2 * linux/fs/jbd2/transaction.c
4 * Written by Stephen C. Tweedie <sct@redhat.com>, 1998
6 * Copyright 1998 Red Hat corp --- All Rights Reserved
8 * This file is part of the Linux kernel and is made available under
9 * the terms of the GNU General Public License, version 2, or at your
10 * option, any later version, incorporated herein by reference.
12 * Generic filesystem transaction handling code; part of the ext2fs
15 * This file manages transactions (compound commits managed by the
16 * journaling code) and handles (individual atomic operations by the
20 #include <linux/time.h>
22 #include <linux/jbd2.h>
23 #include <linux/errno.h>
24 #include <linux/slab.h>
25 #include <linux/timer.h>
27 #include <linux/highmem.h>
28 #include <linux/hrtimer.h>
29 #include <linux/backing-dev.h>
30 #include <linux/bug.h>
31 #include <linux/module.h>
33 #include <trace/events/jbd2.h>
35 static void __jbd2_journal_temp_unlink_buffer(struct journal_head
*jh
);
36 static void __jbd2_journal_unfile_buffer(struct journal_head
*jh
);
38 static struct kmem_cache
*transaction_cache
;
39 int __init
jbd2_journal_init_transaction_cache(void)
41 J_ASSERT(!transaction_cache
);
42 transaction_cache
= kmem_cache_create("jbd2_transaction_s",
43 sizeof(transaction_t
),
45 SLAB_HWCACHE_ALIGN
|SLAB_TEMPORARY
,
47 if (transaction_cache
)
52 void jbd2_journal_destroy_transaction_cache(void)
54 if (transaction_cache
) {
55 kmem_cache_destroy(transaction_cache
);
56 transaction_cache
= NULL
;
60 void jbd2_journal_free_transaction(transaction_t
*transaction
)
62 if (unlikely(ZERO_OR_NULL_PTR(transaction
)))
64 kmem_cache_free(transaction_cache
, transaction
);
68 * jbd2_get_transaction: obtain a new transaction_t object.
70 * Simply allocate and initialise a new transaction. Create it in
71 * RUNNING state and add it to the current journal (which should not
72 * have an existing running transaction: we only make a new transaction
73 * once we have started to commit the old one).
76 * The journal MUST be locked. We don't perform atomic mallocs on the
77 * new transaction and we can't block without protecting against other
78 * processes trying to touch the journal while it is in transition.
82 static transaction_t
*
83 jbd2_get_transaction(journal_t
*journal
, transaction_t
*transaction
)
85 transaction
->t_journal
= journal
;
86 transaction
->t_state
= T_RUNNING
;
87 transaction
->t_start_time
= ktime_get();
88 transaction
->t_tid
= journal
->j_transaction_sequence
++;
89 transaction
->t_expires
= jiffies
+ journal
->j_commit_interval
;
90 spin_lock_init(&transaction
->t_handle_lock
);
91 atomic_set(&transaction
->t_updates
, 0);
92 atomic_set(&transaction
->t_outstanding_credits
, 0);
93 atomic_set(&transaction
->t_handle_count
, 0);
94 INIT_LIST_HEAD(&transaction
->t_inode_list
);
95 INIT_LIST_HEAD(&transaction
->t_private_list
);
97 /* Set up the commit timer for the new transaction. */
98 journal
->j_commit_timer
.expires
= round_jiffies_up(transaction
->t_expires
);
99 add_timer(&journal
->j_commit_timer
);
101 J_ASSERT(journal
->j_running_transaction
== NULL
);
102 journal
->j_running_transaction
= transaction
;
103 transaction
->t_max_wait
= 0;
104 transaction
->t_start
= jiffies
;
105 transaction
->t_requested
= 0;
113 * A handle_t is an object which represents a single atomic update to a
114 * filesystem, and which tracks all of the modifications which form part
115 * of that one update.
119 * Update transaction's maximum wait time, if debugging is enabled.
121 * In order for t_max_wait to be reliable, it must be protected by a
122 * lock. But doing so will mean that start_this_handle() can not be
123 * run in parallel on SMP systems, which limits our scalability. So
124 * unless debugging is enabled, we no longer update t_max_wait, which
125 * means that maximum wait time reported by the jbd2_run_stats
126 * tracepoint will always be zero.
128 static inline void update_t_max_wait(transaction_t
*transaction
,
131 #ifdef CONFIG_JBD2_DEBUG
132 if (jbd2_journal_enable_debug
&&
133 time_after(transaction
->t_start
, ts
)) {
134 ts
= jbd2_time_diff(ts
, transaction
->t_start
);
135 spin_lock(&transaction
->t_handle_lock
);
136 if (ts
> transaction
->t_max_wait
)
137 transaction
->t_max_wait
= ts
;
138 spin_unlock(&transaction
->t_handle_lock
);
144 * start_this_handle: Given a handle, deal with any locking or stalling
145 * needed to make sure that there is enough journal space for the handle
146 * to begin. Attach the handle to a transaction and set up the
147 * transaction's buffer credits.
150 static int start_this_handle(journal_t
*journal
, handle_t
*handle
,
153 transaction_t
*transaction
, *new_transaction
= NULL
;
155 int needed
, need_to_start
;
156 int nblocks
= handle
->h_buffer_credits
;
157 unsigned long ts
= jiffies
;
159 if (nblocks
> journal
->j_max_transaction_buffers
) {
160 printk(KERN_ERR
"JBD2: %s wants too many credits (%d > %d)\n",
161 current
->comm
, nblocks
,
162 journal
->j_max_transaction_buffers
);
167 if (!journal
->j_running_transaction
) {
168 new_transaction
= kmem_cache_zalloc(transaction_cache
,
170 if (!new_transaction
) {
172 * If __GFP_FS is not present, then we may be
173 * being called from inside the fs writeback
174 * layer, so we MUST NOT fail. Since
175 * __GFP_NOFAIL is going away, we will arrange
176 * to retry the allocation ourselves.
178 if ((gfp_mask
& __GFP_FS
) == 0) {
179 congestion_wait(BLK_RW_ASYNC
, HZ
/50);
180 goto alloc_transaction
;
186 jbd_debug(3, "New handle %p going live.\n", handle
);
189 * We need to hold j_state_lock until t_updates has been incremented,
190 * for proper journal barrier handling
193 read_lock(&journal
->j_state_lock
);
194 BUG_ON(journal
->j_flags
& JBD2_UNMOUNT
);
195 if (is_journal_aborted(journal
) ||
196 (journal
->j_errno
!= 0 && !(journal
->j_flags
& JBD2_ACK_ERR
))) {
197 read_unlock(&journal
->j_state_lock
);
198 jbd2_journal_free_transaction(new_transaction
);
202 /* Wait on the journal's transaction barrier if necessary */
203 if (journal
->j_barrier_count
) {
204 read_unlock(&journal
->j_state_lock
);
205 wait_event(journal
->j_wait_transaction_locked
,
206 journal
->j_barrier_count
== 0);
210 if (!journal
->j_running_transaction
) {
211 read_unlock(&journal
->j_state_lock
);
212 if (!new_transaction
)
213 goto alloc_transaction
;
214 write_lock(&journal
->j_state_lock
);
215 if (!journal
->j_running_transaction
&&
216 !journal
->j_barrier_count
) {
217 jbd2_get_transaction(journal
, new_transaction
);
218 new_transaction
= NULL
;
220 write_unlock(&journal
->j_state_lock
);
224 transaction
= journal
->j_running_transaction
;
227 * If the current transaction is locked down for commit, wait for the
228 * lock to be released.
230 if (transaction
->t_state
== T_LOCKED
) {
233 prepare_to_wait(&journal
->j_wait_transaction_locked
,
234 &wait
, TASK_UNINTERRUPTIBLE
);
235 read_unlock(&journal
->j_state_lock
);
237 finish_wait(&journal
->j_wait_transaction_locked
, &wait
);
242 * If there is not enough space left in the log to write all potential
243 * buffers requested by this operation, we need to stall pending a log
244 * checkpoint to free some more log space.
246 needed
= atomic_add_return(nblocks
,
247 &transaction
->t_outstanding_credits
);
249 if (needed
> journal
->j_max_transaction_buffers
) {
251 * If the current transaction is already too large, then start
252 * to commit it: we can then go back and attach this handle to
257 jbd_debug(2, "Handle %p starting new commit...\n", handle
);
258 atomic_sub(nblocks
, &transaction
->t_outstanding_credits
);
259 prepare_to_wait(&journal
->j_wait_transaction_locked
, &wait
,
260 TASK_UNINTERRUPTIBLE
);
261 tid
= transaction
->t_tid
;
262 need_to_start
= !tid_geq(journal
->j_commit_request
, tid
);
263 read_unlock(&journal
->j_state_lock
);
265 jbd2_log_start_commit(journal
, tid
);
267 finish_wait(&journal
->j_wait_transaction_locked
, &wait
);
272 * The commit code assumes that it can get enough log space
273 * without forcing a checkpoint. This is *critical* for
274 * correctness: a checkpoint of a buffer which is also
275 * associated with a committing transaction creates a deadlock,
276 * so commit simply cannot force through checkpoints.
278 * We must therefore ensure the necessary space in the journal
279 * *before* starting to dirty potentially checkpointed buffers
280 * in the new transaction.
282 * The worst part is, any transaction currently committing can
283 * reduce the free space arbitrarily. Be careful to account for
284 * those buffers when checkpointing.
288 * @@@ AKPM: This seems rather over-defensive. We're giving commit
289 * a _lot_ of headroom: 1/4 of the journal plus the size of
290 * the committing transaction. Really, we only need to give it
291 * committing_transaction->t_outstanding_credits plus "enough" for
292 * the log control blocks.
293 * Also, this test is inconsistent with the matching one in
294 * jbd2_journal_extend().
