2 * linux/fs/jbd2/journal.c
4 * Written by Stephen C. Tweedie <sct@redhat.com>, 1998
6 * Copyright 1998 Red Hat corp --- 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 * Generic filesystem journal-writing code; part of the ext2fs
15 * This file manages journals: areas of disk reserved for logging
16 * transactional updates. This includes the kernel journaling thread
17 * which is responsible for scheduling updates to the log.
19 * We do not actually manage the physical storage of the journal in this
20 * file: that is left to a per-journal policy function, which allows us
21 * to store the journal within a filesystem-specified area for ext2
22 * journaling (ext2 can use a reserved inode for storing the log).
25 #include <linux/module.h>
26 #include <linux/time.h>
28 #include <linux/jbd2.h>
29 #include <linux/errno.h>
30 #include <linux/slab.h>
31 #include <linux/init.h>
33 #include <linux/freezer.h>
34 #include <linux/pagemap.h>
35 #include <linux/kthread.h>
36 #include <linux/poison.h>
37 #include <linux/proc_fs.h>
38 #include <linux/seq_file.h>
39 #include <linux/math64.h>
40 #include <linux/hash.h>
41 #include <linux/log2.h>
42 #include <linux/vmalloc.h>
43 #include <linux/backing-dev.h>
44 #include <linux/bitops.h>
45 #include <linux/ratelimit.h>
46 #include <linux/sched/mm.h>
48 #define CREATE_TRACE_POINTS
49 #include <trace/events/jbd2.h>
51 #include <linux/uaccess.h>
54 #ifdef CONFIG_JBD2_DEBUG
55 ushort jbd2_journal_enable_debug __read_mostly
;
56 EXPORT_SYMBOL(jbd2_journal_enable_debug
);
58 module_param_named(jbd2_debug
, jbd2_journal_enable_debug
, ushort
, 0644);
59 MODULE_PARM_DESC(jbd2_debug
, "Debugging level for jbd2");
62 EXPORT_SYMBOL(jbd2_journal_extend
);
63 EXPORT_SYMBOL(jbd2_journal_stop
);
64 EXPORT_SYMBOL(jbd2_journal_lock_updates
);
65 EXPORT_SYMBOL(jbd2_journal_unlock_updates
);
66 EXPORT_SYMBOL(jbd2_journal_get_write_access
);
67 EXPORT_SYMBOL(jbd2_journal_get_create_access
);
68 EXPORT_SYMBOL(jbd2_journal_get_undo_access
);
69 EXPORT_SYMBOL(jbd2_journal_set_triggers
);
70 EXPORT_SYMBOL(jbd2_journal_dirty_metadata
);
71 EXPORT_SYMBOL(jbd2_journal_forget
);
73 EXPORT_SYMBOL(journal_sync_buffer
);
75 EXPORT_SYMBOL(jbd2_journal_flush
);
76 EXPORT_SYMBOL(jbd2_journal_revoke
);
78 EXPORT_SYMBOL(jbd2_journal_init_dev
);
79 EXPORT_SYMBOL(jbd2_journal_init_inode
);
80 EXPORT_SYMBOL(jbd2_journal_check_used_features
);
81 EXPORT_SYMBOL(jbd2_journal_check_available_features
);
82 EXPORT_SYMBOL(jbd2_journal_set_features
);
83 EXPORT_SYMBOL(jbd2_journal_load
);
84 EXPORT_SYMBOL(jbd2_journal_destroy
);
85 EXPORT_SYMBOL(jbd2_journal_abort
);
86 EXPORT_SYMBOL(jbd2_journal_errno
);
87 EXPORT_SYMBOL(jbd2_journal_ack_err
);
88 EXPORT_SYMBOL(jbd2_journal_clear_err
);
89 EXPORT_SYMBOL(jbd2_log_wait_commit
);
90 EXPORT_SYMBOL(jbd2_log_start_commit
);
91 EXPORT_SYMBOL(jbd2_journal_start_commit
);
92 EXPORT_SYMBOL(jbd2_journal_force_commit_nested
);
93 EXPORT_SYMBOL(jbd2_journal_wipe
);
94 EXPORT_SYMBOL(jbd2_journal_blocks_per_page
);
95 EXPORT_SYMBOL(jbd2_journal_invalidatepage
);
96 EXPORT_SYMBOL(jbd2_journal_try_to_free_buffers
);
97 EXPORT_SYMBOL(jbd2_journal_force_commit
);
98 EXPORT_SYMBOL(jbd2_journal_inode_add_write
);
99 EXPORT_SYMBOL(jbd2_journal_inode_add_wait
);
100 EXPORT_SYMBOL(jbd2_journal_init_jbd_inode
);
101 EXPORT_SYMBOL(jbd2_journal_release_jbd_inode
);
102 EXPORT_SYMBOL(jbd2_journal_begin_ordered_truncate
);
103 EXPORT_SYMBOL(jbd2_inode_cache
);
105 static void __journal_abort_soft (journal_t
*journal
, int errno
);
106 static int jbd2_journal_create_slab(size_t slab_size
);
108 #ifdef CONFIG_JBD2_DEBUG
109 void __jbd2_debug(int level
, const char *file
, const char *func
,
110 unsigned int line
, const char *fmt
, ...)
112 struct va_format vaf
;
115 if (level
> jbd2_journal_enable_debug
)
120 printk(KERN_DEBUG
"%s: (%s, %u): %pV\n", file
, func
, line
, &vaf
);
123 EXPORT_SYMBOL(__jbd2_debug
);
126 /* Checksumming functions */
127 static int jbd2_verify_csum_type(journal_t
*j
, journal_superblock_t
*sb
)
129 if (!jbd2_journal_has_csum_v2or3_feature(j
))
132 return sb
->s_checksum_type
== JBD2_CRC32C_CHKSUM
;
135 static __be32
jbd2_superblock_csum(journal_t
*j
, journal_superblock_t
*sb
)
140 old_csum
= sb
->s_checksum
;
142 csum
= jbd2_chksum(j
, ~0, (char *)sb
, sizeof(journal_superblock_t
));
143 sb
->s_checksum
= old_csum
;
145 return cpu_to_be32(csum
);
148 static int jbd2_superblock_csum_verify(journal_t
*j
, journal_superblock_t
*sb
)
150 if (!jbd2_journal_has_csum_v2or3(j
))
153 return sb
->s_checksum
== jbd2_superblock_csum(j
, sb
);
156 static void jbd2_superblock_csum_set(journal_t
*j
, journal_superblock_t
*sb
)
158 if (!jbd2_journal_has_csum_v2or3(j
))
161 sb
->s_checksum
= jbd2_superblock_csum(j
, sb
);
165 * Helper function used to manage commit timeouts
168 static void commit_timeout(unsigned long __data
)
170 struct task_struct
* p
= (struct task_struct
*) __data
;
176 * kjournald2: The main thread function used to manage a logging device
179 * This kernel thread is responsible for two things:
181 * 1) COMMIT: Every so often we need to commit the current state of the
182 * filesystem to disk. The journal thread is responsible for writing
183 * all of the metadata buffers to disk.
185 * 2) CHECKPOINT: We cannot reuse a used section of the log file until all
186 * of the data in that part of the log has been rewritten elsewhere on
187 * the disk. Flushing these old buffers to reclaim space in the log is
188 * known as checkpointing, and this thread is responsible for that job.
191 static int kjournald2(void *arg
)
193 journal_t
*journal
= arg
;
194 transaction_t
*transaction
;
197 * Set up an interval timer which can be used to trigger a commit wakeup
198 * after the commit interval expires
200 setup_timer(&journal
->j_commit_timer
, commit_timeout
,
201 (unsigned long)current
);
205 /* Record that the journal thread is running */
206 journal
->j_task
= current
;
207 wake_up(&journal
->j_wait_done_commit
);
210 * Make sure that no allocations from this kernel thread will ever
211 * recurse to the fs layer because we are responsible for the
212 * transaction commit and any fs involvement might get stuck waiting for
215 memalloc_nofs_save();
218 * And now, wait forever for commit wakeup events.
220 write_lock(&journal
->j_state_lock
);
223 if (journal
->j_flags
& JBD2_UNMOUNT
)
226 jbd_debug(1, "commit_sequence=%d, commit_request=%d\n",
227 journal
->j_commit_sequence
, journal
->j_commit_request
);
229 if (journal
->j_commit_sequence
!= journal
->j_commit_request
) {
230 jbd_debug(1, "OK, requests differ\n");
231 write_unlock(&journal
->j_state_lock
);
232 del_timer_sync(&journal
->j_commit_timer
);
233 jbd2_journal_commit_transaction(journal
);
234 write_lock(&journal
->j_state_lock
);
238 wake_up(&journal
->j_wait_done_commit
);
239 if (freezing(current
)) {
241 * The simpler the better. Flushing journal isn't a
242 * good idea, because that depends on threads that may
243 * be already stopped.
245 jbd_debug(1, "Now suspending kjournald2\n");
246 write_unlock(&journal
->j_state_lock
);
248 write_lock(&journal
->j_state_lock
);
251 * We assume on resume that commits are already there,
255 int should_sleep
= 1;
257 prepare_to_wait(&journal
->j_wait_commit
, &wait
,
259 if (journal
->j_commit_sequence
!= journal
->j_commit_request
)
261 transaction
= journal
->j_running_transaction
;
262 if (transaction
&& time_after_eq(jiffies
,
263 transaction
->t_expires
))
265 if (journal
->j_flags
& JBD2_UNMOUNT
)
268 write_unlock(&journal
->j_state_lock
);
270 write_lock(&journal
->j_state_lock
);
272 finish_wait(&journal
->j_wait_commit
, &wait
);
275 jbd_debug(1, "kjournald2 wakes\n");
278 * Were we woken up by a commit wakeup event?
280 transaction
= journal
->j_running_transaction
;
281 if (transaction
&& time_after_eq(jiffies
, transaction
->t_expires
)) {
282 journal
->j_commit_request
= transaction
->t_tid
;
283 jbd_debug(1, "woke because of timeout\n");
288 del_timer_sync(&journal
->j_commit_timer
);
289 journal
->j_task
= NULL
;
290 wake_up(&journal
->j_wait_done_commit
);
291 jbd_debug(1, "Journal thread exiting.\n");
292 write_unlock(&journal
->j_state_lock
);
296 static int jbd2_journal_start_thread(journal_t
*journal
)
298 struct task_struct
*t
;
300 t
= kthread_run(kjournald2
, journal
, "jbd2/%s",
305 wait_event(journal
->j_wait_done_commit
, journal
->j_task
!= NULL
);
309 static void journal_kill_thread(journal_t
*journal
)
311 write_lock(&journal
->j_state_lock
);
312 journal
->j_flags
|= JBD2_UNMOUNT
;
314 while (journal
->j_task
) {
315 write_unlock(&journal
->j_state_lock
);
316 wake_up(&journal
->j_wait_commit
);
317 wait_event(journal
->j_wait_done_commit
, journal
->j_task
== NULL
);
318 write_lock(&journal
->j_state_lock
);
320 write_unlock(&journal
->j_state_lock
);
324 * jbd2_journal_write_metadata_buffer: write a metadata buffer to the journal.
326 * Writes a metadata buffer to a given disk block. The actual IO is not
327 * performed but a new buffer_head is constructed which labels the data
328 * to be written with the correct destination disk block.
