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>
33 #include <linux/delay.h>
35 #include <cluster/masklog.h>
40 #include "blockcheck.h"
43 #include "extent_map.h"
44 #include "heartbeat.h"
47 #include "localalloc.h"
56 #include "buffer_head_io.h"
57 #include "ocfs2_trace.h"
59 DEFINE_SPINLOCK(trans_inc_lock
);
61 #define ORPHAN_SCAN_SCHEDULE_TIMEOUT 300000
63 static int ocfs2_force_read_journal(struct inode
*inode
);
64 static int ocfs2_recover_node(struct ocfs2_super
*osb
,
65 int node_num
, int slot_num
);
66 static int __ocfs2_recovery_thread(void *arg
);
67 static int ocfs2_commit_cache(struct ocfs2_super
*osb
);
68 static int __ocfs2_wait_on_mount(struct ocfs2_super
*osb
, int quota
);
69 static int ocfs2_journal_toggle_dirty(struct ocfs2_super
*osb
,
70 int dirty
, int replayed
);
71 static int ocfs2_trylock_journal(struct ocfs2_super
*osb
,
73 static int ocfs2_recover_orphans(struct ocfs2_super
*osb
,
75 enum ocfs2_orphan_reco_type orphan_reco_type
);
76 static int ocfs2_commit_thread(void *arg
);
77 static void ocfs2_queue_recovery_completion(struct ocfs2_journal
*journal
,
79 struct ocfs2_dinode
*la_dinode
,
80 struct ocfs2_dinode
*tl_dinode
,
81 struct ocfs2_quota_recovery
*qrec
,
82 enum ocfs2_orphan_reco_type orphan_reco_type
);
84 static inline int ocfs2_wait_on_mount(struct ocfs2_super
*osb
)
86 return __ocfs2_wait_on_mount(osb
, 0);
89 static inline int ocfs2_wait_on_quotas(struct ocfs2_super
*osb
)
91 return __ocfs2_wait_on_mount(osb
, 1);
95 * This replay_map is to track online/offline slots, so we could recover
96 * offline slots during recovery and mount
99 enum ocfs2_replay_state
{
100 REPLAY_UNNEEDED
= 0, /* Replay is not needed, so ignore this map */
101 REPLAY_NEEDED
, /* Replay slots marked in rm_replay_slots */
102 REPLAY_DONE
/* Replay was already queued */
105 struct ocfs2_replay_map
{
106 unsigned int rm_slots
;
107 enum ocfs2_replay_state rm_state
;
108 unsigned char rm_replay_slots
[0];
111 static void ocfs2_replay_map_set_state(struct ocfs2_super
*osb
, int state
)
113 if (!osb
->replay_map
)
116 /* If we've already queued the replay, we don't have any more to do */
117 if (osb
->replay_map
->rm_state
== REPLAY_DONE
)
120 osb
->replay_map
->rm_state
= state
;
123 int ocfs2_compute_replay_slots(struct ocfs2_super
*osb
)
125 struct ocfs2_replay_map
*replay_map
;
128 /* If replay map is already set, we don't do it again */
132 replay_map
= kzalloc(sizeof(struct ocfs2_replay_map
) +
133 (osb
->max_slots
* sizeof(char)), GFP_KERNEL
);
140 spin_lock(&osb
->osb_lock
);
142 replay_map
->rm_slots
= osb
->max_slots
;
143 replay_map
->rm_state
= REPLAY_UNNEEDED
;
145 /* set rm_replay_slots for offline slot(s) */
146 for (i
= 0; i
< replay_map
->rm_slots
; i
++) {
147 if (ocfs2_slot_to_node_num_locked(osb
, i
, &node_num
) == -ENOENT
)
148 replay_map
->rm_replay_slots
[i
] = 1;
151 osb
->replay_map
= replay_map
;
152 spin_unlock(&osb
->osb_lock
);
156 static void ocfs2_queue_replay_slots(struct ocfs2_super
*osb
,
157 enum ocfs2_orphan_reco_type orphan_reco_type
)
159 struct ocfs2_replay_map
*replay_map
= osb
->replay_map
;
165 if (replay_map
->rm_state
!= REPLAY_NEEDED
)
168 for (i
= 0; i
< replay_map
->rm_slots
; i
++)
169 if (replay_map
->rm_replay_slots
[i
])
170 ocfs2_queue_recovery_completion(osb
->journal
, i
, NULL
,
173 replay_map
->rm_state
= REPLAY_DONE
;
176 static void ocfs2_free_replay_slots(struct ocfs2_super
*osb
)
178 struct ocfs2_replay_map
*replay_map
= osb
->replay_map
;
180 if (!osb
->replay_map
)
184 osb
->replay_map
= NULL
;
187 int ocfs2_recovery_init(struct ocfs2_super
*osb
)
189 struct ocfs2_recovery_map
*rm
;
191 mutex_init(&osb
->recovery_lock
);
192 osb
->disable_recovery
= 0;
193 osb
->recovery_thread_task
= NULL
;
194 init_waitqueue_head(&osb
->recovery_event
);
196 rm
= kzalloc(sizeof(struct ocfs2_recovery_map
) +
197 osb
->max_slots
* sizeof(unsigned int),
204 rm
->rm_entries
= (unsigned int *)((char *)rm
+
205 sizeof(struct ocfs2_recovery_map
));
206 osb
->recovery_map
= rm
;
211 /* we can't grab the goofy sem lock from inside wait_event, so we use
212 * memory barriers to make sure that we'll see the null task before
214 static int ocfs2_recovery_thread_running(struct ocfs2_super
*osb
)
217 return osb
->recovery_thread_task
!= NULL
;
220 void ocfs2_recovery_exit(struct ocfs2_super
*osb
)
222 struct ocfs2_recovery_map
*rm
;
224 /* disable any new recovery threads and wait for any currently
225 * running ones to exit. Do this before setting the vol_state. */
226 mutex_lock(&osb
->recovery_lock
);
227 osb
->disable_recovery
= 1;
228 mutex_unlock(&osb
->recovery_lock
);
229 wait_event(osb
->recovery_event
, !ocfs2_recovery_thread_running(osb
));
231 /* At this point, we know that no more recovery threads can be
232 * launched, so wait for any recovery completion work to
235 flush_workqueue(osb
->ocfs2_wq
);
238 * Now that recovery is shut down, and the osb is about to be
239 * freed, the osb_lock is not taken here.
241 rm
= osb
->recovery_map
;
242 /* XXX: Should we bug if there are dirty entries? */
247 static int __ocfs2_recovery_map_test(struct ocfs2_super
*osb
,
248 unsigned int node_num
)
251 struct ocfs2_recovery_map
*rm
= osb
->recovery_map
;
253 assert_spin_locked(&osb
->osb_lock
);
255 for (i
= 0; i
< rm
->rm_used
; i
++) {
256 if (rm
->rm_entries
[i
] == node_num
)
263 /* Behaves like test-and-set. Returns the previous value */
264 static int ocfs2_recovery_map_set(struct ocfs2_super
*osb
,
265 unsigned int node_num
)
267 struct ocfs2_recovery_map
*rm
= osb
->recovery_map
;
269 spin_lock(&osb
->osb_lock
);
270 if (__ocfs2_recovery_map_test(osb
, node_num
)) {
271 spin_unlock(&osb
->osb_lock
);
275 /* XXX: Can this be exploited? Not from o2dlm... */
276 BUG_ON(rm
->rm_used
>= osb
->max_slots
);
278 rm
->rm_entries
[rm
->rm_used
] = node_num
;
280 spin_unlock(&osb
->osb_lock
);
285 static void ocfs2_recovery_map_clear(struct ocfs2_super
*osb
,
286 unsigned int node_num
)
289 struct ocfs2_recovery_map
*rm
= osb
->recovery_map
;
291 spin_lock(&osb
->osb_lock
);
293 for (i
= 0; i
< rm
->rm_used
; i
++) {
294 if (rm
->rm_entries
[i
] == node_num
)
298 if (i
< rm
->rm_used
) {
299 /* XXX: be careful with the pointer math */
300 memmove(&(rm
->rm_entries
[i
]), &(rm
->rm_entries
[i
+ 1]),
301 (rm
->rm_used
- i
- 1) * sizeof(unsigned int));
305 spin_unlock(&osb
->osb_lock
);
308 static int ocfs2_commit_cache(struct ocfs2_super
*osb
)
311 unsigned int flushed
;
312 struct ocfs2_journal
*journal
= NULL
;
314 journal
= osb
->journal
;
316 /* Flush all pending commits and checkpoint the journal. */
317 down_write(&journal
->j_trans_barrier
);
319 flushed
= atomic_read(&journal
->j_num_trans
);
320 trace_ocfs2_commit_cache_begin(flushed
);
322 up_write(&journal
->j_trans_barrier
);
326 jbd2_journal_lock_updates(journal
->j_journal
);
327 status
= jbd2_journal_flush(journal
->j_journal
);
328 jbd2_journal_unlock_updates(journal
->j_journal
);
330 up_write(&journal
->j_trans_barrier
);
335 ocfs2_inc_trans_id(journal
);
337 flushed
= atomic_read(&journal
->j_num_trans
);
338 atomic_set(&journal
->j_num_trans
, 0);
339 up_write(&journal
->j_trans_barrier
);
341 trace_ocfs2_commit_cache_end(journal
->j_trans_id
, flushed
);
343 ocfs2_wake_downconvert_thread(osb
);
344 wake_up(&journal
->j_checkpointed
);
349 handle_t
*ocfs2_start_trans(struct ocfs2_super
*osb
, int max_buffs
)
351 journal_t
*journal
= osb
->journal
->j_journal
;
354 BUG_ON(!osb
|| !osb
->journal
->j_journal
);
356 if (ocfs2_is_hard_readonly(osb
))
357 return ERR_PTR(-EROFS
);
359 BUG_ON(osb
->journal
->j_state
== OCFS2_JOURNAL_FREE
);
360 BUG_ON(max_buffs
<= 0);
362 /* Nested transaction? Just return the handle... */
363 if (journal_current_handle())
364 return jbd2_journal_start(journal
, max_buffs
);
366 sb_start_intwrite(osb
->sb
);
368 down_read(&osb
->journal
->j_trans_barrier
);
370 handle
= jbd2_journal_start(journal
, max_buffs
);
371 if (IS_ERR(handle
)) {
372 up_read(&osb
->journal
->j_trans_barrier
);
373 sb_end_intwrite(osb
->sb
);
375 mlog_errno(PTR_ERR(handle
));
377 if (is_journal_aborted(journal
)) {
378 ocfs2_abort(osb
->sb
, "Detected aborted journal\n");
379 handle
= ERR_PTR(-EROFS
);
382 if (!ocfs2_mount_local(osb
))
383 atomic_inc(&(osb
->journal
->j_num_trans
));
389 int ocfs2_commit_trans(struct ocfs2_super
*osb
,
393 struct ocfs2_journal
*journal
= osb
->journal
;
397 nested
= handle
->h_ref
> 1;
398 ret
= jbd2_journal_stop(handle
);
403 up_read(&journal
->j_trans_barrier
);
404 sb_end_intwrite(osb
->sb
);
411 * 'nblocks' is what you want to add to the current transaction.
