2 * linux/fs/jbd/journal.c
4 * Written by Stephen C. Tweedie <sct@redhat.com>, 1998
6 * Copyright 1998 Red Hat corp --- All Rights Reserved
8 * This file is part of the Linux kernel and is made available under
9 * the terms of the GNU General Public License, version 2, or at your
10 * option, any later version, incorporated herein by reference.
12 * Generic filesystem journal-writing code; part of the ext2fs
15 * This file manages journals: areas of disk reserved for logging
16 * transactional updates. This includes the kernel journaling thread
17 * which is responsible for scheduling updates to the log.
19 * We do not actually manage the physical storage of the journal in this
20 * file: that is left to a per-journal policy function, which allows us
21 * to store the journal within a filesystem-specified area for ext2
22 * journaling (ext2 can use a reserved inode for storing the log).
25 #include <linux/module.h>
26 #include <linux/time.h>
28 #include <linux/jbd.h>
29 #include <linux/errno.h>
30 #include <linux/slab.h>
31 #include <linux/init.h>
33 #include <linux/freezer.h>
34 #include <linux/pagemap.h>
35 #include <linux/kthread.h>
36 #include <linux/poison.h>
37 #include <linux/proc_fs.h>
38 #include <linux/debugfs.h>
39 #include <linux/ratelimit.h>
41 #include <asm/uaccess.h>
44 EXPORT_SYMBOL(journal_start);
45 EXPORT_SYMBOL(journal_restart);
46 EXPORT_SYMBOL(journal_extend);
47 EXPORT_SYMBOL(journal_stop);
48 EXPORT_SYMBOL(journal_lock_updates);
49 EXPORT_SYMBOL(journal_unlock_updates);
50 EXPORT_SYMBOL(journal_get_write_access);
51 EXPORT_SYMBOL(journal_get_create_access);
52 EXPORT_SYMBOL(journal_get_undo_access);
53 EXPORT_SYMBOL(journal_dirty_data);
54 EXPORT_SYMBOL(journal_dirty_metadata);
55 EXPORT_SYMBOL(journal_release_buffer);
56 EXPORT_SYMBOL(journal_forget);
58 EXPORT_SYMBOL(journal_sync_buffer);
60 EXPORT_SYMBOL(journal_flush);
61 EXPORT_SYMBOL(journal_revoke);
63 EXPORT_SYMBOL(journal_init_dev);
64 EXPORT_SYMBOL(journal_init_inode);
65 EXPORT_SYMBOL(journal_update_format);
66 EXPORT_SYMBOL(journal_check_used_features);
67 EXPORT_SYMBOL(journal_check_available_features);
68 EXPORT_SYMBOL(journal_set_features);
69 EXPORT_SYMBOL(journal_create);
70 EXPORT_SYMBOL(journal_load);
71 EXPORT_SYMBOL(journal_destroy);
72 EXPORT_SYMBOL(journal_abort);
73 EXPORT_SYMBOL(journal_errno);
74 EXPORT_SYMBOL(journal_ack_err);
75 EXPORT_SYMBOL(journal_clear_err);
76 EXPORT_SYMBOL(log_wait_commit);
77 EXPORT_SYMBOL(log_start_commit);
78 EXPORT_SYMBOL(journal_start_commit);
79 EXPORT_SYMBOL(journal_force_commit_nested);
80 EXPORT_SYMBOL(journal_wipe);
81 EXPORT_SYMBOL(journal_blocks_per_page);
82 EXPORT_SYMBOL(journal_invalidatepage);
83 EXPORT_SYMBOL(journal_try_to_free_buffers);
84 EXPORT_SYMBOL(journal_force_commit);
86 static int journal_convert_superblock_v1(journal_t *, journal_superblock_t *);
87 static void __journal_abort_soft (journal_t *journal, int errno);
88 static const char *journal_dev_name(journal_t *journal, char *buffer);
91 * Helper function used to manage commit timeouts
94 static void commit_timeout(unsigned long __data)
96 struct task_struct * p = (struct task_struct *) __data;
102 * kjournald: The main thread function used to manage a logging device
105 * This kernel thread is responsible for two things:
107 * 1) COMMIT: Every so often we need to commit the current state of the
108 * filesystem to disk. The journal thread is responsible for writing
109 * all of the metadata buffers to disk.
111 * 2) CHECKPOINT: We cannot reuse a used section of the log file until all
112 * of the data in that part of the log has been rewritten elsewhere on
113 * the disk. Flushing these old buffers to reclaim space in the log is
114 * known as checkpointing, and this thread is responsible for that job.
117 static int kjournald(void *arg)
119 journal_t *journal = arg;
120 transaction_t *transaction;
123 * Set up an interval timer which can be used to trigger a commit wakeup
124 * after the commit interval expires
126 setup_timer(&journal->j_commit_timer, commit_timeout,
127 (unsigned long)current);
129 /* Record that the journal thread is running */
130 journal->j_task = current;
131 wake_up(&journal->j_wait_done_commit);
133 printk(KERN_INFO "kjournald starting. Commit interval %ld seconds\n",
134 journal->j_commit_interval / HZ);
137 * And now, wait forever for commit wakeup events.
139 spin_lock(&journal->j_state_lock);
142 if (journal->j_flags & JFS_UNMOUNT)
145 jbd_debug(1, "commit_sequence=%d, commit_request=%d\n",
146 journal->j_commit_sequence, journal->j_commit_request);
148 if (journal->j_commit_sequence != journal->j_commit_request) {
149 jbd_debug(1, "OK, requests differ\n");
150 spin_unlock(&journal->j_state_lock);
151 del_timer_sync(&journal->j_commit_timer);
152 journal_commit_transaction(journal);
153 spin_lock(&journal->j_state_lock);
157 wake_up(&journal->j_wait_done_commit);
158 if (freezing(current)) {
160 * The simpler the better. Flushing journal isn't a
161 * good idea, because that depends on threads that may
162 * be already stopped.
164 jbd_debug(1, "Now suspending kjournald\n");
165 spin_unlock(&journal->j_state_lock);
167 spin_lock(&journal->j_state_lock);
170 * We assume on resume that commits are already there,
174 int should_sleep = 1;
176 prepare_to_wait(&journal->j_wait_commit, &wait,
178 if (journal->j_commit_sequence != journal->j_commit_request)
180 transaction = journal->j_running_transaction;
181 if (transaction && time_after_eq(jiffies,
182 transaction->t_expires))
184 if (journal->j_flags & JFS_UNMOUNT)
187 spin_unlock(&journal->j_state_lock);
189 spin_lock(&journal->j_state_lock);
191 finish_wait(&journal->j_wait_commit, &wait);
194 jbd_debug(1, "kjournald wakes\n");
197 * Were we woken up by a commit wakeup event?
