1 /* -*- mode: c; c-basic-offset: 8; -*-
2 * vim: noexpandtab sw=8 ts=8 sts=0:
6 * Defines functions of journalling api
8 * Copyright (C) 2003, 2004 Oracle. All rights reserved.
10 * This program is free software; you can redistribute it and/or
11 * modify it under the terms of the GNU General Public
12 * License as published by the Free Software Foundation; either
13 * version 2 of the License, or (at your option) any later version.
15 * This program is distributed in the hope that it will be useful,
16 * but WITHOUT ANY WARRANTY; without even the implied warranty of
17 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
18 * General Public License for more details.
20 * You should have received a copy of the GNU General Public
21 * License along with this program; if not, write to the
22 * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
23 * Boston, MA 021110-1307, USA.
27 #include <linux/types.h>
28 #include <linux/slab.h>
29 #include <linux/highmem.h>
30 #include <linux/kthread.h>
31 #include <linux/time.h>
32 #include <linux/random.h>
34 #include <cluster/masklog.h>
39 #include "blockcheck.h"
42 #include "extent_map.h"
43 #include "heartbeat.h"
46 #include "localalloc.h"
53 #include "buffer_head_io.h"
54 #include "ocfs2_trace.h"
56 DEFINE_SPINLOCK(trans_inc_lock);
58 #define ORPHAN_SCAN_SCHEDULE_TIMEOUT 300000
60 static int ocfs2_force_read_journal(struct inode *inode);
61 static int ocfs2_recover_node(struct ocfs2_super *osb,
62 int node_num, int slot_num);
63 static int __ocfs2_recovery_thread(void *arg);
64 static int ocfs2_commit_cache(struct ocfs2_super *osb);
65 static int __ocfs2_wait_on_mount(struct ocfs2_super *osb, int quota);
66 static int ocfs2_journal_toggle_dirty(struct ocfs2_super *osb,
67 int dirty, int replayed);
68 static int ocfs2_trylock_journal(struct ocfs2_super *osb,
70 static int ocfs2_recover_orphans(struct ocfs2_super *osb,
72 static int ocfs2_commit_thread(void *arg);
73 static void ocfs2_queue_recovery_completion(struct ocfs2_journal *journal,
75 struct ocfs2_dinode *la_dinode,
76 struct ocfs2_dinode *tl_dinode,
77 struct ocfs2_quota_recovery *qrec);
79 static inline int ocfs2_wait_on_mount(struct ocfs2_super *osb)
81 return __ocfs2_wait_on_mount(osb, 0);
84 static inline int ocfs2_wait_on_quotas(struct ocfs2_super *osb)
86 return __ocfs2_wait_on_mount(osb, 1);
90 * This replay_map is to track online/offline slots, so we could recover
91 * offline slots during recovery and mount
94 enum ocfs2_replay_state {
95 REPLAY_UNNEEDED = 0, /* Replay is not needed, so ignore this map */
96 REPLAY_NEEDED, /* Replay slots marked in rm_replay_slots */
97 REPLAY_DONE /* Replay was already queued */
100 struct ocfs2_replay_map {
101 unsigned int rm_slots;
102 enum ocfs2_replay_state rm_state;
103 unsigned char rm_replay_slots[0];
106 void ocfs2_replay_map_set_state(struct ocfs2_super *osb, int state)
108 if (!osb->replay_map)
111 /* If we've already queued the replay, we don't have any more to do */
112 if (osb->replay_map->rm_state == REPLAY_DONE)
115 osb->replay_map->rm_state = state;
118 int ocfs2_compute_replay_slots(struct ocfs2_super *osb)
120 struct ocfs2_replay_map *replay_map;
123 /* If replay map is already set, we don't do it again */
127 replay_map = kzalloc(sizeof(struct ocfs2_replay_map) +
128 (osb->max_slots * sizeof(char)), GFP_KERNEL);
135 spin_lock(&osb->osb_lock);
137 replay_map->rm_slots = osb->max_slots;
138 replay_map->rm_state = REPLAY_UNNEEDED;
140 /* set rm_replay_slots for offline slot(s) */
141 for (i = 0; i < replay_map->rm_slots; i++) {
142 if (ocfs2_slot_to_node_num_locked(osb, i, &node_num) == -ENOENT)
143 replay_map->rm_replay_slots[i] = 1;
146 osb->replay_map = replay_map;
147 spin_unlock(&osb->osb_lock);
151 void ocfs2_queue_replay_slots(struct ocfs2_super *osb)
153 struct ocfs2_replay_map *replay_map = osb->replay_map;
159 if (replay_map->rm_state != REPLAY_NEEDED)
162 for (i = 0; i < replay_map->rm_slots; i++)
163 if (replay_map->rm_replay_slots[i])
164 ocfs2_queue_recovery_completion(osb->journal, i, NULL,
166 replay_map->rm_state = REPLAY_DONE;
169 void ocfs2_free_replay_slots(struct ocfs2_super *osb)
171 struct ocfs2_replay_map *replay_map = osb->replay_map;
173 if (!osb->replay_map)
177 osb->replay_map = NULL;
180 int ocfs2_recovery_init(struct ocfs2_super *osb)
182 struct ocfs2_recovery_map *rm;
184 mutex_init(&osb->recovery_lock);
185 osb->disable_recovery = 0;
186 osb->recovery_thread_task = NULL;
187 init_waitqueue_head(&osb->recovery_event);
189 rm = kzalloc(sizeof(struct ocfs2_recovery_map) +
190 osb->max_slots * sizeof(unsigned int),
197 rm->rm_entries = (unsigned int *)((char *)rm +
198 sizeof(struct ocfs2_recovery_map));
199 osb->recovery_map = rm;
204 /* we can't grab the goofy sem lock from inside wait_event, so we use
205 * memory barriers to make sure that we'll see the null task before
207 static int ocfs2_recovery_thread_running(struct ocfs2_super *osb)
210 return osb->recovery_thread_task != NULL;
213 void ocfs2_recovery_exit(struct ocfs2_super *osb)
215 struct ocfs2_recovery_map *rm;
217 /* disable any new recovery threads and wait for any currently
218 * running ones to exit. Do this before setting the vol_state. */
219 mutex_lock(&osb->recovery_lock);
220 osb->disable_recovery = 1;
221 mutex_unlock(&osb->recovery_lock);
222 wait_event(osb->recovery_event, !ocfs2_recovery_thread_running(osb));
224 /* At this point, we know that no more recovery threads can be
225 * launched, so wait for any recovery completion work to
227 flush_workqueue(ocfs2_wq);
230 * Now that recovery is shut down, and the osb is about to be
231 * freed, the osb_lock is not taken here.
