266 llh->llh_flags &= ~LLOG_F_ZAP_WHEN_EMPTY;
267 rc = llog_trans_destroy(env, loghandle, th);
269 /* Sigh, can not destroy the final plain llog, but
270 * the bitmap has been clearly, so the record can not
271 * be accessed anymore, let's return 0 for now, and
272 * the orphan will be handled by LFSCK. */
273 CERROR("%s: can't destroy empty llog #"DOSTID
275 loghandle->lgh_ctxt->loc_obd->obd_name,
276 POSTID(&loghandle->lgh_id.lgl_oi),
277 loghandle->lgh_id.lgl_ogen, rc);
278 GOTO(out_unlock, rc);
284 mutex_unlock(&loghandle->lgh_hdr_mutex);
285 up_write(&loghandle->lgh_lock);
287 rc1 = dt_trans_stop(env, dt, th);
290 if (rc < 0 && subtract_count) {
291 mutex_lock(&loghandle->lgh_hdr_mutex);
292 loghandle->lgh_hdr->llh_count++;
293 ext2_set_bit(index, LLOG_HDR_BITMAP(llh));
294 mutex_unlock(&loghandle->lgh_hdr_mutex);
299 int llog_read_header(const struct lu_env *env, struct llog_handle *handle,
300 const struct obd_uuid *uuid)
302 struct llog_operations *lop;
306 rc = llog_handle2ops(handle, &lop);
310 if (lop->lop_read_header == NULL)
313 rc = lop->lop_read_header(env, handle);
314 if (rc == LLOG_EEMPTY) {
315 struct llog_log_hdr *llh = handle->lgh_hdr;
317 /* lrh_len should be initialized in llog_init_handle */
318 handle->lgh_last_idx = 0; /* header is record with index 0 */
319 llh->llh_count = 1; /* for the header record */
320 llh->llh_hdr.lrh_type = LLOG_HDR_MAGIC;
321 LASSERT(handle->lgh_ctxt->loc_chunk_size >=
322 LLOG_MIN_CHUNK_SIZE);
323 llh->llh_hdr.lrh_len = handle->lgh_ctxt->loc_chunk_size;
324 llh->llh_hdr.lrh_index = 0;
325 llh->llh_timestamp = cfs_time_current_sec();
327 memcpy(&llh->llh_tgtuuid, uuid,
328 sizeof(llh->llh_tgtuuid));
329 llh->llh_bitmap_offset = offsetof(typeof(*llh), llh_bitmap);
330 /* Since update llog header might also call this function,
331 * let's reset the bitmap to 0 here */
332 memset(LLOG_HDR_BITMAP(llh), 0, llh->llh_hdr.lrh_len -
333 llh->llh_bitmap_offset -
334 sizeof(llh->llh_tail));
335 ext2_set_bit(0, LLOG_HDR_BITMAP(llh));
336 LLOG_HDR_TAIL(llh)->lrt_len = llh->llh_hdr.lrh_len;
337 LLOG_HDR_TAIL(llh)->lrt_index = llh->llh_hdr.lrh_index;
342 EXPORT_SYMBOL(llog_read_header);
344 int llog_init_handle(const struct lu_env *env, struct llog_handle *handle,
345 int flags, struct obd_uuid *uuid)
347 struct llog_log_hdr *llh;
348 enum llog_flag fmt = flags & LLOG_F_EXT_MASK;
350 int chunk_size = handle->lgh_ctxt->loc_chunk_size;
353 LASSERT(handle->lgh_hdr == NULL);
355 LASSERT(chunk_size >= LLOG_MIN_CHUNK_SIZE);
356 OBD_ALLOC_LARGE(llh, chunk_size);
360 handle->lgh_hdr = llh;
361 handle->lgh_hdr_size = chunk_size;
362 /* first assign flags to use llog_client_ops */
363 llh->llh_flags = flags;
364 rc = llog_read_header(env, handle, uuid);
366 if (unlikely((llh->llh_flags & LLOG_F_IS_PLAIN &&
367 flags & LLOG_F_IS_CAT) ||
368 (llh->llh_flags & LLOG_F_IS_CAT &&
369 flags & LLOG_F_IS_PLAIN))) {
370 CERROR("%s: llog type is %s but initializing %s\n",
371 handle->lgh_ctxt->loc_obd->obd_name,
372 llh->llh_flags & LLOG_F_IS_CAT ?
