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
,
93 atomic_read(&journal
->j_reserved_credits
));
94 atomic_set(&transaction
->t_handle_count
, 0);
95 INIT_LIST_HEAD(&transaction
->t_inode_list
);
96 INIT_LIST_HEAD(&transaction
->t_private_list
);
98 /* Set up the commit timer for the new transaction. */
99 journal
->j_commit_timer
.expires
= round_jiffies_up(transaction
->t_expires
);
100 add_timer(&journal
->j_commit_timer
);
102 J_ASSERT(journal
->j_running_transaction
== NULL
);
103 journal
->j_running_transaction
= transaction
;
104 transaction
->t_max_wait
= 0;
105 transaction
->t_start
= jiffies
;
106 transaction
->t_requested
= 0;
114 * A handle_t is an object which represents a single atomic update to a
115 * filesystem, and which tracks all of the modifications which form part
116 * of that one update.
120 * Update transaction's maximum wait time, if debugging is enabled.
122 * In order for t_max_wait to be reliable, it must be protected by a
123 * lock. But doing so will mean that start_this_handle() can not be
124 * run in parallel on SMP systems, which limits our scalability. So
125 * unless debugging is enabled, we no longer update t_max_wait, which
126 * means that maximum wait time reported by the jbd2_run_stats
127 * tracepoint will always be zero.
129 static inline void update_t_max_wait(transaction_t
*transaction
,
132 #ifdef CONFIG_JBD2_DEBUG
133 if (jbd2_journal_enable_debug
&&
134 time_after(transaction
->t_start
, ts
)) {
135 ts
= jbd2_time_diff(ts
, transaction
->t_start
);
136 spin_lock(&transaction
->t_handle_lock
);
137 if (ts
> transaction
->t_max_wait
)
138 transaction
->t_max_wait
= ts
;
139 spin_unlock(&transaction
->t_handle_lock
);
145 * Wait until running transaction passes T_LOCKED state. Also starts the commit
146 * if needed. The function expects running transaction to exist and releases
149 static void wait_transaction_locked(journal_t
*journal
)
150 __releases(journal
->j_state_lock
)
154 tid_t tid
= journal
->j_running_transaction
->t_tid
;
156 prepare_to_wait(&journal
->j_wait_transaction_locked
, &wait
,
157 TASK_UNINTERRUPTIBLE
);
158 need_to_start
= !tid_geq(journal
->j_commit_request
, tid
);
159 read_unlock(&journal
->j_state_lock
);
161 jbd2_log_start_commit(journal
, tid
);
163 finish_wait(&journal
->j_wait_transaction_locked
, &wait
);
166 static void sub_reserved_credits(journal_t
*journal
, int blocks
)
168 atomic_sub(blocks
, &journal
->j_reserved_credits
);
169 wake_up(&journal
->j_wait_reserved
);
173 * Wait until we can add credits for handle to the running transaction. Called
174 * with j_state_lock held for reading. Returns 0 if handle joined the running
175 * transaction. Returns 1 if we had to wait, j_state_lock is dropped, and
178 static int add_transaction_credits(journal_t
*journal
, int blocks
,
181 transaction_t
*t
= journal
->j_running_transaction
;
183 int total
= blocks
+ rsv_blocks
;
186 * If the current transaction is locked down for commit, wait
187 * for the lock to be released.
189 if (t
->t_state
== T_LOCKED
) {
190 wait_transaction_locked(journal
);
195 * If there is not enough space left in the log to write all
196 * potential buffers requested by this operation, we need to
197 * stall pending a log checkpoint to free some more log space.
199 needed
= atomic_add_return(total
, &t
->t_outstanding_credits
);
200 if (needed
> journal
->j_max_transaction_buffers
) {
202 * If the current transaction is already too large,
203 * then start to commit it: we can then go back and
204 * attach this handle to a new transaction.
206 atomic_sub(total
, &t
->t_outstanding_credits
);
207 wait_transaction_locked(journal
);
212 * The commit code assumes that it can get enough log space
213 * without forcing a checkpoint. This is *critical* for
214 * correctness: a checkpoint of a buffer which is also
215 * associated with a committing transaction creates a deadlock,
216 * so commit simply cannot force through checkpoints.
218 * We must therefore ensure the necessary space in the journal
219 * *before* starting to dirty potentially checkpointed buffers
220 * in the new transaction.
222 if (jbd2_log_space_left(journal
) < jbd2_space_needed(journal
)) {
223 atomic_sub(total
, &t
->t_outstanding_credits
);
224 read_unlock(&journal
->j_state_lock
);
225 write_lock(&journal
->j_state_lock
);
226 if (jbd2_log_space_left(journal
) < jbd2_space_needed(journal
))
227 __jbd2_log_wait_for_space(journal
);
228 write_unlock(&journal
->j_state_lock
);
232 /* No reservation? We are done... */
236 needed
= atomic_add_return(rsv_blocks
, &journal
->j_reserved_credits
);
237 /* We allow at most half of a transaction to be reserved */
238 if (needed
> journal
->j_max_transaction_buffers
/ 2) {
239 sub_reserved_credits(journal
, rsv_blocks
);
240 atomic_sub(total
, &t
->t_outstanding_credits
);
241 read_unlock(&journal
->j_state_lock
);
242 wait_event(journal
->j_wait_reserved
,
243 atomic_read(&journal
->j_reserved_credits
) + rsv_blocks
244 <= journal
->j_max_transaction_buffers
/ 2);
251 * start_this_handle: Given a handle, deal with any locking or stalling
252 * needed to make sure that there is enough journal space for the handle
253 * to begin. Attach the handle to a transaction and set up the
254 * transaction's buffer credits.
257 static int start_this_handle(journal_t
*journal
, handle_t
*handle
,
260 transaction_t
*transaction
, *new_transaction
= NULL
;
261 int blocks
= handle
->h_buffer_credits
;
263 unsigned long ts
= jiffies
;
266 * 1/2 of transaction can be reserved so we can practically handle
267 * only 1/2 of maximum transaction size per operation
269 if (WARN_ON(blocks
> journal
->j_max_transaction_buffers
/ 2)) {
270 printk(KERN_ERR
"JBD2: %s wants too many credits (%d > %d)\n",
271 current
->comm
, blocks
,
272 journal
->j_max_transaction_buffers
/ 2);
276 if (handle
->h_rsv_handle
)
277 rsv_blocks
= handle
->h_rsv_handle
->h_buffer_credits
;
280 if (!journal
->j_running_transaction
) {
281 new_transaction
= kmem_cache_zalloc(transaction_cache
,
283 if (!new_transaction
) {
285 * If __GFP_FS is not present, then we may be
286 * being called from inside the fs writeback
287 * layer, so we MUST NOT fail. Since
288 * __GFP_NOFAIL is going away, we will arrange
289 * to retry the allocation ourselves.
291 if ((gfp_mask
& __GFP_FS
) == 0) {
292 congestion_wait(BLK_RW_ASYNC
, HZ
/50);
293 goto alloc_transaction
;
299 jbd_debug(3, "New handle %p going live.\n", handle
);
302 * We need to hold j_state_lock until t_updates has been incremented,
303 * for proper journal barrier handling
306 read_lock(&journal
->j_state_lock
);
307 BUG_ON(journal
->j_flags
& JBD2_UNMOUNT
);
308 if (is_journal_aborted(journal
) ||
309 (journal
->j_errno
!= 0 && !(journal
->j_flags
& JBD2_ACK_ERR
))) {
310 read_unlock(&journal
->j_state_lock
);
311 jbd2_journal_free_transaction(new_transaction
);
316 * Wait on the journal's transaction barrier if necessary. Specifically
317 * we allow reserved handles to proceed because otherwise commit could
318 * deadlock on page writeback not being able to complete.
320 if (!handle
->h_reserved
&& journal
->j_barrier_count
) {
321 read_unlock(&journal
->j_state_lock
);
322 wait_event(journal
->j_wait_transaction_locked
,
323 journal
->j_barrier_count
== 0);
327 if (!journal
->j_running_transaction
) {
328 read_unlock(&journal
->j_state_lock
);
329 if (!new_transaction
)
330 goto alloc_transaction
;
331 write_lock(&journal
->j_state_lock
);
332 if (!journal
->j_running_transaction
&&
333 (handle
->h_reserved
|| !journal
->j_barrier_count
)) {
334 jbd2_get_transaction(journal
, new_transaction
);
335 new_transaction
= NULL
;
337 write_unlock(&journal
->j_state_lock
);
341 transaction
= journal
->j_running_transaction
;
343 if (!handle
->h_reserved
) {
344 /* We may have dropped j_state_lock - restart in that case */
345 if (add_transaction_credits(journal
, blocks
, rsv_blocks
))
349 * We have handle reserved so we are allowed to join T_LOCKED
350 * transaction and we don't have to check for transaction size
353 sub_reserved_credits(journal
, blocks
);
354 handle
->h_reserved
= 0;
357 /* OK, account for the buffers that this operation expects to
358 * use and add the handle to the running transaction.
360 update_t_max_wait(transaction
, ts
);
361 handle
->h_transaction
= transaction
;
362 handle
->h_requested_credits
= blocks
;
363 handle
->h_start_jiffies
= jiffies
;
364 atomic_inc(&transaction
->t_updates
);
365 atomic_inc(&transaction
->t_handle_count
);
366 jbd_debug(4, "Handle %p given %d credits (total %d, free %lu)\n",
368 atomic_read(&transaction
->t_outstanding_credits
),
369 jbd2_log_space_left(journal
));
370 read_unlock(&journal
->j_state_lock
);
372 lock_map_acquire(&handle
->h_lockdep_map
);
373 jbd2_journal_free_transaction(new_transaction
);
377 static struct lock_class_key jbd2_handle_key
;
379 /* Allocate a new handle. This should probably be in a slab... */
380 static handle_t
*new_handle(int nblocks
)
382 handle_t
*handle
= jbd2_alloc_handle(GFP_NOFS
);
385 handle
->h_buffer_credits
= nblocks
;
388 lockdep_init_map(&handle
->h_lockdep_map
, "jbd2_handle",
389 &jbd2_handle_key
, 0);
395 * handle_t *jbd2_journal_start() - Obtain a new handle.