233 rm = osb->recovery_map;
234 /* XXX: Should we bug if there are dirty entries? */
239 static int __ocfs2_recovery_map_test(struct ocfs2_super *osb,
240 unsigned int node_num)
243 struct ocfs2_recovery_map *rm = osb->recovery_map;
245 assert_spin_locked(&osb->osb_lock);
247 for (i = 0; i < rm->rm_used; i++) {
248 if (rm->rm_entries[i] == node_num)
255 /* Behaves like test-and-set. Returns the previous value */
256 static int ocfs2_recovery_map_set(struct ocfs2_super *osb,
257 unsigned int node_num)
259 struct ocfs2_recovery_map *rm = osb->recovery_map;
261 spin_lock(&osb->osb_lock);
262 if (__ocfs2_recovery_map_test(osb, node_num)) {
263 spin_unlock(&osb->osb_lock);
267 /* XXX: Can this be exploited? Not from o2dlm... */
268 BUG_ON(rm->rm_used >= osb->max_slots);
270 rm->rm_entries[rm->rm_used] = node_num;
272 spin_unlock(&osb->osb_lock);
277 static void ocfs2_recovery_map_clear(struct ocfs2_super *osb,
278 unsigned int node_num)
281 struct ocfs2_recovery_map *rm = osb->recovery_map;
283 spin_lock(&osb->osb_lock);
285 for (i = 0; i < rm->rm_used; i++) {
286 if (rm->rm_entries[i] == node_num)
290 if (i < rm->rm_used) {
291 /* XXX: be careful with the pointer math */
292 memmove(&(rm->rm_entries[i]), &(rm->rm_entries[i + 1]),
293 (rm->rm_used - i - 1) * sizeof(unsigned int));
297 spin_unlock(&osb->osb_lock);
300 static int ocfs2_commit_cache(struct ocfs2_super *osb)
303 unsigned int flushed;
304 struct ocfs2_journal *journal = NULL;
306 journal = osb->journal;
308 /* Flush all pending commits and checkpoint the journal. */
309 down_write(&journal->j_trans_barrier);
311 flushed = atomic_read(&journal->j_num_trans);
312 trace_ocfs2_commit_cache_begin(flushed);
314 up_write(&journal->j_trans_barrier);
318 jbd2_journal_lock_updates(journal->j_journal);
319 status = jbd2_journal_flush(journal->j_journal);
320 jbd2_journal_unlock_updates(journal->j_journal);
322 up_write(&journal->j_trans_barrier);
327 ocfs2_inc_trans_id(journal);
329 flushed = atomic_read(&journal->j_num_trans);
330 atomic_set(&journal->j_num_trans, 0);
331 up_write(&journal->j_trans_barrier);
333 trace_ocfs2_commit_cache_end(journal->j_trans_id, flushed);
335 ocfs2_wake_downconvert_thread(osb);
336 wake_up(&journal->j_checkpointed);
341 handle_t *ocfs2_start_trans(struct ocfs2_super *osb, int max_buffs)
343 journal_t *journal = osb->journal->j_journal;
346 BUG_ON(!osb || !osb->journal->j_journal);
348 if (ocfs2_is_hard_readonly(osb))
349 return ERR_PTR(-EROFS);
351 BUG_ON(osb->journal->j_state == OCFS2_JOURNAL_FREE);
352 BUG_ON(max_buffs <= 0);
354 /* Nested transaction? Just return the handle... */
355 if (journal_current_handle())
356 return jbd2_journal_start(journal, max_buffs);
358 sb_start_intwrite(osb->sb);
360 down_read(&osb->journal->j_trans_barrier);
362 handle = jbd2_journal_start(journal, max_buffs);
363 if (IS_ERR(handle)) {
364 up_read(&osb->journal->j_trans_barrier);
365 sb_end_intwrite(osb->sb);
367 mlog_errno(PTR_ERR(handle));
369 if (is_journal_aborted(journal)) {
370 ocfs2_abort(osb->sb, "Detected aborted journal");
371 handle = ERR_PTR(-EROFS);
374 if (!ocfs2_mount_local(osb))
375 atomic_inc(&(osb->journal->j_num_trans));
381 int ocfs2_commit_trans(struct ocfs2_super *osb,
385 struct ocfs2_journal *journal = osb->journal;
389 nested = handle->h_ref > 1;
390 ret = jbd2_journal_stop(handle);
395 up_read(&journal->j_trans_barrier);
396 sb_end_intwrite(osb->sb);
403 * 'nblocks' is what you want to add to the current transaction.
405 * This might call jbd2_journal_restart() which will commit dirty buffers
406 * and then restart the transaction. Before calling
407 * ocfs2_extend_trans(), any changed blocks should have been
408 * dirtied. After calling it, all blocks which need to be changed must
409 * go through another set of journal_access/journal_dirty calls.
411 * WARNING: This will not release any semaphores or disk locks taken
412 * during the transaction, so make sure they were taken *before*
413 * start_trans or we'll have ordering deadlocks.
415 * WARNING2: Note that we do *not* drop j_trans_barrier here. This is
416 * good because transaction ids haven't yet been recorded on the
417 * cluster locks associated with this handle.
419 int ocfs2_extend_trans(handle_t *handle, int nblocks)
421 int status, old_nblocks;
429 old_nblocks = handle->h_buffer_credits;
431 trace_ocfs2_extend_trans(old_nblocks, nblocks);
433 #ifdef CONFIG_OCFS2_DEBUG_FS
436 status = jbd2_journal_extend(handle, nblocks);
444 trace_ocfs2_extend_trans_restart(old_nblocks + nblocks);
445 status = jbd2_journal_restart(handle,
446 old_nblocks + nblocks);
459 * If we have fewer than thresh credits, extend by OCFS2_MAX_TRANS_DATA.
460 * If that fails, restart the transaction & regain write access for the
461 * buffer head which is used for metadata modifications.
462 * Taken from Ext4: extend_or_restart_transaction()
464 int ocfs2_allocate_extend_trans(handle_t *handle, int thresh)
466 int status, old_nblks;
470 old_nblks = handle->h_buffer_credits;
471 trace_ocfs2_allocate_extend_trans(old_nblks, thresh);
473 if (old_nblks < thresh)
476 status = jbd2_journal_extend(handle, OCFS2_MAX_TRANS_DATA);
483 status = jbd2_journal_restart(handle, OCFS2_MAX_TRANS_DATA);
493 struct ocfs2_triggers {
494 struct jbd2_buffer_trigger_type ot_triggers;
498 static inline struct ocfs2_triggers *to_ocfs2_trigger(struct jbd2_buffer_trigger_type *triggers)
500 return container_of(triggers, struct ocfs2_triggers, ot_triggers);
503 static void ocfs2_frozen_trigger(struct jbd2_buffer_trigger_type *triggers,
504 struct buffer_head *bh,
505 void *data, size_t size)
507 struct ocfs2_triggers *ot = to_ocfs2_trigger(triggers);
510 * We aren't guaranteed to have the superblock here, so we
511 * must unconditionally compute the ecc data.
512 * __ocfs2_journal_access() will only set the triggers if
513 * metaecc is enabled.
515 ocfs2_block_check_compute(data, size, data + ot->ot_offset);
519 * Quota blocks have their own trigger because the struct ocfs2_block_check
520 * offset depends on the blocksize.
522 static void ocfs2_dq_frozen_trigger(struct jbd2_buffer_trigger_type *triggers,
523 struct buffer_head *bh,
524 void *data, size_t size)
526 struct ocfs2_disk_dqtrailer *dqt =
527 ocfs2_block_dqtrailer(size, data);
530 * We aren't guaranteed to have the superblock here, so we
531 * must unconditionally compute the ecc data.
532 * __ocfs2_journal_access() will only set the triggers if
533 * metaecc is enabled.
535 ocfs2_block_check_compute(data, size, &dqt->dq_check);
539 * Directory blocks also have their own trigger because the
540 * struct ocfs2_block_check offset depends on the blocksize.
542 static void ocfs2_db_frozen_trigger(struct jbd2_buffer_trigger_type *triggers,
543 struct buffer_head *bh,
544 void *data, size_t size)
546 struct ocfs2_dir_block_trailer *trailer =
547 ocfs2_dir_trailer_from_size(size, data);
550 * We aren't guaranteed to have the superblock here, so we
551 * must unconditionally compute the ecc data.
552 * __ocfs2_journal_access() will only set the triggers if
553 * metaecc is enabled.
