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/debugfs.h>
39 #include <linux/seq_file.h>
40 #include <linux/math64.h>
41 #include <linux/hash.h>
42 #include <linux/log2.h>
43 #include <linux/vmalloc.h>
44 #include <linux/backing-dev.h>
45 #include <linux/bitops.h>
46 #include <linux/ratelimit.h>
48 #define CREATE_TRACE_POINTS
49 #include <trace/events/jbd2.h>
51 #include <asm/uaccess.h>
54 EXPORT_SYMBOL(jbd2_journal_extend
);
55 EXPORT_SYMBOL(jbd2_journal_stop
);
56 EXPORT_SYMBOL(jbd2_journal_lock_updates
);
57 EXPORT_SYMBOL(jbd2_journal_unlock_updates
);
58 EXPORT_SYMBOL(jbd2_journal_get_write_access
);
59 EXPORT_SYMBOL(jbd2_journal_get_create_access
);
60 EXPORT_SYMBOL(jbd2_journal_get_undo_access
);
61 EXPORT_SYMBOL(jbd2_journal_set_triggers
);
62 EXPORT_SYMBOL(jbd2_journal_dirty_metadata
);
63 EXPORT_SYMBOL(jbd2_journal_release_buffer
);
64 EXPORT_SYMBOL(jbd2_journal_forget
);
66 EXPORT_SYMBOL(journal_sync_buffer
);
68 EXPORT_SYMBOL(jbd2_journal_flush
);
69 EXPORT_SYMBOL(jbd2_journal_revoke
);
71 EXPORT_SYMBOL(jbd2_journal_init_dev
);
72 EXPORT_SYMBOL(jbd2_journal_init_inode
);
73 EXPORT_SYMBOL(jbd2_journal_check_used_features
);
74 EXPORT_SYMBOL(jbd2_journal_check_available_features
);
75 EXPORT_SYMBOL(jbd2_journal_set_features
);
76 EXPORT_SYMBOL(jbd2_journal_load
);
77 EXPORT_SYMBOL(jbd2_journal_destroy
);
78 EXPORT_SYMBOL(jbd2_journal_abort
);
79 EXPORT_SYMBOL(jbd2_journal_errno
);
80 EXPORT_SYMBOL(jbd2_journal_ack_err
);
81 EXPORT_SYMBOL(jbd2_journal_clear_err
);
82 EXPORT_SYMBOL(jbd2_log_wait_commit
);
83 EXPORT_SYMBOL(jbd2_log_start_commit
);
84 EXPORT_SYMBOL(jbd2_journal_start_commit
);
85 EXPORT_SYMBOL(jbd2_journal_force_commit_nested
);
86 EXPORT_SYMBOL(jbd2_journal_wipe
);
87 EXPORT_SYMBOL(jbd2_journal_blocks_per_page
);
88 EXPORT_SYMBOL(jbd2_journal_invalidatepage
);
89 EXPORT_SYMBOL(jbd2_journal_try_to_free_buffers
);
90 EXPORT_SYMBOL(jbd2_journal_force_commit
);
91 EXPORT_SYMBOL(jbd2_journal_file_inode
);
92 EXPORT_SYMBOL(jbd2_journal_init_jbd_inode
);
93 EXPORT_SYMBOL(jbd2_journal_release_jbd_inode
);
94 EXPORT_SYMBOL(jbd2_journal_begin_ordered_truncate
);
95 EXPORT_SYMBOL(jbd2_inode_cache
);
97 static void __journal_abort_soft (journal_t
*journal
, int errno
);
98 static int jbd2_journal_create_slab(size_t slab_size
);
101 * Helper function used to manage commit timeouts
104 static void commit_timeout(unsigned long __data
)
106 struct task_struct
* p
= (struct task_struct
*) __data
;
112 * kjournald2: The main thread function used to manage a logging device
115 * This kernel thread is responsible for two things:
117 * 1) COMMIT: Every so often we need to commit the current state of the
118 * filesystem to disk. The journal thread is responsible for writing
119 * all of the metadata buffers to disk.
121 * 2) CHECKPOINT: We cannot reuse a used section of the log file until all
122 * of the data in that part of the log has been rewritten elsewhere on
123 * the disk. Flushing these old buffers to reclaim space in the log is
124 * known as checkpointing, and this thread is responsible for that job.
127 static int kjournald2(void *arg
)
129 journal_t
*journal
= arg
;
130 transaction_t
*transaction
;
133 * Set up an interval timer which can be used to trigger a commit wakeup
134 * after the commit interval expires
136 setup_timer(&journal
->j_commit_timer
, commit_timeout
,
137 (unsigned long)current
);
141 /* Record that the journal thread is running */
142 journal
->j_task
= current
;
143 wake_up(&journal
->j_wait_done_commit
);
146 * And now, wait forever for commit wakeup events.
148 write_lock(&journal
->j_state_lock
);
151 if (journal
->j_flags
& JBD2_UNMOUNT
)
154 jbd_debug(1, "commit_sequence=%d, commit_request=%d\n",
155 journal
->j_commit_sequence
, journal
->j_commit_request
);
157 if (journal
->j_commit_sequence
!= journal
->j_commit_request
) {
158 jbd_debug(1, "OK, requests differ\n");
159 write_unlock(&journal
->j_state_lock
);
160 del_timer_sync(&journal
->j_commit_timer
);
161 jbd2_journal_commit_transaction(journal
);
162 write_lock(&journal
->j_state_lock
);
166 wake_up(&journal
->j_wait_done_commit
);
167 if (freezing(current
)) {
169 * The simpler the better. Flushing journal isn't a
170 * good idea, because that depends on threads that may
171 * be already stopped.
173 jbd_debug(1, "Now suspending kjournald2\n");
174 write_unlock(&journal
->j_state_lock
);
176 write_lock(&journal
->j_state_lock
);
179 * We assume on resume that commits are already there,
183 int should_sleep
= 1;
185 prepare_to_wait(&journal
->j_wait_commit
, &wait
,
187 if (journal
->j_commit_sequence
!= journal
->j_commit_request
)
189 transaction
= journal
->j_running_transaction
;
190 if (transaction
&& time_after_eq(jiffies
,
191 transaction
->t_expires
))
193 if (journal
->j_flags
& JBD2_UNMOUNT
)
196 write_unlock(&journal
->j_state_lock
);
198 write_lock(&journal
->j_state_lock
);
200 finish_wait(&journal
->j_wait_commit
, &wait
);
203 jbd_debug(1, "kjournald2 wakes\n");
206 * Were we woken up by a commit wakeup event?
208 transaction
= journal
->j_running_transaction
;
209 if (transaction
&& time_after_eq(jiffies
, transaction
->t_expires
)) {
210 journal
->j_commit_request
= transaction
->t_tid
;
211 jbd_debug(1, "woke because of timeout\n");
216 write_unlock(&journal
->j_state_lock
);
217 del_timer_sync(&journal
->j_commit_timer
);
218 journal
->j_task
= NULL
;
219 wake_up(&journal
->j_wait_done_commit
);
220 jbd_debug(1, "Journal thread exiting.\n");
224 static int jbd2_journal_start_thread(journal_t
*journal
)
226 struct task_struct
*t
;
228 t
= kthread_run(kjournald2
, journal
, "jbd2/%s",
233 wait_event(journal
->j_wait_done_commit
, journal
->j_task
!= NULL
);
237 static void journal_kill_thread(journal_t
*journal
)
239 write_lock(&journal
->j_state_lock
);
240 journal
->j_flags
|= JBD2_UNMOUNT
;
242 while (journal
->j_task
) {
243 wake_up(&journal
->j_wait_commit
);
244 write_unlock(&journal
->j_state_lock
);
245 wait_event(journal
->j_wait_done_commit
, journal
->j_task
== NULL
);
246 write_lock(&journal
->j_state_lock
);
248 write_unlock(&journal
->j_state_lock
);
252 * jbd2_journal_write_metadata_buffer: write a metadata buffer to the journal.
254 * Writes a metadata buffer to a given disk block. The actual IO is not
255 * performed but a new buffer_head is constructed which labels the data
256 * to be written with the correct destination disk block.
258 * Any magic-number escaping which needs to be done will cause a
259 * copy-out here. If the buffer happens to start with the
260 * JBD2_MAGIC_NUMBER, then we can't write it to the log directly: the
261 * magic number is only written to the log for descripter blocks. In
262 * this case, we copy the data and replace the first word with 0, and we
263 * return a result code which indicates that this buffer needs to be
264 * marked as an escaped buffer in the corresponding log descriptor
265 * block. The missing word can then be restored when the block is read
268 * If the source buffer has already been modified by a new transaction
269 * since we took the last commit snapshot, we use the frozen copy of
270 * that data for IO. If we end up using the existing buffer_head's data
271 * for the write, then we *have* to lock the buffer to prevent anyone
272 * else from using and possibly modifying it while the IO is in
275 * The function returns a pointer to the buffer_heads to be used for IO.
277 * We assume that the journal has already been locked in this function.
284 * Bit 0 set == escape performed on the data
285 * Bit 1 set == buffer copy-out performed (kfree the data after IO)
288 int jbd2_journal_write_metadata_buffer(transaction_t
*transaction
,
289 struct journal_head
*jh_in
,
290 struct journal_head
**jh_out
,
291 unsigned long long blocknr
)
293 int need_copy_out
= 0;
294 int done_copy_out
= 0;
297 struct buffer_head
*new_bh
;
298 struct journal_head
*new_jh
;
299 struct page
*new_page
;
300 unsigned int new_offset
;
301 struct buffer_head
*bh_in
= jh2bh(jh_in
);
302 journal_t
*journal
= transaction
->t_journal
;
305 * The buffer really shouldn't be locked: only the current committing
306 * transaction is allowed to write it, so nobody else is allowed
309 * akpm: except if we're journalling data, and write() output is
310 * also part of a shared mapping, and another thread has
311 * decided to launch a writepage() against this buffer.
313 J_ASSERT_BH(bh_in
, buffer_jbddirty(bh_in
));
316 new_bh
= alloc_buffer_head(GFP_NOFS
);
319 * Failure is not an option, but __GFP_NOFAIL is going
320 * away; so we retry ourselves here.