396 * @journal: Journal to start transaction on.
397 * @nblocks: number of block buffer we might modify
399 * We make sure that the transaction can guarantee at least nblocks of
400 * modified buffers in the log. We block until the log can guarantee
401 * that much space. Additionally, if rsv_blocks > 0, we also create another
402 * handle with rsv_blocks reserved blocks in the journal. This handle is
403 * is stored in h_rsv_handle. It is not attached to any particular transaction
404 * and thus doesn't block transaction commit. If the caller uses this reserved
405 * handle, it has to set h_rsv_handle to NULL as otherwise jbd2_journal_stop()
406 * on the parent handle will dispose the reserved one. Reserved handle has to
407 * be converted to a normal handle using jbd2_journal_start_reserved() before
410 * Return a pointer to a newly allocated handle, or an ERR_PTR() value
413 handle_t
*jbd2__journal_start(journal_t
*journal
, int nblocks
, int rsv_blocks
,
414 gfp_t gfp_mask
, unsigned int type
,
415 unsigned int line_no
)
417 handle_t
*handle
= journal_current_handle();
421 return ERR_PTR(-EROFS
);
424 J_ASSERT(handle
->h_transaction
->t_journal
== journal
);
429 handle
= new_handle(nblocks
);
431 return ERR_PTR(-ENOMEM
);
433 handle_t
*rsv_handle
;
435 rsv_handle
= new_handle(rsv_blocks
);
437 jbd2_free_handle(handle
);
438 return ERR_PTR(-ENOMEM
);
440 rsv_handle
->h_reserved
= 1;
441 rsv_handle
->h_journal
= journal
;
442 handle
->h_rsv_handle
= rsv_handle
;
445 current
->journal_info
= handle
;
447 err
= start_this_handle(journal
, handle
, gfp_mask
);
449 if (handle
->h_rsv_handle
)
450 jbd2_free_handle(handle
->h_rsv_handle
);
451 jbd2_free_handle(handle
);
452 current
->journal_info
= NULL
;
455 handle
->h_type
= type
;
456 handle
->h_line_no
= line_no
;
457 trace_jbd2_handle_start(journal
->j_fs_dev
->bd_dev
,
458 handle
->h_transaction
->t_tid
, type
,
462 EXPORT_SYMBOL(jbd2__journal_start
);
465 handle_t
*jbd2_journal_start(journal_t
*journal
, int nblocks
)
467 return jbd2__journal_start(journal
, nblocks
, 0, GFP_NOFS
, 0, 0);
469 EXPORT_SYMBOL(jbd2_journal_start
);
471 void jbd2_journal_free_reserved(handle_t
*handle
)
473 journal_t
*journal
= handle
->h_journal
;
475 WARN_ON(!handle
->h_reserved
);
476 sub_reserved_credits(journal
, handle
->h_buffer_credits
);
477 jbd2_free_handle(handle
);
479 EXPORT_SYMBOL(jbd2_journal_free_reserved
);
482 * int jbd2_journal_start_reserved(handle_t *handle) - start reserved handle
483 * @handle: handle to start
485 * Start handle that has been previously reserved with jbd2_journal_reserve().
486 * This attaches @handle to the running transaction (or creates one if there's
487 * not transaction running). Unlike jbd2_journal_start() this function cannot
488 * block on journal commit, checkpointing, or similar stuff. It can block on
489 * memory allocation or frozen journal though.
491 * Return 0 on success, non-zero on error - handle is freed in that case.
493 int jbd2_journal_start_reserved(handle_t
*handle
, unsigned int type
,
494 unsigned int line_no
)
496 journal_t
*journal
= handle
->h_journal
;
499 if (WARN_ON(!handle
->h_reserved
)) {
500 /* Someone passed in normal handle? Just stop it. */
501 jbd2_journal_stop(handle
);
505 * Usefulness of mixing of reserved and unreserved handles is
506 * questionable. So far nobody seems to need it so just error out.
508 if (WARN_ON(current
->journal_info
)) {
509 jbd2_journal_free_reserved(handle
);
513 handle
->h_journal
= NULL
;
514 current
->journal_info
= handle
;
516 * GFP_NOFS is here because callers are likely from writeback or
517 * similarly constrained call sites
519 ret
= start_this_handle(journal
, handle
, GFP_NOFS
);
521 current
->journal_info
= NULL
;
522 jbd2_journal_free_reserved(handle
);
524 handle
->h_type
= type
;
525 handle
->h_line_no
= line_no
;
528 EXPORT_SYMBOL(jbd2_journal_start_reserved
);
531 * int jbd2_journal_extend() - extend buffer credits.
532 * @handle: handle to 'extend'
533 * @nblocks: nr blocks to try to extend by.
535 * Some transactions, such as large extends and truncates, can be done
536 * atomically all at once or in several stages. The operation requests
537 * a credit for a number of buffer modications in advance, but can
538 * extend its credit if it needs more.
540 * jbd2_journal_extend tries to give the running handle more buffer credits.
541 * It does not guarantee that allocation - this is a best-effort only.
542 * The calling process MUST be able to deal cleanly with a failure to
545 * Return 0 on success, non-zero on failure.
547 * return code < 0 implies an error
548 * return code > 0 implies normal transaction-full status.
550 int jbd2_journal_extend(handle_t
*handle
, int nblocks
)
552 transaction_t
*transaction
= handle
->h_transaction
;
553 journal_t
*journal
= transaction
->t_journal
;
558 if (is_handle_aborted(handle
))
563 read_lock(&journal
->j_state_lock
);
565 /* Don't extend a locked-down transaction! */
566 if (handle
->h_transaction
->t_state
!= T_RUNNING
) {
567 jbd_debug(3, "denied handle %p %d blocks: "
568 "transaction not running\n", handle
, nblocks
);
572 spin_lock(&transaction
->t_handle_lock
);
573 wanted
= atomic_add_return(nblocks
,
574 &transaction
->t_outstanding_credits
);
576 if (wanted
> journal
->j_max_transaction_buffers
) {
577 jbd_debug(3, "denied handle %p %d blocks: "
578 "transaction too large\n", handle
, nblocks
);
579 atomic_sub(nblocks
, &transaction
->t_outstanding_credits
);
583 if (wanted
+ (wanted
>> JBD2_CONTROL_BLOCKS_SHIFT
) >
584 jbd2_log_space_left(journal
)) {
585 jbd_debug(3, "denied handle %p %d blocks: "
586 "insufficient log space\n", handle
, nblocks
);
587 atomic_sub(nblocks
, &transaction
->t_outstanding_credits
);
591 trace_jbd2_handle_extend(journal
->j_fs_dev
->bd_dev
,
592 handle
->h_transaction
->t_tid
,
593 handle
->h_type
, handle
->h_line_no
,
594 handle
->h_buffer_credits
,
597 handle
->h_buffer_credits
+= nblocks
;
598 handle
->h_requested_credits
+= nblocks
;
601 jbd_debug(3, "extended handle %p by %d\n", handle
, nblocks
);
603 spin_unlock(&transaction
->t_handle_lock
);
605 read_unlock(&journal
->j_state_lock
);
612 * int jbd2_journal_restart() - restart a handle .
613 * @handle: handle to restart
614 * @nblocks: nr credits requested
616 * Restart a handle for a multi-transaction filesystem
619 * If the jbd2_journal_extend() call above fails to grant new buffer credits
620 * to a running handle, a call to jbd2_journal_restart will commit the
621 * handle's transaction so far and reattach the handle to a new
622 * transaction capabable of guaranteeing the requested number of
623 * credits. We preserve reserved handle if there's any attached to the
626 int jbd2__journal_restart(handle_t
*handle
, int nblocks
, gfp_t gfp_mask
)
628 transaction_t
*transaction
= handle
->h_transaction
;
629 journal_t
*journal
= transaction
->t_journal
;
631 int need_to_start
, ret
;
633 /* If we've had an abort of any type, don't even think about
634 * actually doing the restart! */
635 if (is_handle_aborted(handle
))
639 * First unlink the handle from its current transaction, and start the
642 J_ASSERT(atomic_read(&transaction
->t_updates
) > 0);
643 J_ASSERT(journal_current_handle() == handle
);
645 read_lock(&journal
->j_state_lock
);
646 spin_lock(&transaction
->t_handle_lock
);
647 atomic_sub(handle
->h_buffer_credits
,
648 &transaction
->t_outstanding_credits
);
649 if (handle
->h_rsv_handle
) {
650 sub_reserved_credits(journal
,
651 handle
->h_rsv_handle
->h_buffer_credits
);
653 if (atomic_dec_and_test(&transaction
->t_updates
))
654 wake_up(&journal
->j_wait_updates
);
655 tid
= transaction
->t_tid
;
656 spin_unlock(&transaction
->t_handle_lock
);
658 jbd_debug(2, "restarting handle %p\n", handle
);
659 need_to_start
= !tid_geq(journal
->j_commit_request
, tid
);
660 read_unlock(&journal
->j_state_lock
);
662 jbd2_log_start_commit(journal
, tid
);
664 lock_map_release(&handle
->h_lockdep_map
);
665 handle
->h_buffer_credits
= nblocks
;
666 ret
= start_this_handle(journal
, handle
, gfp_mask
);
669 EXPORT_SYMBOL(jbd2__journal_restart
);
672 int jbd2_journal_restart(handle_t
*handle
, int nblocks
)
674 return jbd2__journal_restart(handle
, nblocks
, GFP_NOFS
);
676 EXPORT_SYMBOL(jbd2_journal_restart
);
679 * void jbd2_journal_lock_updates () - establish a transaction barrier.
680 * @journal: Journal to establish a barrier on.
682 * This locks out any further updates from being started, and blocks
683 * until all existing updates have completed, returning only once the
684 * journal is in a quiescent state with no updates running.
686 * The journal lock should not be held on entry.
