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(loghandle->lgh_hdr) + 1;
217 llog_trans_destroy(env, loghandle, th);
219 dt_trans_stop(env, dt, handle);
223 static int llog_cat_refresh(const struct lu_env *env,
224 struct llog_handle *cathandle)
226 struct llog_handle *loghandle;
229 down_write(&cathandle->lgh_lock);
230 list_for_each_entry(loghandle, &cathandle->u.chd.chd_head,
232 if (!llog_exist(loghandle))
235 rc = llog_read_header(env, loghandle, NULL);
240 rc = llog_read_header(env, cathandle, NULL);
242 up_write(&loghandle->lgh_lock);
248 * prepare current/next log for catalog.
250 * if \a *ploghandle is NULL, open it, and declare create, NB, if \a
251 * *ploghandle is remote, create it synchronously here, see comments
254 * \a cathandle->lgh_lock is down_read-ed, it gets down_write-ed if \a
255 * *ploghandle has to be opened.
257 static int llog_cat_prep_log(const struct lu_env *env,
258 struct llog_handle *cathandle,
259 struct llog_handle **ploghandle,
268 if (IS_ERR_OR_NULL(*ploghandle)) {
269 up_read(&cathandle->lgh_lock);
270 down_write(&cathandle->lgh_lock);
272 if (IS_ERR_OR_NULL(*ploghandle)) {
273 struct llog_handle *loghandle;
275 rc = llog_open(env, cathandle->lgh_ctxt, &loghandle,
276 NULL, NULL, LLOG_OPEN_NEW);
278 *ploghandle = loghandle;
279 list_add_tail(&loghandle->u.phd.phd_entry,
280 &cathandle->u.chd.chd_head);
287 rc = llog_exist(*ploghandle);
293 if (dt_object_remote(cathandle->lgh_obj)) {
294 down_write_nested(&(*ploghandle)->lgh_lock, LLOGH_LOG);
295 if (!llog_exist(*ploghandle)) {
296 /* For remote operation, if we put the llog object
297 * creation in the current transaction, then the
298 * llog object will not be created on the remote
299 * target until the transaction stop, if other
300 * operations start before the transaction stop,
301 * and use the same llog object, will be dependent
302 * on the success of this transaction. So let's
303 * create the llog object synchronously here to
304 * remove the dependency. */
305 rc = llog_cat_new_log(env, cathandle, *ploghandle,
308 up_write(&(*ploghandle)->lgh_lock);
310 up_write(&cathandle->lgh_lock);
312 up_read(&cathandle->lgh_lock);
314 rc = llog_cat_refresh(env, cathandle);
315 down_read_nested(&cathandle->lgh_lock,
319 /* *ploghandle might become NULL, restart */
323 up_write(&(*ploghandle)->lgh_lock);
325 struct llog_thread_info *lgi = llog_info(env);
326 struct llog_logid_rec *lirec = &lgi->lgi_logid;
328 rc = llog_declare_create(env, *ploghandle, th);
332 lirec->lid_hdr.lrh_len = sizeof(*lirec);
333 rc = llog_declare_write_rec(env, cathandle, &lirec->lid_hdr, -1,
339 up_write(&cathandle->lgh_lock);
340 down_read_nested(&cathandle->lgh_lock, LLOGH_CAT);
347 /* Open an existent log handle and add it to the open list.
348 * This log handle will be closed when all of the records in it are removed.
350 * Assumes caller has already pushed us into the kernel context and is locking.
351 * We return a lock on the handle to ensure nobody yanks it from us.
