1 /* -*- mode: c; c-basic-offset: 8; -*-
2 * vim: noexpandtab sw=8 ts=8 sts=0:
6 * Defines functions of journalling api
8 * Copyright (C) 2003, 2004 Oracle. All rights reserved.
10 * This program is free software; you can redistribute it and/or
11 * modify it under the terms of the GNU General Public
12 * License as published by the Free Software Foundation; either
13 * version 2 of the License, or (at your option) any later version.
15 * This program is distributed in the hope that it will be useful,
16 * but WITHOUT ANY WARRANTY; without even the implied warranty of
17 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
18 * General Public License for more details.
20 * You should have received a copy of the GNU General Public
21 * License along with this program; if not, write to the
22 * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
23 * Boston, MA 021110-1307, USA.
27 #include <linux/types.h>
28 #include <linux/slab.h>
29 #include <linux/highmem.h>
30 #include <linux/kthread.h>
31 #include <linux/time.h>
32 #include <linux/random.h>
34 #include <cluster/masklog.h>
39 #include "blockcheck.h"
42 #include "extent_map.h"
43 #include "heartbeat.h"
46 #include "localalloc.h"
53 #include "buffer_head_io.h"
54 #include "ocfs2_trace.h"
56 DEFINE_SPINLOCK(trans_inc_lock);
58 #define ORPHAN_SCAN_SCHEDULE_TIMEOUT 300000
60 static int ocfs2_force_read_journal(struct inode *inode);
61 static int ocfs2_recover_node(struct ocfs2_super *osb,
62 int node_num, int slot_num);
63 static int __ocfs2_recovery_thread(void *arg);
64 static int ocfs2_commit_cache(struct ocfs2_super *osb);
65 static int __ocfs2_wait_on_mount(struct ocfs2_super *osb, int quota);
66 static int ocfs2_journal_toggle_dirty(struct ocfs2_super *osb,
67 int dirty, int replayed);
68 static int ocfs2_trylock_journal(struct ocfs2_super *osb,
70 static int ocfs2_recover_orphans(struct ocfs2_super *osb,
72 static int ocfs2_commit_thread(void *arg);
73 static void ocfs2_queue_recovery_completion(struct ocfs2_journal *journal,
75 struct ocfs2_dinode *la_dinode,
76 struct ocfs2_dinode *tl_dinode,
77 struct ocfs2_quota_recovery *qrec);
79 static inline int ocfs2_wait_on_mount(struct ocfs2_super *osb)
81 return __ocfs2_wait_on_mount(osb, 0);
84 static inline int ocfs2_wait_on_quotas(struct ocfs2_super *osb)
86 return __ocfs2_wait_on_mount(osb, 1);
90 * This replay_map is to track online/offline slots, so we could recover
91 * offline slots during recovery and mount
94 enum ocfs2_replay_state {
95 REPLAY_UNNEEDED = 0, /* Replay is not needed, so ignore this map */
96 REPLAY_NEEDED, /* Replay slots marked in rm_replay_slots */
97 REPLAY_DONE /* Replay was already queued */
100 struct ocfs2_replay_map {
101 unsigned int rm_slots;
102 enum ocfs2_replay_state rm_state;
103 unsigned char rm_replay_slots[0];
106 void ocfs2_replay_map_set_state(struct ocfs2_super *osb, int state)
108 if (!osb->replay_map)
111 /* If we've already queued the replay, we don't have any more to do */
112 if (osb->replay_map->rm_state == REPLAY_DONE)
115 osb->replay_map->rm_state = state;
118 int ocfs2_compute_replay_slots(struct ocfs2_super *osb)
120 struct ocfs2_replay_map *replay_map;
123 /* If replay map is already set, we don't do it again */
127 replay_map = kzalloc(sizeof(struct ocfs2_replay_map) +
128 (osb->max_slots * sizeof(char)), GFP_KERNEL);
135 spin_lock(&osb->osb_lock);
137 replay_map->rm_slots = osb->max_slots;
138 replay_map->rm_state = REPLAY_UNNEEDED;
140 /* set rm_replay_slots for offline slot(s) */
141 for (i = 0; i < replay_map->rm_slots; i++) {
142 if (ocfs2_slot_to_node_num_locked(osb, i, &node_num) == -ENOENT)
143 replay_map->rm_replay_slots[i] = 1;
146 osb->replay_map = replay_map;
147 spin_unlock(&osb->osb_lock);
151 void ocfs2_queue_replay_slots(struct ocfs2_super *osb)
153 struct ocfs2_replay_map *replay_map = osb->replay_map;
159 if (replay_map->rm_state != REPLAY_NEEDED)
162 for (i = 0; i < replay_map->rm_slots; i++)
163 if (replay_map->rm_replay_slots[i])
164 ocfs2_queue_recovery_completion(osb->journal, i, NULL,
166 replay_map->rm_state = REPLAY_DONE;
169 void ocfs2_free_replay_slots(struct ocfs2_super *osb)
171 struct ocfs2_replay_map *replay_map = osb->replay_map;
173 if (!osb->replay_map)
177 osb->replay_map = NULL;
180 int ocfs2_recovery_init(struct ocfs2_super *osb)
182 struct ocfs2_recovery_map *rm;
184 mutex_init(&osb->recovery_lock);
185 osb->disable_recovery = 0;
186 osb->recovery_thread_task = NULL;
187 init_waitqueue_head(&osb->recovery_event);
189 rm = kzalloc(sizeof(struct ocfs2_recovery_map) +
190 osb->max_slots * sizeof(unsigned int),
197 rm->rm_entries = (unsigned int *)((char *)rm +
198 sizeof(struct ocfs2_recovery_map));
199 osb->recovery_map = rm;
204 /* we can't grab the goofy sem lock from inside wait_event, so we use
205 * memory barriers to make sure that we'll see the null task before
207 static int ocfs2_recovery_thread_running(struct ocfs2_super *osb)
210 return osb->recovery_thread_task != NULL;
213 void ocfs2_recovery_exit(struct ocfs2_super *osb)
215 struct ocfs2_recovery_map *rm;
217 /* disable any new recovery threads and wait for any currently
218 * running ones to exit. Do this before setting the vol_state. */
219 mutex_lock(&osb->recovery_lock);
220 osb->disable_recovery = 1;
221 mutex_unlock(&osb->recovery_lock);
222 wait_event(osb->recovery_event, !ocfs2_recovery_thread_running(osb));
224 /* At this point, we know that no more recovery threads can be
225 * launched, so wait for any recovery completion work to
227 flush_workqueue(ocfs2_wq);
230 * Now that recovery is shut down, and the osb is about to be
231 * freed, the osb_lock is not taken here.
