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/marker.h>
24 #include <linux/errno.h>
25 #include <linux/slab.h>
28 * Unlink a buffer from a transaction checkpoint list.
30 * Called with j_list_lock held.
32 static inline void __buffer_unlink_first(struct journal_head
*jh
)
34 transaction_t
*transaction
= jh
->b_cp_transaction
;
36 jh
->b_cpnext
->b_cpprev
= jh
->b_cpprev
;
37 jh
->b_cpprev
->b_cpnext
= jh
->b_cpnext
;
38 if (transaction
->t_checkpoint_list
== jh
) {
39 transaction
->t_checkpoint_list
= jh
->b_cpnext
;
40 if (transaction
->t_checkpoint_list
== jh
)
41 transaction
->t_checkpoint_list
= NULL
;
46 * Unlink a buffer from a transaction checkpoint(io) list.
48 * Called with j_list_lock held.
50 static inline void __buffer_unlink(struct journal_head
*jh
)
52 transaction_t
*transaction
= jh
->b_cp_transaction
;
54 __buffer_unlink_first(jh
);
55 if (transaction
->t_checkpoint_io_list
== jh
) {
56 transaction
->t_checkpoint_io_list
= jh
->b_cpnext
;
57 if (transaction
->t_checkpoint_io_list
== jh
)
58 transaction
->t_checkpoint_io_list
= NULL
;
63 * Move a buffer from the checkpoint list to the checkpoint io list
65 * Called with j_list_lock held
67 static inline void __buffer_relink_io(struct journal_head
*jh
)
69 transaction_t
*transaction
= jh
->b_cp_transaction
;
71 __buffer_unlink_first(jh
);
73 if (!transaction
->t_checkpoint_io_list
) {
74 jh
->b_cpnext
= jh
->b_cpprev
= jh
;
76 jh
->b_cpnext
= transaction
->t_checkpoint_io_list
;
77 jh
->b_cpprev
= transaction
->t_checkpoint_io_list
->b_cpprev
;
78 jh
->b_cpprev
->b_cpnext
= jh
;
79 jh
->b_cpnext
->b_cpprev
= jh
;
81 transaction
->t_checkpoint_io_list
= jh
;
85 * Try to release a checkpointed buffer from its transaction.
86 * Returns 1 if we released it and 2 if we also released the
89 * Requires j_list_lock
90 * Called under jbd_lock_bh_state(jh2bh(jh)), and drops it
92 static int __try_to_free_cp_buf(struct journal_head
*jh
)
95 struct buffer_head
*bh
= jh2bh(jh
);
97 if (jh
->b_jlist
== BJ_None
&& !buffer_locked(bh
) && !buffer_dirty(bh
)) {
98 JBUFFER_TRACE(jh
, "remove from checkpoint list");
99 ret
= __jbd2_journal_remove_checkpoint(jh
) + 1;
100 jbd_unlock_bh_state(bh
);
101 jbd2_journal_remove_journal_head(bh
);
102 BUFFER_TRACE(bh
, "release");
105 jbd_unlock_bh_state(bh
);
111 * __jbd2_log_wait_for_space: wait until there is space in the journal.
113 * Called under j-state_lock *only*. It will be unlocked if we have to wait
114 * for a checkpoint to free up some space in the log.
116 void __jbd2_log_wait_for_space(journal_t
*journal
)
119 assert_spin_locked(&journal
->j_state_lock
);
121 nblocks
= jbd_space_needed(journal
);
122 while (__jbd2_log_space_left(journal
) < nblocks
) {
123 if (journal
->j_flags
& JBD2_ABORT
)
125 spin_unlock(&journal
->j_state_lock
);
126 mutex_lock(&journal
->j_checkpoint_mutex
);
129 * Test again, another process may have checkpointed while we
130 * were waiting for the checkpoint lock. If there are no
131 * outstanding transactions there is nothing to checkpoint and
132 * we can't make progress. Abort the journal in this case.
134 spin_lock(&journal
->j_state_lock
);
135 spin_lock(&journal
->j_list_lock
);
136 nblocks
= jbd_space_needed(journal
);
137 if (__jbd2_log_space_left(journal
) < nblocks
) {
138 int chkpt
= journal
->j_checkpoint_transactions
!= NULL
;
140 spin_unlock(&journal
->j_list_lock
);
141 spin_unlock(&journal
->j_state_lock
);
143 jbd2_log_do_checkpoint(journal
);
145 printk(KERN_ERR
"%s: no transactions\n",
147 jbd2_journal_abort(journal
, 0);
150 spin_lock(&journal
->j_state_lock
);
152 spin_unlock(&journal
->j_list_lock
);
154 mutex_unlock(&journal
->j_checkpoint_mutex
);
159 * We were unable to perform jbd_trylock_bh_state() inside j_list_lock.
