4 * Written by Stephen C. Tweedie <sct@redhat.com>, 1998
6 * Copyright 1998 Red Hat corp --- All Rights Reserved
8 * This file is part of the Linux kernel and is made available under
9 * the terms of the GNU General Public License, version 2, or at your
10 * option, any later version, incorporated herein by reference.
12 * Generic filesystem journal-writing code; part of the ext2fs
15 * This file manages journals: areas of disk reserved for logging
16 * transactional updates. This includes the kernel journaling thread
17 * which is responsible for scheduling updates to the log.
19 * We do not actually manage the physical storage of the journal in this
20 * file: that is left to a per-journal policy function, which allows us
21 * to store the journal within a filesystem-specified area for ext2
22 * journaling (ext2 can use a reserved inode for storing the log).
25 #include <linux/module.h>
26 #include <linux/time.h>
28 #include <linux/jbd.h>
29 #include <linux/errno.h>
30 #include <linux/slab.h>
31 #include <linux/smp_lock.h>
32 #include <linux/init.h>
34 #include <linux/suspend.h>
35 #include <linux/pagemap.h>
36 #include <asm/uaccess.h>
38 #include <linux/proc_fs.h>
40 EXPORT_SYMBOL(journal_start);
41 EXPORT_SYMBOL(journal_restart);
42 EXPORT_SYMBOL(journal_extend);
43 EXPORT_SYMBOL(journal_stop);
44 EXPORT_SYMBOL(journal_lock_updates);
45 EXPORT_SYMBOL(journal_unlock_updates);
46 EXPORT_SYMBOL(journal_get_write_access);
47 EXPORT_SYMBOL(journal_get_create_access);
48 EXPORT_SYMBOL(journal_get_undo_access);
49 EXPORT_SYMBOL(journal_dirty_data);
50 EXPORT_SYMBOL(journal_dirty_metadata);
51 EXPORT_SYMBOL(journal_release_buffer);
52 EXPORT_SYMBOL(journal_forget);
54 EXPORT_SYMBOL(journal_sync_buffer);
56 EXPORT_SYMBOL(journal_flush);
57 EXPORT_SYMBOL(journal_revoke);
58 EXPORT_SYMBOL(journal_callback_set);
60 EXPORT_SYMBOL(journal_init_dev);
61 EXPORT_SYMBOL(journal_init_inode);
62 EXPORT_SYMBOL(journal_update_format);
63 EXPORT_SYMBOL(journal_check_used_features);
64 EXPORT_SYMBOL(journal_check_available_features);
65 EXPORT_SYMBOL(journal_set_features);
66 EXPORT_SYMBOL(journal_create);
67 EXPORT_SYMBOL(journal_load);
68 EXPORT_SYMBOL(journal_destroy);
69 EXPORT_SYMBOL(journal_recover);
70 EXPORT_SYMBOL(journal_update_superblock);
71 EXPORT_SYMBOL(journal_abort);
72 EXPORT_SYMBOL(journal_errno);
73 EXPORT_SYMBOL(journal_ack_err);
74 EXPORT_SYMBOL(journal_clear_err);
75 EXPORT_SYMBOL(log_wait_commit);
76 EXPORT_SYMBOL(journal_start_commit);
77 EXPORT_SYMBOL(journal_force_commit_nested);
78 EXPORT_SYMBOL(journal_wipe);
79 EXPORT_SYMBOL(journal_blocks_per_page);
80 EXPORT_SYMBOL(journal_invalidatepage);
81 EXPORT_SYMBOL(journal_try_to_free_buffers);
82 EXPORT_SYMBOL(journal_bmap);
83 EXPORT_SYMBOL(journal_force_commit);
85 static int journal_convert_superblock_v1(journal_t *, journal_superblock_t *);
88 * Helper function used to manage commit timeouts
91 static void commit_timeout(unsigned long __data)
93 struct task_struct * p = (struct task_struct *) __data;
98 /* Static check for data structure consistency. There's no code
99 * invoked --- we'll just get a linker failure if things aren't right.
101 void __journal_internal_check(void)
103 extern void journal_bad_superblock_size(void);
104 if (sizeof(struct journal_superblock_s) != 1024)
105 journal_bad_superblock_size();
109 * kjournald: The main thread function used to manage a logging device
112 * This kernel thread is responsible for two things:
114 * 1) COMMIT: Every so often we need to commit the current state of the
115 * filesystem to disk. The journal thread is responsible for writing
116 * all of the metadata buffers to disk.
118 * 2) CHECKPOINT: We cannot reuse a used section of the log file until all
119 * of the data in that part of the log has been rewritten elsewhere on
120 * the disk. Flushing these old buffers to reclaim space in the log is
121 * known as checkpointing, and this thread is responsible for that job.
124 journal_t *current_journal; // AKPM: debug
126 int kjournald(void *arg)
128 journal_t *journal = (journal_t *) arg;
129 transaction_t *transaction;
130 struct timer_list timer;
132 current_journal = journal;
134 daemonize("kjournald");
136 /* Set up an interval timer which can be used to trigger a
137 commit wakeup after the commit interval expires */
139 timer.data = (unsigned long) current;
140 timer.function = commit_timeout;
141 journal->j_commit_timer = &timer;
143 /* Record that the journal thread is running */
144 journal->j_task = current;
145 wake_up(&journal->j_wait_done_commit);
147 printk(KERN_INFO "kjournald starting. Commit interval %ld seconds\n",
148 journal->j_commit_interval / HZ);
151 * And now, wait forever for commit wakeup events.
153 spin_lock(&journal->j_state_lock);
156 jbd_debug(1, "commit_sequence=%d, commit_request=%d\n",
157 journal->j_commit_sequence, journal->j_commit_request);
159 if (journal->j_commit_sequence != journal->j_commit_request) {
160 jbd_debug(1, "OK, requests differ\n");
161 spin_unlock(&journal->j_state_lock);
162 del_timer_sync(journal->j_commit_timer);
163 journal_commit_transaction(journal);
164 spin_lock(&journal->j_state_lock);
168 wake_up(&journal->j_wait_done_commit);
169 if (current->flags & PF_FREEZE) {
171 * The simpler the better. Flushing journal isn't a
172 * good idea, because that depends on threads that may
173 * be already stopped.
175 jbd_debug(1, "Now suspending kjournald\n");
176 spin_unlock(&journal->j_state_lock);
177 refrigerator(PF_FREEZE);
178 spin_lock(&journal->j_state_lock);
181 * We assume on resume that commits are already there,
185 int should_sleep = 1;
187 prepare_to_wait(&journal->j_wait_commit, &wait,
189 if (journal->j_commit_sequence != journal->j_commit_request)
191 transaction = journal->j_running_transaction;
192 if (transaction && time_after_eq(jiffies,
193 transaction->t_expires))
196 spin_unlock(&journal->j_state_lock);
198 spin_lock(&journal->j_state_lock);
200 finish_wait(&journal->j_wait_commit, &wait);
203 jbd_debug(1, "kjournald wakes\n");
206 * Were we woken up by a commit wakeup event?
