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, "JBD2: 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, "JBD2: 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 * Return 1 if a given transaction has not yet sent barrier request
592 * connected with a transaction commit. If 0 is returned, transaction
593 * may or may not have sent the barrier. Used to avoid sending barrier
594 * twice in common cases.
596 int jbd2_trans_will_send_data_barrier(journal_t *journal, tid_t tid)
599 transaction_t *commit_trans;
601 if (!(journal->j_flags & JBD2_BARRIER))
603 read_lock(&journal->j_state_lock);
604 /* Transaction already committed? */
605 if (tid_geq(journal->j_commit_sequence, tid))
607 commit_trans = journal->j_committing_transaction;
608 if (!commit_trans || commit_trans->t_tid != tid) {
613 * Transaction is being committed and we already proceeded to
614 * submitting a flush to fs partition?
616 if (journal->j_fs_dev != journal->j_dev) {
617 if (!commit_trans->t_need_data_flush ||
618 commit_trans->t_state >= T_COMMIT_DFLUSH)
621 if (commit_trans->t_state >= T_COMMIT_JFLUSH)
626 read_unlock(&journal->j_state_lock);
629 EXPORT_SYMBOL(jbd2_trans_will_send_data_barrier);
632 * Wait for a specified commit to complete.
633 * The caller may not hold the journal lock.
635 int jbd2_log_wait_commit(journal_t *journal, tid_t tid)
639 read_lock(&journal->j_state_lock);
640 #ifdef CONFIG_JBD2_DEBUG
641 if (!tid_geq(journal->j_commit_request, tid)) {
643 "%s: error: j_commit_request=%d, tid=%d\n",
644 __func__, journal->j_commit_request, tid);
647 while (tid_gt(tid, journal->j_commit_sequence)) {
648 jbd_debug(1, "JBD2: want %d, j_commit_sequence=%d\n",
649 tid, journal->j_commit_sequence);
650 wake_up(&journal->j_wait_commit);
651 read_unlock(&journal->j_state_lock);
652 wait_event(journal->j_wait_done_commit,
653 !tid_gt(tid, journal->j_commit_sequence));
654 read_lock(&journal->j_state_lock);
656 read_unlock(&journal->j_state_lock);
658 if (unlikely(is_journal_aborted(journal))) {
659 printk(KERN_EMERG "journal commit I/O error\n");
666 * When this function returns the transaction corresponding to tid
667 * will be completed. If the transaction has currently running, start
668 * committing that transaction before waiting for it to complete. If
669 * the transaction id is stale, it is by definition already completed,
670 * so just return SUCCESS.
672 int jbd2_complete_transaction(journal_t *journal, tid_t tid)
674 int need_to_wait = 1;
676 read_lock(&journal->j_state_lock);
677 if (journal->j_running_transaction &&
678 journal->j_running_transaction->t_tid == tid) {
679 if (journal->j_commit_request != tid) {
680 /* transaction not yet started, so request it */
681 read_unlock(&journal->j_state_lock);
682 jbd2_log_start_commit(journal, tid);
685 } else if (!(journal->j_committing_transaction &&
686 journal->j_committing_transaction->t_tid == tid))
688 read_unlock(&journal->j_state_lock);
692 return jbd2_log_wait_commit(journal, tid);
694 EXPORT_SYMBOL(jbd2_complete_transaction);
697 * Log buffer allocation routines:
700 int jbd2_journal_next_log_block(journal_t *journal, unsigned long long *retp)
702 unsigned long blocknr;
704 write_lock(&journal->j_state_lock);
705 J_ASSERT(journal->j_free > 1);
707 blocknr = journal->j_head;
710 if (journal->j_head == journal->j_last)
711 journal->j_head = journal->j_first;
712 write_unlock(&journal->j_state_lock);
713 return jbd2_journal_bmap(journal, blocknr, retp);
717 * Conversion of logical to physical block numbers for the journal
719 * On external journals the journal blocks are identity-mapped, so
720 * this is a no-op. If needed, we can use j_blk_offset - everything is
723 int jbd2_journal_bmap(journal_t *journal, unsigned long blocknr,
724 unsigned long long *retp)
727 unsigned long long ret;
729 if (journal->j_inode) {
730 ret = bmap(journal->j_inode, blocknr);
734 printk(KERN_ALERT "%s: journal block not found "
735 "at offset %lu on %s\n",
736 __func__, blocknr, journal->j_devname);
738 __journal_abort_soft(journal, err);
741 *retp = blocknr; /* +journal->j_blk_offset */
747 * We play buffer_head aliasing tricks to write data/metadata blocks to
748 * the journal without copying their contents, but for journal
749 * descriptor blocks we do need to generate bona fide buffers.
751 * After the caller of jbd2_journal_get_descriptor_buffer() has finished modifying
752 * the buffer's contents they really should run flush_dcache_page(bh->b_page).
753 * But we don't bother doing that, so there will be coherency problems with
754 * mmaps of blockdevs which hold live JBD-controlled filesystems.
756 struct journal_head *jbd2_journal_get_descriptor_buffer(journal_t *journal)
758 struct buffer_head *bh;
759 unsigned long long blocknr;
762 err = jbd2_journal_next_log_block(journal, &blocknr);
767 bh = __getblk(journal->j_dev, blocknr, journal->j_blocksize);
771 memset(bh->b_data, 0, journal->j_blocksize);
772 set_buffer_uptodate(bh);
774 BUFFER_TRACE(bh, "return this buffer");
775 return jbd2_journal_add_journal_head(bh);
779 * Return tid of the oldest transaction in the journal and block in the journal
780 * where the transaction starts.
782 * If the journal is now empty, return which will be the next transaction ID
783 * we will write and where will that transaction start.
785 * The return value is 0 if journal tail cannot be pushed any further, 1 if
788 int jbd2_journal_get_log_tail(journal_t *journal, tid_t *tid,
789 unsigned long *block)
791 transaction_t *transaction;
794 read_lock(&journal->j_state_lock);
795 spin_lock(&journal->j_list_lock);
796 transaction = journal->j_checkpoint_transactions;
798 *tid = transaction->t_tid;
799 *block = transaction->t_log_start;
800 } else if ((transaction = journal->j_committing_transaction) != NULL) {
801 *tid = transaction->t_tid;
802 *block = transaction->t_log_start;
803 } else if ((transaction = journal->j_running_transaction) != NULL) {
804 *tid = transaction->t_tid;
805 *block = journal->j_head;
807 *tid = journal->j_transaction_sequence;
808 *block = journal->j_head;
810 ret = tid_gt(*tid, journal->j_tail_sequence);
811 spin_unlock(&journal->j_list_lock);
812 read_unlock(&journal->j_state_lock);
818 * Update information in journal structure and in on disk journal superblock
819 * about log tail. This function does not check whether information passed in
820 * really pushes log tail further. It's responsibility of the caller to make
821 * sure provided log tail information is valid (e.g. by holding
822 * j_checkpoint_mutex all the time between computing log tail and calling this
823 * function as is the case with jbd2_cleanup_journal_tail()).
825 * Requires j_checkpoint_mutex
827 void __jbd2_update_log_tail(journal_t *journal, tid_t tid, unsigned long block)
831 BUG_ON(!mutex_is_locked(&journal->j_checkpoint_mutex));
834 * We cannot afford for write to remain in drive's caches since as
835 * soon as we update j_tail, next transaction can start reusing journal
836 * space and if we lose sb update during power failure we'd replay
837 * old transaction with possibly newly overwritten data.