322 congestion_wait(BLK_RW_ASYNC
, HZ
/50);
326 /* keep subsequent assertions sane */
328 init_buffer(new_bh
, NULL
, NULL
);
329 atomic_set(&new_bh
->b_count
, 1);
330 new_jh
= jbd2_journal_add_journal_head(new_bh
); /* This sleeps */
333 * If a new transaction has already done a buffer copy-out, then
334 * we use that version of the data for the commit.
336 jbd_lock_bh_state(bh_in
);
338 if (jh_in
->b_frozen_data
) {
340 new_page
= virt_to_page(jh_in
->b_frozen_data
);
341 new_offset
= offset_in_page(jh_in
->b_frozen_data
);
343 new_page
= jh2bh(jh_in
)->b_page
;
344 new_offset
= offset_in_page(jh2bh(jh_in
)->b_data
);
347 mapped_data
= kmap_atomic(new_page
);
349 * Fire data frozen trigger if data already wasn't frozen. Do this
350 * before checking for escaping, as the trigger may modify the magic
351 * offset. If a copy-out happens afterwards, it will have the correct
352 * data in the buffer.
355 jbd2_buffer_frozen_trigger(jh_in
, mapped_data
+ new_offset
,
361 if (*((__be32
*)(mapped_data
+ new_offset
)) ==
362 cpu_to_be32(JBD2_MAGIC_NUMBER
)) {
366 kunmap_atomic(mapped_data
);
369 * Do we need to do a data copy?
371 if (need_copy_out
&& !done_copy_out
) {
374 jbd_unlock_bh_state(bh_in
);
375 tmp
= jbd2_alloc(bh_in
->b_size
, GFP_NOFS
);
377 jbd2_journal_put_journal_head(new_jh
);
380 jbd_lock_bh_state(bh_in
);
381 if (jh_in
->b_frozen_data
) {
382 jbd2_free(tmp
, bh_in
->b_size
);
386 jh_in
->b_frozen_data
= tmp
;
387 mapped_data
= kmap_atomic(new_page
);
388 memcpy(tmp
, mapped_data
+ new_offset
, jh2bh(jh_in
)->b_size
);
389 kunmap_atomic(mapped_data
);
391 new_page
= virt_to_page(tmp
);
392 new_offset
= offset_in_page(tmp
);
396 * This isn't strictly necessary, as we're using frozen
397 * data for the escaping, but it keeps consistency with
398 * b_frozen_data usage.
400 jh_in
->b_frozen_triggers
= jh_in
->b_triggers
;
404 * Did we need to do an escaping? Now we've done all the
405 * copying, we can finally do so.
408 mapped_data
= kmap_atomic(new_page
);
409 *((unsigned int *)(mapped_data
+ new_offset
)) = 0;
410 kunmap_atomic(mapped_data
);
413 set_bh_page(new_bh
, new_page
, new_offset
);
414 new_jh
->b_transaction
= NULL
;
415 new_bh
->b_size
= jh2bh(jh_in
)->b_size
;
416 new_bh
->b_bdev
= transaction
->t_journal
->j_dev
;
417 new_bh
->b_blocknr
= blocknr
;
418 set_buffer_mapped(new_bh
);
419 set_buffer_dirty(new_bh
);
424 * The to-be-written buffer needs to get moved to the io queue,
425 * and the original buffer whose contents we are shadowing or
426 * copying is moved to the transaction's shadow queue.
428 JBUFFER_TRACE(jh_in
, "file as BJ_Shadow");
429 spin_lock(&journal
->j_list_lock
);
430 __jbd2_journal_file_buffer(jh_in
, transaction
, BJ_Shadow
);
431 spin_unlock(&journal
->j_list_lock
);
432 jbd_unlock_bh_state(bh_in
);
434 JBUFFER_TRACE(new_jh
, "file as BJ_IO");
435 jbd2_journal_file_buffer(new_jh
, transaction
, BJ_IO
);
437 return do_escape
| (done_copy_out
<< 1);
441 * Allocation code for the journal file. Manage the space left in the
442 * journal, so that we can begin checkpointing when appropriate.
446 * __jbd2_log_space_left: Return the number of free blocks left in the journal.
448 * Called with the journal already locked.
450 * Called under j_state_lock
453 int __jbd2_log_space_left(journal_t
*journal
)
455 int left
= journal
->j_free
;
457 /* assert_spin_locked(&journal->j_state_lock); */
460 * Be pessimistic here about the number of those free blocks which
461 * might be required for log descriptor control blocks.
464 #define MIN_LOG_RESERVED_BLOCKS 32 /* Allow for rounding errors */
466 left
-= MIN_LOG_RESERVED_BLOCKS
;
475 * Called with j_state_lock locked for writing.
476 * Returns true if a transaction commit was started.
478 int __jbd2_log_start_commit(journal_t
*journal
, tid_t target
)
481 * The only transaction we can possibly wait upon is the
482 * currently running transaction (if it exists). Otherwise,
483 * the target tid must be an old one.
485 if (journal
->j_running_transaction
&&
486 journal
->j_running_transaction
->t_tid
== target
) {
488 * We want a new commit: OK, mark the request and wakeup the
489 * commit thread. We do _not_ do the commit ourselves.
492 journal
->j_commit_request
= target
;
493 jbd_debug(1, "JBD2: requesting commit %d/%d\n",
494 journal
->j_commit_request
,
495 journal
->j_commit_sequence
);
496 wake_up(&journal
->j_wait_commit
);
498 } else if (!tid_geq(journal
->j_commit_request
, target
))
499 /* This should never happen, but if it does, preserve
500 the evidence before kjournald goes into a loop and
501 increments j_commit_sequence beyond all recognition. */
502 WARN_ONCE(1, "JBD2: bad log_start_commit: %u %u %u %u\n",
503 journal
->j_commit_request
,
504 journal
->j_commit_sequence
,
505 target
, journal
->j_running_transaction
?
506 journal
->j_running_transaction
->t_tid
: 0);
510 int jbd2_log_start_commit(journal_t
*journal
, tid_t tid
)
514 write_lock(&journal
->j_state_lock
);
515 ret
= __jbd2_log_start_commit(journal
, tid
);
516 write_unlock(&journal
->j_state_lock
);
521 * Force and wait upon a commit if the calling process is not within
522 * transaction. This is used for forcing out undo-protected data which contains
523 * bitmaps, when the fs is running out of space.
525 * We can only force the running transaction if we don't have an active handle;
526 * otherwise, we will deadlock.
528 * Returns true if a transaction was started.
530 int jbd2_journal_force_commit_nested(journal_t
*journal
)
532 transaction_t
*transaction
= NULL
;
534 int need_to_start
= 0;
536 read_lock(&journal
->j_state_lock
);
537 if (journal
->j_running_transaction
&& !current
->journal_info
) {
538 transaction
= journal
->j_running_transaction
;
539 if (!tid_geq(journal
->j_commit_request
, transaction
->t_tid
))
541 } else if (journal
->j_committing_transaction
)
542 transaction
= journal
->j_committing_transaction
;
545 read_unlock(&journal
->j_state_lock
);
546 return 0; /* Nothing to retry */
549 tid
= transaction
->t_tid
;
550 read_unlock(&journal
->j_state_lock
);
552 jbd2_log_start_commit(journal
, tid
);
553 jbd2_log_wait_commit(journal
, tid
);
558 * Start a commit of the current running transaction (if any). Returns true
559 * if a transaction is going to be committed (or is currently already
560 * committing), and fills its tid in at *ptid
562 int jbd2_journal_start_commit(journal_t
*journal
, tid_t
*ptid
)
566 write_lock(&journal
->j_state_lock
);
567 if (journal
->j_running_transaction
) {
568 tid_t tid
= journal
->j_running_transaction
->t_tid
;
570 __jbd2_log_start_commit(journal
, tid
);
571 /* There's a running transaction and we've just made sure
572 * it's commit has been scheduled. */
576 } else if (journal
->j_committing_transaction
) {
578 * If ext3_write_super() recently started a commit, then we
579 * have to wait for completion of that transaction
582 *ptid
= journal
->j_committing_transaction
->t_tid
;
585 write_unlock(&journal
->j_state_lock
);
590 * Return 1 if a given transaction has not yet sent barrier request
591 * connected with a transaction commit. If 0 is returned, transaction
592 * may or may not have sent the barrier. Used to avoid sending barrier
593 * twice in common cases.
595 int jbd2_trans_will_send_data_barrier(journal_t
*journal
, tid_t tid
)
598 transaction_t
*commit_trans
;
600 if (!(journal
->j_flags
& JBD2_BARRIER
))
602 read_lock(&journal
->j_state_lock
);
603 /* Transaction already committed? */
604 if (tid_geq(journal
->j_commit_sequence
, tid
))
606 commit_trans
= journal
->j_committing_transaction
;
607 if (!commit_trans
|| commit_trans
->t_tid
!= tid
) {
612 * Transaction is being committed and we already proceeded to
613 * submitting a flush to fs partition?
615 if (journal
->j_fs_dev
!= journal
->j_dev
) {
616 if (!commit_trans
->t_need_data_flush
||
617 commit_trans
->t_state
>= T_COMMIT_DFLUSH
)
620 if (commit_trans
->t_state
>= T_COMMIT_JFLUSH
)
625 read_unlock(&journal
->j_state_lock
);
628 EXPORT_SYMBOL(jbd2_trans_will_send_data_barrier
);
631 * Wait for a specified commit to complete.
632 * The caller may not hold the journal lock.
