2 * linux/fs/jbd/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/jbd.h>
23 #include <linux/errno.h>
24 #include <linux/slab.h>
27 * Unlink a buffer from a transaction checkpoint list.
29 * Called with j_list_lock held.
31 static inline void __buffer_unlink_first(struct journal_head
*jh
)
33 transaction_t
*transaction
= jh
->b_cp_transaction
;
35 jh
->b_cpnext
->b_cpprev
= jh
->b_cpprev
;
36 jh
->b_cpprev
->b_cpnext
= jh
->b_cpnext
;
37 if (transaction
->t_checkpoint_list
== jh
) {
38 transaction
->t_checkpoint_list
= jh
->b_cpnext
;
39 if (transaction
->t_checkpoint_list
== jh
)
40 transaction
->t_checkpoint_list
= NULL
;
45 * Unlink a buffer from a transaction checkpoint(io) list.
47 * Called with j_list_lock held.
49 static inline void __buffer_unlink(struct journal_head
*jh
)
51 transaction_t
*transaction
= jh
->b_cp_transaction
;
53 __buffer_unlink_first(jh
);
54 if (transaction
->t_checkpoint_io_list
== jh
) {
55 transaction
->t_checkpoint_io_list
= jh
->b_cpnext
;
56 if (transaction
->t_checkpoint_io_list
== jh
)
57 transaction
->t_checkpoint_io_list
= NULL
;
62 * Move a buffer from the checkpoint list to the checkpoint io list
64 * Called with j_list_lock held
66 static inline void __buffer_relink_io(struct journal_head
*jh
)
68 transaction_t
*transaction
= jh
->b_cp_transaction
;
70 __buffer_unlink_first(jh
);
72 if (!transaction
->t_checkpoint_io_list
) {
73 jh
->b_cpnext
= jh
->b_cpprev
= jh
;
75 jh
->b_cpnext
= transaction
->t_checkpoint_io_list
;
76 jh
->b_cpprev
= transaction
->t_checkpoint_io_list
->b_cpprev
;
77 jh
->b_cpprev
->b_cpnext
= jh
;
78 jh
->b_cpnext
->b_cpprev
= jh
;
80 transaction
->t_checkpoint_io_list
= jh
;
84 * Try to release a checkpointed buffer from its transaction.
85 * Returns 1 if we released it and 2 if we also released the
88 * Requires j_list_lock
89 * Called under jbd_lock_bh_state(jh2bh(jh)), and drops it
91 static int __try_to_free_cp_buf(struct journal_head
*jh
)
94 struct buffer_head
*bh
= jh2bh(jh
);
96 if (jh
->b_jlist
== BJ_None
&& !buffer_locked(bh
) &&
97 !buffer_dirty(bh
) && !buffer_write_io_error(bh
)) {
98 JBUFFER_TRACE(jh
, "remove from checkpoint list");
99 ret
= __journal_remove_checkpoint(jh
) + 1;
100 jbd_unlock_bh_state(bh
);
101 journal_remove_journal_head(bh
);
102 BUFFER_TRACE(bh
, "release");
105 jbd_unlock_bh_state(bh
);
111 * __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 __log_wait_for_space(journal_t
*journal
)
119 assert_spin_locked(&journal
->j_state_lock
);
121 nblocks
= jbd_space_needed(journal
);
122 while (__log_space_left(journal
) < nblocks
) {
123 if (journal
->j_flags
& JFS_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 (__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 log_do_checkpoint(journal
);
145 printk(KERN_ERR
"%s: no transactions\n",
147 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 * Return 0 on success, and return <0 if some buffers have failed
182 * Called with j_list_lock held.
