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
)
767 char b
[BDEVNAME_SIZE
];
770 "JBD2: Spotted dirty metadata buffer (dev = %s, blocknr = %llu). "
771 "There's a risk of filesystem corruption in case of system "
773 bdevname(bh
->b_bdev
, b
), (unsigned long long)bh
->b_blocknr
);
776 /* Call t_frozen trigger and copy buffer data into jh->b_frozen_data. */
777 static void jbd2_freeze_jh_data(struct journal_head
*jh
)
782 struct buffer_head
*bh
= jh2bh(jh
);
784 J_EXPECT_JH(jh
, buffer_uptodate(bh
), "Possible IO failure.\n");
786 offset
= offset_in_page(bh
->b_data
);
787 source
= kmap_atomic(page
);
788 /* Fire data frozen trigger just before we copy the data */
789 jbd2_buffer_frozen_trigger(jh
, source
+ offset
, jh
->b_triggers
);
790 memcpy(jh
->b_frozen_data
, source
+ offset
, bh
->b_size
);
791 kunmap_atomic(source
);
794 * Now that the frozen data is saved off, we need to store any matching
797 jh
->b_frozen_triggers
= jh
->b_triggers
;
801 * If the buffer is already part of the current transaction, then there
802 * is nothing we need to do. If it is already part of a prior
803 * transaction which we are still committing to disk, then we need to
804 * make sure that we do not overwrite the old copy: we do copy-out to
805 * preserve the copy going to disk. We also account the buffer against
806 * the handle's metadata buffer credits (unless the buffer is already
807 * part of the transaction, that is).
811 do_get_write_access(handle_t
*handle
, struct journal_head
*jh
,
814 struct buffer_head
*bh
;
815 transaction_t
*transaction
= handle
->h_transaction
;
818 char *frozen_buffer
= NULL
;
819 unsigned long start_lock
, time_lock
;
821 if (is_handle_aborted(handle
))
823 journal
= transaction
->t_journal
;
825 jbd_debug(5, "journal_head %p, force_copy %d\n", jh
, force_copy
);
827 JBUFFER_TRACE(jh
, "entry");
831 /* @@@ Need to check for errors here at some point. */
833 start_lock
= jiffies
;
835 jbd_lock_bh_state(bh
);
837 /* If it takes too long to lock the buffer, trace it */
838 time_lock
= jbd2_time_diff(start_lock
, jiffies
);
839 if (time_lock
> HZ
/10)
840 trace_jbd2_lock_buffer_stall(bh
->b_bdev
->bd_dev
,
841 jiffies_to_msecs(time_lock
));
843 /* We now hold the buffer lock so it is safe to query the buffer
844 * state. Is the buffer dirty?
846 * If so, there are two possibilities. The buffer may be
847 * non-journaled, and undergoing a quite legitimate writeback.
848 * Otherwise, it is journaled, and we don't expect dirty buffers
849 * in that state (the buffers should be marked JBD_Dirty
850 * instead.) So either the IO is being done under our own
851 * control and this is a bug, or it's a third party IO such as
852 * dump(8) (which may leave the buffer scheduled for read ---
853 * ie. locked but not dirty) or tune2fs (which may actually have
854 * the buffer dirtied, ugh.) */
856 if (buffer_dirty(bh
)) {
858 * First question: is this buffer already part of the current
859 * transaction or the existing committing transaction?
861 if (jh
->b_transaction
) {
863 jh
->b_transaction
== transaction
||
865 journal
->j_committing_transaction
);
866 if (jh
->b_next_transaction
)
867 J_ASSERT_JH(jh
, jh
->b_next_transaction
==
869 warn_dirty_buffer(bh
);
872 * In any case we need to clean the dirty flag and we must
873 * do it under the buffer lock to be sure we don't race
874 * with running write-out.
876 JBUFFER_TRACE(jh
, "Journalling dirty buffer");
877 clear_buffer_dirty(bh
);
878 set_buffer_jbddirty(bh
);
884 if (is_handle_aborted(handle
)) {
885 jbd_unlock_bh_state(bh
);
891 * The buffer is already part of this transaction if b_transaction or
892 * b_next_transaction points to it
894 if (jh
->b_transaction
== transaction
||
895 jh
->b_next_transaction
== transaction
)
899 * this is the first time this transaction is touching this buffer,
900 * reset the modified flag
905 * If the buffer is not journaled right now, we need to make sure it
906 * doesn't get written to disk before the caller actually commits the
909 if (!jh
->b_transaction
) {
910 JBUFFER_TRACE(jh
, "no transaction");
911 J_ASSERT_JH(jh
, !jh
->b_next_transaction
);
912 JBUFFER_TRACE(jh
, "file as BJ_Reserved");
914 * Make sure all stores to jh (b_modified, b_frozen_data) are
915 * visible before attaching it to the running transaction.
916 * Paired with barrier in jbd2_write_access_granted()
919 spin_lock(&journal
->j_list_lock
);
920 __jbd2_journal_file_buffer(jh
, transaction
, BJ_Reserved
);
921 spin_unlock(&journal
->j_list_lock
);
925 * If there is already a copy-out version of this buffer, then we don't
926 * need to make another one
928 if (jh
->b_frozen_data
) {
929 JBUFFER_TRACE(jh
, "has frozen data");
930 J_ASSERT_JH(jh
, jh
->b_next_transaction
== NULL
);
934 JBUFFER_TRACE(jh
, "owned by older transaction");
935 J_ASSERT_JH(jh
, jh
->b_next_transaction
== NULL
);
936 J_ASSERT_JH(jh
, jh
->b_transaction
== journal
->j_committing_transaction
);
939 * There is one case we have to be very careful about. If the
940 * committing transaction is currently writing this buffer out to disk
941 * and has NOT made a copy-out, then we cannot modify the buffer
942 * contents at all right now. The essence of copy-out is that it is
943 * the extra copy, not the primary copy, which gets journaled. If the
944 * primary copy is already going to disk then we cannot do copy-out
947 if (buffer_shadow(bh
)) {
948 JBUFFER_TRACE(jh
, "on shadow: sleep");
949 jbd_unlock_bh_state(bh
);
950 wait_on_bit_io(&bh
->b_state
, BH_Shadow
, TASK_UNINTERRUPTIBLE
);
955 * Only do the copy if the currently-owning transaction still needs it.
956 * If buffer isn't on BJ_Metadata list, the committing transaction is
957 * past that stage (here we use the fact that BH_Shadow is set under
958 * bh_state lock together with refiling to BJ_Shadow list and at this
959 * point we know the buffer doesn't have BH_Shadow set).
961 * Subtle point, though: if this is a get_undo_access, then we will be
962 * relying on the frozen_data to contain the new value of the
963 * committed_data record after the transaction, so we HAVE to force the
964 * frozen_data copy in that case.
