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 handle
->h_journal
= journal
;
531 jbd2_journal_free_reserved(handle
);
534 handle
->h_type
= type
;
535 handle
->h_line_no
= line_no
;
538 EXPORT_SYMBOL(jbd2_journal_start_reserved
);
541 * int jbd2_journal_extend() - extend buffer credits.
542 * @handle: handle to 'extend'
543 * @nblocks: nr blocks to try to extend by.
545 * Some transactions, such as large extends and truncates, can be done
546 * atomically all at once or in several stages. The operation requests
547 * a credit for a number of buffer modications in advance, but can
548 * extend its credit if it needs more.
550 * jbd2_journal_extend tries to give the running handle more buffer credits.
551 * It does not guarantee that allocation - this is a best-effort only.
552 * The calling process MUST be able to deal cleanly with a failure to
555 * Return 0 on success, non-zero on failure.
557 * return code < 0 implies an error
558 * return code > 0 implies normal transaction-full status.
560 int jbd2_journal_extend(handle_t
*handle
, int nblocks
)
562 transaction_t
*transaction
= handle
->h_transaction
;
567 if (is_handle_aborted(handle
))
569 journal
= transaction
->t_journal
;
573 read_lock(&journal
->j_state_lock
);
575 /* Don't extend a locked-down transaction! */
576 if (transaction
->t_state
!= T_RUNNING
) {
577 jbd_debug(3, "denied handle %p %d blocks: "
578 "transaction not running\n", handle
, nblocks
);
582 spin_lock(&transaction
->t_handle_lock
);
583 wanted
= atomic_add_return(nblocks
,
584 &transaction
->t_outstanding_credits
);
586 if (wanted
> journal
->j_max_transaction_buffers
) {
587 jbd_debug(3, "denied handle %p %d blocks: "
588 "transaction too large\n", handle
, nblocks
);
589 atomic_sub(nblocks
, &transaction
->t_outstanding_credits
);
593 if (wanted
+ (wanted
>> JBD2_CONTROL_BLOCKS_SHIFT
) >
594 jbd2_log_space_left(journal
)) {
595 jbd_debug(3, "denied handle %p %d blocks: "
596 "insufficient log space\n", handle
, nblocks
);
597 atomic_sub(nblocks
, &transaction
->t_outstanding_credits
);
601 trace_jbd2_handle_extend(journal
->j_fs_dev
->bd_dev
,
603 handle
->h_type
, handle
->h_line_no
,
604 handle
->h_buffer_credits
,
607 handle
->h_buffer_credits
+= nblocks
;
608 handle
->h_requested_credits
+= nblocks
;
611 jbd_debug(3, "extended handle %p by %d\n", handle
, nblocks
);
613 spin_unlock(&transaction
->t_handle_lock
);
615 read_unlock(&journal
->j_state_lock
);
621 * int jbd2_journal_restart() - restart a handle .
622 * @handle: handle to restart
623 * @nblocks: nr credits requested
625 * Restart a handle for a multi-transaction filesystem
628 * If the jbd2_journal_extend() call above fails to grant new buffer credits
629 * to a running handle, a call to jbd2_journal_restart will commit the
630 * handle's transaction so far and reattach the handle to a new
631 * transaction capabable of guaranteeing the requested number of
632 * credits. We preserve reserved handle if there's any attached to the
635 int jbd2__journal_restart(handle_t
*handle
, int nblocks
, gfp_t gfp_mask
)
637 transaction_t
*transaction
= handle
->h_transaction
;
640 int need_to_start
, ret
;
642 /* If we've had an abort of any type, don't even think about
643 * actually doing the restart! */
644 if (is_handle_aborted(handle
))
646 journal
= transaction
->t_journal
;
649 * First unlink the handle from its current transaction, and start the
652 J_ASSERT(atomic_read(&transaction
->t_updates
) > 0);
653 J_ASSERT(journal_current_handle() == handle
);
655 read_lock(&journal
->j_state_lock
);
656 spin_lock(&transaction
->t_handle_lock
);
657 atomic_sub(handle
->h_buffer_credits
,
658 &transaction
->t_outstanding_credits
);
659 if (handle
->h_rsv_handle
) {
660 sub_reserved_credits(journal
,
661 handle
->h_rsv_handle
->h_buffer_credits
);
663 if (atomic_dec_and_test(&transaction
->t_updates
))
664 wake_up(&journal
->j_wait_updates
);
665 tid
= transaction
->t_tid
;
666 spin_unlock(&transaction
->t_handle_lock
);
667 handle
->h_transaction
= NULL
;
668 current
->journal_info
= NULL
;
670 jbd_debug(2, "restarting handle %p\n", handle
);
671 need_to_start
= !tid_geq(journal
->j_commit_request
, tid
);
672 read_unlock(&journal
->j_state_lock
);
674 jbd2_log_start_commit(journal
, tid
);
676 lock_map_release(&handle
->h_lockdep_map
);
677 handle
->h_buffer_credits
= nblocks
;
678 ret
= start_this_handle(journal
, handle
, gfp_mask
);
681 EXPORT_SYMBOL(jbd2__journal_restart
);
684 int jbd2_journal_restart(handle_t
*handle
, int nblocks
)
686 return jbd2__journal_restart(handle
, nblocks
, GFP_NOFS
);
688 EXPORT_SYMBOL(jbd2_journal_restart
);
691 * void jbd2_journal_lock_updates () - establish a transaction barrier.
692 * @journal: Journal to establish a barrier on.
694 * This locks out any further updates from being started, and blocks
695 * until all existing updates have completed, returning only once the
696 * journal is in a quiescent state with no updates running.
698 * The journal lock should not be held on entry.
700 void jbd2_journal_lock_updates(journal_t
*journal
)
704 write_lock(&journal
->j_state_lock
);
705 ++journal
->j_barrier_count
;
707 /* Wait until there are no reserved handles */
708 if (atomic_read(&journal
->j_reserved_credits
)) {
709 write_unlock(&journal
->j_state_lock
);
710 wait_event(journal
->j_wait_reserved
,
711 atomic_read(&journal
->j_reserved_credits
) == 0);
712 write_lock(&journal
->j_state_lock
);
715 /* Wait until there are no running updates */
717 transaction_t
*transaction
= journal
->j_running_transaction
;
722 spin_lock(&transaction
->t_handle_lock
);
723 prepare_to_wait(&journal
->j_wait_updates
, &wait
,
724 TASK_UNINTERRUPTIBLE
);
725 if (!atomic_read(&transaction
->t_updates
)) {
726 spin_unlock(&transaction
->t_handle_lock
);
727 finish_wait(&journal
->j_wait_updates
, &wait
);
730 spin_unlock(&transaction
->t_handle_lock
);
731 write_unlock(&journal
->j_state_lock
);
733 finish_wait(&journal
->j_wait_updates
, &wait
);
734 write_lock(&journal
->j_state_lock
);
736 write_unlock(&journal
->j_state_lock
);
739 * We have now established a barrier against other normal updates, but
740 * we also need to barrier against other jbd2_journal_lock_updates() calls
741 * to make sure that we serialise special journal-locked operations
744 mutex_lock(&journal
->j_barrier
);
748 * void jbd2_journal_unlock_updates (journal_t* journal) - release barrier
749 * @journal: Journal to release the barrier on.
751 * Release a transaction barrier obtained with jbd2_journal_lock_updates().
753 * Should be called without the journal lock held.
755 void jbd2_journal_unlock_updates (journal_t
*journal
)
757 J_ASSERT(journal
->j_barrier_count
!= 0);
759 mutex_unlock(&journal
->j_barrier
);
760 write_lock(&journal
->j_state_lock
);
761 --journal
->j_barrier_count
;
762 write_unlock(&journal
->j_state_lock
);
763 wake_up(&journal
->j_wait_transaction_locked
);
766 static void warn_dirty_buffer(struct buffer_head
*bh
)
768 char b
[BDEVNAME_SIZE
];
771 "JBD2: Spotted dirty metadata buffer (dev = %s, blocknr = %llu). "
772 "There's a risk of filesystem corruption in case of system "
774 bdevname(bh
->b_bdev
, b
), (unsigned long long)bh
->b_blocknr
);
777 /* Call t_frozen trigger and copy buffer data into jh->b_frozen_data. */
778 static void jbd2_freeze_jh_data(struct journal_head
*jh
)
783 struct buffer_head
*bh
= jh2bh(jh
);
785 J_EXPECT_JH(jh
, buffer_uptodate(bh
), "Possible IO failure.\n");
787 offset
= offset_in_page(bh
->b_data
);
788 source
= kmap_atomic(page
);
789 /* Fire data frozen trigger just before we copy the data */
790 jbd2_buffer_frozen_trigger(jh
, source
+ offset
, jh
->b_triggers
);
791 memcpy(jh
->b_frozen_data
, source
+ offset
, bh
->b_size
);
792 kunmap_atomic(source
);
795 * Now that the frozen data is saved off, we need to store any matching
798 jh
->b_frozen_triggers
= jh
->b_triggers
;
802 * If the buffer is already part of the current transaction, then there
803 * is nothing we need to do. If it is already part of a prior
804 * transaction which we are still committing to disk, then we need to
805 * make sure that we do not overwrite the old copy: we do copy-out to
806 * preserve the copy going to disk. We also account the buffer against
807 * the handle's metadata buffer credits (unless the buffer is already
808 * part of the transaction, that is).