413 * This might call jbd2_journal_restart() which will commit dirty buffers
414 * and then restart the transaction. Before calling
415 * ocfs2_extend_trans(), any changed blocks should have been
416 * dirtied. After calling it, all blocks which need to be changed must
417 * go through another set of journal_access/journal_dirty calls.
419 * WARNING: This will not release any semaphores or disk locks taken
420 * during the transaction, so make sure they were taken *before*
421 * start_trans or we'll have ordering deadlocks.
423 * WARNING2: Note that we do *not* drop j_trans_barrier here. This is
424 * good because transaction ids haven't yet been recorded on the
425 * cluster locks associated with this handle.
427 int ocfs2_extend_trans(handle_t
*handle
, int nblocks
)
429 int status
, old_nblocks
;
437 old_nblocks
= handle
->h_buffer_credits
;
439 trace_ocfs2_extend_trans(old_nblocks
, nblocks
);
441 #ifdef CONFIG_OCFS2_DEBUG_FS
444 status
= jbd2_journal_extend(handle
, nblocks
);
452 trace_ocfs2_extend_trans_restart(old_nblocks
+ nblocks
);
453 status
= jbd2_journal_restart(handle
,
454 old_nblocks
+ nblocks
);
467 * If we have fewer than thresh credits, extend by OCFS2_MAX_TRANS_DATA.
468 * If that fails, restart the transaction & regain write access for the
469 * buffer head which is used for metadata modifications.
470 * Taken from Ext4: extend_or_restart_transaction()
472 int ocfs2_allocate_extend_trans(handle_t
*handle
, int thresh
)
474 int status
, old_nblks
;
478 old_nblks
= handle
->h_buffer_credits
;
479 trace_ocfs2_allocate_extend_trans(old_nblks
, thresh
);
481 if (old_nblks
< thresh
)
484 status
= jbd2_journal_extend(handle
, OCFS2_MAX_TRANS_DATA
);
491 status
= jbd2_journal_restart(handle
, OCFS2_MAX_TRANS_DATA
);
501 struct ocfs2_triggers
{
502 struct jbd2_buffer_trigger_type ot_triggers
;
506 static inline struct ocfs2_triggers
*to_ocfs2_trigger(struct jbd2_buffer_trigger_type
*triggers
)
508 return container_of(triggers
, struct ocfs2_triggers
, ot_triggers
);
511 static void ocfs2_frozen_trigger(struct jbd2_buffer_trigger_type
*triggers
,
512 struct buffer_head
*bh
,
513 void *data
, size_t size
)
515 struct ocfs2_triggers
*ot
= to_ocfs2_trigger(triggers
);
518 * We aren't guaranteed to have the superblock here, so we
519 * must unconditionally compute the ecc data.
520 * __ocfs2_journal_access() will only set the triggers if
521 * metaecc is enabled.
523 ocfs2_block_check_compute(data
, size
, data
+ ot
->ot_offset
);
527 * Quota blocks have their own trigger because the struct ocfs2_block_check
528 * offset depends on the blocksize.
530 static void ocfs2_dq_frozen_trigger(struct jbd2_buffer_trigger_type
*triggers
,
531 struct buffer_head
*bh
,
532 void *data
, size_t size
)
534 struct ocfs2_disk_dqtrailer
*dqt
=
535 ocfs2_block_dqtrailer(size
, data
);
538 * We aren't guaranteed to have the superblock here, so we
539 * must unconditionally compute the ecc data.
540 * __ocfs2_journal_access() will only set the triggers if
541 * metaecc is enabled.
543 ocfs2_block_check_compute(data
, size
, &dqt
->dq_check
);
547 * Directory blocks also have their own trigger because the
548 * struct ocfs2_block_check offset depends on the blocksize.
550 static void ocfs2_db_frozen_trigger(struct jbd2_buffer_trigger_type
*triggers
,
551 struct buffer_head
*bh
,
552 void *data
, size_t size
)
554 struct ocfs2_dir_block_trailer
*trailer
=
555 ocfs2_dir_trailer_from_size(size
, data
);
558 * We aren't guaranteed to have the superblock here, so we
559 * must unconditionally compute the ecc data.
560 * __ocfs2_journal_access() will only set the triggers if
561 * metaecc is enabled.
563 ocfs2_block_check_compute(data
, size
, &trailer
->db_check
);
566 static void ocfs2_abort_trigger(struct jbd2_buffer_trigger_type
*triggers
,
567 struct buffer_head
*bh
)
570 "ocfs2_abort_trigger called by JBD2. bh = 0x%lx, "
571 "bh->b_blocknr = %llu\n",
573 (unsigned long long)bh
->b_blocknr
);
575 ocfs2_error(bh
->b_bdev
->bd_super
,
576 "JBD2 has aborted our journal, ocfs2 cannot continue\n");
579 static struct ocfs2_triggers di_triggers
= {
581 .t_frozen
= ocfs2_frozen_trigger
,
582 .t_abort
= ocfs2_abort_trigger
,
584 .ot_offset
= offsetof(struct ocfs2_dinode
, i_check
),
587 static struct ocfs2_triggers eb_triggers
= {
589 .t_frozen
= ocfs2_frozen_trigger
,
590 .t_abort
= ocfs2_abort_trigger
,
592 .ot_offset
= offsetof(struct ocfs2_extent_block
, h_check
),
595 static struct ocfs2_triggers rb_triggers
= {
597 .t_frozen
= ocfs2_frozen_trigger
,
598 .t_abort
= ocfs2_abort_trigger
,
600 .ot_offset
= offsetof(struct ocfs2_refcount_block
, rf_check
),
603 static struct ocfs2_triggers gd_triggers
= {
605 .t_frozen
= ocfs2_frozen_trigger
,
606 .t_abort
= ocfs2_abort_trigger
,
608 .ot_offset
= offsetof(struct ocfs2_group_desc
, bg_check
),
611 static struct ocfs2_triggers db_triggers
= {
613 .t_frozen
= ocfs2_db_frozen_trigger
,
614 .t_abort
= ocfs2_abort_trigger
,
618 static struct ocfs2_triggers xb_triggers
= {
620 .t_frozen
= ocfs2_frozen_trigger
,
621 .t_abort
= ocfs2_abort_trigger
,
623 .ot_offset
= offsetof(struct ocfs2_xattr_block
, xb_check
),
626 static struct ocfs2_triggers dq_triggers
= {
628 .t_frozen
= ocfs2_dq_frozen_trigger
,
629 .t_abort
= ocfs2_abort_trigger
,
633 static struct ocfs2_triggers dr_triggers
= {
635 .t_frozen
= ocfs2_frozen_trigger
,
636 .t_abort
= ocfs2_abort_trigger
,
638 .ot_offset
= offsetof(struct ocfs2_dx_root_block
, dr_check
),
641 static struct ocfs2_triggers dl_triggers
= {
643 .t_frozen
= ocfs2_frozen_trigger
,
644 .t_abort
= ocfs2_abort_trigger
,
646 .ot_offset
= offsetof(struct ocfs2_dx_leaf
, dl_check
),
649 static int __ocfs2_journal_access(handle_t
*handle
,
650 struct ocfs2_caching_info
*ci
,
651 struct buffer_head
*bh
,
652 struct ocfs2_triggers
*triggers
,
656 struct ocfs2_super
*osb
=
657 OCFS2_SB(ocfs2_metadata_cache_get_super(ci
));
659 BUG_ON(!ci
|| !ci
->ci_ops
);
663 trace_ocfs2_journal_access(
664 (unsigned long long)ocfs2_metadata_cache_owner(ci
),
665 (unsigned long long)bh
->b_blocknr
, type
, bh
->b_size
);
667 /* we can safely remove this assertion after testing. */
668 if (!buffer_uptodate(bh
)) {
669 mlog(ML_ERROR
, "giving me a buffer that's not uptodate!\n");
670 mlog(ML_ERROR
, "b_blocknr=%llu, b_state=0x%lx\n",
671 (unsigned long long)bh
->b_blocknr
, bh
->b_state
);
675 * A previous transaction with a couple of buffer heads fail
676 * to checkpoint, so all the bhs are marked as BH_Write_EIO.
