2 * linux/fs/jbd2/commit.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 * Journal commit routines for the generic filesystem journaling code;
13 * part of the ext2fs journaling system.
16 #include <linux/time.h>
18 #include <linux/jbd2.h>
19 #include <linux/errno.h>
20 #include <linux/slab.h>
22 #include <linux/pagemap.h>
23 #include <linux/jiffies.h>
24 #include <linux/crc32.h>
25 #include <linux/writeback.h>
26 #include <linux/backing-dev.h>
27 #include <linux/bio.h>
28 #include <linux/blkdev.h>
29 #include <linux/bitops.h>
30 #include <trace/events/jbd2.h>
31 #include <asm/system.h>
34 * Default IO end handler for temporary BJ_IO buffer_heads.
36 static void journal_end_buffer_io_sync(struct buffer_head *bh, int uptodate)
40 set_buffer_uptodate(bh);
42 clear_buffer_uptodate(bh);
47 * When an ext4 file is truncated, it is possible that some pages are not
48 * successfully freed, because they are attached to a committing transaction.
49 * After the transaction commits, these pages are left on the LRU, with no
50 * ->mapping, and with attached buffers. These pages are trivially reclaimable
51 * by the VM, but their apparent absence upsets the VM accounting, and it makes
52 * the numbers in /proc/meminfo look odd.
54 * So here, we have a buffer which has just come off the forget list. Look to
55 * see if we can strip all buffers from the backing page.
57 * Called under lock_journal(), and possibly under journal_datalist_lock. The
58 * caller provided us with a ref against the buffer, and we drop that here.
60 static void release_buffer_page(struct buffer_head *bh)
66 if (atomic_read(&bh->b_count) != 1)
74 /* OK, it's a truncated page */
75 if (!trylock_page(page))
80 try_to_free_buffers(page);
82 page_cache_release(page);
90 * Done it all: now submit the commit record. We should have
91 * cleaned up our previous buffers by now, so if we are in abort
92 * mode we can now just skip the rest of the journal write
95 * Returns 1 if the journal needs to be aborted or 0 on success
97 static int journal_submit_commit_record(journal_t *journal,
98 transaction_t *commit_transaction,
99 struct buffer_head **cbh,
102 struct journal_head *descriptor;
103 struct commit_header *tmp;
104 struct buffer_head *bh;
106 struct timespec now = current_kernel_time();
110 if (is_journal_aborted(journal))
113 descriptor = jbd2_journal_get_descriptor_buffer(journal);
117 bh = jh2bh(descriptor);
119 tmp = (struct commit_header *)bh->b_data;
120 tmp->h_magic = cpu_to_be32(JBD2_MAGIC_NUMBER);
121 tmp->h_blocktype = cpu_to_be32(JBD2_COMMIT_BLOCK);
122 tmp->h_sequence = cpu_to_be32(commit_transaction->t_tid);
123 tmp->h_commit_sec = cpu_to_be64(now.tv_sec);
124 tmp->h_commit_nsec = cpu_to_be32(now.tv_nsec);
126 if (JBD2_HAS_COMPAT_FEATURE(journal,
127 JBD2_FEATURE_COMPAT_CHECKSUM)) {
128 tmp->h_chksum_type = JBD2_CRC32_CHKSUM;
129 tmp->h_chksum_size = JBD2_CRC32_CHKSUM_SIZE;
130 tmp->h_chksum[0] = cpu_to_be32(crc32_sum);
133 JBUFFER_TRACE(descriptor, "submit commit block");
135 clear_buffer_dirty(bh);
136 set_buffer_uptodate(bh);
137 bh->b_end_io = journal_end_buffer_io_sync;
139 if (journal->j_flags & JBD2_BARRIER &&
140 !JBD2_HAS_INCOMPAT_FEATURE(journal,
141 JBD2_FEATURE_INCOMPAT_ASYNC_COMMIT))
142 ret = submit_bh(WRITE_SYNC | WRITE_FLUSH_FUA, bh);
144 ret = submit_bh(WRITE_SYNC, bh);
151 * This function along with journal_submit_commit_record
152 * allows to write the commit record asynchronously.
