2 * linux/fs/jbd2/transaction.c
4 * Written by Stephen C. Tweedie <sct@redhat.com>, 1998
6 * Copyright 1998 Red Hat corp --- All Rights Reserved
8 * This file is part of the Linux kernel and is made available under
9 * the terms of the GNU General Public License, version 2, or at your
10 * option, any later version, incorporated herein by reference.
12 * Generic filesystem transaction handling code; part of the ext2fs
15 * This file manages transactions (compound commits managed by the
16 * journaling code) and handles (individual atomic operations by the
20 #include <linux/time.h>
22 #include <linux/jbd2.h>
23 #include <linux/errno.h>
24 #include <linux/slab.h>
25 #include <linux/timer.h>
27 #include <linux/highmem.h>
28 #include <linux/hrtimer.h>
29 #include <linux/backing-dev.h>
30 #include <linux/bug.h>
31 #include <linux/module.h>
32 #include <linux/sched/mm.h>
34 #include <trace/events/jbd2.h>
36 static void __jbd2_journal_temp_unlink_buffer(struct journal_head *jh);
37 static void __jbd2_journal_unfile_buffer(struct journal_head *jh);
39 static struct kmem_cache *transaction_cache;
40 int __init jbd2_journal_init_transaction_cache(void)
42 J_ASSERT(!transaction_cache);
43 transaction_cache = kmem_cache_create("jbd2_transaction_s",
44 sizeof(transaction_t),
46 SLAB_HWCACHE_ALIGN|SLAB_TEMPORARY,
48 if (transaction_cache)
53 void jbd2_journal_destroy_transaction_cache(void)
55 if (transaction_cache) {
56 kmem_cache_destroy(transaction_cache);
57 transaction_cache = NULL;
61 void jbd2_journal_free_transaction(transaction_t *transaction)
63 if (unlikely(ZERO_OR_NULL_PTR(transaction)))
65 kmem_cache_free(transaction_cache, transaction);
69 * jbd2_get_transaction: obtain a new transaction_t object.
71 * Simply allocate and initialise a new transaction. Create it in
72 * RUNNING state and add it to the current journal (which should not
73 * have an existing running transaction: we only make a new transaction
74 * once we have started to commit the old one).
77 * The journal MUST be locked. We don't perform atomic mallocs on the
78 * new transaction and we can't block without protecting against other
79 * processes trying to touch the journal while it is in transition.
83 static transaction_t *
84 jbd2_get_transaction(journal_t *journal, transaction_t *transaction)
86 transaction->t_journal = journal;
87 transaction->t_state = T_RUNNING;
88 transaction->t_start_time = ktime_get();
89 transaction->t_tid = journal->j_transaction_sequence++;
90 transaction->t_expires = jiffies + journal->j_commit_interval;
91 spin_lock_init(&transaction->t_handle_lock);
92 atomic_set(&transaction->t_updates, 0);
93 atomic_set(&transaction->t_outstanding_credits,
94 atomic_read(&journal->j_reserved_credits));
95 atomic_set(&transaction->t_handle_count, 0);
96 INIT_LIST_HEAD(&transaction->t_inode_list);
97 INIT_LIST_HEAD(&transaction->t_private_list);
99 /* Set up the commit timer for the new transaction. */
100 journal->j_commit_timer.expires = round_jiffies_up(transaction->t_expires);
101 add_timer(&journal->j_commit_timer);
103 J_ASSERT(journal->j_running_transaction == NULL);
104 journal->j_running_transaction = transaction;
105 transaction->t_max_wait = 0;
106 transaction->t_start = jiffies;
107 transaction->t_requested = 0;
115 * A handle_t is an object which represents a single atomic update to a
116 * filesystem, and which tracks all of the modifications which form part
117 * of that one update.
121 * Update transaction's maximum wait time, if debugging is enabled.
123 * In order for t_max_wait to be reliable, it must be protected by a
124 * lock. But doing so will mean that start_this_handle() can not be
125 * run in parallel on SMP systems, which limits our scalability. So
126 * unless debugging is enabled, we no longer update t_max_wait, which
127 * means that maximum wait time reported by the jbd2_run_stats
128 * tracepoint will always be zero.
130 static inline void update_t_max_wait(transaction_t *transaction,
133 #ifdef CONFIG_JBD2_DEBUG
134 if (jbd2_journal_enable_debug &&
135 time_after(transaction->t_start, ts)) {
136 ts = jbd2_time_diff(ts, transaction->t_start);
137 spin_lock(&transaction->t_handle_lock);
138 if (ts > transaction->t_max_wait)
139 transaction->t_max_wait = ts;
140 spin_unlock(&transaction->t_handle_lock);
146 * Wait until running transaction passes T_LOCKED state. Also starts the commit
147 * if needed. The function expects running transaction to exist and releases
150 static void wait_transaction_locked(journal_t *journal)
151 __releases(journal->j_state_lock)
155 tid_t tid = journal->j_running_transaction->t_tid;
157 prepare_to_wait(&journal->j_wait_transaction_locked, &wait,
158 TASK_UNINTERRUPTIBLE);
159 need_to_start = !tid_geq(journal->j_commit_request, tid);
160 read_unlock(&journal->j_state_lock);
162 jbd2_log_start_commit(journal, tid);
163 jbd2_might_wait_for_commit(journal);
165 finish_wait(&journal->j_wait_transaction_locked, &wait);
168 static void sub_reserved_credits(journal_t *journal, int blocks)
170 atomic_sub(blocks, &journal->j_reserved_credits);
171 wake_up(&journal->j_wait_reserved);
175 * Wait until we can add credits for handle to the running transaction. Called
176 * with j_state_lock held for reading. Returns 0 if handle joined the running
177 * transaction. Returns 1 if we had to wait, j_state_lock is dropped, and
180 static int add_transaction_credits(journal_t *journal, int blocks,
183 transaction_t *t = journal->j_running_transaction;
185 int total = blocks + rsv_blocks;
188 * If the current transaction is locked down for commit, wait
189 * for the lock to be released.
191 if (t->t_state == T_LOCKED) {
192 wait_transaction_locked(journal);
197 * If there is not enough space left in the log to write all
198 * potential buffers requested by this operation, we need to
199 * stall pending a log checkpoint to free some more log space.
201 needed = atomic_add_return(total, &t->t_outstanding_credits);
202 if (needed > journal->j_max_transaction_buffers) {
204 * If the current transaction is already too large,
205 * then start to commit it: we can then go back and
206 * attach this handle to a new transaction.
208 atomic_sub(total, &t->t_outstanding_credits);
211 * Is the number of reserved credits in the current transaction too
212 * big to fit this handle? Wait until reserved credits are freed.
214 if (atomic_read(&journal->j_reserved_credits) + total >
215 journal->j_max_transaction_buffers) {
216 read_unlock(&journal->j_state_lock);
217 jbd2_might_wait_for_commit(journal);
218 wait_event(journal->j_wait_reserved,
219 atomic_read(&journal->j_reserved_credits) + total <=
220 journal->j_max_transaction_buffers);
224 wait_transaction_locked(journal);
229 * The commit code assumes that it can get enough log space
230 * without forcing a checkpoint. This is *critical* for
231 * correctness: a checkpoint of a buffer which is also
232 * associated with a committing transaction creates a deadlock,
233 * so commit simply cannot force through checkpoints.
235 * We must therefore ensure the necessary space in the journal
236 * *before* starting to dirty potentially checkpointed buffers
237 * in the new transaction.
239 if (jbd2_log_space_left(journal) < jbd2_space_needed(journal)) {
240 atomic_sub(total, &t->t_outstanding_credits);
241 read_unlock(&journal->j_state_lock);
242 jbd2_might_wait_for_commit(journal);
243 write_lock(&journal->j_state_lock);
244 if (jbd2_log_space_left(journal) < jbd2_space_needed(journal))
245 __jbd2_log_wait_for_space(journal);
246 write_unlock(&journal->j_state_lock);
250 /* No reservation? We are done... */
254 needed = atomic_add_return(rsv_blocks, &journal->j_reserved_credits);
255 /* We allow at most half of a transaction to be reserved */
256 if (needed > journal->j_max_transaction_buffers / 2) {
257 sub_reserved_credits(journal, rsv_blocks);
258 atomic_sub(total, &t->t_outstanding_credits);
259 read_unlock(&journal->j_state_lock);
260 jbd2_might_wait_for_commit(journal);
261 wait_event(journal->j_wait_reserved,
262 atomic_read(&journal->j_reserved_credits) + rsv_blocks
263 <= journal->j_max_transaction_buffers / 2);
270 * start_this_handle: Given a handle, deal with any locking or stalling
271 * needed to make sure that there is enough journal space for the handle
272 * to begin. Attach the handle to a transaction and set up the
273 * transaction's buffer credits.
276 static int start_this_handle(journal_t *journal, handle_t *handle,
279 transaction_t *transaction, *new_transaction = NULL;
280 int blocks = handle->h_buffer_credits;
282 unsigned long ts = jiffies;
284 if (handle->h_rsv_handle)
285 rsv_blocks = handle->h_rsv_handle->h_buffer_credits;
288 * Limit the number of reserved credits to 1/2 of maximum transaction
289 * size and limit the number of total credits to not exceed maximum
290 * transaction size per operation.
292 if ((rsv_blocks > journal->j_max_transaction_buffers / 2) ||
293 (rsv_blocks + blocks > journal->j_max_transaction_buffers)) {
294 printk(KERN_ERR "JBD2: %s wants too many credits "
295 "credits:%d rsv_credits:%d max:%d\n",
296 current->comm, blocks, rsv_blocks,
297 journal->j_max_transaction_buffers);
303 if (!journal->j_running_transaction) {
305 * If __GFP_FS is not present, then we may be being called from
306 * inside the fs writeback layer, so we MUST NOT fail.
308 if ((gfp_mask & __GFP_FS) == 0)
309 gfp_mask |= __GFP_NOFAIL;
310 new_transaction = kmem_cache_zalloc(transaction_cache,
312 if (!new_transaction)
316 jbd_debug(3, "New handle %p going live.\n", handle);
319 * We need to hold j_state_lock until t_updates has been incremented,
320 * for proper journal barrier handling
323 read_lock(&journal->j_state_lock);
324 BUG_ON(journal->j_flags & JBD2_UNMOUNT);
325 if (is_journal_aborted(journal) ||
326 (journal->j_errno != 0 && !(journal->j_flags & JBD2_ACK_ERR))) {
327 read_unlock(&journal->j_state_lock);
328 jbd2_journal_free_transaction(new_transaction);
333 * Wait on the journal's transaction barrier if necessary. Specifically
334 * we allow reserved handles to proceed because otherwise commit could
335 * deadlock on page writeback not being able to complete.
