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
);
209 * Is the number of reserved credits in the current transaction too
210 * big to fit this handle? Wait until reserved credits are freed.
212 if (atomic_read(&journal
->j_reserved_credits
) + total
>
213 journal
->j_max_transaction_buffers
) {
214 read_unlock(&journal
->j_state_lock
);
215 wait_event(journal
->j_wait_reserved
,
216 atomic_read(&journal
->j_reserved_credits
) + total
<=
217 journal
->j_max_transaction_buffers
);
221 wait_transaction_locked(journal
);
226 * The commit code assumes that it can get enough log space
227 * without forcing a checkpoint. This is *critical* for
228 * correctness: a checkpoint of a buffer which is also
229 * associated with a committing transaction creates a deadlock,
230 * so commit simply cannot force through checkpoints.
232 * We must therefore ensure the necessary space in the journal
233 * *before* starting to dirty potentially checkpointed buffers
234 * in the new transaction.
236 if (jbd2_log_space_left(journal
) < jbd2_space_needed(journal
)) {
237 atomic_sub(total
, &t
->t_outstanding_credits
);
238 read_unlock(&journal
->j_state_lock
);
239 write_lock(&journal
->j_state_lock
);
240 if (jbd2_log_space_left(journal
) < jbd2_space_needed(journal
))
241 __jbd2_log_wait_for_space(journal
);
242 write_unlock(&journal
->j_state_lock
);
246 /* No reservation? We are done... */
250 needed
= atomic_add_return(rsv_blocks
, &journal
->j_reserved_credits
);
251 /* We allow at most half of a transaction to be reserved */
252 if (needed
> journal
->j_max_transaction_buffers
/ 2) {
253 sub_reserved_credits(journal
, rsv_blocks
);
254 atomic_sub(total
, &t
->t_outstanding_credits
);
255 read_unlock(&journal
->j_state_lock
);
256 wait_event(journal
->j_wait_reserved
,
257 atomic_read(&journal
->j_reserved_credits
) + rsv_blocks
258 <= journal
->j_max_transaction_buffers
/ 2);
265 * start_this_handle: Given a handle, deal with any locking or stalling
266 * needed to make sure that there is enough journal space for the handle
267 * to begin. Attach the handle to a transaction and set up the
268 * transaction's buffer credits.
271 static int start_this_handle(journal_t
*journal
, handle_t
*handle
,
274 transaction_t
*transaction
, *new_transaction
= NULL
;
275 int blocks
= handle
->h_buffer_credits
;
277 unsigned long ts
= jiffies
;
279 if (handle
->h_rsv_handle
)
280 rsv_blocks
= handle
->h_rsv_handle
->h_buffer_credits
;
283 * Limit the number of reserved credits to 1/2 of maximum transaction
284 * size and limit the number of total credits to not exceed maximum
285 * transaction size per operation.
287 if ((rsv_blocks
> journal
->j_max_transaction_buffers
/ 2) ||
288 (rsv_blocks
+ blocks
> journal
->j_max_transaction_buffers
)) {
289 printk(KERN_ERR
"JBD2: %s wants too many credits "
290 "credits:%d rsv_credits:%d max:%d\n",
291 current
->comm
, blocks
, rsv_blocks
,
292 journal
->j_max_transaction_buffers
);
298 if (!journal
->j_running_transaction
) {
300 * If __GFP_FS is not present, then we may be being called from
301 * inside the fs writeback layer, so we MUST NOT fail.
303 if ((gfp_mask
& __GFP_FS
) == 0)
304 gfp_mask
|= __GFP_NOFAIL
;
305 new_transaction
= kmem_cache_zalloc(transaction_cache
,
307 if (!new_transaction
)
311 jbd_debug(3, "New handle %p going live.\n", handle
);
314 * We need to hold j_state_lock until t_updates has been incremented,
315 * for proper journal barrier handling
318 read_lock(&journal
->j_state_lock
);
319 BUG_ON(journal
->j_flags
& JBD2_UNMOUNT
);
320 if (is_journal_aborted(journal
) ||
321 (journal
->j_errno
!= 0 && !(journal
->j_flags
& JBD2_ACK_ERR
))) {
322 read_unlock(&journal
->j_state_lock
);
323 jbd2_journal_free_transaction(new_transaction
);
328 * Wait on the journal's transaction barrier if necessary. Specifically
329 * we allow reserved handles to proceed because otherwise commit could
330 * deadlock on page writeback not being able to complete.
332 if (!handle
->h_reserved
&& journal
->j_barrier_count
) {
333 read_unlock(&journal
->j_state_lock
);
334 wait_event(journal
->j_wait_transaction_locked
,
335 journal
->j_barrier_count
== 0);
339 if (!journal
->j_running_transaction
) {
340 read_unlock(&journal
->j_state_lock
);
341 if (!new_transaction
)
342 goto alloc_transaction
;
343 write_lock(&journal
->j_state_lock
);
344 if (!journal
->j_running_transaction
&&
345 (handle
->h_reserved
|| !journal
->j_barrier_count
)) {
346 jbd2_get_transaction(journal
, new_transaction
);
347 new_transaction
= NULL
;
349 write_unlock(&journal
->j_state_lock
);
353 transaction
= journal
->j_running_transaction
;
355 if (!handle
->h_reserved
) {
356 /* We may have dropped j_state_lock - restart in that case */
357 if (add_transaction_credits(journal
, blocks
, rsv_blocks
))
361 * We have handle reserved so we are allowed to join T_LOCKED
362 * transaction and we don't have to check for transaction size
365 sub_reserved_credits(journal
, blocks
);
366 handle
->h_reserved
= 0;
369 /* OK, account for the buffers that this operation expects to
370 * use and add the handle to the running transaction.
372 update_t_max_wait(transaction
, ts
);
373 handle
->h_transaction
= transaction
;
374 handle
->h_requested_credits
= blocks
;
375 handle
->h_start_jiffies
= jiffies
;
376 atomic_inc(&transaction
->t_updates
);
377 atomic_inc(&transaction
->t_handle_count
);
378 jbd_debug(4, "Handle %p given %d credits (total %d, free %lu)\n",
380 atomic_read(&transaction
->t_outstanding_credits
),
381 jbd2_log_space_left(journal
));
382 read_unlock(&journal
->j_state_lock
);
383 current
->journal_info
= handle
;
385 lock_map_acquire(&handle
->h_lockdep_map
);
386 jbd2_journal_free_transaction(new_transaction
);
390 static struct lock_class_key jbd2_handle_key
;
392 /* Allocate a new handle. This should probably be in a slab... */
393 static handle_t
*new_handle(int nblocks
)
395 handle_t
*handle
= jbd2_alloc_handle(GFP_NOFS
);
398 handle
->h_buffer_credits
= nblocks
;
401 lockdep_init_map(&handle
->h_lockdep_map
, "jbd2_handle",
402 &jbd2_handle_key
, 0);
408 * handle_t *jbd2_journal_start() - Obtain a new handle.
409 * @journal: Journal to start transaction on.
410 * @nblocks: number of block buffer we might modify
412 * We make sure that the transaction can guarantee at least nblocks of
413 * modified buffers in the log. We block until the log can guarantee
414 * that much space. Additionally, if rsv_blocks > 0, we also create another
415 * handle with rsv_blocks reserved blocks in the journal. This handle is
416 * is stored in h_rsv_handle. It is not attached to any particular transaction
417 * and thus doesn't block transaction commit. If the caller uses this reserved
418 * handle, it has to set h_rsv_handle to NULL as otherwise jbd2_journal_stop()
419 * on the parent handle will dispose the reserved one. Reserved handle has to
420 * be converted to a normal handle using jbd2_journal_start_reserved() before
423 * Return a pointer to a newly allocated handle, or an ERR_PTR() value
426 handle_t
*jbd2__journal_start(journal_t
*journal
, int nblocks
, int rsv_blocks
,
427 gfp_t gfp_mask
, unsigned int type
,
428 unsigned int line_no
)
430 handle_t
*handle
= journal_current_handle();
434 return ERR_PTR(-EROFS
);
437 J_ASSERT(handle
->h_transaction
->t_journal
== journal
);
442 handle
= new_handle(nblocks
);
444 return ERR_PTR(-ENOMEM
);
446 handle_t
*rsv_handle
;
448 rsv_handle
= new_handle(rsv_blocks
);
450 jbd2_free_handle(handle
);
451 return ERR_PTR(-ENOMEM
);
453 rsv_handle
->h_reserved
= 1;
454 rsv_handle
->h_journal
= journal
;
455 handle
->h_rsv_handle
= rsv_handle
;
458 err
= start_this_handle(journal
, handle
, gfp_mask
);
460 if (handle
->h_rsv_handle
)
461 jbd2_free_handle(handle
->h_rsv_handle
);
462 jbd2_free_handle(handle
);
465 handle
->h_type
= type
;
466 handle
->h_line_no
= line_no
;
467 trace_jbd2_handle_start(journal
->j_fs_dev
->bd_dev
,
468 handle
->h_transaction
->t_tid
, type
,
472 EXPORT_SYMBOL(jbd2__journal_start
);
475 handle_t
*jbd2_journal_start(journal_t
*journal
, int nblocks
)
477 return jbd2__journal_start(journal
, nblocks
, 0, GFP_NOFS
, 0, 0);
479 EXPORT_SYMBOL(jbd2_journal_start
);
481 void jbd2_journal_free_reserved(handle_t
*handle
)
483 journal_t
*journal
= handle
->h_journal
;
485 WARN_ON(!handle
->h_reserved
);
486 sub_reserved_credits(journal
, handle
->h_buffer_credits
);
487 jbd2_free_handle(handle
);
489 EXPORT_SYMBOL(jbd2_journal_free_reserved
);
492 * int jbd2_journal_start_reserved(handle_t *handle) - start reserved handle
493 * @handle: handle to start
495 * Start handle that has been previously reserved with jbd2_journal_reserve().
496 * This attaches @handle to the running transaction (or creates one if there's
497 * not transaction running). Unlike jbd2_journal_start() this function cannot
498 * block on journal commit, checkpointing, or similar stuff. It can block on
499 * memory allocation or frozen journal though.
