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/
31 * lustre/obdclass/llog.c
33 * OST<->MDS recovery logging infrastructure.
34 * Invariants in implementation:
35 * - we do not share logs among different OST<->MDS connections, so that
36 * if an OST or MDS fails it need only look at log(s) relevant to itself
38 * Author: Andreas Dilger <adilger@clusterfs.com>
39 * Author: Alex Zhuravlev <bzzz@whamcloud.com>
40 * Author: Mikhail Pershin <tappro@whamcloud.com>
43 #define DEBUG_SUBSYSTEM S_LOG
45 #include <linux/pid_namespace.h>
46 #include <linux/kthread.h>
47 #include <llog_swab.h>
48 #include <lustre_log.h>
49 #include <obd_support.h>
50 #include <obd_class.h>
51 #include "llog_internal.h"
54 * Allocate a new log or catalog handle
55 * Used inside llog_open().
57 static struct llog_handle *llog_alloc_handle(void)
59 struct llog_handle *loghandle;
61 OBD_ALLOC_PTR(loghandle);
62 if (loghandle == NULL)
65 init_rwsem(&loghandle->lgh_lock);
66 mutex_init(&loghandle->lgh_hdr_mutex);
67 init_rwsem(&loghandle->lgh_last_sem);
68 INIT_LIST_HEAD(&loghandle->u.phd.phd_entry);
69 refcount_set(&loghandle->lgh_refcount, 1);
75 * Free llog handle and header data if exists. Used in llog_close() only
77 static void llog_free_handle(struct llog_handle *loghandle)
79 LASSERT(loghandle != NULL);
81 /* failed llog_init_handle */
82 if (loghandle->lgh_hdr == NULL)
85 if (loghandle->lgh_hdr->llh_flags & LLOG_F_IS_PLAIN)
86 LASSERT(list_empty(&loghandle->u.phd.phd_entry));
87 else if (loghandle->lgh_hdr->llh_flags & LLOG_F_IS_CAT)
88 LASSERT(list_empty(&loghandle->u.chd.chd_head));
89 OBD_FREE_LARGE(loghandle->lgh_hdr, loghandle->lgh_hdr_size);
91 OBD_FREE_PTR(loghandle);
94 struct llog_handle *llog_handle_get(struct llog_handle *loghandle)
96 if (refcount_inc_not_zero(&loghandle->lgh_refcount))
101 int llog_handle_put(const struct lu_env *env, struct llog_handle *loghandle)
105 if (refcount_dec_and_test(&loghandle->lgh_refcount)) {
106 const struct llog_operations *lop;
108 rc = llog_handle2ops(loghandle, &lop);
111 rc = lop->lop_close(env, loghandle);
115 llog_free_handle(loghandle);
120 static int llog_declare_destroy(const struct lu_env *env,
121 struct llog_handle *handle,
124 const struct llog_operations *lop;
129 rc = llog_handle2ops(handle, &lop);
132 if (lop->lop_declare_destroy == NULL)
135 rc = lop->lop_declare_destroy(env, handle, th);
140 int llog_trans_destroy(const struct lu_env *env, struct llog_handle *handle,
143 const struct llog_operations *lop;
147 rc = llog_handle2ops(handle, &lop);
150 if (lop->lop_destroy == NULL)
153 LASSERT(handle->lgh_obj != NULL);
154 if (!llog_exist(handle))
157 rc = lop->lop_destroy(env, handle, th);
162 int llog_destroy(const struct lu_env *env, struct llog_handle *handle)
164 const struct llog_operations *lop;
165 struct dt_device *dt;
171 rc = llog_handle2ops(handle, &lop);
174 if (lop->lop_destroy == NULL)
177 if (handle->lgh_obj == NULL) {
178 /* if lgh_obj == NULL, then it is from client side destroy */
179 rc = lop->lop_destroy(env, handle, NULL);
183 if (!llog_exist(handle))
186 dt = lu2dt_dev(handle->lgh_obj->do_lu.lo_dev);
188 if (unlikely(unlikely(dt->dd_rdonly)))
191 th = dt_trans_create(env, dt);
195 rc = llog_declare_destroy(env, handle, th);
199 rc = dt_trans_start_local(env, dt, th);
203 rc = lop->lop_destroy(env, handle, th);
206 dt_trans_stop(env, dt, th);
210 EXPORT_SYMBOL(llog_destroy);
212 /* returns negative on error; 0 if success; 1 if success & log destroyed */
213 int llog_cancel_arr_rec(const struct lu_env *env, struct llog_handle *loghandle,
216 struct llog_thread_info *lgi = llog_info(env);
217 struct dt_device *dt;
218 struct llog_log_hdr *llh;
223 bool subtract_count = false;
227 LASSERT(loghandle != NULL);
228 LASSERT(loghandle->lgh_ctxt != NULL);
229 LASSERT(loghandle->lgh_obj != NULL);
231 llh = loghandle->lgh_hdr;
233 CDEBUG(D_RPCTRACE, "Canceling %d records, first %d in log "DFID"\n",
234 num, index[0], PFID(&loghandle->lgh_id.lgl_oi.oi_fid));
236 dt = lu2dt_dev(loghandle->lgh_obj->do_lu.lo_dev);
238 if (unlikely(unlikely(dt->dd_rdonly)))
241 th = dt_trans_create(env, dt);
245 rc = llog_declare_write_rec(env, loghandle, &llh->llh_hdr, 0, th);
249 if ((llh->llh_flags & LLOG_F_ZAP_WHEN_EMPTY)) {
250 rc = llog_declare_destroy(env, loghandle, th);
255 th->th_wait_submit = 1;
256 rc = dt_trans_start_local(env, dt, th);
260 down_write(&loghandle->lgh_lock);
262 mutex_lock(&loghandle->lgh_hdr_mutex);
263 for (i = 0; i < num; ++i) {
