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.sun.com/software/products/lustre/docs/GPLv2.pdf
20 * Please contact Sun Microsystems, Inc., 4150 Network Circle, Santa Clara,
21 * CA 95054 USA or visit www.sun.com if you need additional information or
27 * Copyright (c) 2003, 2010, Oracle and/or its affiliates. All rights reserved.
28 * Use is subject to license terms.
30 * Copyright (c) 2012, 2015, Intel Corporation.
33 * This file is part of Lustre, http://www.lustre.org/
34 * Lustre is a trademark of Sun Microsystems, Inc.
36 * lustre/obdclass/llog.c
38 * OST<->MDS recovery logging infrastructure.
39 * Invariants in implementation:
40 * - we do not share logs among different OST<->MDS connections, so that
41 * if an OST or MDS fails it need only look at log(s) relevant to itself
43 * Author: Andreas Dilger <adilger@clusterfs.com>
44 * Author: Alex Zhuravlev <bzzz@whamcloud.com>
45 * Author: Mikhail Pershin <tappro@whamcloud.com>
48 #define DEBUG_SUBSYSTEM S_LOG
50 #include <linux/kthread.h>
51 #include <llog_swab.h>
52 #include <lustre_log.h>
53 #include <obd_class.h>
54 #include "llog_internal.h"
56 * Allocate a new log or catalog handle
57 * Used inside llog_open().
59 static struct llog_handle *llog_alloc_handle(void)
61 struct llog_handle *loghandle;
63 OBD_ALLOC_PTR(loghandle);
64 if (loghandle == NULL)
67 init_rwsem(&loghandle->lgh_lock);
68 mutex_init(&loghandle->lgh_hdr_mutex);
69 INIT_LIST_HEAD(&loghandle->u.phd.phd_entry);
70 atomic_set(&loghandle->lgh_refcount, 1);
76 * Free llog handle and header data if exists. Used in llog_close() only
78 static void llog_free_handle(struct llog_handle *loghandle)
80 LASSERT(loghandle != NULL);
82 /* failed llog_init_handle */
83 if (loghandle->lgh_hdr == NULL)
86 if (loghandle->lgh_hdr->llh_flags & LLOG_F_IS_PLAIN)
87 LASSERT(list_empty(&loghandle->u.phd.phd_entry));
88 else if (loghandle->lgh_hdr->llh_flags & LLOG_F_IS_CAT)
89 LASSERT(list_empty(&loghandle->u.chd.chd_head));
90 OBD_FREE_LARGE(loghandle->lgh_hdr, loghandle->lgh_hdr_size);
92 OBD_FREE_PTR(loghandle);
95 void llog_handle_get(struct llog_handle *loghandle)
97 atomic_inc(&loghandle->lgh_refcount);
100 void llog_handle_put(struct llog_handle *loghandle)
102 LASSERT(atomic_read(&loghandle->lgh_refcount) > 0);
103 if (atomic_dec_and_test(&loghandle->lgh_refcount))
104 llog_free_handle(loghandle);
107 static int llog_declare_destroy(const struct lu_env *env,
108 struct llog_handle *handle,
111 struct llog_operations *lop;
116 rc = llog_handle2ops(handle, &lop);
119 if (lop->lop_declare_destroy == NULL)
122 rc = lop->lop_declare_destroy(env, handle, th);
127 int llog_trans_destroy(const struct lu_env *env, struct llog_handle *handle,
130 struct llog_operations *lop;
134 rc = llog_handle2ops(handle, &lop);
137 if (lop->lop_destroy == NULL)
140 LASSERT(handle->lgh_obj != NULL);
141 if (!dt_object_exists(handle->lgh_obj))
144 rc = lop->lop_destroy(env, handle, th);
149 int llog_destroy(const struct lu_env *env, struct llog_handle *handle)
151 struct llog_operations *lop;
152 struct dt_device *dt;
158 rc = llog_handle2ops(handle, &lop);
