gro: Allow tunnel stacking in the case of FOU/GUE
[linux/fpc-iii.git] / fs / jbd2 / transaction.c
blob2abbb2babcaedf9b05f3f743f10b438e4dd4b342
1 /*
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
13 * journaling system.
15 * This file manages transactions (compound commits managed by the
16 * journaling code) and handles (individual atomic operations by the
17 * filesystem).
20 #include <linux/time.h>
21 #include <linux/fs.h>
22 #include <linux/jbd2.h>
23 #include <linux/errno.h>
24 #include <linux/slab.h>
25 #include <linux/timer.h>
26 #include <linux/mm.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,
46 NULL);
47 if (transaction_cache)
48 return 0;
49 return -ENOMEM;
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)))
63 return;
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).
75 * Preconditions:
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;
108 return transaction;
112 * Handle management.
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,
130 unsigned long ts)
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);
141 #endif
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
147 * j_state_lock.
149 static void wait_transaction_locked(journal_t *journal)
150 __releases(journal->j_state_lock)
152 DEFINE_WAIT(wait);
153 int need_to_start;
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);
160 if (need_to_start)
161 jbd2_log_start_commit(journal, tid);
162 schedule();
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
176 * caller must retry.
178 static int add_transaction_credits(journal_t *journal, int blocks,
179 int rsv_blocks)
181 transaction_t *t = journal->j_running_transaction;
182 int needed;
183 int total = blocks + rsv_blocks;
186 * If the current transaction is locked down for commit, wait
187 * for the lock to be released.
189 if (t->t_state == T_LOCKED) {
190 wait_transaction_locked(journal);
191 return 1;
195 * If there is not enough space left in the log to write all
196 * potential buffers requested by this operation, we need to
197 * stall pending a log checkpoint to free some more log space.
199 needed = atomic_add_return(total, &t->t_outstanding_credits);
200 if (needed > journal->j_max_transaction_buffers) {
202 * If the current transaction is already too large,
203 * then start to commit it: we can then go back and
204 * attach this handle to a new transaction.
206 atomic_sub(total, &t->t_outstanding_credits);
207 wait_transaction_locked(journal);
208 return 1;
212 * The commit code assumes that it can get enough log space
213 * without forcing a checkpoint. This is *critical* for
214 * correctness: a checkpoint of a buffer which is also
215 * associated with a committing transaction creates a deadlock,
216 * so commit simply cannot force through checkpoints.
218 * We must therefore ensure the necessary space in the journal
219 * *before* starting to dirty potentially checkpointed buffers
220 * in the new transaction.
222 if (jbd2_log_space_left(journal) < jbd2_space_needed(journal)) {
223 atomic_sub(total, &t->t_outstanding_credits);
224 read_unlock(&journal->j_state_lock);
225 write_lock(&journal->j_state_lock);
226 if (jbd2_log_space_left(journal) < jbd2_space_needed(journal))
227 __jbd2_log_wait_for_space(journal);
228 write_unlock(&journal->j_state_lock);
229 return 1;
232 /* No reservation? We are done... */
233 if (!rsv_blocks)
234 return 0;
236 needed = atomic_add_return(rsv_blocks, &journal->j_reserved_credits);
237 /* We allow at most half of a transaction to be reserved */
238 if (needed > journal->j_max_transaction_buffers / 2) {
239 sub_reserved_credits(journal, rsv_blocks);
240 atomic_sub(total, &t->t_outstanding_credits);
241 read_unlock(&journal->j_state_lock);
242 wait_event(journal->j_wait_reserved,
243 atomic_read(&journal->j_reserved_credits) + rsv_blocks
244 <= journal->j_max_transaction_buffers / 2);
245 return 1;
247 return 0;
251 * start_this_handle: Given a handle, deal with any locking or stalling
252 * needed to make sure that there is enough journal space for the handle
253 * to begin. Attach the handle to a transaction and set up the
254 * transaction's buffer credits.
257 static int start_this_handle(journal_t *journal, handle_t *handle,
258 gfp_t gfp_mask)
260 transaction_t *transaction, *new_transaction = NULL;
261 int blocks = handle->h_buffer_credits;
262 int rsv_blocks = 0;
263 unsigned long ts = jiffies;
266 * 1/2 of transaction can be reserved so we can practically handle
267 * only 1/2 of maximum transaction size per operation
269 if (WARN_ON(blocks > journal->j_max_transaction_buffers / 2)) {
270 printk(KERN_ERR "JBD2: %s wants too many credits (%d > %d)\n",
271 current->comm, blocks,
272 journal->j_max_transaction_buffers / 2);
273 return -ENOSPC;
276 if (handle->h_rsv_handle)
277 rsv_blocks = handle->h_rsv_handle->h_buffer_credits;
279 alloc_transaction:
280 if (!journal->j_running_transaction) {
281 new_transaction = kmem_cache_zalloc(transaction_cache,
282 gfp_mask);
283 if (!new_transaction) {
285 * If __GFP_FS is not present, then we may be
286 * being called from inside the fs writeback
287 * layer, so we MUST NOT fail. Since
288 * __GFP_NOFAIL is going away, we will arrange
289 * to retry the allocation ourselves.
291 if ((gfp_mask & __GFP_FS) == 0) {
292 congestion_wait(BLK_RW_ASYNC, HZ/50);
293 goto alloc_transaction;
295 return -ENOMEM;
299 jbd_debug(3, "New handle %p going live.\n", handle);
302 * We need to hold j_state_lock until t_updates has been incremented,
303 * for proper journal barrier handling
305 repeat:
306 read_lock(&journal->j_state_lock);
307 BUG_ON(journal->j_flags & JBD2_UNMOUNT);
308 if (is_journal_aborted(journal) ||
309 (journal->j_errno != 0 && !(journal->j_flags & JBD2_ACK_ERR))) {
310 read_unlock(&journal->j_state_lock);
311 jbd2_journal_free_transaction(new_transaction);
312 return -EROFS;
316 * Wait on the journal's transaction barrier if necessary. Specifically
317 * we allow reserved handles to proceed because otherwise commit could
318 * deadlock on page writeback not being able to complete.
320 if (!handle->h_reserved && journal->j_barrier_count) {
321 read_unlock(&journal->j_state_lock);
322 wait_event(journal->j_wait_transaction_locked,
323 journal->j_barrier_count == 0);
324 goto repeat;
327 if (!journal->j_running_transaction) {
328 read_unlock(&journal->j_state_lock);
329 if (!new_transaction)
330 goto alloc_transaction;
331 write_lock(&journal->j_state_lock);
332 if (!journal->j_running_transaction &&
333 (handle->h_reserved || !journal->j_barrier_count)) {
334 jbd2_get_transaction(journal, new_transaction);
335 new_transaction = NULL;
337 write_unlock(&journal->j_state_lock);
338 goto repeat;
341 transaction = journal->j_running_transaction;
343 if (!handle->h_reserved) {
344 /* We may have dropped j_state_lock - restart in that case */
345 if (add_transaction_credits(journal, blocks, rsv_blocks))
346 goto repeat;
347 } else {
349 * We have handle reserved so we are allowed to join T_LOCKED
350 * transaction and we don't have to check for transaction size
351 * and journal space.
353 sub_reserved_credits(journal, blocks);
354 handle->h_reserved = 0;
357 /* OK, account for the buffers that this operation expects to
358 * use and add the handle to the running transaction.
360 update_t_max_wait(transaction, ts);
361 handle->h_transaction = transaction;
362 handle->h_requested_credits = blocks;
363 handle->h_start_jiffies = jiffies;
364 atomic_inc(&transaction->t_updates);
365 atomic_inc(&transaction->t_handle_count);
366 jbd_debug(4, "Handle %p given %d credits (total %d, free %lu)\n",
367 handle, blocks,
368 atomic_read(&transaction->t_outstanding_credits),
369 jbd2_log_space_left(journal));
370 read_unlock(&journal->j_state_lock);
371 current->journal_info = handle;
373 lock_map_acquire(&handle->h_lockdep_map);
374 jbd2_journal_free_transaction(new_transaction);
375 return 0;
378 static struct lock_class_key jbd2_handle_key;
380 /* Allocate a new handle. This should probably be in a slab... */
381 static handle_t *new_handle(int nblocks)
383 handle_t *handle = jbd2_alloc_handle(GFP_NOFS);
384 if (!handle)
385 return NULL;
386 handle->h_buffer_credits = nblocks;
387 handle->h_ref = 1;
389 lockdep_init_map(&handle->h_lockdep_map, "jbd2_handle",
390 &jbd2_handle_key, 0);
392 return handle;
396 * handle_t *jbd2_journal_start() - Obtain a new handle.
397 * @journal: Journal to start transaction on.
398 * @nblocks: number of block buffer we might modify
400 * We make sure that the transaction can guarantee at least nblocks of
401 * modified buffers in the log. We block until the log can guarantee
402 * that much space. Additionally, if rsv_blocks > 0, we also create another
403 * handle with rsv_blocks reserved blocks in the journal. This handle is
404 * is stored in h_rsv_handle. It is not attached to any particular transaction
405 * and thus doesn't block transaction commit. If the caller uses this reserved
406 * handle, it has to set h_rsv_handle to NULL as otherwise jbd2_journal_stop()
407 * on the parent handle will dispose the reserved one. Reserved handle has to
408 * be converted to a normal handle using jbd2_journal_start_reserved() before
409 * it can be used.
411 * Return a pointer to a newly allocated handle, or an ERR_PTR() value
412 * on failure.