812 do_get_write_access(handle_t
*handle
, struct journal_head
*jh
,
815 struct buffer_head
*bh
;
816 transaction_t
*transaction
= handle
->h_transaction
;
819 char *frozen_buffer
= NULL
;
820 unsigned long start_lock
, time_lock
;
822 if (is_handle_aborted(handle
))
824 journal
= transaction
->t_journal
;
826 jbd_debug(5, "journal_head %p, force_copy %d\n", jh
, force_copy
);
828 JBUFFER_TRACE(jh
, "entry");
832 /* @@@ Need to check for errors here at some point. */
834 start_lock
= jiffies
;
836 jbd_lock_bh_state(bh
);
838 /* If it takes too long to lock the buffer, trace it */
839 time_lock
= jbd2_time_diff(start_lock
, jiffies
);
840 if (time_lock
> HZ
/10)
841 trace_jbd2_lock_buffer_stall(bh
->b_bdev
->bd_dev
,
842 jiffies_to_msecs(time_lock
));
844 /* We now hold the buffer lock so it is safe to query the buffer
845 * state. Is the buffer dirty?
847 * If so, there are two possibilities. The buffer may be
848 * non-journaled, and undergoing a quite legitimate writeback.
849 * Otherwise, it is journaled, and we don't expect dirty buffers
850 * in that state (the buffers should be marked JBD_Dirty
851 * instead.) So either the IO is being done under our own
852 * control and this is a bug, or it's a third party IO such as
853 * dump(8) (which may leave the buffer scheduled for read ---
854 * ie. locked but not dirty) or tune2fs (which may actually have
855 * the buffer dirtied, ugh.) */
857 if (buffer_dirty(bh
)) {
859 * First question: is this buffer already part of the current
860 * transaction or the existing committing transaction?
862 if (jh
->b_transaction
) {
864 jh
->b_transaction
== transaction
||
866 journal
->j_committing_transaction
);
867 if (jh
->b_next_transaction
)
868 J_ASSERT_JH(jh
, jh
->b_next_transaction
==
870 warn_dirty_buffer(bh
);
873 * In any case we need to clean the dirty flag and we must
874 * do it under the buffer lock to be sure we don't race
875 * with running write-out.
877 JBUFFER_TRACE(jh
, "Journalling dirty buffer");
878 clear_buffer_dirty(bh
);
879 set_buffer_jbddirty(bh
);
885 if (is_handle_aborted(handle
)) {
886 jbd_unlock_bh_state(bh
);
892 * The buffer is already part of this transaction if b_transaction or
893 * b_next_transaction points to it
895 if (jh
->b_transaction
== transaction
||
896 jh
->b_next_transaction
== transaction
)
900 * this is the first time this transaction is touching this buffer,
901 * reset the modified flag
906 * If the buffer is not journaled right now, we need to make sure it
907 * doesn't get written to disk before the caller actually commits the
910 if (!jh
->b_transaction
) {
911 JBUFFER_TRACE(jh
, "no transaction");
912 J_ASSERT_JH(jh
, !jh
->b_next_transaction
);
913 JBUFFER_TRACE(jh
, "file as BJ_Reserved");
915 * Make sure all stores to jh (b_modified, b_frozen_data) are
916 * visible before attaching it to the running transaction.
917 * Paired with barrier in jbd2_write_access_granted()
920 spin_lock(&journal
->j_list_lock
);
921 __jbd2_journal_file_buffer(jh
, transaction
, BJ_Reserved
);
922 spin_unlock(&journal
->j_list_lock
);
926 * If there is already a copy-out version of this buffer, then we don't
927 * need to make another one
929 if (jh
->b_frozen_data
) {
930 JBUFFER_TRACE(jh
, "has frozen data");
931 J_ASSERT_JH(jh
, jh
->b_next_transaction
== NULL
);
935 JBUFFER_TRACE(jh
, "owned by older transaction");
936 J_ASSERT_JH(jh
, jh
->b_next_transaction
== NULL
);
937 J_ASSERT_JH(jh
, jh
->b_transaction
== journal
->j_committing_transaction
);
940 * There is one case we have to be very careful about. If the
941 * committing transaction is currently writing this buffer out to disk
942 * and has NOT made a copy-out, then we cannot modify the buffer
943 * contents at all right now. The essence of copy-out is that it is
944 * the extra copy, not the primary copy, which gets journaled. If the
945 * primary copy is already going to disk then we cannot do copy-out
948 if (buffer_shadow(bh
)) {
949 JBUFFER_TRACE(jh
, "on shadow: sleep");
950 jbd_unlock_bh_state(bh
);
951 wait_on_bit_io(&bh
->b_state
, BH_Shadow
, TASK_UNINTERRUPTIBLE
);
956 * Only do the copy if the currently-owning transaction still needs it.
957 * If buffer isn't on BJ_Metadata list, the committing transaction is
958 * past that stage (here we use the fact that BH_Shadow is set under
959 * bh_state lock together with refiling to BJ_Shadow list and at this
960 * point we know the buffer doesn't have BH_Shadow set).
962 * Subtle point, though: if this is a get_undo_access, then we will be
963 * relying on the frozen_data to contain the new value of the
964 * committed_data record after the transaction, so we HAVE to force the
965 * frozen_data copy in that case.
967 if (jh
->b_jlist
== BJ_Metadata
|| force_copy
) {
968 JBUFFER_TRACE(jh
, "generate frozen data");
969 if (!frozen_buffer
) {
970 JBUFFER_TRACE(jh
, "allocate memory for buffer");
971 jbd_unlock_bh_state(bh
);
972 frozen_buffer
= jbd2_alloc(jh2bh(jh
)->b_size
, GFP_NOFS
);
973 if (!frozen_buffer
) {
974 printk(KERN_ERR
"%s: OOM for frozen_buffer\n",
976 JBUFFER_TRACE(jh
, "oom!");
982 jh
->b_frozen_data
= frozen_buffer
;
983 frozen_buffer
= NULL
;
984 jbd2_freeze_jh_data(jh
);
988 * Make sure all stores to jh (b_modified, b_frozen_data) are visible
989 * before attaching it to the running transaction. Paired with barrier
990 * in jbd2_write_access_granted()
993 jh
->b_next_transaction
= transaction
;
996 jbd_unlock_bh_state(bh
);
999 * If we are about to journal a buffer, then any revoke pending on it is
1002 jbd2_journal_cancel_revoke(handle
, jh
);
1005 if (unlikely(frozen_buffer
)) /* It's usually NULL */
1006 jbd2_free(frozen_buffer
, bh
->b_size
);
1008 JBUFFER_TRACE(jh
, "exit");
1012 /* Fast check whether buffer is already attached to the required transaction */
1013 static bool jbd2_write_access_granted(handle_t
*handle
, struct buffer_head
*bh
,
1016 struct journal_head
*jh
;
1019 /* Dirty buffers require special handling... */
1020 if (buffer_dirty(bh
))
1024 * RCU protects us from dereferencing freed pages. So the checks we do
1025 * are guaranteed not to oops. However the jh slab object can get freed
1026 * & reallocated while we work with it. So we have to be careful. When
1027 * we see jh attached to the running transaction, we know it must stay
1028 * so until the transaction is committed. Thus jh won't be freed and
1029 * will be attached to the same bh while we run. However it can
1030 * happen jh gets freed, reallocated, and attached to the transaction
1031 * just after we get pointer to it from bh. So we have to be careful
1032 * and recheck jh still belongs to our bh before we return success.
1035 if (!buffer_jbd(bh
))
1037 /* This should be bh2jh() but that doesn't work with inline functions */
1038 jh
= READ_ONCE(bh
->b_private
);
1041 /* For undo access buffer must have data copied */
1042 if (undo
&& !jh
->b_committed_data
)
1044 if (READ_ONCE(jh
->b_transaction
) != handle
->h_transaction
&&
1045 READ_ONCE(jh
->b_next_transaction
) != handle
->h_transaction
)
1048 * There are two reasons for the barrier here:
1049 * 1) Make sure to fetch b_bh after we did previous checks so that we
1050 * detect when jh went through free, realloc, attach to transaction
1051 * while we were checking. Paired with implicit barrier in that path.
1052 * 2) So that access to bh done after jbd2_write_access_granted()
1053 * doesn't get reordered and see inconsistent state of concurrent
1054 * do_get_write_access().
1057 if (unlikely(jh
->b_bh
!= bh
))
1066 * int jbd2_journal_get_write_access() - notify intent to modify a buffer for metadata (not data) update.
1067 * @handle: transaction to add buffer modifications to
1068 * @bh: bh to be used for metadata writes
1070 * Returns an error code or 0 on success.
