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);
162 jbd2_might_wait_for_commit(journal);
164 finish_wait(&journal->j_wait_transaction_locked, &wait);
167 static void sub_reserved_credits(journal_t *journal, int blocks)
169 atomic_sub(blocks, &journal->j_reserved_credits);
170 wake_up(&journal->j_wait_reserved);
174 * Wait until we can add credits for handle to the running transaction. Called
175 * with j_state_lock held for reading. Returns 0 if handle joined the running
176 * transaction. Returns 1 if we had to wait, j_state_lock is dropped, and
179 static int add_transaction_credits(journal_t *journal, int blocks,
182 transaction_t *t = journal->j_running_transaction;
184 int total = blocks + rsv_blocks;
187 * If the current transaction is locked down for commit, wait
188 * for the lock to be released.
190 if (t->t_state == T_LOCKED) {
191 wait_transaction_locked(journal);
196 * If there is not enough space left in the log to write all
197 * potential buffers requested by this operation, we need to
198 * stall pending a log checkpoint to free some more log space.
200 needed = atomic_add_return(total, &t->t_outstanding_credits);
201 if (needed > journal->j_max_transaction_buffers) {
203 * If the current transaction is already too large,
204 * then start to commit it: we can then go back and
205 * attach this handle to a new transaction.
207 atomic_sub(total, &t->t_outstanding_credits);
210 * Is the number of reserved credits in the current transaction too
211 * big to fit this handle? Wait until reserved credits are freed.
213 if (atomic_read(&journal->j_reserved_credits) + total >
214 journal->j_max_transaction_buffers) {
215 read_unlock(&journal->j_state_lock);
216 jbd2_might_wait_for_commit(journal);
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 jbd2_might_wait_for_commit(journal);
242 write_lock(&journal->j_state_lock);
243 if (jbd2_log_space_left(journal) < jbd2_space_needed(journal))
244 __jbd2_log_wait_for_space(journal);
245 write_unlock(&journal->j_state_lock);
249 /* No reservation? We are done... */
253 needed = atomic_add_return(rsv_blocks, &journal->j_reserved_credits);
254 /* We allow at most half of a transaction to be reserved */
255 if (needed > journal->j_max_transaction_buffers / 2) {
256 sub_reserved_credits(journal, rsv_blocks);
257 atomic_sub(total, &t->t_outstanding_credits);
258 read_unlock(&journal->j_state_lock);
259 jbd2_might_wait_for_commit(journal);
260 wait_event(journal->j_wait_reserved,
261 atomic_read(&journal->j_reserved_credits) + rsv_blocks
262 <= journal->j_max_transaction_buffers / 2);
269 * start_this_handle: Given a handle, deal with any locking or stalling
270 * needed to make sure that there is enough journal space for the handle
271 * to begin. Attach the handle to a transaction and set up the
272 * transaction's buffer credits.
275 static int start_this_handle(journal_t *journal, handle_t *handle,
278 transaction_t *transaction, *new_transaction = NULL;
279 int blocks = handle->h_buffer_credits;
281 unsigned long ts = jiffies;
283 if (handle->h_rsv_handle)
284 rsv_blocks = handle->h_rsv_handle->h_buffer_credits;
287 * Limit the number of reserved credits to 1/2 of maximum transaction
288 * size and limit the number of total credits to not exceed maximum
289 * transaction size per operation.
291 if ((rsv_blocks > journal->j_max_transaction_buffers / 2) ||
292 (rsv_blocks + blocks > journal->j_max_transaction_buffers)) {
293 printk(KERN_ERR "JBD2: %s wants too many credits "
294 "credits:%d rsv_credits:%d max:%d\n",
295 current->comm, blocks, rsv_blocks,
296 journal->j_max_transaction_buffers);
302 if (!journal->j_running_transaction) {
304 * If __GFP_FS is not present, then we may be being called from
305 * inside the fs writeback layer, so we MUST NOT fail.
307 if ((gfp_mask & __GFP_FS) == 0)
308 gfp_mask |= __GFP_NOFAIL;
309 new_transaction = kmem_cache_zalloc(transaction_cache,
311 if (!new_transaction)
315 jbd_debug(3, "New handle %p going live.\n", handle);
318 * We need to hold j_state_lock until t_updates has been incremented,
319 * for proper journal barrier handling
322 read_lock(&journal->j_state_lock);
323 BUG_ON(journal->j_flags & JBD2_UNMOUNT);
324 if (is_journal_aborted(journal) ||
325 (journal->j_errno != 0 && !(journal->j_flags & JBD2_ACK_ERR))) {
326 read_unlock(&journal->j_state_lock);
327 jbd2_journal_free_transaction(new_transaction);
332 * Wait on the journal's transaction barrier if necessary. Specifically
333 * we allow reserved handles to proceed because otherwise commit could
334 * deadlock on page writeback not being able to complete.
336 if (!handle->h_reserved && journal->j_barrier_count) {
337 read_unlock(&journal->j_state_lock);
338 wait_event(journal->j_wait_transaction_locked,
339 journal->j_barrier_count == 0);
343 if (!journal->j_running_transaction) {
344 read_unlock(&journal->j_state_lock);
345 if (!new_transaction)
346 goto alloc_transaction;
347 write_lock(&journal->j_state_lock);
348 if (!journal->j_running_transaction &&
349 (handle->h_reserved || !journal->j_barrier_count)) {
350 jbd2_get_transaction(journal, new_transaction);
351 new_transaction = NULL;
353 write_unlock(&journal->j_state_lock);
357 transaction = journal->j_running_transaction;
359 if (!handle->h_reserved) {
360 /* We may have dropped j_state_lock - restart in that case */
361 if (add_transaction_credits(journal, blocks, rsv_blocks))
365 * We have handle reserved so we are allowed to join T_LOCKED
366 * transaction and we don't have to check for transaction size
369 sub_reserved_credits(journal, blocks);
370 handle->h_reserved = 0;
373 /* OK, account for the buffers that this operation expects to
374 * use and add the handle to the running transaction.
376 update_t_max_wait(transaction, ts);
377 handle->h_transaction = transaction;
378 handle->h_requested_credits = blocks;
379 handle->h_start_jiffies = jiffies;
380 atomic_inc(&transaction->t_updates);
381 atomic_inc(&transaction->t_handle_count);
382 jbd_debug(4, "Handle %p given %d credits (total %d, free %lu)\n",
384 atomic_read(&transaction->t_outstanding_credits),
385 jbd2_log_space_left(journal));
386 read_unlock(&journal->j_state_lock);
387 current->journal_info = handle;
389 rwsem_acquire_read(&journal->j_trans_commit_map, 0, 0, _THIS_IP_);
390 jbd2_journal_free_transaction(new_transaction);
394 /* Allocate a new handle. This should probably be in a slab... */
395 static handle_t *new_handle(int nblocks)
397 handle_t *handle = jbd2_alloc_handle(GFP_NOFS);
400 handle->h_buffer_credits = nblocks;
407 * handle_t *jbd2_journal_start() - Obtain a new handle.
408 * @journal: Journal to start transaction on.
409 * @nblocks: number of block buffer we might modify
411 * We make sure that the transaction can guarantee at least nblocks of
412 * modified buffers in the log. We block until the log can guarantee
413 * that much space. Additionally, if rsv_blocks > 0, we also create another
414 * handle with rsv_blocks reserved blocks in the journal. This handle is
415 * is stored in h_rsv_handle. It is not attached to any particular transaction
416 * and thus doesn't block transaction commit. If the caller uses this reserved
417 * handle, it has to set h_rsv_handle to NULL as otherwise jbd2_journal_stop()
418 * on the parent handle will dispose the reserved one. Reserved handle has to
419 * be converted to a normal handle using jbd2_journal_start_reserved() before
422 * Return a pointer to a newly allocated handle, or an ERR_PTR() value
425 handle_t *jbd2__journal_start(journal_t *journal, int nblocks, int rsv_blocks,
426 gfp_t gfp_mask, unsigned int type,
427 unsigned int line_no)
429 handle_t *handle = journal_current_handle();
433 return ERR_PTR(-EROFS);
436 J_ASSERT(handle->h_transaction->t_journal == journal);
441 handle = new_handle(nblocks);
443 return ERR_PTR(-ENOMEM);
445 handle_t *rsv_handle;
447 rsv_handle = new_handle(rsv_blocks);
449 jbd2_free_handle(handle);
450 return ERR_PTR(-ENOMEM);
452 rsv_handle->h_reserved = 1;
453 rsv_handle->h_journal = journal;
454 handle->h_rsv_handle = rsv_handle;
457 err = start_this_handle(journal, handle, gfp_mask);
459 if (handle->h_rsv_handle)
460 jbd2_free_handle(handle->h_rsv_handle);
461 jbd2_free_handle(handle);
464 handle->h_type = type;
465 handle->h_line_no = line_no;
466 trace_jbd2_handle_start(journal->j_fs_dev->bd_dev,
467 handle->h_transaction->t_tid, type,
471 EXPORT_SYMBOL(jbd2__journal_start);
474 handle_t *jbd2_journal_start(journal_t *journal, int nblocks)
476 return jbd2__journal_start(journal, nblocks, 0, GFP_NOFS, 0, 0);
478 EXPORT_SYMBOL(jbd2_journal_start);
480 void jbd2_journal_free_reserved(handle_t *handle)
482 journal_t *journal = handle->h_journal;
484 WARN_ON(!handle->h_reserved);
485 sub_reserved_credits(journal, handle->h_buffer_credits);
486 jbd2_free_handle(handle);
488 EXPORT_SYMBOL(jbd2_journal_free_reserved);
491 * int jbd2_journal_start_reserved(handle_t *handle) - start reserved handle
492 * @handle: handle to start
494 * Start handle that has been previously reserved with jbd2_journal_reserve().
