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
);
69 EXPORT_SYMBOL(jbd2_journal_flush
);
70 EXPORT_SYMBOL(jbd2_journal_revoke
);
72 EXPORT_SYMBOL(jbd2_journal_init_dev
);
73 EXPORT_SYMBOL(jbd2_journal_init_inode
);
74 EXPORT_SYMBOL(jbd2_journal_check_used_features
);
75 EXPORT_SYMBOL(jbd2_journal_check_available_features
);
76 EXPORT_SYMBOL(jbd2_journal_set_features
);
77 EXPORT_SYMBOL(jbd2_journal_load
);
78 EXPORT_SYMBOL(jbd2_journal_destroy
);
79 EXPORT_SYMBOL(jbd2_journal_abort
);
80 EXPORT_SYMBOL(jbd2_journal_errno
);
81 EXPORT_SYMBOL(jbd2_journal_ack_err
);
82 EXPORT_SYMBOL(jbd2_journal_clear_err
);
83 EXPORT_SYMBOL(jbd2_log_wait_commit
);
84 EXPORT_SYMBOL(jbd2_log_start_commit
);
85 EXPORT_SYMBOL(jbd2_journal_start_commit
);
86 EXPORT_SYMBOL(jbd2_journal_force_commit_nested
);
87 EXPORT_SYMBOL(jbd2_journal_wipe
);
88 EXPORT_SYMBOL(jbd2_journal_blocks_per_page
);
89 EXPORT_SYMBOL(jbd2_journal_invalidatepage
);
90 EXPORT_SYMBOL(jbd2_journal_try_to_free_buffers
);
91 EXPORT_SYMBOL(jbd2_journal_force_commit
);
92 EXPORT_SYMBOL(jbd2_journal_inode_ranged_write
);
93 EXPORT_SYMBOL(jbd2_journal_inode_ranged_wait
);
94 EXPORT_SYMBOL(jbd2_journal_submit_inode_data_buffers
);
95 EXPORT_SYMBOL(jbd2_journal_finish_inode_data_buffers
);
96 EXPORT_SYMBOL(jbd2_journal_init_jbd_inode
);
97 EXPORT_SYMBOL(jbd2_journal_release_jbd_inode
);
98 EXPORT_SYMBOL(jbd2_journal_begin_ordered_truncate
);
99 EXPORT_SYMBOL(jbd2_inode_cache
);
101 static int jbd2_journal_create_slab(size_t slab_size
);
103 #ifdef CONFIG_JBD2_DEBUG
104 void __jbd2_debug(int level
, const char *file
, const char *func
,
105 unsigned int line
, const char *fmt
, ...)
107 struct va_format vaf
;
110 if (level
> jbd2_journal_enable_debug
)
115 printk(KERN_DEBUG
"%s: (%s, %u): %pV", file
, func
, line
, &vaf
);
118 EXPORT_SYMBOL(__jbd2_debug
);
121 /* Checksumming functions */
122 static int jbd2_verify_csum_type(journal_t
*j
, journal_superblock_t
*sb
)
124 if (!jbd2_journal_has_csum_v2or3_feature(j
))
127 return sb
->s_checksum_type
== JBD2_CRC32C_CHKSUM
;
130 static __be32
jbd2_superblock_csum(journal_t
*j
, journal_superblock_t
*sb
)
135 old_csum
= sb
->s_checksum
;
137 csum
= jbd2_chksum(j
, ~0, (char *)sb
, sizeof(journal_superblock_t
));
138 sb
->s_checksum
= old_csum
;
140 return cpu_to_be32(csum
);
144 * Helper function used to manage commit timeouts
147 static void commit_timeout(struct timer_list
*t
)
149 journal_t
*journal
= from_timer(journal
, t
, j_commit_timer
);
151 wake_up_process(journal
->j_task
);
155 * kjournald2: The main thread function used to manage a logging device
158 * This kernel thread is responsible for two things:
160 * 1) COMMIT: Every so often we need to commit the current state of the
161 * filesystem to disk. The journal thread is responsible for writing
162 * all of the metadata buffers to disk. If a fast commit is ongoing
163 * journal thread waits until it's done and then continues from
166 * 2) CHECKPOINT: We cannot reuse a used section of the log file until all
167 * of the data in that part of the log has been rewritten elsewhere on
168 * the disk. Flushing these old buffers to reclaim space in the log is
169 * known as checkpointing, and this thread is responsible for that job.
172 static int kjournald2(void *arg
)
174 journal_t
*journal
= arg
;
175 transaction_t
*transaction
;
178 * Set up an interval timer which can be used to trigger a commit wakeup
179 * after the commit interval expires
181 timer_setup(&journal
->j_commit_timer
, commit_timeout
, 0);
185 /* Record that the journal thread is running */
186 journal
->j_task
= current
;
187 wake_up(&journal
->j_wait_done_commit
);
190 * Make sure that no allocations from this kernel thread will ever
191 * recurse to the fs layer because we are responsible for the
192 * transaction commit and any fs involvement might get stuck waiting for
195 memalloc_nofs_save();
198 * And now, wait forever for commit wakeup events.
200 write_lock(&journal
->j_state_lock
);
203 if (journal
->j_flags
& JBD2_UNMOUNT
)
206 jbd_debug(1, "commit_sequence=%u, commit_request=%u\n",
207 journal
->j_commit_sequence
, journal
->j_commit_request
);
209 if (journal
->j_commit_sequence
!= journal
->j_commit_request
) {
210 jbd_debug(1, "OK, requests differ\n");
211 write_unlock(&journal
->j_state_lock
);
212 del_timer_sync(&journal
->j_commit_timer
);
213 jbd2_journal_commit_transaction(journal
);
214 write_lock(&journal
->j_state_lock
);
218 wake_up(&journal
->j_wait_done_commit
);
219 if (freezing(current
)) {
221 * The simpler the better. Flushing journal isn't a
222 * good idea, because that depends on threads that may
223 * be already stopped.
225 jbd_debug(1, "Now suspending kjournald2\n");
226 write_unlock(&journal
->j_state_lock
);
228 write_lock(&journal
->j_state_lock
);
231 * We assume on resume that commits are already there,
235 int should_sleep
= 1;
237 prepare_to_wait(&journal
->j_wait_commit
, &wait
,
239 if (journal
->j_commit_sequence
!= journal
->j_commit_request
)
241 transaction
= journal
->j_running_transaction
;
242 if (transaction
&& time_after_eq(jiffies
,
243 transaction
->t_expires
))
245 if (journal
->j_flags
& JBD2_UNMOUNT
)
248 write_unlock(&journal
->j_state_lock
);
250 write_lock(&journal
->j_state_lock
);
252 finish_wait(&journal
->j_wait_commit
, &wait
);
255 jbd_debug(1, "kjournald2 wakes\n");
258 * Were we woken up by a commit wakeup event?
260 transaction
= journal
->j_running_transaction
;
261 if (transaction
&& time_after_eq(jiffies
, transaction
->t_expires
)) {
262 journal
->j_commit_request
= transaction
->t_tid
;
263 jbd_debug(1, "woke because of timeout\n");
268 del_timer_sync(&journal
->j_commit_timer
);
269 journal
->j_task
= NULL
;
270 wake_up(&journal
->j_wait_done_commit
);
271 jbd_debug(1, "Journal thread exiting.\n");
272 write_unlock(&journal
->j_state_lock
);
276 static int jbd2_journal_start_thread(journal_t
*journal
)
278 struct task_struct
*t
;
280 t
= kthread_run(kjournald2
, journal
, "jbd2/%s",
285 wait_event(journal
->j_wait_done_commit
, journal
->j_task
!= NULL
);
289 static void journal_kill_thread(journal_t
*journal
)
291 write_lock(&journal
->j_state_lock
);
292 journal
->j_flags
|= JBD2_UNMOUNT
;
294 while (journal
->j_task
) {
295 write_unlock(&journal
->j_state_lock
);
296 wake_up(&journal
->j_wait_commit
);
297 wait_event(journal
->j_wait_done_commit
, journal
->j_task
== NULL
);
298 write_lock(&journal
->j_state_lock
);
300 write_unlock(&journal
->j_state_lock
);
304 * jbd2_journal_write_metadata_buffer: write a metadata buffer to the journal.
306 * Writes a metadata buffer to a given disk block. The actual IO is not
307 * performed but a new buffer_head is constructed which labels the data
308 * to be written with the correct destination disk block.
310 * Any magic-number escaping which needs to be done will cause a
311 * copy-out here. If the buffer happens to start with the
312 * JBD2_MAGIC_NUMBER, then we can't write it to the log directly: the
313 * magic number is only written to the log for descripter blocks. In
314 * this case, we copy the data and replace the first word with 0, and we
315 * return a result code which indicates that this buffer needs to be
316 * marked as an escaped buffer in the corresponding log descriptor
317 * block. The missing word can then be restored when the block is read
320 * If the source buffer has already been modified by a new transaction
321 * since we took the last commit snapshot, we use the frozen copy of
322 * that data for IO. If we end up using the existing buffer_head's data
323 * for the write, then we have to make sure nobody modifies it while the
324 * IO is in progress. do_get_write_access() handles this.
326 * The function returns a pointer to the buffer_head to be used for IO.
334 * Bit 0 set == escape performed on the data
335 * Bit 1 set == buffer copy-out performed (kfree the data after IO)
338 int jbd2_journal_write_metadata_buffer(transaction_t
*transaction
,
339 struct journal_head
*jh_in
,
340 struct buffer_head
**bh_out
,
343 int need_copy_out
= 0;
344 int done_copy_out
= 0;
347 struct buffer_head
*new_bh
;
348 struct page
*new_page
;
349 unsigned int new_offset
;
350 struct buffer_head
*bh_in
= jh2bh(jh_in
);
351 journal_t
*journal
= transaction
->t_journal
;
354 * The buffer really shouldn't be locked: only the current committing
355 * transaction is allowed to write it, so nobody else is allowed
358 * akpm: except if we're journalling data, and write() output is
359 * also part of a shared mapping, and another thread has
360 * decided to launch a writepage() against this buffer.
362 J_ASSERT_BH(bh_in
, buffer_jbddirty(bh_in
));
364 new_bh
= alloc_buffer_head(GFP_NOFS
|__GFP_NOFAIL
);
366 /* keep subsequent assertions sane */
367 atomic_set(&new_bh
->b_count
, 1);
369 spin_lock(&jh_in
->b_state_lock
);
372 * If a new transaction has already done a buffer copy-out, then
373 * we use that version of the data for the commit.
375 if (jh_in
->b_frozen_data
) {
377 new_page
= virt_to_page(jh_in
->b_frozen_data
);
378 new_offset
= offset_in_page(jh_in
->b_frozen_data
);
380 new_page
= jh2bh(jh_in
)->b_page
;
381 new_offset
= offset_in_page(jh2bh(jh_in
)->b_data
);
384 mapped_data
= kmap_atomic(new_page
);
386 * Fire data frozen trigger if data already wasn't frozen. Do this
387 * before checking for escaping, as the trigger may modify the magic
388 * offset. If a copy-out happens afterwards, it will have the correct
389 * data in the buffer.
392 jbd2_buffer_frozen_trigger(jh_in
, mapped_data
+ new_offset
,
398 if (*((__be32
*)(mapped_data
+ new_offset
)) ==
399 cpu_to_be32(JBD2_MAGIC_NUMBER
)) {
403 kunmap_atomic(mapped_data
);
406 * Do we need to do a data copy?
408 if (need_copy_out
&& !done_copy_out
) {
411 spin_unlock(&jh_in
->b_state_lock
);
412 tmp
= jbd2_alloc(bh_in
->b_size
, GFP_NOFS
);
417 spin_lock(&jh_in
->b_state_lock
);
418 if (jh_in
->b_frozen_data
) {
419 jbd2_free(tmp
, bh_in
->b_size
);
423 jh_in
->b_frozen_data
= tmp
;
424 mapped_data
= kmap_atomic(new_page
);
425 memcpy(tmp
, mapped_data
+ new_offset
, bh_in
->b_size
);
426 kunmap_atomic(mapped_data
);
428 new_page
= virt_to_page(tmp
);
429 new_offset
= offset_in_page(tmp
);
433 * This isn't strictly necessary, as we're using frozen
434 * data for the escaping, but it keeps consistency with
435 * b_frozen_data usage.
437 jh_in
->b_frozen_triggers
= jh_in
->b_triggers
;
441 * Did we need to do an escaping? Now we've done all the
442 * copying, we can finally do so.
445 mapped_data
= kmap_atomic(new_page
);
446 *((unsigned int *)(mapped_data
+ new_offset
)) = 0;
447 kunmap_atomic(mapped_data
);
450 set_bh_page(new_bh
, new_page
, new_offset
);
451 new_bh
->b_size
= bh_in
->b_size
;
452 new_bh
->b_bdev
= journal
->j_dev
;
453 new_bh
->b_blocknr
= blocknr
;
454 new_bh
->b_private
= bh_in
;
455 set_buffer_mapped(new_bh
);
456 set_buffer_dirty(new_bh
);
461 * The to-be-written buffer needs to get moved to the io queue,
462 * and the original buffer whose contents we are shadowing or
463 * copying is moved to the transaction's shadow queue.