208 transaction = journal->j_running_transaction;
209 if (transaction && time_after_eq(jiffies, transaction->t_expires)) {
210 journal->j_commit_request = transaction->t_tid;
211 jbd_debug(1, "woke because of timeout\n");
214 if (!(journal->j_flags & JFS_UNMOUNT))
217 spin_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 void journal_start_thread(journal_t *journal)
227 kernel_thread(kjournald, journal, CLONE_VM|CLONE_FS|CLONE_FILES);
228 wait_event(journal->j_wait_done_commit, journal->j_task != 0);
231 static void journal_kill_thread(journal_t *journal)
233 spin_lock(&journal->j_state_lock);
234 journal->j_flags |= JFS_UNMOUNT;
236 while (journal->j_task) {
237 wake_up(&journal->j_wait_commit);
238 spin_unlock(&journal->j_state_lock);
239 wait_event(journal->j_wait_done_commit, journal->j_task == 0);
240 spin_lock(&journal->j_state_lock);
242 spin_unlock(&journal->j_state_lock);
246 * journal_write_metadata_buffer: write a metadata buffer to the journal.
248 * Writes a metadata buffer to a given disk block. The actual IO is not
249 * performed but a new buffer_head is constructed which labels the data
250 * to be written with the correct destination disk block.
252 * Any magic-number escaping which needs to be done will cause a
253 * copy-out here. If the buffer happens to start with the
254 * JFS_MAGIC_NUMBER, then we can't write it to the log directly: the
255 * magic number is only written to the log for descripter blocks. In
256 * this case, we copy the data and replace the first word with 0, and we
257 * return a result code which indicates that this buffer needs to be
258 * marked as an escaped buffer in the corresponding log descriptor
259 * block. The missing word can then be restored when the block is read
262 * If the source buffer has already been modified by a new transaction
263 * since we took the last commit snapshot, we use the frozen copy of
264 * that data for IO. If we end up using the existing buffer_head's data
265 * for the write, then we *have* to lock the buffer to prevent anyone
266 * else from using and possibly modifying it while the IO is in
269 * The function returns a pointer to the buffer_heads to be used for IO.
271 * We assume that the journal has already been locked in this function.
278 * Bit 0 set == escape performed on the data
279 * Bit 1 set == buffer copy-out performed (kfree the data after IO)
282 int journal_write_metadata_buffer(transaction_t *transaction,
283 struct journal_head *jh_in,
284 struct journal_head **jh_out,
287 int need_copy_out = 0;
288 int done_copy_out = 0;
291 struct buffer_head *new_bh;
292 struct journal_head *new_jh;
293 struct page *new_page;
294 unsigned int new_offset;
295 struct buffer_head *bh_in = jh2bh(jh_in);
298 * The buffer really shouldn't be locked: only the current committing
299 * transaction is allowed to write it, so nobody else is allowed
302 * akpm: except if we're journalling data, and write() output is
303 * also part of a shared mapping, and another thread has
304 * decided to launch a writepage() against this buffer.
306 J_ASSERT_BH(bh_in, buffer_jbddirty(bh_in));
308 new_bh = alloc_buffer_head(GFP_NOFS|__GFP_NOFAIL);
311 * If a new transaction has already done a buffer copy-out, then
312 * we use that version of the data for the commit.
314 jbd_lock_bh_state(bh_in);
316 if (jh_in->b_frozen_data) {
318 new_page = virt_to_page(jh_in->b_frozen_data);
319 new_offset = offset_in_page(jh_in->b_frozen_data);
321 new_page = jh2bh(jh_in)->b_page;
322 new_offset = offset_in_page(jh2bh(jh_in)->b_data);
325 mapped_data = kmap_atomic(new_page, KM_USER0);
329 if (*((__be32 *)(mapped_data + new_offset)) ==
330 cpu_to_be32(JFS_MAGIC_NUMBER)) {
334 kunmap_atomic(mapped_data, KM_USER0);
337 * Do we need to do a data copy?
339 if (need_copy_out && !done_copy_out) {
342 jbd_unlock_bh_state(bh_in);
343 tmp = jbd_rep_kmalloc(bh_in->b_size, GFP_NOFS);
344 jbd_lock_bh_state(bh_in);
345 if (jh_in->b_frozen_data) {
350 jh_in->b_frozen_data = tmp;
351 mapped_data = kmap_atomic(new_page, KM_USER0);
352 memcpy(tmp, mapped_data + new_offset, jh2bh(jh_in)->b_size);
353 kunmap_atomic(mapped_data, KM_USER0);
355 new_page = virt_to_page(tmp);
356 new_offset = offset_in_page(tmp);
361 * Did we need to do an escaping? Now we've done all the
362 * copying, we can finally do so.
365 mapped_data = kmap_atomic(new_page, KM_USER0);
366 *((unsigned int *)(mapped_data + new_offset)) = 0;
367 kunmap_atomic(mapped_data, KM_USER0);
370 /* keep subsequent assertions sane */
372 init_buffer(new_bh, NULL, NULL);
373 atomic_set(&new_bh->b_count, 1);
374 jbd_unlock_bh_state(bh_in);
376 new_jh = journal_add_journal_head(new_bh); /* This sleeps */
378 set_bh_page(new_bh, new_page, new_offset);
379 new_jh->b_transaction = NULL;
380 new_bh->b_size = jh2bh(jh_in)->b_size;
381 new_bh->b_bdev = transaction->t_journal->j_dev;
382 new_bh->b_blocknr = blocknr;
383 set_buffer_mapped(new_bh);
384 set_buffer_dirty(new_bh);
389 * The to-be-written buffer needs to get moved to the io queue,
390 * and the original buffer whose contents we are shadowing or
391 * copying is moved to the transaction's shadow queue.
393 JBUFFER_TRACE(jh_in, "file as BJ_Shadow");
394 journal_file_buffer(jh_in, transaction, BJ_Shadow);
395 JBUFFER_TRACE(new_jh, "file as BJ_IO");
396 journal_file_buffer(new_jh, transaction, BJ_IO);
398 return do_escape | (done_copy_out << 1);
402 * Allocation code for the journal file. Manage the space left in the
403 * journal, so that we can begin checkpointing when appropriate.
407 * __log_space_left: Return the number of free blocks left in the journal.
409 * Called with the journal already locked.
411 * Called under j_state_lock
414 int __log_space_left(journal_t *journal)
416 int left = journal->j_free;
418 assert_spin_locked(&journal->j_state_lock);
421 * Be pessimistic here about the number of those free blocks which
422 * might be required for log descriptor control blocks.
425 #define MIN_LOG_RESERVED_BLOCKS 32 /* Allow for rounding errors */
427 left -= MIN_LOG_RESERVED_BLOCKS;
436 * Called under j_state_lock. Returns true if a transaction was started.
438 int __log_start_commit(journal_t *journal, tid_t target)
441 * Are we already doing a recent enough commit?
443 if (!tid_geq(journal->j_commit_request, target)) {
445 * We want a new commit: OK, mark the request and wakup the
446 * commit thread. We do _not_ do the commit ourselves.
449 journal->j_commit_request = target;
450 jbd_debug(1, "JBD: requesting commit %d/%d\n",
451 journal->j_commit_request,
452 journal->j_commit_sequence);
453 wake_up(&journal->j_wait_commit);
459 int log_start_commit(journal_t *journal, tid_t tid)
463 spin_lock(&journal->j_state_lock);
464 ret = __log_start_commit(journal, tid);
465 spin_unlock(&journal->j_state_lock);
470 * Force and wait upon a commit if the calling process is not within
471 * transaction. This is used for forcing out undo-protected data which contains
472 * bitmaps, when the fs is running out of space.
