1 // SPDX-License-Identifier: GPL-2.0+
3 * linux/fs/jbd2/transaction.c
5 * Written by Stephen C. Tweedie <sct@redhat.com>, 1998
7 * Copyright 1998 Red Hat corp --- All Rights Reserved
9 * Generic filesystem transaction handling code; part of the ext2fs
12 * This file manages transactions (compound commits managed by the
13 * journaling code) and handles (individual atomic operations by the
17 #include <linux/time.h>
19 #include <linux/jbd2.h>
20 #include <linux/errno.h>
21 #include <linux/slab.h>
22 #include <linux/timer.h>
24 #include <linux/highmem.h>
25 #include <linux/hrtimer.h>
26 #include <linux/backing-dev.h>
27 #include <linux/bug.h>
28 #include <linux/module.h>
29 #include <linux/sched/mm.h>
31 #include <trace/events/jbd2.h>
33 static void __jbd2_journal_temp_unlink_buffer(struct journal_head *jh);
34 static void __jbd2_journal_unfile_buffer(struct journal_head *jh);
36 static struct kmem_cache *transaction_cache;
37 int __init jbd2_journal_init_transaction_cache(void)
39 J_ASSERT(!transaction_cache);
40 transaction_cache = kmem_cache_create("jbd2_transaction_s",
41 sizeof(transaction_t),
43 SLAB_HWCACHE_ALIGN|SLAB_TEMPORARY,
45 if (!transaction_cache) {
46 pr_emerg("JBD2: failed to create transaction cache\n");
52 void jbd2_journal_destroy_transaction_cache(void)
54 kmem_cache_destroy(transaction_cache);
55 transaction_cache = NULL;
58 void jbd2_journal_free_transaction(transaction_t *transaction)
60 if (unlikely(ZERO_OR_NULL_PTR(transaction)))
62 kmem_cache_free(transaction_cache, transaction);
66 * Base amount of descriptor blocks we reserve for each transaction.
68 static int jbd2_descriptor_blocks_per_trans(journal_t *journal)
70 int tag_space = journal->j_blocksize - sizeof(journal_header_t);
75 if (jbd2_journal_has_csum_v2or3(journal))
76 tag_space -= sizeof(struct jbd2_journal_block_tail);
77 /* Commit code leaves a slack space of 16 bytes at the end of block */
78 tags_per_block = (tag_space - 16) / journal_tag_bytes(journal);
80 * Revoke descriptors are accounted separately so we need to reserve
81 * space for commit block and normal transaction descriptor blocks.
83 return 1 + DIV_ROUND_UP(journal->j_max_transaction_buffers,
88 * jbd2_get_transaction: obtain a new transaction_t object.
90 * Simply initialise a new transaction. Initialize it in
91 * RUNNING state and add it to the current journal (which should not
92 * have an existing running transaction: we only make a new transaction
93 * once we have started to commit the old one).
96 * The journal MUST be locked. We don't perform atomic mallocs on the
97 * new transaction and we can't block without protecting against other
98 * processes trying to touch the journal while it is in transition.
102 static void jbd2_get_transaction(journal_t *journal,
103 transaction_t *transaction)
105 transaction->t_journal = journal;
106 transaction->t_state = T_RUNNING;
107 transaction->t_start_time = ktime_get();
108 transaction->t_tid = journal->j_transaction_sequence++;
109 transaction->t_expires = jiffies + journal->j_commit_interval;
110 spin_lock_init(&transaction->t_handle_lock);
111 atomic_set(&transaction->t_updates, 0);
112 atomic_set(&transaction->t_outstanding_credits,
113 jbd2_descriptor_blocks_per_trans(journal) +
114 atomic_read(&journal->j_reserved_credits));
115 atomic_set(&transaction->t_outstanding_revokes, 0);
116 atomic_set(&transaction->t_handle_count, 0);
117 INIT_LIST_HEAD(&transaction->t_inode_list);
118 INIT_LIST_HEAD(&transaction->t_private_list);
120 /* Set up the commit timer for the new transaction. */
121 journal->j_commit_timer.expires = round_jiffies_up(transaction->t_expires);
122 add_timer(&journal->j_commit_timer);
124 J_ASSERT(journal->j_running_transaction == NULL);
125 journal->j_running_transaction = transaction;
126 transaction->t_max_wait = 0;
127 transaction->t_start = jiffies;
128 transaction->t_requested = 0;
134 * A handle_t is an object which represents a single atomic update to a
135 * filesystem, and which tracks all of the modifications which form part
136 * of that one update.
140 * Update transaction's maximum wait time, if debugging is enabled.
142 * In order for t_max_wait to be reliable, it must be protected by a
143 * lock. But doing so will mean that start_this_handle() can not be
144 * run in parallel on SMP systems, which limits our scalability. So
145 * unless debugging is enabled, we no longer update t_max_wait, which
146 * means that maximum wait time reported by the jbd2_run_stats
147 * tracepoint will always be zero.
149 static inline void update_t_max_wait(transaction_t *transaction,
152 #ifdef CONFIG_JBD2_DEBUG
153 if (jbd2_journal_enable_debug &&
154 time_after(transaction->t_start, ts)) {
155 ts = jbd2_time_diff(ts, transaction->t_start);
156 spin_lock(&transaction->t_handle_lock);
157 if (ts > transaction->t_max_wait)
158 transaction->t_max_wait = ts;
159 spin_unlock(&transaction->t_handle_lock);
165 * Wait until running transaction passes to T_FLUSH state and new transaction
166 * can thus be started. Also starts the commit if needed. The function expects
167 * running transaction to exist and releases j_state_lock.
169 static void wait_transaction_locked(journal_t *journal)
170 __releases(journal->j_state_lock)
174 tid_t tid = journal->j_running_transaction->t_tid;
176 prepare_to_wait_exclusive(&journal->j_wait_transaction_locked, &wait,
177 TASK_UNINTERRUPTIBLE);
178 need_to_start = !tid_geq(journal->j_commit_request, tid);
179 read_unlock(&journal->j_state_lock);
181 jbd2_log_start_commit(journal, tid);
182 jbd2_might_wait_for_commit(journal);
184 finish_wait(&journal->j_wait_transaction_locked, &wait);
188 * Wait until running transaction transitions from T_SWITCH to T_FLUSH
189 * state and new transaction can thus be started. The function releases
192 static void wait_transaction_switching(journal_t *journal)
193 __releases(journal->j_state_lock)
197 if (WARN_ON(!journal->j_running_transaction ||
198 journal->j_running_transaction->t_state != T_SWITCH)) {
199 read_unlock(&journal->j_state_lock);
202 prepare_to_wait_exclusive(&journal->j_wait_transaction_locked, &wait,
203 TASK_UNINTERRUPTIBLE);
204 read_unlock(&journal->j_state_lock);
206 * We don't call jbd2_might_wait_for_commit() here as there's no
207 * waiting for outstanding handles happening anymore in T_SWITCH state
208 * and handling of reserved handles actually relies on that for
212 finish_wait(&journal->j_wait_transaction_locked, &wait);
215 static void sub_reserved_credits(journal_t *journal, int blocks)
217 atomic_sub(blocks, &journal->j_reserved_credits);
218 wake_up(&journal->j_wait_reserved);
222 * Wait until we can add credits for handle to the running transaction. Called
223 * with j_state_lock held for reading. Returns 0 if handle joined the running
224 * transaction. Returns 1 if we had to wait, j_state_lock is dropped, and
227 static int add_transaction_credits(journal_t *journal, int blocks,
230 transaction_t *t = journal->j_running_transaction;
232 int total = blocks + rsv_blocks;
235 * If the current transaction is locked down for commit, wait
236 * for the lock to be released.
238 if (t->t_state != T_RUNNING) {
239 WARN_ON_ONCE(t->t_state >= T_FLUSH);
240 wait_transaction_locked(journal);
245 * If there is not enough space left in the log to write all
246 * potential buffers requested by this operation, we need to
247 * stall pending a log checkpoint to free some more log space.
249 needed = atomic_add_return(total, &t->t_outstanding_credits);
250 if (needed > journal->j_max_transaction_buffers) {
252 * If the current transaction is already too large,
253 * then start to commit it: we can then go back and
254 * attach this handle to a new transaction.
256 atomic_sub(total, &t->t_outstanding_credits);
259 * Is the number of reserved credits in the current transaction too
260 * big to fit this handle? Wait until reserved credits are freed.
262 if (atomic_read(&journal->j_reserved_credits) + total >
263 journal->j_max_transaction_buffers) {
264 read_unlock(&journal->j_state_lock);
265 jbd2_might_wait_for_commit(journal);
266 wait_event(journal->j_wait_reserved,
267 atomic_read(&journal->j_reserved_credits) + total <=
268 journal->j_max_transaction_buffers);
272 wait_transaction_locked(journal);
277 * The commit code assumes that it can get enough log space
278 * without forcing a checkpoint. This is *critical* for
279 * correctness: a checkpoint of a buffer which is also
280 * associated with a committing transaction creates a deadlock,
281 * so commit simply cannot force through checkpoints.
283 * We must therefore ensure the necessary space in the journal
284 * *before* starting to dirty potentially checkpointed buffers
285 * in the new transaction.
287 if (jbd2_log_space_left(journal) < journal->j_max_transaction_buffers) {
288 atomic_sub(total, &t->t_outstanding_credits);
289 read_unlock(&journal->j_state_lock);
290 jbd2_might_wait_for_commit(journal);
291 write_lock(&journal->j_state_lock);
292 if (jbd2_log_space_left(journal) <
293 journal->j_max_transaction_buffers)
294 __jbd2_log_wait_for_space(journal);
295 write_unlock(&journal->j_state_lock);
299 /* No reservation? We are done... */
303 needed = atomic_add_return(rsv_blocks, &journal->j_reserved_credits);
304 /* We allow at most half of a transaction to be reserved */
305 if (needed > journal->j_max_transaction_buffers / 2) {
306 sub_reserved_credits(journal, rsv_blocks);
307 atomic_sub(total, &t->t_outstanding_credits);
308 read_unlock(&journal->j_state_lock);
309 jbd2_might_wait_for_commit(journal);
310 wait_event(journal->j_wait_reserved,
311 atomic_read(&journal->j_reserved_credits) + rsv_blocks
312 <= journal->j_max_transaction_buffers / 2);
319 * start_this_handle: Given a handle, deal with any locking or stalling
320 * needed to make sure that there is enough journal space for the handle
321 * to begin. Attach the handle to a transaction and set up the
322 * transaction's buffer credits.
325 static int start_this_handle(journal_t *journal, handle_t *handle,
328 transaction_t *transaction, *new_transaction = NULL;
329 int blocks = handle->h_total_credits;
331 unsigned long ts = jiffies;
333 if (handle->h_rsv_handle)
334 rsv_blocks = handle->h_rsv_handle->h_total_credits;
337 * Limit the number of reserved credits to 1/2 of maximum transaction
338 * size and limit the number of total credits to not exceed maximum
339 * transaction size per operation.
341 if ((rsv_blocks > journal->j_max_transaction_buffers / 2) ||
342 (rsv_blocks + blocks > journal->j_max_transaction_buffers)) {
343 printk(KERN_ERR "JBD2: %s wants too many credits "
344 "credits:%d rsv_credits:%d max:%d\n",
345 current->comm, blocks, rsv_blocks,
346 journal->j_max_transaction_buffers);
353 * This check is racy but it is just an optimization of allocating new
354 * transaction early if there are high chances we'll need it. If we
355 * guess wrong, we'll retry or free unused transaction.
