1 // SPDX-License-Identifier: GPL-2.0+
3 * linux/fs/jbd2/checkpoint.c
5 * Written by Stephen C. Tweedie <sct@redhat.com>, 1999
7 * Copyright 1999 Red Hat Software --- All Rights Reserved
9 * Checkpoint routines for the generic filesystem journaling code.
10 * Part of the ext2fs journaling system.
12 * Checkpointing is the process of ensuring that a section of the log is
13 * committed fully to disk, so that that portion of the log can be
17 #include <linux/time.h>
19 #include <linux/jbd2.h>
20 #include <linux/errno.h>
21 #include <linux/slab.h>
22 #include <linux/blkdev.h>
23 #include <trace/events/jbd2.h>
26 * Unlink a buffer from a transaction checkpoint list.
28 * Called with j_list_lock held.
30 static inline void __buffer_unlink_first(struct journal_head
*jh
)
32 transaction_t
*transaction
= jh
->b_cp_transaction
;
34 jh
->b_cpnext
->b_cpprev
= jh
->b_cpprev
;
35 jh
->b_cpprev
->b_cpnext
= jh
->b_cpnext
;
36 if (transaction
->t_checkpoint_list
== jh
) {
37 transaction
->t_checkpoint_list
= jh
->b_cpnext
;
38 if (transaction
->t_checkpoint_list
== jh
)
39 transaction
->t_checkpoint_list
= NULL
;
44 * Unlink a buffer from a transaction checkpoint(io) list.
46 * Called with j_list_lock held.
48 static inline void __buffer_unlink(struct journal_head
*jh
)
50 transaction_t
*transaction
= jh
->b_cp_transaction
;
52 __buffer_unlink_first(jh
);
53 if (transaction
->t_checkpoint_io_list
== jh
) {
54 transaction
->t_checkpoint_io_list
= jh
->b_cpnext
;
55 if (transaction
->t_checkpoint_io_list
== jh
)
56 transaction
->t_checkpoint_io_list
= NULL
;
61 * Move a buffer from the checkpoint list to the checkpoint io list
63 * Called with j_list_lock held
65 static inline void __buffer_relink_io(struct journal_head
*jh
)
67 transaction_t
*transaction
= jh
->b_cp_transaction
;
69 __buffer_unlink_first(jh
);
71 if (!transaction
->t_checkpoint_io_list
) {
72 jh
->b_cpnext
= jh
->b_cpprev
= jh
;
74 jh
->b_cpnext
= transaction
->t_checkpoint_io_list
;
75 jh
->b_cpprev
= transaction
->t_checkpoint_io_list
->b_cpprev
;
76 jh
->b_cpprev
->b_cpnext
= jh
;
77 jh
->b_cpnext
->b_cpprev
= jh
;
79 transaction
->t_checkpoint_io_list
= jh
;
83 * Try to release a checkpointed buffer from its transaction.
84 * Returns 1 if we released it and 2 if we also released the
87 * Requires j_list_lock
89 static int __try_to_free_cp_buf(struct journal_head
*jh
)
92 struct buffer_head
*bh
= jh2bh(jh
);
94 if (jh
->b_transaction
== NULL
&& !buffer_locked(bh
) &&
95 !buffer_dirty(bh
) && !buffer_write_io_error(bh
)) {
96 JBUFFER_TRACE(jh
, "remove from checkpoint list");
97 ret
= __jbd2_journal_remove_checkpoint(jh
) + 1;
103 * __jbd2_log_wait_for_space: wait until there is space in the journal.
105 * Called under j-state_lock *only*. It will be unlocked if we have to wait
106 * for a checkpoint to free up some space in the log.
108 void __jbd2_log_wait_for_space(journal_t
*journal
)
110 int nblocks
, space_left
;
111 /* assert_spin_locked(&journal->j_state_lock); */
113 nblocks
= jbd2_space_needed(journal
);
114 while (jbd2_log_space_left(journal
) < nblocks
) {
115 write_unlock(&journal
->j_state_lock
);
116 mutex_lock_io(&journal
->j_checkpoint_mutex
);
119 * Test again, another process may have checkpointed while we
120 * were waiting for the checkpoint lock. If there are no
121 * transactions ready to be checkpointed, try to recover
122 * journal space by calling cleanup_journal_tail(), and if
123 * that doesn't work, by waiting for the currently committing
124 * transaction to complete. If there is absolutely no way
125 * to make progress, this is either a BUG or corrupted
126 * filesystem, so abort the journal and leave a stack
127 * trace for forensic evidence.