296 if (__jbd2_log_space_left(journal
) < jbd_space_needed(journal
)) {
297 jbd_debug(2, "Handle %p waiting for checkpoint...\n", handle
);
298 atomic_sub(nblocks
, &transaction
->t_outstanding_credits
);
299 read_unlock(&journal
->j_state_lock
);
300 write_lock(&journal
->j_state_lock
);
301 if (__jbd2_log_space_left(journal
) < jbd_space_needed(journal
))
302 __jbd2_log_wait_for_space(journal
);
303 write_unlock(&journal
->j_state_lock
);
307 /* OK, account for the buffers that this operation expects to
308 * use and add the handle to the running transaction.
310 update_t_max_wait(transaction
, ts
);
311 handle
->h_transaction
= transaction
;
312 handle
->h_requested_credits
= nblocks
;
313 handle
->h_start_jiffies
= jiffies
;
314 atomic_inc(&transaction
->t_updates
);
315 atomic_inc(&transaction
->t_handle_count
);
316 jbd_debug(4, "Handle %p given %d credits (total %d, free %d)\n",
318 atomic_read(&transaction
->t_outstanding_credits
),
319 __jbd2_log_space_left(journal
));
320 read_unlock(&journal
->j_state_lock
);
322 lock_map_acquire(&handle
->h_lockdep_map
);
323 jbd2_journal_free_transaction(new_transaction
);
327 static struct lock_class_key jbd2_handle_key
;
329 /* Allocate a new handle. This should probably be in a slab... */
330 static handle_t
*new_handle(int nblocks
)
332 handle_t
*handle
= jbd2_alloc_handle(GFP_NOFS
);
335 handle
->h_buffer_credits
= nblocks
;
338 lockdep_init_map(&handle
->h_lockdep_map
, "jbd2_handle",
339 &jbd2_handle_key
, 0);
345 * handle_t *jbd2_journal_start() - Obtain a new handle.
346 * @journal: Journal to start transaction on.
347 * @nblocks: number of block buffer we might modify
349 * We make sure that the transaction can guarantee at least nblocks of
350 * modified buffers in the log. We block until the log can guarantee
353 * This function is visible to journal users (like ext3fs), so is not
354 * called with the journal already locked.
356 * Return a pointer to a newly allocated handle, or an ERR_PTR() value
359 handle_t
*jbd2__journal_start(journal_t
*journal
, int nblocks
, gfp_t gfp_mask
,
360 unsigned int type
, unsigned int line_no
)
362 handle_t
*handle
= journal_current_handle();
366 return ERR_PTR(-EROFS
);
369 J_ASSERT(handle
->h_transaction
->t_journal
== journal
);
374 handle
= new_handle(nblocks
);
376 return ERR_PTR(-ENOMEM
);
378 current
->journal_info
= handle
;
380 err
= start_this_handle(journal
, handle
, gfp_mask
);
382 jbd2_free_handle(handle
);
383 current
->journal_info
= NULL
;
386 handle
->h_type
= type
;
387 handle
->h_line_no
= line_no
;
388 trace_jbd2_handle_start(journal
->j_fs_dev
->bd_dev
,
389 handle
->h_transaction
->t_tid
, type
,
393 EXPORT_SYMBOL(jbd2__journal_start
);
396 handle_t
*jbd2_journal_start(journal_t
*journal
, int nblocks
)
398 return jbd2__journal_start(journal
, nblocks
, GFP_NOFS
, 0, 0);
400 EXPORT_SYMBOL(jbd2_journal_start
);
404 * int jbd2_journal_extend() - extend buffer credits.
405 * @handle: handle to 'extend'
406 * @nblocks: nr blocks to try to extend by.
408 * Some transactions, such as large extends and truncates, can be done
409 * atomically all at once or in several stages. The operation requests
410 * a credit for a number of buffer modications in advance, but can
411 * extend its credit if it needs more.
413 * jbd2_journal_extend tries to give the running handle more buffer credits.
414 * It does not guarantee that allocation - this is a best-effort only.
415 * The calling process MUST be able to deal cleanly with a failure to
418 * Return 0 on success, non-zero on failure.
420 * return code < 0 implies an error
421 * return code > 0 implies normal transaction-full status.
423 int jbd2_journal_extend(handle_t
*handle
, int nblocks
)
425 transaction_t
*transaction
= handle
->h_transaction
;
426 journal_t
*journal
= transaction
->t_journal
;
431 if (is_handle_aborted(handle
))
436 read_lock(&journal
->j_state_lock
);
438 /* Don't extend a locked-down transaction! */
439 if (handle
->h_transaction
->t_state
!= T_RUNNING
) {
440 jbd_debug(3, "denied handle %p %d blocks: "
441 "transaction not running\n", handle
, nblocks
);
445 spin_lock(&transaction
->t_handle_lock
);
446 wanted
= atomic_read(&transaction
->t_outstanding_credits
) + nblocks
;
448 if (wanted
> journal
->j_max_transaction_buffers
) {
449 jbd_debug(3, "denied handle %p %d blocks: "
450 "transaction too large\n", handle
, nblocks
);
454 if (wanted
> __jbd2_log_space_left(journal
)) {
455 jbd_debug(3, "denied handle %p %d blocks: "
456 "insufficient log space\n", handle
, nblocks
);
460 trace_jbd2_handle_extend(journal
->j_fs_dev
->bd_dev
,
461 handle
->h_transaction
->t_tid
,
462 handle
->h_type
, handle
->h_line_no
,
463 handle
->h_buffer_credits
,
466 handle
->h_buffer_credits
+= nblocks
;
467 handle
->h_requested_credits
+= nblocks
;
468 atomic_add(nblocks
, &transaction
->t_outstanding_credits
);
471 jbd_debug(3, "extended handle %p by %d\n", handle
, nblocks
);
473 spin_unlock(&transaction
->t_handle_lock
);
475 read_unlock(&journal
->j_state_lock
);
482 * int jbd2_journal_restart() - restart a handle .
483 * @handle: handle to restart
484 * @nblocks: nr credits requested
486 * Restart a handle for a multi-transaction filesystem
489 * If the jbd2_journal_extend() call above fails to grant new buffer credits
490 * to a running handle, a call to jbd2_journal_restart will commit the
491 * handle's transaction so far and reattach the handle to a new
492 * transaction capabable of guaranteeing the requested number of
495 int jbd2__journal_restart(handle_t
*handle
, int nblocks
, gfp_t gfp_mask
)
497 transaction_t
*transaction
= handle
->h_transaction
;
498 journal_t
*journal
= transaction
->t_journal
;
500 int need_to_start
, ret
;
502 /* If we've had an abort of any type, don't even think about
503 * actually doing the restart! */
504 if (is_handle_aborted(handle
))
508 * First unlink the handle from its current transaction, and start the
511 J_ASSERT(atomic_read(&transaction
->t_updates
) > 0);
512 J_ASSERT(journal_current_handle() == handle
);
514 read_lock(&journal
->j_state_lock
);
515 spin_lock(&transaction
->t_handle_lock
);
516 atomic_sub(handle
->h_buffer_credits
,
517 &transaction
->t_outstanding_credits
);
518 if (atomic_dec_and_test(&transaction
->t_updates
))
519 wake_up(&journal
->j_wait_updates
);
520 spin_unlock(&transaction
->t_handle_lock
);
522 jbd_debug(2, "restarting handle %p\n", handle
);
523 tid
= transaction
->t_tid
;
524 need_to_start
= !tid_geq(journal
->j_commit_request
, tid
);
525 read_unlock(&journal
->j_state_lock
);
527 jbd2_log_start_commit(journal
, tid
);
529 lock_map_release(&handle
->h_lockdep_map
);
530 handle
->h_buffer_credits
= nblocks
;
531 ret
= start_this_handle(journal
, handle
, gfp_mask
);
534 EXPORT_SYMBOL(jbd2__journal_restart
);
537 int jbd2_journal_restart(handle_t
*handle
, int nblocks
)
539 return jbd2__journal_restart(handle
, nblocks
, GFP_NOFS
);
541 EXPORT_SYMBOL(jbd2_journal_restart
);
544 * void jbd2_journal_lock_updates () - establish a transaction barrier.
545 * @journal: Journal to establish a barrier on.
547 * This locks out any further updates from being started, and blocks
548 * until all existing updates have completed, returning only once the
549 * journal is in a quiescent state with no updates running.
551 * The journal lock should not be held on entry.
553 void jbd2_journal_lock_updates(journal_t
*journal
)
557 write_lock(&journal
->j_state_lock
);
558 ++journal
->j_barrier_count
;
560 /* Wait until there are no running updates */
562 transaction_t
*transaction
= journal
->j_running_transaction
;
567 spin_lock(&transaction
->t_handle_lock
);
568 prepare_to_wait(&journal
->j_wait_updates
, &wait
,
569 TASK_UNINTERRUPTIBLE
);
570 if (!atomic_read(&transaction
->t_updates
)) {
571 spin_unlock(&transaction
->t_handle_lock
);
572 finish_wait(&journal
->j_wait_updates
, &wait
);
575 spin_unlock(&transaction
->t_handle_lock
);
576 write_unlock(&journal
->j_state_lock
);
578 finish_wait(&journal
->j_wait_updates
, &wait
);
579 write_lock(&journal
->j_state_lock
);
581 write_unlock(&journal
->j_state_lock
);
584 * We have now established a barrier against other normal updates, but
585 * we also need to barrier against other jbd2_journal_lock_updates() calls
586 * to make sure that we serialise special journal-locked operations
589 mutex_lock(&journal
->j_barrier
);
593 * void jbd2_journal_unlock_updates (journal_t* journal) - release barrier
594 * @journal: Journal to release the barrier on.
596 * Release a transaction barrier obtained with jbd2_journal_lock_updates().
598 * Should be called without the journal lock held.
