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>
40 #include <linux/math64.h>
41 #include <linux/hash.h>
42 #include <linux/log2.h>
43 #include <linux/vmalloc.h>
44 #include <linux/backing-dev.h>
45 #include <linux/bitops.h>
46 #include <linux/ratelimit.h>
48 #define CREATE_TRACE_POINTS
49 #include <trace/events/jbd2.h>
51 #include <asm/uaccess.h>
53 #include <asm/system.h>
55 EXPORT_SYMBOL(jbd2_journal_extend
);
56 EXPORT_SYMBOL(jbd2_journal_stop
);
57 EXPORT_SYMBOL(jbd2_journal_lock_updates
);
58 EXPORT_SYMBOL(jbd2_journal_unlock_updates
);
59 EXPORT_SYMBOL(jbd2_journal_get_write_access
);
60 EXPORT_SYMBOL(jbd2_journal_get_create_access
);
61 EXPORT_SYMBOL(jbd2_journal_get_undo_access
);
62 EXPORT_SYMBOL(jbd2_journal_set_triggers
);
63 EXPORT_SYMBOL(jbd2_journal_dirty_metadata
);
64 EXPORT_SYMBOL(jbd2_journal_release_buffer
);
65 EXPORT_SYMBOL(jbd2_journal_forget
);
67 EXPORT_SYMBOL(journal_sync_buffer
);
69 EXPORT_SYMBOL(jbd2_journal_flush
);
70 EXPORT_SYMBOL(jbd2_journal_revoke
);
72 EXPORT_SYMBOL(jbd2_journal_init_dev
);
73 EXPORT_SYMBOL(jbd2_journal_init_inode
);
74 EXPORT_SYMBOL(jbd2_journal_update_format
);
75 EXPORT_SYMBOL(jbd2_journal_check_used_features
);
76 EXPORT_SYMBOL(jbd2_journal_check_available_features
);
77 EXPORT_SYMBOL(jbd2_journal_set_features
);
78 EXPORT_SYMBOL(jbd2_journal_load
);
79 EXPORT_SYMBOL(jbd2_journal_destroy
);
80 EXPORT_SYMBOL(jbd2_journal_abort
);
81 EXPORT_SYMBOL(jbd2_journal_errno
);
82 EXPORT_SYMBOL(jbd2_journal_ack_err
);
83 EXPORT_SYMBOL(jbd2_journal_clear_err
);
84 EXPORT_SYMBOL(jbd2_log_wait_commit
);
85 EXPORT_SYMBOL(jbd2_log_start_commit
);
86 EXPORT_SYMBOL(jbd2_journal_start_commit
);
87 EXPORT_SYMBOL(jbd2_journal_force_commit_nested
);
88 EXPORT_SYMBOL(jbd2_journal_wipe
);
89 EXPORT_SYMBOL(jbd2_journal_blocks_per_page
);
90 EXPORT_SYMBOL(jbd2_journal_invalidatepage
);
91 EXPORT_SYMBOL(jbd2_journal_try_to_free_buffers
);
92 EXPORT_SYMBOL(jbd2_journal_force_commit
);
93 EXPORT_SYMBOL(jbd2_journal_file_inode
);
94 EXPORT_SYMBOL(jbd2_journal_init_jbd_inode
);
95 EXPORT_SYMBOL(jbd2_journal_release_jbd_inode
);
96 EXPORT_SYMBOL(jbd2_journal_begin_ordered_truncate
);
97 EXPORT_SYMBOL(jbd2_inode_cache
);
99 static int journal_convert_superblock_v1(journal_t
*, journal_superblock_t
*);
100 static void __journal_abort_soft (journal_t
*journal
, int errno
);
101 static int jbd2_journal_create_slab(size_t slab_size
);
104 * Helper function used to manage commit timeouts
107 static void commit_timeout(unsigned long __data
)
109 struct task_struct
* p
= (struct task_struct
*) __data
;
115 * kjournald2: The main thread function used to manage a logging device
118 * This kernel thread is responsible for two things:
120 * 1) COMMIT: Every so often we need to commit the current state of the
121 * filesystem to disk. The journal thread is responsible for writing
122 * all of the metadata buffers to disk.
124 * 2) CHECKPOINT: We cannot reuse a used section of the log file until all
125 * of the data in that part of the log has been rewritten elsewhere on
126 * the disk. Flushing these old buffers to reclaim space in the log is
127 * known as checkpointing, and this thread is responsible for that job.
130 static int kjournald2(void *arg
)
132 journal_t
*journal
= arg
;
133 transaction_t
*transaction
;
136 * Set up an interval timer which can be used to trigger a commit wakeup
137 * after the commit interval expires
139 setup_timer(&journal
->j_commit_timer
, commit_timeout
,
140 (unsigned long)current
);
142 /* Record that the journal thread is running */
143 journal
->j_task
= current
;
144 wake_up(&journal
->j_wait_done_commit
);
147 * And now, wait forever for commit wakeup events.
149 write_lock(&journal
->j_state_lock
);
152 if (journal
->j_flags
& JBD2_UNMOUNT
)
155 jbd_debug(1, "commit_sequence=%d, commit_request=%d\n",
156 journal
->j_commit_sequence
, journal
->j_commit_request
);
158 if (journal
->j_commit_sequence
!= journal
->j_commit_request
) {
159 jbd_debug(1, "OK, requests differ\n");
160 write_unlock(&journal
->j_state_lock
);
161 del_timer_sync(&journal
->j_commit_timer
);
162 jbd2_journal_commit_transaction(journal
);
163 write_lock(&journal
->j_state_lock
);
167 wake_up(&journal
->j_wait_done_commit
);
168 if (freezing(current
)) {
170 * The simpler the better. Flushing journal isn't a
171 * good idea, because that depends on threads that may
172 * be already stopped.
174 jbd_debug(1, "Now suspending kjournald2\n");
175 write_unlock(&journal
->j_state_lock
);
177 write_lock(&journal
->j_state_lock
);
180 * We assume on resume that commits are already there,
184 int should_sleep
= 1;
186 prepare_to_wait(&journal
->j_wait_commit
, &wait
,
188 if (journal
->j_commit_sequence
!= journal
->j_commit_request
)
190 transaction
= journal
->j_running_transaction
;
191 if (transaction
&& time_after_eq(jiffies
,
192 transaction
->t_expires
))
194 if (journal
->j_flags
& JBD2_UNMOUNT
)
197 write_unlock(&journal
->j_state_lock
);
199 write_lock(&journal
->j_state_lock
);
201 finish_wait(&journal
->j_wait_commit
, &wait
);
204 jbd_debug(1, "kjournald2 wakes\n");
207 * Were we woken up by a commit wakeup event?
209 transaction
= journal
->j_running_transaction
;
210 if (transaction
&& time_after_eq(jiffies
, transaction
->t_expires
)) {
211 journal
->j_commit_request
= transaction
->t_tid
;
212 jbd_debug(1, "woke because of timeout\n");
217 write_unlock(&journal
->j_state_lock
);
218 del_timer_sync(&journal
->j_commit_timer
);
219 journal
->j_task
= NULL
;
220 wake_up(&journal
->j_wait_done_commit
);
221 jbd_debug(1, "Journal thread exiting.\n");
225 static int jbd2_journal_start_thread(journal_t
*journal
)
227 struct task_struct
*t
;
229 t
= kthread_run(kjournald2
, journal
, "jbd2/%s",
234 wait_event(journal
->j_wait_done_commit
, journal
->j_task
!= NULL
);
238 static void journal_kill_thread(journal_t
*journal
)
240 write_lock(&journal
->j_state_lock
);
241 journal
->j_flags
|= JBD2_UNMOUNT
;
243 while (journal
->j_task
) {
244 wake_up(&journal
->j_wait_commit
);
245 write_unlock(&journal
->j_state_lock
);
246 wait_event(journal
->j_wait_done_commit
, journal
->j_task
== NULL
);
247 write_lock(&journal
->j_state_lock
);
249 write_unlock(&journal
->j_state_lock
);
253 * jbd2_journal_write_metadata_buffer: write a metadata buffer to the journal.
255 * Writes a metadata buffer to a given disk block. The actual IO is not
256 * performed but a new buffer_head is constructed which labels the data
257 * to be written with the correct destination disk block.
259 * Any magic-number escaping which needs to be done will cause a
260 * copy-out here. If the buffer happens to start with the
261 * JBD2_MAGIC_NUMBER, then we can't write it to the log directly: the
262 * magic number is only written to the log for descripter blocks. In
263 * this case, we copy the data and replace the first word with 0, and we
264 * return a result code which indicates that this buffer needs to be
265 * marked as an escaped buffer in the corresponding log descriptor
266 * block. The missing word can then be restored when the block is read
269 * If the source buffer has already been modified by a new transaction
270 * since we took the last commit snapshot, we use the frozen copy of
271 * that data for IO. If we end up using the existing buffer_head's data
272 * for the write, then we *have* to lock the buffer to prevent anyone
273 * else from using and possibly modifying it while the IO is in
276 * The function returns a pointer to the buffer_heads to be used for IO.
278 * We assume that the journal has already been locked in this function.
285 * Bit 0 set == escape performed on the data
286 * Bit 1 set == buffer copy-out performed (kfree the data after IO)
289 int jbd2_journal_write_metadata_buffer(transaction_t
*transaction
,
290 struct journal_head
*jh_in
,
291 struct journal_head
**jh_out
,
292 unsigned long long blocknr
)
294 int need_copy_out
= 0;
295 int done_copy_out
= 0;
298 struct buffer_head
*new_bh
;
299 struct journal_head
*new_jh
;
300 struct page
*new_page
;
301 unsigned int new_offset
;
302 struct buffer_head
*bh_in
= jh2bh(jh_in
);
303 journal_t
*journal
= transaction
->t_journal
;
306 * The buffer really shouldn't be locked: only the current committing
307 * transaction is allowed to write it, so nobody else is allowed
310 * akpm: except if we're journalling data, and write() output is
311 * also part of a shared mapping, and another thread has
312 * decided to launch a writepage() against this buffer.