265 CERROR("Can't cancel index 0 which is header\n");
266 GOTO(out_unlock, rc = -EINVAL);
268 if (!__test_and_clear_bit_le(index[i], LLOG_HDR_BITMAP(llh))) {
269 CDEBUG(D_RPCTRACE, "Catalog index %u already clear?\n",
271 GOTO(out_unlock, rc = -ENOENT);
274 loghandle->lgh_hdr->llh_count -= num;
275 subtract_count = true;
277 /* Since llog_process_thread use lgi_cookie, it`s better to save them
278 * and restore after using
280 tmp_lgc_index = lgi->lgi_cookie.lgc_index;
281 /* Pass this index to llog_osd_write_rec(), which will use the index
282 * to only update the necesary bitmap. */
283 lgi->lgi_cookie.lgc_index = index[0];
285 rc = llog_write_rec(env, loghandle, &llh->llh_hdr, (num != 1 ? NULL :
286 &lgi->lgi_cookie), LLOG_HEADER_IDX, th);
287 lgi->lgi_cookie.lgc_index = tmp_lgc_index;
290 GOTO(out_unlock, rc);
292 if ((llh->llh_flags & LLOG_F_ZAP_WHEN_EMPTY) &&
293 (llh->llh_count == 1) &&
294 ((loghandle->lgh_last_idx == LLOG_HDR_BITMAP_SIZE(llh) - 1) ||
295 (loghandle->u.phd.phd_cat_handle != NULL &&
296 loghandle->u.phd.phd_cat_handle->u.chd.chd_current_log !=
298 /* never try to destroy it again */
299 llh->llh_flags &= ~LLOG_F_ZAP_WHEN_EMPTY;
300 rc = llog_trans_destroy(env, loghandle, th);
302 /* Sigh, can not destroy the final plain llog, but
303 * the bitmap has been clearly, so the record can not
304 * be accessed anymore, let's return 0 for now, and
305 * the orphan will be handled by LFSCK. */
306 CERROR("%s: can't destroy empty llog "DFID": rc = %d\n",
307 loghandle2name(loghandle),
308 PFID(&loghandle->lgh_id.lgl_oi.oi_fid), rc);
309 GOTO(out_unlock, rc = 0);
316 /* restore bitmap while holding a mutex */
317 if (subtract_count) {
318 loghandle->lgh_hdr->llh_count += num;
319 subtract_count = false;
321 for (i = i - 1; i >= 0; i--)
322 set_bit_le(index[i], LLOG_HDR_BITMAP(llh));
324 mutex_unlock(&loghandle->lgh_hdr_mutex);
325 up_write(&loghandle->lgh_lock);
327 rc1 = dt_trans_stop(env, dt, th);
331 mutex_lock(&loghandle->lgh_hdr_mutex);
333 loghandle->lgh_hdr->llh_count += num;
334 for (i = i - 1; i >= 0; i--)
335 set_bit_le(index[i], LLOG_HDR_BITMAP(llh));
336 mutex_unlock(&loghandle->lgh_hdr_mutex);
341 int llog_cancel_rec(const struct lu_env *env, struct llog_handle *loghandle,
344 return llog_cancel_arr_rec(env, loghandle, 1, &index);
347 int llog_read_header(const struct lu_env *env, struct llog_handle *handle,
348 const struct obd_uuid *uuid)
350 const struct llog_operations *lop;
354 rc = llog_handle2ops(handle, &lop);
358 if (lop->lop_read_header == NULL)
361 rc = lop->lop_read_header(env, handle);
362 if (rc == LLOG_EEMPTY) {
363 struct llog_log_hdr *llh = handle->lgh_hdr;
365 /* lrh_len should be initialized in llog_init_handle */
366 handle->lgh_last_idx = 0; /* header is record with index 0 */
367 llh->llh_count = 1; /* for the header record */
368 llh->llh_hdr.lrh_type = LLOG_HDR_MAGIC;
369 LASSERT(handle->lgh_ctxt->loc_chunk_size >=
370 LLOG_MIN_CHUNK_SIZE);
371 llh->llh_hdr.lrh_len = handle->lgh_ctxt->loc_chunk_size;
372 llh->llh_hdr.lrh_index = 0;
373 llh->llh_timestamp = ktime_get_real_seconds();
375 memcpy(&llh->llh_tgtuuid, uuid,
376 sizeof(llh->llh_tgtuuid));
377 llh->llh_bitmap_offset = offsetof(typeof(*llh), llh_bitmap);
378 /* Since update llog header might also call this function,
379 * let's reset the bitmap to 0 here */
380 memset(LLOG_HDR_BITMAP(llh), 0, llh->llh_hdr.lrh_len -
381 llh->llh_bitmap_offset -
382 sizeof(llh->llh_tail));
383 set_bit_le(0, LLOG_HDR_BITMAP(llh));
384 LLOG_HDR_TAIL(llh)->lrt_len = llh->llh_hdr.lrh_len;
385 LLOG_HDR_TAIL(llh)->lrt_index = llh->llh_hdr.lrh_index;
390 EXPORT_SYMBOL(llog_read_header);
392 int llog_init_handle(const struct lu_env *env, struct llog_handle *handle,
393 int flags, struct obd_uuid *uuid)
395 struct llog_log_hdr *llh;
396 enum llog_flag fmt = flags & LLOG_F_EXT_MASK;
398 int chunk_size = handle->lgh_ctxt->loc_chunk_size;
401 LASSERT(handle->lgh_hdr == NULL);
403 LASSERT(chunk_size >= LLOG_MIN_CHUNK_SIZE);
404 OBD_ALLOC_LARGE(llh, chunk_size);
408 handle->lgh_hdr = llh;
409 handle->lgh_hdr_size = chunk_size;
410 /* first assign flags to use llog_client_ops */
411 llh->llh_flags = flags;
412 rc = llog_read_header(env, handle, uuid);
414 if (unlikely((llh->llh_flags & LLOG_F_IS_PLAIN &&
415 flags & LLOG_F_IS_CAT) ||
416 (llh->llh_flags & LLOG_F_IS_CAT &&
417 flags & LLOG_F_IS_PLAIN))) {
418 CERROR("%s: llog type is %s but initializing %s\n",
419 loghandle2name(handle),
420 llh->llh_flags & LLOG_F_IS_CAT ?
