2 * linux/fs/jbd2/transaction.c
4 * Written by Stephen C. Tweedie <sct@redhat.com>, 1998
6 * Copyright 1998 Red Hat corp --- All Rights Reserved
8 * This file is part of the Linux kernel and is made available under
9 * the terms of the GNU General Public License, version 2, or at your
10 * option, any later version, incorporated herein by reference.
12 * Generic filesystem transaction handling code; part of the ext2fs
15 * This file manages transactions (compound commits managed by the
16 * journaling code) and handles (individual atomic operations by the
20 #include <linux/time.h>
22 #include <linux/jbd2.h>
23 #include <linux/errno.h>
24 #include <linux/slab.h>
25 #include <linux/timer.h>
27 #include <linux/highmem.h>
28 #include <linux/hrtimer.h>
29 #include <linux/backing-dev.h>
30 #include <linux/bug.h>
31 #include <linux/module.h>
33 #include <trace/events/jbd2.h>
35 static void __jbd2_journal_temp_unlink_buffer(struct journal_head *jh);
36 static void __jbd2_journal_unfile_buffer(struct journal_head *jh);
38 static struct kmem_cache *transaction_cache;
39 int __init jbd2_journal_init_transaction_cache(void)
41 J_ASSERT(!transaction_cache);
42 transaction_cache = kmem_cache_create("jbd2_transaction_s",
43 sizeof(transaction_t),
45 SLAB_HWCACHE_ALIGN|SLAB_TEMPORARY,
47 if (transaction_cache)
52 void jbd2_journal_destroy_transaction_cache(void)
54 if (transaction_cache) {
55 kmem_cache_destroy(transaction_cache);
56 transaction_cache = NULL;
60 void jbd2_journal_free_transaction(transaction_t *transaction)
62 if (unlikely(ZERO_OR_NULL_PTR(transaction)))
64 kmem_cache_free(transaction_cache, transaction);
68 * jbd2_get_transaction: obtain a new transaction_t object.
70 * Simply allocate and initialise a new transaction. Create it in
71 * RUNNING state and add it to the current journal (which should not
72 * have an existing running transaction: we only make a new transaction
73 * once we have started to commit the old one).
76 * The journal MUST be locked. We don't perform atomic mallocs on the
77 * new transaction and we can't block without protecting against other
78 * processes trying to touch the journal while it is in transition.
82 static transaction_t *
83 jbd2_get_transaction(journal_t *journal, transaction_t *transaction)
85 transaction->t_journal = journal;
86 transaction->t_state = T_RUNNING;
87 transaction->t_start_time = ktime_get();
88 transaction->t_tid = journal->j_transaction_sequence++;
89 transaction->t_expires = jiffies + journal->j_commit_interval;
90 spin_lock_init(&transaction->t_handle_lock);
91 atomic_set(&transaction->t_updates, 0);
92 atomic_set(&transaction->t_outstanding_credits,
93 atomic_read(&journal->j_reserved_credits));
94 atomic_set(&transaction->t_handle_count, 0);
95 INIT_LIST_HEAD(&transaction->t_inode_list);
96 INIT_LIST_HEAD(&transaction->t_private_list);
98 /* Set up the commit timer for the new transaction. */
99 journal->j_commit_timer.expires = round_jiffies_up(transaction->t_expires);
100 add_timer(&journal->j_commit_timer);
102 J_ASSERT(journal->j_running_transaction == NULL);
103 journal->j_running_transaction = transaction;
104 transaction->t_max_wait = 0;
105 transaction->t_start = jiffies;
106 transaction->t_requested = 0;
114 * A handle_t is an object which represents a single atomic update to a
115 * filesystem, and which tracks all of the modifications which form part
116 * of that one update.
120 * Update transaction's maximum wait time, if debugging is enabled.
122 * In order for t_max_wait to be reliable, it must be protected by a
123 * lock. But doing so will mean that start_this_handle() can not be
124 * run in parallel on SMP systems, which limits our scalability. So
125 * unless debugging is enabled, we no longer update t_max_wait, which
126 * means that maximum wait time reported by the jbd2_run_stats
127 * tracepoint will always be zero.
129 static inline void update_t_max_wait(transaction_t *transaction,
132 #ifdef CONFIG_JBD2_DEBUG
133 if (jbd2_journal_enable_debug &&
134 time_after(transaction->t_start, ts)) {
135 ts = jbd2_time_diff(ts, transaction->t_start);
136 spin_lock(&transaction->t_handle_lock);
137 if (ts > transaction->t_max_wait)
138 transaction->t_max_wait = ts;
139 spin_unlock(&transaction->t_handle_lock);
145 * Wait until running transaction passes T_LOCKED state. Also starts the commit
146 * if needed. The function expects running transaction to exist and releases
149 static void wait_transaction_locked(journal_t *journal)
150 __releases(journal->j_state_lock)
154 tid_t tid = journal->j_running_transaction->t_tid;
156 prepare_to_wait(&journal->j_wait_transaction_locked, &wait,
157 TASK_UNINTERRUPTIBLE);
158 need_to_start = !tid_geq(journal->j_commit_request, tid);
159 read_unlock(&journal->j_state_lock);
161 jbd2_log_start_commit(journal, tid);
163 finish_wait(&journal->j_wait_transaction_locked, &wait);
166 static void sub_reserved_credits(journal_t *journal, int blocks)
168 atomic_sub(blocks, &journal->j_reserved_credits);
169 wake_up(&journal->j_wait_reserved);
173 * Wait until we can add credits for handle to the running transaction. Called
174 * with j_state_lock held for reading. Returns 0 if handle joined the running
175 * transaction. Returns 1 if we had to wait, j_state_lock is dropped, and
178 static int add_transaction_credits(journal_t *journal, int blocks,
181 transaction_t *t = journal->j_running_transaction;
183 int total = blocks + rsv_blocks;
186 * If the current transaction is locked down for commit, wait
187 * for the lock to be released.
189 if (t->t_state == T_LOCKED) {
190 wait_transaction_locked(journal);
195 * If there is not enough space left in the log to write all
196 * potential buffers requested by this operation, we need to
197 * stall pending a log checkpoint to free some more log space.
199 needed = atomic_add_return(total, &t->t_outstanding_credits);
200 if (needed > journal->j_max_transaction_buffers) {
202 * If the current transaction is already too large,
203 * then start to commit it: we can then go back and
204 * attach this handle to a new transaction.
206 atomic_sub(total, &t->t_outstanding_credits);
207 wait_transaction_locked(journal);
212 * The commit code assumes that it can get enough log space
213 * without forcing a checkpoint. This is *critical* for
214 * correctness: a checkpoint of a buffer which is also
215 * associated with a committing transaction creates a deadlock,
216 * so commit simply cannot force through checkpoints.
218 * We must therefore ensure the necessary space in the journal
219 * *before* starting to dirty potentially checkpointed buffers
220 * in the new transaction.
222 if (jbd2_log_space_left(journal) < jbd2_space_needed(journal)) {
223 atomic_sub(total, &t->t_outstanding_credits);
224 read_unlock(&journal->j_state_lock);
225 write_lock(&journal->j_state_lock);
226 if (jbd2_log_space_left(journal) < jbd2_space_needed(journal))
227 __jbd2_log_wait_for_space(journal);
228 write_unlock(&journal->j_state_lock);
232 /* No reservation? We are done... */
236 needed = atomic_add_return(rsv_blocks, &journal->j_reserved_credits);
237 /* We allow at most half of a transaction to be reserved */
238 if (needed > journal->j_max_transaction_buffers / 2) {
239 sub_reserved_credits(journal, rsv_blocks);
240 atomic_sub(total, &t->t_outstanding_credits);
241 read_unlock(&journal->j_state_lock);
242 wait_event(journal->j_wait_reserved,
243 atomic_read(&journal->j_reserved_credits) + rsv_blocks
244 <= journal->j_max_transaction_buffers / 2);
251 * start_this_handle: Given a handle, deal with any locking or stalling
252 * needed to make sure that there is enough journal space for the handle
253 * to begin. Attach the handle to a transaction and set up the
254 * transaction's buffer credits.
257 static int start_this_handle(journal_t *journal, handle_t *handle,
260 transaction_t *transaction, *new_transaction = NULL;
261 int blocks = handle->h_buffer_credits;
263 unsigned long ts = jiffies;
266 * 1/2 of transaction can be reserved so we can practically handle
267 * only 1/2 of maximum transaction size per operation
269 if (WARN_ON(blocks > journal->j_max_transaction_buffers / 2)) {
270 printk(KERN_ERR "JBD2: %s wants too many credits (%d > %d)\n",
271 current->comm, blocks,
272 journal->j_max_transaction_buffers / 2);
276 if (handle->h_rsv_handle)
277 rsv_blocks = handle->h_rsv_handle->h_buffer_credits;
280 if (!journal->j_running_transaction) {
282 * If __GFP_FS is not present, then we may be being called from
283 * inside the fs writeback layer, so we MUST NOT fail.
285 if ((gfp_mask & __GFP_FS) == 0)
286 gfp_mask |= __GFP_NOFAIL;
287 new_transaction = kmem_cache_zalloc(transaction_cache,
289 if (!new_transaction)
293 jbd_debug(3, "New handle %p going live.\n", handle);
296 * We need to hold j_state_lock until t_updates has been incremented,
297 * for proper journal barrier handling
300 read_lock(&journal->j_state_lock);
301 BUG_ON(journal->j_flags & JBD2_UNMOUNT);
302 if (is_journal_aborted(journal) ||
303 (journal->j_errno != 0 && !(journal->j_flags & JBD2_ACK_ERR))) {
304 read_unlock(&journal->j_state_lock);
305 jbd2_journal_free_transaction(new_transaction);
310 * Wait on the journal's transaction barrier if necessary. Specifically
311 * we allow reserved handles to proceed because otherwise commit could
312 * deadlock on page writeback not being able to complete.