634 int jbd2_log_wait_commit(journal_t
*journal
, tid_t tid
)
638 read_lock(&journal
->j_state_lock
);
639 #ifdef CONFIG_JBD2_DEBUG
640 if (!tid_geq(journal
->j_commit_request
, tid
)) {
642 "%s: error: j_commit_request=%d, tid=%d\n",
643 __func__
, journal
->j_commit_request
, tid
);
646 while (tid_gt(tid
, journal
->j_commit_sequence
)) {
647 jbd_debug(1, "JBD2: want %d, j_commit_sequence=%d\n",
648 tid
, journal
->j_commit_sequence
);
649 wake_up(&journal
->j_wait_commit
);
650 read_unlock(&journal
->j_state_lock
);
651 wait_event(journal
->j_wait_done_commit
,
652 !tid_gt(tid
, journal
->j_commit_sequence
));
653 read_lock(&journal
->j_state_lock
);
655 read_unlock(&journal
->j_state_lock
);
657 if (unlikely(is_journal_aborted(journal
))) {
658 printk(KERN_EMERG
"journal commit I/O error\n");
665 * Log buffer allocation routines:
668 int jbd2_journal_next_log_block(journal_t
*journal
, unsigned long long *retp
)
670 unsigned long blocknr
;
672 write_lock(&journal
->j_state_lock
);
673 J_ASSERT(journal
->j_free
> 1);
675 blocknr
= journal
->j_head
;
678 if (journal
->j_head
== journal
->j_last
)
679 journal
->j_head
= journal
->j_first
;
680 write_unlock(&journal
->j_state_lock
);
681 return jbd2_journal_bmap(journal
, blocknr
, retp
);
685 * Conversion of logical to physical block numbers for the journal
687 * On external journals the journal blocks are identity-mapped, so
688 * this is a no-op. If needed, we can use j_blk_offset - everything is
691 int jbd2_journal_bmap(journal_t
*journal
, unsigned long blocknr
,
692 unsigned long long *retp
)
695 unsigned long long ret
;
697 if (journal
->j_inode
) {
698 ret
= bmap(journal
->j_inode
, blocknr
);
702 printk(KERN_ALERT
"%s: journal block not found "
703 "at offset %lu on %s\n",
704 __func__
, blocknr
, journal
->j_devname
);
706 __journal_abort_soft(journal
, err
);
709 *retp
= blocknr
; /* +journal->j_blk_offset */
715 * We play buffer_head aliasing tricks to write data/metadata blocks to
716 * the journal without copying their contents, but for journal
717 * descriptor blocks we do need to generate bona fide buffers.
719 * After the caller of jbd2_journal_get_descriptor_buffer() has finished modifying
720 * the buffer's contents they really should run flush_dcache_page(bh->b_page).
721 * But we don't bother doing that, so there will be coherency problems with
722 * mmaps of blockdevs which hold live JBD-controlled filesystems.
724 struct journal_head
*jbd2_journal_get_descriptor_buffer(journal_t
*journal
)
726 struct buffer_head
*bh
;
727 unsigned long long blocknr
;
730 err
= jbd2_journal_next_log_block(journal
, &blocknr
);
735 bh
= __getblk(journal
->j_dev
, blocknr
, journal
->j_blocksize
);
739 memset(bh
->b_data
, 0, journal
->j_blocksize
);
740 set_buffer_uptodate(bh
);
742 BUFFER_TRACE(bh
, "return this buffer");
743 return jbd2_journal_add_journal_head(bh
);
747 * Return tid of the oldest transaction in the journal and block in the journal
748 * where the transaction starts.
750 * If the journal is now empty, return which will be the next transaction ID
751 * we will write and where will that transaction start.
753 * The return value is 0 if journal tail cannot be pushed any further, 1 if
756 int jbd2_journal_get_log_tail(journal_t
*journal
, tid_t
*tid
,
757 unsigned long *block
)
759 transaction_t
*transaction
;
762 read_lock(&journal
->j_state_lock
);
763 spin_lock(&journal
->j_list_lock
);
764 transaction
= journal
->j_checkpoint_transactions
;
766 *tid
= transaction
->t_tid
;
767 *block
= transaction
->t_log_start
;
768 } else if ((transaction
= journal
->j_committing_transaction
) != NULL
) {
769 *tid
= transaction
->t_tid
;
770 *block
= transaction
->t_log_start
;
771 } else if ((transaction
= journal
->j_running_transaction
) != NULL
) {
772 *tid
= transaction
->t_tid
;
773 *block
= journal
->j_head
;
775 *tid
= journal
->j_transaction_sequence
;
776 *block
= journal
->j_head
;
778 ret
= tid_gt(*tid
, journal
->j_tail_sequence
);
779 spin_unlock(&journal
->j_list_lock
);
780 read_unlock(&journal
->j_state_lock
);
786 * Update information in journal structure and in on disk journal superblock
787 * about log tail. This function does not check whether information passed in
788 * really pushes log tail further. It's responsibility of the caller to make
789 * sure provided log tail information is valid (e.g. by holding
790 * j_checkpoint_mutex all the time between computing log tail and calling this
791 * function as is the case with jbd2_cleanup_journal_tail()).
793 * Requires j_checkpoint_mutex
795 void __jbd2_update_log_tail(journal_t
*journal
, tid_t tid
, unsigned long block
)
799 BUG_ON(!mutex_is_locked(&journal
->j_checkpoint_mutex
));
802 * We cannot afford for write to remain in drive's caches since as
803 * soon as we update j_tail, next transaction can start reusing journal
804 * space and if we lose sb update during power failure we'd replay
805 * old transaction with possibly newly overwritten data.
807 jbd2_journal_update_sb_log_tail(journal
, tid
, block
, WRITE_FUA
);
808 write_lock(&journal
->j_state_lock
);
809 freed
= block
- journal
->j_tail
;
810 if (block
< journal
->j_tail
)
811 freed
+= journal
->j_last
- journal
->j_first
;
813 trace_jbd2_update_log_tail(journal
, tid
, block
, freed
);
815 "Cleaning journal tail from %d to %d (offset %lu), "
817 journal
->j_tail_sequence
, tid
, block
, freed
);
819 journal
->j_free
+= freed
;
820 journal
->j_tail_sequence
= tid
;
821 journal
->j_tail
= block
;
822 write_unlock(&journal
->j_state_lock
);
826 * This is a variaon of __jbd2_update_log_tail which checks for validity of
827 * provided log tail and locks j_checkpoint_mutex. So it is safe against races
828 * with other threads updating log tail.
830 void jbd2_update_log_tail(journal_t
*journal
, tid_t tid
, unsigned long block
)
832 mutex_lock(&journal
->j_checkpoint_mutex
);
833 if (tid_gt(tid
, journal
->j_tail_sequence
))
834 __jbd2_update_log_tail(journal
, tid
, block
);
835 mutex_unlock(&journal
->j_checkpoint_mutex
);
838 struct jbd2_stats_proc_session
{
840 struct transaction_stats_s
*stats
;
845 static void *jbd2_seq_info_start(struct seq_file
*seq
, loff_t
*pos
)
847 return *pos
? NULL
: SEQ_START_TOKEN
;
850 static void *jbd2_seq_info_next(struct seq_file
*seq
, void *v
, loff_t
*pos
)
855 static int jbd2_seq_info_show(struct seq_file
*seq
, void *v
)
857 struct jbd2_stats_proc_session
*s
= seq
->private;
859 if (v
!= SEQ_START_TOKEN
)
861 seq_printf(seq
, "%lu transaction, each up to %u blocks\n",
863 s
->journal
->j_max_transaction_buffers
);
864 if (s
->stats
->ts_tid
== 0)
866 seq_printf(seq
, "average: \n %ums waiting for transaction\n",
867 jiffies_to_msecs(s
->stats
->run
.rs_wait
/ s
->stats
->ts_tid
));
868 seq_printf(seq
, " %ums running transaction\n",
869 jiffies_to_msecs(s
->stats
->run
.rs_running
/ s
->stats
->ts_tid
));
870 seq_printf(seq
, " %ums transaction was being locked\n",
871 jiffies_to_msecs(s
->stats
->run
.rs_locked
/ s
->stats
->ts_tid
));
872 seq_printf(seq
, " %ums flushing data (in ordered mode)\n",
873 jiffies_to_msecs(s
->stats
->run
.rs_flushing
/ s
->stats
->ts_tid
));
874 seq_printf(seq
, " %ums logging transaction\n",
875 jiffies_to_msecs(s
->stats
->run
.rs_logging
/ s
->stats
->ts_tid
));
876 seq_printf(seq
, " %lluus average transaction commit time\n",
877 div_u64(s
->journal
->j_average_commit_time
, 1000));
878 seq_printf(seq
, " %lu handles per transaction\n",
879 s
->stats
->run
.rs_handle_count
/ s
->stats
->ts_tid
);
880 seq_printf(seq
, " %lu blocks per transaction\n",
881 s
->stats
->run
.rs_blocks
/ s
->stats
->ts_tid
);
882 seq_printf(seq
, " %lu logged blocks per transaction\n",
883 s
->stats
->run
.rs_blocks_logged
/ s
->stats
->ts_tid
);
887 static void jbd2_seq_info_stop(struct seq_file
*seq
, void *v
)
891 static const struct seq_operations jbd2_seq_info_ops
= {
892 .start
= jbd2_seq_info_start
,
893 .next
= jbd2_seq_info_next
,
894 .stop
= jbd2_seq_info_stop
,
895 .show
= jbd2_seq_info_show
,
898 static int jbd2_seq_info_open(struct inode
*inode
, struct file
*file
)
900 journal_t
*journal
= PDE(inode
)->data
;
901 struct jbd2_stats_proc_session
*s
;
904 s
= kmalloc(sizeof(*s
), GFP_KERNEL
);
907 size
= sizeof(struct transaction_stats_s
);
908 s
->stats
= kmalloc(size
, GFP_KERNEL
);
909 if (s
->stats
== NULL
) {
913 spin_lock(&journal
->j_history_lock
);
914 memcpy(s
->stats
, &journal
->j_stats
, size
);
915 s
->journal
= journal
;
916 spin_unlock(&journal
->j_history_lock
);
918 rc
= seq_open(file
, &jbd2_seq_info_ops
);
920 struct seq_file
*m
= file
->private_data
;
930 static int jbd2_seq_info_release(struct inode
*inode
, struct file
*file
)
932 struct seq_file
*seq
= file
->private_data
;
933 struct jbd2_stats_proc_session
*s
= seq
->private;
936 return seq_release(inode
, file
);
939 static const struct file_operations jbd2_seq_info_fops
= {
940 .owner
= THIS_MODULE
,
941 .open
= jbd2_seq_info_open
,
944 .release
= jbd2_seq_info_release
,
947 static struct proc_dir_entry
*proc_jbd2_stats
;
949 static void jbd2_stats_proc_init(journal_t
*journal
)
951 journal
->j_proc_entry
= proc_mkdir(journal
->j_devname
, proc_jbd2_stats
);
952 if (journal
->j_proc_entry
) {
953 proc_create_data("info", S_IRUGO
, journal
->j_proc_entry
,
954 &jbd2_seq_info_fops
, journal
);
958 static void jbd2_stats_proc_exit(journal_t
*journal
)
960 remove_proc_entry("info", journal
->j_proc_entry
);
961 remove_proc_entry(journal
->j_devname
, proc_jbd2_stats
);
965 * Management for journal control blocks: functions to create and
966 * destroy journal_t structures, and to initialise and read existing
967 * journal blocks from disk. */
969 /* First: create and setup a journal_t object in memory. We initialise
970 * very few fields yet: that has to wait until we have created the
971 * journal structures from from scratch, or loaded them from disk. */
973 static journal_t
* journal_init_common (void)
978 journal
= kzalloc(sizeof(*journal
), GFP_KERNEL
);
982 init_waitqueue_head(&journal
->j_wait_transaction_locked
);
983 init_waitqueue_head(&journal
->j_wait_logspace
);
984 init_waitqueue_head(&journal
->j_wait_done_commit
);
985 init_waitqueue_head(&journal
->j_wait_checkpoint
);
986 init_waitqueue_head(&journal
->j_wait_commit
);
987 init_waitqueue_head(&journal
->j_wait_updates
);
988 mutex_init(&journal
->j_barrier
);
989 mutex_init(&journal
->j_checkpoint_mutex
);
990 spin_lock_init(&journal
->j_revoke_lock
);
991 spin_lock_init(&journal
->j_list_lock
);
992 rwlock_init(&journal
->j_state_lock
);
994 journal
->j_commit_interval
= (HZ
* JBD2_DEFAULT_MAX_COMMIT_AGE
);
995 journal
->j_min_batch_time
= 0;
996 journal
->j_max_batch_time
= 15000; /* 15ms */
998 /* The journal is marked for error until we succeed with recovery! */
999 journal
->j_flags
= JBD2_ABORT
;
1001 /* Set up a default-sized revoke table for the new mount. */
1002 err
= jbd2_journal_init_revoke(journal
, JOURNAL_REVOKE_DEFAULT_HASH
);
1008 spin_lock_init(&journal
->j_history_lock
);
1013 /* jbd2_journal_init_dev and jbd2_journal_init_inode:
1015 * Create a journal structure assigned some fixed set of disk blocks to
1016 * the journal. We don't actually touch those disk blocks yet, but we
1017 * need to set up all of the mapping information to tell the journaling
1018 * system where the journal blocks are.