422 flags & LLOG_F_IS_CAT ? "catalog" : "plain");
423 GOTO(out, rc = -EINVAL);
424 } else if (llh->llh_flags &
425 (LLOG_F_IS_PLAIN | LLOG_F_IS_CAT)) {
427 * it is possible to open llog without specifying llog
428 * type so it is taken from llh_flags
430 flags = llh->llh_flags;
432 /* for some reason the llh_flags has no type set */
433 CERROR("llog type is not specified!\n");
434 GOTO(out, rc = -EINVAL);
437 !obd_uuid_equals(uuid, &llh->llh_tgtuuid))) {
438 CERROR("%s: llog uuid mismatch: %s/%s\n",
439 loghandle2name(handle),
441 (char *)llh->llh_tgtuuid.uuid);
442 GOTO(out, rc = -EEXIST);
445 if (flags & LLOG_F_IS_CAT) {
446 LASSERT(list_empty(&handle->u.chd.chd_head));
447 INIT_LIST_HEAD(&handle->u.chd.chd_head);
448 llh->llh_size = sizeof(struct llog_logid_rec);
449 llh->llh_flags |= LLOG_F_IS_FIXSIZE;
450 } else if (!(flags & LLOG_F_IS_PLAIN)) {
451 CERROR("%s: unknown flags: %#x (expected %#x or %#x)\n",
452 loghandle2name(handle), flags, LLOG_F_IS_CAT,
456 llh->llh_flags |= fmt;
459 OBD_FREE_LARGE(llh, chunk_size);
460 handle->lgh_hdr = NULL;
464 EXPORT_SYMBOL(llog_init_handle);
466 int llog_verify_record(const struct llog_handle *llh, struct llog_rec_hdr *rec)
468 int chunk_size = llh->lgh_hdr->llh_hdr.lrh_len;
470 if (rec->lrh_len == 0 || rec->lrh_len > chunk_size) {
471 CERROR("%s: record is too large: %d > %d\n",
472 loghandle2name(llh), rec->lrh_len, chunk_size);
475 if (rec->lrh_index >= LLOG_HDR_BITMAP_SIZE(llh->lgh_hdr)) {
476 CERROR("%s: index is too high: %d\n",
477 loghandle2name(llh), rec->lrh_index);
480 if ((rec->lrh_type & LLOG_OP_MASK) != LLOG_OP_MAGIC) {
481 CERROR("%s: magic %x is bad\n",
482 loghandle2name(llh), rec->lrh_type);
488 EXPORT_SYMBOL(llog_verify_record);
490 static int llog_process_thread(void *arg)
492 struct llog_process_info *lpi = arg;
493 struct llog_handle *loghandle = lpi->lpi_loghandle;
494 struct llog_log_hdr *llh = loghandle->lgh_hdr;
495 struct llog_process_cat_data *cd = lpi->lpi_catdata;
496 struct llog_thread_info *lti;
500 int rc = 0, index = 1, last_index;
502 int last_called_index = 0;
503 bool repeated = false;
504 bool refresh_idx = false;
511 lti = lpi->lpi_env == NULL ? NULL : llog_info(lpi->lpi_env);
513 cur_offset = chunk_size = llh->llh_hdr.lrh_len;
514 /* expect chunk_size to be power of two */
515 LASSERT(is_power_of_2(chunk_size));
517 OBD_ALLOC_LARGE(buf, chunk_size);
519 lpi->lpi_rc = -ENOMEM;
524 last_called_index = cd->lpcd_first_idx;
525 index = cd->lpcd_first_idx + 1;
527 if (cd != NULL && cd->lpcd_last_idx)
528 last_index = cd->lpcd_last_idx;
530 last_index = LLOG_HDR_BITMAP_SIZE(llh) - 1;
533 struct llog_rec_hdr *rec;
534 off_t chunk_offset = 0;
535 unsigned int buf_offset = 0;
539 /* skip records not set in bitmap */
540 while (index <= last_index &&
541 !test_bit_le(index, LLOG_HDR_BITMAP(llh)))
544 /* There are no indices prior the last_index */
545 if (index > last_index)
548 CDEBUG(D_OTHER, "index: %d last_index %d\n", index,
552 /* get the buf with our target record; avoid old garbage */
553 memset(buf, 0, chunk_size);
554 /* the record index for outdated chunk data */
555 /* it is safe to process buffer until saved lgh_last_idx */
556 lh_last_idx = LLOG_HDR_TAIL(llh)->lrt_index;
557 rc = llog_next_block(lpi->lpi_env, loghandle, &saved_index,
558 index, &cur_offset, buf, chunk_size);
560 CDEBUG(D_OTHER, "cur_offset %llu, chunk_offset %llu,"
561 " buf_offset %u, rc = %d\n", cur_offset,
562 (__u64)chunk_offset, buf_offset, rc);
563 /* we`ve tried to reread the chunk, but there is no
565 if (rc == -EIO && repeated && (chunk_offset + buf_offset) ==
571 /* NB: after llog_next_block() call the cur_offset is the
572 * offset of the next block after read one.