414 handle_t *jbd2__journal_start(journal_t *journal, int nblocks, int rsv_blocks,
415 gfp_t gfp_mask, unsigned int type,
416 unsigned int line_no)
418 handle_t *handle = journal_current_handle();
419 int err;
421 if (!journal)
422 return ERR_PTR(-EROFS);
424 if (handle) {
425 J_ASSERT(handle->h_transaction->t_journal == journal);
426 handle->h_ref++;
427 return handle;
430 handle = new_handle(nblocks);
431 if (!handle)
432 return ERR_PTR(-ENOMEM);
433 if (rsv_blocks) {
434 handle_t *rsv_handle;
436 rsv_handle = new_handle(rsv_blocks);
437 if (!rsv_handle) {
438 jbd2_free_handle(handle);
439 return ERR_PTR(-ENOMEM);
441 rsv_handle->h_reserved = 1;
442 rsv_handle->h_journal = journal;
443 handle->h_rsv_handle = rsv_handle;
446 err = start_this_handle(journal, handle, gfp_mask);
447 if (err < 0) {
448 if (handle->h_rsv_handle)
449 jbd2_free_handle(handle->h_rsv_handle);
450 jbd2_free_handle(handle);
451 return ERR_PTR(err);
453 handle->h_type = type;
454 handle->h_line_no = line_no;
455 trace_jbd2_handle_start(journal->j_fs_dev->bd_dev,
456 handle->h_transaction->t_tid, type,
457 line_no, nblocks);
458 return handle;
460 EXPORT_SYMBOL(jbd2__journal_start);
463 handle_t *jbd2_journal_start(journal_t *journal, int nblocks)
465 return jbd2__journal_start(journal, nblocks, 0, GFP_NOFS, 0, 0);
467 EXPORT_SYMBOL(jbd2_journal_start);
469 void jbd2_journal_free_reserved(handle_t *handle)
471 journal_t *journal = handle->h_journal;
473 WARN_ON(!handle->h_reserved);
474 sub_reserved_credits(journal, handle->h_buffer_credits);
475 jbd2_free_handle(handle);
477 EXPORT_SYMBOL(jbd2_journal_free_reserved);
480 * int jbd2_journal_start_reserved(handle_t *handle) - start reserved handle
481 * @handle: handle to start
483 * Start handle that has been previously reserved with jbd2_journal_reserve().
484 * This attaches @handle to the running transaction (or creates one if there's
485 * not transaction running). Unlike jbd2_journal_start() this function cannot
486 * block on journal commit, checkpointing, or similar stuff. It can block on
487 * memory allocation or frozen journal though.
489 * Return 0 on success, non-zero on error - handle is freed in that case.
491 int jbd2_journal_start_reserved(handle_t *handle, unsigned int type,
492 unsigned int line_no)
494 journal_t *journal = handle->h_journal;
495 int ret = -EIO;
497 if (WARN_ON(!handle->h_reserved)) {
498 /* Someone passed in normal handle? Just stop it. */
499 jbd2_journal_stop(handle);
500 return ret;
503 * Usefulness of mixing of reserved and unreserved handles is
504 * questionable. So far nobody seems to need it so just error out.
506 if (WARN_ON(current->journal_info)) {
507 jbd2_journal_free_reserved(handle);
508 return ret;
511 handle->h_journal = NULL;
513 * GFP_NOFS is here because callers are likely from writeback or
514 * similarly constrained call sites
516 ret = start_this_handle(journal, handle, GFP_NOFS);
517 if (ret < 0) {
518 jbd2_journal_free_reserved(handle);
519 return ret;
521 handle->h_type = type;
522 handle->h_line_no = line_no;
523 return 0;
525 EXPORT_SYMBOL(jbd2_journal_start_reserved);
528 * int jbd2_journal_extend() - extend buffer credits.
529 * @handle: handle to 'extend'
530 * @nblocks: nr blocks to try to extend by.
532 * Some transactions, such as large extends and truncates, can be done
533 * atomically all at once or in several stages. The operation requests
534 * a credit for a number of buffer modications in advance, but can
535 * extend its credit if it needs more.
537 * jbd2_journal_extend tries to give the running handle more buffer credits.
538 * It does not guarantee that allocation - this is a best-effort only.
539 * The calling process MUST be able to deal cleanly with a failure to
540 * extend here.
542 * Return 0 on success, non-zero on failure.
544 * return code < 0 implies an error
545 * return code > 0 implies normal transaction-full status.
547 int jbd2_journal_extend(handle_t *handle, int nblocks)
549 transaction_t *transaction = handle->h_transaction;
550 journal_t *journal;
551 int result;
552 int wanted;
554 if (is_handle_aborted(handle))
555 return -EROFS;
556 journal = transaction->t_journal;
558 result = 1;
560 read_lock(&journal->j_state_lock);
562 /* Don't extend a locked-down transaction! */
563 if (transaction->t_state != T_RUNNING) {
564 jbd_debug(3, "denied handle %p %d blocks: "
565 "transaction not running\n", handle, nblocks);
566 goto error_out;
569 spin_lock(&transaction->t_handle_lock);
570 wanted = atomic_add_return(nblocks,
571 &transaction->t_outstanding_credits);
573 if (wanted > journal->j_max_transaction_buffers) {
574 jbd_debug(3, "denied handle %p %d blocks: "
575 "transaction too large\n", handle, nblocks);
576 atomic_sub(nblocks, &transaction->t_outstanding_credits);
577 goto unlock;
580 if (wanted + (wanted >> JBD2_CONTROL_BLOCKS_SHIFT) >
581 jbd2_log_space_left(journal)) {
582 jbd_debug(3, "denied handle %p %d blocks: "
583 "insufficient log space\n", handle, nblocks);
584 atomic_sub(nblocks, &transaction->t_outstanding_credits);
585 goto unlock;
588 trace_jbd2_handle_extend(journal->j_fs_dev->bd_dev,
589 transaction->t_tid,
590 handle->h_type, handle->h_line_no,
591 handle->h_buffer_credits,
592 nblocks);
594 handle->h_buffer_credits += nblocks;
595 handle->h_requested_credits += nblocks;
596 result = 0;
598 jbd_debug(3, "extended handle %p by %d\n", handle, nblocks);
599 unlock:
600 spin_unlock(&transaction->t_handle_lock);
601 error_out:
602 read_unlock(&journal->j_state_lock);
603 return result;
608 * int jbd2_journal_restart() - restart a handle .
609 * @handle: handle to restart
610 * @nblocks: nr credits requested
612 * Restart a handle for a multi-transaction filesystem
613 * operation.
615 * If the jbd2_journal_extend() call above fails to grant new buffer credits
616 * to a running handle, a call to jbd2_journal_restart will commit the
617 * handle's transaction so far and reattach the handle to a new
618 * transaction capabable of guaranteeing the requested number of
619 * credits. We preserve reserved handle if there's any attached to the
620 * passed in handle.
622 int jbd2__journal_restart(handle_t *handle, int nblocks, gfp_t gfp_mask)
624 transaction_t *transaction = handle->h_transaction;
625 journal_t *journal;
626 tid_t tid;
627 int need_to_start, ret;
629 /* If we've had an abort of any type, don't even think about
630 * actually doing the restart! */
631 if (is_handle_aborted(handle))
632 return 0;
633 journal = transaction->t_journal;
636 * First unlink the handle from its current transaction, and start the
637 * commit on that.
639 J_ASSERT(atomic_read(&transaction->t_updates) > 0);
640 J_ASSERT(journal_current_handle() == handle);
642 read_lock(&journal->j_state_lock);
643 spin_lock(&transaction->t_handle_lock);
644 atomic_sub(handle->h_buffer_credits,
645 &transaction->t_outstanding_credits);
646 if (handle->h_rsv_handle) {
647 sub_reserved_credits(journal,
648 handle->h_rsv_handle->h_buffer_credits);
650 if (atomic_dec_and_test(&transaction->t_updates))
651 wake_up(&journal->j_wait_updates);
652 tid = transaction->t_tid;
653 spin_unlock(&transaction->t_handle_lock);
654 handle->h_transaction = NULL;
655 current->journal_info = NULL;
657 jbd_debug(2, "restarting handle %p\n", handle);
658 need_to_start = !tid_geq(journal->j_commit_request, tid);
659 read_unlock(&journal->j_state_lock);
660 if (need_to_start)
661 jbd2_log_start_commit(journal, tid);
663 lock_map_release(&handle->h_lockdep_map);
664 handle->h_buffer_credits = nblocks;
665 ret = start_this_handle(journal, handle, gfp_mask);
666 return ret;
668 EXPORT_SYMBOL(jbd2__journal_restart);
671 int jbd2_journal_restart(handle_t *handle, int nblocks)
673 return jbd2__journal_restart(handle, nblocks, GFP_NOFS);
675 EXPORT_SYMBOL(jbd2_journal_restart);
678 * void jbd2_journal_lock_updates () - establish a transaction barrier.
679 * @journal: Journal to establish a barrier on.
681 * This locks out any further updates from being started, and blocks
682 * until all existing updates have completed, returning only once the
683 * journal is in a quiescent state with no updates running.
685 * The journal lock should not be held on entry.
687 void jbd2_journal_lock_updates(journal_t *journal)
689 DEFINE_WAIT(wait);
691 write_lock(&journal->j_state_lock);
692 ++journal->j_barrier_count;
694 /* Wait until there are no reserved handles */
695 if (atomic_read(&journal->j_reserved_credits)) {
696 write_unlock(&journal->j_state_lock);
697 wait_event(journal->j_wait_reserved,
698 atomic_read(&journal->j_reserved_credits) == 0);
699 write_lock(&journal->j_state_lock);
702 /* Wait until there are no running updates */
703 while (1) {
704 transaction_t *transaction = journal->j_running_transaction;
706 if (!transaction)
707 break;
709 spin_lock(&transaction->t_handle_lock);
710 prepare_to_wait(&journal->j_wait_updates, &wait,
711 TASK_UNINTERRUPTIBLE);
712 if (!atomic_read(&transaction->t_updates)) {
713 spin_unlock(&transaction->t_handle_lock);
714 finish_wait(&journal->j_wait_updates, &wait);
715 break;
717 spin_unlock(&transaction->t_handle_lock);
718 write_unlock(&journal->j_state_lock);
719 schedule();
720 finish_wait(&journal->j_wait_updates, &wait);
721 write_lock(&journal->j_state_lock);
723 write_unlock(&journal->j_state_lock);
726 * We have now established a barrier against other normal updates, but
727 * we also need to barrier against other jbd2_journal_lock_updates() calls
728 * to make sure that we serialise special journal-locked operations
729 * too.
731 mutex_lock(&journal->j_barrier);
735 * void jbd2_journal_unlock_updates (journal_t* journal) - release barrier
736 * @journal: Journal to release the barrier on.
738 * Release a transaction barrier obtained with jbd2_journal_lock_updates().
740 * Should be called without the journal lock held.