474 * We can only force the running transaction if we don't have an active handle;
475 * otherwise, we will deadlock.
477 * Returns true if a transaction was started.
479 int journal_force_commit_nested(journal_t *journal)
481 transaction_t *transaction = NULL;
484 spin_lock(&journal->j_state_lock);
485 if (journal->j_running_transaction && !current->journal_info) {
486 transaction = journal->j_running_transaction;
487 __log_start_commit(journal, transaction->t_tid);
488 } else if (journal->j_committing_transaction)
489 transaction = journal->j_committing_transaction;
492 spin_unlock(&journal->j_state_lock);
493 return 0; /* Nothing to retry */
496 tid = transaction->t_tid;
497 spin_unlock(&journal->j_state_lock);
498 log_wait_commit(journal, tid);
503 * Start a commit of the current running transaction (if any). Returns true
504 * if a transaction was started, and fills its tid in at *ptid
506 int journal_start_commit(journal_t *journal, tid_t *ptid)
510 spin_lock(&journal->j_state_lock);
511 if (journal->j_running_transaction) {
512 tid_t tid = journal->j_running_transaction->t_tid;
514 ret = __log_start_commit(journal, tid);
517 } else if (journal->j_committing_transaction && ptid) {
519 * If ext3_write_super() recently started a commit, then we
520 * have to wait for completion of that transaction
522 *ptid = journal->j_committing_transaction->t_tid;
525 spin_unlock(&journal->j_state_lock);
530 * Wait for a specified commit to complete.
531 * The caller may not hold the journal lock.
533 int log_wait_commit(journal_t *journal, tid_t tid)
537 #ifdef CONFIG_JBD_DEBUG
538 spin_lock(&journal->j_state_lock);
539 if (!tid_geq(journal->j_commit_request, tid)) {
541 "%s: error: j_commit_request=%d, tid=%d\n",
542 __FUNCTION__, journal->j_commit_request, tid);
544 spin_unlock(&journal->j_state_lock);
546 spin_lock(&journal->j_state_lock);
547 while (tid_gt(tid, journal->j_commit_sequence)) {
548 jbd_debug(1, "JBD: want %d, j_commit_sequence=%d\n",
549 tid, journal->j_commit_sequence);
550 wake_up(&journal->j_wait_commit);
551 spin_unlock(&journal->j_state_lock);
552 wait_event(journal->j_wait_done_commit,
553 !tid_gt(tid, journal->j_commit_sequence));
554 spin_lock(&journal->j_state_lock);
556 spin_unlock(&journal->j_state_lock);
558 if (unlikely(is_journal_aborted(journal))) {
559 printk(KERN_EMERG "journal commit I/O error\n");
566 * Log buffer allocation routines:
569 int journal_next_log_block(journal_t *journal, unsigned long *retp)
571 unsigned long blocknr;
573 spin_lock(&journal->j_state_lock);
574 J_ASSERT(journal->j_free > 1);
576 blocknr = journal->j_head;
579 if (journal->j_head == journal->j_last)
580 journal->j_head = journal->j_first;
581 spin_unlock(&journal->j_state_lock);
582 return journal_bmap(journal, blocknr, retp);
586 * Conversion of logical to physical block numbers for the journal
588 * On external journals the journal blocks are identity-mapped, so
589 * this is a no-op. If needed, we can use j_blk_offset - everything is
592 int journal_bmap(journal_t *journal, unsigned long blocknr,
598 if (journal->j_inode) {
599 ret = bmap(journal->j_inode, blocknr);
603 char b[BDEVNAME_SIZE];
605 printk(KERN_ALERT "%s: journal block not found "
606 "at offset %lu on %s\n",
609 bdevname(journal->j_dev, b));
611 __journal_abort_soft(journal, err);
614 *retp = blocknr; /* +journal->j_blk_offset */
620 * We play buffer_head aliasing tricks to write data/metadata blocks to
621 * the journal without copying their contents, but for journal
622 * descriptor blocks we do need to generate bona fide buffers.
624 * After the caller of journal_get_descriptor_buffer() has finished modifying
625 * the buffer's contents they really should run flush_dcache_page(bh->b_page).
626 * But we don't bother doing that, so there will be coherency problems with
627 * mmaps of blockdevs which hold live JBD-controlled filesystems.
629 struct journal_head *journal_get_descriptor_buffer(journal_t *journal)
631 struct buffer_head *bh;
632 unsigned long blocknr;
635 err = journal_next_log_block(journal, &blocknr);
640 bh = __getblk(journal->j_dev, blocknr, journal->j_blocksize);
642 memset(bh->b_data, 0, journal->j_blocksize);
643 set_buffer_uptodate(bh);
645 BUFFER_TRACE(bh, "return this buffer");
646 return journal_add_journal_head(bh);
650 * Management for journal control blocks: functions to create and
651 * destroy journal_t structures, and to initialise and read existing
652 * journal blocks from disk. */
654 /* First: create and setup a journal_t object in memory. We initialise
655 * very few fields yet: that has to wait until we have created the
656 * journal structures from from scratch, or loaded them from disk. */
658 static journal_t * journal_init_common (void)
663 journal = jbd_kmalloc(sizeof(*journal), GFP_KERNEL);
666 memset(journal, 0, sizeof(*journal));
668 init_waitqueue_head(&journal->j_wait_transaction_locked);
669 init_waitqueue_head(&journal->j_wait_logspace);
670 init_waitqueue_head(&journal->j_wait_done_commit);
671 init_waitqueue_head(&journal->j_wait_checkpoint);
672 init_waitqueue_head(&journal->j_wait_commit);
673 init_waitqueue_head(&journal->j_wait_updates);
674 init_MUTEX(&journal->j_barrier);
675 init_MUTEX(&journal->j_checkpoint_sem);
676 spin_lock_init(&journal->j_revoke_lock);
677 spin_lock_init(&journal->j_list_lock);
678 spin_lock_init(&journal->j_state_lock);
680 journal->j_commit_interval = (HZ * JBD_DEFAULT_MAX_COMMIT_AGE);
682 /* The journal is marked for error until we succeed with recovery! */
683 journal->j_flags = JFS_ABORT;
685 /* Set up a default-sized revoke table for the new mount. */
686 err = journal_init_revoke(journal, JOURNAL_REVOKE_DEFAULT_HASH);
696 /* journal_init_dev and journal_init_inode:
698 * Create a journal structure assigned some fixed set of disk blocks to
699 * the journal. We don't actually touch those disk blocks yet, but we
700 * need to set up all of the mapping information to tell the journaling
701 * system where the journal blocks are.
706 * journal_t * journal_init_dev() - creates an initialises a journal structure
707 * @bdev: Block device on which to create the journal
708 * @fs_dev: Device which hold journalled filesystem for this journal.
709 * @start: Block nr Start of journal.
710 * @len: Lenght of the journal in blocks.
711 * @blocksize: blocksize of journalling device
712 * @returns: a newly created journal_t *
714 * journal_init_dev creates a journal which maps a fixed contiguous
715 * range of blocks on an arbitrary block device.