357 if (!data_race(journal->j_running_transaction)) {
359 * If __GFP_FS is not present, then we may be being called from
360 * inside the fs writeback layer, so we MUST NOT fail.
362 if ((gfp_mask & __GFP_FS) == 0)
363 gfp_mask |= __GFP_NOFAIL;
364 new_transaction = kmem_cache_zalloc(transaction_cache,
366 if (!new_transaction)
370 jbd_debug(3, "New handle %p going live.\n", handle);
373 * We need to hold j_state_lock until t_updates has been incremented,
374 * for proper journal barrier handling
377 read_lock(&journal->j_state_lock);
378 BUG_ON(journal->j_flags & JBD2_UNMOUNT);
379 if (is_journal_aborted(journal) ||
380 (journal->j_errno != 0 && !(journal->j_flags & JBD2_ACK_ERR))) {
381 read_unlock(&journal->j_state_lock);
382 jbd2_journal_free_transaction(new_transaction);
387 * Wait on the journal's transaction barrier if necessary. Specifically
388 * we allow reserved handles to proceed because otherwise commit could
389 * deadlock on page writeback not being able to complete.
391 if (!handle->h_reserved && journal->j_barrier_count) {
392 read_unlock(&journal->j_state_lock);
393 wait_event(journal->j_wait_transaction_locked,
394 journal->j_barrier_count == 0);
398 if (!journal->j_running_transaction) {
399 read_unlock(&journal->j_state_lock);
400 if (!new_transaction)
401 goto alloc_transaction;
402 write_lock(&journal->j_state_lock);
403 if (!journal->j_running_transaction &&
404 (handle->h_reserved || !journal->j_barrier_count)) {
405 jbd2_get_transaction(journal, new_transaction);
406 new_transaction = NULL;
408 write_unlock(&journal->j_state_lock);
412 transaction = journal->j_running_transaction;
414 if (!handle->h_reserved) {
415 /* We may have dropped j_state_lock - restart in that case */
416 if (add_transaction_credits(journal, blocks, rsv_blocks))
420 * We have handle reserved so we are allowed to join T_LOCKED
421 * transaction and we don't have to check for transaction size
422 * and journal space. But we still have to wait while running
423 * transaction is being switched to a committing one as it
424 * won't wait for any handles anymore.
426 if (transaction->t_state == T_SWITCH) {
427 wait_transaction_switching(journal);
430 sub_reserved_credits(journal, blocks);
431 handle->h_reserved = 0;
434 /* OK, account for the buffers that this operation expects to
435 * use and add the handle to the running transaction.
437 update_t_max_wait(transaction, ts);
438 handle->h_transaction = transaction;
439 handle->h_requested_credits = blocks;
440 handle->h_revoke_credits_requested = handle->h_revoke_credits;
441 handle->h_start_jiffies = jiffies;
442 atomic_inc(&transaction->t_updates);
443 atomic_inc(&transaction->t_handle_count);
444 jbd_debug(4, "Handle %p given %d credits (total %d, free %lu)\n",
446 atomic_read(&transaction->t_outstanding_credits),
447 jbd2_log_space_left(journal));
448 read_unlock(&journal->j_state_lock);
449 current->journal_info = handle;
451 rwsem_acquire_read(&journal->j_trans_commit_map, 0, 0, _THIS_IP_);
452 jbd2_journal_free_transaction(new_transaction);
454 * Ensure that no allocations done while the transaction is open are
455 * going to recurse back to the fs layer.
457 handle->saved_alloc_context = memalloc_nofs_save();
461 /* Allocate a new handle. This should probably be in a slab... */
462 static handle_t *new_handle(int nblocks)
464 handle_t *handle = jbd2_alloc_handle(GFP_NOFS);
467 handle->h_total_credits = nblocks;
473 handle_t *jbd2__journal_start(journal_t *journal, int nblocks, int rsv_blocks,
474 int revoke_records, gfp_t gfp_mask,
475 unsigned int type, unsigned int line_no)
477 handle_t *handle = journal_current_handle();
481 return ERR_PTR(-EROFS);
484 J_ASSERT(handle->h_transaction->t_journal == journal);
489 nblocks += DIV_ROUND_UP(revoke_records,
490 journal->j_revoke_records_per_block);
491 handle = new_handle(nblocks);
493 return ERR_PTR(-ENOMEM);
495 handle_t *rsv_handle;
497 rsv_handle = new_handle(rsv_blocks);
499 jbd2_free_handle(handle);
500 return ERR_PTR(-ENOMEM);
502 rsv_handle->h_reserved = 1;
503 rsv_handle->h_journal = journal;
504 handle->h_rsv_handle = rsv_handle;
506 handle->h_revoke_credits = revoke_records;
508 err = start_this_handle(journal, handle, gfp_mask);
510 if (handle->h_rsv_handle)
511 jbd2_free_handle(handle->h_rsv_handle);
512 jbd2_free_handle(handle);
515 handle->h_type = type;
516 handle->h_line_no = line_no;
517 trace_jbd2_handle_start(journal->j_fs_dev->bd_dev,
518 handle->h_transaction->t_tid, type,
523 EXPORT_SYMBOL(jbd2__journal_start);
527 * jbd2_journal_start() - Obtain a new handle.
528 * @journal: Journal to start transaction on.
529 * @nblocks: number of block buffer we might modify
531 * We make sure that the transaction can guarantee at least nblocks of
532 * modified buffers in the log. We block until the log can guarantee
533 * that much space. Additionally, if rsv_blocks > 0, we also create another
534 * handle with rsv_blocks reserved blocks in the journal. This handle is
535 * stored in h_rsv_handle. It is not attached to any particular transaction
536 * and thus doesn't block transaction commit. If the caller uses this reserved
537 * handle, it has to set h_rsv_handle to NULL as otherwise jbd2_journal_stop()
538 * on the parent handle will dispose the reserved one. Reserved handle has to
539 * be converted to a normal handle using jbd2_journal_start_reserved() before
542 * Return a pointer to a newly allocated handle, or an ERR_PTR() value
545 handle_t *jbd2_journal_start(journal_t *journal, int nblocks)
547 return jbd2__journal_start(journal, nblocks, 0, 0, GFP_NOFS, 0, 0);
549 EXPORT_SYMBOL(jbd2_journal_start);
551 static void __jbd2_journal_unreserve_handle(handle_t *handle, transaction_t *t)
553 journal_t *journal = handle->h_journal;
555 WARN_ON(!handle->h_reserved);
556 sub_reserved_credits(journal, handle->h_total_credits);
558 atomic_sub(handle->h_total_credits, &t->t_outstanding_credits);
561 void jbd2_journal_free_reserved(handle_t *handle)
563 journal_t *journal = handle->h_journal;
565 /* Get j_state_lock to pin running transaction if it exists */
566 read_lock(&journal->j_state_lock);
567 __jbd2_journal_unreserve_handle(handle, journal->j_running_transaction);
568 read_unlock(&journal->j_state_lock);
569 jbd2_free_handle(handle);
571 EXPORT_SYMBOL(jbd2_journal_free_reserved);
574 * jbd2_journal_start_reserved() - start reserved handle
575 * @handle: handle to start
576 * @type: for handle statistics
577 * @line_no: for handle statistics
579 * Start handle that has been previously reserved with jbd2_journal_reserve().
580 * This attaches @handle to the running transaction (or creates one if there's
581 * not transaction running). Unlike jbd2_journal_start() this function cannot
582 * block on journal commit, checkpointing, or similar stuff. It can block on
583 * memory allocation or frozen journal though.
585 * Return 0 on success, non-zero on error - handle is freed in that case.
587 int jbd2_journal_start_reserved(handle_t *handle, unsigned int type,
588 unsigned int line_no)
590 journal_t *journal = handle->h_journal;
593 if (WARN_ON(!handle->h_reserved)) {
594 /* Someone passed in normal handle? Just stop it. */
595 jbd2_journal_stop(handle);
599 * Usefulness of mixing of reserved and unreserved handles is
600 * questionable. So far nobody seems to need it so just error out.
602 if (WARN_ON(current->journal_info)) {
603 jbd2_journal_free_reserved(handle);
607 handle->h_journal = NULL;
609 * GFP_NOFS is here because callers are likely from writeback or
610 * similarly constrained call sites
612 ret = start_this_handle(journal, handle, GFP_NOFS);
614 handle->h_journal = journal;
615 jbd2_journal_free_reserved(handle);
618 handle->h_type = type;
619 handle->h_line_no = line_no;
620 trace_jbd2_handle_start(journal->j_fs_dev->bd_dev,
621 handle->h_transaction->t_tid, type,
622 line_no, handle->h_total_credits);
625 EXPORT_SYMBOL(jbd2_journal_start_reserved);
628 * jbd2_journal_extend() - extend buffer credits.
629 * @handle: handle to 'extend'
630 * @nblocks: nr blocks to try to extend by.
631 * @revoke_records: number of revoke records to try to extend by.
633 * Some transactions, such as large extends and truncates, can be done
634 * atomically all at once or in several stages. The operation requests
635 * a credit for a number of buffer modifications in advance, but can
636 * extend its credit if it needs more.
638 * jbd2_journal_extend tries to give the running handle more buffer credits.
639 * It does not guarantee that allocation - this is a best-effort only.
640 * The calling process MUST be able to deal cleanly with a failure to
643 * Return 0 on success, non-zero on failure.
645 * return code < 0 implies an error
646 * return code > 0 implies normal transaction-full status.
648 int jbd2_journal_extend(handle_t *handle, int nblocks, int revoke_records)
650 transaction_t *transaction = handle->h_transaction;
655 if (is_handle_aborted(handle))
657 journal = transaction->t_journal;
661 read_lock(&journal->j_state_lock);
663 /* Don't extend a locked-down transaction! */
664 if (transaction->t_state != T_RUNNING) {
665 jbd_debug(3, "denied handle %p %d blocks: "
666 "transaction not running\n", handle, nblocks);
670 nblocks += DIV_ROUND_UP(
671 handle->h_revoke_credits_requested + revoke_records,
672 journal->j_revoke_records_per_block) -
674 handle->h_revoke_credits_requested,
675 journal->j_revoke_records_per_block);
676 spin_lock(&transaction->t_handle_lock);
677 wanted = atomic_add_return(nblocks,
678 &transaction->t_outstanding_credits);
680 if (wanted > journal->j_max_transaction_buffers) {
681 jbd_debug(3, "denied handle %p %d blocks: "
682 "transaction too large\n", handle, nblocks);
683 atomic_sub(nblocks, &transaction->t_outstanding_credits);
687 trace_jbd2_handle_extend(journal->j_fs_dev->bd_dev,
689 handle->h_type, handle->h_line_no,
690 handle->h_total_credits,
693 handle->h_total_credits += nblocks;
694 handle->h_requested_credits += nblocks;
695 handle->h_revoke_credits += revoke_records;
696 handle->h_revoke_credits_requested += revoke_records;
699 jbd_debug(3, "extended handle %p by %d\n", handle, nblocks);
701 spin_unlock(&transaction->t_handle_lock);
703 read_unlock(&journal->j_state_lock);
707 static void stop_this_handle(handle_t *handle)
709 transaction_t *transaction = handle->h_transaction;
710 journal_t *journal = transaction->t_journal;
713 J_ASSERT(journal_current_handle() == handle);
714 J_ASSERT(atomic_read(&transaction->t_updates) > 0);
715 current->journal_info = NULL;
717 * Subtract necessary revoke descriptor blocks from handle credits. We
718 * take care to account only for revoke descriptor blocks the
719 * transaction will really need as large sequences of transactions with
720 * small numbers of revokes are relatively common.