839 jbd2_journal_update_sb_log_tail(journal, tid, block, WRITE_FUA);
840 write_lock(&journal->j_state_lock);
841 freed = block - journal->j_tail;
842 if (block < journal->j_tail)
843 freed += journal->j_last - journal->j_first;
845 trace_jbd2_update_log_tail(journal, tid, block, freed);
847 "Cleaning journal tail from %d to %d (offset %lu), "
849 journal->j_tail_sequence, tid, block, freed);
851 journal->j_free += freed;
852 journal->j_tail_sequence = tid;
853 journal->j_tail = block;
854 write_unlock(&journal->j_state_lock);
857 struct jbd2_stats_proc_session {
859 struct transaction_stats_s *stats;
864 static void *jbd2_seq_info_start(struct seq_file *seq, loff_t *pos)
866 return *pos ? NULL : SEQ_START_TOKEN;
869 static void *jbd2_seq_info_next(struct seq_file *seq, void *v, loff_t *pos)
874 static int jbd2_seq_info_show(struct seq_file *seq, void *v)
876 struct jbd2_stats_proc_session *s = seq->private;
878 if (v != SEQ_START_TOKEN)
880 seq_printf(seq, "%lu transaction, each up to %u blocks\n",
882 s->journal->j_max_transaction_buffers);
883 if (s->stats->ts_tid == 0)
885 seq_printf(seq, "average: \n %ums waiting for transaction\n",
886 jiffies_to_msecs(s->stats->run.rs_wait / s->stats->ts_tid));
887 seq_printf(seq, " %ums running transaction\n",
888 jiffies_to_msecs(s->stats->run.rs_running / s->stats->ts_tid));
889 seq_printf(seq, " %ums transaction was being locked\n",
890 jiffies_to_msecs(s->stats->run.rs_locked / s->stats->ts_tid));
891 seq_printf(seq, " %ums flushing data (in ordered mode)\n",
892 jiffies_to_msecs(s->stats->run.rs_flushing / s->stats->ts_tid));
893 seq_printf(seq, " %ums logging transaction\n",
894 jiffies_to_msecs(s->stats->run.rs_logging / s->stats->ts_tid));
895 seq_printf(seq, " %lluus average transaction commit time\n",
896 div_u64(s->journal->j_average_commit_time, 1000));
897 seq_printf(seq, " %lu handles per transaction\n",
898 s->stats->run.rs_handle_count / s->stats->ts_tid);
899 seq_printf(seq, " %lu blocks per transaction\n",
900 s->stats->run.rs_blocks / s->stats->ts_tid);
901 seq_printf(seq, " %lu logged blocks per transaction\n",
902 s->stats->run.rs_blocks_logged / s->stats->ts_tid);
906 static void jbd2_seq_info_stop(struct seq_file *seq, void *v)
910 static const struct seq_operations jbd2_seq_info_ops = {
911 .start = jbd2_seq_info_start,
912 .next = jbd2_seq_info_next,
913 .stop = jbd2_seq_info_stop,
914 .show = jbd2_seq_info_show,
917 static int jbd2_seq_info_open(struct inode *inode, struct file *file)
919 journal_t *journal = PDE(inode)->data;
920 struct jbd2_stats_proc_session *s;
923 s = kmalloc(sizeof(*s), GFP_KERNEL);
926 size = sizeof(struct transaction_stats_s);
927 s->stats = kmalloc(size, GFP_KERNEL);
928 if (s->stats == NULL) {
932 spin_lock(&journal->j_history_lock);
933 memcpy(s->stats, &journal->j_stats, size);
934 s->journal = journal;
935 spin_unlock(&journal->j_history_lock);
937 rc = seq_open(file, &jbd2_seq_info_ops);
939 struct seq_file *m = file->private_data;
949 static int jbd2_seq_info_release(struct inode *inode, struct file *file)
951 struct seq_file *seq = file->private_data;
952 struct jbd2_stats_proc_session *s = seq->private;
955 return seq_release(inode, file);
958 static const struct file_operations jbd2_seq_info_fops = {
959 .owner = THIS_MODULE,
960 .open = jbd2_seq_info_open,
963 .release = jbd2_seq_info_release,
966 static struct proc_dir_entry *proc_jbd2_stats;
968 static void jbd2_stats_proc_init(journal_t *journal)
970 journal->j_proc_entry = proc_mkdir(journal->j_devname, proc_jbd2_stats);
971 if (journal->j_proc_entry) {
972 proc_create_data("info", S_IRUGO, journal->j_proc_entry,
973 &jbd2_seq_info_fops, journal);
977 static void jbd2_stats_proc_exit(journal_t *journal)
979 remove_proc_entry("info", journal->j_proc_entry);
980 remove_proc_entry(journal->j_devname, proc_jbd2_stats);
984 * Management for journal control blocks: functions to create and
985 * destroy journal_t structures, and to initialise and read existing
986 * journal blocks from disk. */
988 /* First: create and setup a journal_t object in memory. We initialise
989 * very few fields yet: that has to wait until we have created the
990 * journal structures from from scratch, or loaded them from disk. */
992 static journal_t * journal_init_common (void)
997 journal = kzalloc(sizeof(*journal), GFP_KERNEL);
1001 init_waitqueue_head(&journal->j_wait_transaction_locked);
1002 init_waitqueue_head(&journal->j_wait_logspace);
1003 init_waitqueue_head(&journal->j_wait_done_commit);
1004 init_waitqueue_head(&journal->j_wait_checkpoint);
1005 init_waitqueue_head(&journal->j_wait_commit);
1006 init_waitqueue_head(&journal->j_wait_updates);
1007 mutex_init(&journal->j_barrier);
1008 mutex_init(&journal->j_checkpoint_mutex);
1009 spin_lock_init(&journal->j_revoke_lock);
1010 spin_lock_init(&journal->j_list_lock);
1011 rwlock_init(&journal->j_state_lock);
1013 journal->j_commit_interval = (HZ * JBD2_DEFAULT_MAX_COMMIT_AGE);
1014 journal->j_min_batch_time = 0;
1015 journal->j_max_batch_time = 15000; /* 15ms */
1017 /* The journal is marked for error until we succeed with recovery! */
1018 journal->j_flags = JBD2_ABORT;
1020 /* Set up a default-sized revoke table for the new mount. */
1021 err = jbd2_journal_init_revoke(journal, JOURNAL_REVOKE_DEFAULT_HASH);
1027 spin_lock_init(&journal->j_history_lock);
1032 /* jbd2_journal_init_dev and jbd2_journal_init_inode:
1034 * Create a journal structure assigned some fixed set of disk blocks to
1035 * the journal. We don't actually touch those disk blocks yet, but we
1036 * need to set up all of the mapping information to tell the journaling
1037 * system where the journal blocks are.
1042 * journal_t * jbd2_journal_init_dev() - creates and initialises a journal structure
1043 * @bdev: Block device on which to create the journal
1044 * @fs_dev: Device which hold journalled filesystem for this journal.
1045 * @start: Block nr Start of journal.
1046 * @len: Length of the journal in blocks.
1047 * @blocksize: blocksize of journalling device
1049 * Returns: a newly created journal_t *
1051 * jbd2_journal_init_dev creates a journal which maps a fixed contiguous
1052 * range of blocks on an arbitrary block device.
