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>
21 #include <linux/pagemap.h>
22 #include <linux/bio.h>
23 #include <linux/blkdev.h>
24 #include <trace/events/jbd.h>
27 * Default IO end handler for temporary BJ_IO buffer_heads.
29 static void journal_end_buffer_io_sync(struct buffer_head *bh, int uptodate)
33 set_buffer_uptodate(bh);
35 clear_buffer_uptodate(bh);
40 * When an ext3-ordered file is truncated, it is possible that many pages are
41 * not successfully freed, because they are attached to a committing transaction.
42 * After the transaction commits, these pages are left on the LRU, with no
43 * ->mapping, and with attached buffers. These pages are trivially reclaimable
44 * by the VM, but their apparent absence upsets the VM accounting, and it makes
45 * the numbers in /proc/meminfo look odd.
47 * So here, we have a buffer which has just come off the forget list. Look to
48 * see if we can strip all buffers from the backing page.
50 * Called under journal->j_list_lock. The caller provided us with a ref
51 * against the buffer, and we drop that here.
53 static void release_buffer_page(struct buffer_head *bh)
59 if (atomic_read(&bh->b_count) != 1)
67 /* OK, it's a truncated page */
68 if (!trylock_page(page))
73 try_to_free_buffers(page);
75 page_cache_release(page);
83 * Decrement reference counter for data buffer. If it has been marked
84 * 'BH_Freed', release it and the page to which it belongs if possible.
86 static void release_data_buffer(struct buffer_head *bh)
88 if (buffer_freed(bh)) {
89 WARN_ON_ONCE(buffer_dirty(bh));
90 clear_buffer_freed(bh);
91 clear_buffer_mapped(bh);
95 release_buffer_page(bh);
101 * Try to acquire jbd_lock_bh_state() against the buffer, when j_list_lock is
102 * held. For ranking reasons we must trylock. If we lose, schedule away and
103 * return 0. j_list_lock is dropped in this case.
105 static int inverted_lock(journal_t *journal, struct buffer_head *bh)
107 if (!jbd_trylock_bh_state(bh)) {
108 spin_unlock(&journal->j_list_lock);
115 /* Done it all: now write the commit record. We should have
116 * cleaned up our previous buffers by now, so if we are in abort
117 * mode we can now just skip the rest of the journal write
120 * Returns 1 if the journal needs to be aborted or 0 on success
122 static int journal_write_commit_record(journal_t *journal,
123 transaction_t *commit_transaction)
125 struct journal_head *descriptor;
126 struct buffer_head *bh;
127 journal_header_t *header;
130 if (is_journal_aborted(journal))
133 descriptor = journal_get_descriptor_buffer(journal);
137 bh = jh2bh(descriptor);
139 header = (journal_header_t *)(bh->b_data);
140 header->h_magic = cpu_to_be32(JFS_MAGIC_NUMBER);
141 header->h_blocktype = cpu_to_be32(JFS_COMMIT_BLOCK);
142 header->h_sequence = cpu_to_be32(commit_transaction->t_tid);
144 JBUFFER_TRACE(descriptor, "write commit block");
145 set_buffer_dirty(bh);
147 if (journal->j_flags & JFS_BARRIER)
148 ret = __sync_dirty_buffer(bh, WRITE_SYNC | WRITE_FLUSH_FUA);
150 ret = sync_dirty_buffer(bh);
152 put_bh(bh); /* One for getblk() */
153 journal_put_journal_head(descriptor);
155 return (ret == -EIO);
158 static void journal_do_submit_data(struct buffer_head **wbuf, int bufs,
163 for (i = 0; i < bufs; i++) {
164 wbuf[i]->b_end_io = end_buffer_write_sync;
165 /* We use-up our safety reference in submit_bh() */
166 submit_bh(write_op, wbuf[i]);
171 * Submit all the data buffers to disk
173 static int journal_submit_data_buffers(journal_t *journal,
174 transaction_t *commit_transaction,
177 struct journal_head *jh;
178 struct buffer_head *bh;
181 struct buffer_head **wbuf = journal->j_wbuf;
185 * Whenever we unlock the journal and sleep, things can get added
186 * onto ->t_sync_datalist, so we have to keep looping back to
187 * write_out_data until we *know* that the list is empty.
189 * Cleanup any flushed data buffers from the data list. Even in
190 * abort mode, we want to flush this out as soon as possible.