154 static int journal_wait_on_commit_record(journal_t *journal,
155 struct buffer_head *bh)
159 clear_buffer_dirty(bh);
162 if (unlikely(!buffer_uptodate(bh)))
164 put_bh(bh); /* One for getblk() */
165 jbd2_journal_put_journal_head(bh2jh(bh));
171 * write the filemap data using writepage() address_space_operations.
172 * We don't do block allocation here even for delalloc. We don't
173 * use writepages() because with dealyed allocation we may be doing
174 * block allocation in writepages().
176 static int journal_submit_inode_data_buffers(struct address_space *mapping)
179 struct writeback_control wbc = {
180 .sync_mode = WB_SYNC_ALL,
181 .nr_to_write = mapping->nrpages * 2,
183 .range_end = i_size_read(mapping->host),
186 ret = generic_writepages(mapping, &wbc);
191 * Submit all the data buffers of inode associated with the transaction to
194 * We are in a committing transaction. Therefore no new inode can be added to
195 * our inode list. We use JI_COMMIT_RUNNING flag to protect inode we currently
196 * operate on from being released while we write out pages.
198 static int journal_submit_data_buffers(journal_t *journal,
199 transaction_t *commit_transaction)
201 struct jbd2_inode *jinode;
203 struct address_space *mapping;
205 spin_lock(&journal->j_list_lock);
206 list_for_each_entry(jinode, &commit_transaction->t_inode_list, i_list) {
207 mapping = jinode->i_vfs_inode->i_mapping;
208 set_bit(__JI_COMMIT_RUNNING, &jinode->i_flags);
209 spin_unlock(&journal->j_list_lock);
211 * submit the inode data buffers. We use writepage
212 * instead of writepages. Because writepages can do
213 * block allocation with delalloc. We need to write
214 * only allocated blocks here.
216 trace_jbd2_submit_inode_data(jinode->i_vfs_inode);
217 err = journal_submit_inode_data_buffers(mapping);
220 spin_lock(&journal->j_list_lock);
221 J_ASSERT(jinode->i_transaction == commit_transaction);
222 commit_transaction->t_flushed_data_blocks = 1;
223 clear_bit(__JI_COMMIT_RUNNING, &jinode->i_flags);
224 smp_mb__after_clear_bit();
225 wake_up_bit(&jinode->i_flags, __JI_COMMIT_RUNNING);
227 spin_unlock(&journal->j_list_lock);
232 * Wait for data submitted for writeout, refile inodes to proper
233 * transaction if needed.
236 static int journal_finish_inode_data_buffers(journal_t *journal,
237 transaction_t *commit_transaction)
239 struct jbd2_inode *jinode, *next_i;
242 /* For locking, see the comment in journal_submit_data_buffers() */
243 spin_lock(&journal->j_list_lock);
244 list_for_each_entry(jinode, &commit_transaction->t_inode_list, i_list) {
245 set_bit(__JI_COMMIT_RUNNING, &jinode->i_flags);
246 spin_unlock(&journal->j_list_lock);
247 err = filemap_fdatawait(jinode->i_vfs_inode->i_mapping);
250 * Because AS_EIO is cleared by
251 * filemap_fdatawait_range(), set it again so
252 * that user process can get -EIO from fsync().
255 &jinode->i_vfs_inode->i_mapping->flags);
260 spin_lock(&journal->j_list_lock);
261 clear_bit(__JI_COMMIT_RUNNING, &jinode->i_flags);
262 smp_mb__after_clear_bit();
263 wake_up_bit(&jinode->i_flags, __JI_COMMIT_RUNNING);
266 /* Now refile inode to proper lists */
267 list_for_each_entry_safe(jinode, next_i,
268 &commit_transaction->t_inode_list, i_list) {
269 list_del(&jinode->i_list);
270 if (jinode->i_next_transaction) {
271 jinode->i_transaction = jinode->i_next_transaction;
272 jinode->i_next_transaction = NULL;
273 list_add(&jinode->i_list,
274 &jinode->i_transaction->t_inode_list);
276 jinode->i_transaction = NULL;
279 spin_unlock(&journal->j_list_lock);
284 static __u32 jbd2_checksum_data(__u32 crc32_sum, struct buffer_head *bh)
286 struct page *page = bh->b_page;
290 addr = kmap_atomic(page, KM_USER0);
291 checksum = crc32_be(crc32_sum,
292 (void *)(addr + offset_in_page(bh->b_data)), bh->b_size);
293 kunmap_atomic(addr, KM_USER0);
298 static void write_tag_block(int tag_bytes, journal_block_tag_t *tag,
299 unsigned long long block)
301 tag->t_blocknr = cpu_to_be32(block & (u32)~0);
302 if (tag_bytes > JBD2_TAG_SIZE32)
303 tag->t_blocknr_high = cpu_to_be32((block >> 31) >> 1);
307 * jbd2_journal_commit_transaction
309 * The primary function for committing a transaction to the log. This
310 * function is called by the journal thread to begin a complete commit.
