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 */
314 init_buffer(new_bh, NULL, NULL);
315 atomic_set(&new_bh->b_count, 1);
316 new_jh = journal_add_journal_head(new_bh); /* This sleeps */
319 * If a new transaction has already done a buffer copy-out, then
320 * we use that version of the data for the commit.
322 jbd_lock_bh_state(bh_in);
324 if (jh_in->b_frozen_data) {
326 new_page = virt_to_page(jh_in->b_frozen_data);
327 new_offset = offset_in_page(jh_in->b_frozen_data);
329 new_page = jh2bh(jh_in)->b_page;
330 new_offset = offset_in_page(jh2bh(jh_in)->b_data);
333 mapped_data = kmap_atomic(new_page);
337 if (*((__be32 *)(mapped_data + new_offset)) ==
338 cpu_to_be32(JFS_MAGIC_NUMBER)) {
342 kunmap_atomic(mapped_data);
345 * Do we need to do a data copy?
347 if (need_copy_out && !done_copy_out) {
350 jbd_unlock_bh_state(bh_in);
351 tmp = jbd_alloc(bh_in->b_size, GFP_NOFS);
352 jbd_lock_bh_state(bh_in);
353 if (jh_in->b_frozen_data) {
354 jbd_free(tmp, bh_in->b_size);
358 jh_in->b_frozen_data = tmp;
359 mapped_data = kmap_atomic(new_page);
360 memcpy(tmp, mapped_data + new_offset, jh2bh(jh_in)->b_size);
361 kunmap_atomic(mapped_data);
363 new_page = virt_to_page(tmp);
364 new_offset = offset_in_page(tmp);
369 * Did we need to do an escaping? Now we've done all the
370 * copying, we can finally do so.
373 mapped_data = kmap_atomic(new_page);
374 *((unsigned int *)(mapped_data + new_offset)) = 0;
375 kunmap_atomic(mapped_data);
378 set_bh_page(new_bh, new_page, new_offset);
379 new_jh->b_transaction = NULL;
380 new_bh->b_size = jh2bh(jh_in)->b_size;
381 new_bh->b_bdev = transaction->t_journal->j_dev;
382 new_bh->b_blocknr = blocknr;
383 set_buffer_mapped(new_bh);
384 set_buffer_dirty(new_bh);
389 * The to-be-written buffer needs to get moved to the io queue,
390 * and the original buffer whose contents we are shadowing or
391 * copying is moved to the transaction's shadow queue.
393 JBUFFER_TRACE(jh_in, "file as BJ_Shadow");
394 spin_lock(&journal->j_list_lock);
395 __journal_file_buffer(jh_in, transaction, BJ_Shadow);
396 spin_unlock(&journal->j_list_lock);
397 jbd_unlock_bh_state(bh_in);
399 JBUFFER_TRACE(new_jh, "file as BJ_IO");
400 journal_file_buffer(new_jh, transaction, BJ_IO);
402 return do_escape | (done_copy_out << 1);
406 * Allocation code for the journal file. Manage the space left in the
407 * journal, so that we can begin checkpointing when appropriate.
411 * __log_space_left: Return the number of free blocks left in the journal.
413 * Called with the journal already locked.
415 * Called under j_state_lock
418 int __log_space_left(journal_t *journal)
420 int left = journal->j_free;
422 assert_spin_locked(&journal->j_state_lock);
425 * Be pessimistic here about the number of those free blocks which
426 * might be required for log descriptor control blocks.
429 #define MIN_LOG_RESERVED_BLOCKS 32 /* Allow for rounding errors */
431 left -= MIN_LOG_RESERVED_BLOCKS;
440 * Called under j_state_lock. Returns true if a transaction commit was started.
442 int __log_start_commit(journal_t *journal, tid_t target)
445 * The only transaction we can possibly wait upon is the
446 * currently running transaction (if it exists). Otherwise,
447 * the target tid must be an old one.
449 if (journal->j_commit_request != target &&
450 journal->j_running_transaction &&
451 journal->j_running_transaction->t_tid == target) {
453 * We want a new commit: OK, mark the request and wakeup the
454 * commit thread. We do _not_ do the commit ourselves.
457 journal->j_commit_request = target;
458 jbd_debug(1, "JBD: requesting commit %d/%d\n",
459 journal->j_commit_request,
460 journal->j_commit_sequence);
461 wake_up(&journal->j_wait_commit);
463 } else if (!tid_geq(journal->j_commit_request, target))
464 /* This should never happen, but if it does, preserve
465 the evidence before kjournald goes into a loop and
466 increments j_commit_sequence beyond all recognition. */
467 WARN_ONCE(1, "jbd: bad log_start_commit: %u %u %u %u\n",
468 journal->j_commit_request, journal->j_commit_sequence,
469 target, journal->j_running_transaction ?
470 journal->j_running_transaction->t_tid : 0);
474 int log_start_commit(journal_t *journal, tid_t tid)
478 spin_lock(&journal->j_state_lock);
479 ret = __log_start_commit(journal, tid);
480 spin_unlock(&journal->j_state_lock);
485 * Force and wait upon a commit if the calling process is not within
486 * transaction. This is used for forcing out undo-protected data which contains
487 * bitmaps, when the fs is running out of space.
489 * We can only force the running transaction if we don't have an active handle;
490 * otherwise, we will deadlock.
492 * Returns true if a transaction was started.
494 int journal_force_commit_nested(journal_t *journal)
496 transaction_t *transaction = NULL;
499 spin_lock(&journal->j_state_lock);
500 if (journal->j_running_transaction && !current->journal_info) {
501 transaction = journal->j_running_transaction;
502 __log_start_commit(journal, transaction->t_tid);
503 } else if (journal->j_committing_transaction)
504 transaction = journal->j_committing_transaction;
507 spin_unlock(&journal->j_state_lock);
508 return 0; /* Nothing to retry */
511 tid = transaction->t_tid;
512 spin_unlock(&journal->j_state_lock);
513 log_wait_commit(journal, tid);
518 * Start a commit of the current running transaction (if any). Returns true
519 * if a transaction is going to be committed (or is currently already
520 * committing), and fills its tid in at *ptid
522 int journal_start_commit(journal_t *journal, tid_t *ptid)
526 spin_lock(&journal->j_state_lock);
527 if (journal->j_running_transaction) {
528 tid_t tid = journal->j_running_transaction->t_tid;
530 __log_start_commit(journal, tid);
531 /* There's a running transaction and we've just made sure
532 * it's commit has been scheduled. */
536 } else if (journal->j_committing_transaction) {
538 * If commit has been started, then we have to wait for
539 * completion of that transaction.
542 *ptid = journal->j_committing_transaction->t_tid;
545 spin_unlock(&journal->j_state_lock);
550 * Wait for a specified commit to complete.