966 if (jh
->b_jlist
== BJ_Metadata
|| force_copy
) {
967 JBUFFER_TRACE(jh
, "generate frozen data");
968 if (!frozen_buffer
) {
969 JBUFFER_TRACE(jh
, "allocate memory for buffer");
970 jbd_unlock_bh_state(bh
);
971 frozen_buffer
= jbd2_alloc(jh2bh(jh
)->b_size
, GFP_NOFS
);
972 if (!frozen_buffer
) {
973 printk(KERN_ERR
"%s: OOM for frozen_buffer\n",
975 JBUFFER_TRACE(jh
, "oom!");
981 jh
->b_frozen_data
= frozen_buffer
;
982 frozen_buffer
= NULL
;
983 jbd2_freeze_jh_data(jh
);
987 * Make sure all stores to jh (b_modified, b_frozen_data) are visible
988 * before attaching it to the running transaction. Paired with barrier
989 * in jbd2_write_access_granted()
992 jh
->b_next_transaction
= transaction
;
995 jbd_unlock_bh_state(bh
);
998 * If we are about to journal a buffer, then any revoke pending on it is
1001 jbd2_journal_cancel_revoke(handle
, jh
);
1004 if (unlikely(frozen_buffer
)) /* It's usually NULL */
1005 jbd2_free(frozen_buffer
, bh
->b_size
);
1007 JBUFFER_TRACE(jh
, "exit");
1011 /* Fast check whether buffer is already attached to the required transaction */
1012 static bool jbd2_write_access_granted(handle_t
*handle
, struct buffer_head
*bh
)
1014 struct journal_head
*jh
;
1017 /* Dirty buffers require special handling... */
1018 if (buffer_dirty(bh
))
1022 * RCU protects us from dereferencing freed pages. So the checks we do
1023 * are guaranteed not to oops. However the jh slab object can get freed
1024 * & reallocated while we work with it. So we have to be careful. When
1025 * we see jh attached to the running transaction, we know it must stay
1026 * so until the transaction is committed. Thus jh won't be freed and
1027 * will be attached to the same bh while we run. However it can
1028 * happen jh gets freed, reallocated, and attached to the transaction
1029 * just after we get pointer to it from bh. So we have to be careful
1030 * and recheck jh still belongs to our bh before we return success.
1033 if (!buffer_jbd(bh
))
1035 /* This should be bh2jh() but that doesn't work with inline functions */
1036 jh
= READ_ONCE(bh
->b_private
);
1039 if (jh
->b_transaction
!= handle
->h_transaction
&&
1040 jh
->b_next_transaction
!= handle
->h_transaction
)
1043 * There are two reasons for the barrier here:
1044 * 1) Make sure to fetch b_bh after we did previous checks so that we
1045 * detect when jh went through free, realloc, attach to transaction
1046 * while we were checking. Paired with implicit barrier in that path.
1047 * 2) So that access to bh done after jbd2_write_access_granted()
1048 * doesn't get reordered and see inconsistent state of concurrent
1049 * do_get_write_access().
1052 if (unlikely(jh
->b_bh
!= bh
))
1061 * int jbd2_journal_get_write_access() - notify intent to modify a buffer for metadata (not data) update.
1062 * @handle: transaction to add buffer modifications to
1063 * @bh: bh to be used for metadata writes
1065 * Returns an error code or 0 on success.
1067 * In full data journalling mode the buffer may be of type BJ_AsyncData,
1068 * because we're write()ing a buffer which is also part of a shared mapping.
1071 int jbd2_journal_get_write_access(handle_t
*handle
, struct buffer_head
*bh
)
1073 struct journal_head
*jh
;
1076 if (jbd2_write_access_granted(handle
, bh
))
1079 jh
= jbd2_journal_add_journal_head(bh
);
1080 /* We do not want to get caught playing with fields which the
1081 * log thread also manipulates. Make sure that the buffer
1082 * completes any outstanding IO before proceeding. */
1083 rc
= do_get_write_access(handle
, jh
, 0);
1084 jbd2_journal_put_journal_head(jh
);
1090 * When the user wants to journal a newly created buffer_head
1091 * (ie. getblk() returned a new buffer and we are going to populate it
1092 * manually rather than reading off disk), then we need to keep the
1093 * buffer_head locked until it has been completely filled with new
1094 * data. In this case, we should be able to make the assertion that
1095 * the bh is not already part of an existing transaction.
1097 * The buffer should already be locked by the caller by this point.
1098 * There is no lock ranking violation: it was a newly created,
1099 * unlocked buffer beforehand. */
1102 * int jbd2_journal_get_create_access () - notify intent to use newly created bh
1103 * @handle: transaction to new buffer to
1106 * Call this if you create a new bh.
1108 int jbd2_journal_get_create_access(handle_t
*handle
, struct buffer_head
*bh
)
1110 transaction_t
*transaction
= handle
->h_transaction
;
1112 struct journal_head
*jh
= jbd2_journal_add_journal_head(bh
);
1115 jbd_debug(5, "journal_head %p\n", jh
);
1117 if (is_handle_aborted(handle
))
1119 journal
= transaction
->t_journal
;
1122 JBUFFER_TRACE(jh
, "entry");
1124 * The buffer may already belong to this transaction due to pre-zeroing
1125 * in the filesystem's new_block code. It may also be on the previous,
1126 * committing transaction's lists, but it HAS to be in Forget state in
1127 * that case: the transaction must have deleted the buffer for it to be
1130 jbd_lock_bh_state(bh
);
1131 J_ASSERT_JH(jh
, (jh
->b_transaction
== transaction
||
1132 jh
->b_transaction
== NULL
||
1133 (jh
->b_transaction
== journal
->j_committing_transaction
&&
1134 jh
->b_jlist
== BJ_Forget
)));
1136 J_ASSERT_JH(jh
, jh
->b_next_transaction
== NULL
);
1137 J_ASSERT_JH(jh
, buffer_locked(jh2bh(jh
)));
1139 if (jh
->b_transaction
== NULL
) {
1141 * Previous jbd2_journal_forget() could have left the buffer
1142 * with jbddirty bit set because it was being committed. When
1143 * the commit finished, we've filed the buffer for
1144 * checkpointing and marked it dirty. Now we are reallocating
1145 * the buffer so the transaction freeing it must have
1146 * committed and so it's safe to clear the dirty bit.
1148 clear_buffer_dirty(jh2bh(jh
));
1149 /* first access by this transaction */
1152 JBUFFER_TRACE(jh
, "file as BJ_Reserved");
1153 spin_lock(&journal
->j_list_lock
);
1154 __jbd2_journal_file_buffer(jh
, transaction
, BJ_Reserved
);
1155 } else if (jh
->b_transaction
== journal
->j_committing_transaction
) {
1156 /* first access by this transaction */
1159 JBUFFER_TRACE(jh
, "set next transaction");
1160 spin_lock(&journal
->j_list_lock
);
1161 jh
->b_next_transaction
= transaction
;
1163 spin_unlock(&journal
->j_list_lock
);
1164 jbd_unlock_bh_state(bh
);
1167 * akpm: I added this. ext3_alloc_branch can pick up new indirect
1168 * blocks which contain freed but then revoked metadata. We need
1169 * to cancel the revoke in case we end up freeing it yet again
1170 * and the reallocating as data - this would cause a second revoke,
1171 * which hits an assertion error.
1173 JBUFFER_TRACE(jh
, "cancelling revoke");
1174 jbd2_journal_cancel_revoke(handle
, jh
);
1176 jbd2_journal_put_journal_head(jh
);
1181 * int jbd2_journal_get_undo_access() - Notify intent to modify metadata with
1182 * non-rewindable consequences
1183 * @handle: transaction
1184 * @bh: buffer to undo
1186 * Sometimes there is a need to distinguish between metadata which has
1187 * been committed to disk and that which has not. The ext3fs code uses
1188 * this for freeing and allocating space, we have to make sure that we
1189 * do not reuse freed space until the deallocation has been committed,
1190 * since if we overwrote that space we would make the delete
1191 * un-rewindable in case of a crash.
1193 * To deal with that, jbd2_journal_get_undo_access requests write access to a
1194 * buffer for parts of non-rewindable operations such as delete
1195 * operations on the bitmaps. The journaling code must keep a copy of
1196 * the buffer's contents prior to the undo_access call until such time
1197 * as we know that the buffer has definitely been committed to disk.
1199 * We never need to know which transaction the committed data is part
1200 * of, buffers touched here are guaranteed to be dirtied later and so
1201 * will be committed to a new transaction in due course, at which point
1202 * we can discard the old committed data pointer.
1204 * Returns error number or 0 on success.
