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, 2016, 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;
390 if (!llog_is_full(loghandle))
391 GOTO(out_unlock, loghandle);
393 up_write(&loghandle->lgh_lock);
397 /* Sigh, the chd_next_log and chd_current_log is initialized
398 * in declare phase, and we do not serialize the catlog
399 * accessing, so it might be possible the llog creation
400 * thread (see llog_cat_declare_add_rec()) did not create
401 * llog successfully, then the following thread might
402 * meet this situation. */
403 if (IS_ERR_OR_NULL(cathandle->u.chd.chd_next_log)) {
404 CERROR("%s: next log does not exist!\n",
405 cathandle->lgh_ctxt->loc_obd->obd_name);
406 loghandle = ERR_PTR(-EIO);
407 if (cathandle->u.chd.chd_next_log == NULL) {
408 /* Store the error in chd_next_log, so
409 * the following process can get correct
411 cathandle->u.chd.chd_next_log = loghandle;
413 GOTO(out_unlock, loghandle);
416 CDEBUG(D_INODE, "use next log\n");
418 loghandle = cathandle->u.chd.chd_next_log;
419 cathandle->u.chd.chd_current_log = loghandle;
420 cathandle->u.chd.chd_next_log = NULL;
421 down_write_nested(&loghandle->lgh_lock, LLOGH_LOG);
424 up_write(&cathandle->lgh_lock);
429 static int llog_cat_update_header(const struct lu_env *env,
430 struct llog_handle *cathandle)
432 struct llog_handle *loghandle;
437 down_write(&cathandle->lgh_lock);
438 if (!cathandle->lgh_stale) {
439 up_write(&cathandle->lgh_lock);
442 list_for_each_entry(loghandle, &cathandle->u.chd.chd_head,
444 if (!llog_exist(loghandle))
447 rc = llog_read_header(env, loghandle, NULL);
449 up_write(&cathandle->lgh_lock);
453 rc = llog_read_header(env, cathandle, NULL);
455 cathandle->lgh_stale = 0;
456 up_write(&cathandle->lgh_lock);
463 /* Add a single record to the recovery log(s) using a catalog
464 * Returns as llog_write_record
466 * Assumes caller has already pushed us into the kernel context.
468 int llog_cat_add_rec(const struct lu_env *env, struct llog_handle *cathandle,
469 struct llog_rec_hdr *rec, struct llog_cookie *reccookie,
472 struct llog_handle *loghandle;
476 LASSERT(rec->lrh_len <= cathandle->lgh_ctxt->loc_chunk_size);
479 loghandle = llog_cat_current_log(cathandle, th);
480 if (IS_ERR(loghandle))
481 RETURN(PTR_ERR(loghandle));
483 /* loghandle is already locked by llog_cat_current_log() for us */
484 if (!llog_exist(loghandle)) {
485 rc = llog_cat_new_log(env, cathandle, loghandle, th);
487 up_write(&loghandle->lgh_lock);
488 /* nobody should be trying to use this llog */
489 down_write(&cathandle->lgh_lock);
490 if (cathandle->u.chd.chd_current_log == loghandle)
491 cathandle->u.chd.chd_current_log = NULL;
492 up_write(&cathandle->lgh_lock);
496 /* now let's try to add the record */
497 rc = llog_write_rec(env, loghandle, rec, reccookie, LLOG_NEXT_IDX, th);
499 CDEBUG_LIMIT(rc == -ENOSPC ? D_HA : D_ERROR,
500 "llog_write_rec %d: lh=%p\n", rc, loghandle);
501 /* -ENOSPC is returned if no empty records left
502 * and when it's lack of space on the stogage.
