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>
32 #include <linux/sched/mm.h>
34 #include <trace/events/jbd2.h>
36 static void __jbd2_journal_temp_unlink_buffer(struct journal_head *jh);
37 static void __jbd2_journal_unfile_buffer(struct journal_head *jh);
39 static struct kmem_cache *transaction_cache;
40 int __init jbd2_journal_init_transaction_cache(void)
42 J_ASSERT(!transaction_cache);
43 transaction_cache = kmem_cache_create("jbd2_transaction_s",
44 sizeof(transaction_t),
46 SLAB_HWCACHE_ALIGN|SLAB_TEMPORARY,
48 if (transaction_cache)
53 void jbd2_journal_destroy_transaction_cache(void)
55 if (transaction_cache) {
56 kmem_cache_destroy(transaction_cache);
57 transaction_cache = NULL;
61 void jbd2_journal_free_transaction(transaction_t *transaction)
63 if (unlikely(ZERO_OR_NULL_PTR(transaction)))
65 kmem_cache_free(transaction_cache, transaction);
69 * jbd2_get_transaction: obtain a new transaction_t object.
71 * Simply allocate and initialise a new transaction. Create it in
72 * RUNNING state and add it to the current journal (which should not
73 * have an existing running transaction: we only make a new transaction
74 * once we have started to commit the old one).
77 * The journal MUST be locked. We don't perform atomic mallocs on the
78 * new transaction and we can't block without protecting against other
79 * processes trying to touch the journal while it is in transition.
83 static transaction_t *
84 jbd2_get_transaction(journal_t *journal, transaction_t *transaction)
86 transaction->t_journal = journal;
87 transaction->t_state = T_RUNNING;
88 transaction->t_start_time = ktime_get();
89 transaction->t_tid = journal->j_transaction_sequence++;
90 transaction->t_expires = jiffies + journal->j_commit_interval;
91 spin_lock_init(&transaction->t_handle_lock);
92 atomic_set(&transaction->t_updates, 0);
93 atomic_set(&transaction->t_outstanding_credits,
94 atomic_read(&journal->j_reserved_credits));
95 atomic_set(&transaction->t_handle_count, 0);
96 INIT_LIST_HEAD(&transaction->t_inode_list);
97 INIT_LIST_HEAD(&transaction->t_private_list);
99 /* Set up the commit timer for the new transaction. */
100 journal->j_commit_timer.expires = round_jiffies_up(transaction->t_expires);
101 add_timer(&journal->j_commit_timer);
103 J_ASSERT(journal->j_running_transaction == NULL);
104 journal->j_running_transaction = transaction;
105 transaction->t_max_wait = 0;
106 transaction->t_start = jiffies;
107 transaction->t_requested = 0;
115 * A handle_t is an object which represents a single atomic update to a
116 * filesystem, and which tracks all of the modifications which form part
117 * of that one update.
121 * Update transaction's maximum wait time, if debugging is enabled.
123 * In order for t_max_wait to be reliable, it must be protected by a
124 * lock. But doing so will mean that start_this_handle() can not be
125 * run in parallel on SMP systems, which limits our scalability. So
126 * unless debugging is enabled, we no longer update t_max_wait, which
127 * means that maximum wait time reported by the jbd2_run_stats
128 * tracepoint will always be zero.
130 static inline void update_t_max_wait(transaction_t *transaction,
133 #ifdef CONFIG_JBD2_DEBUG
134 if (jbd2_journal_enable_debug &&
135 time_after(transaction->t_start, ts)) {
136 ts = jbd2_time_diff(ts, transaction->t_start);
137 spin_lock(&transaction->t_handle_lock);
138 if (ts > transaction->t_max_wait)
139 transaction->t_max_wait = ts;
140 spin_unlock(&transaction->t_handle_lock);
146 * Wait until running transaction passes T_LOCKED state. Also starts the commit
147 * if needed. The function expects running transaction to exist and releases
150 static void wait_transaction_locked(journal_t *journal)
151 __releases(journal->j_state_lock)
155 tid_t tid = journal->j_running_transaction->t_tid;
157 prepare_to_wait(&journal->j_wait_transaction_locked, &wait,
158 TASK_UNINTERRUPTIBLE);
159 need_to_start = !tid_geq(journal->j_commit_request, tid);
160 read_unlock(&journal->j_state_lock);
162 jbd2_log_start_commit(journal, tid);
163 jbd2_might_wait_for_commit(journal);
165 finish_wait(&journal->j_wait_transaction_locked, &wait);
168 static void sub_reserved_credits(journal_t *journal, int blocks)
170 atomic_sub(blocks, &journal->j_reserved_credits);
171 wake_up(&journal->j_wait_reserved);
175 * Wait until we can add credits for handle to the running transaction. Called
176 * with j_state_lock held for reading. Returns 0 if handle joined the running
177 * transaction. Returns 1 if we had to wait, j_state_lock is dropped, and
180 static int add_transaction_credits(journal_t *journal, int blocks,
183 transaction_t *t = journal->j_running_transaction;
185 int total = blocks + rsv_blocks;
188 * If the current transaction is locked down for commit, wait
189 * for the lock to be released.
191 if (t->t_state == T_LOCKED) {
192 wait_transaction_locked(journal);
197 * If there is not enough space left in the log to write all
198 * potential buffers requested by this operation, we need to
199 * stall pending a log checkpoint to free some more log space.
201 needed = atomic_add_return(total, &t->t_outstanding_credits);
202 if (needed > journal->j_max_transaction_buffers) {
204 * If the current transaction is already too large,
205 * then start to commit it: we can then go back and
206 * attach this handle to a new transaction.
208 atomic_sub(total, &t->t_outstanding_credits);
211 * Is the number of reserved credits in the current transaction too
212 * big to fit this handle? Wait until reserved credits are freed.
214 if (atomic_read(&journal->j_reserved_credits) + total >
215 journal->j_max_transaction_buffers) {
216 read_unlock(&journal->j_state_lock);
217 jbd2_might_wait_for_commit(journal);
218 wait_event(journal->j_wait_reserved,
219 atomic_read(&journal->j_reserved_credits) + total <=
220 journal->j_max_transaction_buffers);
224 wait_transaction_locked(journal);
229 * The commit code assumes that it can get enough log space
230 * without forcing a checkpoint. This is *critical* for
231 * correctness: a checkpoint of a buffer which is also
232 * associated with a committing transaction creates a deadlock,
233 * so commit simply cannot force through checkpoints.
235 * We must therefore ensure the necessary space in the journal
236 * *before* starting to dirty potentially checkpointed buffers
237 * in the new transaction.
239 if (jbd2_log_space_left(journal) < jbd2_space_needed(journal)) {
240 atomic_sub(total, &t->t_outstanding_credits);
241 read_unlock(&journal->j_state_lock);
242 jbd2_might_wait_for_commit(journal);
243 write_lock(&journal->j_state_lock);
244 if (jbd2_log_space_left(journal) < jbd2_space_needed(journal))
245 __jbd2_log_wait_for_space(journal);
246 write_unlock(&journal->j_state_lock);
250 /* No reservation? We are done... */
254 needed = atomic_add_return(rsv_blocks, &journal->j_reserved_credits);
255 /* We allow at most half of a transaction to be reserved */
256 if (needed > journal->j_max_transaction_buffers / 2) {
257 sub_reserved_credits(journal, rsv_blocks);
258 atomic_sub(total, &t->t_outstanding_credits);
259 read_unlock(&journal->j_state_lock);
260 jbd2_might_wait_for_commit(journal);
261 wait_event(journal->j_wait_reserved,
262 atomic_read(&journal->j_reserved_credits) + rsv_blocks
263 <= journal->j_max_transaction_buffers / 2);
270 * start_this_handle: Given a handle, deal with any locking or stalling
271 * needed to make sure that there is enough journal space for the handle
272 * to begin. Attach the handle to a transaction and set up the
273 * transaction's buffer credits.
276 static int start_this_handle(journal_t *journal, handle_t *handle,
279 transaction_t *transaction, *new_transaction = NULL;
280 int blocks = handle->h_buffer_credits;
282 unsigned long ts = jiffies;
284 if (handle->h_rsv_handle)
285 rsv_blocks = handle->h_rsv_handle->h_buffer_credits;
288 * Limit the number of reserved credits to 1/2 of maximum transaction
289 * size and limit the number of total credits to not exceed maximum
290 * transaction size per operation.
292 if ((rsv_blocks > journal->j_max_transaction_buffers / 2) ||
293 (rsv_blocks + blocks > journal->j_max_transaction_buffers)) {
294 printk(KERN_ERR "JBD2: %s wants too many credits "
295 "credits:%d rsv_credits:%d max:%d\n",
296 current->comm, blocks, rsv_blocks,
297 journal->j_max_transaction_buffers);
303 if (!journal->j_running_transaction) {
305 * If __GFP_FS is not present, then we may be being called from
306 * inside the fs writeback layer, so we MUST NOT fail.
308 if ((gfp_mask & __GFP_FS) == 0)
309 gfp_mask |= __GFP_NOFAIL;
310 new_transaction = kmem_cache_zalloc(transaction_cache,
312 if (!new_transaction)
316 jbd_debug(3, "New handle %p going live.\n", handle);
319 * We need to hold j_state_lock until t_updates has been incremented,
320 * for proper journal barrier handling
323 read_lock(&journal->j_state_lock);
324 BUG_ON(journal->j_flags & JBD2_UNMOUNT);
325 if (is_journal_aborted(journal) ||
326 (journal->j_errno != 0 && !(journal->j_flags & JBD2_ACK_ERR))) {
327 read_unlock(&journal->j_state_lock);
328 jbd2_journal_free_transaction(new_transaction);
333 * Wait on the journal's transaction barrier if necessary. Specifically
334 * we allow reserved handles to proceed because otherwise commit could
335 * deadlock on page writeback not being able to complete.
