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>
40 #include <asm/uaccess.h>
43 EXPORT_SYMBOL(journal_start
);
44 EXPORT_SYMBOL(journal_restart
);
45 EXPORT_SYMBOL(journal_extend
);
46 EXPORT_SYMBOL(journal_stop
);
47 EXPORT_SYMBOL(journal_lock_updates
);
48 EXPORT_SYMBOL(journal_unlock_updates
);
49 EXPORT_SYMBOL(journal_get_write_access
);
50 EXPORT_SYMBOL(journal_get_create_access
);
51 EXPORT_SYMBOL(journal_get_undo_access
);
52 EXPORT_SYMBOL(journal_dirty_data
);
53 EXPORT_SYMBOL(journal_dirty_metadata
);
54 EXPORT_SYMBOL(journal_release_buffer
);
55 EXPORT_SYMBOL(journal_forget
);
57 EXPORT_SYMBOL(journal_sync_buffer
);
59 EXPORT_SYMBOL(journal_flush
);
60 EXPORT_SYMBOL(journal_revoke
);
62 EXPORT_SYMBOL(journal_init_dev
);
63 EXPORT_SYMBOL(journal_init_inode
);
64 EXPORT_SYMBOL(journal_update_format
);
65 EXPORT_SYMBOL(journal_check_used_features
);
66 EXPORT_SYMBOL(journal_check_available_features
);
67 EXPORT_SYMBOL(journal_set_features
);
68 EXPORT_SYMBOL(journal_create
);
69 EXPORT_SYMBOL(journal_load
);
70 EXPORT_SYMBOL(journal_destroy
);
71 EXPORT_SYMBOL(journal_abort
);
72 EXPORT_SYMBOL(journal_errno
);
73 EXPORT_SYMBOL(journal_ack_err
);
74 EXPORT_SYMBOL(journal_clear_err
);
75 EXPORT_SYMBOL(log_wait_commit
);
76 EXPORT_SYMBOL(journal_start_commit
);
77 EXPORT_SYMBOL(journal_force_commit_nested
);
78 EXPORT_SYMBOL(journal_wipe
);
79 EXPORT_SYMBOL(journal_blocks_per_page
);
80 EXPORT_SYMBOL(journal_invalidatepage
);
81 EXPORT_SYMBOL(journal_try_to_free_buffers
);
82 EXPORT_SYMBOL(journal_force_commit
);
84 static int journal_convert_superblock_v1(journal_t
*, journal_superblock_t
*);
85 static void __journal_abort_soft (journal_t
*journal
, int errno
);
88 * Helper function used to manage commit timeouts
91 static void commit_timeout(unsigned long __data
)
93 struct task_struct
* p
= (struct task_struct
*) __data
;
99 * kjournald: The main thread function used to manage a logging device
102 * This kernel thread is responsible for two things:
104 * 1) COMMIT: Every so often we need to commit the current state of the
105 * filesystem to disk. The journal thread is responsible for writing
106 * all of the metadata buffers to disk.
108 * 2) CHECKPOINT: We cannot reuse a used section of the log file until all
109 * of the data in that part of the log has been rewritten elsewhere on
110 * the disk. Flushing these old buffers to reclaim space in the log is
111 * known as checkpointing, and this thread is responsible for that job.
114 static int kjournald(void *arg
)
116 journal_t
*journal
= arg
;
117 transaction_t
*transaction
;
120 * Set up an interval timer which can be used to trigger a commit wakeup
121 * after the commit interval expires
123 setup_timer(&journal
->j_commit_timer
, commit_timeout
,
124 (unsigned long)current
);
126 /* Record that the journal thread is running */
127 journal
->j_task
= current
;
128 wake_up(&journal
->j_wait_done_commit
);
130 printk(KERN_INFO
"kjournald starting. Commit interval %ld seconds\n",
131 journal
->j_commit_interval
/ HZ
);
134 * And now, wait forever for commit wakeup events.
136 spin_lock(&journal
->j_state_lock
);
139 if (journal
->j_flags
& JFS_UNMOUNT
)
142 jbd_debug(1, "commit_sequence=%d, commit_request=%d\n",
143 journal
->j_commit_sequence
, journal
->j_commit_request
);
145 if (journal
->j_commit_sequence
!= journal
->j_commit_request
) {
146 jbd_debug(1, "OK, requests differ\n");
147 spin_unlock(&journal
->j_state_lock
);
148 del_timer_sync(&journal
->j_commit_timer
);
149 journal_commit_transaction(journal
);
150 spin_lock(&journal
->j_state_lock
);
154 wake_up(&journal
->j_wait_done_commit
);
155 if (freezing(current
)) {
157 * The simpler the better. Flushing journal isn't a
158 * good idea, because that depends on threads that may
159 * be already stopped.
161 jbd_debug(1, "Now suspending kjournald\n");
162 spin_unlock(&journal
->j_state_lock
);
164 spin_lock(&journal
->j_state_lock
);
167 * We assume on resume that commits are already there,
171 int should_sleep
= 1;
173 prepare_to_wait(&journal
->j_wait_commit
, &wait
,
175 if (journal
->j_commit_sequence
!= journal
->j_commit_request
)
177 transaction
= journal
->j_running_transaction
;
178 if (transaction
&& time_after_eq(jiffies
,
179 transaction
->t_expires
))
181 if (journal
->j_flags
& JFS_UNMOUNT
)
184 spin_unlock(&journal
->j_state_lock
);
186 spin_lock(&journal
->j_state_lock
);
188 finish_wait(&journal
->j_wait_commit
, &wait
);
191 jbd_debug(1, "kjournald wakes\n");
194 * Were we woken up by a commit wakeup event?
196 transaction
= journal
->j_running_transaction
;
197 if (transaction
&& time_after_eq(jiffies
, transaction
->t_expires
)) {
198 journal
->j_commit_request
= transaction
->t_tid
;
199 jbd_debug(1, "woke because of timeout\n");
204 spin_unlock(&journal
->j_state_lock
);
205 del_timer_sync(&journal
->j_commit_timer
);
206 journal
->j_task
= NULL
;
207 wake_up(&journal
->j_wait_done_commit
);
208 jbd_debug(1, "Journal thread exiting.\n");
212 static int journal_start_thread(journal_t
*journal
)
214 struct task_struct
*t
;
216 t
= kthread_run(kjournald
, journal
, "kjournald");
220 wait_event(journal
->j_wait_done_commit
, journal
->j_task
!= NULL
);
224 static void journal_kill_thread(journal_t
*journal
)
226 spin_lock(&journal
->j_state_lock
);
227 journal
->j_flags
|= JFS_UNMOUNT
;
229 while (journal
->j_task
) {
230 wake_up(&journal
->j_wait_commit
);
231 spin_unlock(&journal
->j_state_lock
);
232 wait_event(journal
->j_wait_done_commit
,
233 journal
->j_task
== NULL
);
234 spin_lock(&journal
->j_state_lock
);
236 spin_unlock(&journal
->j_state_lock
);
240 * journal_write_metadata_buffer: write a metadata buffer to the journal.
242 * Writes a metadata buffer to a given disk block. The actual IO is not
243 * performed but a new buffer_head is constructed which labels the data
244 * to be written with the correct destination disk block.
