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"
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 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 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 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 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 loghandle->lgh_ctxt->loc_obd->obd_name,
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 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 handle->lgh_ctxt->loc_obd->obd_name,
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 handle->lgh_ctxt->loc_obd->obd_name,
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 handle->lgh_ctxt->loc_obd->obd_name,
444 flags, LLOG_F_IS_CAT, LLOG_F_IS_PLAIN);
447 llh->llh_flags |= fmt;
450 OBD_FREE_LARGE(llh, chunk_size);
451 handle->lgh_hdr = NULL;
455 EXPORT_SYMBOL(llog_init_handle);
457 static int llog_process_thread(void *arg)
459 struct llog_process_info *lpi = arg;
460 struct llog_handle *loghandle = lpi->lpi_loghandle;
461 struct llog_log_hdr *llh = loghandle->lgh_hdr;
462 struct llog_process_cat_data *cd = lpi->lpi_catdata;
463 struct llog_thread_info *lti;
467 int rc = 0, index = 1, last_index;
469 int last_called_index = 0;
470 bool repeated = false;
477 lti = lpi->lpi_env == NULL ? NULL : llog_info(lpi->lpi_env);
479 cur_offset = chunk_size = llh->llh_hdr.lrh_len;
480 /* expect chunk_size to be power of two */
481 LASSERT(is_power_of_2(chunk_size));
483 OBD_ALLOC_LARGE(buf, chunk_size);
485 lpi->lpi_rc = -ENOMEM;
490 last_called_index = cd->lpcd_first_idx;
491 index = cd->lpcd_first_idx + 1;
493 if (cd != NULL && cd->lpcd_last_idx)
494 last_index = cd->lpcd_last_idx;
496 last_index = LLOG_HDR_BITMAP_SIZE(llh) - 1;
499 struct llog_rec_hdr *rec;
500 off_t chunk_offset = 0;
501 unsigned int buf_offset = 0;
506 /* skip records not set in bitmap */
507 while (index <= last_index &&
508 !test_bit_le(index, LLOG_HDR_BITMAP(llh)))
511 /* There are no indices prior the last_index */
512 if (index > last_index)
515 CDEBUG(D_OTHER, "index: %d last_index %d\n", index,
519 /* get the buf with our target record; avoid old garbage */
520 memset(buf, 0, chunk_size);
521 /* the record index for outdated chunk data */
522 /* it is safe to process buffer until saved lgh_last_idx */
523 lh_last_idx = LLOG_HDR_TAIL(llh)->lrt_index;
524 rc = llog_next_block(lpi->lpi_env, loghandle, &saved_index,
525 index, &cur_offset, buf, chunk_size);
527 CDEBUG(D_OTHER, "cur_offset %llu, chunk_offset %llu,"
528 " buf_offset %u, rc = %d\n", cur_offset,
529 (__u64)chunk_offset, buf_offset, rc);
530 /* we`ve tried to reread the chunk, but there is no
532 if (rc == -EIO && repeated && (chunk_offset + buf_offset) ==
538 /* NB: after llog_next_block() call the cur_offset is the
539 * offset of the next block after read one.
540 * The absolute offset of the current chunk is calculated
541 * from cur_offset value and stored in chunk_offset variable.
543 if ((cur_offset & (chunk_size - 1)) != 0) {
544 partial_chunk = true;
545 chunk_offset = cur_offset & ~(chunk_size - 1);
547 partial_chunk = false;
548 chunk_offset = cur_offset - chunk_size;
551 /* NB: when rec->lrh_len is accessed it is already swabbed
552 * since it is used at the "end" of the loop and the rec
553 * swabbing is done at the beginning of the loop. */
554 for (rec = (struct llog_rec_hdr *)(buf + buf_offset);
555 (char *)rec < buf + chunk_size;
556 rec = llog_rec_hdr_next(rec)) {
558 CDEBUG(D_OTHER, "processing rec 0x%p type %#x\n",
561 if (LLOG_REC_HDR_NEEDS_SWABBING(rec))
562 lustre_swab_llog_rec(rec);
564 CDEBUG(D_OTHER, "after swabbing, type=%#x idx=%d\n",
565 rec->lrh_type, rec->lrh_index);
567 if (index == (synced_idx + 1) &&
568 synced_idx == LLOG_HDR_TAIL(llh)->lrt_index)
571 if (OBD_FAIL_PRECHECK(OBD_FAIL_LLOG_PROCESS_TIMEOUT) &&
572 cfs_fail_val == (unsigned int)
573 (loghandle->lgh_id.lgl_oi.oi.oi_id &
575 OBD_RACE(OBD_FAIL_LLOG_PROCESS_TIMEOUT);
578 /* the bitmap could be changed during processing
579 * records from the chunk. For wrapped catalog
580 * it means we can read deleted record and try to
581 * process it. Check this case and reread the chunk.
