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 atomic_set(&transaction->t_updates, 0);
111 atomic_set(&transaction->t_outstanding_credits,
112 jbd2_descriptor_blocks_per_trans(journal) +
113 atomic_read(&journal->j_reserved_credits));
114 atomic_set(&transaction->t_outstanding_revokes, 0);
115 atomic_set(&transaction->t_handle_count, 0);
116 INIT_LIST_HEAD(&transaction->t_inode_list);
117 INIT_LIST_HEAD(&transaction->t_private_list);
119 /* Set up the commit timer for the new transaction. */
120 journal->j_commit_timer.expires = round_jiffies_up(transaction->t_expires);
121 add_timer(&journal->j_commit_timer);
123 J_ASSERT(journal->j_running_transaction == NULL);
124 journal->j_running_transaction = transaction;
125 transaction->t_max_wait = 0;
126 transaction->t_start = jiffies;
127 transaction->t_requested = 0;
133 * A handle_t is an object which represents a single atomic update to a
134 * filesystem, and which tracks all of the modifications which form part
135 * of that one update.
139 * Update transaction's maximum wait time, if debugging is enabled.
141 * t_max_wait is carefully updated here with use of atomic compare exchange.
142 * Note that there could be multiplre threads trying to do this simultaneously
143 * hence using cmpxchg to avoid any use of locks in this case.
145 static inline void update_t_max_wait(transaction_t *transaction,
148 #ifdef CONFIG_JBD2_DEBUG
149 unsigned long oldts, newts;
150 if (jbd2_journal_enable_debug &&
151 time_after(transaction->t_start, ts)) {
152 newts = jbd2_time_diff(ts, transaction->t_start);
153 oldts = READ_ONCE(transaction->t_max_wait);
154 while (oldts < newts)
155 oldts = cmpxchg(&transaction->t_max_wait, oldts, newts);
161 * Wait until running transaction passes to T_FLUSH state and new transaction
162 * can thus be started. Also starts the commit if needed. The function expects
163 * running transaction to exist and releases j_state_lock.
165 static void wait_transaction_locked(journal_t *journal)
166 __releases(journal->j_state_lock)
170 tid_t tid = journal->j_running_transaction->t_tid;
172 prepare_to_wait_exclusive(&journal->j_wait_transaction_locked, &wait,
173 TASK_UNINTERRUPTIBLE);
174 need_to_start = !tid_geq(journal->j_commit_request, tid);
175 read_unlock(&journal->j_state_lock);
177 jbd2_log_start_commit(journal, tid);
178 jbd2_might_wait_for_commit(journal);
180 finish_wait(&journal->j_wait_transaction_locked, &wait);
184 * Wait until running transaction transitions from T_SWITCH to T_FLUSH
185 * state and new transaction can thus be started. The function releases
188 static void wait_transaction_switching(journal_t *journal)
189 __releases(journal->j_state_lock)
193 if (WARN_ON(!journal->j_running_transaction ||
194 journal->j_running_transaction->t_state != T_SWITCH)) {
195 read_unlock(&journal->j_state_lock);
198 prepare_to_wait_exclusive(&journal->j_wait_transaction_locked, &wait,
199 TASK_UNINTERRUPTIBLE);
200 read_unlock(&journal->j_state_lock);
202 * We don't call jbd2_might_wait_for_commit() here as there's no
203 * waiting for outstanding handles happening anymore in T_SWITCH state
204 * and handling of reserved handles actually relies on that for
208 finish_wait(&journal->j_wait_transaction_locked, &wait);
211 static void sub_reserved_credits(journal_t *journal, int blocks)
213 atomic_sub(blocks, &journal->j_reserved_credits);
214 wake_up(&journal->j_wait_reserved);
218 * Wait until we can add credits for handle to the running transaction. Called
219 * with j_state_lock held for reading. Returns 0 if handle joined the running
220 * transaction. Returns 1 if we had to wait, j_state_lock is dropped, and
223 * Note: because j_state_lock may be dropped depending on the return
224 * value, we need to fake out sparse so ti doesn't complain about a
225 * locking imbalance. Callers of add_transaction_credits will need to
226 * make a similar accomodation.
228 static int add_transaction_credits(journal_t *journal, int blocks,
230 __must_hold(&journal->j_state_lock)
232 transaction_t *t = journal->j_running_transaction;
234 int total = blocks + rsv_blocks;
237 * If the current transaction is locked down for commit, wait
238 * for the lock to be released.
240 if (t->t_state != T_RUNNING) {
241 WARN_ON_ONCE(t->t_state >= T_FLUSH);
242 wait_transaction_locked(journal);
243 __acquire(&journal->j_state_lock); /* fake out sparse */
248 * If there is not enough space left in the log to write all
249 * potential buffers requested by this operation, we need to
250 * stall pending a log checkpoint to free some more log space.
252 needed = atomic_add_return(total, &t->t_outstanding_credits);
253 if (needed > journal->j_max_transaction_buffers) {
255 * If the current transaction is already too large,
256 * then start to commit it: we can then go back and
257 * attach this handle to a new transaction.
259 atomic_sub(total, &t->t_outstanding_credits);
262 * Is the number of reserved credits in the current transaction too
263 * big to fit this handle? Wait until reserved credits are freed.
265 if (atomic_read(&journal->j_reserved_credits) + total >
266 journal->j_max_transaction_buffers) {
267 read_unlock(&journal->j_state_lock);
268 jbd2_might_wait_for_commit(journal);
269 wait_event(journal->j_wait_reserved,
270 atomic_read(&journal->j_reserved_credits) + total <=
271 journal->j_max_transaction_buffers);
272 __acquire(&journal->j_state_lock); /* fake out sparse */
276 wait_transaction_locked(journal);
277 __acquire(&journal->j_state_lock); /* fake out sparse */
282 * The commit code assumes that it can get enough log space
283 * without forcing a checkpoint. This is *critical* for
284 * correctness: a checkpoint of a buffer which is also
285 * associated with a committing transaction creates a deadlock,
286 * so commit simply cannot force through checkpoints.
288 * We must therefore ensure the necessary space in the journal
289 * *before* starting to dirty potentially checkpointed buffers
290 * in the new transaction.
292 if (jbd2_log_space_left(journal) < journal->j_max_transaction_buffers) {
293 atomic_sub(total, &t->t_outstanding_credits);
294 read_unlock(&journal->j_state_lock);
295 jbd2_might_wait_for_commit(journal);
296 write_lock(&journal->j_state_lock);
297 if (jbd2_log_space_left(journal) <
298 journal->j_max_transaction_buffers)
299 __jbd2_log_wait_for_space(journal);
300 write_unlock(&journal->j_state_lock);
301 __acquire(&journal->j_state_lock); /* fake out sparse */
305 /* No reservation? We are done... */
309 needed = atomic_add_return(rsv_blocks, &journal->j_reserved_credits);
310 /* We allow at most half of a transaction to be reserved */
311 if (needed > journal->j_max_transaction_buffers / 2) {
312 sub_reserved_credits(journal, rsv_blocks);
313 atomic_sub(total, &t->t_outstanding_credits);
314 read_unlock(&journal->j_state_lock);
315 jbd2_might_wait_for_commit(journal);
316 wait_event(journal->j_wait_reserved,
317 atomic_read(&journal->j_reserved_credits) + rsv_blocks
318 <= journal->j_max_transaction_buffers / 2);
319 __acquire(&journal->j_state_lock); /* fake out sparse */
326 * start_this_handle: Given a handle, deal with any locking or stalling
327 * needed to make sure that there is enough journal space for the handle
328 * to begin. Attach the handle to a transaction and set up the
329 * transaction's buffer credits.
332 static int start_this_handle(journal_t *journal, handle_t *handle,
335 transaction_t *transaction, *new_transaction = NULL;
336 int blocks = handle->h_total_credits;
338 unsigned long ts = jiffies;
340 if (handle->h_rsv_handle)
341 rsv_blocks = handle->h_rsv_handle->h_total_credits;
344 * Limit the number of reserved credits to 1/2 of maximum transaction
345 * size and limit the number of total credits to not exceed maximum
346 * transaction size per operation.
348 if ((rsv_blocks > journal->j_max_transaction_buffers / 2) ||
349 (rsv_blocks + blocks > journal->j_max_transaction_buffers)) {
350 printk(KERN_ERR "JBD2: %s wants too many credits "
351 "credits:%d rsv_credits:%d max:%d\n",
352 current->comm, blocks, rsv_blocks,
353 journal->j_max_transaction_buffers);
360 * This check is racy but it is just an optimization of allocating new
361 * transaction early if there are high chances we'll need it. If we
362 * guess wrong, we'll retry or free unused transaction.
364 if (!data_race(journal->j_running_transaction)) {
366 * If __GFP_FS is not present, then we may be being called from
367 * inside the fs writeback layer, so we MUST NOT fail.
369 if ((gfp_mask & __GFP_FS) == 0)
370 gfp_mask |= __GFP_NOFAIL;
371 new_transaction = kmem_cache_zalloc(transaction_cache,
373 if (!new_transaction)
377 jbd2_debug(3, "New handle %p going live.\n", handle);
380 * We need to hold j_state_lock until t_updates has been incremented,
381 * for proper journal barrier handling
384 read_lock(&journal->j_state_lock);
385 BUG_ON(journal->j_flags & JBD2_UNMOUNT);
386 if (is_journal_aborted(journal) ||
387 (journal->j_errno != 0 && !(journal->j_flags & JBD2_ACK_ERR))) {
388 read_unlock(&journal->j_state_lock);
389 jbd2_journal_free_transaction(new_transaction);
394 * Wait on the journal's transaction barrier if necessary. Specifically
395 * we allow reserved handles to proceed because otherwise commit could
396 * deadlock on page writeback not being able to complete.
398 if (!handle->h_reserved && journal->j_barrier_count) {
399 read_unlock(&journal->j_state_lock);
400 wait_event(journal->j_wait_transaction_locked,
401 journal->j_barrier_count == 0);
405 if (!journal->j_running_transaction) {
406 read_unlock(&journal->j_state_lock);
407 if (!new_transaction)
408 goto alloc_transaction;
409 write_lock(&journal->j_state_lock);
410 if (!journal->j_running_transaction &&
411 (handle->h_reserved || !journal->j_barrier_count)) {
412 jbd2_get_transaction(journal, new_transaction);
413 new_transaction = NULL;
415 write_unlock(&journal->j_state_lock);
419 transaction = journal->j_running_transaction;
421 if (!handle->h_reserved) {
422 /* We may have dropped j_state_lock - restart in that case */
423 if (add_transaction_credits(journal, blocks, rsv_blocks)) {
425 * add_transaction_credits releases
426 * j_state_lock on a non-zero return
428 __release(&journal->j_state_lock);
433 * We have handle reserved so we are allowed to join T_LOCKED
434 * transaction and we don't have to check for transaction size
435 * and journal space. But we still have to wait while running
436 * transaction is being switched to a committing one as it
437 * won't wait for any handles anymore.
