2 * linux/fs/jbd2/journal.c
4 * Written by Stephen C. Tweedie <sct@redhat.com>, 1998
6 * Copyright 1998 Red Hat corp --- All Rights Reserved
8 * This file is part of the Linux kernel and is made available under
9 * the terms of the GNU General Public License, version 2, or at your
10 * option, any later version, incorporated herein by reference.
12 * Generic filesystem journal-writing code; part of the ext2fs
15 * This file manages journals: areas of disk reserved for logging
16 * transactional updates. This includes the kernel journaling thread
17 * which is responsible for scheduling updates to the log.
19 * We do not actually manage the physical storage of the journal in this
20 * file: that is left to a per-journal policy function, which allows us
21 * to store the journal within a filesystem-specified area for ext2
22 * journaling (ext2 can use a reserved inode for storing the log).
25 #include <linux/module.h>
26 #include <linux/time.h>
28 #include <linux/jbd2.h>
29 #include <linux/errno.h>
30 #include <linux/slab.h>
31 #include <linux/init.h>
33 #include <linux/freezer.h>
34 #include <linux/pagemap.h>
35 #include <linux/kthread.h>
36 #include <linux/poison.h>
37 #include <linux/proc_fs.h>
38 #include <linux/debugfs.h>
39 #include <linux/seq_file.h>
41 #include <asm/uaccess.h>
44 EXPORT_SYMBOL(jbd2_journal_start
);
45 EXPORT_SYMBOL(jbd2_journal_restart
);
46 EXPORT_SYMBOL(jbd2_journal_extend
);
47 EXPORT_SYMBOL(jbd2_journal_stop
);
48 EXPORT_SYMBOL(jbd2_journal_lock_updates
);
49 EXPORT_SYMBOL(jbd2_journal_unlock_updates
);
50 EXPORT_SYMBOL(jbd2_journal_get_write_access
);
51 EXPORT_SYMBOL(jbd2_journal_get_create_access
);
52 EXPORT_SYMBOL(jbd2_journal_get_undo_access
);
53 EXPORT_SYMBOL(jbd2_journal_dirty_metadata
);
54 EXPORT_SYMBOL(jbd2_journal_release_buffer
);
55 EXPORT_SYMBOL(jbd2_journal_forget
);
57 EXPORT_SYMBOL(journal_sync_buffer
);
59 EXPORT_SYMBOL(jbd2_journal_flush
);
60 EXPORT_SYMBOL(jbd2_journal_revoke
);
62 EXPORT_SYMBOL(jbd2_journal_init_dev
);
63 EXPORT_SYMBOL(jbd2_journal_init_inode
);
64 EXPORT_SYMBOL(jbd2_journal_update_format
);
65 EXPORT_SYMBOL(jbd2_journal_check_used_features
);
66 EXPORT_SYMBOL(jbd2_journal_check_available_features
);
67 EXPORT_SYMBOL(jbd2_journal_set_features
);
68 EXPORT_SYMBOL(jbd2_journal_create
);
69 EXPORT_SYMBOL(jbd2_journal_load
);
70 EXPORT_SYMBOL(jbd2_journal_destroy
);
71 EXPORT_SYMBOL(jbd2_journal_abort
);
72 EXPORT_SYMBOL(jbd2_journal_errno
);
73 EXPORT_SYMBOL(jbd2_journal_ack_err
);
74 EXPORT_SYMBOL(jbd2_journal_clear_err
);
75 EXPORT_SYMBOL(jbd2_log_wait_commit
);
76 EXPORT_SYMBOL(jbd2_journal_start_commit
);
77 EXPORT_SYMBOL(jbd2_journal_force_commit_nested
);
78 EXPORT_SYMBOL(jbd2_journal_wipe
);
79 EXPORT_SYMBOL(jbd2_journal_blocks_per_page
);
80 EXPORT_SYMBOL(jbd2_journal_invalidatepage
);
81 EXPORT_SYMBOL(jbd2_journal_try_to_free_buffers
);
82 EXPORT_SYMBOL(jbd2_journal_force_commit
);
83 EXPORT_SYMBOL(jbd2_journal_file_inode
);
84 EXPORT_SYMBOL(jbd2_journal_init_jbd_inode
);
85 EXPORT_SYMBOL(jbd2_journal_release_jbd_inode
);
86 EXPORT_SYMBOL(jbd2_journal_begin_ordered_truncate
);
88 static int journal_convert_superblock_v1(journal_t
*, journal_superblock_t
*);
89 static void __journal_abort_soft (journal_t
*journal
, int errno
);
92 * Helper function used to manage commit timeouts
95 static void commit_timeout(unsigned long __data
)
97 struct task_struct
* p
= (struct task_struct
*) __data
;
103 * kjournald2: The main thread function used to manage a logging device
106 * This kernel thread is responsible for two things:
108 * 1) COMMIT: Every so often we need to commit the current state of the
109 * filesystem to disk. The journal thread is responsible for writing
110 * all of the metadata buffers to disk.
112 * 2) CHECKPOINT: We cannot reuse a used section of the log file until all
113 * of the data in that part of the log has been rewritten elsewhere on
114 * the disk. Flushing these old buffers to reclaim space in the log is
115 * known as checkpointing, and this thread is responsible for that job.
118 static int kjournald2(void *arg
)
120 journal_t
*journal
= arg
;
121 transaction_t
*transaction
;
124 * Set up an interval timer which can be used to trigger a commit wakeup
125 * after the commit interval expires
127 setup_timer(&journal
->j_commit_timer
, commit_timeout
,
128 (unsigned long)current
);
130 /* Record that the journal thread is running */
131 journal
->j_task
= current
;
132 wake_up(&journal
->j_wait_done_commit
);
134 printk(KERN_INFO
"kjournald2 starting. Commit interval %ld seconds\n",
135 journal
->j_commit_interval
/ HZ
);
138 * And now, wait forever for commit wakeup events.
140 spin_lock(&journal
->j_state_lock
);
143 if (journal
->j_flags
& JBD2_UNMOUNT
)
146 jbd_debug(1, "commit_sequence=%d, commit_request=%d\n",
147 journal
->j_commit_sequence
, journal
->j_commit_request
);
149 if (journal
->j_commit_sequence
!= journal
->j_commit_request
) {
150 jbd_debug(1, "OK, requests differ\n");
151 spin_unlock(&journal
->j_state_lock
);
152 del_timer_sync(&journal
->j_commit_timer
);
153 jbd2_journal_commit_transaction(journal
);
154 spin_lock(&journal
->j_state_lock
);
158 wake_up(&journal
->j_wait_done_commit
);
159 if (freezing(current
)) {
161 * The simpler the better. Flushing journal isn't a
162 * good idea, because that depends on threads that may
163 * be already stopped.
165 jbd_debug(1, "Now suspending kjournald2\n");
166 spin_unlock(&journal
->j_state_lock
);
168 spin_lock(&journal
->j_state_lock
);
171 * We assume on resume that commits are already there,
175 int should_sleep
= 1;
177 prepare_to_wait(&journal
->j_wait_commit
, &wait
,
179 if (journal
->j_commit_sequence
!= journal
->j_commit_request
)
181 transaction
= journal
->j_running_transaction
;
182 if (transaction
&& time_after_eq(jiffies
,
183 transaction
->t_expires
))
185 if (journal
->j_flags
& JBD2_UNMOUNT
)
188 spin_unlock(&journal
->j_state_lock
);
190 spin_lock(&journal
->j_state_lock
);
192 finish_wait(&journal
->j_wait_commit
, &wait
);
195 jbd_debug(1, "kjournald2 wakes\n");
198 * Were we woken up by a commit wakeup event?
200 transaction
= journal
->j_running_transaction
;
201 if (transaction
&& time_after_eq(jiffies
, transaction
->t_expires
)) {
202 journal
->j_commit_request
= transaction
->t_tid
;
203 jbd_debug(1, "woke because of timeout\n");
208 spin_unlock(&journal
->j_state_lock
);
209 del_timer_sync(&journal
->j_commit_timer
);
210 journal
->j_task
= NULL
;
211 wake_up(&journal
->j_wait_done_commit
);
212 jbd_debug(1, "Journal thread exiting.\n");
216 static int jbd2_journal_start_thread(journal_t
*journal
)
218 struct task_struct
*t
;
220 t
= kthread_run(kjournald2
, journal
, "kjournald2");
224 wait_event(journal
->j_wait_done_commit
, journal
->j_task
!= NULL
);
228 static void journal_kill_thread(journal_t
*journal
)
230 spin_lock(&journal
->j_state_lock
);
231 journal
->j_flags
|= JBD2_UNMOUNT
;
233 while (journal
->j_task
) {
234 wake_up(&journal
->j_wait_commit
);
235 spin_unlock(&journal
->j_state_lock
);
236 wait_event(journal
->j_wait_done_commit
, journal
->j_task
== NULL
);
237 spin_lock(&journal
->j_state_lock
);
239 spin_unlock(&journal
->j_state_lock
);
243 * jbd2_journal_write_metadata_buffer: write a metadata buffer to the journal.
245 * Writes a metadata buffer to a given disk block. The actual IO is not
246 * performed but a new buffer_head is constructed which labels the data
247 * to be written with the correct destination disk block.
249 * Any magic-number escaping which needs to be done will cause a
250 * copy-out here. If the buffer happens to start with the
251 * JBD2_MAGIC_NUMBER, then we can't write it to the log directly: the
252 * magic number is only written to the log for descripter blocks. In
253 * this case, we copy the data and replace the first word with 0, and we
254 * return a result code which indicates that this buffer needs to be
255 * marked as an escaped buffer in the corresponding log descriptor
256 * block. The missing word can then be restored when the block is read
259 * If the source buffer has already been modified by a new transaction
260 * since we took the last commit snapshot, we use the frozen copy of
261 * that data for IO. If we end up using the existing buffer_head's data
262 * for the write, then we *have* to lock the buffer to prevent anyone
263 * else from using and possibly modifying it while the IO is in
266 * The function returns a pointer to the buffer_heads to be used for IO.
