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>
37 #include <linux/proc_fs.h>
39 EXPORT_SYMBOL(journal_start);
40 EXPORT_SYMBOL(journal_restart);
41 EXPORT_SYMBOL(journal_extend);
42 EXPORT_SYMBOL(journal_stop);
43 EXPORT_SYMBOL(journal_lock_updates);
44 EXPORT_SYMBOL(journal_unlock_updates);
45 EXPORT_SYMBOL(journal_get_write_access);
46 EXPORT_SYMBOL(journal_get_create_access);
47 EXPORT_SYMBOL(journal_get_undo_access);
48 EXPORT_SYMBOL(journal_dirty_data);
49 EXPORT_SYMBOL(journal_dirty_metadata);
50 EXPORT_SYMBOL(journal_release_buffer);
51 EXPORT_SYMBOL(journal_forget);
53 EXPORT_SYMBOL(journal_sync_buffer);
55 EXPORT_SYMBOL(journal_flush);
56 EXPORT_SYMBOL(journal_revoke);
57 EXPORT_SYMBOL(journal_callback_set);
59 EXPORT_SYMBOL(journal_init_dev);
60 EXPORT_SYMBOL(journal_init_inode);
61 EXPORT_SYMBOL(journal_update_format);
62 EXPORT_SYMBOL(journal_check_used_features);
63 EXPORT_SYMBOL(journal_check_available_features);
64 EXPORT_SYMBOL(journal_set_features);
65 EXPORT_SYMBOL(journal_create);
66 EXPORT_SYMBOL(journal_load);
67 EXPORT_SYMBOL(journal_destroy);
68 EXPORT_SYMBOL(journal_recover);
69 EXPORT_SYMBOL(journal_update_superblock);
70 EXPORT_SYMBOL(journal_abort);
71 EXPORT_SYMBOL(journal_errno);
72 EXPORT_SYMBOL(journal_ack_err);
73 EXPORT_SYMBOL(journal_clear_err);
74 EXPORT_SYMBOL(log_wait_commit);
75 EXPORT_SYMBOL(journal_start_commit);
76 EXPORT_SYMBOL(journal_force_commit_nested);
77 EXPORT_SYMBOL(journal_wipe);
78 EXPORT_SYMBOL(journal_blocks_per_page);
79 EXPORT_SYMBOL(journal_invalidatepage);
80 EXPORT_SYMBOL(journal_try_to_free_buffers);
81 EXPORT_SYMBOL(journal_bmap);
82 EXPORT_SYMBOL(journal_force_commit);
84 static int journal_convert_superblock_v1(journal_t *, journal_superblock_t *);
87 * Helper function used to manage commit timeouts
90 static void commit_timeout(unsigned long __data)
92 struct task_struct * p = (struct task_struct *) __data;
97 /* Static check for data structure consistency. There's no code
98 * invoked --- we'll just get a linker failure if things aren't right.
100 void __journal_internal_check(void)
102 extern void journal_bad_superblock_size(void);
103 if (sizeof(struct journal_superblock_s) != 1024)
104 journal_bad_superblock_size();
108 * kjournald: The main thread function used to manage a logging device
111 * This kernel thread is responsible for two things:
113 * 1) COMMIT: Every so often we need to commit the current state of the
114 * filesystem to disk. The journal thread is responsible for writing
115 * all of the metadata buffers to disk.
117 * 2) CHECKPOINT: We cannot reuse a used section of the log file until all
118 * of the data in that part of the log has been rewritten elsewhere on
119 * the disk. Flushing these old buffers to reclaim space in the log is
120 * known as checkpointing, and this thread is responsible for that job.
123 journal_t *current_journal; // AKPM: debug
125 int kjournald(void *arg)
127 journal_t *journal = (journal_t *) arg;
128 transaction_t *transaction;
129 struct timer_list timer;
131 current_journal = journal;
133 daemonize("kjournald");
135 /* Set up an interval timer which can be used to trigger a
136 commit wakeup after the commit interval expires */
138 timer.data = (unsigned long) current;
139 timer.function = commit_timeout;
140 journal->j_commit_timer = &timer;
142 /* Record that the journal thread is running */
143 journal->j_task = current;
144 wake_up(&journal->j_wait_done_commit);
146 printk(KERN_INFO "kjournald starting. Commit interval %ld seconds\n",
147 journal->j_commit_interval / HZ);
150 * And now, wait forever for commit wakeup events.
152 spin_lock(&journal->j_state_lock);
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 spin_unlock(&journal->j_state_lock);
161 del_timer_sync(journal->j_commit_timer);
162 journal_commit_transaction(journal);
163 spin_lock(&journal->j_state_lock);
167 wake_up(&journal->j_wait_done_commit);
168 if (current->flags & PF_FREEZE) {
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 kjournald\n");
175 spin_unlock(&journal->j_state_lock);
176 refrigerator(PF_FREEZE);
177 spin_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))
195 spin_unlock(&journal->j_state_lock);
197 spin_lock(&journal->j_state_lock);
199 finish_wait(&journal->j_wait_commit, &wait);
202 jbd_debug(1, "kjournald wakes\n");
205 * Were we woken up by a commit wakeup event?
207 transaction = journal->j_running_transaction;
208 if (transaction && time_after_eq(jiffies, transaction->t_expires)) {
209 journal->j_commit_request = transaction->t_tid;
210 jbd_debug(1, "woke because of timeout\n");
213 if (!(journal->j_flags & JFS_UNMOUNT))
216 spin_unlock(&journal->j_state_lock);
217 del_timer_sync(journal->j_commit_timer);
218 journal->j_task = NULL;
219 wake_up(&journal->j_wait_done_commit);
220 jbd_debug(1, "Journal thread exiting.\n");
224 static void journal_start_thread(journal_t *journal)
226 kernel_thread(kjournald, journal, CLONE_VM|CLONE_FS|CLONE_FILES);
227 wait_event(journal->j_wait_done_commit, journal->j_task != 0);
230 static void journal_kill_thread(journal_t *journal)
232 spin_lock(&journal->j_state_lock);
233 journal->j_flags |= JFS_UNMOUNT;
235 while (journal->j_task) {
236 wake_up(&journal->j_wait_commit);
237 spin_unlock(&journal->j_state_lock);
238 wait_event(journal->j_wait_done_commit, journal->j_task == 0);
239 spin_lock(&journal->j_state_lock);
241 spin_unlock(&journal->j_state_lock);
245 * journal_write_metadata_buffer: write a metadata buffer to the journal.
247 * Writes a metadata buffer to a given disk block. The actual IO is not
248 * performed but a new buffer_head is constructed which labels the data
249 * to be written with the correct destination disk block.