337 if (!handle->h_reserved && journal->j_barrier_count) {
338 read_unlock(&journal->j_state_lock);
339 wait_event(journal->j_wait_transaction_locked,
340 journal->j_barrier_count == 0);
344 if (!journal->j_running_transaction) {
345 read_unlock(&journal->j_state_lock);
346 if (!new_transaction)
347 goto alloc_transaction;
348 write_lock(&journal->j_state_lock);
349 if (!journal->j_running_transaction &&
350 (handle->h_reserved || !journal->j_barrier_count)) {
351 jbd2_get_transaction(journal, new_transaction);
352 new_transaction = NULL;
354 write_unlock(&journal->j_state_lock);
358 transaction = journal->j_running_transaction;
360 if (!handle->h_reserved) {
361 /* We may have dropped j_state_lock - restart in that case */
362 if (add_transaction_credits(journal, blocks, rsv_blocks))
366 * We have handle reserved so we are allowed to join T_LOCKED
367 * transaction and we don't have to check for transaction size
370 sub_reserved_credits(journal, blocks);
371 handle->h_reserved = 0;
374 /* OK, account for the buffers that this operation expects to
375 * use and add the handle to the running transaction.
377 update_t_max_wait(transaction, ts);
378 handle->h_transaction = transaction;
379 handle->h_requested_credits = blocks;
380 handle->h_start_jiffies = jiffies;
381 atomic_inc(&transaction->t_updates);
382 atomic_inc(&transaction->t_handle_count);
383 jbd_debug(4, "Handle %p given %d credits (total %d, free %lu)\n",
385 atomic_read(&transaction->t_outstanding_credits),
386 jbd2_log_space_left(journal));
387 read_unlock(&journal->j_state_lock);
388 current->journal_info = handle;
390 rwsem_acquire_read(&journal->j_trans_commit_map, 0, 0, _THIS_IP_);
391 jbd2_journal_free_transaction(new_transaction);
393 * Ensure that no allocations done while the transaction is open are
394 * going to recurse back to the fs layer.
396 handle->saved_alloc_context = memalloc_nofs_save();
400 /* Allocate a new handle. This should probably be in a slab... */
401 static handle_t *new_handle(int nblocks)
403 handle_t *handle = jbd2_alloc_handle(GFP_NOFS);
406 handle->h_buffer_credits = nblocks;
413 * handle_t *jbd2_journal_start() - Obtain a new handle.
414 * @journal: Journal to start transaction on.
415 * @nblocks: number of block buffer we might modify
417 * We make sure that the transaction can guarantee at least nblocks of
418 * modified buffers in the log. We block until the log can guarantee
419 * that much space. Additionally, if rsv_blocks > 0, we also create another
420 * handle with rsv_blocks reserved blocks in the journal. This handle is
421 * is stored in h_rsv_handle. It is not attached to any particular transaction
422 * and thus doesn't block transaction commit. If the caller uses this reserved
423 * handle, it has to set h_rsv_handle to NULL as otherwise jbd2_journal_stop()
424 * on the parent handle will dispose the reserved one. Reserved handle has to
425 * be converted to a normal handle using jbd2_journal_start_reserved() before
428 * Return a pointer to a newly allocated handle, or an ERR_PTR() value
431 handle_t *jbd2__journal_start(journal_t *journal, int nblocks, int rsv_blocks,
432 gfp_t gfp_mask, unsigned int type,
433 unsigned int line_no)
435 handle_t *handle = journal_current_handle();
439 return ERR_PTR(-EROFS);
442 J_ASSERT(handle->h_transaction->t_journal == journal);
447 handle = new_handle(nblocks);
449 return ERR_PTR(-ENOMEM);
451 handle_t *rsv_handle;
453 rsv_handle = new_handle(rsv_blocks);
455 jbd2_free_handle(handle);
456 return ERR_PTR(-ENOMEM);
458 rsv_handle->h_reserved = 1;
459 rsv_handle->h_journal = journal;
460 handle->h_rsv_handle = rsv_handle;
463 err = start_this_handle(journal, handle, gfp_mask);
465 if (handle->h_rsv_handle)
466 jbd2_free_handle(handle->h_rsv_handle);
467 jbd2_free_handle(handle);
470 handle->h_type = type;
471 handle->h_line_no = line_no;
472 trace_jbd2_handle_start(journal->j_fs_dev->bd_dev,
473 handle->h_transaction->t_tid, type,
478 EXPORT_SYMBOL(jbd2__journal_start);
481 handle_t *jbd2_journal_start(journal_t *journal, int nblocks)
483 return jbd2__journal_start(journal, nblocks, 0, GFP_NOFS, 0, 0);
485 EXPORT_SYMBOL(jbd2_journal_start);
487 void jbd2_journal_free_reserved(handle_t *handle)
489 journal_t *journal = handle->h_journal;
491 WARN_ON(!handle->h_reserved);
492 sub_reserved_credits(journal, handle->h_buffer_credits);
493 jbd2_free_handle(handle);
495 EXPORT_SYMBOL(jbd2_journal_free_reserved);
498 * int jbd2_journal_start_reserved(handle_t *handle) - start reserved handle
499 * @handle: handle to start
501 * Start handle that has been previously reserved with jbd2_journal_reserve().
502 * This attaches @handle to the running transaction (or creates one if there's
503 * not transaction running). Unlike jbd2_journal_start() this function cannot
504 * block on journal commit, checkpointing, or similar stuff. It can block on
505 * memory allocation or frozen journal though.
507 * Return 0 on success, non-zero on error - handle is freed in that case.
509 int jbd2_journal_start_reserved(handle_t *handle, unsigned int type,
510 unsigned int line_no)
512 journal_t *journal = handle->h_journal;
515 if (WARN_ON(!handle->h_reserved)) {
516 /* Someone passed in normal handle? Just stop it. */
517 jbd2_journal_stop(handle);
521 * Usefulness of mixing of reserved and unreserved handles is
522 * questionable. So far nobody seems to need it so just error out.
524 if (WARN_ON(current->journal_info)) {
525 jbd2_journal_free_reserved(handle);
529 handle->h_journal = NULL;
531 * GFP_NOFS is here because callers are likely from writeback or
532 * similarly constrained call sites
534 ret = start_this_handle(journal, handle, GFP_NOFS);
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
630 * Restart a handle for a multi-transaction filesystem
633 * If the jbd2_journal_extend() call above fails to grant new buffer credits
634 * to a running handle, a call to jbd2_journal_restart will commit the
635 * handle's transaction so far and reattach the handle to a new
636 * transaction capable of guaranteeing the requested number of
637 * credits. We preserve reserved handle if there's any attached to the
640 int jbd2__journal_restart(handle_t *handle, int nblocks, gfp_t gfp_mask)
642 transaction_t *transaction = handle->h_transaction;
645 int need_to_start, ret;
647 /* If we've had an abort of any type, don't even think about
648 * actually doing the restart! */
649 if (is_handle_aborted(handle))
651 journal = transaction->t_journal;
654 * First unlink the handle from its current transaction, and start the
657 J_ASSERT(atomic_read(&transaction->t_updates) > 0);
658 J_ASSERT(journal_current_handle() == handle);
660 read_lock(&journal->j_state_lock);
661 spin_lock(&transaction->t_handle_lock);
662 atomic_sub(handle->h_buffer_credits,
663 &transaction->t_outstanding_credits);
664 if (handle->h_rsv_handle) {
665 sub_reserved_credits(journal,
666 handle->h_rsv_handle->h_buffer_credits);
668 if (atomic_dec_and_test(&transaction->t_updates))
669 wake_up(&journal->j_wait_updates);
670 tid = transaction->t_tid;
671 spin_unlock(&transaction->t_handle_lock);
672 handle->h_transaction = NULL;
673 current->journal_info = NULL;
675 jbd_debug(2, "restarting handle %p\n", handle);
676 need_to_start = !tid_geq(journal->j_commit_request, tid);
677 read_unlock(&journal->j_state_lock);
679 jbd2_log_start_commit(journal, tid);
681 rwsem_release(&journal->j_trans_commit_map, 1, _THIS_IP_);
682 handle->h_buffer_credits = nblocks;
683 ret = start_this_handle(journal, handle, gfp_mask);
686 EXPORT_SYMBOL(jbd2__journal_restart);
689 int jbd2_journal_restart(handle_t *handle, int nblocks)
691 return jbd2__journal_restart(handle, nblocks, GFP_NOFS);
693 EXPORT_SYMBOL(jbd2_journal_restart);
696 * void jbd2_journal_lock_updates () - establish a transaction barrier.
697 * @journal: Journal to establish a barrier on.
699 * This locks out any further updates from being started, and blocks
700 * until all existing updates have completed, returning only once the
701 * journal is in a quiescent state with no updates running.
703 * The journal lock should not be held on entry.
705 void jbd2_journal_lock_updates(journal_t *journal)
709 jbd2_might_wait_for_commit(journal);
711 write_lock(&journal->j_state_lock);
712 ++journal->j_barrier_count;
714 /* Wait until there are no reserved handles */
715 if (atomic_read(&journal->j_reserved_credits)) {
716 write_unlock(&journal->j_state_lock);
717 wait_event(journal->j_wait_reserved,
718 atomic_read(&journal->j_reserved_credits) == 0);
719 write_lock(&journal->j_state_lock);
722 /* Wait until there are no running updates */
724 transaction_t *transaction = journal->j_running_transaction;
729 spin_lock(&transaction->t_handle_lock);
730 prepare_to_wait(&journal->j_wait_updates, &wait,
731 TASK_UNINTERRUPTIBLE);
732 if (!atomic_read(&transaction->t_updates)) {
733 spin_unlock(&transaction->t_handle_lock);
734 finish_wait(&journal->j_wait_updates, &wait);
737 spin_unlock(&transaction->t_handle_lock);
738 write_unlock(&journal->j_state_lock);
740 finish_wait(&journal->j_wait_updates, &wait);
741 write_lock(&journal->j_state_lock);
743 write_unlock(&journal->j_state_lock);
746 * We have now established a barrier against other normal updates, but
747 * we also need to barrier against other jbd2_journal_lock_updates() calls
748 * to make sure that we serialise special journal-locked operations
751 mutex_lock(&journal->j_barrier);
755 * void jbd2_journal_unlock_updates (journal_t* journal) - release barrier
756 * @journal: Journal to release the barrier on.
