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>
19 #include <cluster/masklog.h>
24 #include "blockcheck.h"
27 #include "extent_map.h"
28 #include "heartbeat.h"
31 #include "localalloc.h"
40 #include "buffer_head_io.h"
41 #include "ocfs2_trace.h"
43 DEFINE_SPINLOCK(trans_inc_lock);
45 #define ORPHAN_SCAN_SCHEDULE_TIMEOUT 300000
47 static int ocfs2_force_read_journal(struct inode *inode);
48 static int ocfs2_recover_node(struct ocfs2_super *osb,
49 int node_num, int slot_num);
50 static int __ocfs2_recovery_thread(void *arg);
51 static int ocfs2_commit_cache(struct ocfs2_super *osb);
52 static int __ocfs2_wait_on_mount(struct ocfs2_super *osb, int quota);
53 static int ocfs2_journal_toggle_dirty(struct ocfs2_super *osb,
54 int dirty, int replayed);
55 static int ocfs2_trylock_journal(struct ocfs2_super *osb,
57 static int ocfs2_recover_orphans(struct ocfs2_super *osb,
59 enum ocfs2_orphan_reco_type orphan_reco_type);
60 static int ocfs2_commit_thread(void *arg);
61 static void ocfs2_queue_recovery_completion(struct ocfs2_journal *journal,
63 struct ocfs2_dinode *la_dinode,
64 struct ocfs2_dinode *tl_dinode,
65 struct ocfs2_quota_recovery *qrec,
66 enum ocfs2_orphan_reco_type orphan_reco_type);
68 static inline int ocfs2_wait_on_mount(struct ocfs2_super *osb)
70 return __ocfs2_wait_on_mount(osb, 0);
73 static inline int ocfs2_wait_on_quotas(struct ocfs2_super *osb)
75 return __ocfs2_wait_on_mount(osb, 1);
79 * This replay_map is to track online/offline slots, so we could recover
80 * offline slots during recovery and mount
83 enum ocfs2_replay_state {
84 REPLAY_UNNEEDED = 0, /* Replay is not needed, so ignore this map */
85 REPLAY_NEEDED, /* Replay slots marked in rm_replay_slots */
86 REPLAY_DONE /* Replay was already queued */
89 struct ocfs2_replay_map {
90 unsigned int rm_slots;
91 enum ocfs2_replay_state rm_state;
92 unsigned char rm_replay_slots[];
95 static void ocfs2_replay_map_set_state(struct ocfs2_super *osb, int state)
100 /* If we've already queued the replay, we don't have any more to do */
101 if (osb->replay_map->rm_state == REPLAY_DONE)
104 osb->replay_map->rm_state = state;
107 int ocfs2_compute_replay_slots(struct ocfs2_super *osb)
109 struct ocfs2_replay_map *replay_map;
112 /* If replay map is already set, we don't do it again */
116 replay_map = kzalloc(sizeof(struct ocfs2_replay_map) +
117 (osb->max_slots * sizeof(char)), GFP_KERNEL);
124 spin_lock(&osb->osb_lock);
126 replay_map->rm_slots = osb->max_slots;
127 replay_map->rm_state = REPLAY_UNNEEDED;
129 /* set rm_replay_slots for offline slot(s) */
130 for (i = 0; i < replay_map->rm_slots; i++) {
131 if (ocfs2_slot_to_node_num_locked(osb, i, &node_num) == -ENOENT)
132 replay_map->rm_replay_slots[i] = 1;
135 osb->replay_map = replay_map;
136 spin_unlock(&osb->osb_lock);
140 static void ocfs2_queue_replay_slots(struct ocfs2_super *osb,
141 enum ocfs2_orphan_reco_type orphan_reco_type)
143 struct ocfs2_replay_map *replay_map = osb->replay_map;
149 if (replay_map->rm_state != REPLAY_NEEDED)
152 for (i = 0; i < replay_map->rm_slots; i++)
153 if (replay_map->rm_replay_slots[i])
154 ocfs2_queue_recovery_completion(osb->journal, i, NULL,
157 replay_map->rm_state = REPLAY_DONE;
160 static void ocfs2_free_replay_slots(struct ocfs2_super *osb)
162 struct ocfs2_replay_map *replay_map = osb->replay_map;
164 if (!osb->replay_map)
168 osb->replay_map = NULL;
171 int ocfs2_recovery_init(struct ocfs2_super *osb)
173 struct ocfs2_recovery_map *rm;
175 mutex_init(&osb->recovery_lock);
176 osb->disable_recovery = 0;
177 osb->recovery_thread_task = NULL;
178 init_waitqueue_head(&osb->recovery_event);
180 rm = kzalloc(sizeof(struct ocfs2_recovery_map) +
181 osb->max_slots * sizeof(unsigned int),
188 rm->rm_entries = (unsigned int *)((char *)rm +
189 sizeof(struct ocfs2_recovery_map));
190 osb->recovery_map = rm;
195 /* we can't grab the goofy sem lock from inside wait_event, so we use
196 * memory barriers to make sure that we'll see the null task before
198 static int ocfs2_recovery_thread_running(struct ocfs2_super *osb)
201 return osb->recovery_thread_task != NULL;
204 void ocfs2_recovery_exit(struct ocfs2_super *osb)
206 struct ocfs2_recovery_map *rm;
208 /* disable any new recovery threads and wait for any currently
209 * running ones to exit. Do this before setting the vol_state. */
210 mutex_lock(&osb->recovery_lock);
211 osb->disable_recovery = 1;
212 mutex_unlock(&osb->recovery_lock);
213 wait_event(osb->recovery_event, !ocfs2_recovery_thread_running(osb));
215 /* At this point, we know that no more recovery threads can be
216 * launched, so wait for any recovery completion work to
219 flush_workqueue(osb->ocfs2_wq);
222 * Now that recovery is shut down, and the osb is about to be
223 * freed, the osb_lock is not taken here.
225 rm = osb->recovery_map;
226 /* XXX: Should we bug if there are dirty entries? */
231 static int __ocfs2_recovery_map_test(struct ocfs2_super *osb,
232 unsigned int node_num)
235 struct ocfs2_recovery_map *rm = osb->recovery_map;
237 assert_spin_locked(&osb->osb_lock);
239 for (i = 0; i < rm->rm_used; i++) {
240 if (rm->rm_entries[i] == node_num)
247 /* Behaves like test-and-set. Returns the previous value */
248 static int ocfs2_recovery_map_set(struct ocfs2_super *osb,
249 unsigned int node_num)
251 struct ocfs2_recovery_map *rm = osb->recovery_map;
253 spin_lock(&osb->osb_lock);
254 if (__ocfs2_recovery_map_test(osb, node_num)) {
255 spin_unlock(&osb->osb_lock);
259 /* XXX: Can this be exploited? Not from o2dlm... */
260 BUG_ON(rm->rm_used >= osb->max_slots);
262 rm->rm_entries[rm->rm_used] = node_num;
264 spin_unlock(&osb->osb_lock);
269 static void ocfs2_recovery_map_clear(struct ocfs2_super *osb,
270 unsigned int node_num)
273 struct ocfs2_recovery_map *rm = osb->recovery_map;
275 spin_lock(&osb->osb_lock);
277 for (i = 0; i < rm->rm_used; i++) {
278 if (rm->rm_entries[i] == node_num)
282 if (i < rm->rm_used) {
283 /* XXX: be careful with the pointer math */
284 memmove(&(rm->rm_entries[i]), &(rm->rm_entries[i + 1]),
285 (rm->rm_used - i - 1) * sizeof(unsigned int));
289 spin_unlock(&osb->osb_lock);
292 static int ocfs2_commit_cache(struct ocfs2_super *osb)
295 unsigned int flushed;
296 struct ocfs2_journal *journal = NULL;
298 journal = osb->journal;
300 /* Flush all pending commits and checkpoint the journal. */
301 down_write(&journal->j_trans_barrier);
303 flushed = atomic_read(&journal->j_num_trans);
304 trace_ocfs2_commit_cache_begin(flushed);
306 up_write(&journal->j_trans_barrier);
310 jbd2_journal_lock_updates(journal->j_journal);
311 status = jbd2_journal_flush(journal->j_journal, 0);
312 jbd2_journal_unlock_updates(journal->j_journal);
314 up_write(&journal->j_trans_barrier);
319 ocfs2_inc_trans_id(journal);
321 flushed = atomic_read(&journal->j_num_trans);
322 atomic_set(&journal->j_num_trans, 0);
323 up_write(&journal->j_trans_barrier);
325 trace_ocfs2_commit_cache_end(journal->j_trans_id, flushed);
327 ocfs2_wake_downconvert_thread(osb);
328 wake_up(&journal->j_checkpointed);
333 handle_t *ocfs2_start_trans(struct ocfs2_super *osb, int max_buffs)
335 journal_t *journal = osb->journal->j_journal;
338 BUG_ON(!osb || !osb->journal->j_journal);
340 if (ocfs2_is_hard_readonly(osb))
341 return ERR_PTR(-EROFS);
343 BUG_ON(osb->journal->j_state == OCFS2_JOURNAL_FREE);
344 BUG_ON(max_buffs <= 0);
346 /* Nested transaction? Just return the handle... */
347 if (journal_current_handle())
348 return jbd2_journal_start(journal, max_buffs);
350 sb_start_intwrite(osb->sb);
352 down_read(&osb->journal->j_trans_barrier);
354 handle = jbd2_journal_start(journal, max_buffs);
355 if (IS_ERR(handle)) {
356 up_read(&osb->journal->j_trans_barrier);
357 sb_end_intwrite(osb->sb);
359 mlog_errno(PTR_ERR(handle));
361 if (is_journal_aborted(journal)) {
362 ocfs2_abort(osb->sb, "Detected aborted journal\n");
363 handle = ERR_PTR(-EROFS);
366 if (!ocfs2_mount_local(osb))
367 atomic_inc(&(osb->journal->j_num_trans));
373 int ocfs2_commit_trans(struct ocfs2_super *osb,
377 struct ocfs2_journal *journal = osb->journal;
381 nested = handle->h_ref > 1;
382 ret = jbd2_journal_stop(handle);
387 up_read(&journal->j_trans_barrier);
388 sb_end_intwrite(osb->sb);
395 * 'nblocks' is what you want to add to the current transaction.
