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>
53 #include <asm/system.h>
55 EXPORT_SYMBOL(jbd2_journal_extend
);
56 EXPORT_SYMBOL(jbd2_journal_stop
);
57 EXPORT_SYMBOL(jbd2_journal_lock_updates
);
58 EXPORT_SYMBOL(jbd2_journal_unlock_updates
);
59 EXPORT_SYMBOL(jbd2_journal_get_write_access
);
60 EXPORT_SYMBOL(jbd2_journal_get_create_access
);
61 EXPORT_SYMBOL(jbd2_journal_get_undo_access
);
62 EXPORT_SYMBOL(jbd2_journal_set_triggers
);
63 EXPORT_SYMBOL(jbd2_journal_dirty_metadata
);
64 EXPORT_SYMBOL(jbd2_journal_release_buffer
);
65 EXPORT_SYMBOL(jbd2_journal_forget
);
67 EXPORT_SYMBOL(journal_sync_buffer
);
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_file_inode
);
93 EXPORT_SYMBOL(jbd2_journal_init_jbd_inode
);
94 EXPORT_SYMBOL(jbd2_journal_release_jbd_inode
);
95 EXPORT_SYMBOL(jbd2_journal_begin_ordered_truncate
);
96 EXPORT_SYMBOL(jbd2_inode_cache
);
98 static void __journal_abort_soft (journal_t
*journal
, int errno
);
99 static int jbd2_journal_create_slab(size_t slab_size
);
102 * Helper function used to manage commit timeouts
105 static void commit_timeout(unsigned long __data
)
107 struct task_struct
* p
= (struct task_struct
*) __data
;
113 * kjournald2: The main thread function used to manage a logging device
116 * This kernel thread is responsible for two things:
118 * 1) COMMIT: Every so often we need to commit the current state of the
119 * filesystem to disk. The journal thread is responsible for writing
120 * all of the metadata buffers to disk.
122 * 2) CHECKPOINT: We cannot reuse a used section of the log file until all
123 * of the data in that part of the log has been rewritten elsewhere on
124 * the disk. Flushing these old buffers to reclaim space in the log is
125 * known as checkpointing, and this thread is responsible for that job.
128 static int kjournald2(void *arg
)
130 journal_t
*journal
= arg
;
131 transaction_t
*transaction
;
134 * Set up an interval timer which can be used to trigger a commit wakeup
135 * after the commit interval expires
137 setup_timer(&journal
->j_commit_timer
, commit_timeout
,
138 (unsigned long)current
);
140 /* Record that the journal thread is running */
141 journal
->j_task
= current
;
142 wake_up(&journal
->j_wait_done_commit
);
145 * And now, wait forever for commit wakeup events.
147 write_lock(&journal
->j_state_lock
);
150 if (journal
->j_flags
& JBD2_UNMOUNT
)
153 jbd_debug(1, "commit_sequence=%d, commit_request=%d\n",
154 journal
->j_commit_sequence
, journal
->j_commit_request
);
156 if (journal
->j_commit_sequence
!= journal
->j_commit_request
) {
157 jbd_debug(1, "OK, requests differ\n");
158 write_unlock(&journal
->j_state_lock
);
159 del_timer_sync(&journal
->j_commit_timer
);
160 jbd2_journal_commit_transaction(journal
);
161 write_lock(&journal
->j_state_lock
);
165 wake_up(&journal
->j_wait_done_commit
);
166 if (freezing(current
)) {
168 * The simpler the better. Flushing journal isn't a
169 * good idea, because that depends on threads that may
170 * be already stopped.
172 jbd_debug(1, "Now suspending kjournald2\n");
173 write_unlock(&journal
->j_state_lock
);
175 write_lock(&journal
->j_state_lock
);
178 * We assume on resume that commits are already there,
182 int should_sleep
= 1;
184 prepare_to_wait(&journal
->j_wait_commit
, &wait
,
186 if (journal
->j_commit_sequence
!= journal
->j_commit_request
)
188 transaction
= journal
->j_running_transaction
;
189 if (transaction
&& time_after_eq(jiffies
,
190 transaction
->t_expires
))
192 if (journal
->j_flags
& JBD2_UNMOUNT
)
195 write_unlock(&journal
->j_state_lock
);
197 write_lock(&journal
->j_state_lock
);
199 finish_wait(&journal
->j_wait_commit
, &wait
);
202 jbd_debug(1, "kjournald2 wakes\n");
205 * Were we woken up by a commit wakeup event?
207 transaction
= journal
->j_running_transaction
;
208 if (transaction
&& time_after_eq(jiffies
, transaction
->t_expires
)) {
209 journal
->j_commit_request
= transaction
->t_tid
;
210 jbd_debug(1, "woke because of timeout\n");
215 write_unlock(&journal
->j_state_lock
);
216 del_timer_sync(&journal
->j_commit_timer
);
217 journal
->j_task
= NULL
;
218 wake_up(&journal
->j_wait_done_commit
);
219 jbd_debug(1, "Journal thread exiting.\n");
223 static int jbd2_journal_start_thread(journal_t
*journal
)
225 struct task_struct
*t
;
227 t
= kthread_run(kjournald2
, journal
, "jbd2/%s",
232 wait_event(journal
->j_wait_done_commit
, journal
->j_task
!= NULL
);
236 static void journal_kill_thread(journal_t
*journal
)
238 write_lock(&journal
->j_state_lock
);
239 journal
->j_flags
|= JBD2_UNMOUNT
;
241 while (journal
->j_task
) {
242 wake_up(&journal
->j_wait_commit
);
243 write_unlock(&journal
->j_state_lock
);
244 wait_event(journal
->j_wait_done_commit
, journal
->j_task
== NULL
);
245 write_lock(&journal
->j_state_lock
);
247 write_unlock(&journal
->j_state_lock
);
251 * jbd2_journal_write_metadata_buffer: write a metadata buffer to the journal.
253 * Writes a metadata buffer to a given disk block. The actual IO is not
254 * performed but a new buffer_head is constructed which labels the data
255 * to be written with the correct destination disk block.
257 * Any magic-number escaping which needs to be done will cause a
258 * copy-out here. If the buffer happens to start with the
259 * JBD2_MAGIC_NUMBER, then we can't write it to the log directly: the
260 * magic number is only written to the log for descripter blocks. In
261 * this case, we copy the data and replace the first word with 0, and we
262 * return a result code which indicates that this buffer needs to be
263 * marked as an escaped buffer in the corresponding log descriptor
264 * block. The missing word can then be restored when the block is read
267 * If the source buffer has already been modified by a new transaction
268 * since we took the last commit snapshot, we use the frozen copy of
269 * that data for IO. If we end up using the existing buffer_head's data
270 * for the write, then we *have* to lock the buffer to prevent anyone
271 * else from using and possibly modifying it while the IO is in
274 * The function returns a pointer to the buffer_heads to be used for IO.
276 * We assume that the journal has already been locked in this function.
283 * Bit 0 set == escape performed on the data
284 * Bit 1 set == buffer copy-out performed (kfree the data after IO)
287 int jbd2_journal_write_metadata_buffer(transaction_t
*transaction
,
288 struct journal_head
*jh_in
,
289 struct journal_head
**jh_out
,
290 unsigned long long blocknr
)
292 int need_copy_out
= 0;
293 int done_copy_out
= 0;
296 struct buffer_head
*new_bh
;
297 struct journal_head
*new_jh
;
298 struct page
*new_page
;
299 unsigned int new_offset
;
300 struct buffer_head
*bh_in
= jh2bh(jh_in
);
301 journal_t
*journal
= transaction
->t_journal
;
304 * The buffer really shouldn't be locked: only the current committing
305 * transaction is allowed to write it, so nobody else is allowed
308 * akpm: except if we're journalling data, and write() output is
309 * also part of a shared mapping, and another thread has
310 * decided to launch a writepage() against this buffer.
312 J_ASSERT_BH(bh_in
, buffer_jbddirty(bh_in
));
315 new_bh
= alloc_buffer_head(GFP_NOFS
);
318 * Failure is not an option, but __GFP_NOFAIL is going
319 * away; so we retry ourselves here.
321 congestion_wait(BLK_RW_ASYNC
, HZ
/50);
325 /* keep subsequent assertions sane */
327 init_buffer(new_bh
, NULL
, NULL
);
328 atomic_set(&new_bh
->b_count
, 1);
329 new_jh
= jbd2_journal_add_journal_head(new_bh
); /* This sleeps */
332 * If a new transaction has already done a buffer copy-out, then
333 * we use that version of the data for the commit.
335 jbd_lock_bh_state(bh_in
);
337 if (jh_in
->b_frozen_data
) {
339 new_page
= virt_to_page(jh_in
->b_frozen_data
);
340 new_offset
= offset_in_page(jh_in
->b_frozen_data
);
342 new_page
= jh2bh(jh_in
)->b_page
;
343 new_offset
= offset_in_page(jh2bh(jh_in
)->b_data
);
346 mapped_data
= kmap_atomic(new_page
, KM_USER0
);
348 * Fire data frozen trigger if data already wasn't frozen. Do this
349 * before checking for escaping, as the trigger may modify the magic
350 * offset. If a copy-out happens afterwards, it will have the correct
351 * data in the buffer.
354 jbd2_buffer_frozen_trigger(jh_in
, mapped_data
+ new_offset
,
360 if (*((__be32
*)(mapped_data
+ new_offset
)) ==
361 cpu_to_be32(JBD2_MAGIC_NUMBER
)) {
365 kunmap_atomic(mapped_data
, KM_USER0
);
368 * Do we need to do a data copy?
