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>
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
) && !buffer_dirty(bh
)) {
97 JBUFFER_TRACE(jh
, "remove from checkpoint list");
98 ret
= __jbd2_journal_remove_checkpoint(jh
) + 1;
99 jbd_unlock_bh_state(bh
);
100 jbd2_journal_remove_journal_head(bh
);
101 BUFFER_TRACE(bh
, "release");
104 jbd_unlock_bh_state(bh
);
110 * __jbd2_log_wait_for_space: wait until there is space in the journal.
112 * Called under j-state_lock *only*. It will be unlocked if we have to wait
113 * for a checkpoint to free up some space in the log.
115 void __jbd2_log_wait_for_space(journal_t
*journal
)
118 assert_spin_locked(&journal
->j_state_lock
);
120 nblocks
= jbd_space_needed(journal
);
121 while (__jbd2_log_space_left(journal
) < nblocks
) {
122 if (journal
->j_flags
& JBD2_ABORT
)
124 spin_unlock(&journal
->j_state_lock
);
125 mutex_lock(&journal
->j_checkpoint_mutex
);
128 * Test again, another process may have checkpointed while we
129 * were waiting for the checkpoint lock. If there are no
130 * outstanding transactions there is nothing to checkpoint and
131 * we can't make progress. Abort the journal in this case.
133 spin_lock(&journal
->j_state_lock
);
134 spin_lock(&journal
->j_list_lock
);
135 nblocks
= jbd_space_needed(journal
);
136 if (__jbd2_log_space_left(journal
) < nblocks
) {
137 int chkpt
= journal
->j_checkpoint_transactions
!= NULL
;
139 spin_unlock(&journal
->j_list_lock
);
140 spin_unlock(&journal
->j_state_lock
);
142 jbd2_log_do_checkpoint(journal
);
144 printk(KERN_ERR
"%s: no transactions\n",
146 jbd2_journal_abort(journal
, 0);
149 spin_lock(&journal
->j_state_lock
);
151 spin_unlock(&journal
->j_list_lock
);
153 mutex_unlock(&journal
->j_checkpoint_mutex
);
158 * We were unable to perform jbd_trylock_bh_state() inside j_list_lock.
159 * The caller must restart a list walk. Wait for someone else to run
160 * jbd_unlock_bh_state().
162 static void jbd_sync_bh(journal_t
*journal
, struct buffer_head
*bh
)
163 __releases(journal
->j_list_lock
)
166 spin_unlock(&journal
->j_list_lock
);
167 jbd_lock_bh_state(bh
);
168 jbd_unlock_bh_state(bh
);
173 * Clean up transaction's list of buffers submitted for io.
174 * We wait for any pending IO to complete and remove any clean
175 * buffers. Note that we take the buffers in the opposite ordering
176 * from the one in which they were submitted for IO.
178 * Called with j_list_lock held.
180 static void __wait_cp_io(journal_t
*journal
, transaction_t
*transaction
)
182 struct journal_head
*jh
;
183 struct buffer_head
*bh
;
187 this_tid
= transaction
->t_tid
;
189 /* Did somebody clean up the transaction in the meanwhile? */
190 if (journal
->j_checkpoint_transactions
!= transaction
||
191 transaction
->t_tid
!= this_tid
)
193 while (!released
&& transaction
->t_checkpoint_io_list
) {
194 jh
= transaction
->t_checkpoint_io_list
;
196 if (!jbd_trylock_bh_state(bh
)) {
197 jbd_sync_bh(journal
, bh
);
198 spin_lock(&journal
->j_list_lock
);
201 if (buffer_locked(bh
)) {
202 atomic_inc(&bh
->b_count
);
203 spin_unlock(&journal
->j_list_lock
);
204 jbd_unlock_bh_state(bh
);
206 /* the journal_head may have gone by now */
207 BUFFER_TRACE(bh
, "brelse");
209 spin_lock(&journal
->j_list_lock
);
213 * Now in whatever state the buffer currently is, we know that
214 * it has been written out and so we can drop it from the list
216 released
= __jbd2_journal_remove_checkpoint(jh
);
217 jbd_unlock_bh_state(bh
);
218 jbd2_journal_remove_journal_head(bh
);
226 __flush_batch(journal_t
*journal
, struct buffer_head
**bhs
, int *batch_count
)
230 ll_rw_block(SWRITE
, *batch_count
, bhs
);
231 for (i
= 0; i
< *batch_count
; i
++) {
232 struct buffer_head
*bh
= bhs
[i
];
233 clear_buffer_jwrite(bh
);
234 BUFFER_TRACE(bh
, "brelse");
241 * Try to flush one buffer from the checkpoint list to disk.