1206 int jbd2_journal_get_undo_access(handle_t
*handle
, struct buffer_head
*bh
)
1209 struct journal_head
*jh
;
1210 char *committed_data
= NULL
;
1212 JBUFFER_TRACE(jh
, "entry");
1213 if (jbd2_write_access_granted(handle
, bh
))
1216 jh
= jbd2_journal_add_journal_head(bh
);
1218 * Do this first --- it can drop the journal lock, so we want to
1219 * make sure that obtaining the committed_data is done
1220 * atomically wrt. completion of any outstanding commits.
1222 err
= do_get_write_access(handle
, jh
, 1);
1227 if (!jh
->b_committed_data
) {
1228 committed_data
= jbd2_alloc(jh2bh(jh
)->b_size
, GFP_NOFS
);
1229 if (!committed_data
) {
1230 printk(KERN_ERR
"%s: No memory for committed data\n",
1237 jbd_lock_bh_state(bh
);
1238 if (!jh
->b_committed_data
) {
1239 /* Copy out the current buffer contents into the
1240 * preserved, committed copy. */
1241 JBUFFER_TRACE(jh
, "generate b_committed data");
1242 if (!committed_data
) {
1243 jbd_unlock_bh_state(bh
);
1247 jh
->b_committed_data
= committed_data
;
1248 committed_data
= NULL
;
1249 memcpy(jh
->b_committed_data
, bh
->b_data
, bh
->b_size
);
1251 jbd_unlock_bh_state(bh
);
1253 jbd2_journal_put_journal_head(jh
);
1254 if (unlikely(committed_data
))
1255 jbd2_free(committed_data
, bh
->b_size
);
1260 * void jbd2_journal_set_triggers() - Add triggers for commit writeout
1261 * @bh: buffer to trigger on
1262 * @type: struct jbd2_buffer_trigger_type containing the trigger(s).
1264 * Set any triggers on this journal_head. This is always safe, because
1265 * triggers for a committing buffer will be saved off, and triggers for
1266 * a running transaction will match the buffer in that transaction.
1268 * Call with NULL to clear the triggers.
1270 void jbd2_journal_set_triggers(struct buffer_head
*bh
,
1271 struct jbd2_buffer_trigger_type
*type
)
1273 struct journal_head
*jh
= jbd2_journal_grab_journal_head(bh
);
1277 jh
->b_triggers
= type
;
1278 jbd2_journal_put_journal_head(jh
);
1281 void jbd2_buffer_frozen_trigger(struct journal_head
*jh
, void *mapped_data
,
1282 struct jbd2_buffer_trigger_type
*triggers
)
1284 struct buffer_head
*bh
= jh2bh(jh
);
1286 if (!triggers
|| !triggers
->t_frozen
)
1289 triggers
->t_frozen(triggers
, bh
, mapped_data
, bh
->b_size
);
1292 void jbd2_buffer_abort_trigger(struct journal_head
*jh
,
1293 struct jbd2_buffer_trigger_type
*triggers
)
1295 if (!triggers
|| !triggers
->t_abort
)
1298 triggers
->t_abort(triggers
, jh2bh(jh
));
1302 * int jbd2_journal_dirty_metadata() - mark a buffer as containing dirty metadata
1303 * @handle: transaction to add buffer to.
1304 * @bh: buffer to mark
1306 * mark dirty metadata which needs to be journaled as part of the current
1309 * The buffer must have previously had jbd2_journal_get_write_access()
1310 * called so that it has a valid journal_head attached to the buffer
1313 * The buffer is placed on the transaction's metadata list and is marked
1314 * as belonging to the transaction.
1316 * Returns error number or 0 on success.
1318 * Special care needs to be taken if the buffer already belongs to the
1319 * current committing transaction (in which case we should have frozen
1320 * data present for that commit). In that case, we don't relink the
1321 * buffer: that only gets done when the old transaction finally
1322 * completes its commit.
1324 int jbd2_journal_dirty_metadata(handle_t
*handle
, struct buffer_head
*bh
)
1326 transaction_t
*transaction
= handle
->h_transaction
;
1328 struct journal_head
*jh
;
1331 if (is_handle_aborted(handle
))
1333 if (!buffer_jbd(bh
)) {
1338 * We don't grab jh reference here since the buffer must be part
1339 * of the running transaction.
1343 * This and the following assertions are unreliable since we may see jh
1344 * in inconsistent state unless we grab bh_state lock. But this is
1345 * crucial to catch bugs so let's do a reliable check until the
1346 * lockless handling is fully proven.
1348 if (jh
->b_transaction
!= transaction
&&
1349 jh
->b_next_transaction
!= transaction
) {
1350 jbd_lock_bh_state(bh
);
1351 J_ASSERT_JH(jh
, jh
->b_transaction
== transaction
||
1352 jh
->b_next_transaction
== transaction
);
1353 jbd_unlock_bh_state(bh
);
1355 if (jh
->b_modified
== 1) {
1356 /* If it's in our transaction it must be in BJ_Metadata list. */
1357 if (jh
->b_transaction
== transaction
&&
1358 jh
->b_jlist
!= BJ_Metadata
) {
1359 jbd_lock_bh_state(bh
);
1360 J_ASSERT_JH(jh
, jh
->b_transaction
!= transaction
||
1361 jh
->b_jlist
== BJ_Metadata
);
1362 jbd_unlock_bh_state(bh
);
1367 journal
= transaction
->t_journal
;
1368 jbd_debug(5, "journal_head %p\n", jh
);
1369 JBUFFER_TRACE(jh
, "entry");
1371 jbd_lock_bh_state(bh
);
1373 if (jh
->b_modified
== 0) {
1375 * This buffer's got modified and becoming part
1376 * of the transaction. This needs to be done
1377 * once a transaction -bzzz
1380 if (handle
->h_buffer_credits
<= 0) {
1384 handle
->h_buffer_credits
--;
1388 * fastpath, to avoid expensive locking. If this buffer is already
1389 * on the running transaction's metadata list there is nothing to do.
1390 * Nobody can take it off again because there is a handle open.
1391 * I _think_ we're OK here with SMP barriers - a mistaken decision will
1392 * result in this test being false, so we go in and take the locks.
1394 if (jh
->b_transaction
== transaction
&& jh
->b_jlist
== BJ_Metadata
) {
1395 JBUFFER_TRACE(jh
, "fastpath");
1396 if (unlikely(jh
->b_transaction
!=
1397 journal
->j_running_transaction
)) {
1398 printk(KERN_ERR
"JBD2: %s: "
1399 "jh->b_transaction (%llu, %p, %u) != "
1400 "journal->j_running_transaction (%p, %u)\n",
1402 (unsigned long long) bh
->b_blocknr
,
1404 jh
->b_transaction
? jh
->b_transaction
->t_tid
: 0,
1405 journal
->j_running_transaction
,
1406 journal
->j_running_transaction
?
1407 journal
->j_running_transaction
->t_tid
: 0);
1413 set_buffer_jbddirty(bh
);
1416 * Metadata already on the current transaction list doesn't
1417 * need to be filed. Metadata on another transaction's list must
1418 * be committing, and will be refiled once the commit completes:
1419 * leave it alone for now.
1421 if (jh
->b_transaction
!= transaction
) {
1422 JBUFFER_TRACE(jh
, "already on other transaction");
1423 if (unlikely(((jh
->b_transaction
!=
1424 journal
->j_committing_transaction
)) ||
1425 (jh
->b_next_transaction
!= transaction
))) {
1426 printk(KERN_ERR
"jbd2_journal_dirty_metadata: %s: "
1427 "bad jh for block %llu: "
1428 "transaction (%p, %u), "
1429 "jh->b_transaction (%p, %u), "
1430 "jh->b_next_transaction (%p, %u), jlist %u\n",
1432 (unsigned long long) bh
->b_blocknr
,
1433 transaction
, transaction
->t_tid
,
1436 jh
->b_transaction
->t_tid
: 0,
1437 jh
->b_next_transaction
,
1438 jh
->b_next_transaction
?