501 * Return 0 on success, non-zero on error - handle is freed in that case.
503 int jbd2_journal_start_reserved(handle_t
*handle
, unsigned int type
,
504 unsigned int line_no
)
506 journal_t
*journal
= handle
->h_journal
;
509 if (WARN_ON(!handle
->h_reserved
)) {
510 /* Someone passed in normal handle? Just stop it. */
511 jbd2_journal_stop(handle
);
515 * Usefulness of mixing of reserved and unreserved handles is
516 * questionable. So far nobody seems to need it so just error out.
518 if (WARN_ON(current
->journal_info
)) {
519 jbd2_journal_free_reserved(handle
);
523 handle
->h_journal
= NULL
;
525 * GFP_NOFS is here because callers are likely from writeback or
526 * similarly constrained call sites
528 ret
= start_this_handle(journal
, handle
, GFP_NOFS
);
530 jbd2_journal_free_reserved(handle
);
533 handle
->h_type
= type
;
534 handle
->h_line_no
= line_no
;
537 EXPORT_SYMBOL(jbd2_journal_start_reserved
);
540 * int jbd2_journal_extend() - extend buffer credits.
541 * @handle: handle to 'extend'
542 * @nblocks: nr blocks to try to extend by.
544 * Some transactions, such as large extends and truncates, can be done
545 * atomically all at once or in several stages. The operation requests
546 * a credit for a number of buffer modications in advance, but can
547 * extend its credit if it needs more.
549 * jbd2_journal_extend tries to give the running handle more buffer credits.
550 * It does not guarantee that allocation - this is a best-effort only.
551 * The calling process MUST be able to deal cleanly with a failure to
554 * Return 0 on success, non-zero on failure.
556 * return code < 0 implies an error
557 * return code > 0 implies normal transaction-full status.
559 int jbd2_journal_extend(handle_t
*handle
, int nblocks
)
561 transaction_t
*transaction
= handle
->h_transaction
;
566 if (is_handle_aborted(handle
))
568 journal
= transaction
->t_journal
;
572 read_lock(&journal
->j_state_lock
);
574 /* Don't extend a locked-down transaction! */
575 if (transaction
->t_state
!= T_RUNNING
) {
576 jbd_debug(3, "denied handle %p %d blocks: "
577 "transaction not running\n", handle
, nblocks
);
581 spin_lock(&transaction
->t_handle_lock
);
582 wanted
= atomic_add_return(nblocks
,
583 &transaction
->t_outstanding_credits
);
585 if (wanted
> journal
->j_max_transaction_buffers
) {
586 jbd_debug(3, "denied handle %p %d blocks: "
587 "transaction too large\n", handle
, nblocks
);
588 atomic_sub(nblocks
, &transaction
->t_outstanding_credits
);
592 if (wanted
+ (wanted
>> JBD2_CONTROL_BLOCKS_SHIFT
) >
593 jbd2_log_space_left(journal
)) {
594 jbd_debug(3, "denied handle %p %d blocks: "
595 "insufficient log space\n", handle
, nblocks
);
596 atomic_sub(nblocks
, &transaction
->t_outstanding_credits
);
600 trace_jbd2_handle_extend(journal
->j_fs_dev
->bd_dev
,
602 handle
->h_type
, handle
->h_line_no
,
603 handle
->h_buffer_credits
,
606 handle
->h_buffer_credits
+= nblocks
;
607 handle
->h_requested_credits
+= nblocks
;
610 jbd_debug(3, "extended handle %p by %d\n", handle
, nblocks
);
612 spin_unlock(&transaction
->t_handle_lock
);
614 read_unlock(&journal
->j_state_lock
);
620 * int jbd2_journal_restart() - restart a handle .
621 * @handle: handle to restart
622 * @nblocks: nr credits requested
624 * Restart a handle for a multi-transaction filesystem
627 * If the jbd2_journal_extend() call above fails to grant new buffer credits
628 * to a running handle, a call to jbd2_journal_restart will commit the
629 * handle's transaction so far and reattach the handle to a new
630 * transaction capabable of guaranteeing the requested number of
631 * credits. We preserve reserved handle if there's any attached to the
634 int jbd2__journal_restart(handle_t
*handle
, int nblocks
, gfp_t gfp_mask
)
636 transaction_t
*transaction
= handle
->h_transaction
;
639 int need_to_start
, ret
;
641 /* If we've had an abort of any type, don't even think about
642 * actually doing the restart! */
643 if (is_handle_aborted(handle
))
645 journal
= transaction
->t_journal
;
648 * First unlink the handle from its current transaction, and start the
651 J_ASSERT(atomic_read(&transaction
->t_updates
) > 0);
652 J_ASSERT(journal_current_handle() == handle
);
654 read_lock(&journal
->j_state_lock
);
655 spin_lock(&transaction
->t_handle_lock
);
656 atomic_sub(handle
->h_buffer_credits
,
657 &transaction
->t_outstanding_credits
);
658 if (handle
->h_rsv_handle
) {
659 sub_reserved_credits(journal
,
660 handle
->h_rsv_handle
->h_buffer_credits
);
662 if (atomic_dec_and_test(&transaction
->t_updates
))
663 wake_up(&journal
->j_wait_updates
);
664 tid
= transaction
->t_tid
;
665 spin_unlock(&transaction
->t_handle_lock
);
666 handle
->h_transaction
= NULL
;
667 current
->journal_info
= NULL
;
669 jbd_debug(2, "restarting handle %p\n", handle
);
670 need_to_start
= !tid_geq(journal
->j_commit_request
, tid
);
671 read_unlock(&journal
->j_state_lock
);
673 jbd2_log_start_commit(journal
, tid
);
675 lock_map_release(&handle
->h_lockdep_map
);
676 handle
->h_buffer_credits
= nblocks
;
677 ret
= start_this_handle(journal
, handle
, gfp_mask
);
680 EXPORT_SYMBOL(jbd2__journal_restart
);
683 int jbd2_journal_restart(handle_t
*handle
, int nblocks
)
685 return jbd2__journal_restart(handle
, nblocks
, GFP_NOFS
);
687 EXPORT_SYMBOL(jbd2_journal_restart
);
690 * void jbd2_journal_lock_updates () - establish a transaction barrier.
691 * @journal: Journal to establish a barrier on.
693 * This locks out any further updates from being started, and blocks
694 * until all existing updates have completed, returning only once the
695 * journal is in a quiescent state with no updates running.
697 * The journal lock should not be held on entry.
699 void jbd2_journal_lock_updates(journal_t
*journal
)
703 write_lock(&journal
->j_state_lock
);
704 ++journal
->j_barrier_count
;
706 /* Wait until there are no reserved handles */
707 if (atomic_read(&journal
->j_reserved_credits
)) {
708 write_unlock(&journal
->j_state_lock
);
709 wait_event(journal
->j_wait_reserved
,
710 atomic_read(&journal
->j_reserved_credits
) == 0);
711 write_lock(&journal
->j_state_lock
);
714 /* Wait until there are no running updates */
716 transaction_t
*transaction
= journal
->j_running_transaction
;
721 spin_lock(&transaction
->t_handle_lock
);
722 prepare_to_wait(&journal
->j_wait_updates
, &wait
,
723 TASK_UNINTERRUPTIBLE
);
724 if (!atomic_read(&transaction
->t_updates
)) {
725 spin_unlock(&transaction
->t_handle_lock
);
726 finish_wait(&journal
->j_wait_updates
, &wait
);
729 spin_unlock(&transaction
->t_handle_lock
);
730 write_unlock(&journal
->j_state_lock
);
732 finish_wait(&journal
->j_wait_updates
, &wait
);
733 write_lock(&journal
->j_state_lock
);
735 write_unlock(&journal
->j_state_lock
);
738 * We have now established a barrier against other normal updates, but
739 * we also need to barrier against other jbd2_journal_lock_updates() calls
740 * to make sure that we serialise special journal-locked operations
743 mutex_lock(&journal
->j_barrier
);
747 * void jbd2_journal_unlock_updates (journal_t* journal) - release barrier
748 * @journal: Journal to release the barrier on.
750 * Release a transaction barrier obtained with jbd2_journal_lock_updates().
752 * Should be called without the journal lock held.
754 void jbd2_journal_unlock_updates (journal_t
*journal
)
756 J_ASSERT(journal
->j_barrier_count
!= 0);
758 mutex_unlock(&journal
->j_barrier
);
759 write_lock(&journal
->j_state_lock
);
760 --journal
->j_barrier_count
;
761 write_unlock(&journal
->j_state_lock
);
762 wake_up(&journal
->j_wait_transaction_locked
);
765 static void warn_dirty_buffer(struct buffer_head
*bh
)
768 "JBD2: Spotted dirty metadata buffer (dev = %pg, blocknr = %llu). "
769 "There's a risk of filesystem corruption in case of system "
771 bh
->b_bdev
, (unsigned long long)bh
->b_blocknr
);
774 /* Call t_frozen trigger and copy buffer data into jh->b_frozen_data. */
775 static void jbd2_freeze_jh_data(struct journal_head
*jh
)
780 struct buffer_head
*bh
= jh2bh(jh
);
782 J_EXPECT_JH(jh
, buffer_uptodate(bh
), "Possible IO failure.\n");
784 offset
= offset_in_page(bh
->b_data
);
785 source
= kmap_atomic(page
);
786 /* Fire data frozen trigger just before we copy the data */
787 jbd2_buffer_frozen_trigger(jh
, source
+ offset
, jh
->b_triggers
);
788 memcpy(jh
->b_frozen_data
, source
+ offset
, bh
->b_size
);
789 kunmap_atomic(source
);
792 * Now that the frozen data is saved off, we need to store any matching
795 jh
->b_frozen_triggers
= jh
->b_triggers
;
799 * If the buffer is already part of the current transaction, then there
800 * is nothing we need to do. If it is already part of a prior
801 * transaction which we are still committing to disk, then we need to
802 * make sure that we do not overwrite the old copy: we do copy-out to
803 * preserve the copy going to disk. We also account the buffer against
804 * the handle's metadata buffer credits (unless the buffer is already
805 * part of the transaction, that is).