722 revokes = handle->h_revoke_credits_requested - handle->h_revoke_credits;
724 int t_revokes, revoke_descriptors;
725 int rr_per_blk = journal->j_revoke_records_per_block;
727 WARN_ON_ONCE(DIV_ROUND_UP(revokes, rr_per_blk)
728 > handle->h_total_credits);
729 t_revokes = atomic_add_return(revokes,
730 &transaction->t_outstanding_revokes);
732 DIV_ROUND_UP(t_revokes, rr_per_blk) -
733 DIV_ROUND_UP(t_revokes - revokes, rr_per_blk);
734 handle->h_total_credits -= revoke_descriptors;
736 atomic_sub(handle->h_total_credits,
737 &transaction->t_outstanding_credits);
738 if (handle->h_rsv_handle)
739 __jbd2_journal_unreserve_handle(handle->h_rsv_handle,
741 if (atomic_dec_and_test(&transaction->t_updates))
742 wake_up(&journal->j_wait_updates);
744 rwsem_release(&journal->j_trans_commit_map, _THIS_IP_);
746 * Scope of the GFP_NOFS context is over here and so we can restore the
747 * original alloc context.
749 memalloc_nofs_restore(handle->saved_alloc_context);
753 * jbd2__journal_restart() - restart a handle .
754 * @handle: handle to restart
755 * @nblocks: nr credits requested
756 * @revoke_records: number of revoke record credits requested
757 * @gfp_mask: memory allocation flags (for start_this_handle)
759 * Restart a handle for a multi-transaction filesystem
762 * If the jbd2_journal_extend() call above fails to grant new buffer credits
763 * to a running handle, a call to jbd2_journal_restart will commit the
764 * handle's transaction so far and reattach the handle to a new
765 * transaction capable of guaranteeing the requested number of
766 * credits. We preserve reserved handle if there's any attached to the
769 int jbd2__journal_restart(handle_t *handle, int nblocks, int revoke_records,
772 transaction_t *transaction = handle->h_transaction;
778 /* If we've had an abort of any type, don't even think about
779 * actually doing the restart! */
780 if (is_handle_aborted(handle))
782 journal = transaction->t_journal;
783 tid = transaction->t_tid;
786 * First unlink the handle from its current transaction, and start the
789 jbd_debug(2, "restarting handle %p\n", handle);
790 stop_this_handle(handle);
791 handle->h_transaction = NULL;
794 * TODO: If we use READ_ONCE / WRITE_ONCE for j_commit_request we can
795 * get rid of pointless j_state_lock traffic like this.
797 read_lock(&journal->j_state_lock);
798 need_to_start = !tid_geq(journal->j_commit_request, tid);
799 read_unlock(&journal->j_state_lock);
801 jbd2_log_start_commit(journal, tid);
802 handle->h_total_credits = nblocks +
803 DIV_ROUND_UP(revoke_records,
804 journal->j_revoke_records_per_block);
805 handle->h_revoke_credits = revoke_records;
806 ret = start_this_handle(journal, handle, gfp_mask);
807 trace_jbd2_handle_restart(journal->j_fs_dev->bd_dev,
808 ret ? 0 : handle->h_transaction->t_tid,
809 handle->h_type, handle->h_line_no,
810 handle->h_total_credits);
813 EXPORT_SYMBOL(jbd2__journal_restart);
816 int jbd2_journal_restart(handle_t *handle, int nblocks)
818 return jbd2__journal_restart(handle, nblocks, 0, GFP_NOFS);
820 EXPORT_SYMBOL(jbd2_journal_restart);
823 * jbd2_journal_lock_updates () - establish a transaction barrier.
824 * @journal: Journal to establish a barrier on.
826 * This locks out any further updates from being started, and blocks
827 * until all existing updates have completed, returning only once the
828 * journal is in a quiescent state with no updates running.
830 * The journal lock should not be held on entry.
832 void jbd2_journal_lock_updates(journal_t *journal)
836 jbd2_might_wait_for_commit(journal);
838 write_lock(&journal->j_state_lock);
839 ++journal->j_barrier_count;
841 /* Wait until there are no reserved handles */
842 if (atomic_read(&journal->j_reserved_credits)) {
843 write_unlock(&journal->j_state_lock);
844 wait_event(journal->j_wait_reserved,
845 atomic_read(&journal->j_reserved_credits) == 0);
846 write_lock(&journal->j_state_lock);
849 /* Wait until there are no running updates */
851 transaction_t *transaction = journal->j_running_transaction;
856 spin_lock(&transaction->t_handle_lock);
857 prepare_to_wait(&journal->j_wait_updates, &wait,
858 TASK_UNINTERRUPTIBLE);
859 if (!atomic_read(&transaction->t_updates)) {
860 spin_unlock(&transaction->t_handle_lock);
861 finish_wait(&journal->j_wait_updates, &wait);
864 spin_unlock(&transaction->t_handle_lock);
865 write_unlock(&journal->j_state_lock);
867 finish_wait(&journal->j_wait_updates, &wait);
868 write_lock(&journal->j_state_lock);
870 write_unlock(&journal->j_state_lock);
873 * We have now established a barrier against other normal updates, but
874 * we also need to barrier against other jbd2_journal_lock_updates() calls
875 * to make sure that we serialise special journal-locked operations
878 mutex_lock(&journal->j_barrier);
882 * jbd2_journal_unlock_updates () - release barrier
883 * @journal: Journal to release the barrier on.
885 * Release a transaction barrier obtained with jbd2_journal_lock_updates().
887 * Should be called without the journal lock held.
889 void jbd2_journal_unlock_updates (journal_t *journal)
891 J_ASSERT(journal->j_barrier_count != 0);
893 mutex_unlock(&journal->j_barrier);
894 write_lock(&journal->j_state_lock);
895 --journal->j_barrier_count;
896 write_unlock(&journal->j_state_lock);
897 wake_up_all(&journal->j_wait_transaction_locked);
900 static void warn_dirty_buffer(struct buffer_head *bh)
903 "JBD2: Spotted dirty metadata buffer (dev = %pg, blocknr = %llu). "
904 "There's a risk of filesystem corruption in case of system "
906 bh->b_bdev, (unsigned long long)bh->b_blocknr);
909 /* Call t_frozen trigger and copy buffer data into jh->b_frozen_data. */
910 static void jbd2_freeze_jh_data(struct journal_head *jh)
915 struct buffer_head *bh = jh2bh(jh);
917 J_EXPECT_JH(jh, buffer_uptodate(bh), "Possible IO failure.\n");
919 offset = offset_in_page(bh->b_data);
920 source = kmap_atomic(page);
921 /* Fire data frozen trigger just before we copy the data */
922 jbd2_buffer_frozen_trigger(jh, source + offset, jh->b_triggers);
923 memcpy(jh->b_frozen_data, source + offset, bh->b_size);
924 kunmap_atomic(source);
927 * Now that the frozen data is saved off, we need to store any matching
930 jh->b_frozen_triggers = jh->b_triggers;
934 * If the buffer is already part of the current transaction, then there
935 * is nothing we need to do. If it is already part of a prior
936 * transaction which we are still committing to disk, then we need to
937 * make sure that we do not overwrite the old copy: we do copy-out to
938 * preserve the copy going to disk. We also account the buffer against
939 * the handle's metadata buffer credits (unless the buffer is already
940 * part of the transaction, that is).
944 do_get_write_access(handle_t *handle, struct journal_head *jh,
947 struct buffer_head *bh;
948 transaction_t *transaction = handle->h_transaction;
951 char *frozen_buffer = NULL;
952 unsigned long start_lock, time_lock;
954 journal = transaction->t_journal;
956 jbd_debug(5, "journal_head %p, force_copy %d\n", jh, force_copy);
958 JBUFFER_TRACE(jh, "entry");
962 /* @@@ Need to check for errors here at some point. */
964 start_lock = jiffies;
966 spin_lock(&jh->b_state_lock);
968 /* If it takes too long to lock the buffer, trace it */
969 time_lock = jbd2_time_diff(start_lock, jiffies);
970 if (time_lock > HZ/10)
971 trace_jbd2_lock_buffer_stall(bh->b_bdev->bd_dev,
972 jiffies_to_msecs(time_lock));
974 /* We now hold the buffer lock so it is safe to query the buffer
975 * state. Is the buffer dirty?
977 * If so, there are two possibilities. The buffer may be
978 * non-journaled, and undergoing a quite legitimate writeback.
979 * Otherwise, it is journaled, and we don't expect dirty buffers
980 * in that state (the buffers should be marked JBD_Dirty
981 * instead.) So either the IO is being done under our own
982 * control and this is a bug, or it's a third party IO such as
983 * dump(8) (which may leave the buffer scheduled for read ---
984 * ie. locked but not dirty) or tune2fs (which may actually have
985 * the buffer dirtied, ugh.) */
987 if (buffer_dirty(bh) && jh->b_transaction) {
988 warn_dirty_buffer(bh);
990 * We need to clean the dirty flag and we must do it under the
991 * buffer lock to be sure we don't race with running write-out.
993 JBUFFER_TRACE(jh, "Journalling dirty buffer");
994 clear_buffer_dirty(bh);
996 * The buffer is going to be added to BJ_Reserved list now and
997 * nothing guarantees jbd2_journal_dirty_metadata() will be
998 * ever called for it. So we need to set jbddirty bit here to
999 * make sure the buffer is dirtied and written out when the
1000 * journaling machinery is done with it.
1002 set_buffer_jbddirty(bh);
1006 if (is_handle_aborted(handle)) {
1007 spin_unlock(&jh->b_state_lock);
1014 * The buffer is already part of this transaction if b_transaction or
1015 * b_next_transaction points to it
1017 if (jh->b_transaction == transaction ||
1018 jh->b_next_transaction == transaction) {
1024 * this is the first time this transaction is touching this buffer,
1025 * reset the modified flag
1030 * If the buffer is not journaled right now, we need to make sure it
1031 * doesn't get written to disk before the caller actually commits the
1034 if (!jh->b_transaction) {
1035 JBUFFER_TRACE(jh, "no transaction");
1036 J_ASSERT_JH(jh, !jh->b_next_transaction);
1037 JBUFFER_TRACE(jh, "file as BJ_Reserved");
1039 * Make sure all stores to jh (b_modified, b_frozen_data) are
1040 * visible before attaching it to the running transaction.
1041 * Paired with barrier in jbd2_write_access_granted()
1044 spin_lock(&journal->j_list_lock);
1045 if (test_clear_buffer_dirty(bh)) {
1047 * Execute buffer dirty clearing and jh->b_transaction
1048 * assignment under journal->j_list_lock locked to
1049 * prevent bh being removed from checkpoint list if
1050 * the buffer is in an intermediate state (not dirty
1051 * and jh->b_transaction is NULL).
