2 * linux/fs/jbd2/checkpoint.c
4 * Written by Stephen C. Tweedie <sct@redhat.com>, 1999
6 * Copyright 1999 Red Hat Software --- 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 * Checkpoint routines for the generic filesystem journaling code.
13 * Part of the ext2fs journaling system.
15 * Checkpointing is the process of ensuring that a section of the log is
16 * committed fully to disk, so that that portion of the log can be
20 #include <linux/time.h>
22 #include <linux/jbd2.h>
23 #include <linux/errno.h>
24 #include <linux/slab.h>
25 #include <linux/blkdev.h>
26 #include <trace/events/jbd2.h>
29 * Unlink a buffer from a transaction checkpoint list.
31 * Called with j_list_lock held.
33 static inline void __buffer_unlink_first(struct journal_head
*jh
)
35 transaction_t
*transaction
= jh
->b_cp_transaction
;
37 jh
->b_cpnext
->b_cpprev
= jh
->b_cpprev
;
38 jh
->b_cpprev
->b_cpnext
= jh
->b_cpnext
;
39 if (transaction
->t_checkpoint_list
== jh
) {
40 transaction
->t_checkpoint_list
= jh
->b_cpnext
;
41 if (transaction
->t_checkpoint_list
== jh
)
42 transaction
->t_checkpoint_list
= NULL
;
47 * Unlink a buffer from a transaction checkpoint(io) list.
49 * Called with j_list_lock held.
51 static inline void __buffer_unlink(struct journal_head
*jh
)
53 transaction_t
*transaction
= jh
->b_cp_transaction
;
55 __buffer_unlink_first(jh
);
56 if (transaction
->t_checkpoint_io_list
== jh
) {
57 transaction
->t_checkpoint_io_list
= jh
->b_cpnext
;
58 if (transaction
->t_checkpoint_io_list
== jh
)
59 transaction
->t_checkpoint_io_list
= NULL
;
64 * Move a buffer from the checkpoint list to the checkpoint io list
66 * Called with j_list_lock held
68 static inline void __buffer_relink_io(struct journal_head
*jh
)
70 transaction_t
*transaction
= jh
->b_cp_transaction
;
72 __buffer_unlink_first(jh
);
74 if (!transaction
->t_checkpoint_io_list
) {
75 jh
->b_cpnext
= jh
->b_cpprev
= jh
;
77 jh
->b_cpnext
= transaction
->t_checkpoint_io_list
;
78 jh
->b_cpprev
= transaction
->t_checkpoint_io_list
->b_cpprev
;
79 jh
->b_cpprev
->b_cpnext
= jh
;
80 jh
->b_cpnext
->b_cpprev
= jh
;
82 transaction
->t_checkpoint_io_list
= jh
;
86 * Try to release a checkpointed buffer from its transaction.
87 * Returns 1 if we released it and 2 if we also released the
90 * Requires j_list_lock
91 * Called under jbd_lock_bh_state(jh2bh(jh)), and drops it
93 static int __try_to_free_cp_buf(struct journal_head
*jh
)
96 struct buffer_head
*bh
= jh2bh(jh
);
98 if (jh
->b_jlist
== BJ_None
&& !buffer_locked(bh
) &&
99 !buffer_dirty(bh
) && !buffer_write_io_error(bh
)) {
100 JBUFFER_TRACE(jh
, "remove from checkpoint list");
101 ret
= __jbd2_journal_remove_checkpoint(jh
) + 1;
102 jbd_unlock_bh_state(bh
);
103 jbd2_journal_remove_journal_head(bh
);
104 BUFFER_TRACE(bh
, "release");
107 jbd_unlock_bh_state(bh
);
113 * __jbd2_log_wait_for_space: wait until there is space in the journal.
115 * Called under j-state_lock *only*. It will be unlocked if we have to wait
116 * for a checkpoint to free up some space in the log.
