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;
204 bool subtract_count = false;
208 LASSERT(loghandle != NULL);
209 LASSERT(loghandle->lgh_ctxt != NULL);
210 LASSERT(loghandle->lgh_obj != NULL);
212 llh = loghandle->lgh_hdr;
214 CDEBUG(D_RPCTRACE, "Canceling %d in log "DFID"\n", index,
215 PFID(&loghandle->lgh_id.lgl_oi.oi_fid));
218 CERROR("Can't cancel index 0 which is header\n");
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);
238 th->th_wait_submit = 1;
239 rc = dt_trans_start_local(env, dt, th);
243 down_write(&loghandle->lgh_lock);
245 mutex_lock(&loghandle->lgh_hdr_mutex);
246 if (!ext2_clear_bit(index, LLOG_HDR_BITMAP(llh))) {
247 CDEBUG(D_RPCTRACE, "Catalog index %u already clear?\n", index);
248 GOTO(out_unlock, rc);
251 loghandle->lgh_hdr->llh_count--;
252 subtract_count = true;
254 /* Since llog_process_thread use lgi_cookie, it`s better to save them
255 * and restore after using
257 tmp_lgc_index = lgi->lgi_cookie.lgc_index;
258 /* Pass this index to llog_osd_write_rec(), which will use the index
259 * to only update the necesary bitmap. */
260 lgi->lgi_cookie.lgc_index = index;
262 rc = llog_write_rec(env, loghandle, &llh->llh_hdr, &lgi->lgi_cookie,
263 LLOG_HEADER_IDX, th);
264 lgi->lgi_cookie.lgc_index = tmp_lgc_index;
267 GOTO(out_unlock, rc);
269 if ((llh->llh_flags & LLOG_F_ZAP_WHEN_EMPTY) &&
270 (llh->llh_count == 1) &&
271 ((loghandle->lgh_last_idx == LLOG_HDR_BITMAP_SIZE(llh) - 1) ||
272 (loghandle->u.phd.phd_cat_handle != NULL &&
273 loghandle->u.phd.phd_cat_handle->u.chd.chd_current_log !=
275 /* never try to destroy it again */
276 llh->llh_flags &= ~LLOG_F_ZAP_WHEN_EMPTY;
277 rc = llog_trans_destroy(env, loghandle, th);
279 /* Sigh, can not destroy the final plain llog, but
280 * the bitmap has been clearly, so the record can not
281 * be accessed anymore, let's return 0 for now, and
282 * the orphan will be handled by LFSCK. */
283 CERROR("%s: can't destroy empty llog "DFID": rc = %d\n",
284 loghandle->lgh_ctxt->loc_obd->obd_name,
285 PFID(&loghandle->lgh_id.lgl_oi.oi_fid), rc);
286 GOTO(out_unlock, rc = 0);
292 mutex_unlock(&loghandle->lgh_hdr_mutex);
293 up_write(&loghandle->lgh_lock);
295 rc1 = dt_trans_stop(env, dt, th);
298 if (rc < 0 && subtract_count) {
299 mutex_lock(&loghandle->lgh_hdr_mutex);
300 loghandle->lgh_hdr->llh_count++;
301 ext2_set_bit(index, LLOG_HDR_BITMAP(llh));
302 mutex_unlock(&loghandle->lgh_hdr_mutex);
307 int llog_read_header(const struct lu_env *env, struct llog_handle *handle,
308 const struct obd_uuid *uuid)
310 struct llog_operations *lop;
314 rc = llog_handle2ops(handle, &lop);
318 if (lop->lop_read_header == NULL)
321 rc = lop->lop_read_header(env, handle);
322 if (rc == LLOG_EEMPTY) {
323 struct llog_log_hdr *llh = handle->lgh_hdr;
325 /* lrh_len should be initialized in llog_init_handle */
326 handle->lgh_last_idx = 0; /* header is record with index 0 */
327 llh->llh_count = 1; /* for the header record */
328 llh->llh_hdr.lrh_type = LLOG_HDR_MAGIC;
329 LASSERT(handle->lgh_ctxt->loc_chunk_size >=
330 LLOG_MIN_CHUNK_SIZE);
331 llh->llh_hdr.lrh_len = handle->lgh_ctxt->loc_chunk_size;
332 llh->llh_hdr.lrh_index = 0;
333 llh->llh_timestamp = ktime_get_real_seconds();
335 memcpy(&llh->llh_tgtuuid, uuid,
336 sizeof(llh->llh_tgtuuid));
337 llh->llh_bitmap_offset = offsetof(typeof(*llh), llh_bitmap);
338 /* Since update llog header might also call this function,
339 * let's reset the bitmap to 0 here */
340 memset(LLOG_HDR_BITMAP(llh), 0, llh->llh_hdr.lrh_len -
341 llh->llh_bitmap_offset -
342 sizeof(llh->llh_tail));
343 ext2_set_bit(0, LLOG_HDR_BITMAP(llh));
344 LLOG_HDR_TAIL(llh)->lrt_len = llh->llh_hdr.lrh_len;
345 LLOG_HDR_TAIL(llh)->lrt_index = llh->llh_hdr.lrh_index;
350 EXPORT_SYMBOL(llog_read_header);
352 int llog_init_handle(const struct lu_env *env, struct llog_handle *handle,