314 if (!handle->h_reserved && journal->j_barrier_count) {
315 read_unlock(&journal->j_state_lock);
316 wait_event(journal->j_wait_transaction_locked,
317 journal->j_barrier_count == 0);
321 if (!journal->j_running_transaction) {
322 read_unlock(&journal->j_state_lock);
323 if (!new_transaction)
324 goto alloc_transaction;
325 write_lock(&journal->j_state_lock);
326 if (!journal->j_running_transaction &&
327 (handle->h_reserved || !journal->j_barrier_count)) {
328 jbd2_get_transaction(journal, new_transaction);
329 new_transaction = NULL;
331 write_unlock(&journal->j_state_lock);
335 transaction = journal->j_running_transaction;
337 if (!handle->h_reserved) {
338 /* We may have dropped j_state_lock - restart in that case */
339 if (add_transaction_credits(journal, blocks, rsv_blocks))
343 * We have handle reserved so we are allowed to join T_LOCKED
344 * transaction and we don't have to check for transaction size
347 sub_reserved_credits(journal, blocks);
348 handle->h_reserved = 0;
351 /* OK, account for the buffers that this operation expects to
352 * use and add the handle to the running transaction.
354 update_t_max_wait(transaction, ts);
355 handle->h_transaction = transaction;
356 handle->h_requested_credits = blocks;
357 handle->h_start_jiffies = jiffies;
358 atomic_inc(&transaction->t_updates);
359 atomic_inc(&transaction->t_handle_count);
360 jbd_debug(4, "Handle %p given %d credits (total %d, free %lu)\n",
362 atomic_read(&transaction->t_outstanding_credits),
363 jbd2_log_space_left(journal));
364 read_unlock(&journal->j_state_lock);
365 current->journal_info = handle;
367 lock_map_acquire(&handle->h_lockdep_map);
368 jbd2_journal_free_transaction(new_transaction);
372 static struct lock_class_key jbd2_handle_key;
374 /* Allocate a new handle. This should probably be in a slab... */
375 static handle_t *new_handle(int nblocks)
377 handle_t *handle = jbd2_alloc_handle(GFP_NOFS);
380 handle->h_buffer_credits = nblocks;
383 lockdep_init_map(&handle->h_lockdep_map, "jbd2_handle",
384 &jbd2_handle_key, 0);
390 * handle_t *jbd2_journal_start() - Obtain a new handle.
391 * @journal: Journal to start transaction on.
392 * @nblocks: number of block buffer we might modify
394 * We make sure that the transaction can guarantee at least nblocks of
395 * modified buffers in the log. We block until the log can guarantee
396 * that much space. Additionally, if rsv_blocks > 0, we also create another
397 * handle with rsv_blocks reserved blocks in the journal. This handle is
398 * is stored in h_rsv_handle. It is not attached to any particular transaction
399 * and thus doesn't block transaction commit. If the caller uses this reserved
400 * handle, it has to set h_rsv_handle to NULL as otherwise jbd2_journal_stop()
401 * on the parent handle will dispose the reserved one. Reserved handle has to
402 * be converted to a normal handle using jbd2_journal_start_reserved() before
405 * Return a pointer to a newly allocated handle, or an ERR_PTR() value
408 handle_t *jbd2__journal_start(journal_t *journal, int nblocks, int rsv_blocks,
409 gfp_t gfp_mask, unsigned int type,
410 unsigned int line_no)
412 handle_t *handle = journal_current_handle();
416 return ERR_PTR(-EROFS);
419 J_ASSERT(handle->h_transaction->t_journal == journal);
424 handle = new_handle(nblocks);
426 return ERR_PTR(-ENOMEM);
428 handle_t *rsv_handle;
430 rsv_handle = new_handle(rsv_blocks);
432 jbd2_free_handle(handle);
433 return ERR_PTR(-ENOMEM);
435 rsv_handle->h_reserved = 1;
436 rsv_handle->h_journal = journal;
437 handle->h_rsv_handle = rsv_handle;
440 err = start_this_handle(journal, handle, gfp_mask);
442 if (handle->h_rsv_handle)
443 jbd2_free_handle(handle->h_rsv_handle);
444 jbd2_free_handle(handle);
447 handle->h_type = type;
448 handle->h_line_no = line_no;
449 trace_jbd2_handle_start(journal->j_fs_dev->bd_dev,
450 handle->h_transaction->t_tid, type,
454 EXPORT_SYMBOL(jbd2__journal_start);
457 handle_t *jbd2_journal_start(journal_t *journal, int nblocks)
459 return jbd2__journal_start(journal, nblocks, 0, GFP_NOFS, 0, 0);
461 EXPORT_SYMBOL(jbd2_journal_start);
463 void jbd2_journal_free_reserved(handle_t *handle)
465 journal_t *journal = handle->h_journal;
467 WARN_ON(!handle->h_reserved);
468 sub_reserved_credits(journal, handle->h_buffer_credits);
469 jbd2_free_handle(handle);
471 EXPORT_SYMBOL(jbd2_journal_free_reserved);
474 * int jbd2_journal_start_reserved(handle_t *handle) - start reserved handle
475 * @handle: handle to start
477 * Start handle that has been previously reserved with jbd2_journal_reserve().
478 * This attaches @handle to the running transaction (or creates one if there's
479 * not transaction running). Unlike jbd2_journal_start() this function cannot
480 * block on journal commit, checkpointing, or similar stuff. It can block on
481 * memory allocation or frozen journal though.
483 * Return 0 on success, non-zero on error - handle is freed in that case.
485 int jbd2_journal_start_reserved(handle_t *handle, unsigned int type,
486 unsigned int line_no)
488 journal_t *journal = handle->h_journal;
491 if (WARN_ON(!handle->h_reserved)) {
492 /* Someone passed in normal handle? Just stop it. */
493 jbd2_journal_stop(handle);
497 * Usefulness of mixing of reserved and unreserved handles is
498 * questionable. So far nobody seems to need it so just error out.
500 if (WARN_ON(current->journal_info)) {
501 jbd2_journal_free_reserved(handle);
505 handle->h_journal = NULL;
507 * GFP_NOFS is here because callers are likely from writeback or
508 * similarly constrained call sites
510 ret = start_this_handle(journal, handle, GFP_NOFS);
512 jbd2_journal_free_reserved(handle);
515 handle->h_type = type;
516 handle->h_line_no = line_no;
519 EXPORT_SYMBOL(jbd2_journal_start_reserved);
522 * int jbd2_journal_extend() - extend buffer credits.
523 * @handle: handle to 'extend'
524 * @nblocks: nr blocks to try to extend by.
526 * Some transactions, such as large extends and truncates, can be done
527 * atomically all at once or in several stages. The operation requests
528 * a credit for a number of buffer modications in advance, but can
529 * extend its credit if it needs more.
531 * jbd2_journal_extend tries to give the running handle more buffer credits.
532 * It does not guarantee that allocation - this is a best-effort only.
533 * The calling process MUST be able to deal cleanly with a failure to
536 * Return 0 on success, non-zero on failure.
538 * return code < 0 implies an error
539 * return code > 0 implies normal transaction-full status.
541 int jbd2_journal_extend(handle_t *handle, int nblocks)
543 transaction_t *transaction = handle->h_transaction;
548 if (is_handle_aborted(handle))
550 journal = transaction->t_journal;
554 read_lock(&journal->j_state_lock);
556 /* Don't extend a locked-down transaction! */
557 if (transaction->t_state != T_RUNNING) {
558 jbd_debug(3, "denied handle %p %d blocks: "
559 "transaction not running\n", handle, nblocks);
563 spin_lock(&transaction->t_handle_lock);
564 wanted = atomic_add_return(nblocks,
565 &transaction->t_outstanding_credits);
567 if (wanted > journal->j_max_transaction_buffers) {
568 jbd_debug(3, "denied handle %p %d blocks: "
569 "transaction too large\n", handle, nblocks);
570 atomic_sub(nblocks, &transaction->t_outstanding_credits);
574 if (wanted + (wanted >> JBD2_CONTROL_BLOCKS_SHIFT) >
575 jbd2_log_space_left(journal)) {
576 jbd_debug(3, "denied handle %p %d blocks: "
577 "insufficient log space\n", handle, nblocks);
578 atomic_sub(nblocks, &transaction->t_outstanding_credits);
582 trace_jbd2_handle_extend(journal->j_fs_dev->bd_dev,
584 handle->h_type, handle->h_line_no,
585 handle->h_buffer_credits,
588 handle->h_buffer_credits += nblocks;
589 handle->h_requested_credits += nblocks;
592 jbd_debug(3, "extended handle %p by %d\n", handle, nblocks);
594 spin_unlock(&transaction->t_handle_lock);
596 read_unlock(&journal->j_state_lock);
602 * int jbd2_journal_restart() - restart a handle .