555 ocfs2_block_check_compute(data, size, &trailer->db_check);
558 static void ocfs2_abort_trigger(struct jbd2_buffer_trigger_type *triggers,
559 struct buffer_head *bh)
562 "ocfs2_abort_trigger called by JBD2. bh = 0x%lx, "
563 "bh->b_blocknr = %llu\n",
565 (unsigned long long)bh->b_blocknr);
567 /* We aren't guaranteed to have the superblock here - but if we
568 * don't, it'll just crash. */
569 ocfs2_error(bh->b_assoc_map->host->i_sb,
570 "JBD2 has aborted our journal, ocfs2 cannot continue\n");
573 static struct ocfs2_triggers di_triggers = {
575 .t_frozen = ocfs2_frozen_trigger,
576 .t_abort = ocfs2_abort_trigger,
578 .ot_offset = offsetof(struct ocfs2_dinode, i_check),
581 static struct ocfs2_triggers eb_triggers = {
583 .t_frozen = ocfs2_frozen_trigger,
584 .t_abort = ocfs2_abort_trigger,
586 .ot_offset = offsetof(struct ocfs2_extent_block, h_check),
589 static struct ocfs2_triggers rb_triggers = {
591 .t_frozen = ocfs2_frozen_trigger,
592 .t_abort = ocfs2_abort_trigger,
594 .ot_offset = offsetof(struct ocfs2_refcount_block, rf_check),
597 static struct ocfs2_triggers gd_triggers = {
599 .t_frozen = ocfs2_frozen_trigger,
600 .t_abort = ocfs2_abort_trigger,
602 .ot_offset = offsetof(struct ocfs2_group_desc, bg_check),
605 static struct ocfs2_triggers db_triggers = {
607 .t_frozen = ocfs2_db_frozen_trigger,
608 .t_abort = ocfs2_abort_trigger,
612 static struct ocfs2_triggers xb_triggers = {
614 .t_frozen = ocfs2_frozen_trigger,
615 .t_abort = ocfs2_abort_trigger,
617 .ot_offset = offsetof(struct ocfs2_xattr_block, xb_check),
620 static struct ocfs2_triggers dq_triggers = {
622 .t_frozen = ocfs2_dq_frozen_trigger,
623 .t_abort = ocfs2_abort_trigger,
627 static struct ocfs2_triggers dr_triggers = {
629 .t_frozen = ocfs2_frozen_trigger,
630 .t_abort = ocfs2_abort_trigger,
632 .ot_offset = offsetof(struct ocfs2_dx_root_block, dr_check),
635 static struct ocfs2_triggers dl_triggers = {
637 .t_frozen = ocfs2_frozen_trigger,
638 .t_abort = ocfs2_abort_trigger,
640 .ot_offset = offsetof(struct ocfs2_dx_leaf, dl_check),
643 static int __ocfs2_journal_access(handle_t *handle,
644 struct ocfs2_caching_info *ci,
645 struct buffer_head *bh,
646 struct ocfs2_triggers *triggers,
650 struct ocfs2_super *osb =
651 OCFS2_SB(ocfs2_metadata_cache_get_super(ci));
653 BUG_ON(!ci || !ci->ci_ops);
657 trace_ocfs2_journal_access(
658 (unsigned long long)ocfs2_metadata_cache_owner(ci),
659 (unsigned long long)bh->b_blocknr, type, bh->b_size);
661 /* we can safely remove this assertion after testing. */
662 if (!buffer_uptodate(bh)) {
663 mlog(ML_ERROR, "giving me a buffer that's not uptodate!\n");
664 mlog(ML_ERROR, "b_blocknr=%llu\n",
665 (unsigned long long)bh->b_blocknr);
669 /* Set the current transaction information on the ci so
670 * that the locking code knows whether it can drop it's locks
671 * on this ci or not. We're protected from the commit
672 * thread updating the current transaction id until
673 * ocfs2_commit_trans() because ocfs2_start_trans() took
674 * j_trans_barrier for us. */
675 ocfs2_set_ci_lock_trans(osb->journal, ci);
677 ocfs2_metadata_cache_io_lock(ci);
679 case OCFS2_JOURNAL_ACCESS_CREATE:
680 case OCFS2_JOURNAL_ACCESS_WRITE:
681 status = jbd2_journal_get_write_access(handle, bh);
684 case OCFS2_JOURNAL_ACCESS_UNDO:
685 status = jbd2_journal_get_undo_access(handle, bh);
690 mlog(ML_ERROR, "Unknown access type!\n");
692 if (!status && ocfs2_meta_ecc(osb) && triggers)
693 jbd2_journal_set_triggers(bh, &triggers->ot_triggers);
694 ocfs2_metadata_cache_io_unlock(ci);
697 mlog(ML_ERROR, "Error %d getting %d access to buffer!\n",
703 int ocfs2_journal_access_di(handle_t *handle, struct ocfs2_caching_info *ci,
704 struct buffer_head *bh, int type)
706 return __ocfs2_journal_access(handle, ci, bh, &di_triggers, type);
709 int ocfs2_journal_access_eb(handle_t *handle, struct ocfs2_caching_info *ci,
710 struct buffer_head *bh, int type)
712 return __ocfs2_journal_access(handle, ci, bh, &eb_triggers, type);
715 int ocfs2_journal_access_rb(handle_t *handle, struct ocfs2_caching_info *ci,
716 struct buffer_head *bh, int type)
718 return __ocfs2_journal_access(handle, ci, bh, &rb_triggers,
722 int ocfs2_journal_access_gd(handle_t *handle, struct ocfs2_caching_info *ci,
723 struct buffer_head *bh, int type)
725 return __ocfs2_journal_access(handle, ci, bh, &gd_triggers, type);
728 int ocfs2_journal_access_db(handle_t *handle, struct ocfs2_caching_info *ci,
729 struct buffer_head *bh, int type)
731 return __ocfs2_journal_access(handle, ci, bh, &db_triggers, type);
734 int ocfs2_journal_access_xb(handle_t *handle, struct ocfs2_caching_info *ci,
735 struct buffer_head *bh, int type)
737 return __ocfs2_journal_access(handle, ci, bh, &xb_triggers, type);
740 int ocfs2_journal_access_dq(handle_t *handle, struct ocfs2_caching_info *ci,
741 struct buffer_head *bh, int type)
743 return __ocfs2_journal_access(handle, ci, bh, &dq_triggers, type);
746 int ocfs2_journal_access_dr(handle_t *handle, struct ocfs2_caching_info *ci,
747 struct buffer_head *bh, int type)
749 return __ocfs2_journal_access(handle, ci, bh, &dr_triggers, type);
752 int ocfs2_journal_access_dl(handle_t *handle, struct ocfs2_caching_info *ci,
753 struct buffer_head *bh, int type)
755 return __ocfs2_journal_access(handle, ci, bh, &dl_triggers, type);
758 int ocfs2_journal_access(handle_t *handle, struct ocfs2_caching_info *ci,
759 struct buffer_head *bh, int type)
761 return __ocfs2_journal_access(handle, ci, bh, NULL, type);
764 void ocfs2_journal_dirty(handle_t *handle, struct buffer_head *bh)
768 trace_ocfs2_journal_dirty((unsigned long long)bh->b_blocknr);
770 status = jbd2_journal_dirty_metadata(handle, bh);
774 #define OCFS2_DEFAULT_COMMIT_INTERVAL (HZ * JBD2_DEFAULT_MAX_COMMIT_AGE)
776 void ocfs2_set_journal_params(struct ocfs2_super *osb)
778 journal_t *journal = osb->journal->j_journal;
779 unsigned long commit_interval = OCFS2_DEFAULT_COMMIT_INTERVAL;
781 if (osb->osb_commit_interval)
782 commit_interval = osb->osb_commit_interval;
784 write_lock(&journal->j_state_lock);
785 journal->j_commit_interval = commit_interval;
786 if (osb->s_mount_opt & OCFS2_MOUNT_BARRIER)
787 journal->j_flags |= JBD2_BARRIER;
789 journal->j_flags &= ~JBD2_BARRIER;
790 write_unlock(&journal->j_state_lock);
793 int ocfs2_journal_init(struct ocfs2_journal *journal, int *dirty)
796 struct inode *inode = NULL; /* the journal inode */
797 journal_t *j_journal = NULL;
798 struct ocfs2_dinode *di = NULL;
799 struct buffer_head *bh = NULL;
800 struct ocfs2_super *osb;
805 osb = journal->j_osb;
807 /* already have the inode for our journal */
808 inode = ocfs2_get_system_file_inode(osb, JOURNAL_SYSTEM_INODE,
815 if (is_bad_inode(inode)) {
816 mlog(ML_ERROR, "access error (bad inode)\n");
823 SET_INODE_JOURNAL(inode);
824 OCFS2_I(inode)->ip_open_count++;
826 /* Skip recovery waits here - journal inode metadata never
827 * changes in a live cluster so it can be considered an
828 * exception to the rule. */
829 status = ocfs2_inode_lock_full(inode, &bh, 1, OCFS2_META_LOCK_RECOVERY);
831 if (status != -ERESTARTSYS)
832 mlog(ML_ERROR, "Could not get lock on journal!\n");
837 di = (struct ocfs2_dinode *)bh->b_data;
839 if (i_size_read(inode) < OCFS2_MIN_JOURNAL_SIZE) {
840 mlog(ML_ERROR, "Journal file size (%lld) is too small!\n",
846 trace_ocfs2_journal_init(i_size_read(inode),