1055 journal_t * jbd2_journal_init_dev(struct block_device *bdev,
1056 struct block_device *fs_dev,
1057 unsigned long long start, int len, int blocksize)
1059 journal_t *journal = journal_init_common();
1060 struct buffer_head *bh;
1067 /* journal descriptor can store up to n blocks -bzzz */
1068 journal->j_blocksize = blocksize;
1069 journal->j_dev = bdev;
1070 journal->j_fs_dev = fs_dev;
1071 journal->j_blk_offset = start;
1072 journal->j_maxlen = len;
1073 bdevname(journal->j_dev, journal->j_devname);
1074 p = journal->j_devname;
1075 while ((p = strchr(p, '/')))
1077 jbd2_stats_proc_init(journal);
1078 n = journal->j_blocksize / sizeof(journal_block_tag_t);
1079 journal->j_wbufsize = n;
1080 journal->j_wbuf = kmalloc(n * sizeof(struct buffer_head*), GFP_KERNEL);
1081 if (!journal->j_wbuf) {
1082 printk(KERN_ERR "%s: Can't allocate bhs for commit thread\n",
1087 bh = __getblk(journal->j_dev, start, journal->j_blocksize);
1090 "%s: Cannot get buffer for journal superblock\n",
1094 journal->j_sb_buffer = bh;
1095 journal->j_superblock = (journal_superblock_t *)bh->b_data;
1099 kfree(journal->j_wbuf);
1100 jbd2_stats_proc_exit(journal);
1106 * journal_t * jbd2_journal_init_inode () - creates a journal which maps to a inode.
1107 * @inode: An inode to create the journal in
1109 * jbd2_journal_init_inode creates a journal which maps an on-disk inode as
1110 * the journal. The inode must exist already, must support bmap() and
1111 * must have all data blocks preallocated.
1113 journal_t * jbd2_journal_init_inode (struct inode *inode)
1115 struct buffer_head *bh;
1116 journal_t *journal = journal_init_common();
1120 unsigned long long blocknr;
1125 journal->j_dev = journal->j_fs_dev = inode->i_sb->s_bdev;
1126 journal->j_inode = inode;
1127 bdevname(journal->j_dev, journal->j_devname);
1128 p = journal->j_devname;
1129 while ((p = strchr(p, '/')))
1131 p = journal->j_devname + strlen(journal->j_devname);
1132 sprintf(p, "-%lu", journal->j_inode->i_ino);
1134 "journal %p: inode %s/%ld, size %Ld, bits %d, blksize %ld\n",
1135 journal, inode->i_sb->s_id, inode->i_ino,
1136 (long long) inode->i_size,
1137 inode->i_sb->s_blocksize_bits, inode->i_sb->s_blocksize);
1139 journal->j_maxlen = inode->i_size >> inode->i_sb->s_blocksize_bits;
1140 journal->j_blocksize = inode->i_sb->s_blocksize;
1141 jbd2_stats_proc_init(journal);
1143 /* journal descriptor can store up to n blocks -bzzz */
1144 n = journal->j_blocksize / sizeof(journal_block_tag_t);
1145 journal->j_wbufsize = n;
1146 journal->j_wbuf = kmalloc(n * sizeof(struct buffer_head*), GFP_KERNEL);
1147 if (!journal->j_wbuf) {
1148 printk(KERN_ERR "%s: Can't allocate bhs for commit thread\n",
1153 err = jbd2_journal_bmap(journal, 0, &blocknr);
1154 /* If that failed, give up */
1156 printk(KERN_ERR "%s: Cannot locate journal superblock\n",
1161 bh = __getblk(journal->j_dev, blocknr, journal->j_blocksize);
1164 "%s: Cannot get buffer for journal superblock\n",
1168 journal->j_sb_buffer = bh;
1169 journal->j_superblock = (journal_superblock_t *)bh->b_data;
1173 kfree(journal->j_wbuf);
1174 jbd2_stats_proc_exit(journal);
1180 * If the journal init or create aborts, we need to mark the journal
1181 * superblock as being NULL to prevent the journal destroy from writing
1182 * back a bogus superblock.
1184 static void journal_fail_superblock (journal_t *journal)
1186 struct buffer_head *bh = journal->j_sb_buffer;
1188 journal->j_sb_buffer = NULL;
1192 * Given a journal_t structure, initialise the various fields for
1193 * startup of a new journaling session. We use this both when creating
1194 * a journal, and after recovering an old journal to reset it for
1198 static int journal_reset(journal_t *journal)
1200 journal_superblock_t *sb = journal->j_superblock;
1201 unsigned long long first, last;
1203 first = be32_to_cpu(sb->s_first);
1204 last = be32_to_cpu(sb->s_maxlen);
1205 if (first + JBD2_MIN_JOURNAL_BLOCKS > last + 1) {
1206 printk(KERN_ERR "JBD2: Journal too short (blocks %llu-%llu).\n",
1208 journal_fail_superblock(journal);
1212 journal->j_first = first;
1213 journal->j_last = last;
1215 journal->j_head = first;
1216 journal->j_tail = first;
1217 journal->j_free = last - first;
1219 journal->j_tail_sequence = journal->j_transaction_sequence;
1220 journal->j_commit_sequence = journal->j_transaction_sequence - 1;
1221 journal->j_commit_request = journal->j_commit_sequence;
1223 journal->j_max_transaction_buffers = journal->j_maxlen / 4;
1226 * As a special case, if the on-disk copy is already marked as needing
1227 * no recovery (s_start == 0), then we can safely defer the superblock
1228 * update until the next commit by setting JBD2_FLUSHED. This avoids
1229 * attempting a write to a potential-readonly device.
1231 if (sb->s_start == 0) {
1232 jbd_debug(1, "JBD2: Skipping superblock update on recovered sb "
1233 "(start %ld, seq %d, errno %d)\n",
1234 journal->j_tail, journal->j_tail_sequence,
1236 journal->j_flags |= JBD2_FLUSHED;
1239 * Update log tail information. We use WRITE_FUA since new
1240 * transaction will start reusing journal space and so we
1241 * must make sure information about current log tail is on
1244 jbd2_journal_update_sb_log_tail(journal,
1245 journal->j_tail_sequence,
1249 return jbd2_journal_start_thread(journal);
1252 static void jbd2_write_superblock(journal_t *journal, int write_op)
1254 struct buffer_head *bh = journal->j_sb_buffer;
1257 if (!(journal->j_flags & JBD2_BARRIER))
1258 write_op &= ~(REQ_FUA | REQ_FLUSH);
1260 if (buffer_write_io_error(bh)) {
1262 * Oh, dear. A previous attempt to write the journal
1263 * superblock failed. This could happen because the
1264 * USB device was yanked out. Or it could happen to
1265 * be a transient write error and maybe the block will
1266 * be remapped. Nothing we can do but to retry the
1267 * write and hope for the best.
1269 printk(KERN_ERR "JBD2: previous I/O error detected "
1270 "for journal superblock update for %s.\n",
1271 journal->j_devname);
1272 clear_buffer_write_io_error(bh);
1273 set_buffer_uptodate(bh);
1276 bh->b_end_io = end_buffer_write_sync;
1277 ret = submit_bh(write_op, bh);
1279 if (buffer_write_io_error(bh)) {
1280 clear_buffer_write_io_error(bh);
1281 set_buffer_uptodate(bh);
1285 printk(KERN_ERR "JBD2: Error %d detected when updating "
1286 "journal superblock for %s.\n", ret,
1287 journal->j_devname);
1292 * jbd2_journal_update_sb_log_tail() - Update log tail in journal sb on disk.
1293 * @journal: The journal to update.
1294 * @tail_tid: TID of the new transaction at the tail of the log
1295 * @tail_block: The first block of the transaction at the tail of the log
1296 * @write_op: With which operation should we write the journal sb
1298 * Update a journal's superblock information about log tail and write it to
1299 * disk, waiting for the IO to complete.
