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_support.h>
51 #include <obd_class.h>
52 #include "llog_internal.h"
55 * Allocate a new log or catalog handle
56 * Used inside llog_open().
58 static struct llog_handle *llog_alloc_handle(void)
60 struct llog_handle *loghandle;
62 OBD_ALLOC_PTR(loghandle);
63 if (loghandle == NULL)
66 init_rwsem(&loghandle->lgh_lock);
67 mutex_init(&loghandle->lgh_hdr_mutex);
68 init_rwsem(&loghandle->lgh_last_sem);
69 INIT_LIST_HEAD(&loghandle->u.phd.phd_entry);
70 refcount_set(&loghandle->lgh_refcount, 1);
76 * Free llog handle and header data if exists. Used in llog_close() only
78 static void llog_free_handle(struct llog_handle *loghandle)
80 LASSERT(loghandle != NULL);
82 /* failed llog_init_handle */
83 if (loghandle->lgh_hdr == NULL)
86 if (loghandle->lgh_hdr->llh_flags & LLOG_F_IS_PLAIN)
87 LASSERT(list_empty(&loghandle->u.phd.phd_entry));
88 else if (loghandle->lgh_hdr->llh_flags & LLOG_F_IS_CAT)
89 LASSERT(list_empty(&loghandle->u.chd.chd_head));
90 OBD_FREE_LARGE(loghandle->lgh_hdr, loghandle->lgh_hdr_size);
92 OBD_FREE_PTR(loghandle);
95 struct llog_handle *llog_handle_get(struct llog_handle *loghandle)
97 if (refcount_inc_not_zero(&loghandle->lgh_refcount))
102 int llog_handle_put(const struct lu_env *env, struct llog_handle *loghandle)
106 if (refcount_dec_and_test(&loghandle->lgh_refcount)) {
107 const struct llog_operations *lop;
109 rc = llog_handle2ops(loghandle, &lop);
112 rc = lop->lop_close(env, loghandle);
116 llog_free_handle(loghandle);
121 static int llog_declare_destroy(const struct lu_env *env,
122 struct llog_handle *handle,
125 const struct llog_operations *lop;
130 rc = llog_handle2ops(handle, &lop);
133 if (lop->lop_declare_destroy == NULL)
136 rc = lop->lop_declare_destroy(env, handle, th);
141 int llog_trans_destroy(const struct lu_env *env, struct llog_handle *handle,
144 const struct llog_operations *lop;
148 rc = llog_handle2ops(handle, &lop);
151 if (lop->lop_destroy == NULL)
154 LASSERT(handle->lgh_obj != NULL);
155 if (!llog_exist(handle))
158 rc = lop->lop_destroy(env, handle, th);
163 int llog_destroy(const struct lu_env *env, struct llog_handle *handle)
165 const struct llog_operations *lop;
166 struct dt_device *dt;
172 rc = llog_handle2ops(handle, &lop);
175 if (lop->lop_destroy == NULL)
178 if (handle->lgh_obj == NULL) {
179 /* if lgh_obj == NULL, then it is from client side destroy */
180 rc = lop->lop_destroy(env, handle, NULL);
184 if (!llog_exist(handle))
187 dt = lu2dt_dev(handle->lgh_obj->do_lu.lo_dev);
189 if (unlikely(unlikely(dt->dd_rdonly)))
192 th = dt_trans_create(env, dt);
196 rc = llog_declare_destroy(env, handle, th);
200 rc = dt_trans_start_local(env, dt, th);
204 rc = lop->lop_destroy(env, handle, th);
207 dt_trans_stop(env, dt, th);
211 EXPORT_SYMBOL(llog_destroy);
213 /* returns negative on error; 0 if success; 1 if success & log destroyed */
214 int llog_cancel_arr_rec(const struct lu_env *env, struct llog_handle *loghandle,
217 struct llog_thread_info *lgi = llog_info(env);
218 struct dt_device *dt;
219 struct llog_log_hdr *llh;
224 bool subtract_count = false;
228 LASSERT(loghandle != NULL);
229 LASSERT(loghandle->lgh_ctxt != NULL);
230 LASSERT(loghandle->lgh_obj != NULL);
232 llh = loghandle->lgh_hdr;
234 CDEBUG(D_RPCTRACE, "Canceling %d records, first %d in log "DFID"\n",
235 num, index[0], PFID(&loghandle->lgh_id.lgl_oi.oi_fid));
237 dt = lu2dt_dev(loghandle->lgh_obj->do_lu.lo_dev);
239 if (unlikely(unlikely(dt->dd_rdonly)))
242 th = dt_trans_create(env, dt);
246 rc = llog_declare_write_rec(env, loghandle, &llh->llh_hdr, 0, th);
250 if ((llh->llh_flags & LLOG_F_ZAP_WHEN_EMPTY)) {
251 rc = llog_declare_destroy(env, loghandle, th);
256 th->th_wait_submit = 1;
257 rc = dt_trans_start_local(env, dt, th);
261 down_write(&loghandle->lgh_lock);
263 mutex_lock(&loghandle->lgh_hdr_mutex);
264 for (i = 0; i < num; ++i) {
266 CERROR("Can't cancel index 0 which is header\n");
267 GOTO(out_unlock, rc = -EINVAL);
269 if (!__test_and_clear_bit_le(index[i], LLOG_HDR_BITMAP(llh))) {