600 void jbd2_journal_unlock_updates (journal_t
*journal
)
602 J_ASSERT(journal
->j_barrier_count
!= 0);
604 mutex_unlock(&journal
->j_barrier
);
605 write_lock(&journal
->j_state_lock
);
606 --journal
->j_barrier_count
;
607 write_unlock(&journal
->j_state_lock
);
608 wake_up(&journal
->j_wait_transaction_locked
);
611 static void warn_dirty_buffer(struct buffer_head
*bh
)
613 char b
[BDEVNAME_SIZE
];
616 "JBD2: Spotted dirty metadata buffer (dev = %s, blocknr = %llu). "
617 "There's a risk of filesystem corruption in case of system "
619 bdevname(bh
->b_bdev
, b
), (unsigned long long)bh
->b_blocknr
);
622 static int sleep_on_shadow_bh(void *word
)
629 * If the buffer is already part of the current transaction, then there
630 * is nothing we need to do. If it is already part of a prior
631 * transaction which we are still committing to disk, then we need to
632 * make sure that we do not overwrite the old copy: we do copy-out to
633 * preserve the copy going to disk. We also account the buffer against
634 * the handle's metadata buffer credits (unless the buffer is already
635 * part of the transaction, that is).
639 do_get_write_access(handle_t
*handle
, struct journal_head
*jh
,
642 struct buffer_head
*bh
;
643 transaction_t
*transaction
;
646 char *frozen_buffer
= NULL
;
648 unsigned long start_lock
, time_lock
;
650 if (is_handle_aborted(handle
))
653 transaction
= handle
->h_transaction
;
654 journal
= transaction
->t_journal
;
656 jbd_debug(5, "journal_head %p, force_copy %d\n", jh
, force_copy
);
658 JBUFFER_TRACE(jh
, "entry");
662 /* @@@ Need to check for errors here at some point. */
664 start_lock
= jiffies
;
666 jbd_lock_bh_state(bh
);
668 /* If it takes too long to lock the buffer, trace it */
669 time_lock
= jbd2_time_diff(start_lock
, jiffies
);
670 if (time_lock
> HZ
/10)
671 trace_jbd2_lock_buffer_stall(bh
->b_bdev
->bd_dev
,
672 jiffies_to_msecs(time_lock
));
674 /* We now hold the buffer lock so it is safe to query the buffer
675 * state. Is the buffer dirty?
677 * If so, there are two possibilities. The buffer may be
678 * non-journaled, and undergoing a quite legitimate writeback.
679 * Otherwise, it is journaled, and we don't expect dirty buffers
680 * in that state (the buffers should be marked JBD_Dirty
681 * instead.) So either the IO is being done under our own
682 * control and this is a bug, or it's a third party IO such as
683 * dump(8) (which may leave the buffer scheduled for read ---
684 * ie. locked but not dirty) or tune2fs (which may actually have
685 * the buffer dirtied, ugh.) */
687 if (buffer_dirty(bh
)) {
689 * First question: is this buffer already part of the current
690 * transaction or the existing committing transaction?
692 if (jh
->b_transaction
) {
694 jh
->b_transaction
== transaction
||
696 journal
->j_committing_transaction
);
697 if (jh
->b_next_transaction
)
698 J_ASSERT_JH(jh
, jh
->b_next_transaction
==
700 warn_dirty_buffer(bh
);
703 * In any case we need to clean the dirty flag and we must
704 * do it under the buffer lock to be sure we don't race
705 * with running write-out.
707 JBUFFER_TRACE(jh
, "Journalling dirty buffer");
708 clear_buffer_dirty(bh
);
709 set_buffer_jbddirty(bh
);
715 if (is_handle_aborted(handle
)) {
716 jbd_unlock_bh_state(bh
);
722 * The buffer is already part of this transaction if b_transaction or
723 * b_next_transaction points to it
725 if (jh
->b_transaction
== transaction
||
726 jh
->b_next_transaction
== transaction
)
730 * this is the first time this transaction is touching this buffer,
731 * reset the modified flag
736 * If there is already a copy-out version of this buffer, then we don't
737 * need to make another one
739 if (jh
->b_frozen_data
) {
740 JBUFFER_TRACE(jh
, "has frozen data");
741 J_ASSERT_JH(jh
, jh
->b_next_transaction
== NULL
);
742 jh
->b_next_transaction
= transaction
;
746 /* Is there data here we need to preserve? */
748 if (jh
->b_transaction
&& jh
->b_transaction
!= transaction
) {
749 JBUFFER_TRACE(jh
, "owned by older transaction");
750 J_ASSERT_JH(jh
, jh
->b_next_transaction
== NULL
);
751 J_ASSERT_JH(jh
, jh
->b_transaction
==
752 journal
->j_committing_transaction
);
754 /* There is one case we have to be very careful about.
755 * If the committing transaction is currently writing
756 * this buffer out to disk and has NOT made a copy-out,
757 * then we cannot modify the buffer contents at all
758 * right now. The essence of copy-out is that it is the
759 * extra copy, not the primary copy, which gets
760 * journaled. If the primary copy is already going to
761 * disk then we cannot do copy-out here. */
763 if (buffer_shadow(bh
)) {
764 JBUFFER_TRACE(jh
, "on shadow: sleep");
765 jbd_unlock_bh_state(bh
);
766 wait_on_bit(&bh
->b_state
, BH_Shadow
,
767 sleep_on_shadow_bh
, TASK_UNINTERRUPTIBLE
);
772 * Only do the copy if the currently-owning transaction still
773 * needs it. If buffer isn't on BJ_Metadata list, the
774 * committing transaction is past that stage (here we use the
775 * fact that BH_Shadow is set under bh_state lock together with
776 * refiling to BJ_Shadow list and at this point we know the
777 * buffer doesn't have BH_Shadow set).
779 * Subtle point, though: if this is a get_undo_access,
780 * then we will be relying on the frozen_data to contain
781 * the new value of the committed_data record after the
782 * transaction, so we HAVE to force the frozen_data copy
785 if (jh
->b_jlist
== BJ_Metadata
|| force_copy
) {
786 JBUFFER_TRACE(jh
, "generate frozen data");
787 if (!frozen_buffer
) {
788 JBUFFER_TRACE(jh
, "allocate memory for buffer");
789 jbd_unlock_bh_state(bh
);
791 jbd2_alloc(jh2bh(jh
)->b_size
,
793 if (!frozen_buffer
) {
795 "%s: OOM for frozen_buffer\n",
797 JBUFFER_TRACE(jh
, "oom!");
799 jbd_lock_bh_state(bh
);
804 jh
->b_frozen_data
= frozen_buffer
;
805 frozen_buffer
= NULL
;
808 jh
->b_next_transaction
= transaction
;
813 * Finally, if the buffer is not journaled right now, we need to make
814 * sure it doesn't get written to disk before the caller actually
815 * commits the new data
817 if (!jh
->b_transaction
) {
818 JBUFFER_TRACE(jh
, "no transaction");
819 J_ASSERT_JH(jh
, !jh
->b_next_transaction
);
820 JBUFFER_TRACE(jh
, "file as BJ_Reserved");
821 spin_lock(&journal
->j_list_lock
);
822 __jbd2_journal_file_buffer(jh
, transaction
, BJ_Reserved
);
823 spin_unlock(&journal
->j_list_lock
);
832 J_EXPECT_JH(jh
, buffer_uptodate(jh2bh(jh
)),
833 "Possible IO failure.\n");
834 page
= jh2bh(jh
)->b_page
;
835 offset
= offset_in_page(jh2bh(jh
)->b_data
);
836 source
= kmap_atomic(page
);
837 /* Fire data frozen trigger just before we copy the data */
838 jbd2_buffer_frozen_trigger(jh
, source
+ offset
,
840 memcpy(jh
->b_frozen_data
, source
+offset
, jh2bh(jh
)->b_size
);
841 kunmap_atomic(source
);
844 * Now that the frozen data is saved off, we need to store
845 * any matching triggers.
847 jh
->b_frozen_triggers
= jh
->b_triggers
;
849 jbd_unlock_bh_state(bh
);
852 * If we are about to journal a buffer, then any revoke pending on it is
855 jbd2_journal_cancel_revoke(handle
, jh
);
858 if (unlikely(frozen_buffer
)) /* It's usually NULL */
859 jbd2_free(frozen_buffer
, bh
->b_size
);
861 JBUFFER_TRACE(jh
, "exit");
866 * int jbd2_journal_get_write_access() - notify intent to modify a buffer for metadata (not data) update.
867 * @handle: transaction to add buffer modifications to
868 * @bh: bh to be used for metadata writes
870 * Returns an error code or 0 on success.
872 * In full data journalling mode the buffer may be of type BJ_AsyncData,
873 * because we're write()ing a buffer which is also part of a shared mapping.
876 int jbd2_journal_get_write_access(handle_t
*handle
, struct buffer_head
*bh
)
878 struct journal_head
*jh
= jbd2_journal_add_journal_head(bh
);
881 /* We do not want to get caught playing with fields which the
882 * log thread also manipulates. Make sure that the buffer
883 * completes any outstanding IO before proceeding. */
884 rc
= do_get_write_access(handle
, jh
, 0);
885 jbd2_journal_put_journal_head(jh
);
891 * When the user wants to journal a newly created buffer_head
892 * (ie. getblk() returned a new buffer and we are going to populate it
893 * manually rather than reading off disk), then we need to keep the
894 * buffer_head locked until it has been completely filled with new
895 * data. In this case, we should be able to make the assertion that
896 * the bh is not already part of an existing transaction.
898 * The buffer should already be locked by the caller by this point.
899 * There is no lock ranking violation: it was a newly created,
900 * unlocked buffer beforehand. */
903 * int jbd2_journal_get_create_access () - notify intent to use newly created bh
904 * @handle: transaction to new buffer to
907 * Call this if you create a new bh.