160 * The caller must restart a list walk. Wait for someone else to run
161 * jbd_unlock_bh_state().
163 static void jbd_sync_bh(journal_t
*journal
, struct buffer_head
*bh
)
164 __releases(journal
->j_list_lock
)
167 spin_unlock(&journal
->j_list_lock
);
168 jbd_lock_bh_state(bh
);
169 jbd_unlock_bh_state(bh
);
174 * Clean up transaction's list of buffers submitted for io.
175 * We wait for any pending IO to complete and remove any clean
176 * buffers. Note that we take the buffers in the opposite ordering
177 * from the one in which they were submitted for IO.
179 * Called with j_list_lock held.
181 static void __wait_cp_io(journal_t
*journal
, transaction_t
*transaction
)
183 struct journal_head
*jh
;
184 struct buffer_head
*bh
;
188 this_tid
= transaction
->t_tid
;
190 /* Did somebody clean up the transaction in the meanwhile? */
191 if (journal
->j_checkpoint_transactions
!= transaction
||
192 transaction
->t_tid
!= this_tid
)
194 while (!released
&& transaction
->t_checkpoint_io_list
) {
195 jh
= transaction
->t_checkpoint_io_list
;
197 if (!jbd_trylock_bh_state(bh
)) {
198 jbd_sync_bh(journal
, bh
);
199 spin_lock(&journal
->j_list_lock
);
202 if (buffer_locked(bh
)) {
203 atomic_inc(&bh
->b_count
);
204 spin_unlock(&journal
->j_list_lock
);
205 jbd_unlock_bh_state(bh
);
207 /* the journal_head may have gone by now */
208 BUFFER_TRACE(bh
, "brelse");
210 spin_lock(&journal
->j_list_lock
);
214 * Now in whatever state the buffer currently is, we know that
215 * it has been written out and so we can drop it from the list
217 released
= __jbd2_journal_remove_checkpoint(jh
);
218 jbd_unlock_bh_state(bh
);
219 jbd2_journal_remove_journal_head(bh
);
227 __flush_batch(journal_t
*journal
, struct buffer_head
**bhs
, int *batch_count
)
231 ll_rw_block(SWRITE
, *batch_count
, bhs
);
232 for (i
= 0; i
< *batch_count
; i
++) {
233 struct buffer_head
*bh
= bhs
[i
];
234 clear_buffer_jwrite(bh
);
235 BUFFER_TRACE(bh
, "brelse");
242 * Try to flush one buffer from the checkpoint list to disk.
244 * Return 1 if something happened which requires us to abort the current
245 * scan of the checkpoint list.
247 * Called with j_list_lock held and drops it if 1 is returned
248 * Called under jbd_lock_bh_state(jh2bh(jh)), and drops it
250 static int __process_buffer(journal_t
*journal
, struct journal_head
*jh
,
251 struct buffer_head
**bhs
, int *batch_count
,
252 transaction_t
*transaction
)
254 struct buffer_head
*bh
= jh2bh(jh
);
257 if (buffer_locked(bh
)) {
258 atomic_inc(&bh
->b_count
);
259 spin_unlock(&journal
->j_list_lock
);
260 jbd_unlock_bh_state(bh
);
262 /* the journal_head may have gone by now */
263 BUFFER_TRACE(bh
, "brelse");
266 } else if (jh
->b_transaction
!= NULL
) {
267 transaction_t
*t
= jh
->b_transaction
;
268 tid_t tid
= t
->t_tid
;
270 transaction
->t_chp_stats
.cs_forced_to_close
++;
271 spin_unlock(&journal
->j_list_lock
);
272 jbd_unlock_bh_state(bh
);
273 jbd2_log_start_commit(journal
, tid
);
274 jbd2_log_wait_commit(journal
, tid
);
276 } else if (!buffer_dirty(bh
)) {
277 J_ASSERT_JH(jh
, !buffer_jbddirty(bh
));
278 BUFFER_TRACE(bh
, "remove from checkpoint");
279 __jbd2_journal_remove_checkpoint(jh
);
280 spin_unlock(&journal
->j_list_lock
);
281 jbd_unlock_bh_state(bh
);
282 jbd2_journal_remove_journal_head(bh
);
287 * Important: we are about to write the buffer, and
288 * possibly block, while still holding the journal lock.
