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_cat.c
34 * OST<->MDS recovery logging infrastructure.
36 * Invariants in implementation:
37 * - we do not share logs among different OST<->MDS connections, so that
38 * if an OST or MDS fails it need only look at log(s) relevant to itself
40 * Author: Andreas Dilger <adilger@clusterfs.com>
41 * Author: Alexey Zhuravlev <alexey.zhuravlev@intel.com>
42 * Author: Mikhail Pershin <mike.pershin@intel.com>
45 #define DEBUG_SUBSYSTEM S_LOG
48 #include <obd_class.h>
50 #include "llog_internal.h"
52 /* Create a new log handle and add it to the open list.
53 * This log handle will be closed when all of the records in it are removed.
55 * Assumes caller has already pushed us into the kernel context and is locking.
57 static int llog_cat_new_log(const struct lu_env *env,
58 struct llog_handle *cathandle,
59 struct llog_handle *loghandle,
62 struct llog_thread_info *lgi = llog_info(env);
63 struct llog_logid_rec *rec = &lgi->lgi_logid;
64 struct thandle *handle = NULL;
65 struct dt_device *dt = NULL;
66 struct llog_log_hdr *llh = cathandle->lgh_hdr;
71 index = (cathandle->lgh_last_idx + 1) %
72 (OBD_FAIL_PRECHECK(OBD_FAIL_CAT_RECORDS) ? (cfs_fail_val + 1) :
73 LLOG_HDR_BITMAP_SIZE(llh));
75 /* check that new llog index will not overlap with the first one.
76 * - llh_cat_idx is the index just before the first/oldest still in-use
78 * - lgh_last_idx is the last/newest used index in catalog
80 * When catalog is not wrapped yet then lgh_last_idx is always larger
81 * than llh_cat_idx. After the wrap around lgh_last_idx re-starts
82 * from 0 and llh_cat_idx becomes the upper limit for it
84 * Check if catalog has already wrapped around or not by comparing
85 * last_idx and cat_idx */
86 if ((index == llh->llh_cat_idx + 1 && llh->llh_count > 1) ||
87 (index == 0 && llh->llh_cat_idx == 0)) {
88 if (cathandle->lgh_name == NULL) {
89 CWARN("%s: there are no more free slots in catalog "
91 loghandle->lgh_ctxt->loc_obd->obd_name,
92 PFID(&cathandle->lgh_id.lgl_oi.oi_fid),
93 cathandle->lgh_id.lgl_ogen);
95 CWARN("%s: there are no more free slots in "
97 loghandle->lgh_ctxt->loc_obd->obd_name,
103 if (OBD_FAIL_CHECK(OBD_FAIL_MDS_LLOG_CREATE_FAILED))
106 if (loghandle->lgh_hdr != NULL) {
107 /* If llog object is remote and creation is failed, lgh_hdr
108 * might be left over here, free it first */
109 LASSERT(!llog_exist(loghandle));
110 OBD_FREE_LARGE(loghandle->lgh_hdr, loghandle->lgh_hdr_size);
111 loghandle->lgh_hdr = NULL;
115 dt = lu2dt_dev(cathandle->lgh_obj->do_lu.lo_dev);
117 handle = dt_trans_create(env, dt);
119 RETURN(PTR_ERR(handle));
121 /* Create update llog object synchronously, which
122 * happens during inialization process see
123 * lod_sub_prep_llog(), to make sure the update
124 * llog object is created before corss-MDT writing
125 * updates into the llog object */
126 if (cathandle->lgh_ctxt->loc_flags & LLOG_CTXT_FLAG_NORMAL_FID)
129 handle->th_wait_submit = 1;
131 rc = llog_declare_create(env, loghandle, handle);
135 rec->lid_hdr.lrh_len = sizeof(*rec);
136 rec->lid_hdr.lrh_type = LLOG_LOGID_MAGIC;
137 rec->lid_id = loghandle->lgh_id;
138 rc = llog_declare_write_rec(env, cathandle, &rec->lid_hdr, -1,
143 rc = dt_trans_start_local(env, dt, handle);
150 rc = llog_create(env, loghandle, th);
151 /* if llog is already created, no need to initialize it */
154 } else if (rc != 0) {
155 CERROR("%s: can't create new plain llog in catalog: rc = %d\n",
156 loghandle->lgh_ctxt->loc_obd->obd_name, rc);
160 rc = llog_init_handle(env, loghandle,
161 LLOG_F_IS_PLAIN | LLOG_F_ZAP_WHEN_EMPTY,
162 &cathandle->lgh_hdr->llh_tgtuuid);
166 /* build the record for this log in the catalog */
167 rec->lid_hdr.lrh_len = sizeof(*rec);
168 rec->lid_hdr.lrh_type = LLOG_LOGID_MAGIC;
169 rec->lid_id = loghandle->lgh_id;
171 /* append the new record into catalog. The new index will be
172 * assigned to the record and updated in rec header */
173 rc = llog_write_rec(env, cathandle, &rec->lid_hdr,
174 &loghandle->u.phd.phd_cookie, LLOG_NEXT_IDX, th);
176 GOTO(out_destroy, rc);
178 CDEBUG(D_OTHER, "new plain log "DFID".%u of catalog "DFID"\n",
179 PFID(&loghandle->lgh_id.lgl_oi.oi_fid), rec->lid_hdr.lrh_index,
180 PFID(&cathandle->lgh_id.lgl_oi.oi_fid));
182 loghandle->lgh_hdr->llh_cat_idx = rec->lid_hdr.lrh_index;
184 /* limit max size of plain llog so that space can be
185 * released sooner, especially on small filesystems */
186 /* 2MB for the cases when free space hasn't been learned yet */
187 loghandle->lgh_max_size = 2 << 20;
188 dt = lu2dt_dev(cathandle->lgh_obj->do_lu.lo_dev);
189 rc = dt_statfs(env, dt, &lgi->lgi_statfs);
190 if (rc == 0 && lgi->lgi_statfs.os_bfree > 0) {
191 __u64 freespace = (lgi->lgi_statfs.os_bfree *
192 lgi->lgi_statfs.os_bsize) >> 6;
193 if (freespace < loghandle->lgh_max_size)
194 loghandle->lgh_max_size = freespace;
195 /* shouldn't be > 128MB in any case?
