1 // SPDX-License-Identifier: GPL-2.0+
3 * linux/fs/jbd2/journal.c
5 * Written by Stephen C. Tweedie <sct@redhat.com>, 1998
7 * Copyright 1998 Red Hat corp --- All Rights Reserved
9 * Generic filesystem journal-writing code; part of the ext2fs
12 * This file manages journals: areas of disk reserved for logging
13 * transactional updates. This includes the kernel journaling thread
14 * which is responsible for scheduling updates to the log.
16 * We do not actually manage the physical storage of the journal in this
17 * file: that is left to a per-journal policy function, which allows us
18 * to store the journal within a filesystem-specified area for ext2
19 * journaling (ext2 can use a reserved inode for storing the log).
22 #include <linux/module.h>
23 #include <linux/time.h>
25 #include <linux/jbd2.h>
26 #include <linux/errno.h>
27 #include <linux/slab.h>
28 #include <linux/init.h>
30 #include <linux/freezer.h>
31 #include <linux/pagemap.h>
32 #include <linux/kthread.h>
33 #include <linux/poison.h>
34 #include <linux/proc_fs.h>
35 #include <linux/seq_file.h>
36 #include <linux/math64.h>
37 #include <linux/hash.h>
38 #include <linux/log2.h>
39 #include <linux/vmalloc.h>
40 #include <linux/backing-dev.h>
41 #include <linux/bitops.h>
42 #include <linux/ratelimit.h>
43 #include <linux/sched/mm.h>
45 #define CREATE_TRACE_POINTS
46 #include <trace/events/jbd2.h>
48 #include <linux/uaccess.h>
51 #ifdef CONFIG_JBD2_DEBUG
52 ushort jbd2_journal_enable_debug __read_mostly
;
53 EXPORT_SYMBOL(jbd2_journal_enable_debug
);
55 module_param_named(jbd2_debug
, jbd2_journal_enable_debug
, ushort
, 0644);
56 MODULE_PARM_DESC(jbd2_debug
, "Debugging level for jbd2");
59 EXPORT_SYMBOL(jbd2_journal_extend
);
60 EXPORT_SYMBOL(jbd2_journal_stop
);
61 EXPORT_SYMBOL(jbd2_journal_lock_updates
);
62 EXPORT_SYMBOL(jbd2_journal_unlock_updates
);
63 EXPORT_SYMBOL(jbd2_journal_get_write_access
);
64 EXPORT_SYMBOL(jbd2_journal_get_create_access
);
65 EXPORT_SYMBOL(jbd2_journal_get_undo_access
);
66 EXPORT_SYMBOL(jbd2_journal_set_triggers
);
67 EXPORT_SYMBOL(jbd2_journal_dirty_metadata
);
68 EXPORT_SYMBOL(jbd2_journal_forget
);
70 EXPORT_SYMBOL(journal_sync_buffer
);
72 EXPORT_SYMBOL(jbd2_journal_flush
);
73 EXPORT_SYMBOL(jbd2_journal_revoke
);
75 EXPORT_SYMBOL(jbd2_journal_init_dev
);
76 EXPORT_SYMBOL(jbd2_journal_init_inode
);
77 EXPORT_SYMBOL(jbd2_journal_check_used_features
);
78 EXPORT_SYMBOL(jbd2_journal_check_available_features
);
79 EXPORT_SYMBOL(jbd2_journal_set_features
);
80 EXPORT_SYMBOL(jbd2_journal_load
);
81 EXPORT_SYMBOL(jbd2_journal_destroy
);
82 EXPORT_SYMBOL(jbd2_journal_abort
);
83 EXPORT_SYMBOL(jbd2_journal_errno
);
84 EXPORT_SYMBOL(jbd2_journal_ack_err
);
85 EXPORT_SYMBOL(jbd2_journal_clear_err
);
86 EXPORT_SYMBOL(jbd2_log_wait_commit
);
87 EXPORT_SYMBOL(jbd2_log_start_commit
);
88 EXPORT_SYMBOL(jbd2_journal_start_commit
);
89 EXPORT_SYMBOL(jbd2_journal_force_commit_nested
);
90 EXPORT_SYMBOL(jbd2_journal_wipe
);
91 EXPORT_SYMBOL(jbd2_journal_blocks_per_page
);
92 EXPORT_SYMBOL(jbd2_journal_invalidatepage
);
93 EXPORT_SYMBOL(jbd2_journal_try_to_free_buffers
);
94 EXPORT_SYMBOL(jbd2_journal_force_commit
);
95 EXPORT_SYMBOL(jbd2_journal_inode_add_write
);
96 EXPORT_SYMBOL(jbd2_journal_inode_add_wait
);
97 EXPORT_SYMBOL(jbd2_journal_inode_ranged_write
);
98 EXPORT_SYMBOL(jbd2_journal_inode_ranged_wait
);
99 EXPORT_SYMBOL(jbd2_journal_init_jbd_inode
);
100 EXPORT_SYMBOL(jbd2_journal_release_jbd_inode
);
101 EXPORT_SYMBOL(jbd2_journal_begin_ordered_truncate
);
102 EXPORT_SYMBOL(jbd2_inode_cache
);
104 static void __journal_abort_soft (journal_t
*journal
, int errno
);
105 static int jbd2_journal_create_slab(size_t slab_size
);
107 #ifdef CONFIG_JBD2_DEBUG
108 void __jbd2_debug(int level
, const char *file
, const char *func
,
109 unsigned int line
, const char *fmt
, ...)
111 struct va_format vaf
;
114 if (level
> jbd2_journal_enable_debug
)
119 printk(KERN_DEBUG
"%s: (%s, %u): %pV", file
, func
, line
, &vaf
);
122 EXPORT_SYMBOL(__jbd2_debug
);
125 /* Checksumming functions */
126 static int jbd2_verify_csum_type(journal_t
*j
, journal_superblock_t
*sb
)
128 if (!jbd2_journal_has_csum_v2or3_feature(j
))
131 return sb
->s_checksum_type
== JBD2_CRC32C_CHKSUM
;
134 static __be32
jbd2_superblock_csum(journal_t
*j
, journal_superblock_t
*sb
)
139 old_csum
= sb
->s_checksum
;
141 csum
= jbd2_chksum(j
, ~0, (char *)sb
, sizeof(journal_superblock_t
));
142 sb
->s_checksum
= old_csum
;
144 return cpu_to_be32(csum
);
148 * Helper function used to manage commit timeouts
151 static void commit_timeout(struct timer_list
*t
)
153 journal_t
*journal
= from_timer(journal
, t
, j_commit_timer
);
155 wake_up_process(journal
->j_task
);
159 * kjournald2: The main thread function used to manage a logging device
162 * This kernel thread is responsible for two things:
164 * 1) COMMIT: Every so often we need to commit the current state of the
165 * filesystem to disk. The journal thread is responsible for writing
166 * all of the metadata buffers to disk.
168 * 2) CHECKPOINT: We cannot reuse a used section of the log file until all
169 * of the data in that part of the log has been rewritten elsewhere on
170 * the disk. Flushing these old buffers to reclaim space in the log is
171 * known as checkpointing, and this thread is responsible for that job.
174 static int kjournald2(void *arg
)
176 journal_t
*journal
= arg
;
177 transaction_t
*transaction
;
180 * Set up an interval timer which can be used to trigger a commit wakeup
181 * after the commit interval expires
183 timer_setup(&journal
->j_commit_timer
, commit_timeout
, 0);
187 /* Record that the journal thread is running */
188 journal
->j_task
= current
;
189 wake_up(&journal
->j_wait_done_commit
);
192 * Make sure that no allocations from this kernel thread will ever
193 * recurse to the fs layer because we are responsible for the
194 * transaction commit and any fs involvement might get stuck waiting for
197 memalloc_nofs_save();
200 * And now, wait forever for commit wakeup events.
202 write_lock(&journal
->j_state_lock
);
205 if (journal
->j_flags
& JBD2_UNMOUNT
)
208 jbd_debug(1, "commit_sequence=%u, commit_request=%u\n",
209 journal
->j_commit_sequence
, journal
->j_commit_request
);
211 if (journal
->j_commit_sequence
!= journal
->j_commit_request
) {
212 jbd_debug(1, "OK, requests differ\n");
213 write_unlock(&journal
->j_state_lock
);
214 del_timer_sync(&journal
->j_commit_timer
);
215 jbd2_journal_commit_transaction(journal
);
216 write_lock(&journal
->j_state_lock
);
220 wake_up(&journal
->j_wait_done_commit
);
221 if (freezing(current
)) {
223 * The simpler the better. Flushing journal isn't a
224 * good idea, because that depends on threads that may
225 * be already stopped.
227 jbd_debug(1, "Now suspending kjournald2\n");
228 write_unlock(&journal
->j_state_lock
);
230 write_lock(&journal
->j_state_lock
);
233 * We assume on resume that commits are already there,
237 int should_sleep
= 1;
239 prepare_to_wait(&journal
->j_wait_commit
, &wait
,
241 if (journal
->j_commit_sequence
!= journal
->j_commit_request
)
243 transaction
= journal
->j_running_transaction
;
244 if (transaction
&& time_after_eq(jiffies
,
245 transaction
->t_expires
))
247 if (journal
->j_flags
& JBD2_UNMOUNT
)
250 write_unlock(&journal
->j_state_lock
);
252 write_lock(&journal
->j_state_lock
);
254 finish_wait(&journal
->j_wait_commit
, &wait
);
257 jbd_debug(1, "kjournald2 wakes\n");
260 * Were we woken up by a commit wakeup event?
262 transaction
= journal
->j_running_transaction
;
263 if (transaction
&& time_after_eq(jiffies
, transaction
->t_expires
)) {
264 journal
->j_commit_request
= transaction
->t_tid
;
265 jbd_debug(1, "woke because of timeout\n");
270 del_timer_sync(&journal
->j_commit_timer
);
271 journal
->j_task
= NULL
;
272 wake_up(&journal
->j_wait_done_commit
);
273 jbd_debug(1, "Journal thread exiting.\n");
274 write_unlock(&journal
->j_state_lock
);
278 static int jbd2_journal_start_thread(journal_t
*journal
)
280 struct task_struct
*t
;
282 t
= kthread_run(kjournald2
, journal
, "jbd2/%s",
287 wait_event(journal
->j_wait_done_commit
, journal
->j_task
!= NULL
);
291 static void journal_kill_thread(journal_t
*journal
)
293 write_lock(&journal
->j_state_lock
);
294 journal
->j_flags
|= JBD2_UNMOUNT
;
296 while (journal
->j_task
) {
297 write_unlock(&journal
->j_state_lock
);
298 wake_up(&journal
->j_wait_commit
);
299 wait_event(journal
->j_wait_done_commit
, journal
->j_task
== NULL
);
300 write_lock(&journal
->j_state_lock
);
302 write_unlock(&journal
->j_state_lock
);
306 * jbd2_journal_write_metadata_buffer: write a metadata buffer to the journal.
308 * Writes a metadata buffer to a given disk block. The actual IO is not
309 * performed but a new buffer_head is constructed which labels the data
310 * to be written with the correct destination disk block.
312 * Any magic-number escaping which needs to be done will cause a
313 * copy-out here. If the buffer happens to start with the
314 * JBD2_MAGIC_NUMBER, then we can't write it to the log directly: the
315 * magic number is only written to the log for descripter blocks. In
316 * this case, we copy the data and replace the first word with 0, and we
317 * return a result code which indicates that this buffer needs to be
318 * marked as an escaped buffer in the corresponding log descriptor
319 * block. The missing word can then be restored when the block is read
322 * If the source buffer has already been modified by a new transaction
323 * since we took the last commit snapshot, we use the frozen copy of
324 * that data for IO. If we end up using the existing buffer_head's data
325 * for the write, then we have to make sure nobody modifies it while the
326 * IO is in progress. do_get_write_access() handles this.