337 if (!handle->h_reserved && journal->j_barrier_count) {
338 read_unlock(&journal->j_state_lock);
339 wait_event(journal->j_wait_transaction_locked,
340 journal->j_barrier_count == 0);
344 if (!journal->j_running_transaction) {
345 read_unlock(&journal->j_state_lock);
346 if (!new_transaction)
347 goto alloc_transaction;
348 write_lock(&journal->j_state_lock);
349 if (!journal->j_running_transaction &&
350 (handle->h_reserved || !journal->j_barrier_count)) {
351 jbd2_get_transaction(journal, new_transaction);
352 new_transaction = NULL;
354 write_unlock(&journal->j_state_lock);
358 transaction = journal->j_running_transaction;
360 if (!handle->h_reserved) {
361 /* We may have dropped j_state_lock - restart in that case */
362 if (add_transaction_credits(journal, blocks, rsv_blocks))
366 * We have handle reserved so we are allowed to join T_LOCKED
367 * transaction and we don't have to check for transaction size
370 sub_reserved_credits(journal, blocks);
371 handle->h_reserved = 0;
374 /* OK, account for the buffers that this operation expects to
375 * use and add the handle to the running transaction.
377 update_t_max_wait(transaction, ts);
378 handle->h_transaction = transaction;
379 handle->h_requested_credits = blocks;
380 handle->h_start_jiffies = jiffies;
381 atomic_inc(&transaction->t_updates);
382 atomic_inc(&transaction->t_handle_count);
383 jbd_debug(4, "Handle %p given %d credits (total %d, free %lu)\n",
385 atomic_read(&transaction->t_outstanding_credits),
386 jbd2_log_space_left(journal));
387 read_unlock(&journal->j_state_lock);
388 current->journal_info = handle;
390 rwsem_acquire_read(&journal->j_trans_commit_map, 0, 0, _THIS_IP_);
391 jbd2_journal_free_transaction(new_transaction);
393 * Ensure that no allocations done while the transaction is open are
394 * going to recurse back to the fs layer.
396 handle->saved_alloc_context = memalloc_nofs_save();
400 /* Allocate a new handle. This should probably be in a slab... */
401 static handle_t *new_handle(int nblocks)
403 handle_t *handle = jbd2_alloc_handle(GFP_NOFS);
406 handle->h_buffer_credits = nblocks;
412 handle_t *jbd2__journal_start(journal_t *journal, int nblocks, int rsv_blocks,
413 gfp_t gfp_mask, unsigned int type,
414 unsigned int line_no)
416 handle_t *handle = journal_current_handle();
420 return ERR_PTR(-EROFS);
423 J_ASSERT(handle->h_transaction->t_journal == journal);
428 handle = new_handle(nblocks);
430 return ERR_PTR(-ENOMEM);
432 handle_t *rsv_handle;
434 rsv_handle = new_handle(rsv_blocks);
436 jbd2_free_handle(handle);
437 return ERR_PTR(-ENOMEM);
439 rsv_handle->h_reserved = 1;
440 rsv_handle->h_journal = journal;
441 handle->h_rsv_handle = rsv_handle;
444 err = start_this_handle(journal, handle, gfp_mask);
446 if (handle->h_rsv_handle)
447 jbd2_free_handle(handle->h_rsv_handle);
448 jbd2_free_handle(handle);
451 handle->h_type = type;
452 handle->h_line_no = line_no;
453 trace_jbd2_handle_start(journal->j_fs_dev->bd_dev,
454 handle->h_transaction->t_tid, type,
459 EXPORT_SYMBOL(jbd2__journal_start);
463 * handle_t *jbd2_journal_start() - Obtain a new handle.
464 * @journal: Journal to start transaction on.
465 * @nblocks: number of block buffer we might modify
467 * We make sure that the transaction can guarantee at least nblocks of
468 * modified buffers in the log. We block until the log can guarantee
469 * that much space. Additionally, if rsv_blocks > 0, we also create another
470 * handle with rsv_blocks reserved blocks in the journal. This handle is
471 * is stored in h_rsv_handle. It is not attached to any particular transaction
472 * and thus doesn't block transaction commit. If the caller uses this reserved
473 * handle, it has to set h_rsv_handle to NULL as otherwise jbd2_journal_stop()
474 * on the parent handle will dispose the reserved one. Reserved handle has to
475 * be converted to a normal handle using jbd2_journal_start_reserved() before
478 * Return a pointer to a newly allocated handle, or an ERR_PTR() value
481 handle_t *jbd2_journal_start(journal_t *journal, int nblocks)
483 return jbd2__journal_start(journal, nblocks, 0, GFP_NOFS, 0, 0);
485 EXPORT_SYMBOL(jbd2_journal_start);
487 void jbd2_journal_free_reserved(handle_t *handle)
489 journal_t *journal = handle->h_journal;
491 WARN_ON(!handle->h_reserved);
492 sub_reserved_credits(journal, handle->h_buffer_credits);
493 jbd2_free_handle(handle);
495 EXPORT_SYMBOL(jbd2_journal_free_reserved);
498 * int jbd2_journal_start_reserved() - start reserved handle
499 * @handle: handle to start
500 * @type: for handle statistics
501 * @line_no: for handle statistics
503 * Start handle that has been previously reserved with jbd2_journal_reserve().
504 * This attaches @handle to the running transaction (or creates one if there's
505 * not transaction running). Unlike jbd2_journal_start() this function cannot
506 * block on journal commit, checkpointing, or similar stuff. It can block on
507 * memory allocation or frozen journal though.
509 * Return 0 on success, non-zero on error - handle is freed in that case.
511 int jbd2_journal_start_reserved(handle_t *handle, unsigned int type,
512 unsigned int line_no)
514 journal_t *journal = handle->h_journal;
517 if (WARN_ON(!handle->h_reserved)) {
518 /* Someone passed in normal handle? Just stop it. */
519 jbd2_journal_stop(handle);
523 * Usefulness of mixing of reserved and unreserved handles is
524 * questionable. So far nobody seems to need it so just error out.
526 if (WARN_ON(current->journal_info)) {
527 jbd2_journal_free_reserved(handle);
531 handle->h_journal = NULL;
533 * GFP_NOFS is here because callers are likely from writeback or
534 * similarly constrained call sites
536 ret = start_this_handle(journal, handle, GFP_NOFS);
538 handle->h_journal = journal;
539 jbd2_journal_free_reserved(handle);
542 handle->h_type = type;
543 handle->h_line_no = line_no;
546 EXPORT_SYMBOL(jbd2_journal_start_reserved);
549 * int jbd2_journal_extend() - extend buffer credits.
550 * @handle: handle to 'extend'
551 * @nblocks: nr blocks to try to extend by.
553 * Some transactions, such as large extends and truncates, can be done
554 * atomically all at once or in several stages. The operation requests
555 * a credit for a number of buffer modifications in advance, but can
556 * extend its credit if it needs more.
558 * jbd2_journal_extend tries to give the running handle more buffer credits.
559 * It does not guarantee that allocation - this is a best-effort only.
560 * The calling process MUST be able to deal cleanly with a failure to
563 * Return 0 on success, non-zero on failure.
565 * return code < 0 implies an error
566 * return code > 0 implies normal transaction-full status.
568 int jbd2_journal_extend(handle_t *handle, int nblocks)
570 transaction_t *transaction = handle->h_transaction;
575 if (is_handle_aborted(handle))
577 journal = transaction->t_journal;
581 read_lock(&journal->j_state_lock);
583 /* Don't extend a locked-down transaction! */
584 if (transaction->t_state != T_RUNNING) {
585 jbd_debug(3, "denied handle %p %d blocks: "
586 "transaction not running\n", handle, nblocks);
590 spin_lock(&transaction->t_handle_lock);
591 wanted = atomic_add_return(nblocks,
592 &transaction->t_outstanding_credits);
594 if (wanted > journal->j_max_transaction_buffers) {
595 jbd_debug(3, "denied handle %p %d blocks: "
596 "transaction too large\n", handle, nblocks);
597 atomic_sub(nblocks, &transaction->t_outstanding_credits);
601 if (wanted + (wanted >> JBD2_CONTROL_BLOCKS_SHIFT) >
602 jbd2_log_space_left(journal)) {
603 jbd_debug(3, "denied handle %p %d blocks: "
604 "insufficient log space\n", handle, nblocks);
605 atomic_sub(nblocks, &transaction->t_outstanding_credits);
609 trace_jbd2_handle_extend(journal->j_fs_dev->bd_dev,
611 handle->h_type, handle->h_line_no,
612 handle->h_buffer_credits,
615 handle->h_buffer_credits += nblocks;
616 handle->h_requested_credits += nblocks;
619 jbd_debug(3, "extended handle %p by %d\n", handle, nblocks);
621 spin_unlock(&transaction->t_handle_lock);
623 read_unlock(&journal->j_state_lock);
629 * int jbd2_journal_restart() - restart a handle .
630 * @handle: handle to restart
631 * @nblocks: nr credits requested
632 * @gfp_mask: memory allocation flags (for start_this_handle)
634 * Restart a handle for a multi-transaction filesystem
637 * If the jbd2_journal_extend() call above fails to grant new buffer credits
638 * to a running handle, a call to jbd2_journal_restart will commit the
639 * handle's transaction so far and reattach the handle to a new
640 * transaction capable of guaranteeing the requested number of
641 * credits. We preserve reserved handle if there's any attached to the
644 int jbd2__journal_restart(handle_t *handle, int nblocks, gfp_t gfp_mask)
646 transaction_t *transaction = handle->h_transaction;
649 int need_to_start, ret;
651 /* If we've had an abort of any type, don't even think about
652 * actually doing the restart! */
653 if (is_handle_aborted(handle))
655 journal = transaction->t_journal;
658 * First unlink the handle from its current transaction, and start the
661 J_ASSERT(atomic_read(&transaction->t_updates) > 0);
662 J_ASSERT(journal_current_handle() == handle);
664 read_lock(&journal->j_state_lock);
665 spin_lock(&transaction->t_handle_lock);
666 atomic_sub(handle->h_buffer_credits,
667 &transaction->t_outstanding_credits);
668 if (handle->h_rsv_handle) {
669 sub_reserved_credits(journal,
670 handle->h_rsv_handle->h_buffer_credits);
672 if (atomic_dec_and_test(&transaction->t_updates))
673 wake_up(&journal->j_wait_updates);
674 tid = transaction->t_tid;
675 spin_unlock(&transaction->t_handle_lock);
676 handle->h_transaction = NULL;
677 current->journal_info = NULL;
679 jbd_debug(2, "restarting handle %p\n", handle);
680 need_to_start = !tid_geq(journal->j_commit_request, tid);
681 read_unlock(&journal->j_state_lock);
683 jbd2_log_start_commit(journal, tid);
685 rwsem_release(&journal->j_trans_commit_map, 1, _THIS_IP_);
686 handle->h_buffer_credits = nblocks;
688 * Restore the original nofs context because the journal restart
689 * is basically the same thing as journal stop and start.