909 int jbd2_journal_get_create_access(handle_t
*handle
, struct buffer_head
*bh
)
911 transaction_t
*transaction
= handle
->h_transaction
;
912 journal_t
*journal
= transaction
->t_journal
;
913 struct journal_head
*jh
= jbd2_journal_add_journal_head(bh
);
916 jbd_debug(5, "journal_head %p\n", jh
);
918 if (is_handle_aborted(handle
))
922 JBUFFER_TRACE(jh
, "entry");
924 * The buffer may already belong to this transaction due to pre-zeroing
925 * in the filesystem's new_block code. It may also be on the previous,
926 * committing transaction's lists, but it HAS to be in Forget state in
927 * that case: the transaction must have deleted the buffer for it to be
930 jbd_lock_bh_state(bh
);
931 spin_lock(&journal
->j_list_lock
);
932 J_ASSERT_JH(jh
, (jh
->b_transaction
== transaction
||
933 jh
->b_transaction
== NULL
||
934 (jh
->b_transaction
== journal
->j_committing_transaction
&&
935 jh
->b_jlist
== BJ_Forget
)));
937 J_ASSERT_JH(jh
, jh
->b_next_transaction
== NULL
);
938 J_ASSERT_JH(jh
, buffer_locked(jh2bh(jh
)));
940 if (jh
->b_transaction
== NULL
) {
942 * Previous jbd2_journal_forget() could have left the buffer
943 * with jbddirty bit set because it was being committed. When
944 * the commit finished, we've filed the buffer for
945 * checkpointing and marked it dirty. Now we are reallocating
946 * the buffer so the transaction freeing it must have
947 * committed and so it's safe to clear the dirty bit.
949 clear_buffer_dirty(jh2bh(jh
));
950 /* first access by this transaction */
953 JBUFFER_TRACE(jh
, "file as BJ_Reserved");
954 __jbd2_journal_file_buffer(jh
, transaction
, BJ_Reserved
);
955 } else if (jh
->b_transaction
== journal
->j_committing_transaction
) {
956 /* first access by this transaction */
959 JBUFFER_TRACE(jh
, "set next transaction");
960 jh
->b_next_transaction
= transaction
;
962 spin_unlock(&journal
->j_list_lock
);
963 jbd_unlock_bh_state(bh
);
966 * akpm: I added this. ext3_alloc_branch can pick up new indirect
967 * blocks which contain freed but then revoked metadata. We need
968 * to cancel the revoke in case we end up freeing it yet again
969 * and the reallocating as data - this would cause a second revoke,
970 * which hits an assertion error.
972 JBUFFER_TRACE(jh
, "cancelling revoke");
973 jbd2_journal_cancel_revoke(handle
, jh
);
975 jbd2_journal_put_journal_head(jh
);
980 * int jbd2_journal_get_undo_access() - Notify intent to modify metadata with
981 * non-rewindable consequences
982 * @handle: transaction
983 * @bh: buffer to undo
985 * Sometimes there is a need to distinguish between metadata which has
986 * been committed to disk and that which has not. The ext3fs code uses
987 * this for freeing and allocating space, we have to make sure that we
988 * do not reuse freed space until the deallocation has been committed,
989 * since if we overwrote that space we would make the delete
990 * un-rewindable in case of a crash.
992 * To deal with that, jbd2_journal_get_undo_access requests write access to a
993 * buffer for parts of non-rewindable operations such as delete
994 * operations on the bitmaps. The journaling code must keep a copy of
995 * the buffer's contents prior to the undo_access call until such time
996 * as we know that the buffer has definitely been committed to disk.
998 * We never need to know which transaction the committed data is part
999 * of, buffers touched here are guaranteed to be dirtied later and so
1000 * will be committed to a new transaction in due course, at which point
1001 * we can discard the old committed data pointer.
1003 * Returns error number or 0 on success.
1005 int jbd2_journal_get_undo_access(handle_t
*handle
, struct buffer_head
*bh
)
1008 struct journal_head
*jh
= jbd2_journal_add_journal_head(bh
);
1009 char *committed_data
= NULL
;
1011 JBUFFER_TRACE(jh
, "entry");
1014 * Do this first --- it can drop the journal lock, so we want to
1015 * make sure that obtaining the committed_data is done
1016 * atomically wrt. completion of any outstanding commits.
1018 err
= do_get_write_access(handle
, jh
, 1);
1023 if (!jh
->b_committed_data
) {
1024 committed_data
= jbd2_alloc(jh2bh(jh
)->b_size
, GFP_NOFS
);
1025 if (!committed_data
) {
1026 printk(KERN_EMERG
"%s: No memory for committed data\n",
1033 jbd_lock_bh_state(bh
);
1034 if (!jh
->b_committed_data
) {
1035 /* Copy out the current buffer contents into the
1036 * preserved, committed copy. */
1037 JBUFFER_TRACE(jh
, "generate b_committed data");
1038 if (!committed_data
) {
1039 jbd_unlock_bh_state(bh
);
1043 jh
->b_committed_data
= committed_data
;
1044 committed_data
= NULL
;
1045 memcpy(jh
->b_committed_data
, bh
->b_data
, bh
->b_size
);
1047 jbd_unlock_bh_state(bh
);
1049 jbd2_journal_put_journal_head(jh
);
1050 if (unlikely(committed_data
))
1051 jbd2_free(committed_data
, bh
->b_size
);
1056 * void jbd2_journal_set_triggers() - Add triggers for commit writeout
1057 * @bh: buffer to trigger on
1058 * @type: struct jbd2_buffer_trigger_type containing the trigger(s).
1060 * Set any triggers on this journal_head. This is always safe, because
1061 * triggers for a committing buffer will be saved off, and triggers for
1062 * a running transaction will match the buffer in that transaction.
1064 * Call with NULL to clear the triggers.
1066 void jbd2_journal_set_triggers(struct buffer_head
*bh
,
1067 struct jbd2_buffer_trigger_type
*type
)
1069 struct journal_head
*jh
= jbd2_journal_grab_journal_head(bh
);
1073 jh
->b_triggers
= type
;
1074 jbd2_journal_put_journal_head(jh
);
1077 void jbd2_buffer_frozen_trigger(struct journal_head
*jh
, void *mapped_data
,
1078 struct jbd2_buffer_trigger_type
*triggers
)
1080 struct buffer_head
*bh
= jh2bh(jh
);
1082 if (!triggers
|| !triggers
->t_frozen
)
1085 triggers
->t_frozen(triggers
, bh
, mapped_data
, bh
->b_size
);
1088 void jbd2_buffer_abort_trigger(struct journal_head
*jh
,
1089 struct jbd2_buffer_trigger_type
*triggers
)
1091 if (!triggers
|| !triggers
->t_abort
)
1094 triggers
->t_abort(triggers
, jh2bh(jh
));
1100 * int jbd2_journal_dirty_metadata() - mark a buffer as containing dirty metadata
1101 * @handle: transaction to add buffer to.
1102 * @bh: buffer to mark
1104 * mark dirty metadata which needs to be journaled as part of the current
1107 * The buffer must have previously had jbd2_journal_get_write_access()
1108 * called so that it has a valid journal_head attached to the buffer
1111 * The buffer is placed on the transaction's metadata list and is marked
1112 * as belonging to the transaction.
1114 * Returns error number or 0 on success.
1116 * Special care needs to be taken if the buffer already belongs to the
1117 * current committing transaction (in which case we should have frozen
1118 * data present for that commit). In that case, we don't relink the
1119 * buffer: that only gets done when the old transaction finally
1120 * completes its commit.
1122 int jbd2_journal_dirty_metadata(handle_t
*handle
, struct buffer_head
*bh
)
1124 transaction_t
*transaction
= handle
->h_transaction
;
1125 journal_t
*journal
= transaction
->t_journal
;
1126 struct journal_head
*jh
;
1129 if (is_handle_aborted(handle
))
1131 jh
= jbd2_journal_grab_journal_head(bh
);
1136 jbd_debug(5, "journal_head %p\n", jh
);
1137 JBUFFER_TRACE(jh
, "entry");
1139 jbd_lock_bh_state(bh
);
1141 if (jh
->b_modified
== 0) {
1143 * This buffer's got modified and becoming part
1144 * of the transaction. This needs to be done
1145 * once a transaction -bzzz
1148 J_ASSERT_JH(jh
, handle
->h_buffer_credits
> 0);
1149 handle
->h_buffer_credits
--;
1153 * fastpath, to avoid expensive locking. If this buffer is already
1154 * on the running transaction's metadata list there is nothing to do.
1155 * Nobody can take it off again because there is a handle open.
1156 * I _think_ we're OK here with SMP barriers - a mistaken decision will
1157 * result in this test being false, so we go in and take the locks.
1159 if (jh
->b_transaction
== transaction
&& jh
->b_jlist
== BJ_Metadata
) {
1160 JBUFFER_TRACE(jh
, "fastpath");
1161 if (unlikely(jh
->b_transaction
!=
1162 journal
->j_running_transaction
)) {
1163 printk(KERN_EMERG
"JBD: %s: "
1164 "jh->b_transaction (%llu, %p, %u) != "
1165 "journal->j_running_transaction (%p, %u)",
1167 (unsigned long long) bh
->b_blocknr
,
1169 jh
->b_transaction
? jh
->b_transaction
->t_tid
: 0,
1170 journal
->j_running_transaction
,
1171 journal
->j_running_transaction
?
1172 journal
->j_running_transaction
->t_tid
: 0);
1178 set_buffer_jbddirty(bh
);
1181 * Metadata already on the current transaction list doesn't
1182 * need to be filed. Metadata on another transaction's list must
1183 * be committing, and will be refiled once the commit completes:
1184 * leave it alone for now.
