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 #ifdef CONFIG_JBD2_DEBUG
107 void __jbd2_debug(int level
, const char *file
, const char *func
,
108 unsigned int line
, const char *fmt
, ...)
110 struct va_format vaf
;
113 if (level
> jbd2_journal_enable_debug
)
118 printk(KERN_DEBUG
"%s: (%s, %u): %pV\n", file
, func
, line
, &vaf
);
121 EXPORT_SYMBOL(__jbd2_debug
);
124 /* Checksumming functions */
125 int jbd2_verify_csum_type(journal_t
*j
, journal_superblock_t
*sb
)
127 if (!JBD2_HAS_INCOMPAT_FEATURE(j
, JBD2_FEATURE_INCOMPAT_CSUM_V2
))
130 return sb
->s_checksum_type
== JBD2_CRC32C_CHKSUM
;
133 static __u32
jbd2_superblock_csum(journal_t
*j
, journal_superblock_t
*sb
)
135 __u32 csum
, old_csum
;
137 old_csum
= sb
->s_checksum
;
139 csum
= jbd2_chksum(j
, ~0, (char *)sb
, sizeof(journal_superblock_t
));
140 sb
->s_checksum
= old_csum
;
142 return cpu_to_be32(csum
);
145 int jbd2_superblock_csum_verify(journal_t
*j
, journal_superblock_t
*sb
)
147 if (!JBD2_HAS_INCOMPAT_FEATURE(j
, JBD2_FEATURE_INCOMPAT_CSUM_V2
))
150 return sb
->s_checksum
== jbd2_superblock_csum(j
, sb
);
153 void jbd2_superblock_csum_set(journal_t
*j
, journal_superblock_t
*sb
)
155 if (!JBD2_HAS_INCOMPAT_FEATURE(j
, JBD2_FEATURE_INCOMPAT_CSUM_V2
))
158 sb
->s_checksum
= jbd2_superblock_csum(j
, sb
);
162 * Helper function used to manage commit timeouts
165 static void commit_timeout(unsigned long __data
)
167 struct task_struct
* p
= (struct task_struct
*) __data
;
173 * kjournald2: The main thread function used to manage a logging device
176 * This kernel thread is responsible for two things:
178 * 1) COMMIT: Every so often we need to commit the current state of the
179 * filesystem to disk. The journal thread is responsible for writing
180 * all of the metadata buffers to disk.
182 * 2) CHECKPOINT: We cannot reuse a used section of the log file until all
183 * of the data in that part of the log has been rewritten elsewhere on
184 * the disk. Flushing these old buffers to reclaim space in the log is
185 * known as checkpointing, and this thread is responsible for that job.
188 static int kjournald2(void *arg
)
190 journal_t
*journal
= arg
;
191 transaction_t
*transaction
;
194 * Set up an interval timer which can be used to trigger a commit wakeup
195 * after the commit interval expires
197 setup_timer(&journal
->j_commit_timer
, commit_timeout
,
198 (unsigned long)current
);
202 /* Record that the journal thread is running */
203 journal
->j_task
= current
;
204 wake_up(&journal
->j_wait_done_commit
);
207 * And now, wait forever for commit wakeup events.
209 write_lock(&journal
->j_state_lock
);
212 if (journal
->j_flags
& JBD2_UNMOUNT
)
215 jbd_debug(1, "commit_sequence=%d, commit_request=%d\n",
216 journal
->j_commit_sequence
, journal
->j_commit_request
);
218 if (journal
->j_commit_sequence
!= journal
->j_commit_request
) {
219 jbd_debug(1, "OK, requests differ\n");
220 write_unlock(&journal
->j_state_lock
);
221 del_timer_sync(&journal
->j_commit_timer
);
222 jbd2_journal_commit_transaction(journal
);
223 write_lock(&journal
->j_state_lock
);
227 wake_up(&journal
->j_wait_done_commit
);
228 if (freezing(current
)) {
230 * The simpler the better. Flushing journal isn't a
231 * good idea, because that depends on threads that may
232 * be already stopped.
234 jbd_debug(1, "Now suspending kjournald2\n");
235 write_unlock(&journal
->j_state_lock
);
237 write_lock(&journal
->j_state_lock
);
240 * We assume on resume that commits are already there,
244 int should_sleep
= 1;
246 prepare_to_wait(&journal
->j_wait_commit
, &wait
,
248 if (journal
->j_commit_sequence
!= journal
->j_commit_request
)
250 transaction
= journal
->j_running_transaction
;
251 if (transaction
&& time_after_eq(jiffies
,
252 transaction
->t_expires
))
254 if (journal
->j_flags
& JBD2_UNMOUNT
)
257 write_unlock(&journal
->j_state_lock
);
259 write_lock(&journal
->j_state_lock
);
261 finish_wait(&journal
->j_wait_commit
, &wait
);
264 jbd_debug(1, "kjournald2 wakes\n");
267 * Were we woken up by a commit wakeup event?
269 transaction
= journal
->j_running_transaction
;
270 if (transaction
&& time_after_eq(jiffies
, transaction
->t_expires
)) {
271 journal
->j_commit_request
= transaction
->t_tid
;
272 jbd_debug(1, "woke because of timeout\n");
277 write_unlock(&journal
->j_state_lock
);
278 del_timer_sync(&journal
->j_commit_timer
);
279 journal
->j_task
= NULL
;
280 wake_up(&journal
->j_wait_done_commit
);
281 jbd_debug(1, "Journal thread exiting.\n");
285 static int jbd2_journal_start_thread(journal_t
*journal
)
287 struct task_struct
*t
;
289 t
= kthread_run(kjournald2
, journal
, "jbd2/%s",
294 wait_event(journal
->j_wait_done_commit
, journal
->j_task
!= NULL
);
298 static void journal_kill_thread(journal_t
*journal
)
300 write_lock(&journal
->j_state_lock
);
301 journal
->j_flags
|= JBD2_UNMOUNT
;
303 while (journal
->j_task
) {
304 wake_up(&journal
->j_wait_commit
);
305 write_unlock(&journal
->j_state_lock
);
306 wait_event(journal
->j_wait_done_commit
, journal
->j_task
== NULL
);
307 write_lock(&journal
->j_state_lock
);
309 write_unlock(&journal
->j_state_lock
);
313 * jbd2_journal_write_metadata_buffer: write a metadata buffer to the journal.
315 * Writes a metadata buffer to a given disk block. The actual IO is not
316 * performed but a new buffer_head is constructed which labels the data
317 * to be written with the correct destination disk block.
319 * Any magic-number escaping which needs to be done will cause a
320 * copy-out here. If the buffer happens to start with the
321 * JBD2_MAGIC_NUMBER, then we can't write it to the log directly: the
322 * magic number is only written to the log for descripter blocks. In
323 * this case, we copy the data and replace the first word with 0, and we
324 * return a result code which indicates that this buffer needs to be
325 * marked as an escaped buffer in the corresponding log descriptor
326 * block. The missing word can then be restored when the block is read
329 * If the source buffer has already been modified by a new transaction
330 * since we took the last commit snapshot, we use the frozen copy of
331 * that data for IO. If we end up using the existing buffer_head's data
332 * for the write, then we have to make sure nobody modifies it while the
333 * IO is in progress. do_get_write_access() handles this.
335 * The function returns a pointer to the buffer_head to be used for IO.
343 * Bit 0 set == escape performed on the data
344 * Bit 1 set == buffer copy-out performed (kfree the data after IO)
347 int jbd2_journal_write_metadata_buffer(transaction_t
*transaction
,
348 struct journal_head
*jh_in
,
349 struct buffer_head
**bh_out
,
352 int need_copy_out
= 0;
353 int done_copy_out
= 0;
356 struct buffer_head
*new_bh
;
357 struct page
*new_page
;
358 unsigned int new_offset
;
359 struct buffer_head
*bh_in
= jh2bh(jh_in
);
360 journal_t
*journal
= transaction
->t_journal
;
363 * The buffer really shouldn't be locked: only the current committing
364 * transaction is allowed to write it, so nobody else is allowed
367 * akpm: except if we're journalling data, and write() output is
368 * also part of a shared mapping, and another thread has
369 * decided to launch a writepage() against this buffer.
371 J_ASSERT_BH(bh_in
, buffer_jbddirty(bh_in
));
374 new_bh
= alloc_buffer_head(GFP_NOFS
);
377 * Failure is not an option, but __GFP_NOFAIL is going
378 * away; so we retry ourselves here.
380 congestion_wait(BLK_RW_ASYNC
, HZ
/50);
384 /* keep subsequent assertions sane */
385 atomic_set(&new_bh
->b_count
, 1);
387 jbd_lock_bh_state(bh_in
);
390 * If a new transaction has already done a buffer copy-out, then
391 * we use that version of the data for the commit.
393 if (jh_in
->b_frozen_data
) {
395 new_page
= virt_to_page(jh_in
->b_frozen_data
);
396 new_offset
= offset_in_page(jh_in
->b_frozen_data
);
398 new_page
= jh2bh(jh_in
)->b_page
;
399 new_offset
= offset_in_page(jh2bh(jh_in
)->b_data
);
402 mapped_data
= kmap_atomic(new_page
);
404 * Fire data frozen trigger if data already wasn't frozen. Do this
405 * before checking for escaping, as the trigger may modify the magic
406 * offset. If a copy-out happens afterwards, it will have the correct
407 * data in the buffer.
410 jbd2_buffer_frozen_trigger(jh_in
, mapped_data
+ new_offset
,
416 if (*((__be32
*)(mapped_data
+ new_offset
)) ==
417 cpu_to_be32(JBD2_MAGIC_NUMBER
)) {
421 kunmap_atomic(mapped_data
);
424 * Do we need to do a data copy?
426 if (need_copy_out
&& !done_copy_out
) {
429 jbd_unlock_bh_state(bh_in
);
430 tmp
= jbd2_alloc(bh_in
->b_size
, GFP_NOFS
);
435 jbd_lock_bh_state(bh_in
);
436 if (jh_in
->b_frozen_data
) {
437 jbd2_free(tmp
, bh_in
->b_size
);
441 jh_in
->b_frozen_data
= tmp
;
442 mapped_data
= kmap_atomic(new_page
);
443 memcpy(tmp
, mapped_data
+ new_offset
, bh_in
->b_size
);
444 kunmap_atomic(mapped_data
);
446 new_page
= virt_to_page(tmp
);
447 new_offset
= offset_in_page(tmp
);
451 * This isn't strictly necessary, as we're using frozen
452 * data for the escaping, but it keeps consistency with
453 * b_frozen_data usage.
455 jh_in
->b_frozen_triggers
= jh_in
->b_triggers
;
459 * Did we need to do an escaping? Now we've done all the
460 * copying, we can finally do so.