809 do_get_write_access(handle_t
*handle
, struct journal_head
*jh
,
812 struct buffer_head
*bh
;
813 transaction_t
*transaction
= handle
->h_transaction
;
816 char *frozen_buffer
= NULL
;
817 unsigned long start_lock
, time_lock
;
819 if (is_handle_aborted(handle
))
821 journal
= transaction
->t_journal
;
823 jbd_debug(5, "journal_head %p, force_copy %d\n", jh
, force_copy
);
825 JBUFFER_TRACE(jh
, "entry");
829 /* @@@ Need to check for errors here at some point. */
831 start_lock
= jiffies
;
833 jbd_lock_bh_state(bh
);
835 /* If it takes too long to lock the buffer, trace it */
836 time_lock
= jbd2_time_diff(start_lock
, jiffies
);
837 if (time_lock
> HZ
/10)
838 trace_jbd2_lock_buffer_stall(bh
->b_bdev
->bd_dev
,
839 jiffies_to_msecs(time_lock
));
841 /* We now hold the buffer lock so it is safe to query the buffer
842 * state. Is the buffer dirty?
844 * If so, there are two possibilities. The buffer may be
845 * non-journaled, and undergoing a quite legitimate writeback.
846 * Otherwise, it is journaled, and we don't expect dirty buffers
847 * in that state (the buffers should be marked JBD_Dirty
848 * instead.) So either the IO is being done under our own
849 * control and this is a bug, or it's a third party IO such as
850 * dump(8) (which may leave the buffer scheduled for read ---
851 * ie. locked but not dirty) or tune2fs (which may actually have
852 * the buffer dirtied, ugh.) */
854 if (buffer_dirty(bh
)) {
856 * First question: is this buffer already part of the current
857 * transaction or the existing committing transaction?
859 if (jh
->b_transaction
) {
861 jh
->b_transaction
== transaction
||
863 journal
->j_committing_transaction
);
864 if (jh
->b_next_transaction
)
865 J_ASSERT_JH(jh
, jh
->b_next_transaction
==
867 warn_dirty_buffer(bh
);
870 * In any case we need to clean the dirty flag and we must
871 * do it under the buffer lock to be sure we don't race
872 * with running write-out.
874 JBUFFER_TRACE(jh
, "Journalling dirty buffer");
875 clear_buffer_dirty(bh
);
876 set_buffer_jbddirty(bh
);
882 if (is_handle_aborted(handle
)) {
883 jbd_unlock_bh_state(bh
);
889 * The buffer is already part of this transaction if b_transaction or
890 * b_next_transaction points to it
892 if (jh
->b_transaction
== transaction
||
893 jh
->b_next_transaction
== transaction
)
897 * this is the first time this transaction is touching this buffer,
898 * reset the modified flag
903 * If the buffer is not journaled right now, we need to make sure it
904 * doesn't get written to disk before the caller actually commits the
907 if (!jh
->b_transaction
) {
908 JBUFFER_TRACE(jh
, "no transaction");
909 J_ASSERT_JH(jh
, !jh
->b_next_transaction
);
910 JBUFFER_TRACE(jh
, "file as BJ_Reserved");
912 * Make sure all stores to jh (b_modified, b_frozen_data) are
913 * visible before attaching it to the running transaction.
914 * Paired with barrier in jbd2_write_access_granted()
917 spin_lock(&journal
->j_list_lock
);
918 __jbd2_journal_file_buffer(jh
, transaction
, BJ_Reserved
);
919 spin_unlock(&journal
->j_list_lock
);
923 * If there is already a copy-out version of this buffer, then we don't
924 * need to make another one
926 if (jh
->b_frozen_data
) {
927 JBUFFER_TRACE(jh
, "has frozen data");
928 J_ASSERT_JH(jh
, jh
->b_next_transaction
== NULL
);
932 JBUFFER_TRACE(jh
, "owned by older transaction");
933 J_ASSERT_JH(jh
, jh
->b_next_transaction
== NULL
);
934 J_ASSERT_JH(jh
, jh
->b_transaction
== journal
->j_committing_transaction
);
937 * There is one case we have to be very careful about. If the
938 * committing transaction is currently writing this buffer out to disk
939 * and has NOT made a copy-out, then we cannot modify the buffer
940 * contents at all right now. The essence of copy-out is that it is
941 * the extra copy, not the primary copy, which gets journaled. If the
942 * primary copy is already going to disk then we cannot do copy-out
945 if (buffer_shadow(bh
)) {
946 JBUFFER_TRACE(jh
, "on shadow: sleep");
947 jbd_unlock_bh_state(bh
);
948 wait_on_bit_io(&bh
->b_state
, BH_Shadow
, TASK_UNINTERRUPTIBLE
);
953 * Only do the copy if the currently-owning transaction still needs it.
954 * If buffer isn't on BJ_Metadata list, the committing transaction is
955 * past that stage (here we use the fact that BH_Shadow is set under
956 * bh_state lock together with refiling to BJ_Shadow list and at this
957 * point we know the buffer doesn't have BH_Shadow set).
959 * Subtle point, though: if this is a get_undo_access, then we will be
960 * relying on the frozen_data to contain the new value of the
961 * committed_data record after the transaction, so we HAVE to force the
962 * frozen_data copy in that case.
964 if (jh
->b_jlist
== BJ_Metadata
|| force_copy
) {
965 JBUFFER_TRACE(jh
, "generate frozen data");
966 if (!frozen_buffer
) {
967 JBUFFER_TRACE(jh
, "allocate memory for buffer");
968 jbd_unlock_bh_state(bh
);
969 frozen_buffer
= jbd2_alloc(jh2bh(jh
)->b_size
, GFP_NOFS
);
970 if (!frozen_buffer
) {
971 printk(KERN_ERR
"%s: OOM for frozen_buffer\n",
973 JBUFFER_TRACE(jh
, "oom!");
979 jh
->b_frozen_data
= frozen_buffer
;
980 frozen_buffer
= NULL
;
981 jbd2_freeze_jh_data(jh
);
985 * Make sure all stores to jh (b_modified, b_frozen_data) are visible
986 * before attaching it to the running transaction. Paired with barrier
987 * in jbd2_write_access_granted()
990 jh
->b_next_transaction
= transaction
;
993 jbd_unlock_bh_state(bh
);
996 * If we are about to journal a buffer, then any revoke pending on it is
999 jbd2_journal_cancel_revoke(handle
, jh
);
1002 if (unlikely(frozen_buffer
)) /* It's usually NULL */
1003 jbd2_free(frozen_buffer
, bh
->b_size
);
1005 JBUFFER_TRACE(jh
, "exit");
1009 /* Fast check whether buffer is already attached to the required transaction */
1010 static bool jbd2_write_access_granted(handle_t
*handle
, struct buffer_head
*bh
,
1013 struct journal_head
*jh
;
1016 /* Dirty buffers require special handling... */
1017 if (buffer_dirty(bh
))
1021 * RCU protects us from dereferencing freed pages. So the checks we do
1022 * are guaranteed not to oops. However the jh slab object can get freed
1023 * & reallocated while we work with it. So we have to be careful. When
1024 * we see jh attached to the running transaction, we know it must stay
1025 * so until the transaction is committed. Thus jh won't be freed and
1026 * will be attached to the same bh while we run. However it can
1027 * happen jh gets freed, reallocated, and attached to the transaction
1028 * just after we get pointer to it from bh. So we have to be careful
1029 * and recheck jh still belongs to our bh before we return success.
1032 if (!buffer_jbd(bh
))
1034 /* This should be bh2jh() but that doesn't work with inline functions */
1035 jh
= READ_ONCE(bh
->b_private
);
1038 /* For undo access buffer must have data copied */
1039 if (undo
&& !jh
->b_committed_data
)
1041 if (jh
->b_transaction
!= handle
->h_transaction
&&
1042 jh
->b_next_transaction
!= handle
->h_transaction
)
1045 * There are two reasons for the barrier here:
1046 * 1) Make sure to fetch b_bh after we did previous checks so that we
1047 * detect when jh went through free, realloc, attach to transaction
1048 * while we were checking. Paired with implicit barrier in that path.
1049 * 2) So that access to bh done after jbd2_write_access_granted()
1050 * doesn't get reordered and see inconsistent state of concurrent
1051 * do_get_write_access().
1054 if (unlikely(jh
->b_bh
!= bh
))
1063 * int jbd2_journal_get_write_access() - notify intent to modify a buffer for metadata (not data) update.
1064 * @handle: transaction to add buffer modifications to
1065 * @bh: bh to be used for metadata writes
1067 * Returns an error code or 0 on success.
1069 * In full data journalling mode the buffer may be of type BJ_AsyncData,
1070 * because we're write()ing a buffer which is also part of a shared mapping.
1073 int jbd2_journal_get_write_access(handle_t
*handle
, struct buffer_head
*bh
)
1075 struct journal_head
*jh
;
1078 if (jbd2_write_access_granted(handle
, bh
, false))
1081 jh
= jbd2_journal_add_journal_head(bh
);
1082 /* We do not want to get caught playing with fields which the
1083 * log thread also manipulates. Make sure that the buffer
1084 * completes any outstanding IO before proceeding. */
1085 rc
= do_get_write_access(handle
, jh
, 0);
1086 jbd2_journal_put_journal_head(jh
);
1092 * When the user wants to journal a newly created buffer_head
1093 * (ie. getblk() returned a new buffer and we are going to populate it
1094 * manually rather than reading off disk), then we need to keep the
1095 * buffer_head locked until it has been completely filled with new
1096 * data. In this case, we should be able to make the assertion that
1097 * the bh is not already part of an existing transaction.
1099 * The buffer should already be locked by the caller by this point.
1100 * There is no lock ranking violation: it was a newly created,
1101 * unlocked buffer beforehand. */
1104 * int jbd2_journal_get_create_access () - notify intent to use newly created bh
1105 * @handle: transaction to new buffer to
1108 * Call this if you create a new bh.