397 * This might call jbd2_journal_restart() which will commit dirty buffers
398 * and then restart the transaction. Before calling
399 * ocfs2_extend_trans(), any changed blocks should have been
400 * dirtied. After calling it, all blocks which need to be changed must
401 * go through another set of journal_access/journal_dirty calls.
403 * WARNING: This will not release any semaphores or disk locks taken
404 * during the transaction, so make sure they were taken *before*
405 * start_trans or we'll have ordering deadlocks.
407 * WARNING2: Note that we do *not* drop j_trans_barrier here. This is
408 * good because transaction ids haven't yet been recorded on the
409 * cluster locks associated with this handle.
411 int ocfs2_extend_trans(handle_t *handle, int nblocks)
413 int status, old_nblocks;
421 old_nblocks = jbd2_handle_buffer_credits(handle);
423 trace_ocfs2_extend_trans(old_nblocks, nblocks);
425 #ifdef CONFIG_OCFS2_DEBUG_FS
428 status = jbd2_journal_extend(handle, nblocks, 0);
436 trace_ocfs2_extend_trans_restart(old_nblocks + nblocks);
437 status = jbd2_journal_restart(handle,
438 old_nblocks + nblocks);
451 * If we have fewer than thresh credits, extend by OCFS2_MAX_TRANS_DATA.
452 * If that fails, restart the transaction & regain write access for the
453 * buffer head which is used for metadata modifications.
454 * Taken from Ext4: extend_or_restart_transaction()
456 int ocfs2_allocate_extend_trans(handle_t *handle, int thresh)
458 int status, old_nblks;
462 old_nblks = jbd2_handle_buffer_credits(handle);
463 trace_ocfs2_allocate_extend_trans(old_nblks, thresh);
465 if (old_nblks < thresh)
468 status = jbd2_journal_extend(handle, OCFS2_MAX_TRANS_DATA, 0);
475 status = jbd2_journal_restart(handle, OCFS2_MAX_TRANS_DATA);
485 struct ocfs2_triggers {
486 struct jbd2_buffer_trigger_type ot_triggers;
490 static inline struct ocfs2_triggers *to_ocfs2_trigger(struct jbd2_buffer_trigger_type *triggers)
492 return container_of(triggers, struct ocfs2_triggers, ot_triggers);
495 static void ocfs2_frozen_trigger(struct jbd2_buffer_trigger_type *triggers,
496 struct buffer_head *bh,
497 void *data, size_t size)
499 struct ocfs2_triggers *ot = to_ocfs2_trigger(triggers);
502 * We aren't guaranteed to have the superblock here, so we
503 * must unconditionally compute the ecc data.
504 * __ocfs2_journal_access() will only set the triggers if
505 * metaecc is enabled.
507 ocfs2_block_check_compute(data, size, data + ot->ot_offset);
511 * Quota blocks have their own trigger because the struct ocfs2_block_check
512 * offset depends on the blocksize.
514 static void ocfs2_dq_frozen_trigger(struct jbd2_buffer_trigger_type *triggers,
515 struct buffer_head *bh,
516 void *data, size_t size)
518 struct ocfs2_disk_dqtrailer *dqt =
519 ocfs2_block_dqtrailer(size, data);
522 * We aren't guaranteed to have the superblock here, so we
523 * must unconditionally compute the ecc data.
524 * __ocfs2_journal_access() will only set the triggers if
525 * metaecc is enabled.
527 ocfs2_block_check_compute(data, size, &dqt->dq_check);
531 * Directory blocks also have their own trigger because the
532 * struct ocfs2_block_check offset depends on the blocksize.
534 static void ocfs2_db_frozen_trigger(struct jbd2_buffer_trigger_type *triggers,
535 struct buffer_head *bh,
536 void *data, size_t size)
538 struct ocfs2_dir_block_trailer *trailer =
539 ocfs2_dir_trailer_from_size(size, data);
542 * We aren't guaranteed to have the superblock here, so we
543 * must unconditionally compute the ecc data.
544 * __ocfs2_journal_access() will only set the triggers if
545 * metaecc is enabled.
547 ocfs2_block_check_compute(data, size, &trailer->db_check);
550 static void ocfs2_abort_trigger(struct jbd2_buffer_trigger_type *triggers,
551 struct buffer_head *bh)
554 "ocfs2_abort_trigger called by JBD2. bh = 0x%lx, "
555 "bh->b_blocknr = %llu\n",
557 (unsigned long long)bh->b_blocknr);
559 ocfs2_error(bh->b_bdev->bd_super,
560 "JBD2 has aborted our journal, ocfs2 cannot continue\n");
563 static struct ocfs2_triggers di_triggers = {
565 .t_frozen = ocfs2_frozen_trigger,
566 .t_abort = ocfs2_abort_trigger,
568 .ot_offset = offsetof(struct ocfs2_dinode, i_check),
571 static struct ocfs2_triggers eb_triggers = {
573 .t_frozen = ocfs2_frozen_trigger,
574 .t_abort = ocfs2_abort_trigger,
576 .ot_offset = offsetof(struct ocfs2_extent_block, h_check),
579 static struct ocfs2_triggers rb_triggers = {
581 .t_frozen = ocfs2_frozen_trigger,
582 .t_abort = ocfs2_abort_trigger,
584 .ot_offset = offsetof(struct ocfs2_refcount_block, rf_check),
587 static struct ocfs2_triggers gd_triggers = {
589 .t_frozen = ocfs2_frozen_trigger,
590 .t_abort = ocfs2_abort_trigger,
592 .ot_offset = offsetof(struct ocfs2_group_desc, bg_check),
595 static struct ocfs2_triggers db_triggers = {
597 .t_frozen = ocfs2_db_frozen_trigger,
598 .t_abort = ocfs2_abort_trigger,
602 static struct ocfs2_triggers xb_triggers = {
604 .t_frozen = ocfs2_frozen_trigger,
605 .t_abort = ocfs2_abort_trigger,
607 .ot_offset = offsetof(struct ocfs2_xattr_block, xb_check),
610 static struct ocfs2_triggers dq_triggers = {
612 .t_frozen = ocfs2_dq_frozen_trigger,
613 .t_abort = ocfs2_abort_trigger,
617 static struct ocfs2_triggers dr_triggers = {
619 .t_frozen = ocfs2_frozen_trigger,
620 .t_abort = ocfs2_abort_trigger,
622 .ot_offset = offsetof(struct ocfs2_dx_root_block, dr_check),
625 static struct ocfs2_triggers dl_triggers = {
627 .t_frozen = ocfs2_frozen_trigger,
628 .t_abort = ocfs2_abort_trigger,
630 .ot_offset = offsetof(struct ocfs2_dx_leaf, dl_check),
633 static int __ocfs2_journal_access(handle_t *handle,
634 struct ocfs2_caching_info *ci,
635 struct buffer_head *bh,
636 struct ocfs2_triggers *triggers,
640 struct ocfs2_super *osb =
641 OCFS2_SB(ocfs2_metadata_cache_get_super(ci));
643 BUG_ON(!ci || !ci->ci_ops);
647 trace_ocfs2_journal_access(
648 (unsigned long long)ocfs2_metadata_cache_owner(ci),
649 (unsigned long long)bh->b_blocknr, type, bh->b_size);
651 /* we can safely remove this assertion after testing. */
652 if (!buffer_uptodate(bh)) {
653 mlog(ML_ERROR, "giving me a buffer that's not uptodate!\n");
654 mlog(ML_ERROR, "b_blocknr=%llu, b_state=0x%lx\n",
655 (unsigned long long)bh->b_blocknr, bh->b_state);
659 * A previous transaction with a couple of buffer heads fail
660 * to checkpoint, so all the bhs are marked as BH_Write_EIO.