161 if (lop->lop_destroy == NULL)
164 if (handle->lgh_obj == NULL) {
165 /* if lgh_obj == NULL, then it is from client side destroy */
166 rc = lop->lop_destroy(env, handle, NULL);
170 if (!dt_object_exists(handle->lgh_obj))
173 dt = lu2dt_dev(handle->lgh_obj->do_lu.lo_dev);
175 th = dt_trans_create(env, dt);
179 rc = llog_declare_destroy(env, handle, th);
183 rc = dt_trans_start_local(env, dt, th);
187 rc = lop->lop_destroy(env, handle, th);
190 dt_trans_stop(env, dt, th);
194 EXPORT_SYMBOL(llog_destroy);
196 /* returns negative on error; 0 if success; 1 if success & log destroyed */
197 int llog_cancel_rec(const struct lu_env *env, struct llog_handle *loghandle,
200 struct llog_thread_info *lgi = llog_info(env);
201 struct dt_device *dt;
202 struct llog_log_hdr *llh = loghandle->lgh_hdr;
206 bool subtract_count = false;
210 CDEBUG(D_RPCTRACE, "Canceling %d in log "DOSTID"\n", index,
211 POSTID(&loghandle->lgh_id.lgl_oi));
214 CERROR("Can't cancel index 0 which is header\n");
218 LASSERT(loghandle != NULL);
219 LASSERT(loghandle->lgh_ctxt != NULL);
220 LASSERT(loghandle->lgh_obj != NULL);
222 dt = lu2dt_dev(loghandle->lgh_obj->do_lu.lo_dev);
224 th = dt_trans_create(env, dt);
228 rc = llog_declare_write_rec(env, loghandle, &llh->llh_hdr, index, th);
232 if ((llh->llh_flags & LLOG_F_ZAP_WHEN_EMPTY))
233 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 handle->lgh_write_offset = 0;
320 llh->llh_count = 1; /* for the header record */
321 llh->llh_hdr.lrh_type = LLOG_HDR_MAGIC;
322 LASSERT(handle->lgh_ctxt->loc_chunk_size >=
323 LLOG_MIN_CHUNK_SIZE);
324 llh->llh_hdr.lrh_len = handle->lgh_ctxt->loc_chunk_size;
325 llh->llh_hdr.lrh_index = 0;
326 llh->llh_timestamp = cfs_time_current_sec();
328 memcpy(&llh->llh_tgtuuid, uuid,
329 sizeof(llh->llh_tgtuuid));
330 llh->llh_bitmap_offset = offsetof(typeof(*llh), llh_bitmap);
331 /* Since update llog header might also call this function,
332 * let's reset the bitmap to 0 here */
333 memset(LLOG_HDR_BITMAP(llh), 0, llh->llh_hdr.lrh_len -
334 llh->llh_bitmap_offset -
335 sizeof(llh->llh_tail));
336 ext2_set_bit(0, LLOG_HDR_BITMAP(llh));
337 LLOG_HDR_TAIL(llh)->lrt_len = llh->llh_hdr.lrh_len;
338 LLOG_HDR_TAIL(llh)->lrt_index = llh->llh_hdr.lrh_index;
343 EXPORT_SYMBOL(llog_read_header);
345 int llog_init_handle(const struct lu_env *env, struct llog_handle *handle,
346 int flags, struct obd_uuid *uuid)
348 struct llog_log_hdr *llh;
349 enum llog_flag fmt = flags & LLOG_F_EXT_MASK;
351 int chunk_size = handle->lgh_ctxt->loc_chunk_size;
354 LASSERT(handle->lgh_hdr == NULL);
356 LASSERT(chunk_size >= LLOG_MIN_CHUNK_SIZE);
357 OBD_ALLOC_LARGE(llh, chunk_size);
361 handle->lgh_hdr = llh;
362 handle->lgh_hdr_size = chunk_size;
363 /* first assign flags to use llog_client_ops */
364 llh->llh_flags = flags;
365 rc = llog_read_header(env, handle, uuid);
367 if (unlikely((llh->llh_flags & LLOG_F_IS_PLAIN &&
368 flags & LLOG_F_IS_CAT) ||
369 (llh->llh_flags & LLOG_F_IS_CAT &&
370 flags & LLOG_F_IS_PLAIN))) {
371 CERROR("%s: llog type is %s but initializing %s\n",
372 handle->lgh_ctxt->loc_obd->obd_name,
373 llh->llh_flags & LLOG_F_IS_CAT ?