1053 JBUFFER_TRACE(jh, "Journalling dirty buffer");
1054 set_buffer_jbddirty(bh);
1056 __jbd2_journal_file_buffer(jh, transaction, BJ_Reserved);
1057 spin_unlock(&journal->j_list_lock);
1064 * If there is already a copy-out version of this buffer, then we don't
1065 * need to make another one
1067 if (jh->b_frozen_data) {
1068 JBUFFER_TRACE(jh, "has frozen data");
1069 J_ASSERT_JH(jh, jh->b_next_transaction == NULL);
1073 JBUFFER_TRACE(jh, "owned by older transaction");
1074 J_ASSERT_JH(jh, jh->b_next_transaction == NULL);
1075 J_ASSERT_JH(jh, jh->b_transaction == journal->j_committing_transaction);
1078 * There is one case we have to be very careful about. If the
1079 * committing transaction is currently writing this buffer out to disk
1080 * and has NOT made a copy-out, then we cannot modify the buffer
1081 * contents at all right now. The essence of copy-out is that it is
1082 * the extra copy, not the primary copy, which gets journaled. If the
1083 * primary copy is already going to disk then we cannot do copy-out
1086 if (buffer_shadow(bh)) {
1087 JBUFFER_TRACE(jh, "on shadow: sleep");
1088 spin_unlock(&jh->b_state_lock);
1089 wait_on_bit_io(&bh->b_state, BH_Shadow, TASK_UNINTERRUPTIBLE);
1094 * Only do the copy if the currently-owning transaction still needs it.
1095 * If buffer isn't on BJ_Metadata list, the committing transaction is
1096 * past that stage (here we use the fact that BH_Shadow is set under
1097 * bh_state lock together with refiling to BJ_Shadow list and at this
1098 * point we know the buffer doesn't have BH_Shadow set).
1100 * Subtle point, though: if this is a get_undo_access, then we will be
1101 * relying on the frozen_data to contain the new value of the
1102 * committed_data record after the transaction, so we HAVE to force the
1103 * frozen_data copy in that case.
1105 if (jh->b_jlist == BJ_Metadata || force_copy) {
1106 JBUFFER_TRACE(jh, "generate frozen data");
1107 if (!frozen_buffer) {
1108 JBUFFER_TRACE(jh, "allocate memory for buffer");
1109 spin_unlock(&jh->b_state_lock);
1110 frozen_buffer = jbd2_alloc(jh2bh(jh)->b_size,
1111 GFP_NOFS | __GFP_NOFAIL);
1114 jh->b_frozen_data = frozen_buffer;
1115 frozen_buffer = NULL;
1116 jbd2_freeze_jh_data(jh);
1120 * Make sure all stores to jh (b_modified, b_frozen_data) are visible
1121 * before attaching it to the running transaction. Paired with barrier
1122 * in jbd2_write_access_granted()
1125 jh->b_next_transaction = transaction;
1128 spin_unlock(&jh->b_state_lock);
1131 * If we are about to journal a buffer, then any revoke pending on it is
1134 jbd2_journal_cancel_revoke(handle, jh);
1137 if (unlikely(frozen_buffer)) /* It's usually NULL */
1138 jbd2_free(frozen_buffer, bh->b_size);
1140 JBUFFER_TRACE(jh, "exit");
1144 /* Fast check whether buffer is already attached to the required transaction */
1145 static bool jbd2_write_access_granted(handle_t *handle, struct buffer_head *bh,
1148 struct journal_head *jh;
1151 /* Dirty buffers require special handling... */
1152 if (buffer_dirty(bh))
1156 * RCU protects us from dereferencing freed pages. So the checks we do
1157 * are guaranteed not to oops. However the jh slab object can get freed
1158 * & reallocated while we work with it. So we have to be careful. When
1159 * we see jh attached to the running transaction, we know it must stay
1160 * so until the transaction is committed. Thus jh won't be freed and
1161 * will be attached to the same bh while we run. However it can
1162 * happen jh gets freed, reallocated, and attached to the transaction
1163 * just after we get pointer to it from bh. So we have to be careful
1164 * and recheck jh still belongs to our bh before we return success.
1167 if (!buffer_jbd(bh))
1169 /* This should be bh2jh() but that doesn't work with inline functions */
1170 jh = READ_ONCE(bh->b_private);
1173 /* For undo access buffer must have data copied */
1174 if (undo && !jh->b_committed_data)
1176 if (READ_ONCE(jh->b_transaction) != handle->h_transaction &&
1177 READ_ONCE(jh->b_next_transaction) != handle->h_transaction)
1180 * There are two reasons for the barrier here:
1181 * 1) Make sure to fetch b_bh after we did previous checks so that we
1182 * detect when jh went through free, realloc, attach to transaction
1183 * while we were checking. Paired with implicit barrier in that path.
1184 * 2) So that access to bh done after jbd2_write_access_granted()
1185 * doesn't get reordered and see inconsistent state of concurrent
1186 * do_get_write_access().
1189 if (unlikely(jh->b_bh != bh))
1198 * jbd2_journal_get_write_access() - notify intent to modify a buffer
1199 * for metadata (not data) update.
1200 * @handle: transaction to add buffer modifications to
1201 * @bh: bh to be used for metadata writes
1203 * Returns: error code or 0 on success.
1205 * In full data journalling mode the buffer may be of type BJ_AsyncData,
1206 * because we're ``write()ing`` a buffer which is also part of a shared mapping.
1209 int jbd2_journal_get_write_access(handle_t *handle, struct buffer_head *bh)
1211 struct journal_head *jh;
1214 if (is_handle_aborted(handle))
1217 if (jbd2_write_access_granted(handle, bh, false))
1220 jh = jbd2_journal_add_journal_head(bh);
1221 /* We do not want to get caught playing with fields which the
1222 * log thread also manipulates. Make sure that the buffer
1223 * completes any outstanding IO before proceeding. */
1224 rc = do_get_write_access(handle, jh, 0);
1225 jbd2_journal_put_journal_head(jh);
1231 * When the user wants to journal a newly created buffer_head
1232 * (ie. getblk() returned a new buffer and we are going to populate it
1233 * manually rather than reading off disk), then we need to keep the
1234 * buffer_head locked until it has been completely filled with new
1235 * data. In this case, we should be able to make the assertion that
1236 * the bh is not already part of an existing transaction.
1238 * The buffer should already be locked by the caller by this point.
1239 * There is no lock ranking violation: it was a newly created,
1240 * unlocked buffer beforehand. */
1243 * jbd2_journal_get_create_access () - notify intent to use newly created bh
1244 * @handle: transaction to new buffer to
1247 * Call this if you create a new bh.
1249 int jbd2_journal_get_create_access(handle_t *handle, struct buffer_head *bh)
1251 transaction_t *transaction = handle->h_transaction;
1253 struct journal_head *jh = jbd2_journal_add_journal_head(bh);
1256 jbd_debug(5, "journal_head %p\n", jh);
1258 if (is_handle_aborted(handle))
1260 journal = transaction->t_journal;
1263 JBUFFER_TRACE(jh, "entry");
1265 * The buffer may already belong to this transaction due to pre-zeroing
1266 * in the filesystem's new_block code. It may also be on the previous,
1267 * committing transaction's lists, but it HAS to be in Forget state in
1268 * that case: the transaction must have deleted the buffer for it to be
1271 spin_lock(&jh->b_state_lock);
1272 J_ASSERT_JH(jh, (jh->b_transaction == transaction ||
1273 jh->b_transaction == NULL ||
1274 (jh->b_transaction == journal->j_committing_transaction &&
1275 jh->b_jlist == BJ_Forget)));
1277 J_ASSERT_JH(jh, jh->b_next_transaction == NULL);
1278 J_ASSERT_JH(jh, buffer_locked(jh2bh(jh)));
1280 if (jh->b_transaction == NULL) {
1282 * Previous jbd2_journal_forget() could have left the buffer
1283 * with jbddirty bit set because it was being committed. When
1284 * the commit finished, we've filed the buffer for
1285 * checkpointing and marked it dirty. Now we are reallocating
1286 * the buffer so the transaction freeing it must have
1287 * committed and so it's safe to clear the dirty bit.
1289 clear_buffer_dirty(jh2bh(jh));
1290 /* first access by this transaction */
1293 JBUFFER_TRACE(jh, "file as BJ_Reserved");
1294 spin_lock(&journal->j_list_lock);
1295 __jbd2_journal_file_buffer(jh, transaction, BJ_Reserved);
1296 spin_unlock(&journal->j_list_lock);
1297 } else if (jh->b_transaction == journal->j_committing_transaction) {
1298 /* first access by this transaction */
1301 JBUFFER_TRACE(jh, "set next transaction");
1302 spin_lock(&journal->j_list_lock);
1303 jh->b_next_transaction = transaction;
1304 spin_unlock(&journal->j_list_lock);
1306 spin_unlock(&jh->b_state_lock);
1309 * akpm: I added this. ext3_alloc_branch can pick up new indirect
1310 * blocks which contain freed but then revoked metadata. We need
1311 * to cancel the revoke in case we end up freeing it yet again
1312 * and the reallocating as data - this would cause a second revoke,
1313 * which hits an assertion error.
1315 JBUFFER_TRACE(jh, "cancelling revoke");
1316 jbd2_journal_cancel_revoke(handle, jh);
1318 jbd2_journal_put_journal_head(jh);
1323 * jbd2_journal_get_undo_access() - Notify intent to modify metadata with
1324 * non-rewindable consequences
1325 * @handle: transaction
1326 * @bh: buffer to undo
1328 * Sometimes there is a need to distinguish between metadata which has
1329 * been committed to disk and that which has not. The ext3fs code uses
1330 * this for freeing and allocating space, we have to make sure that we
1331 * do not reuse freed space until the deallocation has been committed,
1332 * since if we overwrote that space we would make the delete
1333 * un-rewindable in case of a crash.
1335 * To deal with that, jbd2_journal_get_undo_access requests write access to a
1336 * buffer for parts of non-rewindable operations such as delete
1337 * operations on the bitmaps. The journaling code must keep a copy of
1338 * the buffer's contents prior to the undo_access call until such time
1339 * as we know that the buffer has definitely been committed to disk.
1341 * We never need to know which transaction the committed data is part
1342 * of, buffers touched here are guaranteed to be dirtied later and so
1343 * will be committed to a new transaction in due course, at which point
1344 * we can discard the old committed data pointer.
1346 * Returns error number or 0 on success.
1348 int jbd2_journal_get_undo_access(handle_t *handle, struct buffer_head *bh)
1351 struct journal_head *jh;
1352 char *committed_data = NULL;
1354 if (is_handle_aborted(handle))
1357 if (jbd2_write_access_granted(handle, bh, true))
1360 jh = jbd2_journal_add_journal_head(bh);
1361 JBUFFER_TRACE(jh, "entry");
1364 * Do this first --- it can drop the journal lock, so we want to
1365 * make sure that obtaining the committed_data is done
1366 * atomically wrt. completion of any outstanding commits.