758 * Release a transaction barrier obtained with jbd2_journal_lock_updates().
760 * Should be called without the journal lock held.
762 void jbd2_journal_unlock_updates (journal_t *journal)
764 J_ASSERT(journal->j_barrier_count != 0);
766 mutex_unlock(&journal->j_barrier);
767 write_lock(&journal->j_state_lock);
768 --journal->j_barrier_count;
769 write_unlock(&journal->j_state_lock);
770 wake_up(&journal->j_wait_transaction_locked);
773 static void warn_dirty_buffer(struct buffer_head *bh)
776 "JBD2: Spotted dirty metadata buffer (dev = %pg, blocknr = %llu). "
777 "There's a risk of filesystem corruption in case of system "
779 bh->b_bdev, (unsigned long long)bh->b_blocknr);
782 /* Call t_frozen trigger and copy buffer data into jh->b_frozen_data. */
783 static void jbd2_freeze_jh_data(struct journal_head *jh)
788 struct buffer_head *bh = jh2bh(jh);
790 J_EXPECT_JH(jh, buffer_uptodate(bh), "Possible IO failure.\n");
792 offset = offset_in_page(bh->b_data);
793 source = kmap_atomic(page);
794 /* Fire data frozen trigger just before we copy the data */
795 jbd2_buffer_frozen_trigger(jh, source + offset, jh->b_triggers);
796 memcpy(jh->b_frozen_data, source + offset, bh->b_size);
797 kunmap_atomic(source);
800 * Now that the frozen data is saved off, we need to store any matching
803 jh->b_frozen_triggers = jh->b_triggers;
807 * If the buffer is already part of the current transaction, then there
808 * is nothing we need to do. If it is already part of a prior
809 * transaction which we are still committing to disk, then we need to
810 * make sure that we do not overwrite the old copy: we do copy-out to
811 * preserve the copy going to disk. We also account the buffer against
812 * the handle's metadata buffer credits (unless the buffer is already
813 * part of the transaction, that is).
817 do_get_write_access(handle_t *handle, struct journal_head *jh,
820 struct buffer_head *bh;
821 transaction_t *transaction = handle->h_transaction;
824 char *frozen_buffer = NULL;
825 unsigned long start_lock, time_lock;
827 if (is_handle_aborted(handle))
829 journal = transaction->t_journal;
831 jbd_debug(5, "journal_head %p, force_copy %d\n", jh, force_copy);
833 JBUFFER_TRACE(jh, "entry");
837 /* @@@ Need to check for errors here at some point. */
839 start_lock = jiffies;
841 jbd_lock_bh_state(bh);
843 /* If it takes too long to lock the buffer, trace it */
844 time_lock = jbd2_time_diff(start_lock, jiffies);
845 if (time_lock > HZ/10)
846 trace_jbd2_lock_buffer_stall(bh->b_bdev->bd_dev,
847 jiffies_to_msecs(time_lock));
849 /* We now hold the buffer lock so it is safe to query the buffer
850 * state. Is the buffer dirty?
852 * If so, there are two possibilities. The buffer may be
853 * non-journaled, and undergoing a quite legitimate writeback.
854 * Otherwise, it is journaled, and we don't expect dirty buffers
855 * in that state (the buffers should be marked JBD_Dirty
856 * instead.) So either the IO is being done under our own
857 * control and this is a bug, or it's a third party IO such as
858 * dump(8) (which may leave the buffer scheduled for read ---
859 * ie. locked but not dirty) or tune2fs (which may actually have
860 * the buffer dirtied, ugh.) */
862 if (buffer_dirty(bh)) {
864 * First question: is this buffer already part of the current
865 * transaction or the existing committing transaction?
867 if (jh->b_transaction) {
869 jh->b_transaction == transaction ||
871 journal->j_committing_transaction);
872 if (jh->b_next_transaction)
873 J_ASSERT_JH(jh, jh->b_next_transaction ==
875 warn_dirty_buffer(bh);
878 * In any case we need to clean the dirty flag and we must
879 * do it under the buffer lock to be sure we don't race
880 * with running write-out.
882 JBUFFER_TRACE(jh, "Journalling dirty buffer");
883 clear_buffer_dirty(bh);
884 set_buffer_jbddirty(bh);
890 if (is_handle_aborted(handle)) {
891 jbd_unlock_bh_state(bh);
897 * The buffer is already part of this transaction if b_transaction or
898 * b_next_transaction points to it
900 if (jh->b_transaction == transaction ||
901 jh->b_next_transaction == transaction)
905 * this is the first time this transaction is touching this buffer,
906 * reset the modified flag
911 * If the buffer is not journaled right now, we need to make sure it
912 * doesn't get written to disk before the caller actually commits the
915 if (!jh->b_transaction) {
916 JBUFFER_TRACE(jh, "no transaction");
917 J_ASSERT_JH(jh, !jh->b_next_transaction);
918 JBUFFER_TRACE(jh, "file as BJ_Reserved");
920 * Make sure all stores to jh (b_modified, b_frozen_data) are
921 * visible before attaching it to the running transaction.
922 * Paired with barrier in jbd2_write_access_granted()
925 spin_lock(&journal->j_list_lock);
926 __jbd2_journal_file_buffer(jh, transaction, BJ_Reserved);
927 spin_unlock(&journal->j_list_lock);
931 * If there is already a copy-out version of this buffer, then we don't
932 * need to make another one
934 if (jh->b_frozen_data) {
935 JBUFFER_TRACE(jh, "has frozen data");
936 J_ASSERT_JH(jh, jh->b_next_transaction == NULL);
940 JBUFFER_TRACE(jh, "owned by older transaction");
941 J_ASSERT_JH(jh, jh->b_next_transaction == NULL);
942 J_ASSERT_JH(jh, jh->b_transaction == journal->j_committing_transaction);
945 * There is one case we have to be very careful about. If the
946 * committing transaction is currently writing this buffer out to disk
947 * and has NOT made a copy-out, then we cannot modify the buffer
948 * contents at all right now. The essence of copy-out is that it is
949 * the extra copy, not the primary copy, which gets journaled. If the
950 * primary copy is already going to disk then we cannot do copy-out
953 if (buffer_shadow(bh)) {
954 JBUFFER_TRACE(jh, "on shadow: sleep");
955 jbd_unlock_bh_state(bh);
956 wait_on_bit_io(&bh->b_state, BH_Shadow, TASK_UNINTERRUPTIBLE);
961 * Only do the copy if the currently-owning transaction still needs it.
962 * If buffer isn't on BJ_Metadata list, the committing transaction is
963 * past that stage (here we use the fact that BH_Shadow is set under
964 * bh_state lock together with refiling to BJ_Shadow list and at this
965 * point we know the buffer doesn't have BH_Shadow set).
967 * Subtle point, though: if this is a get_undo_access, then we will be
968 * relying on the frozen_data to contain the new value of the
969 * committed_data record after the transaction, so we HAVE to force the
970 * frozen_data copy in that case.
972 if (jh->b_jlist == BJ_Metadata || force_copy) {
973 JBUFFER_TRACE(jh, "generate frozen data");
974 if (!frozen_buffer) {
975 JBUFFER_TRACE(jh, "allocate memory for buffer");
976 jbd_unlock_bh_state(bh);
977 frozen_buffer = jbd2_alloc(jh2bh(jh)->b_size,
978 GFP_NOFS | __GFP_NOFAIL);
981 jh->b_frozen_data = frozen_buffer;
982 frozen_buffer = NULL;
983 jbd2_freeze_jh_data(jh);
987 * Make sure all stores to jh (b_modified, b_frozen_data) are visible
988 * before attaching it to the running transaction. Paired with barrier
989 * in jbd2_write_access_granted()
992 jh->b_next_transaction = transaction;
995 jbd_unlock_bh_state(bh);
998 * If we are about to journal a buffer, then any revoke pending on it is
1001 jbd2_journal_cancel_revoke(handle, jh);
1004 if (unlikely(frozen_buffer)) /* It's usually NULL */
1005 jbd2_free(frozen_buffer, bh->b_size);
1007 JBUFFER_TRACE(jh, "exit");
1011 /* Fast check whether buffer is already attached to the required transaction */
1012 static bool jbd2_write_access_granted(handle_t *handle, struct buffer_head *bh,
1015 struct journal_head *jh;
1018 /* Dirty buffers require special handling... */
1019 if (buffer_dirty(bh))
1023 * RCU protects us from dereferencing freed pages. So the checks we do
1024 * are guaranteed not to oops. However the jh slab object can get freed
1025 * & reallocated while we work with it. So we have to be careful. When
1026 * we see jh attached to the running transaction, we know it must stay
1027 * so until the transaction is committed. Thus jh won't be freed and
1028 * will be attached to the same bh while we run. However it can
1029 * happen jh gets freed, reallocated, and attached to the transaction
1030 * just after we get pointer to it from bh. So we have to be careful
1031 * and recheck jh still belongs to our bh before we return success.
1034 if (!buffer_jbd(bh))
1036 /* This should be bh2jh() but that doesn't work with inline functions */
1037 jh = READ_ONCE(bh->b_private);
1040 /* For undo access buffer must have data copied */
1041 if (undo && !jh->b_committed_data)
1043 if (jh->b_transaction != handle->h_transaction &&
1044 jh->b_next_transaction != handle->h_transaction)
1047 * There are two reasons for the barrier here:
1048 * 1) Make sure to fetch b_bh after we did previous checks so that we
1049 * detect when jh went through free, realloc, attach to transaction
1050 * while we were checking. Paired with implicit barrier in that path.
1051 * 2) So that access to bh done after jbd2_write_access_granted()
1052 * doesn't get reordered and see inconsistent state of concurrent
1053 * do_get_write_access().
1056 if (unlikely(jh->b_bh != bh))
1065 * int jbd2_journal_get_write_access() - notify intent to modify a buffer for metadata (not data) update.
1066 * @handle: transaction to add buffer modifications to
1067 * @bh: bh to be used for metadata writes
1069 * Returns an error code or 0 on success.