688 void jbd2_journal_lock_updates(journal_t
*journal
)
692 write_lock(&journal
->j_state_lock
);
693 ++journal
->j_barrier_count
;
695 /* Wait until there are no reserved handles */
696 if (atomic_read(&journal
->j_reserved_credits
)) {
697 write_unlock(&journal
->j_state_lock
);
698 wait_event(journal
->j_wait_reserved
,
699 atomic_read(&journal
->j_reserved_credits
) == 0);
700 write_lock(&journal
->j_state_lock
);
703 /* Wait until there are no running updates */
705 transaction_t
*transaction
= journal
->j_running_transaction
;
710 spin_lock(&transaction
->t_handle_lock
);
711 prepare_to_wait(&journal
->j_wait_updates
, &wait
,
712 TASK_UNINTERRUPTIBLE
);
713 if (!atomic_read(&transaction
->t_updates
)) {
714 spin_unlock(&transaction
->t_handle_lock
);
715 finish_wait(&journal
->j_wait_updates
, &wait
);
718 spin_unlock(&transaction
->t_handle_lock
);
719 write_unlock(&journal
->j_state_lock
);
721 finish_wait(&journal
->j_wait_updates
, &wait
);
722 write_lock(&journal
->j_state_lock
);
724 write_unlock(&journal
->j_state_lock
);
727 * We have now established a barrier against other normal updates, but
728 * we also need to barrier against other jbd2_journal_lock_updates() calls
729 * to make sure that we serialise special journal-locked operations
732 mutex_lock(&journal
->j_barrier
);
736 * void jbd2_journal_unlock_updates (journal_t* journal) - release barrier
737 * @journal: Journal to release the barrier on.
739 * Release a transaction barrier obtained with jbd2_journal_lock_updates().
741 * Should be called without the journal lock held.
743 void jbd2_journal_unlock_updates (journal_t
*journal
)
745 J_ASSERT(journal
->j_barrier_count
!= 0);
747 mutex_unlock(&journal
->j_barrier
);
748 write_lock(&journal
->j_state_lock
);
749 --journal
->j_barrier_count
;
750 write_unlock(&journal
->j_state_lock
);
751 wake_up(&journal
->j_wait_transaction_locked
);
754 static void warn_dirty_buffer(struct buffer_head
*bh
)
756 char b
[BDEVNAME_SIZE
];
759 "JBD2: Spotted dirty metadata buffer (dev = %s, blocknr = %llu). "
760 "There's a risk of filesystem corruption in case of system "
762 bdevname(bh
->b_bdev
, b
), (unsigned long long)bh
->b_blocknr
);
765 static int sleep_on_shadow_bh(void *word
)
772 * If the buffer is already part of the current transaction, then there
773 * is nothing we need to do. If it is already part of a prior
774 * transaction which we are still committing to disk, then we need to
775 * make sure that we do not overwrite the old copy: we do copy-out to
776 * preserve the copy going to disk. We also account the buffer against
777 * the handle's metadata buffer credits (unless the buffer is already
778 * part of the transaction, that is).
782 do_get_write_access(handle_t
*handle
, struct journal_head
*jh
,
785 struct buffer_head
*bh
;
786 transaction_t
*transaction
;
789 char *frozen_buffer
= NULL
;
791 unsigned long start_lock
, time_lock
;
793 if (is_handle_aborted(handle
))
796 transaction
= handle
->h_transaction
;
797 journal
= transaction
->t_journal
;
799 jbd_debug(5, "journal_head %p, force_copy %d\n", jh
, force_copy
);
801 JBUFFER_TRACE(jh
, "entry");
805 /* @@@ Need to check for errors here at some point. */
807 start_lock
= jiffies
;
809 jbd_lock_bh_state(bh
);
811 /* If it takes too long to lock the buffer, trace it */
812 time_lock
= jbd2_time_diff(start_lock
, jiffies
);
813 if (time_lock
> HZ
/10)
814 trace_jbd2_lock_buffer_stall(bh
->b_bdev
->bd_dev
,
815 jiffies_to_msecs(time_lock
));
817 /* We now hold the buffer lock so it is safe to query the buffer
818 * state. Is the buffer dirty?
820 * If so, there are two possibilities. The buffer may be
821 * non-journaled, and undergoing a quite legitimate writeback.
822 * Otherwise, it is journaled, and we don't expect dirty buffers
823 * in that state (the buffers should be marked JBD_Dirty
824 * instead.) So either the IO is being done under our own
825 * control and this is a bug, or it's a third party IO such as
826 * dump(8) (which may leave the buffer scheduled for read ---
827 * ie. locked but not dirty) or tune2fs (which may actually have
828 * the buffer dirtied, ugh.) */
830 if (buffer_dirty(bh
)) {
832 * First question: is this buffer already part of the current
833 * transaction or the existing committing transaction?
835 if (jh
->b_transaction
) {
837 jh
->b_transaction
== transaction
||
839 journal
->j_committing_transaction
);
840 if (jh
->b_next_transaction
)
841 J_ASSERT_JH(jh
, jh
->b_next_transaction
==
843 warn_dirty_buffer(bh
);
846 * In any case we need to clean the dirty flag and we must
847 * do it under the buffer lock to be sure we don't race
848 * with running write-out.
850 JBUFFER_TRACE(jh
, "Journalling dirty buffer");
851 clear_buffer_dirty(bh
);
852 set_buffer_jbddirty(bh
);
858 if (is_handle_aborted(handle
)) {
859 jbd_unlock_bh_state(bh
);
865 * The buffer is already part of this transaction if b_transaction or
866 * b_next_transaction points to it
868 if (jh
->b_transaction
== transaction
||
869 jh
->b_next_transaction
== transaction
)
873 * this is the first time this transaction is touching this buffer,
874 * reset the modified flag
879 * If there is already a copy-out version of this buffer, then we don't
880 * need to make another one
882 if (jh
->b_frozen_data
) {
883 JBUFFER_TRACE(jh
, "has frozen data");
884 J_ASSERT_JH(jh
, jh
->b_next_transaction
== NULL
);
885 jh
->b_next_transaction
= transaction
;
889 /* Is there data here we need to preserve? */
891 if (jh
->b_transaction
&& jh
->b_transaction
!= transaction
) {
892 JBUFFER_TRACE(jh
, "owned by older transaction");
893 J_ASSERT_JH(jh
, jh
->b_next_transaction
== NULL
);
894 J_ASSERT_JH(jh
, jh
->b_transaction
==
895 journal
->j_committing_transaction
);
897 /* There is one case we have to be very careful about.
898 * If the committing transaction is currently writing
899 * this buffer out to disk and has NOT made a copy-out,
900 * then we cannot modify the buffer contents at all
901 * right now. The essence of copy-out is that it is the
902 * extra copy, not the primary copy, which gets
903 * journaled. If the primary copy is already going to
904 * disk then we cannot do copy-out here. */
906 if (buffer_shadow(bh
)) {
907 JBUFFER_TRACE(jh
, "on shadow: sleep");
908 jbd_unlock_bh_state(bh
);
909 wait_on_bit(&bh
->b_state
, BH_Shadow
,
910 sleep_on_shadow_bh
, TASK_UNINTERRUPTIBLE
);
915 * Only do the copy if the currently-owning transaction still
916 * needs it. If buffer isn't on BJ_Metadata list, the
917 * committing transaction is past that stage (here we use the
918 * fact that BH_Shadow is set under bh_state lock together with
919 * refiling to BJ_Shadow list and at this point we know the
920 * buffer doesn't have BH_Shadow set).
922 * Subtle point, though: if this is a get_undo_access,
923 * then we will be relying on the frozen_data to contain
924 * the new value of the committed_data record after the
925 * transaction, so we HAVE to force the frozen_data copy
928 if (jh
->b_jlist
== BJ_Metadata
|| force_copy
) {
929 JBUFFER_TRACE(jh
, "generate frozen data");
930 if (!frozen_buffer
) {
931 JBUFFER_TRACE(jh
, "allocate memory for buffer");
932 jbd_unlock_bh_state(bh
);
934 jbd2_alloc(jh2bh(jh
)->b_size
,
936 if (!frozen_buffer
) {
938 "%s: OOM for frozen_buffer\n",
940 JBUFFER_TRACE(jh
, "oom!");
942 jbd_lock_bh_state(bh
);
947 jh
->b_frozen_data
= frozen_buffer
;
948 frozen_buffer
= NULL
;
951 jh
->b_next_transaction
= transaction
;
956 * Finally, if the buffer is not journaled right now, we need to make
957 * sure it doesn't get written to disk before the caller actually
958 * commits the new data
960 if (!jh
->b_transaction
) {
961 JBUFFER_TRACE(jh
, "no transaction");
962 J_ASSERT_JH(jh
, !jh
->b_next_transaction
);
963 JBUFFER_TRACE(jh
, "file as BJ_Reserved");
964 spin_lock(&journal
->j_list_lock
);
965 __jbd2_journal_file_buffer(jh
, transaction
, BJ_Reserved
);
966 spin_unlock(&journal
->j_list_lock
);
975 J_EXPECT_JH(jh
, buffer_uptodate(jh2bh(jh
)),
976 "Possible IO failure.\n");
977 page
= jh2bh(jh
)->b_page
;
978 offset
= offset_in_page(jh2bh(jh
)->b_data
);
979 source
= kmap_atomic(page
);
980 /* Fire data frozen trigger just before we copy the data */
981 jbd2_buffer_frozen_trigger(jh
, source
+ offset
,
983 memcpy(jh
->b_frozen_data
, source
+offset
, jh2bh(jh
)->b_size
);
984 kunmap_atomic(source
);
987 * Now that the frozen data is saved off, we need to store
988 * any matching triggers.
990 jh
->b_frozen_triggers
= jh
->b_triggers
;
992 jbd_unlock_bh_state(bh
);
995 * If we are about to journal a buffer, then any revoke pending on it is
998 jbd2_journal_cancel_revoke(handle
, jh
);
1001 if (unlikely(frozen_buffer
)) /* It's usually NULL */
1002 jbd2_free(frozen_buffer
, bh
->b_size
);
1004 JBUFFER_TRACE(jh
, "exit");
1009 * int jbd2_journal_get_write_access() - notify intent to modify a buffer for metadata (not data) update.