268 * We assume that the journal has already been locked in this function.
275 * Bit 0 set == escape performed on the data
276 * Bit 1 set == buffer copy-out performed (kfree the data after IO)
279 int jbd2_journal_write_metadata_buffer(transaction_t
*transaction
,
280 struct journal_head
*jh_in
,
281 struct journal_head
**jh_out
,
282 unsigned long long blocknr
)
284 int need_copy_out
= 0;
285 int done_copy_out
= 0;
288 struct buffer_head
*new_bh
;
289 struct journal_head
*new_jh
;
290 struct page
*new_page
;
291 unsigned int new_offset
;
292 struct buffer_head
*bh_in
= jh2bh(jh_in
);
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
);
308 * If a new transaction has already done a buffer copy-out, then
309 * we use that version of the data for the commit.
311 jbd_lock_bh_state(bh_in
);
313 if (jh_in
->b_frozen_data
) {
315 new_page
= virt_to_page(jh_in
->b_frozen_data
);
316 new_offset
= offset_in_page(jh_in
->b_frozen_data
);
318 new_page
= jh2bh(jh_in
)->b_page
;
319 new_offset
= offset_in_page(jh2bh(jh_in
)->b_data
);
322 mapped_data
= kmap_atomic(new_page
, KM_USER0
);
326 if (*((__be32
*)(mapped_data
+ new_offset
)) ==
327 cpu_to_be32(JBD2_MAGIC_NUMBER
)) {
331 kunmap_atomic(mapped_data
, KM_USER0
);
334 * Do we need to do a data copy?
336 if (need_copy_out
&& !done_copy_out
) {
339 jbd_unlock_bh_state(bh_in
);
340 tmp
= jbd2_alloc(bh_in
->b_size
, GFP_NOFS
);
341 jbd_lock_bh_state(bh_in
);
342 if (jh_in
->b_frozen_data
) {
343 jbd2_free(tmp
, bh_in
->b_size
);
347 jh_in
->b_frozen_data
= tmp
;
348 mapped_data
= kmap_atomic(new_page
, KM_USER0
);
349 memcpy(tmp
, mapped_data
+ new_offset
, jh2bh(jh_in
)->b_size
);
350 kunmap_atomic(mapped_data
, KM_USER0
);
352 new_page
= virt_to_page(tmp
);
353 new_offset
= offset_in_page(tmp
);
358 * Did we need to do an escaping? Now we've done all the
359 * copying, we can finally do so.
362 mapped_data
= kmap_atomic(new_page
, KM_USER0
);
363 *((unsigned int *)(mapped_data
+ new_offset
)) = 0;
364 kunmap_atomic(mapped_data
, KM_USER0
);
367 /* keep subsequent assertions sane */
369 init_buffer(new_bh
, NULL
, NULL
);
370 atomic_set(&new_bh
->b_count
, 1);
371 jbd_unlock_bh_state(bh_in
);
373 new_jh
= jbd2_journal_add_journal_head(new_bh
); /* This sleeps */
375 set_bh_page(new_bh
, new_page
, new_offset
);
376 new_jh
->b_transaction
= NULL
;
377 new_bh
->b_size
= jh2bh(jh_in
)->b_size
;
378 new_bh
->b_bdev
= transaction
->t_journal
->j_dev
;
379 new_bh
->b_blocknr
= blocknr
;
380 set_buffer_mapped(new_bh
);
381 set_buffer_dirty(new_bh
);
386 * The to-be-written buffer needs to get moved to the io queue,
387 * and the original buffer whose contents we are shadowing or
388 * copying is moved to the transaction's shadow queue.
390 JBUFFER_TRACE(jh_in
, "file as BJ_Shadow");
391 jbd2_journal_file_buffer(jh_in
, transaction
, BJ_Shadow
);
392 JBUFFER_TRACE(new_jh
, "file as BJ_IO");
393 jbd2_journal_file_buffer(new_jh
, transaction
, BJ_IO
);
395 return do_escape
| (done_copy_out
<< 1);
399 * Allocation code for the journal file. Manage the space left in the
400 * journal, so that we can begin checkpointing when appropriate.
404 * __jbd2_log_space_left: Return the number of free blocks left in the journal.
406 * Called with the journal already locked.
408 * Called under j_state_lock
411 int __jbd2_log_space_left(journal_t
*journal
)
413 int left
= journal
->j_free
;
415 assert_spin_locked(&journal
->j_state_lock
);
418 * Be pessimistic here about the number of those free blocks which
419 * might be required for log descriptor control blocks.
422 #define MIN_LOG_RESERVED_BLOCKS 32 /* Allow for rounding errors */
424 left
-= MIN_LOG_RESERVED_BLOCKS
;
433 * Called under j_state_lock. Returns true if a transaction commit was started.
435 int __jbd2_log_start_commit(journal_t
*journal
, tid_t target
)
438 * Are we already doing a recent enough commit?
440 if (!tid_geq(journal
->j_commit_request
, target
)) {
442 * We want a new commit: OK, mark the request and wakup the
443 * commit thread. We do _not_ do the commit ourselves.
446 journal
->j_commit_request
= target
;
447 jbd_debug(1, "JBD: requesting commit %d/%d\n",
448 journal
->j_commit_request
,
449 journal
->j_commit_sequence
);
450 wake_up(&journal
->j_wait_commit
);
456 int jbd2_log_start_commit(journal_t
*journal
, tid_t tid
)
460 spin_lock(&journal
->j_state_lock
);
461 ret
= __jbd2_log_start_commit(journal
, tid
);
462 spin_unlock(&journal
->j_state_lock
);
467 * Force and wait upon a commit if the calling process is not within
468 * transaction. This is used for forcing out undo-protected data which contains
469 * bitmaps, when the fs is running out of space.
471 * We can only force the running transaction if we don't have an active handle;
472 * otherwise, we will deadlock.
474 * Returns true if a transaction was started.
476 int jbd2_journal_force_commit_nested(journal_t
*journal
)
478 transaction_t
*transaction
= NULL
;
481 spin_lock(&journal
->j_state_lock
);
482 if (journal
->j_running_transaction
&& !current
->journal_info
) {
483 transaction
= journal
->j_running_transaction
;
484 __jbd2_log_start_commit(journal
, transaction
->t_tid
);
485 } else if (journal
->j_committing_transaction
)
486 transaction
= journal
->j_committing_transaction
;
489 spin_unlock(&journal
->j_state_lock
);
490 return 0; /* Nothing to retry */
493 tid
= transaction
->t_tid
;
494 spin_unlock(&journal
->j_state_lock
);
495 jbd2_log_wait_commit(journal
, tid
);
500 * Start a commit of the current running transaction (if any). Returns true
501 * if a transaction is going to be committed (or is currently already
502 * committing), and fills its tid in at *ptid
504 int jbd2_journal_start_commit(journal_t
*journal
, tid_t
*ptid
)
508 spin_lock(&journal
->j_state_lock
);
509 if (journal
->j_running_transaction
) {
510 tid_t tid
= journal
->j_running_transaction
->t_tid
;
512 __jbd2_log_start_commit(journal
, tid
);
513 /* There's a running transaction and we've just made sure
514 * it's commit has been scheduled. */
518 } else if (journal
->j_committing_transaction
) {
520 * If ext3_write_super() recently started a commit, then we
521 * have to wait for completion of that transaction
524 *ptid
= journal
->j_committing_transaction
->t_tid
;
527 spin_unlock(&journal
->j_state_lock
);
532 * Wait for a specified commit to complete.
533 * The caller may not hold the journal lock.
535 int jbd2_log_wait_commit(journal_t
*journal
, tid_t tid
)
539 #ifdef CONFIG_JBD2_DEBUG
540 spin_lock(&journal
->j_state_lock
);
541 if (!tid_geq(journal
->j_commit_request
, tid
)) {
543 "%s: error: j_commit_request=%d, tid=%d\n",
544 __func__
, journal
->j_commit_request
, tid
);
546 spin_unlock(&journal
->j_state_lock
);
548 spin_lock(&journal
->j_state_lock
);
549 while (tid_gt(tid
, journal
->j_commit_sequence
)) {
550 jbd_debug(1, "JBD: want %d, j_commit_sequence=%d\n",
551 tid
, journal
->j_commit_sequence
);
552 wake_up(&journal
->j_wait_commit
);
553 spin_unlock(&journal
->j_state_lock
);
554 wait_event(journal
->j_wait_done_commit
,
555 !tid_gt(tid
, journal
->j_commit_sequence
));
556 spin_lock(&journal
->j_state_lock
);
558 spin_unlock(&journal
->j_state_lock
);
560 if (unlikely(is_journal_aborted(journal
))) {
561 printk(KERN_EMERG
"journal commit I/O error\n");
568 * Log buffer allocation routines:
571 int jbd2_journal_next_log_block(journal_t
*journal
, unsigned long long *retp
)
573 unsigned long blocknr
;
575 spin_lock(&journal
->j_state_lock
);
576 J_ASSERT(journal
->j_free
> 1);
578 blocknr
= journal
->j_head
;
581 if (journal
->j_head
== journal
->j_last
)
582 journal
->j_head
= journal
->j_first
;
583 spin_unlock(&journal
->j_state_lock
);
584 return jbd2_journal_bmap(journal
, blocknr
, retp
);
588 * Conversion of logical to physical block numbers for the journal
590 * On external journals the journal blocks are identity-mapped, so
591 * this is a no-op. If needed, we can use j_blk_offset - everything is
594 int jbd2_journal_bmap(journal_t
*journal
, unsigned long blocknr
,
595 unsigned long long *retp
)
598 unsigned long long ret
;
600 if (journal
->j_inode
) {
601 ret
= bmap(journal
->j_inode
, blocknr
);
605 char b
[BDEVNAME_SIZE
];
607 printk(KERN_ALERT
"%s: journal block not found "
608 "at offset %lu on %s\n",
611 bdevname(journal
->j_dev
, b
));
613 __journal_abort_soft(journal
, err
);
616 *retp
= blocknr
; /* +journal->j_blk_offset */
622 * We play buffer_head aliasing tricks to write data/metadata blocks to
623 * the journal without copying their contents, but for journal
624 * descriptor blocks we do need to generate bona fide buffers.
