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 memset(handle
, 0, sizeof(*handle
));
336 handle
->h_buffer_credits
= nblocks
;
339 lockdep_init_map(&handle
->h_lockdep_map
, "jbd2_handle",
340 &jbd2_handle_key
, 0);
346 * handle_t *jbd2_journal_start() - Obtain a new handle.
347 * @journal: Journal to start transaction on.
348 * @nblocks: number of block buffer we might modify
350 * We make sure that the transaction can guarantee at least nblocks of
351 * modified buffers in the log. We block until the log can guarantee
354 * This function is visible to journal users (like ext3fs), so is not
355 * called with the journal already locked.
357 * Return a pointer to a newly allocated handle, or an ERR_PTR() value
360 handle_t
*jbd2__journal_start(journal_t
*journal
, int nblocks
, gfp_t gfp_mask
,
361 unsigned int type
, unsigned int line_no
)
363 handle_t
*handle
= journal_current_handle();
367 return ERR_PTR(-EROFS
);
370 J_ASSERT(handle
->h_transaction
->t_journal
== journal
);
375 handle
= new_handle(nblocks
);
377 return ERR_PTR(-ENOMEM
);
379 current
->journal_info
= handle
;
381 err
= start_this_handle(journal
, handle
, gfp_mask
);
383 jbd2_free_handle(handle
);
384 current
->journal_info
= NULL
;
385 handle
= ERR_PTR(err
);
387 handle
->h_type
= type
;
388 handle
->h_line_no
= line_no
;
389 trace_jbd2_handle_start(journal
->j_fs_dev
->bd_dev
,
390 handle
->h_transaction
->t_tid
, type
,
394 EXPORT_SYMBOL(jbd2__journal_start
);
397 handle_t
*jbd2_journal_start(journal_t
*journal
, int nblocks
)
399 return jbd2__journal_start(journal
, nblocks
, GFP_NOFS
, 0, 0);
401 EXPORT_SYMBOL(jbd2_journal_start
);
405 * int jbd2_journal_extend() - extend buffer credits.
406 * @handle: handle to 'extend'
407 * @nblocks: nr blocks to try to extend by.
409 * Some transactions, such as large extends and truncates, can be done
410 * atomically all at once or in several stages. The operation requests
411 * a credit for a number of buffer modications in advance, but can
412 * extend its credit if it needs more.
414 * jbd2_journal_extend tries to give the running handle more buffer credits.
415 * It does not guarantee that allocation - this is a best-effort only.
416 * The calling process MUST be able to deal cleanly with a failure to
419 * Return 0 on success, non-zero on failure.
421 * return code < 0 implies an error
422 * return code > 0 implies normal transaction-full status.
424 int jbd2_journal_extend(handle_t
*handle
, int nblocks
)
426 transaction_t
*transaction
= handle
->h_transaction
;
427 journal_t
*journal
= transaction
->t_journal
;
432 if (is_handle_aborted(handle
))
437 read_lock(&journal
->j_state_lock
);
439 /* Don't extend a locked-down transaction! */
440 if (handle
->h_transaction
->t_state
!= T_RUNNING
) {
441 jbd_debug(3, "denied handle %p %d blocks: "
442 "transaction not running\n", handle
, nblocks
);
446 spin_lock(&transaction
->t_handle_lock
);
447 wanted
= atomic_read(&transaction
->t_outstanding_credits
) + nblocks
;
449 if (wanted
> journal
->j_max_transaction_buffers
) {
450 jbd_debug(3, "denied handle %p %d blocks: "
451 "transaction too large\n", handle
, nblocks
);
455 if (wanted
> __jbd2_log_space_left(journal
)) {
456 jbd_debug(3, "denied handle %p %d blocks: "
457 "insufficient log space\n", handle
, nblocks
);
461 trace_jbd2_handle_extend(journal
->j_fs_dev
->bd_dev
,
462 handle
->h_transaction
->t_tid
,
463 handle
->h_type
, handle
->h_line_no
,
464 handle
->h_buffer_credits
,
467 handle
->h_buffer_credits
+= nblocks
;
468 handle
->h_requested_credits
+= nblocks
;
469 atomic_add(nblocks
, &transaction
->t_outstanding_credits
);
472 jbd_debug(3, "extended handle %p by %d\n", handle
, nblocks
);
474 spin_unlock(&transaction
->t_handle_lock
);
476 read_unlock(&journal
->j_state_lock
);
483 * int jbd2_journal_restart() - restart a handle .
484 * @handle: handle to restart
485 * @nblocks: nr credits requested
487 * Restart a handle for a multi-transaction filesystem
490 * If the jbd2_journal_extend() call above fails to grant new buffer credits
491 * to a running handle, a call to jbd2_journal_restart will commit the
492 * handle's transaction so far and reattach the handle to a new
493 * transaction capabable of guaranteeing the requested number of
496 int jbd2__journal_restart(handle_t
*handle
, int nblocks
, gfp_t gfp_mask
)
498 transaction_t
*transaction
= handle
->h_transaction
;
499 journal_t
*journal
= transaction
->t_journal
;
501 int need_to_start
, ret
;
503 /* If we've had an abort of any type, don't even think about
504 * actually doing the restart! */
505 if (is_handle_aborted(handle
))
509 * First unlink the handle from its current transaction, and start the
512 J_ASSERT(atomic_read(&transaction
->t_updates
) > 0);
513 J_ASSERT(journal_current_handle() == handle
);
515 read_lock(&journal
->j_state_lock
);
516 spin_lock(&transaction
->t_handle_lock
);
517 atomic_sub(handle
->h_buffer_credits
,
518 &transaction
->t_outstanding_credits
);
519 if (atomic_dec_and_test(&transaction
->t_updates
))
520 wake_up(&journal
->j_wait_updates
);
521 spin_unlock(&transaction
->t_handle_lock
);
523 jbd_debug(2, "restarting handle %p\n", handle
);
524 tid
= transaction
->t_tid
;
525 need_to_start
= !tid_geq(journal
->j_commit_request
, tid
);
526 read_unlock(&journal
->j_state_lock
);
528 jbd2_log_start_commit(journal
, tid
);
530 lock_map_release(&handle
->h_lockdep_map
);
531 handle
->h_buffer_credits
= nblocks
;
532 ret
= start_this_handle(journal
, handle
, gfp_mask
);
535 EXPORT_SYMBOL(jbd2__journal_restart
);
538 int jbd2_journal_restart(handle_t
*handle
, int nblocks
)
540 return jbd2__journal_restart(handle
, nblocks
, GFP_NOFS
);
542 EXPORT_SYMBOL(jbd2_journal_restart
);
545 * void jbd2_journal_lock_updates () - establish a transaction barrier.
546 * @journal: Journal to establish a barrier on.