270 CDEBUG(D_RPCTRACE, "Catalog index %u already clear?\n",
272 GOTO(out_unlock, rc = -ENOENT);
275 loghandle->lgh_hdr->llh_count -= num;
276 subtract_count = true;
278 /* Since llog_process_thread use lgi_cookie, it`s better to save them
279 * and restore after using
281 tmp_lgc_index = lgi->lgi_cookie.lgc_index;
282 /* Pass this index to llog_osd_write_rec(), which will use the index
283 * to only update the necesary bitmap. */
284 lgi->lgi_cookie.lgc_index = index[0];
286 rc = llog_write_rec(env, loghandle, &llh->llh_hdr, (num != 1 ? NULL :
287 &lgi->lgi_cookie), LLOG_HEADER_IDX, th);
288 lgi->lgi_cookie.lgc_index = tmp_lgc_index;
291 GOTO(out_unlock, rc);
293 if ((llh->llh_flags & LLOG_F_ZAP_WHEN_EMPTY) &&
294 (llh->llh_count == 1) &&
295 ((loghandle->lgh_last_idx == LLOG_HDR_BITMAP_SIZE(llh) - 1) ||
296 (loghandle->u.phd.phd_cat_handle != NULL &&
297 loghandle->u.phd.phd_cat_handle->u.chd.chd_current_log !=
299 /* never try to destroy it again */
300 llh->llh_flags &= ~LLOG_F_ZAP_WHEN_EMPTY;
301 rc = llog_trans_destroy(env, loghandle, th);
303 /* Sigh, can not destroy the final plain llog, but
304 * the bitmap has been clearly, so the record can not
305 * be accessed anymore, let's return 0 for now, and
306 * the orphan will be handled by LFSCK. */
307 CERROR("%s: can't destroy empty llog "DFID": rc = %d\n",
308 loghandle2name(loghandle),
309 PFID(&loghandle->lgh_id.lgl_oi.oi_fid), rc);
310 GOTO(out_unlock, rc = 0);
316 mutex_unlock(&loghandle->lgh_hdr_mutex);
317 up_write(&loghandle->lgh_lock);
319 rc1 = dt_trans_stop(env, dt, th);
323 mutex_lock(&loghandle->lgh_hdr_mutex);
325 loghandle->lgh_hdr->llh_count += num;
326 for (i = i - 1; i >= 0; i--)
327 set_bit_le(index[i], LLOG_HDR_BITMAP(llh));
328 mutex_unlock(&loghandle->lgh_hdr_mutex);
333 int llog_cancel_rec(const struct lu_env *env, struct llog_handle *loghandle,
336 return llog_cancel_arr_rec(env, loghandle, 1, &index);
339 int llog_read_header(const struct lu_env *env, struct llog_handle *handle,
340 const struct obd_uuid *uuid)
342 const struct llog_operations *lop;
346 rc = llog_handle2ops(handle, &lop);
350 if (lop->lop_read_header == NULL)
353 rc = lop->lop_read_header(env, handle);
354 if (rc == LLOG_EEMPTY) {
355 struct llog_log_hdr *llh = handle->lgh_hdr;
357 /* lrh_len should be initialized in llog_init_handle */
358 handle->lgh_last_idx = 0; /* header is record with index 0 */
359 llh->llh_count = 1; /* for the header record */
360 llh->llh_hdr.lrh_type = LLOG_HDR_MAGIC;
361 LASSERT(handle->lgh_ctxt->loc_chunk_size >=
362 LLOG_MIN_CHUNK_SIZE);
363 llh->llh_hdr.lrh_len = handle->lgh_ctxt->loc_chunk_size;
364 llh->llh_hdr.lrh_index = 0;
365 llh->llh_timestamp = ktime_get_real_seconds();
367 memcpy(&llh->llh_tgtuuid, uuid,
368 sizeof(llh->llh_tgtuuid));
369 llh->llh_bitmap_offset = offsetof(typeof(*llh), llh_bitmap);
370 /* Since update llog header might also call this function,
371 * let's reset the bitmap to 0 here */
372 memset(LLOG_HDR_BITMAP(llh), 0, llh->llh_hdr.lrh_len -
373 llh->llh_bitmap_offset -
374 sizeof(llh->llh_tail));
375 set_bit_le(0, LLOG_HDR_BITMAP(llh));
376 LLOG_HDR_TAIL(llh)->lrt_len = llh->llh_hdr.lrh_len;
377 LLOG_HDR_TAIL(llh)->lrt_index = llh->llh_hdr.lrh_index;
382 EXPORT_SYMBOL(llog_read_header);
384 int llog_init_handle(const struct lu_env *env, struct llog_handle *handle,
385 int flags, struct obd_uuid *uuid)
387 struct llog_log_hdr *llh;
388 enum llog_flag fmt = flags & LLOG_F_EXT_MASK;
390 int chunk_size = handle->lgh_ctxt->loc_chunk_size;
393 LASSERT(handle->lgh_hdr == NULL);
395 LASSERT(chunk_size >= LLOG_MIN_CHUNK_SIZE);
396 OBD_ALLOC_LARGE(llh, chunk_size);
400 handle->lgh_hdr = llh;
401 handle->lgh_hdr_size = chunk_size;
402 /* first assign flags to use llog_client_ops */
403 llh->llh_flags = flags;
404 rc = llog_read_header(env, handle, uuid);
406 if (unlikely((llh->llh_flags & LLOG_F_IS_PLAIN &&
407 flags & LLOG_F_IS_CAT) ||
408 (llh->llh_flags & LLOG_F_IS_CAT &&
409 flags & LLOG_F_IS_PLAIN))) {
410 CERROR("%s: llog type is %s but initializing %s\n",
411 loghandle2name(handle),
412 llh->llh_flags & LLOG_F_IS_CAT ?