495 * This attaches @handle to the running transaction (or creates one if there's
496 * not transaction running). Unlike jbd2_journal_start() this function cannot
497 * block on journal commit, checkpointing, or similar stuff. It can block on
498 * memory allocation or frozen journal though.
500 * Return 0 on success, non-zero on error - handle is freed in that case.
502 int jbd2_journal_start_reserved(handle_t *handle, unsigned int type,
503 unsigned int line_no)
505 journal_t *journal = handle->h_journal;
508 if (WARN_ON(!handle->h_reserved)) {
509 /* Someone passed in normal handle? Just stop it. */
510 jbd2_journal_stop(handle);
514 * Usefulness of mixing of reserved and unreserved handles is
515 * questionable. So far nobody seems to need it so just error out.
517 if (WARN_ON(current->journal_info)) {
518 jbd2_journal_free_reserved(handle);
522 handle->h_journal = NULL;
524 * GFP_NOFS is here because callers are likely from writeback or
525 * similarly constrained call sites
527 ret = start_this_handle(journal, handle, GFP_NOFS);
529 jbd2_journal_free_reserved(handle);
532 handle->h_type = type;
533 handle->h_line_no = line_no;
536 EXPORT_SYMBOL(jbd2_journal_start_reserved);
539 * int jbd2_journal_extend() - extend buffer credits.
540 * @handle: handle to 'extend'
541 * @nblocks: nr blocks to try to extend by.
543 * Some transactions, such as large extends and truncates, can be done
544 * atomically all at once or in several stages. The operation requests
545 * a credit for a number of buffer modifications in advance, but can
546 * extend its credit if it needs more.
548 * jbd2_journal_extend tries to give the running handle more buffer credits.
549 * It does not guarantee that allocation - this is a best-effort only.
550 * The calling process MUST be able to deal cleanly with a failure to
553 * Return 0 on success, non-zero on failure.
555 * return code < 0 implies an error
556 * return code > 0 implies normal transaction-full status.
558 int jbd2_journal_extend(handle_t *handle, int nblocks)
560 transaction_t *transaction = handle->h_transaction;
565 if (is_handle_aborted(handle))
567 journal = transaction->t_journal;
571 read_lock(&journal->j_state_lock);
573 /* Don't extend a locked-down transaction! */
574 if (transaction->t_state != T_RUNNING) {
575 jbd_debug(3, "denied handle %p %d blocks: "
576 "transaction not running\n", handle, nblocks);
580 spin_lock(&transaction->t_handle_lock);
581 wanted = atomic_add_return(nblocks,
582 &transaction->t_outstanding_credits);
584 if (wanted > journal->j_max_transaction_buffers) {
585 jbd_debug(3, "denied handle %p %d blocks: "
586 "transaction too large\n", handle, nblocks);
587 atomic_sub(nblocks, &transaction->t_outstanding_credits);
591 if (wanted + (wanted >> JBD2_CONTROL_BLOCKS_SHIFT) >
592 jbd2_log_space_left(journal)) {
593 jbd_debug(3, "denied handle %p %d blocks: "
594 "insufficient log space\n", handle, nblocks);
595 atomic_sub(nblocks, &transaction->t_outstanding_credits);
599 trace_jbd2_handle_extend(journal->j_fs_dev->bd_dev,
601 handle->h_type, handle->h_line_no,
602 handle->h_buffer_credits,
605 handle->h_buffer_credits += nblocks;
606 handle->h_requested_credits += nblocks;
609 jbd_debug(3, "extended handle %p by %d\n", handle, nblocks);
611 spin_unlock(&transaction->t_handle_lock);
613 read_unlock(&journal->j_state_lock);
619 * int jbd2_journal_restart() - restart a handle .
620 * @handle: handle to restart
621 * @nblocks: nr credits requested
623 * Restart a handle for a multi-transaction filesystem
626 * If the jbd2_journal_extend() call above fails to grant new buffer credits
627 * to a running handle, a call to jbd2_journal_restart will commit the
628 * handle's transaction so far and reattach the handle to a new
629 * transaction capable of guaranteeing the requested number of
630 * credits. We preserve reserved handle if there's any attached to the
633 int jbd2__journal_restart(handle_t *handle, int nblocks, gfp_t gfp_mask)
635 transaction_t *transaction = handle->h_transaction;
638 int need_to_start, ret;
640 /* If we've had an abort of any type, don't even think about
641 * actually doing the restart! */
642 if (is_handle_aborted(handle))
644 journal = transaction->t_journal;
647 * First unlink the handle from its current transaction, and start the
650 J_ASSERT(atomic_read(&transaction->t_updates) > 0);
651 J_ASSERT(journal_current_handle() == handle);
653 read_lock(&journal->j_state_lock);
654 spin_lock(&transaction->t_handle_lock);
655 atomic_sub(handle->h_buffer_credits,
656 &transaction->t_outstanding_credits);
657 if (handle->h_rsv_handle) {
658 sub_reserved_credits(journal,
659 handle->h_rsv_handle->h_buffer_credits);
661 if (atomic_dec_and_test(&transaction->t_updates))
662 wake_up(&journal->j_wait_updates);
663 tid = transaction->t_tid;
664 spin_unlock(&transaction->t_handle_lock);
665 handle->h_transaction = NULL;
666 current->journal_info = NULL;
668 jbd_debug(2, "restarting handle %p\n", handle);
669 need_to_start = !tid_geq(journal->j_commit_request, tid);
670 read_unlock(&journal->j_state_lock);
672 jbd2_log_start_commit(journal, tid);
674 rwsem_release(&journal->j_trans_commit_map, 1, _THIS_IP_);
675 handle->h_buffer_credits = nblocks;
676 ret = start_this_handle(journal, handle, gfp_mask);
679 EXPORT_SYMBOL(jbd2__journal_restart);
682 int jbd2_journal_restart(handle_t *handle, int nblocks)
684 return jbd2__journal_restart(handle, nblocks, GFP_NOFS);
686 EXPORT_SYMBOL(jbd2_journal_restart);
689 * void jbd2_journal_lock_updates () - establish a transaction barrier.
690 * @journal: Journal to establish a barrier on.
692 * This locks out any further updates from being started, and blocks
693 * until all existing updates have completed, returning only once the
694 * journal is in a quiescent state with no updates running.
696 * The journal lock should not be held on entry.
698 void jbd2_journal_lock_updates(journal_t *journal)
702 jbd2_might_wait_for_commit(journal);
704 write_lock(&journal->j_state_lock);
705 ++journal->j_barrier_count;
707 /* Wait until there are no reserved handles */
708 if (atomic_read(&journal->j_reserved_credits)) {
709 write_unlock(&journal->j_state_lock);
710 wait_event(journal->j_wait_reserved,
711 atomic_read(&journal->j_reserved_credits) == 0);
712 write_lock(&journal->j_state_lock);
715 /* Wait until there are no running updates */
717 transaction_t *transaction = journal->j_running_transaction;
722 spin_lock(&transaction->t_handle_lock);
723 prepare_to_wait(&journal->j_wait_updates, &wait,
724 TASK_UNINTERRUPTIBLE);
725 if (!atomic_read(&transaction->t_updates)) {
726 spin_unlock(&transaction->t_handle_lock);
727 finish_wait(&journal->j_wait_updates, &wait);
730 spin_unlock(&transaction->t_handle_lock);
731 write_unlock(&journal->j_state_lock);
733 finish_wait(&journal->j_wait_updates, &wait);
734 write_lock(&journal->j_state_lock);
736 write_unlock(&journal->j_state_lock);
739 * We have now established a barrier against other normal updates, but
740 * we also need to barrier against other jbd2_journal_lock_updates() calls
741 * to make sure that we serialise special journal-locked operations
744 mutex_lock(&journal->j_barrier);
748 * void jbd2_journal_unlock_updates (journal_t* journal) - release barrier
749 * @journal: Journal to release the barrier on.
751 * Release a transaction barrier obtained with jbd2_journal_lock_updates().
753 * Should be called without the journal lock held.
755 void jbd2_journal_unlock_updates (journal_t *journal)
757 J_ASSERT(journal->j_barrier_count != 0);
759 mutex_unlock(&journal->j_barrier);
760 write_lock(&journal->j_state_lock);
761 --journal->j_barrier_count;
762 write_unlock(&journal->j_state_lock);
763 wake_up(&journal->j_wait_transaction_locked);
766 static void warn_dirty_buffer(struct buffer_head *bh)
769 "JBD2: Spotted dirty metadata buffer (dev = %pg, blocknr = %llu). "
770 "There's a risk of filesystem corruption in case of system "
772 bh->b_bdev, (unsigned long long)bh->b_blocknr);
775 /* Call t_frozen trigger and copy buffer data into jh->b_frozen_data. */
776 static void jbd2_freeze_jh_data(struct journal_head *jh)
781 struct buffer_head *bh = jh2bh(jh);
783 J_EXPECT_JH(jh, buffer_uptodate(bh), "Possible IO failure.\n");
785 offset = offset_in_page(bh->b_data);
786 source = kmap_atomic(page);
787 /* Fire data frozen trigger just before we copy the data */
788 jbd2_buffer_frozen_trigger(jh, source + offset, jh->b_triggers);
789 memcpy(jh->b_frozen_data, source + offset, bh->b_size);
790 kunmap_atomic(source);
793 * Now that the frozen data is saved off, we need to store any matching
796 jh->b_frozen_triggers = jh->b_triggers;
800 * If the buffer is already part of the current transaction, then there
801 * is nothing we need to do. If it is already part of a prior
802 * transaction which we are still committing to disk, then we need to
803 * make sure that we do not overwrite the old copy: we do copy-out to
804 * preserve the copy going to disk. We also account the buffer against
805 * the handle's metadata buffer credits (unless the buffer is already
806 * part of the transaction, that is).
