1 // SPDX-License-Identifier: GPL-2.0+
3 * linux/fs/jbd2/transaction.c
5 * Written by Stephen C. Tweedie <sct@redhat.com>, 1998
7 * Copyright 1998 Red Hat corp --- All Rights Reserved
9 * Generic filesystem transaction handling code; part of the ext2fs
12 * This file manages transactions (compound commits managed by the
13 * journaling code) and handles (individual atomic operations by the
17 #include <linux/time.h>
19 #include <linux/jbd2.h>
20 #include <linux/errno.h>
21 #include <linux/slab.h>
22 #include <linux/timer.h>
24 #include <linux/highmem.h>
25 #include <linux/hrtimer.h>
26 #include <linux/backing-dev.h>
27 #include <linux/bug.h>
28 #include <linux/module.h>
29 #include <linux/sched/mm.h>
31 #include <trace/events/jbd2.h>
33 static void __jbd2_journal_temp_unlink_buffer(struct journal_head *jh);
34 static void __jbd2_journal_unfile_buffer(struct journal_head *jh);
36 static struct kmem_cache *transaction_cache;
37 int __init jbd2_journal_init_transaction_cache(void)
39 J_ASSERT(!transaction_cache);
40 transaction_cache = kmem_cache_create("jbd2_transaction_s",
41 sizeof(transaction_t),
43 SLAB_HWCACHE_ALIGN|SLAB_TEMPORARY,
45 if (transaction_cache)
50 void jbd2_journal_destroy_transaction_cache(void)
52 if (transaction_cache) {
53 kmem_cache_destroy(transaction_cache);
54 transaction_cache = NULL;
58 void jbd2_journal_free_transaction(transaction_t *transaction)
60 if (unlikely(ZERO_OR_NULL_PTR(transaction)))
62 kmem_cache_free(transaction_cache, transaction);
66 * jbd2_get_transaction: obtain a new transaction_t object.
68 * Simply allocate and initialise a new transaction. Create it in
69 * RUNNING state and add it to the current journal (which should not
70 * have an existing running transaction: we only make a new transaction
71 * once we have started to commit the old one).
74 * The journal MUST be locked. We don't perform atomic mallocs on the
75 * new transaction and we can't block without protecting against other
76 * processes trying to touch the journal while it is in transition.
80 static transaction_t *
81 jbd2_get_transaction(journal_t *journal, transaction_t *transaction)
83 transaction->t_journal = journal;
84 transaction->t_state = T_RUNNING;
85 transaction->t_start_time = ktime_get();
86 transaction->t_tid = journal->j_transaction_sequence++;
87 transaction->t_expires = jiffies + journal->j_commit_interval;
88 spin_lock_init(&transaction->t_handle_lock);
89 atomic_set(&transaction->t_updates, 0);
90 atomic_set(&transaction->t_outstanding_credits,
91 atomic_read(&journal->j_reserved_credits));
92 atomic_set(&transaction->t_handle_count, 0);
93 INIT_LIST_HEAD(&transaction->t_inode_list);
94 INIT_LIST_HEAD(&transaction->t_private_list);
96 /* Set up the commit timer for the new transaction. */
97 journal->j_commit_timer.expires = round_jiffies_up(transaction->t_expires);
98 add_timer(&journal->j_commit_timer);
100 J_ASSERT(journal->j_running_transaction == NULL);
101 journal->j_running_transaction = transaction;
102 transaction->t_max_wait = 0;
103 transaction->t_start = jiffies;
104 transaction->t_requested = 0;
112 * A handle_t is an object which represents a single atomic update to a
113 * filesystem, and which tracks all of the modifications which form part
114 * of that one update.
118 * Update transaction's maximum wait time, if debugging is enabled.
120 * In order for t_max_wait to be reliable, it must be protected by a
121 * lock. But doing so will mean that start_this_handle() can not be
122 * run in parallel on SMP systems, which limits our scalability. So
123 * unless debugging is enabled, we no longer update t_max_wait, which
124 * means that maximum wait time reported by the jbd2_run_stats
125 * tracepoint will always be zero.
127 static inline void update_t_max_wait(transaction_t *transaction,
130 #ifdef CONFIG_JBD2_DEBUG
131 if (jbd2_journal_enable_debug &&
132 time_after(transaction->t_start, ts)) {
133 ts = jbd2_time_diff(ts, transaction->t_start);
134 spin_lock(&transaction->t_handle_lock);
135 if (ts > transaction->t_max_wait)
136 transaction->t_max_wait = ts;
137 spin_unlock(&transaction->t_handle_lock);
143 * Wait until running transaction passes T_LOCKED state. Also starts the commit
144 * if needed. The function expects running transaction to exist and releases
147 static void wait_transaction_locked(journal_t *journal)
148 __releases(journal->j_state_lock)
152 tid_t tid = journal->j_running_transaction->t_tid;
154 prepare_to_wait(&journal->j_wait_transaction_locked, &wait,
155 TASK_UNINTERRUPTIBLE);
156 need_to_start = !tid_geq(journal->j_commit_request, tid);
157 read_unlock(&journal->j_state_lock);
159 jbd2_log_start_commit(journal, tid);
160 jbd2_might_wait_for_commit(journal);
162 finish_wait(&journal->j_wait_transaction_locked, &wait);
165 static void sub_reserved_credits(journal_t *journal, int blocks)
167 atomic_sub(blocks, &journal->j_reserved_credits);
168 wake_up(&journal->j_wait_reserved);
172 * Wait until we can add credits for handle to the running transaction. Called
173 * with j_state_lock held for reading. Returns 0 if handle joined the running
174 * transaction. Returns 1 if we had to wait, j_state_lock is dropped, and
177 static int add_transaction_credits(journal_t *journal, int blocks,
180 transaction_t *t = journal->j_running_transaction;
182 int total = blocks + rsv_blocks;
185 * If the current transaction is locked down for commit, wait
186 * for the lock to be released.
188 if (t->t_state == T_LOCKED) {
189 wait_transaction_locked(journal);
194 * If there is not enough space left in the log to write all
195 * potential buffers requested by this operation, we need to
196 * stall pending a log checkpoint to free some more log space.
198 needed = atomic_add_return(total, &t->t_outstanding_credits);
199 if (needed > journal->j_max_transaction_buffers) {
201 * If the current transaction is already too large,
202 * then start to commit it: we can then go back and
203 * attach this handle to a new transaction.
205 atomic_sub(total, &t->t_outstanding_credits);
208 * Is the number of reserved credits in the current transaction too
209 * big to fit this handle? Wait until reserved credits are freed.
211 if (atomic_read(&journal->j_reserved_credits) + total >
212 journal->j_max_transaction_buffers) {
213 read_unlock(&journal->j_state_lock);
214 jbd2_might_wait_for_commit(journal);
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 jbd2_might_wait_for_commit(journal);
240 write_lock(&journal->j_state_lock);
241 if (jbd2_log_space_left(journal) < jbd2_space_needed(journal))
242 __jbd2_log_wait_for_space(journal);
243 write_unlock(&journal->j_state_lock);
247 /* No reservation? We are done... */
251 needed = atomic_add_return(rsv_blocks, &journal->j_reserved_credits);
252 /* We allow at most half of a transaction to be reserved */
253 if (needed > journal->j_max_transaction_buffers / 2) {
254 sub_reserved_credits(journal, rsv_blocks);
255 atomic_sub(total, &t->t_outstanding_credits);
256 read_unlock(&journal->j_state_lock);
257 jbd2_might_wait_for_commit(journal);
258 wait_event(journal->j_wait_reserved,
259 atomic_read(&journal->j_reserved_credits) + rsv_blocks
260 <= journal->j_max_transaction_buffers / 2);
267 * start_this_handle: Given a handle, deal with any locking or stalling
268 * needed to make sure that there is enough journal space for the handle
269 * to begin. Attach the handle to a transaction and set up the
270 * transaction's buffer credits.
273 static int start_this_handle(journal_t *journal, handle_t *handle,
276 transaction_t *transaction, *new_transaction = NULL;
277 int blocks = handle->h_buffer_credits;
279 unsigned long ts = jiffies;
281 if (handle->h_rsv_handle)
282 rsv_blocks = handle->h_rsv_handle->h_buffer_credits;
285 * Limit the number of reserved credits to 1/2 of maximum transaction
286 * size and limit the number of total credits to not exceed maximum
287 * transaction size per operation.
289 if ((rsv_blocks > journal->j_max_transaction_buffers / 2) ||
290 (rsv_blocks + blocks > journal->j_max_transaction_buffers)) {
291 printk(KERN_ERR "JBD2: %s wants too many credits "
292 "credits:%d rsv_credits:%d max:%d\n",
293 current->comm, blocks, rsv_blocks,
294 journal->j_max_transaction_buffers);
300 if (!journal->j_running_transaction) {
302 * If __GFP_FS is not present, then we may be being called from
303 * inside the fs writeback layer, so we MUST NOT fail.
305 if ((gfp_mask & __GFP_FS) == 0)
306 gfp_mask |= __GFP_NOFAIL;
307 new_transaction = kmem_cache_zalloc(transaction_cache,
309 if (!new_transaction)
313 jbd_debug(3, "New handle %p going live.\n", handle);
316 * We need to hold j_state_lock until t_updates has been incremented,
317 * for proper journal barrier handling
320 read_lock(&journal->j_state_lock);
321 BUG_ON(journal->j_flags & JBD2_UNMOUNT);
322 if (is_journal_aborted(journal) ||
323 (journal->j_errno != 0 && !(journal->j_flags & JBD2_ACK_ERR))) {
324 read_unlock(&journal->j_state_lock);
325 jbd2_journal_free_transaction(new_transaction);
330 * Wait on the journal's transaction barrier if necessary. Specifically
331 * we allow reserved handles to proceed because otherwise commit could
332 * deadlock on page writeback not being able to complete.