503 * there is no point to try again if it's the second
504 * case. many callers (like llog test) expect ENOSPC,
505 * so we preserve this error code, but look for the
506 * actual cause here */
507 if (rc == -ENOSPC && llog_is_full(loghandle))
510 up_write(&loghandle->lgh_lock);
512 if (rc == -ENOBUFS) {
515 CERROR("%s: error on 2nd llog: rc = %d\n",
516 cathandle->lgh_ctxt->loc_obd->obd_name, rc);
521 EXPORT_SYMBOL(llog_cat_add_rec);
523 int llog_cat_declare_add_rec(const struct lu_env *env,
524 struct llog_handle *cathandle,
525 struct llog_rec_hdr *rec, struct thandle *th)
527 struct llog_thread_info *lgi = llog_info(env);
528 struct llog_logid_rec *lirec = &lgi->lgi_logid;
529 struct llog_handle *loghandle, *next;
534 if (cathandle->u.chd.chd_current_log == NULL) {
535 /* declare new plain llog */
536 down_write(&cathandle->lgh_lock);
537 if (cathandle->u.chd.chd_current_log == NULL) {
538 rc = llog_open(env, cathandle->lgh_ctxt, &loghandle,
539 NULL, NULL, LLOG_OPEN_NEW);
541 cathandle->u.chd.chd_current_log = loghandle;
542 list_add_tail(&loghandle->u.phd.phd_entry,
543 &cathandle->u.chd.chd_head);
546 up_write(&cathandle->lgh_lock);
547 } else if (cathandle->u.chd.chd_next_log == NULL ||
548 IS_ERR(cathandle->u.chd.chd_next_log)) {
549 /* declare next plain llog */
550 down_write(&cathandle->lgh_lock);
551 if (cathandle->u.chd.chd_next_log == NULL ||
552 IS_ERR(cathandle->u.chd.chd_next_log)) {
553 rc = llog_open(env, cathandle->lgh_ctxt, &loghandle,
554 NULL, NULL, LLOG_OPEN_NEW);
556 cathandle->u.chd.chd_next_log = loghandle;
557 list_add_tail(&loghandle->u.phd.phd_entry,
558 &cathandle->u.chd.chd_head);
561 up_write(&cathandle->lgh_lock);
566 lirec->lid_hdr.lrh_len = sizeof(*lirec);
568 if (!llog_exist(cathandle->u.chd.chd_current_log)) {
569 if (dt_object_remote(cathandle->lgh_obj)) {
570 /* For remote operation, if we put the llog object
571 * creation in the current transaction, then the
572 * llog object will not be created on the remote
573 * target until the transaction stop, if other
574 * operations start before the transaction stop,
575 * and use the same llog object, will be dependent
576 * on the success of this transaction. So let's
577 * create the llog object synchronously here to
578 * remove the dependency. */
580 down_read_nested(&cathandle->lgh_lock, LLOGH_CAT);
581 loghandle = cathandle->u.chd.chd_current_log;
582 down_write_nested(&loghandle->lgh_lock, LLOGH_LOG);
583 if (cathandle->lgh_stale) {
584 up_write(&loghandle->lgh_lock);
585 up_read(&cathandle->lgh_lock);
586 GOTO(out, rc = -EIO);
588 if (!llog_exist(loghandle)) {
589 rc = llog_cat_new_log(env, cathandle, loghandle,
592 cathandle->lgh_stale = 1;
594 up_write(&loghandle->lgh_lock);
595 up_read(&cathandle->lgh_lock);
597 rc = llog_cat_update_header(env, cathandle);
605 rc = llog_declare_create(env,
606 cathandle->u.chd.chd_current_log, th);
609 llog_declare_write_rec(env, cathandle,
610 &lirec->lid_hdr, -1, th);
615 /* declare records in the llogs */
616 rc = llog_declare_write_rec(env, cathandle->u.chd.chd_current_log,
619 down_write(&cathandle->lgh_lock);
620 if (cathandle->lgh_stale) {
621 up_write(&cathandle->lgh_lock);
622 GOTO(out, rc = -EIO);
625 cathandle->lgh_stale = 1;
626 up_write(&cathandle->lgh_lock);
627 rc = llog_cat_update_header(env, cathandle);
