2 * linux/fs/jbd/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/jbd.h>
29 #include <linux/errno.h>
30 #include <linux/slab.h>
31 #include <linux/init.h>
33 #include <linux/freezer.h>
34 #include <linux/pagemap.h>
35 #include <linux/kthread.h>
36 #include <linux/poison.h>
37 #include <linux/proc_fs.h>
38 #include <linux/debugfs.h>
39 #include <linux/ratelimit.h>
41 #include <asm/uaccess.h>
44 EXPORT_SYMBOL(journal_start
);
45 EXPORT_SYMBOL(journal_restart
);
46 EXPORT_SYMBOL(journal_extend
);
47 EXPORT_SYMBOL(journal_stop
);
48 EXPORT_SYMBOL(journal_lock_updates
);
49 EXPORT_SYMBOL(journal_unlock_updates
);
50 EXPORT_SYMBOL(journal_get_write_access
);
51 EXPORT_SYMBOL(journal_get_create_access
);
52 EXPORT_SYMBOL(journal_get_undo_access
);
53 EXPORT_SYMBOL(journal_dirty_data
);
54 EXPORT_SYMBOL(journal_dirty_metadata
);
55 EXPORT_SYMBOL(journal_release_buffer
);
56 EXPORT_SYMBOL(journal_forget
);
58 EXPORT_SYMBOL(journal_sync_buffer
);
60 EXPORT_SYMBOL(journal_flush
);
61 EXPORT_SYMBOL(journal_revoke
);
63 EXPORT_SYMBOL(journal_init_dev
);
64 EXPORT_SYMBOL(journal_init_inode
);
65 EXPORT_SYMBOL(journal_update_format
);
66 EXPORT_SYMBOL(journal_check_used_features
);
67 EXPORT_SYMBOL(journal_check_available_features
);
68 EXPORT_SYMBOL(journal_set_features
);
69 EXPORT_SYMBOL(journal_create
);
70 EXPORT_SYMBOL(journal_load
);
71 EXPORT_SYMBOL(journal_destroy
);
72 EXPORT_SYMBOL(journal_abort
);
73 EXPORT_SYMBOL(journal_errno
);
74 EXPORT_SYMBOL(journal_ack_err
);
75 EXPORT_SYMBOL(journal_clear_err
);
76 EXPORT_SYMBOL(log_wait_commit
);
77 EXPORT_SYMBOL(log_start_commit
);
78 EXPORT_SYMBOL(journal_start_commit
);
79 EXPORT_SYMBOL(journal_force_commit_nested
);
80 EXPORT_SYMBOL(journal_wipe
);
81 EXPORT_SYMBOL(journal_blocks_per_page
);
82 EXPORT_SYMBOL(journal_invalidatepage
);
83 EXPORT_SYMBOL(journal_try_to_free_buffers
);
84 EXPORT_SYMBOL(journal_force_commit
);
86 static int journal_convert_superblock_v1(journal_t
*, journal_superblock_t
*);
87 static void __journal_abort_soft (journal_t
*journal
, int errno
);
90 * Helper function used to manage commit timeouts
93 static void commit_timeout(unsigned long __data
)
95 struct task_struct
* p
= (struct task_struct
*) __data
;
101 * kjournald: The main thread function used to manage a logging device
104 * This kernel thread is responsible for two things:
106 * 1) COMMIT: Every so often we need to commit the current state of the
107 * filesystem to disk. The journal thread is responsible for writing
108 * all of the metadata buffers to disk.
110 * 2) CHECKPOINT: We cannot reuse a used section of the log file until all
111 * of the data in that part of the log has been rewritten elsewhere on
112 * the disk. Flushing these old buffers to reclaim space in the log is
113 * known as checkpointing, and this thread is responsible for that job.
116 static int kjournald(void *arg
)
118 journal_t
*journal
= arg
;
119 transaction_t
*transaction
;
122 * Set up an interval timer which can be used to trigger a commit wakeup
123 * after the commit interval expires
125 setup_timer(&journal
->j_commit_timer
, commit_timeout
,
126 (unsigned long)current
);
128 /* Record that the journal thread is running */
129 journal
->j_task
= current
;
130 wake_up(&journal
->j_wait_done_commit
);
132 printk(KERN_INFO
"kjournald starting. Commit interval %ld seconds\n",
133 journal
->j_commit_interval
/ HZ
);
136 * And now, wait forever for commit wakeup events.
138 spin_lock(&journal
->j_state_lock
);
141 if (journal
->j_flags
& JFS_UNMOUNT
)
144 jbd_debug(1, "commit_sequence=%d, commit_request=%d\n",
145 journal
->j_commit_sequence
, journal
->j_commit_request
);
147 if (journal
->j_commit_sequence
!= journal
->j_commit_request
) {
148 jbd_debug(1, "OK, requests differ\n");
149 spin_unlock(&journal
->j_state_lock
);
150 del_timer_sync(&journal
->j_commit_timer
);
151 journal_commit_transaction(journal
);
152 spin_lock(&journal
->j_state_lock
);
156 wake_up(&journal
->j_wait_done_commit
);
157 if (freezing(current
)) {
159 * The simpler the better. Flushing journal isn't a
160 * good idea, because that depends on threads that may
161 * be already stopped.
163 jbd_debug(1, "Now suspending kjournald\n");
164 spin_unlock(&journal
->j_state_lock
);
166 spin_lock(&journal
->j_state_lock
);
169 * We assume on resume that commits are already there,
173 int should_sleep
= 1;
175 prepare_to_wait(&journal
->j_wait_commit
, &wait
,
177 if (journal
->j_commit_sequence
!= journal
->j_commit_request
)
179 transaction
= journal
->j_running_transaction
;
180 if (transaction
&& time_after_eq(jiffies
,
181 transaction
->t_expires
))
183 if (journal
->j_flags
& JFS_UNMOUNT
)
186 spin_unlock(&journal
->j_state_lock
);
188 spin_lock(&journal
->j_state_lock
);
190 finish_wait(&journal
->j_wait_commit
, &wait
);
193 jbd_debug(1, "kjournald wakes\n");
196 * Were we woken up by a commit wakeup event?
198 transaction
= journal
->j_running_transaction
;
199 if (transaction
&& time_after_eq(jiffies
, transaction
->t_expires
)) {
200 journal
->j_commit_request
= transaction
->t_tid
;
201 jbd_debug(1, "woke because of timeout\n");
206 spin_unlock(&journal
->j_state_lock
);
207 del_timer_sync(&journal
->j_commit_timer
);
208 journal
->j_task
= NULL
;
209 wake_up(&journal
->j_wait_done_commit
);
210 jbd_debug(1, "Journal thread exiting.\n");
214 static int journal_start_thread(journal_t
*journal
)
216 struct task_struct
*t
;
218 t
= kthread_run(kjournald
, journal
, "kjournald");
222 wait_event(journal
->j_wait_done_commit
, journal
->j_task
!= NULL
);
226 static void journal_kill_thread(journal_t
*journal
)
228 spin_lock(&journal
->j_state_lock
);
229 journal
->j_flags
|= JFS_UNMOUNT
;
231 while (journal
->j_task
) {
232 wake_up(&journal
->j_wait_commit
);
233 spin_unlock(&journal
->j_state_lock
);
234 wait_event(journal
->j_wait_done_commit
,
235 journal
->j_task
== NULL
);
236 spin_lock(&journal
->j_state_lock
);
238 spin_unlock(&journal
->j_state_lock
);
242 * journal_write_metadata_buffer: write a metadata buffer to the journal.
244 * Writes a metadata buffer to a given disk block. The actual IO is not
245 * performed but a new buffer_head is constructed which labels the data
246 * to be written with the correct destination disk block.