718 journal_t * journal_init_dev(struct block_device *bdev,
719 struct block_device *fs_dev,
720 int start, int len, int blocksize)
722 journal_t *journal = journal_init_common();
723 struct buffer_head *bh;
728 journal->j_dev = bdev;
729 journal->j_fs_dev = fs_dev;
730 journal->j_blk_offset = start;
731 journal->j_maxlen = len;
732 journal->j_blocksize = blocksize;
734 bh = __getblk(journal->j_dev, start, journal->j_blocksize);
735 J_ASSERT(bh != NULL);
736 journal->j_sb_buffer = bh;
737 journal->j_superblock = (journal_superblock_t *)bh->b_data;
743 * journal_t * journal_init_inode () - creates a journal which maps to a inode.
744 * @inode: An inode to create the journal in
746 * journal_init_inode creates a journal which maps an on-disk inode as
747 * the journal. The inode must exist already, must support bmap() and
748 * must have all data blocks preallocated.
750 journal_t * journal_init_inode (struct inode *inode)
752 struct buffer_head *bh;
753 journal_t *journal = journal_init_common();
755 unsigned long blocknr;
760 journal->j_dev = journal->j_fs_dev = inode->i_sb->s_bdev;
761 journal->j_inode = inode;
763 "journal %p: inode %s/%ld, size %Ld, bits %d, blksize %ld\n",
764 journal, inode->i_sb->s_id, inode->i_ino,
765 (long long) inode->i_size,
766 inode->i_sb->s_blocksize_bits, inode->i_sb->s_blocksize);
768 journal->j_maxlen = inode->i_size >> inode->i_sb->s_blocksize_bits;
769 journal->j_blocksize = inode->i_sb->s_blocksize;
771 err = journal_bmap(journal, 0, &blocknr);
772 /* If that failed, give up */
774 printk(KERN_ERR "%s: Cannnot locate journal superblock\n",
780 bh = __getblk(journal->j_dev, blocknr, journal->j_blocksize);
781 J_ASSERT(bh != NULL);
782 journal->j_sb_buffer = bh;
783 journal->j_superblock = (journal_superblock_t *)bh->b_data;
789 * If the journal init or create aborts, we need to mark the journal
790 * superblock as being NULL to prevent the journal destroy from writing
791 * back a bogus superblock.
793 static void journal_fail_superblock (journal_t *journal)
795 struct buffer_head *bh = journal->j_sb_buffer;
797 journal->j_sb_buffer = NULL;
801 * Given a journal_t structure, initialise the various fields for
802 * startup of a new journaling session. We use this both when creating
803 * a journal, and after recovering an old journal to reset it for
807 static int journal_reset(journal_t *journal)
809 journal_superblock_t *sb = journal->j_superblock;
810 unsigned int first, last;
812 first = be32_to_cpu(sb->s_first);
813 last = be32_to_cpu(sb->s_maxlen);
815 journal->j_first = first;
816 journal->j_last = last;
818 journal->j_head = first;
819 journal->j_tail = first;
820 journal->j_free = last - first;
822 journal->j_tail_sequence = journal->j_transaction_sequence;
823 journal->j_commit_sequence = journal->j_transaction_sequence - 1;
824 journal->j_commit_request = journal->j_commit_sequence;
826 journal->j_max_transaction_buffers = journal->j_maxlen / 4;
828 /* Add the dynamic fields and write it to disk. */
829 journal_update_superblock(journal, 1);
830 journal_start_thread(journal);
835 * int journal_create() - Initialise the new journal file
836 * @journal: Journal to create. This structure must have been initialised
838 * Given a journal_t structure which tells us which disk blocks we can
839 * use, create a new journal superblock and initialise all of the
840 * journal fields from scratch.
842 int journal_create(journal_t *journal)
844 unsigned long blocknr;
845 struct buffer_head *bh;
846 journal_superblock_t *sb;
849 if (journal->j_maxlen < JFS_MIN_JOURNAL_BLOCKS) {
850 printk (KERN_ERR "Journal length (%d blocks) too short.\n",
852 journal_fail_superblock(journal);
856 if (journal->j_inode == NULL) {
858 * We don't know what block to start at!
861 "%s: creation of journal on external device!\n",
866 /* Zero out the entire journal on disk. We cannot afford to
867 have any blocks on disk beginning with JFS_MAGIC_NUMBER. */
868 jbd_debug(1, "JBD: Zeroing out journal blocks...\n");
869 for (i = 0; i < journal->j_maxlen; i++) {
870 err = journal_bmap(journal, i, &blocknr);
873 bh = __getblk(journal->j_dev, blocknr, journal->j_blocksize);
875 memset (bh->b_data, 0, journal->j_blocksize);
876 BUFFER_TRACE(bh, "marking dirty");
877 mark_buffer_dirty(bh);
878 BUFFER_TRACE(bh, "marking uptodate");
879 set_buffer_uptodate(bh);
884 sync_blockdev(journal->j_dev);
885 jbd_debug(1, "JBD: journal cleared.\n");
887 /* OK, fill in the initial static fields in the new superblock */
888 sb = journal->j_superblock;
890 sb->s_header.h_magic = cpu_to_be32(JFS_MAGIC_NUMBER);
891 sb->s_header.h_blocktype = cpu_to_be32(JFS_SUPERBLOCK_V2);
893 sb->s_blocksize = cpu_to_be32(journal->j_blocksize);
894 sb->s_maxlen = cpu_to_be32(journal->j_maxlen);
895 sb->s_first = cpu_to_be32(1);
897 journal->j_transaction_sequence = 1;
899 journal->j_flags &= ~JFS_ABORT;
900 journal->j_format_version = 2;
902 return journal_reset(journal);
906 * void journal_update_superblock() - Update journal sb on disk.
907 * @journal: The journal to update.
908 * @wait: Set to '0' if you don't want to wait for IO completion.
910 * Update a journal's dynamic superblock fields and write it to disk,
911 * optionally waiting for the IO to complete.
913 void journal_update_superblock(journal_t *journal, int wait)
915 journal_superblock_t *sb = journal->j_superblock;
916 struct buffer_head *bh = journal->j_sb_buffer;
919 * As a special case, if the on-disk copy is already marked as needing
920 * no recovery (s_start == 0) and there are no outstanding transactions
921 * in the filesystem, then we can safely defer the superblock update
922 * until the next commit by setting JFS_FLUSHED. This avoids
923 * attempting a write to a potential-readonly device.
925 if (sb->s_start == 0 && journal->j_tail_sequence ==
926 journal->j_transaction_sequence) {
927 jbd_debug(1,"JBD: Skipping superblock update on recovered sb "
928 "(start %ld, seq %d, errno %d)\n",
929 journal->j_tail, journal->j_tail_sequence,
934 spin_lock(&journal->j_state_lock);
935 jbd_debug(1,"JBD: updating superblock (start %ld, seq %d, errno %d)\n",
936 journal->j_tail, journal->j_tail_sequence, journal->j_errno);
938 sb->s_sequence = cpu_to_be32(journal->j_tail_sequence);
939 sb->s_start = cpu_to_be32(journal->j_tail);
940 sb->s_errno = cpu_to_be32(journal->j_errno);
941 spin_unlock(&journal->j_state_lock);
943 BUFFER_TRACE(bh, "marking dirty");
944 mark_buffer_dirty(bh);
946 sync_dirty_buffer(bh);
948 ll_rw_block(WRITE, 1, &bh);
951 /* If we have just flushed the log (by marking s_start==0), then
952 * any future commit will have to be careful to update the
953 * superblock again to re-record the true start of the log. */
955 spin_lock(&journal->j_state_lock);
957 journal->j_flags &= ~JFS_FLUSHED;
959 journal->j_flags |= JFS_FLUSHED;
960 spin_unlock(&journal->j_state_lock);
964 * Read the superblock for a given journal, performing initial
965 * validation of the format.