573 * The absolute offset of the current chunk is calculated
574 * from cur_offset value and stored in chunk_offset variable.
576 if ((cur_offset & (chunk_size - 1)) != 0)
577 chunk_offset = cur_offset & ~(chunk_size - 1);
579 chunk_offset = cur_offset - chunk_size;
581 /* NB: when rec->lrh_len is accessed it is already swabbed
582 * since it is used at the "end" of the loop and the rec
583 * swabbing is done at the beginning of the loop. */
584 for (rec = (struct llog_rec_hdr *)(buf + buf_offset);
585 (char *)rec < buf + chunk_size;
586 rec = llog_rec_hdr_next(rec)) {
588 CDEBUG(D_OTHER, "processing rec 0x%p type %#x\n",
591 if (LLOG_REC_HDR_NEEDS_SWABBING(rec))
592 lustre_swab_llog_rec(rec);
594 CDEBUG(D_OTHER, "after swabbing, type=%#x idx=%d\n",
595 rec->lrh_type, rec->lrh_index);
597 if (index == (synced_idx + 1) &&
598 synced_idx == LLOG_HDR_TAIL(llh)->lrt_index)
601 if (OBD_FAIL_PRECHECK(OBD_FAIL_LLOG_PROCESS_TIMEOUT) &&
602 cfs_fail_val == (unsigned int)
603 (loghandle->lgh_id.lgl_oi.oi.oi_id &
605 OBD_RACE(OBD_FAIL_LLOG_PROCESS_TIMEOUT);
608 /* the bitmap could be changed during processing
609 * records from the chunk. For wrapped catalog
610 * it means we can read deleted record and try to
611 * process it. Check this case and reread the chunk.
612 * It is safe to process to lh_last_idx, including
613 * lh_last_idx if it was synced. We can not do <=
614 * comparison, cause for wrapped catalog lgh_last_idx
615 * could be less than index. So we detect last index
616 * for processing as index == lh_last_idx+1. But when
617 * catalog is wrapped and full lgh_last_idx=llh_cat_idx,
618 * the first processing index is llh_cat_idx+1.The
619 * exception is !(lgh_last_idx == llh_cat_idx &&
620 * index == llh_cat_idx + 1), and after simplification
622 * lh_last_idx != LLOG_HDR_TAIL(llh)->lrt_index
623 * This exception is working for catalog only.
626 if ((index == lh_last_idx && synced_idx != index) ||
627 (index == (lh_last_idx + 1) &&
628 lh_last_idx != LLOG_HDR_TAIL(llh)->lrt_index) ||
629 (rec->lrh_index == 0 && !repeated)) {
631 /* save offset inside buffer for the re-read */
632 buf_offset = (char *)rec - (char *)buf;
633 cur_offset = chunk_offset;
635 /* We need to be sure lgh_last_idx
636 * record was saved to disk
638 down_read(&loghandle->lgh_last_sem);
639 synced_idx = LLOG_HDR_TAIL(llh)->lrt_index;
640 up_read(&loghandle->lgh_last_sem);
641 CDEBUG(D_OTHER, "synced_idx: %d\n", synced_idx);
648 rc = llog_verify_record(loghandle, rec);
650 CERROR("%s: invalid record in llog "DFID
651 " record for index %d/%d: rc = %d\n",
652 loghandle2name(loghandle),
653 PFID(&loghandle->lgh_id.lgl_oi.oi_fid),
654 rec->lrh_index, index, rc);
656 * the block seem to be corrupted, let's try
657 * with the next one. reset rc to go to the
662 GOTO(repeat, rc = 0);
665 if (rec->lrh_index < index) {
666 CDEBUG(D_OTHER, "skipping lrh_index %d\n",
671 if (rec->lrh_index != index) {
673 * the last time we couldn't parse the block due
674 * to corruption, thus has no idea about the
675 * next index, take it from the block, once.
679 index = rec->lrh_index;
681 CERROR("%s: "DFID" Invalid record: index"
682 " %u but expected %u\n",
683 loghandle2name(loghandle),
684 PFID(&loghandle->lgh_id.lgl_oi.oi_fid),
685 rec->lrh_index, index);
686 GOTO(out, rc = -ERANGE);
691 "lrh_index: %d lrh_len: %d (%d remains)\n",
692 rec->lrh_index, rec->lrh_len,
693 (int)(buf + chunk_size - (char *)rec));
695 /* lgh_cur_offset is used only at llog_test_3 */
696 loghandle->lgh_cur_offset = (char *)rec - (char *)buf +
699 /* if set, process the callback on this record */
700 if (test_bit_le(index, LLOG_HDR_BITMAP(llh))) {
701 struct llog_cookie *lgc;