353 int flags, struct obd_uuid *uuid)
355 struct llog_log_hdr *llh;
356 enum llog_flag fmt = flags & LLOG_F_EXT_MASK;
358 int chunk_size = handle->lgh_ctxt->loc_chunk_size;
361 LASSERT(handle->lgh_hdr == NULL);
363 LASSERT(chunk_size >= LLOG_MIN_CHUNK_SIZE);
364 OBD_ALLOC_LARGE(llh, chunk_size);
368 handle->lgh_hdr = llh;
369 handle->lgh_hdr_size = chunk_size;
370 /* first assign flags to use llog_client_ops */
371 llh->llh_flags = flags;
372 rc = llog_read_header(env, handle, uuid);
374 if (unlikely((llh->llh_flags & LLOG_F_IS_PLAIN &&
375 flags & LLOG_F_IS_CAT) ||
376 (llh->llh_flags & LLOG_F_IS_CAT &&
377 flags & LLOG_F_IS_PLAIN))) {
378 CERROR("%s: llog type is %s but initializing %s\n",
379 handle->lgh_ctxt->loc_obd->obd_name,
380 llh->llh_flags & LLOG_F_IS_CAT ?
382 flags & LLOG_F_IS_CAT ? "catalog" : "plain");
383 GOTO(out, rc = -EINVAL);
384 } else if (llh->llh_flags &
385 (LLOG_F_IS_PLAIN | LLOG_F_IS_CAT)) {
387 * it is possible to open llog without specifying llog
388 * type so it is taken from llh_flags
390 flags = llh->llh_flags;
392 /* for some reason the llh_flags has no type set */
393 CERROR("llog type is not specified!\n");
394 GOTO(out, rc = -EINVAL);
397 !obd_uuid_equals(uuid, &llh->llh_tgtuuid))) {
398 CERROR("%s: llog uuid mismatch: %s/%s\n",
399 handle->lgh_ctxt->loc_obd->obd_name,
401 (char *)llh->llh_tgtuuid.uuid);
402 GOTO(out, rc = -EEXIST);
405 if (flags & LLOG_F_IS_CAT) {
406 LASSERT(list_empty(&handle->u.chd.chd_head));
407 INIT_LIST_HEAD(&handle->u.chd.chd_head);
408 llh->llh_size = sizeof(struct llog_logid_rec);
409 llh->llh_flags |= LLOG_F_IS_FIXSIZE;
410 } else if (!(flags & LLOG_F_IS_PLAIN)) {
411 CERROR("%s: unknown flags: %#x (expected %#x or %#x)\n",
412 handle->lgh_ctxt->loc_obd->obd_name,
413 flags, LLOG_F_IS_CAT, LLOG_F_IS_PLAIN);
416 llh->llh_flags |= fmt;
419 OBD_FREE_LARGE(llh, chunk_size);
420 handle->lgh_hdr = NULL;
424 EXPORT_SYMBOL(llog_init_handle);
426 static int llog_process_thread(void *arg)
428 struct llog_process_info *lpi = arg;
429 struct llog_handle *loghandle = lpi->lpi_loghandle;
430 struct llog_log_hdr *llh = loghandle->lgh_hdr;
431 struct llog_process_cat_data *cd = lpi->lpi_catdata;
432 struct llog_thread_info *lti;
436 int rc = 0, index = 1, last_index;
438 int last_called_index = 0;
439 bool repeated = false;
446 lti = lpi->lpi_env == NULL ? NULL : llog_info(lpi->lpi_env);
448 cur_offset = chunk_size = llh->llh_hdr.lrh_len;
449 /* expect chunk_size to be power of two */
450 LASSERT(is_power_of_2(chunk_size));
452 OBD_ALLOC_LARGE(buf, chunk_size);
454 lpi->lpi_rc = -ENOMEM;
459 last_called_index = cd->lpcd_first_idx;
460 index = cd->lpcd_first_idx + 1;
462 if (cd != NULL && cd->lpcd_last_idx)
463 last_index = cd->lpcd_last_idx;
465 last_index = LLOG_HDR_BITMAP_SIZE(llh) - 1;
468 struct llog_rec_hdr *rec;
469 off_t chunk_offset = 0;
470 unsigned int buf_offset = 0;
474 /* skip records not set in bitmap */
475 while (index <= last_index &&
476 !ext2_test_bit(index, LLOG_HDR_BITMAP(llh)))
479 /* There are no indices prior the last_index */
480 if (index > last_index)
483 CDEBUG(D_OTHER, "index: %d last_index %d\n", index,
487 /* get the buf with our target record; avoid old garbage */
488 memset(buf, 0, chunk_size);
489 /* the record index for outdated chunk data */
490 lh_last_idx = loghandle->lgh_last_idx + 1;
491 rc = llog_next_block(lpi->lpi_env, loghandle, &saved_index,
492 index, &cur_offset, buf, chunk_size);
494 CDEBUG(D_OTHER, "cur_offset %llu, chunk_offset %llu,"
495 " buf_offset %u, rc = %d\n", cur_offset,
496 (__u64)chunk_offset, buf_offset, rc);
497 /* we`ve tried to reread the chunk, but there is no
499 if (rc == -EIO && repeated && (chunk_offset + buf_offset) ==
505 /* NB: after llog_next_block() call the cur_offset is the
506 * offset of the next block after read one.