742 void jbd2_journal_unlock_updates (journal_t *journal)
744 J_ASSERT(journal->j_barrier_count != 0);
746 mutex_unlock(&journal->j_barrier);
747 write_lock(&journal->j_state_lock);
748 --journal->j_barrier_count;
749 write_unlock(&journal->j_state_lock);
750 wake_up(&journal->j_wait_transaction_locked);
753 static void warn_dirty_buffer(struct buffer_head *bh)
755 char b[BDEVNAME_SIZE];
757 printk(KERN_WARNING
758 "JBD2: Spotted dirty metadata buffer (dev = %s, blocknr = %llu). "
759 "There's a risk of filesystem corruption in case of system "
760 "crash.\n",
761 bdevname(bh->b_bdev, b), (unsigned long long)bh->b_blocknr);
765 * If the buffer is already part of the current transaction, then there
766 * is nothing we need to do. If it is already part of a prior
767 * transaction which we are still committing to disk, then we need to
768 * make sure that we do not overwrite the old copy: we do copy-out to
769 * preserve the copy going to disk. We also account the buffer against
770 * the handle's metadata buffer credits (unless the buffer is already
771 * part of the transaction, that is).
774 static int
775 do_get_write_access(handle_t *handle, struct journal_head *jh,
776 int force_copy)
778 struct buffer_head *bh;
779 transaction_t *transaction = handle->h_transaction;
780 journal_t *journal;
781 int error;
782 char *frozen_buffer = NULL;
783 int need_copy = 0;
784 unsigned long start_lock, time_lock;
786 if (is_handle_aborted(handle))
787 return -EROFS;
788 journal = transaction->t_journal;
790 jbd_debug(5, "journal_head %p, force_copy %d\n", jh, force_copy);
792 JBUFFER_TRACE(jh, "entry");
793 repeat:
794 bh = jh2bh(jh);
796 /* @@@ Need to check for errors here at some point. */
798 start_lock = jiffies;
799 lock_buffer(bh);
800 jbd_lock_bh_state(bh);
802 /* If it takes too long to lock the buffer, trace it */
803 time_lock = jbd2_time_diff(start_lock, jiffies);
804 if (time_lock > HZ/10)
805 trace_jbd2_lock_buffer_stall(bh->b_bdev->bd_dev,
806 jiffies_to_msecs(time_lock));
808 /* We now hold the buffer lock so it is safe to query the buffer
809 * state. Is the buffer dirty?
811 * If so, there are two possibilities. The buffer may be
812 * non-journaled, and undergoing a quite legitimate writeback.
813 * Otherwise, it is journaled, and we don't expect dirty buffers
814 * in that state (the buffers should be marked JBD_Dirty
815 * instead.) So either the IO is being done under our own
816 * control and this is a bug, or it's a third party IO such as
817 * dump(8) (which may leave the buffer scheduled for read ---
818 * ie. locked but not dirty) or tune2fs (which may actually have
819 * the buffer dirtied, ugh.) */
821 if (buffer_dirty(bh)) {
823 * First question: is this buffer already part of the current
824 * transaction or the existing committing transaction?
826 if (jh->b_transaction) {
827 J_ASSERT_JH(jh,
828 jh->b_transaction == transaction ||
829 jh->b_transaction ==
830 journal->j_committing_transaction);
831 if (jh->b_next_transaction)
832 J_ASSERT_JH(jh, jh->b_next_transaction ==
833 transaction);
834 warn_dirty_buffer(bh);
837 * In any case we need to clean the dirty flag and we must
838 * do it under the buffer lock to be sure we don't race
839 * with running write-out.
841 JBUFFER_TRACE(jh, "Journalling dirty buffer");
842 clear_buffer_dirty(bh);
843 set_buffer_jbddirty(bh);
846 unlock_buffer(bh);
848 error = -EROFS;
849 if (is_handle_aborted(handle)) {
850 jbd_unlock_bh_state(bh);
851 goto out;
853 error = 0;
856 * The buffer is already part of this transaction if b_transaction or
857 * b_next_transaction points to it
859 if (jh->b_transaction == transaction ||
860 jh->b_next_transaction == transaction)
861 goto done;
864 * this is the first time this transaction is touching this buffer,
865 * reset the modified flag
867 jh->b_modified = 0;
870 * If there is already a copy-out version of this buffer, then we don't
871 * need to make another one
873 if (jh->b_frozen_data) {
874 JBUFFER_TRACE(jh, "has frozen data");
875 J_ASSERT_JH(jh, jh->b_next_transaction == NULL);
876 jh->b_next_transaction = transaction;
877 goto done;
880 /* Is there data here we need to preserve? */
882 if (jh->b_transaction && jh->b_transaction != transaction) {
883 JBUFFER_TRACE(jh, "owned by older transaction");
884 J_ASSERT_JH(jh, jh->b_next_transaction == NULL);
885 J_ASSERT_JH(jh, jh->b_transaction ==
886 journal->j_committing_transaction);
888 /* There is one case we have to be very careful about.
889 * If the committing transaction is currently writing
890 * this buffer out to disk and has NOT made a copy-out,
891 * then we cannot modify the buffer contents at all
892 * right now. The essence of copy-out is that it is the
893 * extra copy, not the primary copy, which gets
894 * journaled. If the primary copy is already going to
895 * disk then we cannot do copy-out here. */
897 if (buffer_shadow(bh)) {
898 JBUFFER_TRACE(jh, "on shadow: sleep");
899 jbd_unlock_bh_state(bh);
900 wait_on_bit_io(&bh->b_state, BH_Shadow,
901 TASK_UNINTERRUPTIBLE);
902 goto repeat;
906 * Only do the copy if the currently-owning transaction still
907 * needs it. If buffer isn't on BJ_Metadata list, the
908 * committing transaction is past that stage (here we use the
909 * fact that BH_Shadow is set under bh_state lock together with
910 * refiling to BJ_Shadow list and at this point we know the
911 * buffer doesn't have BH_Shadow set).
913 * Subtle point, though: if this is a get_undo_access,
914 * then we will be relying on the frozen_data to contain
915 * the new value of the committed_data record after the
916 * transaction, so we HAVE to force the frozen_data copy
917 * in that case.
919 if (jh->b_jlist == BJ_Metadata || force_copy) {
920 JBUFFER_TRACE(jh, "generate frozen data");
921 if (!frozen_buffer) {
922 JBUFFER_TRACE(jh, "allocate memory for buffer");
923 jbd_unlock_bh_state(bh);
924 frozen_buffer =
925 jbd2_alloc(jh2bh(jh)->b_size,
926 GFP_NOFS);
927 if (!frozen_buffer) {
928 printk(KERN_ERR
929 "%s: OOM for frozen_buffer\n",
930 __func__);
931 JBUFFER_TRACE(jh, "oom!");
932 error = -ENOMEM;
933 jbd_lock_bh_state(bh);
934 goto done;
936 goto repeat;
938 jh->b_frozen_data = frozen_buffer;
939 frozen_buffer = NULL;
940 need_copy = 1;
942 jh->b_next_transaction = transaction;
947 * Finally, if the buffer is not journaled right now, we need to make
948 * sure it doesn't get written to disk before the caller actually
949 * commits the new data
951 if (!jh->b_transaction) {
952 JBUFFER_TRACE(jh, "no transaction");
953 J_ASSERT_JH(jh, !jh->b_next_transaction);
954 JBUFFER_TRACE(jh, "file as BJ_Reserved");
955 spin_lock(&journal->j_list_lock);
956 __jbd2_journal_file_buffer(jh, transaction, BJ_Reserved);
957 spin_unlock(&journal->j_list_lock);
960 done:
961 if (need_copy) {
962 struct page *page;
963 int offset;
964 char *source;
966 J_EXPECT_JH(jh, buffer_uptodate(jh2bh(jh)),
967 "Possible IO failure.\n");
968 page = jh2bh(jh)->b_page;
969 offset = offset_in_page(jh2bh(jh)->b_data);
970 source = kmap_atomic(page);
971 /* Fire data frozen trigger just before we copy the data */
972 jbd2_buffer_frozen_trigger(jh, source + offset,
973 jh->b_triggers);
974 memcpy(jh->b_frozen_data, source+offset, jh2bh(jh)->b_size);
975 kunmap_atomic(source);
978 * Now that the frozen data is saved off, we need to store
979 * any matching triggers.
981 jh->b_frozen_triggers = jh->b_triggers;
983 jbd_unlock_bh_state(bh);
986 * If we are about to journal a buffer, then any revoke pending on it is
987 * no longer valid
989 jbd2_journal_cancel_revoke(handle, jh);
991 out:
992 if (unlikely(frozen_buffer)) /* It's usually NULL */
993 jbd2_free(frozen_buffer, bh->b_size);
995 JBUFFER_TRACE(jh, "exit");
996 return error;
1000 * int jbd2_journal_get_write_access() - notify intent to modify a buffer for metadata (not data) update.
1001 * @handle: transaction to add buffer modifications to
1002 * @bh: bh to be used for metadata writes
1004 * Returns an error code or 0 on success.
1006 * In full data journalling mode the buffer may be of type BJ_AsyncData,
1007 * because we're write()ing a buffer which is also part of a shared mapping.
1010 int jbd2_journal_get_write_access(handle_t *handle, struct buffer_head *bh)
1012 struct journal_head *jh = jbd2_journal_add_journal_head(bh);
1013 int rc;
1015 /* We do not want to get caught playing with fields which the
1016 * log thread also manipulates. Make sure that the buffer
1017 * completes any outstanding IO before proceeding. */
1018 rc = do_get_write_access(handle, jh, 0);
1019 jbd2_journal_put_journal_head(jh);
1020 return rc;
1025 * When the user wants to journal a newly created buffer_head
1026 * (ie. getblk() returned a new buffer and we are going to populate it
1027 * manually rather than reading off disk), then we need to keep the
1028 * buffer_head locked until it has been completely filled with new
1029 * data. In this case, we should be able to make the assertion that
1030 * the bh is not already part of an existing transaction.
1032 * The buffer should already be locked by the caller by this point.
1033 * There is no lock ranking violation: it was a newly created,
1034 * unlocked buffer beforehand. */
1037 * int jbd2_journal_get_create_access () - notify intent to use newly created bh
1038 * @handle: transaction to new buffer to
1039 * @bh: new buffer.
1041 * Call this if you create a new bh.
