| blob | d6aeb1f6cfe01809279f09436548b8698e962cd7 |
1 /*
2 * linux/fs/jbd2/commit.c
3 *
4 * Written by Stephen C. Tweedie <sct@redhat.com>, 1998
5 *
6 * Copyright 1998 Red Hat corp --- All Rights Reserved
7 *
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.
11 *
12 * Journal commit routines for the generic filesystem journaling code;
13 * part of the ext2fs journaling system.
14 */
16 #include <linux/time.h>
17 #include <linux/fs.h>
18 #include <linux/jbd2.h>
19 #include <linux/errno.h>
20 #include <linux/slab.h>
21 #include <linux/mm.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 <trace/events/jbd2.h>
31 /*
32 * Default IO end handler for temporary BJ_IO buffer_heads.
33 */
35 {
39 else
42 }
44 /*
45 * When an ext4 file is truncated, it is possible that some pages are not
46 * successfully freed, because they are attached to a committing transaction.
47 * After the transaction commits, these pages are left on the LRU, with no
48 * ->mapping, and with attached buffers. These pages are trivially reclaimable
49 * by the VM, but their apparent absence upsets the VM accounting, and it makes
50 * the numbers in /proc/meminfo look odd.
51 *
52 * So here, we have a buffer which has just come off the forget list. Look to
53 * see if we can strip all buffers from the backing page.
54 *
55 * Called under lock_journal(), and possibly under journal_datalist_lock. The
56 * caller provided us with a ref against the buffer, and we drop that here.
57 */
59 {
72 /* OK, it's a truncated page */
83 nope:
85 }
87 /*
88 * Done it all: now submit the commit record. We should have
89 * cleaned up our previous buffers by now, so if we are in abort
90 * mode we can now just skip the rest of the journal write
91 * entirely.
92 *
93 * Returns 1 if the journal needs to be aborted or 0 on success
94 */
98 __u32 crc32_sum)
99 {
123 JBD2_FEATURE_COMPAT_CHECKSUM)) {
127 }
137 JBD2_FEATURE_INCOMPAT_ASYNC_COMMIT)) {
141 "JBD2: Disabling barriers on %s, "
147 /* And try again, without the barrier */
152 }
155 }
158 }
160 /*
161 * This function along with journal_submit_commit_record
162 * allows to write the commit record asynchronously.
163 */
166 {
169 retry:
174 "JBD2: %s: disabling barries on %s - not supported "
189 }
191 }
199 }
201 /*
202 * write the filemap data using writepage() address_space_operations.
203 * We don't do block allocation here even for delalloc. We don't
204 * use writepages() because with dealyed allocation we may be doing
205 * block allocation in writepages().
206 */
208 {
215 };
219 }
221 /*
222 * Submit all the data buffers of inode associated with the transaction to
223 * disk.
224 *
225 * We are in a committing transaction. Therefore no new inode can be added to
226 * our inode list. We use JI_COMMIT_RUNNING flag to protect inode we currently
227 * operate on from being released while we write out pages.
228 */
231 {
241 /*
242 * submit the inode data buffers. We use writepage
243 * instead of writepages. Because writepages can do
244 * block allocation with delalloc. We need to write
245 * only allocated blocks here.
246 */
256 }
259 }
261 /*
262 * Wait for data submitted for writeout, refile inodes to proper
263 * transaction if needed.
264 *
265 */
268 {
272 /* For locking, see the comment in journal_submit_data_buffers() */
279 /*
280 * Because AS_EIO is cleared by
281 * filemap_fdatawait_range(), set it again so
282 * that user process can get -EIO from fsync().
283 */
289 }
293 }
295 /* Now refile inode to proper lists */
306 }
307 }
311 }
314 {
317 __u32 checksum;
325 }
329 {
333 }
335 /*
336 * jbd2_journal_commit_transaction
337 *
338 * The primary function for committing a transaction to the log. This
339 * function is called by the journal thread to begin a complete commit.
340 */
342 {
351 ktime_t start_time;
352 u64 commit_time;
365 /*
366 * First job: lock down the current transaction and wait for
367 * all outstanding updates to complete.
