2 * linux/fs/jbd2/journal.c
4 * Written by Stephen C. Tweedie <sct@redhat.com>, 1998
6 * Copyright 1998 Red Hat corp --- All Rights Reserved
8 * This file is part of the Linux kernel and is made available under
9 * the terms of the GNU General Public License, version 2, or at your
10 * option, any later version, incorporated herein by reference.
12 * Generic filesystem journal-writing code; part of the ext2fs
15 * This file manages journals: areas of disk reserved for logging
16 * transactional updates. This includes the kernel journaling thread
17 * which is responsible for scheduling updates to the log.
19 * We do not actually manage the physical storage of the journal in this
20 * file: that is left to a per-journal policy function, which allows us
21 * to store the journal within a filesystem-specified area for ext2
22 * journaling (ext2 can use a reserved inode for storing the log).
25 #include <linux/module.h>
26 #include <linux/time.h>
28 #include <linux/jbd2.h>
29 #include <linux/errno.h>
30 #include <linux/slab.h>
31 #include <linux/init.h>
33 #include <linux/freezer.h>
34 #include <linux/pagemap.h>
35 #include <linux/kthread.h>
36 #include <linux/poison.h>
37 #include <linux/proc_fs.h>
38 #include <linux/seq_file.h>
39 #include <linux/math64.h>
40 #include <linux/hash.h>
41 #include <linux/log2.h>
42 #include <linux/vmalloc.h>
43 #include <linux/backing-dev.h>
44 #include <linux/bitops.h>
45 #include <linux/ratelimit.h>
47 #define CREATE_TRACE_POINTS
48 #include <trace/events/jbd2.h>
50 #include <asm/uaccess.h>
53 #ifdef CONFIG_JBD2_DEBUG
54 ushort jbd2_journal_enable_debug __read_mostly
;
55 EXPORT_SYMBOL(jbd2_journal_enable_debug
);
57 module_param_named(jbd2_debug
, jbd2_journal_enable_debug
, ushort
, 0644);
58 MODULE_PARM_DESC(jbd2_debug
, "Debugging level for jbd2");
61 EXPORT_SYMBOL(jbd2_journal_extend
);
62 EXPORT_SYMBOL(jbd2_journal_stop
);
63 EXPORT_SYMBOL(jbd2_journal_lock_updates
);
64 EXPORT_SYMBOL(jbd2_journal_unlock_updates
);
65 EXPORT_SYMBOL(jbd2_journal_get_write_access
);
66 EXPORT_SYMBOL(jbd2_journal_get_create_access
);
67 EXPORT_SYMBOL(jbd2_journal_get_undo_access
);
68 EXPORT_SYMBOL(jbd2_journal_set_triggers
);
69 EXPORT_SYMBOL(jbd2_journal_dirty_metadata
);
70 EXPORT_SYMBOL(jbd2_journal_forget
);
72 EXPORT_SYMBOL(journal_sync_buffer
);
74 EXPORT_SYMBOL(jbd2_journal_flush
);
75 EXPORT_SYMBOL(jbd2_journal_revoke
);
77 EXPORT_SYMBOL(jbd2_journal_init_dev
);
78 EXPORT_SYMBOL(jbd2_journal_init_inode
);
79 EXPORT_SYMBOL(jbd2_journal_check_used_features
);
80 EXPORT_SYMBOL(jbd2_journal_check_available_features
);
81 EXPORT_SYMBOL(jbd2_journal_set_features
);
82 EXPORT_SYMBOL(jbd2_journal_load
);
83 EXPORT_SYMBOL(jbd2_journal_destroy
);
84 EXPORT_SYMBOL(jbd2_journal_abort
);
85 EXPORT_SYMBOL(jbd2_journal_errno
);
86 EXPORT_SYMBOL(jbd2_journal_ack_err
);
87 EXPORT_SYMBOL(jbd2_journal_clear_err
);
88 EXPORT_SYMBOL(jbd2_log_wait_commit
);
89 EXPORT_SYMBOL(jbd2_log_start_commit
);
90 EXPORT_SYMBOL(jbd2_journal_start_commit
);
91 EXPORT_SYMBOL(jbd2_journal_force_commit_nested
);
92 EXPORT_SYMBOL(jbd2_journal_wipe
);
93 EXPORT_SYMBOL(jbd2_journal_blocks_per_page
);
94 EXPORT_SYMBOL(jbd2_journal_invalidatepage
);
95 EXPORT_SYMBOL(jbd2_journal_try_to_free_buffers
);
96 EXPORT_SYMBOL(jbd2_journal_force_commit
);
97 EXPORT_SYMBOL(jbd2_journal_file_inode
);
98 EXPORT_SYMBOL(jbd2_journal_init_jbd_inode
);
99 EXPORT_SYMBOL(jbd2_journal_release_jbd_inode
);
100 EXPORT_SYMBOL(jbd2_journal_begin_ordered_truncate
);
101 EXPORT_SYMBOL(jbd2_inode_cache
);
103 static void __journal_abort_soft (journal_t
*journal
, int errno
);
104 static int jbd2_journal_create_slab(size_t slab_size
);
106 /* Checksumming functions */
107 int jbd2_verify_csum_type(journal_t
*j
, journal_superblock_t
*sb
)
109 if (!JBD2_HAS_INCOMPAT_FEATURE(j
, JBD2_FEATURE_INCOMPAT_CSUM_V2
))
112 return sb
->s_checksum_type
== JBD2_CRC32C_CHKSUM
;
115 static __u32
jbd2_superblock_csum(journal_t
*j
, journal_superblock_t
*sb
)
117 __u32 csum
, old_csum
;
119 old_csum
= sb
->s_checksum
;
121 csum
= jbd2_chksum(j
, ~0, (char *)sb
, sizeof(journal_superblock_t
));
122 sb
->s_checksum
= old_csum
;
124 return cpu_to_be32(csum
);
127 int jbd2_superblock_csum_verify(journal_t
*j
, journal_superblock_t
*sb
)
129 if (!JBD2_HAS_INCOMPAT_FEATURE(j
, JBD2_FEATURE_INCOMPAT_CSUM_V2
))
132 return sb
->s_checksum
== jbd2_superblock_csum(j
, sb
);
135 void jbd2_superblock_csum_set(journal_t
*j
, journal_superblock_t
*sb
)
137 if (!JBD2_HAS_INCOMPAT_FEATURE(j
, JBD2_FEATURE_INCOMPAT_CSUM_V2
))
140 sb
->s_checksum
= jbd2_superblock_csum(j
, sb
);
144 * Helper function used to manage commit timeouts
147 static void commit_timeout(unsigned long __data
)
149 struct task_struct
* p
= (struct task_struct
*) __data
;
155 * kjournald2: The main thread function used to manage a logging device
158 * This kernel thread is responsible for two things:
160 * 1) COMMIT: Every so often we need to commit the current state of the
161 * filesystem to disk. The journal thread is responsible for writing
162 * all of the metadata buffers to disk.
164 * 2) CHECKPOINT: We cannot reuse a used section of the log file until all
165 * of the data in that part of the log has been rewritten elsewhere on
166 * the disk. Flushing these old buffers to reclaim space in the log is
167 * known as checkpointing, and this thread is responsible for that job.
170 static int kjournald2(void *arg
)
172 journal_t
*journal
= arg
;
173 transaction_t
*transaction
;
176 * Set up an interval timer which can be used to trigger a commit wakeup
177 * after the commit interval expires
179 setup_timer(&journal
->j_commit_timer
, commit_timeout
,
180 (unsigned long)current
);
184 /* Record that the journal thread is running */
185 journal
->j_task
= current
;
186 wake_up(&journal
->j_wait_done_commit
);
189 * And now, wait forever for commit wakeup events.
191 write_lock(&journal
->j_state_lock
);
194 if (journal
->j_flags
& JBD2_UNMOUNT
)
197 jbd_debug(1, "commit_sequence=%d, commit_request=%d\n",
198 journal
->j_commit_sequence
, journal
->j_commit_request
);
200 if (journal
->j_commit_sequence
!= journal
->j_commit_request
) {
201 jbd_debug(1, "OK, requests differ\n");
202 write_unlock(&journal
->j_state_lock
);
203 del_timer_sync(&journal
->j_commit_timer
);
204 jbd2_journal_commit_transaction(journal
);
205 write_lock(&journal
->j_state_lock
);
209 wake_up(&journal
->j_wait_done_commit
);
210 if (freezing(current
)) {
212 * The simpler the better. Flushing journal isn't a
213 * good idea, because that depends on threads that may
214 * be already stopped.
216 jbd_debug(1, "Now suspending kjournald2\n");
217 write_unlock(&journal
->j_state_lock
);
219 write_lock(&journal
->j_state_lock
);
222 * We assume on resume that commits are already there,
226 int should_sleep
= 1;
228 prepare_to_wait(&journal
->j_wait_commit
, &wait
,
230 if (journal
->j_commit_sequence
!= journal
->j_commit_request
)
232 transaction
= journal
->j_running_transaction
;
233 if (transaction
&& time_after_eq(jiffies
,
234 transaction
->t_expires
))
236 if (journal
->j_flags
& JBD2_UNMOUNT
)
239 write_unlock(&journal
->j_state_lock
);
241 write_lock(&journal
->j_state_lock
);
243 finish_wait(&journal
->j_wait_commit
, &wait
);
246 jbd_debug(1, "kjournald2 wakes\n");
249 * Were we woken up by a commit wakeup event?
251 transaction
= journal
->j_running_transaction
;
252 if (transaction
&& time_after_eq(jiffies
, transaction
->t_expires
)) {
253 journal
->j_commit_request
= transaction
->t_tid
;
254 jbd_debug(1, "woke because of timeout\n");
259 write_unlock(&journal
->j_state_lock
);
260 del_timer_sync(&journal
->j_commit_timer
);
261 journal
->j_task
= NULL
;
262 wake_up(&journal
->j_wait_done_commit
);
263 jbd_debug(1, "Journal thread exiting.\n");
267 static int jbd2_journal_start_thread(journal_t
*journal
)
269 struct task_struct
*t
;
271 t
= kthread_run(kjournald2
, journal
, "jbd2/%s",
276 wait_event(journal
->j_wait_done_commit
, journal
->j_task
!= NULL
);
280 static void journal_kill_thread(journal_t
*journal
)
282 write_lock(&journal
->j_state_lock
);
283 journal
->j_flags
|= JBD2_UNMOUNT
;
285 while (journal
->j_task
) {
286 wake_up(&journal
->j_wait_commit
);
287 write_unlock(&journal
->j_state_lock
);
288 wait_event(journal
->j_wait_done_commit
, journal
->j_task
== NULL
);
289 write_lock(&journal
->j_state_lock
);
291 write_unlock(&journal
->j_state_lock
);
295 * jbd2_journal_write_metadata_buffer: write a metadata buffer to the journal.
297 * Writes a metadata buffer to a given disk block. The actual IO is not
298 * performed but a new buffer_head is constructed which labels the data
299 * to be written with the correct destination disk block.