677 * For current transaction, the bh is just among those error
678 * bhs which previous transaction handle. We can't just clear
679 * its BH_Write_EIO and reuse directly, since other bhs are
680 * not written to disk yet and that will cause metadata
681 * inconsistency. So we should set fs read-only to avoid
684 if (buffer_write_io_error(bh
) && !buffer_uptodate(bh
)) {
686 return ocfs2_error(osb
->sb
, "A previous attempt to "
687 "write this buffer head failed\n");
692 /* Set the current transaction information on the ci so
693 * that the locking code knows whether it can drop it's locks
694 * on this ci or not. We're protected from the commit
695 * thread updating the current transaction id until
696 * ocfs2_commit_trans() because ocfs2_start_trans() took
697 * j_trans_barrier for us. */
698 ocfs2_set_ci_lock_trans(osb
->journal
, ci
);
700 ocfs2_metadata_cache_io_lock(ci
);
702 case OCFS2_JOURNAL_ACCESS_CREATE
:
703 case OCFS2_JOURNAL_ACCESS_WRITE
:
704 status
= jbd2_journal_get_write_access(handle
, bh
);
707 case OCFS2_JOURNAL_ACCESS_UNDO
:
708 status
= jbd2_journal_get_undo_access(handle
, bh
);
713 mlog(ML_ERROR
, "Unknown access type!\n");
715 if (!status
&& ocfs2_meta_ecc(osb
) && triggers
)
716 jbd2_journal_set_triggers(bh
, &triggers
->ot_triggers
);
717 ocfs2_metadata_cache_io_unlock(ci
);
720 mlog(ML_ERROR
, "Error %d getting %d access to buffer!\n",
726 int ocfs2_journal_access_di(handle_t
*handle
, struct ocfs2_caching_info
*ci
,
727 struct buffer_head
*bh
, int type
)
729 return __ocfs2_journal_access(handle
, ci
, bh
, &di_triggers
, type
);
732 int ocfs2_journal_access_eb(handle_t
*handle
, struct ocfs2_caching_info
*ci
,
733 struct buffer_head
*bh
, int type
)
735 return __ocfs2_journal_access(handle
, ci
, bh
, &eb_triggers
, type
);
738 int ocfs2_journal_access_rb(handle_t
*handle
, struct ocfs2_caching_info
*ci
,
739 struct buffer_head
*bh
, int type
)
741 return __ocfs2_journal_access(handle
, ci
, bh
, &rb_triggers
,
745 int ocfs2_journal_access_gd(handle_t
*handle
, struct ocfs2_caching_info
*ci
,
746 struct buffer_head
*bh
, int type
)
748 return __ocfs2_journal_access(handle
, ci
, bh
, &gd_triggers
, type
);
751 int ocfs2_journal_access_db(handle_t
*handle
, struct ocfs2_caching_info
*ci
,
752 struct buffer_head
*bh
, int type
)
754 return __ocfs2_journal_access(handle
, ci
, bh
, &db_triggers
, type
);
757 int ocfs2_journal_access_xb(handle_t
*handle
, struct ocfs2_caching_info
*ci
,
758 struct buffer_head
*bh
, int type
)
760 return __ocfs2_journal_access(handle
, ci
, bh
, &xb_triggers
, type
);
763 int ocfs2_journal_access_dq(handle_t
*handle
, struct ocfs2_caching_info
*ci
,
764 struct buffer_head
*bh
, int type
)
766 return __ocfs2_journal_access(handle
, ci
, bh
, &dq_triggers
, type
);
769 int ocfs2_journal_access_dr(handle_t
*handle
, struct ocfs2_caching_info
*ci
,
770 struct buffer_head
*bh
, int type
)
772 return __ocfs2_journal_access(handle
, ci
, bh
, &dr_triggers
, type
);
775 int ocfs2_journal_access_dl(handle_t
*handle
, struct ocfs2_caching_info
*ci
,
776 struct buffer_head
*bh
, int type
)
778 return __ocfs2_journal_access(handle
, ci
, bh
, &dl_triggers
, type
);
781 int ocfs2_journal_access(handle_t
*handle
, struct ocfs2_caching_info
*ci
,
782 struct buffer_head
*bh
, int type
)
784 return __ocfs2_journal_access(handle
, ci
, bh
, NULL
, type
);
787 void ocfs2_journal_dirty(handle_t
*handle
, struct buffer_head
*bh
)
791 trace_ocfs2_journal_dirty((unsigned long long)bh
->b_blocknr
);
793 status
= jbd2_journal_dirty_metadata(handle
, bh
);
796 if (!is_handle_aborted(handle
)) {
797 journal_t
*journal
= handle
->h_transaction
->t_journal
;
798 struct super_block
*sb
= bh
->b_bdev
->bd_super
;
800 mlog(ML_ERROR
, "jbd2_journal_dirty_metadata failed. "
801 "Aborting transaction and journal.\n");
802 handle
->h_err
= status
;
803 jbd2_journal_abort_handle(handle
);
804 jbd2_journal_abort(journal
, status
);
805 ocfs2_abort(sb
, "Journal already aborted.\n");
810 #define OCFS2_DEFAULT_COMMIT_INTERVAL (HZ * JBD2_DEFAULT_MAX_COMMIT_AGE)
812 void ocfs2_set_journal_params(struct ocfs2_super
*osb
)
814 journal_t
*journal
= osb
->journal
->j_journal
;
815 unsigned long commit_interval
= OCFS2_DEFAULT_COMMIT_INTERVAL
;
817 if (osb
->osb_commit_interval
)
818 commit_interval
= osb
->osb_commit_interval
;
820 write_lock(&journal
->j_state_lock
);
821 journal
->j_commit_interval
= commit_interval
;
822 if (osb
->s_mount_opt
& OCFS2_MOUNT_BARRIER
)
823 journal
->j_flags
|= JBD2_BARRIER
;
825 journal
->j_flags
&= ~JBD2_BARRIER
;
826 write_unlock(&journal
->j_state_lock
);
829 int ocfs2_journal_init(struct ocfs2_journal
*journal
, int *dirty
)
832 struct inode
*inode
= NULL
; /* the journal inode */
833 journal_t
*j_journal
= NULL
;
834 struct ocfs2_dinode
*di
= NULL
;
835 struct buffer_head
*bh
= NULL
;
836 struct ocfs2_super
*osb
;
841 osb
= journal
->j_osb
;
843 /* already have the inode for our journal */
844 inode
= ocfs2_get_system_file_inode(osb
, JOURNAL_SYSTEM_INODE
,
851 if (is_bad_inode(inode
)) {
852 mlog(ML_ERROR
, "access error (bad inode)\n");
859 SET_INODE_JOURNAL(inode
);
860 OCFS2_I(inode
)->ip_open_count
++;
862 /* Skip recovery waits here - journal inode metadata never
863 * changes in a live cluster so it can be considered an
864 * exception to the rule. */
865 status
= ocfs2_inode_lock_full(inode
, &bh
, 1, OCFS2_META_LOCK_RECOVERY
);
867 if (status
!= -ERESTARTSYS
)
868 mlog(ML_ERROR
, "Could not get lock on journal!\n");
873 di
= (struct ocfs2_dinode
*)bh
->b_data
;
875 if (i_size_read(inode
) < OCFS2_MIN_JOURNAL_SIZE
) {
876 mlog(ML_ERROR
, "Journal file size (%lld) is too small!\n",
882 trace_ocfs2_journal_init(i_size_read(inode
),
883 (unsigned long long)inode
->i_blocks
,
884 OCFS2_I(inode
)->ip_clusters
);
886 /* call the kernels journal init function now */
887 j_journal
= jbd2_journal_init_inode(inode
);
888 if (j_journal
== NULL
) {
889 mlog(ML_ERROR
, "Linux journal layer error\n");
894 trace_ocfs2_journal_init_maxlen(j_journal
->j_maxlen
);
896 *dirty
= (le32_to_cpu(di
->id1
.journal1
.ij_flags
) &
897 OCFS2_JOURNAL_DIRTY_FL
);
899 journal
->j_journal
= j_journal
;
900 journal
->j_inode
= inode
;
903 ocfs2_set_journal_params(osb
);
905 journal
->j_state
= OCFS2_JOURNAL_LOADED
;
911 ocfs2_inode_unlock(inode
, 1);
914 OCFS2_I(inode
)->ip_open_count
--;
922 static void ocfs2_bump_recovery_generation(struct ocfs2_dinode
*di
)
924 le32_add_cpu(&(di
->id1
.journal1
.ij_recovery_generation
), 1);
927 static u32
ocfs2_get_recovery_generation(struct ocfs2_dinode
*di
)
929 return le32_to_cpu(di
->id1
.journal1
.ij_recovery_generation
);
932 static int ocfs2_journal_toggle_dirty(struct ocfs2_super
*osb
,
933 int dirty
, int replayed
)
937 struct ocfs2_journal
*journal
= osb
->journal
;
938 struct buffer_head
*bh
= journal
->j_bh
;
939 struct ocfs2_dinode
*fe
;
941 fe
= (struct ocfs2_dinode
*)bh
->b_data
;
943 /* The journal bh on the osb always comes from ocfs2_journal_init()
944 * and was validated there inside ocfs2_inode_lock_full(). It's a
945 * code bug if we mess it up. */
946 BUG_ON(!OCFS2_IS_VALID_DINODE(fe
));
948 flags
= le32_to_cpu(fe
->id1
.journal1
.ij_flags
);
950 flags
|= OCFS2_JOURNAL_DIRTY_FL
;
952 flags
&= ~OCFS2_JOURNAL_DIRTY_FL
;
953 fe
->id1
.journal1
.ij_flags
= cpu_to_le32(flags
);
956 ocfs2_bump_recovery_generation(fe
);
958 ocfs2_compute_meta_ecc(osb
->sb
, bh
->b_data
, &fe
->i_check
);
959 status
= ocfs2_write_block(osb
, bh
, INODE_CACHE(journal
->j_inode
));
967 * If the journal has been kmalloc'd it needs to be freed after this
970 void ocfs2_journal_shutdown(struct ocfs2_super
*osb
)
972 struct ocfs2_journal
*journal
= NULL
;
974 struct inode
*inode
= NULL
;
975 int num_running_trans
= 0;
979 journal
= osb
->journal
;
983 inode
= journal
->j_inode
;
985 if (journal
->j_state
!= OCFS2_JOURNAL_LOADED
)
988 /* need to inc inode use count - jbd2_journal_destroy will iput. */
992 num_running_trans
= atomic_read(&(osb
->journal
->j_num_trans
));
993 trace_ocfs2_journal_shutdown(num_running_trans
);
995 /* Do a commit_cache here. It will flush our journal, *and*
996 * release any locks that are still held.