375 flags & LLOG_F_IS_CAT ? "catalog" : "plain");
376 GOTO(out, rc = -EINVAL);
377 } else if (llh->llh_flags &
378 (LLOG_F_IS_PLAIN | LLOG_F_IS_CAT)) {
380 * it is possible to open llog without specifying llog
381 * type so it is taken from llh_flags
383 flags = llh->llh_flags;
385 /* for some reason the llh_flags has no type set */
386 CERROR("llog type is not specified!\n");
387 GOTO(out, rc = -EINVAL);
390 !obd_uuid_equals(uuid, &llh->llh_tgtuuid))) {
391 CERROR("%s: llog uuid mismatch: %s/%s\n",
392 handle->lgh_ctxt->loc_obd->obd_name,
394 (char *)llh->llh_tgtuuid.uuid);
395 GOTO(out, rc = -EEXIST);
398 if (flags & LLOG_F_IS_CAT) {
399 LASSERT(list_empty(&handle->u.chd.chd_head));
400 INIT_LIST_HEAD(&handle->u.chd.chd_head);
401 llh->llh_size = sizeof(struct llog_logid_rec);
402 llh->llh_flags |= LLOG_F_IS_FIXSIZE;
403 } else if (!(flags & LLOG_F_IS_PLAIN)) {
404 CERROR("%s: unknown flags: %#x (expected %#x or %#x)\n",
405 handle->lgh_ctxt->loc_obd->obd_name,
406 flags, LLOG_F_IS_CAT, LLOG_F_IS_PLAIN);
409 llh->llh_flags |= fmt;
412 OBD_FREE_LARGE(llh, chunk_size);
413 handle->lgh_hdr = NULL;
417 EXPORT_SYMBOL(llog_init_handle);
419 static int llog_process_thread(void *arg)
421 struct llog_process_info *lpi = arg;
422 struct llog_handle *loghandle = lpi->lpi_loghandle;
423 struct llog_log_hdr *llh = loghandle->lgh_hdr;
424 struct llog_process_cat_data *cd = lpi->lpi_catdata;
427 __u64 cur_offset, tmp_offset;
428 int rc = 0, index = 1, last_index;
430 int last_called_index = 0;
437 cur_offset = chunk_size = llh->llh_hdr.lrh_len;
438 /* expect chunk_size to be power of two */
439 LASSERT(is_power_of_2(chunk_size));
441 OBD_ALLOC_LARGE(buf, chunk_size);
443 lpi->lpi_rc = -ENOMEM;
448 last_called_index = cd->lpcd_first_idx;
449 index = cd->lpcd_first_idx + 1;
451 if (cd != NULL && cd->lpcd_last_idx)
452 last_index = cd->lpcd_last_idx;
454 last_index = LLOG_HDR_BITMAP_SIZE(llh) - 1;
457 struct llog_rec_hdr *rec;
459 unsigned int buf_offset = 0;
462 /* skip records not set in bitmap */
463 while (index <= last_index &&
464 !ext2_test_bit(index, LLOG_HDR_BITMAP(llh)))
467 /* There are no indices prior the last_index */
468 if (index > last_index)
471 CDEBUG(D_OTHER, "index: %d last_index %d\n", index,
475 /* get the buf with our target record; avoid old garbage */
476 memset(buf, 0, chunk_size);
477 rc = llog_next_block(lpi->lpi_env, loghandle, &saved_index,
478 index, &cur_offset, buf, chunk_size);
482 /* NB: after llog_next_block() call the cur_offset is the
483 * offset of the next block after read one.
484 * The absolute offset of the current chunk is calculated
485 * from cur_offset value and stored in chunk_offset variable.