1072 * In full data journalling mode the buffer may be of type BJ_AsyncData,
1073 * because we're write()ing a buffer which is also part of a shared mapping.
1076 int jbd2_journal_get_write_access(handle_t
*handle
, struct buffer_head
*bh
)
1078 struct journal_head
*jh
;
1081 if (jbd2_write_access_granted(handle
, bh
, false))
1084 jh
= jbd2_journal_add_journal_head(bh
);
1085 /* We do not want to get caught playing with fields which the
1086 * log thread also manipulates. Make sure that the buffer
1087 * completes any outstanding IO before proceeding. */
1088 rc
= do_get_write_access(handle
, jh
, 0);
1089 jbd2_journal_put_journal_head(jh
);
1095 * When the user wants to journal a newly created buffer_head
1096 * (ie. getblk() returned a new buffer and we are going to populate it
1097 * manually rather than reading off disk), then we need to keep the
1098 * buffer_head locked until it has been completely filled with new
1099 * data. In this case, we should be able to make the assertion that
1100 * the bh is not already part of an existing transaction.
1102 * The buffer should already be locked by the caller by this point.
1103 * There is no lock ranking violation: it was a newly created,
1104 * unlocked buffer beforehand. */
1107 * int jbd2_journal_get_create_access () - notify intent to use newly created bh
1108 * @handle: transaction to new buffer to
1111 * Call this if you create a new bh.
1113 int jbd2_journal_get_create_access(handle_t
*handle
, struct buffer_head
*bh
)
1115 transaction_t
*transaction
= handle
->h_transaction
;
1117 struct journal_head
*jh
= jbd2_journal_add_journal_head(bh
);
1120 jbd_debug(5, "journal_head %p\n", jh
);
1122 if (is_handle_aborted(handle
))
1124 journal
= transaction
->t_journal
;
1127 JBUFFER_TRACE(jh
, "entry");
1129 * The buffer may already belong to this transaction due to pre-zeroing
1130 * in the filesystem's new_block code. It may also be on the previous,
1131 * committing transaction's lists, but it HAS to be in Forget state in
1132 * that case: the transaction must have deleted the buffer for it to be
1135 jbd_lock_bh_state(bh
);
1136 J_ASSERT_JH(jh
, (jh
->b_transaction
== transaction
||
1137 jh
->b_transaction
== NULL
||
1138 (jh
->b_transaction
== journal
->j_committing_transaction
&&
1139 jh
->b_jlist
== BJ_Forget
)));
1141 J_ASSERT_JH(jh
, jh
->b_next_transaction
== NULL
);
1142 J_ASSERT_JH(jh
, buffer_locked(jh2bh(jh
)));
1144 if (jh
->b_transaction
== NULL
) {
1146 * Previous jbd2_journal_forget() could have left the buffer
1147 * with jbddirty bit set because it was being committed. When
1148 * the commit finished, we've filed the buffer for
1149 * checkpointing and marked it dirty. Now we are reallocating
1150 * the buffer so the transaction freeing it must have
1151 * committed and so it's safe to clear the dirty bit.
1153 clear_buffer_dirty(jh2bh(jh
));
1154 /* first access by this transaction */
1157 JBUFFER_TRACE(jh
, "file as BJ_Reserved");
1158 spin_lock(&journal
->j_list_lock
);
1159 __jbd2_journal_file_buffer(jh
, transaction
, BJ_Reserved
);
1160 spin_unlock(&journal
->j_list_lock
);
1161 } else if (jh
->b_transaction
== journal
->j_committing_transaction
) {
1162 /* first access by this transaction */
1165 JBUFFER_TRACE(jh
, "set next transaction");
1166 spin_lock(&journal
->j_list_lock
);
1167 jh
->b_next_transaction
= transaction
;
1168 spin_unlock(&journal
->j_list_lock
);
1170 jbd_unlock_bh_state(bh
);
1173 * akpm: I added this. ext3_alloc_branch can pick up new indirect
1174 * blocks which contain freed but then revoked metadata. We need
1175 * to cancel the revoke in case we end up freeing it yet again
1176 * and the reallocating as data - this would cause a second revoke,
1177 * which hits an assertion error.
1179 JBUFFER_TRACE(jh
, "cancelling revoke");
1180 jbd2_journal_cancel_revoke(handle
, jh
);
1182 jbd2_journal_put_journal_head(jh
);
1187 * int jbd2_journal_get_undo_access() - Notify intent to modify metadata with
1188 * non-rewindable consequences
1189 * @handle: transaction
1190 * @bh: buffer to undo
1192 * Sometimes there is a need to distinguish between metadata which has
1193 * been committed to disk and that which has not. The ext3fs code uses
1194 * this for freeing and allocating space, we have to make sure that we
1195 * do not reuse freed space until the deallocation has been committed,
1196 * since if we overwrote that space we would make the delete
1197 * un-rewindable in case of a crash.
1199 * To deal with that, jbd2_journal_get_undo_access requests write access to a
1200 * buffer for parts of non-rewindable operations such as delete
1201 * operations on the bitmaps. The journaling code must keep a copy of
1202 * the buffer's contents prior to the undo_access call until such time
1203 * as we know that the buffer has definitely been committed to disk.
1205 * We never need to know which transaction the committed data is part
1206 * of, buffers touched here are guaranteed to be dirtied later and so
1207 * will be committed to a new transaction in due course, at which point
1208 * we can discard the old committed data pointer.
1210 * Returns error number or 0 on success.
1212 int jbd2_journal_get_undo_access(handle_t
*handle
, struct buffer_head
*bh
)
1215 struct journal_head
*jh
;
1216 char *committed_data
= NULL
;
1218 if (jbd2_write_access_granted(handle
, bh
, true))
1221 jh
= jbd2_journal_add_journal_head(bh
);
1222 JBUFFER_TRACE(jh
, "entry");
1225 * Do this first --- it can drop the journal lock, so we want to
1226 * make sure that obtaining the committed_data is done
1227 * atomically wrt. completion of any outstanding commits.
1229 err
= do_get_write_access(handle
, jh
, 1);
1234 if (!jh
->b_committed_data
) {
1235 committed_data
= jbd2_alloc(jh2bh(jh
)->b_size
, GFP_NOFS
);
1236 if (!committed_data
) {
1237 printk(KERN_ERR
"%s: No memory for committed data\n",
1244 jbd_lock_bh_state(bh
);
1245 if (!jh
->b_committed_data
) {
1246 /* Copy out the current buffer contents into the
1247 * preserved, committed copy. */
1248 JBUFFER_TRACE(jh
, "generate b_committed data");
1249 if (!committed_data
) {
1250 jbd_unlock_bh_state(bh
);
1254 jh
->b_committed_data
= committed_data
;
1255 committed_data
= NULL
;
1256 memcpy(jh
->b_committed_data
, bh
->b_data
, bh
->b_size
);
1258 jbd_unlock_bh_state(bh
);
1260 jbd2_journal_put_journal_head(jh
);
1261 if (unlikely(committed_data
))
1262 jbd2_free(committed_data
, bh
->b_size
);
1267 * void jbd2_journal_set_triggers() - Add triggers for commit writeout
1268 * @bh: buffer to trigger on
1269 * @type: struct jbd2_buffer_trigger_type containing the trigger(s).
1271 * Set any triggers on this journal_head. This is always safe, because
1272 * triggers for a committing buffer will be saved off, and triggers for
1273 * a running transaction will match the buffer in that transaction.
1275 * Call with NULL to clear the triggers.
1277 void jbd2_journal_set_triggers(struct buffer_head
*bh
,
1278 struct jbd2_buffer_trigger_type
*type
)
1280 struct journal_head
*jh
= jbd2_journal_grab_journal_head(bh
);
1284 jh
->b_triggers
= type
;
1285 jbd2_journal_put_journal_head(jh
);
1288 void jbd2_buffer_frozen_trigger(struct journal_head
*jh
, void *mapped_data
,
1289 struct jbd2_buffer_trigger_type
*triggers
)
1291 struct buffer_head
*bh
= jh2bh(jh
);
1293 if (!triggers
|| !triggers
->t_frozen
)
1296 triggers
->t_frozen(triggers
, bh
, mapped_data
, bh
->b_size
);
1299 void jbd2_buffer_abort_trigger(struct journal_head
*jh
,
1300 struct jbd2_buffer_trigger_type
*triggers
)
1302 if (!triggers
|| !triggers
->t_abort
)
1305 triggers
->t_abort(triggers
, jh2bh(jh
));
1309 * int jbd2_journal_dirty_metadata() - mark a buffer as containing dirty metadata
1310 * @handle: transaction to add buffer to.
1311 * @bh: buffer to mark
1313 * mark dirty metadata which needs to be journaled as part of the current
1316 * The buffer must have previously had jbd2_journal_get_write_access()
1317 * called so that it has a valid journal_head attached to the buffer
1320 * The buffer is placed on the transaction's metadata list and is marked
1321 * as belonging to the transaction.