248 * Any magic-number escaping which needs to be done will cause a
249 * copy-out here. If the buffer happens to start with the
250 * JFS_MAGIC_NUMBER, then we can't write it to the log directly: the
251 * magic number is only written to the log for descripter blocks. In
252 * this case, we copy the data and replace the first word with 0, and we
253 * return a result code which indicates that this buffer needs to be
254 * marked as an escaped buffer in the corresponding log descriptor
255 * block. The missing word can then be restored when the block is read
258 * If the source buffer has already been modified by a new transaction
259 * since we took the last commit snapshot, we use the frozen copy of
260 * that data for IO. If we end up using the existing buffer_head's data
261 * for the write, then we *have* to lock the buffer to prevent anyone
262 * else from using and possibly modifying it while the IO is in
265 * The function returns a pointer to the buffer_heads to be used for IO.
267 * We assume that the journal has already been locked in this function.
274 * Bit 0 set == escape performed on the data
275 * Bit 1 set == buffer copy-out performed (kfree the data after IO)
278 int journal_write_metadata_buffer(transaction_t
*transaction
,
279 struct journal_head
*jh_in
,
280 struct journal_head
**jh_out
,
281 unsigned int blocknr
)
283 int need_copy_out
= 0;
284 int done_copy_out
= 0;
287 struct buffer_head
*new_bh
;
288 struct journal_head
*new_jh
;
289 struct page
*new_page
;
290 unsigned int new_offset
;
291 struct buffer_head
*bh_in
= jh2bh(jh_in
);
292 journal_t
*journal
= transaction
->t_journal
;
295 * The buffer really shouldn't be locked: only the current committing
296 * transaction is allowed to write it, so nobody else is allowed
299 * akpm: except if we're journalling data, and write() output is
300 * also part of a shared mapping, and another thread has
301 * decided to launch a writepage() against this buffer.
303 J_ASSERT_BH(bh_in
, buffer_jbddirty(bh_in
));
305 new_bh
= alloc_buffer_head(GFP_NOFS
|__GFP_NOFAIL
);
306 /* keep subsequent assertions sane */
308 init_buffer(new_bh
, NULL
, NULL
);
309 atomic_set(&new_bh
->b_count
, 1);
310 new_jh
= journal_add_journal_head(new_bh
); /* This sleeps */
313 * If a new transaction has already done a buffer copy-out, then
314 * we use that version of the data for the commit.
316 jbd_lock_bh_state(bh_in
);
318 if (jh_in
->b_frozen_data
) {
320 new_page
= virt_to_page(jh_in
->b_frozen_data
);
321 new_offset
= offset_in_page(jh_in
->b_frozen_data
);
323 new_page
= jh2bh(jh_in
)->b_page
;
324 new_offset
= offset_in_page(jh2bh(jh_in
)->b_data
);
327 mapped_data
= kmap_atomic(new_page
, KM_USER0
);
331 if (*((__be32
*)(mapped_data
+ new_offset
)) ==
332 cpu_to_be32(JFS_MAGIC_NUMBER
)) {
336 kunmap_atomic(mapped_data
, KM_USER0
);
339 * Do we need to do a data copy?
341 if (need_copy_out
&& !done_copy_out
) {
344 jbd_unlock_bh_state(bh_in
);
345 tmp
= jbd_alloc(bh_in
->b_size
, GFP_NOFS
);
346 jbd_lock_bh_state(bh_in
);
347 if (jh_in
->b_frozen_data
) {
348 jbd_free(tmp
, bh_in
->b_size
);
352 jh_in
->b_frozen_data
= tmp
;
353 mapped_data
= kmap_atomic(new_page
, KM_USER0
);
354 memcpy(tmp
, mapped_data
+ new_offset
, jh2bh(jh_in
)->b_size
);
355 kunmap_atomic(mapped_data
, KM_USER0
);
357 new_page
= virt_to_page(tmp
);
358 new_offset
= offset_in_page(tmp
);
363 * Did we need to do an escaping? Now we've done all the
364 * copying, we can finally do so.
367 mapped_data
= kmap_atomic(new_page
, KM_USER0
);
368 *((unsigned int *)(mapped_data
+ new_offset
)) = 0;
369 kunmap_atomic(mapped_data
, KM_USER0
);
372 set_bh_page(new_bh
, new_page
, new_offset
);
373 new_jh
->b_transaction
= NULL
;
374 new_bh
->b_size
= jh2bh(jh_in
)->b_size
;
375 new_bh
->b_bdev
= transaction
->t_journal
->j_dev
;
376 new_bh
->b_blocknr
= blocknr
;
377 set_buffer_mapped(new_bh
);
378 set_buffer_dirty(new_bh
);
383 * The to-be-written buffer needs to get moved to the io queue,
384 * and the original buffer whose contents we are shadowing or
385 * copying is moved to the transaction's shadow queue.
387 JBUFFER_TRACE(jh_in
, "file as BJ_Shadow");
388 spin_lock(&journal
->j_list_lock
);
389 __journal_file_buffer(jh_in
, transaction
, BJ_Shadow
);
390 spin_unlock(&journal
->j_list_lock
);
391 jbd_unlock_bh_state(bh_in
);
393 JBUFFER_TRACE(new_jh
, "file as BJ_IO");
394 journal_file_buffer(new_jh
, transaction
, BJ_IO
);
396 return do_escape
| (done_copy_out
<< 1);
400 * Allocation code for the journal file. Manage the space left in the
401 * journal, so that we can begin checkpointing when appropriate.
405 * __log_space_left: Return the number of free blocks left in the journal.
407 * Called with the journal already locked.
409 * Called under j_state_lock
412 int __log_space_left(journal_t
*journal
)
414 int left
= journal
->j_free
;
416 assert_spin_locked(&journal
->j_state_lock
);
419 * Be pessimistic here about the number of those free blocks which
420 * might be required for log descriptor control blocks.
423 #define MIN_LOG_RESERVED_BLOCKS 32 /* Allow for rounding errors */
425 left
-= MIN_LOG_RESERVED_BLOCKS
;
434 * Called under j_state_lock. Returns true if a transaction commit was started.
436 int __log_start_commit(journal_t
*journal
, tid_t target
)
439 * Are we already doing a recent enough commit?
441 if (!tid_geq(journal
->j_commit_request
, target
)) {
443 * We want a new commit: OK, mark the request and wakup the
444 * commit thread. We do _not_ do the commit ourselves.
447 journal
->j_commit_request
= target
;
448 jbd_debug(1, "JBD: requesting commit %d/%d\n",
449 journal
->j_commit_request
,
450 journal
->j_commit_sequence
);
451 wake_up(&journal
->j_wait_commit
);
457 int log_start_commit(journal_t
*journal
, tid_t tid
)
461 spin_lock(&journal
->j_state_lock
);
462 ret
= __log_start_commit(journal
, tid
);
463 spin_unlock(&journal
->j_state_lock
);
468 * Force and wait upon a commit if the calling process is not within
469 * transaction. This is used for forcing out undo-protected data which contains
470 * bitmaps, when the fs is running out of space.
472 * We can only force the running transaction if we don't have an active handle;
473 * otherwise, we will deadlock.
475 * Returns true if a transaction was started.
477 int journal_force_commit_nested(journal_t
*journal
)
479 transaction_t
*transaction
= NULL
;
482 spin_lock(&journal
->j_state_lock
);
483 if (journal
->j_running_transaction
&& !current
->journal_info
) {
484 transaction
= journal
->j_running_transaction
;
485 __log_start_commit(journal
, transaction
->t_tid
);
486 } else if (journal
->j_committing_transaction
)
487 transaction
= journal
->j_committing_transaction
;
490 spin_unlock(&journal
->j_state_lock
);
491 return 0; /* Nothing to retry */
494 tid
= transaction
->t_tid
;
495 spin_unlock(&journal
->j_state_lock
);
496 log_wait_commit(journal
, tid
);
501 * Start a commit of the current running transaction (if any). Returns true
502 * if a transaction is going to be committed (or is currently already
503 * committing), and fills its tid in at *ptid
505 int journal_start_commit(journal_t
*journal
, tid_t
*ptid
)
509 spin_lock(&journal
->j_state_lock
);
510 if (journal
->j_running_transaction
) {
511 tid_t tid
= journal
->j_running_transaction
->t_tid
;
513 __log_start_commit(journal
, tid
);
514 /* There's a running transaction and we've just made sure
515 * it's commit has been scheduled. */
519 } else if (journal
->j_committing_transaction
) {
521 * If ext3_write_super() recently started a commit, then we
522 * have to wait for completion of that transaction
525 *ptid
= journal
->j_committing_transaction
->t_tid
;
528 spin_unlock(&journal
->j_state_lock
);
533 * Wait for a specified commit to complete.