439 if (transaction->t_state == T_SWITCH) {
440 wait_transaction_switching(journal);
443 sub_reserved_credits(journal, blocks);
444 handle->h_reserved = 0;
447 /* OK, account for the buffers that this operation expects to
448 * use and add the handle to the running transaction.
450 update_t_max_wait(transaction, ts);
451 handle->h_transaction = transaction;
452 handle->h_requested_credits = blocks;
453 handle->h_revoke_credits_requested = handle->h_revoke_credits;
454 handle->h_start_jiffies = jiffies;
455 atomic_inc(&transaction->t_updates);
456 atomic_inc(&transaction->t_handle_count);
457 jbd2_debug(4, "Handle %p given %d credits (total %d, free %lu)\n",
459 atomic_read(&transaction->t_outstanding_credits),
460 jbd2_log_space_left(journal));
461 read_unlock(&journal->j_state_lock);
462 current->journal_info = handle;
464 rwsem_acquire_read(&journal->j_trans_commit_map, 0, 0, _THIS_IP_);
465 jbd2_journal_free_transaction(new_transaction);
467 * Ensure that no allocations done while the transaction is open are
468 * going to recurse back to the fs layer.
470 handle->saved_alloc_context = memalloc_nofs_save();
474 /* Allocate a new handle. This should probably be in a slab... */
475 static handle_t *new_handle(int nblocks)
477 handle_t *handle = jbd2_alloc_handle(GFP_NOFS);
480 handle->h_total_credits = nblocks;
486 handle_t *jbd2__journal_start(journal_t *journal, int nblocks, int rsv_blocks,
487 int revoke_records, gfp_t gfp_mask,
488 unsigned int type, unsigned int line_no)
490 handle_t *handle = journal_current_handle();
494 return ERR_PTR(-EROFS);
497 J_ASSERT(handle->h_transaction->t_journal == journal);
502 nblocks += DIV_ROUND_UP(revoke_records,
503 journal->j_revoke_records_per_block);
504 handle = new_handle(nblocks);
506 return ERR_PTR(-ENOMEM);
508 handle_t *rsv_handle;
510 rsv_handle = new_handle(rsv_blocks);
512 jbd2_free_handle(handle);
513 return ERR_PTR(-ENOMEM);
515 rsv_handle->h_reserved = 1;
516 rsv_handle->h_journal = journal;
517 handle->h_rsv_handle = rsv_handle;
519 handle->h_revoke_credits = revoke_records;
521 err = start_this_handle(journal, handle, gfp_mask);
523 if (handle->h_rsv_handle)
524 jbd2_free_handle(handle->h_rsv_handle);
525 jbd2_free_handle(handle);
528 handle->h_type = type;
529 handle->h_line_no = line_no;
530 trace_jbd2_handle_start(journal->j_fs_dev->bd_dev,
531 handle->h_transaction->t_tid, type,
536 EXPORT_SYMBOL(jbd2__journal_start);
540 * jbd2_journal_start() - Obtain a new handle.
541 * @journal: Journal to start transaction on.
542 * @nblocks: number of block buffer we might modify
544 * We make sure that the transaction can guarantee at least nblocks of
545 * modified buffers in the log. We block until the log can guarantee
546 * that much space. Additionally, if rsv_blocks > 0, we also create another
547 * handle with rsv_blocks reserved blocks in the journal. This handle is
548 * stored in h_rsv_handle. It is not attached to any particular transaction
549 * and thus doesn't block transaction commit. If the caller uses this reserved
550 * handle, it has to set h_rsv_handle to NULL as otherwise jbd2_journal_stop()
551 * on the parent handle will dispose the reserved one. Reserved handle has to
552 * be converted to a normal handle using jbd2_journal_start_reserved() before
555 * Return a pointer to a newly allocated handle, or an ERR_PTR() value
558 handle_t *jbd2_journal_start(journal_t *journal, int nblocks)
560 return jbd2__journal_start(journal, nblocks, 0, 0, GFP_NOFS, 0, 0);
562 EXPORT_SYMBOL(jbd2_journal_start);
564 static void __jbd2_journal_unreserve_handle(handle_t *handle, transaction_t *t)
566 journal_t *journal = handle->h_journal;
568 WARN_ON(!handle->h_reserved);
569 sub_reserved_credits(journal, handle->h_total_credits);
571 atomic_sub(handle->h_total_credits, &t->t_outstanding_credits);
574 void jbd2_journal_free_reserved(handle_t *handle)
576 journal_t *journal = handle->h_journal;
578 /* Get j_state_lock to pin running transaction if it exists */
579 read_lock(&journal->j_state_lock);
580 __jbd2_journal_unreserve_handle(handle, journal->j_running_transaction);
581 read_unlock(&journal->j_state_lock);
582 jbd2_free_handle(handle);
584 EXPORT_SYMBOL(jbd2_journal_free_reserved);
587 * jbd2_journal_start_reserved() - start reserved handle
588 * @handle: handle to start
589 * @type: for handle statistics
590 * @line_no: for handle statistics
592 * Start handle that has been previously reserved with jbd2_journal_reserve().
593 * This attaches @handle to the running transaction (or creates one if there's
594 * not transaction running). Unlike jbd2_journal_start() this function cannot
595 * block on journal commit, checkpointing, or similar stuff. It can block on
596 * memory allocation or frozen journal though.
598 * Return 0 on success, non-zero on error - handle is freed in that case.
600 int jbd2_journal_start_reserved(handle_t *handle, unsigned int type,
601 unsigned int line_no)
603 journal_t *journal = handle->h_journal;
606 if (WARN_ON(!handle->h_reserved)) {
607 /* Someone passed in normal handle? Just stop it. */
608 jbd2_journal_stop(handle);
612 * Usefulness of mixing of reserved and unreserved handles is
613 * questionable. So far nobody seems to need it so just error out.
615 if (WARN_ON(current->journal_info)) {
616 jbd2_journal_free_reserved(handle);
620 handle->h_journal = NULL;
622 * GFP_NOFS is here because callers are likely from writeback or
623 * similarly constrained call sites
625 ret = start_this_handle(journal, handle, GFP_NOFS);
627 handle->h_journal = journal;
628 jbd2_journal_free_reserved(handle);
631 handle->h_type = type;
632 handle->h_line_no = line_no;
633 trace_jbd2_handle_start(journal->j_fs_dev->bd_dev,
634 handle->h_transaction->t_tid, type,
635 line_no, handle->h_total_credits);
638 EXPORT_SYMBOL(jbd2_journal_start_reserved);
641 * jbd2_journal_extend() - extend buffer credits.
642 * @handle: handle to 'extend'
643 * @nblocks: nr blocks to try to extend by.
644 * @revoke_records: number of revoke records to try to extend by.
646 * Some transactions, such as large extends and truncates, can be done
647 * atomically all at once or in several stages. The operation requests
648 * a credit for a number of buffer modifications in advance, but can
649 * extend its credit if it needs more.
651 * jbd2_journal_extend tries to give the running handle more buffer credits.
652 * It does not guarantee that allocation - this is a best-effort only.
653 * The calling process MUST be able to deal cleanly with a failure to
656 * Return 0 on success, non-zero on failure.
658 * return code < 0 implies an error
659 * return code > 0 implies normal transaction-full status.
661 int jbd2_journal_extend(handle_t *handle, int nblocks, int revoke_records)
663 transaction_t *transaction = handle->h_transaction;
668 if (is_handle_aborted(handle))
670 journal = transaction->t_journal;
674 read_lock(&journal->j_state_lock);
676 /* Don't extend a locked-down transaction! */
677 if (transaction->t_state != T_RUNNING) {
678 jbd2_debug(3, "denied handle %p %d blocks: "
679 "transaction not running\n", handle, nblocks);
683 nblocks += DIV_ROUND_UP(
684 handle->h_revoke_credits_requested + revoke_records,
685 journal->j_revoke_records_per_block) -
687 handle->h_revoke_credits_requested,
688 journal->j_revoke_records_per_block);
689 wanted = atomic_add_return(nblocks,
690 &transaction->t_outstanding_credits);
692 if (wanted > journal->j_max_transaction_buffers) {
693 jbd2_debug(3, "denied handle %p %d blocks: "
694 "transaction too large\n", handle, nblocks);
695 atomic_sub(nblocks, &transaction->t_outstanding_credits);
699 trace_jbd2_handle_extend(journal->j_fs_dev->bd_dev,
701 handle->h_type, handle->h_line_no,
702 handle->h_total_credits,
705 handle->h_total_credits += nblocks;
706 handle->h_requested_credits += nblocks;
707 handle->h_revoke_credits += revoke_records;
708 handle->h_revoke_credits_requested += revoke_records;
711 jbd2_debug(3, "extended handle %p by %d\n", handle, nblocks);
713 read_unlock(&journal->j_state_lock);
717 static void stop_this_handle(handle_t *handle)
719 transaction_t *transaction = handle->h_transaction;
720 journal_t *journal = transaction->t_journal;
723 J_ASSERT(journal_current_handle() == handle);
724 J_ASSERT(atomic_read(&transaction->t_updates) > 0);
725 current->journal_info = NULL;
727 * Subtract necessary revoke descriptor blocks from handle credits. We
728 * take care to account only for revoke descriptor blocks the
729 * transaction will really need as large sequences of transactions with
730 * small numbers of revokes are relatively common.
732 revokes = handle->h_revoke_credits_requested - handle->h_revoke_credits;
734 int t_revokes, revoke_descriptors;
735 int rr_per_blk = journal->j_revoke_records_per_block;
737 WARN_ON_ONCE(DIV_ROUND_UP(revokes, rr_per_blk)
738 > handle->h_total_credits);
739 t_revokes = atomic_add_return(revokes,
740 &transaction->t_outstanding_revokes);
742 DIV_ROUND_UP(t_revokes, rr_per_blk) -
743 DIV_ROUND_UP(t_revokes - revokes, rr_per_blk);
744 handle->h_total_credits -= revoke_descriptors;
746 atomic_sub(handle->h_total_credits,
747 &transaction->t_outstanding_credits);
748 if (handle->h_rsv_handle)
749 __jbd2_journal_unreserve_handle(handle->h_rsv_handle,
751 if (atomic_dec_and_test(&transaction->t_updates))
752 wake_up(&journal->j_wait_updates);
754 rwsem_release(&journal->j_trans_commit_map, _THIS_IP_);
756 * Scope of the GFP_NOFS context is over here and so we can restore the
757 * original alloc context.
759 memalloc_nofs_restore(handle->saved_alloc_context);
763 * jbd2__journal_restart() - restart a handle .
