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 #define CREATE_TRACE_POINTS
42 #include <trace/events/jbd.h>
44 #include <asm/uaccess.h>
47 EXPORT_SYMBOL(journal_start);
48 EXPORT_SYMBOL(journal_restart);
49 EXPORT_SYMBOL(journal_extend);
50 EXPORT_SYMBOL(journal_stop);
51 EXPORT_SYMBOL(journal_lock_updates);
52 EXPORT_SYMBOL(journal_unlock_updates);
53 EXPORT_SYMBOL(journal_get_write_access);
54 EXPORT_SYMBOL(journal_get_create_access);
55 EXPORT_SYMBOL(journal_get_undo_access);
56 EXPORT_SYMBOL(journal_dirty_data);
57 EXPORT_SYMBOL(journal_dirty_metadata);
58 EXPORT_SYMBOL(journal_release_buffer);
59 EXPORT_SYMBOL(journal_forget);
61 EXPORT_SYMBOL(journal_sync_buffer);
63 EXPORT_SYMBOL(journal_flush);
64 EXPORT_SYMBOL(journal_revoke);
66 EXPORT_SYMBOL(journal_init_dev);
67 EXPORT_SYMBOL(journal_init_inode);
68 EXPORT_SYMBOL(journal_update_format);
69 EXPORT_SYMBOL(journal_check_used_features);
70 EXPORT_SYMBOL(journal_check_available_features);
71 EXPORT_SYMBOL(journal_set_features);
72 EXPORT_SYMBOL(journal_create);
73 EXPORT_SYMBOL(journal_load);
74 EXPORT_SYMBOL(journal_destroy);
75 EXPORT_SYMBOL(journal_abort);
76 EXPORT_SYMBOL(journal_errno);
77 EXPORT_SYMBOL(journal_ack_err);
78 EXPORT_SYMBOL(journal_clear_err);
79 EXPORT_SYMBOL(log_wait_commit);
80 EXPORT_SYMBOL(log_start_commit);
81 EXPORT_SYMBOL(journal_start_commit);
82 EXPORT_SYMBOL(journal_force_commit_nested);
83 EXPORT_SYMBOL(journal_wipe);
84 EXPORT_SYMBOL(journal_blocks_per_page);
85 EXPORT_SYMBOL(journal_invalidatepage);
86 EXPORT_SYMBOL(journal_try_to_free_buffers);
87 EXPORT_SYMBOL(journal_force_commit);
89 static int journal_convert_superblock_v1(journal_t *, journal_superblock_t *);
90 static void __journal_abort_soft (journal_t *journal, int errno);
91 static const char *journal_dev_name(journal_t *journal, char *buffer);
94 * Helper function used to manage commit timeouts
97 static void commit_timeout(unsigned long __data)
99 struct task_struct * p = (struct task_struct *) __data;
105 * kjournald: The main thread function used to manage a logging device
108 * This kernel thread is responsible for two things:
110 * 1) COMMIT: Every so often we need to commit the current state of the
111 * filesystem to disk. The journal thread is responsible for writing
112 * all of the metadata buffers to disk.
114 * 2) CHECKPOINT: We cannot reuse a used section of the log file until all
115 * of the data in that part of the log has been rewritten elsewhere on
116 * the disk. Flushing these old buffers to reclaim space in the log is
117 * known as checkpointing, and this thread is responsible for that job.
120 static int kjournald(void *arg)
122 journal_t *journal = arg;
123 transaction_t *transaction;
126 * Set up an interval timer which can be used to trigger a commit wakeup
127 * after the commit interval expires
129 setup_timer(&journal->j_commit_timer, commit_timeout,
130 (unsigned long)current);
134 /* Record that the journal thread is running */
135 journal->j_task = current;
136 wake_up(&journal->j_wait_done_commit);
138 printk(KERN_INFO "kjournald starting. Commit interval %ld seconds\n",
139 journal->j_commit_interval / HZ);
142 * And now, wait forever for commit wakeup events.
144 spin_lock(&journal->j_state_lock);
147 if (journal->j_flags & JFS_UNMOUNT)
150 jbd_debug(1, "commit_sequence=%d, commit_request=%d\n",
151 journal->j_commit_sequence, journal->j_commit_request);
153 if (journal->j_commit_sequence != journal->j_commit_request) {
154 jbd_debug(1, "OK, requests differ\n");
155 spin_unlock(&journal->j_state_lock);
156 del_timer_sync(&journal->j_commit_timer);
157 journal_commit_transaction(journal);
158 spin_lock(&journal->j_state_lock);
162 wake_up(&journal->j_wait_done_commit);
163 if (freezing(current)) {
165 * The simpler the better. Flushing journal isn't a
166 * good idea, because that depends on threads that may
167 * be already stopped.
169 jbd_debug(1, "Now suspending kjournald\n");
170 spin_unlock(&journal->j_state_lock);
172 spin_lock(&journal->j_state_lock);
175 * We assume on resume that commits are already there,
179 int should_sleep = 1;
181 prepare_to_wait(&journal->j_wait_commit, &wait,
183 if (journal->j_commit_sequence != journal->j_commit_request)
185 transaction = journal->j_running_transaction;
186 if (transaction && time_after_eq(jiffies,
187 transaction->t_expires))
189 if (journal->j_flags & JFS_UNMOUNT)
192 spin_unlock(&journal->j_state_lock);
194 spin_lock(&journal->j_state_lock);
196 finish_wait(&journal->j_wait_commit, &wait);
199 jbd_debug(1, "kjournald wakes\n");
202 * Were we woken up by a commit wakeup event?
204 transaction = journal->j_running_transaction;
205 if (transaction && time_after_eq(jiffies, transaction->t_expires)) {
206 journal->j_commit_request = transaction->t_tid;
207 jbd_debug(1, "woke because of timeout\n");
212 spin_unlock(&journal->j_state_lock);
213 del_timer_sync(&journal->j_commit_timer);
214 journal->j_task = NULL;
215 wake_up(&journal->j_wait_done_commit);
216 jbd_debug(1, "Journal thread exiting.\n");
220 static int journal_start_thread(journal_t *journal)
222 struct task_struct *t;
224 t = kthread_run(kjournald, journal, "kjournald");
228 wait_event(journal->j_wait_done_commit, journal->j_task != NULL);
232 static void journal_kill_thread(journal_t *journal)
234 spin_lock(&journal->j_state_lock);
235 journal->j_flags |= JFS_UNMOUNT;
237 while (journal->j_task) {
238 wake_up(&journal->j_wait_commit);
239 spin_unlock(&journal->j_state_lock);
240 wait_event(journal->j_wait_done_commit,
241 journal->j_task == NULL);
242 spin_lock(&journal->j_state_lock);
244 spin_unlock(&journal->j_state_lock);
248 * journal_write_metadata_buffer: write a metadata buffer to the journal.
250 * Writes a metadata buffer to a given disk block. The actual IO is not
251 * performed but a new buffer_head is constructed which labels the data
252 * to be written with the correct destination disk block.
254 * Any magic-number escaping which needs to be done will cause a
255 * copy-out here. If the buffer happens to start with the
256 * JFS_MAGIC_NUMBER, then we can't write it to the log directly: the
257 * magic number is only written to the log for descripter blocks. In
258 * this case, we copy the data and replace the first word with 0, and we
259 * return a result code which indicates that this buffer needs to be
260 * marked as an escaped buffer in the corresponding log descriptor
261 * block. The missing word can then be restored when the block is read
264 * If the source buffer has already been modified by a new transaction
265 * since we took the last commit snapshot, we use the frozen copy of
266 * that data for IO. If we end up using the existing buffer_head's data
267 * for the write, then we *have* to lock the buffer to prevent anyone
268 * else from using and possibly modifying it while the IO is in
271 * The function returns a pointer to the buffer_heads to be used for IO.