534 * The caller may not hold the journal lock.
536 int log_wait_commit(journal_t
*journal
, tid_t tid
)
540 #ifdef CONFIG_JBD_DEBUG
541 spin_lock(&journal
->j_state_lock
);
542 if (!tid_geq(journal
->j_commit_request
, tid
)) {
544 "%s: error: j_commit_request=%d, tid=%d\n",
545 __func__
, journal
->j_commit_request
, tid
);
547 spin_unlock(&journal
->j_state_lock
);
549 spin_lock(&journal
->j_state_lock
);
550 while (tid_gt(tid
, journal
->j_commit_sequence
)) {
551 jbd_debug(1, "JBD: want %d, j_commit_sequence=%d\n",
552 tid
, journal
->j_commit_sequence
);
553 wake_up(&journal
->j_wait_commit
);
554 spin_unlock(&journal
->j_state_lock
);
555 wait_event(journal
->j_wait_done_commit
,
556 !tid_gt(tid
, journal
->j_commit_sequence
));
557 spin_lock(&journal
->j_state_lock
);
559 spin_unlock(&journal
->j_state_lock
);
561 if (unlikely(is_journal_aborted(journal
))) {
562 printk(KERN_EMERG
"journal commit I/O error\n");
569 * Return 1 if a given transaction has not yet sent barrier request
570 * connected with a transaction commit. If 0 is returned, transaction
571 * may or may not have sent the barrier. Used to avoid sending barrier
572 * twice in common cases.
574 int journal_trans_will_send_data_barrier(journal_t
*journal
, tid_t tid
)
577 transaction_t
*commit_trans
;
579 if (!(journal
->j_flags
& JFS_BARRIER
))
581 spin_lock(&journal
->j_state_lock
);
582 /* Transaction already committed? */
583 if (tid_geq(journal
->j_commit_sequence
, tid
))
586 * Transaction is being committed and we already proceeded to
587 * writing commit record?
589 commit_trans
= journal
->j_committing_transaction
;
590 if (commit_trans
&& commit_trans
->t_tid
== tid
&&
591 commit_trans
->t_state
>= T_COMMIT_RECORD
)
595 spin_unlock(&journal
->j_state_lock
);
598 EXPORT_SYMBOL(journal_trans_will_send_data_barrier
);
601 * Log buffer allocation routines:
604 int journal_next_log_block(journal_t
*journal
, unsigned int *retp
)
606 unsigned int blocknr
;
608 spin_lock(&journal
->j_state_lock
);
609 J_ASSERT(journal
->j_free
> 1);
611 blocknr
= journal
->j_head
;
614 if (journal
->j_head
== journal
->j_last
)
615 journal
->j_head
= journal
->j_first
;
616 spin_unlock(&journal
->j_state_lock
);
617 return journal_bmap(journal
, blocknr
, retp
);
621 * Conversion of logical to physical block numbers for the journal
623 * On external journals the journal blocks are identity-mapped, so
624 * this is a no-op. If needed, we can use j_blk_offset - everything is
627 int journal_bmap(journal_t
*journal
, unsigned int blocknr
,
633 if (journal
->j_inode
) {
634 ret
= bmap(journal
->j_inode
, blocknr
);
638 char b
[BDEVNAME_SIZE
];
640 printk(KERN_ALERT
"%s: journal block not found "
641 "at offset %u on %s\n",
644 bdevname(journal
->j_dev
, b
));
646 __journal_abort_soft(journal
, err
);
649 *retp
= blocknr
; /* +journal->j_blk_offset */
655 * We play buffer_head aliasing tricks to write data/metadata blocks to
656 * the journal without copying their contents, but for journal
657 * descriptor blocks we do need to generate bona fide buffers.
659 * After the caller of journal_get_descriptor_buffer() has finished modifying
660 * the buffer's contents they really should run flush_dcache_page(bh->b_page).
661 * But we don't bother doing that, so there will be coherency problems with
662 * mmaps of blockdevs which hold live JBD-controlled filesystems.
664 struct journal_head
*journal_get_descriptor_buffer(journal_t
*journal
)
666 struct buffer_head
*bh
;
667 unsigned int blocknr
;
670 err
= journal_next_log_block(journal
, &blocknr
);
675 bh
= __getblk(journal
->j_dev
, blocknr
, journal
->j_blocksize
);
679 memset(bh
->b_data
, 0, journal
->j_blocksize
);
680 set_buffer_uptodate(bh
);
682 BUFFER_TRACE(bh
, "return this buffer");
683 return journal_add_journal_head(bh
);
687 * Management for journal control blocks: functions to create and
688 * destroy journal_t structures, and to initialise and read existing
689 * journal blocks from disk. */
691 /* First: create and setup a journal_t object in memory. We initialise
692 * very few fields yet: that has to wait until we have created the
693 * journal structures from from scratch, or loaded them from disk. */
695 static journal_t
* journal_init_common (void)
700 journal
= kzalloc(sizeof(*journal
), GFP_KERNEL
);
704 init_waitqueue_head(&journal
->j_wait_transaction_locked
);
705 init_waitqueue_head(&journal
->j_wait_logspace
);
706 init_waitqueue_head(&journal
->j_wait_done_commit
);
707 init_waitqueue_head(&journal
->j_wait_checkpoint
);
708 init_waitqueue_head(&journal
->j_wait_commit
);
709 init_waitqueue_head(&journal
->j_wait_updates
);
710 mutex_init(&journal
->j_barrier
);
711 mutex_init(&journal
->j_checkpoint_mutex
);
712 spin_lock_init(&journal
->j_revoke_lock
);
713 spin_lock_init(&journal
->j_list_lock
);
714 spin_lock_init(&journal
->j_state_lock
);
716 journal
->j_commit_interval
= (HZ
* JBD_DEFAULT_MAX_COMMIT_AGE
);
718 /* The journal is marked for error until we succeed with recovery! */
719 journal
->j_flags
= JFS_ABORT
;
721 /* Set up a default-sized revoke table for the new mount. */
722 err
= journal_init_revoke(journal
, JOURNAL_REVOKE_DEFAULT_HASH
);
732 /* journal_init_dev and journal_init_inode:
734 * Create a journal structure assigned some fixed set of disk blocks to
735 * the journal. We don't actually touch those disk blocks yet, but we
736 * need to set up all of the mapping information to tell the journaling
737 * system where the journal blocks are.
742 * journal_t * journal_init_dev() - creates and initialises a journal structure
743 * @bdev: Block device on which to create the journal
744 * @fs_dev: Device which hold journalled filesystem for this journal.
745 * @start: Block nr Start of journal.
746 * @len: Length of the journal in blocks.
747 * @blocksize: blocksize of journalling device
749 * Returns: a newly created journal_t *
751 * journal_init_dev creates a journal which maps a fixed contiguous
752 * range of blocks on an arbitrary block device.
