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);
315 mutex_unlock(&loghandle->lgh_hdr_mutex);
316 up_write(&loghandle->lgh_lock);
318 rc1 = dt_trans_stop(env, dt, th);
322 mutex_lock(&loghandle->lgh_hdr_mutex);
324 loghandle->lgh_hdr->llh_count += num;
325 for (i = i - 1; i >= 0; i--)
326 set_bit_le(index[i], LLOG_HDR_BITMAP(llh));
327 mutex_unlock(&loghandle->lgh_hdr_mutex);
332 int llog_cancel_rec(const struct lu_env *env, struct llog_handle *loghandle,
335 return llog_cancel_arr_rec(env, loghandle, 1, &index);
338 int llog_read_header(const struct lu_env *env, struct llog_handle *handle,
339 const struct obd_uuid *uuid)
341 const struct llog_operations *lop;
345 rc = llog_handle2ops(handle, &lop);
349 if (lop->lop_read_header == NULL)
352 rc = lop->lop_read_header(env, handle);
353 if (rc == LLOG_EEMPTY) {
354 struct llog_log_hdr *llh = handle->lgh_hdr;
356 /* lrh_len should be initialized in llog_init_handle */
357 handle->lgh_last_idx = 0; /* header is record with index 0 */
358 llh->llh_count = 1; /* for the header record */
359 llh->llh_hdr.lrh_type = LLOG_HDR_MAGIC;
360 LASSERT(handle->lgh_ctxt->loc_chunk_size >=
361 LLOG_MIN_CHUNK_SIZE);
362 llh->llh_hdr.lrh_len = handle->lgh_ctxt->loc_chunk_size;
363 llh->llh_hdr.lrh_index = 0;
364 llh->llh_timestamp = ktime_get_real_seconds();
366 memcpy(&llh->llh_tgtuuid, uuid,
367 sizeof(llh->llh_tgtuuid));
368 llh->llh_bitmap_offset = offsetof(typeof(*llh), llh_bitmap);
369 /* Since update llog header might also call this function,
370 * let's reset the bitmap to 0 here */
371 memset(LLOG_HDR_BITMAP(llh), 0, llh->llh_hdr.lrh_len -
372 llh->llh_bitmap_offset -
373 sizeof(llh->llh_tail));
374 set_bit_le(0, LLOG_HDR_BITMAP(llh));
375 LLOG_HDR_TAIL(llh)->lrt_len = llh->llh_hdr.lrh_len;
376 LLOG_HDR_TAIL(llh)->lrt_index = llh->llh_hdr.lrh_index;
381 EXPORT_SYMBOL(llog_read_header);
383 int llog_init_handle(const struct lu_env *env, struct llog_handle *handle,
384 int flags, struct obd_uuid *uuid)
386 struct llog_log_hdr *llh;
387 enum llog_flag fmt = flags & LLOG_F_EXT_MASK;
389 int chunk_size = handle->lgh_ctxt->loc_chunk_size;
392 LASSERT(handle->lgh_hdr == NULL);
394 LASSERT(chunk_size >= LLOG_MIN_CHUNK_SIZE);
395 OBD_ALLOC_LARGE(llh, chunk_size);
399 handle->lgh_hdr = llh;
400 handle->lgh_hdr_size = chunk_size;
401 /* first assign flags to use llog_client_ops */
402 llh->llh_flags = flags;
403 rc = llog_read_header(env, handle, uuid);
405 if (unlikely((llh->llh_flags & LLOG_F_IS_PLAIN &&
406 flags & LLOG_F_IS_CAT) ||
407 (llh->llh_flags & LLOG_F_IS_CAT &&
408 flags & LLOG_F_IS_PLAIN))) {
409 CERROR("%s: llog type is %s but initializing %s\n",
410 loghandle2name(handle),
411 llh->llh_flags & LLOG_F_IS_CAT ?