705 CDEBUG((llh->llh_flags & LLOG_F_IS_CAT ?
707 "index: %d, lh_last_idx: %d "
708 "synced_idx: %d lgh_last_idx: %d\n",
709 index, lh_last_idx, synced_idx,
710 loghandle->lgh_last_idx);
713 lgc = <i->lgi_cookie;
714 /* store lu_env for recursive calls */
715 tmp_off = lgc->lgc_offset;
716 tmp_idx = lgc->lgc_index;
718 lgc->lgc_offset = (char *)rec -
719 (char *)buf + chunk_offset;
720 lgc->lgc_index = rec->lrh_index;
722 /* using lu_env for passing record offset to
723 * llog_write through various callbacks */
724 rc = lpi->lpi_cb(lpi->lpi_env, loghandle, rec,
726 last_called_index = index;
729 lgc->lgc_offset = tmp_off;
730 lgc->lgc_index = tmp_idx;
733 if (rc == LLOG_PROC_BREAK) {
735 } else if (rc == LLOG_DEL_RECORD) {
736 rc = llog_cancel_rec(lpi->lpi_env,
739 /* Allow parallel cancelling, ENOENT
740 * means record was canceled at another
741 * processing thread or callback
748 /* some stupid callbacks directly cancel records
749 * and delete llog. Check it and stop
751 if (loghandle->lgh_hdr == NULL ||
752 loghandle->lgh_hdr->llh_count == 1)
755 /* exit if the last index is reached */
756 if (index >= last_index)
763 CDEBUG(D_HA, "stop processing %s "DOSTID":%x index %d count %d\n",
764 ((llh->llh_flags & LLOG_F_IS_CAT) ? "catalog" : "plain"),
765 POSTID(&loghandle->lgh_id.lgl_oi), loghandle->lgh_id.lgl_ogen,
766 index, llh->llh_count);
769 cd->lpcd_last_idx = last_called_index;
771 if (unlikely(rc == -EIO && loghandle->lgh_obj != NULL)) {
772 if (dt_object_remote(loghandle->lgh_obj)) {
773 /* If it is remote object, then -EIO might means
774 * disconnection or eviction, let's return -EAGAIN,
775 * so for update recovery log processing, it will
776 * retry until the umount or abort recovery, see
777 * lod_sub_recovery_thread() */
778 CERROR("%s retry remote llog process\n",
779 loghandle2name(loghandle));
782 /* something bad happened to the processing of a local
783 * llog file, probably I/O error or the log got
784 * corrupted to be able to finally release the log we
785 * discard any remaining bits in the header */
786 CERROR("%s: Local llog found corrupted #"DOSTID":%x"
787 " %s index %d count %d\n",
788 loghandle2name(loghandle),
789 POSTID(&loghandle->lgh_id.lgl_oi),
790 loghandle->lgh_id.lgl_ogen,
791 ((llh->llh_flags & LLOG_F_IS_CAT) ? "catalog" :
792 "plain"), index, llh->llh_count);
794 while (index <= last_index) {
795 if (test_bit_le(index,
796 LLOG_HDR_BITMAP(llh)) != 0)
797 llog_cancel_rec(lpi->lpi_env, loghandle,
805 OBD_FREE_LARGE(buf, chunk_size);
810 static int llog_process_thread_daemonize(void *arg)
812 struct llog_process_info *lpi = arg;
815 struct nsproxy *new_ns, *curr_ns = current->nsproxy;
817 task_lock(lpi->lpi_reftask);
818 new_ns = lpi->lpi_reftask->nsproxy;
819 if (curr_ns != new_ns) {
822 current->nsproxy = new_ns;
823 /* XXX: we should call put_nsproxy() instead of
824 * atomic_dec(&ns->count) directly. But put_nsproxy() cannot be
825 * used outside of the kernel itself, because it calls
826 * free_nsproxy() which is not exported by the kernel
827 * (defined in kernel/nsproxy.c) */
829 atomic_dec(&curr_ns->count);
831 task_unlock(lpi->lpi_reftask);
833 /* client env has no keys, tags is just 0 */
834 rc = lu_env_init(&env, LCT_LOCAL | LCT_MG_THREAD);
839 rc = llog_process_thread(arg);
843 complete(&lpi->lpi_completion);
847 int llog_process_or_fork(const struct lu_env *env,
848 struct llog_handle *loghandle,
849 llog_cb_t cb, void *data, void *catdata, bool fork)
851 struct llog_process_info *lpi;
852 struct llog_process_data *d = data;
853 struct llog_process_cat_data *cd = catdata;
854 __u32 flags = loghandle->lgh_hdr->llh_flags;
861 CERROR("cannot alloc pointer\n");
864 lpi->lpi_loghandle = loghandle;
866 lpi->lpi_cbdata = data;
867 lpi->lpi_catdata = catdata;
869 CDEBUG(D_OTHER, "Processing "DFID" flags 0x%03x startcat %d startidx %d first_idx %d last_idx %d\n",
870 PFID(&loghandle->lgh_id.lgl_oi.oi_fid), flags,
871 (flags & LLOG_F_IS_CAT) && d ? d->lpd_startcat : -1,
872 (flags & LLOG_F_IS_CAT) && d ? d->lpd_startidx : -1,
873 cd ? cd->lpcd_first_idx : -1, cd ? cd->lpcd_last_idx : -1);
875 struct task_struct *task;
877 /* The new thread can't use parent env,
878 * init the new one in llog_process_thread_daemonize. */
880 init_completion(&lpi->lpi_completion);
881 /* take reference to current, so that