368 */
370 #ifdef COMMIT_STATS
374 #endif
376 /* Do we need to erase the effects of a prior jbd2_journal_flush? */
382 }
397 /*
398 * Use plugged writes here, since we want to submit several before
399 * we unplug the device. We don't do explicit unplugging in here,
400 * instead we rely on sync_buffer() doing the unplug for us.
401 */
415 TASK_UNINTERRUPTIBLE);
422 }
424 }
430 /*
431 * First thing we are allowed to do is to discard any remaining
432 * BJ_Reserved buffers. Note, it is _not_ permissible to assume
433 * that there are no such buffers: if a large filesystem
434 * operation like a truncate needs to split itself over multiple
435 * transactions, then it may try to do a jbd2_journal_restart() while
436 * there are still BJ_Reserved buffers outstanding. These must
437 * be released cleanly from the current transaction.
438 *
439 * In this case, the filesystem must still reserve write access
440 * again before modifying the buffer in the new transaction, but
441 * we do not require it to remember exactly which old buffers it
442 * has reserved. This is consistent with the existing behaviour
443 * that multiple jbd2_journal_get_write_access() calls to the same
444 * buffer are perfectly permissable.
445 */
449 /*
450 * A jbd2_journal_get_undo_access()+jbd2_journal_release_buffer() may
451 * leave undo-committed data.
452 */
460 }
462 }
464 /*
465 * Now try to drop any written-back buffers from the journal's
466 * checkpoint lists. We do this *before* commit because it potentially
467 * frees some memory
468 */
475 /*
476 * Switch to a new revoke table.
477 */
495 /*
496 * Now start flushing things to disk, in the order they appear
497 * on the transaction lists. Data blocks go first.
498 */
504 write_op);
508 /*
509 * Way to go: we have now written out all of the data for a
510 * transaction! Now comes the tricky part: we need to write out
511 * metadata. Loop over the transaction's entire buffer list:
512 */
533 /* Find the next buffer to be journaled... */
537 /* If we're in abort mode, we just un-journal the buffer and
538 release it. */
548 /* If that was the last one, we need to clean up
549 * any descriptor buffers which may have been
550 * already allocated, even if we are now
551 * aborting. */
555 }
557 /* Make sure we have a descriptor block in which to
558 record the metadata buffer. */
571 }
588 /* Record it so that we can wait for IO
589 completion later */
592 BJ_LogCtl);
593 }
595 /* Where is the buffer to be written? */
598 /* If the block mapping failed, just abandon the buffer
599 and repeat this loop: we'll fall into the
600 refile-on-abort condition above. */
604 }
606 /*
607 * start_this_handle() uses t_outstanding_credits to determine
608 * the free space in the log, but this counter is changed
609 * by jbd2_journal_next_log_block() also.
610 */
613 /* Bump b_count to prevent truncate from stumbling over
614 the shadowed buffer! @@@ This can go if we ever get
615 rid of the BJ_IO/BJ_Shadow pairing of buffers. */
618 /* Make a temporary IO buffer with which to write it out
619 (this will requeue both the metadata buffer and the
620 temporary IO buffer). new_bh goes on BJ_IO*/
623 /*
624 * akpm: jbd2_journal_write_metadata_buffer() sets
625 * new_bh->b_transaction to commit_transaction.
626 * We need to clean this up before we release new_bh
627 * (which is of type BJ_IO)
628 */
635 }
639 /* Record the new block's tag in the current descriptor
640 buffer */
659 }
661 /* If there's no more to do, or if the descriptor is full,
662 let the IO rip! */
670 /* Write an end-of-descriptor marker before
671 submitting the IOs. "tag" still points to
672 the last tag we set up. */
676 start_journal_io:
679 /*
680 * Compute checksum.
681 */
683 JBD2_FEATURE_COMPAT_CHECKSUM)) {
684 crc32_sum =
686 }
693 }
697 /* Force a new descriptor to be generated next
698 time round the loop. */
701 }
702 }
707 "JBD2: Detected IO errors while flushing file data "
712 }
714 /*
715 * If the journal is not located on the file system device,
716 * then we must flush the file system device before we issue
717 * the commit record
718 */
723 BLKDEV_IFL_WAIT);
725 /* Done it all: now write the commit record asynchronously. */
727 JBD2_FEATURE_INCOMPAT_ASYNC_COMMIT)) {
732 }
734 /* Lo and behold: we have just managed to send a transaction to
735 the log. Before we can commit it, wait for the IO so far to
736 complete. Control buffers being written are on the
737 transaction's t_log_list queue, and metadata buffers are on
738 the t_iobuf_list queue.