328 * The function returns a pointer to the buffer_head to be used for IO.
336 * Bit 0 set == escape performed on the data
337 * Bit 1 set == buffer copy-out performed (kfree the data after IO)
340 int jbd2_journal_write_metadata_buffer(transaction_t
*transaction
,
341 struct journal_head
*jh_in
,
342 struct buffer_head
**bh_out
,
345 int need_copy_out
= 0;
346 int done_copy_out
= 0;
349 struct buffer_head
*new_bh
;
350 struct page
*new_page
;
351 unsigned int new_offset
;
352 struct buffer_head
*bh_in
= jh2bh(jh_in
);
353 journal_t
*journal
= transaction
->t_journal
;
356 * The buffer really shouldn't be locked: only the current committing
357 * transaction is allowed to write it, so nobody else is allowed
360 * akpm: except if we're journalling data, and write() output is
361 * also part of a shared mapping, and another thread has
362 * decided to launch a writepage() against this buffer.
364 J_ASSERT_BH(bh_in
, buffer_jbddirty(bh_in
));
366 new_bh
= alloc_buffer_head(GFP_NOFS
|__GFP_NOFAIL
);
368 /* keep subsequent assertions sane */
369 atomic_set(&new_bh
->b_count
, 1);
371 jbd_lock_bh_state(bh_in
);
374 * If a new transaction has already done a buffer copy-out, then
375 * we use that version of the data for the commit.
377 if (jh_in
->b_frozen_data
) {
379 new_page
= virt_to_page(jh_in
->b_frozen_data
);
380 new_offset
= offset_in_page(jh_in
->b_frozen_data
);
382 new_page
= jh2bh(jh_in
)->b_page
;
383 new_offset
= offset_in_page(jh2bh(jh_in
)->b_data
);
386 mapped_data
= kmap_atomic(new_page
);
388 * Fire data frozen trigger if data already wasn't frozen. Do this
389 * before checking for escaping, as the trigger may modify the magic
390 * offset. If a copy-out happens afterwards, it will have the correct
391 * data in the buffer.
394 jbd2_buffer_frozen_trigger(jh_in
, mapped_data
+ new_offset
,
400 if (*((__be32
*)(mapped_data
+ new_offset
)) ==
401 cpu_to_be32(JBD2_MAGIC_NUMBER
)) {
405 kunmap_atomic(mapped_data
);
408 * Do we need to do a data copy?
410 if (need_copy_out
&& !done_copy_out
) {
413 jbd_unlock_bh_state(bh_in
);
414 tmp
= jbd2_alloc(bh_in
->b_size
, GFP_NOFS
);
419 jbd_lock_bh_state(bh_in
);
420 if (jh_in
->b_frozen_data
) {
421 jbd2_free(tmp
, bh_in
->b_size
);
425 jh_in
->b_frozen_data
= tmp
;
426 mapped_data
= kmap_atomic(new_page
);
427 memcpy(tmp
, mapped_data
+ new_offset
, bh_in
->b_size
);
428 kunmap_atomic(mapped_data
);
430 new_page
= virt_to_page(tmp
);
431 new_offset
= offset_in_page(tmp
);
435 * This isn't strictly necessary, as we're using frozen
436 * data for the escaping, but it keeps consistency with
437 * b_frozen_data usage.
439 jh_in
->b_frozen_triggers
= jh_in
->b_triggers
;
443 * Did we need to do an escaping? Now we've done all the
444 * copying, we can finally do so.
447 mapped_data
= kmap_atomic(new_page
);
448 *((unsigned int *)(mapped_data
+ new_offset
)) = 0;
449 kunmap_atomic(mapped_data
);
452 set_bh_page(new_bh
, new_page
, new_offset
);
453 new_bh
->b_size
= bh_in
->b_size
;
454 new_bh
->b_bdev
= journal
->j_dev
;
455 new_bh
->b_blocknr
= blocknr
;
456 new_bh
->b_private
= bh_in
;
457 set_buffer_mapped(new_bh
);
458 set_buffer_dirty(new_bh
);
463 * The to-be-written buffer needs to get moved to the io queue,
464 * and the original buffer whose contents we are shadowing or
465 * copying is moved to the transaction's shadow queue.
467 JBUFFER_TRACE(jh_in
, "file as BJ_Shadow");
468 spin_lock(&journal
->j_list_lock
);
469 __jbd2_journal_file_buffer(jh_in
, transaction
, BJ_Shadow
);
470 spin_unlock(&journal
->j_list_lock
);
471 set_buffer_shadow(bh_in
);
472 jbd_unlock_bh_state(bh_in
);
474 return do_escape
| (done_copy_out
<< 1);
478 * Allocation code for the journal file. Manage the space left in the
479 * journal, so that we can begin checkpointing when appropriate.
483 * Called with j_state_lock locked for writing.
484 * Returns true if a transaction commit was started.
486 int __jbd2_log_start_commit(journal_t
*journal
, tid_t target
)
488 /* Return if the txn has already requested to be committed */
489 if (journal
->j_commit_request
== target
)
493 * The only transaction we can possibly wait upon is the
494 * currently running transaction (if it exists). Otherwise,
495 * the target tid must be an old one.
497 if (journal
->j_running_transaction
&&
498 journal
->j_running_transaction
->t_tid
== target
) {
500 * We want a new commit: OK, mark the request and wakeup the
501 * commit thread. We do _not_ do the commit ourselves.
504 journal
->j_commit_request
= target
;
505 jbd_debug(1, "JBD2: requesting commit %u/%u\n",
506 journal
->j_commit_request
,
507 journal
->j_commit_sequence
);
508 journal
->j_running_transaction
->t_requested
= jiffies
;
509 wake_up(&journal
->j_wait_commit
);
511 } else if (!tid_geq(journal
->j_commit_request
, target
))
512 /* This should never happen, but if it does, preserve
513 the evidence before kjournald goes into a loop and
514 increments j_commit_sequence beyond all recognition. */
515 WARN_ONCE(1, "JBD2: bad log_start_commit: %u %u %u %u\n",
516 journal
->j_commit_request
,
517 journal
->j_commit_sequence
,
518 target
, journal
->j_running_transaction
?
519 journal
->j_running_transaction
->t_tid
: 0);
523 int jbd2_log_start_commit(journal_t
*journal
, tid_t tid
)
527 write_lock(&journal
->j_state_lock
);
528 ret
= __jbd2_log_start_commit(journal
, tid
);
529 write_unlock(&journal
->j_state_lock
);
534 * Force and wait any uncommitted transactions. We can only force the running
535 * transaction if we don't have an active handle, otherwise, we will deadlock.
536 * Returns: <0 in case of error,
537 * 0 if nothing to commit,
538 * 1 if transaction was successfully committed.
540 static int __jbd2_journal_force_commit(journal_t
*journal
)
542 transaction_t
*transaction
= NULL
;
544 int need_to_start
= 0, ret
= 0;
546 read_lock(&journal
->j_state_lock
);
547 if (journal
->j_running_transaction
&& !current
->journal_info
) {
548 transaction
= journal
->j_running_transaction
;
549 if (!tid_geq(journal
->j_commit_request
, transaction
->t_tid
))
551 } else if (journal
->j_committing_transaction
)
552 transaction
= journal
->j_committing_transaction
;
555 /* Nothing to commit */
556 read_unlock(&journal
->j_state_lock
);
559 tid
= transaction
->t_tid
;
560 read_unlock(&journal
->j_state_lock
);
562 jbd2_log_start_commit(journal
, tid
);
563 ret
= jbd2_log_wait_commit(journal
, tid
);
571 * Force and wait upon a commit if the calling process is not within
572 * transaction. This is used for forcing out undo-protected data which contains
573 * bitmaps, when the fs is running out of space.
575 * @journal: journal to force
576 * Returns true if progress was made.
578 int jbd2_journal_force_commit_nested(journal_t
*journal
)
582 ret
= __jbd2_journal_force_commit(journal
);
587 * int journal_force_commit() - force any uncommitted transactions
588 * @journal: journal to force
590 * Caller want unconditional commit. We can only force the running transaction
591 * if we don't have an active handle, otherwise, we will deadlock.
593 int jbd2_journal_force_commit(journal_t
*journal
)
597 J_ASSERT(!current
->journal_info
);
598 ret
= __jbd2_journal_force_commit(journal
);
605 * Start a commit of the current running transaction (if any). Returns true
606 * if a transaction is going to be committed (or is currently already
607 * committing), and fills its tid in at *ptid
609 int jbd2_journal_start_commit(journal_t
*journal
, tid_t
*ptid
)
613 write_lock(&journal
->j_state_lock
);
614 if (journal
->j_running_transaction
) {
615 tid_t tid
= journal
->j_running_transaction
->t_tid
;
617 __jbd2_log_start_commit(journal
, tid
);
618 /* There's a running transaction and we've just made sure
619 * it's commit has been scheduled. */
623 } else if (journal
->j_committing_transaction
) {
625 * If commit has been started, then we have to wait for
626 * completion of that transaction.
629 *ptid
= journal
->j_committing_transaction
->t_tid
;
632 write_unlock(&journal
->j_state_lock
);
637 * Return 1 if a given transaction has not yet sent barrier request
638 * connected with a transaction commit. If 0 is returned, transaction
639 * may or may not have sent the barrier. Used to avoid sending barrier
640 * twice in common cases.
642 int jbd2_trans_will_send_data_barrier(journal_t
*journal
, tid_t tid
)
645 transaction_t
*commit_trans
;
647 if (!(journal
->j_flags
& JBD2_BARRIER
))
649 read_lock(&journal
->j_state_lock
);
650 /* Transaction already committed? */
651 if (tid_geq(journal
->j_commit_sequence
, tid
))
653 commit_trans
= journal
->j_committing_transaction
;
654 if (!commit_trans
|| commit_trans
->t_tid
!= tid
) {
659 * Transaction is being committed and we already proceeded to
660 * submitting a flush to fs partition?
662 if (journal
->j_fs_dev
!= journal
->j_dev
) {
663 if (!commit_trans
->t_need_data_flush
||
664 commit_trans
->t_state
>= T_COMMIT_DFLUSH
)
667 if (commit_trans
->t_state
>= T_COMMIT_JFLUSH
)
672 read_unlock(&journal
->j_state_lock
);
675 EXPORT_SYMBOL(jbd2_trans_will_send_data_barrier
);
678 * Wait for a specified commit to complete.
679 * The caller may not hold the journal lock.
681 int jbd2_log_wait_commit(journal_t
*journal
, tid_t tid
)
685 read_lock(&journal
->j_state_lock
);
686 #ifdef CONFIG_PROVE_LOCKING
688 * Some callers make sure transaction is already committing and in that
689 * case we cannot block on open handles anymore. So don't warn in that
692 if (tid_gt(tid
, journal
->j_commit_sequence
) &&
693 (!journal
->j_committing_transaction
||
694 journal
->j_committing_transaction
->t_tid
!= tid
)) {
695 read_unlock(&journal
->j_state_lock
);
696 jbd2_might_wait_for_commit(journal
);
697 read_lock(&journal
->j_state_lock
);
700 #ifdef CONFIG_JBD2_DEBUG
701 if (!tid_geq(journal
->j_commit_request
, tid
)) {
703 "%s: error: j_commit_request=%u, tid=%u\n",
704 __func__
, journal
->j_commit_request
, tid
);
707 while (tid_gt(tid
, journal
->j_commit_sequence
)) {
708 jbd_debug(1, "JBD2: want %u, j_commit_sequence=%u\n",
709 tid
, journal
->j_commit_sequence
);
710 read_unlock(&journal
->j_state_lock
);
711 wake_up(&journal
->j_wait_commit
);
712 wait_event(journal
->j_wait_done_commit
,
713 !tid_gt(tid
, journal
->j_commit_sequence
));
714 read_lock(&journal
->j_state_lock
);
716 read_unlock(&journal
->j_state_lock
);
718 if (unlikely(is_journal_aborted(journal
)))
723 /* Return 1 when transaction with given tid has already committed. */
724 int jbd2_transaction_committed(journal_t
*journal
, tid_t tid
)
728 read_lock(&journal
->j_state_lock
);
729 if (journal
->j_running_transaction
&&
730 journal
->j_running_transaction
->t_tid
== tid
)
732 if (journal
->j_committing_transaction
&&
733 journal
->j_committing_transaction
->t_tid
== tid
)
735 read_unlock(&journal
->j_state_lock
);
738 EXPORT_SYMBOL(jbd2_transaction_committed
);
741 * When this function returns the transaction corresponding to tid
742 * will be completed. If the transaction has currently running, start
743 * committing that transaction before waiting for it to complete. If
744 * the transaction id is stale, it is by definition already completed,
745 * so just return SUCCESS.