233 rm = osb->recovery_map;
234 /* XXX: Should we bug if there are dirty entries? */
239 static int __ocfs2_recovery_map_test(struct ocfs2_super *osb,
240 unsigned int node_num)
243 struct ocfs2_recovery_map *rm = osb->recovery_map;
245 assert_spin_locked(&osb->osb_lock);
247 for (i = 0; i < rm->rm_used; i++) {
248 if (rm->rm_entries[i] == node_num)
255 /* Behaves like test-and-set. Returns the previous value */
256 static int ocfs2_recovery_map_set(struct ocfs2_super *osb,
257 unsigned int node_num)
259 struct ocfs2_recovery_map *rm = osb->recovery_map;
261 spin_lock(&osb->osb_lock);
262 if (__ocfs2_recovery_map_test(osb, node_num)) {
263 spin_unlock(&osb->osb_lock);
267 /* XXX: Can this be exploited? Not from o2dlm... */
268 BUG_ON(rm->rm_used >= osb->max_slots);
270 rm->rm_entries[rm->rm_used] = node_num;
272 spin_unlock(&osb->osb_lock);
277 static void ocfs2_recovery_map_clear(struct ocfs2_super *osb,
278 unsigned int node_num)
281 struct ocfs2_recovery_map *rm = osb->recovery_map;
283 spin_lock(&osb->osb_lock);
285 for (i = 0; i < rm->rm_used; i++) {
286 if (rm->rm_entries[i] == node_num)
290 if (i < rm->rm_used) {
291 /* XXX: be careful with the pointer math */
292 memmove(&(rm->rm_entries[i]), &(rm->rm_entries[i + 1]),
293 (rm->rm_used - i - 1) * sizeof(unsigned int));
297 spin_unlock(&osb->osb_lock);
300 static int ocfs2_commit_cache(struct ocfs2_super *osb)
303 unsigned int flushed;
304 struct ocfs2_journal *journal = NULL;
306 journal = osb->journal;
308 /* Flush all pending commits and checkpoint the journal. */
309 down_write(&journal->j_trans_barrier);
311 flushed = atomic_read(&journal->j_num_trans);
312 trace_ocfs2_commit_cache_begin(flushed);
314 up_write(&journal->j_trans_barrier);
318 jbd2_journal_lock_updates(journal->j_journal);
319 status = jbd2_journal_flush(journal->j_journal);
320 jbd2_journal_unlock_updates(journal->j_journal);
322 up_write(&journal->j_trans_barrier);
327 ocfs2_inc_trans_id(journal);
329 flushed = atomic_read(&journal->j_num_trans);
330 atomic_set(&journal->j_num_trans, 0);
331 up_write(&journal->j_trans_barrier);
333 trace_ocfs2_commit_cache_end(journal->j_trans_id, flushed);
335 ocfs2_wake_downconvert_thread(osb);
336 wake_up(&journal->j_checkpointed);
341 handle_t *ocfs2_start_trans(struct ocfs2_super *osb, int max_buffs)
343 journal_t *journal = osb->journal->j_journal;
346 BUG_ON(!osb || !osb->journal->j_journal);
348 if (ocfs2_is_hard_readonly(osb))
349 return ERR_PTR(-EROFS);
351 BUG_ON(osb->journal->j_state == OCFS2_JOURNAL_FREE);
352 BUG_ON(max_buffs <= 0);
354 /* Nested transaction? Just return the handle... */
355 if (journal_current_handle())
356 return jbd2_journal_start(journal, max_buffs);
358 sb_start_intwrite(osb->sb);
360 down_read(&osb->journal->j_trans_barrier);
362 handle = jbd2_journal_start(journal, max_buffs);
363 if (IS_ERR(handle)) {
364 up_read(&osb->journal->j_trans_barrier);
365 sb_end_intwrite(osb->sb);
367 mlog_errno(PTR_ERR(handle));
369 if (is_journal_aborted(journal)) {
370 ocfs2_abort(osb->sb, "Detected aborted journal");
371 handle = ERR_PTR(-EROFS);
374 if (!ocfs2_mount_local(osb))
375 atomic_inc(&(osb->journal->j_num_trans));
381 int ocfs2_commit_trans(struct ocfs2_super *osb,
385 struct ocfs2_journal *journal = osb->journal;
389 nested = handle->h_ref > 1;
390 ret = jbd2_journal_stop(handle);
395 up_read(&journal->j_trans_barrier);
396 sb_end_intwrite(osb->sb);
403 * 'nblocks' is what you want to add to the current transaction.
405 * This might call jbd2_journal_restart() which will commit dirty buffers
406 * and then restart the transaction. Before calling
407 * ocfs2_extend_trans(), any changed blocks should have been
408 * dirtied. After calling it, all blocks which need to be changed must
409 * go through another set of journal_access/journal_dirty calls.
411 * WARNING: This will not release any semaphores or disk locks taken
412 * during the transaction, so make sure they were taken *before*
413 * start_trans or we'll have ordering deadlocks.
415 * WARNING2: Note that we do *not* drop j_trans_barrier here. This is
416 * good because transaction ids haven't yet been recorded on the
417 * cluster locks associated with this handle.
419 int ocfs2_extend_trans(handle_t *handle, int nblocks)
421 int status, old_nblocks;
429 old_nblocks = handle->h_buffer_credits;
431 trace_ocfs2_extend_trans(old_nblocks, nblocks);
433 #ifdef CONFIG_OCFS2_DEBUG_FS
436 status = jbd2_journal_extend(handle, nblocks);
444 trace_ocfs2_extend_trans_restart(old_nblocks + nblocks);
445 status = jbd2_journal_restart(handle,
446 old_nblocks + nblocks);
458 struct ocfs2_triggers {
459 struct jbd2_buffer_trigger_type ot_triggers;
463 static inline struct ocfs2_triggers *to_ocfs2_trigger(struct jbd2_buffer_trigger_type *triggers)
465 return container_of(triggers, struct ocfs2_triggers, ot_triggers);
468 static void ocfs2_frozen_trigger(struct jbd2_buffer_trigger_type *triggers,
469 struct buffer_head *bh,
470 void *data, size_t size)
472 struct ocfs2_triggers *ot = to_ocfs2_trigger(triggers);
475 * We aren't guaranteed to have the superblock here, so we
476 * must unconditionally compute the ecc data.
477 * __ocfs2_journal_access() will only set the triggers if
478 * metaecc is enabled.
480 ocfs2_block_check_compute(data, size, data + ot->ot_offset);
484 * Quota blocks have their own trigger because the struct ocfs2_block_check
485 * offset depends on the blocksize.
487 static void ocfs2_dq_frozen_trigger(struct jbd2_buffer_trigger_type *triggers,
488 struct buffer_head *bh,
489 void *data, size_t size)
491 struct ocfs2_disk_dqtrailer *dqt =
492 ocfs2_block_dqtrailer(size, data);
495 * We aren't guaranteed to have the superblock here, so we
496 * must unconditionally compute the ecc data.
497 * __ocfs2_journal_access() will only set the triggers if
498 * metaecc is enabled.
500 ocfs2_block_check_compute(data, size, &dqt->dq_check);
504 * Directory blocks also have their own trigger because the
505 * struct ocfs2_block_check offset depends on the blocksize.
507 static void ocfs2_db_frozen_trigger(struct jbd2_buffer_trigger_type *triggers,
508 struct buffer_head *bh,
509 void *data, size_t size)
511 struct ocfs2_dir_block_trailer *trailer =
512 ocfs2_dir_trailer_from_size(size, data);
515 * We aren't guaranteed to have the superblock here, so we
516 * must unconditionally compute the ecc data.
517 * __ocfs2_journal_access() will only set the triggers if
518 * metaecc is enabled.