1368 err = do_get_write_access(handle, jh, 1);
1373 if (!jh->b_committed_data)
1374 committed_data = jbd2_alloc(jh2bh(jh)->b_size,
1375 GFP_NOFS|__GFP_NOFAIL);
1377 spin_lock(&jh->b_state_lock);
1378 if (!jh->b_committed_data) {
1379 /* Copy out the current buffer contents into the
1380 * preserved, committed copy. */
1381 JBUFFER_TRACE(jh, "generate b_committed data");
1382 if (!committed_data) {
1383 spin_unlock(&jh->b_state_lock);
1387 jh->b_committed_data = committed_data;
1388 committed_data = NULL;
1389 memcpy(jh->b_committed_data, bh->b_data, bh->b_size);
1391 spin_unlock(&jh->b_state_lock);
1393 jbd2_journal_put_journal_head(jh);
1394 if (unlikely(committed_data))
1395 jbd2_free(committed_data, bh->b_size);
1400 * jbd2_journal_set_triggers() - Add triggers for commit writeout
1401 * @bh: buffer to trigger on
1402 * @type: struct jbd2_buffer_trigger_type containing the trigger(s).
1404 * Set any triggers on this journal_head. This is always safe, because
1405 * triggers for a committing buffer will be saved off, and triggers for
1406 * a running transaction will match the buffer in that transaction.
1408 * Call with NULL to clear the triggers.
1410 void jbd2_journal_set_triggers(struct buffer_head *bh,
1411 struct jbd2_buffer_trigger_type *type)
1413 struct journal_head *jh = jbd2_journal_grab_journal_head(bh);
1417 jh->b_triggers = type;
1418 jbd2_journal_put_journal_head(jh);
1421 void jbd2_buffer_frozen_trigger(struct journal_head *jh, void *mapped_data,
1422 struct jbd2_buffer_trigger_type *triggers)
1424 struct buffer_head *bh = jh2bh(jh);
1426 if (!triggers || !triggers->t_frozen)
1429 triggers->t_frozen(triggers, bh, mapped_data, bh->b_size);
1432 void jbd2_buffer_abort_trigger(struct journal_head *jh,
1433 struct jbd2_buffer_trigger_type *triggers)
1435 if (!triggers || !triggers->t_abort)
1438 triggers->t_abort(triggers, jh2bh(jh));
1442 * jbd2_journal_dirty_metadata() - mark a buffer as containing dirty metadata
1443 * @handle: transaction to add buffer to.
1444 * @bh: buffer to mark
1446 * mark dirty metadata which needs to be journaled as part of the current
1449 * The buffer must have previously had jbd2_journal_get_write_access()
1450 * called so that it has a valid journal_head attached to the buffer
1453 * The buffer is placed on the transaction's metadata list and is marked
1454 * as belonging to the transaction.
1456 * Returns error number or 0 on success.
1458 * Special care needs to be taken if the buffer already belongs to the
1459 * current committing transaction (in which case we should have frozen
1460 * data present for that commit). In that case, we don't relink the
1461 * buffer: that only gets done when the old transaction finally
1462 * completes its commit.
1464 int jbd2_journal_dirty_metadata(handle_t *handle, struct buffer_head *bh)
1466 transaction_t *transaction = handle->h_transaction;
1468 struct journal_head *jh;
1471 if (!buffer_jbd(bh))
1475 * We don't grab jh reference here since the buffer must be part
1476 * of the running transaction.
1479 jbd_debug(5, "journal_head %p\n", jh);
1480 JBUFFER_TRACE(jh, "entry");
1483 * This and the following assertions are unreliable since we may see jh
1484 * in inconsistent state unless we grab bh_state lock. But this is
1485 * crucial to catch bugs so let's do a reliable check until the
1486 * lockless handling is fully proven.
1488 if (data_race(jh->b_transaction != transaction &&
1489 jh->b_next_transaction != transaction)) {
1490 spin_lock(&jh->b_state_lock);
1491 J_ASSERT_JH(jh, jh->b_transaction == transaction ||
1492 jh->b_next_transaction == transaction);
1493 spin_unlock(&jh->b_state_lock);
1495 if (jh->b_modified == 1) {
1496 /* If it's in our transaction it must be in BJ_Metadata list. */
1497 if (data_race(jh->b_transaction == transaction &&
1498 jh->b_jlist != BJ_Metadata)) {
1499 spin_lock(&jh->b_state_lock);
1500 if (jh->b_transaction == transaction &&
1501 jh->b_jlist != BJ_Metadata)
1502 pr_err("JBD2: assertion failure: h_type=%u "
1503 "h_line_no=%u block_no=%llu jlist=%u\n",
1504 handle->h_type, handle->h_line_no,
1505 (unsigned long long) bh->b_blocknr,
1507 J_ASSERT_JH(jh, jh->b_transaction != transaction ||
1508 jh->b_jlist == BJ_Metadata);
1509 spin_unlock(&jh->b_state_lock);
1514 journal = transaction->t_journal;
1515 spin_lock(&jh->b_state_lock);
1517 if (is_handle_aborted(handle)) {
1519 * Check journal aborting with @jh->b_state_lock locked,
1520 * since 'jh->b_transaction' could be replaced with
1521 * 'jh->b_next_transaction' during old transaction
1522 * committing if journal aborted, which may fail
1523 * assertion on 'jh->b_frozen_data == NULL'.
1529 if (jh->b_modified == 0) {
1531 * This buffer's got modified and becoming part
1532 * of the transaction. This needs to be done
1533 * once a transaction -bzzz
1535 if (WARN_ON_ONCE(jbd2_handle_buffer_credits(handle) <= 0)) {
1540 handle->h_total_credits--;
1544 * fastpath, to avoid expensive locking. If this buffer is already
1545 * on the running transaction's metadata list there is nothing to do.
1546 * Nobody can take it off again because there is a handle open.
1547 * I _think_ we're OK here with SMP barriers - a mistaken decision will
1548 * result in this test being false, so we go in and take the locks.
1550 if (jh->b_transaction == transaction && jh->b_jlist == BJ_Metadata) {
1551 JBUFFER_TRACE(jh, "fastpath");
1552 if (unlikely(jh->b_transaction !=
1553 journal->j_running_transaction)) {
1554 printk(KERN_ERR "JBD2: %s: "
1555 "jh->b_transaction (%llu, %p, %u) != "
1556 "journal->j_running_transaction (%p, %u)\n",
1558 (unsigned long long) bh->b_blocknr,
1560 jh->b_transaction ? jh->b_transaction->t_tid : 0,
1561 journal->j_running_transaction,
1562 journal->j_running_transaction ?
1563 journal->j_running_transaction->t_tid : 0);
1569 set_buffer_jbddirty(bh);
1572 * Metadata already on the current transaction list doesn't
1573 * need to be filed. Metadata on another transaction's list must
1574 * be committing, and will be refiled once the commit completes:
1575 * leave it alone for now.
1577 if (jh->b_transaction != transaction) {
1578 JBUFFER_TRACE(jh, "already on other transaction");
1579 if (unlikely(((jh->b_transaction !=
1580 journal->j_committing_transaction)) ||
1581 (jh->b_next_transaction != transaction))) {
1582 printk(KERN_ERR "jbd2_journal_dirty_metadata: %s: "
1583 "bad jh for block %llu: "
1584 "transaction (%p, %u), "
1585 "jh->b_transaction (%p, %u), "
1586 "jh->b_next_transaction (%p, %u), jlist %u\n",
1588 (unsigned long long) bh->b_blocknr,
1589 transaction, transaction->t_tid,
1592 jh->b_transaction->t_tid : 0,
1593 jh->b_next_transaction,
1594 jh->b_next_transaction ?
1595 jh->b_next_transaction->t_tid : 0,
1600 /* And this case is illegal: we can't reuse another
1601 * transaction's data buffer, ever. */
1605 /* That test should have eliminated the following case: */
1606 J_ASSERT_JH(jh, jh->b_frozen_data == NULL);
1608 JBUFFER_TRACE(jh, "file as BJ_Metadata");
1609 spin_lock(&journal->j_list_lock);
1610 __jbd2_journal_file_buffer(jh, transaction, BJ_Metadata);
1611 spin_unlock(&journal->j_list_lock);
1613 spin_unlock(&jh->b_state_lock);
1615 JBUFFER_TRACE(jh, "exit");
1620 * jbd2_journal_forget() - bforget() for potentially-journaled buffers.
1621 * @handle: transaction handle
1622 * @bh: bh to 'forget'
1624 * We can only do the bforget if there are no commits pending against the
1625 * buffer. If the buffer is dirty in the current running transaction we
1626 * can safely unlink it.
1628 * bh may not be a journalled buffer at all - it may be a non-JBD
1629 * buffer which came off the hashtable. Check for this.
1631 * Decrements bh->b_count by one.
1633 * Allow this call even if the handle has aborted --- it may be part of
1634 * the caller's cleanup after an abort.
1636 int jbd2_journal_forget(handle_t *handle, struct buffer_head *bh)
1638 transaction_t *transaction = handle->h_transaction;
1640 struct journal_head *jh;
1641 int drop_reserve = 0;
1643 int was_modified = 0;
1645 if (is_handle_aborted(handle))
1647 journal = transaction->t_journal;
1649 BUFFER_TRACE(bh, "entry");
1651 jh = jbd2_journal_grab_journal_head(bh);
1657 spin_lock(&jh->b_state_lock);
1659 /* Critical error: attempting to delete a bitmap buffer, maybe?
1660 * Don't do any jbd operations, and return an error. */
1661 if (!J_EXPECT_JH(jh, !jh->b_committed_data,
1662 "inconsistent data on disk")) {
1667 /* keep track of whether or not this transaction modified us */
1668 was_modified = jh->b_modified;
1671 * The buffer's going from the transaction, we must drop
1672 * all references -bzzz
1676 if (jh->b_transaction == transaction) {
1677 J_ASSERT_JH(jh, !jh->b_frozen_data);
1679 /* If we are forgetting a buffer which is already part
1680 * of this transaction, then we can just drop it from
1681 * the transaction immediately. */
1682 clear_buffer_dirty(bh);
1683 clear_buffer_jbddirty(bh);
1685 JBUFFER_TRACE(jh, "belongs to current transaction: unfile");
1688 * we only want to drop a reference if this transaction
1689 * modified the buffer
1695 * We are no longer going to journal this buffer.
1696 * However, the commit of this transaction is still
1697 * important to the buffer: the delete that we are now
1698 * processing might obsolete an old log entry, so by
1699 * committing, we can satisfy the buffer's checkpoint.
1701 * So, if we have a checkpoint on the buffer, we should
1702 * now refile the buffer on our BJ_Forget list so that
1703 * we know to remove the checkpoint after we commit.
1706 spin_lock(&journal->j_list_lock);
1707 if (jh->b_cp_transaction) {
1708 __jbd2_journal_temp_unlink_buffer(jh);
1709 __jbd2_journal_file_buffer(jh, transaction, BJ_Forget);
1711 __jbd2_journal_unfile_buffer(jh);
1712 jbd2_journal_put_journal_head(jh);
1714 spin_unlock(&journal->j_list_lock);
1715 } else if (jh->b_transaction) {
1716 J_ASSERT_JH(jh, (jh->b_transaction ==
1717 journal->j_committing_transaction));
1718 /* However, if the buffer is still owned by a prior
1719 * (committing) transaction, we can't drop it yet... */
1720 JBUFFER_TRACE(jh, "belongs to older transaction");
1721 /* ... but we CAN drop it from the new transaction through
1722 * marking the buffer as freed and set j_next_transaction to
1723 * the new transaction, so that not only the commit code
1724 * knows it should clear dirty bits when it is done with the
1725 * buffer, but also the buffer can be checkpointed only
1726 * after the new transaction commits. */
1728 set_buffer_freed(bh);
1730 if (!jh->b_next_transaction) {
1731 spin_lock(&journal->j_list_lock);
1732 jh->b_next_transaction = transaction;
1733 spin_unlock(&journal->j_list_lock);
1735 J_ASSERT(jh->b_next_transaction == transaction);
1738 * only drop a reference if this transaction modified
1746 * Finally, if the buffer is not belongs to any
1747 * transaction, we can just drop it now if it has no
1750 spin_lock(&journal->j_list_lock);
1751 if (!jh->b_cp_transaction) {
1752 JBUFFER_TRACE(jh, "belongs to none transaction");
1753 spin_unlock(&journal->j_list_lock);
1758 * Otherwise, if the buffer has been written to disk,
1759 * it is safe to remove the checkpoint and drop it.