661 * For current transaction, the bh is just among those error
662 * bhs which previous transaction handle. We can't just clear
663 * its BH_Write_EIO and reuse directly, since other bhs are
664 * not written to disk yet and that will cause metadata
665 * inconsistency. So we should set fs read-only to avoid
668 if (buffer_write_io_error(bh) && !buffer_uptodate(bh)) {
670 return ocfs2_error(osb->sb, "A previous attempt to "
671 "write this buffer head failed\n");
676 /* Set the current transaction information on the ci so
677 * that the locking code knows whether it can drop it's locks
678 * on this ci or not. We're protected from the commit
679 * thread updating the current transaction id until
680 * ocfs2_commit_trans() because ocfs2_start_trans() took
681 * j_trans_barrier for us. */
682 ocfs2_set_ci_lock_trans(osb->journal, ci);
684 ocfs2_metadata_cache_io_lock(ci);
686 case OCFS2_JOURNAL_ACCESS_CREATE:
687 case OCFS2_JOURNAL_ACCESS_WRITE:
688 status = jbd2_journal_get_write_access(handle, bh);
691 case OCFS2_JOURNAL_ACCESS_UNDO:
692 status = jbd2_journal_get_undo_access(handle, bh);
697 mlog(ML_ERROR, "Unknown access type!\n");
699 if (!status && ocfs2_meta_ecc(osb) && triggers)
700 jbd2_journal_set_triggers(bh, &triggers->ot_triggers);
701 ocfs2_metadata_cache_io_unlock(ci);
704 mlog(ML_ERROR, "Error %d getting %d access to buffer!\n",
710 int ocfs2_journal_access_di(handle_t *handle, struct ocfs2_caching_info *ci,
711 struct buffer_head *bh, int type)
713 return __ocfs2_journal_access(handle, ci, bh, &di_triggers, type);
716 int ocfs2_journal_access_eb(handle_t *handle, struct ocfs2_caching_info *ci,
717 struct buffer_head *bh, int type)
719 return __ocfs2_journal_access(handle, ci, bh, &eb_triggers, type);
722 int ocfs2_journal_access_rb(handle_t *handle, struct ocfs2_caching_info *ci,
723 struct buffer_head *bh, int type)
725 return __ocfs2_journal_access(handle, ci, bh, &rb_triggers,
729 int ocfs2_journal_access_gd(handle_t *handle, struct ocfs2_caching_info *ci,
730 struct buffer_head *bh, int type)
732 return __ocfs2_journal_access(handle, ci, bh, &gd_triggers, type);
735 int ocfs2_journal_access_db(handle_t *handle, struct ocfs2_caching_info *ci,
736 struct buffer_head *bh, int type)
738 return __ocfs2_journal_access(handle, ci, bh, &db_triggers, type);
741 int ocfs2_journal_access_xb(handle_t *handle, struct ocfs2_caching_info *ci,
742 struct buffer_head *bh, int type)
744 return __ocfs2_journal_access(handle, ci, bh, &xb_triggers, type);
747 int ocfs2_journal_access_dq(handle_t *handle, struct ocfs2_caching_info *ci,
748 struct buffer_head *bh, int type)
750 return __ocfs2_journal_access(handle, ci, bh, &dq_triggers, type);
753 int ocfs2_journal_access_dr(handle_t *handle, struct ocfs2_caching_info *ci,
754 struct buffer_head *bh, int type)
756 return __ocfs2_journal_access(handle, ci, bh, &dr_triggers, type);
759 int ocfs2_journal_access_dl(handle_t *handle, struct ocfs2_caching_info *ci,
760 struct buffer_head *bh, int type)
762 return __ocfs2_journal_access(handle, ci, bh, &dl_triggers, type);
765 int ocfs2_journal_access(handle_t *handle, struct ocfs2_caching_info *ci,
766 struct buffer_head *bh, int type)
768 return __ocfs2_journal_access(handle, ci, bh, NULL, type);
771 void ocfs2_journal_dirty(handle_t *handle, struct buffer_head *bh)
775 trace_ocfs2_journal_dirty((unsigned long long)bh->b_blocknr);
777 status = jbd2_journal_dirty_metadata(handle, bh);
780 if (!is_handle_aborted(handle)) {
781 journal_t *journal = handle->h_transaction->t_journal;
782 struct super_block *sb = bh->b_bdev->bd_super;
784 mlog(ML_ERROR, "jbd2_journal_dirty_metadata failed. "
785 "Aborting transaction and journal.\n");
786 handle->h_err = status;
787 jbd2_journal_abort_handle(handle);
788 jbd2_journal_abort(journal, status);
789 ocfs2_abort(sb, "Journal already aborted.\n");
794 #define OCFS2_DEFAULT_COMMIT_INTERVAL (HZ * JBD2_DEFAULT_MAX_COMMIT_AGE)
796 void ocfs2_set_journal_params(struct ocfs2_super *osb)
798 journal_t *journal = osb->journal->j_journal;
799 unsigned long commit_interval = OCFS2_DEFAULT_COMMIT_INTERVAL;
801 if (osb->osb_commit_interval)
802 commit_interval = osb->osb_commit_interval;
804 write_lock(&journal->j_state_lock);
805 journal->j_commit_interval = commit_interval;
806 if (osb->s_mount_opt & OCFS2_MOUNT_BARRIER)
807 journal->j_flags |= JBD2_BARRIER;
809 journal->j_flags &= ~JBD2_BARRIER;
810 write_unlock(&journal->j_state_lock);
813 int ocfs2_journal_init(struct ocfs2_journal *journal, int *dirty)
816 struct inode *inode = NULL; /* the journal inode */
817 journal_t *j_journal = NULL;
818 struct ocfs2_dinode *di = NULL;
819 struct buffer_head *bh = NULL;
820 struct ocfs2_super *osb;
825 osb = journal->j_osb;
827 /* already have the inode for our journal */
828 inode = ocfs2_get_system_file_inode(osb, JOURNAL_SYSTEM_INODE,
835 if (is_bad_inode(inode)) {
836 mlog(ML_ERROR, "access error (bad inode)\n");
843 SET_INODE_JOURNAL(inode);
844 OCFS2_I(inode)->ip_open_count++;
846 /* Skip recovery waits here - journal inode metadata never
847 * changes in a live cluster so it can be considered an
848 * exception to the rule. */
849 status = ocfs2_inode_lock_full(inode, &bh, 1, OCFS2_META_LOCK_RECOVERY);
851 if (status != -ERESTARTSYS)
852 mlog(ML_ERROR, "Could not get lock on journal!\n");
857 di = (struct ocfs2_dinode *)bh->b_data;
859 if (i_size_read(inode) < OCFS2_MIN_JOURNAL_SIZE) {
860 mlog(ML_ERROR, "Journal file size (%lld) is too small!\n",
866 trace_ocfs2_journal_init(i_size_read(inode),
867 (unsigned long long)inode->i_blocks,
868 OCFS2_I(inode)->ip_clusters);
870 /* call the kernels journal init function now */
871 j_journal = jbd2_journal_init_inode(inode);
872 if (j_journal == NULL) {
873 mlog(ML_ERROR, "Linux journal layer error\n");
878 trace_ocfs2_journal_init_maxlen(j_journal->j_total_len);
880 *dirty = (le32_to_cpu(di->id1.journal1.ij_flags) &
881 OCFS2_JOURNAL_DIRTY_FL);
883 journal->j_journal = j_journal;
884 journal->j_journal->j_submit_inode_data_buffers =
885 jbd2_journal_submit_inode_data_buffers;
886 journal->j_journal->j_finish_inode_data_buffers =
887 jbd2_journal_finish_inode_data_buffers;
888 journal->j_inode = inode;
891 ocfs2_set_journal_params(osb);
893 journal->j_state = OCFS2_JOURNAL_LOADED;
899 ocfs2_inode_unlock(inode, 1);
902 OCFS2_I(inode)->ip_open_count--;
910 static void ocfs2_bump_recovery_generation(struct ocfs2_dinode *di)
912 le32_add_cpu(&(di->id1.journal1.ij_recovery_generation), 1);
915 static u32 ocfs2_get_recovery_generation(struct ocfs2_dinode *di)
917 return le32_to_cpu(di->id1.journal1.ij_recovery_generation);
920 static int ocfs2_journal_toggle_dirty(struct ocfs2_super *osb,
921 int dirty, int replayed)
925 struct ocfs2_journal *journal = osb->journal;
926 struct buffer_head *bh = journal->j_bh;
927 struct ocfs2_dinode *fe;
929 fe = (struct ocfs2_dinode *)bh->b_data;
931 /* The journal bh on the osb always comes from ocfs2_journal_init()
932 * and was validated there inside ocfs2_inode_lock_full(). It's a
933 * code bug if we mess it up. */
934 BUG_ON(!OCFS2_IS_VALID_DINODE(fe));
936 flags = le32_to_cpu(fe->id1.journal1.ij_flags);
938 flags |= OCFS2_JOURNAL_DIRTY_FL;
940 flags &= ~OCFS2_JOURNAL_DIRTY_FL;
941 fe->id1.journal1.ij_flags = cpu_to_le32(flags);
944 ocfs2_bump_recovery_generation(fe);
946 ocfs2_compute_meta_ecc(osb->sb, bh->b_data, &fe->i_check);
947 status = ocfs2_write_block(osb, bh, INODE_CACHE(journal->j_inode));
955 * If the journal has been kmalloc'd it needs to be freed after this
958 void ocfs2_journal_shutdown(struct ocfs2_super *osb)
960 struct ocfs2_journal *journal = NULL;
962 struct inode *inode = NULL;
963 int num_running_trans = 0;
967 journal = osb->journal;
971 inode = journal->j_inode;
973 if (journal->j_state != OCFS2_JOURNAL_LOADED)
976 /* need to inc inode use count - jbd2_journal_destroy will iput. */
980 num_running_trans = atomic_read(&(osb->journal->j_num_trans));
981 trace_ocfs2_journal_shutdown(num_running_trans);
983 /* Do a commit_cache here. It will flush our journal, *and*
984 * release any locks that are still held.