548 * This locks out any further updates from being started, and blocks
549 * until all existing updates have completed, returning only once the
550 * journal is in a quiescent state with no updates running.
552 * The journal lock should not be held on entry.
554 void jbd2_journal_lock_updates(journal_t
*journal
)
558 write_lock(&journal
->j_state_lock
);
559 ++journal
->j_barrier_count
;
561 /* Wait until there are no running updates */
563 transaction_t
*transaction
= journal
->j_running_transaction
;
568 spin_lock(&transaction
->t_handle_lock
);
569 prepare_to_wait(&journal
->j_wait_updates
, &wait
,
570 TASK_UNINTERRUPTIBLE
);
571 if (!atomic_read(&transaction
->t_updates
)) {
572 spin_unlock(&transaction
->t_handle_lock
);
573 finish_wait(&journal
->j_wait_updates
, &wait
);
576 spin_unlock(&transaction
->t_handle_lock
);
577 write_unlock(&journal
->j_state_lock
);
579 finish_wait(&journal
->j_wait_updates
, &wait
);
580 write_lock(&journal
->j_state_lock
);
582 write_unlock(&journal
->j_state_lock
);
585 * We have now established a barrier against other normal updates, but
586 * we also need to barrier against other jbd2_journal_lock_updates() calls
587 * to make sure that we serialise special journal-locked operations
590 mutex_lock(&journal
->j_barrier
);
594 * void jbd2_journal_unlock_updates (journal_t* journal) - release barrier
595 * @journal: Journal to release the barrier on.
597 * Release a transaction barrier obtained with jbd2_journal_lock_updates().
599 * Should be called without the journal lock held.
601 void jbd2_journal_unlock_updates (journal_t
*journal
)
603 J_ASSERT(journal
->j_barrier_count
!= 0);
605 mutex_unlock(&journal
->j_barrier
);
606 write_lock(&journal
->j_state_lock
);
607 --journal
->j_barrier_count
;
608 write_unlock(&journal
->j_state_lock
);
609 wake_up(&journal
->j_wait_transaction_locked
);
612 static void warn_dirty_buffer(struct buffer_head
*bh
)
614 char b
[BDEVNAME_SIZE
];
617 "JBD2: Spotted dirty metadata buffer (dev = %s, blocknr = %llu). "
618 "There's a risk of filesystem corruption in case of system "
620 bdevname(bh
->b_bdev
, b
), (unsigned long long)bh
->b_blocknr
);
624 * If the buffer is already part of the current transaction, then there
625 * is nothing we need to do. If it is already part of a prior
626 * transaction which we are still committing to disk, then we need to
627 * make sure that we do not overwrite the old copy: we do copy-out to
628 * preserve the copy going to disk. We also account the buffer against
629 * the handle's metadata buffer credits (unless the buffer is already
630 * part of the transaction, that is).
634 do_get_write_access(handle_t
*handle
, struct journal_head
*jh
,
637 struct buffer_head
*bh
;
638 transaction_t
*transaction
;
641 char *frozen_buffer
= NULL
;
644 if (is_handle_aborted(handle
))
647 transaction
= handle
->h_transaction
;
648 journal
= transaction
->t_journal
;
650 jbd_debug(5, "journal_head %p, force_copy %d\n", jh
, force_copy
);
652 JBUFFER_TRACE(jh
, "entry");
656 /* @@@ Need to check for errors here at some point. */
659 jbd_lock_bh_state(bh
);
661 /* We now hold the buffer lock so it is safe to query the buffer
662 * state. Is the buffer dirty?
664 * If so, there are two possibilities. The buffer may be
665 * non-journaled, and undergoing a quite legitimate writeback.
666 * Otherwise, it is journaled, and we don't expect dirty buffers
667 * in that state (the buffers should be marked JBD_Dirty
668 * instead.) So either the IO is being done under our own
669 * control and this is a bug, or it's a third party IO such as
670 * dump(8) (which may leave the buffer scheduled for read ---
671 * ie. locked but not dirty) or tune2fs (which may actually have
672 * the buffer dirtied, ugh.) */
674 if (buffer_dirty(bh
)) {
676 * First question: is this buffer already part of the current
677 * transaction or the existing committing transaction?
679 if (jh
->b_transaction
) {
681 jh
->b_transaction
== transaction
||
683 journal
->j_committing_transaction
);
684 if (jh
->b_next_transaction
)
685 J_ASSERT_JH(jh
, jh
->b_next_transaction
==
687 warn_dirty_buffer(bh
);
690 * In any case we need to clean the dirty flag and we must
691 * do it under the buffer lock to be sure we don't race
692 * with running write-out.
694 JBUFFER_TRACE(jh
, "Journalling dirty buffer");
695 clear_buffer_dirty(bh
);
696 set_buffer_jbddirty(bh
);
702 if (is_handle_aborted(handle
)) {
703 jbd_unlock_bh_state(bh
);
709 * The buffer is already part of this transaction if b_transaction or
710 * b_next_transaction points to it
712 if (jh
->b_transaction
== transaction
||
713 jh
->b_next_transaction
== transaction
)
717 * this is the first time this transaction is touching this buffer,
718 * reset the modified flag
723 * If there is already a copy-out version of this buffer, then we don't
724 * need to make another one
726 if (jh
->b_frozen_data
) {
727 JBUFFER_TRACE(jh
, "has frozen data");
728 J_ASSERT_JH(jh
, jh
->b_next_transaction
== NULL
);
729 jh
->b_next_transaction
= transaction
;
733 /* Is there data here we need to preserve? */
735 if (jh
->b_transaction
&& jh
->b_transaction
!= transaction
) {
736 JBUFFER_TRACE(jh
, "owned by older transaction");
737 J_ASSERT_JH(jh
, jh
->b_next_transaction
== NULL
);
738 J_ASSERT_JH(jh
, jh
->b_transaction
==
739 journal
->j_committing_transaction
);
741 /* There is one case we have to be very careful about.
742 * If the committing transaction is currently writing
743 * this buffer out to disk and has NOT made a copy-out,
744 * then we cannot modify the buffer contents at all
745 * right now. The essence of copy-out is that it is the
746 * extra copy, not the primary copy, which gets
747 * journaled. If the primary copy is already going to
748 * disk then we cannot do copy-out here. */
750 if (jh
->b_jlist
== BJ_Shadow
) {
751 DEFINE_WAIT_BIT(wait
, &bh
->b_state
, BH_Unshadow
);
752 wait_queue_head_t
*wqh
;
754 wqh
= bit_waitqueue(&bh
->b_state
, BH_Unshadow
);
756 JBUFFER_TRACE(jh
, "on shadow: sleep");
757 jbd_unlock_bh_state(bh
);
758 /* commit wakes up all shadow buffers after IO */
760 prepare_to_wait(wqh
, &wait
.wait
,
761 TASK_UNINTERRUPTIBLE
);
762 if (jh
->b_jlist
!= BJ_Shadow
)
766 finish_wait(wqh
, &wait
.wait
);
770 /* Only do the copy if the currently-owning transaction
771 * still needs it. If it is on the Forget list, the
772 * committing transaction is past that stage. The
773 * buffer had better remain locked during the kmalloc,
774 * but that should be true --- we hold the journal lock
775 * still and the buffer is already on the BUF_JOURNAL
776 * list so won't be flushed.