487 tmp_offset = cur_offset;
488 if (do_div(tmp_offset, chunk_size) != 0) {
489 partial_chunk = true;
490 chunk_offset = cur_offset & ~(chunk_size - 1);
492 partial_chunk = false;
493 chunk_offset = cur_offset - chunk_size;
496 /* NB: when rec->lrh_len is accessed it is already swabbed
497 * since it is used at the "end" of the loop and the rec
498 * swabbing is done at the beginning of the loop. */
499 for (rec = (struct llog_rec_hdr *)(buf + buf_offset);
500 (char *)rec < buf + chunk_size;
501 rec = llog_rec_hdr_next(rec)) {
503 CDEBUG(D_OTHER, "processing rec 0x%p type %#x\n",
506 if (LLOG_REC_HDR_NEEDS_SWABBING(rec))
507 lustre_swab_llog_rec(rec);
509 CDEBUG(D_OTHER, "after swabbing, type=%#x idx=%d\n",
510 rec->lrh_type, rec->lrh_index);
512 /* for partial chunk the end of it is zeroed, check
513 * for index 0 to distinguish it. */
514 if (partial_chunk && rec->lrh_index == 0) {
515 /* concurrent llog_add() might add new records
516 * while llog_processing, check this is not
517 * the case and re-read the current chunk
519 if (index > loghandle->lgh_last_idx)
521 CDEBUG(D_OTHER, "Re-read last llog buffer for "
522 "new records, index %u, last %u\n",
523 index, loghandle->lgh_last_idx);
524 /* save offset inside buffer for the re-read */
525 buf_offset = (char *)rec - (char *)buf;
526 cur_offset = chunk_offset;
530 if (rec->lrh_len == 0 || rec->lrh_len > chunk_size) {
531 CWARN("invalid length %d in llog record for "
532 "index %d/%d\n", rec->lrh_len,
533 rec->lrh_index, index);
534 GOTO(out, rc = -EINVAL);
537 if (rec->lrh_index < index) {
538 CDEBUG(D_OTHER, "skipping lrh_index %d\n",
543 if (rec->lrh_index != index) {
544 CERROR("%s: Invalid record: index %u but "
546 loghandle->lgh_ctxt->loc_obd->obd_name,
547 rec->lrh_index, index);
548 GOTO(out, rc = -ERANGE);
552 "lrh_index: %d lrh_len: %d (%d remains)\n",
553 rec->lrh_index, rec->lrh_len,
554 (int)(buf + chunk_size - (char *)rec));
556 loghandle->lgh_cur_idx = rec->lrh_index;
557 loghandle->lgh_cur_offset = (char *)rec - (char *)buf +
560 /* if set, process the callback on this record */
561 if (ext2_test_bit(index, LLOG_HDR_BITMAP(llh))) {
562 rc = lpi->lpi_cb(lpi->lpi_env, loghandle, rec,
564 last_called_index = index;
565 if (rc == LLOG_PROC_BREAK) {
567 } else if (rc == LLOG_DEL_RECORD) {
568 rc = llog_cancel_rec(lpi->lpi_env,
575 /* exit if the last index is reached */
576 if (index >= last_index)
584 cd->lpcd_last_idx = last_called_index;
586 if (unlikely(rc == -EIO && loghandle->lgh_obj != NULL)) {
587 if (dt_object_remote(loghandle->lgh_obj)) {
588 /* If it is remote object, then -EIO might means
589 * disconnection or eviction, let's return -EAGAIN,
590 * so for update recovery log processing, it will
591 * retry until the umount or abort recovery, see
592 * lod_sub_recovery_thread() */
593 CERROR("%s retry remote llog process\n",
594 loghandle->lgh_ctxt->loc_obd->obd_name);
597 /* something bad happened to the processing of a local
598 * llog file, probably I/O error or the log got
599 * corrupted to be able to finally release the log we
600 * discard any remaining bits in the header */
601 CERROR("Local llog found corrupted\n");
602 while (index <= last_index) {
603 if (ext2_test_bit(index,
604 LLOG_HDR_BITMAP(llh)) != 0)
605 llog_cancel_rec(lpi->lpi_env, loghandle,
613 OBD_FREE_LARGE(buf, chunk_size);