1323 * Returns error number or 0 on success.
1325 * Special care needs to be taken if the buffer already belongs to the
1326 * current committing transaction (in which case we should have frozen
1327 * data present for that commit). In that case, we don't relink the
1328 * buffer: that only gets done when the old transaction finally
1329 * completes its commit.
1331 int jbd2_journal_dirty_metadata(handle_t
*handle
, struct buffer_head
*bh
)
1333 transaction_t
*transaction
= handle
->h_transaction
;
1335 struct journal_head
*jh
;
1338 if (is_handle_aborted(handle
))
1340 if (!buffer_jbd(bh
))
1344 * We don't grab jh reference here since the buffer must be part
1345 * of the running transaction.
1348 jbd_debug(5, "journal_head %p\n", jh
);
1349 JBUFFER_TRACE(jh
, "entry");
1352 * This and the following assertions are unreliable since we may see jh
1353 * in inconsistent state unless we grab bh_state lock. But this is
1354 * crucial to catch bugs so let's do a reliable check until the
1355 * lockless handling is fully proven.
1357 if (jh
->b_transaction
!= transaction
&&
1358 jh
->b_next_transaction
!= transaction
) {
1359 jbd_lock_bh_state(bh
);
1360 J_ASSERT_JH(jh
, jh
->b_transaction
== transaction
||
1361 jh
->b_next_transaction
== transaction
);
1362 jbd_unlock_bh_state(bh
);
1364 if (jh
->b_modified
== 1) {
1365 /* If it's in our transaction it must be in BJ_Metadata list. */
1366 if (jh
->b_transaction
== transaction
&&
1367 jh
->b_jlist
!= BJ_Metadata
) {
1368 jbd_lock_bh_state(bh
);
1369 if (jh
->b_transaction
== transaction
&&
1370 jh
->b_jlist
!= BJ_Metadata
)
1371 pr_err("JBD2: assertion failure: h_type=%u "
1372 "h_line_no=%u block_no=%llu jlist=%u\n",
1373 handle
->h_type
, handle
->h_line_no
,
1374 (unsigned long long) bh
->b_blocknr
,
1376 J_ASSERT_JH(jh
, jh
->b_transaction
!= transaction
||
1377 jh
->b_jlist
== BJ_Metadata
);
1378 jbd_unlock_bh_state(bh
);
1383 journal
= transaction
->t_journal
;
1384 jbd_lock_bh_state(bh
);
1386 if (jh
->b_modified
== 0) {
1388 * This buffer's got modified and becoming part
1389 * of the transaction. This needs to be done
1390 * once a transaction -bzzz
1392 if (handle
->h_buffer_credits
<= 0) {
1397 handle
->h_buffer_credits
--;
1401 * fastpath, to avoid expensive locking. If this buffer is already
1402 * on the running transaction's metadata list there is nothing to do.
1403 * Nobody can take it off again because there is a handle open.
1404 * I _think_ we're OK here with SMP barriers - a mistaken decision will
1405 * result in this test being false, so we go in and take the locks.
1407 if (jh
->b_transaction
== transaction
&& jh
->b_jlist
== BJ_Metadata
) {
1408 JBUFFER_TRACE(jh
, "fastpath");
1409 if (unlikely(jh
->b_transaction
!=
1410 journal
->j_running_transaction
)) {
1411 printk(KERN_ERR
"JBD2: %s: "
1412 "jh->b_transaction (%llu, %p, %u) != "
1413 "journal->j_running_transaction (%p, %u)\n",
1415 (unsigned long long) bh
->b_blocknr
,
1417 jh
->b_transaction
? jh
->b_transaction
->t_tid
: 0,
1418 journal
->j_running_transaction
,
1419 journal
->j_running_transaction
?
1420 journal
->j_running_transaction
->t_tid
: 0);
1426 set_buffer_jbddirty(bh
);
1429 * Metadata already on the current transaction list doesn't
1430 * need to be filed. Metadata on another transaction's list must
1431 * be committing, and will be refiled once the commit completes:
1432 * leave it alone for now.
1434 if (jh
->b_transaction
!= transaction
) {
1435 JBUFFER_TRACE(jh
, "already on other transaction");
1436 if (unlikely(((jh
->b_transaction
!=
1437 journal
->j_committing_transaction
)) ||
1438 (jh
->b_next_transaction
!= transaction
))) {
1439 printk(KERN_ERR
"jbd2_journal_dirty_metadata: %s: "
1440 "bad jh for block %llu: "
1441 "transaction (%p, %u), "
1442 "jh->b_transaction (%p, %u), "
1443 "jh->b_next_transaction (%p, %u), jlist %u\n",
1445 (unsigned long long) bh
->b_blocknr
,
1446 transaction
, transaction
->t_tid
,
1449 jh
->b_transaction
->t_tid
: 0,
1450 jh
->b_next_transaction
,
1451 jh
->b_next_transaction
?
1452 jh
->b_next_transaction
->t_tid
: 0,
1457 /* And this case is illegal: we can't reuse another
1458 * transaction's data buffer, ever. */
1462 /* That test should have eliminated the following case: */
1463 J_ASSERT_JH(jh
, jh
->b_frozen_data
== NULL
);
1465 JBUFFER_TRACE(jh
, "file as BJ_Metadata");
1466 spin_lock(&journal
->j_list_lock
);
1467 __jbd2_journal_file_buffer(jh
, transaction
, BJ_Metadata
);
1468 spin_unlock(&journal
->j_list_lock
);
1470 jbd_unlock_bh_state(bh
);
1472 JBUFFER_TRACE(jh
, "exit");
1477 * void jbd2_journal_forget() - bforget() for potentially-journaled buffers.
1478 * @handle: transaction handle
1479 * @bh: bh to 'forget'
1481 * We can only do the bforget if there are no commits pending against the
1482 * buffer. If the buffer is dirty in the current running transaction we
1483 * can safely unlink it.
1485 * bh may not be a journalled buffer at all - it may be a non-JBD
1486 * buffer which came off the hashtable. Check for this.
1488 * Decrements bh->b_count by one.
1490 * Allow this call even if the handle has aborted --- it may be part of
1491 * the caller's cleanup after an abort.
1493 int jbd2_journal_forget (handle_t
*handle
, struct buffer_head
*bh
)
1495 transaction_t
*transaction
= handle
->h_transaction
;
1497 struct journal_head
*jh
;
1498 int drop_reserve
= 0;
1500 int was_modified
= 0;
1502 if (is_handle_aborted(handle
))
1504 journal
= transaction
->t_journal
;
1506 BUFFER_TRACE(bh
, "entry");
1508 jbd_lock_bh_state(bh
);
1510 if (!buffer_jbd(bh
))
1514 /* Critical error: attempting to delete a bitmap buffer, maybe?
1515 * Don't do any jbd operations, and return an error. */
1516 if (!J_EXPECT_JH(jh
, !jh
->b_committed_data
,
1517 "inconsistent data on disk")) {
1522 /* keep track of whether or not this transaction modified us */
1523 was_modified
= jh
->b_modified
;
1526 * The buffer's going from the transaction, we must drop
1527 * all references -bzzz
1531 if (jh
->b_transaction
== transaction
) {
1532 J_ASSERT_JH(jh
, !jh
->b_frozen_data
);
1534 /* If we are forgetting a buffer which is already part
1535 * of this transaction, then we can just drop it from
1536 * the transaction immediately. */
1537 clear_buffer_dirty(bh
);
1538 clear_buffer_jbddirty(bh
);
1540 JBUFFER_TRACE(jh
, "belongs to current transaction: unfile");
1543 * we only want to drop a reference if this transaction
1544 * modified the buffer
1550 * We are no longer going to journal this buffer.
1551 * However, the commit of this transaction is still
1552 * important to the buffer: the delete that we are now
1553 * processing might obsolete an old log entry, so by
1554 * committing, we can satisfy the buffer's checkpoint.
1556 * So, if we have a checkpoint on the buffer, we should
1557 * now refile the buffer on our BJ_Forget list so that
1558 * we know to remove the checkpoint after we commit.
1561 spin_lock(&journal
->j_list_lock
);
1562 if (jh
->b_cp_transaction
) {
1563 __jbd2_journal_temp_unlink_buffer(jh
);
1564 __jbd2_journal_file_buffer(jh
, transaction
, BJ_Forget
);
1566 __jbd2_journal_unfile_buffer(jh
);
1567 if (!buffer_jbd(bh
)) {
1568 spin_unlock(&journal
->j_list_lock
);
1569 jbd_unlock_bh_state(bh
);
1574 spin_unlock(&journal
->j_list_lock
);
1575 } else if (jh
->b_transaction
) {
1576 J_ASSERT_JH(jh
, (jh
->b_transaction
==
1577 journal
->j_committing_transaction
));
1578 /* However, if the buffer is still owned by a prior
1579 * (committing) transaction, we can't drop it yet... */
1580 JBUFFER_TRACE(jh
, "belongs to older transaction");
1581 /* ... but we CAN drop it from the new transaction through
1582 * marking the buffer as freed and set j_next_transaction to
1583 * the new transaction, so that not only the commit code
1584 * knows it should clear dirty bits when it is done with the
1585 * buffer, but also the buffer can be checkpointed only
1586 * after the new transaction commits. */
1588 set_buffer_freed(bh
);
1590 if (!jh
->b_next_transaction
) {
1591 spin_lock(&journal
->j_list_lock
);
1592 jh
->b_next_transaction
= transaction
;
1593 spin_unlock(&journal
->j_list_lock
);
1595 J_ASSERT(jh
->b_next_transaction
== transaction
);
1598 * only drop a reference if this transaction modified
1607 jbd_unlock_bh_state(bh
);
1611 /* no need to reserve log space for this block -bzzz */
1612 handle
->h_buffer_credits
++;
1618 * int jbd2_journal_stop() - complete a transaction
1619 * @handle: tranaction to complete.