118 void __jbd2_log_wait_for_space(journal_t
*journal
)
120 int nblocks
, space_left
;
121 assert_spin_locked(&journal
->j_state_lock
);
123 nblocks
= jbd_space_needed(journal
);
124 while (__jbd2_log_space_left(journal
) < nblocks
) {
125 if (journal
->j_flags
& JBD2_ABORT
)
127 spin_unlock(&journal
->j_state_lock
);
128 mutex_lock(&journal
->j_checkpoint_mutex
);
131 * Test again, another process may have checkpointed while we
132 * were waiting for the checkpoint lock. If there are no
133 * transactions ready to be checkpointed, try to recover
134 * journal space by calling cleanup_journal_tail(), and if
135 * that doesn't work, by waiting for the currently committing
136 * transaction to complete. If there is absolutely no way
137 * to make progress, this is either a BUG or corrupted
138 * filesystem, so abort the journal and leave a stack
139 * trace for forensic evidence.
141 spin_lock(&journal
->j_state_lock
);
142 spin_lock(&journal
->j_list_lock
);
143 nblocks
= jbd_space_needed(journal
);
144 space_left
= __jbd2_log_space_left(journal
);
145 if (space_left
< nblocks
) {
146 int chkpt
= journal
->j_checkpoint_transactions
!= NULL
;
149 if (journal
->j_committing_transaction
)
150 tid
= journal
->j_committing_transaction
->t_tid
;
151 spin_unlock(&journal
->j_list_lock
);
152 spin_unlock(&journal
->j_state_lock
);
154 jbd2_log_do_checkpoint(journal
);
155 } else if (jbd2_cleanup_journal_tail(journal
) == 0) {
156 /* We were able to recover space; yay! */
159 jbd2_log_wait_commit(journal
, tid
);
161 printk(KERN_ERR
"%s: needed %d blocks and "
162 "only had %d space available\n",
163 __func__
, nblocks
, space_left
);
164 printk(KERN_ERR
"%s: no way to get more "
165 "journal space in %s\n", __func__
,
168 jbd2_journal_abort(journal
, 0);
170 spin_lock(&journal
->j_state_lock
);
172 spin_unlock(&journal
->j_list_lock
);
174 mutex_unlock(&journal
->j_checkpoint_mutex
);
179 * We were unable to perform jbd_trylock_bh_state() inside j_list_lock.
180 * The caller must restart a list walk. Wait for someone else to run
181 * jbd_unlock_bh_state().
183 static void jbd_sync_bh(journal_t
*journal
, struct buffer_head
*bh
)
184 __releases(journal
->j_list_lock
)
187 spin_unlock(&journal
->j_list_lock
);
188 jbd_lock_bh_state(bh
);
189 jbd_unlock_bh_state(bh
);
194 * Clean up transaction's list of buffers submitted for io.
195 * We wait for any pending IO to complete and remove any clean
196 * buffers. Note that we take the buffers in the opposite ordering
197 * from the one in which they were submitted for IO.
199 * Return 0 on success, and return <0 if some buffers have failed
202 * Called with j_list_lock held.
204 static int __wait_cp_io(journal_t
*journal
, transaction_t
*transaction
)
206 struct journal_head
*jh
;
207 struct buffer_head
*bh
;
212 this_tid
= transaction
->t_tid
;
214 /* Did somebody clean up the transaction in the meanwhile? */
215 if (journal
->j_checkpoint_transactions
!= transaction
||
216 transaction
->t_tid
!= this_tid
)
218 while (!released
&& transaction
->t_checkpoint_io_list
) {
219 jh
= transaction
->t_checkpoint_io_list
;
221 if (!jbd_trylock_bh_state(bh
)) {
222 jbd_sync_bh(journal
, bh
);
223 spin_lock(&journal
->j_list_lock
);
226 if (buffer_locked(bh
)) {
227 atomic_inc(&bh
->b_count
);
228 spin_unlock(&journal
->j_list_lock
);
229 jbd_unlock_bh_state(bh
);
231 /* the journal_head may have gone by now */
232 BUFFER_TRACE(bh
, "brelse");
234 spin_lock(&journal
->j_list_lock
);
237 if (unlikely(buffer_write_io_error(bh
)))
241 * Now in whatever state the buffer currently is, we know that
242 * it has been written out and so we can drop it from the list
244 released
= __jbd2_journal_remove_checkpoint(jh
);
245 jbd_unlock_bh_state(bh
);
246 jbd2_journal_remove_journal_head(bh
);
254 __flush_batch(journal_t
*journal
, int *batch_count
)
258 ll_rw_block(SWRITE
, *batch_count
, journal
->j_chkpt_bhs
);
259 for (i
= 0; i
< *batch_count
; i
++) {
260 struct buffer_head
*bh
= journal
->j_chkpt_bhs
[i
];
261 clear_buffer_jwrite(bh
);
262 BUFFER_TRACE(bh
, "brelse");
269 * Try to flush one buffer from the checkpoint list to disk.