1186 if (jh
->b_transaction
!= transaction
) {
1187 JBUFFER_TRACE(jh
, "already on other transaction");
1188 if (unlikely(jh
->b_transaction
!=
1189 journal
->j_committing_transaction
)) {
1190 printk(KERN_EMERG
"JBD: %s: "
1191 "jh->b_transaction (%llu, %p, %u) != "
1192 "journal->j_committing_transaction (%p, %u)",
1194 (unsigned long long) bh
->b_blocknr
,
1196 jh
->b_transaction
? jh
->b_transaction
->t_tid
: 0,
1197 journal
->j_committing_transaction
,
1198 journal
->j_committing_transaction
?
1199 journal
->j_committing_transaction
->t_tid
: 0);
1202 if (unlikely(jh
->b_next_transaction
!= transaction
)) {
1203 printk(KERN_EMERG
"JBD: %s: "
1204 "jh->b_next_transaction (%llu, %p, %u) != "
1205 "transaction (%p, %u)",
1207 (unsigned long long) bh
->b_blocknr
,
1208 jh
->b_next_transaction
,
1209 jh
->b_next_transaction
?
1210 jh
->b_next_transaction
->t_tid
: 0,
1211 transaction
, transaction
->t_tid
);
1214 /* And this case is illegal: we can't reuse another
1215 * transaction's data buffer, ever. */
1219 /* That test should have eliminated the following case: */
1220 J_ASSERT_JH(jh
, jh
->b_frozen_data
== NULL
);
1222 JBUFFER_TRACE(jh
, "file as BJ_Metadata");
1223 spin_lock(&journal
->j_list_lock
);
1224 __jbd2_journal_file_buffer(jh
, handle
->h_transaction
, BJ_Metadata
);
1225 spin_unlock(&journal
->j_list_lock
);
1227 jbd_unlock_bh_state(bh
);
1228 jbd2_journal_put_journal_head(jh
);
1230 JBUFFER_TRACE(jh
, "exit");
1231 WARN_ON(ret
); /* All errors are bugs, so dump the stack */
1236 * void jbd2_journal_forget() - bforget() for potentially-journaled buffers.
1237 * @handle: transaction handle
1238 * @bh: bh to 'forget'
1240 * We can only do the bforget if there are no commits pending against the
1241 * buffer. If the buffer is dirty in the current running transaction we
1242 * can safely unlink it.
1244 * bh may not be a journalled buffer at all - it may be a non-JBD
1245 * buffer which came off the hashtable. Check for this.
1247 * Decrements bh->b_count by one.
1249 * Allow this call even if the handle has aborted --- it may be part of
1250 * the caller's cleanup after an abort.
1252 int jbd2_journal_forget (handle_t
*handle
, struct buffer_head
*bh
)
1254 transaction_t
*transaction
= handle
->h_transaction
;
1255 journal_t
*journal
= transaction
->t_journal
;
1256 struct journal_head
*jh
;
1257 int drop_reserve
= 0;
1259 int was_modified
= 0;
1261 BUFFER_TRACE(bh
, "entry");
1263 jbd_lock_bh_state(bh
);
1264 spin_lock(&journal
->j_list_lock
);
1266 if (!buffer_jbd(bh
))
1270 /* Critical error: attempting to delete a bitmap buffer, maybe?
1271 * Don't do any jbd operations, and return an error. */
1272 if (!J_EXPECT_JH(jh
, !jh
->b_committed_data
,
1273 "inconsistent data on disk")) {
1278 /* keep track of whether or not this transaction modified us */
1279 was_modified
= jh
->b_modified
;
1282 * The buffer's going from the transaction, we must drop
1283 * all references -bzzz
1287 if (jh
->b_transaction
== handle
->h_transaction
) {
1288 J_ASSERT_JH(jh
, !jh
->b_frozen_data
);
1290 /* If we are forgetting a buffer which is already part
1291 * of this transaction, then we can just drop it from
1292 * the transaction immediately. */
1293 clear_buffer_dirty(bh
);
1294 clear_buffer_jbddirty(bh
);
1296 JBUFFER_TRACE(jh
, "belongs to current transaction: unfile");
1299 * we only want to drop a reference if this transaction
1300 * modified the buffer
1306 * We are no longer going to journal this buffer.
1307 * However, the commit of this transaction is still
1308 * important to the buffer: the delete that we are now
1309 * processing might obsolete an old log entry, so by
1310 * committing, we can satisfy the buffer's checkpoint.
1312 * So, if we have a checkpoint on the buffer, we should
1313 * now refile the buffer on our BJ_Forget list so that
1314 * we know to remove the checkpoint after we commit.
1317 if (jh
->b_cp_transaction
) {
1318 __jbd2_journal_temp_unlink_buffer(jh
);
1319 __jbd2_journal_file_buffer(jh
, transaction
, BJ_Forget
);
1321 __jbd2_journal_unfile_buffer(jh
);
1322 if (!buffer_jbd(bh
)) {
1323 spin_unlock(&journal
->j_list_lock
);
1324 jbd_unlock_bh_state(bh
);
1329 } else if (jh
->b_transaction
) {
1330 J_ASSERT_JH(jh
, (jh
->b_transaction
==
1331 journal
->j_committing_transaction
));
1332 /* However, if the buffer is still owned by a prior
1333 * (committing) transaction, we can't drop it yet... */
1334 JBUFFER_TRACE(jh
, "belongs to older transaction");
1335 /* ... but we CAN drop it from the new transaction if we
1336 * have also modified it since the original commit. */
1338 if (jh
->b_next_transaction
) {
1339 J_ASSERT(jh
->b_next_transaction
== transaction
);
1340 jh
->b_next_transaction
= NULL
;
1343 * only drop a reference if this transaction modified
1352 spin_unlock(&journal
->j_list_lock
);
1353 jbd_unlock_bh_state(bh
);
1357 /* no need to reserve log space for this block -bzzz */
1358 handle
->h_buffer_credits
++;
1364 * int jbd2_journal_stop() - complete a transaction
1365 * @handle: tranaction to complete.
1367 * All done for a particular handle.
1369 * There is not much action needed here. We just return any remaining
1370 * buffer credits to the transaction and remove the handle. The only
1371 * complication is that we need to start a commit operation if the
1372 * filesystem is marked for synchronous update.
1374 * jbd2_journal_stop itself will not usually return an error, but it may
1375 * do so in unusual circumstances. In particular, expect it to
1376 * return -EIO if a jbd2_journal_abort has been executed since the
1377 * transaction began.
1379 int jbd2_journal_stop(handle_t
*handle
)
1381 transaction_t
*transaction
= handle
->h_transaction
;
1382 journal_t
*journal
= transaction
->t_journal
;
1383 int err
, wait_for_commit
= 0;
1387 J_ASSERT(journal_current_handle() == handle
);
1389 if (is_handle_aborted(handle
))
1392 J_ASSERT(atomic_read(&transaction
->t_updates
) > 0);
1396 if (--handle
->h_ref
> 0) {
1397 jbd_debug(4, "h_ref %d -> %d\n", handle
->h_ref
+ 1,
1402 jbd_debug(4, "Handle %p going down\n", handle
);
1403 trace_jbd2_handle_stats(journal
->j_fs_dev
->bd_dev
,
1404 handle
->h_transaction
->t_tid
,
1405 handle
->h_type
, handle
->h_line_no
,
1406 jiffies
- handle
->h_start_jiffies
,
1407 handle
->h_sync
, handle
->h_requested_credits
,
1408 (handle
->h_requested_credits
-
1409 handle
->h_buffer_credits
));
1412 * Implement synchronous transaction batching. If the handle
1413 * was synchronous, don't force a commit immediately. Let's
1414 * yield and let another thread piggyback onto this
1415 * transaction. Keep doing that while new threads continue to
1416 * arrive. It doesn't cost much - we're about to run a commit
1417 * and sleep on IO anyway. Speeds up many-threaded, many-dir
1418 * operations by 30x or more...
1420 * We try and optimize the sleep time against what the
1421 * underlying disk can do, instead of having a static sleep
1422 * time. This is useful for the case where our storage is so
1423 * fast that it is more optimal to go ahead and force a flush
1424 * and wait for the transaction to be committed than it is to
1425 * wait for an arbitrary amount of time for new writers to
1426 * join the transaction. We achieve this by measuring how
1427 * long it takes to commit a transaction, and compare it with
1428 * how long this transaction has been running, and if run time
1429 * < commit time then we sleep for the delta and commit. This
1430 * greatly helps super fast disks that would see slowdowns as
1431 * more threads started doing fsyncs.
1433 * But don't do this if this process was the most recent one
1434 * to perform a synchronous write. We do this to detect the
1435 * case where a single process is doing a stream of sync
1436 * writes. No point in waiting for joiners in that case.
1439 if (handle
->h_sync
&& journal
->j_last_sync_writer
!= pid
) {
1440 u64 commit_time
, trans_time
;
1442 journal
->j_last_sync_writer
= pid
;
1444 read_lock(&journal
->j_state_lock
);
1445 commit_time
= journal
->j_average_commit_time
;
1446 read_unlock(&journal
->j_state_lock
);
1448 trans_time
= ktime_to_ns(ktime_sub(ktime_get(),
1449 transaction
->t_start_time
));
1451 commit_time
= max_t(u64
, commit_time
,
1452 1000*journal
->j_min_batch_time
);
1453 commit_time
= min_t(u64
, commit_time
,
1454 1000*journal
->j_max_batch_time
);
1456 if (trans_time
< commit_time
) {
1457 ktime_t expires
= ktime_add_ns(ktime_get(),
1459 set_current_state(TASK_UNINTERRUPTIBLE
);
1460 schedule_hrtimeout(&expires
, HRTIMER_MODE_ABS
);
1465 transaction
->t_synchronous_commit
= 1;
1466 current
->journal_info
= NULL
;
1467 atomic_sub(handle
->h_buffer_credits
,
1468 &transaction
->t_outstanding_credits
);
1471 * If the handle is marked SYNC, we need to set another commit
1472 * going! We also want to force a commit if the current
1473 * transaction is occupying too much of the log, or if the
1474 * transaction is too old now.