273 * We assume that the journal has already been locked in this function.
280 * Bit 0 set == escape performed on the data
281 * Bit 1 set == buffer copy-out performed (kfree the data after IO)
284 int journal_write_metadata_buffer(transaction_t *transaction,
285 struct journal_head *jh_in,
286 struct journal_head **jh_out,
287 unsigned int blocknr)
289 int need_copy_out = 0;
290 int done_copy_out = 0;
293 struct buffer_head *new_bh;
294 struct journal_head *new_jh;
295 struct page *new_page;
296 unsigned int new_offset;
297 struct buffer_head *bh_in = jh2bh(jh_in);
298 journal_t *journal = transaction->t_journal;
301 * The buffer really shouldn't be locked: only the current committing
302 * transaction is allowed to write it, so nobody else is allowed
305 * akpm: except if we're journalling data, and write() output is
306 * also part of a shared mapping, and another thread has
307 * decided to launch a writepage() against this buffer.
309 J_ASSERT_BH(bh_in, buffer_jbddirty(bh_in));
311 new_bh = alloc_buffer_head(GFP_NOFS|__GFP_NOFAIL);
312 /* keep subsequent assertions sane */
313 atomic_set(&new_bh->b_count, 1);
314 new_jh = journal_add_journal_head(new_bh); /* This sleeps */
317 * If a new transaction has already done a buffer copy-out, then
318 * we use that version of the data for the commit.
320 jbd_lock_bh_state(bh_in);
322 if (jh_in->b_frozen_data) {
324 new_page = virt_to_page(jh_in->b_frozen_data);
325 new_offset = offset_in_page(jh_in->b_frozen_data);
327 new_page = jh2bh(jh_in)->b_page;
328 new_offset = offset_in_page(jh2bh(jh_in)->b_data);
331 mapped_data = kmap_atomic(new_page);
335 if (*((__be32 *)(mapped_data + new_offset)) ==
336 cpu_to_be32(JFS_MAGIC_NUMBER)) {
340 kunmap_atomic(mapped_data);
343 * Do we need to do a data copy?
345 if (need_copy_out && !done_copy_out) {
348 jbd_unlock_bh_state(bh_in);
349 tmp = jbd_alloc(bh_in->b_size, GFP_NOFS);
350 jbd_lock_bh_state(bh_in);
351 if (jh_in->b_frozen_data) {
352 jbd_free(tmp, bh_in->b_size);
356 jh_in->b_frozen_data = tmp;
357 mapped_data = kmap_atomic(new_page);
358 memcpy(tmp, mapped_data + new_offset, jh2bh(jh_in)->b_size);
359 kunmap_atomic(mapped_data);
361 new_page = virt_to_page(tmp);
362 new_offset = offset_in_page(tmp);
367 * Did we need to do an escaping? Now we've done all the
368 * copying, we can finally do so.
371 mapped_data = kmap_atomic(new_page);
372 *((unsigned int *)(mapped_data + new_offset)) = 0;
373 kunmap_atomic(mapped_data);
376 set_bh_page(new_bh, new_page, new_offset);
377 new_jh->b_transaction = NULL;
378 new_bh->b_size = jh2bh(jh_in)->b_size;
379 new_bh->b_bdev = transaction->t_journal->j_dev;
380 new_bh->b_blocknr = blocknr;
381 set_buffer_mapped(new_bh);
382 set_buffer_dirty(new_bh);
387 * The to-be-written buffer needs to get moved to the io queue,
388 * and the original buffer whose contents we are shadowing or
389 * copying is moved to the transaction's shadow queue.
391 JBUFFER_TRACE(jh_in, "file as BJ_Shadow");
392 spin_lock(&journal->j_list_lock);
393 __journal_file_buffer(jh_in, transaction, BJ_Shadow);
394 spin_unlock(&journal->j_list_lock);
395 jbd_unlock_bh_state(bh_in);
397 JBUFFER_TRACE(new_jh, "file as BJ_IO");
398 journal_file_buffer(new_jh, transaction, BJ_IO);
400 return do_escape | (done_copy_out << 1);
404 * Allocation code for the journal file. Manage the space left in the
405 * journal, so that we can begin checkpointing when appropriate.
409 * __log_space_left: Return the number of free blocks left in the journal.
411 * Called with the journal already locked.
413 * Called under j_state_lock
416 int __log_space_left(journal_t *journal)
418 int left = journal->j_free;
420 assert_spin_locked(&journal->j_state_lock);
423 * Be pessimistic here about the number of those free blocks which
424 * might be required for log descriptor control blocks.
427 #define MIN_LOG_RESERVED_BLOCKS 32 /* Allow for rounding errors */
429 left -= MIN_LOG_RESERVED_BLOCKS;
438 * Called under j_state_lock. Returns true if a transaction commit was started.
440 int __log_start_commit(journal_t *journal, tid_t target)
443 * The only transaction we can possibly wait upon is the
444 * currently running transaction (if it exists). Otherwise,
445 * the target tid must be an old one.
447 if (journal->j_commit_request != target &&
448 journal->j_running_transaction &&
449 journal->j_running_transaction->t_tid == target) {
451 * We want a new commit: OK, mark the request and wakeup the
452 * commit thread. We do _not_ do the commit ourselves.
455 journal->j_commit_request = target;
456 jbd_debug(1, "JBD: requesting commit %d/%d\n",
457 journal->j_commit_request,
458 journal->j_commit_sequence);
459 wake_up(&journal->j_wait_commit);
461 } else if (!tid_geq(journal->j_commit_request, target))
462 /* This should never happen, but if it does, preserve
463 the evidence before kjournald goes into a loop and
464 increments j_commit_sequence beyond all recognition. */
465 WARN_ONCE(1, "jbd: bad log_start_commit: %u %u %u %u\n",
466 journal->j_commit_request, journal->j_commit_sequence,
467 target, journal->j_running_transaction ?
468 journal->j_running_transaction->t_tid : 0);
472 int log_start_commit(journal_t *journal, tid_t tid)
476 spin_lock(&journal->j_state_lock);
477 ret = __log_start_commit(journal, tid);
478 spin_unlock(&journal->j_state_lock);
483 * Force and wait upon a commit if the calling process is not within
484 * transaction. This is used for forcing out undo-protected data which contains
485 * bitmaps, when the fs is running out of space.
487 * We can only force the running transaction if we don't have an active handle;
488 * otherwise, we will deadlock.
490 * Returns true if a transaction was started.
492 int journal_force_commit_nested(journal_t *journal)
494 transaction_t *transaction = NULL;
497 spin_lock(&journal->j_state_lock);
498 if (journal->j_running_transaction && !current->journal_info) {
499 transaction = journal->j_running_transaction;
500 __log_start_commit(journal, transaction->t_tid);
501 } else if (journal->j_committing_transaction)
502 transaction = journal->j_committing_transaction;
505 spin_unlock(&journal->j_state_lock);
506 return 0; /* Nothing to retry */
509 tid = transaction->t_tid;
510 spin_unlock(&journal->j_state_lock);
511 log_wait_commit(journal, tid);
516 * Start a commit of the current running transaction (if any). Returns true
517 * if a transaction is going to be committed (or is currently already
518 * committing), and fills its tid in at *ptid
520 int journal_start_commit(journal_t *journal, tid_t *ptid)
524 spin_lock(&journal->j_state_lock);
525 if (journal->j_running_transaction) {
526 tid_t tid = journal->j_running_transaction->t_tid;
528 __log_start_commit(journal, tid);
529 /* There's a running transaction and we've just made sure
530 * it's commit has been scheduled. */
534 } else if (journal->j_committing_transaction) {
536 * If commit has been started, then we have to wait for
537 * completion of that transaction.