246 * Any magic-number escaping which needs to be done will cause a
247 * copy-out here. If the buffer happens to start with the
248 * JFS_MAGIC_NUMBER, then we can't write it to the log directly: the
249 * magic number is only written to the log for descripter blocks. In
250 * this case, we copy the data and replace the first word with 0, and we
251 * return a result code which indicates that this buffer needs to be
252 * marked as an escaped buffer in the corresponding log descriptor
253 * block. The missing word can then be restored when the block is read
256 * If the source buffer has already been modified by a new transaction
257 * since we took the last commit snapshot, we use the frozen copy of
258 * that data for IO. If we end up using the existing buffer_head's data
259 * for the write, then we *have* to lock the buffer to prevent anyone
260 * else from using and possibly modifying it while the IO is in
263 * The function returns a pointer to the buffer_heads to be used for IO.
265 * We assume that the journal has already been locked in this function.
272 * Bit 0 set == escape performed on the data
273 * Bit 1 set == buffer copy-out performed (kfree the data after IO)
276 int journal_write_metadata_buffer(transaction_t
*transaction
,
277 struct journal_head
*jh_in
,
278 struct journal_head
**jh_out
,
279 unsigned long blocknr
)
281 int need_copy_out
= 0;
282 int done_copy_out
= 0;
285 struct buffer_head
*new_bh
;
286 struct journal_head
*new_jh
;
287 struct page
*new_page
;
288 unsigned int new_offset
;
289 struct buffer_head
*bh_in
= jh2bh(jh_in
);
292 * The buffer really shouldn't be locked: only the current committing
293 * transaction is allowed to write it, so nobody else is allowed
296 * akpm: except if we're journalling data, and write() output is
297 * also part of a shared mapping, and another thread has
298 * decided to launch a writepage() against this buffer.
300 J_ASSERT_BH(bh_in
, buffer_jbddirty(bh_in
));
302 new_bh
= alloc_buffer_head(GFP_NOFS
|__GFP_NOFAIL
);
305 * If a new transaction has already done a buffer copy-out, then
306 * we use that version of the data for the commit.
308 jbd_lock_bh_state(bh_in
);
310 if (jh_in
->b_frozen_data
) {
312 new_page
= virt_to_page(jh_in
->b_frozen_data
);
313 new_offset
= offset_in_page(jh_in
->b_frozen_data
);
315 new_page
= jh2bh(jh_in
)->b_page
;
316 new_offset
= offset_in_page(jh2bh(jh_in
)->b_data
);
319 mapped_data
= kmap_atomic(new_page
, KM_USER0
);
323 if (*((__be32
*)(mapped_data
+ new_offset
)) ==
324 cpu_to_be32(JFS_MAGIC_NUMBER
)) {
328 kunmap_atomic(mapped_data
, KM_USER0
);
331 * Do we need to do a data copy?
333 if (need_copy_out
&& !done_copy_out
) {
336 jbd_unlock_bh_state(bh_in
);
337 tmp
= jbd_alloc(bh_in
->b_size
, GFP_NOFS
);
338 jbd_lock_bh_state(bh_in
);
339 if (jh_in
->b_frozen_data
) {
340 jbd_free(tmp
, bh_in
->b_size
);
344 jh_in
->b_frozen_data
= tmp
;
345 mapped_data
= kmap_atomic(new_page
, KM_USER0
);
346 memcpy(tmp
, mapped_data
+ new_offset
, jh2bh(jh_in
)->b_size
);
347 kunmap_atomic(mapped_data
, KM_USER0
);
349 new_page
= virt_to_page(tmp
);
350 new_offset
= offset_in_page(tmp
);
355 * Did we need to do an escaping? Now we've done all the
356 * copying, we can finally do so.
359 mapped_data
= kmap_atomic(new_page
, KM_USER0
);
360 *((unsigned int *)(mapped_data
+ new_offset
)) = 0;
361 kunmap_atomic(mapped_data
, KM_USER0
);
364 /* keep subsequent assertions sane */
366 init_buffer(new_bh
, NULL
, NULL
);
367 atomic_set(&new_bh
->b_count
, 1);
368 jbd_unlock_bh_state(bh_in
);
370 new_jh
= journal_add_journal_head(new_bh
); /* This sleeps */
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 journal_file_buffer(jh_in
, transaction
, BJ_Shadow
);
389 JBUFFER_TRACE(new_jh
, "file as BJ_IO");
390 journal_file_buffer(new_jh
, transaction
, BJ_IO
);
392 return do_escape
| (done_copy_out
<< 1);
396 * Allocation code for the journal file. Manage the space left in the
397 * journal, so that we can begin checkpointing when appropriate.
401 * __log_space_left: Return the number of free blocks left in the journal.
403 * Called with the journal already locked.
405 * Called under j_state_lock
408 int __log_space_left(journal_t
*journal
)
410 int left
= journal
->j_free
;
412 assert_spin_locked(&journal
->j_state_lock
);
415 * Be pessimistic here about the number of those free blocks which
416 * might be required for log descriptor control blocks.
419 #define MIN_LOG_RESERVED_BLOCKS 32 /* Allow for rounding errors */
421 left
-= MIN_LOG_RESERVED_BLOCKS
;
430 * Called under j_state_lock. Returns true if a transaction commit was started.
432 int __log_start_commit(journal_t
*journal
, tid_t target
)
435 * Are we already doing a recent enough commit?
437 if (!tid_geq(journal
->j_commit_request
, target
)) {
439 * We want a new commit: OK, mark the request and wakup the
440 * commit thread. We do _not_ do the commit ourselves.
443 journal
->j_commit_request
= target
;
444 jbd_debug(1, "JBD: requesting commit %d/%d\n",
445 journal
->j_commit_request
,
446 journal
->j_commit_sequence
);
447 wake_up(&journal
->j_wait_commit
);
453 int log_start_commit(journal_t
*journal
, tid_t tid
)
457 spin_lock(&journal
->j_state_lock
);
458 ret
= __log_start_commit(journal
, tid
);
459 spin_unlock(&journal
->j_state_lock
);
464 * Force and wait upon a commit if the calling process is not within
465 * transaction. This is used for forcing out undo-protected data which contains
466 * bitmaps, when the fs is running out of space.
468 * We can only force the running transaction if we don't have an active handle;
469 * otherwise, we will deadlock.
471 * Returns true if a transaction was started.
473 int journal_force_commit_nested(journal_t
*journal
)
475 transaction_t
*transaction
= NULL
;
478 spin_lock(&journal
->j_state_lock
);
479 if (journal
->j_running_transaction
&& !current
->journal_info
) {
480 transaction
= journal
->j_running_transaction
;
481 __log_start_commit(journal
, transaction
->t_tid
);
482 } else if (journal
->j_committing_transaction
)
483 transaction
= journal
->j_committing_transaction
;
486 spin_unlock(&journal
->j_state_lock
);
487 return 0; /* Nothing to retry */
490 tid
= transaction
->t_tid
;
491 spin_unlock(&journal
->j_state_lock
);
492 log_wait_commit(journal
, tid
);
497 * Start a commit of the current running transaction (if any). Returns true
498 * if a transaction is going to be committed (or is currently already
499 * committing), and fills its tid in at *ptid
501 int journal_start_commit(journal_t
*journal
, tid_t
*ptid
)
505 spin_lock(&journal
->j_state_lock
);
506 if (journal
->j_running_transaction
) {
507 tid_t tid
= journal
->j_running_transaction
->t_tid
;
509 __log_start_commit(journal
, tid
);
510 /* There's a running transaction and we've just made sure
511 * it's commit has been scheduled. */
515 } else if (journal
->j_committing_transaction
) {
517 * If ext3_write_super() recently started a commit, then we
518 * have to wait for completion of that transaction
521 *ptid
= journal
->j_committing_transaction
->t_tid
;
524 spin_unlock(&journal
->j_state_lock
);
529 * Wait for a specified commit to complete.
530 * The caller may not hold the journal lock.