507 * The absolute offset of the current chunk is calculated
508 * from cur_offset value and stored in chunk_offset variable.
510 if ((cur_offset & (chunk_size - 1)) != 0) {
511 partial_chunk = true;
512 chunk_offset = cur_offset & ~(chunk_size - 1);
514 partial_chunk = false;
515 chunk_offset = cur_offset - chunk_size;
518 /* NB: when rec->lrh_len is accessed it is already swabbed
519 * since it is used at the "end" of the loop and the rec
520 * swabbing is done at the beginning of the loop. */
521 for (rec = (struct llog_rec_hdr *)(buf + buf_offset);
522 (char *)rec < buf + chunk_size;
523 rec = llog_rec_hdr_next(rec)) {
525 CDEBUG(D_OTHER, "processing rec 0x%p type %#x\n",
528 if (LLOG_REC_HDR_NEEDS_SWABBING(rec))
529 lustre_swab_llog_rec(rec);
531 CDEBUG(D_OTHER, "after swabbing, type=%#x idx=%d\n",
532 rec->lrh_type, rec->lrh_index);
534 /* for partial chunk the end of it is zeroed, check
535 * for index 0 to distinguish it. */
536 if (partial_chunk && rec->lrh_index == 0) {
537 /* concurrent llog_add() might add new records
538 * while llog_processing, check this is not
539 * the case and re-read the current chunk
542 /* lgh_last_idx could be less then index
543 * for catalog, if catalog is wrapped */
544 if ((index > loghandle->lgh_last_idx &&
545 !(loghandle->lgh_hdr->llh_flags &
546 LLOG_F_IS_CAT)) || repeated ||
547 (loghandle->lgh_obj != NULL &&
548 dt_object_remote(loghandle->lgh_obj)))
550 /* <2 records means no more records
551 * if the last record we processed was
552 * the final one, then the underlying
553 * object might have been destroyed yet.
554 * we better don't access that.. */
555 mutex_lock(&loghandle->lgh_hdr_mutex);
556 records = loghandle->lgh_hdr->llh_count;
557 mutex_unlock(&loghandle->lgh_hdr_mutex);
560 CDEBUG(D_OTHER, "Re-read last llog buffer for "
561 "new records, index %u, last %u\n",
562 index, loghandle->lgh_last_idx);
563 /* save offset inside buffer for the re-read */
564 buf_offset = (char *)rec - (char *)buf;
565 cur_offset = chunk_offset;
572 if (rec->lrh_len == 0 || rec->lrh_len > chunk_size) {
573 CWARN("%s: invalid length %d in llog "DFID
574 "record for index %d/%d\n",
575 loghandle->lgh_ctxt->loc_obd->obd_name,
577 PFID(&loghandle->lgh_id.lgl_oi.oi_fid),
578 rec->lrh_index, index);
580 GOTO(out, rc = -EINVAL);
583 if (rec->lrh_index < index) {
584 CDEBUG(D_OTHER, "skipping lrh_index %d\n",
589 if (rec->lrh_index != index) {
590 CERROR("%s: "DFID" Invalid record: index %u"
591 " but expected %u\n",
592 loghandle->lgh_ctxt->loc_obd->obd_name,
593 PFID(&loghandle->lgh_id.lgl_oi.oi_fid),
594 rec->lrh_index, index);
595 GOTO(out, rc = -ERANGE);
599 "lrh_index: %d lrh_len: %d (%d remains)\n",
600 rec->lrh_index, rec->lrh_len,
601 (int)(buf + chunk_size - (char *)rec));
603 /* lgh_cur_offset is used only at llog_test_3 */
604 loghandle->lgh_cur_offset = (char *)rec - (char *)buf +
607 /* if set, process the callback on this record */
608 if (ext2_test_bit(index, LLOG_HDR_BITMAP(llh))) {
609 struct llog_cookie *lgc;
612 /* the bitmap could be changed during processing
613 * records from the chunk. For wrapped catalog
614 * it means we can read deleted record and try to
615 * process it. Check this case and reread the chunk.