330 * Any magic-number escaping which needs to be done will cause a
331 * copy-out here. If the buffer happens to start with the
332 * JBD2_MAGIC_NUMBER, then we can't write it to the log directly: the
333 * magic number is only written to the log for descripter blocks. In
334 * this case, we copy the data and replace the first word with 0, and we
335 * return a result code which indicates that this buffer needs to be
336 * marked as an escaped buffer in the corresponding log descriptor
337 * block. The missing word can then be restored when the block is read
340 * If the source buffer has already been modified by a new transaction
341 * since we took the last commit snapshot, we use the frozen copy of
342 * that data for IO. If we end up using the existing buffer_head's data
343 * for the write, then we have to make sure nobody modifies it while the
344 * IO is in progress. do_get_write_access() handles this.
346 * The function returns a pointer to the buffer_head to be used for IO.
354 * Bit 0 set == escape performed on the data
355 * Bit 1 set == buffer copy-out performed (kfree the data after IO)
358 int jbd2_journal_write_metadata_buffer(transaction_t
*transaction
,
359 struct journal_head
*jh_in
,
360 struct buffer_head
**bh_out
,
363 int need_copy_out
= 0;
364 int done_copy_out
= 0;
367 struct buffer_head
*new_bh
;
368 struct page
*new_page
;
369 unsigned int new_offset
;
370 struct buffer_head
*bh_in
= jh2bh(jh_in
);
371 journal_t
*journal
= transaction
->t_journal
;
374 * The buffer really shouldn't be locked: only the current committing
375 * transaction is allowed to write it, so nobody else is allowed
378 * akpm: except if we're journalling data, and write() output is
379 * also part of a shared mapping, and another thread has
380 * decided to launch a writepage() against this buffer.
382 J_ASSERT_BH(bh_in
, buffer_jbddirty(bh_in
));
384 new_bh
= alloc_buffer_head(GFP_NOFS
|__GFP_NOFAIL
);
386 /* keep subsequent assertions sane */
387 atomic_set(&new_bh
->b_count
, 1);
389 jbd_lock_bh_state(bh_in
);
392 * If a new transaction has already done a buffer copy-out, then
393 * we use that version of the data for the commit.
395 if (jh_in
->b_frozen_data
) {
397 new_page
= virt_to_page(jh_in
->b_frozen_data
);
398 new_offset
= offset_in_page(jh_in
->b_frozen_data
);
400 new_page
= jh2bh(jh_in
)->b_page
;
401 new_offset
= offset_in_page(jh2bh(jh_in
)->b_data
);
404 mapped_data
= kmap_atomic(new_page
);
406 * Fire data frozen trigger if data already wasn't frozen. Do this
407 * before checking for escaping, as the trigger may modify the magic
408 * offset. If a copy-out happens afterwards, it will have the correct
409 * data in the buffer.
412 jbd2_buffer_frozen_trigger(jh_in
, mapped_data
+ new_offset
,
418 if (*((__be32
*)(mapped_data
+ new_offset
)) ==
419 cpu_to_be32(JBD2_MAGIC_NUMBER
)) {
423 kunmap_atomic(mapped_data
);
426 * Do we need to do a data copy?
428 if (need_copy_out
&& !done_copy_out
) {
431 jbd_unlock_bh_state(bh_in
);
432 tmp
= jbd2_alloc(bh_in
->b_size
, GFP_NOFS
);
437 jbd_lock_bh_state(bh_in
);
438 if (jh_in
->b_frozen_data
) {
439 jbd2_free(tmp
, bh_in
->b_size
);
443 jh_in
->b_frozen_data
= tmp
;
444 mapped_data
= kmap_atomic(new_page
);
445 memcpy(tmp
, mapped_data
+ new_offset
, bh_in
->b_size
);
446 kunmap_atomic(mapped_data
);
448 new_page
= virt_to_page(tmp
);
449 new_offset
= offset_in_page(tmp
);
453 * This isn't strictly necessary, as we're using frozen
454 * data for the escaping, but it keeps consistency with
455 * b_frozen_data usage.
457 jh_in
->b_frozen_triggers
= jh_in
->b_triggers
;
461 * Did we need to do an escaping? Now we've done all the
462 * copying, we can finally do so.
465 mapped_data
= kmap_atomic(new_page
);
466 *((unsigned int *)(mapped_data
+ new_offset
)) = 0;
467 kunmap_atomic(mapped_data
);
470 set_bh_page(new_bh
, new_page
, new_offset
);
471 new_bh
->b_size
= bh_in
->b_size
;
472 new_bh
->b_bdev
= journal
->j_dev
;
473 new_bh
->b_blocknr
= blocknr
;
474 new_bh
->b_private
= bh_in
;
475 set_buffer_mapped(new_bh
);
476 set_buffer_dirty(new_bh
);
481 * The to-be-written buffer needs to get moved to the io queue,
482 * and the original buffer whose contents we are shadowing or
483 * copying is moved to the transaction's shadow queue.
485 JBUFFER_TRACE(jh_in
, "file as BJ_Shadow");
486 spin_lock(&journal
->j_list_lock
);
487 __jbd2_journal_file_buffer(jh_in
, transaction
, BJ_Shadow
);
488 spin_unlock(&journal
->j_list_lock
);
489 set_buffer_shadow(bh_in
);
490 jbd_unlock_bh_state(bh_in
);
492 return do_escape
| (done_copy_out
<< 1);
496 * Allocation code for the journal file. Manage the space left in the
497 * journal, so that we can begin checkpointing when appropriate.
501 * Called with j_state_lock locked for writing.
502 * Returns true if a transaction commit was started.
504 int __jbd2_log_start_commit(journal_t
*journal
, tid_t target
)
506 /* Return if the txn has already requested to be committed */
507 if (journal
->j_commit_request
== target
)
511 * The only transaction we can possibly wait upon is the
512 * currently running transaction (if it exists). Otherwise,
513 * the target tid must be an old one.
515 if (journal
->j_running_transaction
&&
516 journal
->j_running_transaction
->t_tid
== target
) {
518 * We want a new commit: OK, mark the request and wakeup the
519 * commit thread. We do _not_ do the commit ourselves.
522 journal
->j_commit_request
= target
;
523 jbd_debug(1, "JBD2: requesting commit %d/%d\n",
524 journal
->j_commit_request
,
525 journal
->j_commit_sequence
);
526 journal
->j_running_transaction
->t_requested
= jiffies
;
527 wake_up(&journal
->j_wait_commit
);
529 } else if (!tid_geq(journal
->j_commit_request
, target
))
530 /* This should never happen, but if it does, preserve
531 the evidence before kjournald goes into a loop and
532 increments j_commit_sequence beyond all recognition. */
533 WARN_ONCE(1, "JBD2: bad log_start_commit: %u %u %u %u\n",
534 journal
->j_commit_request
,
535 journal
->j_commit_sequence
,
536 target
, journal
->j_running_transaction
?
537 journal
->j_running_transaction
->t_tid
: 0);
541 int jbd2_log_start_commit(journal_t
*journal
, tid_t tid
)
545 write_lock(&journal
->j_state_lock
);
546 ret
= __jbd2_log_start_commit(journal
, tid
);
547 write_unlock(&journal
->j_state_lock
);
552 * Force and wait any uncommitted transactions. We can only force the running
553 * transaction if we don't have an active handle, otherwise, we will deadlock.
554 * Returns: <0 in case of error,
555 * 0 if nothing to commit,
556 * 1 if transaction was successfully committed.
558 static int __jbd2_journal_force_commit(journal_t
*journal
)
560 transaction_t
*transaction
= NULL
;
562 int need_to_start
= 0, ret
= 0;
564 read_lock(&journal
->j_state_lock
);
565 if (journal
->j_running_transaction
&& !current
->journal_info
) {
566 transaction
= journal
->j_running_transaction
;
567 if (!tid_geq(journal
->j_commit_request
, transaction
->t_tid
))
569 } else if (journal
->j_committing_transaction
)
570 transaction
= journal
->j_committing_transaction
;
573 /* Nothing to commit */
574 read_unlock(&journal
->j_state_lock
);
577 tid
= transaction
->t_tid
;
578 read_unlock(&journal
->j_state_lock
);
580 jbd2_log_start_commit(journal
, tid
);
581 ret
= jbd2_log_wait_commit(journal
, tid
);
589 * Force and wait upon a commit if the calling process is not within
590 * transaction. This is used for forcing out undo-protected data which contains
591 * bitmaps, when the fs is running out of space.
593 * @journal: journal to force
594 * Returns true if progress was made.
596 int jbd2_journal_force_commit_nested(journal_t
*journal
)
600 ret
= __jbd2_journal_force_commit(journal
);
605 * int journal_force_commit() - force any uncommitted transactions
606 * @journal: journal to force
608 * Caller want unconditional commit. We can only force the running transaction
609 * if we don't have an active handle, otherwise, we will deadlock.
611 int jbd2_journal_force_commit(journal_t
*journal
)
615 J_ASSERT(!current
->journal_info
);
616 ret
= __jbd2_journal_force_commit(journal
);
623 * Start a commit of the current running transaction (if any). Returns true
624 * if a transaction is going to be committed (or is currently already
625 * committing), and fills its tid in at *ptid
627 int jbd2_journal_start_commit(journal_t
*journal
, tid_t
*ptid
)
631 write_lock(&journal
->j_state_lock
);
632 if (journal
->j_running_transaction
) {
633 tid_t tid
= journal
->j_running_transaction
->t_tid
;
635 __jbd2_log_start_commit(journal
, tid
);
636 /* There's a running transaction and we've just made sure
637 * it's commit has been scheduled. */
641 } else if (journal
->j_committing_transaction
) {
643 * If commit has been started, then we have to wait for
644 * completion of that transaction.
647 *ptid
= journal
->j_committing_transaction
->t_tid
;
650 write_unlock(&journal
->j_state_lock
);
655 * Return 1 if a given transaction has not yet sent barrier request
656 * connected with a transaction commit. If 0 is returned, transaction
657 * may or may not have sent the barrier. Used to avoid sending barrier
658 * twice in common cases.
660 int jbd2_trans_will_send_data_barrier(journal_t
*journal
, tid_t tid
)
663 transaction_t
*commit_trans
;
665 if (!(journal
->j_flags
& JBD2_BARRIER
))
667 read_lock(&journal
->j_state_lock
);
668 /* Transaction already committed? */
669 if (tid_geq(journal
->j_commit_sequence
, tid
))
671 commit_trans
= journal
->j_committing_transaction
;
672 if (!commit_trans
|| commit_trans
->t_tid
!= tid
) {
677 * Transaction is being committed and we already proceeded to
678 * submitting a flush to fs partition?
680 if (journal
->j_fs_dev
!= journal
->j_dev
) {
681 if (!commit_trans
->t_need_data_flush
||
682 commit_trans
->t_state
>= T_COMMIT_DFLUSH
)
685 if (commit_trans
->t_state
>= T_COMMIT_JFLUSH
)
690 read_unlock(&journal
->j_state_lock
);
693 EXPORT_SYMBOL(jbd2_trans_will_send_data_barrier
);
696 * Wait for a specified commit to complete.
697 * The caller may not hold the journal lock.