1043 int jbd2_journal_get_create_access(handle_t *handle, struct buffer_head *bh)
1045 transaction_t *transaction = handle->h_transaction;
1046 journal_t *journal;
1047 struct journal_head *jh = jbd2_journal_add_journal_head(bh);
1048 int err;
1050 jbd_debug(5, "journal_head %p\n", jh);
1051 err = -EROFS;
1052 if (is_handle_aborted(handle))
1053 goto out;
1054 journal = transaction->t_journal;
1055 err = 0;
1057 JBUFFER_TRACE(jh, "entry");
1059 * The buffer may already belong to this transaction due to pre-zeroing
1060 * in the filesystem's new_block code. It may also be on the previous,
1061 * committing transaction's lists, but it HAS to be in Forget state in
1062 * that case: the transaction must have deleted the buffer for it to be
1063 * reused here.
1065 jbd_lock_bh_state(bh);
1066 J_ASSERT_JH(jh, (jh->b_transaction == transaction ||
1067 jh->b_transaction == NULL ||
1068 (jh->b_transaction == journal->j_committing_transaction &&
1069 jh->b_jlist == BJ_Forget)));
1071 J_ASSERT_JH(jh, jh->b_next_transaction == NULL);
1072 J_ASSERT_JH(jh, buffer_locked(jh2bh(jh)));
1074 if (jh->b_transaction == NULL) {
1076 * Previous jbd2_journal_forget() could have left the buffer
1077 * with jbddirty bit set because it was being committed. When
1078 * the commit finished, we've filed the buffer for
1079 * checkpointing and marked it dirty. Now we are reallocating
1080 * the buffer so the transaction freeing it must have
1081 * committed and so it's safe to clear the dirty bit.
1083 clear_buffer_dirty(jh2bh(jh));
1084 /* first access by this transaction */
1085 jh->b_modified = 0;
1087 JBUFFER_TRACE(jh, "file as BJ_Reserved");
1088 spin_lock(&journal->j_list_lock);
1089 __jbd2_journal_file_buffer(jh, transaction, BJ_Reserved);
1090 spin_unlock(&journal->j_list_lock);
1091 } else if (jh->b_transaction == journal->j_committing_transaction) {
1092 /* first access by this transaction */
1093 jh->b_modified = 0;
1095 JBUFFER_TRACE(jh, "set next transaction");
1096 spin_lock(&journal->j_list_lock);
1097 jh->b_next_transaction = transaction;
1098 spin_unlock(&journal->j_list_lock);
1100 jbd_unlock_bh_state(bh);
1103 * akpm: I added this. ext3_alloc_branch can pick up new indirect
1104 * blocks which contain freed but then revoked metadata. We need
1105 * to cancel the revoke in case we end up freeing it yet again
1106 * and the reallocating as data - this would cause a second revoke,
1107 * which hits an assertion error.
1109 JBUFFER_TRACE(jh, "cancelling revoke");
1110 jbd2_journal_cancel_revoke(handle, jh);
1111 out:
1112 jbd2_journal_put_journal_head(jh);
1113 return err;
1117 * int jbd2_journal_get_undo_access() - Notify intent to modify metadata with
1118 * non-rewindable consequences
1119 * @handle: transaction
1120 * @bh: buffer to undo
1122 * Sometimes there is a need to distinguish between metadata which has
1123 * been committed to disk and that which has not. The ext3fs code uses
1124 * this for freeing and allocating space, we have to make sure that we
1125 * do not reuse freed space until the deallocation has been committed,
1126 * since if we overwrote that space we would make the delete
1127 * un-rewindable in case of a crash.
1129 * To deal with that, jbd2_journal_get_undo_access requests write access to a
1130 * buffer for parts of non-rewindable operations such as delete
1131 * operations on the bitmaps. The journaling code must keep a copy of
1132 * the buffer's contents prior to the undo_access call until such time
1133 * as we know that the buffer has definitely been committed to disk.
1135 * We never need to know which transaction the committed data is part
1136 * of, buffers touched here are guaranteed to be dirtied later and so
1137 * will be committed to a new transaction in due course, at which point
1138 * we can discard the old committed data pointer.
1140 * Returns error number or 0 on success.
1142 int jbd2_journal_get_undo_access(handle_t *handle, struct buffer_head *bh)
1144 int err;
1145 struct journal_head *jh = jbd2_journal_add_journal_head(bh);
1146 char *committed_data = NULL;
1148 JBUFFER_TRACE(jh, "entry");
1151 * Do this first --- it can drop the journal lock, so we want to
1152 * make sure that obtaining the committed_data is done
1153 * atomically wrt. completion of any outstanding commits.
1155 err = do_get_write_access(handle, jh, 1);
1156 if (err)
1157 goto out;
1159 repeat:
1160 if (!jh->b_committed_data) {
1161 committed_data = jbd2_alloc(jh2bh(jh)->b_size, GFP_NOFS);
1162 if (!committed_data) {
1163 printk(KERN_ERR "%s: No memory for committed data\n",
1164 __func__);
1165 err = -ENOMEM;
1166 goto out;
1170 jbd_lock_bh_state(bh);
1171 if (!jh->b_committed_data) {
1172 /* Copy out the current buffer contents into the
1173 * preserved, committed copy. */
1174 JBUFFER_TRACE(jh, "generate b_committed data");
1175 if (!committed_data) {
1176 jbd_unlock_bh_state(bh);
1177 goto repeat;
1180 jh->b_committed_data = committed_data;
1181 committed_data = NULL;
1182 memcpy(jh->b_committed_data, bh->b_data, bh->b_size);
1184 jbd_unlock_bh_state(bh);
1185 out:
1186 jbd2_journal_put_journal_head(jh);
1187 if (unlikely(committed_data))
1188 jbd2_free(committed_data, bh->b_size);
1189 return err;
1193 * void jbd2_journal_set_triggers() - Add triggers for commit writeout
1194 * @bh: buffer to trigger on
1195 * @type: struct jbd2_buffer_trigger_type containing the trigger(s).
1197 * Set any triggers on this journal_head. This is always safe, because
1198 * triggers for a committing buffer will be saved off, and triggers for
1199 * a running transaction will match the buffer in that transaction.
1201 * Call with NULL to clear the triggers.
1203 void jbd2_journal_set_triggers(struct buffer_head *bh,
1204 struct jbd2_buffer_trigger_type *type)
1206 struct journal_head *jh = jbd2_journal_grab_journal_head(bh);
1208 if (WARN_ON(!jh))
1209 return;
1210 jh->b_triggers = type;
1211 jbd2_journal_put_journal_head(jh);
1214 void jbd2_buffer_frozen_trigger(struct journal_head *jh, void *mapped_data,
1215 struct jbd2_buffer_trigger_type *triggers)
1217 struct buffer_head *bh = jh2bh(jh);
1219 if (!triggers || !triggers->t_frozen)
1220 return;
1222 triggers->t_frozen(triggers, bh, mapped_data, bh->b_size);
1225 void jbd2_buffer_abort_trigger(struct journal_head *jh,
1226 struct jbd2_buffer_trigger_type *triggers)
1228 if (!triggers || !triggers->t_abort)
1229 return;
1231 triggers->t_abort(triggers, jh2bh(jh));
1237 * int jbd2_journal_dirty_metadata() - mark a buffer as containing dirty metadata
1238 * @handle: transaction to add buffer to.
1239 * @bh: buffer to mark
1241 * mark dirty metadata which needs to be journaled as part of the current
1242 * transaction.
1244 * The buffer must have previously had jbd2_journal_get_write_access()
1245 * called so that it has a valid journal_head attached to the buffer
1246 * head.
1248 * The buffer is placed on the transaction's metadata list and is marked
1249 * as belonging to the transaction.
1251 * Returns error number or 0 on success.
1253 * Special care needs to be taken if the buffer already belongs to the
1254 * current committing transaction (in which case we should have frozen
1255 * data present for that commit). In that case, we don't relink the
1256 * buffer: that only gets done when the old transaction finally
1257 * completes its commit.
1259 int jbd2_journal_dirty_metadata(handle_t *handle, struct buffer_head *bh)
1261 transaction_t *transaction = handle->h_transaction;
1262 journal_t *journal;
1263 struct journal_head *jh;
1264 int ret = 0;
1266 if (is_handle_aborted(handle))
1267 return -EROFS;
1268 journal = transaction->t_journal;
1269 jh = jbd2_journal_grab_journal_head(bh);
1270 if (!jh) {
1271 ret = -EUCLEAN;
1272 goto out;
1274 jbd_debug(5, "journal_head %p\n", jh);
1275 JBUFFER_TRACE(jh, "entry");
1277 jbd_lock_bh_state(bh);
1279 if (jh->b_modified == 0) {
1281 * This buffer's got modified and becoming part
1282 * of the transaction. This needs to be done
1283 * once a transaction -bzzz
1285 jh->b_modified = 1;
1286 if (handle->h_buffer_credits <= 0) {
1287 ret = -ENOSPC;
1288 goto out_unlock_bh;
1290 handle->h_buffer_credits--;
1294 * fastpath, to avoid expensive locking. If this buffer is already
1295 * on the running transaction's metadata list there is nothing to do.
1296 * Nobody can take it off again because there is a handle open.
1297 * I _think_ we're OK here with SMP barriers - a mistaken decision will
1298 * result in this test being false, so we go in and take the locks.
1300 if (jh->b_transaction == transaction && jh->b_jlist == BJ_Metadata) {
1301 JBUFFER_TRACE(jh, "fastpath");
1302 if (unlikely(jh->b_transaction !=
1303 journal->j_running_transaction)) {
1304 printk(KERN_ERR "JBD2: %s: "
1305 "jh->b_transaction (%llu, %p, %u) != "
1306 "journal->j_running_transaction (%p, %u)\n",
1307 journal->j_devname,
1308 (unsigned long long) bh->b_blocknr,
1309 jh->b_transaction,
1310 jh->b_transaction ? jh->b_transaction->t_tid : 0,
1311 journal->j_running_transaction,
1312 journal->j_running_transaction ?
1313 journal->j_running_transaction->t_tid : 0);
1314 ret = -EINVAL;
1316 goto out_unlock_bh;
1319 set_buffer_jbddirty(bh);
1322 * Metadata already on the current transaction list doesn't
1323 * need to be filed. Metadata on another transaction's list must
1324 * be committing, and will be refiled once the commit completes:
1325 * leave it alone for now.