199 transaction = journal->j_running_transaction;
200 if (transaction && time_after_eq(jiffies, transaction->t_expires)) {
201 journal->j_commit_request = transaction->t_tid;
202 jbd_debug(1, "woke because of timeout\n");
207 spin_unlock(&journal->j_state_lock);
208 del_timer_sync(&journal->j_commit_timer);
209 journal->j_task = NULL;
210 wake_up(&journal->j_wait_done_commit);
211 jbd_debug(1, "Journal thread exiting.\n");
215 static int journal_start_thread(journal_t *journal)
217 struct task_struct *t;
219 t = kthread_run(kjournald, journal, "kjournald");
223 wait_event(journal->j_wait_done_commit, journal->j_task != NULL);
227 static void journal_kill_thread(journal_t *journal)
229 spin_lock(&journal->j_state_lock);
230 journal->j_flags |= JFS_UNMOUNT;
232 while (journal->j_task) {
233 wake_up(&journal->j_wait_commit);
234 spin_unlock(&journal->j_state_lock);
235 wait_event(journal->j_wait_done_commit,
236 journal->j_task == NULL);
237 spin_lock(&journal->j_state_lock);
239 spin_unlock(&journal->j_state_lock);
243 * journal_write_metadata_buffer: write a metadata buffer to the journal.
245 * Writes a metadata buffer to a given disk block. The actual IO is not
246 * performed but a new buffer_head is constructed which labels the data
247 * to be written with the correct destination disk block.
249 * Any magic-number escaping which needs to be done will cause a
250 * copy-out here. If the buffer happens to start with the
251 * JFS_MAGIC_NUMBER, then we can't write it to the log directly: the
252 * magic number is only written to the log for descripter blocks. In
253 * this case, we copy the data and replace the first word with 0, and we
254 * return a result code which indicates that this buffer needs to be
255 * marked as an escaped buffer in the corresponding log descriptor
256 * block. The missing word can then be restored when the block is read
259 * If the source buffer has already been modified by a new transaction
260 * since we took the last commit snapshot, we use the frozen copy of
261 * that data for IO. If we end up using the existing buffer_head's data
262 * for the write, then we *have* to lock the buffer to prevent anyone
263 * else from using and possibly modifying it while the IO is in
266 * The function returns a pointer to the buffer_heads to be used for IO.
268 * We assume that the journal has already been locked in this function.
275 * Bit 0 set == escape performed on the data
276 * Bit 1 set == buffer copy-out performed (kfree the data after IO)
279 int journal_write_metadata_buffer(transaction_t *transaction,
280 struct journal_head *jh_in,
281 struct journal_head **jh_out,
282 unsigned int blocknr)
284 int need_copy_out = 0;
285 int done_copy_out = 0;
288 struct buffer_head *new_bh;
289 struct journal_head *new_jh;
290 struct page *new_page;
291 unsigned int new_offset;
292 struct buffer_head *bh_in = jh2bh(jh_in);
293 journal_t *journal = transaction->t_journal;
296 * The buffer really shouldn't be locked: only the current committing
297 * transaction is allowed to write it, so nobody else is allowed
300 * akpm: except if we're journalling data, and write() output is
301 * also part of a shared mapping, and another thread has
302 * decided to launch a writepage() against this buffer.
304 J_ASSERT_BH(bh_in, buffer_jbddirty(bh_in));
306 new_bh = alloc_buffer_head(GFP_NOFS|__GFP_NOFAIL);
307 /* keep subsequent assertions sane */
309 init_buffer(new_bh, NULL, NULL);
310 atomic_set(&new_bh->b_count, 1);
311 new_jh = journal_add_journal_head(new_bh); /* This sleeps */
314 * If a new transaction has already done a buffer copy-out, then
315 * we use that version of the data for the commit.
317 jbd_lock_bh_state(bh_in);
319 if (jh_in->b_frozen_data) {
321 new_page = virt_to_page(jh_in->b_frozen_data);
322 new_offset = offset_in_page(jh_in->b_frozen_data);
324 new_page = jh2bh(jh_in)->b_page;
325 new_offset = offset_in_page(jh2bh(jh_in)->b_data);
328 mapped_data = kmap_atomic(new_page, KM_USER0);
332 if (*((__be32 *)(mapped_data + new_offset)) ==
333 cpu_to_be32(JFS_MAGIC_NUMBER)) {
337 kunmap_atomic(mapped_data, KM_USER0);
340 * Do we need to do a data copy?
342 if (need_copy_out && !done_copy_out) {
345 jbd_unlock_bh_state(bh_in);
346 tmp = jbd_alloc(bh_in->b_size, GFP_NOFS);
347 jbd_lock_bh_state(bh_in);
348 if (jh_in->b_frozen_data) {
349 jbd_free(tmp, bh_in->b_size);
353 jh_in->b_frozen_data = tmp;
354 mapped_data = kmap_atomic(new_page, KM_USER0);
355 memcpy(tmp, mapped_data + new_offset, jh2bh(jh_in)->b_size);
356 kunmap_atomic(mapped_data, KM_USER0);
358 new_page = virt_to_page(tmp);
359 new_offset = offset_in_page(tmp);
364 * Did we need to do an escaping? Now we've done all the
365 * copying, we can finally do so.
368 mapped_data = kmap_atomic(new_page, KM_USER0);
369 *((unsigned int *)(mapped_data + new_offset)) = 0;
370 kunmap_atomic(mapped_data, KM_USER0);
373 set_bh_page(new_bh, new_page, new_offset);
374 new_jh->b_transaction = NULL;
375 new_bh->b_size = jh2bh(jh_in)->b_size;
376 new_bh->b_bdev = transaction->t_journal->j_dev;
377 new_bh->b_blocknr = blocknr;
378 set_buffer_mapped(new_bh);
379 set_buffer_dirty(new_bh);
384 * The to-be-written buffer needs to get moved to the io queue,
385 * and the original buffer whose contents we are shadowing or
386 * copying is moved to the transaction's shadow queue.
388 JBUFFER_TRACE(jh_in, "file as BJ_Shadow");
389 spin_lock(&journal->j_list_lock);
390 __journal_file_buffer(jh_in, transaction, BJ_Shadow);
391 spin_unlock(&journal->j_list_lock);
392 jbd_unlock_bh_state(bh_in);
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 commit 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 is going to be committed (or is currently already
504 * committing), 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 __log_start_commit(journal, tid);
515 /* There's a running transaction and we've just made sure
516 * it's commit has been scheduled. */
520 } else if (journal->j_committing_transaction) {
522 * If ext3_write_super() recently started a commit, then we
523 * have to wait for completion of that transaction
526 *ptid = journal->j_committing_transaction->t_tid;
529 spin_unlock(&journal->j_state_lock);
534 * Wait for a specified commit to complete.
535 * The caller may not hold the journal lock.
537 int log_wait_commit(journal_t *journal, tid_t tid)
541 #ifdef CONFIG_JBD_DEBUG
542 spin_lock(&journal->j_state_lock);
543 if (!tid_geq(journal->j_commit_request, tid)) {
545 "%s: error: j_commit_request=%d, tid=%d\n",
546 __func__, journal->j_commit_request, tid);
548 spin_unlock(&journal->j_state_lock);
550 spin_lock(&journal->j_state_lock);
551 while (tid_gt(tid, journal->j_commit_sequence)) {
552 jbd_debug(1, "JBD: want %d, j_commit_sequence=%d\n",
553 tid, journal->j_commit_sequence);
554 wake_up(&journal->j_wait_commit);
555 spin_unlock(&journal->j_state_lock);
556 wait_event(journal->j_wait_done_commit,
557 !tid_gt(tid, journal->j_commit_sequence));
558 spin_lock(&journal->j_state_lock);
560 spin_unlock(&journal->j_state_lock);
562 if (unlikely(is_journal_aborted(journal))) {
563 printk(KERN_EMERG "journal commit I/O error\n");
570 * Return 1 if a given transaction has not yet sent barrier request
571 * connected with a transaction commit. If 0 is returned, transaction
572 * may or may not have sent the barrier. Used to avoid sending barrier
573 * twice in common cases.