603 * @handle: handle to restart
604 * @nblocks: nr credits requested
606 * Restart a handle for a multi-transaction filesystem
609 * If the jbd2_journal_extend() call above fails to grant new buffer credits
610 * to a running handle, a call to jbd2_journal_restart will commit the
611 * handle's transaction so far and reattach the handle to a new
612 * transaction capabable of guaranteeing the requested number of
613 * credits. We preserve reserved handle if there's any attached to the
616 int jbd2__journal_restart(handle_t *handle, int nblocks, gfp_t gfp_mask)
618 transaction_t *transaction = handle->h_transaction;
621 int need_to_start, ret;
623 /* If we've had an abort of any type, don't even think about
624 * actually doing the restart! */
625 if (is_handle_aborted(handle))
627 journal = transaction->t_journal;
630 * First unlink the handle from its current transaction, and start the
633 J_ASSERT(atomic_read(&transaction->t_updates) > 0);
634 J_ASSERT(journal_current_handle() == handle);
636 read_lock(&journal->j_state_lock);
637 spin_lock(&transaction->t_handle_lock);
638 atomic_sub(handle->h_buffer_credits,
639 &transaction->t_outstanding_credits);
640 if (handle->h_rsv_handle) {
641 sub_reserved_credits(journal,
642 handle->h_rsv_handle->h_buffer_credits);
644 if (atomic_dec_and_test(&transaction->t_updates))
645 wake_up(&journal->j_wait_updates);
646 tid = transaction->t_tid;
647 spin_unlock(&transaction->t_handle_lock);
648 handle->h_transaction = NULL;
649 current->journal_info = NULL;
651 jbd_debug(2, "restarting handle %p\n", handle);
652 need_to_start = !tid_geq(journal->j_commit_request, tid);
653 read_unlock(&journal->j_state_lock);
655 jbd2_log_start_commit(journal, tid);
657 lock_map_release(&handle->h_lockdep_map);
658 handle->h_buffer_credits = nblocks;
659 ret = start_this_handle(journal, handle, gfp_mask);
662 EXPORT_SYMBOL(jbd2__journal_restart);
665 int jbd2_journal_restart(handle_t *handle, int nblocks)
667 return jbd2__journal_restart(handle, nblocks, GFP_NOFS);
669 EXPORT_SYMBOL(jbd2_journal_restart);
672 * void jbd2_journal_lock_updates () - establish a transaction barrier.
673 * @journal: Journal to establish a barrier on.
675 * This locks out any further updates from being started, and blocks
676 * until all existing updates have completed, returning only once the
677 * journal is in a quiescent state with no updates running.
679 * The journal lock should not be held on entry.
681 void jbd2_journal_lock_updates(journal_t *journal)
685 write_lock(&journal->j_state_lock);
686 ++journal->j_barrier_count;
688 /* Wait until there are no reserved handles */
689 if (atomic_read(&journal->j_reserved_credits)) {
690 write_unlock(&journal->j_state_lock);
691 wait_event(journal->j_wait_reserved,
692 atomic_read(&journal->j_reserved_credits) == 0);
693 write_lock(&journal->j_state_lock);
696 /* Wait until there are no running updates */
698 transaction_t *transaction = journal->j_running_transaction;
703 spin_lock(&transaction->t_handle_lock);
704 prepare_to_wait(&journal->j_wait_updates, &wait,
705 TASK_UNINTERRUPTIBLE);
706 if (!atomic_read(&transaction->t_updates)) {
707 spin_unlock(&transaction->t_handle_lock);
708 finish_wait(&journal->j_wait_updates, &wait);
711 spin_unlock(&transaction->t_handle_lock);
712 write_unlock(&journal->j_state_lock);
714 finish_wait(&journal->j_wait_updates, &wait);
715 write_lock(&journal->j_state_lock);
717 write_unlock(&journal->j_state_lock);
720 * We have now established a barrier against other normal updates, but
721 * we also need to barrier against other jbd2_journal_lock_updates() calls
722 * to make sure that we serialise special journal-locked operations
725 mutex_lock(&journal->j_barrier);
729 * void jbd2_journal_unlock_updates (journal_t* journal) - release barrier
730 * @journal: Journal to release the barrier on.
732 * Release a transaction barrier obtained with jbd2_journal_lock_updates().
734 * Should be called without the journal lock held.
736 void jbd2_journal_unlock_updates (journal_t *journal)
738 J_ASSERT(journal->j_barrier_count != 0);
740 mutex_unlock(&journal->j_barrier);
741 write_lock(&journal->j_state_lock);
742 --journal->j_barrier_count;
743 write_unlock(&journal->j_state_lock);
744 wake_up(&journal->j_wait_transaction_locked);
747 static void warn_dirty_buffer(struct buffer_head *bh)
749 char b[BDEVNAME_SIZE];
752 "JBD2: Spotted dirty metadata buffer (dev = %s, blocknr = %llu). "
753 "There's a risk of filesystem corruption in case of system "
755 bdevname(bh->b_bdev, b), (unsigned long long)bh->b_blocknr);
759 * If the buffer is already part of the current transaction, then there
760 * is nothing we need to do. If it is already part of a prior
761 * transaction which we are still committing to disk, then we need to
762 * make sure that we do not overwrite the old copy: we do copy-out to
763 * preserve the copy going to disk. We also account the buffer against
764 * the handle's metadata buffer credits (unless the buffer is already
765 * part of the transaction, that is).
769 do_get_write_access(handle_t *handle, struct journal_head *jh,
772 struct buffer_head *bh;
773 transaction_t *transaction = handle->h_transaction;
776 char *frozen_buffer = NULL;
778 unsigned long start_lock, time_lock;
780 if (is_handle_aborted(handle))
782 journal = transaction->t_journal;
784 jbd_debug(5, "journal_head %p, force_copy %d\n", jh, force_copy);
786 JBUFFER_TRACE(jh, "entry");
790 /* @@@ Need to check for errors here at some point. */
792 start_lock = jiffies;
794 jbd_lock_bh_state(bh);
796 /* If it takes too long to lock the buffer, trace it */
797 time_lock = jbd2_time_diff(start_lock, jiffies);
798 if (time_lock > HZ/10)
799 trace_jbd2_lock_buffer_stall(bh->b_bdev->bd_dev,
800 jiffies_to_msecs(time_lock));
802 /* We now hold the buffer lock so it is safe to query the buffer
803 * state. Is the buffer dirty?
805 * If so, there are two possibilities. The buffer may be
806 * non-journaled, and undergoing a quite legitimate writeback.
807 * Otherwise, it is journaled, and we don't expect dirty buffers
808 * in that state (the buffers should be marked JBD_Dirty
809 * instead.) So either the IO is being done under our own
810 * control and this is a bug, or it's a third party IO such as
811 * dump(8) (which may leave the buffer scheduled for read ---
812 * ie. locked but not dirty) or tune2fs (which may actually have
813 * the buffer dirtied, ugh.) */
815 if (buffer_dirty(bh)) {
817 * First question: is this buffer already part of the current
818 * transaction or the existing committing transaction?
820 if (jh->b_transaction) {
822 jh->b_transaction == transaction ||
824 journal->j_committing_transaction);
825 if (jh->b_next_transaction)
826 J_ASSERT_JH(jh, jh->b_next_transaction ==
828 warn_dirty_buffer(bh);
831 * In any case we need to clean the dirty flag and we must
832 * do it under the buffer lock to be sure we don't race
833 * with running write-out.
835 JBUFFER_TRACE(jh, "Journalling dirty buffer");
836 clear_buffer_dirty(bh);
837 set_buffer_jbddirty(bh);
843 if (is_handle_aborted(handle)) {
844 jbd_unlock_bh_state(bh);
850 * The buffer is already part of this transaction if b_transaction or
851 * b_next_transaction points to it
853 if (jh->b_transaction == transaction ||
854 jh->b_next_transaction == transaction)
858 * this is the first time this transaction is touching this buffer,
859 * reset the modified flag
864 * If there is already a copy-out version of this buffer, then we don't
865 * need to make another one
867 if (jh->b_frozen_data) {
868 JBUFFER_TRACE(jh, "has frozen data");
869 J_ASSERT_JH(jh, jh->b_next_transaction == NULL);
870 jh->b_next_transaction = transaction;
874 /* Is there data here we need to preserve? */
876 if (jh->b_transaction && jh->b_transaction != transaction) {
877 JBUFFER_TRACE(jh, "owned by older transaction");
878 J_ASSERT_JH(jh, jh->b_next_transaction == NULL);
879 J_ASSERT_JH(jh, jh->b_transaction ==
880 journal->j_committing_transaction);
882 /* There is one case we have to be very careful about.
883 * If the committing transaction is currently writing
884 * this buffer out to disk and has NOT made a copy-out,
885 * then we cannot modify the buffer contents at all
886 * right now. The essence of copy-out is that it is the
887 * extra copy, not the primary copy, which gets
888 * journaled. If the primary copy is already going to
889 * disk then we cannot do copy-out here. */
891 if (buffer_shadow(bh)) {
892 JBUFFER_TRACE(jh, "on shadow: sleep");
893 jbd_unlock_bh_state(bh);
894 wait_on_bit_io(&bh->b_state, BH_Shadow,
895 TASK_UNINTERRUPTIBLE);
900 * Only do the copy if the currently-owning transaction still
901 * needs it. If buffer isn't on BJ_Metadata list, the
902 * committing transaction is past that stage (here we use the
903 * fact that BH_Shadow is set under bh_state lock together with
904 * refiling to BJ_Shadow list and at this point we know the
905 * buffer doesn't have BH_Shadow set).
907 * Subtle point, though: if this is a get_undo_access,
908 * then we will be relying on the frozen_data to contain
909 * the new value of the committed_data record after the
910 * transaction, so we HAVE to force the frozen_data copy
913 if (jh->b_jlist == BJ_Metadata || force_copy) {
914 JBUFFER_TRACE(jh, "generate frozen data");
915 if (!frozen_buffer) {
916 JBUFFER_TRACE(jh, "allocate memory for buffer");
917 jbd_unlock_bh_state(bh);
919 jbd2_alloc(jh2bh(jh)->b_size,
921 if (!frozen_buffer) {
923 "%s: OOM for frozen_buffer\n",
925 JBUFFER_TRACE(jh, "oom!");
927 jbd_lock_bh_state(bh);
932 jh->b_frozen_data = frozen_buffer;
933 frozen_buffer = NULL;
936 jh->b_next_transaction = transaction;
941 * Finally, if the buffer is not journaled right now, we need to make
942 * sure it doesn't get written to disk before the caller actually
943 * commits the new data
945 if (!jh->b_transaction) {
946 JBUFFER_TRACE(jh, "no transaction");
947 J_ASSERT_JH(jh, !jh->b_next_transaction);
948 JBUFFER_TRACE(jh, "file as BJ_Reserved");
949 spin_lock(&journal->j_list_lock);
950 __jbd2_journal_file_buffer(jh, transaction, BJ_Reserved);
951 spin_unlock(&journal->j_list_lock);
960 J_EXPECT_JH(jh, buffer_uptodate(jh2bh(jh)),
961 "Possible IO failure.\n");
962 page = jh2bh(jh)->b_page;
963 offset = offset_in_page(jh2bh(jh)->b_data);
964 source = kmap_atomic(page);
965 /* Fire data frozen trigger just before we copy the data */
966 jbd2_buffer_frozen_trigger(jh, source + offset,
968 memcpy(jh->b_frozen_data, source+offset, jh2bh(jh)->b_size);
969 kunmap_atomic(source);
972 * Now that the frozen data is saved off, we need to store
973 * any matching triggers.