1301 void jbd2_journal_update_sb_log_tail(journal_t *journal, tid_t tail_tid,
1302 unsigned long tail_block, int write_op)
1304 journal_superblock_t *sb = journal->j_superblock;
1306 jbd_debug(1, "JBD2: updating superblock (start %lu, seq %u)\n",
1307 tail_block, tail_tid);
1309 sb->s_sequence = cpu_to_be32(tail_tid);
1310 sb->s_start = cpu_to_be32(tail_block);
1312 jbd2_write_superblock(journal, write_op);
1313 /* Log is no longer empty */
1314 write_lock(&journal->j_state_lock);
1315 WARN_ON(!sb->s_sequence);
1316 journal->j_flags &= ~JBD2_FLUSHED;
1317 write_unlock(&journal->j_state_lock);
1321 * jbd2_mark_journal_empty() - Mark on disk journal as empty.
1322 * @journal: The journal to update.
1324 * Update a journal's dynamic superblock fields to show that journal is empty.
1325 * Write updated superblock to disk waiting for IO to complete.
1327 static void jbd2_mark_journal_empty(journal_t *journal)
1329 journal_superblock_t *sb = journal->j_superblock;
1331 read_lock(&journal->j_state_lock);
1332 jbd_debug(1, "JBD2: Marking journal as empty (seq %d)\n",
1333 journal->j_tail_sequence);
1335 sb->s_sequence = cpu_to_be32(journal->j_tail_sequence);
1336 sb->s_start = cpu_to_be32(0);
1337 read_unlock(&journal->j_state_lock);
1339 jbd2_write_superblock(journal, WRITE_FUA);
1341 /* Log is no longer empty */
1342 write_lock(&journal->j_state_lock);
1343 journal->j_flags |= JBD2_FLUSHED;
1344 write_unlock(&journal->j_state_lock);
1349 * jbd2_journal_update_sb_errno() - Update error in the journal.
1350 * @journal: The journal to update.
1352 * Update a journal's errno. Write updated superblock to disk waiting for IO
1355 static void jbd2_journal_update_sb_errno(journal_t *journal)
1357 journal_superblock_t *sb = journal->j_superblock;
1359 read_lock(&journal->j_state_lock);
1360 jbd_debug(1, "JBD2: updating superblock error (errno %d)\n",
1362 sb->s_errno = cpu_to_be32(journal->j_errno);
1363 read_unlock(&journal->j_state_lock);
1365 jbd2_write_superblock(journal, WRITE_SYNC);
1369 * Read the superblock for a given journal, performing initial
1370 * validation of the format.
1372 static int journal_get_superblock(journal_t *journal)
1374 struct buffer_head *bh;
1375 journal_superblock_t *sb;
1378 bh = journal->j_sb_buffer;
1380 J_ASSERT(bh != NULL);
1381 if (!buffer_uptodate(bh)) {
1382 ll_rw_block(READ, 1, &bh);
1384 if (!buffer_uptodate(bh)) {
1386 "JBD2: IO error reading journal superblock\n");
1391 sb = journal->j_superblock;
1395 if (sb->s_header.h_magic != cpu_to_be32(JBD2_MAGIC_NUMBER) ||
1396 sb->s_blocksize != cpu_to_be32(journal->j_blocksize)) {
1397 printk(KERN_WARNING "JBD2: no valid journal superblock found\n");
1401 switch(be32_to_cpu(sb->s_header.h_blocktype)) {
1402 case JBD2_SUPERBLOCK_V1:
1403 journal->j_format_version = 1;
1405 case JBD2_SUPERBLOCK_V2:
1406 journal->j_format_version = 2;
1409 printk(KERN_WARNING "JBD2: unrecognised superblock format ID\n");
1413 if (be32_to_cpu(sb->s_maxlen) < journal->j_maxlen)
1414 journal->j_maxlen = be32_to_cpu(sb->s_maxlen);
1415 else if (be32_to_cpu(sb->s_maxlen) > journal->j_maxlen) {
1416 printk(KERN_WARNING "JBD2: journal file too short\n");
1420 if (be32_to_cpu(sb->s_first) == 0 ||
1421 be32_to_cpu(sb->s_first) >= journal->j_maxlen) {
1423 "JBD2: Invalid start block of journal: %u\n",
1424 be32_to_cpu(sb->s_first));
1431 journal_fail_superblock(journal);
1436 * Load the on-disk journal superblock and read the key fields into the
1440 static int load_superblock(journal_t *journal)
1443 journal_superblock_t *sb;
1445 err = journal_get_superblock(journal);
1449 sb = journal->j_superblock;
1451 journal->j_tail_sequence = be32_to_cpu(sb->s_sequence);
1452 journal->j_tail = be32_to_cpu(sb->s_start);
1453 journal->j_first = be32_to_cpu(sb->s_first);
1454 journal->j_last = be32_to_cpu(sb->s_maxlen);
1455 journal->j_errno = be32_to_cpu(sb->s_errno);
1462 * int jbd2_journal_load() - Read journal from disk.
1463 * @journal: Journal to act on.
1465 * Given a journal_t structure which tells us which disk blocks contain
1466 * a journal, read the journal from disk to initialise the in-memory
1469 int jbd2_journal_load(journal_t *journal)
1472 journal_superblock_t *sb;
1474 err = load_superblock(journal);
1478 sb = journal->j_superblock;
1479 /* If this is a V2 superblock, then we have to check the
1480 * features flags on it. */
1482 if (journal->j_format_version >= 2) {
1483 if ((sb->s_feature_ro_compat &
1484 ~cpu_to_be32(JBD2_KNOWN_ROCOMPAT_FEATURES)) ||
1485 (sb->s_feature_incompat &
1486 ~cpu_to_be32(JBD2_KNOWN_INCOMPAT_FEATURES))) {
1488 "JBD2: Unrecognised features on journal\n");
1494 * Create a slab for this blocksize
1496 err = jbd2_journal_create_slab(be32_to_cpu(sb->s_blocksize));
1500 /* Let the recovery code check whether it needs to recover any
1501 * data from the journal. */
1502 if (jbd2_journal_recover(journal))
1503 goto recovery_error;
1505 if (journal->j_failed_commit) {
1506 printk(KERN_ERR "JBD2: journal transaction %u on %s "
1507 "is corrupt.\n", journal->j_failed_commit,
1508 journal->j_devname);
1512 /* OK, we've finished with the dynamic journal bits:
1513 * reinitialise the dynamic contents of the superblock in memory
1514 * and reset them on disk. */
1515 if (journal_reset(journal))
1516 goto recovery_error;
1518 journal->j_flags &= ~JBD2_ABORT;
1519 journal->j_flags |= JBD2_LOADED;
1523 printk(KERN_WARNING "JBD2: recovery failed\n");
1528 * void jbd2_journal_destroy() - Release a journal_t structure.
1529 * @journal: Journal to act on.
1531 * Release a journal_t structure once it is no longer in use by the
1533 * Return <0 if we couldn't clean up the journal.
1535 int jbd2_journal_destroy(journal_t *journal)
1539 /* Wait for the commit thread to wake up and die. */
1540 journal_kill_thread(journal);
1542 /* Force a final log commit */
1543 if (journal->j_running_transaction)
1544 jbd2_journal_commit_transaction(journal);
1546 /* Force any old transactions to disk */
1548 /* Totally anal locking here... */
1549 spin_lock(&journal->j_list_lock);
1550 while (journal->j_checkpoint_transactions != NULL) {
1551 spin_unlock(&journal->j_list_lock);
1552 mutex_lock(&journal->j_checkpoint_mutex);
1553 jbd2_log_do_checkpoint(journal);
1554 mutex_unlock(&journal->j_checkpoint_mutex);
1555 spin_lock(&journal->j_list_lock);
1558 J_ASSERT(journal->j_running_transaction == NULL);
1559 J_ASSERT(journal->j_committing_transaction == NULL);
1560 J_ASSERT(journal->j_checkpoint_transactions == NULL);
1561 spin_unlock(&journal->j_list_lock);
1563 if (journal->j_sb_buffer) {
1564 if (!is_journal_aborted(journal))
1565 jbd2_mark_journal_empty(journal);
1568 brelse(journal->j_sb_buffer);
1571 if (journal->j_proc_entry)
1572 jbd2_stats_proc_exit(journal);
1573 if (journal->j_inode)
1574 iput(journal->j_inode);
1575 if (journal->j_revoke)
1576 jbd2_journal_destroy_revoke(journal);
1577 kfree(journal->j_wbuf);
1585 *int jbd2_journal_check_used_features () - Check if features specified are used.