778 * Subtle point, though: if this is a get_undo_access,
779 * then we will be relying on the frozen_data to contain
780 * the new value of the committed_data record after the
781 * transaction, so we HAVE to force the frozen_data copy
784 if (jh
->b_jlist
!= BJ_Forget
|| force_copy
) {
785 JBUFFER_TRACE(jh
, "generate frozen data");
786 if (!frozen_buffer
) {
787 JBUFFER_TRACE(jh
, "allocate memory for buffer");
788 jbd_unlock_bh_state(bh
);
790 jbd2_alloc(jh2bh(jh
)->b_size
,
792 if (!frozen_buffer
) {
794 "%s: OOM for frozen_buffer\n",
796 JBUFFER_TRACE(jh
, "oom!");
798 jbd_lock_bh_state(bh
);
803 jh
->b_frozen_data
= frozen_buffer
;
804 frozen_buffer
= NULL
;
807 jh
->b_next_transaction
= transaction
;
812 * Finally, if the buffer is not journaled right now, we need to make
813 * sure it doesn't get written to disk before the caller actually
814 * commits the new data
816 if (!jh
->b_transaction
) {
817 JBUFFER_TRACE(jh
, "no transaction");
818 J_ASSERT_JH(jh
, !jh
->b_next_transaction
);
819 JBUFFER_TRACE(jh
, "file as BJ_Reserved");
820 spin_lock(&journal
->j_list_lock
);
821 __jbd2_journal_file_buffer(jh
, transaction
, BJ_Reserved
);
822 spin_unlock(&journal
->j_list_lock
);
831 J_EXPECT_JH(jh
, buffer_uptodate(jh2bh(jh
)),
832 "Possible IO failure.\n");
833 page
= jh2bh(jh
)->b_page
;
834 offset
= offset_in_page(jh2bh(jh
)->b_data
);
835 source
= kmap_atomic(page
);
836 /* Fire data frozen trigger just before we copy the data */
837 jbd2_buffer_frozen_trigger(jh
, source
+ offset
,
839 memcpy(jh
->b_frozen_data
, source
+offset
, jh2bh(jh
)->b_size
);
840 kunmap_atomic(source
);
843 * Now that the frozen data is saved off, we need to store
844 * any matching triggers.
846 jh
->b_frozen_triggers
= jh
->b_triggers
;
848 jbd_unlock_bh_state(bh
);
851 * If we are about to journal a buffer, then any revoke pending on it is
854 jbd2_journal_cancel_revoke(handle
, jh
);
857 if (unlikely(frozen_buffer
)) /* It's usually NULL */
858 jbd2_free(frozen_buffer
, bh
->b_size
);
860 JBUFFER_TRACE(jh
, "exit");
865 * int jbd2_journal_get_write_access() - notify intent to modify a buffer for metadata (not data) update.
866 * @handle: transaction to add buffer modifications to
867 * @bh: bh to be used for metadata writes
869 * Returns an error code or 0 on success.
871 * In full data journalling mode the buffer may be of type BJ_AsyncData,
872 * because we're write()ing a buffer which is also part of a shared mapping.
875 int jbd2_journal_get_write_access(handle_t
*handle
, struct buffer_head
*bh
)
877 struct journal_head
*jh
= jbd2_journal_add_journal_head(bh
);
880 /* We do not want to get caught playing with fields which the
881 * log thread also manipulates. Make sure that the buffer
882 * completes any outstanding IO before proceeding. */
883 rc
= do_get_write_access(handle
, jh
, 0);
884 jbd2_journal_put_journal_head(jh
);
890 * When the user wants to journal a newly created buffer_head
891 * (ie. getblk() returned a new buffer and we are going to populate it
892 * manually rather than reading off disk), then we need to keep the
893 * buffer_head locked until it has been completely filled with new
894 * data. In this case, we should be able to make the assertion that
895 * the bh is not already part of an existing transaction.
897 * The buffer should already be locked by the caller by this point.
898 * There is no lock ranking violation: it was a newly created,
899 * unlocked buffer beforehand. */
902 * int jbd2_journal_get_create_access () - notify intent to use newly created bh
903 * @handle: transaction to new buffer to
906 * Call this if you create a new bh.
908 int jbd2_journal_get_create_access(handle_t
*handle
, struct buffer_head
*bh
)
910 transaction_t
*transaction
= handle
->h_transaction
;
911 journal_t
*journal
= transaction
->t_journal
;
912 struct journal_head
*jh
= jbd2_journal_add_journal_head(bh
);
915 jbd_debug(5, "journal_head %p\n", jh
);
917 if (is_handle_aborted(handle
))
921 JBUFFER_TRACE(jh
, "entry");
923 * The buffer may already belong to this transaction due to pre-zeroing
924 * in the filesystem's new_block code. It may also be on the previous,
925 * committing transaction's lists, but it HAS to be in Forget state in
926 * that case: the transaction must have deleted the buffer for it to be
929 jbd_lock_bh_state(bh
);
930 spin_lock(&journal
->j_list_lock
);
931 J_ASSERT_JH(jh
, (jh
->b_transaction
== transaction
||
932 jh
->b_transaction
== NULL
||
933 (jh
->b_transaction
== journal
->j_committing_transaction
&&
934 jh
->b_jlist
== BJ_Forget
)));
936 J_ASSERT_JH(jh
, jh
->b_next_transaction
== NULL
);
937 J_ASSERT_JH(jh
, buffer_locked(jh2bh(jh
)));
939 if (jh
->b_transaction
== NULL
) {
941 * Previous jbd2_journal_forget() could have left the buffer
942 * with jbddirty bit set because it was being committed. When
943 * the commit finished, we've filed the buffer for
944 * checkpointing and marked it dirty. Now we are reallocating
945 * the buffer so the transaction freeing it must have
946 * committed and so it's safe to clear the dirty bit.