184 static int __wait_cp_io(journal_t
*journal
, transaction_t
*transaction
)
186 struct journal_head
*jh
;
187 struct buffer_head
*bh
;
192 this_tid
= transaction
->t_tid
;
194 /* Did somebody clean up the transaction in the meanwhile? */
195 if (journal
->j_checkpoint_transactions
!= transaction
||
196 transaction
->t_tid
!= this_tid
)
198 while (!released
&& transaction
->t_checkpoint_io_list
) {
199 jh
= transaction
->t_checkpoint_io_list
;
201 if (!jbd_trylock_bh_state(bh
)) {
202 jbd_sync_bh(journal
, bh
);
203 spin_lock(&journal
->j_list_lock
);
206 if (buffer_locked(bh
)) {
207 atomic_inc(&bh
->b_count
);
208 spin_unlock(&journal
->j_list_lock
);
209 jbd_unlock_bh_state(bh
);
211 /* the journal_head may have gone by now */
212 BUFFER_TRACE(bh
, "brelse");
214 spin_lock(&journal
->j_list_lock
);
217 if (unlikely(buffer_write_io_error(bh
)))
221 * Now in whatever state the buffer currently is, we know that
222 * it has been written out and so we can drop it from the list
224 released
= __journal_remove_checkpoint(jh
);
225 jbd_unlock_bh_state(bh
);
226 journal_remove_journal_head(bh
);
236 __flush_batch(journal_t
*journal
, struct buffer_head
**bhs
, int *batch_count
)
240 ll_rw_block(SWRITE
, *batch_count
, bhs
);
241 for (i
= 0; i
< *batch_count
; i
++) {
242 struct buffer_head
*bh
= bhs
[i
];
243 clear_buffer_jwrite(bh
);
244 BUFFER_TRACE(bh
, "brelse");
251 * Try to flush one buffer from the checkpoint list to disk.
253 * Return 1 if something happened which requires us to abort the current
254 * scan of the checkpoint list. Return <0 if the buffer has failed to
257 * Called with j_list_lock held and drops it if 1 is returned
258 * Called under jbd_lock_bh_state(jh2bh(jh)), and drops it
260 static int __process_buffer(journal_t
*journal
, struct journal_head
*jh
,
261 struct buffer_head
**bhs
, int *batch_count
)
263 struct buffer_head
*bh
= jh2bh(jh
);
266 if (buffer_locked(bh
)) {
267 atomic_inc(&bh
->b_count
);
268 spin_unlock(&journal
->j_list_lock
);
269 jbd_unlock_bh_state(bh
);
271 /* the journal_head may have gone by now */
272 BUFFER_TRACE(bh
, "brelse");
275 } else if (jh
->b_transaction
!= NULL
) {
276 transaction_t
*t
= jh
->b_transaction
;
277 tid_t tid
= t
->t_tid
;
279 spin_unlock(&journal
->j_list_lock
);
280 jbd_unlock_bh_state(bh
);
281 log_start_commit(journal
, tid
);
282 log_wait_commit(journal
, tid
);
284 } else if (!buffer_dirty(bh
)) {
286 if (unlikely(buffer_write_io_error(bh
)))
288 J_ASSERT_JH(jh
, !buffer_jbddirty(bh
));
289 BUFFER_TRACE(bh
, "remove from checkpoint");
290 __journal_remove_checkpoint(jh
);
291 spin_unlock(&journal
->j_list_lock
);
292 jbd_unlock_bh_state(bh
);
293 journal_remove_journal_head(bh
);
297 * Important: we are about to write the buffer, and
298 * possibly block, while still holding the journal lock.
299 * We cannot afford to let the transaction logic start
300 * messing around with this buffer before we write it to
301 * disk, as that would break recoverability.
303 BUFFER_TRACE(bh
, "queue");
305 J_ASSERT_BH(bh
, !buffer_jwrite(bh
));
306 set_buffer_jwrite(bh
);
307 bhs
[*batch_count
] = bh
;
308 __buffer_relink_io(jh
);
309 jbd_unlock_bh_state(bh
);
311 if (*batch_count
== NR_BATCH
) {
312 spin_unlock(&journal
->j_list_lock
);
313 __flush_batch(journal
, bhs
, batch_count
);
321 * Perform an actual checkpoint. We take the first transaction on the
322 * list of transactions to be checkpointed and send all its buffers
323 * to disk. We submit larger chunks of data at once.
325 * The journal should be locked before calling this function.
326 * Called with j_checkpoint_mutex held.