540 *ptid = journal->j_committing_transaction->t_tid;
543 spin_unlock(&journal->j_state_lock);
548 * Wait for a specified commit to complete.
549 * The caller may not hold the journal lock.
551 int log_wait_commit(journal_t *journal, tid_t tid)
555 #ifdef CONFIG_JBD_DEBUG
556 spin_lock(&journal->j_state_lock);
557 if (!tid_geq(journal->j_commit_request, tid)) {
559 "%s: error: j_commit_request=%d, tid=%d\n",
560 __func__, journal->j_commit_request, tid);
562 spin_unlock(&journal->j_state_lock);
564 spin_lock(&journal->j_state_lock);
565 if (!tid_geq(journal->j_commit_waited, tid))
566 journal->j_commit_waited = tid;
567 while (tid_gt(tid, journal->j_commit_sequence)) {
568 jbd_debug(1, "JBD: want %d, j_commit_sequence=%d\n",
569 tid, journal->j_commit_sequence);
570 wake_up(&journal->j_wait_commit);
571 spin_unlock(&journal->j_state_lock);
572 wait_event(journal->j_wait_done_commit,
573 !tid_gt(tid, journal->j_commit_sequence));
574 spin_lock(&journal->j_state_lock);
576 spin_unlock(&journal->j_state_lock);
578 if (unlikely(is_journal_aborted(journal))) {
579 printk(KERN_EMERG "journal commit I/O error\n");
586 * Return 1 if a given transaction has not yet sent barrier request
587 * connected with a transaction commit. If 0 is returned, transaction
588 * may or may not have sent the barrier. Used to avoid sending barrier
589 * twice in common cases.
591 int journal_trans_will_send_data_barrier(journal_t *journal, tid_t tid)
594 transaction_t *commit_trans;
596 if (!(journal->j_flags & JFS_BARRIER))
598 spin_lock(&journal->j_state_lock);
599 /* Transaction already committed? */
600 if (tid_geq(journal->j_commit_sequence, tid))
603 * Transaction is being committed and we already proceeded to
604 * writing commit record?
606 commit_trans = journal->j_committing_transaction;
607 if (commit_trans && commit_trans->t_tid == tid &&
608 commit_trans->t_state >= T_COMMIT_RECORD)
612 spin_unlock(&journal->j_state_lock);
615 EXPORT_SYMBOL(journal_trans_will_send_data_barrier);
618 * Log buffer allocation routines:
621 int journal_next_log_block(journal_t *journal, unsigned int *retp)
623 unsigned int blocknr;
625 spin_lock(&journal->j_state_lock);
626 J_ASSERT(journal->j_free > 1);
628 blocknr = journal->j_head;
631 if (journal->j_head == journal->j_last)
632 journal->j_head = journal->j_first;
633 spin_unlock(&journal->j_state_lock);
634 return journal_bmap(journal, blocknr, retp);
638 * Conversion of logical to physical block numbers for the journal
640 * On external journals the journal blocks are identity-mapped, so
641 * this is a no-op. If needed, we can use j_blk_offset - everything is
644 int journal_bmap(journal_t *journal, unsigned int blocknr,
650 if (journal->j_inode) {
651 ret = bmap(journal->j_inode, blocknr);
655 char b[BDEVNAME_SIZE];
657 printk(KERN_ALERT "%s: journal block not found "
658 "at offset %u on %s\n",
661 bdevname(journal->j_dev, b));
663 __journal_abort_soft(journal, err);
666 *retp = blocknr; /* +journal->j_blk_offset */
672 * We play buffer_head aliasing tricks to write data/metadata blocks to
673 * the journal without copying their contents, but for journal
674 * descriptor blocks we do need to generate bona fide buffers.
676 * After the caller of journal_get_descriptor_buffer() has finished modifying
677 * the buffer's contents they really should run flush_dcache_page(bh->b_page).
678 * But we don't bother doing that, so there will be coherency problems with
679 * mmaps of blockdevs which hold live JBD-controlled filesystems.
681 struct journal_head *journal_get_descriptor_buffer(journal_t *journal)
683 struct buffer_head *bh;
684 unsigned int blocknr;
687 err = journal_next_log_block(journal, &blocknr);
692 bh = __getblk(journal->j_dev, blocknr, journal->j_blocksize);
696 memset(bh->b_data, 0, journal->j_blocksize);
697 set_buffer_uptodate(bh);
699 BUFFER_TRACE(bh, "return this buffer");
700 return journal_add_journal_head(bh);
704 * Management for journal control blocks: functions to create and
705 * destroy journal_t structures, and to initialise and read existing
706 * journal blocks from disk. */
708 /* First: create and setup a journal_t object in memory. We initialise
709 * very few fields yet: that has to wait until we have created the
710 * journal structures from from scratch, or loaded them from disk. */
712 static journal_t * journal_init_common (void)
717 journal = kzalloc(sizeof(*journal), GFP_KERNEL);
721 init_waitqueue_head(&journal->j_wait_transaction_locked);
722 init_waitqueue_head(&journal->j_wait_logspace);
723 init_waitqueue_head(&journal->j_wait_done_commit);
724 init_waitqueue_head(&journal->j_wait_checkpoint);
725 init_waitqueue_head(&journal->j_wait_commit);
726 init_waitqueue_head(&journal->j_wait_updates);
727 mutex_init(&journal->j_checkpoint_mutex);
728 spin_lock_init(&journal->j_revoke_lock);
729 spin_lock_init(&journal->j_list_lock);
730 spin_lock_init(&journal->j_state_lock);
732 journal->j_commit_interval = (HZ * JBD_DEFAULT_MAX_COMMIT_AGE);
734 /* The journal is marked for error until we succeed with recovery! */
735 journal->j_flags = JFS_ABORT;
737 /* Set up a default-sized revoke table for the new mount. */
738 err = journal_init_revoke(journal, JOURNAL_REVOKE_DEFAULT_HASH);
748 /* journal_init_dev and journal_init_inode:
750 * Create a journal structure assigned some fixed set of disk blocks to
751 * the journal. We don't actually touch those disk blocks yet, but we
752 * need to set up all of the mapping information to tell the journaling
753 * system where the journal blocks are.
758 * journal_t * journal_init_dev() - creates and initialises a journal structure
759 * @bdev: Block device on which to create the journal
760 * @fs_dev: Device which hold journalled filesystem for this journal.
761 * @start: Block nr Start of journal.
762 * @len: Length of the journal in blocks.
763 * @blocksize: blocksize of journalling device
765 * Returns: a newly created journal_t *
767 * journal_init_dev creates a journal which maps a fixed contiguous
768 * range of blocks on an arbitrary block device.
771 journal_t * journal_init_dev(struct block_device *bdev,
772 struct block_device *fs_dev,
773 int start, int len, int blocksize)
775 journal_t *journal = journal_init_common();
776 struct buffer_head *bh;
782 /* journal descriptor can store up to n blocks -bzzz */
783 journal->j_blocksize = blocksize;
784 n = journal->j_blocksize / sizeof(journal_block_tag_t);
785 journal->j_wbufsize = n;
786 journal->j_wbuf = kmalloc(n * sizeof(struct buffer_head*), GFP_KERNEL);
787 if (!journal->j_wbuf) {
788 printk(KERN_ERR "%s: Can't allocate bhs for commit thread\n",
792 journal->j_dev = bdev;
793 journal->j_fs_dev = fs_dev;
794 journal->j_blk_offset = start;
795 journal->j_maxlen = len;
797 bh = __getblk(journal->j_dev, start, journal->j_blocksize);
800 "%s: Cannot get buffer for journal superblock\n",
804 journal->j_sb_buffer = bh;
805 journal->j_superblock = (journal_superblock_t *)bh->b_data;
809 kfree(journal->j_wbuf);
815 * journal_t * journal_init_inode () - creates a journal which maps to a inode.