1110 int jbd2_journal_get_create_access(handle_t
*handle
, struct buffer_head
*bh
)
1112 transaction_t
*transaction
= handle
->h_transaction
;
1114 struct journal_head
*jh
= jbd2_journal_add_journal_head(bh
);
1117 jbd_debug(5, "journal_head %p\n", jh
);
1119 if (is_handle_aborted(handle
))
1121 journal
= transaction
->t_journal
;
1124 JBUFFER_TRACE(jh
, "entry");
1126 * The buffer may already belong to this transaction due to pre-zeroing
1127 * in the filesystem's new_block code. It may also be on the previous,
1128 * committing transaction's lists, but it HAS to be in Forget state in
1129 * that case: the transaction must have deleted the buffer for it to be
1132 jbd_lock_bh_state(bh
);
1133 J_ASSERT_JH(jh
, (jh
->b_transaction
== transaction
||
1134 jh
->b_transaction
== NULL
||
1135 (jh
->b_transaction
== journal
->j_committing_transaction
&&
1136 jh
->b_jlist
== BJ_Forget
)));
1138 J_ASSERT_JH(jh
, jh
->b_next_transaction
== NULL
);
1139 J_ASSERT_JH(jh
, buffer_locked(jh2bh(jh
)));
1141 if (jh
->b_transaction
== NULL
) {
1143 * Previous jbd2_journal_forget() could have left the buffer
1144 * with jbddirty bit set because it was being committed. When
1145 * the commit finished, we've filed the buffer for
1146 * checkpointing and marked it dirty. Now we are reallocating
1147 * the buffer so the transaction freeing it must have
1148 * committed and so it's safe to clear the dirty bit.
1150 clear_buffer_dirty(jh2bh(jh
));
1151 /* first access by this transaction */
1154 JBUFFER_TRACE(jh
, "file as BJ_Reserved");
1155 spin_lock(&journal
->j_list_lock
);
1156 __jbd2_journal_file_buffer(jh
, transaction
, BJ_Reserved
);
1157 } else if (jh
->b_transaction
== journal
->j_committing_transaction
) {
1158 /* first access by this transaction */
1161 JBUFFER_TRACE(jh
, "set next transaction");
1162 spin_lock(&journal
->j_list_lock
);
1163 jh
->b_next_transaction
= transaction
;
1165 spin_unlock(&journal
->j_list_lock
);
1166 jbd_unlock_bh_state(bh
);
1169 * akpm: I added this. ext3_alloc_branch can pick up new indirect
1170 * blocks which contain freed but then revoked metadata. We need
1171 * to cancel the revoke in case we end up freeing it yet again
1172 * and the reallocating as data - this would cause a second revoke,
1173 * which hits an assertion error.
1175 JBUFFER_TRACE(jh
, "cancelling revoke");
1176 jbd2_journal_cancel_revoke(handle
, jh
);
1178 jbd2_journal_put_journal_head(jh
);
1183 * int jbd2_journal_get_undo_access() - Notify intent to modify metadata with
1184 * non-rewindable consequences
1185 * @handle: transaction
1186 * @bh: buffer to undo
1188 * Sometimes there is a need to distinguish between metadata which has
1189 * been committed to disk and that which has not. The ext3fs code uses
1190 * this for freeing and allocating space, we have to make sure that we
1191 * do not reuse freed space until the deallocation has been committed,
1192 * since if we overwrote that space we would make the delete
1193 * un-rewindable in case of a crash.
1195 * To deal with that, jbd2_journal_get_undo_access requests write access to a
1196 * buffer for parts of non-rewindable operations such as delete
1197 * operations on the bitmaps. The journaling code must keep a copy of
1198 * the buffer's contents prior to the undo_access call until such time
1199 * as we know that the buffer has definitely been committed to disk.
1201 * We never need to know which transaction the committed data is part
1202 * of, buffers touched here are guaranteed to be dirtied later and so
1203 * will be committed to a new transaction in due course, at which point
1204 * we can discard the old committed data pointer.
1206 * Returns error number or 0 on success.
1208 int jbd2_journal_get_undo_access(handle_t
*handle
, struct buffer_head
*bh
)
1211 struct journal_head
*jh
;
1212 char *committed_data
= NULL
;
1214 JBUFFER_TRACE(jh
, "entry");
1215 if (jbd2_write_access_granted(handle
, bh
, true))
1218 jh
= jbd2_journal_add_journal_head(bh
);
1220 * Do this first --- it can drop the journal lock, so we want to
1221 * make sure that obtaining the committed_data is done
1222 * atomically wrt. completion of any outstanding commits.
1224 err
= do_get_write_access(handle
, jh
, 1);
1229 if (!jh
->b_committed_data
) {
1230 committed_data
= jbd2_alloc(jh2bh(jh
)->b_size
, GFP_NOFS
);
1231 if (!committed_data
) {
1232 printk(KERN_ERR
"%s: No memory for committed data\n",
1239 jbd_lock_bh_state(bh
);
1240 if (!jh
->b_committed_data
) {
1241 /* Copy out the current buffer contents into the
1242 * preserved, committed copy. */
1243 JBUFFER_TRACE(jh
, "generate b_committed data");
1244 if (!committed_data
) {
1245 jbd_unlock_bh_state(bh
);
1249 jh
->b_committed_data
= committed_data
;
1250 committed_data
= NULL
;
1251 memcpy(jh
->b_committed_data
, bh
->b_data
, bh
->b_size
);
1253 jbd_unlock_bh_state(bh
);
1255 jbd2_journal_put_journal_head(jh
);
1256 if (unlikely(committed_data
))
1257 jbd2_free(committed_data
, bh
->b_size
);
1262 * void jbd2_journal_set_triggers() - Add triggers for commit writeout
1263 * @bh: buffer to trigger on
1264 * @type: struct jbd2_buffer_trigger_type containing the trigger(s).
1266 * Set any triggers on this journal_head. This is always safe, because
1267 * triggers for a committing buffer will be saved off, and triggers for
1268 * a running transaction will match the buffer in that transaction.
1270 * Call with NULL to clear the triggers.
1272 void jbd2_journal_set_triggers(struct buffer_head
*bh
,
1273 struct jbd2_buffer_trigger_type
*type
)
1275 struct journal_head
*jh
= jbd2_journal_grab_journal_head(bh
);
1279 jh
->b_triggers
= type
;
1280 jbd2_journal_put_journal_head(jh
);
1283 void jbd2_buffer_frozen_trigger(struct journal_head
*jh
, void *mapped_data
,
1284 struct jbd2_buffer_trigger_type
*triggers
)
1286 struct buffer_head
*bh
= jh2bh(jh
);
1288 if (!triggers
|| !triggers
->t_frozen
)
1291 triggers
->t_frozen(triggers
, bh
, mapped_data
, bh
->b_size
);
1294 void jbd2_buffer_abort_trigger(struct journal_head
*jh
,
1295 struct jbd2_buffer_trigger_type
*triggers
)
1297 if (!triggers
|| !triggers
->t_abort
)
1300 triggers
->t_abort(triggers
, jh2bh(jh
));
1304 * int jbd2_journal_dirty_metadata() - mark a buffer as containing dirty metadata
1305 * @handle: transaction to add buffer to.
1306 * @bh: buffer to mark
1308 * mark dirty metadata which needs to be journaled as part of the current
1311 * The buffer must have previously had jbd2_journal_get_write_access()
1312 * called so that it has a valid journal_head attached to the buffer
1315 * The buffer is placed on the transaction's metadata list and is marked
1316 * as belonging to the transaction.
1318 * Returns error number or 0 on success.
1320 * Special care needs to be taken if the buffer already belongs to the
1321 * current committing transaction (in which case we should have frozen
1322 * data present for that commit). In that case, we don't relink the
1323 * buffer: that only gets done when the old transaction finally
1324 * completes its commit.
1326 int jbd2_journal_dirty_metadata(handle_t
*handle
, struct buffer_head
*bh
)
1328 transaction_t
*transaction
= handle
->h_transaction
;
1330 struct journal_head
*jh
;
1333 if (is_handle_aborted(handle
))
1335 if (!buffer_jbd(bh
)) {
1340 * We don't grab jh reference here since the buffer must be part
1341 * of the running transaction.
1345 * This and the following assertions are unreliable since we may see jh
1346 * in inconsistent state unless we grab bh_state lock. But this is
1347 * crucial to catch bugs so let's do a reliable check until the
1348 * lockless handling is fully proven.
1350 if (jh
->b_transaction
!= transaction
&&
1351 jh
->b_next_transaction
!= transaction
) {
1352 jbd_lock_bh_state(bh
);
1353 J_ASSERT_JH(jh
, jh
->b_transaction
== transaction
||
1354 jh
->b_next_transaction
== transaction
);
1355 jbd_unlock_bh_state(bh
);
1357 if (jh
->b_modified
== 1) {
1358 /* If it's in our transaction it must be in BJ_Metadata list. */
1359 if (jh
->b_transaction
== transaction
&&
1360 jh
->b_jlist
!= BJ_Metadata
) {
1361 jbd_lock_bh_state(bh
);
1362 J_ASSERT_JH(jh
, jh
->b_transaction
!= transaction
||
1363 jh
->b_jlist
== BJ_Metadata
);
1364 jbd_unlock_bh_state(bh
);
1369 journal
= transaction
->t_journal
;
1370 jbd_debug(5, "journal_head %p\n", jh
);
1371 JBUFFER_TRACE(jh
, "entry");
1373 jbd_lock_bh_state(bh
);
1375 if (jh
->b_modified
== 0) {
1377 * This buffer's got modified and becoming part
1378 * of the transaction. This needs to be done
1379 * once a transaction -bzzz
1382 if (handle
->h_buffer_credits
<= 0) {
1386 handle
->h_buffer_credits
--;
1390 * fastpath, to avoid expensive locking. If this buffer is already
1391 * on the running transaction's metadata list there is nothing to do.
1392 * Nobody can take it off again because there is a handle open.
1393 * I _think_ we're OK here with SMP barriers - a mistaken decision will
1394 * result in this test being false, so we go in and take the locks.