551 * The caller may not hold the journal lock.
553 int log_wait_commit(journal_t *journal, tid_t tid)
557 #ifdef CONFIG_JBD_DEBUG
558 spin_lock(&journal->j_state_lock);
559 if (!tid_geq(journal->j_commit_request, tid)) {
561 "%s: error: j_commit_request=%d, tid=%d\n",
562 __func__, journal->j_commit_request, tid);
564 spin_unlock(&journal->j_state_lock);
566 spin_lock(&journal->j_state_lock);
568 * Not running or committing trans? Must be already committed. This
569 * saves us from waiting for a *long* time when tid overflows.
571 if (!((journal->j_running_transaction &&
572 journal->j_running_transaction->t_tid == tid) ||
573 (journal->j_committing_transaction &&
574 journal->j_committing_transaction->t_tid == tid)))
577 if (!tid_geq(journal->j_commit_waited, tid))
578 journal->j_commit_waited = tid;
579 while (tid_gt(tid, journal->j_commit_sequence)) {
580 jbd_debug(1, "JBD: want %d, j_commit_sequence=%d\n",
581 tid, journal->j_commit_sequence);
582 wake_up(&journal->j_wait_commit);
583 spin_unlock(&journal->j_state_lock);
584 wait_event(journal->j_wait_done_commit,
585 !tid_gt(tid, journal->j_commit_sequence));
586 spin_lock(&journal->j_state_lock);
589 spin_unlock(&journal->j_state_lock);
591 if (unlikely(is_journal_aborted(journal))) {
592 printk(KERN_EMERG "journal commit I/O error\n");
599 * Return 1 if a given transaction has not yet sent barrier request
600 * connected with a transaction commit. If 0 is returned, transaction
601 * may or may not have sent the barrier. Used to avoid sending barrier
602 * twice in common cases.
604 int journal_trans_will_send_data_barrier(journal_t *journal, tid_t tid)
607 transaction_t *commit_trans;
609 if (!(journal->j_flags & JFS_BARRIER))
611 spin_lock(&journal->j_state_lock);
612 /* Transaction already committed? */
613 if (tid_geq(journal->j_commit_sequence, tid))
616 * Transaction is being committed and we already proceeded to
617 * writing commit record?
619 commit_trans = journal->j_committing_transaction;
620 if (commit_trans && commit_trans->t_tid == tid &&
621 commit_trans->t_state >= T_COMMIT_RECORD)
625 spin_unlock(&journal->j_state_lock);
628 EXPORT_SYMBOL(journal_trans_will_send_data_barrier);
631 * Log buffer allocation routines:
634 int journal_next_log_block(journal_t *journal, unsigned int *retp)
636 unsigned int blocknr;
638 spin_lock(&journal->j_state_lock);
639 J_ASSERT(journal->j_free > 1);
641 blocknr = journal->j_head;
644 if (journal->j_head == journal->j_last)
645 journal->j_head = journal->j_first;
646 spin_unlock(&journal->j_state_lock);
647 return journal_bmap(journal, blocknr, retp);
651 * Conversion of logical to physical block numbers for the journal
653 * On external journals the journal blocks are identity-mapped, so
654 * this is a no-op. If needed, we can use j_blk_offset - everything is
657 int journal_bmap(journal_t *journal, unsigned int blocknr,
663 if (journal->j_inode) {
664 ret = bmap(journal->j_inode, blocknr);
668 char b[BDEVNAME_SIZE];
670 printk(KERN_ALERT "%s: journal block not found "
671 "at offset %u on %s\n",
674 bdevname(journal->j_dev, b));
676 __journal_abort_soft(journal, err);
679 *retp = blocknr; /* +journal->j_blk_offset */
685 * We play buffer_head aliasing tricks to write data/metadata blocks to
686 * the journal without copying their contents, but for journal
687 * descriptor blocks we do need to generate bona fide buffers.
689 * After the caller of journal_get_descriptor_buffer() has finished modifying
690 * the buffer's contents they really should run flush_dcache_page(bh->b_page).
691 * But we don't bother doing that, so there will be coherency problems with
692 * mmaps of blockdevs which hold live JBD-controlled filesystems.
694 struct journal_head *journal_get_descriptor_buffer(journal_t *journal)
696 struct buffer_head *bh;
697 unsigned int blocknr;
700 err = journal_next_log_block(journal, &blocknr);
705 bh = __getblk(journal->j_dev, blocknr, journal->j_blocksize);
709 memset(bh->b_data, 0, journal->j_blocksize);
710 set_buffer_uptodate(bh);
712 BUFFER_TRACE(bh, "return this buffer");
713 return journal_add_journal_head(bh);
717 * Management for journal control blocks: functions to create and
718 * destroy journal_t structures, and to initialise and read existing
719 * journal blocks from disk. */
721 /* First: create and setup a journal_t object in memory. We initialise
722 * very few fields yet: that has to wait until we have created the
723 * journal structures from from scratch, or loaded them from disk. */
725 static journal_t * journal_init_common (void)
730 journal = kzalloc(sizeof(*journal), GFP_KERNEL);
734 init_waitqueue_head(&journal->j_wait_transaction_locked);
735 init_waitqueue_head(&journal->j_wait_logspace);
736 init_waitqueue_head(&journal->j_wait_done_commit);
737 init_waitqueue_head(&journal->j_wait_checkpoint);
738 init_waitqueue_head(&journal->j_wait_commit);
739 init_waitqueue_head(&journal->j_wait_updates);
740 mutex_init(&journal->j_checkpoint_mutex);
741 spin_lock_init(&journal->j_revoke_lock);
742 spin_lock_init(&journal->j_list_lock);
743 spin_lock_init(&journal->j_state_lock);
745 journal->j_commit_interval = (HZ * JBD_DEFAULT_MAX_COMMIT_AGE);
747 /* The journal is marked for error until we succeed with recovery! */
748 journal->j_flags = JFS_ABORT;
750 /* Set up a default-sized revoke table for the new mount. */
751 err = journal_init_revoke(journal, JOURNAL_REVOKE_DEFAULT_HASH);
761 /* journal_init_dev and journal_init_inode:
763 * Create a journal structure assigned some fixed set of disk blocks to
764 * the journal. We don't actually touch those disk blocks yet, but we
765 * need to set up all of the mapping information to tell the journaling
766 * system where the journal blocks are.
771 * journal_t * journal_init_dev() - creates and initialises a journal structure
772 * @bdev: Block device on which to create the journal
773 * @fs_dev: Device which hold journalled filesystem for this journal.
774 * @start: Block nr Start of journal.
775 * @len: Length of the journal in blocks.