328 int log_do_checkpoint(journal_t
*journal
)
330 transaction_t
*transaction
;
334 jbd_debug(1, "Start checkpoint\n");
337 * First thing: if there are any transactions in the log which
338 * don't need checkpointing, just eliminate them from the
339 * journal straight away.
341 result
= cleanup_journal_tail(journal
);
342 jbd_debug(1, "cleanup_journal_tail returned %d\n", result
);
347 * OK, we need to start writing disk blocks. Take one transaction
351 spin_lock(&journal
->j_list_lock
);
352 if (!journal
->j_checkpoint_transactions
)
354 transaction
= journal
->j_checkpoint_transactions
;
355 this_tid
= transaction
->t_tid
;
358 * If someone cleaned up this transaction while we slept, we're
359 * done (maybe it's a new transaction, but it fell at the same
362 if (journal
->j_checkpoint_transactions
== transaction
&&
363 transaction
->t_tid
== this_tid
) {
365 struct buffer_head
*bhs
[NR_BATCH
];
366 struct journal_head
*jh
;
369 while (!retry
&& transaction
->t_checkpoint_list
) {
370 struct buffer_head
*bh
;
372 jh
= transaction
->t_checkpoint_list
;
374 if (!jbd_trylock_bh_state(bh
)) {
375 jbd_sync_bh(journal
, bh
);
379 retry
= __process_buffer(journal
, jh
, bhs
,&batch_count
);
380 if (retry
< 0 && !result
)
382 if (!retry
&& (need_resched() ||
383 spin_needbreak(&journal
->j_list_lock
))) {
384 spin_unlock(&journal
->j_list_lock
);
392 spin_unlock(&journal
->j_list_lock
);
395 __flush_batch(journal
, bhs
, &batch_count
);
399 spin_lock(&journal
->j_list_lock
);
403 * Now we have cleaned up the first transaction's checkpoint
404 * list. Let's clean up the second one
406 err
= __wait_cp_io(journal
, transaction
);
411 spin_unlock(&journal
->j_list_lock
);
413 journal_abort(journal
, result
);
415 result
= cleanup_journal_tail(journal
);
417 return (result
< 0) ? result
: 0;
421 * Check the list of checkpoint transactions for the journal to see if
422 * we have already got rid of any since the last update of the log tail
423 * in the journal superblock. If so, we can instantly roll the
424 * superblock forward to remove those transactions from the log.
426 * Return <0 on error, 0 on success, 1 if there was nothing to clean up.
428 * Called with the journal lock held.
430 * This is the only part of the journaling code which really needs to be
431 * aware of transaction aborts. Checkpointing involves writing to the
432 * main filesystem area rather than to the journal, so it can proceed
433 * even in abort state, but we must not update the super block if
434 * checkpointing may have failed. Otherwise, we would lose some metadata
435 * buffers which should be written-back to the filesystem.
438 int cleanup_journal_tail(journal_t
*journal
)
440 transaction_t
* transaction
;
442 unsigned long blocknr
, freed
;
444 if (is_journal_aborted(journal
))
447 /* OK, work out the oldest transaction remaining in the log, and
448 * the log block it starts at.
450 * If the log is now empty, we need to work out which is the
451 * next transaction ID we will write, and where it will
454 spin_lock(&journal
->j_state_lock
);
455 spin_lock(&journal
->j_list_lock
);
456 transaction
= journal
->j_checkpoint_transactions
;
458 first_tid
= transaction
->t_tid
;
459 blocknr
= transaction
->t_log_start
;
460 } else if ((transaction
= journal
->j_committing_transaction
) != NULL
) {
461 first_tid
= transaction
->t_tid
;
462 blocknr
= transaction
->t_log_start
;
463 } else if ((transaction
= journal
->j_running_transaction
) != NULL
) {
464 first_tid
= transaction
->t_tid
;
465 blocknr
= journal
->j_head
;
467 first_tid
= journal
->j_transaction_sequence
;
468 blocknr
= journal
->j_head
;
470 spin_unlock(&journal
->j_list_lock
);
471 J_ASSERT(blocknr
!= 0);
473 /* If the oldest pinned transaction is at the tail of the log
474 already then there's not much we can do right now. */
475 if (journal
->j_tail_sequence
== first_tid
) {
476 spin_unlock(&journal
->j_state_lock
);
480 /* OK, update the superblock to recover the freed space.