755 journal_t
* journal_init_dev(struct block_device
*bdev
,
756 struct block_device
*fs_dev
,
757 int start
, int len
, int blocksize
)
759 journal_t
*journal
= journal_init_common();
760 struct buffer_head
*bh
;
766 /* journal descriptor can store up to n blocks -bzzz */
767 journal
->j_blocksize
= blocksize
;
768 n
= journal
->j_blocksize
/ sizeof(journal_block_tag_t
);
769 journal
->j_wbufsize
= n
;
770 journal
->j_wbuf
= kmalloc(n
* sizeof(struct buffer_head
*), GFP_KERNEL
);
771 if (!journal
->j_wbuf
) {
772 printk(KERN_ERR
"%s: Cant allocate bhs for commit thread\n",
776 journal
->j_dev
= bdev
;
777 journal
->j_fs_dev
= fs_dev
;
778 journal
->j_blk_offset
= start
;
779 journal
->j_maxlen
= len
;
781 bh
= __getblk(journal
->j_dev
, start
, journal
->j_blocksize
);
784 "%s: Cannot get buffer for journal superblock\n",
788 journal
->j_sb_buffer
= bh
;
789 journal
->j_superblock
= (journal_superblock_t
*)bh
->b_data
;
793 kfree(journal
->j_wbuf
);
799 * journal_t * journal_init_inode () - creates a journal which maps to a inode.
800 * @inode: An inode to create the journal in
802 * journal_init_inode creates a journal which maps an on-disk inode as
803 * the journal. The inode must exist already, must support bmap() and
804 * must have all data blocks preallocated.
806 journal_t
* journal_init_inode (struct inode
*inode
)
808 struct buffer_head
*bh
;
809 journal_t
*journal
= journal_init_common();
812 unsigned int blocknr
;
817 journal
->j_dev
= journal
->j_fs_dev
= inode
->i_sb
->s_bdev
;
818 journal
->j_inode
= inode
;
820 "journal %p: inode %s/%ld, size %Ld, bits %d, blksize %ld\n",
821 journal
, inode
->i_sb
->s_id
, inode
->i_ino
,
822 (long long) inode
->i_size
,
823 inode
->i_sb
->s_blocksize_bits
, inode
->i_sb
->s_blocksize
);
825 journal
->j_maxlen
= inode
->i_size
>> inode
->i_sb
->s_blocksize_bits
;
826 journal
->j_blocksize
= inode
->i_sb
->s_blocksize
;
828 /* journal descriptor can store up to n blocks -bzzz */
829 n
= journal
->j_blocksize
/ sizeof(journal_block_tag_t
);
830 journal
->j_wbufsize
= n
;
831 journal
->j_wbuf
= kmalloc(n
* sizeof(struct buffer_head
*), GFP_KERNEL
);
832 if (!journal
->j_wbuf
) {
833 printk(KERN_ERR
"%s: Cant allocate bhs for commit thread\n",
838 err
= journal_bmap(journal
, 0, &blocknr
);
839 /* If that failed, give up */
841 printk(KERN_ERR
"%s: Cannnot locate journal superblock\n",
846 bh
= __getblk(journal
->j_dev
, blocknr
, journal
->j_blocksize
);
849 "%s: Cannot get buffer for journal superblock\n",
853 journal
->j_sb_buffer
= bh
;
854 journal
->j_superblock
= (journal_superblock_t
*)bh
->b_data
;
858 kfree(journal
->j_wbuf
);
864 * If the journal init or create aborts, we need to mark the journal
865 * superblock as being NULL to prevent the journal destroy from writing
866 * back a bogus superblock.
868 static void journal_fail_superblock (journal_t
*journal
)
870 struct buffer_head
*bh
= journal
->j_sb_buffer
;
872 journal
->j_sb_buffer
= NULL
;
876 * Given a journal_t structure, initialise the various fields for
877 * startup of a new journaling session. We use this both when creating
878 * a journal, and after recovering an old journal to reset it for
882 static int journal_reset(journal_t
*journal
)
884 journal_superblock_t
*sb
= journal
->j_superblock
;
885 unsigned int first
, last
;
887 first
= be32_to_cpu(sb
->s_first
);
888 last
= be32_to_cpu(sb
->s_maxlen
);
889 if (first
+ JFS_MIN_JOURNAL_BLOCKS
> last
+ 1) {
890 printk(KERN_ERR
"JBD: Journal too short (blocks %u-%u).\n",
892 journal_fail_superblock(journal
);
896 journal
->j_first
= first
;
897 journal
->j_last
= last
;
899 journal
->j_head
= first
;
900 journal
->j_tail
= first
;
901 journal
->j_free
= last
- first
;
903 journal
->j_tail_sequence
= journal
->j_transaction_sequence
;
904 journal
->j_commit_sequence
= journal
->j_transaction_sequence
- 1;
905 journal
->j_commit_request
= journal
->j_commit_sequence
;
907 journal
->j_max_transaction_buffers
= journal
->j_maxlen
/ 4;
909 /* Add the dynamic fields and write it to disk. */
910 journal_update_superblock(journal
, 1);
911 return journal_start_thread(journal
);
915 * int journal_create() - Initialise the new journal file
916 * @journal: Journal to create. This structure must have been initialised
918 * Given a journal_t structure which tells us which disk blocks we can
919 * use, create a new journal superblock and initialise all of the
920 * journal fields from scratch.
922 int journal_create(journal_t
*journal
)
924 unsigned int blocknr
;
925 struct buffer_head
*bh
;
926 journal_superblock_t
*sb
;
929 if (journal
->j_maxlen
< JFS_MIN_JOURNAL_BLOCKS
) {
930 printk (KERN_ERR
"Journal length (%d blocks) too short.\n",
932 journal_fail_superblock(journal
);
936 if (journal
->j_inode
== NULL
) {
938 * We don't know what block to start at!
941 "%s: creation of journal on external device!\n",
946 /* Zero out the entire journal on disk. We cannot afford to
947 have any blocks on disk beginning with JFS_MAGIC_NUMBER. */
948 jbd_debug(1, "JBD: Zeroing out journal blocks...\n");
949 for (i
= 0; i
< journal
->j_maxlen
; i
++) {
950 err
= journal_bmap(journal
, i
, &blocknr
);
953 bh
= __getblk(journal
->j_dev
, blocknr
, journal
->j_blocksize
);
955 memset (bh
->b_data
, 0, journal
->j_blocksize
);
956 BUFFER_TRACE(bh
, "marking dirty");
957 mark_buffer_dirty(bh
);
958 BUFFER_TRACE(bh
, "marking uptodate");
959 set_buffer_uptodate(bh
);
964 sync_blockdev(journal
->j_dev
);
965 jbd_debug(1, "JBD: journal cleared.\n");
967 /* OK, fill in the initial static fields in the new superblock */
968 sb
= journal
->j_superblock
;
970 sb
->s_header
.h_magic
= cpu_to_be32(JFS_MAGIC_NUMBER
);
971 sb
->s_header
.h_blocktype
= cpu_to_be32(JFS_SUPERBLOCK_V2
);
973 sb
->s_blocksize
= cpu_to_be32(journal
->j_blocksize
);
974 sb
->s_maxlen
= cpu_to_be32(journal
->j_maxlen
);
975 sb
->s_first
= cpu_to_be32(1);
977 journal
->j_transaction_sequence
= 1;
979 journal
->j_flags
&= ~JFS_ABORT
;
980 journal
->j_format_version
= 2;
982 return journal_reset(journal
);
986 * void journal_update_superblock() - Update journal sb on disk.
987 * @journal: The journal to update.
988 * @wait: Set to '0' if you don't want to wait for IO completion.
990 * Update a journal's dynamic superblock fields and write it to disk,
991 * optionally waiting for the IO to complete.
993 void journal_update_superblock(journal_t
*journal
, int wait
)
995 journal_superblock_t
*sb
= journal
->j_superblock
;
996 struct buffer_head
*bh
= journal
->j_sb_buffer
;
999 * As a special case, if the on-disk copy is already marked as needing
1000 * no recovery (s_start == 0) and there are no outstanding transactions
1001 * in the filesystem, then we can safely defer the superblock update
1002 * until the next commit by setting JFS_FLUSHED. This avoids
1003 * attempting a write to a potential-readonly device.