194 spin_lock(&journal->j_list_lock);
196 while (commit_transaction->t_sync_datalist) {
197 jh = commit_transaction->t_sync_datalist;
201 /* Get reference just to make sure buffer does not disappear
202 * when we are forced to drop various locks */
204 /* If the buffer is dirty, we need to submit IO and hence
205 * we need the buffer lock. We try to lock the buffer without
206 * blocking. If we fail, we need to drop j_list_lock and do
207 * blocking lock_buffer().
209 if (buffer_dirty(bh)) {
210 if (!trylock_buffer(bh)) {
211 BUFFER_TRACE(bh, "needs blocking lock");
212 spin_unlock(&journal->j_list_lock);
213 trace_jbd_do_submit_data(journal,
215 /* Write out all data to prevent deadlocks */
216 journal_do_submit_data(wbuf, bufs, write_op);
219 spin_lock(&journal->j_list_lock);
223 /* We have to get bh_state lock. Again out of order, sigh. */
224 if (!inverted_lock(journal, bh)) {
225 jbd_lock_bh_state(bh);
226 spin_lock(&journal->j_list_lock);
228 /* Someone already cleaned up the buffer? */
229 if (!buffer_jbd(bh) || bh2jh(bh) != jh
230 || jh->b_transaction != commit_transaction
231 || jh->b_jlist != BJ_SyncData) {
232 jbd_unlock_bh_state(bh);
235 BUFFER_TRACE(bh, "already cleaned up");
236 release_data_buffer(bh);
239 if (locked && test_clear_buffer_dirty(bh)) {
240 BUFFER_TRACE(bh, "needs writeout, adding to array");
242 __journal_file_buffer(jh, commit_transaction,
244 jbd_unlock_bh_state(bh);
245 if (bufs == journal->j_wbufsize) {
246 spin_unlock(&journal->j_list_lock);
247 trace_jbd_do_submit_data(journal,
249 journal_do_submit_data(wbuf, bufs, write_op);
253 } else if (!locked && buffer_locked(bh)) {
254 __journal_file_buffer(jh, commit_transaction,
256 jbd_unlock_bh_state(bh);
259 BUFFER_TRACE(bh, "writeout complete: unfile");
260 if (unlikely(!buffer_uptodate(bh)))
262 __journal_unfile_buffer(jh);
263 jbd_unlock_bh_state(bh);
266 release_data_buffer(bh);
269 if (need_resched() || spin_needbreak(&journal->j_list_lock)) {
270 spin_unlock(&journal->j_list_lock);
274 spin_unlock(&journal->j_list_lock);
275 trace_jbd_do_submit_data(journal, commit_transaction);
276 journal_do_submit_data(wbuf, bufs, write_op);
282 * journal_commit_transaction
284 * The primary function for committing a transaction to the log. This
285 * function is called by the journal thread to begin a complete commit.
287 void journal_commit_transaction(journal_t *journal)
289 transaction_t *commit_transaction;
290 struct journal_head *jh, *new_jh, *descriptor;
291 struct buffer_head **wbuf = journal->j_wbuf;
295 unsigned int blocknr;
299 journal_header_t *header;
300 journal_block_tag_t *tag = NULL;
305 struct blk_plug plug;
306 int write_op = WRITE;
309 * First job: lock down the current transaction and wait for
310 * all outstanding updates to complete.
313 /* Do we need to erase the effects of a prior journal_flush? */
314 if (journal->j_flags & JFS_FLUSHED) {
315 jbd_debug(3, "super block updated\n");
316 mutex_lock(&journal->j_checkpoint_mutex);
318 * We hold j_checkpoint_mutex so tail cannot change under us.
319 * We don't need any special data guarantees for writing sb
320 * since journal is empty and it is ok for write to be
321 * flushed only with transaction commit.