353 * This takes extra reference on llog_handle via llog_handle_get() and require
354 * this reference to be put by caller using llog_handle_put()
356 int llog_cat_id2handle(const struct lu_env *env, struct llog_handle *cathandle,
357 struct llog_handle **res, struct llog_logid *logid)
359 struct llog_handle *loghandle;
365 if (cathandle == NULL)
368 fmt = cathandle->lgh_hdr->llh_flags & LLOG_F_EXT_MASK;
369 down_write(&cathandle->lgh_lock);
370 list_for_each_entry(loghandle, &cathandle->u.chd.chd_head,
372 struct llog_logid *cgl = &loghandle->lgh_id;
374 if (ostid_id(&cgl->lgl_oi) == ostid_id(&logid->lgl_oi) &&
375 ostid_seq(&cgl->lgl_oi) == ostid_seq(&logid->lgl_oi)) {
376 if (cgl->lgl_ogen != logid->lgl_ogen) {
377 CWARN("%s: log "DFID" generation %x != %x\n",
378 loghandle->lgh_ctxt->loc_obd->obd_name,
379 PFID(&logid->lgl_oi.oi_fid),
380 cgl->lgl_ogen, logid->lgl_ogen);
383 loghandle->u.phd.phd_cat_handle = cathandle;
384 up_write(&cathandle->lgh_lock);
388 up_write(&cathandle->lgh_lock);
390 rc = llog_open(env, cathandle->lgh_ctxt, &loghandle, logid, NULL,
393 CERROR("%s: error opening log id "DFID":%x: rc = %d\n",
394 cathandle->lgh_ctxt->loc_obd->obd_name,
395 PFID(&logid->lgl_oi.oi_fid), logid->lgl_ogen, rc);
399 rc = llog_init_handle(env, loghandle, LLOG_F_IS_PLAIN | fmt, NULL);
401 llog_close(env, loghandle);
406 down_write(&cathandle->lgh_lock);
407 list_add_tail(&loghandle->u.phd.phd_entry, &cathandle->u.chd.chd_head);
408 up_write(&cathandle->lgh_lock);
410 loghandle->u.phd.phd_cat_handle = cathandle;
411 loghandle->u.phd.phd_cookie.lgc_lgl = cathandle->lgh_id;
412 loghandle->u.phd.phd_cookie.lgc_index =
413 loghandle->lgh_hdr->llh_cat_idx;
416 llog_handle_get(loghandle);
421 int llog_cat_close(const struct lu_env *env, struct llog_handle *cathandle)
423 struct llog_handle *loghandle, *n;
428 list_for_each_entry_safe(loghandle, n, &cathandle->u.chd.chd_head,
430 struct llog_log_hdr *llh = loghandle->lgh_hdr;
433 /* unlink open-not-created llogs */
434 list_del_init(&loghandle->u.phd.phd_entry);
435 llh = loghandle->lgh_hdr;
436 if (loghandle->lgh_obj != NULL && llh != NULL &&
437 (llh->llh_flags & LLOG_F_ZAP_WHEN_EMPTY) &&
438 (llh->llh_count == 1)) {
439 rc = llog_destroy(env, loghandle);
441 CERROR("%s: failure destroying log during "
442 "cleanup: rc = %d\n",
443 loghandle->lgh_ctxt->loc_obd->obd_name,
446 index = loghandle->u.phd.phd_cookie.lgc_index;
447 llog_cat_cleanup(env, cathandle, NULL, index);
449 llog_close(env, loghandle);
451 /* if handle was stored in ctxt, remove it too */
452 if (cathandle->lgh_ctxt->loc_handle == cathandle)
453 cathandle->lgh_ctxt->loc_handle = NULL;
454 rc = llog_close(env, cathandle);
457 EXPORT_SYMBOL(llog_cat_close);
460 * lockdep markers for nested struct llog_handle::lgh_lock locking.
467 /** Return the currently active log handle. If the current log handle doesn't
468 * have enough space left for the current record, start a new one.
470 * If reclen is 0, we only want to know what the currently active log is,
471 * otherwise we get a lock on this log so nobody can steal our space.
473 * Assumes caller has already pushed us into the kernel context and is locking.
