2 * linux/fs/jbd2/transaction.c
4 * Written by Stephen C. Tweedie <sct@redhat.com>, 1998
6 * Copyright 1998 Red Hat corp --- All Rights Reserved
8 * This file is part of the Linux kernel and is made available under
9 * the terms of the GNU General Public License, version 2, or at your
10 * option, any later version, incorporated herein by reference.
12 * Generic filesystem transaction handling code; part of the ext2fs
15 * This file manages transactions (compound commits managed by the
16 * journaling code) and handles (individual atomic operations by the
20 #include <linux/time.h>
22 #include <linux/jbd2.h>
23 #include <linux/errno.h>
24 #include <linux/slab.h>
25 #include <linux/timer.h>
27 #include <linux/highmem.h>
28 #include <linux/hrtimer.h>
29 #include <linux/backing-dev.h>
30 #include <linux/bug.h>
31 #include <linux/module.h>
33 #include <trace/events/jbd2.h>
35 static void __jbd2_journal_temp_unlink_buffer(struct journal_head *jh);
36 static void __jbd2_journal_unfile_buffer(struct journal_head *jh);
38 static struct kmem_cache *transaction_cache;
39 int __init jbd2_journal_init_transaction_cache(void)
41 J_ASSERT(!transaction_cache);
42 transaction_cache = kmem_cache_create("jbd2_transaction_s",
43 sizeof(transaction_t),
45 SLAB_HWCACHE_ALIGN|SLAB_TEMPORARY,
47 if (transaction_cache)
52 void jbd2_journal_destroy_transaction_cache(void)
54 if (transaction_cache) {
55 kmem_cache_destroy(transaction_cache);
56 transaction_cache = NULL;
60 void jbd2_journal_free_transaction(transaction_t *transaction)
62 if (unlikely(ZERO_OR_NULL_PTR(transaction)))
64 kmem_cache_free(transaction_cache, transaction);
68 * jbd2_get_transaction: obtain a new transaction_t object.
70 * Simply allocate and initialise a new transaction. Create it in
71 * RUNNING state and add it to the current journal (which should not
72 * have an existing running transaction: we only make a new transaction
73 * once we have started to commit the old one).
76 * The journal MUST be locked. We don't perform atomic mallocs on the
77 * new transaction and we can't block without protecting against other
78 * processes trying to touch the journal while it is in transition.
82 static transaction_t *
83 jbd2_get_transaction(journal_t *journal, transaction_t *transaction)
85 transaction->t_journal = journal;
86 transaction->t_state = T_RUNNING;
87 transaction->t_start_time = ktime_get();
88 transaction->t_tid = journal->j_transaction_sequence++;
89 transaction->t_expires = jiffies + journal->j_commit_interval;
90 spin_lock_init(&transaction->t_handle_lock);
91 atomic_set(&transaction->t_updates, 0);
92 atomic_set(&transaction->t_outstanding_credits,
93 atomic_read(&journal->j_reserved_credits));
94 atomic_set(&transaction->t_handle_count, 0);
95 INIT_LIST_HEAD(&transaction->t_inode_list);
96 INIT_LIST_HEAD(&transaction->t_private_list);
98 /* Set up the commit timer for the new transaction. */
99 journal->j_commit_timer.expires = round_jiffies_up(transaction->t_expires);
100 add_timer(&journal->j_commit_timer);
102 J_ASSERT(journal->j_running_transaction == NULL);
103 journal->j_running_transaction = transaction;
104 transaction->t_max_wait = 0;
105 transaction->t_start = jiffies;
106 transaction->t_requested = 0;
114 * A handle_t is an object which represents a single atomic update to a
115 * filesystem, and which tracks all of the modifications which form part
116 * of that one update.
120 * Update transaction's maximum wait time, if debugging is enabled.
122 * In order for t_max_wait to be reliable, it must be protected by a
123 * lock. But doing so will mean that start_this_handle() can not be
124 * run in parallel on SMP systems, which limits our scalability. So
125 * unless debugging is enabled, we no longer update t_max_wait, which
126 * means that maximum wait time reported by the jbd2_run_stats
127 * tracepoint will always be zero.
129 static inline void update_t_max_wait(transaction_t *transaction,
132 #ifdef CONFIG_JBD2_DEBUG
133 if (jbd2_journal_enable_debug &&
134 time_after(transaction->t_start, ts)) {
135 ts = jbd2_time_diff(ts, transaction->t_start);
136 spin_lock(&transaction->t_handle_lock);
137 if (ts > transaction->t_max_wait)
138 transaction->t_max_wait = ts;
139 spin_unlock(&transaction->t_handle_lock);
145 * Wait until running transaction passes T_LOCKED state. Also starts the commit
146 * if needed. The function expects running transaction to exist and releases
149 static void wait_transaction_locked(journal_t *journal)
150 __releases(journal->j_state_lock)
154 tid_t tid = journal->j_running_transaction->t_tid;
156 prepare_to_wait(&journal->j_wait_transaction_locked, &wait,
157 TASK_UNINTERRUPTIBLE);
158 need_to_start = !tid_geq(journal->j_commit_request, tid);
159 read_unlock(&journal->j_state_lock);
161 jbd2_log_start_commit(journal, tid);
163 finish_wait(&journal->j_wait_transaction_locked, &wait);
166 static void sub_reserved_credits(journal_t *journal, int blocks)
168 atomic_sub(blocks, &journal->j_reserved_credits);
169 wake_up(&journal->j_wait_reserved);
173 * Wait until we can add credits for handle to the running transaction. Called
174 * with j_state_lock held for reading. Returns 0 if handle joined the running
175 * transaction. Returns 1 if we had to wait, j_state_lock is dropped, and
178 static int add_transaction_credits(journal_t *journal, int blocks,
181 transaction_t *t = journal->j_running_transaction;
183 int total = blocks + rsv_blocks;
185 jbd2_might_wait_for_commit(journal);
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 wait_event(journal->j_wait_reserved,
218 atomic_read(&journal->j_reserved_credits) + total <=
219 journal->j_max_transaction_buffers);
223 wait_transaction_locked(journal);
228 * The commit code assumes that it can get enough log space
229 * without forcing a checkpoint. This is *critical* for
230 * correctness: a checkpoint of a buffer which is also
231 * associated with a committing transaction creates a deadlock,
232 * so commit simply cannot force through checkpoints.
234 * We must therefore ensure the necessary space in the journal
235 * *before* starting to dirty potentially checkpointed buffers
236 * in the new transaction.
238 if (jbd2_log_space_left(journal) < jbd2_space_needed(journal)) {
239 atomic_sub(total, &t->t_outstanding_credits);
240 read_unlock(&journal->j_state_lock);
241 write_lock(&journal->j_state_lock);
242 if (jbd2_log_space_left(journal) < jbd2_space_needed(journal))
243 __jbd2_log_wait_for_space(journal);
244 write_unlock(&journal->j_state_lock);
248 /* No reservation? We are done... */
252 needed = atomic_add_return(rsv_blocks, &journal->j_reserved_credits);
253 /* We allow at most half of a transaction to be reserved */
254 if (needed > journal->j_max_transaction_buffers / 2) {
255 sub_reserved_credits(journal, rsv_blocks);
256 atomic_sub(total, &t->t_outstanding_credits);
257 read_unlock(&journal->j_state_lock);
258 wait_event(journal->j_wait_reserved,
259 atomic_read(&journal->j_reserved_credits) + rsv_blocks
260 <= journal->j_max_transaction_buffers / 2);
267 * start_this_handle: Given a handle, deal with any locking or stalling
268 * needed to make sure that there is enough journal space for the handle
269 * to begin. Attach the handle to a transaction and set up the
270 * transaction's buffer credits.
273 static int start_this_handle(journal_t *journal, handle_t *handle,
276 transaction_t *transaction, *new_transaction = NULL;
277 int blocks = handle->h_buffer_credits;
279 unsigned long ts = jiffies;
281 if (handle->h_rsv_handle)
282 rsv_blocks = handle->h_rsv_handle->h_buffer_credits;
285 * Limit the number of reserved credits to 1/2 of maximum transaction
286 * size and limit the number of total credits to not exceed maximum
287 * transaction size per operation.
289 if ((rsv_blocks > journal->j_max_transaction_buffers / 2) ||
290 (rsv_blocks + blocks > journal->j_max_transaction_buffers)) {
291 printk(KERN_ERR "JBD2: %s wants too many credits "
292 "credits:%d rsv_credits:%d max:%d\n",
293 current->comm, blocks, rsv_blocks,
294 journal->j_max_transaction_buffers);
300 if (!journal->j_running_transaction) {
302 * If __GFP_FS is not present, then we may be being called from
303 * inside the fs writeback layer, so we MUST NOT fail.
305 if ((gfp_mask & __GFP_FS) == 0)
306 gfp_mask |= __GFP_NOFAIL;
307 new_transaction = kmem_cache_zalloc(transaction_cache,
309 if (!new_transaction)
313 jbd_debug(3, "New handle %p going live.\n", handle);
316 * We need to hold j_state_lock until t_updates has been incremented,
317 * for proper journal barrier handling
320 read_lock(&journal->j_state_lock);
321 BUG_ON(journal->j_flags & JBD2_UNMOUNT);
322 if (is_journal_aborted(journal) ||
323 (journal->j_errno != 0 && !(journal->j_flags & JBD2_ACK_ERR))) {
324 read_unlock(&journal->j_state_lock);
325 jbd2_journal_free_transaction(new_transaction);
330 * Wait on the journal's transaction barrier if necessary. Specifically
331 * we allow reserved handles to proceed because otherwise commit could
332 * deadlock on page writeback not being able to complete.
334 if (!handle->h_reserved && journal->j_barrier_count) {
335 read_unlock(&journal->j_state_lock);
336 wait_event(journal->j_wait_transaction_locked,
337 journal->j_barrier_count == 0);
341 if (!journal->j_running_transaction) {
342 read_unlock(&journal->j_state_lock);
343 if (!new_transaction)
344 goto alloc_transaction;
345 write_lock(&journal->j_state_lock);
346 if (!journal->j_running_transaction &&
347 (handle->h_reserved || !journal->j_barrier_count)) {
348 jbd2_get_transaction(journal, new_transaction);
349 new_transaction = NULL;
351 write_unlock(&journal->j_state_lock);
355 transaction = journal->j_running_transaction;
357 if (!handle->h_reserved) {
358 /* We may have dropped j_state_lock - restart in that case */
359 if (add_transaction_credits(journal, blocks, rsv_blocks))
363 * We have handle reserved so we are allowed to join T_LOCKED
364 * transaction and we don't have to check for transaction size
367 sub_reserved_credits(journal, blocks);
368 handle->h_reserved = 0;
371 /* OK, account for the buffers that this operation expects to
372 * use and add the handle to the running transaction.