312 void jbd2_journal_commit_transaction(journal_t *journal)
314 struct transaction_stats_s stats;
315 transaction_t *commit_transaction;
316 struct journal_head *jh, *new_jh, *descriptor;
317 struct buffer_head **wbuf = journal->j_wbuf;
321 unsigned long long blocknr;
325 journal_header_t *header;
326 journal_block_tag_t *tag = NULL;
331 int tag_bytes = journal_tag_bytes(journal);
332 struct buffer_head *cbh = NULL; /* For transactional checksums */
333 __u32 crc32_sum = ~0;
334 struct blk_plug plug;
337 * First job: lock down the current transaction and wait for
338 * all outstanding updates to complete.
341 /* Do we need to erase the effects of a prior jbd2_journal_flush? */
342 if (journal->j_flags & JBD2_FLUSHED) {
343 jbd_debug(3, "super block updated\n");
344 jbd2_journal_update_superblock(journal, 1);
346 jbd_debug(3, "superblock not updated\n");
349 J_ASSERT(journal->j_running_transaction != NULL);
350 J_ASSERT(journal->j_committing_transaction == NULL);
352 commit_transaction = journal->j_running_transaction;
353 J_ASSERT(commit_transaction->t_state == T_RUNNING);
355 trace_jbd2_start_commit(journal, commit_transaction);
356 jbd_debug(1, "JBD: starting commit of transaction %d\n",
357 commit_transaction->t_tid);
359 write_lock(&journal->j_state_lock);
360 commit_transaction->t_state = T_LOCKED;
362 trace_jbd2_commit_locking(journal, commit_transaction);
363 stats.run.rs_wait = commit_transaction->t_max_wait;
364 stats.run.rs_locked = jiffies;
365 stats.run.rs_running = jbd2_time_diff(commit_transaction->t_start,
366 stats.run.rs_locked);
368 spin_lock(&commit_transaction->t_handle_lock);
369 while (atomic_read(&commit_transaction->t_updates)) {
372 prepare_to_wait(&journal->j_wait_updates, &wait,
373 TASK_UNINTERRUPTIBLE);
374 if (atomic_read(&commit_transaction->t_updates)) {
375 spin_unlock(&commit_transaction->t_handle_lock);
376 write_unlock(&journal->j_state_lock);
378 write_lock(&journal->j_state_lock);
379 spin_lock(&commit_transaction->t_handle_lock);
381 finish_wait(&journal->j_wait_updates, &wait);
383 spin_unlock(&commit_transaction->t_handle_lock);
385 J_ASSERT (atomic_read(&commit_transaction->t_outstanding_credits) <=
386 journal->j_max_transaction_buffers);
389 * First thing we are allowed to do is to discard any remaining
390 * BJ_Reserved buffers. Note, it is _not_ permissible to assume
391 * that there are no such buffers: if a large filesystem
392 * operation like a truncate needs to split itself over multiple
393 * transactions, then it may try to do a jbd2_journal_restart() while
394 * there are still BJ_Reserved buffers outstanding. These must
395 * be released cleanly from the current transaction.
397 * In this case, the filesystem must still reserve write access
398 * again before modifying the buffer in the new transaction, but
399 * we do not require it to remember exactly which old buffers it
400 * has reserved. This is consistent with the existing behaviour
401 * that multiple jbd2_journal_get_write_access() calls to the same
402 * buffer are perfectly permissible.
404 while (commit_transaction->t_reserved_list) {
405 jh = commit_transaction->t_reserved_list;
406 JBUFFER_TRACE(jh, "reserved, unused: refile");
408 * A jbd2_journal_get_undo_access()+jbd2_journal_release_buffer() may
409 * leave undo-committed data.