985 * set the SHUTDOWN flag and release the trans lock.
986 * the commit thread will take the trans lock for us below. */
987 journal->j_state = OCFS2_JOURNAL_IN_SHUTDOWN;
989 /* The OCFS2_JOURNAL_IN_SHUTDOWN will signal to commit_cache to not
990 * drop the trans_lock (which we want to hold until we
991 * completely destroy the journal. */
992 if (osb->commit_task) {
993 /* Wait for the commit thread */
994 trace_ocfs2_journal_shutdown_wait(osb->commit_task);
995 kthread_stop(osb->commit_task);
996 osb->commit_task = NULL;
999 BUG_ON(atomic_read(&(osb->journal->j_num_trans)) != 0);
1001 if (ocfs2_mount_local(osb)) {
1002 jbd2_journal_lock_updates(journal->j_journal);
1003 status = jbd2_journal_flush(journal->j_journal, 0);
1004 jbd2_journal_unlock_updates(journal->j_journal);
1009 /* Shutdown the kernel journal system */
1010 if (!jbd2_journal_destroy(journal->j_journal) && !status) {
1012 * Do not toggle if flush was unsuccessful otherwise
1013 * will leave dirty metadata in a "clean" journal
1015 status = ocfs2_journal_toggle_dirty(osb, 0, 0);
1019 journal->j_journal = NULL;
1021 OCFS2_I(inode)->ip_open_count--;
1023 /* unlock our journal */
1024 ocfs2_inode_unlock(inode, 1);
1026 brelse(journal->j_bh);
1027 journal->j_bh = NULL;
1029 journal->j_state = OCFS2_JOURNAL_FREE;
1031 // up_write(&journal->j_trans_barrier);
1036 static void ocfs2_clear_journal_error(struct super_block *sb,
1042 olderr = jbd2_journal_errno(journal);
1044 mlog(ML_ERROR, "File system error %d recorded in "
1045 "journal %u.\n", olderr, slot);
1046 mlog(ML_ERROR, "File system on device %s needs checking.\n",
1049 jbd2_journal_ack_err(journal);
1050 jbd2_journal_clear_err(journal);
1054 int ocfs2_journal_load(struct ocfs2_journal *journal, int local, int replayed)
1057 struct ocfs2_super *osb;
1061 osb = journal->j_osb;
1063 status = jbd2_journal_load(journal->j_journal);
1065 mlog(ML_ERROR, "Failed to load journal!\n");
1069 ocfs2_clear_journal_error(osb->sb, journal->j_journal, osb->slot_num);
1072 jbd2_journal_lock_updates(journal->j_journal);
1073 status = jbd2_journal_flush(journal->j_journal, 0);
1074 jbd2_journal_unlock_updates(journal->j_journal);
1079 status = ocfs2_journal_toggle_dirty(osb, 1, replayed);
1085 /* Launch the commit thread */
1087 osb->commit_task = kthread_run(ocfs2_commit_thread, osb,
1088 "ocfs2cmt-%s", osb->uuid_str);
1089 if (IS_ERR(osb->commit_task)) {
1090 status = PTR_ERR(osb->commit_task);
1091 osb->commit_task = NULL;
1092 mlog(ML_ERROR, "unable to launch ocfs2commit thread, "
1093 "error=%d", status);
1097 osb->commit_task = NULL;
1104 /* 'full' flag tells us whether we clear out all blocks or if we just
1105 * mark the journal clean */
1106 int ocfs2_journal_wipe(struct ocfs2_journal *journal, int full)
1112 status = jbd2_journal_wipe(journal->j_journal, full);
1118 status = ocfs2_journal_toggle_dirty(journal->j_osb, 0, 0);
1126 static int ocfs2_recovery_completed(struct ocfs2_super *osb)
1129 struct ocfs2_recovery_map *rm = osb->recovery_map;
1131 spin_lock(&osb->osb_lock);
1132 empty = (rm->rm_used == 0);
1133 spin_unlock(&osb->osb_lock);
1138 void ocfs2_wait_for_recovery(struct ocfs2_super *osb)
1140 wait_event(osb->recovery_event, ocfs2_recovery_completed(osb));
1144 * JBD Might read a cached version of another nodes journal file. We
1145 * don't want this as this file changes often and we get no
1146 * notification on those changes. The only way to be sure that we've
1147 * got the most up to date version of those blocks then is to force
1148 * read them off disk. Just searching through the buffer cache won't
1149 * work as there may be pages backing this file which are still marked
1150 * up to date. We know things can't change on this file underneath us
1151 * as we have the lock by now :)
1153 static int ocfs2_force_read_journal(struct inode *inode)
1157 u64 v_blkno, p_blkno, p_blocks, num_blocks;
1158 struct buffer_head *bh = NULL;
1159 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
1161 num_blocks = ocfs2_blocks_for_bytes(inode->i_sb, i_size_read(inode));
1163 while (v_blkno < num_blocks) {
1164 status = ocfs2_extent_map_get_blocks(inode, v_blkno,
1165 &p_blkno, &p_blocks, NULL);
1171 for (i = 0; i < p_blocks; i++, p_blkno++) {
1172 bh = __find_get_block(osb->sb->s_bdev, p_blkno,
1173 osb->sb->s_blocksize);
1174 /* block not cached. */
1180 /* We are reading journal data which should not
1181 * be put in the uptodate cache.
1183 status = ocfs2_read_blocks_sync(osb, p_blkno, 1, &bh);
1193 v_blkno += p_blocks;
1200 struct ocfs2_la_recovery_item {
1201 struct list_head lri_list;
1203 struct ocfs2_dinode *lri_la_dinode;
1204 struct ocfs2_dinode *lri_tl_dinode;
1205 struct ocfs2_quota_recovery *lri_qrec;
1206 enum ocfs2_orphan_reco_type lri_orphan_reco_type;
1209 /* Does the second half of the recovery process. By this point, the
1210 * node is marked clean and can actually be considered recovered,
1211 * hence it's no longer in the recovery map, but there's still some
1212 * cleanup we can do which shouldn't happen within the recovery thread
1213 * as locking in that context becomes very difficult if we are to take
1214 * recovering nodes into account.
1216 * NOTE: This function can and will sleep on recovery of other nodes
1217 * during cluster locking, just like any other ocfs2 process.