129 write_lock(&journal
->j_state_lock
);
130 if (journal
->j_flags
& JBD2_ABORT
) {
131 mutex_unlock(&journal
->j_checkpoint_mutex
);
134 spin_lock(&journal
->j_list_lock
);
135 nblocks
= jbd2_space_needed(journal
);
136 space_left
= jbd2_log_space_left(journal
);
137 if (space_left
< nblocks
) {
138 int chkpt
= journal
->j_checkpoint_transactions
!= NULL
;
141 if (journal
->j_committing_transaction
)
142 tid
= journal
->j_committing_transaction
->t_tid
;
143 spin_unlock(&journal
->j_list_lock
);
144 write_unlock(&journal
->j_state_lock
);
146 jbd2_log_do_checkpoint(journal
);
147 } else if (jbd2_cleanup_journal_tail(journal
) == 0) {
148 /* We were able to recover space; yay! */
152 * jbd2_journal_commit_transaction() may want
153 * to take the checkpoint_mutex if JBD2_FLUSHED
154 * is set. So we need to temporarily drop it.
156 mutex_unlock(&journal
->j_checkpoint_mutex
);
157 jbd2_log_wait_commit(journal
, tid
);
158 write_lock(&journal
->j_state_lock
);
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 write_lock(&journal
->j_state_lock
);
172 spin_unlock(&journal
->j_list_lock
);
174 mutex_unlock(&journal
->j_checkpoint_mutex
);
179 __flush_batch(journal_t
*journal
, int *batch_count
)
182 struct blk_plug plug
;
184 blk_start_plug(&plug
);
185 for (i
= 0; i
< *batch_count
; i
++)
186 write_dirty_buffer(journal
->j_chkpt_bhs
[i
], REQ_SYNC
);
187 blk_finish_plug(&plug
);
189 for (i
= 0; i
< *batch_count
; i
++) {
190 struct buffer_head
*bh
= journal
->j_chkpt_bhs
[i
];
191 BUFFER_TRACE(bh
, "brelse");
198 * Perform an actual checkpoint. We take the first transaction on the
199 * list of transactions to be checkpointed and send all its buffers
200 * to disk. We submit larger chunks of data at once.
202 * The journal should be locked before calling this function.
203 * Called with j_checkpoint_mutex held.
205 int jbd2_log_do_checkpoint(journal_t
*journal
)
207 struct journal_head
*jh
;
208 struct buffer_head
*bh
;
209 transaction_t
*transaction
;
211 int result
, batch_count
= 0;
213 jbd_debug(1, "Start checkpoint\n");
216 * First thing: if there are any transactions in the log which
217 * don't need checkpointing, just eliminate them from the
218 * journal straight away.
220 result
= jbd2_cleanup_journal_tail(journal
);
221 trace_jbd2_checkpoint(journal
, result
);
222 jbd_debug(1, "cleanup_journal_tail returned %d\n", result
);
227 * OK, we need to start writing disk blocks. Take one transaction
231 spin_lock(&journal
->j_list_lock
);
232 if (!journal
->j_checkpoint_transactions
)
234 transaction
= journal
->j_checkpoint_transactions
;
235 if (transaction
->t_chp_stats
.cs_chp_time
== 0)
236 transaction
->t_chp_stats
.cs_chp_time
= jiffies
;
237 this_tid
= transaction
->t_tid
;
240 * If someone cleaned up this transaction while we slept, we're
241 * done (maybe it's a new transaction, but it fell at the same
244 if (journal
->j_checkpoint_transactions
!= transaction
||
245 transaction
->t_tid
!= this_tid
)
248 /* checkpoint all of the transaction's buffers */
249 while (transaction
->t_checkpoint_list
) {
250 jh
= transaction
->t_checkpoint_list
;
253 if (buffer_locked(bh
)) {
255 spin_unlock(&journal
->j_list_lock
);
257 /* the journal_head may have gone by now */
258 BUFFER_TRACE(bh
, "brelse");
262 if (jh
->b_transaction
!= NULL
) {
263 transaction_t
*t
= jh
->b_transaction
;
264 tid_t tid
= t
->t_tid
;
266 transaction
->t_chp_stats
.cs_forced_to_close
++;
267 spin_unlock(&journal
->j_list_lock
);
268 if (unlikely(journal
->j_flags
& JBD2_UNMOUNT
))
270 * The journal thread is dead; so
271 * starting and waiting for a commit
272 * to finish will cause us to wait for
273 * a _very_ long time.