1439 jh
->b_next_transaction
->t_tid
: 0,
1444 /* And this case is illegal: we can't reuse another
1445 * transaction's data buffer, ever. */
1449 /* That test should have eliminated the following case: */
1450 J_ASSERT_JH(jh
, jh
->b_frozen_data
== NULL
);
1452 JBUFFER_TRACE(jh
, "file as BJ_Metadata");
1453 spin_lock(&journal
->j_list_lock
);
1454 __jbd2_journal_file_buffer(jh
, transaction
, BJ_Metadata
);
1455 spin_unlock(&journal
->j_list_lock
);
1457 jbd_unlock_bh_state(bh
);
1459 JBUFFER_TRACE(jh
, "exit");
1464 * void jbd2_journal_forget() - bforget() for potentially-journaled buffers.
1465 * @handle: transaction handle
1466 * @bh: bh to 'forget'
1468 * We can only do the bforget if there are no commits pending against the
1469 * buffer. If the buffer is dirty in the current running transaction we
1470 * can safely unlink it.
1472 * bh may not be a journalled buffer at all - it may be a non-JBD
1473 * buffer which came off the hashtable. Check for this.
1475 * Decrements bh->b_count by one.
1477 * Allow this call even if the handle has aborted --- it may be part of
1478 * the caller's cleanup after an abort.
1480 int jbd2_journal_forget (handle_t
*handle
, struct buffer_head
*bh
)
1482 transaction_t
*transaction
= handle
->h_transaction
;
1484 struct journal_head
*jh
;
1485 int drop_reserve
= 0;
1487 int was_modified
= 0;
1489 if (is_handle_aborted(handle
))
1491 journal
= transaction
->t_journal
;
1493 BUFFER_TRACE(bh
, "entry");
1495 jbd_lock_bh_state(bh
);
1497 if (!buffer_jbd(bh
))
1501 /* Critical error: attempting to delete a bitmap buffer, maybe?
1502 * Don't do any jbd operations, and return an error. */
1503 if (!J_EXPECT_JH(jh
, !jh
->b_committed_data
,
1504 "inconsistent data on disk")) {
1509 /* keep track of whether or not this transaction modified us */
1510 was_modified
= jh
->b_modified
;
1513 * The buffer's going from the transaction, we must drop
1514 * all references -bzzz
1518 if (jh
->b_transaction
== transaction
) {
1519 J_ASSERT_JH(jh
, !jh
->b_frozen_data
);
1521 /* If we are forgetting a buffer which is already part
1522 * of this transaction, then we can just drop it from
1523 * the transaction immediately. */
1524 clear_buffer_dirty(bh
);
1525 clear_buffer_jbddirty(bh
);
1527 JBUFFER_TRACE(jh
, "belongs to current transaction: unfile");
1530 * we only want to drop a reference if this transaction
1531 * modified the buffer
1537 * We are no longer going to journal this buffer.
1538 * However, the commit of this transaction is still
1539 * important to the buffer: the delete that we are now
1540 * processing might obsolete an old log entry, so by
1541 * committing, we can satisfy the buffer's checkpoint.
1543 * So, if we have a checkpoint on the buffer, we should
1544 * now refile the buffer on our BJ_Forget list so that
1545 * we know to remove the checkpoint after we commit.
1548 spin_lock(&journal
->j_list_lock
);
1549 if (jh
->b_cp_transaction
) {
1550 __jbd2_journal_temp_unlink_buffer(jh
);
1551 __jbd2_journal_file_buffer(jh
, transaction
, BJ_Forget
);
1553 __jbd2_journal_unfile_buffer(jh
);
1554 if (!buffer_jbd(bh
)) {
1555 spin_unlock(&journal
->j_list_lock
);
1556 jbd_unlock_bh_state(bh
);
1561 spin_unlock(&journal
->j_list_lock
);
1562 } else if (jh
->b_transaction
) {
1563 J_ASSERT_JH(jh
, (jh
->b_transaction
==
1564 journal
->j_committing_transaction
));
1565 /* However, if the buffer is still owned by a prior
1566 * (committing) transaction, we can't drop it yet... */
1567 JBUFFER_TRACE(jh
, "belongs to older transaction");
1568 /* ... but we CAN drop it from the new transaction if we
1569 * have also modified it since the original commit. */
1571 if (jh
->b_next_transaction
) {
1572 J_ASSERT(jh
->b_next_transaction
== transaction
);
1573 spin_lock(&journal
->j_list_lock
);
1574 jh
->b_next_transaction
= NULL
;
1575 spin_unlock(&journal
->j_list_lock
);
1578 * only drop a reference if this transaction modified
1587 jbd_unlock_bh_state(bh
);
1591 /* no need to reserve log space for this block -bzzz */
1592 handle
->h_buffer_credits
++;
1598 * int jbd2_journal_stop() - complete a transaction
1599 * @handle: tranaction to complete.
1601 * All done for a particular handle.
1603 * There is not much action needed here. We just return any remaining
1604 * buffer credits to the transaction and remove the handle. The only
1605 * complication is that we need to start a commit operation if the
1606 * filesystem is marked for synchronous update.
1608 * jbd2_journal_stop itself will not usually return an error, but it may
1609 * do so in unusual circumstances. In particular, expect it to
1610 * return -EIO if a jbd2_journal_abort has been executed since the
1611 * transaction began.
1613 int jbd2_journal_stop(handle_t
*handle
)
1615 transaction_t
*transaction
= handle
->h_transaction
;
1617 int err
= 0, wait_for_commit
= 0;
1623 * Handle is already detached from the transaction so
1624 * there is nothing to do other than decrease a refcount,
1625 * or free the handle if refcount drops to zero
1627 if (--handle
->h_ref
> 0) {
1628 jbd_debug(4, "h_ref %d -> %d\n", handle
->h_ref
+ 1,
1632 if (handle
->h_rsv_handle
)
1633 jbd2_free_handle(handle
->h_rsv_handle
);
1637 journal
= transaction
->t_journal
;
1639 J_ASSERT(journal_current_handle() == handle
);
1641 if (is_handle_aborted(handle
))
1644 J_ASSERT(atomic_read(&transaction
->t_updates
) > 0);
1646 if (--handle
->h_ref
> 0) {
1647 jbd_debug(4, "h_ref %d -> %d\n", handle
->h_ref
+ 1,
1652 jbd_debug(4, "Handle %p going down\n", handle
);
1653 trace_jbd2_handle_stats(journal
->j_fs_dev
->bd_dev
,
1655 handle
->h_type
, handle
->h_line_no
,
1656 jiffies
- handle
->h_start_jiffies
,
1657 handle
->h_sync
, handle
->h_requested_credits
,
1658 (handle
->h_requested_credits
-
1659 handle
->h_buffer_credits
));
1662 * Implement synchronous transaction batching. If the handle
1663 * was synchronous, don't force a commit immediately. Let's
1664 * yield and let another thread piggyback onto this
1665 * transaction. Keep doing that while new threads continue to
1666 * arrive. It doesn't cost much - we're about to run a commit
1667 * and sleep on IO anyway. Speeds up many-threaded, many-dir
1668 * operations by 30x or more...
1670 * We try and optimize the sleep time against what the
1671 * underlying disk can do, instead of having a static sleep
1672 * time. This is useful for the case where our storage is so
1673 * fast that it is more optimal to go ahead and force a flush
1674 * and wait for the transaction to be committed than it is to
1675 * wait for an arbitrary amount of time for new writers to
1676 * join the transaction. We achieve this by measuring how
1677 * long it takes to commit a transaction, and compare it with
1678 * how long this transaction has been running, and if run time
1679 * < commit time then we sleep for the delta and commit. This
1680 * greatly helps super fast disks that would see slowdowns as
1681 * more threads started doing fsyncs.