243 * Return 1 if something happened which requires us to abort the current
244 * scan of the checkpoint list.
246 * Called with j_list_lock held and drops it if 1 is returned
247 * Called under jbd_lock_bh_state(jh2bh(jh)), and drops it
249 static int __process_buffer(journal_t
*journal
, struct journal_head
*jh
,
250 struct buffer_head
**bhs
, int *batch_count
,
251 transaction_t
*transaction
)
253 struct buffer_head
*bh
= jh2bh(jh
);
256 if (buffer_locked(bh
)) {
257 atomic_inc(&bh
->b_count
);
258 spin_unlock(&journal
->j_list_lock
);
259 jbd_unlock_bh_state(bh
);
261 /* the journal_head may have gone by now */
262 BUFFER_TRACE(bh
, "brelse");
265 } else if (jh
->b_transaction
!= NULL
) {
266 transaction_t
*t
= jh
->b_transaction
;
267 tid_t tid
= t
->t_tid
;
269 transaction
->t_chp_stats
.cs_forced_to_close
++;
270 spin_unlock(&journal
->j_list_lock
);
271 jbd_unlock_bh_state(bh
);
272 jbd2_log_start_commit(journal
, tid
);
273 jbd2_log_wait_commit(journal
, tid
);
275 } else if (!buffer_dirty(bh
)) {
276 J_ASSERT_JH(jh
, !buffer_jbddirty(bh
));
277 BUFFER_TRACE(bh
, "remove from checkpoint");
278 __jbd2_journal_remove_checkpoint(jh
);
279 spin_unlock(&journal
->j_list_lock
);
280 jbd_unlock_bh_state(bh
);
281 jbd2_journal_remove_journal_head(bh
);
286 * Important: we are about to write the buffer, and
287 * possibly block, while still holding the journal lock.
288 * We cannot afford to let the transaction logic start
289 * messing around with this buffer before we write it to
290 * disk, as that would break recoverability.
292 BUFFER_TRACE(bh
, "queue");
294 J_ASSERT_BH(bh
, !buffer_jwrite(bh
));
295 set_buffer_jwrite(bh
);
296 bhs
[*batch_count
] = bh
;
297 __buffer_relink_io(jh
);
298 jbd_unlock_bh_state(bh
);
299 transaction
->t_chp_stats
.cs_written
++;
301 if (*batch_count
== NR_BATCH
) {
302 spin_unlock(&journal
->j_list_lock
);
303 __flush_batch(journal
, bhs
, batch_count
);
311 * Perform an actual checkpoint. We take the first transaction on the
312 * list of transactions to be checkpointed and send all its buffers
313 * to disk. We submit larger chunks of data at once.
315 * The journal should be locked before calling this function.
317 int jbd2_log_do_checkpoint(journal_t
*journal
)
319 transaction_t
*transaction
;
323 jbd_debug(1, "Start checkpoint\n");
326 * First thing: if there are any transactions in the log which
327 * don't need checkpointing, just eliminate them from the
328 * journal straight away.