314 J_ASSERT_BH(bh_in
, buffer_jbddirty(bh_in
));
317 new_bh
= alloc_buffer_head(GFP_NOFS
);
320 * Failure is not an option, but __GFP_NOFAIL is going
321 * away; so we retry ourselves here.
323 congestion_wait(BLK_RW_ASYNC
, HZ
/50);
327 /* keep subsequent assertions sane */
329 init_buffer(new_bh
, NULL
, NULL
);
330 atomic_set(&new_bh
->b_count
, 1);
331 new_jh
= jbd2_journal_add_journal_head(new_bh
); /* This sleeps */
334 * If a new transaction has already done a buffer copy-out, then
335 * we use that version of the data for the commit.
337 jbd_lock_bh_state(bh_in
);
339 if (jh_in
->b_frozen_data
) {
341 new_page
= virt_to_page(jh_in
->b_frozen_data
);
342 new_offset
= offset_in_page(jh_in
->b_frozen_data
);
344 new_page
= jh2bh(jh_in
)->b_page
;
345 new_offset
= offset_in_page(jh2bh(jh_in
)->b_data
);
348 mapped_data
= kmap_atomic(new_page
, KM_USER0
);
350 * Fire data frozen trigger if data already wasn't frozen. Do this
351 * before checking for escaping, as the trigger may modify the magic
352 * offset. If a copy-out happens afterwards, it will have the correct
353 * data in the buffer.
356 jbd2_buffer_frozen_trigger(jh_in
, mapped_data
+ new_offset
,
362 if (*((__be32
*)(mapped_data
+ new_offset
)) ==
363 cpu_to_be32(JBD2_MAGIC_NUMBER
)) {
367 kunmap_atomic(mapped_data
, KM_USER0
);
370 * Do we need to do a data copy?
372 if (need_copy_out
&& !done_copy_out
) {
375 jbd_unlock_bh_state(bh_in
);
376 tmp
= jbd2_alloc(bh_in
->b_size
, GFP_NOFS
);
378 jbd2_journal_put_journal_head(new_jh
);
381 jbd_lock_bh_state(bh_in
);
382 if (jh_in
->b_frozen_data
) {
383 jbd2_free(tmp
, bh_in
->b_size
);
387 jh_in
->b_frozen_data
= tmp
;
388 mapped_data
= kmap_atomic(new_page
, KM_USER0
);
389 memcpy(tmp
, mapped_data
+ new_offset
, jh2bh(jh_in
)->b_size
);
390 kunmap_atomic(mapped_data
, KM_USER0
);
392 new_page
= virt_to_page(tmp
);
393 new_offset
= offset_in_page(tmp
);
397 * This isn't strictly necessary, as we're using frozen
398 * data for the escaping, but it keeps consistency with
399 * b_frozen_data usage.
401 jh_in
->b_frozen_triggers
= jh_in
->b_triggers
;
405 * Did we need to do an escaping? Now we've done all the
406 * copying, we can finally do so.
409 mapped_data
= kmap_atomic(new_page
, KM_USER0
);
410 *((unsigned int *)(mapped_data
+ new_offset
)) = 0;
411 kunmap_atomic(mapped_data
, KM_USER0
);
414 set_bh_page(new_bh
, new_page
, new_offset
);
415 new_jh
->b_transaction
= NULL
;
416 new_bh
->b_size
= jh2bh(jh_in
)->b_size
;
417 new_bh
->b_bdev
= transaction
->t_journal
->j_dev
;
418 new_bh
->b_blocknr
= blocknr
;
419 set_buffer_mapped(new_bh
);
420 set_buffer_dirty(new_bh
);
425 * The to-be-written buffer needs to get moved to the io queue,
426 * and the original buffer whose contents we are shadowing or
427 * copying is moved to the transaction's shadow queue.
429 JBUFFER_TRACE(jh_in
, "file as BJ_Shadow");
430 spin_lock(&journal
->j_list_lock
);
431 __jbd2_journal_file_buffer(jh_in
, transaction
, BJ_Shadow
);
432 spin_unlock(&journal
->j_list_lock
);
433 jbd_unlock_bh_state(bh_in
);
435 JBUFFER_TRACE(new_jh
, "file as BJ_IO");
436 jbd2_journal_file_buffer(new_jh
, transaction
, BJ_IO
);
438 return do_escape
| (done_copy_out
<< 1);
442 * Allocation code for the journal file. Manage the space left in the
443 * journal, so that we can begin checkpointing when appropriate.
447 * __jbd2_log_space_left: Return the number of free blocks left in the journal.
449 * Called with the journal already locked.
451 * Called under j_state_lock
454 int __jbd2_log_space_left(journal_t
*journal
)
456 int left
= journal
->j_free
;
458 /* assert_spin_locked(&journal->j_state_lock); */
461 * Be pessimistic here about the number of those free blocks which
462 * might be required for log descriptor control blocks.
465 #define MIN_LOG_RESERVED_BLOCKS 32 /* Allow for rounding errors */
467 left
-= MIN_LOG_RESERVED_BLOCKS
;
476 * Called with j_state_lock locked for writing.
477 * Returns true if a transaction commit was started.
479 int __jbd2_log_start_commit(journal_t
*journal
, tid_t target
)
482 * The only transaction we can possibly wait upon is the
483 * currently running transaction (if it exists). Otherwise,
484 * the target tid must be an old one.
486 if (journal
->j_running_transaction
&&
487 journal
->j_running_transaction
->t_tid
== target
) {
489 * We want a new commit: OK, mark the request and wakeup the
490 * commit thread. We do _not_ do the commit ourselves.
493 journal
->j_commit_request
= target
;
494 jbd_debug(1, "JBD: requesting commit %d/%d\n",
495 journal
->j_commit_request
,
496 journal
->j_commit_sequence
);
497 wake_up(&journal
->j_wait_commit
);
499 } else if (!tid_geq(journal
->j_commit_request
, target
))
500 /* This should never happen, but if it does, preserve
501 the evidence before kjournald goes into a loop and
502 increments j_commit_sequence beyond all recognition. */
503 WARN_ONCE(1, "jbd: bad log_start_commit: %u %u %u %u\n",
504 journal
->j_commit_request
,
505 journal
->j_commit_sequence
,
506 target
, journal
->j_running_transaction
?
507 journal
->j_running_transaction
->t_tid
: 0);
511 int jbd2_log_start_commit(journal_t
*journal
, tid_t tid
)
515 write_lock(&journal
->j_state_lock
);
516 ret
= __jbd2_log_start_commit(journal
, tid
);
517 write_unlock(&journal
->j_state_lock
);
522 * Force and wait upon a commit if the calling process is not within
523 * transaction. This is used for forcing out undo-protected data which contains
524 * bitmaps, when the fs is running out of space.
526 * We can only force the running transaction if we don't have an active handle;
527 * otherwise, we will deadlock.
529 * Returns true if a transaction was started.
531 int jbd2_journal_force_commit_nested(journal_t
*journal
)
533 transaction_t
*transaction
= NULL
;
535 int need_to_start
= 0;
537 read_lock(&journal
->j_state_lock
);
538 if (journal
->j_running_transaction
&& !current
->journal_info
) {
539 transaction
= journal
->j_running_transaction
;
540 if (!tid_geq(journal
->j_commit_request
, transaction
->t_tid
))
542 } else if (journal
->j_committing_transaction
)
543 transaction
= journal
->j_committing_transaction
;
546 read_unlock(&journal
->j_state_lock
);
547 return 0; /* Nothing to retry */
550 tid
= transaction
->t_tid
;
551 read_unlock(&journal
->j_state_lock
);
553 jbd2_log_start_commit(journal
, tid
);
554 jbd2_log_wait_commit(journal
, tid
);
559 * Start a commit of the current running transaction (if any). Returns true
560 * if a transaction is going to be committed (or is currently already
561 * committing), and fills its tid in at *ptid
563 int jbd2_journal_start_commit(journal_t
*journal
, tid_t
*ptid
)
567 write_lock(&journal
->j_state_lock
);
568 if (journal
->j_running_transaction
) {
569 tid_t tid
= journal
->j_running_transaction
->t_tid
;
571 __jbd2_log_start_commit(journal
, tid
);
572 /* There's a running transaction and we've just made sure
573 * it's commit has been scheduled. */
577 } else if (journal
->j_committing_transaction
) {
579 * If ext3_write_super() recently started a commit, then we
580 * have to wait for completion of that transaction
583 *ptid
= journal
->j_committing_transaction
->t_tid
;
586 write_unlock(&journal
->j_state_lock
);
591 * Wait for a specified commit to complete.
592 * The caller may not hold the journal lock.
594 int jbd2_log_wait_commit(journal_t
*journal
, tid_t tid
)
598 read_lock(&journal
->j_state_lock
);
599 #ifdef CONFIG_JBD2_DEBUG
600 if (!tid_geq(journal
->j_commit_request
, tid
)) {
602 "%s: error: j_commit_request=%d, tid=%d\n",
603 __func__
, journal
->j_commit_request
, tid
);
606 while (tid_gt(tid
, journal
->j_commit_sequence
)) {
607 jbd_debug(1, "JBD: want %d, j_commit_sequence=%d\n",
608 tid
, journal
->j_commit_sequence
);
609 wake_up(&journal
->j_wait_commit
);
610 read_unlock(&journal
->j_state_lock
);
611 wait_event(journal
->j_wait_done_commit
,
612 !tid_gt(tid
, journal
->j_commit_sequence
));
613 read_lock(&journal
->j_state_lock
);
615 read_unlock(&journal
->j_state_lock
);
617 if (unlikely(is_journal_aborted(journal
))) {
618 printk(KERN_EMERG
"journal commit I/O error\n");
625 * Log buffer allocation routines:
628 int jbd2_journal_next_log_block(journal_t
*journal
, unsigned long long *retp
)
630 unsigned long blocknr
;
632 write_lock(&journal
->j_state_lock
);
633 J_ASSERT(journal
->j_free
> 1);
635 blocknr
= journal
->j_head
;
638 if (journal
->j_head
== journal
->j_last
)
639 journal
->j_head
= journal
->j_first
;
640 write_unlock(&journal
->j_state_lock
);
641 return jbd2_journal_bmap(journal
, blocknr
, retp
);
645 * Conversion of logical to physical block numbers for the journal
647 * On external journals the journal blocks are identity-mapped, so
648 * this is a no-op. If needed, we can use j_blk_offset - everything is
651 int jbd2_journal_bmap(journal_t
*journal
, unsigned long blocknr
,
652 unsigned long long *retp
)
655 unsigned long long ret
;
657 if (journal
->j_inode
) {
658 ret
= bmap(journal
->j_inode
, blocknr
);
662 printk(KERN_ALERT
"%s: journal block not found "
663 "at offset %lu on %s\n",
664 __func__
, blocknr
, journal
->j_devname
);
666 __journal_abort_soft(journal
, err
);
669 *retp
= blocknr
; /* +journal->j_blk_offset */
675 * We play buffer_head aliasing tricks to write data/metadata blocks to
676 * the journal without copying their contents, but for journal
677 * descriptor blocks we do need to generate bona fide buffers.