948 clear_buffer_dirty(jh2bh(jh
));
949 /* first access by this transaction */
952 JBUFFER_TRACE(jh
, "file as BJ_Reserved");
953 __jbd2_journal_file_buffer(jh
, transaction
, BJ_Reserved
);
954 } else if (jh
->b_transaction
== journal
->j_committing_transaction
) {
955 /* first access by this transaction */
958 JBUFFER_TRACE(jh
, "set next transaction");
959 jh
->b_next_transaction
= transaction
;
961 spin_unlock(&journal
->j_list_lock
);
962 jbd_unlock_bh_state(bh
);
965 * akpm: I added this. ext3_alloc_branch can pick up new indirect
966 * blocks which contain freed but then revoked metadata. We need
967 * to cancel the revoke in case we end up freeing it yet again
968 * and the reallocating as data - this would cause a second revoke,
969 * which hits an assertion error.
971 JBUFFER_TRACE(jh
, "cancelling revoke");
972 jbd2_journal_cancel_revoke(handle
, jh
);
974 jbd2_journal_put_journal_head(jh
);
979 * int jbd2_journal_get_undo_access() - Notify intent to modify metadata with
980 * non-rewindable consequences
981 * @handle: transaction
982 * @bh: buffer to undo
984 * Sometimes there is a need to distinguish between metadata which has
985 * been committed to disk and that which has not. The ext3fs code uses
986 * this for freeing and allocating space, we have to make sure that we
987 * do not reuse freed space until the deallocation has been committed,
988 * since if we overwrote that space we would make the delete
989 * un-rewindable in case of a crash.
991 * To deal with that, jbd2_journal_get_undo_access requests write access to a
992 * buffer for parts of non-rewindable operations such as delete
993 * operations on the bitmaps. The journaling code must keep a copy of
994 * the buffer's contents prior to the undo_access call until such time
995 * as we know that the buffer has definitely been committed to disk.
997 * We never need to know which transaction the committed data is part
998 * of, buffers touched here are guaranteed to be dirtied later and so
999 * will be committed to a new transaction in due course, at which point
1000 * we can discard the old committed data pointer.
1002 * Returns error number or 0 on success.
1004 int jbd2_journal_get_undo_access(handle_t
*handle
, struct buffer_head
*bh
)
1007 struct journal_head
*jh
= jbd2_journal_add_journal_head(bh
);
1008 char *committed_data
= NULL
;
1010 JBUFFER_TRACE(jh
, "entry");
1013 * Do this first --- it can drop the journal lock, so we want to
1014 * make sure that obtaining the committed_data is done
1015 * atomically wrt. completion of any outstanding commits.
1017 err
= do_get_write_access(handle
, jh
, 1);
1022 if (!jh
->b_committed_data
) {
1023 committed_data
= jbd2_alloc(jh2bh(jh
)->b_size
, GFP_NOFS
);
1024 if (!committed_data
) {
1025 printk(KERN_EMERG
"%s: No memory for committed data\n",
1032 jbd_lock_bh_state(bh
);
1033 if (!jh
->b_committed_data
) {
1034 /* Copy out the current buffer contents into the
1035 * preserved, committed copy. */
1036 JBUFFER_TRACE(jh
, "generate b_committed data");
1037 if (!committed_data
) {
1038 jbd_unlock_bh_state(bh
);
1042 jh
->b_committed_data
= committed_data
;
1043 committed_data
= NULL
;
1044 memcpy(jh
->b_committed_data
, bh
->b_data
, bh
->b_size
);
1046 jbd_unlock_bh_state(bh
);
1048 jbd2_journal_put_journal_head(jh
);
1049 if (unlikely(committed_data
))
1050 jbd2_free(committed_data
, bh
->b_size
);
1055 * void jbd2_journal_set_triggers() - Add triggers for commit writeout
1056 * @bh: buffer to trigger on
1057 * @type: struct jbd2_buffer_trigger_type containing the trigger(s).
1059 * Set any triggers on this journal_head. This is always safe, because
1060 * triggers for a committing buffer will be saved off, and triggers for
1061 * a running transaction will match the buffer in that transaction.
1063 * Call with NULL to clear the triggers.
1065 void jbd2_journal_set_triggers(struct buffer_head
*bh
,
1066 struct jbd2_buffer_trigger_type
*type
)
1068 struct journal_head
*jh
= bh2jh(bh
);
1070 jh
->b_triggers
= type
;
1073 void jbd2_buffer_frozen_trigger(struct journal_head
*jh
, void *mapped_data
,
1074 struct jbd2_buffer_trigger_type
*triggers
)
1076 struct buffer_head
*bh
= jh2bh(jh
);
1078 if (!triggers
|| !triggers
->t_frozen
)
1081 triggers
->t_frozen(triggers
, bh
, mapped_data
, bh
->b_size
);
1084 void jbd2_buffer_abort_trigger(struct journal_head
*jh
,
1085 struct jbd2_buffer_trigger_type
*triggers
)
1087 if (!triggers
|| !triggers
->t_abort
)
1090 triggers
->t_abort(triggers
, jh2bh(jh
));
1096 * int jbd2_journal_dirty_metadata() - mark a buffer as containing dirty metadata
1097 * @handle: transaction to add buffer to.
1098 * @bh: buffer to mark
1100 * mark dirty metadata which needs to be journaled as part of the current
1103 * The buffer must have previously had jbd2_journal_get_write_access()
1104 * called so that it has a valid journal_head attached to the buffer
1107 * The buffer is placed on the transaction's metadata list and is marked
1108 * as belonging to the transaction.
1110 * Returns error number or 0 on success.
1112 * Special care needs to be taken if the buffer already belongs to the
1113 * current committing transaction (in which case we should have frozen
1114 * data present for that commit). In that case, we don't relink the
1115 * buffer: that only gets done when the old transaction finally
1116 * completes its commit.
1118 int jbd2_journal_dirty_metadata(handle_t
*handle
, struct buffer_head
*bh
)
1120 transaction_t
*transaction
= handle
->h_transaction
;
1121 journal_t
*journal
= transaction
->t_journal
;
1122 struct journal_head
*jh
= bh2jh(bh
);
1125 jbd_debug(5, "journal_head %p\n", jh
);
1126 JBUFFER_TRACE(jh
, "entry");
1127 if (is_handle_aborted(handle
))
1129 if (!buffer_jbd(bh
)) {
1134 jbd_lock_bh_state(bh
);
1136 if (jh
->b_modified
== 0) {
1138 * This buffer's got modified and becoming part
1139 * of the transaction. This needs to be done
1140 * once a transaction -bzzz
1143 J_ASSERT_JH(jh
, handle
->h_buffer_credits
> 0);
1144 handle
->h_buffer_credits
--;
1148 * fastpath, to avoid expensive locking. If this buffer is already
1149 * on the running transaction's metadata list there is nothing to do.
1150 * Nobody can take it off again because there is a handle open.
1151 * I _think_ we're OK here with SMP barriers - a mistaken decision will
1152 * result in this test being false, so we go in and take the locks.