520 ocfs2_block_check_compute(data, size, &trailer->db_check);
523 static void ocfs2_abort_trigger(struct jbd2_buffer_trigger_type *triggers,
524 struct buffer_head *bh)
527 "ocfs2_abort_trigger called by JBD2. bh = 0x%lx, "
528 "bh->b_blocknr = %llu\n",
530 (unsigned long long)bh->b_blocknr);
532 /* We aren't guaranteed to have the superblock here - but if we
533 * don't, it'll just crash. */
534 ocfs2_error(bh->b_assoc_map->host->i_sb,
535 "JBD2 has aborted our journal, ocfs2 cannot continue\n");
538 static struct ocfs2_triggers di_triggers = {
540 .t_frozen = ocfs2_frozen_trigger,
541 .t_abort = ocfs2_abort_trigger,
543 .ot_offset = offsetof(struct ocfs2_dinode, i_check),
546 static struct ocfs2_triggers eb_triggers = {
548 .t_frozen = ocfs2_frozen_trigger,
549 .t_abort = ocfs2_abort_trigger,
551 .ot_offset = offsetof(struct ocfs2_extent_block, h_check),
554 static struct ocfs2_triggers rb_triggers = {
556 .t_frozen = ocfs2_frozen_trigger,
557 .t_abort = ocfs2_abort_trigger,
559 .ot_offset = offsetof(struct ocfs2_refcount_block, rf_check),
562 static struct ocfs2_triggers gd_triggers = {
564 .t_frozen = ocfs2_frozen_trigger,
565 .t_abort = ocfs2_abort_trigger,
567 .ot_offset = offsetof(struct ocfs2_group_desc, bg_check),
570 static struct ocfs2_triggers db_triggers = {
572 .t_frozen = ocfs2_db_frozen_trigger,
573 .t_abort = ocfs2_abort_trigger,
577 static struct ocfs2_triggers xb_triggers = {
579 .t_frozen = ocfs2_frozen_trigger,
580 .t_abort = ocfs2_abort_trigger,
582 .ot_offset = offsetof(struct ocfs2_xattr_block, xb_check),
585 static struct ocfs2_triggers dq_triggers = {
587 .t_frozen = ocfs2_dq_frozen_trigger,
588 .t_abort = ocfs2_abort_trigger,
592 static struct ocfs2_triggers dr_triggers = {
594 .t_frozen = ocfs2_frozen_trigger,
595 .t_abort = ocfs2_abort_trigger,
597 .ot_offset = offsetof(struct ocfs2_dx_root_block, dr_check),
600 static struct ocfs2_triggers dl_triggers = {
602 .t_frozen = ocfs2_frozen_trigger,
603 .t_abort = ocfs2_abort_trigger,
605 .ot_offset = offsetof(struct ocfs2_dx_leaf, dl_check),
608 static int __ocfs2_journal_access(handle_t *handle,
609 struct ocfs2_caching_info *ci,
610 struct buffer_head *bh,
611 struct ocfs2_triggers *triggers,
615 struct ocfs2_super *osb =
616 OCFS2_SB(ocfs2_metadata_cache_get_super(ci));
618 BUG_ON(!ci || !ci->ci_ops);
622 trace_ocfs2_journal_access(
623 (unsigned long long)ocfs2_metadata_cache_owner(ci),
624 (unsigned long long)bh->b_blocknr, type, bh->b_size);
626 /* we can safely remove this assertion after testing. */
627 if (!buffer_uptodate(bh)) {
628 mlog(ML_ERROR, "giving me a buffer that's not uptodate!\n");
629 mlog(ML_ERROR, "b_blocknr=%llu\n",
630 (unsigned long long)bh->b_blocknr);
634 /* Set the current transaction information on the ci so
635 * that the locking code knows whether it can drop it's locks
636 * on this ci or not. We're protected from the commit
637 * thread updating the current transaction id until
638 * ocfs2_commit_trans() because ocfs2_start_trans() took
639 * j_trans_barrier for us. */
640 ocfs2_set_ci_lock_trans(osb->journal, ci);
642 ocfs2_metadata_cache_io_lock(ci);
644 case OCFS2_JOURNAL_ACCESS_CREATE:
645 case OCFS2_JOURNAL_ACCESS_WRITE:
646 status = jbd2_journal_get_write_access(handle, bh);
649 case OCFS2_JOURNAL_ACCESS_UNDO:
650 status = jbd2_journal_get_undo_access(handle, bh);
655 mlog(ML_ERROR, "Unknown access type!\n");
657 if (!status && ocfs2_meta_ecc(osb) && triggers)
658 jbd2_journal_set_triggers(bh, &triggers->ot_triggers);
659 ocfs2_metadata_cache_io_unlock(ci);
662 mlog(ML_ERROR, "Error %d getting %d access to buffer!\n",
668 int ocfs2_journal_access_di(handle_t *handle, struct ocfs2_caching_info *ci,
669 struct buffer_head *bh, int type)
671 return __ocfs2_journal_access(handle, ci, bh, &di_triggers, type);
674 int ocfs2_journal_access_eb(handle_t *handle, struct ocfs2_caching_info *ci,
675 struct buffer_head *bh, int type)
677 return __ocfs2_journal_access(handle, ci, bh, &eb_triggers, type);
680 int ocfs2_journal_access_rb(handle_t *handle, struct ocfs2_caching_info *ci,
681 struct buffer_head *bh, int type)
683 return __ocfs2_journal_access(handle, ci, bh, &rb_triggers,
687 int ocfs2_journal_access_gd(handle_t *handle, struct ocfs2_caching_info *ci,
688 struct buffer_head *bh, int type)
690 return __ocfs2_journal_access(handle, ci, bh, &gd_triggers, type);
693 int ocfs2_journal_access_db(handle_t *handle, struct ocfs2_caching_info *ci,
694 struct buffer_head *bh, int type)
696 return __ocfs2_journal_access(handle, ci, bh, &db_triggers, type);
699 int ocfs2_journal_access_xb(handle_t *handle, struct ocfs2_caching_info *ci,
700 struct buffer_head *bh, int type)
702 return __ocfs2_journal_access(handle, ci, bh, &xb_triggers, type);
705 int ocfs2_journal_access_dq(handle_t *handle, struct ocfs2_caching_info *ci,
706 struct buffer_head *bh, int type)
708 return __ocfs2_journal_access(handle, ci, bh, &dq_triggers, type);
711 int ocfs2_journal_access_dr(handle_t *handle, struct ocfs2_caching_info *ci,
712 struct buffer_head *bh, int type)
714 return __ocfs2_journal_access(handle, ci, bh, &dr_triggers, type);
717 int ocfs2_journal_access_dl(handle_t *handle, struct ocfs2_caching_info *ci,
718 struct buffer_head *bh, int type)
720 return __ocfs2_journal_access(handle, ci, bh, &dl_triggers, type);
723 int ocfs2_journal_access(handle_t *handle, struct ocfs2_caching_info *ci,
724 struct buffer_head *bh, int type)
726 return __ocfs2_journal_access(handle, ci, bh, NULL, type);
729 void ocfs2_journal_dirty(handle_t *handle, struct buffer_head *bh)
733 trace_ocfs2_journal_dirty((unsigned long long)bh->b_blocknr);
735 status = jbd2_journal_dirty_metadata(handle, bh);
739 #define OCFS2_DEFAULT_COMMIT_INTERVAL (HZ * JBD2_DEFAULT_MAX_COMMIT_AGE)
741 void ocfs2_set_journal_params(struct ocfs2_super *osb)
743 journal_t *journal = osb->journal->j_journal;
744 unsigned long commit_interval = OCFS2_DEFAULT_COMMIT_INTERVAL;
746 if (osb->osb_commit_interval)
747 commit_interval = osb->osb_commit_interval;
749 write_lock(&journal->j_state_lock);
750 journal->j_commit_interval = commit_interval;
751 if (osb->s_mount_opt & OCFS2_MOUNT_BARRIER)
752 journal->j_flags |= JBD2_BARRIER;
754 journal->j_flags &= ~JBD2_BARRIER;
755 write_unlock(&journal->j_state_lock);
758 int ocfs2_journal_init(struct ocfs2_journal *journal, int *dirty)
761 struct inode *inode = NULL; /* the journal inode */
762 journal_t *j_journal = NULL;
763 struct ocfs2_dinode *di = NULL;
764 struct buffer_head *bh = NULL;
765 struct ocfs2_super *osb;
770 osb = journal->j_osb;
772 /* already have the inode for our journal */
773 inode = ocfs2_get_system_file_inode(osb, JOURNAL_SYSTEM_INODE,
780 if (is_bad_inode(inode)) {
781 mlog(ML_ERROR, "access error (bad inode)\n");
788 SET_INODE_JOURNAL(inode);
789 OCFS2_I(inode)->ip_open_count++;
791 /* Skip recovery waits here - journal inode metadata never
792 * changes in a live cluster so it can be considered an
793 * exception to the rule. */
794 status = ocfs2_inode_lock_full(inode, &bh, 1, OCFS2_META_LOCK_RECOVERY);
796 if (status != -ERESTARTSYS)
797 mlog(ML_ERROR, "Could not get lock on journal!\n");
802 di = (struct ocfs2_dinode *)bh->b_data;
804 if (inode->i_size < OCFS2_MIN_JOURNAL_SIZE) {
805 mlog(ML_ERROR, "Journal file size (%lld) is too small!\n",
811 trace_ocfs2_journal_init(inode->i_size,
812 (unsigned long long)inode->i_blocks,
813 OCFS2_I(inode)->ip_clusters);
815 /* call the kernels journal init function now */
816 j_journal = jbd2_journal_init_inode(inode);
817 if (j_journal == NULL) {
818 mlog(ML_ERROR, "Linux journal layer error\n");
823 trace_ocfs2_journal_init_maxlen(j_journal->j_maxlen);
825 *dirty = (le32_to_cpu(di->id1.journal1.ij_flags) &
826 OCFS2_JOURNAL_DIRTY_FL);
828 journal->j_journal = j_journal;
829 journal->j_inode = inode;
832 ocfs2_set_journal_params(osb);
834 journal->j_state = OCFS2_JOURNAL_LOADED;
840 ocfs2_inode_unlock(inode, 1);
843 OCFS2_I(inode)->ip_open_count--;
851 static void ocfs2_bump_recovery_generation(struct ocfs2_dinode *di)
853 le32_add_cpu(&(di->id1.journal1.ij_recovery_generation), 1);
856 static u32 ocfs2_get_recovery_generation(struct ocfs2_dinode *di)
858 return le32_to_cpu(di->id1.journal1.ij_recovery_generation);
861 static int ocfs2_journal_toggle_dirty(struct ocfs2_super *osb,
862 int dirty, int replayed)
866 struct ocfs2_journal *journal = osb->journal;
867 struct buffer_head *bh = journal->j_bh;
868 struct ocfs2_dinode *fe;
870 fe = (struct ocfs2_dinode *)bh->b_data;
872 /* The journal bh on the osb always comes from ocfs2_journal_init()
873 * and was validated there inside ocfs2_inode_lock_full(). It's a
874 * code bug if we mess it up. */
875 BUG_ON(!OCFS2_IS_VALID_DINODE(fe));
877 flags = le32_to_cpu(fe->id1.journal1.ij_flags);
879 flags |= OCFS2_JOURNAL_DIRTY_FL;
881 flags &= ~OCFS2_JOURNAL_DIRTY_FL;
882 fe->id1.journal1.ij_flags = cpu_to_le32(flags);
885 ocfs2_bump_recovery_generation(fe);
887 ocfs2_compute_meta_ecc(osb->sb, bh->b_data, &fe->i_check);
888 status = ocfs2_write_block(osb, bh, INODE_CACHE(journal->j_inode));
896 * If the journal has been kmalloc'd it needs to be freed after this
899 void ocfs2_journal_shutdown(struct ocfs2_super *osb)
901 struct ocfs2_journal *journal = NULL;
903 struct inode *inode = NULL;
904 int num_running_trans = 0;
908 journal = osb->journal;
912 inode = journal->j_inode;
914 if (journal->j_state != OCFS2_JOURNAL_LOADED)
917 /* need to inc inode use count - jbd2_journal_destroy will iput. */
921 num_running_trans = atomic_read(&(osb->journal->j_num_trans));
922 trace_ocfs2_journal_shutdown(num_running_trans);
924 /* Do a commit_cache here. It will flush our journal, *and*
925 * release any locks that are still held.