764 * @handle: handle to restart
765 * @nblocks: nr credits requested
766 * @revoke_records: number of revoke record credits requested
767 * @gfp_mask: memory allocation flags (for start_this_handle)
769 * Restart a handle for a multi-transaction filesystem
772 * If the jbd2_journal_extend() call above fails to grant new buffer credits
773 * to a running handle, a call to jbd2_journal_restart will commit the
774 * handle's transaction so far and reattach the handle to a new
775 * transaction capable of guaranteeing the requested number of
776 * credits. We preserve reserved handle if there's any attached to the
779 int jbd2__journal_restart(handle_t *handle, int nblocks, int revoke_records,
782 transaction_t *transaction = handle->h_transaction;
788 /* If we've had an abort of any type, don't even think about
789 * actually doing the restart! */
790 if (is_handle_aborted(handle))
792 journal = transaction->t_journal;
793 tid = transaction->t_tid;
796 * First unlink the handle from its current transaction, and start the
799 jbd2_debug(2, "restarting handle %p\n", handle);
800 stop_this_handle(handle);
801 handle->h_transaction = NULL;
804 * TODO: If we use READ_ONCE / WRITE_ONCE for j_commit_request we can
805 * get rid of pointless j_state_lock traffic like this.
807 read_lock(&journal->j_state_lock);
808 need_to_start = !tid_geq(journal->j_commit_request, tid);
809 read_unlock(&journal->j_state_lock);
811 jbd2_log_start_commit(journal, tid);
812 handle->h_total_credits = nblocks +
813 DIV_ROUND_UP(revoke_records,
814 journal->j_revoke_records_per_block);
815 handle->h_revoke_credits = revoke_records;
816 ret = start_this_handle(journal, handle, gfp_mask);
817 trace_jbd2_handle_restart(journal->j_fs_dev->bd_dev,
818 ret ? 0 : handle->h_transaction->t_tid,
819 handle->h_type, handle->h_line_no,
820 handle->h_total_credits);
823 EXPORT_SYMBOL(jbd2__journal_restart);
826 int jbd2_journal_restart(handle_t *handle, int nblocks)
828 return jbd2__journal_restart(handle, nblocks, 0, GFP_NOFS);
830 EXPORT_SYMBOL(jbd2_journal_restart);
833 * Waits for any outstanding t_updates to finish.
834 * This is called with write j_state_lock held.
836 void jbd2_journal_wait_updates(journal_t *journal)
842 * Note that the running transaction can get freed under us if
843 * this transaction is getting committed in
844 * jbd2_journal_commit_transaction() ->
845 * jbd2_journal_free_transaction(). This can only happen when we
846 * release j_state_lock -> schedule() -> acquire j_state_lock.
847 * Hence we should everytime retrieve new j_running_transaction
848 * value (after j_state_lock release acquire cycle), else it may
849 * lead to use-after-free of old freed transaction.
851 transaction_t *transaction = journal->j_running_transaction;
856 prepare_to_wait(&journal->j_wait_updates, &wait,
857 TASK_UNINTERRUPTIBLE);
858 if (!atomic_read(&transaction->t_updates)) {
859 finish_wait(&journal->j_wait_updates, &wait);
862 write_unlock(&journal->j_state_lock);
864 finish_wait(&journal->j_wait_updates, &wait);
865 write_lock(&journal->j_state_lock);
870 * jbd2_journal_lock_updates () - establish a transaction barrier.
871 * @journal: Journal to establish a barrier on.
873 * This locks out any further updates from being started, and blocks
874 * until all existing updates have completed, returning only once the
875 * journal is in a quiescent state with no updates running.
877 * The journal lock should not be held on entry.
879 void jbd2_journal_lock_updates(journal_t *journal)
881 jbd2_might_wait_for_commit(journal);
883 write_lock(&journal->j_state_lock);
884 ++journal->j_barrier_count;
886 /* Wait until there are no reserved handles */
887 if (atomic_read(&journal->j_reserved_credits)) {
888 write_unlock(&journal->j_state_lock);
889 wait_event(journal->j_wait_reserved,
890 atomic_read(&journal->j_reserved_credits) == 0);
891 write_lock(&journal->j_state_lock);
894 /* Wait until there are no running t_updates */
895 jbd2_journal_wait_updates(journal);
897 write_unlock(&journal->j_state_lock);
900 * We have now established a barrier against other normal updates, but
901 * we also need to barrier against other jbd2_journal_lock_updates() calls
902 * to make sure that we serialise special journal-locked operations
905 mutex_lock(&journal->j_barrier);
909 * jbd2_journal_unlock_updates () - release barrier
910 * @journal: Journal to release the barrier on.
912 * Release a transaction barrier obtained with jbd2_journal_lock_updates().
914 * Should be called without the journal lock held.
916 void jbd2_journal_unlock_updates (journal_t *journal)
918 J_ASSERT(journal->j_barrier_count != 0);
920 mutex_unlock(&journal->j_barrier);
921 write_lock(&journal->j_state_lock);
922 --journal->j_barrier_count;
923 write_unlock(&journal->j_state_lock);
924 wake_up_all(&journal->j_wait_transaction_locked);
927 static void warn_dirty_buffer(struct buffer_head *bh)
930 "JBD2: Spotted dirty metadata buffer (dev = %pg, blocknr = %llu). "
931 "There's a risk of filesystem corruption in case of system "
933 bh->b_bdev, (unsigned long long)bh->b_blocknr);
936 /* Call t_frozen trigger and copy buffer data into jh->b_frozen_data. */
937 static void jbd2_freeze_jh_data(struct journal_head *jh)
942 struct buffer_head *bh = jh2bh(jh);
944 J_EXPECT_JH(jh, buffer_uptodate(bh), "Possible IO failure.\n");
946 offset = offset_in_page(bh->b_data);
947 source = kmap_atomic(page);
948 /* Fire data frozen trigger just before we copy the data */
949 jbd2_buffer_frozen_trigger(jh, source + offset, jh->b_triggers);
950 memcpy(jh->b_frozen_data, source + offset, bh->b_size);
951 kunmap_atomic(source);
954 * Now that the frozen data is saved off, we need to store any matching
957 jh->b_frozen_triggers = jh->b_triggers;
961 * If the buffer is already part of the current transaction, then there
962 * is nothing we need to do. If it is already part of a prior
963 * transaction which we are still committing to disk, then we need to
964 * make sure that we do not overwrite the old copy: we do copy-out to
965 * preserve the copy going to disk. We also account the buffer against
966 * the handle's metadata buffer credits (unless the buffer is already
967 * part of the transaction, that is).
971 do_get_write_access(handle_t *handle, struct journal_head *jh,
974 struct buffer_head *bh;
975 transaction_t *transaction = handle->h_transaction;
978 char *frozen_buffer = NULL;
979 unsigned long start_lock, time_lock;
981 journal = transaction->t_journal;
983 jbd2_debug(5, "journal_head %p, force_copy %d\n", jh, force_copy);
985 JBUFFER_TRACE(jh, "entry");
989 /* @@@ Need to check for errors here at some point. */
991 start_lock = jiffies;
993 spin_lock(&jh->b_state_lock);
995 /* If it takes too long to lock the buffer, trace it */
996 time_lock = jbd2_time_diff(start_lock, jiffies);
997 if (time_lock > HZ/10)
998 trace_jbd2_lock_buffer_stall(bh->b_bdev->bd_dev,
999 jiffies_to_msecs(time_lock));
1001 /* We now hold the buffer lock so it is safe to query the buffer
1002 * state. Is the buffer dirty?
1004 * If so, there are two possibilities. The buffer may be
1005 * non-journaled, and undergoing a quite legitimate writeback.
1006 * Otherwise, it is journaled, and we don't expect dirty buffers
1007 * in that state (the buffers should be marked JBD_Dirty
1008 * instead.) So either the IO is being done under our own
1009 * control and this is a bug, or it's a third party IO such as
1010 * dump(8) (which may leave the buffer scheduled for read ---
1011 * ie. locked but not dirty) or tune2fs (which may actually have
1012 * the buffer dirtied, ugh.) */
1014 if (buffer_dirty(bh) && jh->b_transaction) {
1015 warn_dirty_buffer(bh);
1017 * We need to clean the dirty flag and we must do it under the
1018 * buffer lock to be sure we don't race with running write-out.
1020 JBUFFER_TRACE(jh, "Journalling dirty buffer");
1021 clear_buffer_dirty(bh);
1023 * The buffer is going to be added to BJ_Reserved list now and
1024 * nothing guarantees jbd2_journal_dirty_metadata() will be
1025 * ever called for it. So we need to set jbddirty bit here to
1026 * make sure the buffer is dirtied and written out when the
1027 * journaling machinery is done with it.
1029 set_buffer_jbddirty(bh);
1033 if (is_handle_aborted(handle)) {
1034 spin_unlock(&jh->b_state_lock);
1041 * The buffer is already part of this transaction if b_transaction or
1042 * b_next_transaction points to it
1044 if (jh->b_transaction == transaction ||
1045 jh->b_next_transaction == transaction) {
1051 * this is the first time this transaction is touching this buffer,
1052 * reset the modified flag
1057 * If the buffer is not journaled right now, we need to make sure it
1058 * doesn't get written to disk before the caller actually commits the
1061 if (!jh->b_transaction) {
1062 JBUFFER_TRACE(jh, "no transaction");
1063 J_ASSERT_JH(jh, !jh->b_next_transaction);
1064 JBUFFER_TRACE(jh, "file as BJ_Reserved");
1066 * Make sure all stores to jh (b_modified, b_frozen_data) are
1067 * visible before attaching it to the running transaction.
1068 * Paired with barrier in jbd2_write_access_granted()
1071 spin_lock(&journal->j_list_lock);
1072 if (test_clear_buffer_dirty(bh)) {
1074 * Execute buffer dirty clearing and jh->b_transaction
1075 * assignment under journal->j_list_lock locked to
1076 * prevent bh being removed from checkpoint list if
1077 * the buffer is in an intermediate state (not dirty
1078 * and jh->b_transaction is NULL).
1080 JBUFFER_TRACE(jh, "Journalling dirty buffer");
1081 set_buffer_jbddirty(bh);
1083 __jbd2_journal_file_buffer(jh, transaction, BJ_Reserved);
1084 spin_unlock(&journal->j_list_lock);
1091 * If there is already a copy-out version of this buffer, then we don't
1092 * need to make another one
1094 if (jh->b_frozen_data) {
1095 JBUFFER_TRACE(jh, "has frozen data");
1096 J_ASSERT_JH(jh, jh->b_next_transaction == NULL);
1100 JBUFFER_TRACE(jh, "owned by older transaction");
1101 J_ASSERT_JH(jh, jh->b_next_transaction == NULL);
1102 J_ASSERT_JH(jh, jh->b_transaction == journal->j_committing_transaction);
1105 * There is one case we have to be very careful about. If the
1106 * committing transaction is currently writing this buffer out to disk
1107 * and has NOT made a copy-out, then we cannot modify the buffer
1108 * contents at all right now. The essence of copy-out is that it is
1109 * the extra copy, not the primary copy, which gets journaled. If the
1110 * primary copy is already going to disk then we cannot do copy-out
1113 if (buffer_shadow(bh)) {
1114 JBUFFER_TRACE(jh, "on shadow: sleep");
1115 spin_unlock(&jh->b_state_lock);
1116 wait_on_bit_io(&bh->b_state, BH_Shadow, TASK_UNINTERRUPTIBLE);
1121 * Only do the copy if the currently-owning transaction still needs it.