618 static int llog_process_thread_daemonize(void *arg)
620 struct llog_process_info *lpi = arg;
626 /* client env has no keys, tags is just 0 */
627 rc = lu_env_init(&env, LCT_LOCAL | LCT_MG_THREAD);
632 rc = llog_process_thread(arg);
636 complete(&lpi->lpi_completion);
640 int llog_process_or_fork(const struct lu_env *env,
641 struct llog_handle *loghandle,
642 llog_cb_t cb, void *data, void *catdata, bool fork)
644 struct llog_process_info *lpi;
651 CERROR("cannot alloc pointer\n");
654 lpi->lpi_loghandle = loghandle;
656 lpi->lpi_cbdata = data;
657 lpi->lpi_catdata = catdata;
660 struct task_struct *task;
662 /* The new thread can't use parent env,
663 * init the new one in llog_process_thread_daemonize. */
665 init_completion(&lpi->lpi_completion);
666 task = kthread_run(llog_process_thread_daemonize, lpi,
667 "llog_process_thread");
670 CERROR("%s: cannot start thread: rc = %d\n",
671 loghandle->lgh_ctxt->loc_obd->obd_name, rc);
674 wait_for_completion(&lpi->lpi_completion);
677 llog_process_thread(lpi);
685 EXPORT_SYMBOL(llog_process_or_fork);
687 int llog_process(const struct lu_env *env, struct llog_handle *loghandle,
688 llog_cb_t cb, void *data, void *catdata)
691 rc = llog_process_or_fork(env, loghandle, cb, data, catdata, true);
692 return rc == LLOG_DEL_PLAIN ? 0 : rc;
694 EXPORT_SYMBOL(llog_process);
696 int llog_reverse_process(const struct lu_env *env,
697 struct llog_handle *loghandle, llog_cb_t cb,
698 void *data, void *catdata)
700 struct llog_log_hdr *llh = loghandle->lgh_hdr;
701 struct llog_process_cat_data *cd = catdata;
703 int rc = 0, first_index = 1, index, idx;
704 __u32 chunk_size = llh->llh_hdr.lrh_len;
707 OBD_ALLOC_LARGE(buf, chunk_size);
712 first_index = cd->lpcd_first_idx + 1;
713 if (cd != NULL && cd->lpcd_last_idx)
714 index = cd->lpcd_last_idx;
716 index = LLOG_HDR_BITMAP_SIZE(llh) - 1;
719 struct llog_rec_hdr *rec;
720 struct llog_rec_tail *tail;
722 /* skip records not set in bitmap */
723 while (index >= first_index &&
724 !ext2_test_bit(index, LLOG_HDR_BITMAP(llh)))
727 LASSERT(index >= first_index - 1);
728 if (index == first_index - 1)
731 /* get the buf with our target record; avoid old garbage */
732 memset(buf, 0, chunk_size);
733 rc = llog_prev_block(env, loghandle, index, buf, chunk_size);
738 idx = rec->lrh_index;
739 CDEBUG(D_RPCTRACE, "index %u : idx %u\n", index, idx);
740 while (idx < index) {
741 rec = (void *)rec + rec->lrh_len;
742 if (LLOG_REC_HDR_NEEDS_SWABBING(rec))
743 lustre_swab_llog_rec(rec);
746 LASSERT(idx == index);
747 tail = (void *)rec + rec->lrh_len - sizeof(*tail);
749 /* process records in buffer, starting where we found one */
750 while ((void *)tail > buf) {
751 if (tail->lrt_index == 0)
752 GOTO(out, rc = 0); /* no more records */
754 /* if set, process the callback on this record */
755 if (ext2_test_bit(index, LLOG_HDR_BITMAP(llh))) {
756 rec = (void *)tail - tail->lrt_len +
759 rc = cb(env, loghandle, rec, data);
760 if (rc == LLOG_PROC_BREAK) {
762 } else if (rc == LLOG_DEL_RECORD) {
763 rc = llog_cancel_rec(env, loghandle,
770 /* previous record, still in buffer? */
772 if (index < first_index)
774 tail = (void *)tail - tail->lrt_len;
780 OBD_FREE_LARGE(buf, chunk_size);
783 EXPORT_SYMBOL(llog_reverse_process);
789 * llog_open - open llog, may not exist
790 * llog_exist - check if llog exists