481 * Physical blocks come first: have we wrapped beyond the end of
483 freed
= blocknr
- journal
->j_tail
;
484 if (blocknr
< journal
->j_tail
)
485 freed
= freed
+ journal
->j_last
- journal
->j_first
;
488 "Cleaning journal tail from %d to %d (offset %lu), "
490 journal
->j_tail_sequence
, first_tid
, blocknr
, freed
);
492 journal
->j_free
+= freed
;
493 journal
->j_tail_sequence
= first_tid
;
494 journal
->j_tail
= blocknr
;
495 spin_unlock(&journal
->j_state_lock
);
496 if (!(journal
->j_flags
& JFS_ABORT
))
497 journal_update_superblock(journal
, 1);
502 /* Checkpoint list management */
505 * journal_clean_one_cp_list
507 * Find all the written-back checkpoint buffers in the given list and release them.
509 * Called with the journal locked.
510 * Called with j_list_lock held.
511 * Returns number of bufers reaped (for debug)
514 static int journal_clean_one_cp_list(struct journal_head
*jh
, int *released
)
516 struct journal_head
*last_jh
;
517 struct journal_head
*next_jh
= jh
;
524 last_jh
= jh
->b_cpprev
;
527 next_jh
= jh
->b_cpnext
;
528 /* Use trylock because of the ranking */
529 if (jbd_trylock_bh_state(jh2bh(jh
))) {
530 ret
= __try_to_free_cp_buf(jh
);
540 * This function only frees up some memory
541 * if possible so we dont have an obligation
542 * to finish processing. Bail out if preemption
547 } while (jh
!= last_jh
);
553 * journal_clean_checkpoint_list
555 * Find all the written-back checkpoint buffers in the journal and release them.
557 * Called with the journal locked.
558 * Called with j_list_lock held.
559 * Returns number of buffers reaped (for debug)
562 int __journal_clean_checkpoint_list(journal_t
*journal
)
564 transaction_t
*transaction
, *last_transaction
, *next_transaction
;
568 transaction
= journal
->j_checkpoint_transactions
;
572 last_transaction
= transaction
->t_cpprev
;
573 next_transaction
= transaction
;
575 transaction
= next_transaction
;
576 next_transaction
= transaction
->t_cpnext
;
577 ret
+= journal_clean_one_cp_list(transaction
->
578 t_checkpoint_list
, &released
);
580 * This function only frees up some memory if possible so we
581 * dont have an obligation to finish processing. Bail out if
582 * preemption requested:
589 * It is essential that we are as careful as in the case of
590 * t_checkpoint_list with removing the buffer from the list as
591 * we can possibly see not yet submitted buffers on io_list
593 ret
+= journal_clean_one_cp_list(transaction
->
594 t_checkpoint_io_list
, &released
);
597 } while (transaction
!= last_transaction
);
603 * journal_remove_checkpoint: called after a buffer has been committed
604 * to disk (either by being write-back flushed to disk, or being
605 * committed to the log).
607 * We cannot safely clean a transaction out of the log until all of the
608 * buffer updates committed in that transaction have safely been stored
609 * elsewhere on disk. To achieve this, all of the buffers in a
610 * transaction need to be maintained on the transaction's checkpoint
611 * lists until they have been rewritten, at which point this function is
612 * called to remove the buffer from the existing transaction's
615 * The function returns 1 if it frees the transaction, 0 otherwise.
617 * This function is called with the journal locked.
618 * This function is called with j_list_lock held.