1122 * If buffer isn't on BJ_Metadata list, the committing transaction is
1123 * past that stage (here we use the fact that BH_Shadow is set under
1124 * bh_state lock together with refiling to BJ_Shadow list and at this
1125 * point we know the buffer doesn't have BH_Shadow set).
1127 * Subtle point, though: if this is a get_undo_access, then we will be
1128 * relying on the frozen_data to contain the new value of the
1129 * committed_data record after the transaction, so we HAVE to force the
1130 * frozen_data copy in that case.
1132 if (jh->b_jlist == BJ_Metadata || force_copy) {
1133 JBUFFER_TRACE(jh, "generate frozen data");
1134 if (!frozen_buffer) {
1135 JBUFFER_TRACE(jh, "allocate memory for buffer");
1136 spin_unlock(&jh->b_state_lock);
1137 frozen_buffer = jbd2_alloc(jh2bh(jh)->b_size,
1138 GFP_NOFS | __GFP_NOFAIL);
1141 jh->b_frozen_data = frozen_buffer;
1142 frozen_buffer = NULL;
1143 jbd2_freeze_jh_data(jh);
1147 * Make sure all stores to jh (b_modified, b_frozen_data) are visible
1148 * before attaching it to the running transaction. Paired with barrier
1149 * in jbd2_write_access_granted()
1152 jh->b_next_transaction = transaction;
1155 spin_unlock(&jh->b_state_lock);
1158 * If we are about to journal a buffer, then any revoke pending on it is
1161 jbd2_journal_cancel_revoke(handle, jh);
1164 if (unlikely(frozen_buffer)) /* It's usually NULL */
1165 jbd2_free(frozen_buffer, bh->b_size);
1167 JBUFFER_TRACE(jh, "exit");
1171 /* Fast check whether buffer is already attached to the required transaction */
1172 static bool jbd2_write_access_granted(handle_t *handle, struct buffer_head *bh,
1175 struct journal_head *jh;
1178 /* Dirty buffers require special handling... */
1179 if (buffer_dirty(bh))
1183 * RCU protects us from dereferencing freed pages. So the checks we do
1184 * are guaranteed not to oops. However the jh slab object can get freed
1185 * & reallocated while we work with it. So we have to be careful. When
1186 * we see jh attached to the running transaction, we know it must stay
1187 * so until the transaction is committed. Thus jh won't be freed and
1188 * will be attached to the same bh while we run. However it can
1189 * happen jh gets freed, reallocated, and attached to the transaction
1190 * just after we get pointer to it from bh. So we have to be careful
1191 * and recheck jh still belongs to our bh before we return success.
1194 if (!buffer_jbd(bh))
1196 /* This should be bh2jh() but that doesn't work with inline functions */
1197 jh = READ_ONCE(bh->b_private);
1200 /* For undo access buffer must have data copied */
1201 if (undo && !jh->b_committed_data)
1203 if (READ_ONCE(jh->b_transaction) != handle->h_transaction &&
1204 READ_ONCE(jh->b_next_transaction) != handle->h_transaction)
1207 * There are two reasons for the barrier here:
1208 * 1) Make sure to fetch b_bh after we did previous checks so that we
1209 * detect when jh went through free, realloc, attach to transaction
1210 * while we were checking. Paired with implicit barrier in that path.
1211 * 2) So that access to bh done after jbd2_write_access_granted()
1212 * doesn't get reordered and see inconsistent state of concurrent
1213 * do_get_write_access().
1216 if (unlikely(jh->b_bh != bh))
1225 * jbd2_journal_get_write_access() - notify intent to modify a buffer
1226 * for metadata (not data) update.
1227 * @handle: transaction to add buffer modifications to
1228 * @bh: bh to be used for metadata writes
1230 * Returns: error code or 0 on success.
1232 * In full data journalling mode the buffer may be of type BJ_AsyncData,
1233 * because we're ``write()ing`` a buffer which is also part of a shared mapping.
1236 int jbd2_journal_get_write_access(handle_t *handle, struct buffer_head *bh)
1238 struct journal_head *jh;
1241 if (is_handle_aborted(handle))
1244 if (jbd2_write_access_granted(handle, bh, false))
1247 jh = jbd2_journal_add_journal_head(bh);
1248 /* We do not want to get caught playing with fields which the
1249 * log thread also manipulates. Make sure that the buffer
1250 * completes any outstanding IO before proceeding. */
1251 rc = do_get_write_access(handle, jh, 0);
1252 jbd2_journal_put_journal_head(jh);
1258 * When the user wants to journal a newly created buffer_head
1259 * (ie. getblk() returned a new buffer and we are going to populate it
1260 * manually rather than reading off disk), then we need to keep the
1261 * buffer_head locked until it has been completely filled with new
1262 * data. In this case, we should be able to make the assertion that
1263 * the bh is not already part of an existing transaction.
1265 * The buffer should already be locked by the caller by this point.
1266 * There is no lock ranking violation: it was a newly created,
1267 * unlocked buffer beforehand. */
1270 * jbd2_journal_get_create_access () - notify intent to use newly created bh
1271 * @handle: transaction to new buffer to
1274 * Call this if you create a new bh.
1276 int jbd2_journal_get_create_access(handle_t *handle, struct buffer_head *bh)
1278 transaction_t *transaction = handle->h_transaction;
1280 struct journal_head *jh = jbd2_journal_add_journal_head(bh);
1283 jbd2_debug(5, "journal_head %p\n", jh);
1285 if (is_handle_aborted(handle))
1287 journal = transaction->t_journal;
1290 JBUFFER_TRACE(jh, "entry");
1292 * The buffer may already belong to this transaction due to pre-zeroing
1293 * in the filesystem's new_block code. It may also be on the previous,
1294 * committing transaction's lists, but it HAS to be in Forget state in
1295 * that case: the transaction must have deleted the buffer for it to be
1298 spin_lock(&jh->b_state_lock);
1299 J_ASSERT_JH(jh, (jh->b_transaction == transaction ||
1300 jh->b_transaction == NULL ||
1301 (jh->b_transaction == journal->j_committing_transaction &&
1302 jh->b_jlist == BJ_Forget)));
1304 J_ASSERT_JH(jh, jh->b_next_transaction == NULL);
1305 J_ASSERT_JH(jh, buffer_locked(jh2bh(jh)));
1307 if (jh->b_transaction == NULL) {
1309 * Previous jbd2_journal_forget() could have left the buffer
1310 * with jbddirty bit set because it was being committed. When
1311 * the commit finished, we've filed the buffer for
1312 * checkpointing and marked it dirty. Now we are reallocating
1313 * the buffer so the transaction freeing it must have
1314 * committed and so it's safe to clear the dirty bit.
1316 clear_buffer_dirty(jh2bh(jh));
1317 /* first access by this transaction */
1320 JBUFFER_TRACE(jh, "file as BJ_Reserved");
1321 spin_lock(&journal->j_list_lock);
1322 __jbd2_journal_file_buffer(jh, transaction, BJ_Reserved);
1323 spin_unlock(&journal->j_list_lock);
1324 } else if (jh->b_transaction == journal->j_committing_transaction) {
1325 /* first access by this transaction */
1328 JBUFFER_TRACE(jh, "set next transaction");
1329 spin_lock(&journal->j_list_lock);
1330 jh->b_next_transaction = transaction;
1331 spin_unlock(&journal->j_list_lock);
1333 spin_unlock(&jh->b_state_lock);
1336 * akpm: I added this. ext3_alloc_branch can pick up new indirect
1337 * blocks which contain freed but then revoked metadata. We need
1338 * to cancel the revoke in case we end up freeing it yet again
1339 * and the reallocating as data - this would cause a second revoke,
1340 * which hits an assertion error.
1342 JBUFFER_TRACE(jh, "cancelling revoke");
1343 jbd2_journal_cancel_revoke(handle, jh);
1345 jbd2_journal_put_journal_head(jh);
1350 * jbd2_journal_get_undo_access() - Notify intent to modify metadata with
1351 * non-rewindable consequences
1352 * @handle: transaction
1353 * @bh: buffer to undo
1355 * Sometimes there is a need to distinguish between metadata which has
1356 * been committed to disk and that which has not. The ext3fs code uses
1357 * this for freeing and allocating space, we have to make sure that we
1358 * do not reuse freed space until the deallocation has been committed,
1359 * since if we overwrote that space we would make the delete
1360 * un-rewindable in case of a crash.
1362 * To deal with that, jbd2_journal_get_undo_access requests write access to a
1363 * buffer for parts of non-rewindable operations such as delete
1364 * operations on the bitmaps. The journaling code must keep a copy of
1365 * the buffer's contents prior to the undo_access call until such time
1366 * as we know that the buffer has definitely been committed to disk.
1368 * We never need to know which transaction the committed data is part
1369 * of, buffers touched here are guaranteed to be dirtied later and so
1370 * will be committed to a new transaction in due course, at which point
1371 * we can discard the old committed data pointer.
1373 * Returns error number or 0 on success.
1375 int jbd2_journal_get_undo_access(handle_t *handle, struct buffer_head *bh)
1378 struct journal_head *jh;
1379 char *committed_data = NULL;
1381 if (is_handle_aborted(handle))
1384 if (jbd2_write_access_granted(handle, bh, true))
1387 jh = jbd2_journal_add_journal_head(bh);
1388 JBUFFER_TRACE(jh, "entry");
1391 * Do this first --- it can drop the journal lock, so we want to
1392 * make sure that obtaining the committed_data is done
1393 * atomically wrt. completion of any outstanding commits.
1395 err = do_get_write_access(handle, jh, 1);
1400 if (!jh->b_committed_data)
1401 committed_data = jbd2_alloc(jh2bh(jh)->b_size,
1402 GFP_NOFS|__GFP_NOFAIL);
1404 spin_lock(&jh->b_state_lock);
1405 if (!jh->b_committed_data) {
1406 /* Copy out the current buffer contents into the
1407 * preserved, committed copy. */
1408 JBUFFER_TRACE(jh, "generate b_committed data");
1409 if (!committed_data) {
1410 spin_unlock(&jh->b_state_lock);
1414 jh->b_committed_data = committed_data;
1415 committed_data = NULL;
1416 memcpy(jh->b_committed_data, bh->b_data, bh->b_size);
1418 spin_unlock(&jh->b_state_lock);
1420 jbd2_journal_put_journal_head(jh);
1421 if (unlikely(committed_data))
1422 jbd2_free(committed_data, bh->b_size);
1427 * jbd2_journal_set_triggers() - Add triggers for commit writeout
1428 * @bh: buffer to trigger on
1429 * @type: struct jbd2_buffer_trigger_type containing the trigger(s).
1431 * Set any triggers on this journal_head. This is always safe, because
1432 * triggers for a committing buffer will be saved off, and triggers for
1433 * a running transaction will match the buffer in that transaction.