810 do_get_write_access(handle_t *handle, struct journal_head *jh,
813 struct buffer_head *bh;
814 transaction_t *transaction = handle->h_transaction;
817 char *frozen_buffer = NULL;
818 unsigned long start_lock, time_lock;
820 if (is_handle_aborted(handle))
822 journal = transaction->t_journal;
824 jbd_debug(5, "journal_head %p, force_copy %d\n", jh, force_copy);
826 JBUFFER_TRACE(jh, "entry");
830 /* @@@ Need to check for errors here at some point. */
832 start_lock = jiffies;
834 jbd_lock_bh_state(bh);
836 /* If it takes too long to lock the buffer, trace it */
837 time_lock = jbd2_time_diff(start_lock, jiffies);
838 if (time_lock > HZ/10)
839 trace_jbd2_lock_buffer_stall(bh->b_bdev->bd_dev,
840 jiffies_to_msecs(time_lock));
842 /* We now hold the buffer lock so it is safe to query the buffer
843 * state. Is the buffer dirty?
845 * If so, there are two possibilities. The buffer may be
846 * non-journaled, and undergoing a quite legitimate writeback.
847 * Otherwise, it is journaled, and we don't expect dirty buffers
848 * in that state (the buffers should be marked JBD_Dirty
849 * instead.) So either the IO is being done under our own
850 * control and this is a bug, or it's a third party IO such as
851 * dump(8) (which may leave the buffer scheduled for read ---
852 * ie. locked but not dirty) or tune2fs (which may actually have
853 * the buffer dirtied, ugh.) */
855 if (buffer_dirty(bh)) {
857 * First question: is this buffer already part of the current
858 * transaction or the existing committing transaction?
860 if (jh->b_transaction) {
862 jh->b_transaction == transaction ||
864 journal->j_committing_transaction);
865 if (jh->b_next_transaction)
866 J_ASSERT_JH(jh, jh->b_next_transaction ==
868 warn_dirty_buffer(bh);
871 * In any case we need to clean the dirty flag and we must
872 * do it under the buffer lock to be sure we don't race
873 * with running write-out.
875 JBUFFER_TRACE(jh, "Journalling dirty buffer");
876 clear_buffer_dirty(bh);
877 set_buffer_jbddirty(bh);
883 if (is_handle_aborted(handle)) {
884 jbd_unlock_bh_state(bh);
890 * The buffer is already part of this transaction if b_transaction or
891 * b_next_transaction points to it
893 if (jh->b_transaction == transaction ||
894 jh->b_next_transaction == transaction)
898 * this is the first time this transaction is touching this buffer,
899 * reset the modified flag
904 * If the buffer is not journaled right now, we need to make sure it
905 * doesn't get written to disk before the caller actually commits the
908 if (!jh->b_transaction) {
909 JBUFFER_TRACE(jh, "no transaction");
910 J_ASSERT_JH(jh, !jh->b_next_transaction);
911 JBUFFER_TRACE(jh, "file as BJ_Reserved");
913 * Make sure all stores to jh (b_modified, b_frozen_data) are
914 * visible before attaching it to the running transaction.
915 * Paired with barrier in jbd2_write_access_granted()
918 spin_lock(&journal->j_list_lock);
919 __jbd2_journal_file_buffer(jh, transaction, BJ_Reserved);
920 spin_unlock(&journal->j_list_lock);
924 * If there is already a copy-out version of this buffer, then we don't
925 * need to make another one
927 if (jh->b_frozen_data) {
928 JBUFFER_TRACE(jh, "has frozen data");
929 J_ASSERT_JH(jh, jh->b_next_transaction == NULL);
933 JBUFFER_TRACE(jh, "owned by older transaction");
934 J_ASSERT_JH(jh, jh->b_next_transaction == NULL);
935 J_ASSERT_JH(jh, jh->b_transaction == journal->j_committing_transaction);
938 * There is one case we have to be very careful about. If the
939 * committing transaction is currently writing this buffer out to disk
940 * and has NOT made a copy-out, then we cannot modify the buffer
941 * contents at all right now. The essence of copy-out is that it is
942 * the extra copy, not the primary copy, which gets journaled. If the
943 * primary copy is already going to disk then we cannot do copy-out
946 if (buffer_shadow(bh)) {
947 JBUFFER_TRACE(jh, "on shadow: sleep");
948 jbd_unlock_bh_state(bh);
949 wait_on_bit_io(&bh->b_state, BH_Shadow, TASK_UNINTERRUPTIBLE);
954 * Only do the copy if the currently-owning transaction still needs it.
955 * If buffer isn't on BJ_Metadata list, the committing transaction is
956 * past that stage (here we use the fact that BH_Shadow is set under
957 * bh_state lock together with refiling to BJ_Shadow list and at this
958 * point we know the buffer doesn't have BH_Shadow set).
960 * Subtle point, though: if this is a get_undo_access, then we will be
961 * relying on the frozen_data to contain the new value of the
962 * committed_data record after the transaction, so we HAVE to force the
963 * frozen_data copy in that case.
965 if (jh->b_jlist == BJ_Metadata || force_copy) {
966 JBUFFER_TRACE(jh, "generate frozen data");
967 if (!frozen_buffer) {
968 JBUFFER_TRACE(jh, "allocate memory for buffer");
969 jbd_unlock_bh_state(bh);
970 frozen_buffer = jbd2_alloc(jh2bh(jh)->b_size,
971 GFP_NOFS | __GFP_NOFAIL);
974 jh->b_frozen_data = frozen_buffer;
975 frozen_buffer = NULL;
976 jbd2_freeze_jh_data(jh);
980 * Make sure all stores to jh (b_modified, b_frozen_data) are visible
981 * before attaching it to the running transaction. Paired with barrier
982 * in jbd2_write_access_granted()
985 jh->b_next_transaction = transaction;
988 jbd_unlock_bh_state(bh);
991 * If we are about to journal a buffer, then any revoke pending on it is
994 jbd2_journal_cancel_revoke(handle, jh);
997 if (unlikely(frozen_buffer)) /* It's usually NULL */
998 jbd2_free(frozen_buffer, bh->b_size);
1000 JBUFFER_TRACE(jh, "exit");
1004 /* Fast check whether buffer is already attached to the required transaction */
1005 static bool jbd2_write_access_granted(handle_t *handle, struct buffer_head *bh,
1008 struct journal_head *jh;
1011 /* Dirty buffers require special handling... */
1012 if (buffer_dirty(bh))
1016 * RCU protects us from dereferencing freed pages. So the checks we do
1017 * are guaranteed not to oops. However the jh slab object can get freed
1018 * & reallocated while we work with it. So we have to be careful. When
1019 * we see jh attached to the running transaction, we know it must stay
1020 * so until the transaction is committed. Thus jh won't be freed and
1021 * will be attached to the same bh while we run. However it can
1022 * happen jh gets freed, reallocated, and attached to the transaction
1023 * just after we get pointer to it from bh. So we have to be careful
1024 * and recheck jh still belongs to our bh before we return success.
1027 if (!buffer_jbd(bh))
1029 /* This should be bh2jh() but that doesn't work with inline functions */
1030 jh = READ_ONCE(bh->b_private);
1033 /* For undo access buffer must have data copied */
1034 if (undo && !jh->b_committed_data)
1036 if (jh->b_transaction != handle->h_transaction &&
1037 jh->b_next_transaction != handle->h_transaction)
1040 * There are two reasons for the barrier here:
1041 * 1) Make sure to fetch b_bh after we did previous checks so that we
1042 * detect when jh went through free, realloc, attach to transaction
1043 * while we were checking. Paired with implicit barrier in that path.
1044 * 2) So that access to bh done after jbd2_write_access_granted()
1045 * doesn't get reordered and see inconsistent state of concurrent
1046 * do_get_write_access().
1049 if (unlikely(jh->b_bh != bh))
1058 * int jbd2_journal_get_write_access() - notify intent to modify a buffer for metadata (not data) update.
1059 * @handle: transaction to add buffer modifications to
1060 * @bh: bh to be used for metadata writes
1062 * Returns an error code or 0 on success.
1064 * In full data journalling mode the buffer may be of type BJ_AsyncData,
1065 * because we're write()ing a buffer which is also part of a shared mapping.
1068 int jbd2_journal_get_write_access(handle_t *handle, struct buffer_head *bh)
1070 struct journal_head *jh;
1073 if (jbd2_write_access_granted(handle, bh, false))
1076 jh = jbd2_journal_add_journal_head(bh);
1077 /* We do not want to get caught playing with fields which the
1078 * log thread also manipulates. Make sure that the buffer
1079 * completes any outstanding IO before proceeding. */
1080 rc = do_get_write_access(handle, jh, 0);
1081 jbd2_journal_put_journal_head(jh);
1087 * When the user wants to journal a newly created buffer_head
1088 * (ie. getblk() returned a new buffer and we are going to populate it
1089 * manually rather than reading off disk), then we need to keep the
1090 * buffer_head locked until it has been completely filled with new
1091 * data. In this case, we should be able to make the assertion that
1092 * the bh is not already part of an existing transaction.
1094 * The buffer should already be locked by the caller by this point.
1095 * There is no lock ranking violation: it was a newly created,
1096 * unlocked buffer beforehand. */
1099 * int jbd2_journal_get_create_access () - notify intent to use newly created bh
1100 * @handle: transaction to new buffer to
1103 * Call this if you create a new bh.
