1 // SPDX-License-Identifier: GPL-2.0-or-later
5 * Defines functions of journalling api
7 * Copyright (C) 2003, 2004 Oracle. All rights reserved.
11 #include <linux/types.h>
12 #include <linux/slab.h>
13 #include <linux/highmem.h>
14 #include <linux/kthread.h>
15 #include <linux/time.h>
16 #include <linux/random.h>
17 #include <linux/delay.h>
18 #include <linux/writeback.h>
20 #include <cluster/masklog.h>
25 #include "blockcheck.h"
28 #include "extent_map.h"
29 #include "heartbeat.h"
32 #include "localalloc.h"
41 #include "buffer_head_io.h"
42 #include "ocfs2_trace.h"
44 DEFINE_SPINLOCK(trans_inc_lock);
46 #define ORPHAN_SCAN_SCHEDULE_TIMEOUT 300000
48 static int ocfs2_force_read_journal(struct inode *inode);
49 static int ocfs2_recover_node(struct ocfs2_super *osb,
50 int node_num, int slot_num);
51 static int __ocfs2_recovery_thread(void *arg);
52 static int ocfs2_commit_cache(struct ocfs2_super *osb);
53 static int __ocfs2_wait_on_mount(struct ocfs2_super *osb, int quota);
54 static int ocfs2_journal_toggle_dirty(struct ocfs2_super *osb,
55 int dirty, int replayed);
56 static int ocfs2_trylock_journal(struct ocfs2_super *osb,
58 static int ocfs2_recover_orphans(struct ocfs2_super *osb,
60 enum ocfs2_orphan_reco_type orphan_reco_type);
61 static int ocfs2_commit_thread(void *arg);
62 static void ocfs2_queue_recovery_completion(struct ocfs2_journal *journal,
64 struct ocfs2_dinode *la_dinode,
65 struct ocfs2_dinode *tl_dinode,
66 struct ocfs2_quota_recovery *qrec,
67 enum ocfs2_orphan_reco_type orphan_reco_type);
69 static inline int ocfs2_wait_on_mount(struct ocfs2_super *osb)
71 return __ocfs2_wait_on_mount(osb, 0);
74 static inline int ocfs2_wait_on_quotas(struct ocfs2_super *osb)
76 return __ocfs2_wait_on_mount(osb, 1);
80 * This replay_map is to track online/offline slots, so we could recover
81 * offline slots during recovery and mount
84 enum ocfs2_replay_state {
85 REPLAY_UNNEEDED = 0, /* Replay is not needed, so ignore this map */
86 REPLAY_NEEDED, /* Replay slots marked in rm_replay_slots */
87 REPLAY_DONE /* Replay was already queued */
90 struct ocfs2_replay_map {
91 unsigned int rm_slots;
92 enum ocfs2_replay_state rm_state;
93 unsigned char rm_replay_slots[];
96 static void ocfs2_replay_map_set_state(struct ocfs2_super *osb, int state)
101 /* If we've already queued the replay, we don't have any more to do */
102 if (osb->replay_map->rm_state == REPLAY_DONE)
105 osb->replay_map->rm_state = state;
108 int ocfs2_compute_replay_slots(struct ocfs2_super *osb)
110 struct ocfs2_replay_map *replay_map;
113 /* If replay map is already set, we don't do it again */
117 replay_map = kzalloc(struct_size(replay_map, rm_replay_slots,
125 spin_lock(&osb->osb_lock);
127 replay_map->rm_slots = osb->max_slots;
128 replay_map->rm_state = REPLAY_UNNEEDED;
130 /* set rm_replay_slots for offline slot(s) */
131 for (i = 0; i < replay_map->rm_slots; i++) {
132 if (ocfs2_slot_to_node_num_locked(osb, i, &node_num) == -ENOENT)
133 replay_map->rm_replay_slots[i] = 1;
136 osb->replay_map = replay_map;
137 spin_unlock(&osb->osb_lock);
141 static void ocfs2_queue_replay_slots(struct ocfs2_super *osb,
142 enum ocfs2_orphan_reco_type orphan_reco_type)
144 struct ocfs2_replay_map *replay_map = osb->replay_map;
150 if (replay_map->rm_state != REPLAY_NEEDED)
153 for (i = 0; i < replay_map->rm_slots; i++)
154 if (replay_map->rm_replay_slots[i])
155 ocfs2_queue_recovery_completion(osb->journal, i, NULL,
158 replay_map->rm_state = REPLAY_DONE;
161 void ocfs2_free_replay_slots(struct ocfs2_super *osb)
163 struct ocfs2_replay_map *replay_map = osb->replay_map;
165 if (!osb->replay_map)
169 osb->replay_map = NULL;
172 int ocfs2_recovery_init(struct ocfs2_super *osb)
174 struct ocfs2_recovery_map *rm;
176 mutex_init(&osb->recovery_lock);
177 osb->disable_recovery = 0;
178 osb->recovery_thread_task = NULL;
179 init_waitqueue_head(&osb->recovery_event);
181 rm = kzalloc(struct_size(rm, rm_entries, osb->max_slots),
188 osb->recovery_map = rm;
193 /* we can't grab the goofy sem lock from inside wait_event, so we use
194 * memory barriers to make sure that we'll see the null task before
196 static int ocfs2_recovery_thread_running(struct ocfs2_super *osb)
199 return osb->recovery_thread_task != NULL;
202 void ocfs2_recovery_exit(struct ocfs2_super *osb)
204 struct ocfs2_recovery_map *rm;
206 /* disable any new recovery threads and wait for any currently
207 * running ones to exit. Do this before setting the vol_state. */
208 mutex_lock(&osb->recovery_lock);
209 osb->disable_recovery = 1;
210 mutex_unlock(&osb->recovery_lock);
211 wait_event(osb->recovery_event, !ocfs2_recovery_thread_running(osb));
213 /* At this point, we know that no more recovery threads can be
214 * launched, so wait for any recovery completion work to
217 flush_workqueue(osb->ocfs2_wq);
220 * Now that recovery is shut down, and the osb is about to be
221 * freed, the osb_lock is not taken here.
223 rm = osb->recovery_map;
224 /* XXX: Should we bug if there are dirty entries? */
229 static int __ocfs2_recovery_map_test(struct ocfs2_super *osb,
230 unsigned int node_num)
233 struct ocfs2_recovery_map *rm = osb->recovery_map;
235 assert_spin_locked(&osb->osb_lock);
237 for (i = 0; i < rm->rm_used; i++) {
238 if (rm->rm_entries[i] == node_num)
245 /* Behaves like test-and-set. Returns the previous value */
246 static int ocfs2_recovery_map_set(struct ocfs2_super *osb,
247 unsigned int node_num)
249 struct ocfs2_recovery_map *rm = osb->recovery_map;
251 spin_lock(&osb->osb_lock);
252 if (__ocfs2_recovery_map_test(osb, node_num)) {
253 spin_unlock(&osb->osb_lock);
257 /* XXX: Can this be exploited? Not from o2dlm... */
258 BUG_ON(rm->rm_used >= osb->max_slots);
260 rm->rm_entries[rm->rm_used] = node_num;
262 spin_unlock(&osb->osb_lock);
267 static void ocfs2_recovery_map_clear(struct ocfs2_super *osb,
268 unsigned int node_num)
271 struct ocfs2_recovery_map *rm = osb->recovery_map;
273 spin_lock(&osb->osb_lock);
275 for (i = 0; i < rm->rm_used; i++) {
276 if (rm->rm_entries[i] == node_num)
280 if (i < rm->rm_used) {
281 /* XXX: be careful with the pointer math */
282 memmove(&(rm->rm_entries[i]), &(rm->rm_entries[i + 1]),
283 (rm->rm_used - i - 1) * sizeof(unsigned int));
287 spin_unlock(&osb->osb_lock);
290 static int ocfs2_commit_cache(struct ocfs2_super *osb)
293 unsigned int flushed;
294 struct ocfs2_journal *journal = NULL;
296 journal = osb->journal;
298 /* Flush all pending commits and checkpoint the journal. */
299 down_write(&journal->j_trans_barrier);
301 flushed = atomic_read(&journal->j_num_trans);
302 trace_ocfs2_commit_cache_begin(flushed);
304 up_write(&journal->j_trans_barrier);
308 jbd2_journal_lock_updates(journal->j_journal);
309 status = jbd2_journal_flush(journal->j_journal, 0);
310 jbd2_journal_unlock_updates(journal->j_journal);
312 up_write(&journal->j_trans_barrier);
317 ocfs2_inc_trans_id(journal);
319 flushed = atomic_read(&journal->j_num_trans);
320 atomic_set(&journal->j_num_trans, 0);
321 up_write(&journal->j_trans_barrier);
323 trace_ocfs2_commit_cache_end(journal->j_trans_id, flushed);
325 ocfs2_wake_downconvert_thread(osb);
326 wake_up(&journal->j_checkpointed);
331 handle_t *ocfs2_start_trans(struct ocfs2_super *osb, int max_buffs)
333 journal_t *journal = osb->journal->j_journal;
336 BUG_ON(!osb || !osb->journal->j_journal);
338 if (ocfs2_is_hard_readonly(osb))
339 return ERR_PTR(-EROFS);
341 BUG_ON(osb->journal->j_state == OCFS2_JOURNAL_FREE);
342 BUG_ON(max_buffs <= 0);
344 /* Nested transaction? Just return the handle... */
345 if (journal_current_handle())
346 return jbd2_journal_start(journal, max_buffs);
348 sb_start_intwrite(osb->sb);
350 down_read(&osb->journal->j_trans_barrier);
352 handle = jbd2_journal_start(journal, max_buffs);
353 if (IS_ERR(handle)) {
354 up_read(&osb->journal->j_trans_barrier);
355 sb_end_intwrite(osb->sb);
357 mlog_errno(PTR_ERR(handle));
359 if (is_journal_aborted(journal)) {
360 ocfs2_abort(osb->sb, "Detected aborted journal\n");
361 handle = ERR_PTR(-EROFS);
364 if (!ocfs2_mount_local(osb))
365 atomic_inc(&(osb->journal->j_num_trans));
371 int ocfs2_commit_trans(struct ocfs2_super *osb,
375 struct ocfs2_journal *journal = osb->journal;
379 nested = handle->h_ref > 1;
380 ret = jbd2_journal_stop(handle);
385 up_read(&journal->j_trans_barrier);
386 sb_end_intwrite(osb->sb);
393 * 'nblocks' is what you want to add to the current transaction.