1435 * Call with NULL to clear the triggers.
1437 void jbd2_journal_set_triggers(struct buffer_head *bh,
1438 struct jbd2_buffer_trigger_type *type)
1440 struct journal_head *jh = jbd2_journal_grab_journal_head(bh);
1442 if (WARN_ON_ONCE(!jh))
1444 jh->b_triggers = type;
1445 jbd2_journal_put_journal_head(jh);
1448 void jbd2_buffer_frozen_trigger(struct journal_head *jh, void *mapped_data,
1449 struct jbd2_buffer_trigger_type *triggers)
1451 struct buffer_head *bh = jh2bh(jh);
1453 if (!triggers || !triggers->t_frozen)
1456 triggers->t_frozen(triggers, bh, mapped_data, bh->b_size);
1459 void jbd2_buffer_abort_trigger(struct journal_head *jh,
1460 struct jbd2_buffer_trigger_type *triggers)
1462 if (!triggers || !triggers->t_abort)
1465 triggers->t_abort(triggers, jh2bh(jh));
1469 * jbd2_journal_dirty_metadata() - mark a buffer as containing dirty metadata
1470 * @handle: transaction to add buffer to.
1471 * @bh: buffer to mark
1473 * mark dirty metadata which needs to be journaled as part of the current
1476 * The buffer must have previously had jbd2_journal_get_write_access()
1477 * called so that it has a valid journal_head attached to the buffer
1480 * The buffer is placed on the transaction's metadata list and is marked
1481 * as belonging to the transaction.
1483 * Returns error number or 0 on success.
1485 * Special care needs to be taken if the buffer already belongs to the
1486 * current committing transaction (in which case we should have frozen
1487 * data present for that commit). In that case, we don't relink the
1488 * buffer: that only gets done when the old transaction finally
1489 * completes its commit.
1491 int jbd2_journal_dirty_metadata(handle_t *handle, struct buffer_head *bh)
1493 transaction_t *transaction = handle->h_transaction;
1495 struct journal_head *jh;
1498 if (!buffer_jbd(bh))
1502 * We don't grab jh reference here since the buffer must be part
1503 * of the running transaction.
1506 jbd2_debug(5, "journal_head %p\n", jh);
1507 JBUFFER_TRACE(jh, "entry");
1510 * This and the following assertions are unreliable since we may see jh
1511 * in inconsistent state unless we grab bh_state lock. But this is
1512 * crucial to catch bugs so let's do a reliable check until the
1513 * lockless handling is fully proven.
1515 if (data_race(jh->b_transaction != transaction &&
1516 jh->b_next_transaction != transaction)) {
1517 spin_lock(&jh->b_state_lock);
1518 J_ASSERT_JH(jh, jh->b_transaction == transaction ||
1519 jh->b_next_transaction == transaction);
1520 spin_unlock(&jh->b_state_lock);
1522 if (jh->b_modified == 1) {
1523 /* If it's in our transaction it must be in BJ_Metadata list. */
1524 if (data_race(jh->b_transaction == transaction &&
1525 jh->b_jlist != BJ_Metadata)) {
1526 spin_lock(&jh->b_state_lock);
1527 if (jh->b_transaction == transaction &&
1528 jh->b_jlist != BJ_Metadata)
1529 pr_err("JBD2: assertion failure: h_type=%u "
1530 "h_line_no=%u block_no=%llu jlist=%u\n",
1531 handle->h_type, handle->h_line_no,
1532 (unsigned long long) bh->b_blocknr,
1534 J_ASSERT_JH(jh, jh->b_transaction != transaction ||
1535 jh->b_jlist == BJ_Metadata);
1536 spin_unlock(&jh->b_state_lock);
1541 journal = transaction->t_journal;
1542 spin_lock(&jh->b_state_lock);
1544 if (is_handle_aborted(handle)) {
1546 * Check journal aborting with @jh->b_state_lock locked,
1547 * since 'jh->b_transaction' could be replaced with
1548 * 'jh->b_next_transaction' during old transaction
1549 * committing if journal aborted, which may fail
1550 * assertion on 'jh->b_frozen_data == NULL'.
1556 if (jh->b_modified == 0) {
1558 * This buffer's got modified and becoming part
1559 * of the transaction. This needs to be done
1560 * once a transaction -bzzz
1562 if (WARN_ON_ONCE(jbd2_handle_buffer_credits(handle) <= 0)) {
1567 handle->h_total_credits--;
1571 * fastpath, to avoid expensive locking. If this buffer is already
1572 * on the running transaction's metadata list there is nothing to do.
1573 * Nobody can take it off again because there is a handle open.
1574 * I _think_ we're OK here with SMP barriers - a mistaken decision will
1575 * result in this test being false, so we go in and take the locks.
1577 if (jh->b_transaction == transaction && jh->b_jlist == BJ_Metadata) {
1578 JBUFFER_TRACE(jh, "fastpath");
1579 if (unlikely(jh->b_transaction !=
1580 journal->j_running_transaction)) {
1581 printk(KERN_ERR "JBD2: %s: "
1582 "jh->b_transaction (%llu, %p, %u) != "
1583 "journal->j_running_transaction (%p, %u)\n",
1585 (unsigned long long) bh->b_blocknr,
1587 jh->b_transaction ? jh->b_transaction->t_tid : 0,
1588 journal->j_running_transaction,
1589 journal->j_running_transaction ?
1590 journal->j_running_transaction->t_tid : 0);
1596 set_buffer_jbddirty(bh);
1599 * Metadata already on the current transaction list doesn't
1600 * need to be filed. Metadata on another transaction's list must
1601 * be committing, and will be refiled once the commit completes:
1602 * leave it alone for now.
1604 if (jh->b_transaction != transaction) {
1605 JBUFFER_TRACE(jh, "already on other transaction");
1606 if (unlikely(((jh->b_transaction !=
1607 journal->j_committing_transaction)) ||
1608 (jh->b_next_transaction != transaction))) {
1609 printk(KERN_ERR "jbd2_journal_dirty_metadata: %s: "
1610 "bad jh for block %llu: "
1611 "transaction (%p, %u), "
1612 "jh->b_transaction (%p, %u), "
1613 "jh->b_next_transaction (%p, %u), jlist %u\n",
1615 (unsigned long long) bh->b_blocknr,
1616 transaction, transaction->t_tid,
1619 jh->b_transaction->t_tid : 0,
1620 jh->b_next_transaction,
1621 jh->b_next_transaction ?
1622 jh->b_next_transaction->t_tid : 0,
1627 /* And this case is illegal: we can't reuse another
1628 * transaction's data buffer, ever. */
1632 /* That test should have eliminated the following case: */
1633 J_ASSERT_JH(jh, jh->b_frozen_data == NULL);
1635 JBUFFER_TRACE(jh, "file as BJ_Metadata");
1636 spin_lock(&journal->j_list_lock);
1637 __jbd2_journal_file_buffer(jh, transaction, BJ_Metadata);
1638 spin_unlock(&journal->j_list_lock);
1640 spin_unlock(&jh->b_state_lock);
1642 JBUFFER_TRACE(jh, "exit");
1647 * jbd2_journal_forget() - bforget() for potentially-journaled buffers.
1648 * @handle: transaction handle
1649 * @bh: bh to 'forget'
1651 * We can only do the bforget if there are no commits pending against the
1652 * buffer. If the buffer is dirty in the current running transaction we
1653 * can safely unlink it.
1655 * bh may not be a journalled buffer at all - it may be a non-JBD
1656 * buffer which came off the hashtable. Check for this.
1658 * Decrements bh->b_count by one.
1660 * Allow this call even if the handle has aborted --- it may be part of
1661 * the caller's cleanup after an abort.
1663 int jbd2_journal_forget(handle_t *handle, struct buffer_head *bh)
1665 transaction_t *transaction = handle->h_transaction;
1667 struct journal_head *jh;
1668 int drop_reserve = 0;
1670 int was_modified = 0;
1672 if (is_handle_aborted(handle))
1674 journal = transaction->t_journal;
1676 BUFFER_TRACE(bh, "entry");
1678 jh = jbd2_journal_grab_journal_head(bh);
1684 spin_lock(&jh->b_state_lock);
1686 /* Critical error: attempting to delete a bitmap buffer, maybe?
1687 * Don't do any jbd operations, and return an error. */
1688 if (!J_EXPECT_JH(jh, !jh->b_committed_data,
1689 "inconsistent data on disk")) {
1694 /* keep track of whether or not this transaction modified us */
1695 was_modified = jh->b_modified;
1698 * The buffer's going from the transaction, we must drop
1699 * all references -bzzz
1703 if (jh->b_transaction == transaction) {
1704 J_ASSERT_JH(jh, !jh->b_frozen_data);
1706 /* If we are forgetting a buffer which is already part
1707 * of this transaction, then we can just drop it from
1708 * the transaction immediately. */
1709 clear_buffer_dirty(bh);
1710 clear_buffer_jbddirty(bh);
1712 JBUFFER_TRACE(jh, "belongs to current transaction: unfile");
1715 * we only want to drop a reference if this transaction
1716 * modified the buffer
1722 * We are no longer going to journal this buffer.
1723 * However, the commit of this transaction is still
1724 * important to the buffer: the delete that we are now
1725 * processing might obsolete an old log entry, so by
1726 * committing, we can satisfy the buffer's checkpoint.
1728 * So, if we have a checkpoint on the buffer, we should
1729 * now refile the buffer on our BJ_Forget list so that
1730 * we know to remove the checkpoint after we commit.
1733 spin_lock(&journal->j_list_lock);
1734 if (jh->b_cp_transaction) {
1735 __jbd2_journal_temp_unlink_buffer(jh);
1736 __jbd2_journal_file_buffer(jh, transaction, BJ_Forget);
1738 __jbd2_journal_unfile_buffer(jh);
1739 jbd2_journal_put_journal_head(jh);
1741 spin_unlock(&journal->j_list_lock);
1742 } else if (jh->b_transaction) {
1743 J_ASSERT_JH(jh, (jh->b_transaction ==
1744 journal->j_committing_transaction));
1745 /* However, if the buffer is still owned by a prior
1746 * (committing) transaction, we can't drop it yet... */
1747 JBUFFER_TRACE(jh, "belongs to older transaction");
1748 /* ... but we CAN drop it from the new transaction through
1749 * marking the buffer as freed and set j_next_transaction to
1750 * the new transaction, so that not only the commit code
1751 * knows it should clear dirty bits when it is done with the
1752 * buffer, but also the buffer can be checkpointed only
1753 * after the new transaction commits. */
1755 set_buffer_freed(bh);
1757 if (!jh->b_next_transaction) {
1758 spin_lock(&journal->j_list_lock);
1759 jh->b_next_transaction = transaction;
1760 spin_unlock(&journal->j_list_lock);
1762 J_ASSERT(jh->b_next_transaction == transaction);
1765 * only drop a reference if this transaction modified
1773 * Finally, if the buffer is not belongs to any
1774 * transaction, we can just drop it now if it has no
1777 spin_lock(&journal->j_list_lock);
1778 if (!jh->b_cp_transaction) {
1779 JBUFFER_TRACE(jh, "belongs to none transaction");
1780 spin_unlock(&journal->j_list_lock);
1785 * Otherwise, if the buffer has been written to disk,
1786 * it is safe to remove the checkpoint and drop it.