975 jh->b_frozen_triggers = jh->b_triggers;
977 jbd_unlock_bh_state(bh);
980 * If we are about to journal a buffer, then any revoke pending on it is
983 jbd2_journal_cancel_revoke(handle, jh);
986 if (unlikely(frozen_buffer)) /* It's usually NULL */
987 jbd2_free(frozen_buffer, bh->b_size);
989 JBUFFER_TRACE(jh, "exit");
994 * int jbd2_journal_get_write_access() - notify intent to modify a buffer for metadata (not data) update.
995 * @handle: transaction to add buffer modifications to
996 * @bh: bh to be used for metadata writes
998 * Returns an error code or 0 on success.
1000 * In full data journalling mode the buffer may be of type BJ_AsyncData,
1001 * because we're write()ing a buffer which is also part of a shared mapping.
1004 int jbd2_journal_get_write_access(handle_t *handle, struct buffer_head *bh)
1006 struct journal_head *jh = jbd2_journal_add_journal_head(bh);
1009 /* We do not want to get caught playing with fields which the
1010 * log thread also manipulates. Make sure that the buffer
1011 * completes any outstanding IO before proceeding. */
1012 rc = do_get_write_access(handle, jh, 0);
1013 jbd2_journal_put_journal_head(jh);
1019 * When the user wants to journal a newly created buffer_head
1020 * (ie. getblk() returned a new buffer and we are going to populate it
1021 * manually rather than reading off disk), then we need to keep the
1022 * buffer_head locked until it has been completely filled with new
1023 * data. In this case, we should be able to make the assertion that
1024 * the bh is not already part of an existing transaction.
1026 * The buffer should already be locked by the caller by this point.
1027 * There is no lock ranking violation: it was a newly created,
1028 * unlocked buffer beforehand. */
1031 * int jbd2_journal_get_create_access () - notify intent to use newly created bh
1032 * @handle: transaction to new buffer to
1035 * Call this if you create a new bh.
1037 int jbd2_journal_get_create_access(handle_t *handle, struct buffer_head *bh)
1039 transaction_t *transaction = handle->h_transaction;
1041 struct journal_head *jh = jbd2_journal_add_journal_head(bh);
1044 jbd_debug(5, "journal_head %p\n", jh);
1046 if (is_handle_aborted(handle))
1048 journal = transaction->t_journal;
1051 JBUFFER_TRACE(jh, "entry");
1053 * The buffer may already belong to this transaction due to pre-zeroing
1054 * in the filesystem's new_block code. It may also be on the previous,
1055 * committing transaction's lists, but it HAS to be in Forget state in
1056 * that case: the transaction must have deleted the buffer for it to be
1059 jbd_lock_bh_state(bh);
1060 J_ASSERT_JH(jh, (jh->b_transaction == transaction ||
1061 jh->b_transaction == NULL ||
1062 (jh->b_transaction == journal->j_committing_transaction &&
1063 jh->b_jlist == BJ_Forget)));
1065 J_ASSERT_JH(jh, jh->b_next_transaction == NULL);
1066 J_ASSERT_JH(jh, buffer_locked(jh2bh(jh)));
1068 if (jh->b_transaction == NULL) {
1070 * Previous jbd2_journal_forget() could have left the buffer
1071 * with jbddirty bit set because it was being committed. When
1072 * the commit finished, we've filed the buffer for
1073 * checkpointing and marked it dirty. Now we are reallocating
1074 * the buffer so the transaction freeing it must have
1075 * committed and so it's safe to clear the dirty bit.
1077 clear_buffer_dirty(jh2bh(jh));
1078 /* first access by this transaction */
1081 JBUFFER_TRACE(jh, "file as BJ_Reserved");
1082 spin_lock(&journal->j_list_lock);
1083 __jbd2_journal_file_buffer(jh, transaction, BJ_Reserved);
1084 } else if (jh->b_transaction == journal->j_committing_transaction) {
1085 /* first access by this transaction */
1088 JBUFFER_TRACE(jh, "set next transaction");
1089 spin_lock(&journal->j_list_lock);
1090 jh->b_next_transaction = transaction;
1092 spin_unlock(&journal->j_list_lock);
1093 jbd_unlock_bh_state(bh);
1096 * akpm: I added this. ext3_alloc_branch can pick up new indirect
1097 * blocks which contain freed but then revoked metadata. We need
1098 * to cancel the revoke in case we end up freeing it yet again
1099 * and the reallocating as data - this would cause a second revoke,
1100 * which hits an assertion error.
1102 JBUFFER_TRACE(jh, "cancelling revoke");
1103 jbd2_journal_cancel_revoke(handle, jh);
1105 jbd2_journal_put_journal_head(jh);
1110 * int jbd2_journal_get_undo_access() - Notify intent to modify metadata with
1111 * non-rewindable consequences
1112 * @handle: transaction
1113 * @bh: buffer to undo
1115 * Sometimes there is a need to distinguish between metadata which has
1116 * been committed to disk and that which has not. The ext3fs code uses
1117 * this for freeing and allocating space, we have to make sure that we
1118 * do not reuse freed space until the deallocation has been committed,
1119 * since if we overwrote that space we would make the delete
1120 * un-rewindable in case of a crash.
1122 * To deal with that, jbd2_journal_get_undo_access requests write access to a
1123 * buffer for parts of non-rewindable operations such as delete
1124 * operations on the bitmaps. The journaling code must keep a copy of
1125 * the buffer's contents prior to the undo_access call until such time
1126 * as we know that the buffer has definitely been committed to disk.
1128 * We never need to know which transaction the committed data is part
1129 * of, buffers touched here are guaranteed to be dirtied later and so
1130 * will be committed to a new transaction in due course, at which point
1131 * we can discard the old committed data pointer.
1133 * Returns error number or 0 on success.
1135 int jbd2_journal_get_undo_access(handle_t *handle, struct buffer_head *bh)
1138 struct journal_head *jh = jbd2_journal_add_journal_head(bh);
1139 char *committed_data = NULL;
1141 JBUFFER_TRACE(jh, "entry");
1144 * Do this first --- it can drop the journal lock, so we want to
1145 * make sure that obtaining the committed_data is done
1146 * atomically wrt. completion of any outstanding commits.
1148 err = do_get_write_access(handle, jh, 1);
1153 if (!jh->b_committed_data) {
1154 committed_data = jbd2_alloc(jh2bh(jh)->b_size, GFP_NOFS);
1155 if (!committed_data) {
1156 printk(KERN_ERR "%s: No memory for committed data\n",
1163 jbd_lock_bh_state(bh);
1164 if (!jh->b_committed_data) {
1165 /* Copy out the current buffer contents into the
1166 * preserved, committed copy. */
1167 JBUFFER_TRACE(jh, "generate b_committed data");
1168 if (!committed_data) {
1169 jbd_unlock_bh_state(bh);
1173 jh->b_committed_data = committed_data;
1174 committed_data = NULL;
1175 memcpy(jh->b_committed_data, bh->b_data, bh->b_size);
1177 jbd_unlock_bh_state(bh);
1179 jbd2_journal_put_journal_head(jh);
1180 if (unlikely(committed_data))
1181 jbd2_free(committed_data, bh->b_size);
1186 * void jbd2_journal_set_triggers() - Add triggers for commit writeout
1187 * @bh: buffer to trigger on
1188 * @type: struct jbd2_buffer_trigger_type containing the trigger(s).
1190 * Set any triggers on this journal_head. This is always safe, because
1191 * triggers for a committing buffer will be saved off, and triggers for
1192 * a running transaction will match the buffer in that transaction.
1194 * Call with NULL to clear the triggers.
1196 void jbd2_journal_set_triggers(struct buffer_head *bh,
1197 struct jbd2_buffer_trigger_type *type)
1199 struct journal_head *jh = jbd2_journal_grab_journal_head(bh);
1203 jh->b_triggers = type;
1204 jbd2_journal_put_journal_head(jh);
1207 void jbd2_buffer_frozen_trigger(struct journal_head *jh, void *mapped_data,
1208 struct jbd2_buffer_trigger_type *triggers)
1210 struct buffer_head *bh = jh2bh(jh);
1212 if (!triggers || !triggers->t_frozen)
1215 triggers->t_frozen(triggers, bh, mapped_data, bh->b_size);
1218 void jbd2_buffer_abort_trigger(struct journal_head *jh,
1219 struct jbd2_buffer_trigger_type *triggers)
1221 if (!triggers || !triggers->t_abort)
1224 triggers->t_abort(triggers, jh2bh(jh));
1230 * int jbd2_journal_dirty_metadata() - mark a buffer as containing dirty metadata
1231 * @handle: transaction to add buffer to.
1232 * @bh: buffer to mark
1234 * mark dirty metadata which needs to be journaled as part of the current
1237 * The buffer must have previously had jbd2_journal_get_write_access()
1238 * called so that it has a valid journal_head attached to the buffer
1241 * The buffer is placed on the transaction's metadata list and is marked
1242 * as belonging to the transaction.
1244 * Returns error number or 0 on success.
1246 * Special care needs to be taken if the buffer already belongs to the
1247 * current committing transaction (in which case we should have frozen
1248 * data present for that commit). In that case, we don't relink the
1249 * buffer: that only gets done when the old transaction finally
1250 * completes its commit.