690 * start_this_handle will start a new nofs context.
692 memalloc_nofs_restore(handle->saved_alloc_context);
693 ret = start_this_handle(journal, handle, gfp_mask);
696 EXPORT_SYMBOL(jbd2__journal_restart);
699 int jbd2_journal_restart(handle_t *handle, int nblocks)
701 return jbd2__journal_restart(handle, nblocks, GFP_NOFS);
703 EXPORT_SYMBOL(jbd2_journal_restart);
706 * void jbd2_journal_lock_updates () - establish a transaction barrier.
707 * @journal: Journal to establish a barrier on.
709 * This locks out any further updates from being started, and blocks
710 * until all existing updates have completed, returning only once the
711 * journal is in a quiescent state with no updates running.
713 * The journal lock should not be held on entry.
715 void jbd2_journal_lock_updates(journal_t *journal)
719 jbd2_might_wait_for_commit(journal);
721 write_lock(&journal->j_state_lock);
722 ++journal->j_barrier_count;
724 /* Wait until there are no reserved handles */
725 if (atomic_read(&journal->j_reserved_credits)) {
726 write_unlock(&journal->j_state_lock);
727 wait_event(journal->j_wait_reserved,
728 atomic_read(&journal->j_reserved_credits) == 0);
729 write_lock(&journal->j_state_lock);
732 /* Wait until there are no running updates */
734 transaction_t *transaction = journal->j_running_transaction;
739 spin_lock(&transaction->t_handle_lock);
740 prepare_to_wait(&journal->j_wait_updates, &wait,
741 TASK_UNINTERRUPTIBLE);
742 if (!atomic_read(&transaction->t_updates)) {
743 spin_unlock(&transaction->t_handle_lock);
744 finish_wait(&journal->j_wait_updates, &wait);
747 spin_unlock(&transaction->t_handle_lock);
748 write_unlock(&journal->j_state_lock);
750 finish_wait(&journal->j_wait_updates, &wait);
751 write_lock(&journal->j_state_lock);
753 write_unlock(&journal->j_state_lock);
756 * We have now established a barrier against other normal updates, but
757 * we also need to barrier against other jbd2_journal_lock_updates() calls
758 * to make sure that we serialise special journal-locked operations
761 mutex_lock(&journal->j_barrier);
765 * void jbd2_journal_unlock_updates (journal_t* journal) - release barrier
766 * @journal: Journal to release the barrier on.
768 * Release a transaction barrier obtained with jbd2_journal_lock_updates().
770 * Should be called without the journal lock held.
772 void jbd2_journal_unlock_updates (journal_t *journal)
774 J_ASSERT(journal->j_barrier_count != 0);
776 mutex_unlock(&journal->j_barrier);
777 write_lock(&journal->j_state_lock);
778 --journal->j_barrier_count;
779 write_unlock(&journal->j_state_lock);
780 wake_up(&journal->j_wait_transaction_locked);
783 static void warn_dirty_buffer(struct buffer_head *bh)
786 "JBD2: Spotted dirty metadata buffer (dev = %pg, blocknr = %llu). "
787 "There's a risk of filesystem corruption in case of system "
789 bh->b_bdev, (unsigned long long)bh->b_blocknr);
792 /* Call t_frozen trigger and copy buffer data into jh->b_frozen_data. */
793 static void jbd2_freeze_jh_data(struct journal_head *jh)
798 struct buffer_head *bh = jh2bh(jh);
800 J_EXPECT_JH(jh, buffer_uptodate(bh), "Possible IO failure.\n");
802 offset = offset_in_page(bh->b_data);
803 source = kmap_atomic(page);
804 /* Fire data frozen trigger just before we copy the data */
805 jbd2_buffer_frozen_trigger(jh, source + offset, jh->b_triggers);
806 memcpy(jh->b_frozen_data, source + offset, bh->b_size);
807 kunmap_atomic(source);
810 * Now that the frozen data is saved off, we need to store any matching
813 jh->b_frozen_triggers = jh->b_triggers;
817 * If the buffer is already part of the current transaction, then there
818 * is nothing we need to do. If it is already part of a prior
819 * transaction which we are still committing to disk, then we need to
820 * make sure that we do not overwrite the old copy: we do copy-out to
821 * preserve the copy going to disk. We also account the buffer against
822 * the handle's metadata buffer credits (unless the buffer is already
823 * part of the transaction, that is).
827 do_get_write_access(handle_t *handle, struct journal_head *jh,
830 struct buffer_head *bh;
831 transaction_t *transaction = handle->h_transaction;
834 char *frozen_buffer = NULL;
835 unsigned long start_lock, time_lock;
837 if (is_handle_aborted(handle))
839 journal = transaction->t_journal;
841 jbd_debug(5, "journal_head %p, force_copy %d\n", jh, force_copy);
843 JBUFFER_TRACE(jh, "entry");
847 /* @@@ Need to check for errors here at some point. */
849 start_lock = jiffies;
851 jbd_lock_bh_state(bh);
853 /* If it takes too long to lock the buffer, trace it */
854 time_lock = jbd2_time_diff(start_lock, jiffies);
855 if (time_lock > HZ/10)
856 trace_jbd2_lock_buffer_stall(bh->b_bdev->bd_dev,
857 jiffies_to_msecs(time_lock));
859 /* We now hold the buffer lock so it is safe to query the buffer
860 * state. Is the buffer dirty?
862 * If so, there are two possibilities. The buffer may be
863 * non-journaled, and undergoing a quite legitimate writeback.
864 * Otherwise, it is journaled, and we don't expect dirty buffers
865 * in that state (the buffers should be marked JBD_Dirty
866 * instead.) So either the IO is being done under our own
867 * control and this is a bug, or it's a third party IO such as
868 * dump(8) (which may leave the buffer scheduled for read ---
869 * ie. locked but not dirty) or tune2fs (which may actually have
870 * the buffer dirtied, ugh.) */
872 if (buffer_dirty(bh)) {
874 * First question: is this buffer already part of the current
875 * transaction or the existing committing transaction?
877 if (jh->b_transaction) {
879 jh->b_transaction == transaction ||
881 journal->j_committing_transaction);
882 if (jh->b_next_transaction)
883 J_ASSERT_JH(jh, jh->b_next_transaction ==
885 warn_dirty_buffer(bh);
888 * In any case we need to clean the dirty flag and we must
889 * do it under the buffer lock to be sure we don't race
890 * with running write-out.
892 JBUFFER_TRACE(jh, "Journalling dirty buffer");
893 clear_buffer_dirty(bh);
894 set_buffer_jbddirty(bh);
900 if (is_handle_aborted(handle)) {
901 jbd_unlock_bh_state(bh);
907 * The buffer is already part of this transaction if b_transaction or
908 * b_next_transaction points to it
910 if (jh->b_transaction == transaction ||
911 jh->b_next_transaction == transaction)
915 * this is the first time this transaction is touching this buffer,
916 * reset the modified flag
921 * If the buffer is not journaled right now, we need to make sure it
922 * doesn't get written to disk before the caller actually commits the
925 if (!jh->b_transaction) {
926 JBUFFER_TRACE(jh, "no transaction");
927 J_ASSERT_JH(jh, !jh->b_next_transaction);
928 JBUFFER_TRACE(jh, "file as BJ_Reserved");
930 * Make sure all stores to jh (b_modified, b_frozen_data) are
931 * visible before attaching it to the running transaction.
932 * Paired with barrier in jbd2_write_access_granted()
935 spin_lock(&journal->j_list_lock);
936 __jbd2_journal_file_buffer(jh, transaction, BJ_Reserved);
937 spin_unlock(&journal->j_list_lock);
941 * If there is already a copy-out version of this buffer, then we don't
942 * need to make another one
944 if (jh->b_frozen_data) {
945 JBUFFER_TRACE(jh, "has frozen data");
946 J_ASSERT_JH(jh, jh->b_next_transaction == NULL);
950 JBUFFER_TRACE(jh, "owned by older transaction");
951 J_ASSERT_JH(jh, jh->b_next_transaction == NULL);
952 J_ASSERT_JH(jh, jh->b_transaction == journal->j_committing_transaction);
955 * There is one case we have to be very careful about. If the
956 * committing transaction is currently writing this buffer out to disk
957 * and has NOT made a copy-out, then we cannot modify the buffer
958 * contents at all right now. The essence of copy-out is that it is
959 * the extra copy, not the primary copy, which gets journaled. If the
960 * primary copy is already going to disk then we cannot do copy-out
963 if (buffer_shadow(bh)) {
964 JBUFFER_TRACE(jh, "on shadow: sleep");
965 jbd_unlock_bh_state(bh);
966 wait_on_bit_io(&bh->b_state, BH_Shadow, TASK_UNINTERRUPTIBLE);
971 * Only do the copy if the currently-owning transaction still needs it.
972 * If buffer isn't on BJ_Metadata list, the committing transaction is
973 * past that stage (here we use the fact that BH_Shadow is set under
974 * bh_state lock together with refiling to BJ_Shadow list and at this
975 * point we know the buffer doesn't have BH_Shadow set).
977 * Subtle point, though: if this is a get_undo_access, then we will be
978 * relying on the frozen_data to contain the new value of the
979 * committed_data record after the transaction, so we HAVE to force the
980 * frozen_data copy in that case.
982 if (jh->b_jlist == BJ_Metadata || force_copy) {
983 JBUFFER_TRACE(jh, "generate frozen data");
984 if (!frozen_buffer) {
985 JBUFFER_TRACE(jh, "allocate memory for buffer");
986 jbd_unlock_bh_state(bh);
987 frozen_buffer = jbd2_alloc(jh2bh(jh)->b_size,
988 GFP_NOFS | __GFP_NOFAIL);
991 jh->b_frozen_data = frozen_buffer;
992 frozen_buffer = NULL;
993 jbd2_freeze_jh_data(jh);
997 * Make sure all stores to jh (b_modified, b_frozen_data) are visible
998 * before attaching it to the running transaction. Paired with barrier
999 * in jbd2_write_access_granted()
1002 jh->b_next_transaction = transaction;
1005 jbd_unlock_bh_state(bh);
1008 * If we are about to journal a buffer, then any revoke pending on it is
1011 jbd2_journal_cancel_revoke(handle, jh);
1014 if (unlikely(frozen_buffer)) /* It's usually NULL */
1015 jbd2_free(frozen_buffer, bh->b_size);
1017 JBUFFER_TRACE(jh, "exit");
1021 /* Fast check whether buffer is already attached to the required transaction */
1022 static bool jbd2_write_access_granted(handle_t *handle, struct buffer_head *bh,
1025 struct journal_head *jh;
1028 /* Dirty buffers require special handling... */
1029 if (buffer_dirty(bh))
1033 * RCU protects us from dereferencing freed pages. So the checks we do
1034 * are guaranteed not to oops. However the jh slab object can get freed
1035 * & reallocated while we work with it. So we have to be careful. When
1036 * we see jh attached to the running transaction, we know it must stay
1037 * so until the transaction is committed. Thus jh won't be freed and
1038 * will be attached to the same bh while we run. However it can
1039 * happen jh gets freed, reallocated, and attached to the transaction
1040 * just after we get pointer to it from bh. So we have to be careful
1041 * and recheck jh still belongs to our bh before we return success.