301 * Any magic-number escaping which needs to be done will cause a
302 * copy-out here. If the buffer happens to start with the
303 * JBD2_MAGIC_NUMBER, then we can't write it to the log directly: the
304 * magic number is only written to the log for descripter blocks. In
305 * this case, we copy the data and replace the first word with 0, and we
306 * return a result code which indicates that this buffer needs to be
307 * marked as an escaped buffer in the corresponding log descriptor
308 * block. The missing word can then be restored when the block is read
311 * If the source buffer has already been modified by a new transaction
312 * since we took the last commit snapshot, we use the frozen copy of
313 * that data for IO. If we end up using the existing buffer_head's data
314 * for the write, then we *have* to lock the buffer to prevent anyone
315 * else from using and possibly modifying it while the IO is in
318 * The function returns a pointer to the buffer_heads to be used for IO.
320 * We assume that the journal has already been locked in this function.
327 * Bit 0 set == escape performed on the data
328 * Bit 1 set == buffer copy-out performed (kfree the data after IO)
331 int jbd2_journal_write_metadata_buffer(transaction_t
*transaction
,
332 struct journal_head
*jh_in
,
333 struct journal_head
**jh_out
,
334 unsigned long long blocknr
)
336 int need_copy_out
= 0;
337 int done_copy_out
= 0;
340 struct buffer_head
*new_bh
;
341 struct journal_head
*new_jh
;
342 struct page
*new_page
;
343 unsigned int new_offset
;
344 struct buffer_head
*bh_in
= jh2bh(jh_in
);
345 journal_t
*journal
= transaction
->t_journal
;
348 * The buffer really shouldn't be locked: only the current committing
349 * transaction is allowed to write it, so nobody else is allowed
352 * akpm: except if we're journalling data, and write() output is
353 * also part of a shared mapping, and another thread has
354 * decided to launch a writepage() against this buffer.
356 J_ASSERT_BH(bh_in
, buffer_jbddirty(bh_in
));
359 new_bh
= alloc_buffer_head(GFP_NOFS
);
362 * Failure is not an option, but __GFP_NOFAIL is going
363 * away; so we retry ourselves here.
365 congestion_wait(BLK_RW_ASYNC
, HZ
/50);
369 /* keep subsequent assertions sane */
371 init_buffer(new_bh
, NULL
, NULL
);
372 atomic_set(&new_bh
->b_count
, 1);
373 new_jh
= jbd2_journal_add_journal_head(new_bh
); /* This sleeps */
376 * If a new transaction has already done a buffer copy-out, then
377 * we use that version of the data for the commit.
379 jbd_lock_bh_state(bh_in
);
381 if (jh_in
->b_frozen_data
) {
383 new_page
= virt_to_page(jh_in
->b_frozen_data
);
384 new_offset
= offset_in_page(jh_in
->b_frozen_data
);
386 new_page
= jh2bh(jh_in
)->b_page
;
387 new_offset
= offset_in_page(jh2bh(jh_in
)->b_data
);
390 mapped_data
= kmap_atomic(new_page
);
392 * Fire data frozen trigger if data already wasn't frozen. Do this
393 * before checking for escaping, as the trigger may modify the magic
394 * offset. If a copy-out happens afterwards, it will have the correct
395 * data in the buffer.
398 jbd2_buffer_frozen_trigger(jh_in
, mapped_data
+ new_offset
,
404 if (*((__be32
*)(mapped_data
+ new_offset
)) ==
405 cpu_to_be32(JBD2_MAGIC_NUMBER
)) {
409 kunmap_atomic(mapped_data
);
412 * Do we need to do a data copy?
414 if (need_copy_out
&& !done_copy_out
) {
417 jbd_unlock_bh_state(bh_in
);
418 tmp
= jbd2_alloc(bh_in
->b_size
, GFP_NOFS
);
420 jbd2_journal_put_journal_head(new_jh
);
423 jbd_lock_bh_state(bh_in
);
424 if (jh_in
->b_frozen_data
) {
425 jbd2_free(tmp
, bh_in
->b_size
);
429 jh_in
->b_frozen_data
= tmp
;
430 mapped_data
= kmap_atomic(new_page
);
431 memcpy(tmp
, mapped_data
+ new_offset
, jh2bh(jh_in
)->b_size
);
432 kunmap_atomic(mapped_data
);
434 new_page
= virt_to_page(tmp
);
435 new_offset
= offset_in_page(tmp
);
439 * This isn't strictly necessary, as we're using frozen
440 * data for the escaping, but it keeps consistency with
441 * b_frozen_data usage.
443 jh_in
->b_frozen_triggers
= jh_in
->b_triggers
;
447 * Did we need to do an escaping? Now we've done all the
448 * copying, we can finally do so.
451 mapped_data
= kmap_atomic(new_page
);
452 *((unsigned int *)(mapped_data
+ new_offset
)) = 0;
453 kunmap_atomic(mapped_data
);
456 set_bh_page(new_bh
, new_page
, new_offset
);
457 new_jh
->b_transaction
= NULL
;
458 new_bh
->b_size
= jh2bh(jh_in
)->b_size
;
459 new_bh
->b_bdev
= transaction
->t_journal
->j_dev
;
460 new_bh
->b_blocknr
= blocknr
;
461 set_buffer_mapped(new_bh
);
462 set_buffer_dirty(new_bh
);
467 * The to-be-written buffer needs to get moved to the io queue,
468 * and the original buffer whose contents we are shadowing or
469 * copying is moved to the transaction's shadow queue.
471 JBUFFER_TRACE(jh_in
, "file as BJ_Shadow");
472 spin_lock(&journal
->j_list_lock
);
473 __jbd2_journal_file_buffer(jh_in
, transaction
, BJ_Shadow
);
474 spin_unlock(&journal
->j_list_lock
);
475 jbd_unlock_bh_state(bh_in
);
477 JBUFFER_TRACE(new_jh
, "file as BJ_IO");
478 jbd2_journal_file_buffer(new_jh
, transaction
, BJ_IO
);
480 return do_escape
| (done_copy_out
<< 1);
484 * Allocation code for the journal file. Manage the space left in the
485 * journal, so that we can begin checkpointing when appropriate.
489 * __jbd2_log_space_left: Return the number of free blocks left in the journal.
491 * Called with the journal already locked.
493 * Called under j_state_lock
496 int __jbd2_log_space_left(journal_t
*journal
)
498 int left
= journal
->j_free
;
500 /* assert_spin_locked(&journal->j_state_lock); */
503 * Be pessimistic here about the number of those free blocks which
504 * might be required for log descriptor control blocks.
507 #define MIN_LOG_RESERVED_BLOCKS 32 /* Allow for rounding errors */
509 left
-= MIN_LOG_RESERVED_BLOCKS
;
518 * Called with j_state_lock locked for writing.
519 * Returns true if a transaction commit was started.
521 int __jbd2_log_start_commit(journal_t
*journal
, tid_t target
)
523 /* Return if the txn has already requested to be committed */
524 if (journal
->j_commit_request
== target
)
528 * The only transaction we can possibly wait upon is the
529 * currently running transaction (if it exists). Otherwise,
530 * the target tid must be an old one.
532 if (journal
->j_running_transaction
&&
533 journal
->j_running_transaction
->t_tid
== target
) {
535 * We want a new commit: OK, mark the request and wakeup the
536 * commit thread. We do _not_ do the commit ourselves.
539 journal
->j_commit_request
= target
;
540 jbd_debug(1, "JBD2: requesting commit %d/%d\n",
541 journal
->j_commit_request
,
542 journal
->j_commit_sequence
);
543 journal
->j_running_transaction
->t_requested
= jiffies
;
544 wake_up(&journal
->j_wait_commit
);
546 } else if (!tid_geq(journal
->j_commit_request
, target
))
547 /* This should never happen, but if it does, preserve
548 the evidence before kjournald goes into a loop and
549 increments j_commit_sequence beyond all recognition. */
550 WARN_ONCE(1, "JBD2: bad log_start_commit: %u %u %u %u\n",
551 journal
->j_commit_request
,
552 journal
->j_commit_sequence
,
553 target
, journal
->j_running_transaction
?
554 journal
->j_running_transaction
->t_tid
: 0);
558 int jbd2_log_start_commit(journal_t
*journal
, tid_t tid
)
562 write_lock(&journal
->j_state_lock
);
563 ret
= __jbd2_log_start_commit(journal
, tid
);
564 write_unlock(&journal
->j_state_lock
);
569 * Force and wait upon a commit if the calling process is not within
570 * transaction. This is used for forcing out undo-protected data which contains
571 * bitmaps, when the fs is running out of space.
573 * We can only force the running transaction if we don't have an active handle;
574 * otherwise, we will deadlock.
576 * Returns true if a transaction was started.
578 int jbd2_journal_force_commit_nested(journal_t
*journal
)
580 transaction_t
*transaction
= NULL
;
582 int need_to_start
= 0;
584 read_lock(&journal
->j_state_lock
);
585 if (journal
->j_running_transaction
&& !current
->journal_info
) {
586 transaction
= journal
->j_running_transaction
;
587 if (!tid_geq(journal
->j_commit_request
, transaction
->t_tid
))
589 } else if (journal
->j_committing_transaction
)
590 transaction
= journal
->j_committing_transaction
;
593 read_unlock(&journal
->j_state_lock
);
594 return 0; /* Nothing to retry */
597 tid
= transaction
->t_tid
;
598 read_unlock(&journal
->j_state_lock
);
600 jbd2_log_start_commit(journal
, tid
);
601 jbd2_log_wait_commit(journal
, tid
);
606 * Start a commit of the current running transaction (if any). Returns true
607 * if a transaction is going to be committed (or is currently already
608 * committing), and fills its tid in at *ptid
610 int jbd2_journal_start_commit(journal_t
*journal
, tid_t
*ptid
)
614 write_lock(&journal
->j_state_lock
);
615 if (journal
->j_running_transaction
) {
616 tid_t tid
= journal
->j_running_transaction
->t_tid
;
618 __jbd2_log_start_commit(journal
, tid
);
619 /* There's a running transaction and we've just made sure
620 * it's commit has been scheduled. */
624 } else if (journal
->j_committing_transaction
) {
626 * If commit has been started, then we have to wait for
627 * completion of that transaction.
630 *ptid
= journal
->j_committing_transaction
->t_tid
;
633 write_unlock(&journal
->j_state_lock
);
638 * Return 1 if a given transaction has not yet sent barrier request
639 * connected with a transaction commit. If 0 is returned, transaction
640 * may or may not have sent the barrier. Used to avoid sending barrier
641 * twice in common cases.
643 int jbd2_trans_will_send_data_barrier(journal_t
*journal
, tid_t tid
)
646 transaction_t
*commit_trans
;
648 if (!(journal
->j_flags
& JBD2_BARRIER
))
650 read_lock(&journal
->j_state_lock
);
651 /* Transaction already committed? */
652 if (tid_geq(journal
->j_commit_sequence
, tid
))
654 commit_trans
= journal
->j_committing_transaction
;
655 if (!commit_trans
|| commit_trans
->t_tid
!= tid
) {
660 * Transaction is being committed and we already proceeded to
661 * submitting a flush to fs partition?
663 if (journal
->j_fs_dev
!= journal
->j_dev
) {
664 if (!commit_trans
->t_need_data_flush
||
665 commit_trans
->t_state
>= T_COMMIT_DFLUSH
)
668 if (commit_trans
->t_state
>= T_COMMIT_JFLUSH
)
673 read_unlock(&journal
->j_state_lock
);
676 EXPORT_SYMBOL(jbd2_trans_will_send_data_barrier
);
679 * Wait for a specified commit to complete.