395 * This might call jbd2_journal_restart() which will commit dirty buffers
396 * and then restart the transaction. Before calling
397 * ocfs2_extend_trans(), any changed blocks should have been
398 * dirtied. After calling it, all blocks which need to be changed must
399 * go through another set of journal_access/journal_dirty calls.
401 * WARNING: This will not release any semaphores or disk locks taken
402 * during the transaction, so make sure they were taken *before*
403 * start_trans or we'll have ordering deadlocks.
405 * WARNING2: Note that we do *not* drop j_trans_barrier here. This is
406 * good because transaction ids haven't yet been recorded on the
407 * cluster locks associated with this handle.
409 int ocfs2_extend_trans(handle_t *handle, int nblocks)
411 int status, old_nblocks;
419 old_nblocks = jbd2_handle_buffer_credits(handle);
421 trace_ocfs2_extend_trans(old_nblocks, nblocks);
423 #ifdef CONFIG_OCFS2_DEBUG_FS
426 status = jbd2_journal_extend(handle, nblocks, 0);
434 trace_ocfs2_extend_trans_restart(old_nblocks + nblocks);
435 status = jbd2_journal_restart(handle,
436 old_nblocks + nblocks);
449 * If we have fewer than thresh credits, extend by OCFS2_MAX_TRANS_DATA.
450 * If that fails, restart the transaction & regain write access for the
451 * buffer head which is used for metadata modifications.
452 * Taken from Ext4: extend_or_restart_transaction()
454 int ocfs2_allocate_extend_trans(handle_t *handle, int thresh)
456 int status, old_nblks;
460 old_nblks = jbd2_handle_buffer_credits(handle);
461 trace_ocfs2_allocate_extend_trans(old_nblks, thresh);
463 if (old_nblks < thresh)
466 status = jbd2_journal_extend(handle, OCFS2_MAX_TRANS_DATA, 0);
473 status = jbd2_journal_restart(handle, OCFS2_MAX_TRANS_DATA);
483 struct ocfs2_triggers {
484 struct jbd2_buffer_trigger_type ot_triggers;
488 static inline struct ocfs2_triggers *to_ocfs2_trigger(struct jbd2_buffer_trigger_type *triggers)
490 return container_of(triggers, struct ocfs2_triggers, ot_triggers);
493 static void ocfs2_frozen_trigger(struct jbd2_buffer_trigger_type *triggers,
494 struct buffer_head *bh,
495 void *data, size_t size)
497 struct ocfs2_triggers *ot = to_ocfs2_trigger(triggers);
500 * We aren't guaranteed to have the superblock here, so we
501 * must unconditionally compute the ecc data.
502 * __ocfs2_journal_access() will only set the triggers if
503 * metaecc is enabled.
505 ocfs2_block_check_compute(data, size, data + ot->ot_offset);
509 * Quota blocks have their own trigger because the struct ocfs2_block_check
510 * offset depends on the blocksize.
512 static void ocfs2_dq_frozen_trigger(struct jbd2_buffer_trigger_type *triggers,
513 struct buffer_head *bh,
514 void *data, size_t size)
516 struct ocfs2_disk_dqtrailer *dqt =
517 ocfs2_block_dqtrailer(size, data);
520 * We aren't guaranteed to have the superblock here, so we
521 * must unconditionally compute the ecc data.
522 * __ocfs2_journal_access() will only set the triggers if
523 * metaecc is enabled.
525 ocfs2_block_check_compute(data, size, &dqt->dq_check);
529 * Directory blocks also have their own trigger because the
530 * struct ocfs2_block_check offset depends on the blocksize.
532 static void ocfs2_db_frozen_trigger(struct jbd2_buffer_trigger_type *triggers,
533 struct buffer_head *bh,
534 void *data, size_t size)
536 struct ocfs2_dir_block_trailer *trailer =
537 ocfs2_dir_trailer_from_size(size, data);
540 * We aren't guaranteed to have the superblock here, so we
541 * must unconditionally compute the ecc data.
542 * __ocfs2_journal_access() will only set the triggers if
543 * metaecc is enabled.
545 ocfs2_block_check_compute(data, size, &trailer->db_check);
548 static void ocfs2_abort_trigger(struct jbd2_buffer_trigger_type *triggers,
549 struct buffer_head *bh)
552 "ocfs2_abort_trigger called by JBD2. bh = 0x%lx, "
553 "bh->b_blocknr = %llu\n",
555 (unsigned long long)bh->b_blocknr);
557 ocfs2_error(bh->b_assoc_map->host->i_sb,
558 "JBD2 has aborted our journal, ocfs2 cannot continue\n");
561 static struct ocfs2_triggers di_triggers = {
563 .t_frozen = ocfs2_frozen_trigger,
564 .t_abort = ocfs2_abort_trigger,
566 .ot_offset = offsetof(struct ocfs2_dinode, i_check),
569 static struct ocfs2_triggers eb_triggers = {
571 .t_frozen = ocfs2_frozen_trigger,
572 .t_abort = ocfs2_abort_trigger,
574 .ot_offset = offsetof(struct ocfs2_extent_block, h_check),
577 static struct ocfs2_triggers rb_triggers = {
579 .t_frozen = ocfs2_frozen_trigger,
580 .t_abort = ocfs2_abort_trigger,
582 .ot_offset = offsetof(struct ocfs2_refcount_block, rf_check),
585 static struct ocfs2_triggers gd_triggers = {
587 .t_frozen = ocfs2_frozen_trigger,
588 .t_abort = ocfs2_abort_trigger,
590 .ot_offset = offsetof(struct ocfs2_group_desc, bg_check),
593 static struct ocfs2_triggers db_triggers = {
595 .t_frozen = ocfs2_db_frozen_trigger,
596 .t_abort = ocfs2_abort_trigger,
600 static struct ocfs2_triggers xb_triggers = {
602 .t_frozen = ocfs2_frozen_trigger,
603 .t_abort = ocfs2_abort_trigger,
605 .ot_offset = offsetof(struct ocfs2_xattr_block, xb_check),
608 static struct ocfs2_triggers dq_triggers = {
610 .t_frozen = ocfs2_dq_frozen_trigger,
611 .t_abort = ocfs2_abort_trigger,
615 static struct ocfs2_triggers dr_triggers = {
617 .t_frozen = ocfs2_frozen_trigger,
618 .t_abort = ocfs2_abort_trigger,
620 .ot_offset = offsetof(struct ocfs2_dx_root_block, dr_check),
623 static struct ocfs2_triggers dl_triggers = {
625 .t_frozen = ocfs2_frozen_trigger,
626 .t_abort = ocfs2_abort_trigger,
628 .ot_offset = offsetof(struct ocfs2_dx_leaf, dl_check),
631 static int __ocfs2_journal_access(handle_t *handle,
632 struct ocfs2_caching_info *ci,
633 struct buffer_head *bh,
634 struct ocfs2_triggers *triggers,
638 struct ocfs2_super *osb =
639 OCFS2_SB(ocfs2_metadata_cache_get_super(ci));
641 BUG_ON(!ci || !ci->ci_ops);
645 trace_ocfs2_journal_access(
646 (unsigned long long)ocfs2_metadata_cache_owner(ci),
647 (unsigned long long)bh->b_blocknr, type, bh->b_size);
649 /* we can safely remove this assertion after testing. */
650 if (!buffer_uptodate(bh)) {
651 mlog(ML_ERROR, "giving me a buffer that's not uptodate!\n");
652 mlog(ML_ERROR, "b_blocknr=%llu, b_state=0x%lx\n",
653 (unsigned long long)bh->b_blocknr, bh->b_state);
657 * A previous transaction with a couple of buffer heads fail
658 * to checkpoint, so all the bhs are marked as BH_Write_EIO.