413 flags & LLOG_F_IS_CAT ? "catalog" : "plain");
414 GOTO(out, rc = -EINVAL);
415 } else if (llh->llh_flags &
416 (LLOG_F_IS_PLAIN | LLOG_F_IS_CAT)) {
418 * it is possible to open llog without specifying llog
419 * type so it is taken from llh_flags
421 flags = llh->llh_flags;
423 /* for some reason the llh_flags has no type set */
424 CERROR("llog type is not specified!\n");
425 GOTO(out, rc = -EINVAL);
428 !obd_uuid_equals(uuid, &llh->llh_tgtuuid))) {
429 CERROR("%s: llog uuid mismatch: %s/%s\n",
430 loghandle2name(handle),
432 (char *)llh->llh_tgtuuid.uuid);
433 GOTO(out, rc = -EEXIST);
436 if (flags & LLOG_F_IS_CAT) {
437 LASSERT(list_empty(&handle->u.chd.chd_head));
438 INIT_LIST_HEAD(&handle->u.chd.chd_head);
439 llh->llh_size = sizeof(struct llog_logid_rec);
440 llh->llh_flags |= LLOG_F_IS_FIXSIZE;
441 } else if (!(flags & LLOG_F_IS_PLAIN)) {
442 CERROR("%s: unknown flags: %#x (expected %#x or %#x)\n",
443 loghandle2name(handle), flags, LLOG_F_IS_CAT,
447 llh->llh_flags |= fmt;
450 OBD_FREE_LARGE(llh, chunk_size);
451 handle->lgh_hdr = NULL;
455 EXPORT_SYMBOL(llog_init_handle);
457 int llog_verify_record(const struct llog_handle *llh, struct llog_rec_hdr *rec)
459 int chunk_size = llh->lgh_hdr->llh_hdr.lrh_len;
461 if (rec->lrh_len == 0 || rec->lrh_len > chunk_size) {
462 CERROR("%s: record is too large: %d > %d\n",
463 loghandle2name(llh), rec->lrh_len, chunk_size);
466 if (rec->lrh_index >= LLOG_HDR_BITMAP_SIZE(llh->lgh_hdr)) {
467 CERROR("%s: index is too high: %d\n",
468 loghandle2name(llh), rec->lrh_index);
471 if ((rec->lrh_type & LLOG_OP_MASK) != LLOG_OP_MAGIC) {
472 CERROR("%s: magic %x is bad\n",
473 loghandle2name(llh), rec->lrh_type);
479 EXPORT_SYMBOL(llog_verify_record);
481 static int llog_process_thread(void *arg)
483 struct llog_process_info *lpi = arg;
484 struct llog_handle *loghandle = lpi->lpi_loghandle;
485 struct llog_log_hdr *llh = loghandle->lgh_hdr;
486 struct llog_process_cat_data *cd = lpi->lpi_catdata;
487 struct llog_thread_info *lti;
491 int rc = 0, index = 1, last_index;
493 int last_called_index = 0;
494 bool repeated = false;
495 bool refresh_idx = false;
502 lti = lpi->lpi_env == NULL ? NULL : llog_info(lpi->lpi_env);
504 cur_offset = chunk_size = llh->llh_hdr.lrh_len;
505 /* expect chunk_size to be power of two */
506 LASSERT(is_power_of_2(chunk_size));
508 OBD_ALLOC_LARGE(buf, chunk_size);
510 lpi->lpi_rc = -ENOMEM;
515 last_called_index = cd->lpcd_first_idx;
516 index = cd->lpcd_first_idx + 1;
518 if (cd != NULL && cd->lpcd_last_idx)
519 last_index = cd->lpcd_last_idx;
521 last_index = LLOG_HDR_BITMAP_SIZE(llh) - 1;
524 struct llog_rec_hdr *rec;
525 off_t chunk_offset = 0;
526 unsigned int buf_offset = 0;
530 /* skip records not set in bitmap */
531 while (index <= last_index &&
532 !test_bit_le(index, LLOG_HDR_BITMAP(llh)))
535 /* There are no indices prior the last_index */
536 if (index > last_index)
539 CDEBUG(D_OTHER, "index: %d last_index %d\n", index,
543 /* get the buf with our target record; avoid old garbage */
544 memset(buf, 0, chunk_size);
545 /* the record index for outdated chunk data */
546 /* it is safe to process buffer until saved lgh_last_idx */
547 lh_last_idx = LLOG_HDR_TAIL(llh)->lrt_index;
548 rc = llog_next_block(lpi->lpi_env, loghandle, &saved_index,
549 index, &cur_offset, buf, chunk_size);
551 CDEBUG(D_OTHER, "cur_offset %llu, chunk_offset %llu,"
552 " buf_offset %u, rc = %d\n", cur_offset,
553 (__u64)chunk_offset, buf_offset, rc);
554 /* we`ve tried to reread the chunk, but there is no
556 if (rc == -EIO && repeated && (chunk_offset + buf_offset) ==
562 /* NB: after llog_next_block() call the cur_offset is the
563 * offset of the next block after read one.
564 * The absolute offset of the current chunk is calculated
565 * from cur_offset value and stored in chunk_offset variable.