475 * NOTE: loghandle is write-locked upon successful return
477 static struct llog_handle *llog_cat_current_log(struct llog_handle *cathandle,
480 struct llog_handle *loghandle = NULL;
484 if (OBD_FAIL_CHECK(OBD_FAIL_MDS_LLOG_CREATE_FAILED2)) {
485 down_write_nested(&cathandle->lgh_lock, LLOGH_CAT);
486 GOTO(next, loghandle);
489 down_read_nested(&cathandle->lgh_lock, LLOGH_CAT);
490 loghandle = cathandle->u.chd.chd_current_log;
492 struct llog_log_hdr *llh;
494 down_write_nested(&loghandle->lgh_lock, LLOGH_LOG);
495 llh = loghandle->lgh_hdr;
496 if (llh == NULL || !llog_is_full(loghandle)) {
497 up_read(&cathandle->lgh_lock);
500 up_write(&loghandle->lgh_lock);
503 up_read(&cathandle->lgh_lock);
505 /* time to use next log */
507 /* first, we have to make sure the state hasn't changed */
508 down_write_nested(&cathandle->lgh_lock, LLOGH_CAT);
509 loghandle = cathandle->u.chd.chd_current_log;
511 struct llog_log_hdr *llh;
513 down_write_nested(&loghandle->lgh_lock, LLOGH_LOG);
514 llh = loghandle->lgh_hdr;
515 if (llh == NULL || !llog_is_full(loghandle))
516 GOTO(out_unlock, loghandle);
518 up_write(&loghandle->lgh_lock);
522 /* Sigh, the chd_next_log and chd_current_log is initialized
523 * in declare phase, and we do not serialize the catlog
524 * accessing, so it might be possible the llog creation
525 * thread (see llog_cat_declare_add_rec()) did not create
526 * llog successfully, then the following thread might
527 * meet this situation. */
528 if (IS_ERR_OR_NULL(cathandle->u.chd.chd_next_log)) {
529 CERROR("%s: next log does not exist!\n",
530 cathandle->lgh_ctxt->loc_obd->obd_name);
531 loghandle = ERR_PTR(-EIO);
532 if (cathandle->u.chd.chd_next_log == NULL) {
533 /* Store the error in chd_next_log, so
534 * the following process can get correct
536 cathandle->u.chd.chd_next_log = loghandle;
538 GOTO(out_unlock, loghandle);
541 CDEBUG(D_INODE, "use next log\n");
543 loghandle = cathandle->u.chd.chd_next_log;
544 cathandle->u.chd.chd_current_log = loghandle;
545 cathandle->u.chd.chd_next_log = NULL;
546 down_write_nested(&loghandle->lgh_lock, LLOGH_LOG);
549 up_write(&cathandle->lgh_lock);
554 /* Add a single record to the recovery log(s) using a catalog
555 * Returns as llog_write_record
557 * Assumes caller has already pushed us into the kernel context.
559 int llog_cat_add_rec(const struct lu_env *env, struct llog_handle *cathandle,
560 struct llog_rec_hdr *rec, struct llog_cookie *reccookie,
563 struct llog_handle *loghandle;
567 LASSERT(rec->lrh_len <= cathandle->lgh_ctxt->loc_chunk_size);
570 loghandle = llog_cat_current_log(cathandle, th);
571 if (IS_ERR(loghandle))
572 RETURN(PTR_ERR(loghandle));
574 /* loghandle is already locked by llog_cat_current_log() for us */
575 if (!llog_exist(loghandle)) {
576 rc = llog_cat_new_log(env, cathandle, loghandle, th);
578 up_write(&loghandle->lgh_lock);
579 /* nobody should be trying to use this llog */
580 down_write(&cathandle->lgh_lock);
581 if (cathandle->u.chd.chd_current_log == loghandle)
582 cathandle->u.chd.chd_current_log = NULL;
583 up_write(&cathandle->lgh_lock);
587 /* now let's try to add the record */
588 rc = llog_write_rec(env, loghandle, rec, reccookie, LLOG_NEXT_IDX, th);
590 CDEBUG_LIMIT(rc == -ENOSPC ? D_HA : D_ERROR,
591 "llog_write_rec %d: lh=%p\n", rc, loghandle);
592 /* -ENOSPC is returned if no empty records left
593 * and when it's lack of space on the stogage.