575 int journal_trans_will_send_data_barrier(journal_t *journal, tid_t tid)
578 transaction_t *commit_trans;
580 if (!(journal->j_flags & JFS_BARRIER))
582 spin_lock(&journal->j_state_lock);
583 /* Transaction already committed? */
584 if (tid_geq(journal->j_commit_sequence, tid))
587 * Transaction is being committed and we already proceeded to
588 * writing commit record?
590 commit_trans = journal->j_committing_transaction;
591 if (commit_trans && commit_trans->t_tid == tid &&
592 commit_trans->t_state >= T_COMMIT_RECORD)
596 spin_unlock(&journal->j_state_lock);
599 EXPORT_SYMBOL(journal_trans_will_send_data_barrier);
602 * Log buffer allocation routines:
605 int journal_next_log_block(journal_t *journal, unsigned int *retp)
607 unsigned int blocknr;
609 spin_lock(&journal->j_state_lock);
610 J_ASSERT(journal->j_free > 1);
612 blocknr = journal->j_head;
615 if (journal->j_head == journal->j_last)
616 journal->j_head = journal->j_first;
617 spin_unlock(&journal->j_state_lock);
618 return journal_bmap(journal, blocknr, retp);
622 * Conversion of logical to physical block numbers for the journal
624 * On external journals the journal blocks are identity-mapped, so
625 * this is a no-op. If needed, we can use j_blk_offset - everything is
628 int journal_bmap(journal_t *journal, unsigned int blocknr,
634 if (journal->j_inode) {
635 ret = bmap(journal->j_inode, blocknr);
639 char b[BDEVNAME_SIZE];
641 printk(KERN_ALERT "%s: journal block not found "
642 "at offset %u on %s\n",
645 bdevname(journal->j_dev, b));
647 __journal_abort_soft(journal, err);
650 *retp = blocknr; /* +journal->j_blk_offset */
656 * We play buffer_head aliasing tricks to write data/metadata blocks to
657 * the journal without copying their contents, but for journal
658 * descriptor blocks we do need to generate bona fide buffers.
660 * After the caller of journal_get_descriptor_buffer() has finished modifying
661 * the buffer's contents they really should run flush_dcache_page(bh->b_page).
662 * But we don't bother doing that, so there will be coherency problems with
663 * mmaps of blockdevs which hold live JBD-controlled filesystems.
665 struct journal_head *journal_get_descriptor_buffer(journal_t *journal)
667 struct buffer_head *bh;
668 unsigned int blocknr;
671 err = journal_next_log_block(journal, &blocknr);
676 bh = __getblk(journal->j_dev, blocknr, journal->j_blocksize);
680 memset(bh->b_data, 0, journal->j_blocksize);
681 set_buffer_uptodate(bh);
683 BUFFER_TRACE(bh, "return this buffer");
684 return journal_add_journal_head(bh);
688 * Management for journal control blocks: functions to create and
689 * destroy journal_t structures, and to initialise and read existing
690 * journal blocks from disk. */
692 /* First: create and setup a journal_t object in memory. We initialise
693 * very few fields yet: that has to wait until we have created the
694 * journal structures from from scratch, or loaded them from disk. */
696 static journal_t * journal_init_common (void)
701 journal = kzalloc(sizeof(*journal), GFP_KERNEL);
705 init_waitqueue_head(&journal->j_wait_transaction_locked);
706 init_waitqueue_head(&journal->j_wait_logspace);
707 init_waitqueue_head(&journal->j_wait_done_commit);
708 init_waitqueue_head(&journal->j_wait_checkpoint);
709 init_waitqueue_head(&journal->j_wait_commit);
710 init_waitqueue_head(&journal->j_wait_updates);
711 mutex_init(&journal->j_barrier);
712 mutex_init(&journal->j_checkpoint_mutex);
713 spin_lock_init(&journal->j_revoke_lock);
714 spin_lock_init(&journal->j_list_lock);
715 spin_lock_init(&journal->j_state_lock);
717 journal->j_commit_interval = (HZ * JBD_DEFAULT_MAX_COMMIT_AGE);
719 /* The journal is marked for error until we succeed with recovery! */
720 journal->j_flags = JFS_ABORT;
722 /* Set up a default-sized revoke table for the new mount. */
723 err = journal_init_revoke(journal, JOURNAL_REVOKE_DEFAULT_HASH);
733 /* journal_init_dev and journal_init_inode:
735 * Create a journal structure assigned some fixed set of disk blocks to
736 * the journal. We don't actually touch those disk blocks yet, but we
737 * need to set up all of the mapping information to tell the journaling
738 * system where the journal blocks are.
743 * journal_t * journal_init_dev() - creates and initialises a journal structure
744 * @bdev: Block device on which to create the journal
745 * @fs_dev: Device which hold journalled filesystem for this journal.
746 * @start: Block nr Start of journal.
747 * @len: Length of the journal in blocks.
748 * @blocksize: blocksize of journalling device
750 * Returns: a newly created journal_t *
752 * journal_init_dev creates a journal which maps a fixed contiguous
753 * range of blocks on an arbitrary block device.
756 journal_t * journal_init_dev(struct block_device *bdev,
757 struct block_device *fs_dev,
758 int start, int len, int blocksize)
760 journal_t *journal = journal_init_common();
761 struct buffer_head *bh;
767 /* journal descriptor can store up to n blocks -bzzz */
768 journal->j_blocksize = blocksize;
769 n = journal->j_blocksize / sizeof(journal_block_tag_t);
770 journal->j_wbufsize = n;
771 journal->j_wbuf = kmalloc(n * sizeof(struct buffer_head*), GFP_KERNEL);
772 if (!journal->j_wbuf) {
773 printk(KERN_ERR "%s: Cant allocate bhs for commit thread\n",
777 journal->j_dev = bdev;
778 journal->j_fs_dev = fs_dev;
779 journal->j_blk_offset = start;
780 journal->j_maxlen = len;
782 bh = __getblk(journal->j_dev, start, journal->j_blocksize);
785 "%s: Cannot get buffer for journal superblock\n",
789 journal->j_sb_buffer = bh;
790 journal->j_superblock = (journal_superblock_t *)bh->b_data;
794 kfree(journal->j_wbuf);
800 * journal_t * journal_init_inode () - creates a journal which maps to a inode.
801 * @inode: An inode to create the journal in
803 * journal_init_inode creates a journal which maps an on-disk inode as
804 * the journal. The inode must exist already, must support bmap() and
805 * must have all data blocks preallocated.