196 * it's 256K records of 512 bytes each */
197 if (freespace > (128 << 20))
198 loghandle->lgh_max_size = 128 << 20;
203 if (handle != NULL) {
204 handle->th_result = rc >= 0 ? 0 : rc;
205 dt_trans_stop(env, dt, handle);
210 /* to signal llog_cat_close() it shouldn't try to destroy the llog,
211 * we want to destroy it in this transaction, otherwise the object
212 * becomes an orphan */
213 loghandle->lgh_hdr->llh_flags &= ~LLOG_F_ZAP_WHEN_EMPTY;
214 /* this is to mimic full log, so another llog_cat_current_log()
215 * can skip it and ask for another onet */
216 loghandle->lgh_last_idx = LLOG_HDR_BITMAP_SIZE(llh) + 1;
217 llog_trans_destroy(env, loghandle, th);
221 /* Open an existent log handle and add it to the open list.
222 * This log handle will be closed when all of the records in it are removed.
224 * Assumes caller has already pushed us into the kernel context and is locking.
225 * We return a lock on the handle to ensure nobody yanks it from us.
227 * This takes extra reference on llog_handle via llog_handle_get() and require
228 * this reference to be put by caller using llog_handle_put()
230 int llog_cat_id2handle(const struct lu_env *env, struct llog_handle *cathandle,
231 struct llog_handle **res, struct llog_logid *logid)
233 struct llog_handle *loghandle;
239 if (cathandle == NULL)
242 fmt = cathandle->lgh_hdr->llh_flags & LLOG_F_EXT_MASK;
243 down_write(&cathandle->lgh_lock);
244 list_for_each_entry(loghandle, &cathandle->u.chd.chd_head,
246 struct llog_logid *cgl = &loghandle->lgh_id;
248 if (ostid_id(&cgl->lgl_oi) == ostid_id(&logid->lgl_oi) &&
249 ostid_seq(&cgl->lgl_oi) == ostid_seq(&logid->lgl_oi)) {
250 if (cgl->lgl_ogen != logid->lgl_ogen) {
251 CWARN("%s: log "DFID" generation %x != %x\n",
252 loghandle->lgh_ctxt->loc_obd->obd_name,
253 PFID(&logid->lgl_oi.oi_fid),
254 cgl->lgl_ogen, logid->lgl_ogen);
257 loghandle->u.phd.phd_cat_handle = cathandle;
258 up_write(&cathandle->lgh_lock);
262 up_write(&cathandle->lgh_lock);
264 rc = llog_open(env, cathandle->lgh_ctxt, &loghandle, logid, NULL,
267 CERROR("%s: error opening log id "DFID":%x: rc = %d\n",
268 cathandle->lgh_ctxt->loc_obd->obd_name,
269 PFID(&logid->lgl_oi.oi_fid), logid->lgl_ogen, rc);
273 rc = llog_init_handle(env, loghandle, LLOG_F_IS_PLAIN | fmt, NULL);
275 llog_close(env, loghandle);
280 down_write(&cathandle->lgh_lock);
281 list_add_tail(&loghandle->u.phd.phd_entry, &cathandle->u.chd.chd_head);
282 up_write(&cathandle->lgh_lock);
284 loghandle->u.phd.phd_cat_handle = cathandle;
285 loghandle->u.phd.phd_cookie.lgc_lgl = cathandle->lgh_id;
286 loghandle->u.phd.phd_cookie.lgc_index =
287 loghandle->lgh_hdr->llh_cat_idx;
290 llog_handle_get(loghandle);
295 int llog_cat_close(const struct lu_env *env, struct llog_handle *cathandle)
297 struct llog_handle *loghandle, *n;
302 list_for_each_entry_safe(loghandle, n, &cathandle->u.chd.chd_head,
304 struct llog_log_hdr *llh = loghandle->lgh_hdr;
307 /* unlink open-not-created llogs */
308 list_del_init(&loghandle->u.phd.phd_entry);
309 llh = loghandle->lgh_hdr;
310 if (loghandle->lgh_obj != NULL && llh != NULL &&
311 (llh->llh_flags & LLOG_F_ZAP_WHEN_EMPTY) &&
312 (llh->llh_count == 1)) {
313 rc = llog_destroy(env, loghandle);
315 CERROR("%s: failure destroying log during "
316 "cleanup: rc = %d\n",
317 loghandle->lgh_ctxt->loc_obd->obd_name,
320 index = loghandle->u.phd.phd_cookie.lgc_index;
321 llog_cat_cleanup(env, cathandle, NULL, index);
323 llog_close(env, loghandle);
325 /* if handle was stored in ctxt, remove it too */
326 if (cathandle->lgh_ctxt->loc_handle == cathandle)
327 cathandle->lgh_ctxt->loc_handle = NULL;
328 rc = llog_close(env, cathandle);
331 EXPORT_SYMBOL(llog_cat_close);
334 * lockdep markers for nested struct llog_handle::lgh_lock locking.