1761 if (!buffer_dirty(bh)) {
1762 __jbd2_journal_remove_checkpoint(jh);
1763 spin_unlock(&journal->j_list_lock);
1768 * The buffer is still not written to disk, we should
1769 * attach this buffer to current transaction so that the
1770 * buffer can be checkpointed only after the current
1771 * transaction commits.
1773 clear_buffer_dirty(bh);
1774 __jbd2_journal_file_buffer(jh, transaction, BJ_Forget);
1775 spin_unlock(&journal->j_list_lock);
1779 spin_unlock(&jh->b_state_lock);
1780 jbd2_journal_put_journal_head(jh);
1782 /* no need to reserve log space for this block -bzzz */
1783 handle->h_total_credits++;
1789 * jbd2_journal_stop() - complete a transaction
1790 * @handle: transaction to complete.
1792 * All done for a particular handle.
1794 * There is not much action needed here. We just return any remaining
1795 * buffer credits to the transaction and remove the handle. The only
1796 * complication is that we need to start a commit operation if the
1797 * filesystem is marked for synchronous update.
1799 * jbd2_journal_stop itself will not usually return an error, but it may
1800 * do so in unusual circumstances. In particular, expect it to
1801 * return -EIO if a jbd2_journal_abort has been executed since the
1802 * transaction began.
1804 int jbd2_journal_stop(handle_t *handle)
1806 transaction_t *transaction = handle->h_transaction;
1808 int err = 0, wait_for_commit = 0;
1812 if (--handle->h_ref > 0) {
1813 jbd_debug(4, "h_ref %d -> %d\n", handle->h_ref + 1,
1815 if (is_handle_aborted(handle))
1821 * Handle is already detached from the transaction so there is
1822 * nothing to do other than free the handle.
1824 memalloc_nofs_restore(handle->saved_alloc_context);
1827 journal = transaction->t_journal;
1828 tid = transaction->t_tid;
1830 if (is_handle_aborted(handle))
1833 jbd_debug(4, "Handle %p going down\n", handle);
1834 trace_jbd2_handle_stats(journal->j_fs_dev->bd_dev,
1835 tid, handle->h_type, handle->h_line_no,
1836 jiffies - handle->h_start_jiffies,
1837 handle->h_sync, handle->h_requested_credits,
1838 (handle->h_requested_credits -
1839 handle->h_total_credits));
1842 * Implement synchronous transaction batching. If the handle
1843 * was synchronous, don't force a commit immediately. Let's
1844 * yield and let another thread piggyback onto this
1845 * transaction. Keep doing that while new threads continue to
1846 * arrive. It doesn't cost much - we're about to run a commit
1847 * and sleep on IO anyway. Speeds up many-threaded, many-dir
1848 * operations by 30x or more...
1850 * We try and optimize the sleep time against what the
1851 * underlying disk can do, instead of having a static sleep
1852 * time. This is useful for the case where our storage is so
1853 * fast that it is more optimal to go ahead and force a flush
1854 * and wait for the transaction to be committed than it is to
1855 * wait for an arbitrary amount of time for new writers to
1856 * join the transaction. We achieve this by measuring how
1857 * long it takes to commit a transaction, and compare it with
1858 * how long this transaction has been running, and if run time
1859 * < commit time then we sleep for the delta and commit. This
1860 * greatly helps super fast disks that would see slowdowns as
1861 * more threads started doing fsyncs.
1863 * But don't do this if this process was the most recent one
1864 * to perform a synchronous write. We do this to detect the
1865 * case where a single process is doing a stream of sync
1866 * writes. No point in waiting for joiners in that case.
1868 * Setting max_batch_time to 0 disables this completely.
1871 if (handle->h_sync && journal->j_last_sync_writer != pid &&
1872 journal->j_max_batch_time) {
1873 u64 commit_time, trans_time;
1875 journal->j_last_sync_writer = pid;
1877 read_lock(&journal->j_state_lock);
1878 commit_time = journal->j_average_commit_time;
1879 read_unlock(&journal->j_state_lock);
1881 trans_time = ktime_to_ns(ktime_sub(ktime_get(),
1882 transaction->t_start_time));
1884 commit_time = max_t(u64, commit_time,
1885 1000*journal->j_min_batch_time);
1886 commit_time = min_t(u64, commit_time,
1887 1000*journal->j_max_batch_time);
1889 if (trans_time < commit_time) {
1890 ktime_t expires = ktime_add_ns(ktime_get(),
1892 set_current_state(TASK_UNINTERRUPTIBLE);
1893 schedule_hrtimeout(&expires, HRTIMER_MODE_ABS);
1898 transaction->t_synchronous_commit = 1;
1901 * If the handle is marked SYNC, we need to set another commit
1902 * going! We also want to force a commit if the transaction is too
1905 if (handle->h_sync ||
1906 time_after_eq(jiffies, transaction->t_expires)) {
1907 /* Do this even for aborted journals: an abort still
1908 * completes the commit thread, it just doesn't write
1909 * anything to disk. */
1911 jbd_debug(2, "transaction too old, requesting commit for "
1912 "handle %p\n", handle);
1913 /* This is non-blocking */
1914 jbd2_log_start_commit(journal, tid);
1917 * Special case: JBD2_SYNC synchronous updates require us
1918 * to wait for the commit to complete.
1920 if (handle->h_sync && !(current->flags & PF_MEMALLOC))
1921 wait_for_commit = 1;
1925 * Once stop_this_handle() drops t_updates, the transaction could start
1926 * committing on us and eventually disappear. So we must not
1927 * dereference transaction pointer again after calling
1928 * stop_this_handle().
1930 stop_this_handle(handle);
1932 if (wait_for_commit)
1933 err = jbd2_log_wait_commit(journal, tid);
1936 if (handle->h_rsv_handle)
1937 jbd2_free_handle(handle->h_rsv_handle);
1938 jbd2_free_handle(handle);
1944 * List management code snippets: various functions for manipulating the
1945 * transaction buffer lists.
1950 * Append a buffer to a transaction list, given the transaction's list head
1953 * j_list_lock is held.
1955 * jh->b_state_lock is held.
1959 __blist_add_buffer(struct journal_head **list, struct journal_head *jh)
1962 jh->b_tnext = jh->b_tprev = jh;
1965 /* Insert at the tail of the list to preserve order */
1966 struct journal_head *first = *list, *last = first->b_tprev;
1968 jh->b_tnext = first;
1969 last->b_tnext = first->b_tprev = jh;
1974 * Remove a buffer from a transaction list, given the transaction's list
1977 * Called with j_list_lock held, and the journal may not be locked.
1979 * jh->b_state_lock is held.
1983 __blist_del_buffer(struct journal_head **list, struct journal_head *jh)
1986 *list = jh->b_tnext;
1990 jh->b_tprev->b_tnext = jh->b_tnext;
1991 jh->b_tnext->b_tprev = jh->b_tprev;
1995 * Remove a buffer from the appropriate transaction list.
1997 * Note that this function can *change* the value of
1998 * bh->b_transaction->t_buffers, t_forget, t_shadow_list, t_log_list or
1999 * t_reserved_list. If the caller is holding onto a copy of one of these
2000 * pointers, it could go bad. Generally the caller needs to re-read the
2001 * pointer from the transaction_t.
2003 * Called under j_list_lock.
2005 static void __jbd2_journal_temp_unlink_buffer(struct journal_head *jh)
2007 struct journal_head **list = NULL;
2008 transaction_t *transaction;
2009 struct buffer_head *bh = jh2bh(jh);
2011 lockdep_assert_held(&jh->b_state_lock);
2012 transaction = jh->b_transaction;
2014 assert_spin_locked(&transaction->t_journal->j_list_lock);
2016 J_ASSERT_JH(jh, jh->b_jlist < BJ_Types);
2017 if (jh->b_jlist != BJ_None)
2018 J_ASSERT_JH(jh, transaction != NULL);
2020 switch (jh->b_jlist) {
2024 transaction->t_nr_buffers--;
2025 J_ASSERT_JH(jh, transaction->t_nr_buffers >= 0);
2026 list = &transaction->t_buffers;
2029 list = &transaction->t_forget;
2032 list = &transaction->t_shadow_list;
2035 list = &transaction->t_reserved_list;
2039 __blist_del_buffer(list, jh);
2040 jh->b_jlist = BJ_None;
2041 if (transaction && is_journal_aborted(transaction->t_journal))
2042 clear_buffer_jbddirty(bh);
2043 else if (test_clear_buffer_jbddirty(bh))
2044 mark_buffer_dirty(bh); /* Expose it to the VM */
2048 * Remove buffer from all transactions. The caller is responsible for dropping
2049 * the jh reference that belonged to the transaction.
2051 * Called with bh_state lock and j_list_lock
2053 static void __jbd2_journal_unfile_buffer(struct journal_head *jh)
2055 J_ASSERT_JH(jh, jh->b_transaction != NULL);
2056 J_ASSERT_JH(jh, jh->b_next_transaction == NULL);
2058 __jbd2_journal_temp_unlink_buffer(jh);
2059 jh->b_transaction = NULL;
2062 void jbd2_journal_unfile_buffer(journal_t *journal, struct journal_head *jh)
2064 struct buffer_head *bh = jh2bh(jh);
2066 /* Get reference so that buffer cannot be freed before we unlock it */
2068 spin_lock(&jh->b_state_lock);
2069 spin_lock(&journal->j_list_lock);
2070 __jbd2_journal_unfile_buffer(jh);
2071 spin_unlock(&journal->j_list_lock);
2072 spin_unlock(&jh->b_state_lock);
2073 jbd2_journal_put_journal_head(jh);
2078 * Called from jbd2_journal_try_to_free_buffers().
2080 * Called under jh->b_state_lock
2083 __journal_try_to_free_buffer(journal_t *journal, struct buffer_head *bh)
2085 struct journal_head *jh;
2089 if (buffer_locked(bh) || buffer_dirty(bh))
2092 if (jh->b_next_transaction != NULL || jh->b_transaction != NULL)
2095 spin_lock(&journal->j_list_lock);
2096 if (jh->b_cp_transaction != NULL) {
2097 /* written-back checkpointed metadata buffer */
2098 JBUFFER_TRACE(jh, "remove from checkpoint list");
2099 __jbd2_journal_remove_checkpoint(jh);
2101 spin_unlock(&journal->j_list_lock);
2107 * jbd2_journal_try_to_free_buffers() - try to free page buffers.
2108 * @journal: journal for operation
2109 * @page: to try and free
2111 * For all the buffers on this page,
2112 * if they are fully written out ordered data, move them onto BUF_CLEAN
2113 * so try_to_free_buffers() can reap them.