465 JBUFFER_TRACE(jh_in
, "file as BJ_Shadow");
466 spin_lock(&journal
->j_list_lock
);
467 __jbd2_journal_file_buffer(jh_in
, transaction
, BJ_Shadow
);
468 spin_unlock(&journal
->j_list_lock
);
469 set_buffer_shadow(bh_in
);
470 spin_unlock(&jh_in
->b_state_lock
);
472 return do_escape
| (done_copy_out
<< 1);
476 * Allocation code for the journal file. Manage the space left in the
477 * journal, so that we can begin checkpointing when appropriate.
481 * Called with j_state_lock locked for writing.
482 * Returns true if a transaction commit was started.
484 int __jbd2_log_start_commit(journal_t
*journal
, tid_t target
)
486 /* Return if the txn has already requested to be committed */
487 if (journal
->j_commit_request
== target
)
491 * The only transaction we can possibly wait upon is the
492 * currently running transaction (if it exists). Otherwise,
493 * the target tid must be an old one.
495 if (journal
->j_running_transaction
&&
496 journal
->j_running_transaction
->t_tid
== target
) {
498 * We want a new commit: OK, mark the request and wakeup the
499 * commit thread. We do _not_ do the commit ourselves.
502 journal
->j_commit_request
= target
;
503 jbd_debug(1, "JBD2: requesting commit %u/%u\n",
504 journal
->j_commit_request
,
505 journal
->j_commit_sequence
);
506 journal
->j_running_transaction
->t_requested
= jiffies
;
507 wake_up(&journal
->j_wait_commit
);
509 } else if (!tid_geq(journal
->j_commit_request
, target
))
510 /* This should never happen, but if it does, preserve
511 the evidence before kjournald goes into a loop and
512 increments j_commit_sequence beyond all recognition. */
513 WARN_ONCE(1, "JBD2: bad log_start_commit: %u %u %u %u\n",
514 journal
->j_commit_request
,
515 journal
->j_commit_sequence
,
516 target
, journal
->j_running_transaction
?
517 journal
->j_running_transaction
->t_tid
: 0);
521 int jbd2_log_start_commit(journal_t
*journal
, tid_t tid
)
525 write_lock(&journal
->j_state_lock
);
526 ret
= __jbd2_log_start_commit(journal
, tid
);
527 write_unlock(&journal
->j_state_lock
);
532 * Force and wait any uncommitted transactions. We can only force the running
533 * transaction if we don't have an active handle, otherwise, we will deadlock.
534 * Returns: <0 in case of error,
535 * 0 if nothing to commit,
536 * 1 if transaction was successfully committed.
538 static int __jbd2_journal_force_commit(journal_t
*journal
)
540 transaction_t
*transaction
= NULL
;
542 int need_to_start
= 0, ret
= 0;
544 read_lock(&journal
->j_state_lock
);
545 if (journal
->j_running_transaction
&& !current
->journal_info
) {
546 transaction
= journal
->j_running_transaction
;
547 if (!tid_geq(journal
->j_commit_request
, transaction
->t_tid
))
549 } else if (journal
->j_committing_transaction
)
550 transaction
= journal
->j_committing_transaction
;
553 /* Nothing to commit */
554 read_unlock(&journal
->j_state_lock
);
557 tid
= transaction
->t_tid
;
558 read_unlock(&journal
->j_state_lock
);
560 jbd2_log_start_commit(journal
, tid
);
561 ret
= jbd2_log_wait_commit(journal
, tid
);
569 * jbd2_journal_force_commit_nested - Force and wait upon a commit if the
570 * calling process is not within transaction.
572 * @journal: journal to force
573 * Returns true if progress was made.
575 * This is used for forcing out undo-protected data which contains
576 * bitmaps, when the fs is running out of space.
578 int jbd2_journal_force_commit_nested(journal_t
*journal
)
582 ret
= __jbd2_journal_force_commit(journal
);
587 * jbd2_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
)))
724 * Start a fast commit. If there's an ongoing fast or full commit wait for
725 * it to complete. Returns 0 if a new fast commit was started. Returns -EALREADY
726 * if a fast commit is not needed, either because there's an already a commit
727 * going on or this tid has already been committed. Returns -EINVAL if no jbd2
728 * commit has yet been performed.
730 int jbd2_fc_begin_commit(journal_t
*journal
, tid_t tid
)
732 if (unlikely(is_journal_aborted(journal
)))
735 * Fast commits only allowed if at least one full commit has
738 if (!journal
->j_stats
.ts_tid
)
741 write_lock(&journal
->j_state_lock
);
742 if (tid
<= journal
->j_commit_sequence
) {
743 write_unlock(&journal
->j_state_lock
);
747 if (journal
->j_flags
& JBD2_FULL_COMMIT_ONGOING
||
748 (journal
->j_flags
& JBD2_FAST_COMMIT_ONGOING
)) {
751 prepare_to_wait(&journal
->j_fc_wait
, &wait
,
752 TASK_UNINTERRUPTIBLE
);
753 write_unlock(&journal
->j_state_lock
);
755 finish_wait(&journal
->j_fc_wait
, &wait
);
758 journal
->j_flags
|= JBD2_FAST_COMMIT_ONGOING
;
759 write_unlock(&journal
->j_state_lock
);
763 EXPORT_SYMBOL(jbd2_fc_begin_commit
);
766 * Stop a fast commit. If fallback is set, this function starts commit of
767 * TID tid before any other fast commit can start.
769 static int __jbd2_fc_end_commit(journal_t
*journal
, tid_t tid
, bool fallback
)
771 if (journal
->j_fc_cleanup_callback
)
772 journal
->j_fc_cleanup_callback(journal
, 0);
773 write_lock(&journal
->j_state_lock
);
774 journal
->j_flags
&= ~JBD2_FAST_COMMIT_ONGOING
;
776 journal
->j_flags
|= JBD2_FULL_COMMIT_ONGOING
;
777 write_unlock(&journal
->j_state_lock
);
778 wake_up(&journal
->j_fc_wait
);
780 return jbd2_complete_transaction(journal
, tid
);
784 int jbd2_fc_end_commit(journal_t
*journal
)
786 return __jbd2_fc_end_commit(journal
, 0, false);
788 EXPORT_SYMBOL(jbd2_fc_end_commit
);
790 int jbd2_fc_end_commit_fallback(journal_t
*journal
)
794 read_lock(&journal
->j_state_lock
);
795 tid
= journal
->j_running_transaction
?
796 journal
->j_running_transaction
->t_tid
: 0;
797 read_unlock(&journal
->j_state_lock
);
798 return __jbd2_fc_end_commit(journal
, tid
, true);
800 EXPORT_SYMBOL(jbd2_fc_end_commit_fallback
);
802 /* Return 1 when transaction with given tid has already committed. */
803 int jbd2_transaction_committed(journal_t
*journal
, tid_t tid
)
807 read_lock(&journal
->j_state_lock
);
808 if (journal
->j_running_transaction
&&
809 journal
->j_running_transaction
->t_tid
== tid
)
811 if (journal
->j_committing_transaction
&&
812 journal
->j_committing_transaction
->t_tid
== tid
)
814 read_unlock(&journal
->j_state_lock
);
817 EXPORT_SYMBOL(jbd2_transaction_committed
);
820 * When this function returns the transaction corresponding to tid
821 * will be completed. If the transaction has currently running, start
822 * committing that transaction before waiting for it to complete. If
823 * the transaction id is stale, it is by definition already completed,
824 * so just return SUCCESS.
826 int jbd2_complete_transaction(journal_t
*journal
, tid_t tid
)
828 int need_to_wait
= 1;
830 read_lock(&journal
->j_state_lock
);
831 if (journal
->j_running_transaction
&&
832 journal
->j_running_transaction
->t_tid
== tid
) {
833 if (journal
->j_commit_request
!= tid
) {
834 /* transaction not yet started, so request it */
835 read_unlock(&journal
->j_state_lock
);
836 jbd2_log_start_commit(journal
, tid
);
839 } else if (!(journal
->j_committing_transaction
&&
840 journal
->j_committing_transaction
->t_tid
== tid
))
842 read_unlock(&journal
->j_state_lock
);
846 return jbd2_log_wait_commit(journal
, tid
);
848 EXPORT_SYMBOL(jbd2_complete_transaction
);
851 * Log buffer allocation routines:
854 int jbd2_journal_next_log_block(journal_t
*journal
, unsigned long long *retp
)
856 unsigned long blocknr
;
858 write_lock(&journal
->j_state_lock
);
859 J_ASSERT(journal
->j_free
> 1);
861 blocknr
= journal
->j_head
;
864 if (journal
->j_head
== journal
->j_last
)
865 journal
->j_head
= journal
->j_first
;
866 write_unlock(&journal
->j_state_lock
);
867 return jbd2_journal_bmap(journal
, blocknr
, retp
);
870 /* Map one fast commit buffer for use by the file system */
871 int jbd2_fc_get_buf(journal_t
*journal
, struct buffer_head
**bh_out
)
873 unsigned long long pblock
;
874 unsigned long blocknr
;
876 struct buffer_head
*bh
;
881 if (journal
->j_fc_off
+ journal
->j_fc_first
< journal
->j_fc_last
) {
882 fc_off
= journal
->j_fc_off
;
883 blocknr
= journal
->j_fc_first
+ fc_off
;
892 ret
= jbd2_journal_bmap(journal
, blocknr
, &pblock
);
896 bh
= __getblk(journal
->j_dev
, pblock
, journal
->j_blocksize
);
901 journal
->j_fc_wbuf
[fc_off
] = bh
;
907 EXPORT_SYMBOL(jbd2_fc_get_buf
);
910 * Wait on fast commit buffers that were allocated by jbd2_fc_get_buf
913 int jbd2_fc_wait_bufs(journal_t
*journal
, int num_blks
)
915 struct buffer_head
*bh
;
918 j_fc_off
= journal
->j_fc_off
;
921 * Wait in reverse order to minimize chances of us being woken up before
922 * all IOs have completed
924 for (i
= j_fc_off
- 1; i
>= j_fc_off
- num_blks
; i
--) {
925 bh
= journal
->j_fc_wbuf
[i
];
928 journal
->j_fc_wbuf
[i
] = NULL
;
929 if (unlikely(!buffer_uptodate(bh
)))
935 EXPORT_SYMBOL(jbd2_fc_wait_bufs
);
937 int jbd2_fc_release_bufs(journal_t
*journal
)
939 struct buffer_head
*bh
;
942 j_fc_off
= journal
->j_fc_off
;
944 for (i
= j_fc_off
- 1; i
>= 0; i
--) {
945 bh
= journal
->j_fc_wbuf
[i
];
949 journal
->j_fc_wbuf
[i
] = NULL
;
954 EXPORT_SYMBOL(jbd2_fc_release_bufs
);
957 * Conversion of logical to physical block numbers for the journal
959 * On external journals the journal blocks are identity-mapped, so
960 * this is a no-op. If needed, we can use j_blk_offset - everything is
963 int jbd2_journal_bmap(journal_t
*journal
, unsigned long blocknr
,
964 unsigned long long *retp
)
967 unsigned long long ret
;
970 if (journal
->j_inode
) {
972 ret
= bmap(journal
->j_inode
, &block
);
975 printk(KERN_ALERT
"%s: journal block not found "
976 "at offset %lu on %s\n",
977 __func__
, blocknr
, journal
->j_devname
);
979 jbd2_journal_abort(journal
, err
);
985 *retp
= blocknr
; /* +journal->j_blk_offset */
991 * We play buffer_head aliasing tricks to write data/metadata blocks to
992 * the journal without copying their contents, but for journal
993 * descriptor blocks we do need to generate bona fide buffers.
995 * After the caller of jbd2_journal_get_descriptor_buffer() has finished modifying
996 * the buffer's contents they really should run flush_dcache_page(bh->b_page).
997 * But we don't bother doing that, so there will be coherency problems with
998 * mmaps of blockdevs which hold live JBD-controlled filesystems.