334 if (!handle->h_reserved && journal->j_barrier_count) {
335 read_unlock(&journal->j_state_lock);
336 wait_event(journal->j_wait_transaction_locked,
337 journal->j_barrier_count == 0);
341 if (!journal->j_running_transaction) {
342 read_unlock(&journal->j_state_lock);
343 if (!new_transaction)
344 goto alloc_transaction;
345 write_lock(&journal->j_state_lock);
346 if (!journal->j_running_transaction &&
347 (handle->h_reserved || !journal->j_barrier_count)) {
348 jbd2_get_transaction(journal, new_transaction);
349 new_transaction = NULL;
351 write_unlock(&journal->j_state_lock);
355 transaction = journal->j_running_transaction;
357 if (!handle->h_reserved) {
358 /* We may have dropped j_state_lock - restart in that case */
359 if (add_transaction_credits(journal, blocks, rsv_blocks))
363 * We have handle reserved so we are allowed to join T_LOCKED
364 * transaction and we don't have to check for transaction size
367 sub_reserved_credits(journal, blocks);
368 handle->h_reserved = 0;
371 /* OK, account for the buffers that this operation expects to
372 * use and add the handle to the running transaction.
374 update_t_max_wait(transaction, ts);
375 handle->h_transaction = transaction;
376 handle->h_requested_credits = blocks;
377 handle->h_start_jiffies = jiffies;
378 atomic_inc(&transaction->t_updates);
379 atomic_inc(&transaction->t_handle_count);
380 jbd_debug(4, "Handle %p given %d credits (total %d, free %lu)\n",
382 atomic_read(&transaction->t_outstanding_credits),
383 jbd2_log_space_left(journal));
384 read_unlock(&journal->j_state_lock);
385 current->journal_info = handle;
387 rwsem_acquire_read(&journal->j_trans_commit_map, 0, 0, _THIS_IP_);
388 jbd2_journal_free_transaction(new_transaction);
390 * Ensure that no allocations done while the transaction is open are
391 * going to recurse back to the fs layer.
393 handle->saved_alloc_context = memalloc_nofs_save();
397 /* Allocate a new handle. This should probably be in a slab... */
398 static handle_t *new_handle(int nblocks)
400 handle_t *handle = jbd2_alloc_handle(GFP_NOFS);
403 handle->h_buffer_credits = nblocks;
409 handle_t *jbd2__journal_start(journal_t *journal, int nblocks, int rsv_blocks,
410 gfp_t gfp_mask, unsigned int type,
411 unsigned int line_no)
413 handle_t *handle = journal_current_handle();
417 return ERR_PTR(-EROFS);
420 J_ASSERT(handle->h_transaction->t_journal == journal);
425 handle = new_handle(nblocks);
427 return ERR_PTR(-ENOMEM);
429 handle_t *rsv_handle;
431 rsv_handle = new_handle(rsv_blocks);
433 jbd2_free_handle(handle);
434 return ERR_PTR(-ENOMEM);
436 rsv_handle->h_reserved = 1;
437 rsv_handle->h_journal = journal;
438 handle->h_rsv_handle = rsv_handle;
441 err = start_this_handle(journal, handle, gfp_mask);
443 if (handle->h_rsv_handle)
444 jbd2_free_handle(handle->h_rsv_handle);
445 jbd2_free_handle(handle);
448 handle->h_type = type;
449 handle->h_line_no = line_no;
450 trace_jbd2_handle_start(journal->j_fs_dev->bd_dev,
451 handle->h_transaction->t_tid, type,
456 EXPORT_SYMBOL(jbd2__journal_start);
460 * handle_t *jbd2_journal_start() - Obtain a new handle.
461 * @journal: Journal to start transaction on.
462 * @nblocks: number of block buffer we might modify
464 * We make sure that the transaction can guarantee at least nblocks of
465 * modified buffers in the log. We block until the log can guarantee
466 * that much space. Additionally, if rsv_blocks > 0, we also create another
467 * handle with rsv_blocks reserved blocks in the journal. This handle is
468 * is stored in h_rsv_handle. It is not attached to any particular transaction
469 * and thus doesn't block transaction commit. If the caller uses this reserved
470 * handle, it has to set h_rsv_handle to NULL as otherwise jbd2_journal_stop()
471 * on the parent handle will dispose the reserved one. Reserved handle has to
472 * be converted to a normal handle using jbd2_journal_start_reserved() before
475 * Return a pointer to a newly allocated handle, or an ERR_PTR() value
478 handle_t *jbd2_journal_start(journal_t *journal, int nblocks)
480 return jbd2__journal_start(journal, nblocks, 0, GFP_NOFS, 0, 0);
482 EXPORT_SYMBOL(jbd2_journal_start);
484 void jbd2_journal_free_reserved(handle_t *handle)
486 journal_t *journal = handle->h_journal;
488 WARN_ON(!handle->h_reserved);
489 sub_reserved_credits(journal, handle->h_buffer_credits);
490 jbd2_free_handle(handle);
492 EXPORT_SYMBOL(jbd2_journal_free_reserved);
495 * int jbd2_journal_start_reserved() - start reserved handle
496 * @handle: handle to start
497 * @type: for handle statistics
498 * @line_no: for handle statistics
500 * Start handle that has been previously reserved with jbd2_journal_reserve().
501 * This attaches @handle to the running transaction (or creates one if there's
502 * not transaction running). Unlike jbd2_journal_start() this function cannot
503 * block on journal commit, checkpointing, or similar stuff. It can block on
504 * memory allocation or frozen journal though.
506 * Return 0 on success, non-zero on error - handle is freed in that case.
508 int jbd2_journal_start_reserved(handle_t *handle, unsigned int type,
509 unsigned int line_no)
511 journal_t *journal = handle->h_journal;
514 if (WARN_ON(!handle->h_reserved)) {
515 /* Someone passed in normal handle? Just stop it. */
516 jbd2_journal_stop(handle);
520 * Usefulness of mixing of reserved and unreserved handles is
521 * questionable. So far nobody seems to need it so just error out.
523 if (WARN_ON(current->journal_info)) {
524 jbd2_journal_free_reserved(handle);
528 handle->h_journal = NULL;
530 * GFP_NOFS is here because callers are likely from writeback or
531 * similarly constrained call sites
533 ret = start_this_handle(journal, handle, GFP_NOFS);
535 handle->h_journal = journal;
536 jbd2_journal_free_reserved(handle);
539 handle->h_type = type;
540 handle->h_line_no = line_no;
543 EXPORT_SYMBOL(jbd2_journal_start_reserved);
546 * int jbd2_journal_extend() - extend buffer credits.
547 * @handle: handle to 'extend'
548 * @nblocks: nr blocks to try to extend by.
550 * Some transactions, such as large extends and truncates, can be done
551 * atomically all at once or in several stages. The operation requests
552 * a credit for a number of buffer modifications in advance, but can
553 * extend its credit if it needs more.
555 * jbd2_journal_extend tries to give the running handle more buffer credits.
556 * It does not guarantee that allocation - this is a best-effort only.
557 * The calling process MUST be able to deal cleanly with a failure to
560 * Return 0 on success, non-zero on failure.
562 * return code < 0 implies an error
563 * return code > 0 implies normal transaction-full status.
565 int jbd2_journal_extend(handle_t *handle, int nblocks)
567 transaction_t *transaction = handle->h_transaction;
572 if (is_handle_aborted(handle))
574 journal = transaction->t_journal;
578 read_lock(&journal->j_state_lock);
580 /* Don't extend a locked-down transaction! */
581 if (transaction->t_state != T_RUNNING) {
582 jbd_debug(3, "denied handle %p %d blocks: "
583 "transaction not running\n", handle, nblocks);
587 spin_lock(&transaction->t_handle_lock);
588 wanted = atomic_add_return(nblocks,
589 &transaction->t_outstanding_credits);
591 if (wanted > journal->j_max_transaction_buffers) {
592 jbd_debug(3, "denied handle %p %d blocks: "
593 "transaction too large\n", handle, nblocks);
594 atomic_sub(nblocks, &transaction->t_outstanding_credits);
598 if (wanted + (wanted >> JBD2_CONTROL_BLOCKS_SHIFT) >
599 jbd2_log_space_left(journal)) {
600 jbd_debug(3, "denied handle %p %d blocks: "
601 "insufficient log space\n", handle, nblocks);
602 atomic_sub(nblocks, &transaction->t_outstanding_credits);
606 trace_jbd2_handle_extend(journal->j_fs_dev->bd_dev,
608 handle->h_type, handle->h_line_no,
609 handle->h_buffer_credits,
612 handle->h_buffer_credits += nblocks;
613 handle->h_requested_credits += nblocks;
616 jbd_debug(3, "extended handle %p by %d\n", handle, nblocks);
618 spin_unlock(&transaction->t_handle_lock);
620 read_unlock(&journal->j_state_lock);
626 * int jbd2_journal_restart() - restart a handle .
627 * @handle: handle to restart
628 * @nblocks: nr credits requested
629 * @gfp_mask: memory allocation flags (for start_this_handle)
631 * Restart a handle for a multi-transaction filesystem
634 * If the jbd2_journal_extend() call above fails to grant new buffer credits
635 * to a running handle, a call to jbd2_journal_restart will commit the
636 * handle's transaction so far and reattach the handle to a new
637 * transaction capable of guaranteeing the requested number of
638 * credits. We preserve reserved handle if there's any attached to the
641 int jbd2__journal_restart(handle_t *handle, int nblocks, gfp_t gfp_mask)
643 transaction_t *transaction = handle->h_transaction;
646 int need_to_start, ret;
648 /* If we've had an abort of any type, don't even think about
649 * actually doing the restart! */
650 if (is_handle_aborted(handle))
652 journal = transaction->t_journal;
655 * First unlink the handle from its current transaction, and start the
658 J_ASSERT(atomic_read(&transaction->t_updates) > 0);
659 J_ASSERT(journal_current_handle() == handle);
661 read_lock(&journal->j_state_lock);
662 spin_lock(&transaction->t_handle_lock);
663 atomic_sub(handle->h_buffer_credits,
664 &transaction->t_outstanding_credits);
665 if (handle->h_rsv_handle) {
666 sub_reserved_credits(journal,
667 handle->h_rsv_handle->h_buffer_credits);
669 if (atomic_dec_and_test(&transaction->t_updates))
670 wake_up(&journal->j_wait_updates);
671 tid = transaction->t_tid;
672 spin_unlock(&transaction->t_handle_lock);
673 handle->h_transaction = NULL;
674 current->journal_info = NULL;
676 jbd_debug(2, "restarting handle %p\n", handle);
677 need_to_start = !tid_geq(journal->j_commit_request, tid);
678 read_unlock(&journal->j_state_lock);
680 jbd2_log_start_commit(journal, tid);
682 rwsem_release(&journal->j_trans_commit_map, 1, _THIS_IP_);
683 handle->h_buffer_credits = nblocks;
685 * Restore the original nofs context because the journal restart
686 * is basically the same thing as journal stop and start.