582 * It is safe to process to lh_last_idx, including
583 * lh_last_idx if it was synced. We can not do <=
584 * comparison, cause for wrapped catalog lgh_last_idx
585 * could be less than index. So we detect last index
586 * for processing as index == lh_last_idx+1. But when
587 * catalog is wrapped and full lgh_last_idx=llh_cat_idx,
588 * the first processing index is llh_cat_idx+1.The
589 * exception is !(lgh_last_idx == llh_cat_idx &&
590 * index == llh_cat_idx + 1), and after simplification
592 * lh_last_idx != LLOG_HDR_TAIL(llh)->lrt_index
593 * This exception is working for catalog only.
596 if ((index == lh_last_idx && synced_idx != index) ||
597 (index == (lh_last_idx + 1) &&
598 lh_last_idx != LLOG_HDR_TAIL(llh)->lrt_index) ||
599 (rec->lrh_index == 0 && !repeated)) {
601 /* save offset inside buffer for the re-read */
602 buf_offset = (char *)rec - (char *)buf;
603 cur_offset = chunk_offset;
605 /* We need to be sure lgh_last_idx
606 * record was saved to disk
608 down_read(&loghandle->lgh_last_sem);
609 synced_idx = LLOG_HDR_TAIL(llh)->lrt_index;
610 up_read(&loghandle->lgh_last_sem);
611 CDEBUG(D_OTHER, "synced_idx: %d\n", synced_idx);
618 if (rec->lrh_len == 0 || rec->lrh_len > chunk_size) {
619 CWARN("%s: invalid length %d in llog "DFID
620 "record for index %d/%d\n",
621 loghandle->lgh_ctxt->loc_obd->obd_name,
623 PFID(&loghandle->lgh_id.lgl_oi.oi_fid),
624 rec->lrh_index, index);
626 GOTO(out, rc = -EINVAL);
629 if (rec->lrh_index < index) {
630 CDEBUG(D_OTHER, "skipping lrh_index %d\n",
635 if (rec->lrh_index != index) {
636 CERROR("%s: "DFID" Invalid record: index %u"
637 " but expected %u\n",
638 loghandle->lgh_ctxt->loc_obd->obd_name,
639 PFID(&loghandle->lgh_id.lgl_oi.oi_fid),
640 rec->lrh_index, index);
641 GOTO(out, rc = -ERANGE);
645 "lrh_index: %d lrh_len: %d (%d remains)\n",
646 rec->lrh_index, rec->lrh_len,
647 (int)(buf + chunk_size - (char *)rec));
649 /* lgh_cur_offset is used only at llog_test_3 */
650 loghandle->lgh_cur_offset = (char *)rec - (char *)buf +
653 /* if set, process the callback on this record */
654 if (test_bit_le(index, LLOG_HDR_BITMAP(llh))) {
655 struct llog_cookie *lgc;