1586 * @journal: Journal to check.
1587 * @compat: bitmask of compatible features
1588 * @ro: bitmask of features that force read-only mount
1589 * @incompat: bitmask of incompatible features
1591 * Check whether the journal uses all of a given set of
1592 * features. Return true (non-zero) if it does.
1595 int jbd2_journal_check_used_features (journal_t *journal, unsigned long compat,
1596 unsigned long ro, unsigned long incompat)
1598 journal_superblock_t *sb;
1600 if (!compat && !ro && !incompat)
1602 /* Load journal superblock if it is not loaded yet. */
1603 if (journal->j_format_version == 0 &&
1604 journal_get_superblock(journal) != 0)
1606 if (journal->j_format_version == 1)
1609 sb = journal->j_superblock;
1611 if (((be32_to_cpu(sb->s_feature_compat) & compat) == compat) &&
1612 ((be32_to_cpu(sb->s_feature_ro_compat) & ro) == ro) &&
1613 ((be32_to_cpu(sb->s_feature_incompat) & incompat) == incompat))
1620 * int jbd2_journal_check_available_features() - Check feature set in journalling layer
1621 * @journal: Journal to check.
1622 * @compat: bitmask of compatible features
1623 * @ro: bitmask of features that force read-only mount
1624 * @incompat: bitmask of incompatible features
1626 * Check whether the journaling code supports the use of
1627 * all of a given set of features on this journal. Return true
1628 * (non-zero) if it can. */
1630 int jbd2_journal_check_available_features (journal_t *journal, unsigned long compat,
1631 unsigned long ro, unsigned long incompat)
1633 if (!compat && !ro && !incompat)
1636 /* We can support any known requested features iff the
1637 * superblock is in version 2. Otherwise we fail to support any
1638 * extended sb features. */
1640 if (journal->j_format_version != 2)
1643 if ((compat & JBD2_KNOWN_COMPAT_FEATURES) == compat &&
1644 (ro & JBD2_KNOWN_ROCOMPAT_FEATURES) == ro &&
1645 (incompat & JBD2_KNOWN_INCOMPAT_FEATURES) == incompat)
1652 * int jbd2_journal_set_features () - Mark a given journal feature in the superblock
1653 * @journal: Journal to act on.
1654 * @compat: bitmask of compatible features
1655 * @ro: bitmask of features that force read-only mount
1656 * @incompat: bitmask of incompatible features
1658 * Mark a given journal feature as present on the
1659 * superblock. Returns true if the requested features could be set.
1663 int jbd2_journal_set_features (journal_t *journal, unsigned long compat,
1664 unsigned long ro, unsigned long incompat)
1666 journal_superblock_t *sb;
1668 if (jbd2_journal_check_used_features(journal, compat, ro, incompat))
1671 if (!jbd2_journal_check_available_features(journal, compat, ro, incompat))
1674 jbd_debug(1, "Setting new features 0x%lx/0x%lx/0x%lx\n",
1675 compat, ro, incompat);
1677 sb = journal->j_superblock;
1679 sb->s_feature_compat |= cpu_to_be32(compat);
1680 sb->s_feature_ro_compat |= cpu_to_be32(ro);
1681 sb->s_feature_incompat |= cpu_to_be32(incompat);
1687 * jbd2_journal_clear_features () - Clear a given journal feature in the
1689 * @journal: Journal to act on.
1690 * @compat: bitmask of compatible features
1691 * @ro: bitmask of features that force read-only mount
1692 * @incompat: bitmask of incompatible features
1694 * Clear a given journal feature as present on the
1697 void jbd2_journal_clear_features(journal_t *journal, unsigned long compat,
1698 unsigned long ro, unsigned long incompat)
1700 journal_superblock_t *sb;
1702 jbd_debug(1, "Clear features 0x%lx/0x%lx/0x%lx\n",
1703 compat, ro, incompat);
1705 sb = journal->j_superblock;
1707 sb->s_feature_compat &= ~cpu_to_be32(compat);
1708 sb->s_feature_ro_compat &= ~cpu_to_be32(ro);
1709 sb->s_feature_incompat &= ~cpu_to_be32(incompat);
1711 EXPORT_SYMBOL(jbd2_journal_clear_features);
1714 * int jbd2_journal_update_format () - Update on-disk journal structure.
1715 * @journal: Journal to act on.
1717 * Given an initialised but unloaded journal struct, poke about in the
1718 * on-disk structure to update it to the most recent supported version.
1720 int jbd2_journal_update_format (journal_t *journal)
1722 journal_superblock_t *sb;
1725 err = journal_get_superblock(journal);
1729 sb = journal->j_superblock;
1731 switch (be32_to_cpu(sb->s_header.h_blocktype)) {
1732 case JBD2_SUPERBLOCK_V2:
1734 case JBD2_SUPERBLOCK_V1:
1735 return journal_convert_superblock_v1(journal, sb);
1742 static int journal_convert_superblock_v1(journal_t *journal,
1743 journal_superblock_t *sb)
1745 int offset, blocksize;
1746 struct buffer_head *bh;
1749 "JBD2: Converting superblock from version 1 to 2.\n");
1751 /* Pre-initialise new fields to zero */
1752 offset = ((char *) &(sb->s_feature_compat)) - ((char *) sb);
1753 blocksize = be32_to_cpu(sb->s_blocksize);
1754 memset(&sb->s_feature_compat, 0, blocksize-offset);
1756 sb->s_nr_users = cpu_to_be32(1);
1757 sb->s_header.h_blocktype = cpu_to_be32(JBD2_SUPERBLOCK_V2);
1758 journal->j_format_version = 2;
1760 bh = journal->j_sb_buffer;
1761 BUFFER_TRACE(bh, "marking dirty");
1762 mark_buffer_dirty(bh);
1763 sync_dirty_buffer(bh);
1769 * int jbd2_journal_flush () - Flush journal
1770 * @journal: Journal to act on.
1772 * Flush all data for a given journal to disk and empty the journal.
1773 * Filesystems can use this when remounting readonly to ensure that
1774 * recovery does not need to happen on remount.
1777 int jbd2_journal_flush(journal_t *journal)
1780 transaction_t *transaction = NULL;
1782 write_lock(&journal->j_state_lock);
1784 /* Force everything buffered to the log... */
1785 if (journal->j_running_transaction) {
1786 transaction = journal->j_running_transaction;
1787 __jbd2_log_start_commit(journal, transaction->t_tid);
1788 } else if (journal->j_committing_transaction)
1789 transaction = journal->j_committing_transaction;
1791 /* Wait for the log commit to complete... */
1793 tid_t tid = transaction->t_tid;
1795 write_unlock(&journal->j_state_lock);
1796 jbd2_log_wait_commit(journal, tid);
1798 write_unlock(&journal->j_state_lock);
1801 /* ...and flush everything in the log out to disk. */
1802 spin_lock(&journal->j_list_lock);
1803 while (!err && journal->j_checkpoint_transactions != NULL) {
1804 spin_unlock(&journal->j_list_lock);
1805 mutex_lock(&journal->j_checkpoint_mutex);
1806 err = jbd2_log_do_checkpoint(journal);
1807 mutex_unlock(&journal->j_checkpoint_mutex);
1808 spin_lock(&journal->j_list_lock);
1810 spin_unlock(&journal->j_list_lock);
1812 if (is_journal_aborted(journal))
1815 jbd2_cleanup_journal_tail(journal);
1817 /* Finally, mark the journal as really needing no recovery.