1044 if (!buffer_jbd(bh))
1046 /* This should be bh2jh() but that doesn't work with inline functions */
1047 jh = READ_ONCE(bh->b_private);
1050 /* For undo access buffer must have data copied */
1051 if (undo && !jh->b_committed_data)
1053 if (READ_ONCE(jh->b_transaction) != handle->h_transaction &&
1054 READ_ONCE(jh->b_next_transaction) != handle->h_transaction)
1057 * There are two reasons for the barrier here:
1058 * 1) Make sure to fetch b_bh after we did previous checks so that we
1059 * detect when jh went through free, realloc, attach to transaction
1060 * while we were checking. Paired with implicit barrier in that path.
1061 * 2) So that access to bh done after jbd2_write_access_granted()
1062 * doesn't get reordered and see inconsistent state of concurrent
1063 * do_get_write_access().
1066 if (unlikely(jh->b_bh != bh))
1075 * int jbd2_journal_get_write_access() - notify intent to modify a buffer for metadata (not data) update.
1076 * @handle: transaction to add buffer modifications to
1077 * @bh: bh to be used for metadata writes
1079 * Returns: error code or 0 on success.
1081 * In full data journalling mode the buffer may be of type BJ_AsyncData,
1082 * because we're ``write()ing`` a buffer which is also part of a shared mapping.
1085 int jbd2_journal_get_write_access(handle_t *handle, struct buffer_head *bh)
1087 struct journal_head *jh;
1090 if (jbd2_write_access_granted(handle, bh, false))
1093 jh = jbd2_journal_add_journal_head(bh);
1094 /* We do not want to get caught playing with fields which the
1095 * log thread also manipulates. Make sure that the buffer
1096 * completes any outstanding IO before proceeding. */
1097 rc = do_get_write_access(handle, jh, 0);
1098 jbd2_journal_put_journal_head(jh);
1104 * When the user wants to journal a newly created buffer_head
1105 * (ie. getblk() returned a new buffer and we are going to populate it
1106 * manually rather than reading off disk), then we need to keep the
1107 * buffer_head locked until it has been completely filled with new
1108 * data. In this case, we should be able to make the assertion that
1109 * the bh is not already part of an existing transaction.
1111 * The buffer should already be locked by the caller by this point.
1112 * There is no lock ranking violation: it was a newly created,
1113 * unlocked buffer beforehand. */
1116 * int jbd2_journal_get_create_access () - notify intent to use newly created bh
1117 * @handle: transaction to new buffer to
1120 * Call this if you create a new bh.
1122 int jbd2_journal_get_create_access(handle_t *handle, struct buffer_head *bh)
1124 transaction_t *transaction = handle->h_transaction;
1126 struct journal_head *jh = jbd2_journal_add_journal_head(bh);
1129 jbd_debug(5, "journal_head %p\n", jh);
1131 if (is_handle_aborted(handle))
1133 journal = transaction->t_journal;
1136 JBUFFER_TRACE(jh, "entry");
1138 * The buffer may already belong to this transaction due to pre-zeroing
1139 * in the filesystem's new_block code. It may also be on the previous,
1140 * committing transaction's lists, but it HAS to be in Forget state in
1141 * that case: the transaction must have deleted the buffer for it to be
1144 jbd_lock_bh_state(bh);
1145 J_ASSERT_JH(jh, (jh->b_transaction == transaction ||
1146 jh->b_transaction == NULL ||
1147 (jh->b_transaction == journal->j_committing_transaction &&
1148 jh->b_jlist == BJ_Forget)));
1150 J_ASSERT_JH(jh, jh->b_next_transaction == NULL);
1151 J_ASSERT_JH(jh, buffer_locked(jh2bh(jh)));
1153 if (jh->b_transaction == NULL) {
1155 * Previous jbd2_journal_forget() could have left the buffer
1156 * with jbddirty bit set because it was being committed. When
1157 * the commit finished, we've filed the buffer for
1158 * checkpointing and marked it dirty. Now we are reallocating
1159 * the buffer so the transaction freeing it must have
1160 * committed and so it's safe to clear the dirty bit.
1162 clear_buffer_dirty(jh2bh(jh));
1163 /* first access by this transaction */
1166 JBUFFER_TRACE(jh, "file as BJ_Reserved");
1167 spin_lock(&journal->j_list_lock);
1168 __jbd2_journal_file_buffer(jh, transaction, BJ_Reserved);
1169 spin_unlock(&journal->j_list_lock);
1170 } else if (jh->b_transaction == journal->j_committing_transaction) {
1171 /* first access by this transaction */
1174 JBUFFER_TRACE(jh, "set next transaction");
1175 spin_lock(&journal->j_list_lock);
1176 jh->b_next_transaction = transaction;
1177 spin_unlock(&journal->j_list_lock);
1179 jbd_unlock_bh_state(bh);
1182 * akpm: I added this. ext3_alloc_branch can pick up new indirect
1183 * blocks which contain freed but then revoked metadata. We need
1184 * to cancel the revoke in case we end up freeing it yet again
1185 * and the reallocating as data - this would cause a second revoke,
1186 * which hits an assertion error.
1188 JBUFFER_TRACE(jh, "cancelling revoke");
1189 jbd2_journal_cancel_revoke(handle, jh);
1191 jbd2_journal_put_journal_head(jh);
1196 * int jbd2_journal_get_undo_access() - Notify intent to modify metadata with
1197 * non-rewindable consequences
1198 * @handle: transaction
1199 * @bh: buffer to undo
1201 * Sometimes there is a need to distinguish between metadata which has
1202 * been committed to disk and that which has not. The ext3fs code uses
1203 * this for freeing and allocating space, we have to make sure that we
1204 * do not reuse freed space until the deallocation has been committed,
1205 * since if we overwrote that space we would make the delete
1206 * un-rewindable in case of a crash.
1208 * To deal with that, jbd2_journal_get_undo_access requests write access to a
1209 * buffer for parts of non-rewindable operations such as delete
1210 * operations on the bitmaps. The journaling code must keep a copy of
1211 * the buffer's contents prior to the undo_access call until such time
1212 * as we know that the buffer has definitely been committed to disk.
1214 * We never need to know which transaction the committed data is part
1215 * of, buffers touched here are guaranteed to be dirtied later and so
1216 * will be committed to a new transaction in due course, at which point
1217 * we can discard the old committed data pointer.
1219 * Returns error number or 0 on success.
1221 int jbd2_journal_get_undo_access(handle_t *handle, struct buffer_head *bh)
1224 struct journal_head *jh;
1225 char *committed_data = NULL;
1227 if (jbd2_write_access_granted(handle, bh, true))
1230 jh = jbd2_journal_add_journal_head(bh);
1231 JBUFFER_TRACE(jh, "entry");
1234 * Do this first --- it can drop the journal lock, so we want to
1235 * make sure that obtaining the committed_data is done
1236 * atomically wrt. completion of any outstanding commits.
1238 err = do_get_write_access(handle, jh, 1);
1243 if (!jh->b_committed_data)
1244 committed_data = jbd2_alloc(jh2bh(jh)->b_size,
1245 GFP_NOFS|__GFP_NOFAIL);
1247 jbd_lock_bh_state(bh);
1248 if (!jh->b_committed_data) {
1249 /* Copy out the current buffer contents into the
1250 * preserved, committed copy. */
1251 JBUFFER_TRACE(jh, "generate b_committed data");
1252 if (!committed_data) {
1253 jbd_unlock_bh_state(bh);
1257 jh->b_committed_data = committed_data;
1258 committed_data = NULL;
1259 memcpy(jh->b_committed_data, bh->b_data, bh->b_size);
1261 jbd_unlock_bh_state(bh);
1263 jbd2_journal_put_journal_head(jh);
1264 if (unlikely(committed_data))
1265 jbd2_free(committed_data, bh->b_size);
1270 * void jbd2_journal_set_triggers() - Add triggers for commit writeout
1271 * @bh: buffer to trigger on
1272 * @type: struct jbd2_buffer_trigger_type containing the trigger(s).
1274 * Set any triggers on this journal_head. This is always safe, because
1275 * triggers for a committing buffer will be saved off, and triggers for
1276 * a running transaction will match the buffer in that transaction.
1278 * Call with NULL to clear the triggers.
1280 void jbd2_journal_set_triggers(struct buffer_head *bh,
1281 struct jbd2_buffer_trigger_type *type)
1283 struct journal_head *jh = jbd2_journal_grab_journal_head(bh);
1287 jh->b_triggers = type;
1288 jbd2_journal_put_journal_head(jh);
1291 void jbd2_buffer_frozen_trigger(struct journal_head *jh, void *mapped_data,
1292 struct jbd2_buffer_trigger_type *triggers)
1294 struct buffer_head *bh = jh2bh(jh);
1296 if (!triggers || !triggers->t_frozen)
1299 triggers->t_frozen(triggers, bh, mapped_data, bh->b_size);
1302 void jbd2_buffer_abort_trigger(struct journal_head *jh,
1303 struct jbd2_buffer_trigger_type *triggers)
1305 if (!triggers || !triggers->t_abort)
1308 triggers->t_abort(triggers, jh2bh(jh));
1312 * int jbd2_journal_dirty_metadata() - mark a buffer as containing dirty metadata
1313 * @handle: transaction to add buffer to.
1314 * @bh: buffer to mark
1316 * mark dirty metadata which needs to be journaled as part of the current
1319 * The buffer must have previously had jbd2_journal_get_write_access()
1320 * called so that it has a valid journal_head attached to the buffer
1323 * The buffer is placed on the transaction's metadata list and is marked
1324 * as belonging to the transaction.
1326 * Returns error number or 0 on success.
1328 * Special care needs to be taken if the buffer already belongs to the
1329 * current committing transaction (in which case we should have frozen
1330 * data present for that commit). In that case, we don't relink the
1331 * buffer: that only gets done when the old transaction finally
1332 * completes its commit.
1334 int jbd2_journal_dirty_metadata(handle_t *handle, struct buffer_head *bh)
1336 transaction_t *transaction = handle->h_transaction;
1338 struct journal_head *jh;
1341 if (!buffer_jbd(bh))
1345 * We don't grab jh reference here since the buffer must be part
1346 * of the running transaction.