1476 if (handle
->h_sync
||
1477 (atomic_read(&transaction
->t_outstanding_credits
) >
1478 journal
->j_max_transaction_buffers
) ||
1479 time_after_eq(jiffies
, transaction
->t_expires
)) {
1480 /* Do this even for aborted journals: an abort still
1481 * completes the commit thread, it just doesn't write
1482 * anything to disk. */
1484 jbd_debug(2, "transaction too old, requesting commit for "
1485 "handle %p\n", handle
);
1486 /* This is non-blocking */
1487 jbd2_log_start_commit(journal
, transaction
->t_tid
);
1490 * Special case: JBD2_SYNC synchronous updates require us
1491 * to wait for the commit to complete.
1493 if (handle
->h_sync
&& !(current
->flags
& PF_MEMALLOC
))
1494 wait_for_commit
= 1;
1498 * Once we drop t_updates, if it goes to zero the transaction
1499 * could start committing on us and eventually disappear. So
1500 * once we do this, we must not dereference transaction
1503 tid
= transaction
->t_tid
;
1504 if (atomic_dec_and_test(&transaction
->t_updates
)) {
1505 wake_up(&journal
->j_wait_updates
);
1506 if (journal
->j_barrier_count
)
1507 wake_up(&journal
->j_wait_transaction_locked
);
1510 if (wait_for_commit
)
1511 err
= jbd2_log_wait_commit(journal
, tid
);
1513 lock_map_release(&handle
->h_lockdep_map
);
1515 jbd2_free_handle(handle
);
1520 * int jbd2_journal_force_commit() - force any uncommitted transactions
1521 * @journal: journal to force
1523 * For synchronous operations: force any uncommitted transactions
1524 * to disk. May seem kludgy, but it reuses all the handle batching
1525 * code in a very simple manner.
1527 int jbd2_journal_force_commit(journal_t
*journal
)
1532 handle
= jbd2_journal_start(journal
, 1);
1533 if (IS_ERR(handle
)) {
1534 ret
= PTR_ERR(handle
);
1537 ret
= jbd2_journal_stop(handle
);
1544 * List management code snippets: various functions for manipulating the
1545 * transaction buffer lists.
1550 * Append a buffer to a transaction list, given the transaction's list head
1553 * j_list_lock is held.
1555 * jbd_lock_bh_state(jh2bh(jh)) is held.
1559 __blist_add_buffer(struct journal_head
**list
, struct journal_head
*jh
)
1562 jh
->b_tnext
= jh
->b_tprev
= jh
;
1565 /* Insert at the tail of the list to preserve order */
1566 struct journal_head
*first
= *list
, *last
= first
->b_tprev
;
1568 jh
->b_tnext
= first
;
1569 last
->b_tnext
= first
->b_tprev
= jh
;
1574 * Remove a buffer from a transaction list, given the transaction's list
1577 * Called with j_list_lock held, and the journal may not be locked.
1579 * jbd_lock_bh_state(jh2bh(jh)) is held.
1583 __blist_del_buffer(struct journal_head
**list
, struct journal_head
*jh
)
1586 *list
= jh
->b_tnext
;
1590 jh
->b_tprev
->b_tnext
= jh
->b_tnext
;
1591 jh
->b_tnext
->b_tprev
= jh
->b_tprev
;
1595 * Remove a buffer from the appropriate transaction list.
1597 * Note that this function can *change* the value of
1598 * bh->b_transaction->t_buffers, t_forget, t_shadow_list, t_log_list or
1599 * t_reserved_list. If the caller is holding onto a copy of one of these
1600 * pointers, it could go bad. Generally the caller needs to re-read the
1601 * pointer from the transaction_t.
1603 * Called under j_list_lock.
1605 static void __jbd2_journal_temp_unlink_buffer(struct journal_head
*jh
)
1607 struct journal_head
**list
= NULL
;
1608 transaction_t
*transaction
;
1609 struct buffer_head
*bh
= jh2bh(jh
);
1611 J_ASSERT_JH(jh
, jbd_is_locked_bh_state(bh
));
1612 transaction
= jh
->b_transaction
;
1614 assert_spin_locked(&transaction
->t_journal
->j_list_lock
);
1616 J_ASSERT_JH(jh
, jh
->b_jlist
< BJ_Types
);
1617 if (jh
->b_jlist
!= BJ_None
)
1618 J_ASSERT_JH(jh
, transaction
!= NULL
);
1620 switch (jh
->b_jlist
) {
1624 transaction
->t_nr_buffers
--;
1625 J_ASSERT_JH(jh
, transaction
->t_nr_buffers
>= 0);
1626 list
= &transaction
->t_buffers
;
1629 list
= &transaction
->t_forget
;
1632 list
= &transaction
->t_shadow_list
;
1635 list
= &transaction
->t_reserved_list
;
1639 __blist_del_buffer(list
, jh
);
1640 jh
->b_jlist
= BJ_None
;
1641 if (test_clear_buffer_jbddirty(bh
))
1642 mark_buffer_dirty(bh
); /* Expose it to the VM */
1646 * Remove buffer from all transactions.
1648 * Called with bh_state lock and j_list_lock
1650 * jh and bh may be already freed when this function returns.
1652 static void __jbd2_journal_unfile_buffer(struct journal_head
*jh
)
1654 __jbd2_journal_temp_unlink_buffer(jh
);
1655 jh
->b_transaction
= NULL
;
1656 jbd2_journal_put_journal_head(jh
);
1659 void jbd2_journal_unfile_buffer(journal_t
*journal
, struct journal_head
*jh
)
1661 struct buffer_head
*bh
= jh2bh(jh
);
1663 /* Get reference so that buffer cannot be freed before we unlock it */
1665 jbd_lock_bh_state(bh
);
1666 spin_lock(&journal
->j_list_lock
);
1667 __jbd2_journal_unfile_buffer(jh
);
1668 spin_unlock(&journal
->j_list_lock
);
1669 jbd_unlock_bh_state(bh
);
1674 * Called from jbd2_journal_try_to_free_buffers().
1676 * Called under jbd_lock_bh_state(bh)
1679 __journal_try_to_free_buffer(journal_t
*journal
, struct buffer_head
*bh
)
1681 struct journal_head
*jh
;
1685 if (buffer_locked(bh
) || buffer_dirty(bh
))
1688 if (jh
->b_next_transaction
!= NULL
)
1691 spin_lock(&journal
->j_list_lock
);
1692 if (jh
->b_cp_transaction
!= NULL
&& jh
->b_transaction
== NULL
) {
1693 /* written-back checkpointed metadata buffer */
1694 JBUFFER_TRACE(jh
, "remove from checkpoint list");
1695 __jbd2_journal_remove_checkpoint(jh
);
1697 spin_unlock(&journal
->j_list_lock
);
1703 * int jbd2_journal_try_to_free_buffers() - try to free page buffers.
1704 * @journal: journal for operation
1705 * @page: to try and free
1706 * @gfp_mask: we use the mask to detect how hard should we try to release
1707 * buffers. If __GFP_WAIT and __GFP_FS is set, we wait for commit code to
1708 * release the buffers.
1711 * For all the buffers on this page,
1712 * if they are fully written out ordered data, move them onto BUF_CLEAN
1713 * so try_to_free_buffers() can reap them.
1715 * This function returns non-zero if we wish try_to_free_buffers()
1716 * to be called. We do this if the page is releasable by try_to_free_buffers().
1717 * We also do it if the page has locked or dirty buffers and the caller wants
1718 * us to perform sync or async writeout.
1720 * This complicates JBD locking somewhat. We aren't protected by the
1721 * BKL here. We wish to remove the buffer from its committing or
1722 * running transaction's ->t_datalist via __jbd2_journal_unfile_buffer.
1724 * This may *change* the value of transaction_t->t_datalist, so anyone
1725 * who looks at t_datalist needs to lock against this function.
1727 * Even worse, someone may be doing a jbd2_journal_dirty_data on this
1728 * buffer. So we need to lock against that. jbd2_journal_dirty_data()
1729 * will come out of the lock with the buffer dirty, which makes it
1730 * ineligible for release here.
1732 * Who else is affected by this? hmm... Really the only contender
1733 * is do_get_write_access() - it could be looking at the buffer while
1734 * journal_try_to_free_buffer() is changing its state. But that
1735 * cannot happen because we never reallocate freed data as metadata
1736 * while the data is part of a transaction. Yes?
1738 * Return 0 on failure, 1 on success
1740 int jbd2_journal_try_to_free_buffers(journal_t
*journal
,
1741 struct page
*page
, gfp_t gfp_mask
)
1743 struct buffer_head
*head
;
1744 struct buffer_head
*bh
;
1747 J_ASSERT(PageLocked(page
));
1749 head
= page_buffers(page
);
1752 struct journal_head
*jh
;
1755 * We take our own ref against the journal_head here to avoid
1756 * having to add tons of locking around each instance of
1757 * jbd2_journal_put_journal_head().
1759 jh
= jbd2_journal_grab_journal_head(bh
);
1763 jbd_lock_bh_state(bh
);
1764 __journal_try_to_free_buffer(journal
, bh
);
1765 jbd2_journal_put_journal_head(jh
);
1766 jbd_unlock_bh_state(bh
);
1769 } while ((bh
= bh
->b_this_page
) != head
);
1771 ret
= try_to_free_buffers(page
);
1778 * This buffer is no longer needed. If it is on an older transaction's
1779 * checkpoint list we need to record it on this transaction's forget list
1780 * to pin this buffer (and hence its checkpointing transaction) down until
1781 * this transaction commits. If the buffer isn't on a checkpoint list, we
1783 * Returns non-zero if JBD no longer has an interest in the buffer.
1785 * Called under j_list_lock.
1787 * Called under jbd_lock_bh_state(bh).