1252 int jbd2_journal_dirty_metadata(handle_t *handle, struct buffer_head *bh)
1254 transaction_t *transaction = handle->h_transaction;
1256 struct journal_head *jh;
1259 if (is_handle_aborted(handle))
1261 journal = transaction->t_journal;
1262 jh = jbd2_journal_grab_journal_head(bh);
1267 jbd_debug(5, "journal_head %p\n", jh);
1268 JBUFFER_TRACE(jh, "entry");
1270 jbd_lock_bh_state(bh);
1272 if (jh->b_modified == 0) {
1274 * This buffer's got modified and becoming part
1275 * of the transaction. This needs to be done
1276 * once a transaction -bzzz
1279 if (handle->h_buffer_credits <= 0) {
1283 handle->h_buffer_credits--;
1287 * fastpath, to avoid expensive locking. If this buffer is already
1288 * on the running transaction's metadata list there is nothing to do.
1289 * Nobody can take it off again because there is a handle open.
1290 * I _think_ we're OK here with SMP barriers - a mistaken decision will
1291 * result in this test being false, so we go in and take the locks.
1293 if (jh->b_transaction == transaction && jh->b_jlist == BJ_Metadata) {
1294 JBUFFER_TRACE(jh, "fastpath");
1295 if (unlikely(jh->b_transaction !=
1296 journal->j_running_transaction)) {
1297 printk(KERN_ERR "JBD2: %s: "
1298 "jh->b_transaction (%llu, %p, %u) != "
1299 "journal->j_running_transaction (%p, %u)\n",
1301 (unsigned long long) bh->b_blocknr,
1303 jh->b_transaction ? jh->b_transaction->t_tid : 0,
1304 journal->j_running_transaction,
1305 journal->j_running_transaction ?
1306 journal->j_running_transaction->t_tid : 0);
1312 set_buffer_jbddirty(bh);
1315 * Metadata already on the current transaction list doesn't
1316 * need to be filed. Metadata on another transaction's list must
1317 * be committing, and will be refiled once the commit completes:
1318 * leave it alone for now.
1320 if (jh->b_transaction != transaction) {
1321 JBUFFER_TRACE(jh, "already on other transaction");
1322 if (unlikely(((jh->b_transaction !=
1323 journal->j_committing_transaction)) ||
1324 (jh->b_next_transaction != transaction))) {
1325 printk(KERN_ERR "jbd2_journal_dirty_metadata: %s: "
1326 "bad jh for block %llu: "
1327 "transaction (%p, %u), "
1328 "jh->b_transaction (%p, %u), "
1329 "jh->b_next_transaction (%p, %u), jlist %u\n",
1331 (unsigned long long) bh->b_blocknr,
1332 transaction, transaction->t_tid,
1335 jh->b_transaction->t_tid : 0,
1336 jh->b_next_transaction,
1337 jh->b_next_transaction ?
1338 jh->b_next_transaction->t_tid : 0,
1343 /* And this case is illegal: we can't reuse another
1344 * transaction's data buffer, ever. */
1348 /* That test should have eliminated the following case: */
1349 J_ASSERT_JH(jh, jh->b_frozen_data == NULL);
1351 JBUFFER_TRACE(jh, "file as BJ_Metadata");
1352 spin_lock(&journal->j_list_lock);
1353 __jbd2_journal_file_buffer(jh, transaction, BJ_Metadata);
1354 spin_unlock(&journal->j_list_lock);
1356 jbd_unlock_bh_state(bh);
1357 jbd2_journal_put_journal_head(jh);
1359 JBUFFER_TRACE(jh, "exit");
1364 * void jbd2_journal_forget() - bforget() for potentially-journaled buffers.
1365 * @handle: transaction handle
1366 * @bh: bh to 'forget'
1368 * We can only do the bforget if there are no commits pending against the
1369 * buffer. If the buffer is dirty in the current running transaction we
1370 * can safely unlink it.
1372 * bh may not be a journalled buffer at all - it may be a non-JBD
1373 * buffer which came off the hashtable. Check for this.
1375 * Decrements bh->b_count by one.
1377 * Allow this call even if the handle has aborted --- it may be part of
1378 * the caller's cleanup after an abort.
1380 int jbd2_journal_forget (handle_t *handle, struct buffer_head *bh)
1382 transaction_t *transaction = handle->h_transaction;
1384 struct journal_head *jh;
1385 int drop_reserve = 0;
1387 int was_modified = 0;
1389 if (is_handle_aborted(handle))
1391 journal = transaction->t_journal;
1393 BUFFER_TRACE(bh, "entry");
1395 jbd_lock_bh_state(bh);
1397 if (!buffer_jbd(bh))
1401 /* Critical error: attempting to delete a bitmap buffer, maybe?
1402 * Don't do any jbd operations, and return an error. */
1403 if (!J_EXPECT_JH(jh, !jh->b_committed_data,
1404 "inconsistent data on disk")) {
1409 /* keep track of whether or not this transaction modified us */
1410 was_modified = jh->b_modified;
1413 * The buffer's going from the transaction, we must drop
1414 * all references -bzzz
1418 if (jh->b_transaction == transaction) {
1419 J_ASSERT_JH(jh, !jh->b_frozen_data);
1421 /* If we are forgetting a buffer which is already part
1422 * of this transaction, then we can just drop it from
1423 * the transaction immediately. */
1424 clear_buffer_dirty(bh);
1425 clear_buffer_jbddirty(bh);
1427 JBUFFER_TRACE(jh, "belongs to current transaction: unfile");
1430 * we only want to drop a reference if this transaction
1431 * modified the buffer
1437 * We are no longer going to journal this buffer.
1438 * However, the commit of this transaction is still
1439 * important to the buffer: the delete that we are now
1440 * processing might obsolete an old log entry, so by
1441 * committing, we can satisfy the buffer's checkpoint.
1443 * So, if we have a checkpoint on the buffer, we should
1444 * now refile the buffer on our BJ_Forget list so that
1445 * we know to remove the checkpoint after we commit.
1448 spin_lock(&journal->j_list_lock);
1449 if (jh->b_cp_transaction) {
1450 __jbd2_journal_temp_unlink_buffer(jh);
1451 __jbd2_journal_file_buffer(jh, transaction, BJ_Forget);
1453 __jbd2_journal_unfile_buffer(jh);
1454 if (!buffer_jbd(bh)) {
1455 spin_unlock(&journal->j_list_lock);
1456 jbd_unlock_bh_state(bh);
1461 spin_unlock(&journal->j_list_lock);
1462 } else if (jh->b_transaction) {
1463 J_ASSERT_JH(jh, (jh->b_transaction ==
1464 journal->j_committing_transaction));
1465 /* However, if the buffer is still owned by a prior
1466 * (committing) transaction, we can't drop it yet... */
1467 JBUFFER_TRACE(jh, "belongs to older transaction");
1468 /* ... but we CAN drop it from the new transaction if we
1469 * have also modified it since the original commit. */
1471 if (jh->b_next_transaction) {
1472 J_ASSERT(jh->b_next_transaction == transaction);
1473 spin_lock(&journal->j_list_lock);
1474 jh->b_next_transaction = NULL;
1475 spin_unlock(&journal->j_list_lock);
1478 * only drop a reference if this transaction modified
1487 jbd_unlock_bh_state(bh);
1491 /* no need to reserve log space for this block -bzzz */
1492 handle->h_buffer_credits++;
1498 * int jbd2_journal_stop() - complete a transaction
1499 * @handle: tranaction to complete.
1501 * All done for a particular handle.
1503 * There is not much action needed here. We just return any remaining
1504 * buffer credits to the transaction and remove the handle. The only
1505 * complication is that we need to start a commit operation if the
1506 * filesystem is marked for synchronous update.
1508 * jbd2_journal_stop itself will not usually return an error, but it may
1509 * do so in unusual circumstances. In particular, expect it to
1510 * return -EIO if a jbd2_journal_abort has been executed since the
1511 * transaction began.
1513 int jbd2_journal_stop(handle_t *handle)
1515 transaction_t *transaction = handle->h_transaction;
1517 int err = 0, wait_for_commit = 0;
1523 * Handle is already detached from the transaction so
1524 * there is nothing to do other than decrease a refcount,
1525 * or free the handle if refcount drops to zero
1527 if (--handle->h_ref > 0) {
1528 jbd_debug(4, "h_ref %d -> %d\n", handle->h_ref + 1,
1532 if (handle->h_rsv_handle)
1533 jbd2_free_handle(handle->h_rsv_handle);
1537 journal = transaction->t_journal;
1539 J_ASSERT(journal_current_handle() == handle);
1541 if (is_handle_aborted(handle))
1544 J_ASSERT(atomic_read(&transaction->t_updates) > 0);
1546 if (--handle->h_ref > 0) {
1547 jbd_debug(4, "h_ref %d -> %d\n", handle->h_ref + 1,
1552 jbd_debug(4, "Handle %p going down\n", handle);
1553 trace_jbd2_handle_stats(journal->j_fs_dev->bd_dev,
1555 handle->h_type, handle->h_line_no,
1556 jiffies - handle->h_start_jiffies,
1557 handle->h_sync, handle->h_requested_credits,
1558 (handle->h_requested_credits -
1559 handle->h_buffer_credits));
1562 * Implement synchronous transaction batching. If the handle
1563 * was synchronous, don't force a commit immediately. Let's
1564 * yield and let another thread piggyback onto this
1565 * transaction. Keep doing that while new threads continue to
1566 * arrive. It doesn't cost much - we're about to run a commit
1567 * and sleep on IO anyway. Speeds up many-threaded, many-dir
1568 * operations by 30x or more...