1219 void ocfs2_complete_recovery(struct work_struct *work)
1222 struct ocfs2_journal *journal =
1223 container_of(work, struct ocfs2_journal, j_recovery_work);
1224 struct ocfs2_super *osb = journal->j_osb;
1225 struct ocfs2_dinode *la_dinode, *tl_dinode;
1226 struct ocfs2_la_recovery_item *item, *n;
1227 struct ocfs2_quota_recovery *qrec;
1228 enum ocfs2_orphan_reco_type orphan_reco_type;
1229 LIST_HEAD(tmp_la_list);
1231 trace_ocfs2_complete_recovery(
1232 (unsigned long long)OCFS2_I(journal->j_inode)->ip_blkno);
1234 spin_lock(&journal->j_lock);
1235 list_splice_init(&journal->j_la_cleanups, &tmp_la_list);
1236 spin_unlock(&journal->j_lock);
1238 list_for_each_entry_safe(item, n, &tmp_la_list, lri_list) {
1239 list_del_init(&item->lri_list);
1241 ocfs2_wait_on_quotas(osb);
1243 la_dinode = item->lri_la_dinode;
1244 tl_dinode = item->lri_tl_dinode;
1245 qrec = item->lri_qrec;
1246 orphan_reco_type = item->lri_orphan_reco_type;
1248 trace_ocfs2_complete_recovery_slot(item->lri_slot,
1249 la_dinode ? le64_to_cpu(la_dinode->i_blkno) : 0,
1250 tl_dinode ? le64_to_cpu(tl_dinode->i_blkno) : 0,
1254 ret = ocfs2_complete_local_alloc_recovery(osb,
1263 ret = ocfs2_complete_truncate_log_recovery(osb,
1271 ret = ocfs2_recover_orphans(osb, item->lri_slot,
1277 ret = ocfs2_finish_quota_recovery(osb, qrec,
1281 /* Recovery info is already freed now */
1287 trace_ocfs2_complete_recovery_end(ret);
1290 /* NOTE: This function always eats your references to la_dinode and
1291 * tl_dinode, either manually on error, or by passing them to
1292 * ocfs2_complete_recovery */
1293 static void ocfs2_queue_recovery_completion(struct ocfs2_journal *journal,
1295 struct ocfs2_dinode *la_dinode,
1296 struct ocfs2_dinode *tl_dinode,
1297 struct ocfs2_quota_recovery *qrec,
1298 enum ocfs2_orphan_reco_type orphan_reco_type)
1300 struct ocfs2_la_recovery_item *item;
1302 item = kmalloc(sizeof(struct ocfs2_la_recovery_item), GFP_NOFS);
1304 /* Though we wish to avoid it, we are in fact safe in
1305 * skipping local alloc cleanup as fsck.ocfs2 is more
1306 * than capable of reclaiming unused space. */
1311 ocfs2_free_quota_recovery(qrec);
1313 mlog_errno(-ENOMEM);
1317 INIT_LIST_HEAD(&item->lri_list);
1318 item->lri_la_dinode = la_dinode;
1319 item->lri_slot = slot_num;
1320 item->lri_tl_dinode = tl_dinode;
1321 item->lri_qrec = qrec;
1322 item->lri_orphan_reco_type = orphan_reco_type;
1324 spin_lock(&journal->j_lock);
1325 list_add_tail(&item->lri_list, &journal->j_la_cleanups);
1326 queue_work(journal->j_osb->ocfs2_wq, &journal->j_recovery_work);
1327 spin_unlock(&journal->j_lock);
1330 /* Called by the mount code to queue recovery the last part of
1331 * recovery for it's own and offline slot(s). */
1332 void ocfs2_complete_mount_recovery(struct ocfs2_super *osb)
1334 struct ocfs2_journal *journal = osb->journal;
1336 if (ocfs2_is_hard_readonly(osb))
1339 /* No need to queue up our truncate_log as regular cleanup will catch
1341 ocfs2_queue_recovery_completion(journal, osb->slot_num,
1342 osb->local_alloc_copy, NULL, NULL,
1343 ORPHAN_NEED_TRUNCATE);
1344 ocfs2_schedule_truncate_log_flush(osb, 0);
1346 osb->local_alloc_copy = NULL;
1348 /* queue to recover orphan slots for all offline slots */
1349 ocfs2_replay_map_set_state(osb, REPLAY_NEEDED);
1350 ocfs2_queue_replay_slots(osb, ORPHAN_NEED_TRUNCATE);
1351 ocfs2_free_replay_slots(osb);
1354 void ocfs2_complete_quota_recovery(struct ocfs2_super *osb)
1356 if (osb->quota_rec) {
1357 ocfs2_queue_recovery_completion(osb->journal,
1362 ORPHAN_NEED_TRUNCATE);
1363 osb->quota_rec = NULL;
1367 static int __ocfs2_recovery_thread(void *arg)
1369 int status, node_num, slot_num;
1370 struct ocfs2_super *osb = arg;
1371 struct ocfs2_recovery_map *rm = osb->recovery_map;
1372 int *rm_quota = NULL;
1373 int rm_quota_used = 0, i;
1374 struct ocfs2_quota_recovery *qrec;
1376 /* Whether the quota supported. */
1377 int quota_enabled = OCFS2_HAS_RO_COMPAT_FEATURE(osb->sb,
1378 OCFS2_FEATURE_RO_COMPAT_USRQUOTA)
1379 || OCFS2_HAS_RO_COMPAT_FEATURE(osb->sb,
1380 OCFS2_FEATURE_RO_COMPAT_GRPQUOTA);
1382 status = ocfs2_wait_on_mount(osb);
1387 if (quota_enabled) {
1388 rm_quota = kcalloc(osb->max_slots, sizeof(int), GFP_NOFS);
1395 status = ocfs2_super_lock(osb, 1);
1401 status = ocfs2_compute_replay_slots(osb);
1405 /* queue recovery for our own slot */
1406 ocfs2_queue_recovery_completion(osb->journal, osb->slot_num, NULL,
1407 NULL, NULL, ORPHAN_NO_NEED_TRUNCATE);
1409 spin_lock(&osb->osb_lock);
1410 while (rm->rm_used) {
1411 /* It's always safe to remove entry zero, as we won't
1412 * clear it until ocfs2_recover_node() has succeeded. */
1413 node_num = rm->rm_entries[0];
1414 spin_unlock(&osb->osb_lock);
1415 slot_num = ocfs2_node_num_to_slot(osb, node_num);
1416 trace_ocfs2_recovery_thread_node(node_num, slot_num);
1417 if (slot_num == -ENOENT) {
1422 /* It is a bit subtle with quota recovery. We cannot do it
1423 * immediately because we have to obtain cluster locks from
1424 * quota files and we also don't want to just skip it because
1425 * then quota usage would be out of sync until some node takes
1426 * the slot. So we remember which nodes need quota recovery
1427 * and when everything else is done, we recover quotas. */
1428 if (quota_enabled) {
1429 for (i = 0; i < rm_quota_used
1430 && rm_quota[i] != slot_num; i++)
1433 if (i == rm_quota_used)
1434 rm_quota[rm_quota_used++] = slot_num;
1437 status = ocfs2_recover_node(osb, node_num, slot_num);
1440 ocfs2_recovery_map_clear(osb, node_num);
1443 "Error %d recovering node %d on device (%u,%u)!\n",
1445 MAJOR(osb->sb->s_dev), MINOR(osb->sb->s_dev));
1446 mlog(ML_ERROR, "Volume requires unmount.\n");
1449 spin_lock(&osb->osb_lock);
1451 spin_unlock(&osb->osb_lock);
1452 trace_ocfs2_recovery_thread_end(status);
1454 /* Refresh all journal recovery generations from disk */
1455 status = ocfs2_check_journals_nolocks(osb);
1456 status = (status == -EROFS) ? 0 : status;
1460 /* Now it is right time to recover quotas... We have to do this under
1461 * superblock lock so that no one can start using the slot (and crash)
1462 * before we recover it */
1463 if (quota_enabled) {
1464 for (i = 0; i < rm_quota_used; i++) {
1465 qrec = ocfs2_begin_quota_recovery(osb, rm_quota[i]);
1467 status = PTR_ERR(qrec);
1471 ocfs2_queue_recovery_completion(osb->journal,
1474 ORPHAN_NEED_TRUNCATE);
1478 ocfs2_super_unlock(osb, 1);
1480 /* queue recovery for offline slots */
1481 ocfs2_queue_replay_slots(osb, ORPHAN_NEED_TRUNCATE);
1484 mutex_lock(&osb->recovery_lock);
1485 if (!status && !ocfs2_recovery_completed(osb)) {
1486 mutex_unlock(&osb->recovery_lock);
1490 ocfs2_free_replay_slots(osb);
1491 osb->recovery_thread_task = NULL;
1492 mb(); /* sync with ocfs2_recovery_thread_running */
1493 wake_up(&osb->recovery_event);
1495 mutex_unlock(&osb->recovery_lock);
1500 /* no one is callint kthread_stop() for us so the kthread() api
1501 * requires that we call do_exit(). And it isn't exported, but
1502 * complete_and_exit() seems to be a minimal wrapper around it. */
1503 complete_and_exit(NULL, status);
1506 void ocfs2_recovery_thread(struct ocfs2_super *osb, int node_num)
1508 mutex_lock(&osb->recovery_lock);
1510 trace_ocfs2_recovery_thread(node_num, osb->node_num,
1511 osb->disable_recovery, osb->recovery_thread_task,
1512 osb->disable_recovery ?