776 * @blocksize: blocksize of journalling device
778 * Returns: a newly created journal_t *
780 * journal_init_dev creates a journal which maps a fixed contiguous
781 * range of blocks on an arbitrary block device.
784 journal_t * journal_init_dev(struct block_device *bdev,
785 struct block_device *fs_dev,
786 int start, int len, int blocksize)
788 journal_t *journal = journal_init_common();
789 struct buffer_head *bh;
795 /* journal descriptor can store up to n blocks -bzzz */
796 journal->j_blocksize = blocksize;
797 n = journal->j_blocksize / sizeof(journal_block_tag_t);
798 journal->j_wbufsize = n;
799 journal->j_wbuf = kmalloc(n * sizeof(struct buffer_head*), GFP_KERNEL);
800 if (!journal->j_wbuf) {
801 printk(KERN_ERR "%s: Can't allocate bhs for commit thread\n",
805 journal->j_dev = bdev;
806 journal->j_fs_dev = fs_dev;
807 journal->j_blk_offset = start;
808 journal->j_maxlen = len;
810 bh = __getblk(journal->j_dev, start, journal->j_blocksize);
813 "%s: Cannot get buffer for journal superblock\n",
817 journal->j_sb_buffer = bh;
818 journal->j_superblock = (journal_superblock_t *)bh->b_data;
822 kfree(journal->j_wbuf);
828 * journal_t * journal_init_inode () - creates a journal which maps to a inode.
829 * @inode: An inode to create the journal in
831 * journal_init_inode creates a journal which maps an on-disk inode as
832 * the journal. The inode must exist already, must support bmap() and
833 * must have all data blocks preallocated.
835 journal_t * journal_init_inode (struct inode *inode)
837 struct buffer_head *bh;
838 journal_t *journal = journal_init_common();
841 unsigned int blocknr;
846 journal->j_dev = journal->j_fs_dev = inode->i_sb->s_bdev;
847 journal->j_inode = inode;
849 "journal %p: inode %s/%ld, size %Ld, bits %d, blksize %ld\n",
850 journal, inode->i_sb->s_id, inode->i_ino,
851 (long long) inode->i_size,
852 inode->i_sb->s_blocksize_bits, inode->i_sb->s_blocksize);
854 journal->j_maxlen = inode->i_size >> inode->i_sb->s_blocksize_bits;
855 journal->j_blocksize = inode->i_sb->s_blocksize;
857 /* journal descriptor can store up to n blocks -bzzz */
858 n = journal->j_blocksize / sizeof(journal_block_tag_t);
859 journal->j_wbufsize = n;
860 journal->j_wbuf = kmalloc(n * sizeof(struct buffer_head*), GFP_KERNEL);
861 if (!journal->j_wbuf) {
862 printk(KERN_ERR "%s: Can't allocate bhs for commit thread\n",
867 err = journal_bmap(journal, 0, &blocknr);
868 /* If that failed, give up */
870 printk(KERN_ERR "%s: Cannot locate journal superblock\n",
875 bh = __getblk(journal->j_dev, blocknr, journal->j_blocksize);
878 "%s: Cannot get buffer for journal superblock\n",
882 journal->j_sb_buffer = bh;
883 journal->j_superblock = (journal_superblock_t *)bh->b_data;
887 kfree(journal->j_wbuf);
893 * If the journal init or create aborts, we need to mark the journal
894 * superblock as being NULL to prevent the journal destroy from writing
895 * back a bogus superblock.
897 static void journal_fail_superblock (journal_t *journal)
899 struct buffer_head *bh = journal->j_sb_buffer;
901 journal->j_sb_buffer = NULL;
905 * Given a journal_t structure, initialise the various fields for
906 * startup of a new journaling session. We use this both when creating
907 * a journal, and after recovering an old journal to reset it for
911 static int journal_reset(journal_t *journal)
913 journal_superblock_t *sb = journal->j_superblock;
914 unsigned int first, last;
916 first = be32_to_cpu(sb->s_first);
917 last = be32_to_cpu(sb->s_maxlen);
918 if (first + JFS_MIN_JOURNAL_BLOCKS > last + 1) {
919 printk(KERN_ERR "JBD: Journal too short (blocks %u-%u).\n",
921 journal_fail_superblock(journal);
925 journal->j_first = first;
926 journal->j_last = last;
928 journal->j_head = first;
929 journal->j_tail = first;
930 journal->j_free = last - first;
932 journal->j_tail_sequence = journal->j_transaction_sequence;
933 journal->j_commit_sequence = journal->j_transaction_sequence - 1;
934 journal->j_commit_request = journal->j_commit_sequence;
936 journal->j_max_transaction_buffers = journal->j_maxlen / 4;
939 * As a special case, if the on-disk copy is already marked as needing
940 * no recovery (s_start == 0), then we can safely defer the superblock
941 * update until the next commit by setting JFS_FLUSHED. This avoids
942 * attempting a write to a potential-readonly device.
944 if (sb->s_start == 0) {
945 jbd_debug(1,"JBD: Skipping superblock update on recovered sb "
946 "(start %u, seq %d, errno %d)\n",
947 journal->j_tail, journal->j_tail_sequence,
949 journal->j_flags |= JFS_FLUSHED;
951 /* Lock here to make assertions happy... */
952 mutex_lock(&journal->j_checkpoint_mutex);
954 * Update log tail information. We use WRITE_FUA since new
955 * transaction will start reusing journal space and so we
956 * must make sure information about current log tail is on
959 journal_update_sb_log_tail(journal,
960 journal->j_tail_sequence,
963 mutex_unlock(&journal->j_checkpoint_mutex);
965 return journal_start_thread(journal);
969 * int journal_create() - Initialise the new journal file
970 * @journal: Journal to create. This structure must have been initialised
972 * Given a journal_t structure which tells us which disk blocks we can
973 * use, create a new journal superblock and initialise all of the
974 * journal fields from scratch.
976 int journal_create(journal_t *journal)
978 unsigned int blocknr;
979 struct buffer_head *bh;
980 journal_superblock_t *sb;
983 if (journal->j_maxlen < JFS_MIN_JOURNAL_BLOCKS) {
984 printk (KERN_ERR "Journal length (%d blocks) too short.\n",
986 journal_fail_superblock(journal);
990 if (journal->j_inode == NULL) {
992 * We don't know what block to start at!