1000 struct buffer_head
*
1001 jbd2_journal_get_descriptor_buffer(transaction_t
*transaction
, int type
)
1003 journal_t
*journal
= transaction
->t_journal
;
1004 struct buffer_head
*bh
;
1005 unsigned long long blocknr
;
1006 journal_header_t
*header
;
1009 err
= jbd2_journal_next_log_block(journal
, &blocknr
);
1014 bh
= __getblk(journal
->j_dev
, blocknr
, journal
->j_blocksize
);
1017 atomic_dec(&transaction
->t_outstanding_credits
);
1019 memset(bh
->b_data
, 0, journal
->j_blocksize
);
1020 header
= (journal_header_t
*)bh
->b_data
;
1021 header
->h_magic
= cpu_to_be32(JBD2_MAGIC_NUMBER
);
1022 header
->h_blocktype
= cpu_to_be32(type
);
1023 header
->h_sequence
= cpu_to_be32(transaction
->t_tid
);
1024 set_buffer_uptodate(bh
);
1026 BUFFER_TRACE(bh
, "return this buffer");
1030 void jbd2_descriptor_block_csum_set(journal_t
*j
, struct buffer_head
*bh
)
1032 struct jbd2_journal_block_tail
*tail
;
1035 if (!jbd2_journal_has_csum_v2or3(j
))
1038 tail
= (struct jbd2_journal_block_tail
*)(bh
->b_data
+ j
->j_blocksize
-
1039 sizeof(struct jbd2_journal_block_tail
));
1040 tail
->t_checksum
= 0;
1041 csum
= jbd2_chksum(j
, j
->j_csum_seed
, bh
->b_data
, j
->j_blocksize
);
1042 tail
->t_checksum
= cpu_to_be32(csum
);
1046 * Return tid of the oldest transaction in the journal and block in the journal
1047 * where the transaction starts.
1049 * If the journal is now empty, return which will be the next transaction ID
1050 * we will write and where will that transaction start.
1052 * The return value is 0 if journal tail cannot be pushed any further, 1 if
1055 int jbd2_journal_get_log_tail(journal_t
*journal
, tid_t
*tid
,
1056 unsigned long *block
)
1058 transaction_t
*transaction
;
1061 read_lock(&journal
->j_state_lock
);
1062 spin_lock(&journal
->j_list_lock
);
1063 transaction
= journal
->j_checkpoint_transactions
;
1065 *tid
= transaction
->t_tid
;
1066 *block
= transaction
->t_log_start
;
1067 } else if ((transaction
= journal
->j_committing_transaction
) != NULL
) {
1068 *tid
= transaction
->t_tid
;
1069 *block
= transaction
->t_log_start
;
1070 } else if ((transaction
= journal
->j_running_transaction
) != NULL
) {
1071 *tid
= transaction
->t_tid
;
1072 *block
= journal
->j_head
;
1074 *tid
= journal
->j_transaction_sequence
;
1075 *block
= journal
->j_head
;
1077 ret
= tid_gt(*tid
, journal
->j_tail_sequence
);
1078 spin_unlock(&journal
->j_list_lock
);
1079 read_unlock(&journal
->j_state_lock
);
1085 * Update information in journal structure and in on disk journal superblock
1086 * about log tail. This function does not check whether information passed in
1087 * really pushes log tail further. It's responsibility of the caller to make
1088 * sure provided log tail information is valid (e.g. by holding
1089 * j_checkpoint_mutex all the time between computing log tail and calling this
1090 * function as is the case with jbd2_cleanup_journal_tail()).
1092 * Requires j_checkpoint_mutex
1094 int __jbd2_update_log_tail(journal_t
*journal
, tid_t tid
, unsigned long block
)
1096 unsigned long freed
;
1099 BUG_ON(!mutex_is_locked(&journal
->j_checkpoint_mutex
));
1102 * We cannot afford for write to remain in drive's caches since as
1103 * soon as we update j_tail, next transaction can start reusing journal
1104 * space and if we lose sb update during power failure we'd replay
1105 * old transaction with possibly newly overwritten data.
1107 ret
= jbd2_journal_update_sb_log_tail(journal
, tid
, block
,
1108 REQ_SYNC
| REQ_FUA
);
1112 write_lock(&journal
->j_state_lock
);
1113 freed
= block
- journal
->j_tail
;
1114 if (block
< journal
->j_tail
)
1115 freed
+= journal
->j_last
- journal
->j_first
;
1117 trace_jbd2_update_log_tail(journal
, tid
, block
, freed
);
1119 "Cleaning journal tail from %u to %u (offset %lu), "
1121 journal
->j_tail_sequence
, tid
, block
, freed
);
1123 journal
->j_free
+= freed
;
1124 journal
->j_tail_sequence
= tid
;
1125 journal
->j_tail
= block
;
1126 write_unlock(&journal
->j_state_lock
);
1133 * This is a variation of __jbd2_update_log_tail which checks for validity of
1134 * provided log tail and locks j_checkpoint_mutex. So it is safe against races
1135 * with other threads updating log tail.
1137 void jbd2_update_log_tail(journal_t
*journal
, tid_t tid
, unsigned long block
)
1139 mutex_lock_io(&journal
->j_checkpoint_mutex
);
1140 if (tid_gt(tid
, journal
->j_tail_sequence
))
1141 __jbd2_update_log_tail(journal
, tid
, block
);
1142 mutex_unlock(&journal
->j_checkpoint_mutex
);
1145 struct jbd2_stats_proc_session
{
1147 struct transaction_stats_s
*stats
;
1152 static void *jbd2_seq_info_start(struct seq_file
*seq
, loff_t
*pos
)
1154 return *pos
? NULL
: SEQ_START_TOKEN
;
1157 static void *jbd2_seq_info_next(struct seq_file
*seq
, void *v
, loff_t
*pos
)
1163 static int jbd2_seq_info_show(struct seq_file
*seq
, void *v
)
1165 struct jbd2_stats_proc_session
*s
= seq
->private;
1167 if (v
!= SEQ_START_TOKEN
)
1169 seq_printf(seq
, "%lu transactions (%lu requested), "
1170 "each up to %u blocks\n",
1171 s
->stats
->ts_tid
, s
->stats
->ts_requested
,
1172 s
->journal
->j_max_transaction_buffers
);
1173 if (s
->stats
->ts_tid
== 0)
1175 seq_printf(seq
, "average: \n %ums waiting for transaction\n",
1176 jiffies_to_msecs(s
->stats
->run
.rs_wait
/ s
->stats
->ts_tid
));
1177 seq_printf(seq
, " %ums request delay\n",
1178 (s
->stats
->ts_requested
== 0) ? 0 :
1179 jiffies_to_msecs(s
->stats
->run
.rs_request_delay
/
1180 s
->stats
->ts_requested
));
1181 seq_printf(seq
, " %ums running transaction\n",
1182 jiffies_to_msecs(s
->stats
->run
.rs_running
/ s
->stats
->ts_tid
));
1183 seq_printf(seq
, " %ums transaction was being locked\n",
1184 jiffies_to_msecs(s
->stats
->run
.rs_locked
/ s
->stats
->ts_tid
));
1185 seq_printf(seq
, " %ums flushing data (in ordered mode)\n",
1186 jiffies_to_msecs(s
->stats
->run
.rs_flushing
/ s
->stats
->ts_tid
));
1187 seq_printf(seq
, " %ums logging transaction\n",
1188 jiffies_to_msecs(s
->stats
->run
.rs_logging
/ s
->stats
->ts_tid
));
1189 seq_printf(seq
, " %lluus average transaction commit time\n",
1190 div_u64(s
->journal
->j_average_commit_time
, 1000));
1191 seq_printf(seq
, " %lu handles per transaction\n",
1192 s
->stats
->run
.rs_handle_count
/ s
->stats
->ts_tid
);
1193 seq_printf(seq
, " %lu blocks per transaction\n",
1194 s
->stats
->run
.rs_blocks
/ s
->stats
->ts_tid
);
1195 seq_printf(seq
, " %lu logged blocks per transaction\n",
1196 s
->stats
->run
.rs_blocks_logged
/ s
->stats
->ts_tid
);
1200 static void jbd2_seq_info_stop(struct seq_file
*seq
, void *v
)
1204 static const struct seq_operations jbd2_seq_info_ops
= {
1205 .start
= jbd2_seq_info_start
,
1206 .next
= jbd2_seq_info_next
,
1207 .stop
= jbd2_seq_info_stop
,
1208 .show
= jbd2_seq_info_show
,
1211 static int jbd2_seq_info_open(struct inode
*inode
, struct file
*file
)
1213 journal_t
*journal
= PDE_DATA(inode
);
1214 struct jbd2_stats_proc_session
*s
;
1217 s
= kmalloc(sizeof(*s
), GFP_KERNEL
);
1220 size
= sizeof(struct transaction_stats_s
);
1221 s
->stats
= kmalloc(size
, GFP_KERNEL
);
1222 if (s
->stats
== NULL
) {
1226 spin_lock(&journal
->j_history_lock
);
1227 memcpy(s
->stats
, &journal
->j_stats
, size
);
1228 s
->journal
= journal
;
1229 spin_unlock(&journal
->j_history_lock
);
1231 rc
= seq_open(file
, &jbd2_seq_info_ops
);
1233 struct seq_file
*m
= file
->private_data
;
1243 static int jbd2_seq_info_release(struct inode
*inode
, struct file
*file
)
1245 struct seq_file
*seq
= file
->private_data
;
1246 struct jbd2_stats_proc_session
*s
= seq
->private;
1249 return seq_release(inode
, file
);
1252 static const struct proc_ops jbd2_info_proc_ops
= {
1253 .proc_open
= jbd2_seq_info_open
,
1254 .proc_read
= seq_read
,
1255 .proc_lseek
= seq_lseek
,
1256 .proc_release
= jbd2_seq_info_release
,
1259 static struct proc_dir_entry
*proc_jbd2_stats
;
1261 static void jbd2_stats_proc_init(journal_t
*journal
)
1263 journal
->j_proc_entry
= proc_mkdir(journal
->j_devname
, proc_jbd2_stats
);
1264 if (journal
->j_proc_entry
) {
1265 proc_create_data("info", S_IRUGO
, journal
->j_proc_entry
,
1266 &jbd2_info_proc_ops
, journal
);
1270 static void jbd2_stats_proc_exit(journal_t
*journal
)
1272 remove_proc_entry("info", journal
->j_proc_entry
);
1273 remove_proc_entry(journal
->j_devname
, proc_jbd2_stats
);
1276 /* Minimum size of descriptor tag */
1277 static int jbd2_min_tag_size(void)
1280 * Tag with 32-bit block numbers does not use last four bytes of the
1283 return sizeof(journal_block_tag_t
) - 4;
1287 * jbd2_journal_shrink_scan()
1289 * Scan the checkpointed buffer on the checkpoint list and release the
1292 static unsigned long jbd2_journal_shrink_scan(struct shrinker
*shrink
,
1293 struct shrink_control
*sc
)
1295 journal_t
*journal
= container_of(shrink
, journal_t
, j_shrinker
);
1296 unsigned long nr_to_scan
= sc
->nr_to_scan
;
1297 unsigned long nr_shrunk
;
1298 unsigned long count
;
1300 count
= percpu_counter_read_positive(&journal
->j_checkpoint_jh_count
);
1301 trace_jbd2_shrink_scan_enter(journal
, sc
->nr_to_scan
, count
);
1303 nr_shrunk
= jbd2_journal_shrink_checkpoint_list(journal
, &nr_to_scan
);
1305 count
= percpu_counter_read_positive(&journal
->j_checkpoint_jh_count
);
1306 trace_jbd2_shrink_scan_exit(journal
, nr_to_scan
, nr_shrunk
, count
);
1312 * jbd2_journal_shrink_count()
1314 * Count the number of checkpoint buffers on the checkpoint list.