1005 if (sb
->s_start
== 0 && journal
->j_tail_sequence
==
1006 journal
->j_transaction_sequence
) {
1007 jbd_debug(1,"JBD: Skipping superblock update on recovered sb "
1008 "(start %u, seq %d, errno %d)\n",
1009 journal
->j_tail
, journal
->j_tail_sequence
,
1014 spin_lock(&journal
->j_state_lock
);
1015 jbd_debug(1,"JBD: updating superblock (start %u, seq %d, errno %d)\n",
1016 journal
->j_tail
, journal
->j_tail_sequence
, journal
->j_errno
);
1018 sb
->s_sequence
= cpu_to_be32(journal
->j_tail_sequence
);
1019 sb
->s_start
= cpu_to_be32(journal
->j_tail
);
1020 sb
->s_errno
= cpu_to_be32(journal
->j_errno
);
1021 spin_unlock(&journal
->j_state_lock
);
1023 BUFFER_TRACE(bh
, "marking dirty");
1024 mark_buffer_dirty(bh
);
1026 sync_dirty_buffer(bh
);
1028 write_dirty_buffer(bh
, WRITE
);
1031 /* If we have just flushed the log (by marking s_start==0), then
1032 * any future commit will have to be careful to update the
1033 * superblock again to re-record the true start of the log. */
1035 spin_lock(&journal
->j_state_lock
);
1037 journal
->j_flags
&= ~JFS_FLUSHED
;
1039 journal
->j_flags
|= JFS_FLUSHED
;
1040 spin_unlock(&journal
->j_state_lock
);
1044 * Read the superblock for a given journal, performing initial
1045 * validation of the format.
1048 static int journal_get_superblock(journal_t
*journal
)
1050 struct buffer_head
*bh
;
1051 journal_superblock_t
*sb
;
1054 bh
= journal
->j_sb_buffer
;
1056 J_ASSERT(bh
!= NULL
);
1057 if (!buffer_uptodate(bh
)) {
1058 ll_rw_block(READ
, 1, &bh
);
1060 if (!buffer_uptodate(bh
)) {
1062 "JBD: IO error reading journal superblock\n");
1067 sb
= journal
->j_superblock
;
1071 if (sb
->s_header
.h_magic
!= cpu_to_be32(JFS_MAGIC_NUMBER
) ||
1072 sb
->s_blocksize
!= cpu_to_be32(journal
->j_blocksize
)) {
1073 printk(KERN_WARNING
"JBD: no valid journal superblock found\n");
1077 switch(be32_to_cpu(sb
->s_header
.h_blocktype
)) {
1078 case JFS_SUPERBLOCK_V1
:
1079 journal
->j_format_version
= 1;
1081 case JFS_SUPERBLOCK_V2
:
1082 journal
->j_format_version
= 2;
1085 printk(KERN_WARNING
"JBD: unrecognised superblock format ID\n");
1089 if (be32_to_cpu(sb
->s_maxlen
) < journal
->j_maxlen
)
1090 journal
->j_maxlen
= be32_to_cpu(sb
->s_maxlen
);
1091 else if (be32_to_cpu(sb
->s_maxlen
) > journal
->j_maxlen
) {
1092 printk (KERN_WARNING
"JBD: journal file too short\n");
1099 journal_fail_superblock(journal
);
1104 * Load the on-disk journal superblock and read the key fields into the
1108 static int load_superblock(journal_t
*journal
)
1111 journal_superblock_t
*sb
;
1113 err
= journal_get_superblock(journal
);
1117 sb
= journal
->j_superblock
;
1119 journal
->j_tail_sequence
= be32_to_cpu(sb
->s_sequence
);
1120 journal
->j_tail
= be32_to_cpu(sb
->s_start
);
1121 journal
->j_first
= be32_to_cpu(sb
->s_first
);
1122 journal
->j_last
= be32_to_cpu(sb
->s_maxlen
);
1123 journal
->j_errno
= be32_to_cpu(sb
->s_errno
);
1130 * int journal_load() - Read journal from disk.
1131 * @journal: Journal to act on.
1133 * Given a journal_t structure which tells us which disk blocks contain
1134 * a journal, read the journal from disk to initialise the in-memory
1137 int journal_load(journal_t
*journal
)
1140 journal_superblock_t
*sb
;
1142 err
= load_superblock(journal
);
1146 sb
= journal
->j_superblock
;
1147 /* If this is a V2 superblock, then we have to check the
1148 * features flags on it. */
1150 if (journal
->j_format_version
>= 2) {
1151 if ((sb
->s_feature_ro_compat
&
1152 ~cpu_to_be32(JFS_KNOWN_ROCOMPAT_FEATURES
)) ||
1153 (sb
->s_feature_incompat
&
1154 ~cpu_to_be32(JFS_KNOWN_INCOMPAT_FEATURES
))) {
1155 printk (KERN_WARNING
1156 "JBD: Unrecognised features on journal\n");
1161 /* Let the recovery code check whether it needs to recover any
1162 * data from the journal. */
1163 if (journal_recover(journal
))
1164 goto recovery_error
;
1166 /* OK, we've finished with the dynamic journal bits:
1167 * reinitialise the dynamic contents of the superblock in memory
1168 * and reset them on disk. */
1169 if (journal_reset(journal
))
1170 goto recovery_error
;
1172 journal
->j_flags
&= ~JFS_ABORT
;
1173 journal
->j_flags
|= JFS_LOADED
;
1177 printk (KERN_WARNING
"JBD: recovery failed\n");
1182 * void journal_destroy() - Release a journal_t structure.
1183 * @journal: Journal to act on.
1185 * Release a journal_t structure once it is no longer in use by the
1187 * Return <0 if we couldn't clean up the journal.
1189 int journal_destroy(journal_t
*journal
)
1194 /* Wait for the commit thread to wake up and die. */
1195 journal_kill_thread(journal
);
1197 /* Force a final log commit */
1198 if (journal
->j_running_transaction
)
1199 journal_commit_transaction(journal
);
1201 /* Force any old transactions to disk */
1203 /* Totally anal locking here... */
1204 spin_lock(&journal
->j_list_lock
);
1205 while (journal
->j_checkpoint_transactions
!= NULL
) {
1206 spin_unlock(&journal
->j_list_lock
);
1207 log_do_checkpoint(journal
);
1208 spin_lock(&journal
->j_list_lock
);
1211 J_ASSERT(journal
->j_running_transaction
== NULL
);
1212 J_ASSERT(journal
->j_committing_transaction
== NULL
);
1213 J_ASSERT(journal
->j_checkpoint_transactions
== NULL
);
1214 spin_unlock(&journal
->j_list_lock
);
1216 if (journal
->j_sb_buffer
) {
1217 if (!is_journal_aborted(journal
)) {
1218 /* We can now mark the journal as empty. */
1219 journal
->j_tail
= 0;
1220 journal
->j_tail_sequence
=
1221 ++journal
->j_transaction_sequence
;
1222 journal_update_superblock(journal
, 1);
1226 brelse(journal
->j_sb_buffer
);
1229 if (journal
->j_inode
)
1230 iput(journal
->j_inode
);
1231 if (journal
->j_revoke
)
1232 journal_destroy_revoke(journal
);
1233 kfree(journal
->j_wbuf
);
1241 *int journal_check_used_features () - Check if features specified are used.
1242 * @journal: Journal to check.
1243 * @compat: bitmask of compatible features
1244 * @ro: bitmask of features that force read-only mount
1245 * @incompat: bitmask of incompatible features
1247 * Check whether the journal uses all of a given set of
1248 * features. Return true (non-zero) if it does.
