2 * linux/fs/jbd2/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/jbd2.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/seq_file.h>
40 #include <linux/math64.h>
41 #include <linux/hash.h>
42 #include <linux/log2.h>
43 #include <linux/vmalloc.h>
44 #include <linux/backing-dev.h>
45 #include <linux/bitops.h>
46 #include <linux/ratelimit.h>
48 #define CREATE_TRACE_POINTS
49 #include <trace/events/jbd2.h>
51 #include <asm/uaccess.h>
53 #include <asm/system.h>
55 EXPORT_SYMBOL(jbd2_journal_extend);
56 EXPORT_SYMBOL(jbd2_journal_stop);
57 EXPORT_SYMBOL(jbd2_journal_lock_updates);
58 EXPORT_SYMBOL(jbd2_journal_unlock_updates);
59 EXPORT_SYMBOL(jbd2_journal_get_write_access);
60 EXPORT_SYMBOL(jbd2_journal_get_create_access);
61 EXPORT_SYMBOL(jbd2_journal_get_undo_access);
62 EXPORT_SYMBOL(jbd2_journal_set_triggers);
63 EXPORT_SYMBOL(jbd2_journal_dirty_metadata);
64 EXPORT_SYMBOL(jbd2_journal_release_buffer);
65 EXPORT_SYMBOL(jbd2_journal_forget);
67 EXPORT_SYMBOL(journal_sync_buffer);
69 EXPORT_SYMBOL(jbd2_journal_flush);
70 EXPORT_SYMBOL(jbd2_journal_revoke);
72 EXPORT_SYMBOL(jbd2_journal_init_dev);
73 EXPORT_SYMBOL(jbd2_journal_init_inode);
74 EXPORT_SYMBOL(jbd2_journal_update_format);
75 EXPORT_SYMBOL(jbd2_journal_check_used_features);
76 EXPORT_SYMBOL(jbd2_journal_check_available_features);
77 EXPORT_SYMBOL(jbd2_journal_set_features);
78 EXPORT_SYMBOL(jbd2_journal_load);
79 EXPORT_SYMBOL(jbd2_journal_destroy);
80 EXPORT_SYMBOL(jbd2_journal_abort);
81 EXPORT_SYMBOL(jbd2_journal_errno);
82 EXPORT_SYMBOL(jbd2_journal_ack_err);
83 EXPORT_SYMBOL(jbd2_journal_clear_err);
84 EXPORT_SYMBOL(jbd2_log_wait_commit);
85 EXPORT_SYMBOL(jbd2_log_start_commit);
86 EXPORT_SYMBOL(jbd2_journal_start_commit);
87 EXPORT_SYMBOL(jbd2_journal_force_commit_nested);
88 EXPORT_SYMBOL(jbd2_journal_wipe);
89 EXPORT_SYMBOL(jbd2_journal_blocks_per_page);
90 EXPORT_SYMBOL(jbd2_journal_invalidatepage);
91 EXPORT_SYMBOL(jbd2_journal_try_to_free_buffers);
92 EXPORT_SYMBOL(jbd2_journal_force_commit);
93 EXPORT_SYMBOL(jbd2_journal_file_inode);
94 EXPORT_SYMBOL(jbd2_journal_init_jbd_inode);
95 EXPORT_SYMBOL(jbd2_journal_release_jbd_inode);
96 EXPORT_SYMBOL(jbd2_journal_begin_ordered_truncate);
98 static int journal_convert_superblock_v1(journal_t *, journal_superblock_t *);
99 static void __journal_abort_soft (journal_t *journal, int errno);
100 static int jbd2_journal_create_slab(size_t slab_size);
103 * Helper function used to manage commit timeouts
106 static void commit_timeout(unsigned long __data)
108 struct task_struct * p = (struct task_struct *) __data;
114 * kjournald2: The main thread function used to manage a logging device
117 * This kernel thread is responsible for two things:
119 * 1) COMMIT: Every so often we need to commit the current state of the
120 * filesystem to disk. The journal thread is responsible for writing
121 * all of the metadata buffers to disk.
123 * 2) CHECKPOINT: We cannot reuse a used section of the log file until all
124 * of the data in that part of the log has been rewritten elsewhere on
125 * the disk. Flushing these old buffers to reclaim space in the log is
126 * known as checkpointing, and this thread is responsible for that job.
129 static int kjournald2(void *arg)
131 journal_t *journal = arg;
132 transaction_t *transaction;
135 * Set up an interval timer which can be used to trigger a commit wakeup
136 * after the commit interval expires
138 setup_timer(&journal->j_commit_timer, commit_timeout,
139 (unsigned long)current);
141 /* Record that the journal thread is running */
142 journal->j_task = current;
143 wake_up(&journal->j_wait_done_commit);
146 * And now, wait forever for commit wakeup events.
148 write_lock(&journal->j_state_lock);
151 if (journal->j_flags & JBD2_UNMOUNT)
154 jbd_debug(1, "commit_sequence=%d, commit_request=%d\n",
155 journal->j_commit_sequence, journal->j_commit_request);
157 if (journal->j_commit_sequence != journal->j_commit_request) {
158 jbd_debug(1, "OK, requests differ\n");
159 write_unlock(&journal->j_state_lock);
160 del_timer_sync(&journal->j_commit_timer);
161 jbd2_journal_commit_transaction(journal);
162 write_lock(&journal->j_state_lock);
166 wake_up(&journal->j_wait_done_commit);
167 if (freezing(current)) {
169 * The simpler the better. Flushing journal isn't a
170 * good idea, because that depends on threads that may
171 * be already stopped.
173 jbd_debug(1, "Now suspending kjournald2\n");
174 write_unlock(&journal->j_state_lock);
176 write_lock(&journal->j_state_lock);
179 * We assume on resume that commits are already there,
183 int should_sleep = 1;
185 prepare_to_wait(&journal->j_wait_commit, &wait,
187 if (journal->j_commit_sequence != journal->j_commit_request)
189 transaction = journal->j_running_transaction;
190 if (transaction && time_after_eq(jiffies,
191 transaction->t_expires))
193 if (journal->j_flags & JBD2_UNMOUNT)
196 write_unlock(&journal->j_state_lock);
198 write_lock(&journal->j_state_lock);
200 finish_wait(&journal->j_wait_commit, &wait);
203 jbd_debug(1, "kjournald2 wakes\n");
206 * Were we woken up by a commit wakeup event?