414 flags & LLOG_F_IS_CAT ? "catalog" : "plain");
415 GOTO(out, rc = -EINVAL);
416 } else if (llh->llh_flags &
417 (LLOG_F_IS_PLAIN | LLOG_F_IS_CAT)) {
419 * it is possible to open llog without specifying llog
420 * type so it is taken from llh_flags
422 flags = llh->llh_flags;
424 /* for some reason the llh_flags has no type set */
425 CERROR("llog type is not specified!\n");
426 GOTO(out, rc = -EINVAL);
429 !obd_uuid_equals(uuid, &llh->llh_tgtuuid))) {
430 CERROR("%s: llog uuid mismatch: %s/%s\n",
431 loghandle2name(handle),
433 (char *)llh->llh_tgtuuid.uuid);
434 GOTO(out, rc = -EEXIST);
437 if (flags & LLOG_F_IS_CAT) {
438 LASSERT(list_empty(&handle->u.chd.chd_head));
439 INIT_LIST_HEAD(&handle->u.chd.chd_head);
440 llh->llh_size = sizeof(struct llog_logid_rec);
441 llh->llh_flags |= LLOG_F_IS_FIXSIZE;
442 } else if (!(flags & LLOG_F_IS_PLAIN)) {
443 CERROR("%s: unknown flags: %#x (expected %#x or %#x)\n",
444 loghandle2name(handle), flags, LLOG_F_IS_CAT,
448 llh->llh_flags |= fmt;
451 OBD_FREE_LARGE(llh, chunk_size);
452 handle->lgh_hdr = NULL;
456 EXPORT_SYMBOL(llog_init_handle);
458 int llog_verify_record(const struct llog_handle *llh, struct llog_rec_hdr *rec)
460 int chunk_size = llh->lgh_hdr->llh_hdr.lrh_len;
462 if (rec->lrh_len == 0 || rec->lrh_len > chunk_size) {
463 CERROR("%s: record is too large: %d > %d\n",
464 loghandle2name(llh), rec->lrh_len, chunk_size);
467 if (rec->lrh_index >= LLOG_HDR_BITMAP_SIZE(llh->lgh_hdr)) {
468 CERROR("%s: index is too high: %d\n",
469 loghandle2name(llh), rec->lrh_index);
472 if ((rec->lrh_type & LLOG_OP_MASK) != LLOG_OP_MAGIC) {
473 CERROR("%s: magic %x is bad\n",
474 loghandle2name(llh), rec->lrh_type);
480 EXPORT_SYMBOL(llog_verify_record);
482 static int llog_process_thread(void *arg)
484 struct llog_process_info *lpi = arg;
485 struct llog_handle *loghandle = lpi->lpi_loghandle;
486 struct llog_log_hdr *llh = loghandle->lgh_hdr;
487 struct llog_process_cat_data *cd = lpi->lpi_catdata;
488 struct llog_thread_info *lti;
492 int rc = 0, index = 1, last_index;
494 int last_called_index = 0;
495 bool repeated = false;
496 bool refresh_idx = false;
503 lti = lpi->lpi_env == NULL ? NULL : llog_info(lpi->lpi_env);
505 cur_offset = chunk_size = llh->llh_hdr.lrh_len;
506 /* expect chunk_size to be power of two */
507 LASSERT(is_power_of_2(chunk_size));
509 OBD_ALLOC_LARGE(buf, chunk_size);
511 lpi->lpi_rc = -ENOMEM;
516 last_called_index = cd->lpcd_first_idx;
517 index = cd->lpcd_first_idx + 1;
519 if (cd != NULL && cd->lpcd_last_idx)
520 last_index = cd->lpcd_last_idx;
522 last_index = LLOG_HDR_BITMAP_SIZE(llh) - 1;
525 struct llog_rec_hdr *rec;
526 off_t chunk_offset = 0;
527 unsigned int buf_offset = 0;
531 /* skip records not set in bitmap */
532 while (index <= last_index &&
533 !test_bit_le(index, LLOG_HDR_BITMAP(llh)))
536 /* There are no indices prior the last_index */
537 if (index > last_index)
540 CDEBUG(D_OTHER, "index: %d last_index %d\n", index,
544 /* get the buf with our target record; avoid old garbage */
545 memset(buf, 0, chunk_size);
546 /* the record index for outdated chunk data */
547 /* it is safe to process buffer until saved lgh_last_idx */
548 lh_last_idx = LLOG_HDR_TAIL(llh)->lrt_index;
549 rc = llog_next_block(lpi->lpi_env, loghandle, &saved_index,
550 index, &cur_offset, buf, chunk_size);
552 CDEBUG(D_OTHER, "cur_offset %llu, chunk_offset %llu,"
553 " buf_offset %u, rc = %d\n", cur_offset,
554 (__u64)chunk_offset, buf_offset, rc);
555 /* we`ve tried to reread the chunk, but there is no
557 if (rc == -EIO && repeated && (chunk_offset + buf_offset) ==
563 /* NB: after llog_next_block() call the cur_offset is the
564 * offset of the next block after read one.
565 * The absolute offset of the current chunk is calculated
566 * from cur_offset value and stored in chunk_offset variable.