1349 jbd_debug(5, "journal_head %p\n", jh);
1350 JBUFFER_TRACE(jh, "entry");
1353 * This and the following assertions are unreliable since we may see jh
1354 * in inconsistent state unless we grab bh_state lock. But this is
1355 * crucial to catch bugs so let's do a reliable check until the
1356 * lockless handling is fully proven.
1358 if (jh->b_transaction != transaction &&
1359 jh->b_next_transaction != transaction) {
1360 jbd_lock_bh_state(bh);
1361 J_ASSERT_JH(jh, jh->b_transaction == transaction ||
1362 jh->b_next_transaction == transaction);
1363 jbd_unlock_bh_state(bh);
1365 if (jh->b_modified == 1) {
1366 /* If it's in our transaction it must be in BJ_Metadata list. */
1367 if (jh->b_transaction == transaction &&
1368 jh->b_jlist != BJ_Metadata) {
1369 jbd_lock_bh_state(bh);
1370 if (jh->b_transaction == transaction &&
1371 jh->b_jlist != BJ_Metadata)
1372 pr_err("JBD2: assertion failure: h_type=%u "
1373 "h_line_no=%u block_no=%llu jlist=%u\n",
1374 handle->h_type, handle->h_line_no,
1375 (unsigned long long) bh->b_blocknr,
1377 J_ASSERT_JH(jh, jh->b_transaction != transaction ||
1378 jh->b_jlist == BJ_Metadata);
1379 jbd_unlock_bh_state(bh);
1384 journal = transaction->t_journal;
1385 jbd_lock_bh_state(bh);
1387 if (is_handle_aborted(handle)) {
1389 * Check journal aborting with @jh->b_state_lock locked,
1390 * since 'jh->b_transaction' could be replaced with
1391 * 'jh->b_next_transaction' during old transaction
1392 * committing if journal aborted, which may fail
1393 * assertion on 'jh->b_frozen_data == NULL'.
1399 if (jh->b_modified == 0) {
1401 * This buffer's got modified and becoming part
1402 * of the transaction. This needs to be done
1403 * once a transaction -bzzz
1405 if (handle->h_buffer_credits <= 0) {
1410 handle->h_buffer_credits--;
1414 * fastpath, to avoid expensive locking. If this buffer is already
1415 * on the running transaction's metadata list there is nothing to do.
1416 * Nobody can take it off again because there is a handle open.
1417 * I _think_ we're OK here with SMP barriers - a mistaken decision will
1418 * result in this test being false, so we go in and take the locks.
1420 if (jh->b_transaction == transaction && jh->b_jlist == BJ_Metadata) {
1421 JBUFFER_TRACE(jh, "fastpath");
1422 if (unlikely(jh->b_transaction !=
1423 journal->j_running_transaction)) {
1424 printk(KERN_ERR "JBD2: %s: "
1425 "jh->b_transaction (%llu, %p, %u) != "
1426 "journal->j_running_transaction (%p, %u)\n",
1428 (unsigned long long) bh->b_blocknr,
1430 jh->b_transaction ? jh->b_transaction->t_tid : 0,
1431 journal->j_running_transaction,
1432 journal->j_running_transaction ?
1433 journal->j_running_transaction->t_tid : 0);
1439 set_buffer_jbddirty(bh);
1442 * Metadata already on the current transaction list doesn't
1443 * need to be filed. Metadata on another transaction's list must
1444 * be committing, and will be refiled once the commit completes:
1445 * leave it alone for now.
1447 if (jh->b_transaction != transaction) {
1448 JBUFFER_TRACE(jh, "already on other transaction");
1449 if (unlikely(((jh->b_transaction !=
1450 journal->j_committing_transaction)) ||
1451 (jh->b_next_transaction != transaction))) {
1452 printk(KERN_ERR "jbd2_journal_dirty_metadata: %s: "
1453 "bad jh for block %llu: "
1454 "transaction (%p, %u), "
1455 "jh->b_transaction (%p, %u), "
1456 "jh->b_next_transaction (%p, %u), jlist %u\n",
1458 (unsigned long long) bh->b_blocknr,
1459 transaction, transaction->t_tid,
1462 jh->b_transaction->t_tid : 0,
1463 jh->b_next_transaction,
1464 jh->b_next_transaction ?
1465 jh->b_next_transaction->t_tid : 0,
1470 /* And this case is illegal: we can't reuse another
1471 * transaction's data buffer, ever. */
1475 /* That test should have eliminated the following case: */
1476 J_ASSERT_JH(jh, jh->b_frozen_data == NULL);
1478 JBUFFER_TRACE(jh, "file as BJ_Metadata");
1479 spin_lock(&journal->j_list_lock);
1480 __jbd2_journal_file_buffer(jh, transaction, BJ_Metadata);
1481 spin_unlock(&journal->j_list_lock);
1483 jbd_unlock_bh_state(bh);
1485 JBUFFER_TRACE(jh, "exit");
1490 * void jbd2_journal_forget() - bforget() for potentially-journaled buffers.
1491 * @handle: transaction handle
1492 * @bh: bh to 'forget'
1494 * We can only do the bforget if there are no commits pending against the
1495 * buffer. If the buffer is dirty in the current running transaction we
1496 * can safely unlink it.
1498 * bh may not be a journalled buffer at all - it may be a non-JBD
1499 * buffer which came off the hashtable. Check for this.
1501 * Decrements bh->b_count by one.
1503 * Allow this call even if the handle has aborted --- it may be part of
1504 * the caller's cleanup after an abort.
1506 int jbd2_journal_forget (handle_t *handle, struct buffer_head *bh)
1508 transaction_t *transaction = handle->h_transaction;
1510 struct journal_head *jh;
1511 int drop_reserve = 0;
1513 int was_modified = 0;
1515 if (is_handle_aborted(handle))
1517 journal = transaction->t_journal;
1519 BUFFER_TRACE(bh, "entry");
1521 jbd_lock_bh_state(bh);
1523 if (!buffer_jbd(bh))
1527 /* Critical error: attempting to delete a bitmap buffer, maybe?
1528 * Don't do any jbd operations, and return an error. */
1529 if (!J_EXPECT_JH(jh, !jh->b_committed_data,
1530 "inconsistent data on disk")) {
1535 /* keep track of whether or not this transaction modified us */
1536 was_modified = jh->b_modified;
1539 * The buffer's going from the transaction, we must drop
1540 * all references -bzzz
1544 if (jh->b_transaction == transaction) {
1545 J_ASSERT_JH(jh, !jh->b_frozen_data);
1547 /* If we are forgetting a buffer which is already part
1548 * of this transaction, then we can just drop it from
1549 * the transaction immediately. */
1550 clear_buffer_dirty(bh);
1551 clear_buffer_jbddirty(bh);
1553 JBUFFER_TRACE(jh, "belongs to current transaction: unfile");
1556 * we only want to drop a reference if this transaction
1557 * modified the buffer
1563 * We are no longer going to journal this buffer.
1564 * However, the commit of this transaction is still
1565 * important to the buffer: the delete that we are now
1566 * processing might obsolete an old log entry, so by
1567 * committing, we can satisfy the buffer's checkpoint.
1569 * So, if we have a checkpoint on the buffer, we should
1570 * now refile the buffer on our BJ_Forget list so that
1571 * we know to remove the checkpoint after we commit.
1574 spin_lock(&journal->j_list_lock);
1575 if (jh->b_cp_transaction) {
1576 __jbd2_journal_temp_unlink_buffer(jh);
1577 __jbd2_journal_file_buffer(jh, transaction, BJ_Forget);
1579 __jbd2_journal_unfile_buffer(jh);
1580 if (!buffer_jbd(bh)) {
1581 spin_unlock(&journal->j_list_lock);
1582 jbd_unlock_bh_state(bh);
1587 spin_unlock(&journal->j_list_lock);
1588 } else if (jh->b_transaction) {
1589 J_ASSERT_JH(jh, (jh->b_transaction ==
1590 journal->j_committing_transaction));
1591 /* However, if the buffer is still owned by a prior
1592 * (committing) transaction, we can't drop it yet... */
1593 JBUFFER_TRACE(jh, "belongs to older transaction");
1594 /* ... but we CAN drop it from the new transaction through
1595 * marking the buffer as freed and set j_next_transaction to
1596 * the new transaction, so that not only the commit code
1597 * knows it should clear dirty bits when it is done with the
1598 * buffer, but also the buffer can be checkpointed only
1599 * after the new transaction commits. */
1601 set_buffer_freed(bh);
1603 if (!jh->b_next_transaction) {
1604 spin_lock(&journal->j_list_lock);
1605 jh->b_next_transaction = transaction;
1606 spin_unlock(&journal->j_list_lock);
1608 J_ASSERT(jh->b_next_transaction == transaction);
1611 * only drop a reference if this transaction modified
1620 jbd_unlock_bh_state(bh);
1624 /* no need to reserve log space for this block -bzzz */
1625 handle->h_buffer_credits++;
1631 * int jbd2_journal_stop() - complete a transaction
1632 * @handle: transaction to complete.
1634 * All done for a particular handle.
1636 * There is not much action needed here. We just return any remaining
1637 * buffer credits to the transaction and remove the handle. The only
1638 * complication is that we need to start a commit operation if the
1639 * filesystem is marked for synchronous update.
1641 * jbd2_journal_stop itself will not usually return an error, but it may
1642 * do so in unusual circumstances. In particular, expect it to
1643 * return -EIO if a jbd2_journal_abort has been executed since the
1644 * transaction began.
1646 int jbd2_journal_stop(handle_t *handle)
1648 transaction_t *transaction = handle->h_transaction;
1650 int err = 0, wait_for_commit = 0;
1656 * Handle is already detached from the transaction so
1657 * there is nothing to do other than decrease a refcount,
1658 * or free the handle if refcount drops to zero
1660 if (--handle->h_ref > 0) {
1661 jbd_debug(4, "h_ref %d -> %d\n", handle->h_ref + 1,
1665 if (handle->h_rsv_handle)
1666 jbd2_free_handle(handle->h_rsv_handle);
1670 journal = transaction->t_journal;
1672 J_ASSERT(journal_current_handle() == handle);
1674 if (is_handle_aborted(handle))
1677 J_ASSERT(atomic_read(&transaction->t_updates) > 0);
1679 if (--handle->h_ref > 0) {
1680 jbd_debug(4, "h_ref %d -> %d\n", handle->h_ref + 1,
1685 jbd_debug(4, "Handle %p going down\n", handle);
1686 trace_jbd2_handle_stats(journal->j_fs_dev->bd_dev,
1688 handle->h_type, handle->h_line_no,
1689 jiffies - handle->h_start_jiffies,
1690 handle->h_sync, handle->h_requested_credits,
1691 (handle->h_requested_credits -
1692 handle->h_buffer_credits));
1695 * Implement synchronous transaction batching. If the handle
1696 * was synchronous, don't force a commit immediately. Let's
1697 * yield and let another thread piggyback onto this
1698 * transaction. Keep doing that while new threads continue to
1699 * arrive. It doesn't cost much - we're about to run a commit
1700 * and sleep on IO anyway. Speeds up many-threaded, many-dir
1701 * operations by 30x or more...