374 update_t_max_wait(transaction, ts);
375 handle->h_transaction = transaction;
376 handle->h_requested_credits = blocks;
377 handle->h_start_jiffies = jiffies;
378 atomic_inc(&transaction->t_updates);
379 atomic_inc(&transaction->t_handle_count);
380 jbd_debug(4, "Handle %p given %d credits (total %d, free %lu)\n",
382 atomic_read(&transaction->t_outstanding_credits),
383 jbd2_log_space_left(journal));
384 read_unlock(&journal->j_state_lock);
385 current->journal_info = handle;
387 rwsem_acquire_read(&journal->j_trans_commit_map, 0, 0, _THIS_IP_);
388 jbd2_journal_free_transaction(new_transaction);
392 /* Allocate a new handle. This should probably be in a slab... */
393 static handle_t *new_handle(int nblocks)
395 handle_t *handle = jbd2_alloc_handle(GFP_NOFS);
398 handle->h_buffer_credits = nblocks;
405 * handle_t *jbd2_journal_start() - Obtain a new handle.
406 * @journal: Journal to start transaction on.
407 * @nblocks: number of block buffer we might modify
409 * We make sure that the transaction can guarantee at least nblocks of
410 * modified buffers in the log. We block until the log can guarantee
411 * that much space. Additionally, if rsv_blocks > 0, we also create another
412 * handle with rsv_blocks reserved blocks in the journal. This handle is
413 * is stored in h_rsv_handle. It is not attached to any particular transaction
414 * and thus doesn't block transaction commit. If the caller uses this reserved
415 * handle, it has to set h_rsv_handle to NULL as otherwise jbd2_journal_stop()
416 * on the parent handle will dispose the reserved one. Reserved handle has to
417 * be converted to a normal handle using jbd2_journal_start_reserved() before
420 * Return a pointer to a newly allocated handle, or an ERR_PTR() value
423 handle_t *jbd2__journal_start(journal_t *journal, int nblocks, int rsv_blocks,
424 gfp_t gfp_mask, unsigned int type,
425 unsigned int line_no)
427 handle_t *handle = journal_current_handle();
431 return ERR_PTR(-EROFS);
434 J_ASSERT(handle->h_transaction->t_journal == journal);
439 handle = new_handle(nblocks);
441 return ERR_PTR(-ENOMEM);
443 handle_t *rsv_handle;
445 rsv_handle = new_handle(rsv_blocks);
447 jbd2_free_handle(handle);
448 return ERR_PTR(-ENOMEM);
450 rsv_handle->h_reserved = 1;
451 rsv_handle->h_journal = journal;
452 handle->h_rsv_handle = rsv_handle;
455 err = start_this_handle(journal, handle, gfp_mask);
457 if (handle->h_rsv_handle)
458 jbd2_free_handle(handle->h_rsv_handle);
459 jbd2_free_handle(handle);
462 handle->h_type = type;
463 handle->h_line_no = line_no;
464 trace_jbd2_handle_start(journal->j_fs_dev->bd_dev,
465 handle->h_transaction->t_tid, type,
469 EXPORT_SYMBOL(jbd2__journal_start);
472 handle_t *jbd2_journal_start(journal_t *journal, int nblocks)
474 return jbd2__journal_start(journal, nblocks, 0, GFP_NOFS, 0, 0);
476 EXPORT_SYMBOL(jbd2_journal_start);
478 void jbd2_journal_free_reserved(handle_t *handle)
480 journal_t *journal = handle->h_journal;
482 WARN_ON(!handle->h_reserved);
483 sub_reserved_credits(journal, handle->h_buffer_credits);
484 jbd2_free_handle(handle);
486 EXPORT_SYMBOL(jbd2_journal_free_reserved);
489 * int jbd2_journal_start_reserved(handle_t *handle) - start reserved handle
490 * @handle: handle to start
492 * Start handle that has been previously reserved with jbd2_journal_reserve().
493 * This attaches @handle to the running transaction (or creates one if there's
494 * not transaction running). Unlike jbd2_journal_start() this function cannot
495 * block on journal commit, checkpointing, or similar stuff. It can block on
496 * memory allocation or frozen journal though.
498 * Return 0 on success, non-zero on error - handle is freed in that case.
500 int jbd2_journal_start_reserved(handle_t *handle, unsigned int type,
501 unsigned int line_no)
503 journal_t *journal = handle->h_journal;
506 if (WARN_ON(!handle->h_reserved)) {
507 /* Someone passed in normal handle? Just stop it. */
508 jbd2_journal_stop(handle);
512 * Usefulness of mixing of reserved and unreserved handles is
513 * questionable. So far nobody seems to need it so just error out.
515 if (WARN_ON(current->journal_info)) {
516 jbd2_journal_free_reserved(handle);
520 handle->h_journal = NULL;
522 * GFP_NOFS is here because callers are likely from writeback or
523 * similarly constrained call sites
525 ret = start_this_handle(journal, handle, GFP_NOFS);
527 jbd2_journal_free_reserved(handle);
530 handle->h_type = type;
531 handle->h_line_no = line_no;
534 EXPORT_SYMBOL(jbd2_journal_start_reserved);
537 * int jbd2_journal_extend() - extend buffer credits.
538 * @handle: handle to 'extend'
539 * @nblocks: nr blocks to try to extend by.
541 * Some transactions, such as large extends and truncates, can be done
542 * atomically all at once or in several stages. The operation requests
543 * a credit for a number of buffer modifications in advance, but can
544 * extend its credit if it needs more.
546 * jbd2_journal_extend tries to give the running handle more buffer credits.
547 * It does not guarantee that allocation - this is a best-effort only.
548 * The calling process MUST be able to deal cleanly with a failure to
551 * Return 0 on success, non-zero on failure.
553 * return code < 0 implies an error
554 * return code > 0 implies normal transaction-full status.
556 int jbd2_journal_extend(handle_t *handle, int nblocks)
558 transaction_t *transaction = handle->h_transaction;
563 if (is_handle_aborted(handle))
565 journal = transaction->t_journal;
569 read_lock(&journal->j_state_lock);
571 /* Don't extend a locked-down transaction! */
572 if (transaction->t_state != T_RUNNING) {
573 jbd_debug(3, "denied handle %p %d blocks: "
574 "transaction not running\n", handle, nblocks);
578 spin_lock(&transaction->t_handle_lock);
579 wanted = atomic_add_return(nblocks,
580 &transaction->t_outstanding_credits);
582 if (wanted > journal->j_max_transaction_buffers) {
583 jbd_debug(3, "denied handle %p %d blocks: "
584 "transaction too large\n", handle, nblocks);
585 atomic_sub(nblocks, &transaction->t_outstanding_credits);
589 if (wanted + (wanted >> JBD2_CONTROL_BLOCKS_SHIFT) >
590 jbd2_log_space_left(journal)) {
591 jbd_debug(3, "denied handle %p %d blocks: "
592 "insufficient log space\n", handle, nblocks);
593 atomic_sub(nblocks, &transaction->t_outstanding_credits);
597 trace_jbd2_handle_extend(journal->j_fs_dev->bd_dev,
599 handle->h_type, handle->h_line_no,
600 handle->h_buffer_credits,
603 handle->h_buffer_credits += nblocks;
604 handle->h_requested_credits += nblocks;
607 jbd_debug(3, "extended handle %p by %d\n", handle, nblocks);
609 spin_unlock(&transaction->t_handle_lock);
611 read_unlock(&journal->j_state_lock);
617 * int jbd2_journal_restart() - restart a handle .
618 * @handle: handle to restart
619 * @nblocks: nr credits requested
621 * Restart a handle for a multi-transaction filesystem
624 * If the jbd2_journal_extend() call above fails to grant new buffer credits
625 * to a running handle, a call to jbd2_journal_restart will commit the
626 * handle's transaction so far and reattach the handle to a new
627 * transaction capable of guaranteeing the requested number of
628 * credits. We preserve reserved handle if there's any attached to the
631 int jbd2__journal_restart(handle_t *handle, int nblocks, gfp_t gfp_mask)
633 transaction_t *transaction = handle->h_transaction;
636 int need_to_start, ret;
638 /* If we've had an abort of any type, don't even think about
639 * actually doing the restart! */
640 if (is_handle_aborted(handle))
642 journal = transaction->t_journal;
645 * First unlink the handle from its current transaction, and start the
648 J_ASSERT(atomic_read(&transaction->t_updates) > 0);
649 J_ASSERT(journal_current_handle() == handle);
651 read_lock(&journal->j_state_lock);
652 spin_lock(&transaction->t_handle_lock);
653 atomic_sub(handle->h_buffer_credits,
654 &transaction->t_outstanding_credits);
655 if (handle->h_rsv_handle) {
656 sub_reserved_credits(journal,
657 handle->h_rsv_handle->h_buffer_credits);
659 if (atomic_dec_and_test(&transaction->t_updates))
660 wake_up(&journal->j_wait_updates);
661 tid = transaction->t_tid;
662 spin_unlock(&transaction->t_handle_lock);
663 handle->h_transaction = NULL;
664 current->journal_info = NULL;
666 jbd_debug(2, "restarting handle %p\n", handle);
667 need_to_start = !tid_geq(journal->j_commit_request, tid);
668 read_unlock(&journal->j_state_lock);
670 jbd2_log_start_commit(journal, tid);
672 rwsem_release(&journal->j_trans_commit_map, 1, _THIS_IP_);
673 handle->h_buffer_credits = nblocks;
674 ret = start_this_handle(journal, handle, gfp_mask);
677 EXPORT_SYMBOL(jbd2__journal_restart);
680 int jbd2_journal_restart(handle_t *handle, int nblocks)
682 return jbd2__journal_restart(handle, nblocks, GFP_NOFS);
684 EXPORT_SYMBOL(jbd2_journal_restart);
687 * void jbd2_journal_lock_updates () - establish a transaction barrier.