997 * set the SHUTDOWN flag and release the trans lock.
998 * the commit thread will take the trans lock for us below. */
999 journal
->j_state
= OCFS2_JOURNAL_IN_SHUTDOWN
;
1001 /* The OCFS2_JOURNAL_IN_SHUTDOWN will signal to commit_cache to not
1002 * drop the trans_lock (which we want to hold until we
1003 * completely destroy the journal. */
1004 if (osb
->commit_task
) {
1005 /* Wait for the commit thread */
1006 trace_ocfs2_journal_shutdown_wait(osb
->commit_task
);
1007 kthread_stop(osb
->commit_task
);
1008 osb
->commit_task
= NULL
;
1011 BUG_ON(atomic_read(&(osb
->journal
->j_num_trans
)) != 0);
1013 if (ocfs2_mount_local(osb
)) {
1014 jbd2_journal_lock_updates(journal
->j_journal
);
1015 status
= jbd2_journal_flush(journal
->j_journal
);
1016 jbd2_journal_unlock_updates(journal
->j_journal
);
1021 /* Shutdown the kernel journal system */
1022 if (!jbd2_journal_destroy(journal
->j_journal
) && !status
) {
1024 * Do not toggle if flush was unsuccessful otherwise
1025 * will leave dirty metadata in a "clean" journal
1027 status
= ocfs2_journal_toggle_dirty(osb
, 0, 0);
1031 journal
->j_journal
= NULL
;
1033 OCFS2_I(inode
)->ip_open_count
--;
1035 /* unlock our journal */
1036 ocfs2_inode_unlock(inode
, 1);
1038 brelse(journal
->j_bh
);
1039 journal
->j_bh
= NULL
;
1041 journal
->j_state
= OCFS2_JOURNAL_FREE
;
1043 // up_write(&journal->j_trans_barrier);
1048 static void ocfs2_clear_journal_error(struct super_block
*sb
,
1054 olderr
= jbd2_journal_errno(journal
);
1056 mlog(ML_ERROR
, "File system error %d recorded in "
1057 "journal %u.\n", olderr
, slot
);
1058 mlog(ML_ERROR
, "File system on device %s needs checking.\n",
1061 jbd2_journal_ack_err(journal
);
1062 jbd2_journal_clear_err(journal
);
1066 int ocfs2_journal_load(struct ocfs2_journal
*journal
, int local
, int replayed
)
1069 struct ocfs2_super
*osb
;
1073 osb
= journal
->j_osb
;
1075 status
= jbd2_journal_load(journal
->j_journal
);
1077 mlog(ML_ERROR
, "Failed to load journal!\n");
1081 ocfs2_clear_journal_error(osb
->sb
, journal
->j_journal
, osb
->slot_num
);
1084 jbd2_journal_lock_updates(journal
->j_journal
);
1085 status
= jbd2_journal_flush(journal
->j_journal
);
1086 jbd2_journal_unlock_updates(journal
->j_journal
);
1091 status
= ocfs2_journal_toggle_dirty(osb
, 1, replayed
);
1097 /* Launch the commit thread */
1099 osb
->commit_task
= kthread_run(ocfs2_commit_thread
, osb
,
1100 "ocfs2cmt-%s", osb
->uuid_str
);
1101 if (IS_ERR(osb
->commit_task
)) {
1102 status
= PTR_ERR(osb
->commit_task
);
1103 osb
->commit_task
= NULL
;
1104 mlog(ML_ERROR
, "unable to launch ocfs2commit thread, "
1105 "error=%d", status
);
1109 osb
->commit_task
= NULL
;
1116 /* 'full' flag tells us whether we clear out all blocks or if we just
1117 * mark the journal clean */
1118 int ocfs2_journal_wipe(struct ocfs2_journal
*journal
, int full
)
1124 status
= jbd2_journal_wipe(journal
->j_journal
, full
);
1130 status
= ocfs2_journal_toggle_dirty(journal
->j_osb
, 0, 0);
1138 static int ocfs2_recovery_completed(struct ocfs2_super
*osb
)
1141 struct ocfs2_recovery_map
*rm
= osb
->recovery_map
;
1143 spin_lock(&osb
->osb_lock
);
1144 empty
= (rm
->rm_used
== 0);
1145 spin_unlock(&osb
->osb_lock
);
1150 void ocfs2_wait_for_recovery(struct ocfs2_super
*osb
)
1152 wait_event(osb
->recovery_event
, ocfs2_recovery_completed(osb
));
1156 * JBD Might read a cached version of another nodes journal file. We
1157 * don't want this as this file changes often and we get no
1158 * notification on those changes. The only way to be sure that we've
1159 * got the most up to date version of those blocks then is to force
1160 * read them off disk. Just searching through the buffer cache won't
1161 * work as there may be pages backing this file which are still marked
1162 * up to date. We know things can't change on this file underneath us
1163 * as we have the lock by now :)
1165 static int ocfs2_force_read_journal(struct inode
*inode
)
1169 u64 v_blkno
, p_blkno
, p_blocks
, num_blocks
;
1170 struct buffer_head
*bh
= NULL
;
1171 struct ocfs2_super
*osb
= OCFS2_SB(inode
->i_sb
);
1173 num_blocks
= ocfs2_blocks_for_bytes(inode
->i_sb
, i_size_read(inode
));
1175 while (v_blkno
< num_blocks
) {
1176 status
= ocfs2_extent_map_get_blocks(inode
, v_blkno
,
1177 &p_blkno
, &p_blocks
, NULL
);
1183 for (i
= 0; i
< p_blocks
; i
++, p_blkno
++) {
1184 bh
= __find_get_block(osb
->sb
->s_bdev
, p_blkno
,
1185 osb
->sb
->s_blocksize
);
1186 /* block not cached. */
1192 /* We are reading journal data which should not
1193 * be put in the uptodate cache.
1195 status
= ocfs2_read_blocks_sync(osb
, p_blkno
, 1, &bh
);
1205 v_blkno
+= p_blocks
;
1212 struct ocfs2_la_recovery_item
{
1213 struct list_head lri_list
;
1215 struct ocfs2_dinode
*lri_la_dinode
;
1216 struct ocfs2_dinode
*lri_tl_dinode
;
1217 struct ocfs2_quota_recovery
*lri_qrec
;
1218 enum ocfs2_orphan_reco_type lri_orphan_reco_type
;
1221 /* Does the second half of the recovery process. By this point, the
1222 * node is marked clean and can actually be considered recovered,
1223 * hence it's no longer in the recovery map, but there's still some
1224 * cleanup we can do which shouldn't happen within the recovery thread
1225 * as locking in that context becomes very difficult if we are to take
1226 * recovering nodes into account.
1228 * NOTE: This function can and will sleep on recovery of other nodes
1229 * during cluster locking, just like any other ocfs2 process.
