2 * linux/fs/jbd/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/jbd.h>
19 #include <linux/errno.h>
20 #include <linux/slab.h>
22 #include <linux/pagemap.h>
25 * Default IO end handler for temporary BJ_IO buffer_heads.
27 static void journal_end_buffer_io_sync(struct buffer_head *bh, int uptodate)
31 set_buffer_uptodate(bh);
33 clear_buffer_uptodate(bh);
38 * When an ext3-ordered file is truncated, it is possible that many pages are
39 * not successfully freed, because they are attached to a committing transaction.
40 * After the transaction commits, these pages are left on the LRU, with no
41 * ->mapping, and with attached buffers. These pages are trivially reclaimable
42 * by the VM, but their apparent absence upsets the VM accounting, and it makes
43 * the numbers in /proc/meminfo look odd.
45 * So here, we have a buffer which has just come off the forget list. Look to
46 * see if we can strip all buffers from the backing page.
48 * Called under journal->j_list_lock. The caller provided us with a ref
49 * against the buffer, and we drop that here.
51 static void release_buffer_page(struct buffer_head *bh)
57 if (atomic_read(&bh->b_count) != 1)
65 /* OK, it's a truncated page */
66 if (!trylock_page(page))
71 try_to_free_buffers(page);
73 page_cache_release(page);
81 * Decrement reference counter for data buffer. If it has been marked
82 * 'BH_Freed', release it and the page to which it belongs if possible.
84 static void release_data_buffer(struct buffer_head *bh)
86 if (buffer_freed(bh)) {
87 clear_buffer_freed(bh);
88 release_buffer_page(bh);
94 * Try to acquire jbd_lock_bh_state() against the buffer, when j_list_lock is
95 * held. For ranking reasons we must trylock. If we lose, schedule away and
96 * return 0. j_list_lock is dropped in this case.
98 static int inverted_lock(journal_t *journal, struct buffer_head *bh)
100 if (!jbd_trylock_bh_state(bh)) {
101 spin_unlock(&journal->j_list_lock);
108 /* Done it all: now write the commit record. We should have
109 * cleaned up our previous buffers by now, so if we are in abort
110 * mode we can now just skip the rest of the journal write
113 * Returns 1 if the journal needs to be aborted or 0 on success
115 static int journal_write_commit_record(journal_t *journal,
116 transaction_t *commit_transaction)
118 struct journal_head *descriptor;
119 struct buffer_head *bh;
120 journal_header_t *header;
122 int barrier_done = 0;
124 if (is_journal_aborted(journal))
127 descriptor = journal_get_descriptor_buffer(journal);
131 bh = jh2bh(descriptor);
133 header = (journal_header_t *)(bh->b_data);
134 header->h_magic = cpu_to_be32(JFS_MAGIC_NUMBER);
135 header->h_blocktype = cpu_to_be32(JFS_COMMIT_BLOCK);
136 header->h_sequence = cpu_to_be32(commit_transaction->t_tid);
138 JBUFFER_TRACE(descriptor, "write commit block");
139 set_buffer_dirty(bh);
140 if (journal->j_flags & JFS_BARRIER) {
141 set_buffer_ordered(bh);
144 ret = sync_dirty_buffer(bh);
146 clear_buffer_ordered(bh);
147 /* is it possible for another commit to fail at roughly
148 * the same time as this one? If so, we don't want to
149 * trust the barrier flag in the super, but instead want
150 * to remember if we sent a barrier request
152 if (ret == -EOPNOTSUPP && barrier_done) {
153 char b[BDEVNAME_SIZE];
156 "JBD: barrier-based sync failed on %s - "
157 "disabling barriers\n",
158 bdevname(journal->j_dev, b));
159 spin_lock(&journal->j_state_lock);
160 journal->j_flags &= ~JFS_BARRIER;
161 spin_unlock(&journal->j_state_lock);
163 /* And try again, without the barrier */
164 set_buffer_uptodate(bh);
165 set_buffer_dirty(bh);
166 ret = sync_dirty_buffer(bh);
168 put_bh(bh); /* One for getblk() */
169 journal_put_journal_head(descriptor);
171 return (ret == -EIO);
174 static void journal_do_submit_data(struct buffer_head **wbuf, int bufs)
178 for (i = 0; i < bufs; i++) {
179 wbuf[i]->b_end_io = end_buffer_write_sync;
180 /* We use-up our safety reference in submit_bh() */
181 submit_bh(WRITE, wbuf[i]);
186 * Submit all the data buffers to disk
188 static int journal_submit_data_buffers(journal_t *journal,
189 transaction_t *commit_transaction)
191 struct journal_head *jh;
192 struct buffer_head *bh;
195 struct buffer_head **wbuf = journal->j_wbuf;
199 * Whenever we unlock the journal and sleep, things can get added
200 * onto ->t_sync_datalist, so we have to keep looping back to
201 * write_out_data until we *know* that the list is empty.