271 * Return 1 if something happened which requires us to abort the current
272 * scan of the checkpoint list. Return <0 if the buffer has failed to
275 * Called with j_list_lock held and drops it if 1 is returned
276 * Called under jbd_lock_bh_state(jh2bh(jh)), and drops it
278 static int __process_buffer(journal_t
*journal
, struct journal_head
*jh
,
279 int *batch_count
, transaction_t
*transaction
)
281 struct buffer_head
*bh
= jh2bh(jh
);
284 if (buffer_locked(bh
)) {
285 atomic_inc(&bh
->b_count
);
286 spin_unlock(&journal
->j_list_lock
);
287 jbd_unlock_bh_state(bh
);
289 /* the journal_head may have gone by now */
290 BUFFER_TRACE(bh
, "brelse");
293 } else if (jh
->b_transaction
!= NULL
) {
294 transaction_t
*t
= jh
->b_transaction
;
295 tid_t tid
= t
->t_tid
;
297 transaction
->t_chp_stats
.cs_forced_to_close
++;
298 spin_unlock(&journal
->j_list_lock
);
299 jbd_unlock_bh_state(bh
);
300 jbd2_log_start_commit(journal
, tid
);
301 jbd2_log_wait_commit(journal
, tid
);
303 } else if (!buffer_dirty(bh
)) {
305 if (unlikely(buffer_write_io_error(bh
)))
307 J_ASSERT_JH(jh
, !buffer_jbddirty(bh
));
308 BUFFER_TRACE(bh
, "remove from checkpoint");
309 __jbd2_journal_remove_checkpoint(jh
);
310 spin_unlock(&journal
->j_list_lock
);
311 jbd_unlock_bh_state(bh
);
312 jbd2_journal_remove_journal_head(bh
);
316 * Important: we are about to write the buffer, and
317 * possibly block, while still holding the journal lock.
318 * We cannot afford to let the transaction logic start
319 * messing around with this buffer before we write it to
320 * disk, as that would break recoverability.
322 BUFFER_TRACE(bh
, "queue");
324 J_ASSERT_BH(bh
, !buffer_jwrite(bh
));
325 set_buffer_jwrite(bh
);
326 journal
->j_chkpt_bhs
[*batch_count
] = bh
;
327 __buffer_relink_io(jh
);
328 jbd_unlock_bh_state(bh
);
329 transaction
->t_chp_stats
.cs_written
++;
331 if (*batch_count
== JBD2_NR_BATCH
) {
332 spin_unlock(&journal
->j_list_lock
);
333 __flush_batch(journal
, batch_count
);
341 * Perform an actual checkpoint. We take the first transaction on the
342 * list of transactions to be checkpointed and send all its buffers
343 * to disk. We submit larger chunks of data at once.
345 * The journal should be locked before calling this function.
346 * Called with j_checkpoint_mutex held.
348 int jbd2_log_do_checkpoint(journal_t
*journal
)
350 transaction_t
*transaction
;
354 jbd_debug(1, "Start checkpoint\n");
357 * First thing: if there are any transactions in the log which
358 * don't need checkpointing, just eliminate them from the
359 * journal straight away.