370 if (need_copy_out
&& !done_copy_out
) {
373 jbd_unlock_bh_state(bh_in
);
374 tmp
= jbd2_alloc(bh_in
->b_size
, GFP_NOFS
);
376 jbd2_journal_put_journal_head(new_jh
);
379 jbd_lock_bh_state(bh_in
);
380 if (jh_in
->b_frozen_data
) {
381 jbd2_free(tmp
, bh_in
->b_size
);
385 jh_in
->b_frozen_data
= tmp
;
386 mapped_data
= kmap_atomic(new_page
, KM_USER0
);
387 memcpy(tmp
, mapped_data
+ new_offset
, jh2bh(jh_in
)->b_size
);
388 kunmap_atomic(mapped_data
, KM_USER0
);
390 new_page
= virt_to_page(tmp
);
391 new_offset
= offset_in_page(tmp
);
395 * This isn't strictly necessary, as we're using frozen
396 * data for the escaping, but it keeps consistency with
397 * b_frozen_data usage.
399 jh_in
->b_frozen_triggers
= jh_in
->b_triggers
;
403 * Did we need to do an escaping? Now we've done all the
404 * copying, we can finally do so.
407 mapped_data
= kmap_atomic(new_page
, KM_USER0
);
408 *((unsigned int *)(mapped_data
+ new_offset
)) = 0;
409 kunmap_atomic(mapped_data
, KM_USER0
);
412 set_bh_page(new_bh
, new_page
, new_offset
);
413 new_jh
->b_transaction
= NULL
;
414 new_bh
->b_size
= jh2bh(jh_in
)->b_size
;
415 new_bh
->b_bdev
= transaction
->t_journal
->j_dev
;
416 new_bh
->b_blocknr
= blocknr
;
417 set_buffer_mapped(new_bh
);
418 set_buffer_dirty(new_bh
);
423 * The to-be-written buffer needs to get moved to the io queue,
424 * and the original buffer whose contents we are shadowing or
425 * copying is moved to the transaction's shadow queue.
427 JBUFFER_TRACE(jh_in
, "file as BJ_Shadow");
428 spin_lock(&journal
->j_list_lock
);
429 __jbd2_journal_file_buffer(jh_in
, transaction
, BJ_Shadow
);
430 spin_unlock(&journal
->j_list_lock
);
431 jbd_unlock_bh_state(bh_in
);
433 JBUFFER_TRACE(new_jh
, "file as BJ_IO");
434 jbd2_journal_file_buffer(new_jh
, transaction
, BJ_IO
);
436 return do_escape
| (done_copy_out
<< 1);
440 * Allocation code for the journal file. Manage the space left in the
441 * journal, so that we can begin checkpointing when appropriate.
445 * __jbd2_log_space_left: Return the number of free blocks left in the journal.
447 * Called with the journal already locked.
449 * Called under j_state_lock
452 int __jbd2_log_space_left(journal_t
*journal
)
454 int left
= journal
->j_free
;
456 /* assert_spin_locked(&journal->j_state_lock); */
459 * Be pessimistic here about the number of those free blocks which
460 * might be required for log descriptor control blocks.
463 #define MIN_LOG_RESERVED_BLOCKS 32 /* Allow for rounding errors */
465 left
-= MIN_LOG_RESERVED_BLOCKS
;
474 * Called with j_state_lock locked for writing.
475 * Returns true if a transaction commit was started.
477 int __jbd2_log_start_commit(journal_t
*journal
, tid_t target
)
480 * The only transaction we can possibly wait upon is the
481 * currently running transaction (if it exists). Otherwise,
482 * the target tid must be an old one.
484 if (journal
->j_running_transaction
&&
485 journal
->j_running_transaction
->t_tid
== target
) {
487 * We want a new commit: OK, mark the request and wakeup the
488 * commit thread. We do _not_ do the commit ourselves.
491 journal
->j_commit_request
= target
;
492 jbd_debug(1, "JBD2: requesting commit %d/%d\n",
493 journal
->j_commit_request
,
494 journal
->j_commit_sequence
);
495 wake_up(&journal
->j_wait_commit
);
497 } else if (!tid_geq(journal
->j_commit_request
, target
))
498 /* This should never happen, but if it does, preserve
499 the evidence before kjournald goes into a loop and
500 increments j_commit_sequence beyond all recognition. */
501 WARN_ONCE(1, "JBD2: bad log_start_commit: %u %u %u %u\n",
502 journal
->j_commit_request
,
503 journal
->j_commit_sequence
,
504 target
, journal
->j_running_transaction
?
505 journal
->j_running_transaction
->t_tid
: 0);
509 int jbd2_log_start_commit(journal_t
*journal
, tid_t tid
)
513 write_lock(&journal
->j_state_lock
);
514 ret
= __jbd2_log_start_commit(journal
, tid
);
515 write_unlock(&journal
->j_state_lock
);
520 * Force and wait upon a commit if the calling process is not within
521 * transaction. This is used for forcing out undo-protected data which contains
522 * bitmaps, when the fs is running out of space.
524 * We can only force the running transaction if we don't have an active handle;
525 * otherwise, we will deadlock.
527 * Returns true if a transaction was started.
529 int jbd2_journal_force_commit_nested(journal_t
*journal
)
531 transaction_t
*transaction
= NULL
;
533 int need_to_start
= 0;
535 read_lock(&journal
->j_state_lock
);
536 if (journal
->j_running_transaction
&& !current
->journal_info
) {
537 transaction
= journal
->j_running_transaction
;
538 if (!tid_geq(journal
->j_commit_request
, transaction
->t_tid
))
540 } else if (journal
->j_committing_transaction
)
541 transaction
= journal
->j_committing_transaction
;
544 read_unlock(&journal
->j_state_lock
);
545 return 0; /* Nothing to retry */
548 tid
= transaction
->t_tid
;
549 read_unlock(&journal
->j_state_lock
);
551 jbd2_log_start_commit(journal
, tid
);
552 jbd2_log_wait_commit(journal
, tid
);
557 * Start a commit of the current running transaction (if any). Returns true
558 * if a transaction is going to be committed (or is currently already
559 * committing), and fills its tid in at *ptid
561 int jbd2_journal_start_commit(journal_t
*journal
, tid_t
*ptid
)
565 write_lock(&journal
->j_state_lock
);
566 if (journal
->j_running_transaction
) {
567 tid_t tid
= journal
->j_running_transaction
->t_tid
;
569 __jbd2_log_start_commit(journal
, tid
);
570 /* There's a running transaction and we've just made sure
571 * it's commit has been scheduled. */
575 } else if (journal
->j_committing_transaction
) {
577 * If ext3_write_super() recently started a commit, then we
578 * have to wait for completion of that transaction
581 *ptid
= journal
->j_committing_transaction
->t_tid
;
584 write_unlock(&journal
->j_state_lock
);
589 * Return 1 if a given transaction has not yet sent barrier request
590 * connected with a transaction commit. If 0 is returned, transaction
591 * may or may not have sent the barrier. Used to avoid sending barrier
592 * twice in common cases.
594 int jbd2_trans_will_send_data_barrier(journal_t
*journal
, tid_t tid
)
597 transaction_t
*commit_trans
;
599 if (!(journal
->j_flags
& JBD2_BARRIER
))
601 read_lock(&journal
->j_state_lock
);
602 /* Transaction already committed? */
603 if (tid_geq(journal
->j_commit_sequence
, tid
))
605 commit_trans
= journal
->j_committing_transaction
;
606 if (!commit_trans
|| commit_trans
->t_tid
!= tid
) {
611 * Transaction is being committed and we already proceeded to
612 * submitting a flush to fs partition?
614 if (journal
->j_fs_dev
!= journal
->j_dev
) {
615 if (!commit_trans
->t_need_data_flush
||
616 commit_trans
->t_state
>= T_COMMIT_DFLUSH
)
619 if (commit_trans
->t_state
>= T_COMMIT_JFLUSH
)
624 read_unlock(&journal
->j_state_lock
);
627 EXPORT_SYMBOL(jbd2_trans_will_send_data_barrier
);
630 * Wait for a specified commit to complete.
631 * The caller may not hold the journal lock.
633 int jbd2_log_wait_commit(journal_t
*journal
, tid_t tid
)
637 read_lock(&journal
->j_state_lock
);
638 #ifdef CONFIG_JBD2_DEBUG
639 if (!tid_geq(journal
->j_commit_request
, tid
)) {
641 "%s: error: j_commit_request=%d, tid=%d\n",
642 __func__
, journal
->j_commit_request
, tid
);
645 while (tid_gt(tid
, journal
->j_commit_sequence
)) {
646 jbd_debug(1, "JBD2: want %d, j_commit_sequence=%d\n",
647 tid
, journal
->j_commit_sequence
);
648 wake_up(&journal
->j_wait_commit
);
649 read_unlock(&journal
->j_state_lock
);
650 wait_event(journal
->j_wait_done_commit
,
651 !tid_gt(tid
, journal
->j_commit_sequence
));
652 read_lock(&journal
->j_state_lock
);
654 read_unlock(&journal
->j_state_lock
);
656 if (unlikely(is_journal_aborted(journal
))) {
657 printk(KERN_EMERG
"journal commit I/O error\n");
664 * Log buffer allocation routines:
667 int jbd2_journal_next_log_block(journal_t
*journal
, unsigned long long *retp
)
669 unsigned long blocknr
;
671 write_lock(&journal
->j_state_lock
);
672 J_ASSERT(journal
->j_free
> 1);
674 blocknr
= journal
->j_head
;
677 if (journal
->j_head
== journal
->j_last
)
678 journal
->j_head
= journal
->j_first
;
679 write_unlock(&journal
->j_state_lock
);
680 return jbd2_journal_bmap(journal
, blocknr
, retp
);
684 * Conversion of logical to physical block numbers for the journal
686 * On external journals the journal blocks are identity-mapped, so
687 * this is a no-op. If needed, we can use j_blk_offset - everything is
690 int jbd2_journal_bmap(journal_t
*journal
, unsigned long blocknr
,
691 unsigned long long *retp
)
694 unsigned long long ret
;
696 if (journal
->j_inode
) {
697 ret
= bmap(journal
->j_inode
, blocknr
);
701 printk(KERN_ALERT
"%s: journal block not found "
702 "at offset %lu on %s\n",
703 __func__
, blocknr
, journal
->j_devname
);
705 __journal_abort_soft(journal
, err
);
708 *retp
= blocknr
; /* +journal->j_blk_offset */
714 * We play buffer_head aliasing tricks to write data/metadata blocks to
715 * the journal without copying their contents, but for journal
716 * descriptor blocks we do need to generate bona fide buffers.