688 * @journal: Journal to establish a barrier on.
690 * This locks out any further updates from being started, and blocks
691 * until all existing updates have completed, returning only once the
692 * journal is in a quiescent state with no updates running.
694 * The journal lock should not be held on entry.
696 void jbd2_journal_lock_updates(journal_t *journal)
700 jbd2_might_wait_for_commit(journal);
702 write_lock(&journal->j_state_lock);
703 ++journal->j_barrier_count;
705 /* Wait until there are no reserved handles */
706 if (atomic_read(&journal->j_reserved_credits)) {
707 write_unlock(&journal->j_state_lock);
708 wait_event(journal->j_wait_reserved,
709 atomic_read(&journal->j_reserved_credits) == 0);
710 write_lock(&journal->j_state_lock);
713 /* Wait until there are no running updates */
715 transaction_t *transaction = journal->j_running_transaction;
720 spin_lock(&transaction->t_handle_lock);
721 prepare_to_wait(&journal->j_wait_updates, &wait,
722 TASK_UNINTERRUPTIBLE);
723 if (!atomic_read(&transaction->t_updates)) {
724 spin_unlock(&transaction->t_handle_lock);
725 finish_wait(&journal->j_wait_updates, &wait);
728 spin_unlock(&transaction->t_handle_lock);
729 write_unlock(&journal->j_state_lock);
731 finish_wait(&journal->j_wait_updates, &wait);
732 write_lock(&journal->j_state_lock);
734 write_unlock(&journal->j_state_lock);
737 * We have now established a barrier against other normal updates, but
738 * we also need to barrier against other jbd2_journal_lock_updates() calls
739 * to make sure that we serialise special journal-locked operations
742 mutex_lock(&journal->j_barrier);
746 * void jbd2_journal_unlock_updates (journal_t* journal) - release barrier
747 * @journal: Journal to release the barrier on.
749 * Release a transaction barrier obtained with jbd2_journal_lock_updates().
751 * Should be called without the journal lock held.
753 void jbd2_journal_unlock_updates (journal_t *journal)
755 J_ASSERT(journal->j_barrier_count != 0);
757 mutex_unlock(&journal->j_barrier);
758 write_lock(&journal->j_state_lock);
759 --journal->j_barrier_count;
760 write_unlock(&journal->j_state_lock);
761 wake_up(&journal->j_wait_transaction_locked);
764 static void warn_dirty_buffer(struct buffer_head *bh)
767 "JBD2: Spotted dirty metadata buffer (dev = %pg, blocknr = %llu). "
768 "There's a risk of filesystem corruption in case of system "
770 bh->b_bdev, (unsigned long long)bh->b_blocknr);
773 /* Call t_frozen trigger and copy buffer data into jh->b_frozen_data. */
774 static void jbd2_freeze_jh_data(struct journal_head *jh)
779 struct buffer_head *bh = jh2bh(jh);
781 J_EXPECT_JH(jh, buffer_uptodate(bh), "Possible IO failure.\n");
783 offset = offset_in_page(bh->b_data);
784 source = kmap_atomic(page);
785 /* Fire data frozen trigger just before we copy the data */
786 jbd2_buffer_frozen_trigger(jh, source + offset, jh->b_triggers);
787 memcpy(jh->b_frozen_data, source + offset, bh->b_size);
788 kunmap_atomic(source);
791 * Now that the frozen data is saved off, we need to store any matching
794 jh->b_frozen_triggers = jh->b_triggers;
798 * If the buffer is already part of the current transaction, then there
799 * is nothing we need to do. If it is already part of a prior
800 * transaction which we are still committing to disk, then we need to
801 * make sure that we do not overwrite the old copy: we do copy-out to
802 * preserve the copy going to disk. We also account the buffer against
803 * the handle's metadata buffer credits (unless the buffer is already
804 * part of the transaction, that is).
808 do_get_write_access(handle_t *handle, struct journal_head *jh,
811 struct buffer_head *bh;
812 transaction_t *transaction = handle->h_transaction;
815 char *frozen_buffer = NULL;
816 unsigned long start_lock, time_lock;
818 if (is_handle_aborted(handle))
820 journal = transaction->t_journal;
822 jbd_debug(5, "journal_head %p, force_copy %d\n", jh, force_copy);
824 JBUFFER_TRACE(jh, "entry");
828 /* @@@ Need to check for errors here at some point. */
830 start_lock = jiffies;
832 jbd_lock_bh_state(bh);
834 /* If it takes too long to lock the buffer, trace it */
835 time_lock = jbd2_time_diff(start_lock, jiffies);
836 if (time_lock > HZ/10)
837 trace_jbd2_lock_buffer_stall(bh->b_bdev->bd_dev,
838 jiffies_to_msecs(time_lock));
840 /* We now hold the buffer lock so it is safe to query the buffer
841 * state. Is the buffer dirty?
843 * If so, there are two possibilities. The buffer may be
844 * non-journaled, and undergoing a quite legitimate writeback.
845 * Otherwise, it is journaled, and we don't expect dirty buffers
846 * in that state (the buffers should be marked JBD_Dirty
847 * instead.) So either the IO is being done under our own
848 * control and this is a bug, or it's a third party IO such as
849 * dump(8) (which may leave the buffer scheduled for read ---
850 * ie. locked but not dirty) or tune2fs (which may actually have
851 * the buffer dirtied, ugh.) */
853 if (buffer_dirty(bh)) {
855 * First question: is this buffer already part of the current
856 * transaction or the existing committing transaction?
858 if (jh->b_transaction) {
860 jh->b_transaction == transaction ||
862 journal->j_committing_transaction);
863 if (jh->b_next_transaction)
864 J_ASSERT_JH(jh, jh->b_next_transaction ==
866 warn_dirty_buffer(bh);
869 * In any case we need to clean the dirty flag and we must
870 * do it under the buffer lock to be sure we don't race
871 * with running write-out.
873 JBUFFER_TRACE(jh, "Journalling dirty buffer");
874 clear_buffer_dirty(bh);
875 set_buffer_jbddirty(bh);
881 if (is_handle_aborted(handle)) {
882 jbd_unlock_bh_state(bh);
888 * The buffer is already part of this transaction if b_transaction or
889 * b_next_transaction points to it
891 if (jh->b_transaction == transaction ||
892 jh->b_next_transaction == transaction)
896 * this is the first time this transaction is touching this buffer,
897 * reset the modified flag
902 * If the buffer is not journaled right now, we need to make sure it
903 * doesn't get written to disk before the caller actually commits the
906 if (!jh->b_transaction) {
907 JBUFFER_TRACE(jh, "no transaction");
908 J_ASSERT_JH(jh, !jh->b_next_transaction);
909 JBUFFER_TRACE(jh, "file as BJ_Reserved");
911 * Make sure all stores to jh (b_modified, b_frozen_data) are
912 * visible before attaching it to the running transaction.
913 * Paired with barrier in jbd2_write_access_granted()
916 spin_lock(&journal->j_list_lock);
917 __jbd2_journal_file_buffer(jh, transaction, BJ_Reserved);
918 spin_unlock(&journal->j_list_lock);
922 * If there is already a copy-out version of this buffer, then we don't
923 * need to make another one
925 if (jh->b_frozen_data) {
926 JBUFFER_TRACE(jh, "has frozen data");
927 J_ASSERT_JH(jh, jh->b_next_transaction == NULL);
931 JBUFFER_TRACE(jh, "owned by older transaction");
932 J_ASSERT_JH(jh, jh->b_next_transaction == NULL);
933 J_ASSERT_JH(jh, jh->b_transaction == journal->j_committing_transaction);
936 * There is one case we have to be very careful about. If the
937 * committing transaction is currently writing this buffer out to disk
938 * and has NOT made a copy-out, then we cannot modify the buffer
939 * contents at all right now. The essence of copy-out is that it is
940 * the extra copy, not the primary copy, which gets journaled. If the
941 * primary copy is already going to disk then we cannot do copy-out
944 if (buffer_shadow(bh)) {
945 JBUFFER_TRACE(jh, "on shadow: sleep");
946 jbd_unlock_bh_state(bh);
947 wait_on_bit_io(&bh->b_state, BH_Shadow, TASK_UNINTERRUPTIBLE);
952 * Only do the copy if the currently-owning transaction still needs it.
953 * If buffer isn't on BJ_Metadata list, the committing transaction is
954 * past that stage (here we use the fact that BH_Shadow is set under
955 * bh_state lock together with refiling to BJ_Shadow list and at this
956 * point we know the buffer doesn't have BH_Shadow set).
958 * Subtle point, though: if this is a get_undo_access, then we will be
959 * relying on the frozen_data to contain the new value of the
960 * committed_data record after the transaction, so we HAVE to force the
961 * frozen_data copy in that case.
963 if (jh->b_jlist == BJ_Metadata || force_copy) {
964 JBUFFER_TRACE(jh, "generate frozen data");
965 if (!frozen_buffer) {
966 JBUFFER_TRACE(jh, "allocate memory for buffer");
967 jbd_unlock_bh_state(bh);
968 frozen_buffer = jbd2_alloc(jh2bh(jh)->b_size,
969 GFP_NOFS | __GFP_NOFAIL);
972 jh->b_frozen_data = frozen_buffer;
973 frozen_buffer = NULL;
974 jbd2_freeze_jh_data(jh);
978 * Make sure all stores to jh (b_modified, b_frozen_data) are visible
979 * before attaching it to the running transaction. Paired with barrier
980 * in jbd2_write_access_granted()
983 jh->b_next_transaction = transaction;
986 jbd_unlock_bh_state(bh);
989 * If we are about to journal a buffer, then any revoke pending on it is
992 jbd2_journal_cancel_revoke(handle, jh);
995 if (unlikely(frozen_buffer)) /* It's usually NULL */
996 jbd2_free(frozen_buffer, bh->b_size);
998 JBUFFER_TRACE(jh, "exit");
1002 /* Fast check whether buffer is already attached to the required transaction */
1003 static bool jbd2_write_access_granted(handle_t *handle, struct buffer_head *bh,
1006 struct journal_head *jh;
1009 /* Dirty buffers require special handling... */
1010 if (buffer_dirty(bh))
1014 * RCU protects us from dereferencing freed pages. So the checks we do
1015 * are guaranteed not to oops. However the jh slab object can get freed
1016 * & reallocated while we work with it. So we have to be careful. When
1017 * we see jh attached to the running transaction, we know it must stay
1018 * so until the transaction is committed. Thus jh won't be freed and
1019 * will be attached to the same bh while we run. However it can
1020 * happen jh gets freed, reallocated, and attached to the transaction
1021 * just after we get pointer to it from bh. So we have to be careful
1022 * and recheck jh still belongs to our bh before we return success.