1513 -1 : ocfs2_recovery_map_set(osb, node_num));
1515 if (osb->disable_recovery)
1518 if (osb->recovery_thread_task)
1521 osb->recovery_thread_task = kthread_run(__ocfs2_recovery_thread, osb,
1522 "ocfs2rec-%s", osb->uuid_str);
1523 if (IS_ERR(osb->recovery_thread_task)) {
1524 mlog_errno((int)PTR_ERR(osb->recovery_thread_task));
1525 osb->recovery_thread_task = NULL;
1529 mutex_unlock(&osb->recovery_lock);
1530 wake_up(&osb->recovery_event);
1533 static int ocfs2_read_journal_inode(struct ocfs2_super *osb,
1535 struct buffer_head **bh,
1536 struct inode **ret_inode)
1538 int status = -EACCES;
1539 struct inode *inode = NULL;
1541 BUG_ON(slot_num >= osb->max_slots);
1543 inode = ocfs2_get_system_file_inode(osb, JOURNAL_SYSTEM_INODE,
1545 if (!inode || is_bad_inode(inode)) {
1549 SET_INODE_JOURNAL(inode);
1551 status = ocfs2_read_inode_block_full(inode, bh, OCFS2_BH_IGNORE_CACHE);
1561 if (status || !ret_inode)
1569 /* Does the actual journal replay and marks the journal inode as
1570 * clean. Will only replay if the journal inode is marked dirty. */
1571 static int ocfs2_replay_journal(struct ocfs2_super *osb,
1578 struct inode *inode = NULL;
1579 struct ocfs2_dinode *fe;
1580 journal_t *journal = NULL;
1581 struct buffer_head *bh = NULL;
1584 status = ocfs2_read_journal_inode(osb, slot_num, &bh, &inode);
1590 fe = (struct ocfs2_dinode *)bh->b_data;
1591 slot_reco_gen = ocfs2_get_recovery_generation(fe);
1596 * As the fs recovery is asynchronous, there is a small chance that
1597 * another node mounted (and recovered) the slot before the recovery
1598 * thread could get the lock. To handle that, we dirty read the journal
1599 * inode for that slot to get the recovery generation. If it is
1600 * different than what we expected, the slot has been recovered.
1601 * If not, it needs recovery.
1603 if (osb->slot_recovery_generations[slot_num] != slot_reco_gen) {
1604 trace_ocfs2_replay_journal_recovered(slot_num,
1605 osb->slot_recovery_generations[slot_num], slot_reco_gen);
1606 osb->slot_recovery_generations[slot_num] = slot_reco_gen;
1611 /* Continue with recovery as the journal has not yet been recovered */
1613 status = ocfs2_inode_lock_full(inode, &bh, 1, OCFS2_META_LOCK_RECOVERY);
1615 trace_ocfs2_replay_journal_lock_err(status);
1616 if (status != -ERESTARTSYS)
1617 mlog(ML_ERROR, "Could not lock journal!\n");
1622 fe = (struct ocfs2_dinode *) bh->b_data;
1624 flags = le32_to_cpu(fe->id1.journal1.ij_flags);
1625 slot_reco_gen = ocfs2_get_recovery_generation(fe);
1627 if (!(flags & OCFS2_JOURNAL_DIRTY_FL)) {
1628 trace_ocfs2_replay_journal_skip(node_num);
1629 /* Refresh recovery generation for the slot */
1630 osb->slot_recovery_generations[slot_num] = slot_reco_gen;
1634 /* we need to run complete recovery for offline orphan slots */
1635 ocfs2_replay_map_set_state(osb, REPLAY_NEEDED);
1637 printk(KERN_NOTICE "ocfs2: Begin replay journal (node %d, slot %d) on "\
1638 "device (%u,%u)\n", node_num, slot_num, MAJOR(osb->sb->s_dev),
1639 MINOR(osb->sb->s_dev));
1641 OCFS2_I(inode)->ip_clusters = le32_to_cpu(fe->i_clusters);
1643 status = ocfs2_force_read_journal(inode);
1649 journal = jbd2_journal_init_inode(inode);
1650 if (journal == NULL) {
1651 mlog(ML_ERROR, "Linux journal layer error\n");
1656 status = jbd2_journal_load(journal);
1661 jbd2_journal_destroy(journal);
1665 ocfs2_clear_journal_error(osb->sb, journal, slot_num);
1667 /* wipe the journal */
1668 jbd2_journal_lock_updates(journal);
1669 status = jbd2_journal_flush(journal, 0);
1670 jbd2_journal_unlock_updates(journal);
1674 /* This will mark the node clean */
1675 flags = le32_to_cpu(fe->id1.journal1.ij_flags);
1676 flags &= ~OCFS2_JOURNAL_DIRTY_FL;
1677 fe->id1.journal1.ij_flags = cpu_to_le32(flags);
1679 /* Increment recovery generation to indicate successful recovery */
1680 ocfs2_bump_recovery_generation(fe);
1681 osb->slot_recovery_generations[slot_num] =
1682 ocfs2_get_recovery_generation(fe);
1684 ocfs2_compute_meta_ecc(osb->sb, bh->b_data, &fe->i_check);
1685 status = ocfs2_write_block(osb, bh, INODE_CACHE(inode));
1692 jbd2_journal_destroy(journal);
1694 printk(KERN_NOTICE "ocfs2: End replay journal (node %d, slot %d) on "\
1695 "device (%u,%u)\n", node_num, slot_num, MAJOR(osb->sb->s_dev),
1696 MINOR(osb->sb->s_dev));
1698 /* drop the lock on this nodes journal */
1700 ocfs2_inode_unlock(inode, 1);
1709 * Do the most important parts of node recovery:
1710 * - Replay it's journal
1711 * - Stamp a clean local allocator file
1712 * - Stamp a clean truncate log
1713 * - Mark the node clean
1715 * If this function completes without error, a node in OCFS2 can be
1716 * said to have been safely recovered. As a result, failure during the
1717 * second part of a nodes recovery process (local alloc recovery) is
1718 * far less concerning.
1720 static int ocfs2_recover_node(struct ocfs2_super *osb,
1721 int node_num, int slot_num)
1724 struct ocfs2_dinode *la_copy = NULL;
1725 struct ocfs2_dinode *tl_copy = NULL;
1727 trace_ocfs2_recover_node(node_num, slot_num, osb->node_num);
1729 /* Should not ever be called to recover ourselves -- in that
1730 * case we should've called ocfs2_journal_load instead. */
1731 BUG_ON(osb->node_num == node_num);
1733 status = ocfs2_replay_journal(osb, node_num, slot_num);
1735 if (status == -EBUSY) {
1736 trace_ocfs2_recover_node_skip(slot_num, node_num);
1744 /* Stamp a clean local alloc file AFTER recovering the journal... */
1745 status = ocfs2_begin_local_alloc_recovery(osb, slot_num, &la_copy);
1751 /* An error from begin_truncate_log_recovery is not
1752 * serious enough to warrant halting the rest of
1754 status = ocfs2_begin_truncate_log_recovery(osb, slot_num, &tl_copy);
1758 /* Likewise, this would be a strange but ultimately not so
1759 * harmful place to get an error... */
1760 status = ocfs2_clear_slot(osb, slot_num);
1764 /* This will kfree the memory pointed to by la_copy and tl_copy */
1765 ocfs2_queue_recovery_completion(osb->journal, slot_num, la_copy,
1766 tl_copy, NULL, ORPHAN_NEED_TRUNCATE);
1774 /* Test node liveness by trylocking his journal. If we get the lock,
1775 * we drop it here. Return 0 if we got the lock, -EAGAIN if node is
1776 * still alive (we couldn't get the lock) and < 0 on error. */
1777 static int ocfs2_trylock_journal(struct ocfs2_super *osb,
1781 struct inode *inode = NULL;
1783 inode = ocfs2_get_system_file_inode(osb, JOURNAL_SYSTEM_INODE,
1785 if (inode == NULL) {
1786 mlog(ML_ERROR, "access error\n");
1790 if (is_bad_inode(inode)) {
1791 mlog(ML_ERROR, "access error (bad inode)\n");
1797 SET_INODE_JOURNAL(inode);
1799 flags = OCFS2_META_LOCK_RECOVERY | OCFS2_META_LOCK_NOQUEUE;
1800 status = ocfs2_inode_lock_full(inode, NULL, 1, flags);
1802 if (status != -EAGAIN)
1807 ocfs2_inode_unlock(inode, 1);
1814 /* Call this underneath ocfs2_super_lock. It also assumes that the
1815 * slot info struct has been updated from disk. */
1816 int ocfs2_mark_dead_nodes(struct ocfs2_super *osb)
1818 unsigned int node_num;
1821 struct buffer_head *bh = NULL;
1822 struct ocfs2_dinode *di;
1824 /* This is called with the super block cluster lock, so we
1825 * know that the slot map can't change underneath us. */
1827 for (i = 0; i < osb->max_slots; i++) {
1828 /* Read journal inode to get the recovery generation */
1829 status = ocfs2_read_journal_inode(osb, i, &bh, NULL);
1834 di = (struct ocfs2_dinode *)bh->b_data;
1835 gen = ocfs2_get_recovery_generation(di);
1839 spin_lock(&osb->osb_lock);
1840 osb->slot_recovery_generations[i] = gen;
1842 trace_ocfs2_mark_dead_nodes(i,
1843 osb->slot_recovery_generations[i]);
1845 if (i == osb->slot_num) {
1846 spin_unlock(&osb->osb_lock);
1850 status = ocfs2_slot_to_node_num_locked(osb, i, &node_num);
1851 if (status == -ENOENT) {
1852 spin_unlock(&osb->osb_lock);
1856 if (__ocfs2_recovery_map_test(osb, node_num)) {
1857 spin_unlock(&osb->osb_lock);
1860 spin_unlock(&osb->osb_lock);
1862 /* Ok, we have a slot occupied by another node which
1863 * is not in the recovery map. We trylock his journal
1864 * file here to test if he's alive. */
1865 status = ocfs2_trylock_journal(osb, i);
1867 /* Since we're called from mount, we know that
1868 * the recovery thread can't race us on
1869 * setting / checking the recovery bits. */
1870 ocfs2_recovery_thread(osb, node_num);
1871 } else if ((status < 0) && (status != -EAGAIN)) {
1883 * Scan timer should get fired every ORPHAN_SCAN_SCHEDULE_TIMEOUT. Add some
1884 * randomness to the timeout to minimize multple nodes firing the timer at the
1887 static inline unsigned long ocfs2_orphan_scan_timeout(void)
1891 get_random_bytes(&time, sizeof(time));
1892 time = ORPHAN_SCAN_SCHEDULE_TIMEOUT + (time % 5000);
1893 return msecs_to_jiffies(time);
1897 * ocfs2_queue_orphan_scan calls ocfs2_queue_recovery_completion for
1898 * every slot, queuing a recovery of the slot on the ocfs2_wq thread. This
1899 * is done to catch any orphans that are left over in orphan directories.