718 * After the caller of jbd2_journal_get_descriptor_buffer() has finished modifying
719 * the buffer's contents they really should run flush_dcache_page(bh->b_page).
720 * But we don't bother doing that, so there will be coherency problems with
721 * mmaps of blockdevs which hold live JBD-controlled filesystems.
723 struct journal_head
*jbd2_journal_get_descriptor_buffer(journal_t
*journal
)
725 struct buffer_head
*bh
;
726 unsigned long long blocknr
;
729 err
= jbd2_journal_next_log_block(journal
, &blocknr
);
734 bh
= __getblk(journal
->j_dev
, blocknr
, journal
->j_blocksize
);
738 memset(bh
->b_data
, 0, journal
->j_blocksize
);
739 set_buffer_uptodate(bh
);
741 BUFFER_TRACE(bh
, "return this buffer");
742 return jbd2_journal_add_journal_head(bh
);
746 * Return tid of the oldest transaction in the journal and block in the journal
747 * where the transaction starts.
749 * If the journal is now empty, return which will be the next transaction ID
750 * we will write and where will that transaction start.
752 * The return value is 0 if journal tail cannot be pushed any further, 1 if
755 int jbd2_journal_get_log_tail(journal_t
*journal
, tid_t
*tid
,
756 unsigned long *block
)
758 transaction_t
*transaction
;
761 read_lock(&journal
->j_state_lock
);
762 spin_lock(&journal
->j_list_lock
);
763 transaction
= journal
->j_checkpoint_transactions
;
765 *tid
= transaction
->t_tid
;
766 *block
= transaction
->t_log_start
;
767 } else if ((transaction
= journal
->j_committing_transaction
) != NULL
) {
768 *tid
= transaction
->t_tid
;
769 *block
= transaction
->t_log_start
;
770 } else if ((transaction
= journal
->j_running_transaction
) != NULL
) {
771 *tid
= transaction
->t_tid
;
772 *block
= journal
->j_head
;
774 *tid
= journal
->j_transaction_sequence
;
775 *block
= journal
->j_head
;
777 ret
= tid_gt(*tid
, journal
->j_tail_sequence
);
778 spin_unlock(&journal
->j_list_lock
);
779 read_unlock(&journal
->j_state_lock
);
785 * Update information in journal structure and in on disk journal superblock
786 * about log tail. This function does not check whether information passed in
787 * really pushes log tail further. It's responsibility of the caller to make
788 * sure provided log tail information is valid (e.g. by holding
789 * j_checkpoint_mutex all the time between computing log tail and calling this
790 * function as is the case with jbd2_cleanup_journal_tail()).
792 * Requires j_checkpoint_mutex
794 void __jbd2_update_log_tail(journal_t
*journal
, tid_t tid
, unsigned long block
)
798 BUG_ON(!mutex_is_locked(&journal
->j_checkpoint_mutex
));
801 * We cannot afford for write to remain in drive's caches since as
802 * soon as we update j_tail, next transaction can start reusing journal
803 * space and if we lose sb update during power failure we'd replay
804 * old transaction with possibly newly overwritten data.
806 jbd2_journal_update_sb_log_tail(journal
, tid
, block
, WRITE_FUA
);
807 write_lock(&journal
->j_state_lock
);
808 freed
= block
- journal
->j_tail
;
809 if (block
< journal
->j_tail
)
810 freed
+= journal
->j_last
- journal
->j_first
;
812 trace_jbd2_update_log_tail(journal
, tid
, block
, freed
);
814 "Cleaning journal tail from %d to %d (offset %lu), "
816 journal
->j_tail_sequence
, tid
, block
, freed
);
818 journal
->j_free
+= freed
;
819 journal
->j_tail_sequence
= tid
;
820 journal
->j_tail
= block
;
821 write_unlock(&journal
->j_state_lock
);
824 struct jbd2_stats_proc_session
{
826 struct transaction_stats_s
*stats
;
831 static void *jbd2_seq_info_start(struct seq_file
*seq
, loff_t
*pos
)
833 return *pos
? NULL
: SEQ_START_TOKEN
;
836 static void *jbd2_seq_info_next(struct seq_file
*seq
, void *v
, loff_t
*pos
)
841 static int jbd2_seq_info_show(struct seq_file
*seq
, void *v
)
843 struct jbd2_stats_proc_session
*s
= seq
->private;
845 if (v
!= SEQ_START_TOKEN
)
847 seq_printf(seq
, "%lu transaction, each up to %u blocks\n",
849 s
->journal
->j_max_transaction_buffers
);
850 if (s
->stats
->ts_tid
== 0)
852 seq_printf(seq
, "average: \n %ums waiting for transaction\n",
853 jiffies_to_msecs(s
->stats
->run
.rs_wait
/ s
->stats
->ts_tid
));
854 seq_printf(seq
, " %ums running transaction\n",
855 jiffies_to_msecs(s
->stats
->run
.rs_running
/ s
->stats
->ts_tid
));
856 seq_printf(seq
, " %ums transaction was being locked\n",
857 jiffies_to_msecs(s
->stats
->run
.rs_locked
/ s
->stats
->ts_tid
));
858 seq_printf(seq
, " %ums flushing data (in ordered mode)\n",
859 jiffies_to_msecs(s
->stats
->run
.rs_flushing
/ s
->stats
->ts_tid
));
860 seq_printf(seq
, " %ums logging transaction\n",
861 jiffies_to_msecs(s
->stats
->run
.rs_logging
/ s
->stats
->ts_tid
));
862 seq_printf(seq
, " %lluus average transaction commit time\n",
863 div_u64(s
->journal
->j_average_commit_time
, 1000));
864 seq_printf(seq
, " %lu handles per transaction\n",
865 s
->stats
->run
.rs_handle_count
/ s
->stats
->ts_tid
);
866 seq_printf(seq
, " %lu blocks per transaction\n",
867 s
->stats
->run
.rs_blocks
/ s
->stats
->ts_tid
);
868 seq_printf(seq
, " %lu logged blocks per transaction\n",
869 s
->stats
->run
.rs_blocks_logged
/ s
->stats
->ts_tid
);
873 static void jbd2_seq_info_stop(struct seq_file
*seq
, void *v
)
877 static const struct seq_operations jbd2_seq_info_ops
= {
878 .start
= jbd2_seq_info_start
,
879 .next
= jbd2_seq_info_next
,
880 .stop
= jbd2_seq_info_stop
,
881 .show
= jbd2_seq_info_show
,
884 static int jbd2_seq_info_open(struct inode
*inode
, struct file
*file
)
886 journal_t
*journal
= PDE(inode
)->data
;
887 struct jbd2_stats_proc_session
*s
;
890 s
= kmalloc(sizeof(*s
), GFP_KERNEL
);
893 size
= sizeof(struct transaction_stats_s
);
894 s
->stats
= kmalloc(size
, GFP_KERNEL
);
895 if (s
->stats
== NULL
) {
899 spin_lock(&journal
->j_history_lock
);
900 memcpy(s
->stats
, &journal
->j_stats
, size
);
901 s
->journal
= journal
;
902 spin_unlock(&journal
->j_history_lock
);
904 rc
= seq_open(file
, &jbd2_seq_info_ops
);
906 struct seq_file
*m
= file
->private_data
;
916 static int jbd2_seq_info_release(struct inode
*inode
, struct file
*file
)
918 struct seq_file
*seq
= file
->private_data
;
919 struct jbd2_stats_proc_session
*s
= seq
->private;
922 return seq_release(inode
, file
);
925 static const struct file_operations jbd2_seq_info_fops
= {
926 .owner
= THIS_MODULE
,
927 .open
= jbd2_seq_info_open
,
930 .release
= jbd2_seq_info_release
,
933 static struct proc_dir_entry
*proc_jbd2_stats
;
935 static void jbd2_stats_proc_init(journal_t
*journal
)
937 journal
->j_proc_entry
= proc_mkdir(journal
->j_devname
, proc_jbd2_stats
);
938 if (journal
->j_proc_entry
) {
939 proc_create_data("info", S_IRUGO
, journal
->j_proc_entry
,
940 &jbd2_seq_info_fops
, journal
);
944 static void jbd2_stats_proc_exit(journal_t
*journal
)
946 remove_proc_entry("info", journal
->j_proc_entry
);
947 remove_proc_entry(journal
->j_devname
, proc_jbd2_stats
);
951 * Management for journal control blocks: functions to create and
952 * destroy journal_t structures, and to initialise and read existing
953 * journal blocks from disk. */
955 /* First: create and setup a journal_t object in memory. We initialise
956 * very few fields yet: that has to wait until we have created the
957 * journal structures from from scratch, or loaded them from disk. */
959 static journal_t
* journal_init_common (void)
964 journal
= kzalloc(sizeof(*journal
), GFP_KERNEL
);
968 init_waitqueue_head(&journal
->j_wait_transaction_locked
);
969 init_waitqueue_head(&journal
->j_wait_logspace
);
970 init_waitqueue_head(&journal
->j_wait_done_commit
);
971 init_waitqueue_head(&journal
->j_wait_checkpoint
);
972 init_waitqueue_head(&journal
->j_wait_commit
);
973 init_waitqueue_head(&journal
->j_wait_updates
);
974 mutex_init(&journal
->j_barrier
);
975 mutex_init(&journal
->j_checkpoint_mutex
);
976 spin_lock_init(&journal
->j_revoke_lock
);
977 spin_lock_init(&journal
->j_list_lock
);
978 rwlock_init(&journal
->j_state_lock
);
980 journal
->j_commit_interval
= (HZ
* JBD2_DEFAULT_MAX_COMMIT_AGE
);
981 journal
->j_min_batch_time
= 0;
982 journal
->j_max_batch_time
= 15000; /* 15ms */
984 /* The journal is marked for error until we succeed with recovery! */
985 journal
->j_flags
= JBD2_ABORT
;
987 /* Set up a default-sized revoke table for the new mount. */
988 err
= jbd2_journal_init_revoke(journal
, JOURNAL_REVOKE_DEFAULT_HASH
);
994 spin_lock_init(&journal
->j_history_lock
);
999 /* jbd2_journal_init_dev and jbd2_journal_init_inode:
1001 * Create a journal structure assigned some fixed set of disk blocks to
1002 * the journal. We don't actually touch those disk blocks yet, but we
1003 * need to set up all of the mapping information to tell the journaling
1004 * system where the journal blocks are.