740 Wait for the buffers in reverse order. That way we are
741 less likely to be woken up until all IOs have completed, and
742 so we incur less scheduling load.
743 */
747 /*
748 * akpm: these are BJ_IO, and j_list_lock is not needed.
749 * See __journal_try_to_free_buffer.
750 */
751 wait_for_iobuf:
760 }
772 /*
773 * ->t_iobuf_list should contain only dummy buffer_heads
774 * which were created by jbd2_journal_write_metadata_buffer().
775 */
782 /* We also have to unlock and free the corresponding
783 shadowed buffer */
789 /* The metadata is now released for reuse, but we need
790 to remember it against this transaction so that when
791 we finally commit, we can do any checkpointing
792 required. */
795 /* Wake up any transactions which were waiting for this
796 IO to complete */
800 }
806 /* Here we wait for the revoke record and descriptor record buffers */
807 wait_for_ctlbuf:
816 }
828 /* AKPM: bforget here */
829 }
837 JBD2_FEATURE_INCOMPAT_ASYNC_COMMIT)) {
842 }
846 JBD2_FEATURE_INCOMPAT_ASYNC_COMMIT) &&
849 BLKDEV_IFL_WAIT);
850 }
855 /* End of a transaction! Finally, we can do checkpoint
856 processing: any buffers committed as a result of this
857 transaction can be removed from any checkpoint list it was on
858 before. */
869 restart_loop:
870 /*
871 * As there are other places (journal_unmap_buffer()) adding buffers
872 * to this list we have to be careful and hold the j_list_lock.
873 */
885 /*
886 * If there is undo-protected committed data against
887 * this buffer, then we can remove it now. If it is a
888 * buffer needing such protection, the old frozen_data
889 * field now points to a committed version of the
890 * buffer, so rotate that field to the new committed
891 * data.
892 *
893 * Otherwise, we can just throw away the frozen data now.
894 *
895 * We also know that the frozen data has already fired
896 * its triggers if they exist, so we can clear that too.
897 */
905 }
910 }
918 }
920 /* Only re-checkpoint the buffer_head if it is marked
921 * dirty. If the buffer was added to the BJ_Forget list
922 * by jbd2_journal_forget, it may no longer be dirty and
923 * there's no point in keeping a checkpoint record for
924 * it. */
926 /* A buffer which has been freed while still being
927 * journaled by a previous transaction may end up still
928 * being dirty here, but we want to avoid writing back
929 * that buffer in the future after the "add to orphan"
930 * operation been committed, That's not only a performance
931 * gain, it also stops aliasing problems if the buffer is
932 * left behind for writeback and gets reallocated for another
933 * use in a different page. */
937 }
949 /* The buffer on BJ_Forget list and not jbddirty means
950 * it has been freed by this transaction and hence it
951 * could not have been reallocated until this
952 * transaction has committed. *BUT* it could be
953 * reallocated once we have written all the data to
954 * disk and before we process the buffer on BJ_Forget
955 * list. */
960 /* needs a brelse */
965 }
967 }
969 /*
970 * This is a bit sleazy. We use j_list_lock to protect transition
971 * of a transaction into T_FINISHED state and calling
972 * __jbd2_journal_drop_transaction(). Otherwise we could race with
973 * other checkpointing code processing the transaction...
974 */
977 /*
978 * Now recheck if some buffers did not get attached to the transaction
979 * while the lock was dropped...
980 */
985 }
987 /* Done with this transaction! */
997 /*
998 * File the transaction statistics
999 */
1006 /*
1007 * Calculate overall stats
1008 */
1027 /*
1028 * weight the commit time higher than the average time so we don't
1029 * react too strongly to vast changes in the commit time
1030 */
1034 else
1053 commit_transaction;
1055 commit_transaction;
1056 }
1057 }
1070 }