995 "%s: creation of journal on external device!\n",
1000 /* Zero out the entire journal on disk. We cannot afford to
1001 have any blocks on disk beginning with JFS_MAGIC_NUMBER. */
1002 jbd_debug(1, "JBD: Zeroing out journal blocks...\n");
1003 for (i = 0; i < journal->j_maxlen; i++) {
1004 err = journal_bmap(journal, i, &blocknr);
1007 bh = __getblk(journal->j_dev, blocknr, journal->j_blocksize);
1011 memset (bh->b_data, 0, journal->j_blocksize);
1012 BUFFER_TRACE(bh, "marking dirty");
1013 mark_buffer_dirty(bh);
1014 BUFFER_TRACE(bh, "marking uptodate");
1015 set_buffer_uptodate(bh);
1020 sync_blockdev(journal->j_dev);
1021 jbd_debug(1, "JBD: journal cleared.\n");
1023 /* OK, fill in the initial static fields in the new superblock */
1024 sb = journal->j_superblock;
1026 sb->s_header.h_magic = cpu_to_be32(JFS_MAGIC_NUMBER);
1027 sb->s_header.h_blocktype = cpu_to_be32(JFS_SUPERBLOCK_V2);
1029 sb->s_blocksize = cpu_to_be32(journal->j_blocksize);
1030 sb->s_maxlen = cpu_to_be32(journal->j_maxlen);
1031 sb->s_first = cpu_to_be32(1);
1033 journal->j_transaction_sequence = 1;
1035 journal->j_flags &= ~JFS_ABORT;
1036 journal->j_format_version = 2;
1038 return journal_reset(journal);
1041 static void journal_write_superblock(journal_t *journal, int write_op)
1043 struct buffer_head *bh = journal->j_sb_buffer;
1046 trace_journal_write_superblock(journal, write_op);
1047 if (!(journal->j_flags & JFS_BARRIER))
1048 write_op &= ~(REQ_FUA | REQ_FLUSH);
1050 if (buffer_write_io_error(bh)) {
1051 char b[BDEVNAME_SIZE];
1053 * Oh, dear. A previous attempt to write the journal
1054 * superblock failed. This could happen because the
1055 * USB device was yanked out. Or it could happen to
1056 * be a transient write error and maybe the block will
1057 * be remapped. Nothing we can do but to retry the
1058 * write and hope for the best.
1060 printk(KERN_ERR "JBD: previous I/O error detected "
1061 "for journal superblock update for %s.\n",
1062 journal_dev_name(journal, b));
1063 clear_buffer_write_io_error(bh);
1064 set_buffer_uptodate(bh);
1068 bh->b_end_io = end_buffer_write_sync;
1069 ret = submit_bh(write_op, bh);
1071 if (buffer_write_io_error(bh)) {
1072 clear_buffer_write_io_error(bh);
1073 set_buffer_uptodate(bh);
1077 char b[BDEVNAME_SIZE];
1078 printk(KERN_ERR "JBD: Error %d detected "
1079 "when updating journal superblock for %s.\n",
1080 ret, journal_dev_name(journal, b));
1085 * journal_update_sb_log_tail() - Update log tail in journal sb on disk.
1086 * @journal: The journal to update.
1087 * @tail_tid: TID of the new transaction at the tail of the log
1088 * @tail_block: The first block of the transaction at the tail of the log
1089 * @write_op: With which operation should we write the journal sb
1091 * Update a journal's superblock information about log tail and write it to
1092 * disk, waiting for the IO to complete.
1094 void journal_update_sb_log_tail(journal_t *journal, tid_t tail_tid,
1095 unsigned int tail_block, int write_op)
1097 journal_superblock_t *sb = journal->j_superblock;
1099 BUG_ON(!mutex_is_locked(&journal->j_checkpoint_mutex));
1100 jbd_debug(1,"JBD: updating superblock (start %u, seq %u)\n",
1101 tail_block, tail_tid);
1103 sb->s_sequence = cpu_to_be32(tail_tid);
1104 sb->s_start = cpu_to_be32(tail_block);
1106 journal_write_superblock(journal, write_op);
1108 /* Log is no longer empty */
1109 spin_lock(&journal->j_state_lock);
1110 WARN_ON(!sb->s_sequence);
1111 journal->j_flags &= ~JFS_FLUSHED;
1112 spin_unlock(&journal->j_state_lock);
1116 * mark_journal_empty() - Mark on disk journal as empty.
1117 * @journal: The journal to update.
1119 * Update a journal's dynamic superblock fields to show that journal is empty.
1120 * Write updated superblock to disk waiting for IO to complete.
1122 static void mark_journal_empty(journal_t *journal)
1124 journal_superblock_t *sb = journal->j_superblock;
1126 BUG_ON(!mutex_is_locked(&journal->j_checkpoint_mutex));
1127 spin_lock(&journal->j_state_lock);
1128 /* Is it already empty? */
1129 if (sb->s_start == 0) {
1130 spin_unlock(&journal->j_state_lock);
1133 jbd_debug(1, "JBD: Marking journal as empty (seq %d)\n",
1134 journal->j_tail_sequence);
1136 sb->s_sequence = cpu_to_be32(journal->j_tail_sequence);
1137 sb->s_start = cpu_to_be32(0);
1138 spin_unlock(&journal->j_state_lock);
1140 journal_write_superblock(journal, WRITE_FUA);
1142 spin_lock(&journal->j_state_lock);
1144 journal->j_flags |= JFS_FLUSHED;
1145 spin_unlock(&journal->j_state_lock);
1149 * journal_update_sb_errno() - Update error in the journal.
1150 * @journal: The journal to update.
1152 * Update a journal's errno. Write updated superblock to disk waiting for IO
1155 static void journal_update_sb_errno(journal_t *journal)
1157 journal_superblock_t *sb = journal->j_superblock;
1159 spin_lock(&journal->j_state_lock);
1160 jbd_debug(1, "JBD: updating superblock error (errno %d)\n",
1162 sb->s_errno = cpu_to_be32(journal->j_errno);
1163 spin_unlock(&journal->j_state_lock);
1165 journal_write_superblock(journal, WRITE_SYNC);
1169 * Read the superblock for a given journal, performing initial
1170 * validation of the format.