1683 * But don't do this if this process was the most recent one
1684 * to perform a synchronous write. We do this to detect the
1685 * case where a single process is doing a stream of sync
1686 * writes. No point in waiting for joiners in that case.
1688 * Setting max_batch_time to 0 disables this completely.
1691 if (handle
->h_sync
&& journal
->j_last_sync_writer
!= pid
&&
1692 journal
->j_max_batch_time
) {
1693 u64 commit_time
, trans_time
;
1695 journal
->j_last_sync_writer
= pid
;
1697 read_lock(&journal
->j_state_lock
);
1698 commit_time
= journal
->j_average_commit_time
;
1699 read_unlock(&journal
->j_state_lock
);
1701 trans_time
= ktime_to_ns(ktime_sub(ktime_get(),
1702 transaction
->t_start_time
));
1704 commit_time
= max_t(u64
, commit_time
,
1705 1000*journal
->j_min_batch_time
);
1706 commit_time
= min_t(u64
, commit_time
,
1707 1000*journal
->j_max_batch_time
);
1709 if (trans_time
< commit_time
) {
1710 ktime_t expires
= ktime_add_ns(ktime_get(),
1712 set_current_state(TASK_UNINTERRUPTIBLE
);
1713 schedule_hrtimeout(&expires
, HRTIMER_MODE_ABS
);
1718 transaction
->t_synchronous_commit
= 1;
1719 current
->journal_info
= NULL
;
1720 atomic_sub(handle
->h_buffer_credits
,
1721 &transaction
->t_outstanding_credits
);
1724 * If the handle is marked SYNC, we need to set another commit
1725 * going! We also want to force a commit if the current
1726 * transaction is occupying too much of the log, or if the
1727 * transaction is too old now.
1729 if (handle
->h_sync
||
1730 (atomic_read(&transaction
->t_outstanding_credits
) >
1731 journal
->j_max_transaction_buffers
) ||
1732 time_after_eq(jiffies
, transaction
->t_expires
)) {
1733 /* Do this even for aborted journals: an abort still
1734 * completes the commit thread, it just doesn't write
1735 * anything to disk. */
1737 jbd_debug(2, "transaction too old, requesting commit for "
1738 "handle %p\n", handle
);
1739 /* This is non-blocking */
1740 jbd2_log_start_commit(journal
, transaction
->t_tid
);
1743 * Special case: JBD2_SYNC synchronous updates require us
1744 * to wait for the commit to complete.
1746 if (handle
->h_sync
&& !(current
->flags
& PF_MEMALLOC
))
1747 wait_for_commit
= 1;
1751 * Once we drop t_updates, if it goes to zero the transaction
1752 * could start committing on us and eventually disappear. So
1753 * once we do this, we must not dereference transaction
1756 tid
= transaction
->t_tid
;
1757 if (atomic_dec_and_test(&transaction
->t_updates
)) {
1758 wake_up(&journal
->j_wait_updates
);
1759 if (journal
->j_barrier_count
)
1760 wake_up(&journal
->j_wait_transaction_locked
);
1763 if (wait_for_commit
)
1764 err
= jbd2_log_wait_commit(journal
, tid
);
1766 lock_map_release(&handle
->h_lockdep_map
);
1768 if (handle
->h_rsv_handle
)
1769 jbd2_journal_free_reserved(handle
->h_rsv_handle
);
1771 jbd2_free_handle(handle
);
1777 * List management code snippets: various functions for manipulating the
1778 * transaction buffer lists.
1783 * Append a buffer to a transaction list, given the transaction's list head
1786 * j_list_lock is held.
1788 * jbd_lock_bh_state(jh2bh(jh)) is held.
1792 __blist_add_buffer(struct journal_head
**list
, struct journal_head
*jh
)
1795 jh
->b_tnext
= jh
->b_tprev
= jh
;
1798 /* Insert at the tail of the list to preserve order */
1799 struct journal_head
*first
= *list
, *last
= first
->b_tprev
;
1801 jh
->b_tnext
= first
;
1802 last
->b_tnext
= first
->b_tprev
= jh
;
1807 * Remove a buffer from a transaction list, given the transaction's list
1810 * Called with j_list_lock held, and the journal may not be locked.
1812 * jbd_lock_bh_state(jh2bh(jh)) is held.
1816 __blist_del_buffer(struct journal_head
**list
, struct journal_head
*jh
)
1819 *list
= jh
->b_tnext
;
1823 jh
->b_tprev
->b_tnext
= jh
->b_tnext
;
1824 jh
->b_tnext
->b_tprev
= jh
->b_tprev
;
1828 * Remove a buffer from the appropriate transaction list.
1830 * Note that this function can *change* the value of
1831 * bh->b_transaction->t_buffers, t_forget, t_shadow_list, t_log_list or
1832 * t_reserved_list. If the caller is holding onto a copy of one of these
1833 * pointers, it could go bad. Generally the caller needs to re-read the
1834 * pointer from the transaction_t.
1836 * Called under j_list_lock.
1838 static void __jbd2_journal_temp_unlink_buffer(struct journal_head
*jh
)
1840 struct journal_head
**list
= NULL
;
1841 transaction_t
*transaction
;
1842 struct buffer_head
*bh
= jh2bh(jh
);
1844 J_ASSERT_JH(jh
, jbd_is_locked_bh_state(bh
));
1845 transaction
= jh
->b_transaction
;
1847 assert_spin_locked(&transaction
->t_journal
->j_list_lock
);
1849 J_ASSERT_JH(jh
, jh
->b_jlist
< BJ_Types
);
1850 if (jh
->b_jlist
!= BJ_None
)
1851 J_ASSERT_JH(jh
, transaction
!= NULL
);
1853 switch (jh
->b_jlist
) {
1857 transaction
->t_nr_buffers
--;
1858 J_ASSERT_JH(jh
, transaction
->t_nr_buffers
>= 0);
1859 list
= &transaction
->t_buffers
;
1862 list
= &transaction
->t_forget
;
1865 list
= &transaction
->t_shadow_list
;
1868 list
= &transaction
->t_reserved_list
;
1872 __blist_del_buffer(list
, jh
);
1873 jh
->b_jlist
= BJ_None
;
1874 if (test_clear_buffer_jbddirty(bh
))
1875 mark_buffer_dirty(bh
); /* Expose it to the VM */
1879 * Remove buffer from all transactions.
1881 * Called with bh_state lock and j_list_lock
1883 * jh and bh may be already freed when this function returns.
1885 static void __jbd2_journal_unfile_buffer(struct journal_head
*jh
)
1887 __jbd2_journal_temp_unlink_buffer(jh
);
1888 jh
->b_transaction
= NULL
;
1889 jbd2_journal_put_journal_head(jh
);
1892 void jbd2_journal_unfile_buffer(journal_t
*journal
, struct journal_head
*jh
)
1894 struct buffer_head
*bh
= jh2bh(jh
);
1896 /* Get reference so that buffer cannot be freed before we unlock it */
1898 jbd_lock_bh_state(bh
);
1899 spin_lock(&journal
->j_list_lock
);
1900 __jbd2_journal_unfile_buffer(jh
);
1901 spin_unlock(&journal
->j_list_lock
);
1902 jbd_unlock_bh_state(bh
);
1907 * Called from jbd2_journal_try_to_free_buffers().
1909 * Called under jbd_lock_bh_state(bh)
1912 __journal_try_to_free_buffer(journal_t
*journal
, struct buffer_head
*bh
)
1914 struct journal_head
*jh
;
1918 if (buffer_locked(bh
) || buffer_dirty(bh
))
1921 if (jh
->b_next_transaction
!= NULL
|| jh
->b_transaction
!= NULL
)
1924 spin_lock(&journal
->j_list_lock
);
1925 if (jh
->b_cp_transaction
!= NULL
) {
1926 /* written-back checkpointed metadata buffer */
1927 JBUFFER_TRACE(jh
, "remove from checkpoint list");
1928 __jbd2_journal_remove_checkpoint(jh
);
1930 spin_unlock(&journal
->j_list_lock
);
1936 * int jbd2_journal_try_to_free_buffers() - try to free page buffers.