699 int jbd2_log_wait_commit(journal_t
*journal
, tid_t tid
)
703 read_lock(&journal
->j_state_lock
);
704 #ifdef CONFIG_PROVE_LOCKING
706 * Some callers make sure transaction is already committing and in that
707 * case we cannot block on open handles anymore. So don't warn in that
710 if (tid_gt(tid
, journal
->j_commit_sequence
) &&
711 (!journal
->j_committing_transaction
||
712 journal
->j_committing_transaction
->t_tid
!= tid
)) {
713 read_unlock(&journal
->j_state_lock
);
714 jbd2_might_wait_for_commit(journal
);
715 read_lock(&journal
->j_state_lock
);
718 #ifdef CONFIG_JBD2_DEBUG
719 if (!tid_geq(journal
->j_commit_request
, tid
)) {
721 "%s: error: j_commit_request=%d, tid=%d\n",
722 __func__
, journal
->j_commit_request
, tid
);
725 while (tid_gt(tid
, journal
->j_commit_sequence
)) {
726 jbd_debug(1, "JBD2: want %d, j_commit_sequence=%d\n",
727 tid
, journal
->j_commit_sequence
);
728 read_unlock(&journal
->j_state_lock
);
729 wake_up(&journal
->j_wait_commit
);
730 wait_event(journal
->j_wait_done_commit
,
731 !tid_gt(tid
, journal
->j_commit_sequence
));
732 read_lock(&journal
->j_state_lock
);
734 read_unlock(&journal
->j_state_lock
);
736 if (unlikely(is_journal_aborted(journal
)))
742 * When this function returns the transaction corresponding to tid
743 * will be completed. If the transaction has currently running, start
744 * committing that transaction before waiting for it to complete. If
745 * the transaction id is stale, it is by definition already completed,
746 * so just return SUCCESS.
748 int jbd2_complete_transaction(journal_t
*journal
, tid_t tid
)
750 int need_to_wait
= 1;
752 read_lock(&journal
->j_state_lock
);
753 if (journal
->j_running_transaction
&&
754 journal
->j_running_transaction
->t_tid
== tid
) {
755 if (journal
->j_commit_request
!= tid
) {
756 /* transaction not yet started, so request it */
757 read_unlock(&journal
->j_state_lock
);
758 jbd2_log_start_commit(journal
, tid
);
761 } else if (!(journal
->j_committing_transaction
&&
762 journal
->j_committing_transaction
->t_tid
== tid
))
764 read_unlock(&journal
->j_state_lock
);
768 return jbd2_log_wait_commit(journal
, tid
);
770 EXPORT_SYMBOL(jbd2_complete_transaction
);
773 * Log buffer allocation routines:
776 int jbd2_journal_next_log_block(journal_t
*journal
, unsigned long long *retp
)
778 unsigned long blocknr
;
780 write_lock(&journal
->j_state_lock
);
781 J_ASSERT(journal
->j_free
> 1);
783 blocknr
= journal
->j_head
;
786 if (journal
->j_head
== journal
->j_last
)
787 journal
->j_head
= journal
->j_first
;
788 write_unlock(&journal
->j_state_lock
);
789 return jbd2_journal_bmap(journal
, blocknr
, retp
);
793 * Conversion of logical to physical block numbers for the journal
795 * On external journals the journal blocks are identity-mapped, so
796 * this is a no-op. If needed, we can use j_blk_offset - everything is
799 int jbd2_journal_bmap(journal_t
*journal
, unsigned long blocknr
,
800 unsigned long long *retp
)
803 unsigned long long ret
;
805 if (journal
->j_inode
) {
806 ret
= bmap(journal
->j_inode
, blocknr
);
810 printk(KERN_ALERT
"%s: journal block not found "
811 "at offset %lu on %s\n",
812 __func__
, blocknr
, journal
->j_devname
);
814 __journal_abort_soft(journal
, err
);
817 *retp
= blocknr
; /* +journal->j_blk_offset */
823 * We play buffer_head aliasing tricks to write data/metadata blocks to
824 * the journal without copying their contents, but for journal
825 * descriptor blocks we do need to generate bona fide buffers.
827 * After the caller of jbd2_journal_get_descriptor_buffer() has finished modifying
828 * the buffer's contents they really should run flush_dcache_page(bh->b_page).
829 * But we don't bother doing that, so there will be coherency problems with
830 * mmaps of blockdevs which hold live JBD-controlled filesystems.
833 jbd2_journal_get_descriptor_buffer(transaction_t
*transaction
, int type
)
835 journal_t
*journal
= transaction
->t_journal
;
836 struct buffer_head
*bh
;
837 unsigned long long blocknr
;
838 journal_header_t
*header
;
841 err
= jbd2_journal_next_log_block(journal
, &blocknr
);
846 bh
= __getblk(journal
->j_dev
, blocknr
, journal
->j_blocksize
);
850 memset(bh
->b_data
, 0, journal
->j_blocksize
);
851 header
= (journal_header_t
*)bh
->b_data
;
852 header
->h_magic
= cpu_to_be32(JBD2_MAGIC_NUMBER
);
853 header
->h_blocktype
= cpu_to_be32(type
);
854 header
->h_sequence
= cpu_to_be32(transaction
->t_tid
);
855 set_buffer_uptodate(bh
);
857 BUFFER_TRACE(bh
, "return this buffer");
861 void jbd2_descriptor_block_csum_set(journal_t
*j
, struct buffer_head
*bh
)
863 struct jbd2_journal_block_tail
*tail
;
866 if (!jbd2_journal_has_csum_v2or3(j
))
869 tail
= (struct jbd2_journal_block_tail
*)(bh
->b_data
+ j
->j_blocksize
-
870 sizeof(struct jbd2_journal_block_tail
));
871 tail
->t_checksum
= 0;
872 csum
= jbd2_chksum(j
, j
->j_csum_seed
, bh
->b_data
, j
->j_blocksize
);
873 tail
->t_checksum
= cpu_to_be32(csum
);
877 * Return tid of the oldest transaction in the journal and block in the journal
878 * where the transaction starts.
880 * If the journal is now empty, return which will be the next transaction ID
881 * we will write and where will that transaction start.
883 * The return value is 0 if journal tail cannot be pushed any further, 1 if
886 int jbd2_journal_get_log_tail(journal_t
*journal
, tid_t
*tid
,
887 unsigned long *block
)
889 transaction_t
*transaction
;
892 read_lock(&journal
->j_state_lock
);
893 spin_lock(&journal
->j_list_lock
);
894 transaction
= journal
->j_checkpoint_transactions
;
896 *tid
= transaction
->t_tid
;
897 *block
= transaction
->t_log_start
;
898 } else if ((transaction
= journal
->j_committing_transaction
) != NULL
) {
899 *tid
= transaction
->t_tid
;
900 *block
= transaction
->t_log_start
;
901 } else if ((transaction
= journal
->j_running_transaction
) != NULL
) {
902 *tid
= transaction
->t_tid
;
903 *block
= journal
->j_head
;
905 *tid
= journal
->j_transaction_sequence
;
906 *block
= journal
->j_head
;
908 ret
= tid_gt(*tid
, journal
->j_tail_sequence
);
909 spin_unlock(&journal
->j_list_lock
);
910 read_unlock(&journal
->j_state_lock
);
916 * Update information in journal structure and in on disk journal superblock
917 * about log tail. This function does not check whether information passed in
918 * really pushes log tail further. It's responsibility of the caller to make
919 * sure provided log tail information is valid (e.g. by holding
920 * j_checkpoint_mutex all the time between computing log tail and calling this
921 * function as is the case with jbd2_cleanup_journal_tail()).
923 * Requires j_checkpoint_mutex
925 int __jbd2_update_log_tail(journal_t
*journal
, tid_t tid
, unsigned long block
)
930 BUG_ON(!mutex_is_locked(&journal
->j_checkpoint_mutex
));
933 * We cannot afford for write to remain in drive's caches since as
934 * soon as we update j_tail, next transaction can start reusing journal
935 * space and if we lose sb update during power failure we'd replay
936 * old transaction with possibly newly overwritten data.
938 ret
= jbd2_journal_update_sb_log_tail(journal
, tid
, block
,
943 write_lock(&journal
->j_state_lock
);
944 freed
= block
- journal
->j_tail
;
945 if (block
< journal
->j_tail
)
946 freed
+= journal
->j_last
- journal
->j_first
;
948 trace_jbd2_update_log_tail(journal
, tid
, block
, freed
);
950 "Cleaning journal tail from %d to %d (offset %lu), "
952 journal
->j_tail_sequence
, tid
, block
, freed
);
954 journal
->j_free
+= freed
;
955 journal
->j_tail_sequence
= tid
;
956 journal
->j_tail
= block
;
957 write_unlock(&journal
->j_state_lock
);
964 * This is a variaon of __jbd2_update_log_tail which checks for validity of
965 * provided log tail and locks j_checkpoint_mutex. So it is safe against races
966 * with other threads updating log tail.