1396 if (jh
->b_transaction
== transaction
&& jh
->b_jlist
== BJ_Metadata
) {
1397 JBUFFER_TRACE(jh
, "fastpath");
1398 if (unlikely(jh
->b_transaction
!=
1399 journal
->j_running_transaction
)) {
1400 printk(KERN_ERR
"JBD2: %s: "
1401 "jh->b_transaction (%llu, %p, %u) != "
1402 "journal->j_running_transaction (%p, %u)\n",
1404 (unsigned long long) bh
->b_blocknr
,
1406 jh
->b_transaction
? jh
->b_transaction
->t_tid
: 0,
1407 journal
->j_running_transaction
,
1408 journal
->j_running_transaction
?
1409 journal
->j_running_transaction
->t_tid
: 0);
1415 set_buffer_jbddirty(bh
);
1418 * Metadata already on the current transaction list doesn't
1419 * need to be filed. Metadata on another transaction's list must
1420 * be committing, and will be refiled once the commit completes:
1421 * leave it alone for now.
1423 if (jh
->b_transaction
!= transaction
) {
1424 JBUFFER_TRACE(jh
, "already on other transaction");
1425 if (unlikely(((jh
->b_transaction
!=
1426 journal
->j_committing_transaction
)) ||
1427 (jh
->b_next_transaction
!= transaction
))) {
1428 printk(KERN_ERR
"jbd2_journal_dirty_metadata: %s: "
1429 "bad jh for block %llu: "
1430 "transaction (%p, %u), "
1431 "jh->b_transaction (%p, %u), "
1432 "jh->b_next_transaction (%p, %u), jlist %u\n",
1434 (unsigned long long) bh
->b_blocknr
,
1435 transaction
, transaction
->t_tid
,
1438 jh
->b_transaction
->t_tid
: 0,
1439 jh
->b_next_transaction
,
1440 jh
->b_next_transaction
?
1441 jh
->b_next_transaction
->t_tid
: 0,
1446 /* And this case is illegal: we can't reuse another
1447 * transaction's data buffer, ever. */
1451 /* That test should have eliminated the following case: */
1452 J_ASSERT_JH(jh
, jh
->b_frozen_data
== NULL
);
1454 JBUFFER_TRACE(jh
, "file as BJ_Metadata");
1455 spin_lock(&journal
->j_list_lock
);
1456 __jbd2_journal_file_buffer(jh
, transaction
, BJ_Metadata
);
1457 spin_unlock(&journal
->j_list_lock
);
1459 jbd_unlock_bh_state(bh
);
1461 JBUFFER_TRACE(jh
, "exit");
1466 * void jbd2_journal_forget() - bforget() for potentially-journaled buffers.
1467 * @handle: transaction handle
1468 * @bh: bh to 'forget'
1470 * We can only do the bforget if there are no commits pending against the
1471 * buffer. If the buffer is dirty in the current running transaction we
1472 * can safely unlink it.
1474 * bh may not be a journalled buffer at all - it may be a non-JBD
1475 * buffer which came off the hashtable. Check for this.
1477 * Decrements bh->b_count by one.
1479 * Allow this call even if the handle has aborted --- it may be part of
1480 * the caller's cleanup after an abort.
1482 int jbd2_journal_forget (handle_t
*handle
, struct buffer_head
*bh
)
1484 transaction_t
*transaction
= handle
->h_transaction
;
1486 struct journal_head
*jh
;
1487 int drop_reserve
= 0;
1489 int was_modified
= 0;
1491 if (is_handle_aborted(handle
))
1493 journal
= transaction
->t_journal
;
1495 BUFFER_TRACE(bh
, "entry");
1497 jbd_lock_bh_state(bh
);
1499 if (!buffer_jbd(bh
))
1503 /* Critical error: attempting to delete a bitmap buffer, maybe?
1504 * Don't do any jbd operations, and return an error. */
1505 if (!J_EXPECT_JH(jh
, !jh
->b_committed_data
,
1506 "inconsistent data on disk")) {
1511 /* keep track of whether or not this transaction modified us */
1512 was_modified
= jh
->b_modified
;
1515 * The buffer's going from the transaction, we must drop
1516 * all references -bzzz
1520 if (jh
->b_transaction
== transaction
) {
1521 J_ASSERT_JH(jh
, !jh
->b_frozen_data
);
1523 /* If we are forgetting a buffer which is already part
1524 * of this transaction, then we can just drop it from
1525 * the transaction immediately. */
1526 clear_buffer_dirty(bh
);
1527 clear_buffer_jbddirty(bh
);
1529 JBUFFER_TRACE(jh
, "belongs to current transaction: unfile");
1532 * we only want to drop a reference if this transaction
1533 * modified the buffer
1539 * We are no longer going to journal this buffer.
1540 * However, the commit of this transaction is still
1541 * important to the buffer: the delete that we are now
1542 * processing might obsolete an old log entry, so by
1543 * committing, we can satisfy the buffer's checkpoint.
1545 * So, if we have a checkpoint on the buffer, we should
1546 * now refile the buffer on our BJ_Forget list so that
1547 * we know to remove the checkpoint after we commit.
1550 spin_lock(&journal
->j_list_lock
);
1551 if (jh
->b_cp_transaction
) {
1552 __jbd2_journal_temp_unlink_buffer(jh
);
1553 __jbd2_journal_file_buffer(jh
, transaction
, BJ_Forget
);
1555 __jbd2_journal_unfile_buffer(jh
);
1556 if (!buffer_jbd(bh
)) {
1557 spin_unlock(&journal
->j_list_lock
);
1558 jbd_unlock_bh_state(bh
);
1563 spin_unlock(&journal
->j_list_lock
);
1564 } else if (jh
->b_transaction
) {
1565 J_ASSERT_JH(jh
, (jh
->b_transaction
==
1566 journal
->j_committing_transaction
));
1567 /* However, if the buffer is still owned by a prior
1568 * (committing) transaction, we can't drop it yet... */
1569 JBUFFER_TRACE(jh
, "belongs to older transaction");
1570 /* ... but we CAN drop it from the new transaction if we
1571 * have also modified it since the original commit. */
1573 if (jh
->b_next_transaction
) {
1574 J_ASSERT(jh
->b_next_transaction
== transaction
);
1575 spin_lock(&journal
->j_list_lock
);
1576 jh
->b_next_transaction
= NULL
;
1577 spin_unlock(&journal
->j_list_lock
);
1580 * only drop a reference if this transaction modified
1589 jbd_unlock_bh_state(bh
);
1593 /* no need to reserve log space for this block -bzzz */
1594 handle
->h_buffer_credits
++;
1600 * int jbd2_journal_stop() - complete a transaction
1601 * @handle: tranaction to complete.
1603 * All done for a particular handle.
1605 * There is not much action needed here. We just return any remaining
1606 * buffer credits to the transaction and remove the handle. The only
1607 * complication is that we need to start a commit operation if the
1608 * filesystem is marked for synchronous update.
1610 * jbd2_journal_stop itself will not usually return an error, but it may
1611 * do so in unusual circumstances. In particular, expect it to
1612 * return -EIO if a jbd2_journal_abort has been executed since the
1613 * transaction began.
1615 int jbd2_journal_stop(handle_t
*handle
)
1617 transaction_t
*transaction
= handle
->h_transaction
;
1619 int err
= 0, wait_for_commit
= 0;
1625 * Handle is already detached from the transaction so
1626 * there is nothing to do other than decrease a refcount,
1627 * or free the handle if refcount drops to zero
1629 if (--handle
->h_ref
> 0) {
1630 jbd_debug(4, "h_ref %d -> %d\n", handle
->h_ref
+ 1,
1634 if (handle
->h_rsv_handle
)
1635 jbd2_free_handle(handle
->h_rsv_handle
);
1639 journal
= transaction
->t_journal
;
1641 J_ASSERT(journal_current_handle() == handle
);
1643 if (is_handle_aborted(handle
))
1646 J_ASSERT(atomic_read(&transaction
->t_updates
) > 0);
1648 if (--handle
->h_ref
> 0) {
1649 jbd_debug(4, "h_ref %d -> %d\n", handle
->h_ref
+ 1,
1654 jbd_debug(4, "Handle %p going down\n", handle
);
1655 trace_jbd2_handle_stats(journal
->j_fs_dev
->bd_dev
,
1657 handle
->h_type
, handle
->h_line_no
,
1658 jiffies
- handle
->h_start_jiffies
,
1659 handle
->h_sync
, handle
->h_requested_credits
,
1660 (handle
->h_requested_credits
-
1661 handle
->h_buffer_credits
));
1664 * Implement synchronous transaction batching. If the handle
1665 * was synchronous, don't force a commit immediately. Let's
1666 * yield and let another thread piggyback onto this
1667 * transaction. Keep doing that while new threads continue to
1668 * arrive. It doesn't cost much - we're about to run a commit
1669 * and sleep on IO anyway. Speeds up many-threaded, many-dir
1670 * operations by 30x or more...
1672 * We try and optimize the sleep time against what the
1673 * underlying disk can do, instead of having a static sleep
1674 * time. This is useful for the case where our storage is so
1675 * fast that it is more optimal to go ahead and force a flush
1676 * and wait for the transaction to be committed than it is to
1677 * wait for an arbitrary amount of time for new writers to
1678 * join the transaction. We achieve this by measuring how
1679 * long it takes to commit a transaction, and compare it with
1680 * how long this transaction has been running, and if run time
1681 * < commit time then we sleep for the delta and commit. This
1682 * greatly helps super fast disks that would see slowdowns as
1683 * more threads started doing fsyncs.
1685 * But don't do this if this process was the most recent one
1686 * to perform a synchronous write. We do this to detect the
1687 * case where a single process is doing a stream of sync
1688 * writes. No point in waiting for joiners in that case.
1690 * Setting max_batch_time to 0 disables this completely.