1010 * @handle: transaction to add buffer modifications to
1011 * @bh: bh to be used for metadata writes
1013 * Returns an error code or 0 on success.
1015 * In full data journalling mode the buffer may be of type BJ_AsyncData,
1016 * because we're write()ing a buffer which is also part of a shared mapping.
1019 int jbd2_journal_get_write_access(handle_t
*handle
, struct buffer_head
*bh
)
1021 struct journal_head
*jh
= jbd2_journal_add_journal_head(bh
);
1024 /* We do not want to get caught playing with fields which the
1025 * log thread also manipulates. Make sure that the buffer
1026 * completes any outstanding IO before proceeding. */
1027 rc
= do_get_write_access(handle
, jh
, 0);
1028 jbd2_journal_put_journal_head(jh
);
1034 * When the user wants to journal a newly created buffer_head
1035 * (ie. getblk() returned a new buffer and we are going to populate it
1036 * manually rather than reading off disk), then we need to keep the
1037 * buffer_head locked until it has been completely filled with new
1038 * data. In this case, we should be able to make the assertion that
1039 * the bh is not already part of an existing transaction.
1041 * The buffer should already be locked by the caller by this point.
1042 * There is no lock ranking violation: it was a newly created,
1043 * unlocked buffer beforehand. */
1046 * int jbd2_journal_get_create_access () - notify intent to use newly created bh
1047 * @handle: transaction to new buffer to
1050 * Call this if you create a new bh.
1052 int jbd2_journal_get_create_access(handle_t
*handle
, struct buffer_head
*bh
)
1054 transaction_t
*transaction
= handle
->h_transaction
;
1055 journal_t
*journal
= transaction
->t_journal
;
1056 struct journal_head
*jh
= jbd2_journal_add_journal_head(bh
);
1059 jbd_debug(5, "journal_head %p\n", jh
);
1061 if (is_handle_aborted(handle
))
1065 JBUFFER_TRACE(jh
, "entry");
1067 * The buffer may already belong to this transaction due to pre-zeroing
1068 * in the filesystem's new_block code. It may also be on the previous,
1069 * committing transaction's lists, but it HAS to be in Forget state in
1070 * that case: the transaction must have deleted the buffer for it to be
1073 jbd_lock_bh_state(bh
);
1074 spin_lock(&journal
->j_list_lock
);
1075 J_ASSERT_JH(jh
, (jh
->b_transaction
== transaction
||
1076 jh
->b_transaction
== NULL
||
1077 (jh
->b_transaction
== journal
->j_committing_transaction
&&
1078 jh
->b_jlist
== BJ_Forget
)));
1080 J_ASSERT_JH(jh
, jh
->b_next_transaction
== NULL
);
1081 J_ASSERT_JH(jh
, buffer_locked(jh2bh(jh
)));
1083 if (jh
->b_transaction
== NULL
) {
1085 * Previous jbd2_journal_forget() could have left the buffer
1086 * with jbddirty bit set because it was being committed. When
1087 * the commit finished, we've filed the buffer for
1088 * checkpointing and marked it dirty. Now we are reallocating
1089 * the buffer so the transaction freeing it must have
1090 * committed and so it's safe to clear the dirty bit.
1092 clear_buffer_dirty(jh2bh(jh
));
1093 /* first access by this transaction */
1096 JBUFFER_TRACE(jh
, "file as BJ_Reserved");
1097 __jbd2_journal_file_buffer(jh
, transaction
, BJ_Reserved
);
1098 } else if (jh
->b_transaction
== journal
->j_committing_transaction
) {
1099 /* first access by this transaction */
1102 JBUFFER_TRACE(jh
, "set next transaction");
1103 jh
->b_next_transaction
= transaction
;
1105 spin_unlock(&journal
->j_list_lock
);
1106 jbd_unlock_bh_state(bh
);
1109 * akpm: I added this. ext3_alloc_branch can pick up new indirect
1110 * blocks which contain freed but then revoked metadata. We need
1111 * to cancel the revoke in case we end up freeing it yet again
1112 * and the reallocating as data - this would cause a second revoke,
1113 * which hits an assertion error.
1115 JBUFFER_TRACE(jh
, "cancelling revoke");
1116 jbd2_journal_cancel_revoke(handle
, jh
);
1118 jbd2_journal_put_journal_head(jh
);
1123 * int jbd2_journal_get_undo_access() - Notify intent to modify metadata with
1124 * non-rewindable consequences
1125 * @handle: transaction
1126 * @bh: buffer to undo
1128 * Sometimes there is a need to distinguish between metadata which has
1129 * been committed to disk and that which has not. The ext3fs code uses
1130 * this for freeing and allocating space, we have to make sure that we
1131 * do not reuse freed space until the deallocation has been committed,
1132 * since if we overwrote that space we would make the delete
1133 * un-rewindable in case of a crash.
1135 * To deal with that, jbd2_journal_get_undo_access requests write access to a
1136 * buffer for parts of non-rewindable operations such as delete
1137 * operations on the bitmaps. The journaling code must keep a copy of
1138 * the buffer's contents prior to the undo_access call until such time
1139 * as we know that the buffer has definitely been committed to disk.
1141 * We never need to know which transaction the committed data is part
1142 * of, buffers touched here are guaranteed to be dirtied later and so
1143 * will be committed to a new transaction in due course, at which point
1144 * we can discard the old committed data pointer.
1146 * Returns error number or 0 on success.
1148 int jbd2_journal_get_undo_access(handle_t
*handle
, struct buffer_head
*bh
)
1151 struct journal_head
*jh
= jbd2_journal_add_journal_head(bh
);
1152 char *committed_data
= NULL
;
1154 JBUFFER_TRACE(jh
, "entry");
1157 * Do this first --- it can drop the journal lock, so we want to
1158 * make sure that obtaining the committed_data is done
1159 * atomically wrt. completion of any outstanding commits.
1161 err
= do_get_write_access(handle
, jh
, 1);
1166 if (!jh
->b_committed_data
) {
1167 committed_data
= jbd2_alloc(jh2bh(jh
)->b_size
, GFP_NOFS
);
1168 if (!committed_data
) {
1169 printk(KERN_EMERG
"%s: No memory for committed data\n",
1176 jbd_lock_bh_state(bh
);
1177 if (!jh
->b_committed_data
) {
1178 /* Copy out the current buffer contents into the
1179 * preserved, committed copy. */
1180 JBUFFER_TRACE(jh
, "generate b_committed data");
1181 if (!committed_data
) {
1182 jbd_unlock_bh_state(bh
);
1186 jh
->b_committed_data
= committed_data
;
1187 committed_data
= NULL
;
1188 memcpy(jh
->b_committed_data
, bh
->b_data
, bh
->b_size
);
1190 jbd_unlock_bh_state(bh
);
1192 jbd2_journal_put_journal_head(jh
);
1193 if (unlikely(committed_data
))
1194 jbd2_free(committed_data
, bh
->b_size
);
1199 * void jbd2_journal_set_triggers() - Add triggers for commit writeout
1200 * @bh: buffer to trigger on
1201 * @type: struct jbd2_buffer_trigger_type containing the trigger(s).
1203 * Set any triggers on this journal_head. This is always safe, because
1204 * triggers for a committing buffer will be saved off, and triggers for
1205 * a running transaction will match the buffer in that transaction.
1207 * Call with NULL to clear the triggers.
1209 void jbd2_journal_set_triggers(struct buffer_head
*bh
,
1210 struct jbd2_buffer_trigger_type
*type
)
1212 struct journal_head
*jh
= jbd2_journal_grab_journal_head(bh
);
1216 jh
->b_triggers
= type
;
1217 jbd2_journal_put_journal_head(jh
);
1220 void jbd2_buffer_frozen_trigger(struct journal_head
*jh
, void *mapped_data
,
1221 struct jbd2_buffer_trigger_type
*triggers
)
1223 struct buffer_head
*bh
= jh2bh(jh
);
1225 if (!triggers
|| !triggers
->t_frozen
)
1228 triggers
->t_frozen(triggers
, bh
, mapped_data
, bh
->b_size
);
1231 void jbd2_buffer_abort_trigger(struct journal_head
*jh
,
1232 struct jbd2_buffer_trigger_type
*triggers
)
1234 if (!triggers
|| !triggers
->t_abort
)
1237 triggers
->t_abort(triggers
, jh2bh(jh
));
1243 * int jbd2_journal_dirty_metadata() - mark a buffer as containing dirty metadata
1244 * @handle: transaction to add buffer to.
1245 * @bh: buffer to mark
1247 * mark dirty metadata which needs to be journaled as part of the current
1250 * The buffer must have previously had jbd2_journal_get_write_access()
1251 * called so that it has a valid journal_head attached to the buffer
1254 * The buffer is placed on the transaction's metadata list and is marked
1255 * as belonging to the transaction.
1257 * Returns error number or 0 on success.
1259 * Special care needs to be taken if the buffer already belongs to the
1260 * current committing transaction (in which case we should have frozen
1261 * data present for that commit). In that case, we don't relink the
1262 * buffer: that only gets done when the old transaction finally
1263 * completes its commit.
1265 int jbd2_journal_dirty_metadata(handle_t
*handle
, struct buffer_head
*bh
)
1267 transaction_t
*transaction
= handle
->h_transaction
;
1268 journal_t
*journal
= transaction
->t_journal
;
1269 struct journal_head
*jh
;
1272 if (is_handle_aborted(handle
))
1274 jh
= jbd2_journal_grab_journal_head(bh
);
1279 jbd_debug(5, "journal_head %p\n", jh
);
1280 JBUFFER_TRACE(jh
, "entry");
1282 jbd_lock_bh_state(bh
);
1284 if (jh
->b_modified
== 0) {
1286 * This buffer's got modified and becoming part
1287 * of the transaction. This needs to be done
1288 * once a transaction -bzzz
1291 J_ASSERT_JH(jh
, handle
->h_buffer_credits
> 0);
1292 handle
->h_buffer_credits
--;
1296 * fastpath, to avoid expensive locking. If this buffer is already
1297 * on the running transaction's metadata list there is nothing to do.
1298 * Nobody can take it off again because there is a handle open.