1788 if (jbd2_journal_try_remove_checkpoint(jh) >= 0) {
1789 spin_unlock(&journal->j_list_lock);
1794 * The buffer is still not written to disk, we should
1795 * attach this buffer to current transaction so that the
1796 * buffer can be checkpointed only after the current
1797 * transaction commits.
1799 clear_buffer_dirty(bh);
1800 __jbd2_journal_file_buffer(jh, transaction, BJ_Forget);
1801 spin_unlock(&journal->j_list_lock);
1805 spin_unlock(&jh->b_state_lock);
1806 jbd2_journal_put_journal_head(jh);
1808 /* no need to reserve log space for this block -bzzz */
1809 handle->h_total_credits++;
1815 * jbd2_journal_stop() - complete a transaction
1816 * @handle: transaction to complete.
1818 * All done for a particular handle.
1820 * There is not much action needed here. We just return any remaining
1821 * buffer credits to the transaction and remove the handle. The only
1822 * complication is that we need to start a commit operation if the
1823 * filesystem is marked for synchronous update.
1825 * jbd2_journal_stop itself will not usually return an error, but it may
1826 * do so in unusual circumstances. In particular, expect it to
1827 * return -EIO if a jbd2_journal_abort has been executed since the
1828 * transaction began.
1830 int jbd2_journal_stop(handle_t *handle)
1832 transaction_t *transaction = handle->h_transaction;
1834 int err = 0, wait_for_commit = 0;
1838 if (--handle->h_ref > 0) {
1839 jbd2_debug(4, "h_ref %d -> %d\n", handle->h_ref + 1,
1841 if (is_handle_aborted(handle))
1847 * Handle is already detached from the transaction so there is
1848 * nothing to do other than free the handle.
1850 memalloc_nofs_restore(handle->saved_alloc_context);
1853 journal = transaction->t_journal;
1854 tid = transaction->t_tid;
1856 if (is_handle_aborted(handle))
1859 jbd2_debug(4, "Handle %p going down\n", handle);
1860 trace_jbd2_handle_stats(journal->j_fs_dev->bd_dev,
1861 tid, handle->h_type, handle->h_line_no,
1862 jiffies - handle->h_start_jiffies,
1863 handle->h_sync, handle->h_requested_credits,
1864 (handle->h_requested_credits -
1865 handle->h_total_credits));
1868 * Implement synchronous transaction batching. If the handle
1869 * was synchronous, don't force a commit immediately. Let's
1870 * yield and let another thread piggyback onto this
1871 * transaction. Keep doing that while new threads continue to
1872 * arrive. It doesn't cost much - we're about to run a commit
1873 * and sleep on IO anyway. Speeds up many-threaded, many-dir
1874 * operations by 30x or more...
1876 * We try and optimize the sleep time against what the
1877 * underlying disk can do, instead of having a static sleep
1878 * time. This is useful for the case where our storage is so
1879 * fast that it is more optimal to go ahead and force a flush
1880 * and wait for the transaction to be committed than it is to
1881 * wait for an arbitrary amount of time for new writers to
1882 * join the transaction. We achieve this by measuring how
1883 * long it takes to commit a transaction, and compare it with
1884 * how long this transaction has been running, and if run time
1885 * < commit time then we sleep for the delta and commit. This
1886 * greatly helps super fast disks that would see slowdowns as
1887 * more threads started doing fsyncs.
1889 * But don't do this if this process was the most recent one
1890 * to perform a synchronous write. We do this to detect the
1891 * case where a single process is doing a stream of sync
1892 * writes. No point in waiting for joiners in that case.
1894 * Setting max_batch_time to 0 disables this completely.
1897 if (handle->h_sync && journal->j_last_sync_writer != pid &&
1898 journal->j_max_batch_time) {
1899 u64 commit_time, trans_time;
1901 journal->j_last_sync_writer = pid;
1903 read_lock(&journal->j_state_lock);
1904 commit_time = journal->j_average_commit_time;
1905 read_unlock(&journal->j_state_lock);
1907 trans_time = ktime_to_ns(ktime_sub(ktime_get(),
1908 transaction->t_start_time));
1910 commit_time = max_t(u64, commit_time,
1911 1000*journal->j_min_batch_time);
1912 commit_time = min_t(u64, commit_time,
1913 1000*journal->j_max_batch_time);
1915 if (trans_time < commit_time) {
1916 ktime_t expires = ktime_add_ns(ktime_get(),
1918 set_current_state(TASK_UNINTERRUPTIBLE);
1919 schedule_hrtimeout(&expires, HRTIMER_MODE_ABS);
1924 transaction->t_synchronous_commit = 1;
1927 * If the handle is marked SYNC, we need to set another commit
1928 * going! We also want to force a commit if the transaction is too
1931 if (handle->h_sync ||
1932 time_after_eq(jiffies, transaction->t_expires)) {
1933 /* Do this even for aborted journals: an abort still
1934 * completes the commit thread, it just doesn't write
1935 * anything to disk. */
1937 jbd2_debug(2, "transaction too old, requesting commit for "
1938 "handle %p\n", handle);
1939 /* This is non-blocking */
1940 jbd2_log_start_commit(journal, tid);
1943 * Special case: JBD2_SYNC synchronous updates require us
1944 * to wait for the commit to complete.
1946 if (handle->h_sync && !(current->flags & PF_MEMALLOC))
1947 wait_for_commit = 1;
1951 * Once stop_this_handle() drops t_updates, the transaction could start
1952 * committing on us and eventually disappear. So we must not
1953 * dereference transaction pointer again after calling
1954 * stop_this_handle().
1956 stop_this_handle(handle);
1958 if (wait_for_commit)
1959 err = jbd2_log_wait_commit(journal, tid);
1962 if (handle->h_rsv_handle)
1963 jbd2_free_handle(handle->h_rsv_handle);
1964 jbd2_free_handle(handle);
1970 * List management code snippets: various functions for manipulating the
1971 * transaction buffer lists.
1976 * Append a buffer to a transaction list, given the transaction's list head
1979 * j_list_lock is held.
1981 * jh->b_state_lock is held.
1985 __blist_add_buffer(struct journal_head **list, struct journal_head *jh)
1988 jh->b_tnext = jh->b_tprev = jh;
1991 /* Insert at the tail of the list to preserve order */
1992 struct journal_head *first = *list, *last = first->b_tprev;
1994 jh->b_tnext = first;
1995 last->b_tnext = first->b_tprev = jh;
2000 * Remove a buffer from a transaction list, given the transaction's list
2003 * Called with j_list_lock held, and the journal may not be locked.
2005 * jh->b_state_lock is held.
2009 __blist_del_buffer(struct journal_head **list, struct journal_head *jh)
2012 *list = jh->b_tnext;
2016 jh->b_tprev->b_tnext = jh->b_tnext;
2017 jh->b_tnext->b_tprev = jh->b_tprev;
2021 * Remove a buffer from the appropriate transaction list.
2023 * Note that this function can *change* the value of
2024 * bh->b_transaction->t_buffers, t_forget, t_shadow_list, t_log_list or
2025 * t_reserved_list. If the caller is holding onto a copy of one of these
2026 * pointers, it could go bad. Generally the caller needs to re-read the
2027 * pointer from the transaction_t.
2029 * Called under j_list_lock.
2031 static void __jbd2_journal_temp_unlink_buffer(struct journal_head *jh)
2033 struct journal_head **list = NULL;
2034 transaction_t *transaction;
2035 struct buffer_head *bh = jh2bh(jh);
2037 lockdep_assert_held(&jh->b_state_lock);
2038 transaction = jh->b_transaction;
2040 assert_spin_locked(&transaction->t_journal->j_list_lock);
2042 J_ASSERT_JH(jh, jh->b_jlist < BJ_Types);
2043 if (jh->b_jlist != BJ_None)
2044 J_ASSERT_JH(jh, transaction != NULL);
2046 switch (jh->b_jlist) {
2050 transaction->t_nr_buffers--;
2051 J_ASSERT_JH(jh, transaction->t_nr_buffers >= 0);
2052 list = &transaction->t_buffers;
2055 list = &transaction->t_forget;
2058 list = &transaction->t_shadow_list;
2061 list = &transaction->t_reserved_list;
2065 __blist_del_buffer(list, jh);
2066 jh->b_jlist = BJ_None;
2067 if (transaction && is_journal_aborted(transaction->t_journal))
2068 clear_buffer_jbddirty(bh);
2069 else if (test_clear_buffer_jbddirty(bh))
2070 mark_buffer_dirty(bh); /* Expose it to the VM */
2074 * Remove buffer from all transactions. The caller is responsible for dropping
2075 * the jh reference that belonged to the transaction.
2077 * Called with bh_state lock and j_list_lock
2079 static void __jbd2_journal_unfile_buffer(struct journal_head *jh)
2081 J_ASSERT_JH(jh, jh->b_transaction != NULL);
2082 J_ASSERT_JH(jh, jh->b_next_transaction == NULL);
2084 __jbd2_journal_temp_unlink_buffer(jh);
2085 jh->b_transaction = NULL;
2088 void jbd2_journal_unfile_buffer(journal_t *journal, struct journal_head *jh)
2090 struct buffer_head *bh = jh2bh(jh);
2092 /* Get reference so that buffer cannot be freed before we unlock it */
2094 spin_lock(&jh->b_state_lock);
2095 spin_lock(&journal->j_list_lock);
2096 __jbd2_journal_unfile_buffer(jh);
2097 spin_unlock(&journal->j_list_lock);
2098 spin_unlock(&jh->b_state_lock);
2099 jbd2_journal_put_journal_head(jh);
2104 * Called from jbd2_journal_try_to_free_buffers().
2106 * Called under jh->b_state_lock
2109 __journal_try_to_free_buffer(journal_t *journal, struct buffer_head *bh)
2111 struct journal_head *jh;
2115 if (jh->b_next_transaction != NULL || jh->b_transaction != NULL)
2118 spin_lock(&journal->j_list_lock);
2119 /* Remove written-back checkpointed metadata buffer */
2120 if (jh->b_cp_transaction != NULL)
2121 jbd2_journal_try_remove_checkpoint(jh);
2122 spin_unlock(&journal->j_list_lock);
2127 * jbd2_journal_try_to_free_buffers() - try to free page buffers.
2128 * @journal: journal for operation
2129 * @page: to try and free
2131 * For all the buffers on this page,
2132 * if they are fully written out ordered data, move them onto BUF_CLEAN
2133 * so try_to_free_buffers() can reap them.
2135 * This function returns non-zero if we wish try_to_free_buffers()
2136 * to be called. We do this if the page is releasable by try_to_free_buffers().