594 * there is no point to try again if it's the second
595 * case. many callers (like llog test) expect ENOSPC,
596 * so we preserve this error code, but look for the
597 * actual cause here */
598 if (rc == -ENOSPC && llog_is_full(loghandle))
601 up_write(&loghandle->lgh_lock);
603 if (rc == -ENOBUFS) {
606 CERROR("%s: error on 2nd llog: rc = %d\n",
607 cathandle->lgh_ctxt->loc_obd->obd_name, rc);
612 EXPORT_SYMBOL(llog_cat_add_rec);
614 int llog_cat_declare_add_rec(const struct lu_env *env,
615 struct llog_handle *cathandle,
616 struct llog_rec_hdr *rec, struct thandle *th)
623 down_read_nested(&cathandle->lgh_lock, LLOGH_CAT);
624 rc = llog_cat_prep_log(env, cathandle,
625 &cathandle->u.chd.chd_current_log, th);
629 rc = llog_cat_prep_log(env, cathandle, &cathandle->u.chd.chd_next_log,
634 rc = llog_declare_write_rec(env, cathandle->u.chd.chd_current_log,
636 if (rc == -ESTALE && dt_object_remote(cathandle->lgh_obj)) {
637 up_read(&cathandle->lgh_lock);
638 rc = llog_cat_refresh(env, cathandle);
646 * XXX: we hope for declarations made for existing llog this might be
647 * not correct with some backends where declarations are expected
648 * against specific object like ZFS with full debugging enabled.
650 rc = llog_declare_write_rec(env, cathandle->u.chd.chd_next_log, rec, -1,
654 up_read(&cathandle->lgh_lock);
657 EXPORT_SYMBOL(llog_cat_declare_add_rec);
659 int llog_cat_add(const struct lu_env *env, struct llog_handle *cathandle,
660 struct llog_rec_hdr *rec, struct llog_cookie *reccookie)
662 struct llog_ctxt *ctxt;
663 struct dt_device *dt;
664 struct thandle *th = NULL;
667 ctxt = cathandle->lgh_ctxt;
669 LASSERT(ctxt->loc_exp);
671 LASSERT(cathandle->lgh_obj != NULL);
672 dt = lu2dt_dev(cathandle->lgh_obj->do_lu.lo_dev);
674 th = dt_trans_create(env, dt);
678 rc = llog_cat_declare_add_rec(env, cathandle, rec, th);
682 rc = dt_trans_start_local(env, dt, th);
685 rc = llog_cat_add_rec(env, cathandle, rec, reccookie, th);
687 dt_trans_stop(env, dt, th);
690 EXPORT_SYMBOL(llog_cat_add);
692 /* For each cookie in the cookie array, we clear the log in-use bit and either:
693 * - the log is empty, so mark it free in the catalog header and delete it
694 * - the log is not empty, just write out the log header
696 * The cookies may be in different log files, so we need to get new logs
699 * Assumes caller has already pushed us into the kernel context.
701 int llog_cat_cancel_records(const struct lu_env *env,
702 struct llog_handle *cathandle, int count,
703 struct llog_cookie *cookies)
705 int i, index, rc = 0, failed = 0;
709 for (i = 0; i < count; i++, cookies++) {
710 struct llog_handle *loghandle;
711 struct llog_logid *lgl = &cookies->lgc_lgl;
714 rc = llog_cat_id2handle(env, cathandle, &loghandle, lgl);
716 CDEBUG(D_HA, "%s: cannot find llog for handle "DFID":%x"
718 cathandle->lgh_ctxt->loc_obd->obd_name,
719 PFID(&lgl->lgl_oi.oi_fid), lgl->lgl_ogen, rc);
724 if ((cathandle->lgh_ctxt->loc_flags &
725 LLOG_CTXT_FLAG_NORMAL_FID) && !llog_exist(loghandle)) {
726 /* For update log, some of loghandles of cathandle
727 * might not exist because remote llog creation might
728 * be failed, so let's skip the record cancellation
729 * for these non-exist llogs.