1105 int jbd2_journal_get_create_access(handle_t *handle, struct buffer_head *bh)
1107 transaction_t *transaction = handle->h_transaction;
1109 struct journal_head *jh = jbd2_journal_add_journal_head(bh);
1112 jbd_debug(5, "journal_head %p\n", jh);
1114 if (is_handle_aborted(handle))
1116 journal = transaction->t_journal;
1119 JBUFFER_TRACE(jh, "entry");
1121 * The buffer may already belong to this transaction due to pre-zeroing
1122 * in the filesystem's new_block code. It may also be on the previous,
1123 * committing transaction's lists, but it HAS to be in Forget state in
1124 * that case: the transaction must have deleted the buffer for it to be
1127 jbd_lock_bh_state(bh);
1128 J_ASSERT_JH(jh, (jh->b_transaction == transaction ||
1129 jh->b_transaction == NULL ||
1130 (jh->b_transaction == journal->j_committing_transaction &&
1131 jh->b_jlist == BJ_Forget)));
1133 J_ASSERT_JH(jh, jh->b_next_transaction == NULL);
1134 J_ASSERT_JH(jh, buffer_locked(jh2bh(jh)));
1136 if (jh->b_transaction == NULL) {
1138 * Previous jbd2_journal_forget() could have left the buffer
1139 * with jbddirty bit set because it was being committed. When
1140 * the commit finished, we've filed the buffer for
1141 * checkpointing and marked it dirty. Now we are reallocating
1142 * the buffer so the transaction freeing it must have
1143 * committed and so it's safe to clear the dirty bit.
1145 clear_buffer_dirty(jh2bh(jh));
1146 /* first access by this transaction */
1149 JBUFFER_TRACE(jh, "file as BJ_Reserved");
1150 spin_lock(&journal->j_list_lock);
1151 __jbd2_journal_file_buffer(jh, transaction, BJ_Reserved);
1152 } else if (jh->b_transaction == journal->j_committing_transaction) {
1153 /* first access by this transaction */
1156 JBUFFER_TRACE(jh, "set next transaction");
1157 spin_lock(&journal->j_list_lock);
1158 jh->b_next_transaction = transaction;
1160 spin_unlock(&journal->j_list_lock);
1161 jbd_unlock_bh_state(bh);
1164 * akpm: I added this. ext3_alloc_branch can pick up new indirect
1165 * blocks which contain freed but then revoked metadata. We need
1166 * to cancel the revoke in case we end up freeing it yet again
1167 * and the reallocating as data - this would cause a second revoke,
1168 * which hits an assertion error.
1170 JBUFFER_TRACE(jh, "cancelling revoke");
1171 jbd2_journal_cancel_revoke(handle, jh);
1173 jbd2_journal_put_journal_head(jh);
1178 * int jbd2_journal_get_undo_access() - Notify intent to modify metadata with
1179 * non-rewindable consequences
1180 * @handle: transaction
1181 * @bh: buffer to undo
1183 * Sometimes there is a need to distinguish between metadata which has
1184 * been committed to disk and that which has not. The ext3fs code uses
1185 * this for freeing and allocating space, we have to make sure that we
1186 * do not reuse freed space until the deallocation has been committed,
1187 * since if we overwrote that space we would make the delete
1188 * un-rewindable in case of a crash.
1190 * To deal with that, jbd2_journal_get_undo_access requests write access to a
1191 * buffer for parts of non-rewindable operations such as delete
1192 * operations on the bitmaps. The journaling code must keep a copy of
1193 * the buffer's contents prior to the undo_access call until such time
1194 * as we know that the buffer has definitely been committed to disk.
1196 * We never need to know which transaction the committed data is part
1197 * of, buffers touched here are guaranteed to be dirtied later and so
1198 * will be committed to a new transaction in due course, at which point
1199 * we can discard the old committed data pointer.
1201 * Returns error number or 0 on success.
1203 int jbd2_journal_get_undo_access(handle_t *handle, struct buffer_head *bh)
1206 struct journal_head *jh;
1207 char *committed_data = NULL;
1209 JBUFFER_TRACE(jh, "entry");
1210 if (jbd2_write_access_granted(handle, bh, true))
1213 jh = jbd2_journal_add_journal_head(bh);
1215 * Do this first --- it can drop the journal lock, so we want to
1216 * make sure that obtaining the committed_data is done
1217 * atomically wrt. completion of any outstanding commits.
1219 err = do_get_write_access(handle, jh, 1);
1224 if (!jh->b_committed_data)
1225 committed_data = jbd2_alloc(jh2bh(jh)->b_size,
1226 GFP_NOFS|__GFP_NOFAIL);
1228 jbd_lock_bh_state(bh);
1229 if (!jh->b_committed_data) {
1230 /* Copy out the current buffer contents into the
1231 * preserved, committed copy. */
1232 JBUFFER_TRACE(jh, "generate b_committed data");
1233 if (!committed_data) {
1234 jbd_unlock_bh_state(bh);
1238 jh->b_committed_data = committed_data;
1239 committed_data = NULL;
1240 memcpy(jh->b_committed_data, bh->b_data, bh->b_size);
1242 jbd_unlock_bh_state(bh);
1244 jbd2_journal_put_journal_head(jh);
1245 if (unlikely(committed_data))
1246 jbd2_free(committed_data, bh->b_size);
1251 * void jbd2_journal_set_triggers() - Add triggers for commit writeout
1252 * @bh: buffer to trigger on
1253 * @type: struct jbd2_buffer_trigger_type containing the trigger(s).
1255 * Set any triggers on this journal_head. This is always safe, because
1256 * triggers for a committing buffer will be saved off, and triggers for
1257 * a running transaction will match the buffer in that transaction.
1259 * Call with NULL to clear the triggers.
1261 void jbd2_journal_set_triggers(struct buffer_head *bh,
1262 struct jbd2_buffer_trigger_type *type)
1264 struct journal_head *jh = jbd2_journal_grab_journal_head(bh);
1268 jh->b_triggers = type;
1269 jbd2_journal_put_journal_head(jh);
1272 void jbd2_buffer_frozen_trigger(struct journal_head *jh, void *mapped_data,
1273 struct jbd2_buffer_trigger_type *triggers)
1275 struct buffer_head *bh = jh2bh(jh);
1277 if (!triggers || !triggers->t_frozen)
1280 triggers->t_frozen(triggers, bh, mapped_data, bh->b_size);
1283 void jbd2_buffer_abort_trigger(struct journal_head *jh,
1284 struct jbd2_buffer_trigger_type *triggers)
1286 if (!triggers || !triggers->t_abort)
1289 triggers->t_abort(triggers, jh2bh(jh));
1293 * int jbd2_journal_dirty_metadata() - mark a buffer as containing dirty metadata
1294 * @handle: transaction to add buffer to.
1295 * @bh: buffer to mark
1297 * mark dirty metadata which needs to be journaled as part of the current
1300 * The buffer must have previously had jbd2_journal_get_write_access()
1301 * called so that it has a valid journal_head attached to the buffer
1304 * The buffer is placed on the transaction's metadata list and is marked
1305 * as belonging to the transaction.
1307 * Returns error number or 0 on success.
1309 * Special care needs to be taken if the buffer already belongs to the
1310 * current committing transaction (in which case we should have frozen
1311 * data present for that commit). In that case, we don't relink the
1312 * buffer: that only gets done when the old transaction finally
1313 * completes its commit.
1315 int jbd2_journal_dirty_metadata(handle_t *handle, struct buffer_head *bh)
1317 transaction_t *transaction = handle->h_transaction;
1319 struct journal_head *jh;
1322 if (is_handle_aborted(handle))
1324 if (!buffer_jbd(bh)) {
1329 * We don't grab jh reference here since the buffer must be part
1330 * of the running transaction.
1334 * This and the following assertions are unreliable since we may see jh
1335 * in inconsistent state unless we grab bh_state lock. But this is
1336 * crucial to catch bugs so let's do a reliable check until the
1337 * lockless handling is fully proven.
1339 if (jh->b_transaction != transaction &&
1340 jh->b_next_transaction != transaction) {
1341 jbd_lock_bh_state(bh);
1342 J_ASSERT_JH(jh, jh->b_transaction == transaction ||
1343 jh->b_next_transaction == transaction);
1344 jbd_unlock_bh_state(bh);
1346 if (jh->b_modified == 1) {
1347 /* If it's in our transaction it must be in BJ_Metadata list. */
1348 if (jh->b_transaction == transaction &&
1349 jh->b_jlist != BJ_Metadata) {
1350 jbd_lock_bh_state(bh);
1351 J_ASSERT_JH(jh, jh->b_transaction != transaction ||
1352 jh->b_jlist == BJ_Metadata);
1353 jbd_unlock_bh_state(bh);
1358 journal = transaction->t_journal;
1359 jbd_debug(5, "journal_head %p\n", jh);
1360 JBUFFER_TRACE(jh, "entry");
1362 jbd_lock_bh_state(bh);
1364 if (jh->b_modified == 0) {
1366 * This buffer's got modified and becoming part
1367 * of the transaction. This needs to be done
1368 * once a transaction -bzzz
1371 if (handle->h_buffer_credits <= 0) {
1375 handle->h_buffer_credits--;
1379 * fastpath, to avoid expensive locking. If this buffer is already
1380 * on the running transaction's metadata list there is nothing to do.
1381 * Nobody can take it off again because there is a handle open.
1382 * I _think_ we're OK here with SMP barriers - a mistaken decision will
1383 * result in this test being false, so we go in and take the locks.
1385 if (jh->b_transaction == transaction && jh->b_jlist == BJ_Metadata) {
1386 JBUFFER_TRACE(jh, "fastpath");
1387 if (unlikely(jh->b_transaction !=
1388 journal->j_running_transaction)) {
1389 printk(KERN_ERR "JBD2: %s: "
1390 "jh->b_transaction (%llu, %p, %u) != "
1391 "journal->j_running_transaction (%p, %u)\n",
1393 (unsigned long long) bh->b_blocknr,
1395 jh->b_transaction ? jh->b_transaction->t_tid : 0,
1396 journal->j_running_transaction,
1397 journal->j_running_transaction ?