1023 * journal_t * jbd2_journal_init_dev() - creates and initialises a journal structure
1024 * @bdev: Block device on which to create the journal
1025 * @fs_dev: Device which hold journalled filesystem for this journal.
1026 * @start: Block nr Start of journal.
1027 * @len: Length of the journal in blocks.
1028 * @blocksize: blocksize of journalling device
1030 * Returns: a newly created journal_t *
1032 * jbd2_journal_init_dev creates a journal which maps a fixed contiguous
1033 * range of blocks on an arbitrary block device.
1036 journal_t
* jbd2_journal_init_dev(struct block_device
*bdev
,
1037 struct block_device
*fs_dev
,
1038 unsigned long long start
, int len
, int blocksize
)
1040 journal_t
*journal
= journal_init_common();
1041 struct buffer_head
*bh
;
1048 /* journal descriptor can store up to n blocks -bzzz */
1049 journal
->j_blocksize
= blocksize
;
1050 journal
->j_dev
= bdev
;
1051 journal
->j_fs_dev
= fs_dev
;
1052 journal
->j_blk_offset
= start
;
1053 journal
->j_maxlen
= len
;
1054 bdevname(journal
->j_dev
, journal
->j_devname
);
1055 p
= journal
->j_devname
;
1056 while ((p
= strchr(p
, '/')))
1058 jbd2_stats_proc_init(journal
);
1059 n
= journal
->j_blocksize
/ sizeof(journal_block_tag_t
);
1060 journal
->j_wbufsize
= n
;
1061 journal
->j_wbuf
= kmalloc(n
* sizeof(struct buffer_head
*), GFP_KERNEL
);
1062 if (!journal
->j_wbuf
) {
1063 printk(KERN_ERR
"%s: Can't allocate bhs for commit thread\n",
1068 bh
= __getblk(journal
->j_dev
, start
, journal
->j_blocksize
);
1071 "%s: Cannot get buffer for journal superblock\n",
1075 journal
->j_sb_buffer
= bh
;
1076 journal
->j_superblock
= (journal_superblock_t
*)bh
->b_data
;
1080 kfree(journal
->j_wbuf
);
1081 jbd2_stats_proc_exit(journal
);
1087 * journal_t * jbd2_journal_init_inode () - creates a journal which maps to a inode.
1088 * @inode: An inode to create the journal in
1090 * jbd2_journal_init_inode creates a journal which maps an on-disk inode as
1091 * the journal. The inode must exist already, must support bmap() and
1092 * must have all data blocks preallocated.
1094 journal_t
* jbd2_journal_init_inode (struct inode
*inode
)
1096 struct buffer_head
*bh
;
1097 journal_t
*journal
= journal_init_common();
1101 unsigned long long blocknr
;
1106 journal
->j_dev
= journal
->j_fs_dev
= inode
->i_sb
->s_bdev
;
1107 journal
->j_inode
= inode
;
1108 bdevname(journal
->j_dev
, journal
->j_devname
);
1109 p
= journal
->j_devname
;
1110 while ((p
= strchr(p
, '/')))
1112 p
= journal
->j_devname
+ strlen(journal
->j_devname
);
1113 sprintf(p
, "-%lu", journal
->j_inode
->i_ino
);
1115 "journal %p: inode %s/%ld, size %Ld, bits %d, blksize %ld\n",
1116 journal
, inode
->i_sb
->s_id
, inode
->i_ino
,
1117 (long long) inode
->i_size
,
1118 inode
->i_sb
->s_blocksize_bits
, inode
->i_sb
->s_blocksize
);
1120 journal
->j_maxlen
= inode
->i_size
>> inode
->i_sb
->s_blocksize_bits
;
1121 journal
->j_blocksize
= inode
->i_sb
->s_blocksize
;
1122 jbd2_stats_proc_init(journal
);
1124 /* journal descriptor can store up to n blocks -bzzz */
1125 n
= journal
->j_blocksize
/ sizeof(journal_block_tag_t
);
1126 journal
->j_wbufsize
= n
;
1127 journal
->j_wbuf
= kmalloc(n
* sizeof(struct buffer_head
*), GFP_KERNEL
);
1128 if (!journal
->j_wbuf
) {
1129 printk(KERN_ERR
"%s: Can't allocate bhs for commit thread\n",
1134 err
= jbd2_journal_bmap(journal
, 0, &blocknr
);
1135 /* If that failed, give up */
1137 printk(KERN_ERR
"%s: Cannot locate journal superblock\n",
1142 bh
= __getblk(journal
->j_dev
, blocknr
, journal
->j_blocksize
);
1145 "%s: Cannot get buffer for journal superblock\n",
1149 journal
->j_sb_buffer
= bh
;
1150 journal
->j_superblock
= (journal_superblock_t
*)bh
->b_data
;
1154 kfree(journal
->j_wbuf
);
1155 jbd2_stats_proc_exit(journal
);
1161 * If the journal init or create aborts, we need to mark the journal
1162 * superblock as being NULL to prevent the journal destroy from writing
1163 * back a bogus superblock.
1165 static void journal_fail_superblock (journal_t
*journal
)
1167 struct buffer_head
*bh
= journal
->j_sb_buffer
;
1169 journal
->j_sb_buffer
= NULL
;
1173 * Given a journal_t structure, initialise the various fields for
1174 * startup of a new journaling session. We use this both when creating
1175 * a journal, and after recovering an old journal to reset it for
1179 static int journal_reset(journal_t
*journal
)
1181 journal_superblock_t
*sb
= journal
->j_superblock
;
1182 unsigned long long first
, last
;
1184 first
= be32_to_cpu(sb
->s_first
);
1185 last
= be32_to_cpu(sb
->s_maxlen
);
1186 if (first
+ JBD2_MIN_JOURNAL_BLOCKS
> last
+ 1) {
1187 printk(KERN_ERR
"JBD2: Journal too short (blocks %llu-%llu).\n",
1189 journal_fail_superblock(journal
);
1193 journal
->j_first
= first
;
1194 journal
->j_last
= last
;
1196 journal
->j_head
= first
;
1197 journal
->j_tail
= first
;
1198 journal
->j_free
= last
- first
;
1200 journal
->j_tail_sequence
= journal
->j_transaction_sequence
;
1201 journal
->j_commit_sequence
= journal
->j_transaction_sequence
- 1;
1202 journal
->j_commit_request
= journal
->j_commit_sequence
;
1204 journal
->j_max_transaction_buffers
= journal
->j_maxlen
/ 4;
1207 * As a special case, if the on-disk copy is already marked as needing
1208 * no recovery (s_start == 0), then we can safely defer the superblock
1209 * update until the next commit by setting JBD2_FLUSHED. This avoids
1210 * attempting a write to a potential-readonly device.
1212 if (sb
->s_start
== 0) {
1213 jbd_debug(1, "JBD2: Skipping superblock update on recovered sb "
1214 "(start %ld, seq %d, errno %d)\n",
1215 journal
->j_tail
, journal
->j_tail_sequence
,
1217 journal
->j_flags
|= JBD2_FLUSHED
;
1219 /* Lock here to make assertions happy... */
1220 mutex_lock(&journal
->j_checkpoint_mutex
);
1222 * Update log tail information. We use WRITE_FUA since new
1223 * transaction will start reusing journal space and so we
1224 * must make sure information about current log tail is on
1227 jbd2_journal_update_sb_log_tail(journal
,
1228 journal
->j_tail_sequence
,
1231 mutex_unlock(&journal
->j_checkpoint_mutex
);
1233 return jbd2_journal_start_thread(journal
);
1236 static void jbd2_write_superblock(journal_t
*journal
, int write_op
)
1238 struct buffer_head
*bh
= journal
->j_sb_buffer
;
1241 trace_jbd2_write_superblock(journal
, write_op
);
1242 if (!(journal
->j_flags
& JBD2_BARRIER
))
1243 write_op
&= ~(REQ_FUA
| REQ_FLUSH
);
1245 if (buffer_write_io_error(bh
)) {
1247 * Oh, dear. A previous attempt to write the journal
1248 * superblock failed. This could happen because the
1249 * USB device was yanked out. Or it could happen to
1250 * be a transient write error and maybe the block will
1251 * be remapped. Nothing we can do but to retry the
1252 * write and hope for the best.