1327 if (jh->b_transaction != transaction) {
1328 JBUFFER_TRACE(jh, "already on other transaction");
1329 if (unlikely(((jh->b_transaction !=
1330 journal->j_committing_transaction)) ||
1331 (jh->b_next_transaction != transaction))) {
1332 printk(KERN_ERR "jbd2_journal_dirty_metadata: %s: "
1333 "bad jh for block %llu: "
1334 "transaction (%p, %u), "
1335 "jh->b_transaction (%p, %u), "
1336 "jh->b_next_transaction (%p, %u), jlist %u\n",
1337 journal->j_devname,
1338 (unsigned long long) bh->b_blocknr,
1339 transaction, transaction->t_tid,
1340 jh->b_transaction,
1341 jh->b_transaction ?
1342 jh->b_transaction->t_tid : 0,
1343 jh->b_next_transaction,
1344 jh->b_next_transaction ?
1345 jh->b_next_transaction->t_tid : 0,
1346 jh->b_jlist);
1347 WARN_ON(1);
1348 ret = -EINVAL;
1350 /* And this case is illegal: we can't reuse another
1351 * transaction's data buffer, ever. */
1352 goto out_unlock_bh;
1355 /* That test should have eliminated the following case: */
1356 J_ASSERT_JH(jh, jh->b_frozen_data == NULL);
1358 JBUFFER_TRACE(jh, "file as BJ_Metadata");
1359 spin_lock(&journal->j_list_lock);
1360 __jbd2_journal_file_buffer(jh, transaction, BJ_Metadata);
1361 spin_unlock(&journal->j_list_lock);
1362 out_unlock_bh:
1363 jbd_unlock_bh_state(bh);
1364 jbd2_journal_put_journal_head(jh);
1365 out:
1366 JBUFFER_TRACE(jh, "exit");
1367 return ret;
1371 * void jbd2_journal_forget() - bforget() for potentially-journaled buffers.
1372 * @handle: transaction handle
1373 * @bh: bh to 'forget'
1375 * We can only do the bforget if there are no commits pending against the
1376 * buffer. If the buffer is dirty in the current running transaction we
1377 * can safely unlink it.
1379 * bh may not be a journalled buffer at all - it may be a non-JBD
1380 * buffer which came off the hashtable. Check for this.
1382 * Decrements bh->b_count by one.
1384 * Allow this call even if the handle has aborted --- it may be part of
1385 * the caller's cleanup after an abort.
1387 int jbd2_journal_forget (handle_t *handle, struct buffer_head *bh)
1389 transaction_t *transaction = handle->h_transaction;
1390 journal_t *journal;
1391 struct journal_head *jh;
1392 int drop_reserve = 0;
1393 int err = 0;
1394 int was_modified = 0;
1396 if (is_handle_aborted(handle))
1397 return -EROFS;
1398 journal = transaction->t_journal;
1400 BUFFER_TRACE(bh, "entry");
1402 jbd_lock_bh_state(bh);
1404 if (!buffer_jbd(bh))
1405 goto not_jbd;
1406 jh = bh2jh(bh);
1408 /* Critical error: attempting to delete a bitmap buffer, maybe?
1409 * Don't do any jbd operations, and return an error. */
1410 if (!J_EXPECT_JH(jh, !jh->b_committed_data,
1411 "inconsistent data on disk")) {
1412 err = -EIO;
1413 goto not_jbd;
1416 /* keep track of whether or not this transaction modified us */
1417 was_modified = jh->b_modified;
1420 * The buffer's going from the transaction, we must drop
1421 * all references -bzzz
1423 jh->b_modified = 0;
1425 if (jh->b_transaction == transaction) {
1426 J_ASSERT_JH(jh, !jh->b_frozen_data);
1428 /* If we are forgetting a buffer which is already part
1429 * of this transaction, then we can just drop it from
1430 * the transaction immediately. */
1431 clear_buffer_dirty(bh);
1432 clear_buffer_jbddirty(bh);
1434 JBUFFER_TRACE(jh, "belongs to current transaction: unfile");
1437 * we only want to drop a reference if this transaction
1438 * modified the buffer
1440 if (was_modified)
1441 drop_reserve = 1;
1444 * We are no longer going to journal this buffer.
1445 * However, the commit of this transaction is still
1446 * important to the buffer: the delete that we are now
1447 * processing might obsolete an old log entry, so by
1448 * committing, we can satisfy the buffer's checkpoint.
1450 * So, if we have a checkpoint on the buffer, we should
1451 * now refile the buffer on our BJ_Forget list so that
1452 * we know to remove the checkpoint after we commit.
1455 spin_lock(&journal->j_list_lock);
1456 if (jh->b_cp_transaction) {
1457 __jbd2_journal_temp_unlink_buffer(jh);
1458 __jbd2_journal_file_buffer(jh, transaction, BJ_Forget);
1459 } else {
1460 __jbd2_journal_unfile_buffer(jh);
1461 if (!buffer_jbd(bh)) {
1462 spin_unlock(&journal->j_list_lock);
1463 jbd_unlock_bh_state(bh);
1464 __bforget(bh);
1465 goto drop;
1468 spin_unlock(&journal->j_list_lock);
1469 } else if (jh->b_transaction) {
1470 J_ASSERT_JH(jh, (jh->b_transaction ==
1471 journal->j_committing_transaction));
1472 /* However, if the buffer is still owned by a prior
1473 * (committing) transaction, we can't drop it yet... */
1474 JBUFFER_TRACE(jh, "belongs to older transaction");
1475 /* ... but we CAN drop it from the new transaction if we
1476 * have also modified it since the original commit. */
1478 if (jh->b_next_transaction) {
1479 J_ASSERT(jh->b_next_transaction == transaction);
1480 spin_lock(&journal->j_list_lock);
1481 jh->b_next_transaction = NULL;
1482 spin_unlock(&journal->j_list_lock);
1485 * only drop a reference if this transaction modified
1486 * the buffer
1488 if (was_modified)
1489 drop_reserve = 1;
1493 not_jbd:
1494 jbd_unlock_bh_state(bh);
1495 __brelse(bh);
1496 drop:
1497 if (drop_reserve) {
1498 /* no need to reserve log space for this block -bzzz */
1499 handle->h_buffer_credits++;
1501 return err;
1505 * int jbd2_journal_stop() - complete a transaction
1506 * @handle: tranaction to complete.
1508 * All done for a particular handle.
1510 * There is not much action needed here. We just return any remaining
1511 * buffer credits to the transaction and remove the handle. The only
1512 * complication is that we need to start a commit operation if the
1513 * filesystem is marked for synchronous update.
1515 * jbd2_journal_stop itself will not usually return an error, but it may
1516 * do so in unusual circumstances. In particular, expect it to
1517 * return -EIO if a jbd2_journal_abort has been executed since the
1518 * transaction began.
1520 int jbd2_journal_stop(handle_t *handle)
1522 transaction_t *transaction = handle->h_transaction;
1523 journal_t *journal;
1524 int err = 0, wait_for_commit = 0;
1525 tid_t tid;
1526 pid_t pid;
1528 if (!transaction) {
1530 * Handle is already detached from the transaction so
1531 * there is nothing to do other than decrease a refcount,
1532 * or free the handle if refcount drops to zero
1534 if (--handle->h_ref > 0) {
1535 jbd_debug(4, "h_ref %d -> %d\n", handle->h_ref + 1,
1536 handle->h_ref);
1537 return err;
1538 } else {
1539 if (handle->h_rsv_handle)
1540 jbd2_free_handle(handle->h_rsv_handle);
1541 goto free_and_exit;
1544 journal = transaction->t_journal;
1546 J_ASSERT(journal_current_handle() == handle);
1548 if (is_handle_aborted(handle))
1549 err = -EIO;
1550 else
1551 J_ASSERT(atomic_read(&transaction->t_updates) > 0);
1553 if (--handle->h_ref > 0) {
1554 jbd_debug(4, "h_ref %d -> %d\n", handle->h_ref + 1,
1555 handle->h_ref);
1556 return err;
1559 jbd_debug(4, "Handle %p going down\n", handle);
1560 trace_jbd2_handle_stats(journal->j_fs_dev->bd_dev,
1561 transaction->t_tid,
1562 handle->h_type, handle->h_line_no,
1563 jiffies - handle->h_start_jiffies,
1564 handle->h_sync, handle->h_requested_credits,
1565 (handle->h_requested_credits -
1566 handle->h_buffer_credits));
1569 * Implement synchronous transaction batching. If the handle
1570 * was synchronous, don't force a commit immediately. Let's
1571 * yield and let another thread piggyback onto this
1572 * transaction. Keep doing that while new threads continue to
1573 * arrive. It doesn't cost much - we're about to run a commit
1574 * and sleep on IO anyway. Speeds up many-threaded, many-dir
1575 * operations by 30x or more...
1577 * We try and optimize the sleep time against what the
1578 * underlying disk can do, instead of having a static sleep
1579 * time. This is useful for the case where our storage is so
1580 * fast that it is more optimal to go ahead and force a flush
1581 * and wait for the transaction to be committed than it is to
1582 * wait for an arbitrary amount of time for new writers to
1583 * join the transaction. We achieve this by measuring how
1584 * long it takes to commit a transaction, and compare it with
1585 * how long this transaction has been running, and if run time
1586 * < commit time then we sleep for the delta and commit. This
1587 * greatly helps super fast disks that would see slowdowns as
1588 * more threads started doing fsyncs.
1590 * But don't do this if this process was the most recent one
1591 * to perform a synchronous write. We do this to detect the
1592 * case where a single process is doing a stream of sync
1593 * writes. No point in waiting for joiners in that case.
1595 * Setting max_batch_time to 0 disables this completely.
1597 pid = current->pid;
1598 if (handle->h_sync && journal->j_last_sync_writer != pid &&
1599 journal->j_max_batch_time) {
1600 u64 commit_time, trans_time;
1602 journal->j_last_sync_writer = pid;
1604 read_lock(&journal->j_state_lock);
1605 commit_time = journal->j_average_commit_time;
1606 read_unlock(&journal->j_state_lock);
1608 trans_time = ktime_to_ns(ktime_sub(ktime_get(),
1609 transaction->t_start_time));
1611 commit_time = max_t(u64, commit_time,
1612 1000*journal->j_min_batch_time);
1613 commit_time = min_t(u64, commit_time,
1614 1000*journal->j_max_batch_time);
1616 if (trans_time < commit_time) {
1617 ktime_t expires = ktime_add_ns(ktime_get(),
1618 commit_time);
1619 set_current_state(TASK_UNINTERRUPTIBLE);
1620 schedule_hrtimeout(&expires, HRTIMER_MODE_ABS);
1624 if (handle->h_sync)
1625 transaction->t_synchronous_commit = 1;
1626 current->journal_info = NULL;
1627 atomic_sub(handle->h_buffer_credits,
1628 &transaction->t_outstanding_credits);
1631 * If the handle is marked SYNC, we need to set another commit
1632 * going! We also want to force a commit if the current
1633 * transaction is occupying too much of the log, or if the
1634 * transaction is too old now.
