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
92 static int __try_to_free_cp_buf(struct journal_head
*jh
)
95 struct buffer_head
*bh
= jh2bh(jh
);
97 if (jh
->b_transaction
== NULL
&& !buffer_locked(bh
) &&
98 !buffer_dirty(bh
) && !buffer_write_io_error(bh
)) {
99 JBUFFER_TRACE(jh
, "remove from checkpoint list");
100 ret
= __jbd2_journal_remove_checkpoint(jh
) + 1;
106 * __jbd2_log_wait_for_space: wait until there is space in the journal.
108 * Called under j-state_lock *only*. It will be unlocked if we have to wait
109 * for a checkpoint to free up some space in the log.
111 void __jbd2_log_wait_for_space(journal_t
*journal
)
113 int nblocks
, space_left
;
114 /* assert_spin_locked(&journal->j_state_lock); */
116 nblocks
= jbd2_space_needed(journal
);
117 while (jbd2_log_space_left(journal
) < nblocks
) {
118 write_unlock(&journal
->j_state_lock
);
119 mutex_lock(&journal
->j_checkpoint_mutex
);
122 * Test again, another process may have checkpointed while we
123 * were waiting for the checkpoint lock. If there are no
124 * transactions ready to be checkpointed, try to recover
125 * journal space by calling cleanup_journal_tail(), and if
126 * that doesn't work, by waiting for the currently committing
127 * transaction to complete. If there is absolutely no way
128 * to make progress, this is either a BUG or corrupted
129 * filesystem, so abort the journal and leave a stack
130 * trace for forensic evidence.
132 write_lock(&journal
->j_state_lock
);
133 if (journal
->j_flags
& JBD2_ABORT
) {
134 mutex_unlock(&journal
->j_checkpoint_mutex
);
137 spin_lock(&journal
->j_list_lock
);
138 nblocks
= jbd2_space_needed(journal
);
139 space_left
= jbd2_log_space_left(journal
);
140 if (space_left
< nblocks
) {
141 int chkpt
= journal
->j_checkpoint_transactions
!= NULL
;
144 if (journal
->j_committing_transaction
)
145 tid
= journal
->j_committing_transaction
->t_tid
;
146 spin_unlock(&journal
->j_list_lock
);
147 write_unlock(&journal
->j_state_lock
);
149 jbd2_log_do_checkpoint(journal
);
150 } else if (jbd2_cleanup_journal_tail(journal
) == 0) {
151 /* We were able to recover space; yay! */
155 * jbd2_journal_commit_transaction() may want
156 * to take the checkpoint_mutex if JBD2_FLUSHED
157 * is set. So we need to temporarily drop it.
159 mutex_unlock(&journal
->j_checkpoint_mutex
);
160 jbd2_log_wait_commit(journal
, tid
);
161 write_lock(&journal
->j_state_lock
);
164 printk(KERN_ERR
"%s: needed %d blocks and "
165 "only had %d space available\n",
166 __func__
, nblocks
, space_left
);
167 printk(KERN_ERR
"%s: no way to get more "
168 "journal space in %s\n", __func__
,
171 jbd2_journal_abort(journal
, -EIO
);
173 write_lock(&journal
->j_state_lock
);
175 spin_unlock(&journal
->j_list_lock
);
177 mutex_unlock(&journal
->j_checkpoint_mutex
);
182 __flush_batch(journal_t
*journal
, int *batch_count
)
185 struct blk_plug plug
;
187 blk_start_plug(&plug
);
188 for (i
= 0; i
< *batch_count
; i
++)
189 write_dirty_buffer(journal
->j_chkpt_bhs
[i
], WRITE_SYNC
);
190 blk_finish_plug(&plug
);
192 for (i
= 0; i
< *batch_count
; i
++) {
193 struct buffer_head
*bh
= journal
->j_chkpt_bhs
[i
];
194 BUFFER_TRACE(bh
, "brelse");
201 * Perform an actual checkpoint. We take the first transaction on the
202 * list of transactions to be checkpointed and send all its buffers
203 * to disk. We submit larger chunks of data at once.
205 * The journal should be locked before calling this function.
206 * Called with j_checkpoint_mutex held.
208 int jbd2_log_do_checkpoint(journal_t
*journal
)
210 struct journal_head
*jh
;
211 struct buffer_head
*bh
;
212 transaction_t
*transaction
;
214 int result
, batch_count
= 0;
216 jbd_debug(1, "Start checkpoint\n");
219 * First thing: if there are any transactions in the log which
220 * don't need checkpointing, just eliminate them from the
221 * journal straight away.