659 CDEBUG((llh->llh_flags & LLOG_F_IS_CAT ?
661 "index: %d, lh_last_idx: %d "
662 "synced_idx: %d lgh_last_idx: %d\n",
663 index, lh_last_idx, synced_idx,
664 loghandle->lgh_last_idx);
667 lgc = <i->lgi_cookie;
668 /* store lu_env for recursive calls */
669 tmp_off = lgc->lgc_offset;
670 tmp_idx = lgc->lgc_index;
672 lgc->lgc_offset = (char *)rec -
673 (char *)buf + chunk_offset;
674 lgc->lgc_index = rec->lrh_index;
676 /* using lu_env for passing record offset to
677 * llog_write through various callbacks */
678 rc = lpi->lpi_cb(lpi->lpi_env, loghandle, rec,
680 last_called_index = index;
683 lgc->lgc_offset = tmp_off;
684 lgc->lgc_index = tmp_idx;
687 if (rc == LLOG_PROC_BREAK) {
689 } else if (rc == LLOG_DEL_RECORD) {
690 rc = llog_cancel_rec(lpi->lpi_env,
696 /* some stupid callbacks directly cancel records
697 * and delete llog. Check it and stop
699 if (loghandle->lgh_hdr == NULL ||
700 loghandle->lgh_hdr->llh_count == 1)
703 /* exit if the last index is reached */
704 if (index >= last_index)
711 CDEBUG(D_HA, "stop processing %s "DOSTID":%x index %d count %d\n",
712 ((llh->llh_flags & LLOG_F_IS_CAT) ? "catalog" : "plain"),
713 POSTID(&loghandle->lgh_id.lgl_oi), loghandle->lgh_id.lgl_ogen,
714 index, llh->llh_count);
717 cd->lpcd_last_idx = last_called_index;
719 if (unlikely(rc == -EIO && loghandle->lgh_obj != NULL)) {
720 if (dt_object_remote(loghandle->lgh_obj)) {
721 /* If it is remote object, then -EIO might means
722 * disconnection or eviction, let's return -EAGAIN,
723 * so for update recovery log processing, it will
724 * retry until the umount or abort recovery, see
725 * lod_sub_recovery_thread() */
726 CERROR("%s retry remote llog process\n",
727 loghandle->lgh_ctxt->loc_obd->obd_name);
730 /* something bad happened to the processing of a local
731 * llog file, probably I/O error or the log got
732 * corrupted to be able to finally release the log we
733 * discard any remaining bits in the header */
734 CERROR("%s: Local llog found corrupted #"DOSTID":%x"
735 " %s index %d count %d\n",
736 loghandle->lgh_ctxt->loc_obd->obd_name,
737 POSTID(&loghandle->lgh_id.lgl_oi),
738 loghandle->lgh_id.lgl_ogen,
739 ((llh->llh_flags & LLOG_F_IS_CAT) ? "catalog" :
740 "plain"), index, llh->llh_count);
742 while (index <= last_index) {
743 if (test_bit_le(index,
744 LLOG_HDR_BITMAP(llh)) != 0)
745 llog_cancel_rec(lpi->lpi_env, loghandle,
753 OBD_FREE_LARGE(buf, chunk_size);
758 static int llog_process_thread_daemonize(void *arg)
760 struct llog_process_info *lpi = arg;
763 struct nsproxy *new_ns, *curr_ns = current->nsproxy;
765 task_lock(lpi->lpi_reftask);
766 new_ns = lpi->lpi_reftask->nsproxy;
767 if (curr_ns != new_ns) {
770 current->nsproxy = new_ns;
771 /* XXX: we should call put_nsproxy() instead of
772 * atomic_dec(&ns->count) directly. But put_nsproxy() cannot be
773 * used outside of the kernel itself, because it calls
774 * free_nsproxy() which is not exported by the kernel
775 * (defined in kernel/nsproxy.c) */
777 atomic_dec(&curr_ns->count);
779 task_unlock(lpi->lpi_reftask);
783 /* client env has no keys, tags is just 0 */
784 rc = lu_env_init(&env, LCT_LOCAL | LCT_MG_THREAD);
789 rc = llog_process_thread(arg);
793 complete(&lpi->lpi_completion);
797 int llog_process_or_fork(const struct lu_env *env,
798 struct llog_handle *loghandle,
799 llog_cb_t cb, void *data, void *catdata, bool fork)
801 struct llog_process_info *lpi;
802 struct llog_process_data *d = data;
803 struct llog_process_cat_data *cd = catdata;
810 CERROR("cannot alloc pointer\n");
813 lpi->lpi_loghandle = loghandle;
815 lpi->lpi_cbdata = data;
816 lpi->lpi_catdata = catdata;
818 CDEBUG(D_OTHER, "Processing "DFID" flags 0x%03x startcat %d startidx %d first_idx %d last_idx %d\n",