463 mapped_data
= kmap_atomic(new_page
);
464 *((unsigned int *)(mapped_data
+ new_offset
)) = 0;
465 kunmap_atomic(mapped_data
);
468 set_bh_page(new_bh
, new_page
, new_offset
);
469 new_bh
->b_size
= bh_in
->b_size
;
470 new_bh
->b_bdev
= journal
->j_dev
;
471 new_bh
->b_blocknr
= blocknr
;
472 new_bh
->b_private
= bh_in
;
473 set_buffer_mapped(new_bh
);
474 set_buffer_dirty(new_bh
);
479 * The to-be-written buffer needs to get moved to the io queue,
480 * and the original buffer whose contents we are shadowing or
481 * copying is moved to the transaction's shadow queue.
483 JBUFFER_TRACE(jh_in
, "file as BJ_Shadow");
484 spin_lock(&journal
->j_list_lock
);
485 __jbd2_journal_file_buffer(jh_in
, transaction
, BJ_Shadow
);
486 spin_unlock(&journal
->j_list_lock
);
487 set_buffer_shadow(bh_in
);
488 jbd_unlock_bh_state(bh_in
);
490 return do_escape
| (done_copy_out
<< 1);
494 * Allocation code for the journal file. Manage the space left in the
495 * journal, so that we can begin checkpointing when appropriate.
499 * Called with j_state_lock locked for writing.
500 * Returns true if a transaction commit was started.
502 int __jbd2_log_start_commit(journal_t
*journal
, tid_t target
)
504 /* Return if the txn has already requested to be committed */
505 if (journal
->j_commit_request
== target
)
509 * The only transaction we can possibly wait upon is the
510 * currently running transaction (if it exists). Otherwise,
511 * the target tid must be an old one.
513 if (journal
->j_running_transaction
&&
514 journal
->j_running_transaction
->t_tid
== target
) {
516 * We want a new commit: OK, mark the request and wakeup the
517 * commit thread. We do _not_ do the commit ourselves.
520 journal
->j_commit_request
= target
;
521 jbd_debug(1, "JBD2: requesting commit %d/%d\n",
522 journal
->j_commit_request
,
523 journal
->j_commit_sequence
);
524 journal
->j_running_transaction
->t_requested
= jiffies
;
525 wake_up(&journal
->j_wait_commit
);
527 } else if (!tid_geq(journal
->j_commit_request
, target
))
528 /* This should never happen, but if it does, preserve
529 the evidence before kjournald goes into a loop and
530 increments j_commit_sequence beyond all recognition. */
531 WARN_ONCE(1, "JBD2: bad log_start_commit: %u %u %u %u\n",
532 journal
->j_commit_request
,
533 journal
->j_commit_sequence
,
534 target
, journal
->j_running_transaction
?
535 journal
->j_running_transaction
->t_tid
: 0);
539 int jbd2_log_start_commit(journal_t
*journal
, tid_t tid
)
543 write_lock(&journal
->j_state_lock
);
544 ret
= __jbd2_log_start_commit(journal
, tid
);
545 write_unlock(&journal
->j_state_lock
);
550 * Force and wait any uncommitted transactions. We can only force the running
551 * transaction if we don't have an active handle, otherwise, we will deadlock.
552 * Returns: <0 in case of error,
553 * 0 if nothing to commit,
554 * 1 if transaction was successfully committed.
556 static int __jbd2_journal_force_commit(journal_t
*journal
)
558 transaction_t
*transaction
= NULL
;
560 int need_to_start
= 0, ret
= 0;
562 read_lock(&journal
->j_state_lock
);
563 if (journal
->j_running_transaction
&& !current
->journal_info
) {
564 transaction
= journal
->j_running_transaction
;
565 if (!tid_geq(journal
->j_commit_request
, transaction
->t_tid
))
567 } else if (journal
->j_committing_transaction
)
568 transaction
= journal
->j_committing_transaction
;
571 /* Nothing to commit */
572 read_unlock(&journal
->j_state_lock
);
575 tid
= transaction
->t_tid
;
576 read_unlock(&journal
->j_state_lock
);
578 jbd2_log_start_commit(journal
, tid
);
579 ret
= jbd2_log_wait_commit(journal
, tid
);
587 * Force and wait upon a commit if the calling process is not within
588 * transaction. This is used for forcing out undo-protected data which contains
589 * bitmaps, when the fs is running out of space.
591 * @journal: journal to force
592 * Returns true if progress was made.
594 int jbd2_journal_force_commit_nested(journal_t
*journal
)
598 ret
= __jbd2_journal_force_commit(journal
);
603 * int journal_force_commit() - force any uncommitted transactions
604 * @journal: journal to force
606 * Caller want unconditional commit. We can only force the running transaction
607 * if we don't have an active handle, otherwise, we will deadlock.
609 int jbd2_journal_force_commit(journal_t
*journal
)
613 J_ASSERT(!current
->journal_info
);
614 ret
= __jbd2_journal_force_commit(journal
);
621 * Start a commit of the current running transaction (if any). Returns true
622 * if a transaction is going to be committed (or is currently already
623 * committing), and fills its tid in at *ptid
625 int jbd2_journal_start_commit(journal_t
*journal
, tid_t
*ptid
)
629 write_lock(&journal
->j_state_lock
);
630 if (journal
->j_running_transaction
) {
631 tid_t tid
= journal
->j_running_transaction
->t_tid
;
633 __jbd2_log_start_commit(journal
, tid
);
634 /* There's a running transaction and we've just made sure
635 * it's commit has been scheduled. */
639 } else if (journal
->j_committing_transaction
) {
641 * If commit has been started, then we have to wait for
642 * completion of that transaction.
645 *ptid
= journal
->j_committing_transaction
->t_tid
;
648 write_unlock(&journal
->j_state_lock
);
653 * Return 1 if a given transaction has not yet sent barrier request
654 * connected with a transaction commit. If 0 is returned, transaction
655 * may or may not have sent the barrier. Used to avoid sending barrier
656 * twice in common cases.
658 int jbd2_trans_will_send_data_barrier(journal_t
*journal
, tid_t tid
)
661 transaction_t
*commit_trans
;
663 if (!(journal
->j_flags
& JBD2_BARRIER
))
665 read_lock(&journal
->j_state_lock
);
666 /* Transaction already committed? */
667 if (tid_geq(journal
->j_commit_sequence
, tid
))
669 commit_trans
= journal
->j_committing_transaction
;
670 if (!commit_trans
|| commit_trans
->t_tid
!= tid
) {
675 * Transaction is being committed and we already proceeded to
676 * submitting a flush to fs partition?
678 if (journal
->j_fs_dev
!= journal
->j_dev
) {
679 if (!commit_trans
->t_need_data_flush
||
680 commit_trans
->t_state
>= T_COMMIT_DFLUSH
)
683 if (commit_trans
->t_state
>= T_COMMIT_JFLUSH
)
688 read_unlock(&journal
->j_state_lock
);
691 EXPORT_SYMBOL(jbd2_trans_will_send_data_barrier
);
694 * Wait for a specified commit to complete.
695 * The caller may not hold the journal lock.
697 int jbd2_log_wait_commit(journal_t
*journal
, tid_t tid
)
701 read_lock(&journal
->j_state_lock
);
702 #ifdef CONFIG_JBD2_DEBUG
703 if (!tid_geq(journal
->j_commit_request
, tid
)) {
705 "%s: error: j_commit_request=%d, tid=%d\n",
706 __func__
, journal
->j_commit_request
, tid
);
709 while (tid_gt(tid
, journal
->j_commit_sequence
)) {
710 jbd_debug(1, "JBD2: want %d, j_commit_sequence=%d\n",
711 tid
, journal
->j_commit_sequence
);
712 wake_up(&journal
->j_wait_commit
);
713 read_unlock(&journal
->j_state_lock
);
714 wait_event(journal
->j_wait_done_commit
,
715 !tid_gt(tid
, journal
->j_commit_sequence
));
716 read_lock(&journal
->j_state_lock
);
718 read_unlock(&journal
->j_state_lock
);
720 if (unlikely(is_journal_aborted(journal
))) {
721 printk(KERN_EMERG
"journal commit I/O error\n");
728 * When this function returns the transaction corresponding to tid
729 * will be completed. If the transaction has currently running, start
730 * committing that transaction before waiting for it to complete. If
731 * the transaction id is stale, it is by definition already completed,
732 * so just return SUCCESS.
734 int jbd2_complete_transaction(journal_t
*journal
, tid_t tid
)
736 int need_to_wait
= 1;
738 read_lock(&journal
->j_state_lock
);
739 if (journal
->j_running_transaction
&&
740 journal
->j_running_transaction
->t_tid
== tid
) {
741 if (journal
->j_commit_request
!= tid
) {
742 /* transaction not yet started, so request it */
743 read_unlock(&journal
->j_state_lock
);
744 jbd2_log_start_commit(journal
, tid
);
747 } else if (!(journal
->j_committing_transaction
&&
748 journal
->j_committing_transaction
->t_tid
== tid
))
750 read_unlock(&journal
->j_state_lock
);
754 return jbd2_log_wait_commit(journal
, tid
);
756 EXPORT_SYMBOL(jbd2_complete_transaction
);
759 * Log buffer allocation routines:
762 int jbd2_journal_next_log_block(journal_t
*journal
, unsigned long long *retp
)
764 unsigned long blocknr
;
766 write_lock(&journal
->j_state_lock
);
767 J_ASSERT(journal
->j_free
> 1);
769 blocknr
= journal
->j_head
;
772 if (journal
->j_head
== journal
->j_last
)
773 journal
->j_head
= journal
->j_first
;
774 write_unlock(&journal
->j_state_lock
);
775 return jbd2_journal_bmap(journal
, blocknr
, retp
);
779 * Conversion of logical to physical block numbers for the journal
781 * On external journals the journal blocks are identity-mapped, so
782 * this is a no-op. If needed, we can use j_blk_offset - everything is
785 int jbd2_journal_bmap(journal_t
*journal
, unsigned long blocknr
,
786 unsigned long long *retp
)
789 unsigned long long ret
;
791 if (journal
->j_inode
) {
792 ret
= bmap(journal
->j_inode
, blocknr
);
796 printk(KERN_ALERT
"%s: journal block not found "
797 "at offset %lu on %s\n",
798 __func__
, blocknr
, journal
->j_devname
);
800 __journal_abort_soft(journal
, err
);
803 *retp
= blocknr
; /* +journal->j_blk_offset */
809 * We play buffer_head aliasing tricks to write data/metadata blocks to
810 * the journal without copying their contents, but for journal
811 * descriptor blocks we do need to generate bona fide buffers.