847 (unsigned long long)inode->i_blocks,
848 OCFS2_I(inode)->ip_clusters);
850 /* call the kernels journal init function now */
851 j_journal = jbd2_journal_init_inode(inode);
852 if (j_journal == NULL) {
853 mlog(ML_ERROR, "Linux journal layer error\n");
858 trace_ocfs2_journal_init_maxlen(j_journal->j_maxlen);
860 *dirty = (le32_to_cpu(di->id1.journal1.ij_flags) &
861 OCFS2_JOURNAL_DIRTY_FL);
863 journal->j_journal = j_journal;
864 journal->j_inode = inode;
867 ocfs2_set_journal_params(osb);
869 journal->j_state = OCFS2_JOURNAL_LOADED;
875 ocfs2_inode_unlock(inode, 1);
878 OCFS2_I(inode)->ip_open_count--;
886 static void ocfs2_bump_recovery_generation(struct ocfs2_dinode *di)
888 le32_add_cpu(&(di->id1.journal1.ij_recovery_generation), 1);
891 static u32 ocfs2_get_recovery_generation(struct ocfs2_dinode *di)
893 return le32_to_cpu(di->id1.journal1.ij_recovery_generation);
896 static int ocfs2_journal_toggle_dirty(struct ocfs2_super *osb,
897 int dirty, int replayed)
901 struct ocfs2_journal *journal = osb->journal;
902 struct buffer_head *bh = journal->j_bh;
903 struct ocfs2_dinode *fe;
905 fe = (struct ocfs2_dinode *)bh->b_data;
907 /* The journal bh on the osb always comes from ocfs2_journal_init()
908 * and was validated there inside ocfs2_inode_lock_full(). It's a
909 * code bug if we mess it up. */
910 BUG_ON(!OCFS2_IS_VALID_DINODE(fe));
912 flags = le32_to_cpu(fe->id1.journal1.ij_flags);
914 flags |= OCFS2_JOURNAL_DIRTY_FL;
916 flags &= ~OCFS2_JOURNAL_DIRTY_FL;
917 fe->id1.journal1.ij_flags = cpu_to_le32(flags);
920 ocfs2_bump_recovery_generation(fe);
922 ocfs2_compute_meta_ecc(osb->sb, bh->b_data, &fe->i_check);
923 status = ocfs2_write_block(osb, bh, INODE_CACHE(journal->j_inode));
931 * If the journal has been kmalloc'd it needs to be freed after this
934 void ocfs2_journal_shutdown(struct ocfs2_super *osb)
936 struct ocfs2_journal *journal = NULL;
938 struct inode *inode = NULL;
939 int num_running_trans = 0;
943 journal = osb->journal;
947 inode = journal->j_inode;
949 if (journal->j_state != OCFS2_JOURNAL_LOADED)
952 /* need to inc inode use count - jbd2_journal_destroy will iput. */
956 num_running_trans = atomic_read(&(osb->journal->j_num_trans));
957 trace_ocfs2_journal_shutdown(num_running_trans);
959 /* Do a commit_cache here. It will flush our journal, *and*
960 * release any locks that are still held.
961 * set the SHUTDOWN flag and release the trans lock.
962 * the commit thread will take the trans lock for us below. */
963 journal->j_state = OCFS2_JOURNAL_IN_SHUTDOWN;
965 /* The OCFS2_JOURNAL_IN_SHUTDOWN will signal to commit_cache to not
966 * drop the trans_lock (which we want to hold until we
967 * completely destroy the journal. */
968 if (osb->commit_task) {
969 /* Wait for the commit thread */
970 trace_ocfs2_journal_shutdown_wait(osb->commit_task);
971 kthread_stop(osb->commit_task);
972 osb->commit_task = NULL;
975 BUG_ON(atomic_read(&(osb->journal->j_num_trans)) != 0);
977 if (ocfs2_mount_local(osb)) {
978 jbd2_journal_lock_updates(journal->j_journal);
979 status = jbd2_journal_flush(journal->j_journal);
980 jbd2_journal_unlock_updates(journal->j_journal);
987 * Do not toggle if flush was unsuccessful otherwise
988 * will leave dirty metadata in a "clean" journal
990 status = ocfs2_journal_toggle_dirty(osb, 0, 0);
995 /* Shutdown the kernel journal system */
996 jbd2_journal_destroy(journal->j_journal);
997 journal->j_journal = NULL;
999 OCFS2_I(inode)->ip_open_count--;
1001 /* unlock our journal */
1002 ocfs2_inode_unlock(inode, 1);
1004 brelse(journal->j_bh);
1005 journal->j_bh = NULL;
1007 journal->j_state = OCFS2_JOURNAL_FREE;
1009 // up_write(&journal->j_trans_barrier);
1015 static void ocfs2_clear_journal_error(struct super_block *sb,
1021 olderr = jbd2_journal_errno(journal);
1023 mlog(ML_ERROR, "File system error %d recorded in "
1024 "journal %u.\n", olderr, slot);
1025 mlog(ML_ERROR, "File system on device %s needs checking.\n",
1028 jbd2_journal_ack_err(journal);
1029 jbd2_journal_clear_err(journal);
1033 int ocfs2_journal_load(struct ocfs2_journal *journal, int local, int replayed)
1036 struct ocfs2_super *osb;
1040 osb = journal->j_osb;
1042 status = jbd2_journal_load(journal->j_journal);
1044 mlog(ML_ERROR, "Failed to load journal!\n");
1048 ocfs2_clear_journal_error(osb->sb, journal->j_journal, osb->slot_num);
1050 status = ocfs2_journal_toggle_dirty(osb, 1, replayed);
1056 /* Launch the commit thread */
1058 osb->commit_task = kthread_run(ocfs2_commit_thread, osb,
1060 if (IS_ERR(osb->commit_task)) {
1061 status = PTR_ERR(osb->commit_task);
1062 osb->commit_task = NULL;
1063 mlog(ML_ERROR, "unable to launch ocfs2commit thread, "
1064 "error=%d", status);
1068 osb->commit_task = NULL;
1075 /* 'full' flag tells us whether we clear out all blocks or if we just
1076 * mark the journal clean */
1077 int ocfs2_journal_wipe(struct ocfs2_journal *journal, int full)
1083 status = jbd2_journal_wipe(journal->j_journal, full);
1089 status = ocfs2_journal_toggle_dirty(journal->j_osb, 0, 0);
1097 static int ocfs2_recovery_completed(struct ocfs2_super *osb)
1100 struct ocfs2_recovery_map *rm = osb->recovery_map;
1102 spin_lock(&osb->osb_lock);
1103 empty = (rm->rm_used == 0);
1104 spin_unlock(&osb->osb_lock);
1109 void ocfs2_wait_for_recovery(struct ocfs2_super *osb)
1111 wait_event(osb->recovery_event, ocfs2_recovery_completed(osb));
1115 * JBD Might read a cached version of another nodes journal file. We
1116 * don't want this as this file changes often and we get no
1117 * notification on those changes. The only way to be sure that we've
1118 * got the most up to date version of those blocks then is to force
1119 * read them off disk. Just searching through the buffer cache won't
1120 * work as there may be pages backing this file which are still marked
1121 * up to date. We know things can't change on this file underneath us
1122 * as we have the lock by now :)
1124 static int ocfs2_force_read_journal(struct inode *inode)
1128 u64 v_blkno, p_blkno, p_blocks, num_blocks;
1129 #define CONCURRENT_JOURNAL_FILL 32ULL
1130 struct buffer_head *bhs[CONCURRENT_JOURNAL_FILL];
1132 memset(bhs, 0, sizeof(struct buffer_head *) * CONCURRENT_JOURNAL_FILL);
1134 num_blocks = ocfs2_blocks_for_bytes(inode->i_sb, i_size_read(inode));
1136 while (v_blkno < num_blocks) {
1137 status = ocfs2_extent_map_get_blocks(inode, v_blkno,
1138 &p_blkno, &p_blocks, NULL);
1144 if (p_blocks > CONCURRENT_JOURNAL_FILL)
1145 p_blocks = CONCURRENT_JOURNAL_FILL;
1147 /* We are reading journal data which should not
1148 * be put in the uptodate cache */
1149 status = ocfs2_read_blocks_sync(OCFS2_SB(inode->i_sb),
1150 p_blkno, p_blocks, bhs);
1156 for(i = 0; i < p_blocks; i++) {
1161 v_blkno += p_blocks;
1165 for(i = 0; i < CONCURRENT_JOURNAL_FILL; i++)
1170 struct ocfs2_la_recovery_item {
1171 struct list_head lri_list;
1173 struct ocfs2_dinode *lri_la_dinode;
1174 struct ocfs2_dinode *lri_tl_dinode;
1175 struct ocfs2_quota_recovery *lri_qrec;
1178 /* Does the second half of the recovery process. By this point, the
1179 * node is marked clean and can actually be considered recovered,
1180 * hence it's no longer in the recovery map, but there's still some
1181 * cleanup we can do which shouldn't happen within the recovery thread
1182 * as locking in that context becomes very difficult if we are to take
1183 * recovering nodes into account.