1009 * journal_t * jbd2_journal_init_dev() - creates and initialises a journal structure
1010 * @bdev: Block device on which to create the journal
1011 * @fs_dev: Device which hold journalled filesystem for this journal.
1012 * @start: Block nr Start of journal.
1013 * @len: Length of the journal in blocks.
1014 * @blocksize: blocksize of journalling device
1016 * Returns: a newly created journal_t *
1018 * jbd2_journal_init_dev creates a journal which maps a fixed contiguous
1019 * range of blocks on an arbitrary block device.
1022 journal_t
* jbd2_journal_init_dev(struct block_device
*bdev
,
1023 struct block_device
*fs_dev
,
1024 unsigned long long start
, int len
, int blocksize
)
1026 journal_t
*journal
= journal_init_common();
1027 struct buffer_head
*bh
;
1034 /* journal descriptor can store up to n blocks -bzzz */
1035 journal
->j_blocksize
= blocksize
;
1036 journal
->j_dev
= bdev
;
1037 journal
->j_fs_dev
= fs_dev
;
1038 journal
->j_blk_offset
= start
;
1039 journal
->j_maxlen
= len
;
1040 bdevname(journal
->j_dev
, journal
->j_devname
);
1041 p
= journal
->j_devname
;
1042 while ((p
= strchr(p
, '/')))
1044 jbd2_stats_proc_init(journal
);
1045 n
= journal
->j_blocksize
/ sizeof(journal_block_tag_t
);
1046 journal
->j_wbufsize
= n
;
1047 journal
->j_wbuf
= kmalloc(n
* sizeof(struct buffer_head
*), GFP_KERNEL
);
1048 if (!journal
->j_wbuf
) {
1049 printk(KERN_ERR
"%s: Can't allocate bhs for commit thread\n",
1054 bh
= __getblk(journal
->j_dev
, start
, journal
->j_blocksize
);
1057 "%s: Cannot get buffer for journal superblock\n",
1061 journal
->j_sb_buffer
= bh
;
1062 journal
->j_superblock
= (journal_superblock_t
*)bh
->b_data
;
1066 kfree(journal
->j_wbuf
);
1067 jbd2_stats_proc_exit(journal
);
1073 * journal_t * jbd2_journal_init_inode () - creates a journal which maps to a inode.
1074 * @inode: An inode to create the journal in
1076 * jbd2_journal_init_inode creates a journal which maps an on-disk inode as
1077 * the journal. The inode must exist already, must support bmap() and
1078 * must have all data blocks preallocated.
1080 journal_t
* jbd2_journal_init_inode (struct inode
*inode
)
1082 struct buffer_head
*bh
;
1083 journal_t
*journal
= journal_init_common();
1087 unsigned long long blocknr
;
1092 journal
->j_dev
= journal
->j_fs_dev
= inode
->i_sb
->s_bdev
;
1093 journal
->j_inode
= inode
;
1094 bdevname(journal
->j_dev
, journal
->j_devname
);
1095 p
= journal
->j_devname
;
1096 while ((p
= strchr(p
, '/')))
1098 p
= journal
->j_devname
+ strlen(journal
->j_devname
);
1099 sprintf(p
, "-%lu", journal
->j_inode
->i_ino
);
1101 "journal %p: inode %s/%ld, size %Ld, bits %d, blksize %ld\n",
1102 journal
, inode
->i_sb
->s_id
, inode
->i_ino
,
1103 (long long) inode
->i_size
,
1104 inode
->i_sb
->s_blocksize_bits
, inode
->i_sb
->s_blocksize
);
1106 journal
->j_maxlen
= inode
->i_size
>> inode
->i_sb
->s_blocksize_bits
;
1107 journal
->j_blocksize
= inode
->i_sb
->s_blocksize
;
1108 jbd2_stats_proc_init(journal
);
1110 /* journal descriptor can store up to n blocks -bzzz */
1111 n
= journal
->j_blocksize
/ sizeof(journal_block_tag_t
);
1112 journal
->j_wbufsize
= n
;
1113 journal
->j_wbuf
= kmalloc(n
* sizeof(struct buffer_head
*), GFP_KERNEL
);
1114 if (!journal
->j_wbuf
) {
1115 printk(KERN_ERR
"%s: Can't allocate bhs for commit thread\n",
1120 err
= jbd2_journal_bmap(journal
, 0, &blocknr
);
1121 /* If that failed, give up */
1123 printk(KERN_ERR
"%s: Cannot locate journal superblock\n",
1128 bh
= __getblk(journal
->j_dev
, blocknr
, journal
->j_blocksize
);
1131 "%s: Cannot get buffer for journal superblock\n",
1135 journal
->j_sb_buffer
= bh
;
1136 journal
->j_superblock
= (journal_superblock_t
*)bh
->b_data
;
1140 kfree(journal
->j_wbuf
);
1141 jbd2_stats_proc_exit(journal
);
1147 * If the journal init or create aborts, we need to mark the journal
1148 * superblock as being NULL to prevent the journal destroy from writing
1149 * back a bogus superblock.
1151 static void journal_fail_superblock (journal_t
*journal
)
1153 struct buffer_head
*bh
= journal
->j_sb_buffer
;
1155 journal
->j_sb_buffer
= NULL
;
1159 * Given a journal_t structure, initialise the various fields for
1160 * startup of a new journaling session. We use this both when creating
1161 * a journal, and after recovering an old journal to reset it for
1165 static int journal_reset(journal_t
*journal
)
1167 journal_superblock_t
*sb
= journal
->j_superblock
;
1168 unsigned long long first
, last
;
1170 first
= be32_to_cpu(sb
->s_first
);
1171 last
= be32_to_cpu(sb
->s_maxlen
);
1172 if (first
+ JBD2_MIN_JOURNAL_BLOCKS
> last
+ 1) {
1173 printk(KERN_ERR
"JBD2: Journal too short (blocks %llu-%llu).\n",
1175 journal_fail_superblock(journal
);
1179 journal
->j_first
= first
;
1180 journal
->j_last
= last
;
1182 journal
->j_head
= first
;
1183 journal
->j_tail
= first
;
1184 journal
->j_free
= last
- first
;
1186 journal
->j_tail_sequence
= journal
->j_transaction_sequence
;
1187 journal
->j_commit_sequence
= journal
->j_transaction_sequence
- 1;
1188 journal
->j_commit_request
= journal
->j_commit_sequence
;
1190 journal
->j_max_transaction_buffers
= journal
->j_maxlen
/ 4;
1193 * As a special case, if the on-disk copy is already marked as needing
1194 * no recovery (s_start == 0), then we can safely defer the superblock
1195 * update until the next commit by setting JBD2_FLUSHED. This avoids
1196 * attempting a write to a potential-readonly device.
1198 if (sb
->s_start
== 0) {
1199 jbd_debug(1, "JBD2: Skipping superblock update on recovered sb "
1200 "(start %ld, seq %d, errno %d)\n",
1201 journal
->j_tail
, journal
->j_tail_sequence
,
1203 journal
->j_flags
|= JBD2_FLUSHED
;
1205 /* Lock here to make assertions happy... */
1206 mutex_lock(&journal
->j_checkpoint_mutex
);
1208 * Update log tail information. We use WRITE_FUA since new
1209 * transaction will start reusing journal space and so we
1210 * must make sure information about current log tail is on
1213 jbd2_journal_update_sb_log_tail(journal
,
1214 journal
->j_tail_sequence
,
1217 mutex_unlock(&journal
->j_checkpoint_mutex
);
1219 return jbd2_journal_start_thread(journal
);
1222 static void jbd2_write_superblock(journal_t
*journal
, int write_op
)
1224 struct buffer_head
*bh
= journal
->j_sb_buffer
;
1227 trace_jbd2_write_superblock(journal
, write_op
);
1228 if (!(journal
->j_flags
& JBD2_BARRIER
))
1229 write_op
&= ~(REQ_FUA
| REQ_FLUSH
);
1231 if (buffer_write_io_error(bh
)) {
1233 * Oh, dear. A previous attempt to write the journal
1234 * superblock failed. This could happen because the
1235 * USB device was yanked out. Or it could happen to
1236 * be a transient write error and maybe the block will
1237 * be remapped. Nothing we can do but to retry the
1238 * write and hope for the best.
1240 printk(KERN_ERR
"JBD2: previous I/O error detected "
1241 "for journal superblock update for %s.\n",
1242 journal
->j_devname
);
1243 clear_buffer_write_io_error(bh
);
1244 set_buffer_uptodate(bh
);
1247 bh
->b_end_io
= end_buffer_write_sync
;
1248 ret
= submit_bh(write_op
, bh
);
1250 if (buffer_write_io_error(bh
)) {
1251 clear_buffer_write_io_error(bh
);
1252 set_buffer_uptodate(bh
);
1256 printk(KERN_ERR
"JBD2: Error %d detected when updating "
1257 "journal superblock for %s.\n", ret
,
1258 journal
->j_devname
);
1263 * jbd2_journal_update_sb_log_tail() - Update log tail in journal sb on disk.
1264 * @journal: The journal to update.
1265 * @tail_tid: TID of the new transaction at the tail of the log
1266 * @tail_block: The first block of the transaction at the tail of the log
1267 * @write_op: With which operation should we write the journal sb
1269 * Update a journal's superblock information about log tail and write it to
1270 * disk, waiting for the IO to complete.