1818 * This sets s_start==0 in the underlying superblock, which is
1819 * the magic code for a fully-recovered superblock. Any future
1820 * commits of data to the journal will restore the current
1822 jbd2_mark_journal_empty(journal);
1823 write_lock(&journal->j_state_lock);
1824 J_ASSERT(!journal->j_running_transaction);
1825 J_ASSERT(!journal->j_committing_transaction);
1826 J_ASSERT(!journal->j_checkpoint_transactions);
1827 J_ASSERT(journal->j_head == journal->j_tail);
1828 J_ASSERT(journal->j_tail_sequence == journal->j_transaction_sequence);
1829 write_unlock(&journal->j_state_lock);
1834 * int jbd2_journal_wipe() - Wipe journal contents
1835 * @journal: Journal to act on.
1836 * @write: flag (see below)
1838 * Wipe out all of the contents of a journal, safely. This will produce
1839 * a warning if the journal contains any valid recovery information.
1840 * Must be called between journal_init_*() and jbd2_journal_load().
1842 * If 'write' is non-zero, then we wipe out the journal on disk; otherwise
1843 * we merely suppress recovery.
1846 int jbd2_journal_wipe(journal_t *journal, int write)
1850 J_ASSERT (!(journal->j_flags & JBD2_LOADED));
1852 err = load_superblock(journal);
1856 if (!journal->j_tail)
1859 printk(KERN_WARNING "JBD2: %s recovery information on journal\n",
1860 write ? "Clearing" : "Ignoring");
1862 err = jbd2_journal_skip_recovery(journal);
1864 jbd2_mark_journal_empty(journal);
1871 * Journal abort has very specific semantics, which we describe
1872 * for journal abort.
1874 * Two internal functions, which provide abort to the jbd layer
1879 * Quick version for internal journal use (doesn't lock the journal).
1880 * Aborts hard --- we mark the abort as occurred, but do _nothing_ else,
1881 * and don't attempt to make any other journal updates.
1883 void __jbd2_journal_abort_hard(journal_t *journal)
1885 transaction_t *transaction;
1887 if (journal->j_flags & JBD2_ABORT)
1890 printk(KERN_ERR "Aborting journal on device %s.\n",
1891 journal->j_devname);
1893 write_lock(&journal->j_state_lock);
1894 journal->j_flags |= JBD2_ABORT;
1895 transaction = journal->j_running_transaction;
1897 __jbd2_log_start_commit(journal, transaction->t_tid);
1898 write_unlock(&journal->j_state_lock);
1901 /* Soft abort: record the abort error status in the journal superblock,
1902 * but don't do any other IO. */
1903 static void __journal_abort_soft (journal_t *journal, int errno)
1905 if (journal->j_flags & JBD2_ABORT)
1908 if (!journal->j_errno)
1909 journal->j_errno = errno;
1911 __jbd2_journal_abort_hard(journal);
1914 jbd2_journal_update_sb_errno(journal);
1918 * void jbd2_journal_abort () - Shutdown the journal immediately.
1919 * @journal: the journal to shutdown.
1920 * @errno: an error number to record in the journal indicating
1921 * the reason for the shutdown.
1923 * Perform a complete, immediate shutdown of the ENTIRE
1924 * journal (not of a single transaction). This operation cannot be
1925 * undone without closing and reopening the journal.
1927 * The jbd2_journal_abort function is intended to support higher level error
1928 * recovery mechanisms such as the ext2/ext3 remount-readonly error
1931 * Journal abort has very specific semantics. Any existing dirty,
1932 * unjournaled buffers in the main filesystem will still be written to
1933 * disk by bdflush, but the journaling mechanism will be suspended
1934 * immediately and no further transaction commits will be honoured.
1936 * Any dirty, journaled buffers will be written back to disk without
1937 * hitting the journal. Atomicity cannot be guaranteed on an aborted
1938 * filesystem, but we _do_ attempt to leave as much data as possible
1939 * behind for fsck to use for cleanup.
1941 * Any attempt to get a new transaction handle on a journal which is in
1942 * ABORT state will just result in an -EROFS error return. A
1943 * jbd2_journal_stop on an existing handle will return -EIO if we have
1944 * entered abort state during the update.
1946 * Recursive transactions are not disturbed by journal abort until the
1947 * final jbd2_journal_stop, which will receive the -EIO error.
1949 * Finally, the jbd2_journal_abort call allows the caller to supply an errno
1950 * which will be recorded (if possible) in the journal superblock. This
1951 * allows a client to record failure conditions in the middle of a
1952 * transaction without having to complete the transaction to record the
1953 * failure to disk. ext3_error, for example, now uses this
1956 * Errors which originate from within the journaling layer will NOT
1957 * supply an errno; a null errno implies that absolutely no further
1958 * writes are done to the journal (unless there are any already in
1963 void jbd2_journal_abort(journal_t *journal, int errno)
1965 __journal_abort_soft(journal, errno);
1969 * int jbd2_journal_errno () - returns the journal's error state.
1970 * @journal: journal to examine.
1972 * This is the errno number set with jbd2_journal_abort(), the last
1973 * time the journal was mounted - if the journal was stopped
1974 * without calling abort this will be 0.
1976 * If the journal has been aborted on this mount time -EROFS will
1979 int jbd2_journal_errno(journal_t *journal)
1983 read_lock(&journal->j_state_lock);
1984 if (journal->j_flags & JBD2_ABORT)
1987 err = journal->j_errno;
1988 read_unlock(&journal->j_state_lock);
1993 * int jbd2_journal_clear_err () - clears the journal's error state
1994 * @journal: journal to act on.
1996 * An error must be cleared or acked to take a FS out of readonly
1999 int jbd2_journal_clear_err(journal_t *journal)
2003 write_lock(&journal->j_state_lock);
2004 if (journal->j_flags & JBD2_ABORT)
2007 journal->j_errno = 0;
2008 write_unlock(&journal->j_state_lock);
2013 * void jbd2_journal_ack_err() - Ack journal err.
2014 * @journal: journal to act on.
2016 * An error must be cleared or acked to take a FS out of readonly
2019 void jbd2_journal_ack_err(journal_t *journal)
2021 write_lock(&journal->j_state_lock);
2022 if (journal->j_errno)
2023 journal->j_flags |= JBD2_ACK_ERR;
2024 write_unlock(&journal->j_state_lock);
2027 int jbd2_journal_blocks_per_page(struct inode *inode)
2029 return 1 << (PAGE_CACHE_SHIFT - inode->i_sb->s_blocksize_bits);
2033 * helper functions to deal with 32 or 64bit block numbers.
2035 size_t journal_tag_bytes(journal_t *journal)
2037 if (JBD2_HAS_INCOMPAT_FEATURE(journal, JBD2_FEATURE_INCOMPAT_64BIT))
2038 return JBD2_TAG_SIZE64;
2040 return JBD2_TAG_SIZE32;
2044 * JBD memory management
2046 * These functions are used to allocate block-sized chunks of memory
2047 * used for making copies of buffer_head data. Very often it will be
2048 * page-sized chunks of data, but sometimes it will be in
2049 * sub-page-size chunks. (For example, 16k pages on Power systems
2050 * with a 4k block file system.) For blocks smaller than a page, we
2051 * use a SLAB allocator. There are slab caches for each block size,
2052 * which are allocated at mount time, if necessary, and we only free
2053 * (all of) the slab caches when/if the jbd2 module is unloaded. For
2054 * this reason we don't need to a mutex to protect access to
2055 * jbd2_slab[] allocating or releasing memory; only in
2056 * jbd2_journal_create_slab().