1251 int journal_check_used_features (journal_t
*journal
, unsigned long compat
,
1252 unsigned long ro
, unsigned long incompat
)
1254 journal_superblock_t
*sb
;
1256 if (!compat
&& !ro
&& !incompat
)
1258 if (journal
->j_format_version
== 1)
1261 sb
= journal
->j_superblock
;
1263 if (((be32_to_cpu(sb
->s_feature_compat
) & compat
) == compat
) &&
1264 ((be32_to_cpu(sb
->s_feature_ro_compat
) & ro
) == ro
) &&
1265 ((be32_to_cpu(sb
->s_feature_incompat
) & incompat
) == incompat
))
1272 * int journal_check_available_features() - Check feature set in journalling layer
1273 * @journal: Journal to check.
1274 * @compat: bitmask of compatible features
1275 * @ro: bitmask of features that force read-only mount
1276 * @incompat: bitmask of incompatible features
1278 * Check whether the journaling code supports the use of
1279 * all of a given set of features on this journal. Return true
1280 * (non-zero) if it can. */
1282 int journal_check_available_features (journal_t
*journal
, unsigned long compat
,
1283 unsigned long ro
, unsigned long incompat
)
1285 if (!compat
&& !ro
&& !incompat
)
1288 /* We can support any known requested features iff the
1289 * superblock is in version 2. Otherwise we fail to support any
1290 * extended sb features. */
1292 if (journal
->j_format_version
!= 2)
1295 if ((compat
& JFS_KNOWN_COMPAT_FEATURES
) == compat
&&
1296 (ro
& JFS_KNOWN_ROCOMPAT_FEATURES
) == ro
&&
1297 (incompat
& JFS_KNOWN_INCOMPAT_FEATURES
) == incompat
)
1304 * int journal_set_features () - Mark a given journal feature in the superblock
1305 * @journal: Journal to act on.
1306 * @compat: bitmask of compatible features
1307 * @ro: bitmask of features that force read-only mount
1308 * @incompat: bitmask of incompatible features
1310 * Mark a given journal feature as present on the
1311 * superblock. Returns true if the requested features could be set.
1315 int journal_set_features (journal_t
*journal
, unsigned long compat
,
1316 unsigned long ro
, unsigned long incompat
)
1318 journal_superblock_t
*sb
;
1320 if (journal_check_used_features(journal
, compat
, ro
, incompat
))
1323 if (!journal_check_available_features(journal
, compat
, ro
, incompat
))
1326 jbd_debug(1, "Setting new features 0x%lx/0x%lx/0x%lx\n",
1327 compat
, ro
, incompat
);
1329 sb
= journal
->j_superblock
;
1331 sb
->s_feature_compat
|= cpu_to_be32(compat
);
1332 sb
->s_feature_ro_compat
|= cpu_to_be32(ro
);
1333 sb
->s_feature_incompat
|= cpu_to_be32(incompat
);
1340 * int journal_update_format () - Update on-disk journal structure.
1341 * @journal: Journal to act on.
1343 * Given an initialised but unloaded journal struct, poke about in the
1344 * on-disk structure to update it to the most recent supported version.
1346 int journal_update_format (journal_t
*journal
)
1348 journal_superblock_t
*sb
;
1351 err
= journal_get_superblock(journal
);
1355 sb
= journal
->j_superblock
;
1357 switch (be32_to_cpu(sb
->s_header
.h_blocktype
)) {
1358 case JFS_SUPERBLOCK_V2
:
1360 case JFS_SUPERBLOCK_V1
:
1361 return journal_convert_superblock_v1(journal
, sb
);
1368 static int journal_convert_superblock_v1(journal_t
*journal
,
1369 journal_superblock_t
*sb
)
1371 int offset
, blocksize
;
1372 struct buffer_head
*bh
;
1375 "JBD: Converting superblock from version 1 to 2.\n");
1377 /* Pre-initialise new fields to zero */
1378 offset
= ((char *) &(sb
->s_feature_compat
)) - ((char *) sb
);
1379 blocksize
= be32_to_cpu(sb
->s_blocksize
);
1380 memset(&sb
->s_feature_compat
, 0, blocksize
-offset
);
1382 sb
->s_nr_users
= cpu_to_be32(1);
1383 sb
->s_header
.h_blocktype
= cpu_to_be32(JFS_SUPERBLOCK_V2
);
1384 journal
->j_format_version
= 2;
1386 bh
= journal
->j_sb_buffer
;
1387 BUFFER_TRACE(bh
, "marking dirty");
1388 mark_buffer_dirty(bh
);
1389 sync_dirty_buffer(bh
);
1395 * int journal_flush () - Flush journal
1396 * @journal: Journal to act on.
1398 * Flush all data for a given journal to disk and empty the journal.
1399 * Filesystems can use this when remounting readonly to ensure that
1400 * recovery does not need to happen on remount.
1403 int journal_flush(journal_t
*journal
)
1406 transaction_t
*transaction
= NULL
;
1407 unsigned int old_tail
;
1409 spin_lock(&journal
->j_state_lock
);
1411 /* Force everything buffered to the log... */
1412 if (journal
->j_running_transaction
) {
1413 transaction
= journal
->j_running_transaction
;
1414 __log_start_commit(journal
, transaction
->t_tid
);
1415 } else if (journal
->j_committing_transaction
)
1416 transaction
= journal
->j_committing_transaction
;
1418 /* Wait for the log commit to complete... */
1420 tid_t tid
= transaction
->t_tid
;
1422 spin_unlock(&journal
->j_state_lock
);
1423 log_wait_commit(journal
, tid
);
1425 spin_unlock(&journal
->j_state_lock
);
1428 /* ...and flush everything in the log out to disk. */
1429 spin_lock(&journal
->j_list_lock
);
1430 while (!err
&& journal
->j_checkpoint_transactions
!= NULL
) {
1431 spin_unlock(&journal
->j_list_lock
);
1432 mutex_lock(&journal
->j_checkpoint_mutex
);
1433 err
= log_do_checkpoint(journal
);
1434 mutex_unlock(&journal
->j_checkpoint_mutex
);
1435 spin_lock(&journal
->j_list_lock
);
1437 spin_unlock(&journal
->j_list_lock
);
1439 if (is_journal_aborted(journal
))
1442 cleanup_journal_tail(journal
);
1444 /* Finally, mark the journal as really needing no recovery.
1445 * This sets s_start==0 in the underlying superblock, which is
1446 * the magic code for a fully-recovered superblock. Any future
1447 * commits of data to the journal will restore the current
1449 spin_lock(&journal
->j_state_lock
);
1450 old_tail
= journal
->j_tail
;
1451 journal
->j_tail
= 0;
1452 spin_unlock(&journal
->j_state_lock
);
1453 journal_update_superblock(journal
, 1);
1454 spin_lock(&journal
->j_state_lock
);
1455 journal
->j_tail
= old_tail
;
1457 J_ASSERT(!journal
->j_running_transaction
);
1458 J_ASSERT(!journal
->j_committing_transaction
);
1459 J_ASSERT(!journal
->j_checkpoint_transactions
);
1460 J_ASSERT(journal
->j_head
== journal
->j_tail
);
1461 J_ASSERT(journal
->j_tail_sequence
== journal
->j_transaction_sequence
);
1462 spin_unlock(&journal
->j_state_lock
);
1467 * int journal_wipe() - Wipe journal contents
1468 * @journal: Journal to act on.
1469 * @write: flag (see below)
1471 * Wipe out all of the contents of a journal, safely. This will produce
1472 * a warning if the journal contains any valid recovery information.
1473 * Must be called between journal_init_*() and journal_load().
1475 * If 'write' is non-zero, then we wipe out the journal on disk; otherwise
1476 * we merely suppress recovery.
1479 int journal_wipe(journal_t
*journal
, int write
)
1483 J_ASSERT (!(journal
->j_flags
& JFS_LOADED
));
1485 err
= load_superblock(journal
);
1489 if (!journal
->j_tail
)
1492 printk (KERN_WARNING
"JBD: %s recovery information on journal\n",
1493 write
? "Clearing" : "Ignoring");
1495 err
= journal_skip_recovery(journal
);
1497 journal_update_superblock(journal
, 1);
1504 * journal_dev_name: format a character string to describe on what
1505 * device this journal is present.