361 result
= jbd2_cleanup_journal_tail(journal
);
362 trace_jbd2_checkpoint(journal
, result
);
363 jbd_debug(1, "cleanup_journal_tail returned %d\n", result
);
368 * OK, we need to start writing disk blocks. Take one transaction
372 spin_lock(&journal
->j_list_lock
);
373 if (!journal
->j_checkpoint_transactions
)
375 transaction
= journal
->j_checkpoint_transactions
;
376 if (transaction
->t_chp_stats
.cs_chp_time
== 0)
377 transaction
->t_chp_stats
.cs_chp_time
= jiffies
;
378 this_tid
= transaction
->t_tid
;
381 * If someone cleaned up this transaction while we slept, we're
382 * done (maybe it's a new transaction, but it fell at the same
385 if (journal
->j_checkpoint_transactions
== transaction
&&
386 transaction
->t_tid
== this_tid
) {
388 struct journal_head
*jh
;
391 while (!retry
&& transaction
->t_checkpoint_list
) {
392 struct buffer_head
*bh
;
394 jh
= transaction
->t_checkpoint_list
;
396 if (!jbd_trylock_bh_state(bh
)) {
397 jbd_sync_bh(journal
, bh
);
401 retry
= __process_buffer(journal
, jh
, &batch_count
,
403 if (retry
< 0 && !result
)
405 if (!retry
&& (need_resched() ||
406 spin_needbreak(&journal
->j_list_lock
))) {
407 spin_unlock(&journal
->j_list_lock
);
415 spin_unlock(&journal
->j_list_lock
);
418 __flush_batch(journal
, &batch_count
);
422 spin_lock(&journal
->j_list_lock
);
426 * Now we have cleaned up the first transaction's checkpoint
427 * list. Let's clean up the second one
429 err
= __wait_cp_io(journal
, transaction
);
434 spin_unlock(&journal
->j_list_lock
);
436 jbd2_journal_abort(journal
, result
);
438 result
= jbd2_cleanup_journal_tail(journal
);
440 return (result
< 0) ? result
: 0;
444 * Check the list of checkpoint transactions for the journal to see if
445 * we have already got rid of any since the last update of the log tail
446 * in the journal superblock. If so, we can instantly roll the
447 * superblock forward to remove those transactions from the log.
449 * Return <0 on error, 0 on success, 1 if there was nothing to clean up.
451 * Called with the journal lock held.
453 * This is the only part of the journaling code which really needs to be
454 * aware of transaction aborts. Checkpointing involves writing to the
455 * main filesystem area rather than to the journal, so it can proceed
456 * even in abort state, but we must not update the super block if
457 * checkpointing may have failed. Otherwise, we would lose some metadata
458 * buffers which should be written-back to the filesystem.
461 int jbd2_cleanup_journal_tail(journal_t
*journal
)
463 transaction_t
* transaction
;
465 unsigned long blocknr
, freed
;
467 if (is_journal_aborted(journal
))
470 /* OK, work out the oldest transaction remaining in the log, and
471 * the log block it starts at.
473 * If the log is now empty, we need to work out which is the
474 * next transaction ID we will write, and where it will
477 spin_lock(&journal
->j_state_lock
);
478 spin_lock(&journal
->j_list_lock
);
479 transaction
= journal
->j_checkpoint_transactions
;
481 first_tid
= transaction
->t_tid
;
482 blocknr
= transaction
->t_log_start
;
483 } else if ((transaction
= journal
->j_committing_transaction
) != NULL
) {
484 first_tid
= transaction
->t_tid
;
485 blocknr
= transaction
->t_log_start
;
486 } else if ((transaction
= journal
->j_running_transaction
) != NULL
) {
487 first_tid
= transaction
->t_tid
;
488 blocknr
= journal
->j_head
;
490 first_tid
= journal
->j_transaction_sequence
;
491 blocknr
= journal
->j_head
;
493 spin_unlock(&journal
->j_list_lock
);
494 J_ASSERT(blocknr
!= 0);
496 /* If the oldest pinned transaction is at the tail of the log
497 already then there's not much we can do right now. */
498 if (journal
->j_tail_sequence
== first_tid
) {
499 spin_unlock(&journal
->j_state_lock
);
503 /* OK, update the superblock to recover the freed space.
504 * Physical blocks come first: have we wrapped beyond the end of
506 freed
= blocknr
- journal
->j_tail
;
507 if (blocknr
< journal
->j_tail
)
508 freed
= freed
+ journal
->j_last
- journal
->j_first
;
510 trace_jbd2_cleanup_journal_tail(journal
, first_tid
, blocknr
, freed
);
512 "Cleaning journal tail from %d to %d (offset %lu), "
514 journal
->j_tail_sequence
, first_tid
, blocknr
, freed
);
516 journal
->j_free
+= freed
;
517 journal
->j_tail_sequence
= first_tid
;
518 journal
->j_tail
= blocknr
;
519 spin_unlock(&journal
->j_state_lock
);
522 * If there is an external journal, we need to make sure that
523 * any data blocks that were recently written out --- perhaps
524 * by jbd2_log_do_checkpoint() --- are flushed out before we
525 * drop the transactions from the external journal. It's
526 * unlikely this will be necessary, especially with a
527 * appropriately sized journal, but we need this to guarantee
528 * correctness. Fortunately jbd2_cleanup_journal_tail()
529 * doesn't get called all that often.