411 if (jh->b_committed_data) {
412 struct buffer_head *bh = jh2bh(jh);
414 jbd_lock_bh_state(bh);
415 jbd2_free(jh->b_committed_data, bh->b_size);
416 jh->b_committed_data = NULL;
417 jbd_unlock_bh_state(bh);
419 jbd2_journal_refile_buffer(journal, jh);
423 * Now try to drop any written-back buffers from the journal's
424 * checkpoint lists. We do this *before* commit because it potentially
427 spin_lock(&journal->j_list_lock);
428 __jbd2_journal_clean_checkpoint_list(journal);
429 spin_unlock(&journal->j_list_lock);
431 jbd_debug (3, "JBD: commit phase 1\n");
434 * Switch to a new revoke table.
436 jbd2_journal_switch_revoke_table(journal);
438 trace_jbd2_commit_flushing(journal, commit_transaction);
439 stats.run.rs_flushing = jiffies;
440 stats.run.rs_locked = jbd2_time_diff(stats.run.rs_locked,
441 stats.run.rs_flushing);
443 commit_transaction->t_state = T_FLUSH;
444 journal->j_committing_transaction = commit_transaction;
445 journal->j_running_transaction = NULL;
446 start_time = ktime_get();
447 commit_transaction->t_log_start = journal->j_head;
448 wake_up(&journal->j_wait_transaction_locked);
449 write_unlock(&journal->j_state_lock);
451 jbd_debug (3, "JBD: commit phase 2\n");
454 * Now start flushing things to disk, in the order they appear
455 * on the transaction lists. Data blocks go first.
457 err = journal_submit_data_buffers(journal, commit_transaction);
459 jbd2_journal_abort(journal, err);
461 blk_start_plug(&plug);
462 jbd2_journal_write_revoke_records(journal, commit_transaction,
464 blk_finish_plug(&plug);
466 jbd_debug(3, "JBD: commit phase 2\n");
469 * Way to go: we have now written out all of the data for a
470 * transaction! Now comes the tricky part: we need to write out
471 * metadata. Loop over the transaction's entire buffer list:
473 write_lock(&journal->j_state_lock);
474 commit_transaction->t_state = T_COMMIT;
475 write_unlock(&journal->j_state_lock);
477 trace_jbd2_commit_logging(journal, commit_transaction);
478 stats.run.rs_logging = jiffies;
479 stats.run.rs_flushing = jbd2_time_diff(stats.run.rs_flushing,
480 stats.run.rs_logging);
481 stats.run.rs_blocks =
482 atomic_read(&commit_transaction->t_outstanding_credits);
483 stats.run.rs_blocks_logged = 0;
485 J_ASSERT(commit_transaction->t_nr_buffers <=
486 atomic_read(&commit_transaction->t_outstanding_credits));
491 blk_start_plug(&plug);
492 while (commit_transaction->t_buffers) {
494 /* Find the next buffer to be journaled... */
496 jh = commit_transaction->t_buffers;
498 /* If we're in abort mode, we just un-journal the buffer and
501 if (is_journal_aborted(journal)) {
502 clear_buffer_jbddirty(jh2bh(jh));
503 JBUFFER_TRACE(jh, "journal is aborting: refile");
504 jbd2_buffer_abort_trigger(jh,
506 jh->b_frozen_triggers :
508 jbd2_journal_refile_buffer(journal, jh);
509 /* If that was the last one, we need to clean up
510 * any descriptor buffers which may have been
511 * already allocated, even if we are now
513 if (!commit_transaction->t_buffers)
514 goto start_journal_io;
518 /* Make sure we have a descriptor block in which to
519 record the metadata buffer. */
522 struct buffer_head *bh;
524 J_ASSERT (bufs == 0);
526 jbd_debug(4, "JBD: get descriptor\n");
528 descriptor = jbd2_journal_get_descriptor_buffer(journal);
530 jbd2_journal_abort(journal, -EIO);
534 bh = jh2bh(descriptor);
535 jbd_debug(4, "JBD: got buffer %llu (%p)\n",
536 (unsigned long long)bh->b_blocknr, bh->b_data);
537 header = (journal_header_t *)&bh->b_data[0];
538 header->h_magic = cpu_to_be32(JBD2_MAGIC_NUMBER);
539 header->h_blocktype = cpu_to_be32(JBD2_DESCRIPTOR_BLOCK);
540 header->h_sequence = cpu_to_be32(commit_transaction->t_tid);
542 tagp = &bh->b_data[sizeof(journal_header_t)];
543 space_left = bh->b_size - sizeof(journal_header_t);
545 set_buffer_jwrite(bh);
546 set_buffer_dirty(bh);
549 /* Record it so that we can wait for IO
551 BUFFER_TRACE(bh, "ph3: file as descriptor");
552 jbd2_journal_file_buffer(descriptor, commit_transaction,
556 /* Where is the buffer to be written? */
558 err = jbd2_journal_next_log_block(journal, &blocknr);
559 /* If the block mapping failed, just abandon the buffer
560 and repeat this loop: we'll fall into the
561 refile-on-abort condition above. */
563 jbd2_journal_abort(journal, err);
568 * start_this_handle() uses t_outstanding_credits to determine
569 * the free space in the log, but this counter is changed
570 * by jbd2_journal_next_log_block() also.