659 * For current transaction, the bh is just among those error
660 * bhs which previous transaction handle. We can't just clear
661 * its BH_Write_EIO and reuse directly, since other bhs are
662 * not written to disk yet and that will cause metadata
663 * inconsistency. So we should set fs read-only to avoid
666 if (buffer_write_io_error(bh) && !buffer_uptodate(bh)) {
668 return ocfs2_error(osb->sb, "A previous attempt to "
669 "write this buffer head failed\n");
674 /* Set the current transaction information on the ci so
675 * that the locking code knows whether it can drop it's locks
676 * on this ci or not. We're protected from the commit
677 * thread updating the current transaction id until
678 * ocfs2_commit_trans() because ocfs2_start_trans() took
679 * j_trans_barrier for us. */
680 ocfs2_set_ci_lock_trans(osb->journal, ci);
682 ocfs2_metadata_cache_io_lock(ci);
684 case OCFS2_JOURNAL_ACCESS_CREATE:
685 case OCFS2_JOURNAL_ACCESS_WRITE:
686 status = jbd2_journal_get_write_access(handle, bh);
689 case OCFS2_JOURNAL_ACCESS_UNDO:
690 status = jbd2_journal_get_undo_access(handle, bh);
695 mlog(ML_ERROR, "Unknown access type!\n");
697 if (!status && ocfs2_meta_ecc(osb) && triggers)
698 jbd2_journal_set_triggers(bh, &triggers->ot_triggers);
699 ocfs2_metadata_cache_io_unlock(ci);
702 mlog(ML_ERROR, "Error %d getting %d access to buffer!\n",
708 int ocfs2_journal_access_di(handle_t *handle, struct ocfs2_caching_info *ci,
709 struct buffer_head *bh, int type)
711 return __ocfs2_journal_access(handle, ci, bh, &di_triggers, type);
714 int ocfs2_journal_access_eb(handle_t *handle, struct ocfs2_caching_info *ci,
715 struct buffer_head *bh, int type)
717 return __ocfs2_journal_access(handle, ci, bh, &eb_triggers, type);
720 int ocfs2_journal_access_rb(handle_t *handle, struct ocfs2_caching_info *ci,
721 struct buffer_head *bh, int type)
723 return __ocfs2_journal_access(handle, ci, bh, &rb_triggers,
727 int ocfs2_journal_access_gd(handle_t *handle, struct ocfs2_caching_info *ci,
728 struct buffer_head *bh, int type)
730 return __ocfs2_journal_access(handle, ci, bh, &gd_triggers, type);
733 int ocfs2_journal_access_db(handle_t *handle, struct ocfs2_caching_info *ci,
734 struct buffer_head *bh, int type)
736 return __ocfs2_journal_access(handle, ci, bh, &db_triggers, type);
739 int ocfs2_journal_access_xb(handle_t *handle, struct ocfs2_caching_info *ci,
740 struct buffer_head *bh, int type)
742 return __ocfs2_journal_access(handle, ci, bh, &xb_triggers, type);
745 int ocfs2_journal_access_dq(handle_t *handle, struct ocfs2_caching_info *ci,
746 struct buffer_head *bh, int type)
748 return __ocfs2_journal_access(handle, ci, bh, &dq_triggers, type);
751 int ocfs2_journal_access_dr(handle_t *handle, struct ocfs2_caching_info *ci,
752 struct buffer_head *bh, int type)
754 return __ocfs2_journal_access(handle, ci, bh, &dr_triggers, type);
757 int ocfs2_journal_access_dl(handle_t *handle, struct ocfs2_caching_info *ci,
758 struct buffer_head *bh, int type)
760 return __ocfs2_journal_access(handle, ci, bh, &dl_triggers, type);
763 int ocfs2_journal_access(handle_t *handle, struct ocfs2_caching_info *ci,
764 struct buffer_head *bh, int type)
766 return __ocfs2_journal_access(handle, ci, bh, NULL, type);
769 void ocfs2_journal_dirty(handle_t *handle, struct buffer_head *bh)
773 trace_ocfs2_journal_dirty((unsigned long long)bh->b_blocknr);
775 status = jbd2_journal_dirty_metadata(handle, bh);
778 if (!is_handle_aborted(handle)) {
779 journal_t *journal = handle->h_transaction->t_journal;
781 mlog(ML_ERROR, "jbd2_journal_dirty_metadata failed. "
782 "Aborting transaction and journal.\n");
783 handle->h_err = status;
784 jbd2_journal_abort_handle(handle);
785 jbd2_journal_abort(journal, status);
786 ocfs2_abort(bh->b_assoc_map->host->i_sb,
787 "Journal already aborted.\n");
792 #define OCFS2_DEFAULT_COMMIT_INTERVAL (HZ * JBD2_DEFAULT_MAX_COMMIT_AGE)
794 void ocfs2_set_journal_params(struct ocfs2_super *osb)
796 journal_t *journal = osb->journal->j_journal;
797 unsigned long commit_interval = OCFS2_DEFAULT_COMMIT_INTERVAL;
799 if (osb->osb_commit_interval)
800 commit_interval = osb->osb_commit_interval;
802 write_lock(&journal->j_state_lock);
803 journal->j_commit_interval = commit_interval;
804 if (osb->s_mount_opt & OCFS2_MOUNT_BARRIER)
805 journal->j_flags |= JBD2_BARRIER;
807 journal->j_flags &= ~JBD2_BARRIER;
808 write_unlock(&journal->j_state_lock);
812 * alloc & initialize skeleton for journal structure.
813 * ocfs2_journal_init() will make fs have journal ability.
815 int ocfs2_journal_alloc(struct ocfs2_super *osb)
818 struct ocfs2_journal *journal;
820 journal = kzalloc(sizeof(struct ocfs2_journal), GFP_KERNEL);
822 mlog(ML_ERROR, "unable to alloc journal\n");
826 osb->journal = journal;
827 journal->j_osb = osb;
829 atomic_set(&journal->j_num_trans, 0);
830 init_rwsem(&journal->j_trans_barrier);
831 init_waitqueue_head(&journal->j_checkpointed);
832 spin_lock_init(&journal->j_lock);
833 journal->j_trans_id = 1UL;
834 INIT_LIST_HEAD(&journal->j_la_cleanups);
835 INIT_WORK(&journal->j_recovery_work, ocfs2_complete_recovery);
836 journal->j_state = OCFS2_JOURNAL_FREE;
842 static int ocfs2_journal_submit_inode_data_buffers(struct jbd2_inode *jinode)
844 struct address_space *mapping = jinode->i_vfs_inode->i_mapping;
845 struct writeback_control wbc = {
846 .sync_mode = WB_SYNC_ALL,
847 .nr_to_write = mapping->nrpages * 2,
848 .range_start = jinode->i_dirty_start,
849 .range_end = jinode->i_dirty_end,
852 return filemap_fdatawrite_wbc(mapping, &wbc);
855 int ocfs2_journal_init(struct ocfs2_super *osb, int *dirty)
858 struct inode *inode = NULL; /* the journal inode */
859 journal_t *j_journal = NULL;
860 struct ocfs2_journal *journal = osb->journal;
861 struct ocfs2_dinode *di = NULL;
862 struct buffer_head *bh = NULL;
866 /* already have the inode for our journal */
867 inode = ocfs2_get_system_file_inode(osb, JOURNAL_SYSTEM_INODE,
874 if (is_bad_inode(inode)) {
875 mlog(ML_ERROR, "access error (bad inode)\n");
882 SET_INODE_JOURNAL(inode);
883 OCFS2_I(inode)->ip_open_count++;
885 /* Skip recovery waits here - journal inode metadata never
886 * changes in a live cluster so it can be considered an
887 * exception to the rule. */
888 status = ocfs2_inode_lock_full(inode, &bh, 1, OCFS2_META_LOCK_RECOVERY);
890 if (status != -ERESTARTSYS)
891 mlog(ML_ERROR, "Could not get lock on journal!\n");
896 di = (struct ocfs2_dinode *)bh->b_data;
898 if (i_size_read(inode) < OCFS2_MIN_JOURNAL_SIZE) {
899 mlog(ML_ERROR, "Journal file size (%lld) is too small!\n",
905 trace_ocfs2_journal_init(i_size_read(inode),
906 (unsigned long long)inode->i_blocks,
907 OCFS2_I(inode)->ip_clusters);
909 /* call the kernels journal init function now */
910 j_journal = jbd2_journal_init_inode(inode);
911 if (IS_ERR(j_journal)) {
912 mlog(ML_ERROR, "Linux journal layer error\n");
913 status = PTR_ERR(j_journal);
917 trace_ocfs2_journal_init_maxlen(j_journal->j_total_len);
919 *dirty = (le32_to_cpu(di->id1.journal1.ij_flags) &
920 OCFS2_JOURNAL_DIRTY_FL);
922 journal->j_journal = j_journal;
923 journal->j_journal->j_submit_inode_data_buffers =
924 ocfs2_journal_submit_inode_data_buffers;