251 * Any magic-number escaping which needs to be done will cause a
252 * copy-out here. If the buffer happens to start with the
253 * JFS_MAGIC_NUMBER, then we can't write it to the log directly: the
254 * magic number is only written to the log for descripter blocks. In
255 * this case, we copy the data and replace the first word with 0, and we
256 * return a result code which indicates that this buffer needs to be
257 * marked as an escaped buffer in the corresponding log descriptor
258 * block. The missing word can then be restored when the block is read
261 * If the source buffer has already been modified by a new transaction
262 * since we took the last commit snapshot, we use the frozen copy of
263 * that data for IO. If we end up using the existing buffer_head's data
264 * for the write, then we *have* to lock the buffer to prevent anyone
265 * else from using and possibly modifying it while the IO is in
268 * The function returns a pointer to the buffer_heads to be used for IO.
270 * We assume that the journal has already been locked in this function.
277 * Bit 0 set == escape performed on the data
278 * Bit 1 set == buffer copy-out performed (kfree the data after IO)
281 int journal_write_metadata_buffer(transaction_t *transaction,
282 struct journal_head *jh_in,
283 struct journal_head **jh_out,
286 int need_copy_out = 0;
287 int done_copy_out = 0;
290 struct buffer_head *new_bh;
291 struct journal_head *new_jh;
292 struct page *new_page;
293 unsigned int new_offset;
294 struct buffer_head *bh_in = jh2bh(jh_in);
297 * The buffer really shouldn't be locked: only the current committing
298 * transaction is allowed to write it, so nobody else is allowed
301 * akpm: except if we're journalling data, and write() output is
302 * also part of a shared mapping, and another thread has
303 * decided to launch a writepage() against this buffer.
305 J_ASSERT_BH(bh_in, buffer_jbddirty(bh_in));
307 new_bh = alloc_buffer_head(GFP_NOFS|__GFP_NOFAIL);
310 * If a new transaction has already done a buffer copy-out, then
311 * we use that version of the data for the commit.
313 jbd_lock_bh_state(bh_in);
315 if (jh_in->b_frozen_data) {
317 new_page = virt_to_page(jh_in->b_frozen_data);
318 new_offset = offset_in_page(jh_in->b_frozen_data);
320 new_page = jh2bh(jh_in)->b_page;
321 new_offset = offset_in_page(jh2bh(jh_in)->b_data);
324 mapped_data = kmap_atomic(new_page, KM_USER0);
328 if (*((unsigned int *)(mapped_data + new_offset)) ==
329 htonl(JFS_MAGIC_NUMBER)) {
333 kunmap_atomic(mapped_data, KM_USER0);
336 * Do we need to do a data copy?
338 if (need_copy_out && !done_copy_out) {
341 jbd_unlock_bh_state(bh_in);
342 tmp = jbd_rep_kmalloc(bh_in->b_size, GFP_NOFS);
343 jbd_lock_bh_state(bh_in);
344 if (jh_in->b_frozen_data) {
349 jh_in->b_frozen_data = tmp;
350 mapped_data = kmap_atomic(new_page, KM_USER0);
351 memcpy(tmp, mapped_data + new_offset, jh2bh(jh_in)->b_size);
352 kunmap_atomic(mapped_data, KM_USER0);
354 new_page = virt_to_page(tmp);
355 new_offset = offset_in_page(tmp);
360 * Did we need to do an escaping? Now we've done all the
361 * copying, we can finally do so.
364 mapped_data = kmap_atomic(new_page, KM_USER0);
365 *((unsigned int *)(mapped_data + new_offset)) = 0;
366 kunmap_atomic(mapped_data, KM_USER0);
369 /* keep subsequent assertions sane */
371 init_buffer(new_bh, NULL, NULL);
372 atomic_set(&new_bh->b_count, 1);
373 jbd_unlock_bh_state(bh_in);
375 new_jh = journal_add_journal_head(new_bh); /* This sleeps */
377 set_bh_page(new_bh, new_page, new_offset);
378 new_jh->b_transaction = NULL;
379 new_bh->b_size = jh2bh(jh_in)->b_size;
380 new_bh->b_bdev = transaction->t_journal->j_dev;
381 new_bh->b_blocknr = blocknr;
382 set_buffer_mapped(new_bh);
383 set_buffer_dirty(new_bh);
388 * The to-be-written buffer needs to get moved to the io queue,
389 * and the original buffer whose contents we are shadowing or
390 * copying is moved to the transaction's shadow queue.
392 JBUFFER_TRACE(jh_in, "file as BJ_Shadow");
393 journal_file_buffer(jh_in, transaction, BJ_Shadow);
394 JBUFFER_TRACE(new_jh, "file as BJ_IO");
395 journal_file_buffer(new_jh, transaction, BJ_IO);
397 return do_escape | (done_copy_out << 1);
401 * Allocation code for the journal file. Manage the space left in the
402 * journal, so that we can begin checkpointing when appropriate.
406 * __log_space_left: Return the number of free blocks left in the journal.
408 * Called with the journal already locked.
410 * Called under j_state_lock
413 int __log_space_left(journal_t *journal)
415 int left = journal->j_free;
417 assert_spin_locked(&journal->j_state_lock);
420 * Be pessimistic here about the number of those free blocks which
421 * might be required for log descriptor control blocks.
424 #define MIN_LOG_RESERVED_BLOCKS 32 /* Allow for rounding errors */
426 left -= MIN_LOG_RESERVED_BLOCKS;
435 * Called under j_state_lock. Returns true if a transaction was started.
437 int __log_start_commit(journal_t *journal, tid_t target)
440 * Are we already doing a recent enough commit?
442 if (!tid_geq(journal->j_commit_request, target)) {
444 * We want a new commit: OK, mark the request and wakup the
445 * commit thread. We do _not_ do the commit ourselves.
448 journal->j_commit_request = target;
449 jbd_debug(1, "JBD: requesting commit %d/%d\n",
450 journal->j_commit_request,
451 journal->j_commit_sequence);
452 wake_up(&journal->j_wait_commit);
458 int log_start_commit(journal_t *journal, tid_t tid)
462 spin_lock(&journal->j_state_lock);
463 ret = __log_start_commit(journal, tid);
464 spin_unlock(&journal->j_state_lock);
469 * Force and wait upon a commit if the calling process is not within
470 * transaction. This is used for forcing out undo-protected data which contains
471 * bitmaps, when the fs is running out of space.
473 * We can only force the running transaction if we don't have an active handle;
474 * otherwise, we will deadlock.
476 * Returns true if a transaction was started.
478 int journal_force_commit_nested(journal_t *journal)
480 transaction_t *transaction = NULL;
483 spin_lock(&journal->j_state_lock);
484 if (journal->j_running_transaction && !current->journal_info) {
485 transaction = journal->j_running_transaction;
486 __log_start_commit(journal, transaction->t_tid);
487 } else if (journal->j_committing_transaction)
488 transaction = journal->j_committing_transaction;
491 spin_unlock(&journal->j_state_lock);
492 return 0; /* Nothing to retry */
495 tid = transaction->t_tid;
496 spin_unlock(&journal->j_state_lock);
497 log_wait_commit(journal, tid);
502 * Start a commit of the current running transaction (if any). Returns true
503 * if a transaction was started, and fills its tid in at *ptid
505 int journal_start_commit(journal_t *journal, tid_t *ptid)
509 spin_lock(&journal->j_state_lock);
510 if (journal->j_running_transaction) {
511 tid_t tid = journal->j_running_transaction->t_tid;
513 ret = __log_start_commit(journal, tid);
516 } else if (journal->j_committing_transaction && ptid) {
518 * If ext3_write_super() recently started a commit, then we
519 * have to wait for completion of that transaction
521 *ptid = journal->j_committing_transaction->t_tid;
524 spin_unlock(&journal->j_state_lock);
529 * Wait for a specified commit to complete.