1316 static unsigned long jbd2_journal_shrink_count(struct shrinker
*shrink
,
1317 struct shrink_control
*sc
)
1319 journal_t
*journal
= container_of(shrink
, journal_t
, j_shrinker
);
1320 unsigned long count
;
1322 count
= percpu_counter_read_positive(&journal
->j_checkpoint_jh_count
);
1323 trace_jbd2_shrink_count(journal
, sc
->nr_to_scan
, count
);
1329 * Management for journal control blocks: functions to create and
1330 * destroy journal_t structures, and to initialise and read existing
1331 * journal blocks from disk. */
1333 /* First: create and setup a journal_t object in memory. We initialise
1334 * very few fields yet: that has to wait until we have created the
1335 * journal structures from from scratch, or loaded them from disk. */
1337 static journal_t
*journal_init_common(struct block_device
*bdev
,
1338 struct block_device
*fs_dev
,
1339 unsigned long long start
, int len
, int blocksize
)
1341 static struct lock_class_key jbd2_trans_commit_key
;
1344 struct buffer_head
*bh
;
1347 journal
= kzalloc(sizeof(*journal
), GFP_KERNEL
);
1351 init_waitqueue_head(&journal
->j_wait_transaction_locked
);
1352 init_waitqueue_head(&journal
->j_wait_done_commit
);
1353 init_waitqueue_head(&journal
->j_wait_commit
);
1354 init_waitqueue_head(&journal
->j_wait_updates
);
1355 init_waitqueue_head(&journal
->j_wait_reserved
);
1356 init_waitqueue_head(&journal
->j_fc_wait
);
1357 mutex_init(&journal
->j_abort_mutex
);
1358 mutex_init(&journal
->j_barrier
);
1359 mutex_init(&journal
->j_checkpoint_mutex
);
1360 spin_lock_init(&journal
->j_revoke_lock
);
1361 spin_lock_init(&journal
->j_list_lock
);
1362 rwlock_init(&journal
->j_state_lock
);
1364 journal
->j_commit_interval
= (HZ
* JBD2_DEFAULT_MAX_COMMIT_AGE
);
1365 journal
->j_min_batch_time
= 0;
1366 journal
->j_max_batch_time
= 15000; /* 15ms */
1367 atomic_set(&journal
->j_reserved_credits
, 0);
1369 /* The journal is marked for error until we succeed with recovery! */
1370 journal
->j_flags
= JBD2_ABORT
;
1372 /* Set up a default-sized revoke table for the new mount. */
1373 err
= jbd2_journal_init_revoke(journal
, JOURNAL_REVOKE_DEFAULT_HASH
);
1377 spin_lock_init(&journal
->j_history_lock
);
1379 lockdep_init_map(&journal
->j_trans_commit_map
, "jbd2_handle",
1380 &jbd2_trans_commit_key
, 0);
1382 /* journal descriptor can store up to n blocks -bzzz */
1383 journal
->j_blocksize
= blocksize
;
1384 journal
->j_dev
= bdev
;
1385 journal
->j_fs_dev
= fs_dev
;
1386 journal
->j_blk_offset
= start
;
1387 journal
->j_total_len
= len
;
1388 /* We need enough buffers to write out full descriptor block. */
1389 n
= journal
->j_blocksize
/ jbd2_min_tag_size();
1390 journal
->j_wbufsize
= n
;
1391 journal
->j_fc_wbuf
= NULL
;
1392 journal
->j_wbuf
= kmalloc_array(n
, sizeof(struct buffer_head
*),
1394 if (!journal
->j_wbuf
)
1397 bh
= getblk_unmovable(journal
->j_dev
, start
, journal
->j_blocksize
);
1399 pr_err("%s: Cannot get buffer for journal superblock\n",
1403 journal
->j_sb_buffer
= bh
;
1404 journal
->j_superblock
= (journal_superblock_t
*)bh
->b_data
;
1406 journal
->j_shrink_transaction
= NULL
;
1407 journal
->j_shrinker
.scan_objects
= jbd2_journal_shrink_scan
;
1408 journal
->j_shrinker
.count_objects
= jbd2_journal_shrink_count
;
1409 journal
->j_shrinker
.seeks
= DEFAULT_SEEKS
;
1410 journal
->j_shrinker
.batch
= journal
->j_max_transaction_buffers
;
1412 if (percpu_counter_init(&journal
->j_checkpoint_jh_count
, 0, GFP_KERNEL
))
1415 if (register_shrinker(&journal
->j_shrinker
)) {
1416 percpu_counter_destroy(&journal
->j_checkpoint_jh_count
);
1422 brelse(journal
->j_sb_buffer
);
1423 kfree(journal
->j_wbuf
);
1424 jbd2_journal_destroy_revoke(journal
);
1429 /* jbd2_journal_init_dev and jbd2_journal_init_inode:
1431 * Create a journal structure assigned some fixed set of disk blocks to
1432 * the journal. We don't actually touch those disk blocks yet, but we
1433 * need to set up all of the mapping information to tell the journaling
1434 * system where the journal blocks are.
1439 * journal_t * jbd2_journal_init_dev() - creates and initialises a journal structure
1440 * @bdev: Block device on which to create the journal
1441 * @fs_dev: Device which hold journalled filesystem for this journal.
1442 * @start: Block nr Start of journal.
1443 * @len: Length of the journal in blocks.
1444 * @blocksize: blocksize of journalling device
1446 * Returns: a newly created journal_t *
1448 * jbd2_journal_init_dev creates a journal which maps a fixed contiguous
1449 * range of blocks on an arbitrary block device.
1452 journal_t
*jbd2_journal_init_dev(struct block_device
*bdev
,
1453 struct block_device
*fs_dev
,
1454 unsigned long long start
, int len
, int blocksize
)
1458 journal
= journal_init_common(bdev
, fs_dev
, start
, len
, blocksize
);
1462 bdevname(journal
->j_dev
, journal
->j_devname
);
1463 strreplace(journal
->j_devname
, '/', '!');
1464 jbd2_stats_proc_init(journal
);
1470 * journal_t * jbd2_journal_init_inode () - creates a journal which maps to a inode.
1471 * @inode: An inode to create the journal in
1473 * jbd2_journal_init_inode creates a journal which maps an on-disk inode as
1474 * the journal. The inode must exist already, must support bmap() and
1475 * must have all data blocks preallocated.
1477 journal_t
*jbd2_journal_init_inode(struct inode
*inode
)
1485 err
= bmap(inode
, &blocknr
);
1487 if (err
|| !blocknr
) {
1488 pr_err("%s: Cannot locate journal superblock\n",
1493 jbd_debug(1, "JBD2: inode %s/%ld, size %lld, bits %d, blksize %ld\n",
1494 inode
->i_sb
->s_id
, inode
->i_ino
, (long long) inode
->i_size
,
1495 inode
->i_sb
->s_blocksize_bits
, inode
->i_sb
->s_blocksize
);
1497 journal
= journal_init_common(inode
->i_sb
->s_bdev
, inode
->i_sb
->s_bdev
,
1498 blocknr
, inode
->i_size
>> inode
->i_sb
->s_blocksize_bits
,
1499 inode
->i_sb
->s_blocksize
);
1503 journal
->j_inode
= inode
;
1504 bdevname(journal
->j_dev
, journal
->j_devname
);
1505 p
= strreplace(journal
->j_devname
, '/', '!');
1506 sprintf(p
, "-%lu", journal
->j_inode
->i_ino
);
1507 jbd2_stats_proc_init(journal
);
1513 * If the journal init or create aborts, we need to mark the journal
1514 * superblock as being NULL to prevent the journal destroy from writing
1515 * back a bogus superblock.
1517 static void journal_fail_superblock(journal_t
*journal
)
1519 struct buffer_head
*bh
= journal
->j_sb_buffer
;
1521 journal
->j_sb_buffer
= NULL
;
1525 * Given a journal_t structure, initialise the various fields for
1526 * startup of a new journaling session. We use this both when creating
1527 * a journal, and after recovering an old journal to reset it for
1531 static int journal_reset(journal_t
*journal
)
1533 journal_superblock_t
*sb
= journal
->j_superblock
;
1534 unsigned long long first
, last
;
1536 first
= be32_to_cpu(sb
->s_first
);
1537 last
= be32_to_cpu(sb
->s_maxlen
);
1538 if (first
+ JBD2_MIN_JOURNAL_BLOCKS
> last
+ 1) {
1539 printk(KERN_ERR
"JBD2: Journal too short (blocks %llu-%llu).\n",
1541 journal_fail_superblock(journal
);
1545 journal
->j_first
= first
;
1546 journal
->j_last
= last
;
1548 journal
->j_head
= journal
->j_first
;
1549 journal
->j_tail
= journal
->j_first
;
1550 journal
->j_free
= journal
->j_last
- journal
->j_first
;
1552 journal
->j_tail_sequence
= journal
->j_transaction_sequence
;
1553 journal
->j_commit_sequence
= journal
->j_transaction_sequence
- 1;
1554 journal
->j_commit_request
= journal
->j_commit_sequence
;
1556 journal
->j_max_transaction_buffers
= jbd2_journal_get_max_txn_bufs(journal
);
1559 * Now that journal recovery is done, turn fast commits off here. This
1560 * way, if fast commit was enabled before the crash but if now FS has
1561 * disabled it, we don't enable fast commits.
1563 jbd2_clear_feature_fast_commit(journal
);
1566 * As a special case, if the on-disk copy is already marked as needing
1567 * no recovery (s_start == 0), then we can safely defer the superblock
1568 * update until the next commit by setting JBD2_FLUSHED. This avoids
1569 * attempting a write to a potential-readonly device.
1571 if (sb
->s_start
== 0) {
1572 jbd_debug(1, "JBD2: Skipping superblock update on recovered sb "
1573 "(start %ld, seq %u, errno %d)\n",
1574 journal
->j_tail
, journal
->j_tail_sequence
,
1576 journal
->j_flags
|= JBD2_FLUSHED
;
1578 /* Lock here to make assertions happy... */
1579 mutex_lock_io(&journal
->j_checkpoint_mutex
);
1581 * Update log tail information. We use REQ_FUA since new
1582 * transaction will start reusing journal space and so we
1583 * must make sure information about current log tail is on
1586 jbd2_journal_update_sb_log_tail(journal
,
1587 journal
->j_tail_sequence
,
1589 REQ_SYNC
| REQ_FUA
);
1590 mutex_unlock(&journal
->j_checkpoint_mutex
);
1592 return jbd2_journal_start_thread(journal
);
1596 * This function expects that the caller will have locked the journal
1597 * buffer head, and will return with it unlocked
1599 static int jbd2_write_superblock(journal_t
*journal
, int write_flags
)
1601 struct buffer_head
*bh
= journal
->j_sb_buffer
;
1602 journal_superblock_t
*sb
= journal
->j_superblock
;
1605 /* Buffer got discarded which means block device got invalidated */
1606 if (!buffer_mapped(bh
)) {
1611 trace_jbd2_write_superblock(journal
, write_flags
);
1612 if (!(journal
->j_flags
& JBD2_BARRIER
))
1613 write_flags
&= ~(REQ_FUA
| REQ_PREFLUSH
);
1614 if (buffer_write_io_error(bh
)) {
1616 * Oh, dear. A previous attempt to write the journal
1617 * superblock failed. This could happen because the
1618 * USB device was yanked out. Or it could happen to
1619 * be a transient write error and maybe the block will
1620 * be remapped. Nothing we can do but to retry the
1621 * write and hope for the best.
1623 printk(KERN_ERR
"JBD2: previous I/O error detected "
1624 "for journal superblock update for %s.\n",
1625 journal
->j_devname
);
1626 clear_buffer_write_io_error(bh
);
1627 set_buffer_uptodate(bh
);
1629 if (jbd2_journal_has_csum_v2or3(journal
))
1630 sb
->s_checksum
= jbd2_superblock_csum(journal
, sb
);
1632 bh
->b_end_io
= end_buffer_write_sync
;
1633 ret
= submit_bh(REQ_OP_WRITE
, write_flags
, bh
);
1635 if (buffer_write_io_error(bh
)) {
1636 clear_buffer_write_io_error(bh
);
1637 set_buffer_uptodate(bh
);
1641 printk(KERN_ERR
"JBD2: Error %d detected when updating "
1642 "journal superblock for %s.\n", ret
,
1643 journal
->j_devname
);
1644 if (!is_journal_aborted(journal
))
1645 jbd2_journal_abort(journal
, ret
);
1652 * jbd2_journal_update_sb_log_tail() - Update log tail in journal sb on disk.
1653 * @journal: The journal to update.
1654 * @tail_tid: TID of the new transaction at the tail of the log
1655 * @tail_block: The first block of the transaction at the tail of the log
1656 * @write_op: With which operation should we write the journal sb
1658 * Update a journal's superblock information about log tail and write it to
1659 * disk, waiting for the IO to complete.
1661 int jbd2_journal_update_sb_log_tail(journal_t
*journal
, tid_t tail_tid
,
1662 unsigned long tail_block
, int write_op
)
1664 journal_superblock_t
*sb
= journal
->j_superblock
;
1667 if (is_journal_aborted(journal
))
1669 if (test_bit(JBD2_CHECKPOINT_IO_ERROR
, &journal
->j_atomic_flags
)) {
1670 jbd2_journal_abort(journal
, -EIO
);
1674 BUG_ON(!mutex_is_locked(&journal
->j_checkpoint_mutex
));
1675 jbd_debug(1, "JBD2: updating superblock (start %lu, seq %u)\n",
1676 tail_block
, tail_tid
);
1678 lock_buffer(journal
->j_sb_buffer
);
1679 sb
->s_sequence
= cpu_to_be32(tail_tid
);
1680 sb
->s_start
= cpu_to_be32(tail_block
);
1682 ret
= jbd2_write_superblock(journal
, write_op
);
1686 /* Log is no longer empty */
1687 write_lock(&journal
->j_state_lock
);
1688 WARN_ON(!sb
->s_sequence
);
1689 journal
->j_flags
&= ~JBD2_FLUSHED
;
1690 write_unlock(&journal
->j_state_lock
);
1697 * jbd2_mark_journal_empty() - Mark on disk journal as empty.
1698 * @journal: The journal to update.
1699 * @write_op: With which operation should we write the journal sb
1701 * Update a journal's dynamic superblock fields to show that journal is empty.
1702 * Write updated superblock to disk waiting for IO to complete.