925 journal->j_journal->j_finish_inode_data_buffers =
926 jbd2_journal_finish_inode_data_buffers;
927 journal->j_inode = inode;
930 ocfs2_set_journal_params(osb);
932 journal->j_state = OCFS2_JOURNAL_LOADED;
938 ocfs2_inode_unlock(inode, 1);
941 OCFS2_I(inode)->ip_open_count--;
949 static void ocfs2_bump_recovery_generation(struct ocfs2_dinode *di)
951 le32_add_cpu(&(di->id1.journal1.ij_recovery_generation), 1);
954 static u32 ocfs2_get_recovery_generation(struct ocfs2_dinode *di)
956 return le32_to_cpu(di->id1.journal1.ij_recovery_generation);
959 static int ocfs2_journal_toggle_dirty(struct ocfs2_super *osb,
960 int dirty, int replayed)
964 struct ocfs2_journal *journal = osb->journal;
965 struct buffer_head *bh = journal->j_bh;
966 struct ocfs2_dinode *fe;
968 fe = (struct ocfs2_dinode *)bh->b_data;
970 /* The journal bh on the osb always comes from ocfs2_journal_init()
971 * and was validated there inside ocfs2_inode_lock_full(). It's a
972 * code bug if we mess it up. */
973 BUG_ON(!OCFS2_IS_VALID_DINODE(fe));
975 flags = le32_to_cpu(fe->id1.journal1.ij_flags);
977 flags |= OCFS2_JOURNAL_DIRTY_FL;
979 flags &= ~OCFS2_JOURNAL_DIRTY_FL;
980 fe->id1.journal1.ij_flags = cpu_to_le32(flags);
983 ocfs2_bump_recovery_generation(fe);
985 ocfs2_compute_meta_ecc(osb->sb, bh->b_data, &fe->i_check);
986 status = ocfs2_write_block(osb, bh, INODE_CACHE(journal->j_inode));
994 * If the journal has been kmalloc'd it needs to be freed after this
997 void ocfs2_journal_shutdown(struct ocfs2_super *osb)
999 struct ocfs2_journal *journal = NULL;
1001 struct inode *inode = NULL;
1002 int num_running_trans = 0;
1006 journal = osb->journal;
1010 inode = journal->j_inode;
1012 if (journal->j_state != OCFS2_JOURNAL_LOADED)
1015 /* need to inc inode use count - jbd2_journal_destroy will iput. */
1019 num_running_trans = atomic_read(&(osb->journal->j_num_trans));
1020 trace_ocfs2_journal_shutdown(num_running_trans);
1022 /* Do a commit_cache here. It will flush our journal, *and*
1023 * release any locks that are still held.
1024 * set the SHUTDOWN flag and release the trans lock.
1025 * the commit thread will take the trans lock for us below. */
1026 journal->j_state = OCFS2_JOURNAL_IN_SHUTDOWN;
1028 /* The OCFS2_JOURNAL_IN_SHUTDOWN will signal to commit_cache to not
1029 * drop the trans_lock (which we want to hold until we
1030 * completely destroy the journal. */
1031 if (osb->commit_task) {
1032 /* Wait for the commit thread */
1033 trace_ocfs2_journal_shutdown_wait(osb->commit_task);
1034 kthread_stop(osb->commit_task);
1035 osb->commit_task = NULL;
1038 BUG_ON(atomic_read(&(osb->journal->j_num_trans)) != 0);
1040 if (ocfs2_mount_local(osb)) {
1041 jbd2_journal_lock_updates(journal->j_journal);
1042 status = jbd2_journal_flush(journal->j_journal, 0);
1043 jbd2_journal_unlock_updates(journal->j_journal);
1048 /* Shutdown the kernel journal system */
1049 if (!jbd2_journal_destroy(journal->j_journal) && !status) {
1051 * Do not toggle if flush was unsuccessful otherwise
1052 * will leave dirty metadata in a "clean" journal
1054 status = ocfs2_journal_toggle_dirty(osb, 0, 0);
1058 journal->j_journal = NULL;
1060 OCFS2_I(inode)->ip_open_count--;
1062 /* unlock our journal */
1063 ocfs2_inode_unlock(inode, 1);
1065 brelse(journal->j_bh);
1066 journal->j_bh = NULL;
1068 journal->j_state = OCFS2_JOURNAL_FREE;
1073 osb->journal = NULL;
1076 static void ocfs2_clear_journal_error(struct super_block *sb,
1082 olderr = jbd2_journal_errno(journal);
1084 mlog(ML_ERROR, "File system error %d recorded in "
1085 "journal %u.\n", olderr, slot);
1086 mlog(ML_ERROR, "File system on device %s needs checking.\n",
1089 jbd2_journal_ack_err(journal);
1090 jbd2_journal_clear_err(journal);
1094 int ocfs2_journal_load(struct ocfs2_journal *journal, int local, int replayed)
1097 struct ocfs2_super *osb;
1101 osb = journal->j_osb;
1103 status = jbd2_journal_load(journal->j_journal);
1105 mlog(ML_ERROR, "Failed to load journal!\n");
1109 ocfs2_clear_journal_error(osb->sb, journal->j_journal, osb->slot_num);
1112 jbd2_journal_lock_updates(journal->j_journal);
1113 status = jbd2_journal_flush(journal->j_journal, 0);
1114 jbd2_journal_unlock_updates(journal->j_journal);
1119 status = ocfs2_journal_toggle_dirty(osb, 1, replayed);
1125 /* Launch the commit thread */
1127 osb->commit_task = kthread_run(ocfs2_commit_thread, osb,
1128 "ocfs2cmt-%s", osb->uuid_str);
1129 if (IS_ERR(osb->commit_task)) {
1130 status = PTR_ERR(osb->commit_task);
1131 osb->commit_task = NULL;
1132 mlog(ML_ERROR, "unable to launch ocfs2commit thread, "
1133 "error=%d", status);
1137 osb->commit_task = NULL;
1144 /* 'full' flag tells us whether we clear out all blocks or if we just
1145 * mark the journal clean */
1146 int ocfs2_journal_wipe(struct ocfs2_journal *journal, int full)
1152 status = jbd2_journal_wipe(journal->j_journal, full);
1158 status = ocfs2_journal_toggle_dirty(journal->j_osb, 0, 0);
1166 static int ocfs2_recovery_completed(struct ocfs2_super *osb)
1169 struct ocfs2_recovery_map *rm = osb->recovery_map;
1171 spin_lock(&osb->osb_lock);
1172 empty = (rm->rm_used == 0);
1173 spin_unlock(&osb->osb_lock);
1178 void ocfs2_wait_for_recovery(struct ocfs2_super *osb)
1180 wait_event(osb->recovery_event, ocfs2_recovery_completed(osb));
1184 * JBD Might read a cached version of another nodes journal file. We
1185 * don't want this as this file changes often and we get no
1186 * notification on those changes. The only way to be sure that we've
1187 * got the most up to date version of those blocks then is to force
1188 * read them off disk. Just searching through the buffer cache won't
1189 * work as there may be pages backing this file which are still marked
1190 * up to date. We know things can't change on this file underneath us
1191 * as we have the lock by now :)
1193 static int ocfs2_force_read_journal(struct inode *inode)
1197 u64 v_blkno, p_blkno, p_blocks, num_blocks;
1198 struct buffer_head *bh = NULL;
1199 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
1201 num_blocks = ocfs2_blocks_for_bytes(inode->i_sb, i_size_read(inode));
1203 while (v_blkno < num_blocks) {
1204 status = ocfs2_extent_map_get_blocks(inode, v_blkno,
1205 &p_blkno, &p_blocks, NULL);
1211 for (i = 0; i < p_blocks; i++, p_blkno++) {
1212 bh = __find_get_block(osb->sb->s_bdev, p_blkno,
1213 osb->sb->s_blocksize);
1214 /* block not cached. */
1220 /* We are reading journal data which should not
1221 * be put in the uptodate cache.
1223 status = ocfs2_read_blocks_sync(osb, p_blkno, 1, &bh);
1233 v_blkno += p_blocks;
1240 struct ocfs2_la_recovery_item {
1241 struct list_head lri_list;
1243 struct ocfs2_dinode *lri_la_dinode;
1244 struct ocfs2_dinode *lri_tl_dinode;
1245 struct ocfs2_quota_recovery *lri_qrec;
1246 enum ocfs2_orphan_reco_type lri_orphan_reco_type;
1249 /* Does the second half of the recovery process. By this point, the
1250 * node is marked clean and can actually be considered recovered,
1251 * hence it's no longer in the recovery map, but there's still some
1252 * cleanup we can do which shouldn't happen within the recovery thread
1253 * as locking in that context becomes very difficult if we are to take
1254 * recovering nodes into account.