1693 if (handle
->h_sync
&& journal
->j_last_sync_writer
!= pid
&&
1694 journal
->j_max_batch_time
) {
1695 u64 commit_time
, trans_time
;
1697 journal
->j_last_sync_writer
= pid
;
1699 read_lock(&journal
->j_state_lock
);
1700 commit_time
= journal
->j_average_commit_time
;
1701 read_unlock(&journal
->j_state_lock
);
1703 trans_time
= ktime_to_ns(ktime_sub(ktime_get(),
1704 transaction
->t_start_time
));
1706 commit_time
= max_t(u64
, commit_time
,
1707 1000*journal
->j_min_batch_time
);
1708 commit_time
= min_t(u64
, commit_time
,
1709 1000*journal
->j_max_batch_time
);
1711 if (trans_time
< commit_time
) {
1712 ktime_t expires
= ktime_add_ns(ktime_get(),
1714 set_current_state(TASK_UNINTERRUPTIBLE
);
1715 schedule_hrtimeout(&expires
, HRTIMER_MODE_ABS
);
1720 transaction
->t_synchronous_commit
= 1;
1721 current
->journal_info
= NULL
;
1722 atomic_sub(handle
->h_buffer_credits
,
1723 &transaction
->t_outstanding_credits
);
1726 * If the handle is marked SYNC, we need to set another commit
1727 * going! We also want to force a commit if the current
1728 * transaction is occupying too much of the log, or if the
1729 * transaction is too old now.
1731 if (handle
->h_sync
||
1732 (atomic_read(&transaction
->t_outstanding_credits
) >
1733 journal
->j_max_transaction_buffers
) ||
1734 time_after_eq(jiffies
, transaction
->t_expires
)) {
1735 /* Do this even for aborted journals: an abort still
1736 * completes the commit thread, it just doesn't write
1737 * anything to disk. */
1739 jbd_debug(2, "transaction too old, requesting commit for "
1740 "handle %p\n", handle
);
1741 /* This is non-blocking */
1742 jbd2_log_start_commit(journal
, transaction
->t_tid
);
1745 * Special case: JBD2_SYNC synchronous updates require us
1746 * to wait for the commit to complete.
1748 if (handle
->h_sync
&& !(current
->flags
& PF_MEMALLOC
))
1749 wait_for_commit
= 1;
1753 * Once we drop t_updates, if it goes to zero the transaction
1754 * could start committing on us and eventually disappear. So
1755 * once we do this, we must not dereference transaction
1758 tid
= transaction
->t_tid
;
1759 if (atomic_dec_and_test(&transaction
->t_updates
)) {
1760 wake_up(&journal
->j_wait_updates
);
1761 if (journal
->j_barrier_count
)
1762 wake_up(&journal
->j_wait_transaction_locked
);
1765 if (wait_for_commit
)
1766 err
= jbd2_log_wait_commit(journal
, tid
);
1768 lock_map_release(&handle
->h_lockdep_map
);
1770 if (handle
->h_rsv_handle
)
1771 jbd2_journal_free_reserved(handle
->h_rsv_handle
);
1773 jbd2_free_handle(handle
);
1779 * List management code snippets: various functions for manipulating the
1780 * transaction buffer lists.
1785 * Append a buffer to a transaction list, given the transaction's list head
1788 * j_list_lock is held.
1790 * jbd_lock_bh_state(jh2bh(jh)) is held.
1794 __blist_add_buffer(struct journal_head
**list
, struct journal_head
*jh
)
1797 jh
->b_tnext
= jh
->b_tprev
= jh
;
1800 /* Insert at the tail of the list to preserve order */
1801 struct journal_head
*first
= *list
, *last
= first
->b_tprev
;
1803 jh
->b_tnext
= first
;
1804 last
->b_tnext
= first
->b_tprev
= jh
;
1809 * Remove a buffer from a transaction list, given the transaction's list
1812 * Called with j_list_lock held, and the journal may not be locked.
1814 * jbd_lock_bh_state(jh2bh(jh)) is held.
1818 __blist_del_buffer(struct journal_head
**list
, struct journal_head
*jh
)
1821 *list
= jh
->b_tnext
;
1825 jh
->b_tprev
->b_tnext
= jh
->b_tnext
;
1826 jh
->b_tnext
->b_tprev
= jh
->b_tprev
;
1830 * Remove a buffer from the appropriate transaction list.
1832 * Note that this function can *change* the value of
1833 * bh->b_transaction->t_buffers, t_forget, t_shadow_list, t_log_list or
1834 * t_reserved_list. If the caller is holding onto a copy of one of these
1835 * pointers, it could go bad. Generally the caller needs to re-read the
1836 * pointer from the transaction_t.
1838 * Called under j_list_lock.
1840 static void __jbd2_journal_temp_unlink_buffer(struct journal_head
*jh
)
1842 struct journal_head
**list
= NULL
;
1843 transaction_t
*transaction
;
1844 struct buffer_head
*bh
= jh2bh(jh
);
1846 J_ASSERT_JH(jh
, jbd_is_locked_bh_state(bh
));
1847 transaction
= jh
->b_transaction
;
1849 assert_spin_locked(&transaction
->t_journal
->j_list_lock
);
1851 J_ASSERT_JH(jh
, jh
->b_jlist
< BJ_Types
);
1852 if (jh
->b_jlist
!= BJ_None
)
1853 J_ASSERT_JH(jh
, transaction
!= NULL
);
1855 switch (jh
->b_jlist
) {
1859 transaction
->t_nr_buffers
--;
1860 J_ASSERT_JH(jh
, transaction
->t_nr_buffers
>= 0);
1861 list
= &transaction
->t_buffers
;
1864 list
= &transaction
->t_forget
;
1867 list
= &transaction
->t_shadow_list
;
1870 list
= &transaction
->t_reserved_list
;
1874 __blist_del_buffer(list
, jh
);
1875 jh
->b_jlist
= BJ_None
;
1876 if (test_clear_buffer_jbddirty(bh
))
1877 mark_buffer_dirty(bh
); /* Expose it to the VM */
1881 * Remove buffer from all transactions.
1883 * Called with bh_state lock and j_list_lock
1885 * jh and bh may be already freed when this function returns.
1887 static void __jbd2_journal_unfile_buffer(struct journal_head
*jh
)
1889 __jbd2_journal_temp_unlink_buffer(jh
);
1890 jh
->b_transaction
= NULL
;
1891 jbd2_journal_put_journal_head(jh
);
1894 void jbd2_journal_unfile_buffer(journal_t
*journal
, struct journal_head
*jh
)
1896 struct buffer_head
*bh
= jh2bh(jh
);
1898 /* Get reference so that buffer cannot be freed before we unlock it */
1900 jbd_lock_bh_state(bh
);
1901 spin_lock(&journal
->j_list_lock
);
1902 __jbd2_journal_unfile_buffer(jh
);
1903 spin_unlock(&journal
->j_list_lock
);
1904 jbd_unlock_bh_state(bh
);
1909 * Called from jbd2_journal_try_to_free_buffers().
1911 * Called under jbd_lock_bh_state(bh)
1914 __journal_try_to_free_buffer(journal_t
*journal
, struct buffer_head
*bh
)
1916 struct journal_head
*jh
;
1920 if (buffer_locked(bh
) || buffer_dirty(bh
))
1923 if (jh
->b_next_transaction
!= NULL
|| jh
->b_transaction
!= NULL
)
1926 spin_lock(&journal
->j_list_lock
);
1927 if (jh
->b_cp_transaction
!= NULL
) {
1928 /* written-back checkpointed metadata buffer */
1929 JBUFFER_TRACE(jh
, "remove from checkpoint list");
1930 __jbd2_journal_remove_checkpoint(jh
);
1932 spin_unlock(&journal
->j_list_lock
);
1938 * int jbd2_journal_try_to_free_buffers() - try to free page buffers.
1939 * @journal: journal for operation
1940 * @page: to try and free
1941 * @gfp_mask: we use the mask to detect how hard should we try to release
1942 * buffers. If __GFP_DIRECT_RECLAIM and __GFP_FS is set, we wait for commit
1943 * code to release the buffers.
1946 * For all the buffers on this page,
1947 * if they are fully written out ordered data, move them onto BUF_CLEAN
1948 * so try_to_free_buffers() can reap them.
1950 * This function returns non-zero if we wish try_to_free_buffers()
1951 * to be called. We do this if the page is releasable by try_to_free_buffers().
1952 * We also do it if the page has locked or dirty buffers and the caller wants
1953 * us to perform sync or async writeout.
1955 * This complicates JBD locking somewhat. We aren't protected by the
1956 * BKL here. We wish to remove the buffer from its committing or
1957 * running transaction's ->t_datalist via __jbd2_journal_unfile_buffer.
1959 * This may *change* the value of transaction_t->t_datalist, so anyone
1960 * who looks at t_datalist needs to lock against this function.
1962 * Even worse, someone may be doing a jbd2_journal_dirty_data on this
1963 * buffer. So we need to lock against that. jbd2_journal_dirty_data()
1964 * will come out of the lock with the buffer dirty, which makes it
1965 * ineligible for release here.
1967 * Who else is affected by this? hmm... Really the only contender
1968 * is do_get_write_access() - it could be looking at the buffer while
1969 * journal_try_to_free_buffer() is changing its state. But that
1970 * cannot happen because we never reallocate freed data as metadata
1971 * while the data is part of a transaction. Yes?
1973 * Return 0 on failure, 1 on success
1975 int jbd2_journal_try_to_free_buffers(journal_t
*journal
,
1976 struct page
*page
, gfp_t gfp_mask
)
1978 struct buffer_head
*head
;
1979 struct buffer_head
*bh
;
1982 J_ASSERT(PageLocked(page
));
1984 head
= page_buffers(page
);
1987 struct journal_head
*jh
;
1990 * We take our own ref against the journal_head here to avoid
1991 * having to add tons of locking around each instance of
1992 * jbd2_journal_put_journal_head().
1994 jh
= jbd2_journal_grab_journal_head(bh
);
1998 jbd_lock_bh_state(bh
);
1999 __journal_try_to_free_buffer(journal
, bh
);
2000 jbd2_journal_put_journal_head(jh
);
2001 jbd_unlock_bh_state(bh
);
2004 } while ((bh
= bh
->b_this_page
) != head
);
2006 ret
= try_to_free_buffers(page
);
2013 * This buffer is no longer needed. If it is on an older transaction's
2014 * checkpoint list we need to record it on this transaction's forget list
2015 * to pin this buffer (and hence its checkpointing transaction) down until
2016 * this transaction commits. If the buffer isn't on a checkpoint list, we
2018 * Returns non-zero if JBD no longer has an interest in the buffer.