680 * The caller may not hold the journal lock.
682 int jbd2_log_wait_commit(journal_t
*journal
, tid_t tid
)
686 read_lock(&journal
->j_state_lock
);
687 #ifdef CONFIG_JBD2_DEBUG
688 if (!tid_geq(journal
->j_commit_request
, tid
)) {
690 "%s: error: j_commit_request=%d, tid=%d\n",
691 __func__
, journal
->j_commit_request
, tid
);
694 while (tid_gt(tid
, journal
->j_commit_sequence
)) {
695 jbd_debug(1, "JBD2: want %d, j_commit_sequence=%d\n",
696 tid
, journal
->j_commit_sequence
);
697 wake_up(&journal
->j_wait_commit
);
698 read_unlock(&journal
->j_state_lock
);
699 wait_event(journal
->j_wait_done_commit
,
700 !tid_gt(tid
, journal
->j_commit_sequence
));
701 read_lock(&journal
->j_state_lock
);
703 read_unlock(&journal
->j_state_lock
);
705 if (unlikely(is_journal_aborted(journal
))) {
706 printk(KERN_EMERG
"journal commit I/O error\n");
713 * When this function returns the transaction corresponding to tid
714 * will be completed. If the transaction has currently running, start
715 * committing that transaction before waiting for it to complete. If
716 * the transaction id is stale, it is by definition already completed,
717 * so just return SUCCESS.
719 int jbd2_complete_transaction(journal_t
*journal
, tid_t tid
)
721 int need_to_wait
= 1;
723 read_lock(&journal
->j_state_lock
);
724 if (journal
->j_running_transaction
&&
725 journal
->j_running_transaction
->t_tid
== tid
) {
726 if (journal
->j_commit_request
!= tid
) {
727 /* transaction not yet started, so request it */
728 read_unlock(&journal
->j_state_lock
);
729 jbd2_log_start_commit(journal
, tid
);
732 } else if (!(journal
->j_committing_transaction
&&
733 journal
->j_committing_transaction
->t_tid
== tid
))
735 read_unlock(&journal
->j_state_lock
);
739 return jbd2_log_wait_commit(journal
, tid
);
741 EXPORT_SYMBOL(jbd2_complete_transaction
);
744 * Log buffer allocation routines:
747 int jbd2_journal_next_log_block(journal_t
*journal
, unsigned long long *retp
)
749 unsigned long blocknr
;
751 write_lock(&journal
->j_state_lock
);
752 J_ASSERT(journal
->j_free
> 1);
754 blocknr
= journal
->j_head
;
757 if (journal
->j_head
== journal
->j_last
)
758 journal
->j_head
= journal
->j_first
;
759 write_unlock(&journal
->j_state_lock
);
760 return jbd2_journal_bmap(journal
, blocknr
, retp
);
764 * Conversion of logical to physical block numbers for the journal
766 * On external journals the journal blocks are identity-mapped, so
767 * this is a no-op. If needed, we can use j_blk_offset - everything is
770 int jbd2_journal_bmap(journal_t
*journal
, unsigned long blocknr
,
771 unsigned long long *retp
)
774 unsigned long long ret
;
776 if (journal
->j_inode
) {
777 ret
= bmap(journal
->j_inode
, blocknr
);
781 printk(KERN_ALERT
"%s: journal block not found "
782 "at offset %lu on %s\n",
783 __func__
, blocknr
, journal
->j_devname
);
785 __journal_abort_soft(journal
, err
);
788 *retp
= blocknr
; /* +journal->j_blk_offset */
794 * We play buffer_head aliasing tricks to write data/metadata blocks to
795 * the journal without copying their contents, but for journal
796 * descriptor blocks we do need to generate bona fide buffers.
798 * After the caller of jbd2_journal_get_descriptor_buffer() has finished modifying
799 * the buffer's contents they really should run flush_dcache_page(bh->b_page).
800 * But we don't bother doing that, so there will be coherency problems with
801 * mmaps of blockdevs which hold live JBD-controlled filesystems.
803 struct journal_head
*jbd2_journal_get_descriptor_buffer(journal_t
*journal
)
805 struct buffer_head
*bh
;
806 unsigned long long blocknr
;
809 err
= jbd2_journal_next_log_block(journal
, &blocknr
);
814 bh
= __getblk(journal
->j_dev
, blocknr
, journal
->j_blocksize
);
818 memset(bh
->b_data
, 0, journal
->j_blocksize
);
819 set_buffer_uptodate(bh
);
821 BUFFER_TRACE(bh
, "return this buffer");
822 return jbd2_journal_add_journal_head(bh
);
826 * Return tid of the oldest transaction in the journal and block in the journal
827 * where the transaction starts.
829 * If the journal is now empty, return which will be the next transaction ID
830 * we will write and where will that transaction start.
832 * The return value is 0 if journal tail cannot be pushed any further, 1 if
835 int jbd2_journal_get_log_tail(journal_t
*journal
, tid_t
*tid
,
836 unsigned long *block
)
838 transaction_t
*transaction
;
841 read_lock(&journal
->j_state_lock
);
842 spin_lock(&journal
->j_list_lock
);
843 transaction
= journal
->j_checkpoint_transactions
;
845 *tid
= transaction
->t_tid
;
846 *block
= transaction
->t_log_start
;
847 } else if ((transaction
= journal
->j_committing_transaction
) != NULL
) {
848 *tid
= transaction
->t_tid
;
849 *block
= transaction
->t_log_start
;
850 } else if ((transaction
= journal
->j_running_transaction
) != NULL
) {
851 *tid
= transaction
->t_tid
;
852 *block
= journal
->j_head
;
854 *tid
= journal
->j_transaction_sequence
;
855 *block
= journal
->j_head
;
857 ret
= tid_gt(*tid
, journal
->j_tail_sequence
);
858 spin_unlock(&journal
->j_list_lock
);
859 read_unlock(&journal
->j_state_lock
);
865 * Update information in journal structure and in on disk journal superblock
866 * about log tail. This function does not check whether information passed in
867 * really pushes log tail further. It's responsibility of the caller to make
868 * sure provided log tail information is valid (e.g. by holding
869 * j_checkpoint_mutex all the time between computing log tail and calling this
870 * function as is the case with jbd2_cleanup_journal_tail()).
872 * Requires j_checkpoint_mutex
874 void __jbd2_update_log_tail(journal_t
*journal
, tid_t tid
, unsigned long block
)
878 BUG_ON(!mutex_is_locked(&journal
->j_checkpoint_mutex
));
881 * We cannot afford for write to remain in drive's caches since as
882 * soon as we update j_tail, next transaction can start reusing journal
883 * space and if we lose sb update during power failure we'd replay
884 * old transaction with possibly newly overwritten data.
886 jbd2_journal_update_sb_log_tail(journal
, tid
, block
, WRITE_FUA
);
887 write_lock(&journal
->j_state_lock
);
888 freed
= block
- journal
->j_tail
;
889 if (block
< journal
->j_tail
)
890 freed
+= journal
->j_last
- journal
->j_first
;
892 trace_jbd2_update_log_tail(journal
, tid
, block
, freed
);
894 "Cleaning journal tail from %d to %d (offset %lu), "
896 journal
->j_tail_sequence
, tid
, block
, freed
);
898 journal
->j_free
+= freed
;
899 journal
->j_tail_sequence
= tid
;
900 journal
->j_tail
= block
;
901 write_unlock(&journal
->j_state_lock
);
905 * This is a variaon of __jbd2_update_log_tail which checks for validity of
906 * provided log tail and locks j_checkpoint_mutex. So it is safe against races
907 * with other threads updating log tail.
909 void jbd2_update_log_tail(journal_t
*journal
, tid_t tid
, unsigned long block
)
911 mutex_lock(&journal
->j_checkpoint_mutex
);
912 if (tid_gt(tid
, journal
->j_tail_sequence
))
913 __jbd2_update_log_tail(journal
, tid
, block
);
914 mutex_unlock(&journal
->j_checkpoint_mutex
);
917 struct jbd2_stats_proc_session
{
919 struct transaction_stats_s
*stats
;
924 static void *jbd2_seq_info_start(struct seq_file
*seq
, loff_t
*pos
)
926 return *pos
? NULL
: SEQ_START_TOKEN
;
929 static void *jbd2_seq_info_next(struct seq_file
*seq
, void *v
, loff_t
*pos
)
934 static int jbd2_seq_info_show(struct seq_file
*seq
, void *v
)
936 struct jbd2_stats_proc_session
*s
= seq
->private;
938 if (v
!= SEQ_START_TOKEN
)
940 seq_printf(seq
, "%lu transactions (%lu requested), "
941 "each up to %u blocks\n",
942 s
->stats
->ts_tid
, s
->stats
->ts_requested
,
943 s
->journal
->j_max_transaction_buffers
);
944 if (s
->stats
->ts_tid
== 0)
946 seq_printf(seq
, "average: \n %ums waiting for transaction\n",
947 jiffies_to_msecs(s
->stats
->run
.rs_wait
/ s
->stats
->ts_tid
));
948 seq_printf(seq
, " %ums request delay\n",
949 (s
->stats
->ts_requested
== 0) ? 0 :
950 jiffies_to_msecs(s
->stats
->run
.rs_request_delay
/
951 s
->stats
->ts_requested
));
952 seq_printf(seq
, " %ums running transaction\n",
953 jiffies_to_msecs(s
->stats
->run
.rs_running
/ s
->stats
->ts_tid
));
954 seq_printf(seq
, " %ums transaction was being locked\n",
955 jiffies_to_msecs(s
->stats
->run
.rs_locked
/ s
->stats
->ts_tid
));
956 seq_printf(seq
, " %ums flushing data (in ordered mode)\n",
957 jiffies_to_msecs(s
->stats
->run
.rs_flushing
/ s
->stats
->ts_tid
));
958 seq_printf(seq
, " %ums logging transaction\n",
959 jiffies_to_msecs(s
->stats
->run
.rs_logging
/ s
->stats
->ts_tid
));
960 seq_printf(seq
, " %lluus average transaction commit time\n",
961 div_u64(s
->journal
->j_average_commit_time
, 1000));
962 seq_printf(seq
, " %lu handles per transaction\n",
963 s
->stats
->run
.rs_handle_count
/ s
->stats
->ts_tid
);
964 seq_printf(seq
, " %lu blocks per transaction\n",
965 s
->stats
->run
.rs_blocks
/ s
->stats
->ts_tid
);
966 seq_printf(seq
, " %lu logged blocks per transaction\n",
967 s
->stats
->run
.rs_blocks_logged
/ s
->stats
->ts_tid
);
971 static void jbd2_seq_info_stop(struct seq_file
*seq
, void *v
)
975 static const struct seq_operations jbd2_seq_info_ops
= {
976 .start
= jbd2_seq_info_start
,
977 .next
= jbd2_seq_info_next
,
978 .stop
= jbd2_seq_info_stop
,
979 .show
= jbd2_seq_info_show
,
982 static int jbd2_seq_info_open(struct inode
*inode
, struct file
*file
)
984 journal_t
*journal
= PDE(inode
)->data
;
985 struct jbd2_stats_proc_session
*s
;
988 s
= kmalloc(sizeof(*s
), GFP_KERNEL
);
991 size
= sizeof(struct transaction_stats_s
);
992 s
->stats
= kmalloc(size
, GFP_KERNEL
);
993 if (s
->stats
== NULL
) {
997 spin_lock(&journal
->j_history_lock
);
998 memcpy(s
->stats
, &journal
->j_stats
, size
);
999 s
->journal