1071 * In full data journalling mode the buffer may be of type BJ_AsyncData,
1072 * because we're write()ing a buffer which is also part of a shared mapping.
1075 int jbd2_journal_get_write_access(handle_t *handle, struct buffer_head *bh)
1077 struct journal_head *jh;
1080 if (jbd2_write_access_granted(handle, bh, false))
1083 jh = jbd2_journal_add_journal_head(bh);
1084 /* We do not want to get caught playing with fields which the
1085 * log thread also manipulates. Make sure that the buffer
1086 * completes any outstanding IO before proceeding. */
1087 rc = do_get_write_access(handle, jh, 0);
1088 jbd2_journal_put_journal_head(jh);
1094 * When the user wants to journal a newly created buffer_head
1095 * (ie. getblk() returned a new buffer and we are going to populate it
1096 * manually rather than reading off disk), then we need to keep the
1097 * buffer_head locked until it has been completely filled with new
1098 * data. In this case, we should be able to make the assertion that
1099 * the bh is not already part of an existing transaction.
1101 * The buffer should already be locked by the caller by this point.
1102 * There is no lock ranking violation: it was a newly created,
1103 * unlocked buffer beforehand. */
1106 * int jbd2_journal_get_create_access () - notify intent to use newly created bh
1107 * @handle: transaction to new buffer to
1110 * Call this if you create a new bh.
1112 int jbd2_journal_get_create_access(handle_t *handle, struct buffer_head *bh)
1114 transaction_t *transaction = handle->h_transaction;
1116 struct journal_head *jh = jbd2_journal_add_journal_head(bh);
1119 jbd_debug(5, "journal_head %p\n", jh);
1121 if (is_handle_aborted(handle))
1123 journal = transaction->t_journal;
1126 JBUFFER_TRACE(jh, "entry");
1128 * The buffer may already belong to this transaction due to pre-zeroing
1129 * in the filesystem's new_block code. It may also be on the previous,
1130 * committing transaction's lists, but it HAS to be in Forget state in
1131 * that case: the transaction must have deleted the buffer for it to be
1134 jbd_lock_bh_state(bh);
1135 J_ASSERT_JH(jh, (jh->b_transaction == transaction ||
1136 jh->b_transaction == NULL ||
1137 (jh->b_transaction == journal->j_committing_transaction &&
1138 jh->b_jlist == BJ_Forget)));
1140 J_ASSERT_JH(jh, jh->b_next_transaction == NULL);
1141 J_ASSERT_JH(jh, buffer_locked(jh2bh(jh)));
1143 if (jh->b_transaction == NULL) {
1145 * Previous jbd2_journal_forget() could have left the buffer
1146 * with jbddirty bit set because it was being committed. When
1147 * the commit finished, we've filed the buffer for
1148 * checkpointing and marked it dirty. Now we are reallocating
1149 * the buffer so the transaction freeing it must have
1150 * committed and so it's safe to clear the dirty bit.
1152 clear_buffer_dirty(jh2bh(jh));
1153 /* first access by this transaction */
1156 JBUFFER_TRACE(jh, "file as BJ_Reserved");
1157 spin_lock(&journal->j_list_lock);
1158 __jbd2_journal_file_buffer(jh, transaction, BJ_Reserved);
1159 spin_unlock(&journal->j_list_lock);
1160 } else if (jh->b_transaction == journal->j_committing_transaction) {
1161 /* first access by this transaction */
1164 JBUFFER_TRACE(jh, "set next transaction");
1165 spin_lock(&journal->j_list_lock);
1166 jh->b_next_transaction = transaction;
1167 spin_unlock(&journal->j_list_lock);
1169 jbd_unlock_bh_state(bh);
1172 * akpm: I added this. ext3_alloc_branch can pick up new indirect
1173 * blocks which contain freed but then revoked metadata. We need
1174 * to cancel the revoke in case we end up freeing it yet again
1175 * and the reallocating as data - this would cause a second revoke,
1176 * which hits an assertion error.
1178 JBUFFER_TRACE(jh, "cancelling revoke");
1179 jbd2_journal_cancel_revoke(handle, jh);
1181 jbd2_journal_put_journal_head(jh);
1186 * int jbd2_journal_get_undo_access() - Notify intent to modify metadata with
1187 * non-rewindable consequences
1188 * @handle: transaction
1189 * @bh: buffer to undo
1191 * Sometimes there is a need to distinguish between metadata which has
1192 * been committed to disk and that which has not. The ext3fs code uses
1193 * this for freeing and allocating space, we have to make sure that we
1194 * do not reuse freed space until the deallocation has been committed,
1195 * since if we overwrote that space we would make the delete
1196 * un-rewindable in case of a crash.
1198 * To deal with that, jbd2_journal_get_undo_access requests write access to a
1199 * buffer for parts of non-rewindable operations such as delete
1200 * operations on the bitmaps. The journaling code must keep a copy of
1201 * the buffer's contents prior to the undo_access call until such time
1202 * as we know that the buffer has definitely been committed to disk.
1204 * We never need to know which transaction the committed data is part
1205 * of, buffers touched here are guaranteed to be dirtied later and so
1206 * will be committed to a new transaction in due course, at which point
1207 * we can discard the old committed data pointer.
1209 * Returns error number or 0 on success.
1211 int jbd2_journal_get_undo_access(handle_t *handle, struct buffer_head *bh)
1214 struct journal_head *jh;
1215 char *committed_data = NULL;
1217 JBUFFER_TRACE(jh, "entry");
1218 if (jbd2_write_access_granted(handle, bh, true))
1221 jh = jbd2_journal_add_journal_head(bh);
1223 * Do this first --- it can drop the journal lock, so we want to
1224 * make sure that obtaining the committed_data is done
1225 * atomically wrt. completion of any outstanding commits.
1227 err = do_get_write_access(handle, jh, 1);
1232 if (!jh->b_committed_data)
1233 committed_data = jbd2_alloc(jh2bh(jh)->b_size,
1234 GFP_NOFS|__GFP_NOFAIL);
1236 jbd_lock_bh_state(bh);
1237 if (!jh->b_committed_data) {
1238 /* Copy out the current buffer contents into the
1239 * preserved, committed copy. */
1240 JBUFFER_TRACE(jh, "generate b_committed data");
1241 if (!committed_data) {
1242 jbd_unlock_bh_state(bh);
1246 jh->b_committed_data = committed_data;
1247 committed_data = NULL;
1248 memcpy(jh->b_committed_data, bh->b_data, bh->b_size);
1250 jbd_unlock_bh_state(bh);
1252 jbd2_journal_put_journal_head(jh);
1253 if (unlikely(committed_data))
1254 jbd2_free(committed_data, bh->b_size);
1259 * void jbd2_journal_set_triggers() - Add triggers for commit writeout
1260 * @bh: buffer to trigger on
1261 * @type: struct jbd2_buffer_trigger_type containing the trigger(s).
1263 * Set any triggers on this journal_head. This is always safe, because
1264 * triggers for a committing buffer will be saved off, and triggers for
1265 * a running transaction will match the buffer in that transaction.
1267 * Call with NULL to clear the triggers.
1269 void jbd2_journal_set_triggers(struct buffer_head *bh,
1270 struct jbd2_buffer_trigger_type *type)
1272 struct journal_head *jh = jbd2_journal_grab_journal_head(bh);
1276 jh->b_triggers = type;
1277 jbd2_journal_put_journal_head(jh);
1280 void jbd2_buffer_frozen_trigger(struct journal_head *jh, void *mapped_data,
1281 struct jbd2_buffer_trigger_type *triggers)
1283 struct buffer_head *bh = jh2bh(jh);
1285 if (!triggers || !triggers->t_frozen)
1288 triggers->t_frozen(triggers, bh, mapped_data, bh->b_size);
1291 void jbd2_buffer_abort_trigger(struct journal_head *jh,
1292 struct jbd2_buffer_trigger_type *triggers)
1294 if (!triggers || !triggers->t_abort)
1297 triggers->t_abort(triggers, jh2bh(jh));
1301 * int jbd2_journal_dirty_metadata() - mark a buffer as containing dirty metadata
1302 * @handle: transaction to add buffer to.
1303 * @bh: buffer to mark
1305 * mark dirty metadata which needs to be journaled as part of the current
1308 * The buffer must have previously had jbd2_journal_get_write_access()
1309 * called so that it has a valid journal_head attached to the buffer
1312 * The buffer is placed on the transaction's metadata list and is marked
1313 * as belonging to the transaction.
1315 * Returns error number or 0 on success.
1317 * Special care needs to be taken if the buffer already belongs to the
1318 * current committing transaction (in which case we should have frozen
1319 * data present for that commit). In that case, we don't relink the
1320 * buffer: that only gets done when the old transaction finally
1321 * completes its commit.
1323 int jbd2_journal_dirty_metadata(handle_t *handle, struct buffer_head *bh)
1325 transaction_t *transaction = handle->h_transaction;
1327 struct journal_head *jh;
1330 if (is_handle_aborted(handle))
1332 if (!buffer_jbd(bh)) {
1337 * We don't grab jh reference here since the buffer must be part
1338 * of the running transaction.
1342 * This and the following assertions are unreliable since we may see jh
1343 * in inconsistent state unless we grab bh_state lock. But this is
1344 * crucial to catch bugs so let's do a reliable check until the
1345 * lockless handling is fully proven.
1347 if (jh->b_transaction != transaction &&
1348 jh->b_next_transaction != transaction) {
1349 jbd_lock_bh_state(bh);
1350 J_ASSERT_JH(jh, jh->b_transaction == transaction ||
1351 jh->b_next_transaction == transaction);
1352 jbd_unlock_bh_state(bh);
1354 if (jh->b_modified == 1) {
1355 /* If it's in our transaction it must be in BJ_Metadata list. */
1356 if (jh->b_transaction == transaction &&
1357 jh->b_jlist != BJ_Metadata) {
1358 jbd_lock_bh_state(bh);
1359 J_ASSERT_JH(jh, jh->b_transaction != transaction ||
1360 jh->b_jlist == BJ_Metadata);
1361 jbd_unlock_bh_state(bh);
1366 journal = transaction->t_journal;
1367 jbd_debug(5, "journal_head %p\n", jh);
1368 JBUFFER_TRACE(jh, "entry");
1370 jbd_lock_bh_state(bh);
1372 if (jh->b_modified == 0) {
1374 * This buffer's got modified and becoming part
1375 * of the transaction. This needs to be done
1376 * once a transaction -bzzz
1379 if (handle->h_buffer_credits <= 0) {
1383 handle->h_buffer_credits--;
1387 * fastpath, to avoid expensive locking. If this buffer is already
1388 * on the running transaction's metadata list there is nothing to do.