374 flags & LLOG_F_IS_CAT ? "catalog" : "plain");
375 GOTO(out, rc = -EINVAL);
376 } else if (llh->llh_flags &
377 (LLOG_F_IS_PLAIN | LLOG_F_IS_CAT)) {
379 * it is possible to open llog without specifying llog
380 * type so it is taken from llh_flags
382 flags = llh->llh_flags;
384 /* for some reason the llh_flags has no type set */
385 CERROR("llog type is not specified!\n");
386 GOTO(out, rc = -EINVAL);
389 !obd_uuid_equals(uuid, &llh->llh_tgtuuid))) {
390 CERROR("%s: llog uuid mismatch: %s/%s\n",
391 handle->lgh_ctxt->loc_obd->obd_name,
393 (char *)llh->llh_tgtuuid.uuid);
394 GOTO(out, rc = -EEXIST);
397 if (flags & LLOG_F_IS_CAT) {
398 LASSERT(list_empty(&handle->u.chd.chd_head));
399 INIT_LIST_HEAD(&handle->u.chd.chd_head);
400 llh->llh_size = sizeof(struct llog_logid_rec);
401 llh->llh_flags |= LLOG_F_IS_FIXSIZE;
402 } else if (!(flags & LLOG_F_IS_PLAIN)) {
403 CERROR("%s: unknown flags: %#x (expected %#x or %#x)\n",
404 handle->lgh_ctxt->loc_obd->obd_name,
405 flags, LLOG_F_IS_CAT, LLOG_F_IS_PLAIN);
408 llh->llh_flags |= fmt;
411 OBD_FREE_LARGE(llh, chunk_size);
412 handle->lgh_hdr = NULL;
416 EXPORT_SYMBOL(llog_init_handle);
418 static int llog_process_thread(void *arg)
420 struct llog_process_info *lpi = arg;
421 struct llog_handle *loghandle = lpi->lpi_loghandle;
422 struct llog_log_hdr *llh = loghandle->lgh_hdr;
423 struct llog_process_cat_data *cd = lpi->lpi_catdata;
426 __u64 cur_offset, tmp_offset;
427 int rc = 0, index = 1, last_index;
429 int last_called_index = 0;
436 cur_offset = chunk_size = llh->llh_hdr.lrh_len;
437 /* expect chunk_size to be power of two */
438 LASSERT(is_power_of_2(chunk_size));
440 OBD_ALLOC_LARGE(buf, chunk_size);
442 lpi->lpi_rc = -ENOMEM;
447 last_called_index = cd->lpcd_first_idx;
448 index = cd->lpcd_first_idx + 1;
450 if (cd != NULL && cd->lpcd_last_idx)
451 last_index = cd->lpcd_last_idx;
453 last_index = LLOG_HDR_BITMAP_SIZE(llh) - 1;
456 struct llog_rec_hdr *rec;
458 unsigned int buf_offset = 0;
461 /* skip records not set in bitmap */
462 while (index <= last_index &&
463 !ext2_test_bit(index, LLOG_HDR_BITMAP(llh)))
466 /* There are no indices prior the last_index */
467 if (index > last_index)
470 CDEBUG(D_OTHER, "index: %d last_index %d\n", index,
474 /* get the buf with our target record; avoid old garbage */
475 memset(buf, 0, chunk_size);
476 rc = llog_next_block(lpi->lpi_env, loghandle, &saved_index,
477 index, &cur_offset, buf, chunk_size);
481 /* NB: after llog_next_block() call the cur_offset is the
482 * offset of the next block after read one.