568 if ((cur_offset & (chunk_size - 1)) != 0)
569 chunk_offset = cur_offset & ~(chunk_size - 1);
571 chunk_offset = cur_offset - chunk_size;
573 /* NB: when rec->lrh_len is accessed it is already swabbed
574 * since it is used at the "end" of the loop and the rec
575 * swabbing is done at the beginning of the loop. */
576 for (rec = (struct llog_rec_hdr *)(buf + buf_offset);
577 (char *)rec < buf + chunk_size;
578 rec = llog_rec_hdr_next(rec)) {
580 CDEBUG(D_OTHER, "processing rec 0x%p type %#x\n",
583 if (LLOG_REC_HDR_NEEDS_SWABBING(rec))
584 lustre_swab_llog_rec(rec);
586 CDEBUG(D_OTHER, "after swabbing, type=%#x idx=%d\n",
587 rec->lrh_type, rec->lrh_index);
589 if (index == (synced_idx + 1) &&
590 synced_idx == LLOG_HDR_TAIL(llh)->lrt_index)
593 if (OBD_FAIL_PRECHECK(OBD_FAIL_LLOG_PROCESS_TIMEOUT) &&
594 cfs_fail_val == (unsigned int)
595 (loghandle->lgh_id.lgl_oi.oi.oi_id &
597 OBD_RACE(OBD_FAIL_LLOG_PROCESS_TIMEOUT);
600 /* the bitmap could be changed during processing
601 * records from the chunk. For wrapped catalog
602 * it means we can read deleted record and try to
603 * process it. Check this case and reread the chunk.
604 * It is safe to process to lh_last_idx, including
605 * lh_last_idx if it was synced. We can not do <=
606 * comparison, cause for wrapped catalog lgh_last_idx
607 * could be less than index. So we detect last index
608 * for processing as index == lh_last_idx+1. But when
609 * catalog is wrapped and full lgh_last_idx=llh_cat_idx,
610 * the first processing index is llh_cat_idx+1.The
611 * exception is !(lgh_last_idx == llh_cat_idx &&
612 * index == llh_cat_idx + 1), and after simplification
614 * lh_last_idx != LLOG_HDR_TAIL(llh)->lrt_index
615 * This exception is working for catalog only.
618 if ((index == lh_last_idx && synced_idx != index) ||
619 (index == (lh_last_idx + 1) &&
620 lh_last_idx != LLOG_HDR_TAIL(llh)->lrt_index) ||
621 (rec->lrh_index == 0 && !repeated)) {
623 /* save offset inside buffer for the re-read */
624 buf_offset = (char *)rec - (char *)buf;
625 cur_offset = chunk_offset;
627 /* We need to be sure lgh_last_idx
628 * record was saved to disk
630 down_read(&loghandle->lgh_last_sem);
631 synced_idx = LLOG_HDR_TAIL(llh)->lrt_index;
632 up_read(&loghandle->lgh_last_sem);
633 CDEBUG(D_OTHER, "synced_idx: %d\n", synced_idx);
640 rc = llog_verify_record(loghandle, rec);
642 CERROR("%s: invalid record in llog "DFID
643 " record for index %d/%d: rc = %d\n",
644 loghandle2name(loghandle),
645 PFID(&loghandle->lgh_id.lgl_oi.oi_fid),
646 rec->lrh_index, index, rc);
648 * the block seem to be corrupted, let's try
649 * with the next one. reset rc to go to the
654 GOTO(repeat, rc = 0);
657 if (rec->lrh_index < index) {
658 CDEBUG(D_OTHER, "skipping lrh_index %d\n",
663 if (rec->lrh_index != index) {
665 * the last time we couldn't parse the block due
666 * to corruption, thus has no idea about the
667 * next index, take it from the block, once.
671 index = rec->lrh_index;
673 CERROR("%s: "DFID" Invalid record: index"
674 " %u but expected %u\n",
675 loghandle2name(loghandle),
676 PFID(&loghandle->lgh_id.lgl_oi.oi_fid),
677 rec->lrh_index, index);
678 GOTO(out, rc = -ERANGE);
683 "lrh_index: %d lrh_len: %d (%d remains)\n",
684 rec->lrh_index, rec->lrh_len,
685 (int)(buf + chunk_size - (char *)rec));
687 /* lgh_cur_offset is used only at llog_test_3 */
688 loghandle->lgh_cur_offset = (char *)rec - (char *)buf +
691 /* if set, process the callback on this record */
692 if (test_bit_le(index, LLOG_HDR_BITMAP(llh))) {
693 struct llog_cookie *lgc;