968 void jbd2_update_log_tail(journal_t
*journal
, tid_t tid
, unsigned long block
)
970 mutex_lock_io(&journal
->j_checkpoint_mutex
);
971 if (tid_gt(tid
, journal
->j_tail_sequence
))
972 __jbd2_update_log_tail(journal
, tid
, block
);
973 mutex_unlock(&journal
->j_checkpoint_mutex
);
976 struct jbd2_stats_proc_session
{
978 struct transaction_stats_s
*stats
;
983 static void *jbd2_seq_info_start(struct seq_file
*seq
, loff_t
*pos
)
985 return *pos
? NULL
: SEQ_START_TOKEN
;
988 static void *jbd2_seq_info_next(struct seq_file
*seq
, void *v
, loff_t
*pos
)
993 static int jbd2_seq_info_show(struct seq_file
*seq
, void *v
)
995 struct jbd2_stats_proc_session
*s
= seq
->private;
997 if (v
!= SEQ_START_TOKEN
)
999 seq_printf(seq
, "%lu transactions (%lu requested), "
1000 "each up to %u blocks\n",
1001 s
->stats
->ts_tid
, s
->stats
->ts_requested
,
1002 s
->journal
->j_max_transaction_buffers
);
1003 if (s
->stats
->ts_tid
== 0)
1005 seq_printf(seq
, "average: \n %ums waiting for transaction\n",
1006 jiffies_to_msecs(s
->stats
->run
.rs_wait
/ s
->stats
->ts_tid
));
1007 seq_printf(seq
, " %ums request delay\n",
1008 (s
->stats
->ts_requested
== 0) ? 0 :
1009 jiffies_to_msecs(s
->stats
->run
.rs_request_delay
/
1010 s
->stats
->ts_requested
));
1011 seq_printf(seq
, " %ums running transaction\n",
1012 jiffies_to_msecs(s
->stats
->run
.rs_running
/ s
->stats
->ts_tid
));
1013 seq_printf(seq
, " %ums transaction was being locked\n",
1014 jiffies_to_msecs(s
->stats
->run
.rs_locked
/ s
->stats
->ts_tid
));
1015 seq_printf(seq
, " %ums flushing data (in ordered mode)\n",
1016 jiffies_to_msecs(s
->stats
->run
.rs_flushing
/ s
->stats
->ts_tid
));
1017 seq_printf(seq
, " %ums logging transaction\n",
1018 jiffies_to_msecs(s
->stats
->run
.rs_logging
/ s
->stats
->ts_tid
));
1019 seq_printf(seq
, " %lluus average transaction commit time\n",
1020 div_u64(s
->journal
->j_average_commit_time
, 1000));
1021 seq_printf(seq
, " %lu handles per transaction\n",
1022 s
->stats
->run
.rs_handle_count
/ s
->stats
->ts_tid
);
1023 seq_printf(seq
, " %lu blocks per transaction\n",
1024 s
->stats
->run
.rs_blocks
/ s
->stats
->ts_tid
);
1025 seq_printf(seq
, " %lu logged blocks per transaction\n",
1026 s
->stats
->run
.rs_blocks_logged
/ s
->stats
->ts_tid
);
1030 static void jbd2_seq_info_stop(struct seq_file
*seq
, void *v
)
1034 static const struct seq_operations jbd2_seq_info_ops
= {
1035 .start
= jbd2_seq_info_start
,
1036 .next
= jbd2_seq_info_next
,
1037 .stop
= jbd2_seq_info_stop
,
1038 .show
= jbd2_seq_info_show
,
1041 static int jbd2_seq_info_open(struct inode
*inode
, struct file
*file
)
1043 journal_t
*journal
= PDE_DATA(inode
);
1044 struct jbd2_stats_proc_session
*s
;
1047 s
= kmalloc(sizeof(*s
), GFP_KERNEL
);
1050 size
= sizeof(struct transaction_stats_s
);
1051 s
->stats
= kmalloc(size
, GFP_KERNEL
);
1052 if (s
->stats
== NULL
) {
1056 spin_lock(&journal
->j_history_lock
);
1057 memcpy(s
->stats
, &journal
->j_stats
, size
);
1058 s
->journal
= journal
;
1059 spin_unlock(&journal
->j_history_lock
);
1061 rc
= seq_open(file
, &jbd2_seq_info_ops
);
1063 struct seq_file
*m
= file
->private_data
;
1073 static int jbd2_seq_info_release(struct inode
*inode
, struct file
*file
)
1075 struct seq_file
*seq
= file
->private_data
;
1076 struct jbd2_stats_proc_session
*s
= seq
->private;
1079 return seq_release(inode
, file
);
1082 static const struct file_operations jbd2_seq_info_fops
= {
1083 .owner
= THIS_MODULE
,
1084 .open
= jbd2_seq_info_open
,
1086 .llseek
= seq_lseek
,
1087 .release
= jbd2_seq_info_release
,
1090 static struct proc_dir_entry
*proc_jbd2_stats
;
1092 static void jbd2_stats_proc_init(journal_t
*journal
)
1094 journal
->j_proc_entry
= proc_mkdir(journal
->j_devname
, proc_jbd2_stats
);
1095 if (journal
->j_proc_entry
) {
1096 proc_create_data("info", S_IRUGO
, journal
->j_proc_entry
,
1097 &jbd2_seq_info_fops
, journal
);
1101 static void jbd2_stats_proc_exit(journal_t
*journal
)
1103 remove_proc_entry("info", journal
->j_proc_entry
);
1104 remove_proc_entry(journal
->j_devname
, proc_jbd2_stats
);
1108 * Management for journal control blocks: functions to create and
1109 * destroy journal_t structures, and to initialise and read existing
1110 * journal blocks from disk. */
1112 /* First: create and setup a journal_t object in memory. We initialise
1113 * very few fields yet: that has to wait until we have created the
1114 * journal structures from from scratch, or loaded them from disk. */
1116 static journal_t
*journal_init_common(struct block_device
*bdev
,
1117 struct block_device
*fs_dev
,
1118 unsigned long long start
, int len
, int blocksize
)
1120 static struct lock_class_key jbd2_trans_commit_key
;
1123 struct buffer_head
*bh
;
1126 journal
= kzalloc(sizeof(*journal
), GFP_KERNEL
);
1130 init_waitqueue_head(&journal
->j_wait_transaction_locked
);
1131 init_waitqueue_head(&journal
->j_wait_done_commit
);
1132 init_waitqueue_head(&journal
->j_wait_commit
);
1133 init_waitqueue_head(&journal
->j_wait_updates
);
1134 init_waitqueue_head(&journal
->j_wait_reserved
);
1135 mutex_init(&journal
->j_barrier
);
1136 mutex_init(&journal
->j_checkpoint_mutex
);
1137 spin_lock_init(&journal
->j_revoke_lock
);
1138 spin_lock_init(&journal
->j_list_lock
);
1139 rwlock_init(&journal
->j_state_lock
);
1141 journal
->j_commit_interval
= (HZ
* JBD2_DEFAULT_MAX_COMMIT_AGE
);
1142 journal
->j_min_batch_time
= 0;
1143 journal
->j_max_batch_time
= 15000; /* 15ms */
1144 atomic_set(&journal
->j_reserved_credits
, 0);
1146 /* The journal is marked for error until we succeed with recovery! */
1147 journal
->j_flags
= JBD2_ABORT
;
1149 /* Set up a default-sized revoke table for the new mount. */
1150 err
= jbd2_journal_init_revoke(journal
, JOURNAL_REVOKE_DEFAULT_HASH
);
1154 spin_lock_init(&journal
->j_history_lock
);
1156 lockdep_init_map(&journal
->j_trans_commit_map
, "jbd2_handle",
1157 &jbd2_trans_commit_key
, 0);
1159 /* journal descriptor can store up to n blocks -bzzz */
1160 journal
->j_blocksize
= blocksize
;
1161 journal
->j_dev
= bdev
;
1162 journal
->j_fs_dev
= fs_dev
;
1163 journal
->j_blk_offset
= start
;
1164 journal
->j_maxlen
= len
;
1165 n
= journal
->j_blocksize
/ sizeof(journal_block_tag_t
);
1166 journal
->j_wbufsize
= n
;
1167 journal
->j_wbuf
= kmalloc_array(n
, sizeof(struct buffer_head
*),
1169 if (!journal
->j_wbuf
)
1172 bh
= getblk_unmovable(journal
->j_dev
, start
, journal
->j_blocksize
);
1174 pr_err("%s: Cannot get buffer for journal superblock\n",
1178 journal
->j_sb_buffer
= bh
;
1179 journal
->j_superblock
= (journal_superblock_t
*)bh
->b_data
;
1184 kfree(journal
->j_wbuf
);
1185 jbd2_journal_destroy_revoke(journal
);
1190 /* jbd2_journal_init_dev and jbd2_journal_init_inode:
1192 * Create a journal structure assigned some fixed set of disk blocks to
1193 * the journal. We don't actually touch those disk blocks yet, but we
1194 * need to set up all of the mapping information to tell the journaling
1195 * system where the journal blocks are.
1200 * journal_t * jbd2_journal_init_dev() - creates and initialises a journal structure
1201 * @bdev: Block device on which to create the journal
1202 * @fs_dev: Device which hold journalled filesystem for this journal.
1203 * @start: Block nr Start of journal.
1204 * @len: Length of the journal in blocks.
1205 * @blocksize: blocksize of journalling device
1207 * Returns: a newly created journal_t *
1209 * jbd2_journal_init_dev creates a journal which maps a fixed contiguous
1210 * range of blocks on an arbitrary block device.
1213 journal_t
*jbd2_journal_init_dev(struct block_device
*bdev
,
1214 struct block_device
*fs_dev
,
1215 unsigned long long start
, int len
, int blocksize
)
1219 journal
= journal_init_common(bdev
, fs_dev
, start
, len
, blocksize
);
1223 bdevname(journal
->j_dev
, journal
->j_devname
);
1224 strreplace(journal
->j_devname
, '/', '!');
1225 jbd2_stats_proc_init(journal
);
1231 * journal_t * jbd2_journal_init_inode () - creates a journal which maps to a inode.
1232 * @inode: An inode to create the journal in
1234 * jbd2_journal_init_inode creates a journal which maps an on-disk inode as
1235 * the journal. The inode must exist already, must support bmap() and
1236 * must have all data blocks preallocated.
1238 journal_t
*jbd2_journal_init_inode(struct inode
*inode
)
1242 unsigned long long blocknr
;
1244 blocknr
= bmap(inode
, 0);
1246 pr_err("%s: Cannot locate journal superblock\n",
1251 jbd_debug(1, "JBD2: inode %s/%ld, size %lld, bits %d, blksize %ld\n",
1252 inode
->i_sb
->s_id
, inode
->i_ino
, (long long) inode
->i_size
,
1253 inode
->i_sb
->s_blocksize_bits
, inode
->i_sb
->s_blocksize
);
1255 journal
= journal_init_common(inode
->i_sb
->s_bdev
, inode
->i_sb
->s_bdev
,
1256 blocknr
, inode
->i_size
>> inode
->i_sb
->s_blocksize_bits
,
1257 inode
->i_sb
->s_blocksize
);
1261 journal
->j_inode
= inode
;
1262 bdevname(journal
->j_dev
, journal
->j_devname
);
1263 p
= strreplace(journal
->j_devname
, '/', '!');
1264 sprintf(p
, "-%lu", journal
->j_inode
->i_ino
);
1265 jbd2_stats_proc_init(journal
);
1271 * If the journal init or create aborts, we need to mark the journal
1272 * superblock as being NULL to prevent the journal destroy from writing
1273 * back a bogus superblock.
1275 static void journal_fail_superblock (journal_t
*journal
)
1277 struct buffer_head
*bh
= journal
->j_sb_buffer
;
1279 journal
->j_sb_buffer
= NULL
;
1283 * Given a journal_t structure, initialise the various fields for
1284 * startup of a new journaling session. We use this both when creating
1285 * a journal, and after recovering an old journal to reset it for
1289 static int journal_reset(journal_t
*journal
)
1291 journal_superblock_t
*sb
= journal
->j_superblock
;
1292 unsigned long long first
, last
;
1294 first
= be32_to_cpu(sb
->s_first
);
1295 last
= be32_to_cpu(sb
->s_maxlen
);
1296 if (first
+ JBD2_MIN_JOURNAL_BLOCKS
> last
+ 1) {
1297 printk(KERN_ERR
"JBD2: Journal too short (blocks %llu-%llu).\n",
1299 journal_fail_superblock(journal
);
1303 journal
->j_first
= first
;
1304 journal
->j_last
= last
;
1306 journal
->j_head
= first
;
1307 journal
->j_tail
= first
;
1308 journal
->j_free
= last
- first
;
1310 journal
->j_tail_sequence
= journal
->j_transaction_sequence
;
1311 journal
->j_commit_sequence
= journal
->j_transaction_sequence
- 1;
1312 journal
->j_commit_request
= journal
->j_commit_sequence
;
1314 journal
->j_max_transaction_buffers
= journal
->j_maxlen
/ 4;
1317 * As a special case, if the on-disk copy is already marked as needing
1318 * no recovery (s_start == 0), then we can safely defer the superblock
1319 * update until the next commit by setting JBD2_FLUSHED. This avoids
1320 * attempting a write to a potential-readonly device.
1322 if (sb
->s_start
== 0) {
1323 jbd_debug(1, "JBD2: Skipping superblock update on recovered sb "
1324 "(start %ld, seq %d, errno %d)\n",
1325 journal
->j_tail
, journal
->j_tail_sequence
,
1327 journal
->j_flags
|= JBD2_FLUSHED
;
1329 /* Lock here to make assertions happy... */
1330 mutex_lock_io(&journal
->j_checkpoint_mutex
);
1332 * Update log tail information. We use REQ_FUA since new
1333 * transaction will start reusing journal space and so we
1334 * must make sure information about current log tail is on
1337 jbd2_journal_update_sb_log_tail(journal
,
1338 journal
->j_tail_sequence
,
1340 REQ_SYNC
| REQ_FUA
);
1341 mutex_unlock(&journal
->j_checkpoint_mutex
);
1343 return jbd2_journal_start_thread(journal
);
1346 static int jbd2_write_superblock(journal_t
*journal
, int write_flags
)
1348 struct buffer_head
*bh
= journal
->j_sb_buffer
;
1349 journal_superblock_t
*sb
= journal
->j_superblock
;
1352 trace_jbd2_write_superblock(journal
, write_flags
);
1353 if (!(journal
->j_flags
& JBD2_BARRIER
))
1354 write_flags
&= ~(REQ_FUA
| REQ_PREFLUSH
);
1356 if (buffer_write_io_error(bh
)) {
1358 * Oh, dear. A previous attempt to write the journal
1359 * superblock failed. This could happen because the
1360 * USB device was yanked out. Or it could happen to
1361 * be a transient write error and maybe the block will
1362 * be remapped. Nothing we can do but to retry the
1363 * write and hope for the best.