1508 static const char *journal_dev_name(journal_t
*journal
, char *buffer
)
1510 struct block_device
*bdev
;
1512 if (journal
->j_inode
)
1513 bdev
= journal
->j_inode
->i_sb
->s_bdev
;
1515 bdev
= journal
->j_dev
;
1517 return bdevname(bdev
, buffer
);
1521 * Journal abort has very specific semantics, which we describe
1522 * for journal abort.
1524 * Two internal function, which provide abort to te jbd layer
1529 * Quick version for internal journal use (doesn't lock the journal).
1530 * Aborts hard --- we mark the abort as occurred, but do _nothing_ else,
1531 * and don't attempt to make any other journal updates.
1533 static void __journal_abort_hard(journal_t
*journal
)
1535 transaction_t
*transaction
;
1536 char b
[BDEVNAME_SIZE
];
1538 if (journal
->j_flags
& JFS_ABORT
)
1541 printk(KERN_ERR
"Aborting journal on device %s.\n",
1542 journal_dev_name(journal
, b
));
1544 spin_lock(&journal
->j_state_lock
);
1545 journal
->j_flags
|= JFS_ABORT
;
1546 transaction
= journal
->j_running_transaction
;
1548 __log_start_commit(journal
, transaction
->t_tid
);
1549 spin_unlock(&journal
->j_state_lock
);
1552 /* Soft abort: record the abort error status in the journal superblock,
1553 * but don't do any other IO. */
1554 static void __journal_abort_soft (journal_t
*journal
, int errno
)
1556 if (journal
->j_flags
& JFS_ABORT
)
1559 if (!journal
->j_errno
)
1560 journal
->j_errno
= errno
;
1562 __journal_abort_hard(journal
);
1565 journal_update_superblock(journal
, 1);
1569 * void journal_abort () - Shutdown the journal immediately.
1570 * @journal: the journal to shutdown.
1571 * @errno: an error number to record in the journal indicating
1572 * the reason for the shutdown.
1574 * Perform a complete, immediate shutdown of the ENTIRE
1575 * journal (not of a single transaction). This operation cannot be
1576 * undone without closing and reopening the journal.
1578 * The journal_abort function is intended to support higher level error
1579 * recovery mechanisms such as the ext2/ext3 remount-readonly error
1582 * Journal abort has very specific semantics. Any existing dirty,
1583 * unjournaled buffers in the main filesystem will still be written to
1584 * disk by bdflush, but the journaling mechanism will be suspended
1585 * immediately and no further transaction commits will be honoured.
1587 * Any dirty, journaled buffers will be written back to disk without
1588 * hitting the journal. Atomicity cannot be guaranteed on an aborted
1589 * filesystem, but we _do_ attempt to leave as much data as possible
1590 * behind for fsck to use for cleanup.
1592 * Any attempt to get a new transaction handle on a journal which is in
1593 * ABORT state will just result in an -EROFS error return. A
1594 * journal_stop on an existing handle will return -EIO if we have
1595 * entered abort state during the update.
1597 * Recursive transactions are not disturbed by journal abort until the
1598 * final journal_stop, which will receive the -EIO error.
1600 * Finally, the journal_abort call allows the caller to supply an errno
1601 * which will be recorded (if possible) in the journal superblock. This
1602 * allows a client to record failure conditions in the middle of a
1603 * transaction without having to complete the transaction to record the
1604 * failure to disk. ext3_error, for example, now uses this
1607 * Errors which originate from within the journaling layer will NOT
1608 * supply an errno; a null errno implies that absolutely no further
1609 * writes are done to the journal (unless there are any already in
1614 void journal_abort(journal_t
*journal
, int errno
)
1616 __journal_abort_soft(journal
, errno
);
1620 * int journal_errno () - returns the journal's error state.
1621 * @journal: journal to examine.
1623 * This is the errno numbet set with journal_abort(), the last
1624 * time the journal was mounted - if the journal was stopped
1625 * without calling abort this will be 0.
1627 * If the journal has been aborted on this mount time -EROFS will
1630 int journal_errno(journal_t
*journal
)
1634 spin_lock(&journal
->j_state_lock
);
1635 if (journal
->j_flags
& JFS_ABORT
)
1638 err
= journal
->j_errno
;
1639 spin_unlock(&journal
->j_state_lock
);
1644 * int journal_clear_err () - clears the journal's error state
1645 * @journal: journal to act on.
1647 * An error must be cleared or Acked to take a FS out of readonly
1650 int journal_clear_err(journal_t
*journal
)
1654 spin_lock(&journal
->j_state_lock
);
1655 if (journal
->j_flags
& JFS_ABORT
)
1658 journal
->j_errno
= 0;
1659 spin_unlock(&journal
->j_state_lock
);
1664 * void journal_ack_err() - Ack journal err.
1665 * @journal: journal to act on.
1667 * An error must be cleared or Acked to take a FS out of readonly
1670 void journal_ack_err(journal_t
*journal
)
1672 spin_lock(&journal
->j_state_lock
);
1673 if (journal
->j_errno
)
1674 journal
->j_flags
|= JFS_ACK_ERR
;
1675 spin_unlock(&journal
->j_state_lock
);
1678 int journal_blocks_per_page(struct inode
*inode
)
1680 return 1 << (PAGE_CACHE_SHIFT
- inode
->i_sb
->s_blocksize_bits
);
1684 * Journal_head storage management
1686 static struct kmem_cache
*journal_head_cache
;
1687 #ifdef CONFIG_JBD_DEBUG
1688 static atomic_t nr_journal_heads
= ATOMIC_INIT(0);
1691 static int journal_init_journal_head_cache(void)
1695 J_ASSERT(journal_head_cache
== NULL
);
1696 journal_head_cache
= kmem_cache_create("journal_head",
1697 sizeof(struct journal_head
),
1699 SLAB_TEMPORARY
, /* flags */
1702 if (!journal_head_cache
) {
1704 printk(KERN_EMERG
"JBD: no memory for journal_head cache\n");
1709 static void journal_destroy_journal_head_cache(void)
1711 if (journal_head_cache
) {
1712 kmem_cache_destroy(journal_head_cache
);
1713 journal_head_cache
= NULL
;
1718 * journal_head splicing and dicing
1720 static struct journal_head
*journal_alloc_journal_head(void)
1722 struct journal_head
*ret
;
1724 #ifdef CONFIG_JBD_DEBUG
1725 atomic_inc(&nr_journal_heads
);
1727 ret
= kmem_cache_alloc(journal_head_cache
, GFP_NOFS
);
1729 jbd_debug(1, "out of memory for journal_head\n");
1730 printk_ratelimited(KERN_NOTICE
"ENOMEM in %s, retrying.\n",
1733 while (ret
== NULL
) {
1735 ret
= kmem_cache_alloc(journal_head_cache
, GFP_NOFS
);
1741 static void journal_free_journal_head(struct journal_head
*jh
)
1743 #ifdef CONFIG_JBD_DEBUG
1744 atomic_dec(&nr_journal_heads
);
1745 memset(jh
, JBD_POISON_FREE
, sizeof(*jh
));
1747 kmem_cache_free(journal_head_cache
, jh
);
1751 * A journal_head is attached to a buffer_head whenever JBD has an
1752 * interest in the buffer.
1754 * Whenever a buffer has an attached journal_head, its ->b_state:BH_JBD bit
1755 * is set. This bit is tested in core kernel code where we need to take
1756 * JBD-specific actions. Testing the zeroness of ->b_private is not reliable
1759 * When a buffer has its BH_JBD bit set, its ->b_count is elevated by one.
1761 * When a buffer has its BH_JBD bit set it is immune from being released by
1762 * core kernel code, mainly via ->b_count.