679 * After the caller of jbd2_journal_get_descriptor_buffer() has finished modifying
680 * the buffer's contents they really should run flush_dcache_page(bh->b_page).
681 * But we don't bother doing that, so there will be coherency problems with
682 * mmaps of blockdevs which hold live JBD-controlled filesystems.
684 struct journal_head
*jbd2_journal_get_descriptor_buffer(journal_t
*journal
)
686 struct buffer_head
*bh
;
687 unsigned long long blocknr
;
690 err
= jbd2_journal_next_log_block(journal
, &blocknr
);
695 bh
= __getblk(journal
->j_dev
, blocknr
, journal
->j_blocksize
);
699 memset(bh
->b_data
, 0, journal
->j_blocksize
);
700 set_buffer_uptodate(bh
);
702 BUFFER_TRACE(bh
, "return this buffer");
703 return jbd2_journal_add_journal_head(bh
);
706 struct jbd2_stats_proc_session
{
708 struct transaction_stats_s
*stats
;
713 static void *jbd2_seq_info_start(struct seq_file
*seq
, loff_t
*pos
)
715 return *pos
? NULL
: SEQ_START_TOKEN
;
718 static void *jbd2_seq_info_next(struct seq_file
*seq
, void *v
, loff_t
*pos
)
723 static int jbd2_seq_info_show(struct seq_file
*seq
, void *v
)
725 struct jbd2_stats_proc_session
*s
= seq
->private;
727 if (v
!= SEQ_START_TOKEN
)
729 seq_printf(seq
, "%lu transaction, each up to %u blocks\n",
731 s
->journal
->j_max_transaction_buffers
);
732 if (s
->stats
->ts_tid
== 0)
734 seq_printf(seq
, "average: \n %ums waiting for transaction\n",
735 jiffies_to_msecs(s
->stats
->run
.rs_wait
/ s
->stats
->ts_tid
));
736 seq_printf(seq
, " %ums running transaction\n",
737 jiffies_to_msecs(s
->stats
->run
.rs_running
/ s
->stats
->ts_tid
));
738 seq_printf(seq
, " %ums transaction was being locked\n",
739 jiffies_to_msecs(s
->stats
->run
.rs_locked
/ s
->stats
->ts_tid
));
740 seq_printf(seq
, " %ums flushing data (in ordered mode)\n",
741 jiffies_to_msecs(s
->stats
->run
.rs_flushing
/ s
->stats
->ts_tid
));
742 seq_printf(seq
, " %ums logging transaction\n",
743 jiffies_to_msecs(s
->stats
->run
.rs_logging
/ s
->stats
->ts_tid
));
744 seq_printf(seq
, " %lluus average transaction commit time\n",
745 div_u64(s
->journal
->j_average_commit_time
, 1000));
746 seq_printf(seq
, " %lu handles per transaction\n",
747 s
->stats
->run
.rs_handle_count
/ s
->stats
->ts_tid
);
748 seq_printf(seq
, " %lu blocks per transaction\n",
749 s
->stats
->run
.rs_blocks
/ s
->stats
->ts_tid
);
750 seq_printf(seq
, " %lu logged blocks per transaction\n",
751 s
->stats
->run
.rs_blocks_logged
/ s
->stats
->ts_tid
);
755 static void jbd2_seq_info_stop(struct seq_file
*seq
, void *v
)
759 static const struct seq_operations jbd2_seq_info_ops
= {
760 .start
= jbd2_seq_info_start
,
761 .next
= jbd2_seq_info_next
,
762 .stop
= jbd2_seq_info_stop
,
763 .show
= jbd2_seq_info_show
,
766 static int jbd2_seq_info_open(struct inode
*inode
, struct file
*file
)
768 journal_t
*journal
= PDE(inode
)->data
;
769 struct jbd2_stats_proc_session
*s
;
772 s
= kmalloc(sizeof(*s
), GFP_KERNEL
);
775 size
= sizeof(struct transaction_stats_s
);
776 s
->stats
= kmalloc(size
, GFP_KERNEL
);
777 if (s
->stats
== NULL
) {
781 spin_lock(&journal
->j_history_lock
);
782 memcpy(s
->stats
, &journal
->j_stats
, size
);
783 s
->journal
= journal
;
784 spin_unlock(&journal
->j_history_lock
);
786 rc
= seq_open(file
, &jbd2_seq_info_ops
);
788 struct seq_file
*m
= file
->private_data
;
798 static int jbd2_seq_info_release(struct inode
*inode
, struct file
*file
)
800 struct seq_file
*seq
= file
->private_data
;
801 struct jbd2_stats_proc_session
*s
= seq
->private;
804 return seq_release(inode
, file
);
807 static const struct file_operations jbd2_seq_info_fops
= {
808 .owner
= THIS_MODULE
,
809 .open
= jbd2_seq_info_open
,
812 .release
= jbd2_seq_info_release
,
815 static struct proc_dir_entry
*proc_jbd2_stats
;
817 static void jbd2_stats_proc_init(journal_t
*journal
)
819 journal
->j_proc_entry
= proc_mkdir(journal
->j_devname
, proc_jbd2_stats
);
820 if (journal
->j_proc_entry
) {
821 proc_create_data("info", S_IRUGO
, journal
->j_proc_entry
,
822 &jbd2_seq_info_fops
, journal
);
826 static void jbd2_stats_proc_exit(journal_t
*journal
)
828 remove_proc_entry("info", journal
->j_proc_entry
);
829 remove_proc_entry(journal
->j_devname
, proc_jbd2_stats
);
833 * Management for journal control blocks: functions to create and
834 * destroy journal_t structures, and to initialise and read existing
835 * journal blocks from disk. */
837 /* First: create and setup a journal_t object in memory. We initialise
838 * very few fields yet: that has to wait until we have created the
839 * journal structures from from scratch, or loaded them from disk. */
841 static journal_t
* journal_init_common (void)
846 journal
= kzalloc(sizeof(*journal
), GFP_KERNEL
);
850 init_waitqueue_head(&journal
->j_wait_transaction_locked
);
851 init_waitqueue_head(&journal
->j_wait_logspace
);
852 init_waitqueue_head(&journal
->j_wait_done_commit
);
853 init_waitqueue_head(&journal
->j_wait_checkpoint
);
854 init_waitqueue_head(&journal
->j_wait_commit
);
855 init_waitqueue_head(&journal
->j_wait_updates
);
856 mutex_init(&journal
->j_barrier
);
857 mutex_init(&journal
->j_checkpoint_mutex
);
858 spin_lock_init(&journal
->j_revoke_lock
);
859 spin_lock_init(&journal
->j_list_lock
);
860 rwlock_init(&journal
->j_state_lock
);
862 journal
->j_commit_interval
= (HZ
* JBD2_DEFAULT_MAX_COMMIT_AGE
);
863 journal
->j_min_batch_time
= 0;
864 journal
->j_max_batch_time
= 15000; /* 15ms */
866 /* The journal is marked for error until we succeed with recovery! */
867 journal
->j_flags
= JBD2_ABORT
;
869 /* Set up a default-sized revoke table for the new mount. */
870 err
= jbd2_journal_init_revoke(journal
, JOURNAL_REVOKE_DEFAULT_HASH
);
876 spin_lock_init(&journal
->j_history_lock
);
881 /* jbd2_journal_init_dev and jbd2_journal_init_inode:
883 * Create a journal structure assigned some fixed set of disk blocks to
884 * the journal. We don't actually touch those disk blocks yet, but we
885 * need to set up all of the mapping information to tell the journaling
886 * system where the journal blocks are.
891 * journal_t * jbd2_journal_init_dev() - creates and initialises a journal structure
892 * @bdev: Block device on which to create the journal
893 * @fs_dev: Device which hold journalled filesystem for this journal.
894 * @start: Block nr Start of journal.
895 * @len: Length of the journal in blocks.
896 * @blocksize: blocksize of journalling device
898 * Returns: a newly created journal_t *
900 * jbd2_journal_init_dev creates a journal which maps a fixed contiguous
901 * range of blocks on an arbitrary block device.