926 * set the SHUTDOWN flag and release the trans lock.
927 * the commit thread will take the trans lock for us below. */
928 journal->j_state = OCFS2_JOURNAL_IN_SHUTDOWN;
930 /* The OCFS2_JOURNAL_IN_SHUTDOWN will signal to commit_cache to not
931 * drop the trans_lock (which we want to hold until we
932 * completely destroy the journal. */
933 if (osb->commit_task) {
934 /* Wait for the commit thread */
935 trace_ocfs2_journal_shutdown_wait(osb->commit_task);
936 kthread_stop(osb->commit_task);
937 osb->commit_task = NULL;
940 BUG_ON(atomic_read(&(osb->journal->j_num_trans)) != 0);
942 if (ocfs2_mount_local(osb)) {
943 jbd2_journal_lock_updates(journal->j_journal);
944 status = jbd2_journal_flush(journal->j_journal);
945 jbd2_journal_unlock_updates(journal->j_journal);
952 * Do not toggle if flush was unsuccessful otherwise
953 * will leave dirty metadata in a "clean" journal
955 status = ocfs2_journal_toggle_dirty(osb, 0, 0);
960 /* Shutdown the kernel journal system */
961 jbd2_journal_destroy(journal->j_journal);
962 journal->j_journal = NULL;
964 OCFS2_I(inode)->ip_open_count--;
966 /* unlock our journal */
967 ocfs2_inode_unlock(inode, 1);
969 brelse(journal->j_bh);
970 journal->j_bh = NULL;
972 journal->j_state = OCFS2_JOURNAL_FREE;
974 // up_write(&journal->j_trans_barrier);
980 static void ocfs2_clear_journal_error(struct super_block *sb,
986 olderr = jbd2_journal_errno(journal);
988 mlog(ML_ERROR, "File system error %d recorded in "
989 "journal %u.\n", olderr, slot);
990 mlog(ML_ERROR, "File system on device %s needs checking.\n",
993 jbd2_journal_ack_err(journal);
994 jbd2_journal_clear_err(journal);
998 int ocfs2_journal_load(struct ocfs2_journal *journal, int local, int replayed)
1001 struct ocfs2_super *osb;
1005 osb = journal->j_osb;
1007 status = jbd2_journal_load(journal->j_journal);
1009 mlog(ML_ERROR, "Failed to load journal!\n");
1013 ocfs2_clear_journal_error(osb->sb, journal->j_journal, osb->slot_num);
1015 status = ocfs2_journal_toggle_dirty(osb, 1, replayed);
1021 /* Launch the commit thread */
1023 osb->commit_task = kthread_run(ocfs2_commit_thread, osb,
1025 if (IS_ERR(osb->commit_task)) {
1026 status = PTR_ERR(osb->commit_task);
1027 osb->commit_task = NULL;
1028 mlog(ML_ERROR, "unable to launch ocfs2commit thread, "
1029 "error=%d", status);
1033 osb->commit_task = NULL;
1040 /* 'full' flag tells us whether we clear out all blocks or if we just
1041 * mark the journal clean */
1042 int ocfs2_journal_wipe(struct ocfs2_journal *journal, int full)
1048 status = jbd2_journal_wipe(journal->j_journal, full);
1054 status = ocfs2_journal_toggle_dirty(journal->j_osb, 0, 0);
1062 static int ocfs2_recovery_completed(struct ocfs2_super *osb)
1065 struct ocfs2_recovery_map *rm = osb->recovery_map;
1067 spin_lock(&osb->osb_lock);
1068 empty = (rm->rm_used == 0);
1069 spin_unlock(&osb->osb_lock);
1074 void ocfs2_wait_for_recovery(struct ocfs2_super *osb)
1076 wait_event(osb->recovery_event, ocfs2_recovery_completed(osb));
1080 * JBD Might read a cached version of another nodes journal file. We
1081 * don't want this as this file changes often and we get no
1082 * notification on those changes. The only way to be sure that we've
1083 * got the most up to date version of those blocks then is to force
1084 * read them off disk. Just searching through the buffer cache won't
1085 * work as there may be pages backing this file which are still marked
1086 * up to date. We know things can't change on this file underneath us
1087 * as we have the lock by now :)
1089 static int ocfs2_force_read_journal(struct inode *inode)
1093 u64 v_blkno, p_blkno, p_blocks, num_blocks;
1094 #define CONCURRENT_JOURNAL_FILL 32ULL
1095 struct buffer_head *bhs[CONCURRENT_JOURNAL_FILL];
1097 memset(bhs, 0, sizeof(struct buffer_head *) * CONCURRENT_JOURNAL_FILL);
1099 num_blocks = ocfs2_blocks_for_bytes(inode->i_sb, inode->i_size);
1101 while (v_blkno < num_blocks) {
1102 status = ocfs2_extent_map_get_blocks(inode, v_blkno,
1103 &p_blkno, &p_blocks, NULL);
1109 if (p_blocks > CONCURRENT_JOURNAL_FILL)
1110 p_blocks = CONCURRENT_JOURNAL_FILL;
1112 /* We are reading journal data which should not
1113 * be put in the uptodate cache */
1114 status = ocfs2_read_blocks_sync(OCFS2_SB(inode->i_sb),
1115 p_blkno, p_blocks, bhs);
1121 for(i = 0; i < p_blocks; i++) {
1126 v_blkno += p_blocks;
1130 for(i = 0; i < CONCURRENT_JOURNAL_FILL; i++)
1135 struct ocfs2_la_recovery_item {
1136 struct list_head lri_list;
1138 struct ocfs2_dinode *lri_la_dinode;
1139 struct ocfs2_dinode *lri_tl_dinode;
1140 struct ocfs2_quota_recovery *lri_qrec;
1143 /* Does the second half of the recovery process. By this point, the
1144 * node is marked clean and can actually be considered recovered,
1145 * hence it's no longer in the recovery map, but there's still some
1146 * cleanup we can do which shouldn't happen within the recovery thread
1147 * as locking in that context becomes very difficult if we are to take
1148 * recovering nodes into account.