341 /** Return the currently active log handle. If the current log handle doesn't
342 * have enough space left for the current record, start a new one.
344 * If reclen is 0, we only want to know what the currently active log is,
345 * otherwise we get a lock on this log so nobody can steal our space.
347 * Assumes caller has already pushed us into the kernel context and is locking.
349 * NOTE: loghandle is write-locked upon successful return
351 static struct llog_handle *llog_cat_current_log(struct llog_handle *cathandle,
354 struct llog_handle *loghandle = NULL;
358 if (OBD_FAIL_CHECK(OBD_FAIL_MDS_LLOG_CREATE_FAILED2)) {
359 down_write_nested(&cathandle->lgh_lock, LLOGH_CAT);
360 GOTO(next, loghandle);
363 down_read_nested(&cathandle->lgh_lock, LLOGH_CAT);
364 loghandle = cathandle->u.chd.chd_current_log;
366 struct llog_log_hdr *llh;
368 down_write_nested(&loghandle->lgh_lock, LLOGH_LOG);
369 llh = loghandle->lgh_hdr;
370 if (llh == NULL || !llog_is_full(loghandle)) {
371 up_read(&cathandle->lgh_lock);
374 up_write(&loghandle->lgh_lock);
377 up_read(&cathandle->lgh_lock);
379 /* time to use next log */
381 /* first, we have to make sure the state hasn't changed */
382 down_write_nested(&cathandle->lgh_lock, LLOGH_CAT);
383 loghandle = cathandle->u.chd.chd_current_log;
385 struct llog_log_hdr *llh;
387 down_write_nested(&loghandle->lgh_lock, LLOGH_LOG);
388 llh = loghandle->lgh_hdr;
389 if (llh == NULL || !llog_is_full(loghandle))
390 GOTO(out_unlock, loghandle);
392 up_write(&loghandle->lgh_lock);
396 /* Sigh, the chd_next_log and chd_current_log is initialized
397 * in declare phase, and we do not serialize the catlog
398 * accessing, so it might be possible the llog creation
399 * thread (see llog_cat_declare_add_rec()) did not create
400 * llog successfully, then the following thread might
401 * meet this situation. */
402 if (IS_ERR_OR_NULL(cathandle->u.chd.chd_next_log)) {
403 CERROR("%s: next log does not exist!\n",
404 cathandle->lgh_ctxt->loc_obd->obd_name);
405 loghandle = ERR_PTR(-EIO);
406 if (cathandle->u.chd.chd_next_log == NULL) {
407 /* Store the error in chd_next_log, so
408 * the following process can get correct
410 cathandle->u.chd.chd_next_log = loghandle;
412 GOTO(out_unlock, loghandle);
415 CDEBUG(D_INODE, "use next log\n");
417 loghandle = cathandle->u.chd.chd_next_log;
418 cathandle->u.chd.chd_current_log = loghandle;
419 cathandle->u.chd.chd_next_log = NULL;
420 down_write_nested(&loghandle->lgh_lock, LLOGH_LOG);
423 up_write(&cathandle->lgh_lock);
428 static int llog_cat_update_header(const struct lu_env *env,
429 struct llog_handle *cathandle)
431 struct llog_handle *loghandle;
436 down_write(&cathandle->lgh_lock);
437 if (!cathandle->lgh_stale) {
438 up_write(&cathandle->lgh_lock);
441 list_for_each_entry(loghandle, &cathandle->u.chd.chd_head,
443 if (!llog_exist(loghandle))
446 rc = llog_read_header(env, loghandle, NULL);
448 up_write(&cathandle->lgh_lock);
452 rc = llog_read_header(env, cathandle, NULL);
454 cathandle->lgh_stale = 0;
455 up_write(&cathandle->lgh_lock);
462 /* Add a single record to the recovery log(s) using a catalog
463 * Returns as llog_write_record
465 * Assumes caller has already pushed us into the kernel context.