1254 printk(KERN_ERR
"JBD2: previous I/O error detected "
1255 "for journal superblock update for %s.\n",
1256 journal
->j_devname
);
1257 clear_buffer_write_io_error(bh
);
1258 set_buffer_uptodate(bh
);
1261 bh
->b_end_io
= end_buffer_write_sync
;
1262 ret
= submit_bh(write_op
, bh
);
1264 if (buffer_write_io_error(bh
)) {
1265 clear_buffer_write_io_error(bh
);
1266 set_buffer_uptodate(bh
);
1270 printk(KERN_ERR
"JBD2: Error %d detected when updating "
1271 "journal superblock for %s.\n", ret
,
1272 journal
->j_devname
);
1277 * jbd2_journal_update_sb_log_tail() - Update log tail in journal sb on disk.
1278 * @journal: The journal to update.
1279 * @tail_tid: TID of the new transaction at the tail of the log
1280 * @tail_block: The first block of the transaction at the tail of the log
1281 * @write_op: With which operation should we write the journal sb
1283 * Update a journal's superblock information about log tail and write it to
1284 * disk, waiting for the IO to complete.
1286 void jbd2_journal_update_sb_log_tail(journal_t
*journal
, tid_t tail_tid
,
1287 unsigned long tail_block
, int write_op
)
1289 journal_superblock_t
*sb
= journal
->j_superblock
;
1291 BUG_ON(!mutex_is_locked(&journal
->j_checkpoint_mutex
));
1292 jbd_debug(1, "JBD2: updating superblock (start %lu, seq %u)\n",
1293 tail_block
, tail_tid
);
1295 sb
->s_sequence
= cpu_to_be32(tail_tid
);
1296 sb
->s_start
= cpu_to_be32(tail_block
);
1298 jbd2_write_superblock(journal
, write_op
);
1300 /* Log is no longer empty */
1301 write_lock(&journal
->j_state_lock
);
1302 WARN_ON(!sb
->s_sequence
);
1303 journal
->j_flags
&= ~JBD2_FLUSHED
;
1304 write_unlock(&journal
->j_state_lock
);
1308 * jbd2_mark_journal_empty() - Mark on disk journal as empty.
1309 * @journal: The journal to update.
1311 * Update a journal's dynamic superblock fields to show that journal is empty.
1312 * Write updated superblock to disk waiting for IO to complete.
1314 static void jbd2_mark_journal_empty(journal_t
*journal
)
1316 journal_superblock_t
*sb
= journal
->j_superblock
;
1318 BUG_ON(!mutex_is_locked(&journal
->j_checkpoint_mutex
));
1319 read_lock(&journal
->j_state_lock
);
1320 jbd_debug(1, "JBD2: Marking journal as empty (seq %d)\n",
1321 journal
->j_tail_sequence
);
1323 sb
->s_sequence
= cpu_to_be32(journal
->j_tail_sequence
);
1324 sb
->s_start
= cpu_to_be32(0);
1325 read_unlock(&journal
->j_state_lock
);
1327 jbd2_write_superblock(journal
, WRITE_FUA
);
1329 /* Log is no longer empty */
1330 write_lock(&journal
->j_state_lock
);
1331 journal
->j_flags
|= JBD2_FLUSHED
;
1332 write_unlock(&journal
->j_state_lock
);
1337 * jbd2_journal_update_sb_errno() - Update error in the journal.
1338 * @journal: The journal to update.
1340 * Update a journal's errno. Write updated superblock to disk waiting for IO
1343 static void jbd2_journal_update_sb_errno(journal_t
*journal
)
1345 journal_superblock_t
*sb
= journal
->j_superblock
;
1347 read_lock(&journal
->j_state_lock
);
1348 jbd_debug(1, "JBD2: updating superblock error (errno %d)\n",
1350 sb
->s_errno
= cpu_to_be32(journal
->j_errno
);
1351 read_unlock(&journal
->j_state_lock
);
1353 jbd2_write_superblock(journal
, WRITE_SYNC
);
1357 * Read the superblock for a given journal, performing initial
1358 * validation of the format.
1360 static int journal_get_superblock(journal_t
*journal
)
1362 struct buffer_head
*bh
;
1363 journal_superblock_t
*sb
;
1366 bh
= journal
->j_sb_buffer
;
1368 J_ASSERT(bh
!= NULL
);
1369 if (!buffer_uptodate(bh
)) {
1370 ll_rw_block(READ
, 1, &bh
);
1372 if (!buffer_uptodate(bh
)) {
1374 "JBD2: IO error reading journal superblock\n");
1379 sb
= journal
->j_superblock
;
1383 if (sb
->s_header
.h_magic
!= cpu_to_be32(JBD2_MAGIC_NUMBER
) ||
1384 sb
->s_blocksize
!= cpu_to_be32(journal
->j_blocksize
)) {
1385 printk(KERN_WARNING
"JBD2: no valid journal superblock found\n");
1389 switch(be32_to_cpu(sb
->s_header
.h_blocktype
)) {
1390 case JBD2_SUPERBLOCK_V1
:
1391 journal
->j_format_version
= 1;
1393 case JBD2_SUPERBLOCK_V2
:
1394 journal
->j_format_version
= 2;
1397 printk(KERN_WARNING
"JBD2: unrecognised superblock format ID\n");
1401 if (be32_to_cpu(sb
->s_maxlen
) < journal
->j_maxlen
)
1402 journal
->j_maxlen
= be32_to_cpu(sb
->s_maxlen
);
1403 else if (be32_to_cpu(sb
->s_maxlen
) > journal
->j_maxlen
) {
1404 printk(KERN_WARNING
"JBD2: journal file too short\n");
1408 if (be32_to_cpu(sb
->s_first
) == 0 ||
1409 be32_to_cpu(sb
->s_first
) >= journal
->j_maxlen
) {
1411 "JBD2: Invalid start block of journal: %u\n",
1412 be32_to_cpu(sb
->s_first
));
1419 journal_fail_superblock(journal
);
1424 * Load the on-disk journal superblock and read the key fields into the
1428 static int load_superblock(journal_t
*journal
)
1431 journal_superblock_t
*sb
;
1433 err
= journal_get_superblock(journal
);
1437 sb
= journal
->j_superblock
;
1439 journal
->j_tail_sequence
= be32_to_cpu(sb
->s_sequence
);
1440 journal
->j_tail
= be32_to_cpu(sb
->s_start
);
1441 journal
->j_first
= be32_to_cpu(sb
->s_first
);
1442 journal
->j_last
= be32_to_cpu(sb
->s_maxlen
);
1443 journal
->j_errno
= be32_to_cpu(sb
->s_errno
);
1450 * int jbd2_journal_load() - Read journal from disk.
1451 * @journal: Journal to act on.
1453 * Given a journal_t structure which tells us which disk blocks contain
1454 * a journal, read the journal from disk to initialise the in-memory
1457 int jbd2_journal_load(journal_t
*journal
)
1460 journal_superblock_t
*sb
;
1462 err
= load_superblock(journal
);
1466 sb
= journal
->j_superblock
;
1467 /* If this is a V2 superblock, then we have to check the
1468 * features flags on it. */
1470 if (journal
->j_format_version
>= 2) {
1471 if ((sb
->s_feature_ro_compat
&
1472 ~cpu_to_be32(JBD2_KNOWN_ROCOMPAT_FEATURES
)) ||
1473 (sb
->s_feature_incompat
&
1474 ~cpu_to_be32(JBD2_KNOWN_INCOMPAT_FEATURES
))) {
1476 "JBD2: Unrecognised features on journal\n");
1482 * Create a slab for this blocksize
1484 err
= jbd2_journal_create_slab(be32_to_cpu(sb
->s_blocksize
));
1488 /* Let the recovery code check whether it needs to recover any
1489 * data from the journal. */
1490 if (jbd2_journal_recover(journal
))
1491 goto recovery_error
;
1493 if (journal
->j_failed_commit
) {
1494 printk(KERN_ERR
"JBD2: journal transaction %u on %s "
1495 "is corrupt.\n", journal
->j_failed_commit
,
1496 journal
->j_devname
);
1500 /* OK, we've finished with the dynamic journal bits:
1501 * reinitialise the dynamic contents of the superblock in memory
1502 * and reset them on disk. */
1503 if (journal_reset(journal
))
1504 goto recovery_error
;
1506 journal
->j_flags
&= ~JBD2_ABORT
;
1507 journal
->j_flags
|= JBD2_LOADED
;
1511 printk(KERN_WARNING
"JBD2: recovery failed\n");
1516 * void jbd2_journal_destroy() - Release a journal_t structure.
1517 * @journal: Journal to act on.
1519 * Release a journal_t structure once it is no longer in use by the
1521 * Return <0 if we couldn't clean up the journal.
1523 int jbd2_journal_destroy(journal_t
*journal
)
1527 /* Wait for the commit thread to wake up and die. */
1528 journal_kill_thread(journal
);
1530 /* Force a final log commit */
1531 if (journal
->j_running_transaction
)
1532 jbd2_journal_commit_transaction(journal
);
1534 /* Force any old transactions to disk */
1536 /* Totally anal locking here... */
1537 spin_lock(&journal
->j_list_lock
);
1538 while (journal
->j_checkpoint_transactions
!= NULL
) {
1539 spin_unlock(&journal
->j_list_lock
);
1540 mutex_lock(&journal
->j_checkpoint_mutex
);
1541 jbd2_log_do_checkpoint(journal
);
1542 mutex_unlock(&journal
->j_checkpoint_mutex
);
1543 spin_lock(&journal
->j_list_lock
);
1546 J_ASSERT(journal
->j_running_transaction
== NULL
);
1547 J_ASSERT(journal
->j_committing_transaction
== NULL
);
1548 J_ASSERT(journal
->j_checkpoint_transactions
== NULL
);
1549 spin_unlock(&journal
->j_list_lock
);
1551 if (journal
->j_sb_buffer
) {
1552 if (!is_journal_aborted(journal
)) {
1553 mutex_lock(&journal
->j_checkpoint_mutex
);
1554 jbd2_mark_journal_empty(journal
);
1555 mutex_unlock(&journal
->j_checkpoint_mutex
);
1558 brelse(journal
->j_sb_buffer
);
1561 if (journal
->j_proc_entry
)
1562 jbd2_stats_proc_exit(journal
);
1563 if (journal
->j_inode
)
1564 iput(journal
->j_inode
);
1565 if (journal
->j_revoke
)
1566 jbd2_journal_destroy_revoke(journal
);
1567 kfree(journal
->j_wbuf
);
1575 *int jbd2_journal_check_used_features () - Check if features specified are used.
