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, 2017, 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;
203 bool subtract_count = false;
207 LASSERT(loghandle != NULL);
208 LASSERT(loghandle->lgh_ctxt != NULL);
209 LASSERT(loghandle->lgh_obj != NULL);
211 llh = loghandle->lgh_hdr;
213 CDEBUG(D_RPCTRACE, "Canceling %d in log "DFID"\n", index,
214 PFID(&loghandle->lgh_id.lgl_oi.oi_fid));
217 CERROR("Can't cancel index 0 which is header\n");
221 dt = lu2dt_dev(loghandle->lgh_obj->do_lu.lo_dev);
223 th = dt_trans_create(env, dt);
227 rc = llog_declare_write_rec(env, loghandle, &llh->llh_hdr, index, th);
231 if ((llh->llh_flags & LLOG_F_ZAP_WHEN_EMPTY)) {
232 rc = llog_declare_destroy(env, loghandle, th);
237 th->th_wait_submit = 1;
238 rc = dt_trans_start_local(env, dt, th);
242 down_write(&loghandle->lgh_lock);
244 mutex_lock(&loghandle->lgh_hdr_mutex);
245 if (!ext2_clear_bit(index, LLOG_HDR_BITMAP(llh))) {
246 CDEBUG(D_RPCTRACE, "Catalog index %u already clear?\n", index);
247 GOTO(out_unlock, rc);
250 loghandle->lgh_hdr->llh_count--;
251 subtract_count = true;
252 /* Pass this index to llog_osd_write_rec(), which will use the index
253 * to only update the necesary bitmap. */
254 lgi->lgi_cookie.lgc_index = index;
256 rc = llog_write_rec(env, loghandle, &llh->llh_hdr, &lgi->lgi_cookie,
257 LLOG_HEADER_IDX, th);
259 GOTO(out_unlock, rc);
261 if ((llh->llh_flags & LLOG_F_ZAP_WHEN_EMPTY) &&
262 (llh->llh_count == 1) &&
263 ((loghandle->lgh_last_idx == LLOG_HDR_BITMAP_SIZE(llh) - 1) ||
264 (loghandle->u.phd.phd_cat_handle != NULL &&
265 loghandle->u.phd.phd_cat_handle->u.chd.chd_current_log !=
267 /* never try to destroy it again */
268 llh->llh_flags &= ~LLOG_F_ZAP_WHEN_EMPTY;
269 rc = llog_trans_destroy(env, loghandle, th);
271 /* Sigh, can not destroy the final plain llog, but
272 * the bitmap has been clearly, so the record can not
273 * be accessed anymore, let's return 0 for now, and
274 * the orphan will be handled by LFSCK. */
275 CERROR("%s: can't destroy empty llog "DFID": rc = %d\n",
276 loghandle->lgh_ctxt->loc_obd->obd_name,
277 PFID(&loghandle->lgh_id.lgl_oi.oi_fid), rc);
278 GOTO(out_unlock, rc = 0);
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 = ktime_get_real_seconds();
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;
424 struct llog_thread_info *lti;
428 int rc = 0, index = 1, last_index;
430 int last_called_index = 0;
431 bool repeated = false;
438 lti = lpi->lpi_env == NULL ? NULL : llog_info(lpi->lpi_env);
440 cur_offset = chunk_size = llh->llh_hdr.lrh_len;
441 /* expect chunk_size to be power of two */
442 LASSERT(is_power_of_2(chunk_size));
444 OBD_ALLOC_LARGE(buf, chunk_size);
446 lpi->lpi_rc = -ENOMEM;
451 last_called_index = cd->lpcd_first_idx;
452 index = cd->lpcd_first_idx + 1;
454 if (cd != NULL && cd->lpcd_last_idx)
455 last_index = cd->lpcd_last_idx;
457 last_index = LLOG_HDR_BITMAP_SIZE(llh) - 1;