323 journal_update_sb_log_tail(journal, journal->j_tail_sequence,
324 journal->j_tail, WRITE_SYNC);
325 mutex_unlock(&journal->j_checkpoint_mutex);
327 jbd_debug(3, "superblock not updated\n");
330 J_ASSERT(journal->j_running_transaction != NULL);
331 J_ASSERT(journal->j_committing_transaction == NULL);
333 commit_transaction = journal->j_running_transaction;
334 J_ASSERT(commit_transaction->t_state == T_RUNNING);
336 trace_jbd_start_commit(journal, commit_transaction);
337 jbd_debug(1, "JBD: starting commit of transaction %d\n",
338 commit_transaction->t_tid);
340 spin_lock(&journal->j_state_lock);
341 commit_transaction->t_state = T_LOCKED;
343 trace_jbd_commit_locking(journal, commit_transaction);
344 spin_lock(&commit_transaction->t_handle_lock);
345 while (commit_transaction->t_updates) {
348 prepare_to_wait(&journal->j_wait_updates, &wait,
349 TASK_UNINTERRUPTIBLE);
350 if (commit_transaction->t_updates) {
351 spin_unlock(&commit_transaction->t_handle_lock);
352 spin_unlock(&journal->j_state_lock);
354 spin_lock(&journal->j_state_lock);
355 spin_lock(&commit_transaction->t_handle_lock);
357 finish_wait(&journal->j_wait_updates, &wait);
359 spin_unlock(&commit_transaction->t_handle_lock);
361 J_ASSERT (commit_transaction->t_outstanding_credits <=
362 journal->j_max_transaction_buffers);
365 * First thing we are allowed to do is to discard any remaining
366 * BJ_Reserved buffers. Note, it is _not_ permissible to assume
367 * that there are no such buffers: if a large filesystem
368 * operation like a truncate needs to split itself over multiple
369 * transactions, then it may try to do a journal_restart() while
370 * there are still BJ_Reserved buffers outstanding. These must
371 * be released cleanly from the current transaction.
373 * In this case, the filesystem must still reserve write access
374 * again before modifying the buffer in the new transaction, but
375 * we do not require it to remember exactly which old buffers it
376 * has reserved. This is consistent with the existing behaviour
377 * that multiple journal_get_write_access() calls to the same
378 * buffer are perfectly permissible.
380 while (commit_transaction->t_reserved_list) {
381 jh = commit_transaction->t_reserved_list;
382 JBUFFER_TRACE(jh, "reserved, unused: refile");
384 * A journal_get_undo_access()+journal_release_buffer() may
385 * leave undo-committed data.
387 if (jh->b_committed_data) {
388 struct buffer_head *bh = jh2bh(jh);
390 jbd_lock_bh_state(bh);
391 jbd_free(jh->b_committed_data, bh->b_size);
392 jh->b_committed_data = NULL;
393 jbd_unlock_bh_state(bh);
395 journal_refile_buffer(journal, jh);
399 * Now try to drop any written-back buffers from the journal's
400 * checkpoint lists. We do this *before* commit because it potentially
403 spin_lock(&journal->j_list_lock);
404 __journal_clean_checkpoint_list(journal);
405 spin_unlock(&journal->j_list_lock);
407 jbd_debug (3, "JBD: commit phase 1\n");
410 * Clear revoked flag to reflect there is no revoked buffers
411 * in the next transaction which is going to be started.
413 journal_clear_buffer_revoked_flags(journal);
416 * Switch to a new revoke table.
418 journal_switch_revoke_table(journal);
420 trace_jbd_commit_flushing(journal, commit_transaction);
421 commit_transaction->t_state = T_FLUSH;
422 journal->j_committing_transaction = commit_transaction;
423 journal->j_running_transaction = NULL;
424 start_time = ktime_get();
425 commit_transaction->t_log_start = journal->j_head;
426 wake_up(&journal->j_wait_transaction_locked);
427 spin_unlock(&journal->j_state_lock);
429 jbd_debug (3, "JBD: commit phase 2\n");
431 if (tid_geq(journal->j_commit_waited, commit_transaction->t_tid))
432 write_op = WRITE_SYNC;
435 * Now start flushing things to disk, in the order they appear
436 * on the transaction lists. Data blocks go first.
438 blk_start_plug(&plug);
439 err = journal_submit_data_buffers(journal, commit_transaction,
441 blk_finish_plug(&plug);
444 * Wait for all previously submitted IO to complete.