1272 void jbd2_journal_update_sb_log_tail(journal_t
*journal
, tid_t tail_tid
,
1273 unsigned long tail_block
, int write_op
)
1275 journal_superblock_t
*sb
= journal
->j_superblock
;
1277 BUG_ON(!mutex_is_locked(&journal
->j_checkpoint_mutex
));
1278 jbd_debug(1, "JBD2: updating superblock (start %lu, seq %u)\n",
1279 tail_block
, tail_tid
);
1281 sb
->s_sequence
= cpu_to_be32(tail_tid
);
1282 sb
->s_start
= cpu_to_be32(tail_block
);
1284 jbd2_write_superblock(journal
, write_op
);
1286 /* Log is no longer empty */
1287 write_lock(&journal
->j_state_lock
);
1288 WARN_ON(!sb
->s_sequence
);
1289 journal
->j_flags
&= ~JBD2_FLUSHED
;
1290 write_unlock(&journal
->j_state_lock
);
1294 * jbd2_mark_journal_empty() - Mark on disk journal as empty.
1295 * @journal: The journal to update.
1297 * Update a journal's dynamic superblock fields to show that journal is empty.
1298 * Write updated superblock to disk waiting for IO to complete.
1300 static void jbd2_mark_journal_empty(journal_t
*journal
)
1302 journal_superblock_t
*sb
= journal
->j_superblock
;
1304 BUG_ON(!mutex_is_locked(&journal
->j_checkpoint_mutex
));
1305 read_lock(&journal
->j_state_lock
);
1306 jbd_debug(1, "JBD2: Marking journal as empty (seq %d)\n",
1307 journal
->j_tail_sequence
);
1309 sb
->s_sequence
= cpu_to_be32(journal
->j_tail_sequence
);
1310 sb
->s_start
= cpu_to_be32(0);
1311 read_unlock(&journal
->j_state_lock
);
1313 jbd2_write_superblock(journal
, WRITE_FUA
);
1315 /* Log is no longer empty */
1316 write_lock(&journal
->j_state_lock
);
1317 journal
->j_flags
|= JBD2_FLUSHED
;
1318 write_unlock(&journal
->j_state_lock
);
1323 * jbd2_journal_update_sb_errno() - Update error in the journal.
1324 * @journal: The journal to update.
1326 * Update a journal's errno. Write updated superblock to disk waiting for IO
1329 static void jbd2_journal_update_sb_errno(journal_t
*journal
)
1331 journal_superblock_t
*sb
= journal
->j_superblock
;
1333 read_lock(&journal
->j_state_lock
);
1334 jbd_debug(1, "JBD2: updating superblock error (errno %d)\n",
1336 sb
->s_errno
= cpu_to_be32(journal
->j_errno
);
1337 read_unlock(&journal
->j_state_lock
);
1339 jbd2_write_superblock(journal
, WRITE_SYNC
);
1343 * Read the superblock for a given journal, performing initial
1344 * validation of the format.
1346 static int journal_get_superblock(journal_t
*journal
)
1348 struct buffer_head
*bh
;
1349 journal_superblock_t
*sb
;
1352 bh
= journal
->j_sb_buffer
;
1354 J_ASSERT(bh
!= NULL
);
1355 if (!buffer_uptodate(bh
)) {
1356 ll_rw_block(READ
, 1, &bh
);
1358 if (!buffer_uptodate(bh
)) {
1360 "JBD2: IO error reading journal superblock\n");
1365 sb
= journal
->j_superblock
;
1369 if (sb
->s_header
.h_magic
!= cpu_to_be32(JBD2_MAGIC_NUMBER
) ||
1370 sb
->s_blocksize
!= cpu_to_be32(journal
->j_blocksize
)) {
1371 printk(KERN_WARNING
"JBD2: no valid journal superblock found\n");
1375 switch(be32_to_cpu(sb
->s_header
.h_blocktype
)) {
1376 case JBD2_SUPERBLOCK_V1
:
1377 journal
->j_format_version
= 1;
1379 case JBD2_SUPERBLOCK_V2
:
1380 journal
->j_format_version
= 2;
1383 printk(KERN_WARNING
"JBD2: unrecognised superblock format ID\n");
1387 if (be32_to_cpu(sb
->s_maxlen
) < journal
->j_maxlen
)
1388 journal
->j_maxlen
= be32_to_cpu(sb
->s_maxlen
);
1389 else if (be32_to_cpu(sb
->s_maxlen
) > journal
->j_maxlen
) {
1390 printk(KERN_WARNING
"JBD2: journal file too short\n");
1394 if (be32_to_cpu(sb
->s_first
) == 0 ||
1395 be32_to_cpu(sb
->s_first
) >= journal
->j_maxlen
) {
1397 "JBD2: Invalid start block of journal: %u\n",
1398 be32_to_cpu(sb
->s_first
));
1405 journal_fail_superblock(journal
);
1410 * Load the on-disk journal superblock and read the key fields into the
1414 static int load_superblock(journal_t
*journal
)
1417 journal_superblock_t
*sb
;
1419 err
= journal_get_superblock(journal
);
1423 sb
= journal
->j_superblock
;
1425 journal
->j_tail_sequence
= be32_to_cpu(sb
->s_sequence
);
1426 journal
->j_tail
= be32_to_cpu(sb
->s_start
);
1427 journal
->j_first
= be32_to_cpu(sb
->s_first
);
1428 journal
->j_last
= be32_to_cpu(sb
->s_maxlen
);
1429 journal
->j_errno
= be32_to_cpu(sb
->s_errno
);
1436 * int jbd2_journal_load() - Read journal from disk.
1437 * @journal: Journal to act on.
1439 * Given a journal_t structure which tells us which disk blocks contain
1440 * a journal, read the journal from disk to initialise the in-memory
1443 int jbd2_journal_load(journal_t
*journal
)
1446 journal_superblock_t
*sb
;
1448 err
= load_superblock(journal
);
1452 sb
= journal
->j_superblock
;
1453 /* If this is a V2 superblock, then we have to check the
1454 * features flags on it. */
1456 if (journal
->j_format_version
>= 2) {
1457 if ((sb
->s_feature_ro_compat
&
1458 ~cpu_to_be32(JBD2_KNOWN_ROCOMPAT_FEATURES
)) ||
1459 (sb
->s_feature_incompat
&
1460 ~cpu_to_be32(JBD2_KNOWN_INCOMPAT_FEATURES
))) {
1462 "JBD2: Unrecognised features on journal\n");
1468 * Create a slab for this blocksize
1470 err
= jbd2_journal_create_slab(be32_to_cpu(sb
->s_blocksize
));
1474 /* Let the recovery code check whether it needs to recover any
1475 * data from the journal. */
1476 if (jbd2_journal_recover(journal
))
1477 goto recovery_error
;
1479 if (journal
->j_failed_commit
) {
1480 printk(KERN_ERR
"JBD2: journal transaction %u on %s "
1481 "is corrupt.\n", journal
->j_failed_commit
,
1482 journal
->j_devname
);
1486 /* OK, we've finished with the dynamic journal bits:
1487 * reinitialise the dynamic contents of the superblock in memory
1488 * and reset them on disk. */
1489 if (journal_reset(journal
))
1490 goto recovery_error
;
1492 journal
->j_flags
&= ~JBD2_ABORT
;
1493 journal
->j_flags
|= JBD2_LOADED
;
1497 printk(KERN_WARNING
"JBD2: recovery failed\n");
1502 * void jbd2_journal_destroy() - Release a journal_t structure.
1503 * @journal: Journal to act on.
1505 * Release a journal_t structure once it is no longer in use by the
1507 * Return <0 if we couldn't clean up the journal.
1509 int jbd2_journal_destroy(journal_t
*journal
)
1513 /* Wait for the commit thread to wake up and die. */
1514 journal_kill_thread(journal
);
1516 /* Force a final log commit */
1517 if (journal
->j_running_transaction
)
1518 jbd2_journal_commit_transaction(journal
);
1520 /* Force any old transactions to disk */
1522 /* Totally anal locking here... */
1523 spin_lock(&journal
->j_list_lock
);
1524 while (journal
->j_checkpoint_transactions
!= NULL
) {
1525 spin_unlock(&journal
->j_list_lock
);
1526 mutex_lock(&journal
->j_checkpoint_mutex
);
1527 jbd2_log_do_checkpoint(journal
);
1528 mutex_unlock(&journal
->j_checkpoint_mutex
);
1529 spin_lock(&journal
->j_list_lock
);
1532 J_ASSERT(journal
->j_running_transaction
== NULL
);
1533 J_ASSERT(journal
->j_committing_transaction
== NULL
);
1534 J_ASSERT(journal
->j_checkpoint_transactions
== NULL
);
1535 spin_unlock(&journal
->j_list_lock
);
1537 if (journal
->j_sb_buffer
) {
1538 if (!is_journal_aborted(journal
)) {
1539 mutex_lock(&journal
->j_checkpoint_mutex
);
1540 jbd2_mark_journal_empty(journal
);
1541 mutex_unlock(&journal
->j_checkpoint_mutex
);
1544 brelse(journal
->j_sb_buffer
);
1547 if (journal
->j_proc_entry
)
1548 jbd2_stats_proc_exit(journal
);
1549 if (journal
->j_inode
)
1550 iput(journal
->j_inode
);
1551 if (journal
->j_revoke
)
1552 jbd2_journal_destroy_revoke(journal
);
1553 kfree(journal
->j_wbuf
);
1561 *int jbd2_journal_check_used_features () - Check if features specified are used.
1562 * @journal: Journal to check.
1563 * @compat: bitmask of compatible features
1564 * @ro: bitmask of features that force read-only mount
1565 * @incompat: bitmask of incompatible features
1567 * Check whether the journal uses all of a given set of
1568 * features. Return true (non-zero) if it does.