747 int jbd2_complete_transaction(journal_t
*journal
, tid_t tid
)
749 int need_to_wait
= 1;
751 read_lock(&journal
->j_state_lock
);
752 if (journal
->j_running_transaction
&&
753 journal
->j_running_transaction
->t_tid
== tid
) {
754 if (journal
->j_commit_request
!= tid
) {
755 /* transaction not yet started, so request it */
756 read_unlock(&journal
->j_state_lock
);
757 jbd2_log_start_commit(journal
, tid
);
760 } else if (!(journal
->j_committing_transaction
&&
761 journal
->j_committing_transaction
->t_tid
== tid
))
763 read_unlock(&journal
->j_state_lock
);
767 return jbd2_log_wait_commit(journal
, tid
);
769 EXPORT_SYMBOL(jbd2_complete_transaction
);
772 * Log buffer allocation routines:
775 int jbd2_journal_next_log_block(journal_t
*journal
, unsigned long long *retp
)
777 unsigned long blocknr
;
779 write_lock(&journal
->j_state_lock
);
780 J_ASSERT(journal
->j_free
> 1);
782 blocknr
= journal
->j_head
;
785 if (journal
->j_head
== journal
->j_last
)
786 journal
->j_head
= journal
->j_first
;
787 write_unlock(&journal
->j_state_lock
);
788 return jbd2_journal_bmap(journal
, blocknr
, retp
);
792 * Conversion of logical to physical block numbers for the journal
794 * On external journals the journal blocks are identity-mapped, so
795 * this is a no-op. If needed, we can use j_blk_offset - everything is
798 int jbd2_journal_bmap(journal_t
*journal
, unsigned long blocknr
,
799 unsigned long long *retp
)
802 unsigned long long ret
;
804 if (journal
->j_inode
) {
805 ret
= bmap(journal
->j_inode
, blocknr
);
809 printk(KERN_ALERT
"%s: journal block not found "
810 "at offset %lu on %s\n",
811 __func__
, blocknr
, journal
->j_devname
);
813 __journal_abort_soft(journal
, err
);
816 *retp
= blocknr
; /* +journal->j_blk_offset */
822 * We play buffer_head aliasing tricks to write data/metadata blocks to
823 * the journal without copying their contents, but for journal
824 * descriptor blocks we do need to generate bona fide buffers.
826 * After the caller of jbd2_journal_get_descriptor_buffer() has finished modifying
827 * the buffer's contents they really should run flush_dcache_page(bh->b_page).
828 * But we don't bother doing that, so there will be coherency problems with
829 * mmaps of blockdevs which hold live JBD-controlled filesystems.
832 jbd2_journal_get_descriptor_buffer(transaction_t
*transaction
, int type
)
834 journal_t
*journal
= transaction
->t_journal
;
835 struct buffer_head
*bh
;
836 unsigned long long blocknr
;
837 journal_header_t
*header
;
840 err
= jbd2_journal_next_log_block(journal
, &blocknr
);
845 bh
= __getblk(journal
->j_dev
, blocknr
, journal
->j_blocksize
);
849 memset(bh
->b_data
, 0, journal
->j_blocksize
);
850 header
= (journal_header_t
*)bh
->b_data
;
851 header
->h_magic
= cpu_to_be32(JBD2_MAGIC_NUMBER
);
852 header
->h_blocktype
= cpu_to_be32(type
);
853 header
->h_sequence
= cpu_to_be32(transaction
->t_tid
);
854 set_buffer_uptodate(bh
);
856 BUFFER_TRACE(bh
, "return this buffer");
860 void jbd2_descriptor_block_csum_set(journal_t
*j
, struct buffer_head
*bh
)
862 struct jbd2_journal_block_tail
*tail
;
865 if (!jbd2_journal_has_csum_v2or3(j
))
868 tail
= (struct jbd2_journal_block_tail
*)(bh
->b_data
+ j
->j_blocksize
-
869 sizeof(struct jbd2_journal_block_tail
));
870 tail
->t_checksum
= 0;
871 csum
= jbd2_chksum(j
, j
->j_csum_seed
, bh
->b_data
, j
->j_blocksize
);
872 tail
->t_checksum
= cpu_to_be32(csum
);
876 * Return tid of the oldest transaction in the journal and block in the journal
877 * where the transaction starts.
879 * If the journal is now empty, return which will be the next transaction ID
880 * we will write and where will that transaction start.
882 * The return value is 0 if journal tail cannot be pushed any further, 1 if
885 int jbd2_journal_get_log_tail(journal_t
*journal
, tid_t
*tid
,
886 unsigned long *block
)
888 transaction_t
*transaction
;
891 read_lock(&journal
->j_state_lock
);
892 spin_lock(&journal
->j_list_lock
);
893 transaction
= journal
->j_checkpoint_transactions
;
895 *tid
= transaction
->t_tid
;
896 *block
= transaction
->t_log_start
;
897 } else if ((transaction
= journal
->j_committing_transaction
) != NULL
) {
898 *tid
= transaction
->t_tid
;
899 *block
= transaction
->t_log_start
;
900 } else if ((transaction
= journal
->j_running_transaction
) != NULL
) {
901 *tid
= transaction
->t_tid
;
902 *block
= journal
->j_head
;
904 *tid
= journal
->j_transaction_sequence
;
905 *block
= journal
->j_head
;
907 ret
= tid_gt(*tid
, journal
->j_tail_sequence
);
908 spin_unlock(&journal
->j_list_lock
);
909 read_unlock(&journal
->j_state_lock
);
915 * Update information in journal structure and in on disk journal superblock
916 * about log tail. This function does not check whether information passed in
917 * really pushes log tail further. It's responsibility of the caller to make
918 * sure provided log tail information is valid (e.g. by holding
919 * j_checkpoint_mutex all the time between computing log tail and calling this
920 * function as is the case with jbd2_cleanup_journal_tail()).
922 * Requires j_checkpoint_mutex
924 int __jbd2_update_log_tail(journal_t
*journal
, tid_t tid
, unsigned long block
)
929 BUG_ON(!mutex_is_locked(&journal
->j_checkpoint_mutex
));
932 * We cannot afford for write to remain in drive's caches since as
933 * soon as we update j_tail, next transaction can start reusing journal
934 * space and if we lose sb update during power failure we'd replay
935 * old transaction with possibly newly overwritten data.
937 ret
= jbd2_journal_update_sb_log_tail(journal
, tid
, block
,
942 write_lock(&journal
->j_state_lock
);
943 freed
= block
- journal
->j_tail
;
944 if (block
< journal
->j_tail
)
945 freed
+= journal
->j_last
- journal
->j_first
;
947 trace_jbd2_update_log_tail(journal
, tid
, block
, freed
);
949 "Cleaning journal tail from %u to %u (offset %lu), "
951 journal
->j_tail_sequence
, tid
, block
, freed
);
953 journal
->j_free
+= freed
;
954 journal
->j_tail_sequence
= tid
;
955 journal
->j_tail
= block
;
956 write_unlock(&journal
->j_state_lock
);
963 * This is a variation of __jbd2_update_log_tail which checks for validity of
964 * provided log tail and locks j_checkpoint_mutex. So it is safe against races
965 * with other threads updating log tail.