635 next = cathandle->u.chd.chd_next_log;
636 if (!IS_ERR_OR_NULL(next)) {
637 if (!llog_exist(next)) {
638 if (dt_object_remote(cathandle->lgh_obj)) {
639 /* For remote operation, if we put the llog
640 * object creation in the current transaction,
641 * then the llog object will not be created on
642 * the remote target until the transaction stop,
643 * if other operations start before the
644 * transaction stop, and use the same llog
645 * object, will be dependent on the success of
646 * this transaction. So let's create the llog
647 * object synchronously here to remove the
649 down_write_nested(&cathandle->lgh_lock,
651 next = cathandle->u.chd.chd_next_log;
652 if (IS_ERR_OR_NULL(next)) {
653 /* Sigh, another thread just tried,
654 * let's fail as well */
655 up_write(&cathandle->lgh_lock);
663 down_write_nested(&next->lgh_lock, LLOGH_LOG);
664 if (!llog_exist(next)) {
665 rc = llog_cat_new_log(env, cathandle,
668 cathandle->u.chd.chd_next_log =
671 up_write(&next->lgh_lock);
672 up_write(&cathandle->lgh_lock);
676 rc = llog_declare_create(env, next, th);
677 llog_declare_write_rec(env, cathandle,
678 &lirec->lid_hdr, -1, th);
681 /* XXX: we hope for declarations made for existing llog
682 * this might be not correct with some backends
683 * where declarations are expected against specific
684 * object like ZFS with full debugging enabled */
685 /*llog_declare_write_rec(env, next, rec, -1, th);*/
690 EXPORT_SYMBOL(llog_cat_declare_add_rec);
692 int llog_cat_add(const struct lu_env *env, struct llog_handle *cathandle,
693 struct llog_rec_hdr *rec, struct llog_cookie *reccookie)
695 struct llog_ctxt *ctxt;
696 struct dt_device *dt;
697 struct thandle *th = NULL;
700 ctxt = cathandle->lgh_ctxt;
702 LASSERT(ctxt->loc_exp);
704 LASSERT(cathandle->lgh_obj != NULL);
705 dt = lu2dt_dev(cathandle->lgh_obj->do_lu.lo_dev);
707 th = dt_trans_create(env, dt);
711 rc = llog_cat_declare_add_rec(env, cathandle, rec, th);
715 rc = dt_trans_start_local(env, dt, th);
718 rc = llog_cat_add_rec(env, cathandle, rec, reccookie, th);
720 dt_trans_stop(env, dt, th);
723 EXPORT_SYMBOL(llog_cat_add);
725 /* For each cookie in the cookie array, we clear the log in-use bit and either:
726 * - the log is empty, so mark it free in the catalog header and delete it
727 * - the log is not empty, just write out the log header
729 * The cookies may be in different log files, so we need to get new logs
732 * Assumes caller has already pushed us into the kernel context.
734 int llog_cat_cancel_records(const struct lu_env *env,
735 struct llog_handle *cathandle, int count,
736 struct llog_cookie *cookies)
738 int i, index, rc = 0, failed = 0;
742 for (i = 0; i < count; i++, cookies++) {
743 struct llog_handle *loghandle;
744 struct llog_logid *lgl = &cookies->lgc_lgl;
747 rc = llog_cat_id2handle(env, cathandle, &loghandle, lgl);
749 CDEBUG(D_HA, "%s: cannot find llog for handle "DFID":%x"
751 cathandle->lgh_ctxt->loc_obd->obd_name,
752 PFID(&lgl->lgl_oi.oi_fid), lgl->lgl_ogen, rc);
757 if ((cathandle->lgh_ctxt->loc_flags &
758 LLOG_CTXT_FLAG_NORMAL_FID) && !llog_exist(loghandle)) {
759 /* For update log, some of loghandles of cathandle
760 * might not exist because remote llog creation might
761 * be failed, so let's skip the record cancellation
762 * for these non-exist llogs.