616 * Checking the race with llog_add the bit is set
617 * after incrementation of lgh_last_idx */
618 if (index == lh_last_idx &&
620 (loghandle->lgh_last_idx + 1)) {
621 /* save offset inside buffer for
623 buf_offset = (char *)rec - (char *)buf;
624 cur_offset = chunk_offset;
630 lgc = <i->lgi_cookie;
631 /* store lu_env for recursive calls */
632 tmp_off = lgc->lgc_offset;
633 tmp_idx = lgc->lgc_index;
635 lgc->lgc_offset = (char *)rec -
636 (char *)buf + chunk_offset;
637 lgc->lgc_index = rec->lrh_index;
639 /* using lu_env for passing record offset to
640 * llog_write through various callbacks */
641 rc = lpi->lpi_cb(lpi->lpi_env, loghandle, rec,
643 last_called_index = index;
646 lgc->lgc_offset = tmp_off;
647 lgc->lgc_index = tmp_idx;
650 if (rc == LLOG_PROC_BREAK) {
652 } else if (rc == LLOG_DEL_RECORD) {
653 rc = llog_cancel_rec(lpi->lpi_env,
659 /* some stupid callbacks directly cancel records
660 * and delete llog. Check it and stop
662 if (loghandle->lgh_hdr == NULL ||
663 loghandle->lgh_hdr->llh_count == 1)
666 /* exit if the last index is reached */
667 if (index >= last_index)
675 cd->lpcd_last_idx = last_called_index;
677 if (unlikely(rc == -EIO && loghandle->lgh_obj != NULL)) {
678 if (dt_object_remote(loghandle->lgh_obj)) {
679 /* If it is remote object, then -EIO might means
680 * disconnection or eviction, let's return -EAGAIN,
681 * so for update recovery log processing, it will
682 * retry until the umount or abort recovery, see
683 * lod_sub_recovery_thread() */
684 CERROR("%s retry remote llog process\n",
685 loghandle->lgh_ctxt->loc_obd->obd_name);
688 /* something bad happened to the processing of a local
689 * llog file, probably I/O error or the log got
690 * corrupted to be able to finally release the log we
691 * discard any remaining bits in the header */
692 CERROR("%s: Local llog found corrupted #"DOSTID":%x"
693 " %s index %d count %d\n",
694 loghandle->lgh_ctxt->loc_obd->obd_name,
695 POSTID(&loghandle->lgh_id.lgl_oi),
696 loghandle->lgh_id.lgl_ogen,
697 ((llh->llh_flags & LLOG_F_IS_CAT) ? "catalog" :
698 "plain"), index, llh->llh_count);
700 while (index <= last_index) {
701 if (ext2_test_bit(index,
702 LLOG_HDR_BITMAP(llh)) != 0)
703 llog_cancel_rec(lpi->lpi_env, loghandle,
711 OBD_FREE_LARGE(buf, chunk_size);
716 static int llog_process_thread_daemonize(void *arg)
718 struct llog_process_info *lpi = arg;
721 struct nsproxy *new_ns, *curr_ns = current->nsproxy;
723 task_lock(lpi->lpi_reftask);
724 new_ns = lpi->lpi_reftask->nsproxy;
725 if (curr_ns != new_ns) {
728 current->nsproxy = new_ns;
729 /* XXX: we should call put_nsproxy() instead of
730 * atomic_dec(&ns->count) directly. But put_nsproxy() cannot be
731 * used outside of the kernel itself, because it calls
732 * free_nsproxy() which is not exported by the kernel
733 * (defined in kernel/nsproxy.c) */
734 atomic_dec(&curr_ns->count);
736 task_unlock(lpi->lpi_reftask);
740 /* client env has no keys, tags is just 0 */
741 rc = lu_env_init(&env, LCT_LOCAL | LCT_MG_THREAD);
746 rc = llog_process_thread(arg);
750 complete(&lpi->lpi_completion);
754 int llog_process_or_fork(const struct lu_env *env,
755 struct llog_handle *loghandle,
756 llog_cb_t cb, void *data, void *catdata, bool fork)
758 struct llog_process_info *lpi;
765 CERROR("cannot alloc pointer\n");
768 lpi->lpi_loghandle = loghandle;
770 lpi->lpi_cbdata = data;
771 lpi->lpi_catdata = catdata;