532 int log_wait_commit(journal_t
*journal
, tid_t tid
)
536 #ifdef CONFIG_JBD_DEBUG
537 spin_lock(&journal
->j_state_lock
);
538 if (!tid_geq(journal
->j_commit_request
, tid
)) {
540 "%s: error: j_commit_request=%d, tid=%d\n",
541 __func__
, journal
->j_commit_request
, tid
);
543 spin_unlock(&journal
->j_state_lock
);
545 spin_lock(&journal
->j_state_lock
);
546 while (tid_gt(tid
, journal
->j_commit_sequence
)) {
547 jbd_debug(1, "JBD: want %d, j_commit_sequence=%d\n",
548 tid
, journal
->j_commit_sequence
);
549 wake_up(&journal
->j_wait_commit
);
550 spin_unlock(&journal
->j_state_lock
);
551 wait_event(journal
->j_wait_done_commit
,
552 !tid_gt(tid
, journal
->j_commit_sequence
));
553 spin_lock(&journal
->j_state_lock
);
555 spin_unlock(&journal
->j_state_lock
);
557 if (unlikely(is_journal_aborted(journal
))) {
558 printk(KERN_EMERG
"journal commit I/O error\n");
565 * Log buffer allocation routines:
568 int journal_next_log_block(journal_t
*journal
, unsigned long *retp
)
570 unsigned long blocknr
;
572 spin_lock(&journal
->j_state_lock
);
573 J_ASSERT(journal
->j_free
> 1);
575 blocknr
= journal
->j_head
;
578 if (journal
->j_head
== journal
->j_last
)
579 journal
->j_head
= journal
->j_first
;
580 spin_unlock(&journal
->j_state_lock
);
581 return journal_bmap(journal
, blocknr
, retp
);
585 * Conversion of logical to physical block numbers for the journal
587 * On external journals the journal blocks are identity-mapped, so
588 * this is a no-op. If needed, we can use j_blk_offset - everything is
591 int journal_bmap(journal_t
*journal
, unsigned long blocknr
,
597 if (journal
->j_inode
) {
598 ret
= bmap(journal
->j_inode
, blocknr
);
602 char b
[BDEVNAME_SIZE
];
604 printk(KERN_ALERT
"%s: journal block not found "
605 "at offset %lu on %s\n",
608 bdevname(journal
->j_dev
, b
));
610 __journal_abort_soft(journal
, err
);
613 *retp
= blocknr
; /* +journal->j_blk_offset */
619 * We play buffer_head aliasing tricks to write data/metadata blocks to
620 * the journal without copying their contents, but for journal
621 * descriptor blocks we do need to generate bona fide buffers.
623 * After the caller of journal_get_descriptor_buffer() has finished modifying
624 * the buffer's contents they really should run flush_dcache_page(bh->b_page).
625 * But we don't bother doing that, so there will be coherency problems with
626 * mmaps of blockdevs which hold live JBD-controlled filesystems.
628 struct journal_head
*journal_get_descriptor_buffer(journal_t
*journal
)
630 struct buffer_head
*bh
;
631 unsigned long blocknr
;
634 err
= journal_next_log_block(journal
, &blocknr
);
639 bh
= __getblk(journal
->j_dev
, blocknr
, journal
->j_blocksize
);
643 memset(bh
->b_data
, 0, journal
->j_blocksize
);
644 set_buffer_uptodate(bh
);
646 BUFFER_TRACE(bh
, "return this buffer");
647 return journal_add_journal_head(bh
);
651 * Management for journal control blocks: functions to create and
652 * destroy journal_t structures, and to initialise and read existing
653 * journal blocks from disk. */
655 /* First: create and setup a journal_t object in memory. We initialise
656 * very few fields yet: that has to wait until we have created the
657 * journal structures from from scratch, or loaded them from disk. */
659 static journal_t
* journal_init_common (void)
664 journal
= kzalloc(sizeof(*journal
), GFP_KERNEL
);
668 init_waitqueue_head(&journal
->j_wait_transaction_locked
);
669 init_waitqueue_head(&journal
->j_wait_logspace
);
670 init_waitqueue_head(&journal
->j_wait_done_commit
);
671 init_waitqueue_head(&journal
->j_wait_checkpoint
);
672 init_waitqueue_head(&journal
->j_wait_commit
);
673 init_waitqueue_head(&journal
->j_wait_updates
);
674 mutex_init(&journal
->j_barrier
);
675 mutex_init(&journal
->j_checkpoint_mutex
);
676 spin_lock_init(&journal
->j_revoke_lock
);
677 spin_lock_init(&journal
->j_list_lock
);
678 spin_lock_init(&journal
->j_state_lock
);
680 journal
->j_commit_interval
= (HZ
* JBD_DEFAULT_MAX_COMMIT_AGE
);
682 /* The journal is marked for error until we succeed with recovery! */
683 journal
->j_flags
= JFS_ABORT
;
685 /* Set up a default-sized revoke table for the new mount. */
686 err
= journal_init_revoke(journal
, JOURNAL_REVOKE_DEFAULT_HASH
);
696 /* journal_init_dev and journal_init_inode:
698 * Create a journal structure assigned some fixed set of disk blocks to
699 * the journal. We don't actually touch those disk blocks yet, but we
700 * need to set up all of the mapping information to tell the journaling
701 * system where the journal blocks are.
706 * journal_t * journal_init_dev() - creates and initialises a journal structure
707 * @bdev: Block device on which to create the journal
708 * @fs_dev: Device which hold journalled filesystem for this journal.
709 * @start: Block nr Start of journal.
710 * @len: Length of the journal in blocks.
711 * @blocksize: blocksize of journalling device
713 * Returns: a newly created journal_t *
715 * journal_init_dev creates a journal which maps a fixed contiguous
716 * range of blocks on an arbitrary block device.
719 journal_t
* journal_init_dev(struct block_device
*bdev
,
720 struct block_device
*fs_dev
,
721 int start
, int len
, int blocksize
)
723 journal_t
*journal
= journal_init_common();
724 struct buffer_head
*bh
;
730 /* journal descriptor can store up to n blocks -bzzz */
731 journal
->j_blocksize
= blocksize
;
732 n
= journal
->j_blocksize
/ sizeof(journal_block_tag_t
);
733 journal
->j_wbufsize
= n
;
734 journal
->j_wbuf
= kmalloc(n
* sizeof(struct buffer_head
*), GFP_KERNEL
);
735 if (!journal
->j_wbuf
) {
736 printk(KERN_ERR
"%s: Cant allocate bhs for commit thread\n",
740 journal
->j_dev
= bdev
;
741 journal
->j_fs_dev
= fs_dev
;
742 journal
->j_blk_offset
= start
;
743 journal
->j_maxlen
= len
;
745 bh
= __getblk(journal
->j_dev
, start
, journal
->j_blocksize
);
748 "%s: Cannot get buffer for journal superblock\n",
752 journal
->j_sb_buffer
= bh
;
753 journal
->j_superblock
= (journal_superblock_t
*)bh
->b_data
;
762 * journal_t * journal_init_inode () - creates a journal which maps to a inode.
763 * @inode: An inode to create the journal in
765 * journal_init_inode creates a journal which maps an on-disk inode as
766 * the journal. The inode must exist already, must support bmap() and
767 * must have all data blocks preallocated.