530 * The caller may not hold the journal lock.
532 int log_wait_commit(journal_t *journal, tid_t tid)
536 #ifdef CONFIG_JBD_DEBUG
537 spin_lock(&journal->j_state_lock);
538 if (!tid_geq(journal->j_commit_request, tid)) {
540 "%s: error: j_commit_request=%d, tid=%d\n",
541 __FUNCTION__, journal->j_commit_request, tid);
543 spin_unlock(&journal->j_state_lock);
545 spin_lock(&journal->j_state_lock);
546 while (tid_gt(tid, journal->j_commit_sequence)) {
547 jbd_debug(1, "JBD: want %d, j_commit_sequence=%d\n",
548 tid, journal->j_commit_sequence);
549 wake_up(&journal->j_wait_commit);
550 spin_unlock(&journal->j_state_lock);
551 wait_event(journal->j_wait_done_commit,
552 !tid_gt(tid, journal->j_commit_sequence));
553 spin_lock(&journal->j_state_lock);
555 spin_unlock(&journal->j_state_lock);
557 if (unlikely(is_journal_aborted(journal))) {
558 printk(KERN_EMERG "journal commit I/O error\n");
565 * Log buffer allocation routines:
568 int journal_next_log_block(journal_t *journal, unsigned long *retp)
570 unsigned long blocknr;
572 spin_lock(&journal->j_state_lock);
573 J_ASSERT(journal->j_free > 1);
575 blocknr = journal->j_head;
578 if (journal->j_head == journal->j_last)
579 journal->j_head = journal->j_first;
580 spin_unlock(&journal->j_state_lock);
581 return journal_bmap(journal, blocknr, retp);
585 * Conversion of logical to physical block numbers for the journal
587 * On external journals the journal blocks are identity-mapped, so
588 * this is a no-op. If needed, we can use j_blk_offset - everything is
591 int journal_bmap(journal_t *journal, unsigned long blocknr,
597 if (journal->j_inode) {
598 ret = bmap(journal->j_inode, blocknr);
602 char b[BDEVNAME_SIZE];
604 printk(KERN_ALERT "%s: journal block not found "
605 "at offset %lu on %s\n",
608 bdevname(journal->j_dev, b));
610 __journal_abort_soft(journal, err);
613 *retp = blocknr; /* +journal->j_blk_offset */
619 * We play buffer_head aliasing tricks to write data/metadata blocks to
620 * the journal without copying their contents, but for journal
621 * descriptor blocks we do need to generate bona fide buffers.
624 struct journal_head * journal_get_descriptor_buffer(journal_t *journal)
626 struct buffer_head *bh;
627 unsigned long blocknr;
630 err = journal_next_log_block(journal, &blocknr);
635 bh = __getblk(journal->j_dev, blocknr, journal->j_blocksize);
636 memset(bh->b_data, 0, journal->j_blocksize);
637 bh->b_state |= (1 << BH_Dirty);
638 BUFFER_TRACE(bh, "return this buffer");
639 return journal_add_journal_head(bh);
643 * Management for journal control blocks: functions to create and
644 * destroy journal_t structures, and to initialise and read existing
645 * journal blocks from disk. */
647 /* First: create and setup a journal_t object in memory. We initialise
648 * very few fields yet: that has to wait until we have created the
649 * journal structures from from scratch, or loaded them from disk. */
651 static journal_t * journal_init_common (void)
656 journal = jbd_kmalloc(sizeof(*journal), GFP_KERNEL);
659 memset(journal, 0, sizeof(*journal));
661 init_waitqueue_head(&journal->j_wait_transaction_locked);
662 init_waitqueue_head(&journal->j_wait_logspace);
663 init_waitqueue_head(&journal->j_wait_done_commit);
664 init_waitqueue_head(&journal->j_wait_checkpoint);
665 init_waitqueue_head(&journal->j_wait_commit);
666 init_waitqueue_head(&journal->j_wait_updates);
667 init_MUTEX(&journal->j_barrier);
668 init_MUTEX(&journal->j_checkpoint_sem);
669 spin_lock_init(&journal->j_revoke_lock);
670 spin_lock_init(&journal->j_list_lock);
671 spin_lock_init(&journal->j_state_lock);
673 journal->j_commit_interval = (HZ * JBD_DEFAULT_MAX_COMMIT_AGE);
675 /* The journal is marked for error until we succeed with recovery! */
676 journal->j_flags = JFS_ABORT;
678 /* Set up a default-sized revoke table for the new mount. */
679 err = journal_init_revoke(journal, JOURNAL_REVOKE_DEFAULT_HASH);
689 /* journal_init_dev and journal_init_inode:
691 * Create a journal structure assigned some fixed set of disk blocks to
692 * the journal. We don't actually touch those disk blocks yet, but we
693 * need to set up all of the mapping information to tell the journaling
694 * system where the journal blocks are.
699 * journal_t * journal_init_dev() - creates an initialises a journal structure
700 * @bdev: Block device on which to create the journal
701 * @fs_dev: Device which hold journalled filesystem for this journal.
702 * @start: Block nr Start of journal.
703 * @len: Lenght of the journal in blocks.
704 * @blocksize: blocksize of journalling device
705 * @returns: a newly created journal_t *
707 * journal_init_dev creates a journal which maps a fixed contiguous
708 * range of blocks on an arbitrary block device.
711 journal_t * journal_init_dev(struct block_device *bdev,
712 struct block_device *fs_dev,
713 int start, int len, int blocksize)
715 journal_t *journal = journal_init_common();
716 struct buffer_head *bh;
721 journal->j_dev = bdev;
722 journal->j_fs_dev = fs_dev;
723 journal->j_blk_offset = start;
724 journal->j_maxlen = len;
725 journal->j_blocksize = blocksize;
727 bh = __getblk(journal->j_dev, start, journal->j_blocksize);
728 J_ASSERT(bh != NULL);
729 journal->j_sb_buffer = bh;
730 journal->j_superblock = (journal_superblock_t *)bh->b_data;
736 * journal_t * journal_init_inode () - creates a journal which maps to a inode.
737 * @inode: An inode to create the journal in
739 * journal_init_inode creates a journal which maps an on-disk inode as
740 * the journal. The inode must exist already, must support bmap() and
741 * must have all data blocks preallocated.