208 transaction = journal->j_running_transaction;
209 if (transaction && time_after_eq(jiffies, transaction->t_expires)) {
210 journal->j_commit_request = transaction->t_tid;
211 jbd_debug(1, "woke because of timeout\n");
216 write_unlock(&journal->j_state_lock);
217 del_timer_sync(&journal->j_commit_timer);
218 journal->j_task = NULL;
219 wake_up(&journal->j_wait_done_commit);
220 jbd_debug(1, "Journal thread exiting.\n");
224 static int jbd2_journal_start_thread(journal_t *journal)
226 struct task_struct *t;
228 t = kthread_run(kjournald2, journal, "jbd2/%s",
233 wait_event(journal->j_wait_done_commit, journal->j_task != NULL);
237 static void journal_kill_thread(journal_t *journal)
239 write_lock(&journal->j_state_lock);
240 journal->j_flags |= JBD2_UNMOUNT;
242 while (journal->j_task) {
243 wake_up(&journal->j_wait_commit);
244 write_unlock(&journal->j_state_lock);
245 wait_event(journal->j_wait_done_commit, journal->j_task == NULL);
246 write_lock(&journal->j_state_lock);
248 write_unlock(&journal->j_state_lock);
252 * jbd2_journal_write_metadata_buffer: write a metadata buffer to the journal.
254 * Writes a metadata buffer to a given disk block. The actual IO is not
255 * performed but a new buffer_head is constructed which labels the data
256 * to be written with the correct destination disk block.
258 * Any magic-number escaping which needs to be done will cause a
259 * copy-out here. If the buffer happens to start with the
260 * JBD2_MAGIC_NUMBER, then we can't write it to the log directly: the
261 * magic number is only written to the log for descripter blocks. In
262 * this case, we copy the data and replace the first word with 0, and we
263 * return a result code which indicates that this buffer needs to be
264 * marked as an escaped buffer in the corresponding log descriptor
265 * block. The missing word can then be restored when the block is read
268 * If the source buffer has already been modified by a new transaction
269 * since we took the last commit snapshot, we use the frozen copy of
270 * that data for IO. If we end up using the existing buffer_head's data
271 * for the write, then we *have* to lock the buffer to prevent anyone
272 * else from using and possibly modifying it while the IO is in
275 * The function returns a pointer to the buffer_heads to be used for IO.
277 * We assume that the journal has already been locked in this function.
284 * Bit 0 set == escape performed on the data
285 * Bit 1 set == buffer copy-out performed (kfree the data after IO)
288 int jbd2_journal_write_metadata_buffer(transaction_t *transaction,
289 struct journal_head *jh_in,
290 struct journal_head **jh_out,
291 unsigned long long blocknr)
293 int need_copy_out = 0;
294 int done_copy_out = 0;
297 struct buffer_head *new_bh;
298 struct journal_head *new_jh;
299 struct page *new_page;
300 unsigned int new_offset;
301 struct buffer_head *bh_in = jh2bh(jh_in);
302 journal_t *journal = transaction->t_journal;
305 * The buffer really shouldn't be locked: only the current committing
306 * transaction is allowed to write it, so nobody else is allowed
309 * akpm: except if we're journalling data, and write() output is
310 * also part of a shared mapping, and another thread has
311 * decided to launch a writepage() against this buffer.
313 J_ASSERT_BH(bh_in, buffer_jbddirty(bh_in));
316 new_bh = alloc_buffer_head(GFP_NOFS);
319 * Failure is not an option, but __GFP_NOFAIL is going
320 * away; so we retry ourselves here.
322 congestion_wait(BLK_RW_ASYNC, HZ/50);
326 /* keep subsequent assertions sane */
328 init_buffer(new_bh, NULL, NULL);
329 atomic_set(&new_bh->b_count, 1);
330 new_jh = jbd2_journal_add_journal_head(new_bh); /* This sleeps */
333 * If a new transaction has already done a buffer copy-out, then
334 * we use that version of the data for the commit.
336 jbd_lock_bh_state(bh_in);
338 if (jh_in->b_frozen_data) {
340 new_page = virt_to_page(jh_in->b_frozen_data);
341 new_offset = offset_in_page(jh_in->b_frozen_data);
343 new_page = jh2bh(jh_in)->b_page;
344 new_offset = offset_in_page(jh2bh(jh_in)->b_data);
347 mapped_data = kmap_atomic(new_page, KM_USER0);
349 * Fire data frozen trigger if data already wasn't frozen. Do this
350 * before checking for escaping, as the trigger may modify the magic
351 * offset. If a copy-out happens afterwards, it will have the correct
352 * data in the buffer.
355 jbd2_buffer_frozen_trigger(jh_in, mapped_data + new_offset,
361 if (*((__be32 *)(mapped_data + new_offset)) ==
362 cpu_to_be32(JBD2_MAGIC_NUMBER)) {
366 kunmap_atomic(mapped_data, KM_USER0);
369 * Do we need to do a data copy?
371 if (need_copy_out && !done_copy_out) {
374 jbd_unlock_bh_state(bh_in);
375 tmp = jbd2_alloc(bh_in->b_size, GFP_NOFS);
377 jbd2_journal_put_journal_head(new_jh);
380 jbd_lock_bh_state(bh_in);
381 if (jh_in->b_frozen_data) {
382 jbd2_free(tmp, bh_in->b_size);
386 jh_in->b_frozen_data = tmp;
387 mapped_data = kmap_atomic(new_page, KM_USER0);
388 memcpy(tmp, mapped_data + new_offset, jh2bh(jh_in)->b_size);
389 kunmap_atomic(mapped_data, KM_USER0);
391 new_page = virt_to_page(tmp);
392 new_offset = offset_in_page(tmp);
396 * This isn't strictly necessary, as we're using frozen
397 * data for the escaping, but it keeps consistency with
398 * b_frozen_data usage.