= journal
;
1000 spin_unlock(&journal
->j_history_lock
);
1002 rc
= seq_open(file
, &jbd2_seq_info_ops
);
1004 struct seq_file
*m
= file
->private_data
;
1014 static int jbd2_seq_info_release(struct inode
*inode
, struct file
*file
)
1016 struct seq_file
*seq
= file
->private_data
;
1017 struct jbd2_stats_proc_session
*s
= seq
->private;
1020 return seq_release(inode
, file
);
1023 static const struct file_operations jbd2_seq_info_fops
= {
1024 .owner
= THIS_MODULE
,
1025 .open
= jbd2_seq_info_open
,
1027 .llseek
= seq_lseek
,
1028 .release
= jbd2_seq_info_release
,
1031 static struct proc_dir_entry
*proc_jbd2_stats
;
1033 static void jbd2_stats_proc_init(journal_t
*journal
)
1035 journal
->j_proc_entry
= proc_mkdir(journal
->j_devname
, proc_jbd2_stats
);
1036 if (journal
->j_proc_entry
) {
1037 proc_create_data("info", S_IRUGO
, journal
->j_proc_entry
,
1038 &jbd2_seq_info_fops
, journal
);
1042 static void jbd2_stats_proc_exit(journal_t
*journal
)
1044 remove_proc_entry("info", journal
->j_proc_entry
);
1045 remove_proc_entry(journal
->j_devname
, proc_jbd2_stats
);
1049 * Management for journal control blocks: functions to create and
1050 * destroy journal_t structures, and to initialise and read existing
1051 * journal blocks from disk. */
1053 /* First: create and setup a journal_t object in memory. We initialise
1054 * very few fields yet: that has to wait until we have created the
1055 * journal structures from from scratch, or loaded them from disk. */
1057 static journal_t
* journal_init_common (void)
1062 journal
= kzalloc(sizeof(*journal
), GFP_KERNEL
);
1066 init_waitqueue_head(&journal
->j_wait_transaction_locked
);
1067 init_waitqueue_head(&journal
->j_wait_logspace
);
1068 init_waitqueue_head(&journal
->j_wait_done_commit
);
1069 init_waitqueue_head(&journal
->j_wait_checkpoint
);
1070 init_waitqueue_head(&journal
->j_wait_commit
);
1071 init_waitqueue_head(&journal
->j_wait_updates
);
1072 mutex_init(&journal
->j_barrier
);
1073 mutex_init(&journal
->j_checkpoint_mutex
);
1074 spin_lock_init(&journal
->j_revoke_lock
);
1075 spin_lock_init(&journal
->j_list_lock
);
1076 rwlock_init(&journal
->j_state_lock
);
1078 journal
->j_commit_interval
= (HZ
* JBD2_DEFAULT_MAX_COMMIT_AGE
);
1079 journal
->j_min_batch_time
= 0;
1080 journal
->j_max_batch_time
= 15000; /* 15ms */
1082 /* The journal is marked for error until we succeed with recovery! */
1083 journal
->j_flags
= JBD2_ABORT
;
1085 /* Set up a default-sized revoke table for the new mount. */
1086 err
= jbd2_journal_init_revoke(journal
, JOURNAL_REVOKE_DEFAULT_HASH
);
1092 spin_lock_init(&journal
->j_history_lock
);
1097 /* jbd2_journal_init_dev and jbd2_journal_init_inode:
1099 * Create a journal structure assigned some fixed set of disk blocks to
1100 * the journal. We don't actually touch those disk blocks yet, but we
1101 * need to set up all of the mapping information to tell the journaling
1102 * system where the journal blocks are.
1107 * journal_t * jbd2_journal_init_dev() - creates and initialises a journal structure
1108 * @bdev: Block device on which to create the journal
1109 * @fs_dev: Device which hold journalled filesystem for this journal.
1110 * @start: Block nr Start of journal.
1111 * @len: Length of the journal in blocks.
1112 * @blocksize: blocksize of journalling device
1114 * Returns: a newly created journal_t *
1116 * jbd2_journal_init_dev creates a journal which maps a fixed contiguous
1117 * range of blocks on an arbitrary block device.
1120 journal_t
* jbd2_journal_init_dev(struct block_device
*bdev
,
1121 struct block_device
*fs_dev
,
1122 unsigned long long start
, int len
, int blocksize
)
1124 journal_t
*journal
= journal_init_common();
1125 struct buffer_head
*bh
;
1132 /* journal descriptor can store up to n blocks -bzzz */
1133 journal
->j_blocksize
= blocksize
;
1134 journal
->j_dev
= bdev
;
1135 journal
->j_fs_dev
= fs_dev
;
1136 journal
->j_blk_offset
= start
;
1137 journal
->j_maxlen
= len
;
1138 bdevname(journal
->j_dev
, journal
->j_devname
);
1139 p
= journal
->j_devname
;
1140 while ((p
= strchr(p
, '/')))
1142 jbd2_stats_proc_init(journal
);
1143 n
= journal
->j_blocksize
/ sizeof(journal_block_tag_t
);
1144 journal
->j_wbufsize
= n
;
1145 journal
->j_wbuf
= kmalloc(n
* sizeof(struct buffer_head
*), GFP_KERNEL
);
1146 if (!journal
->j_wbuf
) {
1147 printk(KERN_ERR
"%s: Can't allocate bhs for commit thread\n",
1152 bh
= __getblk(journal
->j_dev
, start
, journal
->j_blocksize
);
1155 "%s: Cannot get buffer for journal superblock\n",
1159 journal
->j_sb_buffer
= bh
;
1160 journal
->j_superblock
= (journal_superblock_t
*)bh
->b_data
;
1164 kfree(journal
->j_wbuf
);
1165 jbd2_stats_proc_exit(journal
);
1171 * journal_t * jbd2_journal_init_inode () - creates a journal which maps to a inode.
1172 * @inode: An inode to create the journal in
1174 * jbd2_journal_init_inode creates a journal which maps an on-disk inode as
1175 * the journal. The inode must exist already, must support bmap() and
1176 * must have all data blocks preallocated.
1178 journal_t
* jbd2_journal_init_inode (struct inode
*inode
)
1180 struct buffer_head
*bh
;
1181 journal_t
*journal
= journal_init_common();
1185 unsigned long long blocknr
;
1190 journal
->j_dev
= journal
->j_fs_dev
= inode
->i_sb
->s_bdev
;
1191 journal
->j_inode
= inode
;
1192 bdevname(journal
->j_dev
, journal
->j_devname
);
1193 p
= journal
->j_devname
;
1194 while ((p
= strchr(p
, '/')))
1196 p
= journal
->j_devname
+ strlen(journal
->j_devname
);
1197 sprintf(p
, "-%lu", journal
->j_inode
->i_ino
);
1199 "journal %p: inode %s/%ld, size %Ld, bits %d, blksize %ld\n",
1200 journal
, inode
->i_sb
->s_id
, inode
->i_ino
,
1201 (long long) inode
->i_size
,
1202 inode
->i_sb
->s_blocksize_bits
, inode
->i_sb
->s_blocksize
);
1204 journal
->j_maxlen
= inode
->i_size
>> inode
->i_sb
->s_blocksize_bits
;
1205 journal
->j_blocksize
= inode
->i_sb
->s_blocksize
;
1206 jbd2_stats_proc_init(journal
);
1208 /* journal descriptor can store up to n blocks -bzzz */
1209 n
= journal
->j_blocksize
/ sizeof(journal_block_tag_t
);
1210 journal
->j_wbufsize
= n
;
1211 journal
->j_wbuf
= kmalloc(n
* sizeof(struct buffer_head
*), GFP_KERNEL
);
1212 if (!journal
->j_wbuf
) {
1213 printk(KERN_ERR
"%s: Can't allocate bhs for commit thread\n",
1218 err
= jbd2_journal_bmap(journal
, 0, &blocknr
);
1219 /* If that failed, give up */
1221 printk(KERN_ERR
"%s: Cannot locate journal superblock\n",
1226 bh
= __getblk(journal
->j_dev
, blocknr
, journal
->j_blocksize
);
1229 "%s: Cannot get buffer for journal superblock\n",
1233 journal
->j_sb_buffer
= bh
;
1234 journal
->j_superblock
= (journal_superblock_t
*)bh
->b_data
;
1238 kfree(journal
->j_wbuf
);
1239 jbd2_stats_proc_exit(journal
);
1245 * If the journal init or create aborts, we need to mark the journal
1246 * superblock as being NULL to prevent the journal destroy from writing
1247 * back a bogus superblock.
1249 static void journal_fail_superblock (journal_t
*journal
)
1251 struct buffer_head
*bh
= journal
->j_sb_buffer
;
1253 journal
->j_sb_buffer
= NULL
;
1257 * Given a journal_t structure, initialise the various fields for
1258 * startup of a new journaling session. We use this both when creating
1259 * a journal, and after recovering an old journal to reset it for
1263 static int journal_reset(journal_t
*journal
)
1265 journal_superblock_t
*sb
= journal
->j_superblock
;
1266 unsigned long long first
, last
;
1268 first
= be32_to_cpu(sb
->s_first
);
1269 last
= be32_to_cpu(sb
->s_maxlen
);
1270 if (first
+ JBD2_MIN_JOURNAL_BLOCKS
> last
+ 1) {
1271 printk(KERN_ERR
"JBD2: Journal too short (blocks %llu-%llu).\n",
1273 journal_fail_superblock(journal
);
1277 journal
->j_first
= first
;
1278 journal
->j_last
= last
;
1280 journal
->j_head
= first
;
1281 journal
->j_tail
= first
;
1282 journal
->j_free
= last
- first
;
1284 journal
->j_tail_sequence
= journal
->j_transaction_sequence
;
1285 journal
->j_commit_sequence
= journal
->j_transaction_sequence
- 1;
1286 journal
->j_commit_request
= journal
->j_commit_sequence
;
1288 journal
->j_max_transaction_buffers
= journal
->j_maxlen
/ 4;
1291 * As a special case, if the on-disk copy is already marked as needing
1292 * no recovery (s_start == 0), then we can safely defer the superblock
1293 * update until the next commit by setting JBD2_FLUSHED. This avoids
1294 * attempting a write to a potential-readonly device.
1296 if (sb
->s_start
== 0) {
1297 jbd_debug(1, "JBD2: Skipping superblock update on recovered sb "
1298 "(start %ld, seq %d, errno %d)\n",
1299 journal
->j_tail
, journal
->j_tail_sequence
,
1301 journal
->j_flags
|= JBD2_FLUSHED
;
1303 /* Lock here to make assertions happy... */
1304 mutex_lock(&journal
->j_checkpoint_mutex
);
1306 * Update log tail information. We use WRITE_FUA since new
1307 * transaction will start reusing journal space and so we
1308 * must make sure information about current log tail is on
1311 jbd2_journal_update_sb_log_tail(journal
,
1312 journal
->j_tail_sequence
,
1315 mutex_unlock(&journal
->j_checkpoint_mutex
);
1317 return jbd2_journal_start_thread(journal
);
1320 static void jbd2_write_superblock(journal_t
*journal
, int write_op
)
1322 struct buffer_head
*bh
= journal
->j_sb_buffer
;
1325 trace_jbd2_write_superblock(journal
, write_op
);
1326 if (!(journal
->j_flags
& JBD2_BARRIER
))
1327 write_op
&= ~(REQ_FUA
| REQ_FLUSH
);
1329 if (buffer_write_io_error(bh
)) {
1331 * Oh, dear. A previous attempt to write the journal
1332 * superblock failed. This could happen because the
1333 * USB device was yanked out. Or it could happen to
1334 * be a transient write error and maybe the block will
1335 * be remapped. Nothing we can do but to retry the
1336 * write and hope for the best.