2137 * We also do it if the page has locked or dirty buffers and the caller wants
2138 * us to perform sync or async writeout.
2140 * This complicates JBD locking somewhat. We aren't protected by the
2141 * BKL here. We wish to remove the buffer from its committing or
2142 * running transaction's ->t_datalist via __jbd2_journal_unfile_buffer.
2144 * This may *change* the value of transaction_t->t_datalist, so anyone
2145 * who looks at t_datalist needs to lock against this function.
2147 * Even worse, someone may be doing a jbd2_journal_dirty_data on this
2148 * buffer. So we need to lock against that. jbd2_journal_dirty_data()
2149 * will come out of the lock with the buffer dirty, which makes it
2150 * ineligible for release here.
2152 * Who else is affected by this? hmm... Really the only contender
2153 * is do_get_write_access() - it could be looking at the buffer while
2154 * journal_try_to_free_buffer() is changing its state. But that
2155 * cannot happen because we never reallocate freed data as metadata
2156 * while the data is part of a transaction. Yes?
2158 * Return 0 on failure, 1 on success
2160 int jbd2_journal_try_to_free_buffers(journal_t *journal, struct page *page)
2162 struct buffer_head *head;
2163 struct buffer_head *bh;
2166 J_ASSERT(PageLocked(page));
2168 head = page_buffers(page);
2171 struct journal_head *jh;
2174 * We take our own ref against the journal_head here to avoid
2175 * having to add tons of locking around each instance of
2176 * jbd2_journal_put_journal_head().
2178 jh = jbd2_journal_grab_journal_head(bh);
2182 spin_lock(&jh->b_state_lock);
2183 __journal_try_to_free_buffer(journal, bh);
2184 spin_unlock(&jh->b_state_lock);
2185 jbd2_journal_put_journal_head(jh);
2188 } while ((bh = bh->b_this_page) != head);
2190 ret = try_to_free_buffers(page);
2196 * This buffer is no longer needed. If it is on an older transaction's
2197 * checkpoint list we need to record it on this transaction's forget list
2198 * to pin this buffer (and hence its checkpointing transaction) down until
2199 * this transaction commits. If the buffer isn't on a checkpoint list, we
2201 * Returns non-zero if JBD no longer has an interest in the buffer.
2203 * Called under j_list_lock.
2205 * Called under jh->b_state_lock.
2207 static int __dispose_buffer(struct journal_head *jh, transaction_t *transaction)
2210 struct buffer_head *bh = jh2bh(jh);
2212 if (jh->b_cp_transaction) {
2213 JBUFFER_TRACE(jh, "on running+cp transaction");
2214 __jbd2_journal_temp_unlink_buffer(jh);
2216 * We don't want to write the buffer anymore, clear the
2217 * bit so that we don't confuse checks in
2218 * __journal_file_buffer
2220 clear_buffer_dirty(bh);
2221 __jbd2_journal_file_buffer(jh, transaction, BJ_Forget);
2224 JBUFFER_TRACE(jh, "on running transaction");
2225 __jbd2_journal_unfile_buffer(jh);
2226 jbd2_journal_put_journal_head(jh);
2232 * jbd2_journal_invalidatepage
2234 * This code is tricky. It has a number of cases to deal with.
2236 * There are two invariants which this code relies on:
2238 * i_size must be updated on disk before we start calling invalidatepage on the
2241 * This is done in ext3 by defining an ext3_setattr method which
2242 * updates i_size before truncate gets going. By maintaining this
2243 * invariant, we can be sure that it is safe to throw away any buffers
2244 * attached to the current transaction: once the transaction commits,
2245 * we know that the data will not be needed.
2247 * Note however that we can *not* throw away data belonging to the
2248 * previous, committing transaction!
2250 * Any disk blocks which *are* part of the previous, committing
2251 * transaction (and which therefore cannot be discarded immediately) are
2252 * not going to be reused in the new running transaction
2254 * The bitmap committed_data images guarantee this: any block which is
2255 * allocated in one transaction and removed in the next will be marked
2256 * as in-use in the committed_data bitmap, so cannot be reused until
2257 * the next transaction to delete the block commits. This means that
2258 * leaving committing buffers dirty is quite safe: the disk blocks
2259 * cannot be reallocated to a different file and so buffer aliasing is
2263 * The above applies mainly to ordered data mode. In writeback mode we
2264 * don't make guarantees about the order in which data hits disk --- in
2265 * particular we don't guarantee that new dirty data is flushed before
2266 * transaction commit --- so it is always safe just to discard data
2267 * immediately in that mode. --sct
2271 * The journal_unmap_buffer helper function returns zero if the buffer
2272 * concerned remains pinned as an anonymous buffer belonging to an older
2275 * We're outside-transaction here. Either or both of j_running_transaction
2276 * and j_committing_transaction may be NULL.
2278 static int journal_unmap_buffer(journal_t *journal, struct buffer_head *bh,
2281 transaction_t *transaction;
2282 struct journal_head *jh;
2285 BUFFER_TRACE(bh, "entry");
2288 * It is safe to proceed here without the j_list_lock because the
2289 * buffers cannot be stolen by try_to_free_buffers as long as we are
2290 * holding the page lock. --sct
2293 jh = jbd2_journal_grab_journal_head(bh);
2295 goto zap_buffer_unlocked;
2297 /* OK, we have data buffer in journaled mode */
2298 write_lock(&journal->j_state_lock);
2299 spin_lock(&jh->b_state_lock);
2300 spin_lock(&journal->j_list_lock);
2303 * We cannot remove the buffer from checkpoint lists until the
2304 * transaction adding inode to orphan list (let's call it T)
2305 * is committed. Otherwise if the transaction changing the
2306 * buffer would be cleaned from the journal before T is
2307 * committed, a crash will cause that the correct contents of
2308 * the buffer will be lost. On the other hand we have to
2309 * clear the buffer dirty bit at latest at the moment when the
2310 * transaction marking the buffer as freed in the filesystem
2311 * structures is committed because from that moment on the
2312 * block can be reallocated and used by a different page.
2313 * Since the block hasn't been freed yet but the inode has
2314 * already been added to orphan list, it is safe for us to add
2315 * the buffer to BJ_Forget list of the newest transaction.
2317 * Also we have to clear buffer_mapped flag of a truncated buffer
2318 * because the buffer_head may be attached to the page straddling
2319 * i_size (can happen only when blocksize < pagesize) and thus the
2320 * buffer_head can be reused when the file is extended again. So we end
2321 * up keeping around invalidated buffers attached to transactions'
2322 * BJ_Forget list just to stop checkpointing code from cleaning up
2323 * the transaction this buffer was modified in.
2325 transaction = jh->b_transaction;
2326 if (transaction == NULL) {
2327 /* First case: not on any transaction. If it
2328 * has no checkpoint link, then we can zap it:
2329 * it's a writeback-mode buffer so we don't care
2330 * if it hits disk safely. */
2331 if (!jh->b_cp_transaction) {
2332 JBUFFER_TRACE(jh, "not on any transaction: zap");
2336 if (!buffer_dirty(bh)) {
2337 /* bdflush has written it. We can drop it now */
2338 __jbd2_journal_remove_checkpoint(jh);
2342 /* OK, it must be in the journal but still not
2343 * written fully to disk: it's metadata or
2344 * journaled data... */
2346 if (journal->j_running_transaction) {
2347 /* ... and once the current transaction has
2348 * committed, the buffer won't be needed any
2350 JBUFFER_TRACE(jh, "checkpointed: add to BJ_Forget");
2351 may_free = __dispose_buffer(jh,
2352 journal->j_running_transaction);
2355 /* There is no currently-running transaction. So the
2356 * orphan record which we wrote for this file must have
2357 * passed into commit. We must attach this buffer to
2358 * the committing transaction, if it exists. */
2359 if (journal->j_committing_transaction) {
2360 JBUFFER_TRACE(jh, "give to committing trans");
2361 may_free = __dispose_buffer(jh,
2362 journal->j_committing_transaction);
2365 /* The orphan record's transaction has
2366 * committed. We can cleanse this buffer */
2367 clear_buffer_jbddirty(bh);
2368 __jbd2_journal_remove_checkpoint(jh);
2372 } else if (transaction == journal->j_committing_transaction) {
2373 JBUFFER_TRACE(jh, "on committing transaction");
2375 * The buffer is committing, we simply cannot touch
2376 * it. If the page is straddling i_size we have to wait
2377 * for commit and try again.
2380 spin_unlock(&journal->j_list_lock);
2381 spin_unlock(&jh->b_state_lock);
2382 write_unlock(&journal->j_state_lock);
2383 jbd2_journal_put_journal_head(jh);
2384 /* Already zapped buffer? Nothing to do... */
2390 * OK, buffer won't be reachable after truncate. We just clear
2391 * b_modified to not confuse transaction credit accounting, and
2392 * set j_next_transaction to the running transaction (if there
2393 * is one) and mark buffer as freed so that commit code knows
2394 * it should clear dirty bits when it is done with the buffer.
2396 set_buffer_freed(bh);
2397 if (journal->j_running_transaction && buffer_jbddirty(bh))
2398 jh->b_next_transaction = journal->j_running_transaction;
2400 spin_unlock(&journal->j_list_lock);
2401 spin_unlock(&jh->b_state_lock);
2402 write_unlock(&journal->j_state_lock);
2403 jbd2_journal_put_journal_head(jh);
2406 /* Good, the buffer belongs to the running transaction.
2407 * We are writing our own transaction's data, not any
2408 * previous one's, so it is safe to throw it away
2409 * (remember that we expect the filesystem to have set
2410 * i_size already for this truncate so recovery will not
2411 * expose the disk blocks we are discarding here.) */
2412 J_ASSERT_JH(jh, transaction == journal->j_running_transaction);
2413 JBUFFER_TRACE(jh, "on running transaction");
2414 may_free = __dispose_buffer(jh, transaction);
2419 * This is tricky. Although the buffer is truncated, it may be reused
2420 * if blocksize < pagesize and it is attached to the page straddling
2421 * EOF. Since the buffer might have been added to BJ_Forget list of the
2422 * running transaction, journal_get_write_access() won't clear
2423 * b_modified and credit accounting gets confused. So clear b_modified
2427 spin_unlock(&journal->j_list_lock);
2428 spin_unlock(&jh->b_state_lock);
2429 write_unlock(&journal->j_state_lock);
2430 jbd2_journal_put_journal_head(jh);
2431 zap_buffer_unlocked:
2432 clear_buffer_dirty(bh);
2433 J_ASSERT_BH(bh, !buffer_jbddirty(bh));
2434 clear_buffer_mapped(bh);
2435 clear_buffer_req(bh);
2436 clear_buffer_new(bh);
2437 clear_buffer_delay(bh);
2438 clear_buffer_unwritten(bh);
2444 * jbd2_journal_invalidatepage()
2445 * @journal: journal to use for flush...