400 jh_in->b_frozen_triggers = jh_in->b_triggers;
404 * Did we need to do an escaping? Now we've done all the
405 * copying, we can finally do so.
408 mapped_data = kmap_atomic(new_page, KM_USER0);
409 *((unsigned int *)(mapped_data + new_offset)) = 0;
410 kunmap_atomic(mapped_data, KM_USER0);
413 set_bh_page(new_bh, new_page, new_offset);
414 new_jh->b_transaction = NULL;
415 new_bh->b_size = jh2bh(jh_in)->b_size;
416 new_bh->b_bdev = transaction->t_journal->j_dev;
417 new_bh->b_blocknr = blocknr;
418 set_buffer_mapped(new_bh);
419 set_buffer_dirty(new_bh);
424 * The to-be-written buffer needs to get moved to the io queue,
425 * and the original buffer whose contents we are shadowing or
426 * copying is moved to the transaction's shadow queue.
428 JBUFFER_TRACE(jh_in, "file as BJ_Shadow");
429 spin_lock(&journal->j_list_lock);
430 __jbd2_journal_file_buffer(jh_in, transaction, BJ_Shadow);
431 spin_unlock(&journal->j_list_lock);
432 jbd_unlock_bh_state(bh_in);
434 JBUFFER_TRACE(new_jh, "file as BJ_IO");
435 jbd2_journal_file_buffer(new_jh, transaction, BJ_IO);
437 return do_escape | (done_copy_out << 1);
441 * Allocation code for the journal file. Manage the space left in the
442 * journal, so that we can begin checkpointing when appropriate.
446 * __jbd2_log_space_left: Return the number of free blocks left in the journal.
448 * Called with the journal already locked.
450 * Called under j_state_lock
453 int __jbd2_log_space_left(journal_t *journal)
455 int left = journal->j_free;
457 /* assert_spin_locked(&journal->j_state_lock); */
460 * Be pessimistic here about the number of those free blocks which
461 * might be required for log descriptor control blocks.
464 #define MIN_LOG_RESERVED_BLOCKS 32 /* Allow for rounding errors */
466 left -= MIN_LOG_RESERVED_BLOCKS;
475 * Called under j_state_lock. Returns true if a transaction commit was started.
477 int __jbd2_log_start_commit(journal_t *journal, tid_t target)
480 * Are we already doing a recent enough commit?
482 if (!tid_geq(journal->j_commit_request, target)) {
484 * We want a new commit: OK, mark the request and wakeup the
485 * commit thread. We do _not_ do the commit ourselves.
488 journal->j_commit_request = target;
489 jbd_debug(1, "JBD: requesting commit %d/%d\n",
490 journal->j_commit_request,
491 journal->j_commit_sequence);
492 wake_up(&journal->j_wait_commit);
498 int jbd2_log_start_commit(journal_t *journal, tid_t tid)
502 write_lock(&journal->j_state_lock);
503 ret = __jbd2_log_start_commit(journal, tid);
504 write_unlock(&journal->j_state_lock);
509 * Force and wait upon a commit if the calling process is not within
510 * transaction. This is used for forcing out undo-protected data which contains
511 * bitmaps, when the fs is running out of space.
513 * We can only force the running transaction if we don't have an active handle;
514 * otherwise, we will deadlock.
516 * Returns true if a transaction was started.
518 int jbd2_journal_force_commit_nested(journal_t *journal)
520 transaction_t *transaction = NULL;
523 read_lock(&journal->j_state_lock);
524 if (journal->j_running_transaction && !current->journal_info) {
525 transaction = journal->j_running_transaction;
526 __jbd2_log_start_commit(journal, transaction->t_tid);
527 } else if (journal->j_committing_transaction)
528 transaction = journal->j_committing_transaction;
531 read_unlock(&journal->j_state_lock);
532 return 0; /* Nothing to retry */
535 tid = transaction->t_tid;
536 read_unlock(&journal->j_state_lock);
537 jbd2_log_wait_commit(journal, tid);
542 * Start a commit of the current running transaction (if any). Returns true
543 * if a transaction is going to be committed (or is currently already
544 * committing), and fills its tid in at *ptid
546 int jbd2_journal_start_commit(journal_t *journal, tid_t *ptid)
550 write_lock(&journal->j_state_lock);
551 if (journal->j_running_transaction) {
552 tid_t tid = journal->j_running_transaction->t_tid;
554 __jbd2_log_start_commit(journal, tid);
555 /* There's a running transaction and we've just made sure
556 * it's commit has been scheduled. */
560 } else if (journal->j_committing_transaction) {
562 * If ext3_write_super() recently started a commit, then we
563 * have to wait for completion of that transaction
566 *ptid = journal->j_committing_transaction->t_tid;
569 write_unlock(&journal->j_state_lock);
574 * Wait for a specified commit to complete.
575 * The caller may not hold the journal lock.