1154 if (jh
->b_transaction
== transaction
&& jh
->b_jlist
== BJ_Metadata
) {
1155 JBUFFER_TRACE(jh
, "fastpath");
1156 if (unlikely(jh
->b_transaction
!=
1157 journal
->j_running_transaction
)) {
1158 printk(KERN_EMERG
"JBD: %s: "
1159 "jh->b_transaction (%llu, %p, %u) != "
1160 "journal->j_running_transaction (%p, %u)",
1162 (unsigned long long) bh
->b_blocknr
,
1164 jh
->b_transaction
? jh
->b_transaction
->t_tid
: 0,
1165 journal
->j_running_transaction
,
1166 journal
->j_running_transaction
?
1167 journal
->j_running_transaction
->t_tid
: 0);
1173 set_buffer_jbddirty(bh
);
1176 * Metadata already on the current transaction list doesn't
1177 * need to be filed. Metadata on another transaction's list must
1178 * be committing, and will be refiled once the commit completes:
1179 * leave it alone for now.
1181 if (jh
->b_transaction
!= transaction
) {
1182 JBUFFER_TRACE(jh
, "already on other transaction");
1183 if (unlikely(jh
->b_transaction
!=
1184 journal
->j_committing_transaction
)) {
1185 printk(KERN_EMERG
"JBD: %s: "
1186 "jh->b_transaction (%llu, %p, %u) != "
1187 "journal->j_committing_transaction (%p, %u)",
1189 (unsigned long long) bh
->b_blocknr
,
1191 jh
->b_transaction
? jh
->b_transaction
->t_tid
: 0,
1192 journal
->j_committing_transaction
,
1193 journal
->j_committing_transaction
?
1194 journal
->j_committing_transaction
->t_tid
: 0);
1197 if (unlikely(jh
->b_next_transaction
!= transaction
)) {
1198 printk(KERN_EMERG
"JBD: %s: "
1199 "jh->b_next_transaction (%llu, %p, %u) != "
1200 "transaction (%p, %u)",
1202 (unsigned long long) bh
->b_blocknr
,
1203 jh
->b_next_transaction
,
1204 jh
->b_next_transaction
?
1205 jh
->b_next_transaction
->t_tid
: 0,
1206 transaction
, transaction
->t_tid
);
1209 /* And this case is illegal: we can't reuse another
1210 * transaction's data buffer, ever. */
1214 /* That test should have eliminated the following case: */
1215 J_ASSERT_JH(jh
, jh
->b_frozen_data
== NULL
);
1217 JBUFFER_TRACE(jh
, "file as BJ_Metadata");
1218 spin_lock(&journal
->j_list_lock
);
1219 __jbd2_journal_file_buffer(jh
, handle
->h_transaction
, BJ_Metadata
);
1220 spin_unlock(&journal
->j_list_lock
);
1222 jbd_unlock_bh_state(bh
);
1224 JBUFFER_TRACE(jh
, "exit");
1225 WARN_ON(ret
); /* All errors are bugs, so dump the stack */
1230 * void jbd2_journal_forget() - bforget() for potentially-journaled buffers.
1231 * @handle: transaction handle
1232 * @bh: bh to 'forget'
1234 * We can only do the bforget if there are no commits pending against the
1235 * buffer. If the buffer is dirty in the current running transaction we
1236 * can safely unlink it.
1238 * bh may not be a journalled buffer at all - it may be a non-JBD
1239 * buffer which came off the hashtable. Check for this.
1241 * Decrements bh->b_count by one.
1243 * Allow this call even if the handle has aborted --- it may be part of
1244 * the caller's cleanup after an abort.
1246 int jbd2_journal_forget (handle_t
*handle
, struct buffer_head
*bh
)
1248 transaction_t
*transaction
= handle
->h_transaction
;
1249 journal_t
*journal
= transaction
->t_journal
;
1250 struct journal_head
*jh
;
1251 int drop_reserve
= 0;
1253 int was_modified
= 0;
1255 BUFFER_TRACE(bh
, "entry");
1257 jbd_lock_bh_state(bh
);
1258 spin_lock(&journal
->j_list_lock
);
1260 if (!buffer_jbd(bh
))
1264 /* Critical error: attempting to delete a bitmap buffer, maybe?
1265 * Don't do any jbd operations, and return an error. */
1266 if (!J_EXPECT_JH(jh
, !jh
->b_committed_data
,
1267 "inconsistent data on disk")) {
1272 /* keep track of whether or not this transaction modified us */
1273 was_modified
= jh
->b_modified
;
1276 * The buffer's going from the transaction, we must drop
1277 * all references -bzzz
1281 if (jh
->b_transaction
== handle
->h_transaction
) {
1282 J_ASSERT_JH(jh
, !jh
->b_frozen_data
);
1284 /* If we are forgetting a buffer which is already part
1285 * of this transaction, then we can just drop it from
1286 * the transaction immediately. */
1287 clear_buffer_dirty(bh
);
1288 clear_buffer_jbddirty(bh
);
1290 JBUFFER_TRACE(jh
, "belongs to current transaction: unfile");
1293 * we only want to drop a reference if this transaction
1294 * modified the buffer
1300 * We are no longer going to journal this buffer.
1301 * However, the commit of this transaction is still
1302 * important to the buffer: the delete that we are now
1303 * processing might obsolete an old log entry, so by
1304 * committing, we can satisfy the buffer's checkpoint.
1306 * So, if we have a checkpoint on the buffer, we should
1307 * now refile the buffer on our BJ_Forget list so that
1308 * we know to remove the checkpoint after we commit.
1311 if (jh
->b_cp_transaction
) {
1312 __jbd2_journal_temp_unlink_buffer(jh
);
1313 __jbd2_journal_file_buffer(jh
, transaction
, BJ_Forget
);
1315 __jbd2_journal_unfile_buffer(jh
);
1316 if (!buffer_jbd(bh
)) {
1317 spin_unlock(&journal
->j_list_lock
);
1318 jbd_unlock_bh_state(bh
);
1323 } else if (jh
->b_transaction
) {
1324 J_ASSERT_JH(jh
, (jh
->b_transaction
==
1325 journal
->j_committing_transaction
));
1326 /* However, if the buffer is still owned by a prior
1327 * (committing) transaction, we can't drop it yet... */
1328 JBUFFER_TRACE(jh
, "belongs to older transaction");
1329 /* ... but we CAN drop it from the new transaction if we
1330 * have also modified it since the original commit. */
1332 if (jh
->b_next_transaction
) {
1333 J_ASSERT(jh
->b_next_transaction
== transaction
);
1334 jh
->b_next_transaction
= NULL
;
1337 * only drop a reference if this transaction modified
1346 spin_unlock(&journal
->j_list_lock
);
1347 jbd_unlock_bh_state(bh
);
1351 /* no need to reserve log space for this block -bzzz */
1352 handle
->h_buffer_credits
++;
1358 * int jbd2_journal_stop() - complete a transaction
1359 * @handle: tranaction to complete.