626 * After the caller of jbd2_journal_get_descriptor_buffer() has finished modifying
627 * the buffer's contents they really should run flush_dcache_page(bh->b_page).
628 * But we don't bother doing that, so there will be coherency problems with
629 * mmaps of blockdevs which hold live JBD-controlled filesystems.
631 struct journal_head
*jbd2_journal_get_descriptor_buffer(journal_t
*journal
)
633 struct buffer_head
*bh
;
634 unsigned long long blocknr
;
637 err
= jbd2_journal_next_log_block(journal
, &blocknr
);
642 bh
= __getblk(journal
->j_dev
, blocknr
, journal
->j_blocksize
);
644 memset(bh
->b_data
, 0, journal
->j_blocksize
);
645 set_buffer_uptodate(bh
);
647 BUFFER_TRACE(bh
, "return this buffer");
648 return jbd2_journal_add_journal_head(bh
);
651 struct jbd2_stats_proc_session
{
653 struct transaction_stats_s
*stats
;
658 static void *jbd2_history_skip_empty(struct jbd2_stats_proc_session
*s
,
659 struct transaction_stats_s
*ts
,
662 if (ts
== s
->stats
+ s
->max
)
664 if (!first
&& ts
== s
->stats
+ s
->start
)
666 while (ts
->ts_type
== 0) {
668 if (ts
== s
->stats
+ s
->max
)
670 if (ts
== s
->stats
+ s
->start
)
677 static void *jbd2_seq_history_start(struct seq_file
*seq
, loff_t
*pos
)
679 struct jbd2_stats_proc_session
*s
= seq
->private;
680 struct transaction_stats_s
*ts
;
684 return SEQ_START_TOKEN
;
685 ts
= jbd2_history_skip_empty(s
, s
->stats
+ s
->start
, 1);
690 ts
= jbd2_history_skip_empty(s
, ++ts
, 0);
698 static void *jbd2_seq_history_next(struct seq_file
*seq
, void *v
, loff_t
*pos
)
700 struct jbd2_stats_proc_session
*s
= seq
->private;
701 struct transaction_stats_s
*ts
= v
;
704 if (v
== SEQ_START_TOKEN
)
705 return jbd2_history_skip_empty(s
, s
->stats
+ s
->start
, 1);
707 return jbd2_history_skip_empty(s
, ++ts
, 0);
710 static int jbd2_seq_history_show(struct seq_file
*seq
, void *v
)
712 struct transaction_stats_s
*ts
= v
;
713 if (v
== SEQ_START_TOKEN
) {
714 seq_printf(seq
, "%-4s %-5s %-5s %-5s %-5s %-5s %-5s %-6s %-5s "
715 "%-5s %-5s %-5s %-5s %-5s\n", "R/C", "tid",
716 "wait", "run", "lock", "flush", "log", "hndls",
717 "block", "inlog", "ctime", "write", "drop",
721 if (ts
->ts_type
== JBD2_STATS_RUN
)
722 seq_printf(seq
, "%-4s %-5lu %-5u %-5u %-5u %-5u %-5u "
723 "%-6lu %-5lu %-5lu\n", "R", ts
->ts_tid
,
724 jiffies_to_msecs(ts
->u
.run
.rs_wait
),
725 jiffies_to_msecs(ts
->u
.run
.rs_running
),
726 jiffies_to_msecs(ts
->u
.run
.rs_locked
),
727 jiffies_to_msecs(ts
->u
.run
.rs_flushing
),
728 jiffies_to_msecs(ts
->u
.run
.rs_logging
),
729 ts
->u
.run
.rs_handle_count
,
731 ts
->u
.run
.rs_blocks_logged
);
732 else if (ts
->ts_type
== JBD2_STATS_CHECKPOINT
)
733 seq_printf(seq
, "%-4s %-5lu %48s %-5u %-5lu %-5lu %-5lu\n",
734 "C", ts
->ts_tid
, " ",
735 jiffies_to_msecs(ts
->u
.chp
.cs_chp_time
),
736 ts
->u
.chp
.cs_written
, ts
->u
.chp
.cs_dropped
,
737 ts
->u
.chp
.cs_forced_to_close
);
743 static void jbd2_seq_history_stop(struct seq_file
*seq
, void *v
)
747 static struct seq_operations jbd2_seq_history_ops
= {
748 .start
= jbd2_seq_history_start
,
749 .next
= jbd2_seq_history_next
,
750 .stop
= jbd2_seq_history_stop
,
751 .show
= jbd2_seq_history_show
,
754 static int jbd2_seq_history_open(struct inode
*inode
, struct file
*file
)
756 journal_t
*journal
= PDE(inode
)->data
;
757 struct jbd2_stats_proc_session
*s
;
760 s
= kmalloc(sizeof(*s
), GFP_KERNEL
);
763 size
= sizeof(struct transaction_stats_s
) * journal
->j_history_max
;
764 s
->stats
= kmalloc(size
, GFP_KERNEL
);
765 if (s
->stats
== NULL
) {
769 spin_lock(&journal
->j_history_lock
);
770 memcpy(s
->stats
, journal
->j_history
, size
);
771 s
->max
= journal
->j_history_max
;
772 s
->start
= journal
->j_history_cur
% s
->max
;
773 spin_unlock(&journal
->j_history_lock
);
775 rc
= seq_open(file
, &jbd2_seq_history_ops
);
777 struct seq_file
*m
= file
->private_data
;
787 static int jbd2_seq_history_release(struct inode
*inode
, struct file
*file
)
789 struct seq_file
*seq
= file
->private_data
;
790 struct jbd2_stats_proc_session
*s
= seq
->private;
794 return seq_release(inode
, file
);
797 static struct file_operations jbd2_seq_history_fops
= {
798 .owner
= THIS_MODULE
,
799 .open
= jbd2_seq_history_open
,
802 .release
= jbd2_seq_history_release
,
805 static void *jbd2_seq_info_start(struct seq_file
*seq
, loff_t
*pos
)
807 return *pos
? NULL
: SEQ_START_TOKEN
;
810 static void *jbd2_seq_info_next(struct seq_file
*seq
, void *v
, loff_t
*pos
)
815 static int jbd2_seq_info_show(struct seq_file
*seq
, void *v
)
817 struct jbd2_stats_proc_session
*s
= seq
->private;
819 if (v
!= SEQ_START_TOKEN
)
821 seq_printf(seq
, "%lu transaction, each upto %u blocks\n",
823 s
->journal
->j_max_transaction_buffers
);
824 if (s
->stats
->ts_tid
== 0)
826 seq_printf(seq
, "average: \n %ums waiting for transaction\n",
827 jiffies_to_msecs(s
->stats
->u
.run
.rs_wait
/ s
->stats
->ts_tid
));
828 seq_printf(seq
, " %ums running transaction\n",
829 jiffies_to_msecs(s
->stats
->u
.run
.rs_running
/ s
->stats
->ts_tid
));
830 seq_printf(seq
, " %ums transaction was being locked\n",
831 jiffies_to_msecs(s
->stats
->u
.run
.rs_locked
/ s
->stats
->ts_tid
));
832 seq_printf(seq
, " %ums flushing data (in ordered mode)\n",
833 jiffies_to_msecs(s
->stats
->u
.run
.rs_flushing
/ s
->stats
->ts_tid
));
834 seq_printf(seq
, " %ums logging transaction\n",
835 jiffies_to_msecs(s
->stats
->u
.run
.rs_logging
/ s
->stats
->ts_tid
));
836 seq_printf(seq
, " %lu handles per transaction\n",
837 s
->stats
->u
.run
.rs_handle_count
/ s
->stats
->ts_tid
);
838 seq_printf(seq
, " %lu blocks per transaction\n",
839 s
->stats
->u
.run
.rs_blocks
/ s
->stats
->ts_tid
);
840 seq_printf(seq
, " %lu logged blocks per transaction\n",
841 s
->stats
->u
.run
.rs_blocks_logged
/ s
->stats
->ts_tid
);
845 static void jbd2_seq_info_stop(struct seq_file
*seq
, void *v
)
849 static struct seq_operations jbd2_seq_info_ops
= {
850 .start
= jbd2_seq_info_start
,
851 .next
= jbd2_seq_info_next
,
852 .stop
= jbd2_seq_info_stop
,
853 .show
= jbd2_seq_info_show
,
856 static int jbd2_seq_info_open(struct inode
*inode
, struct file
*file
)
858 journal_t
*journal
= PDE(inode
)->data
;
859 struct jbd2_stats_proc_session
*s
;
862 s
= kmalloc(sizeof(*s
), GFP_KERNEL
);
865 size
= sizeof(struct transaction_stats_s
);
866 s
->stats
= kmalloc(size
, GFP_KERNEL
);
867 if (s
->stats
== NULL
) {
871 spin_lock(&journal
->j_history_lock
);
872 memcpy(s
->stats
, &journal
->j_stats
, size
);
873 s
->journal
= journal
;
874 spin_unlock(&journal
->j_history_lock
);
876 rc
= seq_open(file
, &jbd2_seq_info_ops
);
878 struct seq_file
*m
= file
->private_data
;
888 static int jbd2_seq_info_release(struct inode
*inode
, struct file
*file
)
890 struct seq_file
*seq
= file
->private_data
;
891 struct jbd2_stats_proc_session
*s
= seq
->private;
894 return seq_release(inode
, file
);
897 static struct file_operations jbd2_seq_info_fops
= {
898 .owner
= THIS_MODULE
,
899 .open
= jbd2_seq_info_open
,
902 .release
= jbd2_seq_info_release
,
905 static struct proc_dir_entry
*proc_jbd2_stats
;
907 static void jbd2_stats_proc_init(journal_t
*journal
)
909 journal
->j_proc_entry
= proc_mkdir(journal
->j_devname
, proc_jbd2_stats
);
910 if (journal
->j_proc_entry
) {
911 proc_create_data("history", S_IRUGO
, journal
->j_proc_entry
,
912 &jbd2_seq_history_fops
, journal
);
913 proc_create_data("info", S_IRUGO
, journal
->j_proc_entry
,
914 &jbd2_seq_info_fops
, journal
);
918 static void jbd2_stats_proc_exit(journal_t
*journal
)
920 remove_proc_entry("info", journal
->j_proc_entry
);
921 remove_proc_entry("history", journal
->j_proc_entry
);
922 remove_proc_entry(journal
->j_devname
, proc_jbd2_stats
);
925 static void journal_init_stats(journal_t
*journal
)