531 if ((journal
->j_fs_dev
!= journal
->j_dev
) &&
532 (journal
->j_flags
& JBD2_BARRIER
))
533 blkdev_issue_flush(journal
->j_fs_dev
, GFP_KERNEL
, NULL
,
535 if (!(journal
->j_flags
& JBD2_ABORT
))
536 jbd2_journal_update_superblock(journal
, 1);
541 /* Checkpoint list management */
544 * journal_clean_one_cp_list
546 * Find all the written-back checkpoint buffers in the given list and release them.
548 * Called with the journal locked.
549 * Called with j_list_lock held.
550 * Returns number of bufers reaped (for debug)
553 static int journal_clean_one_cp_list(struct journal_head
*jh
, int *released
)
555 struct journal_head
*last_jh
;
556 struct journal_head
*next_jh
= jh
;
563 last_jh
= jh
->b_cpprev
;
566 next_jh
= jh
->b_cpnext
;
567 /* Use trylock because of the ranking */
568 if (jbd_trylock_bh_state(jh2bh(jh
))) {
569 ret
= __try_to_free_cp_buf(jh
);
579 * This function only frees up some memory
580 * if possible so we dont have an obligation
581 * to finish processing. Bail out if preemption
586 } while (jh
!= last_jh
);
592 * journal_clean_checkpoint_list
594 * Find all the written-back checkpoint buffers in the journal and release them.
596 * Called with the journal locked.
597 * Called with j_list_lock held.
598 * Returns number of buffers reaped (for debug)
601 int __jbd2_journal_clean_checkpoint_list(journal_t
*journal
)
603 transaction_t
*transaction
, *last_transaction
, *next_transaction
;
607 transaction
= journal
->j_checkpoint_transactions
;
611 last_transaction
= transaction
->t_cpprev
;
612 next_transaction
= transaction
;
614 transaction
= next_transaction
;
615 next_transaction
= transaction
->t_cpnext
;
616 ret
+= journal_clean_one_cp_list(transaction
->
617 t_checkpoint_list
, &released
);
619 * This function only frees up some memory if possible so we
620 * dont have an obligation to finish processing. Bail out if
621 * preemption requested:
628 * It is essential that we are as careful as in the case of
629 * t_checkpoint_list with removing the buffer from the list as
630 * we can possibly see not yet submitted buffers on io_list
632 ret
+= journal_clean_one_cp_list(transaction
->
633 t_checkpoint_io_list
, &released
);
636 } while (transaction
!= last_transaction
);
642 * journal_remove_checkpoint: called after a buffer has been committed
643 * to disk (either by being write-back flushed to disk, or being
644 * committed to the log).
646 * We cannot safely clean a transaction out of the log until all of the
647 * buffer updates committed in that transaction have safely been stored
648 * elsewhere on disk. To achieve this, all of the buffers in a
649 * transaction need to be maintained on the transaction's checkpoint
650 * lists until they have been rewritten, at which point this function is
651 * called to remove the buffer from the existing transaction's
654 * The function returns 1 if it frees the transaction, 0 otherwise.
656 * This function is called with the journal locked.
657 * This function is called with j_list_lock held.
658 * This function is called with jbd_lock_bh_state(jh2bh(jh))
661 int __jbd2_journal_remove_checkpoint(struct journal_head
*jh
)
663 struct transaction_chp_stats_s
*stats
;
664 transaction_t
*transaction
;
668 JBUFFER_TRACE(jh
, "entry");
670 if ((transaction
= jh
->b_cp_transaction
) == NULL
) {
671 JBUFFER_TRACE(jh
, "not on transaction");
674 journal
= transaction
->t_journal
;
677 jh
->b_cp_transaction
= NULL
;
679 if (transaction
->t_checkpoint_list
!= NULL
||
680 transaction
->t_checkpoint_io_list
!= NULL
)
682 JBUFFER_TRACE(jh
, "transaction has no more buffers");
685 * There is one special case to worry about: if we have just pulled the
686 * buffer off a running or committing transaction's checkpoing list,
687 * then even if the checkpoint list is empty, the transaction obviously
690 * The locking here around t_state is a bit sleazy.