816 * @inode: An inode to create the journal in
818 * journal_init_inode creates a journal which maps an on-disk inode as
819 * the journal. The inode must exist already, must support bmap() and
820 * must have all data blocks preallocated.
822 journal_t * journal_init_inode (struct inode *inode)
824 struct buffer_head *bh;
825 journal_t *journal = journal_init_common();
828 unsigned int blocknr;
833 journal->j_dev = journal->j_fs_dev = inode->i_sb->s_bdev;
834 journal->j_inode = inode;
836 "journal %p: inode %s/%ld, size %Ld, bits %d, blksize %ld\n",
837 journal, inode->i_sb->s_id, inode->i_ino,
838 (long long) inode->i_size,
839 inode->i_sb->s_blocksize_bits, inode->i_sb->s_blocksize);
841 journal->j_maxlen = inode->i_size >> inode->i_sb->s_blocksize_bits;
842 journal->j_blocksize = inode->i_sb->s_blocksize;
844 /* journal descriptor can store up to n blocks -bzzz */
845 n = journal->j_blocksize / sizeof(journal_block_tag_t);
846 journal->j_wbufsize = n;
847 journal->j_wbuf = kmalloc(n * sizeof(struct buffer_head*), GFP_KERNEL);
848 if (!journal->j_wbuf) {
849 printk(KERN_ERR "%s: Can't allocate bhs for commit thread\n",
854 err = journal_bmap(journal, 0, &blocknr);
855 /* If that failed, give up */
857 printk(KERN_ERR "%s: Cannot locate journal superblock\n",
862 bh = __getblk(journal->j_dev, blocknr, journal->j_blocksize);
865 "%s: Cannot get buffer for journal superblock\n",
869 journal->j_sb_buffer = bh;
870 journal->j_superblock = (journal_superblock_t *)bh->b_data;
874 kfree(journal->j_wbuf);
880 * If the journal init or create aborts, we need to mark the journal
881 * superblock as being NULL to prevent the journal destroy from writing
882 * back a bogus superblock.
884 static void journal_fail_superblock (journal_t *journal)
886 struct buffer_head *bh = journal->j_sb_buffer;
888 journal->j_sb_buffer = NULL;
892 * Given a journal_t structure, initialise the various fields for
893 * startup of a new journaling session. We use this both when creating
894 * a journal, and after recovering an old journal to reset it for
898 static int journal_reset(journal_t *journal)
900 journal_superblock_t *sb = journal->j_superblock;
901 unsigned int first, last;
903 first = be32_to_cpu(sb->s_first);
904 last = be32_to_cpu(sb->s_maxlen);
905 if (first + JFS_MIN_JOURNAL_BLOCKS > last + 1) {
906 printk(KERN_ERR "JBD: Journal too short (blocks %u-%u).\n",
908 journal_fail_superblock(journal);
912 journal->j_first = first;
913 journal->j_last = last;
915 journal->j_head = first;
916 journal->j_tail = first;
917 journal->j_free = last - first;
919 journal->j_tail_sequence = journal->j_transaction_sequence;
920 journal->j_commit_sequence = journal->j_transaction_sequence - 1;
921 journal->j_commit_request = journal->j_commit_sequence;
923 journal->j_max_transaction_buffers = journal->j_maxlen / 4;
926 * As a special case, if the on-disk copy is already marked as needing
927 * no recovery (s_start == 0), then we can safely defer the superblock
928 * update until the next commit by setting JFS_FLUSHED. This avoids
929 * attempting a write to a potential-readonly device.
931 if (sb->s_start == 0) {
932 jbd_debug(1,"JBD: Skipping superblock update on recovered sb "
933 "(start %u, seq %d, errno %d)\n",
934 journal->j_tail, journal->j_tail_sequence,
936 journal->j_flags |= JFS_FLUSHED;
938 /* Lock here to make assertions happy... */
939 mutex_lock(&journal->j_checkpoint_mutex);
941 * Update log tail information. We use WRITE_FUA since new
942 * transaction will start reusing journal space and so we
943 * must make sure information about current log tail is on
946 journal_update_sb_log_tail(journal,
947 journal->j_tail_sequence,
950 mutex_unlock(&journal->j_checkpoint_mutex);
952 return journal_start_thread(journal);
956 * int journal_create() - Initialise the new journal file
957 * @journal: Journal to create. This structure must have been initialised
959 * Given a journal_t structure which tells us which disk blocks we can
960 * use, create a new journal superblock and initialise all of the
961 * journal fields from scratch.
963 int journal_create(journal_t *journal)
965 unsigned int blocknr;
966 struct buffer_head *bh;
967 journal_superblock_t *sb;
970 if (journal->j_maxlen < JFS_MIN_JOURNAL_BLOCKS) {
971 printk (KERN_ERR "Journal length (%d blocks) too short.\n",
973 journal_fail_superblock(journal);
977 if (journal->j_inode == NULL) {
979 * We don't know what block to start at!
982 "%s: creation of journal on external device!\n",
987 /* Zero out the entire journal on disk. We cannot afford to
988 have any blocks on disk beginning with JFS_MAGIC_NUMBER. */
989 jbd_debug(1, "JBD: Zeroing out journal blocks...\n");
990 for (i = 0; i < journal->j_maxlen; i++) {
991 err = journal_bmap(journal, i, &blocknr);
994 bh = __getblk(journal->j_dev, blocknr, journal->j_blocksize);
998 memset (bh->b_data, 0, journal->j_blocksize);
999 BUFFER_TRACE(bh, "marking dirty");
1000 mark_buffer_dirty(bh);
1001 BUFFER_TRACE(bh, "marking uptodate");
1002 set_buffer_uptodate(bh);
1007 sync_blockdev(journal->j_dev);
1008 jbd_debug(1, "JBD: journal cleared.\n");
1010 /* OK, fill in the initial static fields in the new superblock */
1011 sb = journal->j_superblock;
1013 sb->s_header.h_magic = cpu_to_be32(JFS_MAGIC_NUMBER);
1014 sb->s_header.h_blocktype = cpu_to_be32(JFS_SUPERBLOCK_V2);
1016 sb->s_blocksize = cpu_to_be32(journal->j_blocksize);
1017 sb->s_maxlen = cpu_to_be32(journal->j_maxlen);
1018 sb->s_first = cpu_to_be32(1);
1020 journal->j_transaction_sequence = 1;
1022 journal->j_flags &= ~JFS_ABORT;
1023 journal->j_format_version = 2;
1025 return journal_reset(journal);
1028 static void journal_write_superblock(journal_t *journal, int write_op)
1030 struct buffer_head *bh = journal->j_sb_buffer;
1033 trace_journal_write_superblock(journal, write_op);
1034 if (!(journal->j_flags & JFS_BARRIER))
1035 write_op &= ~(REQ_FUA | REQ_FLUSH);
1037 if (buffer_write_io_error(bh)) {
1038 char b[BDEVNAME_SIZE];
1040 * Oh, dear. A previous attempt to write the journal
1041 * superblock failed. This could happen because the
1042 * USB device was yanked out. Or it could happen to
1043 * be a transient write error and maybe the block will
1044 * be remapped. Nothing we can do but to retry the
1045 * write and hope for the best.
1047 printk(KERN_ERR "JBD: previous I/O error detected "
1048 "for journal superblock update for %s.\n",
1049 journal_dev_name(journal, b));
1050 clear_buffer_write_io_error(bh);
1051 set_buffer_uptodate(bh);
1055 bh->b_end_io = end_buffer_write_sync;
1056 ret = submit_bh(write_op, bh);
1058 if (buffer_write_io_error(bh)) {
1059 clear_buffer_write_io_error(bh);
1060 set_buffer_uptodate(bh);
1064 char b[BDEVNAME_SIZE];
1065 printk(KERN_ERR "JBD: Error %d detected "
1066 "when updating journal superblock for %s.\n",
1067 ret, journal_dev_name(journal, b));
1072 * journal_update_sb_log_tail() - Update log tail in journal sb on disk.