697 CDEBUG((llh->llh_flags & LLOG_F_IS_CAT ?
699 "index: %d, lh_last_idx: %d "
700 "synced_idx: %d lgh_last_idx: %d\n",
701 index, lh_last_idx, synced_idx,
702 loghandle->lgh_last_idx);
705 lgc = <i->lgi_cookie;
706 /* store lu_env for recursive calls */
707 tmp_off = lgc->lgc_offset;
708 tmp_idx = lgc->lgc_index;
710 lgc->lgc_offset = (char *)rec -
711 (char *)buf + chunk_offset;
712 lgc->lgc_index = rec->lrh_index;
714 /* using lu_env for passing record offset to
715 * llog_write through various callbacks */
716 rc = lpi->lpi_cb(lpi->lpi_env, loghandle, rec,
718 last_called_index = index;
721 lgc->lgc_offset = tmp_off;
722 lgc->lgc_index = tmp_idx;
725 if (rc == LLOG_PROC_BREAK) {
727 } else if (rc == LLOG_DEL_RECORD) {
728 rc = llog_cancel_rec(lpi->lpi_env,
734 /* some stupid callbacks directly cancel records
735 * and delete llog. Check it and stop
737 if (loghandle->lgh_hdr == NULL ||
738 loghandle->lgh_hdr->llh_count == 1)
741 /* exit if the last index is reached */
742 if (index >= last_index)
749 CDEBUG(D_HA, "stop processing %s "DOSTID":%x index %d count %d\n",
750 ((llh->llh_flags & LLOG_F_IS_CAT) ? "catalog" : "plain"),
751 POSTID(&loghandle->lgh_id.lgl_oi), loghandle->lgh_id.lgl_ogen,
752 index, llh->llh_count);
755 cd->lpcd_last_idx = last_called_index;
757 if (unlikely(rc == -EIO && loghandle->lgh_obj != NULL)) {
758 if (dt_object_remote(loghandle->lgh_obj)) {
759 /* If it is remote object, then -EIO might means
760 * disconnection or eviction, let's return -EAGAIN,
761 * so for update recovery log processing, it will
762 * retry until the umount or abort recovery, see
763 * lod_sub_recovery_thread() */
764 CERROR("%s retry remote llog process\n",
765 loghandle2name(loghandle));
768 /* something bad happened to the processing of a local
769 * llog file, probably I/O error or the log got
770 * corrupted to be able to finally release the log we
771 * discard any remaining bits in the header */
772 CERROR("%s: Local llog found corrupted #"DOSTID":%x"
773 " %s index %d count %d\n",
774 loghandle2name(loghandle),
775 POSTID(&loghandle->lgh_id.lgl_oi),
776 loghandle->lgh_id.lgl_ogen,
777 ((llh->llh_flags & LLOG_F_IS_CAT) ? "catalog" :
778 "plain"), index, llh->llh_count);
780 while (index <= last_index) {
781 if (test_bit_le(index,
782 LLOG_HDR_BITMAP(llh)) != 0)
783 llog_cancel_rec(lpi->lpi_env, loghandle,
791 OBD_FREE_LARGE(buf, chunk_size);
796 static int llog_process_thread_daemonize(void *arg)
798 struct llog_process_info *lpi = arg;
801 struct nsproxy *new_ns, *curr_ns = current->nsproxy;
803 task_lock(lpi->lpi_reftask);
804 new_ns = lpi->lpi_reftask->nsproxy;
805 if (curr_ns != new_ns) {
808 current->nsproxy = new_ns;
809 /* XXX: we should call put_nsproxy() instead of
810 * atomic_dec(&ns->count) directly. But put_nsproxy() cannot be
811 * used outside of the kernel itself, because it calls
812 * free_nsproxy() which is not exported by the kernel
813 * (defined in kernel/nsproxy.c) */
815 atomic_dec(&curr_ns->count);
817 task_unlock(lpi->lpi_reftask);
819 /* client env has no keys, tags is just 0 */
820 rc = lu_env_init(&env, LCT_LOCAL | LCT_MG_THREAD);
825 rc = llog_process_thread(arg);
829 complete(&lpi->lpi_completion);
833 int llog_process_or_fork(const struct lu_env *env,
834 struct llog_handle *loghandle,
835 llog_cb_t cb, void *data, void *catdata, bool fork)
837 struct llog_process_info *lpi;
838 struct llog_process_data *d = data;
839 struct llog_process_cat_data *cd = catdata;
840 __u32 flags = loghandle->lgh_hdr->llh_flags;
847 CERROR("cannot alloc pointer\n");
850 lpi->lpi_loghandle = loghandle;
852 lpi->lpi_cbdata = data;
853 lpi->lpi_catdata = catdata;
855 CDEBUG(D_OTHER, "Processing "DFID" flags 0x%03x startcat %d startidx %d first_idx %d last_idx %d\n",
856 PFID(&loghandle->lgh_id.lgl_oi.oi_fid), flags,
857 (flags & LLOG_F_IS_CAT) && d ? d->lpd_startcat : -1,
858 (flags & LLOG_F_IS_CAT) && d ? d->lpd_startidx : -1,
859 cd ? cd->lpcd_first_idx : -1, cd ? cd->lpcd_last_idx : -1);
861 struct task_struct *task;
863 /* The new thread can't use parent env,
864 * init the new one in llog_process_thread_daemonize. */
866 init_completion(&lpi->lpi_completion);
867 /* take reference to current, so that
868 * llog_process_thread_daemonize() can use it to switch to