968 static int journal_get_superblock(journal_t *journal)
970 struct buffer_head *bh;
971 journal_superblock_t *sb;
974 bh = journal->j_sb_buffer;
976 J_ASSERT(bh != NULL);
977 if (!buffer_uptodate(bh)) {
978 ll_rw_block(READ, 1, &bh);
980 if (!buffer_uptodate(bh)) {
982 "JBD: IO error reading journal superblock\n");
987 sb = journal->j_superblock;
991 if (sb->s_header.h_magic != cpu_to_be32(JFS_MAGIC_NUMBER) ||
992 sb->s_blocksize != cpu_to_be32(journal->j_blocksize)) {
993 printk(KERN_WARNING "JBD: no valid journal superblock found\n");
997 switch(be32_to_cpu(sb->s_header.h_blocktype)) {
998 case JFS_SUPERBLOCK_V1:
999 journal->j_format_version = 1;
1001 case JFS_SUPERBLOCK_V2:
1002 journal->j_format_version = 2;
1005 printk(KERN_WARNING "JBD: unrecognised superblock format ID\n");
1009 if (be32_to_cpu(sb->s_maxlen) < journal->j_maxlen)
1010 journal->j_maxlen = be32_to_cpu(sb->s_maxlen);
1011 else if (be32_to_cpu(sb->s_maxlen) > journal->j_maxlen) {
1012 printk (KERN_WARNING "JBD: journal file too short\n");
1019 journal_fail_superblock(journal);
1024 * Load the on-disk journal superblock and read the key fields into the
1028 static int load_superblock(journal_t *journal)
1031 journal_superblock_t *sb;
1033 err = journal_get_superblock(journal);
1037 sb = journal->j_superblock;
1039 journal->j_tail_sequence = be32_to_cpu(sb->s_sequence);
1040 journal->j_tail = be32_to_cpu(sb->s_start);
1041 journal->j_first = be32_to_cpu(sb->s_first);
1042 journal->j_last = be32_to_cpu(sb->s_maxlen);
1043 journal->j_errno = be32_to_cpu(sb->s_errno);
1050 * int journal_load() - Read journal from disk.
1051 * @journal: Journal to act on.
1053 * Given a journal_t structure which tells us which disk blocks contain
1054 * a journal, read the journal from disk to initialise the in-memory
1057 int journal_load(journal_t *journal)
1061 err = load_superblock(journal);
1065 /* If this is a V2 superblock, then we have to check the
1066 * features flags on it. */
1068 if (journal->j_format_version >= 2) {
1069 journal_superblock_t *sb = journal->j_superblock;
1071 if ((sb->s_feature_ro_compat &
1072 ~cpu_to_be32(JFS_KNOWN_ROCOMPAT_FEATURES)) ||
1073 (sb->s_feature_incompat &
1074 ~cpu_to_be32(JFS_KNOWN_INCOMPAT_FEATURES))) {
1075 printk (KERN_WARNING
1076 "JBD: Unrecognised features on journal\n");
1081 /* Let the recovery code check whether it needs to recover any
1082 * data from the journal. */
1083 if (journal_recover(journal))
1084 goto recovery_error;
1086 /* OK, we've finished with the dynamic journal bits:
1087 * reinitialise the dynamic contents of the superblock in memory
1088 * and reset them on disk. */
1089 if (journal_reset(journal))
1090 goto recovery_error;
1092 journal->j_flags &= ~JFS_ABORT;
1093 journal->j_flags |= JFS_LOADED;
1097 printk (KERN_WARNING "JBD: recovery failed\n");
1102 * void journal_destroy() - Release a journal_t structure.
1103 * @journal: Journal to act on.
1105 * Release a journal_t structure once it is no longer in use by the
1108 void journal_destroy(journal_t *journal)
1110 /* Wait for the commit thread to wake up and die. */
1111 journal_kill_thread(journal);
1113 /* Force a final log commit */
1114 if (journal->j_running_transaction)
1115 journal_commit_transaction(journal);
1117 /* Force any old transactions to disk */
1119 /* Totally anal locking here... */
1120 spin_lock(&journal->j_list_lock);
1121 while (journal->j_checkpoint_transactions != NULL) {
1122 spin_unlock(&journal->j_list_lock);
1123 log_do_checkpoint(journal);
1124 spin_lock(&journal->j_list_lock);
1127 J_ASSERT(journal->j_running_transaction == NULL);
1128 J_ASSERT(journal->j_committing_transaction == NULL);
1129 J_ASSERT(journal->j_checkpoint_transactions == NULL);
1130 spin_unlock(&journal->j_list_lock);
1132 /* We can now mark the journal as empty. */
1133 journal->j_tail = 0;
1134 journal->j_tail_sequence = ++journal->j_transaction_sequence;
1135 if (journal->j_sb_buffer) {
1136 journal_update_superblock(journal, 1);
1137 brelse(journal->j_sb_buffer);
1140 if (journal->j_inode)
1141 iput(journal->j_inode);
1142 if (journal->j_revoke)
1143 journal_destroy_revoke(journal);
1149 *int journal_check_used_features () - Check if features specified are used.
1151 * Check whether the journal uses all of a given set of
1152 * features. Return true (non-zero) if it does.
1155 int journal_check_used_features (journal_t *journal, unsigned long compat,
1156 unsigned long ro, unsigned long incompat)
1158 journal_superblock_t *sb;
1160 if (!compat && !ro && !incompat)
1162 if (journal->j_format_version == 1)
1165 sb = journal->j_superblock;
1167 if (((be32_to_cpu(sb->s_feature_compat) & compat) == compat) &&
1168 ((be32_to_cpu(sb->s_feature_ro_compat) & ro) == ro) &&
1169 ((be32_to_cpu(sb->s_feature_incompat) & incompat) == incompat))
1176 * int journal_check_available_features() - Check feature set in journalling layer
1178 * Check whether the journaling code supports the use of
1179 * all of a given set of features on this journal. Return true
1180 * (non-zero) if it can. */
1182 int journal_check_available_features (journal_t *journal, unsigned long compat,
1183 unsigned long ro, unsigned long incompat)
1185 journal_superblock_t *sb;
1187 if (!compat && !ro && !incompat)
1190 sb = journal->j_superblock;
1192 /* We can support any known requested features iff the
1193 * superblock is in version 2. Otherwise we fail to support any
1194 * extended sb features. */
1196 if (journal->j_format_version != 2)
1199 if ((compat & JFS_KNOWN_COMPAT_FEATURES) == compat &&
1200 (ro & JFS_KNOWN_ROCOMPAT_FEATURES) == ro &&
1201 (incompat & JFS_KNOWN_INCOMPAT_FEATURES) == incompat)
1208 * int journal_set_features () - Mark a given journal feature in the superblock
1210 * Mark a given journal feature as present on the
1211 * superblock. Returns true if the requested features could be set.