1570 * We try and optimize the sleep time against what the
1571 * underlying disk can do, instead of having a static sleep
1572 * time. This is useful for the case where our storage is so
1573 * fast that it is more optimal to go ahead and force a flush
1574 * and wait for the transaction to be committed than it is to
1575 * wait for an arbitrary amount of time for new writers to
1576 * join the transaction. We achieve this by measuring how
1577 * long it takes to commit a transaction, and compare it with
1578 * how long this transaction has been running, and if run time
1579 * < commit time then we sleep for the delta and commit. This
1580 * greatly helps super fast disks that would see slowdowns as
1581 * more threads started doing fsyncs.
1583 * But don't do this if this process was the most recent one
1584 * to perform a synchronous write. We do this to detect the
1585 * case where a single process is doing a stream of sync
1586 * writes. No point in waiting for joiners in that case.
1588 * Setting max_batch_time to 0 disables this completely.
1591 if (handle->h_sync && journal->j_last_sync_writer != pid &&
1592 journal->j_max_batch_time) {
1593 u64 commit_time, trans_time;
1595 journal->j_last_sync_writer = pid;
1597 read_lock(&journal->j_state_lock);
1598 commit_time = journal->j_average_commit_time;
1599 read_unlock(&journal->j_state_lock);
1601 trans_time = ktime_to_ns(ktime_sub(ktime_get(),
1602 transaction->t_start_time));
1604 commit_time = max_t(u64, commit_time,
1605 1000*journal->j_min_batch_time);
1606 commit_time = min_t(u64, commit_time,
1607 1000*journal->j_max_batch_time);
1609 if (trans_time < commit_time) {
1610 ktime_t expires = ktime_add_ns(ktime_get(),
1612 set_current_state(TASK_UNINTERRUPTIBLE);
1613 schedule_hrtimeout(&expires, HRTIMER_MODE_ABS);
1618 transaction->t_synchronous_commit = 1;
1619 current->journal_info = NULL;
1620 atomic_sub(handle->h_buffer_credits,
1621 &transaction->t_outstanding_credits);
1624 * If the handle is marked SYNC, we need to set another commit
1625 * going! We also want to force a commit if the current
1626 * transaction is occupying too much of the log, or if the
1627 * transaction is too old now.
1629 if (handle->h_sync ||
1630 (atomic_read(&transaction->t_outstanding_credits) >
1631 journal->j_max_transaction_buffers) ||
1632 time_after_eq(jiffies, transaction->t_expires)) {
1633 /* Do this even for aborted journals: an abort still
1634 * completes the commit thread, it just doesn't write
1635 * anything to disk. */
1637 jbd_debug(2, "transaction too old, requesting commit for "
1638 "handle %p\n", handle);
1639 /* This is non-blocking */
1640 jbd2_log_start_commit(journal, transaction->t_tid);
1643 * Special case: JBD2_SYNC synchronous updates require us
1644 * to wait for the commit to complete.
1646 if (handle->h_sync && !(current->flags & PF_MEMALLOC))
1647 wait_for_commit = 1;
1651 * Once we drop t_updates, if it goes to zero the transaction
1652 * could start committing on us and eventually disappear. So
1653 * once we do this, we must not dereference transaction
1656 tid = transaction->t_tid;
1657 if (atomic_dec_and_test(&transaction->t_updates)) {
1658 wake_up(&journal->j_wait_updates);
1659 if (journal->j_barrier_count)
1660 wake_up(&journal->j_wait_transaction_locked);
1663 if (wait_for_commit)
1664 err = jbd2_log_wait_commit(journal, tid);
1666 lock_map_release(&handle->h_lockdep_map);
1668 if (handle->h_rsv_handle)
1669 jbd2_journal_free_reserved(handle->h_rsv_handle);
1671 jbd2_free_handle(handle);
1677 * List management code snippets: various functions for manipulating the
1678 * transaction buffer lists.
1683 * Append a buffer to a transaction list, given the transaction's list head
1686 * j_list_lock is held.
1688 * jbd_lock_bh_state(jh2bh(jh)) is held.
1692 __blist_add_buffer(struct journal_head **list, struct journal_head *jh)
1695 jh->b_tnext = jh->b_tprev = jh;
1698 /* Insert at the tail of the list to preserve order */
1699 struct journal_head *first = *list, *last = first->b_tprev;
1701 jh->b_tnext = first;
1702 last->b_tnext = first->b_tprev = jh;
1707 * Remove a buffer from a transaction list, given the transaction's list
1710 * Called with j_list_lock held, and the journal may not be locked.
1712 * jbd_lock_bh_state(jh2bh(jh)) is held.
1716 __blist_del_buffer(struct journal_head **list, struct journal_head *jh)
1719 *list = jh->b_tnext;
1723 jh->b_tprev->b_tnext = jh->b_tnext;
1724 jh->b_tnext->b_tprev = jh->b_tprev;
1728 * Remove a buffer from the appropriate transaction list.
1730 * Note that this function can *change* the value of
1731 * bh->b_transaction->t_buffers, t_forget, t_shadow_list, t_log_list or
1732 * t_reserved_list. If the caller is holding onto a copy of one of these
1733 * pointers, it could go bad. Generally the caller needs to re-read the
1734 * pointer from the transaction_t.
1736 * Called under j_list_lock.
1738 static void __jbd2_journal_temp_unlink_buffer(struct journal_head *jh)
1740 struct journal_head **list = NULL;
1741 transaction_t *transaction;
1742 struct buffer_head *bh = jh2bh(jh);
1744 J_ASSERT_JH(jh, jbd_is_locked_bh_state(bh));
1745 transaction = jh->b_transaction;
1747 assert_spin_locked(&transaction->t_journal->j_list_lock);
1749 J_ASSERT_JH(jh, jh->b_jlist < BJ_Types);
1750 if (jh->b_jlist != BJ_None)
1751 J_ASSERT_JH(jh, transaction != NULL);
1753 switch (jh->b_jlist) {
1757 transaction->t_nr_buffers--;
1758 J_ASSERT_JH(jh, transaction->t_nr_buffers >= 0);
1759 list = &transaction->t_buffers;
1762 list = &transaction->t_forget;
1765 list = &transaction->t_shadow_list;
1768 list = &transaction->t_reserved_list;
1772 __blist_del_buffer(list, jh);
1773 jh->b_jlist = BJ_None;
1774 if (test_clear_buffer_jbddirty(bh))
1775 mark_buffer_dirty(bh); /* Expose it to the VM */
1779 * Remove buffer from all transactions.
1781 * Called with bh_state lock and j_list_lock
1783 * jh and bh may be already freed when this function returns.
1785 static void __jbd2_journal_unfile_buffer(struct journal_head *jh)
1787 __jbd2_journal_temp_unlink_buffer(jh);
1788 jh->b_transaction = NULL;
1789 jbd2_journal_put_journal_head(jh);
1792 void jbd2_journal_unfile_buffer(journal_t *journal, struct journal_head *jh)
1794 struct buffer_head *bh = jh2bh(jh);
1796 /* Get reference so that buffer cannot be freed before we unlock it */
1798 jbd_lock_bh_state(bh);
1799 spin_lock(&journal->j_list_lock);
1800 __jbd2_journal_unfile_buffer(jh);
1801 spin_unlock(&journal->j_list_lock);
1802 jbd_unlock_bh_state(bh);
1807 * Called from jbd2_journal_try_to_free_buffers().
1809 * Called under jbd_lock_bh_state(bh)
1812 __journal_try_to_free_buffer(journal_t *journal, struct buffer_head *bh)
1814 struct journal_head *jh;
1818 if (buffer_locked(bh) || buffer_dirty(bh))
1821 if (jh->b_next_transaction != NULL || jh->b_transaction != NULL)
1824 spin_lock(&journal->j_list_lock);
1825 if (jh->b_cp_transaction != NULL) {
1826 /* written-back checkpointed metadata buffer */
1827 JBUFFER_TRACE(jh, "remove from checkpoint list");
1828 __jbd2_journal_remove_checkpoint(jh);
1830 spin_unlock(&journal->j_list_lock);
1836 * int jbd2_journal_try_to_free_buffers() - try to free page buffers.
1837 * @journal: journal for operation
1838 * @page: to try and free
1839 * @gfp_mask: we use the mask to detect how hard should we try to release
1840 * buffers. If __GFP_WAIT and __GFP_FS is set, we wait for commit code to
1841 * release the buffers.
1844 * For all the buffers on this page,
1845 * if they are fully written out ordered data, move them onto BUF_CLEAN
1846 * so try_to_free_buffers() can reap them.
1848 * This function returns non-zero if we wish try_to_free_buffers()
1849 * to be called. We do this if the page is releasable by try_to_free_buffers().
1850 * We also do it if the page has locked or dirty buffers and the caller wants
1851 * us to perform sync or async writeout.
1853 * This complicates JBD locking somewhat. We aren't protected by the
1854 * BKL here. We wish to remove the buffer from its committing or
1855 * running transaction's ->t_datalist via __jbd2_journal_unfile_buffer.
1857 * This may *change* the value of transaction_t->t_datalist, so anyone
1858 * who looks at t_datalist needs to lock against this function.
1860 * Even worse, someone may be doing a jbd2_journal_dirty_data on this
1861 * buffer. So we need to lock against that. jbd2_journal_dirty_data()
1862 * will come out of the lock with the buffer dirty, which makes it
1863 * ineligible for release here.
1865 * Who else is affected by this? hmm... Really the only contender
1866 * is do_get_write_access() - it could be looking at the buffer while
1867 * journal_try_to_free_buffer() is changing its state. But that
1868 * cannot happen because we never reallocate freed data as metadata
1869 * while the data is part of a transaction. Yes?
1871 * Return 0 on failure, 1 on success
1873 int jbd2_journal_try_to_free_buffers(journal_t *journal,
1874 struct page *page, gfp_t gfp_mask)
1876 struct buffer_head *head;
1877 struct buffer_head *bh;
1880 J_ASSERT(PageLocked(page));
1882 head = page_buffers(page);
1885 struct journal_head *jh;
1888 * We take our own ref against the journal_head here to avoid
1889 * having to add tons of locking around each instance of
1890 * jbd2_journal_put_journal_head().