1299 * I _think_ we're OK here with SMP barriers - a mistaken decision will
1300 * result in this test being false, so we go in and take the locks.
1302 if (jh
->b_transaction
== transaction
&& jh
->b_jlist
== BJ_Metadata
) {
1303 JBUFFER_TRACE(jh
, "fastpath");
1304 if (unlikely(jh
->b_transaction
!=
1305 journal
->j_running_transaction
)) {
1306 printk(KERN_EMERG
"JBD: %s: "
1307 "jh->b_transaction (%llu, %p, %u) != "
1308 "journal->j_running_transaction (%p, %u)",
1310 (unsigned long long) bh
->b_blocknr
,
1312 jh
->b_transaction
? jh
->b_transaction
->t_tid
: 0,
1313 journal
->j_running_transaction
,
1314 journal
->j_running_transaction
?
1315 journal
->j_running_transaction
->t_tid
: 0);
1321 set_buffer_jbddirty(bh
);
1324 * Metadata already on the current transaction list doesn't
1325 * need to be filed. Metadata on another transaction's list must
1326 * be committing, and will be refiled once the commit completes:
1327 * leave it alone for now.
1329 if (jh
->b_transaction
!= transaction
) {
1330 JBUFFER_TRACE(jh
, "already on other transaction");
1331 if (unlikely(jh
->b_transaction
!=
1332 journal
->j_committing_transaction
)) {
1333 printk(KERN_EMERG
"JBD: %s: "
1334 "jh->b_transaction (%llu, %p, %u) != "
1335 "journal->j_committing_transaction (%p, %u)",
1337 (unsigned long long) bh
->b_blocknr
,
1339 jh
->b_transaction
? jh
->b_transaction
->t_tid
: 0,
1340 journal
->j_committing_transaction
,
1341 journal
->j_committing_transaction
?
1342 journal
->j_committing_transaction
->t_tid
: 0);
1345 if (unlikely(jh
->b_next_transaction
!= transaction
)) {
1346 printk(KERN_EMERG
"JBD: %s: "
1347 "jh->b_next_transaction (%llu, %p, %u) != "
1348 "transaction (%p, %u)",
1350 (unsigned long long) bh
->b_blocknr
,
1351 jh
->b_next_transaction
,
1352 jh
->b_next_transaction
?
1353 jh
->b_next_transaction
->t_tid
: 0,
1354 transaction
, transaction
->t_tid
);
1357 /* And this case is illegal: we can't reuse another
1358 * transaction's data buffer, ever. */
1362 /* That test should have eliminated the following case: */
1363 J_ASSERT_JH(jh
, jh
->b_frozen_data
== NULL
);
1365 JBUFFER_TRACE(jh
, "file as BJ_Metadata");
1366 spin_lock(&journal
->j_list_lock
);
1367 __jbd2_journal_file_buffer(jh
, handle
->h_transaction
, BJ_Metadata
);
1368 spin_unlock(&journal
->j_list_lock
);
1370 jbd_unlock_bh_state(bh
);
1371 jbd2_journal_put_journal_head(jh
);
1373 JBUFFER_TRACE(jh
, "exit");
1374 WARN_ON(ret
); /* All errors are bugs, so dump the stack */
1379 * void jbd2_journal_forget() - bforget() for potentially-journaled buffers.
1380 * @handle: transaction handle
1381 * @bh: bh to 'forget'
1383 * We can only do the bforget if there are no commits pending against the
1384 * buffer. If the buffer is dirty in the current running transaction we
1385 * can safely unlink it.
1387 * bh may not be a journalled buffer at all - it may be a non-JBD
1388 * buffer which came off the hashtable. Check for this.
1390 * Decrements bh->b_count by one.
1392 * Allow this call even if the handle has aborted --- it may be part of
1393 * the caller's cleanup after an abort.
1395 int jbd2_journal_forget (handle_t
*handle
, struct buffer_head
*bh
)
1397 transaction_t
*transaction
= handle
->h_transaction
;
1398 journal_t
*journal
= transaction
->t_journal
;
1399 struct journal_head
*jh
;
1400 int drop_reserve
= 0;
1402 int was_modified
= 0;
1404 BUFFER_TRACE(bh
, "entry");
1406 jbd_lock_bh_state(bh
);
1407 spin_lock(&journal
->j_list_lock
);
1409 if (!buffer_jbd(bh
))
1413 /* Critical error: attempting to delete a bitmap buffer, maybe?
1414 * Don't do any jbd operations, and return an error. */
1415 if (!J_EXPECT_JH(jh
, !jh
->b_committed_data
,
1416 "inconsistent data on disk")) {
1421 /* keep track of whether or not this transaction modified us */
1422 was_modified
= jh
->b_modified
;
1425 * The buffer's going from the transaction, we must drop
1426 * all references -bzzz
1430 if (jh
->b_transaction
== handle
->h_transaction
) {
1431 J_ASSERT_JH(jh
, !jh
->b_frozen_data
);
1433 /* If we are forgetting a buffer which is already part
1434 * of this transaction, then we can just drop it from
1435 * the transaction immediately. */
1436 clear_buffer_dirty(bh
);
1437 clear_buffer_jbddirty(bh
);
1439 JBUFFER_TRACE(jh
, "belongs to current transaction: unfile");
1442 * we only want to drop a reference if this transaction
1443 * modified the buffer
1449 * We are no longer going to journal this buffer.
1450 * However, the commit of this transaction is still
1451 * important to the buffer: the delete that we are now
1452 * processing might obsolete an old log entry, so by
1453 * committing, we can satisfy the buffer's checkpoint.
1455 * So, if we have a checkpoint on the buffer, we should
1456 * now refile the buffer on our BJ_Forget list so that
1457 * we know to remove the checkpoint after we commit.
1460 if (jh
->b_cp_transaction
) {
1461 __jbd2_journal_temp_unlink_buffer(jh
);
1462 __jbd2_journal_file_buffer(jh
, transaction
, BJ_Forget
);
1464 __jbd2_journal_unfile_buffer(jh
);
1465 if (!buffer_jbd(bh
)) {
1466 spin_unlock(&journal
->j_list_lock
);
1467 jbd_unlock_bh_state(bh
);
1472 } else if (jh
->b_transaction
) {
1473 J_ASSERT_JH(jh
, (jh
->b_transaction
==
1474 journal
->j_committing_transaction
));
1475 /* However, if the buffer is still owned by a prior
1476 * (committing) transaction, we can't drop it yet... */
1477 JBUFFER_TRACE(jh
, "belongs to older transaction");
1478 /* ... but we CAN drop it from the new transaction if we
1479 * have also modified it since the original commit. */
1481 if (jh
->b_next_transaction
) {
1482 J_ASSERT(jh
->b_next_transaction
== transaction
);
1483 jh
->b_next_transaction
= NULL
;
1486 * only drop a reference if this transaction modified
1495 spin_unlock(&journal
->j_list_lock
);
1496 jbd_unlock_bh_state(bh
);
1500 /* no need to reserve log space for this block -bzzz */
1501 handle
->h_buffer_credits
++;
1507 * int jbd2_journal_stop() - complete a transaction
1508 * @handle: tranaction to complete.
1510 * All done for a particular handle.
1512 * There is not much action needed here. We just return any remaining
1513 * buffer credits to the transaction and remove the handle. The only
1514 * complication is that we need to start a commit operation if the
1515 * filesystem is marked for synchronous update.
1517 * jbd2_journal_stop itself will not usually return an error, but it may
1518 * do so in unusual circumstances. In particular, expect it to
1519 * return -EIO if a jbd2_journal_abort has been executed since the
1520 * transaction began.
1522 int jbd2_journal_stop(handle_t
*handle
)
1524 transaction_t
*transaction
= handle
->h_transaction
;
1525 journal_t
*journal
= transaction
->t_journal
;
1526 int err
, wait_for_commit
= 0;
1530 J_ASSERT(journal_current_handle() == handle
);
1532 if (is_handle_aborted(handle
))
1535 J_ASSERT(atomic_read(&transaction
->t_updates
) > 0);
1539 if (--handle
->h_ref
> 0) {
1540 jbd_debug(4, "h_ref %d -> %d\n", handle
->h_ref
+ 1,
1545 jbd_debug(4, "Handle %p going down\n", handle
);
1546 trace_jbd2_handle_stats(journal
->j_fs_dev
->bd_dev
,
1547 handle
->h_transaction
->t_tid
,
1548 handle
->h_type
, handle
->h_line_no
,
1549 jiffies
- handle
->h_start_jiffies
,
1550 handle
->h_sync
, handle
->h_requested_credits
,
1551 (handle
->h_requested_credits
-
1552 handle
->h_buffer_credits
));
1555 * Implement synchronous transaction batching. If the handle
1556 * was synchronous, don't force a commit immediately. Let's
1557 * yield and let another thread piggyback onto this
1558 * transaction. Keep doing that while new threads continue to
1559 * arrive. It doesn't cost much - we're about to run a commit
1560 * and sleep on IO anyway. Speeds up many-threaded, many-dir
1561 * operations by 30x or more...
1563 * We try and optimize the sleep time against what the
1564 * underlying disk can do, instead of having a static sleep
1565 * time. This is useful for the case where our storage is so
1566 * fast that it is more optimal to go ahead and force a flush
1567 * and wait for the transaction to be committed than it is to
1568 * wait for an arbitrary amount of time for new writers to
1569 * join the transaction. We achieve this by measuring how
1570 * long it takes to commit a transaction, and compare it with
1571 * how long this transaction has been running, and if run time
1572 * < commit time then we sleep for the delta and commit. This
1573 * greatly helps super fast disks that would see slowdowns as
1574 * more threads started doing fsyncs.
1576 * But don't do this if this process was the most recent one
1577 * to perform a synchronous write. We do this to detect the
1578 * case where a single process is doing a stream of sync
1579 * writes. No point in waiting for joiners in that case.