882 * llog_process_thread_daemonize() can use it to switch to
883 * namespace associated with current */
884 lpi->lpi_reftask = current;
885 task = kthread_run(llog_process_thread_daemonize, lpi,
886 "llog_process_thread");
889 CERROR("%s: cannot start thread: rc = %d\n",
890 loghandle2name(loghandle), rc);
893 wait_for_completion(&lpi->lpi_completion);
896 llog_process_thread(lpi);
904 EXPORT_SYMBOL(llog_process_or_fork);
906 int llog_process(const struct lu_env *env, struct llog_handle *loghandle,
907 llog_cb_t cb, void *data, void *catdata)
910 rc = llog_process_or_fork(env, loghandle, cb, data, catdata, true);
911 return rc == LLOG_DEL_PLAIN ? 0 : rc;
913 EXPORT_SYMBOL(llog_process);
915 static inline const struct cred *llog_raise_resource(void)
917 struct cred *cred = NULL;
919 if (cap_raised(current_cap(), CAP_SYS_RESOURCE))
922 cred = prepare_creds();
926 cap_raise(cred->cap_effective, CAP_SYS_RESOURCE);
927 return override_creds(cred);
930 static inline void llog_restore_resource(const struct cred *old_cred)
933 revert_creds(old_cred);
936 int llog_reverse_process(const struct lu_env *env,
937 struct llog_handle *loghandle, llog_cb_t cb,
938 void *data, void *catdata)
940 struct llog_log_hdr *llh = loghandle->lgh_hdr;
941 struct llog_process_cat_data *cd = catdata;
943 int rc = 0, first_index = 1, index, idx;
944 __u32 chunk_size = llh->llh_hdr.lrh_len;
947 OBD_ALLOC_LARGE(buf, chunk_size);
952 first_index = cd->lpcd_first_idx + 1;
953 if (cd != NULL && cd->lpcd_last_idx)
954 index = cd->lpcd_last_idx;
956 index = LLOG_HDR_BITMAP_SIZE(llh) - 1;
959 struct llog_rec_hdr *rec;
960 struct llog_rec_tail *tail;
962 /* skip records not set in bitmap */
963 while (index >= first_index &&
964 !test_bit_le(index, LLOG_HDR_BITMAP(llh)))
967 LASSERT(index >= first_index - 1);
968 if (index == first_index - 1)
971 /* get the buf with our target record; avoid old garbage */
972 memset(buf, 0, chunk_size);
973 rc = llog_prev_block(env, loghandle, index, buf, chunk_size);
978 idx = rec->lrh_index;
979 CDEBUG(D_RPCTRACE, "index %u : idx %u\n", index, idx);
980 while (idx < index) {
981 rec = (void *)rec + rec->lrh_len;
982 if (LLOG_REC_HDR_NEEDS_SWABBING(rec))
983 lustre_swab_llog_rec(rec);
986 LASSERT(idx == index);
987 tail = (void *)rec + rec->lrh_len - sizeof(*tail);
989 /* process records in buffer, starting where we found one */
990 while ((void *)tail > buf) {
991 if (tail->lrt_index == 0)
992 GOTO(out, rc = 0); /* no more records */
994 /* if set, process the callback on this record */
995 if (test_bit_le(index, LLOG_HDR_BITMAP(llh))) {
996 rec = (void *)tail - tail->lrt_len +
999 rc = cb(env, loghandle, rec, data);
1000 if (rc == LLOG_PROC_BREAK) {
1002 } else if (rc == LLOG_DEL_RECORD) {
1003 rc = llog_cancel_rec(env, loghandle,
1010 /* previous record, still in buffer? */
1012 if (index < first_index)
1014 tail = (void *)tail - tail->lrt_len;
1020 OBD_FREE_LARGE(buf, chunk_size);
1023 EXPORT_SYMBOL(llog_reverse_process);
1029 * llog_open - open llog, may not exist
1030 * llog_exist - check if llog exists
1031 * llog_close - close opened llog, pair for open, frees llog_handle
1032 * llog_declare_create - declare llog creation
1033 * llog_create - create new llog on disk, need transaction handle
1034 * llog_declare_write_rec - declaration of llog write
1035 * llog_write_rec - write llog record on disk, need transaction handle
1036 * llog_declare_add - declare llog catalog record addition
1037 * llog_add - add llog record in catalog, need transaction handle
1039 int llog_exist(struct llog_handle *loghandle)
1041 const struct llog_operations *lop;
1046 rc = llog_handle2ops(loghandle, &lop);
1049 if (lop->lop_exist == NULL)
1050 RETURN(-EOPNOTSUPP);
1052 rc = lop->lop_exist(loghandle);
1055 EXPORT_SYMBOL(llog_exist);
1057 int llog_declare_create(const struct lu_env *env,
1058 struct llog_handle *loghandle, struct thandle *th)
1060 const struct cred *old_cred;
1061 const struct llog_operations *lop;
1066 rc = llog_handle2ops(loghandle, &lop);
1069 if (lop->lop_declare_create == NULL)
1070 RETURN(-EOPNOTSUPP);
1072 old_cred = llog_raise_resource();
1073 rc = lop->lop_declare_create(env, loghandle, th);
1074 llog_restore_resource(old_cred);
1078 int llog_create(const struct lu_env *env, struct llog_handle *handle,
1081 const struct cred *old_cred;