203 * Cleanup any flushed data buffers from the data list. Even in
204 * abort mode, we want to flush this out as soon as possible.
208 spin_lock(&journal->j_list_lock);
210 while (commit_transaction->t_sync_datalist) {
211 jh = commit_transaction->t_sync_datalist;
215 /* Get reference just to make sure buffer does not disappear
216 * when we are forced to drop various locks */
218 /* If the buffer is dirty, we need to submit IO and hence
219 * we need the buffer lock. We try to lock the buffer without
220 * blocking. If we fail, we need to drop j_list_lock and do
221 * blocking lock_buffer().
223 if (buffer_dirty(bh)) {
224 if (!trylock_buffer(bh)) {
225 BUFFER_TRACE(bh, "needs blocking lock");
226 spin_unlock(&journal->j_list_lock);
227 /* Write out all data to prevent deadlocks */
228 journal_do_submit_data(wbuf, bufs);
231 spin_lock(&journal->j_list_lock);
235 /* We have to get bh_state lock. Again out of order, sigh. */
236 if (!inverted_lock(journal, bh)) {
237 jbd_lock_bh_state(bh);
238 spin_lock(&journal->j_list_lock);
240 /* Someone already cleaned up the buffer? */
242 || jh->b_transaction != commit_transaction
243 || jh->b_jlist != BJ_SyncData) {
244 jbd_unlock_bh_state(bh);
247 BUFFER_TRACE(bh, "already cleaned up");
248 release_data_buffer(bh);
251 if (locked && test_clear_buffer_dirty(bh)) {
252 BUFFER_TRACE(bh, "needs writeout, adding to array");
254 __journal_file_buffer(jh, commit_transaction,
256 jbd_unlock_bh_state(bh);
257 if (bufs == journal->j_wbufsize) {
258 spin_unlock(&journal->j_list_lock);
259 journal_do_submit_data(wbuf, bufs);
263 } else if (!locked && buffer_locked(bh)) {
264 __journal_file_buffer(jh, commit_transaction,
266 jbd_unlock_bh_state(bh);
269 BUFFER_TRACE(bh, "writeout complete: unfile");
270 if (unlikely(!buffer_uptodate(bh)))
272 __journal_unfile_buffer(jh);
273 jbd_unlock_bh_state(bh);
276 journal_remove_journal_head(bh);
277 /* One for our safety reference, other for
278 * journal_remove_journal_head() */
280 release_data_buffer(bh);
283 if (need_resched() || spin_needbreak(&journal->j_list_lock)) {
284 spin_unlock(&journal->j_list_lock);
288 spin_unlock(&journal->j_list_lock);
289 journal_do_submit_data(wbuf, bufs);
295 * journal_commit_transaction
297 * The primary function for committing a transaction to the log. This
298 * function is called by the journal thread to begin a complete commit.
300 void journal_commit_transaction(journal_t *journal)
302 transaction_t *commit_transaction;
303 struct journal_head *jh, *new_jh, *descriptor;
304 struct buffer_head **wbuf = journal->j_wbuf;
308 unsigned long blocknr;
310 journal_header_t *header;
311 journal_block_tag_t *tag = NULL;
318 * First job: lock down the current transaction and wait for
319 * all outstanding updates to complete.