1398 journal->j_running_transaction->t_tid : 0);
1404 set_buffer_jbddirty(bh);
1407 * Metadata already on the current transaction list doesn't
1408 * need to be filed. Metadata on another transaction's list must
1409 * be committing, and will be refiled once the commit completes:
1410 * leave it alone for now.
1412 if (jh->b_transaction != transaction) {
1413 JBUFFER_TRACE(jh, "already on other transaction");
1414 if (unlikely(((jh->b_transaction !=
1415 journal->j_committing_transaction)) ||
1416 (jh->b_next_transaction != transaction))) {
1417 printk(KERN_ERR "jbd2_journal_dirty_metadata: %s: "
1418 "bad jh for block %llu: "
1419 "transaction (%p, %u), "
1420 "jh->b_transaction (%p, %u), "
1421 "jh->b_next_transaction (%p, %u), jlist %u\n",
1423 (unsigned long long) bh->b_blocknr,
1424 transaction, transaction->t_tid,
1427 jh->b_transaction->t_tid : 0,
1428 jh->b_next_transaction,
1429 jh->b_next_transaction ?
1430 jh->b_next_transaction->t_tid : 0,
1435 /* And this case is illegal: we can't reuse another
1436 * transaction's data buffer, ever. */
1440 /* That test should have eliminated the following case: */
1441 J_ASSERT_JH(jh, jh->b_frozen_data == NULL);
1443 JBUFFER_TRACE(jh, "file as BJ_Metadata");
1444 spin_lock(&journal->j_list_lock);
1445 __jbd2_journal_file_buffer(jh, transaction, BJ_Metadata);
1446 spin_unlock(&journal->j_list_lock);
1448 jbd_unlock_bh_state(bh);
1450 JBUFFER_TRACE(jh, "exit");
1455 * void jbd2_journal_forget() - bforget() for potentially-journaled buffers.
1456 * @handle: transaction handle
1457 * @bh: bh to 'forget'
1459 * We can only do the bforget if there are no commits pending against the
1460 * buffer. If the buffer is dirty in the current running transaction we
1461 * can safely unlink it.
1463 * bh may not be a journalled buffer at all - it may be a non-JBD
1464 * buffer which came off the hashtable. Check for this.
1466 * Decrements bh->b_count by one.
1468 * Allow this call even if the handle has aborted --- it may be part of
1469 * the caller's cleanup after an abort.
1471 int jbd2_journal_forget (handle_t *handle, struct buffer_head *bh)
1473 transaction_t *transaction = handle->h_transaction;
1475 struct journal_head *jh;
1476 int drop_reserve = 0;
1478 int was_modified = 0;
1480 if (is_handle_aborted(handle))
1482 journal = transaction->t_journal;
1484 BUFFER_TRACE(bh, "entry");
1486 jbd_lock_bh_state(bh);
1488 if (!buffer_jbd(bh))
1492 /* Critical error: attempting to delete a bitmap buffer, maybe?
1493 * Don't do any jbd operations, and return an error. */
1494 if (!J_EXPECT_JH(jh, !jh->b_committed_data,
1495 "inconsistent data on disk")) {
1500 /* keep track of whether or not this transaction modified us */
1501 was_modified = jh->b_modified;
1504 * The buffer's going from the transaction, we must drop
1505 * all references -bzzz
1509 if (jh->b_transaction == transaction) {
1510 J_ASSERT_JH(jh, !jh->b_frozen_data);
1512 /* If we are forgetting a buffer which is already part
1513 * of this transaction, then we can just drop it from
1514 * the transaction immediately. */
1515 clear_buffer_dirty(bh);
1516 clear_buffer_jbddirty(bh);
1518 JBUFFER_TRACE(jh, "belongs to current transaction: unfile");
1521 * we only want to drop a reference if this transaction
1522 * modified the buffer
1528 * We are no longer going to journal this buffer.
1529 * However, the commit of this transaction is still
1530 * important to the buffer: the delete that we are now
1531 * processing might obsolete an old log entry, so by
1532 * committing, we can satisfy the buffer's checkpoint.
1534 * So, if we have a checkpoint on the buffer, we should
1535 * now refile the buffer on our BJ_Forget list so that
1536 * we know to remove the checkpoint after we commit.
1539 spin_lock(&journal->j_list_lock);
1540 if (jh->b_cp_transaction) {
1541 __jbd2_journal_temp_unlink_buffer(jh);
1542 __jbd2_journal_file_buffer(jh, transaction, BJ_Forget);
1544 __jbd2_journal_unfile_buffer(jh);
1545 if (!buffer_jbd(bh)) {
1546 spin_unlock(&journal->j_list_lock);
1547 jbd_unlock_bh_state(bh);
1552 spin_unlock(&journal->j_list_lock);
1553 } else if (jh->b_transaction) {
1554 J_ASSERT_JH(jh, (jh->b_transaction ==
1555 journal->j_committing_transaction));
1556 /* However, if the buffer is still owned by a prior
1557 * (committing) transaction, we can't drop it yet... */
1558 JBUFFER_TRACE(jh, "belongs to older transaction");
1559 /* ... but we CAN drop it from the new transaction if we
1560 * have also modified it since the original commit. */
1562 if (jh->b_next_transaction) {
1563 J_ASSERT(jh->b_next_transaction == transaction);
1564 spin_lock(&journal->j_list_lock);
1565 jh->b_next_transaction = NULL;
1566 spin_unlock(&journal->j_list_lock);
1569 * only drop a reference if this transaction modified
1578 jbd_unlock_bh_state(bh);
1582 /* no need to reserve log space for this block -bzzz */
1583 handle->h_buffer_credits++;
1589 * int jbd2_journal_stop() - complete a transaction
1590 * @handle: transaction to complete.
1592 * All done for a particular handle.
1594 * There is not much action needed here. We just return any remaining
1595 * buffer credits to the transaction and remove the handle. The only
1596 * complication is that we need to start a commit operation if the
1597 * filesystem is marked for synchronous update.
1599 * jbd2_journal_stop itself will not usually return an error, but it may
1600 * do so in unusual circumstances. In particular, expect it to
1601 * return -EIO if a jbd2_journal_abort has been executed since the
1602 * transaction began.
1604 int jbd2_journal_stop(handle_t *handle)
1606 transaction_t *transaction = handle->h_transaction;
1608 int err = 0, wait_for_commit = 0;
1614 * Handle is already detached from the transaction so
1615 * there is nothing to do other than decrease a refcount,
1616 * or free the handle if refcount drops to zero
1618 if (--handle->h_ref > 0) {
1619 jbd_debug(4, "h_ref %d -> %d\n", handle->h_ref + 1,
1623 if (handle->h_rsv_handle)
1624 jbd2_free_handle(handle->h_rsv_handle);
1628 journal = transaction->t_journal;
1630 J_ASSERT(journal_current_handle() == handle);
1632 if (is_handle_aborted(handle))
1635 J_ASSERT(atomic_read(&transaction->t_updates) > 0);
1637 if (--handle->h_ref > 0) {
1638 jbd_debug(4, "h_ref %d -> %d\n", handle->h_ref + 1,
1643 jbd_debug(4, "Handle %p going down\n", handle);
1644 trace_jbd2_handle_stats(journal->j_fs_dev->bd_dev,
1646 handle->h_type, handle->h_line_no,
1647 jiffies - handle->h_start_jiffies,
1648 handle->h_sync, handle->h_requested_credits,
1649 (handle->h_requested_credits -
1650 handle->h_buffer_credits));
1653 * Implement synchronous transaction batching. If the handle
1654 * was synchronous, don't force a commit immediately. Let's
1655 * yield and let another thread piggyback onto this
1656 * transaction. Keep doing that while new threads continue to
1657 * arrive. It doesn't cost much - we're about to run a commit
1658 * and sleep on IO anyway. Speeds up many-threaded, many-dir
1659 * operations by 30x or more...
1661 * We try and optimize the sleep time against what the
1662 * underlying disk can do, instead of having a static sleep
1663 * time. This is useful for the case where our storage is so
1664 * fast that it is more optimal to go ahead and force a flush
1665 * and wait for the transaction to be committed than it is to
1666 * wait for an arbitrary amount of time for new writers to
1667 * join the transaction. We achieve this by measuring how
1668 * long it takes to commit a transaction, and compare it with
1669 * how long this transaction has been running, and if run time
1670 * < commit time then we sleep for the delta and commit. This
1671 * greatly helps super fast disks that would see slowdowns as
1672 * more threads started doing fsyncs.
1674 * But don't do this if this process was the most recent one
1675 * to perform a synchronous write. We do this to detect the
1676 * case where a single process is doing a stream of sync
1677 * writes. No point in waiting for joiners in that case.
1679 * Setting max_batch_time to 0 disables this completely.
1682 if (handle->h_sync && journal->j_last_sync_writer != pid &&
1683 journal->j_max_batch_time) {
1684 u64 commit_time, trans_time;
1686 journal->j_last_sync_writer = pid;
1688 read_lock(&journal->j_state_lock);
1689 commit_time = journal->j_average_commit_time;
1690 read_unlock(&journal->j_state_lock);
1692 trans_time = ktime_to_ns(ktime_sub(ktime_get(),
1693 transaction->t_start_time));
1695 commit_time = max_t(u64, commit_time,
1696 1000*journal->j_min_batch_time);
1697 commit_time = min_t(u64, commit_time,
1698 1000*journal->j_max_batch_time);
1700 if (trans_time < commit_time) {
1701 ktime_t expires = ktime_add_ns(ktime_get(),
1703 set_current_state(TASK_UNINTERRUPTIBLE);
1704 schedule_hrtimeout(&expires, HRTIMER_MODE_ABS);
1709 transaction->t_synchronous_commit = 1;
1710 current->journal_info = NULL;
1711 atomic_sub(handle->h_buffer_credits,
1712 &transaction->t_outstanding_credits);
1715 * If the handle is marked SYNC, we need to set another commit
1716 * going! We also want to force a commit if the current
1717 * transaction is occupying too much of the log, or if the
1718 * transaction is too old now.