687 * start_this_handle will start a new nofs context.
689 memalloc_nofs_restore(handle->saved_alloc_context);
690 ret = start_this_handle(journal, handle, gfp_mask);
693 EXPORT_SYMBOL(jbd2__journal_restart);
696 int jbd2_journal_restart(handle_t *handle, int nblocks)
698 return jbd2__journal_restart(handle, nblocks, GFP_NOFS);
700 EXPORT_SYMBOL(jbd2_journal_restart);
703 * void jbd2_journal_lock_updates () - establish a transaction barrier.
704 * @journal: Journal to establish a barrier on.
706 * This locks out any further updates from being started, and blocks
707 * until all existing updates have completed, returning only once the
708 * journal is in a quiescent state with no updates running.
710 * The journal lock should not be held on entry.
712 void jbd2_journal_lock_updates(journal_t *journal)
716 jbd2_might_wait_for_commit(journal);
718 write_lock(&journal->j_state_lock);
719 ++journal->j_barrier_count;
721 /* Wait until there are no reserved handles */
722 if (atomic_read(&journal->j_reserved_credits)) {
723 write_unlock(&journal->j_state_lock);
724 wait_event(journal->j_wait_reserved,
725 atomic_read(&journal->j_reserved_credits) == 0);
726 write_lock(&journal->j_state_lock);
729 /* Wait until there are no running updates */
731 transaction_t *transaction = journal->j_running_transaction;
736 spin_lock(&transaction->t_handle_lock);
737 prepare_to_wait(&journal->j_wait_updates, &wait,
738 TASK_UNINTERRUPTIBLE);
739 if (!atomic_read(&transaction->t_updates)) {
740 spin_unlock(&transaction->t_handle_lock);
741 finish_wait(&journal->j_wait_updates, &wait);
744 spin_unlock(&transaction->t_handle_lock);
745 write_unlock(&journal->j_state_lock);
747 finish_wait(&journal->j_wait_updates, &wait);
748 write_lock(&journal->j_state_lock);
750 write_unlock(&journal->j_state_lock);
753 * We have now established a barrier against other normal updates, but
754 * we also need to barrier against other jbd2_journal_lock_updates() calls
755 * to make sure that we serialise special journal-locked operations
758 mutex_lock(&journal->j_barrier);
762 * void jbd2_journal_unlock_updates (journal_t* journal) - release barrier
763 * @journal: Journal to release the barrier on.
765 * Release a transaction barrier obtained with jbd2_journal_lock_updates().
767 * Should be called without the journal lock held.
769 void jbd2_journal_unlock_updates (journal_t *journal)
771 J_ASSERT(journal->j_barrier_count != 0);
773 mutex_unlock(&journal->j_barrier);
774 write_lock(&journal->j_state_lock);
775 --journal->j_barrier_count;
776 write_unlock(&journal->j_state_lock);
777 wake_up(&journal->j_wait_transaction_locked);
780 static void warn_dirty_buffer(struct buffer_head *bh)
783 "JBD2: Spotted dirty metadata buffer (dev = %pg, blocknr = %llu). "
784 "There's a risk of filesystem corruption in case of system "
786 bh->b_bdev, (unsigned long long)bh->b_blocknr);
789 /* Call t_frozen trigger and copy buffer data into jh->b_frozen_data. */
790 static void jbd2_freeze_jh_data(struct journal_head *jh)
795 struct buffer_head *bh = jh2bh(jh);
797 J_EXPECT_JH(jh, buffer_uptodate(bh), "Possible IO failure.\n");
799 offset = offset_in_page(bh->b_data);
800 source = kmap_atomic(page);
801 /* Fire data frozen trigger just before we copy the data */
802 jbd2_buffer_frozen_trigger(jh, source + offset, jh->b_triggers);
803 memcpy(jh->b_frozen_data, source + offset, bh->b_size);
804 kunmap_atomic(source);
807 * Now that the frozen data is saved off, we need to store any matching
810 jh->b_frozen_triggers = jh->b_triggers;
814 * If the buffer is already part of the current transaction, then there
815 * is nothing we need to do. If it is already part of a prior
816 * transaction which we are still committing to disk, then we need to
817 * make sure that we do not overwrite the old copy: we do copy-out to
818 * preserve the copy going to disk. We also account the buffer against
819 * the handle's metadata buffer credits (unless the buffer is already
820 * part of the transaction, that is).
824 do_get_write_access(handle_t *handle, struct journal_head *jh,
827 struct buffer_head *bh;
828 transaction_t *transaction = handle->h_transaction;
831 char *frozen_buffer = NULL;
832 unsigned long start_lock, time_lock;
834 if (is_handle_aborted(handle))
836 journal = transaction->t_journal;
838 jbd_debug(5, "journal_head %p, force_copy %d\n", jh, force_copy);
840 JBUFFER_TRACE(jh, "entry");
844 /* @@@ Need to check for errors here at some point. */
846 start_lock = jiffies;
848 jbd_lock_bh_state(bh);
850 /* If it takes too long to lock the buffer, trace it */
851 time_lock = jbd2_time_diff(start_lock, jiffies);
852 if (time_lock > HZ/10)
853 trace_jbd2_lock_buffer_stall(bh->b_bdev->bd_dev,
854 jiffies_to_msecs(time_lock));
856 /* We now hold the buffer lock so it is safe to query the buffer
857 * state. Is the buffer dirty?
859 * If so, there are two possibilities. The buffer may be
860 * non-journaled, and undergoing a quite legitimate writeback.
861 * Otherwise, it is journaled, and we don't expect dirty buffers
862 * in that state (the buffers should be marked JBD_Dirty
863 * instead.) So either the IO is being done under our own
864 * control and this is a bug, or it's a third party IO such as
865 * dump(8) (which may leave the buffer scheduled for read ---
866 * ie. locked but not dirty) or tune2fs (which may actually have
867 * the buffer dirtied, ugh.) */
869 if (buffer_dirty(bh)) {
871 * First question: is this buffer already part of the current
872 * transaction or the existing committing transaction?
874 if (jh->b_transaction) {
876 jh->b_transaction == transaction ||
878 journal->j_committing_transaction);
879 if (jh->b_next_transaction)
880 J_ASSERT_JH(jh, jh->b_next_transaction ==
882 warn_dirty_buffer(bh);
885 * In any case we need to clean the dirty flag and we must
886 * do it under the buffer lock to be sure we don't race
887 * with running write-out.
889 JBUFFER_TRACE(jh, "Journalling dirty buffer");
890 clear_buffer_dirty(bh);
891 set_buffer_jbddirty(bh);
897 if (is_handle_aborted(handle)) {
898 jbd_unlock_bh_state(bh);
904 * The buffer is already part of this transaction if b_transaction or
905 * b_next_transaction points to it
907 if (jh->b_transaction == transaction ||
908 jh->b_next_transaction == transaction)
912 * this is the first time this transaction is touching this buffer,
913 * reset the modified flag
918 * If the buffer is not journaled right now, we need to make sure it
919 * doesn't get written to disk before the caller actually commits the
922 if (!jh->b_transaction) {
923 JBUFFER_TRACE(jh, "no transaction");
924 J_ASSERT_JH(jh, !jh->b_next_transaction);
925 JBUFFER_TRACE(jh, "file as BJ_Reserved");
927 * Make sure all stores to jh (b_modified, b_frozen_data) are
928 * visible before attaching it to the running transaction.
929 * Paired with barrier in jbd2_write_access_granted()
932 spin_lock(&journal->j_list_lock);
933 __jbd2_journal_file_buffer(jh, transaction, BJ_Reserved);
934 spin_unlock(&journal->j_list_lock);
938 * If there is already a copy-out version of this buffer, then we don't
939 * need to make another one
941 if (jh->b_frozen_data) {
942 JBUFFER_TRACE(jh, "has frozen data");
943 J_ASSERT_JH(jh, jh->b_next_transaction == NULL);
947 JBUFFER_TRACE(jh, "owned by older transaction");
948 J_ASSERT_JH(jh, jh->b_next_transaction == NULL);
949 J_ASSERT_JH(jh, jh->b_transaction == journal->j_committing_transaction);
952 * There is one case we have to be very careful about. If the
953 * committing transaction is currently writing this buffer out to disk
954 * and has NOT made a copy-out, then we cannot modify the buffer
955 * contents at all right now. The essence of copy-out is that it is
956 * the extra copy, not the primary copy, which gets journaled. If the
957 * primary copy is already going to disk then we cannot do copy-out
960 if (buffer_shadow(bh)) {
961 JBUFFER_TRACE(jh, "on shadow: sleep");
962 jbd_unlock_bh_state(bh);
963 wait_on_bit_io(&bh->b_state, BH_Shadow, TASK_UNINTERRUPTIBLE);
968 * Only do the copy if the currently-owning transaction still needs it.
969 * If buffer isn't on BJ_Metadata list, the committing transaction is
970 * past that stage (here we use the fact that BH_Shadow is set under
971 * bh_state lock together with refiling to BJ_Shadow list and at this
972 * point we know the buffer doesn't have BH_Shadow set).
974 * Subtle point, though: if this is a get_undo_access, then we will be
975 * relying on the frozen_data to contain the new value of the
976 * committed_data record after the transaction, so we HAVE to force the
977 * frozen_data copy in that case.
979 if (jh->b_jlist == BJ_Metadata || force_copy) {
980 JBUFFER_TRACE(jh, "generate frozen data");
981 if (!frozen_buffer) {
982 JBUFFER_TRACE(jh, "allocate memory for buffer");
983 jbd_unlock_bh_state(bh);
984 frozen_buffer = jbd2_alloc(jh2bh(jh)->b_size,
985 GFP_NOFS | __GFP_NOFAIL);
988 jh->b_frozen_data = frozen_buffer;
989 frozen_buffer = NULL;
990 jbd2_freeze_jh_data(jh);
994 * Make sure all stores to jh (b_modified, b_frozen_data) are visible
995 * before attaching it to the running transaction. Paired with barrier
996 * in jbd2_write_access_granted()
999 jh->b_next_transaction = transaction;
1002 jbd_unlock_bh_state(bh);
1005 * If we are about to journal a buffer, then any revoke pending on it is
1008 jbd2_journal_cancel_revoke(handle, jh);
1011 if (unlikely(frozen_buffer)) /* It's usually NULL */
1012 jbd2_free(frozen_buffer, bh->b_size);
1014 JBUFFER_TRACE(jh, "exit");
1018 /* Fast check whether buffer is already attached to the required transaction */
1019 static bool jbd2_write_access_granted(handle_t *handle, struct buffer_head *bh,
1022 struct journal_head *jh;
1025 /* Dirty buffers require special handling... */
1026 if (buffer_dirty(bh))
1030 * RCU protects us from dereferencing freed pages. So the checks we do
1031 * are guaranteed not to oops. However the jh slab object can get freed
1032 * & reallocated while we work with it. So we have to be careful. When
1033 * we see jh attached to the running transaction, we know it must stay
1034 * so until the transaction is committed. Thus jh won't be freed and
1035 * will be attached to the same bh while we run. However it can
1036 * happen jh gets freed, reallocated, and attached to the transaction
1037 * just after we get pointer to it from bh. So we have to be careful
1038 * and recheck jh still belongs to our bh before we return success.