1025 if (!buffer_jbd(bh))
1027 /* This should be bh2jh() but that doesn't work with inline functions */
1028 jh = READ_ONCE(bh->b_private);
1031 /* For undo access buffer must have data copied */
1032 if (undo && !jh->b_committed_data)
1034 if (jh->b_transaction != handle->h_transaction &&
1035 jh->b_next_transaction != handle->h_transaction)
1038 * There are two reasons for the barrier here:
1039 * 1) Make sure to fetch b_bh after we did previous checks so that we
1040 * detect when jh went through free, realloc, attach to transaction
1041 * while we were checking. Paired with implicit barrier in that path.
1042 * 2) So that access to bh done after jbd2_write_access_granted()
1043 * doesn't get reordered and see inconsistent state of concurrent
1044 * do_get_write_access().
1047 if (unlikely(jh->b_bh != bh))
1056 * int jbd2_journal_get_write_access() - notify intent to modify a buffer for metadata (not data) update.
1057 * @handle: transaction to add buffer modifications to
1058 * @bh: bh to be used for metadata writes
1060 * Returns an error code or 0 on success.
1062 * In full data journalling mode the buffer may be of type BJ_AsyncData,
1063 * because we're write()ing a buffer which is also part of a shared mapping.
1066 int jbd2_journal_get_write_access(handle_t *handle, struct buffer_head *bh)
1068 struct journal_head *jh;
1071 if (jbd2_write_access_granted(handle, bh, false))
1074 jh = jbd2_journal_add_journal_head(bh);
1075 /* We do not want to get caught playing with fields which the
1076 * log thread also manipulates. Make sure that the buffer
1077 * completes any outstanding IO before proceeding. */
1078 rc = do_get_write_access(handle, jh, 0);
1079 jbd2_journal_put_journal_head(jh);
1085 * When the user wants to journal a newly created buffer_head
1086 * (ie. getblk() returned a new buffer and we are going to populate it
1087 * manually rather than reading off disk), then we need to keep the
1088 * buffer_head locked until it has been completely filled with new
1089 * data. In this case, we should be able to make the assertion that
1090 * the bh is not already part of an existing transaction.
1092 * The buffer should already be locked by the caller by this point.
1093 * There is no lock ranking violation: it was a newly created,
1094 * unlocked buffer beforehand. */
1097 * int jbd2_journal_get_create_access () - notify intent to use newly created bh
1098 * @handle: transaction to new buffer to
1101 * Call this if you create a new bh.
1103 int jbd2_journal_get_create_access(handle_t *handle, struct buffer_head *bh)
1105 transaction_t *transaction = handle->h_transaction;
1107 struct journal_head *jh = jbd2_journal_add_journal_head(bh);
1110 jbd_debug(5, "journal_head %p\n", jh);
1112 if (is_handle_aborted(handle))
1114 journal = transaction->t_journal;
1117 JBUFFER_TRACE(jh, "entry");
1119 * The buffer may already belong to this transaction due to pre-zeroing
1120 * in the filesystem's new_block code. It may also be on the previous,
1121 * committing transaction's lists, but it HAS to be in Forget state in
1122 * that case: the transaction must have deleted the buffer for it to be
1125 jbd_lock_bh_state(bh);
1126 J_ASSERT_JH(jh, (jh->b_transaction == transaction ||
1127 jh->b_transaction == NULL ||
1128 (jh->b_transaction == journal->j_committing_transaction &&
1129 jh->b_jlist == BJ_Forget)));
1131 J_ASSERT_JH(jh, jh->b_next_transaction == NULL);
1132 J_ASSERT_JH(jh, buffer_locked(jh2bh(jh)));
1134 if (jh->b_transaction == NULL) {
1136 * Previous jbd2_journal_forget() could have left the buffer
1137 * with jbddirty bit set because it was being committed. When
1138 * the commit finished, we've filed the buffer for
1139 * checkpointing and marked it dirty. Now we are reallocating
1140 * the buffer so the transaction freeing it must have
1141 * committed and so it's safe to clear the dirty bit.
1143 clear_buffer_dirty(jh2bh(jh));
1144 /* first access by this transaction */
1147 JBUFFER_TRACE(jh, "file as BJ_Reserved");
1148 spin_lock(&journal->j_list_lock);
1149 __jbd2_journal_file_buffer(jh, transaction, BJ_Reserved);
1150 } else if (jh->b_transaction == journal->j_committing_transaction) {
1151 /* first access by this transaction */
1154 JBUFFER_TRACE(jh, "set next transaction");
1155 spin_lock(&journal->j_list_lock);
1156 jh->b_next_transaction = transaction;
1158 spin_unlock(&journal->j_list_lock);
1159 jbd_unlock_bh_state(bh);
1162 * akpm: I added this. ext3_alloc_branch can pick up new indirect
1163 * blocks which contain freed but then revoked metadata. We need
1164 * to cancel the revoke in case we end up freeing it yet again
1165 * and the reallocating as data - this would cause a second revoke,
1166 * which hits an assertion error.
1168 JBUFFER_TRACE(jh, "cancelling revoke");
1169 jbd2_journal_cancel_revoke(handle, jh);
1171 jbd2_journal_put_journal_head(jh);
1176 * int jbd2_journal_get_undo_access() - Notify intent to modify metadata with
1177 * non-rewindable consequences
1178 * @handle: transaction
1179 * @bh: buffer to undo
1181 * Sometimes there is a need to distinguish between metadata which has
1182 * been committed to disk and that which has not. The ext3fs code uses
1183 * this for freeing and allocating space, we have to make sure that we
1184 * do not reuse freed space until the deallocation has been committed,
1185 * since if we overwrote that space we would make the delete
1186 * un-rewindable in case of a crash.
1188 * To deal with that, jbd2_journal_get_undo_access requests write access to a
1189 * buffer for parts of non-rewindable operations such as delete
1190 * operations on the bitmaps. The journaling code must keep a copy of
1191 * the buffer's contents prior to the undo_access call until such time
1192 * as we know that the buffer has definitely been committed to disk.
1194 * We never need to know which transaction the committed data is part
1195 * of, buffers touched here are guaranteed to be dirtied later and so
1196 * will be committed to a new transaction in due course, at which point
1197 * we can discard the old committed data pointer.
1199 * Returns error number or 0 on success.
1201 int jbd2_journal_get_undo_access(handle_t *handle, struct buffer_head *bh)
1204 struct journal_head *jh;
1205 char *committed_data = NULL;
1207 JBUFFER_TRACE(jh, "entry");
1208 if (jbd2_write_access_granted(handle, bh, true))
1211 jh = jbd2_journal_add_journal_head(bh);
1213 * Do this first --- it can drop the journal lock, so we want to
1214 * make sure that obtaining the committed_data is done
1215 * atomically wrt. completion of any outstanding commits.
1217 err = do_get_write_access(handle, jh, 1);
1222 if (!jh->b_committed_data)
1223 committed_data = jbd2_alloc(jh2bh(jh)->b_size,
1224 GFP_NOFS|__GFP_NOFAIL);
1226 jbd_lock_bh_state(bh);
1227 if (!jh->b_committed_data) {
1228 /* Copy out the current buffer contents into the
1229 * preserved, committed copy. */
1230 JBUFFER_TRACE(jh, "generate b_committed data");
1231 if (!committed_data) {
1232 jbd_unlock_bh_state(bh);
1236 jh->b_committed_data = committed_data;
1237 committed_data = NULL;
1238 memcpy(jh->b_committed_data, bh->b_data, bh->b_size);
1240 jbd_unlock_bh_state(bh);
1242 jbd2_journal_put_journal_head(jh);
1243 if (unlikely(committed_data))
1244 jbd2_free(committed_data, bh->b_size);
1249 * void jbd2_journal_set_triggers() - Add triggers for commit writeout
1250 * @bh: buffer to trigger on
1251 * @type: struct jbd2_buffer_trigger_type containing the trigger(s).
1253 * Set any triggers on this journal_head. This is always safe, because
1254 * triggers for a committing buffer will be saved off, and triggers for
1255 * a running transaction will match the buffer in that transaction.
1257 * Call with NULL to clear the triggers.
1259 void jbd2_journal_set_triggers(struct buffer_head *bh,
1260 struct jbd2_buffer_trigger_type *type)
1262 struct journal_head *jh = jbd2_journal_grab_journal_head(bh);
1266 jh->b_triggers = type;
1267 jbd2_journal_put_journal_head(jh);
1270 void jbd2_buffer_frozen_trigger(struct journal_head *jh, void *mapped_data,
1271 struct jbd2_buffer_trigger_type *triggers)
1273 struct buffer_head *bh = jh2bh(jh);
1275 if (!triggers || !triggers->t_frozen)
1278 triggers->t_frozen(triggers, bh, mapped_data, bh->b_size);
1281 void jbd2_buffer_abort_trigger(struct journal_head *jh,
1282 struct jbd2_buffer_trigger_type *triggers)
1284 if (!triggers || !triggers->t_abort)
1287 triggers->t_abort(triggers, jh2bh(jh));
1291 * int jbd2_journal_dirty_metadata() - mark a buffer as containing dirty metadata
1292 * @handle: transaction to add buffer to.
1293 * @bh: buffer to mark
1295 * mark dirty metadata which needs to be journaled as part of the current
1298 * The buffer must have previously had jbd2_journal_get_write_access()
1299 * called so that it has a valid journal_head attached to the buffer
1302 * The buffer is placed on the transaction's metadata list and is marked
1303 * as belonging to the transaction.
1305 * Returns error number or 0 on success.
1307 * Special care needs to be taken if the buffer already belongs to the
1308 * current committing transaction (in which case we should have frozen
1309 * data present for that commit). In that case, we don't relink the
1310 * buffer: that only gets done when the old transaction finally
1311 * completes its commit.