1582 if (handle
->h_sync
&& journal
->j_last_sync_writer
!= pid
) {
1583 u64 commit_time
, trans_time
;
1585 journal
->j_last_sync_writer
= pid
;
1587 read_lock(&journal
->j_state_lock
);
1588 commit_time
= journal
->j_average_commit_time
;
1589 read_unlock(&journal
->j_state_lock
);
1591 trans_time
= ktime_to_ns(ktime_sub(ktime_get(),
1592 transaction
->t_start_time
));
1594 commit_time
= max_t(u64
, commit_time
,
1595 1000*journal
->j_min_batch_time
);
1596 commit_time
= min_t(u64
, commit_time
,
1597 1000*journal
->j_max_batch_time
);
1599 if (trans_time
< commit_time
) {
1600 ktime_t expires
= ktime_add_ns(ktime_get(),
1602 set_current_state(TASK_UNINTERRUPTIBLE
);
1603 schedule_hrtimeout(&expires
, HRTIMER_MODE_ABS
);
1608 transaction
->t_synchronous_commit
= 1;
1609 current
->journal_info
= NULL
;
1610 atomic_sub(handle
->h_buffer_credits
,
1611 &transaction
->t_outstanding_credits
);
1614 * If the handle is marked SYNC, we need to set another commit
1615 * going! We also want to force a commit if the current
1616 * transaction is occupying too much of the log, or if the
1617 * transaction is too old now.
1619 if (handle
->h_sync
||
1620 (atomic_read(&transaction
->t_outstanding_credits
) >
1621 journal
->j_max_transaction_buffers
) ||
1622 time_after_eq(jiffies
, transaction
->t_expires
)) {
1623 /* Do this even for aborted journals: an abort still
1624 * completes the commit thread, it just doesn't write
1625 * anything to disk. */
1627 jbd_debug(2, "transaction too old, requesting commit for "
1628 "handle %p\n", handle
);
1629 /* This is non-blocking */
1630 jbd2_log_start_commit(journal
, transaction
->t_tid
);
1633 * Special case: JBD2_SYNC synchronous updates require us
1634 * to wait for the commit to complete.
1636 if (handle
->h_sync
&& !(current
->flags
& PF_MEMALLOC
))
1637 wait_for_commit
= 1;
1641 * Once we drop t_updates, if it goes to zero the transaction
1642 * could start committing on us and eventually disappear. So
1643 * once we do this, we must not dereference transaction
1646 tid
= transaction
->t_tid
;
1647 if (atomic_dec_and_test(&transaction
->t_updates
)) {
1648 wake_up(&journal
->j_wait_updates
);
1649 if (journal
->j_barrier_count
)
1650 wake_up(&journal
->j_wait_transaction_locked
);
1653 if (wait_for_commit
)
1654 err
= jbd2_log_wait_commit(journal
, tid
);
1656 lock_map_release(&handle
->h_lockdep_map
);
1658 if (handle
->h_rsv_handle
)
1659 jbd2_journal_free_reserved(handle
->h_rsv_handle
);
1660 jbd2_free_handle(handle
);
1666 * List management code snippets: various functions for manipulating the
1667 * transaction buffer lists.
1672 * Append a buffer to a transaction list, given the transaction's list head
1675 * j_list_lock is held.
1677 * jbd_lock_bh_state(jh2bh(jh)) is held.
1681 __blist_add_buffer(struct journal_head
**list
, struct journal_head
*jh
)
1684 jh
->b_tnext
= jh
->b_tprev
= jh
;
1687 /* Insert at the tail of the list to preserve order */
1688 struct journal_head
*first
= *list
, *last
= first
->b_tprev
;
1690 jh
->b_tnext
= first
;
1691 last
->b_tnext
= first
->b_tprev
= jh
;
1696 * Remove a buffer from a transaction list, given the transaction's list
1699 * Called with j_list_lock held, and the journal may not be locked.
1701 * jbd_lock_bh_state(jh2bh(jh)) is held.
1705 __blist_del_buffer(struct journal_head
**list
, struct journal_head
*jh
)
1708 *list
= jh
->b_tnext
;
1712 jh
->b_tprev
->b_tnext
= jh
->b_tnext
;
1713 jh
->b_tnext
->b_tprev
= jh
->b_tprev
;
1717 * Remove a buffer from the appropriate transaction list.
1719 * Note that this function can *change* the value of
1720 * bh->b_transaction->t_buffers, t_forget, t_shadow_list, t_log_list or
1721 * t_reserved_list. If the caller is holding onto a copy of one of these
1722 * pointers, it could go bad. Generally the caller needs to re-read the
1723 * pointer from the transaction_t.
1725 * Called under j_list_lock.
1727 static void __jbd2_journal_temp_unlink_buffer(struct journal_head
*jh
)
1729 struct journal_head
**list
= NULL
;
1730 transaction_t
*transaction
;
1731 struct buffer_head
*bh
= jh2bh(jh
);
1733 J_ASSERT_JH(jh
, jbd_is_locked_bh_state(bh
));
1734 transaction
= jh
->b_transaction
;
1736 assert_spin_locked(&transaction
->t_journal
->j_list_lock
);
1738 J_ASSERT_JH(jh
, jh
->b_jlist
< BJ_Types
);
1739 if (jh
->b_jlist
!= BJ_None
)
1740 J_ASSERT_JH(jh
, transaction
!= NULL
);
1742 switch (jh
->b_jlist
) {
1746 transaction
->t_nr_buffers
--;
1747 J_ASSERT_JH(jh
, transaction
->t_nr_buffers
>= 0);
1748 list
= &transaction
->t_buffers
;
1751 list
= &transaction
->t_forget
;
1754 list
= &transaction
->t_shadow_list
;
1757 list
= &transaction
->t_reserved_list
;
1761 __blist_del_buffer(list
, jh
);
1762 jh
->b_jlist
= BJ_None
;
1763 if (test_clear_buffer_jbddirty(bh
))
1764 mark_buffer_dirty(bh
); /* Expose it to the VM */
1768 * Remove buffer from all transactions.
1770 * Called with bh_state lock and j_list_lock
1772 * jh and bh may be already freed when this function returns.
1774 static void __jbd2_journal_unfile_buffer(struct journal_head
*jh
)
1776 __jbd2_journal_temp_unlink_buffer(jh
);
1777 jh
->b_transaction
= NULL
;
1778 jbd2_journal_put_journal_head(jh
);
1781 void jbd2_journal_unfile_buffer(journal_t
*journal
, struct journal_head
*jh
)
1783 struct buffer_head
*bh
= jh2bh(jh
);
1785 /* Get reference so that buffer cannot be freed before we unlock it */
1787 jbd_lock_bh_state(bh
);
1788 spin_lock(&journal
->j_list_lock
);
1789 __jbd2_journal_unfile_buffer(jh
);
1790 spin_unlock(&journal
->j_list_lock
);
1791 jbd_unlock_bh_state(bh
);
1796 * Called from jbd2_journal_try_to_free_buffers().
1798 * Called under jbd_lock_bh_state(bh)
1801 __journal_try_to_free_buffer(journal_t
*journal
, struct buffer_head
*bh
)
1803 struct journal_head
*jh
;
1807 if (buffer_locked(bh
) || buffer_dirty(bh
))
1810 if (jh
->b_next_transaction
!= NULL
)
1813 spin_lock(&journal
->j_list_lock
);
1814 if (jh
->b_cp_transaction
!= NULL
&& jh
->b_transaction
== NULL
) {
1815 /* written-back checkpointed metadata buffer */
1816 JBUFFER_TRACE(jh
, "remove from checkpoint list");
1817 __jbd2_journal_remove_checkpoint(jh
);
1819 spin_unlock(&journal
->j_list_lock
);
1825 * int jbd2_journal_try_to_free_buffers() - try to free page buffers.
1826 * @journal: journal for operation
1827 * @page: to try and free
1828 * @gfp_mask: we use the mask to detect how hard should we try to release
1829 * buffers. If __GFP_WAIT and __GFP_FS is set, we wait for commit code to
1830 * release the buffers.
1833 * For all the buffers on this page,
1834 * if they are fully written out ordered data, move them onto BUF_CLEAN
1835 * so try_to_free_buffers() can reap them.
1837 * This function returns non-zero if we wish try_to_free_buffers()
1838 * to be called. We do this if the page is releasable by try_to_free_buffers().
1839 * We also do it if the page has locked or dirty buffers and the caller wants
1840 * us to perform sync or async writeout.
1842 * This complicates JBD locking somewhat. We aren't protected by the
1843 * BKL here. We wish to remove the buffer from its committing or
1844 * running transaction's ->t_datalist via __jbd2_journal_unfile_buffer.
1846 * This may *change* the value of transaction_t->t_datalist, so anyone
1847 * who looks at t_datalist needs to lock against this function.
1849 * Even worse, someone may be doing a jbd2_journal_dirty_data on this
1850 * buffer. So we need to lock against that. jbd2_journal_dirty_data()
1851 * will come out of the lock with the buffer dirty, which makes it
1852 * ineligible for release here.
1854 * Who else is affected by this? hmm... Really the only contender
1855 * is do_get_write_access() - it could be looking at the buffer while
1856 * journal_try_to_free_buffer() is changing its state. But that
1857 * cannot happen because we never reallocate freed data as metadata
1858 * while the data is part of a transaction. Yes?
1860 * Return 0 on failure, 1 on success
1862 int jbd2_journal_try_to_free_buffers(journal_t
*journal
,
1863 struct page
*page
, gfp_t gfp_mask
)
1865 struct buffer_head
*head
;
1866 struct buffer_head
*bh
;
1869 J_ASSERT(PageLocked(page
));
1871 head
= page_buffers(page
);
1874 struct journal_head
*jh
;
1877 * We take our own ref against the journal_head here to avoid
1878 * having to add tons of locking around each instance of
1879 * jbd2_journal_put_journal_head().
1881 jh
= jbd2_journal_grab_journal_head(bh
);
1885 jbd_lock_bh_state(bh
);
1886 __journal_try_to_free_buffer(journal
, bh
);
1887 jbd2_journal_put_journal_head(jh
);
1888 jbd_unlock_bh_state(bh
);
1891 } while ((bh
= bh
->b_this_page
) != head
);
1893 ret
= try_to_free_buffers(page
);
1900 * This buffer is no longer needed. If it is on an older transaction's
1901 * checkpoint list we need to record it on this transaction's forget list
1902 * to pin this buffer (and hence its checkpointing transaction) down until
1903 * this transaction commits. If the buffer isn't on a checkpoint list, we
1905 * Returns non-zero if JBD no longer has an interest in the buffer.