289 * We cannot afford to let the transaction logic start
290 * messing around with this buffer before we write it to
291 * disk, as that would break recoverability.
293 BUFFER_TRACE(bh
, "queue");
295 J_ASSERT_BH(bh
, !buffer_jwrite(bh
));
296 set_buffer_jwrite(bh
);
297 bhs
[*batch_count
] = bh
;
298 __buffer_relink_io(jh
);
299 jbd_unlock_bh_state(bh
);
300 transaction
->t_chp_stats
.cs_written
++;
302 if (*batch_count
== NR_BATCH
) {
303 spin_unlock(&journal
->j_list_lock
);
304 __flush_batch(journal
, bhs
, batch_count
);
312 * Perform an actual checkpoint. We take the first transaction on the
313 * list of transactions to be checkpointed and send all its buffers
314 * to disk. We submit larger chunks of data at once.
316 * The journal should be locked before calling this function.
318 int jbd2_log_do_checkpoint(journal_t
*journal
)
320 transaction_t
*transaction
;
324 jbd_debug(1, "Start checkpoint\n");
327 * First thing: if there are any transactions in the log which
328 * don't need checkpointing, just eliminate them from the
329 * journal straight away.
331 result
= jbd2_cleanup_journal_tail(journal
);
332 trace_mark(jbd2_checkpoint
, "dev %s need_checkpoint %d",
333 journal
->j_devname
, result
);
334 jbd_debug(1, "cleanup_journal_tail returned %d\n", result
);
339 * OK, we need to start writing disk blocks. Take one transaction
342 spin_lock(&journal
->j_list_lock
);
343 if (!journal
->j_checkpoint_transactions
)
345 transaction
= journal
->j_checkpoint_transactions
;
346 if (transaction
->t_chp_stats
.cs_chp_time
== 0)
347 transaction
->t_chp_stats
.cs_chp_time
= jiffies
;
348 this_tid
= transaction
->t_tid
;
351 * If someone cleaned up this transaction while we slept, we're
352 * done (maybe it's a new transaction, but it fell at the same
355 if (journal
->j_checkpoint_transactions
== transaction
&&
356 transaction
->t_tid
== this_tid
) {
358 struct buffer_head
*bhs
[NR_BATCH
];
359 struct journal_head
*jh
;
362 while (!retry
&& transaction
->t_checkpoint_list
) {
363 struct buffer_head
*bh
;
365 jh
= transaction
->t_checkpoint_list
;
367 if (!jbd_trylock_bh_state(bh
)) {
368 jbd_sync_bh(journal
, bh
);
372 retry
= __process_buffer(journal
, jh
, bhs
, &batch_count
,
374 if (!retry
&& (need_resched() ||
375 spin_needbreak(&journal
->j_list_lock
))) {
376 spin_unlock(&journal
->j_list_lock
);
384 spin_unlock(&journal
->j_list_lock
);
387 __flush_batch(journal
, bhs
, &batch_count
);
391 spin_lock(&journal
->j_list_lock
);
395 * Now we have cleaned up the first transaction's checkpoint
396 * list. Let's clean up the second one
398 __wait_cp_io(journal
, transaction
);
401 spin_unlock(&journal
->j_list_lock
);
402 result
= jbd2_cleanup_journal_tail(journal
);
409 * Check the list of checkpoint transactions for the journal to see if
410 * we have already got rid of any since the last update of the log tail
411 * in the journal superblock. If so, we can instantly roll the
412 * superblock forward to remove those transactions from the log.
414 * Return <0 on error, 0 on success, 1 if there was nothing to clean up.
416 * Called with the journal lock held.
418 * This is the only part of the journaling code which really needs to be
419 * aware of transaction aborts. Checkpointing involves writing to the
420 * main filesystem area rather than to the journal, so it can proceed
421 * even in abort state, but we must not update the journal superblock if
422 * we have an abort error outstanding.
425 int jbd2_cleanup_journal_tail(journal_t
*journal
)
427 transaction_t
* transaction
;
429 unsigned long blocknr
, freed
;
431 /* OK, work out the oldest transaction remaining in the log, and
432 * the log block it starts at.