813 * After the caller of jbd2_journal_get_descriptor_buffer() has finished modifying
814 * the buffer's contents they really should run flush_dcache_page(bh->b_page).
815 * But we don't bother doing that, so there will be coherency problems with
816 * mmaps of blockdevs which hold live JBD-controlled filesystems.
818 struct buffer_head
*jbd2_journal_get_descriptor_buffer(journal_t
*journal
)
820 struct buffer_head
*bh
;
821 unsigned long long blocknr
;
824 err
= jbd2_journal_next_log_block(journal
, &blocknr
);
829 bh
= __getblk(journal
->j_dev
, blocknr
, journal
->j_blocksize
);
833 memset(bh
->b_data
, 0, journal
->j_blocksize
);
834 set_buffer_uptodate(bh
);
836 BUFFER_TRACE(bh
, "return this buffer");
841 * Return tid of the oldest transaction in the journal and block in the journal
842 * where the transaction starts.
844 * If the journal is now empty, return which will be the next transaction ID
845 * we will write and where will that transaction start.
847 * The return value is 0 if journal tail cannot be pushed any further, 1 if
850 int jbd2_journal_get_log_tail(journal_t
*journal
, tid_t
*tid
,
851 unsigned long *block
)
853 transaction_t
*transaction
;
856 read_lock(&journal
->j_state_lock
);
857 spin_lock(&journal
->j_list_lock
);
858 transaction
= journal
->j_checkpoint_transactions
;
860 *tid
= transaction
->t_tid
;
861 *block
= transaction
->t_log_start
;
862 } else if ((transaction
= journal
->j_committing_transaction
) != NULL
) {
863 *tid
= transaction
->t_tid
;
864 *block
= transaction
->t_log_start
;
865 } else if ((transaction
= journal
->j_running_transaction
) != NULL
) {
866 *tid
= transaction
->t_tid
;
867 *block
= journal
->j_head
;
869 *tid
= journal
->j_transaction_sequence
;
870 *block
= journal
->j_head
;
872 ret
= tid_gt(*tid
, journal
->j_tail_sequence
);
873 spin_unlock(&journal
->j_list_lock
);
874 read_unlock(&journal
->j_state_lock
);
880 * Update information in journal structure and in on disk journal superblock
881 * about log tail. This function does not check whether information passed in
882 * really pushes log tail further. It's responsibility of the caller to make
883 * sure provided log tail information is valid (e.g. by holding
884 * j_checkpoint_mutex all the time between computing log tail and calling this
885 * function as is the case with jbd2_cleanup_journal_tail()).
887 * Requires j_checkpoint_mutex
889 void __jbd2_update_log_tail(journal_t
*journal
, tid_t tid
, unsigned long block
)
893 BUG_ON(!mutex_is_locked(&journal
->j_checkpoint_mutex
));
896 * We cannot afford for write to remain in drive's caches since as
897 * soon as we update j_tail, next transaction can start reusing journal
898 * space and if we lose sb update during power failure we'd replay
899 * old transaction with possibly newly overwritten data.
901 jbd2_journal_update_sb_log_tail(journal
, tid
, block
, WRITE_FUA
);
902 write_lock(&journal
->j_state_lock
);
903 freed
= block
- journal
->j_tail
;
904 if (block
< journal
->j_tail
)
905 freed
+= journal
->j_last
- journal
->j_first
;
907 trace_jbd2_update_log_tail(journal
, tid
, block
, freed
);
909 "Cleaning journal tail from %d to %d (offset %lu), "
911 journal
->j_tail_sequence
, tid
, block
, freed
);
913 journal
->j_free
+= freed
;
914 journal
->j_tail_sequence
= tid
;
915 journal
->j_tail
= block
;
916 write_unlock(&journal
->j_state_lock
);
920 * This is a variaon of __jbd2_update_log_tail which checks for validity of
921 * provided log tail and locks j_checkpoint_mutex. So it is safe against races
922 * with other threads updating log tail.
924 void jbd2_update_log_tail(journal_t
*journal
, tid_t tid
, unsigned long block
)
926 mutex_lock(&journal
->j_checkpoint_mutex
);
927 if (tid_gt(tid
, journal
->j_tail_sequence
))
928 __jbd2_update_log_tail(journal
, tid
, block
);
929 mutex_unlock(&journal
->j_checkpoint_mutex
);
932 struct jbd2_stats_proc_session
{
934 struct transaction_stats_s
*stats
;
939 static void *jbd2_seq_info_start(struct seq_file
*seq
, loff_t
*pos
)
941 return *pos
? NULL
: SEQ_START_TOKEN
;
944 static void *jbd2_seq_info_next(struct seq_file
*seq
, void *v
, loff_t
*pos
)
949 static int jbd2_seq_info_show(struct seq_file
*seq
, void *v
)
951 struct jbd2_stats_proc_session
*s
= seq
->private;
953 if (v
!= SEQ_START_TOKEN
)
955 seq_printf(seq
, "%lu transactions (%lu requested), "
956 "each up to %u blocks\n",
957 s
->stats
->ts_tid
, s
->stats
->ts_requested
,
958 s
->journal
->j_max_transaction_buffers
);
959 if (s
->stats
->ts_tid
== 0)
961 seq_printf(seq
, "average: \n %ums waiting for transaction\n",
962 jiffies_to_msecs(s
->stats
->run
.rs_wait
/ s
->stats
->ts_tid
));
963 seq_printf(seq
, " %ums request delay\n",
964 (s
->stats
->ts_requested
== 0) ? 0 :
965 jiffies_to_msecs(s
->stats
->run
.rs_request_delay
/
966 s
->stats
->ts_requested
));
967 seq_printf(seq
, " %ums running transaction\n",
968 jiffies_to_msecs(s
->stats
->run
.rs_running
/ s
->stats
->ts_tid
));
969 seq_printf(seq
, " %ums transaction was being locked\n",
970 jiffies_to_msecs(s
->stats
->run
.rs_locked
/ s
->stats
->ts_tid
));
971 seq_printf(seq
, " %ums flushing data (in ordered mode)\n",
972 jiffies_to_msecs(s
->stats
->run
.rs_flushing
/ s
->stats
->ts_tid
));
973 seq_printf(seq
, " %ums logging transaction\n",
974 jiffies_to_msecs(s
->stats
->run
.rs_logging
/ s
->stats
->ts_tid
));
975 seq_printf(seq
, " %lluus average transaction commit time\n",
976 div_u64(s
->journal
->j_average_commit_time
, 1000));
977 seq_printf(seq
, " %lu handles per transaction\n",
978 s
->stats
->run
.rs_handle_count
/ s
->stats
->ts_tid
);
979 seq_printf(seq
, " %lu blocks per transaction\n",
980 s
->stats
->run
.rs_blocks
/ s
->stats
->ts_tid
);
981 seq_printf(seq
, " %lu logged blocks per transaction\n",
982 s
->stats
->run
.rs_blocks_logged
/ s
->stats
->ts_tid
);
986 static void jbd2_seq_info_stop(struct seq_file
*seq
, void *v
)
990 static const struct seq_operations jbd2_seq_info_ops
= {
991 .start
= jbd2_seq_info_start
,
992 .next
= jbd2_seq_info_next
,
993 .stop
= jbd2_seq_info_stop
,
994 .show
= jbd2_seq_info_show
,
997 static int jbd2_seq_info_open(struct inode
*inode
, struct file
*file
)
999 journal_t
*journal
= PDE_DATA(inode
);
1000 struct jbd2_stats_proc_session
*s
;
1003 s
= kmalloc(sizeof(*s
), GFP_KERNEL
);
1006 size
= sizeof(struct transaction_stats_s
);
1007 s
->stats
= kmalloc(size
, GFP_KERNEL
);
1008 if (s
->stats
== NULL
) {
1012 spin_lock(&journal
->j_history_lock
);
1013 memcpy(s
->stats
, &journal
->j_stats
, size
);
1014 s
->journal
= journal
;
1015 spin_unlock(&journal
->j_history_lock
);
1017 rc
= seq_open(file
, &jbd2_seq_info_ops
);
1019 struct seq_file
*m
= file
->private_data
;
1029 static int jbd2_seq_info_release(struct inode
*inode
, struct file
*file
)
1031 struct seq_file
*seq
= file
->private_data
;
1032 struct jbd2_stats_proc_session
*s
= seq
->private;
1035 return seq_release(inode
, file
);
1038 static const struct file_operations jbd2_seq_info_fops
= {
1039 .owner
= THIS_MODULE
,
1040 .open
= jbd2_seq_info_open
,
1042 .llseek
= seq_lseek
,
1043 .release
= jbd2_seq_info_release
,
1046 static struct proc_dir_entry
*proc_jbd2_stats
;
1048 static void jbd2_stats_proc_init(journal_t
*journal
)
1050 journal
->j_proc_entry
= proc_mkdir(journal
->j_devname
, proc_jbd2_stats
);
1051 if (journal
->j_proc_entry
) {
1052 proc_create_data("info", S_IRUGO
, journal
->j_proc_entry
,
1053 &jbd2_seq_info_fops
, journal
);
1057 static void jbd2_stats_proc_exit(journal_t
*journal
)
1059 remove_proc_entry("info", journal
->j_proc_entry
);
1060 remove_proc_entry(journal
->j_devname
, proc_jbd2_stats
);
1064 * Management for journal control blocks: functions to create and
1065 * destroy journal_t structures, and to initialise and read existing
1066 * journal blocks from disk. */
1068 /* First: create and setup a journal_t object in memory. We initialise
1069 * very few fields yet: that has to wait until we have created the
1070 * journal structures from from scratch, or loaded them from disk. */
1072 static journal_t
* journal_init_common (void)
1077 journal
= kzalloc(sizeof(*journal
), GFP_KERNEL
);
1081 init_waitqueue_head(&journal
->j_wait_transaction_locked
);
1082 init_waitqueue_head(&journal
->j_wait_done_commit
);
1083 init_waitqueue_head(&journal
->j_wait_commit
);
1084 init_waitqueue_head(&journal
->j_wait_updates
);
1085 init_waitqueue_head(&journal
->j_wait_reserved
);
1086 mutex_init(&journal
->j_barrier
);
1087 mutex_init(&journal
->j_checkpoint_mutex
);
1088 spin_lock_init(&journal
->j_revoke_lock
);
1089 spin_lock_init(&journal
->j_list_lock
);
1090 rwlock_init(&journal
->j_state_lock
);
1092 journal
->j_commit_interval
= (HZ
* JBD2_DEFAULT_MAX_COMMIT_AGE
);
1093 journal
->j_min_batch_time
= 0;
1094 journal
->j_max_batch_time
= 15000; /* 15ms */
1095 atomic_set(&journal
->j_reserved_credits
, 0);
1097 /* The journal is marked for error until we succeed with recovery! */
1098 journal
->j_flags
= JBD2_ABORT
;
1100 /* Set up a default-sized revoke table for the new mount. */
1101 err
= jbd2_journal_init_revoke(journal
, JOURNAL_REVOKE_DEFAULT_HASH
);
1107 spin_lock_init(&journal
->j_history_lock
);
1112 /* jbd2_journal_init_dev and jbd2_journal_init_inode:
1114 * Create a journal structure assigned some fixed set of disk blocks to
1115 * the journal. We don't actually touch those disk blocks yet, but we
1116 * need to set up all of the mapping information to tell the journaling
1117 * system where the journal blocks are.