929 if (!proc_jbd2_stats
)
932 journal
->j_history_max
= 100;
933 size
= sizeof(struct transaction_stats_s
) * journal
->j_history_max
;
934 journal
->j_history
= kzalloc(size
, GFP_KERNEL
);
935 if (!journal
->j_history
) {
936 journal
->j_history_max
= 0;
939 spin_lock_init(&journal
->j_history_lock
);
943 * Management for journal control blocks: functions to create and
944 * destroy journal_t structures, and to initialise and read existing
945 * journal blocks from disk. */
947 /* First: create and setup a journal_t object in memory. We initialise
948 * very few fields yet: that has to wait until we have created the
949 * journal structures from from scratch, or loaded them from disk. */
951 static journal_t
* journal_init_common (void)
956 journal
= kzalloc(sizeof(*journal
), GFP_KERNEL
|__GFP_NOFAIL
);
960 init_waitqueue_head(&journal
->j_wait_transaction_locked
);
961 init_waitqueue_head(&journal
->j_wait_logspace
);
962 init_waitqueue_head(&journal
->j_wait_done_commit
);
963 init_waitqueue_head(&journal
->j_wait_checkpoint
);
964 init_waitqueue_head(&journal
->j_wait_commit
);
965 init_waitqueue_head(&journal
->j_wait_updates
);
966 mutex_init(&journal
->j_barrier
);
967 mutex_init(&journal
->j_checkpoint_mutex
);
968 spin_lock_init(&journal
->j_revoke_lock
);
969 spin_lock_init(&journal
->j_list_lock
);
970 spin_lock_init(&journal
->j_state_lock
);
972 journal
->j_commit_interval
= (HZ
* JBD2_DEFAULT_MAX_COMMIT_AGE
);
974 /* The journal is marked for error until we succeed with recovery! */
975 journal
->j_flags
= JBD2_ABORT
;
977 /* Set up a default-sized revoke table for the new mount. */
978 err
= jbd2_journal_init_revoke(journal
, JOURNAL_REVOKE_DEFAULT_HASH
);
984 journal_init_stats(journal
);
991 /* jbd2_journal_init_dev and jbd2_journal_init_inode:
993 * Create a journal structure assigned some fixed set of disk blocks to
994 * the journal. We don't actually touch those disk blocks yet, but we
995 * need to set up all of the mapping information to tell the journaling
996 * system where the journal blocks are.
1001 * journal_t * jbd2_journal_init_dev() - creates and initialises a journal structure
1002 * @bdev: Block device on which to create the journal
1003 * @fs_dev: Device which hold journalled filesystem for this journal.
1004 * @start: Block nr Start of journal.
1005 * @len: Length of the journal in blocks.
1006 * @blocksize: blocksize of journalling device
1008 * Returns: a newly created journal_t *
1010 * jbd2_journal_init_dev creates a journal which maps a fixed contiguous
1011 * range of blocks on an arbitrary block device.
1014 journal_t
* jbd2_journal_init_dev(struct block_device
*bdev
,
1015 struct block_device
*fs_dev
,
1016 unsigned long long start
, int len
, int blocksize
)
1018 journal_t
*journal
= journal_init_common();
1019 struct buffer_head
*bh
;
1026 /* journal descriptor can store up to n blocks -bzzz */
1027 journal
->j_blocksize
= blocksize
;
1028 n
= journal
->j_blocksize
/ sizeof(journal_block_tag_t
);
1029 journal
->j_wbufsize
= n
;
1030 journal
->j_wbuf
= kmalloc(n
* sizeof(struct buffer_head
*), GFP_KERNEL
);
1031 if (!journal
->j_wbuf
) {
1032 printk(KERN_ERR
"%s: Cant allocate bhs for commit thread\n",
1038 journal
->j_dev
= bdev
;
1039 journal
->j_fs_dev
= fs_dev
;
1040 journal
->j_blk_offset
= start
;
1041 journal
->j_maxlen
= len
;
1042 bdevname(journal
->j_dev
, journal
->j_devname
);
1043 p
= journal
->j_devname
;
1044 while ((p
= strchr(p
, '/')))
1046 jbd2_stats_proc_init(journal
);
1048 bh
= __getblk(journal
->j_dev
, start
, journal
->j_blocksize
);
1049 J_ASSERT(bh
!= NULL
);
1050 journal
->j_sb_buffer
= bh
;
1051 journal
->j_superblock
= (journal_superblock_t
*)bh
->b_data
;
1057 * journal_t * jbd2_journal_init_inode () - creates a journal which maps to a inode.
1058 * @inode: An inode to create the journal in
1060 * jbd2_journal_init_inode creates a journal which maps an on-disk inode as
1061 * the journal. The inode must exist already, must support bmap() and
1062 * must have all data blocks preallocated.
1064 journal_t
* jbd2_journal_init_inode (struct inode
*inode
)
1066 struct buffer_head
*bh
;
1067 journal_t
*journal
= journal_init_common();
1071 unsigned long long blocknr
;
1076 journal
->j_dev
= journal
->j_fs_dev
= inode
->i_sb
->s_bdev
;
1077 journal
->j_inode
= inode
;
1078 bdevname(journal
->j_dev
, journal
->j_devname
);
1079 p
= journal
->j_devname
;
1080 while ((p
= strchr(p
, '/')))
1082 p
= journal
->j_devname
+ strlen(journal
->j_devname
);
1083 sprintf(p
, ":%lu", journal
->j_inode
->i_ino
);
1085 "journal %p: inode %s/%ld, size %Ld, bits %d, blksize %ld\n",
1086 journal
, inode
->i_sb
->s_id
, inode
->i_ino
,
1087 (long long) inode
->i_size
,
1088 inode
->i_sb
->s_blocksize_bits
, inode
->i_sb
->s_blocksize
);
1090 journal
->j_maxlen
= inode
->i_size
>> inode
->i_sb
->s_blocksize_bits
;
1091 journal
->j_blocksize
= inode
->i_sb
->s_blocksize
;
1092 jbd2_stats_proc_init(journal
);
1094 /* journal descriptor can store up to n blocks -bzzz */
1095 n
= journal
->j_blocksize
/ sizeof(journal_block_tag_t
);
1096 journal
->j_wbufsize
= n
;
1097 journal
->j_wbuf
= kmalloc(n
* sizeof(struct buffer_head
*), GFP_KERNEL
);
1098 if (!journal
->j_wbuf
) {
1099 printk(KERN_ERR
"%s: Cant allocate bhs for commit thread\n",
1105 err
= jbd2_journal_bmap(journal
, 0, &blocknr
);
1106 /* If that failed, give up */
1108 printk(KERN_ERR
"%s: Cannnot locate journal superblock\n",
1114 bh
= __getblk(journal
->j_dev
, blocknr
, journal
->j_blocksize
);
1115 J_ASSERT(bh
!= NULL
);
1116 journal
->j_sb_buffer
= bh
;
1117 journal
->j_superblock
= (journal_superblock_t
*)bh
->b_data
;
1123 * If the journal init or create aborts, we need to mark the journal
1124 * superblock as being NULL to prevent the journal destroy from writing
1125 * back a bogus superblock.
1127 static void journal_fail_superblock (journal_t
*journal
)
1129 struct buffer_head
*bh
= journal
->j_sb_buffer
;
1131 journal
->j_sb_buffer
= NULL
;
1135 * Given a journal_t structure, initialise the various fields for
1136 * startup of a new journaling session. We use this both when creating
1137 * a journal, and after recovering an old journal to reset it for
1141 static int journal_reset(journal_t
*journal
)
1143 journal_superblock_t
*sb
= journal
->j_superblock
;
1144 unsigned long long first
, last
;
1146 first
= be32_to_cpu(sb
->s_first
);
1147 last
= be32_to_cpu(sb
->s_maxlen
);
1149 journal
->j_first
= first
;
1150 journal
->j_last
= last
;
1152 journal
->j_head
= first
;
1153 journal
->j_tail
= first
;
1154 journal
->j_free
= last
- first
;
1156 journal
->j_tail_sequence
= journal
->j_transaction_sequence
;
1157 journal
->j_commit_sequence
= journal
->j_transaction_sequence
- 1;
1158 journal
->j_commit_request
= journal
->j_commit_sequence
;
1160 journal
->j_max_transaction_buffers
= journal
->j_maxlen
/ 4;
1162 /* Add the dynamic fields and write it to disk. */
1163 jbd2_journal_update_superblock(journal
, 1);
1164 return jbd2_journal_start_thread(journal
);
1168 * int jbd2_journal_create() - Initialise the new journal file
1169 * @journal: Journal to create. This structure must have been initialised
1171 * Given a journal_t structure which tells us which disk blocks we can
1172 * use, create a new journal superblock and initialise all of the
1173 * journal fields from scratch.
1175 int jbd2_journal_create(journal_t
*journal
)
1177 unsigned long long blocknr
;
1178 struct buffer_head
*bh
;
1179 journal_superblock_t
*sb
;
1182 if (journal
->j_maxlen
< JBD2_MIN_JOURNAL_BLOCKS
) {
1183 printk (KERN_ERR
"Journal length (%d blocks) too short.\n",
1185 journal_fail_superblock(journal
);
1189 if (journal
->j_inode
== NULL
) {
1191 * We don't know what block to start at!