1256 * NOTE: This function can and will sleep on recovery of other nodes
1257 * during cluster locking, just like any other ocfs2 process.
1259 void ocfs2_complete_recovery(struct work_struct *work)
1262 struct ocfs2_journal *journal =
1263 container_of(work, struct ocfs2_journal, j_recovery_work);
1264 struct ocfs2_super *osb = journal->j_osb;
1265 struct ocfs2_dinode *la_dinode, *tl_dinode;
1266 struct ocfs2_la_recovery_item *item, *n;
1267 struct ocfs2_quota_recovery *qrec;
1268 enum ocfs2_orphan_reco_type orphan_reco_type;
1269 LIST_HEAD(tmp_la_list);
1271 trace_ocfs2_complete_recovery(
1272 (unsigned long long)OCFS2_I(journal->j_inode)->ip_blkno);
1274 spin_lock(&journal->j_lock);
1275 list_splice_init(&journal->j_la_cleanups, &tmp_la_list);
1276 spin_unlock(&journal->j_lock);
1278 list_for_each_entry_safe(item, n, &tmp_la_list, lri_list) {
1279 list_del_init(&item->lri_list);
1281 ocfs2_wait_on_quotas(osb);
1283 la_dinode = item->lri_la_dinode;
1284 tl_dinode = item->lri_tl_dinode;
1285 qrec = item->lri_qrec;
1286 orphan_reco_type = item->lri_orphan_reco_type;
1288 trace_ocfs2_complete_recovery_slot(item->lri_slot,
1289 la_dinode ? le64_to_cpu(la_dinode->i_blkno) : 0,
1290 tl_dinode ? le64_to_cpu(tl_dinode->i_blkno) : 0,
1294 ret = ocfs2_complete_local_alloc_recovery(osb,
1303 ret = ocfs2_complete_truncate_log_recovery(osb,
1311 ret = ocfs2_recover_orphans(osb, item->lri_slot,
1317 ret = ocfs2_finish_quota_recovery(osb, qrec,
1321 /* Recovery info is already freed now */
1327 trace_ocfs2_complete_recovery_end(ret);
1330 /* NOTE: This function always eats your references to la_dinode and
1331 * tl_dinode, either manually on error, or by passing them to
1332 * ocfs2_complete_recovery */
1333 static void ocfs2_queue_recovery_completion(struct ocfs2_journal *journal,
1335 struct ocfs2_dinode *la_dinode,
1336 struct ocfs2_dinode *tl_dinode,
1337 struct ocfs2_quota_recovery *qrec,
1338 enum ocfs2_orphan_reco_type orphan_reco_type)
1340 struct ocfs2_la_recovery_item *item;
1342 item = kmalloc(sizeof(struct ocfs2_la_recovery_item), GFP_NOFS);
1344 /* Though we wish to avoid it, we are in fact safe in
1345 * skipping local alloc cleanup as fsck.ocfs2 is more
1346 * than capable of reclaiming unused space. */
1351 ocfs2_free_quota_recovery(qrec);
1353 mlog_errno(-ENOMEM);
1357 INIT_LIST_HEAD(&item->lri_list);
1358 item->lri_la_dinode = la_dinode;
1359 item->lri_slot = slot_num;
1360 item->lri_tl_dinode = tl_dinode;
1361 item->lri_qrec = qrec;
1362 item->lri_orphan_reco_type = orphan_reco_type;
1364 spin_lock(&journal->j_lock);
1365 list_add_tail(&item->lri_list, &journal->j_la_cleanups);
1366 queue_work(journal->j_osb->ocfs2_wq, &journal->j_recovery_work);
1367 spin_unlock(&journal->j_lock);
1370 /* Called by the mount code to queue recovery the last part of
1371 * recovery for it's own and offline slot(s). */
1372 void ocfs2_complete_mount_recovery(struct ocfs2_super *osb)
1374 struct ocfs2_journal *journal = osb->journal;
1376 if (ocfs2_is_hard_readonly(osb))
1379 /* No need to queue up our truncate_log as regular cleanup will catch
1381 ocfs2_queue_recovery_completion(journal, osb->slot_num,
1382 osb->local_alloc_copy, NULL, NULL,
1383 ORPHAN_NEED_TRUNCATE);
1384 ocfs2_schedule_truncate_log_flush(osb, 0);
1386 osb->local_alloc_copy = NULL;
1388 /* queue to recover orphan slots for all offline slots */
1389 ocfs2_replay_map_set_state(osb, REPLAY_NEEDED);
1390 ocfs2_queue_replay_slots(osb, ORPHAN_NEED_TRUNCATE);
1391 ocfs2_free_replay_slots(osb);
1394 void ocfs2_complete_quota_recovery(struct ocfs2_super *osb)
1396 if (osb->quota_rec) {
1397 ocfs2_queue_recovery_completion(osb->journal,
1402 ORPHAN_NEED_TRUNCATE);
1403 osb->quota_rec = NULL;
1407 static int __ocfs2_recovery_thread(void *arg)
1409 int status, node_num, slot_num;
1410 struct ocfs2_super *osb = arg;
1411 struct ocfs2_recovery_map *rm = osb->recovery_map;
1412 int *rm_quota = NULL;
1413 int rm_quota_used = 0, i;
1414 struct ocfs2_quota_recovery *qrec;
1416 /* Whether the quota supported. */
1417 int quota_enabled = OCFS2_HAS_RO_COMPAT_FEATURE(osb->sb,
1418 OCFS2_FEATURE_RO_COMPAT_USRQUOTA)
1419 || OCFS2_HAS_RO_COMPAT_FEATURE(osb->sb,
1420 OCFS2_FEATURE_RO_COMPAT_GRPQUOTA);
1422 status = ocfs2_wait_on_mount(osb);
1427 if (quota_enabled) {
1428 rm_quota = kcalloc(osb->max_slots, sizeof(int), GFP_NOFS);
1435 status = ocfs2_super_lock(osb, 1);
1441 status = ocfs2_compute_replay_slots(osb);
1445 /* queue recovery for our own slot */
1446 ocfs2_queue_recovery_completion(osb->journal, osb->slot_num, NULL,
1447 NULL, NULL, ORPHAN_NO_NEED_TRUNCATE);
1449 spin_lock(&osb->osb_lock);
1450 while (rm->rm_used) {
1451 /* It's always safe to remove entry zero, as we won't
1452 * clear it until ocfs2_recover_node() has succeeded. */
1453 node_num = rm->rm_entries[0];
1454 spin_unlock(&osb->osb_lock);
1455 slot_num = ocfs2_node_num_to_slot(osb, node_num);
1456 trace_ocfs2_recovery_thread_node(node_num, slot_num);
1457 if (slot_num == -ENOENT) {
1462 /* It is a bit subtle with quota recovery. We cannot do it
1463 * immediately because we have to obtain cluster locks from
1464 * quota files and we also don't want to just skip it because
1465 * then quota usage would be out of sync until some node takes
1466 * the slot. So we remember which nodes need quota recovery
1467 * and when everything else is done, we recover quotas. */
1468 if (quota_enabled) {
1469 for (i = 0; i < rm_quota_used
1470 && rm_quota[i] != slot_num; i++)
1473 if (i == rm_quota_used)
1474 rm_quota[rm_quota_used++] = slot_num;
1477 status = ocfs2_recover_node(osb, node_num, slot_num);
1480 ocfs2_recovery_map_clear(osb, node_num);
1483 "Error %d recovering node %d on device (%u,%u)!\n",
1485 MAJOR(osb->sb->s_dev), MINOR(osb->sb->s_dev));
1486 mlog(ML_ERROR, "Volume requires unmount.\n");
1489 spin_lock(&osb->osb_lock);
1491 spin_unlock(&osb->osb_lock);
1492 trace_ocfs2_recovery_thread_end(status);
1494 /* Refresh all journal recovery generations from disk */
1495 status = ocfs2_check_journals_nolocks(osb);
1496 status = (status == -EROFS) ? 0 : status;
1500 /* Now it is right time to recover quotas... We have to do this under
1501 * superblock lock so that no one can start using the slot (and crash)
1502 * before we recover it */
1503 if (quota_enabled) {
1504 for (i = 0; i < rm_quota_used; i++) {
1505 qrec = ocfs2_begin_quota_recovery(osb, rm_quota[i]);
1507 status = PTR_ERR(qrec);
1511 ocfs2_queue_recovery_completion(osb->journal,
1514 ORPHAN_NEED_TRUNCATE);
1518 ocfs2_super_unlock(osb, 1);
1520 /* queue recovery for offline slots */
1521 ocfs2_queue_replay_slots(osb, ORPHAN_NEED_TRUNCATE);
1524 mutex_lock(&osb->recovery_lock);
1525 if (!status && !ocfs2_recovery_completed(osb)) {
1526 mutex_unlock(&osb->recovery_lock);
1530 ocfs2_free_replay_slots(osb);
1531 osb->recovery_thread_task = NULL;
1532 mb(); /* sync with ocfs2_recovery_thread_running */
1533 wake_up(&osb->recovery_event);
1535 mutex_unlock(&osb->recovery_lock);
1543 void ocfs2_recovery_thread(struct ocfs2_super *osb, int node_num)
1545 mutex_lock(&osb->recovery_lock);
1547 trace_ocfs2_recovery_thread(node_num, osb->node_num,
1548 osb->disable_recovery, osb->recovery_thread_task,
1549 osb->disable_recovery ?