1901 * It scans all slots, even ones that are in use. It does so to handle the
1902 * case described below:
1904 * Node 1 has an inode it was using. The dentry went away due to memory
1905 * pressure. Node 1 closes the inode, but it's on the free list. The node
1906 * has the open lock.
1907 * Node 2 unlinks the inode. It grabs the dentry lock to notify others,
1908 * but node 1 has no dentry and doesn't get the message. It trylocks the
1909 * open lock, sees that another node has a PR, and does nothing.
1910 * Later node 2 runs its orphan dir. It igets the inode, trylocks the
1911 * open lock, sees the PR still, and does nothing.
1912 * Basically, we have to trigger an orphan iput on node 1. The only way
1913 * for this to happen is if node 1 runs node 2's orphan dir.
1915 * ocfs2_queue_orphan_scan gets called every ORPHAN_SCAN_SCHEDULE_TIMEOUT
1916 * seconds. It gets an EX lock on os_lockres and checks sequence number
1917 * stored in LVB. If the sequence number has changed, it means some other
1918 * node has done the scan. This node skips the scan and tracks the
1919 * sequence number. If the sequence number didn't change, it means a scan
1920 * hasn't happened. The node queues a scan and increments the
1921 * sequence number in the LVB.
1923 static void ocfs2_queue_orphan_scan(struct ocfs2_super *osb)
1925 struct ocfs2_orphan_scan *os;
1929 os = &osb->osb_orphan_scan;
1931 if (atomic_read(&os->os_state) == ORPHAN_SCAN_INACTIVE)
1934 trace_ocfs2_queue_orphan_scan_begin(os->os_count, os->os_seqno,
1935 atomic_read(&os->os_state));
1937 status = ocfs2_orphan_scan_lock(osb, &seqno);
1939 if (status != -EAGAIN)
1944 /* Do no queue the tasks if the volume is being umounted */
1945 if (atomic_read(&os->os_state) == ORPHAN_SCAN_INACTIVE)
1948 if (os->os_seqno != seqno) {
1949 os->os_seqno = seqno;
1953 for (i = 0; i < osb->max_slots; i++)
1954 ocfs2_queue_recovery_completion(osb->journal, i, NULL, NULL,
1955 NULL, ORPHAN_NO_NEED_TRUNCATE);
1957 * We queued a recovery on orphan slots, increment the sequence
1958 * number and update LVB so other node will skip the scan for a while
1962 os->os_scantime = ktime_get_seconds();
1964 ocfs2_orphan_scan_unlock(osb, seqno);
1966 trace_ocfs2_queue_orphan_scan_end(os->os_count, os->os_seqno,
1967 atomic_read(&os->os_state));
1971 /* Worker task that gets fired every ORPHAN_SCAN_SCHEDULE_TIMEOUT millsec */
1972 static void ocfs2_orphan_scan_work(struct work_struct *work)
1974 struct ocfs2_orphan_scan *os;
1975 struct ocfs2_super *osb;
1977 os = container_of(work, struct ocfs2_orphan_scan,
1978 os_orphan_scan_work.work);
1981 mutex_lock(&os->os_lock);
1982 ocfs2_queue_orphan_scan(osb);
1983 if (atomic_read(&os->os_state) == ORPHAN_SCAN_ACTIVE)
1984 queue_delayed_work(osb->ocfs2_wq, &os->os_orphan_scan_work,
1985 ocfs2_orphan_scan_timeout());
1986 mutex_unlock(&os->os_lock);
1989 void ocfs2_orphan_scan_stop(struct ocfs2_super *osb)
1991 struct ocfs2_orphan_scan *os;
1993 os = &osb->osb_orphan_scan;
1994 if (atomic_read(&os->os_state) == ORPHAN_SCAN_ACTIVE) {
1995 atomic_set(&os->os_state, ORPHAN_SCAN_INACTIVE);
1996 mutex_lock(&os->os_lock);
1997 cancel_delayed_work(&os->os_orphan_scan_work);
1998 mutex_unlock(&os->os_lock);
2002 void ocfs2_orphan_scan_init(struct ocfs2_super *osb)
2004 struct ocfs2_orphan_scan *os;
2006 os = &osb->osb_orphan_scan;
2010 mutex_init(&os->os_lock);
2011 INIT_DELAYED_WORK(&os->os_orphan_scan_work, ocfs2_orphan_scan_work);
2014 void ocfs2_orphan_scan_start(struct ocfs2_super *osb)
2016 struct ocfs2_orphan_scan *os;
2018 os = &osb->osb_orphan_scan;
2019 os->os_scantime = ktime_get_seconds();
2020 if (ocfs2_is_hard_readonly(osb) || ocfs2_mount_local(osb))
2021 atomic_set(&os->os_state, ORPHAN_SCAN_INACTIVE);
2023 atomic_set(&os->os_state, ORPHAN_SCAN_ACTIVE);
2024 queue_delayed_work(osb->ocfs2_wq, &os->os_orphan_scan_work,
2025 ocfs2_orphan_scan_timeout());
2029 struct ocfs2_orphan_filldir_priv {
2030 struct dir_context ctx;
2032 struct ocfs2_super *osb;
2033 enum ocfs2_orphan_reco_type orphan_reco_type;
2036 static int ocfs2_orphan_filldir(struct dir_context *ctx, const char *name,
2037 int name_len, loff_t pos, u64 ino,
2040 struct ocfs2_orphan_filldir_priv *p =
2041 container_of(ctx, struct ocfs2_orphan_filldir_priv, ctx);
2044 if (name_len == 1 && !strncmp(".", name, 1))
2046 if (name_len == 2 && !strncmp("..", name, 2))
2049 /* do not include dio entry in case of orphan scan */
2050 if ((p->orphan_reco_type == ORPHAN_NO_NEED_TRUNCATE) &&
2051 (!strncmp(name, OCFS2_DIO_ORPHAN_PREFIX,
2052 OCFS2_DIO_ORPHAN_PREFIX_LEN)))
2055 /* Skip bad inodes so that recovery can continue */
2056 iter = ocfs2_iget(p->osb, ino,
2057 OCFS2_FI_FLAG_ORPHAN_RECOVERY, 0);
2061 if (!strncmp(name, OCFS2_DIO_ORPHAN_PREFIX,
2062 OCFS2_DIO_ORPHAN_PREFIX_LEN))
2063 OCFS2_I(iter)->ip_flags |= OCFS2_INODE_DIO_ORPHAN_ENTRY;
2065 /* Skip inodes which are already added to recover list, since dio may
2066 * happen concurrently with unlink/rename */
2067 if (OCFS2_I(iter)->ip_next_orphan) {
2072 trace_ocfs2_orphan_filldir((unsigned long long)OCFS2_I(iter)->ip_blkno);
2073 /* No locking is required for the next_orphan queue as there
2074 * is only ever a single process doing orphan recovery. */
2075 OCFS2_I(iter)->ip_next_orphan = p->head;
2081 static int ocfs2_queue_orphans(struct ocfs2_super *osb,
2083 struct inode **head,
2084 enum ocfs2_orphan_reco_type orphan_reco_type)
2087 struct inode *orphan_dir_inode = NULL;
2088 struct ocfs2_orphan_filldir_priv priv = {
2089 .ctx.actor = ocfs2_orphan_filldir,
2092 .orphan_reco_type = orphan_reco_type
2095 orphan_dir_inode = ocfs2_get_system_file_inode(osb,
2096 ORPHAN_DIR_SYSTEM_INODE,
2098 if (!orphan_dir_inode) {
2104 inode_lock(orphan_dir_inode);
2105 status = ocfs2_inode_lock(orphan_dir_inode, NULL, 0);
2111 status = ocfs2_dir_foreach(orphan_dir_inode, &priv.ctx);
2120 ocfs2_inode_unlock(orphan_dir_inode, 0);
2122 inode_unlock(orphan_dir_inode);
2123 iput(orphan_dir_inode);
2127 static int ocfs2_orphan_recovery_can_continue(struct ocfs2_super *osb,