732 CDEBUG(D_HA, "%s: llog "DFID":%x does not exist"
734 cathandle->lgh_ctxt->loc_obd->obd_name,
735 PFID(&lgl->lgl_oi.oi_fid), lgl->lgl_ogen, lrc);
742 lrc = llog_cancel_rec(env, loghandle, cookies->lgc_index);
743 if (lrc == LLOG_DEL_PLAIN) { /* log has been destroyed */
744 index = loghandle->u.phd.phd_cookie.lgc_index;
745 lrc = llog_cat_cleanup(env, cathandle, loghandle,
749 } else if (lrc == -ENOENT) {
750 if (rc == 0) /* ENOENT shouldn't rewrite any error */
752 } else if (lrc < 0) {
757 llog_handle_put(loghandle);
760 CERROR("%s: fail to cancel %d of %d llog-records: rc = %d\n",
761 cathandle->lgh_ctxt->loc_obd->obd_name, failed, count,
766 EXPORT_SYMBOL(llog_cat_cancel_records);
768 static int llog_cat_process_common(const struct lu_env *env,
769 struct llog_handle *cat_llh,
770 struct llog_rec_hdr *rec,
771 struct llog_handle **llhp)
773 struct llog_logid_rec *lir = container_of(rec, typeof(*lir), lid_hdr);
774 struct llog_log_hdr *hdr;
778 if (rec->lrh_type != le32_to_cpu(LLOG_LOGID_MAGIC)) {
780 CWARN("%s: invalid record in catalog "DFID":%x: rc = %d\n",
781 cat_llh->lgh_ctxt->loc_obd->obd_name,
782 PFID(&cat_llh->lgh_id.lgl_oi.oi_fid),
783 cat_llh->lgh_id.lgl_ogen, rc);
786 CDEBUG(D_HA, "processing log "DFID":%x at index %u of catalog "DFID"\n",
787 PFID(&lir->lid_id.lgl_oi.oi_fid), lir->lid_id.lgl_ogen,
788 le32_to_cpu(rec->lrh_index),
789 PFID(&cat_llh->lgh_id.lgl_oi.oi_fid));
791 rc = llog_cat_id2handle(env, cat_llh, llhp, &lir->lid_id);
793 /* After a server crash, a stub of index record in catlog could
794 * be kept, because plain log destroy + catlog index record
795 * deletion are not atomic. So we end up with an index but no
796 * actual record. Destroy the index and move on. */
797 if (rc == -ENOENT || rc == -ESTALE)
798 rc = LLOG_DEL_RECORD;
800 CWARN("%s: can't find llog handle "DFID":%x: rc = %d\n",
801 cat_llh->lgh_ctxt->loc_obd->obd_name,
802 PFID(&lir->lid_id.lgl_oi.oi_fid),
803 lir->lid_id.lgl_ogen, rc);
808 /* clean old empty llogs, do not consider current llog in use */
809 /* ignore remote (lgh_obj == NULL) llogs */
810 hdr = (*llhp)->lgh_hdr;
811 if ((hdr->llh_flags & LLOG_F_ZAP_WHEN_EMPTY) &&
812 hdr->llh_count == 1 && cat_llh->lgh_obj != NULL &&
813 *llhp != cat_llh->u.chd.chd_current_log) {
814 rc = llog_destroy(env, *llhp);
816 CWARN("%s: can't destroy empty log "DFID": rc = %d\n",
817 (*llhp)->lgh_ctxt->loc_obd->obd_name,
818 PFID(&lir->lid_id.lgl_oi.oi_fid), rc);