904 journal_t
* jbd2_journal_init_dev(struct block_device
*bdev
,
905 struct block_device
*fs_dev
,
906 unsigned long long start
, int len
, int blocksize
)
908 journal_t
*journal
= journal_init_common();
909 struct buffer_head
*bh
;
916 /* journal descriptor can store up to n blocks -bzzz */
917 journal
->j_blocksize
= blocksize
;
918 journal
->j_dev
= bdev
;
919 journal
->j_fs_dev
= fs_dev
;
920 journal
->j_blk_offset
= start
;
921 journal
->j_maxlen
= len
;
922 bdevname(journal
->j_dev
, journal
->j_devname
);
923 p
= journal
->j_devname
;
924 while ((p
= strchr(p
, '/')))
926 jbd2_stats_proc_init(journal
);
927 n
= journal
->j_blocksize
/ sizeof(journal_block_tag_t
);
928 journal
->j_wbufsize
= n
;
929 journal
->j_wbuf
= kmalloc(n
* sizeof(struct buffer_head
*), GFP_KERNEL
);
930 if (!journal
->j_wbuf
) {
931 printk(KERN_ERR
"%s: Can't allocate bhs for commit thread\n",
936 bh
= __getblk(journal
->j_dev
, start
, journal
->j_blocksize
);
939 "%s: Cannot get buffer for journal superblock\n",
943 journal
->j_sb_buffer
= bh
;
944 journal
->j_superblock
= (journal_superblock_t
*)bh
->b_data
;
948 kfree(journal
->j_wbuf
);
949 jbd2_stats_proc_exit(journal
);
955 * journal_t * jbd2_journal_init_inode () - creates a journal which maps to a inode.
956 * @inode: An inode to create the journal in
958 * jbd2_journal_init_inode creates a journal which maps an on-disk inode as
959 * the journal. The inode must exist already, must support bmap() and
960 * must have all data blocks preallocated.
962 journal_t
* jbd2_journal_init_inode (struct inode
*inode
)
964 struct buffer_head
*bh
;
965 journal_t
*journal
= journal_init_common();
969 unsigned long long blocknr
;
974 journal
->j_dev
= journal
->j_fs_dev
= inode
->i_sb
->s_bdev
;
975 journal
->j_inode
= inode
;
976 bdevname(journal
->j_dev
, journal
->j_devname
);
977 p
= journal
->j_devname
;
978 while ((p
= strchr(p
, '/')))
980 p
= journal
->j_devname
+ strlen(journal
->j_devname
);
981 sprintf(p
, "-%lu", journal
->j_inode
->i_ino
);
983 "journal %p: inode %s/%ld, size %Ld, bits %d, blksize %ld\n",
984 journal
, inode
->i_sb
->s_id
, inode
->i_ino
,
985 (long long) inode
->i_size
,
986 inode
->i_sb
->s_blocksize_bits
, inode
->i_sb
->s_blocksize
);
988 journal
->j_maxlen
= inode
->i_size
>> inode
->i_sb
->s_blocksize_bits
;
989 journal
->j_blocksize
= inode
->i_sb
->s_blocksize
;
990 jbd2_stats_proc_init(journal
);
992 /* journal descriptor can store up to n blocks -bzzz */
993 n
= journal
->j_blocksize
/ sizeof(journal_block_tag_t
);
994 journal
->j_wbufsize
= n
;
995 journal
->j_wbuf
= kmalloc(n
* sizeof(struct buffer_head
*), GFP_KERNEL
);
996 if (!journal
->j_wbuf
) {
997 printk(KERN_ERR
"%s: Can't allocate bhs for commit thread\n",
1002 err
= jbd2_journal_bmap(journal
, 0, &blocknr
);
1003 /* If that failed, give up */
1005 printk(KERN_ERR
"%s: Cannot locate journal superblock\n",
1010 bh
= __getblk(journal
->j_dev
, blocknr
, journal
->j_blocksize
);
1013 "%s: Cannot get buffer for journal superblock\n",
1017 journal
->j_sb_buffer
= bh
;
1018 journal
->j_superblock
= (journal_superblock_t
*)bh
->b_data
;
1022 kfree(journal
->j_wbuf
);
1023 jbd2_stats_proc_exit(journal
);
1029 * If the journal init or create aborts, we need to mark the journal
1030 * superblock as being NULL to prevent the journal destroy from writing
1031 * back a bogus superblock.
1033 static void journal_fail_superblock (journal_t
*journal
)
1035 struct buffer_head
*bh
= journal
->j_sb_buffer
;
1037 journal
->j_sb_buffer
= NULL
;
1041 * Given a journal_t structure, initialise the various fields for
1042 * startup of a new journaling session. We use this both when creating
1043 * a journal, and after recovering an old journal to reset it for
1047 static int journal_reset(journal_t
*journal
)
1049 journal_superblock_t
*sb
= journal
->j_superblock
;
1050 unsigned long long first
, last
;
1052 first
= be32_to_cpu(sb
->s_first
);
1053 last
= be32_to_cpu(sb
->s_maxlen
);
1054 if (first
+ JBD2_MIN_JOURNAL_BLOCKS
> last
+ 1) {
1055 printk(KERN_ERR
"JBD: Journal too short (blocks %llu-%llu).\n",
1057 journal_fail_superblock(journal
);
1061 journal
->j_first
= first
;
1062 journal
->j_last
= last
;
1064 journal
->j_head
= first
;
1065 journal
->j_tail
= first
;
1066 journal
->j_free
= last
- first
;
1068 journal
->j_tail_sequence
= journal
->j_transaction_sequence
;
1069 journal
->j_commit_sequence
= journal
->j_transaction_sequence
- 1;
1070 journal
->j_commit_request
= journal
->j_commit_sequence
;
1072 journal
->j_max_transaction_buffers
= journal
->j_maxlen
/ 4;
1074 /* Add the dynamic fields and write it to disk. */
1075 jbd2_journal_update_superblock(journal
, 1);
1076 return jbd2_journal_start_thread(journal
);
1080 * void jbd2_journal_update_superblock() - Update journal sb on disk.
1081 * @journal: The journal to update.
1082 * @wait: Set to '0' if you don't want to wait for IO completion.
1084 * Update a journal's dynamic superblock fields and write it to disk,
1085 * optionally waiting for the IO to complete.
1087 void jbd2_journal_update_superblock(journal_t
*journal
, int wait
)
1089 journal_superblock_t
*sb
= journal
->j_superblock
;
1090 struct buffer_head
*bh
= journal
->j_sb_buffer
;
1093 * As a special case, if the on-disk copy is already marked as needing
1094 * no recovery (s_start == 0) and there are no outstanding transactions
1095 * in the filesystem, then we can safely defer the superblock update
1096 * until the next commit by setting JBD2_FLUSHED. This avoids
1097 * attempting a write to a potential-readonly device.
1099 if (sb
->s_start
== 0 && journal
->j_tail_sequence
==
1100 journal
->j_transaction_sequence
) {
1101 jbd_debug(1,"JBD: Skipping superblock update on recovered sb "
1102 "(start %ld, seq %d, errno %d)\n",
1103 journal
->j_tail
, journal
->j_tail_sequence
,
1108 if (buffer_write_io_error(bh
)) {
1110 * Oh, dear. A previous attempt to write the journal
1111 * superblock failed. This could happen because the
1112 * USB device was yanked out. Or it could happen to
1113 * be a transient write error and maybe the block will
1114 * be remapped. Nothing we can do but to retry the
1115 * write and hope for the best.
1117 printk(KERN_ERR
"JBD2: previous I/O error detected "
1118 "for journal superblock update for %s.\n",
1119 journal
->j_devname
);
1120 clear_buffer_write_io_error(bh
);
1121 set_buffer_uptodate(bh
);
1124 read_lock(&journal
->j_state_lock
);
1125 jbd_debug(1,"JBD: updating superblock (start %ld, seq %d, errno %d)\n",
1126 journal
->j_tail
, journal
->j_tail_sequence
, journal
->j_errno
);
1128 sb
->s_sequence
= cpu_to_be32(journal
->j_tail_sequence
);
1129 sb
->s_start
= cpu_to_be32(journal
->j_tail
);
1130 sb
->s_errno
= cpu_to_be32(journal
->j_errno
);
1131 read_unlock(&journal
->j_state_lock
);
1133 BUFFER_TRACE(bh
, "marking dirty");
1134 mark_buffer_dirty(bh
);
1136 sync_dirty_buffer(bh
);
1137 if (buffer_write_io_error(bh
)) {
1138 printk(KERN_ERR
"JBD2: I/O error detected "
1139 "when updating journal superblock for %s.\n",
1140 journal
->j_devname
);
1141 clear_buffer_write_io_error(bh
);
1142 set_buffer_uptodate(bh
);
1145 write_dirty_buffer(bh
, WRITE
);
1148 /* If we have just flushed the log (by marking s_start==0), then
1149 * any future commit will have to be careful to update the
1150 * superblock again to re-record the true start of the log. */
1152 write_lock(&journal
->j_state_lock
);
1154 journal
->j_flags
&= ~JBD2_FLUSHED
;
1156 journal
->j_flags
|= JBD2_FLUSHED
;
1157 write_unlock(&journal
->j_state_lock
);
1161 * Read the superblock for a given journal, performing initial
1162 * validation of the format.
1165 static int journal_get_superblock(journal_t
*journal
)
1167 struct buffer_head
*bh
;
1168 journal_superblock_t
*sb
;
1171 bh
= journal
->j_sb_buffer
;
1173 J_ASSERT(bh
!= NULL
);
1174 if (!buffer_uptodate(bh
)) {
1175 ll_rw_block(READ
, 1, &bh
);
1177 if (!buffer_uptodate(bh
)) {
1179 "JBD: IO error reading journal superblock\n");
1184 sb
= journal
->j_superblock
;
1188 if (sb
->s_header
.h_magic
!= cpu_to_be32(JBD2_MAGIC_NUMBER
) ||
1189 sb
->s_blocksize
!= cpu_to_be32(journal
->j_blocksize
)) {
1190 printk(KERN_WARNING
"JBD: no valid journal superblock found\n");
1194 switch(be32_to_cpu(sb
->s_header
.h_blocktype
)) {
1195 case JBD2_SUPERBLOCK_V1
:
1196 journal
->j_format_version
= 1;
1198 case JBD2_SUPERBLOCK_V2
:
1199 journal
->j_format_version
= 2;
1202 printk(KERN_WARNING
"JBD: unrecognised superblock format ID\n");
1206 if (be32_to_cpu(sb
->s_maxlen
) < journal
->j_maxlen
)
1207 journal
->j_maxlen
= be32_to_cpu(sb
->s_maxlen
);
1208 else if (be32_to_cpu(sb
->s_maxlen
) > journal
->j_maxlen
) {
1209 printk (KERN_WARNING
"JBD: journal file too short\n");
1216 journal_fail_superblock(journal
);
1221 * Load the on-disk journal superblock and read the key fields into the
1225 static int load_superblock(journal_t
*journal
)
1228 journal_superblock_t
*sb
;
1230 err
= journal_get_superblock(journal
);
1234 sb
= journal
->j_superblock
;
1236 journal
->j_tail_sequence
= be32_to_cpu(sb
->s_sequence
);
1237 journal
->j_tail
= be32_to_cpu(sb
->s_start
);
1238 journal
->j_first
= be32_to_cpu(sb
->s_first
);
1239 journal
->j_last
= be32_to_cpu(sb
->s_maxlen
);
1240 journal
->j_errno
= be32_to_cpu(sb
->s_errno
);
1247 * int jbd2_journal_load() - Read journal from disk.