1122 * journal_t * jbd2_journal_init_dev() - creates and initialises a journal structure
1123 * @bdev: Block device on which to create the journal
1124 * @fs_dev: Device which hold journalled filesystem for this journal.
1125 * @start: Block nr Start of journal.
1126 * @len: Length of the journal in blocks.
1127 * @blocksize: blocksize of journalling device
1129 * Returns: a newly created journal_t *
1131 * jbd2_journal_init_dev creates a journal which maps a fixed contiguous
1132 * range of blocks on an arbitrary block device.
1135 journal_t
* jbd2_journal_init_dev(struct block_device
*bdev
,
1136 struct block_device
*fs_dev
,
1137 unsigned long long start
, int len
, int blocksize
)
1139 journal_t
*journal
= journal_init_common();
1140 struct buffer_head
*bh
;
1147 /* journal descriptor can store up to n blocks -bzzz */
1148 journal
->j_blocksize
= blocksize
;
1149 journal
->j_dev
= bdev
;
1150 journal
->j_fs_dev
= fs_dev
;
1151 journal
->j_blk_offset
= start
;
1152 journal
->j_maxlen
= len
;
1153 bdevname(journal
->j_dev
, journal
->j_devname
);
1154 p
= journal
->j_devname
;
1155 while ((p
= strchr(p
, '/')))
1157 jbd2_stats_proc_init(journal
);
1158 n
= journal
->j_blocksize
/ sizeof(journal_block_tag_t
);
1159 journal
->j_wbufsize
= n
;
1160 journal
->j_wbuf
= kmalloc(n
* sizeof(struct buffer_head
*), GFP_KERNEL
);
1161 if (!journal
->j_wbuf
) {
1162 printk(KERN_ERR
"%s: Can't allocate bhs for commit thread\n",
1167 bh
= __getblk(journal
->j_dev
, start
, journal
->j_blocksize
);
1170 "%s: Cannot get buffer for journal superblock\n",
1174 journal
->j_sb_buffer
= bh
;
1175 journal
->j_superblock
= (journal_superblock_t
*)bh
->b_data
;
1179 kfree(journal
->j_wbuf
);
1180 jbd2_stats_proc_exit(journal
);
1186 * journal_t * jbd2_journal_init_inode () - creates a journal which maps to a inode.
1187 * @inode: An inode to create the journal in
1189 * jbd2_journal_init_inode creates a journal which maps an on-disk inode as
1190 * the journal. The inode must exist already, must support bmap() and
1191 * must have all data blocks preallocated.
1193 journal_t
* jbd2_journal_init_inode (struct inode
*inode
)
1195 struct buffer_head
*bh
;
1196 journal_t
*journal
= journal_init_common();
1200 unsigned long long blocknr
;
1205 journal
->j_dev
= journal
->j_fs_dev
= inode
->i_sb
->s_bdev
;
1206 journal
->j_inode
= inode
;
1207 bdevname(journal
->j_dev
, journal
->j_devname
);
1208 p
= journal
->j_devname
;
1209 while ((p
= strchr(p
, '/')))
1211 p
= journal
->j_devname
+ strlen(journal
->j_devname
);
1212 sprintf(p
, "-%lu", journal
->j_inode
->i_ino
);
1214 "journal %p: inode %s/%ld, size %Ld, bits %d, blksize %ld\n",
1215 journal
, inode
->i_sb
->s_id
, inode
->i_ino
,
1216 (long long) inode
->i_size
,
1217 inode
->i_sb
->s_blocksize_bits
, inode
->i_sb
->s_blocksize
);
1219 journal
->j_maxlen
= inode
->i_size
>> inode
->i_sb
->s_blocksize_bits
;
1220 journal
->j_blocksize
= inode
->i_sb
->s_blocksize
;
1221 jbd2_stats_proc_init(journal
);
1223 /* journal descriptor can store up to n blocks -bzzz */
1224 n
= journal
->j_blocksize
/ sizeof(journal_block_tag_t
);
1225 journal
->j_wbufsize
= n
;
1226 journal
->j_wbuf
= kmalloc(n
* sizeof(struct buffer_head
*), GFP_KERNEL
);
1227 if (!journal
->j_wbuf
) {
1228 printk(KERN_ERR
"%s: Can't allocate bhs for commit thread\n",
1233 err
= jbd2_journal_bmap(journal
, 0, &blocknr
);
1234 /* If that failed, give up */
1236 printk(KERN_ERR
"%s: Cannot locate journal superblock\n",
1241 bh
= __getblk(journal
->j_dev
, blocknr
, journal
->j_blocksize
);
1244 "%s: Cannot get buffer for journal superblock\n",
1248 journal
->j_sb_buffer
= bh
;
1249 journal
->j_superblock
= (journal_superblock_t
*)bh
->b_data
;
1253 kfree(journal
->j_wbuf
);
1254 jbd2_stats_proc_exit(journal
);
1260 * If the journal init or create aborts, we need to mark the journal
1261 * superblock as being NULL to prevent the journal destroy from writing
1262 * back a bogus superblock.
1264 static void journal_fail_superblock (journal_t
*journal
)
1266 struct buffer_head
*bh
= journal
->j_sb_buffer
;
1268 journal
->j_sb_buffer
= NULL
;
1272 * Given a journal_t structure, initialise the various fields for
1273 * startup of a new journaling session. We use this both when creating
1274 * a journal, and after recovering an old journal to reset it for
1278 static int journal_reset(journal_t
*journal
)
1280 journal_superblock_t
*sb
= journal
->j_superblock
;
1281 unsigned long long first
, last
;
1283 first
= be32_to_cpu(sb
->s_first
);
1284 last
= be32_to_cpu(sb
->s_maxlen
);
1285 if (first
+ JBD2_MIN_JOURNAL_BLOCKS
> last
+ 1) {
1286 printk(KERN_ERR
"JBD2: Journal too short (blocks %llu-%llu).\n",
1288 journal_fail_superblock(journal
);
1292 journal
->j_first
= first
;
1293 journal
->j_last
= last
;
1295 journal
->j_head
= first
;
1296 journal
->j_tail
= first
;
1297 journal
->j_free
= last
- first
;
1299 journal
->j_tail_sequence
= journal
->j_transaction_sequence
;
1300 journal
->j_commit_sequence
= journal
->j_transaction_sequence
- 1;
1301 journal
->j_commit_request
= journal
->j_commit_sequence
;
1303 journal
->j_max_transaction_buffers
= journal
->j_maxlen
/ 4;
1306 * As a special case, if the on-disk copy is already marked as needing
1307 * no recovery (s_start == 0), then we can safely defer the superblock
1308 * update until the next commit by setting JBD2_FLUSHED. This avoids
1309 * attempting a write to a potential-readonly device.
1311 if (sb
->s_start
== 0) {
1312 jbd_debug(1, "JBD2: Skipping superblock update on recovered sb "
1313 "(start %ld, seq %d, errno %d)\n",
1314 journal
->j_tail
, journal
->j_tail_sequence
,
1316 journal
->j_flags
|= JBD2_FLUSHED
;
1318 /* Lock here to make assertions happy... */
1319 mutex_lock(&journal
->j_checkpoint_mutex
);
1321 * Update log tail information. We use WRITE_FUA since new
1322 * transaction will start reusing journal space and so we
1323 * must make sure information about current log tail is on
1326 jbd2_journal_update_sb_log_tail(journal
,
1327 journal
->j_tail_sequence
,
1330 mutex_unlock(&journal
->j_checkpoint_mutex
);
1332 return jbd2_journal_start_thread(journal
);
1335 static void jbd2_write_superblock(journal_t
*journal
, int write_op
)
1337 struct buffer_head
*bh
= journal
->j_sb_buffer
;
1338 journal_superblock_t
*sb
= journal
->j_superblock
;
1341 trace_jbd2_write_superblock(journal
, write_op
);
1342 if (!(journal
->j_flags
& JBD2_BARRIER
))
1343 write_op
&= ~(REQ_FUA
| REQ_FLUSH
);
1345 if (buffer_write_io_error(bh
)) {
1347 * Oh, dear. A previous attempt to write the journal
1348 * superblock failed. This could happen because the
1349 * USB device was yanked out. Or it could happen to
1350 * be a transient write error and maybe the block will
1351 * be remapped. Nothing we can do but to retry the
1352 * write and hope for the best.
1354 printk(KERN_ERR
"JBD2: previous I/O error detected "
1355 "for journal superblock update for %s.\n",
1356 journal
->j_devname
);
1357 clear_buffer_write_io_error(bh
);
1358 set_buffer_uptodate(bh
);
1360 jbd2_superblock_csum_set(journal
, sb
);
1362 bh
->b_end_io
= end_buffer_write_sync
;
1363 ret
= submit_bh(write_op
, bh
);
1365 if (buffer_write_io_error(bh
)) {
1366 clear_buffer_write_io_error(bh
);
1367 set_buffer_uptodate(bh
);
1371 printk(KERN_ERR
"JBD2: Error %d detected when updating "
1372 "journal superblock for %s.\n", ret
,
1373 journal
->j_devname
);
1378 * jbd2_journal_update_sb_log_tail() - Update log tail in journal sb on disk.
1379 * @journal: The journal to update.
1380 * @tail_tid: TID of the new transaction at the tail of the log
1381 * @tail_block: The first block of the transaction at the tail of the log
1382 * @write_op: With which operation should we write the journal sb
1384 * Update a journal's superblock information about log tail and write it to
1385 * disk, waiting for the IO to complete.
1387 void jbd2_journal_update_sb_log_tail(journal_t
*journal
, tid_t tail_tid
,
1388 unsigned long tail_block
, int write_op
)
1390 journal_superblock_t
*sb
= journal
->j_superblock
;
1392 BUG_ON(!mutex_is_locked(&journal
->j_checkpoint_mutex
));
1393 jbd_debug(1, "JBD2: updating superblock (start %lu, seq %u)\n",
1394 tail_block
, tail_tid
);
1396 sb
->s_sequence
= cpu_to_be32(tail_tid
);
1397 sb
->s_start
= cpu_to_be32(tail_block
);
1399 jbd2_write_superblock(journal
, write_op
);
1401 /* Log is no longer empty */
1402 write_lock(&journal
->j_state_lock
);
1403 WARN_ON(!sb
->s_sequence
);
1404 journal
->j_flags
&= ~JBD2_FLUSHED
;
1405 write_unlock(&journal
->j_state_lock
);
1409 * jbd2_mark_journal_empty() - Mark on disk journal as empty.