2115 * This function returns non-zero if we wish try_to_free_buffers()
2116 * to be called. We do this if the page is releasable by try_to_free_buffers().
2117 * We also do it if the page has locked or dirty buffers and the caller wants
2118 * us to perform sync or async writeout.
2120 * This complicates JBD locking somewhat. We aren't protected by the
2121 * BKL here. We wish to remove the buffer from its committing or
2122 * running transaction's ->t_datalist via __jbd2_journal_unfile_buffer.
2124 * This may *change* the value of transaction_t->t_datalist, so anyone
2125 * who looks at t_datalist needs to lock against this function.
2127 * Even worse, someone may be doing a jbd2_journal_dirty_data on this
2128 * buffer. So we need to lock against that. jbd2_journal_dirty_data()
2129 * will come out of the lock with the buffer dirty, which makes it
2130 * ineligible for release here.
2132 * Who else is affected by this? hmm... Really the only contender
2133 * is do_get_write_access() - it could be looking at the buffer while
2134 * journal_try_to_free_buffer() is changing its state. But that
2135 * cannot happen because we never reallocate freed data as metadata
2136 * while the data is part of a transaction. Yes?
2138 * Return 0 on failure, 1 on success
2140 int jbd2_journal_try_to_free_buffers(journal_t *journal, struct page *page)
2142 struct buffer_head *head;
2143 struct buffer_head *bh;
2144 bool has_write_io_error = false;
2147 J_ASSERT(PageLocked(page));
2149 head = page_buffers(page);
2152 struct journal_head *jh;
2155 * We take our own ref against the journal_head here to avoid
2156 * having to add tons of locking around each instance of
2157 * jbd2_journal_put_journal_head().
2159 jh = jbd2_journal_grab_journal_head(bh);
2163 spin_lock(&jh->b_state_lock);
2164 __journal_try_to_free_buffer(journal, bh);
2165 spin_unlock(&jh->b_state_lock);
2166 jbd2_journal_put_journal_head(jh);
2171 * If we free a metadata buffer which has been failed to
2172 * write out, the jbd2 checkpoint procedure will not detect
2173 * this failure and may lead to filesystem inconsistency
2174 * after cleanup journal tail.
2176 if (buffer_write_io_error(bh)) {
2177 pr_err("JBD2: Error while async write back metadata bh %llu.",
2178 (unsigned long long)bh->b_blocknr);
2179 has_write_io_error = true;
2181 } while ((bh = bh->b_this_page) != head);
2183 ret = try_to_free_buffers(page);
2186 if (has_write_io_error)
2187 jbd2_journal_abort(journal, -EIO);
2193 * This buffer is no longer needed. If it is on an older transaction's
2194 * checkpoint list we need to record it on this transaction's forget list
2195 * to pin this buffer (and hence its checkpointing transaction) down until
2196 * this transaction commits. If the buffer isn't on a checkpoint list, we
2198 * Returns non-zero if JBD no longer has an interest in the buffer.
2200 * Called under j_list_lock.
2202 * Called under jh->b_state_lock.
2204 static int __dispose_buffer(struct journal_head *jh, transaction_t *transaction)
2207 struct buffer_head *bh = jh2bh(jh);
2209 if (jh->b_cp_transaction) {
2210 JBUFFER_TRACE(jh, "on running+cp transaction");
2211 __jbd2_journal_temp_unlink_buffer(jh);
2213 * We don't want to write the buffer anymore, clear the
2214 * bit so that we don't confuse checks in
2215 * __journal_file_buffer
2217 clear_buffer_dirty(bh);
2218 __jbd2_journal_file_buffer(jh, transaction, BJ_Forget);
2221 JBUFFER_TRACE(jh, "on running transaction");
2222 __jbd2_journal_unfile_buffer(jh);
2223 jbd2_journal_put_journal_head(jh);
2229 * jbd2_journal_invalidatepage
2231 * This code is tricky. It has a number of cases to deal with.
2233 * There are two invariants which this code relies on:
2235 * i_size must be updated on disk before we start calling invalidatepage on the
2238 * This is done in ext3 by defining an ext3_setattr method which
2239 * updates i_size before truncate gets going. By maintaining this
2240 * invariant, we can be sure that it is safe to throw away any buffers
2241 * attached to the current transaction: once the transaction commits,
2242 * we know that the data will not be needed.
2244 * Note however that we can *not* throw away data belonging to the
2245 * previous, committing transaction!
2247 * Any disk blocks which *are* part of the previous, committing
2248 * transaction (and which therefore cannot be discarded immediately) are
2249 * not going to be reused in the new running transaction
2251 * The bitmap committed_data images guarantee this: any block which is
2252 * allocated in one transaction and removed in the next will be marked
2253 * as in-use in the committed_data bitmap, so cannot be reused until
2254 * the next transaction to delete the block commits. This means that
2255 * leaving committing buffers dirty is quite safe: the disk blocks
2256 * cannot be reallocated to a different file and so buffer aliasing is
2260 * The above applies mainly to ordered data mode. In writeback mode we
2261 * don't make guarantees about the order in which data hits disk --- in
2262 * particular we don't guarantee that new dirty data is flushed before
2263 * transaction commit --- so it is always safe just to discard data
2264 * immediately in that mode. --sct
2268 * The journal_unmap_buffer helper function returns zero if the buffer
2269 * concerned remains pinned as an anonymous buffer belonging to an older
2272 * We're outside-transaction here. Either or both of j_running_transaction
2273 * and j_committing_transaction may be NULL.
2275 static int journal_unmap_buffer(journal_t *journal, struct buffer_head *bh,
2278 transaction_t *transaction;
2279 struct journal_head *jh;
2282 BUFFER_TRACE(bh, "entry");
2285 * It is safe to proceed here without the j_list_lock because the
2286 * buffers cannot be stolen by try_to_free_buffers as long as we are
2287 * holding the page lock. --sct
2290 jh = jbd2_journal_grab_journal_head(bh);
2292 goto zap_buffer_unlocked;
2294 /* OK, we have data buffer in journaled mode */
2295 write_lock(&journal->j_state_lock);
2296 spin_lock(&jh->b_state_lock);
2297 spin_lock(&journal->j_list_lock);
2300 * We cannot remove the buffer from checkpoint lists until the
2301 * transaction adding inode to orphan list (let's call it T)
2302 * is committed. Otherwise if the transaction changing the
2303 * buffer would be cleaned from the journal before T is
2304 * committed, a crash will cause that the correct contents of
2305 * the buffer will be lost. On the other hand we have to
2306 * clear the buffer dirty bit at latest at the moment when the
2307 * transaction marking the buffer as freed in the filesystem
2308 * structures is committed because from that moment on the
2309 * block can be reallocated and used by a different page.
2310 * Since the block hasn't been freed yet but the inode has
2311 * already been added to orphan list, it is safe for us to add
2312 * the buffer to BJ_Forget list of the newest transaction.
2314 * Also we have to clear buffer_mapped flag of a truncated buffer
2315 * because the buffer_head may be attached to the page straddling
2316 * i_size (can happen only when blocksize < pagesize) and thus the
2317 * buffer_head can be reused when the file is extended again. So we end
2318 * up keeping around invalidated buffers attached to transactions'
2319 * BJ_Forget list just to stop checkpointing code from cleaning up
2320 * the transaction this buffer was modified in.
2322 transaction = jh->b_transaction;
2323 if (transaction == NULL) {
2324 /* First case: not on any transaction. If it
2325 * has no checkpoint link, then we can zap it:
2326 * it's a writeback-mode buffer so we don't care
2327 * if it hits disk safely. */
2328 if (!jh->b_cp_transaction) {
2329 JBUFFER_TRACE(jh, "not on any transaction: zap");
2333 if (!buffer_dirty(bh)) {
2334 /* bdflush has written it. We can drop it now */
2335 __jbd2_journal_remove_checkpoint(jh);
2339 /* OK, it must be in the journal but still not
2340 * written fully to disk: it's metadata or
2341 * journaled data... */
2343 if (journal->j_running_transaction) {
2344 /* ... and once the current transaction has
2345 * committed, the buffer won't be needed any
2347 JBUFFER_TRACE(jh, "checkpointed: add to BJ_Forget");
2348 may_free = __dispose_buffer(jh,
2349 journal->j_running_transaction);
2352 /* There is no currently-running transaction. So the
2353 * orphan record which we wrote for this file must have
2354 * passed into commit. We must attach this buffer to
2355 * the committing transaction, if it exists. */
2356 if (journal->j_committing_transaction) {
2357 JBUFFER_TRACE(jh, "give to committing trans");
2358 may_free = __dispose_buffer(jh,
2359 journal->j_committing_transaction);
2362 /* The orphan record's transaction has
2363 * committed. We can cleanse this buffer */
2364 clear_buffer_jbddirty(bh);
2365 __jbd2_journal_remove_checkpoint(jh);
2369 } else if (transaction == journal->j_committing_transaction) {
2370 JBUFFER_TRACE(jh, "on committing transaction");
2372 * The buffer is committing, we simply cannot touch
2373 * it. If the page is straddling i_size we have to wait
2374 * for commit and try again.
2377 spin_unlock(&journal->j_list_lock);
2378 spin_unlock(&jh->b_state_lock);
2379 write_unlock(&journal->j_state_lock);
2380 jbd2_journal_put_journal_head(jh);
2381 /* Already zapped buffer? Nothing to do... */
2387 * OK, buffer won't be reachable after truncate. We just clear
2388 * b_modified to not confuse transaction credit accounting, and
2389 * set j_next_transaction to the running transaction (if there
2390 * is one) and mark buffer as freed so that commit code knows
2391 * it should clear dirty bits when it is done with the buffer.
2393 set_buffer_freed(bh);
2394 if (journal->j_running_transaction && buffer_jbddirty(bh))
2395 jh->b_next_transaction = journal->j_running_transaction;
2397 spin_unlock(&journal->j_list_lock);
2398 spin_unlock(&jh->b_state_lock);
2399 write_unlock(&journal->j_state_lock);
2400 jbd2_journal_put_journal_head(jh);
2403 /* Good, the buffer belongs to the running transaction.
2404 * We are writing our own transaction's data, not any
2405 * previous one's, so it is safe to throw it away
2406 * (remember that we expect the filesystem to have set
2407 * i_size already for this truncate so recovery will not
2408 * expose the disk blocks we are discarding here.) */
2409 J_ASSERT_JH(jh, transaction == journal->j_running_transaction);
2410 JBUFFER_TRACE(jh, "on running transaction");
2411 may_free = __dispose_buffer(jh, transaction);
2416 * This is tricky. Although the buffer is truncated, it may be reused
2417 * if blocksize < pagesize and it is attached to the page straddling
2418 * EOF. Since the buffer might have been added to BJ_Forget list of the
2419 * running transaction, journal_get_write_access() won't clear
2420 * b_modified and credit accounting gets confused. So clear b_modified
2424 spin_unlock(&journal->j_list_lock);
2425 spin_unlock(&jh->b_state_lock);
2426 write_unlock(&journal->j_state_lock);
2427 jbd2_journal_put_journal_head(jh);
2428 zap_buffer_unlocked:
2429 clear_buffer_dirty(bh);
2430 J_ASSERT_BH(bh, !buffer_jbddirty(bh));
2431 clear_buffer_mapped(bh);
2432 clear_buffer_req(bh);
2433 clear_buffer_new(bh);
2434 clear_buffer_delay(bh);
2435 clear_buffer_unwritten(bh);
2441 * jbd2_journal_invalidatepage()
2442 * @journal: journal to use for flush...
