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 static int jbd2_verify_csum_type(journal_t
*j
, journal_superblock_t
*sb
)
127 if (!jbd2_journal_has_csum_v2or3_feature(j
))
130 return sb
->s_checksum_type
== JBD2_CRC32C_CHKSUM
;
133 static __be32
jbd2_superblock_csum(journal_t
*j
, journal_superblock_t
*sb
)
138 old_csum
= sb
->s_checksum
;
140 csum
= jbd2_chksum(j
, ~0, (char *)sb
, sizeof(journal_superblock_t
));
141 sb
->s_checksum
= old_csum
;
143 return cpu_to_be32(csum
);
146 static int jbd2_superblock_csum_verify(journal_t
*j
, journal_superblock_t
*sb
)
148 if (!jbd2_journal_has_csum_v2or3(j
))
151 return sb
->s_checksum
== jbd2_superblock_csum(j
, sb
);
154 static void jbd2_superblock_csum_set(journal_t
*j
, journal_superblock_t
*sb
)
156 if (!jbd2_journal_has_csum_v2or3(j
))
159 sb
->s_checksum
= jbd2_superblock_csum(j
, sb
);
163 * Helper function used to manage commit timeouts
166 static void commit_timeout(unsigned long __data
)
168 struct task_struct
* p
= (struct task_struct
*) __data
;
174 * kjournald2: The main thread function used to manage a logging device
177 * This kernel thread is responsible for two things:
179 * 1) COMMIT: Every so often we need to commit the current state of the
180 * filesystem to disk. The journal thread is responsible for writing
181 * all of the metadata buffers to disk.
183 * 2) CHECKPOINT: We cannot reuse a used section of the log file until all
184 * of the data in that part of the log has been rewritten elsewhere on
185 * the disk. Flushing these old buffers to reclaim space in the log is
186 * known as checkpointing, and this thread is responsible for that job.
189 static int kjournald2(void *arg
)
191 journal_t
*journal
= arg
;
192 transaction_t
*transaction
;
195 * Set up an interval timer which can be used to trigger a commit wakeup
196 * after the commit interval expires
198 setup_timer(&journal
->j_commit_timer
, commit_timeout
,
199 (unsigned long)current
);
203 /* Record that the journal thread is running */
204 journal
->j_task
= current
;
205 wake_up(&journal
->j_wait_done_commit
);
208 * And now, wait forever for commit wakeup events.
210 write_lock(&journal
->j_state_lock
);
213 if (journal
->j_flags
& JBD2_UNMOUNT
)
216 jbd_debug(1, "commit_sequence=%d, commit_request=%d\n",
217 journal
->j_commit_sequence
, journal
->j_commit_request
);
219 if (journal
->j_commit_sequence
!= journal
->j_commit_request
) {
220 jbd_debug(1, "OK, requests differ\n");
221 write_unlock(&journal
->j_state_lock
);
222 del_timer_sync(&journal
->j_commit_timer
);
223 jbd2_journal_commit_transaction(journal
);
224 write_lock(&journal
->j_state_lock
);
228 wake_up(&journal
->j_wait_done_commit
);
229 if (freezing(current
)) {
231 * The simpler the better. Flushing journal isn't a
232 * good idea, because that depends on threads that may
233 * be already stopped.
235 jbd_debug(1, "Now suspending kjournald2\n");
236 write_unlock(&journal
->j_state_lock
);
238 write_lock(&journal
->j_state_lock
);
241 * We assume on resume that commits are already there,
245 int should_sleep
= 1;
247 prepare_to_wait(&journal
->j_wait_commit
, &wait
,
249 if (journal
->j_commit_sequence
!= journal
->j_commit_request
)
251 transaction
= journal
->j_running_transaction
;
252 if (transaction
&& time_after_eq(jiffies
,
253 transaction
->t_expires
))
255 if (journal
->j_flags
& JBD2_UNMOUNT
)
258 write_unlock(&journal
->j_state_lock
);
260 write_lock(&journal
->j_state_lock
);
262 finish_wait(&journal
->j_wait_commit
, &wait
);
265 jbd_debug(1, "kjournald2 wakes\n");
268 * Were we woken up by a commit wakeup event?
270 transaction
= journal
->j_running_transaction
;
271 if (transaction
&& time_after_eq(jiffies
, transaction
->t_expires
)) {
272 journal
->j_commit_request
= transaction
->t_tid
;
273 jbd_debug(1, "woke because of timeout\n");
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");
282 write_unlock(&journal
->j_state_lock
);
286 static int jbd2_journal_start_thread(journal_t
*journal
)
288 struct task_struct
*t
;
290 t
= kthread_run(kjournald2
, journal
, "jbd2/%s",
295 wait_event(journal
->j_wait_done_commit
, journal
->j_task
!= NULL
);
299 static void journal_kill_thread(journal_t
*journal
)
301 write_lock(&journal
->j_state_lock
);
302 journal
->j_flags
|= JBD2_UNMOUNT
;
304 while (journal
->j_task
) {
305 write_unlock(&journal
->j_state_lock
);
306 wake_up(&journal
->j_wait_commit
);
307 wait_event(journal
->j_wait_done_commit
, journal
->j_task
== NULL
);
308 write_lock(&journal
->j_state_lock
);
310 write_unlock(&journal
->j_state_lock
);
314 * jbd2_journal_write_metadata_buffer: write a metadata buffer to the journal.
316 * Writes a metadata buffer to a given disk block. The actual IO is not
317 * performed but a new buffer_head is constructed which labels the data
318 * to be written with the correct destination disk block.
320 * Any magic-number escaping which needs to be done will cause a
321 * copy-out here. If the buffer happens to start with the
322 * JBD2_MAGIC_NUMBER, then we can't write it to the log directly: the
323 * magic number is only written to the log for descripter blocks. In
324 * this case, we copy the data and replace the first word with 0, and we
325 * return a result code which indicates that this buffer needs to be
326 * marked as an escaped buffer in the corresponding log descriptor
327 * block. The missing word can then be restored when the block is read
330 * If the source buffer has already been modified by a new transaction
331 * since we took the last commit snapshot, we use the frozen copy of
332 * that data for IO. If we end up using the existing buffer_head's data
333 * for the write, then we have to make sure nobody modifies it while the
334 * IO is in progress. do_get_write_access() handles this.
336 * The function returns a pointer to the buffer_head to be used for IO.
344 * Bit 0 set == escape performed on the data
345 * Bit 1 set == buffer copy-out performed (kfree the data after IO)
348 int jbd2_journal_write_metadata_buffer(transaction_t
*transaction
,
349 struct journal_head
*jh_in
,
350 struct buffer_head
**bh_out
,
353 int need_copy_out
= 0;
354 int done_copy_out
= 0;
357 struct buffer_head
*new_bh
;
358 struct page
*new_page
;
359 unsigned int new_offset
;
360 struct buffer_head
*bh_in
= jh2bh(jh_in
);
361 journal_t
*journal
= transaction
->t_journal
;
364 * The buffer really shouldn't be locked: only the current committing
365 * transaction is allowed to write it, so nobody else is allowed
368 * akpm: except if we're journalling data, and write() output is
369 * also part of a shared mapping, and another thread has
370 * decided to launch a writepage() against this buffer.
372 J_ASSERT_BH(bh_in
, buffer_jbddirty(bh_in
));
374 new_bh
= alloc_buffer_head(GFP_NOFS
|__GFP_NOFAIL
);
376 /* keep subsequent assertions sane */
377 atomic_set(&new_bh
->b_count
, 1);
379 jbd_lock_bh_state(bh_in
);
382 * If a new transaction has already done a buffer copy-out, then
383 * we use that version of the data for the commit.
385 if (jh_in
->b_frozen_data
) {
387 new_page
= virt_to_page(jh_in
->b_frozen_data
);
388 new_offset
= offset_in_page(jh_in
->b_frozen_data
);
390 new_page
= jh2bh(jh_in
)->b_page
;
391 new_offset
= offset_in_page(jh2bh(jh_in
)->b_data
);
394 mapped_data
= kmap_atomic(new_page
);
396 * Fire data frozen trigger if data already wasn't frozen. Do this
397 * before checking for escaping, as the trigger may modify the magic
398 * offset. If a copy-out happens afterwards, it will have the correct
399 * data in the buffer.
402 jbd2_buffer_frozen_trigger(jh_in
, mapped_data
+ new_offset
,
408 if (*((__be32
*)(mapped_data
+ new_offset
)) ==
409 cpu_to_be32(JBD2_MAGIC_NUMBER
)) {
413 kunmap_atomic(mapped_data
);
416 * Do we need to do a data copy?
418 if (need_copy_out
&& !done_copy_out
) {
421 jbd_unlock_bh_state(bh_in
);
422 tmp
= jbd2_alloc(bh_in
->b_size
, GFP_NOFS
);
427 jbd_lock_bh_state(bh_in
);
428 if (jh_in
->b_frozen_data
) {
429 jbd2_free(tmp
, bh_in
->b_size
);
433 jh_in
->b_frozen_data
= tmp
;
434 mapped_data
= kmap_atomic(new_page
);
435 memcpy(tmp
, mapped_data
+ new_offset
, bh_in
->b_size
);
436 kunmap_atomic(mapped_data
);
438 new_page
= virt_to_page(tmp
);
439 new_offset
= offset_in_page(tmp
);
443 * This isn't strictly necessary, as we're using frozen
444 * data for the escaping, but it keeps consistency with
445 * b_frozen_data usage.
447 jh_in
->b_frozen_triggers
= jh_in
->b_triggers
;
451 * Did we need to do an escaping? Now we've done all the
452 * copying, we can finally do so.
455 mapped_data
= kmap_atomic(new_page
);
456 *((unsigned int *)(mapped_data
+ new_offset
)) = 0;
457 kunmap_atomic(mapped_data
);
460 set_bh_page(new_bh
, new_page
, new_offset
);
461 new_bh
->b_size
= bh_in
->b_size
;
462 new_bh
->b_bdev
= journal
->j_dev
;
463 new_bh
->b_blocknr
= blocknr
;
464 new_bh
->b_private
= bh_in
;
465 set_buffer_mapped(new_bh
);
466 set_buffer_dirty(new_bh
);
471 * The to-be-written buffer needs to get moved to the io queue,
472 * and the original buffer whose contents we are shadowing or
473 * copying is moved to the transaction's shadow queue.
475 JBUFFER_TRACE(jh_in
, "file as BJ_Shadow");
476 spin_lock(&journal
->j_list_lock
);
477 __jbd2_journal_file_buffer(jh_in
, transaction
, BJ_Shadow
);
478 spin_unlock(&journal
->j_list_lock
);
479 set_buffer_shadow(bh_in
);
480 jbd_unlock_bh_state(bh_in
);
482 return do_escape
| (done_copy_out
<< 1);
486 * Allocation code for the journal file. Manage the space left in the
487 * journal, so that we can begin checkpointing when appropriate.