1703 * We try and optimize the sleep time against what the
1704 * underlying disk can do, instead of having a static sleep
1705 * time. This is useful for the case where our storage is so
1706 * fast that it is more optimal to go ahead and force a flush
1707 * and wait for the transaction to be committed than it is to
1708 * wait for an arbitrary amount of time for new writers to
1709 * join the transaction. We achieve this by measuring how
1710 * long it takes to commit a transaction, and compare it with
1711 * how long this transaction has been running, and if run time
1712 * < commit time then we sleep for the delta and commit. This
1713 * greatly helps super fast disks that would see slowdowns as
1714 * more threads started doing fsyncs.
1716 * But don't do this if this process was the most recent one
1717 * to perform a synchronous write. We do this to detect the
1718 * case where a single process is doing a stream of sync
1719 * writes. No point in waiting for joiners in that case.
1721 * Setting max_batch_time to 0 disables this completely.
1724 if (handle->h_sync && journal->j_last_sync_writer != pid &&
1725 journal->j_max_batch_time) {
1726 u64 commit_time, trans_time;
1728 journal->j_last_sync_writer = pid;
1730 read_lock(&journal->j_state_lock);
1731 commit_time = journal->j_average_commit_time;
1732 read_unlock(&journal->j_state_lock);
1734 trans_time = ktime_to_ns(ktime_sub(ktime_get(),
1735 transaction->t_start_time));
1737 commit_time = max_t(u64, commit_time,
1738 1000*journal->j_min_batch_time);
1739 commit_time = min_t(u64, commit_time,
1740 1000*journal->j_max_batch_time);
1742 if (trans_time < commit_time) {
1743 ktime_t expires = ktime_add_ns(ktime_get(),
1745 set_current_state(TASK_UNINTERRUPTIBLE);
1746 schedule_hrtimeout(&expires, HRTIMER_MODE_ABS);
1751 transaction->t_synchronous_commit = 1;
1752 current->journal_info = NULL;
1753 atomic_sub(handle->h_buffer_credits,
1754 &transaction->t_outstanding_credits);
1757 * If the handle is marked SYNC, we need to set another commit
1758 * going! We also want to force a commit if the current
1759 * transaction is occupying too much of the log, or if the
1760 * transaction is too old now.
1762 if (handle->h_sync ||
1763 (atomic_read(&transaction->t_outstanding_credits) >
1764 journal->j_max_transaction_buffers) ||
1765 time_after_eq(jiffies, transaction->t_expires)) {
1766 /* Do this even for aborted journals: an abort still
1767 * completes the commit thread, it just doesn't write
1768 * anything to disk. */
1770 jbd_debug(2, "transaction too old, requesting commit for "
1771 "handle %p\n", handle);
1772 /* This is non-blocking */
1773 jbd2_log_start_commit(journal, transaction->t_tid);
1776 * Special case: JBD2_SYNC synchronous updates require us
1777 * to wait for the commit to complete.
1779 if (handle->h_sync && !(current->flags & PF_MEMALLOC))
1780 wait_for_commit = 1;
1784 * Once we drop t_updates, if it goes to zero the transaction
1785 * could start committing on us and eventually disappear. So
1786 * once we do this, we must not dereference transaction
1789 tid = transaction->t_tid;
1790 if (atomic_dec_and_test(&transaction->t_updates)) {
1791 wake_up(&journal->j_wait_updates);
1792 if (journal->j_barrier_count)
1793 wake_up(&journal->j_wait_transaction_locked);
1796 rwsem_release(&journal->j_trans_commit_map, 1, _THIS_IP_);
1798 if (wait_for_commit)
1799 err = jbd2_log_wait_commit(journal, tid);
1801 if (handle->h_rsv_handle)
1802 jbd2_journal_free_reserved(handle->h_rsv_handle);
1805 * Scope of the GFP_NOFS context is over here and so we can restore the
1806 * original alloc context.
1808 memalloc_nofs_restore(handle->saved_alloc_context);
1809 jbd2_free_handle(handle);
1815 * List management code snippets: various functions for manipulating the
1816 * transaction buffer lists.
1821 * Append a buffer to a transaction list, given the transaction's list head
1824 * j_list_lock is held.
1826 * jbd_lock_bh_state(jh2bh(jh)) is held.
1830 __blist_add_buffer(struct journal_head **list, struct journal_head *jh)
1833 jh->b_tnext = jh->b_tprev = jh;
1836 /* Insert at the tail of the list to preserve order */
1837 struct journal_head *first = *list, *last = first->b_tprev;
1839 jh->b_tnext = first;
1840 last->b_tnext = first->b_tprev = jh;
1845 * Remove a buffer from a transaction list, given the transaction's list
1848 * Called with j_list_lock held, and the journal may not be locked.
1850 * jbd_lock_bh_state(jh2bh(jh)) is held.
1854 __blist_del_buffer(struct journal_head **list, struct journal_head *jh)
1857 *list = jh->b_tnext;
1861 jh->b_tprev->b_tnext = jh->b_tnext;
1862 jh->b_tnext->b_tprev = jh->b_tprev;
1866 * Remove a buffer from the appropriate transaction list.
1868 * Note that this function can *change* the value of
1869 * bh->b_transaction->t_buffers, t_forget, t_shadow_list, t_log_list or
1870 * t_reserved_list. If the caller is holding onto a copy of one of these
1871 * pointers, it could go bad. Generally the caller needs to re-read the
1872 * pointer from the transaction_t.
1874 * Called under j_list_lock.
1876 static void __jbd2_journal_temp_unlink_buffer(struct journal_head *jh)
1878 struct journal_head **list = NULL;
1879 transaction_t *transaction;
1880 struct buffer_head *bh = jh2bh(jh);
1882 J_ASSERT_JH(jh, jbd_is_locked_bh_state(bh));
1883 transaction = jh->b_transaction;
1885 assert_spin_locked(&transaction->t_journal->j_list_lock);
1887 J_ASSERT_JH(jh, jh->b_jlist < BJ_Types);
1888 if (jh->b_jlist != BJ_None)
1889 J_ASSERT_JH(jh, transaction != NULL);
1891 switch (jh->b_jlist) {
1895 transaction->t_nr_buffers--;
1896 J_ASSERT_JH(jh, transaction->t_nr_buffers >= 0);
1897 list = &transaction->t_buffers;
1900 list = &transaction->t_forget;
1903 list = &transaction->t_shadow_list;
1906 list = &transaction->t_reserved_list;
1910 __blist_del_buffer(list, jh);
1911 jh->b_jlist = BJ_None;
1912 if (transaction && is_journal_aborted(transaction->t_journal))
1913 clear_buffer_jbddirty(bh);
1914 else if (test_clear_buffer_jbddirty(bh))
1915 mark_buffer_dirty(bh); /* Expose it to the VM */
1919 * Remove buffer from all transactions.
1921 * Called with bh_state lock and j_list_lock
1923 * jh and bh may be already freed when this function returns.
1925 static void __jbd2_journal_unfile_buffer(struct journal_head *jh)
1927 J_ASSERT_JH(jh, jh->b_transaction != NULL);
1928 J_ASSERT_JH(jh, jh->b_next_transaction == NULL);
1930 __jbd2_journal_temp_unlink_buffer(jh);
1931 jh->b_transaction = NULL;
1932 jbd2_journal_put_journal_head(jh);
1935 void jbd2_journal_unfile_buffer(journal_t *journal, struct journal_head *jh)
1937 struct buffer_head *bh = jh2bh(jh);
1939 /* Get reference so that buffer cannot be freed before we unlock it */
1941 jbd_lock_bh_state(bh);
1942 spin_lock(&journal->j_list_lock);
1943 __jbd2_journal_unfile_buffer(jh);
1944 spin_unlock(&journal->j_list_lock);
1945 jbd_unlock_bh_state(bh);
1950 * Called from jbd2_journal_try_to_free_buffers().
1952 * Called under jbd_lock_bh_state(bh)
1955 __journal_try_to_free_buffer(journal_t *journal, struct buffer_head *bh)
1957 struct journal_head *jh;
1961 if (buffer_locked(bh) || buffer_dirty(bh))
1964 if (jh->b_next_transaction != NULL || jh->b_transaction != NULL)
1967 spin_lock(&journal->j_list_lock);
1968 if (jh->b_cp_transaction != NULL) {
1969 /* written-back checkpointed metadata buffer */
1970 JBUFFER_TRACE(jh, "remove from checkpoint list");
1971 __jbd2_journal_remove_checkpoint(jh);
1973 spin_unlock(&journal->j_list_lock);
1979 * int jbd2_journal_try_to_free_buffers() - try to free page buffers.
1980 * @journal: journal for operation
1981 * @page: to try and free
1982 * @gfp_mask: we use the mask to detect how hard should we try to release
1983 * buffers. If __GFP_DIRECT_RECLAIM and __GFP_FS is set, we wait for commit
1984 * code to release the buffers.
1987 * For all the buffers on this page,
1988 * if they are fully written out ordered data, move them onto BUF_CLEAN
1989 * so try_to_free_buffers() can reap them.
1991 * This function returns non-zero if we wish try_to_free_buffers()
1992 * to be called. We do this if the page is releasable by try_to_free_buffers().
1993 * We also do it if the page has locked or dirty buffers and the caller wants
1994 * us to perform sync or async writeout.
1996 * This complicates JBD locking somewhat. We aren't protected by the
1997 * BKL here. We wish to remove the buffer from its committing or
1998 * running transaction's ->t_datalist via __jbd2_journal_unfile_buffer.
2000 * This may *change* the value of transaction_t->t_datalist, so anyone
2001 * who looks at t_datalist needs to lock against this function.
2003 * Even worse, someone may be doing a jbd2_journal_dirty_data on this
2004 * buffer. So we need to lock against that. jbd2_journal_dirty_data()
2005 * will come out of the lock with the buffer dirty, which makes it
2006 * ineligible for release here.
2008 * Who else is affected by this? hmm... Really the only contender
2009 * is do_get_write_access() - it could be looking at the buffer while
2010 * journal_try_to_free_buffer() is changing its state. But that
2011 * cannot happen because we never reallocate freed data as metadata
2012 * while the data is part of a transaction. Yes?
2014 * Return 0 on failure, 1 on success
2016 int jbd2_journal_try_to_free_buffers(journal_t *journal,
2017 struct page *page, gfp_t gfp_mask)
2019 struct buffer_head *head;
2020 struct buffer_head *bh;
2021 bool has_write_io_error = false;
2024 J_ASSERT(PageLocked(page));
2026 head = page_buffers(page);
2029 struct journal_head *jh;
2032 * We take our own ref against the journal_head here to avoid
2033 * having to add tons of locking around each instance of
2034 * jbd2_journal_put_journal_head().