819 PFID(&loghandle->lgh_id.lgl_oi.oi_fid),
820 loghandle->lgh_hdr->llh_flags, d ? d->lpd_startcat : -1,
821 d ? d->lpd_startidx : -1, cd ? cd->lpcd_first_idx : -1,
822 cd ? cd->lpcd_last_idx : -1);
824 struct task_struct *task;
826 /* The new thread can't use parent env,
827 * init the new one in llog_process_thread_daemonize. */
829 init_completion(&lpi->lpi_completion);
830 /* take reference to current, so that
831 * llog_process_thread_daemonize() can use it to switch to
832 * namespace associated with current */
833 lpi->lpi_reftask = current;
834 task = kthread_run(llog_process_thread_daemonize, lpi,
835 "llog_process_thread");
838 CERROR("%s: cannot start thread: rc = %d\n",
839 loghandle->lgh_ctxt->loc_obd->obd_name, rc);
842 wait_for_completion(&lpi->lpi_completion);
845 llog_process_thread(lpi);
853 EXPORT_SYMBOL(llog_process_or_fork);
855 int llog_process(const struct lu_env *env, struct llog_handle *loghandle,
856 llog_cb_t cb, void *data, void *catdata)
859 rc = llog_process_or_fork(env, loghandle, cb, data, catdata, true);
860 return rc == LLOG_DEL_PLAIN ? 0 : rc;
862 EXPORT_SYMBOL(llog_process);
864 static inline const struct cred *llog_raise_resource(void)
866 struct cred *cred = NULL;
868 if (cap_raised(current_cap(), CAP_SYS_RESOURCE))
871 cred = prepare_creds();
875 cap_raise(cred->cap_effective, CAP_SYS_RESOURCE);
876 return override_creds(cred);
879 static inline void llog_restore_resource(const struct cred *old_cred)
882 revert_creds(old_cred);
885 int llog_reverse_process(const struct lu_env *env,
886 struct llog_handle *loghandle, llog_cb_t cb,
887 void *data, void *catdata)
889 struct llog_log_hdr *llh = loghandle->lgh_hdr;
890 struct llog_process_cat_data *cd = catdata;
892 int rc = 0, first_index = 1, index, idx;
893 __u32 chunk_size = llh->llh_hdr.lrh_len;
896 OBD_ALLOC_LARGE(buf, chunk_size);
901 first_index = cd->lpcd_first_idx + 1;
902 if (cd != NULL && cd->lpcd_last_idx)
903 index = cd->lpcd_last_idx;
905 index = LLOG_HDR_BITMAP_SIZE(llh) - 1;
908 struct llog_rec_hdr *rec;
909 struct llog_rec_tail *tail;
911 /* skip records not set in bitmap */
912 while (index >= first_index &&
913 !test_bit_le(index, LLOG_HDR_BITMAP(llh)))
916 LASSERT(index >= first_index - 1);
917 if (index == first_index - 1)
920 /* get the buf with our target record; avoid old garbage */
921 memset(buf, 0, chunk_size);
922 rc = llog_prev_block(env, loghandle, index, buf, chunk_size);
927 idx = rec->lrh_index;
928 CDEBUG(D_RPCTRACE, "index %u : idx %u\n", index, idx);
929 while (idx < index) {
930 rec = (void *)rec + rec->lrh_len;
931 if (LLOG_REC_HDR_NEEDS_SWABBING(rec))
932 lustre_swab_llog_rec(rec);
935 LASSERT(idx == index);
936 tail = (void *)rec + rec->lrh_len - sizeof(*tail);
938 /* process records in buffer, starting where we found one */
939 while ((void *)tail > buf) {
940 if (tail->lrt_index == 0)
941 GOTO(out, rc = 0); /* no more records */
943 /* if set, process the callback on this record */
944 if (test_bit_le(index, LLOG_HDR_BITMAP(llh))) {
945 rec = (void *)tail - tail->lrt_len +
948 rc = cb(env, loghandle, rec, data);
949 if (rc == LLOG_PROC_BREAK) {
951 } else if (rc == LLOG_DEL_RECORD) {
952 rc = llog_cancel_rec(env, loghandle,
959 /* previous record, still in buffer? */
961 if (index < first_index)
963 tail = (void *)tail - tail->lrt_len;
969 OBD_FREE_LARGE(buf, chunk_size);
972 EXPORT_SYMBOL(llog_reverse_process);
978 * llog_open - open llog, may not exist
979 * llog_exist - check if llog exists
980 * llog_close - close opened llog, pair for open, frees llog_handle
981 * llog_declare_create - declare llog creation
982 * llog_create - create new llog on disk, need transaction handle