1576 * @journal: Journal to check.
1577 * @compat: bitmask of compatible features
1578 * @ro: bitmask of features that force read-only mount
1579 * @incompat: bitmask of incompatible features
1581 * Check whether the journal uses all of a given set of
1582 * features. Return true (non-zero) if it does.
1585 int jbd2_journal_check_used_features (journal_t
*journal
, unsigned long compat
,
1586 unsigned long ro
, unsigned long incompat
)
1588 journal_superblock_t
*sb
;
1590 if (!compat
&& !ro
&& !incompat
)
1592 /* Load journal superblock if it is not loaded yet. */
1593 if (journal
->j_format_version
== 0 &&
1594 journal_get_superblock(journal
) != 0)
1596 if (journal
->j_format_version
== 1)
1599 sb
= journal
->j_superblock
;
1601 if (((be32_to_cpu(sb
->s_feature_compat
) & compat
) == compat
) &&
1602 ((be32_to_cpu(sb
->s_feature_ro_compat
) & ro
) == ro
) &&
1603 ((be32_to_cpu(sb
->s_feature_incompat
) & incompat
) == incompat
))
1610 * int jbd2_journal_check_available_features() - Check feature set in journalling layer
1611 * @journal: Journal to check.
1612 * @compat: bitmask of compatible features
1613 * @ro: bitmask of features that force read-only mount
1614 * @incompat: bitmask of incompatible features
1616 * Check whether the journaling code supports the use of
1617 * all of a given set of features on this journal. Return true
1618 * (non-zero) if it can. */
1620 int jbd2_journal_check_available_features (journal_t
*journal
, unsigned long compat
,
1621 unsigned long ro
, unsigned long incompat
)
1623 if (!compat
&& !ro
&& !incompat
)
1626 /* We can support any known requested features iff the
1627 * superblock is in version 2. Otherwise we fail to support any
1628 * extended sb features. */
1630 if (journal
->j_format_version
!= 2)
1633 if ((compat
& JBD2_KNOWN_COMPAT_FEATURES
) == compat
&&
1634 (ro
& JBD2_KNOWN_ROCOMPAT_FEATURES
) == ro
&&
1635 (incompat
& JBD2_KNOWN_INCOMPAT_FEATURES
) == incompat
)
1642 * int jbd2_journal_set_features () - Mark a given journal feature in the superblock
1643 * @journal: Journal to act on.
1644 * @compat: bitmask of compatible features
1645 * @ro: bitmask of features that force read-only mount
1646 * @incompat: bitmask of incompatible features
1648 * Mark a given journal feature as present on the
1649 * superblock. Returns true if the requested features could be set.
1653 int jbd2_journal_set_features (journal_t
*journal
, unsigned long compat
,
1654 unsigned long ro
, unsigned long incompat
)
1656 journal_superblock_t
*sb
;
1658 if (jbd2_journal_check_used_features(journal
, compat
, ro
, incompat
))
1661 if (!jbd2_journal_check_available_features(journal
, compat
, ro
, incompat
))
1664 jbd_debug(1, "Setting new features 0x%lx/0x%lx/0x%lx\n",
1665 compat
, ro
, incompat
);
1667 sb
= journal
->j_superblock
;
1669 sb
->s_feature_compat
|= cpu_to_be32(compat
);
1670 sb
->s_feature_ro_compat
|= cpu_to_be32(ro
);
1671 sb
->s_feature_incompat
|= cpu_to_be32(incompat
);
1677 * jbd2_journal_clear_features () - Clear a given journal feature in the
1679 * @journal: Journal to act on.
1680 * @compat: bitmask of compatible features
1681 * @ro: bitmask of features that force read-only mount
1682 * @incompat: bitmask of incompatible features
1684 * Clear a given journal feature as present on the
1687 void jbd2_journal_clear_features(journal_t
*journal
, unsigned long compat
,
1688 unsigned long ro
, unsigned long incompat
)
1690 journal_superblock_t
*sb
;
1692 jbd_debug(1, "Clear features 0x%lx/0x%lx/0x%lx\n",
1693 compat
, ro
, incompat
);
1695 sb
= journal
->j_superblock
;
1697 sb
->s_feature_compat
&= ~cpu_to_be32(compat
);
1698 sb
->s_feature_ro_compat
&= ~cpu_to_be32(ro
);
1699 sb
->s_feature_incompat
&= ~cpu_to_be32(incompat
);
1701 EXPORT_SYMBOL(jbd2_journal_clear_features
);
1704 * int jbd2_journal_flush () - Flush journal
1705 * @journal: Journal to act on.
1707 * Flush all data for a given journal to disk and empty the journal.
1708 * Filesystems can use this when remounting readonly to ensure that
1709 * recovery does not need to happen on remount.
1712 int jbd2_journal_flush(journal_t
*journal
)
1715 transaction_t
*transaction
= NULL
;
1717 write_lock(&journal
->j_state_lock
);
1719 /* Force everything buffered to the log... */
1720 if (journal
->j_running_transaction
) {
1721 transaction
= journal
->j_running_transaction
;
1722 __jbd2_log_start_commit(journal
, transaction
->t_tid
);
1723 } else if (journal
->j_committing_transaction
)
1724 transaction
= journal
->j_committing_transaction
;
1726 /* Wait for the log commit to complete... */
1728 tid_t tid
= transaction
->t_tid
;
1730 write_unlock(&journal
->j_state_lock
);
1731 jbd2_log_wait_commit(journal
, tid
);
1733 write_unlock(&journal
->j_state_lock
);
1736 /* ...and flush everything in the log out to disk. */
1737 spin_lock(&journal
->j_list_lock
);
1738 while (!err
&& journal
->j_checkpoint_transactions
!= NULL
) {
1739 spin_unlock(&journal
->j_list_lock
);
1740 mutex_lock(&journal
->j_checkpoint_mutex
);
1741 err
= jbd2_log_do_checkpoint(journal
);
1742 mutex_unlock(&journal
->j_checkpoint_mutex
);
1743 spin_lock(&journal
->j_list_lock
);
1745 spin_unlock(&journal
->j_list_lock
);
1747 if (is_journal_aborted(journal
))
1750 mutex_lock(&journal
->j_checkpoint_mutex
);
1751 jbd2_cleanup_journal_tail(journal
);
1753 /* Finally, mark the journal as really needing no recovery.
1754 * This sets s_start==0 in the underlying superblock, which is
1755 * the magic code for a fully-recovered superblock. Any future
1756 * commits of data to the journal will restore the current
1758 jbd2_mark_journal_empty(journal
);
1759 mutex_unlock(&journal
->j_checkpoint_mutex
);
1760 write_lock(&journal
->j_state_lock
);
1761 J_ASSERT(!journal
->j_running_transaction
);
1762 J_ASSERT(!journal
->j_committing_transaction
);
1763 J_ASSERT(!journal
->j_checkpoint_transactions
);
1764 J_ASSERT(journal
->j_head
== journal
->j_tail
);
1765 J_ASSERT(journal
->j_tail_sequence
== journal
->j_transaction_sequence
);
1766 write_unlock(&journal
->j_state_lock
);
1771 * int jbd2_journal_wipe() - Wipe journal contents
1772 * @journal: Journal to act on.
1773 * @write: flag (see below)
1775 * Wipe out all of the contents of a journal, safely. This will produce
1776 * a warning if the journal contains any valid recovery information.
1777 * Must be called between journal_init_*() and jbd2_journal_load().
1779 * If 'write' is non-zero, then we wipe out the journal on disk; otherwise
1780 * we merely suppress recovery.
1783 int jbd2_journal_wipe(journal_t
*journal
, int write
)
1787 J_ASSERT (!(journal
->j_flags
& JBD2_LOADED
));
1789 err
= load_superblock(journal
);
1793 if (!journal
->j_tail
)
1796 printk(KERN_WARNING
"JBD2: %s recovery information on journal\n",
1797 write
? "Clearing" : "Ignoring");
1799 err
= jbd2_journal_skip_recovery(journal
);
1801 /* Lock to make assertions happy... */
1802 mutex_lock(&journal
->j_checkpoint_mutex
);
1803 jbd2_mark_journal_empty(journal
);
1804 mutex_unlock(&journal
->j_checkpoint_mutex
);
1812 * Journal abort has very specific semantics, which we describe
1813 * for journal abort.
1815 * Two internal functions, which provide abort to the jbd layer
1820 * Quick version for internal journal use (doesn't lock the journal).
1821 * Aborts hard --- we mark the abort as occurred, but do _nothing_ else,
1822 * and don't attempt to make any other journal updates.
1824 void __jbd2_journal_abort_hard(journal_t
*journal
)
1826 transaction_t
*transaction
;
1828 if (journal
->j_flags
& JBD2_ABORT
)
1831 printk(KERN_ERR
"Aborting journal on device %s.\n",
1832 journal
->j_devname
);
1834 write_lock(&journal
->j_state_lock
);
1835 journal
->j_flags
|= JBD2_ABORT
;
1836 transaction
= journal
->j_running_transaction
;
1838 __jbd2_log_start_commit(journal
, transaction
->t_tid
);
1839 write_unlock(&journal
->j_state_lock
);
1842 /* Soft abort: record the abort error status in the journal superblock,
1843 * but don't do any other IO. */
1844 static void __journal_abort_soft (journal_t
*journal
, int errno
)
1846 if (journal
->j_flags
& JBD2_ABORT
)
1849 if (!journal
->j_errno
)
1850 journal
->j_errno
= errno
;
1852 __jbd2_journal_abort_hard(journal
);
1855 jbd2_journal_update_sb_errno(journal
);
1859 * void jbd2_journal_abort () - Shutdown the journal immediately.
1860 * @journal: the journal to shutdown.
1861 * @errno: an error number to record in the journal indicating
1862 * the reason for the shutdown.
