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>
43 #include <asm/div64.h>
45 EXPORT_SYMBOL(jbd2_journal_start
);
46 EXPORT_SYMBOL(jbd2_journal_restart
);
47 EXPORT_SYMBOL(jbd2_journal_extend
);
48 EXPORT_SYMBOL(jbd2_journal_stop
);
49 EXPORT_SYMBOL(jbd2_journal_lock_updates
);
50 EXPORT_SYMBOL(jbd2_journal_unlock_updates
);
51 EXPORT_SYMBOL(jbd2_journal_get_write_access
);
52 EXPORT_SYMBOL(jbd2_journal_get_create_access
);
53 EXPORT_SYMBOL(jbd2_journal_get_undo_access
);
54 EXPORT_SYMBOL(jbd2_journal_dirty_metadata
);
55 EXPORT_SYMBOL(jbd2_journal_release_buffer
);
56 EXPORT_SYMBOL(jbd2_journal_forget
);
58 EXPORT_SYMBOL(journal_sync_buffer
);
60 EXPORT_SYMBOL(jbd2_journal_flush
);
61 EXPORT_SYMBOL(jbd2_journal_revoke
);
63 EXPORT_SYMBOL(jbd2_journal_init_dev
);
64 EXPORT_SYMBOL(jbd2_journal_init_inode
);
65 EXPORT_SYMBOL(jbd2_journal_update_format
);
66 EXPORT_SYMBOL(jbd2_journal_check_used_features
);
67 EXPORT_SYMBOL(jbd2_journal_check_available_features
);
68 EXPORT_SYMBOL(jbd2_journal_set_features
);
69 EXPORT_SYMBOL(jbd2_journal_create
);
70 EXPORT_SYMBOL(jbd2_journal_load
);
71 EXPORT_SYMBOL(jbd2_journal_destroy
);
72 EXPORT_SYMBOL(jbd2_journal_abort
);
73 EXPORT_SYMBOL(jbd2_journal_errno
);
74 EXPORT_SYMBOL(jbd2_journal_ack_err
);
75 EXPORT_SYMBOL(jbd2_journal_clear_err
);
76 EXPORT_SYMBOL(jbd2_log_wait_commit
);
77 EXPORT_SYMBOL(jbd2_journal_start_commit
);
78 EXPORT_SYMBOL(jbd2_journal_force_commit_nested
);
79 EXPORT_SYMBOL(jbd2_journal_wipe
);
80 EXPORT_SYMBOL(jbd2_journal_blocks_per_page
);
81 EXPORT_SYMBOL(jbd2_journal_invalidatepage
);
82 EXPORT_SYMBOL(jbd2_journal_try_to_free_buffers
);
83 EXPORT_SYMBOL(jbd2_journal_force_commit
);
84 EXPORT_SYMBOL(jbd2_journal_file_inode
);
85 EXPORT_SYMBOL(jbd2_journal_init_jbd_inode
);
86 EXPORT_SYMBOL(jbd2_journal_release_jbd_inode
);
87 EXPORT_SYMBOL(jbd2_journal_begin_ordered_truncate
);
89 static int journal_convert_superblock_v1(journal_t
*, journal_superblock_t
*);
90 static void __journal_abort_soft (journal_t
*journal
, int errno
);
93 * Helper function used to manage commit timeouts
96 static void commit_timeout(unsigned long __data
)
98 struct task_struct
* p
= (struct task_struct
*) __data
;
104 * kjournald2: The main thread function used to manage a logging device
107 * This kernel thread is responsible for two things:
109 * 1) COMMIT: Every so often we need to commit the current state of the
110 * filesystem to disk. The journal thread is responsible for writing
111 * all of the metadata buffers to disk.
113 * 2) CHECKPOINT: We cannot reuse a used section of the log file until all
114 * of the data in that part of the log has been rewritten elsewhere on
115 * the disk. Flushing these old buffers to reclaim space in the log is
116 * known as checkpointing, and this thread is responsible for that job.
119 static int kjournald2(void *arg
)
121 journal_t
*journal
= arg
;
122 transaction_t
*transaction
;
125 * Set up an interval timer which can be used to trigger a commit wakeup
126 * after the commit interval expires
128 setup_timer(&journal
->j_commit_timer
, commit_timeout
,
129 (unsigned long)current
);
131 /* Record that the journal thread is running */
132 journal
->j_task
= current
;
133 wake_up(&journal
->j_wait_done_commit
);
135 printk(KERN_INFO
"kjournald2 starting. Commit interval %ld seconds\n",
136 journal
->j_commit_interval
/ HZ
);
139 * And now, wait forever for commit wakeup events.
141 spin_lock(&journal
->j_state_lock
);
144 if (journal
->j_flags
& JBD2_UNMOUNT
)
147 jbd_debug(1, "commit_sequence=%d, commit_request=%d\n",
148 journal
->j_commit_sequence
, journal
->j_commit_request
);
150 if (journal
->j_commit_sequence
!= journal
->j_commit_request
) {
151 jbd_debug(1, "OK, requests differ\n");
152 spin_unlock(&journal
->j_state_lock
);
153 del_timer_sync(&journal
->j_commit_timer
);
154 jbd2_journal_commit_transaction(journal
);
155 spin_lock(&journal
->j_state_lock
);
159 wake_up(&journal
->j_wait_done_commit
);
160 if (freezing(current
)) {
162 * The simpler the better. Flushing journal isn't a
163 * good idea, because that depends on threads that may
164 * be already stopped.
166 jbd_debug(1, "Now suspending kjournald2\n");
167 spin_unlock(&journal
->j_state_lock
);
169 spin_lock(&journal
->j_state_lock
);
172 * We assume on resume that commits are already there,
176 int should_sleep
= 1;
178 prepare_to_wait(&journal
->j_wait_commit
, &wait
,
180 if (journal
->j_commit_sequence
!= journal
->j_commit_request
)
182 transaction
= journal
->j_running_transaction
;
183 if (transaction
&& time_after_eq(jiffies
,
184 transaction
->t_expires
))
186 if (journal
->j_flags
& JBD2_UNMOUNT
)
189 spin_unlock(&journal
->j_state_lock
);
191 spin_lock(&journal
->j_state_lock
);
193 finish_wait(&journal
->j_wait_commit
, &wait
);
196 jbd_debug(1, "kjournald2 wakes\n");
199 * Were we woken up by a commit wakeup event?
201 transaction
= journal
->j_running_transaction
;
202 if (transaction
&& time_after_eq(jiffies
, transaction
->t_expires
)) {
203 journal
->j_commit_request
= transaction
->t_tid
;
204 jbd_debug(1, "woke because of timeout\n");
209 spin_unlock(&journal
->j_state_lock
);
210 del_timer_sync(&journal
->j_commit_timer
);
211 journal
->j_task
= NULL
;
212 wake_up(&journal
->j_wait_done_commit
);
213 jbd_debug(1, "Journal thread exiting.\n");
217 static int jbd2_journal_start_thread(journal_t
*journal
)
219 struct task_struct
*t
;
221 t
= kthread_run(kjournald2
, journal
, "kjournald2");
225 wait_event(journal
->j_wait_done_commit
, journal
->j_task
!= NULL
);
229 static void journal_kill_thread(journal_t
*journal
)
231 spin_lock(&journal
->j_state_lock
);
232 journal
->j_flags
|= JBD2_UNMOUNT
;
234 while (journal
->j_task
) {
235 wake_up(&journal
->j_wait_commit
);
236 spin_unlock(&journal
->j_state_lock
);
237 wait_event(journal
->j_wait_done_commit
, journal
->j_task
== NULL
);
238 spin_lock(&journal
->j_state_lock
);
240 spin_unlock(&journal
->j_state_lock
);
244 * jbd2_journal_write_metadata_buffer: write a metadata buffer to the journal.
246 * Writes a metadata buffer to a given disk block. The actual IO is not
247 * performed but a new buffer_head is constructed which labels the data
248 * to be written with the correct destination disk block.
250 * Any magic-number escaping which needs to be done will cause a
251 * copy-out here. If the buffer happens to start with the
252 * JBD2_MAGIC_NUMBER, then we can't write it to the log directly: the
253 * magic number is only written to the log for descripter blocks. In
254 * this case, we copy the data and replace the first word with 0, and we
255 * return a result code which indicates that this buffer needs to be
256 * marked as an escaped buffer in the corresponding log descriptor
257 * block. The missing word can then be restored when the block is read
260 * If the source buffer has already been modified by a new transaction
261 * since we took the last commit snapshot, we use the frozen copy of
262 * that data for IO. If we end up using the existing buffer_head's data
263 * for the write, then we *have* to lock the buffer to prevent anyone
264 * else from using and possibly modifying it while the IO is in
267 * The function returns a pointer to the buffer_heads to be used for IO.
269 * We assume that the journal has already been locked in this function.
276 * Bit 0 set == escape performed on the data
277 * Bit 1 set == buffer copy-out performed (kfree the data after IO)
280 int jbd2_journal_write_metadata_buffer(transaction_t
*transaction
,
281 struct journal_head
*jh_in
,
282 struct journal_head
**jh_out
,
283 unsigned long long blocknr
)
285 int need_copy_out
= 0;
286 int done_copy_out
= 0;
289 struct buffer_head
*new_bh
;
290 struct journal_head
*new_jh
;
291 struct page
*new_page
;
292 unsigned int new_offset
;
293 struct buffer_head
*bh_in
= jh2bh(jh_in
);
296 * The buffer really shouldn't be locked: only the current committing
297 * transaction is allowed to write it, so nobody else is allowed
300 * akpm: except if we're journalling data, and write() output is
301 * also part of a shared mapping, and another thread has
302 * decided to launch a writepage() against this buffer.
304 J_ASSERT_BH(bh_in
, buffer_jbddirty(bh_in
));
306 new_bh
= alloc_buffer_head(GFP_NOFS
|__GFP_NOFAIL
);
309 * If a new transaction has already done a buffer copy-out, then
310 * we use that version of the data for the commit.