967 void jbd2_update_log_tail(journal_t
*journal
, tid_t tid
, unsigned long block
)
969 mutex_lock_io(&journal
->j_checkpoint_mutex
);
970 if (tid_gt(tid
, journal
->j_tail_sequence
))
971 __jbd2_update_log_tail(journal
, tid
, block
);
972 mutex_unlock(&journal
->j_checkpoint_mutex
);
975 struct jbd2_stats_proc_session
{
977 struct transaction_stats_s
*stats
;
982 static void *jbd2_seq_info_start(struct seq_file
*seq
, loff_t
*pos
)
984 return *pos
? NULL
: SEQ_START_TOKEN
;
987 static void *jbd2_seq_info_next(struct seq_file
*seq
, void *v
, loff_t
*pos
)
992 static int jbd2_seq_info_show(struct seq_file
*seq
, void *v
)
994 struct jbd2_stats_proc_session
*s
= seq
->private;
996 if (v
!= SEQ_START_TOKEN
)
998 seq_printf(seq
, "%lu transactions (%lu requested), "
999 "each up to %u blocks\n",
1000 s
->stats
->ts_tid
, s
->stats
->ts_requested
,
1001 s
->journal
->j_max_transaction_buffers
);
1002 if (s
->stats
->ts_tid
== 0)
1004 seq_printf(seq
, "average: \n %ums waiting for transaction\n",
1005 jiffies_to_msecs(s
->stats
->run
.rs_wait
/ s
->stats
->ts_tid
));
1006 seq_printf(seq
, " %ums request delay\n",
1007 (s
->stats
->ts_requested
== 0) ? 0 :
1008 jiffies_to_msecs(s
->stats
->run
.rs_request_delay
/
1009 s
->stats
->ts_requested
));
1010 seq_printf(seq
, " %ums running transaction\n",
1011 jiffies_to_msecs(s
->stats
->run
.rs_running
/ s
->stats
->ts_tid
));
1012 seq_printf(seq
, " %ums transaction was being locked\n",
1013 jiffies_to_msecs(s
->stats
->run
.rs_locked
/ s
->stats
->ts_tid
));
1014 seq_printf(seq
, " %ums flushing data (in ordered mode)\n",
1015 jiffies_to_msecs(s
->stats
->run
.rs_flushing
/ s
->stats
->ts_tid
));
1016 seq_printf(seq
, " %ums logging transaction\n",
1017 jiffies_to_msecs(s
->stats
->run
.rs_logging
/ s
->stats
->ts_tid
));
1018 seq_printf(seq
, " %lluus average transaction commit time\n",
1019 div_u64(s
->journal
->j_average_commit_time
, 1000));
1020 seq_printf(seq
, " %lu handles per transaction\n",
1021 s
->stats
->run
.rs_handle_count
/ s
->stats
->ts_tid
);
1022 seq_printf(seq
, " %lu blocks per transaction\n",
1023 s
->stats
->run
.rs_blocks
/ s
->stats
->ts_tid
);
1024 seq_printf(seq
, " %lu logged blocks per transaction\n",
1025 s
->stats
->run
.rs_blocks_logged
/ s
->stats
->ts_tid
);
1029 static void jbd2_seq_info_stop(struct seq_file
*seq
, void *v
)
1033 static const struct seq_operations jbd2_seq_info_ops
= {
1034 .start
= jbd2_seq_info_start
,
1035 .next
= jbd2_seq_info_next
,
1036 .stop
= jbd2_seq_info_stop
,
1037 .show
= jbd2_seq_info_show
,
1040 static int jbd2_seq_info_open(struct inode
*inode
, struct file
*file
)
1042 journal_t
*journal
= PDE_DATA(inode
);
1043 struct jbd2_stats_proc_session
*s
;
1046 s
= kmalloc(sizeof(*s
), GFP_KERNEL
);
1049 size
= sizeof(struct transaction_stats_s
);
1050 s
->stats
= kmalloc(size
, GFP_KERNEL
);
1051 if (s
->stats
== NULL
) {
1055 spin_lock(&journal
->j_history_lock
);
1056 memcpy(s
->stats
, &journal
->j_stats
, size
);
1057 s
->journal
= journal
;
1058 spin_unlock(&journal
->j_history_lock
);
1060 rc
= seq_open(file
, &jbd2_seq_info_ops
);
1062 struct seq_file
*m
= file
->private_data
;
1072 static int jbd2_seq_info_release(struct inode
*inode
, struct file
*file
)
1074 struct seq_file
*seq
= file
->private_data
;
1075 struct jbd2_stats_proc_session
*s
= seq
->private;
1078 return seq_release(inode
, file
);
1081 static const struct file_operations jbd2_seq_info_fops
= {
1082 .owner
= THIS_MODULE
,
1083 .open
= jbd2_seq_info_open
,
1085 .llseek
= seq_lseek
,
1086 .release
= jbd2_seq_info_release
,
1089 static struct proc_dir_entry
*proc_jbd2_stats
;
1091 static void jbd2_stats_proc_init(journal_t
*journal
)
1093 journal
->j_proc_entry
= proc_mkdir(journal
->j_devname
, proc_jbd2_stats
);
1094 if (journal
->j_proc_entry
) {
1095 proc_create_data("info", S_IRUGO
, journal
->j_proc_entry
,
1096 &jbd2_seq_info_fops
, journal
);
1100 static void jbd2_stats_proc_exit(journal_t
*journal
)
1102 remove_proc_entry("info", journal
->j_proc_entry
);
1103 remove_proc_entry(journal
->j_devname
, proc_jbd2_stats
);
1107 * Management for journal control blocks: functions to create and
1108 * destroy journal_t structures, and to initialise and read existing
1109 * journal blocks from disk. */
1111 /* First: create and setup a journal_t object in memory. We initialise
1112 * very few fields yet: that has to wait until we have created the
1113 * journal structures from from scratch, or loaded them from disk. */
1115 static journal_t
*journal_init_common(struct block_device
*bdev
,
1116 struct block_device
*fs_dev
,
1117 unsigned long long start
, int len
, int blocksize
)
1119 static struct lock_class_key jbd2_trans_commit_key
;
1122 struct buffer_head
*bh
;
1125 journal
= kzalloc(sizeof(*journal
), GFP_KERNEL
);
1129 init_waitqueue_head(&journal
->j_wait_transaction_locked
);
1130 init_waitqueue_head(&journal
->j_wait_done_commit
);
1131 init_waitqueue_head(&journal
->j_wait_commit
);
1132 init_waitqueue_head(&journal
->j_wait_updates
);
1133 init_waitqueue_head(&journal
->j_wait_reserved
);
1134 mutex_init(&journal
->j_barrier
);
1135 mutex_init(&journal
->j_checkpoint_mutex
);
1136 spin_lock_init(&journal
->j_revoke_lock
);
1137 spin_lock_init(&journal
->j_list_lock
);
1138 rwlock_init(&journal
->j_state_lock
);
1140 journal
->j_commit_interval
= (HZ
* JBD2_DEFAULT_MAX_COMMIT_AGE
);
1141 journal
->j_min_batch_time
= 0;
1142 journal
->j_max_batch_time
= 15000; /* 15ms */
1143 atomic_set(&journal
->j_reserved_credits
, 0);
1145 /* The journal is marked for error until we succeed with recovery! */
1146 journal
->j_flags
= JBD2_ABORT
;
1148 /* Set up a default-sized revoke table for the new mount. */
1149 err
= jbd2_journal_init_revoke(journal
, JOURNAL_REVOKE_DEFAULT_HASH
);
1153 spin_lock_init(&journal
->j_history_lock
);
1155 lockdep_init_map(&journal
->j_trans_commit_map
, "jbd2_handle",
1156 &jbd2_trans_commit_key
, 0);
1158 /* journal descriptor can store up to n blocks -bzzz */
1159 journal
->j_blocksize
= blocksize
;
1160 journal
->j_dev
= bdev
;
1161 journal
->j_fs_dev
= fs_dev
;
1162 journal
->j_blk_offset
= start
;
1163 journal
->j_maxlen
= len
;
1164 n
= journal
->j_blocksize
/ sizeof(journal_block_tag_t
);
1165 journal
->j_wbufsize
= n
;
1166 journal
->j_wbuf
= kmalloc_array(n
, sizeof(struct buffer_head
*),
1168 if (!journal
->j_wbuf
)
1171 bh
= getblk_unmovable(journal
->j_dev
, start
, journal
->j_blocksize
);
1173 pr_err("%s: Cannot get buffer for journal superblock\n",
1177 journal
->j_sb_buffer
= bh
;
1178 journal
->j_superblock
= (journal_superblock_t
*)bh
->b_data
;
1183 kfree(journal
->j_wbuf
);
1184 jbd2_journal_destroy_revoke(journal
);
1189 /* jbd2_journal_init_dev and jbd2_journal_init_inode:
1191 * Create a journal structure assigned some fixed set of disk blocks to
1192 * the journal. We don't actually touch those disk blocks yet, but we
1193 * need to set up all of the mapping information to tell the journaling
1194 * system where the journal blocks are.
1199 * journal_t * jbd2_journal_init_dev() - creates and initialises a journal structure
1200 * @bdev: Block device on which to create the journal
1201 * @fs_dev: Device which hold journalled filesystem for this journal.
1202 * @start: Block nr Start of journal.
1203 * @len: Length of the journal in blocks.
1204 * @blocksize: blocksize of journalling device
1206 * Returns: a newly created journal_t *
1208 * jbd2_journal_init_dev creates a journal which maps a fixed contiguous
1209 * range of blocks on an arbitrary block device.
1212 journal_t
*jbd2_journal_init_dev(struct block_device
*bdev
,
1213 struct block_device
*fs_dev
,
1214 unsigned long long start
, int len
, int blocksize
)
1218 journal
= journal_init_common(bdev
, fs_dev
, start
, len
, blocksize
);
1222 bdevname(journal
->j_dev
, journal
->j_devname
);
1223 strreplace(journal
->j_devname
, '/', '!');
1224 jbd2_stats_proc_init(journal
);
1230 * journal_t * jbd2_journal_init_inode () - creates a journal which maps to a inode.
1231 * @inode: An inode to create the journal in
1233 * jbd2_journal_init_inode creates a journal which maps an on-disk inode as
1234 * the journal. The inode must exist already, must support bmap() and
1235 * must have all data blocks preallocated.
1237 journal_t
*jbd2_journal_init_inode(struct inode
*inode
)
1241 unsigned long long blocknr
;
1243 blocknr
= bmap(inode
, 0);
1245 pr_err("%s: Cannot locate journal superblock\n",
1250 jbd_debug(1, "JBD2: inode %s/%ld, size %lld, bits %d, blksize %ld\n",
1251 inode
->i_sb
->s_id
, inode
->i_ino
, (long long) inode
->i_size
,
1252 inode
->i_sb
->s_blocksize_bits
, inode
->i_sb
->s_blocksize
);
1254 journal
= journal_init_common(inode
->i_sb
->s_bdev
, inode
->i_sb
->s_bdev
,
1255 blocknr
, inode
->i_size
>> inode
->i_sb
->s_blocksize_bits
,
1256 inode
->i_sb
->s_blocksize
);
1260 journal
->j_inode
= inode
;
1261 bdevname(journal
->j_dev
, journal
->j_devname
);
1262 p
= strreplace(journal
->j_devname
, '/', '!');
1263 sprintf(p
, "-%lu", journal
->j_inode
->i_ino
);
1264 jbd2_stats_proc_init(journal
);
1270 * If the journal init or create aborts, we need to mark the journal
1271 * superblock as being NULL to prevent the journal destroy from writing
1272 * back a bogus superblock.
1274 static void journal_fail_superblock (journal_t
*journal
)
1276 struct buffer_head
*bh
= journal
->j_sb_buffer
;
1278 journal
->j_sb_buffer
= NULL
;
1282 * Given a journal_t structure, initialise the various fields for
1283 * startup of a new journaling session. We use this both when creating
1284 * a journal, and after recovering an old journal to reset it for
1288 static int journal_reset(journal_t
*journal
)
1290 journal_superblock_t
*sb
= journal
->j_superblock
;
1291 unsigned long long first
, last
;
1293 first
= be32_to_cpu(sb
->s_first
);
1294 last
= be32_to_cpu(sb
->s_maxlen
);
1295 if (first
+ JBD2_MIN_JOURNAL_BLOCKS
> last
+ 1) {
1296 printk(KERN_ERR
"JBD2: Journal too short (blocks %llu-%llu).\n",
1298 journal_fail_superblock(journal
);
1302 journal
->j_first
= first
;
1303 journal
->j_last
= last
;
1305 journal
->j_head
= first
;
1306 journal
->j_tail
= first
;
1307 journal
->j_free
= last
- first
;
1309 journal
->j_tail_sequence
= journal
->j_transaction_sequence
;
1310 journal
->j_commit_sequence
= journal
->j_transaction_sequence
- 1;
1311 journal
->j_commit_request
= journal
->j_commit_sequence
;
1313 journal
->j_max_transaction_buffers
= journal
->j_maxlen
/ 4;
1316 * As a special case, if the on-disk copy is already marked as needing
1317 * no recovery (s_start == 0), then we can safely defer the superblock
1318 * update until the next commit by setting JBD2_FLUSHED. This avoids
1319 * attempting a write to a potential-readonly device.
1321 if (sb
->s_start
== 0) {
1322 jbd_debug(1, "JBD2: Skipping superblock update on recovered sb "
1323 "(start %ld, seq %u, errno %d)\n",
1324 journal
->j_tail
, journal
->j_tail_sequence
,
1326 journal
->j_flags
|= JBD2_FLUSHED
;
1328 /* Lock here to make assertions happy... */
1329 mutex_lock_io(&journal
->j_checkpoint_mutex
);
1331 * Update log tail information. We use REQ_FUA since new
1332 * transaction will start reusing journal space and so we
1333 * must make sure information about current log tail is on
1336 jbd2_journal_update_sb_log_tail(journal
,
1337 journal
->j_tail_sequence
,
1339 REQ_SYNC
| REQ_FUA
);
1340 mutex_unlock(&journal
->j_checkpoint_mutex
);
1342 return jbd2_journal_start_thread(journal
);
1346 * This function expects that the caller will have locked the journal
1347 * buffer head, and will return with it unlocked
1349 static int jbd2_write_superblock(journal_t
*journal
, int write_flags
)
1351 struct buffer_head
*bh
= journal
->j_sb_buffer
;
1352 journal_superblock_t
*sb
= journal
->j_superblock
;
1355 /* Buffer got discarded which means block device got invalidated */
1356 if (!buffer_mapped(bh
))
1359 trace_jbd2_write_superblock(journal
, write_flags
);
1360 if (!(journal
->j_flags
& JBD2_BARRIER
))
1361 write_flags
&= ~(REQ_FUA
| REQ_PREFLUSH
);
1362 if (buffer_write_io_error(bh
)) {
1364 * Oh, dear. A previous attempt to write the journal
1365 * superblock failed. This could happen because the
1366 * USB device was yanked out. Or it could happen to
1367 * be a transient write error and maybe the block will
1368 * be remapped. Nothing we can do but to retry the
1369 * write and hope for the best.