1073 * @journal: The journal to update.
1074 * @tail_tid: TID of the new transaction at the tail of the log
1075 * @tail_block: The first block of the transaction at the tail of the log
1076 * @write_op: With which operation should we write the journal sb
1078 * Update a journal's superblock information about log tail and write it to
1079 * disk, waiting for the IO to complete.
1081 void journal_update_sb_log_tail(journal_t *journal, tid_t tail_tid,
1082 unsigned int tail_block, int write_op)
1084 journal_superblock_t *sb = journal->j_superblock;
1086 BUG_ON(!mutex_is_locked(&journal->j_checkpoint_mutex));
1087 jbd_debug(1,"JBD: updating superblock (start %u, seq %u)\n",
1088 tail_block, tail_tid);
1090 sb->s_sequence = cpu_to_be32(tail_tid);
1091 sb->s_start = cpu_to_be32(tail_block);
1093 journal_write_superblock(journal, write_op);
1095 /* Log is no longer empty */
1096 spin_lock(&journal->j_state_lock);
1097 WARN_ON(!sb->s_sequence);
1098 journal->j_flags &= ~JFS_FLUSHED;
1099 spin_unlock(&journal->j_state_lock);
1103 * mark_journal_empty() - Mark on disk journal as empty.
1104 * @journal: The journal to update.
1106 * Update a journal's dynamic superblock fields to show that journal is empty.
1107 * Write updated superblock to disk waiting for IO to complete.
1109 static void mark_journal_empty(journal_t *journal)
1111 journal_superblock_t *sb = journal->j_superblock;
1113 BUG_ON(!mutex_is_locked(&journal->j_checkpoint_mutex));
1114 spin_lock(&journal->j_state_lock);
1115 /* Is it already empty? */
1116 if (sb->s_start == 0) {
1117 spin_unlock(&journal->j_state_lock);
1120 jbd_debug(1, "JBD: Marking journal as empty (seq %d)\n",
1121 journal->j_tail_sequence);
1123 sb->s_sequence = cpu_to_be32(journal->j_tail_sequence);
1124 sb->s_start = cpu_to_be32(0);
1125 spin_unlock(&journal->j_state_lock);
1127 journal_write_superblock(journal, WRITE_FUA);
1129 spin_lock(&journal->j_state_lock);
1131 journal->j_flags |= JFS_FLUSHED;
1132 spin_unlock(&journal->j_state_lock);
1136 * journal_update_sb_errno() - Update error in the journal.
1137 * @journal: The journal to update.
1139 * Update a journal's errno. Write updated superblock to disk waiting for IO
1142 static void journal_update_sb_errno(journal_t *journal)
1144 journal_superblock_t *sb = journal->j_superblock;
1146 spin_lock(&journal->j_state_lock);
1147 jbd_debug(1, "JBD: updating superblock error (errno %d)\n",
1149 sb->s_errno = cpu_to_be32(journal->j_errno);
1150 spin_unlock(&journal->j_state_lock);
1152 journal_write_superblock(journal, WRITE_SYNC);
1156 * Read the superblock for a given journal, performing initial
1157 * validation of the format.
1160 static int journal_get_superblock(journal_t *journal)
1162 struct buffer_head *bh;
1163 journal_superblock_t *sb;
1166 bh = journal->j_sb_buffer;
1168 J_ASSERT(bh != NULL);
1169 if (!buffer_uptodate(bh)) {
1170 ll_rw_block(READ, 1, &bh);
1172 if (!buffer_uptodate(bh)) {
1174 "JBD: IO error reading journal superblock\n");
1179 sb = journal->j_superblock;
1183 if (sb->s_header.h_magic != cpu_to_be32(JFS_MAGIC_NUMBER) ||
1184 sb->s_blocksize != cpu_to_be32(journal->j_blocksize)) {
1185 printk(KERN_WARNING "JBD: no valid journal superblock found\n");
1189 switch(be32_to_cpu(sb->s_header.h_blocktype)) {
1190 case JFS_SUPERBLOCK_V1:
1191 journal->j_format_version = 1;
1193 case JFS_SUPERBLOCK_V2:
1194 journal->j_format_version = 2;
1197 printk(KERN_WARNING "JBD: unrecognised superblock format ID\n");
1201 if (be32_to_cpu(sb->s_maxlen) < journal->j_maxlen)
1202 journal->j_maxlen = be32_to_cpu(sb->s_maxlen);
1203 else if (be32_to_cpu(sb->s_maxlen) > journal->j_maxlen) {
1204 printk (KERN_WARNING "JBD: journal file too short\n");
1208 if (be32_to_cpu(sb->s_first) == 0 ||
1209 be32_to_cpu(sb->s_first) >= journal->j_maxlen) {
1211 "JBD: Invalid start block of journal: %u\n",
1212 be32_to_cpu(sb->s_first));
1219 journal_fail_superblock(journal);
1224 * Load the on-disk journal superblock and read the key fields into the
1228 static int load_superblock(journal_t *journal)
1231 journal_superblock_t *sb;
1233 err = journal_get_superblock(journal);
1237 sb = journal->j_superblock;
1239 journal->j_tail_sequence = be32_to_cpu(sb->s_sequence);
1240 journal->j_tail = be32_to_cpu(sb->s_start);
1241 journal->j_first = be32_to_cpu(sb->s_first);
1242 journal->j_last = be32_to_cpu(sb->s_maxlen);
1243 journal->j_errno = be32_to_cpu(sb->s_errno);
1250 * int journal_load() - Read journal from disk.
1251 * @journal: Journal to act on.
1253 * Given a journal_t structure which tells us which disk blocks contain
1254 * a journal, read the journal from disk to initialise the in-memory
1257 int journal_load(journal_t *journal)
1260 journal_superblock_t *sb;
1262 err = load_superblock(journal);
1266 sb = journal->j_superblock;
1267 /* If this is a V2 superblock, then we have to check the
1268 * features flags on it. */
1270 if (journal->j_format_version >= 2) {
1271 if ((sb->s_feature_ro_compat &
1272 ~cpu_to_be32(JFS_KNOWN_ROCOMPAT_FEATURES)) ||
1273 (sb->s_feature_incompat &
1274 ~cpu_to_be32(JFS_KNOWN_INCOMPAT_FEATURES))) {
1275 printk (KERN_WARNING
1276 "JBD: Unrecognised features on journal\n");
1281 /* Let the recovery code check whether it needs to recover any
1282 * data from the journal. */
1283 if (journal_recover(journal))
1284 goto recovery_error;
1286 /* OK, we've finished with the dynamic journal bits:
1287 * reinitialise the dynamic contents of the superblock in memory
1288 * and reset them on disk. */
1289 if (journal_reset(journal))
1290 goto recovery_error;
1292 journal->j_flags &= ~JFS_ABORT;
1293 journal->j_flags |= JFS_LOADED;
1297 printk (KERN_WARNING "JBD: recovery failed\n");
1302 * void journal_destroy() - Release a journal_t structure.
1303 * @journal: Journal to act on.
1305 * Release a journal_t structure once it is no longer in use by the
1307 * Return <0 if we couldn't clean up the journal.