491 * Called with j_state_lock locked for writing.
492 * Returns true if a transaction commit was started.
494 int __jbd2_log_start_commit(journal_t
*journal
, tid_t target
)
496 /* Return if the txn has already requested to be committed */
497 if (journal
->j_commit_request
== target
)
501 * The only transaction we can possibly wait upon is the
502 * currently running transaction (if it exists). Otherwise,
503 * the target tid must be an old one.
505 if (journal
->j_running_transaction
&&
506 journal
->j_running_transaction
->t_tid
== target
) {
508 * We want a new commit: OK, mark the request and wakeup the
509 * commit thread. We do _not_ do the commit ourselves.
512 journal
->j_commit_request
= target
;
513 jbd_debug(1, "JBD2: requesting commit %d/%d\n",
514 journal
->j_commit_request
,
515 journal
->j_commit_sequence
);
516 journal
->j_running_transaction
->t_requested
= jiffies
;
517 wake_up(&journal
->j_wait_commit
);
519 } else if (!tid_geq(journal
->j_commit_request
, target
))
520 /* This should never happen, but if it does, preserve
521 the evidence before kjournald goes into a loop and
522 increments j_commit_sequence beyond all recognition. */
523 WARN_ONCE(1, "JBD2: bad log_start_commit: %u %u %u %u\n",
524 journal
->j_commit_request
,
525 journal
->j_commit_sequence
,
526 target
, journal
->j_running_transaction
?
527 journal
->j_running_transaction
->t_tid
: 0);
531 int jbd2_log_start_commit(journal_t
*journal
, tid_t tid
)
535 write_lock(&journal
->j_state_lock
);
536 ret
= __jbd2_log_start_commit(journal
, tid
);
537 write_unlock(&journal
->j_state_lock
);
542 * Force and wait any uncommitted transactions. We can only force the running
543 * transaction if we don't have an active handle, otherwise, we will deadlock.
544 * Returns: <0 in case of error,
545 * 0 if nothing to commit,
546 * 1 if transaction was successfully committed.
548 static int __jbd2_journal_force_commit(journal_t
*journal
)
550 transaction_t
*transaction
= NULL
;
552 int need_to_start
= 0, ret
= 0;
554 read_lock(&journal
->j_state_lock
);
555 if (journal
->j_running_transaction
&& !current
->journal_info
) {
556 transaction
= journal
->j_running_transaction
;
557 if (!tid_geq(journal
->j_commit_request
, transaction
->t_tid
))
559 } else if (journal
->j_committing_transaction
)
560 transaction
= journal
->j_committing_transaction
;
563 /* Nothing to commit */
564 read_unlock(&journal
->j_state_lock
);
567 tid
= transaction
->t_tid
;
568 read_unlock(&journal
->j_state_lock
);
570 jbd2_log_start_commit(journal
, tid
);
571 ret
= jbd2_log_wait_commit(journal
, tid
);
579 * Force and wait upon a commit if the calling process is not within
580 * transaction. This is used for forcing out undo-protected data which contains
581 * bitmaps, when the fs is running out of space.
583 * @journal: journal to force
584 * Returns true if progress was made.
586 int jbd2_journal_force_commit_nested(journal_t
*journal
)
590 ret
= __jbd2_journal_force_commit(journal
);
595 * int journal_force_commit() - force any uncommitted transactions
596 * @journal: journal to force
598 * Caller want unconditional commit. We can only force the running transaction
599 * if we don't have an active handle, otherwise, we will deadlock.
601 int jbd2_journal_force_commit(journal_t
*journal
)
605 J_ASSERT(!current
->journal_info
);
606 ret
= __jbd2_journal_force_commit(journal
);
613 * Start a commit of the current running transaction (if any). Returns true
614 * if a transaction is going to be committed (or is currently already
615 * committing), and fills its tid in at *ptid
617 int jbd2_journal_start_commit(journal_t
*journal
, tid_t
*ptid
)
621 write_lock(&journal
->j_state_lock
);
622 if (journal
->j_running_transaction
) {
623 tid_t tid
= journal
->j_running_transaction
->t_tid
;
625 __jbd2_log_start_commit(journal
, tid
);
626 /* There's a running transaction and we've just made sure
627 * it's commit has been scheduled. */
631 } else if (journal
->j_committing_transaction
) {
633 * If commit has been started, then we have to wait for
634 * completion of that transaction.
637 *ptid
= journal
->j_committing_transaction
->t_tid
;
640 write_unlock(&journal
->j_state_lock
);
645 * Return 1 if a given transaction has not yet sent barrier request
646 * connected with a transaction commit. If 0 is returned, transaction
647 * may or may not have sent the barrier. Used to avoid sending barrier
648 * twice in common cases.
650 int jbd2_trans_will_send_data_barrier(journal_t
*journal
, tid_t tid
)
653 transaction_t
*commit_trans
;
655 if (!(journal
->j_flags
& JBD2_BARRIER
))
657 read_lock(&journal
->j_state_lock
);
658 /* Transaction already committed? */
659 if (tid_geq(journal
->j_commit_sequence
, tid
))
661 commit_trans
= journal
->j_committing_transaction
;
662 if (!commit_trans
|| commit_trans
->t_tid
!= tid
) {
667 * Transaction is being committed and we already proceeded to
668 * submitting a flush to fs partition?
670 if (journal
->j_fs_dev
!= journal
->j_dev
) {
671 if (!commit_trans
->t_need_data_flush
||
672 commit_trans
->t_state
>= T_COMMIT_DFLUSH
)
675 if (commit_trans
->t_state
>= T_COMMIT_JFLUSH
)
680 read_unlock(&journal
->j_state_lock
);
683 EXPORT_SYMBOL(jbd2_trans_will_send_data_barrier
);
686 * Wait for a specified commit to complete.
687 * The caller may not hold the journal lock.
689 int jbd2_log_wait_commit(journal_t
*journal
, tid_t tid
)
693 read_lock(&journal
->j_state_lock
);
694 #ifdef CONFIG_JBD2_DEBUG
695 if (!tid_geq(journal
->j_commit_request
, tid
)) {
697 "%s: error: j_commit_request=%d, tid=%d\n",
698 __func__
, journal
->j_commit_request
, tid
);
701 while (tid_gt(tid
, journal
->j_commit_sequence
)) {
702 jbd_debug(1, "JBD2: want %d, j_commit_sequence=%d\n",
703 tid
, journal
->j_commit_sequence
);
704 read_unlock(&journal
->j_state_lock
);
705 wake_up(&journal
->j_wait_commit
);
706 wait_event(journal
->j_wait_done_commit
,
707 !tid_gt(tid
, journal
->j_commit_sequence
));
708 read_lock(&journal
->j_state_lock
);
710 read_unlock(&journal
->j_state_lock
);
712 if (unlikely(is_journal_aborted(journal
)))
718 * When this function returns the transaction corresponding to tid
719 * will be completed. If the transaction has currently running, start
720 * committing that transaction before waiting for it to complete. If
721 * the transaction id is stale, it is by definition already completed,
722 * so just return SUCCESS.
724 int jbd2_complete_transaction(journal_t
*journal
, tid_t tid
)
726 int need_to_wait
= 1;
728 read_lock(&journal
->j_state_lock
);
729 if (journal
->j_running_transaction
&&
730 journal
->j_running_transaction
->t_tid
== tid
) {
731 if (journal
->j_commit_request
!= tid
) {
732 /* transaction not yet started, so request it */
733 read_unlock(&journal
->j_state_lock
);
734 jbd2_log_start_commit(journal
, tid
);
737 } else if (!(journal
->j_committing_transaction
&&
738 journal
->j_committing_transaction
->t_tid
== tid
))
740 read_unlock(&journal
->j_state_lock
);
744 return jbd2_log_wait_commit(journal
, tid
);
746 EXPORT_SYMBOL(jbd2_complete_transaction
);
749 * Log buffer allocation routines:
752 int jbd2_journal_next_log_block(journal_t
*journal
, unsigned long long *retp
)
754 unsigned long blocknr
;
756 write_lock(&journal
->j_state_lock
);
757 J_ASSERT(journal
->j_free
> 1);
759 blocknr
= journal
->j_head
;
762 if (journal
->j_head
== journal
->j_last
)
763 journal
->j_head
= journal
->j_first
;
764 write_unlock(&journal
->j_state_lock
);
765 return jbd2_journal_bmap(journal
, blocknr
, retp
);
769 * Conversion of logical to physical block numbers for the journal
771 * On external journals the journal blocks are identity-mapped, so
772 * this is a no-op. If needed, we can use j_blk_offset - everything is
775 int jbd2_journal_bmap(journal_t
*journal
, unsigned long blocknr
,
776 unsigned long long *retp
)
779 unsigned long long ret
;
781 if (journal
->j_inode
) {
782 ret
= bmap(journal
->j_inode
, blocknr
);
786 printk(KERN_ALERT
"%s: journal block not found "
787 "at offset %lu on %s\n",
788 __func__
, blocknr
, journal
->j_devname
);
790 __journal_abort_soft(journal
, err
);
793 *retp
= blocknr
; /* +journal->j_blk_offset */
799 * We play buffer_head aliasing tricks to write data/metadata blocks to
800 * the journal without copying their contents, but for journal
801 * descriptor blocks we do need to generate bona fide buffers.
803 * After the caller of jbd2_journal_get_descriptor_buffer() has finished modifying
804 * the buffer's contents they really should run flush_dcache_page(bh->b_page).
805 * But we don't bother doing that, so there will be coherency problems with
806 * mmaps of blockdevs which hold live JBD-controlled filesystems.
808 struct buffer_head
*jbd2_journal_get_descriptor_buffer(journal_t
*journal
)
810 struct buffer_head
*bh
;
811 unsigned long long blocknr
;
814 err
= jbd2_journal_next_log_block(journal
, &blocknr
);
819 bh
= __getblk(journal
->j_dev
, blocknr
, journal
->j_blocksize
);
823 memset(bh
->b_data
, 0, journal
->j_blocksize
);
824 set_buffer_uptodate(bh
);
826 BUFFER_TRACE(bh
, "return this buffer");
831 * Return tid of the oldest transaction in the journal and block in the journal
832 * where the transaction starts.
834 * If the journal is now empty, return which will be the next transaction ID
835 * we will write and where will that transaction start.