1636 if (handle->h_sync ||
1637 (atomic_read(&transaction->t_outstanding_credits) >
1638 journal->j_max_transaction_buffers) ||
1639 time_after_eq(jiffies, transaction->t_expires)) {
1640 /* Do this even for aborted journals: an abort still
1641 * completes the commit thread, it just doesn't write
1642 * anything to disk. */
1644 jbd_debug(2, "transaction too old, requesting commit for "
1645 "handle %p\n", handle);
1646 /* This is non-blocking */
1647 jbd2_log_start_commit(journal, transaction->t_tid);
1650 * Special case: JBD2_SYNC synchronous updates require us
1651 * to wait for the commit to complete.
1653 if (handle->h_sync && !(current->flags & PF_MEMALLOC))
1654 wait_for_commit = 1;
1658 * Once we drop t_updates, if it goes to zero the transaction
1659 * could start committing on us and eventually disappear. So
1660 * once we do this, we must not dereference transaction
1661 * pointer again.
1663 tid = transaction->t_tid;
1664 if (atomic_dec_and_test(&transaction->t_updates)) {
1665 wake_up(&journal->j_wait_updates);
1666 if (journal->j_barrier_count)
1667 wake_up(&journal->j_wait_transaction_locked);
1670 if (wait_for_commit)
1671 err = jbd2_log_wait_commit(journal, tid);
1673 lock_map_release(&handle->h_lockdep_map);
1675 if (handle->h_rsv_handle)
1676 jbd2_journal_free_reserved(handle->h_rsv_handle);
1677 free_and_exit:
1678 jbd2_free_handle(handle);
1679 return err;
1684 * List management code snippets: various functions for manipulating the
1685 * transaction buffer lists.
1690 * Append a buffer to a transaction list, given the transaction's list head
1691 * pointer.
1693 * j_list_lock is held.
1695 * jbd_lock_bh_state(jh2bh(jh)) is held.
1698 static inline void
1699 __blist_add_buffer(struct journal_head **list, struct journal_head *jh)
1701 if (!*list) {
1702 jh->b_tnext = jh->b_tprev = jh;
1703 *list = jh;
1704 } else {
1705 /* Insert at the tail of the list to preserve order */
1706 struct journal_head *first = *list, *last = first->b_tprev;
1707 jh->b_tprev = last;
1708 jh->b_tnext = first;
1709 last->b_tnext = first->b_tprev = jh;
1714 * Remove a buffer from a transaction list, given the transaction's list
1715 * head pointer.
1717 * Called with j_list_lock held, and the journal may not be locked.
1719 * jbd_lock_bh_state(jh2bh(jh)) is held.
1722 static inline void
1723 __blist_del_buffer(struct journal_head **list, struct journal_head *jh)
1725 if (*list == jh) {
1726 *list = jh->b_tnext;
1727 if (*list == jh)
1728 *list = NULL;
1730 jh->b_tprev->b_tnext = jh->b_tnext;
1731 jh->b_tnext->b_tprev = jh->b_tprev;
1735 * Remove a buffer from the appropriate transaction list.
1737 * Note that this function can *change* the value of
1738 * bh->b_transaction->t_buffers, t_forget, t_shadow_list, t_log_list or
1739 * t_reserved_list. If the caller is holding onto a copy of one of these
1740 * pointers, it could go bad. Generally the caller needs to re-read the
1741 * pointer from the transaction_t.
1743 * Called under j_list_lock.
1745 static void __jbd2_journal_temp_unlink_buffer(struct journal_head *jh)
1747 struct journal_head **list = NULL;
1748 transaction_t *transaction;
1749 struct buffer_head *bh = jh2bh(jh);
1751 J_ASSERT_JH(jh, jbd_is_locked_bh_state(bh));
1752 transaction = jh->b_transaction;
1753 if (transaction)
1754 assert_spin_locked(&transaction->t_journal->j_list_lock);
1756 J_ASSERT_JH(jh, jh->b_jlist < BJ_Types);
1757 if (jh->b_jlist != BJ_None)
1758 J_ASSERT_JH(jh, transaction != NULL);
1760 switch (jh->b_jlist) {
1761 case BJ_None:
1762 return;
1763 case BJ_Metadata:
1764 transaction->t_nr_buffers--;
1765 J_ASSERT_JH(jh, transaction->t_nr_buffers >= 0);
1766 list = &transaction->t_buffers;
1767 break;
1768 case BJ_Forget:
1769 list = &transaction->t_forget;
1770 break;
1771 case BJ_Shadow:
1772 list = &transaction->t_shadow_list;
1773 break;
1774 case BJ_Reserved:
1775 list = &transaction->t_reserved_list;
1776 break;
1779 __blist_del_buffer(list, jh);
1780 jh->b_jlist = BJ_None;
1781 if (test_clear_buffer_jbddirty(bh))
1782 mark_buffer_dirty(bh); /* Expose it to the VM */
1786 * Remove buffer from all transactions.
1788 * Called with bh_state lock and j_list_lock
1790 * jh and bh may be already freed when this function returns.
1792 static void __jbd2_journal_unfile_buffer(struct journal_head *jh)
1794 __jbd2_journal_temp_unlink_buffer(jh);
1795 jh->b_transaction = NULL;
1796 jbd2_journal_put_journal_head(jh);
1799 void jbd2_journal_unfile_buffer(journal_t *journal, struct journal_head *jh)
1801 struct buffer_head *bh = jh2bh(jh);
1803 /* Get reference so that buffer cannot be freed before we unlock it */
1804 get_bh(bh);
1805 jbd_lock_bh_state(bh);
1806 spin_lock(&journal->j_list_lock);
1807 __jbd2_journal_unfile_buffer(jh);
1808 spin_unlock(&journal->j_list_lock);
1809 jbd_unlock_bh_state(bh);
1810 __brelse(bh);
1814 * Called from jbd2_journal_try_to_free_buffers().
1816 * Called under jbd_lock_bh_state(bh)
1818 static void
1819 __journal_try_to_free_buffer(journal_t *journal, struct buffer_head *bh)
1821 struct journal_head *jh;
1823 jh = bh2jh(bh);
1825 if (buffer_locked(bh) || buffer_dirty(bh))
1826 goto out;
1828 if (jh->b_next_transaction != NULL || jh->b_transaction != NULL)
1829 goto out;
1831 spin_lock(&journal->j_list_lock);
1832 if (jh->b_cp_transaction != NULL) {
1833 /* written-back checkpointed metadata buffer */
1834 JBUFFER_TRACE(jh, "remove from checkpoint list");
1835 __jbd2_journal_remove_checkpoint(jh);
1837 spin_unlock(&journal->j_list_lock);
1838 out:
1839 return;
1843 * int jbd2_journal_try_to_free_buffers() - try to free page buffers.
1844 * @journal: journal for operation
1845 * @page: to try and free
1846 * @gfp_mask: we use the mask to detect how hard should we try to release
1847 * buffers. If __GFP_WAIT and __GFP_FS is set, we wait for commit code to
1848 * release the buffers.
1851 * For all the buffers on this page,
1852 * if they are fully written out ordered data, move them onto BUF_CLEAN
1853 * so try_to_free_buffers() can reap them.
1855 * This function returns non-zero if we wish try_to_free_buffers()
1856 * to be called. We do this if the page is releasable by try_to_free_buffers().
1857 * We also do it if the page has locked or dirty buffers and the caller wants
1858 * us to perform sync or async writeout.
1860 * This complicates JBD locking somewhat. We aren't protected by the
1861 * BKL here. We wish to remove the buffer from its committing or
1862 * running transaction's ->t_datalist via __jbd2_journal_unfile_buffer.
1864 * This may *change* the value of transaction_t->t_datalist, so anyone
1865 * who looks at t_datalist needs to lock against this function.
1867 * Even worse, someone may be doing a jbd2_journal_dirty_data on this
1868 * buffer. So we need to lock against that. jbd2_journal_dirty_data()
1869 * will come out of the lock with the buffer dirty, which makes it
1870 * ineligible for release here.
1872 * Who else is affected by this? hmm... Really the only contender
1873 * is do_get_write_access() - it could be looking at the buffer while
1874 * journal_try_to_free_buffer() is changing its state. But that
1875 * cannot happen because we never reallocate freed data as metadata
1876 * while the data is part of a transaction. Yes?
1878 * Return 0 on failure, 1 on success
1880 int jbd2_journal_try_to_free_buffers(journal_t *journal,
1881 struct page *page, gfp_t gfp_mask)
1883 struct buffer_head *head;
1884 struct buffer_head *bh;
1885 int ret = 0;
1887 J_ASSERT(PageLocked(page));
1889 head = page_buffers(page);
1890 bh = head;
1891 do {
1892 struct journal_head *jh;
1895 * We take our own ref against the journal_head here to avoid
1896 * having to add tons of locking around each instance of
1897 * jbd2_journal_put_journal_head().
1899 jh = jbd2_journal_grab_journal_head(bh);
1900 if (!jh)
1901 continue;
1903 jbd_lock_bh_state(bh);
1904 __journal_try_to_free_buffer(journal, bh);
1905 jbd2_journal_put_journal_head(jh);
1906 jbd_unlock_bh_state(bh);
1907 if (buffer_jbd(bh))
1908 goto busy;
1909 } while ((bh = bh->b_this_page) != head);
1911 ret = try_to_free_buffers(page);
1913 busy:
1914 return ret;
1918 * This buffer is no longer needed. If it is on an older transaction's
1919 * checkpoint list we need to record it on this transaction's forget list
1920 * to pin this buffer (and hence its checkpointing transaction) down until
1921 * this transaction commits. If the buffer isn't on a checkpoint list, we
1922 * release it.
1923 * Returns non-zero if JBD no longer has an interest in the buffer.
1925 * Called under j_list_lock.
1927 * Called under jbd_lock_bh_state(bh).