743 journal_t * journal_init_inode (struct inode *inode)
745 struct buffer_head *bh;
746 journal_t *journal = journal_init_common();
748 unsigned long blocknr;
753 journal->j_dev = journal->j_fs_dev = inode->i_sb->s_bdev;
754 journal->j_inode = inode;
756 "journal %p: inode %s/%ld, size %Ld, bits %d, blksize %ld\n",
757 journal, inode->i_sb->s_id, inode->i_ino,
758 (long long) inode->i_size,
759 inode->i_sb->s_blocksize_bits, inode->i_sb->s_blocksize);
761 journal->j_maxlen = inode->i_size >> inode->i_sb->s_blocksize_bits;
762 journal->j_blocksize = inode->i_sb->s_blocksize;
764 err = journal_bmap(journal, 0, &blocknr);
765 /* If that failed, give up */
767 printk(KERN_ERR "%s: Cannnot locate journal superblock\n",
773 bh = __getblk(journal->j_dev, blocknr, journal->j_blocksize);
774 J_ASSERT(bh != NULL);
775 journal->j_sb_buffer = bh;
776 journal->j_superblock = (journal_superblock_t *)bh->b_data;
782 * If the journal init or create aborts, we need to mark the journal
783 * superblock as being NULL to prevent the journal destroy from writing
784 * back a bogus superblock.
786 static void journal_fail_superblock (journal_t *journal)
788 struct buffer_head *bh = journal->j_sb_buffer;
790 journal->j_sb_buffer = NULL;
794 * Given a journal_t structure, initialise the various fields for
795 * startup of a new journaling session. We use this both when creating
796 * a journal, and after recovering an old journal to reset it for
800 static int journal_reset(journal_t *journal)
802 journal_superblock_t *sb = journal->j_superblock;
803 unsigned int first, last;
805 first = ntohl(sb->s_first);
806 last = ntohl(sb->s_maxlen);
808 journal->j_first = first;
809 journal->j_last = last;
811 journal->j_head = first;
812 journal->j_tail = first;
813 journal->j_free = last - first;
815 journal->j_tail_sequence = journal->j_transaction_sequence;
816 journal->j_commit_sequence = journal->j_transaction_sequence - 1;
817 journal->j_commit_request = journal->j_commit_sequence;
819 journal->j_max_transaction_buffers = journal->j_maxlen / 4;
821 /* Add the dynamic fields and write it to disk. */
822 journal_update_superblock(journal, 1);
823 journal_start_thread(journal);
828 * int journal_create() - Initialise the new journal file
829 * @journal: Journal to create. This structure must have been initialised
831 * Given a journal_t structure which tells us which disk blocks we can
832 * use, create a new journal superblock and initialise all of the
833 * journal fields from scratch.
835 int journal_create(journal_t *journal)
837 unsigned long blocknr;
838 struct buffer_head *bh;
839 journal_superblock_t *sb;
842 if (journal->j_maxlen < JFS_MIN_JOURNAL_BLOCKS) {
843 printk (KERN_ERR "Journal length (%d blocks) too short.\n",
845 journal_fail_superblock(journal);
849 if (journal->j_inode == NULL) {
851 * We don't know what block to start at!
854 "%s: creation of journal on external device!\n",
859 /* Zero out the entire journal on disk. We cannot afford to
860 have any blocks on disk beginning with JFS_MAGIC_NUMBER. */
861 jbd_debug(1, "JBD: Zeroing out journal blocks...\n");
862 for (i = 0; i < journal->j_maxlen; i++) {
863 err = journal_bmap(journal, i, &blocknr);
866 bh = __getblk(journal->j_dev, blocknr, journal->j_blocksize);
868 memset (bh->b_data, 0, journal->j_blocksize);
869 BUFFER_TRACE(bh, "marking dirty");
870 mark_buffer_dirty(bh);
871 BUFFER_TRACE(bh, "marking uptodate");
872 set_buffer_uptodate(bh);
877 sync_blockdev(journal->j_dev);
878 jbd_debug(1, "JBD: journal cleared.\n");
880 /* OK, fill in the initial static fields in the new superblock */
881 sb = journal->j_superblock;
883 sb->s_header.h_magic = htonl(JFS_MAGIC_NUMBER);
884 sb->s_header.h_blocktype = htonl(JFS_SUPERBLOCK_V2);
886 sb->s_blocksize = htonl(journal->j_blocksize);
887 sb->s_maxlen = htonl(journal->j_maxlen);
888 sb->s_first = htonl(1);
890 journal->j_transaction_sequence = 1;
892 journal->j_flags &= ~JFS_ABORT;
893 journal->j_format_version = 2;
895 return journal_reset(journal);
899 * void journal_update_superblock() - Update journal sb on disk.
900 * @journal: The journal to update.
901 * @wait: Set to '0' if you don't want to wait for IO completion.
903 * Update a journal's dynamic superblock fields and write it to disk,
904 * optionally waiting for the IO to complete.
906 void journal_update_superblock(journal_t *journal, int wait)
908 journal_superblock_t *sb = journal->j_superblock;
909 struct buffer_head *bh = journal->j_sb_buffer;
912 * As a special case, if the on-disk copy is already marked as needing
913 * no recovery (s_start == 0) and there are no outstanding transactions
914 * in the filesystem, then we can safely defer the superblock update
915 * until the next commit by setting JFS_FLUSHED. This avoids
916 * attempting a write to a potential-readonly device.
918 if (sb->s_start == 0 && journal->j_tail_sequence ==
919 journal->j_transaction_sequence) {
920 jbd_debug(1,"JBD: Skipping superblock update on recovered sb "
921 "(start %ld, seq %d, errno %d)\n",
922 journal->j_tail, journal->j_tail_sequence,
927 spin_lock(&journal->j_state_lock);
928 jbd_debug(1,"JBD: updating superblock (start %ld, seq %d, errno %d)\n",
929 journal->j_tail, journal->j_tail_sequence, journal->j_errno);
931 sb->s_sequence = htonl(journal->j_tail_sequence);
932 sb->s_start = htonl(journal->j_tail);
933 sb->s_errno = htonl(journal->j_errno);
934 spin_unlock(&journal->j_state_lock);
936 BUFFER_TRACE(bh, "marking dirty");
937 mark_buffer_dirty(bh);
939 sync_dirty_buffer(bh);
941 ll_rw_block(WRITE, 1, &bh);
944 /* If we have just flushed the log (by marking s_start==0), then
945 * any future commit will have to be careful to update the
946 * superblock again to re-record the true start of the log. */
948 spin_lock(&journal->j_state_lock);
950 journal->j_flags &= ~JFS_FLUSHED;
952 journal->j_flags |= JFS_FLUSHED;
953 spin_unlock(&journal->j_state_lock);
957 * Read the superblock for a given journal, performing initial
958 * validation of the format.