825 static int llog_cat_process_cb(const struct lu_env *env,
826 struct llog_handle *cat_llh,
827 struct llog_rec_hdr *rec, void *data)
829 struct llog_process_data *d = data;
830 struct llog_handle *llh = NULL;
834 rc = llog_cat_process_common(env, cat_llh, rec, &llh);
838 if (rec->lrh_index < d->lpd_startcat) {
839 /* Skip processing of the logs until startcat */
841 } else if (d->lpd_startidx > 0) {
842 struct llog_process_cat_data cd;
844 cd.lpcd_first_idx = d->lpd_startidx;
845 cd.lpcd_last_idx = 0;
846 rc = llog_process_or_fork(env, llh, d->lpd_cb, d->lpd_data,
848 /* Continue processing the next log from idx 0 */
851 rc = llog_process_or_fork(env, llh, d->lpd_cb, d->lpd_data,
856 /* The empty plain log was destroyed while processing */
857 if (rc == LLOG_DEL_PLAIN) {
858 rc = llog_cat_cleanup(env, cat_llh, llh,
859 llh->u.phd.phd_cookie.lgc_index);
860 } else if (rc == LLOG_DEL_RECORD) {
861 /* clear wrong catalog entry */
862 rc = llog_cat_cleanup(env, cat_llh, NULL, rec->lrh_index);
866 llog_handle_put(llh);
871 int llog_cat_process_or_fork(const struct lu_env *env,
872 struct llog_handle *cat_llh, llog_cb_t cat_cb,
873 llog_cb_t cb, void *data, int startcat,
874 int startidx, bool fork)
876 struct llog_process_data d;
877 struct llog_log_hdr *llh = cat_llh->lgh_hdr;
882 LASSERT(llh->llh_flags & LLOG_F_IS_CAT);
885 d.lpd_startcat = (startcat == LLOG_CAT_FIRST ? 0 : startcat);
886 d.lpd_startidx = startidx;
888 if (llh->llh_cat_idx >= cat_llh->lgh_last_idx &&
889 llh->llh_count > 1) {
890 struct llog_process_cat_data cd;
892 CWARN("%s: catlog "DFID" crosses index zero\n",
893 cat_llh->lgh_ctxt->loc_obd->obd_name,
894 PFID(&cat_llh->lgh_id.lgl_oi.oi_fid));
895 /*startcat = 0 is default value for general processing */
896 if ((startcat != LLOG_CAT_FIRST &&
897 startcat >= llh->llh_cat_idx) || !startcat) {
898 /* processing the catalog part at the end */
899 cd.lpcd_first_idx = (startcat ? startcat :
901 if (OBD_FAIL_PRECHECK(OBD_FAIL_CAT_RECORDS))
902 cd.lpcd_last_idx = cfs_fail_val;
904 cd.lpcd_last_idx = 0;
905 rc = llog_process_or_fork(env, cat_llh, cat_cb,
907 /* Reset the startcat becasue it has already reached
914 /* processing the catalog part at the begining */
915 cd.lpcd_first_idx = (startcat == LLOG_CAT_FIRST) ? 0 : startcat;
916 /* Note, the processing will stop at the lgh_last_idx value,
917 * and it could be increased during processing. So records
918 * between current lgh_last_idx and lgh_last_idx in future
919 * would left unprocessed.