1389 * Nobody can take it off again because there is a handle open.
1390 * I _think_ we're OK here with SMP barriers - a mistaken decision will
1391 * result in this test being false, so we go in and take the locks.
1393 if (jh->b_transaction == transaction && jh->b_jlist == BJ_Metadata) {
1394 JBUFFER_TRACE(jh, "fastpath");
1395 if (unlikely(jh->b_transaction !=
1396 journal->j_running_transaction)) {
1397 printk(KERN_ERR "JBD2: %s: "
1398 "jh->b_transaction (%llu, %p, %u) != "
1399 "journal->j_running_transaction (%p, %u)\n",
1401 (unsigned long long) bh->b_blocknr,
1403 jh->b_transaction ? jh->b_transaction->t_tid : 0,
1404 journal->j_running_transaction,
1405 journal->j_running_transaction ?
1406 journal->j_running_transaction->t_tid : 0);
1412 set_buffer_jbddirty(bh);
1415 * Metadata already on the current transaction list doesn't
1416 * need to be filed. Metadata on another transaction's list must
1417 * be committing, and will be refiled once the commit completes:
1418 * leave it alone for now.
1420 if (jh->b_transaction != transaction) {
1421 JBUFFER_TRACE(jh, "already on other transaction");
1422 if (unlikely(((jh->b_transaction !=
1423 journal->j_committing_transaction)) ||
1424 (jh->b_next_transaction != transaction))) {
1425 printk(KERN_ERR "jbd2_journal_dirty_metadata: %s: "
1426 "bad jh for block %llu: "
1427 "transaction (%p, %u), "
1428 "jh->b_transaction (%p, %u), "
1429 "jh->b_next_transaction (%p, %u), jlist %u\n",
1431 (unsigned long long) bh->b_blocknr,
1432 transaction, transaction->t_tid,
1435 jh->b_transaction->t_tid : 0,
1436 jh->b_next_transaction,
1437 jh->b_next_transaction ?
1438 jh->b_next_transaction->t_tid : 0,
1443 /* And this case is illegal: we can't reuse another
1444 * transaction's data buffer, ever. */
1448 /* That test should have eliminated the following case: */
1449 J_ASSERT_JH(jh, jh->b_frozen_data == NULL);
1451 JBUFFER_TRACE(jh, "file as BJ_Metadata");
1452 spin_lock(&journal->j_list_lock);
1453 __jbd2_journal_file_buffer(jh, transaction, BJ_Metadata);
1454 spin_unlock(&journal->j_list_lock);
1456 jbd_unlock_bh_state(bh);
1458 JBUFFER_TRACE(jh, "exit");
1463 * void jbd2_journal_forget() - bforget() for potentially-journaled buffers.
1464 * @handle: transaction handle
1465 * @bh: bh to 'forget'
1467 * We can only do the bforget if there are no commits pending against the
1468 * buffer. If the buffer is dirty in the current running transaction we
1469 * can safely unlink it.
1471 * bh may not be a journalled buffer at all - it may be a non-JBD
1472 * buffer which came off the hashtable. Check for this.
1474 * Decrements bh->b_count by one.
1476 * Allow this call even if the handle has aborted --- it may be part of
1477 * the caller's cleanup after an abort.
1479 int jbd2_journal_forget (handle_t *handle, struct buffer_head *bh)
1481 transaction_t *transaction = handle->h_transaction;
1483 struct journal_head *jh;
1484 int drop_reserve = 0;
1486 int was_modified = 0;
1488 if (is_handle_aborted(handle))
1490 journal = transaction->t_journal;
1492 BUFFER_TRACE(bh, "entry");
1494 jbd_lock_bh_state(bh);
1496 if (!buffer_jbd(bh))
1500 /* Critical error: attempting to delete a bitmap buffer, maybe?
1501 * Don't do any jbd operations, and return an error. */
1502 if (!J_EXPECT_JH(jh, !jh->b_committed_data,
1503 "inconsistent data on disk")) {
1508 /* keep track of whether or not this transaction modified us */
1509 was_modified = jh->b_modified;
1512 * The buffer's going from the transaction, we must drop
1513 * all references -bzzz
1517 if (jh->b_transaction == transaction) {
1518 J_ASSERT_JH(jh, !jh->b_frozen_data);
1520 /* If we are forgetting a buffer which is already part
1521 * of this transaction, then we can just drop it from
1522 * the transaction immediately. */
1523 clear_buffer_dirty(bh);
1524 clear_buffer_jbddirty(bh);
1526 JBUFFER_TRACE(jh, "belongs to current transaction: unfile");
1529 * we only want to drop a reference if this transaction
1530 * modified the buffer
1536 * We are no longer going to journal this buffer.
1537 * However, the commit of this transaction is still
1538 * important to the buffer: the delete that we are now
1539 * processing might obsolete an old log entry, so by
1540 * committing, we can satisfy the buffer's checkpoint.
1542 * So, if we have a checkpoint on the buffer, we should
1543 * now refile the buffer on our BJ_Forget list so that
1544 * we know to remove the checkpoint after we commit.
1547 spin_lock(&journal->j_list_lock);
1548 if (jh->b_cp_transaction) {
1549 __jbd2_journal_temp_unlink_buffer(jh);
1550 __jbd2_journal_file_buffer(jh, transaction, BJ_Forget);
1552 __jbd2_journal_unfile_buffer(jh);
1553 if (!buffer_jbd(bh)) {
1554 spin_unlock(&journal->j_list_lock);
1555 jbd_unlock_bh_state(bh);
1560 spin_unlock(&journal->j_list_lock);
1561 } else if (jh->b_transaction) {
1562 J_ASSERT_JH(jh, (jh->b_transaction ==
1563 journal->j_committing_transaction));
1564 /* However, if the buffer is still owned by a prior
1565 * (committing) transaction, we can't drop it yet... */
1566 JBUFFER_TRACE(jh, "belongs to older transaction");
1567 /* ... but we CAN drop it from the new transaction if we
1568 * have also modified it since the original commit. */
1570 if (jh->b_next_transaction) {
1571 J_ASSERT(jh->b_next_transaction == transaction);
1572 spin_lock(&journal->j_list_lock);
1573 jh->b_next_transaction = NULL;
1574 spin_unlock(&journal->j_list_lock);
1577 * only drop a reference if this transaction modified
1586 jbd_unlock_bh_state(bh);
1590 /* no need to reserve log space for this block -bzzz */
1591 handle->h_buffer_credits++;
1597 * int jbd2_journal_stop() - complete a transaction
1598 * @handle: transaction to complete.
1600 * All done for a particular handle.
1602 * There is not much action needed here. We just return any remaining
1603 * buffer credits to the transaction and remove the handle. The only
1604 * complication is that we need to start a commit operation if the
1605 * filesystem is marked for synchronous update.
1607 * jbd2_journal_stop itself will not usually return an error, but it may
1608 * do so in unusual circumstances. In particular, expect it to
1609 * return -EIO if a jbd2_journal_abort has been executed since the
1610 * transaction began.
1612 int jbd2_journal_stop(handle_t *handle)
1614 transaction_t *transaction = handle->h_transaction;
1616 int err = 0, wait_for_commit = 0;
1622 * Handle is already detached from the transaction so
1623 * there is nothing to do other than decrease a refcount,
1624 * or free the handle if refcount drops to zero
1626 if (--handle->h_ref > 0) {
1627 jbd_debug(4, "h_ref %d -> %d\n", handle->h_ref + 1,
1631 if (handle->h_rsv_handle)
1632 jbd2_free_handle(handle->h_rsv_handle);
1636 journal = transaction->t_journal;
1638 J_ASSERT(journal_current_handle() == handle);
1640 if (is_handle_aborted(handle))
1643 J_ASSERT(atomic_read(&transaction->t_updates) > 0);
1645 if (--handle->h_ref > 0) {
1646 jbd_debug(4, "h_ref %d -> %d\n", handle->h_ref + 1,
1651 jbd_debug(4, "Handle %p going down\n", handle);
1652 trace_jbd2_handle_stats(journal->j_fs_dev->bd_dev,
1654 handle->h_type, handle->h_line_no,
1655 jiffies - handle->h_start_jiffies,
1656 handle->h_sync, handle->h_requested_credits,
1657 (handle->h_requested_credits -
1658 handle->h_buffer_credits));
1661 * Implement synchronous transaction batching. If the handle
1662 * was synchronous, don't force a commit immediately. Let's
1663 * yield and let another thread piggyback onto this
1664 * transaction. Keep doing that while new threads continue to
1665 * arrive. It doesn't cost much - we're about to run a commit
1666 * and sleep on IO anyway. Speeds up many-threaded, many-dir
1667 * operations by 30x or more...
1669 * We try and optimize the sleep time against what the
1670 * underlying disk can do, instead of having a static sleep
1671 * time. This is useful for the case where our storage is so
1672 * fast that it is more optimal to go ahead and force a flush
1673 * and wait for the transaction to be committed than it is to
1674 * wait for an arbitrary amount of time for new writers to
1675 * join the transaction. We achieve this by measuring how
1676 * long it takes to commit a transaction, and compare it with
1677 * how long this transaction has been running, and if run time
1678 * < commit time then we sleep for the delta and commit. This
1679 * greatly helps super fast disks that would see slowdowns as
1680 * more threads started doing fsyncs.
1682 * But don't do this if this process was the most recent one
1683 * to perform a synchronous write. We do this to detect the
1684 * case where a single process is doing a stream of sync
1685 * writes. No point in waiting for joiners in that case.