1041 if (!buffer_jbd(bh))
1043 /* This should be bh2jh() but that doesn't work with inline functions */
1044 jh = READ_ONCE(bh->b_private);
1047 /* For undo access buffer must have data copied */
1048 if (undo && !jh->b_committed_data)
1050 if (jh->b_transaction != handle->h_transaction &&
1051 jh->b_next_transaction != handle->h_transaction)
1054 * There are two reasons for the barrier here:
1055 * 1) Make sure to fetch b_bh after we did previous checks so that we
1056 * detect when jh went through free, realloc, attach to transaction
1057 * while we were checking. Paired with implicit barrier in that path.
1058 * 2) So that access to bh done after jbd2_write_access_granted()
1059 * doesn't get reordered and see inconsistent state of concurrent
1060 * do_get_write_access().
1063 if (unlikely(jh->b_bh != bh))
1072 * int jbd2_journal_get_write_access() - notify intent to modify a buffer for metadata (not data) update.
1073 * @handle: transaction to add buffer modifications to
1074 * @bh: bh to be used for metadata writes
1076 * Returns: error code or 0 on success.
1078 * In full data journalling mode the buffer may be of type BJ_AsyncData,
1079 * because we're ``write()ing`` a buffer which is also part of a shared mapping.
1082 int jbd2_journal_get_write_access(handle_t *handle, struct buffer_head *bh)
1084 struct journal_head *jh;
1087 if (jbd2_write_access_granted(handle, bh, false))
1090 jh = jbd2_journal_add_journal_head(bh);
1091 /* We do not want to get caught playing with fields which the
1092 * log thread also manipulates. Make sure that the buffer
1093 * completes any outstanding IO before proceeding. */
1094 rc = do_get_write_access(handle, jh, 0);
1095 jbd2_journal_put_journal_head(jh);
1101 * When the user wants to journal a newly created buffer_head
1102 * (ie. getblk() returned a new buffer and we are going to populate it
1103 * manually rather than reading off disk), then we need to keep the
1104 * buffer_head locked until it has been completely filled with new
1105 * data. In this case, we should be able to make the assertion that
1106 * the bh is not already part of an existing transaction.
1108 * The buffer should already be locked by the caller by this point.
1109 * There is no lock ranking violation: it was a newly created,
1110 * unlocked buffer beforehand. */
1113 * int jbd2_journal_get_create_access () - notify intent to use newly created bh
1114 * @handle: transaction to new buffer to
1117 * Call this if you create a new bh.
1119 int jbd2_journal_get_create_access(handle_t *handle, struct buffer_head *bh)
1121 transaction_t *transaction = handle->h_transaction;
1123 struct journal_head *jh = jbd2_journal_add_journal_head(bh);
1126 jbd_debug(5, "journal_head %p\n", jh);
1128 if (is_handle_aborted(handle))
1130 journal = transaction->t_journal;
1133 JBUFFER_TRACE(jh, "entry");
1135 * The buffer may already belong to this transaction due to pre-zeroing
1136 * in the filesystem's new_block code. It may also be on the previous,
1137 * committing transaction's lists, but it HAS to be in Forget state in
1138 * that case: the transaction must have deleted the buffer for it to be
1141 jbd_lock_bh_state(bh);
1142 J_ASSERT_JH(jh, (jh->b_transaction == transaction ||
1143 jh->b_transaction == NULL ||
1144 (jh->b_transaction == journal->j_committing_transaction &&
1145 jh->b_jlist == BJ_Forget)));
1147 J_ASSERT_JH(jh, jh->b_next_transaction == NULL);
1148 J_ASSERT_JH(jh, buffer_locked(jh2bh(jh)));
1150 if (jh->b_transaction == NULL) {
1152 * Previous jbd2_journal_forget() could have left the buffer
1153 * with jbddirty bit set because it was being committed. When
1154 * the commit finished, we've filed the buffer for
1155 * checkpointing and marked it dirty. Now we are reallocating
1156 * the buffer so the transaction freeing it must have
1157 * committed and so it's safe to clear the dirty bit.
1159 clear_buffer_dirty(jh2bh(jh));
1160 /* first access by this transaction */
1163 JBUFFER_TRACE(jh, "file as BJ_Reserved");
1164 spin_lock(&journal->j_list_lock);
1165 __jbd2_journal_file_buffer(jh, transaction, BJ_Reserved);
1166 spin_unlock(&journal->j_list_lock);
1167 } else if (jh->b_transaction == journal->j_committing_transaction) {
1168 /* first access by this transaction */
1171 JBUFFER_TRACE(jh, "set next transaction");
1172 spin_lock(&journal->j_list_lock);
1173 jh->b_next_transaction = transaction;
1174 spin_unlock(&journal->j_list_lock);
1176 jbd_unlock_bh_state(bh);
1179 * akpm: I added this. ext3_alloc_branch can pick up new indirect
1180 * blocks which contain freed but then revoked metadata. We need
1181 * to cancel the revoke in case we end up freeing it yet again
1182 * and the reallocating as data - this would cause a second revoke,
1183 * which hits an assertion error.
1185 JBUFFER_TRACE(jh, "cancelling revoke");
1186 jbd2_journal_cancel_revoke(handle, jh);
1188 jbd2_journal_put_journal_head(jh);
1193 * int jbd2_journal_get_undo_access() - Notify intent to modify metadata with
1194 * non-rewindable consequences
1195 * @handle: transaction
1196 * @bh: buffer to undo
1198 * Sometimes there is a need to distinguish between metadata which has
1199 * been committed to disk and that which has not. The ext3fs code uses
1200 * this for freeing and allocating space, we have to make sure that we
1201 * do not reuse freed space until the deallocation has been committed,
1202 * since if we overwrote that space we would make the delete
1203 * un-rewindable in case of a crash.
1205 * To deal with that, jbd2_journal_get_undo_access requests write access to a
1206 * buffer for parts of non-rewindable operations such as delete
1207 * operations on the bitmaps. The journaling code must keep a copy of
1208 * the buffer's contents prior to the undo_access call until such time
1209 * as we know that the buffer has definitely been committed to disk.
1211 * We never need to know which transaction the committed data is part
1212 * of, buffers touched here are guaranteed to be dirtied later and so
1213 * will be committed to a new transaction in due course, at which point
1214 * we can discard the old committed data pointer.
1216 * Returns error number or 0 on success.
1218 int jbd2_journal_get_undo_access(handle_t *handle, struct buffer_head *bh)
1221 struct journal_head *jh;
1222 char *committed_data = NULL;
1224 JBUFFER_TRACE(jh, "entry");
1225 if (jbd2_write_access_granted(handle, bh, true))
1228 jh = jbd2_journal_add_journal_head(bh);
1230 * Do this first --- it can drop the journal lock, so we want to
1231 * make sure that obtaining the committed_data is done
1232 * atomically wrt. completion of any outstanding commits.
1234 err = do_get_write_access(handle, jh, 1);
1239 if (!jh->b_committed_data)
1240 committed_data = jbd2_alloc(jh2bh(jh)->b_size,
1241 GFP_NOFS|__GFP_NOFAIL);
1243 jbd_lock_bh_state(bh);
1244 if (!jh->b_committed_data) {
1245 /* Copy out the current buffer contents into the
1246 * preserved, committed copy. */
1247 JBUFFER_TRACE(jh, "generate b_committed data");
1248 if (!committed_data) {
1249 jbd_unlock_bh_state(bh);
1253 jh->b_committed_data = committed_data;
1254 committed_data = NULL;
1255 memcpy(jh->b_committed_data, bh->b_data, bh->b_size);
1257 jbd_unlock_bh_state(bh);
1259 jbd2_journal_put_journal_head(jh);
1260 if (unlikely(committed_data))
1261 jbd2_free(committed_data, bh->b_size);
1266 * void jbd2_journal_set_triggers() - Add triggers for commit writeout
1267 * @bh: buffer to trigger on
1268 * @type: struct jbd2_buffer_trigger_type containing the trigger(s).
1270 * Set any triggers on this journal_head. This is always safe, because
1271 * triggers for a committing buffer will be saved off, and triggers for
1272 * a running transaction will match the buffer in that transaction.
1274 * Call with NULL to clear the triggers.
1276 void jbd2_journal_set_triggers(struct buffer_head *bh,
1277 struct jbd2_buffer_trigger_type *type)
1279 struct journal_head *jh = jbd2_journal_grab_journal_head(bh);
1283 jh->b_triggers = type;
1284 jbd2_journal_put_journal_head(jh);
1287 void jbd2_buffer_frozen_trigger(struct journal_head *jh, void *mapped_data,
1288 struct jbd2_buffer_trigger_type *triggers)
1290 struct buffer_head *bh = jh2bh(jh);
1292 if (!triggers || !triggers->t_frozen)
1295 triggers->t_frozen(triggers, bh, mapped_data, bh->b_size);
1298 void jbd2_buffer_abort_trigger(struct journal_head *jh,
1299 struct jbd2_buffer_trigger_type *triggers)
1301 if (!triggers || !triggers->t_abort)
1304 triggers->t_abort(triggers, jh2bh(jh));
1308 * int jbd2_journal_dirty_metadata() - mark a buffer as containing dirty metadata
1309 * @handle: transaction to add buffer to.
1310 * @bh: buffer to mark
1312 * mark dirty metadata which needs to be journaled as part of the current
1315 * The buffer must have previously had jbd2_journal_get_write_access()
1316 * called so that it has a valid journal_head attached to the buffer
1319 * The buffer is placed on the transaction's metadata list and is marked
1320 * as belonging to the transaction.