1338 printk(KERN_ERR
"JBD2: previous I/O error detected "
1339 "for journal superblock update for %s.\n",
1340 journal
->j_devname
);
1341 clear_buffer_write_io_error(bh
);
1342 set_buffer_uptodate(bh
);
1345 bh
->b_end_io
= end_buffer_write_sync
;
1346 ret
= submit_bh(write_op
, bh
);
1348 if (buffer_write_io_error(bh
)) {
1349 clear_buffer_write_io_error(bh
);
1350 set_buffer_uptodate(bh
);
1354 printk(KERN_ERR
"JBD2: Error %d detected when updating "
1355 "journal superblock for %s.\n", ret
,
1356 journal
->j_devname
);
1361 * jbd2_journal_update_sb_log_tail() - Update log tail in journal sb on disk.
1362 * @journal: The journal to update.
1363 * @tail_tid: TID of the new transaction at the tail of the log
1364 * @tail_block: The first block of the transaction at the tail of the log
1365 * @write_op: With which operation should we write the journal sb
1367 * Update a journal's superblock information about log tail and write it to
1368 * disk, waiting for the IO to complete.
1370 void jbd2_journal_update_sb_log_tail(journal_t
*journal
, tid_t tail_tid
,
1371 unsigned long tail_block
, int write_op
)
1373 journal_superblock_t
*sb
= journal
->j_superblock
;
1375 BUG_ON(!mutex_is_locked(&journal
->j_checkpoint_mutex
));
1376 jbd_debug(1, "JBD2: updating superblock (start %lu, seq %u)\n",
1377 tail_block
, tail_tid
);
1379 sb
->s_sequence
= cpu_to_be32(tail_tid
);
1380 sb
->s_start
= cpu_to_be32(tail_block
);
1382 jbd2_write_superblock(journal
, write_op
);
1384 /* Log is no longer empty */
1385 write_lock(&journal
->j_state_lock
);
1386 WARN_ON(!sb
->s_sequence
);
1387 journal
->j_flags
&= ~JBD2_FLUSHED
;
1388 write_unlock(&journal
->j_state_lock
);
1392 * jbd2_mark_journal_empty() - Mark on disk journal as empty.
1393 * @journal: The journal to update.
1395 * Update a journal's dynamic superblock fields to show that journal is empty.
1396 * Write updated superblock to disk waiting for IO to complete.
1398 static void jbd2_mark_journal_empty(journal_t
*journal
)
1400 journal_superblock_t
*sb
= journal
->j_superblock
;
1402 BUG_ON(!mutex_is_locked(&journal
->j_checkpoint_mutex
));
1403 read_lock(&journal
->j_state_lock
);
1404 /* Is it already empty? */
1405 if (sb
->s_start
== 0) {
1406 read_unlock(&journal
->j_state_lock
);
1409 jbd_debug(1, "JBD2: Marking journal as empty (seq %d)\n",
1410 journal
->j_tail_sequence
);
1412 sb
->s_sequence
= cpu_to_be32(journal
->j_tail_sequence
);
1413 sb
->s_start
= cpu_to_be32(0);
1414 read_unlock(&journal
->j_state_lock
);
1416 jbd2_write_superblock(journal
, WRITE_FUA
);
1418 /* Log is no longer empty */
1419 write_lock(&journal
->j_state_lock
);
1420 journal
->j_flags
|= JBD2_FLUSHED
;
1421 write_unlock(&journal
->j_state_lock
);
1426 * jbd2_journal_update_sb_errno() - Update error in the journal.
1427 * @journal: The journal to update.
1429 * Update a journal's errno. Write updated superblock to disk waiting for IO
1432 void jbd2_journal_update_sb_errno(journal_t
*journal
)
1434 journal_superblock_t
*sb
= journal
->j_superblock
;
1436 read_lock(&journal
->j_state_lock
);
1437 jbd_debug(1, "JBD2: updating superblock error (errno %d)\n",
1439 sb
->s_errno
= cpu_to_be32(journal
->j_errno
);
1440 jbd2_superblock_csum_set(journal
, sb
);
1441 read_unlock(&journal
->j_state_lock
);
1443 jbd2_write_superblock(journal
, WRITE_SYNC
);
1445 EXPORT_SYMBOL(jbd2_journal_update_sb_errno
);
1448 * Read the superblock for a given journal, performing initial
1449 * validation of the format.
1451 static int journal_get_superblock(journal_t
*journal
)
1453 struct buffer_head
*bh
;
1454 journal_superblock_t
*sb
;
1457 bh
= journal
->j_sb_buffer
;
1459 J_ASSERT(bh
!= NULL
);
1460 if (!buffer_uptodate(bh
)) {
1461 ll_rw_block(READ
, 1, &bh
);
1463 if (!buffer_uptodate(bh
)) {
1465 "JBD2: IO error reading journal superblock\n");
1470 if (buffer_verified(bh
))
1473 sb
= journal
->j_superblock
;
1477 if (sb
->s_header
.h_magic
!= cpu_to_be32(JBD2_MAGIC_NUMBER
) ||
1478 sb
->s_blocksize
!= cpu_to_be32(journal
->j_blocksize
)) {
1479 printk(KERN_WARNING
"JBD2: no valid journal superblock found\n");
1483 switch(be32_to_cpu(sb
->s_header
.h_blocktype
)) {
1484 case JBD2_SUPERBLOCK_V1
:
1485 journal
->j_format_version
= 1;
1487 case JBD2_SUPERBLOCK_V2
:
1488 journal
->j_format_version
= 2;
1491 printk(KERN_WARNING
"JBD2: unrecognised superblock format ID\n");
1495 if (be32_to_cpu(sb
->s_maxlen
) < journal
->j_maxlen
)
1496 journal
->j_maxlen
= be32_to_cpu(sb
->s_maxlen
);
1497 else if (be32_to_cpu(sb
->s_maxlen
) > journal
->j_maxlen
) {
1498 printk(KERN_WARNING
"JBD2: journal file too short\n");
1502 if (be32_to_cpu(sb
->s_first
) == 0 ||
1503 be32_to_cpu(sb
->s_first
) >= journal
->j_maxlen
) {
1505 "JBD2: Invalid start block of journal: %u\n",
1506 be32_to_cpu(sb
->s_first
));
1510 if (JBD2_HAS_COMPAT_FEATURE(journal
, JBD2_FEATURE_COMPAT_CHECKSUM
) &&
1511 JBD2_HAS_INCOMPAT_FEATURE(journal
, JBD2_FEATURE_INCOMPAT_CSUM_V2
)) {
1512 /* Can't have checksum v1 and v2 on at the same time! */
1513 printk(KERN_ERR
"JBD: Can't enable checksumming v1 and v2 "
1514 "at the same time!\n");
1518 if (!jbd2_verify_csum_type(journal
, sb
)) {
1519 printk(KERN_ERR
"JBD: Unknown checksum type\n");
1523 /* Load the checksum driver */
1524 if (JBD2_HAS_INCOMPAT_FEATURE(journal
, JBD2_FEATURE_INCOMPAT_CSUM_V2
)) {
1525 journal
->j_chksum_driver
= crypto_alloc_shash("crc32c", 0, 0);
1526 if (IS_ERR(journal
->j_chksum_driver
)) {
1527 printk(KERN_ERR
"JBD: Cannot load crc32c driver.\n");
1528 err
= PTR_ERR(journal
->j_chksum_driver
);
1529 journal
->j_chksum_driver
= NULL
;
1534 /* Check superblock checksum */
1535 if (!jbd2_superblock_csum_verify(journal
, sb
)) {
1536 printk(KERN_ERR
"JBD: journal checksum error\n");
1540 /* Precompute checksum seed for all metadata */
1541 if (JBD2_HAS_INCOMPAT_FEATURE(journal
, JBD2_FEATURE_INCOMPAT_CSUM_V2
))
1542 journal
->j_csum_seed
= jbd2_chksum(journal
, ~0, sb
->s_uuid
,
1543 sizeof(sb
->s_uuid
));
1545 set_buffer_verified(bh
);
1550 journal_fail_superblock(journal
);
1555 * Load the on-disk journal superblock and read the key fields into the
1559 static int load_superblock(journal_t
*journal
)
1562 journal_superblock_t
*sb
;
1564 err
= journal_get_superblock(journal
);
1568 sb
= journal
->j_superblock
;
1570 journal
->j_tail_sequence
= be32_to_cpu(sb
->s_sequence
);
1571 journal
->j_tail
= be32_to_cpu(sb
->s_start
);
1572 journal
->j_first
= be32_to_cpu(sb
->s_first
);
1573 journal
->j_last
= be32_to_cpu(sb
->s_maxlen
);
1574 journal
->j_errno
= be32_to_cpu(sb
->s_errno
);
1581 * int jbd2_journal_load() - Read journal from disk.
1582 * @journal: Journal to act on.
1584 * Given a journal_t structure which tells us which disk blocks contain
1585 * a journal, read the journal from disk to initialise the in-memory
1588 int jbd2_journal_load(journal_t
*journal
)
1591 journal_superblock_t
*sb
;
1593 err
= load_superblock(journal
);
1597 sb
= journal
->j_superblock
;
1598 /* If this is a V2 superblock, then we have to check the
1599 * features flags on it. */
1601 if (journal
->j_format_version
>= 2) {
1602 if ((sb
->s_feature_ro_compat
&
1603 ~cpu_to_be32(JBD2_KNOWN_ROCOMPAT_FEATURES
)) ||
1604 (sb
->s_feature_incompat
&
1605 ~cpu_to_be32(JBD2_KNOWN_INCOMPAT_FEATURES
))) {
1607 "JBD2: Unrecognised features on journal\n");
1613 * Create a slab for this blocksize
1615 err
= jbd2_journal_create_slab(be32_to_cpu(sb
->s_blocksize
));
1619 /* Let the recovery code check whether it needs to recover any
1620 * data from the journal. */
1621 if (jbd2_journal_recover(journal
))
1622 goto recovery_error
;
1624 if (journal
->j_failed_commit
) {
1625 printk(KERN_ERR
"JBD2: journal transaction %u on %s "
1626 "is corrupt.\n", journal
->j_failed_commit
,
1627 journal
->j_devname
);
1631 /* OK, we've finished with the dynamic journal bits:
1632 * reinitialise the dynamic contents of the superblock in memory
1633 * and reset them on disk. */
1634 if (journal_reset(journal
))
1635 goto recovery_error
;
1637 journal
->j_flags
&= ~JBD2_ABORT
;
1638 journal
->j_flags
|= JBD2_LOADED
;
1642 printk(KERN_WARNING
"JBD2: recovery failed\n");
1647 * void jbd2_journal_destroy() - Release a journal_t structure.