223 result
= jbd2_cleanup_journal_tail(journal
);
224 trace_jbd2_checkpoint(journal
, result
);
225 jbd_debug(1, "cleanup_journal_tail returned %d\n", result
);
230 * OK, we need to start writing disk blocks. Take one transaction
234 spin_lock(&journal
->j_list_lock
);
235 if (!journal
->j_checkpoint_transactions
)
237 transaction
= journal
->j_checkpoint_transactions
;
238 if (transaction
->t_chp_stats
.cs_chp_time
== 0)
239 transaction
->t_chp_stats
.cs_chp_time
= jiffies
;
240 this_tid
= transaction
->t_tid
;
243 * If someone cleaned up this transaction while we slept, we're
244 * done (maybe it's a new transaction, but it fell at the same
247 if (journal
->j_checkpoint_transactions
!= transaction
||
248 transaction
->t_tid
!= this_tid
)
251 /* checkpoint all of the transaction's buffers */
252 while (transaction
->t_checkpoint_list
) {
253 jh
= transaction
->t_checkpoint_list
;
256 if (buffer_locked(bh
)) {
258 spin_unlock(&journal
->j_list_lock
);
260 /* the journal_head may have gone by now */
261 BUFFER_TRACE(bh
, "brelse");
265 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 if (unlikely(journal
->j_flags
& JBD2_UNMOUNT
))
273 * The journal thread is dead; so
274 * starting and waiting for a commit
275 * to finish will cause us to wait for
276 * a _very_ long time.
279 "JBD2: %s: Waiting for Godot: block %llu\n",
280 journal
->j_devname
, (unsigned long long) bh
->b_blocknr
);
282 jbd2_log_start_commit(journal
, tid
);
283 jbd2_log_wait_commit(journal
, tid
);
286 if (!buffer_dirty(bh
)) {
287 if (unlikely(buffer_write_io_error(bh
)) && !result
)
289 BUFFER_TRACE(bh
, "remove from checkpoint");
290 if (__jbd2_journal_remove_checkpoint(jh
))
291 /* The transaction was released; we're done */
296 * Important: we are about to write the buffer, and
297 * possibly block, while still holding the journal
298 * lock. We cannot afford to let the transaction
299 * logic start messing around with this buffer before
300 * we write it to disk, as that would break
303 BUFFER_TRACE(bh
, "queue");
305 J_ASSERT_BH(bh
, !buffer_jwrite(bh
));
306 journal
->j_chkpt_bhs
[batch_count
++] = bh
;
307 __buffer_relink_io(jh
);
308 transaction
->t_chp_stats
.cs_written
++;
309 if ((batch_count
== JBD2_NR_BATCH
) ||
311 spin_needbreak(&journal
->j_list_lock
))
312 goto unlock_and_flush
;
317 spin_unlock(&journal
->j_list_lock
);
320 __flush_batch(journal
, &batch_count
);
321 spin_lock(&journal
->j_list_lock
);
326 * Now we issued all of the transaction's buffers, let's deal
327 * with the buffers that are out for I/O.
330 /* Did somebody clean up the transaction in the meanwhile? */
331 if (journal
->j_checkpoint_transactions
!= transaction
||
332 transaction
->t_tid
!= this_tid
)
335 while (transaction
->t_checkpoint_io_list
) {
336 jh
= transaction
->t_checkpoint_io_list
;
338 if (buffer_locked(bh
)) {
340 spin_unlock(&journal
->j_list_lock
);
342 /* the journal_head may have gone by now */
343 BUFFER_TRACE(bh
, "brelse");
345 spin_lock(&journal
->j_list_lock
);
348 if (unlikely(buffer_write_io_error(bh
)) && !result
)
352 * Now in whatever state the buffer currently is, we
353 * know that it has been written out and so we can
354 * drop it from the list
356 if (__jbd2_journal_remove_checkpoint(jh
))
360 spin_unlock(&journal
->j_list_lock
);
362 jbd2_journal_abort(journal
, result
);
364 result
= jbd2_cleanup_journal_tail(journal
);
366 return (result
< 0) ? result
: 0;
370 * Check the list of checkpoint transactions for the journal to see if
371 * we have already got rid of any since the last update of the log tail
372 * in the journal superblock. If so, we can instantly roll the
373 * superblock forward to remove those transactions from the log.
375 * Return <0 on error, 0 on success, 1 if there was nothing to clean up.
377 * Called with the journal lock held.
379 * This is the only part of the journaling code which really needs to be
380 * aware of transaction aborts. Checkpointing involves writing to the
381 * main filesystem area rather than to the journal, so it can proceed
382 * even in abort state, but we must not update the super block if
383 * checkpointing may have failed. Otherwise, we would lose some metadata
384 * buffers which should be written-back to the filesystem.