983 * llog_declare_write_rec - declaration of llog write
984 * llog_write_rec - write llog record on disk, need transaction handle
985 * llog_declare_add - declare llog catalog record addition
986 * llog_add - add llog record in catalog, need transaction handle
988 int llog_exist(struct llog_handle *loghandle)
990 struct llog_operations *lop;
995 rc = llog_handle2ops(loghandle, &lop);
998 if (lop->lop_exist == NULL)
1001 rc = lop->lop_exist(loghandle);
1004 EXPORT_SYMBOL(llog_exist);
1006 int llog_declare_create(const struct lu_env *env,
1007 struct llog_handle *loghandle, struct thandle *th)
1009 const struct cred *old_cred;
1010 struct llog_operations *lop;
1015 rc = llog_handle2ops(loghandle, &lop);
1018 if (lop->lop_declare_create == NULL)
1019 RETURN(-EOPNOTSUPP);
1021 old_cred = llog_raise_resource();
1022 rc = lop->lop_declare_create(env, loghandle, th);
1023 llog_restore_resource(old_cred);
1027 int llog_create(const struct lu_env *env, struct llog_handle *handle,
1030 const struct cred *old_cred;
1031 struct llog_operations *lop;
1036 rc = llog_handle2ops(handle, &lop);
1039 if (lop->lop_create == NULL)
1040 RETURN(-EOPNOTSUPP);
1042 old_cred = llog_raise_resource();
1043 rc = lop->lop_create(env, handle, th);
1044 llog_restore_resource(old_cred);
1048 int llog_declare_write_rec(const struct lu_env *env,
1049 struct llog_handle *handle,
1050 struct llog_rec_hdr *rec, int idx,
1053 const struct cred *old_cred;
1054 struct llog_operations *lop;
1059 rc = llog_handle2ops(handle, &lop);
1063 if (lop->lop_declare_write_rec == NULL)
1064 RETURN(-EOPNOTSUPP);
1066 old_cred = llog_raise_resource();
1067 rc = lop->lop_declare_write_rec(env, handle, rec, idx, th);
1068 llog_restore_resource(old_cred);
1072 int llog_write_rec(const struct lu_env *env, struct llog_handle *handle,
1073 struct llog_rec_hdr *rec, struct llog_cookie *logcookies,
1074 int idx, struct thandle *th)
1076 const struct cred *old_cred;
1077 struct llog_operations *lop;
1082 /* API sanity checks */
1083 if (handle == NULL) {
1084 CERROR("loghandle is missed\n");
1086 } else if (handle->lgh_obj == NULL) {
1087 CERROR("loghandle %p with NULL object\n",
1090 } else if (th == NULL) {
1091 CERROR("%s: missed transaction handle\n",
1092 handle->lgh_obj->do_lu.lo_dev->ld_obd->obd_name);
1094 } else if (handle->lgh_hdr == NULL) {
1095 CERROR("%s: loghandle %p with no header\n",
1096 handle->lgh_obj->do_lu.lo_dev->ld_obd->obd_name,
1101 rc = llog_handle2ops(handle, &lop);
1105 if (lop->lop_write_rec == NULL)
1106 RETURN(-EOPNOTSUPP);
1108 buflen = rec->lrh_len;
1109 LASSERT(cfs_size_round(buflen) == buflen);
1111 old_cred = llog_raise_resource();
1112 rc = lop->lop_write_rec(env, handle, rec, logcookies, idx, th);
1113 llog_restore_resource(old_cred);
1117 int llog_add(const struct lu_env *env, struct llog_handle *lgh,
1118 struct llog_rec_hdr *rec, struct llog_cookie *logcookies,
1121 const struct cred *old_cred;
1126 if (lgh->lgh_logops->lop_add == NULL)
1127 RETURN(-EOPNOTSUPP);
1129 old_cred = llog_raise_resource();
1130 rc = lgh->lgh_logops->lop_add(env, lgh, rec, logcookies, th);
1131 llog_restore_resource(old_cred);
1134 EXPORT_SYMBOL(llog_add);
1136 int llog_declare_add(const struct lu_env *env, struct llog_handle *lgh,
1137 struct llog_rec_hdr *rec, struct thandle *th)
1139 const struct cred *old_cred;
1144 if (lgh->lgh_logops->lop_declare_add == NULL)
1145 RETURN(-EOPNOTSUPP);
1147 old_cred = llog_raise_resource();
1148 rc = lgh->lgh_logops->lop_declare_add(env, lgh, rec, th);
1149 llog_restore_resource(old_cred);
1152 EXPORT_SYMBOL(llog_declare_add);
1155 * Helper function to open llog or create it if doesn't exist.