2036 jh = jbd2_journal_grab_journal_head(bh);
2040 jbd_lock_bh_state(bh);
2041 __journal_try_to_free_buffer(journal, bh);
2042 jbd2_journal_put_journal_head(jh);
2043 jbd_unlock_bh_state(bh);
2048 * If we free a metadata buffer which has been failed to
2049 * write out, the jbd2 checkpoint procedure will not detect
2050 * this failure and may lead to filesystem inconsistency
2051 * after cleanup journal tail.
2053 if (buffer_write_io_error(bh)) {
2054 pr_err("JBD2: Error while async write back metadata bh %llu.",
2055 (unsigned long long)bh->b_blocknr);
2056 has_write_io_error = true;
2058 } while ((bh = bh->b_this_page) != head);
2060 ret = try_to_free_buffers(page);
2063 if (has_write_io_error)
2064 jbd2_journal_abort(journal, -EIO);
2070 * This buffer is no longer needed. If it is on an older transaction's
2071 * checkpoint list we need to record it on this transaction's forget list
2072 * to pin this buffer (and hence its checkpointing transaction) down until
2073 * this transaction commits. If the buffer isn't on a checkpoint list, we
2075 * Returns non-zero if JBD no longer has an interest in the buffer.
2077 * Called under j_list_lock.
2079 * Called under jbd_lock_bh_state(bh).
2081 static int __dispose_buffer(struct journal_head *jh, transaction_t *transaction)
2084 struct buffer_head *bh = jh2bh(jh);
2086 if (jh->b_cp_transaction) {
2087 JBUFFER_TRACE(jh, "on running+cp transaction");
2088 __jbd2_journal_temp_unlink_buffer(jh);
2090 * We don't want to write the buffer anymore, clear the
2091 * bit so that we don't confuse checks in
2092 * __journal_file_buffer
2094 clear_buffer_dirty(bh);
2095 __jbd2_journal_file_buffer(jh, transaction, BJ_Forget);
2098 JBUFFER_TRACE(jh, "on running transaction");
2099 __jbd2_journal_unfile_buffer(jh);
2105 * jbd2_journal_invalidatepage
2107 * This code is tricky. It has a number of cases to deal with.
2109 * There are two invariants which this code relies on:
2111 * i_size must be updated on disk before we start calling invalidatepage on the
2114 * This is done in ext3 by defining an ext3_setattr method which
2115 * updates i_size before truncate gets going. By maintaining this
2116 * invariant, we can be sure that it is safe to throw away any buffers
2117 * attached to the current transaction: once the transaction commits,
2118 * we know that the data will not be needed.
2120 * Note however that we can *not* throw away data belonging to the
2121 * previous, committing transaction!
2123 * Any disk blocks which *are* part of the previous, committing
2124 * transaction (and which therefore cannot be discarded immediately) are
2125 * not going to be reused in the new running transaction
2127 * The bitmap committed_data images guarantee this: any block which is
2128 * allocated in one transaction and removed in the next will be marked
2129 * as in-use in the committed_data bitmap, so cannot be reused until
2130 * the next transaction to delete the block commits. This means that
2131 * leaving committing buffers dirty is quite safe: the disk blocks
2132 * cannot be reallocated to a different file and so buffer aliasing is
2136 * The above applies mainly to ordered data mode. In writeback mode we
2137 * don't make guarantees about the order in which data hits disk --- in
2138 * particular we don't guarantee that new dirty data is flushed before
2139 * transaction commit --- so it is always safe just to discard data
2140 * immediately in that mode. --sct
2144 * The journal_unmap_buffer helper function returns zero if the buffer
2145 * concerned remains pinned as an anonymous buffer belonging to an older
2148 * We're outside-transaction here. Either or both of j_running_transaction
2149 * and j_committing_transaction may be NULL.
2151 static int journal_unmap_buffer(journal_t *journal, struct buffer_head *bh,
2154 transaction_t *transaction;
2155 struct journal_head *jh;
2158 BUFFER_TRACE(bh, "entry");
2161 * It is safe to proceed here without the j_list_lock because the
2162 * buffers cannot be stolen by try_to_free_buffers as long as we are
2163 * holding the page lock. --sct
2166 if (!buffer_jbd(bh))
2167 goto zap_buffer_unlocked;
2169 /* OK, we have data buffer in journaled mode */
2170 write_lock(&journal->j_state_lock);
2171 jbd_lock_bh_state(bh);
2172 spin_lock(&journal->j_list_lock);
2174 jh = jbd2_journal_grab_journal_head(bh);
2176 goto zap_buffer_no_jh;
2179 * We cannot remove the buffer from checkpoint lists until the
2180 * transaction adding inode to orphan list (let's call it T)
2181 * is committed. Otherwise if the transaction changing the
2182 * buffer would be cleaned from the journal before T is
2183 * committed, a crash will cause that the correct contents of
2184 * the buffer will be lost. On the other hand we have to
2185 * clear the buffer dirty bit at latest at the moment when the
2186 * transaction marking the buffer as freed in the filesystem
2187 * structures is committed because from that moment on the
2188 * block can be reallocated and used by a different page.
2189 * Since the block hasn't been freed yet but the inode has
2190 * already been added to orphan list, it is safe for us to add
2191 * the buffer to BJ_Forget list of the newest transaction.
2193 * Also we have to clear buffer_mapped flag of a truncated buffer
2194 * because the buffer_head may be attached to the page straddling
2195 * i_size (can happen only when blocksize < pagesize) and thus the
2196 * buffer_head can be reused when the file is extended again. So we end
2197 * up keeping around invalidated buffers attached to transactions'
2198 * BJ_Forget list just to stop checkpointing code from cleaning up
2199 * the transaction this buffer was modified in.
2201 transaction = jh->b_transaction;
2202 if (transaction == NULL) {
2203 /* First case: not on any transaction. If it
2204 * has no checkpoint link, then we can zap it:
2205 * it's a writeback-mode buffer so we don't care
2206 * if it hits disk safely. */
2207 if (!jh->b_cp_transaction) {
2208 JBUFFER_TRACE(jh, "not on any transaction: zap");
2212 if (!buffer_dirty(bh)) {
2213 /* bdflush has written it. We can drop it now */
2214 __jbd2_journal_remove_checkpoint(jh);
2218 /* OK, it must be in the journal but still not
2219 * written fully to disk: it's metadata or
2220 * journaled data... */
2222 if (journal->j_running_transaction) {
2223 /* ... and once the current transaction has
2224 * committed, the buffer won't be needed any
2226 JBUFFER_TRACE(jh, "checkpointed: add to BJ_Forget");
2227 may_free = __dispose_buffer(jh,
2228 journal->j_running_transaction);
2231 /* There is no currently-running transaction. So the
2232 * orphan record which we wrote for this file must have
2233 * passed into commit. We must attach this buffer to
2234 * the committing transaction, if it exists. */
2235 if (journal->j_committing_transaction) {
2236 JBUFFER_TRACE(jh, "give to committing trans");
2237 may_free = __dispose_buffer(jh,
2238 journal->j_committing_transaction);
2241 /* The orphan record's transaction has
2242 * committed. We can cleanse this buffer */
2243 clear_buffer_jbddirty(bh);
2244 __jbd2_journal_remove_checkpoint(jh);
2248 } else if (transaction == journal->j_committing_transaction) {
2249 JBUFFER_TRACE(jh, "on committing transaction");
2251 * The buffer is committing, we simply cannot touch
2252 * it. If the page is straddling i_size we have to wait
2253 * for commit and try again.
2256 jbd2_journal_put_journal_head(jh);
2257 spin_unlock(&journal->j_list_lock);
2258 jbd_unlock_bh_state(bh);
2259 write_unlock(&journal->j_state_lock);
2263 * OK, buffer won't be reachable after truncate. We just clear
2264 * b_modified to not confuse transaction credit accounting, and
2265 * set j_next_transaction to the running transaction (if there
2266 * is one) and mark buffer as freed so that commit code knows
2267 * it should clear dirty bits when it is done with the buffer.
2269 set_buffer_freed(bh);
2270 if (journal->j_running_transaction && buffer_jbddirty(bh))
2271 jh->b_next_transaction = journal->j_running_transaction;
2273 jbd2_journal_put_journal_head(jh);
2274 spin_unlock(&journal->j_list_lock);
2275 jbd_unlock_bh_state(bh);
2276 write_unlock(&journal->j_state_lock);
2279 /* Good, the buffer belongs to the running transaction.
2280 * We are writing our own transaction's data, not any
2281 * previous one's, so it is safe to throw it away
2282 * (remember that we expect the filesystem to have set
2283 * i_size already for this truncate so recovery will not
2284 * expose the disk blocks we are discarding here.) */
2285 J_ASSERT_JH(jh, transaction == journal->j_running_transaction);
2286 JBUFFER_TRACE(jh, "on running transaction");
2287 may_free = __dispose_buffer(jh, transaction);
2292 * This is tricky. Although the buffer is truncated, it may be reused
2293 * if blocksize < pagesize and it is attached to the page straddling
2294 * EOF. Since the buffer might have been added to BJ_Forget list of the
2295 * running transaction, journal_get_write_access() won't clear
2296 * b_modified and credit accounting gets confused. So clear b_modified
2300 jbd2_journal_put_journal_head(jh);
2302 spin_unlock(&journal->j_list_lock);
2303 jbd_unlock_bh_state(bh);
2304 write_unlock(&journal->j_state_lock);
2305 zap_buffer_unlocked:
2306 clear_buffer_dirty(bh);
2307 J_ASSERT_BH(bh, !buffer_jbddirty(bh));
2308 clear_buffer_mapped(bh);
2309 clear_buffer_req(bh);
2310 clear_buffer_new(bh);
2311 clear_buffer_delay(bh);
2312 clear_buffer_unwritten(bh);
2318 * void jbd2_journal_invalidatepage()
2319 * @journal: journal to use for flush...
2320 * @page: page to flush
2321 * @offset: start of the range to invalidate
2322 * @length: length of the range to invalidate
2324 * Reap page buffers containing data after in the specified range in page.