276 "JBD2: %s: Waiting for Godot: block %llu\n",
277 journal
->j_devname
, (unsigned long long) bh
->b_blocknr
);
280 __flush_batch(journal
, &batch_count
);
281 jbd2_log_start_commit(journal
, tid
);
283 * jbd2_journal_commit_transaction() may want
284 * to take the checkpoint_mutex if JBD2_FLUSHED
285 * is set, jbd2_update_log_tail() called by
286 * jbd2_journal_commit_transaction() may also take
287 * checkpoint_mutex. So we need to temporarily
290 mutex_unlock(&journal
->j_checkpoint_mutex
);
291 jbd2_log_wait_commit(journal
, tid
);
292 mutex_lock_io(&journal
->j_checkpoint_mutex
);
293 spin_lock(&journal
->j_list_lock
);
296 if (!buffer_dirty(bh
)) {
297 if (unlikely(buffer_write_io_error(bh
)) && !result
)
299 BUFFER_TRACE(bh
, "remove from checkpoint");
300 if (__jbd2_journal_remove_checkpoint(jh
))
301 /* The transaction was released; we're done */
306 * Important: we are about to write the buffer, and
307 * possibly block, while still holding the journal
308 * lock. We cannot afford to let the transaction
309 * logic start messing around with this buffer before
310 * we write it to disk, as that would break
313 BUFFER_TRACE(bh
, "queue");
315 J_ASSERT_BH(bh
, !buffer_jwrite(bh
));
316 journal
->j_chkpt_bhs
[batch_count
++] = bh
;
317 __buffer_relink_io(jh
);
318 transaction
->t_chp_stats
.cs_written
++;
319 if ((batch_count
== JBD2_NR_BATCH
) ||
321 spin_needbreak(&journal
->j_list_lock
))
322 goto unlock_and_flush
;
327 spin_unlock(&journal
->j_list_lock
);
330 __flush_batch(journal
, &batch_count
);
331 spin_lock(&journal
->j_list_lock
);
336 * Now we issued all of the transaction's buffers, let's deal
337 * with the buffers that are out for I/O.
340 /* Did somebody clean up the transaction in the meanwhile? */
341 if (journal
->j_checkpoint_transactions
!= transaction
||
342 transaction
->t_tid
!= this_tid
)
345 while (transaction
->t_checkpoint_io_list
) {
346 jh
= transaction
->t_checkpoint_io_list
;
348 if (buffer_locked(bh
)) {
350 spin_unlock(&journal
->j_list_lock
);
352 /* the journal_head may have gone by now */
353 BUFFER_TRACE(bh
, "brelse");
355 spin_lock(&journal
->j_list_lock
);
358 if (unlikely(buffer_write_io_error(bh
)) && !result
)
362 * Now in whatever state the buffer currently is, we
363 * know that it has been written out and so we can
364 * drop it from the list
366 if (__jbd2_journal_remove_checkpoint(jh
))
370 spin_unlock(&journal
->j_list_lock
);
372 jbd2_journal_abort(journal
, result
);
374 result
= jbd2_cleanup_journal_tail(journal
);
376 return (result
< 0) ? result
: 0;
380 * Check the list of checkpoint transactions for the journal to see if
381 * we have already got rid of any since the last update of the log tail
382 * in the journal superblock. If so, we can instantly roll the
383 * superblock forward to remove those transactions from the log.
385 * Return <0 on error, 0 on success, 1 if there was nothing to clean up.
387 * Called with the journal lock held.
389 * This is the only part of the journaling code which really needs to be
390 * aware of transaction aborts. Checkpointing involves writing to the
391 * main filesystem area rather than to the journal, so it can proceed
392 * even in abort state, but we must not update the super block if
393 * checkpointing may have failed. Otherwise, we would lose some metadata
394 * buffers which should be written-back to the filesystem.
397 int jbd2_cleanup_journal_tail(journal_t
*journal
)
400 unsigned long blocknr
;
402 if (is_journal_aborted(journal
))
405 if (!jbd2_journal_get_log_tail(journal
, &first_tid
, &blocknr
))
407 J_ASSERT(blocknr
!= 0);
410 * We need to make sure that any blocks that were recently written out
411 * --- perhaps by jbd2_log_do_checkpoint() --- are flushed out before
412 * we drop the transactions from the journal. It's unlikely this will
413 * be necessary, especially with an appropriately sized journal, but we
414 * need this to guarantee correctness. Fortunately
415 * jbd2_cleanup_journal_tail() doesn't get called all that often.