460 struct llog_rec_hdr *rec;
461 off_t chunk_offset = 0;
462 unsigned int buf_offset = 0;
466 /* skip records not set in bitmap */
467 while (index <= last_index &&
468 !ext2_test_bit(index, LLOG_HDR_BITMAP(llh)))
471 /* There are no indices prior the last_index */
472 if (index > last_index)
475 CDEBUG(D_OTHER, "index: %d last_index %d\n", index,
479 /* get the buf with our target record; avoid old garbage */
480 memset(buf, 0, chunk_size);
481 /* the record index for outdated chunk data */
482 lh_last_idx = loghandle->lgh_last_idx + 1;
483 rc = llog_next_block(lpi->lpi_env, loghandle, &saved_index,
484 index, &cur_offset, buf, chunk_size);
486 CDEBUG(D_OTHER, "cur_offset %llu, chunk_offset %llu,"
487 " buf_offset %u, rc = %d\n", cur_offset,
488 (__u64)chunk_offset, buf_offset, rc);
489 /* we`ve tried to reread the chunk, but there is no
491 if (rc == -EIO && repeated && (chunk_offset + buf_offset) ==
497 /* NB: after llog_next_block() call the cur_offset is the
498 * offset of the next block after read one.
499 * The absolute offset of the current chunk is calculated
500 * from cur_offset value and stored in chunk_offset variable.
502 if ((cur_offset & (chunk_size - 1)) != 0) {
503 partial_chunk = true;
504 chunk_offset = cur_offset & ~(chunk_size - 1);
506 partial_chunk = false;
507 chunk_offset = cur_offset - chunk_size;
510 /* NB: when rec->lrh_len is accessed it is already swabbed
511 * since it is used at the "end" of the loop and the rec
512 * swabbing is done at the beginning of the loop. */
513 for (rec = (struct llog_rec_hdr *)(buf + buf_offset);
514 (char *)rec < buf + chunk_size;
515 rec = llog_rec_hdr_next(rec)) {
517 CDEBUG(D_OTHER, "processing rec 0x%p type %#x\n",
520 if (LLOG_REC_HDR_NEEDS_SWABBING(rec))
521 lustre_swab_llog_rec(rec);
523 CDEBUG(D_OTHER, "after swabbing, type=%#x idx=%d\n",
524 rec->lrh_type, rec->lrh_index);
526 /* the bitmap could be changed during processing
527 * records from the chunk. For wrapped catalog
528 * it means we can read deleted record and try to
529 * process it. Check this case and reread the chunk. */
531 /* for partial chunk the end of it is zeroed, check
532 * for index 0 to distinguish it. */
533 if ((partial_chunk && rec->lrh_index == 0) ||
534 (index == lh_last_idx &&
535 lh_last_idx != (loghandle->lgh_last_idx + 1))) {
536 /* concurrent llog_add() might add new records
537 * while llog_processing, check this is not
538 * the case and re-read the current chunk
541 /* lgh_last_idx could be less then index
542 * for catalog, if catalog is wrapped */
543 if ((index > loghandle->lgh_last_idx &&
544 !(loghandle->lgh_hdr->llh_flags &
545 LLOG_F_IS_CAT)) || repeated ||
546 (loghandle->lgh_obj != NULL &&
547 dt_object_remote(loghandle->lgh_obj)))
549 /* <2 records means no more records
550 * if the last record we processed was
551 * the final one, then the underlying
552 * object might have been destroyed yet.