921 cd.lpcd_last_idx = cat_llh->lgh_last_idx;
922 rc = llog_process_or_fork(env, cat_llh, cat_cb,
925 rc = llog_process_or_fork(env, cat_llh, cat_cb,
931 EXPORT_SYMBOL(llog_cat_process_or_fork);
933 int llog_cat_process(const struct lu_env *env, struct llog_handle *cat_llh,
934 llog_cb_t cb, void *data, int startcat, int startidx)
936 return llog_cat_process_or_fork(env, cat_llh, llog_cat_process_cb,
937 cb, data, startcat, startidx, false);
939 EXPORT_SYMBOL(llog_cat_process);
941 static int llog_cat_size_cb(const struct lu_env *env,
942 struct llog_handle *cat_llh,
943 struct llog_rec_hdr *rec, void *data)
945 struct llog_process_data *d = data;
946 struct llog_handle *llh = NULL;
947 __u64 *cum_size = d->lpd_data;
952 rc = llog_cat_process_common(env, cat_llh, rec, &llh);
954 if (rc == LLOG_DEL_PLAIN) {
955 /* empty log was deleted, don't count it */
956 rc = llog_cat_cleanup(env, cat_llh, llh,
957 llh->u.phd.phd_cookie.lgc_index);
958 } else if (rc == LLOG_DEL_RECORD) {
959 /* clear wrong catalog entry */
960 rc = llog_cat_cleanup(env, cat_llh, NULL, rec->lrh_index);
962 size = llog_size(env, llh);
965 CDEBUG(D_INFO, "Add llog entry "DFID" size=%llu, tot=%llu\n",
966 PFID(&llh->lgh_id.lgl_oi.oi_fid), size, *cum_size);
970 llog_handle_put(llh);
975 __u64 llog_cat_size(const struct lu_env *env, struct llog_handle *cat_llh)
977 __u64 size = llog_size(env, cat_llh);
979 llog_cat_process_or_fork(env, cat_llh, llog_cat_size_cb,
980 NULL, &size, 0, 0, false);
984 EXPORT_SYMBOL(llog_cat_size);
986 /* currently returns the number of "free" entries in catalog,
987 * ie the available entries for a new plain LLOG file creation,
988 * even if catalog has wrapped
990 __u32 llog_cat_free_space(struct llog_handle *cat_llh)
992 /* simulate almost full Catalog */
993 if (OBD_FAIL_CHECK(OBD_FAIL_CAT_FREE_RECORDS))
996 if (cat_llh->lgh_hdr->llh_count == 1)
997 return LLOG_HDR_BITMAP_SIZE(cat_llh->lgh_hdr) - 1;
999 if (cat_llh->lgh_last_idx > cat_llh->lgh_hdr->llh_cat_idx)
1000 return LLOG_HDR_BITMAP_SIZE(cat_llh->lgh_hdr) - 1 +
1001 cat_llh->lgh_hdr->llh_cat_idx - cat_llh->lgh_last_idx;
1003 /* catalog is presently wrapped */
1004 return cat_llh->lgh_hdr->llh_cat_idx - cat_llh->lgh_last_idx;
1006 EXPORT_SYMBOL(llog_cat_free_space);
1008 static int llog_cat_reverse_process_cb(const struct lu_env *env,
1009 struct llog_handle *cat_llh,
1010 struct llog_rec_hdr *rec, void *data)
1012 struct llog_process_data *d = data;
1013 struct llog_handle *llh;
1017 rc = llog_cat_process_common(env, cat_llh, rec, &llh);
1019 /* The empty plain log was destroyed while processing */
1020 if (rc == LLOG_DEL_PLAIN) {
1021 rc = llog_cat_cleanup(env, cat_llh, llh,
1022 llh->u.phd.phd_cookie.lgc_index);
1023 } else if (rc == LLOG_DEL_RECORD) {
1024 /* clear wrong catalog entry */
1025 rc = llog_cat_cleanup(env, cat_llh, NULL, rec->lrh_index);
1030 rc = llog_reverse_process(env, llh, d->lpd_cb, d->lpd_data, NULL);
1032 /* The empty plain was destroyed while processing */
1033 if (rc == LLOG_DEL_PLAIN)