577 int jbd2_log_wait_commit(journal_t *journal, tid_t tid)
581 read_lock(&journal->j_state_lock);
582 #ifdef CONFIG_JBD2_DEBUG
583 if (!tid_geq(journal->j_commit_request, tid)) {
585 "%s: error: j_commit_request=%d, tid=%d\n",
586 __func__, journal->j_commit_request, tid);
589 while (tid_gt(tid, journal->j_commit_sequence)) {
590 jbd_debug(1, "JBD: want %d, j_commit_sequence=%d\n",
591 tid, journal->j_commit_sequence);
592 wake_up(&journal->j_wait_commit);
593 read_unlock(&journal->j_state_lock);
594 wait_event(journal->j_wait_done_commit,
595 !tid_gt(tid, journal->j_commit_sequence));
596 read_lock(&journal->j_state_lock);
598 read_unlock(&journal->j_state_lock);
600 if (unlikely(is_journal_aborted(journal))) {
601 printk(KERN_EMERG "journal commit I/O error\n");
608 * Log buffer allocation routines:
611 int jbd2_journal_next_log_block(journal_t *journal, unsigned long long *retp)
613 unsigned long blocknr;
615 write_lock(&journal->j_state_lock);
616 J_ASSERT(journal->j_free > 1);
618 blocknr = journal->j_head;
621 if (journal->j_head == journal->j_last)
622 journal->j_head = journal->j_first;
623 write_unlock(&journal->j_state_lock);
624 return jbd2_journal_bmap(journal, blocknr, retp);
628 * Conversion of logical to physical block numbers for the journal
630 * On external journals the journal blocks are identity-mapped, so
631 * this is a no-op. If needed, we can use j_blk_offset - everything is
634 int jbd2_journal_bmap(journal_t *journal, unsigned long blocknr,
635 unsigned long long *retp)
638 unsigned long long ret;
640 if (journal->j_inode) {
641 ret = bmap(journal->j_inode, blocknr);
645 printk(KERN_ALERT "%s: journal block not found "
646 "at offset %lu on %s\n",
647 __func__, blocknr, journal->j_devname);
649 __journal_abort_soft(journal, err);
652 *retp = blocknr; /* +journal->j_blk_offset */
658 * We play buffer_head aliasing tricks to write data/metadata blocks to
659 * the journal without copying their contents, but for journal
660 * descriptor blocks we do need to generate bona fide buffers.
662 * After the caller of jbd2_journal_get_descriptor_buffer() has finished modifying
663 * the buffer's contents they really should run flush_dcache_page(bh->b_page).
664 * But we don't bother doing that, so there will be coherency problems with
665 * mmaps of blockdevs which hold live JBD-controlled filesystems.
667 struct journal_head *jbd2_journal_get_descriptor_buffer(journal_t *journal)
669 struct buffer_head *bh;
670 unsigned long long blocknr;
673 err = jbd2_journal_next_log_block(journal, &blocknr);
678 bh = __getblk(journal->j_dev, blocknr, journal->j_blocksize);
682 memset(bh->b_data, 0, journal->j_blocksize);
683 set_buffer_uptodate(bh);
685 BUFFER_TRACE(bh, "return this buffer");
686 return jbd2_journal_add_journal_head(bh);
689 struct jbd2_stats_proc_session {
691 struct transaction_stats_s *stats;
696 static void *jbd2_seq_info_start(struct seq_file *seq, loff_t *pos)
698 return *pos ? NULL : SEQ_START_TOKEN;
701 static void *jbd2_seq_info_next(struct seq_file *seq, void *v, loff_t *pos)
706 static int jbd2_seq_info_show(struct seq_file *seq, void *v)
708 struct jbd2_stats_proc_session *s = seq->private;
710 if (v != SEQ_START_TOKEN)
712 seq_printf(seq, "%lu transaction, each up to %u blocks\n",
714 s->journal->j_max_transaction_buffers);
715 if (s->stats->ts_tid == 0)
717 seq_printf(seq, "average: \n %ums waiting for transaction\n",
718 jiffies_to_msecs(s->stats->run.rs_wait / s->stats->ts_tid));
719 seq_printf(seq, " %ums running transaction\n",
720 jiffies_to_msecs(s->stats->run.rs_running / s->stats->ts_tid));
721 seq_printf(seq, " %ums transaction was being locked\n",
722 jiffies_to_msecs(s->stats->run.rs_locked / s->stats->ts_tid));
723 seq_printf(seq, " %ums flushing data (in ordered mode)\n",
724 jiffies_to_msecs(s->stats->run.rs_flushing / s->stats->ts_tid));
725 seq_printf(seq, " %ums logging transaction\n",
726 jiffies_to_msecs(s->stats->run.rs_logging / s->stats->ts_tid));
727 seq_printf(seq, " %lluus average transaction commit time\n",
728 div_u64(s->journal->j_average_commit_time, 1000));
729 seq_printf(seq, " %lu handles per transaction\n",
730 s->stats->run.rs_handle_count / s->stats->ts_tid);
731 seq_printf(seq, " %lu blocks per transaction\n",
732 s->stats->run.rs_blocks / s->stats->ts_tid);
733 seq_printf(seq, " %lu logged blocks per transaction\n",
734 s->stats->run.rs_blocks_logged / s->stats->ts_tid);
738 static void jbd2_seq_info_stop(struct seq_file *seq, void *v)
742 static const struct seq_operations jbd2_seq_info_ops = {
743 .start = jbd2_seq_info_start,
744 .next = jbd2_seq_info_next,
745 .stop = jbd2_seq_info_stop,
746 .show = jbd2_seq_info_show,
749 static int jbd2_seq_info_open(struct inode *inode, struct file *file)
751 journal_t *journal = PDE(inode)->data;
752 struct jbd2_stats_proc_session *s;
755 s = kmalloc(sizeof(*s), GFP_KERNEL);
758 size = sizeof(struct transaction_stats_s);
759 s->stats = kmalloc(size, GFP_KERNEL);
760 if (s->stats == NULL) {
764 spin_lock(&journal->j_history_lock);
765 memcpy(s->stats, &journal->j_stats, size);
766 s->journal = journal;
767 spin_unlock(&journal->j_history_lock);
769 rc = seq_open(file, &jbd2_seq_info_ops);
771 struct seq_file *m = file->private_data;
781 static int jbd2_seq_info_release(struct inode *inode, struct file *file)
783 struct seq_file *seq = file->private_data;
784 struct jbd2_stats_proc_session *s = seq->private;
787 return seq_release(inode, file);
790 static const struct file_operations jbd2_seq_info_fops = {
791 .owner = THIS_MODULE,
792 .open = jbd2_seq_info_open,
795 .release = jbd2_seq_info_release,
798 static struct proc_dir_entry *proc_jbd2_stats;
800 static void jbd2_stats_proc_init(journal_t *journal)
802 journal->j_proc_entry = proc_mkdir(journal->j_devname, proc_jbd2_stats);
803 if (journal->j_proc_entry) {