572 atomic_dec(&commit_transaction->t_outstanding_credits);
574 /* Bump b_count to prevent truncate from stumbling over
575 the shadowed buffer! @@@ This can go if we ever get
576 rid of the BJ_IO/BJ_Shadow pairing of buffers. */
577 atomic_inc(&jh2bh(jh)->b_count);
579 /* Make a temporary IO buffer with which to write it out
580 (this will requeue both the metadata buffer and the
581 temporary IO buffer). new_bh goes on BJ_IO*/
583 set_bit(BH_JWrite, &jh2bh(jh)->b_state);
585 * akpm: jbd2_journal_write_metadata_buffer() sets
586 * new_bh->b_transaction to commit_transaction.
587 * We need to clean this up before we release new_bh
588 * (which is of type BJ_IO)
590 JBUFFER_TRACE(jh, "ph3: write metadata");
591 flags = jbd2_journal_write_metadata_buffer(commit_transaction,
592 jh, &new_jh, blocknr);
594 jbd2_journal_abort(journal, flags);
597 set_bit(BH_JWrite, &jh2bh(new_jh)->b_state);
598 wbuf[bufs++] = jh2bh(new_jh);
600 /* Record the new block's tag in the current descriptor
605 tag_flag |= JBD2_FLAG_ESCAPE;
607 tag_flag |= JBD2_FLAG_SAME_UUID;
609 tag = (journal_block_tag_t *) tagp;
610 write_tag_block(tag_bytes, tag, jh2bh(jh)->b_blocknr);
611 tag->t_flags = cpu_to_be32(tag_flag);
613 space_left -= tag_bytes;
616 memcpy (tagp, journal->j_uuid, 16);
622 /* If there's no more to do, or if the descriptor is full,
625 if (bufs == journal->j_wbufsize ||
626 commit_transaction->t_buffers == NULL ||
627 space_left < tag_bytes + 16) {
629 jbd_debug(4, "JBD: Submit %d IOs\n", bufs);
631 /* Write an end-of-descriptor marker before
632 submitting the IOs. "tag" still points to
633 the last tag we set up. */
635 tag->t_flags |= cpu_to_be32(JBD2_FLAG_LAST_TAG);
638 for (i = 0; i < bufs; i++) {
639 struct buffer_head *bh = wbuf[i];
643 if (JBD2_HAS_COMPAT_FEATURE(journal,
644 JBD2_FEATURE_COMPAT_CHECKSUM)) {
646 jbd2_checksum_data(crc32_sum, bh);
650 clear_buffer_dirty(bh);
651 set_buffer_uptodate(bh);
652 bh->b_end_io = journal_end_buffer_io_sync;
653 submit_bh(WRITE_SYNC, bh);
656 stats.run.rs_blocks_logged += bufs;
658 /* Force a new descriptor to be generated next
659 time round the loop. */
665 err = journal_finish_inode_data_buffers(journal, commit_transaction);
668 "JBD2: Detected IO errors while flushing file data "
669 "on %s\n", journal->j_devname);
670 if (journal->j_flags & JBD2_ABORT_ON_SYNCDATA_ERR)
671 jbd2_journal_abort(journal, err);
676 * If the journal is not located on the file system device,
677 * then we must flush the file system device before we issue
680 if (commit_transaction->t_flushed_data_blocks &&
681 (journal->j_fs_dev != journal->j_dev) &&
682 (journal->j_flags & JBD2_BARRIER))
683 blkdev_issue_flush(journal->j_fs_dev, GFP_KERNEL, NULL);
685 /* Done it all: now write the commit record asynchronously. */
686 if (JBD2_HAS_INCOMPAT_FEATURE(journal,
687 JBD2_FEATURE_INCOMPAT_ASYNC_COMMIT)) {
688 err = journal_submit_commit_record(journal, commit_transaction,
691 __jbd2_journal_abort_hard(journal);
694 blk_finish_plug(&plug);
696 /* Lo and behold: we have just managed to send a transaction to
697 the log. Before we can commit it, wait for the IO so far to
698 complete. Control buffers being written are on the
699 transaction's t_log_list queue, and metadata buffers are on
700 the t_iobuf_list queue.