1720 if (handle->h_sync ||
1721 (atomic_read(&transaction->t_outstanding_credits) >
1722 journal->j_max_transaction_buffers) ||
1723 time_after_eq(jiffies, transaction->t_expires)) {
1724 /* Do this even for aborted journals: an abort still
1725 * completes the commit thread, it just doesn't write
1726 * anything to disk. */
1728 jbd_debug(2, "transaction too old, requesting commit for "
1729 "handle %p\n", handle);
1730 /* This is non-blocking */
1731 jbd2_log_start_commit(journal, transaction->t_tid);
1734 * Special case: JBD2_SYNC synchronous updates require us
1735 * to wait for the commit to complete.
1737 if (handle->h_sync && !(current->flags & PF_MEMALLOC))
1738 wait_for_commit = 1;
1742 * Once we drop t_updates, if it goes to zero the transaction
1743 * could start committing on us and eventually disappear. So
1744 * once we do this, we must not dereference transaction
1747 tid = transaction->t_tid;
1748 if (atomic_dec_and_test(&transaction->t_updates)) {
1749 wake_up(&journal->j_wait_updates);
1750 if (journal->j_barrier_count)
1751 wake_up(&journal->j_wait_transaction_locked);
1754 rwsem_release(&journal->j_trans_commit_map, 1, _THIS_IP_);
1756 if (wait_for_commit)
1757 err = jbd2_log_wait_commit(journal, tid);
1759 if (handle->h_rsv_handle)
1760 jbd2_journal_free_reserved(handle->h_rsv_handle);
1762 jbd2_free_handle(handle);
1768 * List management code snippets: various functions for manipulating the
1769 * transaction buffer lists.
1774 * Append a buffer to a transaction list, given the transaction's list head
1777 * j_list_lock is held.
1779 * jbd_lock_bh_state(jh2bh(jh)) is held.
1783 __blist_add_buffer(struct journal_head **list, struct journal_head *jh)
1786 jh->b_tnext = jh->b_tprev = jh;
1789 /* Insert at the tail of the list to preserve order */
1790 struct journal_head *first = *list, *last = first->b_tprev;
1792 jh->b_tnext = first;
1793 last->b_tnext = first->b_tprev = jh;
1798 * Remove a buffer from a transaction list, given the transaction's list
1801 * Called with j_list_lock held, and the journal may not be locked.
1803 * jbd_lock_bh_state(jh2bh(jh)) is held.
1807 __blist_del_buffer(struct journal_head **list, struct journal_head *jh)
1810 *list = jh->b_tnext;
1814 jh->b_tprev->b_tnext = jh->b_tnext;
1815 jh->b_tnext->b_tprev = jh->b_tprev;
1819 * Remove a buffer from the appropriate transaction list.
1821 * Note that this function can *change* the value of
1822 * bh->b_transaction->t_buffers, t_forget, t_shadow_list, t_log_list or
1823 * t_reserved_list. If the caller is holding onto a copy of one of these
1824 * pointers, it could go bad. Generally the caller needs to re-read the
1825 * pointer from the transaction_t.
1827 * Called under j_list_lock.
1829 static void __jbd2_journal_temp_unlink_buffer(struct journal_head *jh)
1831 struct journal_head **list = NULL;
1832 transaction_t *transaction;
1833 struct buffer_head *bh = jh2bh(jh);
1835 J_ASSERT_JH(jh, jbd_is_locked_bh_state(bh));
1836 transaction = jh->b_transaction;
1838 assert_spin_locked(&transaction->t_journal->j_list_lock);
1840 J_ASSERT_JH(jh, jh->b_jlist < BJ_Types);
1841 if (jh->b_jlist != BJ_None)
1842 J_ASSERT_JH(jh, transaction != NULL);
1844 switch (jh->b_jlist) {
1848 transaction->t_nr_buffers--;
1849 J_ASSERT_JH(jh, transaction->t_nr_buffers >= 0);
1850 list = &transaction->t_buffers;
1853 list = &transaction->t_forget;
1856 list = &transaction->t_shadow_list;
1859 list = &transaction->t_reserved_list;
1863 __blist_del_buffer(list, jh);
1864 jh->b_jlist = BJ_None;
1865 if (test_clear_buffer_jbddirty(bh))
1866 mark_buffer_dirty(bh); /* Expose it to the VM */
1870 * Remove buffer from all transactions.
1872 * Called with bh_state lock and j_list_lock
1874 * jh and bh may be already freed when this function returns.
1876 static void __jbd2_journal_unfile_buffer(struct journal_head *jh)
1878 __jbd2_journal_temp_unlink_buffer(jh);
1879 jh->b_transaction = NULL;
1880 jbd2_journal_put_journal_head(jh);
1883 void jbd2_journal_unfile_buffer(journal_t *journal, struct journal_head *jh)
1885 struct buffer_head *bh = jh2bh(jh);
1887 /* Get reference so that buffer cannot be freed before we unlock it */
1889 jbd_lock_bh_state(bh);
1890 spin_lock(&journal->j_list_lock);
1891 __jbd2_journal_unfile_buffer(jh);
1892 spin_unlock(&journal->j_list_lock);
1893 jbd_unlock_bh_state(bh);
1898 * Called from jbd2_journal_try_to_free_buffers().
1900 * Called under jbd_lock_bh_state(bh)
1903 __journal_try_to_free_buffer(journal_t *journal, struct buffer_head *bh)
1905 struct journal_head *jh;
1909 if (buffer_locked(bh) || buffer_dirty(bh))
1912 if (jh->b_next_transaction != NULL || jh->b_transaction != NULL)
1915 spin_lock(&journal->j_list_lock);
1916 if (jh->b_cp_transaction != NULL) {
1917 /* written-back checkpointed metadata buffer */
1918 JBUFFER_TRACE(jh, "remove from checkpoint list");
1919 __jbd2_journal_remove_checkpoint(jh);
1921 spin_unlock(&journal->j_list_lock);
1927 * int jbd2_journal_try_to_free_buffers() - try to free page buffers.
1928 * @journal: journal for operation
1929 * @page: to try and free
1930 * @gfp_mask: we use the mask to detect how hard should we try to release
1931 * buffers. If __GFP_DIRECT_RECLAIM and __GFP_FS is set, we wait for commit
1932 * code to release the buffers.
1935 * For all the buffers on this page,
1936 * if they are fully written out ordered data, move them onto BUF_CLEAN
1937 * so try_to_free_buffers() can reap them.
1939 * This function returns non-zero if we wish try_to_free_buffers()
1940 * to be called. We do this if the page is releasable by try_to_free_buffers().
1941 * We also do it if the page has locked or dirty buffers and the caller wants
1942 * us to perform sync or async writeout.
1944 * This complicates JBD locking somewhat. We aren't protected by the
1945 * BKL here. We wish to remove the buffer from its committing or
1946 * running transaction's ->t_datalist via __jbd2_journal_unfile_buffer.
1948 * This may *change* the value of transaction_t->t_datalist, so anyone
1949 * who looks at t_datalist needs to lock against this function.
1951 * Even worse, someone may be doing a jbd2_journal_dirty_data on this
1952 * buffer. So we need to lock against that. jbd2_journal_dirty_data()
1953 * will come out of the lock with the buffer dirty, which makes it
1954 * ineligible for release here.
1956 * Who else is affected by this? hmm... Really the only contender
1957 * is do_get_write_access() - it could be looking at the buffer while
1958 * journal_try_to_free_buffer() is changing its state. But that
1959 * cannot happen because we never reallocate freed data as metadata
1960 * while the data is part of a transaction. Yes?
1962 * Return 0 on failure, 1 on success
1964 int jbd2_journal_try_to_free_buffers(journal_t *journal,
1965 struct page *page, gfp_t gfp_mask)
1967 struct buffer_head *head;
1968 struct buffer_head *bh;
1971 J_ASSERT(PageLocked(page));
1973 head = page_buffers(page);
1976 struct journal_head *jh;
1979 * We take our own ref against the journal_head here to avoid
1980 * having to add tons of locking around each instance of
1981 * jbd2_journal_put_journal_head().
1983 jh = jbd2_journal_grab_journal_head(bh);
1987 jbd_lock_bh_state(bh);
1988 __journal_try_to_free_buffer(journal, bh);
1989 jbd2_journal_put_journal_head(jh);
1990 jbd_unlock_bh_state(bh);
1993 } while ((bh = bh->b_this_page) != head);
1995 ret = try_to_free_buffers(page);
2002 * This buffer is no longer needed. If it is on an older transaction's
2003 * checkpoint list we need to record it on this transaction's forget list
2004 * to pin this buffer (and hence its checkpointing transaction) down until
2005 * this transaction commits. If the buffer isn't on a checkpoint list, we
2007 * Returns non-zero if JBD no longer has an interest in the buffer.
2009 * Called under j_list_lock.
2011 * Called under jbd_lock_bh_state(bh).