1621 * All done for a particular handle.
1623 * There is not much action needed here. We just return any remaining
1624 * buffer credits to the transaction and remove the handle. The only
1625 * complication is that we need to start a commit operation if the
1626 * filesystem is marked for synchronous update.
1628 * jbd2_journal_stop itself will not usually return an error, but it may
1629 * do so in unusual circumstances. In particular, expect it to
1630 * return -EIO if a jbd2_journal_abort has been executed since the
1631 * transaction began.
1633 int jbd2_journal_stop(handle_t
*handle
)
1635 transaction_t
*transaction
= handle
->h_transaction
;
1637 int err
= 0, wait_for_commit
= 0;
1643 * Handle is already detached from the transaction so
1644 * there is nothing to do other than decrease a refcount,
1645 * or free the handle if refcount drops to zero
1647 if (--handle
->h_ref
> 0) {
1648 jbd_debug(4, "h_ref %d -> %d\n", handle
->h_ref
+ 1,
1652 if (handle
->h_rsv_handle
)
1653 jbd2_free_handle(handle
->h_rsv_handle
);
1657 journal
= transaction
->t_journal
;
1659 J_ASSERT(journal_current_handle() == handle
);
1661 if (is_handle_aborted(handle
))
1664 J_ASSERT(atomic_read(&transaction
->t_updates
) > 0);
1666 if (--handle
->h_ref
> 0) {
1667 jbd_debug(4, "h_ref %d -> %d\n", handle
->h_ref
+ 1,
1672 jbd_debug(4, "Handle %p going down\n", handle
);
1673 trace_jbd2_handle_stats(journal
->j_fs_dev
->bd_dev
,
1675 handle
->h_type
, handle
->h_line_no
,
1676 jiffies
- handle
->h_start_jiffies
,
1677 handle
->h_sync
, handle
->h_requested_credits
,
1678 (handle
->h_requested_credits
-
1679 handle
->h_buffer_credits
));
1682 * Implement synchronous transaction batching. If the handle
1683 * was synchronous, don't force a commit immediately. Let's
1684 * yield and let another thread piggyback onto this
1685 * transaction. Keep doing that while new threads continue to
1686 * arrive. It doesn't cost much - we're about to run a commit
1687 * and sleep on IO anyway. Speeds up many-threaded, many-dir
1688 * operations by 30x or more...
1690 * We try and optimize the sleep time against what the
1691 * underlying disk can do, instead of having a static sleep
1692 * time. This is useful for the case where our storage is so
1693 * fast that it is more optimal to go ahead and force a flush
1694 * and wait for the transaction to be committed than it is to
1695 * wait for an arbitrary amount of time for new writers to
1696 * join the transaction. We achieve this by measuring how
1697 * long it takes to commit a transaction, and compare it with
1698 * how long this transaction has been running, and if run time
1699 * < commit time then we sleep for the delta and commit. This
1700 * greatly helps super fast disks that would see slowdowns as
1701 * more threads started doing fsyncs.
1703 * But don't do this if this process was the most recent one
1704 * to perform a synchronous write. We do this to detect the
1705 * case where a single process is doing a stream of sync
1706 * writes. No point in waiting for joiners in that case.
1708 * Setting max_batch_time to 0 disables this completely.
1711 if (handle
->h_sync
&& journal
->j_last_sync_writer
!= pid
&&
1712 journal
->j_max_batch_time
) {
1713 u64 commit_time
, trans_time
;
1715 journal
->j_last_sync_writer
= pid
;
1717 read_lock(&journal
->j_state_lock
);
1718 commit_time
= journal
->j_average_commit_time
;
1719 read_unlock(&journal
->j_state_lock
);
1721 trans_time
= ktime_to_ns(ktime_sub(ktime_get(),
1722 transaction
->t_start_time
));
1724 commit_time
= max_t(u64
, commit_time
,
1725 1000*journal
->j_min_batch_time
);
1726 commit_time
= min_t(u64
, commit_time
,
1727 1000*journal
->j_max_batch_time
);
1729 if (trans_time
< commit_time
) {
1730 ktime_t expires
= ktime_add_ns(ktime_get(),
1732 set_current_state(TASK_UNINTERRUPTIBLE
);
1733 schedule_hrtimeout(&expires
, HRTIMER_MODE_ABS
);
1738 transaction
->t_synchronous_commit
= 1;
1739 current
->journal_info
= NULL
;
1740 atomic_sub(handle
->h_buffer_credits
,
1741 &transaction
->t_outstanding_credits
);
1744 * If the handle is marked SYNC, we need to set another commit
1745 * going! We also want to force a commit if the current
1746 * transaction is occupying too much of the log, or if the
1747 * transaction is too old now.
1749 if (handle
->h_sync
||
1750 (atomic_read(&transaction
->t_outstanding_credits
) >
1751 journal
->j_max_transaction_buffers
) ||
1752 time_after_eq(jiffies
, transaction
->t_expires
)) {
1753 /* Do this even for aborted journals: an abort still
1754 * completes the commit thread, it just doesn't write
1755 * anything to disk. */
1757 jbd_debug(2, "transaction too old, requesting commit for "
1758 "handle %p\n", handle
);
1759 /* This is non-blocking */
1760 jbd2_log_start_commit(journal
, transaction
->t_tid
);
1763 * Special case: JBD2_SYNC synchronous updates require us
1764 * to wait for the commit to complete.
1766 if (handle
->h_sync
&& !(current
->flags
& PF_MEMALLOC
))
1767 wait_for_commit
= 1;
1771 * Once we drop t_updates, if it goes to zero the transaction
1772 * could start committing on us and eventually disappear. So
1773 * once we do this, we must not dereference transaction
1776 tid
= transaction
->t_tid
;
1777 if (atomic_dec_and_test(&transaction
->t_updates
)) {
1778 wake_up(&journal
->j_wait_updates
);
1779 if (journal
->j_barrier_count
)
1780 wake_up(&journal
->j_wait_transaction_locked
);
1783 if (wait_for_commit
)
1784 err
= jbd2_log_wait_commit(journal
, tid
);
1786 lock_map_release(&handle
->h_lockdep_map
);
1788 if (handle
->h_rsv_handle
)
1789 jbd2_journal_free_reserved(handle
->h_rsv_handle
);
1791 jbd2_free_handle(handle
);
1797 * List management code snippets: various functions for manipulating the
1798 * transaction buffer lists.
1803 * Append a buffer to a transaction list, given the transaction's list head
1806 * j_list_lock is held.
1808 * jbd_lock_bh_state(jh2bh(jh)) is held.
1812 __blist_add_buffer(struct journal_head
**list
, struct journal_head
*jh
)
1815 jh
->b_tnext
= jh
->b_tprev
= jh
;
1818 /* Insert at the tail of the list to preserve order */
1819 struct journal_head
*first
= *list
, *last
= first
->b_tprev
;
1821 jh
->b_tnext
= first
;
1822 last
->b_tnext
= first
->b_tprev
= jh
;
1827 * Remove a buffer from a transaction list, given the transaction's list
1830 * Called with j_list_lock held, and the journal may not be locked.
1832 * jbd_lock_bh_state(jh2bh(jh)) is held.
1836 __blist_del_buffer(struct journal_head
**list
, struct journal_head
*jh
)
1839 *list
= jh
->b_tnext
;
1843 jh
->b_tprev
->b_tnext
= jh
->b_tnext
;
1844 jh
->b_tnext
->b_tprev
= jh
->b_tprev
;
1848 * Remove a buffer from the appropriate transaction list.
1850 * Note that this function can *change* the value of
1851 * bh->b_transaction->t_buffers, t_forget, t_shadow_list, t_log_list or
1852 * t_reserved_list. If the caller is holding onto a copy of one of these
1853 * pointers, it could go bad. Generally the caller needs to re-read the
1854 * pointer from the transaction_t.
1856 * Called under j_list_lock.