765 CDEBUG(D_HA, "%s: llog "DFID":%x does not exist"
767 cathandle->lgh_ctxt->loc_obd->obd_name,
768 PFID(&lgl->lgl_oi.oi_fid), lgl->lgl_ogen, lrc);
775 lrc = llog_cancel_rec(env, loghandle, cookies->lgc_index);
776 if (lrc == LLOG_DEL_PLAIN) { /* log has been destroyed */
777 index = loghandle->u.phd.phd_cookie.lgc_index;
778 lrc = llog_cat_cleanup(env, cathandle, loghandle,
782 } else if (lrc == -ENOENT) {
783 if (rc == 0) /* ENOENT shouldn't rewrite any error */
785 } else if (lrc < 0) {
790 llog_handle_put(loghandle);
793 CERROR("%s: fail to cancel %d of %d llog-records: rc = %d\n",
794 cathandle->lgh_ctxt->loc_obd->obd_name, failed, count,
799 EXPORT_SYMBOL(llog_cat_cancel_records);
801 static int llog_cat_process_cb(const struct lu_env *env,
802 struct llog_handle *cat_llh,
803 struct llog_rec_hdr *rec, void *data)
805 struct llog_process_data *d = data;
806 struct llog_logid_rec *lir = (struct llog_logid_rec *)rec;
807 struct llog_handle *llh;
808 struct llog_log_hdr *hdr;
812 if (rec->lrh_type != LLOG_LOGID_MAGIC) {
813 CERROR("invalid record in catalog\n");
816 CDEBUG(D_HA, "processing log "DFID":%x at index %u of catalog "
817 DFID"\n", PFID(&lir->lid_id.lgl_oi.oi_fid), lir->lid_id.lgl_ogen,
818 rec->lrh_index, PFID(&cat_llh->lgh_id.lgl_oi.oi_fid));
820 rc = llog_cat_id2handle(env, cat_llh, &llh, &lir->lid_id);
822 CERROR("%s: cannot find handle for llog "DFID": rc = %d\n",
823 cat_llh->lgh_ctxt->loc_obd->obd_name,
824 PFID(&lir->lid_id.lgl_oi.oi_fid), rc);
825 if (rc == -ENOENT || rc == -ESTALE) {
826 /* After a server crash, a stub of index
827 * record in catlog could be kept, because
828 * plain log destroy + catlog index record
829 * deletion are not atomic. So we end up with
830 * an index but no actual record. Destroy the
831 * index and move on. */
832 rc = llog_cat_cleanup(env, cat_llh, NULL,
839 /* clean old empty llogs, do not consider current llog in use */
840 /* ignore remote (lgh_obj=NULL) llogs */
842 if ((hdr->llh_flags & LLOG_F_ZAP_WHEN_EMPTY) &&
843 hdr->llh_count == 1 && cat_llh->lgh_obj != NULL &&
844 llh != cat_llh->u.chd.chd_current_log) {
845 rc = llog_destroy(env, llh);
847 CERROR("%s: fail to destroy empty log: rc = %d\n",
848 llh->lgh_ctxt->loc_obd->obd_name, rc);
849 GOTO(out, rc = LLOG_DEL_PLAIN);
852 if (rec->lrh_index < d->lpd_startcat) {
853 /* Skip processing of the logs until startcat */
855 } else if (d->lpd_startidx > 0) {
856 struct llog_process_cat_data cd;
858 cd.lpcd_first_idx = d->lpd_startidx;
859 cd.lpcd_last_idx = 0;
860 rc = llog_process_or_fork(env, llh, d->lpd_cb, d->lpd_data,
862 /* Continue processing the next log from idx 0 */
865 rc = llog_process_or_fork(env, llh, d->lpd_cb, d->lpd_data,
870 /* The empty plain log was destroyed while processing */
871 if (rc == LLOG_DEL_PLAIN)
872 rc = llog_cat_cleanup(env, cat_llh, llh,
873 llh->u.phd.phd_cookie.lgc_index);
874 llog_handle_put(llh);
879 int llog_cat_process_or_fork(const struct lu_env *env,
880 struct llog_handle *cat_llh, llog_cb_t cat_cb,
881 llog_cb_t cb, void *data, int startcat,
882 int startidx, bool fork)
884 struct llog_process_data d;
885 struct llog_log_hdr *llh = cat_llh->lgh_hdr;
889 LASSERT(llh->llh_flags & LLOG_F_IS_CAT);
892 d.lpd_startcat = startcat;
893 d.lpd_startidx = startidx;
895 if (llh->llh_cat_idx >= cat_llh->lgh_last_idx &&
896 llh->llh_count > 1) {
897 struct llog_process_cat_data cd;