1864 * Perform a complete, immediate shutdown of the ENTIRE
1865 * journal (not of a single transaction). This operation cannot be
1866 * undone without closing and reopening the journal.
1868 * The jbd2_journal_abort function is intended to support higher level error
1869 * recovery mechanisms such as the ext2/ext3 remount-readonly error
1872 * Journal abort has very specific semantics. Any existing dirty,
1873 * unjournaled buffers in the main filesystem will still be written to
1874 * disk by bdflush, but the journaling mechanism will be suspended
1875 * immediately and no further transaction commits will be honoured.
1877 * Any dirty, journaled buffers will be written back to disk without
1878 * hitting the journal. Atomicity cannot be guaranteed on an aborted
1879 * filesystem, but we _do_ attempt to leave as much data as possible
1880 * behind for fsck to use for cleanup.
1882 * Any attempt to get a new transaction handle on a journal which is in
1883 * ABORT state will just result in an -EROFS error return. A
1884 * jbd2_journal_stop on an existing handle will return -EIO if we have
1885 * entered abort state during the update.
1887 * Recursive transactions are not disturbed by journal abort until the
1888 * final jbd2_journal_stop, which will receive the -EIO error.
1890 * Finally, the jbd2_journal_abort call allows the caller to supply an errno
1891 * which will be recorded (if possible) in the journal superblock. This
1892 * allows a client to record failure conditions in the middle of a
1893 * transaction without having to complete the transaction to record the
1894 * failure to disk. ext3_error, for example, now uses this
1897 * Errors which originate from within the journaling layer will NOT
1898 * supply an errno; a null errno implies that absolutely no further
1899 * writes are done to the journal (unless there are any already in
1904 void jbd2_journal_abort(journal_t
*journal
, int errno
)
1906 __journal_abort_soft(journal
, errno
);
1910 * int jbd2_journal_errno () - returns the journal's error state.
1911 * @journal: journal to examine.
1913 * This is the errno number set with jbd2_journal_abort(), the last
1914 * time the journal was mounted - if the journal was stopped
1915 * without calling abort this will be 0.
1917 * If the journal has been aborted on this mount time -EROFS will
1920 int jbd2_journal_errno(journal_t
*journal
)
1924 read_lock(&journal
->j_state_lock
);
1925 if (journal
->j_flags
& JBD2_ABORT
)
1928 err
= journal
->j_errno
;
1929 read_unlock(&journal
->j_state_lock
);
1934 * int jbd2_journal_clear_err () - clears the journal's error state
1935 * @journal: journal to act on.
1937 * An error must be cleared or acked to take a FS out of readonly
1940 int jbd2_journal_clear_err(journal_t
*journal
)
1944 write_lock(&journal
->j_state_lock
);
1945 if (journal
->j_flags
& JBD2_ABORT
)
1948 journal
->j_errno
= 0;
1949 write_unlock(&journal
->j_state_lock
);
1954 * void jbd2_journal_ack_err() - Ack journal err.
1955 * @journal: journal to act on.
1957 * An error must be cleared or acked to take a FS out of readonly
1960 void jbd2_journal_ack_err(journal_t
*journal
)
1962 write_lock(&journal
->j_state_lock
);
1963 if (journal
->j_errno
)
1964 journal
->j_flags
|= JBD2_ACK_ERR
;
1965 write_unlock(&journal
->j_state_lock
);
1968 int jbd2_journal_blocks_per_page(struct inode
*inode
)
1970 return 1 << (PAGE_CACHE_SHIFT
- inode
->i_sb
->s_blocksize_bits
);
1974 * helper functions to deal with 32 or 64bit block numbers.
1976 size_t journal_tag_bytes(journal_t
*journal
)
1978 if (JBD2_HAS_INCOMPAT_FEATURE(journal
, JBD2_FEATURE_INCOMPAT_64BIT
))
1979 return JBD2_TAG_SIZE64
;
1981 return JBD2_TAG_SIZE32
;
1985 * JBD memory management
1987 * These functions are used to allocate block-sized chunks of memory
1988 * used for making copies of buffer_head data. Very often it will be
1989 * page-sized chunks of data, but sometimes it will be in
1990 * sub-page-size chunks. (For example, 16k pages on Power systems
1991 * with a 4k block file system.) For blocks smaller than a page, we
1992 * use a SLAB allocator. There are slab caches for each block size,
1993 * which are allocated at mount time, if necessary, and we only free
1994 * (all of) the slab caches when/if the jbd2 module is unloaded. For
1995 * this reason we don't need to a mutex to protect access to
1996 * jbd2_slab[] allocating or releasing memory; only in
1997 * jbd2_journal_create_slab().
1999 #define JBD2_MAX_SLABS 8
2000 static struct kmem_cache
*jbd2_slab
[JBD2_MAX_SLABS
];
2002 static const char *jbd2_slab_names
[JBD2_MAX_SLABS
] = {
2003 "jbd2_1k", "jbd2_2k", "jbd2_4k", "jbd2_8k",
2004 "jbd2_16k", "jbd2_32k", "jbd2_64k", "jbd2_128k"
2008 static void jbd2_journal_destroy_slabs(void)
2012 for (i
= 0; i
< JBD2_MAX_SLABS
; i
++) {
2014 kmem_cache_destroy(jbd2_slab
[i
]);
2015 jbd2_slab
[i
] = NULL
;
2019 static int jbd2_journal_create_slab(size_t size
)
2021 static DEFINE_MUTEX(jbd2_slab_create_mutex
);
2022 int i
= order_base_2(size
) - 10;
2025 if (size
== PAGE_SIZE
)
2028 if (i
>= JBD2_MAX_SLABS
)
2031 if (unlikely(i
< 0))
2033 mutex_lock(&jbd2_slab_create_mutex
);
2035 mutex_unlock(&jbd2_slab_create_mutex
);
2036 return 0; /* Already created */
2039 slab_size
= 1 << (i
+10);
2040 jbd2_slab
[i
] = kmem_cache_create(jbd2_slab_names
[i
], slab_size
,
2041 slab_size
, 0, NULL
);
2042 mutex_unlock(&jbd2_slab_create_mutex
);
2043 if (!jbd2_slab
[i
]) {
2044 printk(KERN_EMERG
"JBD2: no memory for jbd2_slab cache\n");
2050 static struct kmem_cache
*get_slab(size_t size
)
2052 int i
= order_base_2(size
) - 10;
2054 BUG_ON(i
>= JBD2_MAX_SLABS
);
2055 if (unlikely(i
< 0))
2057 BUG_ON(jbd2_slab
[i
] == NULL
);
2058 return jbd2_slab
[i
];
2061 void *jbd2_alloc(size_t size
, gfp_t flags
)
2065 BUG_ON(size
& (size
-1)); /* Must be a power of 2 */
2067 flags
|= __GFP_REPEAT
;
2068 if (size
== PAGE_SIZE
)
2069 ptr
= (void *)__get_free_pages(flags
, 0);
2070 else if (size
> PAGE_SIZE
) {
2071 int order
= get_order(size
);
2074 ptr
= (void *)__get_free_pages(flags
, order
);
2076 ptr
= vmalloc(size
);
2078 ptr
= kmem_cache_alloc(get_slab(size
), flags
);
2080 /* Check alignment; SLUB has gotten this wrong in the past,
2081 * and this can lead to user data corruption! */
2082 BUG_ON(((unsigned long) ptr
) & (size
-1));
2087 void jbd2_free(void *ptr
, size_t size
)
2089 if (size
== PAGE_SIZE
) {
2090 free_pages((unsigned long)ptr
, 0);
2093 if (size
> PAGE_SIZE
) {
2094 int order
= get_order(size
);
2097 free_pages((unsigned long)ptr
, order
);
2102 kmem_cache_free(get_slab(size
), ptr
);
2106 * Journal_head storage management
2108 static struct kmem_cache
*jbd2_journal_head_cache
;
2109 #ifdef CONFIG_JBD2_DEBUG
2110 static atomic_t nr_journal_heads
= ATOMIC_INIT(0);
2113 static int jbd2_journal_init_journal_head_cache(void)
2117 J_ASSERT(jbd2_journal_head_cache
== NULL
);
2118 jbd2_journal_head_cache
= kmem_cache_create("jbd2_journal_head",
2119 sizeof(struct journal_head
),
2121 SLAB_TEMPORARY
, /* flags */
2124 if (!jbd2_journal_head_cache
) {
2126 printk(KERN_EMERG
"JBD2: no memory for journal_head cache\n");
2131 static void jbd2_journal_destroy_journal_head_cache(void)
2133 if (jbd2_journal_head_cache
) {
2134 kmem_cache_destroy(jbd2_journal_head_cache
);
2135 jbd2_journal_head_cache
= NULL
;
2140 * journal_head splicing and dicing
2142 static struct journal_head
*journal_alloc_journal_head(void)
2144 struct journal_head
*ret
;
2146 #ifdef CONFIG_JBD2_DEBUG
2147 atomic_inc(&nr_journal_heads
);
2149 ret
= kmem_cache_alloc(jbd2_journal_head_cache
, GFP_NOFS
);
2151 jbd_debug(1, "out of memory for journal_head\n");
2152 pr_notice_ratelimited("ENOMEM in %s, retrying.\n", __func__
);
2155 ret
= kmem_cache_alloc(jbd2_journal_head_cache
, GFP_NOFS
);
2161 static void journal_free_journal_head(struct journal_head
*jh
)
2163 #ifdef CONFIG_JBD2_DEBUG
2164 atomic_dec(&nr_journal_heads
);
2165 memset(jh
, JBD2_POISON_FREE
, sizeof(*jh
));
2167 kmem_cache_free(jbd2_journal_head_cache
, jh
);
2171 * A journal_head is attached to a buffer_head whenever JBD has an
2172 * interest in the buffer.
2174 * Whenever a buffer has an attached journal_head, its ->b_state:BH_JBD bit
2175 * is set. This bit is tested in core kernel code where we need to take
2176 * JBD-specific actions. Testing the zeroness of ->b_private is not reliable
2179 * When a buffer has its BH_JBD bit set, its ->b_count is elevated by one.
2181 * When a buffer has its BH_JBD bit set it is immune from being released by
2182 * core kernel code, mainly via ->b_count.
2184 * A journal_head is detached from its buffer_head when the journal_head's
2185 * b_jcount reaches zero. Running transaction (b_transaction) and checkpoint
2186 * transaction (b_cp_transaction) hold their references to b_jcount.