2058 #define JBD2_MAX_SLABS 8
2059 static struct kmem_cache *jbd2_slab[JBD2_MAX_SLABS];
2061 static const char *jbd2_slab_names[JBD2_MAX_SLABS] = {
2062 "jbd2_1k", "jbd2_2k", "jbd2_4k", "jbd2_8k",
2063 "jbd2_16k", "jbd2_32k", "jbd2_64k", "jbd2_128k"
2067 static void jbd2_journal_destroy_slabs(void)
2071 for (i = 0; i < JBD2_MAX_SLABS; i++) {
2073 kmem_cache_destroy(jbd2_slab[i]);
2074 jbd2_slab[i] = NULL;
2078 static int jbd2_journal_create_slab(size_t size)
2080 static DEFINE_MUTEX(jbd2_slab_create_mutex);
2081 int i = order_base_2(size) - 10;
2084 if (size == PAGE_SIZE)
2087 if (i >= JBD2_MAX_SLABS)
2090 if (unlikely(i < 0))
2092 mutex_lock(&jbd2_slab_create_mutex);
2094 mutex_unlock(&jbd2_slab_create_mutex);
2095 return 0; /* Already created */
2098 slab_size = 1 << (i+10);
2099 jbd2_slab[i] = kmem_cache_create(jbd2_slab_names[i], slab_size,
2100 slab_size, 0, NULL);
2101 mutex_unlock(&jbd2_slab_create_mutex);
2102 if (!jbd2_slab[i]) {
2103 printk(KERN_EMERG "JBD2: no memory for jbd2_slab cache\n");
2109 static struct kmem_cache *get_slab(size_t size)
2111 int i = order_base_2(size) - 10;
2113 BUG_ON(i >= JBD2_MAX_SLABS);
2114 if (unlikely(i < 0))
2116 BUG_ON(jbd2_slab[i] == NULL);
2117 return jbd2_slab[i];
2120 void *jbd2_alloc(size_t size, gfp_t flags)
2124 BUG_ON(size & (size-1)); /* Must be a power of 2 */
2126 flags |= __GFP_REPEAT;
2127 if (size == PAGE_SIZE)
2128 ptr = (void *)__get_free_pages(flags, 0);
2129 else if (size > PAGE_SIZE) {
2130 int order = get_order(size);
2133 ptr = (void *)__get_free_pages(flags, order);
2135 ptr = vmalloc(size);
2137 ptr = kmem_cache_alloc(get_slab(size), flags);
2139 /* Check alignment; SLUB has gotten this wrong in the past,
2140 * and this can lead to user data corruption! */
2141 BUG_ON(((unsigned long) ptr) & (size-1));
2146 void jbd2_free(void *ptr, size_t size)
2148 if (size == PAGE_SIZE) {
2149 free_pages((unsigned long)ptr, 0);
2152 if (size > PAGE_SIZE) {
2153 int order = get_order(size);
2156 free_pages((unsigned long)ptr, order);
2161 kmem_cache_free(get_slab(size), ptr);
2165 * Journal_head storage management
2167 static struct kmem_cache *jbd2_journal_head_cache;
2168 #ifdef CONFIG_JBD2_DEBUG
2169 static atomic_t nr_journal_heads = ATOMIC_INIT(0);
2172 static int journal_init_jbd2_journal_head_cache(void)
2176 J_ASSERT(jbd2_journal_head_cache == NULL);
2177 jbd2_journal_head_cache = kmem_cache_create("jbd2_journal_head",
2178 sizeof(struct journal_head),
2180 SLAB_TEMPORARY, /* flags */
2183 if (!jbd2_journal_head_cache) {
2185 printk(KERN_EMERG "JBD2: no memory for journal_head cache\n");
2190 static void jbd2_journal_destroy_jbd2_journal_head_cache(void)
2192 if (jbd2_journal_head_cache) {
2193 kmem_cache_destroy(jbd2_journal_head_cache);
2194 jbd2_journal_head_cache = NULL;
2199 * journal_head splicing and dicing
2201 static struct journal_head *journal_alloc_journal_head(void)
2203 struct journal_head *ret;
2205 #ifdef CONFIG_JBD2_DEBUG
2206 atomic_inc(&nr_journal_heads);
2208 ret = kmem_cache_alloc(jbd2_journal_head_cache, GFP_NOFS);
2210 jbd_debug(1, "out of memory for journal_head\n");
2211 pr_notice_ratelimited("ENOMEM in %s, retrying.\n", __func__);
2214 ret = kmem_cache_alloc(jbd2_journal_head_cache, GFP_NOFS);
2220 static void journal_free_journal_head(struct journal_head *jh)
2222 #ifdef CONFIG_JBD2_DEBUG
2223 atomic_dec(&nr_journal_heads);
2224 memset(jh, JBD2_POISON_FREE, sizeof(*jh));
2226 kmem_cache_free(jbd2_journal_head_cache, jh);
2230 * A journal_head is attached to a buffer_head whenever JBD has an
2231 * interest in the buffer.
2233 * Whenever a buffer has an attached journal_head, its ->b_state:BH_JBD bit
2234 * is set. This bit is tested in core kernel code where we need to take
2235 * JBD-specific actions. Testing the zeroness of ->b_private is not reliable
2238 * When a buffer has its BH_JBD bit set, its ->b_count is elevated by one.
2240 * When a buffer has its BH_JBD bit set it is immune from being released by
2241 * core kernel code, mainly via ->b_count.
2243 * A journal_head is detached from its buffer_head when the journal_head's
2244 * b_jcount reaches zero. Running transaction (b_transaction) and checkpoint
2245 * transaction (b_cp_transaction) hold their references to b_jcount.
2247 * Various places in the kernel want to attach a journal_head to a buffer_head
2248 * _before_ attaching the journal_head to a transaction. To protect the
2249 * journal_head in this situation, jbd2_journal_add_journal_head elevates the
2250 * journal_head's b_jcount refcount by one. The caller must call
2251 * jbd2_journal_put_journal_head() to undo this.
2253 * So the typical usage would be:
2255 * (Attach a journal_head if needed. Increments b_jcount)
2256 * struct journal_head *jh = jbd2_journal_add_journal_head(bh);
2258 * (Get another reference for transaction)
2259 * jbd2_journal_grab_journal_head(bh);
2260 * jh->b_transaction = xxx;
2261 * (Put original reference)
2262 * jbd2_journal_put_journal_head(jh);
2266 * Give a buffer_head a journal_head.