1313 int jbd2_journal_dirty_metadata(handle_t *handle, struct buffer_head *bh)
1315 transaction_t *transaction = handle->h_transaction;
1317 struct journal_head *jh;
1320 if (is_handle_aborted(handle))
1322 if (!buffer_jbd(bh)) {
1327 * We don't grab jh reference here since the buffer must be part
1328 * of the running transaction.
1332 * This and the following assertions are unreliable since we may see jh
1333 * in inconsistent state unless we grab bh_state lock. But this is
1334 * crucial to catch bugs so let's do a reliable check until the
1335 * lockless handling is fully proven.
1337 if (jh->b_transaction != transaction &&
1338 jh->b_next_transaction != transaction) {
1339 jbd_lock_bh_state(bh);
1340 J_ASSERT_JH(jh, jh->b_transaction == transaction ||
1341 jh->b_next_transaction == transaction);
1342 jbd_unlock_bh_state(bh);
1344 if (jh->b_modified == 1) {
1345 /* If it's in our transaction it must be in BJ_Metadata list. */
1346 if (jh->b_transaction == transaction &&
1347 jh->b_jlist != BJ_Metadata) {
1348 jbd_lock_bh_state(bh);
1349 J_ASSERT_JH(jh, jh->b_transaction != transaction ||
1350 jh->b_jlist == BJ_Metadata);
1351 jbd_unlock_bh_state(bh);
1356 journal = transaction->t_journal;
1357 jbd_debug(5, "journal_head %p\n", jh);
1358 JBUFFER_TRACE(jh, "entry");
1360 jbd_lock_bh_state(bh);
1362 if (jh->b_modified == 0) {
1364 * This buffer's got modified and becoming part
1365 * of the transaction. This needs to be done
1366 * once a transaction -bzzz
1369 if (handle->h_buffer_credits <= 0) {
1373 handle->h_buffer_credits--;
1377 * fastpath, to avoid expensive locking. If this buffer is already
1378 * on the running transaction's metadata list there is nothing to do.
1379 * Nobody can take it off again because there is a handle open.
1380 * I _think_ we're OK here with SMP barriers - a mistaken decision will
1381 * result in this test being false, so we go in and take the locks.
1383 if (jh->b_transaction == transaction && jh->b_jlist == BJ_Metadata) {
1384 JBUFFER_TRACE(jh, "fastpath");
1385 if (unlikely(jh->b_transaction !=
1386 journal->j_running_transaction)) {
1387 printk(KERN_ERR "JBD2: %s: "
1388 "jh->b_transaction (%llu, %p, %u) != "
1389 "journal->j_running_transaction (%p, %u)\n",
1391 (unsigned long long) bh->b_blocknr,
1393 jh->b_transaction ? jh->b_transaction->t_tid : 0,
1394 journal->j_running_transaction,
1395 journal->j_running_transaction ?
1396 journal->j_running_transaction->t_tid : 0);
1402 set_buffer_jbddirty(bh);
1405 * Metadata already on the current transaction list doesn't
1406 * need to be filed. Metadata on another transaction's list must
1407 * be committing, and will be refiled once the commit completes:
1408 * leave it alone for now.
1410 if (jh->b_transaction != transaction) {
1411 JBUFFER_TRACE(jh, "already on other transaction");
1412 if (unlikely(((jh->b_transaction !=
1413 journal->j_committing_transaction)) ||
1414 (jh->b_next_transaction != transaction))) {
1415 printk(KERN_ERR "jbd2_journal_dirty_metadata: %s: "
1416 "bad jh for block %llu: "
1417 "transaction (%p, %u), "
1418 "jh->b_transaction (%p, %u), "
1419 "jh->b_next_transaction (%p, %u), jlist %u\n",
1421 (unsigned long long) bh->b_blocknr,
1422 transaction, transaction->t_tid,
1425 jh->b_transaction->t_tid : 0,
1426 jh->b_next_transaction,
1427 jh->b_next_transaction ?
1428 jh->b_next_transaction->t_tid : 0,
1433 /* And this case is illegal: we can't reuse another
1434 * transaction's data buffer, ever. */
1438 /* That test should have eliminated the following case: */
1439 J_ASSERT_JH(jh, jh->b_frozen_data == NULL);
1441 JBUFFER_TRACE(jh, "file as BJ_Metadata");
1442 spin_lock(&journal->j_list_lock);
1443 __jbd2_journal_file_buffer(jh, transaction, BJ_Metadata);
1444 spin_unlock(&journal->j_list_lock);
1446 jbd_unlock_bh_state(bh);
1448 JBUFFER_TRACE(jh, "exit");
1453 * void jbd2_journal_forget() - bforget() for potentially-journaled buffers.
1454 * @handle: transaction handle
1455 * @bh: bh to 'forget'
1457 * We can only do the bforget if there are no commits pending against the
1458 * buffer. If the buffer is dirty in the current running transaction we
1459 * can safely unlink it.
1461 * bh may not be a journalled buffer at all - it may be a non-JBD
1462 * buffer which came off the hashtable. Check for this.
1464 * Decrements bh->b_count by one.
1466 * Allow this call even if the handle has aborted --- it may be part of
1467 * the caller's cleanup after an abort.
1469 int jbd2_journal_forget (handle_t *handle, struct buffer_head *bh)
1471 transaction_t *transaction = handle->h_transaction;
1473 struct journal_head *jh;
1474 int drop_reserve = 0;
1476 int was_modified = 0;
1478 if (is_handle_aborted(handle))
1480 journal = transaction->t_journal;
1482 BUFFER_TRACE(bh, "entry");
1484 jbd_lock_bh_state(bh);
1486 if (!buffer_jbd(bh))
1490 /* Critical error: attempting to delete a bitmap buffer, maybe?
1491 * Don't do any jbd operations, and return an error. */
1492 if (!J_EXPECT_JH(jh, !jh->b_committed_data,
1493 "inconsistent data on disk")) {
1498 /* keep track of whether or not this transaction modified us */
1499 was_modified = jh->b_modified;
1502 * The buffer's going from the transaction, we must drop
1503 * all references -bzzz
1507 if (jh->b_transaction == transaction) {
1508 J_ASSERT_JH(jh, !jh->b_frozen_data);
1510 /* If we are forgetting a buffer which is already part
1511 * of this transaction, then we can just drop it from
1512 * the transaction immediately. */
1513 clear_buffer_dirty(bh);
1514 clear_buffer_jbddirty(bh);
1516 JBUFFER_TRACE(jh, "belongs to current transaction: unfile");
1519 * we only want to drop a reference if this transaction
1520 * modified the buffer
1526 * We are no longer going to journal this buffer.
1527 * However, the commit of this transaction is still
1528 * important to the buffer: the delete that we are now
1529 * processing might obsolete an old log entry, so by
1530 * committing, we can satisfy the buffer's checkpoint.
1532 * So, if we have a checkpoint on the buffer, we should
1533 * now refile the buffer on our BJ_Forget list so that
1534 * we know to remove the checkpoint after we commit.
1537 spin_lock(&journal->j_list_lock);
1538 if (jh->b_cp_transaction) {
1539 __jbd2_journal_temp_unlink_buffer(jh);
1540 __jbd2_journal_file_buffer(jh, transaction, BJ_Forget);
1542 __jbd2_journal_unfile_buffer(jh);
1543 if (!buffer_jbd(bh)) {
1544 spin_unlock(&journal->j_list_lock);
1545 jbd_unlock_bh_state(bh);
1550 spin_unlock(&journal->j_list_lock);
1551 } else if (jh->b_transaction) {
1552 J_ASSERT_JH(jh, (jh->b_transaction ==
1553 journal->j_committing_transaction));
1554 /* However, if the buffer is still owned by a prior
1555 * (committing) transaction, we can't drop it yet... */
1556 JBUFFER_TRACE(jh, "belongs to older transaction");
1557 /* ... but we CAN drop it from the new transaction if we
1558 * have also modified it since the original commit. */
1560 if (jh->b_next_transaction) {
1561 J_ASSERT(jh->b_next_transaction == transaction);
1562 spin_lock(&journal->j_list_lock);
1563 jh->b_next_transaction = NULL;
1564 spin_unlock(&journal->j_list_lock);
1567 * only drop a reference if this transaction modified
1576 jbd_unlock_bh_state(bh);
1580 /* no need to reserve log space for this block -bzzz */
1581 handle->h_buffer_credits++;
1587 * int jbd2_journal_stop() - complete a transaction
1588 * @handle: transaction to complete.
1590 * All done for a particular handle.
1592 * There is not much action needed here. We just return any remaining
1593 * buffer credits to the transaction and remove the handle. The only
1594 * complication is that we need to start a commit operation if the
1595 * filesystem is marked for synchronous update.
1597 * jbd2_journal_stop itself will not usually return an error, but it may
1598 * do so in unusual circumstances. In particular, expect it to
1599 * return -EIO if a jbd2_journal_abort has been executed since the
1600 * transaction began.
1602 int jbd2_journal_stop(handle_t *handle)
1604 transaction_t *transaction = handle->h_transaction;
1606 int err = 0, wait_for_commit = 0;
1612 * Handle is already detached from the transaction so
1613 * there is nothing to do other than decrease a refcount,
1614 * or free the handle if refcount drops to zero
1616 if (--handle->h_ref > 0) {
1617 jbd_debug(4, "h_ref %d -> %d\n", handle->h_ref + 1,
1621 if (handle->h_rsv_handle)
1622 jbd2_free_handle(handle->h_rsv_handle);
1626 journal = transaction->t_journal;
1628 J_ASSERT(journal_current_handle() == handle);
1630 if (is_handle_aborted(handle))
1633 J_ASSERT(atomic_read(&transaction->t_updates) > 0);
1635 if (--handle->h_ref > 0) {
1636 jbd_debug(4, "h_ref %d -> %d\n", handle->h_ref + 1,
1641 jbd_debug(4, "Handle %p going down\n", handle);
1642 trace_jbd2_handle_stats(journal->j_fs_dev->bd_dev,
1644 handle->h_type, handle->h_line_no,
1645 jiffies - handle->h_start_jiffies,
1646 handle->h_sync, handle->h_requested_credits,
1647 (handle->h_requested_credits -
1648 handle->h_buffer_credits));
1651 * Implement synchronous transaction batching. If the handle
1652 * was synchronous, don't force a commit immediately. Let's
1653 * yield and let another thread piggyback onto this
1654 * transaction. Keep doing that while new threads continue to
1655 * arrive. It doesn't cost much - we're about to run a commit
1656 * and sleep on IO anyway. Speeds up many-threaded, many-dir
1657 * operations by 30x or more...