1322 * Returns error number or 0 on success.
1324 * Special care needs to be taken if the buffer already belongs to the
1325 * current committing transaction (in which case we should have frozen
1326 * data present for that commit). In that case, we don't relink the
1327 * buffer: that only gets done when the old transaction finally
1328 * completes its commit.
1330 int jbd2_journal_dirty_metadata(handle_t *handle, struct buffer_head *bh)
1332 transaction_t *transaction = handle->h_transaction;
1334 struct journal_head *jh;
1337 if (is_handle_aborted(handle))
1339 if (!buffer_jbd(bh)) {
1344 * We don't grab jh reference here since the buffer must be part
1345 * of the running transaction.
1349 * This and the following assertions are unreliable since we may see jh
1350 * in inconsistent state unless we grab bh_state lock. But this is
1351 * crucial to catch bugs so let's do a reliable check until the
1352 * lockless handling is fully proven.
1354 if (jh->b_transaction != transaction &&
1355 jh->b_next_transaction != transaction) {
1356 jbd_lock_bh_state(bh);
1357 J_ASSERT_JH(jh, jh->b_transaction == transaction ||
1358 jh->b_next_transaction == transaction);
1359 jbd_unlock_bh_state(bh);
1361 if (jh->b_modified == 1) {
1362 /* If it's in our transaction it must be in BJ_Metadata list. */
1363 if (jh->b_transaction == transaction &&
1364 jh->b_jlist != BJ_Metadata) {
1365 jbd_lock_bh_state(bh);
1366 J_ASSERT_JH(jh, jh->b_transaction != transaction ||
1367 jh->b_jlist == BJ_Metadata);
1368 jbd_unlock_bh_state(bh);
1373 journal = transaction->t_journal;
1374 jbd_debug(5, "journal_head %p\n", jh);
1375 JBUFFER_TRACE(jh, "entry");
1377 jbd_lock_bh_state(bh);
1379 if (jh->b_modified == 0) {
1381 * This buffer's got modified and becoming part
1382 * of the transaction. This needs to be done
1383 * once a transaction -bzzz
1386 if (handle->h_buffer_credits <= 0) {
1390 handle->h_buffer_credits--;
1394 * fastpath, to avoid expensive locking. If this buffer is already
1395 * on the running transaction's metadata list there is nothing to do.
1396 * Nobody can take it off again because there is a handle open.
1397 * I _think_ we're OK here with SMP barriers - a mistaken decision will
1398 * result in this test being false, so we go in and take the locks.
1400 if (jh->b_transaction == transaction && jh->b_jlist == BJ_Metadata) {
1401 JBUFFER_TRACE(jh, "fastpath");
1402 if (unlikely(jh->b_transaction !=
1403 journal->j_running_transaction)) {
1404 printk(KERN_ERR "JBD2: %s: "
1405 "jh->b_transaction (%llu, %p, %u) != "
1406 "journal->j_running_transaction (%p, %u)\n",
1408 (unsigned long long) bh->b_blocknr,
1410 jh->b_transaction ? jh->b_transaction->t_tid : 0,
1411 journal->j_running_transaction,
1412 journal->j_running_transaction ?
1413 journal->j_running_transaction->t_tid : 0);
1419 set_buffer_jbddirty(bh);
1422 * Metadata already on the current transaction list doesn't
1423 * need to be filed. Metadata on another transaction's list must
1424 * be committing, and will be refiled once the commit completes:
1425 * leave it alone for now.
1427 if (jh->b_transaction != transaction) {
1428 JBUFFER_TRACE(jh, "already on other transaction");
1429 if (unlikely(((jh->b_transaction !=
1430 journal->j_committing_transaction)) ||
1431 (jh->b_next_transaction != transaction))) {
1432 printk(KERN_ERR "jbd2_journal_dirty_metadata: %s: "
1433 "bad jh for block %llu: "
1434 "transaction (%p, %u), "
1435 "jh->b_transaction (%p, %u), "
1436 "jh->b_next_transaction (%p, %u), jlist %u\n",
1438 (unsigned long long) bh->b_blocknr,
1439 transaction, transaction->t_tid,
1442 jh->b_transaction->t_tid : 0,
1443 jh->b_next_transaction,
1444 jh->b_next_transaction ?
1445 jh->b_next_transaction->t_tid : 0,
1450 /* And this case is illegal: we can't reuse another
1451 * transaction's data buffer, ever. */
1455 /* That test should have eliminated the following case: */
1456 J_ASSERT_JH(jh, jh->b_frozen_data == NULL);
1458 JBUFFER_TRACE(jh, "file as BJ_Metadata");
1459 spin_lock(&journal->j_list_lock);
1460 __jbd2_journal_file_buffer(jh, transaction, BJ_Metadata);
1461 spin_unlock(&journal->j_list_lock);
1463 jbd_unlock_bh_state(bh);
1465 JBUFFER_TRACE(jh, "exit");
1470 * void jbd2_journal_forget() - bforget() for potentially-journaled buffers.
1471 * @handle: transaction handle
1472 * @bh: bh to 'forget'
1474 * We can only do the bforget if there are no commits pending against the
1475 * buffer. If the buffer is dirty in the current running transaction we
1476 * can safely unlink it.
1478 * bh may not be a journalled buffer at all - it may be a non-JBD
1479 * buffer which came off the hashtable. Check for this.
1481 * Decrements bh->b_count by one.
1483 * Allow this call even if the handle has aborted --- it may be part of
1484 * the caller's cleanup after an abort.
1486 int jbd2_journal_forget (handle_t *handle, struct buffer_head *bh)
1488 transaction_t *transaction = handle->h_transaction;
1490 struct journal_head *jh;
1491 int drop_reserve = 0;
1493 int was_modified = 0;
1495 if (is_handle_aborted(handle))
1497 journal = transaction->t_journal;
1499 BUFFER_TRACE(bh, "entry");
1501 jbd_lock_bh_state(bh);
1503 if (!buffer_jbd(bh))
1507 /* Critical error: attempting to delete a bitmap buffer, maybe?
1508 * Don't do any jbd operations, and return an error. */
1509 if (!J_EXPECT_JH(jh, !jh->b_committed_data,
1510 "inconsistent data on disk")) {
1515 /* keep track of whether or not this transaction modified us */
1516 was_modified = jh->b_modified;
1519 * The buffer's going from the transaction, we must drop
1520 * all references -bzzz
1524 if (jh->b_transaction == transaction) {
1525 J_ASSERT_JH(jh, !jh->b_frozen_data);
1527 /* If we are forgetting a buffer which is already part
1528 * of this transaction, then we can just drop it from
1529 * the transaction immediately. */
1530 clear_buffer_dirty(bh);
1531 clear_buffer_jbddirty(bh);
1533 JBUFFER_TRACE(jh, "belongs to current transaction: unfile");
1536 * we only want to drop a reference if this transaction
1537 * modified the buffer
1543 * We are no longer going to journal this buffer.
1544 * However, the commit of this transaction is still
1545 * important to the buffer: the delete that we are now
1546 * processing might obsolete an old log entry, so by
1547 * committing, we can satisfy the buffer's checkpoint.
1549 * So, if we have a checkpoint on the buffer, we should
1550 * now refile the buffer on our BJ_Forget list so that
1551 * we know to remove the checkpoint after we commit.
1554 spin_lock(&journal->j_list_lock);
1555 if (jh->b_cp_transaction) {
1556 __jbd2_journal_temp_unlink_buffer(jh);
1557 __jbd2_journal_file_buffer(jh, transaction, BJ_Forget);
1559 __jbd2_journal_unfile_buffer(jh);
1560 if (!buffer_jbd(bh)) {
1561 spin_unlock(&journal->j_list_lock);
1562 jbd_unlock_bh_state(bh);
1567 spin_unlock(&journal->j_list_lock);
1568 } else if (jh->b_transaction) {
1569 J_ASSERT_JH(jh, (jh->b_transaction ==
1570 journal->j_committing_transaction));
1571 /* However, if the buffer is still owned by a prior
1572 * (committing) transaction, we can't drop it yet... */
1573 JBUFFER_TRACE(jh, "belongs to older transaction");
1574 /* ... but we CAN drop it from the new transaction if we
1575 * have also modified it since the original commit. */
1577 if (jh->b_next_transaction) {
1578 J_ASSERT(jh->b_next_transaction == transaction);
1579 spin_lock(&journal->j_list_lock);
1580 jh->b_next_transaction = NULL;
1581 spin_unlock(&journal->j_list_lock);
1584 * only drop a reference if this transaction modified
1593 jbd_unlock_bh_state(bh);
1597 /* no need to reserve log space for this block -bzzz */
1598 handle->h_buffer_credits++;
1604 * int jbd2_journal_stop() - complete a transaction
1605 * @handle: transaction to complete.
1607 * All done for a particular handle.
1609 * There is not much action needed here. We just return any remaining
1610 * buffer credits to the transaction and remove the handle. The only
1611 * complication is that we need to start a commit operation if the
1612 * filesystem is marked for synchronous update.
1614 * jbd2_journal_stop itself will not usually return an error, but it may
1615 * do so in unusual circumstances. In particular, expect it to
1616 * return -EIO if a jbd2_journal_abort has been executed since the
1617 * transaction began.
1619 int jbd2_journal_stop(handle_t *handle)
1621 transaction_t *transaction = handle->h_transaction;
1623 int err = 0, wait_for_commit = 0;
1629 * Handle is already detached from the transaction so
1630 * there is nothing to do other than decrease a refcount,
1631 * or free the handle if refcount drops to zero
1633 if (--handle->h_ref > 0) {
1634 jbd_debug(4, "h_ref %d -> %d\n", handle->h_ref + 1,
1638 if (handle->h_rsv_handle)
1639 jbd2_free_handle(handle->h_rsv_handle);
1643 journal = transaction->t_journal;
1645 J_ASSERT(journal_current_handle() == handle);
1647 if (is_handle_aborted(handle))
1650 J_ASSERT(atomic_read(&transaction->t_updates) > 0);
1652 if (--handle->h_ref > 0) {
1653 jbd_debug(4, "h_ref %d -> %d\n", handle->h_ref + 1,
1658 jbd_debug(4, "Handle %p going down\n", handle);
1659 trace_jbd2_handle_stats(journal->j_fs_dev->bd_dev,
1661 handle->h_type, handle->h_line_no,
1662 jiffies - handle->h_start_jiffies,
1663 handle->h_sync, handle->h_requested_credits,
1664 (handle->h_requested_credits -
1665 handle->h_buffer_credits));
1668 * Implement synchronous transaction batching. If the handle
1669 * was synchronous, don't force a commit immediately. Let's
1670 * yield and let another thread piggyback onto this
1671 * transaction. Keep doing that while new threads continue to
1672 * arrive. It doesn't cost much - we're about to run a commit
1673 * and sleep on IO anyway. Speeds up many-threaded, many-dir
1674 * operations by 30x or more...