774 struct task_struct *task;
776 /* The new thread can't use parent env,
777 * init the new one in llog_process_thread_daemonize. */
779 init_completion(&lpi->lpi_completion);
780 /* take reference to current, so that
781 * llog_process_thread_daemonize() can use it to switch to
782 * namespace associated with current */
783 lpi->lpi_reftask = current;
784 task = kthread_run(llog_process_thread_daemonize, lpi,
785 "llog_process_thread");
788 CERROR("%s: cannot start thread: rc = %d\n",
789 loghandle->lgh_ctxt->loc_obd->obd_name, rc);
792 wait_for_completion(&lpi->lpi_completion);
795 llog_process_thread(lpi);
803 EXPORT_SYMBOL(llog_process_or_fork);
805 int llog_process(const struct lu_env *env, struct llog_handle *loghandle,
806 llog_cb_t cb, void *data, void *catdata)
809 rc = llog_process_or_fork(env, loghandle, cb, data, catdata, true);
810 return rc == LLOG_DEL_PLAIN ? 0 : rc;
812 EXPORT_SYMBOL(llog_process);
814 int llog_reverse_process(const struct lu_env *env,
815 struct llog_handle *loghandle, llog_cb_t cb,
816 void *data, void *catdata)
818 struct llog_log_hdr *llh = loghandle->lgh_hdr;
819 struct llog_process_cat_data *cd = catdata;
821 int rc = 0, first_index = 1, index, idx;
822 __u32 chunk_size = llh->llh_hdr.lrh_len;
825 OBD_ALLOC_LARGE(buf, chunk_size);
830 first_index = cd->lpcd_first_idx + 1;
831 if (cd != NULL && cd->lpcd_last_idx)
832 index = cd->lpcd_last_idx;
834 index = LLOG_HDR_BITMAP_SIZE(llh) - 1;
837 struct llog_rec_hdr *rec;
838 struct llog_rec_tail *tail;
840 /* skip records not set in bitmap */
841 while (index >= first_index &&
842 !ext2_test_bit(index, LLOG_HDR_BITMAP(llh)))
845 LASSERT(index >= first_index - 1);
846 if (index == first_index - 1)
849 /* get the buf with our target record; avoid old garbage */
850 memset(buf, 0, chunk_size);
851 rc = llog_prev_block(env, loghandle, index, buf, chunk_size);
856 idx = rec->lrh_index;
857 CDEBUG(D_RPCTRACE, "index %u : idx %u\n", index, idx);
858 while (idx < index) {
859 rec = (void *)rec + rec->lrh_len;
860 if (LLOG_REC_HDR_NEEDS_SWABBING(rec))
861 lustre_swab_llog_rec(rec);
864 LASSERT(idx == index);
865 tail = (void *)rec + rec->lrh_len - sizeof(*tail);
867 /* process records in buffer, starting where we found one */
868 while ((void *)tail > buf) {
869 if (tail->lrt_index == 0)
870 GOTO(out, rc = 0); /* no more records */
872 /* if set, process the callback on this record */
873 if (ext2_test_bit(index, LLOG_HDR_BITMAP(llh))) {
874 rec = (void *)tail - tail->lrt_len +
877 rc = cb(env, loghandle, rec, data);
878 if (rc == LLOG_PROC_BREAK) {
880 } else if (rc == LLOG_DEL_RECORD) {
881 rc = llog_cancel_rec(env, loghandle,
888 /* previous record, still in buffer? */
890 if (index < first_index)
892 tail = (void *)tail - tail->lrt_len;
898 OBD_FREE_LARGE(buf, chunk_size);
901 EXPORT_SYMBOL(llog_reverse_process);
907 * llog_open - open llog, may not exist
908 * llog_exist - check if llog exists
909 * llog_close - close opened llog, pair for open, frees llog_handle
910 * llog_declare_create - declare llog creation
911 * llog_create - create new llog on disk, need transaction handle
912 * llog_declare_write_rec - declaration of llog write
913 * llog_write_rec - write llog record on disk, need transaction handle
914 * llog_declare_add - declare llog catalog record addition
915 * llog_add - add llog record in catalog, need transaction handle
917 int llog_exist(struct llog_handle *loghandle)