2325 * Can return -EBUSY if buffers are part of the committing transaction and
2326 * the page is straddling i_size. Caller then has to wait for current commit
2329 int jbd2_journal_invalidatepage(journal_t *journal,
2331 unsigned int offset,
2332 unsigned int length)
2334 struct buffer_head *head, *bh, *next;
2335 unsigned int stop = offset + length;
2336 unsigned int curr_off = 0;
2337 int partial_page = (offset || length < PAGE_SIZE);
2341 if (!PageLocked(page))
2343 if (!page_has_buffers(page))
2346 BUG_ON(stop > PAGE_SIZE || stop < length);
2348 /* We will potentially be playing with lists other than just the
2349 * data lists (especially for journaled data mode), so be
2350 * cautious in our locking. */
2352 head = bh = page_buffers(page);
2354 unsigned int next_off = curr_off + bh->b_size;
2355 next = bh->b_this_page;
2357 if (next_off > stop)
2360 if (offset <= curr_off) {
2361 /* This block is wholly outside the truncation point */
2363 ret = journal_unmap_buffer(journal, bh, partial_page);
2369 curr_off = next_off;
2372 } while (bh != head);
2374 if (!partial_page) {
2375 if (may_free && try_to_free_buffers(page))
2376 J_ASSERT(!page_has_buffers(page));
2382 * File a buffer on the given transaction list.
2384 void __jbd2_journal_file_buffer(struct journal_head *jh,
2385 transaction_t *transaction, int jlist)
2387 struct journal_head **list = NULL;
2389 struct buffer_head *bh = jh2bh(jh);
2391 J_ASSERT_JH(jh, jbd_is_locked_bh_state(bh));
2392 assert_spin_locked(&transaction->t_journal->j_list_lock);
2394 J_ASSERT_JH(jh, jh->b_jlist < BJ_Types);
2395 J_ASSERT_JH(jh, jh->b_transaction == transaction ||
2396 jh->b_transaction == NULL);
2398 if (jh->b_transaction && jh->b_jlist == jlist)
2401 if (jlist == BJ_Metadata || jlist == BJ_Reserved ||
2402 jlist == BJ_Shadow || jlist == BJ_Forget) {
2404 * For metadata buffers, we track dirty bit in buffer_jbddirty
2405 * instead of buffer_dirty. We should not see a dirty bit set
2406 * here because we clear it in do_get_write_access but e.g.
2407 * tune2fs can modify the sb and set the dirty bit at any time
2408 * so we try to gracefully handle that.
2410 if (buffer_dirty(bh))
2411 warn_dirty_buffer(bh);
2412 if (test_clear_buffer_dirty(bh) ||
2413 test_clear_buffer_jbddirty(bh))
2417 if (jh->b_transaction)
2418 __jbd2_journal_temp_unlink_buffer(jh);
2420 jbd2_journal_grab_journal_head(bh);
2421 jh->b_transaction = transaction;
2425 J_ASSERT_JH(jh, !jh->b_committed_data);
2426 J_ASSERT_JH(jh, !jh->b_frozen_data);
2429 transaction->t_nr_buffers++;
2430 list = &transaction->t_buffers;
2433 list = &transaction->t_forget;
2436 list = &transaction->t_shadow_list;
2439 list = &transaction->t_reserved_list;
2443 __blist_add_buffer(list, jh);
2444 jh->b_jlist = jlist;
2447 set_buffer_jbddirty(bh);
2450 void jbd2_journal_file_buffer(struct journal_head *jh,
2451 transaction_t *transaction, int jlist)
2453 jbd_lock_bh_state(jh2bh(jh));
2454 spin_lock(&transaction->t_journal->j_list_lock);
2455 __jbd2_journal_file_buffer(jh, transaction, jlist);
2456 spin_unlock(&transaction->t_journal->j_list_lock);
2457 jbd_unlock_bh_state(jh2bh(jh));
2461 * Remove a buffer from its current buffer list in preparation for
2462 * dropping it from its current transaction entirely. If the buffer has
2463 * already started to be used by a subsequent transaction, refile the
2464 * buffer on that transaction's metadata list.
2466 * Called under j_list_lock
2467 * Called under jbd_lock_bh_state(jh2bh(jh))
2469 * jh and bh may be already free when this function returns
2471 void __jbd2_journal_refile_buffer(struct journal_head *jh)
2473 int was_dirty, jlist;
2474 struct buffer_head *bh = jh2bh(jh);
2476 J_ASSERT_JH(jh, jbd_is_locked_bh_state(bh));
2477 if (jh->b_transaction)
2478 assert_spin_locked(&jh->b_transaction->t_journal->j_list_lock);
2480 /* If the buffer is now unused, just drop it. */
2481 if (jh->b_next_transaction == NULL) {
2482 __jbd2_journal_unfile_buffer(jh);
2487 * It has been modified by a later transaction: add it to the new
2488 * transaction's metadata list.
2491 was_dirty = test_clear_buffer_jbddirty(bh);
2492 __jbd2_journal_temp_unlink_buffer(jh);
2495 * b_transaction must be set, otherwise the new b_transaction won't
2496 * be holding jh reference
2498 J_ASSERT_JH(jh, jh->b_transaction != NULL);
2501 * We set b_transaction here because b_next_transaction will inherit
2502 * our jh reference and thus __jbd2_journal_file_buffer() must not
2505 WRITE_ONCE(jh->b_transaction, jh->b_next_transaction);
2506 WRITE_ONCE(jh->b_next_transaction, NULL);
2507 if (buffer_freed(bh))
2509 else if (jh->b_modified)
2510 jlist = BJ_Metadata;
2512 jlist = BJ_Reserved;
2513 __jbd2_journal_file_buffer(jh, jh->b_transaction, jlist);
2514 J_ASSERT_JH(jh, jh->b_transaction->t_state == T_RUNNING);
2517 set_buffer_jbddirty(bh);
2521 * __jbd2_journal_refile_buffer() with necessary locking added. We take our
2522 * bh reference so that we can safely unlock bh.
2524 * The jh and bh may be freed by this call.
2526 void jbd2_journal_refile_buffer(journal_t *journal, struct journal_head *jh)
2528 struct buffer_head *bh = jh2bh(jh);
2530 /* Get reference so that buffer cannot be freed before we unlock it */
2532 jbd_lock_bh_state(bh);
2533 spin_lock(&journal->j_list_lock);
2534 __jbd2_journal_refile_buffer(jh);
2535 jbd_unlock_bh_state(bh);
2536 spin_unlock(&journal->j_list_lock);
2541 * File inode in the inode list of the handle's transaction
2543 static int jbd2_journal_file_inode(handle_t *handle, struct jbd2_inode *jinode,
2544 unsigned long flags, loff_t start_byte, loff_t end_byte)
2546 transaction_t *transaction = handle->h_transaction;
2549 if (is_handle_aborted(handle))
2551 journal = transaction->t_journal;
2553 jbd_debug(4, "Adding inode %lu, tid:%d\n", jinode->i_vfs_inode->i_ino,
2554 transaction->t_tid);
2556 spin_lock(&journal->j_list_lock);
2557 jinode->i_flags |= flags;
2559 if (jinode->i_dirty_end) {
2560 jinode->i_dirty_start = min(jinode->i_dirty_start, start_byte);
2561 jinode->i_dirty_end = max(jinode->i_dirty_end, end_byte);
2563 jinode->i_dirty_start = start_byte;
2564 jinode->i_dirty_end = end_byte;
2567 /* Is inode already attached where we need it? */
2568 if (jinode->i_transaction == transaction ||
2569 jinode->i_next_transaction == transaction)
2573 * We only ever set this variable to 1 so the test is safe. Since
2574 * t_need_data_flush is likely to be set, we do the test to save some
2575 * cacheline bouncing
2577 if (!transaction->t_need_data_flush)
2578 transaction->t_need_data_flush = 1;
2579 /* On some different transaction's list - should be
2580 * the committing one */
2581 if (jinode->i_transaction) {
2582 J_ASSERT(jinode->i_next_transaction == NULL);
2583 J_ASSERT(jinode->i_transaction ==
2584 journal->j_committing_transaction);
2585 jinode->i_next_transaction = transaction;
2588 /* Not on any transaction list... */
2589 J_ASSERT(!jinode->i_next_transaction);
2590 jinode->i_transaction = transaction;
2591 list_add(&jinode->i_list, &transaction->t_inode_list);
2593 spin_unlock(&journal->j_list_lock);
2598 int jbd2_journal_inode_add_write(handle_t *handle, struct jbd2_inode *jinode)
2600 return jbd2_journal_file_inode(handle, jinode,
2601 JI_WRITE_DATA | JI_WAIT_DATA, 0, LLONG_MAX);
2604 int jbd2_journal_inode_add_wait(handle_t *handle, struct jbd2_inode *jinode)
2606 return jbd2_journal_file_inode(handle, jinode, JI_WAIT_DATA, 0,
2610 int jbd2_journal_inode_ranged_write(handle_t *handle,
2611 struct jbd2_inode *jinode, loff_t start_byte, loff_t length)
2613 return jbd2_journal_file_inode(handle, jinode,
2614 JI_WRITE_DATA | JI_WAIT_DATA, start_byte,
2615 start_byte + length - 1);
2618 int jbd2_journal_inode_ranged_wait(handle_t *handle, struct jbd2_inode *jinode,
2619 loff_t start_byte, loff_t length)
2621 return jbd2_journal_file_inode(handle, jinode, JI_WAIT_DATA,
2622 start_byte, start_byte + length - 1);
2626 * File truncate and transaction commit interact with each other in a
2627 * non-trivial way. If a transaction writing data block A is
2628 * committing, we cannot discard the data by truncate until we have
2629 * written them. Otherwise if we crashed after the transaction with
2630 * write has committed but before the transaction with truncate has
2631 * committed, we could see stale data in block A. This function is a
2632 * helper to solve this problem. It starts writeout of the truncated
2633 * part in case it is in the committing transaction.
2635 * Filesystem code must call this function when inode is journaled in
2636 * ordered mode before truncation happens and after the inode has been
2637 * placed on orphan list with the new inode size. The second condition
2638 * avoids the race that someone writes new data and we start
2639 * committing the transaction after this function has been called but
2640 * before a transaction for truncate is started (and furthermore it
2641 * allows us to optimize the case where the addition to orphan list
2642 * happens in the same transaction as write --- we don't have to write
2643 * any data in such case).
2645 int jbd2_journal_begin_ordered_truncate(journal_t *journal,
2646 struct jbd2_inode *jinode,
2649 transaction_t *inode_trans, *commit_trans;
2652 /* This is a quick check to avoid locking if not necessary */
2653 if (!jinode->i_transaction)
2655 /* Locks are here just to force reading of recent values, it is
2656 * enough that the transaction was not committing before we started
2657 * a transaction adding the inode to orphan list */
2658 read_lock(&journal->j_state_lock);
2659 commit_trans = journal->j_committing_transaction;
2660 read_unlock(&journal->j_state_lock);
2661 spin_lock(&journal->j_list_lock);
2662 inode_trans = jinode->i_transaction;
2663 spin_unlock(&journal->j_list_lock);
2664 if (inode_trans == commit_trans) {
2665 ret = filemap_fdatawrite_range(jinode->i_vfs_inode->i_mapping,
2666 new_size, LLONG_MAX);
2668 jbd2_journal_abort(journal, ret);