330 result
= jbd2_cleanup_journal_tail(journal
);
331 jbd_debug(1, "cleanup_journal_tail returned %d\n", result
);
336 * OK, we need to start writing disk blocks. Take one transaction
339 spin_lock(&journal
->j_list_lock
);
340 if (!journal
->j_checkpoint_transactions
)
342 transaction
= journal
->j_checkpoint_transactions
;
343 if (transaction
->t_chp_stats
.cs_chp_time
== 0)
344 transaction
->t_chp_stats
.cs_chp_time
= jiffies
;
345 this_tid
= transaction
->t_tid
;
348 * If someone cleaned up this transaction while we slept, we're
349 * done (maybe it's a new transaction, but it fell at the same
352 if (journal
->j_checkpoint_transactions
== transaction
&&
353 transaction
->t_tid
== this_tid
) {
355 struct buffer_head
*bhs
[NR_BATCH
];
356 struct journal_head
*jh
;
359 while (!retry
&& transaction
->t_checkpoint_list
) {
360 struct buffer_head
*bh
;
362 jh
= transaction
->t_checkpoint_list
;
364 if (!jbd_trylock_bh_state(bh
)) {
365 jbd_sync_bh(journal
, bh
);
369 retry
= __process_buffer(journal
, jh
, bhs
, &batch_count
,
371 if (!retry
&& (need_resched() ||
372 spin_needbreak(&journal
->j_list_lock
))) {
373 spin_unlock(&journal
->j_list_lock
);
381 spin_unlock(&journal
->j_list_lock
);
384 __flush_batch(journal
, bhs
, &batch_count
);
388 spin_lock(&journal
->j_list_lock
);
392 * Now we have cleaned up the first transaction's checkpoint
393 * list. Let's clean up the second one
395 __wait_cp_io(journal
, transaction
);
398 spin_unlock(&journal
->j_list_lock
);
399 result
= jbd2_cleanup_journal_tail(journal
);
406 * Check the list of checkpoint transactions for the journal to see if
407 * we have already got rid of any since the last update of the log tail
408 * in the journal superblock. If so, we can instantly roll the
409 * superblock forward to remove those transactions from the log.
411 * Return <0 on error, 0 on success, 1 if there was nothing to clean up.
413 * Called with the journal lock held.
415 * This is the only part of the journaling code which really needs to be
416 * aware of transaction aborts. Checkpointing involves writing to the
417 * main filesystem area rather than to the journal, so it can proceed
418 * even in abort state, but we must not update the journal superblock if
419 * we have an abort error outstanding.
422 int jbd2_cleanup_journal_tail(journal_t
*journal
)
424 transaction_t
* transaction
;
426 unsigned long blocknr
, freed
;
428 /* OK, work out the oldest transaction remaining in the log, and
429 * the log block it starts at.
431 * If the log is now empty, we need to work out which is the
432 * next transaction ID we will write, and where it will
435 spin_lock(&journal
->j_state_lock
);
436 spin_lock(&journal
->j_list_lock
);
437 transaction
= journal
->j_checkpoint_transactions
;
439 first_tid
= transaction
->t_tid
;
440 blocknr
= transaction
->t_log_start
;
441 } else if ((transaction
= journal
->j_committing_transaction
) != NULL
) {
442 first_tid
= transaction
->t_tid
;
443 blocknr
= transaction
->t_log_start
;
444 } else if ((transaction
= journal
->j_running_transaction
) != NULL
) {
445 first_tid
= transaction
->t_tid
;
446 blocknr
= journal
->j_head
;
448 first_tid
= journal
->j_transaction_sequence
;
449 blocknr
= journal
->j_head
;
451 spin_unlock(&journal
->j_list_lock
);
452 J_ASSERT(blocknr
!= 0);
454 /* If the oldest pinned transaction is at the tail of the log
455 already then there's not much we can do right now. */
456 if (journal
->j_tail_sequence
== first_tid
) {
457 spin_unlock(&journal
->j_state_lock
);
461 /* OK, update the superblock to recover the freed space.