1215 int journal_set_features (journal_t *journal, unsigned long compat,
1216 unsigned long ro, unsigned long incompat)
1218 journal_superblock_t *sb;
1220 if (journal_check_used_features(journal, compat, ro, incompat))
1223 if (!journal_check_available_features(journal, compat, ro, incompat))
1226 jbd_debug(1, "Setting new features 0x%lx/0x%lx/0x%lx\n",
1227 compat, ro, incompat);
1229 sb = journal->j_superblock;
1231 sb->s_feature_compat |= cpu_to_be32(compat);
1232 sb->s_feature_ro_compat |= cpu_to_be32(ro);
1233 sb->s_feature_incompat |= cpu_to_be32(incompat);
1240 * int journal_update_format () - Update on-disk journal structure.
1242 * Given an initialised but unloaded journal struct, poke about in the
1243 * on-disk structure to update it to the most recent supported version.
1245 int journal_update_format (journal_t *journal)
1247 journal_superblock_t *sb;
1250 err = journal_get_superblock(journal);
1254 sb = journal->j_superblock;
1256 switch (be32_to_cpu(sb->s_header.h_blocktype)) {
1257 case JFS_SUPERBLOCK_V2:
1259 case JFS_SUPERBLOCK_V1:
1260 return journal_convert_superblock_v1(journal, sb);
1267 static int journal_convert_superblock_v1(journal_t *journal,
1268 journal_superblock_t *sb)
1270 int offset, blocksize;
1271 struct buffer_head *bh;
1274 "JBD: Converting superblock from version 1 to 2.\n");
1276 /* Pre-initialise new fields to zero */
1277 offset = ((char *) &(sb->s_feature_compat)) - ((char *) sb);
1278 blocksize = be32_to_cpu(sb->s_blocksize);
1279 memset(&sb->s_feature_compat, 0, blocksize-offset);
1281 sb->s_nr_users = cpu_to_be32(1);
1282 sb->s_header.h_blocktype = cpu_to_be32(JFS_SUPERBLOCK_V2);
1283 journal->j_format_version = 2;
1285 bh = journal->j_sb_buffer;
1286 BUFFER_TRACE(bh, "marking dirty");
1287 mark_buffer_dirty(bh);
1288 sync_dirty_buffer(bh);
1294 * int journal_flush () - Flush journal
1295 * @journal: Journal to act on.
1297 * Flush all data for a given journal to disk and empty the journal.
1298 * Filesystems can use this when remounting readonly to ensure that
1299 * recovery does not need to happen on remount.
1302 int journal_flush(journal_t *journal)
1305 transaction_t *transaction = NULL;
1306 unsigned long old_tail;
1308 spin_lock(&journal->j_state_lock);
1310 /* Force everything buffered to the log... */
1311 if (journal->j_running_transaction) {
1312 transaction = journal->j_running_transaction;
1313 __log_start_commit(journal, transaction->t_tid);
1314 } else if (journal->j_committing_transaction)
1315 transaction = journal->j_committing_transaction;
1317 /* Wait for the log commit to complete... */
1319 tid_t tid = transaction->t_tid;
1321 spin_unlock(&journal->j_state_lock);
1322 log_wait_commit(journal, tid);
1324 spin_unlock(&journal->j_state_lock);
1327 /* ...and flush everything in the log out to disk. */
1328 spin_lock(&journal->j_list_lock);
1329 while (!err && journal->j_checkpoint_transactions != NULL) {
1330 spin_unlock(&journal->j_list_lock);
1331 err = log_do_checkpoint(journal);
1332 spin_lock(&journal->j_list_lock);
1334 spin_unlock(&journal->j_list_lock);
1335 cleanup_journal_tail(journal);
1337 /* Finally, mark the journal as really needing no recovery.
1338 * This sets s_start==0 in the underlying superblock, which is
1339 * the magic code for a fully-recovered superblock. Any future
1340 * commits of data to the journal will restore the current
1342 spin_lock(&journal->j_state_lock);
1343 old_tail = journal->j_tail;
1344 journal->j_tail = 0;
1345 spin_unlock(&journal->j_state_lock);
1346 journal_update_superblock(journal, 1);
1347 spin_lock(&journal->j_state_lock);
1348 journal->j_tail = old_tail;
1350 J_ASSERT(!journal->j_running_transaction);
1351 J_ASSERT(!journal->j_committing_transaction);
1352 J_ASSERT(!journal->j_checkpoint_transactions);
1353 J_ASSERT(journal->j_head == journal->j_tail);
1354 J_ASSERT(journal->j_tail_sequence == journal->j_transaction_sequence);
1355 spin_unlock(&journal->j_state_lock);
1360 * int journal_wipe() - Wipe journal contents
1361 * @journal: Journal to act on.
1362 * @write: flag (see below)
1364 * Wipe out all of the contents of a journal, safely. This will produce
1365 * a warning if the journal contains any valid recovery information.
1366 * Must be called between journal_init_*() and journal_load().
1368 * If 'write' is non-zero, then we wipe out the journal on disk; otherwise
1369 * we merely suppress recovery.
1372 int journal_wipe(journal_t *journal, int write)
1374 journal_superblock_t *sb;
1377 J_ASSERT (!(journal->j_flags & JFS_LOADED));
1379 err = load_superblock(journal);
1383 sb = journal->j_superblock;
1385 if (!journal->j_tail)
1388 printk (KERN_WARNING "JBD: %s recovery information on journal\n",
1389 write ? "Clearing" : "Ignoring");
1391 err = journal_skip_recovery(journal);
1393 journal_update_superblock(journal, 1);
1400 * journal_dev_name: format a character string to describe on what
1401 * device this journal is present.
1404 const char *journal_dev_name(journal_t *journal, char *buffer)
1406 struct block_device *bdev;
1408 if (journal->j_inode)
1409 bdev = journal->j_inode->i_sb->s_bdev;
1411 bdev = journal->j_dev;
1413 return bdevname(bdev, buffer);
1417 * Journal abort has very specific semantics, which we describe
1418 * for journal abort.
1420 * Two internal function, which provide abort to te jbd layer
1425 * Quick version for internal journal use (doesn't lock the journal).
1426 * Aborts hard --- we mark the abort as occurred, but do _nothing_ else,
1427 * and don't attempt to make any other journal updates.
1429 void __journal_abort_hard(journal_t *journal)
1431 transaction_t *transaction;
1432 char b[BDEVNAME_SIZE];
1434 if (journal->j_flags & JFS_ABORT)
1437 printk(KERN_ERR "Aborting journal on device %s.\n",
1438 journal_dev_name(journal, b));
1440 spin_lock(&journal->j_state_lock);
1441 journal->j_flags |= JFS_ABORT;
1442 transaction = journal->j_running_transaction;
1444 __log_start_commit(journal, transaction->t_tid);
1445 spin_unlock(&journal->j_state_lock);
1448 /* Soft abort: record the abort error status in the journal superblock,
1449 * but don't do any other IO. */
1450 void __journal_abort_soft (journal_t *journal, int errno)
1452 if (journal->j_flags & JFS_ABORT)
1455 if (!journal->j_errno)
1456 journal->j_errno = errno;
1458 __journal_abort_hard(journal);
1461 journal_update_superblock(journal, 1);
1465 * void journal_abort () - Shutdown the journal immediately.