1907 * Called under j_list_lock.
1909 * Called under jbd_lock_bh_state(bh).
1911 static int __dispose_buffer(struct journal_head
*jh
, transaction_t
*transaction
)
1914 struct buffer_head
*bh
= jh2bh(jh
);
1916 if (jh
->b_cp_transaction
) {
1917 JBUFFER_TRACE(jh
, "on running+cp transaction");
1918 __jbd2_journal_temp_unlink_buffer(jh
);
1920 * We don't want to write the buffer anymore, clear the
1921 * bit so that we don't confuse checks in
1922 * __journal_file_buffer
1924 clear_buffer_dirty(bh
);
1925 __jbd2_journal_file_buffer(jh
, transaction
, BJ_Forget
);
1928 JBUFFER_TRACE(jh
, "on running transaction");
1929 __jbd2_journal_unfile_buffer(jh
);
1935 * jbd2_journal_invalidatepage
1937 * This code is tricky. It has a number of cases to deal with.
1939 * There are two invariants which this code relies on:
1941 * i_size must be updated on disk before we start calling invalidatepage on the
1944 * This is done in ext3 by defining an ext3_setattr method which
1945 * updates i_size before truncate gets going. By maintaining this
1946 * invariant, we can be sure that it is safe to throw away any buffers
1947 * attached to the current transaction: once the transaction commits,
1948 * we know that the data will not be needed.
1950 * Note however that we can *not* throw away data belonging to the
1951 * previous, committing transaction!
1953 * Any disk blocks which *are* part of the previous, committing
1954 * transaction (and which therefore cannot be discarded immediately) are
1955 * not going to be reused in the new running transaction
1957 * The bitmap committed_data images guarantee this: any block which is
1958 * allocated in one transaction and removed in the next will be marked
1959 * as in-use in the committed_data bitmap, so cannot be reused until
1960 * the next transaction to delete the block commits. This means that
1961 * leaving committing buffers dirty is quite safe: the disk blocks
1962 * cannot be reallocated to a different file and so buffer aliasing is
1966 * The above applies mainly to ordered data mode. In writeback mode we
1967 * don't make guarantees about the order in which data hits disk --- in
1968 * particular we don't guarantee that new dirty data is flushed before
1969 * transaction commit --- so it is always safe just to discard data
1970 * immediately in that mode. --sct
1974 * The journal_unmap_buffer helper function returns zero if the buffer
1975 * concerned remains pinned as an anonymous buffer belonging to an older
1978 * We're outside-transaction here. Either or both of j_running_transaction
1979 * and j_committing_transaction may be NULL.
1981 static int journal_unmap_buffer(journal_t
*journal
, struct buffer_head
*bh
,
1984 transaction_t
*transaction
;
1985 struct journal_head
*jh
;
1988 BUFFER_TRACE(bh
, "entry");
1991 * It is safe to proceed here without the j_list_lock because the
1992 * buffers cannot be stolen by try_to_free_buffers as long as we are
1993 * holding the page lock. --sct
1996 if (!buffer_jbd(bh
))
1997 goto zap_buffer_unlocked
;
1999 /* OK, we have data buffer in journaled mode */
2000 write_lock(&journal
->j_state_lock
);
2001 jbd_lock_bh_state(bh
);
2002 spin_lock(&journal
->j_list_lock
);
2004 jh
= jbd2_journal_grab_journal_head(bh
);
2006 goto zap_buffer_no_jh
;
2009 * We cannot remove the buffer from checkpoint lists until the
2010 * transaction adding inode to orphan list (let's call it T)
2011 * is committed. Otherwise if the transaction changing the
2012 * buffer would be cleaned from the journal before T is
2013 * committed, a crash will cause that the correct contents of
2014 * the buffer will be lost. On the other hand we have to
2015 * clear the buffer dirty bit at latest at the moment when the
2016 * transaction marking the buffer as freed in the filesystem
2017 * structures is committed because from that moment on the
2018 * block can be reallocated and used by a different page.
2019 * Since the block hasn't been freed yet but the inode has
2020 * already been added to orphan list, it is safe for us to add
2021 * the buffer to BJ_Forget list of the newest transaction.
2023 * Also we have to clear buffer_mapped flag of a truncated buffer
2024 * because the buffer_head may be attached to the page straddling
2025 * i_size (can happen only when blocksize < pagesize) and thus the
2026 * buffer_head can be reused when the file is extended again. So we end
2027 * up keeping around invalidated buffers attached to transactions'
2028 * BJ_Forget list just to stop checkpointing code from cleaning up
2029 * the transaction this buffer was modified in.
2031 transaction
= jh
->b_transaction
;
2032 if (transaction
== NULL
) {
2033 /* First case: not on any transaction. If it
2034 * has no checkpoint link, then we can zap it:
2035 * it's a writeback-mode buffer so we don't care
2036 * if it hits disk safely. */
2037 if (!jh
->b_cp_transaction
) {
2038 JBUFFER_TRACE(jh
, "not on any transaction: zap");
2042 if (!buffer_dirty(bh
)) {
2043 /* bdflush has written it. We can drop it now */
2047 /* OK, it must be in the journal but still not
2048 * written fully to disk: it's metadata or
2049 * journaled data... */
2051 if (journal
->j_running_transaction
) {
2052 /* ... and once the current transaction has
2053 * committed, the buffer won't be needed any
2055 JBUFFER_TRACE(jh
, "checkpointed: add to BJ_Forget");
2056 may_free
= __dispose_buffer(jh
,
2057 journal
->j_running_transaction
);
2060 /* There is no currently-running transaction. So the
2061 * orphan record which we wrote for this file must have
2062 * passed into commit. We must attach this buffer to
2063 * the committing transaction, if it exists. */
2064 if (journal
->j_committing_transaction
) {
2065 JBUFFER_TRACE(jh
, "give to committing trans");
2066 may_free
= __dispose_buffer(jh
,
2067 journal
->j_committing_transaction
);
2070 /* The orphan record's transaction has
2071 * committed. We can cleanse this buffer */
2072 clear_buffer_jbddirty(bh
);
2076 } else if (transaction
== journal
->j_committing_transaction
) {
2077 JBUFFER_TRACE(jh
, "on committing transaction");
2079 * The buffer is committing, we simply cannot touch
2080 * it. If the page is straddling i_size we have to wait
2081 * for commit and try again.
2084 jbd2_journal_put_journal_head(jh
);
2085 spin_unlock(&journal
->j_list_lock
);
2086 jbd_unlock_bh_state(bh
);
2087 write_unlock(&journal
->j_state_lock
);
2091 * OK, buffer won't be reachable after truncate. We just set
2092 * j_next_transaction to the running transaction (if there is
2093 * one) and mark buffer as freed so that commit code knows it
2094 * should clear dirty bits when it is done with the buffer.
2096 set_buffer_freed(bh
);
2097 if (journal
->j_running_transaction
&& buffer_jbddirty(bh
))
2098 jh
->b_next_transaction
= journal
->j_running_transaction
;
2099 jbd2_journal_put_journal_head(jh
);
2100 spin_unlock(&journal
->j_list_lock
);
2101 jbd_unlock_bh_state(bh
);
2102 write_unlock(&journal
->j_state_lock
);
2105 /* Good, the buffer belongs to the running transaction.
2106 * We are writing our own transaction's data, not any
2107 * previous one's, so it is safe to throw it away
2108 * (remember that we expect the filesystem to have set
2109 * i_size already for this truncate so recovery will not
2110 * expose the disk blocks we are discarding here.) */
2111 J_ASSERT_JH(jh
, transaction
== journal
->j_running_transaction
);
2112 JBUFFER_TRACE(jh
, "on running transaction");
2113 may_free
= __dispose_buffer(jh
, transaction
);
2118 * This is tricky. Although the buffer is truncated, it may be reused
2119 * if blocksize < pagesize and it is attached to the page straddling
2120 * EOF. Since the buffer might have been added to BJ_Forget list of the
2121 * running transaction, journal_get_write_access() won't clear
2122 * b_modified and credit accounting gets confused. So clear b_modified
2126 jbd2_journal_put_journal_head(jh
);
2128 spin_unlock(&journal
->j_list_lock
);
2129 jbd_unlock_bh_state(bh
);
2130 write_unlock(&journal
->j_state_lock
);
2131 zap_buffer_unlocked
:
2132 clear_buffer_dirty(bh
);
2133 J_ASSERT_BH(bh
, !buffer_jbddirty(bh
));
2134 clear_buffer_mapped(bh
);
2135 clear_buffer_req(bh
);
2136 clear_buffer_new(bh
);
2137 clear_buffer_delay(bh
);
2138 clear_buffer_unwritten(bh
);
2144 * void jbd2_journal_invalidatepage()
2145 * @journal: journal to use for flush...
2146 * @page: page to flush
2147 * @offset: start of the range to invalidate
2148 * @length: length of the range to invalidate
2150 * Reap page buffers containing data after in the specified range in page.
2151 * Can return -EBUSY if buffers are part of the committing transaction and
2152 * the page is straddling i_size. Caller then has to wait for current commit
2155 int jbd2_journal_invalidatepage(journal_t
*journal
,
2157 unsigned int offset
,
2158 unsigned int length
)
2160 struct buffer_head
*head
, *bh
, *next
;
2161 unsigned int stop
= offset
+ length
;
2162 unsigned int curr_off
= 0;
2163 int partial_page
= (offset
|| length
< PAGE_CACHE_SIZE
);
2167 if (!PageLocked(page
))
2169 if (!page_has_buffers(page
))
2172 BUG_ON(stop
> PAGE_CACHE_SIZE
|| stop
< length
);
2174 /* We will potentially be playing with lists other than just the
2175 * data lists (especially for journaled data mode), so be
2176 * cautious in our locking. */
2178 head
= bh
= page_buffers(page
);
2180 unsigned int next_off
= curr_off
+ bh
->b_size
;
2181 next
= bh
->b_this_page
;
2183 if (next_off
> stop
)
2186 if (offset
<= curr_off
) {
2187 /* This block is wholly outside the truncation point */
2189 ret
= journal_unmap_buffer(journal
, bh
, partial_page
);
2195 curr_off
= next_off
;
2198 } while (bh
!= head
);
2200 if (!partial_page
) {
2201 if (may_free
&& try_to_free_buffers(page
))
2202 J_ASSERT(!page_has_buffers(page
));
2208 * File a buffer on the given transaction list.