1082 const struct llog_operations *lop;
1087 rc = llog_handle2ops(handle, &lop);
1090 if (lop->lop_create == NULL)
1091 RETURN(-EOPNOTSUPP);
1093 old_cred = llog_raise_resource();
1094 rc = lop->lop_create(env, handle, th);
1095 llog_restore_resource(old_cred);
1099 int llog_declare_write_rec(const struct lu_env *env,
1100 struct llog_handle *handle,
1101 struct llog_rec_hdr *rec, int idx,
1104 const struct cred *old_cred;
1105 const struct llog_operations *lop;
1110 rc = llog_handle2ops(handle, &lop);
1114 if (lop->lop_declare_write_rec == NULL)
1115 RETURN(-EOPNOTSUPP);
1117 old_cred = llog_raise_resource();
1118 rc = lop->lop_declare_write_rec(env, handle, rec, idx, th);
1119 llog_restore_resource(old_cred);
1123 int llog_write_rec(const struct lu_env *env, struct llog_handle *handle,
1124 struct llog_rec_hdr *rec, struct llog_cookie *logcookies,
1125 int idx, struct thandle *th)
1127 const struct cred *old_cred;
1128 const struct llog_operations *lop;
1133 /* API sanity checks */
1134 if (handle == NULL) {
1135 CERROR("loghandle is missed\n");
1137 } else if (handle->lgh_obj == NULL) {
1138 CERROR("loghandle %p with NULL object\n",
1141 } else if (th == NULL) {
1142 CERROR("%s: missed transaction handle\n",
1143 loghandle2name(handle));
1145 } else if (handle->lgh_hdr == NULL) {
1146 CERROR("%s: loghandle %p with no header\n",
1147 loghandle2name(handle), handle);
1151 rc = llog_handle2ops(handle, &lop);
1155 if (lop->lop_write_rec == NULL)
1156 RETURN(-EOPNOTSUPP);
1158 buflen = rec->lrh_len;
1159 LASSERT(cfs_size_round(buflen) == buflen);
1161 old_cred = llog_raise_resource();
1162 rc = lop->lop_write_rec(env, handle, rec, logcookies, idx, th);
1163 llog_restore_resource(old_cred);
1167 int llog_add(const struct lu_env *env, struct llog_handle *lgh,
1168 struct llog_rec_hdr *rec, struct llog_cookie *logcookies,
1171 const struct cred *old_cred;
1176 if (lgh->lgh_logops->lop_add == NULL)
1177 RETURN(-EOPNOTSUPP);
1179 old_cred = llog_raise_resource();
1180 rc = lgh->lgh_logops->lop_add(env, lgh, rec, logcookies, th);
1181 llog_restore_resource(old_cred);
1184 EXPORT_SYMBOL(llog_add);
1186 int llog_declare_add(const struct lu_env *env, struct llog_handle *lgh,
1187 struct llog_rec_hdr *rec, struct thandle *th)
1189 const struct cred *old_cred;
1194 if (lgh->lgh_logops->lop_declare_add == NULL)
1195 RETURN(-EOPNOTSUPP);
1197 old_cred = llog_raise_resource();
1198 rc = lgh->lgh_logops->lop_declare_add(env, lgh, rec, th);
1199 llog_restore_resource(old_cred);
1202 EXPORT_SYMBOL(llog_declare_add);
1205 * Helper function to open llog or create it if doesn't exist.
1206 * It hides all transaction handling from caller.
1208 int llog_open_create(const struct lu_env *env, struct llog_ctxt *ctxt,
1209 struct llog_handle **res, struct llog_logid *logid,
1212 struct dt_device *d;
1218 rc = llog_open(env, ctxt, res, logid, name, LLOG_OPEN_NEW);
1222 if (llog_exist(*res))
1225 LASSERT((*res)->lgh_obj != NULL);
1227 d = lu2dt_dev((*res)->lgh_obj->do_lu.lo_dev);
1229 if (unlikely(unlikely(d->dd_rdonly)))
1232 th = dt_trans_create(env, d);
1234 GOTO(out, rc = PTR_ERR(th));
1236 /* Create update llog object synchronously, which
1237 * happens during inialization process see
1238 * lod_sub_prep_llog(), to make sure the update
1239 * llog object is created before corss-MDT writing
1240 * updates into the llog object */
1241 if (ctxt->loc_flags & LLOG_CTXT_FLAG_NORMAL_FID)
1244 th->th_wait_submit = 1;
1245 rc = llog_declare_create(env, *res, th);
1247 rc = dt_trans_start_local(env, d, th);
1249 rc = llog_create(env, *res, th);
1251 dt_trans_stop(env, d, th);
1254 llog_close(env, *res);
1257 EXPORT_SYMBOL(llog_open_create);
1260 * Helper function to delete existent llog.
1262 int llog_erase(const struct lu_env *env, struct llog_ctxt *ctxt,
1263 struct llog_logid *logid, char *name)
1265 struct llog_handle *handle;
1270 /* nothing to erase */
1271 if (name == NULL && logid == NULL)
1274 rc = llog_open(env, ctxt, &handle, logid, name, LLOG_OPEN_EXISTS);
1278 rc = llog_init_handle(env, handle, LLOG_F_IS_PLAIN, NULL);
1280 rc = llog_destroy(env, handle);
1282 rc2 = llog_close(env, handle);
1287 EXPORT_SYMBOL(llog_erase);
1290 * Helper function for write record in llog.
1291 * It hides all transaction handling from caller.
1292 * Valid only with local llog.