791 * llog_close - close opened llog, pair for open, frees llog_handle
792 * llog_declare_create - declare llog creation
793 * llog_create - create new llog on disk, need transaction handle
794 * llog_declare_write_rec - declaration of llog write
795 * llog_write_rec - write llog record on disk, need transaction handle
796 * llog_declare_add - declare llog catalog record addition
797 * llog_add - add llog record in catalog, need transaction handle
799 int llog_exist(struct llog_handle *loghandle)
801 struct llog_operations *lop;
806 rc = llog_handle2ops(loghandle, &lop);
809 if (lop->lop_exist == NULL)
812 rc = lop->lop_exist(loghandle);
815 EXPORT_SYMBOL(llog_exist);
817 int llog_declare_create(const struct lu_env *env,
818 struct llog_handle *loghandle, struct thandle *th)
820 struct llog_operations *lop;
825 rc = llog_handle2ops(loghandle, &lop);
828 if (lop->lop_declare_create == NULL)
831 raised = cfs_cap_raised(CFS_CAP_SYS_RESOURCE);
833 cfs_cap_raise(CFS_CAP_SYS_RESOURCE);
834 rc = lop->lop_declare_create(env, loghandle, th);
836 cfs_cap_lower(CFS_CAP_SYS_RESOURCE);
840 int llog_create(const struct lu_env *env, struct llog_handle *handle,
843 struct llog_operations *lop;
848 rc = llog_handle2ops(handle, &lop);
851 if (lop->lop_create == NULL)
854 raised = cfs_cap_raised(CFS_CAP_SYS_RESOURCE);
856 cfs_cap_raise(CFS_CAP_SYS_RESOURCE);
857 rc = lop->lop_create(env, handle, th);
859 cfs_cap_lower(CFS_CAP_SYS_RESOURCE);
863 int llog_declare_write_rec(const struct lu_env *env,
864 struct llog_handle *handle,
865 struct llog_rec_hdr *rec, int idx,
868 struct llog_operations *lop;
873 rc = llog_handle2ops(handle, &lop);
877 if (lop->lop_declare_write_rec == NULL)
880 raised = cfs_cap_raised(CFS_CAP_SYS_RESOURCE);
882 cfs_cap_raise(CFS_CAP_SYS_RESOURCE);
883 rc = lop->lop_declare_write_rec(env, handle, rec, idx, th);
885 cfs_cap_lower(CFS_CAP_SYS_RESOURCE);
889 int llog_write_rec(const struct lu_env *env, struct llog_handle *handle,
890 struct llog_rec_hdr *rec, struct llog_cookie *logcookies,
891 int idx, struct thandle *th)
893 struct llog_operations *lop;
894 int raised, rc, buflen;
898 rc = llog_handle2ops(handle, &lop);
903 if (lop->lop_write_rec == NULL)
906 buflen = rec->lrh_len;
907 LASSERT(cfs_size_round(buflen) == buflen);
909 raised = cfs_cap_raised(CFS_CAP_SYS_RESOURCE);
911 cfs_cap_raise(CFS_CAP_SYS_RESOURCE);
912 rc = lop->lop_write_rec(env, handle, rec, logcookies, idx, th);
914 cfs_cap_lower(CFS_CAP_SYS_RESOURCE);
918 int llog_add(const struct lu_env *env, struct llog_handle *lgh,
919 struct llog_rec_hdr *rec, struct llog_cookie *logcookies,
926 if (lgh->lgh_logops->lop_add == NULL)
929 raised = cfs_cap_raised(CFS_CAP_SYS_RESOURCE);
931 cfs_cap_raise(CFS_CAP_SYS_RESOURCE);
932 rc = lgh->lgh_logops->lop_add(env, lgh, rec, logcookies, th);
934 cfs_cap_lower(CFS_CAP_SYS_RESOURCE);
937 EXPORT_SYMBOL(llog_add);
939 int llog_declare_add(const struct lu_env *env, struct llog_handle *lgh,
940 struct llog_rec_hdr *rec, struct thandle *th)
946 if (lgh->lgh_logops->lop_declare_add == NULL)
949 raised = cfs_cap_raised(CFS_CAP_SYS_RESOURCE);
951 cfs_cap_raise(CFS_CAP_SYS_RESOURCE);
952 rc = lgh->lgh_logops->lop_declare_add(env, lgh, rec, th);
954 cfs_cap_lower(CFS_CAP_SYS_RESOURCE);
957 EXPORT_SYMBOL(llog_declare_add);
960 * Helper function to open llog or create it if doesn't exist.