467 int llog_cat_add_rec(const struct lu_env *env, struct llog_handle *cathandle,
468 struct llog_rec_hdr *rec, struct llog_cookie *reccookie,
471 struct llog_handle *loghandle;
475 LASSERT(rec->lrh_len <= cathandle->lgh_ctxt->loc_chunk_size);
478 loghandle = llog_cat_current_log(cathandle, th);
479 if (IS_ERR(loghandle))
480 RETURN(PTR_ERR(loghandle));
482 /* loghandle is already locked by llog_cat_current_log() for us */
483 if (!llog_exist(loghandle)) {
484 rc = llog_cat_new_log(env, cathandle, loghandle, th);
486 up_write(&loghandle->lgh_lock);
487 /* nobody should be trying to use this llog */
488 down_write(&cathandle->lgh_lock);
489 /* only reset current log if still room in catalog, to
490 * avoid unnecessarily and racy creation of new and
491 * partially initialized llog_handle
493 if ((cathandle->u.chd.chd_current_log == loghandle) &&
495 cathandle->u.chd.chd_current_log = NULL;
496 up_write(&cathandle->lgh_lock);
500 /* now let's try to add the record */
501 rc = llog_write_rec(env, loghandle, rec, reccookie, LLOG_NEXT_IDX, th);
503 CDEBUG_LIMIT(rc == -ENOSPC ? D_HA : D_ERROR,
504 "llog_write_rec %d: lh=%p\n", rc, loghandle);
505 /* -ENOSPC is returned if no empty records left
506 * and when it's lack of space on the stogage.
507 * there is no point to try again if it's the second
508 * case. many callers (like llog test) expect ENOSPC,
509 * so we preserve this error code, but look for the
510 * actual cause here */
511 if (rc == -ENOSPC && llog_is_full(loghandle))
514 up_write(&loghandle->lgh_lock);
516 if (rc == -ENOBUFS) {
519 CERROR("%s: error on 2nd llog: rc = %d\n",
520 cathandle->lgh_ctxt->loc_obd->obd_name, rc);
525 EXPORT_SYMBOL(llog_cat_add_rec);
527 int llog_cat_declare_add_rec(const struct lu_env *env,
528 struct llog_handle *cathandle,
529 struct llog_rec_hdr *rec, struct thandle *th)
531 struct llog_thread_info *lgi = llog_info(env);
532 struct llog_logid_rec *lirec = &lgi->lgi_logid;
533 struct llog_handle *loghandle, *next;
538 if (cathandle->u.chd.chd_current_log == NULL) {
539 /* declare new plain llog */
540 down_write(&cathandle->lgh_lock);
541 if (cathandle->u.chd.chd_current_log == NULL) {
542 rc = llog_open(env, cathandle->lgh_ctxt, &loghandle,
543 NULL, NULL, LLOG_OPEN_NEW);
545 cathandle->u.chd.chd_current_log = loghandle;
546 list_add_tail(&loghandle->u.phd.phd_entry,
547 &cathandle->u.chd.chd_head);
550 up_write(&cathandle->lgh_lock);
551 } else if (cathandle->u.chd.chd_next_log == NULL ||
552 IS_ERR(cathandle->u.chd.chd_next_log)) {
553 /* declare next plain llog */
554 down_write(&cathandle->lgh_lock);
555 if (cathandle->u.chd.chd_next_log == NULL ||
556 IS_ERR(cathandle->u.chd.chd_next_log)) {
557 rc = llog_open(env, cathandle->lgh_ctxt, &loghandle,
558 NULL, NULL, LLOG_OPEN_NEW);
560 cathandle->u.chd.chd_next_log = loghandle;
561 list_add_tail(&loghandle->u.phd.phd_entry,
562 &cathandle->u.chd.chd_head);
565 up_write(&cathandle->lgh_lock);
570 lirec->lid_hdr.lrh_len = sizeof(*lirec);
572 if (!llog_exist(cathandle->u.chd.chd_current_log)) {
573 if (dt_object_remote(cathandle->lgh_obj)) {
574 /* For remote operation, if we put the llog object
575 * creation in the current transaction, then the
576 * llog object will not be created on the remote
577 * target until the transaction stop, if other
578 * operations start before the transaction stop,
579 * and use the same llog object, will be dependent
580 * on the success of this transaction. So let's
581 * create the llog object synchronously here to
582 * remove the dependency. */
584 down_read_nested(&cathandle->lgh_lock, LLOGH_CAT);
585 loghandle = cathandle->u.chd.chd_current_log;
586 down_write_nested(&loghandle->lgh_lock, LLOGH_LOG);
587 if (cathandle->lgh_stale) {
588 up_write(&loghandle->lgh_lock);
589 up_read(&cathandle->lgh_lock);
590 GOTO(out, rc = -EIO);
592 if (!llog_exist(loghandle)) {
593 rc = llog_cat_new_log(env, cathandle, loghandle,
596 cathandle->lgh_stale = 1;