1571 int jbd2_journal_check_used_features (journal_t
*journal
, unsigned long compat
,
1572 unsigned long ro
, unsigned long incompat
)
1574 journal_superblock_t
*sb
;
1576 if (!compat
&& !ro
&& !incompat
)
1578 /* Load journal superblock if it is not loaded yet. */
1579 if (journal
->j_format_version
== 0 &&
1580 journal_get_superblock(journal
) != 0)
1582 if (journal
->j_format_version
== 1)
1585 sb
= journal
->j_superblock
;
1587 if (((be32_to_cpu(sb
->s_feature_compat
) & compat
) == compat
) &&
1588 ((be32_to_cpu(sb
->s_feature_ro_compat
) & ro
) == ro
) &&
1589 ((be32_to_cpu(sb
->s_feature_incompat
) & incompat
) == incompat
))
1596 * int jbd2_journal_check_available_features() - Check feature set in journalling layer
1597 * @journal: Journal to check.
1598 * @compat: bitmask of compatible features
1599 * @ro: bitmask of features that force read-only mount
1600 * @incompat: bitmask of incompatible features
1602 * Check whether the journaling code supports the use of
1603 * all of a given set of features on this journal. Return true
1604 * (non-zero) if it can. */
1606 int jbd2_journal_check_available_features (journal_t
*journal
, unsigned long compat
,
1607 unsigned long ro
, unsigned long incompat
)
1609 if (!compat
&& !ro
&& !incompat
)
1612 /* We can support any known requested features iff the
1613 * superblock is in version 2. Otherwise we fail to support any
1614 * extended sb features. */
1616 if (journal
->j_format_version
!= 2)
1619 if ((compat
& JBD2_KNOWN_COMPAT_FEATURES
) == compat
&&
1620 (ro
& JBD2_KNOWN_ROCOMPAT_FEATURES
) == ro
&&
1621 (incompat
& JBD2_KNOWN_INCOMPAT_FEATURES
) == incompat
)
1628 * int jbd2_journal_set_features () - Mark a given journal feature in the superblock
1629 * @journal: Journal to act on.
1630 * @compat: bitmask of compatible features
1631 * @ro: bitmask of features that force read-only mount
1632 * @incompat: bitmask of incompatible features
1634 * Mark a given journal feature as present on the
1635 * superblock. Returns true if the requested features could be set.
1639 int jbd2_journal_set_features (journal_t
*journal
, unsigned long compat
,
1640 unsigned long ro
, unsigned long incompat
)
1642 journal_superblock_t
*sb
;
1644 if (jbd2_journal_check_used_features(journal
, compat
, ro
, incompat
))
1647 if (!jbd2_journal_check_available_features(journal
, compat
, ro
, incompat
))
1650 jbd_debug(1, "Setting new features 0x%lx/0x%lx/0x%lx\n",
1651 compat
, ro
, incompat
);
1653 sb
= journal
->j_superblock
;
1655 sb
->s_feature_compat
|= cpu_to_be32(compat
);
1656 sb
->s_feature_ro_compat
|= cpu_to_be32(ro
);
1657 sb
->s_feature_incompat
|= cpu_to_be32(incompat
);
1663 * jbd2_journal_clear_features () - Clear a given journal feature in the
1665 * @journal: Journal to act on.
1666 * @compat: bitmask of compatible features
1667 * @ro: bitmask of features that force read-only mount
1668 * @incompat: bitmask of incompatible features
1670 * Clear a given journal feature as present on the
1673 void jbd2_journal_clear_features(journal_t
*journal
, unsigned long compat
,
1674 unsigned long ro
, unsigned long incompat
)
1676 journal_superblock_t
*sb
;
1678 jbd_debug(1, "Clear features 0x%lx/0x%lx/0x%lx\n",
1679 compat
, ro
, incompat
);
1681 sb
= journal
->j_superblock
;
1683 sb
->s_feature_compat
&= ~cpu_to_be32(compat
);
1684 sb
->s_feature_ro_compat
&= ~cpu_to_be32(ro
);
1685 sb
->s_feature_incompat
&= ~cpu_to_be32(incompat
);
1687 EXPORT_SYMBOL(jbd2_journal_clear_features
);
1690 * int jbd2_journal_flush () - Flush journal
1691 * @journal: Journal to act on.
1693 * Flush all data for a given journal to disk and empty the journal.
1694 * Filesystems can use this when remounting readonly to ensure that
1695 * recovery does not need to happen on remount.
1698 int jbd2_journal_flush(journal_t
*journal
)
1701 transaction_t
*transaction
= NULL
;
1703 write_lock(&journal
->j_state_lock
);
1705 /* Force everything buffered to the log... */
1706 if (journal
->j_running_transaction
) {
1707 transaction
= journal
->j_running_transaction
;
1708 __jbd2_log_start_commit(journal
, transaction
->t_tid
);
1709 } else if (journal
->j_committing_transaction
)
1710 transaction
= journal
->j_committing_transaction
;
1712 /* Wait for the log commit to complete... */
1714 tid_t tid
= transaction
->t_tid
;
1716 write_unlock(&journal
->j_state_lock
);
1717 jbd2_log_wait_commit(journal
, tid
);
1719 write_unlock(&journal
->j_state_lock
);
1722 /* ...and flush everything in the log out to disk. */
1723 spin_lock(&journal
->j_list_lock
);
1724 while (!err
&& journal
->j_checkpoint_transactions
!= NULL
) {
1725 spin_unlock(&journal
->j_list_lock
);
1726 mutex_lock(&journal
->j_checkpoint_mutex
);
1727 err
= jbd2_log_do_checkpoint(journal
);
1728 mutex_unlock(&journal
->j_checkpoint_mutex
);
1729 spin_lock(&journal
->j_list_lock
);
1731 spin_unlock(&journal
->j_list_lock
);
1733 if (is_journal_aborted(journal
))
1736 mutex_lock(&journal
->j_checkpoint_mutex
);
1737 jbd2_cleanup_journal_tail(journal
);
1739 /* Finally, mark the journal as really needing no recovery.
1740 * This sets s_start==0 in the underlying superblock, which is
1741 * the magic code for a fully-recovered superblock. Any future
1742 * commits of data to the journal will restore the current
1744 jbd2_mark_journal_empty(journal
);
1745 mutex_unlock(&journal
->j_checkpoint_mutex
);
1746 write_lock(&journal
->j_state_lock
);
1747 J_ASSERT(!journal
->j_running_transaction
);
1748 J_ASSERT(!journal
->j_committing_transaction
);
1749 J_ASSERT(!journal
->j_checkpoint_transactions
);
1750 J_ASSERT(journal
->j_head
== journal
->j_tail
);
1751 J_ASSERT(journal
->j_tail_sequence
== journal
->j_transaction_sequence
);
1752 write_unlock(&journal
->j_state_lock
);
1757 * int jbd2_journal_wipe() - Wipe journal contents
1758 * @journal: Journal to act on.
1759 * @write: flag (see below)
1761 * Wipe out all of the contents of a journal, safely. This will produce
1762 * a warning if the journal contains any valid recovery information.
1763 * Must be called between journal_init_*() and jbd2_journal_load().
1765 * If 'write' is non-zero, then we wipe out the journal on disk; otherwise
1766 * we merely suppress recovery.
1769 int jbd2_journal_wipe(journal_t
*journal
, int write
)
1773 J_ASSERT (!(journal
->j_flags
& JBD2_LOADED
));
1775 err
= load_superblock(journal
);
1779 if (!journal
->j_tail
)
1782 printk(KERN_WARNING
"JBD2: %s recovery information on journal\n",
1783 write
? "Clearing" : "Ignoring");
1785 err
= jbd2_journal_skip_recovery(journal
);
1787 /* Lock to make assertions happy... */
1788 mutex_lock(&journal
->j_checkpoint_mutex
);
1789 jbd2_mark_journal_empty(journal
);
1790 mutex_unlock(&journal
->j_checkpoint_mutex
);
1798 * Journal abort has very specific semantics, which we describe
1799 * for journal abort.
1801 * Two internal functions, which provide abort to the jbd layer
1806 * Quick version for internal journal use (doesn't lock the journal).
1807 * Aborts hard --- we mark the abort as occurred, but do _nothing_ else,
1808 * and don't attempt to make any other journal updates.
1810 void __jbd2_journal_abort_hard(journal_t
*journal
)
1812 transaction_t
*transaction
;
1814 if (journal
->j_flags
& JBD2_ABORT
)
1817 printk(KERN_ERR
"Aborting journal on device %s.\n",
1818 journal
->j_devname
);
1820 write_lock(&journal
->j_state_lock
);
1821 journal
->j_flags
|= JBD2_ABORT
;
1822 transaction
= journal
->j_running_transaction
;
1824 __jbd2_log_start_commit(journal
, transaction
->t_tid
);
1825 write_unlock(&journal
->j_state_lock
);
1828 /* Soft abort: record the abort error status in the journal superblock,
1829 * but don't do any other IO. */
1830 static void __journal_abort_soft (journal_t
*journal
, int errno
)
1832 if (journal
->j_flags
& JBD2_ABORT
)
1835 if (!journal
->j_errno
)
1836 journal
->j_errno
= errno
;
1838 __jbd2_journal_abort_hard(journal
);
1841 jbd2_journal_update_sb_errno(journal
);
1845 * void jbd2_journal_abort () - Shutdown the journal immediately.
1846 * @journal: the journal to shutdown.
1847 * @errno: an error number to record in the journal indicating
1848 * the reason for the shutdown.
1850 * Perform a complete, immediate shutdown of the ENTIRE
1851 * journal (not of a single transaction). This operation cannot be
1852 * undone without closing and reopening the journal.
1854 * The jbd2_journal_abort function is intended to support higher level error
1855 * recovery mechanisms such as the ext2/ext3 remount-readonly error
1858 * Journal abort has very specific semantics. Any existing dirty,
1859 * unjournaled buffers in the main filesystem will still be written to
1860 * disk by bdflush, but the journaling mechanism will be suspended
1861 * immediately and no further transaction commits will be honoured.
1863 * Any dirty, journaled buffers will be written back to disk without
1864 * hitting the journal. Atomicity cannot be guaranteed on an aborted
1865 * filesystem, but we _do_ attempt to leave as much data as possible
1866 * behind for fsck to use for cleanup.