1150 * NOTE: This function can and will sleep on recovery of other nodes
1151 * during cluster locking, just like any other ocfs2 process.
1153 void ocfs2_complete_recovery(struct work_struct *work)
1156 struct ocfs2_journal *journal =
1157 container_of(work, struct ocfs2_journal, j_recovery_work);
1158 struct ocfs2_super *osb = journal->j_osb;
1159 struct ocfs2_dinode *la_dinode, *tl_dinode;
1160 struct ocfs2_la_recovery_item *item, *n;
1161 struct ocfs2_quota_recovery *qrec;
1162 LIST_HEAD(tmp_la_list);
1164 trace_ocfs2_complete_recovery(
1165 (unsigned long long)OCFS2_I(journal->j_inode)->ip_blkno);
1167 spin_lock(&journal->j_lock);
1168 list_splice_init(&journal->j_la_cleanups, &tmp_la_list);
1169 spin_unlock(&journal->j_lock);
1171 list_for_each_entry_safe(item, n, &tmp_la_list, lri_list) {
1172 list_del_init(&item->lri_list);
1174 ocfs2_wait_on_quotas(osb);
1176 la_dinode = item->lri_la_dinode;
1177 tl_dinode = item->lri_tl_dinode;
1178 qrec = item->lri_qrec;
1180 trace_ocfs2_complete_recovery_slot(item->lri_slot,
1181 la_dinode ? le64_to_cpu(la_dinode->i_blkno) : 0,
1182 tl_dinode ? le64_to_cpu(tl_dinode->i_blkno) : 0,
1186 ret = ocfs2_complete_local_alloc_recovery(osb,
1195 ret = ocfs2_complete_truncate_log_recovery(osb,
1203 ret = ocfs2_recover_orphans(osb, item->lri_slot);
1208 ret = ocfs2_finish_quota_recovery(osb, qrec,
1212 /* Recovery info is already freed now */
1218 trace_ocfs2_complete_recovery_end(ret);
1221 /* NOTE: This function always eats your references to la_dinode and
1222 * tl_dinode, either manually on error, or by passing them to
1223 * ocfs2_complete_recovery */
1224 static void ocfs2_queue_recovery_completion(struct ocfs2_journal *journal,
1226 struct ocfs2_dinode *la_dinode,
1227 struct ocfs2_dinode *tl_dinode,
1228 struct ocfs2_quota_recovery *qrec)
1230 struct ocfs2_la_recovery_item *item;
1232 item = kmalloc(sizeof(struct ocfs2_la_recovery_item), GFP_NOFS);
1234 /* Though we wish to avoid it, we are in fact safe in
1235 * skipping local alloc cleanup as fsck.ocfs2 is more
1236 * than capable of reclaiming unused space. */
1241 ocfs2_free_quota_recovery(qrec);
1243 mlog_errno(-ENOMEM);
1247 INIT_LIST_HEAD(&item->lri_list);
1248 item->lri_la_dinode = la_dinode;
1249 item->lri_slot = slot_num;
1250 item->lri_tl_dinode = tl_dinode;
1251 item->lri_qrec = qrec;
1253 spin_lock(&journal->j_lock);
1254 list_add_tail(&item->lri_list, &journal->j_la_cleanups);
1255 queue_work(ocfs2_wq, &journal->j_recovery_work);
1256 spin_unlock(&journal->j_lock);
1259 /* Called by the mount code to queue recovery the last part of
1260 * recovery for it's own and offline slot(s). */
1261 void ocfs2_complete_mount_recovery(struct ocfs2_super *osb)
1263 struct ocfs2_journal *journal = osb->journal;
1265 if (ocfs2_is_hard_readonly(osb))
1268 /* No need to queue up our truncate_log as regular cleanup will catch
1270 ocfs2_queue_recovery_completion(journal, osb->slot_num,
1271 osb->local_alloc_copy, NULL, NULL);
1272 ocfs2_schedule_truncate_log_flush(osb, 0);
1274 osb->local_alloc_copy = NULL;
1277 /* queue to recover orphan slots for all offline slots */
1278 ocfs2_replay_map_set_state(osb, REPLAY_NEEDED);
1279 ocfs2_queue_replay_slots(osb);
1280 ocfs2_free_replay_slots(osb);
1283 void ocfs2_complete_quota_recovery(struct ocfs2_super *osb)
1285 if (osb->quota_rec) {
1286 ocfs2_queue_recovery_completion(osb->journal,
1291 osb->quota_rec = NULL;
1295 static int __ocfs2_recovery_thread(void *arg)
1297 int status, node_num, slot_num;
1298 struct ocfs2_super *osb = arg;
1299 struct ocfs2_recovery_map *rm = osb->recovery_map;
1300 int *rm_quota = NULL;
1301 int rm_quota_used = 0, i;
1302 struct ocfs2_quota_recovery *qrec;
1304 status = ocfs2_wait_on_mount(osb);
1309 rm_quota = kzalloc(osb->max_slots * sizeof(int), GFP_NOFS);
1315 status = ocfs2_super_lock(osb, 1);
1321 status = ocfs2_compute_replay_slots(osb);
1325 /* queue recovery for our own slot */
1326 ocfs2_queue_recovery_completion(osb->journal, osb->slot_num, NULL,
1329 spin_lock(&osb->osb_lock);
1330 while (rm->rm_used) {
1331 /* It's always safe to remove entry zero, as we won't
1332 * clear it until ocfs2_recover_node() has succeeded. */
1333 node_num = rm->rm_entries[0];
1334 spin_unlock(&osb->osb_lock);
1335 slot_num = ocfs2_node_num_to_slot(osb, node_num);
1336 trace_ocfs2_recovery_thread_node(node_num, slot_num);
1337 if (slot_num == -ENOENT) {
1342 /* It is a bit subtle with quota recovery. We cannot do it
1343 * immediately because we have to obtain cluster locks from
1344 * quota files and we also don't want to just skip it because
1345 * then quota usage would be out of sync until some node takes
1346 * the slot. So we remember which nodes need quota recovery
1347 * and when everything else is done, we recover quotas. */
1348 for (i = 0; i < rm_quota_used && rm_quota[i] != slot_num; i++);
1349 if (i == rm_quota_used)
1350 rm_quota[rm_quota_used++] = slot_num;
1352 status = ocfs2_recover_node(osb, node_num, slot_num);
1355 ocfs2_recovery_map_clear(osb, node_num);
1358 "Error %d recovering node %d on device (%u,%u)!\n",
1360 MAJOR(osb->sb->s_dev), MINOR(osb->sb->s_dev));
1361 mlog(ML_ERROR, "Volume requires unmount.\n");
1364 spin_lock(&osb->osb_lock);
1366 spin_unlock(&osb->osb_lock);
1367 trace_ocfs2_recovery_thread_end(status);
1369 /* Refresh all journal recovery generations from disk */
1370 status = ocfs2_check_journals_nolocks(osb);
1371 status = (status == -EROFS) ? 0 : status;
1375 /* Now it is right time to recover quotas... We have to do this under
1376 * superblock lock so that no one can start using the slot (and crash)
1377 * before we recover it */
1378 for (i = 0; i < rm_quota_used; i++) {
1379 qrec = ocfs2_begin_quota_recovery(osb, rm_quota[i]);
1381 status = PTR_ERR(qrec);
1385 ocfs2_queue_recovery_completion(osb->journal, rm_quota[i],
1389 ocfs2_super_unlock(osb, 1);
1391 /* queue recovery for offline slots */
1392 ocfs2_queue_replay_slots(osb);
1395 mutex_lock(&osb->recovery_lock);
1396 if (!status && !ocfs2_recovery_completed(osb)) {
1397 mutex_unlock(&osb->recovery_lock);
1401 ocfs2_free_replay_slots(osb);
1402 osb->recovery_thread_task = NULL;
1403 mb(); /* sync with ocfs2_recovery_thread_running */
1404 wake_up(&osb->recovery_event);
1406 mutex_unlock(&osb->recovery_lock);
1410 /* no one is callint kthread_stop() for us so the kthread() api
1411 * requires that we call do_exit(). And it isn't exported, but
1412 * complete_and_exit() seems to be a minimal wrapper around it. */
1413 complete_and_exit(NULL, status);
1417 void ocfs2_recovery_thread(struct ocfs2_super *osb, int node_num)
1419 mutex_lock(&osb->recovery_lock);
1421 trace_ocfs2_recovery_thread(node_num, osb->node_num,
1422 osb->disable_recovery, osb->recovery_thread_task,
1423 osb->disable_recovery ?