1371 printk(KERN_ERR
"JBD2: previous I/O error detected "
1372 "for journal superblock update for %s.\n",
1373 journal
->j_devname
);
1374 clear_buffer_write_io_error(bh
);
1375 set_buffer_uptodate(bh
);
1377 if (jbd2_journal_has_csum_v2or3(journal
))
1378 sb
->s_checksum
= jbd2_superblock_csum(journal
, sb
);
1380 bh
->b_end_io
= end_buffer_write_sync
;
1381 ret
= submit_bh(REQ_OP_WRITE
, write_flags
, bh
);
1383 if (buffer_write_io_error(bh
)) {
1384 clear_buffer_write_io_error(bh
);
1385 set_buffer_uptodate(bh
);
1389 printk(KERN_ERR
"JBD2: Error %d detected when updating "
1390 "journal superblock for %s.\n", ret
,
1391 journal
->j_devname
);
1392 jbd2_journal_abort(journal
, ret
);
1399 * jbd2_journal_update_sb_log_tail() - Update log tail in journal sb on disk.
1400 * @journal: The journal to update.
1401 * @tail_tid: TID of the new transaction at the tail of the log
1402 * @tail_block: The first block of the transaction at the tail of the log
1403 * @write_op: With which operation should we write the journal sb
1405 * Update a journal's superblock information about log tail and write it to
1406 * disk, waiting for the IO to complete.
1408 int jbd2_journal_update_sb_log_tail(journal_t
*journal
, tid_t tail_tid
,
1409 unsigned long tail_block
, int write_op
)
1411 journal_superblock_t
*sb
= journal
->j_superblock
;
1414 if (is_journal_aborted(journal
))
1417 BUG_ON(!mutex_is_locked(&journal
->j_checkpoint_mutex
));
1418 jbd_debug(1, "JBD2: updating superblock (start %lu, seq %u)\n",
1419 tail_block
, tail_tid
);
1421 lock_buffer(journal
->j_sb_buffer
);
1422 sb
->s_sequence
= cpu_to_be32(tail_tid
);
1423 sb
->s_start
= cpu_to_be32(tail_block
);
1425 ret
= jbd2_write_superblock(journal
, write_op
);
1429 /* Log is no longer empty */
1430 write_lock(&journal
->j_state_lock
);
1431 WARN_ON(!sb
->s_sequence
);
1432 journal
->j_flags
&= ~JBD2_FLUSHED
;
1433 write_unlock(&journal
->j_state_lock
);
1440 * jbd2_mark_journal_empty() - Mark on disk journal as empty.
1441 * @journal: The journal to update.
1442 * @write_op: With which operation should we write the journal sb
1444 * Update a journal's dynamic superblock fields to show that journal is empty.
1445 * Write updated superblock to disk waiting for IO to complete.
1447 static void jbd2_mark_journal_empty(journal_t
*journal
, int write_op
)
1449 journal_superblock_t
*sb
= journal
->j_superblock
;
1451 BUG_ON(!mutex_is_locked(&journal
->j_checkpoint_mutex
));
1452 lock_buffer(journal
->j_sb_buffer
);
1453 if (sb
->s_start
== 0) { /* Is it already empty? */
1454 unlock_buffer(journal
->j_sb_buffer
);
1458 jbd_debug(1, "JBD2: Marking journal as empty (seq %u)\n",
1459 journal
->j_tail_sequence
);
1461 sb
->s_sequence
= cpu_to_be32(journal
->j_tail_sequence
);
1462 sb
->s_start
= cpu_to_be32(0);
1464 jbd2_write_superblock(journal
, write_op
);
1466 /* Log is no longer empty */
1467 write_lock(&journal
->j_state_lock
);
1468 journal
->j_flags
|= JBD2_FLUSHED
;
1469 write_unlock(&journal
->j_state_lock
);
1474 * jbd2_journal_update_sb_errno() - Update error in the journal.
1475 * @journal: The journal to update.
1477 * Update a journal's errno. Write updated superblock to disk waiting for IO
1480 void jbd2_journal_update_sb_errno(journal_t
*journal
)
1482 journal_superblock_t
*sb
= journal
->j_superblock
;
1485 lock_buffer(journal
->j_sb_buffer
);
1486 errcode
= journal
->j_errno
;
1487 if (errcode
== -ESHUTDOWN
)
1489 jbd_debug(1, "JBD2: updating superblock error (errno %d)\n", errcode
);
1490 sb
->s_errno
= cpu_to_be32(errcode
);
1492 jbd2_write_superblock(journal
, REQ_SYNC
| REQ_FUA
);
1494 EXPORT_SYMBOL(jbd2_journal_update_sb_errno
);
1497 * Read the superblock for a given journal, performing initial
1498 * validation of the format.
1500 static int journal_get_superblock(journal_t
*journal
)
1502 struct buffer_head
*bh
;
1503 journal_superblock_t
*sb
;
1506 bh
= journal
->j_sb_buffer
;
1508 J_ASSERT(bh
!= NULL
);
1509 if (!buffer_uptodate(bh
)) {
1510 ll_rw_block(REQ_OP_READ
, 0, 1, &bh
);
1512 if (!buffer_uptodate(bh
)) {
1514 "JBD2: IO error reading journal superblock\n");
1519 if (buffer_verified(bh
))
1522 sb
= journal
->j_superblock
;
1526 if (sb
->s_header
.h_magic
!= cpu_to_be32(JBD2_MAGIC_NUMBER
) ||
1527 sb
->s_blocksize
!= cpu_to_be32(journal
->j_blocksize
)) {
1528 printk(KERN_WARNING
"JBD2: no valid journal superblock found\n");
1532 switch(be32_to_cpu(sb
->s_header
.h_blocktype
)) {
1533 case JBD2_SUPERBLOCK_V1
:
1534 journal
->j_format_version
= 1;
1536 case JBD2_SUPERBLOCK_V2
:
1537 journal
->j_format_version
= 2;
1540 printk(KERN_WARNING
"JBD2: unrecognised superblock format ID\n");
1544 if (be32_to_cpu(sb
->s_maxlen
) < journal
->j_maxlen
)
1545 journal
->j_maxlen
= be32_to_cpu(sb
->s_maxlen
);
1546 else if (be32_to_cpu(sb
->s_maxlen
) > journal
->j_maxlen
) {
1547 printk(KERN_WARNING
"JBD2: journal file too short\n");
1551 if (be32_to_cpu(sb
->s_first
) == 0 ||
1552 be32_to_cpu(sb
->s_first
) >= journal
->j_maxlen
) {
1554 "JBD2: Invalid start block of journal: %u\n",
1555 be32_to_cpu(sb
->s_first
));
1559 if (jbd2_has_feature_csum2(journal
) &&
1560 jbd2_has_feature_csum3(journal
)) {
1561 /* Can't have checksum v2 and v3 at the same time! */
1562 printk(KERN_ERR
"JBD2: Can't enable checksumming v2 and v3 "
1563 "at the same time!\n");
1567 if (jbd2_journal_has_csum_v2or3_feature(journal
) &&
1568 jbd2_has_feature_checksum(journal
)) {
1569 /* Can't have checksum v1 and v2 on at the same time! */
1570 printk(KERN_ERR
"JBD2: Can't enable checksumming v1 and v2/3 "
1571 "at the same time!\n");
1575 if (!jbd2_verify_csum_type(journal
, sb
)) {
1576 printk(KERN_ERR
"JBD2: Unknown checksum type\n");
1580 /* Load the checksum driver */
1581 if (jbd2_journal_has_csum_v2or3_feature(journal
)) {
1582 journal
->j_chksum_driver
= crypto_alloc_shash("crc32c", 0, 0);
1583 if (IS_ERR(journal
->j_chksum_driver
)) {
1584 printk(KERN_ERR
"JBD2: Cannot load crc32c driver.\n");
1585 err
= PTR_ERR(journal
->j_chksum_driver
);
1586 journal
->j_chksum_driver
= NULL
;
1591 if (jbd2_journal_has_csum_v2or3(journal
)) {
1592 /* Check superblock checksum */
1593 if (sb
->s_checksum
!= jbd2_superblock_csum(journal
, sb
)) {
1594 printk(KERN_ERR
"JBD2: journal checksum error\n");
1599 /* Precompute checksum seed for all metadata */
1600 journal
->j_csum_seed
= jbd2_chksum(journal
, ~0, sb
->s_uuid
,
1601 sizeof(sb
->s_uuid
));
1604 set_buffer_verified(bh
);
1609 journal_fail_superblock(journal
);
1614 * Load the on-disk journal superblock and read the key fields into the
1618 static int load_superblock(journal_t
*journal
)
1621 journal_superblock_t
*sb
;
1623 err
= journal_get_superblock(journal
);
1627 sb
= journal
->j_superblock
;
1629 journal
->j_tail_sequence
= be32_to_cpu(sb
->s_sequence
);
1630 journal
->j_tail
= be32_to_cpu(sb
->s_start
);
1631 journal
->j_first
= be32_to_cpu(sb
->s_first
);
1632 journal
->j_last
= be32_to_cpu(sb
->s_maxlen
);
1633 journal
->j_errno
= be32_to_cpu(sb
->s_errno
);
1640 * int jbd2_journal_load() - Read journal from disk.
1641 * @journal: Journal to act on.
1643 * Given a journal_t structure which tells us which disk blocks contain
1644 * a journal, read the journal from disk to initialise the in-memory
1647 int jbd2_journal_load(journal_t
*journal
)
1650 journal_superblock_t
*sb
;
1652 err
= load_superblock(journal
);
1656 sb
= journal
->j_superblock
;
1657 /* If this is a V2 superblock, then we have to check the
1658 * features flags on it. */
1660 if (journal
->j_format_version
>= 2) {
1661 if ((sb
->s_feature_ro_compat
&
1662 ~cpu_to_be32(JBD2_KNOWN_ROCOMPAT_FEATURES
)) ||
1663 (sb
->s_feature_incompat
&
1664 ~cpu_to_be32(JBD2_KNOWN_INCOMPAT_FEATURES
))) {
1666 "JBD2: Unrecognised features on journal\n");
1672 * Create a slab for this blocksize
1674 err
= jbd2_journal_create_slab(be32_to_cpu(sb
->s_blocksize
));
1678 /* Let the recovery code check whether it needs to recover any
1679 * data from the journal. */
1680 if (jbd2_journal_recover(journal
))
1681 goto recovery_error
;
1683 if (journal
->j_failed_commit
) {
1684 printk(KERN_ERR
"JBD2: journal transaction %u on %s "
1685 "is corrupt.\n", journal
->j_failed_commit
,
1686 journal
->j_devname
);
1687 return -EFSCORRUPTED
;
1690 /* OK, we've finished with the dynamic journal bits:
1691 * reinitialise the dynamic contents of the superblock in memory
1692 * and reset them on disk. */
1693 if (journal_reset(journal
))
1694 goto recovery_error
;
1696 journal
->j_flags
&= ~JBD2_ABORT
;
1697 journal
->j_flags
|= JBD2_LOADED
;
1701 printk(KERN_WARNING
"JBD2: recovery failed\n");
1706 * void jbd2_journal_destroy() - Release a journal_t structure.
1707 * @journal: Journal to act on.
1709 * Release a journal_t structure once it is no longer in use by the
1711 * Return <0 if we couldn't clean up the journal.