837 * The return value is 0 if journal tail cannot be pushed any further, 1 if
840 int jbd2_journal_get_log_tail(journal_t
*journal
, tid_t
*tid
,
841 unsigned long *block
)
843 transaction_t
*transaction
;
846 read_lock(&journal
->j_state_lock
);
847 spin_lock(&journal
->j_list_lock
);
848 transaction
= journal
->j_checkpoint_transactions
;
850 *tid
= transaction
->t_tid
;
851 *block
= transaction
->t_log_start
;
852 } else if ((transaction
= journal
->j_committing_transaction
) != NULL
) {
853 *tid
= transaction
->t_tid
;
854 *block
= transaction
->t_log_start
;
855 } else if ((transaction
= journal
->j_running_transaction
) != NULL
) {
856 *tid
= transaction
->t_tid
;
857 *block
= journal
->j_head
;
859 *tid
= journal
->j_transaction_sequence
;
860 *block
= journal
->j_head
;
862 ret
= tid_gt(*tid
, journal
->j_tail_sequence
);
863 spin_unlock(&journal
->j_list_lock
);
864 read_unlock(&journal
->j_state_lock
);
870 * Update information in journal structure and in on disk journal superblock
871 * about log tail. This function does not check whether information passed in
872 * really pushes log tail further. It's responsibility of the caller to make
873 * sure provided log tail information is valid (e.g. by holding
874 * j_checkpoint_mutex all the time between computing log tail and calling this
875 * function as is the case with jbd2_cleanup_journal_tail()).
877 * Requires j_checkpoint_mutex
879 int __jbd2_update_log_tail(journal_t
*journal
, tid_t tid
, unsigned long block
)
884 BUG_ON(!mutex_is_locked(&journal
->j_checkpoint_mutex
));
887 * We cannot afford for write to remain in drive's caches since as
888 * soon as we update j_tail, next transaction can start reusing journal
889 * space and if we lose sb update during power failure we'd replay
890 * old transaction with possibly newly overwritten data.
892 ret
= jbd2_journal_update_sb_log_tail(journal
, tid
, block
, WRITE_FUA
);
896 write_lock(&journal
->j_state_lock
);
897 freed
= block
- journal
->j_tail
;
898 if (block
< journal
->j_tail
)
899 freed
+= journal
->j_last
- journal
->j_first
;
901 trace_jbd2_update_log_tail(journal
, tid
, block
, freed
);
903 "Cleaning journal tail from %d to %d (offset %lu), "
905 journal
->j_tail_sequence
, tid
, block
, freed
);
907 journal
->j_free
+= freed
;
908 journal
->j_tail_sequence
= tid
;
909 journal
->j_tail
= block
;
910 write_unlock(&journal
->j_state_lock
);
917 * This is a variation of __jbd2_update_log_tail which checks for validity of
918 * provided log tail and locks j_checkpoint_mutex. So it is safe against races
919 * with other threads updating log tail.
921 void jbd2_update_log_tail(journal_t
*journal
, tid_t tid
, unsigned long block
)
923 mutex_lock(&journal
->j_checkpoint_mutex
);
924 if (tid_gt(tid
, journal
->j_tail_sequence
))
925 __jbd2_update_log_tail(journal
, tid
, block
);
926 mutex_unlock(&journal
->j_checkpoint_mutex
);
929 struct jbd2_stats_proc_session
{
931 struct transaction_stats_s
*stats
;
936 static void *jbd2_seq_info_start(struct seq_file
*seq
, loff_t
*pos
)
938 return *pos
? NULL
: SEQ_START_TOKEN
;
941 static void *jbd2_seq_info_next(struct seq_file
*seq
, void *v
, loff_t
*pos
)
946 static int jbd2_seq_info_show(struct seq_file
*seq
, void *v
)
948 struct jbd2_stats_proc_session
*s
= seq
->private;
950 if (v
!= SEQ_START_TOKEN
)
952 seq_printf(seq
, "%lu transactions (%lu requested), "
953 "each up to %u blocks\n",
954 s
->stats
->ts_tid
, s
->stats
->ts_requested
,
955 s
->journal
->j_max_transaction_buffers
);
956 if (s
->stats
->ts_tid
== 0)
958 seq_printf(seq
, "average: \n %ums waiting for transaction\n",
959 jiffies_to_msecs(s
->stats
->run
.rs_wait
/ s
->stats
->ts_tid
));
960 seq_printf(seq
, " %ums request delay\n",
961 (s
->stats
->ts_requested
== 0) ? 0 :
962 jiffies_to_msecs(s
->stats
->run
.rs_request_delay
/
963 s
->stats
->ts_requested
));
964 seq_printf(seq
, " %ums running transaction\n",
965 jiffies_to_msecs(s
->stats
->run
.rs_running
/ s
->stats
->ts_tid
));
966 seq_printf(seq
, " %ums transaction was being locked\n",
967 jiffies_to_msecs(s
->stats
->run
.rs_locked
/ s
->stats
->ts_tid
));
968 seq_printf(seq
, " %ums flushing data (in ordered mode)\n",
969 jiffies_to_msecs(s
->stats
->run
.rs_flushing
/ s
->stats
->ts_tid
));
970 seq_printf(seq
, " %ums logging transaction\n",
971 jiffies_to_msecs(s
->stats
->run
.rs_logging
/ s
->stats
->ts_tid
));
972 seq_printf(seq
, " %lluus average transaction commit time\n",
973 div_u64(s
->journal
->j_average_commit_time
, 1000));
974 seq_printf(seq
, " %lu handles per transaction\n",
975 s
->stats
->run
.rs_handle_count
/ s
->stats
->ts_tid
);
976 seq_printf(seq
, " %lu blocks per transaction\n",
977 s
->stats
->run
.rs_blocks
/ s
->stats
->ts_tid
);
978 seq_printf(seq
, " %lu logged blocks per transaction\n",
979 s
->stats
->run
.rs_blocks_logged
/ s
->stats
->ts_tid
);
983 static void jbd2_seq_info_stop(struct seq_file
*seq
, void *v
)
987 static const struct seq_operations jbd2_seq_info_ops
= {
988 .start
= jbd2_seq_info_start
,
989 .next
= jbd2_seq_info_next
,
990 .stop
= jbd2_seq_info_stop
,
991 .show
= jbd2_seq_info_show
,
994 static int jbd2_seq_info_open(struct inode
*inode
, struct file
*file
)
996 journal_t
*journal
= PDE_DATA(inode
);
997 struct jbd2_stats_proc_session
*s
;
1000 s
= kmalloc(sizeof(*s
), GFP_KERNEL
);
1003 size
= sizeof(struct transaction_stats_s
);
1004 s
->stats
= kmalloc(size
, GFP_KERNEL
);
1005 if (s
->stats
== NULL
) {
1009 spin_lock(&journal
->j_history_lock
);
1010 memcpy(s
->stats
, &journal
->j_stats
, size
);
1011 s
->journal
= journal
;
1012 spin_unlock(&journal
->j_history_lock
);
1014 rc
= seq_open(file
, &jbd2_seq_info_ops
);
1016 struct seq_file
*m
= file
->private_data
;
1026 static int jbd2_seq_info_release(struct inode
*inode
, struct file
*file
)
1028 struct seq_file
*seq
= file
->private_data
;
1029 struct jbd2_stats_proc_session
*s
= seq
->private;
1032 return seq_release(inode
, file
);
1035 static const struct file_operations jbd2_seq_info_fops
= {
1036 .owner
= THIS_MODULE
,
1037 .open
= jbd2_seq_info_open
,
1039 .llseek
= seq_lseek
,
1040 .release
= jbd2_seq_info_release
,
1043 static struct proc_dir_entry
*proc_jbd2_stats
;
1045 static void jbd2_stats_proc_init(journal_t
*journal
)
1047 journal
->j_proc_entry
= proc_mkdir(journal
->j_devname
, proc_jbd2_stats
);
1048 if (journal
->j_proc_entry
) {
1049 proc_create_data("info", S_IRUGO
, journal
->j_proc_entry
,
1050 &jbd2_seq_info_fops
, journal
);
1054 static void jbd2_stats_proc_exit(journal_t
*journal
)
1056 remove_proc_entry("info", journal
->j_proc_entry
);
1057 remove_proc_entry(journal
->j_devname
, proc_jbd2_stats
);
1061 * Management for journal control blocks: functions to create and
1062 * destroy journal_t structures, and to initialise and read existing
1063 * journal blocks from disk. */
1065 /* First: create and setup a journal_t object in memory. We initialise
1066 * very few fields yet: that has to wait until we have created the
1067 * journal structures from from scratch, or loaded them from disk. */
1069 static journal_t
* journal_init_common (void)
1074 journal
= kzalloc(sizeof(*journal
), GFP_KERNEL
);
1078 init_waitqueue_head(&journal
->j_wait_transaction_locked
);
1079 init_waitqueue_head(&journal
->j_wait_done_commit
);
1080 init_waitqueue_head(&journal
->j_wait_commit
);
1081 init_waitqueue_head(&journal
->j_wait_updates
);
1082 init_waitqueue_head(&journal
->j_wait_reserved
);
1083 mutex_init(&journal
->j_barrier
);
1084 mutex_init(&journal
->j_checkpoint_mutex
);
1085 spin_lock_init(&journal
->j_revoke_lock
);
1086 spin_lock_init(&journal
->j_list_lock
);
1087 rwlock_init(&journal
->j_state_lock
);
1089 journal
->j_commit_interval
= (HZ
* JBD2_DEFAULT_MAX_COMMIT_AGE
);
1090 journal
->j_min_batch_time
= 0;
1091 journal
->j_max_batch_time
= 15000; /* 15ms */
1092 atomic_set(&journal
->j_reserved_credits
, 0);
1094 /* The journal is marked for error until we succeed with recovery! */
1095 journal
->j_flags
= JBD2_ABORT
;
1097 /* Set up a default-sized revoke table for the new mount. */
1098 err
= jbd2_journal_init_revoke(journal
, JOURNAL_REVOKE_DEFAULT_HASH
);
1104 spin_lock_init(&journal
->j_history_lock
);
1109 /* jbd2_journal_init_dev and jbd2_journal_init_inode:
1111 * Create a journal structure assigned some fixed set of disk blocks to
1112 * the journal. We don't actually touch those disk blocks yet, but we
1113 * need to set up all of the mapping information to tell the journaling
1114 * system where the journal blocks are.
1119 * journal_t * jbd2_journal_init_dev() - creates and initialises a journal structure
1120 * @bdev: Block device on which to create the journal
1121 * @fs_dev: Device which hold journalled filesystem for this journal.
1122 * @start: Block nr Start of journal.
1123 * @len: Length of the journal in blocks.