807 journal_t * journal_init_inode (struct inode *inode)
809 struct buffer_head *bh;
810 journal_t *journal = journal_init_common();
813 unsigned int blocknr;
818 journal->j_dev = journal->j_fs_dev = inode->i_sb->s_bdev;
819 journal->j_inode = inode;
821 "journal %p: inode %s/%ld, size %Ld, bits %d, blksize %ld\n",
822 journal, inode->i_sb->s_id, inode->i_ino,
823 (long long) inode->i_size,
824 inode->i_sb->s_blocksize_bits, inode->i_sb->s_blocksize);
826 journal->j_maxlen = inode->i_size >> inode->i_sb->s_blocksize_bits;
827 journal->j_blocksize = inode->i_sb->s_blocksize;
829 /* journal descriptor can store up to n blocks -bzzz */
830 n = journal->j_blocksize / sizeof(journal_block_tag_t);
831 journal->j_wbufsize = n;
832 journal->j_wbuf = kmalloc(n * sizeof(struct buffer_head*), GFP_KERNEL);
833 if (!journal->j_wbuf) {
834 printk(KERN_ERR "%s: Cant allocate bhs for commit thread\n",
839 err = journal_bmap(journal, 0, &blocknr);
840 /* If that failed, give up */
842 printk(KERN_ERR "%s: Cannnot locate journal superblock\n",
847 bh = __getblk(journal->j_dev, blocknr, journal->j_blocksize);
850 "%s: Cannot get buffer for journal superblock\n",
854 journal->j_sb_buffer = bh;
855 journal->j_superblock = (journal_superblock_t *)bh->b_data;
859 kfree(journal->j_wbuf);
865 * If the journal init or create aborts, we need to mark the journal
866 * superblock as being NULL to prevent the journal destroy from writing
867 * back a bogus superblock.
869 static void journal_fail_superblock (journal_t *journal)
871 struct buffer_head *bh = journal->j_sb_buffer;
873 journal->j_sb_buffer = NULL;
877 * Given a journal_t structure, initialise the various fields for
878 * startup of a new journaling session. We use this both when creating
879 * a journal, and after recovering an old journal to reset it for
883 static int journal_reset(journal_t *journal)
885 journal_superblock_t *sb = journal->j_superblock;
886 unsigned int first, last;
888 first = be32_to_cpu(sb->s_first);
889 last = be32_to_cpu(sb->s_maxlen);
890 if (first + JFS_MIN_JOURNAL_BLOCKS > last + 1) {
891 printk(KERN_ERR "JBD: Journal too short (blocks %u-%u).\n",
893 journal_fail_superblock(journal);
897 journal->j_first = first;
898 journal->j_last = last;
900 journal->j_head = first;
901 journal->j_tail = first;
902 journal->j_free = last - first;
904 journal->j_tail_sequence = journal->j_transaction_sequence;
905 journal->j_commit_sequence = journal->j_transaction_sequence - 1;
906 journal->j_commit_request = journal->j_commit_sequence;
908 journal->j_max_transaction_buffers = journal->j_maxlen / 4;
910 /* Add the dynamic fields and write it to disk. */
911 journal_update_superblock(journal, 1);
912 return journal_start_thread(journal);
916 * int journal_create() - Initialise the new journal file
917 * @journal: Journal to create. This structure must have been initialised
919 * Given a journal_t structure which tells us which disk blocks we can
920 * use, create a new journal superblock and initialise all of the
921 * journal fields from scratch.
923 int journal_create(journal_t *journal)
925 unsigned int blocknr;
926 struct buffer_head *bh;
927 journal_superblock_t *sb;
930 if (journal->j_maxlen < JFS_MIN_JOURNAL_BLOCKS) {
931 printk (KERN_ERR "Journal length (%d blocks) too short.\n",
933 journal_fail_superblock(journal);
937 if (journal->j_inode == NULL) {
939 * We don't know what block to start at!
942 "%s: creation of journal on external device!\n",
947 /* Zero out the entire journal on disk. We cannot afford to
948 have any blocks on disk beginning with JFS_MAGIC_NUMBER. */
949 jbd_debug(1, "JBD: Zeroing out journal blocks...\n");
950 for (i = 0; i < journal->j_maxlen; i++) {
951 err = journal_bmap(journal, i, &blocknr);
954 bh = __getblk(journal->j_dev, blocknr, journal->j_blocksize);
956 memset (bh->b_data, 0, journal->j_blocksize);
957 BUFFER_TRACE(bh, "marking dirty");
958 mark_buffer_dirty(bh);
959 BUFFER_TRACE(bh, "marking uptodate");
960 set_buffer_uptodate(bh);
965 sync_blockdev(journal->j_dev);
966 jbd_debug(1, "JBD: journal cleared.\n");
968 /* OK, fill in the initial static fields in the new superblock */
969 sb = journal->j_superblock;
971 sb->s_header.h_magic = cpu_to_be32(JFS_MAGIC_NUMBER);
972 sb->s_header.h_blocktype = cpu_to_be32(JFS_SUPERBLOCK_V2);
974 sb->s_blocksize = cpu_to_be32(journal->j_blocksize);
975 sb->s_maxlen = cpu_to_be32(journal->j_maxlen);
976 sb->s_first = cpu_to_be32(1);
978 journal->j_transaction_sequence = 1;
980 journal->j_flags &= ~JFS_ABORT;
981 journal->j_format_version = 2;
983 return journal_reset(journal);
987 * void journal_update_superblock() - Update journal sb on disk.
988 * @journal: The journal to update.
989 * @wait: Set to '0' if you don't want to wait for IO completion.
991 * Update a journal's dynamic superblock fields and write it to disk,
992 * optionally waiting for the IO to complete.
994 void journal_update_superblock(journal_t *journal, int wait)
996 journal_superblock_t *sb = journal->j_superblock;
997 struct buffer_head *bh = journal->j_sb_buffer;
1000 * As a special case, if the on-disk copy is already marked as needing
1001 * no recovery (s_start == 0) and there are no outstanding transactions
1002 * in the filesystem, then we can safely defer the superblock update
1003 * until the next commit by setting JFS_FLUSHED. This avoids
1004 * attempting a write to a potential-readonly device.
1006 if (sb->s_start == 0 && journal->j_tail_sequence ==
1007 journal->j_transaction_sequence) {
1008 jbd_debug(1,"JBD: Skipping superblock update on recovered sb "
1009 "(start %u, seq %d, errno %d)\n",
1010 journal->j_tail, journal->j_tail_sequence,
1015 if (buffer_write_io_error(bh)) {
1016 char b[BDEVNAME_SIZE];
1018 * Oh, dear. A previous attempt to write the journal
1019 * superblock failed. This could happen because the
1020 * USB device was yanked out. Or it could happen to
1021 * be a transient write error and maybe the block will
1022 * be remapped. Nothing we can do but to retry the
1023 * write and hope for the best.