553 * we better don't access that.. */
554 mutex_lock(&loghandle->lgh_hdr_mutex);
555 records = loghandle->lgh_hdr->llh_count;
556 mutex_unlock(&loghandle->lgh_hdr_mutex);
559 CDEBUG(D_OTHER, "Re-read last llog buffer for "
560 "new records, index %u, last %u\n",
561 index, loghandle->lgh_last_idx);
562 /* save offset inside buffer for the re-read */
563 buf_offset = (char *)rec - (char *)buf;
564 cur_offset = chunk_offset;
571 if (rec->lrh_len == 0 || rec->lrh_len > chunk_size) {
572 CWARN("%s: invalid length %d in llog "DFID
573 "record for index %d/%d\n",
574 loghandle->lgh_ctxt->loc_obd->obd_name,
576 PFID(&loghandle->lgh_id.lgl_oi.oi_fid),
577 rec->lrh_index, index);
579 GOTO(out, rc = -EINVAL);
582 if (rec->lrh_index < index) {
583 CDEBUG(D_OTHER, "skipping lrh_index %d\n",
588 if (rec->lrh_index != index) {
589 CERROR("%s: "DFID" Invalid record: index %u"
590 " but expected %u\n",
591 loghandle->lgh_ctxt->loc_obd->obd_name,
592 PFID(&loghandle->lgh_id.lgl_oi.oi_fid),
593 rec->lrh_index, index);
594 GOTO(out, rc = -ERANGE);
598 "lrh_index: %d lrh_len: %d (%d remains)\n",
599 rec->lrh_index, rec->lrh_len,
600 (int)(buf + chunk_size - (char *)rec));
602 /* lgh_cur_offset is used only at llog_test_3 */
603 loghandle->lgh_cur_offset = (char *)rec - (char *)buf +
606 /* if set, process the callback on this record */
607 if (ext2_test_bit(index, LLOG_HDR_BITMAP(llh))) {
608 struct llog_cookie *lgc;
613 lgc = <i->lgi_cookie;
614 /* store lu_env for recursive calls */
615 tmp_off = lgc->lgc_offset;
616 tmp_idx = lgc->lgc_index;
618 lgc->lgc_offset = (char *)rec -
619 (char *)buf + chunk_offset;
620 lgc->lgc_index = rec->lrh_index;
622 /* using lu_env for passing record offset to
623 * llog_write through various callbacks */
624 rc = lpi->lpi_cb(lpi->lpi_env, loghandle, rec,
626 last_called_index = index;
629 lgc->lgc_offset = tmp_off;
630 lgc->lgc_index = tmp_idx;
633 if (rc == LLOG_PROC_BREAK) {
635 } else if (rc == LLOG_DEL_RECORD) {
636 rc = llog_cancel_rec(lpi->lpi_env,
642 /* some stupid callbacks directly cancel records
643 * and delete llog. Check it and stop
645 if (loghandle->lgh_hdr == NULL ||
646 loghandle->lgh_hdr->llh_count == 1)
649 /* exit if the last index is reached */
650 if (index >= last_index)
658 cd->lpcd_last_idx = last_called_index;
660 if (unlikely(rc == -EIO && loghandle->lgh_obj != NULL)) {
661 if (dt_object_remote(loghandle->lgh_obj)) {
662 /* If it is remote object, then -EIO might means
663 * disconnection or eviction, let's return -EAGAIN,
664 * so for update recovery log processing, it will
665 * retry until the umount or abort recovery, see
666 * lod_sub_recovery_thread() */
667 CERROR("%s retry remote llog process\n",
668 loghandle->lgh_ctxt->loc_obd->obd_name);
671 /* something bad happened to the processing of a local
672 * llog file, probably I/O error or the log got
673 * corrupted to be able to finally release the log we
674 * discard any remaining bits in the header */
675 CERROR("%s: Local llog found corrupted #"DOSTID":%x"
676 " %s index %d count %d\n",
677 loghandle->lgh_ctxt->loc_obd->obd_name,
678 POSTID(&loghandle->lgh_id.lgl_oi),