1185 * NOTE: This function can and will sleep on recovery of other nodes
1186 * during cluster locking, just like any other ocfs2 process.
1188 void ocfs2_complete_recovery(struct work_struct *work)
1191 struct ocfs2_journal *journal =
1192 container_of(work, struct ocfs2_journal, j_recovery_work);
1193 struct ocfs2_super *osb = journal->j_osb;
1194 struct ocfs2_dinode *la_dinode, *tl_dinode;
1195 struct ocfs2_la_recovery_item *item, *n;
1196 struct ocfs2_quota_recovery *qrec;
1197 LIST_HEAD(tmp_la_list);
1199 trace_ocfs2_complete_recovery(
1200 (unsigned long long)OCFS2_I(journal->j_inode)->ip_blkno);
1202 spin_lock(&journal->j_lock);
1203 list_splice_init(&journal->j_la_cleanups, &tmp_la_list);
1204 spin_unlock(&journal->j_lock);
1206 list_for_each_entry_safe(item, n, &tmp_la_list, lri_list) {
1207 list_del_init(&item->lri_list);
1209 ocfs2_wait_on_quotas(osb);
1211 la_dinode = item->lri_la_dinode;
1212 tl_dinode = item->lri_tl_dinode;
1213 qrec = item->lri_qrec;
1215 trace_ocfs2_complete_recovery_slot(item->lri_slot,
1216 la_dinode ? le64_to_cpu(la_dinode->i_blkno) : 0,
1217 tl_dinode ? le64_to_cpu(tl_dinode->i_blkno) : 0,
1221 ret = ocfs2_complete_local_alloc_recovery(osb,
1230 ret = ocfs2_complete_truncate_log_recovery(osb,
1238 ret = ocfs2_recover_orphans(osb, item->lri_slot);
1243 ret = ocfs2_finish_quota_recovery(osb, qrec,
1247 /* Recovery info is already freed now */
1253 trace_ocfs2_complete_recovery_end(ret);
1256 /* NOTE: This function always eats your references to la_dinode and
1257 * tl_dinode, either manually on error, or by passing them to
1258 * ocfs2_complete_recovery */
1259 static void ocfs2_queue_recovery_completion(struct ocfs2_journal *journal,
1261 struct ocfs2_dinode *la_dinode,
1262 struct ocfs2_dinode *tl_dinode,
1263 struct ocfs2_quota_recovery *qrec)
1265 struct ocfs2_la_recovery_item *item;
1267 item = kmalloc(sizeof(struct ocfs2_la_recovery_item), GFP_NOFS);
1269 /* Though we wish to avoid it, we are in fact safe in
1270 * skipping local alloc cleanup as fsck.ocfs2 is more
1271 * than capable of reclaiming unused space. */
1276 ocfs2_free_quota_recovery(qrec);
1278 mlog_errno(-ENOMEM);
1282 INIT_LIST_HEAD(&item->lri_list);
1283 item->lri_la_dinode = la_dinode;
1284 item->lri_slot = slot_num;
1285 item->lri_tl_dinode = tl_dinode;
1286 item->lri_qrec = qrec;
1288 spin_lock(&journal->j_lock);
1289 list_add_tail(&item->lri_list, &journal->j_la_cleanups);
1290 queue_work(ocfs2_wq, &journal->j_recovery_work);
1291 spin_unlock(&journal->j_lock);
1294 /* Called by the mount code to queue recovery the last part of
1295 * recovery for it's own and offline slot(s). */
1296 void ocfs2_complete_mount_recovery(struct ocfs2_super *osb)
1298 struct ocfs2_journal *journal = osb->journal;
1300 if (ocfs2_is_hard_readonly(osb))
1303 /* No need to queue up our truncate_log as regular cleanup will catch
1305 ocfs2_queue_recovery_completion(journal, osb->slot_num,
1306 osb->local_alloc_copy, NULL, NULL);
1307 ocfs2_schedule_truncate_log_flush(osb, 0);
1309 osb->local_alloc_copy = NULL;
1312 /* queue to recover orphan slots for all offline slots */
1313 ocfs2_replay_map_set_state(osb, REPLAY_NEEDED);
1314 ocfs2_queue_replay_slots(osb);
1315 ocfs2_free_replay_slots(osb);
1318 void ocfs2_complete_quota_recovery(struct ocfs2_super *osb)
1320 if (osb->quota_rec) {
1321 ocfs2_queue_recovery_completion(osb->journal,
1326 osb->quota_rec = NULL;
1330 static int __ocfs2_recovery_thread(void *arg)
1332 int status, node_num, slot_num;
1333 struct ocfs2_super *osb = arg;
1334 struct ocfs2_recovery_map *rm = osb->recovery_map;
1335 int *rm_quota = NULL;
1336 int rm_quota_used = 0, i;
1337 struct ocfs2_quota_recovery *qrec;
1339 status = ocfs2_wait_on_mount(osb);
1344 rm_quota = kzalloc(osb->max_slots * sizeof(int), GFP_NOFS);
1350 status = ocfs2_super_lock(osb, 1);
1356 status = ocfs2_compute_replay_slots(osb);
1360 /* queue recovery for our own slot */
1361 ocfs2_queue_recovery_completion(osb->journal, osb->slot_num, NULL,
1364 spin_lock(&osb->osb_lock);
1365 while (rm->rm_used) {
1366 /* It's always safe to remove entry zero, as we won't
1367 * clear it until ocfs2_recover_node() has succeeded. */
1368 node_num = rm->rm_entries[0];
1369 spin_unlock(&osb->osb_lock);
1370 slot_num = ocfs2_node_num_to_slot(osb, node_num);
1371 trace_ocfs2_recovery_thread_node(node_num, slot_num);
1372 if (slot_num == -ENOENT) {
1377 /* It is a bit subtle with quota recovery. We cannot do it
1378 * immediately because we have to obtain cluster locks from
1379 * quota files and we also don't want to just skip it because
1380 * then quota usage would be out of sync until some node takes
1381 * the slot. So we remember which nodes need quota recovery
1382 * and when everything else is done, we recover quotas. */
1383 for (i = 0; i < rm_quota_used && rm_quota[i] != slot_num; i++);
1384 if (i == rm_quota_used)
1385 rm_quota[rm_quota_used++] = slot_num;
1387 status = ocfs2_recover_node(osb, node_num, slot_num);
1390 ocfs2_recovery_map_clear(osb, node_num);
1393 "Error %d recovering node %d on device (%u,%u)!\n",
1395 MAJOR(osb->sb->s_dev), MINOR(osb->sb->s_dev));
1396 mlog(ML_ERROR, "Volume requires unmount.\n");
1399 spin_lock(&osb->osb_lock);
1401 spin_unlock(&osb->osb_lock);
1402 trace_ocfs2_recovery_thread_end(status);
1404 /* Refresh all journal recovery generations from disk */
1405 status = ocfs2_check_journals_nolocks(osb);
1406 status = (status == -EROFS) ? 0 : status;
1410 /* Now it is right time to recover quotas... We have to do this under
1411 * superblock lock so that no one can start using the slot (and crash)
1412 * before we recover it */
1413 for (i = 0; i < rm_quota_used; i++) {
1414 qrec = ocfs2_begin_quota_recovery(osb, rm_quota[i]);
1416 status = PTR_ERR(qrec);
1420 ocfs2_queue_recovery_completion(osb->journal, rm_quota[i],
1424 ocfs2_super_unlock(osb, 1);
1426 /* queue recovery for offline slots */
1427 ocfs2_queue_replay_slots(osb);
1430 mutex_lock(&osb->recovery_lock);
1431 if (!status && !ocfs2_recovery_completed(osb)) {
1432 mutex_unlock(&osb->recovery_lock);
1436 ocfs2_free_replay_slots(osb);
1437 osb->recovery_thread_task = NULL;
1438 mb(); /* sync with ocfs2_recovery_thread_running */
1439 wake_up(&osb->recovery_event);
1441 mutex_unlock(&osb->recovery_lock);
1445 /* no one is callint kthread_stop() for us so the kthread() api
1446 * requires that we call do_exit(). And it isn't exported, but
1447 * complete_and_exit() seems to be a minimal wrapper around it. */
1448 complete_and_exit(NULL, status);
1452 void ocfs2_recovery_thread(struct ocfs2_super *osb, int node_num)
1454 mutex_lock(&osb->recovery_lock);
1456 trace_ocfs2_recovery_thread(node_num, osb->node_num,
1457 osb->disable_recovery, osb->recovery_thread_task,
1458 osb->disable_recovery ?