387 int jbd2_cleanup_journal_tail(journal_t
*journal
)
390 unsigned long blocknr
;
392 if (is_journal_aborted(journal
))
395 if (!jbd2_journal_get_log_tail(journal
, &first_tid
, &blocknr
))
397 J_ASSERT(blocknr
!= 0);
400 * We need to make sure that any blocks that were recently written out
401 * --- perhaps by jbd2_log_do_checkpoint() --- are flushed out before
402 * we drop the transactions from the journal. It's unlikely this will
403 * be necessary, especially with an appropriately sized journal, but we
404 * need this to guarantee correctness. Fortunately
405 * jbd2_cleanup_journal_tail() doesn't get called all that often.
407 if (journal
->j_flags
& JBD2_BARRIER
)
408 blkdev_issue_flush(journal
->j_fs_dev
, GFP_NOFS
, NULL
);
410 return __jbd2_update_log_tail(journal
, first_tid
, blocknr
);
414 /* Checkpoint list management */
417 * journal_clean_one_cp_list
419 * Find all the written-back checkpoint buffers in the given list and
420 * release them. If 'destroy' is set, clean all buffers unconditionally.
422 * Called with j_list_lock held.
423 * Returns 1 if we freed the transaction, 0 otherwise.
425 static int journal_clean_one_cp_list(struct journal_head
*jh
, bool destroy
)
427 struct journal_head
*last_jh
;
428 struct journal_head
*next_jh
= jh
;
434 last_jh
= jh
->b_cpprev
;
437 next_jh
= jh
->b_cpnext
;
439 ret
= __try_to_free_cp_buf(jh
);
441 ret
= __jbd2_journal_remove_checkpoint(jh
) + 1;
447 * This function only frees up some memory
448 * if possible so we dont have an obligation
449 * to finish processing. Bail out if preemption
454 } while (jh
!= last_jh
);
460 * journal_clean_checkpoint_list
462 * Find all the written-back checkpoint buffers in the journal and release them.
463 * If 'destroy' is set, release all buffers unconditionally.
465 * Called with j_list_lock held.
467 void __jbd2_journal_clean_checkpoint_list(journal_t
*journal
, bool destroy
)
469 transaction_t
*transaction
, *last_transaction
, *next_transaction
;
472 transaction
= journal
->j_checkpoint_transactions
;
476 last_transaction
= transaction
->t_cpprev
;
477 next_transaction
= transaction
;
479 transaction
= next_transaction
;
480 next_transaction
= transaction
->t_cpnext
;
481 ret
= journal_clean_one_cp_list(transaction
->t_checkpoint_list
,
484 * This function only frees up some memory if possible so we
485 * dont have an obligation to finish processing. Bail out if
486 * preemption requested:
493 * It is essential that we are as careful as in the case of
494 * t_checkpoint_list with removing the buffer from the list as
495 * we can possibly see not yet submitted buffers on io_list
497 ret
= journal_clean_one_cp_list(transaction
->
498 t_checkpoint_io_list
, destroy
);
502 * Stop scanning if we couldn't free the transaction. This
503 * avoids pointless scanning of transactions which still
504 * weren't checkpointed.
508 } while (transaction
!= last_transaction
);
512 * Remove buffers from all checkpoint lists as journal is aborted and we just
513 * need to free memory
515 void jbd2_journal_destroy_checkpoint(journal_t
*journal
)
518 * We loop because __jbd2_journal_clean_checkpoint_list() may abort
519 * early due to a need of rescheduling.
522 spin_lock(&journal
->j_list_lock
);
523 if (!journal
->j_checkpoint_transactions
) {
524 spin_unlock(&journal
->j_list_lock
);
527 __jbd2_journal_clean_checkpoint_list(journal
, true);
528 spin_unlock(&journal
->j_list_lock
);
534 * journal_remove_checkpoint: called after a buffer has been committed
535 * to disk (either by being write-back flushed to disk, or being
536 * committed to the log).
538 * We cannot safely clean a transaction out of the log until all of the
539 * buffer updates committed in that transaction have safely been stored
540 * elsewhere on disk. To achieve this, all of the buffers in a
541 * transaction need to be maintained on the transaction's checkpoint
542 * lists until they have been rewritten, at which point this function is
543 * called to remove the buffer from the existing transaction's
546 * The function returns 1 if it frees the transaction, 0 otherwise.
547 * The function can free jh and bh.
549 * This function is called with j_list_lock held.