804 proc_create_data("info", S_IRUGO, journal->j_proc_entry,
805 &jbd2_seq_info_fops, journal);
809 static void jbd2_stats_proc_exit(journal_t *journal)
811 remove_proc_entry("info", journal->j_proc_entry);
812 remove_proc_entry(journal->j_devname, proc_jbd2_stats);
816 * Management for journal control blocks: functions to create and
817 * destroy journal_t structures, and to initialise and read existing
818 * journal blocks from disk. */
820 /* First: create and setup a journal_t object in memory. We initialise
821 * very few fields yet: that has to wait until we have created the
822 * journal structures from from scratch, or loaded them from disk. */
824 static journal_t * journal_init_common (void)
829 journal = kzalloc(sizeof(*journal), GFP_KERNEL);
833 init_waitqueue_head(&journal->j_wait_transaction_locked);
834 init_waitqueue_head(&journal->j_wait_logspace);
835 init_waitqueue_head(&journal->j_wait_done_commit);
836 init_waitqueue_head(&journal->j_wait_checkpoint);
837 init_waitqueue_head(&journal->j_wait_commit);
838 init_waitqueue_head(&journal->j_wait_updates);
839 mutex_init(&journal->j_barrier);
840 mutex_init(&journal->j_checkpoint_mutex);
841 spin_lock_init(&journal->j_revoke_lock);
842 spin_lock_init(&journal->j_list_lock);
843 rwlock_init(&journal->j_state_lock);
845 journal->j_commit_interval = (HZ * JBD2_DEFAULT_MAX_COMMIT_AGE);
846 journal->j_min_batch_time = 0;
847 journal->j_max_batch_time = 15000; /* 15ms */
849 /* The journal is marked for error until we succeed with recovery! */
850 journal->j_flags = JBD2_ABORT;
852 /* Set up a default-sized revoke table for the new mount. */
853 err = jbd2_journal_init_revoke(journal, JOURNAL_REVOKE_DEFAULT_HASH);
859 spin_lock_init(&journal->j_history_lock);
864 /* jbd2_journal_init_dev and jbd2_journal_init_inode:
866 * Create a journal structure assigned some fixed set of disk blocks to
867 * the journal. We don't actually touch those disk blocks yet, but we
868 * need to set up all of the mapping information to tell the journaling
869 * system where the journal blocks are.
874 * journal_t * jbd2_journal_init_dev() - creates and initialises a journal structure
875 * @bdev: Block device on which to create the journal
876 * @fs_dev: Device which hold journalled filesystem for this journal.
877 * @start: Block nr Start of journal.
878 * @len: Length of the journal in blocks.
879 * @blocksize: blocksize of journalling device
881 * Returns: a newly created journal_t *
883 * jbd2_journal_init_dev creates a journal which maps a fixed contiguous
884 * range of blocks on an arbitrary block device.
887 journal_t * jbd2_journal_init_dev(struct block_device *bdev,
888 struct block_device *fs_dev,
889 unsigned long long start, int len, int blocksize)
891 journal_t *journal = journal_init_common();
892 struct buffer_head *bh;
899 /* journal descriptor can store up to n blocks -bzzz */
900 journal->j_blocksize = blocksize;
901 journal->j_dev = bdev;
902 journal->j_fs_dev = fs_dev;
903 journal->j_blk_offset = start;
904 journal->j_maxlen = len;
905 bdevname(journal->j_dev, journal->j_devname);
906 p = journal->j_devname;
907 while ((p = strchr(p, '/')))
909 jbd2_stats_proc_init(journal);
910 n = journal->j_blocksize / sizeof(journal_block_tag_t);
911 journal->j_wbufsize = n;
912 journal->j_wbuf = kmalloc(n * sizeof(struct buffer_head*), GFP_KERNEL);
913 if (!journal->j_wbuf) {
914 printk(KERN_ERR "%s: Cant allocate bhs for commit thread\n",
919 bh = __getblk(journal->j_dev, start, journal->j_blocksize);
922 "%s: Cannot get buffer for journal superblock\n",
926 journal->j_sb_buffer = bh;
927 journal->j_superblock = (journal_superblock_t *)bh->b_data;
931 kfree(journal->j_wbuf);
932 jbd2_stats_proc_exit(journal);
938 * journal_t * jbd2_journal_init_inode () - creates a journal which maps to a inode.
939 * @inode: An inode to create the journal in
941 * jbd2_journal_init_inode creates a journal which maps an on-disk inode as
942 * the journal. The inode must exist already, must support bmap() and
943 * must have all data blocks preallocated.
945 journal_t * jbd2_journal_init_inode (struct inode *inode)
947 struct buffer_head *bh;
948 journal_t *journal = journal_init_common();
952 unsigned long long blocknr;
957 journal->j_dev = journal->j_fs_dev = inode->i_sb->s_bdev;
958 journal->j_inode = inode;
959 bdevname(journal->j_dev, journal->j_devname);
960 p = journal->j_devname;
961 while ((p = strchr(p, '/')))
963 p = journal->j_devname + strlen(journal->j_devname);
964 sprintf(p, "-%lu", journal->j_inode->i_ino);
966 "journal %p: inode %s/%ld, size %Ld, bits %d, blksize %ld\n",
967 journal, inode->i_sb->s_id, inode->i_ino,
968 (long long) inode->i_size,
969 inode->i_sb->s_blocksize_bits, inode->i_sb->s_blocksize);
971 journal->j_maxlen = inode->i_size >> inode->i_sb->s_blocksize_bits;
972 journal->j_blocksize = inode->i_sb->s_blocksize;
973 jbd2_stats_proc_init(journal);
975 /* journal descriptor can store up to n blocks -bzzz */
976 n = journal->j_blocksize / sizeof(journal_block_tag_t);
977 journal->j_wbufsize = n;
978 journal->j_wbuf = kmalloc(n * sizeof(struct buffer_head*), GFP_KERNEL);
979 if (!journal->j_wbuf) {
980 printk(KERN_ERR "%s: Cant allocate bhs for commit thread\n",
985 err = jbd2_journal_bmap(journal, 0, &blocknr);
986 /* If that failed, give up */
988 printk(KERN_ERR "%s: Cannnot locate journal superblock\n",
993 bh = __getblk(journal->j_dev, blocknr, journal->j_blocksize);
996 "%s: Cannot get buffer for journal superblock\n",
1000 journal->j_sb_buffer = bh;
1001 journal->j_superblock = (journal_superblock_t *)bh->b_data;
1005 kfree(journal->j_wbuf);
1006 jbd2_stats_proc_exit(journal);
1012 * If the journal init or create aborts, we need to mark the journal
1013 * superblock as being NULL to prevent the journal destroy from writing
1014 * back a bogus superblock.