567 if ((cur_offset & (chunk_size - 1)) != 0)
568 chunk_offset = cur_offset & ~(chunk_size - 1);
570 chunk_offset = cur_offset - chunk_size;
572 /* NB: when rec->lrh_len is accessed it is already swabbed
573 * since it is used at the "end" of the loop and the rec
574 * swabbing is done at the beginning of the loop. */
575 for (rec = (struct llog_rec_hdr *)(buf + buf_offset);
576 (char *)rec < buf + chunk_size;
577 rec = llog_rec_hdr_next(rec)) {
579 CDEBUG(D_OTHER, "processing rec 0x%p type %#x\n",
582 if (LLOG_REC_HDR_NEEDS_SWABBING(rec))
583 lustre_swab_llog_rec(rec);
585 CDEBUG(D_OTHER, "after swabbing, type=%#x idx=%d\n",
586 rec->lrh_type, rec->lrh_index);
588 if (index == (synced_idx + 1) &&
589 synced_idx == LLOG_HDR_TAIL(llh)->lrt_index)
592 if (OBD_FAIL_PRECHECK(OBD_FAIL_LLOG_PROCESS_TIMEOUT) &&
593 cfs_fail_val == (unsigned int)
594 (loghandle->lgh_id.lgl_oi.oi.oi_id &
596 OBD_RACE(OBD_FAIL_LLOG_PROCESS_TIMEOUT);
599 /* the bitmap could be changed during processing
600 * records from the chunk. For wrapped catalog
601 * it means we can read deleted record and try to
602 * process it. Check this case and reread the chunk.
603 * It is safe to process to lh_last_idx, including
604 * lh_last_idx if it was synced. We can not do <=
605 * comparison, cause for wrapped catalog lgh_last_idx
606 * could be less than index. So we detect last index
607 * for processing as index == lh_last_idx+1. But when
608 * catalog is wrapped and full lgh_last_idx=llh_cat_idx,
609 * the first processing index is llh_cat_idx+1.The
610 * exception is !(lgh_last_idx == llh_cat_idx &&
611 * index == llh_cat_idx + 1), and after simplification
613 * lh_last_idx != LLOG_HDR_TAIL(llh)->lrt_index
614 * This exception is working for catalog only.
617 if ((index == lh_last_idx && synced_idx != index) ||
618 (index == (lh_last_idx + 1) &&
619 lh_last_idx != LLOG_HDR_TAIL(llh)->lrt_index) ||
620 (rec->lrh_index == 0 && !repeated)) {
622 /* save offset inside buffer for the re-read */
623 buf_offset = (char *)rec - (char *)buf;
624 cur_offset = chunk_offset;
626 /* We need to be sure lgh_last_idx
627 * record was saved to disk
629 down_read(&loghandle->lgh_last_sem);
630 synced_idx = LLOG_HDR_TAIL(llh)->lrt_index;
631 up_read(&loghandle->lgh_last_sem);
632 CDEBUG(D_OTHER, "synced_idx: %d\n", synced_idx);
639 rc = llog_verify_record(loghandle, rec);
641 CERROR("%s: invalid record in llog "DFID
642 " record for index %d/%d: rc = %d\n",
643 loghandle2name(loghandle),
644 PFID(&loghandle->lgh_id.lgl_oi.oi_fid),
645 rec->lrh_index, index, rc);
647 * the block seem to be corrupted, let's try
648 * with the next one. reset rc to go to the
653 GOTO(repeat, rc = 0);
656 if (rec->lrh_index < index) {
657 CDEBUG(D_OTHER, "skipping lrh_index %d\n",
662 if (rec->lrh_index != index) {
664 * the last time we couldn't parse the block due
665 * to corruption, thus has no idea about the
666 * next index, take it from the block, once.
670 index = rec->lrh_index;
672 CERROR("%s: "DFID" Invalid record: index"
673 " %u but expected %u\n",
674 loghandle2name(loghandle),
675 PFID(&loghandle->lgh_id.lgl_oi.oi_fid),
676 rec->lrh_index, index);
677 GOTO(out, rc = -ERANGE);
682 "lrh_index: %d lrh_len: %d (%d remains)\n",
683 rec->lrh_index, rec->lrh_len,
684 (int)(buf + chunk_size - (char *)rec));
686 /* lgh_cur_offset is used only at llog_test_3 */
687 loghandle->lgh_cur_offset = (char *)rec - (char *)buf +
690 /* if set, process the callback on this record */
691 if (test_bit_le(index, LLOG_HDR_BITMAP(llh))) {
692 struct llog_cookie *lgc;