1173 static int journal_get_superblock(journal_t *journal)
1175 struct buffer_head *bh;
1176 journal_superblock_t *sb;
1179 bh = journal->j_sb_buffer;
1181 J_ASSERT(bh != NULL);
1182 if (!buffer_uptodate(bh)) {
1183 ll_rw_block(READ, 1, &bh);
1185 if (!buffer_uptodate(bh)) {
1187 "JBD: IO error reading journal superblock\n");
1192 sb = journal->j_superblock;
1196 if (sb->s_header.h_magic != cpu_to_be32(JFS_MAGIC_NUMBER) ||
1197 sb->s_blocksize != cpu_to_be32(journal->j_blocksize)) {
1198 printk(KERN_WARNING "JBD: no valid journal superblock found\n");
1202 switch(be32_to_cpu(sb->s_header.h_blocktype)) {
1203 case JFS_SUPERBLOCK_V1:
1204 journal->j_format_version = 1;
1206 case JFS_SUPERBLOCK_V2:
1207 journal->j_format_version = 2;
1210 printk(KERN_WARNING "JBD: unrecognised superblock format ID\n");
1214 if (be32_to_cpu(sb->s_maxlen) < journal->j_maxlen)
1215 journal->j_maxlen = be32_to_cpu(sb->s_maxlen);
1216 else if (be32_to_cpu(sb->s_maxlen) > journal->j_maxlen) {
1217 printk (KERN_WARNING "JBD: journal file too short\n");
1221 if (be32_to_cpu(sb->s_first) == 0 ||
1222 be32_to_cpu(sb->s_first) >= journal->j_maxlen) {
1224 "JBD: Invalid start block of journal: %u\n",
1225 be32_to_cpu(sb->s_first));
1232 journal_fail_superblock(journal);
1237 * Load the on-disk journal superblock and read the key fields into the
1241 static int load_superblock(journal_t *journal)
1244 journal_superblock_t *sb;
1246 err = journal_get_superblock(journal);
1250 sb = journal->j_superblock;
1252 journal->j_tail_sequence = be32_to_cpu(sb->s_sequence);
1253 journal->j_tail = be32_to_cpu(sb->s_start);
1254 journal->j_first = be32_to_cpu(sb->s_first);
1255 journal->j_last = be32_to_cpu(sb->s_maxlen);
1256 journal->j_errno = be32_to_cpu(sb->s_errno);
1263 * int journal_load() - Read journal from disk.
1264 * @journal: Journal to act on.
1266 * Given a journal_t structure which tells us which disk blocks contain
1267 * a journal, read the journal from disk to initialise the in-memory
1270 int journal_load(journal_t *journal)
1273 journal_superblock_t *sb;
1275 err = load_superblock(journal);
1279 sb = journal->j_superblock;
1280 /* If this is a V2 superblock, then we have to check the
1281 * features flags on it. */
1283 if (journal->j_format_version >= 2) {
1284 if ((sb->s_feature_ro_compat &
1285 ~cpu_to_be32(JFS_KNOWN_ROCOMPAT_FEATURES)) ||
1286 (sb->s_feature_incompat &
1287 ~cpu_to_be32(JFS_KNOWN_INCOMPAT_FEATURES))) {
1288 printk (KERN_WARNING
1289 "JBD: Unrecognised features on journal\n");
1294 /* Let the recovery code check whether it needs to recover any
1295 * data from the journal. */
1296 if (journal_recover(journal))
1297 goto recovery_error;
1299 /* OK, we've finished with the dynamic journal bits:
1300 * reinitialise the dynamic contents of the superblock in memory
1301 * and reset them on disk. */
1302 if (journal_reset(journal))
1303 goto recovery_error;
1305 journal->j_flags &= ~JFS_ABORT;
1306 journal->j_flags |= JFS_LOADED;
1310 printk (KERN_WARNING "JBD: recovery failed\n");
1315 * void journal_destroy() - Release a journal_t structure.
1316 * @journal: Journal to act on.
1318 * Release a journal_t structure once it is no longer in use by the
1320 * Return <0 if we couldn't clean up the journal.
1322 int journal_destroy(journal_t *journal)
1327 /* Wait for the commit thread to wake up and die. */
1328 journal_kill_thread(journal);
1330 /* Force a final log commit */
1331 if (journal->j_running_transaction)
1332 journal_commit_transaction(journal);
1334 /* Force any old transactions to disk */
1336 /* We cannot race with anybody but must keep assertions happy */
1337 mutex_lock(&journal->j_checkpoint_mutex);
1338 /* Totally anal locking here... */
1339 spin_lock(&journal->j_list_lock);
1340 while (journal->j_checkpoint_transactions != NULL) {
1341 spin_unlock(&journal->j_list_lock);
1342 log_do_checkpoint(journal);
1343 spin_lock(&journal->j_list_lock);
1346 J_ASSERT(journal->j_running_transaction == NULL);
1347 J_ASSERT(journal->j_committing_transaction == NULL);
1348 J_ASSERT(journal->j_checkpoint_transactions == NULL);
1349 spin_unlock(&journal->j_list_lock);
1351 if (journal->j_sb_buffer) {
1352 if (!is_journal_aborted(journal)) {
1353 journal->j_tail_sequence =
1354 ++journal->j_transaction_sequence;
1355 mark_journal_empty(journal);
1358 brelse(journal->j_sb_buffer);
1360 mutex_unlock(&journal->j_checkpoint_mutex);
1362 if (journal->j_inode)
1363 iput(journal->j_inode);
1364 if (journal->j_revoke)
1365 journal_destroy_revoke(journal);
1366 kfree(journal->j_wbuf);
1374 *int journal_check_used_features () - Check if features specified are used.
1375 * @journal: Journal to check.
1376 * @compat: bitmask of compatible features
1377 * @ro: bitmask of features that force read-only mount
1378 * @incompat: bitmask of incompatible features
1380 * Check whether the journal uses all of a given set of
1381 * features. Return true (non-zero) if it does.
1384 int journal_check_used_features (journal_t *journal, unsigned long compat,
1385 unsigned long ro, unsigned long incompat)
1387 journal_superblock_t *sb;
1389 if (!compat && !ro && !incompat)
1391 if (journal->j_format_version == 1)
1394 sb = journal->j_superblock;
1396 if (((be32_to_cpu(sb->s_feature_compat) & compat) == compat) &&
1397 ((be32_to_cpu(sb->s_feature_ro_compat) & ro) == ro) &&
1398 ((be32_to_cpu(sb->s_feature_incompat) & incompat) == incompat))
1405 * int journal_check_available_features() - Check feature set in journalling layer
1406 * @journal: Journal to check.
1407 * @compat: bitmask of compatible features
1408 * @ro: bitmask of features that force read-only mount
1409 * @incompat: bitmask of incompatible features
1411 * Check whether the journaling code supports the use of
1412 * all of a given set of features on this journal. Return true
1413 * (non-zero) if it can. */
1415 int journal_check_available_features (journal_t *journal, unsigned long compat,
1416 unsigned long ro, unsigned long incompat)
1418 if (!compat && !ro && !incompat)
1421 /* We can support any known requested features iff the
1422 * superblock is in version 2. Otherwise we fail to support any
1423 * extended sb features. */
1425 if (journal->j_format_version != 2)
1428 if ((compat & JFS_KNOWN_COMPAT_FEATURES) == compat &&
1429 (ro & JFS_KNOWN_ROCOMPAT_FEATURES) == ro &&
1430 (incompat & JFS_KNOWN_INCOMPAT_FEATURES) == incompat)
1437 * int journal_set_features () - Mark a given journal feature in the superblock
1438 * @journal: Journal to act on.