1868 * Any attempt to get a new transaction handle on a journal which is in
1869 * ABORT state will just result in an -EROFS error return. A
1870 * jbd2_journal_stop on an existing handle will return -EIO if we have
1871 * entered abort state during the update.
1873 * Recursive transactions are not disturbed by journal abort until the
1874 * final jbd2_journal_stop, which will receive the -EIO error.
1876 * Finally, the jbd2_journal_abort call allows the caller to supply an errno
1877 * which will be recorded (if possible) in the journal superblock. This
1878 * allows a client to record failure conditions in the middle of a
1879 * transaction without having to complete the transaction to record the
1880 * failure to disk. ext3_error, for example, now uses this
1883 * Errors which originate from within the journaling layer will NOT
1884 * supply an errno; a null errno implies that absolutely no further
1885 * writes are done to the journal (unless there are any already in
1890 void jbd2_journal_abort(journal_t
*journal
, int errno
)
1892 __journal_abort_soft(journal
, errno
);
1896 * int jbd2_journal_errno () - returns the journal's error state.
1897 * @journal: journal to examine.
1899 * This is the errno number set with jbd2_journal_abort(), the last
1900 * time the journal was mounted - if the journal was stopped
1901 * without calling abort this will be 0.
1903 * If the journal has been aborted on this mount time -EROFS will
1906 int jbd2_journal_errno(journal_t
*journal
)
1910 read_lock(&journal
->j_state_lock
);
1911 if (journal
->j_flags
& JBD2_ABORT
)
1914 err
= journal
->j_errno
;
1915 read_unlock(&journal
->j_state_lock
);
1920 * int jbd2_journal_clear_err () - clears the journal's error state
1921 * @journal: journal to act on.
1923 * An error must be cleared or acked to take a FS out of readonly
1926 int jbd2_journal_clear_err(journal_t
*journal
)
1930 write_lock(&journal
->j_state_lock
);
1931 if (journal
->j_flags
& JBD2_ABORT
)
1934 journal
->j_errno
= 0;
1935 write_unlock(&journal
->j_state_lock
);
1940 * void jbd2_journal_ack_err() - Ack journal err.
1941 * @journal: journal to act on.
1943 * An error must be cleared or acked to take a FS out of readonly
1946 void jbd2_journal_ack_err(journal_t
*journal
)
1948 write_lock(&journal
->j_state_lock
);
1949 if (journal
->j_errno
)
1950 journal
->j_flags
|= JBD2_ACK_ERR
;
1951 write_unlock(&journal
->j_state_lock
);
1954 int jbd2_journal_blocks_per_page(struct inode
*inode
)
1956 return 1 << (PAGE_CACHE_SHIFT
- inode
->i_sb
->s_blocksize_bits
);
1960 * helper functions to deal with 32 or 64bit block numbers.
1962 size_t journal_tag_bytes(journal_t
*journal
)
1964 if (JBD2_HAS_INCOMPAT_FEATURE(journal
, JBD2_FEATURE_INCOMPAT_64BIT
))
1965 return JBD2_TAG_SIZE64
;
1967 return JBD2_TAG_SIZE32
;
1971 * JBD memory management
1973 * These functions are used to allocate block-sized chunks of memory
1974 * used for making copies of buffer_head data. Very often it will be
1975 * page-sized chunks of data, but sometimes it will be in
1976 * sub-page-size chunks. (For example, 16k pages on Power systems
1977 * with a 4k block file system.) For blocks smaller than a page, we
1978 * use a SLAB allocator. There are slab caches for each block size,
1979 * which are allocated at mount time, if necessary, and we only free
1980 * (all of) the slab caches when/if the jbd2 module is unloaded. For
1981 * this reason we don't need to a mutex to protect access to
1982 * jbd2_slab[] allocating or releasing memory; only in
1983 * jbd2_journal_create_slab().
1985 #define JBD2_MAX_SLABS 8
1986 static struct kmem_cache
*jbd2_slab
[JBD2_MAX_SLABS
];
1988 static const char *jbd2_slab_names
[JBD2_MAX_SLABS
] = {
1989 "jbd2_1k", "jbd2_2k", "jbd2_4k", "jbd2_8k",
1990 "jbd2_16k", "jbd2_32k", "jbd2_64k", "jbd2_128k"
1994 static void jbd2_journal_destroy_slabs(void)
1998 for (i
= 0; i
< JBD2_MAX_SLABS
; i
++) {
2000 kmem_cache_destroy(jbd2_slab
[i
]);
2001 jbd2_slab
[i
] = NULL
;
2005 static int jbd2_journal_create_slab(size_t size
)
2007 static DEFINE_MUTEX(jbd2_slab_create_mutex
);
2008 int i
= order_base_2(size
) - 10;
2011 if (size
== PAGE_SIZE
)
2014 if (i
>= JBD2_MAX_SLABS
)
2017 if (unlikely(i
< 0))
2019 mutex_lock(&jbd2_slab_create_mutex
);
2021 mutex_unlock(&jbd2_slab_create_mutex
);
2022 return 0; /* Already created */
2025 slab_size
= 1 << (i
+10);
2026 jbd2_slab
[i
] = kmem_cache_create(jbd2_slab_names
[i
], slab_size
,
2027 slab_size
, 0, NULL
);
2028 mutex_unlock(&jbd2_slab_create_mutex
);
2029 if (!jbd2_slab
[i
]) {
2030 printk(KERN_EMERG
"JBD2: no memory for jbd2_slab cache\n");
2036 static struct kmem_cache
*get_slab(size_t size
)
2038 int i
= order_base_2(size
) - 10;
2040 BUG_ON(i
>= JBD2_MAX_SLABS
);
2041 if (unlikely(i
< 0))
2043 BUG_ON(jbd2_slab
[i
] == NULL
);
2044 return jbd2_slab
[i
];
2047 void *jbd2_alloc(size_t size
, gfp_t flags
)
2051 BUG_ON(size
& (size
-1)); /* Must be a power of 2 */
2053 flags
|= __GFP_REPEAT
;
2054 if (size
== PAGE_SIZE
)
2055 ptr
= (void *)__get_free_pages(flags
, 0);
2056 else if (size
> PAGE_SIZE
) {
2057 int order
= get_order(size
);
2060 ptr
= (void *)__get_free_pages(flags
, order
);
2062 ptr
= vmalloc(size
);
2064 ptr
= kmem_cache_alloc(get_slab(size
), flags
);
2066 /* Check alignment; SLUB has gotten this wrong in the past,
2067 * and this can lead to user data corruption! */
2068 BUG_ON(((unsigned long) ptr
) & (size
-1));
2073 void jbd2_free(void *ptr
, size_t size
)
2075 if (size
== PAGE_SIZE
) {
2076 free_pages((unsigned long)ptr
, 0);
2079 if (size
> PAGE_SIZE
) {
2080 int order
= get_order(size
);
2083 free_pages((unsigned long)ptr
, order
);
2088 kmem_cache_free(get_slab(size
), ptr
);
2092 * Journal_head storage management
2094 static struct kmem_cache
*jbd2_journal_head_cache
;
2095 #ifdef CONFIG_JBD2_DEBUG
2096 static atomic_t nr_journal_heads
= ATOMIC_INIT(0);
2099 static int jbd2_journal_init_journal_head_cache(void)
2103 J_ASSERT(jbd2_journal_head_cache
== NULL
);
2104 jbd2_journal_head_cache
= kmem_cache_create("jbd2_journal_head",
2105 sizeof(struct journal_head
),
2107 SLAB_TEMPORARY
, /* flags */
2110 if (!jbd2_journal_head_cache
) {
2112 printk(KERN_EMERG
"JBD2: no memory for journal_head cache\n");
2117 static void jbd2_journal_destroy_journal_head_cache(void)
2119 if (jbd2_journal_head_cache
) {
2120 kmem_cache_destroy(jbd2_journal_head_cache
);
2121 jbd2_journal_head_cache
= NULL
;
2126 * journal_head splicing and dicing
2128 static struct journal_head
*journal_alloc_journal_head(void)
2130 struct journal_head
*ret
;
2132 #ifdef CONFIG_JBD2_DEBUG
2133 atomic_inc(&nr_journal_heads
);
2135 ret
= kmem_cache_alloc(jbd2_journal_head_cache
, GFP_NOFS
);
2137 jbd_debug(1, "out of memory for journal_head\n");
2138 pr_notice_ratelimited("ENOMEM in %s, retrying.\n", __func__
);
2141 ret
= kmem_cache_alloc(jbd2_journal_head_cache
, GFP_NOFS
);
2147 static void journal_free_journal_head(struct journal_head
*jh
)
2149 #ifdef CONFIG_JBD2_DEBUG
2150 atomic_dec(&nr_journal_heads
);
2151 memset(jh
, JBD2_POISON_FREE
, sizeof(*jh
));
2153 kmem_cache_free(jbd2_journal_head_cache
, jh
);
2157 * A journal_head is attached to a buffer_head whenever JBD has an
2158 * interest in the buffer.
2160 * Whenever a buffer has an attached journal_head, its ->b_state:BH_JBD bit
2161 * is set. This bit is tested in core kernel code where we need to take
2162 * JBD-specific actions. Testing the zeroness of ->b_private is not reliable
2165 * When a buffer has its BH_JBD bit set, its ->b_count is elevated by one.
2167 * When a buffer has its BH_JBD bit set it is immune from being released by
2168 * core kernel code, mainly via ->b_count.
2170 * A journal_head is detached from its buffer_head when the journal_head's
2171 * b_jcount reaches zero. Running transaction (b_transaction) and checkpoint
2172 * transaction (b_cp_transaction) hold their references to b_jcount.
2174 * Various places in the kernel want to attach a journal_head to a buffer_head
2175 * _before_ attaching the journal_head to a transaction. To protect the
2176 * journal_head in this situation, jbd2_journal_add_journal_head elevates the
2177 * journal_head's b_jcount refcount by one. The caller must call
2178 * jbd2_journal_put_journal_head() to undo this.