1704 static void jbd2_mark_journal_empty(journal_t
*journal
, int write_op
)
1706 journal_superblock_t
*sb
= journal
->j_superblock
;
1707 bool had_fast_commit
= false;
1709 BUG_ON(!mutex_is_locked(&journal
->j_checkpoint_mutex
));
1710 lock_buffer(journal
->j_sb_buffer
);
1711 if (sb
->s_start
== 0) { /* Is it already empty? */
1712 unlock_buffer(journal
->j_sb_buffer
);
1716 jbd_debug(1, "JBD2: Marking journal as empty (seq %u)\n",
1717 journal
->j_tail_sequence
);
1719 sb
->s_sequence
= cpu_to_be32(journal
->j_tail_sequence
);
1720 sb
->s_start
= cpu_to_be32(0);
1721 if (jbd2_has_feature_fast_commit(journal
)) {
1723 * When journal is clean, no need to commit fast commit flag and
1724 * make file system incompatible with older kernels.
1726 jbd2_clear_feature_fast_commit(journal
);
1727 had_fast_commit
= true;
1730 jbd2_write_superblock(journal
, write_op
);
1732 if (had_fast_commit
)
1733 jbd2_set_feature_fast_commit(journal
);
1735 /* Log is no longer empty */
1736 write_lock(&journal
->j_state_lock
);
1737 journal
->j_flags
|= JBD2_FLUSHED
;
1738 write_unlock(&journal
->j_state_lock
);
1742 * __jbd2_journal_erase() - Discard or zeroout journal blocks (excluding superblock)
1743 * @journal: The journal to erase.
1744 * @flags: A discard/zeroout request is sent for each physically contigous
1745 * region of the journal. Either JBD2_JOURNAL_FLUSH_DISCARD or
1746 * JBD2_JOURNAL_FLUSH_ZEROOUT must be set to determine which operation
1749 * Note: JBD2_JOURNAL_FLUSH_ZEROOUT attempts to use hardware offload. Zeroes
1750 * will be explicitly written if no hardware offload is available, see
1751 * blkdev_issue_zeroout for more details.
1753 static int __jbd2_journal_erase(journal_t
*journal
, unsigned int flags
)
1756 unsigned long block
, log_offset
; /* logical */
1757 unsigned long long phys_block
, block_start
, block_stop
; /* physical */
1758 loff_t byte_start
, byte_stop
, byte_count
;
1759 struct request_queue
*q
= bdev_get_queue(journal
->j_dev
);
1761 /* flags must be set to either discard or zeroout */
1762 if ((flags
& ~JBD2_JOURNAL_FLUSH_VALID
) || !flags
||
1763 ((flags
& JBD2_JOURNAL_FLUSH_DISCARD
) &&
1764 (flags
& JBD2_JOURNAL_FLUSH_ZEROOUT
)))
1770 if ((flags
& JBD2_JOURNAL_FLUSH_DISCARD
) && !blk_queue_discard(q
))
1774 * lookup block mapping and issue discard/zeroout for each
1777 log_offset
= be32_to_cpu(journal
->j_superblock
->s_first
);
1778 block_start
= ~0ULL;
1779 for (block
= log_offset
; block
< journal
->j_total_len
; block
++) {
1780 err
= jbd2_journal_bmap(journal
, block
, &phys_block
);
1782 pr_err("JBD2: bad block at offset %lu", block
);
1786 if (block_start
== ~0ULL) {
1787 block_start
= phys_block
;
1788 block_stop
= block_start
- 1;
1792 * last block not contiguous with current block,
1793 * process last contiguous region and return to this block on
1796 if (phys_block
!= block_stop
+ 1) {
1801 * if this isn't the last block of journal,
1802 * no need to process now because next block may also
1803 * be part of this contiguous region
1805 if (block
!= journal
->j_total_len
- 1)
1810 * end of contiguous region or this is last block of journal,
1811 * take care of the region
1813 byte_start
= block_start
* journal
->j_blocksize
;
1814 byte_stop
= block_stop
* journal
->j_blocksize
;
1815 byte_count
= (block_stop
- block_start
+ 1) *
1816 journal
->j_blocksize
;
1818 truncate_inode_pages_range(journal
->j_dev
->bd_inode
->i_mapping
,
1819 byte_start
, byte_stop
);
1821 if (flags
& JBD2_JOURNAL_FLUSH_DISCARD
) {
1822 err
= blkdev_issue_discard(journal
->j_dev
,
1823 byte_start
>> SECTOR_SHIFT
,
1824 byte_count
>> SECTOR_SHIFT
,
1826 } else if (flags
& JBD2_JOURNAL_FLUSH_ZEROOUT
) {
1827 err
= blkdev_issue_zeroout(journal
->j_dev
,
1828 byte_start
>> SECTOR_SHIFT
,
1829 byte_count
>> SECTOR_SHIFT
,
1833 if (unlikely(err
!= 0)) {
1834 pr_err("JBD2: (error %d) unable to wipe journal at physical blocks %llu - %llu",
1835 err
, block_start
, block_stop
);
1839 /* reset start and stop after processing a region */
1840 block_start
= ~0ULL;
1843 return blkdev_issue_flush(journal
->j_dev
);
1847 * jbd2_journal_update_sb_errno() - Update error in the journal.
1848 * @journal: The journal to update.
1850 * Update a journal's errno. Write updated superblock to disk waiting for IO
1853 void jbd2_journal_update_sb_errno(journal_t
*journal
)
1855 journal_superblock_t
*sb
= journal
->j_superblock
;
1858 lock_buffer(journal
->j_sb_buffer
);
1859 errcode
= journal
->j_errno
;
1860 if (errcode
== -ESHUTDOWN
)
1862 jbd_debug(1, "JBD2: updating superblock error (errno %d)\n", errcode
);
1863 sb
->s_errno
= cpu_to_be32(errcode
);
1865 jbd2_write_superblock(journal
, REQ_SYNC
| REQ_FUA
);
1867 EXPORT_SYMBOL(jbd2_journal_update_sb_errno
);
1869 static int journal_revoke_records_per_block(journal_t
*journal
)
1872 int space
= journal
->j_blocksize
- sizeof(jbd2_journal_revoke_header_t
);
1874 if (jbd2_has_feature_64bit(journal
))
1879 if (jbd2_journal_has_csum_v2or3(journal
))
1880 space
-= sizeof(struct jbd2_journal_block_tail
);
1881 return space
/ record_size
;
1885 * Read the superblock for a given journal, performing initial
1886 * validation of the format.
1888 static int journal_get_superblock(journal_t
*journal
)
1890 struct buffer_head
*bh
;
1891 journal_superblock_t
*sb
;
1894 bh
= journal
->j_sb_buffer
;
1896 J_ASSERT(bh
!= NULL
);
1897 if (!buffer_uptodate(bh
)) {
1898 ll_rw_block(REQ_OP_READ
, 0, 1, &bh
);
1900 if (!buffer_uptodate(bh
)) {
1902 "JBD2: IO error reading journal superblock\n");
1907 if (buffer_verified(bh
))
1910 sb
= journal
->j_superblock
;
1914 if (sb
->s_header
.h_magic
!= cpu_to_be32(JBD2_MAGIC_NUMBER
) ||
1915 sb
->s_blocksize
!= cpu_to_be32(journal
->j_blocksize
)) {
1916 printk(KERN_WARNING
"JBD2: no valid journal superblock found\n");
1920 switch(be32_to_cpu(sb
->s_header
.h_blocktype
)) {
1921 case JBD2_SUPERBLOCK_V1
:
1922 journal
->j_format_version
= 1;
1924 case JBD2_SUPERBLOCK_V2
:
1925 journal
->j_format_version
= 2;
1928 printk(KERN_WARNING
"JBD2: unrecognised superblock format ID\n");
1932 if (be32_to_cpu(sb
->s_maxlen
) < journal
->j_total_len
)
1933 journal
->j_total_len
= be32_to_cpu(sb
->s_maxlen
);
1934 else if (be32_to_cpu(sb
->s_maxlen
) > journal
->j_total_len
) {
1935 printk(KERN_WARNING
"JBD2: journal file too short\n");
1939 if (be32_to_cpu(sb
->s_first
) == 0 ||
1940 be32_to_cpu(sb
->s_first
) >= journal
->j_total_len
) {
1942 "JBD2: Invalid start block of journal: %u\n",
1943 be32_to_cpu(sb
->s_first
));
1947 if (jbd2_has_feature_csum2(journal
) &&
1948 jbd2_has_feature_csum3(journal
)) {
1949 /* Can't have checksum v2 and v3 at the same time! */
1950 printk(KERN_ERR
"JBD2: Can't enable checksumming v2 and v3 "
1951 "at the same time!\n");
1955 if (jbd2_journal_has_csum_v2or3_feature(journal
) &&
1956 jbd2_has_feature_checksum(journal
)) {
1957 /* Can't have checksum v1 and v2 on at the same time! */
1958 printk(KERN_ERR
"JBD2: Can't enable checksumming v1 and v2/3 "
1959 "at the same time!\n");
1963 if (!jbd2_verify_csum_type(journal
, sb
)) {
1964 printk(KERN_ERR
"JBD2: Unknown checksum type\n");
1968 /* Load the checksum driver */
1969 if (jbd2_journal_has_csum_v2or3_feature(journal
)) {
1970 journal
->j_chksum_driver
= crypto_alloc_shash("crc32c", 0, 0);
1971 if (IS_ERR(journal
->j_chksum_driver
)) {
1972 printk(KERN_ERR
"JBD2: Cannot load crc32c driver.\n");
1973 err
= PTR_ERR(journal
->j_chksum_driver
);
1974 journal
->j_chksum_driver
= NULL
;
1979 if (jbd2_journal_has_csum_v2or3(journal
)) {
1980 /* Check superblock checksum */
1981 if (sb
->s_checksum
!= jbd2_superblock_csum(journal
, sb
)) {
1982 printk(KERN_ERR
"JBD2: journal checksum error\n");
1987 /* Precompute checksum seed for all metadata */
1988 journal
->j_csum_seed
= jbd2_chksum(journal
, ~0, sb
->s_uuid
,
1989 sizeof(sb
->s_uuid
));
1992 journal
->j_revoke_records_per_block
=
1993 journal_revoke_records_per_block(journal
);
1994 set_buffer_verified(bh
);
1999 journal_fail_superblock(journal
);
2004 * Load the on-disk journal superblock and read the key fields into the
2008 static int load_superblock(journal_t
*journal
)
2011 journal_superblock_t
*sb
;
2014 err
= journal_get_superblock(journal
);
2018 sb
= journal
->j_superblock
;
2020 journal
->j_tail_sequence
= be32_to_cpu(sb
->s_sequence
);
2021 journal
->j_tail
= be32_to_cpu(sb
->s_start
);
2022 journal
->j_first
= be32_to_cpu(sb
->s_first
);
2023 journal
->j_errno
= be32_to_cpu(sb
->s_errno
);
2024 journal
->j_last
= be32_to_cpu(sb
->s_maxlen
);
2026 if (jbd2_has_feature_fast_commit(journal
)) {
2027 journal
->j_fc_last
= be32_to_cpu(sb
->s_maxlen
);
2028 num_fc_blocks
= jbd2_journal_get_num_fc_blks(sb
);
2029 if (journal
->j_last
- num_fc_blocks
>= JBD2_MIN_JOURNAL_BLOCKS
)
2030 journal
->j_last
= journal
->j_fc_last
- num_fc_blocks
;
2031 journal
->j_fc_first
= journal
->j_last
+ 1;
2032 journal
->j_fc_off
= 0;
2040 * jbd2_journal_load() - Read journal from disk.
2041 * @journal: Journal to act on.
2043 * Given a journal_t structure which tells us which disk blocks contain
2044 * a journal, read the journal from disk to initialise the in-memory
2047 int jbd2_journal_load(journal_t
*journal
)
2050 journal_superblock_t
*sb
;
2052 err
= load_superblock(journal
);
2056 sb
= journal
->j_superblock
;
2057 /* If this is a V2 superblock, then we have to check the
2058 * features flags on it. */
2060 if (journal
->j_format_version
>= 2) {
2061 if ((sb
->s_feature_ro_compat
&
2062 ~cpu_to_be32(JBD2_KNOWN_ROCOMPAT_FEATURES
)) ||
2063 (sb
->s_feature_incompat
&
2064 ~cpu_to_be32(JBD2_KNOWN_INCOMPAT_FEATURES
))) {
2066 "JBD2: Unrecognised features on journal\n");
2072 * Create a slab for this blocksize
2074 err
= jbd2_journal_create_slab(be32_to_cpu(sb
->s_blocksize
));
2078 /* Let the recovery code check whether it needs to recover any
2079 * data from the journal. */
2080 if (jbd2_journal_recover(journal
))
2081 goto recovery_error
;
2083 if (journal
->j_failed_commit
) {
2084 printk(KERN_ERR
"JBD2: journal transaction %u on %s "
2085 "is corrupt.\n", journal
->j_failed_commit
,
2086 journal
->j_devname
);
2087 return -EFSCORRUPTED
;
2090 * clear JBD2_ABORT flag initialized in journal_init_common
2091 * here to update log tail information with the newest seq.