483 * The absolute offset of the current chunk is calculated
484 * from cur_offset value and stored in chunk_offset variable.
486 tmp_offset = cur_offset;
487 if (do_div(tmp_offset, chunk_size) != 0) {
488 partial_chunk = true;
489 chunk_offset = cur_offset & ~(chunk_size - 1);
491 partial_chunk = false;
492 chunk_offset = cur_offset - chunk_size;
495 /* NB: when rec->lrh_len is accessed it is already swabbed
496 * since it is used at the "end" of the loop and the rec
497 * swabbing is done at the beginning of the loop. */
498 for (rec = (struct llog_rec_hdr *)(buf + buf_offset);
499 (char *)rec < buf + chunk_size;
500 rec = llog_rec_hdr_next(rec)) {
502 CDEBUG(D_OTHER, "processing rec 0x%p type %#x\n",
505 if (LLOG_REC_HDR_NEEDS_SWABBING(rec))
506 lustre_swab_llog_rec(rec);
508 CDEBUG(D_OTHER, "after swabbing, type=%#x idx=%d\n",
509 rec->lrh_type, rec->lrh_index);
511 /* for partial chunk the end of it is zeroed, check
512 * for index 0 to distinguish it. */
513 if (partial_chunk && rec->lrh_index == 0) {
514 /* concurrent llog_add() might add new records
515 * while llog_processing, check this is not
516 * the case and re-read the current chunk
519 if (index > loghandle->lgh_last_idx)
521 /* <2 records means no more records
522 * if the last record we processed was
523 * the final one, then the underlying
524 * object might have been destroyed yet.
525 * we better don't access that.. */
526 mutex_lock(&loghandle->lgh_hdr_mutex);
527 records = loghandle->lgh_hdr->llh_count;
528 mutex_unlock(&loghandle->lgh_hdr_mutex);
531 CDEBUG(D_OTHER, "Re-read last llog buffer for "
532 "new records, index %u, last %u\n",
533 index, loghandle->lgh_last_idx);
534 /* save offset inside buffer for the re-read */
535 buf_offset = (char *)rec - (char *)buf;
536 cur_offset = chunk_offset;
540 if (rec->lrh_len == 0 || rec->lrh_len > chunk_size) {
541 CWARN("invalid length %d in llog record for "
542 "index %d/%d\n", rec->lrh_len,
543 rec->lrh_index, index);
544 GOTO(out, rc = -EINVAL);
547 if (rec->lrh_index < index) {
548 CDEBUG(D_OTHER, "skipping lrh_index %d\n",
553 if (rec->lrh_index != index) {
554 CERROR("%s: Invalid record: index %u but "
556 loghandle->lgh_ctxt->loc_obd->obd_name,
557 rec->lrh_index, index);
558 GOTO(out, rc = -ERANGE);
562 "lrh_index: %d lrh_len: %d (%d remains)\n",
563 rec->lrh_index, rec->lrh_len,
564 (int)(buf + chunk_size - (char *)rec));
566 loghandle->lgh_cur_idx = rec->lrh_index;
567 loghandle->lgh_cur_offset = (char *)rec - (char *)buf +
570 /* if set, process the callback on this record */
571 if (ext2_test_bit(index, LLOG_HDR_BITMAP(llh))) {
572 rc = lpi->lpi_cb(lpi->lpi_env, loghandle, rec,
574 last_called_index = index;
575 if (rc == LLOG_PROC_BREAK) {
577 } else if (rc == LLOG_DEL_RECORD) {
578 rc = llog_cancel_rec(lpi->lpi_env,
585 /* exit if the last index is reached */
586 if (index >= last_index)
594 cd->lpcd_last_idx = last_called_index;