769 journal_t
* journal_init_inode (struct inode
*inode
)
771 struct buffer_head
*bh
;
772 journal_t
*journal
= journal_init_common();
775 unsigned long blocknr
;
780 journal
->j_dev
= journal
->j_fs_dev
= inode
->i_sb
->s_bdev
;
781 journal
->j_inode
= inode
;
783 "journal %p: inode %s/%ld, size %Ld, bits %d, blksize %ld\n",
784 journal
, inode
->i_sb
->s_id
, inode
->i_ino
,
785 (long long) inode
->i_size
,
786 inode
->i_sb
->s_blocksize_bits
, inode
->i_sb
->s_blocksize
);
788 journal
->j_maxlen
= inode
->i_size
>> inode
->i_sb
->s_blocksize_bits
;
789 journal
->j_blocksize
= inode
->i_sb
->s_blocksize
;
791 /* journal descriptor can store up to n blocks -bzzz */
792 n
= journal
->j_blocksize
/ sizeof(journal_block_tag_t
);
793 journal
->j_wbufsize
= n
;
794 journal
->j_wbuf
= kmalloc(n
* sizeof(struct buffer_head
*), GFP_KERNEL
);
795 if (!journal
->j_wbuf
) {
796 printk(KERN_ERR
"%s: Cant allocate bhs for commit thread\n",
801 err
= journal_bmap(journal
, 0, &blocknr
);
802 /* If that failed, give up */
804 printk(KERN_ERR
"%s: Cannnot locate journal superblock\n",
809 bh
= __getblk(journal
->j_dev
, blocknr
, journal
->j_blocksize
);
812 "%s: Cannot get buffer for journal superblock\n",
816 journal
->j_sb_buffer
= bh
;
817 journal
->j_superblock
= (journal_superblock_t
*)bh
->b_data
;
826 * If the journal init or create aborts, we need to mark the journal
827 * superblock as being NULL to prevent the journal destroy from writing
828 * back a bogus superblock.
830 static void journal_fail_superblock (journal_t
*journal
)
832 struct buffer_head
*bh
= journal
->j_sb_buffer
;
834 journal
->j_sb_buffer
= NULL
;
838 * Given a journal_t structure, initialise the various fields for
839 * startup of a new journaling session. We use this both when creating
840 * a journal, and after recovering an old journal to reset it for
844 static int journal_reset(journal_t
*journal
)
846 journal_superblock_t
*sb
= journal
->j_superblock
;
847 unsigned long first
, last
;
849 first
= be32_to_cpu(sb
->s_first
);
850 last
= be32_to_cpu(sb
->s_maxlen
);
851 if (first
+ JFS_MIN_JOURNAL_BLOCKS
> last
+ 1) {
852 printk(KERN_ERR
"JBD: Journal too short (blocks %lu-%lu).\n",
854 journal_fail_superblock(journal
);
858 journal
->j_first
= first
;
859 journal
->j_last
= last
;
861 journal
->j_head
= first
;
862 journal
->j_tail
= first
;
863 journal
->j_free
= last
- first
;
865 journal
->j_tail_sequence
= journal
->j_transaction_sequence
;
866 journal
->j_commit_sequence
= journal
->j_transaction_sequence
- 1;
867 journal
->j_commit_request
= journal
->j_commit_sequence
;
869 journal
->j_max_transaction_buffers
= journal
->j_maxlen
/ 4;
871 /* Add the dynamic fields and write it to disk. */
872 journal_update_superblock(journal
, 1);
873 return journal_start_thread(journal
);
877 * int journal_create() - Initialise the new journal file
878 * @journal: Journal to create. This structure must have been initialised
880 * Given a journal_t structure which tells us which disk blocks we can
881 * use, create a new journal superblock and initialise all of the
882 * journal fields from scratch.
884 int journal_create(journal_t
*journal
)
886 unsigned long blocknr
;
887 struct buffer_head
*bh
;
888 journal_superblock_t
*sb
;
891 if (journal
->j_maxlen
< JFS_MIN_JOURNAL_BLOCKS
) {
892 printk (KERN_ERR
"Journal length (%d blocks) too short.\n",
894 journal_fail_superblock(journal
);
898 if (journal
->j_inode
== NULL
) {
900 * We don't know what block to start at!
903 "%s: creation of journal on external device!\n",
908 /* Zero out the entire journal on disk. We cannot afford to
909 have any blocks on disk beginning with JFS_MAGIC_NUMBER. */
910 jbd_debug(1, "JBD: Zeroing out journal blocks...\n");
911 for (i
= 0; i
< journal
->j_maxlen
; i
++) {
912 err
= journal_bmap(journal
, i
, &blocknr
);
915 bh
= __getblk(journal
->j_dev
, blocknr
, journal
->j_blocksize
);
917 memset (bh
->b_data
, 0, journal
->j_blocksize
);
918 BUFFER_TRACE(bh
, "marking dirty");
919 mark_buffer_dirty(bh
);
920 BUFFER_TRACE(bh
, "marking uptodate");
921 set_buffer_uptodate(bh
);
926 sync_blockdev(journal
->j_dev
);
927 jbd_debug(1, "JBD: journal cleared.\n");
929 /* OK, fill in the initial static fields in the new superblock */
930 sb
= journal
->j_superblock
;
932 sb
->s_header
.h_magic
= cpu_to_be32(JFS_MAGIC_NUMBER
);
933 sb
->s_header
.h_blocktype
= cpu_to_be32(JFS_SUPERBLOCK_V2
);
935 sb
->s_blocksize
= cpu_to_be32(journal
->j_blocksize
);
936 sb
->s_maxlen
= cpu_to_be32(journal
->j_maxlen
);
937 sb
->s_first
= cpu_to_be32(1);
939 journal
->j_transaction_sequence
= 1;
941 journal
->j_flags
&= ~JFS_ABORT
;
942 journal
->j_format_version
= 2;
944 return journal_reset(journal
);
948 * void journal_update_superblock() - Update journal sb on disk.
949 * @journal: The journal to update.
950 * @wait: Set to '0' if you don't want to wait for IO completion.
952 * Update a journal's dynamic superblock fields and write it to disk,
953 * optionally waiting for the IO to complete.
955 void journal_update_superblock(journal_t
*journal
, int wait
)
957 journal_superblock_t
*sb
= journal
->j_superblock
;
958 struct buffer_head
*bh
= journal
->j_sb_buffer
;
961 * As a special case, if the on-disk copy is already marked as needing
962 * no recovery (s_start == 0) and there are no outstanding transactions
963 * in the filesystem, then we can safely defer the superblock update
964 * until the next commit by setting JFS_FLUSHED. This avoids
965 * attempting a write to a potential-readonly device.
967 if (sb
->s_start
== 0 && journal
->j_tail_sequence
==
968 journal
->j_transaction_sequence
) {
969 jbd_debug(1,"JBD: Skipping superblock update on recovered sb "
970 "(start %ld, seq %d, errno %d)\n",
971 journal
->j_tail
, journal
->j_tail_sequence
,
976 spin_lock(&journal
->j_state_lock
);
977 jbd_debug(1,"JBD: updating superblock (start %ld, seq %d, errno %d)\n",
978 journal
->j_tail
, journal
->j_tail_sequence
, journal
->j_errno
);
980 sb
->s_sequence
= cpu_to_be32(journal
->j_tail_sequence
);
981 sb
->s_start
= cpu_to_be32(journal
->j_tail
);
982 sb
->s_errno
= cpu_to_be32(journal
->j_errno
);
983 spin_unlock(&journal
->j_state_lock
);
985 BUFFER_TRACE(bh
, "marking dirty");
986 mark_buffer_dirty(bh
);
988 sync_dirty_buffer(bh
);
990 ll_rw_block(SWRITE
, 1, &bh
);
993 /* If we have just flushed the log (by marking s_start==0), then
994 * any future commit will have to be careful to update the
995 * superblock again to re-record the true start of the log. */
997 spin_lock(&journal
->j_state_lock
);
999 journal
->j_flags
&= ~JFS_FLUSHED
;
1001 journal
->j_flags
|= JFS_FLUSHED
;
1002 spin_unlock(&journal
->j_state_lock
);
1006 * Read the superblock for a given journal, performing initial
1007 * validation of the format.