598 up_write(&loghandle->lgh_lock);
599 up_read(&cathandle->lgh_lock);
601 rc = llog_cat_update_header(env, cathandle);
609 rc = llog_declare_create(env,
610 cathandle->u.chd.chd_current_log, th);
613 llog_declare_write_rec(env, cathandle,
614 &lirec->lid_hdr, -1, th);
619 /* declare records in the llogs */
620 rc = llog_declare_write_rec(env, cathandle->u.chd.chd_current_log,
623 down_write(&cathandle->lgh_lock);
624 if (cathandle->lgh_stale) {
625 up_write(&cathandle->lgh_lock);
626 GOTO(out, rc = -EIO);
629 cathandle->lgh_stale = 1;
630 up_write(&cathandle->lgh_lock);
631 rc = llog_cat_update_header(env, cathandle);
639 next = cathandle->u.chd.chd_next_log;
640 if (!IS_ERR_OR_NULL(next)) {
641 if (!llog_exist(next)) {
642 if (dt_object_remote(cathandle->lgh_obj)) {
643 /* For remote operation, if we put the llog
644 * object creation in the current transaction,
645 * then the llog object will not be created on
646 * the remote target until the transaction stop,
647 * if other operations start before the
648 * transaction stop, and use the same llog
649 * object, will be dependent on the success of
650 * this transaction. So let's create the llog
651 * object synchronously here to remove the
653 down_write_nested(&cathandle->lgh_lock,
655 next = cathandle->u.chd.chd_next_log;
656 if (IS_ERR_OR_NULL(next)) {
657 /* Sigh, another thread just tried,
658 * let's fail as well */
659 up_write(&cathandle->lgh_lock);
667 down_write_nested(&next->lgh_lock, LLOGH_LOG);
668 if (!llog_exist(next)) {
669 rc = llog_cat_new_log(env, cathandle,
672 cathandle->u.chd.chd_next_log =
675 up_write(&next->lgh_lock);
676 up_write(&cathandle->lgh_lock);
680 rc = llog_declare_create(env, next, th);
681 llog_declare_write_rec(env, cathandle,
682 &lirec->lid_hdr, -1, th);
685 /* XXX: we hope for declarations made for existing llog
686 * this might be not correct with some backends
687 * where declarations are expected against specific
688 * object like ZFS with full debugging enabled */
689 /*llog_declare_write_rec(env, next, rec, -1, th);*/
694 EXPORT_SYMBOL(llog_cat_declare_add_rec);
696 int llog_cat_add(const struct lu_env *env, struct llog_handle *cathandle,
697 struct llog_rec_hdr *rec, struct llog_cookie *reccookie)
699 struct llog_ctxt *ctxt;
700 struct dt_device *dt;
701 struct thandle *th = NULL;
704 ctxt = cathandle->lgh_ctxt;
706 LASSERT(ctxt->loc_exp);
708 LASSERT(cathandle->lgh_obj != NULL);
709 dt = lu2dt_dev(cathandle->lgh_obj->do_lu.lo_dev);
711 th = dt_trans_create(env, dt);
715 rc = llog_cat_declare_add_rec(env, cathandle, rec, th);
719 rc = dt_trans_start_local(env, dt, th);
722 rc = llog_cat_add_rec(env, cathandle, rec, reccookie, th);
724 dt_trans_stop(env, dt, th);
727 EXPORT_SYMBOL(llog_cat_add);
729 /* For each cookie in the cookie array, we clear the log in-use bit and either:
730 * - the log is empty, so mark it free in the catalog header and delete it
731 * - the log is not empty, just write out the log header
733 * The cookies may be in different log files, so we need to get new logs
736 * Assumes caller has already pushed us into the kernel context.
738 int llog_cat_cancel_records(const struct lu_env *env,
739 struct llog_handle *cathandle, int count,
740 struct llog_cookie *cookies)
742 int i, index, rc = 0, failed = 0;
746 for (i = 0; i < count; i++, cookies++) {
747 struct llog_handle *loghandle;
748 struct llog_logid *lgl = &cookies->lgc_lgl;
751 rc = llog_cat_id2handle(env, cathandle, &loghandle, lgl);
753 CDEBUG(D_HA, "%s: cannot find llog for handle "DFID":%x"
755 cathandle->lgh_ctxt->loc_obd->obd_name,
756 PFID(&lgl->lgl_oi.oi_fid), lgl->lgl_ogen, rc);
761 if ((cathandle->lgh_ctxt->loc_flags &
762 LLOG_CTXT_FLAG_NORMAL_FID) && !llog_exist(loghandle)) {
763 /* For update log, some of loghandles of cathandle
764 * might not exist because remote llog creation might
765 * be failed, so let's skip the record cancellation
766 * for these non-exist llogs.