696 CDEBUG((llh->llh_flags & LLOG_F_IS_CAT ?
698 "index: %d, lh_last_idx: %d "
699 "synced_idx: %d lgh_last_idx: %d\n",
700 index, lh_last_idx, synced_idx,
701 loghandle->lgh_last_idx);
704 lgc = <i->lgi_cookie;
705 /* store lu_env for recursive calls */
706 tmp_off = lgc->lgc_offset;
707 tmp_idx = lgc->lgc_index;
709 lgc->lgc_offset = (char *)rec -
710 (char *)buf + chunk_offset;
711 lgc->lgc_index = rec->lrh_index;
713 /* using lu_env for passing record offset to
714 * llog_write through various callbacks */
715 rc = lpi->lpi_cb(lpi->lpi_env, loghandle, rec,
717 last_called_index = index;
720 lgc->lgc_offset = tmp_off;
721 lgc->lgc_index = tmp_idx;
724 if (rc == LLOG_PROC_BREAK) {
726 } else if (rc == LLOG_DEL_RECORD) {
727 rc = llog_cancel_rec(lpi->lpi_env,
733 /* some stupid callbacks directly cancel records
734 * and delete llog. Check it and stop
736 if (loghandle->lgh_hdr == NULL ||
737 loghandle->lgh_hdr->llh_count == 1)
740 /* exit if the last index is reached */
741 if (index >= last_index)
748 CDEBUG(D_HA, "stop processing %s "DOSTID":%x index %d count %d\n",
749 ((llh->llh_flags & LLOG_F_IS_CAT) ? "catalog" : "plain"),
750 POSTID(&loghandle->lgh_id.lgl_oi), loghandle->lgh_id.lgl_ogen,
751 index, llh->llh_count);
754 cd->lpcd_last_idx = last_called_index;
756 if (unlikely(rc == -EIO && loghandle->lgh_obj != NULL)) {
757 if (dt_object_remote(loghandle->lgh_obj)) {
758 /* If it is remote object, then -EIO might means
759 * disconnection or eviction, let's return -EAGAIN,
760 * so for update recovery log processing, it will
761 * retry until the umount or abort recovery, see
762 * lod_sub_recovery_thread() */
763 CERROR("%s retry remote llog process\n",
764 loghandle2name(loghandle));
767 /* something bad happened to the processing of a local
768 * llog file, probably I/O error or the log got
769 * corrupted to be able to finally release the log we
770 * discard any remaining bits in the header */
771 CERROR("%s: Local llog found corrupted #"DOSTID":%x"
772 " %s index %d count %d\n",
773 loghandle2name(loghandle),
774 POSTID(&loghandle->lgh_id.lgl_oi),
775 loghandle->lgh_id.lgl_ogen,
776 ((llh->llh_flags & LLOG_F_IS_CAT) ? "catalog" :
777 "plain"), index, llh->llh_count);
779 while (index <= last_index) {
780 if (test_bit_le(index,
781 LLOG_HDR_BITMAP(llh)) != 0)
782 llog_cancel_rec(lpi->lpi_env, loghandle,
790 OBD_FREE_LARGE(buf, chunk_size);
795 static int llog_process_thread_daemonize(void *arg)
797 struct llog_process_info *lpi = arg;
800 struct nsproxy *new_ns, *curr_ns = current->nsproxy;
802 task_lock(lpi->lpi_reftask);
803 new_ns = lpi->lpi_reftask->nsproxy;
804 if (curr_ns != new_ns) {
807 current->nsproxy = new_ns;
808 /* XXX: we should call put_nsproxy() instead of
809 * atomic_dec(&ns->count) directly. But put_nsproxy() cannot be
810 * used outside of the kernel itself, because it calls
811 * free_nsproxy() which is not exported by the kernel
812 * (defined in kernel/nsproxy.c) */
814 atomic_dec(&curr_ns->count);
816 task_unlock(lpi->lpi_reftask);
818 /* client env has no keys, tags is just 0 */
819 rc = lu_env_init(&env, LCT_LOCAL | LCT_MG_THREAD);
824 rc = llog_process_thread(arg);
828 complete(&lpi->lpi_completion);
832 int llog_process_or_fork(const struct lu_env *env,
833 struct llog_handle *loghandle,
834 llog_cb_t cb, void *data, void *catdata, bool fork)
836 struct llog_process_info *lpi;
837 struct llog_process_data *d = data;
838 struct llog_process_cat_data *cd = catdata;
839 __u32 flags = loghandle->lgh_hdr->llh_flags;
846 CERROR("cannot alloc pointer\n");
849 lpi->lpi_loghandle = loghandle;
851 lpi->lpi_cbdata = data;
852 lpi->lpi_catdata = catdata;
854 CDEBUG(D_OTHER, "Processing "DFID" flags 0x%03x startcat %d startidx %d first_idx %d last_idx %d\n",