446 spin_lock(&journal->j_list_lock);
447 while (commit_transaction->t_locked_list) {
448 struct buffer_head *bh;
450 jh = commit_transaction->t_locked_list->b_tprev;
453 if (buffer_locked(bh)) {
454 spin_unlock(&journal->j_list_lock);
456 spin_lock(&journal->j_list_lock);
458 if (unlikely(!buffer_uptodate(bh))) {
459 if (!trylock_page(bh->b_page)) {
460 spin_unlock(&journal->j_list_lock);
461 lock_page(bh->b_page);
462 spin_lock(&journal->j_list_lock);
464 if (bh->b_page->mapping)
465 set_bit(AS_EIO, &bh->b_page->mapping->flags);
467 unlock_page(bh->b_page);
468 SetPageError(bh->b_page);
471 if (!inverted_lock(journal, bh)) {
473 spin_lock(&journal->j_list_lock);
476 if (buffer_jbd(bh) && bh2jh(bh) == jh &&
477 jh->b_transaction == commit_transaction &&
478 jh->b_jlist == BJ_Locked)
479 __journal_unfile_buffer(jh);
480 jbd_unlock_bh_state(bh);
481 release_data_buffer(bh);
482 cond_resched_lock(&journal->j_list_lock);
484 spin_unlock(&journal->j_list_lock);
487 char b[BDEVNAME_SIZE];
490 "JBD: Detected IO errors while flushing file data "
491 "on %s\n", bdevname(journal->j_fs_dev, b));
492 if (journal->j_flags & JFS_ABORT_ON_SYNCDATA_ERR)
493 journal_abort(journal, err);
497 blk_start_plug(&plug);
499 journal_write_revoke_records(journal, commit_transaction, write_op);
502 * If we found any dirty or locked buffers, then we should have
503 * looped back up to the write_out_data label. If there weren't
504 * any then journal_clean_data_list should have wiped the list
505 * clean by now, so check that it is in fact empty.
507 J_ASSERT (commit_transaction->t_sync_datalist == NULL);
509 jbd_debug (3, "JBD: commit phase 3\n");
512 * Way to go: we have now written out all of the data for a
513 * transaction! Now comes the tricky part: we need to write out
514 * metadata. Loop over the transaction's entire buffer list:
516 spin_lock(&journal->j_state_lock);
517 commit_transaction->t_state = T_COMMIT;
518 spin_unlock(&journal->j_state_lock);
520 trace_jbd_commit_logging(journal, commit_transaction);
521 J_ASSERT(commit_transaction->t_nr_buffers <=
522 commit_transaction->t_outstanding_credits);
526 while (commit_transaction->t_buffers) {
528 /* Find the next buffer to be journaled... */
530 jh = commit_transaction->t_buffers;
532 /* If we're in abort mode, we just un-journal the buffer and
535 if (is_journal_aborted(journal)) {
536 clear_buffer_jbddirty(jh2bh(jh));
537 JBUFFER_TRACE(jh, "journal is aborting: refile");
538 journal_refile_buffer(journal, jh);
539 /* If that was the last one, we need to clean up
540 * any descriptor buffers which may have been
541 * already allocated, even if we are now
543 if (!commit_transaction->t_buffers)
544 goto start_journal_io;
548 /* Make sure we have a descriptor block in which to
549 record the metadata buffer. */
552 struct buffer_head *bh;
554 J_ASSERT (bufs == 0);
556 jbd_debug(4, "JBD: get descriptor\n");
558 descriptor = journal_get_descriptor_buffer(journal);
560 journal_abort(journal, -EIO);
564 bh = jh2bh(descriptor);
565 jbd_debug(4, "JBD: got buffer %llu (%p)\n",
566 (unsigned long long)bh->b_blocknr, bh->b_data);
567 header = (journal_header_t *)&bh->b_data[0];
568 header->h_magic = cpu_to_be32(JFS_MAGIC_NUMBER);
569 header->h_blocktype = cpu_to_be32(JFS_DESCRIPTOR_BLOCK);
570 header->h_sequence = cpu_to_be32(commit_transaction->t_tid);
572 tagp = &bh->b_data[sizeof(journal_header_t)];
573 space_left = bh->b_size - sizeof(journal_header_t);
575 set_buffer_jwrite(bh);
576 set_buffer_dirty(bh);
579 /* Record it so that we can wait for IO
581 BUFFER_TRACE(bh, "ph3: file as descriptor");
582 journal_file_buffer(descriptor, commit_transaction,
586 /* Where is the buffer to be written? */
588 err = journal_next_log_block(journal, &blocknr);
589 /* If the block mapping failed, just abandon the buffer
590 and repeat this loop: we'll fall into the
591 refile-on-abort condition above. */
593 journal_abort(journal, err);
598 * start_this_handle() uses t_outstanding_credits to determine
599 * the free space in the log, but this counter is changed
600 * by journal_next_log_block() also.