1676 * We try and optimize the sleep time against what the
1677 * underlying disk can do, instead of having a static sleep
1678 * time. This is useful for the case where our storage is so
1679 * fast that it is more optimal to go ahead and force a flush
1680 * and wait for the transaction to be committed than it is to
1681 * wait for an arbitrary amount of time for new writers to
1682 * join the transaction. We achieve this by measuring how
1683 * long it takes to commit a transaction, and compare it with
1684 * how long this transaction has been running, and if run time
1685 * < commit time then we sleep for the delta and commit. This
1686 * greatly helps super fast disks that would see slowdowns as
1687 * more threads started doing fsyncs.
1689 * But don't do this if this process was the most recent one
1690 * to perform a synchronous write. We do this to detect the
1691 * case where a single process is doing a stream of sync
1692 * writes. No point in waiting for joiners in that case.
1694 * Setting max_batch_time to 0 disables this completely.
1697 if (handle->h_sync && journal->j_last_sync_writer != pid &&
1698 journal->j_max_batch_time) {
1699 u64 commit_time, trans_time;
1701 journal->j_last_sync_writer = pid;
1703 read_lock(&journal->j_state_lock);
1704 commit_time = journal->j_average_commit_time;
1705 read_unlock(&journal->j_state_lock);
1707 trans_time = ktime_to_ns(ktime_sub(ktime_get(),
1708 transaction->t_start_time));
1710 commit_time = max_t(u64, commit_time,
1711 1000*journal->j_min_batch_time);
1712 commit_time = min_t(u64, commit_time,
1713 1000*journal->j_max_batch_time);
1715 if (trans_time < commit_time) {
1716 ktime_t expires = ktime_add_ns(ktime_get(),
1718 set_current_state(TASK_UNINTERRUPTIBLE);
1719 schedule_hrtimeout(&expires, HRTIMER_MODE_ABS);
1724 transaction->t_synchronous_commit = 1;
1725 current->journal_info = NULL;
1726 atomic_sub(handle->h_buffer_credits,
1727 &transaction->t_outstanding_credits);
1730 * If the handle is marked SYNC, we need to set another commit
1731 * going! We also want to force a commit if the current
1732 * transaction is occupying too much of the log, or if the
1733 * transaction is too old now.
1735 if (handle->h_sync ||
1736 (atomic_read(&transaction->t_outstanding_credits) >
1737 journal->j_max_transaction_buffers) ||
1738 time_after_eq(jiffies, transaction->t_expires)) {
1739 /* Do this even for aborted journals: an abort still
1740 * completes the commit thread, it just doesn't write
1741 * anything to disk. */
1743 jbd_debug(2, "transaction too old, requesting commit for "
1744 "handle %p\n", handle);
1745 /* This is non-blocking */
1746 jbd2_log_start_commit(journal, transaction->t_tid);
1749 * Special case: JBD2_SYNC synchronous updates require us
1750 * to wait for the commit to complete.
1752 if (handle->h_sync && !(current->flags & PF_MEMALLOC))
1753 wait_for_commit = 1;
1757 * Once we drop t_updates, if it goes to zero the transaction
1758 * could start committing on us and eventually disappear. So
1759 * once we do this, we must not dereference transaction
1762 tid = transaction->t_tid;
1763 if (atomic_dec_and_test(&transaction->t_updates)) {
1764 wake_up(&journal->j_wait_updates);
1765 if (journal->j_barrier_count)
1766 wake_up(&journal->j_wait_transaction_locked);
1769 rwsem_release(&journal->j_trans_commit_map, 1, _THIS_IP_);
1771 if (wait_for_commit)
1772 err = jbd2_log_wait_commit(journal, tid);
1774 if (handle->h_rsv_handle)
1775 jbd2_journal_free_reserved(handle->h_rsv_handle);
1778 * Scope of the GFP_NOFS context is over here and so we can restore the
1779 * original alloc context.
1781 memalloc_nofs_restore(handle->saved_alloc_context);
1782 jbd2_free_handle(handle);
1788 * List management code snippets: various functions for manipulating the
1789 * transaction buffer lists.
1794 * Append a buffer to a transaction list, given the transaction's list head
1797 * j_list_lock is held.
1799 * jbd_lock_bh_state(jh2bh(jh)) is held.
1803 __blist_add_buffer(struct journal_head **list, struct journal_head *jh)
1806 jh->b_tnext = jh->b_tprev = jh;
1809 /* Insert at the tail of the list to preserve order */
1810 struct journal_head *first = *list, *last = first->b_tprev;
1812 jh->b_tnext = first;
1813 last->b_tnext = first->b_tprev = jh;
1818 * Remove a buffer from a transaction list, given the transaction's list
1821 * Called with j_list_lock held, and the journal may not be locked.
1823 * jbd_lock_bh_state(jh2bh(jh)) is held.
1827 __blist_del_buffer(struct journal_head **list, struct journal_head *jh)
1830 *list = jh->b_tnext;
1834 jh->b_tprev->b_tnext = jh->b_tnext;
1835 jh->b_tnext->b_tprev = jh->b_tprev;
1839 * Remove a buffer from the appropriate transaction list.
1841 * Note that this function can *change* the value of
1842 * bh->b_transaction->t_buffers, t_forget, t_shadow_list, t_log_list or
1843 * t_reserved_list. If the caller is holding onto a copy of one of these
1844 * pointers, it could go bad. Generally the caller needs to re-read the
1845 * pointer from the transaction_t.
1847 * Called under j_list_lock.
1849 static void __jbd2_journal_temp_unlink_buffer(struct journal_head *jh)
1851 struct journal_head **list = NULL;
1852 transaction_t *transaction;
1853 struct buffer_head *bh = jh2bh(jh);
1855 J_ASSERT_JH(jh, jbd_is_locked_bh_state(bh));
1856 transaction = jh->b_transaction;
1858 assert_spin_locked(&transaction->t_journal->j_list_lock);
1860 J_ASSERT_JH(jh, jh->b_jlist < BJ_Types);
1861 if (jh->b_jlist != BJ_None)
1862 J_ASSERT_JH(jh, transaction != NULL);
1864 switch (jh->b_jlist) {
1868 transaction->t_nr_buffers--;
1869 J_ASSERT_JH(jh, transaction->t_nr_buffers >= 0);
1870 list = &transaction->t_buffers;
1873 list = &transaction->t_forget;
1876 list = &transaction->t_shadow_list;
1879 list = &transaction->t_reserved_list;
1883 __blist_del_buffer(list, jh);
1884 jh->b_jlist = BJ_None;
1885 if (transaction && is_journal_aborted(transaction->t_journal))
1886 clear_buffer_jbddirty(bh);
1887 else if (test_clear_buffer_jbddirty(bh))
1888 mark_buffer_dirty(bh); /* Expose it to the VM */
1892 * Remove buffer from all transactions.
1894 * Called with bh_state lock and j_list_lock
1896 * jh and bh may be already freed when this function returns.
1898 static void __jbd2_journal_unfile_buffer(struct journal_head *jh)
1900 __jbd2_journal_temp_unlink_buffer(jh);
1901 jh->b_transaction = NULL;
1902 jbd2_journal_put_journal_head(jh);
1905 void jbd2_journal_unfile_buffer(journal_t *journal, struct journal_head *jh)
1907 struct buffer_head *bh = jh2bh(jh);
1909 /* Get reference so that buffer cannot be freed before we unlock it */
1911 jbd_lock_bh_state(bh);
1912 spin_lock(&journal->j_list_lock);
1913 __jbd2_journal_unfile_buffer(jh);
1914 spin_unlock(&journal->j_list_lock);
1915 jbd_unlock_bh_state(bh);
1920 * Called from jbd2_journal_try_to_free_buffers().
1922 * Called under jbd_lock_bh_state(bh)
1925 __journal_try_to_free_buffer(journal_t *journal, struct buffer_head *bh)
1927 struct journal_head *jh;
1931 if (buffer_locked(bh) || buffer_dirty(bh))
1934 if (jh->b_next_transaction != NULL || jh->b_transaction != NULL)
1937 spin_lock(&journal->j_list_lock);
1938 if (jh->b_cp_transaction != NULL) {
1939 /* written-back checkpointed metadata buffer */
1940 JBUFFER_TRACE(jh, "remove from checkpoint list");
1941 __jbd2_journal_remove_checkpoint(jh);
1943 spin_unlock(&journal->j_list_lock);
1949 * int jbd2_journal_try_to_free_buffers() - try to free page buffers.
1950 * @journal: journal for operation
1951 * @page: to try and free
1952 * @gfp_mask: we use the mask to detect how hard should we try to release
1953 * buffers. If __GFP_DIRECT_RECLAIM and __GFP_FS is set, we wait for commit
1954 * code to release the buffers.
1957 * For all the buffers on this page,
1958 * if they are fully written out ordered data, move them onto BUF_CLEAN
1959 * so try_to_free_buffers() can reap them.
1961 * This function returns non-zero if we wish try_to_free_buffers()
1962 * to be called. We do this if the page is releasable by try_to_free_buffers().
1963 * We also do it if the page has locked or dirty buffers and the caller wants
1964 * us to perform sync or async writeout.
1966 * This complicates JBD locking somewhat. We aren't protected by the
1967 * BKL here. We wish to remove the buffer from its committing or
1968 * running transaction's ->t_datalist via __jbd2_journal_unfile_buffer.
1970 * This may *change* the value of transaction_t->t_datalist, so anyone
1971 * who looks at t_datalist needs to lock against this function.
1973 * Even worse, someone may be doing a jbd2_journal_dirty_data on this
1974 * buffer. So we need to lock against that. jbd2_journal_dirty_data()
1975 * will come out of the lock with the buffer dirty, which makes it
1976 * ineligible for release here.
1978 * Who else is affected by this? hmm... Really the only contender
1979 * is do_get_write_access() - it could be looking at the buffer while
1980 * journal_try_to_free_buffer() is changing its state. But that
1981 * cannot happen because we never reallocate freed data as metadata
1982 * while the data is part of a transaction. Yes?