1010 static int journal_get_superblock(journal_t
*journal
)
1012 struct buffer_head
*bh
;
1013 journal_superblock_t
*sb
;
1016 bh
= journal
->j_sb_buffer
;
1018 J_ASSERT(bh
!= NULL
);
1019 if (!buffer_uptodate(bh
)) {
1020 ll_rw_block(READ
, 1, &bh
);
1022 if (!buffer_uptodate(bh
)) {
1024 "JBD: IO error reading journal superblock\n");
1029 sb
= journal
->j_superblock
;
1033 if (sb
->s_header
.h_magic
!= cpu_to_be32(JFS_MAGIC_NUMBER
) ||
1034 sb
->s_blocksize
!= cpu_to_be32(journal
->j_blocksize
)) {
1035 printk(KERN_WARNING
"JBD: no valid journal superblock found\n");
1039 switch(be32_to_cpu(sb
->s_header
.h_blocktype
)) {
1040 case JFS_SUPERBLOCK_V1
:
1041 journal
->j_format_version
= 1;
1043 case JFS_SUPERBLOCK_V2
:
1044 journal
->j_format_version
= 2;
1047 printk(KERN_WARNING
"JBD: unrecognised superblock format ID\n");
1051 if (be32_to_cpu(sb
->s_maxlen
) < journal
->j_maxlen
)
1052 journal
->j_maxlen
= be32_to_cpu(sb
->s_maxlen
);
1053 else if (be32_to_cpu(sb
->s_maxlen
) > journal
->j_maxlen
) {
1054 printk (KERN_WARNING
"JBD: journal file too short\n");
1061 journal_fail_superblock(journal
);
1066 * Load the on-disk journal superblock and read the key fields into the
1070 static int load_superblock(journal_t
*journal
)
1073 journal_superblock_t
*sb
;
1075 err
= journal_get_superblock(journal
);
1079 sb
= journal
->j_superblock
;
1081 journal
->j_tail_sequence
= be32_to_cpu(sb
->s_sequence
);
1082 journal
->j_tail
= be32_to_cpu(sb
->s_start
);
1083 journal
->j_first
= be32_to_cpu(sb
->s_first
);
1084 journal
->j_last
= be32_to_cpu(sb
->s_maxlen
);
1085 journal
->j_errno
= be32_to_cpu(sb
->s_errno
);
1092 * int journal_load() - Read journal from disk.
1093 * @journal: Journal to act on.
1095 * Given a journal_t structure which tells us which disk blocks contain
1096 * a journal, read the journal from disk to initialise the in-memory
1099 int journal_load(journal_t
*journal
)
1102 journal_superblock_t
*sb
;
1104 err
= load_superblock(journal
);
1108 sb
= journal
->j_superblock
;
1109 /* If this is a V2 superblock, then we have to check the
1110 * features flags on it. */
1112 if (journal
->j_format_version
>= 2) {
1113 if ((sb
->s_feature_ro_compat
&
1114 ~cpu_to_be32(JFS_KNOWN_ROCOMPAT_FEATURES
)) ||
1115 (sb
->s_feature_incompat
&
1116 ~cpu_to_be32(JFS_KNOWN_INCOMPAT_FEATURES
))) {
1117 printk (KERN_WARNING
1118 "JBD: Unrecognised features on journal\n");
1123 /* Let the recovery code check whether it needs to recover any
1124 * data from the journal. */
1125 if (journal_recover(journal
))
1126 goto recovery_error
;
1128 /* OK, we've finished with the dynamic journal bits:
1129 * reinitialise the dynamic contents of the superblock in memory
1130 * and reset them on disk. */
1131 if (journal_reset(journal
))
1132 goto recovery_error
;
1134 journal
->j_flags
&= ~JFS_ABORT
;
1135 journal
->j_flags
|= JFS_LOADED
;
1139 printk (KERN_WARNING
"JBD: recovery failed\n");
1144 * void journal_destroy() - Release a journal_t structure.
1145 * @journal: Journal to act on.
1147 * Release a journal_t structure once it is no longer in use by the
1149 * Return <0 if we couldn't clean up the journal.
1151 int journal_destroy(journal_t
*journal
)
1155 /* Wait for the commit thread to wake up and die. */
1156 journal_kill_thread(journal
);
1158 /* Force a final log commit */
1159 if (journal
->j_running_transaction
)
1160 journal_commit_transaction(journal
);
1162 /* Force any old transactions to disk */
1164 /* Totally anal locking here... */
1165 spin_lock(&journal
->j_list_lock
);
1166 while (journal
->j_checkpoint_transactions
!= NULL
) {
1167 spin_unlock(&journal
->j_list_lock
);
1168 log_do_checkpoint(journal
);
1169 spin_lock(&journal
->j_list_lock
);
1172 J_ASSERT(journal
->j_running_transaction
== NULL
);
1173 J_ASSERT(journal
->j_committing_transaction
== NULL
);
1174 J_ASSERT(journal
->j_checkpoint_transactions
== NULL
);
1175 spin_unlock(&journal
->j_list_lock
);
1177 if (journal
->j_sb_buffer
) {
1178 if (!is_journal_aborted(journal
)) {
1179 /* We can now mark the journal as empty. */
1180 journal
->j_tail
= 0;
1181 journal
->j_tail_sequence
=
1182 ++journal
->j_transaction_sequence
;
1183 journal_update_superblock(journal
, 1);
1187 brelse(journal
->j_sb_buffer
);
1190 if (journal
->j_inode
)
1191 iput(journal
->j_inode
);
1192 if (journal
->j_revoke
)
1193 journal_destroy_revoke(journal
);
1194 kfree(journal
->j_wbuf
);
1202 *int journal_check_used_features () - Check if features specified are used.
1203 * @journal: Journal to check.
1204 * @compat: bitmask of compatible features
1205 * @ro: bitmask of features that force read-only mount
1206 * @incompat: bitmask of incompatible features
1208 * Check whether the journal uses all of a given set of
1209 * features. Return true (non-zero) if it does.
1212 int journal_check_used_features (journal_t
*journal
, unsigned long compat
,
1213 unsigned long ro
, unsigned long incompat
)
1215 journal_superblock_t
*sb
;
1217 if (!compat
&& !ro
&& !incompat
)
1219 if (journal
->j_format_version
== 1)
1222 sb
= journal
->j_superblock
;
1224 if (((be32_to_cpu(sb
->s_feature_compat
) & compat
) == compat
) &&
1225 ((be32_to_cpu(sb
->s_feature_ro_compat
) & ro
) == ro
) &&
1226 ((be32_to_cpu(sb
->s_feature_incompat
) & incompat
) == incompat
))
1233 * int journal_check_available_features() - Check feature set in journalling layer
1234 * @journal: Journal to check.
1235 * @compat: bitmask of compatible features
1236 * @ro: bitmask of features that force read-only mount
1237 * @incompat: bitmask of incompatible features
1239 * Check whether the journaling code supports the use of
1240 * all of a given set of features on this journal. Return true
1241 * (non-zero) if it can. */
1243 int journal_check_available_features (journal_t
*journal
, unsigned long compat
,
1244 unsigned long ro
, unsigned long incompat
)
1246 journal_superblock_t
*sb
;
1248 if (!compat
&& !ro
&& !incompat
)
1251 sb
= journal
->j_superblock
;
1253 /* We can support any known requested features iff the
1254 * superblock is in version 2. Otherwise we fail to support any
1255 * extended sb features. */
1257 if (journal
->j_format_version
!= 2)
1260 if ((compat
& JFS_KNOWN_COMPAT_FEATURES
) == compat
&&
1261 (ro
& JFS_KNOWN_ROCOMPAT_FEATURES
) == ro
&&
1262 (incompat
& JFS_KNOWN_INCOMPAT_FEATURES
) == incompat
)
1269 * int journal_set_features () - Mark a given journal feature in the superblock
1270 * @journal: Journal to act on.