312 jbd_lock_bh_state(bh_in
);
314 if (jh_in
->b_frozen_data
) {
316 new_page
= virt_to_page(jh_in
->b_frozen_data
);
317 new_offset
= offset_in_page(jh_in
->b_frozen_data
);
319 new_page
= jh2bh(jh_in
)->b_page
;
320 new_offset
= offset_in_page(jh2bh(jh_in
)->b_data
);
323 mapped_data
= kmap_atomic(new_page
, KM_USER0
);
327 if (*((__be32
*)(mapped_data
+ new_offset
)) ==
328 cpu_to_be32(JBD2_MAGIC_NUMBER
)) {
332 kunmap_atomic(mapped_data
, KM_USER0
);
335 * Do we need to do a data copy?
337 if (need_copy_out
&& !done_copy_out
) {
340 jbd_unlock_bh_state(bh_in
);
341 tmp
= jbd2_alloc(bh_in
->b_size
, GFP_NOFS
);
342 jbd_lock_bh_state(bh_in
);
343 if (jh_in
->b_frozen_data
) {
344 jbd2_free(tmp
, bh_in
->b_size
);
348 jh_in
->b_frozen_data
= tmp
;
349 mapped_data
= kmap_atomic(new_page
, KM_USER0
);
350 memcpy(tmp
, mapped_data
+ new_offset
, jh2bh(jh_in
)->b_size
);
351 kunmap_atomic(mapped_data
, KM_USER0
);
353 new_page
= virt_to_page(tmp
);
354 new_offset
= offset_in_page(tmp
);
359 * Did we need to do an escaping? Now we've done all the
360 * copying, we can finally do so.
363 mapped_data
= kmap_atomic(new_page
, KM_USER0
);
364 *((unsigned int *)(mapped_data
+ new_offset
)) = 0;
365 kunmap_atomic(mapped_data
, KM_USER0
);
368 /* keep subsequent assertions sane */
370 init_buffer(new_bh
, NULL
, NULL
);
371 atomic_set(&new_bh
->b_count
, 1);
372 jbd_unlock_bh_state(bh_in
);
374 new_jh
= jbd2_journal_add_journal_head(new_bh
); /* This sleeps */
376 set_bh_page(new_bh
, new_page
, new_offset
);
377 new_jh
->b_transaction
= NULL
;
378 new_bh
->b_size
= jh2bh(jh_in
)->b_size
;
379 new_bh
->b_bdev
= transaction
->t_journal
->j_dev
;
380 new_bh
->b_blocknr
= blocknr
;
381 set_buffer_mapped(new_bh
);
382 set_buffer_dirty(new_bh
);
387 * The to-be-written buffer needs to get moved to the io queue,
388 * and the original buffer whose contents we are shadowing or
389 * copying is moved to the transaction's shadow queue.
391 JBUFFER_TRACE(jh_in
, "file as BJ_Shadow");
392 jbd2_journal_file_buffer(jh_in
, transaction
, BJ_Shadow
);
393 JBUFFER_TRACE(new_jh
, "file as BJ_IO");
394 jbd2_journal_file_buffer(new_jh
, transaction
, BJ_IO
);
396 return do_escape
| (done_copy_out
<< 1);
400 * Allocation code for the journal file. Manage the space left in the
401 * journal, so that we can begin checkpointing when appropriate.
405 * __jbd2_log_space_left: Return the number of free blocks left in the journal.
407 * Called with the journal already locked.
409 * Called under j_state_lock
412 int __jbd2_log_space_left(journal_t
*journal
)
414 int left
= journal
->j_free
;
416 assert_spin_locked(&journal
->j_state_lock
);
419 * Be pessimistic here about the number of those free blocks which
420 * might be required for log descriptor control blocks.
423 #define MIN_LOG_RESERVED_BLOCKS 32 /* Allow for rounding errors */
425 left
-= MIN_LOG_RESERVED_BLOCKS
;
434 * Called under j_state_lock. Returns true if a transaction was started.
436 int __jbd2_log_start_commit(journal_t
*journal
, tid_t target
)
439 * Are we already doing a recent enough commit?
441 if (!tid_geq(journal
->j_commit_request
, target
)) {
443 * We want a new commit: OK, mark the request and wakup the
444 * commit thread. We do _not_ do the commit ourselves.
447 journal
->j_commit_request
= target
;
448 jbd_debug(1, "JBD: requesting commit %d/%d\n",
449 journal
->j_commit_request
,
450 journal
->j_commit_sequence
);
451 wake_up(&journal
->j_wait_commit
);
457 int jbd2_log_start_commit(journal_t
*journal
, tid_t tid
)
461 spin_lock(&journal
->j_state_lock
);
462 ret
= __jbd2_log_start_commit(journal
, tid
);
463 spin_unlock(&journal
->j_state_lock
);
468 * Force and wait upon a commit if the calling process is not within
469 * transaction. This is used for forcing out undo-protected data which contains
470 * bitmaps, when the fs is running out of space.
472 * We can only force the running transaction if we don't have an active handle;
473 * otherwise, we will deadlock.
475 * Returns true if a transaction was started.
477 int jbd2_journal_force_commit_nested(journal_t
*journal
)
479 transaction_t
*transaction
= NULL
;
482 spin_lock(&journal
->j_state_lock
);
483 if (journal
->j_running_transaction
&& !current
->journal_info
) {
484 transaction
= journal
->j_running_transaction
;
485 __jbd2_log_start_commit(journal
, transaction
->t_tid
);
486 } else if (journal
->j_committing_transaction
)
487 transaction
= journal
->j_committing_transaction
;
490 spin_unlock(&journal
->j_state_lock
);
491 return 0; /* Nothing to retry */
494 tid
= transaction
->t_tid
;
495 spin_unlock(&journal
->j_state_lock
);
496 jbd2_log_wait_commit(journal
, tid
);
501 * Start a commit of the current running transaction (if any). Returns true
502 * if a transaction was started, 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 ret
= __jbd2_log_start_commit(journal
, tid
);
515 } else if (journal
->j_committing_transaction
&& ptid
) {
517 * If ext3_write_super() recently started a commit, then we
518 * have to wait for completion of that transaction
520 *ptid
= journal
->j_committing_transaction
->t_tid
;
523 spin_unlock(&journal
->j_state_lock
);
528 * Wait for a specified commit to complete.
529 * The caller may not hold the journal lock.
531 int jbd2_log_wait_commit(journal_t
*journal
, tid_t tid
)
535 #ifdef CONFIG_JBD2_DEBUG
536 spin_lock(&journal
->j_state_lock
);
537 if (!tid_geq(journal
->j_commit_request
, tid
)) {
539 "%s: error: j_commit_request=%d, tid=%d\n",
540 __func__
, journal
->j_commit_request
, tid
);
542 spin_unlock(&journal
->j_state_lock
);
544 spin_lock(&journal
->j_state_lock
);
545 while (tid_gt(tid
, journal
->j_commit_sequence
)) {
546 jbd_debug(1, "JBD: want %d, j_commit_sequence=%d\n",
547 tid
, journal
->j_commit_sequence
);
548 wake_up(&journal
->j_wait_commit
);
549 spin_unlock(&journal
->j_state_lock
);
550 wait_event(journal
->j_wait_done_commit
,
551 !tid_gt(tid
, journal
->j_commit_sequence
));
552 spin_lock(&journal
->j_state_lock
);
554 spin_unlock(&journal
->j_state_lock
);
556 if (unlikely(is_journal_aborted(journal
))) {
557 printk(KERN_EMERG
"journal commit I/O error\n");
564 * Log buffer allocation routines:
567 int jbd2_journal_next_log_block(journal_t
*journal
, unsigned long long *retp
)
569 unsigned long blocknr
;
571 spin_lock(&journal
->j_state_lock
);
572 J_ASSERT(journal
->j_free
> 1);
574 blocknr
= journal
->j_head
;
577 if (journal
->j_head
== journal
->j_last
)
578 journal
->j_head
= journal
->j_first
;
579 spin_unlock(&journal
->j_state_lock
);
580 return jbd2_journal_bmap(journal
, blocknr
, retp
);
584 * Conversion of logical to physical block numbers for the journal
586 * On external journals the journal blocks are identity-mapped, so
587 * this is a no-op. If needed, we can use j_blk_offset - everything is
590 int jbd2_journal_bmap(journal_t
*journal
, unsigned long blocknr
,
591 unsigned long long *retp
)
594 unsigned long long ret
;
596 if (journal
->j_inode
) {
597 ret
= bmap(journal
->j_inode
, blocknr
);
601 printk(KERN_ALERT
"%s: journal block not found "
602 "at offset %lu on %s\n",
603 __func__
, blocknr
, journal
->j_devname
);
605 __journal_abort_soft(journal
, err
);
608 *retp
= blocknr
; /* +journal->j_blk_offset */
614 * We play buffer_head aliasing tricks to write data/metadata blocks to
615 * the journal without copying their contents, but for journal
616 * descriptor blocks we do need to generate bona fide buffers.
618 * After the caller of jbd2_journal_get_descriptor_buffer() has finished modifying
619 * the buffer's contents they really should run flush_dcache_page(bh->b_page).
620 * But we don't bother doing that, so there will be coherency problems with
621 * mmaps of blockdevs which hold live JBD-controlled filesystems.