702 Wait for the buffers in reverse order. That way we are
703 less likely to be woken up until all IOs have completed, and
704 so we incur less scheduling load.
707 jbd_debug(3, "JBD: commit phase 3\n");
710 * akpm: these are BJ_IO, and j_list_lock is not needed.
711 * See __journal_try_to_free_buffer.
714 while (commit_transaction->t_iobuf_list != NULL) {
715 struct buffer_head *bh;
717 jh = commit_transaction->t_iobuf_list->b_tprev;
719 if (buffer_locked(bh)) {
726 if (unlikely(!buffer_uptodate(bh)))
729 clear_buffer_jwrite(bh);
731 JBUFFER_TRACE(jh, "ph4: unfile after journal write");
732 jbd2_journal_unfile_buffer(journal, jh);
735 * ->t_iobuf_list should contain only dummy buffer_heads
736 * which were created by jbd2_journal_write_metadata_buffer().
738 BUFFER_TRACE(bh, "dumping temporary bh");
739 jbd2_journal_put_journal_head(jh);
741 J_ASSERT_BH(bh, atomic_read(&bh->b_count) == 0);
742 free_buffer_head(bh);
744 /* We also have to unlock and free the corresponding
746 jh = commit_transaction->t_shadow_list->b_tprev;
748 clear_bit(BH_JWrite, &bh->b_state);
749 J_ASSERT_BH(bh, buffer_jbddirty(bh));
751 /* The metadata is now released for reuse, but we need
752 to remember it against this transaction so that when
753 we finally commit, we can do any checkpointing
755 JBUFFER_TRACE(jh, "file as BJ_Forget");
756 jbd2_journal_file_buffer(jh, commit_transaction, BJ_Forget);
757 /* Wake up any transactions which were waiting for this
759 wake_up_bit(&bh->b_state, BH_Unshadow);
760 JBUFFER_TRACE(jh, "brelse shadowed buffer");
764 J_ASSERT (commit_transaction->t_shadow_list == NULL);
766 jbd_debug(3, "JBD: commit phase 4\n");
768 /* Here we wait for the revoke record and descriptor record buffers */
770 while (commit_transaction->t_log_list != NULL) {
771 struct buffer_head *bh;
773 jh = commit_transaction->t_log_list->b_tprev;
775 if (buffer_locked(bh)) {
777 goto wait_for_ctlbuf;
780 goto wait_for_ctlbuf;
782 if (unlikely(!buffer_uptodate(bh)))
785 BUFFER_TRACE(bh, "ph5: control buffer writeout done: unfile");
786 clear_buffer_jwrite(bh);
787 jbd2_journal_unfile_buffer(journal, jh);
788 jbd2_journal_put_journal_head(jh);
789 __brelse(bh); /* One for getblk */
790 /* AKPM: bforget here */
794 jbd2_journal_abort(journal, err);
796 jbd_debug(3, "JBD: commit phase 5\n");
798 if (!JBD2_HAS_INCOMPAT_FEATURE(journal,
799 JBD2_FEATURE_INCOMPAT_ASYNC_COMMIT)) {
800 err = journal_submit_commit_record(journal, commit_transaction,
803 __jbd2_journal_abort_hard(journal);
806 err = journal_wait_on_commit_record(journal, cbh);
807 if (JBD2_HAS_INCOMPAT_FEATURE(journal,
808 JBD2_FEATURE_INCOMPAT_ASYNC_COMMIT) &&
809 journal->j_flags & JBD2_BARRIER) {
810 blkdev_issue_flush(journal->j_dev, GFP_KERNEL, NULL);
814 jbd2_journal_abort(journal, err);
816 /* End of a transaction! Finally, we can do checkpoint
817 processing: any buffers committed as a result of this
818 transaction can be removed from any checkpoint list it was on
821 jbd_debug(3, "JBD: commit phase 6\n");
823 J_ASSERT(list_empty(&commit_transaction->t_inode_list));
824 J_ASSERT(commit_transaction->t_buffers == NULL);
825 J_ASSERT(commit_transaction->t_checkpoint_list == NULL);
826 J_ASSERT(commit_transaction->t_iobuf_list == NULL);
827 J_ASSERT(commit_transaction->t_shadow_list == NULL);
828 J_ASSERT(commit_transaction->t_log_list == NULL);
832 * As there are other places (journal_unmap_buffer()) adding buffers
833 * to this list we have to be careful and hold the j_list_lock.