1271 * @compat: bitmask of compatible features
1272 * @ro: bitmask of features that force read-only mount
1273 * @incompat: bitmask of incompatible features
1275 * Mark a given journal feature as present on the
1276 * superblock. Returns true if the requested features could be set.
1280 int journal_set_features (journal_t
*journal
, unsigned long compat
,
1281 unsigned long ro
, unsigned long incompat
)
1283 journal_superblock_t
*sb
;
1285 if (journal_check_used_features(journal
, compat
, ro
, incompat
))
1288 if (!journal_check_available_features(journal
, compat
, ro
, incompat
))
1291 jbd_debug(1, "Setting new features 0x%lx/0x%lx/0x%lx\n",
1292 compat
, ro
, incompat
);
1294 sb
= journal
->j_superblock
;
1296 sb
->s_feature_compat
|= cpu_to_be32(compat
);
1297 sb
->s_feature_ro_compat
|= cpu_to_be32(ro
);
1298 sb
->s_feature_incompat
|= cpu_to_be32(incompat
);
1305 * int journal_update_format () - Update on-disk journal structure.
1306 * @journal: Journal to act on.
1308 * Given an initialised but unloaded journal struct, poke about in the
1309 * on-disk structure to update it to the most recent supported version.
1311 int journal_update_format (journal_t
*journal
)
1313 journal_superblock_t
*sb
;
1316 err
= journal_get_superblock(journal
);
1320 sb
= journal
->j_superblock
;
1322 switch (be32_to_cpu(sb
->s_header
.h_blocktype
)) {
1323 case JFS_SUPERBLOCK_V2
:
1325 case JFS_SUPERBLOCK_V1
:
1326 return journal_convert_superblock_v1(journal
, sb
);
1333 static int journal_convert_superblock_v1(journal_t
*journal
,
1334 journal_superblock_t
*sb
)
1336 int offset
, blocksize
;
1337 struct buffer_head
*bh
;
1340 "JBD: Converting superblock from version 1 to 2.\n");
1342 /* Pre-initialise new fields to zero */
1343 offset
= ((char *) &(sb
->s_feature_compat
)) - ((char *) sb
);
1344 blocksize
= be32_to_cpu(sb
->s_blocksize
);
1345 memset(&sb
->s_feature_compat
, 0, blocksize
-offset
);
1347 sb
->s_nr_users
= cpu_to_be32(1);
1348 sb
->s_header
.h_blocktype
= cpu_to_be32(JFS_SUPERBLOCK_V2
);
1349 journal
->j_format_version
= 2;
1351 bh
= journal
->j_sb_buffer
;
1352 BUFFER_TRACE(bh
, "marking dirty");
1353 mark_buffer_dirty(bh
);
1354 sync_dirty_buffer(bh
);
1360 * int journal_flush () - Flush journal
1361 * @journal: Journal to act on.
1363 * Flush all data for a given journal to disk and empty the journal.
1364 * Filesystems can use this when remounting readonly to ensure that
1365 * recovery does not need to happen on remount.
1368 int journal_flush(journal_t
*journal
)
1371 transaction_t
*transaction
= NULL
;
1372 unsigned long old_tail
;
1374 spin_lock(&journal
->j_state_lock
);
1376 /* Force everything buffered to the log... */
1377 if (journal
->j_running_transaction
) {
1378 transaction
= journal
->j_running_transaction
;
1379 __log_start_commit(journal
, transaction
->t_tid
);
1380 } else if (journal
->j_committing_transaction
)
1381 transaction
= journal
->j_committing_transaction
;
1383 /* Wait for the log commit to complete... */
1385 tid_t tid
= transaction
->t_tid
;
1387 spin_unlock(&journal
->j_state_lock
);
1388 log_wait_commit(journal
, tid
);
1390 spin_unlock(&journal
->j_state_lock
);
1393 /* ...and flush everything in the log out to disk. */
1394 spin_lock(&journal
->j_list_lock
);
1395 while (!err
&& journal
->j_checkpoint_transactions
!= NULL
) {
1396 spin_unlock(&journal
->j_list_lock
);
1397 mutex_lock(&journal
->j_checkpoint_mutex
);
1398 err
= log_do_checkpoint(journal
);
1399 mutex_unlock(&journal
->j_checkpoint_mutex
);
1400 spin_lock(&journal
->j_list_lock
);
1402 spin_unlock(&journal
->j_list_lock
);
1404 if (is_journal_aborted(journal
))
1407 cleanup_journal_tail(journal
);
1409 /* Finally, mark the journal as really needing no recovery.
1410 * This sets s_start==0 in the underlying superblock, which is
1411 * the magic code for a fully-recovered superblock. Any future
1412 * commits of data to the journal will restore the current
1414 spin_lock(&journal
->j_state_lock
);
1415 old_tail
= journal
->j_tail
;
1416 journal
->j_tail
= 0;
1417 spin_unlock(&journal
->j_state_lock
);
1418 journal_update_superblock(journal
, 1);
1419 spin_lock(&journal
->j_state_lock
);
1420 journal
->j_tail
= old_tail
;
1422 J_ASSERT(!journal
->j_running_transaction
);
1423 J_ASSERT(!journal
->j_committing_transaction
);
1424 J_ASSERT(!journal
->j_checkpoint_transactions
);
1425 J_ASSERT(journal
->j_head
== journal
->j_tail
);
1426 J_ASSERT(journal
->j_tail_sequence
== journal
->j_transaction_sequence
);
1427 spin_unlock(&journal
->j_state_lock
);
1432 * int journal_wipe() - Wipe journal contents
1433 * @journal: Journal to act on.
1434 * @write: flag (see below)
1436 * Wipe out all of the contents of a journal, safely. This will produce
1437 * a warning if the journal contains any valid recovery information.
1438 * Must be called between journal_init_*() and journal_load().
1440 * If 'write' is non-zero, then we wipe out the journal on disk; otherwise
1441 * we merely suppress recovery.
1444 int journal_wipe(journal_t
*journal
, int write
)
1446 journal_superblock_t
*sb
;
1449 J_ASSERT (!(journal
->j_flags
& JFS_LOADED
));
1451 err
= load_superblock(journal
);
1455 sb
= journal
->j_superblock
;
1457 if (!journal
->j_tail
)
1460 printk (KERN_WARNING
"JBD: %s recovery information on journal\n",
1461 write
? "Clearing" : "Ignoring");
1463 err
= journal_skip_recovery(journal
);
1465 journal_update_superblock(journal
, 1);
1472 * journal_dev_name: format a character string to describe on what
1473 * device this journal is present.
1476 static const char *journal_dev_name(journal_t
*journal
, char *buffer
)
1478 struct block_device
*bdev
;
1480 if (journal
->j_inode
)
1481 bdev
= journal
->j_inode
->i_sb
->s_bdev
;
1483 bdev
= journal
->j_dev
;
1485 return bdevname(bdev
, buffer
);
1489 * Journal abort has very specific semantics, which we describe
1490 * for journal abort.
1492 * Two internal function, which provide abort to te jbd layer
1497 * Quick version for internal journal use (doesn't lock the journal).