323 spin_lock(&journal->j_list_lock);
324 summarise_journal_usage(journal);
325 spin_unlock(&journal->j_list_lock);
328 /* Do we need to erase the effects of a prior journal_flush? */
329 if (journal->j_flags & JFS_FLUSHED) {
330 jbd_debug(3, "super block updated\n");
331 journal_update_superblock(journal, 1);
333 jbd_debug(3, "superblock not updated\n");
336 J_ASSERT(journal->j_running_transaction != NULL);
337 J_ASSERT(journal->j_committing_transaction == NULL);
339 commit_transaction = journal->j_running_transaction;
340 J_ASSERT(commit_transaction->t_state == T_RUNNING);
342 jbd_debug(1, "JBD: starting commit of transaction %d\n",
343 commit_transaction->t_tid);
345 spin_lock(&journal->j_state_lock);
346 commit_transaction->t_state = T_LOCKED;
348 spin_lock(&commit_transaction->t_handle_lock);
349 while (commit_transaction->t_updates) {
352 prepare_to_wait(&journal->j_wait_updates, &wait,
353 TASK_UNINTERRUPTIBLE);
354 if (commit_transaction->t_updates) {
355 spin_unlock(&commit_transaction->t_handle_lock);
356 spin_unlock(&journal->j_state_lock);
358 spin_lock(&journal->j_state_lock);
359 spin_lock(&commit_transaction->t_handle_lock);
361 finish_wait(&journal->j_wait_updates, &wait);
363 spin_unlock(&commit_transaction->t_handle_lock);
365 J_ASSERT (commit_transaction->t_outstanding_credits <=
366 journal->j_max_transaction_buffers);
369 * First thing we are allowed to do is to discard any remaining
370 * BJ_Reserved buffers. Note, it is _not_ permissible to assume
371 * that there are no such buffers: if a large filesystem
372 * operation like a truncate needs to split itself over multiple
373 * transactions, then it may try to do a journal_restart() while
374 * there are still BJ_Reserved buffers outstanding. These must
375 * be released cleanly from the current transaction.
377 * In this case, the filesystem must still reserve write access
378 * again before modifying the buffer in the new transaction, but
379 * we do not require it to remember exactly which old buffers it
380 * has reserved. This is consistent with the existing behaviour
381 * that multiple journal_get_write_access() calls to the same
382 * buffer are perfectly permissable.
384 while (commit_transaction->t_reserved_list) {
385 jh = commit_transaction->t_reserved_list;
386 JBUFFER_TRACE(jh, "reserved, unused: refile");
388 * A journal_get_undo_access()+journal_release_buffer() may
389 * leave undo-committed data.
391 if (jh->b_committed_data) {
392 struct buffer_head *bh = jh2bh(jh);
394 jbd_lock_bh_state(bh);
395 jbd_free(jh->b_committed_data, bh->b_size);
396 jh->b_committed_data = NULL;
397 jbd_unlock_bh_state(bh);
399 journal_refile_buffer(journal, jh);
403 * Now try to drop any written-back buffers from the journal's
404 * checkpoint lists. We do this *before* commit because it potentially
407 spin_lock(&journal->j_list_lock);
408 __journal_clean_checkpoint_list(journal);
409 spin_unlock(&journal->j_list_lock);
411 jbd_debug (3, "JBD: commit phase 1\n");
414 * Switch to a new revoke table.
416 journal_switch_revoke_table(journal);
418 commit_transaction->t_state = T_FLUSH;
419 journal->j_committing_transaction = commit_transaction;
420 journal->j_running_transaction = NULL;
421 commit_transaction->t_log_start = journal->j_head;
422 wake_up(&journal->j_wait_transaction_locked);
423 spin_unlock(&journal->j_state_lock);
425 jbd_debug (3, "JBD: commit phase 2\n");
428 * Now start flushing things to disk, in the order they appear
429 * on the transaction lists. Data blocks go first.
431 err = journal_submit_data_buffers(journal, commit_transaction);
434 * Wait for all previously submitted IO to complete.