869 * namespace associated with current */
870 lpi->lpi_reftask = current;
871 task = kthread_run(llog_process_thread_daemonize, lpi,
872 "llog_process_thread");
875 CERROR("%s: cannot start thread: rc = %d\n",
876 loghandle2name(loghandle), rc);
879 wait_for_completion(&lpi->lpi_completion);
882 llog_process_thread(lpi);
890 EXPORT_SYMBOL(llog_process_or_fork);
892 int llog_process(const struct lu_env *env, struct llog_handle *loghandle,
893 llog_cb_t cb, void *data, void *catdata)
896 rc = llog_process_or_fork(env, loghandle, cb, data, catdata, true);
897 return rc == LLOG_DEL_PLAIN ? 0 : rc;
899 EXPORT_SYMBOL(llog_process);
901 static inline const struct cred *llog_raise_resource(void)
903 struct cred *cred = NULL;
905 if (cap_raised(current_cap(), CAP_SYS_RESOURCE))
908 cred = prepare_creds();
912 cap_raise(cred->cap_effective, CAP_SYS_RESOURCE);
913 return override_creds(cred);
916 static inline void llog_restore_resource(const struct cred *old_cred)
919 revert_creds(old_cred);
922 int llog_reverse_process(const struct lu_env *env,
923 struct llog_handle *loghandle, llog_cb_t cb,
924 void *data, void *catdata)
926 struct llog_log_hdr *llh = loghandle->lgh_hdr;
927 struct llog_process_cat_data *cd = catdata;
929 int rc = 0, first_index = 1, index, idx;
930 __u32 chunk_size = llh->llh_hdr.lrh_len;
933 OBD_ALLOC_LARGE(buf, chunk_size);
938 first_index = cd->lpcd_first_idx + 1;
939 if (cd != NULL && cd->lpcd_last_idx)
940 index = cd->lpcd_last_idx;
942 index = LLOG_HDR_BITMAP_SIZE(llh) - 1;
945 struct llog_rec_hdr *rec;
946 struct llog_rec_tail *tail;
948 /* skip records not set in bitmap */
949 while (index >= first_index &&
950 !test_bit_le(index, LLOG_HDR_BITMAP(llh)))
953 LASSERT(index >= first_index - 1);
954 if (index == first_index - 1)
957 /* get the buf with our target record; avoid old garbage */
958 memset(buf, 0, chunk_size);
959 rc = llog_prev_block(env, loghandle, index, buf, chunk_size);
964 idx = rec->lrh_index;
965 CDEBUG(D_RPCTRACE, "index %u : idx %u\n", index, idx);
966 while (idx < index) {
967 rec = (void *)rec + rec->lrh_len;
968 if (LLOG_REC_HDR_NEEDS_SWABBING(rec))
969 lustre_swab_llog_rec(rec);
972 LASSERT(idx == index);
973 tail = (void *)rec + rec->lrh_len - sizeof(*tail);
975 /* process records in buffer, starting where we found one */
976 while ((void *)tail > buf) {
977 if (tail->lrt_index == 0)
978 GOTO(out, rc = 0); /* no more records */
980 /* if set, process the callback on this record */
981 if (test_bit_le(index, LLOG_HDR_BITMAP(llh))) {
982 rec = (void *)tail - tail->lrt_len +
985 rc = cb(env, loghandle, rec, data);
986 if (rc == LLOG_PROC_BREAK) {
988 } else if (rc == LLOG_DEL_RECORD) {
989 rc = llog_cancel_rec(env, loghandle,
996 /* previous record, still in buffer? */
998 if (index < first_index)
1000 tail = (void *)tail - tail->lrt_len;
1006 OBD_FREE_LARGE(buf, chunk_size);
1009 EXPORT_SYMBOL(llog_reverse_process);
1015 * llog_open - open llog, may not exist
1016 * llog_exist - check if llog exists
1017 * llog_close - close opened llog, pair for open, frees llog_handle
1018 * llog_declare_create - declare llog creation
1019 * llog_create - create new llog on disk, need transaction handle
1020 * llog_declare_write_rec - declaration of llog write
1021 * llog_write_rec - write llog record on disk, need transaction handle
1022 * llog_declare_add - declare llog catalog record addition
1023 * llog_add - add llog record in catalog, need transaction handle
1025 int llog_exist(struct llog_handle *loghandle)
1027 const struct llog_operations *lop;
1032 rc = llog_handle2ops(loghandle, &lop);
1035 if (lop->lop_exist == NULL)
1036 RETURN(-EOPNOTSUPP);
1038 rc = lop->lop_exist(loghandle);
1041 EXPORT_SYMBOL(llog_exist);
1043 int llog_declare_create(const struct lu_env *env,
1044 struct llog_handle *loghandle, struct thandle *th)
1046 const struct cred *old_cred;
1047 const struct llog_operations *lop;
1052 rc = llog_handle2ops(loghandle, &lop);
1055 if (lop->lop_declare_create == NULL)
1056 RETURN(-EOPNOTSUPP);
1058 old_cred = llog_raise_resource();
1059 rc = lop->lop_declare_create(env, loghandle, th);
1060 llog_restore_resource(old_cred);
1064 int llog_create(const struct lu_env *env, struct llog_handle *handle,
1067 const struct cred *old_cred;
1068 const struct llog_operations *lop;