1365 printk(KERN_ERR
"JBD2: previous I/O error detected "
1366 "for journal superblock update for %s.\n",
1367 journal
->j_devname
);
1368 clear_buffer_write_io_error(bh
);
1369 set_buffer_uptodate(bh
);
1371 jbd2_superblock_csum_set(journal
, sb
);
1373 bh
->b_end_io
= end_buffer_write_sync
;
1374 ret
= submit_bh(REQ_OP_WRITE
, write_flags
, bh
);
1376 if (buffer_write_io_error(bh
)) {
1377 clear_buffer_write_io_error(bh
);
1378 set_buffer_uptodate(bh
);
1382 printk(KERN_ERR
"JBD2: Error %d detected when updating "
1383 "journal superblock for %s.\n", ret
,
1384 journal
->j_devname
);
1385 jbd2_journal_abort(journal
, ret
);
1392 * jbd2_journal_update_sb_log_tail() - Update log tail in journal sb on disk.
1393 * @journal: The journal to update.
1394 * @tail_tid: TID of the new transaction at the tail of the log
1395 * @tail_block: The first block of the transaction at the tail of the log
1396 * @write_op: With which operation should we write the journal sb
1398 * Update a journal's superblock information about log tail and write it to
1399 * disk, waiting for the IO to complete.
1401 int jbd2_journal_update_sb_log_tail(journal_t
*journal
, tid_t tail_tid
,
1402 unsigned long tail_block
, int write_op
)
1404 journal_superblock_t
*sb
= journal
->j_superblock
;
1407 BUG_ON(!mutex_is_locked(&journal
->j_checkpoint_mutex
));
1408 jbd_debug(1, "JBD2: updating superblock (start %lu, seq %u)\n",
1409 tail_block
, tail_tid
);
1411 sb
->s_sequence
= cpu_to_be32(tail_tid
);
1412 sb
->s_start
= cpu_to_be32(tail_block
);
1414 ret
= jbd2_write_superblock(journal
, write_op
);
1418 /* Log is no longer empty */
1419 write_lock(&journal
->j_state_lock
);
1420 WARN_ON(!sb
->s_sequence
);
1421 journal
->j_flags
&= ~JBD2_FLUSHED
;
1422 write_unlock(&journal
->j_state_lock
);
1429 * jbd2_mark_journal_empty() - Mark on disk journal as empty.
1430 * @journal: The journal to update.
1431 * @write_op: With which operation should we write the journal sb
1433 * Update a journal's dynamic superblock fields to show that journal is empty.
1434 * Write updated superblock to disk waiting for IO to complete.
1436 static void jbd2_mark_journal_empty(journal_t
*journal
, int write_op
)
1438 journal_superblock_t
*sb
= journal
->j_superblock
;
1440 BUG_ON(!mutex_is_locked(&journal
->j_checkpoint_mutex
));
1441 read_lock(&journal
->j_state_lock
);
1442 /* Is it already empty? */
1443 if (sb
->s_start
== 0) {
1444 read_unlock(&journal
->j_state_lock
);
1447 jbd_debug(1, "JBD2: Marking journal as empty (seq %d)\n",
1448 journal
->j_tail_sequence
);
1450 sb
->s_sequence
= cpu_to_be32(journal
->j_tail_sequence
);
1451 sb
->s_start
= cpu_to_be32(0);
1452 read_unlock(&journal
->j_state_lock
);
1454 jbd2_write_superblock(journal
, write_op
);
1456 /* Log is no longer empty */
1457 write_lock(&journal
->j_state_lock
);
1458 journal
->j_flags
|= JBD2_FLUSHED
;
1459 write_unlock(&journal
->j_state_lock
);
1464 * jbd2_journal_update_sb_errno() - Update error in the journal.
1465 * @journal: The journal to update.
1467 * Update a journal's errno. Write updated superblock to disk waiting for IO
1470 void jbd2_journal_update_sb_errno(journal_t
*journal
)
1472 journal_superblock_t
*sb
= journal
->j_superblock
;
1474 read_lock(&journal
->j_state_lock
);
1475 jbd_debug(1, "JBD2: updating superblock error (errno %d)\n",
1477 sb
->s_errno
= cpu_to_be32(journal
->j_errno
);
1478 read_unlock(&journal
->j_state_lock
);
1480 jbd2_write_superblock(journal
, REQ_SYNC
| REQ_FUA
);
1482 EXPORT_SYMBOL(jbd2_journal_update_sb_errno
);
1485 * Read the superblock for a given journal, performing initial
1486 * validation of the format.
1488 static int journal_get_superblock(journal_t
*journal
)
1490 struct buffer_head
*bh
;
1491 journal_superblock_t
*sb
;
1494 bh
= journal
->j_sb_buffer
;
1496 J_ASSERT(bh
!= NULL
);
1497 if (!buffer_uptodate(bh
)) {
1498 ll_rw_block(REQ_OP_READ
, 0, 1, &bh
);
1500 if (!buffer_uptodate(bh
)) {
1502 "JBD2: IO error reading journal superblock\n");
1507 if (buffer_verified(bh
))
1510 sb
= journal
->j_superblock
;
1514 if (sb
->s_header
.h_magic
!= cpu_to_be32(JBD2_MAGIC_NUMBER
) ||
1515 sb
->s_blocksize
!= cpu_to_be32(journal
->j_blocksize
)) {
1516 printk(KERN_WARNING
"JBD2: no valid journal superblock found\n");
1520 switch(be32_to_cpu(sb
->s_header
.h_blocktype
)) {
1521 case JBD2_SUPERBLOCK_V1
:
1522 journal
->j_format_version
= 1;
1524 case JBD2_SUPERBLOCK_V2
:
1525 journal
->j_format_version
= 2;
1528 printk(KERN_WARNING
"JBD2: unrecognised superblock format ID\n");
1532 if (be32_to_cpu(sb
->s_maxlen
) < journal
->j_maxlen
)
1533 journal
->j_maxlen
= be32_to_cpu(sb
->s_maxlen
);
1534 else if (be32_to_cpu(sb
->s_maxlen
) > journal
->j_maxlen
) {
1535 printk(KERN_WARNING
"JBD2: journal file too short\n");
1539 if (be32_to_cpu(sb
->s_first
) == 0 ||
1540 be32_to_cpu(sb
->s_first
) >= journal
->j_maxlen
) {
1542 "JBD2: Invalid start block of journal: %u\n",
1543 be32_to_cpu(sb
->s_first
));
1547 if (jbd2_has_feature_csum2(journal
) &&
1548 jbd2_has_feature_csum3(journal
)) {
1549 /* Can't have checksum v2 and v3 at the same time! */
1550 printk(KERN_ERR
"JBD2: Can't enable checksumming v2 and v3 "
1551 "at the same time!\n");
1555 if (jbd2_journal_has_csum_v2or3_feature(journal
) &&
1556 jbd2_has_feature_checksum(journal
)) {
1557 /* Can't have checksum v1 and v2 on at the same time! */
1558 printk(KERN_ERR
"JBD2: Can't enable checksumming v1 and v2/3 "
1559 "at the same time!\n");
1563 if (!jbd2_verify_csum_type(journal
, sb
)) {
1564 printk(KERN_ERR
"JBD2: Unknown checksum type\n");
1568 /* Load the checksum driver */
1569 if (jbd2_journal_has_csum_v2or3_feature(journal
)) {
1570 journal
->j_chksum_driver
= crypto_alloc_shash("crc32c", 0, 0);
1571 if (IS_ERR(journal
->j_chksum_driver
)) {
1572 printk(KERN_ERR
"JBD2: Cannot load crc32c driver.\n");
1573 err
= PTR_ERR(journal
->j_chksum_driver
);
1574 journal
->j_chksum_driver
= NULL
;
1579 /* Check superblock checksum */
1580 if (!jbd2_superblock_csum_verify(journal
, sb
)) {
1581 printk(KERN_ERR
"JBD2: journal checksum error\n");
1586 /* Precompute checksum seed for all metadata */
1587 if (jbd2_journal_has_csum_v2or3(journal
))
1588 journal
->j_csum_seed
= jbd2_chksum(journal
, ~0, sb
->s_uuid
,
1589 sizeof(sb
->s_uuid
));
1591 set_buffer_verified(bh
);
1596 journal_fail_superblock(journal
);
1601 * Load the on-disk journal superblock and read the key fields into the
1605 static int load_superblock(journal_t
*journal
)
1608 journal_superblock_t
*sb
;
1610 err
= journal_get_superblock(journal
);
1614 sb
= journal
->j_superblock
;
1616 journal
->j_tail_sequence
= be32_to_cpu(sb
->s_sequence
);
1617 journal
->j_tail
= be32_to_cpu(sb
->s_start
);
1618 journal
->j_first
= be32_to_cpu(sb
->s_first
);
1619 journal
->j_last
= be32_to_cpu(sb
->s_maxlen
);
1620 journal
->j_errno
= be32_to_cpu(sb
->s_errno
);
1627 * int jbd2_journal_load() - Read journal from disk.
1628 * @journal: Journal to act on.
1630 * Given a journal_t structure which tells us which disk blocks contain
1631 * a journal, read the journal from disk to initialise the in-memory
1634 int jbd2_journal_load(journal_t
*journal
)
1637 journal_superblock_t
*sb
;
1639 err
= load_superblock(journal
);
1643 sb
= journal
->j_superblock
;
1644 /* If this is a V2 superblock, then we have to check the
1645 * features flags on it. */
1647 if (journal
->j_format_version
>= 2) {
1648 if ((sb
->s_feature_ro_compat
&
1649 ~cpu_to_be32(JBD2_KNOWN_ROCOMPAT_FEATURES
)) ||
1650 (sb
->s_feature_incompat
&
1651 ~cpu_to_be32(JBD2_KNOWN_INCOMPAT_FEATURES
))) {
1653 "JBD2: Unrecognised features on journal\n");
1659 * Create a slab for this blocksize
1661 err
= jbd2_journal_create_slab(be32_to_cpu(sb
->s_blocksize
));
1665 /* Let the recovery code check whether it needs to recover any
1666 * data from the journal. */
1667 if (jbd2_journal_recover(journal
))
1668 goto recovery_error
;
1670 if (journal
->j_failed_commit
) {
1671 printk(KERN_ERR
"JBD2: journal transaction %u on %s "
1672 "is corrupt.\n", journal
->j_failed_commit
,
1673 journal
->j_devname
);
1674 return -EFSCORRUPTED
;
1677 /* OK, we've finished with the dynamic journal bits:
1678 * reinitialise the dynamic contents of the superblock in memory
1679 * and reset them on disk. */
1680 if (journal_reset(journal
))
1681 goto recovery_error
;
1683 journal
->j_flags
&= ~JBD2_ABORT
;
1684 journal
->j_flags
|= JBD2_LOADED
;
1688 printk(KERN_WARNING
"JBD2: recovery failed\n");
1693 * void jbd2_journal_destroy() - Release a journal_t structure.