855 PFID(&loghandle->lgh_id.lgl_oi.oi_fid), flags,
856 (flags & LLOG_F_IS_CAT) && d ? d->lpd_startcat : -1,
857 (flags & LLOG_F_IS_CAT) && d ? d->lpd_startidx : -1,
858 cd ? cd->lpcd_first_idx : -1, cd ? cd->lpcd_last_idx : -1);
860 struct task_struct *task;
862 /* The new thread can't use parent env,
863 * init the new one in llog_process_thread_daemonize. */
865 init_completion(&lpi->lpi_completion);
866 /* take reference to current, so that
867 * llog_process_thread_daemonize() can use it to switch to
868 * namespace associated with current */
869 lpi->lpi_reftask = current;
870 task = kthread_run(llog_process_thread_daemonize, lpi,
871 "llog_process_thread");
874 CERROR("%s: cannot start thread: rc = %d\n",
875 loghandle2name(loghandle), rc);
878 wait_for_completion(&lpi->lpi_completion);
881 llog_process_thread(lpi);
889 EXPORT_SYMBOL(llog_process_or_fork);
891 int llog_process(const struct lu_env *env, struct llog_handle *loghandle,
892 llog_cb_t cb, void *data, void *catdata)
895 rc = llog_process_or_fork(env, loghandle, cb, data, catdata, true);
896 return rc == LLOG_DEL_PLAIN ? 0 : rc;
898 EXPORT_SYMBOL(llog_process);
900 static inline const struct cred *llog_raise_resource(void)
902 struct cred *cred = NULL;
904 if (cap_raised(current_cap(), CAP_SYS_RESOURCE))
907 cred = prepare_creds();
911 cap_raise(cred->cap_effective, CAP_SYS_RESOURCE);
912 return override_creds(cred);
915 static inline void llog_restore_resource(const struct cred *old_cred)
918 revert_creds(old_cred);
921 int llog_reverse_process(const struct lu_env *env,
922 struct llog_handle *loghandle, llog_cb_t cb,
923 void *data, void *catdata)
925 struct llog_log_hdr *llh = loghandle->lgh_hdr;
926 struct llog_process_cat_data *cd = catdata;
928 int rc = 0, first_index = 1, index, idx;
929 __u32 chunk_size = llh->llh_hdr.lrh_len;
932 OBD_ALLOC_LARGE(buf, chunk_size);
937 first_index = cd->lpcd_first_idx + 1;
938 if (cd != NULL && cd->lpcd_last_idx)
939 index = cd->lpcd_last_idx;
941 index = LLOG_HDR_BITMAP_SIZE(llh) - 1;
944 struct llog_rec_hdr *rec;
945 struct llog_rec_tail *tail;
947 /* skip records not set in bitmap */
948 while (index >= first_index &&
949 !test_bit_le(index, LLOG_HDR_BITMAP(llh)))
952 LASSERT(index >= first_index - 1);
953 if (index == first_index - 1)
956 /* get the buf with our target record; avoid old garbage */
957 memset(buf, 0, chunk_size);
958 rc = llog_prev_block(env, loghandle, index, buf, chunk_size);
963 idx = rec->lrh_index;
964 CDEBUG(D_RPCTRACE, "index %u : idx %u\n", index, idx);
965 while (idx < index) {
966 rec = (void *)rec + rec->lrh_len;
967 if (LLOG_REC_HDR_NEEDS_SWABBING(rec))
968 lustre_swab_llog_rec(rec);
971 LASSERT(idx == index);
972 tail = (void *)rec + rec->lrh_len - sizeof(*tail);
974 /* process records in buffer, starting where we found one */
975 while ((void *)tail > buf) {
976 if (tail->lrt_index == 0)
977 GOTO(out, rc = 0); /* no more records */
979 /* if set, process the callback on this record */
980 if (test_bit_le(index, LLOG_HDR_BITMAP(llh))) {
981 rec = (void *)tail - tail->lrt_len +
984 rc = cb(env, loghandle, rec, data);
985 if (rc == LLOG_PROC_BREAK) {
987 } else if (rc == LLOG_DEL_RECORD) {
988 rc = llog_cancel_rec(env, loghandle,
995 /* previous record, still in buffer? */
997 if (index < first_index)
999 tail = (void *)tail - tail->lrt_len;
1005 OBD_FREE_LARGE(buf, chunk_size);
1008 EXPORT_SYMBOL(llog_reverse_process);
1014 * llog_open - open llog, may not exist
1015 * llog_exist - check if llog exists
1016 * llog_close - close opened llog, pair for open, frees llog_handle
1017 * llog_declare_create - declare llog creation
1018 * llog_create - create new llog on disk, need transaction handle