1789 static int __dispose_buffer(struct journal_head
*jh
, transaction_t
*transaction
)
1792 struct buffer_head
*bh
= jh2bh(jh
);
1794 if (jh
->b_cp_transaction
) {
1795 JBUFFER_TRACE(jh
, "on running+cp transaction");
1796 __jbd2_journal_temp_unlink_buffer(jh
);
1798 * We don't want to write the buffer anymore, clear the
1799 * bit so that we don't confuse checks in
1800 * __journal_file_buffer
1802 clear_buffer_dirty(bh
);
1803 __jbd2_journal_file_buffer(jh
, transaction
, BJ_Forget
);
1806 JBUFFER_TRACE(jh
, "on running transaction");
1807 __jbd2_journal_unfile_buffer(jh
);
1813 * jbd2_journal_invalidatepage
1815 * This code is tricky. It has a number of cases to deal with.
1817 * There are two invariants which this code relies on:
1819 * i_size must be updated on disk before we start calling invalidatepage on the
1822 * This is done in ext3 by defining an ext3_setattr method which
1823 * updates i_size before truncate gets going. By maintaining this
1824 * invariant, we can be sure that it is safe to throw away any buffers
1825 * attached to the current transaction: once the transaction commits,
1826 * we know that the data will not be needed.
1828 * Note however that we can *not* throw away data belonging to the
1829 * previous, committing transaction!
1831 * Any disk blocks which *are* part of the previous, committing
1832 * transaction (and which therefore cannot be discarded immediately) are
1833 * not going to be reused in the new running transaction
1835 * The bitmap committed_data images guarantee this: any block which is
1836 * allocated in one transaction and removed in the next will be marked
1837 * as in-use in the committed_data bitmap, so cannot be reused until
1838 * the next transaction to delete the block commits. This means that
1839 * leaving committing buffers dirty is quite safe: the disk blocks
1840 * cannot be reallocated to a different file and so buffer aliasing is
1844 * The above applies mainly to ordered data mode. In writeback mode we
1845 * don't make guarantees about the order in which data hits disk --- in
1846 * particular we don't guarantee that new dirty data is flushed before
1847 * transaction commit --- so it is always safe just to discard data
1848 * immediately in that mode. --sct
1852 * The journal_unmap_buffer helper function returns zero if the buffer
1853 * concerned remains pinned as an anonymous buffer belonging to an older
1856 * We're outside-transaction here. Either or both of j_running_transaction
1857 * and j_committing_transaction may be NULL.
1859 static int journal_unmap_buffer(journal_t
*journal
, struct buffer_head
*bh
,
1862 transaction_t
*transaction
;
1863 struct journal_head
*jh
;
1866 BUFFER_TRACE(bh
, "entry");
1869 * It is safe to proceed here without the j_list_lock because the
1870 * buffers cannot be stolen by try_to_free_buffers as long as we are
1871 * holding the page lock. --sct
1874 if (!buffer_jbd(bh
))
1875 goto zap_buffer_unlocked
;
1877 /* OK, we have data buffer in journaled mode */
1878 write_lock(&journal
->j_state_lock
);
1879 jbd_lock_bh_state(bh
);
1880 spin_lock(&journal
->j_list_lock
);
1882 jh
= jbd2_journal_grab_journal_head(bh
);
1884 goto zap_buffer_no_jh
;
1887 * We cannot remove the buffer from checkpoint lists until the
1888 * transaction adding inode to orphan list (let's call it T)
1889 * is committed. Otherwise if the transaction changing the
1890 * buffer would be cleaned from the journal before T is
1891 * committed, a crash will cause that the correct contents of
1892 * the buffer will be lost. On the other hand we have to
1893 * clear the buffer dirty bit at latest at the moment when the
1894 * transaction marking the buffer as freed in the filesystem
1895 * structures is committed because from that moment on the
1896 * block can be reallocated and used by a different page.
1897 * Since the block hasn't been freed yet but the inode has
1898 * already been added to orphan list, it is safe for us to add
1899 * the buffer to BJ_Forget list of the newest transaction.
1901 * Also we have to clear buffer_mapped flag of a truncated buffer
1902 * because the buffer_head may be attached to the page straddling
1903 * i_size (can happen only when blocksize < pagesize) and thus the
1904 * buffer_head can be reused when the file is extended again. So we end
1905 * up keeping around invalidated buffers attached to transactions'
1906 * BJ_Forget list just to stop checkpointing code from cleaning up
1907 * the transaction this buffer was modified in.
1909 transaction
= jh
->b_transaction
;
1910 if (transaction
== NULL
) {
1911 /* First case: not on any transaction. If it
1912 * has no checkpoint link, then we can zap it:
1913 * it's a writeback-mode buffer so we don't care
1914 * if it hits disk safely. */
1915 if (!jh
->b_cp_transaction
) {
1916 JBUFFER_TRACE(jh
, "not on any transaction: zap");
1920 if (!buffer_dirty(bh
)) {
1921 /* bdflush has written it. We can drop it now */
1925 /* OK, it must be in the journal but still not
1926 * written fully to disk: it's metadata or
1927 * journaled data... */
1929 if (journal
->j_running_transaction
) {
1930 /* ... and once the current transaction has
1931 * committed, the buffer won't be needed any
1933 JBUFFER_TRACE(jh
, "checkpointed: add to BJ_Forget");
1934 may_free
= __dispose_buffer(jh
,
1935 journal
->j_running_transaction
);
1938 /* There is no currently-running transaction. So the
1939 * orphan record which we wrote for this file must have
1940 * passed into commit. We must attach this buffer to
1941 * the committing transaction, if it exists. */
1942 if (journal
->j_committing_transaction
) {
1943 JBUFFER_TRACE(jh
, "give to committing trans");
1944 may_free
= __dispose_buffer(jh
,
1945 journal
->j_committing_transaction
);
1948 /* The orphan record's transaction has
1949 * committed. We can cleanse this buffer */
1950 clear_buffer_jbddirty(bh
);
1954 } else if (transaction
== journal
->j_committing_transaction
) {
1955 JBUFFER_TRACE(jh
, "on committing transaction");
1957 * The buffer is committing, we simply cannot touch
1958 * it. If the page is straddling i_size we have to wait
1959 * for commit and try again.
1962 jbd2_journal_put_journal_head(jh
);
1963 spin_unlock(&journal
->j_list_lock
);
1964 jbd_unlock_bh_state(bh
);
1965 write_unlock(&journal
->j_state_lock
);
1969 * OK, buffer won't be reachable after truncate. We just set
1970 * j_next_transaction to the running transaction (if there is
1971 * one) and mark buffer as freed so that commit code knows it
1972 * should clear dirty bits when it is done with the buffer.
1974 set_buffer_freed(bh
);
1975 if (journal
->j_running_transaction
&& buffer_jbddirty(bh
))
1976 jh
->b_next_transaction
= journal
->j_running_transaction
;
1977 jbd2_journal_put_journal_head(jh
);
1978 spin_unlock(&journal
->j_list_lock
);
1979 jbd_unlock_bh_state(bh
);
1980 write_unlock(&journal
->j_state_lock
);
1983 /* Good, the buffer belongs to the running transaction.
1984 * We are writing our own transaction's data, not any
1985 * previous one's, so it is safe to throw it away
1986 * (remember that we expect the filesystem to have set
1987 * i_size already for this truncate so recovery will not
1988 * expose the disk blocks we are discarding here.) */
1989 J_ASSERT_JH(jh
, transaction
== journal
->j_running_transaction
);
1990 JBUFFER_TRACE(jh
, "on running transaction");
1991 may_free
= __dispose_buffer(jh
, transaction
);
1996 * This is tricky. Although the buffer is truncated, it may be reused
1997 * if blocksize < pagesize and it is attached to the page straddling
1998 * EOF. Since the buffer might have been added to BJ_Forget list of the
1999 * running transaction, journal_get_write_access() won't clear
2000 * b_modified and credit accounting gets confused. So clear b_modified
2004 jbd2_journal_put_journal_head(jh
);
2006 spin_unlock(&journal
->j_list_lock
);
2007 jbd_unlock_bh_state(bh
);
2008 write_unlock(&journal
->j_state_lock
);
2009 zap_buffer_unlocked
:
2010 clear_buffer_dirty(bh
);
2011 J_ASSERT_BH(bh
, !buffer_jbddirty(bh
));
2012 clear_buffer_mapped(bh
);
2013 clear_buffer_req(bh
);
2014 clear_buffer_new(bh
);
2015 clear_buffer_delay(bh
);
2016 clear_buffer_unwritten(bh
);
2022 * void jbd2_journal_invalidatepage()
2023 * @journal: journal to use for flush...
2024 * @page: page to flush
2025 * @offset: start of the range to invalidate
2026 * @length: length of the range to invalidate
2028 * Reap page buffers containing data after in the specified range in page.
2029 * Can return -EBUSY if buffers are part of the committing transaction and
2030 * the page is straddling i_size. Caller then has to wait for current commit
2033 int jbd2_journal_invalidatepage(journal_t
*journal
,
2035 unsigned int offset
,
2036 unsigned int length
)
2038 struct buffer_head
*head
, *bh
, *next
;
2039 unsigned int stop
= offset
+ length
;
2040 unsigned int curr_off
= 0;
2041 int partial_page
= (offset
|| length
< PAGE_CACHE_SIZE
);
2045 if (!PageLocked(page
))
2047 if (!page_has_buffers(page
))
2050 BUG_ON(stop
> PAGE_CACHE_SIZE
|| stop
< length
);
2052 /* We will potentially be playing with lists other than just the
2053 * data lists (especially for journaled data mode), so be
2054 * cautious in our locking. */
2056 head
= bh
= page_buffers(page
);
2058 unsigned int next_off
= curr_off
+ bh
->b_size
;
2059 next
= bh
->b_this_page
;
2061 if (next_off
> stop
)
2064 if (offset
<= curr_off
) {
2065 /* This block is wholly outside the truncation point */
2067 ret
= journal_unmap_buffer(journal
, bh
, partial_page
);
2073 curr_off
= next_off
;
2076 } while (bh
!= head
);
2078 if (!partial_page
) {
2079 if (may_free
&& try_to_free_buffers(page
))
2080 J_ASSERT(!page_has_buffers(page
));
2086 * File a buffer on the given transaction list.