623 struct journal_head
*jbd2_journal_get_descriptor_buffer(journal_t
*journal
)
625 struct buffer_head
*bh
;
626 unsigned long long blocknr
;
629 err
= jbd2_journal_next_log_block(journal
, &blocknr
);
634 bh
= __getblk(journal
->j_dev
, blocknr
, journal
->j_blocksize
);
636 memset(bh
->b_data
, 0, journal
->j_blocksize
);
637 set_buffer_uptodate(bh
);
639 BUFFER_TRACE(bh
, "return this buffer");
640 return jbd2_journal_add_journal_head(bh
);
643 struct jbd2_stats_proc_session
{
645 struct transaction_stats_s
*stats
;
650 static void *jbd2_history_skip_empty(struct jbd2_stats_proc_session
*s
,
651 struct transaction_stats_s
*ts
,
654 if (ts
== s
->stats
+ s
->max
)
656 if (!first
&& ts
== s
->stats
+ s
->start
)
658 while (ts
->ts_type
== 0) {
660 if (ts
== s
->stats
+ s
->max
)
662 if (ts
== s
->stats
+ s
->start
)
669 static void *jbd2_seq_history_start(struct seq_file
*seq
, loff_t
*pos
)
671 struct jbd2_stats_proc_session
*s
= seq
->private;
672 struct transaction_stats_s
*ts
;
676 return SEQ_START_TOKEN
;
677 ts
= jbd2_history_skip_empty(s
, s
->stats
+ s
->start
, 1);
682 ts
= jbd2_history_skip_empty(s
, ++ts
, 0);
690 static void *jbd2_seq_history_next(struct seq_file
*seq
, void *v
, loff_t
*pos
)
692 struct jbd2_stats_proc_session
*s
= seq
->private;
693 struct transaction_stats_s
*ts
= v
;
696 if (v
== SEQ_START_TOKEN
)
697 return jbd2_history_skip_empty(s
, s
->stats
+ s
->start
, 1);
699 return jbd2_history_skip_empty(s
, ++ts
, 0);
702 static int jbd2_seq_history_show(struct seq_file
*seq
, void *v
)
704 struct transaction_stats_s
*ts
= v
;
705 if (v
== SEQ_START_TOKEN
) {
706 seq_printf(seq
, "%-4s %-5s %-5s %-5s %-5s %-5s %-5s %-6s %-5s "
707 "%-5s %-5s %-5s %-5s %-5s\n", "R/C", "tid",
708 "wait", "run", "lock", "flush", "log", "hndls",
709 "block", "inlog", "ctime", "write", "drop",
713 if (ts
->ts_type
== JBD2_STATS_RUN
)
714 seq_printf(seq
, "%-4s %-5lu %-5u %-5u %-5u %-5u %-5u "
715 "%-6lu %-5lu %-5lu\n", "R", ts
->ts_tid
,
716 jiffies_to_msecs(ts
->u
.run
.rs_wait
),
717 jiffies_to_msecs(ts
->u
.run
.rs_running
),
718 jiffies_to_msecs(ts
->u
.run
.rs_locked
),
719 jiffies_to_msecs(ts
->u
.run
.rs_flushing
),
720 jiffies_to_msecs(ts
->u
.run
.rs_logging
),
721 ts
->u
.run
.rs_handle_count
,
723 ts
->u
.run
.rs_blocks_logged
);
724 else if (ts
->ts_type
== JBD2_STATS_CHECKPOINT
)
725 seq_printf(seq
, "%-4s %-5lu %48s %-5u %-5lu %-5lu %-5lu\n",
726 "C", ts
->ts_tid
, " ",
727 jiffies_to_msecs(ts
->u
.chp
.cs_chp_time
),
728 ts
->u
.chp
.cs_written
, ts
->u
.chp
.cs_dropped
,
729 ts
->u
.chp
.cs_forced_to_close
);
735 static void jbd2_seq_history_stop(struct seq_file
*seq
, void *v
)
739 static struct seq_operations jbd2_seq_history_ops
= {
740 .start
= jbd2_seq_history_start
,
741 .next
= jbd2_seq_history_next
,
742 .stop
= jbd2_seq_history_stop
,
743 .show
= jbd2_seq_history_show
,
746 static int jbd2_seq_history_open(struct inode
*inode
, struct file
*file
)
748 journal_t
*journal
= PDE(inode
)->data
;
749 struct jbd2_stats_proc_session
*s
;
752 s
= kmalloc(sizeof(*s
), GFP_KERNEL
);
755 size
= sizeof(struct transaction_stats_s
) * journal
->j_history_max
;
756 s
->stats
= kmalloc(size
, GFP_KERNEL
);
757 if (s
->stats
== NULL
) {
761 spin_lock(&journal
->j_history_lock
);
762 memcpy(s
->stats
, journal
->j_history
, size
);
763 s
->max
= journal
->j_history_max
;
764 s
->start
= journal
->j_history_cur
% s
->max
;
765 spin_unlock(&journal
->j_history_lock
);
767 rc
= seq_open(file
, &jbd2_seq_history_ops
);
769 struct seq_file
*m
= file
->private_data
;
779 static int jbd2_seq_history_release(struct inode
*inode
, struct file
*file
)
781 struct seq_file
*seq
= file
->private_data
;
782 struct jbd2_stats_proc_session
*s
= seq
->private;
786 return seq_release(inode
, file
);
789 static struct file_operations jbd2_seq_history_fops
= {
790 .owner
= THIS_MODULE
,
791 .open
= jbd2_seq_history_open
,
794 .release
= jbd2_seq_history_release
,
797 static void *jbd2_seq_info_start(struct seq_file
*seq
, loff_t
*pos
)
799 return *pos
? NULL
: SEQ_START_TOKEN
;
802 static void *jbd2_seq_info_next(struct seq_file
*seq
, void *v
, loff_t
*pos
)
807 static int jbd2_seq_info_show(struct seq_file
*seq
, void *v
)
809 struct jbd2_stats_proc_session
*s
= seq
->private;
811 if (v
!= SEQ_START_TOKEN
)
813 seq_printf(seq
, "%lu transaction, each upto %u blocks\n",
815 s
->journal
->j_max_transaction_buffers
);
816 if (s
->stats
->ts_tid
== 0)
818 seq_printf(seq
, "average: \n %ums waiting for transaction\n",
819 jiffies_to_msecs(s
->stats
->u
.run
.rs_wait
/ s
->stats
->ts_tid
));
820 seq_printf(seq
, " %ums running transaction\n",
821 jiffies_to_msecs(s
->stats
->u
.run
.rs_running
/ s
->stats
->ts_tid
));
822 seq_printf(seq
, " %ums transaction was being locked\n",
823 jiffies_to_msecs(s
->stats
->u
.run
.rs_locked
/ s
->stats
->ts_tid
));
824 seq_printf(seq
, " %ums flushing data (in ordered mode)\n",
825 jiffies_to_msecs(s
->stats
->u
.run
.rs_flushing
/ s
->stats
->ts_tid
));
826 seq_printf(seq
, " %ums logging transaction\n",
827 jiffies_to_msecs(s
->stats
->u
.run
.rs_logging
/ s
->stats
->ts_tid
));
828 seq_printf(seq
, " %luus average transaction commit time\n",
829 do_div(s
->journal
->j_average_commit_time
, 1000));
830 seq_printf(seq
, " %lu handles per transaction\n",
831 s
->stats
->u
.run
.rs_handle_count
/ s
->stats
->ts_tid
);
832 seq_printf(seq
, " %lu blocks per transaction\n",
833 s
->stats
->u
.run
.rs_blocks
/ s
->stats
->ts_tid
);
834 seq_printf(seq
, " %lu logged blocks per transaction\n",
835 s
->stats
->u
.run
.rs_blocks_logged
/ s
->stats
->ts_tid
);
839 static void jbd2_seq_info_stop(struct seq_file
*seq
, void *v
)
843 static struct seq_operations jbd2_seq_info_ops
= {
844 .start
= jbd2_seq_info_start
,
845 .next
= jbd2_seq_info_next
,
846 .stop
= jbd2_seq_info_stop
,
847 .show
= jbd2_seq_info_show
,
850 static int jbd2_seq_info_open(struct inode
*inode
, struct file
*file
)
852 journal_t
*journal
= PDE(inode
)->data
;
853 struct jbd2_stats_proc_session
*s
;
856 s
= kmalloc(sizeof(*s
), GFP_KERNEL
);
859 size
= sizeof(struct transaction_stats_s
);
860 s
->stats
= kmalloc(size
, GFP_KERNEL
);
861 if (s
->stats
== NULL
) {
865 spin_lock(&journal
->j_history_lock
);
866 memcpy(s
->stats
, &journal
->j_stats
, size
);
867 s
->journal
= journal
;
868 spin_unlock(&journal
->j_history_lock
);
870 rc
= seq_open(file
, &jbd2_seq_info_ops
);
872 struct seq_file
*m
= file
->private_data
;
882 static int jbd2_seq_info_release(struct inode
*inode
, struct file
*file
)
884 struct seq_file
*seq
= file
->private_data
;
885 struct jbd2_stats_proc_session
*s
= seq
->private;
888 return seq_release(inode
, file
);
891 static struct file_operations jbd2_seq_info_fops
= {
892 .owner
= THIS_MODULE
,
893 .open
= jbd2_seq_info_open
,
896 .release
= jbd2_seq_info_release
,
899 static struct proc_dir_entry
*proc_jbd2_stats
;
901 static void jbd2_stats_proc_init(journal_t
*journal
)
903 journal
->j_proc_entry
= proc_mkdir(journal
->j_devname
, proc_jbd2_stats
);
904 if (journal
->j_proc_entry
) {
905 proc_create_data("history", S_IRUGO
, journal
->j_proc_entry
,
906 &jbd2_seq_history_fops
, journal
);
907 proc_create_data("info", S_IRUGO
, journal
->j_proc_entry
,
908 &jbd2_seq_info_fops
, journal
);
912 static void jbd2_stats_proc_exit(journal_t
*journal
)
914 remove_proc_entry("info", journal
->j_proc_entry
);
915 remove_proc_entry("history", journal
->j_proc_entry
);
916 remove_proc_entry(journal
->j_devname
, proc_jbd2_stats
);
919 static void journal_init_stats(journal_t
*journal
)
923 if (!proc_jbd2_stats
)
926 journal
->j_history_max
= 100;
927 size
= sizeof(struct transaction_stats_s
) * journal
->j_history_max
;
928 journal
->j_history
= kzalloc(size
, GFP_KERNEL
);
929 if (!journal
->j_history
) {
930 journal
->j_history_max
= 0;
933 spin_lock_init(&journal
->j_history_lock
);
937 * Management for journal control blocks: functions to create and
938 * destroy journal_t structures, and to initialise and read existing
939 * journal blocks from disk. */
941 /* First: create and setup a journal_t object in memory. We initialise
942 * very few fields yet: that has to wait until we have created the
943 * journal structures from from scratch, or loaded them from disk. */
945 static journal_t
* journal_init_common (void)
950 journal
= kzalloc(sizeof(*journal
), GFP_KERNEL
|__GFP_NOFAIL
);
954 init_waitqueue_head(&journal
->j_wait_transaction_locked
);
955 init_waitqueue_head(&journal
->j_wait_logspace
);
956 init_waitqueue_head(&journal
->j_wait_done_commit
);
957 init_waitqueue_head(&journal
->j_wait_checkpoint
);
958 init_waitqueue_head(&journal
->j_wait_commit
);
959 init_waitqueue_head(&journal
->j_wait_updates
);
960 mutex_init(&journal
->j_barrier
);
961 mutex_init(&journal
->j_checkpoint_mutex
);
962 spin_lock_init(&journal
->j_revoke_lock
);
963 spin_lock_init(&journal
->j_list_lock
);
964 spin_lock_init(&journal
->j_state_lock
);
966 journal
->j_commit_interval
= (HZ
* JBD2_DEFAULT_MAX_COMMIT_AGE
);
967 journal
->j_min_batch_time
= 0;
968 journal
->j_max_batch_time
= 15000; /* 15ms */
970 /* The journal is marked for error until we succeed with recovery! */
971 journal
->j_flags
= JBD2_ABORT
;
973 /* Set up a default-sized revoke table for the new mount. */
974 err
= jbd2_journal_init_revoke(journal
, JOURNAL_REVOKE_DEFAULT_HASH
);
980 journal_init_stats(journal
);
987 /* jbd2_journal_init_dev and jbd2_journal_init_inode:
989 * Create a journal structure assigned some fixed set of disk blocks to
990 * the journal. We don't actually touch those disk blocks yet, but we
991 * need to set up all of the mapping information to tell the journaling
992 * system where the journal blocks are.