1019 * llog_declare_write_rec - declaration of llog write
1020 * llog_write_rec - write llog record on disk, need transaction handle
1021 * llog_declare_add - declare llog catalog record addition
1022 * llog_add - add llog record in catalog, need transaction handle
1024 int llog_exist(struct llog_handle *loghandle)
1026 const struct llog_operations *lop;
1031 rc = llog_handle2ops(loghandle, &lop);
1034 if (lop->lop_exist == NULL)
1035 RETURN(-EOPNOTSUPP);
1037 rc = lop->lop_exist(loghandle);
1040 EXPORT_SYMBOL(llog_exist);
1042 int llog_declare_create(const struct lu_env *env,
1043 struct llog_handle *loghandle, struct thandle *th)
1045 const struct cred *old_cred;
1046 const struct llog_operations *lop;
1051 rc = llog_handle2ops(loghandle, &lop);
1054 if (lop->lop_declare_create == NULL)
1055 RETURN(-EOPNOTSUPP);
1057 old_cred = llog_raise_resource();
1058 rc = lop->lop_declare_create(env, loghandle, th);
1059 llog_restore_resource(old_cred);
1063 int llog_create(const struct lu_env *env, struct llog_handle *handle,
1066 const struct cred *old_cred;
1067 const struct llog_operations *lop;
1072 rc = llog_handle2ops(handle, &lop);
1075 if (lop->lop_create == NULL)
1076 RETURN(-EOPNOTSUPP);
1078 old_cred = llog_raise_resource();
1079 rc = lop->lop_create(env, handle, th);
1080 llog_restore_resource(old_cred);
1084 int llog_declare_write_rec(const struct lu_env *env,
1085 struct llog_handle *handle,
1086 struct llog_rec_hdr *rec, int idx,
1089 const struct cred *old_cred;
1090 const struct llog_operations *lop;
1095 rc = llog_handle2ops(handle, &lop);
1099 if (lop->lop_declare_write_rec == NULL)
1100 RETURN(-EOPNOTSUPP);
1102 old_cred = llog_raise_resource();
1103 rc = lop->lop_declare_write_rec(env, handle, rec, idx, th);
1104 llog_restore_resource(old_cred);
1108 int llog_write_rec(const struct lu_env *env, struct llog_handle *handle,
1109 struct llog_rec_hdr *rec, struct llog_cookie *logcookies,
1110 int idx, struct thandle *th)
1112 const struct cred *old_cred;
1113 const struct llog_operations *lop;
1118 /* API sanity checks */
1119 if (handle == NULL) {
1120 CERROR("loghandle is missed\n");
1122 } else if (handle->lgh_obj == NULL) {
1123 CERROR("loghandle %p with NULL object\n",
1126 } else if (th == NULL) {
1127 CERROR("%s: missed transaction handle\n",
1128 loghandle2name(handle));
1130 } else if (handle->lgh_hdr == NULL) {
1131 CERROR("%s: loghandle %p with no header\n",
1132 loghandle2name(handle), handle);
1136 rc = llog_handle2ops(handle, &lop);
1140 if (lop->lop_write_rec == NULL)
1141 RETURN(-EOPNOTSUPP);
1143 buflen = rec->lrh_len;
1144 LASSERT(cfs_size_round(buflen) == buflen);
1146 old_cred = llog_raise_resource();
1147 rc = lop->lop_write_rec(env, handle, rec, logcookies, idx, th);
1148 llog_restore_resource(old_cred);
1152 int llog_add(const struct lu_env *env, struct llog_handle *lgh,
1153 struct llog_rec_hdr *rec, struct llog_cookie *logcookies,
1156 const struct cred *old_cred;
1161 if (lgh->lgh_logops->lop_add == NULL)
1162 RETURN(-EOPNOTSUPP);
1164 old_cred = llog_raise_resource();
1165 rc = lgh->lgh_logops->lop_add(env, lgh, rec, logcookies, th);
1166 llog_restore_resource(old_cred);
1169 EXPORT_SYMBOL(llog_add);
1171 int llog_declare_add(const struct lu_env *env, struct llog_handle *lgh,
1172 struct llog_rec_hdr *rec, struct thandle *th)
1174 const struct cred *old_cred;
1179 if (lgh->lgh_logops->lop_declare_add == NULL)
1180 RETURN(-EOPNOTSUPP);
1182 old_cred = llog_raise_resource();
1183 rc = lgh->lgh_logops->lop_declare_add(env, lgh, rec, th);
1184 llog_restore_resource(old_cred);
1187 EXPORT_SYMBOL(llog_declare_add);
1190 * Helper function to open llog or create it if doesn't exist.