1025 printk(KERN_ERR "JBD: previous I/O error detected "
1026 "for journal superblock update for %s.\n",
1027 journal_dev_name(journal, b));
1028 clear_buffer_write_io_error(bh);
1029 set_buffer_uptodate(bh);
1032 spin_lock(&journal->j_state_lock);
1033 jbd_debug(1,"JBD: updating superblock (start %u, seq %d, errno %d)\n",
1034 journal->j_tail, journal->j_tail_sequence, journal->j_errno);
1036 sb->s_sequence = cpu_to_be32(journal->j_tail_sequence);
1037 sb->s_start = cpu_to_be32(journal->j_tail);
1038 sb->s_errno = cpu_to_be32(journal->j_errno);
1039 spin_unlock(&journal->j_state_lock);
1041 BUFFER_TRACE(bh, "marking dirty");
1042 mark_buffer_dirty(bh);
1044 sync_dirty_buffer(bh);
1045 if (buffer_write_io_error(bh)) {
1046 char b[BDEVNAME_SIZE];
1047 printk(KERN_ERR "JBD: I/O error detected "
1048 "when updating journal superblock for %s.\n",
1049 journal_dev_name(journal, b));
1050 clear_buffer_write_io_error(bh);
1051 set_buffer_uptodate(bh);
1054 write_dirty_buffer(bh, WRITE);
1057 /* If we have just flushed the log (by marking s_start==0), then
1058 * any future commit will have to be careful to update the
1059 * superblock again to re-record the true start of the log. */
1061 spin_lock(&journal->j_state_lock);
1063 journal->j_flags &= ~JFS_FLUSHED;
1065 journal->j_flags |= JFS_FLUSHED;
1066 spin_unlock(&journal->j_state_lock);
1070 * Read the superblock for a given journal, performing initial
1071 * validation of the format.
1074 static int journal_get_superblock(journal_t *journal)
1076 struct buffer_head *bh;
1077 journal_superblock_t *sb;
1080 bh = journal->j_sb_buffer;
1082 J_ASSERT(bh != NULL);
1083 if (!buffer_uptodate(bh)) {
1084 ll_rw_block(READ, 1, &bh);
1086 if (!buffer_uptodate(bh)) {
1088 "JBD: IO error reading journal superblock\n");
1093 sb = journal->j_superblock;
1097 if (sb->s_header.h_magic != cpu_to_be32(JFS_MAGIC_NUMBER) ||
1098 sb->s_blocksize != cpu_to_be32(journal->j_blocksize)) {
1099 printk(KERN_WARNING "JBD: no valid journal superblock found\n");
1103 switch(be32_to_cpu(sb->s_header.h_blocktype)) {
1104 case JFS_SUPERBLOCK_V1:
1105 journal->j_format_version = 1;
1107 case JFS_SUPERBLOCK_V2:
1108 journal->j_format_version = 2;
1111 printk(KERN_WARNING "JBD: unrecognised superblock format ID\n");
1115 if (be32_to_cpu(sb->s_maxlen) < journal->j_maxlen)
1116 journal->j_maxlen = be32_to_cpu(sb->s_maxlen);
1117 else if (be32_to_cpu(sb->s_maxlen) > journal->j_maxlen) {
1118 printk (KERN_WARNING "JBD: journal file too short\n");
1125 journal_fail_superblock(journal);
1130 * Load the on-disk journal superblock and read the key fields into the
1134 static int load_superblock(journal_t *journal)
1137 journal_superblock_t *sb;
1139 err = journal_get_superblock(journal);
1143 sb = journal->j_superblock;
1145 journal->j_tail_sequence = be32_to_cpu(sb->s_sequence);
1146 journal->j_tail = be32_to_cpu(sb->s_start);
1147 journal->j_first = be32_to_cpu(sb->s_first);
1148 journal->j_last = be32_to_cpu(sb->s_maxlen);
1149 journal->j_errno = be32_to_cpu(sb->s_errno);
1156 * int journal_load() - Read journal from disk.
1157 * @journal: Journal to act on.
1159 * Given a journal_t structure which tells us which disk blocks contain
1160 * a journal, read the journal from disk to initialise the in-memory
1163 int journal_load(journal_t *journal)
1166 journal_superblock_t *sb;
1168 err = load_superblock(journal);
1172 sb = journal->j_superblock;
1173 /* If this is a V2 superblock, then we have to check the
1174 * features flags on it. */
1176 if (journal->j_format_version >= 2) {
1177 if ((sb->s_feature_ro_compat &
1178 ~cpu_to_be32(JFS_KNOWN_ROCOMPAT_FEATURES)) ||
1179 (sb->s_feature_incompat &
1180 ~cpu_to_be32(JFS_KNOWN_INCOMPAT_FEATURES))) {
1181 printk (KERN_WARNING
1182 "JBD: Unrecognised features on journal\n");
1187 /* Let the recovery code check whether it needs to recover any
1188 * data from the journal. */
1189 if (journal_recover(journal))
1190 goto recovery_error;
1192 /* OK, we've finished with the dynamic journal bits:
1193 * reinitialise the dynamic contents of the superblock in memory
1194 * and reset them on disk. */
1195 if (journal_reset(journal))
1196 goto recovery_error;
1198 journal->j_flags &= ~JFS_ABORT;
1199 journal->j_flags |= JFS_LOADED;
1203 printk (KERN_WARNING "JBD: recovery failed\n");
1208 * void journal_destroy() - Release a journal_t structure.
1209 * @journal: Journal to act on.
1211 * Release a journal_t structure once it is no longer in use by the
1213 * Return <0 if we couldn't clean up the journal.
1215 int journal_destroy(journal_t *journal)
1220 /* Wait for the commit thread to wake up and die. */
1221 journal_kill_thread(journal);
1223 /* Force a final log commit */
1224 if (journal->j_running_transaction)
1225 journal_commit_transaction(journal);
1227 /* Force any old transactions to disk */
1229 /* Totally anal locking here... */
1230 spin_lock(&journal->j_list_lock);
1231 while (journal->j_checkpoint_transactions != NULL) {
1232 spin_unlock(&journal->j_list_lock);
1233 log_do_checkpoint(journal);
1234 spin_lock(&journal->j_list_lock);
1237 J_ASSERT(journal->j_running_transaction == NULL);
1238 J_ASSERT(journal->j_committing_transaction == NULL);
1239 J_ASSERT(journal->j_checkpoint_transactions == NULL);
1240 spin_unlock(&journal->j_list_lock);
1242 if (journal->j_sb_buffer) {
1243 if (!is_journal_aborted(journal)) {
1244 /* We can now mark the journal as empty. */
1245 journal->j_tail = 0;
1246 journal->j_tail_sequence =
1247 ++journal->j_transaction_sequence;
1248 journal_update_superblock(journal, 1);
1252 brelse(journal->j_sb_buffer);
1255 if (journal->j_inode)
1256 iput(journal->j_inode);
1257 if (journal->j_revoke)
1258 journal_destroy_revoke(journal);
1259 kfree(journal->j_wbuf);
1267 *int journal_check_used_features () - Check if features specified are used.
1268 * @journal: Journal to check.
1269 * @compat: bitmask of compatible features
1270 * @ro: bitmask of features that force read-only mount
1271 * @incompat: bitmask of incompatible features
1273 * Check whether the journal uses all of a given set of
1274 * features. Return true (non-zero) if it does.
1277 int journal_check_used_features (journal_t *journal, unsigned long compat,
1278 unsigned long ro, unsigned long incompat)
1280 journal_superblock_t *sb;
1282 if (!compat && !ro && !incompat)
1284 if (journal->j_format_version == 1)
1287 sb = journal->j_superblock;
1289 if (((be32_to_cpu(sb->s_feature_compat) & compat) == compat) &&
1290 ((be32_to_cpu(sb->s_feature_ro_compat) & ro) == ro) &&
1291 ((be32_to_cpu(sb->s_feature_incompat) & incompat) == incompat))
1298 * int journal_check_available_features() - Check feature set in journalling layer
1299 * @journal: Journal to check.