1248 * @journal: Journal to act on.
1250 * Given a journal_t structure which tells us which disk blocks contain
1251 * a journal, read the journal from disk to initialise the in-memory
1254 int jbd2_journal_load(journal_t
*journal
)
1257 journal_superblock_t
*sb
;
1259 err
= load_superblock(journal
);
1263 sb
= journal
->j_superblock
;
1264 /* If this is a V2 superblock, then we have to check the
1265 * features flags on it. */
1267 if (journal
->j_format_version
>= 2) {
1268 if ((sb
->s_feature_ro_compat
&
1269 ~cpu_to_be32(JBD2_KNOWN_ROCOMPAT_FEATURES
)) ||
1270 (sb
->s_feature_incompat
&
1271 ~cpu_to_be32(JBD2_KNOWN_INCOMPAT_FEATURES
))) {
1272 printk (KERN_WARNING
1273 "JBD: Unrecognised features on journal\n");
1279 * Create a slab for this blocksize
1281 err
= jbd2_journal_create_slab(be32_to_cpu(sb
->s_blocksize
));
1285 /* Let the recovery code check whether it needs to recover any
1286 * data from the journal. */
1287 if (jbd2_journal_recover(journal
))
1288 goto recovery_error
;
1290 if (journal
->j_failed_commit
) {
1291 printk(KERN_ERR
"JBD2: journal transaction %u on %s "
1292 "is corrupt.\n", journal
->j_failed_commit
,
1293 journal
->j_devname
);
1297 /* OK, we've finished with the dynamic journal bits:
1298 * reinitialise the dynamic contents of the superblock in memory
1299 * and reset them on disk. */
1300 if (journal_reset(journal
))
1301 goto recovery_error
;
1303 journal
->j_flags
&= ~JBD2_ABORT
;
1304 journal
->j_flags
|= JBD2_LOADED
;
1308 printk (KERN_WARNING
"JBD: recovery failed\n");
1313 * void jbd2_journal_destroy() - Release a journal_t structure.
1314 * @journal: Journal to act on.
1316 * Release a journal_t structure once it is no longer in use by the
1318 * Return <0 if we couldn't clean up the journal.
1320 int jbd2_journal_destroy(journal_t
*journal
)
1324 /* Wait for the commit thread to wake up and die. */
1325 journal_kill_thread(journal
);
1327 /* Force a final log commit */
1328 if (journal
->j_running_transaction
)
1329 jbd2_journal_commit_transaction(journal
);
1331 /* Force any old transactions to disk */
1333 /* Totally anal locking here... */
1334 spin_lock(&journal
->j_list_lock
);
1335 while (journal
->j_checkpoint_transactions
!= NULL
) {
1336 spin_unlock(&journal
->j_list_lock
);
1337 mutex_lock(&journal
->j_checkpoint_mutex
);
1338 jbd2_log_do_checkpoint(journal
);
1339 mutex_unlock(&journal
->j_checkpoint_mutex
);
1340 spin_lock(&journal
->j_list_lock
);
1343 J_ASSERT(journal
->j_running_transaction
== NULL
);
1344 J_ASSERT(journal
->j_committing_transaction
== NULL
);
1345 J_ASSERT(journal
->j_checkpoint_transactions
== NULL
);
1346 spin_unlock(&journal
->j_list_lock
);
1348 if (journal
->j_sb_buffer
) {
1349 if (!is_journal_aborted(journal
)) {
1350 /* We can now mark the journal as empty. */
1351 journal
->j_tail
= 0;
1352 journal
->j_tail_sequence
=
1353 ++journal
->j_transaction_sequence
;
1354 jbd2_journal_update_superblock(journal
, 1);
1358 brelse(journal
->j_sb_buffer
);
1361 if (journal
->j_proc_entry
)
1362 jbd2_stats_proc_exit(journal
);
1363 if (journal
->j_inode
)
1364 iput(journal
->j_inode
);
1365 if (journal
->j_revoke
)
1366 jbd2_journal_destroy_revoke(journal
);
1367 kfree(journal
->j_wbuf
);
1375 *int jbd2_journal_check_used_features () - Check if features specified are used.
1376 * @journal: Journal to check.
1377 * @compat: bitmask of compatible features
1378 * @ro: bitmask of features that force read-only mount
1379 * @incompat: bitmask of incompatible features
1381 * Check whether the journal uses all of a given set of
1382 * features. Return true (non-zero) if it does.
1385 int jbd2_journal_check_used_features (journal_t
*journal
, unsigned long compat
,
1386 unsigned long ro
, unsigned long incompat
)
1388 journal_superblock_t
*sb
;
1390 if (!compat
&& !ro
&& !incompat
)
1392 /* Load journal superblock if it is not loaded yet. */
1393 if (journal
->j_format_version
== 0 &&
1394 journal_get_superblock(journal
) != 0)
1396 if (journal
->j_format_version
== 1)
1399 sb
= journal
->j_superblock
;
1401 if (((be32_to_cpu(sb
->s_feature_compat
) & compat
) == compat
) &&
1402 ((be32_to_cpu(sb
->s_feature_ro_compat
) & ro
) == ro
) &&
1403 ((be32_to_cpu(sb
->s_feature_incompat
) & incompat
) == incompat
))
1410 * int jbd2_journal_check_available_features() - Check feature set in journalling layer
1411 * @journal: Journal to check.
1412 * @compat: bitmask of compatible features
1413 * @ro: bitmask of features that force read-only mount
1414 * @incompat: bitmask of incompatible features
1416 * Check whether the journaling code supports the use of
1417 * all of a given set of features on this journal. Return true
1418 * (non-zero) if it can. */
1420 int jbd2_journal_check_available_features (journal_t
*journal
, unsigned long compat
,
1421 unsigned long ro
, unsigned long incompat
)
1423 if (!compat
&& !ro
&& !incompat
)
1426 /* We can support any known requested features iff the
1427 * superblock is in version 2. Otherwise we fail to support any
1428 * extended sb features. */
1430 if (journal
->j_format_version
!= 2)
1433 if ((compat
& JBD2_KNOWN_COMPAT_FEATURES
) == compat
&&
1434 (ro
& JBD2_KNOWN_ROCOMPAT_FEATURES
) == ro
&&
1435 (incompat
& JBD2_KNOWN_INCOMPAT_FEATURES
) == incompat
)
1442 * int jbd2_journal_set_features () - Mark a given journal feature in the superblock
1443 * @journal: Journal to act on.
1444 * @compat: bitmask of compatible features
1445 * @ro: bitmask of features that force read-only mount
1446 * @incompat: bitmask of incompatible features
1448 * Mark a given journal feature as present on the
1449 * superblock. Returns true if the requested features could be set.
1453 int jbd2_journal_set_features (journal_t
*journal
, unsigned long compat
,
1454 unsigned long ro
, unsigned long incompat
)
1456 journal_superblock_t
*sb
;
1458 if (jbd2_journal_check_used_features(journal
, compat
, ro
, incompat
))
1461 if (!jbd2_journal_check_available_features(journal
, compat
, ro
, incompat
))
1464 jbd_debug(1, "Setting new features 0x%lx/0x%lx/0x%lx\n",
1465 compat
, ro
, incompat
);
1467 sb
= journal
->j_superblock
;
1469 sb
->s_feature_compat
|= cpu_to_be32(compat
);
1470 sb
->s_feature_ro_compat
|= cpu_to_be32(ro
);
1471 sb
->s_feature_incompat
|= cpu_to_be32(incompat
);
1477 * jbd2_journal_clear_features () - Clear a given journal feature in the
1479 * @journal: Journal to act on.
1480 * @compat: bitmask of compatible features
1481 * @ro: bitmask of features that force read-only mount
1482 * @incompat: bitmask of incompatible features
1484 * Clear a given journal feature as present on the
1487 void jbd2_journal_clear_features(journal_t
*journal
, unsigned long compat
,
1488 unsigned long ro
, unsigned long incompat
)
1490 journal_superblock_t
*sb
;
1492 jbd_debug(1, "Clear features 0x%lx/0x%lx/0x%lx\n",
1493 compat
, ro
, incompat
);
1495 sb
= journal
->j_superblock
;
1497 sb
->s_feature_compat
&= ~cpu_to_be32(compat
);
1498 sb
->s_feature_ro_compat
&= ~cpu_to_be32(ro
);
1499 sb
->s_feature_incompat
&= ~cpu_to_be32(incompat
);
1501 EXPORT_SYMBOL(jbd2_journal_clear_features
);
1504 * int jbd2_journal_update_format () - Update on-disk journal structure.
1505 * @journal: Journal to act on.
1507 * Given an initialised but unloaded journal struct, poke about in the
1508 * on-disk structure to update it to the most recent supported version.