1191 * It hides all transaction handling from caller.
1193 int llog_open_create(const struct lu_env *env, struct llog_ctxt *ctxt,
1194 struct llog_handle **res, struct llog_logid *logid,
1197 struct dt_device *d;
1203 rc = llog_open(env, ctxt, res, logid, name, LLOG_OPEN_NEW);
1207 if (llog_exist(*res))
1210 LASSERT((*res)->lgh_obj != NULL);
1212 d = lu2dt_dev((*res)->lgh_obj->do_lu.lo_dev);
1214 if (unlikely(unlikely(d->dd_rdonly)))
1217 th = dt_trans_create(env, d);
1219 GOTO(out, rc = PTR_ERR(th));
1221 /* Create update llog object synchronously, which
1222 * happens during inialization process see
1223 * lod_sub_prep_llog(), to make sure the update
1224 * llog object is created before corss-MDT writing
1225 * updates into the llog object */
1226 if (ctxt->loc_flags & LLOG_CTXT_FLAG_NORMAL_FID)
1229 th->th_wait_submit = 1;
1230 rc = llog_declare_create(env, *res, th);
1232 rc = dt_trans_start_local(env, d, th);
1234 rc = llog_create(env, *res, th);
1236 dt_trans_stop(env, d, th);
1239 llog_close(env, *res);
1242 EXPORT_SYMBOL(llog_open_create);
1245 * Helper function to delete existent llog.
1247 int llog_erase(const struct lu_env *env, struct llog_ctxt *ctxt,
1248 struct llog_logid *logid, char *name)
1250 struct llog_handle *handle;
1255 /* nothing to erase */
1256 if (name == NULL && logid == NULL)
1259 rc = llog_open(env, ctxt, &handle, logid, name, LLOG_OPEN_EXISTS);
1263 rc = llog_init_handle(env, handle, LLOG_F_IS_PLAIN, NULL);
1265 rc = llog_destroy(env, handle);
1267 rc2 = llog_close(env, handle);
1272 EXPORT_SYMBOL(llog_erase);
1275 * Helper function for write record in llog.
1276 * It hides all transaction handling from caller.
1277 * Valid only with local llog.
1279 int llog_write(const struct lu_env *env, struct llog_handle *loghandle,
1280 struct llog_rec_hdr *rec, int idx)
1282 struct dt_device *dt;
1290 LASSERT(loghandle->lgh_ctxt);
1291 LASSERT(loghandle->lgh_obj != NULL);
1293 dt = lu2dt_dev(loghandle->lgh_obj->do_lu.lo_dev);
1295 if (unlikely(unlikely(dt->dd_rdonly)))
1298 th = dt_trans_create(env, dt);
1300 RETURN(PTR_ERR(th));
1302 rc = llog_declare_write_rec(env, loghandle, rec, idx, th);
1304 GOTO(out_trans, rc);
1306 th->th_wait_submit = 1;
1307 rc = dt_trans_start_local(env, dt, th);
1309 GOTO(out_trans, rc);
1311 need_cookie = !(idx == LLOG_HEADER_IDX || idx == LLOG_NEXT_IDX);
1313 down_write(&loghandle->lgh_lock);
1315 struct llog_thread_info *lti = llog_info(env);
1317 /* cookie comes from llog_process_thread */
1318 rc = llog_write_rec(env, loghandle, rec, <i->lgi_cookie,
1319 rec->lrh_index, th);
1320 /* upper layer didn`t pass cookie so change rc */
1321 rc = (rc == 1 ? 0 : rc);
1323 rc = llog_write_rec(env, loghandle, rec, NULL, idx, th);
1326 up_write(&loghandle->lgh_lock);
1328 dt_trans_stop(env, dt, th);
1331 EXPORT_SYMBOL(llog_write);
1333 int llog_open(const struct lu_env *env, struct llog_ctxt *ctxt,
1334 struct llog_handle **lgh, struct llog_logid *logid,
1335 char *name, enum llog_open_param open_param)
1337 const struct cred *old_cred;
1343 LASSERT(ctxt->loc_logops);
1345 if (ctxt->loc_logops->lop_open == NULL) {
1347 RETURN(-EOPNOTSUPP);
1350 *lgh = llog_alloc_handle();
1353 (*lgh)->lgh_ctxt = ctxt;
1354 (*lgh)->lgh_logops = ctxt->loc_logops;
1356 old_cred = llog_raise_resource();
1357 rc = ctxt->loc_logops->lop_open(env, *lgh, logid, name, open_param);
1358 llog_restore_resource(old_cred);
1360 llog_free_handle(*lgh);
1365 EXPORT_SYMBOL(llog_open);
1367 int llog_close(const struct lu_env *env, struct llog_handle *loghandle)
1369 return llog_handle_put(env, loghandle);
1371 EXPORT_SYMBOL(llog_close);
1374 * Helper function to get the llog size in records. It is used by MGS
1375 * mostly to check that config llog exists and contains data.