1410 * @journal: The journal to update.
1412 * Update a journal's dynamic superblock fields to show that journal is empty.
1413 * Write updated superblock to disk waiting for IO to complete.
1415 static void jbd2_mark_journal_empty(journal_t
*journal
)
1417 journal_superblock_t
*sb
= journal
->j_superblock
;
1419 BUG_ON(!mutex_is_locked(&journal
->j_checkpoint_mutex
));
1420 read_lock(&journal
->j_state_lock
);
1421 /* Is it already empty? */
1422 if (sb
->s_start
== 0) {
1423 read_unlock(&journal
->j_state_lock
);
1426 jbd_debug(1, "JBD2: Marking journal as empty (seq %d)\n",
1427 journal
->j_tail_sequence
);
1429 sb
->s_sequence
= cpu_to_be32(journal
->j_tail_sequence
);
1430 sb
->s_start
= cpu_to_be32(0);
1431 read_unlock(&journal
->j_state_lock
);
1433 jbd2_write_superblock(journal
, WRITE_FUA
);
1435 /* Log is no longer empty */
1436 write_lock(&journal
->j_state_lock
);
1437 journal
->j_flags
|= JBD2_FLUSHED
;
1438 write_unlock(&journal
->j_state_lock
);
1443 * jbd2_journal_update_sb_errno() - Update error in the journal.
1444 * @journal: The journal to update.
1446 * Update a journal's errno. Write updated superblock to disk waiting for IO
1449 void jbd2_journal_update_sb_errno(journal_t
*journal
)
1451 journal_superblock_t
*sb
= journal
->j_superblock
;
1453 read_lock(&journal
->j_state_lock
);
1454 jbd_debug(1, "JBD2: updating superblock error (errno %d)\n",
1456 sb
->s_errno
= cpu_to_be32(journal
->j_errno
);
1457 read_unlock(&journal
->j_state_lock
);
1459 jbd2_write_superblock(journal
, WRITE_SYNC
);
1461 EXPORT_SYMBOL(jbd2_journal_update_sb_errno
);
1464 * Read the superblock for a given journal, performing initial
1465 * validation of the format.
1467 static int journal_get_superblock(journal_t
*journal
)
1469 struct buffer_head
*bh
;
1470 journal_superblock_t
*sb
;
1473 bh
= journal
->j_sb_buffer
;
1475 J_ASSERT(bh
!= NULL
);
1476 if (!buffer_uptodate(bh
)) {
1477 ll_rw_block(READ
, 1, &bh
);
1479 if (!buffer_uptodate(bh
)) {
1481 "JBD2: IO error reading journal superblock\n");
1486 if (buffer_verified(bh
))
1489 sb
= journal
->j_superblock
;
1493 if (sb
->s_header
.h_magic
!= cpu_to_be32(JBD2_MAGIC_NUMBER
) ||
1494 sb
->s_blocksize
!= cpu_to_be32(journal
->j_blocksize
)) {
1495 printk(KERN_WARNING
"JBD2: no valid journal superblock found\n");
1499 switch(be32_to_cpu(sb
->s_header
.h_blocktype
)) {
1500 case JBD2_SUPERBLOCK_V1
:
1501 journal
->j_format_version
= 1;
1503 case JBD2_SUPERBLOCK_V2
:
1504 journal
->j_format_version
= 2;
1507 printk(KERN_WARNING
"JBD2: unrecognised superblock format ID\n");
1511 if (be32_to_cpu(sb
->s_maxlen
) < journal
->j_maxlen
)
1512 journal
->j_maxlen
= be32_to_cpu(sb
->s_maxlen
);
1513 else if (be32_to_cpu(sb
->s_maxlen
) > journal
->j_maxlen
) {
1514 printk(KERN_WARNING
"JBD2: journal file too short\n");
1518 if (be32_to_cpu(sb
->s_first
) == 0 ||
1519 be32_to_cpu(sb
->s_first
) >= journal
->j_maxlen
) {
1521 "JBD2: Invalid start block of journal: %u\n",
1522 be32_to_cpu(sb
->s_first
));
1526 if (JBD2_HAS_COMPAT_FEATURE(journal
, JBD2_FEATURE_COMPAT_CHECKSUM
) &&
1527 JBD2_HAS_INCOMPAT_FEATURE(journal
, JBD2_FEATURE_INCOMPAT_CSUM_V2
)) {
1528 /* Can't have checksum v1 and v2 on at the same time! */
1529 printk(KERN_ERR
"JBD: Can't enable checksumming v1 and v2 "
1530 "at the same time!\n");
1534 if (!jbd2_verify_csum_type(journal
, sb
)) {
1535 printk(KERN_ERR
"JBD: Unknown checksum type\n");
1539 /* Load the checksum driver */
1540 if (JBD2_HAS_INCOMPAT_FEATURE(journal
, JBD2_FEATURE_INCOMPAT_CSUM_V2
)) {
1541 journal
->j_chksum_driver
= crypto_alloc_shash("crc32c", 0, 0);
1542 if (IS_ERR(journal
->j_chksum_driver
)) {
1543 printk(KERN_ERR
"JBD: Cannot load crc32c driver.\n");
1544 err
= PTR_ERR(journal
->j_chksum_driver
);
1545 journal
->j_chksum_driver
= NULL
;
1550 /* Check superblock checksum */
1551 if (!jbd2_superblock_csum_verify(journal
, sb
)) {
1552 printk(KERN_ERR
"JBD: journal checksum error\n");
1556 /* Precompute checksum seed for all metadata */
1557 if (JBD2_HAS_INCOMPAT_FEATURE(journal
, JBD2_FEATURE_INCOMPAT_CSUM_V2
))
1558 journal
->j_csum_seed
= jbd2_chksum(journal
, ~0, sb
->s_uuid
,
1559 sizeof(sb
->s_uuid
));
1561 set_buffer_verified(bh
);
1566 journal_fail_superblock(journal
);
1571 * Load the on-disk journal superblock and read the key fields into the
1575 static int load_superblock(journal_t
*journal
)
1578 journal_superblock_t
*sb
;
1580 err
= journal_get_superblock(journal
);
1584 sb
= journal
->j_superblock
;
1586 journal
->j_tail_sequence
= be32_to_cpu(sb
->s_sequence
);
1587 journal
->j_tail
= be32_to_cpu(sb
->s_start
);
1588 journal
->j_first
= be32_to_cpu(sb
->s_first
);
1589 journal
->j_last
= be32_to_cpu(sb
->s_maxlen
);
1590 journal
->j_errno
= be32_to_cpu(sb
->s_errno
);
1597 * int jbd2_journal_load() - Read journal from disk.
1598 * @journal: Journal to act on.
1600 * Given a journal_t structure which tells us which disk blocks contain
1601 * a journal, read the journal from disk to initialise the in-memory
1604 int jbd2_journal_load(journal_t
*journal
)
1607 journal_superblock_t
*sb
;
1609 err
= load_superblock(journal
);
1613 sb
= journal
->j_superblock
;
1614 /* If this is a V2 superblock, then we have to check the
1615 * features flags on it. */
1617 if (journal
->j_format_version
>= 2) {
1618 if ((sb
->s_feature_ro_compat
&
1619 ~cpu_to_be32(JBD2_KNOWN_ROCOMPAT_FEATURES
)) ||
1620 (sb
->s_feature_incompat
&
1621 ~cpu_to_be32(JBD2_KNOWN_INCOMPAT_FEATURES
))) {
1623 "JBD2: Unrecognised features on journal\n");
1629 * Create a slab for this blocksize
1631 err
= jbd2_journal_create_slab(be32_to_cpu(sb
->s_blocksize
));
1635 /* Let the recovery code check whether it needs to recover any
1636 * data from the journal. */
1637 if (jbd2_journal_recover(journal
))
1638 goto recovery_error
;
1640 if (journal
->j_failed_commit
) {
1641 printk(KERN_ERR
"JBD2: journal transaction %u on %s "
1642 "is corrupt.\n", journal
->j_failed_commit
,
1643 journal
->j_devname
);
1647 /* OK, we've finished with the dynamic journal bits:
1648 * reinitialise the dynamic contents of the superblock in memory
1649 * and reset them on disk. */
1650 if (journal_reset(journal
))
1651 goto recovery_error
;
1653 journal
->j_flags
&= ~JBD2_ABORT
;
1654 journal
->j_flags
|= JBD2_LOADED
;
1658 printk(KERN_WARNING
"JBD2: recovery failed\n");
1663 * void jbd2_journal_destroy() - Release a journal_t structure.
1664 * @journal: Journal to act on.
1666 * Release a journal_t structure once it is no longer in use by the
1668 * Return <0 if we couldn't clean up the journal.
1670 int jbd2_journal_destroy(journal_t
*journal
)
1674 /* Wait for the commit thread to wake up and die. */
1675 journal_kill_thread(journal
);
1677 /* Force a final log commit */
1678 if (journal
->j_running_transaction
)
1679 jbd2_journal_commit_transaction(journal
);
1681 /* Force any old transactions to disk */
1683 /* Totally anal locking here... */
1684 spin_lock(&journal
->j_list_lock
);
1685 while (journal
->j_checkpoint_transactions
!= NULL
) {
1686 spin_unlock(&journal
->j_list_lock
);
1687 mutex_lock(&journal
->j_checkpoint_mutex
);
1688 jbd2_log_do_checkpoint(journal
);
1689 mutex_unlock(&journal
->j_checkpoint_mutex
);
1690 spin_lock(&journal
->j_list_lock
);
1693 J_ASSERT(journal
->j_running_transaction
== NULL
);
1694 J_ASSERT(journal
->j_committing_transaction
== NULL
);
1695 J_ASSERT(journal
->j_checkpoint_transactions
== NULL
);
1696 spin_unlock(&journal
->j_list_lock
);
1698 if (journal
->j_sb_buffer
) {
1699 if (!is_journal_aborted(journal
)) {
1700 mutex_lock(&journal
->j_checkpoint_mutex
);
1701 jbd2_mark_journal_empty(journal
);
1702 mutex_unlock(&journal
->j_checkpoint_mutex
);
1705 brelse(journal
->j_sb_buffer
);
1708 if (journal
->j_proc_entry
)
1709 jbd2_stats_proc_exit(journal
);
1710 if (journal
->j_inode
)
1711 iput(journal
->j_inode
);
1712 if (journal
->j_revoke
)
1713 jbd2_journal_destroy_revoke(journal
);
1714 if (journal
->j_chksum_driver
)
1715 crypto_free_shash(journal
->j_chksum_driver
);
1716 kfree(journal
->j_wbuf
);
1724 *int jbd2_journal_check_used_features () - Check if features specified are used.