1073 rc = llog_handle2ops(handle, &lop);
1076 if (lop->lop_create == NULL)
1077 RETURN(-EOPNOTSUPP);
1079 old_cred = llog_raise_resource();
1080 rc = lop->lop_create(env, handle, th);
1081 llog_restore_resource(old_cred);
1085 int llog_declare_write_rec(const struct lu_env *env,
1086 struct llog_handle *handle,
1087 struct llog_rec_hdr *rec, int idx,
1090 const struct cred *old_cred;
1091 const struct llog_operations *lop;
1096 rc = llog_handle2ops(handle, &lop);
1100 if (lop->lop_declare_write_rec == NULL)
1101 RETURN(-EOPNOTSUPP);
1103 old_cred = llog_raise_resource();
1104 rc = lop->lop_declare_write_rec(env, handle, rec, idx, th);
1105 llog_restore_resource(old_cred);
1109 int llog_write_rec(const struct lu_env *env, struct llog_handle *handle,
1110 struct llog_rec_hdr *rec, struct llog_cookie *logcookies,
1111 int idx, struct thandle *th)
1113 const struct cred *old_cred;
1114 const struct llog_operations *lop;
1119 /* API sanity checks */
1120 if (handle == NULL) {
1121 CERROR("loghandle is missed\n");
1123 } else if (handle->lgh_obj == NULL) {
1124 CERROR("loghandle %p with NULL object\n",
1127 } else if (th == NULL) {
1128 CERROR("%s: missed transaction handle\n",
1129 loghandle2name(handle));
1131 } else if (handle->lgh_hdr == NULL) {
1132 CERROR("%s: loghandle %p with no header\n",
1133 loghandle2name(handle), handle);
1137 rc = llog_handle2ops(handle, &lop);
1141 if (lop->lop_write_rec == NULL)
1142 RETURN(-EOPNOTSUPP);
1144 buflen = rec->lrh_len;
1145 LASSERT(cfs_size_round(buflen) == buflen);
1147 old_cred = llog_raise_resource();
1148 rc = lop->lop_write_rec(env, handle, rec, logcookies, idx, th);
1149 llog_restore_resource(old_cred);
1153 int llog_add(const struct lu_env *env, struct llog_handle *lgh,
1154 struct llog_rec_hdr *rec, struct llog_cookie *logcookies,
1157 const struct cred *old_cred;
1162 if (lgh->lgh_logops->lop_add == NULL)
1163 RETURN(-EOPNOTSUPP);
1165 old_cred = llog_raise_resource();
1166 rc = lgh->lgh_logops->lop_add(env, lgh, rec, logcookies, th);
1167 llog_restore_resource(old_cred);
1170 EXPORT_SYMBOL(llog_add);
1172 int llog_declare_add(const struct lu_env *env, struct llog_handle *lgh,
1173 struct llog_rec_hdr *rec, struct thandle *th)
1175 const struct cred *old_cred;
1180 if (lgh->lgh_logops->lop_declare_add == NULL)
1181 RETURN(-EOPNOTSUPP);
1183 old_cred = llog_raise_resource();
1184 rc = lgh->lgh_logops->lop_declare_add(env, lgh, rec, th);
1185 llog_restore_resource(old_cred);
1188 EXPORT_SYMBOL(llog_declare_add);
1191 * Helper function to open llog or create it if doesn't exist.
1192 * It hides all transaction handling from caller.
1194 int llog_open_create(const struct lu_env *env, struct llog_ctxt *ctxt,
1195 struct llog_handle **res, struct llog_logid *logid,
1198 struct dt_device *d;
1204 rc = llog_open(env, ctxt, res, logid, name, LLOG_OPEN_NEW);
1208 if (llog_exist(*res))
1211 LASSERT((*res)->lgh_obj != NULL);
1213 d = lu2dt_dev((*res)->lgh_obj->do_lu.lo_dev);
1215 if (unlikely(unlikely(d->dd_rdonly)))
1218 th = dt_trans_create(env, d);
1220 GOTO(out, rc = PTR_ERR(th));
1222 /* Create update llog object synchronously, which
1223 * happens during inialization process see
1224 * lod_sub_prep_llog(), to make sure the update
1225 * llog object is created before corss-MDT writing
1226 * updates into the llog object */
1227 if (ctxt->loc_flags & LLOG_CTXT_FLAG_NORMAL_FID)
1230 th->th_wait_submit = 1;
1231 rc = llog_declare_create(env, *res, th);
1233 rc = dt_trans_start_local(env, d, th);
1235 rc = llog_create(env, *res, th);
1237 dt_trans_stop(env, d, th);
1240 llog_close(env, *res);
1243 EXPORT_SYMBOL(llog_open_create);
1246 * Helper function to delete existent llog.
1248 int llog_erase(const struct lu_env *env, struct llog_ctxt *ctxt,
1249 struct llog_logid *logid, char *name)
1251 struct llog_handle *handle;
1256 /* nothing to erase */
1257 if (name == NULL && logid == NULL)
1260 rc = llog_open(env, ctxt, &handle, logid, name, LLOG_OPEN_EXISTS);
1264 rc = llog_init_handle(env, handle, LLOG_F_IS_PLAIN, NULL);
1266 rc = llog_destroy(env, handle);
1268 rc2 = llog_close(env, handle);
1273 EXPORT_SYMBOL(llog_erase);
1276 * Helper function for write record in llog.
1277 * It hides all transaction handling from caller.
1278 * Valid only with local llog.