1892 jh = jbd2_journal_grab_journal_head(bh);
1896 jbd_lock_bh_state(bh);
1897 __journal_try_to_free_buffer(journal, bh);
1898 jbd2_journal_put_journal_head(jh);
1899 jbd_unlock_bh_state(bh);
1902 } while ((bh = bh->b_this_page) != head);
1904 ret = try_to_free_buffers(page);
1911 * This buffer is no longer needed. If it is on an older transaction's
1912 * checkpoint list we need to record it on this transaction's forget list
1913 * to pin this buffer (and hence its checkpointing transaction) down until
1914 * this transaction commits. If the buffer isn't on a checkpoint list, we
1916 * Returns non-zero if JBD no longer has an interest in the buffer.
1918 * Called under j_list_lock.
1920 * Called under jbd_lock_bh_state(bh).
1922 static int __dispose_buffer(struct journal_head *jh, transaction_t *transaction)
1925 struct buffer_head *bh = jh2bh(jh);
1927 if (jh->b_cp_transaction) {
1928 JBUFFER_TRACE(jh, "on running+cp transaction");
1929 __jbd2_journal_temp_unlink_buffer(jh);
1931 * We don't want to write the buffer anymore, clear the
1932 * bit so that we don't confuse checks in
1933 * __journal_file_buffer
1935 clear_buffer_dirty(bh);
1936 __jbd2_journal_file_buffer(jh, transaction, BJ_Forget);
1939 JBUFFER_TRACE(jh, "on running transaction");
1940 __jbd2_journal_unfile_buffer(jh);
1946 * jbd2_journal_invalidatepage
1948 * This code is tricky. It has a number of cases to deal with.
1950 * There are two invariants which this code relies on:
1952 * i_size must be updated on disk before we start calling invalidatepage on the
1955 * This is done in ext3 by defining an ext3_setattr method which
1956 * updates i_size before truncate gets going. By maintaining this
1957 * invariant, we can be sure that it is safe to throw away any buffers
1958 * attached to the current transaction: once the transaction commits,
1959 * we know that the data will not be needed.
1961 * Note however that we can *not* throw away data belonging to the
1962 * previous, committing transaction!
1964 * Any disk blocks which *are* part of the previous, committing
1965 * transaction (and which therefore cannot be discarded immediately) are
1966 * not going to be reused in the new running transaction
1968 * The bitmap committed_data images guarantee this: any block which is
1969 * allocated in one transaction and removed in the next will be marked
1970 * as in-use in the committed_data bitmap, so cannot be reused until
1971 * the next transaction to delete the block commits. This means that
1972 * leaving committing buffers dirty is quite safe: the disk blocks
1973 * cannot be reallocated to a different file and so buffer aliasing is
1977 * The above applies mainly to ordered data mode. In writeback mode we
1978 * don't make guarantees about the order in which data hits disk --- in
1979 * particular we don't guarantee that new dirty data is flushed before
1980 * transaction commit --- so it is always safe just to discard data
1981 * immediately in that mode. --sct
1985 * The journal_unmap_buffer helper function returns zero if the buffer
1986 * concerned remains pinned as an anonymous buffer belonging to an older
1989 * We're outside-transaction here. Either or both of j_running_transaction
1990 * and j_committing_transaction may be NULL.
1992 static int journal_unmap_buffer(journal_t *journal, struct buffer_head *bh,
1995 transaction_t *transaction;
1996 struct journal_head *jh;
1999 BUFFER_TRACE(bh, "entry");
2002 * It is safe to proceed here without the j_list_lock because the
2003 * buffers cannot be stolen by try_to_free_buffers as long as we are
2004 * holding the page lock. --sct
2007 if (!buffer_jbd(bh))
2008 goto zap_buffer_unlocked;
2010 /* OK, we have data buffer in journaled mode */
2011 write_lock(&journal->j_state_lock);
2012 jbd_lock_bh_state(bh);
2013 spin_lock(&journal->j_list_lock);
2015 jh = jbd2_journal_grab_journal_head(bh);
2017 goto zap_buffer_no_jh;
2020 * We cannot remove the buffer from checkpoint lists until the
2021 * transaction adding inode to orphan list (let's call it T)
2022 * is committed. Otherwise if the transaction changing the
2023 * buffer would be cleaned from the journal before T is
2024 * committed, a crash will cause that the correct contents of
2025 * the buffer will be lost. On the other hand we have to
2026 * clear the buffer dirty bit at latest at the moment when the
2027 * transaction marking the buffer as freed in the filesystem
2028 * structures is committed because from that moment on the
2029 * block can be reallocated and used by a different page.
2030 * Since the block hasn't been freed yet but the inode has
2031 * already been added to orphan list, it is safe for us to add
2032 * the buffer to BJ_Forget list of the newest transaction.
2034 * Also we have to clear buffer_mapped flag of a truncated buffer
2035 * because the buffer_head may be attached to the page straddling
2036 * i_size (can happen only when blocksize < pagesize) and thus the
2037 * buffer_head can be reused when the file is extended again. So we end
2038 * up keeping around invalidated buffers attached to transactions'
2039 * BJ_Forget list just to stop checkpointing code from cleaning up
2040 * the transaction this buffer was modified in.
2042 transaction = jh->b_transaction;
2043 if (transaction == NULL) {
2044 /* First case: not on any transaction. If it
2045 * has no checkpoint link, then we can zap it:
2046 * it's a writeback-mode buffer so we don't care
2047 * if it hits disk safely. */
2048 if (!jh->b_cp_transaction) {
2049 JBUFFER_TRACE(jh, "not on any transaction: zap");
2053 if (!buffer_dirty(bh)) {
2054 /* bdflush has written it. We can drop it now */
2058 /* OK, it must be in the journal but still not
2059 * written fully to disk: it's metadata or
2060 * journaled data... */
2062 if (journal->j_running_transaction) {
2063 /* ... and once the current transaction has
2064 * committed, the buffer won't be needed any
2066 JBUFFER_TRACE(jh, "checkpointed: add to BJ_Forget");
2067 may_free = __dispose_buffer(jh,
2068 journal->j_running_transaction);
2071 /* There is no currently-running transaction. So the
2072 * orphan record which we wrote for this file must have
2073 * passed into commit. We must attach this buffer to
2074 * the committing transaction, if it exists. */
2075 if (journal->j_committing_transaction) {
2076 JBUFFER_TRACE(jh, "give to committing trans");
2077 may_free = __dispose_buffer(jh,
2078 journal->j_committing_transaction);
2081 /* The orphan record's transaction has
2082 * committed. We can cleanse this buffer */
2083 clear_buffer_jbddirty(bh);
2087 } else if (transaction == journal->j_committing_transaction) {
2088 JBUFFER_TRACE(jh, "on committing transaction");
2090 * The buffer is committing, we simply cannot touch
2091 * it. If the page is straddling i_size we have to wait
2092 * for commit and try again.
2095 jbd2_journal_put_journal_head(jh);
2096 spin_unlock(&journal->j_list_lock);
2097 jbd_unlock_bh_state(bh);
2098 write_unlock(&journal->j_state_lock);
2102 * OK, buffer won't be reachable after truncate. We just set
2103 * j_next_transaction to the running transaction (if there is
2104 * one) and mark buffer as freed so that commit code knows it
2105 * should clear dirty bits when it is done with the buffer.
2107 set_buffer_freed(bh);
2108 if (journal->j_running_transaction && buffer_jbddirty(bh))
2109 jh->b_next_transaction = journal->j_running_transaction;
2110 jbd2_journal_put_journal_head(jh);
2111 spin_unlock(&journal->j_list_lock);
2112 jbd_unlock_bh_state(bh);
2113 write_unlock(&journal->j_state_lock);
2116 /* Good, the buffer belongs to the running transaction.
2117 * We are writing our own transaction's data, not any
2118 * previous one's, so it is safe to throw it away
2119 * (remember that we expect the filesystem to have set
2120 * i_size already for this truncate so recovery will not
2121 * expose the disk blocks we are discarding here.) */
2122 J_ASSERT_JH(jh, transaction == journal->j_running_transaction);
2123 JBUFFER_TRACE(jh, "on running transaction");
2124 may_free = __dispose_buffer(jh, transaction);
2129 * This is tricky. Although the buffer is truncated, it may be reused
2130 * if blocksize < pagesize and it is attached to the page straddling
2131 * EOF. Since the buffer might have been added to BJ_Forget list of the
2132 * running transaction, journal_get_write_access() won't clear
2133 * b_modified and credit accounting gets confused. So clear b_modified
2137 jbd2_journal_put_journal_head(jh);
2139 spin_unlock(&journal->j_list_lock);
2140 jbd_unlock_bh_state(bh);
2141 write_unlock(&journal->j_state_lock);
2142 zap_buffer_unlocked:
2143 clear_buffer_dirty(bh);
2144 J_ASSERT_BH(bh, !buffer_jbddirty(bh));
2145 clear_buffer_mapped(bh);
2146 clear_buffer_req(bh);
2147 clear_buffer_new(bh);
2148 clear_buffer_delay(bh);
2149 clear_buffer_unwritten(bh);
2155 * void jbd2_journal_invalidatepage()
2156 * @journal: journal to use for flush...
2157 * @page: page to flush
2158 * @offset: start of the range to invalidate
2159 * @length: length of the range to invalidate
2161 * Reap page buffers containing data after in the specified range in page.