2093 journal
->j_flags
&= ~JBD2_ABORT
;
2095 /* OK, we've finished with the dynamic journal bits:
2096 * reinitialise the dynamic contents of the superblock in memory
2097 * and reset them on disk. */
2098 if (journal_reset(journal
))
2099 goto recovery_error
;
2101 journal
->j_flags
|= JBD2_LOADED
;
2105 printk(KERN_WARNING
"JBD2: recovery failed\n");
2110 * jbd2_journal_destroy() - Release a journal_t structure.
2111 * @journal: Journal to act on.
2113 * Release a journal_t structure once it is no longer in use by the
2115 * Return <0 if we couldn't clean up the journal.
2117 int jbd2_journal_destroy(journal_t
*journal
)
2121 /* Wait for the commit thread to wake up and die. */
2122 journal_kill_thread(journal
);
2124 /* Force a final log commit */
2125 if (journal
->j_running_transaction
)
2126 jbd2_journal_commit_transaction(journal
);
2128 /* Force any old transactions to disk */
2130 /* Totally anal locking here... */
2131 spin_lock(&journal
->j_list_lock
);
2132 while (journal
->j_checkpoint_transactions
!= NULL
) {
2133 spin_unlock(&journal
->j_list_lock
);
2134 mutex_lock_io(&journal
->j_checkpoint_mutex
);
2135 err
= jbd2_log_do_checkpoint(journal
);
2136 mutex_unlock(&journal
->j_checkpoint_mutex
);
2138 * If checkpointing failed, just free the buffers to avoid
2142 jbd2_journal_destroy_checkpoint(journal
);
2143 spin_lock(&journal
->j_list_lock
);
2146 spin_lock(&journal
->j_list_lock
);
2149 J_ASSERT(journal
->j_running_transaction
== NULL
);
2150 J_ASSERT(journal
->j_committing_transaction
== NULL
);
2151 J_ASSERT(journal
->j_checkpoint_transactions
== NULL
);
2152 spin_unlock(&journal
->j_list_lock
);
2155 * OK, all checkpoint transactions have been checked, now check the
2156 * write out io error flag and abort the journal if some buffer failed
2157 * to write back to the original location, otherwise the filesystem
2158 * may become inconsistent.
2160 if (!is_journal_aborted(journal
) &&
2161 test_bit(JBD2_CHECKPOINT_IO_ERROR
, &journal
->j_atomic_flags
))
2162 jbd2_journal_abort(journal
, -EIO
);
2164 if (journal
->j_sb_buffer
) {
2165 if (!is_journal_aborted(journal
)) {
2166 mutex_lock_io(&journal
->j_checkpoint_mutex
);
2168 write_lock(&journal
->j_state_lock
);
2169 journal
->j_tail_sequence
=
2170 ++journal
->j_transaction_sequence
;
2171 write_unlock(&journal
->j_state_lock
);
2173 jbd2_mark_journal_empty(journal
,
2174 REQ_SYNC
| REQ_PREFLUSH
| REQ_FUA
);
2175 mutex_unlock(&journal
->j_checkpoint_mutex
);
2178 brelse(journal
->j_sb_buffer
);
2181 if (journal
->j_shrinker
.flags
& SHRINKER_REGISTERED
) {
2182 percpu_counter_destroy(&journal
->j_checkpoint_jh_count
);
2183 unregister_shrinker(&journal
->j_shrinker
);
2185 if (journal
->j_proc_entry
)
2186 jbd2_stats_proc_exit(journal
);
2187 iput(journal
->j_inode
);
2188 if (journal
->j_revoke
)
2189 jbd2_journal_destroy_revoke(journal
);
2190 if (journal
->j_chksum_driver
)
2191 crypto_free_shash(journal
->j_chksum_driver
);
2192 kfree(journal
->j_fc_wbuf
);
2193 kfree(journal
->j_wbuf
);
2201 * jbd2_journal_check_used_features() - Check if features specified are used.
2202 * @journal: Journal to check.
2203 * @compat: bitmask of compatible features
2204 * @ro: bitmask of features that force read-only mount
2205 * @incompat: bitmask of incompatible features
2207 * Check whether the journal uses all of a given set of
2208 * features. Return true (non-zero) if it does.
2211 int jbd2_journal_check_used_features(journal_t
*journal
, unsigned long compat
,
2212 unsigned long ro
, unsigned long incompat
)
2214 journal_superblock_t
*sb
;
2216 if (!compat
&& !ro
&& !incompat
)
2218 /* Load journal superblock if it is not loaded yet. */
2219 if (journal
->j_format_version
== 0 &&
2220 journal_get_superblock(journal
) != 0)
2222 if (journal
->j_format_version
== 1)
2225 sb
= journal
->j_superblock
;
2227 if (((be32_to_cpu(sb
->s_feature_compat
) & compat
) == compat
) &&
2228 ((be32_to_cpu(sb
->s_feature_ro_compat
) & ro
) == ro
) &&
2229 ((be32_to_cpu(sb
->s_feature_incompat
) & incompat
) == incompat
))
2236 * jbd2_journal_check_available_features() - Check feature set in journalling layer
2237 * @journal: Journal to check.
2238 * @compat: bitmask of compatible features
2239 * @ro: bitmask of features that force read-only mount
2240 * @incompat: bitmask of incompatible features
2242 * Check whether the journaling code supports the use of
2243 * all of a given set of features on this journal. Return true
2244 * (non-zero) if it can. */
2246 int jbd2_journal_check_available_features(journal_t
*journal
, unsigned long compat
,
2247 unsigned long ro
, unsigned long incompat
)
2249 if (!compat
&& !ro
&& !incompat
)
2252 /* We can support any known requested features iff the
2253 * superblock is in version 2. Otherwise we fail to support any
2254 * extended sb features. */
2256 if (journal
->j_format_version
!= 2)
2259 if ((compat
& JBD2_KNOWN_COMPAT_FEATURES
) == compat
&&
2260 (ro
& JBD2_KNOWN_ROCOMPAT_FEATURES
) == ro
&&
2261 (incompat
& JBD2_KNOWN_INCOMPAT_FEATURES
) == incompat
)
2268 jbd2_journal_initialize_fast_commit(journal_t
*journal
)
2270 journal_superblock_t
*sb
= journal
->j_superblock
;
2271 unsigned long long num_fc_blks
;
2273 num_fc_blks
= jbd2_journal_get_num_fc_blks(sb
);
2274 if (journal
->j_last
- num_fc_blks
< JBD2_MIN_JOURNAL_BLOCKS
)
2277 /* Are we called twice? */
2278 WARN_ON(journal
->j_fc_wbuf
!= NULL
);
2279 journal
->j_fc_wbuf
= kmalloc_array(num_fc_blks
,
2280 sizeof(struct buffer_head
*), GFP_KERNEL
);
2281 if (!journal
->j_fc_wbuf
)
2284 journal
->j_fc_wbufsize
= num_fc_blks
;
2285 journal
->j_fc_last
= journal
->j_last
;
2286 journal
->j_last
= journal
->j_fc_last
- num_fc_blks
;
2287 journal
->j_fc_first
= journal
->j_last
+ 1;
2288 journal
->j_fc_off
= 0;
2289 journal
->j_free
= journal
->j_last
- journal
->j_first
;
2290 journal
->j_max_transaction_buffers
=
2291 jbd2_journal_get_max_txn_bufs(journal
);
2297 * jbd2_journal_set_features() - Mark a given journal feature in the superblock
2298 * @journal: Journal to act on.
2299 * @compat: bitmask of compatible features
2300 * @ro: bitmask of features that force read-only mount
2301 * @incompat: bitmask of incompatible features
2303 * Mark a given journal feature as present on the
2304 * superblock. Returns true if the requested features could be set.
2308 int jbd2_journal_set_features(journal_t
*journal
, unsigned long compat
,
2309 unsigned long ro
, unsigned long incompat
)
2311 #define INCOMPAT_FEATURE_ON(f) \
2312 ((incompat & (f)) && !(sb->s_feature_incompat & cpu_to_be32(f)))
2313 #define COMPAT_FEATURE_ON(f) \
2314 ((compat & (f)) && !(sb->s_feature_compat & cpu_to_be32(f)))
2315 journal_superblock_t
*sb
;
2317 if (jbd2_journal_check_used_features(journal
, compat
, ro
, incompat
))
2320 if (!jbd2_journal_check_available_features(journal
, compat
, ro
, incompat
))
2323 /* If enabling v2 checksums, turn on v3 instead */
2324 if (incompat
& JBD2_FEATURE_INCOMPAT_CSUM_V2
) {
2325 incompat
&= ~JBD2_FEATURE_INCOMPAT_CSUM_V2
;
2326 incompat
|= JBD2_FEATURE_INCOMPAT_CSUM_V3
;
2329 /* Asking for checksumming v3 and v1? Only give them v3. */
2330 if (incompat
& JBD2_FEATURE_INCOMPAT_CSUM_V3
&&
2331 compat
& JBD2_FEATURE_COMPAT_CHECKSUM
)
2332 compat
&= ~JBD2_FEATURE_COMPAT_CHECKSUM
;
2334 jbd_debug(1, "Setting new features 0x%lx/0x%lx/0x%lx\n",
2335 compat
, ro
, incompat
);
2337 sb
= journal
->j_superblock
;
2339 if (incompat
& JBD2_FEATURE_INCOMPAT_FAST_COMMIT
) {
2340 if (jbd2_journal_initialize_fast_commit(journal
)) {
2341 pr_err("JBD2: Cannot enable fast commits.\n");
2346 /* Load the checksum driver if necessary */
2347 if ((journal
->j_chksum_driver
== NULL
) &&
2348 INCOMPAT_FEATURE_ON(JBD2_FEATURE_INCOMPAT_CSUM_V3
)) {
2349 journal
->j_chksum_driver
= crypto_alloc_shash("crc32c", 0, 0);
2350 if (IS_ERR(journal
->j_chksum_driver
)) {
2351 printk(KERN_ERR
"JBD2: Cannot load crc32c driver.\n");
2352 journal
->j_chksum_driver
= NULL
;
2355 /* Precompute checksum seed for all metadata */
2356 journal
->j_csum_seed
= jbd2_chksum(journal
, ~0, sb
->s_uuid
,
2357 sizeof(sb
->s_uuid
));
2360 lock_buffer(journal
->j_sb_buffer
);
2362 /* If enabling v3 checksums, update superblock */
2363 if (INCOMPAT_FEATURE_ON(JBD2_FEATURE_INCOMPAT_CSUM_V3
)) {
2364 sb
->s_checksum_type
= JBD2_CRC32C_CHKSUM
;
2365 sb
->s_feature_compat
&=
2366 ~cpu_to_be32(JBD2_FEATURE_COMPAT_CHECKSUM
);
2369 /* If enabling v1 checksums, downgrade superblock */
2370 if (COMPAT_FEATURE_ON(JBD2_FEATURE_COMPAT_CHECKSUM
))
2371 sb
->s_feature_incompat
&=
2372 ~cpu_to_be32(JBD2_FEATURE_INCOMPAT_CSUM_V2
|
2373 JBD2_FEATURE_INCOMPAT_CSUM_V3
);
2375 sb
->s_feature_compat
|= cpu_to_be32(compat
);
2376 sb
->s_feature_ro_compat
|= cpu_to_be32(ro
);
2377 sb
->s_feature_incompat
|= cpu_to_be32(incompat
);
2378 unlock_buffer(journal
->j_sb_buffer
);
2379 journal
->j_revoke_records_per_block
=
2380 journal_revoke_records_per_block(journal
);
2383 #undef COMPAT_FEATURE_ON
2384 #undef INCOMPAT_FEATURE_ON
2388 * jbd2_journal_clear_features() - Clear a given journal feature in the
2390 * @journal: Journal to act on.
2391 * @compat: bitmask of compatible features
2392 * @ro: bitmask of features that force read-only mount
2393 * @incompat: bitmask of incompatible features
2395 * Clear a given journal feature as present on the
2398 void jbd2_journal_clear_features(journal_t
*journal
, unsigned long compat
,
2399 unsigned long ro
, unsigned long incompat
)
2401 journal_superblock_t
*sb
;
2403 jbd_debug(1, "Clear features 0x%lx/0x%lx/0x%lx\n",
2404 compat
, ro
, incompat
);
2406 sb
= journal
->j_superblock
;
2408 sb
->s_feature_compat
&= ~cpu_to_be32(compat
);
2409 sb
->s_feature_ro_compat
&= ~cpu_to_be32(ro
);
2410 sb
->s_feature_incompat
&= ~cpu_to_be32(incompat
);
2411 journal
->j_revoke_records_per_block
=
2412 journal_revoke_records_per_block(journal
);
2414 EXPORT_SYMBOL(jbd2_journal_clear_features
);
2417 * jbd2_journal_flush() - Flush journal
2418 * @journal: Journal to act on.