1550 -1 : ocfs2_recovery_map_set(osb, node_num));
1552 if (osb->disable_recovery)
1555 if (osb->recovery_thread_task)
1558 osb->recovery_thread_task = kthread_run(__ocfs2_recovery_thread, osb,
1559 "ocfs2rec-%s", osb->uuid_str);
1560 if (IS_ERR(osb->recovery_thread_task)) {
1561 mlog_errno((int)PTR_ERR(osb->recovery_thread_task));
1562 osb->recovery_thread_task = NULL;
1566 mutex_unlock(&osb->recovery_lock);
1567 wake_up(&osb->recovery_event);
1570 static int ocfs2_read_journal_inode(struct ocfs2_super *osb,
1572 struct buffer_head **bh,
1573 struct inode **ret_inode)
1575 int status = -EACCES;
1576 struct inode *inode = NULL;
1578 BUG_ON(slot_num >= osb->max_slots);
1580 inode = ocfs2_get_system_file_inode(osb, JOURNAL_SYSTEM_INODE,
1582 if (!inode || is_bad_inode(inode)) {
1586 SET_INODE_JOURNAL(inode);
1588 status = ocfs2_read_inode_block_full(inode, bh, OCFS2_BH_IGNORE_CACHE);
1598 if (status || !ret_inode)
1606 /* Does the actual journal replay and marks the journal inode as
1607 * clean. Will only replay if the journal inode is marked dirty. */
1608 static int ocfs2_replay_journal(struct ocfs2_super *osb,
1615 struct inode *inode = NULL;
1616 struct ocfs2_dinode *fe;
1617 journal_t *journal = NULL;
1618 struct buffer_head *bh = NULL;
1621 status = ocfs2_read_journal_inode(osb, slot_num, &bh, &inode);
1627 fe = (struct ocfs2_dinode *)bh->b_data;
1628 slot_reco_gen = ocfs2_get_recovery_generation(fe);
1633 * As the fs recovery is asynchronous, there is a small chance that
1634 * another node mounted (and recovered) the slot before the recovery
1635 * thread could get the lock. To handle that, we dirty read the journal
1636 * inode for that slot to get the recovery generation. If it is
1637 * different than what we expected, the slot has been recovered.
1638 * If not, it needs recovery.
1640 if (osb->slot_recovery_generations[slot_num] != slot_reco_gen) {
1641 trace_ocfs2_replay_journal_recovered(slot_num,
1642 osb->slot_recovery_generations[slot_num], slot_reco_gen);
1643 osb->slot_recovery_generations[slot_num] = slot_reco_gen;
1648 /* Continue with recovery as the journal has not yet been recovered */
1650 status = ocfs2_inode_lock_full(inode, &bh, 1, OCFS2_META_LOCK_RECOVERY);
1652 trace_ocfs2_replay_journal_lock_err(status);
1653 if (status != -ERESTARTSYS)
1654 mlog(ML_ERROR, "Could not lock journal!\n");
1659 fe = (struct ocfs2_dinode *) bh->b_data;
1661 flags = le32_to_cpu(fe->id1.journal1.ij_flags);
1662 slot_reco_gen = ocfs2_get_recovery_generation(fe);
1664 if (!(flags & OCFS2_JOURNAL_DIRTY_FL)) {
1665 trace_ocfs2_replay_journal_skip(node_num);
1666 /* Refresh recovery generation for the slot */
1667 osb->slot_recovery_generations[slot_num] = slot_reco_gen;
1671 /* we need to run complete recovery for offline orphan slots */
1672 ocfs2_replay_map_set_state(osb, REPLAY_NEEDED);
1674 printk(KERN_NOTICE "ocfs2: Begin replay journal (node %d, slot %d) on "\
1675 "device (%u,%u)\n", node_num, slot_num, MAJOR(osb->sb->s_dev),
1676 MINOR(osb->sb->s_dev));
1678 OCFS2_I(inode)->ip_clusters = le32_to_cpu(fe->i_clusters);
1680 status = ocfs2_force_read_journal(inode);
1686 journal = jbd2_journal_init_inode(inode);
1687 if (IS_ERR(journal)) {
1688 mlog(ML_ERROR, "Linux journal layer error\n");
1689 status = PTR_ERR(journal);
1693 status = jbd2_journal_load(journal);
1696 BUG_ON(!igrab(inode));
1697 jbd2_journal_destroy(journal);
1701 ocfs2_clear_journal_error(osb->sb, journal, slot_num);
1703 /* wipe the journal */
1704 jbd2_journal_lock_updates(journal);
1705 status = jbd2_journal_flush(journal, 0);
1706 jbd2_journal_unlock_updates(journal);
1710 /* This will mark the node clean */
1711 flags = le32_to_cpu(fe->id1.journal1.ij_flags);
1712 flags &= ~OCFS2_JOURNAL_DIRTY_FL;
1713 fe->id1.journal1.ij_flags = cpu_to_le32(flags);
1715 /* Increment recovery generation to indicate successful recovery */
1716 ocfs2_bump_recovery_generation(fe);
1717 osb->slot_recovery_generations[slot_num] =
1718 ocfs2_get_recovery_generation(fe);
1720 ocfs2_compute_meta_ecc(osb->sb, bh->b_data, &fe->i_check);
1721 status = ocfs2_write_block(osb, bh, INODE_CACHE(inode));
1725 BUG_ON(!igrab(inode));
1727 jbd2_journal_destroy(journal);
1729 printk(KERN_NOTICE "ocfs2: End replay journal (node %d, slot %d) on "\
1730 "device (%u,%u)\n", node_num, slot_num, MAJOR(osb->sb->s_dev),
1731 MINOR(osb->sb->s_dev));
1733 /* drop the lock on this nodes journal */
1735 ocfs2_inode_unlock(inode, 1);
1744 * Do the most important parts of node recovery:
1745 * - Replay it's journal
1746 * - Stamp a clean local allocator file
1747 * - Stamp a clean truncate log
1748 * - Mark the node clean
1750 * If this function completes without error, a node in OCFS2 can be
1751 * said to have been safely recovered. As a result, failure during the
1752 * second part of a nodes recovery process (local alloc recovery) is
1753 * far less concerning.