1858 static void __jbd2_journal_temp_unlink_buffer(struct journal_head
*jh
)
1860 struct journal_head
**list
= NULL
;
1861 transaction_t
*transaction
;
1862 struct buffer_head
*bh
= jh2bh(jh
);
1864 J_ASSERT_JH(jh
, jbd_is_locked_bh_state(bh
));
1865 transaction
= jh
->b_transaction
;
1867 assert_spin_locked(&transaction
->t_journal
->j_list_lock
);
1869 J_ASSERT_JH(jh
, jh
->b_jlist
< BJ_Types
);
1870 if (jh
->b_jlist
!= BJ_None
)
1871 J_ASSERT_JH(jh
, transaction
!= NULL
);
1873 switch (jh
->b_jlist
) {
1877 transaction
->t_nr_buffers
--;
1878 J_ASSERT_JH(jh
, transaction
->t_nr_buffers
>= 0);
1879 list
= &transaction
->t_buffers
;
1882 list
= &transaction
->t_forget
;
1885 list
= &transaction
->t_shadow_list
;
1888 list
= &transaction
->t_reserved_list
;
1892 __blist_del_buffer(list
, jh
);
1893 jh
->b_jlist
= BJ_None
;
1894 if (transaction
&& is_journal_aborted(transaction
->t_journal
))
1895 clear_buffer_jbddirty(bh
);
1896 else if (test_clear_buffer_jbddirty(bh
))
1897 mark_buffer_dirty(bh
); /* Expose it to the VM */
1901 * Remove buffer from all transactions.
1903 * Called with bh_state lock and j_list_lock
1905 * jh and bh may be already freed when this function returns.
1907 static void __jbd2_journal_unfile_buffer(struct journal_head
*jh
)
1909 J_ASSERT_JH(jh
, jh
->b_transaction
!= NULL
);
1910 J_ASSERT_JH(jh
, jh
->b_next_transaction
== NULL
);
1912 __jbd2_journal_temp_unlink_buffer(jh
);
1913 jh
->b_transaction
= NULL
;
1914 jbd2_journal_put_journal_head(jh
);
1917 void jbd2_journal_unfile_buffer(journal_t
*journal
, struct journal_head
*jh
)
1919 struct buffer_head
*bh
= jh2bh(jh
);
1921 /* Get reference so that buffer cannot be freed before we unlock it */
1923 jbd_lock_bh_state(bh
);
1924 spin_lock(&journal
->j_list_lock
);
1925 __jbd2_journal_unfile_buffer(jh
);
1926 spin_unlock(&journal
->j_list_lock
);
1927 jbd_unlock_bh_state(bh
);
1932 * Called from jbd2_journal_try_to_free_buffers().
1934 * Called under jbd_lock_bh_state(bh)
1937 __journal_try_to_free_buffer(journal_t
*journal
, struct buffer_head
*bh
)
1939 struct journal_head
*jh
;
1943 if (buffer_locked(bh
) || buffer_dirty(bh
))
1946 if (jh
->b_next_transaction
!= NULL
|| jh
->b_transaction
!= NULL
)
1949 spin_lock(&journal
->j_list_lock
);
1950 if (jh
->b_cp_transaction
!= NULL
) {
1951 /* written-back checkpointed metadata buffer */
1952 JBUFFER_TRACE(jh
, "remove from checkpoint list");
1953 __jbd2_journal_remove_checkpoint(jh
);
1955 spin_unlock(&journal
->j_list_lock
);
1961 * int jbd2_journal_try_to_free_buffers() - try to free page buffers.
1962 * @journal: journal for operation
1963 * @page: to try and free
1964 * @gfp_mask: we use the mask to detect how hard should we try to release
1965 * buffers. If __GFP_DIRECT_RECLAIM and __GFP_FS is set, we wait for commit
1966 * code to release the buffers.
1969 * For all the buffers on this page,
1970 * if they are fully written out ordered data, move them onto BUF_CLEAN
1971 * so try_to_free_buffers() can reap them.
1973 * This function returns non-zero if we wish try_to_free_buffers()
1974 * to be called. We do this if the page is releasable by try_to_free_buffers().
1975 * We also do it if the page has locked or dirty buffers and the caller wants
1976 * us to perform sync or async writeout.
1978 * This complicates JBD locking somewhat. We aren't protected by the
1979 * BKL here. We wish to remove the buffer from its committing or
1980 * running transaction's ->t_datalist via __jbd2_journal_unfile_buffer.
1982 * This may *change* the value of transaction_t->t_datalist, so anyone
1983 * who looks at t_datalist needs to lock against this function.
1985 * Even worse, someone may be doing a jbd2_journal_dirty_data on this
1986 * buffer. So we need to lock against that. jbd2_journal_dirty_data()
1987 * will come out of the lock with the buffer dirty, which makes it
1988 * ineligible for release here.
1990 * Who else is affected by this? hmm... Really the only contender
1991 * is do_get_write_access() - it could be looking at the buffer while
1992 * journal_try_to_free_buffer() is changing its state. But that
1993 * cannot happen because we never reallocate freed data as metadata
1994 * while the data is part of a transaction. Yes?
1996 * Return 0 on failure, 1 on success
1998 int jbd2_journal_try_to_free_buffers(journal_t
*journal
,
1999 struct page
*page
, gfp_t gfp_mask
)
2001 struct buffer_head
*head
;
2002 struct buffer_head
*bh
;
2003 bool has_write_io_error
= false;
2006 J_ASSERT(PageLocked(page
));
2008 head
= page_buffers(page
);
2011 struct journal_head
*jh
;
2014 * We take our own ref against the journal_head here to avoid
2015 * having to add tons of locking around each instance of
2016 * jbd2_journal_put_journal_head().
2018 jh
= jbd2_journal_grab_journal_head(bh
);
2022 jbd_lock_bh_state(bh
);
2023 __journal_try_to_free_buffer(journal
, bh
);
2024 jbd2_journal_put_journal_head(jh
);
2025 jbd_unlock_bh_state(bh
);
2030 * If we free a metadata buffer which has been failed to
2031 * write out, the jbd2 checkpoint procedure will not detect
2032 * this failure and may lead to filesystem inconsistency
2033 * after cleanup journal tail.
2035 if (buffer_write_io_error(bh
)) {
2036 pr_err("JBD2: Error while async write back metadata bh %llu.",
2037 (unsigned long long)bh
->b_blocknr
);
2038 has_write_io_error
= true;
2040 } while ((bh
= bh
->b_this_page
) != head
);
2042 ret
= try_to_free_buffers(page
);
2045 if (has_write_io_error
)
2046 jbd2_journal_abort(journal
, -EIO
);
2052 * This buffer is no longer needed. If it is on an older transaction's
2053 * checkpoint list we need to record it on this transaction's forget list
2054 * to pin this buffer (and hence its checkpointing transaction) down until
2055 * this transaction commits. If the buffer isn't on a checkpoint list, we
2057 * Returns non-zero if JBD no longer has an interest in the buffer.
2059 * Called under j_list_lock.
2061 * Called under jbd_lock_bh_state(bh).
2063 static int __dispose_buffer(struct journal_head
*jh
, transaction_t
*transaction
)
2066 struct buffer_head
*bh
= jh2bh(jh
);
2068 if (jh
->b_cp_transaction
) {
2069 JBUFFER_TRACE(jh
, "on running+cp transaction");
2070 __jbd2_journal_temp_unlink_buffer(jh
);
2072 * We don't want to write the buffer anymore, clear the
2073 * bit so that we don't confuse checks in
2074 * __journal_file_buffer
2076 clear_buffer_dirty(bh
);
2077 __jbd2_journal_file_buffer(jh
, transaction
, BJ_Forget
);
2080 JBUFFER_TRACE(jh
, "on running transaction");
2081 __jbd2_journal_unfile_buffer(jh
);
2087 * jbd2_journal_invalidatepage
2089 * This code is tricky. It has a number of cases to deal with.
2091 * There are two invariants which this code relies on:
2093 * i_size must be updated on disk before we start calling invalidatepage on the
2096 * This is done in ext3 by defining an ext3_setattr method which
2097 * updates i_size before truncate gets going. By maintaining this
2098 * invariant, we can be sure that it is safe to throw away any buffers
2099 * attached to the current transaction: once the transaction commits,
2100 * we know that the data will not be needed.
2102 * Note however that we can *not* throw away data belonging to the
2103 * previous, committing transaction!
2105 * Any disk blocks which *are* part of the previous, committing
2106 * transaction (and which therefore cannot be discarded immediately) are
2107 * not going to be reused in the new running transaction
2109 * The bitmap committed_data images guarantee this: any block which is
2110 * allocated in one transaction and removed in the next will be marked
2111 * as in-use in the committed_data bitmap, so cannot be reused until
2112 * the next transaction to delete the block commits. This means that
2113 * leaving committing buffers dirty is quite safe: the disk blocks
2114 * cannot be reallocated to a different file and so buffer aliasing is
2118 * The above applies mainly to ordered data mode. In writeback mode we
2119 * don't make guarantees about the order in which data hits disk --- in
2120 * particular we don't guarantee that new dirty data is flushed before
2121 * transaction commit --- so it is always safe just to discard data
2122 * immediately in that mode. --sct
2126 * The journal_unmap_buffer helper function returns zero if the buffer
2127 * concerned remains pinned as an anonymous buffer belonging to an older
2130 * We're outside-transaction here. Either or both of j_running_transaction
2131 * and j_committing_transaction may be NULL.