899 CWARN("%s: catlog "DFID" crosses index zero\n",
900 cat_llh->lgh_ctxt->loc_obd->obd_name,
901 PFID(&cat_llh->lgh_id.lgl_oi.oi_fid));
903 cd.lpcd_first_idx = llh->llh_cat_idx;
904 cd.lpcd_last_idx = 0;
905 rc = llog_process_or_fork(env, cat_llh, cat_cb,
910 cd.lpcd_first_idx = 0;
911 cd.lpcd_last_idx = cat_llh->lgh_last_idx;
912 rc = llog_process_or_fork(env, cat_llh, cat_cb,
915 rc = llog_process_or_fork(env, cat_llh, cat_cb,
922 int llog_cat_process(const struct lu_env *env, struct llog_handle *cat_llh,
923 llog_cb_t cb, void *data, int startcat, int startidx)
925 return llog_cat_process_or_fork(env, cat_llh, llog_cat_process_cb,
926 cb, data, startcat, startidx, false);
928 EXPORT_SYMBOL(llog_cat_process);
930 static int llog_cat_size_cb(const struct lu_env *env,
931 struct llog_handle *cat_llh,
932 struct llog_rec_hdr *rec, void *data)
934 struct llog_process_data *d = data;
935 struct llog_logid_rec *lir = (struct llog_logid_rec *)rec;
936 struct llog_handle *llh;
938 __u64 *cum_size = d->lpd_data;
942 if (rec->lrh_type != LLOG_LOGID_MAGIC) {
943 CERROR("%s: invalid record in catalog, rc = %d\n",
944 cat_llh->lgh_ctxt->loc_obd->obd_name, -EINVAL);
947 CDEBUG(D_HA, "processing log "DFID":%x at index %u of catalog "
948 DFID"\n", PFID(&lir->lid_id.lgl_oi.oi_fid), lir->lid_id.lgl_ogen,
949 rec->lrh_index, PFID(&cat_llh->lgh_id.lgl_oi.oi_fid));
951 rc = llog_cat_id2handle(env, cat_llh, &llh, &lir->lid_id);
953 CWARN("%s: cannot find handle for llog "DFID": rc = %d\n",
954 cat_llh->lgh_ctxt->loc_obd->obd_name,
955 PFID(&lir->lid_id.lgl_oi.oi_fid), rc);
958 size = llog_size(env, llh);
961 CDEBUG(D_INFO, "Add llog entry "DFID" size %llu\n",
962 PFID(&llh->lgh_id.lgl_oi.oi_fid), size);
964 llog_handle_put(llh);
970 __u64 llog_cat_size(const struct lu_env *env, struct llog_handle *cat_llh)
972 __u64 size = llog_size(env, cat_llh);
974 llog_cat_process_or_fork(env, cat_llh, llog_cat_size_cb,
975 NULL, &size, 0, 0, false);
979 EXPORT_SYMBOL(llog_cat_size);
981 static int llog_cat_reverse_process_cb(const struct lu_env *env,
982 struct llog_handle *cat_llh,
983 struct llog_rec_hdr *rec, void *data)
985 struct llog_process_data *d = data;
986 struct llog_logid_rec *lir = (struct llog_logid_rec *)rec;
987 struct llog_handle *llh;
988 struct llog_log_hdr *hdr;
991 if (le32_to_cpu(rec->lrh_type) != LLOG_LOGID_MAGIC) {
992 CERROR("invalid record in catalog\n");
995 CDEBUG(D_HA, "processing log "DFID":%x at index %u of catalog "
996 DFID"\n", PFID(&lir->lid_id.lgl_oi.oi_fid), lir->lid_id.lgl_ogen,
997 le32_to_cpu(rec->lrh_index),
998 PFID(&cat_llh->lgh_id.lgl_oi.oi_fid));
1000 rc = llog_cat_id2handle(env, cat_llh, &llh, &lir->lid_id);
1002 CERROR("%s: cannot find handle for llog "DFID": rc = %d\n",
1003 cat_llh->lgh_ctxt->loc_obd->obd_name,
1004 PFID(&lir->lid_id.lgl_oi.oi_fid), rc);
1005 if (rc == -ENOENT || rc == -ESTALE) {
1006 /* After a server crash, a stub of index
1007 * record in catlog could be kept, because
1008 * plain log destroy + catlog index record
1009 * deletion are not atomic. So we end up with
1010 * an index but no actual record. Destroy the
1011 * index and move on. */
1012 rc = llog_cat_cleanup(env, cat_llh, NULL,
1019 /* clean old empty llogs, do not consider current llog in use */
1021 if ((hdr->llh_flags & LLOG_F_ZAP_WHEN_EMPTY) &&
1022 hdr->llh_count == 1 &&
1023 llh != cat_llh->u.chd.chd_current_log) {