1659 * We try and optimize the sleep time against what the
1660 * underlying disk can do, instead of having a static sleep
1661 * time. This is useful for the case where our storage is so
1662 * fast that it is more optimal to go ahead and force a flush
1663 * and wait for the transaction to be committed than it is to
1664 * wait for an arbitrary amount of time for new writers to
1665 * join the transaction. We achieve this by measuring how
1666 * long it takes to commit a transaction, and compare it with
1667 * how long this transaction has been running, and if run time
1668 * < commit time then we sleep for the delta and commit. This
1669 * greatly helps super fast disks that would see slowdowns as
1670 * more threads started doing fsyncs.
1672 * But don't do this if this process was the most recent one
1673 * to perform a synchronous write. We do this to detect the
1674 * case where a single process is doing a stream of sync
1675 * writes. No point in waiting for joiners in that case.
1677 * Setting max_batch_time to 0 disables this completely.
1680 if (handle->h_sync && journal->j_last_sync_writer != pid &&
1681 journal->j_max_batch_time) {
1682 u64 commit_time, trans_time;
1684 journal->j_last_sync_writer = pid;
1686 read_lock(&journal->j_state_lock);
1687 commit_time = journal->j_average_commit_time;
1688 read_unlock(&journal->j_state_lock);
1690 trans_time = ktime_to_ns(ktime_sub(ktime_get(),
1691 transaction->t_start_time));
1693 commit_time = max_t(u64, commit_time,
1694 1000*journal->j_min_batch_time);
1695 commit_time = min_t(u64, commit_time,
1696 1000*journal->j_max_batch_time);
1698 if (trans_time < commit_time) {
1699 ktime_t expires = ktime_add_ns(ktime_get(),
1701 set_current_state(TASK_UNINTERRUPTIBLE);
1702 schedule_hrtimeout(&expires, HRTIMER_MODE_ABS);
1707 transaction->t_synchronous_commit = 1;
1708 current->journal_info = NULL;
1709 atomic_sub(handle->h_buffer_credits,
1710 &transaction->t_outstanding_credits);
1713 * If the handle is marked SYNC, we need to set another commit
1714 * going! We also want to force a commit if the current
1715 * transaction is occupying too much of the log, or if the
1716 * transaction is too old now.
1718 if (handle->h_sync ||
1719 (atomic_read(&transaction->t_outstanding_credits) >
1720 journal->j_max_transaction_buffers) ||
1721 time_after_eq(jiffies, transaction->t_expires)) {
1722 /* Do this even for aborted journals: an abort still
1723 * completes the commit thread, it just doesn't write
1724 * anything to disk. */
1726 jbd_debug(2, "transaction too old, requesting commit for "
1727 "handle %p\n", handle);
1728 /* This is non-blocking */
1729 jbd2_log_start_commit(journal, transaction->t_tid);
1732 * Special case: JBD2_SYNC synchronous updates require us
1733 * to wait for the commit to complete.
1735 if (handle->h_sync && !(current->flags & PF_MEMALLOC))
1736 wait_for_commit = 1;
1740 * Once we drop t_updates, if it goes to zero the transaction
1741 * could start committing on us and eventually disappear. So
1742 * once we do this, we must not dereference transaction
1745 tid = transaction->t_tid;
1746 if (atomic_dec_and_test(&transaction->t_updates)) {
1747 wake_up(&journal->j_wait_updates);
1748 if (journal->j_barrier_count)
1749 wake_up(&journal->j_wait_transaction_locked);
1752 rwsem_release(&journal->j_trans_commit_map, 1, _THIS_IP_);
1754 if (wait_for_commit)
1755 err = jbd2_log_wait_commit(journal, tid);
1757 if (handle->h_rsv_handle)
1758 jbd2_journal_free_reserved(handle->h_rsv_handle);
1760 jbd2_free_handle(handle);
1766 * List management code snippets: various functions for manipulating the
1767 * transaction buffer lists.
1772 * Append a buffer to a transaction list, given the transaction's list head
1775 * j_list_lock is held.
1777 * jbd_lock_bh_state(jh2bh(jh)) is held.
1781 __blist_add_buffer(struct journal_head **list, struct journal_head *jh)
1784 jh->b_tnext = jh->b_tprev = jh;
1787 /* Insert at the tail of the list to preserve order */
1788 struct journal_head *first = *list, *last = first->b_tprev;
1790 jh->b_tnext = first;
1791 last->b_tnext = first->b_tprev = jh;
1796 * Remove a buffer from a transaction list, given the transaction's list
1799 * Called with j_list_lock held, and the journal may not be locked.
1801 * jbd_lock_bh_state(jh2bh(jh)) is held.
1805 __blist_del_buffer(struct journal_head **list, struct journal_head *jh)
1808 *list = jh->b_tnext;
1812 jh->b_tprev->b_tnext = jh->b_tnext;
1813 jh->b_tnext->b_tprev = jh->b_tprev;
1817 * Remove a buffer from the appropriate transaction list.
1819 * Note that this function can *change* the value of
1820 * bh->b_transaction->t_buffers, t_forget, t_shadow_list, t_log_list or
1821 * t_reserved_list. If the caller is holding onto a copy of one of these
1822 * pointers, it could go bad. Generally the caller needs to re-read the
1823 * pointer from the transaction_t.
1825 * Called under j_list_lock.
1827 static void __jbd2_journal_temp_unlink_buffer(struct journal_head *jh)
1829 struct journal_head **list = NULL;
1830 transaction_t *transaction;
1831 struct buffer_head *bh = jh2bh(jh);
1833 J_ASSERT_JH(jh, jbd_is_locked_bh_state(bh));
1834 transaction = jh->b_transaction;
1836 assert_spin_locked(&transaction->t_journal->j_list_lock);
1838 J_ASSERT_JH(jh, jh->b_jlist < BJ_Types);
1839 if (jh->b_jlist != BJ_None)
1840 J_ASSERT_JH(jh, transaction != NULL);
1842 switch (jh->b_jlist) {
1846 transaction->t_nr_buffers--;
1847 J_ASSERT_JH(jh, transaction->t_nr_buffers >= 0);
1848 list = &transaction->t_buffers;
1851 list = &transaction->t_forget;
1854 list = &transaction->t_shadow_list;
1857 list = &transaction->t_reserved_list;
1861 __blist_del_buffer(list, jh);
1862 jh->b_jlist = BJ_None;
1863 if (test_clear_buffer_jbddirty(bh))
1864 mark_buffer_dirty(bh); /* Expose it to the VM */
1868 * Remove buffer from all transactions.
1870 * Called with bh_state lock and j_list_lock
1872 * jh and bh may be already freed when this function returns.
1874 static void __jbd2_journal_unfile_buffer(struct journal_head *jh)
1876 __jbd2_journal_temp_unlink_buffer(jh);
1877 jh->b_transaction = NULL;
1878 jbd2_journal_put_journal_head(jh);
1881 void jbd2_journal_unfile_buffer(journal_t *journal, struct journal_head *jh)
1883 struct buffer_head *bh = jh2bh(jh);
1885 /* Get reference so that buffer cannot be freed before we unlock it */
1887 jbd_lock_bh_state(bh);
1888 spin_lock(&journal->j_list_lock);
1889 __jbd2_journal_unfile_buffer(jh);
1890 spin_unlock(&journal->j_list_lock);
1891 jbd_unlock_bh_state(bh);
1896 * Called from jbd2_journal_try_to_free_buffers().
1898 * Called under jbd_lock_bh_state(bh)
1901 __journal_try_to_free_buffer(journal_t *journal, struct buffer_head *bh)
1903 struct journal_head *jh;
1907 if (buffer_locked(bh) || buffer_dirty(bh))
1910 if (jh->b_next_transaction != NULL || jh->b_transaction != NULL)
1913 spin_lock(&journal->j_list_lock);
1914 if (jh->b_cp_transaction != NULL) {
1915 /* written-back checkpointed metadata buffer */
1916 JBUFFER_TRACE(jh, "remove from checkpoint list");
1917 __jbd2_journal_remove_checkpoint(jh);
1919 spin_unlock(&journal->j_list_lock);
1925 * int jbd2_journal_try_to_free_buffers() - try to free page buffers.
1926 * @journal: journal for operation
1927 * @page: to try and free
1928 * @gfp_mask: we use the mask to detect how hard should we try to release
1929 * buffers. If __GFP_DIRECT_RECLAIM and __GFP_FS is set, we wait for commit
1930 * code to release the buffers.
1933 * For all the buffers on this page,
1934 * if they are fully written out ordered data, move them onto BUF_CLEAN
1935 * so try_to_free_buffers() can reap them.
1937 * This function returns non-zero if we wish try_to_free_buffers()
1938 * to be called. We do this if the page is releasable by try_to_free_buffers().
1939 * We also do it if the page has locked or dirty buffers and the caller wants
1940 * us to perform sync or async writeout.
1942 * This complicates JBD locking somewhat. We aren't protected by the
1943 * BKL here. We wish to remove the buffer from its committing or
1944 * running transaction's ->t_datalist via __jbd2_journal_unfile_buffer.
1946 * This may *change* the value of transaction_t->t_datalist, so anyone
1947 * who looks at t_datalist needs to lock against this function.
1949 * Even worse, someone may be doing a jbd2_journal_dirty_data on this
1950 * buffer. So we need to lock against that. jbd2_journal_dirty_data()
1951 * will come out of the lock with the buffer dirty, which makes it
1952 * ineligible for release here.
1954 * Who else is affected by this? hmm... Really the only contender
1955 * is do_get_write_access() - it could be looking at the buffer while
1956 * journal_try_to_free_buffer() is changing its state. But that
1957 * cannot happen because we never reallocate freed data as metadata
1958 * while the data is part of a transaction. Yes?
1960 * Return 0 on failure, 1 on success
1962 int jbd2_journal_try_to_free_buffers(journal_t *journal,
1963 struct page *page, gfp_t gfp_mask)
1965 struct buffer_head *head;
1966 struct buffer_head *bh;
1969 J_ASSERT(PageLocked(page));
1971 head = page_buffers(page);
1974 struct journal_head *jh;
1977 * We take our own ref against the journal_head here to avoid
1978 * having to add tons of locking around each instance of
1979 * jbd2_journal_put_journal_head().