1713 int jbd2_journal_destroy(journal_t
*journal
)
1717 /* Wait for the commit thread to wake up and die. */
1718 journal_kill_thread(journal
);
1720 /* Force a final log commit */
1721 if (journal
->j_running_transaction
)
1722 jbd2_journal_commit_transaction(journal
);
1724 /* Force any old transactions to disk */
1726 /* Totally anal locking here... */
1727 spin_lock(&journal
->j_list_lock
);
1728 while (journal
->j_checkpoint_transactions
!= NULL
) {
1729 spin_unlock(&journal
->j_list_lock
);
1730 mutex_lock_io(&journal
->j_checkpoint_mutex
);
1731 err
= jbd2_log_do_checkpoint(journal
);
1732 mutex_unlock(&journal
->j_checkpoint_mutex
);
1734 * If checkpointing failed, just free the buffers to avoid
1738 jbd2_journal_destroy_checkpoint(journal
);
1739 spin_lock(&journal
->j_list_lock
);
1742 spin_lock(&journal
->j_list_lock
);
1745 J_ASSERT(journal
->j_running_transaction
== NULL
);
1746 J_ASSERT(journal
->j_committing_transaction
== NULL
);
1747 J_ASSERT(journal
->j_checkpoint_transactions
== NULL
);
1748 spin_unlock(&journal
->j_list_lock
);
1750 if (journal
->j_sb_buffer
) {
1751 if (!is_journal_aborted(journal
)) {
1752 mutex_lock_io(&journal
->j_checkpoint_mutex
);
1754 write_lock(&journal
->j_state_lock
);
1755 journal
->j_tail_sequence
=
1756 ++journal
->j_transaction_sequence
;
1757 write_unlock(&journal
->j_state_lock
);
1759 jbd2_mark_journal_empty(journal
,
1760 REQ_SYNC
| REQ_PREFLUSH
| REQ_FUA
);
1761 mutex_unlock(&journal
->j_checkpoint_mutex
);
1764 brelse(journal
->j_sb_buffer
);
1767 if (journal
->j_proc_entry
)
1768 jbd2_stats_proc_exit(journal
);
1769 iput(journal
->j_inode
);
1770 if (journal
->j_revoke
)
1771 jbd2_journal_destroy_revoke(journal
);
1772 if (journal
->j_chksum_driver
)
1773 crypto_free_shash(journal
->j_chksum_driver
);
1774 kfree(journal
->j_wbuf
);
1782 *int jbd2_journal_check_used_features () - Check if features specified are used.
1783 * @journal: Journal to check.
1784 * @compat: bitmask of compatible features
1785 * @ro: bitmask of features that force read-only mount
1786 * @incompat: bitmask of incompatible features
1788 * Check whether the journal uses all of a given set of
1789 * features. Return true (non-zero) if it does.
1792 int jbd2_journal_check_used_features (journal_t
*journal
, unsigned long compat
,
1793 unsigned long ro
, unsigned long incompat
)
1795 journal_superblock_t
*sb
;
1797 if (!compat
&& !ro
&& !incompat
)
1799 /* Load journal superblock if it is not loaded yet. */
1800 if (journal
->j_format_version
== 0 &&
1801 journal_get_superblock(journal
) != 0)
1803 if (journal
->j_format_version
== 1)
1806 sb
= journal
->j_superblock
;
1808 if (((be32_to_cpu(sb
->s_feature_compat
) & compat
) == compat
) &&
1809 ((be32_to_cpu(sb
->s_feature_ro_compat
) & ro
) == ro
) &&
1810 ((be32_to_cpu(sb
->s_feature_incompat
) & incompat
) == incompat
))
1817 * int jbd2_journal_check_available_features() - Check feature set in journalling layer
1818 * @journal: Journal to check.
1819 * @compat: bitmask of compatible features
1820 * @ro: bitmask of features that force read-only mount
1821 * @incompat: bitmask of incompatible features
1823 * Check whether the journaling code supports the use of
1824 * all of a given set of features on this journal. Return true
1825 * (non-zero) if it can. */
1827 int jbd2_journal_check_available_features (journal_t
*journal
, unsigned long compat
,
1828 unsigned long ro
, unsigned long incompat
)
1830 if (!compat
&& !ro
&& !incompat
)
1833 /* We can support any known requested features iff the
1834 * superblock is in version 2. Otherwise we fail to support any
1835 * extended sb features. */
1837 if (journal
->j_format_version
!= 2)
1840 if ((compat
& JBD2_KNOWN_COMPAT_FEATURES
) == compat
&&
1841 (ro
& JBD2_KNOWN_ROCOMPAT_FEATURES
) == ro
&&
1842 (incompat
& JBD2_KNOWN_INCOMPAT_FEATURES
) == incompat
)
1849 * int jbd2_journal_set_features () - Mark a given journal feature in the superblock
1850 * @journal: Journal to act on.
1851 * @compat: bitmask of compatible features
1852 * @ro: bitmask of features that force read-only mount
1853 * @incompat: bitmask of incompatible features
1855 * Mark a given journal feature as present on the
1856 * superblock. Returns true if the requested features could be set.
1860 int jbd2_journal_set_features (journal_t
*journal
, unsigned long compat
,
1861 unsigned long ro
, unsigned long incompat
)
1863 #define INCOMPAT_FEATURE_ON(f) \
1864 ((incompat & (f)) && !(sb->s_feature_incompat & cpu_to_be32(f)))
1865 #define COMPAT_FEATURE_ON(f) \
1866 ((compat & (f)) && !(sb->s_feature_compat & cpu_to_be32(f)))
1867 journal_superblock_t
*sb
;
1869 if (jbd2_journal_check_used_features(journal
, compat
, ro
, incompat
))
1872 if (!jbd2_journal_check_available_features(journal
, compat
, ro
, incompat
))
1875 /* If enabling v2 checksums, turn on v3 instead */
1876 if (incompat
& JBD2_FEATURE_INCOMPAT_CSUM_V2
) {
1877 incompat
&= ~JBD2_FEATURE_INCOMPAT_CSUM_V2
;
1878 incompat
|= JBD2_FEATURE_INCOMPAT_CSUM_V3
;
1881 /* Asking for checksumming v3 and v1? Only give them v3. */
1882 if (incompat
& JBD2_FEATURE_INCOMPAT_CSUM_V3
&&
1883 compat
& JBD2_FEATURE_COMPAT_CHECKSUM
)
1884 compat
&= ~JBD2_FEATURE_COMPAT_CHECKSUM
;
1886 jbd_debug(1, "Setting new features 0x%lx/0x%lx/0x%lx\n",
1887 compat
, ro
, incompat
);
1889 sb
= journal
->j_superblock
;
1891 /* Load the checksum driver if necessary */
1892 if ((journal
->j_chksum_driver
== NULL
) &&
1893 INCOMPAT_FEATURE_ON(JBD2_FEATURE_INCOMPAT_CSUM_V3
)) {
1894 journal
->j_chksum_driver
= crypto_alloc_shash("crc32c", 0, 0);
1895 if (IS_ERR(journal
->j_chksum_driver
)) {
1896 printk(KERN_ERR
"JBD2: Cannot load crc32c driver.\n");
1897 journal
->j_chksum_driver
= NULL
;
1900 /* Precompute checksum seed for all metadata */
1901 journal
->j_csum_seed
= jbd2_chksum(journal
, ~0, sb
->s_uuid
,
1902 sizeof(sb
->s_uuid
));
1905 lock_buffer(journal
->j_sb_buffer
);
1907 /* If enabling v3 checksums, update superblock */
1908 if (INCOMPAT_FEATURE_ON(JBD2_FEATURE_INCOMPAT_CSUM_V3
)) {
1909 sb
->s_checksum_type
= JBD2_CRC32C_CHKSUM
;
1910 sb
->s_feature_compat
&=
1911 ~cpu_to_be32(JBD2_FEATURE_COMPAT_CHECKSUM
);
1914 /* If enabling v1 checksums, downgrade superblock */
1915 if (COMPAT_FEATURE_ON(JBD2_FEATURE_COMPAT_CHECKSUM
))
1916 sb
->s_feature_incompat
&=
1917 ~cpu_to_be32(JBD2_FEATURE_INCOMPAT_CSUM_V2
|
1918 JBD2_FEATURE_INCOMPAT_CSUM_V3
);
1920 sb
->s_feature_compat
|= cpu_to_be32(compat
);
1921 sb
->s_feature_ro_compat
|= cpu_to_be32(ro
);
1922 sb
->s_feature_incompat
|= cpu_to_be32(incompat
);
1923 unlock_buffer(journal
->j_sb_buffer
);
1926 #undef COMPAT_FEATURE_ON
1927 #undef INCOMPAT_FEATURE_ON
1931 * jbd2_journal_clear_features () - Clear a given journal feature in the
1933 * @journal: Journal to act on.
1934 * @compat: bitmask of compatible features
1935 * @ro: bitmask of features that force read-only mount
1936 * @incompat: bitmask of incompatible features
1938 * Clear a given journal feature as present on the
1941 void jbd2_journal_clear_features(journal_t
*journal
, unsigned long compat
,
1942 unsigned long ro
, unsigned long incompat
)
1944 journal_superblock_t
*sb
;
1946 jbd_debug(1, "Clear features 0x%lx/0x%lx/0x%lx\n",
1947 compat
, ro
, incompat
);
1949 sb
= journal
->j_superblock
;
1951 sb
->s_feature_compat
&= ~cpu_to_be32(compat
);
1952 sb
->s_feature_ro_compat
&= ~cpu_to_be32(ro
);
1953 sb
->s_feature_incompat
&= ~cpu_to_be32(incompat
);
1955 EXPORT_SYMBOL(jbd2_journal_clear_features
);
1958 * int jbd2_journal_flush () - Flush journal
1959 * @journal: Journal to act on.
1961 * Flush all data for a given journal to disk and empty the journal.
1962 * Filesystems can use this when remounting readonly to ensure that
1963 * recovery does not need to happen on remount.
1966 int jbd2_journal_flush(journal_t
*journal
)
1969 transaction_t
*transaction
= NULL
;
1971 write_lock(&journal
->j_state_lock
);
1973 /* Force everything buffered to the log... */
1974 if (journal
->j_running_transaction
) {
1975 transaction
= journal
->j_running_transaction
;
1976 __jbd2_log_start_commit(journal
, transaction
->t_tid
);
1977 } else if (journal
->j_committing_transaction
)
1978 transaction
= journal
->j_committing_transaction
;
1980 /* Wait for the log commit to complete... */
1982 tid_t tid
= transaction
->t_tid
;
1984 write_unlock(&journal
->j_state_lock
);
1985 jbd2_log_wait_commit(journal
, tid
);
1987 write_unlock(&journal
->j_state_lock
);
1990 /* ...and flush everything in the log out to disk. */
1991 spin_lock(&journal
->j_list_lock
);
1992 while (!err
&& journal
->j_checkpoint_transactions
!= NULL
) {
1993 spin_unlock(&journal
->j_list_lock
);
1994 mutex_lock_io(&journal
->j_checkpoint_mutex
);
1995 err
= jbd2_log_do_checkpoint(journal
);
1996 mutex_unlock(&journal
->j_checkpoint_mutex
);
1997 spin_lock(&journal
->j_list_lock
);
1999 spin_unlock(&journal
->j_list_lock
);
2001 if (is_journal_aborted(journal
))
2004 mutex_lock_io(&journal
->j_checkpoint_mutex
);
2006 err
= jbd2_cleanup_journal_tail(journal
);
2008 mutex_unlock(&journal
->j_checkpoint_mutex
);
2014 /* Finally, mark the journal as really needing no recovery.