1124 * @blocksize: blocksize of journalling device
1126 * Returns: a newly created journal_t *
1128 * jbd2_journal_init_dev creates a journal which maps a fixed contiguous
1129 * range of blocks on an arbitrary block device.
1132 journal_t
* jbd2_journal_init_dev(struct block_device
*bdev
,
1133 struct block_device
*fs_dev
,
1134 unsigned long long start
, int len
, int blocksize
)
1136 journal_t
*journal
= journal_init_common();
1137 struct buffer_head
*bh
;
1143 /* journal descriptor can store up to n blocks -bzzz */
1144 journal
->j_blocksize
= blocksize
;
1145 journal
->j_dev
= bdev
;
1146 journal
->j_fs_dev
= fs_dev
;
1147 journal
->j_blk_offset
= start
;
1148 journal
->j_maxlen
= len
;
1149 bdevname(journal
->j_dev
, journal
->j_devname
);
1150 strreplace(journal
->j_devname
, '/', '!');
1151 jbd2_stats_proc_init(journal
);
1152 n
= journal
->j_blocksize
/ sizeof(journal_block_tag_t
);
1153 journal
->j_wbufsize
= n
;
1154 journal
->j_wbuf
= kmalloc(n
* sizeof(struct buffer_head
*), GFP_KERNEL
);
1155 if (!journal
->j_wbuf
) {
1156 printk(KERN_ERR
"%s: Can't allocate bhs for commit thread\n",
1161 bh
= __getblk(journal
->j_dev
, start
, journal
->j_blocksize
);
1164 "%s: Cannot get buffer for journal superblock\n",
1168 journal
->j_sb_buffer
= bh
;
1169 journal
->j_superblock
= (journal_superblock_t
*)bh
->b_data
;
1173 kfree(journal
->j_wbuf
);
1174 jbd2_stats_proc_exit(journal
);
1180 * journal_t * jbd2_journal_init_inode () - creates a journal which maps to a inode.
1181 * @inode: An inode to create the journal in
1183 * jbd2_journal_init_inode creates a journal which maps an on-disk inode as
1184 * the journal. The inode must exist already, must support bmap() and
1185 * must have all data blocks preallocated.
1187 journal_t
* jbd2_journal_init_inode (struct inode
*inode
)
1189 struct buffer_head
*bh
;
1190 journal_t
*journal
= journal_init_common();
1194 unsigned long long blocknr
;
1199 journal
->j_dev
= journal
->j_fs_dev
= inode
->i_sb
->s_bdev
;
1200 journal
->j_inode
= inode
;
1201 bdevname(journal
->j_dev
, journal
->j_devname
);
1202 p
= strreplace(journal
->j_devname
, '/', '!');
1203 sprintf(p
, "-%lu", journal
->j_inode
->i_ino
);
1205 "journal %p: inode %s/%ld, size %Ld, bits %d, blksize %ld\n",
1206 journal
, inode
->i_sb
->s_id
, inode
->i_ino
,
1207 (long long) inode
->i_size
,
1208 inode
->i_sb
->s_blocksize_bits
, inode
->i_sb
->s_blocksize
);
1210 journal
->j_maxlen
= inode
->i_size
>> inode
->i_sb
->s_blocksize_bits
;
1211 journal
->j_blocksize
= inode
->i_sb
->s_blocksize
;
1212 jbd2_stats_proc_init(journal
);
1214 /* journal descriptor can store up to n blocks -bzzz */
1215 n
= journal
->j_blocksize
/ sizeof(journal_block_tag_t
);
1216 journal
->j_wbufsize
= n
;
1217 journal
->j_wbuf
= kmalloc(n
* sizeof(struct buffer_head
*), GFP_KERNEL
);
1218 if (!journal
->j_wbuf
) {
1219 printk(KERN_ERR
"%s: Can't allocate bhs for commit thread\n",
1224 err
= jbd2_journal_bmap(journal
, 0, &blocknr
);
1225 /* If that failed, give up */
1227 printk(KERN_ERR
"%s: Cannot locate journal superblock\n",
1232 bh
= getblk_unmovable(journal
->j_dev
, blocknr
, journal
->j_blocksize
);
1235 "%s: Cannot get buffer for journal superblock\n",
1239 journal
->j_sb_buffer
= bh
;
1240 journal
->j_superblock
= (journal_superblock_t
*)bh
->b_data
;
1244 kfree(journal
->j_wbuf
);
1245 jbd2_stats_proc_exit(journal
);
1251 * If the journal init or create aborts, we need to mark the journal
1252 * superblock as being NULL to prevent the journal destroy from writing
1253 * back a bogus superblock.
1255 static void journal_fail_superblock (journal_t
*journal
)
1257 struct buffer_head
*bh
= journal
->j_sb_buffer
;
1259 journal
->j_sb_buffer
= NULL
;
1263 * Given a journal_t structure, initialise the various fields for
1264 * startup of a new journaling session. We use this both when creating
1265 * a journal, and after recovering an old journal to reset it for
1269 static int journal_reset(journal_t
*journal
)
1271 journal_superblock_t
*sb
= journal
->j_superblock
;
1272 unsigned long long first
, last
;
1274 first
= be32_to_cpu(sb
->s_first
);
1275 last
= be32_to_cpu(sb
->s_maxlen
);
1276 if (first
+ JBD2_MIN_JOURNAL_BLOCKS
> last
+ 1) {
1277 printk(KERN_ERR
"JBD2: Journal too short (blocks %llu-%llu).\n",
1279 journal_fail_superblock(journal
);
1283 journal
->j_first
= first
;
1284 journal
->j_last
= last
;
1286 journal
->j_head
= first
;
1287 journal
->j_tail
= first
;
1288 journal
->j_free
= last
- first
;
1290 journal
->j_tail_sequence
= journal
->j_transaction_sequence
;
1291 journal
->j_commit_sequence
= journal
->j_transaction_sequence
- 1;
1292 journal
->j_commit_request
= journal
->j_commit_sequence
;
1294 journal
->j_max_transaction_buffers
= journal
->j_maxlen
/ 4;
1297 * As a special case, if the on-disk copy is already marked as needing
1298 * no recovery (s_start == 0), then we can safely defer the superblock
1299 * update until the next commit by setting JBD2_FLUSHED. This avoids
1300 * attempting a write to a potential-readonly device.
1302 if (sb
->s_start
== 0) {
1303 jbd_debug(1, "JBD2: Skipping superblock update on recovered sb "
1304 "(start %ld, seq %d, errno %d)\n",
1305 journal
->j_tail
, journal
->j_tail_sequence
,
1307 journal
->j_flags
|= JBD2_FLUSHED
;
1309 /* Lock here to make assertions happy... */
1310 mutex_lock(&journal
->j_checkpoint_mutex
);
1312 * Update log tail information. We use WRITE_FUA since new
1313 * transaction will start reusing journal space and so we
1314 * must make sure information about current log tail is on
1317 jbd2_journal_update_sb_log_tail(journal
,
1318 journal
->j_tail_sequence
,
1321 mutex_unlock(&journal
->j_checkpoint_mutex
);
1323 return jbd2_journal_start_thread(journal
);
1326 static int jbd2_write_superblock(journal_t
*journal
, int write_op
)
1328 struct buffer_head
*bh
= journal
->j_sb_buffer
;
1329 journal_superblock_t
*sb
= journal
->j_superblock
;
1332 trace_jbd2_write_superblock(journal
, write_op
);
1333 if (!(journal
->j_flags
& JBD2_BARRIER
))
1334 write_op
&= ~(REQ_FUA
| REQ_FLUSH
);
1336 if (buffer_write_io_error(bh
)) {
1338 * Oh, dear. A previous attempt to write the journal
1339 * superblock failed. This could happen because the
1340 * USB device was yanked out. Or it could happen to
1341 * be a transient write error and maybe the block will
1342 * be remapped. Nothing we can do but to retry the
1343 * write and hope for the best.
1345 printk(KERN_ERR
"JBD2: previous I/O error detected "
1346 "for journal superblock update for %s.\n",
1347 journal
->j_devname
);
1348 clear_buffer_write_io_error(bh
);
1349 set_buffer_uptodate(bh
);
1351 jbd2_superblock_csum_set(journal
, sb
);
1353 bh
->b_end_io
= end_buffer_write_sync
;
1354 ret
= submit_bh(write_op
, bh
);
1356 if (buffer_write_io_error(bh
)) {
1357 clear_buffer_write_io_error(bh
);
1358 set_buffer_uptodate(bh
);
1362 printk(KERN_ERR
"JBD2: Error %d detected when updating "
1363 "journal superblock for %s.\n", ret
,
1364 journal
->j_devname
);
1365 jbd2_journal_abort(journal
, ret
);
1372 * jbd2_journal_update_sb_log_tail() - Update log tail in journal sb on disk.
1373 * @journal: The journal to update.
1374 * @tail_tid: TID of the new transaction at the tail of the log
1375 * @tail_block: The first block of the transaction at the tail of the log
1376 * @write_op: With which operation should we write the journal sb
1378 * Update a journal's superblock information about log tail and write it to
1379 * disk, waiting for the IO to complete.
1381 int jbd2_journal_update_sb_log_tail(journal_t
*journal
, tid_t tail_tid
,
1382 unsigned long tail_block
, int write_op
)
1384 journal_superblock_t
*sb
= journal
->j_superblock
;
1387 if (is_journal_aborted(journal
))
1390 BUG_ON(!mutex_is_locked(&journal
->j_checkpoint_mutex
));
1391 jbd_debug(1, "JBD2: updating superblock (start %lu, seq %u)\n",
1392 tail_block
, tail_tid
);
1394 sb
->s_sequence
= cpu_to_be32(tail_tid
);
1395 sb
->s_start
= cpu_to_be32(tail_block
);
1397 ret
= 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
);
1412 * jbd2_mark_journal_empty() - Mark on disk journal as empty.
1413 * @journal: The journal to update.
1414 * @write_op: With which operation should we write the journal sb
1416 * Update a journal's dynamic superblock fields to show that journal is empty.
1417 * Write updated superblock to disk waiting for IO to complete.
1419 static void jbd2_mark_journal_empty(journal_t
*journal
, int write_op
)
1421 journal_superblock_t
*sb
= journal
->j_superblock
;
1423 BUG_ON(!mutex_is_locked(&journal
->j_checkpoint_mutex
));
1424 read_lock(&journal
->j_state_lock
);
1425 /* Is it already empty? */
1426 if (sb
->s_start
== 0) {
1427 read_unlock(&journal
->j_state_lock
);
1430 jbd_debug(1, "JBD2: Marking journal as empty (seq %d)\n",
1431 journal
->j_tail_sequence
);
1433 sb
->s_sequence
= cpu_to_be32(journal
->j_tail_sequence
);
1434 sb
->s_start
= cpu_to_be32(0);
1435 read_unlock(&journal
->j_state_lock
);
1437 jbd2_write_superblock(journal
, write_op
);
1439 /* Log is no longer empty */
1440 write_lock(&journal
->j_state_lock
);
1441 journal
->j_flags
|= JBD2_FLUSHED
;
1442 write_unlock(&journal
->j_state_lock
);
1447 * jbd2_journal_update_sb_errno() - Update error in the journal.