1937 * @journal: journal for operation
1938 * @page: to try and free
1939 * @gfp_mask: we use the mask to detect how hard should we try to release
1940 * buffers. If __GFP_WAIT and __GFP_FS is set, we wait for commit code to
1941 * release the buffers.
1944 * For all the buffers on this page,
1945 * if they are fully written out ordered data, move them onto BUF_CLEAN
1946 * so try_to_free_buffers() can reap them.
1948 * This function returns non-zero if we wish try_to_free_buffers()
1949 * to be called. We do this if the page is releasable by try_to_free_buffers().
1950 * We also do it if the page has locked or dirty buffers and the caller wants
1951 * us to perform sync or async writeout.
1953 * This complicates JBD locking somewhat. We aren't protected by the
1954 * BKL here. We wish to remove the buffer from its committing or
1955 * running transaction's ->t_datalist via __jbd2_journal_unfile_buffer.
1957 * This may *change* the value of transaction_t->t_datalist, so anyone
1958 * who looks at t_datalist needs to lock against this function.
1960 * Even worse, someone may be doing a jbd2_journal_dirty_data on this
1961 * buffer. So we need to lock against that. jbd2_journal_dirty_data()
1962 * will come out of the lock with the buffer dirty, which makes it
1963 * ineligible for release here.
1965 * Who else is affected by this? hmm... Really the only contender
1966 * is do_get_write_access() - it could be looking at the buffer while
1967 * journal_try_to_free_buffer() is changing its state. But that
1968 * cannot happen because we never reallocate freed data as metadata
1969 * while the data is part of a transaction. Yes?
1971 * Return 0 on failure, 1 on success
1973 int jbd2_journal_try_to_free_buffers(journal_t
*journal
,
1974 struct page
*page
, gfp_t gfp_mask
)
1976 struct buffer_head
*head
;
1977 struct buffer_head
*bh
;
1980 J_ASSERT(PageLocked(page
));
1982 head
= page_buffers(page
);
1985 struct journal_head
*jh
;
1988 * We take our own ref against the journal_head here to avoid
1989 * having to add tons of locking around each instance of
1990 * jbd2_journal_put_journal_head().
1992 jh
= jbd2_journal_grab_journal_head(bh
);
1996 jbd_lock_bh_state(bh
);
1997 __journal_try_to_free_buffer(journal
, bh
);
1998 jbd2_journal_put_journal_head(jh
);
1999 jbd_unlock_bh_state(bh
);
2002 } while ((bh
= bh
->b_this_page
) != head
);
2004 ret
= try_to_free_buffers(page
);
2011 * This buffer is no longer needed. If it is on an older transaction's
2012 * checkpoint list we need to record it on this transaction's forget list
2013 * to pin this buffer (and hence its checkpointing transaction) down until
2014 * this transaction commits. If the buffer isn't on a checkpoint list, we
2016 * Returns non-zero if JBD no longer has an interest in the buffer.
2018 * Called under j_list_lock.
2020 * Called under jbd_lock_bh_state(bh).
2022 static int __dispose_buffer(struct journal_head
*jh
, transaction_t
*transaction
)
2025 struct buffer_head
*bh
= jh2bh(jh
);
2027 if (jh
->b_cp_transaction
) {
2028 JBUFFER_TRACE(jh
, "on running+cp transaction");
2029 __jbd2_journal_temp_unlink_buffer(jh
);
2031 * We don't want to write the buffer anymore, clear the
2032 * bit so that we don't confuse checks in
2033 * __journal_file_buffer
2035 clear_buffer_dirty(bh
);
2036 __jbd2_journal_file_buffer(jh
, transaction
, BJ_Forget
);
2039 JBUFFER_TRACE(jh
, "on running transaction");
2040 __jbd2_journal_unfile_buffer(jh
);
2046 * jbd2_journal_invalidatepage
2048 * This code is tricky. It has a number of cases to deal with.
2050 * There are two invariants which this code relies on:
2052 * i_size must be updated on disk before we start calling invalidatepage on the
2055 * This is done in ext3 by defining an ext3_setattr method which
2056 * updates i_size before truncate gets going. By maintaining this
2057 * invariant, we can be sure that it is safe to throw away any buffers
2058 * attached to the current transaction: once the transaction commits,
2059 * we know that the data will not be needed.
2061 * Note however that we can *not* throw away data belonging to the
2062 * previous, committing transaction!
2064 * Any disk blocks which *are* part of the previous, committing
2065 * transaction (and which therefore cannot be discarded immediately) are
2066 * not going to be reused in the new running transaction
2068 * The bitmap committed_data images guarantee this: any block which is
2069 * allocated in one transaction and removed in the next will be marked
2070 * as in-use in the committed_data bitmap, so cannot be reused until
2071 * the next transaction to delete the block commits. This means that
2072 * leaving committing buffers dirty is quite safe: the disk blocks
2073 * cannot be reallocated to a different file and so buffer aliasing is
2077 * The above applies mainly to ordered data mode. In writeback mode we
2078 * don't make guarantees about the order in which data hits disk --- in
2079 * particular we don't guarantee that new dirty data is flushed before
2080 * transaction commit --- so it is always safe just to discard data
2081 * immediately in that mode. --sct
2085 * The journal_unmap_buffer helper function returns zero if the buffer
2086 * concerned remains pinned as an anonymous buffer belonging to an older
2089 * We're outside-transaction here. Either or both of j_running_transaction
2090 * and j_committing_transaction may be NULL.
2092 static int journal_unmap_buffer(journal_t
*journal
, struct buffer_head
*bh
,
2095 transaction_t
*transaction
;
2096 struct journal_head
*jh
;
2099 BUFFER_TRACE(bh
, "entry");
2102 * It is safe to proceed here without the j_list_lock because the
2103 * buffers cannot be stolen by try_to_free_buffers as long as we are
2104 * holding the page lock. --sct
2107 if (!buffer_jbd(bh
))
2108 goto zap_buffer_unlocked
;
2110 /* OK, we have data buffer in journaled mode */
2111 write_lock(&journal
->j_state_lock
);
2112 jbd_lock_bh_state(bh
);
2113 spin_lock(&journal
->j_list_lock
);
2115 jh
= jbd2_journal_grab_journal_head(bh
);
2117 goto zap_buffer_no_jh
;
2120 * We cannot remove the buffer from checkpoint lists until the
2121 * transaction adding inode to orphan list (let's call it T)
2122 * is committed. Otherwise if the transaction changing the
2123 * buffer would be cleaned from the journal before T is
2124 * committed, a crash will cause that the correct contents of
2125 * the buffer will be lost. On the other hand we have to
2126 * clear the buffer dirty bit at latest at the moment when the
2127 * transaction marking the buffer as freed in the filesystem
2128 * structures is committed because from that moment on the
2129 * block can be reallocated and used by a different page.
2130 * Since the block hasn't been freed yet but the inode has
2131 * already been added to orphan list, it is safe for us to add
2132 * the buffer to BJ_Forget list of the newest transaction.
2134 * Also we have to clear buffer_mapped flag of a truncated buffer
2135 * because the buffer_head may be attached to the page straddling
2136 * i_size (can happen only when blocksize < pagesize) and thus the
2137 * buffer_head can be reused when the file is extended again. So we end
2138 * up keeping around invalidated buffers attached to transactions'
2139 * BJ_Forget list just to stop checkpointing code from cleaning up
2140 * the transaction this buffer was modified in.