997 * journal_t * jbd2_journal_init_dev() - creates and initialises a journal structure
998 * @bdev: Block device on which to create the journal
999 * @fs_dev: Device which hold journalled filesystem for this journal.
1000 * @start: Block nr Start of journal.
1001 * @len: Length of the journal in blocks.
1002 * @blocksize: blocksize of journalling device
1004 * Returns: a newly created journal_t *
1006 * jbd2_journal_init_dev creates a journal which maps a fixed contiguous
1007 * range of blocks on an arbitrary block device.
1010 journal_t
* jbd2_journal_init_dev(struct block_device
*bdev
,
1011 struct block_device
*fs_dev
,
1012 unsigned long long start
, int len
, int blocksize
)
1014 journal_t
*journal
= journal_init_common();
1015 struct buffer_head
*bh
;
1022 /* journal descriptor can store up to n blocks -bzzz */
1023 journal
->j_blocksize
= blocksize
;
1024 n
= journal
->j_blocksize
/ sizeof(journal_block_tag_t
);
1025 journal
->j_wbufsize
= n
;
1026 journal
->j_wbuf
= kmalloc(n
* sizeof(struct buffer_head
*), GFP_KERNEL
);
1027 if (!journal
->j_wbuf
) {
1028 printk(KERN_ERR
"%s: Cant allocate bhs for commit thread\n",
1034 journal
->j_dev
= bdev
;
1035 journal
->j_fs_dev
= fs_dev
;
1036 journal
->j_blk_offset
= start
;
1037 journal
->j_maxlen
= len
;
1038 bdevname(journal
->j_dev
, journal
->j_devname
);
1039 p
= journal
->j_devname
;
1040 while ((p
= strchr(p
, '/')))
1042 jbd2_stats_proc_init(journal
);
1044 bh
= __getblk(journal
->j_dev
, start
, journal
->j_blocksize
);
1045 J_ASSERT(bh
!= NULL
);
1046 journal
->j_sb_buffer
= bh
;
1047 journal
->j_superblock
= (journal_superblock_t
*)bh
->b_data
;
1053 * journal_t * jbd2_journal_init_inode () - creates a journal which maps to a inode.
1054 * @inode: An inode to create the journal in
1056 * jbd2_journal_init_inode creates a journal which maps an on-disk inode as
1057 * the journal. The inode must exist already, must support bmap() and
1058 * must have all data blocks preallocated.
1060 journal_t
* jbd2_journal_init_inode (struct inode
*inode
)
1062 struct buffer_head
*bh
;
1063 journal_t
*journal
= journal_init_common();
1067 unsigned long long blocknr
;
1072 journal
->j_dev
= journal
->j_fs_dev
= inode
->i_sb
->s_bdev
;
1073 journal
->j_inode
= inode
;
1074 bdevname(journal
->j_dev
, journal
->j_devname
);
1075 p
= journal
->j_devname
;
1076 while ((p
= strchr(p
, '/')))
1078 p
= journal
->j_devname
+ strlen(journal
->j_devname
);
1079 sprintf(p
, ":%lu", journal
->j_inode
->i_ino
);
1081 "journal %p: inode %s/%ld, size %Ld, bits %d, blksize %ld\n",
1082 journal
, inode
->i_sb
->s_id
, inode
->i_ino
,
1083 (long long) inode
->i_size
,
1084 inode
->i_sb
->s_blocksize_bits
, inode
->i_sb
->s_blocksize
);
1086 journal
->j_maxlen
= inode
->i_size
>> inode
->i_sb
->s_blocksize_bits
;
1087 journal
->j_blocksize
= inode
->i_sb
->s_blocksize
;
1088 jbd2_stats_proc_init(journal
);
1090 /* journal descriptor can store up to n blocks -bzzz */
1091 n
= journal
->j_blocksize
/ sizeof(journal_block_tag_t
);
1092 journal
->j_wbufsize
= n
;
1093 journal
->j_wbuf
= kmalloc(n
* sizeof(struct buffer_head
*), GFP_KERNEL
);
1094 if (!journal
->j_wbuf
) {
1095 printk(KERN_ERR
"%s: Cant allocate bhs for commit thread\n",
1097 jbd2_stats_proc_exit(journal
);
1102 err
= jbd2_journal_bmap(journal
, 0, &blocknr
);
1103 /* If that failed, give up */
1105 printk(KERN_ERR
"%s: Cannnot locate journal superblock\n",
1107 jbd2_stats_proc_exit(journal
);
1112 bh
= __getblk(journal
->j_dev
, blocknr
, journal
->j_blocksize
);
1113 J_ASSERT(bh
!= NULL
);
1114 journal
->j_sb_buffer
= bh
;
1115 journal
->j_superblock
= (journal_superblock_t
*)bh
->b_data
;
1121 * If the journal init or create aborts, we need to mark the journal
1122 * superblock as being NULL to prevent the journal destroy from writing
1123 * back a bogus superblock.
1125 static void journal_fail_superblock (journal_t
*journal
)
1127 struct buffer_head
*bh
= journal
->j_sb_buffer
;
1129 journal
->j_sb_buffer
= NULL
;
1133 * Given a journal_t structure, initialise the various fields for
1134 * startup of a new journaling session. We use this both when creating
1135 * a journal, and after recovering an old journal to reset it for
1139 static int journal_reset(journal_t
*journal
)
1141 journal_superblock_t
*sb
= journal
->j_superblock
;
1142 unsigned long long first
, last
;
1144 first
= be32_to_cpu(sb
->s_first
);
1145 last
= be32_to_cpu(sb
->s_maxlen
);
1147 journal
->j_first
= first
;
1148 journal
->j_last
= last
;
1150 journal
->j_head
= first
;
1151 journal
->j_tail
= first
;
1152 journal
->j_free
= last
- first
;
1154 journal
->j_tail_sequence
= journal
->j_transaction_sequence
;
1155 journal
->j_commit_sequence
= journal
->j_transaction_sequence
- 1;
1156 journal
->j_commit_request
= journal
->j_commit_sequence
;
1158 journal
->j_max_transaction_buffers
= journal
->j_maxlen
/ 4;
1160 /* Add the dynamic fields and write it to disk. */
1161 jbd2_journal_update_superblock(journal
, 1);
1162 return jbd2_journal_start_thread(journal
);
1166 * int jbd2_journal_create() - Initialise the new journal file
1167 * @journal: Journal to create. This structure must have been initialised
1169 * Given a journal_t structure which tells us which disk blocks we can
1170 * use, create a new journal superblock and initialise all of the
1171 * journal fields from scratch.
1173 int jbd2_journal_create(journal_t
*journal
)
1175 unsigned long long blocknr
;
1176 struct buffer_head
*bh
;
1177 journal_superblock_t
*sb
;
1180 if (journal
->j_maxlen
< JBD2_MIN_JOURNAL_BLOCKS
) {
1181 printk (KERN_ERR
"Journal length (%d blocks) too short.\n",
1183 journal_fail_superblock(journal
);
1187 if (journal
->j_inode
== NULL
) {
1189 * We don't know what block to start at!