1156 * It hides all transaction handling from caller.
1158 int llog_open_create(const struct lu_env *env, struct llog_ctxt *ctxt,
1159 struct llog_handle **res, struct llog_logid *logid,
1162 struct dt_device *d;
1168 rc = llog_open(env, ctxt, res, logid, name, LLOG_OPEN_NEW);
1172 if (llog_exist(*res))
1175 LASSERT((*res)->lgh_obj != NULL);
1177 d = lu2dt_dev((*res)->lgh_obj->do_lu.lo_dev);
1179 if (unlikely(unlikely(d->dd_rdonly)))
1182 th = dt_trans_create(env, d);
1184 GOTO(out, rc = PTR_ERR(th));
1186 /* Create update llog object synchronously, which
1187 * happens during inialization process see
1188 * lod_sub_prep_llog(), to make sure the update
1189 * llog object is created before corss-MDT writing
1190 * updates into the llog object */
1191 if (ctxt->loc_flags & LLOG_CTXT_FLAG_NORMAL_FID)
1194 th->th_wait_submit = 1;
1195 rc = llog_declare_create(env, *res, th);
1197 rc = dt_trans_start_local(env, d, th);
1199 rc = llog_create(env, *res, th);
1201 dt_trans_stop(env, d, th);
1204 llog_close(env, *res);
1207 EXPORT_SYMBOL(llog_open_create);
1210 * Helper function to delete existent llog.
1212 int llog_erase(const struct lu_env *env, struct llog_ctxt *ctxt,
1213 struct llog_logid *logid, char *name)
1215 struct llog_handle *handle;
1220 /* nothing to erase */
1221 if (name == NULL && logid == NULL)
1224 rc = llog_open(env, ctxt, &handle, logid, name, LLOG_OPEN_EXISTS);
1228 rc = llog_init_handle(env, handle, LLOG_F_IS_PLAIN, NULL);
1230 rc = llog_destroy(env, handle);
1232 rc2 = llog_close(env, handle);
1237 EXPORT_SYMBOL(llog_erase);
1240 * Helper function for write record in llog.
1241 * It hides all transaction handling from caller.
1242 * Valid only with local llog.
1244 int llog_write(const struct lu_env *env, struct llog_handle *loghandle,
1245 struct llog_rec_hdr *rec, int idx)
1247 struct dt_device *dt;
1255 LASSERT(loghandle->lgh_ctxt);
1256 LASSERT(loghandle->lgh_obj != NULL);
1258 dt = lu2dt_dev(loghandle->lgh_obj->do_lu.lo_dev);
1260 if (unlikely(unlikely(dt->dd_rdonly)))
1263 th = dt_trans_create(env, dt);
1265 RETURN(PTR_ERR(th));
1267 rc = llog_declare_write_rec(env, loghandle, rec, idx, th);
1269 GOTO(out_trans, rc);
1271 th->th_wait_submit = 1;
1272 rc = dt_trans_start_local(env, dt, th);
1274 GOTO(out_trans, rc);
1276 need_cookie = !(idx == LLOG_HEADER_IDX || idx == LLOG_NEXT_IDX);
1278 down_write(&loghandle->lgh_lock);
1280 struct llog_thread_info *lti = llog_info(env);
1282 /* cookie comes from llog_process_thread */
1283 rc = llog_write_rec(env, loghandle, rec, <i->lgi_cookie,
1284 rec->lrh_index, th);
1285 /* upper layer didn`t pass cookie so change rc */
1286 rc = (rc == 1 ? 0 : rc);
1288 rc = llog_write_rec(env, loghandle, rec, NULL, idx, th);
1291 up_write(&loghandle->lgh_lock);
1293 dt_trans_stop(env, dt, th);
1296 EXPORT_SYMBOL(llog_write);
1298 int llog_open(const struct lu_env *env, struct llog_ctxt *ctxt,
1299 struct llog_handle **lgh, struct llog_logid *logid,
1300 char *name, enum llog_open_param open_param)
1302 const struct cred *old_cred;
1308 LASSERT(ctxt->loc_logops);
1310 if (ctxt->loc_logops->lop_open == NULL) {
1312 RETURN(-EOPNOTSUPP);
1315 *lgh = llog_alloc_handle();
1318 (*lgh)->lgh_ctxt = ctxt;
1319 (*lgh)->lgh_logops = ctxt->loc_logops;
1321 old_cred = llog_raise_resource();
1322 rc = ctxt->loc_logops->lop_open(env, *lgh, logid, name, open_param);
1323 llog_restore_resource(old_cred);
1325 llog_free_handle(*lgh);
1330 EXPORT_SYMBOL(llog_open);
1332 int llog_close(const struct lu_env *env, struct llog_handle *loghandle)
1334 return llog_handle_put(env, loghandle);
1336 EXPORT_SYMBOL(llog_close);
1339 * Helper function to get the llog size in records. It is used by MGS
1340 * mostly to check that config llog exists and contains data.