769 CDEBUG(D_HA, "%s: llog "DFID":%x does not exist"
771 cathandle->lgh_ctxt->loc_obd->obd_name,
772 PFID(&lgl->lgl_oi.oi_fid), lgl->lgl_ogen, lrc);
779 lrc = llog_cancel_rec(env, loghandle, cookies->lgc_index);
780 if (lrc == LLOG_DEL_PLAIN) { /* log has been destroyed */
781 index = loghandle->u.phd.phd_cookie.lgc_index;
782 lrc = llog_cat_cleanup(env, cathandle, loghandle,
786 } else if (lrc == -ENOENT) {
787 if (rc == 0) /* ENOENT shouldn't rewrite any error */
789 } else if (lrc < 0) {
794 llog_handle_put(loghandle);
797 CERROR("%s: fail to cancel %d of %d llog-records: rc = %d\n",
798 cathandle->lgh_ctxt->loc_obd->obd_name, failed, count,
803 EXPORT_SYMBOL(llog_cat_cancel_records);
805 static int llog_cat_process_common(const struct lu_env *env,
806 struct llog_handle *cat_llh,
807 struct llog_rec_hdr *rec,
808 struct llog_handle **llhp)
810 struct llog_logid_rec *lir = container_of(rec, typeof(*lir), lid_hdr);
811 struct llog_log_hdr *hdr;
815 if (rec->lrh_type != le32_to_cpu(LLOG_LOGID_MAGIC)) {
817 CWARN("%s: invalid record in catalog "DFID":%x: rc = %d\n",
818 cat_llh->lgh_ctxt->loc_obd->obd_name,
819 PFID(&cat_llh->lgh_id.lgl_oi.oi_fid),
820 cat_llh->lgh_id.lgl_ogen, rc);
823 CDEBUG(D_HA, "processing log "DFID":%x at index %u of catalog "DFID"\n",
824 PFID(&lir->lid_id.lgl_oi.oi_fid), lir->lid_id.lgl_ogen,
825 le32_to_cpu(rec->lrh_index),
826 PFID(&cat_llh->lgh_id.lgl_oi.oi_fid));
828 rc = llog_cat_id2handle(env, cat_llh, llhp, &lir->lid_id);
830 /* After a server crash, a stub of index record in catlog could
831 * be kept, because plain log destroy + catlog index record
832 * deletion are not atomic. So we end up with an index but no
833 * actual record. Destroy the index and move on. */
834 if (rc == -ENOENT || rc == -ESTALE)
835 rc = LLOG_DEL_RECORD;
837 CWARN("%s: can't find llog handle "DFID":%x: rc = %d\n",
838 cat_llh->lgh_ctxt->loc_obd->obd_name,
839 PFID(&lir->lid_id.lgl_oi.oi_fid),
840 lir->lid_id.lgl_ogen, rc);
845 /* clean old empty llogs, do not consider current llog in use */
846 /* ignore remote (lgh_obj == NULL) llogs */
847 hdr = (*llhp)->lgh_hdr;
848 if ((hdr->llh_flags & LLOG_F_ZAP_WHEN_EMPTY) &&
849 hdr->llh_count == 1 && cat_llh->lgh_obj != NULL &&
850 *llhp != cat_llh->u.chd.chd_current_log) {
851 rc = llog_destroy(env, *llhp);
853 CWARN("%s: can't destroy empty log "DFID": rc = %d\n",
854 (*llhp)->lgh_ctxt->loc_obd->obd_name,
855 PFID(&lir->lid_id.lgl_oi.oi_fid), rc);
862 static int llog_cat_process_cb(const struct lu_env *env,
863 struct llog_handle *cat_llh,
864 struct llog_rec_hdr *rec, void *data)
866 struct llog_process_data *d = data;
867 struct llog_handle *llh = NULL;
871 rc = llog_cat_process_common(env, cat_llh, rec, &llh);
875 if (rec->lrh_index < d->lpd_startcat) {
876 /* Skip processing of the logs until startcat */
878 } else if (d->lpd_startidx > 0) {
879 struct llog_process_cat_data cd;
881 cd.lpcd_first_idx = d->lpd_startidx;
882 cd.lpcd_last_idx = 0;
883 rc = llog_process_or_fork(env, llh, d->lpd_cb, d->lpd_data,
885 /* Continue processing the next log from idx 0 */
888 rc = llog_process_or_fork(env, llh, d->lpd_cb, d->lpd_data,
893 /* The empty plain log was destroyed while processing */
894 if (rc == LLOG_DEL_PLAIN) {
895 rc = llog_cat_cleanup(env, cat_llh, llh,
896 llh->u.phd.phd_cookie.lgc_index);
897 } else if (rc == LLOG_DEL_RECORD) {
898 /* clear wrong catalog entry */
899 rc = llog_cat_cleanup(env, cat_llh, NULL, rec->lrh_index);