2210 void __jbd2_journal_file_buffer(struct journal_head
*jh
,
2211 transaction_t
*transaction
, int jlist
)
2213 struct journal_head
**list
= NULL
;
2215 struct buffer_head
*bh
= jh2bh(jh
);
2217 J_ASSERT_JH(jh
, jbd_is_locked_bh_state(bh
));
2218 assert_spin_locked(&transaction
->t_journal
->j_list_lock
);
2220 J_ASSERT_JH(jh
, jh
->b_jlist
< BJ_Types
);
2221 J_ASSERT_JH(jh
, jh
->b_transaction
== transaction
||
2222 jh
->b_transaction
== NULL
);
2224 if (jh
->b_transaction
&& jh
->b_jlist
== jlist
)
2227 if (jlist
== BJ_Metadata
|| jlist
== BJ_Reserved
||
2228 jlist
== BJ_Shadow
|| jlist
== BJ_Forget
) {
2230 * For metadata buffers, we track dirty bit in buffer_jbddirty
2231 * instead of buffer_dirty. We should not see a dirty bit set
2232 * here because we clear it in do_get_write_access but e.g.
2233 * tune2fs can modify the sb and set the dirty bit at any time
2234 * so we try to gracefully handle that.
2236 if (buffer_dirty(bh
))
2237 warn_dirty_buffer(bh
);
2238 if (test_clear_buffer_dirty(bh
) ||
2239 test_clear_buffer_jbddirty(bh
))
2243 if (jh
->b_transaction
)
2244 __jbd2_journal_temp_unlink_buffer(jh
);
2246 jbd2_journal_grab_journal_head(bh
);
2247 jh
->b_transaction
= transaction
;
2251 J_ASSERT_JH(jh
, !jh
->b_committed_data
);
2252 J_ASSERT_JH(jh
, !jh
->b_frozen_data
);
2255 transaction
->t_nr_buffers
++;
2256 list
= &transaction
->t_buffers
;
2259 list
= &transaction
->t_forget
;
2262 list
= &transaction
->t_shadow_list
;
2265 list
= &transaction
->t_reserved_list
;
2269 __blist_add_buffer(list
, jh
);
2270 jh
->b_jlist
= jlist
;
2273 set_buffer_jbddirty(bh
);
2276 void jbd2_journal_file_buffer(struct journal_head
*jh
,
2277 transaction_t
*transaction
, int jlist
)
2279 jbd_lock_bh_state(jh2bh(jh
));
2280 spin_lock(&transaction
->t_journal
->j_list_lock
);
2281 __jbd2_journal_file_buffer(jh
, transaction
, jlist
);
2282 spin_unlock(&transaction
->t_journal
->j_list_lock
);
2283 jbd_unlock_bh_state(jh2bh(jh
));
2287 * Remove a buffer from its current buffer list in preparation for
2288 * dropping it from its current transaction entirely. If the buffer has
2289 * already started to be used by a subsequent transaction, refile the
2290 * buffer on that transaction's metadata list.
2292 * Called under j_list_lock
2293 * Called under jbd_lock_bh_state(jh2bh(jh))
2295 * jh and bh may be already free when this function returns
2297 void __jbd2_journal_refile_buffer(struct journal_head
*jh
)
2299 int was_dirty
, jlist
;
2300 struct buffer_head
*bh
= jh2bh(jh
);
2302 J_ASSERT_JH(jh
, jbd_is_locked_bh_state(bh
));
2303 if (jh
->b_transaction
)
2304 assert_spin_locked(&jh
->b_transaction
->t_journal
->j_list_lock
);
2306 /* If the buffer is now unused, just drop it. */
2307 if (jh
->b_next_transaction
== NULL
) {
2308 __jbd2_journal_unfile_buffer(jh
);
2313 * It has been modified by a later transaction: add it to the new
2314 * transaction's metadata list.
2317 was_dirty
= test_clear_buffer_jbddirty(bh
);
2318 __jbd2_journal_temp_unlink_buffer(jh
);
2320 * We set b_transaction here because b_next_transaction will inherit
2321 * our jh reference and thus __jbd2_journal_file_buffer() must not
2324 jh
->b_transaction
= jh
->b_next_transaction
;
2325 jh
->b_next_transaction
= NULL
;
2326 if (buffer_freed(bh
))
2328 else if (jh
->b_modified
)
2329 jlist
= BJ_Metadata
;
2331 jlist
= BJ_Reserved
;
2332 __jbd2_journal_file_buffer(jh
, jh
->b_transaction
, jlist
);
2333 J_ASSERT_JH(jh
, jh
->b_transaction
->t_state
== T_RUNNING
);
2336 set_buffer_jbddirty(bh
);
2340 * __jbd2_journal_refile_buffer() with necessary locking added. We take our
2341 * bh reference so that we can safely unlock bh.
2343 * The jh and bh may be freed by this call.
2345 void jbd2_journal_refile_buffer(journal_t
*journal
, struct journal_head
*jh
)
2347 struct buffer_head
*bh
= jh2bh(jh
);
2349 /* Get reference so that buffer cannot be freed before we unlock it */
2351 jbd_lock_bh_state(bh
);
2352 spin_lock(&journal
->j_list_lock
);
2353 __jbd2_journal_refile_buffer(jh
);
2354 jbd_unlock_bh_state(bh
);
2355 spin_unlock(&journal
->j_list_lock
);
2360 * File inode in the inode list of the handle's transaction
2362 int jbd2_journal_file_inode(handle_t
*handle
, struct jbd2_inode
*jinode
)
2364 transaction_t
*transaction
= handle
->h_transaction
;
2365 journal_t
*journal
= transaction
->t_journal
;
2367 if (is_handle_aborted(handle
))
2370 jbd_debug(4, "Adding inode %lu, tid:%d\n", jinode
->i_vfs_inode
->i_ino
,
2371 transaction
->t_tid
);
2374 * First check whether inode isn't already on the transaction's
2375 * lists without taking the lock. Note that this check is safe
2376 * without the lock as we cannot race with somebody removing inode
2377 * from the transaction. The reason is that we remove inode from the
2378 * transaction only in journal_release_jbd_inode() and when we commit
2379 * the transaction. We are guarded from the first case by holding
2380 * a reference to the inode. We are safe against the second case
2381 * because if jinode->i_transaction == transaction, commit code
2382 * cannot touch the transaction because we hold reference to it,
2383 * and if jinode->i_next_transaction == transaction, commit code
2384 * will only file the inode where we want it.
2386 if (jinode
->i_transaction
== transaction
||
2387 jinode
->i_next_transaction
== transaction
)
2390 spin_lock(&journal
->j_list_lock
);
2392 if (jinode
->i_transaction
== transaction
||
2393 jinode
->i_next_transaction
== transaction
)
2397 * We only ever set this variable to 1 so the test is safe. Since
2398 * t_need_data_flush is likely to be set, we do the test to save some
2399 * cacheline bouncing
2401 if (!transaction
->t_need_data_flush
)
2402 transaction
->t_need_data_flush
= 1;
2403 /* On some different transaction's list - should be
2404 * the committing one */
2405 if (jinode
->i_transaction
) {
2406 J_ASSERT(jinode
->i_next_transaction
== NULL
);
2407 J_ASSERT(jinode
->i_transaction
==
2408 journal
->j_committing_transaction
);
2409 jinode
->i_next_transaction
= transaction
;
2412 /* Not on any transaction list... */
2413 J_ASSERT(!jinode
->i_next_transaction
);
2414 jinode
->i_transaction
= transaction
;
2415 list_add(&jinode
->i_list
, &transaction
->t_inode_list
);
2417 spin_unlock(&journal
->j_list_lock
);
2423 * File truncate and transaction commit interact with each other in a
2424 * non-trivial way. If a transaction writing data block A is
2425 * committing, we cannot discard the data by truncate until we have
2426 * written them. Otherwise if we crashed after the transaction with
2427 * write has committed but before the transaction with truncate has
2428 * committed, we could see stale data in block A. This function is a
2429 * helper to solve this problem. It starts writeout of the truncated
2430 * part in case it is in the committing transaction.
2432 * Filesystem code must call this function when inode is journaled in
2433 * ordered mode before truncation happens and after the inode has been
2434 * placed on orphan list with the new inode size. The second condition
2435 * avoids the race that someone writes new data and we start
2436 * committing the transaction after this function has been called but
2437 * before a transaction for truncate is started (and furthermore it
2438 * allows us to optimize the case where the addition to orphan list
2439 * happens in the same transaction as write --- we don't have to write
2440 * any data in such case).
2442 int jbd2_journal_begin_ordered_truncate(journal_t
*journal
,
2443 struct jbd2_inode
*jinode
,
2446 transaction_t
*inode_trans
, *commit_trans
;
2449 /* This is a quick check to avoid locking if not necessary */
2450 if (!jinode
->i_transaction
)
2452 /* Locks are here just to force reading of recent values, it is
2453 * enough that the transaction was not committing before we started
2454 * a transaction adding the inode to orphan list */
2455 read_lock(&journal
->j_state_lock
);
2456 commit_trans
= journal
->j_committing_transaction
;
2457 read_unlock(&journal
->j_state_lock
);
2458 spin_lock(&journal
->j_list_lock
);
2459 inode_trans
= jinode
->i_transaction
;
2460 spin_unlock(&journal
->j_list_lock
);
2461 if (inode_trans
== commit_trans
) {
2462 ret
= filemap_fdatawrite_range(jinode
->i_vfs_inode
->i_mapping
,
2463 new_size
, LLONG_MAX
);
2465 jbd2_journal_abort(journal
, ret
);