434 * If the log is now empty, we need to work out which is the
435 * next transaction ID we will write, and where it will
438 spin_lock(&journal
->j_state_lock
);
439 spin_lock(&journal
->j_list_lock
);
440 transaction
= journal
->j_checkpoint_transactions
;
442 first_tid
= transaction
->t_tid
;
443 blocknr
= transaction
->t_log_start
;
444 } else if ((transaction
= journal
->j_committing_transaction
) != NULL
) {
445 first_tid
= transaction
->t_tid
;
446 blocknr
= transaction
->t_log_start
;
447 } else if ((transaction
= journal
->j_running_transaction
) != NULL
) {
448 first_tid
= transaction
->t_tid
;
449 blocknr
= journal
->j_head
;
451 first_tid
= journal
->j_transaction_sequence
;
452 blocknr
= journal
->j_head
;
454 spin_unlock(&journal
->j_list_lock
);
455 J_ASSERT(blocknr
!= 0);
457 /* If the oldest pinned transaction is at the tail of the log
458 already then there's not much we can do right now. */
459 if (journal
->j_tail_sequence
== first_tid
) {
460 spin_unlock(&journal
->j_state_lock
);
464 /* OK, update the superblock to recover the freed space.
465 * Physical blocks come first: have we wrapped beyond the end of
467 freed
= blocknr
- journal
->j_tail
;
468 if (blocknr
< journal
->j_tail
)
469 freed
= freed
+ journal
->j_last
- journal
->j_first
;
472 "Cleaning journal tail from %d to %d (offset %lu), "
474 journal
->j_tail_sequence
, first_tid
, blocknr
, freed
);
476 journal
->j_free
+= freed
;
477 journal
->j_tail_sequence
= first_tid
;
478 journal
->j_tail
= blocknr
;
479 spin_unlock(&journal
->j_state_lock
);
480 if (!(journal
->j_flags
& JBD2_ABORT
))
481 jbd2_journal_update_superblock(journal
, 1);
486 /* Checkpoint list management */
489 * journal_clean_one_cp_list
491 * Find all the written-back checkpoint buffers in the given list and release them.
493 * Called with the journal locked.
494 * Called with j_list_lock held.
495 * Returns number of bufers reaped (for debug)
498 static int journal_clean_one_cp_list(struct journal_head
*jh
, int *released
)
500 struct journal_head
*last_jh
;
501 struct journal_head
*next_jh
= jh
;
508 last_jh
= jh
->b_cpprev
;
511 next_jh
= jh
->b_cpnext
;
512 /* Use trylock because of the ranking */
513 if (jbd_trylock_bh_state(jh2bh(jh
))) {
514 ret
= __try_to_free_cp_buf(jh
);
524 * This function only frees up some memory
525 * if possible so we dont have an obligation
526 * to finish processing. Bail out if preemption
531 } while (jh
!= last_jh
);
537 * journal_clean_checkpoint_list
539 * Find all the written-back checkpoint buffers in the journal and release them.
541 * Called with the journal locked.
542 * Called with j_list_lock held.
543 * Returns number of buffers reaped (for debug)
546 int __jbd2_journal_clean_checkpoint_list(journal_t
*journal
)
548 transaction_t
*transaction
, *last_transaction
, *next_transaction
;
552 transaction
= journal
->j_checkpoint_transactions
;
556 last_transaction
= transaction
->t_cpprev
;
557 next_transaction
= transaction
;
559 transaction
= next_transaction
;
560 next_transaction
= transaction
->t_cpnext
;
561 ret
+= journal_clean_one_cp_list(transaction
->
562 t_checkpoint_list
, &released
);
564 * This function only frees up some memory if possible so we
565 * dont have an obligation to finish processing. Bail out if
566 * preemption requested:
573 * It is essential that we are as careful as in the case of
574 * t_checkpoint_list with removing the buffer from the list as
575 * we can possibly see not yet submitted buffers on io_list
577 ret
+= journal_clean_one_cp_list(transaction
->
578 t_checkpoint_io_list
, &released
);
581 } while (transaction
!= last_transaction
);
587 * journal_remove_checkpoint: called after a buffer has been committed
588 * to disk (either by being write-back flushed to disk, or being
589 * committed to the log).
591 * We cannot safely clean a transaction out of the log until all of the
592 * buffer updates committed in that transaction have safely been stored
593 * elsewhere on disk. To achieve this, all of the buffers in a
594 * transaction need to be maintained on the transaction's checkpoint
595 * lists until they have been rewritten, at which point this function is
596 * called to remove the buffer from the existing transaction's
599 * The function returns 1 if it frees the transaction, 0 otherwise.
601 * This function is called with the journal locked.
602 * This function is called with j_list_lock held.