619 * This function is called with jbd_lock_bh_state(jh2bh(jh))
622 int __journal_remove_checkpoint(struct journal_head
*jh
)
624 transaction_t
*transaction
;
628 JBUFFER_TRACE(jh
, "entry");
630 if ((transaction
= jh
->b_cp_transaction
) == NULL
) {
631 JBUFFER_TRACE(jh
, "not on transaction");
634 journal
= transaction
->t_journal
;
637 jh
->b_cp_transaction
= NULL
;
639 if (transaction
->t_checkpoint_list
!= NULL
||
640 transaction
->t_checkpoint_io_list
!= NULL
)
642 JBUFFER_TRACE(jh
, "transaction has no more buffers");
645 * There is one special case to worry about: if we have just pulled the
646 * buffer off a running or committing transaction's checkpoing list,
647 * then even if the checkpoint list is empty, the transaction obviously
650 * The locking here around t_state is a bit sleazy.
651 * See the comment at the end of journal_commit_transaction().
653 if (transaction
->t_state
!= T_FINISHED
) {
654 JBUFFER_TRACE(jh
, "belongs to running/committing transaction");
658 /* OK, that was the last buffer for the transaction: we can now
659 safely remove this transaction from the log */
661 __journal_drop_transaction(journal
, transaction
);
663 /* Just in case anybody was waiting for more transactions to be
665 wake_up(&journal
->j_wait_logspace
);
668 JBUFFER_TRACE(jh
, "exit");
673 * journal_insert_checkpoint: put a committed buffer onto a checkpoint
674 * list so that we know when it is safe to clean the transaction out of
677 * Called with the journal locked.
678 * Called with j_list_lock held.
680 void __journal_insert_checkpoint(struct journal_head
*jh
,
681 transaction_t
*transaction
)
683 JBUFFER_TRACE(jh
, "entry");
684 J_ASSERT_JH(jh
, buffer_dirty(jh2bh(jh
)) || buffer_jbddirty(jh2bh(jh
)));
685 J_ASSERT_JH(jh
, jh
->b_cp_transaction
== NULL
);
687 jh
->b_cp_transaction
= transaction
;
689 if (!transaction
->t_checkpoint_list
) {
690 jh
->b_cpnext
= jh
->b_cpprev
= jh
;
692 jh
->b_cpnext
= transaction
->t_checkpoint_list
;
693 jh
->b_cpprev
= transaction
->t_checkpoint_list
->b_cpprev
;
694 jh
->b_cpprev
->b_cpnext
= jh
;
695 jh
->b_cpnext
->b_cpprev
= jh
;
697 transaction
->t_checkpoint_list
= jh
;
701 * We've finished with this transaction structure: adios...
703 * The transaction must have no links except for the checkpoint by this
706 * Called with the journal locked.
707 * Called with j_list_lock held.
710 void __journal_drop_transaction(journal_t
*journal
, transaction_t
*transaction
)
712 assert_spin_locked(&journal
->j_list_lock
);
713 if (transaction
->t_cpnext
) {
714 transaction
->t_cpnext
->t_cpprev
= transaction
->t_cpprev
;
715 transaction
->t_cpprev
->t_cpnext
= transaction
->t_cpnext
;
716 if (journal
->j_checkpoint_transactions
== transaction
)
717 journal
->j_checkpoint_transactions
=
718 transaction
->t_cpnext
;
719 if (journal
->j_checkpoint_transactions
== transaction
)
720 journal
->j_checkpoint_transactions
= NULL
;
723 J_ASSERT(transaction
->t_state
== T_FINISHED
);
724 J_ASSERT(transaction
->t_buffers
== NULL
);
725 J_ASSERT(transaction
->t_sync_datalist
== NULL
);
726 J_ASSERT(transaction
->t_forget
== NULL
);
727 J_ASSERT(transaction
->t_iobuf_list
== NULL
);
728 J_ASSERT(transaction
->t_shadow_list
== NULL
);
729 J_ASSERT(transaction
->t_log_list
== NULL
);
730 J_ASSERT(transaction
->t_checkpoint_list
== NULL
);
731 J_ASSERT(transaction
->t_checkpoint_io_list
== NULL
);
732 J_ASSERT(transaction
->t_updates
== 0);
733 J_ASSERT(journal
->j_committing_transaction
!= transaction
);
734 J_ASSERT(journal
->j_running_transaction
!= transaction
);
736 jbd_debug(1, "Dropping transaction %d, all done\n", transaction
->t_tid
);