2162 * Can return -EBUSY if buffers are part of the committing transaction and
2163 * the page is straddling i_size. Caller then has to wait for current commit
2166 int jbd2_journal_invalidatepage(journal_t *journal,
2168 unsigned int offset,
2169 unsigned int length)
2171 struct buffer_head *head, *bh, *next;
2172 unsigned int stop = offset + length;
2173 unsigned int curr_off = 0;
2174 int partial_page = (offset || length < PAGE_CACHE_SIZE);
2178 if (!PageLocked(page))
2180 if (!page_has_buffers(page))
2183 BUG_ON(stop > PAGE_CACHE_SIZE || stop < length);
2185 /* We will potentially be playing with lists other than just the
2186 * data lists (especially for journaled data mode), so be
2187 * cautious in our locking. */
2189 head = bh = page_buffers(page);
2191 unsigned int next_off = curr_off + bh->b_size;
2192 next = bh->b_this_page;
2194 if (next_off > stop)
2197 if (offset <= curr_off) {
2198 /* This block is wholly outside the truncation point */
2200 ret = journal_unmap_buffer(journal, bh, partial_page);
2206 curr_off = next_off;
2209 } while (bh != head);
2211 if (!partial_page) {
2212 if (may_free && try_to_free_buffers(page))
2213 J_ASSERT(!page_has_buffers(page));
2219 * File a buffer on the given transaction list.
2221 void __jbd2_journal_file_buffer(struct journal_head *jh,
2222 transaction_t *transaction, int jlist)
2224 struct journal_head **list = NULL;
2226 struct buffer_head *bh = jh2bh(jh);
2228 J_ASSERT_JH(jh, jbd_is_locked_bh_state(bh));
2229 assert_spin_locked(&transaction->t_journal->j_list_lock);
2231 J_ASSERT_JH(jh, jh->b_jlist < BJ_Types);
2232 J_ASSERT_JH(jh, jh->b_transaction == transaction ||
2233 jh->b_transaction == NULL);
2235 if (jh->b_transaction && jh->b_jlist == jlist)
2238 if (jlist == BJ_Metadata || jlist == BJ_Reserved ||
2239 jlist == BJ_Shadow || jlist == BJ_Forget) {
2241 * For metadata buffers, we track dirty bit in buffer_jbddirty
2242 * instead of buffer_dirty. We should not see a dirty bit set
2243 * here because we clear it in do_get_write_access but e.g.
2244 * tune2fs can modify the sb and set the dirty bit at any time
2245 * so we try to gracefully handle that.
2247 if (buffer_dirty(bh))
2248 warn_dirty_buffer(bh);
2249 if (test_clear_buffer_dirty(bh) ||
2250 test_clear_buffer_jbddirty(bh))
2254 if (jh->b_transaction)
2255 __jbd2_journal_temp_unlink_buffer(jh);
2257 jbd2_journal_grab_journal_head(bh);
2258 jh->b_transaction = transaction;
2262 J_ASSERT_JH(jh, !jh->b_committed_data);
2263 J_ASSERT_JH(jh, !jh->b_frozen_data);
2266 transaction->t_nr_buffers++;
2267 list = &transaction->t_buffers;
2270 list = &transaction->t_forget;
2273 list = &transaction->t_shadow_list;
2276 list = &transaction->t_reserved_list;
2280 __blist_add_buffer(list, jh);
2281 jh->b_jlist = jlist;
2284 set_buffer_jbddirty(bh);
2287 void jbd2_journal_file_buffer(struct journal_head *jh,
2288 transaction_t *transaction, int jlist)
2290 jbd_lock_bh_state(jh2bh(jh));
2291 spin_lock(&transaction->t_journal->j_list_lock);
2292 __jbd2_journal_file_buffer(jh, transaction, jlist);
2293 spin_unlock(&transaction->t_journal->j_list_lock);
2294 jbd_unlock_bh_state(jh2bh(jh));
2298 * Remove a buffer from its current buffer list in preparation for
2299 * dropping it from its current transaction entirely. If the buffer has
2300 * already started to be used by a subsequent transaction, refile the
2301 * buffer on that transaction's metadata list.
2303 * Called under j_list_lock
2304 * Called under jbd_lock_bh_state(jh2bh(jh))
2306 * jh and bh may be already free when this function returns
2308 void __jbd2_journal_refile_buffer(struct journal_head *jh)
2310 int was_dirty, jlist;
2311 struct buffer_head *bh = jh2bh(jh);
2313 J_ASSERT_JH(jh, jbd_is_locked_bh_state(bh));
2314 if (jh->b_transaction)
2315 assert_spin_locked(&jh->b_transaction->t_journal->j_list_lock);
2317 /* If the buffer is now unused, just drop it. */
2318 if (jh->b_next_transaction == NULL) {
2319 __jbd2_journal_unfile_buffer(jh);
2324 * It has been modified by a later transaction: add it to the new
2325 * transaction's metadata list.
2328 was_dirty = test_clear_buffer_jbddirty(bh);
2329 __jbd2_journal_temp_unlink_buffer(jh);
2331 * We set b_transaction here because b_next_transaction will inherit
2332 * our jh reference and thus __jbd2_journal_file_buffer() must not
2335 jh->b_transaction = jh->b_next_transaction;
2336 jh->b_next_transaction = NULL;
2337 if (buffer_freed(bh))
2339 else if (jh->b_modified)
2340 jlist = BJ_Metadata;
2342 jlist = BJ_Reserved;
2343 __jbd2_journal_file_buffer(jh, jh->b_transaction, jlist);
2344 J_ASSERT_JH(jh, jh->b_transaction->t_state == T_RUNNING);
2347 set_buffer_jbddirty(bh);
2351 * __jbd2_journal_refile_buffer() with necessary locking added. We take our
2352 * bh reference so that we can safely unlock bh.
2354 * The jh and bh may be freed by this call.
2356 void jbd2_journal_refile_buffer(journal_t *journal, struct journal_head *jh)
2358 struct buffer_head *bh = jh2bh(jh);
2360 /* Get reference so that buffer cannot be freed before we unlock it */
2362 jbd_lock_bh_state(bh);
2363 spin_lock(&journal->j_list_lock);
2364 __jbd2_journal_refile_buffer(jh);
2365 jbd_unlock_bh_state(bh);
2366 spin_unlock(&journal->j_list_lock);
2371 * File inode in the inode list of the handle's transaction
2373 int jbd2_journal_file_inode(handle_t *handle, struct jbd2_inode *jinode)
2375 transaction_t *transaction = handle->h_transaction;
2378 if (is_handle_aborted(handle))
2380 journal = transaction->t_journal;
2382 jbd_debug(4, "Adding inode %lu, tid:%d\n", jinode->i_vfs_inode->i_ino,
2383 transaction->t_tid);
2386 * First check whether inode isn't already on the transaction's
2387 * lists without taking the lock. Note that this check is safe
2388 * without the lock as we cannot race with somebody removing inode
2389 * from the transaction. The reason is that we remove inode from the
2390 * transaction only in journal_release_jbd_inode() and when we commit
2391 * the transaction. We are guarded from the first case by holding
2392 * a reference to the inode. We are safe against the second case
2393 * because if jinode->i_transaction == transaction, commit code
2394 * cannot touch the transaction because we hold reference to it,
2395 * and if jinode->i_next_transaction == transaction, commit code
2396 * will only file the inode where we want it.
2398 if (jinode->i_transaction == transaction ||
2399 jinode->i_next_transaction == transaction)
2402 spin_lock(&journal->j_list_lock);
2404 if (jinode->i_transaction == transaction ||
2405 jinode->i_next_transaction == transaction)
2409 * We only ever set this variable to 1 so the test is safe. Since
2410 * t_need_data_flush is likely to be set, we do the test to save some
2411 * cacheline bouncing
2413 if (!transaction->t_need_data_flush)
2414 transaction->t_need_data_flush = 1;
2415 /* On some different transaction's list - should be
2416 * the committing one */
2417 if (jinode->i_transaction) {
2418 J_ASSERT(jinode->i_next_transaction == NULL);
2419 J_ASSERT(jinode->i_transaction ==
2420 journal->j_committing_transaction);
2421 jinode->i_next_transaction = transaction;
2424 /* Not on any transaction list... */
2425 J_ASSERT(!jinode->i_next_transaction);
2426 jinode->i_transaction = transaction;
2427 list_add(&jinode->i_list, &transaction->t_inode_list);
2429 spin_unlock(&journal->j_list_lock);
2435 * File truncate and transaction commit interact with each other in a
2436 * non-trivial way. If a transaction writing data block A is
2437 * committing, we cannot discard the data by truncate until we have
2438 * written them. Otherwise if we crashed after the transaction with
2439 * write has committed but before the transaction with truncate has
2440 * committed, we could see stale data in block A. This function is a
2441 * helper to solve this problem. It starts writeout of the truncated
2442 * part in case it is in the committing transaction.
2444 * Filesystem code must call this function when inode is journaled in
2445 * ordered mode before truncation happens and after the inode has been
2446 * placed on orphan list with the new inode size. The second condition
2447 * avoids the race that someone writes new data and we start
2448 * committing the transaction after this function has been called but
2449 * before a transaction for truncate is started (and furthermore it
2450 * allows us to optimize the case where the addition to orphan list
2451 * happens in the same transaction as write --- we don't have to write
2452 * any data in such case).
2454 int jbd2_journal_begin_ordered_truncate(journal_t *journal,
2455 struct jbd2_inode *jinode,
2458 transaction_t *inode_trans, *commit_trans;
2461 /* This is a quick check to avoid locking if not necessary */
2462 if (!jinode->i_transaction)
2464 /* Locks are here just to force reading of recent values, it is
2465 * enough that the transaction was not committing before we started
2466 * a transaction adding the inode to orphan list */
2467 read_lock(&journal->j_state_lock);
2468 commit_trans = journal->j_committing_transaction;
2469 read_unlock(&journal->j_state_lock);
2470 spin_lock(&journal->j_list_lock);
2471 inode_trans = jinode->i_transaction;
2472 spin_unlock(&journal->j_list_lock);
2473 if (inode_trans == commit_trans) {
2474 ret = filemap_fdatawrite_range(jinode->i_vfs_inode->i_mapping,
2475 new_size, LLONG_MAX);
2477 jbd2_journal_abort(journal, ret);