596 if (unlikely(rc == -EIO && loghandle->lgh_obj != NULL)) {
597 if (dt_object_remote(loghandle->lgh_obj)) {
598 /* If it is remote object, then -EIO might means
599 * disconnection or eviction, let's return -EAGAIN,
600 * so for update recovery log processing, it will
601 * retry until the umount or abort recovery, see
602 * lod_sub_recovery_thread() */
603 CERROR("%s retry remote llog process\n",
604 loghandle->lgh_ctxt->loc_obd->obd_name);
607 /* something bad happened to the processing of a local
608 * llog file, probably I/O error or the log got
609 * corrupted to be able to finally release the log we
610 * discard any remaining bits in the header */
611 CERROR("Local llog found corrupted\n");
612 while (index <= last_index) {
613 if (ext2_test_bit(index,
614 LLOG_HDR_BITMAP(llh)) != 0)
615 llog_cancel_rec(lpi->lpi_env, loghandle,
623 OBD_FREE_LARGE(buf, chunk_size);
628 static int llog_process_thread_daemonize(void *arg)
630 struct llog_process_info *lpi = arg;
633 struct nsproxy *new_ns, *curr_ns = current->nsproxy;
635 task_lock(lpi->lpi_reftask);
636 new_ns = lpi->lpi_reftask->nsproxy;
637 if (curr_ns != new_ns) {
640 current->nsproxy = new_ns;
641 /* XXX: we should call put_nsproxy() instead of
642 * atomic_dec(&ns->count) directly. But put_nsproxy() cannot be
643 * used outside of the kernel itself, because it calls
644 * free_nsproxy() which is not exported by the kernel
645 * (defined in kernel/nsproxy.c) */
646 atomic_dec(&curr_ns->count);
648 task_unlock(lpi->lpi_reftask);
652 /* client env has no keys, tags is just 0 */
653 rc = lu_env_init(&env, LCT_LOCAL | LCT_MG_THREAD);
658 rc = llog_process_thread(arg);
662 complete(&lpi->lpi_completion);
666 int llog_process_or_fork(const struct lu_env *env,
667 struct llog_handle *loghandle,
668 llog_cb_t cb, void *data, void *catdata, bool fork)
670 struct llog_process_info *lpi;
677 CERROR("cannot alloc pointer\n");
680 lpi->lpi_loghandle = loghandle;
682 lpi->lpi_cbdata = data;
683 lpi->lpi_catdata = catdata;
686 struct task_struct *task;
688 /* The new thread can't use parent env,
689 * init the new one in llog_process_thread_daemonize. */
691 init_completion(&lpi->lpi_completion);
692 /* take reference to current, so that
693 * llog_process_thread_daemonize() can use it to switch to
694 * namespace associated with current */
695 lpi->lpi_reftask = current;
696 task = kthread_run(llog_process_thread_daemonize, lpi,
697 "llog_process_thread");
700 CERROR("%s: cannot start thread: rc = %d\n",
701 loghandle->lgh_ctxt->loc_obd->obd_name, rc);
704 wait_for_completion(&lpi->lpi_completion);
707 llog_process_thread(lpi);
715 EXPORT_SYMBOL(llog_process_or_fork);
717 int llog_process(const struct lu_env *env, struct llog_handle *loghandle,
718 llog_cb_t cb, void *data, void *catdata)
721 rc = llog_process_or_fork(env, loghandle, cb, data, catdata, true);
722 return rc == LLOG_DEL_PLAIN ? 0 : rc;
724 EXPORT_SYMBOL(llog_process);