1424 -1 : ocfs2_recovery_map_set(osb, node_num));
1426 if (osb->disable_recovery)
1429 if (osb->recovery_thread_task)
1432 osb->recovery_thread_task = kthread_run(__ocfs2_recovery_thread, osb,
1434 if (IS_ERR(osb->recovery_thread_task)) {
1435 mlog_errno((int)PTR_ERR(osb->recovery_thread_task));
1436 osb->recovery_thread_task = NULL;
1440 mutex_unlock(&osb->recovery_lock);
1441 wake_up(&osb->recovery_event);
1444 static int ocfs2_read_journal_inode(struct ocfs2_super *osb,
1446 struct buffer_head **bh,
1447 struct inode **ret_inode)
1449 int status = -EACCES;
1450 struct inode *inode = NULL;
1452 BUG_ON(slot_num >= osb->max_slots);
1454 inode = ocfs2_get_system_file_inode(osb, JOURNAL_SYSTEM_INODE,
1456 if (!inode || is_bad_inode(inode)) {
1460 SET_INODE_JOURNAL(inode);
1462 status = ocfs2_read_inode_block_full(inode, bh, OCFS2_BH_IGNORE_CACHE);
1472 if (status || !ret_inode)
1480 /* Does the actual journal replay and marks the journal inode as
1481 * clean. Will only replay if the journal inode is marked dirty. */
1482 static int ocfs2_replay_journal(struct ocfs2_super *osb,
1489 struct inode *inode = NULL;
1490 struct ocfs2_dinode *fe;
1491 journal_t *journal = NULL;
1492 struct buffer_head *bh = NULL;
1495 status = ocfs2_read_journal_inode(osb, slot_num, &bh, &inode);
1501 fe = (struct ocfs2_dinode *)bh->b_data;
1502 slot_reco_gen = ocfs2_get_recovery_generation(fe);
1507 * As the fs recovery is asynchronous, there is a small chance that
1508 * another node mounted (and recovered) the slot before the recovery
1509 * thread could get the lock. To handle that, we dirty read the journal
1510 * inode for that slot to get the recovery generation. If it is
1511 * different than what we expected, the slot has been recovered.
1512 * If not, it needs recovery.
1514 if (osb->slot_recovery_generations[slot_num] != slot_reco_gen) {
1515 trace_ocfs2_replay_journal_recovered(slot_num,
1516 osb->slot_recovery_generations[slot_num], slot_reco_gen);
1517 osb->slot_recovery_generations[slot_num] = slot_reco_gen;
1522 /* Continue with recovery as the journal has not yet been recovered */
1524 status = ocfs2_inode_lock_full(inode, &bh, 1, OCFS2_META_LOCK_RECOVERY);
1526 trace_ocfs2_replay_journal_lock_err(status);
1527 if (status != -ERESTARTSYS)
1528 mlog(ML_ERROR, "Could not lock journal!\n");
1533 fe = (struct ocfs2_dinode *) bh->b_data;
1535 flags = le32_to_cpu(fe->id1.journal1.ij_flags);
1536 slot_reco_gen = ocfs2_get_recovery_generation(fe);
1538 if (!(flags & OCFS2_JOURNAL_DIRTY_FL)) {
1539 trace_ocfs2_replay_journal_skip(node_num);
1540 /* Refresh recovery generation for the slot */
1541 osb->slot_recovery_generations[slot_num] = slot_reco_gen;
1545 /* we need to run complete recovery for offline orphan slots */
1546 ocfs2_replay_map_set_state(osb, REPLAY_NEEDED);
1548 printk(KERN_NOTICE "ocfs2: Begin replay journal (node %d, slot %d) on "\
1549 "device (%u,%u)\n", node_num, slot_num, MAJOR(osb->sb->s_dev),
1550 MINOR(osb->sb->s_dev));
1552 OCFS2_I(inode)->ip_clusters = le32_to_cpu(fe->i_clusters);
1554 status = ocfs2_force_read_journal(inode);
1560 journal = jbd2_journal_init_inode(inode);
1561 if (journal == NULL) {
1562 mlog(ML_ERROR, "Linux journal layer error\n");
1567 status = jbd2_journal_load(journal);
1572 jbd2_journal_destroy(journal);
1576 ocfs2_clear_journal_error(osb->sb, journal, slot_num);
1578 /* wipe the journal */
1579 jbd2_journal_lock_updates(journal);
1580 status = jbd2_journal_flush(journal);
1581 jbd2_journal_unlock_updates(journal);
1585 /* This will mark the node clean */
1586 flags = le32_to_cpu(fe->id1.journal1.ij_flags);
1587 flags &= ~OCFS2_JOURNAL_DIRTY_FL;
1588 fe->id1.journal1.ij_flags = cpu_to_le32(flags);
1590 /* Increment recovery generation to indicate successful recovery */
1591 ocfs2_bump_recovery_generation(fe);
1592 osb->slot_recovery_generations[slot_num] =
1593 ocfs2_get_recovery_generation(fe);
1595 ocfs2_compute_meta_ecc(osb->sb, bh->b_data, &fe->i_check);
1596 status = ocfs2_write_block(osb, bh, INODE_CACHE(inode));
1603 jbd2_journal_destroy(journal);
1605 printk(KERN_NOTICE "ocfs2: End replay journal (node %d, slot %d) on "\
1606 "device (%u,%u)\n", node_num, slot_num, MAJOR(osb->sb->s_dev),
1607 MINOR(osb->sb->s_dev));
1609 /* drop the lock on this nodes journal */
1611 ocfs2_inode_unlock(inode, 1);
1622 * Do the most important parts of node recovery:
1623 * - Replay it's journal
1624 * - Stamp a clean local allocator file
1625 * - Stamp a clean truncate log
1626 * - Mark the node clean
1628 * If this function completes without error, a node in OCFS2 can be
1629 * said to have been safely recovered. As a result, failure during the
1630 * second part of a nodes recovery process (local alloc recovery) is
1631 * far less concerning.