679 loghandle->lgh_id.lgl_ogen,
680 ((llh->llh_flags & LLOG_F_IS_CAT) ? "catalog" :
681 "plain"), index, llh->llh_count);
683 while (index <= last_index) {
684 if (ext2_test_bit(index,
685 LLOG_HDR_BITMAP(llh)) != 0)
686 llog_cancel_rec(lpi->lpi_env, loghandle,
694 OBD_FREE_LARGE(buf, chunk_size);
699 static int llog_process_thread_daemonize(void *arg)
701 struct llog_process_info *lpi = arg;
704 struct nsproxy *new_ns, *curr_ns = current->nsproxy;
706 task_lock(lpi->lpi_reftask);
707 new_ns = lpi->lpi_reftask->nsproxy;
708 if (curr_ns != new_ns) {
711 current->nsproxy = new_ns;
712 /* XXX: we should call put_nsproxy() instead of
713 * atomic_dec(&ns->count) directly. But put_nsproxy() cannot be
714 * used outside of the kernel itself, because it calls
715 * free_nsproxy() which is not exported by the kernel
716 * (defined in kernel/nsproxy.c) */
717 atomic_dec(&curr_ns->count);
719 task_unlock(lpi->lpi_reftask);
723 /* client env has no keys, tags is just 0 */
724 rc = lu_env_init(&env, LCT_LOCAL | LCT_MG_THREAD);
729 rc = llog_process_thread(arg);
733 complete(&lpi->lpi_completion);
737 int llog_process_or_fork(const struct lu_env *env,
738 struct llog_handle *loghandle,
739 llog_cb_t cb, void *data, void *catdata, bool fork)
741 struct llog_process_info *lpi;
748 CERROR("cannot alloc pointer\n");
751 lpi->lpi_loghandle = loghandle;
753 lpi->lpi_cbdata = data;
754 lpi->lpi_catdata = catdata;
757 struct task_struct *task;
759 /* The new thread can't use parent env,
760 * init the new one in llog_process_thread_daemonize. */
762 init_completion(&lpi->lpi_completion);
763 /* take reference to current, so that
764 * llog_process_thread_daemonize() can use it to switch to
765 * namespace associated with current */
766 lpi->lpi_reftask = current;
767 task = kthread_run(llog_process_thread_daemonize, lpi,
768 "llog_process_thread");
771 CERROR("%s: cannot start thread: rc = %d\n",
772 loghandle->lgh_ctxt->loc_obd->obd_name, rc);
775 wait_for_completion(&lpi->lpi_completion);
778 llog_process_thread(lpi);
786 EXPORT_SYMBOL(llog_process_or_fork);
788 int llog_process(const struct lu_env *env, struct llog_handle *loghandle,
789 llog_cb_t cb, void *data, void *catdata)
792 rc = llog_process_or_fork(env, loghandle, cb, data, catdata, true);
793 return rc == LLOG_DEL_PLAIN ? 0 : rc;
795 EXPORT_SYMBOL(llog_process);
797 int llog_reverse_process(const struct lu_env *env,
798 struct llog_handle *loghandle, llog_cb_t cb,
799 void *data, void *catdata)
801 struct llog_log_hdr *llh = loghandle->lgh_hdr;
802 struct llog_process_cat_data *cd = catdata;
804 int rc = 0, first_index = 1, index, idx;
805 __u32 chunk_size = llh->llh_hdr.lrh_len;
808 OBD_ALLOC_LARGE(buf, chunk_size);
813 first_index = cd->lpcd_first_idx + 1;
814 if (cd != NULL && cd->lpcd_last_idx)
815 index = cd->lpcd_last_idx;
817 index = LLOG_HDR_BITMAP_SIZE(llh) - 1;
820 struct llog_rec_hdr *rec;
821 struct llog_rec_tail *tail;
823 /* skip records not set in bitmap */
824 while (index >= first_index &&
825 !ext2_test_bit(index, LLOG_HDR_BITMAP(llh)))
828 LASSERT(index >= first_index - 1);