1231 void ocfs2_complete_recovery(struct work_struct
*work
)
1234 struct ocfs2_journal
*journal
=
1235 container_of(work
, struct ocfs2_journal
, j_recovery_work
);
1236 struct ocfs2_super
*osb
= journal
->j_osb
;
1237 struct ocfs2_dinode
*la_dinode
, *tl_dinode
;
1238 struct ocfs2_la_recovery_item
*item
, *n
;
1239 struct ocfs2_quota_recovery
*qrec
;
1240 enum ocfs2_orphan_reco_type orphan_reco_type
;
1241 LIST_HEAD(tmp_la_list
);
1243 trace_ocfs2_complete_recovery(
1244 (unsigned long long)OCFS2_I(journal
->j_inode
)->ip_blkno
);
1246 spin_lock(&journal
->j_lock
);
1247 list_splice_init(&journal
->j_la_cleanups
, &tmp_la_list
);
1248 spin_unlock(&journal
->j_lock
);
1250 list_for_each_entry_safe(item
, n
, &tmp_la_list
, lri_list
) {
1251 list_del_init(&item
->lri_list
);
1253 ocfs2_wait_on_quotas(osb
);
1255 la_dinode
= item
->lri_la_dinode
;
1256 tl_dinode
= item
->lri_tl_dinode
;
1257 qrec
= item
->lri_qrec
;
1258 orphan_reco_type
= item
->lri_orphan_reco_type
;
1260 trace_ocfs2_complete_recovery_slot(item
->lri_slot
,
1261 la_dinode
? le64_to_cpu(la_dinode
->i_blkno
) : 0,
1262 tl_dinode
? le64_to_cpu(tl_dinode
->i_blkno
) : 0,
1266 ret
= ocfs2_complete_local_alloc_recovery(osb
,
1275 ret
= ocfs2_complete_truncate_log_recovery(osb
,
1283 ret
= ocfs2_recover_orphans(osb
, item
->lri_slot
,
1289 ret
= ocfs2_finish_quota_recovery(osb
, qrec
,
1293 /* Recovery info is already freed now */
1299 trace_ocfs2_complete_recovery_end(ret
);
1302 /* NOTE: This function always eats your references to la_dinode and
1303 * tl_dinode, either manually on error, or by passing them to
1304 * ocfs2_complete_recovery */
1305 static void ocfs2_queue_recovery_completion(struct ocfs2_journal
*journal
,
1307 struct ocfs2_dinode
*la_dinode
,
1308 struct ocfs2_dinode
*tl_dinode
,
1309 struct ocfs2_quota_recovery
*qrec
,
1310 enum ocfs2_orphan_reco_type orphan_reco_type
)
1312 struct ocfs2_la_recovery_item
*item
;
1314 item
= kmalloc(sizeof(struct ocfs2_la_recovery_item
), GFP_NOFS
);
1316 /* Though we wish to avoid it, we are in fact safe in
1317 * skipping local alloc cleanup as fsck.ocfs2 is more
1318 * than capable of reclaiming unused space. */
1323 ocfs2_free_quota_recovery(qrec
);
1325 mlog_errno(-ENOMEM
);
1329 INIT_LIST_HEAD(&item
->lri_list
);
1330 item
->lri_la_dinode
= la_dinode
;
1331 item
->lri_slot
= slot_num
;
1332 item
->lri_tl_dinode
= tl_dinode
;
1333 item
->lri_qrec
= qrec
;
1334 item
->lri_orphan_reco_type
= orphan_reco_type
;
1336 spin_lock(&journal
->j_lock
);
1337 list_add_tail(&item
->lri_list
, &journal
->j_la_cleanups
);
1338 queue_work(journal
->j_osb
->ocfs2_wq
, &journal
->j_recovery_work
);
1339 spin_unlock(&journal
->j_lock
);
1342 /* Called by the mount code to queue recovery the last part of
1343 * recovery for it's own and offline slot(s). */
1344 void ocfs2_complete_mount_recovery(struct ocfs2_super
*osb
)
1346 struct ocfs2_journal
*journal
= osb
->journal
;
1348 if (ocfs2_is_hard_readonly(osb
))
1351 /* No need to queue up our truncate_log as regular cleanup will catch
1353 ocfs2_queue_recovery_completion(journal
, osb
->slot_num
,
1354 osb
->local_alloc_copy
, NULL
, NULL
,
1355 ORPHAN_NEED_TRUNCATE
);
1356 ocfs2_schedule_truncate_log_flush(osb
, 0);
1358 osb
->local_alloc_copy
= NULL
;
1360 /* queue to recover orphan slots for all offline slots */
1361 ocfs2_replay_map_set_state(osb
, REPLAY_NEEDED
);
1362 ocfs2_queue_replay_slots(osb
, ORPHAN_NEED_TRUNCATE
);
1363 ocfs2_free_replay_slots(osb
);
1366 void ocfs2_complete_quota_recovery(struct ocfs2_super
*osb
)
1368 if (osb
->quota_rec
) {
1369 ocfs2_queue_recovery_completion(osb
->journal
,
1374 ORPHAN_NEED_TRUNCATE
);
1375 osb
->quota_rec
= NULL
;
1379 static int __ocfs2_recovery_thread(void *arg
)
1381 int status
, node_num
, slot_num
;
1382 struct ocfs2_super
*osb
= arg
;
1383 struct ocfs2_recovery_map
*rm
= osb
->recovery_map
;
1384 int *rm_quota
= NULL
;
1385 int rm_quota_used
= 0, i
;
1386 struct ocfs2_quota_recovery
*qrec
;
1388 status
= ocfs2_wait_on_mount(osb
);
1393 rm_quota
= kzalloc(osb
->max_slots
* sizeof(int), GFP_NOFS
);
1399 status
= ocfs2_super_lock(osb
, 1);
1405 status
= ocfs2_compute_replay_slots(osb
);
1409 /* queue recovery for our own slot */
1410 ocfs2_queue_recovery_completion(osb
->journal
, osb
->slot_num
, NULL
,
1411 NULL
, NULL
, ORPHAN_NO_NEED_TRUNCATE
);
1413 spin_lock(&osb
->osb_lock
);
1414 while (rm
->rm_used
) {
1415 /* It's always safe to remove entry zero, as we won't
1416 * clear it until ocfs2_recover_node() has succeeded. */
1417 node_num
= rm
->rm_entries
[0];
1418 spin_unlock(&osb
->osb_lock
);
1419 slot_num
= ocfs2_node_num_to_slot(osb
, node_num
);
1420 trace_ocfs2_recovery_thread_node(node_num
, slot_num
);
1421 if (slot_num
== -ENOENT
) {
1426 /* It is a bit subtle with quota recovery. We cannot do it
1427 * immediately because we have to obtain cluster locks from
1428 * quota files and we also don't want to just skip it because
1429 * then quota usage would be out of sync until some node takes
1430 * the slot. So we remember which nodes need quota recovery
1431 * and when everything else is done, we recover quotas. */
1432 for (i
= 0; i
< rm_quota_used
&& rm_quota
[i
] != slot_num
; i
++);
1433 if (i
== rm_quota_used
)
1434 rm_quota
[rm_quota_used
++] = slot_num
;
1436 status
= ocfs2_recover_node(osb
, node_num
, slot_num
);
1439 ocfs2_recovery_map_clear(osb
, node_num
);
1442 "Error %d recovering node %d on device (%u,%u)!\n",
1444 MAJOR(osb
->sb
->s_dev
), MINOR(osb
->sb
->s_dev
));
1445 mlog(ML_ERROR
, "Volume requires unmount.\n");
1448 spin_lock(&osb
->osb_lock
);
1450 spin_unlock(&osb
->osb_lock
);
1451 trace_ocfs2_recovery_thread_end(status
);
1453 /* Refresh all journal recovery generations from disk */
1454 status
= ocfs2_check_journals_nolocks(osb
);
1455 status
= (status
== -EROFS
) ? 0 : status
;
1459 /* Now it is right time to recover quotas... We have to do this under
1460 * superblock lock so that no one can start using the slot (and crash)
1461 * before we recover it */
1462 for (i
= 0; i
< rm_quota_used
; i
++) {
1463 qrec
= ocfs2_begin_quota_recovery(osb
, rm_quota
[i
]);
1465 status
= PTR_ERR(qrec
);
1469 ocfs2_queue_recovery_completion(osb
->journal
, rm_quota
[i
],
1471 ORPHAN_NEED_TRUNCATE
);
1474 ocfs2_super_unlock(osb
, 1);
1476 /* queue recovery for offline slots */
1477 ocfs2_queue_replay_slots(osb
, ORPHAN_NEED_TRUNCATE
);
1480 mutex_lock(&osb
->recovery_lock
);
1481 if (!status
&& !ocfs2_recovery_completed(osb
)) {
1482 mutex_unlock(&osb
->recovery_lock
);
1486 ocfs2_free_replay_slots(osb
);
1487 osb
->recovery_thread_task
= NULL
;
1488 mb(); /* sync with ocfs2_recovery_thread_running */
1489 wake_up(&osb
->recovery_event
);
1491 mutex_unlock(&osb
->recovery_lock
);
1495 /* no one is callint kthread_stop() for us so the kthread() api
1496 * requires that we call do_exit(). And it isn't exported, but
1497 * complete_and_exit() seems to be a minimal wrapper around it. */
1498 complete_and_exit(NULL
, status
);
1501 void ocfs2_recovery_thread(struct ocfs2_super
*osb
, int node_num
)
1503 mutex_lock(&osb
->recovery_lock
);
1505 trace_ocfs2_recovery_thread(node_num
, osb
->node_num
,
1506 osb
->disable_recovery
, osb
->recovery_thread_task
,
1507 osb
->disable_recovery
?