436 spin_lock(&journal->j_list_lock);
437 while (commit_transaction->t_locked_list) {
438 struct buffer_head *bh;
440 jh = commit_transaction->t_locked_list->b_tprev;
443 if (buffer_locked(bh)) {
444 spin_unlock(&journal->j_list_lock);
446 spin_lock(&journal->j_list_lock);
448 if (unlikely(!buffer_uptodate(bh))) {
449 if (!trylock_page(bh->b_page)) {
450 spin_unlock(&journal->j_list_lock);
451 lock_page(bh->b_page);
452 spin_lock(&journal->j_list_lock);
454 if (bh->b_page->mapping)
455 set_bit(AS_EIO, &bh->b_page->mapping->flags);
457 unlock_page(bh->b_page);
458 SetPageError(bh->b_page);
461 if (!inverted_lock(journal, bh)) {
463 spin_lock(&journal->j_list_lock);
466 if (buffer_jbd(bh) && jh->b_jlist == BJ_Locked) {
467 __journal_unfile_buffer(jh);
468 jbd_unlock_bh_state(bh);
469 journal_remove_journal_head(bh);
472 jbd_unlock_bh_state(bh);
474 release_data_buffer(bh);
475 cond_resched_lock(&journal->j_list_lock);
477 spin_unlock(&journal->j_list_lock);
480 char b[BDEVNAME_SIZE];
483 "JBD: Detected IO errors while flushing file data "
484 "on %s\n", bdevname(journal->j_fs_dev, b));
488 journal_write_revoke_records(journal, commit_transaction);
491 * If we found any dirty or locked buffers, then we should have
492 * looped back up to the write_out_data label. If there weren't
493 * any then journal_clean_data_list should have wiped the list
494 * clean by now, so check that it is in fact empty.
496 J_ASSERT (commit_transaction->t_sync_datalist == NULL);
498 jbd_debug (3, "JBD: commit phase 3\n");
501 * Way to go: we have now written out all of the data for a
502 * transaction! Now comes the tricky part: we need to write out
503 * metadata. Loop over the transaction's entire buffer list:
505 spin_lock(&journal->j_state_lock);
506 commit_transaction->t_state = T_COMMIT;
507 spin_unlock(&journal->j_state_lock);
509 J_ASSERT(commit_transaction->t_nr_buffers <=
510 commit_transaction->t_outstanding_credits);
514 while (commit_transaction->t_buffers) {
516 /* Find the next buffer to be journaled... */
518 jh = commit_transaction->t_buffers;
520 /* If we're in abort mode, we just un-journal the buffer and
523 if (is_journal_aborted(journal)) {
524 clear_buffer_jbddirty(jh2bh(jh));
525 JBUFFER_TRACE(jh, "journal is aborting: refile");
526 journal_refile_buffer(journal, jh);
527 /* If that was the last one, we need to clean up
528 * any descriptor buffers which may have been
529 * already allocated, even if we are now
531 if (!commit_transaction->t_buffers)
532 goto start_journal_io;
536 /* Make sure we have a descriptor block in which to
537 record the metadata buffer. */
540 struct buffer_head *bh;
542 J_ASSERT (bufs == 0);
544 jbd_debug(4, "JBD: get descriptor\n");
546 descriptor = journal_get_descriptor_buffer(journal);
548 journal_abort(journal, -EIO);
552 bh = jh2bh(descriptor);
553 jbd_debug(4, "JBD: got buffer %llu (%p)\n",
554 (unsigned long long)bh->b_blocknr, bh->b_data);
555 header = (journal_header_t *)&bh->b_data[0];
556 header->h_magic = cpu_to_be32(JFS_MAGIC_NUMBER);
557 header->h_blocktype = cpu_to_be32(JFS_DESCRIPTOR_BLOCK);
558 header->h_sequence = cpu_to_be32(commit_transaction->t_tid);
560 tagp = &bh->b_data[sizeof(journal_header_t)];
561 space_left = bh->b_size - sizeof(journal_header_t);
563 set_buffer_jwrite(bh);
564 set_buffer_dirty(bh);
567 /* Record it so that we can wait for IO
569 BUFFER_TRACE(bh, "ph3: file as descriptor");
570 journal_file_buffer(descriptor, commit_transaction,
574 /* Where is the buffer to be written? */
576 err = journal_next_log_block(journal, &blocknr);
577 /* If the block mapping failed, just abandon the buffer
578 and repeat this loop: we'll fall into the
579 refile-on-abort condition above. */
581 journal_abort(journal, err);
586 * start_this_handle() uses t_outstanding_credits to determine
587 * the free space in the log, but this counter is changed
588 * by journal_next_log_block() also.