1459 -1 : ocfs2_recovery_map_set(osb, node_num));
1461 if (osb->disable_recovery)
1464 if (osb->recovery_thread_task)
1467 osb->recovery_thread_task = kthread_run(__ocfs2_recovery_thread, osb,
1469 if (IS_ERR(osb->recovery_thread_task)) {
1470 mlog_errno((int)PTR_ERR(osb->recovery_thread_task));
1471 osb->recovery_thread_task = NULL;
1475 mutex_unlock(&osb->recovery_lock);
1476 wake_up(&osb->recovery_event);
1479 static int ocfs2_read_journal_inode(struct ocfs2_super *osb,
1481 struct buffer_head **bh,
1482 struct inode **ret_inode)
1484 int status = -EACCES;
1485 struct inode *inode = NULL;
1487 BUG_ON(slot_num >= osb->max_slots);
1489 inode = ocfs2_get_system_file_inode(osb, JOURNAL_SYSTEM_INODE,
1491 if (!inode || is_bad_inode(inode)) {
1495 SET_INODE_JOURNAL(inode);
1497 status = ocfs2_read_inode_block_full(inode, bh, OCFS2_BH_IGNORE_CACHE);
1507 if (status || !ret_inode)
1515 /* Does the actual journal replay and marks the journal inode as
1516 * clean. Will only replay if the journal inode is marked dirty. */
1517 static int ocfs2_replay_journal(struct ocfs2_super *osb,
1524 struct inode *inode = NULL;
1525 struct ocfs2_dinode *fe;
1526 journal_t *journal = NULL;
1527 struct buffer_head *bh = NULL;
1530 status = ocfs2_read_journal_inode(osb, slot_num, &bh, &inode);
1536 fe = (struct ocfs2_dinode *)bh->b_data;
1537 slot_reco_gen = ocfs2_get_recovery_generation(fe);
1542 * As the fs recovery is asynchronous, there is a small chance that
1543 * another node mounted (and recovered) the slot before the recovery
1544 * thread could get the lock. To handle that, we dirty read the journal
1545 * inode for that slot to get the recovery generation. If it is
1546 * different than what we expected, the slot has been recovered.
1547 * If not, it needs recovery.
1549 if (osb->slot_recovery_generations[slot_num] != slot_reco_gen) {
1550 trace_ocfs2_replay_journal_recovered(slot_num,
1551 osb->slot_recovery_generations[slot_num], slot_reco_gen);
1552 osb->slot_recovery_generations[slot_num] = slot_reco_gen;
1557 /* Continue with recovery as the journal has not yet been recovered */
1559 status = ocfs2_inode_lock_full(inode, &bh, 1, OCFS2_META_LOCK_RECOVERY);
1561 trace_ocfs2_replay_journal_lock_err(status);
1562 if (status != -ERESTARTSYS)
1563 mlog(ML_ERROR, "Could not lock journal!\n");
1568 fe = (struct ocfs2_dinode *) bh->b_data;
1570 flags = le32_to_cpu(fe->id1.journal1.ij_flags);
1571 slot_reco_gen = ocfs2_get_recovery_generation(fe);
1573 if (!(flags & OCFS2_JOURNAL_DIRTY_FL)) {
1574 trace_ocfs2_replay_journal_skip(node_num);
1575 /* Refresh recovery generation for the slot */
1576 osb->slot_recovery_generations[slot_num] = slot_reco_gen;
1580 /* we need to run complete recovery for offline orphan slots */
1581 ocfs2_replay_map_set_state(osb, REPLAY_NEEDED);
1583 printk(KERN_NOTICE "ocfs2: Begin replay journal (node %d, slot %d) on "\
1584 "device (%u,%u)\n", node_num, slot_num, MAJOR(osb->sb->s_dev),
1585 MINOR(osb->sb->s_dev));
1587 OCFS2_I(inode)->ip_clusters = le32_to_cpu(fe->i_clusters);
1589 status = ocfs2_force_read_journal(inode);
1595 journal = jbd2_journal_init_inode(inode);
1596 if (journal == NULL) {
1597 mlog(ML_ERROR, "Linux journal layer error\n");
1602 status = jbd2_journal_load(journal);
1607 jbd2_journal_destroy(journal);
1611 ocfs2_clear_journal_error(osb->sb, journal, slot_num);
1613 /* wipe the journal */
1614 jbd2_journal_lock_updates(journal);
1615 status = jbd2_journal_flush(journal);
1616 jbd2_journal_unlock_updates(journal);
1620 /* This will mark the node clean */
1621 flags = le32_to_cpu(fe->id1.journal1.ij_flags);
1622 flags &= ~OCFS2_JOURNAL_DIRTY_FL;
1623 fe->id1.journal1.ij_flags = cpu_to_le32(flags);
1625 /* Increment recovery generation to indicate successful recovery */
1626 ocfs2_bump_recovery_generation(fe);
1627 osb->slot_recovery_generations[slot_num] =
1628 ocfs2_get_recovery_generation(fe);
1630 ocfs2_compute_meta_ecc(osb->sb, bh->b_data, &fe->i_check);
1631 status = ocfs2_write_block(osb, bh, INODE_CACHE(inode));
1638 jbd2_journal_destroy(journal);
1640 printk(KERN_NOTICE "ocfs2: End replay journal (node %d, slot %d) on "\
1641 "device (%u,%u)\n", node_num, slot_num, MAJOR(osb->sb->s_dev),
1642 MINOR(osb->sb->s_dev));
1644 /* drop the lock on this nodes journal */
1646 ocfs2_inode_unlock(inode, 1);
1657 * Do the most important parts of node recovery:
1658 * - Replay it's journal
1659 * - Stamp a clean local allocator file
1660 * - Stamp a clean truncate log
1661 * - Mark the node clean
1663 * If this function completes without error, a node in OCFS2 can be
1664 * said to have been safely recovered. As a result, failure during the
1665 * second part of a nodes recovery process (local alloc recovery) is
1666 * far less concerning.