1929 static int __dispose_buffer(struct journal_head *jh, transaction_t *transaction)
1931 int may_free = 1;
1932 struct buffer_head *bh = jh2bh(jh);
1934 if (jh->b_cp_transaction) {
1935 JBUFFER_TRACE(jh, "on running+cp transaction");
1936 __jbd2_journal_temp_unlink_buffer(jh);
1938 * We don't want to write the buffer anymore, clear the
1939 * bit so that we don't confuse checks in
1940 * __journal_file_buffer
1942 clear_buffer_dirty(bh);
1943 __jbd2_journal_file_buffer(jh, transaction, BJ_Forget);
1944 may_free = 0;
1945 } else {
1946 JBUFFER_TRACE(jh, "on running transaction");
1947 __jbd2_journal_unfile_buffer(jh);
1949 return may_free;
1953 * jbd2_journal_invalidatepage
1955 * This code is tricky. It has a number of cases to deal with.
1957 * There are two invariants which this code relies on:
1959 * i_size must be updated on disk before we start calling invalidatepage on the
1960 * data.
1962 * This is done in ext3 by defining an ext3_setattr method which
1963 * updates i_size before truncate gets going. By maintaining this
1964 * invariant, we can be sure that it is safe to throw away any buffers
1965 * attached to the current transaction: once the transaction commits,
1966 * we know that the data will not be needed.
1968 * Note however that we can *not* throw away data belonging to the
1969 * previous, committing transaction!
1971 * Any disk blocks which *are* part of the previous, committing
1972 * transaction (and which therefore cannot be discarded immediately) are
1973 * not going to be reused in the new running transaction
1975 * The bitmap committed_data images guarantee this: any block which is
1976 * allocated in one transaction and removed in the next will be marked
1977 * as in-use in the committed_data bitmap, so cannot be reused until
1978 * the next transaction to delete the block commits. This means that
1979 * leaving committing buffers dirty is quite safe: the disk blocks
1980 * cannot be reallocated to a different file and so buffer aliasing is
1981 * not possible.
1984 * The above applies mainly to ordered data mode. In writeback mode we
1985 * don't make guarantees about the order in which data hits disk --- in
1986 * particular we don't guarantee that new dirty data is flushed before
1987 * transaction commit --- so it is always safe just to discard data
1988 * immediately in that mode. --sct
1992 * The journal_unmap_buffer helper function returns zero if the buffer
1993 * concerned remains pinned as an anonymous buffer belonging to an older
1994 * transaction.
1996 * We're outside-transaction here. Either or both of j_running_transaction
1997 * and j_committing_transaction may be NULL.
1999 static int journal_unmap_buffer(journal_t *journal, struct buffer_head *bh,
2000 int partial_page)
2002 transaction_t *transaction;
2003 struct journal_head *jh;
2004 int may_free = 1;
2006 BUFFER_TRACE(bh, "entry");
2009 * It is safe to proceed here without the j_list_lock because the
2010 * buffers cannot be stolen by try_to_free_buffers as long as we are
2011 * holding the page lock. --sct
2014 if (!buffer_jbd(bh))
2015 goto zap_buffer_unlocked;
2017 /* OK, we have data buffer in journaled mode */
2018 write_lock(&journal->j_state_lock);
2019 jbd_lock_bh_state(bh);
2020 spin_lock(&journal->j_list_lock);
2022 jh = jbd2_journal_grab_journal_head(bh);
2023 if (!jh)
2024 goto zap_buffer_no_jh;
2027 * We cannot remove the buffer from checkpoint lists until the
2028 * transaction adding inode to orphan list (let's call it T)
2029 * is committed. Otherwise if the transaction changing the
2030 * buffer would be cleaned from the journal before T is
2031 * committed, a crash will cause that the correct contents of
2032 * the buffer will be lost. On the other hand we have to
2033 * clear the buffer dirty bit at latest at the moment when the
2034 * transaction marking the buffer as freed in the filesystem
2035 * structures is committed because from that moment on the
2036 * block can be reallocated and used by a different page.
2037 * Since the block hasn't been freed yet but the inode has
2038 * already been added to orphan list, it is safe for us to add
2039 * the buffer to BJ_Forget list of the newest transaction.
2041 * Also we have to clear buffer_mapped flag of a truncated buffer
2042 * because the buffer_head may be attached to the page straddling
2043 * i_size (can happen only when blocksize < pagesize) and thus the
2044 * buffer_head can be reused when the file is extended again. So we end
2045 * up keeping around invalidated buffers attached to transactions'
2046 * BJ_Forget list just to stop checkpointing code from cleaning up
2047 * the transaction this buffer was modified in.
2049 transaction = jh->b_transaction;
2050 if (transaction == NULL) {
2051 /* First case: not on any transaction. If it
2052 * has no checkpoint link, then we can zap it:
2053 * it's a writeback-mode buffer so we don't care
2054 * if it hits disk safely. */
2055 if (!jh->b_cp_transaction) {
2056 JBUFFER_TRACE(jh, "not on any transaction: zap");
2057 goto zap_buffer;
2060 if (!buffer_dirty(bh)) {
2061 /* bdflush has written it. We can drop it now */
2062 goto zap_buffer;
2065 /* OK, it must be in the journal but still not
2066 * written fully to disk: it's metadata or
2067 * journaled data... */
2069 if (journal->j_running_transaction) {
2070 /* ... and once the current transaction has
2071 * committed, the buffer won't be needed any
2072 * longer. */
2073 JBUFFER_TRACE(jh, "checkpointed: add to BJ_Forget");
2074 may_free = __dispose_buffer(jh,
2075 journal->j_running_transaction);
2076 goto zap_buffer;
2077 } else {
2078 /* There is no currently-running transaction. So the
2079 * orphan record which we wrote for this file must have
2080 * passed into commit. We must attach this buffer to
2081 * the committing transaction, if it exists. */
2082 if (journal->j_committing_transaction) {
2083 JBUFFER_TRACE(jh, "give to committing trans");
2084 may_free = __dispose_buffer(jh,
2085 journal->j_committing_transaction);
2086 goto zap_buffer;
2087 } else {
2088 /* The orphan record's transaction has
2089 * committed. We can cleanse this buffer */
2090 clear_buffer_jbddirty(bh);
2091 goto zap_buffer;
2094 } else if (transaction == journal->j_committing_transaction) {
2095 JBUFFER_TRACE(jh, "on committing transaction");
2097 * The buffer is committing, we simply cannot touch
2098 * it. If the page is straddling i_size we have to wait
2099 * for commit and try again.
2101 if (partial_page) {
2102 jbd2_journal_put_journal_head(jh);
2103 spin_unlock(&journal->j_list_lock);
2104 jbd_unlock_bh_state(bh);
2105 write_unlock(&journal->j_state_lock);
2106 return -EBUSY;
2109 * OK, buffer won't be reachable after truncate. We just set
2110 * j_next_transaction to the running transaction (if there is
2111 * one) and mark buffer as freed so that commit code knows it
2112 * should clear dirty bits when it is done with the buffer.
2114 set_buffer_freed(bh);
2115 if (journal->j_running_transaction && buffer_jbddirty(bh))
2116 jh->b_next_transaction = journal->j_running_transaction;
2117 jbd2_journal_put_journal_head(jh);
2118 spin_unlock(&journal->j_list_lock);
2119 jbd_unlock_bh_state(bh);
2120 write_unlock(&journal->j_state_lock);
2121 return 0;
2122 } else {
2123 /* Good, the buffer belongs to the running transaction.
2124 * We are writing our own transaction's data, not any
2125 * previous one's, so it is safe to throw it away
2126 * (remember that we expect the filesystem to have set
2127 * i_size already for this truncate so recovery will not
2128 * expose the disk blocks we are discarding here.) */
2129 J_ASSERT_JH(jh, transaction == journal->j_running_transaction);
2130 JBUFFER_TRACE(jh, "on running transaction");
2131 may_free = __dispose_buffer(jh, transaction);
2134 zap_buffer:
2136 * This is tricky. Although the buffer is truncated, it may be reused
2137 * if blocksize < pagesize and it is attached to the page straddling
2138 * EOF. Since the buffer might have been added to BJ_Forget list of the
2139 * running transaction, journal_get_write_access() won't clear
2140 * b_modified and credit accounting gets confused. So clear b_modified
2141 * here.
2143 jh->b_modified = 0;
2144 jbd2_journal_put_journal_head(jh);
2145 zap_buffer_no_jh:
2146 spin_unlock(&journal->j_list_lock);
2147 jbd_unlock_bh_state(bh);
2148 write_unlock(&journal->j_state_lock);
2149 zap_buffer_unlocked:
2150 clear_buffer_dirty(bh);
2151 J_ASSERT_BH(bh, !buffer_jbddirty(bh));
2152 clear_buffer_mapped(bh);
2153 clear_buffer_req(bh);
2154 clear_buffer_new(bh);
2155 clear_buffer_delay(bh);
2156 clear_buffer_unwritten(bh);
2157 bh->b_bdev = NULL;
2158 return may_free;
2162 * void jbd2_journal_invalidatepage()
2163 * @journal: journal to use for flush...
2164 * @page: page to flush
2165 * @offset: start of the range to invalidate
2166 * @length: length of the range to invalidate
2168 * Reap page buffers containing data after in the specified range in page.
2169 * Can return -EBUSY if buffers are part of the committing transaction and
2170 * the page is straddling i_size. Caller then has to wait for current commit
2171 * and try again.
2173 int jbd2_journal_invalidatepage(journal_t *journal,
2174 struct page *page,
2175 unsigned int offset,
2176 unsigned int length)
2178 struct buffer_head *head, *bh, *next;
2179 unsigned int stop = offset + length;
2180 unsigned int curr_off = 0;
2181 int partial_page = (offset || length < PAGE_CACHE_SIZE);
2182 int may_free = 1;
2183 int ret = 0;
2185 if (!PageLocked(page))
2186 BUG();
2187 if (!page_has_buffers(page))
2188 return 0;
2190 BUG_ON(stop > PAGE_CACHE_SIZE || stop < length);
2192 /* We will potentially be playing with lists other than just the
2193 * data lists (especially for journaled data mode), so be
2194 * cautious in our locking. */
2196 head = bh = page_buffers(page);
2197 do {
2198 unsigned int next_off = curr_off + bh->b_size;
2199 next = bh->b_this_page;
2201 if (next_off > stop)
2202 return 0;
2204 if (offset <= curr_off) {
2205 /* This block is wholly outside the truncation point */
2206 lock_buffer(bh);
2207 ret = journal_unmap_buffer(journal, bh, partial_page);
2208 unlock_buffer(bh);
2209 if (ret < 0)
2210 return ret;
2211 may_free &= ret;
2213 curr_off = next_off;
2214 bh = next;
2216 } while (bh != head);
2218 if (!partial_page) {
2219 if (may_free && try_to_free_buffers(page))
2220 J_ASSERT(!page_has_buffers(page));
2222 return 0;
2226 * File a buffer on the given transaction list.