1648 * @journal: Journal to act on.
1650 * Release a journal_t structure once it is no longer in use by the
1652 * Return <0 if we couldn't clean up the journal.
1654 int jbd2_journal_destroy(journal_t
*journal
)
1658 /* Wait for the commit thread to wake up and die. */
1659 journal_kill_thread(journal
);
1661 /* Force a final log commit */
1662 if (journal
->j_running_transaction
)
1663 jbd2_journal_commit_transaction(journal
);
1665 /* Force any old transactions to disk */
1667 /* Totally anal locking here... */
1668 spin_lock(&journal
->j_list_lock
);
1669 while (journal
->j_checkpoint_transactions
!= NULL
) {
1670 spin_unlock(&journal
->j_list_lock
);
1671 mutex_lock(&journal
->j_checkpoint_mutex
);
1672 jbd2_log_do_checkpoint(journal
);
1673 mutex_unlock(&journal
->j_checkpoint_mutex
);
1674 spin_lock(&journal
->j_list_lock
);
1677 J_ASSERT(journal
->j_running_transaction
== NULL
);
1678 J_ASSERT(journal
->j_committing_transaction
== NULL
);
1679 J_ASSERT(journal
->j_checkpoint_transactions
== NULL
);
1680 spin_unlock(&journal
->j_list_lock
);
1682 if (journal
->j_sb_buffer
) {
1683 if (!is_journal_aborted(journal
)) {
1684 mutex_lock(&journal
->j_checkpoint_mutex
);
1685 jbd2_mark_journal_empty(journal
);
1686 mutex_unlock(&journal
->j_checkpoint_mutex
);
1689 brelse(journal
->j_sb_buffer
);
1692 if (journal
->j_proc_entry
)
1693 jbd2_stats_proc_exit(journal
);
1694 if (journal
->j_inode
)
1695 iput(journal
->j_inode
);
1696 if (journal
->j_revoke
)
1697 jbd2_journal_destroy_revoke(journal
);
1698 if (journal
->j_chksum_driver
)
1699 crypto_free_shash(journal
->j_chksum_driver
);
1700 kfree(journal
->j_wbuf
);
1708 *int jbd2_journal_check_used_features () - Check if features specified are used.
1709 * @journal: Journal to check.
1710 * @compat: bitmask of compatible features
1711 * @ro: bitmask of features that force read-only mount
1712 * @incompat: bitmask of incompatible features
1714 * Check whether the journal uses all of a given set of
1715 * features. Return true (non-zero) if it does.
1718 int jbd2_journal_check_used_features (journal_t
*journal
, unsigned long compat
,
1719 unsigned long ro
, unsigned long incompat
)
1721 journal_superblock_t
*sb
;
1723 if (!compat
&& !ro
&& !incompat
)
1725 /* Load journal superblock if it is not loaded yet. */
1726 if (journal
->j_format_version
== 0 &&
1727 journal_get_superblock(journal
) != 0)
1729 if (journal
->j_format_version
== 1)
1732 sb
= journal
->j_superblock
;
1734 if (((be32_to_cpu(sb
->s_feature_compat
) & compat
) == compat
) &&
1735 ((be32_to_cpu(sb
->s_feature_ro_compat
) & ro
) == ro
) &&
1736 ((be32_to_cpu(sb
->s_feature_incompat
) & incompat
) == incompat
))
1743 * int jbd2_journal_check_available_features() - Check feature set in journalling layer
1744 * @journal: Journal to check.
1745 * @compat: bitmask of compatible features
1746 * @ro: bitmask of features that force read-only mount
1747 * @incompat: bitmask of incompatible features
1749 * Check whether the journaling code supports the use of
1750 * all of a given set of features on this journal. Return true
1751 * (non-zero) if it can. */
1753 int jbd2_journal_check_available_features (journal_t
*journal
, unsigned long compat
,
1754 unsigned long ro
, unsigned long incompat
)
1756 if (!compat
&& !ro
&& !incompat
)
1759 /* We can support any known requested features iff the
1760 * superblock is in version 2. Otherwise we fail to support any
1761 * extended sb features. */
1763 if (journal
->j_format_version
!= 2)
1766 if ((compat
& JBD2_KNOWN_COMPAT_FEATURES
) == compat
&&
1767 (ro
& JBD2_KNOWN_ROCOMPAT_FEATURES
) == ro
&&
1768 (incompat
& JBD2_KNOWN_INCOMPAT_FEATURES
) == incompat
)
1775 * int jbd2_journal_set_features () - Mark a given journal feature in the superblock
1776 * @journal: Journal to act on.
1777 * @compat: bitmask of compatible features
1778 * @ro: bitmask of features that force read-only mount
1779 * @incompat: bitmask of incompatible features
1781 * Mark a given journal feature as present on the
1782 * superblock. Returns true if the requested features could be set.
1786 int jbd2_journal_set_features (journal_t
*journal
, unsigned long compat
,
1787 unsigned long ro
, unsigned long incompat
)
1789 #define INCOMPAT_FEATURE_ON(f) \
1790 ((incompat & (f)) && !(sb->s_feature_incompat & cpu_to_be32(f)))
1791 #define COMPAT_FEATURE_ON(f) \
1792 ((compat & (f)) && !(sb->s_feature_compat & cpu_to_be32(f)))
1793 journal_superblock_t
*sb
;
1795 if (jbd2_journal_check_used_features(journal
, compat
, ro
, incompat
))
1798 if (!jbd2_journal_check_available_features(journal
, compat
, ro
, incompat
))
1801 /* Asking for checksumming v2 and v1? Only give them v2. */
1802 if (incompat
& JBD2_FEATURE_INCOMPAT_CSUM_V2
&&
1803 compat
& JBD2_FEATURE_COMPAT_CHECKSUM
)
1804 compat
&= ~JBD2_FEATURE_COMPAT_CHECKSUM
;
1806 jbd_debug(1, "Setting new features 0x%lx/0x%lx/0x%lx\n",
1807 compat
, ro
, incompat
);
1809 sb
= journal
->j_superblock
;
1811 /* If enabling v2 checksums, update superblock */
1812 if (INCOMPAT_FEATURE_ON(JBD2_FEATURE_INCOMPAT_CSUM_V2
)) {
1813 sb
->s_checksum_type
= JBD2_CRC32C_CHKSUM
;
1814 sb
->s_feature_compat
&=
1815 ~cpu_to_be32(JBD2_FEATURE_COMPAT_CHECKSUM
);
1817 /* Load the checksum driver */
1818 if (journal
->j_chksum_driver
== NULL
) {
1819 journal
->j_chksum_driver
= crypto_alloc_shash("crc32c",
1821 if (IS_ERR(journal
->j_chksum_driver
)) {
1822 printk(KERN_ERR
"JBD: Cannot load crc32c "
1824 journal
->j_chksum_driver
= NULL
;
1829 /* Precompute checksum seed for all metadata */
1830 if (JBD2_HAS_INCOMPAT_FEATURE(journal
,
1831 JBD2_FEATURE_INCOMPAT_CSUM_V2
))
1832 journal
->j_csum_seed
= jbd2_chksum(journal
, ~0,
1834 sizeof(sb
->s_uuid
));
1837 /* If enabling v1 checksums, downgrade superblock */
1838 if (COMPAT_FEATURE_ON(JBD2_FEATURE_COMPAT_CHECKSUM
))
1839 sb
->s_feature_incompat
&=
1840 ~cpu_to_be32(JBD2_FEATURE_INCOMPAT_CSUM_V2
);
1842 sb
->s_feature_compat
|= cpu_to_be32(compat
);
1843 sb
->s_feature_ro_compat
|= cpu_to_be32(ro
);
1844 sb
->s_feature_incompat
|= cpu_to_be32(incompat
);
1847 #undef COMPAT_FEATURE_ON
1848 #undef INCOMPAT_FEATURE_ON
1852 * jbd2_journal_clear_features () - Clear a given journal feature in the
1854 * @journal: Journal to act on.
1855 * @compat: bitmask of compatible features
1856 * @ro: bitmask of features that force read-only mount
1857 * @incompat: bitmask of incompatible features
1859 * Clear a given journal feature as present on the
1862 void jbd2_journal_clear_features(journal_t
*journal
, unsigned long compat
,
1863 unsigned long ro
, unsigned long incompat
)
1865 journal_superblock_t
*sb
;
1867 jbd_debug(1, "Clear features 0x%lx/0x%lx/0x%lx\n",
1868 compat
, ro
, incompat
);
1870 sb
= journal
->j_superblock
;
1872 sb
->s_feature_compat
&= ~cpu_to_be32(compat
);
1873 sb
->s_feature_ro_compat
&= ~cpu_to_be32(ro
);
1874 sb
->s_feature_incompat
&= ~cpu_to_be32(incompat
);
1876 EXPORT_SYMBOL(jbd2_journal_clear_features
);
1879 * int jbd2_journal_flush () - Flush journal
1880 * @journal: Journal to act on.
1882 * Flush all data for a given journal to disk and empty the journal.
1883 * Filesystems can use this when remounting readonly to ensure that
1884 * recovery does not need to happen on remount.
1887 int jbd2_journal_flush(journal_t
*journal
)
1890 transaction_t
*transaction
= NULL
;
1892 write_lock(&journal
->j_state_lock
);
1894 /* Force everything buffered to the log... */
1895 if (journal
->j_running_transaction
) {
1896 transaction
= journal
->j_running_transaction
;
1897 __jbd2_log_start_commit(journal
, transaction
->t_tid
);
1898 } else if (journal
->j_committing_transaction
)
1899 transaction
= journal
->j_committing_transaction
;
1901 /* Wait for the log commit to complete... */
1903 tid_t tid
= transaction
->t_tid
;
1905 write_unlock(&journal
->j_state_lock
);
1906 jbd2_log_wait_commit(journal
, tid
);
1908 write_unlock(&journal
->j_state_lock
);
1911 /* ...and flush everything in the log out to disk. */
1912 spin_lock(&journal
->j_list_lock
);
1913 while (!err
&& journal
->j_checkpoint_transactions
!= NULL
) {
1914 spin_unlock(&journal
->j_list_lock
);
1915 mutex_lock(&journal
->j_checkpoint_mutex
);
1916 err
= jbd2_log_do_checkpoint(journal
);
1917 mutex_unlock(&journal
->j_checkpoint_mutex
);
1918 spin_lock(&journal
->j_list_lock
);
1920 spin_unlock(&journal
->j_list_lock
);
1922 if (is_journal_aborted(journal
))
1925 mutex_lock(&journal
->j_checkpoint_mutex
);
1926 jbd2_cleanup_journal_tail(journal
);
1928 /* Finally, mark the journal as really needing no recovery.
1929 * This sets s_start==0 in the underlying superblock, which is
1930 * the magic code for a fully-recovered superblock. Any future
1931 * commits of data to the journal will restore the current
1933 jbd2_mark_journal_empty(journal
);
1934 mutex_unlock(&journal
->j_checkpoint_mutex
);
1935 write_lock(&journal
->j_state_lock
);
1936 J_ASSERT(!journal
->j_running_transaction
);
1937 J_ASSERT(!journal
->j_committing_transaction
);
1938 J_ASSERT(!journal
->j_checkpoint_transactions
);
1939 J_ASSERT(journal
->j_head
== journal
->j_tail
);
1940 J_ASSERT(journal
->j_tail_sequence
== journal
->j_transaction_sequence
);
1941 write_unlock(&journal
->j_state_lock
);
1946 * int jbd2_journal_wipe() - Wipe journal contents
1947 * @journal: Journal to act on.