2013 static int __dispose_buffer(struct journal_head *jh, transaction_t *transaction)
2016 struct buffer_head *bh = jh2bh(jh);
2018 if (jh->b_cp_transaction) {
2019 JBUFFER_TRACE(jh, "on running+cp transaction");
2020 __jbd2_journal_temp_unlink_buffer(jh);
2022 * We don't want to write the buffer anymore, clear the
2023 * bit so that we don't confuse checks in
2024 * __journal_file_buffer
2026 clear_buffer_dirty(bh);
2027 __jbd2_journal_file_buffer(jh, transaction, BJ_Forget);
2030 JBUFFER_TRACE(jh, "on running transaction");
2031 __jbd2_journal_unfile_buffer(jh);
2037 * jbd2_journal_invalidatepage
2039 * This code is tricky. It has a number of cases to deal with.
2041 * There are two invariants which this code relies on:
2043 * i_size must be updated on disk before we start calling invalidatepage on the
2046 * This is done in ext3 by defining an ext3_setattr method which
2047 * updates i_size before truncate gets going. By maintaining this
2048 * invariant, we can be sure that it is safe to throw away any buffers
2049 * attached to the current transaction: once the transaction commits,
2050 * we know that the data will not be needed.
2052 * Note however that we can *not* throw away data belonging to the
2053 * previous, committing transaction!
2055 * Any disk blocks which *are* part of the previous, committing
2056 * transaction (and which therefore cannot be discarded immediately) are
2057 * not going to be reused in the new running transaction
2059 * The bitmap committed_data images guarantee this: any block which is
2060 * allocated in one transaction and removed in the next will be marked
2061 * as in-use in the committed_data bitmap, so cannot be reused until
2062 * the next transaction to delete the block commits. This means that
2063 * leaving committing buffers dirty is quite safe: the disk blocks
2064 * cannot be reallocated to a different file and so buffer aliasing is
2068 * The above applies mainly to ordered data mode. In writeback mode we
2069 * don't make guarantees about the order in which data hits disk --- in
2070 * particular we don't guarantee that new dirty data is flushed before
2071 * transaction commit --- so it is always safe just to discard data
2072 * immediately in that mode. --sct
2076 * The journal_unmap_buffer helper function returns zero if the buffer
2077 * concerned remains pinned as an anonymous buffer belonging to an older
2080 * We're outside-transaction here. Either or both of j_running_transaction
2081 * and j_committing_transaction may be NULL.
2083 static int journal_unmap_buffer(journal_t *journal, struct buffer_head *bh,
2086 transaction_t *transaction;
2087 struct journal_head *jh;
2090 BUFFER_TRACE(bh, "entry");
2093 * It is safe to proceed here without the j_list_lock because the
2094 * buffers cannot be stolen by try_to_free_buffers as long as we are
2095 * holding the page lock. --sct
2098 if (!buffer_jbd(bh))
2099 goto zap_buffer_unlocked;
2101 /* OK, we have data buffer in journaled mode */
2102 write_lock(&journal->j_state_lock);
2103 jbd_lock_bh_state(bh);
2104 spin_lock(&journal->j_list_lock);
2106 jh = jbd2_journal_grab_journal_head(bh);
2108 goto zap_buffer_no_jh;
2111 * We cannot remove the buffer from checkpoint lists until the
2112 * transaction adding inode to orphan list (let's call it T)
2113 * is committed. Otherwise if the transaction changing the
2114 * buffer would be cleaned from the journal before T is
2115 * committed, a crash will cause that the correct contents of
2116 * the buffer will be lost. On the other hand we have to
2117 * clear the buffer dirty bit at latest at the moment when the
2118 * transaction marking the buffer as freed in the filesystem
2119 * structures is committed because from that moment on the
2120 * block can be reallocated and used by a different page.
2121 * Since the block hasn't been freed yet but the inode has
2122 * already been added to orphan list, it is safe for us to add
2123 * the buffer to BJ_Forget list of the newest transaction.
2125 * Also we have to clear buffer_mapped flag of a truncated buffer
2126 * because the buffer_head may be attached to the page straddling
2127 * i_size (can happen only when blocksize < pagesize) and thus the
2128 * buffer_head can be reused when the file is extended again. So we end
2129 * up keeping around invalidated buffers attached to transactions'
2130 * BJ_Forget list just to stop checkpointing code from cleaning up
2131 * the transaction this buffer was modified in.
2133 transaction = jh->b_transaction;
2134 if (transaction == NULL) {
2135 /* First case: not on any transaction. If it
2136 * has no checkpoint link, then we can zap it:
2137 * it's a writeback-mode buffer so we don't care
2138 * if it hits disk safely. */
2139 if (!jh->b_cp_transaction) {
2140 JBUFFER_TRACE(jh, "not on any transaction: zap");
2144 if (!buffer_dirty(bh)) {
2145 /* bdflush has written it. We can drop it now */
2146 __jbd2_journal_remove_checkpoint(jh);
2150 /* OK, it must be in the journal but still not
2151 * written fully to disk: it's metadata or
2152 * journaled data... */
2154 if (journal->j_running_transaction) {
2155 /* ... and once the current transaction has
2156 * committed, the buffer won't be needed any
2158 JBUFFER_TRACE(jh, "checkpointed: add to BJ_Forget");
2159 may_free = __dispose_buffer(jh,
2160 journal->j_running_transaction);
2163 /* There is no currently-running transaction. So the
2164 * orphan record which we wrote for this file must have
2165 * passed into commit. We must attach this buffer to
2166 * the committing transaction, if it exists. */
2167 if (journal->j_committing_transaction) {
2168 JBUFFER_TRACE(jh, "give to committing trans");
2169 may_free = __dispose_buffer(jh,
2170 journal->j_committing_transaction);
2173 /* The orphan record's transaction has
2174 * committed. We can cleanse this buffer */
2175 clear_buffer_jbddirty(bh);
2176 __jbd2_journal_remove_checkpoint(jh);
2180 } else if (transaction == journal->j_committing_transaction) {
2181 JBUFFER_TRACE(jh, "on committing transaction");
2183 * The buffer is committing, we simply cannot touch
2184 * it. If the page is straddling i_size we have to wait
2185 * for commit and try again.
2188 jbd2_journal_put_journal_head(jh);
2189 spin_unlock(&journal->j_list_lock);
2190 jbd_unlock_bh_state(bh);
2191 write_unlock(&journal->j_state_lock);
2195 * OK, buffer won't be reachable after truncate. We just set
2196 * j_next_transaction to the running transaction (if there is
2197 * one) and mark buffer as freed so that commit code knows it
2198 * should clear dirty bits when it is done with the buffer.
2200 set_buffer_freed(bh);
2201 if (journal->j_running_transaction && buffer_jbddirty(bh))
2202 jh->b_next_transaction = journal->j_running_transaction;
2203 jbd2_journal_put_journal_head(jh);
2204 spin_unlock(&journal->j_list_lock);
2205 jbd_unlock_bh_state(bh);
2206 write_unlock(&journal->j_state_lock);
2209 /* Good, the buffer belongs to the running transaction.
2210 * We are writing our own transaction's data, not any
2211 * previous one's, so it is safe to throw it away
2212 * (remember that we expect the filesystem to have set
2213 * i_size already for this truncate so recovery will not
2214 * expose the disk blocks we are discarding here.) */
2215 J_ASSERT_JH(jh, transaction == journal->j_running_transaction);
2216 JBUFFER_TRACE(jh, "on running transaction");
2217 may_free = __dispose_buffer(jh, transaction);
2222 * This is tricky. Although the buffer is truncated, it may be reused
2223 * if blocksize < pagesize and it is attached to the page straddling
2224 * EOF. Since the buffer might have been added to BJ_Forget list of the
2225 * running transaction, journal_get_write_access() won't clear
2226 * b_modified and credit accounting gets confused. So clear b_modified
2230 jbd2_journal_put_journal_head(jh);
2232 spin_unlock(&journal->j_list_lock);
2233 jbd_unlock_bh_state(bh);
2234 write_unlock(&journal->j_state_lock);
2235 zap_buffer_unlocked:
2236 clear_buffer_dirty(bh);
2237 J_ASSERT_BH(bh, !buffer_jbddirty(bh));
2238 clear_buffer_mapped(bh);
2239 clear_buffer_req(bh);
2240 clear_buffer_new(bh);
2241 clear_buffer_delay(bh);
2242 clear_buffer_unwritten(bh);
2248 * void jbd2_journal_invalidatepage()
2249 * @journal: journal to use for flush...
2250 * @page: page to flush
2251 * @offset: start of the range to invalidate
2252 * @length: length of the range to invalidate
2254 * Reap page buffers containing data after in the specified range in page.
2255 * Can return -EBUSY if buffers are part of the committing transaction and
2256 * the page is straddling i_size. Caller then has to wait for current commit
2259 int jbd2_journal_invalidatepage(journal_t *journal,
2261 unsigned int offset,
2262 unsigned int length)
2264 struct buffer_head *head, *bh, *next;
2265 unsigned int stop = offset + length;
2266 unsigned int curr_off = 0;
2267 int partial_page = (offset || length < PAGE_SIZE);
2271 if (!PageLocked(page))
2273 if (!page_has_buffers(page))
2276 BUG_ON(stop > PAGE_SIZE || stop < length);
2278 /* We will potentially be playing with lists other than just the
2279 * data lists (especially for journaled data mode), so be
2280 * cautious in our locking. */
2282 head = bh = page_buffers(page);
2284 unsigned int next_off = curr_off + bh->b_size;
2285 next = bh->b_this_page;
2287 if (next_off > stop)
2290 if (offset <= curr_off) {
2291 /* This block is wholly outside the truncation point */
2293 ret = journal_unmap_buffer(journal, bh, partial_page);
2299 curr_off = next_off;
2302 } while (bh != head);
2304 if (!partial_page) {
2305 if (may_free && try_to_free_buffers(page))
2306 J_ASSERT(!page_has_buffers(page));
2312 * File a buffer on the given transaction list.