919 struct llog_operations *lop;
924 rc = llog_handle2ops(loghandle, &lop);
927 if (lop->lop_exist == NULL)
930 rc = lop->lop_exist(loghandle);
933 EXPORT_SYMBOL(llog_exist);
935 int llog_declare_create(const struct lu_env *env,
936 struct llog_handle *loghandle, struct thandle *th)
938 struct llog_operations *lop;
943 rc = llog_handle2ops(loghandle, &lop);
946 if (lop->lop_declare_create == NULL)
949 raised = cfs_cap_raised(CFS_CAP_SYS_RESOURCE);
951 cfs_cap_raise(CFS_CAP_SYS_RESOURCE);
952 rc = lop->lop_declare_create(env, loghandle, th);
954 cfs_cap_lower(CFS_CAP_SYS_RESOURCE);
958 int llog_create(const struct lu_env *env, struct llog_handle *handle,
961 struct llog_operations *lop;
966 rc = llog_handle2ops(handle, &lop);
969 if (lop->lop_create == NULL)
972 raised = cfs_cap_raised(CFS_CAP_SYS_RESOURCE);
974 cfs_cap_raise(CFS_CAP_SYS_RESOURCE);
975 rc = lop->lop_create(env, handle, th);
977 cfs_cap_lower(CFS_CAP_SYS_RESOURCE);
981 int llog_declare_write_rec(const struct lu_env *env,
982 struct llog_handle *handle,
983 struct llog_rec_hdr *rec, int idx,
986 struct llog_operations *lop;
991 rc = llog_handle2ops(handle, &lop);
995 if (lop->lop_declare_write_rec == NULL)
998 raised = cfs_cap_raised(CFS_CAP_SYS_RESOURCE);
1000 cfs_cap_raise(CFS_CAP_SYS_RESOURCE);
1001 rc = lop->lop_declare_write_rec(env, handle, rec, idx, th);
1003 cfs_cap_lower(CFS_CAP_SYS_RESOURCE);
1007 int llog_write_rec(const struct lu_env *env, struct llog_handle *handle,
1008 struct llog_rec_hdr *rec, struct llog_cookie *logcookies,
1009 int idx, struct thandle *th)
1011 struct llog_operations *lop;
1012 int raised, rc, buflen;
1016 /* API sanity checks */
1017 if (handle == NULL) {
1018 CERROR("loghandle is missed\n");
1020 } else if (handle->lgh_obj == NULL) {
1021 CERROR("loghandle %p with NULL object\n",
1024 } else if (th == NULL) {
1025 CERROR("%s: missed transaction handle\n",
1026 handle->lgh_obj->do_lu.lo_dev->ld_obd->obd_name);
1028 } else if (handle->lgh_hdr == NULL) {
1029 CERROR("%s: loghandle %p with no header\n",
1030 handle->lgh_obj->do_lu.lo_dev->ld_obd->obd_name,
1035 rc = llog_handle2ops(handle, &lop);
1039 if (lop->lop_write_rec == NULL)
1040 RETURN(-EOPNOTSUPP);
1042 buflen = rec->lrh_len;
1043 LASSERT(cfs_size_round(buflen) == buflen);
1045 raised = cfs_cap_raised(CFS_CAP_SYS_RESOURCE);
1047 cfs_cap_raise(CFS_CAP_SYS_RESOURCE);
1048 rc = lop->lop_write_rec(env, handle, rec, logcookies, idx, th);
1050 cfs_cap_lower(CFS_CAP_SYS_RESOURCE);
1054 int llog_add(const struct lu_env *env, struct llog_handle *lgh,
1055 struct llog_rec_hdr *rec, struct llog_cookie *logcookies,
1062 if (lgh->lgh_logops->lop_add == NULL)
1063 RETURN(-EOPNOTSUPP);
1065 raised = cfs_cap_raised(CFS_CAP_SYS_RESOURCE);
1067 cfs_cap_raise(CFS_CAP_SYS_RESOURCE);
1068 rc = lgh->lgh_logops->lop_add(env, lgh, rec, logcookies, th);
1070 cfs_cap_lower(CFS_CAP_SYS_RESOURCE);
1073 EXPORT_SYMBOL(llog_add);
1075 int llog_declare_add(const struct lu_env *env, struct llog_handle *lgh,
1076 struct llog_rec_hdr *rec, struct thandle *th)
1082 if (lgh->lgh_logops->lop_declare_add == NULL)
1083 RETURN(-EOPNOTSUPP);
1085 raised = cfs_cap_raised(CFS_CAP_SYS_RESOURCE);
1087 cfs_cap_raise(CFS_CAP_SYS_RESOURCE);
1088 rc = lgh->lgh_logops->lop_declare_add(env, lgh, rec, th);
1090 cfs_cap_lower(CFS_CAP_SYS_RESOURCE);
1093 EXPORT_SYMBOL(llog_declare_add);
1096 * Helper function to open llog or create it if doesn't exist.