1981 jh = jbd2_journal_grab_journal_head(bh);
1985 jbd_lock_bh_state(bh);
1986 __journal_try_to_free_buffer(journal, bh);
1987 jbd2_journal_put_journal_head(jh);
1988 jbd_unlock_bh_state(bh);
1991 } while ((bh = bh->b_this_page) != head);
1993 ret = try_to_free_buffers(page);
2000 * This buffer is no longer needed. If it is on an older transaction's
2001 * checkpoint list we need to record it on this transaction's forget list
2002 * to pin this buffer (and hence its checkpointing transaction) down until
2003 * this transaction commits. If the buffer isn't on a checkpoint list, we
2005 * Returns non-zero if JBD no longer has an interest in the buffer.
2007 * Called under j_list_lock.
2009 * Called under jbd_lock_bh_state(bh).
2011 static int __dispose_buffer(struct journal_head *jh, transaction_t *transaction)
2014 struct buffer_head *bh = jh2bh(jh);
2016 if (jh->b_cp_transaction) {
2017 JBUFFER_TRACE(jh, "on running+cp transaction");
2018 __jbd2_journal_temp_unlink_buffer(jh);
2020 * We don't want to write the buffer anymore, clear the
2021 * bit so that we don't confuse checks in
2022 * __journal_file_buffer
2024 clear_buffer_dirty(bh);
2025 __jbd2_journal_file_buffer(jh, transaction, BJ_Forget);
2028 JBUFFER_TRACE(jh, "on running transaction");
2029 __jbd2_journal_unfile_buffer(jh);
2035 * jbd2_journal_invalidatepage
2037 * This code is tricky. It has a number of cases to deal with.
2039 * There are two invariants which this code relies on:
2041 * i_size must be updated on disk before we start calling invalidatepage on the
2044 * This is done in ext3 by defining an ext3_setattr method which
2045 * updates i_size before truncate gets going. By maintaining this
2046 * invariant, we can be sure that it is safe to throw away any buffers
2047 * attached to the current transaction: once the transaction commits,
2048 * we know that the data will not be needed.
2050 * Note however that we can *not* throw away data belonging to the
2051 * previous, committing transaction!
2053 * Any disk blocks which *are* part of the previous, committing
2054 * transaction (and which therefore cannot be discarded immediately) are
2055 * not going to be reused in the new running transaction
2057 * The bitmap committed_data images guarantee this: any block which is
2058 * allocated in one transaction and removed in the next will be marked
2059 * as in-use in the committed_data bitmap, so cannot be reused until
2060 * the next transaction to delete the block commits. This means that
2061 * leaving committing buffers dirty is quite safe: the disk blocks
2062 * cannot be reallocated to a different file and so buffer aliasing is
2066 * The above applies mainly to ordered data mode. In writeback mode we
2067 * don't make guarantees about the order in which data hits disk --- in
2068 * particular we don't guarantee that new dirty data is flushed before
2069 * transaction commit --- so it is always safe just to discard data
2070 * immediately in that mode. --sct
2074 * The journal_unmap_buffer helper function returns zero if the buffer
2075 * concerned remains pinned as an anonymous buffer belonging to an older
2078 * We're outside-transaction here. Either or both of j_running_transaction
2079 * and j_committing_transaction may be NULL.
2081 static int journal_unmap_buffer(journal_t *journal, struct buffer_head *bh,
2084 transaction_t *transaction;
2085 struct journal_head *jh;
2088 BUFFER_TRACE(bh, "entry");
2091 * It is safe to proceed here without the j_list_lock because the
2092 * buffers cannot be stolen by try_to_free_buffers as long as we are
2093 * holding the page lock. --sct
2096 if (!buffer_jbd(bh))
2097 goto zap_buffer_unlocked;
2099 /* OK, we have data buffer in journaled mode */
2100 write_lock(&journal->j_state_lock);
2101 jbd_lock_bh_state(bh);
2102 spin_lock(&journal->j_list_lock);
2104 jh = jbd2_journal_grab_journal_head(bh);
2106 goto zap_buffer_no_jh;
2109 * We cannot remove the buffer from checkpoint lists until the
2110 * transaction adding inode to orphan list (let's call it T)
2111 * is committed. Otherwise if the transaction changing the
2112 * buffer would be cleaned from the journal before T is
2113 * committed, a crash will cause that the correct contents of
2114 * the buffer will be lost. On the other hand we have to
2115 * clear the buffer dirty bit at latest at the moment when the
2116 * transaction marking the buffer as freed in the filesystem
2117 * structures is committed because from that moment on the
2118 * block can be reallocated and used by a different page.
2119 * Since the block hasn't been freed yet but the inode has
2120 * already been added to orphan list, it is safe for us to add
2121 * the buffer to BJ_Forget list of the newest transaction.
2123 * Also we have to clear buffer_mapped flag of a truncated buffer
2124 * because the buffer_head may be attached to the page straddling
2125 * i_size (can happen only when blocksize < pagesize) and thus the
2126 * buffer_head can be reused when the file is extended again. So we end
2127 * up keeping around invalidated buffers attached to transactions'
2128 * BJ_Forget list just to stop checkpointing code from cleaning up
2129 * the transaction this buffer was modified in.
2131 transaction = jh->b_transaction;
2132 if (transaction == NULL) {
2133 /* First case: not on any transaction. If it
2134 * has no checkpoint link, then we can zap it:
2135 * it's a writeback-mode buffer so we don't care
2136 * if it hits disk safely. */
2137 if (!jh->b_cp_transaction) {
2138 JBUFFER_TRACE(jh, "not on any transaction: zap");
2142 if (!buffer_dirty(bh)) {
2143 /* bdflush has written it. We can drop it now */
2144 __jbd2_journal_remove_checkpoint(jh);
2148 /* OK, it must be in the journal but still not
2149 * written fully to disk: it's metadata or
2150 * journaled data... */
2152 if (journal->j_running_transaction) {
2153 /* ... and once the current transaction has
2154 * committed, the buffer won't be needed any
2156 JBUFFER_TRACE(jh, "checkpointed: add to BJ_Forget");
2157 may_free = __dispose_buffer(jh,
2158 journal->j_running_transaction);
2161 /* There is no currently-running transaction. So the
2162 * orphan record which we wrote for this file must have
2163 * passed into commit. We must attach this buffer to
2164 * the committing transaction, if it exists. */
2165 if (journal->j_committing_transaction) {
2166 JBUFFER_TRACE(jh, "give to committing trans");
2167 may_free = __dispose_buffer(jh,
2168 journal->j_committing_transaction);
2171 /* The orphan record's transaction has
2172 * committed. We can cleanse this buffer */
2173 clear_buffer_jbddirty(bh);
2174 __jbd2_journal_remove_checkpoint(jh);
2178 } else if (transaction == journal->j_committing_transaction) {
2179 JBUFFER_TRACE(jh, "on committing transaction");
2181 * The buffer is committing, we simply cannot touch
2182 * it. If the page is straddling i_size we have to wait
2183 * for commit and try again.
2186 jbd2_journal_put_journal_head(jh);
2187 spin_unlock(&journal->j_list_lock);
2188 jbd_unlock_bh_state(bh);
2189 write_unlock(&journal->j_state_lock);
2193 * OK, buffer won't be reachable after truncate. We just set
2194 * j_next_transaction to the running transaction (if there is
2195 * one) and mark buffer as freed so that commit code knows it
2196 * should clear dirty bits when it is done with the buffer.
2198 set_buffer_freed(bh);
2199 if (journal->j_running_transaction && buffer_jbddirty(bh))
2200 jh->b_next_transaction = journal->j_running_transaction;
2201 jbd2_journal_put_journal_head(jh);
2202 spin_unlock(&journal->j_list_lock);
2203 jbd_unlock_bh_state(bh);
2204 write_unlock(&journal->j_state_lock);
2207 /* Good, the buffer belongs to the running transaction.
2208 * We are writing our own transaction's data, not any
2209 * previous one's, so it is safe to throw it away
2210 * (remember that we expect the filesystem to have set
2211 * i_size already for this truncate so recovery will not
2212 * expose the disk blocks we are discarding here.) */
2213 J_ASSERT_JH(jh, transaction == journal->j_running_transaction);
2214 JBUFFER_TRACE(jh, "on running transaction");
2215 may_free = __dispose_buffer(jh, transaction);
2220 * This is tricky. Although the buffer is truncated, it may be reused
2221 * if blocksize < pagesize and it is attached to the page straddling
2222 * EOF. Since the buffer might have been added to BJ_Forget list of the
2223 * running transaction, journal_get_write_access() won't clear
2224 * b_modified and credit accounting gets confused. So clear b_modified
2228 jbd2_journal_put_journal_head(jh);
2230 spin_unlock(&journal->j_list_lock);
2231 jbd_unlock_bh_state(bh);
2232 write_unlock(&journal->j_state_lock);
2233 zap_buffer_unlocked:
2234 clear_buffer_dirty(bh);
2235 J_ASSERT_BH(bh, !buffer_jbddirty(bh));
2236 clear_buffer_mapped(bh);
2237 clear_buffer_req(bh);
2238 clear_buffer_new(bh);
2239 clear_buffer_delay(bh);
2240 clear_buffer_unwritten(bh);
2246 * void jbd2_journal_invalidatepage()
2247 * @journal: journal to use for flush...
2248 * @page: page to flush
2249 * @offset: start of the range to invalidate
2250 * @length: length of the range to invalidate
2252 * Reap page buffers containing data after in the specified range in page.