602 commit_transaction->t_outstanding_credits--;
604 /* Bump b_count to prevent truncate from stumbling over
605 the shadowed buffer! @@@ This can go if we ever get
606 rid of the BJ_IO/BJ_Shadow pairing of buffers. */
609 /* Make a temporary IO buffer with which to write it out
610 (this will requeue both the metadata buffer and the
611 temporary IO buffer). new_bh goes on BJ_IO*/
613 set_buffer_jwrite(jh2bh(jh));
615 * akpm: journal_write_metadata_buffer() sets
616 * new_bh->b_transaction to commit_transaction.
617 * We need to clean this up before we release new_bh
618 * (which is of type BJ_IO)
620 JBUFFER_TRACE(jh, "ph3: write metadata");
621 flags = journal_write_metadata_buffer(commit_transaction,
622 jh, &new_jh, blocknr);
623 set_buffer_jwrite(jh2bh(new_jh));
624 wbuf[bufs++] = jh2bh(new_jh);
626 /* Record the new block's tag in the current descriptor
631 tag_flag |= JFS_FLAG_ESCAPE;
633 tag_flag |= JFS_FLAG_SAME_UUID;
635 tag = (journal_block_tag_t *) tagp;
636 tag->t_blocknr = cpu_to_be32(jh2bh(jh)->b_blocknr);
637 tag->t_flags = cpu_to_be32(tag_flag);
638 tagp += sizeof(journal_block_tag_t);
639 space_left -= sizeof(journal_block_tag_t);
642 memcpy (tagp, journal->j_uuid, 16);
648 /* If there's no more to do, or if the descriptor is full,
651 if (bufs == journal->j_wbufsize ||
652 commit_transaction->t_buffers == NULL ||
653 space_left < sizeof(journal_block_tag_t) + 16) {
655 jbd_debug(4, "JBD: Submit %d IOs\n", bufs);
657 /* Write an end-of-descriptor marker before
658 submitting the IOs. "tag" still points to
659 the last tag we set up. */
661 tag->t_flags |= cpu_to_be32(JFS_FLAG_LAST_TAG);
664 for (i = 0; i < bufs; i++) {
665 struct buffer_head *bh = wbuf[i];
667 clear_buffer_dirty(bh);
668 set_buffer_uptodate(bh);
669 bh->b_end_io = journal_end_buffer_io_sync;
670 submit_bh(write_op, bh);
674 /* Force a new descriptor to be generated next
675 time round the loop. */
681 blk_finish_plug(&plug);
683 /* Lo and behold: we have just managed to send a transaction to
684 the log. Before we can commit it, wait for the IO so far to
685 complete. Control buffers being written are on the
686 transaction's t_log_list queue, and metadata buffers are on
687 the t_iobuf_list queue.
689 Wait for the buffers in reverse order. That way we are
690 less likely to be woken up until all IOs have completed, and
691 so we incur less scheduling load.
694 jbd_debug(3, "JBD: commit phase 4\n");
697 * akpm: these are BJ_IO, and j_list_lock is not needed.
698 * See __journal_try_to_free_buffer.
701 while (commit_transaction->t_iobuf_list != NULL) {
702 struct buffer_head *bh;
704 jh = commit_transaction->t_iobuf_list->b_tprev;
706 if (buffer_locked(bh)) {
713 if (unlikely(!buffer_uptodate(bh)))
716 clear_buffer_jwrite(bh);
718 JBUFFER_TRACE(jh, "ph4: unfile after journal write");
719 journal_unfile_buffer(journal, jh);
722 * ->t_iobuf_list should contain only dummy buffer_heads
723 * which were created by journal_write_metadata_buffer().
725 BUFFER_TRACE(bh, "dumping temporary bh");
726 journal_put_journal_head(jh);
728 J_ASSERT_BH(bh, atomic_read(&bh->b_count) == 0);
729 free_buffer_head(bh);
731 /* We also have to unlock and free the corresponding
733 jh = commit_transaction->t_shadow_list->b_tprev;
735 clear_buffer_jwrite(bh);
736 J_ASSERT_BH(bh, buffer_jbddirty(bh));
738 /* The metadata is now released for reuse, but we need
739 to remember it against this transaction so that when
740 we finally commit, we can do any checkpointing
742 JBUFFER_TRACE(jh, "file as BJ_Forget");
743 journal_file_buffer(jh, commit_transaction, BJ_Forget);
745 * Wake up any transactions which were waiting for this
746 * IO to complete. The barrier must be here so that changes
747 * by journal_file_buffer() take effect before wake_up_bit()
748 * does the waitqueue check.