462 * Physical blocks come first: have we wrapped beyond the end of
464 freed
= blocknr
- journal
->j_tail
;
465 if (blocknr
< journal
->j_tail
)
466 freed
= freed
+ journal
->j_last
- journal
->j_first
;
469 "Cleaning journal tail from %d to %d (offset %lu), "
471 journal
->j_tail_sequence
, first_tid
, blocknr
, freed
);
473 journal
->j_free
+= freed
;
474 journal
->j_tail_sequence
= first_tid
;
475 journal
->j_tail
= blocknr
;
476 spin_unlock(&journal
->j_state_lock
);
477 if (!(journal
->j_flags
& JBD2_ABORT
))
478 jbd2_journal_update_superblock(journal
, 1);
483 /* Checkpoint list management */
486 * journal_clean_one_cp_list
488 * Find all the written-back checkpoint buffers in the given list and release them.
490 * Called with the journal locked.
491 * Called with j_list_lock held.
492 * Returns number of bufers reaped (for debug)
495 static int journal_clean_one_cp_list(struct journal_head
*jh
, int *released
)
497 struct journal_head
*last_jh
;
498 struct journal_head
*next_jh
= jh
;
505 last_jh
= jh
->b_cpprev
;
508 next_jh
= jh
->b_cpnext
;
509 /* Use trylock because of the ranking */
510 if (jbd_trylock_bh_state(jh2bh(jh
))) {
511 ret
= __try_to_free_cp_buf(jh
);
521 * This function only frees up some memory
522 * if possible so we dont have an obligation
523 * to finish processing. Bail out if preemption
528 } while (jh
!= last_jh
);
534 * journal_clean_checkpoint_list
536 * Find all the written-back checkpoint buffers in the journal and release them.
538 * Called with the journal locked.
539 * Called with j_list_lock held.
540 * Returns number of buffers reaped (for debug)
543 int __jbd2_journal_clean_checkpoint_list(journal_t
*journal
)
545 transaction_t
*transaction
, *last_transaction
, *next_transaction
;
549 transaction
= journal
->j_checkpoint_transactions
;
553 last_transaction
= transaction
->t_cpprev
;
554 next_transaction
= transaction
;
556 transaction
= next_transaction
;
557 next_transaction
= transaction
->t_cpnext
;
558 ret
+= journal_clean_one_cp_list(transaction
->
559 t_checkpoint_list
, &released
);
561 * This function only frees up some memory if possible so we
562 * dont have an obligation to finish processing. Bail out if
563 * preemption requested:
570 * It is essential that we are as careful as in the case of
571 * t_checkpoint_list with removing the buffer from the list as
572 * we can possibly see not yet submitted buffers on io_list
574 ret
+= journal_clean_one_cp_list(transaction
->
575 t_checkpoint_io_list
, &released
);
578 } while (transaction
!= last_transaction
);
584 * journal_remove_checkpoint: called after a buffer has been committed
585 * to disk (either by being write-back flushed to disk, or being
586 * committed to the log).
588 * We cannot safely clean a transaction out of the log until all of the
589 * buffer updates committed in that transaction have safely been stored
590 * elsewhere on disk. To achieve this, all of the buffers in a
591 * transaction need to be maintained on the transaction's checkpoint
592 * lists until they have been rewritten, at which point this function is
593 * called to remove the buffer from the existing transaction's
596 * The function returns 1 if it frees the transaction, 0 otherwise.
598 * This function is called with the journal locked.
599 * This function is called with j_list_lock held.
600 * This function is called with jbd_lock_bh_state(jh2bh(jh))
603 int __jbd2_journal_remove_checkpoint(struct journal_head
*jh
)
605 transaction_t
*transaction
;
609 JBUFFER_TRACE(jh
, "entry");
611 if ((transaction
= jh
->b_cp_transaction
) == NULL
) {
612 JBUFFER_TRACE(jh
, "not on transaction");
615 journal
= transaction
->t_journal
;
618 jh
->b_cp_transaction
= NULL
;
620 if (transaction
->t_checkpoint_list
!= NULL
||
621 transaction
->t_checkpoint_io_list
!= NULL
)
623 JBUFFER_TRACE(jh
, "transaction has no more buffers");
626 * There is one special case to worry about: if we have just pulled the
627 * buffer off a running or committing transaction's checkpoing list,
628 * then even if the checkpoint list is empty, the transaction obviously
631 * The locking here around t_state is a bit sleazy.