590 commit_transaction->t_outstanding_credits--;
592 /* Bump b_count to prevent truncate from stumbling over
593 the shadowed buffer! @@@ This can go if we ever get
594 rid of the BJ_IO/BJ_Shadow pairing of buffers. */
595 atomic_inc(&jh2bh(jh)->b_count);
597 /* Make a temporary IO buffer with which to write it out
598 (this will requeue both the metadata buffer and the
599 temporary IO buffer). new_bh goes on BJ_IO*/
601 set_bit(BH_JWrite, &jh2bh(jh)->b_state);
603 * akpm: journal_write_metadata_buffer() sets
604 * new_bh->b_transaction to commit_transaction.
605 * We need to clean this up before we release new_bh
606 * (which is of type BJ_IO)
608 JBUFFER_TRACE(jh, "ph3: write metadata");
609 flags = journal_write_metadata_buffer(commit_transaction,
610 jh, &new_jh, blocknr);
611 set_bit(BH_JWrite, &jh2bh(new_jh)->b_state);
612 wbuf[bufs++] = jh2bh(new_jh);
614 /* Record the new block's tag in the current descriptor
619 tag_flag |= JFS_FLAG_ESCAPE;
621 tag_flag |= JFS_FLAG_SAME_UUID;
623 tag = (journal_block_tag_t *) tagp;
624 tag->t_blocknr = cpu_to_be32(jh2bh(jh)->b_blocknr);
625 tag->t_flags = cpu_to_be32(tag_flag);
626 tagp += sizeof(journal_block_tag_t);
627 space_left -= sizeof(journal_block_tag_t);
630 memcpy (tagp, journal->j_uuid, 16);
636 /* If there's no more to do, or if the descriptor is full,
639 if (bufs == journal->j_wbufsize ||
640 commit_transaction->t_buffers == NULL ||
641 space_left < sizeof(journal_block_tag_t) + 16) {
643 jbd_debug(4, "JBD: Submit %d IOs\n", bufs);
645 /* Write an end-of-descriptor marker before
646 submitting the IOs. "tag" still points to
647 the last tag we set up. */
649 tag->t_flags |= cpu_to_be32(JFS_FLAG_LAST_TAG);
652 for (i = 0; i < bufs; i++) {
653 struct buffer_head *bh = wbuf[i];
655 clear_buffer_dirty(bh);
656 set_buffer_uptodate(bh);
657 bh->b_end_io = journal_end_buffer_io_sync;
658 submit_bh(WRITE, bh);
662 /* Force a new descriptor to be generated next
663 time round the loop. */
669 /* Lo and behold: we have just managed to send a transaction to
670 the log. Before we can commit it, wait for the IO so far to
671 complete. Control buffers being written are on the
672 transaction's t_log_list queue, and metadata buffers are on
673 the t_iobuf_list queue.
675 Wait for the buffers in reverse order. That way we are
676 less likely to be woken up until all IOs have completed, and
677 so we incur less scheduling load.
680 jbd_debug(3, "JBD: commit phase 4\n");
683 * akpm: these are BJ_IO, and j_list_lock is not needed.
684 * See __journal_try_to_free_buffer.
687 while (commit_transaction->t_iobuf_list != NULL) {
688 struct buffer_head *bh;
690 jh = commit_transaction->t_iobuf_list->b_tprev;
692 if (buffer_locked(bh)) {
699 if (unlikely(!buffer_uptodate(bh)))
702 clear_buffer_jwrite(bh);
704 JBUFFER_TRACE(jh, "ph4: unfile after journal write");
705 journal_unfile_buffer(journal, jh);
708 * ->t_iobuf_list should contain only dummy buffer_heads
709 * which were created by journal_write_metadata_buffer().