2133 static int journal_unmap_buffer(journal_t
*journal
, struct buffer_head
*bh
,
2136 transaction_t
*transaction
;
2137 struct journal_head
*jh
;
2140 BUFFER_TRACE(bh
, "entry");
2143 * It is safe to proceed here without the j_list_lock because the
2144 * buffers cannot be stolen by try_to_free_buffers as long as we are
2145 * holding the page lock. --sct
2148 if (!buffer_jbd(bh
))
2149 goto zap_buffer_unlocked
;
2151 /* OK, we have data buffer in journaled mode */
2152 write_lock(&journal
->j_state_lock
);
2153 jbd_lock_bh_state(bh
);
2154 spin_lock(&journal
->j_list_lock
);
2156 jh
= jbd2_journal_grab_journal_head(bh
);
2158 goto zap_buffer_no_jh
;
2161 * We cannot remove the buffer from checkpoint lists until the
2162 * transaction adding inode to orphan list (let's call it T)
2163 * is committed. Otherwise if the transaction changing the
2164 * buffer would be cleaned from the journal before T is
2165 * committed, a crash will cause that the correct contents of
2166 * the buffer will be lost. On the other hand we have to
2167 * clear the buffer dirty bit at latest at the moment when the
2168 * transaction marking the buffer as freed in the filesystem
2169 * structures is committed because from that moment on the
2170 * block can be reallocated and used by a different page.
2171 * Since the block hasn't been freed yet but the inode has
2172 * already been added to orphan list, it is safe for us to add
2173 * the buffer to BJ_Forget list of the newest transaction.
2175 * Also we have to clear buffer_mapped flag of a truncated buffer
2176 * because the buffer_head may be attached to the page straddling
2177 * i_size (can happen only when blocksize < pagesize) and thus the
2178 * buffer_head can be reused when the file is extended again. So we end
2179 * up keeping around invalidated buffers attached to transactions'
2180 * BJ_Forget list just to stop checkpointing code from cleaning up
2181 * the transaction this buffer was modified in.
2183 transaction
= jh
->b_transaction
;
2184 if (transaction
== NULL
) {
2185 /* First case: not on any transaction. If it
2186 * has no checkpoint link, then we can zap it:
2187 * it's a writeback-mode buffer so we don't care
2188 * if it hits disk safely. */
2189 if (!jh
->b_cp_transaction
) {
2190 JBUFFER_TRACE(jh
, "not on any transaction: zap");
2194 if (!buffer_dirty(bh
)) {
2195 /* bdflush has written it. We can drop it now */
2196 __jbd2_journal_remove_checkpoint(jh
);
2200 /* OK, it must be in the journal but still not
2201 * written fully to disk: it's metadata or
2202 * journaled data... */
2204 if (journal
->j_running_transaction
) {
2205 /* ... and once the current transaction has
2206 * committed, the buffer won't be needed any
2208 JBUFFER_TRACE(jh
, "checkpointed: add to BJ_Forget");
2209 may_free
= __dispose_buffer(jh
,
2210 journal
->j_running_transaction
);
2213 /* There is no currently-running transaction. So the
2214 * orphan record which we wrote for this file must have
2215 * passed into commit. We must attach this buffer to
2216 * the committing transaction, if it exists. */
2217 if (journal
->j_committing_transaction
) {
2218 JBUFFER_TRACE(jh
, "give to committing trans");
2219 may_free
= __dispose_buffer(jh
,
2220 journal
->j_committing_transaction
);
2223 /* The orphan record's transaction has
2224 * committed. We can cleanse this buffer */
2225 clear_buffer_jbddirty(bh
);
2226 __jbd2_journal_remove_checkpoint(jh
);
2230 } else if (transaction
== journal
->j_committing_transaction
) {
2231 JBUFFER_TRACE(jh
, "on committing transaction");
2233 * The buffer is committing, we simply cannot touch
2234 * it. If the page is straddling i_size we have to wait
2235 * for commit and try again.
2238 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
);
2245 * OK, buffer won't be reachable after truncate. We just clear
2246 * b_modified to not confuse transaction credit accounting, and
2247 * set j_next_transaction to the running transaction (if there
2248 * is one) and mark buffer as freed so that commit code knows
2249 * it should clear dirty bits when it is done with the buffer.
2251 set_buffer_freed(bh
);
2252 if (journal
->j_running_transaction
&& buffer_jbddirty(bh
))
2253 jh
->b_next_transaction
= journal
->j_running_transaction
;
2255 jbd2_journal_put_journal_head(jh
);
2256 spin_unlock(&journal
->j_list_lock
);
2257 jbd_unlock_bh_state(bh
);
2258 write_unlock(&journal
->j_state_lock
);
2261 /* Good, the buffer belongs to the running transaction.
2262 * We are writing our own transaction's data, not any
2263 * previous one's, so it is safe to throw it away
2264 * (remember that we expect the filesystem to have set
2265 * i_size already for this truncate so recovery will not
2266 * expose the disk blocks we are discarding here.) */
2267 J_ASSERT_JH(jh
, transaction
== journal
->j_running_transaction
);
2268 JBUFFER_TRACE(jh
, "on running transaction");
2269 may_free
= __dispose_buffer(jh
, transaction
);
2274 * This is tricky. Although the buffer is truncated, it may be reused
2275 * if blocksize < pagesize and it is attached to the page straddling
2276 * EOF. Since the buffer might have been added to BJ_Forget list of the
2277 * running transaction, journal_get_write_access() won't clear
2278 * b_modified and credit accounting gets confused. So clear b_modified
2282 jbd2_journal_put_journal_head(jh
);
2284 spin_unlock(&journal
->j_list_lock
);
2285 jbd_unlock_bh_state(bh
);
2286 write_unlock(&journal
->j_state_lock
);
2287 zap_buffer_unlocked
:
2288 clear_buffer_dirty(bh
);
2289 J_ASSERT_BH(bh
, !buffer_jbddirty(bh
));
2290 clear_buffer_mapped(bh
);
2291 clear_buffer_req(bh
);
2292 clear_buffer_new(bh
);
2293 clear_buffer_delay(bh
);
2294 clear_buffer_unwritten(bh
);
2300 * void jbd2_journal_invalidatepage()
2301 * @journal: journal to use for flush...
2302 * @page: page to flush
2303 * @offset: start of the range to invalidate
2304 * @length: length of the range to invalidate
2306 * Reap page buffers containing data after in the specified range in page.
2307 * Can return -EBUSY if buffers are part of the committing transaction and
2308 * the page is straddling i_size. Caller then has to wait for current commit
2311 int jbd2_journal_invalidatepage(journal_t
*journal
,
2313 unsigned int offset
,
2314 unsigned int length
)
2316 struct buffer_head
*head
, *bh
, *next
;
2317 unsigned int stop
= offset
+ length
;
2318 unsigned int curr_off
= 0;
2319 int partial_page
= (offset
|| length
< PAGE_CACHE_SIZE
);
2323 if (!PageLocked(page
))
2325 if (!page_has_buffers(page
))
2328 BUG_ON(stop
> PAGE_CACHE_SIZE
|| stop
< length
);
2330 /* We will potentially be playing with lists other than just the
2331 * data lists (especially for journaled data mode), so be
2332 * cautious in our locking. */
2334 head
= bh
= page_buffers(page
);
2336 unsigned int next_off
= curr_off
+ bh
->b_size
;
2337 next
= bh
->b_this_page
;
2339 if (next_off
> stop
)
2342 if (offset
<= curr_off
) {
2343 /* This block is wholly outside the truncation point */
2345 ret
= journal_unmap_buffer(journal
, bh
, partial_page
);
2351 curr_off
= next_off
;
2354 } while (bh
!= head
);
2356 if (!partial_page
) {
2357 if (may_free
&& try_to_free_buffers(page
))
2358 J_ASSERT(!page_has_buffers(page
));
2364 * File a buffer on the given transaction list.
2366 void __jbd2_journal_file_buffer(struct journal_head
*jh
,
2367 transaction_t
*transaction
, int jlist
)
2369 struct journal_head
**list
= NULL
;
2371 struct buffer_head
*bh
= jh2bh(jh
);
2373 J_ASSERT_JH(jh
, jbd_is_locked_bh_state(bh
));
2374 assert_spin_locked(&transaction
->t_journal
->j_list_lock
);
2376 J_ASSERT_JH(jh
, jh
->b_jlist
< BJ_Types
);
2377 J_ASSERT_JH(jh
, jh
->b_transaction
== transaction
||
2378 jh
->b_transaction
== NULL
);
2380 if (jh
->b_transaction
&& jh
->b_jlist
== jlist
)
2383 if (jlist
== BJ_Metadata
|| jlist
== BJ_Reserved
||
2384 jlist
== BJ_Shadow
|| jlist
== BJ_Forget
) {
2386 * For metadata buffers, we track dirty bit in buffer_jbddirty
2387 * instead of buffer_dirty. We should not see a dirty bit set
2388 * here because we clear it in do_get_write_access but e.g.