551 int __jbd2_journal_remove_checkpoint(struct journal_head
*jh
)
553 struct transaction_chp_stats_s
*stats
;
554 transaction_t
*transaction
;
558 JBUFFER_TRACE(jh
, "entry");
560 if ((transaction
= jh
->b_cp_transaction
) == NULL
) {
561 JBUFFER_TRACE(jh
, "not on transaction");
564 journal
= transaction
->t_journal
;
566 JBUFFER_TRACE(jh
, "removing from transaction");
568 jh
->b_cp_transaction
= NULL
;
569 jbd2_journal_put_journal_head(jh
);
571 if (transaction
->t_checkpoint_list
!= NULL
||
572 transaction
->t_checkpoint_io_list
!= NULL
)
576 * There is one special case to worry about: if we have just pulled the
577 * buffer off a running or committing transaction's checkpoing list,
578 * then even if the checkpoint list is empty, the transaction obviously
581 * The locking here around t_state is a bit sleazy.
582 * See the comment at the end of jbd2_journal_commit_transaction().
584 if (transaction
->t_state
!= T_FINISHED
)
587 /* OK, that was the last buffer for the transaction: we can now
588 safely remove this transaction from the log */
589 stats
= &transaction
->t_chp_stats
;
590 if (stats
->cs_chp_time
)
591 stats
->cs_chp_time
= jbd2_time_diff(stats
->cs_chp_time
,
593 trace_jbd2_checkpoint_stats(journal
->j_fs_dev
->bd_dev
,
594 transaction
->t_tid
, stats
);
596 __jbd2_journal_drop_transaction(journal
, transaction
);
597 jbd2_journal_free_transaction(transaction
);
604 * journal_insert_checkpoint: put a committed buffer onto a checkpoint
605 * list so that we know when it is safe to clean the transaction out of
608 * Called with the journal locked.
609 * Called with j_list_lock held.
611 void __jbd2_journal_insert_checkpoint(struct journal_head
*jh
,
612 transaction_t
*transaction
)
614 JBUFFER_TRACE(jh
, "entry");
615 J_ASSERT_JH(jh
, buffer_dirty(jh2bh(jh
)) || buffer_jbddirty(jh2bh(jh
)));
616 J_ASSERT_JH(jh
, jh
->b_cp_transaction
== NULL
);
618 /* Get reference for checkpointing transaction */
619 jbd2_journal_grab_journal_head(jh2bh(jh
));
620 jh
->b_cp_transaction
= transaction
;
622 if (!transaction
->t_checkpoint_list
) {
623 jh
->b_cpnext
= jh
->b_cpprev
= jh
;
625 jh
->b_cpnext
= transaction
->t_checkpoint_list
;
626 jh
->b_cpprev
= transaction
->t_checkpoint_list
->b_cpprev
;
627 jh
->b_cpprev
->b_cpnext
= jh
;
628 jh
->b_cpnext
->b_cpprev
= jh
;
630 transaction
->t_checkpoint_list
= jh
;
634 * We've finished with this transaction structure: adios...
636 * The transaction must have no links except for the checkpoint by this
639 * Called with the journal locked.
640 * Called with j_list_lock held.
643 void __jbd2_journal_drop_transaction(journal_t
*journal
, transaction_t
*transaction
)
645 assert_spin_locked(&journal
->j_list_lock
);
646 if (transaction
->t_cpnext
) {
647 transaction
->t_cpnext
->t_cpprev
= transaction
->t_cpprev
;
648 transaction
->t_cpprev
->t_cpnext
= transaction
->t_cpnext
;
649 if (journal
->j_checkpoint_transactions
== transaction
)
650 journal
->j_checkpoint_transactions
=
651 transaction
->t_cpnext
;
652 if (journal
->j_checkpoint_transactions
== transaction
)
653 journal
->j_checkpoint_transactions
= NULL
;
656 J_ASSERT(transaction
->t_state
== T_FINISHED
);
657 J_ASSERT(transaction
->t_buffers
== NULL
);
658 J_ASSERT(transaction
->t_forget
== NULL
);
659 J_ASSERT(transaction
->t_shadow_list
== NULL
);
660 J_ASSERT(transaction
->t_checkpoint_list
== NULL
);
661 J_ASSERT(transaction
->t_checkpoint_io_list
== NULL
);
662 J_ASSERT(atomic_read(&transaction
->t_updates
) == 0);
663 J_ASSERT(journal
->j_committing_transaction
!= transaction
);
664 J_ASSERT(journal
->j_running_transaction
!= transaction
);
666 trace_jbd2_drop_transaction(journal
, transaction
);
668 jbd_debug(1, "Dropping transaction %d, all done\n", transaction
->t_tid
);