2419 * @flags: optional operation on the journal blocks after the flush (see below)
2421 * Flush all data for a given journal to disk and empty the journal.
2422 * Filesystems can use this when remounting readonly to ensure that
2423 * recovery does not need to happen on remount. Optionally, a discard or zeroout
2424 * can be issued on the journal blocks after flushing.
2427 * JBD2_JOURNAL_FLUSH_DISCARD: issues discards for the journal blocks
2428 * JBD2_JOURNAL_FLUSH_ZEROOUT: issues zeroouts for the journal blocks
2430 int jbd2_journal_flush(journal_t
*journal
, unsigned int flags
)
2433 transaction_t
*transaction
= NULL
;
2435 write_lock(&journal
->j_state_lock
);
2437 /* Force everything buffered to the log... */
2438 if (journal
->j_running_transaction
) {
2439 transaction
= journal
->j_running_transaction
;
2440 __jbd2_log_start_commit(journal
, transaction
->t_tid
);
2441 } else if (journal
->j_committing_transaction
)
2442 transaction
= journal
->j_committing_transaction
;
2444 /* Wait for the log commit to complete... */
2446 tid_t tid
= transaction
->t_tid
;
2448 write_unlock(&journal
->j_state_lock
);
2449 jbd2_log_wait_commit(journal
, tid
);
2451 write_unlock(&journal
->j_state_lock
);
2454 /* ...and flush everything in the log out to disk. */
2455 spin_lock(&journal
->j_list_lock
);
2456 while (!err
&& journal
->j_checkpoint_transactions
!= NULL
) {
2457 spin_unlock(&journal
->j_list_lock
);
2458 mutex_lock_io(&journal
->j_checkpoint_mutex
);
2459 err
= jbd2_log_do_checkpoint(journal
);
2460 mutex_unlock(&journal
->j_checkpoint_mutex
);
2461 spin_lock(&journal
->j_list_lock
);
2463 spin_unlock(&journal
->j_list_lock
);
2465 if (is_journal_aborted(journal
))
2468 mutex_lock_io(&journal
->j_checkpoint_mutex
);
2470 err
= jbd2_cleanup_journal_tail(journal
);
2472 mutex_unlock(&journal
->j_checkpoint_mutex
);
2478 /* Finally, mark the journal as really needing no recovery.
2479 * This sets s_start==0 in the underlying superblock, which is
2480 * the magic code for a fully-recovered superblock. Any future
2481 * commits of data to the journal will restore the current
2483 jbd2_mark_journal_empty(journal
, REQ_SYNC
| REQ_FUA
);
2486 err
= __jbd2_journal_erase(journal
, flags
);
2488 mutex_unlock(&journal
->j_checkpoint_mutex
);
2489 write_lock(&journal
->j_state_lock
);
2490 J_ASSERT(!journal
->j_running_transaction
);
2491 J_ASSERT(!journal
->j_committing_transaction
);
2492 J_ASSERT(!journal
->j_checkpoint_transactions
);
2493 J_ASSERT(journal
->j_head
== journal
->j_tail
);
2494 J_ASSERT(journal
->j_tail_sequence
== journal
->j_transaction_sequence
);
2495 write_unlock(&journal
->j_state_lock
);
2501 * jbd2_journal_wipe() - Wipe journal contents
2502 * @journal: Journal to act on.
2503 * @write: flag (see below)
2505 * Wipe out all of the contents of a journal, safely. This will produce
2506 * a warning if the journal contains any valid recovery information.
2507 * Must be called between journal_init_*() and jbd2_journal_load().
2509 * If 'write' is non-zero, then we wipe out the journal on disk; otherwise
2510 * we merely suppress recovery.
2513 int jbd2_journal_wipe(journal_t
*journal
, int write
)
2517 J_ASSERT (!(journal
->j_flags
& JBD2_LOADED
));
2519 err
= load_superblock(journal
);
2523 if (!journal
->j_tail
)
2526 printk(KERN_WARNING
"JBD2: %s recovery information on journal\n",
2527 write
? "Clearing" : "Ignoring");
2529 err
= jbd2_journal_skip_recovery(journal
);
2531 /* Lock to make assertions happy... */
2532 mutex_lock_io(&journal
->j_checkpoint_mutex
);
2533 jbd2_mark_journal_empty(journal
, REQ_SYNC
| REQ_FUA
);
2534 mutex_unlock(&journal
->j_checkpoint_mutex
);
2542 * jbd2_journal_abort () - Shutdown the journal immediately.
2543 * @journal: the journal to shutdown.
2544 * @errno: an error number to record in the journal indicating
2545 * the reason for the shutdown.
2547 * Perform a complete, immediate shutdown of the ENTIRE
2548 * journal (not of a single transaction). This operation cannot be
2549 * undone without closing and reopening the journal.
2551 * The jbd2_journal_abort function is intended to support higher level error
2552 * recovery mechanisms such as the ext2/ext3 remount-readonly error
2555 * Journal abort has very specific semantics. Any existing dirty,
2556 * unjournaled buffers in the main filesystem will still be written to
2557 * disk by bdflush, but the journaling mechanism will be suspended
2558 * immediately and no further transaction commits will be honoured.
2560 * Any dirty, journaled buffers will be written back to disk without
2561 * hitting the journal. Atomicity cannot be guaranteed on an aborted
2562 * filesystem, but we _do_ attempt to leave as much data as possible
2563 * behind for fsck to use for cleanup.
2565 * Any attempt to get a new transaction handle on a journal which is in
2566 * ABORT state will just result in an -EROFS error return. A
2567 * jbd2_journal_stop on an existing handle will return -EIO if we have
2568 * entered abort state during the update.
2570 * Recursive transactions are not disturbed by journal abort until the
2571 * final jbd2_journal_stop, which will receive the -EIO error.
2573 * Finally, the jbd2_journal_abort call allows the caller to supply an errno
2574 * which will be recorded (if possible) in the journal superblock. This
2575 * allows a client to record failure conditions in the middle of a
2576 * transaction without having to complete the transaction to record the
2577 * failure to disk. ext3_error, for example, now uses this
2582 void jbd2_journal_abort(journal_t
*journal
, int errno
)
2584 transaction_t
*transaction
;
2587 * Lock the aborting procedure until everything is done, this avoid
2588 * races between filesystem's error handling flow (e.g. ext4_abort()),
2589 * ensure panic after the error info is written into journal's
2592 mutex_lock(&journal
->j_abort_mutex
);
2594 * ESHUTDOWN always takes precedence because a file system check
2595 * caused by any other journal abort error is not required after
2596 * a shutdown triggered.
2598 write_lock(&journal
->j_state_lock
);
2599 if (journal
->j_flags
& JBD2_ABORT
) {
2600 int old_errno
= journal
->j_errno
;
2602 write_unlock(&journal
->j_state_lock
);
2603 if (old_errno
!= -ESHUTDOWN
&& errno
== -ESHUTDOWN
) {
2604 journal
->j_errno
= errno
;
2605 jbd2_journal_update_sb_errno(journal
);
2607 mutex_unlock(&journal
->j_abort_mutex
);
2612 * Mark the abort as occurred and start current running transaction
2613 * to release all journaled buffer.
2615 pr_err("Aborting journal on device %s.\n", journal
->j_devname
);
2617 journal
->j_flags
|= JBD2_ABORT
;
2618 journal
->j_errno
= errno
;
2619 transaction
= journal
->j_running_transaction
;
2621 __jbd2_log_start_commit(journal
, transaction
->t_tid
);
2622 write_unlock(&journal
->j_state_lock
);
2625 * Record errno to the journal super block, so that fsck and jbd2
2626 * layer could realise that a filesystem check is needed.
2628 jbd2_journal_update_sb_errno(journal
);
2629 mutex_unlock(&journal
->j_abort_mutex
);
2633 * jbd2_journal_errno() - returns the journal's error state.
2634 * @journal: journal to examine.
2636 * This is the errno number set with jbd2_journal_abort(), the last
2637 * time the journal was mounted - if the journal was stopped
2638 * without calling abort this will be 0.
2640 * If the journal has been aborted on this mount time -EROFS will
2643 int jbd2_journal_errno(journal_t
*journal
)
2647 read_lock(&journal
->j_state_lock
);
2648 if (journal
->j_flags
& JBD2_ABORT
)
2651 err
= journal
->j_errno
;
2652 read_unlock(&journal
->j_state_lock
);
2657 * jbd2_journal_clear_err() - clears the journal's error state
2658 * @journal: journal to act on.
2660 * An error must be cleared or acked to take a FS out of readonly
2663 int jbd2_journal_clear_err(journal_t
*journal
)
2667 write_lock(&journal
->j_state_lock
);
2668 if (journal
->j_flags
& JBD2_ABORT
)
2671 journal
->j_errno
= 0;
2672 write_unlock(&journal
->j_state_lock
);
2677 * jbd2_journal_ack_err() - Ack journal err.
2678 * @journal: journal to act on.
2680 * An error must be cleared or acked to take a FS out of readonly
2683 void jbd2_journal_ack_err(journal_t
*journal
)
2685 write_lock(&journal
->j_state_lock
);
2686 if (journal
->j_errno
)
2687 journal
->j_flags
|= JBD2_ACK_ERR
;
2688 write_unlock(&journal
->j_state_lock
);
2691 int jbd2_journal_blocks_per_page(struct inode
*inode
)
2693 return 1 << (PAGE_SHIFT
- inode
->i_sb
->s_blocksize_bits
);
2697 * helper functions to deal with 32 or 64bit block numbers.
2699 size_t journal_tag_bytes(journal_t
*journal
)
2703 if (jbd2_has_feature_csum3(journal
))
2704 return sizeof(journal_block_tag3_t
);
2706 sz
= sizeof(journal_block_tag_t
);
2708 if (jbd2_has_feature_csum2(journal
))
2709 sz
+= sizeof(__u16
);
2711 if (jbd2_has_feature_64bit(journal
))
2714 return sz
- sizeof(__u32
);
2718 * JBD memory management
2720 * These functions are used to allocate block-sized chunks of memory
2721 * used for making copies of buffer_head data. Very often it will be
2722 * page-sized chunks of data, but sometimes it will be in
2723 * sub-page-size chunks. (For example, 16k pages on Power systems
2724 * with a 4k block file system.) For blocks smaller than a page, we
2725 * use a SLAB allocator. There are slab caches for each block size,
2726 * which are allocated at mount time, if necessary, and we only free
2727 * (all of) the slab caches when/if the jbd2 module is unloaded. For
2728 * this reason we don't need to a mutex to protect access to
2729 * jbd2_slab[] allocating or releasing memory; only in
2730 * jbd2_journal_create_slab().