1764 * A journal_head may be detached from its buffer_head when the journal_head's
1765 * b_transaction, b_cp_transaction and b_next_transaction pointers are NULL.
1766 * Various places in JBD call journal_remove_journal_head() to indicate that the
1767 * journal_head can be dropped if needed.
1769 * Various places in the kernel want to attach a journal_head to a buffer_head
1770 * _before_ attaching the journal_head to a transaction. To protect the
1771 * journal_head in this situation, journal_add_journal_head elevates the
1772 * journal_head's b_jcount refcount by one. The caller must call
1773 * journal_put_journal_head() to undo this.
1775 * So the typical usage would be:
1777 * (Attach a journal_head if needed. Increments b_jcount)
1778 * struct journal_head *jh = journal_add_journal_head(bh);
1780 * jh->b_transaction = xxx;
1781 * journal_put_journal_head(jh);
1783 * Now, the journal_head's b_jcount is zero, but it is safe from being released
1784 * because it has a non-zero b_transaction.
1788 * Give a buffer_head a journal_head.
1790 * Doesn't need the journal lock.
1793 struct journal_head
*journal_add_journal_head(struct buffer_head
*bh
)
1795 struct journal_head
*jh
;
1796 struct journal_head
*new_jh
= NULL
;
1799 if (!buffer_jbd(bh
)) {
1800 new_jh
= journal_alloc_journal_head();
1801 memset(new_jh
, 0, sizeof(*new_jh
));
1804 jbd_lock_bh_journal_head(bh
);
1805 if (buffer_jbd(bh
)) {
1809 (atomic_read(&bh
->b_count
) > 0) ||
1810 (bh
->b_page
&& bh
->b_page
->mapping
));
1813 jbd_unlock_bh_journal_head(bh
);
1818 new_jh
= NULL
; /* We consumed it */
1823 BUFFER_TRACE(bh
, "added journal_head");
1826 jbd_unlock_bh_journal_head(bh
);
1828 journal_free_journal_head(new_jh
);
1829 return bh
->b_private
;
1833 * Grab a ref against this buffer_head's journal_head. If it ended up not
1834 * having a journal_head, return NULL
1836 struct journal_head
*journal_grab_journal_head(struct buffer_head
*bh
)
1838 struct journal_head
*jh
= NULL
;
1840 jbd_lock_bh_journal_head(bh
);
1841 if (buffer_jbd(bh
)) {
1845 jbd_unlock_bh_journal_head(bh
);
1849 static void __journal_remove_journal_head(struct buffer_head
*bh
)
1851 struct journal_head
*jh
= bh2jh(bh
);
1853 J_ASSERT_JH(jh
, jh
->b_jcount
>= 0);
1856 if (jh
->b_jcount
== 0) {
1857 if (jh
->b_transaction
== NULL
&&
1858 jh
->b_next_transaction
== NULL
&&
1859 jh
->b_cp_transaction
== NULL
) {
1860 J_ASSERT_JH(jh
, jh
->b_jlist
== BJ_None
);
1861 J_ASSERT_BH(bh
, buffer_jbd(bh
));
1862 J_ASSERT_BH(bh
, jh2bh(jh
) == bh
);
1863 BUFFER_TRACE(bh
, "remove journal_head");
1864 if (jh
->b_frozen_data
) {
1865 printk(KERN_WARNING
"%s: freeing "
1868 jbd_free(jh
->b_frozen_data
, bh
->b_size
);
1870 if (jh
->b_committed_data
) {
1871 printk(KERN_WARNING
"%s: freeing "
1872 "b_committed_data\n",
1874 jbd_free(jh
->b_committed_data
, bh
->b_size
);
1876 bh
->b_private
= NULL
;
1877 jh
->b_bh
= NULL
; /* debug, really */
1878 clear_buffer_jbd(bh
);
1880 journal_free_journal_head(jh
);
1882 BUFFER_TRACE(bh
, "journal_head was locked");
1888 * journal_remove_journal_head(): if the buffer isn't attached to a transaction
1889 * and has a zero b_jcount then remove and release its journal_head. If we did
1890 * see that the buffer is not used by any transaction we also "logically"
1891 * decrement ->b_count.
1893 * We in fact take an additional increment on ->b_count as a convenience,
1894 * because the caller usually wants to do additional things with the bh
1895 * after calling here.
1896 * The caller of journal_remove_journal_head() *must* run __brelse(bh) at some
1897 * time. Once the caller has run __brelse(), the buffer is eligible for
1898 * reaping by try_to_free_buffers().
1900 void journal_remove_journal_head(struct buffer_head
*bh
)
1902 jbd_lock_bh_journal_head(bh
);
1903 __journal_remove_journal_head(bh
);
1904 jbd_unlock_bh_journal_head(bh
);
1908 * Drop a reference on the passed journal_head. If it fell to zero then try to
1909 * release the journal_head from the buffer_head.
1911 void journal_put_journal_head(struct journal_head
*jh
)
1913 struct buffer_head
*bh
= jh2bh(jh
);
1915 jbd_lock_bh_journal_head(bh
);
1916 J_ASSERT_JH(jh
, jh
->b_jcount
> 0);
1918 if (!jh
->b_jcount
&& !jh
->b_transaction
) {
1919 __journal_remove_journal_head(bh
);
1922 jbd_unlock_bh_journal_head(bh
);
1928 #ifdef CONFIG_JBD_DEBUG
1930 u8 journal_enable_debug __read_mostly
;
1931 EXPORT_SYMBOL(journal_enable_debug
);
1933 static struct dentry
*jbd_debugfs_dir
;
1934 static struct dentry
*jbd_debug
;
1936 static void __init
jbd_create_debugfs_entry(void)
1938 jbd_debugfs_dir
= debugfs_create_dir("jbd", NULL
);
1939 if (jbd_debugfs_dir
)
1940 jbd_debug
= debugfs_create_u8("jbd-debug", S_IRUGO
| S_IWUSR
,
1942 &journal_enable_debug
);
1945 static void __exit
jbd_remove_debugfs_entry(void)
1947 debugfs_remove(jbd_debug
);
1948 debugfs_remove(jbd_debugfs_dir
);
1953 static inline void jbd_create_debugfs_entry(void)
1957 static inline void jbd_remove_debugfs_entry(void)
1963 struct kmem_cache
*jbd_handle_cache
;
1965 static int __init
journal_init_handle_cache(void)
1967 jbd_handle_cache
= kmem_cache_create("journal_handle",
1970 SLAB_TEMPORARY
, /* flags */
1972 if (jbd_handle_cache
== NULL
) {
1973 printk(KERN_EMERG
"JBD: failed to create handle cache\n");
1979 static void journal_destroy_handle_cache(void)
1981 if (jbd_handle_cache
)
1982 kmem_cache_destroy(jbd_handle_cache
);
1986 * Module startup and shutdown
1989 static int __init
journal_init_caches(void)
1993 ret
= journal_init_revoke_caches();
1995 ret
= journal_init_journal_head_cache();
1997 ret
= journal_init_handle_cache();
2001 static void journal_destroy_caches(void)
2003 journal_destroy_revoke_caches();
2004 journal_destroy_journal_head_cache();
2005 journal_destroy_handle_cache();
2008 static int __init
journal_init(void)
2012 BUILD_BUG_ON(sizeof(struct journal_superblock_s
) != 1024);
2014 ret
= journal_init_caches();
2016 journal_destroy_caches();
2017 jbd_create_debugfs_entry();
2021 static void __exit
journal_exit(void)
2023 #ifdef CONFIG_JBD_DEBUG
2024 int n
= atomic_read(&nr_journal_heads
);
2026 printk(KERN_EMERG
"JBD: leaked %d journal_heads!\n", n
);
2028 jbd_remove_debugfs_entry();
2029 journal_destroy_caches();
2032 MODULE_LICENSE("GPL");
2033 module_init(journal_init
);
2034 module_exit(journal_exit
);