1725 * @journal: Journal to check.
1726 * @compat: bitmask of compatible features
1727 * @ro: bitmask of features that force read-only mount
1728 * @incompat: bitmask of incompatible features
1730 * Check whether the journal uses all of a given set of
1731 * features. Return true (non-zero) if it does.
1734 int jbd2_journal_check_used_features (journal_t
*journal
, unsigned long compat
,
1735 unsigned long ro
, unsigned long incompat
)
1737 journal_superblock_t
*sb
;
1739 if (!compat
&& !ro
&& !incompat
)
1741 /* Load journal superblock if it is not loaded yet. */
1742 if (journal
->j_format_version
== 0 &&
1743 journal_get_superblock(journal
) != 0)
1745 if (journal
->j_format_version
== 1)
1748 sb
= journal
->j_superblock
;
1750 if (((be32_to_cpu(sb
->s_feature_compat
) & compat
) == compat
) &&
1751 ((be32_to_cpu(sb
->s_feature_ro_compat
) & ro
) == ro
) &&
1752 ((be32_to_cpu(sb
->s_feature_incompat
) & incompat
) == incompat
))
1759 * int jbd2_journal_check_available_features() - Check feature set in journalling layer
1760 * @journal: Journal to check.
1761 * @compat: bitmask of compatible features
1762 * @ro: bitmask of features that force read-only mount
1763 * @incompat: bitmask of incompatible features
1765 * Check whether the journaling code supports the use of
1766 * all of a given set of features on this journal. Return true
1767 * (non-zero) if it can. */
1769 int jbd2_journal_check_available_features (journal_t
*journal
, unsigned long compat
,
1770 unsigned long ro
, unsigned long incompat
)
1772 if (!compat
&& !ro
&& !incompat
)
1775 /* We can support any known requested features iff the
1776 * superblock is in version 2. Otherwise we fail to support any
1777 * extended sb features. */
1779 if (journal
->j_format_version
!= 2)
1782 if ((compat
& JBD2_KNOWN_COMPAT_FEATURES
) == compat
&&
1783 (ro
& JBD2_KNOWN_ROCOMPAT_FEATURES
) == ro
&&
1784 (incompat
& JBD2_KNOWN_INCOMPAT_FEATURES
) == incompat
)
1791 * int jbd2_journal_set_features () - Mark a given journal feature in the superblock
1792 * @journal: Journal to act on.
1793 * @compat: bitmask of compatible features
1794 * @ro: bitmask of features that force read-only mount
1795 * @incompat: bitmask of incompatible features
1797 * Mark a given journal feature as present on the
1798 * superblock. Returns true if the requested features could be set.
1802 int jbd2_journal_set_features (journal_t
*journal
, unsigned long compat
,
1803 unsigned long ro
, unsigned long incompat
)
1805 #define INCOMPAT_FEATURE_ON(f) \
1806 ((incompat & (f)) && !(sb->s_feature_incompat & cpu_to_be32(f)))
1807 #define COMPAT_FEATURE_ON(f) \
1808 ((compat & (f)) && !(sb->s_feature_compat & cpu_to_be32(f)))
1809 journal_superblock_t
*sb
;
1811 if (jbd2_journal_check_used_features(journal
, compat
, ro
, incompat
))
1814 if (!jbd2_journal_check_available_features(journal
, compat
, ro
, incompat
))
1817 /* Asking for checksumming v2 and v1? Only give them v2. */
1818 if (incompat
& JBD2_FEATURE_INCOMPAT_CSUM_V2
&&
1819 compat
& JBD2_FEATURE_COMPAT_CHECKSUM
)
1820 compat
&= ~JBD2_FEATURE_COMPAT_CHECKSUM
;
1822 jbd_debug(1, "Setting new features 0x%lx/0x%lx/0x%lx\n",
1823 compat
, ro
, incompat
);
1825 sb
= journal
->j_superblock
;
1827 /* If enabling v2 checksums, update superblock */
1828 if (INCOMPAT_FEATURE_ON(JBD2_FEATURE_INCOMPAT_CSUM_V2
)) {
1829 sb
->s_checksum_type
= JBD2_CRC32C_CHKSUM
;
1830 sb
->s_feature_compat
&=
1831 ~cpu_to_be32(JBD2_FEATURE_COMPAT_CHECKSUM
);
1833 /* Load the checksum driver */
1834 if (journal
->j_chksum_driver
== NULL
) {
1835 journal
->j_chksum_driver
= crypto_alloc_shash("crc32c",
1837 if (IS_ERR(journal
->j_chksum_driver
)) {
1838 printk(KERN_ERR
"JBD: Cannot load crc32c "
1840 journal
->j_chksum_driver
= NULL
;
1845 /* Precompute checksum seed for all metadata */
1846 if (JBD2_HAS_INCOMPAT_FEATURE(journal
,
1847 JBD2_FEATURE_INCOMPAT_CSUM_V2
))
1848 journal
->j_csum_seed
= jbd2_chksum(journal
, ~0,
1850 sizeof(sb
->s_uuid
));
1853 /* If enabling v1 checksums, downgrade superblock */
1854 if (COMPAT_FEATURE_ON(JBD2_FEATURE_COMPAT_CHECKSUM
))
1855 sb
->s_feature_incompat
&=
1856 ~cpu_to_be32(JBD2_FEATURE_INCOMPAT_CSUM_V2
);
1858 sb
->s_feature_compat
|= cpu_to_be32(compat
);
1859 sb
->s_feature_ro_compat
|= cpu_to_be32(ro
);
1860 sb
->s_feature_incompat
|= cpu_to_be32(incompat
);
1863 #undef COMPAT_FEATURE_ON
1864 #undef INCOMPAT_FEATURE_ON
1868 * jbd2_journal_clear_features () - Clear a given journal feature in the
1870 * @journal: Journal to act on.
1871 * @compat: bitmask of compatible features
1872 * @ro: bitmask of features that force read-only mount
1873 * @incompat: bitmask of incompatible features
1875 * Clear a given journal feature as present on the
1878 void jbd2_journal_clear_features(journal_t
*journal
, unsigned long compat
,
1879 unsigned long ro
, unsigned long incompat
)
1881 journal_superblock_t
*sb
;
1883 jbd_debug(1, "Clear features 0x%lx/0x%lx/0x%lx\n",
1884 compat
, ro
, incompat
);
1886 sb
= journal
->j_superblock
;
1888 sb
->s_feature_compat
&= ~cpu_to_be32(compat
);
1889 sb
->s_feature_ro_compat
&= ~cpu_to_be32(ro
);
1890 sb
->s_feature_incompat
&= ~cpu_to_be32(incompat
);
1892 EXPORT_SYMBOL(jbd2_journal_clear_features
);
1895 * int jbd2_journal_flush () - Flush journal
1896 * @journal: Journal to act on.
1898 * Flush all data for a given journal to disk and empty the journal.
1899 * Filesystems can use this when remounting readonly to ensure that
1900 * recovery does not need to happen on remount.
1903 int jbd2_journal_flush(journal_t
*journal
)
1906 transaction_t
*transaction
= NULL
;
1908 write_lock(&journal
->j_state_lock
);
1910 /* Force everything buffered to the log... */
1911 if (journal
->j_running_transaction
) {
1912 transaction
= journal
->j_running_transaction
;
1913 __jbd2_log_start_commit(journal
, transaction
->t_tid
);
1914 } else if (journal
->j_committing_transaction
)
1915 transaction
= journal
->j_committing_transaction
;
1917 /* Wait for the log commit to complete... */
1919 tid_t tid
= transaction
->t_tid
;
1921 write_unlock(&journal
->j_state_lock
);
1922 jbd2_log_wait_commit(journal
, tid
);
1924 write_unlock(&journal
->j_state_lock
);
1927 /* ...and flush everything in the log out to disk. */
1928 spin_lock(&journal
->j_list_lock
);
1929 while (!err
&& journal
->j_checkpoint_transactions
!= NULL
) {
1930 spin_unlock(&journal
->j_list_lock
);
1931 mutex_lock(&journal
->j_checkpoint_mutex
);
1932 err
= jbd2_log_do_checkpoint(journal
);
1933 mutex_unlock(&journal
->j_checkpoint_mutex
);
1934 spin_lock(&journal
->j_list_lock
);
1936 spin_unlock(&journal
->j_list_lock
);
1938 if (is_journal_aborted(journal
))
1941 mutex_lock(&journal
->j_checkpoint_mutex
);
1942 jbd2_cleanup_journal_tail(journal
);
1944 /* Finally, mark the journal as really needing no recovery.
1945 * This sets s_start==0 in the underlying superblock, which is
1946 * the magic code for a fully-recovered superblock. Any future
1947 * commits of data to the journal will restore the current
1949 jbd2_mark_journal_empty(journal
);
1950 mutex_unlock(&journal
->j_checkpoint_mutex
);
1951 write_lock(&journal
->j_state_lock
);
1952 J_ASSERT(!journal
->j_running_transaction
);
1953 J_ASSERT(!journal
->j_committing_transaction
);
1954 J_ASSERT(!journal
->j_checkpoint_transactions
);
1955 J_ASSERT(journal
->j_head
== journal
->j_tail
);
1956 J_ASSERT(journal
->j_tail_sequence
== journal
->j_transaction_sequence
);
1957 write_unlock(&journal
->j_state_lock
);
1962 * int jbd2_journal_wipe() - Wipe journal contents
1963 * @journal: Journal to act on.
1964 * @write: flag (see below)
1966 * Wipe out all of the contents of a journal, safely. This will produce
1967 * a warning if the journal contains any valid recovery information.
1968 * Must be called between journal_init_*() and jbd2_journal_load().
1970 * If 'write' is non-zero, then we wipe out the journal on disk; otherwise
1971 * we merely suppress recovery.
1974 int jbd2_journal_wipe(journal_t
*journal
, int write
)
1978 J_ASSERT (!(journal
->j_flags
& JBD2_LOADED
));
1980 err
= load_superblock(journal
);
1984 if (!journal
->j_tail
)
1987 printk(KERN_WARNING
"JBD2: %s recovery information on journal\n",
1988 write
? "Clearing" : "Ignoring");
1990 err
= jbd2_journal_skip_recovery(journal
);
1992 /* Lock to make assertions happy... */
1993 mutex_lock(&journal
->j_checkpoint_mutex
);
1994 jbd2_mark_journal_empty(journal
);
1995 mutex_unlock(&journal
->j_checkpoint_mutex
);
2003 * Journal abort has very specific semantics, which we describe
2004 * for journal abort.