961 static int journal_get_superblock(journal_t *journal)
963 struct buffer_head *bh;
964 journal_superblock_t *sb;
967 bh = journal->j_sb_buffer;
969 J_ASSERT(bh != NULL);
970 if (!buffer_uptodate(bh)) {
971 ll_rw_block(READ, 1, &bh);
973 if (!buffer_uptodate(bh)) {
975 "JBD: IO error reading journal superblock\n");
980 sb = journal->j_superblock;
984 if (sb->s_header.h_magic != htonl(JFS_MAGIC_NUMBER) ||
985 sb->s_blocksize != htonl(journal->j_blocksize)) {
986 printk(KERN_WARNING "JBD: no valid journal superblock found\n");
990 switch(ntohl(sb->s_header.h_blocktype)) {
991 case JFS_SUPERBLOCK_V1:
992 journal->j_format_version = 1;
994 case JFS_SUPERBLOCK_V2:
995 journal->j_format_version = 2;
998 printk(KERN_WARNING "JBD: unrecognised superblock format ID\n");
1002 if (ntohl(sb->s_maxlen) < journal->j_maxlen)
1003 journal->j_maxlen = ntohl(sb->s_maxlen);
1004 else if (ntohl(sb->s_maxlen) > journal->j_maxlen) {
1005 printk (KERN_WARNING "JBD: journal file too short\n");
1012 journal_fail_superblock(journal);
1017 * Load the on-disk journal superblock and read the key fields into the
1021 static int load_superblock(journal_t *journal)
1024 journal_superblock_t *sb;
1026 err = journal_get_superblock(journal);
1030 sb = journal->j_superblock;
1032 journal->j_tail_sequence = ntohl(sb->s_sequence);
1033 journal->j_tail = ntohl(sb->s_start);
1034 journal->j_first = ntohl(sb->s_first);
1035 journal->j_last = ntohl(sb->s_maxlen);
1036 journal->j_errno = ntohl(sb->s_errno);
1043 * int journal_load() - Read journal from disk.
1044 * @journal: Journal to act on.
1046 * Given a journal_t structure which tells us which disk blocks contain
1047 * a journal, read the journal from disk to initialise the in-memory
1050 int journal_load(journal_t *journal)
1054 err = load_superblock(journal);
1058 /* If this is a V2 superblock, then we have to check the
1059 * features flags on it. */
1061 if (journal->j_format_version >= 2) {
1062 journal_superblock_t *sb = journal->j_superblock;
1064 if ((sb->s_feature_ro_compat &
1065 ~cpu_to_be32(JFS_KNOWN_ROCOMPAT_FEATURES)) ||
1066 (sb->s_feature_incompat &
1067 ~cpu_to_be32(JFS_KNOWN_INCOMPAT_FEATURES))) {
1068 printk (KERN_WARNING
1069 "JBD: Unrecognised features on journal\n");
1074 /* Let the recovery code check whether it needs to recover any
1075 * data from the journal. */
1076 if (journal_recover(journal))
1077 goto recovery_error;
1079 /* OK, we've finished with the dynamic journal bits:
1080 * reinitialise the dynamic contents of the superblock in memory
1081 * and reset them on disk. */
1082 if (journal_reset(journal))
1083 goto recovery_error;
1085 journal->j_flags &= ~JFS_ABORT;
1086 journal->j_flags |= JFS_LOADED;
1090 printk (KERN_WARNING "JBD: recovery failed\n");
1095 * void journal_destroy() - Release a journal_t structure.
1096 * @journal: Journal to act on.
1098 * Release a journal_t structure once it is no longer in use by the
1101 void journal_destroy(journal_t *journal)
1103 /* Wait for the commit thread to wake up and die. */
1104 journal_kill_thread(journal);
1106 /* Force a final log commit */
1107 if (journal->j_running_transaction)
1108 journal_commit_transaction(journal);
1110 /* Force any old transactions to disk */
1112 /* Totally anal locking here... */
1113 spin_lock(&journal->j_list_lock);
1114 while (journal->j_checkpoint_transactions != NULL) {
1115 spin_unlock(&journal->j_list_lock);
1116 log_do_checkpoint(journal);
1117 spin_lock(&journal->j_list_lock);
1120 J_ASSERT(journal->j_running_transaction == NULL);
1121 J_ASSERT(journal->j_committing_transaction == NULL);
1122 J_ASSERT(journal->j_checkpoint_transactions == NULL);
1123 spin_unlock(&journal->j_list_lock);
1125 /* We can now mark the journal as empty. */
1126 journal->j_tail = 0;
1127 journal->j_tail_sequence = ++journal->j_transaction_sequence;
1128 if (journal->j_sb_buffer) {
1129 journal_update_superblock(journal, 1);
1130 brelse(journal->j_sb_buffer);
1133 if (journal->j_inode)
1134 iput(journal->j_inode);
1135 if (journal->j_revoke)
1136 journal_destroy_revoke(journal);
1142 *int journal_check_used_features () - Check if features specified are used.
1144 * Check whether the journal uses all of a given set of
1145 * features. Return true (non-zero) if it does.
1148 int journal_check_used_features (journal_t *journal, unsigned long compat,
1149 unsigned long ro, unsigned long incompat)
1151 journal_superblock_t *sb;
1153 if (!compat && !ro && !incompat)
1155 if (journal->j_format_version == 1)
1158 sb = journal->j_superblock;
1160 if (((be32_to_cpu(sb->s_feature_compat) & compat) == compat) &&
1161 ((be32_to_cpu(sb->s_feature_ro_compat) & ro) == ro) &&
1162 ((be32_to_cpu(sb->s_feature_incompat) & incompat) == incompat))
1169 * int journal_check_available_features() - Check feature set in journalling layer
1171 * Check whether the journaling code supports the use of
1172 * all of a given set of features on this journal. Return true
1173 * (non-zero) if it can. */
1175 int journal_check_available_features (journal_t *journal, unsigned long compat,
1176 unsigned long ro, unsigned long incompat)
1178 journal_superblock_t *sb;
1180 if (!compat && !ro && !incompat)
1183 sb = journal->j_superblock;
1185 /* We can support any known requested features iff the
1186 * superblock is in version 2. Otherwise we fail to support any
1187 * extended sb features. */
1189 if (journal->j_format_version != 2)
1192 if ((compat & JFS_KNOWN_COMPAT_FEATURES) == compat &&
1193 (ro & JFS_KNOWN_ROCOMPAT_FEATURES) == ro &&
1194 (incompat & JFS_KNOWN_INCOMPAT_FEATURES) == incompat)
1201 * int journal_set_features () - Mark a given journal feature in the superblock
1203 * Mark a given journal feature as present on the
1204 * superblock. Returns true if the requested features could be set.
1208 int journal_set_features (journal_t *journal, unsigned long compat,
1209 unsigned long ro, unsigned long incompat)
1211 journal_superblock_t *sb;
1213 if (journal_check_used_features(journal, compat, ro, incompat))
1216 if (!journal_check_available_features(journal, compat, ro, incompat))
1219 jbd_debug(1, "Setting new features 0x%lx/0x%lx/0x%lx\n",
1220 compat, ro, incompat);
1222 sb = journal->j_superblock;
1224 sb->s_feature_compat |= cpu_to_be32(compat);
1225 sb->s_feature_ro_compat |= cpu_to_be32(ro);
1226 sb->s_feature_incompat |= cpu_to_be32(incompat);
1233 * int journal_update_format () - Update on-disk journal structure.