1984 * Return 0 on failure, 1 on success
1986 int jbd2_journal_try_to_free_buffers(journal_t *journal,
1987 struct page *page, gfp_t gfp_mask)
1989 struct buffer_head *head;
1990 struct buffer_head *bh;
1993 J_ASSERT(PageLocked(page));
1995 head = page_buffers(page);
1998 struct journal_head *jh;
2001 * We take our own ref against the journal_head here to avoid
2002 * having to add tons of locking around each instance of
2003 * jbd2_journal_put_journal_head().
2005 jh = jbd2_journal_grab_journal_head(bh);
2009 jbd_lock_bh_state(bh);
2010 __journal_try_to_free_buffer(journal, bh);
2011 jbd2_journal_put_journal_head(jh);
2012 jbd_unlock_bh_state(bh);
2015 } while ((bh = bh->b_this_page) != head);
2017 ret = try_to_free_buffers(page);
2024 * This buffer is no longer needed. If it is on an older transaction's
2025 * checkpoint list we need to record it on this transaction's forget list
2026 * to pin this buffer (and hence its checkpointing transaction) down until
2027 * this transaction commits. If the buffer isn't on a checkpoint list, we
2029 * Returns non-zero if JBD no longer has an interest in the buffer.
2031 * Called under j_list_lock.
2033 * Called under jbd_lock_bh_state(bh).
2035 static int __dispose_buffer(struct journal_head *jh, transaction_t *transaction)
2038 struct buffer_head *bh = jh2bh(jh);
2040 if (jh->b_cp_transaction) {
2041 JBUFFER_TRACE(jh, "on running+cp transaction");
2042 __jbd2_journal_temp_unlink_buffer(jh);
2044 * We don't want to write the buffer anymore, clear the
2045 * bit so that we don't confuse checks in
2046 * __journal_file_buffer
2048 clear_buffer_dirty(bh);
2049 __jbd2_journal_file_buffer(jh, transaction, BJ_Forget);
2052 JBUFFER_TRACE(jh, "on running transaction");
2053 __jbd2_journal_unfile_buffer(jh);
2059 * jbd2_journal_invalidatepage
2061 * This code is tricky. It has a number of cases to deal with.
2063 * There are two invariants which this code relies on:
2065 * i_size must be updated on disk before we start calling invalidatepage on the
2068 * This is done in ext3 by defining an ext3_setattr method which
2069 * updates i_size before truncate gets going. By maintaining this
2070 * invariant, we can be sure that it is safe to throw away any buffers
2071 * attached to the current transaction: once the transaction commits,
2072 * we know that the data will not be needed.
2074 * Note however that we can *not* throw away data belonging to the
2075 * previous, committing transaction!
2077 * Any disk blocks which *are* part of the previous, committing
2078 * transaction (and which therefore cannot be discarded immediately) are
2079 * not going to be reused in the new running transaction
2081 * The bitmap committed_data images guarantee this: any block which is
2082 * allocated in one transaction and removed in the next will be marked
2083 * as in-use in the committed_data bitmap, so cannot be reused until
2084 * the next transaction to delete the block commits. This means that
2085 * leaving committing buffers dirty is quite safe: the disk blocks
2086 * cannot be reallocated to a different file and so buffer aliasing is
2090 * The above applies mainly to ordered data mode. In writeback mode we
2091 * don't make guarantees about the order in which data hits disk --- in
2092 * particular we don't guarantee that new dirty data is flushed before
2093 * transaction commit --- so it is always safe just to discard data
2094 * immediately in that mode. --sct
2098 * The journal_unmap_buffer helper function returns zero if the buffer
2099 * concerned remains pinned as an anonymous buffer belonging to an older
2102 * We're outside-transaction here. Either or both of j_running_transaction
2103 * and j_committing_transaction may be NULL.
2105 static int journal_unmap_buffer(journal_t *journal, struct buffer_head *bh,
2108 transaction_t *transaction;
2109 struct journal_head *jh;
2112 BUFFER_TRACE(bh, "entry");
2115 * It is safe to proceed here without the j_list_lock because the
2116 * buffers cannot be stolen by try_to_free_buffers as long as we are
2117 * holding the page lock. --sct
2120 if (!buffer_jbd(bh))
2121 goto zap_buffer_unlocked;
2123 /* OK, we have data buffer in journaled mode */
2124 write_lock(&journal->j_state_lock);
2125 jbd_lock_bh_state(bh);
2126 spin_lock(&journal->j_list_lock);
2128 jh = jbd2_journal_grab_journal_head(bh);
2130 goto zap_buffer_no_jh;
2133 * We cannot remove the buffer from checkpoint lists until the
2134 * transaction adding inode to orphan list (let's call it T)
2135 * is committed. Otherwise if the transaction changing the
2136 * buffer would be cleaned from the journal before T is
2137 * committed, a crash will cause that the correct contents of
2138 * the buffer will be lost. On the other hand we have to
2139 * clear the buffer dirty bit at latest at the moment when the
2140 * transaction marking the buffer as freed in the filesystem
2141 * structures is committed because from that moment on the
2142 * block can be reallocated and used by a different page.
2143 * Since the block hasn't been freed yet but the inode has
2144 * already been added to orphan list, it is safe for us to add
2145 * the buffer to BJ_Forget list of the newest transaction.
2147 * Also we have to clear buffer_mapped flag of a truncated buffer
2148 * because the buffer_head may be attached to the page straddling
2149 * i_size (can happen only when blocksize < pagesize) and thus the
2150 * buffer_head can be reused when the file is extended again. So we end
2151 * up keeping around invalidated buffers attached to transactions'
2152 * BJ_Forget list just to stop checkpointing code from cleaning up
2153 * the transaction this buffer was modified in.
2155 transaction = jh->b_transaction;
2156 if (transaction == NULL) {
2157 /* First case: not on any transaction. If it
2158 * has no checkpoint link, then we can zap it:
2159 * it's a writeback-mode buffer so we don't care
2160 * if it hits disk safely. */
2161 if (!jh->b_cp_transaction) {
2162 JBUFFER_TRACE(jh, "not on any transaction: zap");
2166 if (!buffer_dirty(bh)) {
2167 /* bdflush has written it. We can drop it now */
2168 __jbd2_journal_remove_checkpoint(jh);
2172 /* OK, it must be in the journal but still not
2173 * written fully to disk: it's metadata or
2174 * journaled data... */
2176 if (journal->j_running_transaction) {
2177 /* ... and once the current transaction has
2178 * committed, the buffer won't be needed any
2180 JBUFFER_TRACE(jh, "checkpointed: add to BJ_Forget");
2181 may_free = __dispose_buffer(jh,
2182 journal->j_running_transaction);
2185 /* There is no currently-running transaction. So the
2186 * orphan record which we wrote for this file must have
2187 * passed into commit. We must attach this buffer to
2188 * the committing transaction, if it exists. */
2189 if (journal->j_committing_transaction) {
2190 JBUFFER_TRACE(jh, "give to committing trans");
2191 may_free = __dispose_buffer(jh,
2192 journal->j_committing_transaction);
2195 /* The orphan record's transaction has
2196 * committed. We can cleanse this buffer */
2197 clear_buffer_jbddirty(bh);
2198 __jbd2_journal_remove_checkpoint(jh);
2202 } else if (transaction == journal->j_committing_transaction) {
2203 JBUFFER_TRACE(jh, "on committing transaction");
2205 * The buffer is committing, we simply cannot touch
2206 * it. If the page is straddling i_size we have to wait
2207 * for commit and try again.
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);
2217 * OK, buffer won't be reachable after truncate. We just set
2218 * j_next_transaction to the running transaction (if there is
2219 * one) and mark buffer as freed so that commit code knows it
2220 * should clear dirty bits when it is done with the buffer.
2222 set_buffer_freed(bh);
2223 if (journal->j_running_transaction && buffer_jbddirty(bh))
2224 jh->b_next_transaction = journal->j_running_transaction;
2225 jbd2_journal_put_journal_head(jh);
2226 spin_unlock(&journal->j_list_lock);
2227 jbd_unlock_bh_state(bh);
2228 write_unlock(&journal->j_state_lock);
2231 /* Good, the buffer belongs to the running transaction.
2232 * We are writing our own transaction's data, not any
2233 * previous one's, so it is safe to throw it away
2234 * (remember that we expect the filesystem to have set
2235 * i_size already for this truncate so recovery will not
2236 * expose the disk blocks we are discarding here.) */
2237 J_ASSERT_JH(jh, transaction == journal->j_running_transaction);
2238 JBUFFER_TRACE(jh, "on running transaction");
2239 may_free = __dispose_buffer(jh, transaction);
2244 * This is tricky. Although the buffer is truncated, it may be reused
2245 * if blocksize < pagesize and it is attached to the page straddling
2246 * EOF. Since the buffer might have been added to BJ_Forget list of the
2247 * running transaction, journal_get_write_access() won't clear
2248 * b_modified and credit accounting gets confused. So clear b_modified
2252 jbd2_journal_put_journal_head(jh);
2254 spin_unlock(&journal->j_list_lock);
2255 jbd_unlock_bh_state(bh);
2256 write_unlock(&journal->j_state_lock);
2257 zap_buffer_unlocked:
2258 clear_buffer_dirty(bh);
2259 J_ASSERT_BH(bh, !buffer_jbddirty(bh));
2260 clear_buffer_mapped(bh);
2261 clear_buffer_req(bh);
2262 clear_buffer_new(bh);
2263 clear_buffer_delay(bh);
2264 clear_buffer_unwritten(bh);
2270 * void jbd2_journal_invalidatepage()
2271 * @journal: journal to use for flush...
2272 * @page: page to flush
2273 * @offset: start of the range to invalidate
2274 * @length: length of the range to invalidate
2276 * Reap page buffers containing data after in the specified range in page.