1448 * @journal: The journal to update.
1450 * Update a journal's errno. Write updated superblock to disk waiting for IO
1453 void jbd2_journal_update_sb_errno(journal_t
*journal
)
1455 journal_superblock_t
*sb
= journal
->j_superblock
;
1457 read_lock(&journal
->j_state_lock
);
1458 jbd_debug(1, "JBD2: updating superblock error (errno %d)\n",
1460 sb
->s_errno
= cpu_to_be32(journal
->j_errno
);
1461 read_unlock(&journal
->j_state_lock
);
1463 jbd2_write_superblock(journal
, WRITE_FUA
);
1465 EXPORT_SYMBOL(jbd2_journal_update_sb_errno
);
1468 * Read the superblock for a given journal, performing initial
1469 * validation of the format.
1471 static int journal_get_superblock(journal_t
*journal
)
1473 struct buffer_head
*bh
;
1474 journal_superblock_t
*sb
;
1477 bh
= journal
->j_sb_buffer
;
1479 J_ASSERT(bh
!= NULL
);
1480 if (!buffer_uptodate(bh
)) {
1481 ll_rw_block(READ
, 1, &bh
);
1483 if (!buffer_uptodate(bh
)) {
1485 "JBD2: IO error reading journal superblock\n");
1490 if (buffer_verified(bh
))
1493 sb
= journal
->j_superblock
;
1497 if (sb
->s_header
.h_magic
!= cpu_to_be32(JBD2_MAGIC_NUMBER
) ||
1498 sb
->s_blocksize
!= cpu_to_be32(journal
->j_blocksize
)) {
1499 printk(KERN_WARNING
"JBD2: no valid journal superblock found\n");
1503 switch(be32_to_cpu(sb
->s_header
.h_blocktype
)) {
1504 case JBD2_SUPERBLOCK_V1
:
1505 journal
->j_format_version
= 1;
1507 case JBD2_SUPERBLOCK_V2
:
1508 journal
->j_format_version
= 2;
1511 printk(KERN_WARNING
"JBD2: unrecognised superblock format ID\n");
1515 if (be32_to_cpu(sb
->s_maxlen
) < journal
->j_maxlen
)
1516 journal
->j_maxlen
= be32_to_cpu(sb
->s_maxlen
);
1517 else if (be32_to_cpu(sb
->s_maxlen
) > journal
->j_maxlen
) {
1518 printk(KERN_WARNING
"JBD2: journal file too short\n");
1522 if (be32_to_cpu(sb
->s_first
) == 0 ||
1523 be32_to_cpu(sb
->s_first
) >= journal
->j_maxlen
) {
1525 "JBD2: Invalid start block of journal: %u\n",
1526 be32_to_cpu(sb
->s_first
));
1530 if (jbd2_has_feature_csum2(journal
) &&
1531 jbd2_has_feature_csum3(journal
)) {
1532 /* Can't have checksum v2 and v3 at the same time! */
1533 printk(KERN_ERR
"JBD2: Can't enable checksumming v2 and v3 "
1534 "at the same time!\n");
1538 if (jbd2_journal_has_csum_v2or3_feature(journal
) &&
1539 jbd2_has_feature_checksum(journal
)) {
1540 /* Can't have checksum v1 and v2 on at the same time! */
1541 printk(KERN_ERR
"JBD2: Can't enable checksumming v1 and v2/3 "
1542 "at the same time!\n");
1546 if (!jbd2_verify_csum_type(journal
, sb
)) {
1547 printk(KERN_ERR
"JBD2: Unknown checksum type\n");
1551 /* Load the checksum driver */
1552 if (jbd2_journal_has_csum_v2or3_feature(journal
)) {
1553 journal
->j_chksum_driver
= crypto_alloc_shash("crc32c", 0, 0);
1554 if (IS_ERR(journal
->j_chksum_driver
)) {
1555 printk(KERN_ERR
"JBD2: Cannot load crc32c driver.\n");
1556 err
= PTR_ERR(journal
->j_chksum_driver
);
1557 journal
->j_chksum_driver
= NULL
;
1562 /* Check superblock checksum */
1563 if (!jbd2_superblock_csum_verify(journal
, sb
)) {
1564 printk(KERN_ERR
"JBD2: journal checksum error\n");
1569 /* Precompute checksum seed for all metadata */
1570 if (jbd2_journal_has_csum_v2or3(journal
))
1571 journal
->j_csum_seed
= jbd2_chksum(journal
, ~0, sb
->s_uuid
,
1572 sizeof(sb
->s_uuid
));
1574 set_buffer_verified(bh
);
1579 journal_fail_superblock(journal
);
1584 * Load the on-disk journal superblock and read the key fields into the
1588 static int load_superblock(journal_t
*journal
)
1591 journal_superblock_t
*sb
;
1593 err
= journal_get_superblock(journal
);
1597 sb
= journal
->j_superblock
;
1599 journal
->j_tail_sequence
= be32_to_cpu(sb
->s_sequence
);
1600 journal
->j_tail
= be32_to_cpu(sb
->s_start
);
1601 journal
->j_first
= be32_to_cpu(sb
->s_first
);
1602 journal
->j_last
= be32_to_cpu(sb
->s_maxlen
);
1603 journal
->j_errno
= be32_to_cpu(sb
->s_errno
);
1610 * int jbd2_journal_load() - Read journal from disk.
1611 * @journal: Journal to act on.
1613 * Given a journal_t structure which tells us which disk blocks contain
1614 * a journal, read the journal from disk to initialise the in-memory
1617 int jbd2_journal_load(journal_t
*journal
)
1620 journal_superblock_t
*sb
;
1622 err
= load_superblock(journal
);
1626 sb
= journal
->j_superblock
;
1627 /* If this is a V2 superblock, then we have to check the
1628 * features flags on it. */
1630 if (journal
->j_format_version
>= 2) {
1631 if ((sb
->s_feature_ro_compat
&
1632 ~cpu_to_be32(JBD2_KNOWN_ROCOMPAT_FEATURES
)) ||
1633 (sb
->s_feature_incompat
&
1634 ~cpu_to_be32(JBD2_KNOWN_INCOMPAT_FEATURES
))) {
1636 "JBD2: Unrecognised features on journal\n");
1642 * Create a slab for this blocksize
1644 err
= jbd2_journal_create_slab(be32_to_cpu(sb
->s_blocksize
));
1648 /* Let the recovery code check whether it needs to recover any
1649 * data from the journal. */
1650 if (jbd2_journal_recover(journal
))
1651 goto recovery_error
;
1653 if (journal
->j_failed_commit
) {
1654 printk(KERN_ERR
"JBD2: journal transaction %u on %s "
1655 "is corrupt.\n", journal
->j_failed_commit
,
1656 journal
->j_devname
);
1657 return -EFSCORRUPTED
;
1660 /* OK, we've finished with the dynamic journal bits:
1661 * reinitialise the dynamic contents of the superblock in memory
1662 * and reset them on disk. */
1663 if (journal_reset(journal
))
1664 goto recovery_error
;
1666 journal
->j_flags
&= ~JBD2_ABORT
;
1667 journal
->j_flags
|= JBD2_LOADED
;
1671 printk(KERN_WARNING
"JBD2: recovery failed\n");
1676 * void jbd2_journal_destroy() - Release a journal_t structure.
1677 * @journal: Journal to act on.
1679 * Release a journal_t structure once it is no longer in use by the
1681 * Return <0 if we couldn't clean up the journal.
1683 int jbd2_journal_destroy(journal_t
*journal
)
1687 /* Wait for the commit thread to wake up and die. */
1688 journal_kill_thread(journal
);
1690 /* Force a final log commit */
1691 if (journal
->j_running_transaction
)
1692 jbd2_journal_commit_transaction(journal
);
1694 /* Force any old transactions to disk */
1696 /* Totally anal locking here... */
1697 spin_lock(&journal
->j_list_lock
);
1698 while (journal
->j_checkpoint_transactions
!= NULL
) {
1699 spin_unlock(&journal
->j_list_lock
);
1700 mutex_lock(&journal
->j_checkpoint_mutex
);
1701 err
= jbd2_log_do_checkpoint(journal
);
1702 mutex_unlock(&journal
->j_checkpoint_mutex
);
1704 * If checkpointing failed, just free the buffers to avoid
1708 jbd2_journal_destroy_checkpoint(journal
);
1709 spin_lock(&journal
->j_list_lock
);
1712 spin_lock(&journal
->j_list_lock
);
1715 J_ASSERT(journal
->j_running_transaction
== NULL
);
1716 J_ASSERT(journal
->j_committing_transaction
== NULL
);
1717 J_ASSERT(journal
->j_checkpoint_transactions
== NULL
);
1718 spin_unlock(&journal
->j_list_lock
);
1720 if (journal
->j_sb_buffer
) {
1721 if (!is_journal_aborted(journal
)) {
1722 mutex_lock(&journal
->j_checkpoint_mutex
);
1724 write_lock(&journal
->j_state_lock
);
1725 journal
->j_tail_sequence
=
1726 ++journal
->j_transaction_sequence
;
1727 write_unlock(&journal
->j_state_lock
);
1729 jbd2_mark_journal_empty(journal
, WRITE_FLUSH_FUA
);
1730 mutex_unlock(&journal
->j_checkpoint_mutex
);
1733 brelse(journal
->j_sb_buffer
);
1736 if (journal
->j_proc_entry
)
1737 jbd2_stats_proc_exit(journal
);
1738 iput(journal
->j_inode
);
1739 if (journal
->j_revoke
)
1740 jbd2_journal_destroy_revoke(journal
);
1741 if (journal
->j_chksum_driver
)
1742 crypto_free_shash(journal
->j_chksum_driver
);
1743 kfree(journal
->j_wbuf
);
1751 *int jbd2_journal_check_used_features () - Check if features specified are used.
1752 * @journal: Journal to check.
1753 * @compat: bitmask of compatible features
1754 * @ro: bitmask of features that force read-only mount
1755 * @incompat: bitmask of incompatible features
1757 * Check whether the journal uses all of a given set of
1758 * features. Return true (non-zero) if it does.