2253 * Can return -EBUSY if buffers are part of the committing transaction and
2254 * the page is straddling i_size. Caller then has to wait for current commit
2257 int jbd2_journal_invalidatepage(journal_t *journal,
2259 unsigned int offset,
2260 unsigned int length)
2262 struct buffer_head *head, *bh, *next;
2263 unsigned int stop = offset + length;
2264 unsigned int curr_off = 0;
2265 int partial_page = (offset || length < PAGE_SIZE);
2269 if (!PageLocked(page))
2271 if (!page_has_buffers(page))
2274 BUG_ON(stop > PAGE_SIZE || stop < length);
2276 /* We will potentially be playing with lists other than just the
2277 * data lists (especially for journaled data mode), so be
2278 * cautious in our locking. */
2280 head = bh = page_buffers(page);
2282 unsigned int next_off = curr_off + bh->b_size;
2283 next = bh->b_this_page;
2285 if (next_off > stop)
2288 if (offset <= curr_off) {
2289 /* This block is wholly outside the truncation point */
2291 ret = journal_unmap_buffer(journal, bh, partial_page);
2297 curr_off = next_off;
2300 } while (bh != head);
2302 if (!partial_page) {
2303 if (may_free && try_to_free_buffers(page))
2304 J_ASSERT(!page_has_buffers(page));
2310 * File a buffer on the given transaction list.
2312 void __jbd2_journal_file_buffer(struct journal_head *jh,
2313 transaction_t *transaction, int jlist)
2315 struct journal_head **list = NULL;
2317 struct buffer_head *bh = jh2bh(jh);
2319 J_ASSERT_JH(jh, jbd_is_locked_bh_state(bh));
2320 assert_spin_locked(&transaction->t_journal->j_list_lock);
2322 J_ASSERT_JH(jh, jh->b_jlist < BJ_Types);
2323 J_ASSERT_JH(jh, jh->b_transaction == transaction ||
2324 jh->b_transaction == NULL);
2326 if (jh->b_transaction && jh->b_jlist == jlist)
2329 if (jlist == BJ_Metadata || jlist == BJ_Reserved ||
2330 jlist == BJ_Shadow || jlist == BJ_Forget) {
2332 * For metadata buffers, we track dirty bit in buffer_jbddirty
2333 * instead of buffer_dirty. We should not see a dirty bit set
2334 * here because we clear it in do_get_write_access but e.g.
2335 * tune2fs can modify the sb and set the dirty bit at any time
2336 * so we try to gracefully handle that.
2338 if (buffer_dirty(bh))
2339 warn_dirty_buffer(bh);
2340 if (test_clear_buffer_dirty(bh) ||
2341 test_clear_buffer_jbddirty(bh))
2345 if (jh->b_transaction)
2346 __jbd2_journal_temp_unlink_buffer(jh);
2348 jbd2_journal_grab_journal_head(bh);
2349 jh->b_transaction = transaction;
2353 J_ASSERT_JH(jh, !jh->b_committed_data);
2354 J_ASSERT_JH(jh, !jh->b_frozen_data);
2357 transaction->t_nr_buffers++;
2358 list = &transaction->t_buffers;
2361 list = &transaction->t_forget;
2364 list = &transaction->t_shadow_list;
2367 list = &transaction->t_reserved_list;
2371 __blist_add_buffer(list, jh);
2372 jh->b_jlist = jlist;
2375 set_buffer_jbddirty(bh);
2378 void jbd2_journal_file_buffer(struct journal_head *jh,
2379 transaction_t *transaction, int jlist)
2381 jbd_lock_bh_state(jh2bh(jh));
2382 spin_lock(&transaction->t_journal->j_list_lock);
2383 __jbd2_journal_file_buffer(jh, transaction, jlist);
2384 spin_unlock(&transaction->t_journal->j_list_lock);
2385 jbd_unlock_bh_state(jh2bh(jh));
2389 * Remove a buffer from its current buffer list in preparation for
2390 * dropping it from its current transaction entirely. If the buffer has
2391 * already started to be used by a subsequent transaction, refile the
2392 * buffer on that transaction's metadata list.
2394 * Called under j_list_lock
2395 * Called under jbd_lock_bh_state(jh2bh(jh))
2397 * jh and bh may be already free when this function returns
2399 void __jbd2_journal_refile_buffer(struct journal_head *jh)
2401 int was_dirty, jlist;
2402 struct buffer_head *bh = jh2bh(jh);
2404 J_ASSERT_JH(jh, jbd_is_locked_bh_state(bh));
2405 if (jh->b_transaction)
2406 assert_spin_locked(&jh->b_transaction->t_journal->j_list_lock);
2408 /* If the buffer is now unused, just drop it. */
2409 if (jh->b_next_transaction == NULL) {
2410 __jbd2_journal_unfile_buffer(jh);
2415 * It has been modified by a later transaction: add it to the new
2416 * transaction's metadata list.
2419 was_dirty = test_clear_buffer_jbddirty(bh);
2420 __jbd2_journal_temp_unlink_buffer(jh);
2422 * We set b_transaction here because b_next_transaction will inherit
2423 * our jh reference and thus __jbd2_journal_file_buffer() must not
2426 jh->b_transaction = jh->b_next_transaction;
2427 jh->b_next_transaction = NULL;
2428 if (buffer_freed(bh))
2430 else if (jh->b_modified)
2431 jlist = BJ_Metadata;
2433 jlist = BJ_Reserved;
2434 __jbd2_journal_file_buffer(jh, jh->b_transaction, jlist);
2435 J_ASSERT_JH(jh, jh->b_transaction->t_state == T_RUNNING);
2438 set_buffer_jbddirty(bh);
2442 * __jbd2_journal_refile_buffer() with necessary locking added. We take our
2443 * bh reference so that we can safely unlock bh.
2445 * The jh and bh may be freed by this call.
2447 void jbd2_journal_refile_buffer(journal_t *journal, struct journal_head *jh)
2449 struct buffer_head *bh = jh2bh(jh);
2451 /* Get reference so that buffer cannot be freed before we unlock it */
2453 jbd_lock_bh_state(bh);
2454 spin_lock(&journal->j_list_lock);
2455 __jbd2_journal_refile_buffer(jh);
2456 jbd_unlock_bh_state(bh);
2457 spin_unlock(&journal->j_list_lock);
2462 * File inode in the inode list of the handle's transaction
2464 static int jbd2_journal_file_inode(handle_t *handle, struct jbd2_inode *jinode,
2465 unsigned long flags)
2467 transaction_t *transaction = handle->h_transaction;
2470 if (is_handle_aborted(handle))
2472 journal = transaction->t_journal;
2474 jbd_debug(4, "Adding inode %lu, tid:%d\n", jinode->i_vfs_inode->i_ino,
2475 transaction->t_tid);
2478 * First check whether inode isn't already on the transaction's
2479 * lists without taking the lock. Note that this check is safe
2480 * without the lock as we cannot race with somebody removing inode
2481 * from the transaction. The reason is that we remove inode from the
2482 * transaction only in journal_release_jbd_inode() and when we commit
2483 * the transaction. We are guarded from the first case by holding
2484 * a reference to the inode. We are safe against the second case
2485 * because if jinode->i_transaction == transaction, commit code
2486 * cannot touch the transaction because we hold reference to it,
2487 * and if jinode->i_next_transaction == transaction, commit code
2488 * will only file the inode where we want it.
2490 if ((jinode->i_transaction == transaction ||
2491 jinode->i_next_transaction == transaction) &&
2492 (jinode->i_flags & flags) == flags)
2495 spin_lock(&journal->j_list_lock);
2496 jinode->i_flags |= flags;
2497 /* Is inode already attached where we need it? */
2498 if (jinode->i_transaction == transaction ||
2499 jinode->i_next_transaction == transaction)
2503 * We only ever set this variable to 1 so the test is safe. Since
2504 * t_need_data_flush is likely to be set, we do the test to save some
2505 * cacheline bouncing
2507 if (!transaction->t_need_data_flush)
2508 transaction->t_need_data_flush = 1;
2509 /* On some different transaction's list - should be
2510 * the committing one */
2511 if (jinode->i_transaction) {
2512 J_ASSERT(jinode->i_next_transaction == NULL);
2513 J_ASSERT(jinode->i_transaction ==
2514 journal->j_committing_transaction);
2515 jinode->i_next_transaction = transaction;
2518 /* Not on any transaction list... */
2519 J_ASSERT(!jinode->i_next_transaction);
2520 jinode->i_transaction = transaction;
2521 list_add(&jinode->i_list, &transaction->t_inode_list);
2523 spin_unlock(&journal->j_list_lock);
2528 int jbd2_journal_inode_add_write(handle_t *handle, struct jbd2_inode *jinode)
2530 return jbd2_journal_file_inode(handle, jinode,
2531 JI_WRITE_DATA | JI_WAIT_DATA);
2534 int jbd2_journal_inode_add_wait(handle_t *handle, struct jbd2_inode *jinode)
2536 return jbd2_journal_file_inode(handle, jinode, JI_WAIT_DATA);
2540 * File truncate and transaction commit interact with each other in a
2541 * non-trivial way. If a transaction writing data block A is
2542 * committing, we cannot discard the data by truncate until we have
2543 * written them. Otherwise if we crashed after the transaction with
2544 * write has committed but before the transaction with truncate has
2545 * committed, we could see stale data in block A. This function is a
2546 * helper to solve this problem. It starts writeout of the truncated
2547 * part in case it is in the committing transaction.
2549 * Filesystem code must call this function when inode is journaled in
2550 * ordered mode before truncation happens and after the inode has been
2551 * placed on orphan list with the new inode size. The second condition
2552 * avoids the race that someone writes new data and we start
2553 * committing the transaction after this function has been called but
2554 * before a transaction for truncate is started (and furthermore it
2555 * allows us to optimize the case where the addition to orphan list
2556 * happens in the same transaction as write --- we don't have to write
2557 * any data in such case).
2559 int jbd2_journal_begin_ordered_truncate(journal_t *journal,
2560 struct jbd2_inode *jinode,
2563 transaction_t *inode_trans, *commit_trans;
2566 /* This is a quick check to avoid locking if not necessary */
2567 if (!jinode->i_transaction)
2569 /* Locks are here just to force reading of recent values, it is
2570 * enough that the transaction was not committing before we started
2571 * a transaction adding the inode to orphan list */
2572 read_lock(&journal->j_state_lock);
2573 commit_trans = journal->j_committing_transaction;
2574 read_unlock(&journal->j_state_lock);
2575 spin_lock(&journal->j_list_lock);
2576 inode_trans = jinode->i_transaction;
2577 spin_unlock(&journal->j_list_lock);
2578 if (inode_trans == commit_trans) {
2579 ret = filemap_fdatawrite_range(jinode->i_vfs_inode->i_mapping,
2580 new_size, LLONG_MAX);
2582 jbd2_journal_abort(journal, ret);