1508 -1 : ocfs2_recovery_map_set(osb
, node_num
));
1510 if (osb
->disable_recovery
)
1513 if (osb
->recovery_thread_task
)
1516 osb
->recovery_thread_task
= kthread_run(__ocfs2_recovery_thread
, osb
,
1517 "ocfs2rec-%s", osb
->uuid_str
);
1518 if (IS_ERR(osb
->recovery_thread_task
)) {
1519 mlog_errno((int)PTR_ERR(osb
->recovery_thread_task
));
1520 osb
->recovery_thread_task
= NULL
;
1524 mutex_unlock(&osb
->recovery_lock
);
1525 wake_up(&osb
->recovery_event
);
1528 static int ocfs2_read_journal_inode(struct ocfs2_super
*osb
,
1530 struct buffer_head
**bh
,
1531 struct inode
**ret_inode
)
1533 int status
= -EACCES
;
1534 struct inode
*inode
= NULL
;
1536 BUG_ON(slot_num
>= osb
->max_slots
);
1538 inode
= ocfs2_get_system_file_inode(osb
, JOURNAL_SYSTEM_INODE
,
1540 if (!inode
|| is_bad_inode(inode
)) {
1544 SET_INODE_JOURNAL(inode
);
1546 status
= ocfs2_read_inode_block_full(inode
, bh
, OCFS2_BH_IGNORE_CACHE
);
1556 if (status
|| !ret_inode
)
1564 /* Does the actual journal replay and marks the journal inode as
1565 * clean. Will only replay if the journal inode is marked dirty. */
1566 static int ocfs2_replay_journal(struct ocfs2_super
*osb
,
1573 struct inode
*inode
= NULL
;
1574 struct ocfs2_dinode
*fe
;
1575 journal_t
*journal
= NULL
;
1576 struct buffer_head
*bh
= NULL
;
1579 status
= ocfs2_read_journal_inode(osb
, slot_num
, &bh
, &inode
);
1585 fe
= (struct ocfs2_dinode
*)bh
->b_data
;
1586 slot_reco_gen
= ocfs2_get_recovery_generation(fe
);
1591 * As the fs recovery is asynchronous, there is a small chance that
1592 * another node mounted (and recovered) the slot before the recovery
1593 * thread could get the lock. To handle that, we dirty read the journal
1594 * inode for that slot to get the recovery generation. If it is
1595 * different than what we expected, the slot has been recovered.
1596 * If not, it needs recovery.
1598 if (osb
->slot_recovery_generations
[slot_num
] != slot_reco_gen
) {
1599 trace_ocfs2_replay_journal_recovered(slot_num
,
1600 osb
->slot_recovery_generations
[slot_num
], slot_reco_gen
);
1601 osb
->slot_recovery_generations
[slot_num
] = slot_reco_gen
;
1606 /* Continue with recovery as the journal has not yet been recovered */
1608 status
= ocfs2_inode_lock_full(inode
, &bh
, 1, OCFS2_META_LOCK_RECOVERY
);
1610 trace_ocfs2_replay_journal_lock_err(status
);
1611 if (status
!= -ERESTARTSYS
)
1612 mlog(ML_ERROR
, "Could not lock journal!\n");
1617 fe
= (struct ocfs2_dinode
*) bh
->b_data
;
1619 flags
= le32_to_cpu(fe
->id1
.journal1
.ij_flags
);
1620 slot_reco_gen
= ocfs2_get_recovery_generation(fe
);
1622 if (!(flags
& OCFS2_JOURNAL_DIRTY_FL
)) {
1623 trace_ocfs2_replay_journal_skip(node_num
);
1624 /* Refresh recovery generation for the slot */
1625 osb
->slot_recovery_generations
[slot_num
] = slot_reco_gen
;
1629 /* we need to run complete recovery for offline orphan slots */
1630 ocfs2_replay_map_set_state(osb
, REPLAY_NEEDED
);
1632 printk(KERN_NOTICE
"ocfs2: Begin replay journal (node %d, slot %d) on "\
1633 "device (%u,%u)\n", node_num
, slot_num
, MAJOR(osb
->sb
->s_dev
),
1634 MINOR(osb
->sb
->s_dev
));
1636 OCFS2_I(inode
)->ip_clusters
= le32_to_cpu(fe
->i_clusters
);
1638 status
= ocfs2_force_read_journal(inode
);
1644 journal
= jbd2_journal_init_inode(inode
);
1645 if (journal
== NULL
) {
1646 mlog(ML_ERROR
, "Linux journal layer error\n");
1651 status
= jbd2_journal_load(journal
);
1656 jbd2_journal_destroy(journal
);
1660 ocfs2_clear_journal_error(osb
->sb
, journal
, slot_num
);
1662 /* wipe the journal */
1663 jbd2_journal_lock_updates(journal
);
1664 status
= jbd2_journal_flush(journal
);
1665 jbd2_journal_unlock_updates(journal
);
1669 /* This will mark the node clean */
1670 flags
= le32_to_cpu(fe
->id1
.journal1
.ij_flags
);
1671 flags
&= ~OCFS2_JOURNAL_DIRTY_FL
;
1672 fe
->id1
.journal1
.ij_flags
= cpu_to_le32(flags
);
1674 /* Increment recovery generation to indicate successful recovery */
1675 ocfs2_bump_recovery_generation(fe
);
1676 osb
->slot_recovery_generations
[slot_num
] =
1677 ocfs2_get_recovery_generation(fe
);
1679 ocfs2_compute_meta_ecc(osb
->sb
, bh
->b_data
, &fe
->i_check
);
1680 status
= ocfs2_write_block(osb
, bh
, INODE_CACHE(inode
));
1687 jbd2_journal_destroy(journal
);
1689 printk(KERN_NOTICE
"ocfs2: End replay journal (node %d, slot %d) on "\
1690 "device (%u,%u)\n", node_num
, slot_num
, MAJOR(osb
->sb
->s_dev
),
1691 MINOR(osb
->sb
->s_dev
));
1693 /* drop the lock on this nodes journal */
1695 ocfs2_inode_unlock(inode
, 1);
1704 * Do the most important parts of node recovery:
1705 * - Replay it's journal
1706 * - Stamp a clean local allocator file
1707 * - Stamp a clean truncate log
1708 * - Mark the node clean
1710 * If this function completes without error, a node in OCFS2 can be
1711 * said to have been safely recovered. As a result, failure during the
1712 * second part of a nodes recovery process (local alloc recovery) is
1713 * far less concerning.
1715 static int ocfs2_recover_node(struct ocfs2_super
*osb
,
1716 int node_num
, int slot_num
)
1719 struct ocfs2_dinode
*la_copy
= NULL
;
1720 struct ocfs2_dinode
*tl_copy
= NULL
;
1722 trace_ocfs2_recover_node(node_num
, slot_num
, osb
->node_num
);
1724 /* Should not ever be called to recover ourselves -- in that
1725 * case we should've called ocfs2_journal_load instead. */
1726 BUG_ON(osb
->node_num
== node_num
);
1728 status
= ocfs2_replay_journal(osb
, node_num
, slot_num
);
1730 if (status
== -EBUSY
) {
1731 trace_ocfs2_recover_node_skip(slot_num
, node_num
);
1739 /* Stamp a clean local alloc file AFTER recovering the journal... */
1740 status
= ocfs2_begin_local_alloc_recovery(osb
, slot_num
, &la_copy
);
1746 /* An error from begin_truncate_log_recovery is not
1747 * serious enough to warrant halting the rest of
1749 status
= ocfs2_begin_truncate_log_recovery(osb
, slot_num
, &tl_copy
);
1753 /* Likewise, this would be a strange but ultimately not so
1754 * harmful place to get an error... */
1755 status
= ocfs2_clear_slot(osb
, slot_num
);
1759 /* This will kfree the memory pointed to by la_copy and tl_copy */
1760 ocfs2_queue_recovery_completion(osb
->journal
, slot_num
, la_copy
,
1761 tl_copy
, NULL
, ORPHAN_NEED_TRUNCATE
);
1769 /* Test node liveness by trylocking his journal. If we get the lock,
1770 * we drop it here. Return 0 if we got the lock, -EAGAIN if node is
1771 * still alive (we couldn't get the lock) and < 0 on error. */
1772 static int ocfs2_trylock_journal(struct ocfs2_super
*osb
,
1776 struct inode
*inode
= NULL
;
1778 inode
= ocfs2_get_system_file_inode(osb
, JOURNAL_SYSTEM_INODE
,
1780 if (inode
== NULL
) {
1781 mlog(ML_ERROR
, "access error\n");
1785 if (is_bad_inode(inode
)) {
1786 mlog(ML_ERROR
, "access error (bad inode)\n");
1792 SET_INODE_JOURNAL(inode
);
1794 flags
= OCFS2_META_LOCK_RECOVERY
| OCFS2_META_LOCK_NOQUEUE
;
1795 status
= ocfs2_inode_lock_full(inode
, NULL
, 1, flags
);
1797 if (status
!= -EAGAIN
)
1802 ocfs2_inode_unlock(inode
, 1);
1809 /* Call this underneath ocfs2_super_lock. It also assumes that the
1810 * slot info struct has been updated from disk. */
1811 int ocfs2_mark_dead_nodes(struct ocfs2_super
*osb
)
1813 unsigned int node_num
;
1816 struct buffer_head
*bh
= NULL
;
1817 struct ocfs2_dinode
*di
;
1819 /* This is called with the super block cluster lock, so we
1820 * know that the slot map can't change underneath us. */
1822 for (i
= 0; i
< osb
->max_slots
; i
++) {
1823 /* Read journal inode to get the recovery generation */
1824 status
= ocfs2_read_journal_inode(osb
, i
, &bh
, NULL
);
1829 di
= (struct ocfs2_dinode
*)bh
->b_data
;
1830 gen
= ocfs2_get_recovery_generation(di
);
1834 spin_lock(&osb
->osb_lock
);
1835 osb
->slot_recovery_generations
[i
] = gen
;
1837 trace_ocfs2_mark_dead_nodes(i
,
1838 osb
->slot_recovery_generations
[i
]);
1840 if (i
== osb
->slot_num
) {
1841 spin_unlock(&osb
->osb_lock
);
1845 status
= ocfs2_slot_to_node_num_locked(osb
, i
, &node_num
);
1846 if (status
== -ENOENT
) {
1847 spin_unlock(&osb
->osb_lock
);
1851 if (__ocfs2_recovery_map_test(osb
, node_num
)) {
1852 spin_unlock(&osb
->osb_lock
);
1855 spin_unlock(&osb
->osb_lock
);
1857 /* Ok, we have a slot occupied by another node which
1858 * is not in the recovery map. We trylock his journal
1859 * file here to test if he's alive. */
1860 status
= ocfs2_trylock_journal(osb
, i
);
1862 /* Since we're called from mount, we know that
1863 * the recovery thread can't race us on
1864 * setting / checking the recovery bits. */
1865 ocfs2_recovery_thread(osb
, node_num
);
1866 } else if ((status
< 0) && (status
!= -EAGAIN
)) {
1878 * Scan timer should get fired every ORPHAN_SCAN_SCHEDULE_TIMEOUT. Add some
1879 * randomness to the timeout to minimize multple nodes firing the timer at the
1882 static inline unsigned long ocfs2_orphan_scan_timeout(void)
1886 get_random_bytes(&time
, sizeof(time
));
1887 time
= ORPHAN_SCAN_SCHEDULE_TIMEOUT
+ (time
% 5000);
1888 return msecs_to_jiffies(time
);
1892 * ocfs2_queue_orphan_scan calls ocfs2_queue_recovery_completion for
1893 * every slot, queuing a recovery of the slot on the ocfs2_wq thread. This
1894 * is done to catch any orphans that are left over in orphan directories.