1694 * @journal: Journal to act on.
1696 * Release a journal_t structure once it is no longer in use by the
1698 * Return <0 if we couldn't clean up the journal.
1700 int jbd2_journal_destroy(journal_t
*journal
)
1704 /* Wait for the commit thread to wake up and die. */
1705 journal_kill_thread(journal
);
1707 /* Force a final log commit */
1708 if (journal
->j_running_transaction
)
1709 jbd2_journal_commit_transaction(journal
);
1711 /* Force any old transactions to disk */
1713 /* Totally anal locking here... */
1714 spin_lock(&journal
->j_list_lock
);
1715 while (journal
->j_checkpoint_transactions
!= NULL
) {
1716 spin_unlock(&journal
->j_list_lock
);
1717 mutex_lock_io(&journal
->j_checkpoint_mutex
);
1718 err
= jbd2_log_do_checkpoint(journal
);
1719 mutex_unlock(&journal
->j_checkpoint_mutex
);
1721 * If checkpointing failed, just free the buffers to avoid
1725 jbd2_journal_destroy_checkpoint(journal
);
1726 spin_lock(&journal
->j_list_lock
);
1729 spin_lock(&journal
->j_list_lock
);
1732 J_ASSERT(journal
->j_running_transaction
== NULL
);
1733 J_ASSERT(journal
->j_committing_transaction
== NULL
);
1734 J_ASSERT(journal
->j_checkpoint_transactions
== NULL
);
1735 spin_unlock(&journal
->j_list_lock
);
1737 if (journal
->j_sb_buffer
) {
1738 if (!is_journal_aborted(journal
)) {
1739 mutex_lock_io(&journal
->j_checkpoint_mutex
);
1741 write_lock(&journal
->j_state_lock
);
1742 journal
->j_tail_sequence
=
1743 ++journal
->j_transaction_sequence
;
1744 write_unlock(&journal
->j_state_lock
);
1746 jbd2_mark_journal_empty(journal
,
1747 REQ_SYNC
| REQ_PREFLUSH
| REQ_FUA
);
1748 mutex_unlock(&journal
->j_checkpoint_mutex
);
1751 brelse(journal
->j_sb_buffer
);
1754 if (journal
->j_proc_entry
)
1755 jbd2_stats_proc_exit(journal
);
1756 iput(journal
->j_inode
);
1757 if (journal
->j_revoke
)
1758 jbd2_journal_destroy_revoke(journal
);
1759 if (journal
->j_chksum_driver
)
1760 crypto_free_shash(journal
->j_chksum_driver
);
1761 kfree(journal
->j_wbuf
);
1769 *int jbd2_journal_check_used_features () - Check if features specified are used.
1770 * @journal: Journal to check.
1771 * @compat: bitmask of compatible features
1772 * @ro: bitmask of features that force read-only mount
1773 * @incompat: bitmask of incompatible features
1775 * Check whether the journal uses all of a given set of
1776 * features. Return true (non-zero) if it does.
1779 int jbd2_journal_check_used_features (journal_t
*journal
, unsigned long compat
,
1780 unsigned long ro
, unsigned long incompat
)
1782 journal_superblock_t
*sb
;
1784 if (!compat
&& !ro
&& !incompat
)
1786 /* Load journal superblock if it is not loaded yet. */
1787 if (journal
->j_format_version
== 0 &&
1788 journal_get_superblock(journal
) != 0)
1790 if (journal
->j_format_version
== 1)
1793 sb
= journal
->j_superblock
;
1795 if (((be32_to_cpu(sb
->s_feature_compat
) & compat
) == compat
) &&
1796 ((be32_to_cpu(sb
->s_feature_ro_compat
) & ro
) == ro
) &&
1797 ((be32_to_cpu(sb
->s_feature_incompat
) & incompat
) == incompat
))
1804 * int jbd2_journal_check_available_features() - Check feature set in journalling layer
1805 * @journal: Journal to check.
1806 * @compat: bitmask of compatible features
1807 * @ro: bitmask of features that force read-only mount
1808 * @incompat: bitmask of incompatible features
1810 * Check whether the journaling code supports the use of
1811 * all of a given set of features on this journal. Return true
1812 * (non-zero) if it can. */
1814 int jbd2_journal_check_available_features (journal_t
*journal
, unsigned long compat
,
1815 unsigned long ro
, unsigned long incompat
)
1817 if (!compat
&& !ro
&& !incompat
)
1820 /* We can support any known requested features iff the
1821 * superblock is in version 2. Otherwise we fail to support any
1822 * extended sb features. */
1824 if (journal
->j_format_version
!= 2)
1827 if ((compat
& JBD2_KNOWN_COMPAT_FEATURES
) == compat
&&
1828 (ro
& JBD2_KNOWN_ROCOMPAT_FEATURES
) == ro
&&
1829 (incompat
& JBD2_KNOWN_INCOMPAT_FEATURES
) == incompat
)
1836 * int jbd2_journal_set_features () - Mark a given journal feature in the superblock
1837 * @journal: Journal to act on.
1838 * @compat: bitmask of compatible features
1839 * @ro: bitmask of features that force read-only mount
1840 * @incompat: bitmask of incompatible features
1842 * Mark a given journal feature as present on the
1843 * superblock. Returns true if the requested features could be set.
1847 int jbd2_journal_set_features (journal_t
*journal
, unsigned long compat
,
1848 unsigned long ro
, unsigned long incompat
)
1850 #define INCOMPAT_FEATURE_ON(f) \
1851 ((incompat & (f)) && !(sb->s_feature_incompat & cpu_to_be32(f)))
1852 #define COMPAT_FEATURE_ON(f) \
1853 ((compat & (f)) && !(sb->s_feature_compat & cpu_to_be32(f)))
1854 journal_superblock_t
*sb
;
1856 if (jbd2_journal_check_used_features(journal
, compat
, ro
, incompat
))
1859 if (!jbd2_journal_check_available_features(journal
, compat
, ro
, incompat
))
1862 /* If enabling v2 checksums, turn on v3 instead */
1863 if (incompat
& JBD2_FEATURE_INCOMPAT_CSUM_V2
) {
1864 incompat
&= ~JBD2_FEATURE_INCOMPAT_CSUM_V2
;
1865 incompat
|= JBD2_FEATURE_INCOMPAT_CSUM_V3
;
1868 /* Asking for checksumming v3 and v1? Only give them v3. */
1869 if (incompat
& JBD2_FEATURE_INCOMPAT_CSUM_V3
&&
1870 compat
& JBD2_FEATURE_COMPAT_CHECKSUM
)
1871 compat
&= ~JBD2_FEATURE_COMPAT_CHECKSUM
;
1873 jbd_debug(1, "Setting new features 0x%lx/0x%lx/0x%lx\n",
1874 compat
, ro
, incompat
);
1876 sb
= journal
->j_superblock
;
1878 /* If enabling v3 checksums, update superblock */
1879 if (INCOMPAT_FEATURE_ON(JBD2_FEATURE_INCOMPAT_CSUM_V3
)) {
1880 sb
->s_checksum_type
= JBD2_CRC32C_CHKSUM
;
1881 sb
->s_feature_compat
&=
1882 ~cpu_to_be32(JBD2_FEATURE_COMPAT_CHECKSUM
);
1884 /* Load the checksum driver */
1885 if (journal
->j_chksum_driver
== NULL
) {
1886 journal
->j_chksum_driver
= crypto_alloc_shash("crc32c",
1888 if (IS_ERR(journal
->j_chksum_driver
)) {
1889 printk(KERN_ERR
"JBD2: Cannot load crc32c "
1891 journal
->j_chksum_driver
= NULL
;
1895 /* Precompute checksum seed for all metadata */
1896 journal
->j_csum_seed
= jbd2_chksum(journal
, ~0,
1898 sizeof(sb
->s_uuid
));
1902 /* If enabling v1 checksums, downgrade superblock */
1903 if (COMPAT_FEATURE_ON(JBD2_FEATURE_COMPAT_CHECKSUM
))
1904 sb
->s_feature_incompat
&=
1905 ~cpu_to_be32(JBD2_FEATURE_INCOMPAT_CSUM_V2
|
1906 JBD2_FEATURE_INCOMPAT_CSUM_V3
);
1908 sb
->s_feature_compat
|= cpu_to_be32(compat
);
1909 sb
->s_feature_ro_compat
|= cpu_to_be32(ro
);
1910 sb
->s_feature_incompat
|= cpu_to_be32(incompat
);
1913 #undef COMPAT_FEATURE_ON
1914 #undef INCOMPAT_FEATURE_ON
1918 * jbd2_journal_clear_features () - Clear a given journal feature in the
1920 * @journal: Journal to act on.
1921 * @compat: bitmask of compatible features
1922 * @ro: bitmask of features that force read-only mount
1923 * @incompat: bitmask of incompatible features
1925 * Clear a given journal feature as present on the
1928 void jbd2_journal_clear_features(journal_t
*journal
, unsigned long compat
,
1929 unsigned long ro
, unsigned long incompat
)
1931 journal_superblock_t
*sb
;
1933 jbd_debug(1, "Clear features 0x%lx/0x%lx/0x%lx\n",
1934 compat
, ro
, incompat
);
1936 sb
= journal
->j_superblock
;
1938 sb
->s_feature_compat
&= ~cpu_to_be32(compat
);
1939 sb
->s_feature_ro_compat
&= ~cpu_to_be32(ro
);
1940 sb
->s_feature_incompat
&= ~cpu_to_be32(incompat
);
1942 EXPORT_SYMBOL(jbd2_journal_clear_features
);
1945 * int jbd2_journal_flush () - Flush journal
1946 * @journal: Journal to act on.
1948 * Flush all data for a given journal to disk and empty the journal.
1949 * Filesystems can use this when remounting readonly to ensure that
1950 * recovery does not need to happen on remount.
1953 int jbd2_journal_flush(journal_t
*journal
)
1956 transaction_t
*transaction
= NULL
;
1958 write_lock(&journal
->j_state_lock
);
1960 /* Force everything buffered to the log... */
1961 if (journal
->j_running_transaction
) {
1962 transaction
= journal
->j_running_transaction
;
1963 __jbd2_log_start_commit(journal
, transaction
->t_tid
);
1964 } else if (journal
->j_committing_transaction
)
1965 transaction
= journal
->j_committing_transaction
;
1967 /* Wait for the log commit to complete... */
1969 tid_t tid
= transaction
->t_tid
;
1971 write_unlock(&journal
->j_state_lock
);
1972 jbd2_log_wait_commit(journal
, tid
);
1974 write_unlock(&journal
->j_state_lock
);
1977 /* ...and flush everything in the log out to disk. */
1978 spin_lock(&journal
->j_list_lock
);
1979 while (!err
&& journal
->j_checkpoint_transactions
!= NULL
) {
1980 spin_unlock(&journal
->j_list_lock
);
1981 mutex_lock_io(&journal
->j_checkpoint_mutex
);
1982 err
= jbd2_log_do_checkpoint(journal
);
1983 mutex_unlock(&journal
->j_checkpoint_mutex
);
1984 spin_lock(&journal
->j_list_lock
);
1986 spin_unlock(&journal
->j_list_lock
);
1988 if (is_journal_aborted(journal
))
1991 mutex_lock_io(&journal
->j_checkpoint_mutex
);
1993 err
= jbd2_cleanup_journal_tail(journal
);
1995 mutex_unlock(&journal
->j_checkpoint_mutex
);
2001 /* Finally, mark the journal as really needing no recovery.