903 llog_handle_put(llh);
908 int llog_cat_process_or_fork(const struct lu_env *env,
909 struct llog_handle *cat_llh, llog_cb_t cat_cb,
910 llog_cb_t cb, void *data, int startcat,
911 int startidx, bool fork)
913 struct llog_process_data d;
914 struct llog_log_hdr *llh = cat_llh->lgh_hdr;
918 LASSERT(llh->llh_flags & LLOG_F_IS_CAT);
921 d.lpd_startcat = startcat;
922 d.lpd_startidx = startidx;
924 if (llh->llh_cat_idx >= cat_llh->lgh_last_idx &&
925 llh->llh_count > 1) {
926 struct llog_process_cat_data cd;
928 CWARN("%s: catlog "DFID" crosses index zero\n",
929 cat_llh->lgh_ctxt->loc_obd->obd_name,
930 PFID(&cat_llh->lgh_id.lgl_oi.oi_fid));
932 cd.lpcd_first_idx = llh->llh_cat_idx;
933 cd.lpcd_last_idx = 0;
934 rc = llog_process_or_fork(env, cat_llh, cat_cb,
939 cd.lpcd_first_idx = 0;
940 cd.lpcd_last_idx = cat_llh->lgh_last_idx;
941 rc = llog_process_or_fork(env, cat_llh, cat_cb,
944 rc = llog_process_or_fork(env, cat_llh, cat_cb,
951 int llog_cat_process(const struct lu_env *env, struct llog_handle *cat_llh,
952 llog_cb_t cb, void *data, int startcat, int startidx)
954 return llog_cat_process_or_fork(env, cat_llh, llog_cat_process_cb,
955 cb, data, startcat, startidx, false);
957 EXPORT_SYMBOL(llog_cat_process);
959 static int llog_cat_size_cb(const struct lu_env *env,
960 struct llog_handle *cat_llh,
961 struct llog_rec_hdr *rec, void *data)
963 struct llog_process_data *d = data;
964 struct llog_handle *llh = NULL;
965 __u64 *cum_size = d->lpd_data;
970 rc = llog_cat_process_common(env, cat_llh, rec, &llh);
972 if (rc == LLOG_DEL_PLAIN) {
973 /* empty log was deleted, don't count it */
974 rc = llog_cat_cleanup(env, cat_llh, llh,
975 llh->u.phd.phd_cookie.lgc_index);
976 } else if (rc == LLOG_DEL_RECORD) {
977 /* clear wrong catalog entry */
978 rc = llog_cat_cleanup(env, cat_llh, NULL, rec->lrh_index);
980 size = llog_size(env, llh);
983 CDEBUG(D_INFO, "Add llog entry "DFID" size=%llu, tot=%llu\n",
984 PFID(&llh->lgh_id.lgl_oi.oi_fid), size, *cum_size);
988 llog_handle_put(llh);
993 __u64 llog_cat_size(const struct lu_env *env, struct llog_handle *cat_llh)
995 __u64 size = llog_size(env, cat_llh);
997 llog_cat_process_or_fork(env, cat_llh, llog_cat_size_cb,
998 NULL, &size, 0, 0, false);
1002 EXPORT_SYMBOL(llog_cat_size);
1004 /* currently returns the number of "free" entries in catalog,
1005 * ie the available entries for a new plain LLOG file creation,
1006 * even if catalog has wrapped
1008 __u32 llog_cat_free_space(struct llog_handle *cat_llh)
1010 /* simulate almost full Catalog */
1011 if (OBD_FAIL_CHECK(OBD_FAIL_CAT_FREE_RECORDS))
1012 return cfs_fail_val;
1014 if (cat_llh->lgh_hdr->llh_count == 1)
1015 return LLOG_HDR_BITMAP_SIZE(cat_llh->lgh_hdr) - 1;
1017 if (cat_llh->lgh_last_idx > cat_llh->lgh_hdr->llh_cat_idx)
1018 return LLOG_HDR_BITMAP_SIZE(cat_llh->lgh_hdr) - 1 +
1019 cat_llh->lgh_hdr->llh_cat_idx - cat_llh->lgh_last_idx;
1021 /* catalog is presently wrapped */
1022 return cat_llh->lgh_hdr->llh_cat_idx - cat_llh->lgh_last_idx;
1024 EXPORT_SYMBOL(llog_cat_free_space);
1026 static int llog_cat_reverse_process_cb(const struct lu_env *env,
1027 struct llog_handle *cat_llh,
1028 struct llog_rec_hdr *rec, void *data)
1030 struct llog_process_data *d = data;
1031 struct llog_handle *llh;
1035 rc = llog_cat_process_common(env, cat_llh, rec, &llh);
1037 /* The empty plain log was destroyed while processing */
1038 if (rc == LLOG_DEL_PLAIN) {