961 * It hides all transaction handling from caller.
963 int llog_open_create(const struct lu_env *env, struct llog_ctxt *ctxt,
964 struct llog_handle **res, struct llog_logid *logid,
973 rc = llog_open(env, ctxt, res, logid, name, LLOG_OPEN_NEW);
977 if (llog_exist(*res))
980 LASSERT((*res)->lgh_obj != NULL);
982 d = lu2dt_dev((*res)->lgh_obj->do_lu.lo_dev);
984 th = dt_trans_create(env, d);
986 GOTO(out, rc = PTR_ERR(th));
988 /* Create update llog object synchronously, which
989 * happens during inialization process see
990 * lod_sub_prep_llog(), to make sure the update
991 * llog object is created before corss-MDT writing
992 * updates into the llog object */
993 if (ctxt->loc_flags & LLOG_CTXT_FLAG_NORMAL_FID)
996 th->th_wait_submit = 1;
997 rc = llog_declare_create(env, *res, th);
999 rc = dt_trans_start_local(env, d, th);
1001 rc = llog_create(env, *res, th);
1003 dt_trans_stop(env, d, th);
1006 llog_close(env, *res);
1009 EXPORT_SYMBOL(llog_open_create);
1012 * Helper function to delete existent llog.
1014 int llog_erase(const struct lu_env *env, struct llog_ctxt *ctxt,
1015 struct llog_logid *logid, char *name)
1017 struct llog_handle *handle;
1022 /* nothing to erase */
1023 if (name == NULL && logid == NULL)
1026 rc = llog_open(env, ctxt, &handle, logid, name, LLOG_OPEN_EXISTS);
1030 rc = llog_init_handle(env, handle, LLOG_F_IS_PLAIN, NULL);
1032 rc = llog_destroy(env, handle);
1034 rc2 = llog_close(env, handle);
1039 EXPORT_SYMBOL(llog_erase);
1042 * Helper function for write record in llog.
1043 * It hides all transaction handling from caller.
1044 * Valid only with local llog.
1046 int llog_write(const struct lu_env *env, struct llog_handle *loghandle,
1047 struct llog_rec_hdr *rec, int idx)
1049 struct dt_device *dt;
1056 LASSERT(loghandle->lgh_ctxt);
1057 LASSERT(loghandle->lgh_obj != NULL);
1059 dt = lu2dt_dev(loghandle->lgh_obj->do_lu.lo_dev);
1061 th = dt_trans_create(env, dt);
1063 RETURN(PTR_ERR(th));
1065 rc = llog_declare_write_rec(env, loghandle, rec, idx, th);
1067 GOTO(out_trans, rc);
1069 th->th_wait_submit = 1;
1070 rc = dt_trans_start_local(env, dt, th);
1072 GOTO(out_trans, rc);
1074 down_write(&loghandle->lgh_lock);
1075 rc = llog_write_rec(env, loghandle, rec, NULL, idx, th);
1076 up_write(&loghandle->lgh_lock);
1078 dt_trans_stop(env, dt, th);
1081 EXPORT_SYMBOL(llog_write);
1083 int llog_open(const struct lu_env *env, struct llog_ctxt *ctxt,
1084 struct llog_handle **lgh, struct llog_logid *logid,
1085 char *name, enum llog_open_param open_param)
1093 LASSERT(ctxt->loc_logops);
1095 if (ctxt->loc_logops->lop_open == NULL) {
1097 RETURN(-EOPNOTSUPP);
1100 *lgh = llog_alloc_handle();
1103 (*lgh)->lgh_ctxt = ctxt;
1104 (*lgh)->lgh_logops = ctxt->loc_logops;
1106 raised = cfs_cap_raised(CFS_CAP_SYS_RESOURCE);
1108 cfs_cap_raise(CFS_CAP_SYS_RESOURCE);
1109 rc = ctxt->loc_logops->lop_open(env, *lgh, logid, name, open_param);
1111 cfs_cap_lower(CFS_CAP_SYS_RESOURCE);
1113 llog_free_handle(*lgh);
1118 EXPORT_SYMBOL(llog_open);
1120 int llog_close(const struct lu_env *env, struct llog_handle *loghandle)
1122 struct llog_operations *lop;
1127 rc = llog_handle2ops(loghandle, &lop);
1130 if (lop->lop_close == NULL)
1131 GOTO(out, rc = -EOPNOTSUPP);
1132 rc = lop->lop_close(env, loghandle);
1134 llog_handle_put(loghandle);
1137 EXPORT_SYMBOL(llog_close);
1140 * Helper function to get the llog size in records. It is used by MGS
1141 * mostly to check that config llog exists and contains data.