1194 "%s: creation of journal on external device!\n",
1199 /* Zero out the entire journal on disk. We cannot afford to
1200 have any blocks on disk beginning with JBD2_MAGIC_NUMBER. */
1201 jbd_debug(1, "JBD: Zeroing out journal blocks...\n");
1202 for (i
= 0; i
< journal
->j_maxlen
; i
++) {
1203 err
= jbd2_journal_bmap(journal
, i
, &blocknr
);
1206 bh
= __getblk(journal
->j_dev
, blocknr
, journal
->j_blocksize
);
1208 memset (bh
->b_data
, 0, journal
->j_blocksize
);
1209 BUFFER_TRACE(bh
, "marking dirty");
1210 mark_buffer_dirty(bh
);
1211 BUFFER_TRACE(bh
, "marking uptodate");
1212 set_buffer_uptodate(bh
);
1217 sync_blockdev(journal
->j_dev
);
1218 jbd_debug(1, "JBD: journal cleared.\n");
1220 /* OK, fill in the initial static fields in the new superblock */
1221 sb
= journal
->j_superblock
;
1223 sb
->s_header
.h_magic
= cpu_to_be32(JBD2_MAGIC_NUMBER
);
1224 sb
->s_header
.h_blocktype
= cpu_to_be32(JBD2_SUPERBLOCK_V2
);
1226 sb
->s_blocksize
= cpu_to_be32(journal
->j_blocksize
);
1227 sb
->s_maxlen
= cpu_to_be32(journal
->j_maxlen
);
1228 sb
->s_first
= cpu_to_be32(1);
1230 journal
->j_transaction_sequence
= 1;
1232 journal
->j_flags
&= ~JBD2_ABORT
;
1233 journal
->j_format_version
= 2;
1235 return journal_reset(journal
);
1239 * void jbd2_journal_update_superblock() - Update journal sb on disk.
1240 * @journal: The journal to update.
1241 * @wait: Set to '0' if you don't want to wait for IO completion.
1243 * Update a journal's dynamic superblock fields and write it to disk,
1244 * optionally waiting for the IO to complete.
1246 void jbd2_journal_update_superblock(journal_t
*journal
, int wait
)
1248 journal_superblock_t
*sb
= journal
->j_superblock
;
1249 struct buffer_head
*bh
= journal
->j_sb_buffer
;
1252 * As a special case, if the on-disk copy is already marked as needing
1253 * no recovery (s_start == 0) and there are no outstanding transactions
1254 * in the filesystem, then we can safely defer the superblock update
1255 * until the next commit by setting JBD2_FLUSHED. This avoids
1256 * attempting a write to a potential-readonly device.
1258 if (sb
->s_start
== 0 && journal
->j_tail_sequence
==
1259 journal
->j_transaction_sequence
) {
1260 jbd_debug(1,"JBD: Skipping superblock update on recovered sb "
1261 "(start %ld, seq %d, errno %d)\n",
1262 journal
->j_tail
, journal
->j_tail_sequence
,
1267 if (buffer_write_io_error(bh
)) {
1269 * Oh, dear. A previous attempt to write the journal
1270 * superblock failed. This could happen because the
1271 * USB device was yanked out. Or it could happen to
1272 * be a transient write error and maybe the block will
1273 * be remapped. Nothing we can do but to retry the
1274 * write and hope for the best.
1276 printk(KERN_ERR
"JBD2: previous I/O error detected "
1277 "for journal superblock update for %s.\n",
1278 journal
->j_devname
);
1279 clear_buffer_write_io_error(bh
);
1280 set_buffer_uptodate(bh
);
1283 spin_lock(&journal
->j_state_lock
);
1284 jbd_debug(1,"JBD: updating superblock (start %ld, seq %d, errno %d)\n",
1285 journal
->j_tail
, journal
->j_tail_sequence
, journal
->j_errno
);
1287 sb
->s_sequence
= cpu_to_be32(journal
->j_tail_sequence
);
1288 sb
->s_start
= cpu_to_be32(journal
->j_tail
);
1289 sb
->s_errno
= cpu_to_be32(journal
->j_errno
);
1290 spin_unlock(&journal
->j_state_lock
);
1292 BUFFER_TRACE(bh
, "marking dirty");
1293 mark_buffer_dirty(bh
);
1295 sync_dirty_buffer(bh
);
1296 if (buffer_write_io_error(bh
)) {
1297 printk(KERN_ERR
"JBD2: I/O error detected "
1298 "when updating journal superblock for %s.\n",
1299 journal
->j_devname
);
1300 clear_buffer_write_io_error(bh
);
1301 set_buffer_uptodate(bh
);
1304 ll_rw_block(SWRITE
, 1, &bh
);
1307 /* If we have just flushed the log (by marking s_start==0), then
1308 * any future commit will have to be careful to update the
1309 * superblock again to re-record the true start of the log. */
1311 spin_lock(&journal
->j_state_lock
);
1313 journal
->j_flags
&= ~JBD2_FLUSHED
;
1315 journal
->j_flags
|= JBD2_FLUSHED
;
1316 spin_unlock(&journal
->j_state_lock
);
1320 * Read the superblock for a given journal, performing initial
1321 * validation of the format.
1324 static int journal_get_superblock(journal_t
*journal
)
1326 struct buffer_head
*bh
;
1327 journal_superblock_t
*sb
;
1330 bh
= journal
->j_sb_buffer
;
1332 J_ASSERT(bh
!= NULL
);
1333 if (!buffer_uptodate(bh
)) {
1334 ll_rw_block(READ
, 1, &bh
);
1336 if (!buffer_uptodate(bh
)) {
1338 "JBD: IO error reading journal superblock\n");
1343 sb
= journal
->j_superblock
;
1347 if (sb
->s_header
.h_magic
!= cpu_to_be32(JBD2_MAGIC_NUMBER
) ||
1348 sb
->s_blocksize
!= cpu_to_be32(journal
->j_blocksize
)) {
1349 printk(KERN_WARNING
"JBD: no valid journal superblock found\n");
1353 switch(be32_to_cpu(sb
->s_header
.h_blocktype
)) {
1354 case JBD2_SUPERBLOCK_V1
:
1355 journal
->j_format_version
= 1;
1357 case JBD2_SUPERBLOCK_V2
:
1358 journal
->j_format_version
= 2;
1361 printk(KERN_WARNING
"JBD: unrecognised superblock format ID\n");
1365 if (be32_to_cpu(sb
->s_maxlen
) < journal
->j_maxlen
)
1366 journal
->j_maxlen
= be32_to_cpu(sb
->s_maxlen
);
1367 else if (be32_to_cpu(sb
->s_maxlen
) > journal
->j_maxlen
) {
1368 printk (KERN_WARNING
"JBD: journal file too short\n");
1375 journal_fail_superblock(journal
);
1380 * Load the on-disk journal superblock and read the key fields into the
1384 static int load_superblock(journal_t
*journal
)
1387 journal_superblock_t
*sb
;
1389 err
= journal_get_superblock(journal
);
1393 sb
= journal
->j_superblock
;
1395 journal
->j_tail_sequence
= be32_to_cpu(sb
->s_sequence
);
1396 journal
->j_tail
= be32_to_cpu(sb
->s_start
);
1397 journal
->j_first
= be32_to_cpu(sb
->s_first
);
1398 journal
->j_last
= be32_to_cpu(sb
->s_maxlen
);
1399 journal
->j_errno
= be32_to_cpu(sb
->s_errno
);
1406 * int jbd2_journal_load() - Read journal from disk.
1407 * @journal: Journal to act on.
1409 * Given a journal_t structure which tells us which disk blocks contain
1410 * a journal, read the journal from disk to initialise the in-memory
1413 int jbd2_journal_load(journal_t
*journal
)
1416 journal_superblock_t
*sb
;
1418 err
= load_superblock(journal
);
1422 sb
= journal
->j_superblock
;
1423 /* If this is a V2 superblock, then we have to check the
1424 * features flags on it. */
1426 if (journal
->j_format_version
>= 2) {
1427 if ((sb
->s_feature_ro_compat
&
1428 ~cpu_to_be32(JBD2_KNOWN_ROCOMPAT_FEATURES
)) ||
1429 (sb
->s_feature_incompat
&
1430 ~cpu_to_be32(JBD2_KNOWN_INCOMPAT_FEATURES
))) {
1431 printk (KERN_WARNING
1432 "JBD: Unrecognised features on journal\n");
1437 /* Let the recovery code check whether it needs to recover any
1438 * data from the journal. */
1439 if (jbd2_journal_recover(journal
))
1440 goto recovery_error
;
1442 /* OK, we've finished with the dynamic journal bits:
1443 * reinitialise the dynamic contents of the superblock in memory
1444 * and reset them on disk. */
1445 if (journal_reset(journal
))
1446 goto recovery_error
;
1448 journal
->j_flags
&= ~JBD2_ABORT
;
1449 journal
->j_flags
|= JBD2_LOADED
;
1453 printk (KERN_WARNING
"JBD: recovery failed\n");
1458 * void jbd2_journal_destroy() - Release a journal_t structure.
1459 * @journal: Journal to act on.
1461 * Release a journal_t structure once it is no longer in use by the
1464 void jbd2_journal_destroy(journal_t
*journal
)
1466 /* Wait for the commit thread to wake up and die. */
1467 journal_kill_thread(journal
);
1469 /* Force a final log commit */
1470 if (journal
->j_running_transaction
)
1471 jbd2_journal_commit_transaction(journal
);
1473 /* Force any old transactions to disk */
1475 /* Totally anal locking here... */
1476 spin_lock(&journal
->j_list_lock
);
1477 while (journal
->j_checkpoint_transactions
!= NULL
) {
1478 spin_unlock(&journal
->j_list_lock
);
1479 jbd2_log_do_checkpoint(journal
);
1480 spin_lock(&journal
->j_list_lock
);
1483 J_ASSERT(journal
->j_running_transaction
== NULL
);
1484 J_ASSERT(journal
->j_committing_transaction
== NULL
);
1485 J_ASSERT(journal
->j_checkpoint_transactions
== NULL
);
1486 spin_unlock(&journal
->j_list_lock
);
1488 /* We can now mark the journal as empty. */
1489 journal
->j_tail
= 0;
1490 journal
->j_tail_sequence
= ++journal
->j_transaction_sequence
;
1491 if (journal
->j_sb_buffer
) {
1492 jbd2_journal_update_superblock(journal
, 1);
1493 brelse(journal
->j_sb_buffer
);
1496 if (journal
->j_proc_entry
)
1497 jbd2_stats_proc_exit(journal
);
1498 if (journal
->j_inode
)
1499 iput(journal
->j_inode
);
1500 if (journal
->j_revoke
)
1501 jbd2_journal_destroy_revoke(journal
);
1502 kfree(journal
->j_wbuf
);
1508 *int jbd2_journal_check_used_features () - Check if features specified are used.