711 BUFFER_TRACE(bh, "dumping temporary bh");
712 journal_put_journal_head(jh);
714 J_ASSERT_BH(bh, atomic_read(&bh->b_count) == 0);
715 free_buffer_head(bh);
717 /* We also have to unlock and free the corresponding
719 jh = commit_transaction->t_shadow_list->b_tprev;
721 clear_bit(BH_JWrite, &bh->b_state);
722 J_ASSERT_BH(bh, buffer_jbddirty(bh));
724 /* The metadata is now released for reuse, but we need
725 to remember it against this transaction so that when
726 we finally commit, we can do any checkpointing
728 JBUFFER_TRACE(jh, "file as BJ_Forget");
729 journal_file_buffer(jh, commit_transaction, BJ_Forget);
730 /* Wake up any transactions which were waiting for this
732 wake_up_bit(&bh->b_state, BH_Unshadow);
733 JBUFFER_TRACE(jh, "brelse shadowed buffer");
737 J_ASSERT (commit_transaction->t_shadow_list == NULL);
739 jbd_debug(3, "JBD: commit phase 5\n");
741 /* Here we wait for the revoke record and descriptor record buffers */
743 while (commit_transaction->t_log_list != NULL) {
744 struct buffer_head *bh;
746 jh = commit_transaction->t_log_list->b_tprev;
748 if (buffer_locked(bh)) {
750 goto wait_for_ctlbuf;
753 goto wait_for_ctlbuf;
755 if (unlikely(!buffer_uptodate(bh)))
758 BUFFER_TRACE(bh, "ph5: control buffer writeout done: unfile");
759 clear_buffer_jwrite(bh);
760 journal_unfile_buffer(journal, jh);
761 journal_put_journal_head(jh);
762 __brelse(bh); /* One for getblk */
763 /* AKPM: bforget here */
767 journal_abort(journal, err);
769 jbd_debug(3, "JBD: commit phase 6\n");
771 if (journal_write_commit_record(journal, commit_transaction))
775 journal_abort(journal, err);
777 /* End of a transaction! Finally, we can do checkpoint
778 processing: any buffers committed as a result of this
779 transaction can be removed from any checkpoint list it was on
782 jbd_debug(3, "JBD: commit phase 7\n");
784 J_ASSERT(commit_transaction->t_sync_datalist == NULL);
785 J_ASSERT(commit_transaction->t_buffers == NULL);
786 J_ASSERT(commit_transaction->t_checkpoint_list == NULL);
787 J_ASSERT(commit_transaction->t_iobuf_list == NULL);
788 J_ASSERT(commit_transaction->t_shadow_list == NULL);
789 J_ASSERT(commit_transaction->t_log_list == NULL);
793 * As there are other places (journal_unmap_buffer()) adding buffers
794 * to this list we have to be careful and hold the j_list_lock.
796 spin_lock(&journal->j_list_lock);
797 while (commit_transaction->t_forget) {
798 transaction_t *cp_transaction;
799 struct buffer_head *bh;
801 jh = commit_transaction->t_forget;
802 spin_unlock(&journal->j_list_lock);
804 jbd_lock_bh_state(bh);
805 J_ASSERT_JH(jh, jh->b_transaction == commit_transaction ||
806 jh->b_transaction == journal->j_running_transaction);
809 * If there is undo-protected committed data against
810 * this buffer, then we can remove it now. If it is a
811 * buffer needing such protection, the old frozen_data
812 * field now points to a committed version of the
813 * buffer, so rotate that field to the new committed
816 * Otherwise, we can just throw away the frozen data now.