2142 transaction
= jh
->b_transaction
;
2143 if (transaction
== NULL
) {
2144 /* First case: not on any transaction. If it
2145 * has no checkpoint link, then we can zap it:
2146 * it's a writeback-mode buffer so we don't care
2147 * if it hits disk safely. */
2148 if (!jh
->b_cp_transaction
) {
2149 JBUFFER_TRACE(jh
, "not on any transaction: zap");
2153 if (!buffer_dirty(bh
)) {
2154 /* bdflush has written it. We can drop it now */
2158 /* OK, it must be in the journal but still not
2159 * written fully to disk: it's metadata or
2160 * journaled data... */
2162 if (journal
->j_running_transaction
) {
2163 /* ... and once the current transaction has
2164 * committed, the buffer won't be needed any
2166 JBUFFER_TRACE(jh
, "checkpointed: add to BJ_Forget");
2167 may_free
= __dispose_buffer(jh
,
2168 journal
->j_running_transaction
);
2171 /* There is no currently-running transaction. So the
2172 * orphan record which we wrote for this file must have
2173 * passed into commit. We must attach this buffer to
2174 * the committing transaction, if it exists. */
2175 if (journal
->j_committing_transaction
) {
2176 JBUFFER_TRACE(jh
, "give to committing trans");
2177 may_free
= __dispose_buffer(jh
,
2178 journal
->j_committing_transaction
);
2181 /* The orphan record's transaction has
2182 * committed. We can cleanse this buffer */
2183 clear_buffer_jbddirty(bh
);
2187 } else if (transaction
== journal
->j_committing_transaction
) {
2188 JBUFFER_TRACE(jh
, "on committing transaction");
2190 * The buffer is committing, we simply cannot touch
2191 * it. If the page is straddling i_size we have to wait
2192 * for commit and try again.
2195 jbd2_journal_put_journal_head(jh
);
2196 spin_unlock(&journal
->j_list_lock
);
2197 jbd_unlock_bh_state(bh
);
2198 write_unlock(&journal
->j_state_lock
);
2202 * OK, buffer won't be reachable after truncate. We just set
2203 * j_next_transaction to the running transaction (if there is
2204 * one) and mark buffer as freed so that commit code knows it
2205 * should clear dirty bits when it is done with the buffer.
2207 set_buffer_freed(bh
);
2208 if (journal
->j_running_transaction
&& buffer_jbddirty(bh
))
2209 jh
->b_next_transaction
= journal
->j_running_transaction
;
2210 jbd2_journal_put_journal_head(jh
);
2211 spin_unlock(&journal
->j_list_lock
);
2212 jbd_unlock_bh_state(bh
);
2213 write_unlock(&journal
->j_state_lock
);
2216 /* Good, the buffer belongs to the running transaction.
2217 * We are writing our own transaction's data, not any
2218 * previous one's, so it is safe to throw it away
2219 * (remember that we expect the filesystem to have set
2220 * i_size already for this truncate so recovery will not
2221 * expose the disk blocks we are discarding here.) */
2222 J_ASSERT_JH(jh
, transaction
== journal
->j_running_transaction
);
2223 JBUFFER_TRACE(jh
, "on running transaction");
2224 may_free
= __dispose_buffer(jh
, transaction
);
2229 * This is tricky. Although the buffer is truncated, it may be reused
2230 * if blocksize < pagesize and it is attached to the page straddling
2231 * EOF. Since the buffer might have been added to BJ_Forget list of the
2232 * running transaction, journal_get_write_access() won't clear
2233 * b_modified and credit accounting gets confused. So clear b_modified
2237 jbd2_journal_put_journal_head(jh
);
2239 spin_unlock(&journal
->j_list_lock
);
2240 jbd_unlock_bh_state(bh
);
2241 write_unlock(&journal
->j_state_lock
);
2242 zap_buffer_unlocked
:
2243 clear_buffer_dirty(bh
);
2244 J_ASSERT_BH(bh
, !buffer_jbddirty(bh
));
2245 clear_buffer_mapped(bh
);
2246 clear_buffer_req(bh
);
2247 clear_buffer_new(bh
);
2248 clear_buffer_delay(bh
);
2249 clear_buffer_unwritten(bh
);
2255 * void jbd2_journal_invalidatepage()
2256 * @journal: journal to use for flush...
2257 * @page: page to flush
2258 * @offset: start of the range to invalidate
2259 * @length: length of the range to invalidate
2261 * Reap page buffers containing data after in the specified range in page.
2262 * Can return -EBUSY if buffers are part of the committing transaction and
2263 * the page is straddling i_size. Caller then has to wait for current commit
2266 int jbd2_journal_invalidatepage(journal_t
*journal
,
2268 unsigned int offset
,
2269 unsigned int length
)
2271 struct buffer_head
*head
, *bh
, *next
;
2272 unsigned int stop
= offset
+ length
;
2273 unsigned int curr_off
= 0;
2274 int partial_page
= (offset
|| length
< PAGE_CACHE_SIZE
);
2278 if (!PageLocked(page
))
2280 if (!page_has_buffers(page
))
2283 BUG_ON(stop
> PAGE_CACHE_SIZE
|| stop
< length
);
2285 /* We will potentially be playing with lists other than just the
2286 * data lists (especially for journaled data mode), so be
2287 * cautious in our locking. */
2289 head
= bh
= page_buffers(page
);
2291 unsigned int next_off
= curr_off
+ bh
->b_size
;
2292 next
= bh
->b_this_page
;
2294 if (next_off
> stop
)
2297 if (offset
<= curr_off
) {
2298 /* This block is wholly outside the truncation point */
2300 ret
= journal_unmap_buffer(journal
, bh
, partial_page
);
2306 curr_off
= next_off
;
2309 } while (bh
!= head
);
2311 if (!partial_page
) {
2312 if (may_free
&& try_to_free_buffers(page
))
2313 J_ASSERT(!page_has_buffers(page
));
2319 * File a buffer on the given transaction list.
2321 void __jbd2_journal_file_buffer(struct journal_head
*jh
,
2322 transaction_t
*transaction
, int jlist
)
2324 struct journal_head
**list
= NULL
;
2326 struct buffer_head
*bh
= jh2bh(jh
);
2328 J_ASSERT_JH(jh
, jbd_is_locked_bh_state(bh
));
2329 assert_spin_locked(&transaction
->t_journal
->j_list_lock
);
2331 J_ASSERT_JH(jh
, jh
->b_jlist
< BJ_Types
);
2332 J_ASSERT_JH(jh
, jh
->b_transaction
== transaction
||
2333 jh
->b_transaction
== NULL
);
2335 if (jh
->b_transaction
&& jh
->b_jlist
== jlist
)
2338 if (jlist
== BJ_Metadata
|| jlist
== BJ_Reserved
||
2339 jlist
== BJ_Shadow
|| jlist
== BJ_Forget
) {
2341 * For metadata buffers, we track dirty bit in buffer_jbddirty
2342 * instead of buffer_dirty. We should not see a dirty bit set
2343 * here because we clear it in do_get_write_access but e.g.
2344 * tune2fs can modify the sb and set the dirty bit at any time
2345 * so we try to gracefully handle that.