2132 spin_lock(&osb->osb_lock);
2133 ret = !osb->osb_orphan_wipes[slot];
2134 spin_unlock(&osb->osb_lock);
2138 static void ocfs2_mark_recovering_orphan_dir(struct ocfs2_super *osb,
2141 spin_lock(&osb->osb_lock);
2142 /* Mark ourselves such that new processes in delete_inode()
2143 * know to quit early. */
2144 ocfs2_node_map_set_bit(osb, &osb->osb_recovering_orphan_dirs, slot);
2145 while (osb->osb_orphan_wipes[slot]) {
2146 /* If any processes are already in the middle of an
2147 * orphan wipe on this dir, then we need to wait for
2149 spin_unlock(&osb->osb_lock);
2150 wait_event_interruptible(osb->osb_wipe_event,
2151 ocfs2_orphan_recovery_can_continue(osb, slot));
2152 spin_lock(&osb->osb_lock);
2154 spin_unlock(&osb->osb_lock);
2157 static void ocfs2_clear_recovering_orphan_dir(struct ocfs2_super *osb,
2160 ocfs2_node_map_clear_bit(osb, &osb->osb_recovering_orphan_dirs, slot);
2164 * Orphan recovery. Each mounted node has it's own orphan dir which we
2165 * must run during recovery. Our strategy here is to build a list of
2166 * the inodes in the orphan dir and iget/iput them. The VFS does
2167 * (most) of the rest of the work.
2169 * Orphan recovery can happen at any time, not just mount so we have a
2170 * couple of extra considerations.
2172 * - We grab as many inodes as we can under the orphan dir lock -
2173 * doing iget() outside the orphan dir risks getting a reference on
2175 * - We must be sure not to deadlock with other processes on the
2176 * system wanting to run delete_inode(). This can happen when they go
2177 * to lock the orphan dir and the orphan recovery process attempts to
2178 * iget() inside the orphan dir lock. This can be avoided by
2179 * advertising our state to ocfs2_delete_inode().
2181 static int ocfs2_recover_orphans(struct ocfs2_super *osb,
2183 enum ocfs2_orphan_reco_type orphan_reco_type)
2186 struct inode *inode = NULL;
2188 struct ocfs2_inode_info *oi;
2189 struct buffer_head *di_bh = NULL;
2190 struct ocfs2_dinode *di = NULL;
2192 trace_ocfs2_recover_orphans(slot);
2194 ocfs2_mark_recovering_orphan_dir(osb, slot);
2195 ret = ocfs2_queue_orphans(osb, slot, &inode, orphan_reco_type);
2196 ocfs2_clear_recovering_orphan_dir(osb, slot);
2198 /* Error here should be noted, but we want to continue with as
2199 * many queued inodes as we've got. */
2204 oi = OCFS2_I(inode);
2205 trace_ocfs2_recover_orphans_iput(
2206 (unsigned long long)oi->ip_blkno);
2208 iter = oi->ip_next_orphan;
2209 oi->ip_next_orphan = NULL;
2211 if (oi->ip_flags & OCFS2_INODE_DIO_ORPHAN_ENTRY) {
2213 ret = ocfs2_rw_lock(inode, 1);
2219 * We need to take and drop the inode lock to
2220 * force read inode from disk.
2222 ret = ocfs2_inode_lock(inode, &di_bh, 1);
2228 di = (struct ocfs2_dinode *)di_bh->b_data;
2230 if (di->i_flags & cpu_to_le32(OCFS2_DIO_ORPHANED_FL)) {
2231 ret = ocfs2_truncate_file(inode, di_bh,
2232 i_size_read(inode));
2239 ret = ocfs2_del_inode_from_orphan(osb, inode,
2245 ocfs2_inode_unlock(inode, 1);
2249 ocfs2_rw_unlock(inode, 1);
2251 inode_unlock(inode);
2253 /* clear dio flag in ocfs2_inode_info */
2254 oi->ip_flags &= ~OCFS2_INODE_DIO_ORPHAN_ENTRY;
2256 spin_lock(&oi->ip_lock);
2257 /* Set the proper information to get us going into
2258 * ocfs2_delete_inode. */
2259 oi->ip_flags |= OCFS2_INODE_MAYBE_ORPHANED;
2260 spin_unlock(&oi->ip_lock);
2270 static int __ocfs2_wait_on_mount(struct ocfs2_super *osb, int quota)
2272 /* This check is good because ocfs2 will wait on our recovery
2273 * thread before changing it to something other than MOUNTED
2275 wait_event(osb->osb_mount_event,
2276 (!quota && atomic_read(&osb->vol_state) == VOLUME_MOUNTED) ||
2277 atomic_read(&osb->vol_state) == VOLUME_MOUNTED_QUOTAS ||
2278 atomic_read(&osb->vol_state) == VOLUME_DISABLED);
2280 /* If there's an error on mount, then we may never get to the
2281 * MOUNTED flag, but this is set right before
2282 * dismount_volume() so we can trust it. */
2283 if (atomic_read(&osb->vol_state) == VOLUME_DISABLED) {
2284 trace_ocfs2_wait_on_mount(VOLUME_DISABLED);
2285 mlog(0, "mount error, exiting!\n");
2292 static int ocfs2_commit_thread(void *arg)
2295 struct ocfs2_super *osb = arg;
2296 struct ocfs2_journal *journal = osb->journal;
2298 /* we can trust j_num_trans here because _should_stop() is only set in
2299 * shutdown and nobody other than ourselves should be able to start
2300 * transactions. committing on shutdown might take a few iterations
2301 * as final transactions put deleted inodes on the list */
2302 while (!(kthread_should_stop() &&
2303 atomic_read(&journal->j_num_trans) == 0)) {
2305 wait_event_interruptible(osb->checkpoint_event,
2306 atomic_read(&journal->j_num_trans)
2307 || kthread_should_stop());
2309 status = ocfs2_commit_cache(osb);
2311 static unsigned long abort_warn_time;
2313 /* Warn about this once per minute */
2314 if (printk_timed_ratelimit(&abort_warn_time, 60*HZ))
2315 mlog(ML_ERROR, "status = %d, journal is "
2316 "already aborted.\n", status);
2318 * After ocfs2_commit_cache() fails, j_num_trans has a
2319 * non-zero value. Sleep here to avoid a busy-wait
2322 msleep_interruptible(1000);
2325 if (kthread_should_stop() && atomic_read(&journal->j_num_trans)){
2327 "commit_thread: %u transactions pending on "
2329 atomic_read(&journal->j_num_trans));
2336 /* Reads all the journal inodes without taking any cluster locks. Used
2337 * for hard readonly access to determine whether any journal requires
2338 * recovery. Also used to refresh the recovery generation numbers after
2339 * a journal has been recovered by another node.
2341 int ocfs2_check_journals_nolocks(struct ocfs2_super *osb)
2345 struct buffer_head *di_bh = NULL;
2346 struct ocfs2_dinode *di;
2347 int journal_dirty = 0;
2349 for(slot = 0; slot < osb->max_slots; slot++) {
2350 ret = ocfs2_read_journal_inode(osb, slot, &di_bh, NULL);
2356 di = (struct ocfs2_dinode *) di_bh->b_data;
2358 osb->slot_recovery_generations[slot] =
2359 ocfs2_get_recovery_generation(di);
2361 if (le32_to_cpu(di->id1.journal1.ij_flags) &
2362 OCFS2_JOURNAL_DIRTY_FL)