1439 * @compat: bitmask of compatible features
1440 * @ro: bitmask of features that force read-only mount
1441 * @incompat: bitmask of incompatible features
1443 * Mark a given journal feature as present on the
1444 * superblock. Returns true if the requested features could be set.
1448 int journal_set_features (journal_t *journal, unsigned long compat,
1449 unsigned long ro, unsigned long incompat)
1451 journal_superblock_t *sb;
1453 if (journal_check_used_features(journal, compat, ro, incompat))
1456 if (!journal_check_available_features(journal, compat, ro, incompat))
1459 jbd_debug(1, "Setting new features 0x%lx/0x%lx/0x%lx\n",
1460 compat, ro, incompat);
1462 sb = journal->j_superblock;
1464 sb->s_feature_compat |= cpu_to_be32(compat);
1465 sb->s_feature_ro_compat |= cpu_to_be32(ro);
1466 sb->s_feature_incompat |= cpu_to_be32(incompat);
1473 * int journal_update_format () - Update on-disk journal structure.
1474 * @journal: Journal to act on.
1476 * Given an initialised but unloaded journal struct, poke about in the
1477 * on-disk structure to update it to the most recent supported version.
1479 int journal_update_format (journal_t *journal)
1481 journal_superblock_t *sb;
1484 err = journal_get_superblock(journal);
1488 sb = journal->j_superblock;
1490 switch (be32_to_cpu(sb->s_header.h_blocktype)) {
1491 case JFS_SUPERBLOCK_V2:
1493 case JFS_SUPERBLOCK_V1:
1494 return journal_convert_superblock_v1(journal, sb);
1501 static int journal_convert_superblock_v1(journal_t *journal,
1502 journal_superblock_t *sb)
1504 int offset, blocksize;
1505 struct buffer_head *bh;
1508 "JBD: Converting superblock from version 1 to 2.\n");
1510 /* Pre-initialise new fields to zero */
1511 offset = ((char *) &(sb->s_feature_compat)) - ((char *) sb);
1512 blocksize = be32_to_cpu(sb->s_blocksize);
1513 memset(&sb->s_feature_compat, 0, blocksize-offset);
1515 sb->s_nr_users = cpu_to_be32(1);
1516 sb->s_header.h_blocktype = cpu_to_be32(JFS_SUPERBLOCK_V2);
1517 journal->j_format_version = 2;
1519 bh = journal->j_sb_buffer;
1520 BUFFER_TRACE(bh, "marking dirty");
1521 mark_buffer_dirty(bh);
1522 sync_dirty_buffer(bh);
1528 * int journal_flush () - Flush journal
1529 * @journal: Journal to act on.
1531 * Flush all data for a given journal to disk and empty the journal.
1532 * Filesystems can use this when remounting readonly to ensure that
1533 * recovery does not need to happen on remount.
1536 int journal_flush(journal_t *journal)
1539 transaction_t *transaction = NULL;
1541 spin_lock(&journal->j_state_lock);
1543 /* Force everything buffered to the log... */
1544 if (journal->j_running_transaction) {
1545 transaction = journal->j_running_transaction;
1546 __log_start_commit(journal, transaction->t_tid);
1547 } else if (journal->j_committing_transaction)
1548 transaction = journal->j_committing_transaction;
1550 /* Wait for the log commit to complete... */
1552 tid_t tid = transaction->t_tid;
1554 spin_unlock(&journal->j_state_lock);
1555 log_wait_commit(journal, tid);
1557 spin_unlock(&journal->j_state_lock);
1560 /* ...and flush everything in the log out to disk. */
1561 spin_lock(&journal->j_list_lock);
1562 while (!err && journal->j_checkpoint_transactions != NULL) {
1563 spin_unlock(&journal->j_list_lock);
1564 mutex_lock(&journal->j_checkpoint_mutex);
1565 err = log_do_checkpoint(journal);
1566 mutex_unlock(&journal->j_checkpoint_mutex);
1567 spin_lock(&journal->j_list_lock);
1569 spin_unlock(&journal->j_list_lock);
1571 if (is_journal_aborted(journal))
1574 mutex_lock(&journal->j_checkpoint_mutex);
1575 cleanup_journal_tail(journal);
1577 /* Finally, mark the journal as really needing no recovery.
1578 * This sets s_start==0 in the underlying superblock, which is
1579 * the magic code for a fully-recovered superblock. Any future
1580 * commits of data to the journal will restore the current
1582 mark_journal_empty(journal);
1583 mutex_unlock(&journal->j_checkpoint_mutex);
1584 spin_lock(&journal->j_state_lock);
1585 J_ASSERT(!journal->j_running_transaction);
1586 J_ASSERT(!journal->j_committing_transaction);
1587 J_ASSERT(!journal->j_checkpoint_transactions);
1588 J_ASSERT(journal->j_head == journal->j_tail);
1589 J_ASSERT(journal->j_tail_sequence == journal->j_transaction_sequence);
1590 spin_unlock(&journal->j_state_lock);
1595 * int journal_wipe() - Wipe journal contents
1596 * @journal: Journal to act on.
1597 * @write: flag (see below)
1599 * Wipe out all of the contents of a journal, safely. This will produce
1600 * a warning if the journal contains any valid recovery information.
1601 * Must be called between journal_init_*() and journal_load().
1603 * If 'write' is non-zero, then we wipe out the journal on disk; otherwise
1604 * we merely suppress recovery.
1607 int journal_wipe(journal_t *journal, int write)
1611 J_ASSERT (!(journal->j_flags & JFS_LOADED));
1613 err = load_superblock(journal);
1617 if (!journal->j_tail)
1620 printk (KERN_WARNING "JBD: %s recovery information on journal\n",
1621 write ? "Clearing" : "Ignoring");
1623 err = journal_skip_recovery(journal);
1625 /* Lock to make assertions happy... */
1626 mutex_lock(&journal->j_checkpoint_mutex);
1627 mark_journal_empty(journal);
1628 mutex_unlock(&journal->j_checkpoint_mutex);
1636 * journal_dev_name: format a character string to describe on what
1637 * device this journal is present.
1640 static const char *journal_dev_name(journal_t *journal, char *buffer)
1642 struct block_device *bdev;
1644 if (journal->j_inode)
1645 bdev = journal->j_inode->i_sb->s_bdev;
1647 bdev = journal->j_dev;
1649 return bdevname(bdev, buffer);
1653 * Journal abort has very specific semantics, which we describe
1654 * for journal abort.
1656 * Two internal function, which provide abort to te jbd layer
1661 * Quick version for internal journal use (doesn't lock the journal).