691 * See the comment at the end of jbd2_journal_commit_transaction().
693 if (transaction
->t_state
!= T_FINISHED
) {
694 JBUFFER_TRACE(jh
, "belongs to running/committing transaction");
698 /* OK, that was the last buffer for the transaction: we can now
699 safely remove this transaction from the log */
700 stats
= &transaction
->t_chp_stats
;
701 if (stats
->cs_chp_time
)
702 stats
->cs_chp_time
= jbd2_time_diff(stats
->cs_chp_time
,
704 trace_jbd2_checkpoint_stats(journal
->j_fs_dev
->bd_dev
,
705 transaction
->t_tid
, stats
);
707 __jbd2_journal_drop_transaction(journal
, transaction
);
710 /* Just in case anybody was waiting for more transactions to be
712 wake_up(&journal
->j_wait_logspace
);
715 JBUFFER_TRACE(jh
, "exit");
720 * journal_insert_checkpoint: put a committed buffer onto a checkpoint
721 * list so that we know when it is safe to clean the transaction out of
724 * Called with the journal locked.
725 * Called with j_list_lock held.
727 void __jbd2_journal_insert_checkpoint(struct journal_head
*jh
,
728 transaction_t
*transaction
)
730 JBUFFER_TRACE(jh
, "entry");
731 J_ASSERT_JH(jh
, buffer_dirty(jh2bh(jh
)) || buffer_jbddirty(jh2bh(jh
)));
732 J_ASSERT_JH(jh
, jh
->b_cp_transaction
== NULL
);
734 jh
->b_cp_transaction
= transaction
;
736 if (!transaction
->t_checkpoint_list
) {
737 jh
->b_cpnext
= jh
->b_cpprev
= jh
;
739 jh
->b_cpnext
= transaction
->t_checkpoint_list
;
740 jh
->b_cpprev
= transaction
->t_checkpoint_list
->b_cpprev
;
741 jh
->b_cpprev
->b_cpnext
= jh
;
742 jh
->b_cpnext
->b_cpprev
= jh
;
744 transaction
->t_checkpoint_list
= jh
;
748 * We've finished with this transaction structure: adios...
750 * The transaction must have no links except for the checkpoint by this
753 * Called with the journal locked.
754 * Called with j_list_lock held.
757 void __jbd2_journal_drop_transaction(journal_t
*journal
, transaction_t
*transaction
)
759 assert_spin_locked(&journal
->j_list_lock
);
760 if (transaction
->t_cpnext
) {
761 transaction
->t_cpnext
->t_cpprev
= transaction
->t_cpprev
;
762 transaction
->t_cpprev
->t_cpnext
= transaction
->t_cpnext
;
763 if (journal
->j_checkpoint_transactions
== transaction
)
764 journal
->j_checkpoint_transactions
=
765 transaction
->t_cpnext
;
766 if (journal
->j_checkpoint_transactions
== transaction
)
767 journal
->j_checkpoint_transactions
= NULL
;
770 J_ASSERT(transaction
->t_state
== T_FINISHED
);
771 J_ASSERT(transaction
->t_buffers
== NULL
);
772 J_ASSERT(transaction
->t_forget
== NULL
);
773 J_ASSERT(transaction
->t_iobuf_list
== NULL
);
774 J_ASSERT(transaction
->t_shadow_list
== NULL
);
775 J_ASSERT(transaction
->t_log_list
== NULL
);
776 J_ASSERT(transaction
->t_checkpoint_list
== NULL
);
777 J_ASSERT(transaction
->t_checkpoint_io_list
== NULL
);
778 J_ASSERT(transaction
->t_updates
== 0);
779 J_ASSERT(journal
->j_committing_transaction
!= transaction
);
780 J_ASSERT(journal
->j_running_transaction
!= transaction
);
782 jbd_debug(1, "Dropping transaction %d, all done\n", transaction
->t_tid
);