1097 * It hides all transaction handling from caller.
1099 int llog_open_create(const struct lu_env *env, struct llog_ctxt *ctxt,
1100 struct llog_handle **res, struct llog_logid *logid,
1103 struct dt_device *d;
1109 rc = llog_open(env, ctxt, res, logid, name, LLOG_OPEN_NEW);
1113 if (llog_exist(*res))
1116 LASSERT((*res)->lgh_obj != NULL);
1118 d = lu2dt_dev((*res)->lgh_obj->do_lu.lo_dev);
1120 th = dt_trans_create(env, d);
1122 GOTO(out, rc = PTR_ERR(th));
1124 /* Create update llog object synchronously, which
1125 * happens during inialization process see
1126 * lod_sub_prep_llog(), to make sure the update
1127 * llog object is created before corss-MDT writing
1128 * updates into the llog object */
1129 if (ctxt->loc_flags & LLOG_CTXT_FLAG_NORMAL_FID)
1132 th->th_wait_submit = 1;
1133 rc = llog_declare_create(env, *res, th);
1135 rc = dt_trans_start_local(env, d, th);
1137 rc = llog_create(env, *res, th);
1139 dt_trans_stop(env, d, th);
1142 llog_close(env, *res);
1145 EXPORT_SYMBOL(llog_open_create);
1148 * Helper function to delete existent llog.
1150 int llog_erase(const struct lu_env *env, struct llog_ctxt *ctxt,
1151 struct llog_logid *logid, char *name)
1153 struct llog_handle *handle;
1158 /* nothing to erase */
1159 if (name == NULL && logid == NULL)
1162 rc = llog_open(env, ctxt, &handle, logid, name, LLOG_OPEN_EXISTS);
1166 rc = llog_init_handle(env, handle, LLOG_F_IS_PLAIN, NULL);
1168 rc = llog_destroy(env, handle);
1170 rc2 = llog_close(env, handle);
1175 EXPORT_SYMBOL(llog_erase);
1178 * Helper function for write record in llog.
1179 * It hides all transaction handling from caller.
1180 * Valid only with local llog.
1182 int llog_write(const struct lu_env *env, struct llog_handle *loghandle,
1183 struct llog_rec_hdr *rec, int idx)
1185 struct dt_device *dt;
1193 LASSERT(loghandle->lgh_ctxt);
1194 LASSERT(loghandle->lgh_obj != NULL);
1196 dt = lu2dt_dev(loghandle->lgh_obj->do_lu.lo_dev);
1198 th = dt_trans_create(env, dt);
1200 RETURN(PTR_ERR(th));
1202 rc = llog_declare_write_rec(env, loghandle, rec, idx, th);
1204 GOTO(out_trans, rc);
1206 th->th_wait_submit = 1;
1207 rc = dt_trans_start_local(env, dt, th);
1209 GOTO(out_trans, rc);
1211 need_cookie = !(idx == LLOG_HEADER_IDX || idx == LLOG_NEXT_IDX);
1213 down_write(&loghandle->lgh_lock);
1215 struct llog_thread_info *lti = llog_info(env);
1217 /* cookie comes from llog_process_thread */
1218 rc = llog_write_rec(env, loghandle, rec, <i->lgi_cookie,
1219 rec->lrh_index, th);
1220 /* upper layer didn`t pass cookie so change rc */
1221 rc = (rc == 1 ? 0 : rc);
1223 rc = llog_write_rec(env, loghandle, rec, NULL, idx, th);
1226 up_write(&loghandle->lgh_lock);
1228 dt_trans_stop(env, dt, th);
1231 EXPORT_SYMBOL(llog_write);
1233 int llog_open(const struct lu_env *env, struct llog_ctxt *ctxt,
1234 struct llog_handle **lgh, struct llog_logid *logid,
1235 char *name, enum llog_open_param open_param)
1243 LASSERT(ctxt->loc_logops);
1245 if (ctxt->loc_logops->lop_open == NULL) {
1247 RETURN(-EOPNOTSUPP);
1250 *lgh = llog_alloc_handle();
1253 (*lgh)->lgh_ctxt = ctxt;
1254 (*lgh)->lgh_logops = ctxt->loc_logops;
1256 raised = cfs_cap_raised(CFS_CAP_SYS_RESOURCE);
1258 cfs_cap_raise(CFS_CAP_SYS_RESOURCE);
1259 rc = ctxt->loc_logops->lop_open(env, *lgh, logid, name, open_param);
1261 cfs_cap_lower(CFS_CAP_SYS_RESOURCE);
1263 llog_free_handle(*lgh);
1268 EXPORT_SYMBOL(llog_open);
1270 int llog_close(const struct lu_env *env, struct llog_handle *loghandle)
1272 struct llog_operations *lop;
1277 rc = llog_handle2ops(loghandle, &lop);
1280 if (lop->lop_close == NULL)
1281 GOTO(out, rc = -EOPNOTSUPP);
1282 rc = lop->lop_close(env, loghandle);
1284 llog_handle_put(loghandle);
1287 EXPORT_SYMBOL(llog_close);
1290 * Helper function to get the llog size in records. It is used by MGS
1291 * mostly to check that config llog exists and contains data.