829 if (index == first_index - 1)
832 /* get the buf with our target record; avoid old garbage */
833 memset(buf, 0, chunk_size);
834 rc = llog_prev_block(env, loghandle, index, buf, chunk_size);
839 idx = rec->lrh_index;
840 CDEBUG(D_RPCTRACE, "index %u : idx %u\n", index, idx);
841 while (idx < index) {
842 rec = (void *)rec + rec->lrh_len;
843 if (LLOG_REC_HDR_NEEDS_SWABBING(rec))
844 lustre_swab_llog_rec(rec);
847 LASSERT(idx == index);
848 tail = (void *)rec + rec->lrh_len - sizeof(*tail);
850 /* process records in buffer, starting where we found one */
851 while ((void *)tail > buf) {
852 if (tail->lrt_index == 0)
853 GOTO(out, rc = 0); /* no more records */
855 /* if set, process the callback on this record */
856 if (ext2_test_bit(index, LLOG_HDR_BITMAP(llh))) {
857 rec = (void *)tail - tail->lrt_len +
860 rc = cb(env, loghandle, rec, data);
861 if (rc == LLOG_PROC_BREAK) {
863 } else if (rc == LLOG_DEL_RECORD) {
864 rc = llog_cancel_rec(env, loghandle,
871 /* previous record, still in buffer? */
873 if (index < first_index)
875 tail = (void *)tail - tail->lrt_len;
881 OBD_FREE_LARGE(buf, chunk_size);
884 EXPORT_SYMBOL(llog_reverse_process);
890 * llog_open - open llog, may not exist
891 * llog_exist - check if llog exists
892 * llog_close - close opened llog, pair for open, frees llog_handle
893 * llog_declare_create - declare llog creation
894 * llog_create - create new llog on disk, need transaction handle
895 * llog_declare_write_rec - declaration of llog write
896 * llog_write_rec - write llog record on disk, need transaction handle
897 * llog_declare_add - declare llog catalog record addition
898 * llog_add - add llog record in catalog, need transaction handle
900 int llog_exist(struct llog_handle *loghandle)
902 struct llog_operations *lop;
907 rc = llog_handle2ops(loghandle, &lop);
910 if (lop->lop_exist == NULL)
913 rc = lop->lop_exist(loghandle);
916 EXPORT_SYMBOL(llog_exist);
918 int llog_declare_create(const struct lu_env *env,
919 struct llog_handle *loghandle, struct thandle *th)
921 struct llog_operations *lop;
926 rc = llog_handle2ops(loghandle, &lop);
929 if (lop->lop_declare_create == NULL)
932 raised = cfs_cap_raised(CFS_CAP_SYS_RESOURCE);
934 cfs_cap_raise(CFS_CAP_SYS_RESOURCE);
935 rc = lop->lop_declare_create(env, loghandle, th);
937 cfs_cap_lower(CFS_CAP_SYS_RESOURCE);
941 int llog_create(const struct lu_env *env, struct llog_handle *handle,
944 struct llog_operations *lop;
949 rc = llog_handle2ops(handle, &lop);
952 if (lop->lop_create == NULL)
955 raised = cfs_cap_raised(CFS_CAP_SYS_RESOURCE);
957 cfs_cap_raise(CFS_CAP_SYS_RESOURCE);
958 rc = lop->lop_create(env, handle, th);
960 cfs_cap_lower(CFS_CAP_SYS_RESOURCE);
964 int llog_declare_write_rec(const struct lu_env *env,
965 struct llog_handle *handle,
966 struct llog_rec_hdr *rec, int idx,
969 struct llog_operations *lop;
974 rc = llog_handle2ops(handle, &lop);
978 if (lop->lop_declare_write_rec == NULL)
981 raised = cfs_cap_raised(CFS_CAP_SYS_RESOURCE);
983 cfs_cap_raise(CFS_CAP_SYS_RESOURCE);
984 rc = lop->lop_declare_write_rec(env, handle, rec, idx, th);