1235 * Given an initialised but unloaded journal struct, poke about in the
1236 * on-disk structure to update it to the most recent supported version.
1238 int journal_update_format (journal_t *journal)
1240 journal_superblock_t *sb;
1243 err = journal_get_superblock(journal);
1247 sb = journal->j_superblock;
1249 switch (ntohl(sb->s_header.h_blocktype)) {
1250 case JFS_SUPERBLOCK_V2:
1252 case JFS_SUPERBLOCK_V1:
1253 return journal_convert_superblock_v1(journal, sb);
1260 static int journal_convert_superblock_v1(journal_t *journal,
1261 journal_superblock_t *sb)
1263 int offset, blocksize;
1264 struct buffer_head *bh;
1267 "JBD: Converting superblock from version 1 to 2.\n");
1269 /* Pre-initialise new fields to zero */
1270 offset = ((char *) &(sb->s_feature_compat)) - ((char *) sb);
1271 blocksize = ntohl(sb->s_blocksize);
1272 memset(&sb->s_feature_compat, 0, blocksize-offset);
1274 sb->s_nr_users = cpu_to_be32(1);
1275 sb->s_header.h_blocktype = cpu_to_be32(JFS_SUPERBLOCK_V2);
1276 journal->j_format_version = 2;
1278 bh = journal->j_sb_buffer;
1279 BUFFER_TRACE(bh, "marking dirty");
1280 mark_buffer_dirty(bh);
1281 sync_dirty_buffer(bh);
1287 * int journal_flush () - Flush journal
1288 * @journal: Journal to act on.
1290 * Flush all data for a given journal to disk and empty the journal.
1291 * Filesystems can use this when remounting readonly to ensure that
1292 * recovery does not need to happen on remount.
1295 int journal_flush(journal_t *journal)
1298 transaction_t *transaction = NULL;
1299 unsigned long old_tail;
1301 spin_lock(&journal->j_state_lock);
1303 /* Force everything buffered to the log... */
1304 if (journal->j_running_transaction) {
1305 transaction = journal->j_running_transaction;
1306 __log_start_commit(journal, transaction->t_tid);
1307 } else if (journal->j_committing_transaction)
1308 transaction = journal->j_committing_transaction;
1310 /* Wait for the log commit to complete... */
1312 tid_t tid = transaction->t_tid;
1314 spin_unlock(&journal->j_state_lock);
1315 log_wait_commit(journal, tid);
1317 spin_unlock(&journal->j_state_lock);
1320 /* ...and flush everything in the log out to disk. */
1321 spin_lock(&journal->j_list_lock);
1322 while (!err && journal->j_checkpoint_transactions != NULL) {
1323 spin_unlock(&journal->j_list_lock);
1324 err = log_do_checkpoint(journal);
1325 spin_lock(&journal->j_list_lock);
1327 spin_unlock(&journal->j_list_lock);
1328 cleanup_journal_tail(journal);
1330 /* Finally, mark the journal as really needing no recovery.
1331 * This sets s_start==0 in the underlying superblock, which is
1332 * the magic code for a fully-recovered superblock. Any future
1333 * commits of data to the journal will restore the current
1335 spin_lock(&journal->j_state_lock);
1336 old_tail = journal->j_tail;
1337 journal->j_tail = 0;
1338 spin_unlock(&journal->j_state_lock);
1339 journal_update_superblock(journal, 1);
1340 spin_lock(&journal->j_state_lock);
1341 journal->j_tail = old_tail;
1343 J_ASSERT(!journal->j_running_transaction);
1344 J_ASSERT(!journal->j_committing_transaction);
1345 J_ASSERT(!journal->j_checkpoint_transactions);
1346 J_ASSERT(journal->j_head == journal->j_tail);
1347 J_ASSERT(journal->j_tail_sequence == journal->j_transaction_sequence);
1348 spin_unlock(&journal->j_state_lock);
1353 * int journal_wipe() - Wipe journal contents
1354 * @journal: Journal to act on.
1355 * @write: flag (see below)
1357 * Wipe out all of the contents of a journal, safely. This will produce
1358 * a warning if the journal contains any valid recovery information.
1359 * Must be called between journal_init_*() and journal_load().
1361 * If 'write' is non-zero, then we wipe out the journal on disk; otherwise
1362 * we merely suppress recovery.
1365 int journal_wipe(journal_t *journal, int write)
1367 journal_superblock_t *sb;
1370 J_ASSERT (!(journal->j_flags & JFS_LOADED));
1372 err = load_superblock(journal);
1376 sb = journal->j_superblock;
1378 if (!journal->j_tail)
1381 printk (KERN_WARNING "JBD: %s recovery information on journal\n",
1382 write ? "Clearing" : "Ignoring");
1384 err = journal_skip_recovery(journal);
1386 journal_update_superblock(journal, 1);
1393 * journal_dev_name: format a character string to describe on what
1394 * device this journal is present.
1397 const char *journal_dev_name(journal_t *journal, char *buffer)
1399 struct block_device *bdev;
1401 if (journal->j_inode)
1402 bdev = journal->j_inode->i_sb->s_bdev;
1404 bdev = journal->j_dev;
1406 return bdevname(bdev, buffer);
1410 * Journal abort has very specific semantics, which we describe
1411 * for journal abort.
1413 * Two internal function, which provide abort to te jbd layer
1418 * Quick version for internal journal use (doesn't lock the journal).
1419 * Aborts hard --- we mark the abort as occurred, but do _nothing_ else,
1420 * and don't attempt to make any other journal updates.
1422 void __journal_abort_hard(journal_t *journal)
1424 transaction_t *transaction;
1425 char b[BDEVNAME_SIZE];
1427 if (journal->j_flags & JFS_ABORT)
1430 printk(KERN_ERR "Aborting journal on device %s.\n",
1431 journal_dev_name(journal, b));
1433 spin_lock(&journal->j_state_lock);
1434 journal->j_flags |= JFS_ABORT;
1435 transaction = journal->j_running_transaction;
1437 __log_start_commit(journal, transaction->t_tid);
1438 spin_unlock(&journal->j_state_lock);
1441 /* Soft abort: record the abort error status in the journal superblock,
1442 * but don't do any other IO. */
1443 void __journal_abort_soft (journal_t *journal, int errno)
1445 if (journal->j_flags & JFS_ABORT)
1448 if (!journal->j_errno)
1449 journal->j_errno = errno;
1451 __journal_abort_hard(journal);
1454 journal_update_superblock(journal, 1);
1458 * void journal_abort () - Shutdown the journal immediately.
1459 * @journal: the journal to shutdown.
1460 * @errno: an error number to record in the journal indicating
1461 * the reason for the shutdown.