2228 void __jbd2_journal_file_buffer(struct journal_head *jh,
2229 transaction_t *transaction, int jlist)
2231 struct journal_head **list = NULL;
2232 int was_dirty = 0;
2233 struct buffer_head *bh = jh2bh(jh);
2235 J_ASSERT_JH(jh, jbd_is_locked_bh_state(bh));
2236 assert_spin_locked(&transaction->t_journal->j_list_lock);
2238 J_ASSERT_JH(jh, jh->b_jlist < BJ_Types);
2239 J_ASSERT_JH(jh, jh->b_transaction == transaction ||
2240 jh->b_transaction == NULL);
2242 if (jh->b_transaction && jh->b_jlist == jlist)
2243 return;
2245 if (jlist == BJ_Metadata || jlist == BJ_Reserved ||
2246 jlist == BJ_Shadow || jlist == BJ_Forget) {
2248 * For metadata buffers, we track dirty bit in buffer_jbddirty
2249 * instead of buffer_dirty. We should not see a dirty bit set
2250 * here because we clear it in do_get_write_access but e.g.
2251 * tune2fs can modify the sb and set the dirty bit at any time
2252 * so we try to gracefully handle that.
2254 if (buffer_dirty(bh))
2255 warn_dirty_buffer(bh);
2256 if (test_clear_buffer_dirty(bh) ||
2257 test_clear_buffer_jbddirty(bh))
2258 was_dirty = 1;
2261 if (jh->b_transaction)
2262 __jbd2_journal_temp_unlink_buffer(jh);
2263 else
2264 jbd2_journal_grab_journal_head(bh);
2265 jh->b_transaction = transaction;
2267 switch (jlist) {
2268 case BJ_None:
2269 J_ASSERT_JH(jh, !jh->b_committed_data);
2270 J_ASSERT_JH(jh, !jh->b_frozen_data);
2271 return;
2272 case BJ_Metadata:
2273 transaction->t_nr_buffers++;
2274 list = &transaction->t_buffers;
2275 break;
2276 case BJ_Forget:
2277 list = &transaction->t_forget;
2278 break;
2279 case BJ_Shadow:
2280 list = &transaction->t_shadow_list;
2281 break;
2282 case BJ_Reserved:
2283 list = &transaction->t_reserved_list;
2284 break;
2287 __blist_add_buffer(list, jh);
2288 jh->b_jlist = jlist;
2290 if (was_dirty)
2291 set_buffer_jbddirty(bh);
2294 void jbd2_journal_file_buffer(struct journal_head *jh,
2295 transaction_t *transaction, int jlist)
2297 jbd_lock_bh_state(jh2bh(jh));
2298 spin_lock(&transaction->t_journal->j_list_lock);
2299 __jbd2_journal_file_buffer(jh, transaction, jlist);
2300 spin_unlock(&transaction->t_journal->j_list_lock);
2301 jbd_unlock_bh_state(jh2bh(jh));
2305 * Remove a buffer from its current buffer list in preparation for
2306 * dropping it from its current transaction entirely. If the buffer has
2307 * already started to be used by a subsequent transaction, refile the
2308 * buffer on that transaction's metadata list.
2310 * Called under j_list_lock
2311 * Called under jbd_lock_bh_state(jh2bh(jh))
2313 * jh and bh may be already free when this function returns
2315 void __jbd2_journal_refile_buffer(struct journal_head *jh)
2317 int was_dirty, jlist;
2318 struct buffer_head *bh = jh2bh(jh);
2320 J_ASSERT_JH(jh, jbd_is_locked_bh_state(bh));
2321 if (jh->b_transaction)
2322 assert_spin_locked(&jh->b_transaction->t_journal->j_list_lock);
2324 /* If the buffer is now unused, just drop it. */
2325 if (jh->b_next_transaction == NULL) {
2326 __jbd2_journal_unfile_buffer(jh);
2327 return;
2331 * It has been modified by a later transaction: add it to the new
2332 * transaction's metadata list.
2335 was_dirty = test_clear_buffer_jbddirty(bh);
2336 __jbd2_journal_temp_unlink_buffer(jh);
2338 * We set b_transaction here because b_next_transaction will inherit
2339 * our jh reference and thus __jbd2_journal_file_buffer() must not
2340 * take a new one.
2342 jh->b_transaction = jh->b_next_transaction;
2343 jh->b_next_transaction = NULL;
2344 if (buffer_freed(bh))
2345 jlist = BJ_Forget;
2346 else if (jh->b_modified)
2347 jlist = BJ_Metadata;
2348 else
2349 jlist = BJ_Reserved;
2350 __jbd2_journal_file_buffer(jh, jh->b_transaction, jlist);
2351 J_ASSERT_JH(jh, jh->b_transaction->t_state == T_RUNNING);
2353 if (was_dirty)
2354 set_buffer_jbddirty(bh);
2358 * __jbd2_journal_refile_buffer() with necessary locking added. We take our
2359 * bh reference so that we can safely unlock bh.
2361 * The jh and bh may be freed by this call.
2363 void jbd2_journal_refile_buffer(journal_t *journal, struct journal_head *jh)
2365 struct buffer_head *bh = jh2bh(jh);
2367 /* Get reference so that buffer cannot be freed before we unlock it */
2368 get_bh(bh);
2369 jbd_lock_bh_state(bh);
2370 spin_lock(&journal->j_list_lock);
2371 __jbd2_journal_refile_buffer(jh);
2372 jbd_unlock_bh_state(bh);
2373 spin_unlock(&journal->j_list_lock);
2374 __brelse(bh);
2378 * File inode in the inode list of the handle's transaction
2380 int jbd2_journal_file_inode(handle_t *handle, struct jbd2_inode *jinode)
2382 transaction_t *transaction = handle->h_transaction;
2383 journal_t *journal;
2385 if (is_handle_aborted(handle))
2386 return -EROFS;
2387 journal = transaction->t_journal;
2389 jbd_debug(4, "Adding inode %lu, tid:%d\n", jinode->i_vfs_inode->i_ino,
2390 transaction->t_tid);
2393 * First check whether inode isn't already on the transaction's
2394 * lists without taking the lock. Note that this check is safe
2395 * without the lock as we cannot race with somebody removing inode
2396 * from the transaction. The reason is that we remove inode from the
2397 * transaction only in journal_release_jbd_inode() and when we commit
2398 * the transaction. We are guarded from the first case by holding
2399 * a reference to the inode. We are safe against the second case
2400 * because if jinode->i_transaction == transaction, commit code
2401 * cannot touch the transaction because we hold reference to it,
2402 * and if jinode->i_next_transaction == transaction, commit code
2403 * will only file the inode where we want it.
2405 if (jinode->i_transaction == transaction ||
2406 jinode->i_next_transaction == transaction)
2407 return 0;
2409 spin_lock(&journal->j_list_lock);
2411 if (jinode->i_transaction == transaction ||
2412 jinode->i_next_transaction == transaction)
2413 goto done;
2416 * We only ever set this variable to 1 so the test is safe. Since
2417 * t_need_data_flush is likely to be set, we do the test to save some
2418 * cacheline bouncing
2420 if (!transaction->t_need_data_flush)
2421 transaction->t_need_data_flush = 1;
2422 /* On some different transaction's list - should be
2423 * the committing one */
2424 if (jinode->i_transaction) {
2425 J_ASSERT(jinode->i_next_transaction == NULL);
2426 J_ASSERT(jinode->i_transaction ==
2427 journal->j_committing_transaction);
2428 jinode->i_next_transaction = transaction;
2429 goto done;
2431 /* Not on any transaction list... */
2432 J_ASSERT(!jinode->i_next_transaction);
2433 jinode->i_transaction = transaction;
2434 list_add(&jinode->i_list, &transaction->t_inode_list);
2435 done:
2436 spin_unlock(&journal->j_list_lock);
2438 return 0;
2442 * File truncate and transaction commit interact with each other in a
2443 * non-trivial way. If a transaction writing data block A is
2444 * committing, we cannot discard the data by truncate until we have
2445 * written them. Otherwise if we crashed after the transaction with
2446 * write has committed but before the transaction with truncate has
2447 * committed, we could see stale data in block A. This function is a
2448 * helper to solve this problem. It starts writeout of the truncated
2449 * part in case it is in the committing transaction.
2451 * Filesystem code must call this function when inode is journaled in
2452 * ordered mode before truncation happens and after the inode has been
2453 * placed on orphan list with the new inode size. The second condition
2454 * avoids the race that someone writes new data and we start
2455 * committing the transaction after this function has been called but
2456 * before a transaction for truncate is started (and furthermore it
2457 * allows us to optimize the case where the addition to orphan list
2458 * happens in the same transaction as write --- we don't have to write
2459 * any data in such case).
2461 int jbd2_journal_begin_ordered_truncate(journal_t *journal,
2462 struct jbd2_inode *jinode,
2463 loff_t new_size)
2465 transaction_t *inode_trans, *commit_trans;
2466 int ret = 0;
2468 /* This is a quick check to avoid locking if not necessary */
2469 if (!jinode->i_transaction)
2470 goto out;
2471 /* Locks are here just to force reading of recent values, it is
2472 * enough that the transaction was not committing before we started
2473 * a transaction adding the inode to orphan list */
2474 read_lock(&journal->j_state_lock);
2475 commit_trans = journal->j_committing_transaction;
2476 read_unlock(&journal->j_state_lock);
2477 spin_lock(&journal->j_list_lock);
2478 inode_trans = jinode->i_transaction;
2479 spin_unlock(&journal->j_list_lock);
2480 if (inode_trans == commit_trans) {
2481 ret = filemap_fdatawrite_range(jinode->i_vfs_inode->i_mapping,
2482 new_size, LLONG_MAX);
2483 if (ret)
2484 jbd2_journal_abort(journal, ret);
2486 out:
2487 return ret;