1300 * @compat: bitmask of compatible features
1301 * @ro: bitmask of features that force read-only mount
1302 * @incompat: bitmask of incompatible features
1304 * Check whether the journaling code supports the use of
1305 * all of a given set of features on this journal. Return true
1306 * (non-zero) if it can. */
1308 int journal_check_available_features (journal_t *journal, unsigned long compat,
1309 unsigned long ro, unsigned long incompat)
1311 if (!compat && !ro && !incompat)
1314 /* We can support any known requested features iff the
1315 * superblock is in version 2. Otherwise we fail to support any
1316 * extended sb features. */
1318 if (journal->j_format_version != 2)
1321 if ((compat & JFS_KNOWN_COMPAT_FEATURES) == compat &&
1322 (ro & JFS_KNOWN_ROCOMPAT_FEATURES) == ro &&
1323 (incompat & JFS_KNOWN_INCOMPAT_FEATURES) == incompat)
1330 * int journal_set_features () - Mark a given journal feature in the superblock
1331 * @journal: Journal to act on.
1332 * @compat: bitmask of compatible features
1333 * @ro: bitmask of features that force read-only mount
1334 * @incompat: bitmask of incompatible features
1336 * Mark a given journal feature as present on the
1337 * superblock. Returns true if the requested features could be set.
1341 int journal_set_features (journal_t *journal, unsigned long compat,
1342 unsigned long ro, unsigned long incompat)
1344 journal_superblock_t *sb;
1346 if (journal_check_used_features(journal, compat, ro, incompat))
1349 if (!journal_check_available_features(journal, compat, ro, incompat))
1352 jbd_debug(1, "Setting new features 0x%lx/0x%lx/0x%lx\n",
1353 compat, ro, incompat);
1355 sb = journal->j_superblock;
1357 sb->s_feature_compat |= cpu_to_be32(compat);
1358 sb->s_feature_ro_compat |= cpu_to_be32(ro);
1359 sb->s_feature_incompat |= cpu_to_be32(incompat);
1366 * int journal_update_format () - Update on-disk journal structure.
1367 * @journal: Journal to act on.
1369 * Given an initialised but unloaded journal struct, poke about in the
1370 * on-disk structure to update it to the most recent supported version.
1372 int journal_update_format (journal_t *journal)
1374 journal_superblock_t *sb;
1377 err = journal_get_superblock(journal);
1381 sb = journal->j_superblock;
1383 switch (be32_to_cpu(sb->s_header.h_blocktype)) {
1384 case JFS_SUPERBLOCK_V2:
1386 case JFS_SUPERBLOCK_V1:
1387 return journal_convert_superblock_v1(journal, sb);
1394 static int journal_convert_superblock_v1(journal_t *journal,
1395 journal_superblock_t *sb)
1397 int offset, blocksize;
1398 struct buffer_head *bh;
1401 "JBD: Converting superblock from version 1 to 2.\n");
1403 /* Pre-initialise new fields to zero */
1404 offset = ((char *) &(sb->s_feature_compat)) - ((char *) sb);
1405 blocksize = be32_to_cpu(sb->s_blocksize);
1406 memset(&sb->s_feature_compat, 0, blocksize-offset);
1408 sb->s_nr_users = cpu_to_be32(1);
1409 sb->s_header.h_blocktype = cpu_to_be32(JFS_SUPERBLOCK_V2);
1410 journal->j_format_version = 2;
1412 bh = journal->j_sb_buffer;
1413 BUFFER_TRACE(bh, "marking dirty");
1414 mark_buffer_dirty(bh);
1415 sync_dirty_buffer(bh);
1421 * int journal_flush () - Flush journal
1422 * @journal: Journal to act on.
1424 * Flush all data for a given journal to disk and empty the journal.
1425 * Filesystems can use this when remounting readonly to ensure that
1426 * recovery does not need to happen on remount.
1429 int journal_flush(journal_t *journal)
1432 transaction_t *transaction = NULL;
1433 unsigned int old_tail;
1435 spin_lock(&journal->j_state_lock);
1437 /* Force everything buffered to the log... */
1438 if (journal->j_running_transaction) {
1439 transaction = journal->j_running_transaction;
1440 __log_start_commit(journal, transaction->t_tid);
1441 } else if (journal->j_committing_transaction)
1442 transaction = journal->j_committing_transaction;
1444 /* Wait for the log commit to complete... */
1446 tid_t tid = transaction->t_tid;
1448 spin_unlock(&journal->j_state_lock);
1449 log_wait_commit(journal, tid);
1451 spin_unlock(&journal->j_state_lock);
1454 /* ...and flush everything in the log out to disk. */
1455 spin_lock(&journal->j_list_lock);
1456 while (!err && journal->j_checkpoint_transactions != NULL) {
1457 spin_unlock(&journal->j_list_lock);
1458 mutex_lock(&journal->j_checkpoint_mutex);
1459 err = log_do_checkpoint(journal);
1460 mutex_unlock(&journal->j_checkpoint_mutex);
1461 spin_lock(&journal->j_list_lock);
1463 spin_unlock(&journal->j_list_lock);
1465 if (is_journal_aborted(journal))
1468 cleanup_journal_tail(journal);
1470 /* Finally, mark the journal as really needing no recovery.
1471 * This sets s_start==0 in the underlying superblock, which is
1472 * the magic code for a fully-recovered superblock. Any future
1473 * commits of data to the journal will restore the current
1475 spin_lock(&journal->j_state_lock);
1476 old_tail = journal->j_tail;
1477 journal->j_tail = 0;
1478 spin_unlock(&journal->j_state_lock);
1479 journal_update_superblock(journal, 1);
1480 spin_lock(&journal->j_state_lock);
1481 journal->j_tail = old_tail;
1483 J_ASSERT(!journal->j_running_transaction);
1484 J_ASSERT(!journal->j_committing_transaction);
1485 J_ASSERT(!journal->j_checkpoint_transactions);
1486 J_ASSERT(journal->j_head == journal->j_tail);
1487 J_ASSERT(journal->j_tail_sequence == journal->j_transaction_sequence);
1488 spin_unlock(&journal->j_state_lock);
1493 * int journal_wipe() - Wipe journal contents
1494 * @journal: Journal to act on.
1495 * @write: flag (see below)
1497 * Wipe out all of the contents of a journal, safely. This will produce
1498 * a warning if the journal contains any valid recovery information.
1499 * Must be called between journal_init_*() and journal_load().
1501 * If 'write' is non-zero, then we wipe out the journal on disk; otherwise
1502 * we merely suppress recovery.
1505 int journal_wipe(journal_t *journal, int write)
1509 J_ASSERT (!(journal->j_flags & JFS_LOADED));
1511 err = load_superblock(journal);
1515 if (!journal->j_tail)
1518 printk (KERN_WARNING "JBD: %s recovery information on journal\n",
1519 write ? "Clearing" : "Ignoring");
1521 err = journal_skip_recovery(journal);
1523 journal_update_superblock(journal, 1);
1530 * journal_dev_name: format a character string to describe on what
1531 * device this journal is present.
1534 static const char *journal_dev_name(journal_t *journal, char *buffer)
1536 struct block_device *bdev;
1538 if (journal->j_inode)
1539 bdev = journal->j_inode->i_sb->s_bdev;
1541 bdev = journal->j_dev;
1543 return bdevname(bdev, buffer);
1547 * Journal abort has very specific semantics, which we describe
1548 * for journal abort.