1016 static void journal_fail_superblock (journal_t *journal)
1018 struct buffer_head *bh = journal->j_sb_buffer;
1020 journal->j_sb_buffer = NULL;
1024 * Given a journal_t structure, initialise the various fields for
1025 * startup of a new journaling session. We use this both when creating
1026 * a journal, and after recovering an old journal to reset it for
1030 static int journal_reset(journal_t *journal)
1032 journal_superblock_t *sb = journal->j_superblock;
1033 unsigned long long first, last;
1035 first = be32_to_cpu(sb->s_first);
1036 last = be32_to_cpu(sb->s_maxlen);
1037 if (first + JBD2_MIN_JOURNAL_BLOCKS > last + 1) {
1038 printk(KERN_ERR "JBD: Journal too short (blocks %llu-%llu).\n",
1040 journal_fail_superblock(journal);
1044 journal->j_first = first;
1045 journal->j_last = last;
1047 journal->j_head = first;
1048 journal->j_tail = first;
1049 journal->j_free = last - first;
1051 journal->j_tail_sequence = journal->j_transaction_sequence;
1052 journal->j_commit_sequence = journal->j_transaction_sequence - 1;
1053 journal->j_commit_request = journal->j_commit_sequence;
1055 journal->j_max_transaction_buffers = journal->j_maxlen / 4;
1057 /* Add the dynamic fields and write it to disk. */
1058 jbd2_journal_update_superblock(journal, 1);
1059 return jbd2_journal_start_thread(journal);
1063 * void jbd2_journal_update_superblock() - Update journal sb on disk.
1064 * @journal: The journal to update.
1065 * @wait: Set to '0' if you don't want to wait for IO completion.
1067 * Update a journal's dynamic superblock fields and write it to disk,
1068 * optionally waiting for the IO to complete.
1070 void jbd2_journal_update_superblock(journal_t *journal, int wait)
1072 journal_superblock_t *sb = journal->j_superblock;
1073 struct buffer_head *bh = journal->j_sb_buffer;
1076 * As a special case, if the on-disk copy is already marked as needing
1077 * no recovery (s_start == 0) and there are no outstanding transactions
1078 * in the filesystem, then we can safely defer the superblock update
1079 * until the next commit by setting JBD2_FLUSHED. This avoids
1080 * attempting a write to a potential-readonly device.
1082 if (sb->s_start == 0 && journal->j_tail_sequence ==
1083 journal->j_transaction_sequence) {
1084 jbd_debug(1,"JBD: Skipping superblock update on recovered sb "
1085 "(start %ld, seq %d, errno %d)\n",
1086 journal->j_tail, journal->j_tail_sequence,
1091 if (buffer_write_io_error(bh)) {
1093 * Oh, dear. A previous attempt to write the journal
1094 * superblock failed. This could happen because the
1095 * USB device was yanked out. Or it could happen to
1096 * be a transient write error and maybe the block will
1097 * be remapped. Nothing we can do but to retry the
1098 * write and hope for the best.
1100 printk(KERN_ERR "JBD2: previous I/O error detected "
1101 "for journal superblock update for %s.\n",
1102 journal->j_devname);
1103 clear_buffer_write_io_error(bh);
1104 set_buffer_uptodate(bh);
1107 read_lock(&journal->j_state_lock);
1108 jbd_debug(1,"JBD: updating superblock (start %ld, seq %d, errno %d)\n",
1109 journal->j_tail, journal->j_tail_sequence, journal->j_errno);
1111 sb->s_sequence = cpu_to_be32(journal->j_tail_sequence);
1112 sb->s_start = cpu_to_be32(journal->j_tail);
1113 sb->s_errno = cpu_to_be32(journal->j_errno);
1114 read_unlock(&journal->j_state_lock);
1116 BUFFER_TRACE(bh, "marking dirty");
1117 mark_buffer_dirty(bh);
1119 sync_dirty_buffer(bh);
1120 if (buffer_write_io_error(bh)) {
1121 printk(KERN_ERR "JBD2: I/O error detected "
1122 "when updating journal superblock for %s.\n",
1123 journal->j_devname);
1124 clear_buffer_write_io_error(bh);
1125 set_buffer_uptodate(bh);
1128 write_dirty_buffer(bh, WRITE);
1131 /* If we have just flushed the log (by marking s_start==0), then
1132 * any future commit will have to be careful to update the
1133 * superblock again to re-record the true start of the log. */
1135 write_lock(&journal->j_state_lock);
1137 journal->j_flags &= ~JBD2_FLUSHED;
1139 journal->j_flags |= JBD2_FLUSHED;
1140 write_unlock(&journal->j_state_lock);
1144 * Read the superblock for a given journal, performing initial
1145 * validation of the format.