986 cfs_cap_lower(CFS_CAP_SYS_RESOURCE);
990 int llog_write_rec(const struct lu_env *env, struct llog_handle *handle,
991 struct llog_rec_hdr *rec, struct llog_cookie *logcookies,
992 int idx, struct thandle *th)
994 struct llog_operations *lop;
995 int raised, rc, buflen;
999 /* API sanity checks */
1000 if (handle == NULL) {
1001 CERROR("loghandle is missed\n");
1003 } else if (handle->lgh_obj == NULL) {
1004 CERROR("loghandle %p with NULL object\n",
1007 } else if (th == NULL) {
1008 CERROR("%s: missed transaction handle\n",
1009 handle->lgh_obj->do_lu.lo_dev->ld_obd->obd_name);
1011 } else if (handle->lgh_hdr == NULL) {
1012 CERROR("%s: loghandle %p with no header\n",
1013 handle->lgh_obj->do_lu.lo_dev->ld_obd->obd_name,
1018 rc = llog_handle2ops(handle, &lop);
1022 if (lop->lop_write_rec == NULL)
1023 RETURN(-EOPNOTSUPP);
1025 buflen = rec->lrh_len;
1026 LASSERT(cfs_size_round(buflen) == buflen);
1028 raised = cfs_cap_raised(CFS_CAP_SYS_RESOURCE);
1030 cfs_cap_raise(CFS_CAP_SYS_RESOURCE);
1031 rc = lop->lop_write_rec(env, handle, rec, logcookies, idx, th);
1033 cfs_cap_lower(CFS_CAP_SYS_RESOURCE);
1037 int llog_add(const struct lu_env *env, struct llog_handle *lgh,
1038 struct llog_rec_hdr *rec, struct llog_cookie *logcookies,
1045 if (lgh->lgh_logops->lop_add == NULL)
1046 RETURN(-EOPNOTSUPP);
1048 raised = cfs_cap_raised(CFS_CAP_SYS_RESOURCE);
1050 cfs_cap_raise(CFS_CAP_SYS_RESOURCE);
1051 rc = lgh->lgh_logops->lop_add(env, lgh, rec, logcookies, th);
1053 cfs_cap_lower(CFS_CAP_SYS_RESOURCE);
1056 EXPORT_SYMBOL(llog_add);
1058 int llog_declare_add(const struct lu_env *env, struct llog_handle *lgh,
1059 struct llog_rec_hdr *rec, struct thandle *th)
1065 if (lgh->lgh_logops->lop_declare_add == NULL)
1066 RETURN(-EOPNOTSUPP);
1068 raised = cfs_cap_raised(CFS_CAP_SYS_RESOURCE);
1070 cfs_cap_raise(CFS_CAP_SYS_RESOURCE);
1071 rc = lgh->lgh_logops->lop_declare_add(env, lgh, rec, th);
1073 cfs_cap_lower(CFS_CAP_SYS_RESOURCE);
1076 EXPORT_SYMBOL(llog_declare_add);
1079 * Helper function to open llog or create it if doesn't exist.
1080 * It hides all transaction handling from caller.
1082 int llog_open_create(const struct lu_env *env, struct llog_ctxt *ctxt,
1083 struct llog_handle **res, struct llog_logid *logid,
1086 struct dt_device *d;
1092 rc = llog_open(env, ctxt, res, logid, name, LLOG_OPEN_NEW);
1096 if (llog_exist(*res))
1099 LASSERT((*res)->lgh_obj != NULL);
1101 d = lu2dt_dev((*res)->lgh_obj->do_lu.lo_dev);
1103 th = dt_trans_create(env, d);
1105 GOTO(out, rc = PTR_ERR(th));
1107 /* Create update llog object synchronously, which
1108 * happens during inialization process see
1109 * lod_sub_prep_llog(), to make sure the update
1110 * llog object is created before corss-MDT writing
1111 * updates into the llog object */
1112 if (ctxt->loc_flags & LLOG_CTXT_FLAG_NORMAL_FID)
1115 th->th_wait_submit = 1;
1116 rc = llog_declare_create(env, *res, th);
1118 rc = dt_trans_start_local(env, d, th);
1120 rc = llog_create(env, *res, th);
1122 dt_trans_stop(env, d, th);
1125 llog_close(env, *res);
1128 EXPORT_SYMBOL(llog_open_create);
1131 * Helper function to delete existent llog.