726 int llog_reverse_process(const struct lu_env *env,
727 struct llog_handle *loghandle, llog_cb_t cb,
728 void *data, void *catdata)
730 struct llog_log_hdr *llh = loghandle->lgh_hdr;
731 struct llog_process_cat_data *cd = catdata;
733 int rc = 0, first_index = 1, index, idx;
734 __u32 chunk_size = llh->llh_hdr.lrh_len;
737 OBD_ALLOC_LARGE(buf, chunk_size);
742 first_index = cd->lpcd_first_idx + 1;
743 if (cd != NULL && cd->lpcd_last_idx)
744 index = cd->lpcd_last_idx;
746 index = LLOG_HDR_BITMAP_SIZE(llh) - 1;
749 struct llog_rec_hdr *rec;
750 struct llog_rec_tail *tail;
752 /* skip records not set in bitmap */
753 while (index >= first_index &&
754 !ext2_test_bit(index, LLOG_HDR_BITMAP(llh)))
757 LASSERT(index >= first_index - 1);
758 if (index == first_index - 1)
761 /* get the buf with our target record; avoid old garbage */
762 memset(buf, 0, chunk_size);
763 rc = llog_prev_block(env, loghandle, index, buf, chunk_size);
768 idx = rec->lrh_index;
769 CDEBUG(D_RPCTRACE, "index %u : idx %u\n", index, idx);
770 while (idx < index) {
771 rec = (void *)rec + rec->lrh_len;
772 if (LLOG_REC_HDR_NEEDS_SWABBING(rec))
773 lustre_swab_llog_rec(rec);
776 LASSERT(idx == index);
777 tail = (void *)rec + rec->lrh_len - sizeof(*tail);
779 /* process records in buffer, starting where we found one */
780 while ((void *)tail > buf) {
781 if (tail->lrt_index == 0)
782 GOTO(out, rc = 0); /* no more records */
784 /* if set, process the callback on this record */
785 if (ext2_test_bit(index, LLOG_HDR_BITMAP(llh))) {
786 rec = (void *)tail - tail->lrt_len +
789 rc = cb(env, loghandle, rec, data);
790 if (rc == LLOG_PROC_BREAK) {
792 } else if (rc == LLOG_DEL_RECORD) {
793 rc = llog_cancel_rec(env, loghandle,
800 /* previous record, still in buffer? */
802 if (index < first_index)
804 tail = (void *)tail - tail->lrt_len;
810 OBD_FREE_LARGE(buf, chunk_size);
813 EXPORT_SYMBOL(llog_reverse_process);
819 * llog_open - open llog, may not exist
820 * llog_exist - check if llog exists
821 * llog_close - close opened llog, pair for open, frees llog_handle
822 * llog_declare_create - declare llog creation
823 * llog_create - create new llog on disk, need transaction handle
824 * llog_declare_write_rec - declaration of llog write
825 * llog_write_rec - write llog record on disk, need transaction handle
826 * llog_declare_add - declare llog catalog record addition
827 * llog_add - add llog record in catalog, need transaction handle
829 int llog_exist(struct llog_handle *loghandle)
831 struct llog_operations *lop;
836 rc = llog_handle2ops(loghandle, &lop);
839 if (lop->lop_exist == NULL)
842 rc = lop->lop_exist(loghandle);
845 EXPORT_SYMBOL(llog_exist);
847 int llog_declare_create(const struct lu_env *env,
848 struct llog_handle *loghandle, struct thandle *th)
850 struct llog_operations *lop;
855 rc = llog_handle2ops(loghandle, &lop);
858 if (lop->lop_declare_create == NULL)
861 raised = cfs_cap_raised(CFS_CAP_SYS_RESOURCE);