1280 int llog_write(const struct lu_env *env, struct llog_handle *loghandle,
1281 struct llog_rec_hdr *rec, int idx)
1283 struct dt_device *dt;
1291 LASSERT(loghandle->lgh_ctxt);
1292 LASSERT(loghandle->lgh_obj != NULL);
1294 dt = lu2dt_dev(loghandle->lgh_obj->do_lu.lo_dev);
1296 if (unlikely(unlikely(dt->dd_rdonly)))
1299 th = dt_trans_create(env, dt);
1301 RETURN(PTR_ERR(th));
1303 rc = llog_declare_write_rec(env, loghandle, rec, idx, th);
1305 GOTO(out_trans, rc);
1307 th->th_wait_submit = 1;
1308 rc = dt_trans_start_local(env, dt, th);
1310 GOTO(out_trans, rc);
1312 need_cookie = !(idx == LLOG_HEADER_IDX || idx == LLOG_NEXT_IDX);
1314 down_write(&loghandle->lgh_lock);
1316 struct llog_thread_info *lti = llog_info(env);
1318 /* cookie comes from llog_process_thread */
1319 rc = llog_write_rec(env, loghandle, rec, <i->lgi_cookie,
1320 rec->lrh_index, th);
1321 /* upper layer didn`t pass cookie so change rc */
1322 rc = (rc == 1 ? 0 : rc);
1324 rc = llog_write_rec(env, loghandle, rec, NULL, idx, th);
1327 up_write(&loghandle->lgh_lock);
1329 dt_trans_stop(env, dt, th);
1332 EXPORT_SYMBOL(llog_write);
1334 int llog_open(const struct lu_env *env, struct llog_ctxt *ctxt,
1335 struct llog_handle **lgh, struct llog_logid *logid,
1336 char *name, enum llog_open_param open_param)
1338 const struct cred *old_cred;
1344 LASSERT(ctxt->loc_logops);
1346 if (ctxt->loc_logops->lop_open == NULL) {
1348 RETURN(-EOPNOTSUPP);
1351 *lgh = llog_alloc_handle();
1354 (*lgh)->lgh_ctxt = ctxt;
1355 (*lgh)->lgh_logops = ctxt->loc_logops;
1357 old_cred = llog_raise_resource();
1358 rc = ctxt->loc_logops->lop_open(env, *lgh, logid, name, open_param);
1359 llog_restore_resource(old_cred);
1361 llog_free_handle(*lgh);
1366 EXPORT_SYMBOL(llog_open);
1368 int llog_close(const struct lu_env *env, struct llog_handle *loghandle)
1370 return llog_handle_put(env, loghandle);
1372 EXPORT_SYMBOL(llog_close);
1375 * Helper function to get the llog size in records. It is used by MGS
1376 * mostly to check that config llog exists and contains data.
1378 * \param[in] env execution environment
1379 * \param[in] ctxt llog context
1380 * \param[in] name llog name
1382 * \retval true if there are records in llog besides a header
1383 * \retval false on error or llog without records
1385 int llog_is_empty(const struct lu_env *env, struct llog_ctxt *ctxt,
1388 struct llog_handle *llh;
1391 rc = llog_open(env, ctxt, &llh, NULL, name, LLOG_OPEN_EXISTS);
1393 if (likely(rc == -ENOENT))
1398 rc = llog_init_handle(env, llh, LLOG_F_IS_PLAIN, NULL);
1400 GOTO(out_close, rc);
1401 rc = llog_get_size(llh);
1404 llog_close(env, llh);
1406 /* The header is record 1, the llog is still considered as empty
1407 * if there is only header */
1410 EXPORT_SYMBOL(llog_is_empty);
1412 int llog_copy_handler(const struct lu_env *env, struct llog_handle *llh,
1413 struct llog_rec_hdr *rec, void *data)
1415 struct llog_handle *copy_llh = data;
1417 /* Append all records */
1418 return llog_write(env, copy_llh, rec, LLOG_NEXT_IDX);
1421 /* backup plain llog */
1422 int llog_backup(const struct lu_env *env, struct obd_device *obd,
1423 struct llog_ctxt *ctxt, struct llog_ctxt *bctxt,
1424 char *name, char *backup)
1426 struct llog_handle *llh, *bllh;
1431 /* open original log */
1432 rc = llog_open(env, ctxt, &llh, NULL, name, LLOG_OPEN_EXISTS);
1434 /* the -ENOENT case is also reported to the caller
1435 * but silently so it should handle that if needed.
1438 CERROR("%s: failed to open log %s: rc = %d\n",
1439 obd->obd_name, name, rc);
1443 rc = llog_init_handle(env, llh, LLOG_F_IS_PLAIN, NULL);
1445 GOTO(out_close, rc);
1447 /* Make sure there's no old backup log */
1448 rc = llog_erase(env, bctxt, NULL, backup);
1449 if (rc < 0 && rc != -ENOENT)
1450 GOTO(out_close, rc);
1452 /* open backup log */
1453 rc = llog_open_create(env, bctxt, &bllh, NULL, backup);
1455 CERROR("%s: failed to open backup logfile %s: rc = %d\n",
1456 obd->obd_name, backup, rc);
1457 GOTO(out_close, rc);
1460 /* check that backup llog is not the same object as original one */
1461 if (llh->lgh_obj == bllh->lgh_obj) {
1462 CERROR("%s: backup llog %s to itself (%s), objects %p/%p\n",
1463 obd->obd_name, name, backup, llh->lgh_obj,
1465 GOTO(out_backup, rc = -EEXIST);
1468 rc = llog_init_handle(env, bllh, LLOG_F_IS_PLAIN, NULL);
1470 GOTO(out_backup, rc);
1472 /* Copy log record by record */
1473 rc = llog_process_or_fork(env, llh, llog_copy_handler, (void *)bllh,
1476 CERROR("%s: failed to backup log %s: rc = %d\n",
1477 obd->obd_name, name, rc);
1479 llog_close(env, bllh);
1481 llog_close(env, llh);
1484 EXPORT_SYMBOL(llog_backup);
1486 /* Get size of llog */
1487 __u64 llog_size(const struct lu_env *env, struct llog_handle *llh)
1492 rc = llh->lgh_obj->do_ops->do_attr_get(env, llh->lgh_obj, &la);
1494 CERROR("%s: attr_get failed for "DFID": rc = %d\n",
1495 loghandle2name(llh), PFID(&llh->lgh_id.lgl_oi.oi_fid),
1502 EXPORT_SYMBOL(llog_size);