1148 static int journal_get_superblock(journal_t *journal)
1150 struct buffer_head *bh;
1151 journal_superblock_t *sb;
1154 bh = journal->j_sb_buffer;
1156 J_ASSERT(bh != NULL);
1157 if (!buffer_uptodate(bh)) {
1158 ll_rw_block(READ, 1, &bh);
1160 if (!buffer_uptodate(bh)) {
1162 "JBD: IO error reading journal superblock\n");
1167 sb = journal->j_superblock;
1171 if (sb->s_header.h_magic != cpu_to_be32(JBD2_MAGIC_NUMBER) ||
1172 sb->s_blocksize != cpu_to_be32(journal->j_blocksize)) {
1173 printk(KERN_WARNING "JBD: no valid journal superblock found\n");
1177 switch(be32_to_cpu(sb->s_header.h_blocktype)) {
1178 case JBD2_SUPERBLOCK_V1:
1179 journal->j_format_version = 1;
1181 case JBD2_SUPERBLOCK_V2:
1182 journal->j_format_version = 2;
1185 printk(KERN_WARNING "JBD: unrecognised superblock format ID\n");
1189 if (be32_to_cpu(sb->s_maxlen) < journal->j_maxlen)
1190 journal->j_maxlen = be32_to_cpu(sb->s_maxlen);
1191 else if (be32_to_cpu(sb->s_maxlen) > journal->j_maxlen) {
1192 printk (KERN_WARNING "JBD: journal file too short\n");
1199 journal_fail_superblock(journal);
1204 * Load the on-disk journal superblock and read the key fields into the
1208 static int load_superblock(journal_t *journal)
1211 journal_superblock_t *sb;
1213 err = journal_get_superblock(journal);
1217 sb = journal->j_superblock;
1219 journal->j_tail_sequence = be32_to_cpu(sb->s_sequence);
1220 journal->j_tail = be32_to_cpu(sb->s_start);
1221 journal->j_first = be32_to_cpu(sb->s_first);
1222 journal->j_last = be32_to_cpu(sb->s_maxlen);
1223 journal->j_errno = be32_to_cpu(sb->s_errno);
1230 * int jbd2_journal_load() - Read journal from disk.
1231 * @journal: Journal to act on.
1233 * Given a journal_t structure which tells us which disk blocks contain
1234 * a journal, read the journal from disk to initialise the in-memory
1237 int jbd2_journal_load(journal_t *journal)
1240 journal_superblock_t *sb;
1242 err = load_superblock(journal);
1246 sb = journal->j_superblock;
1247 /* If this is a V2 superblock, then we have to check the
1248 * features flags on it. */
1250 if (journal->j_format_version >= 2) {
1251 if ((sb->s_feature_ro_compat &
1252 ~cpu_to_be32(JBD2_KNOWN_ROCOMPAT_FEATURES)) ||
1253 (sb->s_feature_incompat &
1254 ~cpu_to_be32(JBD2_KNOWN_INCOMPAT_FEATURES))) {
1255 printk (KERN_WARNING
1256 "JBD: Unrecognised features on journal\n");
1262 * Create a slab for this blocksize
1264 err = jbd2_journal_create_slab(be32_to_cpu(sb->s_blocksize));
1268 /* Let the recovery code check whether it needs to recover any
1269 * data from the journal. */
1270 if (jbd2_journal_recover(journal))
1271 goto recovery_error;
1273 if (journal->j_failed_commit) {
1274 printk(KERN_ERR "JBD2: journal transaction %u on %s "
1275 "is corrupt.\n", journal->j_failed_commit,
1276 journal->j_devname);
1280 /* OK, we've finished with the dynamic journal bits:
1281 * reinitialise the dynamic contents of the superblock in memory
1282 * and reset them on disk. */
1283 if (journal_reset(journal))
1284 goto recovery_error;
1286 journal->j_flags &= ~JBD2_ABORT;
1287 journal->j_flags |= JBD2_LOADED;
1291 printk (KERN_WARNING "JBD: recovery failed\n");
1296 * void jbd2_journal_destroy() - Release a journal_t structure.
1297 * @journal: Journal to act on.
1299 * Release a journal_t structure once it is no longer in use by the
1301 * Return <0 if we couldn't clean up the journal.
1303 int jbd2_journal_destroy(journal_t *journal)
1307 /* Wait for the commit thread to wake up and die. */
1308 journal_kill_thread(journal);
1310 /* Force a final log commit */
1311 if (journal->j_running_transaction)
1312 jbd2_journal_commit_transaction(journal);
1314 /* Force any old transactions to disk */
1316 /* Totally anal locking here... */
1317 spin_lock(&journal->j_list_lock);
1318 while (journal->j_checkpoint_transactions != NULL) {
1319 spin_unlock(&journal->j_list_lock);
1320 mutex_lock(&journal->j_checkpoint_mutex);
1321 jbd2_log_do_checkpoint(journal);
1322 mutex_unlock(&journal->j_checkpoint_mutex);
1323 spin_lock(&journal->j_list_lock);
1326 J_ASSERT(journal->j_running_transaction == NULL);
1327 J_ASSERT(journal->j_committing_transaction == NULL);
1328 J_ASSERT(journal->j_checkpoint_transactions == NULL);
1329 spin_unlock(&journal->j_list_lock);
1331 if (journal->j_sb_buffer) {
1332 if (!is_journal_aborted(journal)) {
1333 /* We can now mark the journal as empty. */
1334 journal->j_tail = 0;
1335 journal->j_tail_sequence =
1336 ++journal->j_transaction_sequence;
1337 jbd2_journal_update_superblock(journal, 1);
1341 brelse(journal->j_sb_buffer);
1344 if (journal->j_proc_entry)
1345 jbd2_stats_proc_exit(journal);
1346 if (journal->j_inode)
1347 iput(journal->j_inode);
1348 if (journal->j_revoke)
1349 jbd2_journal_destroy_revoke(journal);
1350 kfree(journal->j_wbuf);
1358 *int jbd2_journal_check_used_features () - Check if features specified are used.
1359 * @journal: Journal to check.
1360 * @compat: bitmask of compatible features
1361 * @ro: bitmask of features that force read-only mount
1362 * @incompat: bitmask of incompatible features
1364 * Check whether the journal uses all of a given set of
1365 * features. Return true (non-zero) if it does.