1463 * Perform a complete, immediate shutdown of the ENTIRE
1464 * journal (not of a single transaction). This operation cannot be
1465 * undone without closing and reopening the journal.
1467 * The journal_abort function is intended to support higher level error
1468 * recovery mechanisms such as the ext2/ext3 remount-readonly error
1471 * Journal abort has very specific semantics. Any existing dirty,
1472 * unjournaled buffers in the main filesystem will still be written to
1473 * disk by bdflush, but the journaling mechanism will be suspended
1474 * immediately and no further transaction commits will be honoured.
1476 * Any dirty, journaled buffers will be written back to disk without
1477 * hitting the journal. Atomicity cannot be guaranteed on an aborted
1478 * filesystem, but we _do_ attempt to leave as much data as possible
1479 * behind for fsck to use for cleanup.
1481 * Any attempt to get a new transaction handle on a journal which is in
1482 * ABORT state will just result in an -EROFS error return. A
1483 * journal_stop on an existing handle will return -EIO if we have
1484 * entered abort state during the update.
1486 * Recursive transactions are not disturbed by journal abort until the
1487 * final journal_stop, which will receive the -EIO error.
1489 * Finally, the journal_abort call allows the caller to supply an errno
1490 * which will be recorded (if possible) in the journal superblock. This
1491 * allows a client to record failure conditions in the middle of a
1492 * transaction without having to complete the transaction to record the
1493 * failure to disk. ext3_error, for example, now uses this
1496 * Errors which originate from within the journaling layer will NOT
1497 * supply an errno; a null errno implies that absolutely no further
1498 * writes are done to the journal (unless there are any already in
1503 void journal_abort(journal_t *journal, int errno)
1505 __journal_abort_soft(journal, errno);
1509 * int journal_errno () - returns the journal's error state.
1510 * @journal: journal to examine.
1512 * This is the errno numbet set with journal_abort(), the last
1513 * time the journal was mounted - if the journal was stopped
1514 * without calling abort this will be 0.
1516 * If the journal has been aborted on this mount time -EROFS will
1519 int journal_errno(journal_t *journal)
1523 spin_lock(&journal->j_state_lock);
1524 if (journal->j_flags & JFS_ABORT)
1527 err = journal->j_errno;
1528 spin_unlock(&journal->j_state_lock);
1533 * int journal_clear_err () - clears the journal's error state
1535 * An error must be cleared or Acked to take a FS out of readonly
1538 int journal_clear_err(journal_t *journal)
1542 spin_lock(&journal->j_state_lock);
1543 if (journal->j_flags & JFS_ABORT)
1546 journal->j_errno = 0;
1547 spin_unlock(&journal->j_state_lock);
1552 * void journal_ack_err() - Ack journal err.
1554 * An error must be cleared or Acked to take a FS out of readonly
1557 void journal_ack_err(journal_t *journal)
1559 spin_lock(&journal->j_state_lock);
1560 if (journal->j_errno)
1561 journal->j_flags |= JFS_ACK_ERR;
1562 spin_unlock(&journal->j_state_lock);
1565 int journal_blocks_per_page(struct inode *inode)
1567 return 1 << (PAGE_CACHE_SHIFT - inode->i_sb->s_blocksize_bits);
1571 * Simple support for retying memory allocations. Introduced to help to
1572 * debug different VM deadlock avoidance strategies.
1575 * Simple support for retying memory allocations. Introduced to help to
1576 * debug different VM deadlock avoidance strategies.
1578 void * __jbd_kmalloc (const char *where, size_t size, int flags, int retry)
1580 return kmalloc(size, flags | (retry ? __GFP_NOFAIL : 0));
1584 * Journal_head storage management
1586 static kmem_cache_t *journal_head_cache;
1587 #ifdef CONFIG_JBD_DEBUG
1588 static atomic_t nr_journal_heads = ATOMIC_INIT(0);
1591 static int journal_init_journal_head_cache(void)
1595 J_ASSERT(journal_head_cache == 0);
1596 journal_head_cache = kmem_cache_create("journal_head",
1597 sizeof(struct journal_head),
1603 if (journal_head_cache == 0) {
1605 printk(KERN_EMERG "JBD: no memory for journal_head cache\n");
1610 static void journal_destroy_journal_head_cache(void)
1612 J_ASSERT(journal_head_cache != NULL);
1613 kmem_cache_destroy(journal_head_cache);
1614 journal_head_cache = 0;
1618 * journal_head splicing and dicing
1620 static struct journal_head *journal_alloc_journal_head(void)
1622 struct journal_head *ret;
1623 static unsigned long last_warning;
1625 #ifdef CONFIG_JBD_DEBUG
1626 atomic_inc(&nr_journal_heads);
1628 ret = kmem_cache_alloc(journal_head_cache, GFP_NOFS);
1630 jbd_debug(1, "out of memory for journal_head\n");
1631 if (time_after(jiffies, last_warning + 5*HZ)) {
1632 printk(KERN_NOTICE "ENOMEM in %s, retrying.\n",
1634 last_warning = jiffies;
1638 ret = kmem_cache_alloc(journal_head_cache, GFP_NOFS);
1644 static void journal_free_journal_head(struct journal_head *jh)
1646 #ifdef CONFIG_JBD_DEBUG
1647 atomic_dec(&nr_journal_heads);
1648 memset(jh, 0x5b, sizeof(*jh));
1650 kmem_cache_free(journal_head_cache, jh);
1654 * A journal_head is attached to a buffer_head whenever JBD has an
1655 * interest in the buffer.
1657 * Whenever a buffer has an attached journal_head, its ->b_state:BH_JBD bit
1658 * is set. This bit is tested in core kernel code where we need to take
1659 * JBD-specific actions. Testing the zeroness of ->b_private is not reliable
1662 * When a buffer has its BH_JBD bit set, its ->b_count is elevated by one.
1664 * When a buffer has its BH_JBD bit set it is immune from being released by
1665 * core kernel code, mainly via ->b_count.
1667 * A journal_head may be detached from its buffer_head when the journal_head's
1668 * b_transaction, b_cp_transaction and b_next_transaction pointers are NULL.
1669 * Various places in JBD call journal_remove_journal_head() to indicate that the
1670 * journal_head can be dropped if needed.
1672 * Various places in the kernel want to attach a journal_head to a buffer_head
1673 * _before_ attaching the journal_head to a transaction. To protect the
1674 * journal_head in this situation, journal_add_journal_head elevates the
1675 * journal_head's b_jcount refcount by one. The caller must call
1676 * journal_put_journal_head() to undo this.
1678 * So the typical usage would be:
1680 * (Attach a journal_head if needed. Increments b_jcount)
1681 * struct journal_head *jh = journal_add_journal_head(bh);
1683 * jh->b_transaction = xxx;
1684 * journal_put_journal_head(jh);
1686 * Now, the journal_head's b_jcount is zero, but it is safe from being released
1687 * because it has a non-zero b_transaction.
1691 * Give a buffer_head a journal_head.
1693 * Doesn't need the journal lock.