835 spin_lock(&journal->j_list_lock);
836 while (commit_transaction->t_forget) {
837 transaction_t *cp_transaction;
838 struct buffer_head *bh;
840 jh = commit_transaction->t_forget;
841 spin_unlock(&journal->j_list_lock);
843 jbd_lock_bh_state(bh);
844 J_ASSERT_JH(jh, jh->b_transaction == commit_transaction);
847 * If there is undo-protected committed data against
848 * this buffer, then we can remove it now. If it is a
849 * buffer needing such protection, the old frozen_data
850 * field now points to a committed version of the
851 * buffer, so rotate that field to the new committed
854 * Otherwise, we can just throw away the frozen data now.
856 * We also know that the frozen data has already fired
857 * its triggers if they exist, so we can clear that too.
859 if (jh->b_committed_data) {
860 jbd2_free(jh->b_committed_data, bh->b_size);
861 jh->b_committed_data = NULL;
862 if (jh->b_frozen_data) {
863 jh->b_committed_data = jh->b_frozen_data;
864 jh->b_frozen_data = NULL;
865 jh->b_frozen_triggers = NULL;
867 } else if (jh->b_frozen_data) {
868 jbd2_free(jh->b_frozen_data, bh->b_size);
869 jh->b_frozen_data = NULL;
870 jh->b_frozen_triggers = NULL;
873 spin_lock(&journal->j_list_lock);
874 cp_transaction = jh->b_cp_transaction;
875 if (cp_transaction) {
876 JBUFFER_TRACE(jh, "remove from old cp transaction");
877 cp_transaction->t_chp_stats.cs_dropped++;
878 __jbd2_journal_remove_checkpoint(jh);
881 /* Only re-checkpoint the buffer_head if it is marked
882 * dirty. If the buffer was added to the BJ_Forget list
883 * by jbd2_journal_forget, it may no longer be dirty and
884 * there's no point in keeping a checkpoint record for
887 /* A buffer which has been freed while still being
888 * journaled by a previous transaction may end up still
889 * being dirty here, but we want to avoid writing back
890 * that buffer in the future after the "add to orphan"
891 * operation been committed, That's not only a performance
892 * gain, it also stops aliasing problems if the buffer is
893 * left behind for writeback and gets reallocated for another
894 * use in a different page. */
895 if (buffer_freed(bh) && !jh->b_next_transaction) {
896 clear_buffer_freed(bh);
897 clear_buffer_jbddirty(bh);
900 if (buffer_jbddirty(bh)) {
901 JBUFFER_TRACE(jh, "add to new checkpointing trans");
902 __jbd2_journal_insert_checkpoint(jh, commit_transaction);
903 if (is_journal_aborted(journal))
904 clear_buffer_jbddirty(bh);
905 JBUFFER_TRACE(jh, "refile for checkpoint writeback");
906 __jbd2_journal_refile_buffer(jh);
907 jbd_unlock_bh_state(bh);
909 J_ASSERT_BH(bh, !buffer_dirty(bh));
910 /* The buffer on BJ_Forget list and not jbddirty means
911 * it has been freed by this transaction and hence it
912 * could not have been reallocated until this
913 * transaction has committed. *BUT* it could be
914 * reallocated once we have written all the data to
915 * disk and before we process the buffer on BJ_Forget
917 JBUFFER_TRACE(jh, "refile or unfile freed buffer");
918 __jbd2_journal_refile_buffer(jh);
919 if (!jh->b_transaction) {
920 jbd_unlock_bh_state(bh);
922 jbd2_journal_remove_journal_head(bh);
923 release_buffer_page(bh);
925 jbd_unlock_bh_state(bh);
927 cond_resched_lock(&journal->j_list_lock);
929 spin_unlock(&journal->j_list_lock);
931 * This is a bit sleazy. We use j_list_lock to protect transition
932 * of a transaction into T_FINISHED state and calling
933 * __jbd2_journal_drop_transaction(). Otherwise we could race with
934 * other checkpointing code processing the transaction...