863 cfs_cap_raise(CFS_CAP_SYS_RESOURCE);
864 rc = lop->lop_declare_create(env, loghandle, th);
866 cfs_cap_lower(CFS_CAP_SYS_RESOURCE);
870 int llog_create(const struct lu_env *env, struct llog_handle *handle,
873 struct llog_operations *lop;
878 rc = llog_handle2ops(handle, &lop);
881 if (lop->lop_create == NULL)
884 raised = cfs_cap_raised(CFS_CAP_SYS_RESOURCE);
886 cfs_cap_raise(CFS_CAP_SYS_RESOURCE);
887 rc = lop->lop_create(env, handle, th);
889 cfs_cap_lower(CFS_CAP_SYS_RESOURCE);
893 int llog_declare_write_rec(const struct lu_env *env,
894 struct llog_handle *handle,
895 struct llog_rec_hdr *rec, int idx,
898 struct llog_operations *lop;
903 rc = llog_handle2ops(handle, &lop);
907 if (lop->lop_declare_write_rec == NULL)
910 raised = cfs_cap_raised(CFS_CAP_SYS_RESOURCE);
912 cfs_cap_raise(CFS_CAP_SYS_RESOURCE);
913 rc = lop->lop_declare_write_rec(env, handle, rec, idx, th);
915 cfs_cap_lower(CFS_CAP_SYS_RESOURCE);
919 int llog_write_rec(const struct lu_env *env, struct llog_handle *handle,
920 struct llog_rec_hdr *rec, struct llog_cookie *logcookies,
921 int idx, struct thandle *th)
923 struct llog_operations *lop;
924 int raised, rc, buflen;
928 /* API sanity checks */
929 if (handle == NULL) {
930 CERROR("loghandle is missed\n");
932 } else if (handle->lgh_obj == NULL) {
933 CERROR("loghandle %p with NULL object\n",
936 } else if (th == NULL) {
937 CERROR("%s: missed transaction handle\n",
938 handle->lgh_obj->do_lu.lo_dev->ld_obd->obd_name);
940 } else if (handle->lgh_hdr == NULL) {
941 CERROR("%s: loghandle %p with no header\n",
942 handle->lgh_obj->do_lu.lo_dev->ld_obd->obd_name,
947 rc = llog_handle2ops(handle, &lop);
951 if (lop->lop_write_rec == NULL)
954 buflen = rec->lrh_len;
955 LASSERT(cfs_size_round(buflen) == buflen);
957 raised = cfs_cap_raised(CFS_CAP_SYS_RESOURCE);
959 cfs_cap_raise(CFS_CAP_SYS_RESOURCE);
960 rc = lop->lop_write_rec(env, handle, rec, logcookies, idx, th);
962 cfs_cap_lower(CFS_CAP_SYS_RESOURCE);
966 int llog_add(const struct lu_env *env, struct llog_handle *lgh,
967 struct llog_rec_hdr *rec, struct llog_cookie *logcookies,
974 if (lgh->lgh_logops->lop_add == NULL)
977 raised = cfs_cap_raised(CFS_CAP_SYS_RESOURCE);
979 cfs_cap_raise(CFS_CAP_SYS_RESOURCE);
980 rc = lgh->lgh_logops->lop_add(env, lgh, rec, logcookies, th);
982 cfs_cap_lower(CFS_CAP_SYS_RESOURCE);
985 EXPORT_SYMBOL(llog_add);
987 int llog_declare_add(const struct lu_env *env, struct llog_handle *lgh,
988 struct llog_rec_hdr *rec, struct thandle *th)
994 if (lgh->lgh_logops->lop_declare_add == NULL)
997 raised = cfs_cap_raised(CFS_CAP_SYS_RESOURCE);
999 cfs_cap_raise(CFS_CAP_SYS_RESOURCE);
1000 rc = lgh->lgh_logops->lop_declare_add(env, lgh, rec, th);
1002 cfs_cap_lower(CFS_CAP_SYS_RESOURCE);
1005 EXPORT_SYMBOL(llog_declare_add);
1008 * Helper function to open llog or create it if doesn't exist.
1009 * It hides all transaction handling from caller.