1309 int journal_destroy(journal_t *journal)
1314 /* Wait for the commit thread to wake up and die. */
1315 journal_kill_thread(journal);
1317 /* Force a final log commit */
1318 if (journal->j_running_transaction)
1319 journal_commit_transaction(journal);
1321 /* Force any old transactions to disk */
1323 /* We cannot race with anybody but must keep assertions happy */
1324 mutex_lock(&journal->j_checkpoint_mutex);
1325 /* Totally anal locking here... */
1326 spin_lock(&journal->j_list_lock);
1327 while (journal->j_checkpoint_transactions != NULL) {
1328 spin_unlock(&journal->j_list_lock);
1329 log_do_checkpoint(journal);
1330 spin_lock(&journal->j_list_lock);
1333 J_ASSERT(journal->j_running_transaction == NULL);
1334 J_ASSERT(journal->j_committing_transaction == NULL);
1335 J_ASSERT(journal->j_checkpoint_transactions == NULL);
1336 spin_unlock(&journal->j_list_lock);
1338 if (journal->j_sb_buffer) {
1339 if (!is_journal_aborted(journal)) {
1340 journal->j_tail_sequence =
1341 ++journal->j_transaction_sequence;
1342 mark_journal_empty(journal);
1345 brelse(journal->j_sb_buffer);
1347 mutex_unlock(&journal->j_checkpoint_mutex);
1349 if (journal->j_inode)
1350 iput(journal->j_inode);
1351 if (journal->j_revoke)
1352 journal_destroy_revoke(journal);
1353 kfree(journal->j_wbuf);
1361 *int journal_check_used_features () - Check if features specified are used.
1362 * @journal: Journal to check.
1363 * @compat: bitmask of compatible features
1364 * @ro: bitmask of features that force read-only mount
1365 * @incompat: bitmask of incompatible features
1367 * Check whether the journal uses all of a given set of
1368 * features. Return true (non-zero) if it does.
1371 int journal_check_used_features (journal_t *journal, unsigned long compat,
1372 unsigned long ro, unsigned long incompat)
1374 journal_superblock_t *sb;
1376 if (!compat && !ro && !incompat)
1378 if (journal->j_format_version == 1)
1381 sb = journal->j_superblock;
1383 if (((be32_to_cpu(sb->s_feature_compat) & compat) == compat) &&
1384 ((be32_to_cpu(sb->s_feature_ro_compat) & ro) == ro) &&
1385 ((be32_to_cpu(sb->s_feature_incompat) & incompat) == incompat))
1392 * int journal_check_available_features() - Check feature set in journalling layer
1393 * @journal: Journal to check.
1394 * @compat: bitmask of compatible features
1395 * @ro: bitmask of features that force read-only mount
1396 * @incompat: bitmask of incompatible features
1398 * Check whether the journaling code supports the use of
1399 * all of a given set of features on this journal. Return true
1400 * (non-zero) if it can. */
1402 int journal_check_available_features (journal_t *journal, unsigned long compat,
1403 unsigned long ro, unsigned long incompat)
1405 if (!compat && !ro && !incompat)
1408 /* We can support any known requested features iff the
1409 * superblock is in version 2. Otherwise we fail to support any
1410 * extended sb features. */
1412 if (journal->j_format_version != 2)
1415 if ((compat & JFS_KNOWN_COMPAT_FEATURES) == compat &&
1416 (ro & JFS_KNOWN_ROCOMPAT_FEATURES) == ro &&
1417 (incompat & JFS_KNOWN_INCOMPAT_FEATURES) == incompat)
1424 * int journal_set_features () - Mark a given journal feature in the superblock
1425 * @journal: Journal to act on.
1426 * @compat: bitmask of compatible features
1427 * @ro: bitmask of features that force read-only mount
1428 * @incompat: bitmask of incompatible features
1430 * Mark a given journal feature as present on the
1431 * superblock. Returns true if the requested features could be set.
1435 int journal_set_features (journal_t *journal, unsigned long compat,
1436 unsigned long ro, unsigned long incompat)
1438 journal_superblock_t *sb;
1440 if (journal_check_used_features(journal, compat, ro, incompat))
1443 if (!journal_check_available_features(journal, compat, ro, incompat))
1446 jbd_debug(1, "Setting new features 0x%lx/0x%lx/0x%lx\n",
1447 compat, ro, incompat);
1449 sb = journal->j_superblock;
1451 sb->s_feature_compat |= cpu_to_be32(compat);
1452 sb->s_feature_ro_compat |= cpu_to_be32(ro);
1453 sb->s_feature_incompat |= cpu_to_be32(incompat);
1460 * int journal_update_format () - Update on-disk journal structure.
1461 * @journal: Journal to act on.
1463 * Given an initialised but unloaded journal struct, poke about in the
1464 * on-disk structure to update it to the most recent supported version.
1466 int journal_update_format (journal_t *journal)
1468 journal_superblock_t *sb;
1471 err = journal_get_superblock(journal);
1475 sb = journal->j_superblock;
1477 switch (be32_to_cpu(sb->s_header.h_blocktype)) {
1478 case JFS_SUPERBLOCK_V2:
1480 case JFS_SUPERBLOCK_V1:
1481 return journal_convert_superblock_v1(journal, sb);
1488 static int journal_convert_superblock_v1(journal_t *journal,
1489 journal_superblock_t *sb)
1491 int offset, blocksize;
1492 struct buffer_head *bh;
1495 "JBD: Converting superblock from version 1 to 2.\n");
1497 /* Pre-initialise new fields to zero */
1498 offset = ((char *) &(sb->s_feature_compat)) - ((char *) sb);
1499 blocksize = be32_to_cpu(sb->s_blocksize);
1500 memset(&sb->s_feature_compat, 0, blocksize-offset);
1502 sb->s_nr_users = cpu_to_be32(1);
1503 sb->s_header.h_blocktype = cpu_to_be32(JFS_SUPERBLOCK_V2);
1504 journal->j_format_version = 2;
1506 bh = journal->j_sb_buffer;
1507 BUFFER_TRACE(bh, "marking dirty");
1508 mark_buffer_dirty(bh);
1509 sync_dirty_buffer(bh);
1515 * int journal_flush () - Flush journal
1516 * @journal: Journal to act on.
1518 * Flush all data for a given journal to disk and empty the journal.
1519 * Filesystems can use this when remounting readonly to ensure that
1520 * recovery does not need to happen on remount.
1523 int journal_flush(journal_t *journal)
1526 transaction_t *transaction = NULL;
1528 spin_lock(&journal->j_state_lock);
1530 /* Force everything buffered to the log... */
1531 if (journal->j_running_transaction) {
1532 transaction = journal->j_running_transaction;
1533 __log_start_commit(journal, transaction->t_tid);
1534 } else if (journal->j_committing_transaction)
1535 transaction = journal->j_committing_transaction;
1537 /* Wait for the log commit to complete... */
1539 tid_t tid = transaction->t_tid;
1541 spin_unlock(&journal->j_state_lock);
1542 log_wait_commit(journal, tid);
1544 spin_unlock(&journal->j_state_lock);
1547 /* ...and flush everything in the log out to disk. */
1548 spin_lock(&journal->j_list_lock);
1549 while (!err && journal->j_checkpoint_transactions != NULL) {
1550 spin_unlock(&journal->j_list_lock);
1551 mutex_lock(&journal->j_checkpoint_mutex);
1552 err = log_do_checkpoint(journal);
1553 mutex_unlock(&journal->j_checkpoint_mutex);
1554 spin_lock(&journal->j_list_lock);
1556 spin_unlock(&journal->j_list_lock);
1558 if (is_journal_aborted(journal))
1561 mutex_lock(&journal->j_checkpoint_mutex);
1562 cleanup_journal_tail(journal);
1564 /* Finally, mark the journal as really needing no recovery.