1662 * Aborts hard --- we mark the abort as occurred, but do _nothing_ else,
1663 * and don't attempt to make any other journal updates.
1665 static void __journal_abort_hard(journal_t *journal)
1667 transaction_t *transaction;
1668 char b[BDEVNAME_SIZE];
1670 if (journal->j_flags & JFS_ABORT)
1673 printk(KERN_ERR "Aborting journal on device %s.\n",
1674 journal_dev_name(journal, b));
1676 spin_lock(&journal->j_state_lock);
1677 journal->j_flags |= JFS_ABORT;
1678 transaction = journal->j_running_transaction;
1680 __log_start_commit(journal, transaction->t_tid);
1681 spin_unlock(&journal->j_state_lock);
1684 /* Soft abort: record the abort error status in the journal superblock,
1685 * but don't do any other IO. */
1686 static void __journal_abort_soft (journal_t *journal, int errno)
1688 if (journal->j_flags & JFS_ABORT)
1691 if (!journal->j_errno)
1692 journal->j_errno = errno;
1694 __journal_abort_hard(journal);
1697 journal_update_sb_errno(journal);
1701 * void journal_abort () - Shutdown the journal immediately.
1702 * @journal: the journal to shutdown.
1703 * @errno: an error number to record in the journal indicating
1704 * the reason for the shutdown.
1706 * Perform a complete, immediate shutdown of the ENTIRE
1707 * journal (not of a single transaction). This operation cannot be
1708 * undone without closing and reopening the journal.
1710 * The journal_abort function is intended to support higher level error
1711 * recovery mechanisms such as the ext2/ext3 remount-readonly error
1714 * Journal abort has very specific semantics. Any existing dirty,
1715 * unjournaled buffers in the main filesystem will still be written to
1716 * disk by bdflush, but the journaling mechanism will be suspended
1717 * immediately and no further transaction commits will be honoured.
1719 * Any dirty, journaled buffers will be written back to disk without
1720 * hitting the journal. Atomicity cannot be guaranteed on an aborted
1721 * filesystem, but we _do_ attempt to leave as much data as possible
1722 * behind for fsck to use for cleanup.
1724 * Any attempt to get a new transaction handle on a journal which is in
1725 * ABORT state will just result in an -EROFS error return. A
1726 * journal_stop on an existing handle will return -EIO if we have
1727 * entered abort state during the update.
1729 * Recursive transactions are not disturbed by journal abort until the
1730 * final journal_stop, which will receive the -EIO error.
1732 * Finally, the journal_abort call allows the caller to supply an errno
1733 * which will be recorded (if possible) in the journal superblock. This
1734 * allows a client to record failure conditions in the middle of a
1735 * transaction without having to complete the transaction to record the
1736 * failure to disk. ext3_error, for example, now uses this
1739 * Errors which originate from within the journaling layer will NOT
1740 * supply an errno; a null errno implies that absolutely no further
1741 * writes are done to the journal (unless there are any already in
1746 void journal_abort(journal_t *journal, int errno)
1748 __journal_abort_soft(journal, errno);
1752 * int journal_errno () - returns the journal's error state.
1753 * @journal: journal to examine.
1755 * This is the errno numbet set with journal_abort(), the last
1756 * time the journal was mounted - if the journal was stopped
1757 * without calling abort this will be 0.
1759 * If the journal has been aborted on this mount time -EROFS will
1762 int journal_errno(journal_t *journal)
1766 spin_lock(&journal->j_state_lock);
1767 if (journal->j_flags & JFS_ABORT)
1770 err = journal->j_errno;
1771 spin_unlock(&journal->j_state_lock);
1776 * int journal_clear_err () - clears the journal's error state
1777 * @journal: journal to act on.
1779 * An error must be cleared or Acked to take a FS out of readonly
1782 int journal_clear_err(journal_t *journal)
1786 spin_lock(&journal->j_state_lock);
1787 if (journal->j_flags & JFS_ABORT)
1790 journal->j_errno = 0;
1791 spin_unlock(&journal->j_state_lock);
1796 * void journal_ack_err() - Ack journal err.
1797 * @journal: journal to act on.
1799 * An error must be cleared or Acked to take a FS out of readonly
1802 void journal_ack_err(journal_t *journal)
1804 spin_lock(&journal->j_state_lock);
1805 if (journal->j_errno)
1806 journal->j_flags |= JFS_ACK_ERR;
1807 spin_unlock(&journal->j_state_lock);
1810 int journal_blocks_per_page(struct inode *inode)
1812 return 1 << (PAGE_CACHE_SHIFT - inode->i_sb->s_blocksize_bits);
1816 * Journal_head storage management
1818 static struct kmem_cache *journal_head_cache;
1819 #ifdef CONFIG_JBD_DEBUG
1820 static atomic_t nr_journal_heads = ATOMIC_INIT(0);
1823 static int journal_init_journal_head_cache(void)
1827 J_ASSERT(journal_head_cache == NULL);
1828 journal_head_cache = kmem_cache_create("journal_head",
1829 sizeof(struct journal_head),
1831 SLAB_TEMPORARY, /* flags */
1834 if (!journal_head_cache) {
1836 printk(KERN_EMERG "JBD: no memory for journal_head cache\n");
1841 static void journal_destroy_journal_head_cache(void)
1843 if (journal_head_cache) {
1844 kmem_cache_destroy(journal_head_cache);
1845 journal_head_cache = NULL;
1850 * journal_head splicing and dicing
1852 static struct journal_head *journal_alloc_journal_head(void)
1854 struct journal_head *ret;
1856 #ifdef CONFIG_JBD_DEBUG
1857 atomic_inc(&nr_journal_heads);
1859 ret = kmem_cache_zalloc(journal_head_cache, GFP_NOFS);
1861 jbd_debug(1, "out of memory for journal_head\n");
1862 printk_ratelimited(KERN_NOTICE "ENOMEM in %s, retrying.\n",
1865 while (ret == NULL) {
1867 ret = kmem_cache_zalloc(journal_head_cache, GFP_NOFS);
1873 static void journal_free_journal_head(struct journal_head *jh)
1875 #ifdef CONFIG_JBD_DEBUG
1876 atomic_dec(&nr_journal_heads);
1877 memset(jh, JBD_POISON_FREE, sizeof(*jh));
1879 kmem_cache_free(journal_head_cache, jh);
1883 * A journal_head is attached to a buffer_head whenever JBD has an
1884 * interest in the buffer.
1886 * Whenever a buffer has an attached journal_head, its ->b_state:BH_JBD bit
1887 * is set. This bit is tested in core kernel code where we need to take
1888 * JBD-specific actions. Testing the zeroness of ->b_private is not reliable
1891 * When a buffer has its BH_JBD bit set, its ->b_count is elevated by one.