1034 rc = llog_cat_cleanup(env, cat_llh, llh,
1035 llh->u.phd.phd_cookie.lgc_index);
1037 llog_handle_put(llh);
1041 int llog_cat_reverse_process(const struct lu_env *env,
1042 struct llog_handle *cat_llh,
1043 llog_cb_t cb, void *data)
1045 struct llog_process_data d;
1046 struct llog_process_cat_data cd;
1047 struct llog_log_hdr *llh = cat_llh->lgh_hdr;
1051 LASSERT(llh->llh_flags & LLOG_F_IS_CAT);
1055 if (llh->llh_cat_idx >= cat_llh->lgh_last_idx &&
1056 llh->llh_count > 1) {
1057 CWARN("%s: catalog "DFID" crosses index zero\n",
1058 cat_llh->lgh_ctxt->loc_obd->obd_name,
1059 PFID(&cat_llh->lgh_id.lgl_oi.oi_fid));
1061 cd.lpcd_first_idx = 0;
1062 cd.lpcd_last_idx = cat_llh->lgh_last_idx;
1063 rc = llog_reverse_process(env, cat_llh,
1064 llog_cat_reverse_process_cb,
1069 cd.lpcd_first_idx = le32_to_cpu(llh->llh_cat_idx);
1070 cd.lpcd_last_idx = 0;
1071 rc = llog_reverse_process(env, cat_llh,
1072 llog_cat_reverse_process_cb,
1075 rc = llog_reverse_process(env, cat_llh,
1076 llog_cat_reverse_process_cb,
1082 EXPORT_SYMBOL(llog_cat_reverse_process);
1084 static int llog_cat_set_first_idx(struct llog_handle *cathandle, int idx)
1086 struct llog_log_hdr *llh = cathandle->lgh_hdr;
1091 bitmap_size = LLOG_HDR_BITMAP_SIZE(llh);
1093 * The llh_cat_idx equals to the first used index minus 1
1094 * so if we canceled the first index then llh_cat_idx
1097 if (llh->llh_cat_idx == (idx - 1)) {
1098 llh->llh_cat_idx = idx;
1100 while (idx != cathandle->lgh_last_idx) {
1101 idx = (idx + 1) % bitmap_size;
1102 if (!ext2_test_bit(idx, LLOG_HDR_BITMAP(llh))) {
1103 /* update llh_cat_idx for each unset bit,
1104 * expecting the next one is set */
1105 llh->llh_cat_idx = idx;
1106 } else if (idx == 0) {
1107 /* skip header bit */
1108 llh->llh_cat_idx = 0;
1111 /* the first index is found */
1116 CDEBUG(D_RPCTRACE, "catlog "DFID" first idx %u, last_idx %u\n",
1117 PFID(&cathandle->lgh_id.lgl_oi.oi_fid),
1118 llh->llh_cat_idx, cathandle->lgh_last_idx);
1124 /* Cleanup deleted plain llog traces from catalog */
1125 int llog_cat_cleanup(const struct lu_env *env, struct llog_handle *cathandle,
1126 struct llog_handle *loghandle, int index)
1131 if (loghandle != NULL) {
1132 /* remove destroyed llog from catalog list and
1133 * chd_current_log variable */
1134 down_write(&cathandle->lgh_lock);
1135 if (cathandle->u.chd.chd_current_log == loghandle)
1136 cathandle->u.chd.chd_current_log = NULL;
1137 list_del_init(&loghandle->u.phd.phd_entry);
1138 up_write(&cathandle->lgh_lock);
1139 LASSERT(index == loghandle->u.phd.phd_cookie.lgc_index);
1140 /* llog was opened and keep in a list, close it now */
1141 llog_close(env, loghandle);
1144 /* do not attempt to cleanup on-disk llog if on client side */
1145 if (cathandle->lgh_obj == NULL)
1148 /* remove plain llog entry from catalog by index */
1149 llog_cat_set_first_idx(cathandle, index);
1150 rc = llog_cancel_rec(env, cathandle, index);
1152 CDEBUG(D_HA, "cancel plain log at index %u of catalog "DFID"\n",
1153 index, PFID(&cathandle->lgh_id.lgl_oi.oi_fid));