1192 "%s: creation of journal on external device!\n",
1197 /* Zero out the entire journal on disk. We cannot afford to
1198 have any blocks on disk beginning with JBD2_MAGIC_NUMBER. */
1199 jbd_debug(1, "JBD: Zeroing out journal blocks...\n");
1200 for (i
= 0; i
< journal
->j_maxlen
; i
++) {
1201 err
= jbd2_journal_bmap(journal
, i
, &blocknr
);
1204 bh
= __getblk(journal
->j_dev
, blocknr
, journal
->j_blocksize
);
1206 memset (bh
->b_data
, 0, journal
->j_blocksize
);
1207 BUFFER_TRACE(bh
, "marking dirty");
1208 mark_buffer_dirty(bh
);
1209 BUFFER_TRACE(bh
, "marking uptodate");
1210 set_buffer_uptodate(bh
);
1215 sync_blockdev(journal
->j_dev
);
1216 jbd_debug(1, "JBD: journal cleared.\n");
1218 /* OK, fill in the initial static fields in the new superblock */
1219 sb
= journal
->j_superblock
;
1221 sb
->s_header
.h_magic
= cpu_to_be32(JBD2_MAGIC_NUMBER
);
1222 sb
->s_header
.h_blocktype
= cpu_to_be32(JBD2_SUPERBLOCK_V2
);
1224 sb
->s_blocksize
= cpu_to_be32(journal
->j_blocksize
);
1225 sb
->s_maxlen
= cpu_to_be32(journal
->j_maxlen
);
1226 sb
->s_first
= cpu_to_be32(1);
1228 journal
->j_transaction_sequence
= 1;
1230 journal
->j_flags
&= ~JBD2_ABORT
;
1231 journal
->j_format_version
= 2;
1233 return journal_reset(journal
);
1237 * void jbd2_journal_update_superblock() - Update journal sb on disk.
1238 * @journal: The journal to update.
1239 * @wait: Set to '0' if you don't want to wait for IO completion.
1241 * Update a journal's dynamic superblock fields and write it to disk,
1242 * optionally waiting for the IO to complete.
1244 void jbd2_journal_update_superblock(journal_t
*journal
, int wait
)
1246 journal_superblock_t
*sb
= journal
->j_superblock
;
1247 struct buffer_head
*bh
= journal
->j_sb_buffer
;
1250 * As a special case, if the on-disk copy is already marked as needing
1251 * no recovery (s_start == 0) and there are no outstanding transactions
1252 * in the filesystem, then we can safely defer the superblock update
1253 * until the next commit by setting JBD2_FLUSHED. This avoids
1254 * attempting a write to a potential-readonly device.
1256 if (sb
->s_start
== 0 && journal
->j_tail_sequence
==
1257 journal
->j_transaction_sequence
) {
1258 jbd_debug(1,"JBD: Skipping superblock update on recovered sb "
1259 "(start %ld, seq %d, errno %d)\n",
1260 journal
->j_tail
, journal
->j_tail_sequence
,
1265 if (buffer_write_io_error(bh
)) {
1267 * Oh, dear. A previous attempt to write the journal
1268 * superblock failed. This could happen because the
1269 * USB device was yanked out. Or it could happen to
1270 * be a transient write error and maybe the block will
1271 * be remapped. Nothing we can do but to retry the
1272 * write and hope for the best.
1274 printk(KERN_ERR
"JBD2: previous I/O error detected "
1275 "for journal superblock update for %s.\n",
1276 journal
->j_devname
);
1277 clear_buffer_write_io_error(bh
);
1278 set_buffer_uptodate(bh
);
1281 spin_lock(&journal
->j_state_lock
);
1282 jbd_debug(1,"JBD: updating superblock (start %ld, seq %d, errno %d)\n",
1283 journal
->j_tail
, journal
->j_tail_sequence
, journal
->j_errno
);
1285 sb
->s_sequence
= cpu_to_be32(journal
->j_tail_sequence
);
1286 sb
->s_start
= cpu_to_be32(journal
->j_tail
);
1287 sb
->s_errno
= cpu_to_be32(journal
->j_errno
);
1288 spin_unlock(&journal
->j_state_lock
);
1290 BUFFER_TRACE(bh
, "marking dirty");
1291 mark_buffer_dirty(bh
);
1293 sync_dirty_buffer(bh
);
1294 if (buffer_write_io_error(bh
)) {
1295 printk(KERN_ERR
"JBD2: I/O error detected "
1296 "when updating journal superblock for %s.\n",
1297 journal
->j_devname
);
1298 clear_buffer_write_io_error(bh
);
1299 set_buffer_uptodate(bh
);
1302 ll_rw_block(SWRITE
, 1, &bh
);
1305 /* If we have just flushed the log (by marking s_start==0), then
1306 * any future commit will have to be careful to update the
1307 * superblock again to re-record the true start of the log. */
1309 spin_lock(&journal
->j_state_lock
);
1311 journal
->j_flags
&= ~JBD2_FLUSHED
;
1313 journal
->j_flags
|= JBD2_FLUSHED
;
1314 spin_unlock(&journal
->j_state_lock
);
1318 * Read the superblock for a given journal, performing initial
1319 * validation of the format.
1322 static int journal_get_superblock(journal_t
*journal
)
1324 struct buffer_head
*bh
;
1325 journal_superblock_t
*sb
;
1328 bh
= journal
->j_sb_buffer
;
1330 J_ASSERT(bh
!= NULL
);
1331 if (!buffer_uptodate(bh
)) {
1332 ll_rw_block(READ
, 1, &bh
);
1334 if (!buffer_uptodate(bh
)) {
1336 "JBD: IO error reading journal superblock\n");
1341 sb
= journal
->j_superblock
;
1345 if (sb
->s_header
.h_magic
!= cpu_to_be32(JBD2_MAGIC_NUMBER
) ||
1346 sb
->s_blocksize
!= cpu_to_be32(journal
->j_blocksize
)) {
1347 printk(KERN_WARNING
"JBD: no valid journal superblock found\n");
1351 switch(be32_to_cpu(sb
->s_header
.h_blocktype
)) {
1352 case JBD2_SUPERBLOCK_V1
:
1353 journal
->j_format_version
= 1;
1355 case JBD2_SUPERBLOCK_V2
:
1356 journal
->j_format_version
= 2;
1359 printk(KERN_WARNING
"JBD: unrecognised superblock format ID\n");
1363 if (be32_to_cpu(sb
->s_maxlen
) < journal
->j_maxlen
)
1364 journal
->j_maxlen
= be32_to_cpu(sb
->s_maxlen
);
1365 else if (be32_to_cpu(sb
->s_maxlen
) > journal
->j_maxlen
) {
1366 printk (KERN_WARNING
"JBD: journal file too short\n");
1373 journal_fail_superblock(journal
);
1378 * Load the on-disk journal superblock and read the key fields into the
1382 static int load_superblock(journal_t
*journal
)
1385 journal_superblock_t
*sb
;
1387 err
= journal_get_superblock(journal
);
1391 sb
= journal
->j_superblock
;
1393 journal
->j_tail_sequence
= be32_to_cpu(sb
->s_sequence
);
1394 journal
->j_tail
= be32_to_cpu(sb
->s_start
);
1395 journal
->j_first
= be32_to_cpu(sb
->s_first
);
1396 journal
->j_last
= be32_to_cpu(sb
->s_maxlen
);
1397 journal
->j_errno
= be32_to_cpu(sb
->s_errno
);
1404 * int jbd2_journal_load() - Read journal from disk.
1405 * @journal: Journal to act on.
1407 * Given a journal_t structure which tells us which disk blocks contain
1408 * a journal, read the journal from disk to initialise the in-memory
1411 int jbd2_journal_load(journal_t
*journal
)
1414 journal_superblock_t
*sb
;
1416 err
= load_superblock(journal
);
1420 sb
= journal
->j_superblock
;
1421 /* If this is a V2 superblock, then we have to check the
1422 * features flags on it. */
1424 if (journal
->j_format_version
>= 2) {
1425 if ((sb
->s_feature_ro_compat
&
1426 ~cpu_to_be32(JBD2_KNOWN_ROCOMPAT_FEATURES
)) ||
1427 (sb
->s_feature_incompat
&
1428 ~cpu_to_be32(JBD2_KNOWN_INCOMPAT_FEATURES
))) {
1429 printk (KERN_WARNING
1430 "JBD: Unrecognised features on journal\n");
1435 /* Let the recovery code check whether it needs to recover any
1436 * data from the journal. */
1437 if (jbd2_journal_recover(journal
))
1438 goto recovery_error
;
1440 /* OK, we've finished with the dynamic journal bits:
1441 * reinitialise the dynamic contents of the superblock in memory
1442 * and reset them on disk. */
1443 if (journal_reset(journal
))
1444 goto recovery_error
;
1446 journal
->j_flags
&= ~JBD2_ABORT
;
1447 journal
->j_flags
|= JBD2_LOADED
;
1451 printk (KERN_WARNING
"JBD: recovery failed\n");
1456 * void jbd2_journal_destroy() - Release a journal_t structure.
1457 * @journal: Journal to act on.
1459 * Release a journal_t structure once it is no longer in use by the
1461 * Return <0 if we couldn't clean up the journal.
1463 int jbd2_journal_destroy(journal_t
*journal
)
1467 /* Wait for the commit thread to wake up and die. */
1468 journal_kill_thread(journal
);
1470 /* Force a final log commit */
1471 if (journal
->j_running_transaction
)
1472 jbd2_journal_commit_transaction(journal
);
1474 /* Force any old transactions to disk */
1476 /* Totally anal locking here... */
1477 spin_lock(&journal
->j_list_lock
);
1478 while (journal
->j_checkpoint_transactions
!= NULL
) {
1479 spin_unlock(&journal
->j_list_lock
);
1480 mutex_lock(&journal
->j_checkpoint_mutex
);
1481 jbd2_log_do_checkpoint(journal
);
1482 mutex_unlock(&journal
->j_checkpoint_mutex
);
1483 spin_lock(&journal
->j_list_lock
);
1486 J_ASSERT(journal
->j_running_transaction
== NULL
);
1487 J_ASSERT(journal
->j_committing_transaction
== NULL
);
1488 J_ASSERT(journal
->j_checkpoint_transactions
== NULL
);
1489 spin_unlock(&journal
->j_list_lock
);
1491 if (journal
->j_sb_buffer
) {
1492 if (!is_journal_aborted(journal
)) {
1493 /* We can now mark the journal as empty. */
1494 journal
->j_tail
= 0;
1495 journal
->j_tail_sequence
=
1496 ++journal
->j_transaction_sequence
;
1497 jbd2_journal_update_superblock(journal
, 1);
1501 brelse(journal
->j_sb_buffer
);
1504 if (journal
->j_proc_entry
)
1505 jbd2_stats_proc_exit(journal
);
1506 if (journal
->j_inode
)
1507 iput(journal
->j_inode
);
1508 if (journal
->j_revoke
)
1509 jbd2_journal_destroy_revoke(journal
);
1510 kfree(journal
->j_wbuf
);
1518 *int jbd2_journal_check_used_features () - Check if features specified are used.