2020 * Called under j_list_lock.
2022 * Called under jbd_lock_bh_state(bh).
2024 static int __dispose_buffer(struct journal_head
*jh
, transaction_t
*transaction
)
2027 struct buffer_head
*bh
= jh2bh(jh
);
2029 if (jh
->b_cp_transaction
) {
2030 JBUFFER_TRACE(jh
, "on running+cp transaction");
2031 __jbd2_journal_temp_unlink_buffer(jh
);
2033 * We don't want to write the buffer anymore, clear the
2034 * bit so that we don't confuse checks in
2035 * __journal_file_buffer
2037 clear_buffer_dirty(bh
);
2038 __jbd2_journal_file_buffer(jh
, transaction
, BJ_Forget
);
2041 JBUFFER_TRACE(jh
, "on running transaction");
2042 __jbd2_journal_unfile_buffer(jh
);
2048 * jbd2_journal_invalidatepage
2050 * This code is tricky. It has a number of cases to deal with.
2052 * There are two invariants which this code relies on:
2054 * i_size must be updated on disk before we start calling invalidatepage on the
2057 * This is done in ext3 by defining an ext3_setattr method which
2058 * updates i_size before truncate gets going. By maintaining this
2059 * invariant, we can be sure that it is safe to throw away any buffers
2060 * attached to the current transaction: once the transaction commits,
2061 * we know that the data will not be needed.
2063 * Note however that we can *not* throw away data belonging to the
2064 * previous, committing transaction!
2066 * Any disk blocks which *are* part of the previous, committing
2067 * transaction (and which therefore cannot be discarded immediately) are
2068 * not going to be reused in the new running transaction
2070 * The bitmap committed_data images guarantee this: any block which is
2071 * allocated in one transaction and removed in the next will be marked
2072 * as in-use in the committed_data bitmap, so cannot be reused until
2073 * the next transaction to delete the block commits. This means that
2074 * leaving committing buffers dirty is quite safe: the disk blocks
2075 * cannot be reallocated to a different file and so buffer aliasing is
2079 * The above applies mainly to ordered data mode. In writeback mode we
2080 * don't make guarantees about the order in which data hits disk --- in
2081 * particular we don't guarantee that new dirty data is flushed before
2082 * transaction commit --- so it is always safe just to discard data
2083 * immediately in that mode. --sct
2087 * The journal_unmap_buffer helper function returns zero if the buffer
2088 * concerned remains pinned as an anonymous buffer belonging to an older
2091 * We're outside-transaction here. Either or both of j_running_transaction
2092 * and j_committing_transaction may be NULL.
2094 static int journal_unmap_buffer(journal_t
*journal
, struct buffer_head
*bh
,
2097 transaction_t
*transaction
;
2098 struct journal_head
*jh
;
2101 BUFFER_TRACE(bh
, "entry");
2104 * It is safe to proceed here without the j_list_lock because the
2105 * buffers cannot be stolen by try_to_free_buffers as long as we are
2106 * holding the page lock. --sct
2109 if (!buffer_jbd(bh
))
2110 goto zap_buffer_unlocked
;
2112 /* OK, we have data buffer in journaled mode */
2113 write_lock(&journal
->j_state_lock
);
2114 jbd_lock_bh_state(bh
);
2115 spin_lock(&journal
->j_list_lock
);
2117 jh
= jbd2_journal_grab_journal_head(bh
);
2119 goto zap_buffer_no_jh
;
2122 * We cannot remove the buffer from checkpoint lists until the
2123 * transaction adding inode to orphan list (let's call it T)
2124 * is committed. Otherwise if the transaction changing the
2125 * buffer would be cleaned from the journal before T is
2126 * committed, a crash will cause that the correct contents of
2127 * the buffer will be lost. On the other hand we have to
2128 * clear the buffer dirty bit at latest at the moment when the
2129 * transaction marking the buffer as freed in the filesystem
2130 * structures is committed because from that moment on the
2131 * block can be reallocated and used by a different page.
2132 * Since the block hasn't been freed yet but the inode has
2133 * already been added to orphan list, it is safe for us to add
2134 * the buffer to BJ_Forget list of the newest transaction.
2136 * Also we have to clear buffer_mapped flag of a truncated buffer
2137 * because the buffer_head may be attached to the page straddling
2138 * i_size (can happen only when blocksize < pagesize) and thus the
2139 * buffer_head can be reused when the file is extended again. So we end
2140 * up keeping around invalidated buffers attached to transactions'
2141 * BJ_Forget list just to stop checkpointing code from cleaning up
2142 * the transaction this buffer was modified in.
2144 transaction
= jh
->b_transaction
;
2145 if (transaction
== NULL
) {
2146 /* First case: not on any transaction. If it
2147 * has no checkpoint link, then we can zap it:
2148 * it's a writeback-mode buffer so we don't care
2149 * if it hits disk safely. */
2150 if (!jh
->b_cp_transaction
) {
2151 JBUFFER_TRACE(jh
, "not on any transaction: zap");
2155 if (!buffer_dirty(bh
)) {
2156 /* bdflush has written it. We can drop it now */
2157 __jbd2_journal_remove_checkpoint(jh
);
2161 /* OK, it must be in the journal but still not
2162 * written fully to disk: it's metadata or
2163 * journaled data... */
2165 if (journal
->j_running_transaction
) {
2166 /* ... and once the current transaction has
2167 * committed, the buffer won't be needed any
2169 JBUFFER_TRACE(jh
, "checkpointed: add to BJ_Forget");
2170 may_free
= __dispose_buffer(jh
,
2171 journal
->j_running_transaction
);
2174 /* There is no currently-running transaction. So the
2175 * orphan record which we wrote for this file must have
2176 * passed into commit. We must attach this buffer to
2177 * the committing transaction, if it exists. */
2178 if (journal
->j_committing_transaction
) {
2179 JBUFFER_TRACE(jh
, "give to committing trans");
2180 may_free
= __dispose_buffer(jh
,
2181 journal
->j_committing_transaction
);
2184 /* The orphan record's transaction has
2185 * committed. We can cleanse this buffer */
2186 clear_buffer_jbddirty(bh
);
2187 __jbd2_journal_remove_checkpoint(jh
);
2191 } else if (transaction
== journal
->j_committing_transaction
) {
2192 JBUFFER_TRACE(jh
, "on committing transaction");
2194 * The buffer is committing, we simply cannot touch
2195 * it. If the page is straddling i_size we have to wait
2196 * for commit and try again.
2199 jbd2_journal_put_journal_head(jh
);
2200 spin_unlock(&journal
->j_list_lock
);
2201 jbd_unlock_bh_state(bh
);
2202 write_unlock(&journal
->j_state_lock
);
2206 * OK, buffer won't be reachable after truncate. We just set
2207 * j_next_transaction to the running transaction (if there is
2208 * one) and mark buffer as freed so that commit code knows it
2209 * should clear dirty bits when it is done with the buffer.
2211 set_buffer_freed(bh
);
2212 if (journal
->j_running_transaction
&& buffer_jbddirty(bh
))
2213 jh
->b_next_transaction
= journal
->j_running_transaction
;
2214 jbd2_journal_put_journal_head(jh
);
2215 spin_unlock(&journal
->j_list_lock
);
2216 jbd_unlock_bh_state(bh
);
2217 write_unlock(&journal
->j_state_lock
);
2220 /* Good, the buffer belongs to the running transaction.
2221 * We are writing our own transaction's data, not any
2222 * previous one's, so it is safe to throw it away
2223 * (remember that we expect the filesystem to have set
2224 * i_size already for this truncate so recovery will not
2225 * expose the disk blocks we are discarding here.) */
2226 J_ASSERT_JH(jh
, transaction
== journal
->j_running_transaction
);
2227 JBUFFER_TRACE(jh
, "on running transaction");
2228 may_free
= __dispose_buffer(jh
, transaction
);
2233 * This is tricky. Although the buffer is truncated, it may be reused
2234 * if blocksize < pagesize and it is attached to the page straddling
2235 * EOF. Since the buffer might have been added to BJ_Forget list of the
2236 * running transaction, journal_get_write_access() won't clear
2237 * b_modified and credit accounting gets confused. So clear b_modified
2241 jbd2_journal_put_journal_head(jh
);
2243 spin_unlock(&journal
->j_list_lock
);
2244 jbd_unlock_bh_state(bh
);
2245 write_unlock(&journal
->j_state_lock
);
2246 zap_buffer_unlocked
:
2247 clear_buffer_dirty(bh
);
2248 J_ASSERT_BH(bh
, !buffer_jbddirty(bh
));
2249 clear_buffer_mapped(bh
);
2250 clear_buffer_req(bh
);
2251 clear_buffer_new(bh
);
2252 clear_buffer_delay(bh
);
2253 clear_buffer_unwritten(bh
);
2259 * void jbd2_journal_invalidatepage()
2260 * @journal: journal to use for flush...
2261 * @page: page to flush
2262 * @offset: start of the range to invalidate
2263 * @length: length of the range to invalidate
2265 * Reap page buffers containing data after in the specified range in page.
2266 * Can return -EBUSY if buffers are part of the committing transaction and
2267 * the page is straddling i_size. Caller then has to wait for current commit
2270 int jbd2_journal_invalidatepage(journal_t
*journal
,
2272 unsigned int offset
,
2273 unsigned int length
)
2275 struct buffer_head
*head
, *bh
, *next
;
2276 unsigned int stop
= offset
+ length
;
2277 unsigned int curr_off
= 0;
2278 int partial_page
= (offset
|| length
< PAGE_CACHE_SIZE
);
2282 if (!PageLocked(page
))
2284 if (!page_has_buffers(page
))
2287 BUG_ON(stop
> PAGE_CACHE_SIZE
|| stop
< length
);
2289 /* We will potentially be playing with lists other than just the
2290 * data lists (especially for journaled data mode), so be
2291 * cautious in our locking. */
2293 head
= bh
= page_buffers(page
);
2295 unsigned int next_off
= curr_off
+ bh
->b_size
;
2296 next
= bh
->b_this_page
;
2298 if (next_off
> stop
)
2301 if (offset
<= curr_off
) {
2302 /* This block is wholly outside the truncation point */
2304 ret
= journal_unmap_buffer(journal
, bh
, partial_page
);
2310 curr_off
= next_off
;
2313 } while (bh
!= head
);
2315 if (!partial_page
) {
2316 if (may_free
&& try_to_free_buffers(page
))
2317 J_ASSERT(!page_has_buffers(page
));
2323 * File a buffer on the given transaction list.