2446 * @page: page to flush
2447 * @offset: start of the range to invalidate
2448 * @length: length of the range to invalidate
2450 * Reap page buffers containing data after in the specified range in page.
2451 * Can return -EBUSY if buffers are part of the committing transaction and
2452 * the page is straddling i_size. Caller then has to wait for current commit
2455 int jbd2_journal_invalidatepage(journal_t *journal,
2457 unsigned int offset,
2458 unsigned int length)
2460 struct buffer_head *head, *bh, *next;
2461 unsigned int stop = offset + length;
2462 unsigned int curr_off = 0;
2463 int partial_page = (offset || length < PAGE_SIZE);
2467 if (!PageLocked(page))
2469 if (!page_has_buffers(page))
2472 BUG_ON(stop > PAGE_SIZE || stop < length);
2474 /* We will potentially be playing with lists other than just the
2475 * data lists (especially for journaled data mode), so be
2476 * cautious in our locking. */
2478 head = bh = page_buffers(page);
2480 unsigned int next_off = curr_off + bh->b_size;
2481 next = bh->b_this_page;
2483 if (next_off > stop)
2486 if (offset <= curr_off) {
2487 /* This block is wholly outside the truncation point */
2489 ret = journal_unmap_buffer(journal, bh, partial_page);
2495 curr_off = next_off;
2498 } while (bh != head);
2500 if (!partial_page) {
2501 if (may_free && try_to_free_buffers(page))
2502 J_ASSERT(!page_has_buffers(page));
2508 * File a buffer on the given transaction list.
2510 void __jbd2_journal_file_buffer(struct journal_head *jh,
2511 transaction_t *transaction, int jlist)
2513 struct journal_head **list = NULL;
2515 struct buffer_head *bh = jh2bh(jh);
2517 lockdep_assert_held(&jh->b_state_lock);
2518 assert_spin_locked(&transaction->t_journal->j_list_lock);
2520 J_ASSERT_JH(jh, jh->b_jlist < BJ_Types);
2521 J_ASSERT_JH(jh, jh->b_transaction == transaction ||
2522 jh->b_transaction == NULL);
2524 if (jh->b_transaction && jh->b_jlist == jlist)
2527 if (jlist == BJ_Metadata || jlist == BJ_Reserved ||
2528 jlist == BJ_Shadow || jlist == BJ_Forget) {
2530 * For metadata buffers, we track dirty bit in buffer_jbddirty
2531 * instead of buffer_dirty. We should not see a dirty bit set
2532 * here because we clear it in do_get_write_access but e.g.
2533 * tune2fs can modify the sb and set the dirty bit at any time
2534 * so we try to gracefully handle that.
2536 if (buffer_dirty(bh))
2537 warn_dirty_buffer(bh);
2538 if (test_clear_buffer_dirty(bh) ||
2539 test_clear_buffer_jbddirty(bh))
2543 if (jh->b_transaction)
2544 __jbd2_journal_temp_unlink_buffer(jh);
2546 jbd2_journal_grab_journal_head(bh);
2547 jh->b_transaction = transaction;
2551 J_ASSERT_JH(jh, !jh->b_committed_data);
2552 J_ASSERT_JH(jh, !jh->b_frozen_data);
2555 transaction->t_nr_buffers++;
2556 list = &transaction->t_buffers;
2559 list = &transaction->t_forget;
2562 list = &transaction->t_shadow_list;
2565 list = &transaction->t_reserved_list;
2569 __blist_add_buffer(list, jh);
2570 jh->b_jlist = jlist;
2573 set_buffer_jbddirty(bh);
2576 void jbd2_journal_file_buffer(struct journal_head *jh,
2577 transaction_t *transaction, int jlist)
2579 spin_lock(&jh->b_state_lock);
2580 spin_lock(&transaction->t_journal->j_list_lock);
2581 __jbd2_journal_file_buffer(jh, transaction, jlist);
2582 spin_unlock(&transaction->t_journal->j_list_lock);
2583 spin_unlock(&jh->b_state_lock);
2587 * Remove a buffer from its current buffer list in preparation for
2588 * dropping it from its current transaction entirely. If the buffer has
2589 * already started to be used by a subsequent transaction, refile the
2590 * buffer on that transaction's metadata list.
2592 * Called under j_list_lock
2593 * Called under jh->b_state_lock
2595 * When this function returns true, there's no next transaction to refile to
2596 * and the caller has to drop jh reference through
2597 * jbd2_journal_put_journal_head().
2599 bool __jbd2_journal_refile_buffer(struct journal_head *jh)
2601 int was_dirty, jlist;
2602 struct buffer_head *bh = jh2bh(jh);
2604 lockdep_assert_held(&jh->b_state_lock);
2605 if (jh->b_transaction)
2606 assert_spin_locked(&jh->b_transaction->t_journal->j_list_lock);
2608 /* If the buffer is now unused, just drop it. */
2609 if (jh->b_next_transaction == NULL) {
2610 __jbd2_journal_unfile_buffer(jh);
2615 * It has been modified by a later transaction: add it to the new
2616 * transaction's metadata list.
2619 was_dirty = test_clear_buffer_jbddirty(bh);
2620 __jbd2_journal_temp_unlink_buffer(jh);
2623 * b_transaction must be set, otherwise the new b_transaction won't
2624 * be holding jh reference
2626 J_ASSERT_JH(jh, jh->b_transaction != NULL);
2629 * We set b_transaction here because b_next_transaction will inherit
2630 * our jh reference and thus __jbd2_journal_file_buffer() must not
2633 WRITE_ONCE(jh->b_transaction, jh->b_next_transaction);
2634 WRITE_ONCE(jh->b_next_transaction, NULL);
2635 if (buffer_freed(bh))
2637 else if (jh->b_modified)
2638 jlist = BJ_Metadata;
2640 jlist = BJ_Reserved;
2641 __jbd2_journal_file_buffer(jh, jh->b_transaction, jlist);
2642 J_ASSERT_JH(jh, jh->b_transaction->t_state == T_RUNNING);
2645 set_buffer_jbddirty(bh);
2650 * __jbd2_journal_refile_buffer() with necessary locking added. We take our
2651 * bh reference so that we can safely unlock bh.
2653 * The jh and bh may be freed by this call.
2655 void jbd2_journal_refile_buffer(journal_t *journal, struct journal_head *jh)
2659 spin_lock(&jh->b_state_lock);
2660 spin_lock(&journal->j_list_lock);
2661 drop = __jbd2_journal_refile_buffer(jh);
2662 spin_unlock(&jh->b_state_lock);
2663 spin_unlock(&journal->j_list_lock);
2665 jbd2_journal_put_journal_head(jh);
2669 * File inode in the inode list of the handle's transaction
2671 static int jbd2_journal_file_inode(handle_t *handle, struct jbd2_inode *jinode,
2672 unsigned long flags, loff_t start_byte, loff_t end_byte)
2674 transaction_t *transaction = handle->h_transaction;
2677 if (is_handle_aborted(handle))
2679 journal = transaction->t_journal;
2681 jbd2_debug(4, "Adding inode %lu, tid:%d\n", jinode->i_vfs_inode->i_ino,
2682 transaction->t_tid);
2684 spin_lock(&journal->j_list_lock);
2685 jinode->i_flags |= flags;
2687 if (jinode->i_dirty_end) {
2688 jinode->i_dirty_start = min(jinode->i_dirty_start, start_byte);
2689 jinode->i_dirty_end = max(jinode->i_dirty_end, end_byte);
2691 jinode->i_dirty_start = start_byte;
2692 jinode->i_dirty_end = end_byte;
2695 /* Is inode already attached where we need it? */
2696 if (jinode->i_transaction == transaction ||
2697 jinode->i_next_transaction == transaction)
2701 * We only ever set this variable to 1 so the test is safe. Since
2702 * t_need_data_flush is likely to be set, we do the test to save some
2703 * cacheline bouncing
2705 if (!transaction->t_need_data_flush)
2706 transaction->t_need_data_flush = 1;
2707 /* On some different transaction's list - should be
2708 * the committing one */
2709 if (jinode->i_transaction) {
2710 J_ASSERT(jinode->i_next_transaction == NULL);
2711 J_ASSERT(jinode->i_transaction ==
2712 journal->j_committing_transaction);
2713 jinode->i_next_transaction = transaction;
2716 /* Not on any transaction list... */
2717 J_ASSERT(!jinode->i_next_transaction);
2718 jinode->i_transaction = transaction;
2719 list_add(&jinode->i_list, &transaction->t_inode_list);
2721 spin_unlock(&journal->j_list_lock);
2726 int jbd2_journal_inode_ranged_write(handle_t *handle,
2727 struct jbd2_inode *jinode, loff_t start_byte, loff_t length)
2729 return jbd2_journal_file_inode(handle, jinode,
2730 JI_WRITE_DATA | JI_WAIT_DATA, start_byte,
2731 start_byte + length - 1);
2734 int jbd2_journal_inode_ranged_wait(handle_t *handle, struct jbd2_inode *jinode,
2735 loff_t start_byte, loff_t length)
2737 return jbd2_journal_file_inode(handle, jinode, JI_WAIT_DATA,
2738 start_byte, start_byte + length - 1);
2742 * File truncate and transaction commit interact with each other in a
2743 * non-trivial way. If a transaction writing data block A is
2744 * committing, we cannot discard the data by truncate until we have
2745 * written them. Otherwise if we crashed after the transaction with
2746 * write has committed but before the transaction with truncate has
2747 * committed, we could see stale data in block A. This function is a
2748 * helper to solve this problem. It starts writeout of the truncated
2749 * part in case it is in the committing transaction.
2751 * Filesystem code must call this function when inode is journaled in
2752 * ordered mode before truncation happens and after the inode has been
2753 * placed on orphan list with the new inode size. The second condition
2754 * avoids the race that someone writes new data and we start
2755 * committing the transaction after this function has been called but
2756 * before a transaction for truncate is started (and furthermore it
2757 * allows us to optimize the case where the addition to orphan list
2758 * happens in the same transaction as write --- we don't have to write
2759 * any data in such case).
2761 int jbd2_journal_begin_ordered_truncate(journal_t *journal,
2762 struct jbd2_inode *jinode,
2765 transaction_t *inode_trans, *commit_trans;
2768 /* This is a quick check to avoid locking if not necessary */
2769 if (!jinode->i_transaction)
2771 /* Locks are here just to force reading of recent values, it is
2772 * enough that the transaction was not committing before we started
2773 * a transaction adding the inode to orphan list */
2774 read_lock(&journal->j_state_lock);
2775 commit_trans = journal->j_committing_transaction;
2776 read_unlock(&journal->j_state_lock);
2777 spin_lock(&journal->j_list_lock);
2778 inode_trans = jinode->i_transaction;
2779 spin_unlock(&journal->j_list_lock);
2780 if (inode_trans == commit_trans) {
2781 ret = filemap_fdatawrite_range(jinode->i_vfs_inode->i_mapping,
2782 new_size, LLONG_MAX);
2784 jbd2_journal_abort(journal, ret);