1294 int llog_write(const struct lu_env *env, struct llog_handle *loghandle,
1295 struct llog_rec_hdr *rec, int idx)
1297 struct dt_device *dt;
1305 LASSERT(loghandle->lgh_ctxt);
1306 LASSERT(loghandle->lgh_obj != NULL);
1308 dt = lu2dt_dev(loghandle->lgh_obj->do_lu.lo_dev);
1310 if (unlikely(unlikely(dt->dd_rdonly)))
1313 th = dt_trans_create(env, dt);
1315 RETURN(PTR_ERR(th));
1317 rc = llog_declare_write_rec(env, loghandle, rec, idx, th);
1319 GOTO(out_trans, rc);
1321 th->th_wait_submit = 1;
1322 rc = dt_trans_start_local(env, dt, th);
1324 GOTO(out_trans, rc);
1326 need_cookie = !(idx == LLOG_HEADER_IDX || idx == LLOG_NEXT_IDX);
1328 down_write(&loghandle->lgh_lock);
1330 struct llog_thread_info *lti = llog_info(env);
1332 /* cookie comes from llog_process_thread */
1333 rc = llog_write_rec(env, loghandle, rec, <i->lgi_cookie,
1334 rec->lrh_index, th);
1335 /* upper layer didn`t pass cookie so change rc */
1336 rc = (rc == 1 ? 0 : rc);
1338 rc = llog_write_rec(env, loghandle, rec, NULL, idx, th);
1341 up_write(&loghandle->lgh_lock);
1343 dt_trans_stop(env, dt, th);
1346 EXPORT_SYMBOL(llog_write);
1348 int llog_open(const struct lu_env *env, struct llog_ctxt *ctxt,
1349 struct llog_handle **lgh, struct llog_logid *logid,
1350 char *name, enum llog_open_param open_param)
1352 const struct cred *old_cred;
1358 LASSERT(ctxt->loc_logops);
1360 if (ctxt->loc_logops->lop_open == NULL) {
1362 RETURN(-EOPNOTSUPP);
1365 *lgh = llog_alloc_handle();
1368 (*lgh)->lgh_ctxt = ctxt;
1369 (*lgh)->lgh_logops = ctxt->loc_logops;
1371 old_cred = llog_raise_resource();
1372 rc = ctxt->loc_logops->lop_open(env, *lgh, logid, name, open_param);
1373 llog_restore_resource(old_cred);
1375 llog_free_handle(*lgh);
1380 EXPORT_SYMBOL(llog_open);
1382 int llog_close(const struct lu_env *env, struct llog_handle *loghandle)
1384 return llog_handle_put(env, loghandle);
1386 EXPORT_SYMBOL(llog_close);
1389 * Helper function to get the llog size in records. It is used by MGS
1390 * mostly to check that config llog exists and contains data.
1392 * \param[in] env execution environment
1393 * \param[in] ctxt llog context
1394 * \param[in] name llog name
1396 * \retval true if there are records in llog besides a header
1397 * \retval false on error or llog without records
1399 int llog_is_empty(const struct lu_env *env, struct llog_ctxt *ctxt,
1402 struct llog_handle *llh;
1405 rc = llog_open(env, ctxt, &llh, NULL, name, LLOG_OPEN_EXISTS);
1407 if (likely(rc == -ENOENT))
1412 rc = llog_init_handle(env, llh, LLOG_F_IS_PLAIN, NULL);
1414 GOTO(out_close, rc);
1415 rc = llog_get_size(llh);
1418 llog_close(env, llh);
1420 /* The header is record 1, the llog is still considered as empty
1421 * if there is only header */
1424 EXPORT_SYMBOL(llog_is_empty);
1426 int llog_copy_handler(const struct lu_env *env, struct llog_handle *llh,
1427 struct llog_rec_hdr *rec, void *data)
1429 struct llog_handle *copy_llh = data;
1431 /* Append all records */
1432 return llog_write(env, copy_llh, rec, LLOG_NEXT_IDX);
1435 /* backup plain llog */
1436 int llog_backup(const struct lu_env *env, struct obd_device *obd,
1437 struct llog_ctxt *ctxt, struct llog_ctxt *bctxt,
1438 char *name, char *backup)
1440 struct llog_handle *llh, *bllh;
1445 /* open original log */
1446 rc = llog_open(env, ctxt, &llh, NULL, name, LLOG_OPEN_EXISTS);
1448 /* the -ENOENT case is also reported to the caller
1449 * but silently so it should handle that if needed.
1452 CERROR("%s: failed to open log %s: rc = %d\n",
1453 obd->obd_name, name, rc);
1457 rc = llog_init_handle(env, llh, LLOG_F_IS_PLAIN, NULL);
1459 GOTO(out_close, rc);
1461 /* Make sure there's no old backup log */
1462 rc = llog_erase(env, bctxt, NULL, backup);
1463 if (rc < 0 && rc != -ENOENT)
1464 GOTO(out_close, rc);
1466 /* open backup log */
1467 rc = llog_open_create(env, bctxt, &bllh, NULL, backup);
1469 CERROR("%s: failed to open backup logfile %s: rc = %d\n",
1470 obd->obd_name, backup, rc);
1471 GOTO(out_close, rc);
1474 /* check that backup llog is not the same object as original one */
1475 if (llh->lgh_obj == bllh->lgh_obj) {
1476 CERROR("%s: backup llog %s to itself (%s), objects %p/%p\n",
1477 obd->obd_name, name, backup, llh->lgh_obj,
1479 GOTO(out_backup, rc = -EEXIST);
1482 rc = llog_init_handle(env, bllh, LLOG_F_IS_PLAIN, NULL);
1484 GOTO(out_backup, rc);
1486 /* Copy log record by record */
1487 rc = llog_process_or_fork(env, llh, llog_copy_handler, (void *)bllh,
1490 CERROR("%s: failed to backup log %s: rc = %d\n",
1491 obd->obd_name, name, rc);
1493 llog_close(env, bllh);
1495 llog_close(env, llh);
1498 EXPORT_SYMBOL(llog_backup);
1500 /* Get size of llog */
1501 __u64 llog_size(const struct lu_env *env, struct llog_handle *llh)
1506 rc = llh->lgh_obj->do_ops->do_attr_get(env, llh->lgh_obj, &la);
1508 CERROR("%s: attr_get failed for "DFID": rc = %d\n",
1509 loghandle2name(llh), PFID(&llh->lgh_id.lgl_oi.oi_fid),
1516 EXPORT_SYMBOL(llog_size);