2732 #define JBD2_MAX_SLABS 8
2733 static struct kmem_cache
*jbd2_slab
[JBD2_MAX_SLABS
];
2735 static const char *jbd2_slab_names
[JBD2_MAX_SLABS
] = {
2736 "jbd2_1k", "jbd2_2k", "jbd2_4k", "jbd2_8k",
2737 "jbd2_16k", "jbd2_32k", "jbd2_64k", "jbd2_128k"
2741 static void jbd2_journal_destroy_slabs(void)
2745 for (i
= 0; i
< JBD2_MAX_SLABS
; i
++) {
2746 kmem_cache_destroy(jbd2_slab
[i
]);
2747 jbd2_slab
[i
] = NULL
;
2751 static int jbd2_journal_create_slab(size_t size
)
2753 static DEFINE_MUTEX(jbd2_slab_create_mutex
);
2754 int i
= order_base_2(size
) - 10;
2757 if (size
== PAGE_SIZE
)
2760 if (i
>= JBD2_MAX_SLABS
)
2763 if (unlikely(i
< 0))
2765 mutex_lock(&jbd2_slab_create_mutex
);
2767 mutex_unlock(&jbd2_slab_create_mutex
);
2768 return 0; /* Already created */
2771 slab_size
= 1 << (i
+10);
2772 jbd2_slab
[i
] = kmem_cache_create(jbd2_slab_names
[i
], slab_size
,
2773 slab_size
, 0, NULL
);
2774 mutex_unlock(&jbd2_slab_create_mutex
);
2775 if (!jbd2_slab
[i
]) {
2776 printk(KERN_EMERG
"JBD2: no memory for jbd2_slab cache\n");
2782 static struct kmem_cache
*get_slab(size_t size
)
2784 int i
= order_base_2(size
) - 10;
2786 BUG_ON(i
>= JBD2_MAX_SLABS
);
2787 if (unlikely(i
< 0))
2789 BUG_ON(jbd2_slab
[i
] == NULL
);
2790 return jbd2_slab
[i
];
2793 void *jbd2_alloc(size_t size
, gfp_t flags
)
2797 BUG_ON(size
& (size
-1)); /* Must be a power of 2 */
2799 if (size
< PAGE_SIZE
)
2800 ptr
= kmem_cache_alloc(get_slab(size
), flags
);
2802 ptr
= (void *)__get_free_pages(flags
, get_order(size
));
2804 /* Check alignment; SLUB has gotten this wrong in the past,
2805 * and this can lead to user data corruption! */
2806 BUG_ON(((unsigned long) ptr
) & (size
-1));
2811 void jbd2_free(void *ptr
, size_t size
)
2813 if (size
< PAGE_SIZE
)
2814 kmem_cache_free(get_slab(size
), ptr
);
2816 free_pages((unsigned long)ptr
, get_order(size
));
2820 * Journal_head storage management
2822 static struct kmem_cache
*jbd2_journal_head_cache
;
2823 #ifdef CONFIG_JBD2_DEBUG
2824 static atomic_t nr_journal_heads
= ATOMIC_INIT(0);
2827 static int __init
jbd2_journal_init_journal_head_cache(void)
2829 J_ASSERT(!jbd2_journal_head_cache
);
2830 jbd2_journal_head_cache
= kmem_cache_create("jbd2_journal_head",
2831 sizeof(struct journal_head
),
2833 SLAB_TEMPORARY
| SLAB_TYPESAFE_BY_RCU
,
2835 if (!jbd2_journal_head_cache
) {
2836 printk(KERN_EMERG
"JBD2: no memory for journal_head cache\n");
2842 static void jbd2_journal_destroy_journal_head_cache(void)
2844 kmem_cache_destroy(jbd2_journal_head_cache
);
2845 jbd2_journal_head_cache
= NULL
;
2849 * journal_head splicing and dicing
2851 static struct journal_head
*journal_alloc_journal_head(void)
2853 struct journal_head
*ret
;
2855 #ifdef CONFIG_JBD2_DEBUG
2856 atomic_inc(&nr_journal_heads
);
2858 ret
= kmem_cache_zalloc(jbd2_journal_head_cache
, GFP_NOFS
);
2860 jbd_debug(1, "out of memory for journal_head\n");
2861 pr_notice_ratelimited("ENOMEM in %s, retrying.\n", __func__
);
2862 ret
= kmem_cache_zalloc(jbd2_journal_head_cache
,
2863 GFP_NOFS
| __GFP_NOFAIL
);
2866 spin_lock_init(&ret
->b_state_lock
);
2870 static void journal_free_journal_head(struct journal_head
*jh
)
2872 #ifdef CONFIG_JBD2_DEBUG
2873 atomic_dec(&nr_journal_heads
);
2874 memset(jh
, JBD2_POISON_FREE
, sizeof(*jh
));
2876 kmem_cache_free(jbd2_journal_head_cache
, jh
);
2880 * A journal_head is attached to a buffer_head whenever JBD has an
2881 * interest in the buffer.
2883 * Whenever a buffer has an attached journal_head, its ->b_state:BH_JBD bit
2884 * is set. This bit is tested in core kernel code where we need to take
2885 * JBD-specific actions. Testing the zeroness of ->b_private is not reliable
2888 * When a buffer has its BH_JBD bit set, its ->b_count is elevated by one.
2890 * When a buffer has its BH_JBD bit set it is immune from being released by
2891 * core kernel code, mainly via ->b_count.
2893 * A journal_head is detached from its buffer_head when the journal_head's
2894 * b_jcount reaches zero. Running transaction (b_transaction) and checkpoint
2895 * transaction (b_cp_transaction) hold their references to b_jcount.
2897 * Various places in the kernel want to attach a journal_head to a buffer_head
2898 * _before_ attaching the journal_head to a transaction. To protect the
2899 * journal_head in this situation, jbd2_journal_add_journal_head elevates the
2900 * journal_head's b_jcount refcount by one. The caller must call
2901 * jbd2_journal_put_journal_head() to undo this.
2903 * So the typical usage would be:
2905 * (Attach a journal_head if needed. Increments b_jcount)
2906 * struct journal_head *jh = jbd2_journal_add_journal_head(bh);
2908 * (Get another reference for transaction)
2909 * jbd2_journal_grab_journal_head(bh);
2910 * jh->b_transaction = xxx;
2911 * (Put original reference)
2912 * jbd2_journal_put_journal_head(jh);
2916 * Give a buffer_head a journal_head.
2920 struct journal_head
*jbd2_journal_add_journal_head(struct buffer_head
*bh
)
2922 struct journal_head
*jh
;
2923 struct journal_head
*new_jh
= NULL
;
2926 if (!buffer_jbd(bh
))
2927 new_jh
= journal_alloc_journal_head();
2929 jbd_lock_bh_journal_head(bh
);
2930 if (buffer_jbd(bh
)) {
2934 (atomic_read(&bh
->b_count
) > 0) ||
2935 (bh
->b_page
&& bh
->b_page
->mapping
));
2938 jbd_unlock_bh_journal_head(bh
);
2943 new_jh
= NULL
; /* We consumed it */
2948 BUFFER_TRACE(bh
, "added journal_head");
2951 jbd_unlock_bh_journal_head(bh
);
2953 journal_free_journal_head(new_jh
);
2954 return bh
->b_private
;
2958 * Grab a ref against this buffer_head's journal_head. If it ended up not
2959 * having a journal_head, return NULL
2961 struct journal_head
*jbd2_journal_grab_journal_head(struct buffer_head
*bh
)
2963 struct journal_head
*jh
= NULL
;
2965 jbd_lock_bh_journal_head(bh
);
2966 if (buffer_jbd(bh
)) {
2970 jbd_unlock_bh_journal_head(bh
);
2974 static void __journal_remove_journal_head(struct buffer_head
*bh
)
2976 struct journal_head
*jh
= bh2jh(bh
);
2978 J_ASSERT_JH(jh
, jh
->b_transaction
== NULL
);
2979 J_ASSERT_JH(jh
, jh
->b_next_transaction
== NULL
);
2980 J_ASSERT_JH(jh
, jh
->b_cp_transaction
== NULL
);
2981 J_ASSERT_JH(jh
, jh
->b_jlist
== BJ_None
);
2982 J_ASSERT_BH(bh
, buffer_jbd(bh
));
2983 J_ASSERT_BH(bh
, jh2bh(jh
) == bh
);
2984 BUFFER_TRACE(bh
, "remove journal_head");
2986 /* Unlink before dropping the lock */
2987 bh
->b_private
= NULL
;
2988 jh
->b_bh
= NULL
; /* debug, really */
2989 clear_buffer_jbd(bh
);
2992 static void journal_release_journal_head(struct journal_head
*jh
, size_t b_size
)
2994 if (jh
->b_frozen_data
) {
2995 printk(KERN_WARNING
"%s: freeing b_frozen_data\n", __func__
);
2996 jbd2_free(jh
->b_frozen_data
, b_size
);
2998 if (jh
->b_committed_data
) {
2999 printk(KERN_WARNING
"%s: freeing b_committed_data\n", __func__
);
3000 jbd2_free(jh
->b_committed_data
, b_size
);
3002 journal_free_journal_head(jh
);
3006 * Drop a reference on the passed journal_head. If it fell to zero then
3007 * release the journal_head from the buffer_head.
3009 void jbd2_journal_put_journal_head(struct journal_head
*jh
)
3011 struct buffer_head
*bh
= jh2bh(jh
);
3013 jbd_lock_bh_journal_head(bh
);
3014 J_ASSERT_JH(jh
, jh
->b_jcount
> 0);
3016 if (!jh
->b_jcount
) {
3017 __journal_remove_journal_head(bh
);
3018 jbd_unlock_bh_journal_head(bh
);
3019 journal_release_journal_head(jh
, bh
->b_size
);
3022 jbd_unlock_bh_journal_head(bh
);
3027 * Initialize jbd inode head
3029 void jbd2_journal_init_jbd_inode(struct jbd2_inode
*jinode
, struct inode
*inode
)
3031 jinode
->i_transaction
= NULL
;
3032 jinode
->i_next_transaction
= NULL
;
3033 jinode
->i_vfs_inode
= inode
;
3034 jinode
->i_flags
= 0;
3035 jinode
->i_dirty_start
= 0;
3036 jinode
->i_dirty_end
= 0;
3037 INIT_LIST_HEAD(&jinode
->i_list
);
3041 * Function to be called before we start removing inode from memory (i.e.,
3042 * clear_inode() is a fine place to be called from). It removes inode from
3043 * transaction's lists.
3045 void jbd2_journal_release_jbd_inode(journal_t
*journal
,
3046 struct jbd2_inode
*jinode
)
3051 spin_lock(&journal
->j_list_lock
);
3052 /* Is commit writing out inode - we have to wait */
3053 if (jinode
->i_flags
& JI_COMMIT_RUNNING
) {
3054 wait_queue_head_t
*wq
;
3055 DEFINE_WAIT_BIT(wait
, &jinode
->i_flags
, __JI_COMMIT_RUNNING
);
3056 wq
= bit_waitqueue(&jinode
->i_flags
, __JI_COMMIT_RUNNING
);
3057 prepare_to_wait(wq
, &wait
.wq_entry
, TASK_UNINTERRUPTIBLE
);
3058 spin_unlock(&journal
->j_list_lock
);
3060 finish_wait(wq
, &wait
.wq_entry
);
3064 if (jinode
->i_transaction
) {
3065 list_del(&jinode
->i_list
);
3066 jinode
->i_transaction
= NULL
;
3068 spin_unlock(&journal
->j_list_lock
);
3072 #ifdef CONFIG_PROC_FS
3074 #define JBD2_STATS_PROC_NAME "fs/jbd2"
3076 static void __init
jbd2_create_jbd_stats_proc_entry(void)
3078 proc_jbd2_stats
= proc_mkdir(JBD2_STATS_PROC_NAME
, NULL
);
3081 static void __exit
jbd2_remove_jbd_stats_proc_entry(void)
3083 if (proc_jbd2_stats
)
3084 remove_proc_entry(JBD2_STATS_PROC_NAME
, NULL
);
3089 #define jbd2_create_jbd_stats_proc_entry() do {} while (0)
3090 #define jbd2_remove_jbd_stats_proc_entry() do {} while (0)
3094 struct kmem_cache
*jbd2_handle_cache
, *jbd2_inode_cache
;
3096 static int __init
jbd2_journal_init_inode_cache(void)
3098 J_ASSERT(!jbd2_inode_cache
);
3099 jbd2_inode_cache
= KMEM_CACHE(jbd2_inode
, 0);
3100 if (!jbd2_inode_cache
) {
3101 pr_emerg("JBD2: failed to create inode cache\n");
3107 static int __init
jbd2_journal_init_handle_cache(void)
3109 J_ASSERT(!jbd2_handle_cache
);
3110 jbd2_handle_cache
= KMEM_CACHE(jbd2_journal_handle
, SLAB_TEMPORARY
);
3111 if (!jbd2_handle_cache
) {
3112 printk(KERN_EMERG
"JBD2: failed to create handle cache\n");
3118 static void jbd2_journal_destroy_inode_cache(void)
3120 kmem_cache_destroy(jbd2_inode_cache
);
3121 jbd2_inode_cache
= NULL
;
3124 static void jbd2_journal_destroy_handle_cache(void)
3126 kmem_cache_destroy(jbd2_handle_cache
);
3127 jbd2_handle_cache
= NULL
;
3131 * Module startup and shutdown
3134 static int __init
journal_init_caches(void)
3138 ret
= jbd2_journal_init_revoke_record_cache();
3140 ret
= jbd2_journal_init_revoke_table_cache();
3142 ret
= jbd2_journal_init_journal_head_cache();
3144 ret
= jbd2_journal_init_handle_cache();
3146 ret
= jbd2_journal_init_inode_cache();
3148 ret
= jbd2_journal_init_transaction_cache();
3152 static void jbd2_journal_destroy_caches(void)
3154 jbd2_journal_destroy_revoke_record_cache();
3155 jbd2_journal_destroy_revoke_table_cache();
3156 jbd2_journal_destroy_journal_head_cache();
3157 jbd2_journal_destroy_handle_cache();
3158 jbd2_journal_destroy_inode_cache();
3159 jbd2_journal_destroy_transaction_cache();
3160 jbd2_journal_destroy_slabs();
3163 static int __init
journal_init(void)
3167 BUILD_BUG_ON(sizeof(struct journal_superblock_s
) != 1024);
3169 ret
= journal_init_caches();
3171 jbd2_create_jbd_stats_proc_entry();
3173 jbd2_journal_destroy_caches();
3178 static void __exit
journal_exit(void)
3180 #ifdef CONFIG_JBD2_DEBUG
3181 int n
= atomic_read(&nr_journal_heads
);
3183 printk(KERN_ERR
"JBD2: leaked %d journal_heads!\n", n
);
3185 jbd2_remove_jbd_stats_proc_entry();
3186 jbd2_journal_destroy_caches();
3189 MODULE_LICENSE("GPL");
3190 module_init(journal_init
);
3191 module_exit(journal_exit
);