2314 void __jbd2_journal_file_buffer(struct journal_head *jh,
2315 transaction_t *transaction, int jlist)
2317 struct journal_head **list = NULL;
2319 struct buffer_head *bh = jh2bh(jh);
2321 J_ASSERT_JH(jh, jbd_is_locked_bh_state(bh));
2322 assert_spin_locked(&transaction->t_journal->j_list_lock);
2324 J_ASSERT_JH(jh, jh->b_jlist < BJ_Types);
2325 J_ASSERT_JH(jh, jh->b_transaction == transaction ||
2326 jh->b_transaction == NULL);
2328 if (jh->b_transaction && jh->b_jlist == jlist)
2331 if (jlist == BJ_Metadata || jlist == BJ_Reserved ||
2332 jlist == BJ_Shadow || jlist == BJ_Forget) {
2334 * For metadata buffers, we track dirty bit in buffer_jbddirty
2335 * instead of buffer_dirty. We should not see a dirty bit set
2336 * here because we clear it in do_get_write_access but e.g.
2337 * tune2fs can modify the sb and set the dirty bit at any time
2338 * so we try to gracefully handle that.
2340 if (buffer_dirty(bh))
2341 warn_dirty_buffer(bh);
2342 if (test_clear_buffer_dirty(bh) ||
2343 test_clear_buffer_jbddirty(bh))
2347 if (jh->b_transaction)
2348 __jbd2_journal_temp_unlink_buffer(jh);
2350 jbd2_journal_grab_journal_head(bh);
2351 jh->b_transaction = transaction;
2355 J_ASSERT_JH(jh, !jh->b_committed_data);
2356 J_ASSERT_JH(jh, !jh->b_frozen_data);
2359 transaction->t_nr_buffers++;
2360 list = &transaction->t_buffers;
2363 list = &transaction->t_forget;
2366 list = &transaction->t_shadow_list;
2369 list = &transaction->t_reserved_list;
2373 __blist_add_buffer(list, jh);
2374 jh->b_jlist = jlist;
2377 set_buffer_jbddirty(bh);
2380 void jbd2_journal_file_buffer(struct journal_head *jh,
2381 transaction_t *transaction, int jlist)
2383 jbd_lock_bh_state(jh2bh(jh));
2384 spin_lock(&transaction->t_journal->j_list_lock);
2385 __jbd2_journal_file_buffer(jh, transaction, jlist);
2386 spin_unlock(&transaction->t_journal->j_list_lock);
2387 jbd_unlock_bh_state(jh2bh(jh));
2391 * Remove a buffer from its current buffer list in preparation for
2392 * dropping it from its current transaction entirely. If the buffer has
2393 * already started to be used by a subsequent transaction, refile the
2394 * buffer on that transaction's metadata list.
2396 * Called under j_list_lock
2397 * Called under jbd_lock_bh_state(jh2bh(jh))
2399 * jh and bh may be already free when this function returns
2401 void __jbd2_journal_refile_buffer(struct journal_head *jh)
2403 int was_dirty, jlist;
2404 struct buffer_head *bh = jh2bh(jh);
2406 J_ASSERT_JH(jh, jbd_is_locked_bh_state(bh));
2407 if (jh->b_transaction)
2408 assert_spin_locked(&jh->b_transaction->t_journal->j_list_lock);
2410 /* If the buffer is now unused, just drop it. */
2411 if (jh->b_next_transaction == NULL) {
2412 __jbd2_journal_unfile_buffer(jh);
2417 * It has been modified by a later transaction: add it to the new
2418 * transaction's metadata list.
2421 was_dirty = test_clear_buffer_jbddirty(bh);
2422 __jbd2_journal_temp_unlink_buffer(jh);
2424 * We set b_transaction here because b_next_transaction will inherit
2425 * our jh reference and thus __jbd2_journal_file_buffer() must not
2428 jh->b_transaction = jh->b_next_transaction;
2429 jh->b_next_transaction = NULL;
2430 if (buffer_freed(bh))
2432 else if (jh->b_modified)
2433 jlist = BJ_Metadata;
2435 jlist = BJ_Reserved;
2436 __jbd2_journal_file_buffer(jh, jh->b_transaction, jlist);
2437 J_ASSERT_JH(jh, jh->b_transaction->t_state == T_RUNNING);
2440 set_buffer_jbddirty(bh);
2444 * __jbd2_journal_refile_buffer() with necessary locking added. We take our
2445 * bh reference so that we can safely unlock bh.
2447 * The jh and bh may be freed by this call.
2449 void jbd2_journal_refile_buffer(journal_t *journal, struct journal_head *jh)
2451 struct buffer_head *bh = jh2bh(jh);
2453 /* Get reference so that buffer cannot be freed before we unlock it */
2455 jbd_lock_bh_state(bh);
2456 spin_lock(&journal->j_list_lock);
2457 __jbd2_journal_refile_buffer(jh);
2458 jbd_unlock_bh_state(bh);
2459 spin_unlock(&journal->j_list_lock);
2464 * File inode in the inode list of the handle's transaction
2466 static int jbd2_journal_file_inode(handle_t *handle, struct jbd2_inode *jinode,
2467 unsigned long flags)
2469 transaction_t *transaction = handle->h_transaction;
2472 if (is_handle_aborted(handle))
2474 journal = transaction->t_journal;
2476 jbd_debug(4, "Adding inode %lu, tid:%d\n", jinode->i_vfs_inode->i_ino,
2477 transaction->t_tid);
2480 * First check whether inode isn't already on the transaction's
2481 * lists without taking the lock. Note that this check is safe
2482 * without the lock as we cannot race with somebody removing inode
2483 * from the transaction. The reason is that we remove inode from the
2484 * transaction only in journal_release_jbd_inode() and when we commit
2485 * the transaction. We are guarded from the first case by holding
2486 * a reference to the inode. We are safe against the second case
2487 * because if jinode->i_transaction == transaction, commit code
2488 * cannot touch the transaction because we hold reference to it,
2489 * and if jinode->i_next_transaction == transaction, commit code
2490 * will only file the inode where we want it.
2492 if ((jinode->i_transaction == transaction ||
2493 jinode->i_next_transaction == transaction) &&
2494 (jinode->i_flags & flags) == flags)
2497 spin_lock(&journal->j_list_lock);
2498 jinode->i_flags |= flags;
2499 /* Is inode already attached where we need it? */
2500 if (jinode->i_transaction == transaction ||
2501 jinode->i_next_transaction == transaction)
2505 * We only ever set this variable to 1 so the test is safe. Since
2506 * t_need_data_flush is likely to be set, we do the test to save some
2507 * cacheline bouncing
2509 if (!transaction->t_need_data_flush)
2510 transaction->t_need_data_flush = 1;
2511 /* On some different transaction's list - should be
2512 * the committing one */
2513 if (jinode->i_transaction) {
2514 J_ASSERT(jinode->i_next_transaction == NULL);
2515 J_ASSERT(jinode->i_transaction ==
2516 journal->j_committing_transaction);
2517 jinode->i_next_transaction = transaction;
2520 /* Not on any transaction list... */
2521 J_ASSERT(!jinode->i_next_transaction);
2522 jinode->i_transaction = transaction;
2523 list_add(&jinode->i_list, &transaction->t_inode_list);
2525 spin_unlock(&journal->j_list_lock);
2530 int jbd2_journal_inode_add_write(handle_t *handle, struct jbd2_inode *jinode)
2532 return jbd2_journal_file_inode(handle, jinode,
2533 JI_WRITE_DATA | JI_WAIT_DATA);
2536 int jbd2_journal_inode_add_wait(handle_t *handle, struct jbd2_inode *jinode)
2538 return jbd2_journal_file_inode(handle, jinode, JI_WAIT_DATA);
2542 * File truncate and transaction commit interact with each other in a
2543 * non-trivial way. If a transaction writing data block A is
2544 * committing, we cannot discard the data by truncate until we have
2545 * written them. Otherwise if we crashed after the transaction with
2546 * write has committed but before the transaction with truncate has
2547 * committed, we could see stale data in block A. This function is a
2548 * helper to solve this problem. It starts writeout of the truncated
2549 * part in case it is in the committing transaction.
2551 * Filesystem code must call this function when inode is journaled in
2552 * ordered mode before truncation happens and after the inode has been
2553 * placed on orphan list with the new inode size. The second condition
2554 * avoids the race that someone writes new data and we start
2555 * committing the transaction after this function has been called but
2556 * before a transaction for truncate is started (and furthermore it
2557 * allows us to optimize the case where the addition to orphan list
2558 * happens in the same transaction as write --- we don't have to write
2559 * any data in such case).
2561 int jbd2_journal_begin_ordered_truncate(journal_t *journal,
2562 struct jbd2_inode *jinode,
2565 transaction_t *inode_trans, *commit_trans;
2568 /* This is a quick check to avoid locking if not necessary */
2569 if (!jinode->i_transaction)
2571 /* Locks are here just to force reading of recent values, it is
2572 * enough that the transaction was not committing before we started
2573 * a transaction adding the inode to orphan list */
2574 read_lock(&journal->j_state_lock);
2575 commit_trans = journal->j_committing_transaction;
2576 read_unlock(&journal->j_state_lock);
2577 spin_lock(&journal->j_list_lock);
2578 inode_trans = jinode->i_transaction;
2579 spin_unlock(&journal->j_list_lock);
2580 if (inode_trans == commit_trans) {
2581 ret = filemap_fdatawrite_range(jinode->i_vfs_inode->i_mapping,
2582 new_size, LLONG_MAX);
2584 jbd2_journal_abort(journal, ret);