1011 int llog_open_create(const struct lu_env *env, struct llog_ctxt *ctxt,
1012 struct llog_handle **res, struct llog_logid *logid,
1015 struct dt_device *d;
1021 rc = llog_open(env, ctxt, res, logid, name, LLOG_OPEN_NEW);
1025 if (llog_exist(*res))
1028 LASSERT((*res)->lgh_obj != NULL);
1030 d = lu2dt_dev((*res)->lgh_obj->do_lu.lo_dev);
1032 th = dt_trans_create(env, d);
1034 GOTO(out, rc = PTR_ERR(th));
1036 /* Create update llog object synchronously, which
1037 * happens during inialization process see
1038 * lod_sub_prep_llog(), to make sure the update
1039 * llog object is created before corss-MDT writing
1040 * updates into the llog object */
1041 if (ctxt->loc_flags & LLOG_CTXT_FLAG_NORMAL_FID)
1044 th->th_wait_submit = 1;
1045 rc = llog_declare_create(env, *res, th);
1047 rc = dt_trans_start_local(env, d, th);
1049 rc = llog_create(env, *res, th);
1051 dt_trans_stop(env, d, th);
1054 llog_close(env, *res);
1057 EXPORT_SYMBOL(llog_open_create);
1060 * Helper function to delete existent llog.
1062 int llog_erase(const struct lu_env *env, struct llog_ctxt *ctxt,
1063 struct llog_logid *logid, char *name)
1065 struct llog_handle *handle;
1070 /* nothing to erase */
1071 if (name == NULL && logid == NULL)
1074 rc = llog_open(env, ctxt, &handle, logid, name, LLOG_OPEN_EXISTS);
1078 rc = llog_init_handle(env, handle, LLOG_F_IS_PLAIN, NULL);
1080 rc = llog_destroy(env, handle);
1082 rc2 = llog_close(env, handle);
1087 EXPORT_SYMBOL(llog_erase);
1090 * Helper function for write record in llog.
1091 * It hides all transaction handling from caller.
1092 * Valid only with local llog.
1094 int llog_write(const struct lu_env *env, struct llog_handle *loghandle,
1095 struct llog_rec_hdr *rec, int idx)
1097 struct dt_device *dt;
1104 LASSERT(loghandle->lgh_ctxt);
1105 LASSERT(loghandle->lgh_obj != NULL);
1107 dt = lu2dt_dev(loghandle->lgh_obj->do_lu.lo_dev);
1109 th = dt_trans_create(env, dt);
1111 RETURN(PTR_ERR(th));
1113 rc = llog_declare_write_rec(env, loghandle, rec, idx, th);
1115 GOTO(out_trans, rc);
1117 th->th_wait_submit = 1;
1118 rc = dt_trans_start_local(env, dt, th);
1120 GOTO(out_trans, rc);
1122 down_write(&loghandle->lgh_lock);
1123 rc = llog_write_rec(env, loghandle, rec, NULL, idx, th);
1124 up_write(&loghandle->lgh_lock);
1126 dt_trans_stop(env, dt, th);
1129 EXPORT_SYMBOL(llog_write);
1131 int llog_open(const struct lu_env *env, struct llog_ctxt *ctxt,
1132 struct llog_handle **lgh, struct llog_logid *logid,
1133 char *name, enum llog_open_param open_param)
1141 LASSERT(ctxt->loc_logops);
1143 if (ctxt->loc_logops->lop_open == NULL) {
1145 RETURN(-EOPNOTSUPP);
1148 *lgh = llog_alloc_handle();
1151 (*lgh)->lgh_ctxt = ctxt;
1152 (*lgh)->lgh_logops = ctxt->loc_logops;
1154 raised = cfs_cap_raised(CFS_CAP_SYS_RESOURCE);
1156 cfs_cap_raise(CFS_CAP_SYS_RESOURCE);
1157 rc = ctxt->loc_logops->lop_open(env, *lgh, logid, name, open_param);
1159 cfs_cap_lower(CFS_CAP_SYS_RESOURCE);
1161 llog_free_handle(*lgh);
1166 EXPORT_SYMBOL(llog_open);
1168 int llog_close(const struct lu_env *env, struct llog_handle *loghandle)
1170 struct llog_operations *lop;
1175 rc = llog_handle2ops(loghandle, &lop);
1178 if (lop->lop_close == NULL)
1179 GOTO(out, rc = -EOPNOTSUPP);
1180 rc = lop->lop_close(env, loghandle);
1182 llog_handle_put(loghandle);
1185 EXPORT_SYMBOL(llog_close);
1188 * Helper function to get the llog size in records. It is used by MGS
1189 * mostly to check that config llog exists and contains data.