1668 static int ocfs2_recover_node(struct ocfs2_super *osb,
1669 int node_num, int slot_num)
1672 struct ocfs2_dinode *la_copy = NULL;
1673 struct ocfs2_dinode *tl_copy = NULL;
1675 trace_ocfs2_recover_node(node_num, slot_num, osb->node_num);
1677 /* Should not ever be called to recover ourselves -- in that
1678 * case we should've called ocfs2_journal_load instead. */
1679 BUG_ON(osb->node_num == node_num);
1681 status = ocfs2_replay_journal(osb, node_num, slot_num);
1683 if (status == -EBUSY) {
1684 trace_ocfs2_recover_node_skip(slot_num, node_num);
1692 /* Stamp a clean local alloc file AFTER recovering the journal... */
1693 status = ocfs2_begin_local_alloc_recovery(osb, slot_num, &la_copy);
1699 /* An error from begin_truncate_log_recovery is not
1700 * serious enough to warrant halting the rest of
1702 status = ocfs2_begin_truncate_log_recovery(osb, slot_num, &tl_copy);
1706 /* Likewise, this would be a strange but ultimately not so
1707 * harmful place to get an error... */
1708 status = ocfs2_clear_slot(osb, slot_num);
1712 /* This will kfree the memory pointed to by la_copy and tl_copy */
1713 ocfs2_queue_recovery_completion(osb->journal, slot_num, la_copy,
1722 /* Test node liveness by trylocking his journal. If we get the lock,
1723 * we drop it here. Return 0 if we got the lock, -EAGAIN if node is
1724 * still alive (we couldn't get the lock) and < 0 on error. */
1725 static int ocfs2_trylock_journal(struct ocfs2_super *osb,
1729 struct inode *inode = NULL;
1731 inode = ocfs2_get_system_file_inode(osb, JOURNAL_SYSTEM_INODE,
1733 if (inode == NULL) {
1734 mlog(ML_ERROR, "access error\n");
1738 if (is_bad_inode(inode)) {
1739 mlog(ML_ERROR, "access error (bad inode)\n");
1745 SET_INODE_JOURNAL(inode);
1747 flags = OCFS2_META_LOCK_RECOVERY | OCFS2_META_LOCK_NOQUEUE;
1748 status = ocfs2_inode_lock_full(inode, NULL, 1, flags);
1750 if (status != -EAGAIN)
1755 ocfs2_inode_unlock(inode, 1);
1763 /* Call this underneath ocfs2_super_lock. It also assumes that the
1764 * slot info struct has been updated from disk. */
1765 int ocfs2_mark_dead_nodes(struct ocfs2_super *osb)
1767 unsigned int node_num;
1770 struct buffer_head *bh = NULL;
1771 struct ocfs2_dinode *di;
1773 /* This is called with the super block cluster lock, so we
1774 * know that the slot map can't change underneath us. */
1776 for (i = 0; i < osb->max_slots; i++) {
1777 /* Read journal inode to get the recovery generation */
1778 status = ocfs2_read_journal_inode(osb, i, &bh, NULL);
1783 di = (struct ocfs2_dinode *)bh->b_data;
1784 gen = ocfs2_get_recovery_generation(di);
1788 spin_lock(&osb->osb_lock);
1789 osb->slot_recovery_generations[i] = gen;
1791 trace_ocfs2_mark_dead_nodes(i,
1792 osb->slot_recovery_generations[i]);
1794 if (i == osb->slot_num) {
1795 spin_unlock(&osb->osb_lock);
1799 status = ocfs2_slot_to_node_num_locked(osb, i, &node_num);
1800 if (status == -ENOENT) {
1801 spin_unlock(&osb->osb_lock);
1805 if (__ocfs2_recovery_map_test(osb, node_num)) {
1806 spin_unlock(&osb->osb_lock);
1809 spin_unlock(&osb->osb_lock);
1811 /* Ok, we have a slot occupied by another node which
1812 * is not in the recovery map. We trylock his journal
1813 * file here to test if he's alive. */
1814 status = ocfs2_trylock_journal(osb, i);
1816 /* Since we're called from mount, we know that
1817 * the recovery thread can't race us on
1818 * setting / checking the recovery bits. */
1819 ocfs2_recovery_thread(osb, node_num);
1820 } else if ((status < 0) && (status != -EAGAIN)) {
1832 * Scan timer should get fired every ORPHAN_SCAN_SCHEDULE_TIMEOUT. Add some
1833 * randomness to the timeout to minimize multple nodes firing the timer at the
1836 static inline unsigned long ocfs2_orphan_scan_timeout(void)
1840 get_random_bytes(&time, sizeof(time));
1841 time = ORPHAN_SCAN_SCHEDULE_TIMEOUT + (time % 5000);
1842 return msecs_to_jiffies(time);
1846 * ocfs2_queue_orphan_scan calls ocfs2_queue_recovery_completion for
1847 * every slot, queuing a recovery of the slot on the ocfs2_wq thread. This
1848 * is done to catch any orphans that are left over in orphan directories.
1850 * It scans all slots, even ones that are in use. It does so to handle the
1851 * case described below:
1853 * Node 1 has an inode it was using. The dentry went away due to memory
1854 * pressure. Node 1 closes the inode, but it's on the free list. The node
1855 * has the open lock.
1856 * Node 2 unlinks the inode. It grabs the dentry lock to notify others,
1857 * but node 1 has no dentry and doesn't get the message. It trylocks the
1858 * open lock, sees that another node has a PR, and does nothing.
1859 * Later node 2 runs its orphan dir. It igets the inode, trylocks the
1860 * open lock, sees the PR still, and does nothing.
1861 * Basically, we have to trigger an orphan iput on node 1. The only way
1862 * for this to happen is if node 1 runs node 2's orphan dir.
1864 * ocfs2_queue_orphan_scan gets called every ORPHAN_SCAN_SCHEDULE_TIMEOUT
1865 * seconds. It gets an EX lock on os_lockres and checks sequence number
1866 * stored in LVB. If the sequence number has changed, it means some other
1867 * node has done the scan. This node skips the scan and tracks the
1868 * sequence number. If the sequence number didn't change, it means a scan
1869 * hasn't happened. The node queues a scan and increments the
1870 * sequence number in the LVB.
1872 void ocfs2_queue_orphan_scan(struct ocfs2_super *osb)
1874 struct ocfs2_orphan_scan *os;
1878 os = &osb->osb_orphan_scan;
1880 if (atomic_read(&os->os_state) == ORPHAN_SCAN_INACTIVE)
1883 trace_ocfs2_queue_orphan_scan_begin(os->os_count, os->os_seqno,
1884 atomic_read(&os->os_state));
1886 status = ocfs2_orphan_scan_lock(osb, &seqno);
1888 if (status != -EAGAIN)
1893 /* Do no queue the tasks if the volume is being umounted */
1894 if (atomic_read(&os->os_state) == ORPHAN_SCAN_INACTIVE)
1897 if (os->os_seqno != seqno) {
1898 os->os_seqno = seqno;
1902 for (i = 0; i < osb->max_slots; i++)
1903 ocfs2_queue_recovery_completion(osb->journal, i, NULL, NULL,
1906 * We queued a recovery on orphan slots, increment the sequence
1907 * number and update LVB so other node will skip the scan for a while
1911 os->os_scantime = CURRENT_TIME;
1913 ocfs2_orphan_scan_unlock(osb, seqno);
1915 trace_ocfs2_queue_orphan_scan_end(os->os_count, os->os_seqno,
1916 atomic_read(&os->os_state));
1920 /* Worker task that gets fired every ORPHAN_SCAN_SCHEDULE_TIMEOUT millsec */
1921 void ocfs2_orphan_scan_work(struct work_struct *work)
1923 struct ocfs2_orphan_scan *os;
1924 struct ocfs2_super *osb;
1926 os = container_of(work, struct ocfs2_orphan_scan,
1927 os_orphan_scan_work.work);
1930 mutex_lock(&os->os_lock);
1931 ocfs2_queue_orphan_scan(osb);
1932 if (atomic_read(&os->os_state) == ORPHAN_SCAN_ACTIVE)
1933 queue_delayed_work(ocfs2_wq, &os->os_orphan_scan_work,
1934 ocfs2_orphan_scan_timeout());
1935 mutex_unlock(&os->os_lock);
1938 void ocfs2_orphan_scan_stop(struct ocfs2_super *osb)
1940 struct ocfs2_orphan_scan *os;
1942 os = &osb->osb_orphan_scan;
1943 if (atomic_read(&os->os_state) == ORPHAN_SCAN_ACTIVE) {
1944 atomic_set(&os->os_state, ORPHAN_SCAN_INACTIVE);
1945 mutex_lock(&os->os_lock);
1946 cancel_delayed_work(&os->os_orphan_scan_work);
1947 mutex_unlock(&os->os_lock);