1896 * It scans all slots, even ones that are in use. It does so to handle the
1897 * case described below:
1899 * Node 1 has an inode it was using. The dentry went away due to memory
1900 * pressure. Node 1 closes the inode, but it's on the free list. The node
1901 * has the open lock.
1902 * Node 2 unlinks the inode. It grabs the dentry lock to notify others,
1903 * but node 1 has no dentry and doesn't get the message. It trylocks the
1904 * open lock, sees that another node has a PR, and does nothing.
1905 * Later node 2 runs its orphan dir. It igets the inode, trylocks the
1906 * open lock, sees the PR still, and does nothing.
1907 * Basically, we have to trigger an orphan iput on node 1. The only way
1908 * for this to happen is if node 1 runs node 2's orphan dir.
1910 * ocfs2_queue_orphan_scan gets called every ORPHAN_SCAN_SCHEDULE_TIMEOUT
1911 * seconds. It gets an EX lock on os_lockres and checks sequence number
1912 * stored in LVB. If the sequence number has changed, it means some other
1913 * node has done the scan. This node skips the scan and tracks the
1914 * sequence number. If the sequence number didn't change, it means a scan
1915 * hasn't happened. The node queues a scan and increments the
1916 * sequence number in the LVB.
1918 static void ocfs2_queue_orphan_scan(struct ocfs2_super
*osb
)
1920 struct ocfs2_orphan_scan
*os
;
1924 os
= &osb
->osb_orphan_scan
;
1926 if (atomic_read(&os
->os_state
) == ORPHAN_SCAN_INACTIVE
)
1929 trace_ocfs2_queue_orphan_scan_begin(os
->os_count
, os
->os_seqno
,
1930 atomic_read(&os
->os_state
));
1932 status
= ocfs2_orphan_scan_lock(osb
, &seqno
);
1934 if (status
!= -EAGAIN
)
1939 /* Do no queue the tasks if the volume is being umounted */
1940 if (atomic_read(&os
->os_state
) == ORPHAN_SCAN_INACTIVE
)
1943 if (os
->os_seqno
!= seqno
) {
1944 os
->os_seqno
= seqno
;
1948 for (i
= 0; i
< osb
->max_slots
; i
++)
1949 ocfs2_queue_recovery_completion(osb
->journal
, i
, NULL
, NULL
,
1950 NULL
, ORPHAN_NO_NEED_TRUNCATE
);
1952 * We queued a recovery on orphan slots, increment the sequence
1953 * number and update LVB so other node will skip the scan for a while
1957 os
->os_scantime
= ktime_get_seconds();
1959 ocfs2_orphan_scan_unlock(osb
, seqno
);
1961 trace_ocfs2_queue_orphan_scan_end(os
->os_count
, os
->os_seqno
,
1962 atomic_read(&os
->os_state
));
1966 /* Worker task that gets fired every ORPHAN_SCAN_SCHEDULE_TIMEOUT millsec */
1967 static void ocfs2_orphan_scan_work(struct work_struct
*work
)
1969 struct ocfs2_orphan_scan
*os
;
1970 struct ocfs2_super
*osb
;
1972 os
= container_of(work
, struct ocfs2_orphan_scan
,
1973 os_orphan_scan_work
.work
);
1976 mutex_lock(&os
->os_lock
);
1977 ocfs2_queue_orphan_scan(osb
);
1978 if (atomic_read(&os
->os_state
) == ORPHAN_SCAN_ACTIVE
)
1979 queue_delayed_work(osb
->ocfs2_wq
, &os
->os_orphan_scan_work
,
1980 ocfs2_orphan_scan_timeout());
1981 mutex_unlock(&os
->os_lock
);
1984 void ocfs2_orphan_scan_stop(struct ocfs2_super
*osb
)
1986 struct ocfs2_orphan_scan
*os
;
1988 os
= &osb
->osb_orphan_scan
;
1989 if (atomic_read(&os
->os_state
) == ORPHAN_SCAN_ACTIVE
) {
1990 atomic_set(&os
->os_state
, ORPHAN_SCAN_INACTIVE
);
1991 mutex_lock(&os
->os_lock
);
1992 cancel_delayed_work(&os
->os_orphan_scan_work
);
1993 mutex_unlock(&os
->os_lock
);
1997 void ocfs2_orphan_scan_init(struct ocfs2_super
*osb
)
1999 struct ocfs2_orphan_scan
*os
;
2001 os
= &osb
->osb_orphan_scan
;
2005 mutex_init(&os
->os_lock
);
2006 INIT_DELAYED_WORK(&os
->os_orphan_scan_work
, ocfs2_orphan_scan_work
);
2009 void ocfs2_orphan_scan_start(struct ocfs2_super
*osb
)
2011 struct ocfs2_orphan_scan
*os
;
2013 os
= &osb
->osb_orphan_scan
;
2014 os
->os_scantime
= ktime_get_seconds();
2015 if (ocfs2_is_hard_readonly(osb
) || ocfs2_mount_local(osb
))
2016 atomic_set(&os
->os_state
, ORPHAN_SCAN_INACTIVE
);
2018 atomic_set(&os
->os_state
, ORPHAN_SCAN_ACTIVE
);
2019 queue_delayed_work(osb
->ocfs2_wq
, &os
->os_orphan_scan_work
,
2020 ocfs2_orphan_scan_timeout());
2024 struct ocfs2_orphan_filldir_priv
{
2025 struct dir_context ctx
;
2027 struct ocfs2_super
*osb
;
2028 enum ocfs2_orphan_reco_type orphan_reco_type
;
2031 static int ocfs2_orphan_filldir(struct dir_context
*ctx
, const char *name
,
2032 int name_len
, loff_t pos
, u64 ino
,
2035 struct ocfs2_orphan_filldir_priv
*p
=
2036 container_of(ctx
, struct ocfs2_orphan_filldir_priv
, ctx
);
2039 if (name_len
== 1 && !strncmp(".", name
, 1))
2041 if (name_len
== 2 && !strncmp("..", name
, 2))
2044 /* do not include dio entry in case of orphan scan */
2045 if ((p
->orphan_reco_type
== ORPHAN_NO_NEED_TRUNCATE
) &&
2046 (!strncmp(name
, OCFS2_DIO_ORPHAN_PREFIX
,
2047 OCFS2_DIO_ORPHAN_PREFIX_LEN
)))
2050 /* Skip bad inodes so that recovery can continue */
2051 iter
= ocfs2_iget(p
->osb
, ino
,
2052 OCFS2_FI_FLAG_ORPHAN_RECOVERY
, 0);
2056 if (!strncmp(name
, OCFS2_DIO_ORPHAN_PREFIX
,
2057 OCFS2_DIO_ORPHAN_PREFIX_LEN
))
2058 OCFS2_I(iter
)->ip_flags
|= OCFS2_INODE_DIO_ORPHAN_ENTRY
;
2060 /* Skip inodes which are already added to recover list, since dio may
2061 * happen concurrently with unlink/rename */
2062 if (OCFS2_I(iter
)->ip_next_orphan
) {
2067 trace_ocfs2_orphan_filldir((unsigned long long)OCFS2_I(iter
)->ip_blkno
);
2068 /* No locking is required for the next_orphan queue as there
2069 * is only ever a single process doing orphan recovery. */
2070 OCFS2_I(iter
)->ip_next_orphan
= p
->head
;
2076 static int ocfs2_queue_orphans(struct ocfs2_super
*osb
,
2078 struct inode
**head
,
2079 enum ocfs2_orphan_reco_type orphan_reco_type
)
2082 struct inode
*orphan_dir_inode
= NULL
;
2083 struct ocfs2_orphan_filldir_priv priv
= {
2084 .ctx
.actor
= ocfs2_orphan_filldir
,
2087 .orphan_reco_type
= orphan_reco_type
2090 orphan_dir_inode
= ocfs2_get_system_file_inode(osb
,
2091 ORPHAN_DIR_SYSTEM_INODE
,
2093 if (!orphan_dir_inode
) {
2099 inode_lock(orphan_dir_inode
);
2100 status
= ocfs2_inode_lock(orphan_dir_inode
, NULL
, 0);
2106 status
= ocfs2_dir_foreach(orphan_dir_inode
, &priv
.ctx
);
2115 ocfs2_inode_unlock(orphan_dir_inode
, 0);
2117 inode_unlock(orphan_dir_inode
);
2118 iput(orphan_dir_inode
);
2122 static int ocfs2_orphan_recovery_can_continue(struct ocfs2_super