2006 * Two internal functions, which provide abort to the jbd layer
2011 * Quick version for internal journal use (doesn't lock the journal).
2012 * Aborts hard --- we mark the abort as occurred, but do _nothing_ else,
2013 * and don't attempt to make any other journal updates.
2015 void __jbd2_journal_abort_hard(journal_t
*journal
)
2017 transaction_t
*transaction
;
2019 if (journal
->j_flags
& JBD2_ABORT
)
2022 printk(KERN_ERR
"Aborting journal on device %s.\n",
2023 journal
->j_devname
);
2025 write_lock(&journal
->j_state_lock
);
2026 journal
->j_flags
|= JBD2_ABORT
;
2027 transaction
= journal
->j_running_transaction
;
2029 __jbd2_log_start_commit(journal
, transaction
->t_tid
);
2030 write_unlock(&journal
->j_state_lock
);
2033 /* Soft abort: record the abort error status in the journal superblock,
2034 * but don't do any other IO. */
2035 static void __journal_abort_soft (journal_t
*journal
, int errno
)
2037 if (journal
->j_flags
& JBD2_ABORT
)
2040 if (!journal
->j_errno
)
2041 journal
->j_errno
= errno
;
2043 __jbd2_journal_abort_hard(journal
);
2046 jbd2_journal_update_sb_errno(journal
);
2050 * void jbd2_journal_abort () - Shutdown the journal immediately.
2051 * @journal: the journal to shutdown.
2052 * @errno: an error number to record in the journal indicating
2053 * the reason for the shutdown.
2055 * Perform a complete, immediate shutdown of the ENTIRE
2056 * journal (not of a single transaction). This operation cannot be
2057 * undone without closing and reopening the journal.
2059 * The jbd2_journal_abort function is intended to support higher level error
2060 * recovery mechanisms such as the ext2/ext3 remount-readonly error
2063 * Journal abort has very specific semantics. Any existing dirty,
2064 * unjournaled buffers in the main filesystem will still be written to
2065 * disk by bdflush, but the journaling mechanism will be suspended
2066 * immediately and no further transaction commits will be honoured.
2068 * Any dirty, journaled buffers will be written back to disk without
2069 * hitting the journal. Atomicity cannot be guaranteed on an aborted
2070 * filesystem, but we _do_ attempt to leave as much data as possible
2071 * behind for fsck to use for cleanup.
2073 * Any attempt to get a new transaction handle on a journal which is in
2074 * ABORT state will just result in an -EROFS error return. A
2075 * jbd2_journal_stop on an existing handle will return -EIO if we have
2076 * entered abort state during the update.
2078 * Recursive transactions are not disturbed by journal abort until the
2079 * final jbd2_journal_stop, which will receive the -EIO error.
2081 * Finally, the jbd2_journal_abort call allows the caller to supply an errno
2082 * which will be recorded (if possible) in the journal superblock. This
2083 * allows a client to record failure conditions in the middle of a
2084 * transaction without having to complete the transaction to record the
2085 * failure to disk. ext3_error, for example, now uses this
2088 * Errors which originate from within the journaling layer will NOT
2089 * supply an errno; a null errno implies that absolutely no further
2090 * writes are done to the journal (unless there are any already in
2095 void jbd2_journal_abort(journal_t
*journal
, int errno
)
2097 __journal_abort_soft(journal
, errno
);
2101 * int jbd2_journal_errno () - returns the journal's error state.
2102 * @journal: journal to examine.
2104 * This is the errno number set with jbd2_journal_abort(), the last
2105 * time the journal was mounted - if the journal was stopped
2106 * without calling abort this will be 0.
2108 * If the journal has been aborted on this mount time -EROFS will
2111 int jbd2_journal_errno(journal_t
*journal
)
2115 read_lock(&journal
->j_state_lock
);
2116 if (journal
->j_flags
& JBD2_ABORT
)
2119 err
= journal
->j_errno
;
2120 read_unlock(&journal
->j_state_lock
);
2125 * int jbd2_journal_clear_err () - clears the journal's error state
2126 * @journal: journal to act on.
2128 * An error must be cleared or acked to take a FS out of readonly
2131 int jbd2_journal_clear_err(journal_t
*journal
)
2135 write_lock(&journal
->j_state_lock
);
2136 if (journal
->j_flags
& JBD2_ABORT
)
2139 journal
->j_errno
= 0;
2140 write_unlock(&journal
->j_state_lock
);
2145 * void jbd2_journal_ack_err() - Ack journal err.
2146 * @journal: journal to act on.
2148 * An error must be cleared or acked to take a FS out of readonly
2151 void jbd2_journal_ack_err(journal_t
*journal
)
2153 write_lock(&journal
->j_state_lock
);
2154 if (journal
->j_errno
)
2155 journal
->j_flags
|= JBD2_ACK_ERR
;
2156 write_unlock(&journal
->j_state_lock
);
2159 int jbd2_journal_blocks_per_page(struct inode
*inode
)
2161 return 1 << (PAGE_CACHE_SHIFT
- inode
->i_sb
->s_blocksize_bits
);
2165 * helper functions to deal with 32 or 64bit block numbers.
2167 size_t journal_tag_bytes(journal_t
*journal
)
2169 journal_block_tag_t tag
;
2172 if (JBD2_HAS_INCOMPAT_FEATURE(journal
, JBD2_FEATURE_INCOMPAT_CSUM_V2
))
2173 x
+= sizeof(tag
.t_checksum
);
2175 if (JBD2_HAS_INCOMPAT_FEATURE(journal
, JBD2_FEATURE_INCOMPAT_64BIT
))
2176 return x
+ JBD2_TAG_SIZE64
;
2178 return x
+ JBD2_TAG_SIZE32
;
2182 * JBD memory management
2184 * These functions are used to allocate block-sized chunks of memory
2185 * used for making copies of buffer_head data. Very often it will be
2186 * page-sized chunks of data, but sometimes it will be in
2187 * sub-page-size chunks. (For example, 16k pages on Power systems
2188 * with a 4k block file system.) For blocks smaller than a page, we
2189 * use a SLAB allocator. There are slab caches for each block size,
2190 * which are allocated at mount time, if necessary, and we only free
2191 * (all of) the slab caches when/if the jbd2 module is unloaded. For
2192 * this reason we don't need to a mutex to protect access to
2193 * jbd2_slab[] allocating or releasing memory; only in
2194 * jbd2_journal_create_slab().
2196 #define JBD2_MAX_SLABS 8
2197 static struct kmem_cache
*jbd2_slab
[JBD2_MAX_SLABS
];
2199 static const char *jbd2_slab_names
[JBD2_MAX_SLABS
] = {
2200 "jbd2_1k", "jbd2_2k", "jbd2_4k", "jbd2_8k",
2201 "jbd2_16k", "jbd2_32k", "jbd2_64k", "jbd2_128k"
2205 static void jbd2_journal_destroy_slabs(void)
2209 for (i
= 0; i
< JBD2_MAX_SLABS
; i
++) {
2211 kmem_cache_destroy(jbd2_slab
[i
]);
2212 jbd2_slab
[i
] = NULL
;
2216 static int jbd2_journal_create_slab(size_t size
)
2218 static DEFINE_MUTEX(jbd2_slab_create_mutex
);
2219 int i
= order_base_2(size
) - 10;
2222 if (size
== PAGE_SIZE
)
2225 if (i
>= JBD2_MAX_SLABS
)
2228 if (unlikely(i
< 0))
2230 mutex_lock(&jbd2_slab_create_mutex
);
2232 mutex_unlock(&jbd2_slab_create_mutex
);
2233 return 0; /* Already created */
2236 slab_size
= 1 << (i
+10);
2237 jbd2_slab
[i
] = kmem_cache_create(jbd2_slab_names
[i
], slab_size
,
2238 slab_size
, 0, NULL
);
2239 mutex_unlock(&jbd2_slab_create_mutex
);
2240 if (!jbd2_slab
[i
]) {
2241 printk(KERN_EMERG
"JBD2: no memory for jbd2_slab cache\n");
2247 static struct kmem_cache
*get_slab(size_t size
)
2249 int i
= order_base_2(size
) - 10;
2251 BUG_ON(i
>= JBD2_MAX_SLABS
);
2252 if (unlikely(i
< 0))
2254 BUG_ON(jbd2_slab
[i
] == NULL
);
2255 return jbd2_slab
[i
];
2258 void *jbd2_alloc(size_t size
, gfp_t flags
)
2262 BUG_ON(size
& (size
-1)); /* Must be a power of 2 */
2264 flags
|= __GFP_REPEAT
;
2265 if (size
== PAGE_SIZE
)
2266 ptr
= (void *)__get_free_pages(flags
, 0);
2267 else if (size
> PAGE_SIZE
) {
2268 int order
= get_order(size
);
2271 ptr
= (void *)__get_free_pages(flags
, order
);
2273 ptr
= vmalloc(size
);
2275 ptr
= kmem_cache_alloc(get_slab(size
), flags
);
2277 /* Check alignment; SLUB has gotten this wrong in the past,
2278 * and this can lead to user data corruption! */
2279 BUG_ON(((unsigned long) ptr
) & (size
-1));
2284 void jbd2_free(void *ptr
, size_t size
)
2286 if (size
== PAGE_SIZE
) {
2287 free_pages((unsigned long)ptr
, 0);
2290 if (size
> PAGE_SIZE
) {
2291 int order
= get_order(size
);
2294 free_pages((unsigned long)ptr
, order
);
2299 kmem_cache_free(get_slab(size
), ptr
);
2303 * Journal_head storage management
2305 static struct kmem_cache
*jbd2_journal_head_cache
;
2306 #ifdef CONFIG_JBD2_DEBUG
2307 static atomic_t nr_journal_heads
= ATOMIC_INIT(0);
2310 static int jbd2_journal_init_journal_head_cache(void)
2314 J_ASSERT(jbd2_journal_head_cache
== NULL
);
2315 jbd2_journal_head_cache
= kmem_cache_create("jbd2_journal_head",
2316 sizeof(struct journal_head
),
2318 SLAB_TEMPORARY
, /* flags */
2321 if (!jbd2_journal_head_cache
) {
2323 printk(KERN_EMERG
"JBD2: no memory for journal_head cache\n");
2328 static void jbd2_journal_destroy_journal_head_cache(void)
2330 if (jbd2_journal_head_cache
) {
2331 kmem_cache_destroy(jbd2_journal_head_cache
);
2332 jbd2_journal_head_cache
= NULL
;
2337 * journal_head splicing and dicing
2339 static struct journal_head
*journal_alloc_journal_head(void)
2341 struct journal_head
*ret
;
2343 #ifdef CONFIG_JBD2_DEBUG
2344 atomic_inc(&nr_journal_heads
);
2346 ret
= kmem_cache_zalloc(jbd2_journal_head_cache
, GFP_NOFS
);
2348 jbd_debug(1, "out of memory for journal_head\n");
2349 pr_notice_ratelimited("ENOMEM in %s, retrying.\n", __func__
);
2352 ret
= kmem_cache_zalloc(jbd2_journal_head_cache
, GFP_NOFS
);
2358 static void journal_free_journal_head(struct journal_head
*jh
)
2360 #ifdef CONFIG_JBD2_DEBUG
2361 atomic_dec(&nr_journal_heads
);
2362 memset(jh
, JBD2_POISON_FREE
, sizeof(*jh
));
2364 kmem_cache_free(jbd2_journal_head_cache
, jh
);
2368 * A journal_head is attached to a buffer_head whenever JBD has an
2369 * interest in the buffer.