1377 * \param[in] env execution environment
1378 * \param[in] ctxt llog context
1379 * \param[in] name llog name
1381 * \retval true if there are records in llog besides a header
1382 * \retval false on error or llog without records
1384 int llog_is_empty(const struct lu_env *env, struct llog_ctxt *ctxt,
1387 struct llog_handle *llh;
1390 rc = llog_open(env, ctxt, &llh, NULL, name, LLOG_OPEN_EXISTS);
1392 if (likely(rc == -ENOENT))
1397 rc = llog_init_handle(env, llh, LLOG_F_IS_PLAIN, NULL);
1399 GOTO(out_close, rc);
1400 rc = llog_get_size(llh);
1403 llog_close(env, llh);
1405 /* The header is record 1, the llog is still considered as empty
1406 * if there is only header */
1409 EXPORT_SYMBOL(llog_is_empty);
1411 int llog_copy_handler(const struct lu_env *env, struct llog_handle *llh,
1412 struct llog_rec_hdr *rec, void *data)
1414 struct llog_handle *copy_llh = data;
1416 /* Append all records */
1417 return llog_write(env, copy_llh, rec, LLOG_NEXT_IDX);
1420 /* backup plain llog */
1421 int llog_backup(const struct lu_env *env, struct obd_device *obd,
1422 struct llog_ctxt *ctxt, struct llog_ctxt *bctxt,
1423 char *name, char *backup)
1425 struct llog_handle *llh, *bllh;
1430 /* open original log */
1431 rc = llog_open(env, ctxt, &llh, NULL, name, LLOG_OPEN_EXISTS);
1433 /* the -ENOENT case is also reported to the caller
1434 * but silently so it should handle that if needed.
1437 CERROR("%s: failed to open log %s: rc = %d\n",
1438 obd->obd_name, name, rc);
1442 rc = llog_init_handle(env, llh, LLOG_F_IS_PLAIN, NULL);
1444 GOTO(out_close, rc);
1446 /* Make sure there's no old backup log */
1447 rc = llog_erase(env, bctxt, NULL, backup);
1448 if (rc < 0 && rc != -ENOENT)
1449 GOTO(out_close, rc);
1451 /* open backup log */
1452 rc = llog_open_create(env, bctxt, &bllh, NULL, backup);
1454 CERROR("%s: failed to open backup logfile %s: rc = %d\n",
1455 obd->obd_name, backup, rc);
1456 GOTO(out_close, rc);
1459 /* check that backup llog is not the same object as original one */
1460 if (llh->lgh_obj == bllh->lgh_obj) {
1461 CERROR("%s: backup llog %s to itself (%s), objects %p/%p\n",
1462 obd->obd_name, name, backup, llh->lgh_obj,
1464 GOTO(out_backup, rc = -EEXIST);
1467 rc = llog_init_handle(env, bllh, LLOG_F_IS_PLAIN, NULL);
1469 GOTO(out_backup, rc);
1471 /* Copy log record by record */
1472 rc = llog_process_or_fork(env, llh, llog_copy_handler, (void *)bllh,
1475 CERROR("%s: failed to backup log %s: rc = %d\n",
1476 obd->obd_name, name, rc);
1478 llog_close(env, bllh);
1480 llog_close(env, llh);
1483 EXPORT_SYMBOL(llog_backup);
1485 /* Get size of llog */
1486 __u64 llog_size(const struct lu_env *env, struct llog_handle *llh)
1491 rc = llh->lgh_obj->do_ops->do_attr_get(env, llh->lgh_obj, &la);
1493 CERROR("%s: attr_get failed for "DFID": rc = %d\n",
1494 loghandle2name(llh), PFID(&llh->lgh_id.lgl_oi.oi_fid),
1501 EXPORT_SYMBOL(llog_size);