751 wake_up_bit(&bh->b_state, BH_Unshadow);
752 JBUFFER_TRACE(jh, "brelse shadowed buffer");
756 J_ASSERT (commit_transaction->t_shadow_list == NULL);
758 jbd_debug(3, "JBD: commit phase 5\n");
760 /* Here we wait for the revoke record and descriptor record buffers */
762 while (commit_transaction->t_log_list != NULL) {
763 struct buffer_head *bh;
765 jh = commit_transaction->t_log_list->b_tprev;
767 if (buffer_locked(bh)) {
769 goto wait_for_ctlbuf;
772 goto wait_for_ctlbuf;
774 if (unlikely(!buffer_uptodate(bh)))
777 BUFFER_TRACE(bh, "ph5: control buffer writeout done: unfile");
778 clear_buffer_jwrite(bh);
779 journal_unfile_buffer(journal, jh);
780 journal_put_journal_head(jh);
781 __brelse(bh); /* One for getblk */
782 /* AKPM: bforget here */
786 journal_abort(journal, err);
788 jbd_debug(3, "JBD: commit phase 6\n");
790 /* All metadata is written, now write commit record and do cleanup */
791 spin_lock(&journal->j_state_lock);
792 J_ASSERT(commit_transaction->t_state == T_COMMIT);
793 commit_transaction->t_state = T_COMMIT_RECORD;
794 spin_unlock(&journal->j_state_lock);
796 if (journal_write_commit_record(journal, commit_transaction))
800 journal_abort(journal, err);
802 /* End of a transaction! Finally, we can do checkpoint
803 processing: any buffers committed as a result of this
804 transaction can be removed from any checkpoint list it was on
807 jbd_debug(3, "JBD: commit phase 7\n");
809 J_ASSERT(commit_transaction->t_sync_datalist == NULL);
810 J_ASSERT(commit_transaction->t_buffers == NULL);
811 J_ASSERT(commit_transaction->t_checkpoint_list == NULL);
812 J_ASSERT(commit_transaction->t_iobuf_list == NULL);
813 J_ASSERT(commit_transaction->t_shadow_list == NULL);
814 J_ASSERT(commit_transaction->t_log_list == NULL);
818 * As there are other places (journal_unmap_buffer()) adding buffers
819 * to this list we have to be careful and hold the j_list_lock.
821 spin_lock(&journal->j_list_lock);
822 while (commit_transaction->t_forget) {
823 transaction_t *cp_transaction;
824 struct buffer_head *bh;
827 jh = commit_transaction->t_forget;
828 spin_unlock(&journal->j_list_lock);
831 * Get a reference so that bh cannot be freed before we are
835 jbd_lock_bh_state(bh);
836 J_ASSERT_JH(jh, jh->b_transaction == commit_transaction ||
837 jh->b_transaction == journal->j_running_transaction);
840 * If there is undo-protected committed data against
841 * this buffer, then we can remove it now. If it is a
842 * buffer needing such protection, the old frozen_data
843 * field now points to a committed version of the
844 * buffer, so rotate that field to the new committed
847 * Otherwise, we can just throw away the frozen data now.
849 if (jh->b_committed_data) {
850 jbd_free(jh->b_committed_data, bh->b_size);
851 jh->b_committed_data = NULL;
852 if (jh->b_frozen_data) {
853 jh->b_committed_data = jh->b_frozen_data;
854 jh->b_frozen_data = NULL;
856 } else if (jh->b_frozen_data) {
857 jbd_free(jh->b_frozen_data, bh->b_size);
858 jh->b_frozen_data = NULL;
861 spin_lock(&journal->j_list_lock);
862 cp_transaction = jh->b_cp_transaction;
863 if (cp_transaction) {
864 JBUFFER_TRACE(jh, "remove from old cp transaction");
865 __journal_remove_checkpoint(jh);
868 /* Only re-checkpoint the buffer_head if it is marked
869 * dirty. If the buffer was added to the BJ_Forget list
870 * by journal_forget, it may no longer be dirty and
871 * there's no point in keeping a checkpoint record for
875 * A buffer which has been freed while still being journaled by
876 * a previous transaction.