1510 int jbd2_journal_update_format (journal_t
*journal
)
1512 journal_superblock_t
*sb
;
1515 err
= journal_get_superblock(journal
);
1519 sb
= journal
->j_superblock
;
1521 switch (be32_to_cpu(sb
->s_header
.h_blocktype
)) {
1522 case JBD2_SUPERBLOCK_V2
:
1524 case JBD2_SUPERBLOCK_V1
:
1525 return journal_convert_superblock_v1(journal
, sb
);
1532 static int journal_convert_superblock_v1(journal_t
*journal
,
1533 journal_superblock_t
*sb
)
1535 int offset
, blocksize
;
1536 struct buffer_head
*bh
;
1539 "JBD: Converting superblock from version 1 to 2.\n");
1541 /* Pre-initialise new fields to zero */
1542 offset
= ((char *) &(sb
->s_feature_compat
)) - ((char *) sb
);
1543 blocksize
= be32_to_cpu(sb
->s_blocksize
);
1544 memset(&sb
->s_feature_compat
, 0, blocksize
-offset
);
1546 sb
->s_nr_users
= cpu_to_be32(1);
1547 sb
->s_header
.h_blocktype
= cpu_to_be32(JBD2_SUPERBLOCK_V2
);
1548 journal
->j_format_version
= 2;
1550 bh
= journal
->j_sb_buffer
;
1551 BUFFER_TRACE(bh
, "marking dirty");
1552 mark_buffer_dirty(bh
);
1553 sync_dirty_buffer(bh
);
1559 * int jbd2_journal_flush () - Flush journal
1560 * @journal: Journal to act on.
1562 * Flush all data for a given journal to disk and empty the journal.
1563 * Filesystems can use this when remounting readonly to ensure that
1564 * recovery does not need to happen on remount.
1567 int jbd2_journal_flush(journal_t
*journal
)
1570 transaction_t
*transaction
= NULL
;
1571 unsigned long old_tail
;
1573 write_lock(&journal
->j_state_lock
);
1575 /* Force everything buffered to the log... */
1576 if (journal
->j_running_transaction
) {
1577 transaction
= journal
->j_running_transaction
;
1578 __jbd2_log_start_commit(journal
, transaction
->t_tid
);
1579 } else if (journal
->j_committing_transaction
)
1580 transaction
= journal
->j_committing_transaction
;
1582 /* Wait for the log commit to complete... */
1584 tid_t tid
= transaction
->t_tid
;
1586 write_unlock(&journal
->j_state_lock
);
1587 jbd2_log_wait_commit(journal
, tid
);
1589 write_unlock(&journal
->j_state_lock
);
1592 /* ...and flush everything in the log out to disk. */
1593 spin_lock(&journal
->j_list_lock
);
1594 while (!err
&& journal
->j_checkpoint_transactions
!= NULL
) {
1595 spin_unlock(&journal
->j_list_lock
);
1596 mutex_lock(&journal
->j_checkpoint_mutex
);
1597 err
= jbd2_log_do_checkpoint(journal
);
1598 mutex_unlock(&journal
->j_checkpoint_mutex
);
1599 spin_lock(&journal
->j_list_lock
);
1601 spin_unlock(&journal
->j_list_lock
);
1603 if (is_journal_aborted(journal
))
1606 jbd2_cleanup_journal_tail(journal
);
1608 /* Finally, mark the journal as really needing no recovery.
1609 * This sets s_start==0 in the underlying superblock, which is
1610 * the magic code for a fully-recovered superblock. Any future
1611 * commits of data to the journal will restore the current
1613 write_lock(&journal
->j_state_lock
);
1614 old_tail
= journal
->j_tail
;
1615 journal
->j_tail
= 0;
1616 write_unlock(&journal
->j_state_lock
);
1617 jbd2_journal_update_superblock(journal
, 1);
1618 write_lock(&journal
->j_state_lock
);
1619 journal
->j_tail
= old_tail
;
1621 J_ASSERT(!journal
->j_running_transaction
);
1622 J_ASSERT(!journal
->j_committing_transaction
);
1623 J_ASSERT(!journal
->j_checkpoint_transactions
);
1624 J_ASSERT(journal
->j_head
== journal
->j_tail
);
1625 J_ASSERT(journal
->j_tail_sequence
== journal
->j_transaction_sequence
);
1626 write_unlock(&journal
->j_state_lock
);
1631 * int jbd2_journal_wipe() - Wipe journal contents
1632 * @journal: Journal to act on.
1633 * @write: flag (see below)
1635 * Wipe out all of the contents of a journal, safely. This will produce
1636 * a warning if the journal contains any valid recovery information.
1637 * Must be called between journal_init_*() and jbd2_journal_load().
1639 * If 'write' is non-zero, then we wipe out the journal on disk; otherwise
1640 * we merely suppress recovery.
1643 int jbd2_journal_wipe(journal_t
*journal
, int write
)
1647 J_ASSERT (!(journal
->j_flags
& JBD2_LOADED
));
1649 err
= load_superblock(journal
);
1653 if (!journal
->j_tail
)
1656 printk (KERN_WARNING
"JBD: %s recovery information on journal\n",
1657 write
? "Clearing" : "Ignoring");
1659 err
= jbd2_journal_skip_recovery(journal
);
1661 jbd2_journal_update_superblock(journal
, 1);
1668 * Journal abort has very specific semantics, which we describe
1669 * for journal abort.
1671 * Two internal functions, which provide abort to the jbd layer
1676 * Quick version for internal journal use (doesn't lock the journal).
1677 * Aborts hard --- we mark the abort as occurred, but do _nothing_ else,
1678 * and don't attempt to make any other journal updates.
1680 void __jbd2_journal_abort_hard(journal_t
*journal
)
1682 transaction_t
*transaction
;
1684 if (journal
->j_flags
& JBD2_ABORT
)
1687 printk(KERN_ERR
"Aborting journal on device %s.\n",
1688 journal
->j_devname
);
1690 write_lock(&journal
->j_state_lock
);
1691 journal
->j_flags
|= JBD2_ABORT
;
1692 transaction
= journal
->j_running_transaction
;
1694 __jbd2_log_start_commit(journal
, transaction
->t_tid
);
1695 write_unlock(&journal
->j_state_lock
);
1698 /* Soft abort: record the abort error status in the journal superblock,
1699 * but don't do any other IO. */
1700 static void __journal_abort_soft (journal_t
*journal
, int errno
)
1702 if (journal
->j_flags
& JBD2_ABORT
)
1705 if (!journal
->j_errno
)
1706 journal
->j_errno
= errno
;
1708 __jbd2_journal_abort_hard(journal
);
1711 jbd2_journal_update_superblock(journal
, 1);
1715 * void jbd2_journal_abort () - Shutdown the journal immediately.
1716 * @journal: the journal to shutdown.
1717 * @errno: an error number to record in the journal indicating
1718 * the reason for the shutdown.
1720 * Perform a complete, immediate shutdown of the ENTIRE
1721 * journal (not of a single transaction). This operation cannot be
1722 * undone without closing and reopening the journal.
1724 * The jbd2_journal_abort function is intended to support higher level error
1725 * recovery mechanisms such as the ext2/ext3 remount-readonly error
1728 * Journal abort has very specific semantics. Any existing dirty,
1729 * unjournaled buffers in the main filesystem will still be written to
1730 * disk by bdflush, but the journaling mechanism will be suspended
1731 * immediately and no further transaction commits will be honoured.
1733 * Any dirty, journaled buffers will be written back to disk without
1734 * hitting the journal. Atomicity cannot be guaranteed on an aborted
1735 * filesystem, but we _do_ attempt to leave as much data as possible
1736 * behind for fsck to use for cleanup.
1738 * Any attempt to get a new transaction handle on a journal which is in
1739 * ABORT state will just result in an -EROFS error return. A
1740 * jbd2_journal_stop on an existing handle will return -EIO if we have
1741 * entered abort state during the update.
1743 * Recursive transactions are not disturbed by journal abort until the
1744 * final jbd2_journal_stop, which will receive the -EIO error.
1746 * Finally, the jbd2_journal_abort call allows the caller to supply an errno
1747 * which will be recorded (if possible) in the journal superblock. This
1748 * allows a client to record failure conditions in the middle of a
1749 * transaction without having to complete the transaction to record the
1750 * failure to disk. ext3_error, for example, now uses this
1753 * Errors which originate from within the journaling layer will NOT
1754 * supply an errno; a null errno implies that absolutely no further
1755 * writes are done to the journal (unless there are any already in
1760 void jbd2_journal_abort(journal_t
*journal
, int errno
)
1762 __journal_abort_soft(journal
, errno
);
1766 * int jbd2_journal_errno () - returns the journal's error state.
1767 * @journal: journal to examine.
1769 * This is the errno number set with jbd2_journal_abort(), the last
1770 * time the journal was mounted - if the journal was stopped
1771 * without calling abort this will be 0.
1773 * If the journal has been aborted on this mount time -EROFS will
1776 int jbd2_journal_errno(journal_t
*journal
)
1780 read_lock(&journal
->j_state_lock
);
1781 if (journal
->j_flags
& JBD2_ABORT
)
1784 err
= journal
->j_errno
;
1785 read_unlock(&journal
->j_state_lock
);
1790 * int jbd2_journal_clear_err () - clears the journal's error state
1791 * @journal: journal to act on.
1793 * An error must be cleared or acked to take a FS out of readonly
1796 int jbd2_journal_clear_err(journal_t
*journal
)
1800 write_lock(&journal
->j_state_lock
);
1801 if (journal
->j_flags
& JBD2_ABORT
)
1804 journal
->j_errno
= 0;
1805 write_unlock(&journal
->j_state_lock
);
1810 * void jbd2_journal_ack_err() - Ack journal err.
1811 * @journal: journal to act on.
1813 * An error must be cleared or acked to take a FS out of readonly
1816 void jbd2_journal_ack_err(journal_t
*journal
)
1818 write_lock(&journal
->j_state_lock
);
1819 if (journal
->j_errno
)
1820 journal
->j_flags
|= JBD2_ACK_ERR
;
1821 write_unlock(&journal
->j_state_lock
);
1824 int jbd2_journal_blocks_per_page(struct inode
*inode
)
1826 return 1 << (PAGE_CACHE_SHIFT
- inode
->i_sb
->s_blocksize_bits
);
1830 * helper functions to deal with 32 or 64bit block numbers.