1519 * @journal: Journal to check.
1520 * @compat: bitmask of compatible features
1521 * @ro: bitmask of features that force read-only mount
1522 * @incompat: bitmask of incompatible features
1524 * Check whether the journal uses all of a given set of
1525 * features. Return true (non-zero) if it does.
1528 int jbd2_journal_check_used_features (journal_t
*journal
, unsigned long compat
,
1529 unsigned long ro
, unsigned long incompat
)
1531 journal_superblock_t
*sb
;
1533 if (!compat
&& !ro
&& !incompat
)
1535 if (journal
->j_format_version
== 1)
1538 sb
= journal
->j_superblock
;
1540 if (((be32_to_cpu(sb
->s_feature_compat
) & compat
) == compat
) &&
1541 ((be32_to_cpu(sb
->s_feature_ro_compat
) & ro
) == ro
) &&
1542 ((be32_to_cpu(sb
->s_feature_incompat
) & incompat
) == incompat
))
1549 * int jbd2_journal_check_available_features() - Check feature set in journalling layer
1550 * @journal: Journal to check.
1551 * @compat: bitmask of compatible features
1552 * @ro: bitmask of features that force read-only mount
1553 * @incompat: bitmask of incompatible features
1555 * Check whether the journaling code supports the use of
1556 * all of a given set of features on this journal. Return true
1557 * (non-zero) if it can. */
1559 int jbd2_journal_check_available_features (journal_t
*journal
, unsigned long compat
,
1560 unsigned long ro
, unsigned long incompat
)
1562 journal_superblock_t
*sb
;
1564 if (!compat
&& !ro
&& !incompat
)
1567 sb
= journal
->j_superblock
;
1569 /* We can support any known requested features iff the
1570 * superblock is in version 2. Otherwise we fail to support any
1571 * extended sb features. */
1573 if (journal
->j_format_version
!= 2)
1576 if ((compat
& JBD2_KNOWN_COMPAT_FEATURES
) == compat
&&
1577 (ro
& JBD2_KNOWN_ROCOMPAT_FEATURES
) == ro
&&
1578 (incompat
& JBD2_KNOWN_INCOMPAT_FEATURES
) == incompat
)
1585 * int jbd2_journal_set_features () - Mark a given journal feature in the superblock
1586 * @journal: Journal to act on.
1587 * @compat: bitmask of compatible features
1588 * @ro: bitmask of features that force read-only mount
1589 * @incompat: bitmask of incompatible features
1591 * Mark a given journal feature as present on the
1592 * superblock. Returns true if the requested features could be set.
1596 int jbd2_journal_set_features (journal_t
*journal
, unsigned long compat
,
1597 unsigned long ro
, unsigned long incompat
)
1599 journal_superblock_t
*sb
;
1601 if (jbd2_journal_check_used_features(journal
, compat
, ro
, incompat
))
1604 if (!jbd2_journal_check_available_features(journal
, compat
, ro
, incompat
))
1607 jbd_debug(1, "Setting new features 0x%lx/0x%lx/0x%lx\n",
1608 compat
, ro
, incompat
);
1610 sb
= journal
->j_superblock
;
1612 sb
->s_feature_compat
|= cpu_to_be32(compat
);
1613 sb
->s_feature_ro_compat
|= cpu_to_be32(ro
);
1614 sb
->s_feature_incompat
|= cpu_to_be32(incompat
);
1620 * jbd2_journal_clear_features () - Clear a given journal feature in the
1622 * @journal: Journal to act on.
1623 * @compat: bitmask of compatible features
1624 * @ro: bitmask of features that force read-only mount
1625 * @incompat: bitmask of incompatible features
1627 * Clear a given journal feature as present on the
1630 void jbd2_journal_clear_features(journal_t
*journal
, unsigned long compat
,
1631 unsigned long ro
, unsigned long incompat
)
1633 journal_superblock_t
*sb
;
1635 jbd_debug(1, "Clear features 0x%lx/0x%lx/0x%lx\n",
1636 compat
, ro
, incompat
);
1638 sb
= journal
->j_superblock
;
1640 sb
->s_feature_compat
&= ~cpu_to_be32(compat
);
1641 sb
->s_feature_ro_compat
&= ~cpu_to_be32(ro
);
1642 sb
->s_feature_incompat
&= ~cpu_to_be32(incompat
);
1644 EXPORT_SYMBOL(jbd2_journal_clear_features
);
1647 * int jbd2_journal_update_format () - Update on-disk journal structure.
1648 * @journal: Journal to act on.
1650 * Given an initialised but unloaded journal struct, poke about in the
1651 * on-disk structure to update it to the most recent supported version.
1653 int jbd2_journal_update_format (journal_t
*journal
)
1655 journal_superblock_t
*sb
;
1658 err
= journal_get_superblock(journal
);
1662 sb
= journal
->j_superblock
;
1664 switch (be32_to_cpu(sb
->s_header
.h_blocktype
)) {
1665 case JBD2_SUPERBLOCK_V2
:
1667 case JBD2_SUPERBLOCK_V1
:
1668 return journal_convert_superblock_v1(journal
, sb
);
1675 static int journal_convert_superblock_v1(journal_t
*journal
,
1676 journal_superblock_t
*sb
)
1678 int offset
, blocksize
;
1679 struct buffer_head
*bh
;
1682 "JBD: Converting superblock from version 1 to 2.\n");
1684 /* Pre-initialise new fields to zero */
1685 offset
= ((char *) &(sb
->s_feature_compat
)) - ((char *) sb
);
1686 blocksize
= be32_to_cpu(sb
->s_blocksize
);
1687 memset(&sb
->s_feature_compat
, 0, blocksize
-offset
);
1689 sb
->s_nr_users
= cpu_to_be32(1);
1690 sb
->s_header
.h_blocktype
= cpu_to_be32(JBD2_SUPERBLOCK_V2
);
1691 journal
->j_format_version
= 2;
1693 bh
= journal
->j_sb_buffer
;
1694 BUFFER_TRACE(bh
, "marking dirty");
1695 mark_buffer_dirty(bh
);
1696 sync_dirty_buffer(bh
);
1702 * int jbd2_journal_flush () - Flush journal
1703 * @journal: Journal to act on.
1705 * Flush all data for a given journal to disk and empty the journal.
1706 * Filesystems can use this when remounting readonly to ensure that
1707 * recovery does not need to happen on remount.
1710 int jbd2_journal_flush(journal_t
*journal
)
1713 transaction_t
*transaction
= NULL
;
1714 unsigned long old_tail
;
1716 spin_lock(&journal
->j_state_lock
);
1718 /* Force everything buffered to the log... */
1719 if (journal
->j_running_transaction
) {
1720 transaction
= journal
->j_running_transaction
;
1721 __jbd2_log_start_commit(journal
, transaction
->t_tid
);
1722 } else if (journal
->j_committing_transaction
)
1723 transaction
= journal
->j_committing_transaction
;
1725 /* Wait for the log commit to complete... */
1727 tid_t tid
= transaction
->t_tid
;
1729 spin_unlock(&journal
->j_state_lock
);
1730 jbd2_log_wait_commit(journal
, tid
);
1732 spin_unlock(&journal
->j_state_lock
);
1735 /* ...and flush everything in the log out to disk. */
1736 spin_lock(&journal
->j_list_lock
);
1737 while (!err
&& journal
->j_checkpoint_transactions
!= NULL
) {
1738 spin_unlock(&journal
->j_list_lock
);
1739 mutex_lock(&journal
->j_checkpoint_mutex
);
1740 err
= jbd2_log_do_checkpoint(journal
);
1741 mutex_unlock(&journal
->j_checkpoint_mutex
);
1742 spin_lock(&journal
->j_list_lock
);
1744 spin_unlock(&journal
->j_list_lock
);
1746 if (is_journal_aborted(journal
))
1749 jbd2_cleanup_journal_tail(journal
);
1751 /* Finally, mark the journal as really needing no recovery.
1752 * This sets s_start==0 in the underlying superblock, which is
1753 * the magic code for a fully-recovered superblock. Any future
1754 * commits of data to the journal will restore the current
1756 spin_lock(&journal
->j_state_lock
);
1757 old_tail
= journal
->j_tail
;
1758 journal
->j_tail
= 0;
1759 spin_unlock(&journal
->j_state_lock
);
1760 jbd2_journal_update_superblock(journal
, 1);
1761 spin_lock(&journal
->j_state_lock
);
1762 journal
->j_tail
= old_tail
;
1764 J_ASSERT(!journal
->j_running_transaction
);
1765 J_ASSERT(!journal
->j_committing_transaction
);
1766 J_ASSERT(!journal
->j_checkpoint_transactions
);
1767 J_ASSERT(journal
->j_head
== journal
->j_tail
);
1768 J_ASSERT(journal
->j_tail_sequence
== journal
->j_transaction_sequence
);
1769 spin_unlock(&journal
->j_state_lock
);
1774 * int jbd2_journal_wipe() - Wipe journal contents
1775 * @journal: Journal to act on.
1776 * @write: flag (see below)
1778 * Wipe out all of the contents of a journal, safely. This will produce
1779 * a warning if the journal contains any valid recovery information.
1780 * Must be called between journal_init_*() and jbd2_journal_load().