1565 * This sets s_start==0 in the underlying superblock, which is
1566 * the magic code for a fully-recovered superblock. Any future
1567 * commits of data to the journal will restore the current
1569 mark_journal_empty(journal);
1570 mutex_unlock(&journal->j_checkpoint_mutex);
1571 spin_lock(&journal->j_state_lock);
1572 J_ASSERT(!journal->j_running_transaction);
1573 J_ASSERT(!journal->j_committing_transaction);
1574 J_ASSERT(!journal->j_checkpoint_transactions);
1575 J_ASSERT(journal->j_head == journal->j_tail);
1576 J_ASSERT(journal->j_tail_sequence == journal->j_transaction_sequence);
1577 spin_unlock(&journal->j_state_lock);
1582 * int journal_wipe() - Wipe journal contents
1583 * @journal: Journal to act on.
1584 * @write: flag (see below)
1586 * Wipe out all of the contents of a journal, safely. This will produce
1587 * a warning if the journal contains any valid recovery information.
1588 * Must be called between journal_init_*() and journal_load().
1590 * If 'write' is non-zero, then we wipe out the journal on disk; otherwise
1591 * we merely suppress recovery.
1594 int journal_wipe(journal_t *journal, int write)
1598 J_ASSERT (!(journal->j_flags & JFS_LOADED));
1600 err = load_superblock(journal);
1604 if (!journal->j_tail)
1607 printk (KERN_WARNING "JBD: %s recovery information on journal\n",
1608 write ? "Clearing" : "Ignoring");
1610 err = journal_skip_recovery(journal);
1612 /* Lock to make assertions happy... */
1613 mutex_lock(&journal->j_checkpoint_mutex);
1614 mark_journal_empty(journal);
1615 mutex_unlock(&journal->j_checkpoint_mutex);
1623 * journal_dev_name: format a character string to describe on what
1624 * device this journal is present.
1627 static const char *journal_dev_name(journal_t *journal, char *buffer)
1629 struct block_device *bdev;
1631 if (journal->j_inode)
1632 bdev = journal->j_inode->i_sb->s_bdev;
1634 bdev = journal->j_dev;
1636 return bdevname(bdev, buffer);
1640 * Journal abort has very specific semantics, which we describe
1641 * for journal abort.
1643 * Two internal function, which provide abort to te jbd layer
1648 * Quick version for internal journal use (doesn't lock the journal).
1649 * Aborts hard --- we mark the abort as occurred, but do _nothing_ else,
1650 * and don't attempt to make any other journal updates.
1652 static void __journal_abort_hard(journal_t *journal)
1654 transaction_t *transaction;
1655 char b[BDEVNAME_SIZE];
1657 if (journal->j_flags & JFS_ABORT)
1660 printk(KERN_ERR "Aborting journal on device %s.\n",
1661 journal_dev_name(journal, b));
1663 spin_lock(&journal->j_state_lock);
1664 journal->j_flags |= JFS_ABORT;
1665 transaction = journal->j_running_transaction;
1667 __log_start_commit(journal, transaction->t_tid);
1668 spin_unlock(&journal->j_state_lock);
1671 /* Soft abort: record the abort error status in the journal superblock,
1672 * but don't do any other IO. */
1673 static void __journal_abort_soft (journal_t *journal, int errno)
1675 if (journal->j_flags & JFS_ABORT)
1678 if (!journal->j_errno)
1679 journal->j_errno = errno;
1681 __journal_abort_hard(journal);
1684 journal_update_sb_errno(journal);
1688 * void journal_abort () - Shutdown the journal immediately.
1689 * @journal: the journal to shutdown.
1690 * @errno: an error number to record in the journal indicating
1691 * the reason for the shutdown.
1693 * Perform a complete, immediate shutdown of the ENTIRE
1694 * journal (not of a single transaction). This operation cannot be
1695 * undone without closing and reopening the journal.
1697 * The journal_abort function is intended to support higher level error
1698 * recovery mechanisms such as the ext2/ext3 remount-readonly error
1701 * Journal abort has very specific semantics. Any existing dirty,
1702 * unjournaled buffers in the main filesystem will still be written to
1703 * disk by bdflush, but the journaling mechanism will be suspended
1704 * immediately and no further transaction commits will be honoured.
1706 * Any dirty, journaled buffers will be written back to disk without
1707 * hitting the journal. Atomicity cannot be guaranteed on an aborted
1708 * filesystem, but we _do_ attempt to leave as much data as possible
1709 * behind for fsck to use for cleanup.
1711 * Any attempt to get a new transaction handle on a journal which is in
1712 * ABORT state will just result in an -EROFS error return. A
1713 * journal_stop on an existing handle will return -EIO if we have
1714 * entered abort state during the update.
1716 * Recursive transactions are not disturbed by journal abort until the
1717 * final journal_stop, which will receive the -EIO error.
1719 * Finally, the journal_abort call allows the caller to supply an errno
1720 * which will be recorded (if possible) in the journal superblock. This
1721 * allows a client to record failure conditions in the middle of a
1722 * transaction without having to complete the transaction to record the
1723 * failure to disk. ext3_error, for example, now uses this
1726 * Errors which originate from within the journaling layer will NOT
1727 * supply an errno; a null errno implies that absolutely no further
1728 * writes are done to the journal (unless there are any already in
1733 void journal_abort(journal_t *journal, int errno)
1735 __journal_abort_soft(journal, errno);
1739 * int journal_errno () - returns the journal's error state.
1740 * @journal: journal to examine.
1742 * This is the errno numbet set with journal_abort(), the last
1743 * time the journal was mounted - if the journal was stopped
1744 * without calling abort this will be 0.
1746 * If the journal has been aborted on this mount time -EROFS will
1749 int journal_errno(journal_t *journal)
1753 spin_lock(&journal->j_state_lock);
1754 if (journal->j_flags & JFS_ABORT)
1757 err = journal->j_errno;
1758 spin_unlock(&journal->j_state_lock);
1763 * int journal_clear_err () - clears the journal's error state
1764 * @journal: journal to act on.
1766 * An error must be cleared or Acked to take a FS out of readonly
1769 int journal_clear_err(journal_t *journal)
1773 spin_lock(&journal->j_state_lock);
1774 if (journal->j_flags & JFS_ABORT)
1777 journal->j_errno = 0;
1778 spin_unlock(&journal->j_state_lock);
1783 * void journal_ack_err() - Ack journal err.
1784 * @journal: journal to act on.
1786 * An error must be cleared or Acked to take a FS out of readonly
1789 void journal_ack_err(journal_t *journal)
1791 spin_lock(&journal->j_state_lock);
1792 if (journal->j_errno)
1793 journal->j_flags |= JFS_ACK_ERR;
1794 spin_unlock(&journal->j_state_lock);
1797 int journal_blocks_per_page(struct inode *inode)
1799 return 1 << (PAGE_CACHE_SHIFT - inode->i_sb->s_blocksize_bits);
1803 * Journal_head storage management
1805 static struct kmem_cache *journal_head_cache;
1806 #ifdef CONFIG_JBD_DEBUG
1807 static atomic_t nr_journal_heads = ATOMIC_INIT(0);
1810 static int journal_init_journal_head_cache(void)
1814 J_ASSERT(journal_head_cache == NULL);
1815 journal_head_cache = kmem_cache_create("journal_head",
1816 sizeof(struct journal_head),
1818 SLAB_TEMPORARY, /* flags */
1821 if (!journal_head_cache) {
1823 printk(KERN_EMERG "JBD: no memory for journal_head cache\n");
1828 static void journal_destroy_journal_head_cache(void)
1830 if (journal_head_cache) {
1831 kmem_cache_destroy(journal_head_cache);
1832 journal_head_cache = NULL;
1837 * journal_head splicing and dicing
1839 static struct journal_head *journal_alloc_journal_head(void)
1841 struct journal_head *ret;
1843 #ifdef CONFIG_JBD_DEBUG
1844 atomic_inc(&nr_journal_heads);
1846 ret = kmem_cache_alloc(journal_head_cache, GFP_NOFS);
1848 jbd_debug(1, "out of memory for journal_head\n");
1849 printk_ratelimited(KERN_NOTICE "ENOMEM in %s, retrying.\n",
1852 while (ret == NULL) {
1854 ret = kmem_cache_alloc(journal_head_cache, GFP_NOFS);
1860 static void journal_free_journal_head(struct journal_head *jh)
1862 #ifdef CONFIG_JBD_DEBUG
1863 atomic_dec(&nr_journal_heads);
1864 memset(jh, JBD_POISON_FREE, sizeof(*jh));
1866 kmem_cache_free(journal_head_cache, jh);
1870 * A journal_head is attached to a buffer_head whenever JBD has an
1871 * interest in the buffer.