417 if (journal
->j_flags
& JBD2_BARRIER
)
418 blkdev_issue_flush(journal
->j_fs_dev
, GFP_NOFS
, NULL
);
420 return __jbd2_update_log_tail(journal
, first_tid
, blocknr
);
424 /* Checkpoint list management */
427 * journal_clean_one_cp_list
429 * Find all the written-back checkpoint buffers in the given list and
430 * release them. If 'destroy' is set, clean all buffers unconditionally.
432 * Called with j_list_lock held.
433 * Returns 1 if we freed the transaction, 0 otherwise.
435 static int journal_clean_one_cp_list(struct journal_head
*jh
, bool destroy
)
437 struct journal_head
*last_jh
;
438 struct journal_head
*next_jh
= jh
;
444 last_jh
= jh
->b_cpprev
;
447 next_jh
= jh
->b_cpnext
;
449 ret
= __try_to_free_cp_buf(jh
);
451 ret
= __jbd2_journal_remove_checkpoint(jh
) + 1;
457 * This function only frees up some memory
458 * if possible so we dont have an obligation
459 * to finish processing. Bail out if preemption
464 } while (jh
!= last_jh
);
470 * journal_clean_checkpoint_list
472 * Find all the written-back checkpoint buffers in the journal and release them.
473 * If 'destroy' is set, release all buffers unconditionally.
475 * Called with j_list_lock held.
477 void __jbd2_journal_clean_checkpoint_list(journal_t
*journal
, bool destroy
)
479 transaction_t
*transaction
, *last_transaction
, *next_transaction
;
482 transaction
= journal
->j_checkpoint_transactions
;
486 last_transaction
= transaction
->t_cpprev
;
487 next_transaction
= transaction
;
489 transaction
= next_transaction
;
490 next_transaction
= transaction
->t_cpnext
;
491 ret
= journal_clean_one_cp_list(transaction
->t_checkpoint_list
,
494 * This function only frees up some memory if possible so we
495 * dont have an obligation to finish processing. Bail out if
496 * preemption requested:
503 * It is essential that we are as careful as in the case of
504 * t_checkpoint_list with removing the buffer from the list as
505 * we can possibly see not yet submitted buffers on io_list
507 ret
= journal_clean_one_cp_list(transaction
->
508 t_checkpoint_io_list
, destroy
);
512 * Stop scanning if we couldn't free the transaction. This
513 * avoids pointless scanning of transactions which still
514 * weren't checkpointed.
518 } while (transaction
!= last_transaction
);
522 * Remove buffers from all checkpoint lists as journal is aborted and we just
523 * need to free memory
525 void jbd2_journal_destroy_checkpoint(journal_t
*journal
)
528 * We loop because __jbd2_journal_clean_checkpoint_list() may abort
529 * early due to a need of rescheduling.
532 spin_lock(&journal
->j_list_lock
);
533 if (!journal
->j_checkpoint_transactions
) {
534 spin_unlock(&journal
->j_list_lock
);
537 __jbd2_journal_clean_checkpoint_list(journal
, true);
538 spin_unlock(&journal
->j_list_lock
);
544 * journal_remove_checkpoint: called after a buffer has been committed
545 * to disk (either by being write-back flushed to disk, or being
546 * committed to the log).
548 * We cannot safely clean a transaction out of the log until all of the
549 * buffer updates committed in that transaction have safely been stored
550 * elsewhere on disk. To achieve this, all of the buffers in a
551 * transaction need to be maintained on the transaction's checkpoint
552 * lists until they have been rewritten, at which point this function is
553 * called to remove the buffer from the existing transaction's
556 * The function returns 1 if it frees the transaction, 0 otherwise.
557 * The function can free jh and bh.
559 * This function is called with j_list_lock held.