1293 * \param[in] env execution environment
1294 * \param[in] ctxt llog context
1295 * \param[in] name llog name
1297 * \retval true if there are records in llog besides a header
1298 * \retval false on error or llog without records
1300 int llog_is_empty(const struct lu_env *env, struct llog_ctxt *ctxt,
1303 struct llog_handle *llh;
1306 rc = llog_open(env, ctxt, &llh, NULL, name, LLOG_OPEN_EXISTS);
1308 if (likely(rc == -ENOENT))
1313 rc = llog_init_handle(env, llh, LLOG_F_IS_PLAIN, NULL);
1315 GOTO(out_close, rc);
1316 rc = llog_get_size(llh);
1319 llog_close(env, llh);
1321 /* The header is record 1, the llog is still considered as empty
1322 * if there is only header */
1325 EXPORT_SYMBOL(llog_is_empty);
1327 int llog_copy_handler(const struct lu_env *env, struct llog_handle *llh,
1328 struct llog_rec_hdr *rec, void *data)
1330 struct llog_handle *copy_llh = data;
1332 /* Append all records */
1333 return llog_write(env, copy_llh, rec, LLOG_NEXT_IDX);
1336 /* backup plain llog */
1337 int llog_backup(const struct lu_env *env, struct obd_device *obd,
1338 struct llog_ctxt *ctxt, struct llog_ctxt *bctxt,
1339 char *name, char *backup)
1341 struct llog_handle *llh, *bllh;
1346 /* open original log */
1347 rc = llog_open(env, ctxt, &llh, NULL, name, LLOG_OPEN_EXISTS);
1349 /* the -ENOENT case is also reported to the caller
1350 * but silently so it should handle that if needed.
1353 CERROR("%s: failed to open log %s: rc = %d\n",
1354 obd->obd_name, name, rc);
1358 rc = llog_init_handle(env, llh, LLOG_F_IS_PLAIN, NULL);
1360 GOTO(out_close, rc);
1362 /* Make sure there's no old backup log */
1363 rc = llog_erase(env, bctxt, NULL, backup);
1364 if (rc < 0 && rc != -ENOENT)
1365 GOTO(out_close, rc);
1367 /* open backup log */
1368 rc = llog_open_create(env, bctxt, &bllh, NULL, backup);
1370 CERROR("%s: failed to open backup logfile %s: rc = %d\n",
1371 obd->obd_name, backup, rc);
1372 GOTO(out_close, rc);
1375 /* check that backup llog is not the same object as original one */
1376 if (llh->lgh_obj == bllh->lgh_obj) {
1377 CERROR("%s: backup llog %s to itself (%s), objects %p/%p\n",
1378 obd->obd_name, name, backup, llh->lgh_obj,
1380 GOTO(out_backup, rc = -EEXIST);
1383 rc = llog_init_handle(env, bllh, LLOG_F_IS_PLAIN, NULL);
1385 GOTO(out_backup, rc);
1387 /* Copy log record by record */
1388 rc = llog_process_or_fork(env, llh, llog_copy_handler, (void *)bllh,
1391 CERROR("%s: failed to backup log %s: rc = %d\n",
1392 obd->obd_name, name, rc);
1394 llog_close(env, bllh);
1396 llog_close(env, llh);
1399 EXPORT_SYMBOL(llog_backup);
1401 /* Get size of llog */
1402 __u64 llog_size(const struct lu_env *env, struct llog_handle *llh)
1407 rc = llh->lgh_obj->do_ops->do_attr_get(env, llh->lgh_obj, &la);
1409 CERROR("%s: attr_get failed for "DFID": rc = %d\n",
1410 llh->lgh_ctxt->loc_obd->obd_name,
1411 PFID(&llh->lgh_id.lgl_oi.oi_fid), rc);
1417 EXPORT_SYMBOL(llog_size);