1024 rc = llog_destroy(env, llh);
1026 CERROR("%s: fail to destroy empty log: rc = %d\n",
1027 llh->lgh_ctxt->loc_obd->obd_name, rc);
1028 GOTO(out, rc = LLOG_DEL_PLAIN);
1031 rc = llog_reverse_process(env, llh, d->lpd_cb, d->lpd_data, NULL);
1034 /* The empty plain was destroyed while processing */
1035 if (rc == LLOG_DEL_PLAIN)
1036 rc = llog_cat_cleanup(env, cat_llh, llh,
1037 llh->u.phd.phd_cookie.lgc_index);
1039 llog_handle_put(llh);
1043 int llog_cat_reverse_process(const struct lu_env *env,
1044 struct llog_handle *cat_llh,
1045 llog_cb_t cb, void *data)
1047 struct llog_process_data d;
1048 struct llog_process_cat_data cd;
1049 struct llog_log_hdr *llh = cat_llh->lgh_hdr;
1053 LASSERT(llh->llh_flags & LLOG_F_IS_CAT);
1057 if (llh->llh_cat_idx >= cat_llh->lgh_last_idx &&
1058 llh->llh_count > 1) {
1059 CWARN("%s: catalog "DFID" crosses index zero\n",
1060 cat_llh->lgh_ctxt->loc_obd->obd_name,
1061 PFID(&cat_llh->lgh_id.lgl_oi.oi_fid));
1063 cd.lpcd_first_idx = 0;
1064 cd.lpcd_last_idx = cat_llh->lgh_last_idx;
1065 rc = llog_reverse_process(env, cat_llh,
1066 llog_cat_reverse_process_cb,
1071 cd.lpcd_first_idx = le32_to_cpu(llh->llh_cat_idx);
1072 cd.lpcd_last_idx = 0;
1073 rc = llog_reverse_process(env, cat_llh,
1074 llog_cat_reverse_process_cb,
1077 rc = llog_reverse_process(env, cat_llh,
1078 llog_cat_reverse_process_cb,
1084 EXPORT_SYMBOL(llog_cat_reverse_process);
1086 static int llog_cat_set_first_idx(struct llog_handle *cathandle, int idx)
1088 struct llog_log_hdr *llh = cathandle->lgh_hdr;
1093 bitmap_size = LLOG_HDR_BITMAP_SIZE(llh);
1095 * The llh_cat_idx equals to the first used index minus 1
1096 * so if we canceled the first index then llh_cat_idx
1099 if (llh->llh_cat_idx == (idx - 1)) {
1100 llh->llh_cat_idx = idx;
1102 while (idx != cathandle->lgh_last_idx) {
1103 idx = (idx + 1) % bitmap_size;
1104 if (!ext2_test_bit(idx, LLOG_HDR_BITMAP(llh))) {
1105 /* update llh_cat_idx for each unset bit,
1106 * expecting the next one is set */
1107 llh->llh_cat_idx = idx;
1108 } else if (idx == 0) {
1109 /* skip header bit */
1110 llh->llh_cat_idx = 0;
1113 /* the first index is found */
1118 CDEBUG(D_RPCTRACE, "catlog "DFID" first idx %u, last_idx %u\n",
1119 PFID(&cathandle->lgh_id.lgl_oi.oi_fid),
1120 llh->llh_cat_idx, cathandle->lgh_last_idx);
1126 /* Cleanup deleted plain llog traces from catalog */
1127 int llog_cat_cleanup(const struct lu_env *env, struct llog_handle *cathandle,
1128 struct llog_handle *loghandle, int index)
1133 if (loghandle != NULL) {
1134 /* remove destroyed llog from catalog list and
1135 * chd_current_log variable */
1136 down_write(&cathandle->lgh_lock);
1137 if (cathandle->u.chd.chd_current_log == loghandle)
1138 cathandle->u.chd.chd_current_log = NULL;
1139 list_del_init(&loghandle->u.phd.phd_entry);
1140 up_write(&cathandle->lgh_lock);
1141 LASSERT(index == loghandle->u.phd.phd_cookie.lgc_index);
1142 /* llog was opened and keep in a list, close it now */
1143 llog_close(env, loghandle);
1146 /* do not attempt to cleanup on-disk llog if on client side */
1147 if (cathandle->lgh_obj == NULL)
1150 /* remove plain llog entry from catalog by index */
1151 llog_cat_set_first_idx(cathandle, index);
1152 rc = llog_cancel_rec(env, cathandle, index);
1154 CDEBUG(D_HA, "cancel plain log at index %u of catalog "DFID"\n",
1155 index, PFID(&cathandle->lgh_id.lgl_oi.oi_fid));