1498 * Aborts hard --- we mark the abort as occurred, but do _nothing_ else,
1499 * and don't attempt to make any other journal updates.
1501 static void __journal_abort_hard(journal_t
*journal
)
1503 transaction_t
*transaction
;
1504 char b
[BDEVNAME_SIZE
];
1506 if (journal
->j_flags
& JFS_ABORT
)
1509 printk(KERN_ERR
"Aborting journal on device %s.\n",
1510 journal_dev_name(journal
, b
));
1512 spin_lock(&journal
->j_state_lock
);
1513 journal
->j_flags
|= JFS_ABORT
;
1514 transaction
= journal
->j_running_transaction
;
1516 __log_start_commit(journal
, transaction
->t_tid
);
1517 spin_unlock(&journal
->j_state_lock
);
1520 /* Soft abort: record the abort error status in the journal superblock,
1521 * but don't do any other IO. */
1522 static void __journal_abort_soft (journal_t
*journal
, int errno
)
1524 if (journal
->j_flags
& JFS_ABORT
)
1527 if (!journal
->j_errno
)
1528 journal
->j_errno
= errno
;
1530 __journal_abort_hard(journal
);
1533 journal_update_superblock(journal
, 1);
1537 * void journal_abort () - Shutdown the journal immediately.
1538 * @journal: the journal to shutdown.
1539 * @errno: an error number to record in the journal indicating
1540 * the reason for the shutdown.
1542 * Perform a complete, immediate shutdown of the ENTIRE
1543 * journal (not of a single transaction). This operation cannot be
1544 * undone without closing and reopening the journal.
1546 * The journal_abort function is intended to support higher level error
1547 * recovery mechanisms such as the ext2/ext3 remount-readonly error
1550 * Journal abort has very specific semantics. Any existing dirty,
1551 * unjournaled buffers in the main filesystem will still be written to
1552 * disk by bdflush, but the journaling mechanism will be suspended
1553 * immediately and no further transaction commits will be honoured.
1555 * Any dirty, journaled buffers will be written back to disk without
1556 * hitting the journal. Atomicity cannot be guaranteed on an aborted
1557 * filesystem, but we _do_ attempt to leave as much data as possible
1558 * behind for fsck to use for cleanup.
1560 * Any attempt to get a new transaction handle on a journal which is in
1561 * ABORT state will just result in an -EROFS error return. A
1562 * journal_stop on an existing handle will return -EIO if we have
1563 * entered abort state during the update.
1565 * Recursive transactions are not disturbed by journal abort until the
1566 * final journal_stop, which will receive the -EIO error.
1568 * Finally, the journal_abort call allows the caller to supply an errno
1569 * which will be recorded (if possible) in the journal superblock. This
1570 * allows a client to record failure conditions in the middle of a
1571 * transaction without having to complete the transaction to record the
1572 * failure to disk. ext3_error, for example, now uses this
1575 * Errors which originate from within the journaling layer will NOT
1576 * supply an errno; a null errno implies that absolutely no further
1577 * writes are done to the journal (unless there are any already in
1582 void journal_abort(journal_t
*journal
, int errno
)
1584 __journal_abort_soft(journal
, errno
);
1588 * int journal_errno () - returns the journal's error state.
1589 * @journal: journal to examine.
1591 * This is the errno numbet set with journal_abort(), the last
1592 * time the journal was mounted - if the journal was stopped
1593 * without calling abort this will be 0.
1595 * If the journal has been aborted on this mount time -EROFS will
1598 int journal_errno(journal_t
*journal
)
1602 spin_lock(&journal
->j_state_lock
);
1603 if (journal
->j_flags
& JFS_ABORT
)
1606 err
= journal
->j_errno
;
1607 spin_unlock(&journal
->j_state_lock
);
1612 * int journal_clear_err () - clears the journal's error state
1613 * @journal: journal to act on.
1615 * An error must be cleared or Acked to take a FS out of readonly
1618 int journal_clear_err(journal_t
*journal
)
1622 spin_lock(&journal
->j_state_lock
);
1623 if (journal
->j_flags
& JFS_ABORT
)
1626 journal
->j_errno
= 0;
1627 spin_unlock(&journal
->j_state_lock
);
1632 * void journal_ack_err() - Ack journal err.
1633 * @journal: journal to act on.
1635 * An error must be cleared or Acked to take a FS out of readonly
1638 void journal_ack_err(journal_t
*journal
)
1640 spin_lock(&journal
->j_state_lock
);
1641 if (journal
->j_errno
)
1642 journal
->j_flags
|= JFS_ACK_ERR
;
1643 spin_unlock(&journal
->j_state_lock
);
1646 int journal_blocks_per_page(struct inode
*inode
)
1648 return 1 << (PAGE_CACHE_SHIFT
- inode
->i_sb
->s_blocksize_bits
);
1652 * Journal_head storage management
1654 static struct kmem_cache
*journal_head_cache
;
1655 #ifdef CONFIG_JBD_DEBUG
1656 static atomic_t nr_journal_heads
= ATOMIC_INIT(0);
1659 static int journal_init_journal_head_cache(void)
1663 J_ASSERT(journal_head_cache
== NULL
);
1664 journal_head_cache
= kmem_cache_create("journal_head",
1665 sizeof(struct journal_head
),
1667 SLAB_TEMPORARY
, /* flags */
1670 if (!journal_head_cache
) {
1672 printk(KERN_EMERG
"JBD: no memory for journal_head cache\n");
1677 static void journal_destroy_journal_head_cache(void)
1679 if (journal_head_cache
) {
1680 kmem_cache_destroy(journal_head_cache
);
1681 journal_head_cache
= NULL
;
1686 * journal_head splicing and dicing
1688 static struct journal_head
*journal_alloc_journal_head(void)
1690 struct journal_head
*ret
;
1691 static unsigned long last_warning
;
1693 #ifdef CONFIG_JBD_DEBUG
1694 atomic_inc(&nr_journal_heads
);
1696 ret
= kmem_cache_alloc(journal_head_cache
, GFP_NOFS
);
1698 jbd_debug(1, "out of memory for journal_head\n");
1699 if (time_after(jiffies
, last_warning
+ 5*HZ
)) {
1700 printk(KERN_NOTICE
"ENOMEM in %s, retrying.\n",
1702 last_warning
= jiffies
;
1704 while (ret
== NULL
) {
1706 ret
= kmem_cache_alloc(journal_head_cache
, GFP_NOFS
);
1712 static void journal_free_journal_head(struct journal_head
*jh
)
1714 #ifdef CONFIG_JBD_DEBUG
1715 atomic_dec(&nr_journal_heads
);
1716 memset(jh
, JBD_POISON_FREE
, sizeof(*jh
));
1718 kmem_cache_free(journal_head_cache
, jh
);
1722 * A journal_head is attached to a buffer_head whenever JBD has an
1723 * interest in the buffer.
1725 * Whenever a buffer has an attached journal_head, its ->b_state:BH_JBD bit
1726 * is set. This bit is tested in core kernel code where we need to take
1727 * JBD-specific actions. Testing the zeroness of ->b_private is not reliable
1730 * When a buffer has its BH_JBD bit set, its ->b_count is elevated by one.
1732 * When a buffer has its BH_JBD bit set it is immune from being released by
1733 * core kernel code, mainly via ->b_count.
1735 * A journal_head may be detached from its buffer_head when the journal_head's
1736 * b_transaction, b_cp_transaction and b_next_transaction pointers are NULL.
1737 * Various places in JBD call journal_remove_journal_head() to indicate that the
1738 * journal_head can be dropped if needed.