2347 if (buffer_dirty(bh
))
2348 warn_dirty_buffer(bh
);
2349 if (test_clear_buffer_dirty(bh
) ||
2350 test_clear_buffer_jbddirty(bh
))
2354 if (jh
->b_transaction
)
2355 __jbd2_journal_temp_unlink_buffer(jh
);
2357 jbd2_journal_grab_journal_head(bh
);
2358 jh
->b_transaction
= transaction
;
2362 J_ASSERT_JH(jh
, !jh
->b_committed_data
);
2363 J_ASSERT_JH(jh
, !jh
->b_frozen_data
);
2366 transaction
->t_nr_buffers
++;
2367 list
= &transaction
->t_buffers
;
2370 list
= &transaction
->t_forget
;
2373 list
= &transaction
->t_shadow_list
;
2376 list
= &transaction
->t_reserved_list
;
2380 __blist_add_buffer(list
, jh
);
2381 jh
->b_jlist
= jlist
;
2384 set_buffer_jbddirty(bh
);
2387 void jbd2_journal_file_buffer(struct journal_head
*jh
,
2388 transaction_t
*transaction
, int jlist
)
2390 jbd_lock_bh_state(jh2bh(jh
));
2391 spin_lock(&transaction
->t_journal
->j_list_lock
);
2392 __jbd2_journal_file_buffer(jh
, transaction
, jlist
);
2393 spin_unlock(&transaction
->t_journal
->j_list_lock
);
2394 jbd_unlock_bh_state(jh2bh(jh
));
2398 * Remove a buffer from its current buffer list in preparation for
2399 * dropping it from its current transaction entirely. If the buffer has
2400 * already started to be used by a subsequent transaction, refile the
2401 * buffer on that transaction's metadata list.
2403 * Called under j_list_lock
2404 * Called under jbd_lock_bh_state(jh2bh(jh))
2406 * jh and bh may be already free when this function returns
2408 void __jbd2_journal_refile_buffer(struct journal_head
*jh
)
2410 int was_dirty
, jlist
;
2411 struct buffer_head
*bh
= jh2bh(jh
);
2413 J_ASSERT_JH(jh
, jbd_is_locked_bh_state(bh
));
2414 if (jh
->b_transaction
)
2415 assert_spin_locked(&jh
->b_transaction
->t_journal
->j_list_lock
);
2417 /* If the buffer is now unused, just drop it. */
2418 if (jh
->b_next_transaction
== NULL
) {
2419 __jbd2_journal_unfile_buffer(jh
);
2424 * It has been modified by a later transaction: add it to the new
2425 * transaction's metadata list.
2428 was_dirty
= test_clear_buffer_jbddirty(bh
);
2429 __jbd2_journal_temp_unlink_buffer(jh
);
2431 * We set b_transaction here because b_next_transaction will inherit
2432 * our jh reference and thus __jbd2_journal_file_buffer() must not
2435 jh
->b_transaction
= jh
->b_next_transaction
;
2436 jh
->b_next_transaction
= NULL
;
2437 if (buffer_freed(bh
))
2439 else if (jh
->b_modified
)
2440 jlist
= BJ_Metadata
;
2442 jlist
= BJ_Reserved
;
2443 __jbd2_journal_file_buffer(jh
, jh
->b_transaction
, jlist
);
2444 J_ASSERT_JH(jh
, jh
->b_transaction
->t_state
== T_RUNNING
);
2447 set_buffer_jbddirty(bh
);
2451 * __jbd2_journal_refile_buffer() with necessary locking added. We take our
2452 * bh reference so that we can safely unlock bh.
2454 * The jh and bh may be freed by this call.
2456 void jbd2_journal_refile_buffer(journal_t
*journal
, struct journal_head
*jh
)
2458 struct buffer_head
*bh
= jh2bh(jh
);
2460 /* Get reference so that buffer cannot be freed before we unlock it */
2462 jbd_lock_bh_state(bh
);
2463 spin_lock(&journal
->j_list_lock
);
2464 __jbd2_journal_refile_buffer(jh
);
2465 jbd_unlock_bh_state(bh
);
2466 spin_unlock(&journal
->j_list_lock
);
2471 * File inode in the inode list of the handle's transaction
2473 int jbd2_journal_file_inode(handle_t
*handle
, struct jbd2_inode
*jinode
)
2475 transaction_t
*transaction
= handle
->h_transaction
;
2478 if (is_handle_aborted(handle
))
2480 journal
= transaction
->t_journal
;
2482 jbd_debug(4, "Adding inode %lu, tid:%d\n", jinode
->i_vfs_inode
->i_ino
,
2483 transaction
->t_tid
);
2486 * First check whether inode isn't already on the transaction's
2487 * lists without taking the lock. Note that this check is safe
2488 * without the lock as we cannot race with somebody removing inode
2489 * from the transaction. The reason is that we remove inode from the
2490 * transaction only in journal_release_jbd_inode() and when we commit
2491 * the transaction. We are guarded from the first case by holding
2492 * a reference to the inode. We are safe against the second case
2493 * because if jinode->i_transaction == transaction, commit code
2494 * cannot touch the transaction because we hold reference to it,
2495 * and if jinode->i_next_transaction == transaction, commit code
2496 * will only file the inode where we want it.
2498 if (jinode
->i_transaction
== transaction
||
2499 jinode
->i_next_transaction
== transaction
)
2502 spin_lock(&journal
->j_list_lock
);
2504 if (jinode
->i_transaction
== transaction
||
2505 jinode
->i_next_transaction
== transaction
)
2509 * We only ever set this variable to 1 so the test is safe. Since
2510 * t_need_data_flush is likely to be set, we do the test to save some
2511 * cacheline bouncing
2513 if (!transaction
->t_need_data_flush
)
2514 transaction
->t_need_data_flush
= 1;
2515 /* On some different transaction's list - should be
2516 * the committing one */
2517 if (jinode
->i_transaction
) {
2518 J_ASSERT(jinode
->i_next_transaction
== NULL
);
2519 J_ASSERT(jinode
->i_transaction
==
2520 journal
->j_committing_transaction
);
2521 jinode
->i_next_transaction
= transaction
;
2524 /* Not on any transaction list... */
2525 J_ASSERT(!jinode
->i_next_transaction
);
2526 jinode
->i_transaction
= transaction
;
2527 list_add(&jinode
->i_list
, &transaction
->t_inode_list
);
2529 spin_unlock(&journal
->j_list_lock
);
2535 * File truncate and transaction commit interact with each other in a
2536 * non-trivial way. If a transaction writing data block A is
2537 * committing, we cannot discard the data by truncate until we have
2538 * written them. Otherwise if we crashed after the transaction with
2539 * write has committed but before the transaction with truncate has
2540 * committed, we could see stale data in block A. This function is a
2541 * helper to solve this problem. It starts writeout of the truncated
2542 * part in case it is in the committing transaction.
2544 * Filesystem code must call this function when inode is journaled in
2545 * ordered mode before truncation happens and after the inode has been
2546 * placed on orphan list with the new inode size. The second condition
2547 * avoids the race that someone writes new data and we start
2548 * committing the transaction after this function has been called but
2549 * before a transaction for truncate is started (and furthermore it
2550 * allows us to optimize the case where the addition to orphan list
2551 * happens in the same transaction as write --- we don't have to write
2552 * any data in such case).
2554 int jbd2_journal_begin_ordered_truncate(journal_t
*journal
,
2555 struct jbd2_inode
*jinode
,
2558 transaction_t
*inode_trans
, *commit_trans
;
2561 /* This is a quick check to avoid locking if not necessary */
2562 if (!jinode
->i_transaction
)
2564 /* Locks are here just to force reading of recent values, it is
2565 * enough that the transaction was not committing before we started
2566 * a transaction adding the inode to orphan list */
2567 read_lock(&journal
->j_state_lock
);
2568 commit_trans
= journal
->j_committing_transaction
;
2569 read_unlock(&journal
->j_state_lock
);
2570 spin_lock(&journal
->j_list_lock
);
2571 inode_trans
= jinode
->i_transaction
;
2572 spin_unlock(&journal
->j_list_lock
);
2573 if (inode_trans
== commit_trans
) {
2574 ret
= filemap_fdatawrite_range(jinode
->i_vfs_inode
->i_mapping
,
2575 new_size
, LLONG_MAX
);
2577 jbd2_journal_abort(journal
, ret
);