878 if (buffer_freed(bh)) {
880 * If the running transaction is the one containing
881 * "add to orphan" operation (b_next_transaction !=
882 * NULL), we have to wait for that transaction to
883 * commit before we can really get rid of the buffer.
884 * So just clear b_modified to not confuse transaction
885 * credit accounting and refile the buffer to
886 * BJ_Forget of the running transaction. If the just
887 * committed transaction contains "add to orphan"
888 * operation, we can completely invalidate the buffer
889 * now. We are rather throughout in that since the
890 * buffer may be still accessible when blocksize <
891 * pagesize and it is attached to the last partial
895 if (!jh->b_next_transaction) {
896 clear_buffer_freed(bh);
897 clear_buffer_jbddirty(bh);
898 clear_buffer_mapped(bh);
899 clear_buffer_new(bh);
900 clear_buffer_req(bh);
905 if (buffer_jbddirty(bh)) {
906 JBUFFER_TRACE(jh, "add to new checkpointing trans");
907 __journal_insert_checkpoint(jh, commit_transaction);
908 if (is_journal_aborted(journal))
909 clear_buffer_jbddirty(bh);
911 J_ASSERT_BH(bh, !buffer_dirty(bh));
913 * The buffer on BJ_Forget list and not jbddirty means
914 * it has been freed by this transaction and hence it
915 * could not have been reallocated until this
916 * transaction has committed. *BUT* it could be
917 * reallocated once we have written all the data to
918 * disk and before we process the buffer on BJ_Forget
921 if (!jh->b_next_transaction)
924 JBUFFER_TRACE(jh, "refile or unfile freed buffer");
925 __journal_refile_buffer(jh);
926 jbd_unlock_bh_state(bh);
928 release_buffer_page(bh);
931 cond_resched_lock(&journal->j_list_lock);
933 spin_unlock(&journal->j_list_lock);
935 * This is a bit sleazy. We use j_list_lock to protect transition
936 * of a transaction into T_FINISHED state and calling
937 * __journal_drop_transaction(). Otherwise we could race with
938 * other checkpointing code processing the transaction...
940 spin_lock(&journal->j_state_lock);
941 spin_lock(&journal->j_list_lock);
943 * Now recheck if some buffers did not get attached to the transaction
944 * while the lock was dropped...
946 if (commit_transaction->t_forget) {
947 spin_unlock(&journal->j_list_lock);
948 spin_unlock(&journal->j_state_lock);
952 /* Done with this transaction! */
954 jbd_debug(3, "JBD: commit phase 8\n");
956 J_ASSERT(commit_transaction->t_state == T_COMMIT_RECORD);
958 commit_transaction->t_state = T_FINISHED;
959 J_ASSERT(commit_transaction == journal->j_committing_transaction);
960 journal->j_commit_sequence = commit_transaction->t_tid;
961 journal->j_committing_transaction = NULL;
962 commit_time = ktime_to_ns(ktime_sub(ktime_get(), start_time));
965 * weight the commit time higher than the average time so we don't
966 * react too strongly to vast changes in commit time
968 if (likely(journal->j_average_commit_time))
969 journal->j_average_commit_time = (commit_time*3 +
970 journal->j_average_commit_time) / 4;
972 journal->j_average_commit_time = commit_time;
974 spin_unlock(&journal->j_state_lock);
976 if (commit_transaction->t_checkpoint_list == NULL &&
977 commit_transaction->t_checkpoint_io_list == NULL) {
978 __journal_drop_transaction(journal, commit_transaction);
980 if (journal->j_checkpoint_transactions == NULL) {
981 journal->j_checkpoint_transactions = commit_transaction;
982 commit_transaction->t_cpnext = commit_transaction;
983 commit_transaction->t_cpprev = commit_transaction;
985 commit_transaction->t_cpnext =
986 journal->j_checkpoint_transactions;
987 commit_transaction->t_cpprev =
988 commit_transaction->t_cpnext->t_cpprev;
989 commit_transaction->t_cpnext->t_cpprev =
991 commit_transaction->t_cpprev->t_cpnext =
995 spin_unlock(&journal->j_list_lock);
997 trace_jbd_end_commit(journal, commit_transaction);
998 jbd_debug(1, "JBD: commit %d complete, head %d\n",
999 journal->j_commit_sequence, journal->j_tail_sequence);
1001 wake_up(&journal->j_wait_done_commit);