2015 * This sets s_start==0 in the underlying superblock, which is
2016 * the magic code for a fully-recovered superblock. Any future
2017 * commits of data to the journal will restore the current
2019 jbd2_mark_journal_empty(journal
, REQ_SYNC
| REQ_FUA
);
2020 mutex_unlock(&journal
->j_checkpoint_mutex
);
2021 write_lock(&journal
->j_state_lock
);
2022 J_ASSERT(!journal
->j_running_transaction
);
2023 J_ASSERT(!journal
->j_committing_transaction
);
2024 J_ASSERT(!journal
->j_checkpoint_transactions
);
2025 J_ASSERT(journal
->j_head
== journal
->j_tail
);
2026 J_ASSERT(journal
->j_tail_sequence
== journal
->j_transaction_sequence
);
2027 write_unlock(&journal
->j_state_lock
);
2033 * int jbd2_journal_wipe() - Wipe journal contents
2034 * @journal: Journal to act on.
2035 * @write: flag (see below)
2037 * Wipe out all of the contents of a journal, safely. This will produce
2038 * a warning if the journal contains any valid recovery information.
2039 * Must be called between journal_init_*() and jbd2_journal_load().
2041 * If 'write' is non-zero, then we wipe out the journal on disk; otherwise
2042 * we merely suppress recovery.
2045 int jbd2_journal_wipe(journal_t
*journal
, int write
)
2049 J_ASSERT (!(journal
->j_flags
& JBD2_LOADED
));
2051 err
= load_superblock(journal
);
2055 if (!journal
->j_tail
)
2058 printk(KERN_WARNING
"JBD2: %s recovery information on journal\n",
2059 write
? "Clearing" : "Ignoring");
2061 err
= jbd2_journal_skip_recovery(journal
);
2063 /* Lock to make assertions happy... */
2064 mutex_lock_io(&journal
->j_checkpoint_mutex
);
2065 jbd2_mark_journal_empty(journal
, REQ_SYNC
| REQ_FUA
);
2066 mutex_unlock(&journal
->j_checkpoint_mutex
);
2074 * Journal abort has very specific semantics, which we describe
2075 * for journal abort.
2077 * Two internal functions, which provide abort to the jbd layer
2082 * Quick version for internal journal use (doesn't lock the journal).
2083 * Aborts hard --- we mark the abort as occurred, but do _nothing_ else,
2084 * and don't attempt to make any other journal updates.
2086 void __jbd2_journal_abort_hard(journal_t
*journal
)
2088 transaction_t
*transaction
;
2090 if (journal
->j_flags
& JBD2_ABORT
)
2093 printk(KERN_ERR
"Aborting journal on device %s.\n",
2094 journal
->j_devname
);
2096 write_lock(&journal
->j_state_lock
);
2097 journal
->j_flags
|= JBD2_ABORT
;
2098 transaction
= journal
->j_running_transaction
;
2100 __jbd2_log_start_commit(journal
, transaction
->t_tid
);
2101 write_unlock(&journal
->j_state_lock
);
2104 /* Soft abort: record the abort error status in the journal superblock,
2105 * but don't do any other IO. */
2106 static void __journal_abort_soft (journal_t
*journal
, int errno
)
2110 write_lock(&journal
->j_state_lock
);
2111 old_errno
= journal
->j_errno
;
2112 if (!journal
->j_errno
|| errno
== -ESHUTDOWN
)
2113 journal
->j_errno
= errno
;
2115 if (journal
->j_flags
& JBD2_ABORT
) {
2116 write_unlock(&journal
->j_state_lock
);
2117 if (!old_errno
&& old_errno
!= -ESHUTDOWN
&&
2118 errno
== -ESHUTDOWN
)
2119 jbd2_journal_update_sb_errno(journal
);
2122 write_unlock(&journal
->j_state_lock
);
2124 __jbd2_journal_abort_hard(journal
);
2127 jbd2_journal_update_sb_errno(journal
);
2128 write_lock(&journal
->j_state_lock
);
2129 journal
->j_flags
|= JBD2_REC_ERR
;
2130 write_unlock(&journal
->j_state_lock
);
2135 * void jbd2_journal_abort () - Shutdown the journal immediately.
2136 * @journal: the journal to shutdown.
2137 * @errno: an error number to record in the journal indicating
2138 * the reason for the shutdown.
2140 * Perform a complete, immediate shutdown of the ENTIRE
2141 * journal (not of a single transaction). This operation cannot be
2142 * undone without closing and reopening the journal.
2144 * The jbd2_journal_abort function is intended to support higher level error
2145 * recovery mechanisms such as the ext2/ext3 remount-readonly error
2148 * Journal abort has very specific semantics. Any existing dirty,
2149 * unjournaled buffers in the main filesystem will still be written to
2150 * disk by bdflush, but the journaling mechanism will be suspended
2151 * immediately and no further transaction commits will be honoured.
2153 * Any dirty, journaled buffers will be written back to disk without
2154 * hitting the journal. Atomicity cannot be guaranteed on an aborted
2155 * filesystem, but we _do_ attempt to leave as much data as possible
2156 * behind for fsck to use for cleanup.
2158 * Any attempt to get a new transaction handle on a journal which is in
2159 * ABORT state will just result in an -EROFS error return. A
2160 * jbd2_journal_stop on an existing handle will return -EIO if we have
2161 * entered abort state during the update.
2163 * Recursive transactions are not disturbed by journal abort until the
2164 * final jbd2_journal_stop, which will receive the -EIO error.
2166 * Finally, the jbd2_journal_abort call allows the caller to supply an errno
2167 * which will be recorded (if possible) in the journal superblock. This
2168 * allows a client to record failure conditions in the middle of a
2169 * transaction without having to complete the transaction to record the
2170 * failure to disk. ext3_error, for example, now uses this
2173 * Errors which originate from within the journaling layer will NOT
2174 * supply an errno; a null errno implies that absolutely no further
2175 * writes are done to the journal (unless there are any already in
2180 void jbd2_journal_abort(journal_t
*journal
, int errno
)
2182 __journal_abort_soft(journal
, errno
);
2186 * int jbd2_journal_errno () - returns the journal's error state.
2187 * @journal: journal to examine.
2189 * This is the errno number set with jbd2_journal_abort(), the last
2190 * time the journal was mounted - if the journal was stopped
2191 * without calling abort this will be 0.
2193 * If the journal has been aborted on this mount time -EROFS will
2196 int jbd2_journal_errno(journal_t
*journal
)
2200 read_lock(&journal
->j_state_lock
);
2201 if (journal
->j_flags
& JBD2_ABORT
)
2204 err
= journal
->j_errno
;
2205 read_unlock(&journal
->j_state_lock
);
2210 * int jbd2_journal_clear_err () - clears the journal's error state
2211 * @journal: journal to act on.
2213 * An error must be cleared or acked to take a FS out of readonly
2216 int jbd2_journal_clear_err(journal_t
*journal
)
2220 write_lock(&journal
->j_state_lock
);
2221 if (journal
->j_flags
& JBD2_ABORT
)
2224 journal
->j_errno
= 0;
2225 write_unlock(&journal
->j_state_lock
);
2230 * void jbd2_journal_ack_err() - Ack journal err.
2231 * @journal: journal to act on.
2233 * An error must be cleared or acked to take a FS out of readonly
2236 void jbd2_journal_ack_err(journal_t
*journal
)
2238 write_lock(&journal
->j_state_lock
);
2239 if (journal
->j_errno
)
2240 journal
->j_flags
|= JBD2_ACK_ERR
;
2241 write_unlock(&journal
->j_state_lock
);
2244 int jbd2_journal_blocks_per_page(struct inode
*inode
)
2246 return 1 << (PAGE_SHIFT
- inode
->i_sb
->s_blocksize_bits
);
2250 * helper functions to deal with 32 or 64bit block numbers.
2252 size_t journal_tag_bytes(journal_t
*journal
)
2256 if (jbd2_has_feature_csum3(journal
))
2257 return sizeof(journal_block_tag3_t
);
2259 sz
= sizeof(journal_block_tag_t
);
2261 if (jbd2_has_feature_csum2(journal
))
2262 sz
+= sizeof(__u16
);
2264 if (jbd2_has_feature_64bit(journal
))
2267 return sz
- sizeof(__u32
);
2271 * JBD memory management
2273 * These functions are used to allocate block-sized chunks of memory
2274 * used for making copies of buffer_head data. Very often it will be
2275 * page-sized chunks of data, but sometimes it will be in
2276 * sub-page-size chunks. (For example, 16k pages on Power systems
2277 * with a 4k block file system.) For blocks smaller than a page, we
2278 * use a SLAB allocator. There are slab caches for each block size,
2279 * which are allocated at mount time, if necessary, and we only free
2280 * (all of) the slab caches when/if the jbd2 module is unloaded. For
2281 * this reason we don't need to a mutex to protect access to
2282 * jbd2_slab[] allocating or releasing memory; only in
2283 * jbd2_journal_create_slab().
2285 #define JBD2_MAX_SLABS 8
2286 static struct kmem_cache
*jbd2_slab
[JBD2_MAX_SLABS
];
2288 static const char *jbd2_slab_names
[JBD2_MAX_SLABS
] = {
2289 "jbd2_1k", "jbd2_2k", "jbd2_4k", "jbd2_8k",
2290 "jbd2_16k", "jbd2_32k", "jbd2_64k", "jbd2_128k"
2294 static void jbd2_journal_destroy_slabs(void)
2298 for (i
= 0; i
< JBD2_MAX_SLABS
; i
++) {
2299 kmem_cache_destroy(jbd2_slab
[i
]);
2300 jbd2_slab
[i
] = NULL
;
2304 static int jbd2_journal_create_slab(size_t size
)
2306 static DEFINE_MUTEX(jbd2_slab_create_mutex
);
2307 int i
= order_base_2(size
) - 10;
2310 if (size
== PAGE_SIZE
)
2313 if (i
>= JBD2_MAX_SLABS
)
2316 if (unlikely(i
< 0))
2318 mutex_lock(&jbd2_slab_create_mutex
);
2320 mutex_unlock(&jbd2_slab_create_mutex
);
2321 return 0; /* Already created */
2324 slab_size
= 1 << (i
+10);
2325 jbd2_slab
[i
] = kmem_cache_create(jbd2_slab_names
[i
], slab_size
,
2326 slab_size
, 0, NULL
);
2327 mutex_unlock(&jbd2_slab_create_mutex
);
2328 if (!jbd2_slab
[i
]) {
2329 printk(KERN_EMERG
"JBD2: no memory for jbd2_slab cache\n");
2335 static struct kmem_cache
*get_slab(size_t size
)
2337 int i
= order_base_2(size
) - 10;
2339 BUG_ON(i
>= JBD2_MAX_SLABS
);
2340 if (unlikely(i
< 0))
2342 BUG_ON(jbd2_slab
[i
] == NULL
);
2343 return jbd2_slab
[i
];
2346 void *jbd2_alloc(size_t size
, gfp_t flags
)
2350 BUG_ON(size
& (size
-1)); /* Must be a power of 2 */
2352 if (size
< PAGE_SIZE
)
2353 ptr
= kmem_cache_alloc(get_slab(size
), flags
);
2355 ptr
= (void *)__get_free_pages(flags
, get_order(size
));
2357 /* Check alignment; SLUB has gotten this wrong in the past,
2358 * and this can lead to user data corruption! */
2359 BUG_ON(((unsigned long) ptr
) & (size
-1));
2364 void jbd2_free(void *ptr
, size_t size
)
2366 if (size
< PAGE_SIZE
)
2367 kmem_cache_free(get_slab(size
), ptr
);
2369 free_pages((unsigned long)ptr
, get_order(size
));
2373 * Journal_head storage management
2375 static struct kmem_cache
*jbd2_journal_head_cache
;
2376 #ifdef CONFIG_JBD2_DEBUG
2377 static atomic_t nr_journal_heads
= ATOMIC_INIT(0);
2380 static int __init
jbd2_journal_init_journal_head_cache(void)
2382 J_ASSERT(!jbd2_journal_head_cache
);
2383 jbd2_journal_head_cache
= kmem_cache_create("jbd2_journal_head",
2384 sizeof(struct journal_head
),
2386 SLAB_TEMPORARY
| SLAB_TYPESAFE_BY_RCU
,
2388 if (!jbd2_journal_head_cache
) {
2389 printk(KERN_EMERG
"JBD2: no memory for journal_head cache\n");
2395 static void jbd2_journal_destroy_journal_head_cache(void)
2397 kmem_cache_destroy(jbd2_journal_head_cache
);
2398 jbd2_journal_head_cache
= NULL
;
2402 * journal_head splicing and dicing
2404 static struct journal_head
*journal_alloc_journal_head(void)
2406 struct journal_head
*ret
;
2408 #ifdef CONFIG_JBD2_DEBUG
2409 atomic_inc(&nr_journal_heads
);
2411 ret
= kmem_cache_zalloc(jbd2_journal_head_cache
, GFP_NOFS
);
2413 jbd_debug(1, "out of memory for journal_head\n");
2414 pr_notice_ratelimited("ENOMEM in %s, retrying.\n", __func__
);
2415 ret
= kmem_cache_zalloc(jbd2_journal_head_cache
,
2416 GFP_NOFS
| __GFP_NOFAIL
);
2421 static void journal_free_journal_head(struct journal_head
*jh
)
2423 #ifdef CONFIG_JBD2_DEBUG
2424 atomic_dec(&nr_journal_heads
);
2425 memset(jh
, JBD2_POISON_FREE
, sizeof(*jh
));
2427 kmem_cache_free(jbd2_journal_head_cache
, jh
);
2431 * A journal_head is attached to a buffer_head whenever JBD has an
2432 * interest in the buffer.