1633 static int ocfs2_recover_node(struct ocfs2_super *osb,
1634 int node_num, int slot_num)
1637 struct ocfs2_dinode *la_copy = NULL;
1638 struct ocfs2_dinode *tl_copy = NULL;
1640 trace_ocfs2_recover_node(node_num, slot_num, osb->node_num);
1642 /* Should not ever be called to recover ourselves -- in that
1643 * case we should've called ocfs2_journal_load instead. */
1644 BUG_ON(osb->node_num == node_num);
1646 status = ocfs2_replay_journal(osb, node_num, slot_num);
1648 if (status == -EBUSY) {
1649 trace_ocfs2_recover_node_skip(slot_num, node_num);
1657 /* Stamp a clean local alloc file AFTER recovering the journal... */
1658 status = ocfs2_begin_local_alloc_recovery(osb, slot_num, &la_copy);
1664 /* An error from begin_truncate_log_recovery is not
1665 * serious enough to warrant halting the rest of
1667 status = ocfs2_begin_truncate_log_recovery(osb, slot_num, &tl_copy);
1671 /* Likewise, this would be a strange but ultimately not so
1672 * harmful place to get an error... */
1673 status = ocfs2_clear_slot(osb, slot_num);
1677 /* This will kfree the memory pointed to by la_copy and tl_copy */
1678 ocfs2_queue_recovery_completion(osb->journal, slot_num, la_copy,
1687 /* Test node liveness by trylocking his journal. If we get the lock,
1688 * we drop it here. Return 0 if we got the lock, -EAGAIN if node is
1689 * still alive (we couldn't get the lock) and < 0 on error. */
1690 static int ocfs2_trylock_journal(struct ocfs2_super *osb,
1694 struct inode *inode = NULL;
1696 inode = ocfs2_get_system_file_inode(osb, JOURNAL_SYSTEM_INODE,
1698 if (inode == NULL) {
1699 mlog(ML_ERROR, "access error\n");
1703 if (is_bad_inode(inode)) {
1704 mlog(ML_ERROR, "access error (bad inode)\n");
1710 SET_INODE_JOURNAL(inode);
1712 flags = OCFS2_META_LOCK_RECOVERY | OCFS2_META_LOCK_NOQUEUE;
1713 status = ocfs2_inode_lock_full(inode, NULL, 1, flags);
1715 if (status != -EAGAIN)
1720 ocfs2_inode_unlock(inode, 1);
1728 /* Call this underneath ocfs2_super_lock. It also assumes that the
1729 * slot info struct has been updated from disk. */
1730 int ocfs2_mark_dead_nodes(struct ocfs2_super *osb)
1732 unsigned int node_num;
1735 struct buffer_head *bh = NULL;
1736 struct ocfs2_dinode *di;
1738 /* This is called with the super block cluster lock, so we
1739 * know that the slot map can't change underneath us. */
1741 for (i = 0; i < osb->max_slots; i++) {
1742 /* Read journal inode to get the recovery generation */
1743 status = ocfs2_read_journal_inode(osb, i, &bh, NULL);
1748 di = (struct ocfs2_dinode *)bh->b_data;
1749 gen = ocfs2_get_recovery_generation(di);
1753 spin_lock(&osb->osb_lock);
1754 osb->slot_recovery_generations[i] = gen;
1756 trace_ocfs2_mark_dead_nodes(i,
1757 osb->slot_recovery_generations[i]);
1759 if (i == osb->slot_num) {
1760 spin_unlock(&osb->osb_lock);
1764 status = ocfs2_slot_to_node_num_locked(osb, i, &node_num);
1765 if (status == -ENOENT) {
1766 spin_unlock(&osb->osb_lock);
1770 if (__ocfs2_recovery_map_test(osb, node_num)) {
1771 spin_unlock(&osb->osb_lock);
1774 spin_unlock(&osb->osb_lock);
1776 /* Ok, we have a slot occupied by another node which
1777 * is not in the recovery map. We trylock his journal
1778 * file here to test if he's alive. */
1779 status = ocfs2_trylock_journal(osb, i);
1781 /* Since we're called from mount, we know that
1782 * the recovery thread can't race us on
1783 * setting / checking the recovery bits. */
1784 ocfs2_recovery_thread(osb, node_num);
1785 } else if ((status < 0) && (status != -EAGAIN)) {
1797 * Scan timer should get fired every ORPHAN_SCAN_SCHEDULE_TIMEOUT. Add some
1798 * randomness to the timeout to minimize multple nodes firing the timer at the
1801 static inline unsigned long ocfs2_orphan_scan_timeout(void)
1805 get_random_bytes(&time, sizeof(time));
1806 time = ORPHAN_SCAN_SCHEDULE_TIMEOUT + (time % 5000);
1807 return msecs_to_jiffies(time);
1811 * ocfs2_queue_orphan_scan calls ocfs2_queue_recovery_completion for
1812 * every slot, queuing a recovery of the slot on the ocfs2_wq thread. This
1813 * is done to catch any orphans that are left over in orphan directories.
1815 * It scans all slots, even ones that are in use. It does so to handle the
1816 * case described below:
1818 * Node 1 has an inode it was using. The dentry went away due to memory
1819 * pressure. Node 1 closes the inode, but it's on the free list. The node
1820 * has the open lock.
1821 * Node 2 unlinks the inode. It grabs the dentry lock to notify others,
1822 * but node 1 has no dentry and doesn't get the message. It trylocks the
1823 * open lock, sees that another node has a PR, and does nothing.
1824 * Later node 2 runs its orphan dir. It igets the inode, trylocks the
1825 * open lock, sees the PR still, and does nothing.
1826 * Basically, we have to trigger an orphan iput on node 1. The only way
1827 * for this to happen is if node 1 runs node 2's orphan dir.
1829 * ocfs2_queue_orphan_scan gets called every ORPHAN_SCAN_SCHEDULE_TIMEOUT
1830 * seconds. It gets an EX lock on os_lockres and checks sequence number
1831 * stored in LVB. If the sequence number has changed, it means some other
1832 * node has done the scan. This node skips the scan and tracks the
1833 * sequence number. If the sequence number didn't change, it means a scan
1834 * hasn't happened. The node queues a scan and increments the
1835 * sequence number in the LVB.
1837 void ocfs2_queue_orphan_scan(struct ocfs2_super *osb)
1839 struct ocfs2_orphan_scan *os;
1843 os = &osb->osb_orphan_scan;
1845 if (atomic_read(&os->os_state) == ORPHAN_SCAN_INACTIVE)
1848 trace_ocfs2_queue_orphan_scan_begin(os->os_count, os->os_seqno,
1849 atomic_read(&os->os_state));
1851 status = ocfs2_orphan_scan_lock(osb, &seqno);
1853 if (status != -EAGAIN)
1858 /* Do no queue the tasks if the volume is being umounted */
1859 if (atomic_read(&os->os_state) == ORPHAN_SCAN_INACTIVE)
1862 if (os->os_seqno != seqno) {
1863 os->os_seqno = seqno;
1867 for (i = 0; i < osb->max_slots; i++)
1868 ocfs2_queue_recovery_completion(osb->journal, i, NULL, NULL,
1871 * We queued a recovery on orphan slots, increment the sequence
1872 * number and update LVB so other node will skip the scan for a while
1876 os->os_scantime = CURRENT_TIME;
1878 ocfs2_orphan_scan_unlock(osb, seqno);
1880 trace_ocfs2_queue_orphan_scan_end(os->os_count, os->os_seqno,
1881 atomic_read(&os->os_state));
1885 /* Worker task that gets fired every ORPHAN_SCAN_SCHEDULE_TIMEOUT millsec */
1886 void ocfs2_orphan_scan_work(struct work_struct *work)
1888 struct ocfs2_orphan_scan *os;
1889 struct ocfs2_super *osb;
1891 os = container_of(work, struct ocfs2_orphan_scan,
1892 os_orphan_scan_work.work);
1895 mutex_lock(&os->os_lock);
1896 ocfs2_queue_orphan_scan(osb);
1897 if (atomic_read(&os->os_state) == ORPHAN_SCAN_ACTIVE)
1898 queue_delayed_work(ocfs2_wq, &os->os_orphan_scan_work,
1899 ocfs2_orphan_scan_timeout());
1900 mutex_unlock(&os->os_lock);
1903 void ocfs2_orphan_scan_stop(struct ocfs2_super *osb)
1905 struct ocfs2_orphan_scan *os;
1907 os = &osb->osb_orphan_scan;
1908 if (atomic_read(&os->os_state) == ORPHAN_SCAN_ACTIVE) {
1909 atomic_set(&os->os_state, ORPHAN_SCAN_INACTIVE);
1910 mutex_lock(&os->os_lock);
1911 cancel_delayed_work(&os->os_orphan_scan_work);
1912 mutex_unlock(&os->os_lock);