603 * This function is called with jbd_lock_bh_state(jh2bh(jh))
606 int __jbd2_journal_remove_checkpoint(struct journal_head
*jh
)
608 transaction_t
*transaction
;
612 JBUFFER_TRACE(jh
, "entry");
614 if ((transaction
= jh
->b_cp_transaction
) == NULL
) {
615 JBUFFER_TRACE(jh
, "not on transaction");
618 journal
= transaction
->t_journal
;
621 jh
->b_cp_transaction
= NULL
;
623 if (transaction
->t_checkpoint_list
!= NULL
||
624 transaction
->t_checkpoint_io_list
!= NULL
)
626 JBUFFER_TRACE(jh
, "transaction has no more buffers");
629 * There is one special case to worry about: if we have just pulled the
630 * buffer off a running or committing transaction's checkpoing list,
631 * then even if the checkpoint list is empty, the transaction obviously
634 * The locking here around t_state is a bit sleazy.
635 * See the comment at the end of jbd2_journal_commit_transaction().
637 if (transaction
->t_state
!= T_FINISHED
) {
638 JBUFFER_TRACE(jh
, "belongs to running/committing transaction");
642 /* OK, that was the last buffer for the transaction: we can now
643 safely remove this transaction from the log */
645 __jbd2_journal_drop_transaction(journal
, transaction
);
647 /* Just in case anybody was waiting for more transactions to be
649 wake_up(&journal
->j_wait_logspace
);
652 JBUFFER_TRACE(jh
, "exit");
657 * journal_insert_checkpoint: put a committed buffer onto a checkpoint
658 * list so that we know when it is safe to clean the transaction out of
661 * Called with the journal locked.
662 * Called with j_list_lock held.
664 void __jbd2_journal_insert_checkpoint(struct journal_head
*jh
,
665 transaction_t
*transaction
)
667 JBUFFER_TRACE(jh
, "entry");
668 J_ASSERT_JH(jh
, buffer_dirty(jh2bh(jh
)) || buffer_jbddirty(jh2bh(jh
)));
669 J_ASSERT_JH(jh
, jh
->b_cp_transaction
== NULL
);
671 jh
->b_cp_transaction
= transaction
;
673 if (!transaction
->t_checkpoint_list
) {
674 jh
->b_cpnext
= jh
->b_cpprev
= jh
;
676 jh
->b_cpnext
= transaction
->t_checkpoint_list
;
677 jh
->b_cpprev
= transaction
->t_checkpoint_list
->b_cpprev
;
678 jh
->b_cpprev
->b_cpnext
= jh
;
679 jh
->b_cpnext
->b_cpprev
= jh
;
681 transaction
->t_checkpoint_list
= jh
;
685 * We've finished with this transaction structure: adios...
687 * The transaction must have no links except for the checkpoint by this
690 * Called with the journal locked.
691 * Called with j_list_lock held.
694 void __jbd2_journal_drop_transaction(journal_t
*journal
, transaction_t
*transaction
)
696 assert_spin_locked(&journal
->j_list_lock
);
697 if (transaction
->t_cpnext
) {
698 transaction
->t_cpnext
->t_cpprev
= transaction
->t_cpprev
;
699 transaction
->t_cpprev
->t_cpnext
= transaction
->t_cpnext
;
700 if (journal
->j_checkpoint_transactions
== transaction
)
701 journal
->j_checkpoint_transactions
=
702 transaction
->t_cpnext
;
703 if (journal
->j_checkpoint_transactions
== transaction
)
704 journal
->j_checkpoint_transactions
= NULL
;
707 J_ASSERT(transaction
->t_state
== T_FINISHED
);
708 J_ASSERT(transaction
->t_buffers
== NULL
);
709 J_ASSERT(transaction
->t_forget
== NULL
);
710 J_ASSERT(transaction
->t_iobuf_list
== NULL
);
711 J_ASSERT(transaction
->t_shadow_list
== NULL
);
712 J_ASSERT(transaction
->t_log_list
== NULL
);
713 J_ASSERT(transaction
->t_checkpoint_list
== NULL
);
714 J_ASSERT(transaction
->t_checkpoint_io_list
== NULL
);
715 J_ASSERT(transaction
->t_updates
== 0);
716 J_ASSERT(journal
->j_committing_transaction
!= transaction
);
717 J_ASSERT(journal
->j_running_transaction
!= transaction
);
719 jbd_debug(1, "Dropping transaction %d, all done\n", transaction
->t_tid
);