2434 * Whenever a buffer has an attached journal_head, its ->b_state:BH_JBD bit
2435 * is set. This bit is tested in core kernel code where we need to take
2436 * JBD-specific actions. Testing the zeroness of ->b_private is not reliable
2439 * When a buffer has its BH_JBD bit set, its ->b_count is elevated by one.
2441 * When a buffer has its BH_JBD bit set it is immune from being released by
2442 * core kernel code, mainly via ->b_count.
2444 * A journal_head is detached from its buffer_head when the journal_head's
2445 * b_jcount reaches zero. Running transaction (b_transaction) and checkpoint
2446 * transaction (b_cp_transaction) hold their references to b_jcount.
2448 * Various places in the kernel want to attach a journal_head to a buffer_head
2449 * _before_ attaching the journal_head to a transaction. To protect the
2450 * journal_head in this situation, jbd2_journal_add_journal_head elevates the
2451 * journal_head's b_jcount refcount by one. The caller must call
2452 * jbd2_journal_put_journal_head() to undo this.
2454 * So the typical usage would be:
2456 * (Attach a journal_head if needed. Increments b_jcount)
2457 * struct journal_head *jh = jbd2_journal_add_journal_head(bh);
2459 * (Get another reference for transaction)
2460 * jbd2_journal_grab_journal_head(bh);
2461 * jh->b_transaction = xxx;
2462 * (Put original reference)
2463 * jbd2_journal_put_journal_head(jh);
2467 * Give a buffer_head a journal_head.
2471 struct journal_head
*jbd2_journal_add_journal_head(struct buffer_head
*bh
)
2473 struct journal_head
*jh
;
2474 struct journal_head
*new_jh
= NULL
;
2477 if (!buffer_jbd(bh
))
2478 new_jh
= journal_alloc_journal_head();
2480 jbd_lock_bh_journal_head(bh
);
2481 if (buffer_jbd(bh
)) {
2485 (atomic_read(&bh
->b_count
) > 0) ||
2486 (bh
->b_page
&& bh
->b_page
->mapping
));
2489 jbd_unlock_bh_journal_head(bh
);
2494 new_jh
= NULL
; /* We consumed it */
2499 BUFFER_TRACE(bh
, "added journal_head");
2502 jbd_unlock_bh_journal_head(bh
);
2504 journal_free_journal_head(new_jh
);
2505 return bh
->b_private
;
2509 * Grab a ref against this buffer_head's journal_head. If it ended up not
2510 * having a journal_head, return NULL
2512 struct journal_head
*jbd2_journal_grab_journal_head(struct buffer_head
*bh
)
2514 struct journal_head
*jh
= NULL
;
2516 jbd_lock_bh_journal_head(bh
);
2517 if (buffer_jbd(bh
)) {
2521 jbd_unlock_bh_journal_head(bh
);
2525 static void __journal_remove_journal_head(struct buffer_head
*bh
)
2527 struct journal_head
*jh
= bh2jh(bh
);
2529 J_ASSERT_JH(jh
, jh
->b_jcount
>= 0);
2530 J_ASSERT_JH(jh
, jh
->b_transaction
== NULL
);
2531 J_ASSERT_JH(jh
, jh
->b_next_transaction
== NULL
);
2532 J_ASSERT_JH(jh
, jh
->b_cp_transaction
== NULL
);
2533 J_ASSERT_JH(jh
, jh
->b_jlist
== BJ_None
);
2534 J_ASSERT_BH(bh
, buffer_jbd(bh
));
2535 J_ASSERT_BH(bh
, jh2bh(jh
) == bh
);
2536 BUFFER_TRACE(bh
, "remove journal_head");
2537 if (jh
->b_frozen_data
) {
2538 printk(KERN_WARNING
"%s: freeing b_frozen_data\n", __func__
);
2539 jbd2_free(jh
->b_frozen_data
, bh
->b_size
);
2541 if (jh
->b_committed_data
) {
2542 printk(KERN_WARNING
"%s: freeing b_committed_data\n", __func__
);
2543 jbd2_free(jh
->b_committed_data
, bh
->b_size
);
2545 bh
->b_private
= NULL
;
2546 jh
->b_bh
= NULL
; /* debug, really */
2547 clear_buffer_jbd(bh
);
2548 journal_free_journal_head(jh
);
2552 * Drop a reference on the passed journal_head. If it fell to zero then
2553 * release the journal_head from the buffer_head.
2555 void jbd2_journal_put_journal_head(struct journal_head
*jh
)
2557 struct buffer_head
*bh
= jh2bh(jh
);
2559 jbd_lock_bh_journal_head(bh
);
2560 J_ASSERT_JH(jh
, jh
->b_jcount
> 0);
2562 if (!jh
->b_jcount
) {
2563 __journal_remove_journal_head(bh
);
2564 jbd_unlock_bh_journal_head(bh
);
2567 jbd_unlock_bh_journal_head(bh
);
2571 * Initialize jbd inode head
2573 void jbd2_journal_init_jbd_inode(struct jbd2_inode
*jinode
, struct inode
*inode
)
2575 jinode
->i_transaction
= NULL
;
2576 jinode
->i_next_transaction
= NULL
;
2577 jinode
->i_vfs_inode
= inode
;
2578 jinode
->i_flags
= 0;
2579 jinode
->i_dirty_start
= 0;
2580 jinode
->i_dirty_end
= 0;
2581 INIT_LIST_HEAD(&jinode
->i_list
);
2585 * Function to be called before we start removing inode from memory (i.e.,
2586 * clear_inode() is a fine place to be called from). It removes inode from
2587 * transaction's lists.
2589 void jbd2_journal_release_jbd_inode(journal_t
*journal
,
2590 struct jbd2_inode
*jinode
)
2595 spin_lock(&journal
->j_list_lock
);
2596 /* Is commit writing out inode - we have to wait */
2597 if (jinode
->i_flags
& JI_COMMIT_RUNNING
) {
2598 wait_queue_head_t
*wq
;
2599 DEFINE_WAIT_BIT(wait
, &jinode
->i_flags
, __JI_COMMIT_RUNNING
);
2600 wq
= bit_waitqueue(&jinode
->i_flags
, __JI_COMMIT_RUNNING
);
2601 prepare_to_wait(wq
, &wait
.wq_entry
, TASK_UNINTERRUPTIBLE
);
2602 spin_unlock(&journal
->j_list_lock
);
2604 finish_wait(wq
, &wait
.wq_entry
);
2608 if (jinode
->i_transaction
) {
2609 list_del(&jinode
->i_list
);
2610 jinode
->i_transaction
= NULL
;
2612 spin_unlock(&journal
->j_list_lock
);
2616 #ifdef CONFIG_PROC_FS
2618 #define JBD2_STATS_PROC_NAME "fs/jbd2"
2620 static void __init
jbd2_create_jbd_stats_proc_entry(void)
2622 proc_jbd2_stats
= proc_mkdir(JBD2_STATS_PROC_NAME
, NULL
);
2625 static void __exit
jbd2_remove_jbd_stats_proc_entry(void)
2627 if (proc_jbd2_stats
)
2628 remove_proc_entry(JBD2_STATS_PROC_NAME
, NULL
);
2633 #define jbd2_create_jbd_stats_proc_entry() do {} while (0)
2634 #define jbd2_remove_jbd_stats_proc_entry() do {} while (0)
2638 struct kmem_cache
*jbd2_handle_cache
, *jbd2_inode_cache
;
2640 static int __init
jbd2_journal_init_inode_cache(void)
2642 J_ASSERT(!jbd2_inode_cache
);
2643 jbd2_inode_cache
= KMEM_CACHE(jbd2_inode
, 0);
2644 if (!jbd2_inode_cache
) {
2645 pr_emerg("JBD2: failed to create inode cache\n");
2651 static int __init
jbd2_journal_init_handle_cache(void)
2653 J_ASSERT(!jbd2_handle_cache
);
2654 jbd2_handle_cache
= KMEM_CACHE(jbd2_journal_handle
, SLAB_TEMPORARY
);
2655 if (!jbd2_handle_cache
) {
2656 printk(KERN_EMERG
"JBD2: failed to create handle cache\n");
2662 static void jbd2_journal_destroy_inode_cache(void)
2664 kmem_cache_destroy(jbd2_inode_cache
);
2665 jbd2_inode_cache
= NULL
;
2668 static void jbd2_journal_destroy_handle_cache(void)
2670 kmem_cache_destroy(jbd2_handle_cache
);
2671 jbd2_handle_cache
= NULL
;
2675 * Module startup and shutdown
2678 static int __init
journal_init_caches(void)
2682 ret
= jbd2_journal_init_revoke_record_cache();
2684 ret
= jbd2_journal_init_revoke_table_cache();
2686 ret
= jbd2_journal_init_journal_head_cache();
2688 ret
= jbd2_journal_init_handle_cache();
2690 ret
= jbd2_journal_init_inode_cache();
2692 ret
= jbd2_journal_init_transaction_cache();
2696 static void jbd2_journal_destroy_caches(void)
2698 jbd2_journal_destroy_revoke_record_cache();
2699 jbd2_journal_destroy_revoke_table_cache();
2700 jbd2_journal_destroy_journal_head_cache();
2701 jbd2_journal_destroy_handle_cache();
2702 jbd2_journal_destroy_inode_cache();
2703 jbd2_journal_destroy_transaction_cache();
2704 jbd2_journal_destroy_slabs();
2707 static int __init
journal_init(void)
2711 BUILD_BUG_ON(sizeof(struct journal_superblock_s
) != 1024);
2713 ret
= journal_init_caches();
2715 jbd2_create_jbd_stats_proc_entry();
2717 jbd2_journal_destroy_caches();
2722 static void __exit
journal_exit(void)
2724 #ifdef CONFIG_JBD2_DEBUG
2725 int n
= atomic_read(&nr_journal_heads
);
2727 printk(KERN_ERR
"JBD2: leaked %d journal_heads!\n", n
);
2729 jbd2_remove_jbd_stats_proc_entry();
2730 jbd2_journal_destroy_caches();
2733 MODULE_LICENSE("GPL");
2734 module_init(journal_init
);
2735 module_exit(journal_exit
);