1948 * @write: flag (see below)
1950 * Wipe out all of the contents of a journal, safely. This will produce
1951 * a warning if the journal contains any valid recovery information.
1952 * Must be called between journal_init_*() and jbd2_journal_load().
1954 * If 'write' is non-zero, then we wipe out the journal on disk; otherwise
1955 * we merely suppress recovery.
1958 int jbd2_journal_wipe(journal_t
*journal
, int write
)
1962 J_ASSERT (!(journal
->j_flags
& JBD2_LOADED
));
1964 err
= load_superblock(journal
);
1968 if (!journal
->j_tail
)
1971 printk(KERN_WARNING
"JBD2: %s recovery information on journal\n",
1972 write
? "Clearing" : "Ignoring");
1974 err
= jbd2_journal_skip_recovery(journal
);
1976 /* Lock to make assertions happy... */
1977 mutex_lock(&journal
->j_checkpoint_mutex
);
1978 jbd2_mark_journal_empty(journal
);
1979 mutex_unlock(&journal
->j_checkpoint_mutex
);
1987 * Journal abort has very specific semantics, which we describe
1988 * for journal abort.
1990 * Two internal functions, which provide abort to the jbd layer
1995 * Quick version for internal journal use (doesn't lock the journal).
1996 * Aborts hard --- we mark the abort as occurred, but do _nothing_ else,
1997 * and don't attempt to make any other journal updates.
1999 void __jbd2_journal_abort_hard(journal_t
*journal
)
2001 transaction_t
*transaction
;
2003 if (journal
->j_flags
& JBD2_ABORT
)
2006 printk(KERN_ERR
"Aborting journal on device %s.\n",
2007 journal
->j_devname
);
2009 write_lock(&journal
->j_state_lock
);
2010 journal
->j_flags
|= JBD2_ABORT
;
2011 transaction
= journal
->j_running_transaction
;
2013 __jbd2_log_start_commit(journal
, transaction
->t_tid
);
2014 write_unlock(&journal
->j_state_lock
);
2017 /* Soft abort: record the abort error status in the journal superblock,
2018 * but don't do any other IO. */
2019 static void __journal_abort_soft (journal_t
*journal
, int errno
)
2021 if (journal
->j_flags
& JBD2_ABORT
)
2024 if (!journal
->j_errno
)
2025 journal
->j_errno
= errno
;
2027 __jbd2_journal_abort_hard(journal
);
2030 jbd2_journal_update_sb_errno(journal
);
2034 * void jbd2_journal_abort () - Shutdown the journal immediately.
2035 * @journal: the journal to shutdown.
2036 * @errno: an error number to record in the journal indicating
2037 * the reason for the shutdown.
2039 * Perform a complete, immediate shutdown of the ENTIRE
2040 * journal (not of a single transaction). This operation cannot be
2041 * undone without closing and reopening the journal.
2043 * The jbd2_journal_abort function is intended to support higher level error
2044 * recovery mechanisms such as the ext2/ext3 remount-readonly error
2047 * Journal abort has very specific semantics. Any existing dirty,
2048 * unjournaled buffers in the main filesystem will still be written to
2049 * disk by bdflush, but the journaling mechanism will be suspended
2050 * immediately and no further transaction commits will be honoured.
2052 * Any dirty, journaled buffers will be written back to disk without
2053 * hitting the journal. Atomicity cannot be guaranteed on an aborted
2054 * filesystem, but we _do_ attempt to leave as much data as possible
2055 * behind for fsck to use for cleanup.
2057 * Any attempt to get a new transaction handle on a journal which is in
2058 * ABORT state will just result in an -EROFS error return. A
2059 * jbd2_journal_stop on an existing handle will return -EIO if we have
2060 * entered abort state during the update.
2062 * Recursive transactions are not disturbed by journal abort until the
2063 * final jbd2_journal_stop, which will receive the -EIO error.
2065 * Finally, the jbd2_journal_abort call allows the caller to supply an errno
2066 * which will be recorded (if possible) in the journal superblock. This
2067 * allows a client to record failure conditions in the middle of a
2068 * transaction without having to complete the transaction to record the
2069 * failure to disk. ext3_error, for example, now uses this
2072 * Errors which originate from within the journaling layer will NOT
2073 * supply an errno; a null errno implies that absolutely no further
2074 * writes are done to the journal (unless there are any already in
2079 void jbd2_journal_abort(journal_t
*journal
, int errno
)
2081 __journal_abort_soft(journal
, errno
);
2085 * int jbd2_journal_errno () - returns the journal's error state.
2086 * @journal: journal to examine.
2088 * This is the errno number set with jbd2_journal_abort(), the last
2089 * time the journal was mounted - if the journal was stopped
2090 * without calling abort this will be 0.
2092 * If the journal has been aborted on this mount time -EROFS will
2095 int jbd2_journal_errno(journal_t
*journal
)
2099 read_lock(&journal
->j_state_lock
);
2100 if (journal
->j_flags
& JBD2_ABORT
)
2103 err
= journal
->j_errno
;
2104 read_unlock(&journal
->j_state_lock
);
2109 * int jbd2_journal_clear_err () - clears the journal's error state
2110 * @journal: journal to act on.
2112 * An error must be cleared or acked to take a FS out of readonly
2115 int jbd2_journal_clear_err(journal_t
*journal
)
2119 write_lock(&journal
->j_state_lock
);
2120 if (journal
->j_flags
& JBD2_ABORT
)
2123 journal
->j_errno
= 0;
2124 write_unlock(&journal
->j_state_lock
);
2129 * void jbd2_journal_ack_err() - Ack journal err.
2130 * @journal: journal to act on.
2132 * An error must be cleared or acked to take a FS out of readonly
2135 void jbd2_journal_ack_err(journal_t
*journal
)
2137 write_lock(&journal
->j_state_lock
);
2138 if (journal
->j_errno
)
2139 journal
->j_flags
|= JBD2_ACK_ERR
;
2140 write_unlock(&journal
->j_state_lock
);
2143 int jbd2_journal_blocks_per_page(struct inode
*inode
)
2145 return 1 << (PAGE_CACHE_SHIFT
- inode
->i_sb
->s_blocksize_bits
);
2149 * helper functions to deal with 32 or 64bit block numbers.
2151 size_t journal_tag_bytes(journal_t
*journal
)
2153 journal_block_tag_t tag
;
2156 if (JBD2_HAS_INCOMPAT_FEATURE(journal
, JBD2_FEATURE_INCOMPAT_CSUM_V2
))
2157 x
+= sizeof(tag
.t_checksum
);
2159 if (JBD2_HAS_INCOMPAT_FEATURE(journal
, JBD2_FEATURE_INCOMPAT_64BIT
))
2160 return x
+ JBD2_TAG_SIZE64
;
2162 return x
+ JBD2_TAG_SIZE32
;
2166 * JBD memory management
2168 * These functions are used to allocate block-sized chunks of memory
2169 * used for making copies of buffer_head data. Very often it will be
2170 * page-sized chunks of data, but sometimes it will be in
2171 * sub-page-size chunks. (For example, 16k pages on Power systems
2172 * with a 4k block file system.) For blocks smaller than a page, we
2173 * use a SLAB allocator. There are slab caches for each block size,
2174 * which are allocated at mount time, if necessary, and we only free
2175 * (all of) the slab caches when/if the jbd2 module is unloaded. For
2176 * this reason we don't need to a mutex to protect access to
2177 * jbd2_slab[] allocating or releasing memory; only in
2178 * jbd2_journal_create_slab().
2180 #define JBD2_MAX_SLABS 8
2181 static struct kmem_cache
*jbd2_slab
[JBD2_MAX_SLABS
];
2183 static const char *jbd2_slab_names
[JBD2_MAX_SLABS
] = {
2184 "jbd2_1k", "jbd2_2k", "jbd2_4k", "jbd2_8k",
2185 "jbd2_16k", "jbd2_32k", "jbd2_64k", "jbd2_128k"
2189 static void jbd2_journal_destroy_slabs(void)
2193 for (i
= 0; i
< JBD2_MAX_SLABS
; i
++) {
2195 kmem_cache_destroy(jbd2_slab
[i
]);
2196 jbd2_slab
[i
] = NULL
;
2200 static int jbd2_journal_create_slab(size_t size
)
2202 static DEFINE_MUTEX(jbd2_slab_create_mutex
);
2203 int i
= order_base_2(size
) - 10;
2206 if (size
== PAGE_SIZE
)
2209 if (i
>= JBD2_MAX_SLABS
)
2212 if (unlikely(i
< 0))
2214 mutex_lock(&jbd2_slab_create_mutex
);
2216 mutex_unlock(&jbd2_slab_create_mutex
);
2217 return 0; /* Already created */
2220 slab_size
= 1 << (i
+10);
2221 jbd2_slab
[i
] = kmem_cache_create(jbd2_slab_names
[i
], slab_size
,
2222 slab_size
, 0, NULL
);
2223 mutex_unlock(&jbd2_slab_create_mutex
);
2224 if (!jbd2_slab
[i
]) {
2225 printk(KERN_EMERG
"JBD2: no memory for jbd2_slab cache\n");
2231 static struct kmem_cache
*get_slab(size_t size
)
2233 int i
= order_base_2(size
) - 10;
2235 BUG_ON(i
>= JBD2_MAX_SLABS
);
2236 if (unlikely(i
< 0))
2238 BUG_ON(jbd2_slab
[i
] == NULL
);
2239 return jbd2_slab
[i
];
2242 void *jbd2_alloc(size_t size
, gfp_t flags
)
2246 BUG_ON(size
& (size
-1)); /* Must be a power of 2 */
2248 flags
|= __GFP_REPEAT
;
2249 if (size
== PAGE_SIZE
)
2250 ptr
= (void *)__get_free_pages(flags
, 0);
2251 else if (size
> PAGE_SIZE
) {
2252 int order
= get_order(size
);
2255 ptr
= (void *)__get_free_pages(flags
, order
);
2257 ptr
= vmalloc(size
);
2259 ptr
= kmem_cache_alloc(get_slab(size
), flags
);
2261 /* Check alignment; SLUB has gotten this wrong in the past,
2262 * and this can lead to user data corruption! */
2263 BUG_ON(((unsigned long) ptr
) & (size
-1));
2268 void jbd2_free(void *ptr
, size_t size
)
2270 if (size
== PAGE_SIZE
) {
2271 free_pages((unsigned long)ptr
, 0);
2274 if (size
> PAGE_SIZE
) {
2275 int order
= get_order(size
);
2278 free_pages((unsigned long)ptr
, order
);
2283 kmem_cache_free(get_slab(size
), ptr
);
2287 * Journal_head storage management
2289 static struct kmem_cache
*jbd2_journal_head_cache
;
2290 #ifdef CONFIG_JBD2_DEBUG
2291 static atomic_t nr_journal_heads
= ATOMIC_INIT(0);
2294 static int jbd2_journal_init_journal_head_cache(void)
2298 J_ASSERT(jbd2_journal_head_cache
== NULL
);
2299 jbd2_journal_head_cache
= kmem_cache_create("jbd2_journal_head",
2300 sizeof(struct journal_head
),
2302 SLAB_TEMPORARY
, /* flags */
2305 if (!jbd2_journal_head_cache
) {
2307 printk(KERN_EMERG
"JBD2: no memory for journal_head cache\n");
2312 static void jbd2_journal_destroy_journal_head_cache(void)
2314 if (jbd2_journal_head_cache
) {
2315 kmem_cache_destroy(jbd2_journal_head_cache
);
2316 jbd2_journal_head_cache
= NULL
;
2321 * journal_head splicing and dicing
2323 static struct journal_head
*journal_alloc_journal_head(void)
2325 struct journal_head
*ret
;
2327 #ifdef CONFIG_JBD2_DEBUG
2328 atomic_inc(&nr_journal_heads
);
2330 ret
= kmem_cache_alloc(jbd2_journal_head_cache
, GFP_NOFS
);
2332 jbd_debug(1, "out of memory for journal_head\n");
2333 pr_notice_ratelimited("ENOMEM in %s, retrying.\n", __func__
);
2336 ret
= kmem_cache_alloc(jbd2_journal_head_cache
, GFP_NOFS
);
2342 static void journal_free_journal_head(struct journal_head
*jh
)
2344 #ifdef CONFIG_JBD2_DEBUG
2345 atomic_dec(&nr_journal_heads
);
2346 memset(jh
, JBD2_POISON_FREE
, sizeof(*jh
));
2348 kmem_cache_free(jbd2_journal_head_cache
, jh
);
2352 * A journal_head is attached to a buffer_head whenever JBD has an
2353 * interest in the buffer.