1361 * All done for a particular handle.
1363 * There is not much action needed here. We just return any remaining
1364 * buffer credits to the transaction and remove the handle. The only
1365 * complication is that we need to start a commit operation if the
1366 * filesystem is marked for synchronous update.
1368 * jbd2_journal_stop itself will not usually return an error, but it may
1369 * do so in unusual circumstances. In particular, expect it to
1370 * return -EIO if a jbd2_journal_abort has been executed since the
1371 * transaction began.
1373 int jbd2_journal_stop(handle_t
*handle
)
1375 transaction_t
*transaction
= handle
->h_transaction
;
1376 journal_t
*journal
= transaction
->t_journal
;
1377 int err
, wait_for_commit
= 0;
1381 J_ASSERT(journal_current_handle() == handle
);
1383 if (is_handle_aborted(handle
))
1386 J_ASSERT(atomic_read(&transaction
->t_updates
) > 0);
1390 if (--handle
->h_ref
> 0) {
1391 jbd_debug(4, "h_ref %d -> %d\n", handle
->h_ref
+ 1,
1396 jbd_debug(4, "Handle %p going down\n", handle
);
1397 trace_jbd2_handle_stats(journal
->j_fs_dev
->bd_dev
,
1398 handle
->h_transaction
->t_tid
,
1399 handle
->h_type
, handle
->h_line_no
,
1400 jiffies
- handle
->h_start_jiffies
,
1401 handle
->h_sync
, handle
->h_requested_credits
,
1402 (handle
->h_requested_credits
-
1403 handle
->h_buffer_credits
));
1406 * Implement synchronous transaction batching. If the handle
1407 * was synchronous, don't force a commit immediately. Let's
1408 * yield and let another thread piggyback onto this
1409 * transaction. Keep doing that while new threads continue to
1410 * arrive. It doesn't cost much - we're about to run a commit
1411 * and sleep on IO anyway. Speeds up many-threaded, many-dir
1412 * operations by 30x or more...
1414 * We try and optimize the sleep time against what the
1415 * underlying disk can do, instead of having a static sleep
1416 * time. This is useful for the case where our storage is so
1417 * fast that it is more optimal to go ahead and force a flush
1418 * and wait for the transaction to be committed than it is to
1419 * wait for an arbitrary amount of time for new writers to
1420 * join the transaction. We achieve this by measuring how
1421 * long it takes to commit a transaction, and compare it with
1422 * how long this transaction has been running, and if run time
1423 * < commit time then we sleep for the delta and commit. This
1424 * greatly helps super fast disks that would see slowdowns as
1425 * more threads started doing fsyncs.
1427 * But don't do this if this process was the most recent one
1428 * to perform a synchronous write. We do this to detect the
1429 * case where a single process is doing a stream of sync
1430 * writes. No point in waiting for joiners in that case.
1433 if (handle
->h_sync
&& journal
->j_last_sync_writer
!= pid
) {
1434 u64 commit_time
, trans_time
;
1436 journal
->j_last_sync_writer
= pid
;
1438 read_lock(&journal
->j_state_lock
);
1439 commit_time
= journal
->j_average_commit_time
;
1440 read_unlock(&journal
->j_state_lock
);
1442 trans_time
= ktime_to_ns(ktime_sub(ktime_get(),
1443 transaction
->t_start_time
));
1445 commit_time
= max_t(u64
, commit_time
,
1446 1000*journal
->j_min_batch_time
);
1447 commit_time
= min_t(u64
, commit_time
,
1448 1000*journal
->j_max_batch_time
);
1450 if (trans_time
< commit_time
) {
1451 ktime_t expires
= ktime_add_ns(ktime_get(),
1453 set_current_state(TASK_UNINTERRUPTIBLE
);
1454 schedule_hrtimeout(&expires
, HRTIMER_MODE_ABS
);
1459 transaction
->t_synchronous_commit
= 1;
1460 current
->journal_info
= NULL
;
1461 atomic_sub(handle
->h_buffer_credits
,
1462 &transaction
->t_outstanding_credits
);
1465 * If the handle is marked SYNC, we need to set another commit
1466 * going! We also want to force a commit if the current
1467 * transaction is occupying too much of the log, or if the
1468 * transaction is too old now.
1470 if (handle
->h_sync
||
1471 (atomic_read(&transaction
->t_outstanding_credits
) >
1472 journal
->j_max_transaction_buffers
) ||
1473 time_after_eq(jiffies
, transaction
->t_expires
)) {
1474 /* Do this even for aborted journals: an abort still
1475 * completes the commit thread, it just doesn't write
1476 * anything to disk. */
1478 jbd_debug(2, "transaction too old, requesting commit for "
1479 "handle %p\n", handle
);
1480 /* This is non-blocking */
1481 jbd2_log_start_commit(journal
, transaction
->t_tid
);
1484 * Special case: JBD2_SYNC synchronous updates require us
1485 * to wait for the commit to complete.
1487 if (handle
->h_sync
&& !(current
->flags
& PF_MEMALLOC
))
1488 wait_for_commit
= 1;
1492 * Once we drop t_updates, if it goes to zero the transaction
1493 * could start committing on us and eventually disappear. So
1494 * once we do this, we must not dereference transaction
1497 tid
= transaction
->t_tid
;
1498 if (atomic_dec_and_test(&transaction
->t_updates
)) {
1499 wake_up(&journal
->j_wait_updates
);
1500 if (journal
->j_barrier_count
)
1501 wake_up(&journal
->j_wait_transaction_locked
);
1504 if (wait_for_commit
)
1505 err
= jbd2_log_wait_commit(journal
, tid
);
1507 lock_map_release(&handle
->h_lockdep_map
);
1509 jbd2_free_handle(handle
);
1514 * int jbd2_journal_force_commit() - force any uncommitted transactions
1515 * @journal: journal to force
1517 * For synchronous operations: force any uncommitted transactions
1518 * to disk. May seem kludgy, but it reuses all the handle batching
1519 * code in a very simple manner.
1521 int jbd2_journal_force_commit(journal_t
*journal
)
1526 handle
= jbd2_journal_start(journal
, 1);
1527 if (IS_ERR(handle
)) {
1528 ret
= PTR_ERR(handle
);
1531 ret
= jbd2_journal_stop(handle
);
1538 * List management code snippets: various functions for manipulating the
1539 * transaction buffer lists.
1544 * Append a buffer to a transaction list, given the transaction's list head
1547 * j_list_lock is held.
1549 * jbd_lock_bh_state(jh2bh(jh)) is held.