2443 * @page: page to flush
2444 * @offset: start of the range to invalidate
2445 * @length: length of the range to invalidate
2447 * Reap page buffers containing data after in the specified range in page.
2448 * Can return -EBUSY if buffers are part of the committing transaction and
2449 * the page is straddling i_size. Caller then has to wait for current commit
2452 int jbd2_journal_invalidatepage(journal_t *journal,
2454 unsigned int offset,
2455 unsigned int length)
2457 struct buffer_head *head, *bh, *next;
2458 unsigned int stop = offset + length;
2459 unsigned int curr_off = 0;
2460 int partial_page = (offset || length < PAGE_SIZE);
2464 if (!PageLocked(page))
2466 if (!page_has_buffers(page))
2469 BUG_ON(stop > PAGE_SIZE || stop < length);
2471 /* We will potentially be playing with lists other than just the
2472 * data lists (especially for journaled data mode), so be
2473 * cautious in our locking. */
2475 head = bh = page_buffers(page);
2477 unsigned int next_off = curr_off + bh->b_size;
2478 next = bh->b_this_page;
2480 if (next_off > stop)
2483 if (offset <= curr_off) {
2484 /* This block is wholly outside the truncation point */
2486 ret = journal_unmap_buffer(journal, bh, partial_page);
2492 curr_off = next_off;
2495 } while (bh != head);
2497 if (!partial_page) {
2498 if (may_free && try_to_free_buffers(page))
2499 J_ASSERT(!page_has_buffers(page));
2505 * File a buffer on the given transaction list.
2507 void __jbd2_journal_file_buffer(struct journal_head *jh,
2508 transaction_t *transaction, int jlist)
2510 struct journal_head **list = NULL;
2512 struct buffer_head *bh = jh2bh(jh);
2514 lockdep_assert_held(&jh->b_state_lock);
2515 assert_spin_locked(&transaction->t_journal->j_list_lock);
2517 J_ASSERT_JH(jh, jh->b_jlist < BJ_Types);
2518 J_ASSERT_JH(jh, jh->b_transaction == transaction ||
2519 jh->b_transaction == NULL);
2521 if (jh->b_transaction && jh->b_jlist == jlist)
2524 if (jlist == BJ_Metadata || jlist == BJ_Reserved ||
2525 jlist == BJ_Shadow || jlist == BJ_Forget) {
2527 * For metadata buffers, we track dirty bit in buffer_jbddirty
2528 * instead of buffer_dirty. We should not see a dirty bit set
2529 * here because we clear it in do_get_write_access but e.g.
2530 * tune2fs can modify the sb and set the dirty bit at any time
2531 * so we try to gracefully handle that.
2533 if (buffer_dirty(bh))
2534 warn_dirty_buffer(bh);
2535 if (test_clear_buffer_dirty(bh) ||
2536 test_clear_buffer_jbddirty(bh))
2540 if (jh->b_transaction)
2541 __jbd2_journal_temp_unlink_buffer(jh);
2543 jbd2_journal_grab_journal_head(bh);
2544 jh->b_transaction = transaction;
2548 J_ASSERT_JH(jh, !jh->b_committed_data);
2549 J_ASSERT_JH(jh, !jh->b_frozen_data);
2552 transaction->t_nr_buffers++;
2553 list = &transaction->t_buffers;
2556 list = &transaction->t_forget;
2559 list = &transaction->t_shadow_list;
2562 list = &transaction->t_reserved_list;
2566 __blist_add_buffer(list, jh);
2567 jh->b_jlist = jlist;
2570 set_buffer_jbddirty(bh);
2573 void jbd2_journal_file_buffer(struct journal_head *jh,
2574 transaction_t *transaction, int jlist)
2576 spin_lock(&jh->b_state_lock);
2577 spin_lock(&transaction->t_journal->j_list_lock);
2578 __jbd2_journal_file_buffer(jh, transaction, jlist);
2579 spin_unlock(&transaction->t_journal->j_list_lock);
2580 spin_unlock(&jh->b_state_lock);
2584 * Remove a buffer from its current buffer list in preparation for
2585 * dropping it from its current transaction entirely. If the buffer has
2586 * already started to be used by a subsequent transaction, refile the
2587 * buffer on that transaction's metadata list.
2589 * Called under j_list_lock
2590 * Called under jh->b_state_lock
2592 * When this function returns true, there's no next transaction to refile to
2593 * and the caller has to drop jh reference through
2594 * jbd2_journal_put_journal_head().
2596 bool __jbd2_journal_refile_buffer(struct journal_head *jh)
2598 int was_dirty, jlist;
2599 struct buffer_head *bh = jh2bh(jh);
2601 lockdep_assert_held(&jh->b_state_lock);
2602 if (jh->b_transaction)
2603 assert_spin_locked(&jh->b_transaction->t_journal->j_list_lock);
2605 /* If the buffer is now unused, just drop it. */
2606 if (jh->b_next_transaction == NULL) {
2607 __jbd2_journal_unfile_buffer(jh);
2612 * It has been modified by a later transaction: add it to the new
2613 * transaction's metadata list.
2616 was_dirty = test_clear_buffer_jbddirty(bh);
2617 __jbd2_journal_temp_unlink_buffer(jh);
2620 * b_transaction must be set, otherwise the new b_transaction won't
2621 * be holding jh reference
2623 J_ASSERT_JH(jh, jh->b_transaction != NULL);
2626 * We set b_transaction here because b_next_transaction will inherit
2627 * our jh reference and thus __jbd2_journal_file_buffer() must not
2630 WRITE_ONCE(jh->b_transaction, jh->b_next_transaction);
2631 WRITE_ONCE(jh->b_next_transaction, NULL);
2632 if (buffer_freed(bh))
2634 else if (jh->b_modified)
2635 jlist = BJ_Metadata;
2637 jlist = BJ_Reserved;
2638 __jbd2_journal_file_buffer(jh, jh->b_transaction, jlist);
2639 J_ASSERT_JH(jh, jh->b_transaction->t_state == T_RUNNING);
2642 set_buffer_jbddirty(bh);
2647 * __jbd2_journal_refile_buffer() with necessary locking added. We take our
2648 * bh reference so that we can safely unlock bh.
2650 * The jh and bh may be freed by this call.
2652 void jbd2_journal_refile_buffer(journal_t *journal, struct journal_head *jh)
2656 spin_lock(&jh->b_state_lock);
2657 spin_lock(&journal->j_list_lock);
2658 drop = __jbd2_journal_refile_buffer(jh);
2659 spin_unlock(&jh->b_state_lock);
2660 spin_unlock(&journal->j_list_lock);
2662 jbd2_journal_put_journal_head(jh);
2666 * File inode in the inode list of the handle's transaction
2668 static int jbd2_journal_file_inode(handle_t *handle, struct jbd2_inode *jinode,
2669 unsigned long flags, loff_t start_byte, loff_t end_byte)
2671 transaction_t *transaction = handle->h_transaction;
2674 if (is_handle_aborted(handle))
2676 journal = transaction->t_journal;
2678 jbd_debug(4, "Adding inode %lu, tid:%d\n", jinode->i_vfs_inode->i_ino,
2679 transaction->t_tid);
2681 spin_lock(&journal->j_list_lock);
2682 jinode->i_flags |= flags;
2684 if (jinode->i_dirty_end) {
2685 jinode->i_dirty_start = min(jinode->i_dirty_start, start_byte);
2686 jinode->i_dirty_end = max(jinode->i_dirty_end, end_byte);
2688 jinode->i_dirty_start = start_byte;
2689 jinode->i_dirty_end = end_byte;
2692 /* Is inode already attached where we need it? */
2693 if (jinode->i_transaction == transaction ||
2694 jinode->i_next_transaction == transaction)
2698 * We only ever set this variable to 1 so the test is safe. Since
2699 * t_need_data_flush is likely to be set, we do the test to save some
2700 * cacheline bouncing
2702 if (!transaction->t_need_data_flush)
2703 transaction->t_need_data_flush = 1;
2704 /* On some different transaction's list - should be
2705 * the committing one */
2706 if (jinode->i_transaction) {
2707 J_ASSERT(jinode->i_next_transaction == NULL);
2708 J_ASSERT(jinode->i_transaction ==
2709 journal->j_committing_transaction);
2710 jinode->i_next_transaction = transaction;
2713 /* Not on any transaction list... */
2714 J_ASSERT(!jinode->i_next_transaction);
2715 jinode->i_transaction = transaction;
2716 list_add(&jinode->i_list, &transaction->t_inode_list);
2718 spin_unlock(&journal->j_list_lock);
2723 int jbd2_journal_inode_ranged_write(handle_t *handle,
2724 struct jbd2_inode *jinode, loff_t start_byte, loff_t length)
2726 return jbd2_journal_file_inode(handle, jinode,
2727 JI_WRITE_DATA | JI_WAIT_DATA, start_byte,
2728 start_byte + length - 1);
2731 int jbd2_journal_inode_ranged_wait(handle_t *handle, struct jbd2_inode *jinode,
2732 loff_t start_byte, loff_t length)
2734 return jbd2_journal_file_inode(handle, jinode, JI_WAIT_DATA,
2735 start_byte, start_byte + length - 1);
2739 * File truncate and transaction commit interact with each other in a
2740 * non-trivial way. If a transaction writing data block A is
2741 * committing, we cannot discard the data by truncate until we have
2742 * written them. Otherwise if we crashed after the transaction with
2743 * write has committed but before the transaction with truncate has
2744 * committed, we could see stale data in block A. This function is a
2745 * helper to solve this problem. It starts writeout of the truncated
2746 * part in case it is in the committing transaction.
2748 * Filesystem code must call this function when inode is journaled in
2749 * ordered mode before truncation happens and after the inode has been
2750 * placed on orphan list with the new inode size. The second condition
2751 * avoids the race that someone writes new data and we start
2752 * committing the transaction after this function has been called but
2753 * before a transaction for truncate is started (and furthermore it
2754 * allows us to optimize the case where the addition to orphan list
2755 * happens in the same transaction as write --- we don't have to write
2756 * any data in such case).
2758 int jbd2_journal_begin_ordered_truncate(journal_t *journal,
2759 struct jbd2_inode *jinode,
2762 transaction_t *inode_trans, *commit_trans;
2765 /* This is a quick check to avoid locking if not necessary */
2766 if (!jinode->i_transaction)
2768 /* Locks are here just to force reading of recent values, it is
2769 * enough that the transaction was not committing before we started
2770 * a transaction adding the inode to orphan list */
2771 read_lock(&journal->j_state_lock);
2772 commit_trans = journal->j_committing_transaction;
2773 read_unlock(&journal->j_state_lock);
2774 spin_lock(&journal->j_list_lock);
2775 inode_trans = jinode->i_transaction;
2776 spin_unlock(&journal->j_list_lock);
2777 if (inode_trans == commit_trans) {
2778 ret = filemap_fdatawrite_range(jinode->i_vfs_inode->i_mapping,
2779 new_size, LLONG_MAX);
2781 jbd2_journal_abort(journal, ret);