1466 * @journal: the journal to shutdown.
1467 * @errno: an error number to record in the journal indicating
1468 * the reason for the shutdown.
1470 * Perform a complete, immediate shutdown of the ENTIRE
1471 * journal (not of a single transaction). This operation cannot be
1472 * undone without closing and reopening the journal.
1474 * The journal_abort function is intended to support higher level error
1475 * recovery mechanisms such as the ext2/ext3 remount-readonly error
1478 * Journal abort has very specific semantics. Any existing dirty,
1479 * unjournaled buffers in the main filesystem will still be written to
1480 * disk by bdflush, but the journaling mechanism will be suspended
1481 * immediately and no further transaction commits will be honoured.
1483 * Any dirty, journaled buffers will be written back to disk without
1484 * hitting the journal. Atomicity cannot be guaranteed on an aborted
1485 * filesystem, but we _do_ attempt to leave as much data as possible
1486 * behind for fsck to use for cleanup.
1488 * Any attempt to get a new transaction handle on a journal which is in
1489 * ABORT state will just result in an -EROFS error return. A
1490 * journal_stop on an existing handle will return -EIO if we have
1491 * entered abort state during the update.
1493 * Recursive transactions are not disturbed by journal abort until the
1494 * final journal_stop, which will receive the -EIO error.
1496 * Finally, the journal_abort call allows the caller to supply an errno
1497 * which will be recorded (if possible) in the journal superblock. This
1498 * allows a client to record failure conditions in the middle of a
1499 * transaction without having to complete the transaction to record the
1500 * failure to disk. ext3_error, for example, now uses this
1503 * Errors which originate from within the journaling layer will NOT
1504 * supply an errno; a null errno implies that absolutely no further
1505 * writes are done to the journal (unless there are any already in
1510 void journal_abort(journal_t *journal, int errno)
1512 __journal_abort_soft(journal, errno);
1516 * int journal_errno () - returns the journal's error state.
1517 * @journal: journal to examine.
1519 * This is the errno numbet set with journal_abort(), the last
1520 * time the journal was mounted - if the journal was stopped
1521 * without calling abort this will be 0.
1523 * If the journal has been aborted on this mount time -EROFS will
1526 int journal_errno(journal_t *journal)
1530 spin_lock(&journal->j_state_lock);
1531 if (journal->j_flags & JFS_ABORT)
1534 err = journal->j_errno;
1535 spin_unlock(&journal->j_state_lock);
1540 * int journal_clear_err () - clears the journal's error state
1542 * An error must be cleared or Acked to take a FS out of readonly
1545 int journal_clear_err(journal_t *journal)
1549 spin_lock(&journal->j_state_lock);
1550 if (journal->j_flags & JFS_ABORT)
1553 journal->j_errno = 0;
1554 spin_unlock(&journal->j_state_lock);
1559 * void journal_ack_err() - Ack journal err.
1561 * An error must be cleared or Acked to take a FS out of readonly
1564 void journal_ack_err(journal_t *journal)
1566 spin_lock(&journal->j_state_lock);
1567 if (journal->j_errno)
1568 journal->j_flags |= JFS_ACK_ERR;
1569 spin_unlock(&journal->j_state_lock);
1572 int journal_blocks_per_page(struct inode *inode)
1574 return 1 << (PAGE_CACHE_SHIFT - inode->i_sb->s_blocksize_bits);
1578 * Simple support for retying memory allocations. Introduced to help to
1579 * debug different VM deadlock avoidance strategies.
1582 * Simple support for retying memory allocations. Introduced to help to
1583 * debug different VM deadlock avoidance strategies.
1585 void * __jbd_kmalloc (const char *where, size_t size, int flags, int retry)
1587 return kmalloc(size, flags | (retry ? __GFP_NOFAIL : 0));
1591 * Journal_head storage management
1593 static kmem_cache_t *journal_head_cache;
1594 #ifdef CONFIG_JBD_DEBUG
1595 static atomic_t nr_journal_heads = ATOMIC_INIT(0);
1598 static int journal_init_journal_head_cache(void)
1602 J_ASSERT(journal_head_cache == 0);
1603 journal_head_cache = kmem_cache_create("journal_head",
1604 sizeof(struct journal_head),
1610 if (journal_head_cache == 0) {
1612 printk(KERN_EMERG "JBD: no memory for journal_head cache\n");
1617 static void journal_destroy_journal_head_cache(void)
1619 J_ASSERT(journal_head_cache != NULL);
1620 kmem_cache_destroy(journal_head_cache);
1621 journal_head_cache = NULL;
1625 * journal_head splicing and dicing
1627 static struct journal_head *journal_alloc_journal_head(void)
1629 struct journal_head *ret;
1630 static unsigned long last_warning;
1632 #ifdef CONFIG_JBD_DEBUG
1633 atomic_inc(&nr_journal_heads);
1635 ret = kmem_cache_alloc(journal_head_cache, GFP_NOFS);
1637 jbd_debug(1, "out of memory for journal_head\n");
1638 if (time_after(jiffies, last_warning + 5*HZ)) {
1639 printk(KERN_NOTICE "ENOMEM in %s, retrying.\n",
1641 last_warning = jiffies;
1645 ret = kmem_cache_alloc(journal_head_cache, GFP_NOFS);
1651 static void journal_free_journal_head(struct journal_head *jh)
1653 #ifdef CONFIG_JBD_DEBUG
1654 atomic_dec(&nr_journal_heads);
1655 memset(jh, 0x5b, sizeof(*jh));
1657 kmem_cache_free(journal_head_cache, jh);
1661 * A journal_head is attached to a buffer_head whenever JBD has an
1662 * interest in the buffer.
1664 * Whenever a buffer has an attached journal_head, its ->b_state:BH_JBD bit
1665 * is set. This bit is tested in core kernel code where we need to take
1666 * JBD-specific actions. Testing the zeroness of ->b_private is not reliable
1669 * When a buffer has its BH_JBD bit set, its ->b_count is elevated by one.
1671 * When a buffer has its BH_JBD bit set it is immune from being released by
1672 * core kernel code, mainly via ->b_count.
1674 * A journal_head may be detached from its buffer_head when the journal_head's
1675 * b_transaction, b_cp_transaction and b_next_transaction pointers are NULL.
1676 * Various places in JBD call journal_remove_journal_head() to indicate that the
1677 * journal_head can be dropped if needed.
1679 * Various places in the kernel want to attach a journal_head to a buffer_head
1680 * _before_ attaching the journal_head to a transaction. To protect the
1681 * journal_head in this situation, journal_add_journal_head elevates the
1682 * journal_head's b_jcount refcount by one. The caller must call
1683 * journal_put_journal_head() to undo this.