1143 * \param[in] env execution environment
1144 * \param[in] ctxt llog context
1145 * \param[in] name llog name
1147 * \retval true if there are records in llog besides a header
1148 * \retval false on error or llog without records
1150 int llog_is_empty(const struct lu_env *env, struct llog_ctxt *ctxt,
1153 struct llog_handle *llh;
1156 rc = llog_open(env, ctxt, &llh, NULL, name, LLOG_OPEN_EXISTS);
1158 if (likely(rc == -ENOENT))
1163 rc = llog_init_handle(env, llh, LLOG_F_IS_PLAIN, NULL);
1165 GOTO(out_close, rc);
1166 rc = llog_get_size(llh);
1169 llog_close(env, llh);
1171 /* The header is record 1, the llog is still considered as empty
1172 * if there is only header */
1175 EXPORT_SYMBOL(llog_is_empty);
1177 int llog_copy_handler(const struct lu_env *env, struct llog_handle *llh,
1178 struct llog_rec_hdr *rec, void *data)
1180 struct llog_handle *copy_llh = data;
1182 /* Append all records */
1183 return llog_write(env, copy_llh, rec, LLOG_NEXT_IDX);
1186 /* backup plain llog */
1187 int llog_backup(const struct lu_env *env, struct obd_device *obd,
1188 struct llog_ctxt *ctxt, struct llog_ctxt *bctxt,
1189 char *name, char *backup)
1191 struct llog_handle *llh, *bllh;
1196 /* open original log */
1197 rc = llog_open(env, ctxt, &llh, NULL, name, LLOG_OPEN_EXISTS);
1199 /* the -ENOENT case is also reported to the caller
1200 * but silently so it should handle that if needed.
1203 CERROR("%s: failed to open log %s: rc = %d\n",
1204 obd->obd_name, name, rc);
1208 rc = llog_init_handle(env, llh, LLOG_F_IS_PLAIN, NULL);
1210 GOTO(out_close, rc);
1212 /* Make sure there's no old backup log */
1213 rc = llog_erase(env, bctxt, NULL, backup);
1214 if (rc < 0 && rc != -ENOENT)
1215 GOTO(out_close, rc);
1217 /* open backup log */
1218 rc = llog_open_create(env, bctxt, &bllh, NULL, backup);
1220 CERROR("%s: failed to open backup logfile %s: rc = %d\n",
1221 obd->obd_name, backup, rc);
1222 GOTO(out_close, rc);
1225 /* check that backup llog is not the same object as original one */
1226 if (llh->lgh_obj == bllh->lgh_obj) {
1227 CERROR("%s: backup llog %s to itself (%s), objects %p/%p\n",
1228 obd->obd_name, name, backup, llh->lgh_obj,
1230 GOTO(out_backup, rc = -EEXIST);
1233 rc = llog_init_handle(env, bllh, LLOG_F_IS_PLAIN, NULL);
1235 GOTO(out_backup, rc);
1237 /* Copy log record by record */
1238 rc = llog_process_or_fork(env, llh, llog_copy_handler, (void *)bllh,
1241 CERROR("%s: failed to backup log %s: rc = %d\n",
1242 obd->obd_name, name, rc);
1244 llog_close(env, bllh);
1246 llog_close(env, llh);
1249 EXPORT_SYMBOL(llog_backup);
1251 /* Get size of llog */
1252 __u64 llog_size(const struct lu_env *env, struct llog_handle *llh)
1257 rc = llh->lgh_obj->do_ops->do_attr_get(env, llh->lgh_obj, &la);
1259 CERROR("%s: attr_get failed, rc = %d\n",
1260 llh->lgh_ctxt->loc_obd->obd_name, rc);
1266 EXPORT_SYMBOL(llog_size);