1761 int jbd2_journal_check_used_features (journal_t
*journal
, unsigned long compat
,
1762 unsigned long ro
, unsigned long incompat
)
1764 journal_superblock_t
*sb
;
1766 if (!compat
&& !ro
&& !incompat
)
1768 /* Load journal superblock if it is not loaded yet. */
1769 if (journal
->j_format_version
== 0 &&
1770 journal_get_superblock(journal
) != 0)
1772 if (journal
->j_format_version
== 1)
1775 sb
= journal
->j_superblock
;
1777 if (((be32_to_cpu(sb
->s_feature_compat
) & compat
) == compat
) &&
1778 ((be32_to_cpu(sb
->s_feature_ro_compat
) & ro
) == ro
) &&
1779 ((be32_to_cpu(sb
->s_feature_incompat
) & incompat
) == incompat
))
1786 * int jbd2_journal_check_available_features() - Check feature set in journalling layer
1787 * @journal: Journal to check.
1788 * @compat: bitmask of compatible features
1789 * @ro: bitmask of features that force read-only mount
1790 * @incompat: bitmask of incompatible features
1792 * Check whether the journaling code supports the use of
1793 * all of a given set of features on this journal. Return true
1794 * (non-zero) if it can. */
1796 int jbd2_journal_check_available_features (journal_t
*journal
, unsigned long compat
,
1797 unsigned long ro
, unsigned long incompat
)
1799 if (!compat
&& !ro
&& !incompat
)
1802 /* We can support any known requested features iff the
1803 * superblock is in version 2. Otherwise we fail to support any
1804 * extended sb features. */
1806 if (journal
->j_format_version
!= 2)
1809 if ((compat
& JBD2_KNOWN_COMPAT_FEATURES
) == compat
&&
1810 (ro
& JBD2_KNOWN_ROCOMPAT_FEATURES
) == ro
&&
1811 (incompat
& JBD2_KNOWN_INCOMPAT_FEATURES
) == incompat
)
1818 * int jbd2_journal_set_features () - Mark a given journal feature in the superblock
1819 * @journal: Journal to act on.
1820 * @compat: bitmask of compatible features
1821 * @ro: bitmask of features that force read-only mount
1822 * @incompat: bitmask of incompatible features
1824 * Mark a given journal feature as present on the
1825 * superblock. Returns true if the requested features could be set.
1829 int jbd2_journal_set_features (journal_t
*journal
, unsigned long compat
,
1830 unsigned long ro
, unsigned long incompat
)
1832 #define INCOMPAT_FEATURE_ON(f) \
1833 ((incompat & (f)) && !(sb->s_feature_incompat & cpu_to_be32(f)))
1834 #define COMPAT_FEATURE_ON(f) \
1835 ((compat & (f)) && !(sb->s_feature_compat & cpu_to_be32(f)))
1836 journal_superblock_t
*sb
;
1838 if (jbd2_journal_check_used_features(journal
, compat
, ro
, incompat
))
1841 if (!jbd2_journal_check_available_features(journal
, compat
, ro
, incompat
))
1844 /* If enabling v2 checksums, turn on v3 instead */
1845 if (incompat
& JBD2_FEATURE_INCOMPAT_CSUM_V2
) {
1846 incompat
&= ~JBD2_FEATURE_INCOMPAT_CSUM_V2
;
1847 incompat
|= JBD2_FEATURE_INCOMPAT_CSUM_V3
;
1850 /* Asking for checksumming v3 and v1? Only give them v3. */
1851 if (incompat
& JBD2_FEATURE_INCOMPAT_CSUM_V3
&&
1852 compat
& JBD2_FEATURE_COMPAT_CHECKSUM
)
1853 compat
&= ~JBD2_FEATURE_COMPAT_CHECKSUM
;
1855 jbd_debug(1, "Setting new features 0x%lx/0x%lx/0x%lx\n",
1856 compat
, ro
, incompat
);
1858 sb
= journal
->j_superblock
;
1860 /* If enabling v3 checksums, update superblock */
1861 if (INCOMPAT_FEATURE_ON(JBD2_FEATURE_INCOMPAT_CSUM_V3
)) {
1862 sb
->s_checksum_type
= JBD2_CRC32C_CHKSUM
;
1863 sb
->s_feature_compat
&=
1864 ~cpu_to_be32(JBD2_FEATURE_COMPAT_CHECKSUM
);
1866 /* Load the checksum driver */
1867 if (journal
->j_chksum_driver
== NULL
) {
1868 journal
->j_chksum_driver
= crypto_alloc_shash("crc32c",
1870 if (IS_ERR(journal
->j_chksum_driver
)) {
1871 printk(KERN_ERR
"JBD2: Cannot load crc32c "
1873 journal
->j_chksum_driver
= NULL
;
1877 /* Precompute checksum seed for all metadata */
1878 journal
->j_csum_seed
= jbd2_chksum(journal
, ~0,
1880 sizeof(sb
->s_uuid
));
1884 /* If enabling v1 checksums, downgrade superblock */
1885 if (COMPAT_FEATURE_ON(JBD2_FEATURE_COMPAT_CHECKSUM
))
1886 sb
->s_feature_incompat
&=
1887 ~cpu_to_be32(JBD2_FEATURE_INCOMPAT_CSUM_V2
|
1888 JBD2_FEATURE_INCOMPAT_CSUM_V3
);
1890 sb
->s_feature_compat
|= cpu_to_be32(compat
);
1891 sb
->s_feature_ro_compat
|= cpu_to_be32(ro
);
1892 sb
->s_feature_incompat
|= cpu_to_be32(incompat
);
1895 #undef COMPAT_FEATURE_ON
1896 #undef INCOMPAT_FEATURE_ON
1900 * jbd2_journal_clear_features () - Clear a given journal feature in the
1902 * @journal: Journal to act on.
1903 * @compat: bitmask of compatible features
1904 * @ro: bitmask of features that force read-only mount
1905 * @incompat: bitmask of incompatible features
1907 * Clear a given journal feature as present on the
1910 void jbd2_journal_clear_features(journal_t
*journal
, unsigned long compat
,
1911 unsigned long ro
, unsigned long incompat
)
1913 journal_superblock_t
*sb
;
1915 jbd_debug(1, "Clear features 0x%lx/0x%lx/0x%lx\n",
1916 compat
, ro
, incompat
);
1918 sb
= journal
->j_superblock
;
1920 sb
->s_feature_compat
&= ~cpu_to_be32(compat
);
1921 sb
->s_feature_ro_compat
&= ~cpu_to_be32(ro
);
1922 sb
->s_feature_incompat
&= ~cpu_to_be32(incompat
);
1924 EXPORT_SYMBOL(jbd2_journal_clear_features
);
1927 * int jbd2_journal_flush () - Flush journal
1928 * @journal: Journal to act on.
1930 * Flush all data for a given journal to disk and empty the journal.
1931 * Filesystems can use this when remounting readonly to ensure that
1932 * recovery does not need to happen on remount.
1935 int jbd2_journal_flush(journal_t
*journal
)
1938 transaction_t
*transaction
= NULL
;
1940 write_lock(&journal
->j_state_lock
);
1942 /* Force everything buffered to the log... */
1943 if (journal
->j_running_transaction
) {
1944 transaction
= journal
->j_running_transaction
;
1945 __jbd2_log_start_commit(journal
, transaction
->t_tid
);
1946 } else if (journal
->j_committing_transaction
)
1947 transaction
= journal
->j_committing_transaction
;
1949 /* Wait for the log commit to complete... */
1951 tid_t tid
= transaction
->t_tid
;
1953 write_unlock(&journal
->j_state_lock
);
1954 jbd2_log_wait_commit(journal
, tid
);
1956 write_unlock(&journal
->j_state_lock
);
1959 /* ...and flush everything in the log out to disk. */
1960 spin_lock(&journal
->j_list_lock
);
1961 while (!err
&& journal
->j_checkpoint_transactions
!= NULL
) {
1962 spin_unlock(&journal
->j_list_lock
);
1963 mutex_lock(&journal
->j_checkpoint_mutex
);
1964 err
= jbd2_log_do_checkpoint(journal
);
1965 mutex_unlock(&journal
->j_checkpoint_mutex
);
1966 spin_lock(&journal
->j_list_lock
);
1968 spin_unlock(&journal
->j_list_lock
);
1970 if (is_journal_aborted(journal
))
1973 mutex_lock(&journal
->j_checkpoint_mutex
);
1975 err
= jbd2_cleanup_journal_tail(journal
);
1977 mutex_unlock(&journal
->j_checkpoint_mutex
);
1983 /* Finally, mark the journal as really needing no recovery.
1984 * This sets s_start==0 in the underlying superblock, which is
1985 * the magic code for a fully-recovered superblock. Any future
1986 * commits of data to the journal will restore the current
1988 jbd2_mark_journal_empty(journal
, WRITE_FUA
);
1989 mutex_unlock(&journal
->j_checkpoint_mutex
);
1990 write_lock(&journal
->j_state_lock
);
1991 J_ASSERT(!journal
->j_running_transaction
);
1992 J_ASSERT(!journal
->j_committing_transaction
);
1993 J_ASSERT(!journal
->j_checkpoint_transactions
);
1994 J_ASSERT(journal
->j_head
== journal
->j_tail
);
1995 J_ASSERT(journal
->j_tail_sequence
== journal
->j_transaction_sequence
);
1996 write_unlock(&journal
->j_state_lock
);
2002 * int jbd2_journal_wipe() - Wipe journal contents
2003 * @journal: Journal to act on.
2004 * @write: flag (see below)
2006 * Wipe out all of the contents of a journal, safely. This will produce
2007 * a warning if the journal contains any valid recovery information.
2008 * Must be called between journal_init_*() and jbd2_journal_load().
2010 * If 'write' is non-zero, then we wipe out the journal on disk; otherwise
2011 * we merely suppress recovery.
2014 int jbd2_journal_wipe(journal_t
*journal
, int write
)
2018 J_ASSERT (!(journal
->j_flags
& JBD2_LOADED
));
2020 err
= load_superblock(journal
);
2024 if (!journal
->j_tail
)
2027 printk(KERN_WARNING
"JBD2: %s recovery information on journal\n",
2028 write
? "Clearing" : "Ignoring");
2030 err
= jbd2_journal_skip_recovery(journal
);
2032 /* Lock to make assertions happy... */
2033 mutex_lock(&journal
->j_checkpoint_mutex
);
2034 jbd2_mark_journal_empty(journal
, WRITE_FUA
);
2035 mutex_unlock(&journal
->j_checkpoint_mutex
);
2043 * Journal abort has very specific semantics, which we describe
2044 * for journal abort.