1832 size_t journal_tag_bytes(journal_t
*journal
)
1834 if (JBD2_HAS_INCOMPAT_FEATURE(journal
, JBD2_FEATURE_INCOMPAT_64BIT
))
1835 return JBD2_TAG_SIZE64
;
1837 return JBD2_TAG_SIZE32
;
1841 * JBD memory management
1843 * These functions are used to allocate block-sized chunks of memory
1844 * used for making copies of buffer_head data. Very often it will be
1845 * page-sized chunks of data, but sometimes it will be in
1846 * sub-page-size chunks. (For example, 16k pages on Power systems
1847 * with a 4k block file system.) For blocks smaller than a page, we
1848 * use a SLAB allocator. There are slab caches for each block size,
1849 * which are allocated at mount time, if necessary, and we only free
1850 * (all of) the slab caches when/if the jbd2 module is unloaded. For
1851 * this reason we don't need to a mutex to protect access to
1852 * jbd2_slab[] allocating or releasing memory; only in
1853 * jbd2_journal_create_slab().
1855 #define JBD2_MAX_SLABS 8
1856 static struct kmem_cache
*jbd2_slab
[JBD2_MAX_SLABS
];
1858 static const char *jbd2_slab_names
[JBD2_MAX_SLABS
] = {
1859 "jbd2_1k", "jbd2_2k", "jbd2_4k", "jbd2_8k",
1860 "jbd2_16k", "jbd2_32k", "jbd2_64k", "jbd2_128k"
1864 static void jbd2_journal_destroy_slabs(void)
1868 for (i
= 0; i
< JBD2_MAX_SLABS
; i
++) {
1870 kmem_cache_destroy(jbd2_slab
[i
]);
1871 jbd2_slab
[i
] = NULL
;
1875 static int jbd2_journal_create_slab(size_t size
)
1877 static DEFINE_MUTEX(jbd2_slab_create_mutex
);
1878 int i
= order_base_2(size
) - 10;
1881 if (size
== PAGE_SIZE
)
1884 if (i
>= JBD2_MAX_SLABS
)
1887 if (unlikely(i
< 0))
1889 mutex_lock(&jbd2_slab_create_mutex
);
1891 mutex_unlock(&jbd2_slab_create_mutex
);
1892 return 0; /* Already created */
1895 slab_size
= 1 << (i
+10);
1896 jbd2_slab
[i
] = kmem_cache_create(jbd2_slab_names
[i
], slab_size
,
1897 slab_size
, 0, NULL
);
1898 mutex_unlock(&jbd2_slab_create_mutex
);
1899 if (!jbd2_slab
[i
]) {
1900 printk(KERN_EMERG
"JBD2: no memory for jbd2_slab cache\n");
1906 static struct kmem_cache
*get_slab(size_t size
)
1908 int i
= order_base_2(size
) - 10;
1910 BUG_ON(i
>= JBD2_MAX_SLABS
);
1911 if (unlikely(i
< 0))
1913 BUG_ON(jbd2_slab
[i
] == NULL
);
1914 return jbd2_slab
[i
];
1917 void *jbd2_alloc(size_t size
, gfp_t flags
)
1921 BUG_ON(size
& (size
-1)); /* Must be a power of 2 */
1923 flags
|= __GFP_REPEAT
;
1924 if (size
== PAGE_SIZE
)
1925 ptr
= (void *)__get_free_pages(flags
, 0);
1926 else if (size
> PAGE_SIZE
) {
1927 int order
= get_order(size
);
1930 ptr
= (void *)__get_free_pages(flags
, order
);
1932 ptr
= vmalloc(size
);
1934 ptr
= kmem_cache_alloc(get_slab(size
), flags
);
1936 /* Check alignment; SLUB has gotten this wrong in the past,
1937 * and this can lead to user data corruption! */
1938 BUG_ON(((unsigned long) ptr
) & (size
-1));
1943 void jbd2_free(void *ptr
, size_t size
)
1945 if (size
== PAGE_SIZE
) {
1946 free_pages((unsigned long)ptr
, 0);
1949 if (size
> PAGE_SIZE
) {
1950 int order
= get_order(size
);
1953 free_pages((unsigned long)ptr
, order
);
1958 kmem_cache_free(get_slab(size
), ptr
);
1962 * Journal_head storage management
1964 static struct kmem_cache
*jbd2_journal_head_cache
;
1965 #ifdef CONFIG_JBD2_DEBUG
1966 static atomic_t nr_journal_heads
= ATOMIC_INIT(0);
1969 static int journal_init_jbd2_journal_head_cache(void)
1973 J_ASSERT(jbd2_journal_head_cache
== NULL
);
1974 jbd2_journal_head_cache
= kmem_cache_create("jbd2_journal_head",
1975 sizeof(struct journal_head
),
1977 SLAB_TEMPORARY
, /* flags */
1980 if (!jbd2_journal_head_cache
) {
1982 printk(KERN_EMERG
"JBD: no memory for journal_head cache\n");
1987 static void jbd2_journal_destroy_jbd2_journal_head_cache(void)
1989 if (jbd2_journal_head_cache
) {
1990 kmem_cache_destroy(jbd2_journal_head_cache
);
1991 jbd2_journal_head_cache
= NULL
;
1996 * journal_head splicing and dicing
1998 static struct journal_head
*journal_alloc_journal_head(void)
2000 struct journal_head
*ret
;
2002 #ifdef CONFIG_JBD2_DEBUG
2003 atomic_inc(&nr_journal_heads
);
2005 ret
= kmem_cache_alloc(jbd2_journal_head_cache
, GFP_NOFS
);
2007 jbd_debug(1, "out of memory for journal_head\n");
2008 pr_notice_ratelimited("ENOMEM in %s, retrying.\n", __func__
);
2011 ret
= kmem_cache_alloc(jbd2_journal_head_cache
, GFP_NOFS
);
2017 static void journal_free_journal_head(struct journal_head
*jh
)
2019 #ifdef CONFIG_JBD2_DEBUG
2020 atomic_dec(&nr_journal_heads
);
2021 memset(jh
, JBD2_POISON_FREE
, sizeof(*jh
));
2023 kmem_cache_free(jbd2_journal_head_cache
, jh
);
2027 * A journal_head is attached to a buffer_head whenever JBD has an
2028 * interest in the buffer.
2030 * Whenever a buffer has an attached journal_head, its ->b_state:BH_JBD bit
2031 * is set. This bit is tested in core kernel code where we need to take
2032 * JBD-specific actions. Testing the zeroness of ->b_private is not reliable
2035 * When a buffer has its BH_JBD bit set, its ->b_count is elevated by one.
2037 * When a buffer has its BH_JBD bit set it is immune from being released by
2038 * core kernel code, mainly via ->b_count.
2040 * A journal_head may be detached from its buffer_head when the journal_head's
2041 * b_transaction, b_cp_transaction and b_next_transaction pointers are NULL.
2042 * Various places in JBD call jbd2_journal_remove_journal_head() to indicate that the
2043 * journal_head can be dropped if needed.
2045 * Various places in the kernel want to attach a journal_head to a buffer_head
2046 * _before_ attaching the journal_head to a transaction. To protect the
2047 * journal_head in this situation, jbd2_journal_add_journal_head elevates the
2048 * journal_head's b_jcount refcount by one. The caller must call
2049 * jbd2_journal_put_journal_head() to undo this.
2051 * So the typical usage would be:
2053 * (Attach a journal_head if needed. Increments b_jcount)
2054 * struct journal_head *jh = jbd2_journal_add_journal_head(bh);
2056 * jh->b_transaction = xxx;
2057 * jbd2_journal_put_journal_head(jh);
2059 * Now, the journal_head's b_jcount is zero, but it is safe from being released
2060 * because it has a non-zero b_transaction.
2064 * Give a buffer_head a journal_head.
2066 * Doesn't need the journal lock.
2069 struct journal_head
*jbd2_journal_add_journal_head(struct buffer_head
*bh
)
2071 struct journal_head
*jh
;
2072 struct journal_head
*new_jh
= NULL
;
2075 if (!buffer_jbd(bh
)) {
2076 new_jh
= journal_alloc_journal_head();
2077 memset(new_jh
, 0, sizeof(*new_jh
));
2080 jbd_lock_bh_journal_head(bh
);
2081 if (buffer_jbd(bh
)) {
2085 (atomic_read(&bh
->b_count
) > 0) ||
2086 (bh
->b_page
&& bh
->b_page
->mapping
));
2089 jbd_unlock_bh_journal_head(bh
);
2094 new_jh
= NULL
; /* We consumed it */
2099 BUFFER_TRACE(bh
, "added journal_head");
2102 jbd_unlock_bh_journal_head(bh
);
2104 journal_free_journal_head(new_jh
);
2105 return bh
->b_private
;
2109 * Grab a ref against this buffer_head's journal_head. If it ended up not
2110 * having a journal_head, return NULL
2112 struct journal_head
*jbd2_journal_grab_journal_head(struct buffer_head
*bh
)
2114 struct journal_head
*jh
= NULL
;
2116 jbd_lock_bh_journal_head(bh
);
2117 if (buffer_jbd(bh
)) {
2121 jbd_unlock_bh_journal_head(bh
);
2125 static void __journal_remove_journal_head(struct buffer_head
*bh
)
2127 struct journal_head
*jh
= bh2jh(bh
);
2129 J_ASSERT_JH(jh
, jh
->b_jcount
>= 0);
2132 if (jh
->b_jcount
== 0) {
2133 if (jh
->b_transaction
== NULL
&&
2134 jh
->b_next_transaction
== NULL
&&
2135 jh
->b_cp_transaction
== NULL
) {
2136 J_ASSERT_JH(jh
, jh
->b_jlist
== BJ_None
);
2137 J_ASSERT_BH(bh
, buffer_jbd(bh
));
2138 J_ASSERT_BH(bh
, jh2bh(jh
) == bh
);
2139 BUFFER_TRACE(bh
, "remove journal_head");
2140 if (jh
->b_frozen_data
) {
2141 printk(KERN_WARNING
"%s: freeing "
2144 jbd2_free(jh
->b_frozen_data
, bh
->b_size
);
2146 if (jh
->b_committed_data
) {
2147 printk(KERN_WARNING
"%s: freeing "
2148 "b_committed_data\n",
2150 jbd2_free(jh
->b_committed_data
, bh
->b_size
);
2152 bh
->b_private
= NULL
;
2153 jh
->b_bh
= NULL
; /* debug, really */
2154 clear_buffer_jbd(bh
);
2156 journal_free_journal_head(jh
);
2158 BUFFER_TRACE(bh
, "journal_head was locked");
2164 * jbd2_journal_remove_journal_head(): if the buffer isn't attached to a transaction
2165 * and has a zero b_jcount then remove and release its journal_head. If we did
2166 * see that the buffer is not used by any transaction we also "logically"
2167 * decrement ->b_count.