*osb
,
2127 spin_lock(&osb
->osb_lock
);
2128 ret
= !osb
->osb_orphan_wipes
[slot
];
2129 spin_unlock(&osb
->osb_lock
);
2133 static void ocfs2_mark_recovering_orphan_dir(struct ocfs2_super
*osb
,
2136 spin_lock(&osb
->osb_lock
);
2137 /* Mark ourselves such that new processes in delete_inode()
2138 * know to quit early. */
2139 ocfs2_node_map_set_bit(osb
, &osb
->osb_recovering_orphan_dirs
, slot
);
2140 while (osb
->osb_orphan_wipes
[slot
]) {
2141 /* If any processes are already in the middle of an
2142 * orphan wipe on this dir, then we need to wait for
2144 spin_unlock(&osb
->osb_lock
);
2145 wait_event_interruptible(osb
->osb_wipe_event
,
2146 ocfs2_orphan_recovery_can_continue(osb
, slot
));
2147 spin_lock(&osb
->osb_lock
);
2149 spin_unlock(&osb
->osb_lock
);
2152 static void ocfs2_clear_recovering_orphan_dir(struct ocfs2_super
*osb
,
2155 ocfs2_node_map_clear_bit(osb
, &osb
->osb_recovering_orphan_dirs
, slot
);
2159 * Orphan recovery. Each mounted node has it's own orphan dir which we
2160 * must run during recovery. Our strategy here is to build a list of
2161 * the inodes in the orphan dir and iget/iput them. The VFS does
2162 * (most) of the rest of the work.
2164 * Orphan recovery can happen at any time, not just mount so we have a
2165 * couple of extra considerations.
2167 * - We grab as many inodes as we can under the orphan dir lock -
2168 * doing iget() outside the orphan dir risks getting a reference on
2170 * - We must be sure not to deadlock with other processes on the
2171 * system wanting to run delete_inode(). This can happen when they go
2172 * to lock the orphan dir and the orphan recovery process attempts to
2173 * iget() inside the orphan dir lock. This can be avoided by
2174 * advertising our state to ocfs2_delete_inode().
2176 static int ocfs2_recover_orphans(struct ocfs2_super
*osb
,
2178 enum ocfs2_orphan_reco_type orphan_reco_type
)
2181 struct inode
*inode
= NULL
;
2183 struct ocfs2_inode_info
*oi
;
2184 struct buffer_head
*di_bh
= NULL
;
2185 struct ocfs2_dinode
*di
= NULL
;
2187 trace_ocfs2_recover_orphans(slot
);
2189 ocfs2_mark_recovering_orphan_dir(osb
, slot
);
2190 ret
= ocfs2_queue_orphans(osb
, slot
, &inode
, orphan_reco_type
);
2191 ocfs2_clear_recovering_orphan_dir(osb
, slot
);
2193 /* Error here should be noted, but we want to continue with as
2194 * many queued inodes as we've got. */
2199 oi
= OCFS2_I(inode
);
2200 trace_ocfs2_recover_orphans_iput(
2201 (unsigned long long)oi
->ip_blkno
);
2203 iter
= oi
->ip_next_orphan
;
2204 oi
->ip_next_orphan
= NULL
;
2206 if (oi
->ip_flags
& OCFS2_INODE_DIO_ORPHAN_ENTRY
) {
2208 ret
= ocfs2_rw_lock(inode
, 1);
2214 * We need to take and drop the inode lock to
2215 * force read inode from disk.
2217 ret
= ocfs2_inode_lock(inode
, &di_bh
, 1);
2223 di
= (struct ocfs2_dinode
*)di_bh
->b_data
;
2225 if (di
->i_flags
& cpu_to_le32(OCFS2_DIO_ORPHANED_FL
)) {
2226 ret
= ocfs2_truncate_file(inode
, di_bh
,
2227 i_size_read(inode
));
2234 ret
= ocfs2_del_inode_from_orphan(osb
, inode
,
2240 ocfs2_inode_unlock(inode
, 1);
2244 ocfs2_rw_unlock(inode
, 1);
2246 inode_unlock(inode
);
2248 /* clear dio flag in ocfs2_inode_info */
2249 oi
->ip_flags
&= ~OCFS2_INODE_DIO_ORPHAN_ENTRY
;
2251 spin_lock(&oi
->ip_lock
);
2252 /* Set the proper information to get us going into
2253 * ocfs2_delete_inode. */
2254 oi
->ip_flags
|= OCFS2_INODE_MAYBE_ORPHANED
;
2255 spin_unlock(&oi
->ip_lock
);
2265 static int __ocfs2_wait_on_mount(struct ocfs2_super
*osb
, int quota
)
2267 /* This check is good because ocfs2 will wait on our recovery
2268 * thread before changing it to something other than MOUNTED
2270 wait_event(osb
->osb_mount_event
,
2271 (!quota
&& atomic_read(&osb
->vol_state
) == VOLUME_MOUNTED
) ||
2272 atomic_read(&osb
->vol_state
) == VOLUME_MOUNTED_QUOTAS
||
2273 atomic_read(&osb
->vol_state
) == VOLUME_DISABLED
);
2275 /* If there's an error on mount, then we may never get to the
2276 * MOUNTED flag, but this is set right before
2277 * dismount_volume() so we can trust it. */
2278 if (atomic_read(&osb
->vol_state
) == VOLUME_DISABLED
) {
2279 trace_ocfs2_wait_on_mount(VOLUME_DISABLED
);
2280 mlog(0, "mount error, exiting!\n");
2287 static int ocfs2_commit_thread(void *arg
)
2290 struct ocfs2_super
*osb
= arg
;
2291 struct ocfs2_journal
*journal
= osb
->journal
;
2293 /* we can trust j_num_trans here because _should_stop() is only set in
2294 * shutdown and nobody other than ourselves should be able to start
2295 * transactions. committing on shutdown might take a few iterations
2296 * as final transactions put deleted inodes on the list */
2297 while (!(kthread_should_stop() &&
2298 atomic_read(&journal
->j_num_trans
) == 0)) {
2300 wait_event_interruptible(osb
->checkpoint_event
,
2301 atomic_read(&journal
->j_num_trans
)
2302 || kthread_should_stop());
2304 status
= ocfs2_commit_cache(osb
);
2306 static unsigned long abort_warn_time
;
2308 /* Warn about this once per minute */
2309 if (printk_timed_ratelimit(&abort_warn_time
, 60*HZ
))
2310 mlog(ML_ERROR
, "status = %d, journal is "
2311 "already aborted.\n", status
);
2313 * After ocfs2_commit_cache() fails, j_num_trans has a
2314 * non-zero value. Sleep here to avoid a busy-wait
2317 msleep_interruptible(1000);
2320 if (kthread_should_stop() && atomic_read(&journal
->j_num_trans
)){
2322 "commit_thread: %u transactions pending on "
2324 atomic_read(&journal
->j_num_trans
));
2331 /* Reads all the journal inodes without taking any cluster locks. Used
2332 * for hard readonly access to determine whether any journal requires
2333 * recovery. Also used to refresh the recovery generation numbers after
2334 * a journal has been recovered by another node.
2336 int ocfs2_check_journals_nolocks(struct ocfs2_super
*osb
)
2340 struct buffer_head
*di_bh
= NULL
;
2341 struct ocfs2_dinode
*di
;
2342 int journal_dirty
= 0;
2344 for(slot
= 0; slot
< osb
->max_slots
; slot
++) {
2345 ret
= ocfs2_read_journal_inode(osb
, slot
, &di_bh
, NULL
);
2351 di
= (struct ocfs2_dinode
*) di_bh
->b_data
;
2353 osb
->slot_recovery_generations
[slot
] =
2354 ocfs2_get_recovery_generation(di
);
2356 if (le32_to_cpu(di
->id1
.journal1
.ij_flags
) &
2357 OCFS2_JOURNAL_DIRTY_FL
)