2371 * Whenever a buffer has an attached journal_head, its ->b_state:BH_JBD bit
2372 * is set. This bit is tested in core kernel code where we need to take
2373 * JBD-specific actions. Testing the zeroness of ->b_private is not reliable
2376 * When a buffer has its BH_JBD bit set, its ->b_count is elevated by one.
2378 * When a buffer has its BH_JBD bit set it is immune from being released by
2379 * core kernel code, mainly via ->b_count.
2381 * A journal_head is detached from its buffer_head when the journal_head's
2382 * b_jcount reaches zero. Running transaction (b_transaction) and checkpoint
2383 * transaction (b_cp_transaction) hold their references to b_jcount.
2385 * Various places in the kernel want to attach a journal_head to a buffer_head
2386 * _before_ attaching the journal_head to a transaction. To protect the
2387 * journal_head in this situation, jbd2_journal_add_journal_head elevates the
2388 * journal_head's b_jcount refcount by one. The caller must call
2389 * jbd2_journal_put_journal_head() to undo this.
2391 * So the typical usage would be:
2393 * (Attach a journal_head if needed. Increments b_jcount)
2394 * struct journal_head *jh = jbd2_journal_add_journal_head(bh);
2396 * (Get another reference for transaction)
2397 * jbd2_journal_grab_journal_head(bh);
2398 * jh->b_transaction = xxx;
2399 * (Put original reference)
2400 * jbd2_journal_put_journal_head(jh);
2404 * Give a buffer_head a journal_head.
2408 struct journal_head
*jbd2_journal_add_journal_head(struct buffer_head
*bh
)
2410 struct journal_head
*jh
;
2411 struct journal_head
*new_jh
= NULL
;
2414 if (!buffer_jbd(bh
))
2415 new_jh
= journal_alloc_journal_head();
2417 jbd_lock_bh_journal_head(bh
);
2418 if (buffer_jbd(bh
)) {
2422 (atomic_read(&bh
->b_count
) > 0) ||
2423 (bh
->b_page
&& bh
->b_page
->mapping
));
2426 jbd_unlock_bh_journal_head(bh
);
2431 new_jh
= NULL
; /* We consumed it */
2436 BUFFER_TRACE(bh
, "added journal_head");
2439 jbd_unlock_bh_journal_head(bh
);
2441 journal_free_journal_head(new_jh
);
2442 return bh
->b_private
;
2446 * Grab a ref against this buffer_head's journal_head. If it ended up not
2447 * having a journal_head, return NULL
2449 struct journal_head
*jbd2_journal_grab_journal_head(struct buffer_head
*bh
)
2451 struct journal_head
*jh
= NULL
;
2453 jbd_lock_bh_journal_head(bh
);
2454 if (buffer_jbd(bh
)) {
2458 jbd_unlock_bh_journal_head(bh
);
2462 static void __journal_remove_journal_head(struct buffer_head
*bh
)
2464 struct journal_head
*jh
= bh2jh(bh
);
2466 J_ASSERT_JH(jh
, jh
->b_jcount
>= 0);
2467 J_ASSERT_JH(jh
, jh
->b_transaction
== NULL
);
2468 J_ASSERT_JH(jh
, jh
->b_next_transaction
== NULL
);
2469 J_ASSERT_JH(jh
, jh
->b_cp_transaction
== NULL
);
2470 J_ASSERT_JH(jh
, jh
->b_jlist
== BJ_None
);
2471 J_ASSERT_BH(bh
, buffer_jbd(bh
));
2472 J_ASSERT_BH(bh
, jh2bh(jh
) == bh
);
2473 BUFFER_TRACE(bh
, "remove journal_head");
2474 if (jh
->b_frozen_data
) {
2475 printk(KERN_WARNING
"%s: freeing b_frozen_data\n", __func__
);
2476 jbd2_free(jh
->b_frozen_data
, bh
->b_size
);
2478 if (jh
->b_committed_data
) {
2479 printk(KERN_WARNING
"%s: freeing b_committed_data\n", __func__
);
2480 jbd2_free(jh
->b_committed_data
, bh
->b_size
);
2482 bh
->b_private
= NULL
;
2483 jh
->b_bh
= NULL
; /* debug, really */
2484 clear_buffer_jbd(bh
);
2485 journal_free_journal_head(jh
);
2489 * Drop a reference on the passed journal_head. If it fell to zero then
2490 * release the journal_head from the buffer_head.
2492 void jbd2_journal_put_journal_head(struct journal_head
*jh
)
2494 struct buffer_head
*bh
= jh2bh(jh
);
2496 jbd_lock_bh_journal_head(bh
);
2497 J_ASSERT_JH(jh
, jh
->b_jcount
> 0);
2499 if (!jh
->b_jcount
) {
2500 __journal_remove_journal_head(bh
);
2501 jbd_unlock_bh_journal_head(bh
);
2504 jbd_unlock_bh_journal_head(bh
);
2508 * Initialize jbd inode head
2510 void jbd2_journal_init_jbd_inode(struct jbd2_inode
*jinode
, struct inode
*inode
)
2512 jinode
->i_transaction
= NULL
;
2513 jinode
->i_next_transaction
= NULL
;
2514 jinode
->i_vfs_inode
= inode
;
2515 jinode
->i_flags
= 0;
2516 INIT_LIST_HEAD(&jinode
->i_list
);
2520 * Function to be called before we start removing inode from memory (i.e.,
2521 * clear_inode() is a fine place to be called from). It removes inode from
2522 * transaction's lists.
2524 void jbd2_journal_release_jbd_inode(journal_t
*journal
,
2525 struct jbd2_inode
*jinode
)
2530 spin_lock(&journal
->j_list_lock
);
2531 /* Is commit writing out inode - we have to wait */
2532 if (test_bit(__JI_COMMIT_RUNNING
, &jinode
->i_flags
)) {
2533 wait_queue_head_t
*wq
;
2534 DEFINE_WAIT_BIT(wait
, &jinode
->i_flags
, __JI_COMMIT_RUNNING
);
2535 wq
= bit_waitqueue(&jinode
->i_flags
, __JI_COMMIT_RUNNING
);
2536 prepare_to_wait(wq
, &wait
.wait
, TASK_UNINTERRUPTIBLE
);
2537 spin_unlock(&journal
->j_list_lock
);
2539 finish_wait(wq
, &wait
.wait
);
2543 if (jinode
->i_transaction
) {
2544 list_del(&jinode
->i_list
);
2545 jinode
->i_transaction
= NULL
;
2547 spin_unlock(&journal
->j_list_lock
);
2551 #ifdef CONFIG_PROC_FS
2553 #define JBD2_STATS_PROC_NAME "fs/jbd2"
2555 static void __init
jbd2_create_jbd_stats_proc_entry(void)
2557 proc_jbd2_stats
= proc_mkdir(JBD2_STATS_PROC_NAME
, NULL
);
2560 static void __exit
jbd2_remove_jbd_stats_proc_entry(void)
2562 if (proc_jbd2_stats
)
2563 remove_proc_entry(JBD2_STATS_PROC_NAME
, NULL
);
2568 #define jbd2_create_jbd_stats_proc_entry() do {} while (0)
2569 #define jbd2_remove_jbd_stats_proc_entry() do {} while (0)
2573 struct kmem_cache
*jbd2_handle_cache
, *jbd2_inode_cache
;
2575 static int __init
jbd2_journal_init_handle_cache(void)
2577 jbd2_handle_cache
= KMEM_CACHE(jbd2_journal_handle
, SLAB_TEMPORARY
);
2578 if (jbd2_handle_cache
== NULL
) {
2579 printk(KERN_EMERG
"JBD2: failed to create handle cache\n");
2582 jbd2_inode_cache
= KMEM_CACHE(jbd2_inode
, 0);
2583 if (jbd2_inode_cache
== NULL
) {
2584 printk(KERN_EMERG
"JBD2: failed to create inode cache\n");
2585 kmem_cache_destroy(jbd2_handle_cache
);
2591 static void jbd2_journal_destroy_handle_cache(void)
2593 if (jbd2_handle_cache
)
2594 kmem_cache_destroy(jbd2_handle_cache
);
2595 if (jbd2_inode_cache
)
2596 kmem_cache_destroy(jbd2_inode_cache
);
2601 * Module startup and shutdown
2604 static int __init
journal_init_caches(void)
2608 ret
= jbd2_journal_init_revoke_caches();
2610 ret
= jbd2_journal_init_journal_head_cache();
2612 ret
= jbd2_journal_init_handle_cache();
2614 ret
= jbd2_journal_init_transaction_cache();
2618 static void jbd2_journal_destroy_caches(void)
2620 jbd2_journal_destroy_revoke_caches();
2621 jbd2_journal_destroy_journal_head_cache();
2622 jbd2_journal_destroy_handle_cache();
2623 jbd2_journal_destroy_transaction_cache();
2624 jbd2_journal_destroy_slabs();
2627 static int __init
journal_init(void)
2631 BUILD_BUG_ON(sizeof(struct journal_superblock_s
) != 1024);
2633 ret
= journal_init_caches();
2635 jbd2_create_jbd_stats_proc_entry();
2637 jbd2_journal_destroy_caches();
2642 static void __exit
journal_exit(void)
2644 #ifdef CONFIG_JBD2_DEBUG
2645 int n
= atomic_read(&nr_journal_heads
);
2647 printk(KERN_EMERG
"JBD2: leaked %d journal_heads!\n", n
);
2649 jbd2_remove_jbd_stats_proc_entry();
2650 jbd2_journal_destroy_caches();
2653 MODULE_LICENSE("GPL");
2654 module_init(journal_init
);
2655 module_exit(journal_exit
);