1687 * Setting max_batch_time to 0 disables this completely.
1690 if (handle->h_sync && journal->j_last_sync_writer != pid &&
1691 journal->j_max_batch_time) {
1692 u64 commit_time, trans_time;
1694 journal->j_last_sync_writer = pid;
1696 read_lock(&journal->j_state_lock);
1697 commit_time = journal->j_average_commit_time;
1698 read_unlock(&journal->j_state_lock);
1700 trans_time = ktime_to_ns(ktime_sub(ktime_get(),
1701 transaction->t_start_time));
1703 commit_time = max_t(u64, commit_time,
1704 1000*journal->j_min_batch_time);
1705 commit_time = min_t(u64, commit_time,
1706 1000*journal->j_max_batch_time);
1708 if (trans_time < commit_time) {
1709 ktime_t expires = ktime_add_ns(ktime_get(),
1711 set_current_state(TASK_UNINTERRUPTIBLE);
1712 schedule_hrtimeout(&expires, HRTIMER_MODE_ABS);
1717 transaction->t_synchronous_commit = 1;
1718 current->journal_info = NULL;
1719 atomic_sub(handle->h_buffer_credits,
1720 &transaction->t_outstanding_credits);
1723 * If the handle is marked SYNC, we need to set another commit
1724 * going! We also want to force a commit if the current
1725 * transaction is occupying too much of the log, or if the
1726 * transaction is too old now.
1728 if (handle->h_sync ||
1729 (atomic_read(&transaction->t_outstanding_credits) >
1730 journal->j_max_transaction_buffers) ||
1731 time_after_eq(jiffies, transaction->t_expires)) {
1732 /* Do this even for aborted journals: an abort still
1733 * completes the commit thread, it just doesn't write
1734 * anything to disk. */
1736 jbd_debug(2, "transaction too old, requesting commit for "
1737 "handle %p\n", handle);
1738 /* This is non-blocking */
1739 jbd2_log_start_commit(journal, transaction->t_tid);
1742 * Special case: JBD2_SYNC synchronous updates require us
1743 * to wait for the commit to complete.
1745 if (handle->h_sync && !(current->flags & PF_MEMALLOC))
1746 wait_for_commit = 1;
1750 * Once we drop t_updates, if it goes to zero the transaction
1751 * could start committing on us and eventually disappear. So
1752 * once we do this, we must not dereference transaction
1755 tid = transaction->t_tid;
1756 if (atomic_dec_and_test(&transaction->t_updates)) {
1757 wake_up(&journal->j_wait_updates);
1758 if (journal->j_barrier_count)
1759 wake_up(&journal->j_wait_transaction_locked);
1762 rwsem_release(&journal->j_trans_commit_map, 1, _THIS_IP_);
1764 if (wait_for_commit)
1765 err = jbd2_log_wait_commit(journal, tid);
1767 if (handle->h_rsv_handle)
1768 jbd2_journal_free_reserved(handle->h_rsv_handle);
1771 * Scope of the GFP_NOFS context is over here and so we can restore the
1772 * original alloc context.
1774 memalloc_nofs_restore(handle->saved_alloc_context);
1775 jbd2_free_handle(handle);
1781 * List management code snippets: various functions for manipulating the
1782 * transaction buffer lists.
1787 * Append a buffer to a transaction list, given the transaction's list head
1790 * j_list_lock is held.
1792 * jbd_lock_bh_state(jh2bh(jh)) is held.
1796 __blist_add_buffer(struct journal_head **list, struct journal_head *jh)
1799 jh->b_tnext = jh->b_tprev = jh;
1802 /* Insert at the tail of the list to preserve order */
1803 struct journal_head *first = *list, *last = first->b_tprev;
1805 jh->b_tnext = first;
1806 last->b_tnext = first->b_tprev = jh;
1811 * Remove a buffer from a transaction list, given the transaction's list
1814 * Called with j_list_lock held, and the journal may not be locked.
1816 * jbd_lock_bh_state(jh2bh(jh)) is held.
1820 __blist_del_buffer(struct journal_head **list, struct journal_head *jh)
1823 *list = jh->b_tnext;
1827 jh->b_tprev->b_tnext = jh->b_tnext;
1828 jh->b_tnext->b_tprev = jh->b_tprev;
1832 * Remove a buffer from the appropriate transaction list.
1834 * Note that this function can *change* the value of
1835 * bh->b_transaction->t_buffers, t_forget, t_shadow_list, t_log_list or
1836 * t_reserved_list. If the caller is holding onto a copy of one of these
1837 * pointers, it could go bad. Generally the caller needs to re-read the
1838 * pointer from the transaction_t.
1840 * Called under j_list_lock.
1842 static void __jbd2_journal_temp_unlink_buffer(struct journal_head *jh)
1844 struct journal_head **list = NULL;
1845 transaction_t *transaction;
1846 struct buffer_head *bh = jh2bh(jh);
1848 J_ASSERT_JH(jh, jbd_is_locked_bh_state(bh));
1849 transaction = jh->b_transaction;
1851 assert_spin_locked(&transaction->t_journal->j_list_lock);
1853 J_ASSERT_JH(jh, jh->b_jlist < BJ_Types);
1854 if (jh->b_jlist != BJ_None)
1855 J_ASSERT_JH(jh, transaction != NULL);
1857 switch (jh->b_jlist) {
1861 transaction->t_nr_buffers--;
1862 J_ASSERT_JH(jh, transaction->t_nr_buffers >= 0);
1863 list = &transaction->t_buffers;
1866 list = &transaction->t_forget;
1869 list = &transaction->t_shadow_list;
1872 list = &transaction->t_reserved_list;
1876 __blist_del_buffer(list, jh);
1877 jh->b_jlist = BJ_None;
1878 if (transaction && is_journal_aborted(transaction->t_journal))
1879 clear_buffer_jbddirty(bh);
1880 else if (test_clear_buffer_jbddirty(bh))
1881 mark_buffer_dirty(bh); /* Expose it to the VM */
1885 * Remove buffer from all transactions.
1887 * Called with bh_state lock and j_list_lock
1889 * jh and bh may be already freed when this function returns.
1891 static void __jbd2_journal_unfile_buffer(struct journal_head *jh)
1893 __jbd2_journal_temp_unlink_buffer(jh);
1894 jh->b_transaction = NULL;
1895 jbd2_journal_put_journal_head(jh);
1898 void jbd2_journal_unfile_buffer(journal_t *journal, struct journal_head *jh)
1900 struct buffer_head *bh = jh2bh(jh);
1902 /* Get reference so that buffer cannot be freed before we unlock it */
1904 jbd_lock_bh_state(bh);
1905 spin_lock(&journal->j_list_lock);
1906 __jbd2_journal_unfile_buffer(jh);
1907 spin_unlock(&journal->j_list_lock);
1908 jbd_unlock_bh_state(bh);
1913 * Called from jbd2_journal_try_to_free_buffers().
1915 * Called under jbd_lock_bh_state(bh)
1918 __journal_try_to_free_buffer(journal_t *journal, struct buffer_head *bh)
1920 struct journal_head *jh;
1924 if (buffer_locked(bh) || buffer_dirty(bh))
1927 if (jh->b_next_transaction != NULL || jh->b_transaction != NULL)
1930 spin_lock(&journal->j_list_lock);
1931 if (jh->b_cp_transaction != NULL) {
1932 /* written-back checkpointed metadata buffer */
1933 JBUFFER_TRACE(jh, "remove from checkpoint list");
1934 __jbd2_journal_remove_checkpoint(jh);
1936 spin_unlock(&journal->j_list_lock);
1942 * int jbd2_journal_try_to_free_buffers() - try to free page buffers.
1943 * @journal: journal for operation
1944 * @page: to try and free
1945 * @gfp_mask: we use the mask to detect how hard should we try to release
1946 * buffers. If __GFP_DIRECT_RECLAIM and __GFP_FS is set, we wait for commit
1947 * code to release the buffers.
1950 * For all the buffers on this page,
1951 * if they are fully written out ordered data, move them onto BUF_CLEAN
1952 * so try_to_free_buffers() can reap them.
1954 * This function returns non-zero if we wish try_to_free_buffers()
1955 * to be called. We do this if the page is releasable by try_to_free_buffers().
1956 * We also do it if the page has locked or dirty buffers and the caller wants
1957 * us to perform sync or async writeout.
1959 * This complicates JBD locking somewhat. We aren't protected by the
1960 * BKL here. We wish to remove the buffer from its committing or
1961 * running transaction's ->t_datalist via __jbd2_journal_unfile_buffer.
1963 * This may *change* the value of transaction_t->t_datalist, so anyone
1964 * who looks at t_datalist needs to lock against this function.
1966 * Even worse, someone may be doing a jbd2_journal_dirty_data on this
1967 * buffer. So we need to lock against that. jbd2_journal_dirty_data()
1968 * will come out of the lock with the buffer dirty, which makes it
1969 * ineligible for release here.
1971 * Who else is affected by this? hmm... Really the only contender
1972 * is do_get_write_access() - it could be looking at the buffer while
1973 * journal_try_to_free_buffer() is changing its state. But that
1974 * cannot happen because we never reallocate freed data as metadata
1975 * while the data is part of a transaction. Yes?
1977 * Return 0 on failure, 1 on success
1979 int jbd2_journal_try_to_free_buffers(journal_t *journal,
1980 struct page *page, gfp_t gfp_mask)
1982 struct buffer_head *head;
1983 struct buffer_head *bh;
1986 J_ASSERT(PageLocked(page));
1988 head = page_buffers(page);
1991 struct journal_head *jh;
1994 * We take our own ref against the journal_head here to avoid
1995 * having to add tons of locking around each instance of
1996 * jbd2_journal_put_journal_head().