2046 * Two internal functions, which provide abort to the jbd layer
2051 * Quick version for internal journal use (doesn't lock the journal).
2052 * Aborts hard --- we mark the abort as occurred, but do _nothing_ else,
2053 * and don't attempt to make any other journal updates.
2055 void __jbd2_journal_abort_hard(journal_t
*journal
)
2057 transaction_t
*transaction
;
2059 if (journal
->j_flags
& JBD2_ABORT
)
2062 printk(KERN_ERR
"Aborting journal on device %s.\n",
2063 journal
->j_devname
);
2065 write_lock(&journal
->j_state_lock
);
2066 journal
->j_flags
|= JBD2_ABORT
;
2067 transaction
= journal
->j_running_transaction
;
2069 __jbd2_log_start_commit(journal
, transaction
->t_tid
);
2070 write_unlock(&journal
->j_state_lock
);
2073 /* Soft abort: record the abort error status in the journal superblock,
2074 * but don't do any other IO. */
2075 static void __journal_abort_soft (journal_t
*journal
, int errno
)
2077 if (journal
->j_flags
& JBD2_ABORT
)
2080 if (!journal
->j_errno
)
2081 journal
->j_errno
= errno
;
2083 __jbd2_journal_abort_hard(journal
);
2086 jbd2_journal_update_sb_errno(journal
);
2087 write_lock(&journal
->j_state_lock
);
2088 journal
->j_flags
|= JBD2_REC_ERR
;
2089 write_unlock(&journal
->j_state_lock
);
2094 * void jbd2_journal_abort () - Shutdown the journal immediately.
2095 * @journal: the journal to shutdown.
2096 * @errno: an error number to record in the journal indicating
2097 * the reason for the shutdown.
2099 * Perform a complete, immediate shutdown of the ENTIRE
2100 * journal (not of a single transaction). This operation cannot be
2101 * undone without closing and reopening the journal.
2103 * The jbd2_journal_abort function is intended to support higher level error
2104 * recovery mechanisms such as the ext2/ext3 remount-readonly error
2107 * Journal abort has very specific semantics. Any existing dirty,
2108 * unjournaled buffers in the main filesystem will still be written to
2109 * disk by bdflush, but the journaling mechanism will be suspended
2110 * immediately and no further transaction commits will be honoured.
2112 * Any dirty, journaled buffers will be written back to disk without
2113 * hitting the journal. Atomicity cannot be guaranteed on an aborted
2114 * filesystem, but we _do_ attempt to leave as much data as possible
2115 * behind for fsck to use for cleanup.
2117 * Any attempt to get a new transaction handle on a journal which is in
2118 * ABORT state will just result in an -EROFS error return. A
2119 * jbd2_journal_stop on an existing handle will return -EIO if we have
2120 * entered abort state during the update.
2122 * Recursive transactions are not disturbed by journal abort until the
2123 * final jbd2_journal_stop, which will receive the -EIO error.
2125 * Finally, the jbd2_journal_abort call allows the caller to supply an errno
2126 * which will be recorded (if possible) in the journal superblock. This
2127 * allows a client to record failure conditions in the middle of a
2128 * transaction without having to complete the transaction to record the
2129 * failure to disk. ext3_error, for example, now uses this
2132 * Errors which originate from within the journaling layer will NOT
2133 * supply an errno; a null errno implies that absolutely no further
2134 * writes are done to the journal (unless there are any already in
2139 void jbd2_journal_abort(journal_t
*journal
, int errno
)
2141 __journal_abort_soft(journal
, errno
);
2145 * int jbd2_journal_errno () - returns the journal's error state.
2146 * @journal: journal to examine.
2148 * This is the errno number set with jbd2_journal_abort(), the last
2149 * time the journal was mounted - if the journal was stopped
2150 * without calling abort this will be 0.
2152 * If the journal has been aborted on this mount time -EROFS will
2155 int jbd2_journal_errno(journal_t
*journal
)
2159 read_lock(&journal
->j_state_lock
);
2160 if (journal
->j_flags
& JBD2_ABORT
)
2163 err
= journal
->j_errno
;
2164 read_unlock(&journal
->j_state_lock
);
2169 * int jbd2_journal_clear_err () - clears the journal's error state
2170 * @journal: journal to act on.
2172 * An error must be cleared or acked to take a FS out of readonly
2175 int jbd2_journal_clear_err(journal_t
*journal
)
2179 write_lock(&journal
->j_state_lock
);
2180 if (journal
->j_flags
& JBD2_ABORT
)
2183 journal
->j_errno
= 0;
2184 write_unlock(&journal
->j_state_lock
);
2189 * void jbd2_journal_ack_err() - Ack journal err.
2190 * @journal: journal to act on.
2192 * An error must be cleared or acked to take a FS out of readonly
2195 void jbd2_journal_ack_err(journal_t
*journal
)
2197 write_lock(&journal
->j_state_lock
);
2198 if (journal
->j_errno
)
2199 journal
->j_flags
|= JBD2_ACK_ERR
;
2200 write_unlock(&journal
->j_state_lock
);
2203 int jbd2_journal_blocks_per_page(struct inode
*inode
)
2205 return 1 << (PAGE_CACHE_SHIFT
- inode
->i_sb
->s_blocksize_bits
);
2209 * helper functions to deal with 32 or 64bit block numbers.
2211 size_t journal_tag_bytes(journal_t
*journal
)
2215 if (jbd2_has_feature_csum3(journal
))
2216 return sizeof(journal_block_tag3_t
);
2218 sz
= sizeof(journal_block_tag_t
);
2220 if (jbd2_has_feature_csum2(journal
))
2221 sz
+= sizeof(__u16
);
2223 if (jbd2_has_feature_64bit(journal
))
2226 return sz
- sizeof(__u32
);
2230 * JBD memory management
2232 * These functions are used to allocate block-sized chunks of memory
2233 * used for making copies of buffer_head data. Very often it will be
2234 * page-sized chunks of data, but sometimes it will be in
2235 * sub-page-size chunks. (For example, 16k pages on Power systems
2236 * with a 4k block file system.) For blocks smaller than a page, we
2237 * use a SLAB allocator. There are slab caches for each block size,
2238 * which are allocated at mount time, if necessary, and we only free
2239 * (all of) the slab caches when/if the jbd2 module is unloaded. For
2240 * this reason we don't need to a mutex to protect access to
2241 * jbd2_slab[] allocating or releasing memory; only in
2242 * jbd2_journal_create_slab().
2244 #define JBD2_MAX_SLABS 8
2245 static struct kmem_cache
*jbd2_slab
[JBD2_MAX_SLABS
];
2247 static const char *jbd2_slab_names
[JBD2_MAX_SLABS
] = {
2248 "jbd2_1k", "jbd2_2k", "jbd2_4k", "jbd2_8k",
2249 "jbd2_16k", "jbd2_32k", "jbd2_64k", "jbd2_128k"
2253 static void jbd2_journal_destroy_slabs(void)
2257 for (i
= 0; i
< JBD2_MAX_SLABS
; i
++) {
2259 kmem_cache_destroy(jbd2_slab
[i
]);
2260 jbd2_slab
[i
] = NULL
;
2264 static int jbd2_journal_create_slab(size_t size
)
2266 static DEFINE_MUTEX(jbd2_slab_create_mutex
);
2267 int i
= order_base_2(size
) - 10;
2270 if (size
== PAGE_SIZE
)
2273 if (i
>= JBD2_MAX_SLABS
)
2276 if (unlikely(i
< 0))
2278 mutex_lock(&jbd2_slab_create_mutex
);
2280 mutex_unlock(&jbd2_slab_create_mutex
);
2281 return 0; /* Already created */
2284 slab_size
= 1 << (i
+10);
2285 jbd2_slab
[i
] = kmem_cache_create(jbd2_slab_names
[i
], slab_size
,
2286 slab_size
, 0, NULL
);
2287 mutex_unlock(&jbd2_slab_create_mutex
);
2288 if (!jbd2_slab
[i
]) {
2289 printk(KERN_EMERG
"JBD2: no memory for jbd2_slab cache\n");
2295 static struct kmem_cache
*get_slab(size_t size
)
2297 int i
= order_base_2(size
) - 10;
2299 BUG_ON(i
>= JBD2_MAX_SLABS
);
2300 if (unlikely(i
< 0))
2302 BUG_ON(jbd2_slab
[i
] == NULL
);
2303 return jbd2_slab
[i
];
2306 void *jbd2_alloc(size_t size
, gfp_t flags
)
2310 BUG_ON(size
& (size
-1)); /* Must be a power of 2 */
2312 flags
|= __GFP_REPEAT
;
2313 if (size
== PAGE_SIZE
)
2314 ptr
= (void *)__get_free_pages(flags
, 0);
2315 else if (size
> PAGE_SIZE
) {
2316 int order
= get_order(size
);
2319 ptr
= (void *)__get_free_pages(flags
, order
);
2321 ptr
= vmalloc(size
);
2323 ptr
= kmem_cache_alloc(get_slab(size
), flags
);
2325 /* Check alignment; SLUB has gotten this wrong in the past,
2326 * and this can lead to user data corruption! */
2327 BUG_ON(((unsigned long) ptr
) & (size
-1));
2332 void jbd2_free(void *ptr
, size_t size
)
2334 if (size
== PAGE_SIZE
) {
2335 free_pages((unsigned long)ptr
, 0);
2338 if (size
> PAGE_SIZE
) {
2339 int order
= get_order(size
);
2342 free_pages((unsigned long)ptr
, order
);
2347 kmem_cache_free(get_slab(size
), ptr
);
2351 * Journal_head storage management
2353 static struct kmem_cache
*jbd2_journal_head_cache
;
2354 #ifdef CONFIG_JBD2_DEBUG
2355 static atomic_t nr_journal_heads
= ATOMIC_INIT(0);
2358 static int jbd2_journal_init_journal_head_cache(void)
2362 J_ASSERT(jbd2_journal_head_cache
== NULL
);
2363 jbd2_journal_head_cache
= kmem_cache_create("jbd2_journal_head",
2364 sizeof(struct journal_head
),
2366 SLAB_TEMPORARY
| SLAB_DESTROY_BY_RCU
,
2369 if (!jbd2_journal_head_cache
) {
2371 printk(KERN_EMERG
"JBD2: no memory for journal_head cache\n");
2376 static void jbd2_journal_destroy_journal_head_cache(void)
2378 if (jbd2_journal_head_cache
) {
2379 kmem_cache_destroy(jbd2_journal_head_cache
);
2380 jbd2_journal_head_cache
= NULL
;
2385 * journal_head splicing and dicing
2387 static struct journal_head
*journal_alloc_journal_head(void)
2389 struct journal_head
*ret
;
2391 #ifdef CONFIG_JBD2_DEBUG
2392 atomic_inc(&nr_journal_heads
);
2394 ret
= kmem_cache_zalloc(jbd2_journal_head_cache
, GFP_NOFS
);
2396 jbd_debug(1, "out of memory for journal_head\n");
2397 pr_notice_ratelimited("ENOMEM in %s, retrying.\n", __func__
);
2398 ret
= kmem_cache_zalloc(jbd2_journal_head_cache
,
2399 GFP_NOFS
| __GFP_NOFAIL
);
2404 static void journal_free_journal_head(struct journal_head
*jh
)
2406 #ifdef CONFIG_JBD2_DEBUG
2407 atomic_dec(&nr_journal_heads
);
2408 memset(jh
, JBD2_POISON_FREE
, sizeof(*jh
));
2410 kmem_cache_free(jbd2_journal_head_cache
, jh
);
2414 * A journal_head is attached to a buffer_head whenever JBD has an
2415 * interest in the buffer.