936 write_lock(&journal->j_state_lock);
937 spin_lock(&journal->j_list_lock);
939 * Now recheck if some buffers did not get attached to the transaction
940 * while the lock was dropped...
942 if (commit_transaction->t_forget) {
943 spin_unlock(&journal->j_list_lock);
944 write_unlock(&journal->j_state_lock);
948 /* Done with this transaction! */
950 jbd_debug(3, "JBD: commit phase 7\n");
952 J_ASSERT(commit_transaction->t_state == T_COMMIT);
954 commit_transaction->t_start = jiffies;
955 stats.run.rs_logging = jbd2_time_diff(stats.run.rs_logging,
956 commit_transaction->t_start);
959 * File the transaction statistics
961 stats.ts_tid = commit_transaction->t_tid;
962 stats.run.rs_handle_count =
963 atomic_read(&commit_transaction->t_handle_count);
964 trace_jbd2_run_stats(journal->j_fs_dev->bd_dev,
965 commit_transaction->t_tid, &stats.run);
968 * Calculate overall stats
970 spin_lock(&journal->j_history_lock);
971 journal->j_stats.ts_tid++;
972 journal->j_stats.run.rs_wait += stats.run.rs_wait;
973 journal->j_stats.run.rs_running += stats.run.rs_running;
974 journal->j_stats.run.rs_locked += stats.run.rs_locked;
975 journal->j_stats.run.rs_flushing += stats.run.rs_flushing;
976 journal->j_stats.run.rs_logging += stats.run.rs_logging;
977 journal->j_stats.run.rs_handle_count += stats.run.rs_handle_count;
978 journal->j_stats.run.rs_blocks += stats.run.rs_blocks;
979 journal->j_stats.run.rs_blocks_logged += stats.run.rs_blocks_logged;
980 spin_unlock(&journal->j_history_lock);
982 commit_transaction->t_state = T_FINISHED;
983 J_ASSERT(commit_transaction == journal->j_committing_transaction);
984 journal->j_commit_sequence = commit_transaction->t_tid;
985 journal->j_committing_transaction = NULL;
986 commit_time = ktime_to_ns(ktime_sub(ktime_get(), start_time));
989 * weight the commit time higher than the average time so we don't
990 * react too strongly to vast changes in the commit time
992 if (likely(journal->j_average_commit_time))
993 journal->j_average_commit_time = (commit_time +
994 journal->j_average_commit_time*3) / 4;
996 journal->j_average_commit_time = commit_time;
997 write_unlock(&journal->j_state_lock);
999 if (commit_transaction->t_checkpoint_list == NULL &&
1000 commit_transaction->t_checkpoint_io_list == NULL) {
1001 __jbd2_journal_drop_transaction(journal, commit_transaction);
1004 if (journal->j_checkpoint_transactions == NULL) {
1005 journal->j_checkpoint_transactions = commit_transaction;
1006 commit_transaction->t_cpnext = commit_transaction;
1007 commit_transaction->t_cpprev = commit_transaction;
1009 commit_transaction->t_cpnext =
1010 journal->j_checkpoint_transactions;
1011 commit_transaction->t_cpprev =
1012 commit_transaction->t_cpnext->t_cpprev;
1013 commit_transaction->t_cpnext->t_cpprev =
1015 commit_transaction->t_cpprev->t_cpnext =
1019 spin_unlock(&journal->j_list_lock);
1021 if (journal->j_commit_callback)
1022 journal->j_commit_callback(journal, commit_transaction);
1024 trace_jbd2_end_commit(journal, commit_transaction);
1025 jbd_debug(1, "JBD: commit %d complete, head %d\n",
1026 journal->j_commit_sequence, journal->j_tail_sequence);
1028 kfree(commit_transaction);
1030 wake_up(&journal->j_wait_done_commit);