2277 * Can return -EBUSY if buffers are part of the committing transaction and
2278 * the page is straddling i_size. Caller then has to wait for current commit
2281 int jbd2_journal_invalidatepage(journal_t *journal,
2283 unsigned int offset,
2284 unsigned int length)
2286 struct buffer_head *head, *bh, *next;
2287 unsigned int stop = offset + length;
2288 unsigned int curr_off = 0;
2289 int partial_page = (offset || length < PAGE_SIZE);
2293 if (!PageLocked(page))
2295 if (!page_has_buffers(page))
2298 BUG_ON(stop > PAGE_SIZE || stop < length);
2300 /* We will potentially be playing with lists other than just the
2301 * data lists (especially for journaled data mode), so be
2302 * cautious in our locking. */
2304 head = bh = page_buffers(page);
2306 unsigned int next_off = curr_off + bh->b_size;
2307 next = bh->b_this_page;
2309 if (next_off > stop)
2312 if (offset <= curr_off) {
2313 /* This block is wholly outside the truncation point */
2315 ret = journal_unmap_buffer(journal, bh, partial_page);
2321 curr_off = next_off;
2324 } while (bh != head);
2326 if (!partial_page) {
2327 if (may_free && try_to_free_buffers(page))
2328 J_ASSERT(!page_has_buffers(page));
2334 * File a buffer on the given transaction list.
2336 void __jbd2_journal_file_buffer(struct journal_head *jh,
2337 transaction_t *transaction, int jlist)
2339 struct journal_head **list = NULL;
2341 struct buffer_head *bh = jh2bh(jh);
2343 J_ASSERT_JH(jh, jbd_is_locked_bh_state(bh));
2344 assert_spin_locked(&transaction->t_journal->j_list_lock);
2346 J_ASSERT_JH(jh, jh->b_jlist < BJ_Types);
2347 J_ASSERT_JH(jh, jh->b_transaction == transaction ||
2348 jh->b_transaction == NULL);
2350 if (jh->b_transaction && jh->b_jlist == jlist)
2353 if (jlist == BJ_Metadata || jlist == BJ_Reserved ||
2354 jlist == BJ_Shadow || jlist == BJ_Forget) {
2356 * For metadata buffers, we track dirty bit in buffer_jbddirty
2357 * instead of buffer_dirty. We should not see a dirty bit set
2358 * here because we clear it in do_get_write_access but e.g.
2359 * tune2fs can modify the sb and set the dirty bit at any time
2360 * so we try to gracefully handle that.
2362 if (buffer_dirty(bh))
2363 warn_dirty_buffer(bh);
2364 if (test_clear_buffer_dirty(bh) ||
2365 test_clear_buffer_jbddirty(bh))
2369 if (jh->b_transaction)
2370 __jbd2_journal_temp_unlink_buffer(jh);
2372 jbd2_journal_grab_journal_head(bh);
2373 jh->b_transaction = transaction;
2377 J_ASSERT_JH(jh, !jh->b_committed_data);
2378 J_ASSERT_JH(jh, !jh->b_frozen_data);
2381 transaction->t_nr_buffers++;
2382 list = &transaction->t_buffers;
2385 list = &transaction->t_forget;
2388 list = &transaction->t_shadow_list;
2391 list = &transaction->t_reserved_list;
2395 __blist_add_buffer(list, jh);
2396 jh->b_jlist = jlist;
2399 set_buffer_jbddirty(bh);
2402 void jbd2_journal_file_buffer(struct journal_head *jh,
2403 transaction_t *transaction, int jlist)
2405 jbd_lock_bh_state(jh2bh(jh));
2406 spin_lock(&transaction->t_journal->j_list_lock);
2407 __jbd2_journal_file_buffer(jh, transaction, jlist);
2408 spin_unlock(&transaction->t_journal->j_list_lock);
2409 jbd_unlock_bh_state(jh2bh(jh));
2413 * Remove a buffer from its current buffer list in preparation for
2414 * dropping it from its current transaction entirely. If the buffer has
2415 * already started to be used by a subsequent transaction, refile the
2416 * buffer on that transaction's metadata list.
2418 * Called under j_list_lock
2419 * Called under jbd_lock_bh_state(jh2bh(jh))
2421 * jh and bh may be already free when this function returns
2423 void __jbd2_journal_refile_buffer(struct journal_head *jh)
2425 int was_dirty, jlist;
2426 struct buffer_head *bh = jh2bh(jh);
2428 J_ASSERT_JH(jh, jbd_is_locked_bh_state(bh));
2429 if (jh->b_transaction)
2430 assert_spin_locked(&jh->b_transaction->t_journal->j_list_lock);
2432 /* If the buffer is now unused, just drop it. */
2433 if (jh->b_next_transaction == NULL) {
2434 __jbd2_journal_unfile_buffer(jh);
2439 * It has been modified by a later transaction: add it to the new
2440 * transaction's metadata list.
2443 was_dirty = test_clear_buffer_jbddirty(bh);
2444 __jbd2_journal_temp_unlink_buffer(jh);
2446 * We set b_transaction here because b_next_transaction will inherit
2447 * our jh reference and thus __jbd2_journal_file_buffer() must not
2450 jh->b_transaction = jh->b_next_transaction;
2451 jh->b_next_transaction = NULL;
2452 if (buffer_freed(bh))
2454 else if (jh->b_modified)
2455 jlist = BJ_Metadata;
2457 jlist = BJ_Reserved;
2458 __jbd2_journal_file_buffer(jh, jh->b_transaction, jlist);
2459 J_ASSERT_JH(jh, jh->b_transaction->t_state == T_RUNNING);
2462 set_buffer_jbddirty(bh);
2466 * __jbd2_journal_refile_buffer() with necessary locking added. We take our
2467 * bh reference so that we can safely unlock bh.
2469 * The jh and bh may be freed by this call.
2471 void jbd2_journal_refile_buffer(journal_t *journal, struct journal_head *jh)
2473 struct buffer_head *bh = jh2bh(jh);
2475 /* Get reference so that buffer cannot be freed before we unlock it */
2477 jbd_lock_bh_state(bh);
2478 spin_lock(&journal->j_list_lock);
2479 __jbd2_journal_refile_buffer(jh);
2480 jbd_unlock_bh_state(bh);
2481 spin_unlock(&journal->j_list_lock);
2486 * File inode in the inode list of the handle's transaction
2488 static int jbd2_journal_file_inode(handle_t *handle, struct jbd2_inode *jinode,
2489 unsigned long flags)
2491 transaction_t *transaction = handle->h_transaction;
2494 if (is_handle_aborted(handle))
2496 journal = transaction->t_journal;
2498 jbd_debug(4, "Adding inode %lu, tid:%d\n", jinode->i_vfs_inode->i_ino,
2499 transaction->t_tid);
2502 * First check whether inode isn't already on the transaction's
2503 * lists without taking the lock. Note that this check is safe
2504 * without the lock as we cannot race with somebody removing inode
2505 * from the transaction. The reason is that we remove inode from the
2506 * transaction only in journal_release_jbd_inode() and when we commit
2507 * the transaction. We are guarded from the first case by holding
2508 * a reference to the inode. We are safe against the second case
2509 * because if jinode->i_transaction == transaction, commit code
2510 * cannot touch the transaction because we hold reference to it,
2511 * and if jinode->i_next_transaction == transaction, commit code
2512 * will only file the inode where we want it.
2514 if ((jinode->i_transaction == transaction ||
2515 jinode->i_next_transaction == transaction) &&
2516 (jinode->i_flags & flags) == flags)
2519 spin_lock(&journal->j_list_lock);
2520 jinode->i_flags |= flags;
2521 /* Is inode already attached where we need it? */
2522 if (jinode->i_transaction == transaction ||
2523 jinode->i_next_transaction == transaction)
2527 * We only ever set this variable to 1 so the test is safe. Since
2528 * t_need_data_flush is likely to be set, we do the test to save some
2529 * cacheline bouncing
2531 if (!transaction->t_need_data_flush)
2532 transaction->t_need_data_flush = 1;
2533 /* On some different transaction's list - should be
2534 * the committing one */
2535 if (jinode->i_transaction) {
2536 J_ASSERT(jinode->i_next_transaction == NULL);
2537 J_ASSERT(jinode->i_transaction ==
2538 journal->j_committing_transaction);
2539 jinode->i_next_transaction = transaction;
2542 /* Not on any transaction list... */
2543 J_ASSERT(!jinode->i_next_transaction);
2544 jinode->i_transaction = transaction;
2545 list_add(&jinode->i_list, &transaction->t_inode_list);
2547 spin_unlock(&journal->j_list_lock);
2552 int jbd2_journal_inode_add_write(handle_t *handle, struct jbd2_inode *jinode)
2554 return jbd2_journal_file_inode(handle, jinode,
2555 JI_WRITE_DATA | JI_WAIT_DATA);
2558 int jbd2_journal_inode_add_wait(handle_t *handle, struct jbd2_inode *jinode)
2560 return jbd2_journal_file_inode(handle, jinode, JI_WAIT_DATA);
2564 * File truncate and transaction commit interact with each other in a
2565 * non-trivial way. If a transaction writing data block A is
2566 * committing, we cannot discard the data by truncate until we have
2567 * written them. Otherwise if we crashed after the transaction with
2568 * write has committed but before the transaction with truncate has
2569 * committed, we could see stale data in block A. This function is a
2570 * helper to solve this problem. It starts writeout of the truncated
2571 * part in case it is in the committing transaction.
2573 * Filesystem code must call this function when inode is journaled in
2574 * ordered mode before truncation happens and after the inode has been
2575 * placed on orphan list with the new inode size. The second condition
2576 * avoids the race that someone writes new data and we start
2577 * committing the transaction after this function has been called but
2578 * before a transaction for truncate is started (and furthermore it
2579 * allows us to optimize the case where the addition to orphan list
2580 * happens in the same transaction as write --- we don't have to write
2581 * any data in such case).
2583 int jbd2_journal_begin_ordered_truncate(journal_t *journal,
2584 struct jbd2_inode *jinode,
2587 transaction_t *inode_trans, *commit_trans;
2590 /* This is a quick check to avoid locking if not necessary */
2591 if (!jinode->i_transaction)
2593 /* Locks are here just to force reading of recent values, it is
2594 * enough that the transaction was not committing before we started
2595 * a transaction adding the inode to orphan list */
2596 read_lock(&journal->j_state_lock);
2597 commit_trans = journal->j_committing_transaction;
2598 read_unlock(&journal->j_state_lock);
2599 spin_lock(&journal->j_list_lock);
2600 inode_trans = jinode->i_transaction;
2601 spin_unlock(&journal->j_list_lock);
2602 if (inode_trans == commit_trans) {
2603 ret = filemap_fdatawrite_range(jinode->i_vfs_inode->i_mapping,
2604 new_size, LLONG_MAX);
2606 jbd2_journal_abort(journal, ret);