2188 * Various places in the kernel want to attach a journal_head to a buffer_head
2189 * _before_ attaching the journal_head to a transaction. To protect the
2190 * journal_head in this situation, jbd2_journal_add_journal_head elevates the
2191 * journal_head's b_jcount refcount by one. The caller must call
2192 * jbd2_journal_put_journal_head() to undo this.
2194 * So the typical usage would be:
2196 * (Attach a journal_head if needed. Increments b_jcount)
2197 * struct journal_head *jh = jbd2_journal_add_journal_head(bh);
2199 * (Get another reference for transaction)
2200 * jbd2_journal_grab_journal_head(bh);
2201 * jh->b_transaction = xxx;
2202 * (Put original reference)
2203 * jbd2_journal_put_journal_head(jh);
2207 * Give a buffer_head a journal_head.
2211 struct journal_head
*jbd2_journal_add_journal_head(struct buffer_head
*bh
)
2213 struct journal_head
*jh
;
2214 struct journal_head
*new_jh
= NULL
;
2217 if (!buffer_jbd(bh
)) {
2218 new_jh
= journal_alloc_journal_head();
2219 memset(new_jh
, 0, sizeof(*new_jh
));
2222 jbd_lock_bh_journal_head(bh
);
2223 if (buffer_jbd(bh
)) {
2227 (atomic_read(&bh
->b_count
) > 0) ||
2228 (bh
->b_page
&& bh
->b_page
->mapping
));
2231 jbd_unlock_bh_journal_head(bh
);
2236 new_jh
= NULL
; /* We consumed it */
2241 BUFFER_TRACE(bh
, "added journal_head");
2244 jbd_unlock_bh_journal_head(bh
);
2246 journal_free_journal_head(new_jh
);
2247 return bh
->b_private
;
2251 * Grab a ref against this buffer_head's journal_head. If it ended up not
2252 * having a journal_head, return NULL
2254 struct journal_head
*jbd2_journal_grab_journal_head(struct buffer_head
*bh
)
2256 struct journal_head
*jh
= NULL
;
2258 jbd_lock_bh_journal_head(bh
);
2259 if (buffer_jbd(bh
)) {
2263 jbd_unlock_bh_journal_head(bh
);
2267 static void __journal_remove_journal_head(struct buffer_head
*bh
)
2269 struct journal_head
*jh
= bh2jh(bh
);
2271 J_ASSERT_JH(jh
, jh
->b_jcount
>= 0);
2272 J_ASSERT_JH(jh
, jh
->b_transaction
== NULL
);
2273 J_ASSERT_JH(jh
, jh
->b_next_transaction
== NULL
);
2274 J_ASSERT_JH(jh
, jh
->b_cp_transaction
== NULL
);
2275 J_ASSERT_JH(jh
, jh
->b_jlist
== BJ_None
);
2276 J_ASSERT_BH(bh
, buffer_jbd(bh
));
2277 J_ASSERT_BH(bh
, jh2bh(jh
) == bh
);
2278 BUFFER_TRACE(bh
, "remove journal_head");
2279 if (jh
->b_frozen_data
) {
2280 printk(KERN_WARNING
"%s: freeing b_frozen_data\n", __func__
);
2281 jbd2_free(jh
->b_frozen_data
, bh
->b_size
);
2283 if (jh
->b_committed_data
) {
2284 printk(KERN_WARNING
"%s: freeing b_committed_data\n", __func__
);
2285 jbd2_free(jh
->b_committed_data
, bh
->b_size
);
2287 bh
->b_private
= NULL
;
2288 jh
->b_bh
= NULL
; /* debug, really */
2289 clear_buffer_jbd(bh
);
2290 journal_free_journal_head(jh
);
2294 * Drop a reference on the passed journal_head. If it fell to zero then
2295 * release the journal_head from the buffer_head.
2297 void jbd2_journal_put_journal_head(struct journal_head
*jh
)
2299 struct buffer_head
*bh
= jh2bh(jh
);
2301 jbd_lock_bh_journal_head(bh
);
2302 J_ASSERT_JH(jh
, jh
->b_jcount
> 0);
2304 if (!jh
->b_jcount
) {
2305 __journal_remove_journal_head(bh
);
2306 jbd_unlock_bh_journal_head(bh
);
2309 jbd_unlock_bh_journal_head(bh
);
2313 * Initialize jbd inode head
2315 void jbd2_journal_init_jbd_inode(struct jbd2_inode
*jinode
, struct inode
*inode
)
2317 jinode
->i_transaction
= NULL
;
2318 jinode
->i_next_transaction
= NULL
;
2319 jinode
->i_vfs_inode
= inode
;
2320 jinode
->i_flags
= 0;
2321 INIT_LIST_HEAD(&jinode
->i_list
);
2325 * Function to be called before we start removing inode from memory (i.e.,
2326 * clear_inode() is a fine place to be called from). It removes inode from
2327 * transaction's lists.
2329 void jbd2_journal_release_jbd_inode(journal_t
*journal
,
2330 struct jbd2_inode
*jinode
)
2335 spin_lock(&journal
->j_list_lock
);
2336 /* Is commit writing out inode - we have to wait */
2337 if (test_bit(__JI_COMMIT_RUNNING
, &jinode
->i_flags
)) {
2338 wait_queue_head_t
*wq
;
2339 DEFINE_WAIT_BIT(wait
, &jinode
->i_flags
, __JI_COMMIT_RUNNING
);
2340 wq
= bit_waitqueue(&jinode
->i_flags
, __JI_COMMIT_RUNNING
);
2341 prepare_to_wait(wq
, &wait
.wait
, TASK_UNINTERRUPTIBLE
);
2342 spin_unlock(&journal
->j_list_lock
);
2344 finish_wait(wq
, &wait
.wait
);
2348 if (jinode
->i_transaction
) {
2349 list_del(&jinode
->i_list
);
2350 jinode
->i_transaction
= NULL
;
2352 spin_unlock(&journal
->j_list_lock
);
2358 #ifdef CONFIG_JBD2_DEBUG
2359 u8 jbd2_journal_enable_debug __read_mostly
;
2360 EXPORT_SYMBOL(jbd2_journal_enable_debug
);
2362 #define JBD2_DEBUG_NAME "jbd2-debug"
2364 static struct dentry
*jbd2_debugfs_dir
;
2365 static struct dentry
*jbd2_debug
;
2367 static void __init
jbd2_create_debugfs_entry(void)
2369 jbd2_debugfs_dir
= debugfs_create_dir("jbd2", NULL
);
2370 if (jbd2_debugfs_dir
)
2371 jbd2_debug
= debugfs_create_u8(JBD2_DEBUG_NAME
,
2374 &jbd2_journal_enable_debug
);
2377 static void __exit
jbd2_remove_debugfs_entry(void)
2379 debugfs_remove(jbd2_debug
);
2380 debugfs_remove(jbd2_debugfs_dir
);
2385 static void __init
jbd2_create_debugfs_entry(void)
2389 static void __exit
jbd2_remove_debugfs_entry(void)
2395 #ifdef CONFIG_PROC_FS
2397 #define JBD2_STATS_PROC_NAME "fs/jbd2"
2399 static void __init
jbd2_create_jbd_stats_proc_entry(void)
2401 proc_jbd2_stats
= proc_mkdir(JBD2_STATS_PROC_NAME
, NULL
);
2404 static void __exit
jbd2_remove_jbd_stats_proc_entry(void)
2406 if (proc_jbd2_stats
)
2407 remove_proc_entry(JBD2_STATS_PROC_NAME
, NULL
);
2412 #define jbd2_create_jbd_stats_proc_entry() do {} while (0)
2413 #define jbd2_remove_jbd_stats_proc_entry() do {} while (0)
2417 struct kmem_cache
*jbd2_handle_cache
, *jbd2_inode_cache
;
2419 static int __init
jbd2_journal_init_handle_cache(void)
2421 jbd2_handle_cache
= KMEM_CACHE(jbd2_journal_handle
, SLAB_TEMPORARY
);
2422 if (jbd2_handle_cache
== NULL
) {
2423 printk(KERN_EMERG
"JBD2: failed to create handle cache\n");
2426 jbd2_inode_cache
= KMEM_CACHE(jbd2_inode
, 0);
2427 if (jbd2_inode_cache
== NULL
) {
2428 printk(KERN_EMERG
"JBD2: failed to create inode cache\n");
2429 kmem_cache_destroy(jbd2_handle_cache
);
2435 static void jbd2_journal_destroy_handle_cache(void)
2437 if (jbd2_handle_cache
)
2438 kmem_cache_destroy(jbd2_handle_cache
);
2439 if (jbd2_inode_cache
)
2440 kmem_cache_destroy(jbd2_inode_cache
);
2445 * Module startup and shutdown
2448 static int __init
journal_init_caches(void)
2452 ret
= jbd2_journal_init_revoke_caches();
2454 ret
= jbd2_journal_init_journal_head_cache();
2456 ret
= jbd2_journal_init_handle_cache();
2458 ret
= jbd2_journal_init_transaction_cache();
2462 static void jbd2_journal_destroy_caches(void)
2464 jbd2_journal_destroy_revoke_caches();
2465 jbd2_journal_destroy_journal_head_cache();
2466 jbd2_journal_destroy_handle_cache();
2467 jbd2_journal_destroy_transaction_cache();
2468 jbd2_journal_destroy_slabs();
2471 static int __init
journal_init(void)
2475 BUILD_BUG_ON(sizeof(struct journal_superblock_s
) != 1024);
2477 ret
= journal_init_caches();
2479 jbd2_create_debugfs_entry();
2480 jbd2_create_jbd_stats_proc_entry();
2482 jbd2_journal_destroy_caches();
2487 static void __exit
journal_exit(void)
2489 #ifdef CONFIG_JBD2_DEBUG
2490 int n
= atomic_read(&nr_journal_heads
);
2492 printk(KERN_EMERG
"JBD2: leaked %d journal_heads!\n", n
);
2494 jbd2_remove_debugfs_entry();
2495 jbd2_remove_jbd_stats_proc_entry();
2496 jbd2_journal_destroy_caches();
2499 MODULE_LICENSE("GPL");
2500 module_init(journal_init
);
2501 module_exit(journal_exit
);