1893 * When a buffer has its BH_JBD bit set it is immune from being released by
1894 * core kernel code, mainly via ->b_count.
1896 * A journal_head is detached from its buffer_head when the journal_head's
1897 * b_jcount reaches zero. Running transaction (b_transaction) and checkpoint
1898 * transaction (b_cp_transaction) hold their references to b_jcount.
1900 * Various places in the kernel want to attach a journal_head to a buffer_head
1901 * _before_ attaching the journal_head to a transaction. To protect the
1902 * journal_head in this situation, journal_add_journal_head elevates the
1903 * journal_head's b_jcount refcount by one. The caller must call
1904 * journal_put_journal_head() to undo this.
1906 * So the typical usage would be:
1908 * (Attach a journal_head if needed. Increments b_jcount)
1909 * struct journal_head *jh = journal_add_journal_head(bh);
1911 * (Get another reference for transaction)
1912 * journal_grab_journal_head(bh);
1913 * jh->b_transaction = xxx;
1914 * (Put original reference)
1915 * journal_put_journal_head(jh);
1919 * Give a buffer_head a journal_head.
1923 struct journal_head *journal_add_journal_head(struct buffer_head *bh)
1925 struct journal_head *jh;
1926 struct journal_head *new_jh = NULL;
1929 if (!buffer_jbd(bh))
1930 new_jh = journal_alloc_journal_head();
1932 jbd_lock_bh_journal_head(bh);
1933 if (buffer_jbd(bh)) {
1937 (atomic_read(&bh->b_count) > 0) ||
1938 (bh->b_page && bh->b_page->mapping));
1941 jbd_unlock_bh_journal_head(bh);
1946 new_jh = NULL; /* We consumed it */
1951 BUFFER_TRACE(bh, "added journal_head");
1954 jbd_unlock_bh_journal_head(bh);
1956 journal_free_journal_head(new_jh);
1957 return bh->b_private;
1961 * Grab a ref against this buffer_head's journal_head. If it ended up not
1962 * having a journal_head, return NULL
1964 struct journal_head *journal_grab_journal_head(struct buffer_head *bh)
1966 struct journal_head *jh = NULL;
1968 jbd_lock_bh_journal_head(bh);
1969 if (buffer_jbd(bh)) {
1973 jbd_unlock_bh_journal_head(bh);
1977 static void __journal_remove_journal_head(struct buffer_head *bh)
1979 struct journal_head *jh = bh2jh(bh);
1981 J_ASSERT_JH(jh, jh->b_jcount >= 0);
1982 J_ASSERT_JH(jh, jh->b_transaction == NULL);
1983 J_ASSERT_JH(jh, jh->b_next_transaction == NULL);
1984 J_ASSERT_JH(jh, jh->b_cp_transaction == NULL);
1985 J_ASSERT_JH(jh, jh->b_jlist == BJ_None);
1986 J_ASSERT_BH(bh, buffer_jbd(bh));
1987 J_ASSERT_BH(bh, jh2bh(jh) == bh);
1988 BUFFER_TRACE(bh, "remove journal_head");
1989 if (jh->b_frozen_data) {
1990 printk(KERN_WARNING "%s: freeing b_frozen_data\n", __func__);
1991 jbd_free(jh->b_frozen_data, bh->b_size);
1993 if (jh->b_committed_data) {
1994 printk(KERN_WARNING "%s: freeing b_committed_data\n", __func__);
1995 jbd_free(jh->b_committed_data, bh->b_size);
1997 bh->b_private = NULL;
1998 jh->b_bh = NULL; /* debug, really */
1999 clear_buffer_jbd(bh);
2000 journal_free_journal_head(jh);
2004 * Drop a reference on the passed journal_head. If it fell to zero then
2005 * release the journal_head from the buffer_head.
2007 void journal_put_journal_head(struct journal_head *jh)
2009 struct buffer_head *bh = jh2bh(jh);
2011 jbd_lock_bh_journal_head(bh);
2012 J_ASSERT_JH(jh, jh->b_jcount > 0);
2014 if (!jh->b_jcount) {
2015 __journal_remove_journal_head(bh);
2016 jbd_unlock_bh_journal_head(bh);
2019 jbd_unlock_bh_journal_head(bh);
2025 #ifdef CONFIG_JBD_DEBUG
2027 u8 journal_enable_debug __read_mostly;
2028 EXPORT_SYMBOL(journal_enable_debug);
2030 static struct dentry *jbd_debugfs_dir;
2031 static struct dentry *jbd_debug;
2033 static void __init jbd_create_debugfs_entry(void)
2035 jbd_debugfs_dir = debugfs_create_dir("jbd", NULL);
2036 if (jbd_debugfs_dir)
2037 jbd_debug = debugfs_create_u8("jbd-debug", S_IRUGO | S_IWUSR,
2039 &journal_enable_debug);
2042 static void __exit jbd_remove_debugfs_entry(void)
2044 debugfs_remove(jbd_debug);
2045 debugfs_remove(jbd_debugfs_dir);
2050 static inline void jbd_create_debugfs_entry(void)
2054 static inline void jbd_remove_debugfs_entry(void)
2060 struct kmem_cache *jbd_handle_cache;
2062 static int __init journal_init_handle_cache(void)
2064 jbd_handle_cache = kmem_cache_create("journal_handle",
2067 SLAB_TEMPORARY, /* flags */
2069 if (jbd_handle_cache == NULL) {
2070 printk(KERN_EMERG "JBD: failed to create handle cache\n");
2076 static void journal_destroy_handle_cache(void)
2078 if (jbd_handle_cache)
2079 kmem_cache_destroy(jbd_handle_cache);
2083 * Module startup and shutdown
2086 static int __init journal_init_caches(void)
2090 ret = journal_init_revoke_caches();
2092 ret = journal_init_journal_head_cache();
2094 ret = journal_init_handle_cache();
2098 static void journal_destroy_caches(void)
2100 journal_destroy_revoke_caches();
2101 journal_destroy_journal_head_cache();
2102 journal_destroy_handle_cache();
2105 static int __init journal_init(void)
2109 BUILD_BUG_ON(sizeof(struct journal_superblock_s) != 1024);
2111 ret = journal_init_caches();
2113 journal_destroy_caches();
2114 jbd_create_debugfs_entry();
2118 static void __exit journal_exit(void)
2120 #ifdef CONFIG_JBD_DEBUG
2121 int n = atomic_read(&nr_journal_heads);
2123 printk(KERN_EMERG "JBD: leaked %d journal_heads!\n", n);
2125 jbd_remove_debugfs_entry();
2126 journal_destroy_caches();
2129 MODULE_LICENSE("GPL");
2130 module_init(journal_init);
2131 module_exit(journal_exit);