1685 * So the typical usage would be:
1687 * (Attach a journal_head if needed. Increments b_jcount)
1688 * struct journal_head *jh = journal_add_journal_head(bh);
1690 * jh->b_transaction = xxx;
1691 * journal_put_journal_head(jh);
1693 * Now, the journal_head's b_jcount is zero, but it is safe from being released
1694 * because it has a non-zero b_transaction.
1698 * Give a buffer_head a journal_head.
1700 * Doesn't need the journal lock.
1703 struct journal_head *journal_add_journal_head(struct buffer_head *bh)
1705 struct journal_head *jh;
1706 struct journal_head *new_jh = NULL;
1709 if (!buffer_jbd(bh)) {
1710 new_jh = journal_alloc_journal_head();
1711 memset(new_jh, 0, sizeof(*new_jh));
1714 jbd_lock_bh_journal_head(bh);
1715 if (buffer_jbd(bh)) {
1719 (atomic_read(&bh->b_count) > 0) ||
1720 (bh->b_page && bh->b_page->mapping));
1723 jbd_unlock_bh_journal_head(bh);
1728 new_jh = NULL; /* We consumed it */
1733 BUFFER_TRACE(bh, "added journal_head");
1736 jbd_unlock_bh_journal_head(bh);
1738 journal_free_journal_head(new_jh);
1739 return bh->b_private;
1743 * Grab a ref against this buffer_head's journal_head. If it ended up not
1744 * having a journal_head, return NULL
1746 struct journal_head *journal_grab_journal_head(struct buffer_head *bh)
1748 struct journal_head *jh = NULL;
1750 jbd_lock_bh_journal_head(bh);
1751 if (buffer_jbd(bh)) {
1755 jbd_unlock_bh_journal_head(bh);
1759 static void __journal_remove_journal_head(struct buffer_head *bh)
1761 struct journal_head *jh = bh2jh(bh);
1763 J_ASSERT_JH(jh, jh->b_jcount >= 0);
1766 if (jh->b_jcount == 0) {
1767 if (jh->b_transaction == NULL &&
1768 jh->b_next_transaction == NULL &&
1769 jh->b_cp_transaction == NULL) {
1770 J_ASSERT_BH(bh, buffer_jbd(bh));
1771 J_ASSERT_BH(bh, jh2bh(jh) == bh);
1772 BUFFER_TRACE(bh, "remove journal_head");
1773 if (jh->b_frozen_data) {
1774 printk(KERN_WARNING "%s: freeing "
1777 kfree(jh->b_frozen_data);
1779 if (jh->b_committed_data) {
1780 printk(KERN_WARNING "%s: freeing "
1781 "b_committed_data\n",
1783 kfree(jh->b_committed_data);
1785 bh->b_private = NULL;
1786 jh->b_bh = NULL; /* debug, really */
1787 clear_buffer_jbd(bh);
1789 journal_free_journal_head(jh);
1791 BUFFER_TRACE(bh, "journal_head was locked");
1797 * journal_remove_journal_head(): if the buffer isn't attached to a transaction
1798 * and has a zero b_jcount then remove and release its journal_head. If we did
1799 * see that the buffer is not used by any transaction we also "logically"
1800 * decrement ->b_count.
1802 * We in fact take an additional increment on ->b_count as a convenience,
1803 * because the caller usually wants to do additional things with the bh
1804 * after calling here.
1805 * The caller of journal_remove_journal_head() *must* run __brelse(bh) at some
1806 * time. Once the caller has run __brelse(), the buffer is eligible for
1807 * reaping by try_to_free_buffers().
1809 void journal_remove_journal_head(struct buffer_head *bh)
1811 jbd_lock_bh_journal_head(bh);
1812 __journal_remove_journal_head(bh);
1813 jbd_unlock_bh_journal_head(bh);
1817 * Drop a reference on the passed journal_head. If it fell to zero then try to
1818 * release the journal_head from the buffer_head.
1820 void journal_put_journal_head(struct journal_head *jh)
1822 struct buffer_head *bh = jh2bh(jh);
1824 jbd_lock_bh_journal_head(bh);
1825 J_ASSERT_JH(jh, jh->b_jcount > 0);
1827 if (!jh->b_jcount && !jh->b_transaction) {
1828 __journal_remove_journal_head(bh);
1831 jbd_unlock_bh_journal_head(bh);
1837 #if defined(CONFIG_JBD_DEBUG)
1838 int journal_enable_debug;
1839 EXPORT_SYMBOL(journal_enable_debug);
1842 #if defined(CONFIG_JBD_DEBUG) && defined(CONFIG_PROC_FS)
1844 static struct proc_dir_entry *proc_jbd_debug;
1846 int read_jbd_debug(char *page, char **start, off_t off,
1847 int count, int *eof, void *data)
1851 ret = sprintf(page + off, "%d\n", journal_enable_debug);
1856 int write_jbd_debug(struct file *file, const char __user *buffer,
1857 unsigned long count, void *data)
1861 if (count > ARRAY_SIZE(buf) - 1)
1862 count = ARRAY_SIZE(buf) - 1;
1863 if (copy_from_user(buf, buffer, count))
1865 buf[ARRAY_SIZE(buf) - 1] = '\0';
1866 journal_enable_debug = simple_strtoul(buf, NULL, 10);
1870 #define JBD_PROC_NAME "sys/fs/jbd-debug"
1872 static void __init create_jbd_proc_entry(void)
1874 proc_jbd_debug = create_proc_entry(JBD_PROC_NAME, 0644, NULL);
1875 if (proc_jbd_debug) {
1876 /* Why is this so hard? */
1877 proc_jbd_debug->read_proc = read_jbd_debug;
1878 proc_jbd_debug->write_proc = write_jbd_debug;
1882 static void __exit remove_jbd_proc_entry(void)
1885 remove_proc_entry(JBD_PROC_NAME, NULL);
1890 #define create_jbd_proc_entry() do {} while (0)
1891 #define remove_jbd_proc_entry() do {} while (0)
1895 kmem_cache_t *jbd_handle_cache;
1897 static int __init journal_init_handle_cache(void)
1899 jbd_handle_cache = kmem_cache_create("journal_handle",
1905 if (jbd_handle_cache == NULL) {
1906 printk(KERN_EMERG "JBD: failed to create handle cache\n");
1912 static void journal_destroy_handle_cache(void)
1914 if (jbd_handle_cache)
1915 kmem_cache_destroy(jbd_handle_cache);
1919 * Module startup and shutdown
1922 static int __init journal_init_caches(void)
1926 ret = journal_init_revoke_caches();
1928 ret = journal_init_journal_head_cache();
1930 ret = journal_init_handle_cache();
1934 static void journal_destroy_caches(void)
1936 journal_destroy_revoke_caches();
1937 journal_destroy_journal_head_cache();
1938 journal_destroy_handle_cache();
1941 static int __init journal_init(void)
1945 ret = journal_init_caches();
1947 journal_destroy_caches();
1948 create_jbd_proc_entry();
1952 static void __exit journal_exit(void)
1954 #ifdef CONFIG_JBD_DEBUG
1955 int n = atomic_read(&nr_journal_heads);
1957 printk(KERN_EMERG "JBD: leaked %d journal_heads!\n", n);
1959 remove_jbd_proc_entry();
1960 journal_destroy_caches();
1963 MODULE_LICENSE("GPL");
1964 module_init(journal_init);
1965 module_exit(journal_exit);