1755 static int ocfs2_recover_node(struct ocfs2_super *osb,
1756 int node_num, int slot_num)
1759 struct ocfs2_dinode *la_copy = NULL;
1760 struct ocfs2_dinode *tl_copy = NULL;
1762 trace_ocfs2_recover_node(node_num, slot_num, osb->node_num);
1764 /* Should not ever be called to recover ourselves -- in that
1765 * case we should've called ocfs2_journal_load instead. */
1766 BUG_ON(osb->node_num == node_num);
1768 status = ocfs2_replay_journal(osb, node_num, slot_num);
1770 if (status == -EBUSY) {
1771 trace_ocfs2_recover_node_skip(slot_num, node_num);
1779 /* Stamp a clean local alloc file AFTER recovering the journal... */
1780 status = ocfs2_begin_local_alloc_recovery(osb, slot_num, &la_copy);
1786 /* An error from begin_truncate_log_recovery is not
1787 * serious enough to warrant halting the rest of
1789 status = ocfs2_begin_truncate_log_recovery(osb, slot_num, &tl_copy);
1793 /* Likewise, this would be a strange but ultimately not so
1794 * harmful place to get an error... */
1795 status = ocfs2_clear_slot(osb, slot_num);
1799 /* This will kfree the memory pointed to by la_copy and tl_copy */
1800 ocfs2_queue_recovery_completion(osb->journal, slot_num, la_copy,
1801 tl_copy, NULL, ORPHAN_NEED_TRUNCATE);
1809 /* Test node liveness by trylocking his journal. If we get the lock,
1810 * we drop it here. Return 0 if we got the lock, -EAGAIN if node is
1811 * still alive (we couldn't get the lock) and < 0 on error. */
1812 static int ocfs2_trylock_journal(struct ocfs2_super *osb,
1816 struct inode *inode = NULL;
1818 inode = ocfs2_get_system_file_inode(osb, JOURNAL_SYSTEM_INODE,
1820 if (inode == NULL) {
1821 mlog(ML_ERROR, "access error\n");
1825 if (is_bad_inode(inode)) {
1826 mlog(ML_ERROR, "access error (bad inode)\n");
1832 SET_INODE_JOURNAL(inode);
1834 flags = OCFS2_META_LOCK_RECOVERY | OCFS2_META_LOCK_NOQUEUE;
1835 status = ocfs2_inode_lock_full(inode, NULL, 1, flags);
1837 if (status != -EAGAIN)
1842 ocfs2_inode_unlock(inode, 1);
1849 /* Call this underneath ocfs2_super_lock. It also assumes that the
1850 * slot info struct has been updated from disk. */
1851 int ocfs2_mark_dead_nodes(struct ocfs2_super *osb)
1853 unsigned int node_num;
1856 struct buffer_head *bh = NULL;
1857 struct ocfs2_dinode *di;
1859 /* This is called with the super block cluster lock, so we
1860 * know that the slot map can't change underneath us. */
1862 for (i = 0; i < osb->max_slots; i++) {
1863 /* Read journal inode to get the recovery generation */
1864 status = ocfs2_read_journal_inode(osb, i, &bh, NULL);
1869 di = (struct ocfs2_dinode *)bh->b_data;
1870 gen = ocfs2_get_recovery_generation(di);
1874 spin_lock(&osb->osb_lock);
1875 osb->slot_recovery_generations[i] = gen;
1877 trace_ocfs2_mark_dead_nodes(i,
1878 osb->slot_recovery_generations[i]);
1880 if (i == osb->slot_num) {
1881 spin_unlock(&osb->osb_lock);
1885 status = ocfs2_slot_to_node_num_locked(osb, i, &node_num);
1886 if (status == -ENOENT) {
1887 spin_unlock(&osb->osb_lock);
1891 if (__ocfs2_recovery_map_test(osb, node_num)) {
1892 spin_unlock(&osb->osb_lock);
1895 spin_unlock(&osb->osb_lock);
1897 /* Ok, we have a slot occupied by another node which
1898 * is not in the recovery map. We trylock his journal
1899 * file here to test if he's alive. */
1900 status = ocfs2_trylock_journal(osb, i);
1902 /* Since we're called from mount, we know that
1903 * the recovery thread can't race us on
1904 * setting / checking the recovery bits. */
1905 ocfs2_recovery_thread(osb, node_num);
1906 } else if ((status < 0) && (status != -EAGAIN)) {
1918 * Scan timer should get fired every ORPHAN_SCAN_SCHEDULE_TIMEOUT. Add some
1919 * randomness to the timeout to minimize multple nodes firing the timer at the
1922 static inline unsigned long ocfs2_orphan_scan_timeout(void)
1926 get_random_bytes(&time, sizeof(time));
1927 time = ORPHAN_SCAN_SCHEDULE_TIMEOUT + (time % 5000);
1928 return msecs_to_jiffies(time);
1932 * ocfs2_queue_orphan_scan calls ocfs2_queue_recovery_completion for
1933 * every slot, queuing a recovery of the slot on the ocfs2_wq thread. This
1934 * is done to catch any orphans that are left over in orphan directories.
1936 * It scans all slots, even ones that are in use. It does so to handle the
1937 * case described below:
1939 * Node 1 has an inode it was using. The dentry went away due to memory
1940 * pressure. Node 1 closes the inode, but it's on the free list. The node
1941 * has the open lock.
1942 * Node 2 unlinks the inode. It grabs the dentry lock to notify others,
1943 * but node 1 has no dentry and doesn't get the message. It trylocks the
1944 * open lock, sees that another node has a PR, and does nothing.
1945 * Later node 2 runs its orphan dir. It igets the inode, trylocks the
1946 * open lock, sees the PR still, and does nothing.
1947 * Basically, we have to trigger an orphan iput on node 1. The only way
1948 * for this to happen is if node 1 runs node 2's orphan dir.
1950 * ocfs2_queue_orphan_scan gets called every ORPHAN_SCAN_SCHEDULE_TIMEOUT
1951 * seconds. It gets an EX lock on os_lockres and checks sequence number
1952 * stored in LVB. If the sequence number has changed, it means some other
1953 * node has done the scan. This node skips the scan and tracks the
1954 * sequence number. If the sequence number didn't change, it means a scan
1955 * hasn't happened. The node queues a scan and increments the
1956 * sequence number in the LVB.
1958 static void ocfs2_queue_orphan_scan(struct ocfs2_super *osb)
1960 struct ocfs2_orphan_scan *os;
1964 os = &osb->osb_orphan_scan;
1966 if (atomic_read(&os->os_state) == ORPHAN_SCAN_INACTIVE)
1969 trace_ocfs2_queue_orphan_scan_begin(os->os_count, os->os_seqno,
1970 atomic_read(&os->os_state));
1972 status = ocfs2_orphan_scan_lock(osb, &seqno);
1974 if (status != -EAGAIN)
1979 /* Do no queue the tasks if the volume is being umounted */
1980 if (atomic_read(&os->os_state) == ORPHAN_SCAN_INACTIVE)
1983 if (os->os_seqno != seqno) {
1984 os->os_seqno = seqno;
1988 for (i = 0; i < osb->max_slots; i++)
1989 ocfs2_queue_recovery_completion(osb->journal, i, NULL, NULL,
1990 NULL, ORPHAN_NO_NEED_TRUNCATE);
1992 * We queued a recovery on orphan slots, increment the sequence
1993 * number and update LVB so other node will skip the scan for a while
1997 os->os_scantime = ktime_get_seconds();
1999 ocfs2_orphan_scan_unlock(osb, seqno);
2001 trace_ocfs2_queue_orphan_scan_end(os->os_count, os->os_seqno,
2002 atomic_read(&os->os_state));
2006 /* Worker task that gets fired every ORPHAN_SCAN_SCHEDULE_TIMEOUT millsec */
2007 static void ocfs2_orphan_scan_work(struct work_struct *work)
2009 struct ocfs2_orphan_scan *os;
2010 struct ocfs2_super *osb;
2012 os = container_of(work, struct ocfs2_orphan_scan,
2013 os_orphan_scan_work.work);
2016 mutex_lock(&os->os_lock);
2017 ocfs2_queue_orphan_scan(osb);
2018 if (atomic_read(&os->os_state) == ORPHAN_SCAN_ACTIVE)
2019 queue_delayed_work(osb->ocfs2_wq, &os->os_orphan_scan_work,
2020 ocfs2_orphan_scan_timeout());
2021 mutex_unlock(&os->os_lock);
2024 void ocfs2_orphan_scan_stop(struct ocfs2_super *osb)
2026 struct ocfs2_orphan_scan *os;
2028 os = &osb->osb_orphan_scan;
2029 if (atomic_read(&os->os_state) == ORPHAN_SCAN_ACTIVE) {
2030 atomic_set(&os->os_state, ORPHAN_SCAN_INACTIVE);
2031 mutex_lock(&os->os_lock);
2032 cancel_delayed_work(&os->os_orphan_scan_work);
2033 mutex_unlock(&os->os_lock);
2037 void ocfs2_orphan_scan_init(struct ocfs2_super *osb)
2039 struct ocfs2_orphan_scan *os;
2041 os = &osb->osb_orphan_scan;
2045 mutex_init(&os->os_lock);
2046 INIT_DELAYED_WORK(&os->os_orphan_scan_work, ocfs2_orphan_scan_work);
2049 void ocfs2_orphan_scan_start(struct ocfs2_super *osb)
2051 struct ocfs2_orphan_scan *os;
2053 os = &osb->osb_orphan_scan;
2054 os->os_scantime = ktime_get_seconds();
2055 if (ocfs2_is_hard_readonly(osb) || ocfs2_mount_local(osb))
2056 atomic_set(&os->os_state, ORPHAN_SCAN_INACTIVE);
2058 atomic_set(&os->os_state, ORPHAN_SCAN_ACTIVE);
2059 queue_delayed_work(osb->ocfs2_wq, &os->os_orphan_scan_work,
2060 ocfs2_orphan_scan_timeout());
2064 struct ocfs2_orphan_filldir_priv {
2065 struct dir_context ctx;
2067 struct ocfs2_super *osb;