1998 jh = jbd2_journal_grab_journal_head(bh);
2002 jbd_lock_bh_state(bh);
2003 __journal_try_to_free_buffer(journal, bh);
2004 jbd2_journal_put_journal_head(jh);
2005 jbd_unlock_bh_state(bh);
2008 } while ((bh = bh->b_this_page) != head);
2010 ret = try_to_free_buffers(page);
2017 * This buffer is no longer needed. If it is on an older transaction's
2018 * checkpoint list we need to record it on this transaction's forget list
2019 * to pin this buffer (and hence its checkpointing transaction) down until
2020 * this transaction commits. If the buffer isn't on a checkpoint list, we
2022 * Returns non-zero if JBD no longer has an interest in the buffer.
2024 * Called under j_list_lock.
2026 * Called under jbd_lock_bh_state(bh).
2028 static int __dispose_buffer(struct journal_head *jh, transaction_t *transaction)
2031 struct buffer_head *bh = jh2bh(jh);
2033 if (jh->b_cp_transaction) {
2034 JBUFFER_TRACE(jh, "on running+cp transaction");
2035 __jbd2_journal_temp_unlink_buffer(jh);
2037 * We don't want to write the buffer anymore, clear the
2038 * bit so that we don't confuse checks in
2039 * __journal_file_buffer
2041 clear_buffer_dirty(bh);
2042 __jbd2_journal_file_buffer(jh, transaction, BJ_Forget);
2045 JBUFFER_TRACE(jh, "on running transaction");
2046 __jbd2_journal_unfile_buffer(jh);
2052 * jbd2_journal_invalidatepage
2054 * This code is tricky. It has a number of cases to deal with.
2056 * There are two invariants which this code relies on:
2058 * i_size must be updated on disk before we start calling invalidatepage on the
2061 * This is done in ext3 by defining an ext3_setattr method which
2062 * updates i_size before truncate gets going. By maintaining this
2063 * invariant, we can be sure that it is safe to throw away any buffers
2064 * attached to the current transaction: once the transaction commits,
2065 * we know that the data will not be needed.
2067 * Note however that we can *not* throw away data belonging to the
2068 * previous, committing transaction!
2070 * Any disk blocks which *are* part of the previous, committing
2071 * transaction (and which therefore cannot be discarded immediately) are
2072 * not going to be reused in the new running transaction
2074 * The bitmap committed_data images guarantee this: any block which is
2075 * allocated in one transaction and removed in the next will be marked
2076 * as in-use in the committed_data bitmap, so cannot be reused until
2077 * the next transaction to delete the block commits. This means that
2078 * leaving committing buffers dirty is quite safe: the disk blocks
2079 * cannot be reallocated to a different file and so buffer aliasing is
2083 * The above applies mainly to ordered data mode. In writeback mode we
2084 * don't make guarantees about the order in which data hits disk --- in
2085 * particular we don't guarantee that new dirty data is flushed before
2086 * transaction commit --- so it is always safe just to discard data
2087 * immediately in that mode. --sct
2091 * The journal_unmap_buffer helper function returns zero if the buffer
2092 * concerned remains pinned as an anonymous buffer belonging to an older
2095 * We're outside-transaction here. Either or both of j_running_transaction
2096 * and j_committing_transaction may be NULL.
2098 static int journal_unmap_buffer(journal_t *journal, struct buffer_head *bh,
2101 transaction_t *transaction;
2102 struct journal_head *jh;
2105 BUFFER_TRACE(bh, "entry");
2108 * It is safe to proceed here without the j_list_lock because the
2109 * buffers cannot be stolen by try_to_free_buffers as long as we are
2110 * holding the page lock. --sct
2113 if (!buffer_jbd(bh))
2114 goto zap_buffer_unlocked;
2116 /* OK, we have data buffer in journaled mode */
2117 write_lock(&journal->j_state_lock);
2118 jbd_lock_bh_state(bh);
2119 spin_lock(&journal->j_list_lock);
2121 jh = jbd2_journal_grab_journal_head(bh);
2123 goto zap_buffer_no_jh;
2126 * We cannot remove the buffer from checkpoint lists until the
2127 * transaction adding inode to orphan list (let's call it T)
2128 * is committed. Otherwise if the transaction changing the
2129 * buffer would be cleaned from the journal before T is
2130 * committed, a crash will cause that the correct contents of
2131 * the buffer will be lost. On the other hand we have to
2132 * clear the buffer dirty bit at latest at the moment when the
2133 * transaction marking the buffer as freed in the filesystem
2134 * structures is committed because from that moment on the
2135 * block can be reallocated and used by a different page.
2136 * Since the block hasn't been freed yet but the inode has
2137 * already been added to orphan list, it is safe for us to add
2138 * the buffer to BJ_Forget list of the newest transaction.
2140 * Also we have to clear buffer_mapped flag of a truncated buffer
2141 * because the buffer_head may be attached to the page straddling
2142 * i_size (can happen only when blocksize < pagesize) and thus the
2143 * buffer_head can be reused when the file is extended again. So we end
2144 * up keeping around invalidated buffers attached to transactions'
2145 * BJ_Forget list just to stop checkpointing code from cleaning up
2146 * the transaction this buffer was modified in.
2148 transaction = jh->b_transaction;
2149 if (transaction == NULL) {
2150 /* First case: not on any transaction. If it
2151 * has no checkpoint link, then we can zap it:
2152 * it's a writeback-mode buffer so we don't care
2153 * if it hits disk safely. */
2154 if (!jh->b_cp_transaction) {
2155 JBUFFER_TRACE(jh, "not on any transaction: zap");
2159 if (!buffer_dirty(bh)) {
2160 /* bdflush has written it. We can drop it now */
2161 __jbd2_journal_remove_checkpoint(jh);
2165 /* OK, it must be in the journal but still not
2166 * written fully to disk: it's metadata or
2167 * journaled data... */
2169 if (journal->j_running_transaction) {
2170 /* ... and once the current transaction has
2171 * committed, the buffer won't be needed any
2173 JBUFFER_TRACE(jh, "checkpointed: add to BJ_Forget");
2174 may_free = __dispose_buffer(jh,
2175 journal->j_running_transaction);
2178 /* There is no currently-running transaction. So the
2179 * orphan record which we wrote for this file must have
2180 * passed into commit. We must attach this buffer to
2181 * the committing transaction, if it exists. */
2182 if (journal->j_committing_transaction) {
2183 JBUFFER_TRACE(jh, "give to committing trans");
2184 may_free = __dispose_buffer(jh,
2185 journal->j_committing_transaction);
2188 /* The orphan record's transaction has
2189 * committed. We can cleanse this buffer */
2190 clear_buffer_jbddirty(bh);
2191 __jbd2_journal_remove_checkpoint(jh);
2195 } else if (transaction == journal->j_committing_transaction) {
2196 JBUFFER_TRACE(jh, "on committing transaction");
2198 * The buffer is committing, we simply cannot touch
2199 * it. If the page is straddling i_size we have to wait
2200 * for commit and try again.
2203 jbd2_journal_put_journal_head(jh);
2204 spin_unlock(&journal->j_list_lock);
2205 jbd_unlock_bh_state(bh);
2206 write_unlock(&journal->j_state_lock);
2210 * OK, buffer won't be reachable after truncate. We just set
2211 * j_next_transaction to the running transaction (if there is
2212 * one) and mark buffer as freed so that commit code knows it
2213 * should clear dirty bits when it is done with the buffer.
2215 set_buffer_freed(bh);
2216 if (journal->j_running_transaction && buffer_jbddirty(bh))
2217 jh->b_next_transaction = journal->j_running_transaction;
2218 jbd2_journal_put_journal_head(jh);
2219 spin_unlock(&journal->j_list_lock);
2220 jbd_unlock_bh_state(bh);
2221 write_unlock(&journal->j_state_lock);
2224 /* Good, the buffer belongs to the running transaction.
2225 * We are writing our own transaction's data, not any
2226 * previous one's, so it is safe to throw it away
2227 * (remember that we expect the filesystem to have set
2228 * i_size already for this truncate so recovery will not
2229 * expose the disk blocks we are discarding here.) */
2230 J_ASSERT_JH(jh, transaction == journal->j_running_transaction);
2231 JBUFFER_TRACE(jh, "on running transaction");
2232 may_free = __dispose_buffer(jh, transaction);
2237 * This is tricky. Although the buffer is truncated, it may be reused
2238 * if blocksize < pagesize and it is attached to the page straddling
2239 * EOF. Since the buffer might have been added to BJ_Forget list of the
2240 * running transaction, journal_get_write_access() won't clear
2241 * b_modified and credit accounting gets confused. So clear b_modified
2245 jbd2_journal_put_journal_head(jh);
2247 spin_unlock(&journal->j_list_lock);
2248 jbd_unlock_bh_state(bh);
2249 write_unlock(&journal->j_state_lock);
2250 zap_buffer_unlocked:
2251 clear_buffer_dirty(bh);
2252 J_ASSERT_BH(bh, !buffer_jbddirty(bh));
2253 clear_buffer_mapped(bh);
2254 clear_buffer_req(bh);
2255 clear_buffer_new(bh);
2256 clear_buffer_delay(bh);
2257 clear_buffer_unwritten(bh);
2263 * void jbd2_journal_invalidatepage()
2264 * @journal: journal to use for flush...
2265 * @page: page to flush
2266 * @offset: start of the range to invalidate
2267 * @length: length of the range to invalidate
2269 * Reap page buffers containing data after in the specified range in page.
2270 * Can return -EBUSY if buffers are part of the committing transaction and
2271 * the page is straddling i_size. Caller then has to wait for current commit
2274 int jbd2_journal_invalidatepage(journal_t *journal,
2276 unsigned int offset,
2277 unsigned int length)
2279 struct buffer_head *head, *bh, *next;
2280 unsigned int stop = offset + length;
2281 unsigned int curr_off = 0;
2282 int partial_page = (offset || length < PAGE_SIZE);
2286 if (!PageLocked(page))
2288 if (!page_has_buffers(page))
2291 BUG_ON(stop > PAGE_SIZE || stop < length);
2293 /* We will potentially be playing with lists other than just the
2294 * data lists (especially for journaled data mode), so be
2295 * cautious in our locking. */
2297 head = bh = page_buffers(page);
2299 unsigned int next_off = curr_off + bh->b_size;
2300 next = bh->b_this_page;
2302 if (next_off > stop)
2305 if (offset <= curr_off) {
2306 /* This block is wholly outside the truncation point */
2308 ret = journal_unmap_buffer(journal, bh, partial_page);
2314 curr_off = next_off;
2317 } while (bh != head);
2319 if (!partial_page) {
2320 if (may_free && try_to_free_buffers(page))
2321 J_ASSERT(!page_has_buffers(page));
2327 * File a buffer on the given transaction list.