2169 * We in fact take an additional increment on ->b_count as a convenience,
2170 * because the caller usually wants to do additional things with the bh
2171 * after calling here.
2172 * The caller of jbd2_journal_remove_journal_head() *must* run __brelse(bh) at some
2173 * time. Once the caller has run __brelse(), the buffer is eligible for
2174 * reaping by try_to_free_buffers().
2176 void jbd2_journal_remove_journal_head(struct buffer_head
*bh
)
2178 jbd_lock_bh_journal_head(bh
);
2179 __journal_remove_journal_head(bh
);
2180 jbd_unlock_bh_journal_head(bh
);
2184 * Drop a reference on the passed journal_head. If it fell to zero then try to
2185 * release the journal_head from the buffer_head.
2187 void jbd2_journal_put_journal_head(struct journal_head
*jh
)
2189 struct buffer_head
*bh
= jh2bh(jh
);
2191 jbd_lock_bh_journal_head(bh
);
2192 J_ASSERT_JH(jh
, jh
->b_jcount
> 0);
2194 if (!jh
->b_jcount
&& !jh
->b_transaction
) {
2195 __journal_remove_journal_head(bh
);
2198 jbd_unlock_bh_journal_head(bh
);
2202 * Initialize jbd inode head
2204 void jbd2_journal_init_jbd_inode(struct jbd2_inode
*jinode
, struct inode
*inode
)
2206 jinode
->i_transaction
= NULL
;
2207 jinode
->i_next_transaction
= NULL
;
2208 jinode
->i_vfs_inode
= inode
;
2209 jinode
->i_flags
= 0;
2210 INIT_LIST_HEAD(&jinode
->i_list
);
2214 * Function to be called before we start removing inode from memory (i.e.,
2215 * clear_inode() is a fine place to be called from). It removes inode from
2216 * transaction's lists.
2218 void jbd2_journal_release_jbd_inode(journal_t
*journal
,
2219 struct jbd2_inode
*jinode
)
2224 spin_lock(&journal
->j_list_lock
);
2225 /* Is commit writing out inode - we have to wait */
2226 if (test_bit(__JI_COMMIT_RUNNING
, &jinode
->i_flags
)) {
2227 wait_queue_head_t
*wq
;
2228 DEFINE_WAIT_BIT(wait
, &jinode
->i_flags
, __JI_COMMIT_RUNNING
);
2229 wq
= bit_waitqueue(&jinode
->i_flags
, __JI_COMMIT_RUNNING
);
2230 prepare_to_wait(wq
, &wait
.wait
, TASK_UNINTERRUPTIBLE
);
2231 spin_unlock(&journal
->j_list_lock
);
2233 finish_wait(wq
, &wait
.wait
);
2237 if (jinode
->i_transaction
) {
2238 list_del(&jinode
->i_list
);
2239 jinode
->i_transaction
= NULL
;
2241 spin_unlock(&journal
->j_list_lock
);
2247 #ifdef CONFIG_JBD2_DEBUG
2248 u8 jbd2_journal_enable_debug __read_mostly
;
2249 EXPORT_SYMBOL(jbd2_journal_enable_debug
);
2251 #define JBD2_DEBUG_NAME "jbd2-debug"
2253 static struct dentry
*jbd2_debugfs_dir
;
2254 static struct dentry
*jbd2_debug
;
2256 static void __init
jbd2_create_debugfs_entry(void)
2258 jbd2_debugfs_dir
= debugfs_create_dir("jbd2", NULL
);
2259 if (jbd2_debugfs_dir
)
2260 jbd2_debug
= debugfs_create_u8(JBD2_DEBUG_NAME
,
2263 &jbd2_journal_enable_debug
);
2266 static void __exit
jbd2_remove_debugfs_entry(void)
2268 debugfs_remove(jbd2_debug
);
2269 debugfs_remove(jbd2_debugfs_dir
);
2274 static void __init
jbd2_create_debugfs_entry(void)
2278 static void __exit
jbd2_remove_debugfs_entry(void)
2284 #ifdef CONFIG_PROC_FS
2286 #define JBD2_STATS_PROC_NAME "fs/jbd2"
2288 static void __init
jbd2_create_jbd_stats_proc_entry(void)
2290 proc_jbd2_stats
= proc_mkdir(JBD2_STATS_PROC_NAME
, NULL
);
2293 static void __exit
jbd2_remove_jbd_stats_proc_entry(void)
2295 if (proc_jbd2_stats
)
2296 remove_proc_entry(JBD2_STATS_PROC_NAME
, NULL
);
2301 #define jbd2_create_jbd_stats_proc_entry() do {} while (0)
2302 #define jbd2_remove_jbd_stats_proc_entry() do {} while (0)
2306 struct kmem_cache
*jbd2_handle_cache
, *jbd2_inode_cache
;
2308 static int __init
journal_init_handle_cache(void)
2310 jbd2_handle_cache
= KMEM_CACHE(jbd2_journal_handle
, SLAB_TEMPORARY
);
2311 if (jbd2_handle_cache
== NULL
) {
2312 printk(KERN_EMERG
"JBD2: failed to create handle cache\n");
2315 jbd2_inode_cache
= KMEM_CACHE(jbd2_inode
, 0);
2316 if (jbd2_inode_cache
== NULL
) {
2317 printk(KERN_EMERG
"JBD2: failed to create inode cache\n");
2318 kmem_cache_destroy(jbd2_handle_cache
);
2324 static void jbd2_journal_destroy_handle_cache(void)
2326 if (jbd2_handle_cache
)
2327 kmem_cache_destroy(jbd2_handle_cache
);
2328 if (jbd2_inode_cache
)
2329 kmem_cache_destroy(jbd2_inode_cache
);
2334 * Module startup and shutdown
2337 static int __init
journal_init_caches(void)
2341 ret
= jbd2_journal_init_revoke_caches();
2343 ret
= journal_init_jbd2_journal_head_cache();
2345 ret
= journal_init_handle_cache();
2349 static void jbd2_journal_destroy_caches(void)
2351 jbd2_journal_destroy_revoke_caches();
2352 jbd2_journal_destroy_jbd2_journal_head_cache();
2353 jbd2_journal_destroy_handle_cache();
2354 jbd2_journal_destroy_slabs();
2357 static int __init
journal_init(void)
2361 BUILD_BUG_ON(sizeof(struct journal_superblock_s
) != 1024);
2363 ret
= journal_init_caches();
2365 jbd2_create_debugfs_entry();
2366 jbd2_create_jbd_stats_proc_entry();
2368 jbd2_journal_destroy_caches();
2373 static void __exit
journal_exit(void)
2375 #ifdef CONFIG_JBD2_DEBUG
2376 int n
= atomic_read(&nr_journal_heads
);
2378 printk(KERN_EMERG
"JBD: leaked %d journal_heads!\n", n
);
2380 jbd2_remove_debugfs_entry();
2381 jbd2_remove_jbd_stats_proc_entry();
2382 jbd2_journal_destroy_caches();
2386 * jbd2_dev_to_name is a utility function used by the jbd2 and ext4
2387 * tracing infrastructure to map a dev_t to a device name.
2389 * The caller should use rcu_read_lock() in order to make sure the
2390 * device name stays valid until its done with it. We use
2391 * rcu_read_lock() as well to make sure we're safe in case the caller
2392 * gets sloppy, and because rcu_read_lock() is cheap and can be safely
2395 struct devname_cache
{
2396 struct rcu_head rcu
;
2398 char devname
[BDEVNAME_SIZE
];
2400 #define CACHE_SIZE_BITS 6
2401 static struct devname_cache
*devcache
[1 << CACHE_SIZE_BITS
];
2402 static DEFINE_SPINLOCK(devname_cache_lock
);
2404 static void free_devcache(struct rcu_head
*rcu
)
2409 const char *jbd2_dev_to_name(dev_t device
)
2411 int i
= hash_32(device
, CACHE_SIZE_BITS
);
2413 struct block_device
*bd
;
2414 static struct devname_cache
*new_dev
;
2417 if (devcache
[i
] && devcache
[i
]->device
== device
) {
2418 ret
= devcache
[i
]->devname
;
2424 new_dev
= kmalloc(sizeof(struct devname_cache
), GFP_KERNEL
);
2426 return "NODEV-ALLOCFAILURE"; /* Something non-NULL */
2428 spin_lock(&devname_cache_lock
);
2430 if (devcache
[i
]->device
== device
) {
2433 ret
= devcache
[i
]->devname
;
2434 spin_unlock(&devname_cache_lock
);
2437 call_rcu(&devcache
[i
]->rcu
, free_devcache
);
2439 devcache
[i
] = new_dev
;
2440 devcache
[i
]->device
= device
;
2442 bdevname(bd
, devcache
[i
]->devname
);
2445 __bdevname(device
, devcache
[i
]->devname
);
2446 ret
= devcache
[i
]->devname
;
2447 spin_unlock(&devname_cache_lock
);
2450 EXPORT_SYMBOL(jbd2_dev_to_name
);
2452 MODULE_LICENSE("GPL");
2453 module_init(journal_init
);
2454 module_exit(journal_exit
);