2088 void __jbd2_journal_file_buffer(struct journal_head
*jh
,
2089 transaction_t
*transaction
, int jlist
)
2091 struct journal_head
**list
= NULL
;
2093 struct buffer_head
*bh
= jh2bh(jh
);
2095 J_ASSERT_JH(jh
, jbd_is_locked_bh_state(bh
));
2096 assert_spin_locked(&transaction
->t_journal
->j_list_lock
);
2098 J_ASSERT_JH(jh
, jh
->b_jlist
< BJ_Types
);
2099 J_ASSERT_JH(jh
, jh
->b_transaction
== transaction
||
2100 jh
->b_transaction
== NULL
);
2102 if (jh
->b_transaction
&& jh
->b_jlist
== jlist
)
2105 if (jlist
== BJ_Metadata
|| jlist
== BJ_Reserved
||
2106 jlist
== BJ_Shadow
|| jlist
== BJ_Forget
) {
2108 * For metadata buffers, we track dirty bit in buffer_jbddirty
2109 * instead of buffer_dirty. We should not see a dirty bit set
2110 * here because we clear it in do_get_write_access but e.g.
2111 * tune2fs can modify the sb and set the dirty bit at any time
2112 * so we try to gracefully handle that.
2114 if (buffer_dirty(bh
))
2115 warn_dirty_buffer(bh
);
2116 if (test_clear_buffer_dirty(bh
) ||
2117 test_clear_buffer_jbddirty(bh
))
2121 if (jh
->b_transaction
)
2122 __jbd2_journal_temp_unlink_buffer(jh
);
2124 jbd2_journal_grab_journal_head(bh
);
2125 jh
->b_transaction
= transaction
;
2129 J_ASSERT_JH(jh
, !jh
->b_committed_data
);
2130 J_ASSERT_JH(jh
, !jh
->b_frozen_data
);
2133 transaction
->t_nr_buffers
++;
2134 list
= &transaction
->t_buffers
;
2137 list
= &transaction
->t_forget
;
2140 list
= &transaction
->t_shadow_list
;
2143 list
= &transaction
->t_reserved_list
;
2147 __blist_add_buffer(list
, jh
);
2148 jh
->b_jlist
= jlist
;
2151 set_buffer_jbddirty(bh
);
2154 void jbd2_journal_file_buffer(struct journal_head
*jh
,
2155 transaction_t
*transaction
, int jlist
)
2157 jbd_lock_bh_state(jh2bh(jh
));
2158 spin_lock(&transaction
->t_journal
->j_list_lock
);
2159 __jbd2_journal_file_buffer(jh
, transaction
, jlist
);
2160 spin_unlock(&transaction
->t_journal
->j_list_lock
);
2161 jbd_unlock_bh_state(jh2bh(jh
));
2165 * Remove a buffer from its current buffer list in preparation for
2166 * dropping it from its current transaction entirely. If the buffer has
2167 * already started to be used by a subsequent transaction, refile the
2168 * buffer on that transaction's metadata list.
2170 * Called under j_list_lock
2171 * Called under jbd_lock_bh_state(jh2bh(jh))
2173 * jh and bh may be already free when this function returns
2175 void __jbd2_journal_refile_buffer(struct journal_head
*jh
)
2177 int was_dirty
, jlist
;
2178 struct buffer_head
*bh
= jh2bh(jh
);
2180 J_ASSERT_JH(jh
, jbd_is_locked_bh_state(bh
));
2181 if (jh
->b_transaction
)
2182 assert_spin_locked(&jh
->b_transaction
->t_journal
->j_list_lock
);
2184 /* If the buffer is now unused, just drop it. */
2185 if (jh
->b_next_transaction
== NULL
) {
2186 __jbd2_journal_unfile_buffer(jh
);
2191 * It has been modified by a later transaction: add it to the new
2192 * transaction's metadata list.
2195 was_dirty
= test_clear_buffer_jbddirty(bh
);
2196 __jbd2_journal_temp_unlink_buffer(jh
);
2198 * We set b_transaction here because b_next_transaction will inherit
2199 * our jh reference and thus __jbd2_journal_file_buffer() must not
2202 jh
->b_transaction
= jh
->b_next_transaction
;
2203 jh
->b_next_transaction
= NULL
;
2204 if (buffer_freed(bh
))
2206 else if (jh
->b_modified
)
2207 jlist
= BJ_Metadata
;
2209 jlist
= BJ_Reserved
;
2210 __jbd2_journal_file_buffer(jh
, jh
->b_transaction
, jlist
);
2211 J_ASSERT_JH(jh
, jh
->b_transaction
->t_state
== T_RUNNING
);
2214 set_buffer_jbddirty(bh
);
2218 * __jbd2_journal_refile_buffer() with necessary locking added. We take our
2219 * bh reference so that we can safely unlock bh.
2221 * The jh and bh may be freed by this call.
2223 void jbd2_journal_refile_buffer(journal_t
*journal
, struct journal_head
*jh
)
2225 struct buffer_head
*bh
= jh2bh(jh
);
2227 /* Get reference so that buffer cannot be freed before we unlock it */
2229 jbd_lock_bh_state(bh
);
2230 spin_lock(&journal
->j_list_lock
);
2231 __jbd2_journal_refile_buffer(jh
);
2232 jbd_unlock_bh_state(bh
);
2233 spin_unlock(&journal
->j_list_lock
);
2238 * File inode in the inode list of the handle's transaction
2240 int jbd2_journal_file_inode(handle_t
*handle
, struct jbd2_inode
*jinode
)
2242 transaction_t
*transaction
= handle
->h_transaction
;
2243 journal_t
*journal
= transaction
->t_journal
;
2245 if (is_handle_aborted(handle
))
2248 jbd_debug(4, "Adding inode %lu, tid:%d\n", jinode
->i_vfs_inode
->i_ino
,
2249 transaction
->t_tid
);
2252 * First check whether inode isn't already on the transaction's
2253 * lists without taking the lock. Note that this check is safe
2254 * without the lock as we cannot race with somebody removing inode
2255 * from the transaction. The reason is that we remove inode from the
2256 * transaction only in journal_release_jbd_inode() and when we commit
2257 * the transaction. We are guarded from the first case by holding
2258 * a reference to the inode. We are safe against the second case
2259 * because if jinode->i_transaction == transaction, commit code
2260 * cannot touch the transaction because we hold reference to it,
2261 * and if jinode->i_next_transaction == transaction, commit code
2262 * will only file the inode where we want it.
2264 if (jinode
->i_transaction
== transaction
||
2265 jinode
->i_next_transaction
== transaction
)
2268 spin_lock(&journal
->j_list_lock
);
2270 if (jinode
->i_transaction
== transaction
||
2271 jinode
->i_next_transaction
== transaction
)
2275 * We only ever set this variable to 1 so the test is safe. Since
2276 * t_need_data_flush is likely to be set, we do the test to save some
2277 * cacheline bouncing
2279 if (!transaction
->t_need_data_flush
)
2280 transaction
->t_need_data_flush
= 1;
2281 /* On some different transaction's list - should be
2282 * the committing one */
2283 if (jinode
->i_transaction
) {
2284 J_ASSERT(jinode
->i_next_transaction
== NULL
);
2285 J_ASSERT(jinode
->i_transaction
==
2286 journal
->j_committing_transaction
);
2287 jinode
->i_next_transaction
= transaction
;
2290 /* Not on any transaction list... */
2291 J_ASSERT(!jinode
->i_next_transaction
);
2292 jinode
->i_transaction
= transaction
;
2293 list_add(&jinode
->i_list
, &transaction
->t_inode_list
);
2295 spin_unlock(&journal
->j_list_lock
);
2301 * File truncate and transaction commit interact with each other in a
2302 * non-trivial way. If a transaction writing data block A is
2303 * committing, we cannot discard the data by truncate until we have
2304 * written them. Otherwise if we crashed after the transaction with
2305 * write has committed but before the transaction with truncate has
2306 * committed, we could see stale data in block A. This function is a
2307 * helper to solve this problem. It starts writeout of the truncated
2308 * part in case it is in the committing transaction.
2310 * Filesystem code must call this function when inode is journaled in
2311 * ordered mode before truncation happens and after the inode has been
2312 * placed on orphan list with the new inode size. The second condition
2313 * avoids the race that someone writes new data and we start
2314 * committing the transaction after this function has been called but
2315 * before a transaction for truncate is started (and furthermore it
2316 * allows us to optimize the case where the addition to orphan list
2317 * happens in the same transaction as write --- we don't have to write
2318 * any data in such case).
2320 int jbd2_journal_begin_ordered_truncate(journal_t
*journal
,
2321 struct jbd2_inode
*jinode
,
2324 transaction_t
*inode_trans
, *commit_trans
;
2327 /* This is a quick check to avoid locking if not necessary */
2328 if (!jinode
->i_transaction
)
2330 /* Locks are here just to force reading of recent values, it is
2331 * enough that the transaction was not committing before we started
2332 * a transaction adding the inode to orphan list */
2333 read_lock(&journal
->j_state_lock
);
2334 commit_trans
= journal
->j_committing_transaction
;
2335 read_unlock(&journal
->j_state_lock
);
2336 spin_lock(&journal
->j_list_lock
);
2337 inode_trans
= jinode
->i_transaction
;
2338 spin_unlock(&journal
->j_list_lock
);
2339 if (inode_trans
== commit_trans
) {
2340 ret
= filemap_fdatawrite_range(jinode
->i_vfs_inode
->i_mapping
,
2341 new_size
, LLONG_MAX
);
2343 jbd2_journal_abort(journal
, ret
);