2325 void __jbd2_journal_file_buffer(struct journal_head
*jh
,
2326 transaction_t
*transaction
, int jlist
)
2328 struct journal_head
**list
= NULL
;
2330 struct buffer_head
*bh
= jh2bh(jh
);
2332 J_ASSERT_JH(jh
, jbd_is_locked_bh_state(bh
));
2333 assert_spin_locked(&transaction
->t_journal
->j_list_lock
);
2335 J_ASSERT_JH(jh
, jh
->b_jlist
< BJ_Types
);
2336 J_ASSERT_JH(jh
, jh
->b_transaction
== transaction
||
2337 jh
->b_transaction
== NULL
);
2339 if (jh
->b_transaction
&& jh
->b_jlist
== jlist
)
2342 if (jlist
== BJ_Metadata
|| jlist
== BJ_Reserved
||
2343 jlist
== BJ_Shadow
|| jlist
== BJ_Forget
) {
2345 * For metadata buffers, we track dirty bit in buffer_jbddirty
2346 * instead of buffer_dirty. We should not see a dirty bit set
2347 * here because we clear it in do_get_write_access but e.g.
2348 * tune2fs can modify the sb and set the dirty bit at any time
2349 * so we try to gracefully handle that.
2351 if (buffer_dirty(bh
))
2352 warn_dirty_buffer(bh
);
2353 if (test_clear_buffer_dirty(bh
) ||
2354 test_clear_buffer_jbddirty(bh
))
2358 if (jh
->b_transaction
)
2359 __jbd2_journal_temp_unlink_buffer(jh
);
2361 jbd2_journal_grab_journal_head(bh
);
2362 jh
->b_transaction
= transaction
;
2366 J_ASSERT_JH(jh
, !jh
->b_committed_data
);
2367 J_ASSERT_JH(jh
, !jh
->b_frozen_data
);
2370 transaction
->t_nr_buffers
++;
2371 list
= &transaction
->t_buffers
;
2374 list
= &transaction
->t_forget
;
2377 list
= &transaction
->t_shadow_list
;
2380 list
= &transaction
->t_reserved_list
;
2384 __blist_add_buffer(list
, jh
);
2385 jh
->b_jlist
= jlist
;
2388 set_buffer_jbddirty(bh
);
2391 void jbd2_journal_file_buffer(struct journal_head
*jh
,
2392 transaction_t
*transaction
, int jlist
)
2394 jbd_lock_bh_state(jh2bh(jh
));
2395 spin_lock(&transaction
->t_journal
->j_list_lock
);
2396 __jbd2_journal_file_buffer(jh
, transaction
, jlist
);
2397 spin_unlock(&transaction
->t_journal
->j_list_lock
);
2398 jbd_unlock_bh_state(jh2bh(jh
));
2402 * Remove a buffer from its current buffer list in preparation for
2403 * dropping it from its current transaction entirely. If the buffer has
2404 * already started to be used by a subsequent transaction, refile the
2405 * buffer on that transaction's metadata list.
2407 * Called under j_list_lock
2408 * Called under jbd_lock_bh_state(jh2bh(jh))
2410 * jh and bh may be already free when this function returns
2412 void __jbd2_journal_refile_buffer(struct journal_head
*jh
)
2414 int was_dirty
, jlist
;
2415 struct buffer_head
*bh
= jh2bh(jh
);
2417 J_ASSERT_JH(jh
, jbd_is_locked_bh_state(bh
));
2418 if (jh
->b_transaction
)
2419 assert_spin_locked(&jh
->b_transaction
->t_journal
->j_list_lock
);
2421 /* If the buffer is now unused, just drop it. */
2422 if (jh
->b_next_transaction
== NULL
) {
2423 __jbd2_journal_unfile_buffer(jh
);
2428 * It has been modified by a later transaction: add it to the new
2429 * transaction's metadata list.
2432 was_dirty
= test_clear_buffer_jbddirty(bh
);
2433 __jbd2_journal_temp_unlink_buffer(jh
);
2435 * We set b_transaction here because b_next_transaction will inherit
2436 * our jh reference and thus __jbd2_journal_file_buffer() must not
2439 jh
->b_transaction
= jh
->b_next_transaction
;
2440 jh
->b_next_transaction
= NULL
;
2441 if (buffer_freed(bh
))
2443 else if (jh
->b_modified
)
2444 jlist
= BJ_Metadata
;
2446 jlist
= BJ_Reserved
;
2447 __jbd2_journal_file_buffer(jh
, jh
->b_transaction
, jlist
);
2448 J_ASSERT_JH(jh
, jh
->b_transaction
->t_state
== T_RUNNING
);
2451 set_buffer_jbddirty(bh
);
2455 * __jbd2_journal_refile_buffer() with necessary locking added. We take our
2456 * bh reference so that we can safely unlock bh.
2458 * The jh and bh may be freed by this call.
2460 void jbd2_journal_refile_buffer(journal_t
*journal
, struct journal_head
*jh
)
2462 struct buffer_head
*bh
= jh2bh(jh
);
2464 /* Get reference so that buffer cannot be freed before we unlock it */
2466 jbd_lock_bh_state(bh
);
2467 spin_lock(&journal
->j_list_lock
);
2468 __jbd2_journal_refile_buffer(jh
);
2469 jbd_unlock_bh_state(bh
);
2470 spin_unlock(&journal
->j_list_lock
);
2475 * File inode in the inode list of the handle's transaction
2477 int jbd2_journal_file_inode(handle_t
*handle
, struct jbd2_inode
*jinode
)
2479 transaction_t
*transaction
= handle
->h_transaction
;
2482 if (is_handle_aborted(handle
))
2484 journal
= transaction
->t_journal
;
2486 jbd_debug(4, "Adding inode %lu, tid:%d\n", jinode
->i_vfs_inode
->i_ino
,
2487 transaction
->t_tid
);
2490 * First check whether inode isn't already on the transaction's
2491 * lists without taking the lock. Note that this check is safe
2492 * without the lock as we cannot race with somebody removing inode
2493 * from the transaction. The reason is that we remove inode from the
2494 * transaction only in journal_release_jbd_inode() and when we commit
2495 * the transaction. We are guarded from the first case by holding
2496 * a reference to the inode. We are safe against the second case
2497 * because if jinode->i_transaction == transaction, commit code
2498 * cannot touch the transaction because we hold reference to it,
2499 * and if jinode->i_next_transaction == transaction, commit code
2500 * will only file the inode where we want it.
2502 if (jinode
->i_transaction
== transaction
||
2503 jinode
->i_next_transaction
== transaction
)
2506 spin_lock(&journal
->j_list_lock
);
2508 if (jinode
->i_transaction
== transaction
||
2509 jinode
->i_next_transaction
== transaction
)
2513 * We only ever set this variable to 1 so the test is safe. Since
2514 * t_need_data_flush is likely to be set, we do the test to save some
2515 * cacheline bouncing
2517 if (!transaction
->t_need_data_flush
)
2518 transaction
->t_need_data_flush
= 1;
2519 /* On some different transaction's list - should be
2520 * the committing one */
2521 if (jinode
->i_transaction
) {
2522 J_ASSERT(jinode
->i_next_transaction
== NULL
);
2523 J_ASSERT(jinode
->i_transaction
==
2524 journal
->j_committing_transaction
);
2525 jinode
->i_next_transaction
= transaction
;
2528 /* Not on any transaction list... */
2529 J_ASSERT(!jinode
->i_next_transaction
);
2530 jinode
->i_transaction
= transaction
;
2531 list_add(&jinode
->i_list
, &transaction
->t_inode_list
);
2533 spin_unlock(&journal
->j_list_lock
);
2539 * File truncate and transaction commit interact with each other in a
2540 * non-trivial way. If a transaction writing data block A is
2541 * committing, we cannot discard the data by truncate until we have
2542 * written them. Otherwise if we crashed after the transaction with
2543 * write has committed but before the transaction with truncate has
2544 * committed, we could see stale data in block A. This function is a
2545 * helper to solve this problem. It starts writeout of the truncated
2546 * part in case it is in the committing transaction.
2548 * Filesystem code must call this function when inode is journaled in
2549 * ordered mode before truncation happens and after the inode has been
2550 * placed on orphan list with the new inode size. The second condition
2551 * avoids the race that someone writes new data and we start
2552 * committing the transaction after this function has been called but
2553 * before a transaction for truncate is started (and furthermore it
2554 * allows us to optimize the case where the addition to orphan list
2555 * happens in the same transaction as write --- we don't have to write
2556 * any data in such case).
2558 int jbd2_journal_begin_ordered_truncate(journal_t
*journal
,
2559 struct jbd2_inode
*jinode
,
2562 transaction_t
*inode_trans
, *commit_trans
;
2565 /* This is a quick check to avoid locking if not necessary */
2566 if (!jinode
->i_transaction
)
2568 /* Locks are here just to force reading of recent values, it is
2569 * enough that the transaction was not committing before we started
2570 * a transaction adding the inode to orphan list */
2571 read_lock(&journal
->j_state_lock
);
2572 commit_trans
= journal
->j_committing_transaction
;
2573 read_unlock(&journal
->j_state_lock
);
2574 spin_lock(&journal
->j_list_lock
);
2575 inode_trans
= jinode
->i_transaction
;
2576 spin_unlock(&journal
->j_list_lock
);
2577 if (inode_trans
== commit_trans
) {
2578 ret
= filemap_fdatawrite_range(jinode
->i_vfs_inode
->i_mapping
,
2579 new_size
, LLONG_MAX
);
2581 jbd2_journal_abort(journal
, ret
);