1133 int llog_erase(const struct lu_env *env, struct llog_ctxt *ctxt,
1134 struct llog_logid *logid, char *name)
1136 struct llog_handle *handle;
1141 /* nothing to erase */
1142 if (name == NULL && logid == NULL)
1145 rc = llog_open(env, ctxt, &handle, logid, name, LLOG_OPEN_EXISTS);
1149 rc = llog_init_handle(env, handle, LLOG_F_IS_PLAIN, NULL);
1151 rc = llog_destroy(env, handle);
1153 rc2 = llog_close(env, handle);
1158 EXPORT_SYMBOL(llog_erase);
1161 * Helper function for write record in llog.
1162 * It hides all transaction handling from caller.
1163 * Valid only with local llog.
1165 int llog_write(const struct lu_env *env, struct llog_handle *loghandle,
1166 struct llog_rec_hdr *rec, int idx)
1168 struct dt_device *dt;
1176 LASSERT(loghandle->lgh_ctxt);
1177 LASSERT(loghandle->lgh_obj != NULL);
1179 dt = lu2dt_dev(loghandle->lgh_obj->do_lu.lo_dev);
1181 th = dt_trans_create(env, dt);
1183 RETURN(PTR_ERR(th));
1185 rc = llog_declare_write_rec(env, loghandle, rec, idx, th);
1187 GOTO(out_trans, rc);
1189 th->th_wait_submit = 1;
1190 rc = dt_trans_start_local(env, dt, th);
1192 GOTO(out_trans, rc);
1194 need_cookie = !(idx == LLOG_HEADER_IDX || idx == LLOG_NEXT_IDX);
1196 down_write(&loghandle->lgh_lock);
1198 struct llog_thread_info *lti = llog_info(env);
1200 /* cookie comes from llog_process_thread */
1201 rc = llog_write_rec(env, loghandle, rec, <i->lgi_cookie,
1202 rec->lrh_index, th);
1203 /* upper layer didn`t pass cookie so change rc */
1204 rc = (rc == 1 ? 0 : rc);
1206 rc = llog_write_rec(env, loghandle, rec, NULL, idx, th);
1209 up_write(&loghandle->lgh_lock);
1211 dt_trans_stop(env, dt, th);
1214 EXPORT_SYMBOL(llog_write);
1216 int llog_open(const struct lu_env *env, struct llog_ctxt *ctxt,
1217 struct llog_handle **lgh, struct llog_logid *logid,
1218 char *name, enum llog_open_param open_param)
1226 LASSERT(ctxt->loc_logops);
1228 if (ctxt->loc_logops->lop_open == NULL) {
1230 RETURN(-EOPNOTSUPP);
1233 *lgh = llog_alloc_handle();
1236 (*lgh)->lgh_ctxt = ctxt;
1237 (*lgh)->lgh_logops = ctxt->loc_logops;
1239 raised = cfs_cap_raised(CFS_CAP_SYS_RESOURCE);
1241 cfs_cap_raise(CFS_CAP_SYS_RESOURCE);
1242 rc = ctxt->loc_logops->lop_open(env, *lgh, logid, name, open_param);
1244 cfs_cap_lower(CFS_CAP_SYS_RESOURCE);
1246 llog_free_handle(*lgh);
1251 EXPORT_SYMBOL(llog_open);
1253 int llog_close(const struct lu_env *env, struct llog_handle *loghandle)
1255 struct llog_operations *lop;
1260 rc = llog_handle2ops(loghandle, &lop);
1263 if (lop->lop_close == NULL)
1264 GOTO(out, rc = -EOPNOTSUPP);
1265 rc = lop->lop_close(env, loghandle);
1267 llog_handle_put(loghandle);
1270 EXPORT_SYMBOL(llog_close);
1273 * Helper function to get the llog size in records. It is used by MGS
1274 * mostly to check that config llog exists and contains data.