1951 void ocfs2_orphan_scan_init(struct ocfs2_super *osb)
1953 struct ocfs2_orphan_scan *os;
1955 os = &osb->osb_orphan_scan;
1959 mutex_init(&os->os_lock);
1960 INIT_DELAYED_WORK(&os->os_orphan_scan_work, ocfs2_orphan_scan_work);
1963 void ocfs2_orphan_scan_start(struct ocfs2_super *osb)
1965 struct ocfs2_orphan_scan *os;
1967 os = &osb->osb_orphan_scan;
1968 os->os_scantime = CURRENT_TIME;
1969 if (ocfs2_is_hard_readonly(osb) || ocfs2_mount_local(osb))
1970 atomic_set(&os->os_state, ORPHAN_SCAN_INACTIVE);
1972 atomic_set(&os->os_state, ORPHAN_SCAN_ACTIVE);
1973 queue_delayed_work(ocfs2_wq, &os->os_orphan_scan_work,
1974 ocfs2_orphan_scan_timeout());
1978 struct ocfs2_orphan_filldir_priv {
1979 struct dir_context ctx;
1981 struct ocfs2_super *osb;
1984 static int ocfs2_orphan_filldir(void *priv, const char *name, int name_len,
1985 loff_t pos, u64 ino, unsigned type)
1987 struct ocfs2_orphan_filldir_priv *p = priv;
1990 if (name_len == 1 && !strncmp(".", name, 1))
1992 if (name_len == 2 && !strncmp("..", name, 2))
1995 /* Skip bad inodes so that recovery can continue */
1996 iter = ocfs2_iget(p->osb, ino,
1997 OCFS2_FI_FLAG_ORPHAN_RECOVERY, 0);
2001 trace_ocfs2_orphan_filldir((unsigned long long)OCFS2_I(iter)->ip_blkno);
2002 /* No locking is required for the next_orphan queue as there
2003 * is only ever a single process doing orphan recovery. */
2004 OCFS2_I(iter)->ip_next_orphan = p->head;
2010 static int ocfs2_queue_orphans(struct ocfs2_super *osb,
2012 struct inode **head)
2015 struct inode *orphan_dir_inode = NULL;
2016 struct ocfs2_orphan_filldir_priv priv = {
2017 .ctx.actor = ocfs2_orphan_filldir,
2022 orphan_dir_inode = ocfs2_get_system_file_inode(osb,
2023 ORPHAN_DIR_SYSTEM_INODE,
2025 if (!orphan_dir_inode) {
2031 mutex_lock(&orphan_dir_inode->i_mutex);
2032 status = ocfs2_inode_lock(orphan_dir_inode, NULL, 0);
2038 status = ocfs2_dir_foreach(orphan_dir_inode, &priv.ctx);
2047 ocfs2_inode_unlock(orphan_dir_inode, 0);
2049 mutex_unlock(&orphan_dir_inode->i_mutex);
2050 iput(orphan_dir_inode);
2054 static int ocfs2_orphan_recovery_can_continue(struct ocfs2_super *osb,
2059 spin_lock(&osb->osb_lock);
2060 ret = !osb->osb_orphan_wipes[slot];
2061 spin_unlock(&osb->osb_lock);
2065 static void ocfs2_mark_recovering_orphan_dir(struct ocfs2_super *osb,
2068 spin_lock(&osb->osb_lock);
2069 /* Mark ourselves such that new processes in delete_inode()
2070 * know to quit early. */
2071 ocfs2_node_map_set_bit(osb, &osb->osb_recovering_orphan_dirs, slot);
2072 while (osb->osb_orphan_wipes[slot]) {
2073 /* If any processes are already in the middle of an
2074 * orphan wipe on this dir, then we need to wait for
2076 spin_unlock(&osb->osb_lock);
2077 wait_event_interruptible(osb->osb_wipe_event,
2078 ocfs2_orphan_recovery_can_continue(osb, slot));
2079 spin_lock(&osb->osb_lock);
2081 spin_unlock(&osb->osb_lock);
2084 static void ocfs2_clear_recovering_orphan_dir(struct ocfs2_super *osb,
2087 ocfs2_node_map_clear_bit(osb, &osb->osb_recovering_orphan_dirs, slot);
2091 * Orphan recovery. Each mounted node has it's own orphan dir which we
2092 * must run during recovery. Our strategy here is to build a list of
2093 * the inodes in the orphan dir and iget/iput them. The VFS does
2094 * (most) of the rest of the work.
2096 * Orphan recovery can happen at any time, not just mount so we have a
2097 * couple of extra considerations.
2099 * - We grab as many inodes as we can under the orphan dir lock -
2100 * doing iget() outside the orphan dir risks getting a reference on
2102 * - We must be sure not to deadlock with other processes on the
2103 * system wanting to run delete_inode(). This can happen when they go
2104 * to lock the orphan dir and the orphan recovery process attempts to
2105 * iget() inside the orphan dir lock. This can be avoided by
2106 * advertising our state to ocfs2_delete_inode().
2108 static int ocfs2_recover_orphans(struct ocfs2_super *osb,
2112 struct inode *inode = NULL;
2114 struct ocfs2_inode_info *oi;
2116 trace_ocfs2_recover_orphans(slot);
2118 ocfs2_mark_recovering_orphan_dir(osb, slot);
2119 ret = ocfs2_queue_orphans(osb, slot, &inode);
2120 ocfs2_clear_recovering_orphan_dir(osb, slot);
2122 /* Error here should be noted, but we want to continue with as
2123 * many queued inodes as we've got. */
2128 oi = OCFS2_I(inode);
2129 trace_ocfs2_recover_orphans_iput(
2130 (unsigned long long)oi->ip_blkno);
2132 iter = oi->ip_next_orphan;
2134 spin_lock(&oi->ip_lock);
2135 /* The remote delete code may have set these on the
2136 * assumption that the other node would wipe them
2137 * successfully. If they are still in the node's
2138 * orphan dir, we need to reset that state. */
2139 oi->ip_flags &= ~(OCFS2_INODE_DELETED|OCFS2_INODE_SKIP_DELETE);
2141 /* Set the proper information to get us going into
2142 * ocfs2_delete_inode. */
2143 oi->ip_flags |= OCFS2_INODE_MAYBE_ORPHANED;
2144 spin_unlock(&oi->ip_lock);
2154 static int __ocfs2_wait_on_mount(struct ocfs2_super *osb, int quota)
2156 /* This check is good because ocfs2 will wait on our recovery
2157 * thread before changing it to something other than MOUNTED
2159 wait_event(osb->osb_mount_event,
2160 (!quota && atomic_read(&osb->vol_state) == VOLUME_MOUNTED) ||
2161 atomic_read(&osb->vol_state) == VOLUME_MOUNTED_QUOTAS ||
2162 atomic_read(&osb->vol_state) == VOLUME_DISABLED);
2164 /* If there's an error on mount, then we may never get to the
2165 * MOUNTED flag, but this is set right before
2166 * dismount_volume() so we can trust it. */
2167 if (atomic_read(&osb->vol_state) == VOLUME_DISABLED) {
2168 trace_ocfs2_wait_on_mount(VOLUME_DISABLED);
2169 mlog(0, "mount error, exiting!\n");
2176 static int ocfs2_commit_thread(void *arg)
2179 struct ocfs2_super *osb = arg;
2180 struct ocfs2_journal *journal = osb->journal;
2182 /* we can trust j_num_trans here because _should_stop() is only set in
2183 * shutdown and nobody other than ourselves should be able to start
2184 * transactions. committing on shutdown might take a few iterations
2185 * as final transactions put deleted inodes on the list */
2186 while (!(kthread_should_stop() &&
2187 atomic_read(&journal->j_num_trans) == 0)) {
2189 wait_event_interruptible(osb->checkpoint_event,
2190 atomic_read(&journal->j_num_trans)
2191 || kthread_should_stop());
2193 status = ocfs2_commit_cache(osb);
2197 if (kthread_should_stop() && atomic_read(&journal->j_num_trans)){
2199 "commit_thread: %u transactions pending on "
2201 atomic_read(&journal->j_num_trans));
2208 /* Reads all the journal inodes without taking any cluster locks. Used
2209 * for hard readonly access to determine whether any journal requires
2210 * recovery. Also used to refresh the recovery generation numbers after
2211 * a journal has been recovered by another node.
2213 int ocfs2_check_journals_nolocks(struct ocfs2_super *osb)
2217 struct buffer_head *di_bh = NULL;
2218 struct ocfs2_dinode *di;
2219 int journal_dirty = 0;
2221 for(slot = 0; slot < osb->max_slots; slot++) {
2222 ret = ocfs2_read_journal_inode(osb, slot, &di_bh, NULL);
2228 di = (struct ocfs2_dinode *) di_bh->b_data;
2230 osb->slot_recovery_generations[slot] =
2231 ocfs2_get_recovery_generation(di);
2233 if (le32_to_cpu(di->id1.journal1.ij_flags) &
2234 OCFS2_JOURNAL_DIRTY_FL)