1191 * \param[in] env execution environment
1192 * \param[in] ctxt llog context
1193 * \param[in] name llog name
1195 * \retval true if there are records in llog besides a header
1196 * \retval false on error or llog without records
1198 int llog_is_empty(const struct lu_env *env, struct llog_ctxt *ctxt,
1201 struct llog_handle *llh;
1204 rc = llog_open(env, ctxt, &llh, NULL, name, LLOG_OPEN_EXISTS);
1206 if (likely(rc == -ENOENT))
1211 rc = llog_init_handle(env, llh, LLOG_F_IS_PLAIN, NULL);
1213 GOTO(out_close, rc);
1214 rc = llog_get_size(llh);
1217 llog_close(env, llh);
1219 /* The header is record 1, the llog is still considered as empty
1220 * if there is only header */
1223 EXPORT_SYMBOL(llog_is_empty);
1225 int llog_copy_handler(const struct lu_env *env, struct llog_handle *llh,
1226 struct llog_rec_hdr *rec, void *data)
1228 struct llog_handle *copy_llh = data;
1230 /* Append all records */
1231 return llog_write(env, copy_llh, rec, LLOG_NEXT_IDX);
1234 /* backup plain llog */
1235 int llog_backup(const struct lu_env *env, struct obd_device *obd,
1236 struct llog_ctxt *ctxt, struct llog_ctxt *bctxt,
1237 char *name, char *backup)
1239 struct llog_handle *llh, *bllh;
1244 /* open original log */
1245 rc = llog_open(env, ctxt, &llh, NULL, name, LLOG_OPEN_EXISTS);
1247 /* the -ENOENT case is also reported to the caller
1248 * but silently so it should handle that if needed.
1251 CERROR("%s: failed to open log %s: rc = %d\n",
1252 obd->obd_name, name, rc);
1256 rc = llog_init_handle(env, llh, LLOG_F_IS_PLAIN, NULL);
1258 GOTO(out_close, rc);
1260 /* Make sure there's no old backup log */
1261 rc = llog_erase(env, bctxt, NULL, backup);
1262 if (rc < 0 && rc != -ENOENT)
1263 GOTO(out_close, rc);
1265 /* open backup log */
1266 rc = llog_open_create(env, bctxt, &bllh, NULL, backup);
1268 CERROR("%s: failed to open backup logfile %s: rc = %d\n",
1269 obd->obd_name, backup, rc);
1270 GOTO(out_close, rc);
1273 /* check that backup llog is not the same object as original one */
1274 if (llh->lgh_obj == bllh->lgh_obj) {
1275 CERROR("%s: backup llog %s to itself (%s), objects %p/%p\n",
1276 obd->obd_name, name, backup, llh->lgh_obj,
1278 GOTO(out_backup, rc = -EEXIST);
1281 rc = llog_init_handle(env, bllh, LLOG_F_IS_PLAIN, NULL);
1283 GOTO(out_backup, rc);
1285 /* Copy log record by record */
1286 rc = llog_process_or_fork(env, llh, llog_copy_handler, (void *)bllh,
1289 CERROR("%s: failed to backup log %s: rc = %d\n",
1290 obd->obd_name, name, rc);
1292 llog_close(env, bllh);
1294 llog_close(env, llh);
1297 EXPORT_SYMBOL(llog_backup);
1299 /* Get size of llog */
1300 __u64 llog_size(const struct lu_env *env, struct llog_handle *llh)
1305 rc = llh->lgh_obj->do_ops->do_attr_get(env, llh->lgh_obj, &la);
1307 CERROR("%s: attr_get failed, rc = %d\n",
1308 llh->lgh_ctxt->loc_obd->obd_name, rc);
1314 EXPORT_SYMBOL(llog_size);