1740 * Various places in the kernel want to attach a journal_head to a buffer_head
1741 * _before_ attaching the journal_head to a transaction. To protect the
1742 * journal_head in this situation, journal_add_journal_head elevates the
1743 * journal_head's b_jcount refcount by one. The caller must call
1744 * journal_put_journal_head() to undo this.
1746 * So the typical usage would be:
1748 * (Attach a journal_head if needed. Increments b_jcount)
1749 * struct journal_head *jh = journal_add_journal_head(bh);
1751 * jh->b_transaction = xxx;
1752 * journal_put_journal_head(jh);
1754 * Now, the journal_head's b_jcount is zero, but it is safe from being released
1755 * because it has a non-zero b_transaction.
1759 * Give a buffer_head a journal_head.
1761 * Doesn't need the journal lock.
1764 struct journal_head
*journal_add_journal_head(struct buffer_head
*bh
)
1766 struct journal_head
*jh
;
1767 struct journal_head
*new_jh
= NULL
;
1770 if (!buffer_jbd(bh
)) {
1771 new_jh
= journal_alloc_journal_head();
1772 memset(new_jh
, 0, sizeof(*new_jh
));
1775 jbd_lock_bh_journal_head(bh
);
1776 if (buffer_jbd(bh
)) {
1780 (atomic_read(&bh
->b_count
) > 0) ||
1781 (bh
->b_page
&& bh
->b_page
->mapping
));
1784 jbd_unlock_bh_journal_head(bh
);
1789 new_jh
= NULL
; /* We consumed it */
1794 BUFFER_TRACE(bh
, "added journal_head");
1797 jbd_unlock_bh_journal_head(bh
);
1799 journal_free_journal_head(new_jh
);
1800 return bh
->b_private
;
1804 * Grab a ref against this buffer_head's journal_head. If it ended up not
1805 * having a journal_head, return NULL
1807 struct journal_head
*journal_grab_journal_head(struct buffer_head
*bh
)
1809 struct journal_head
*jh
= NULL
;
1811 jbd_lock_bh_journal_head(bh
);
1812 if (buffer_jbd(bh
)) {
1816 jbd_unlock_bh_journal_head(bh
);
1820 static void __journal_remove_journal_head(struct buffer_head
*bh
)
1822 struct journal_head
*jh
= bh2jh(bh
);
1824 J_ASSERT_JH(jh
, jh
->b_jcount
>= 0);
1827 if (jh
->b_jcount
== 0) {
1828 if (jh
->b_transaction
== NULL
&&
1829 jh
->b_next_transaction
== NULL
&&
1830 jh
->b_cp_transaction
== NULL
) {
1831 J_ASSERT_JH(jh
, jh
->b_jlist
== BJ_None
);
1832 J_ASSERT_BH(bh
, buffer_jbd(bh
));
1833 J_ASSERT_BH(bh
, jh2bh(jh
) == bh
);
1834 BUFFER_TRACE(bh
, "remove journal_head");
1835 if (jh
->b_frozen_data
) {
1836 printk(KERN_WARNING
"%s: freeing "
1839 jbd_free(jh
->b_frozen_data
, bh
->b_size
);
1841 if (jh
->b_committed_data
) {
1842 printk(KERN_WARNING
"%s: freeing "
1843 "b_committed_data\n",
1845 jbd_free(jh
->b_committed_data
, bh
->b_size
);
1847 bh
->b_private
= NULL
;
1848 jh
->b_bh
= NULL
; /* debug, really */
1849 clear_buffer_jbd(bh
);
1851 journal_free_journal_head(jh
);
1853 BUFFER_TRACE(bh
, "journal_head was locked");
1859 * journal_remove_journal_head(): if the buffer isn't attached to a transaction
1860 * and has a zero b_jcount then remove and release its journal_head. If we did
1861 * see that the buffer is not used by any transaction we also "logically"
1862 * decrement ->b_count.
1864 * We in fact take an additional increment on ->b_count as a convenience,
1865 * because the caller usually wants to do additional things with the bh
1866 * after calling here.
1867 * The caller of journal_remove_journal_head() *must* run __brelse(bh) at some
1868 * time. Once the caller has run __brelse(), the buffer is eligible for
1869 * reaping by try_to_free_buffers().
1871 void journal_remove_journal_head(struct buffer_head
*bh
)
1873 jbd_lock_bh_journal_head(bh
);
1874 __journal_remove_journal_head(bh
);
1875 jbd_unlock_bh_journal_head(bh
);
1879 * Drop a reference on the passed journal_head. If it fell to zero then try to
1880 * release the journal_head from the buffer_head.
1882 void journal_put_journal_head(struct journal_head
*jh
)
1884 struct buffer_head
*bh
= jh2bh(jh
);
1886 jbd_lock_bh_journal_head(bh
);
1887 J_ASSERT_JH(jh
, jh
->b_jcount
> 0);
1889 if (!jh
->b_jcount
&& !jh
->b_transaction
) {
1890 __journal_remove_journal_head(bh
);
1893 jbd_unlock_bh_journal_head(bh
);
1899 #ifdef CONFIG_JBD_DEBUG
1901 u8 journal_enable_debug __read_mostly
;
1902 EXPORT_SYMBOL(journal_enable_debug
);
1904 static struct dentry
*jbd_debugfs_dir
;
1905 static struct dentry
*jbd_debug
;
1907 static void __init
jbd_create_debugfs_entry(void)
1909 jbd_debugfs_dir
= debugfs_create_dir("jbd", NULL
);
1910 if (jbd_debugfs_dir
)
1911 jbd_debug
= debugfs_create_u8("jbd-debug", S_IRUGO
,
1913 &journal_enable_debug
);
1916 static void __exit
jbd_remove_debugfs_entry(void)
1918 debugfs_remove(jbd_debug
);
1919 debugfs_remove(jbd_debugfs_dir
);
1924 static inline void jbd_create_debugfs_entry(void)
1928 static inline void jbd_remove_debugfs_entry(void)
1934 struct kmem_cache
*jbd_handle_cache
;
1936 static int __init
journal_init_handle_cache(void)
1938 jbd_handle_cache
= kmem_cache_create("journal_handle",
1941 SLAB_TEMPORARY
, /* flags */
1943 if (jbd_handle_cache
== NULL
) {
1944 printk(KERN_EMERG
"JBD: failed to create handle cache\n");
1950 static void journal_destroy_handle_cache(void)
1952 if (jbd_handle_cache
)
1953 kmem_cache_destroy(jbd_handle_cache
);
1957 * Module startup and shutdown
1960 static int __init
journal_init_caches(void)
1964 ret
= journal_init_revoke_caches();
1966 ret
= journal_init_journal_head_cache();
1968 ret
= journal_init_handle_cache();
1972 static void journal_destroy_caches(void)
1974 journal_destroy_revoke_caches();
1975 journal_destroy_journal_head_cache();
1976 journal_destroy_handle_cache();
1979 static int __init
journal_init(void)
1983 BUILD_BUG_ON(sizeof(struct journal_superblock_s
) != 1024);
1985 ret
= journal_init_caches();
1987 journal_destroy_caches();
1988 jbd_create_debugfs_entry();
1992 static void __exit
journal_exit(void)
1994 #ifdef CONFIG_JBD_DEBUG
1995 int n
= atomic_read(&nr_journal_heads
);
1997 printk(KERN_EMERG
"JBD: leaked %d journal_heads!\n", n
);
1999 jbd_remove_debugfs_entry();
2000 journal_destroy_caches();
2003 MODULE_LICENSE("GPL");
2004 module_init(journal_init
);
2005 module_exit(journal_exit
);