1276 * \param[in] env execution environment
1277 * \param[in] ctxt llog context
1278 * \param[in] name llog name
1280 * \retval true if there are records in llog besides a header
1281 * \retval false on error or llog without records
1283 int llog_is_empty(const struct lu_env *env, struct llog_ctxt *ctxt,
1286 struct llog_handle *llh;
1289 rc = llog_open(env, ctxt, &llh, NULL, name, LLOG_OPEN_EXISTS);
1291 if (likely(rc == -ENOENT))
1296 rc = llog_init_handle(env, llh, LLOG_F_IS_PLAIN, NULL);
1298 GOTO(out_close, rc);
1299 rc = llog_get_size(llh);
1302 llog_close(env, llh);
1304 /* The header is record 1, the llog is still considered as empty
1305 * if there is only header */
1308 EXPORT_SYMBOL(llog_is_empty);
1310 int llog_copy_handler(const struct lu_env *env, struct llog_handle *llh,
1311 struct llog_rec_hdr *rec, void *data)
1313 struct llog_handle *copy_llh = data;
1315 /* Append all records */
1316 return llog_write(env, copy_llh, rec, LLOG_NEXT_IDX);
1319 /* backup plain llog */
1320 int llog_backup(const struct lu_env *env, struct obd_device *obd,
1321 struct llog_ctxt *ctxt, struct llog_ctxt *bctxt,
1322 char *name, char *backup)
1324 struct llog_handle *llh, *bllh;
1329 /* open original log */
1330 rc = llog_open(env, ctxt, &llh, NULL, name, LLOG_OPEN_EXISTS);
1332 /* the -ENOENT case is also reported to the caller
1333 * but silently so it should handle that if needed.
1336 CERROR("%s: failed to open log %s: rc = %d\n",
1337 obd->obd_name, name, rc);
1341 rc = llog_init_handle(env, llh, LLOG_F_IS_PLAIN, NULL);
1343 GOTO(out_close, rc);
1345 /* Make sure there's no old backup log */
1346 rc = llog_erase(env, bctxt, NULL, backup);
1347 if (rc < 0 && rc != -ENOENT)
1348 GOTO(out_close, rc);
1350 /* open backup log */
1351 rc = llog_open_create(env, bctxt, &bllh, NULL, backup);
1353 CERROR("%s: failed to open backup logfile %s: rc = %d\n",
1354 obd->obd_name, backup, rc);
1355 GOTO(out_close, rc);
1358 /* check that backup llog is not the same object as original one */
1359 if (llh->lgh_obj == bllh->lgh_obj) {
1360 CERROR("%s: backup llog %s to itself (%s), objects %p/%p\n",
1361 obd->obd_name, name, backup, llh->lgh_obj,
1363 GOTO(out_backup, rc = -EEXIST);
1366 rc = llog_init_handle(env, bllh, LLOG_F_IS_PLAIN, NULL);
1368 GOTO(out_backup, rc);
1370 /* Copy log record by record */
1371 rc = llog_process_or_fork(env, llh, llog_copy_handler, (void *)bllh,
1374 CERROR("%s: failed to backup log %s: rc = %d\n",
1375 obd->obd_name, name, rc);
1377 llog_close(env, bllh);
1379 llog_close(env, llh);
1382 EXPORT_SYMBOL(llog_backup);
1384 /* Get size of llog */
1385 __u64 llog_size(const struct lu_env *env, struct llog_handle *llh)
1390 rc = llh->lgh_obj->do_ops->do_attr_get(env, llh->lgh_obj, &la);
1392 CERROR("%s: attr_get failed for "DFID": rc = %d\n",
1393 llh->lgh_ctxt->loc_obd->obd_name,
1394 PFID(&llh->lgh_id.lgl_oi.oi_fid), rc);
1400 EXPORT_SYMBOL(llog_size);