1553 __blist_add_buffer(struct journal_head
**list
, struct journal_head
*jh
)
1556 jh
->b_tnext
= jh
->b_tprev
= jh
;
1559 /* Insert at the tail of the list to preserve order */
1560 struct journal_head
*first
= *list
, *last
= first
->b_tprev
;
1562 jh
->b_tnext
= first
;
1563 last
->b_tnext
= first
->b_tprev
= jh
;
1568 * Remove a buffer from a transaction list, given the transaction's list
1571 * Called with j_list_lock held, and the journal may not be locked.
1573 * jbd_lock_bh_state(jh2bh(jh)) is held.
1577 __blist_del_buffer(struct journal_head
**list
, struct journal_head
*jh
)
1580 *list
= jh
->b_tnext
;
1584 jh
->b_tprev
->b_tnext
= jh
->b_tnext
;
1585 jh
->b_tnext
->b_tprev
= jh
->b_tprev
;
1589 * Remove a buffer from the appropriate transaction list.
1591 * Note that this function can *change* the value of
1592 * bh->b_transaction->t_buffers, t_forget, t_iobuf_list, t_shadow_list,
1593 * t_log_list or t_reserved_list. If the caller is holding onto a copy of one
1594 * of these pointers, it could go bad. Generally the caller needs to re-read
1595 * the pointer from the transaction_t.
1597 * Called under j_list_lock.
1599 static void __jbd2_journal_temp_unlink_buffer(struct journal_head
*jh
)
1601 struct journal_head
**list
= NULL
;
1602 transaction_t
*transaction
;
1603 struct buffer_head
*bh
= jh2bh(jh
);
1605 J_ASSERT_JH(jh
, jbd_is_locked_bh_state(bh
));
1606 transaction
= jh
->b_transaction
;
1608 assert_spin_locked(&transaction
->t_journal
->j_list_lock
);
1610 J_ASSERT_JH(jh
, jh
->b_jlist
< BJ_Types
);
1611 if (jh
->b_jlist
!= BJ_None
)
1612 J_ASSERT_JH(jh
, transaction
!= NULL
);
1614 switch (jh
->b_jlist
) {
1618 transaction
->t_nr_buffers
--;
1619 J_ASSERT_JH(jh
, transaction
->t_nr_buffers
>= 0);
1620 list
= &transaction
->t_buffers
;
1623 list
= &transaction
->t_forget
;
1626 list
= &transaction
->t_iobuf_list
;
1629 list
= &transaction
->t_shadow_list
;
1632 list
= &transaction
->t_log_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: length of page to invalidate.
2027 * Reap page buffers containing data after offset in page. Can return -EBUSY
2028 * if buffers are part of the committing transaction and the page is straddling
2029 * i_size. Caller then has to wait for current commit and try again.
2031 int jbd2_journal_invalidatepage(journal_t
*journal
,
2033 unsigned long offset
)
2035 struct buffer_head
*head
, *bh
, *next
;
2036 unsigned int curr_off
= 0;
2040 if (!PageLocked(page
))
2042 if (!page_has_buffers(page
))
2045 /* We will potentially be playing with lists other than just the
2046 * data lists (especially for journaled data mode), so be
2047 * cautious in our locking. */
2049 head
= bh
= page_buffers(page
);
2051 unsigned int next_off
= curr_off
+ bh
->b_size
;
2052 next
= bh
->b_this_page
;
2054 if (offset
<= curr_off
) {
2055 /* This block is wholly outside the truncation point */
2057 ret
= journal_unmap_buffer(journal
, bh
, offset
> 0);
2063 curr_off
= next_off
;
2066 } while (bh
!= head
);
2069 if (may_free
&& try_to_free_buffers(page
))
2070 J_ASSERT(!page_has_buffers(page
));
2076 * File a buffer on the given transaction list.
2078 void __jbd2_journal_file_buffer(struct journal_head
*jh
,
2079 transaction_t
*transaction
, int jlist
)
2081 struct journal_head
**list
= NULL
;
2083 struct buffer_head
*bh
= jh2bh(jh
);
2085 J_ASSERT_JH(jh
, jbd_is_locked_bh_state(bh
));
2086 assert_spin_locked(&transaction
->t_journal
->j_list_lock
);
2088 J_ASSERT_JH(jh
, jh
->b_jlist
< BJ_Types
);
2089 J_ASSERT_JH(jh
, jh
->b_transaction
== transaction
||
2090 jh
->b_transaction
== NULL
);
2092 if (jh
->b_transaction
&& jh
->b_jlist
== jlist
)
2095 if (jlist
== BJ_Metadata
|| jlist
== BJ_Reserved
||
2096 jlist
== BJ_Shadow
|| jlist
== BJ_Forget
) {
2098 * For metadata buffers, we track dirty bit in buffer_jbddirty
2099 * instead of buffer_dirty. We should not see a dirty bit set
2100 * here because we clear it in do_get_write_access but e.g.
2101 * tune2fs can modify the sb and set the dirty bit at any time
2102 * so we try to gracefully handle that.
2104 if (buffer_dirty(bh
))
2105 warn_dirty_buffer(bh
);
2106 if (test_clear_buffer_dirty(bh
) ||
2107 test_clear_buffer_jbddirty(bh
))
2111 if (jh
->b_transaction
)
2112 __jbd2_journal_temp_unlink_buffer(jh
);
2114 jbd2_journal_grab_journal_head(bh
);
2115 jh
->b_transaction
= transaction
;
2119 J_ASSERT_JH(jh
, !jh
->b_committed_data
);
2120 J_ASSERT_JH(jh
, !jh
->b_frozen_data
);
2123 transaction
->t_nr_buffers
++;
2124 list
= &transaction
->t_buffers
;
2127 list
= &transaction
->t_forget
;
2130 list
= &transaction
->t_iobuf_list
;
2133 list
= &transaction
->t_shadow_list
;
2136 list
= &transaction
->t_log_list
;
2139 list
= &transaction
->t_reserved_list
;
2143 __blist_add_buffer(list
, jh
);
2144 jh
->b_jlist
= jlist
;
2147 set_buffer_jbddirty(bh
);
2150 void jbd2_journal_file_buffer(struct journal_head
*jh
,
2151 transaction_t
*transaction
, int jlist
)
2153 jbd_lock_bh_state(jh2bh(jh
));
2154 spin_lock(&transaction
->t_journal
->j_list_lock
);
2155 __jbd2_journal_file_buffer(jh
, transaction
, jlist
);
2156 spin_unlock(&transaction
->t_journal
->j_list_lock
);
2157 jbd_unlock_bh_state(jh2bh(jh
));
2161 * Remove a buffer from its current buffer list in preparation for
2162 * dropping it from its current transaction entirely. If the buffer has
2163 * already started to be used by a subsequent transaction, refile the
2164 * buffer on that transaction's metadata list.