2180 * So the typical usage would be:
2182 * (Attach a journal_head if needed. Increments b_jcount)
2183 * struct journal_head *jh = jbd2_journal_add_journal_head(bh);
2185 * (Get another reference for transaction)
2186 * jbd2_journal_grab_journal_head(bh);
2187 * jh->b_transaction = xxx;
2188 * (Put original reference)
2189 * jbd2_journal_put_journal_head(jh);
2193 * Give a buffer_head a journal_head.
2197 struct journal_head
*jbd2_journal_add_journal_head(struct buffer_head
*bh
)
2199 struct journal_head
*jh
;
2200 struct journal_head
*new_jh
= NULL
;
2203 if (!buffer_jbd(bh
)) {
2204 new_jh
= journal_alloc_journal_head();
2205 memset(new_jh
, 0, sizeof(*new_jh
));
2208 jbd_lock_bh_journal_head(bh
);
2209 if (buffer_jbd(bh
)) {
2213 (atomic_read(&bh
->b_count
) > 0) ||
2214 (bh
->b_page
&& bh
->b_page
->mapping
));
2217 jbd_unlock_bh_journal_head(bh
);
2222 new_jh
= NULL
; /* We consumed it */
2227 BUFFER_TRACE(bh
, "added journal_head");
2230 jbd_unlock_bh_journal_head(bh
);
2232 journal_free_journal_head(new_jh
);
2233 return bh
->b_private
;
2237 * Grab a ref against this buffer_head's journal_head. If it ended up not
2238 * having a journal_head, return NULL
2240 struct journal_head
*jbd2_journal_grab_journal_head(struct buffer_head
*bh
)
2242 struct journal_head
*jh
= NULL
;
2244 jbd_lock_bh_journal_head(bh
);
2245 if (buffer_jbd(bh
)) {
2249 jbd_unlock_bh_journal_head(bh
);
2253 static void __journal_remove_journal_head(struct buffer_head
*bh
)
2255 struct journal_head
*jh
= bh2jh(bh
);
2257 J_ASSERT_JH(jh
, jh
->b_jcount
>= 0);
2258 J_ASSERT_JH(jh
, jh
->b_transaction
== NULL
);
2259 J_ASSERT_JH(jh
, jh
->b_next_transaction
== NULL
);
2260 J_ASSERT_JH(jh
, jh
->b_cp_transaction
== NULL
);
2261 J_ASSERT_JH(jh
, jh
->b_jlist
== BJ_None
);
2262 J_ASSERT_BH(bh
, buffer_jbd(bh
));
2263 J_ASSERT_BH(bh
, jh2bh(jh
) == bh
);
2264 BUFFER_TRACE(bh
, "remove journal_head");
2265 if (jh
->b_frozen_data
) {
2266 printk(KERN_WARNING
"%s: freeing b_frozen_data\n", __func__
);
2267 jbd2_free(jh
->b_frozen_data
, bh
->b_size
);
2269 if (jh
->b_committed_data
) {
2270 printk(KERN_WARNING
"%s: freeing b_committed_data\n", __func__
);
2271 jbd2_free(jh
->b_committed_data
, bh
->b_size
);
2273 bh
->b_private
= NULL
;
2274 jh
->b_bh
= NULL
; /* debug, really */
2275 clear_buffer_jbd(bh
);
2276 journal_free_journal_head(jh
);
2280 * Drop a reference on the passed journal_head. If it fell to zero then
2281 * release the journal_head from the buffer_head.
2283 void jbd2_journal_put_journal_head(struct journal_head
*jh
)
2285 struct buffer_head
*bh
= jh2bh(jh
);
2287 jbd_lock_bh_journal_head(bh
);
2288 J_ASSERT_JH(jh
, jh
->b_jcount
> 0);
2290 if (!jh
->b_jcount
) {
2291 __journal_remove_journal_head(bh
);
2292 jbd_unlock_bh_journal_head(bh
);
2295 jbd_unlock_bh_journal_head(bh
);
2299 * Initialize jbd inode head
2301 void jbd2_journal_init_jbd_inode(struct jbd2_inode
*jinode
, struct inode
*inode
)
2303 jinode
->i_transaction
= NULL
;
2304 jinode
->i_next_transaction
= NULL
;
2305 jinode
->i_vfs_inode
= inode
;
2306 jinode
->i_flags
= 0;
2307 INIT_LIST_HEAD(&jinode
->i_list
);
2311 * Function to be called before we start removing inode from memory (i.e.,
2312 * clear_inode() is a fine place to be called from). It removes inode from
2313 * transaction's lists.
2315 void jbd2_journal_release_jbd_inode(journal_t
*journal
,
2316 struct jbd2_inode
*jinode
)
2321 spin_lock(&journal
->j_list_lock
);
2322 /* Is commit writing out inode - we have to wait */
2323 if (test_bit(__JI_COMMIT_RUNNING
, &jinode
->i_flags
)) {
2324 wait_queue_head_t
*wq
;
2325 DEFINE_WAIT_BIT(wait
, &jinode
->i_flags
, __JI_COMMIT_RUNNING
);
2326 wq
= bit_waitqueue(&jinode
->i_flags
, __JI_COMMIT_RUNNING
);
2327 prepare_to_wait(wq
, &wait
.wait
, TASK_UNINTERRUPTIBLE
);
2328 spin_unlock(&journal
->j_list_lock
);
2330 finish_wait(wq
, &wait
.wait
);
2334 if (jinode
->i_transaction
) {
2335 list_del(&jinode
->i_list
);
2336 jinode
->i_transaction
= NULL
;
2338 spin_unlock(&journal
->j_list_lock
);
2344 #ifdef CONFIG_JBD2_DEBUG
2345 u8 jbd2_journal_enable_debug __read_mostly
;
2346 EXPORT_SYMBOL(jbd2_journal_enable_debug
);
2348 #define JBD2_DEBUG_NAME "jbd2-debug"
2350 static struct dentry
*jbd2_debugfs_dir
;
2351 static struct dentry
*jbd2_debug
;
2353 static void __init
jbd2_create_debugfs_entry(void)
2355 jbd2_debugfs_dir
= debugfs_create_dir("jbd2", NULL
);
2356 if (jbd2_debugfs_dir
)
2357 jbd2_debug
= debugfs_create_u8(JBD2_DEBUG_NAME
,
2360 &jbd2_journal_enable_debug
);
2363 static void __exit
jbd2_remove_debugfs_entry(void)
2365 debugfs_remove(jbd2_debug
);
2366 debugfs_remove(jbd2_debugfs_dir
);
2371 static void __init
jbd2_create_debugfs_entry(void)
2375 static void __exit
jbd2_remove_debugfs_entry(void)
2381 #ifdef CONFIG_PROC_FS
2383 #define JBD2_STATS_PROC_NAME "fs/jbd2"
2385 static void __init
jbd2_create_jbd_stats_proc_entry(void)
2387 proc_jbd2_stats
= proc_mkdir(JBD2_STATS_PROC_NAME
, NULL
);
2390 static void __exit
jbd2_remove_jbd_stats_proc_entry(void)
2392 if (proc_jbd2_stats
)
2393 remove_proc_entry(JBD2_STATS_PROC_NAME
, NULL
);
2398 #define jbd2_create_jbd_stats_proc_entry() do {} while (0)
2399 #define jbd2_remove_jbd_stats_proc_entry() do {} while (0)
2403 struct kmem_cache
*jbd2_handle_cache
, *jbd2_inode_cache
;
2405 static int __init
jbd2_journal_init_handle_cache(void)
2407 jbd2_handle_cache
= KMEM_CACHE(jbd2_journal_handle
, SLAB_TEMPORARY
);
2408 if (jbd2_handle_cache
== NULL
) {
2409 printk(KERN_EMERG
"JBD2: failed to create handle cache\n");
2412 jbd2_inode_cache
= KMEM_CACHE(jbd2_inode
, 0);
2413 if (jbd2_inode_cache
== NULL
) {
2414 printk(KERN_EMERG
"JBD2: failed to create inode cache\n");
2415 kmem_cache_destroy(jbd2_handle_cache
);
2421 static void jbd2_journal_destroy_handle_cache(void)
2423 if (jbd2_handle_cache
)
2424 kmem_cache_destroy(jbd2_handle_cache
);
2425 if (jbd2_inode_cache
)
2426 kmem_cache_destroy(jbd2_inode_cache
);
2431 * Module startup and shutdown
2434 static int __init
journal_init_caches(void)
2438 ret
= jbd2_journal_init_revoke_caches();
2440 ret
= jbd2_journal_init_journal_head_cache();
2442 ret
= jbd2_journal_init_handle_cache();
2444 ret
= jbd2_journal_init_transaction_cache();
2448 static void jbd2_journal_destroy_caches(void)
2450 jbd2_journal_destroy_revoke_caches();
2451 jbd2_journal_destroy_journal_head_cache();
2452 jbd2_journal_destroy_handle_cache();
2453 jbd2_journal_destroy_transaction_cache();
2454 jbd2_journal_destroy_slabs();
2457 static int __init
journal_init(void)
2461 BUILD_BUG_ON(sizeof(struct journal_superblock_s
) != 1024);
2463 ret
= journal_init_caches();
2465 jbd2_create_debugfs_entry();
2466 jbd2_create_jbd_stats_proc_entry();
2468 jbd2_journal_destroy_caches();
2473 static void __exit
journal_exit(void)
2475 #ifdef CONFIG_JBD2_DEBUG
2476 int n
= atomic_read(&nr_journal_heads
);
2478 printk(KERN_EMERG
"JBD2: leaked %d journal_heads!\n", n
);
2480 jbd2_remove_debugfs_entry();
2481 jbd2_remove_jbd_stats_proc_entry();
2482 jbd2_journal_destroy_caches();
2485 MODULE_LICENSE("GPL");
2486 module_init(journal_init
);
2487 module_exit(journal_exit
);