2417 * Whenever a buffer has an attached journal_head, its ->b_state:BH_JBD bit
2418 * is set. This bit is tested in core kernel code where we need to take
2419 * JBD-specific actions. Testing the zeroness of ->b_private is not reliable
2422 * When a buffer has its BH_JBD bit set, its ->b_count is elevated by one.
2424 * When a buffer has its BH_JBD bit set it is immune from being released by
2425 * core kernel code, mainly via ->b_count.
2427 * A journal_head is detached from its buffer_head when the journal_head's
2428 * b_jcount reaches zero. Running transaction (b_transaction) and checkpoint
2429 * transaction (b_cp_transaction) hold their references to b_jcount.
2431 * Various places in the kernel want to attach a journal_head to a buffer_head
2432 * _before_ attaching the journal_head to a transaction. To protect the
2433 * journal_head in this situation, jbd2_journal_add_journal_head elevates the
2434 * journal_head's b_jcount refcount by one. The caller must call
2435 * jbd2_journal_put_journal_head() to undo this.
2437 * So the typical usage would be:
2439 * (Attach a journal_head if needed. Increments b_jcount)
2440 * struct journal_head *jh = jbd2_journal_add_journal_head(bh);
2442 * (Get another reference for transaction)
2443 * jbd2_journal_grab_journal_head(bh);
2444 * jh->b_transaction = xxx;
2445 * (Put original reference)
2446 * jbd2_journal_put_journal_head(jh);
2450 * Give a buffer_head a journal_head.
2454 struct journal_head
*jbd2_journal_add_journal_head(struct buffer_head
*bh
)
2456 struct journal_head
*jh
;
2457 struct journal_head
*new_jh
= NULL
;
2460 if (!buffer_jbd(bh
))
2461 new_jh
= journal_alloc_journal_head();
2463 jbd_lock_bh_journal_head(bh
);
2464 if (buffer_jbd(bh
)) {
2468 (atomic_read(&bh
->b_count
) > 0) ||
2469 (bh
->b_page
&& bh
->b_page
->mapping
));
2472 jbd_unlock_bh_journal_head(bh
);
2477 new_jh
= NULL
; /* We consumed it */
2482 BUFFER_TRACE(bh
, "added journal_head");
2485 jbd_unlock_bh_journal_head(bh
);
2487 journal_free_journal_head(new_jh
);
2488 return bh
->b_private
;
2492 * Grab a ref against this buffer_head's journal_head. If it ended up not
2493 * having a journal_head, return NULL
2495 struct journal_head
*jbd2_journal_grab_journal_head(struct buffer_head
*bh
)
2497 struct journal_head
*jh
= NULL
;
2499 jbd_lock_bh_journal_head(bh
);
2500 if (buffer_jbd(bh
)) {
2504 jbd_unlock_bh_journal_head(bh
);
2508 static void __journal_remove_journal_head(struct buffer_head
*bh
)
2510 struct journal_head
*jh
= bh2jh(bh
);
2512 J_ASSERT_JH(jh
, jh
->b_jcount
>= 0);
2513 J_ASSERT_JH(jh
, jh
->b_transaction
== NULL
);
2514 J_ASSERT_JH(jh
, jh
->b_next_transaction
== NULL
);
2515 J_ASSERT_JH(jh
, jh
->b_cp_transaction
== NULL
);
2516 J_ASSERT_JH(jh
, jh
->b_jlist
== BJ_None
);
2517 J_ASSERT_BH(bh
, buffer_jbd(bh
));
2518 J_ASSERT_BH(bh
, jh2bh(jh
) == bh
);
2519 BUFFER_TRACE(bh
, "remove journal_head");
2520 if (jh
->b_frozen_data
) {
2521 printk(KERN_WARNING
"%s: freeing b_frozen_data\n", __func__
);
2522 jbd2_free(jh
->b_frozen_data
, bh
->b_size
);
2524 if (jh
->b_committed_data
) {
2525 printk(KERN_WARNING
"%s: freeing b_committed_data\n", __func__
);
2526 jbd2_free(jh
->b_committed_data
, bh
->b_size
);
2528 bh
->b_private
= NULL
;
2529 jh
->b_bh
= NULL
; /* debug, really */
2530 clear_buffer_jbd(bh
);
2531 journal_free_journal_head(jh
);
2535 * Drop a reference on the passed journal_head. If it fell to zero then
2536 * release the journal_head from the buffer_head.
2538 void jbd2_journal_put_journal_head(struct journal_head
*jh
)
2540 struct buffer_head
*bh
= jh2bh(jh
);
2542 jbd_lock_bh_journal_head(bh
);
2543 J_ASSERT_JH(jh
, jh
->b_jcount
> 0);
2545 if (!jh
->b_jcount
) {
2546 __journal_remove_journal_head(bh
);
2547 jbd_unlock_bh_journal_head(bh
);
2550 jbd_unlock_bh_journal_head(bh
);
2554 * Initialize jbd inode head
2556 void jbd2_journal_init_jbd_inode(struct jbd2_inode
*jinode
, struct inode
*inode
)
2558 jinode
->i_transaction
= NULL
;
2559 jinode
->i_next_transaction
= NULL
;
2560 jinode
->i_vfs_inode
= inode
;
2561 jinode
->i_flags
= 0;
2562 INIT_LIST_HEAD(&jinode
->i_list
);
2566 * Function to be called before we start removing inode from memory (i.e.,
2567 * clear_inode() is a fine place to be called from). It removes inode from
2568 * transaction's lists.
2570 void jbd2_journal_release_jbd_inode(journal_t
*journal
,
2571 struct jbd2_inode
*jinode
)
2576 spin_lock(&journal
->j_list_lock
);
2577 /* Is commit writing out inode - we have to wait */
2578 if (test_bit(__JI_COMMIT_RUNNING
, &jinode
->i_flags
)) {
2579 wait_queue_head_t
*wq
;
2580 DEFINE_WAIT_BIT(wait
, &jinode
->i_flags
, __JI_COMMIT_RUNNING
);
2581 wq
= bit_waitqueue(&jinode
->i_flags
, __JI_COMMIT_RUNNING
);
2582 prepare_to_wait(wq
, &wait
.wait
, TASK_UNINTERRUPTIBLE
);
2583 spin_unlock(&journal
->j_list_lock
);
2585 finish_wait(wq
, &wait
.wait
);
2589 if (jinode
->i_transaction
) {
2590 list_del(&jinode
->i_list
);
2591 jinode
->i_transaction
= NULL
;
2593 spin_unlock(&journal
->j_list_lock
);
2597 #ifdef CONFIG_PROC_FS
2599 #define JBD2_STATS_PROC_NAME "fs/jbd2"
2601 static void __init
jbd2_create_jbd_stats_proc_entry(void)
2603 proc_jbd2_stats
= proc_mkdir(JBD2_STATS_PROC_NAME
, NULL
);
2606 static void __exit
jbd2_remove_jbd_stats_proc_entry(void)
2608 if (proc_jbd2_stats
)
2609 remove_proc_entry(JBD2_STATS_PROC_NAME
, NULL
);
2614 #define jbd2_create_jbd_stats_proc_entry() do {} while (0)
2615 #define jbd2_remove_jbd_stats_proc_entry() do {} while (0)
2619 struct kmem_cache
*jbd2_handle_cache
, *jbd2_inode_cache
;
2621 static int __init
jbd2_journal_init_handle_cache(void)
2623 jbd2_handle_cache
= KMEM_CACHE(jbd2_journal_handle
, SLAB_TEMPORARY
);
2624 if (jbd2_handle_cache
== NULL
) {
2625 printk(KERN_EMERG
"JBD2: failed to create handle cache\n");
2628 jbd2_inode_cache
= KMEM_CACHE(jbd2_inode
, 0);
2629 if (jbd2_inode_cache
== NULL
) {
2630 printk(KERN_EMERG
"JBD2: failed to create inode cache\n");
2631 kmem_cache_destroy(jbd2_handle_cache
);
2637 static void jbd2_journal_destroy_handle_cache(void)
2639 if (jbd2_handle_cache
)
2640 kmem_cache_destroy(jbd2_handle_cache
);
2641 if (jbd2_inode_cache
)
2642 kmem_cache_destroy(jbd2_inode_cache
);
2647 * Module startup and shutdown
2650 static int __init
journal_init_caches(void)
2654 ret
= jbd2_journal_init_revoke_caches();
2656 ret
= jbd2_journal_init_journal_head_cache();
2658 ret
= jbd2_journal_init_handle_cache();
2660 ret
= jbd2_journal_init_transaction_cache();
2664 static void jbd2_journal_destroy_caches(void)
2666 jbd2_journal_destroy_revoke_caches();
2667 jbd2_journal_destroy_journal_head_cache();
2668 jbd2_journal_destroy_handle_cache();
2669 jbd2_journal_destroy_transaction_cache();
2670 jbd2_journal_destroy_slabs();
2673 static int __init
journal_init(void)
2677 BUILD_BUG_ON(sizeof(struct journal_superblock_s
) != 1024);
2679 ret
= journal_init_caches();
2681 jbd2_create_jbd_stats_proc_entry();
2683 jbd2_journal_destroy_caches();
2688 static void __exit
journal_exit(void)
2690 #ifdef CONFIG_JBD2_DEBUG
2691 int n
= atomic_read(&nr_journal_heads
);
2693 printk(KERN_ERR
"JBD2: leaked %d journal_heads!\n", n
);
2695 jbd2_remove_jbd_stats_proc_entry();
2696 jbd2_journal_destroy_caches();
2699 MODULE_LICENSE("GPL");
2700 module_init(journal_init
);
2701 module_exit(journal_exit
);