1509 * @journal: Journal to check.
1510 * @compat: bitmask of compatible features
1511 * @ro: bitmask of features that force read-only mount
1512 * @incompat: bitmask of incompatible features
1514 * Check whether the journal uses all of a given set of
1515 * features. Return true (non-zero) if it does.
1518 int jbd2_journal_check_used_features (journal_t
*journal
, unsigned long compat
,
1519 unsigned long ro
, unsigned long incompat
)
1521 journal_superblock_t
*sb
;
1523 if (!compat
&& !ro
&& !incompat
)
1525 if (journal
->j_format_version
== 1)
1528 sb
= journal
->j_superblock
;
1530 if (((be32_to_cpu(sb
->s_feature_compat
) & compat
) == compat
) &&
1531 ((be32_to_cpu(sb
->s_feature_ro_compat
) & ro
) == ro
) &&
1532 ((be32_to_cpu(sb
->s_feature_incompat
) & incompat
) == incompat
))
1539 * int jbd2_journal_check_available_features() - Check feature set in journalling layer
1540 * @journal: Journal to check.
1541 * @compat: bitmask of compatible features
1542 * @ro: bitmask of features that force read-only mount
1543 * @incompat: bitmask of incompatible features
1545 * Check whether the journaling code supports the use of
1546 * all of a given set of features on this journal. Return true
1547 * (non-zero) if it can. */
1549 int jbd2_journal_check_available_features (journal_t
*journal
, unsigned long compat
,
1550 unsigned long ro
, unsigned long incompat
)
1552 journal_superblock_t
*sb
;
1554 if (!compat
&& !ro
&& !incompat
)
1557 sb
= journal
->j_superblock
;
1559 /* We can support any known requested features iff the
1560 * superblock is in version 2. Otherwise we fail to support any
1561 * extended sb features. */
1563 if (journal
->j_format_version
!= 2)
1566 if ((compat
& JBD2_KNOWN_COMPAT_FEATURES
) == compat
&&
1567 (ro
& JBD2_KNOWN_ROCOMPAT_FEATURES
) == ro
&&
1568 (incompat
& JBD2_KNOWN_INCOMPAT_FEATURES
) == incompat
)
1575 * int jbd2_journal_set_features () - Mark a given journal feature in the superblock
1576 * @journal: Journal to act on.
1577 * @compat: bitmask of compatible features
1578 * @ro: bitmask of features that force read-only mount
1579 * @incompat: bitmask of incompatible features
1581 * Mark a given journal feature as present on the
1582 * superblock. Returns true if the requested features could be set.
1586 int jbd2_journal_set_features (journal_t
*journal
, unsigned long compat
,
1587 unsigned long ro
, unsigned long incompat
)
1589 journal_superblock_t
*sb
;
1591 if (jbd2_journal_check_used_features(journal
, compat
, ro
, incompat
))
1594 if (!jbd2_journal_check_available_features(journal
, compat
, ro
, incompat
))
1597 jbd_debug(1, "Setting new features 0x%lx/0x%lx/0x%lx\n",
1598 compat
, ro
, incompat
);
1600 sb
= journal
->j_superblock
;
1602 sb
->s_feature_compat
|= cpu_to_be32(compat
);
1603 sb
->s_feature_ro_compat
|= cpu_to_be32(ro
);
1604 sb
->s_feature_incompat
|= cpu_to_be32(incompat
);
1610 * jbd2_journal_clear_features () - Clear a given journal feature in the
1612 * @journal: Journal to act on.
1613 * @compat: bitmask of compatible features
1614 * @ro: bitmask of features that force read-only mount
1615 * @incompat: bitmask of incompatible features
1617 * Clear a given journal feature as present on the
1620 void jbd2_journal_clear_features(journal_t
*journal
, unsigned long compat
,
1621 unsigned long ro
, unsigned long incompat
)
1623 journal_superblock_t
*sb
;
1625 jbd_debug(1, "Clear features 0x%lx/0x%lx/0x%lx\n",
1626 compat
, ro
, incompat
);
1628 sb
= journal
->j_superblock
;
1630 sb
->s_feature_compat
&= ~cpu_to_be32(compat
);
1631 sb
->s_feature_ro_compat
&= ~cpu_to_be32(ro
);
1632 sb
->s_feature_incompat
&= ~cpu_to_be32(incompat
);
1634 EXPORT_SYMBOL(jbd2_journal_clear_features
);
1637 * int jbd2_journal_update_format () - Update on-disk journal structure.
1638 * @journal: Journal to act on.
1640 * Given an initialised but unloaded journal struct, poke about in the
1641 * on-disk structure to update it to the most recent supported version.
1643 int jbd2_journal_update_format (journal_t
*journal
)
1645 journal_superblock_t
*sb
;
1648 err
= journal_get_superblock(journal
);
1652 sb
= journal
->j_superblock
;
1654 switch (be32_to_cpu(sb
->s_header
.h_blocktype
)) {
1655 case JBD2_SUPERBLOCK_V2
:
1657 case JBD2_SUPERBLOCK_V1
:
1658 return journal_convert_superblock_v1(journal
, sb
);
1665 static int journal_convert_superblock_v1(journal_t
*journal
,
1666 journal_superblock_t
*sb
)
1668 int offset
, blocksize
;
1669 struct buffer_head
*bh
;
1672 "JBD: Converting superblock from version 1 to 2.\n");
1674 /* Pre-initialise new fields to zero */
1675 offset
= ((char *) &(sb
->s_feature_compat
)) - ((char *) sb
);
1676 blocksize
= be32_to_cpu(sb
->s_blocksize
);
1677 memset(&sb
->s_feature_compat
, 0, blocksize
-offset
);
1679 sb
->s_nr_users
= cpu_to_be32(1);
1680 sb
->s_header
.h_blocktype
= cpu_to_be32(JBD2_SUPERBLOCK_V2
);
1681 journal
->j_format_version
= 2;
1683 bh
= journal
->j_sb_buffer
;
1684 BUFFER_TRACE(bh
, "marking dirty");
1685 mark_buffer_dirty(bh
);
1686 sync_dirty_buffer(bh
);
1692 * int jbd2_journal_flush () - Flush journal
1693 * @journal: Journal to act on.
1695 * Flush all data for a given journal to disk and empty the journal.
1696 * Filesystems can use this when remounting readonly to ensure that
1697 * recovery does not need to happen on remount.
1700 int jbd2_journal_flush(journal_t
*journal
)
1703 transaction_t
*transaction
= NULL
;
1704 unsigned long old_tail
;
1706 spin_lock(&journal
->j_state_lock
);
1708 /* Force everything buffered to the log... */
1709 if (journal
->j_running_transaction
) {
1710 transaction
= journal
->j_running_transaction
;
1711 __jbd2_log_start_commit(journal
, transaction
->t_tid
);
1712 } else if (journal
->j_committing_transaction
)
1713 transaction
= journal
->j_committing_transaction
;
1715 /* Wait for the log commit to complete... */
1717 tid_t tid
= transaction
->t_tid
;
1719 spin_unlock(&journal
->j_state_lock
);
1720 jbd2_log_wait_commit(journal
, tid
);
1722 spin_unlock(&journal
->j_state_lock
);
1725 /* ...and flush everything in the log out to disk. */
1726 spin_lock(&journal
->j_list_lock
);
1727 while (!err
&& journal
->j_checkpoint_transactions
!= NULL
) {
1728 spin_unlock(&journal
->j_list_lock
);
1729 err
= jbd2_log_do_checkpoint(journal
);
1730 spin_lock(&journal
->j_list_lock
);
1732 spin_unlock(&journal
->j_list_lock
);
1733 jbd2_cleanup_journal_tail(journal
);
1735 /* Finally, mark the journal as really needing no recovery.
1736 * This sets s_start==0 in the underlying superblock, which is
1737 * the magic code for a fully-recovered superblock. Any future
1738 * commits of data to the journal will restore the current
1740 spin_lock(&journal
->j_state_lock
);
1741 old_tail
= journal
->j_tail
;
1742 journal
->j_tail
= 0;
1743 spin_unlock(&journal
->j_state_lock
);
1744 jbd2_journal_update_superblock(journal
, 1);
1745 spin_lock(&journal
->j_state_lock
);
1746 journal
->j_tail
= old_tail
;
1748 J_ASSERT(!journal
->j_running_transaction
);
1749 J_ASSERT(!journal
->j_committing_transaction
);
1750 J_ASSERT(!journal
->j_checkpoint_transactions
);
1751 J_ASSERT(journal
->j_head
== journal
->j_tail
);
1752 J_ASSERT(journal
->j_tail_sequence
== journal
->j_transaction_sequence
);
1753 spin_unlock(&journal
->j_state_lock
);
1758 * int jbd2_journal_wipe() - Wipe journal contents
1759 * @journal: Journal to act on.
1760 * @write: flag (see below)
1762 * Wipe out all of the contents of a journal, safely. This will produce
1763 * a warning if the journal contains any valid recovery information.
1764 * Must be called between journal_init_*() and jbd2_journal_load().
1766 * If 'write' is non-zero, then we wipe out the journal on disk; otherwise
1767 * we merely suppress recovery.