2068 enum ocfs2_orphan_reco_type orphan_reco_type;
2071 static bool ocfs2_orphan_filldir(struct dir_context *ctx, const char *name,
2072 int name_len, loff_t pos, u64 ino,
2075 struct ocfs2_orphan_filldir_priv *p =
2076 container_of(ctx, struct ocfs2_orphan_filldir_priv, ctx);
2079 if (name_len == 1 && !strncmp(".", name, 1))
2081 if (name_len == 2 && !strncmp("..", name, 2))
2084 /* do not include dio entry in case of orphan scan */
2085 if ((p->orphan_reco_type == ORPHAN_NO_NEED_TRUNCATE) &&
2086 (!strncmp(name, OCFS2_DIO_ORPHAN_PREFIX,
2087 OCFS2_DIO_ORPHAN_PREFIX_LEN)))
2090 /* Skip bad inodes so that recovery can continue */
2091 iter = ocfs2_iget(p->osb, ino,
2092 OCFS2_FI_FLAG_ORPHAN_RECOVERY, 0);
2096 if (!strncmp(name, OCFS2_DIO_ORPHAN_PREFIX,
2097 OCFS2_DIO_ORPHAN_PREFIX_LEN))
2098 OCFS2_I(iter)->ip_flags |= OCFS2_INODE_DIO_ORPHAN_ENTRY;
2100 /* Skip inodes which are already added to recover list, since dio may
2101 * happen concurrently with unlink/rename */
2102 if (OCFS2_I(iter)->ip_next_orphan) {
2107 trace_ocfs2_orphan_filldir((unsigned long long)OCFS2_I(iter)->ip_blkno);
2108 /* No locking is required for the next_orphan queue as there
2109 * is only ever a single process doing orphan recovery. */
2110 OCFS2_I(iter)->ip_next_orphan = p->head;
2116 static int ocfs2_queue_orphans(struct ocfs2_super *osb,
2118 struct inode **head,
2119 enum ocfs2_orphan_reco_type orphan_reco_type)
2122 struct inode *orphan_dir_inode = NULL;
2123 struct ocfs2_orphan_filldir_priv priv = {
2124 .ctx.actor = ocfs2_orphan_filldir,
2127 .orphan_reco_type = orphan_reco_type
2130 orphan_dir_inode = ocfs2_get_system_file_inode(osb,
2131 ORPHAN_DIR_SYSTEM_INODE,
2133 if (!orphan_dir_inode) {
2139 inode_lock(orphan_dir_inode);
2140 status = ocfs2_inode_lock(orphan_dir_inode, NULL, 0);
2146 status = ocfs2_dir_foreach(orphan_dir_inode, &priv.ctx);
2155 ocfs2_inode_unlock(orphan_dir_inode, 0);
2157 inode_unlock(orphan_dir_inode);
2158 iput(orphan_dir_inode);
2162 static int ocfs2_orphan_recovery_can_continue(struct ocfs2_super *osb,
2167 spin_lock(&osb->osb_lock);
2168 ret = !osb->osb_orphan_wipes[slot];
2169 spin_unlock(&osb->osb_lock);
2173 static void ocfs2_mark_recovering_orphan_dir(struct ocfs2_super *osb,
2176 spin_lock(&osb->osb_lock);
2177 /* Mark ourselves such that new processes in delete_inode()
2178 * know to quit early. */
2179 ocfs2_node_map_set_bit(osb, &osb->osb_recovering_orphan_dirs, slot);
2180 while (osb->osb_orphan_wipes[slot]) {
2181 /* If any processes are already in the middle of an
2182 * orphan wipe on this dir, then we need to wait for
2184 spin_unlock(&osb->osb_lock);
2185 wait_event_interruptible(osb->osb_wipe_event,
2186 ocfs2_orphan_recovery_can_continue(osb, slot));
2187 spin_lock(&osb->osb_lock);
2189 spin_unlock(&osb->osb_lock);
2192 static void ocfs2_clear_recovering_orphan_dir(struct ocfs2_super *osb,
2195 ocfs2_node_map_clear_bit(osb, &osb->osb_recovering_orphan_dirs, slot);
2199 * Orphan recovery. Each mounted node has it's own orphan dir which we
2200 * must run during recovery. Our strategy here is to build a list of
2201 * the inodes in the orphan dir and iget/iput them. The VFS does
2202 * (most) of the rest of the work.
2204 * Orphan recovery can happen at any time, not just mount so we have a
2205 * couple of extra considerations.
2207 * - We grab as many inodes as we can under the orphan dir lock -
2208 * doing iget() outside the orphan dir risks getting a reference on
2210 * - We must be sure not to deadlock with other processes on the
2211 * system wanting to run delete_inode(). This can happen when they go
2212 * to lock the orphan dir and the orphan recovery process attempts to
2213 * iget() inside the orphan dir lock. This can be avoided by
2214 * advertising our state to ocfs2_delete_inode().
2216 static int ocfs2_recover_orphans(struct ocfs2_super *osb,
2218 enum ocfs2_orphan_reco_type orphan_reco_type)
2221 struct inode *inode = NULL;
2223 struct ocfs2_inode_info *oi;
2224 struct buffer_head *di_bh = NULL;
2225 struct ocfs2_dinode *di = NULL;
2227 trace_ocfs2_recover_orphans(slot);
2229 ocfs2_mark_recovering_orphan_dir(osb, slot);
2230 ret = ocfs2_queue_orphans(osb, slot, &inode, orphan_reco_type);
2231 ocfs2_clear_recovering_orphan_dir(osb, slot);
2233 /* Error here should be noted, but we want to continue with as
2234 * many queued inodes as we've got. */
2239 oi = OCFS2_I(inode);
2240 trace_ocfs2_recover_orphans_iput(
2241 (unsigned long long)oi->ip_blkno);
2243 iter = oi->ip_next_orphan;
2244 oi->ip_next_orphan = NULL;
2246 if (oi->ip_flags & OCFS2_INODE_DIO_ORPHAN_ENTRY) {
2248 ret = ocfs2_rw_lock(inode, 1);
2254 * We need to take and drop the inode lock to
2255 * force read inode from disk.
2257 ret = ocfs2_inode_lock(inode, &di_bh, 1);
2263 di = (struct ocfs2_dinode *)di_bh->b_data;
2265 if (di->i_flags & cpu_to_le32(OCFS2_DIO_ORPHANED_FL)) {
2266 ret = ocfs2_truncate_file(inode, di_bh,
2267 i_size_read(inode));
2274 ret = ocfs2_del_inode_from_orphan(osb, inode,
2280 ocfs2_inode_unlock(inode, 1);
2284 ocfs2_rw_unlock(inode, 1);
2286 inode_unlock(inode);
2288 /* clear dio flag in ocfs2_inode_info */
2289 oi->ip_flags &= ~OCFS2_INODE_DIO_ORPHAN_ENTRY;
2291 spin_lock(&oi->ip_lock);
2292 /* Set the proper information to get us going into
2293 * ocfs2_delete_inode. */
2294 oi->ip_flags |= OCFS2_INODE_MAYBE_ORPHANED;
2295 spin_unlock(&oi->ip_lock);
2305 static int __ocfs2_wait_on_mount(struct ocfs2_super *osb, int quota)
2307 /* This check is good because ocfs2 will wait on our recovery
2308 * thread before changing it to something other than MOUNTED
2310 wait_event(osb->osb_mount_event,
2311 (!quota && atomic_read(&osb->vol_state) == VOLUME_MOUNTED) ||
2312 atomic_read(&osb->vol_state) == VOLUME_MOUNTED_QUOTAS ||
2313 atomic_read(&osb->vol_state) == VOLUME_DISABLED);
2315 /* If there's an error on mount, then we may never get to the
2316 * MOUNTED flag, but this is set right before
2317 * dismount_volume() so we can trust it. */
2318 if (atomic_read(&osb->vol_state) == VOLUME_DISABLED) {
2319 trace_ocfs2_wait_on_mount(VOLUME_DISABLED);
2320 mlog(0, "mount error, exiting!\n");
2327 static int ocfs2_commit_thread(void *arg)
2330 struct ocfs2_super *osb = arg;
2331 struct ocfs2_journal *journal = osb->journal;
2333 /* we can trust j_num_trans here because _should_stop() is only set in
2334 * shutdown and nobody other than ourselves should be able to start
2335 * transactions. committing on shutdown might take a few iterations
2336 * as final transactions put deleted inodes on the list */
2337 while (!(kthread_should_stop() &&
2338 atomic_read(&journal->j_num_trans) == 0)) {
2340 wait_event_interruptible(osb->checkpoint_event,
2341 atomic_read(&journal->j_num_trans)
2342 || kthread_should_stop());
2344 status = ocfs2_commit_cache(osb);
2346 static unsigned long abort_warn_time;
2348 /* Warn about this once per minute */
2349 if (printk_timed_ratelimit(&abort_warn_time, 60*HZ))
2350 mlog(ML_ERROR, "status = %d, journal is "
2351 "already aborted.\n", status);
2353 * After ocfs2_commit_cache() fails, j_num_trans has a
2354 * non-zero value. Sleep here to avoid a busy-wait
2357 msleep_interruptible(1000);
2360 if (kthread_should_stop() && atomic_read(&journal->j_num_trans)){
2362 "commit_thread: %u transactions pending on "
2364 atomic_read(&journal->j_num_trans));
2371 /* Reads all the journal inodes without taking any cluster locks. Used
2372 * for hard readonly access to determine whether any journal requires
2373 * recovery. Also used to refresh the recovery generation numbers after
2374 * a journal has been recovered by another node.
2376 int ocfs2_check_journals_nolocks(struct ocfs2_super *osb)
2380 struct buffer_head *di_bh = NULL;
2381 struct ocfs2_dinode *di;
2382 int journal_dirty = 0;
2384 for(slot = 0; slot < osb->max_slots; slot++) {
2385 ret = ocfs2_read_journal_inode(osb, slot, &di_bh, NULL);
2391 di = (struct ocfs2_dinode *) di_bh->b_data;
2393 osb->slot_recovery_generations[slot] =
2394 ocfs2_get_recovery_generation(di);
2396 if (le32_to_cpu(di->id1.journal1.ij_flags) &
2397 OCFS2_JOURNAL_DIRTY_FL)