1039 rc = llog_cat_cleanup(env, cat_llh, llh,
1040 llh->u.phd.phd_cookie.lgc_index);
1041 } else if (rc == LLOG_DEL_RECORD) {
1042 /* clear wrong catalog entry */
1043 rc = llog_cat_cleanup(env, cat_llh, NULL, rec->lrh_index);
1048 rc = llog_reverse_process(env, llh, d->lpd_cb, d->lpd_data, NULL);
1050 /* The empty plain was destroyed while processing */
1051 if (rc == LLOG_DEL_PLAIN)
1052 rc = llog_cat_cleanup(env, cat_llh, llh,
1053 llh->u.phd.phd_cookie.lgc_index);
1055 llog_handle_put(llh);
1059 int llog_cat_reverse_process(const struct lu_env *env,
1060 struct llog_handle *cat_llh,
1061 llog_cb_t cb, void *data)
1063 struct llog_process_data d;
1064 struct llog_process_cat_data cd;
1065 struct llog_log_hdr *llh = cat_llh->lgh_hdr;
1069 LASSERT(llh->llh_flags & LLOG_F_IS_CAT);
1073 if (llh->llh_cat_idx >= cat_llh->lgh_last_idx &&
1074 llh->llh_count > 1) {
1075 CWARN("%s: catalog "DFID" crosses index zero\n",
1076 cat_llh->lgh_ctxt->loc_obd->obd_name,
1077 PFID(&cat_llh->lgh_id.lgl_oi.oi_fid));
1079 cd.lpcd_first_idx = 0;
1080 cd.lpcd_last_idx = cat_llh->lgh_last_idx;
1081 rc = llog_reverse_process(env, cat_llh,
1082 llog_cat_reverse_process_cb,
1087 cd.lpcd_first_idx = le32_to_cpu(llh->llh_cat_idx);
1088 cd.lpcd_last_idx = 0;
1089 rc = llog_reverse_process(env, cat_llh,
1090 llog_cat_reverse_process_cb,
1093 rc = llog_reverse_process(env, cat_llh,
1094 llog_cat_reverse_process_cb,
1100 EXPORT_SYMBOL(llog_cat_reverse_process);
1102 static int llog_cat_set_first_idx(struct llog_handle *cathandle, int idx)
1104 struct llog_log_hdr *llh = cathandle->lgh_hdr;
1109 bitmap_size = LLOG_HDR_BITMAP_SIZE(llh);
1111 * The llh_cat_idx equals to the first used index minus 1
1112 * so if we canceled the first index then llh_cat_idx
1115 if (llh->llh_cat_idx == (idx - 1)) {
1116 llh->llh_cat_idx = idx;
1118 while (idx != cathandle->lgh_last_idx) {
1119 idx = (idx + 1) % bitmap_size;
1120 if (!ext2_test_bit(idx, LLOG_HDR_BITMAP(llh))) {
1121 /* update llh_cat_idx for each unset bit,
1122 * expecting the next one is set */
1123 llh->llh_cat_idx = idx;
1124 } else if (idx == 0) {
1125 /* skip header bit */
1126 llh->llh_cat_idx = 0;
1129 /* the first index is found */
1134 CDEBUG(D_RPCTRACE, "catlog "DFID" first idx %u, last_idx %u\n",
1135 PFID(&cathandle->lgh_id.lgl_oi.oi_fid),
1136 llh->llh_cat_idx, cathandle->lgh_last_idx);
1142 /* Cleanup deleted plain llog traces from catalog */
1143 int llog_cat_cleanup(const struct lu_env *env, struct llog_handle *cathandle,
1144 struct llog_handle *loghandle, int index)
1149 if (loghandle != NULL) {
1150 /* remove destroyed llog from catalog list and
1151 * chd_current_log variable */
1152 down_write(&cathandle->lgh_lock);
1153 if (cathandle->u.chd.chd_current_log == loghandle)
1154 cathandle->u.chd.chd_current_log = NULL;
1155 list_del_init(&loghandle->u.phd.phd_entry);
1156 up_write(&cathandle->lgh_lock);
1157 LASSERT(index == loghandle->u.phd.phd_cookie.lgc_index);
1158 /* llog was opened and keep in a list, close it now */
1159 llog_close(env, loghandle);
1162 /* do not attempt to cleanup on-disk llog if on client side */
1163 if (cathandle->lgh_obj == NULL)
1166 /* remove plain llog entry from catalog by index */
1167 llog_cat_set_first_idx(cathandle, index);
1168 rc = llog_cancel_rec(env, cathandle, index);
1170 CDEBUG(D_HA, "cancel plain log at index %u of catalog "DFID"\n",
1171 index, PFID(&cathandle->lgh_id.lgl_oi.oi_fid));