818 if (jh->b_committed_data) {
819 jbd_free(jh->b_committed_data, bh->b_size);
820 jh->b_committed_data = NULL;
821 if (jh->b_frozen_data) {
822 jh->b_committed_data = jh->b_frozen_data;
823 jh->b_frozen_data = NULL;
825 } else if (jh->b_frozen_data) {
826 jbd_free(jh->b_frozen_data, bh->b_size);
827 jh->b_frozen_data = NULL;
830 spin_lock(&journal->j_list_lock);
831 cp_transaction = jh->b_cp_transaction;
832 if (cp_transaction) {
833 JBUFFER_TRACE(jh, "remove from old cp transaction");
834 __journal_remove_checkpoint(jh);
837 /* Only re-checkpoint the buffer_head if it is marked
838 * dirty. If the buffer was added to the BJ_Forget list
839 * by journal_forget, it may no longer be dirty and
840 * there's no point in keeping a checkpoint record for
843 /* A buffer which has been freed while still being
844 * journaled by a previous transaction may end up still
845 * being dirty here, but we want to avoid writing back
846 * that buffer in the future now that the last use has
847 * been committed. That's not only a performance gain,
848 * it also stops aliasing problems if the buffer is left
849 * behind for writeback and gets reallocated for another
850 * use in a different page. */
851 if (buffer_freed(bh)) {
852 clear_buffer_freed(bh);
853 clear_buffer_jbddirty(bh);
856 if (buffer_jbddirty(bh)) {
857 JBUFFER_TRACE(jh, "add to new checkpointing trans");
858 __journal_insert_checkpoint(jh, commit_transaction);
859 if (is_journal_aborted(journal))
860 clear_buffer_jbddirty(bh);
861 JBUFFER_TRACE(jh, "refile for checkpoint writeback");
862 __journal_refile_buffer(jh);
863 jbd_unlock_bh_state(bh);
865 J_ASSERT_BH(bh, !buffer_dirty(bh));
866 /* The buffer on BJ_Forget list and not jbddirty means
867 * it has been freed by this transaction and hence it
868 * could not have been reallocated until this
869 * transaction has committed. *BUT* it could be
870 * reallocated once we have written all the data to
871 * disk and before we process the buffer on BJ_Forget
873 JBUFFER_TRACE(jh, "refile or unfile freed buffer");
874 __journal_refile_buffer(jh);
875 if (!jh->b_transaction) {
876 jbd_unlock_bh_state(bh);
878 journal_remove_journal_head(bh);
879 release_buffer_page(bh);
881 jbd_unlock_bh_state(bh);
883 cond_resched_lock(&journal->j_list_lock);
885 spin_unlock(&journal->j_list_lock);
887 * This is a bit sleazy. We use j_list_lock to protect transition
888 * of a transaction into T_FINISHED state and calling
889 * __journal_drop_transaction(). Otherwise we could race with
890 * other checkpointing code processing the transaction...
892 spin_lock(&journal->j_state_lock);
893 spin_lock(&journal->j_list_lock);
895 * Now recheck if some buffers did not get attached to the transaction
896 * while the lock was dropped...
898 if (commit_transaction->t_forget) {
899 spin_unlock(&journal->j_list_lock);
900 spin_unlock(&journal->j_state_lock);
904 /* Done with this transaction! */
906 jbd_debug(3, "JBD: commit phase 8\n");
908 J_ASSERT(commit_transaction->t_state == T_COMMIT);
910 commit_transaction->t_state = T_FINISHED;
911 J_ASSERT(commit_transaction == journal->j_committing_transaction);
912 journal->j_commit_sequence = commit_transaction->t_tid;
913 journal->j_committing_transaction = NULL;
914 spin_unlock(&journal->j_state_lock);
916 if (commit_transaction->t_checkpoint_list == NULL &&
917 commit_transaction->t_checkpoint_io_list == NULL) {
918 __journal_drop_transaction(journal, commit_transaction);
920 if (journal->j_checkpoint_transactions == NULL) {
921 journal->j_checkpoint_transactions = commit_transaction;
922 commit_transaction->t_cpnext = commit_transaction;
923 commit_transaction->t_cpprev = commit_transaction;
925 commit_transaction->t_cpnext =
926 journal->j_checkpoint_transactions;
927 commit_transaction->t_cpprev =
928 commit_transaction->t_cpnext->t_cpprev;
929 commit_transaction->t_cpnext->t_cpprev =
931 commit_transaction->t_cpprev->t_cpnext =
935 spin_unlock(&journal->j_list_lock);
937 jbd_debug(1, "JBD: commit %d complete, head %d\n",
938 journal->j_commit_sequence, journal->j_tail_sequence);
940 wake_up(&journal->j_wait_done_commit);