1782 * If 'write' is non-zero, then we wipe out the journal on disk; otherwise
1783 * we merely suppress recovery.
1786 int jbd2_journal_wipe(journal_t
*journal
, int write
)
1788 journal_superblock_t
*sb
;
1791 J_ASSERT (!(journal
->j_flags
& JBD2_LOADED
));
1793 err
= load_superblock(journal
);
1797 sb
= journal
->j_superblock
;
1799 if (!journal
->j_tail
)
1802 printk (KERN_WARNING
"JBD: %s recovery information on journal\n",
1803 write
? "Clearing" : "Ignoring");
1805 err
= jbd2_journal_skip_recovery(journal
);
1807 jbd2_journal_update_superblock(journal
, 1);
1814 * Journal abort has very specific semantics, which we describe
1815 * for journal abort.
1817 * Two internal function, which provide abort to te jbd layer
1822 * Quick version for internal journal use (doesn't lock the journal).
1823 * Aborts hard --- we mark the abort as occurred, but do _nothing_ else,
1824 * and don't attempt to make any other journal updates.
1826 void __jbd2_journal_abort_hard(journal_t
*journal
)
1828 transaction_t
*transaction
;
1830 if (journal
->j_flags
& JBD2_ABORT
)
1833 printk(KERN_ERR
"Aborting journal on device %s.\n",
1834 journal
->j_devname
);
1836 spin_lock(&journal
->j_state_lock
);
1837 journal
->j_flags
|= JBD2_ABORT
;
1838 transaction
= journal
->j_running_transaction
;
1840 __jbd2_log_start_commit(journal
, transaction
->t_tid
);
1841 spin_unlock(&journal
->j_state_lock
);
1844 /* Soft abort: record the abort error status in the journal superblock,
1845 * but don't do any other IO. */
1846 static void __journal_abort_soft (journal_t
*journal
, int errno
)
1848 if (journal
->j_flags
& JBD2_ABORT
)
1851 if (!journal
->j_errno
)
1852 journal
->j_errno
= errno
;
1854 __jbd2_journal_abort_hard(journal
);
1857 jbd2_journal_update_superblock(journal
, 1);
1861 * void jbd2_journal_abort () - Shutdown the journal immediately.
1862 * @journal: the journal to shutdown.
1863 * @errno: an error number to record in the journal indicating
1864 * the reason for the shutdown.
1866 * Perform a complete, immediate shutdown of the ENTIRE
1867 * journal (not of a single transaction). This operation cannot be
1868 * undone without closing and reopening the journal.
1870 * The jbd2_journal_abort function is intended to support higher level error
1871 * recovery mechanisms such as the ext2/ext3 remount-readonly error
1874 * Journal abort has very specific semantics. Any existing dirty,
1875 * unjournaled buffers in the main filesystem will still be written to
1876 * disk by bdflush, but the journaling mechanism will be suspended
1877 * immediately and no further transaction commits will be honoured.
1879 * Any dirty, journaled buffers will be written back to disk without
1880 * hitting the journal. Atomicity cannot be guaranteed on an aborted
1881 * filesystem, but we _do_ attempt to leave as much data as possible
1882 * behind for fsck to use for cleanup.
1884 * Any attempt to get a new transaction handle on a journal which is in
1885 * ABORT state will just result in an -EROFS error return. A
1886 * jbd2_journal_stop on an existing handle will return -EIO if we have
1887 * entered abort state during the update.
1889 * Recursive transactions are not disturbed by journal abort until the
1890 * final jbd2_journal_stop, which will receive the -EIO error.
1892 * Finally, the jbd2_journal_abort call allows the caller to supply an errno
1893 * which will be recorded (if possible) in the journal superblock. This
1894 * allows a client to record failure conditions in the middle of a
1895 * transaction without having to complete the transaction to record the
1896 * failure to disk. ext3_error, for example, now uses this
1899 * Errors which originate from within the journaling layer will NOT
1900 * supply an errno; a null errno implies that absolutely no further
1901 * writes are done to the journal (unless there are any already in
1906 void jbd2_journal_abort(journal_t
*journal
, int errno
)
1908 __journal_abort_soft(journal
, errno
);
1912 * int jbd2_journal_errno () - returns the journal's error state.
1913 * @journal: journal to examine.
1915 * This is the errno numbet set with jbd2_journal_abort(), the last
1916 * time the journal was mounted - if the journal was stopped
1917 * without calling abort this will be 0.
1919 * If the journal has been aborted on this mount time -EROFS will
1922 int jbd2_journal_errno(journal_t
*journal
)
1926 spin_lock(&journal
->j_state_lock
);
1927 if (journal
->j_flags
& JBD2_ABORT
)
1930 err
= journal
->j_errno
;
1931 spin_unlock(&journal
->j_state_lock
);
1936 * int jbd2_journal_clear_err () - clears the journal's error state
1937 * @journal: journal to act on.
1939 * An error must be cleared or Acked to take a FS out of readonly
1942 int jbd2_journal_clear_err(journal_t
*journal
)
1946 spin_lock(&journal
->j_state_lock
);
1947 if (journal
->j_flags
& JBD2_ABORT
)
1950 journal
->j_errno
= 0;
1951 spin_unlock(&journal
->j_state_lock
);
1956 * void jbd2_journal_ack_err() - Ack journal err.
1957 * @journal: journal to act on.
1959 * An error must be cleared or Acked to take a FS out of readonly
1962 void jbd2_journal_ack_err(journal_t
*journal
)
1964 spin_lock(&journal
->j_state_lock
);
1965 if (journal
->j_errno
)
1966 journal
->j_flags
|= JBD2_ACK_ERR
;
1967 spin_unlock(&journal
->j_state_lock
);
1970 int jbd2_journal_blocks_per_page(struct inode
*inode
)
1972 return 1 << (PAGE_CACHE_SHIFT
- inode
->i_sb
->s_blocksize_bits
);
1976 * helper functions to deal with 32 or 64bit block numbers.
1978 size_t journal_tag_bytes(journal_t
*journal
)
1980 if (JBD2_HAS_INCOMPAT_FEATURE(journal
, JBD2_FEATURE_INCOMPAT_64BIT
))
1981 return JBD2_TAG_SIZE64
;
1983 return JBD2_TAG_SIZE32
;
1987 * Journal_head storage management
1989 static struct kmem_cache
*jbd2_journal_head_cache
;
1990 #ifdef CONFIG_JBD2_DEBUG
1991 static atomic_t nr_journal_heads
= ATOMIC_INIT(0);
1994 static int journal_init_jbd2_journal_head_cache(void)
1998 J_ASSERT(jbd2_journal_head_cache
== NULL
);
1999 jbd2_journal_head_cache
= kmem_cache_create("jbd2_journal_head",
2000 sizeof(struct journal_head
),
2002 SLAB_TEMPORARY
, /* flags */
2005 if (!jbd2_journal_head_cache
) {
2007 printk(KERN_EMERG
"JBD: no memory for journal_head cache\n");
2012 static void jbd2_journal_destroy_jbd2_journal_head_cache(void)
2014 if (jbd2_journal_head_cache
) {
2015 kmem_cache_destroy(jbd2_journal_head_cache
);
2016 jbd2_journal_head_cache
= NULL
;
2021 * journal_head splicing and dicing
2023 static struct journal_head
*journal_alloc_journal_head(void)
2025 struct journal_head
*ret
;
2026 static unsigned long last_warning
;
2028 #ifdef CONFIG_JBD2_DEBUG
2029 atomic_inc(&nr_journal_heads
);
2031 ret
= kmem_cache_alloc(jbd2_journal_head_cache
, GFP_NOFS
);
2033 jbd_debug(1, "out of memory for journal_head\n");
2034 if (time_after(jiffies
, last_warning
+ 5*HZ
)) {
2035 printk(KERN_NOTICE
"ENOMEM in %s, retrying.\n",
2037 last_warning
= jiffies
;
2041 ret
= kmem_cache_alloc(jbd2_journal_head_cache
, GFP_NOFS
);
2047 static void journal_free_journal_head(struct journal_head
*jh
)
2049 #ifdef CONFIG_JBD2_DEBUG
2050 atomic_dec(&nr_journal_heads
);
2051 memset(jh
, JBD2_POISON_FREE
, sizeof(*jh
));
2053 kmem_cache_free(jbd2_journal_head_cache
, jh
);
2057 * A journal_head is attached to a buffer_head whenever JBD has an
2058 * interest in the buffer.
2060 * Whenever a buffer has an attached journal_head, its ->b_state:BH_JBD bit
2061 * is set. This bit is tested in core kernel code where we need to take
2062 * JBD-specific actions. Testing the zeroness of ->b_private is not reliable
2065 * When a buffer has its BH_JBD bit set, its ->b_count is elevated by one.
2067 * When a buffer has its BH_JBD bit set it is immune from being released by
2068 * core kernel code, mainly via ->b_count.
2070 * A journal_head may be detached from its buffer_head when the journal_head's
2071 * b_transaction, b_cp_transaction and b_next_transaction pointers are NULL.
2072 * Various places in JBD call jbd2_journal_remove_journal_head() to indicate that the
2073 * journal_head can be dropped if needed.
2075 * Various places in the kernel want to attach a journal_head to a buffer_head
2076 * _before_ attaching the journal_head to a transaction. To protect the
2077 * journal_head in this situation, jbd2_journal_add_journal_head elevates the
2078 * journal_head's b_jcount refcount by one. The caller must call
2079 * jbd2_journal_put_journal_head() to undo this.
2081 * So the typical usage would be:
2083 * (Attach a journal_head if needed. Increments b_jcount)
2084 * struct journal_head *jh = jbd2_journal_add_journal_head(bh);
2086 * jh->b_transaction = xxx;
2087 * jbd2_journal_put_journal_head(jh);
2089 * Now, the journal_head's b_jcount is zero, but it is safe from being released
2090 * because it has a non-zero b_transaction.