1770 int jbd2_journal_wipe(journal_t
*journal
, int write
)
1772 journal_superblock_t
*sb
;
1775 J_ASSERT (!(journal
->j_flags
& JBD2_LOADED
));
1777 err
= load_superblock(journal
);
1781 sb
= journal
->j_superblock
;
1783 if (!journal
->j_tail
)
1786 printk (KERN_WARNING
"JBD: %s recovery information on journal\n",
1787 write
? "Clearing" : "Ignoring");
1789 err
= jbd2_journal_skip_recovery(journal
);
1791 jbd2_journal_update_superblock(journal
, 1);
1798 * journal_dev_name: format a character string to describe on what
1799 * device this journal is present.
1802 static const char *journal_dev_name(journal_t
*journal
, char *buffer
)
1804 struct block_device
*bdev
;
1806 if (journal
->j_inode
)
1807 bdev
= journal
->j_inode
->i_sb
->s_bdev
;
1809 bdev
= journal
->j_dev
;
1811 return bdevname(bdev
, buffer
);
1815 * Journal abort has very specific semantics, which we describe
1816 * for journal abort.
1818 * Two internal function, which provide abort to te jbd layer
1823 * Quick version for internal journal use (doesn't lock the journal).
1824 * Aborts hard --- we mark the abort as occurred, but do _nothing_ else,
1825 * and don't attempt to make any other journal updates.
1827 void __jbd2_journal_abort_hard(journal_t
*journal
)
1829 transaction_t
*transaction
;
1830 char b
[BDEVNAME_SIZE
];
1832 if (journal
->j_flags
& JBD2_ABORT
)
1835 printk(KERN_ERR
"Aborting journal on device %s.\n",
1836 journal_dev_name(journal
, b
));
1838 spin_lock(&journal
->j_state_lock
);
1839 journal
->j_flags
|= JBD2_ABORT
;
1840 transaction
= journal
->j_running_transaction
;
1842 __jbd2_log_start_commit(journal
, transaction
->t_tid
);
1843 spin_unlock(&journal
->j_state_lock
);
1846 /* Soft abort: record the abort error status in the journal superblock,
1847 * but don't do any other IO. */
1848 static void __journal_abort_soft (journal_t
*journal
, int errno
)
1850 if (journal
->j_flags
& JBD2_ABORT
)
1853 if (!journal
->j_errno
)
1854 journal
->j_errno
= errno
;
1856 __jbd2_journal_abort_hard(journal
);
1859 jbd2_journal_update_superblock(journal
, 1);
1863 * void jbd2_journal_abort () - Shutdown the journal immediately.
1864 * @journal: the journal to shutdown.
1865 * @errno: an error number to record in the journal indicating
1866 * the reason for the shutdown.
1868 * Perform a complete, immediate shutdown of the ENTIRE
1869 * journal (not of a single transaction). This operation cannot be
1870 * undone without closing and reopening the journal.
1872 * The jbd2_journal_abort function is intended to support higher level error
1873 * recovery mechanisms such as the ext2/ext3 remount-readonly error
1876 * Journal abort has very specific semantics. Any existing dirty,
1877 * unjournaled buffers in the main filesystem will still be written to
1878 * disk by bdflush, but the journaling mechanism will be suspended
1879 * immediately and no further transaction commits will be honoured.
1881 * Any dirty, journaled buffers will be written back to disk without
1882 * hitting the journal. Atomicity cannot be guaranteed on an aborted
1883 * filesystem, but we _do_ attempt to leave as much data as possible
1884 * behind for fsck to use for cleanup.
1886 * Any attempt to get a new transaction handle on a journal which is in
1887 * ABORT state will just result in an -EROFS error return. A
1888 * jbd2_journal_stop on an existing handle will return -EIO if we have
1889 * entered abort state during the update.
1891 * Recursive transactions are not disturbed by journal abort until the
1892 * final jbd2_journal_stop, which will receive the -EIO error.
1894 * Finally, the jbd2_journal_abort call allows the caller to supply an errno
1895 * which will be recorded (if possible) in the journal superblock. This
1896 * allows a client to record failure conditions in the middle of a
1897 * transaction without having to complete the transaction to record the
1898 * failure to disk. ext3_error, for example, now uses this
1901 * Errors which originate from within the journaling layer will NOT
1902 * supply an errno; a null errno implies that absolutely no further
1903 * writes are done to the journal (unless there are any already in
1908 void jbd2_journal_abort(journal_t
*journal
, int errno
)
1910 __journal_abort_soft(journal
, errno
);
1914 * int jbd2_journal_errno () - returns the journal's error state.
1915 * @journal: journal to examine.
1917 * This is the errno numbet set with jbd2_journal_abort(), the last
1918 * time the journal was mounted - if the journal was stopped
1919 * without calling abort this will be 0.
1921 * If the journal has been aborted on this mount time -EROFS will
1924 int jbd2_journal_errno(journal_t
*journal
)
1928 spin_lock(&journal
->j_state_lock
);
1929 if (journal
->j_flags
& JBD2_ABORT
)
1932 err
= journal
->j_errno
;
1933 spin_unlock(&journal
->j_state_lock
);
1938 * int jbd2_journal_clear_err () - clears the journal's error state
1939 * @journal: journal to act on.
1941 * An error must be cleared or Acked to take a FS out of readonly
1944 int jbd2_journal_clear_err(journal_t
*journal
)
1948 spin_lock(&journal
->j_state_lock
);
1949 if (journal
->j_flags
& JBD2_ABORT
)
1952 journal
->j_errno
= 0;
1953 spin_unlock(&journal
->j_state_lock
);
1958 * void jbd2_journal_ack_err() - Ack journal err.
1959 * @journal: journal to act on.
1961 * An error must be cleared or Acked to take a FS out of readonly
1964 void jbd2_journal_ack_err(journal_t
*journal
)
1966 spin_lock(&journal
->j_state_lock
);
1967 if (journal
->j_errno
)
1968 journal
->j_flags
|= JBD2_ACK_ERR
;
1969 spin_unlock(&journal
->j_state_lock
);
1972 int jbd2_journal_blocks_per_page(struct inode
*inode
)
1974 return 1 << (PAGE_CACHE_SHIFT
- inode
->i_sb
->s_blocksize_bits
);
1978 * helper functions to deal with 32 or 64bit block numbers.
1980 size_t journal_tag_bytes(journal_t
*journal
)
1982 if (JBD2_HAS_INCOMPAT_FEATURE(journal
, JBD2_FEATURE_INCOMPAT_64BIT
))
1983 return JBD2_TAG_SIZE64
;
1985 return JBD2_TAG_SIZE32
;
1989 * Journal_head storage management
1991 static struct kmem_cache
*jbd2_journal_head_cache
;
1992 #ifdef CONFIG_JBD2_DEBUG
1993 static atomic_t nr_journal_heads
= ATOMIC_INIT(0);
1996 static int journal_init_jbd2_journal_head_cache(void)
2000 J_ASSERT(jbd2_journal_head_cache
== NULL
);
2001 jbd2_journal_head_cache
= kmem_cache_create("jbd2_journal_head",
2002 sizeof(struct journal_head
),
2004 SLAB_TEMPORARY
, /* flags */
2007 if (!jbd2_journal_head_cache
) {
2009 printk(KERN_EMERG
"JBD: no memory for journal_head cache\n");
2014 static void jbd2_journal_destroy_jbd2_journal_head_cache(void)
2016 if (jbd2_journal_head_cache
) {
2017 kmem_cache_destroy(jbd2_journal_head_cache
);
2018 jbd2_journal_head_cache
= NULL
;
2023 * journal_head splicing and dicing
2025 static struct journal_head
*journal_alloc_journal_head(void)
2027 struct journal_head
*ret
;
2028 static unsigned long last_warning
;
2030 #ifdef CONFIG_JBD2_DEBUG
2031 atomic_inc(&nr_journal_heads
);
2033 ret
= kmem_cache_alloc(jbd2_journal_head_cache
, GFP_NOFS
);
2035 jbd_debug(1, "out of memory for journal_head\n");
2036 if (time_after(jiffies
, last_warning
+ 5*HZ
)) {
2037 printk(KERN_NOTICE
"ENOMEM in %s, retrying.\n",
2039 last_warning
= jiffies
;
2043 ret
= kmem_cache_alloc(jbd2_journal_head_cache
, GFP_NOFS
);
2049 static void journal_free_journal_head(struct journal_head
*jh
)
2051 #ifdef CONFIG_JBD2_DEBUG
2052 atomic_dec(&nr_journal_heads
);
2053 memset(jh
, JBD2_POISON_FREE
, sizeof(*jh
));
2055 kmem_cache_free(jbd2_journal_head_cache
, jh
);
2059 * A journal_head is attached to a buffer_head whenever JBD has an
2060 * interest in the buffer.
2062 * Whenever a buffer has an attached journal_head, its ->b_state:BH_JBD bit
2063 * is set. This bit is tested in core kernel code where we need to take
2064 * JBD-specific actions. Testing the zeroness of ->b_private is not reliable
2067 * When a buffer has its BH_JBD bit set, its ->b_count is elevated by one.
2069 * When a buffer has its BH_JBD bit set it is immune from being released by
2070 * core kernel code, mainly via ->b_count.
2072 * A journal_head may be detached from its buffer_head when the journal_head's
2073 * b_transaction, b_cp_transaction and b_next_transaction pointers are NULL.
2074 * Various places in JBD call jbd2_journal_remove_journal_head() to indicate that the
2075 * journal_head can be dropped if needed.
2077 * Various places in the kernel want to attach a journal_head to a buffer_head
2078 * _before_ attaching the journal_head to a transaction. To protect the
2079 * journal_head in this situation, jbd2_journal_add_journal_head elevates the
2080 * journal_head's b_jcount refcount by one. The caller must call
2081 * jbd2_journal_put_journal_head() to undo this.
2083 * So the typical usage would be:
2085 * (Attach a journal_head if needed. Increments b_jcount)
2086 * struct journal_head *jh = jbd2_journal_add_journal_head(bh);
2088 * jh->b_transaction = xxx;
2089 * jbd2_journal_put_journal_head(jh);
2091 * Now, the journal_head's b_jcount is zero, but it is safe from being released
2092 * because it has a non-zero b_transaction.