1550 * Two internal function, which provide abort to te jbd layer
1555 * Quick version for internal journal use (doesn't lock the journal).
1556 * Aborts hard --- we mark the abort as occurred, but do _nothing_ else,
1557 * and don't attempt to make any other journal updates.
1559 static void __journal_abort_hard(journal_t *journal)
1561 transaction_t *transaction;
1562 char b[BDEVNAME_SIZE];
1564 if (journal->j_flags & JFS_ABORT)
1567 printk(KERN_ERR "Aborting journal on device %s.\n",
1568 journal_dev_name(journal, b));
1570 spin_lock(&journal->j_state_lock);
1571 journal->j_flags |= JFS_ABORT;
1572 transaction = journal->j_running_transaction;
1574 __log_start_commit(journal, transaction->t_tid);
1575 spin_unlock(&journal->j_state_lock);
1578 /* Soft abort: record the abort error status in the journal superblock,
1579 * but don't do any other IO. */
1580 static void __journal_abort_soft (journal_t *journal, int errno)
1582 if (journal->j_flags & JFS_ABORT)
1585 if (!journal->j_errno)
1586 journal->j_errno = errno;
1588 __journal_abort_hard(journal);
1591 journal_update_superblock(journal, 1);
1595 * void journal_abort () - Shutdown the journal immediately.
1596 * @journal: the journal to shutdown.
1597 * @errno: an error number to record in the journal indicating
1598 * the reason for the shutdown.
1600 * Perform a complete, immediate shutdown of the ENTIRE
1601 * journal (not of a single transaction). This operation cannot be
1602 * undone without closing and reopening the journal.
1604 * The journal_abort function is intended to support higher level error
1605 * recovery mechanisms such as the ext2/ext3 remount-readonly error
1608 * Journal abort has very specific semantics. Any existing dirty,
1609 * unjournaled buffers in the main filesystem will still be written to
1610 * disk by bdflush, but the journaling mechanism will be suspended
1611 * immediately and no further transaction commits will be honoured.
1613 * Any dirty, journaled buffers will be written back to disk without
1614 * hitting the journal. Atomicity cannot be guaranteed on an aborted
1615 * filesystem, but we _do_ attempt to leave as much data as possible
1616 * behind for fsck to use for cleanup.
1618 * Any attempt to get a new transaction handle on a journal which is in
1619 * ABORT state will just result in an -EROFS error return. A
1620 * journal_stop on an existing handle will return -EIO if we have
1621 * entered abort state during the update.
1623 * Recursive transactions are not disturbed by journal abort until the
1624 * final journal_stop, which will receive the -EIO error.
1626 * Finally, the journal_abort call allows the caller to supply an errno
1627 * which will be recorded (if possible) in the journal superblock. This
1628 * allows a client to record failure conditions in the middle of a
1629 * transaction without having to complete the transaction to record the
1630 * failure to disk. ext3_error, for example, now uses this
1633 * Errors which originate from within the journaling layer will NOT
1634 * supply an errno; a null errno implies that absolutely no further
1635 * writes are done to the journal (unless there are any already in
1640 void journal_abort(journal_t *journal, int errno)
1642 __journal_abort_soft(journal, errno);
1646 * int journal_errno () - returns the journal's error state.
1647 * @journal: journal to examine.
1649 * This is the errno numbet set with journal_abort(), the last
1650 * time the journal was mounted - if the journal was stopped
1651 * without calling abort this will be 0.
1653 * If the journal has been aborted on this mount time -EROFS will
1656 int journal_errno(journal_t *journal)
1660 spin_lock(&journal->j_state_lock);
1661 if (journal->j_flags & JFS_ABORT)
1664 err = journal->j_errno;
1665 spin_unlock(&journal->j_state_lock);
1670 * int journal_clear_err () - clears the journal's error state
1671 * @journal: journal to act on.
1673 * An error must be cleared or Acked to take a FS out of readonly
1676 int journal_clear_err(journal_t *journal)
1680 spin_lock(&journal->j_state_lock);
1681 if (journal->j_flags & JFS_ABORT)
1684 journal->j_errno = 0;
1685 spin_unlock(&journal->j_state_lock);
1690 * void journal_ack_err() - Ack journal err.
1691 * @journal: journal to act on.
1693 * An error must be cleared or Acked to take a FS out of readonly
1696 void journal_ack_err(journal_t *journal)
1698 spin_lock(&journal->j_state_lock);
1699 if (journal->j_errno)
1700 journal->j_flags |= JFS_ACK_ERR;
1701 spin_unlock(&journal->j_state_lock);
1704 int journal_blocks_per_page(struct inode *inode)
1706 return 1 << (PAGE_CACHE_SHIFT - inode->i_sb->s_blocksize_bits);
1710 * Journal_head storage management
1712 static struct kmem_cache *journal_head_cache;
1713 #ifdef CONFIG_JBD_DEBUG
1714 static atomic_t nr_journal_heads = ATOMIC_INIT(0);
1717 static int journal_init_journal_head_cache(void)
1721 J_ASSERT(journal_head_cache == NULL);
1722 journal_head_cache = kmem_cache_create("journal_head",
1723 sizeof(struct journal_head),
1725 SLAB_TEMPORARY, /* flags */
1728 if (!journal_head_cache) {
1730 printk(KERN_EMERG "JBD: no memory for journal_head cache\n");
1735 static void journal_destroy_journal_head_cache(void)
1737 if (journal_head_cache) {
1738 kmem_cache_destroy(journal_head_cache);
1739 journal_head_cache = NULL;
1744 * journal_head splicing and dicing
1746 static struct journal_head *journal_alloc_journal_head(void)
1748 struct journal_head *ret;
1750 #ifdef CONFIG_JBD_DEBUG
1751 atomic_inc(&nr_journal_heads);
1753 ret = kmem_cache_alloc(journal_head_cache, GFP_NOFS);
1755 jbd_debug(1, "out of memory for journal_head\n");
1756 printk_ratelimited(KERN_NOTICE "ENOMEM in %s, retrying.\n",
1759 while (ret == NULL) {
1761 ret = kmem_cache_alloc(journal_head_cache, GFP_NOFS);
1767 static void journal_free_journal_head(struct journal_head *jh)
1769 #ifdef CONFIG_JBD_DEBUG
1770 atomic_dec(&nr_journal_heads);
1771 memset(jh, JBD_POISON_FREE, sizeof(*jh));
1773 kmem_cache_free(journal_head_cache, jh);
1777 * A journal_head is attached to a buffer_head whenever JBD has an
1778 * interest in the buffer.
1780 * Whenever a buffer has an attached journal_head, its ->b_state:BH_JBD bit
1781 * is set. This bit is tested in core kernel code where we need to take
1782 * JBD-specific actions. Testing the zeroness of ->b_private is not reliable
1785 * When a buffer has its BH_JBD bit set, its ->b_count is elevated by one.
1787 * When a buffer has its BH_JBD bit set it is immune from being released by
1788 * core kernel code, mainly via ->b_count.
1790 * A journal_head may be detached from its buffer_head when the journal_head's
1791 * b_transaction, b_cp_transaction and b_next_transaction pointers are NULL.
1792 * Various places in JBD call journal_remove_journal_head() to indicate that the
1793 * journal_head can be dropped if needed.