2002 * This sets s_start==0 in the underlying superblock, which is
2003 * the magic code for a fully-recovered superblock. Any future
2004 * commits of data to the journal will restore the current
2006 jbd2_mark_journal_empty(journal
, REQ_SYNC
| REQ_FUA
);
2007 mutex_unlock(&journal
->j_checkpoint_mutex
);
2008 write_lock(&journal
->j_state_lock
);
2009 J_ASSERT(!journal
->j_running_transaction
);
2010 J_ASSERT(!journal
->j_committing_transaction
);
2011 J_ASSERT(!journal
->j_checkpoint_transactions
);
2012 J_ASSERT(journal
->j_head
== journal
->j_tail
);
2013 J_ASSERT(journal
->j_tail_sequence
== journal
->j_transaction_sequence
);
2014 write_unlock(&journal
->j_state_lock
);
2020 * int jbd2_journal_wipe() - Wipe journal contents
2021 * @journal: Journal to act on.
2022 * @write: flag (see below)
2024 * Wipe out all of the contents of a journal, safely. This will produce
2025 * a warning if the journal contains any valid recovery information.
2026 * Must be called between journal_init_*() and jbd2_journal_load().
2028 * If 'write' is non-zero, then we wipe out the journal on disk; otherwise
2029 * we merely suppress recovery.
2032 int jbd2_journal_wipe(journal_t
*journal
, int write
)
2036 J_ASSERT (!(journal
->j_flags
& JBD2_LOADED
));
2038 err
= load_superblock(journal
);
2042 if (!journal
->j_tail
)
2045 printk(KERN_WARNING
"JBD2: %s recovery information on journal\n",
2046 write
? "Clearing" : "Ignoring");
2048 err
= jbd2_journal_skip_recovery(journal
);
2050 /* Lock to make assertions happy... */
2051 mutex_lock(&journal
->j_checkpoint_mutex
);
2052 jbd2_mark_journal_empty(journal
, REQ_SYNC
| REQ_FUA
);
2053 mutex_unlock(&journal
->j_checkpoint_mutex
);
2061 * Journal abort has very specific semantics, which we describe
2062 * for journal abort.
2064 * Two internal functions, which provide abort to the jbd layer
2069 * Quick version for internal journal use (doesn't lock the journal).
2070 * Aborts hard --- we mark the abort as occurred, but do _nothing_ else,
2071 * and don't attempt to make any other journal updates.
2073 void __jbd2_journal_abort_hard(journal_t
*journal
)
2075 transaction_t
*transaction
;
2077 if (journal
->j_flags
& JBD2_ABORT
)
2080 printk(KERN_ERR
"Aborting journal on device %s.\n",
2081 journal
->j_devname
);
2083 write_lock(&journal
->j_state_lock
);
2084 journal
->j_flags
|= JBD2_ABORT
;
2085 transaction
= journal
->j_running_transaction
;
2087 __jbd2_log_start_commit(journal
, transaction
->t_tid
);
2088 write_unlock(&journal
->j_state_lock
);
2091 /* Soft abort: record the abort error status in the journal superblock,
2092 * but don't do any other IO. */
2093 static void __journal_abort_soft (journal_t
*journal
, int errno
)
2095 if (journal
->j_flags
& JBD2_ABORT
)
2098 if (!journal
->j_errno
)
2099 journal
->j_errno
= errno
;
2101 __jbd2_journal_abort_hard(journal
);
2104 jbd2_journal_update_sb_errno(journal
);
2105 write_lock(&journal
->j_state_lock
);
2106 journal
->j_flags
|= JBD2_REC_ERR
;
2107 write_unlock(&journal
->j_state_lock
);
2112 * void jbd2_journal_abort () - Shutdown the journal immediately.
2113 * @journal: the journal to shutdown.
2114 * @errno: an error number to record in the journal indicating
2115 * the reason for the shutdown.
2117 * Perform a complete, immediate shutdown of the ENTIRE
2118 * journal (not of a single transaction). This operation cannot be
2119 * undone without closing and reopening the journal.
2121 * The jbd2_journal_abort function is intended to support higher level error
2122 * recovery mechanisms such as the ext2/ext3 remount-readonly error
2125 * Journal abort has very specific semantics. Any existing dirty,
2126 * unjournaled buffers in the main filesystem will still be written to
2127 * disk by bdflush, but the journaling mechanism will be suspended
2128 * immediately and no further transaction commits will be honoured.
2130 * Any dirty, journaled buffers will be written back to disk without
2131 * hitting the journal. Atomicity cannot be guaranteed on an aborted
2132 * filesystem, but we _do_ attempt to leave as much data as possible
2133 * behind for fsck to use for cleanup.
2135 * Any attempt to get a new transaction handle on a journal which is in
2136 * ABORT state will just result in an -EROFS error return. A
2137 * jbd2_journal_stop on an existing handle will return -EIO if we have
2138 * entered abort state during the update.
2140 * Recursive transactions are not disturbed by journal abort until the
2141 * final jbd2_journal_stop, which will receive the -EIO error.
2143 * Finally, the jbd2_journal_abort call allows the caller to supply an errno
2144 * which will be recorded (if possible) in the journal superblock. This
2145 * allows a client to record failure conditions in the middle of a
2146 * transaction without having to complete the transaction to record the
2147 * failure to disk. ext3_error, for example, now uses this
2150 * Errors which originate from within the journaling layer will NOT
2151 * supply an errno; a null errno implies that absolutely no further
2152 * writes are done to the journal (unless there are any already in
2157 void jbd2_journal_abort(journal_t
*journal
, int errno
)
2159 __journal_abort_soft(journal
, errno
);
2163 * int jbd2_journal_errno () - returns the journal's error state.
2164 * @journal: journal to examine.
2166 * This is the errno number set with jbd2_journal_abort(), the last
2167 * time the journal was mounted - if the journal was stopped
2168 * without calling abort this will be 0.
2170 * If the journal has been aborted on this mount time -EROFS will
2173 int jbd2_journal_errno(journal_t
*journal
)
2177 read_lock(&journal
->j_state_lock
);
2178 if (journal
->j_flags
& JBD2_ABORT
)
2181 err
= journal
->j_errno
;
2182 read_unlock(&journal
->j_state_lock
);
2187 * int jbd2_journal_clear_err () - clears the journal's error state
2188 * @journal: journal to act on.
2190 * An error must be cleared or acked to take a FS out of readonly
2193 int jbd2_journal_clear_err(journal_t
*journal
)
2197 write_lock(&journal
->j_state_lock
);
2198 if (journal
->j_flags
& JBD2_ABORT
)
2201 journal
->j_errno
= 0;
2202 write_unlock(&journal
->j_state_lock
);
2207 * void jbd2_journal_ack_err() - Ack journal err.
2208 * @journal: journal to act on.
2210 * An error must be cleared or acked to take a FS out of readonly
2213 void jbd2_journal_ack_err(journal_t
*journal
)
2215 write_lock(&journal
->j_state_lock
);
2216 if (journal
->j_errno
)
2217 journal
->j_flags
|= JBD2_ACK_ERR
;
2218 write_unlock(&journal
->j_state_lock
);
2221 int jbd2_journal_blocks_per_page(struct inode
*inode
)
2223 return 1 << (PAGE_SHIFT
- inode
->i_sb
->s_blocksize_bits
);
2227 * helper functions to deal with 32 or 64bit block numbers.
2229 size_t journal_tag_bytes(journal_t
*journal
)
2233 if (jbd2_has_feature_csum3(journal
))
2234 return sizeof(journal_block_tag3_t
);
2236 sz
= sizeof(journal_block_tag_t
);
2238 if (jbd2_has_feature_csum2(journal
))
2239 sz
+= sizeof(__u16
);
2241 if (jbd2_has_feature_64bit(journal
))
2244 return sz
- sizeof(__u32
);
2248 * JBD memory management
2250 * These functions are used to allocate block-sized chunks of memory
2251 * used for making copies of buffer_head data. Very often it will be
2252 * page-sized chunks of data, but sometimes it will be in
2253 * sub-page-size chunks. (For example, 16k pages on Power systems
2254 * with a 4k block file system.) For blocks smaller than a page, we
2255 * use a SLAB allocator. There are slab caches for each block size,
2256 * which are allocated at mount time, if necessary, and we only free
2257 * (all of) the slab caches when/if the jbd2 module is unloaded. For
2258 * this reason we don't need to a mutex to protect access to
2259 * jbd2_slab[] allocating or releasing memory; only in
2260 * jbd2_journal_create_slab().
2262 #define JBD2_MAX_SLABS 8
2263 static struct kmem_cache
*jbd2_slab
[JBD2_MAX_SLABS
];
2265 static const char *jbd2_slab_names
[JBD2_MAX_SLABS
] = {
2266 "jbd2_1k", "jbd2_2k", "jbd2_4k", "jbd2_8k",
2267 "jbd2_16k", "jbd2_32k", "jbd2_64k", "jbd2_128k"
2271 static void jbd2_journal_destroy_slabs(void)
2275 for (i
= 0; i
< JBD2_MAX_SLABS
; i
++) {
2277 kmem_cache_destroy(jbd2_slab
[i
]);
2278 jbd2_slab
[i
] = NULL
;
2282 static int jbd2_journal_create_slab(size_t size
)
2284 static DEFINE_MUTEX(jbd2_slab_create_mutex
);
2285 int i
= order_base_2(size
) - 10;
2288 if (size
== PAGE_SIZE
)
2291 if (i
>= JBD2_MAX_SLABS
)
2294 if (unlikely(i
< 0))
2296 mutex_lock(&jbd2_slab_create_mutex
);
2298 mutex_unlock(&jbd2_slab_create_mutex
);
2299 return 0; /* Already created */
2302 slab_size
= 1 << (i
+10);
2303 jbd2_slab
[i
] = kmem_cache_create(jbd2_slab_names
[i
], slab_size
,
2304 slab_size
, 0, NULL
);
2305 mutex_unlock(&jbd2_slab_create_mutex
);
2306 if (!jbd2_slab
[i
]) {
2307 printk(KERN_EMERG
"JBD2: no memory for jbd2_slab cache\n");
2313 static struct kmem_cache
*get_slab(size_t size
)
2315 int i
= order_base_2(size
) - 10;
2317 BUG_ON(i
>= JBD2_MAX_SLABS
);
2318 if (unlikely(i
< 0))
2320 BUG_ON(jbd2_slab
[i
] == NULL
);
2321 return jbd2_slab
[i
];
2324 void *jbd2_alloc(size_t size
, gfp_t flags
)
2328 BUG_ON(size
& (size
-1)); /* Must be a power of 2 */
2330 if (size
< PAGE_SIZE
)
2331 ptr
= kmem_cache_alloc(get_slab(size
), flags
);
2333 ptr
= (void *)__get_free_pages(flags
, get_order(size
));
2335 /* Check alignment; SLUB has gotten this wrong in the past,
2336 * and this can lead to user data corruption! */
2337 BUG_ON(((unsigned long) ptr
) & (size
-1));
2342 void jbd2_free(void *ptr
, size_t size
)
2344 if (size
< PAGE_SIZE
)
2345 kmem_cache_free(get_slab(size
), ptr
);
2347 free_pages((unsigned long)ptr
, get_order(size
));
2351 * Journal_head storage management
2353 static struct kmem_cache
*jbd2_journal_head_cache
;
2354 #ifdef CONFIG_JBD2_DEBUG
2355 static atomic_t nr_journal_heads
= ATOMIC_INIT(0);
2358 static int jbd2_journal_init_journal_head_cache(void)
2362 J_ASSERT(jbd2_journal_head_cache
== NULL
);
2363 jbd2_journal_head_cache
= kmem_cache_create("jbd2_journal_head",
2364 sizeof(struct journal_head
),
2366 SLAB_TEMPORARY
| SLAB_TYPESAFE_BY_RCU
,
2369 if (!jbd2_journal_head_cache
) {
2371 printk(KERN_EMERG
"JBD2: no memory for journal_head cache\n");
2376 static void jbd2_journal_destroy_journal_head_cache(void)
2378 if (jbd2_journal_head_cache
) {
2379 kmem_cache_destroy(jbd2_journal_head_cache
);
2380 jbd2_journal_head_cache
= NULL
;
2385 * journal_head splicing and dicing
2387 static struct journal_head
*journal_alloc_journal_head(void)
2389 struct journal_head
*ret
;
2391 #ifdef CONFIG_JBD2_DEBUG
2392 atomic_inc(&nr_journal_heads
);
2394 ret
= kmem_cache_zalloc(jbd2_journal_head_cache
, GFP_NOFS
);
2396 jbd_debug(1, "out of memory for journal_head\n");
2397 pr_notice_ratelimited("ENOMEM in %s, retrying.\n", __func__
);
2398 ret
= kmem_cache_zalloc(jbd2_journal_head_cache
,
2399 GFP_NOFS
| __GFP_NOFAIL
);
2404 static void journal_free_journal_head(struct journal_head
*jh
)
2406 #ifdef CONFIG_JBD2_DEBUG
2407 atomic_dec(&nr_journal_heads
);
2408 memset(jh
, JBD2_POISON_FREE
, sizeof(*jh
));
2410 kmem_cache_free(jbd2_journal_head_cache
, jh
);
2414 * A journal_head is attached to a buffer_head whenever JBD has an
2415 * interest in the buffer.