2329 void __jbd2_journal_file_buffer(struct journal_head *jh,
2330 transaction_t *transaction, int jlist)
2332 struct journal_head **list = NULL;
2334 struct buffer_head *bh = jh2bh(jh);
2336 J_ASSERT_JH(jh, jbd_is_locked_bh_state(bh));
2337 assert_spin_locked(&transaction->t_journal->j_list_lock);
2339 J_ASSERT_JH(jh, jh->b_jlist < BJ_Types);
2340 J_ASSERT_JH(jh, jh->b_transaction == transaction ||
2341 jh->b_transaction == NULL);
2343 if (jh->b_transaction && jh->b_jlist == jlist)
2346 if (jlist == BJ_Metadata || jlist == BJ_Reserved ||
2347 jlist == BJ_Shadow || jlist == BJ_Forget) {
2349 * For metadata buffers, we track dirty bit in buffer_jbddirty
2350 * instead of buffer_dirty. We should not see a dirty bit set
2351 * here because we clear it in do_get_write_access but e.g.
2352 * tune2fs can modify the sb and set the dirty bit at any time
2353 * so we try to gracefully handle that.
2355 if (buffer_dirty(bh))
2356 warn_dirty_buffer(bh);
2357 if (test_clear_buffer_dirty(bh) ||
2358 test_clear_buffer_jbddirty(bh))
2362 if (jh->b_transaction)
2363 __jbd2_journal_temp_unlink_buffer(jh);
2365 jbd2_journal_grab_journal_head(bh);
2366 jh->b_transaction = transaction;
2370 J_ASSERT_JH(jh, !jh->b_committed_data);
2371 J_ASSERT_JH(jh, !jh->b_frozen_data);
2374 transaction->t_nr_buffers++;
2375 list = &transaction->t_buffers;
2378 list = &transaction->t_forget;
2381 list = &transaction->t_shadow_list;
2384 list = &transaction->t_reserved_list;
2388 __blist_add_buffer(list, jh);
2389 jh->b_jlist = jlist;
2392 set_buffer_jbddirty(bh);
2395 void jbd2_journal_file_buffer(struct journal_head *jh,
2396 transaction_t *transaction, int jlist)
2398 jbd_lock_bh_state(jh2bh(jh));
2399 spin_lock(&transaction->t_journal->j_list_lock);
2400 __jbd2_journal_file_buffer(jh, transaction, jlist);
2401 spin_unlock(&transaction->t_journal->j_list_lock);
2402 jbd_unlock_bh_state(jh2bh(jh));
2406 * Remove a buffer from its current buffer list in preparation for
2407 * dropping it from its current transaction entirely. If the buffer has
2408 * already started to be used by a subsequent transaction, refile the
2409 * buffer on that transaction's metadata list.
2411 * Called under j_list_lock
2412 * Called under jbd_lock_bh_state(jh2bh(jh))
2414 * jh and bh may be already free when this function returns
2416 void __jbd2_journal_refile_buffer(struct journal_head *jh)
2418 int was_dirty, jlist;
2419 struct buffer_head *bh = jh2bh(jh);
2421 J_ASSERT_JH(jh, jbd_is_locked_bh_state(bh));
2422 if (jh->b_transaction)
2423 assert_spin_locked(&jh->b_transaction->t_journal->j_list_lock);
2425 /* If the buffer is now unused, just drop it. */
2426 if (jh->b_next_transaction == NULL) {
2427 __jbd2_journal_unfile_buffer(jh);
2432 * It has been modified by a later transaction: add it to the new
2433 * transaction's metadata list.
2436 was_dirty = test_clear_buffer_jbddirty(bh);
2437 __jbd2_journal_temp_unlink_buffer(jh);
2439 * We set b_transaction here because b_next_transaction will inherit
2440 * our jh reference and thus __jbd2_journal_file_buffer() must not
2443 jh->b_transaction = jh->b_next_transaction;
2444 jh->b_next_transaction = NULL;
2445 if (buffer_freed(bh))
2447 else if (jh->b_modified)
2448 jlist = BJ_Metadata;
2450 jlist = BJ_Reserved;
2451 __jbd2_journal_file_buffer(jh, jh->b_transaction, jlist);
2452 J_ASSERT_JH(jh, jh->b_transaction->t_state == T_RUNNING);
2455 set_buffer_jbddirty(bh);
2459 * __jbd2_journal_refile_buffer() with necessary locking added. We take our
2460 * bh reference so that we can safely unlock bh.
2462 * The jh and bh may be freed by this call.
2464 void jbd2_journal_refile_buffer(journal_t *journal, struct journal_head *jh)
2466 struct buffer_head *bh = jh2bh(jh);
2468 /* Get reference so that buffer cannot be freed before we unlock it */
2470 jbd_lock_bh_state(bh);
2471 spin_lock(&journal->j_list_lock);
2472 __jbd2_journal_refile_buffer(jh);
2473 jbd_unlock_bh_state(bh);
2474 spin_unlock(&journal->j_list_lock);
2479 * File inode in the inode list of the handle's transaction
2481 static int jbd2_journal_file_inode(handle_t *handle, struct jbd2_inode *jinode,
2482 unsigned long flags)
2484 transaction_t *transaction = handle->h_transaction;
2487 if (is_handle_aborted(handle))
2489 journal = transaction->t_journal;
2491 jbd_debug(4, "Adding inode %lu, tid:%d\n", jinode->i_vfs_inode->i_ino,
2492 transaction->t_tid);
2495 * First check whether inode isn't already on the transaction's
2496 * lists without taking the lock. Note that this check is safe
2497 * without the lock as we cannot race with somebody removing inode
2498 * from the transaction. The reason is that we remove inode from the
2499 * transaction only in journal_release_jbd_inode() and when we commit
2500 * the transaction. We are guarded from the first case by holding
2501 * a reference to the inode. We are safe against the second case
2502 * because if jinode->i_transaction == transaction, commit code
2503 * cannot touch the transaction because we hold reference to it,
2504 * and if jinode->i_next_transaction == transaction, commit code
2505 * will only file the inode where we want it.
2507 if ((jinode->i_transaction == transaction ||
2508 jinode->i_next_transaction == transaction) &&
2509 (jinode->i_flags & flags) == flags)
2512 spin_lock(&journal->j_list_lock);
2513 jinode->i_flags |= flags;
2514 /* Is inode already attached where we need it? */
2515 if (jinode->i_transaction == transaction ||
2516 jinode->i_next_transaction == transaction)
2520 * We only ever set this variable to 1 so the test is safe. Since
2521 * t_need_data_flush is likely to be set, we do the test to save some
2522 * cacheline bouncing
2524 if (!transaction->t_need_data_flush)
2525 transaction->t_need_data_flush = 1;
2526 /* On some different transaction's list - should be
2527 * the committing one */
2528 if (jinode->i_transaction) {
2529 J_ASSERT(jinode->i_next_transaction == NULL);
2530 J_ASSERT(jinode->i_transaction ==
2531 journal->j_committing_transaction);
2532 jinode->i_next_transaction = transaction;
2535 /* Not on any transaction list... */
2536 J_ASSERT(!jinode->i_next_transaction);
2537 jinode->i_transaction = transaction;
2538 list_add(&jinode->i_list, &transaction->t_inode_list);
2540 spin_unlock(&journal->j_list_lock);
2545 int jbd2_journal_inode_add_write(handle_t *handle, struct jbd2_inode *jinode)
2547 return jbd2_journal_file_inode(handle, jinode,
2548 JI_WRITE_DATA | JI_WAIT_DATA);
2551 int jbd2_journal_inode_add_wait(handle_t *handle, struct jbd2_inode *jinode)
2553 return jbd2_journal_file_inode(handle, jinode, JI_WAIT_DATA);
2557 * File truncate and transaction commit interact with each other in a
2558 * non-trivial way. If a transaction writing data block A is
2559 * committing, we cannot discard the data by truncate until we have
2560 * written them. Otherwise if we crashed after the transaction with
2561 * write has committed but before the transaction with truncate has
2562 * committed, we could see stale data in block A. This function is a
2563 * helper to solve this problem. It starts writeout of the truncated
2564 * part in case it is in the committing transaction.
2566 * Filesystem code must call this function when inode is journaled in
2567 * ordered mode before truncation happens and after the inode has been
2568 * placed on orphan list with the new inode size. The second condition
2569 * avoids the race that someone writes new data and we start
2570 * committing the transaction after this function has been called but
2571 * before a transaction for truncate is started (and furthermore it
2572 * allows us to optimize the case where the addition to orphan list
2573 * happens in the same transaction as write --- we don't have to write
2574 * any data in such case).
2576 int jbd2_journal_begin_ordered_truncate(journal_t *journal,
2577 struct jbd2_inode *jinode,
2580 transaction_t *inode_trans, *commit_trans;
2583 /* This is a quick check to avoid locking if not necessary */
2584 if (!jinode->i_transaction)
2586 /* Locks are here just to force reading of recent values, it is
2587 * enough that the transaction was not committing before we started
2588 * a transaction adding the inode to orphan list */
2589 read_lock(&journal->j_state_lock);
2590 commit_trans = journal->j_committing_transaction;
2591 read_unlock(&journal->j_state_lock);
2592 spin_lock(&journal->j_list_lock);
2593 inode_trans = jinode->i_transaction;
2594 spin_unlock(&journal->j_list_lock);
2595 if (inode_trans == commit_trans) {
2596 ret = filemap_fdatawrite_range(jinode->i_vfs_inode->i_mapping,
2597 new_size, LLONG_MAX);
2599 jbd2_journal_abort(journal, ret);