632 * See the comment at the end of jbd2_journal_commit_transaction().
634 if (transaction
->t_state
!= T_FINISHED
) {
635 JBUFFER_TRACE(jh
, "belongs to running/committing transaction");
639 /* OK, that was the last buffer for the transaction: we can now
640 safely remove this transaction from the log */
642 __jbd2_journal_drop_transaction(journal
, transaction
);
644 /* Just in case anybody was waiting for more transactions to be
646 wake_up(&journal
->j_wait_logspace
);
649 JBUFFER_TRACE(jh
, "exit");
654 * journal_insert_checkpoint: put a committed buffer onto a checkpoint
655 * list so that we know when it is safe to clean the transaction out of
658 * Called with the journal locked.
659 * Called with j_list_lock held.
661 void __jbd2_journal_insert_checkpoint(struct journal_head
*jh
,
662 transaction_t
*transaction
)
664 JBUFFER_TRACE(jh
, "entry");
665 J_ASSERT_JH(jh
, buffer_dirty(jh2bh(jh
)) || buffer_jbddirty(jh2bh(jh
)));
666 J_ASSERT_JH(jh
, jh
->b_cp_transaction
== NULL
);
668 jh
->b_cp_transaction
= transaction
;
670 if (!transaction
->t_checkpoint_list
) {
671 jh
->b_cpnext
= jh
->b_cpprev
= jh
;
673 jh
->b_cpnext
= transaction
->t_checkpoint_list
;
674 jh
->b_cpprev
= transaction
->t_checkpoint_list
->b_cpprev
;
675 jh
->b_cpprev
->b_cpnext
= jh
;
676 jh
->b_cpnext
->b_cpprev
= jh
;
678 transaction
->t_checkpoint_list
= jh
;
682 * We've finished with this transaction structure: adios...
684 * The transaction must have no links except for the checkpoint by this
687 * Called with the journal locked.
688 * Called with j_list_lock held.
691 void __jbd2_journal_drop_transaction(journal_t
*journal
, transaction_t
*transaction
)
693 assert_spin_locked(&journal
->j_list_lock
);
694 if (transaction
->t_cpnext
) {
695 transaction
->t_cpnext
->t_cpprev
= transaction
->t_cpprev
;
696 transaction
->t_cpprev
->t_cpnext
= transaction
->t_cpnext
;
697 if (journal
->j_checkpoint_transactions
== transaction
)
698 journal
->j_checkpoint_transactions
=
699 transaction
->t_cpnext
;
700 if (journal
->j_checkpoint_transactions
== transaction
)
701 journal
->j_checkpoint_transactions
= NULL
;
704 J_ASSERT(transaction
->t_state
== T_FINISHED
);
705 J_ASSERT(transaction
->t_buffers
== NULL
);
706 J_ASSERT(transaction
->t_forget
== NULL
);
707 J_ASSERT(transaction
->t_iobuf_list
== NULL
);
708 J_ASSERT(transaction
->t_shadow_list
== NULL
);
709 J_ASSERT(transaction
->t_log_list
== NULL
);
710 J_ASSERT(transaction
->t_checkpoint_list
== NULL
);
711 J_ASSERT(transaction
->t_checkpoint_io_list
== NULL
);
712 J_ASSERT(transaction
->t_updates
== 0);
713 J_ASSERT(journal
->j_committing_transaction
!= transaction
);
714 J_ASSERT(journal
->j_running_transaction
!= transaction
);
716 jbd_debug(1, "Dropping transaction %d, all done\n", transaction
->t_tid
);