2096 * Give a buffer_head a journal_head.
2098 * Doesn't need the journal lock.
2101 struct journal_head
*jbd2_journal_add_journal_head(struct buffer_head
*bh
)
2103 struct journal_head
*jh
;
2104 struct journal_head
*new_jh
= NULL
;
2107 if (!buffer_jbd(bh
)) {
2108 new_jh
= journal_alloc_journal_head();
2109 memset(new_jh
, 0, sizeof(*new_jh
));
2112 jbd_lock_bh_journal_head(bh
);
2113 if (buffer_jbd(bh
)) {
2117 (atomic_read(&bh
->b_count
) > 0) ||
2118 (bh
->b_page
&& bh
->b_page
->mapping
));
2121 jbd_unlock_bh_journal_head(bh
);
2126 new_jh
= NULL
; /* We consumed it */
2131 BUFFER_TRACE(bh
, "added journal_head");
2134 jbd_unlock_bh_journal_head(bh
);
2136 journal_free_journal_head(new_jh
);
2137 return bh
->b_private
;
2141 * Grab a ref against this buffer_head's journal_head. If it ended up not
2142 * having a journal_head, return NULL
2144 struct journal_head
*jbd2_journal_grab_journal_head(struct buffer_head
*bh
)
2146 struct journal_head
*jh
= NULL
;
2148 jbd_lock_bh_journal_head(bh
);
2149 if (buffer_jbd(bh
)) {
2153 jbd_unlock_bh_journal_head(bh
);
2157 static void __journal_remove_journal_head(struct buffer_head
*bh
)
2159 struct journal_head
*jh
= bh2jh(bh
);
2161 J_ASSERT_JH(jh
, jh
->b_jcount
>= 0);
2164 if (jh
->b_jcount
== 0) {
2165 if (jh
->b_transaction
== NULL
&&
2166 jh
->b_next_transaction
== NULL
&&
2167 jh
->b_cp_transaction
== NULL
) {
2168 J_ASSERT_JH(jh
, jh
->b_jlist
== BJ_None
);
2169 J_ASSERT_BH(bh
, buffer_jbd(bh
));
2170 J_ASSERT_BH(bh
, jh2bh(jh
) == bh
);
2171 BUFFER_TRACE(bh
, "remove journal_head");
2172 if (jh
->b_frozen_data
) {
2173 printk(KERN_WARNING
"%s: freeing "
2176 jbd2_free(jh
->b_frozen_data
, bh
->b_size
);
2178 if (jh
->b_committed_data
) {
2179 printk(KERN_WARNING
"%s: freeing "
2180 "b_committed_data\n",
2182 jbd2_free(jh
->b_committed_data
, bh
->b_size
);
2184 bh
->b_private
= NULL
;
2185 jh
->b_bh
= NULL
; /* debug, really */
2186 clear_buffer_jbd(bh
);
2188 journal_free_journal_head(jh
);
2190 BUFFER_TRACE(bh
, "journal_head was locked");
2196 * jbd2_journal_remove_journal_head(): if the buffer isn't attached to a transaction
2197 * and has a zero b_jcount then remove and release its journal_head. If we did
2198 * see that the buffer is not used by any transaction we also "logically"
2199 * decrement ->b_count.
2201 * We in fact take an additional increment on ->b_count as a convenience,
2202 * because the caller usually wants to do additional things with the bh
2203 * after calling here.
2204 * The caller of jbd2_journal_remove_journal_head() *must* run __brelse(bh) at some
2205 * time. Once the caller has run __brelse(), the buffer is eligible for
2206 * reaping by try_to_free_buffers().
2208 void jbd2_journal_remove_journal_head(struct buffer_head
*bh
)
2210 jbd_lock_bh_journal_head(bh
);
2211 __journal_remove_journal_head(bh
);
2212 jbd_unlock_bh_journal_head(bh
);
2216 * Drop a reference on the passed journal_head. If it fell to zero then try to
2217 * release the journal_head from the buffer_head.
2219 void jbd2_journal_put_journal_head(struct journal_head
*jh
)
2221 struct buffer_head
*bh
= jh2bh(jh
);
2223 jbd_lock_bh_journal_head(bh
);
2224 J_ASSERT_JH(jh
, jh
->b_jcount
> 0);
2226 if (!jh
->b_jcount
&& !jh
->b_transaction
) {
2227 __journal_remove_journal_head(bh
);
2230 jbd_unlock_bh_journal_head(bh
);
2234 * Initialize jbd inode head
2236 void jbd2_journal_init_jbd_inode(struct jbd2_inode
*jinode
, struct inode
*inode
)
2238 jinode
->i_transaction
= NULL
;
2239 jinode
->i_next_transaction
= NULL
;
2240 jinode
->i_vfs_inode
= inode
;
2241 jinode
->i_flags
= 0;
2242 INIT_LIST_HEAD(&jinode
->i_list
);
2246 * Function to be called before we start removing inode from memory (i.e.,
2247 * clear_inode() is a fine place to be called from). It removes inode from
2248 * transaction's lists.
2250 void jbd2_journal_release_jbd_inode(journal_t
*journal
,
2251 struct jbd2_inode
*jinode
)
2258 spin_lock(&journal
->j_list_lock
);
2259 /* Is commit writing out inode - we have to wait */
2260 if (jinode
->i_flags
& JI_COMMIT_RUNNING
) {
2261 wait_queue_head_t
*wq
;
2262 DEFINE_WAIT_BIT(wait
, &jinode
->i_flags
, __JI_COMMIT_RUNNING
);
2263 wq
= bit_waitqueue(&jinode
->i_flags
, __JI_COMMIT_RUNNING
);
2264 prepare_to_wait(wq
, &wait
.wait
, TASK_UNINTERRUPTIBLE
);
2265 spin_unlock(&journal
->j_list_lock
);
2267 finish_wait(wq
, &wait
.wait
);
2271 /* Do we need to wait for data writeback? */
2272 if (journal
->j_committing_transaction
== jinode
->i_transaction
)
2274 if (jinode
->i_transaction
) {
2275 list_del(&jinode
->i_list
);
2276 jinode
->i_transaction
= NULL
;
2278 spin_unlock(&journal
->j_list_lock
);
2284 #ifdef CONFIG_JBD2_DEBUG
2285 u8 jbd2_journal_enable_debug __read_mostly
;
2286 EXPORT_SYMBOL(jbd2_journal_enable_debug
);
2288 #define JBD2_DEBUG_NAME "jbd2-debug"
2290 static struct dentry
*jbd2_debugfs_dir
;
2291 static struct dentry
*jbd2_debug
;
2293 static void __init
jbd2_create_debugfs_entry(void)
2295 jbd2_debugfs_dir
= debugfs_create_dir("jbd2", NULL
);
2296 if (jbd2_debugfs_dir
)
2297 jbd2_debug
= debugfs_create_u8(JBD2_DEBUG_NAME
, S_IRUGO
,
2299 &jbd2_journal_enable_debug
);
2302 static void __exit
jbd2_remove_debugfs_entry(void)
2304 debugfs_remove(jbd2_debug
);
2305 debugfs_remove(jbd2_debugfs_dir
);
2310 static void __init
jbd2_create_debugfs_entry(void)
2314 static void __exit
jbd2_remove_debugfs_entry(void)
2320 #ifdef CONFIG_PROC_FS
2322 #define JBD2_STATS_PROC_NAME "fs/jbd2"
2324 static void __init
jbd2_create_jbd_stats_proc_entry(void)
2326 proc_jbd2_stats
= proc_mkdir(JBD2_STATS_PROC_NAME
, NULL
);
2329 static void __exit
jbd2_remove_jbd_stats_proc_entry(void)
2331 if (proc_jbd2_stats
)
2332 remove_proc_entry(JBD2_STATS_PROC_NAME
, NULL
);
2337 #define jbd2_create_jbd_stats_proc_entry() do {} while (0)
2338 #define jbd2_remove_jbd_stats_proc_entry() do {} while (0)
2342 struct kmem_cache
*jbd2_handle_cache
;
2344 static int __init
journal_init_handle_cache(void)
2346 jbd2_handle_cache
= kmem_cache_create("jbd2_journal_handle",
2349 SLAB_TEMPORARY
, /* flags */
2351 if (jbd2_handle_cache
== NULL
) {
2352 printk(KERN_EMERG
"JBD: failed to create handle cache\n");
2358 static void jbd2_journal_destroy_handle_cache(void)
2360 if (jbd2_handle_cache
)
2361 kmem_cache_destroy(jbd2_handle_cache
);
2365 * Module startup and shutdown
2368 static int __init
journal_init_caches(void)
2372 ret
= jbd2_journal_init_revoke_caches();
2374 ret
= journal_init_jbd2_journal_head_cache();
2376 ret
= journal_init_handle_cache();
2380 static void jbd2_journal_destroy_caches(void)
2382 jbd2_journal_destroy_revoke_caches();
2383 jbd2_journal_destroy_jbd2_journal_head_cache();
2384 jbd2_journal_destroy_handle_cache();
2387 static int __init
journal_init(void)
2391 BUILD_BUG_ON(sizeof(struct journal_superblock_s
) != 1024);
2393 ret
= journal_init_caches();
2395 jbd2_create_debugfs_entry();
2396 jbd2_create_jbd_stats_proc_entry();
2398 jbd2_journal_destroy_caches();
2403 static void __exit
journal_exit(void)
2405 #ifdef CONFIG_JBD2_DEBUG
2406 int n
= atomic_read(&nr_journal_heads
);
2408 printk(KERN_EMERG
"JBD: leaked %d journal_heads!\n", n
);
2410 jbd2_remove_debugfs_entry();
2411 jbd2_remove_jbd_stats_proc_entry();
2412 jbd2_journal_destroy_caches();
2415 MODULE_LICENSE("GPL");
2416 module_init(journal_init
);
2417 module_exit(journal_exit
);