1795 * Various places in the kernel want to attach a journal_head to a buffer_head
1796 * _before_ attaching the journal_head to a transaction. To protect the
1797 * journal_head in this situation, journal_add_journal_head elevates the
1798 * journal_head's b_jcount refcount by one. The caller must call
1799 * journal_put_journal_head() to undo this.
1801 * So the typical usage would be:
1803 * (Attach a journal_head if needed. Increments b_jcount)
1804 * struct journal_head *jh = journal_add_journal_head(bh);
1806 * jh->b_transaction = xxx;
1807 * journal_put_journal_head(jh);
1809 * Now, the journal_head's b_jcount is zero, but it is safe from being released
1810 * because it has a non-zero b_transaction.
1814 * Give a buffer_head a journal_head.
1816 * Doesn't need the journal lock.
1819 struct journal_head *journal_add_journal_head(struct buffer_head *bh)
1821 struct journal_head *jh;
1822 struct journal_head *new_jh = NULL;
1825 if (!buffer_jbd(bh)) {
1826 new_jh = journal_alloc_journal_head();
1827 memset(new_jh, 0, sizeof(*new_jh));
1830 jbd_lock_bh_journal_head(bh);
1831 if (buffer_jbd(bh)) {
1835 (atomic_read(&bh->b_count) > 0) ||
1836 (bh->b_page && bh->b_page->mapping));
1839 jbd_unlock_bh_journal_head(bh);
1844 new_jh = NULL; /* We consumed it */
1849 BUFFER_TRACE(bh, "added journal_head");
1852 jbd_unlock_bh_journal_head(bh);
1854 journal_free_journal_head(new_jh);
1855 return bh->b_private;
1859 * Grab a ref against this buffer_head's journal_head. If it ended up not
1860 * having a journal_head, return NULL
1862 struct journal_head *journal_grab_journal_head(struct buffer_head *bh)
1864 struct journal_head *jh = NULL;
1866 jbd_lock_bh_journal_head(bh);
1867 if (buffer_jbd(bh)) {
1871 jbd_unlock_bh_journal_head(bh);
1875 static void __journal_remove_journal_head(struct buffer_head *bh)
1877 struct journal_head *jh = bh2jh(bh);
1879 J_ASSERT_JH(jh, jh->b_jcount >= 0);
1882 if (jh->b_jcount == 0) {
1883 if (jh->b_transaction == NULL &&
1884 jh->b_next_transaction == NULL &&
1885 jh->b_cp_transaction == NULL) {
1886 J_ASSERT_JH(jh, jh->b_jlist == BJ_None);
1887 J_ASSERT_BH(bh, buffer_jbd(bh));
1888 J_ASSERT_BH(bh, jh2bh(jh) == bh);
1889 BUFFER_TRACE(bh, "remove journal_head");
1890 if (jh->b_frozen_data) {
1891 printk(KERN_WARNING "%s: freeing "
1894 jbd_free(jh->b_frozen_data, bh->b_size);
1896 if (jh->b_committed_data) {
1897 printk(KERN_WARNING "%s: freeing "
1898 "b_committed_data\n",
1900 jbd_free(jh->b_committed_data, bh->b_size);
1902 bh->b_private = NULL;
1903 jh->b_bh = NULL; /* debug, really */
1904 clear_buffer_jbd(bh);
1906 journal_free_journal_head(jh);
1908 BUFFER_TRACE(bh, "journal_head was locked");
1914 * journal_remove_journal_head(): if the buffer isn't attached to a transaction
1915 * and has a zero b_jcount then remove and release its journal_head. If we did
1916 * see that the buffer is not used by any transaction we also "logically"
1917 * decrement ->b_count.
1919 * We in fact take an additional increment on ->b_count as a convenience,
1920 * because the caller usually wants to do additional things with the bh
1921 * after calling here.
1922 * The caller of journal_remove_journal_head() *must* run __brelse(bh) at some
1923 * time. Once the caller has run __brelse(), the buffer is eligible for
1924 * reaping by try_to_free_buffers().
1926 void journal_remove_journal_head(struct buffer_head *bh)
1928 jbd_lock_bh_journal_head(bh);
1929 __journal_remove_journal_head(bh);
1930 jbd_unlock_bh_journal_head(bh);
1934 * Drop a reference on the passed journal_head. If it fell to zero then try to
1935 * release the journal_head from the buffer_head.
1937 void journal_put_journal_head(struct journal_head *jh)
1939 struct buffer_head *bh = jh2bh(jh);
1941 jbd_lock_bh_journal_head(bh);
1942 J_ASSERT_JH(jh, jh->b_jcount > 0);
1944 if (!jh->b_jcount && !jh->b_transaction) {
1945 __journal_remove_journal_head(bh);
1948 jbd_unlock_bh_journal_head(bh);
1954 #ifdef CONFIG_JBD_DEBUG
1956 u8 journal_enable_debug __read_mostly;
1957 EXPORT_SYMBOL(journal_enable_debug);
1959 static struct dentry *jbd_debugfs_dir;
1960 static struct dentry *jbd_debug;
1962 static void __init jbd_create_debugfs_entry(void)
1964 jbd_debugfs_dir = debugfs_create_dir("jbd", NULL);
1965 if (jbd_debugfs_dir)
1966 jbd_debug = debugfs_create_u8("jbd-debug", S_IRUGO | S_IWUSR,
1968 &journal_enable_debug);
1971 static void __exit jbd_remove_debugfs_entry(void)
1973 debugfs_remove(jbd_debug);
1974 debugfs_remove(jbd_debugfs_dir);
1979 static inline void jbd_create_debugfs_entry(void)
1983 static inline void jbd_remove_debugfs_entry(void)
1989 struct kmem_cache *jbd_handle_cache;
1991 static int __init journal_init_handle_cache(void)
1993 jbd_handle_cache = kmem_cache_create("journal_handle",
1996 SLAB_TEMPORARY, /* flags */
1998 if (jbd_handle_cache == NULL) {
1999 printk(KERN_EMERG "JBD: failed to create handle cache\n");
2005 static void journal_destroy_handle_cache(void)
2007 if (jbd_handle_cache)
2008 kmem_cache_destroy(jbd_handle_cache);
2012 * Module startup and shutdown
2015 static int __init journal_init_caches(void)
2019 ret = journal_init_revoke_caches();
2021 ret = journal_init_journal_head_cache();
2023 ret = journal_init_handle_cache();
2027 static void journal_destroy_caches(void)
2029 journal_destroy_revoke_caches();
2030 journal_destroy_journal_head_cache();
2031 journal_destroy_handle_cache();
2034 static int __init journal_init(void)
2038 BUILD_BUG_ON(sizeof(struct journal_superblock_s) != 1024);
2040 ret = journal_init_caches();
2042 journal_destroy_caches();
2043 jbd_create_debugfs_entry();
2047 static void __exit journal_exit(void)
2049 #ifdef CONFIG_JBD_DEBUG
2050 int n = atomic_read(&nr_journal_heads);
2052 printk(KERN_EMERG "JBD: leaked %d journal_heads!\n", n);
2054 jbd_remove_debugfs_entry();
2055 journal_destroy_caches();
2058 MODULE_LICENSE("GPL");
2059 module_init(journal_init);
2060 module_exit(journal_exit);