2094 * Give a buffer_head a journal_head.
2096 * Doesn't need the journal lock.
2099 struct journal_head
*jbd2_journal_add_journal_head(struct buffer_head
*bh
)
2101 struct journal_head
*jh
;
2102 struct journal_head
*new_jh
= NULL
;
2105 if (!buffer_jbd(bh
)) {
2106 new_jh
= journal_alloc_journal_head();
2107 memset(new_jh
, 0, sizeof(*new_jh
));
2110 jbd_lock_bh_journal_head(bh
);
2111 if (buffer_jbd(bh
)) {
2115 (atomic_read(&bh
->b_count
) > 0) ||
2116 (bh
->b_page
&& bh
->b_page
->mapping
));
2119 jbd_unlock_bh_journal_head(bh
);
2124 new_jh
= NULL
; /* We consumed it */
2129 BUFFER_TRACE(bh
, "added journal_head");
2132 jbd_unlock_bh_journal_head(bh
);
2134 journal_free_journal_head(new_jh
);
2135 return bh
->b_private
;
2139 * Grab a ref against this buffer_head's journal_head. If it ended up not
2140 * having a journal_head, return NULL
2142 struct journal_head
*jbd2_journal_grab_journal_head(struct buffer_head
*bh
)
2144 struct journal_head
*jh
= NULL
;
2146 jbd_lock_bh_journal_head(bh
);
2147 if (buffer_jbd(bh
)) {
2151 jbd_unlock_bh_journal_head(bh
);
2155 static void __journal_remove_journal_head(struct buffer_head
*bh
)
2157 struct journal_head
*jh
= bh2jh(bh
);
2159 J_ASSERT_JH(jh
, jh
->b_jcount
>= 0);
2162 if (jh
->b_jcount
== 0) {
2163 if (jh
->b_transaction
== NULL
&&
2164 jh
->b_next_transaction
== NULL
&&
2165 jh
->b_cp_transaction
== NULL
) {
2166 J_ASSERT_JH(jh
, jh
->b_jlist
== BJ_None
);
2167 J_ASSERT_BH(bh
, buffer_jbd(bh
));
2168 J_ASSERT_BH(bh
, jh2bh(jh
) == bh
);
2169 BUFFER_TRACE(bh
, "remove journal_head");
2170 if (jh
->b_frozen_data
) {
2171 printk(KERN_WARNING
"%s: freeing "
2174 jbd2_free(jh
->b_frozen_data
, bh
->b_size
);
2176 if (jh
->b_committed_data
) {
2177 printk(KERN_WARNING
"%s: freeing "
2178 "b_committed_data\n",
2180 jbd2_free(jh
->b_committed_data
, bh
->b_size
);
2182 bh
->b_private
= NULL
;
2183 jh
->b_bh
= NULL
; /* debug, really */
2184 clear_buffer_jbd(bh
);
2186 journal_free_journal_head(jh
);
2188 BUFFER_TRACE(bh
, "journal_head was locked");
2194 * jbd2_journal_remove_journal_head(): if the buffer isn't attached to a transaction
2195 * and has a zero b_jcount then remove and release its journal_head. If we did
2196 * see that the buffer is not used by any transaction we also "logically"
2197 * decrement ->b_count.
2199 * We in fact take an additional increment on ->b_count as a convenience,
2200 * because the caller usually wants to do additional things with the bh
2201 * after calling here.
2202 * The caller of jbd2_journal_remove_journal_head() *must* run __brelse(bh) at some
2203 * time. Once the caller has run __brelse(), the buffer is eligible for
2204 * reaping by try_to_free_buffers().
2206 void jbd2_journal_remove_journal_head(struct buffer_head
*bh
)
2208 jbd_lock_bh_journal_head(bh
);
2209 __journal_remove_journal_head(bh
);
2210 jbd_unlock_bh_journal_head(bh
);
2214 * Drop a reference on the passed journal_head. If it fell to zero then try to
2215 * release the journal_head from the buffer_head.
2217 void jbd2_journal_put_journal_head(struct journal_head
*jh
)
2219 struct buffer_head
*bh
= jh2bh(jh
);
2221 jbd_lock_bh_journal_head(bh
);
2222 J_ASSERT_JH(jh
, jh
->b_jcount
> 0);
2224 if (!jh
->b_jcount
&& !jh
->b_transaction
) {
2225 __journal_remove_journal_head(bh
);
2228 jbd_unlock_bh_journal_head(bh
);
2232 * Initialize jbd inode head
2234 void jbd2_journal_init_jbd_inode(struct jbd2_inode
*jinode
, struct inode
*inode
)
2236 jinode
->i_transaction
= NULL
;
2237 jinode
->i_next_transaction
= NULL
;
2238 jinode
->i_vfs_inode
= inode
;
2239 jinode
->i_flags
= 0;
2240 INIT_LIST_HEAD(&jinode
->i_list
);
2244 * Function to be called before we start removing inode from memory (i.e.,
2245 * clear_inode() is a fine place to be called from). It removes inode from
2246 * transaction's lists.
2248 void jbd2_journal_release_jbd_inode(journal_t
*journal
,
2249 struct jbd2_inode
*jinode
)
2256 spin_lock(&journal
->j_list_lock
);
2257 /* Is commit writing out inode - we have to wait */
2258 if (jinode
->i_flags
& JI_COMMIT_RUNNING
) {
2259 wait_queue_head_t
*wq
;
2260 DEFINE_WAIT_BIT(wait
, &jinode
->i_flags
, __JI_COMMIT_RUNNING
);
2261 wq
= bit_waitqueue(&jinode
->i_flags
, __JI_COMMIT_RUNNING
);
2262 prepare_to_wait(wq
, &wait
.wait
, TASK_UNINTERRUPTIBLE
);
2263 spin_unlock(&journal
->j_list_lock
);
2265 finish_wait(wq
, &wait
.wait
);
2269 /* Do we need to wait for data writeback? */
2270 if (journal
->j_committing_transaction
== jinode
->i_transaction
)
2272 if (jinode
->i_transaction
) {
2273 list_del(&jinode
->i_list
);
2274 jinode
->i_transaction
= NULL
;
2276 spin_unlock(&journal
->j_list_lock
);
2282 #ifdef CONFIG_JBD2_DEBUG
2283 u8 jbd2_journal_enable_debug __read_mostly
;
2284 EXPORT_SYMBOL(jbd2_journal_enable_debug
);
2286 #define JBD2_DEBUG_NAME "jbd2-debug"
2288 static struct dentry
*jbd2_debugfs_dir
;
2289 static struct dentry
*jbd2_debug
;
2291 static void __init
jbd2_create_debugfs_entry(void)
2293 jbd2_debugfs_dir
= debugfs_create_dir("jbd2", NULL
);
2294 if (jbd2_debugfs_dir
)
2295 jbd2_debug
= debugfs_create_u8(JBD2_DEBUG_NAME
, S_IRUGO
,
2297 &jbd2_journal_enable_debug
);
2300 static void __exit
jbd2_remove_debugfs_entry(void)
2302 debugfs_remove(jbd2_debug
);
2303 debugfs_remove(jbd2_debugfs_dir
);
2308 static void __init
jbd2_create_debugfs_entry(void)
2312 static void __exit
jbd2_remove_debugfs_entry(void)
2318 #ifdef CONFIG_PROC_FS
2320 #define JBD2_STATS_PROC_NAME "fs/jbd2"
2322 static void __init
jbd2_create_jbd_stats_proc_entry(void)
2324 proc_jbd2_stats
= proc_mkdir(JBD2_STATS_PROC_NAME
, NULL
);
2327 static void __exit
jbd2_remove_jbd_stats_proc_entry(void)
2329 if (proc_jbd2_stats
)
2330 remove_proc_entry(JBD2_STATS_PROC_NAME
, NULL
);
2335 #define jbd2_create_jbd_stats_proc_entry() do {} while (0)
2336 #define jbd2_remove_jbd_stats_proc_entry() do {} while (0)
2340 struct kmem_cache
*jbd2_handle_cache
;
2342 static int __init
journal_init_handle_cache(void)
2344 jbd2_handle_cache
= kmem_cache_create("jbd2_journal_handle",
2347 SLAB_TEMPORARY
, /* flags */
2349 if (jbd2_handle_cache
== NULL
) {
2350 printk(KERN_EMERG
"JBD: failed to create handle cache\n");
2356 static void jbd2_journal_destroy_handle_cache(void)
2358 if (jbd2_handle_cache
)
2359 kmem_cache_destroy(jbd2_handle_cache
);
2363 * Module startup and shutdown
2366 static int __init
journal_init_caches(void)
2370 ret
= jbd2_journal_init_revoke_caches();
2372 ret
= journal_init_jbd2_journal_head_cache();
2374 ret
= journal_init_handle_cache();
2378 static void jbd2_journal_destroy_caches(void)
2380 jbd2_journal_destroy_revoke_caches();
2381 jbd2_journal_destroy_jbd2_journal_head_cache();
2382 jbd2_journal_destroy_handle_cache();
2385 static int __init
journal_init(void)
2389 BUILD_BUG_ON(sizeof(struct journal_superblock_s
) != 1024);
2391 ret
= journal_init_caches();
2393 jbd2_create_debugfs_entry();
2394 jbd2_create_jbd_stats_proc_entry();
2396 jbd2_journal_destroy_caches();
2401 static void __exit
journal_exit(void)
2403 #ifdef CONFIG_JBD2_DEBUG
2404 int n
= atomic_read(&nr_journal_heads
);
2406 printk(KERN_EMERG
"JBD: leaked %d journal_heads!\n", n
);
2408 jbd2_remove_debugfs_entry();
2409 jbd2_remove_jbd_stats_proc_entry();
2410 jbd2_journal_destroy_caches();
2413 MODULE_LICENSE("GPL");
2414 module_init(journal_init
);
2415 module_exit(journal_exit
);