1696 struct journal_head *journal_add_journal_head(struct buffer_head *bh)
1698 struct journal_head *jh;
1699 struct journal_head *new_jh = NULL;
1702 if (!buffer_jbd(bh)) {
1703 new_jh = journal_alloc_journal_head();
1704 memset(new_jh, 0, sizeof(*new_jh));
1707 jbd_lock_bh_journal_head(bh);
1708 if (buffer_jbd(bh)) {
1712 (atomic_read(&bh->b_count) > 0) ||
1713 (bh->b_page && bh->b_page->mapping));
1716 jbd_unlock_bh_journal_head(bh);
1721 new_jh = NULL; /* We consumed it */
1726 BUFFER_TRACE(bh, "added journal_head");
1729 jbd_unlock_bh_journal_head(bh);
1731 journal_free_journal_head(new_jh);
1732 return bh->b_private;
1736 * Grab a ref against this buffer_head's journal_head. If it ended up not
1737 * having a journal_head, return NULL
1739 struct journal_head *journal_grab_journal_head(struct buffer_head *bh)
1741 struct journal_head *jh = NULL;
1743 jbd_lock_bh_journal_head(bh);
1744 if (buffer_jbd(bh)) {
1748 jbd_unlock_bh_journal_head(bh);
1752 static void __journal_remove_journal_head(struct buffer_head *bh)
1754 struct journal_head *jh = bh2jh(bh);
1756 J_ASSERT_JH(jh, jh->b_jcount >= 0);
1759 if (jh->b_jcount == 0) {
1760 if (jh->b_transaction == NULL &&
1761 jh->b_next_transaction == NULL &&
1762 jh->b_cp_transaction == NULL) {
1763 J_ASSERT_BH(bh, buffer_jbd(bh));
1764 J_ASSERT_BH(bh, jh2bh(jh) == bh);
1765 BUFFER_TRACE(bh, "remove journal_head");
1766 if (jh->b_frozen_data) {
1767 printk(KERN_WARNING "%s: freeing "
1770 kfree(jh->b_frozen_data);
1772 if (jh->b_committed_data) {
1773 printk(KERN_WARNING "%s: freeing "
1774 "b_committed_data\n",
1776 kfree(jh->b_committed_data);
1778 bh->b_private = NULL;
1779 jh->b_bh = NULL; /* debug, really */
1780 clear_buffer_jbd(bh);
1782 journal_free_journal_head(jh);
1784 BUFFER_TRACE(bh, "journal_head was locked");
1790 * journal_remove_journal_head(): if the buffer isn't attached to a transaction
1791 * and has a zero b_jcount then remove and release its journal_head. If we did
1792 * see that the buffer is not used by any transaction we also "logically"
1793 * decrement ->b_count.
1795 * We in fact take an additional increment on ->b_count as a convenience,
1796 * because the caller usually wants to do additional things with the bh
1797 * after calling here.
1798 * The caller of journal_remove_journal_head() *must* run __brelse(bh) at some
1799 * time. Once the caller has run __brelse(), the buffer is eligible for
1800 * reaping by try_to_free_buffers().
1802 void journal_remove_journal_head(struct buffer_head *bh)
1804 jbd_lock_bh_journal_head(bh);
1805 __journal_remove_journal_head(bh);
1806 jbd_unlock_bh_journal_head(bh);
1810 * Drop a reference on the passed journal_head. If it fell to zero then try to
1811 * release the journal_head from the buffer_head.
1813 void journal_put_journal_head(struct journal_head *jh)
1815 struct buffer_head *bh = jh2bh(jh);
1817 jbd_lock_bh_journal_head(bh);
1818 J_ASSERT_JH(jh, jh->b_jcount > 0);
1820 if (!jh->b_jcount && !jh->b_transaction) {
1821 __journal_remove_journal_head(bh);
1824 jbd_unlock_bh_journal_head(bh);
1830 #if defined(CONFIG_JBD_DEBUG)
1831 int journal_enable_debug;
1832 EXPORT_SYMBOL(journal_enable_debug);
1835 #if defined(CONFIG_JBD_DEBUG) && defined(CONFIG_PROC_FS)
1837 static struct proc_dir_entry *proc_jbd_debug;
1839 int read_jbd_debug(char *page, char **start, off_t off,
1840 int count, int *eof, void *data)
1844 ret = sprintf(page + off, "%d\n", journal_enable_debug);
1849 int write_jbd_debug(struct file *file, const char __user *buffer,
1850 unsigned long count, void *data)
1854 if (count > ARRAY_SIZE(buf) - 1)
1855 count = ARRAY_SIZE(buf) - 1;
1856 if (copy_from_user(buf, buffer, count))
1858 buf[ARRAY_SIZE(buf) - 1] = '\0';
1859 journal_enable_debug = simple_strtoul(buf, NULL, 10);
1863 #define JBD_PROC_NAME "sys/fs/jbd-debug"
1865 static void __init create_jbd_proc_entry(void)
1867 proc_jbd_debug = create_proc_entry(JBD_PROC_NAME, 0644, NULL);
1868 if (proc_jbd_debug) {
1869 /* Why is this so hard? */
1870 proc_jbd_debug->read_proc = read_jbd_debug;
1871 proc_jbd_debug->write_proc = write_jbd_debug;
1875 static void __exit remove_jbd_proc_entry(void)
1878 remove_proc_entry(JBD_PROC_NAME, NULL);
1883 #define create_jbd_proc_entry() do {} while (0)
1884 #define remove_jbd_proc_entry() do {} while (0)
1888 kmem_cache_t *jbd_handle_cache;
1890 static int __init journal_init_handle_cache(void)
1892 jbd_handle_cache = kmem_cache_create("journal_handle",
1898 if (jbd_handle_cache == NULL) {
1899 printk(KERN_EMERG "JBD: failed to create handle cache\n");
1905 static void journal_destroy_handle_cache(void)
1907 if (jbd_handle_cache)
1908 kmem_cache_destroy(jbd_handle_cache);
1912 * Module startup and shutdown
1915 static int __init journal_init_caches(void)
1919 ret = journal_init_revoke_caches();
1921 ret = journal_init_journal_head_cache();
1923 ret = journal_init_handle_cache();
1927 static void journal_destroy_caches(void)
1929 journal_destroy_revoke_caches();
1930 journal_destroy_journal_head_cache();
1931 journal_destroy_handle_cache();
1934 static int __init journal_init(void)
1938 ret = journal_init_caches();
1940 journal_destroy_caches();
1941 create_jbd_proc_entry();
1945 static void __exit journal_exit(void)
1947 #ifdef CONFIG_JBD_DEBUG
1948 int n = atomic_read(&nr_journal_heads);
1950 printk(KERN_EMERG "JBD: leaked %d journal_heads!\n", n);
1952 remove_jbd_proc_entry();
1953 journal_destroy_caches();
1956 MODULE_LICENSE("GPL");
1957 module_init(journal_init);
1958 module_exit(journal_exit);