1342 * \param[in] env execution environment
1343 * \param[in] ctxt llog context
1344 * \param[in] name llog name
1346 * \retval true if there are records in llog besides a header
1347 * \retval false on error or llog without records
1349 int llog_is_empty(const struct lu_env *env, struct llog_ctxt *ctxt,
1352 struct llog_handle *llh;
1355 rc = llog_open(env, ctxt, &llh, NULL, name, LLOG_OPEN_EXISTS);
1357 if (likely(rc == -ENOENT))
1362 rc = llog_init_handle(env, llh, LLOG_F_IS_PLAIN, NULL);
1364 GOTO(out_close, rc);
1365 rc = llog_get_size(llh);
1368 llog_close(env, llh);
1370 /* The header is record 1, the llog is still considered as empty
1371 * if there is only header */
1374 EXPORT_SYMBOL(llog_is_empty);
1376 int llog_copy_handler(const struct lu_env *env, struct llog_handle *llh,
1377 struct llog_rec_hdr *rec, void *data)
1379 struct llog_handle *copy_llh = data;
1381 /* Append all records */
1382 return llog_write(env, copy_llh, rec, LLOG_NEXT_IDX);
1385 /* backup plain llog */
1386 int llog_backup(const struct lu_env *env, struct obd_device *obd,
1387 struct llog_ctxt *ctxt, struct llog_ctxt *bctxt,
1388 char *name, char *backup)
1390 struct llog_handle *llh, *bllh;
1395 /* open original log */
1396 rc = llog_open(env, ctxt, &llh, NULL, name, LLOG_OPEN_EXISTS);
1398 /* the -ENOENT case is also reported to the caller
1399 * but silently so it should handle that if needed.
1402 CERROR("%s: failed to open log %s: rc = %d\n",
1403 obd->obd_name, name, rc);
1407 rc = llog_init_handle(env, llh, LLOG_F_IS_PLAIN, NULL);
1409 GOTO(out_close, rc);
1411 /* Make sure there's no old backup log */
1412 rc = llog_erase(env, bctxt, NULL, backup);
1413 if (rc < 0 && rc != -ENOENT)
1414 GOTO(out_close, rc);
1416 /* open backup log */
1417 rc = llog_open_create(env, bctxt, &bllh, NULL, backup);
1419 CERROR("%s: failed to open backup logfile %s: rc = %d\n",
1420 obd->obd_name, backup, rc);
1421 GOTO(out_close, rc);
1424 /* check that backup llog is not the same object as original one */
1425 if (llh->lgh_obj == bllh->lgh_obj) {
1426 CERROR("%s: backup llog %s to itself (%s), objects %p/%p\n",
1427 obd->obd_name, name, backup, llh->lgh_obj,
1429 GOTO(out_backup, rc = -EEXIST);
1432 rc = llog_init_handle(env, bllh, LLOG_F_IS_PLAIN, NULL);
1434 GOTO(out_backup, rc);
1436 /* Copy log record by record */
1437 rc = llog_process_or_fork(env, llh, llog_copy_handler, (void *)bllh,
1440 CERROR("%s: failed to backup log %s: rc = %d\n",
1441 obd->obd_name, name, rc);
1443 llog_close(env, bllh);
1445 llog_close(env, llh);
1448 EXPORT_SYMBOL(llog_backup);
1450 /* Get size of llog */
1451 __u64 llog_size(const struct lu_env *env, struct llog_handle *llh)
1456 rc = llh->lgh_obj->do_ops->do_attr_get(env, llh->lgh_obj, &la);
1458 CERROR("%s: attr_get failed for "DFID": rc = %d\n",
1459 llh->lgh_ctxt->loc_obd->obd_name,
1460 PFID(&llh->lgh_id.lgl_oi.oi_fid), rc);
1466 EXPORT_SYMBOL(llog_size);