2389 * tune2fs can modify the sb and set the dirty bit at any time
2390 * so we try to gracefully handle that.
2392 if (buffer_dirty(bh
))
2393 warn_dirty_buffer(bh
);
2394 if (test_clear_buffer_dirty(bh
) ||
2395 test_clear_buffer_jbddirty(bh
))
2399 if (jh
->b_transaction
)
2400 __jbd2_journal_temp_unlink_buffer(jh
);
2402 jbd2_journal_grab_journal_head(bh
);
2403 jh
->b_transaction
= transaction
;
2407 J_ASSERT_JH(jh
, !jh
->b_committed_data
);
2408 J_ASSERT_JH(jh
, !jh
->b_frozen_data
);
2411 transaction
->t_nr_buffers
++;
2412 list
= &transaction
->t_buffers
;
2415 list
= &transaction
->t_forget
;
2418 list
= &transaction
->t_shadow_list
;
2421 list
= &transaction
->t_reserved_list
;
2425 __blist_add_buffer(list
, jh
);
2426 jh
->b_jlist
= jlist
;
2429 set_buffer_jbddirty(bh
);
2432 void jbd2_journal_file_buffer(struct journal_head
*jh
,
2433 transaction_t
*transaction
, int jlist
)
2435 jbd_lock_bh_state(jh2bh(jh
));
2436 spin_lock(&transaction
->t_journal
->j_list_lock
);
2437 __jbd2_journal_file_buffer(jh
, transaction
, jlist
);
2438 spin_unlock(&transaction
->t_journal
->j_list_lock
);
2439 jbd_unlock_bh_state(jh2bh(jh
));
2443 * Remove a buffer from its current buffer list in preparation for
2444 * dropping it from its current transaction entirely. If the buffer has
2445 * already started to be used by a subsequent transaction, refile the
2446 * buffer on that transaction's metadata list.
2448 * Called under j_list_lock
2449 * Called under jbd_lock_bh_state(jh2bh(jh))
2451 * jh and bh may be already free when this function returns
2453 void __jbd2_journal_refile_buffer(struct journal_head
*jh
)
2455 int was_dirty
, jlist
;
2456 struct buffer_head
*bh
= jh2bh(jh
);
2458 J_ASSERT_JH(jh
, jbd_is_locked_bh_state(bh
));
2459 if (jh
->b_transaction
)
2460 assert_spin_locked(&jh
->b_transaction
->t_journal
->j_list_lock
);
2462 /* If the buffer is now unused, just drop it. */
2463 if (jh
->b_next_transaction
== NULL
) {
2464 __jbd2_journal_unfile_buffer(jh
);
2469 * It has been modified by a later transaction: add it to the new
2470 * transaction's metadata list.
2473 was_dirty
= test_clear_buffer_jbddirty(bh
);
2474 __jbd2_journal_temp_unlink_buffer(jh
);
2477 * b_transaction must be set, otherwise the new b_transaction won't
2478 * be holding jh reference
2480 J_ASSERT_JH(jh
, jh
->b_transaction
!= NULL
);
2483 * We set b_transaction here because b_next_transaction will inherit
2484 * our jh reference and thus __jbd2_journal_file_buffer() must not
2487 WRITE_ONCE(jh
->b_transaction
, jh
->b_next_transaction
);
2488 WRITE_ONCE(jh
->b_next_transaction
, NULL
);
2489 if (buffer_freed(bh
))
2491 else if (jh
->b_modified
)
2492 jlist
= BJ_Metadata
;
2494 jlist
= BJ_Reserved
;
2495 __jbd2_journal_file_buffer(jh
, jh
->b_transaction
, jlist
);
2496 J_ASSERT_JH(jh
, jh
->b_transaction
->t_state
== T_RUNNING
);
2499 set_buffer_jbddirty(bh
);
2503 * __jbd2_journal_refile_buffer() with necessary locking added. We take our
2504 * bh reference so that we can safely unlock bh.
2506 * The jh and bh may be freed by this call.
2508 void jbd2_journal_refile_buffer(journal_t
*journal
, struct journal_head
*jh
)
2510 struct buffer_head
*bh
= jh2bh(jh
);
2512 /* Get reference so that buffer cannot be freed before we unlock it */
2514 jbd_lock_bh_state(bh
);
2515 spin_lock(&journal
->j_list_lock
);
2516 __jbd2_journal_refile_buffer(jh
);
2517 jbd_unlock_bh_state(bh
);
2518 spin_unlock(&journal
->j_list_lock
);
2523 * File inode in the inode list of the handle's transaction
2525 int jbd2_journal_file_inode(handle_t
*handle
, struct jbd2_inode
*jinode
)
2527 transaction_t
*transaction
= handle
->h_transaction
;
2530 if (is_handle_aborted(handle
))
2532 journal
= transaction
->t_journal
;
2534 jbd_debug(4, "Adding inode %lu, tid:%d\n", jinode
->i_vfs_inode
->i_ino
,
2535 transaction
->t_tid
);
2538 * First check whether inode isn't already on the transaction's
2539 * lists without taking the lock. Note that this check is safe
2540 * without the lock as we cannot race with somebody removing inode
2541 * from the transaction. The reason is that we remove inode from the
2542 * transaction only in journal_release_jbd_inode() and when we commit
2543 * the transaction. We are guarded from the first case by holding
2544 * a reference to the inode. We are safe against the second case
2545 * because if jinode->i_transaction == transaction, commit code
2546 * cannot touch the transaction because we hold reference to it,
2547 * and if jinode->i_next_transaction == transaction, commit code
2548 * will only file the inode where we want it.
2550 if (jinode
->i_transaction
== transaction
||
2551 jinode
->i_next_transaction
== transaction
)
2554 spin_lock(&journal
->j_list_lock
);
2556 if (jinode
->i_transaction
== transaction
||
2557 jinode
->i_next_transaction
== transaction
)
2561 * We only ever set this variable to 1 so the test is safe. Since
2562 * t_need_data_flush is likely to be set, we do the test to save some
2563 * cacheline bouncing
2565 if (!transaction
->t_need_data_flush
)
2566 transaction
->t_need_data_flush
= 1;
2567 /* On some different transaction's list - should be
2568 * the committing one */
2569 if (jinode
->i_transaction
) {
2570 J_ASSERT(jinode
->i_next_transaction
== NULL
);
2571 J_ASSERT(jinode
->i_transaction
==
2572 journal
->j_committing_transaction
);
2573 jinode
->i_next_transaction
= transaction
;
2576 /* Not on any transaction list... */
2577 J_ASSERT(!jinode
->i_next_transaction
);
2578 jinode
->i_transaction
= transaction
;
2579 list_add(&jinode
->i_list
, &transaction
->t_inode_list
);
2581 spin_unlock(&journal
->j_list_lock
);
2587 * File truncate and transaction commit interact with each other in a
2588 * non-trivial way. If a transaction writing data block A is
2589 * committing, we cannot discard the data by truncate until we have
2590 * written them. Otherwise if we crashed after the transaction with
2591 * write has committed but before the transaction with truncate has
2592 * committed, we could see stale data in block A. This function is a
2593 * helper to solve this problem. It starts writeout of the truncated
2594 * part in case it is in the committing transaction.
2596 * Filesystem code must call this function when inode is journaled in
2597 * ordered mode before truncation happens and after the inode has been
2598 * placed on orphan list with the new inode size. The second condition
2599 * avoids the race that someone writes new data and we start
2600 * committing the transaction after this function has been called but
2601 * before a transaction for truncate is started (and furthermore it
2602 * allows us to optimize the case where the addition to orphan list
2603 * happens in the same transaction as write --- we don't have to write
2604 * any data in such case).
2606 int jbd2_journal_begin_ordered_truncate(journal_t
*journal
,
2607 struct jbd2_inode
*jinode
,
2610 transaction_t
*inode_trans
, *commit_trans
;
2613 /* This is a quick check to avoid locking if not necessary */
2614 if (!jinode
->i_transaction
)
2616 /* Locks are here just to force reading of recent values, it is
2617 * enough that the transaction was not committing before we started
2618 * a transaction adding the inode to orphan list */
2619 read_lock(&journal
->j_state_lock
);
2620 commit_trans
= journal
->j_committing_transaction
;
2621 read_unlock(&journal
->j_state_lock
);
2622 spin_lock(&journal
->j_list_lock
);
2623 inode_trans
= jinode
->i_transaction
;
2624 spin_unlock(&journal
->j_list_lock
);
2625 if (inode_trans
== commit_trans
) {
2626 ret
= filemap_fdatawrite_range(jinode
->i_vfs_inode
->i_mapping
,
2627 new_size
, LLONG_MAX
);
2629 jbd2_journal_abort(journal
, ret
);