1873 * Whenever a buffer has an attached journal_head, its ->b_state:BH_JBD bit
1874 * is set. This bit is tested in core kernel code where we need to take
1875 * JBD-specific actions. Testing the zeroness of ->b_private is not reliable
1878 * When a buffer has its BH_JBD bit set, its ->b_count is elevated by one.
1880 * When a buffer has its BH_JBD bit set it is immune from being released by
1881 * core kernel code, mainly via ->b_count.
1883 * A journal_head is detached from its buffer_head when the journal_head's
1884 * b_jcount reaches zero. Running transaction (b_transaction) and checkpoint
1885 * transaction (b_cp_transaction) hold their references to b_jcount.
1887 * Various places in the kernel want to attach a journal_head to a buffer_head
1888 * _before_ attaching the journal_head to a transaction. To protect the
1889 * journal_head in this situation, journal_add_journal_head elevates the
1890 * journal_head's b_jcount refcount by one. The caller must call
1891 * journal_put_journal_head() to undo this.
1893 * So the typical usage would be:
1895 * (Attach a journal_head if needed. Increments b_jcount)
1896 * struct journal_head *jh = journal_add_journal_head(bh);
1898 * (Get another reference for transaction)
1899 * journal_grab_journal_head(bh);
1900 * jh->b_transaction = xxx;
1901 * (Put original reference)
1902 * journal_put_journal_head(jh);
1906 * Give a buffer_head a journal_head.
1910 struct journal_head *journal_add_journal_head(struct buffer_head *bh)
1912 struct journal_head *jh;
1913 struct journal_head *new_jh = NULL;
1916 if (!buffer_jbd(bh)) {
1917 new_jh = journal_alloc_journal_head();
1918 memset(new_jh, 0, sizeof(*new_jh));
1921 jbd_lock_bh_journal_head(bh);
1922 if (buffer_jbd(bh)) {
1926 (atomic_read(&bh->b_count) > 0) ||
1927 (bh->b_page && bh->b_page->mapping));
1930 jbd_unlock_bh_journal_head(bh);
1935 new_jh = NULL; /* We consumed it */
1940 BUFFER_TRACE(bh, "added journal_head");
1943 jbd_unlock_bh_journal_head(bh);
1945 journal_free_journal_head(new_jh);
1946 return bh->b_private;
1950 * Grab a ref against this buffer_head's journal_head. If it ended up not
1951 * having a journal_head, return NULL
1953 struct journal_head *journal_grab_journal_head(struct buffer_head *bh)
1955 struct journal_head *jh = NULL;
1957 jbd_lock_bh_journal_head(bh);
1958 if (buffer_jbd(bh)) {
1962 jbd_unlock_bh_journal_head(bh);
1966 static void __journal_remove_journal_head(struct buffer_head *bh)
1968 struct journal_head *jh = bh2jh(bh);
1970 J_ASSERT_JH(jh, jh->b_jcount >= 0);
1971 J_ASSERT_JH(jh, jh->b_transaction == NULL);
1972 J_ASSERT_JH(jh, jh->b_next_transaction == NULL);
1973 J_ASSERT_JH(jh, jh->b_cp_transaction == NULL);
1974 J_ASSERT_JH(jh, jh->b_jlist == BJ_None);
1975 J_ASSERT_BH(bh, buffer_jbd(bh));
1976 J_ASSERT_BH(bh, jh2bh(jh) == bh);
1977 BUFFER_TRACE(bh, "remove journal_head");
1978 if (jh->b_frozen_data) {
1979 printk(KERN_WARNING "%s: freeing b_frozen_data\n", __func__);
1980 jbd_free(jh->b_frozen_data, bh->b_size);
1982 if (jh->b_committed_data) {
1983 printk(KERN_WARNING "%s: freeing b_committed_data\n", __func__);
1984 jbd_free(jh->b_committed_data, bh->b_size);
1986 bh->b_private = NULL;
1987 jh->b_bh = NULL; /* debug, really */
1988 clear_buffer_jbd(bh);
1989 journal_free_journal_head(jh);
1993 * Drop a reference on the passed journal_head. If it fell to zero then
1994 * release the journal_head from the buffer_head.
1996 void journal_put_journal_head(struct journal_head *jh)
1998 struct buffer_head *bh = jh2bh(jh);
2000 jbd_lock_bh_journal_head(bh);
2001 J_ASSERT_JH(jh, jh->b_jcount > 0);
2003 if (!jh->b_jcount) {
2004 __journal_remove_journal_head(bh);
2005 jbd_unlock_bh_journal_head(bh);
2008 jbd_unlock_bh_journal_head(bh);
2014 #ifdef CONFIG_JBD_DEBUG
2016 u8 journal_enable_debug __read_mostly;
2017 EXPORT_SYMBOL(journal_enable_debug);
2019 static struct dentry *jbd_debugfs_dir;
2020 static struct dentry *jbd_debug;
2022 static void __init jbd_create_debugfs_entry(void)
2024 jbd_debugfs_dir = debugfs_create_dir("jbd", NULL);
2025 if (jbd_debugfs_dir)
2026 jbd_debug = debugfs_create_u8("jbd-debug", S_IRUGO | S_IWUSR,
2028 &journal_enable_debug);
2031 static void __exit jbd_remove_debugfs_entry(void)
2033 debugfs_remove(jbd_debug);
2034 debugfs_remove(jbd_debugfs_dir);
2039 static inline void jbd_create_debugfs_entry(void)
2043 static inline void jbd_remove_debugfs_entry(void)
2049 struct kmem_cache *jbd_handle_cache;
2051 static int __init journal_init_handle_cache(void)
2053 jbd_handle_cache = kmem_cache_create("journal_handle",
2056 SLAB_TEMPORARY, /* flags */
2058 if (jbd_handle_cache == NULL) {
2059 printk(KERN_EMERG "JBD: failed to create handle cache\n");
2065 static void journal_destroy_handle_cache(void)
2067 if (jbd_handle_cache)
2068 kmem_cache_destroy(jbd_handle_cache);
2072 * Module startup and shutdown
2075 static int __init journal_init_caches(void)
2079 ret = journal_init_revoke_caches();
2081 ret = journal_init_journal_head_cache();
2083 ret = journal_init_handle_cache();
2087 static void journal_destroy_caches(void)
2089 journal_destroy_revoke_caches();
2090 journal_destroy_journal_head_cache();
2091 journal_destroy_handle_cache();
2094 static int __init journal_init(void)
2098 BUILD_BUG_ON(sizeof(struct journal_superblock_s) != 1024);
2100 ret = journal_init_caches();
2102 journal_destroy_caches();
2103 jbd_create_debugfs_entry();
2107 static void __exit journal_exit(void)
2109 #ifdef CONFIG_JBD_DEBUG
2110 int n = atomic_read(&nr_journal_heads);
2112 printk(KERN_EMERG "JBD: leaked %d journal_heads!\n", n);
2114 jbd_remove_debugfs_entry();
2115 journal_destroy_caches();
2118 MODULE_LICENSE("GPL");
2119 module_init(journal_init);
2120 module_exit(journal_exit);