2166 * Called under j_list_lock
2167 * Called under jbd_lock_bh_state(jh2bh(jh))
2169 * jh and bh may be already free when this function returns
2171 void __jbd2_journal_refile_buffer(struct journal_head
*jh
)
2173 int was_dirty
, jlist
;
2174 struct buffer_head
*bh
= jh2bh(jh
);
2176 J_ASSERT_JH(jh
, jbd_is_locked_bh_state(bh
));
2177 if (jh
->b_transaction
)
2178 assert_spin_locked(&jh
->b_transaction
->t_journal
->j_list_lock
);
2180 /* If the buffer is now unused, just drop it. */
2181 if (jh
->b_next_transaction
== NULL
) {
2182 __jbd2_journal_unfile_buffer(jh
);
2187 * It has been modified by a later transaction: add it to the new
2188 * transaction's metadata list.
2191 was_dirty
= test_clear_buffer_jbddirty(bh
);
2192 __jbd2_journal_temp_unlink_buffer(jh
);
2194 * We set b_transaction here because b_next_transaction will inherit
2195 * our jh reference and thus __jbd2_journal_file_buffer() must not
2198 jh
->b_transaction
= jh
->b_next_transaction
;
2199 jh
->b_next_transaction
= NULL
;
2200 if (buffer_freed(bh
))
2202 else if (jh
->b_modified
)
2203 jlist
= BJ_Metadata
;
2205 jlist
= BJ_Reserved
;
2206 __jbd2_journal_file_buffer(jh
, jh
->b_transaction
, jlist
);
2207 J_ASSERT_JH(jh
, jh
->b_transaction
->t_state
== T_RUNNING
);
2210 set_buffer_jbddirty(bh
);
2214 * __jbd2_journal_refile_buffer() with necessary locking added. We take our
2215 * bh reference so that we can safely unlock bh.
2217 * The jh and bh may be freed by this call.
2219 void jbd2_journal_refile_buffer(journal_t
*journal
, struct journal_head
*jh
)
2221 struct buffer_head
*bh
= jh2bh(jh
);
2223 /* Get reference so that buffer cannot be freed before we unlock it */
2225 jbd_lock_bh_state(bh
);
2226 spin_lock(&journal
->j_list_lock
);
2227 __jbd2_journal_refile_buffer(jh
);
2228 jbd_unlock_bh_state(bh
);
2229 spin_unlock(&journal
->j_list_lock
);
2234 * File inode in the inode list of the handle's transaction
2236 int jbd2_journal_file_inode(handle_t
*handle
, struct jbd2_inode
*jinode
)
2238 transaction_t
*transaction
= handle
->h_transaction
;
2239 journal_t
*journal
= transaction
->t_journal
;
2241 if (is_handle_aborted(handle
))
2244 jbd_debug(4, "Adding inode %lu, tid:%d\n", jinode
->i_vfs_inode
->i_ino
,
2245 transaction
->t_tid
);
2248 * First check whether inode isn't already on the transaction's
2249 * lists without taking the lock. Note that this check is safe
2250 * without the lock as we cannot race with somebody removing inode
2251 * from the transaction. The reason is that we remove inode from the
2252 * transaction only in journal_release_jbd_inode() and when we commit
2253 * the transaction. We are guarded from the first case by holding
2254 * a reference to the inode. We are safe against the second case
2255 * because if jinode->i_transaction == transaction, commit code
2256 * cannot touch the transaction because we hold reference to it,
2257 * and if jinode->i_next_transaction == transaction, commit code
2258 * will only file the inode where we want it.
2260 if (jinode
->i_transaction
== transaction
||
2261 jinode
->i_next_transaction
== transaction
)
2264 spin_lock(&journal
->j_list_lock
);
2266 if (jinode
->i_transaction
== transaction
||
2267 jinode
->i_next_transaction
== transaction
)
2271 * We only ever set this variable to 1 so the test is safe. Since
2272 * t_need_data_flush is likely to be set, we do the test to save some
2273 * cacheline bouncing
2275 if (!transaction
->t_need_data_flush
)
2276 transaction
->t_need_data_flush
= 1;
2277 /* On some different transaction's list - should be
2278 * the committing one */
2279 if (jinode
->i_transaction
) {
2280 J_ASSERT(jinode
->i_next_transaction
== NULL
);
2281 J_ASSERT(jinode
->i_transaction
==
2282 journal
->j_committing_transaction
);
2283 jinode
->i_next_transaction
= transaction
;
2286 /* Not on any transaction list... */
2287 J_ASSERT(!jinode
->i_next_transaction
);
2288 jinode
->i_transaction
= transaction
;
2289 list_add(&jinode
->i_list
, &transaction
->t_inode_list
);
2291 spin_unlock(&journal
->j_list_lock
);
2297 * File truncate and transaction commit interact with each other in a
2298 * non-trivial way. If a transaction writing data block A is
2299 * committing, we cannot discard the data by truncate until we have
2300 * written them. Otherwise if we crashed after the transaction with
2301 * write has committed but before the transaction with truncate has
2302 * committed, we could see stale data in block A. This function is a
2303 * helper to solve this problem. It starts writeout of the truncated
2304 * part in case it is in the committing transaction.
2306 * Filesystem code must call this function when inode is journaled in
2307 * ordered mode before truncation happens and after the inode has been
2308 * placed on orphan list with the new inode size. The second condition
2309 * avoids the race that someone writes new data and we start
2310 * committing the transaction after this function has been called but
2311 * before a transaction for truncate is started (and furthermore it
2312 * allows us to optimize the case where the addition to orphan list
2313 * happens in the same transaction as write --- we don't have to write
2314 * any data in such case).
2316 int jbd2_journal_begin_ordered_truncate(journal_t
*journal
,
2317 struct jbd2_inode
*jinode
,
2320 transaction_t
*inode_trans
, *commit_trans
;
2323 /* This is a quick check to avoid locking if not necessary */
2324 if (!jinode
->i_transaction
)
2326 /* Locks are here just to force reading of recent values, it is
2327 * enough that the transaction was not committing before we started
2328 * a transaction adding the inode to orphan list */
2329 read_lock(&journal
->j_state_lock
);
2330 commit_trans
= journal
->j_committing_transaction
;
2331 read_unlock(&journal
->j_state_lock
);
2332 spin_lock(&journal
->j_list_lock
);
2333 inode_trans
= jinode
->i_transaction
;
2334 spin_unlock(&journal
->j_list_lock
);
2335 if (inode_trans
== commit_trans
) {
2336 ret
= filemap_fdatawrite_range(jinode
->i_vfs_inode
->i_mapping
,
2337 new_size
, LLONG_MAX
);
2339 jbd2_journal_abort(journal
, ret
);