561 int __jbd2_journal_remove_checkpoint(struct journal_head
*jh
)
563 struct transaction_chp_stats_s
*stats
;
564 transaction_t
*transaction
;
568 JBUFFER_TRACE(jh
, "entry");
570 if ((transaction
= jh
->b_cp_transaction
) == NULL
) {
571 JBUFFER_TRACE(jh
, "not on transaction");
574 journal
= transaction
->t_journal
;
576 JBUFFER_TRACE(jh
, "removing from transaction");
578 jh
->b_cp_transaction
= NULL
;
579 jbd2_journal_put_journal_head(jh
);
581 if (transaction
->t_checkpoint_list
!= NULL
||
582 transaction
->t_checkpoint_io_list
!= NULL
)
586 * There is one special case to worry about: if we have just pulled the
587 * buffer off a running or committing transaction's checkpoing list,
588 * then even if the checkpoint list is empty, the transaction obviously
591 * The locking here around t_state is a bit sleazy.
592 * See the comment at the end of jbd2_journal_commit_transaction().
594 if (transaction
->t_state
!= T_FINISHED
)
597 /* OK, that was the last buffer for the transaction: we can now
598 safely remove this transaction from the log */
599 stats
= &transaction
->t_chp_stats
;
600 if (stats
->cs_chp_time
)
601 stats
->cs_chp_time
= jbd2_time_diff(stats
->cs_chp_time
,
603 trace_jbd2_checkpoint_stats(journal
->j_fs_dev
->bd_dev
,
604 transaction
->t_tid
, stats
);
606 __jbd2_journal_drop_transaction(journal
, transaction
);
607 jbd2_journal_free_transaction(transaction
);
614 * journal_insert_checkpoint: put a committed buffer onto a checkpoint
615 * list so that we know when it is safe to clean the transaction out of
618 * Called with the journal locked.
619 * Called with j_list_lock held.
621 void __jbd2_journal_insert_checkpoint(struct journal_head
*jh
,
622 transaction_t
*transaction
)
624 JBUFFER_TRACE(jh
, "entry");
625 J_ASSERT_JH(jh
, buffer_dirty(jh2bh(jh
)) || buffer_jbddirty(jh2bh(jh
)));
626 J_ASSERT_JH(jh
, jh
->b_cp_transaction
== NULL
);
628 /* Get reference for checkpointing transaction */
629 jbd2_journal_grab_journal_head(jh2bh(jh
));
630 jh
->b_cp_transaction
= transaction
;
632 if (!transaction
->t_checkpoint_list
) {
633 jh
->b_cpnext
= jh
->b_cpprev
= jh
;
635 jh
->b_cpnext
= transaction
->t_checkpoint_list
;
636 jh
->b_cpprev
= transaction
->t_checkpoint_list
->b_cpprev
;
637 jh
->b_cpprev
->b_cpnext
= jh
;
638 jh
->b_cpnext
->b_cpprev
= jh
;
640 transaction
->t_checkpoint_list
= jh
;
644 * We've finished with this transaction structure: adios...
646 * The transaction must have no links except for the checkpoint by this
649 * Called with the journal locked.
650 * Called with j_list_lock held.
653 void __jbd2_journal_drop_transaction(journal_t
*journal
, transaction_t
*transaction
)
655 assert_spin_locked(&journal
->j_list_lock
);
656 if (transaction
->t_cpnext
) {
657 transaction
->t_cpnext
->t_cpprev
= transaction
->t_cpprev
;
658 transaction
->t_cpprev
->t_cpnext
= transaction
->t_cpnext
;
659 if (journal
->j_checkpoint_transactions
== transaction
)
660 journal
->j_checkpoint_transactions
=
661 transaction
->t_cpnext
;
662 if (journal
->j_checkpoint_transactions
== transaction
)
663 journal
->j_checkpoint_transactions
= NULL
;
666 J_ASSERT(transaction
->t_state
== T_FINISHED
);
667 J_ASSERT(transaction
->t_buffers
== NULL
);
668 J_ASSERT(transaction
->t_forget
== NULL
);
669 J_ASSERT(transaction
->t_shadow_list
== NULL
);
670 J_ASSERT(transaction
->t_checkpoint_list
== NULL
);
671 J_ASSERT(transaction
->t_checkpoint_io_list
== NULL
);
672 J_ASSERT(atomic_read(&transaction
->t_updates
) == 0);
673 J_ASSERT(journal
->j_committing_transaction
!= transaction
);
674 J_ASSERT(journal
->j_running_transaction
!= transaction
);
676 trace_jbd2_drop_transaction(journal
, transaction
);
678 jbd_debug(1, "Dropping transaction %d, all done\n", transaction
->t_tid
);