2417 * Whenever a buffer has an attached journal_head, its ->b_state:BH_JBD bit
2418 * is set. This bit is tested in core kernel code where we need to take
2419 * JBD-specific actions. Testing the zeroness of ->b_private is not reliable
2422 * When a buffer has its BH_JBD bit set, its ->b_count is elevated by one.
2424 * When a buffer has its BH_JBD bit set it is immune from being released by
2425 * core kernel code, mainly via ->b_count.
2427 * A journal_head is detached from its buffer_head when the journal_head's
2428 * b_jcount reaches zero. Running transaction (b_transaction) and checkpoint
2429 * transaction (b_cp_transaction) hold their references to b_jcount.
2431 * Various places in the kernel want to attach a journal_head to a buffer_head
2432 * _before_ attaching the journal_head to a transaction. To protect the
2433 * journal_head in this situation, jbd2_journal_add_journal_head elevates the
2434 * journal_head's b_jcount refcount by one. The caller must call
2435 * jbd2_journal_put_journal_head() to undo this.
2437 * So the typical usage would be:
2439 * (Attach a journal_head if needed. Increments b_jcount)
2440 * struct journal_head *jh = jbd2_journal_add_journal_head(bh);
2442 * (Get another reference for transaction)
2443 * jbd2_journal_grab_journal_head(bh);
2444 * jh->b_transaction = xxx;
2445 * (Put original reference)
2446 * jbd2_journal_put_journal_head(jh);
2450 * Give a buffer_head a journal_head.
2454 struct journal_head
*jbd2_journal_add_journal_head(struct buffer_head
*bh
)
2456 struct journal_head
*jh
;
2457 struct journal_head
*new_jh
= NULL
;
2460 if (!buffer_jbd(bh
))
2461 new_jh
= journal_alloc_journal_head();
2463 jbd_lock_bh_journal_head(bh
);
2464 if (buffer_jbd(bh
)) {
2468 (atomic_read(&bh
->b_count
) > 0) ||
2469 (bh
->b_page
&& bh
->b_page
->mapping
));
2472 jbd_unlock_bh_journal_head(bh
);
2477 new_jh
= NULL
; /* We consumed it */
2482 BUFFER_TRACE(bh
, "added journal_head");
2485 jbd_unlock_bh_journal_head(bh
);
2487 journal_free_journal_head(new_jh
);
2488 return bh
->b_private
;
2492 * Grab a ref against this buffer_head's journal_head. If it ended up not
2493 * having a journal_head, return NULL
2495 struct journal_head
*jbd2_journal_grab_journal_head(struct buffer_head
*bh
)
2497 struct journal_head
*jh
= NULL
;
2499 jbd_lock_bh_journal_head(bh
);
2500 if (buffer_jbd(bh
)) {
2504 jbd_unlock_bh_journal_head(bh
);
2508 static void __journal_remove_journal_head(struct buffer_head
*bh
)
2510 struct journal_head
*jh
= bh2jh(bh
);
2512 J_ASSERT_JH(jh
, jh
->b_jcount
>= 0);
2513 J_ASSERT_JH(jh
, jh
->b_transaction
== NULL
);
2514 J_ASSERT_JH(jh
, jh
->b_next_transaction
== NULL
);
2515 J_ASSERT_JH(jh
, jh
->b_cp_transaction
== NULL
);
2516 J_ASSERT_JH(jh
, jh
->b_jlist
== BJ_None
);
2517 J_ASSERT_BH(bh
, buffer_jbd(bh
));
2518 J_ASSERT_BH(bh
, jh2bh(jh
) == bh
);
2519 BUFFER_TRACE(bh
, "remove journal_head");
2520 if (jh
->b_frozen_data
) {
2521 printk(KERN_WARNING
"%s: freeing b_frozen_data\n", __func__
);
2522 jbd2_free(jh
->b_frozen_data
, bh
->b_size
);
2524 if (jh
->b_committed_data
) {
2525 printk(KERN_WARNING
"%s: freeing b_committed_data\n", __func__
);
2526 jbd2_free(jh
->b_committed_data
, bh
->b_size
);
2528 bh
->b_private
= NULL
;
2529 jh
->b_bh
= NULL
; /* debug, really */
2530 clear_buffer_jbd(bh
);
2531 journal_free_journal_head(jh
);
2535 * Drop a reference on the passed journal_head. If it fell to zero then
2536 * release the journal_head from the buffer_head.
2538 void jbd2_journal_put_journal_head(struct journal_head
*jh
)
2540 struct buffer_head
*bh
= jh2bh(jh
);
2542 jbd_lock_bh_journal_head(bh
);
2543 J_ASSERT_JH(jh
, jh
->b_jcount
> 0);
2545 if (!jh
->b_jcount
) {
2546 __journal_remove_journal_head(bh
);
2547 jbd_unlock_bh_journal_head(bh
);
2550 jbd_unlock_bh_journal_head(bh
);
2554 * Initialize jbd inode head
2556 void jbd2_journal_init_jbd_inode(struct jbd2_inode
*jinode
, struct inode
*inode
)
2558 jinode
->i_transaction
= NULL
;
2559 jinode
->i_next_transaction
= NULL
;
2560 jinode
->i_vfs_inode
= inode
;
2561 jinode
->i_flags
= 0;
2562 INIT_LIST_HEAD(&jinode
->i_list
);
2566 * Function to be called before we start removing inode from memory (i.e.,
2567 * clear_inode() is a fine place to be called from). It removes inode from
2568 * transaction's lists.
2570 void jbd2_journal_release_jbd_inode(journal_t
*journal
,
2571 struct jbd2_inode
*jinode
)
2576 spin_lock(&journal
->j_list_lock
);
2577 /* Is commit writing out inode - we have to wait */
2578 if (jinode
->i_flags
& JI_COMMIT_RUNNING
) {
2579 wait_queue_head_t
*wq
;
2580 DEFINE_WAIT_BIT(wait
, &jinode
->i_flags
, __JI_COMMIT_RUNNING
);
2581 wq
= bit_waitqueue(&jinode
->i_flags
, __JI_COMMIT_RUNNING
);
2582 prepare_to_wait(wq
, &wait
.wait
, TASK_UNINTERRUPTIBLE
);
2583 spin_unlock(&journal
->j_list_lock
);
2585 finish_wait(wq
, &wait
.wait
);
2589 if (jinode
->i_transaction
) {
2590 list_del(&jinode
->i_list
);
2591 jinode
->i_transaction
= NULL
;
2593 spin_unlock(&journal
->j_list_lock
);
2597 #ifdef CONFIG_PROC_FS
2599 #define JBD2_STATS_PROC_NAME "fs/jbd2"
2601 static void __init
jbd2_create_jbd_stats_proc_entry(void)
2603 proc_jbd2_stats
= proc_mkdir(JBD2_STATS_PROC_NAME
, NULL
);
2606 static void __exit
jbd2_remove_jbd_stats_proc_entry(void)
2608 if (proc_jbd2_stats
)
2609 remove_proc_entry(JBD2_STATS_PROC_NAME
, NULL
);
2614 #define jbd2_create_jbd_stats_proc_entry() do {} while (0)
2615 #define jbd2_remove_jbd_stats_proc_entry() do {} while (0)
2619 struct kmem_cache
*jbd2_handle_cache
, *jbd2_inode_cache
;
2621 static int __init
jbd2_journal_init_handle_cache(void)
2623 jbd2_handle_cache
= KMEM_CACHE(jbd2_journal_handle
, SLAB_TEMPORARY
);
2624 if (jbd2_handle_cache
== NULL
) {
2625 printk(KERN_EMERG
"JBD2: failed to create handle cache\n");
2628 jbd2_inode_cache
= KMEM_CACHE(jbd2_inode
, 0);
2629 if (jbd2_inode_cache
== NULL
) {
2630 printk(KERN_EMERG
"JBD2: failed to create inode cache\n");
2631 kmem_cache_destroy(jbd2_handle_cache
);
2637 static void jbd2_journal_destroy_handle_cache(void)
2639 if (jbd2_handle_cache
)
2640 kmem_cache_destroy(jbd2_handle_cache
);
2641 if (jbd2_inode_cache
)
2642 kmem_cache_destroy(jbd2_inode_cache
);
2647 * Module startup and shutdown
2650 static int __init
journal_init_caches(void)
2654 ret
= jbd2_journal_init_revoke_caches();
2656 ret
= jbd2_journal_init_journal_head_cache();
2658 ret
= jbd2_journal_init_handle_cache();
2660 ret
= jbd2_journal_init_transaction_cache();
2664 static void jbd2_journal_destroy_caches(void)
2666 jbd2_journal_destroy_revoke_caches();
2667 jbd2_journal_destroy_journal_head_cache();
2668 jbd2_journal_destroy_handle_cache();
2669 jbd2_journal_destroy_transaction_cache();
2670 jbd2_journal_destroy_slabs();
2673 static int __init
journal_init(void)
2677 BUILD_BUG_ON(sizeof(struct journal_superblock_s
) != 1024);
2679 ret
= journal_init_caches();
2681 jbd2_create_jbd_stats_proc_entry();
2683 jbd2_journal_destroy_caches();
2688 static void __exit
journal_exit(void)
2690 #ifdef CONFIG_JBD2_DEBUG
2691 int n
= atomic_read(&nr_journal_heads
);
2693 printk(KERN_ERR
"JBD2: leaked %d journal_heads!\n", n
);
2695 jbd2_remove_jbd_stats_proc_entry();
2696 jbd2_journal_destroy_caches();
2699 MODULE_LICENSE("GPL");
2700 module_init(journal_init
);
2701 module_exit(journal_exit
);