2270 struct journal_head *jbd2_journal_add_journal_head(struct buffer_head *bh)
2272 struct journal_head *jh;
2273 struct journal_head *new_jh = NULL;
2276 if (!buffer_jbd(bh)) {
2277 new_jh = journal_alloc_journal_head();
2278 memset(new_jh, 0, sizeof(*new_jh));
2281 jbd_lock_bh_journal_head(bh);
2282 if (buffer_jbd(bh)) {
2286 (atomic_read(&bh->b_count) > 0) ||
2287 (bh->b_page && bh->b_page->mapping));
2290 jbd_unlock_bh_journal_head(bh);
2295 new_jh = NULL; /* We consumed it */
2300 BUFFER_TRACE(bh, "added journal_head");
2303 jbd_unlock_bh_journal_head(bh);
2305 journal_free_journal_head(new_jh);
2306 return bh->b_private;
2310 * Grab a ref against this buffer_head's journal_head. If it ended up not
2311 * having a journal_head, return NULL
2313 struct journal_head *jbd2_journal_grab_journal_head(struct buffer_head *bh)
2315 struct journal_head *jh = NULL;
2317 jbd_lock_bh_journal_head(bh);
2318 if (buffer_jbd(bh)) {
2322 jbd_unlock_bh_journal_head(bh);
2326 static void __journal_remove_journal_head(struct buffer_head *bh)
2328 struct journal_head *jh = bh2jh(bh);
2330 J_ASSERT_JH(jh, jh->b_jcount >= 0);
2331 J_ASSERT_JH(jh, jh->b_transaction == NULL);
2332 J_ASSERT_JH(jh, jh->b_next_transaction == NULL);
2333 J_ASSERT_JH(jh, jh->b_cp_transaction == NULL);
2334 J_ASSERT_JH(jh, jh->b_jlist == BJ_None);
2335 J_ASSERT_BH(bh, buffer_jbd(bh));
2336 J_ASSERT_BH(bh, jh2bh(jh) == bh);
2337 BUFFER_TRACE(bh, "remove journal_head");
2338 if (jh->b_frozen_data) {
2339 printk(KERN_WARNING "%s: freeing b_frozen_data\n", __func__);
2340 jbd2_free(jh->b_frozen_data, bh->b_size);
2342 if (jh->b_committed_data) {
2343 printk(KERN_WARNING "%s: freeing b_committed_data\n", __func__);
2344 jbd2_free(jh->b_committed_data, bh->b_size);
2346 bh->b_private = NULL;
2347 jh->b_bh = NULL; /* debug, really */
2348 clear_buffer_jbd(bh);
2349 journal_free_journal_head(jh);
2353 * Drop a reference on the passed journal_head. If it fell to zero then
2354 * release the journal_head from the buffer_head.
2356 void jbd2_journal_put_journal_head(struct journal_head *jh)
2358 struct buffer_head *bh = jh2bh(jh);
2360 jbd_lock_bh_journal_head(bh);
2361 J_ASSERT_JH(jh, jh->b_jcount > 0);
2363 if (!jh->b_jcount) {
2364 __journal_remove_journal_head(bh);
2365 jbd_unlock_bh_journal_head(bh);
2368 jbd_unlock_bh_journal_head(bh);
2372 * Initialize jbd inode head
2374 void jbd2_journal_init_jbd_inode(struct jbd2_inode *jinode, struct inode *inode)
2376 jinode->i_transaction = NULL;
2377 jinode->i_next_transaction = NULL;
2378 jinode->i_vfs_inode = inode;
2379 jinode->i_flags = 0;
2380 INIT_LIST_HEAD(&jinode->i_list);
2384 * Function to be called before we start removing inode from memory (i.e.,
2385 * clear_inode() is a fine place to be called from). It removes inode from
2386 * transaction's lists.
2388 void jbd2_journal_release_jbd_inode(journal_t *journal,
2389 struct jbd2_inode *jinode)
2394 spin_lock(&journal->j_list_lock);
2395 /* Is commit writing out inode - we have to wait */
2396 if (test_bit(__JI_COMMIT_RUNNING, &jinode->i_flags)) {
2397 wait_queue_head_t *wq;
2398 DEFINE_WAIT_BIT(wait, &jinode->i_flags, __JI_COMMIT_RUNNING);
2399 wq = bit_waitqueue(&jinode->i_flags, __JI_COMMIT_RUNNING);
2400 prepare_to_wait(wq, &wait.wait, TASK_UNINTERRUPTIBLE);
2401 spin_unlock(&journal->j_list_lock);
2403 finish_wait(wq, &wait.wait);
2407 if (jinode->i_transaction) {
2408 list_del(&jinode->i_list);
2409 jinode->i_transaction = NULL;
2411 spin_unlock(&journal->j_list_lock);
2417 #ifdef CONFIG_JBD2_DEBUG
2418 u8 jbd2_journal_enable_debug __read_mostly;
2419 EXPORT_SYMBOL(jbd2_journal_enable_debug);
2421 #define JBD2_DEBUG_NAME "jbd2-debug"
2423 static struct dentry *jbd2_debugfs_dir;
2424 static struct dentry *jbd2_debug;
2426 static void __init jbd2_create_debugfs_entry(void)
2428 jbd2_debugfs_dir = debugfs_create_dir("jbd2", NULL);
2429 if (jbd2_debugfs_dir)
2430 jbd2_debug = debugfs_create_u8(JBD2_DEBUG_NAME,
2433 &jbd2_journal_enable_debug);
2436 static void __exit jbd2_remove_debugfs_entry(void)
2438 debugfs_remove(jbd2_debug);
2439 debugfs_remove(jbd2_debugfs_dir);
2444 static void __init jbd2_create_debugfs_entry(void)
2448 static void __exit jbd2_remove_debugfs_entry(void)
2454 #ifdef CONFIG_PROC_FS
2456 #define JBD2_STATS_PROC_NAME "fs/jbd2"
2458 static void __init jbd2_create_jbd_stats_proc_entry(void)
2460 proc_jbd2_stats = proc_mkdir(JBD2_STATS_PROC_NAME, NULL);
2463 static void __exit jbd2_remove_jbd_stats_proc_entry(void)
2465 if (proc_jbd2_stats)
2466 remove_proc_entry(JBD2_STATS_PROC_NAME, NULL);
2471 #define jbd2_create_jbd_stats_proc_entry() do {} while (0)
2472 #define jbd2_remove_jbd_stats_proc_entry() do {} while (0)
2476 struct kmem_cache *jbd2_handle_cache, *jbd2_inode_cache;
2478 static int __init journal_init_handle_cache(void)
2480 jbd2_handle_cache = KMEM_CACHE(jbd2_journal_handle, SLAB_TEMPORARY);
2481 if (jbd2_handle_cache == NULL) {
2482 printk(KERN_EMERG "JBD2: failed to create handle cache\n");
2485 jbd2_inode_cache = KMEM_CACHE(jbd2_inode, 0);
2486 if (jbd2_inode_cache == NULL) {
2487 printk(KERN_EMERG "JBD2: failed to create inode cache\n");
2488 kmem_cache_destroy(jbd2_handle_cache);
2494 static void jbd2_journal_destroy_handle_cache(void)
2496 if (jbd2_handle_cache)
2497 kmem_cache_destroy(jbd2_handle_cache);
2498 if (jbd2_inode_cache)
2499 kmem_cache_destroy(jbd2_inode_cache);
2504 * Module startup and shutdown
2507 static int __init journal_init_caches(void)
2511 ret = jbd2_journal_init_revoke_caches();
2513 ret = journal_init_jbd2_journal_head_cache();
2515 ret = journal_init_handle_cache();
2519 static void jbd2_journal_destroy_caches(void)
2521 jbd2_journal_destroy_revoke_caches();
2522 jbd2_journal_destroy_jbd2_journal_head_cache();
2523 jbd2_journal_destroy_handle_cache();
2524 jbd2_journal_destroy_slabs();
2527 static int __init journal_init(void)
2531 BUILD_BUG_ON(sizeof(struct journal_superblock_s) != 1024);
2533 ret = journal_init_caches();
2535 jbd2_create_debugfs_entry();
2536 jbd2_create_jbd_stats_proc_entry();
2538 jbd2_journal_destroy_caches();
2543 static void __exit journal_exit(void)
2545 #ifdef CONFIG_JBD2_DEBUG
2546 int n = atomic_read(&nr_journal_heads);
2548 printk(KERN_EMERG "JBD2: leaked %d journal_heads!\n", n);
2550 jbd2_remove_debugfs_entry();
2551 jbd2_remove_jbd_stats_proc_entry();
2552 jbd2_journal_destroy_caches();
2555 MODULE_LICENSE("GPL");
2556 module_init(journal_init);
2557 module_exit(journal_exit);