2355 * Whenever a buffer has an attached journal_head, its ->b_state:BH_JBD bit
2356 * is set. This bit is tested in core kernel code where we need to take
2357 * JBD-specific actions. Testing the zeroness of ->b_private is not reliable
2360 * When a buffer has its BH_JBD bit set, its ->b_count is elevated by one.
2362 * When a buffer has its BH_JBD bit set it is immune from being released by
2363 * core kernel code, mainly via ->b_count.
2365 * A journal_head is detached from its buffer_head when the journal_head's
2366 * b_jcount reaches zero. Running transaction (b_transaction) and checkpoint
2367 * transaction (b_cp_transaction) hold their references to b_jcount.
2369 * Various places in the kernel want to attach a journal_head to a buffer_head
2370 * _before_ attaching the journal_head to a transaction. To protect the
2371 * journal_head in this situation, jbd2_journal_add_journal_head elevates the
2372 * journal_head's b_jcount refcount by one. The caller must call
2373 * jbd2_journal_put_journal_head() to undo this.
2375 * So the typical usage would be:
2377 * (Attach a journal_head if needed. Increments b_jcount)
2378 * struct journal_head *jh = jbd2_journal_add_journal_head(bh);
2380 * (Get another reference for transaction)
2381 * jbd2_journal_grab_journal_head(bh);
2382 * jh->b_transaction = xxx;
2383 * (Put original reference)
2384 * jbd2_journal_put_journal_head(jh);
2388 * Give a buffer_head a journal_head.
2392 struct journal_head
*jbd2_journal_add_journal_head(struct buffer_head
*bh
)
2394 struct journal_head
*jh
;
2395 struct journal_head
*new_jh
= NULL
;
2398 if (!buffer_jbd(bh
)) {
2399 new_jh
= journal_alloc_journal_head();
2400 memset(new_jh
, 0, sizeof(*new_jh
));
2403 jbd_lock_bh_journal_head(bh
);
2404 if (buffer_jbd(bh
)) {
2408 (atomic_read(&bh
->b_count
) > 0) ||
2409 (bh
->b_page
&& bh
->b_page
->mapping
));
2412 jbd_unlock_bh_journal_head(bh
);
2417 new_jh
= NULL
; /* We consumed it */
2422 BUFFER_TRACE(bh
, "added journal_head");
2425 jbd_unlock_bh_journal_head(bh
);
2427 journal_free_journal_head(new_jh
);
2428 return bh
->b_private
;
2432 * Grab a ref against this buffer_head's journal_head. If it ended up not
2433 * having a journal_head, return NULL
2435 struct journal_head
*jbd2_journal_grab_journal_head(struct buffer_head
*bh
)
2437 struct journal_head
*jh
= NULL
;
2439 jbd_lock_bh_journal_head(bh
);
2440 if (buffer_jbd(bh
)) {
2444 jbd_unlock_bh_journal_head(bh
);
2448 static void __journal_remove_journal_head(struct buffer_head
*bh
)
2450 struct journal_head
*jh
= bh2jh(bh
);
2452 J_ASSERT_JH(jh
, jh
->b_jcount
>= 0);
2453 J_ASSERT_JH(jh
, jh
->b_transaction
== NULL
);
2454 J_ASSERT_JH(jh
, jh
->b_next_transaction
== NULL
);
2455 J_ASSERT_JH(jh
, jh
->b_cp_transaction
== NULL
);
2456 J_ASSERT_JH(jh
, jh
->b_jlist
== BJ_None
);
2457 J_ASSERT_BH(bh
, buffer_jbd(bh
));
2458 J_ASSERT_BH(bh
, jh2bh(jh
) == bh
);
2459 BUFFER_TRACE(bh
, "remove journal_head");
2460 if (jh
->b_frozen_data
) {
2461 printk(KERN_WARNING
"%s: freeing b_frozen_data\n", __func__
);
2462 jbd2_free(jh
->b_frozen_data
, bh
->b_size
);
2464 if (jh
->b_committed_data
) {
2465 printk(KERN_WARNING
"%s: freeing b_committed_data\n", __func__
);
2466 jbd2_free(jh
->b_committed_data
, bh
->b_size
);
2468 bh
->b_private
= NULL
;
2469 jh
->b_bh
= NULL
; /* debug, really */
2470 clear_buffer_jbd(bh
);
2471 journal_free_journal_head(jh
);
2475 * Drop a reference on the passed journal_head. If it fell to zero then
2476 * release the journal_head from the buffer_head.
2478 void jbd2_journal_put_journal_head(struct journal_head
*jh
)
2480 struct buffer_head
*bh
= jh2bh(jh
);
2482 jbd_lock_bh_journal_head(bh
);
2483 J_ASSERT_JH(jh
, jh
->b_jcount
> 0);
2485 if (!jh
->b_jcount
) {
2486 __journal_remove_journal_head(bh
);
2487 jbd_unlock_bh_journal_head(bh
);
2490 jbd_unlock_bh_journal_head(bh
);
2494 * Initialize jbd inode head
2496 void jbd2_journal_init_jbd_inode(struct jbd2_inode
*jinode
, struct inode
*inode
)
2498 jinode
->i_transaction
= NULL
;
2499 jinode
->i_next_transaction
= NULL
;
2500 jinode
->i_vfs_inode
= inode
;
2501 jinode
->i_flags
= 0;
2502 INIT_LIST_HEAD(&jinode
->i_list
);
2506 * Function to be called before we start removing inode from memory (i.e.,
2507 * clear_inode() is a fine place to be called from). It removes inode from
2508 * transaction's lists.
2510 void jbd2_journal_release_jbd_inode(journal_t
*journal
,
2511 struct jbd2_inode
*jinode
)
2516 spin_lock(&journal
->j_list_lock
);
2517 /* Is commit writing out inode - we have to wait */
2518 if (test_bit(__JI_COMMIT_RUNNING
, &jinode
->i_flags
)) {
2519 wait_queue_head_t
*wq
;
2520 DEFINE_WAIT_BIT(wait
, &jinode
->i_flags
, __JI_COMMIT_RUNNING
);
2521 wq
= bit_waitqueue(&jinode
->i_flags
, __JI_COMMIT_RUNNING
);
2522 prepare_to_wait(wq
, &wait
.wait
, TASK_UNINTERRUPTIBLE
);
2523 spin_unlock(&journal
->j_list_lock
);
2525 finish_wait(wq
, &wait
.wait
);
2529 if (jinode
->i_transaction
) {
2530 list_del(&jinode
->i_list
);
2531 jinode
->i_transaction
= NULL
;
2533 spin_unlock(&journal
->j_list_lock
);
2537 #ifdef CONFIG_PROC_FS
2539 #define JBD2_STATS_PROC_NAME "fs/jbd2"
2541 static void __init
jbd2_create_jbd_stats_proc_entry(void)
2543 proc_jbd2_stats
= proc_mkdir(JBD2_STATS_PROC_NAME
, NULL
);
2546 static void __exit
jbd2_remove_jbd_stats_proc_entry(void)
2548 if (proc_jbd2_stats
)
2549 remove_proc_entry(JBD2_STATS_PROC_NAME
, NULL
);
2554 #define jbd2_create_jbd_stats_proc_entry() do {} while (0)
2555 #define jbd2_remove_jbd_stats_proc_entry() do {} while (0)
2559 struct kmem_cache
*jbd2_handle_cache
, *jbd2_inode_cache
;
2561 static int __init
jbd2_journal_init_handle_cache(void)
2563 jbd2_handle_cache
= KMEM_CACHE(jbd2_journal_handle
, SLAB_TEMPORARY
);
2564 if (jbd2_handle_cache
== NULL
) {
2565 printk(KERN_EMERG
"JBD2: failed to create handle cache\n");
2568 jbd2_inode_cache
= KMEM_CACHE(jbd2_inode
, 0);
2569 if (jbd2_inode_cache
== NULL
) {
2570 printk(KERN_EMERG
"JBD2: failed to create inode cache\n");
2571 kmem_cache_destroy(jbd2_handle_cache
);
2577 static void jbd2_journal_destroy_handle_cache(void)
2579 if (jbd2_handle_cache
)
2580 kmem_cache_destroy(jbd2_handle_cache
);
2581 if (jbd2_inode_cache
)
2582 kmem_cache_destroy(jbd2_inode_cache
);
2587 * Module startup and shutdown
2590 static int __init
journal_init_caches(void)
2594 ret
= jbd2_journal_init_revoke_caches();
2596 ret
= jbd2_journal_init_journal_head_cache();
2598 ret
= jbd2_journal_init_handle_cache();
2600 ret
= jbd2_journal_init_transaction_cache();
2604 static void jbd2_journal_destroy_caches(void)
2606 jbd2_journal_destroy_revoke_caches();
2607 jbd2_journal_destroy_journal_head_cache();
2608 jbd2_journal_destroy_handle_cache();
2609 jbd2_journal_destroy_transaction_cache();
2610 jbd2_journal_destroy_slabs();
2613 static int __init
journal_init(void)
2617 BUILD_BUG_ON(sizeof(struct journal_superblock_s
) != 1024);
2619 ret
= journal_init_caches();
2621 jbd2_create_jbd_stats_proc_entry();
2623 jbd2_journal_destroy_caches();
2628 static void __exit
journal_exit(void)
2630 #ifdef CONFIG_JBD2_DEBUG
2631 int n
= atomic_read(&nr_journal_heads
);
2633 printk(KERN_EMERG
"JBD2: leaked %d journal_heads!\n", n
);
2635 jbd2_remove_jbd_stats_proc_entry();
2636 jbd2_journal_destroy_caches();
2639 MODULE_LICENSE("GPL");
2640 module_init(journal_init
);
2641 module_exit(journal_exit
);