1916 void ocfs2_orphan_scan_init(struct ocfs2_super *osb)
1918 struct ocfs2_orphan_scan *os;
1920 os = &osb->osb_orphan_scan;
1924 mutex_init(&os->os_lock);
1925 INIT_DELAYED_WORK(&os->os_orphan_scan_work, ocfs2_orphan_scan_work);
1928 void ocfs2_orphan_scan_start(struct ocfs2_super *osb)
1930 struct ocfs2_orphan_scan *os;
1932 os = &osb->osb_orphan_scan;
1933 os->os_scantime = CURRENT_TIME;
1934 if (ocfs2_is_hard_readonly(osb) || ocfs2_mount_local(osb))
1935 atomic_set(&os->os_state, ORPHAN_SCAN_INACTIVE);
1937 atomic_set(&os->os_state, ORPHAN_SCAN_ACTIVE);
1938 queue_delayed_work(ocfs2_wq, &os->os_orphan_scan_work,
1939 ocfs2_orphan_scan_timeout());
1943 struct ocfs2_orphan_filldir_priv {
1944 struct dir_context ctx;
1946 struct ocfs2_super *osb;
1949 static int ocfs2_orphan_filldir(void *priv, const char *name, int name_len,
1950 loff_t pos, u64 ino, unsigned type)
1952 struct ocfs2_orphan_filldir_priv *p = priv;
1955 if (name_len == 1 && !strncmp(".", name, 1))
1957 if (name_len == 2 && !strncmp("..", name, 2))
1960 /* Skip bad inodes so that recovery can continue */
1961 iter = ocfs2_iget(p->osb, ino,
1962 OCFS2_FI_FLAG_ORPHAN_RECOVERY, 0);
1966 trace_ocfs2_orphan_filldir((unsigned long long)OCFS2_I(iter)->ip_blkno);
1967 /* No locking is required for the next_orphan queue as there
1968 * is only ever a single process doing orphan recovery. */
1969 OCFS2_I(iter)->ip_next_orphan = p->head;
1975 static int ocfs2_queue_orphans(struct ocfs2_super *osb,
1977 struct inode **head)
1980 struct inode *orphan_dir_inode = NULL;
1981 struct ocfs2_orphan_filldir_priv priv = {
1982 .ctx.actor = ocfs2_orphan_filldir,
1987 orphan_dir_inode = ocfs2_get_system_file_inode(osb,
1988 ORPHAN_DIR_SYSTEM_INODE,
1990 if (!orphan_dir_inode) {
1996 mutex_lock(&orphan_dir_inode->i_mutex);
1997 status = ocfs2_inode_lock(orphan_dir_inode, NULL, 0);
2003 status = ocfs2_dir_foreach(orphan_dir_inode, &priv.ctx);
2012 ocfs2_inode_unlock(orphan_dir_inode, 0);
2014 mutex_unlock(&orphan_dir_inode->i_mutex);
2015 iput(orphan_dir_inode);
2019 static int ocfs2_orphan_recovery_can_continue(struct ocfs2_super *osb,
2024 spin_lock(&osb->osb_lock);
2025 ret = !osb->osb_orphan_wipes[slot];
2026 spin_unlock(&osb->osb_lock);
2030 static void ocfs2_mark_recovering_orphan_dir(struct ocfs2_super *osb,
2033 spin_lock(&osb->osb_lock);
2034 /* Mark ourselves such that new processes in delete_inode()
2035 * know to quit early. */
2036 ocfs2_node_map_set_bit(osb, &osb->osb_recovering_orphan_dirs, slot);
2037 while (osb->osb_orphan_wipes[slot]) {
2038 /* If any processes are already in the middle of an
2039 * orphan wipe on this dir, then we need to wait for
2041 spin_unlock(&osb->osb_lock);
2042 wait_event_interruptible(osb->osb_wipe_event,
2043 ocfs2_orphan_recovery_can_continue(osb, slot));
2044 spin_lock(&osb->osb_lock);
2046 spin_unlock(&osb->osb_lock);
2049 static void ocfs2_clear_recovering_orphan_dir(struct ocfs2_super *osb,
2052 ocfs2_node_map_clear_bit(osb, &osb->osb_recovering_orphan_dirs, slot);
2056 * Orphan recovery. Each mounted node has it's own orphan dir which we
2057 * must run during recovery. Our strategy here is to build a list of
2058 * the inodes in the orphan dir and iget/iput them. The VFS does
2059 * (most) of the rest of the work.
2061 * Orphan recovery can happen at any time, not just mount so we have a
2062 * couple of extra considerations.
2064 * - We grab as many inodes as we can under the orphan dir lock -
2065 * doing iget() outside the orphan dir risks getting a reference on
2067 * - We must be sure not to deadlock with other processes on the
2068 * system wanting to run delete_inode(). This can happen when they go
2069 * to lock the orphan dir and the orphan recovery process attempts to
2070 * iget() inside the orphan dir lock. This can be avoided by
2071 * advertising our state to ocfs2_delete_inode().
2073 static int ocfs2_recover_orphans(struct ocfs2_super *osb,
2077 struct inode *inode = NULL;
2079 struct ocfs2_inode_info *oi;
2081 trace_ocfs2_recover_orphans(slot);
2083 ocfs2_mark_recovering_orphan_dir(osb, slot);
2084 ret = ocfs2_queue_orphans(osb, slot, &inode);
2085 ocfs2_clear_recovering_orphan_dir(osb, slot);
2087 /* Error here should be noted, but we want to continue with as
2088 * many queued inodes as we've got. */
2093 oi = OCFS2_I(inode);
2094 trace_ocfs2_recover_orphans_iput(
2095 (unsigned long long)oi->ip_blkno);
2097 iter = oi->ip_next_orphan;
2099 spin_lock(&oi->ip_lock);
2100 /* The remote delete code may have set these on the
2101 * assumption that the other node would wipe them
2102 * successfully. If they are still in the node's
2103 * orphan dir, we need to reset that state. */
2104 oi->ip_flags &= ~(OCFS2_INODE_DELETED|OCFS2_INODE_SKIP_DELETE);
2106 /* Set the proper information to get us going into
2107 * ocfs2_delete_inode. */
2108 oi->ip_flags |= OCFS2_INODE_MAYBE_ORPHANED;
2109 spin_unlock(&oi->ip_lock);
2119 static int __ocfs2_wait_on_mount(struct ocfs2_super *osb, int quota)
2121 /* This check is good because ocfs2 will wait on our recovery
2122 * thread before changing it to something other than MOUNTED
2124 wait_event(osb->osb_mount_event,
2125 (!quota && atomic_read(&osb->vol_state) == VOLUME_MOUNTED) ||
2126 atomic_read(&osb->vol_state) == VOLUME_MOUNTED_QUOTAS ||
2127 atomic_read(&osb->vol_state) == VOLUME_DISABLED);
2129 /* If there's an error on mount, then we may never get to the
2130 * MOUNTED flag, but this is set right before
2131 * dismount_volume() so we can trust it. */
2132 if (atomic_read(&osb->vol_state) == VOLUME_DISABLED) {
2133 trace_ocfs2_wait_on_mount(VOLUME_DISABLED);
2134 mlog(0, "mount error, exiting!\n");
2141 static int ocfs2_commit_thread(void *arg)
2144 struct ocfs2_super *osb = arg;
2145 struct ocfs2_journal *journal = osb->journal;
2147 /* we can trust j_num_trans here because _should_stop() is only set in
2148 * shutdown and nobody other than ourselves should be able to start
2149 * transactions. committing on shutdown might take a few iterations
2150 * as final transactions put deleted inodes on the list */
2151 while (!(kthread_should_stop() &&
2152 atomic_read(&journal->j_num_trans) == 0)) {
2154 wait_event_interruptible(osb->checkpoint_event,
2155 atomic_read(&journal->j_num_trans)
2156 || kthread_should_stop());
2158 status = ocfs2_commit_cache(osb);
2162 if (kthread_should_stop() && atomic_read(&journal->j_num_trans)){
2164 "commit_thread: %u transactions pending on "
2166 atomic_read(&journal->j_num_trans));
2173 /* Reads all the journal inodes without taking any cluster locks. Used
2174 * for hard readonly access to determine whether any journal requires
2175 * recovery. Also used to refresh the recovery generation numbers after
2176 * a journal has been recovered by another node.
2178 int ocfs2_check_journals_nolocks(struct ocfs2_super *osb)
2182 struct buffer_head *di_bh = NULL;
2183 struct ocfs2_dinode *di;
2184 int journal_dirty = 0;
2186 for(slot = 0; slot < osb->max_slots; slot++) {
2187 ret = ocfs2_read_journal_inode(osb, slot, &di_bh, NULL);
2193 di = (struct ocfs2_dinode *) di_bh->b_data;
2195 osb->slot_recovery_generations[slot] =
2196 ocfs2_get_recovery_generation(di);
2198 if (le32_to_cpu(di->id1.journal1.ij_flags) &
2199 OCFS2_JOURNAL_DIRTY_FL)