1368 int jbd2_journal_check_used_features (journal_t *journal, unsigned long compat,
1369 unsigned long ro, unsigned long incompat)
1371 journal_superblock_t *sb;
1373 if (!compat && !ro && !incompat)
1375 /* Load journal superblock if it is not loaded yet. */
1376 if (journal->j_format_version == 0 &&
1377 journal_get_superblock(journal) != 0)
1379 if (journal->j_format_version == 1)
1382 sb = journal->j_superblock;
1384 if (((be32_to_cpu(sb->s_feature_compat) & compat) == compat) &&
1385 ((be32_to_cpu(sb->s_feature_ro_compat) & ro) == ro) &&
1386 ((be32_to_cpu(sb->s_feature_incompat) & incompat) == incompat))
1393 * int jbd2_journal_check_available_features() - Check feature set in journalling layer
1394 * @journal: Journal to check.
1395 * @compat: bitmask of compatible features
1396 * @ro: bitmask of features that force read-only mount
1397 * @incompat: bitmask of incompatible features
1399 * Check whether the journaling code supports the use of
1400 * all of a given set of features on this journal. Return true
1401 * (non-zero) if it can. */
1403 int jbd2_journal_check_available_features (journal_t *journal, unsigned long compat,
1404 unsigned long ro, unsigned long incompat)
1406 if (!compat && !ro && !incompat)
1409 /* We can support any known requested features iff the
1410 * superblock is in version 2. Otherwise we fail to support any
1411 * extended sb features. */
1413 if (journal->j_format_version != 2)
1416 if ((compat & JBD2_KNOWN_COMPAT_FEATURES) == compat &&
1417 (ro & JBD2_KNOWN_ROCOMPAT_FEATURES) == ro &&
1418 (incompat & JBD2_KNOWN_INCOMPAT_FEATURES) == incompat)
1425 * int jbd2_journal_set_features () - Mark a given journal feature in the superblock
1426 * @journal: Journal to act on.
1427 * @compat: bitmask of compatible features
1428 * @ro: bitmask of features that force read-only mount
1429 * @incompat: bitmask of incompatible features
1431 * Mark a given journal feature as present on the
1432 * superblock. Returns true if the requested features could be set.
1436 int jbd2_journal_set_features (journal_t *journal, unsigned long compat,
1437 unsigned long ro, unsigned long incompat)
1439 journal_superblock_t *sb;
1441 if (jbd2_journal_check_used_features(journal, compat, ro, incompat))
1444 if (!jbd2_journal_check_available_features(journal, compat, ro, incompat))
1447 jbd_debug(1, "Setting new features 0x%lx/0x%lx/0x%lx\n",
1448 compat, ro, incompat);
1450 sb = journal->j_superblock;
1452 sb->s_feature_compat |= cpu_to_be32(compat);
1453 sb->s_feature_ro_compat |= cpu_to_be32(ro);
1454 sb->s_feature_incompat |= cpu_to_be32(incompat);
1460 * jbd2_journal_clear_features () - Clear a given journal feature in the
1462 * @journal: Journal to act on.
1463 * @compat: bitmask of compatible features
1464 * @ro: bitmask of features that force read-only mount
1465 * @incompat: bitmask of incompatible features
1467 * Clear a given journal feature as present on the
1470 void jbd2_journal_clear_features(journal_t *journal, unsigned long compat,
1471 unsigned long ro, unsigned long incompat)
1473 journal_superblock_t *sb;
1475 jbd_debug(1, "Clear features 0x%lx/0x%lx/0x%lx\n",
1476 compat, ro, incompat);
1478 sb = journal->j_superblock;
1480 sb->s_feature_compat &= ~cpu_to_be32(compat);
1481 sb->s_feature_ro_compat &= ~cpu_to_be32(ro);
1482 sb->s_feature_incompat &= ~cpu_to_be32(incompat);
1484 EXPORT_SYMBOL(jbd2_journal_clear_features);
1487 * int jbd2_journal_update_format () - Update on-disk journal structure.
1488 * @journal: Journal to act on.
1490 * Given an initialised but unloaded journal struct, poke about in the
1491 * on-disk structure to update it to the most recent supported version.
1493 int jbd2_journal_update_format (journal_t *journal)
1495 journal_superblock_t *sb;
1498 err = journal_get_superblock(journal);
1502 sb = journal->j_superblock;
1504 switch (be32_to_cpu(sb->s_header.h_blocktype)) {
1505 case JBD2_SUPERBLOCK_V2:
1507 case JBD2_SUPERBLOCK_V1:
1508 return journal_convert_superblock_v1(journal, sb);
1515 static int journal_convert_superblock_v1(journal_t *journal,
1516 journal_superblock_t *sb)
1518 int offset, blocksize;
1519 struct buffer_head *bh;
1522 "JBD: Converting superblock from version 1 to 2.\n");
1524 /* Pre-initialise new fields to zero */
1525 offset = ((char *) &(sb->s_feature_compat)) - ((char *) sb);
1526 blocksize = be32_to_cpu(sb->s_blocksize);
1527 memset(&sb->s_feature_compat, 0, blocksize-offset);
1529 sb->s_nr_users = cpu_to_be32(1);
1530 sb->s_header.h_blocktype = cpu_to_be32(JBD2_SUPERBLOCK_V2);
1531 journal->j_format_version = 2;
1533 bh = journal->j_sb_buffer;
1534 BUFFER_TRACE(bh, "marking dirty");
1535 mark_buffer_dirty(bh);
1536 sync_dirty_buffer(bh);
1542 * int jbd2_journal_flush () - Flush journal
1543 * @journal: Journal to act on.
1545 * Flush all data for a given journal to disk and empty the journal.
1546 * Filesystems can use this when remounting readonly to ensure that
1547 * recovery does not need to happen on remount.
1550 int jbd2_journal_flush(journal_t *journal)
1553 transaction_t *transaction = NULL;
1554 unsigned long old_tail;
1556 write_lock(&journal->j_state_lock);
1558 /* Force everything buffered to the log... */
1559 if (journal->j_running_transaction) {
1560 transaction = journal->j_running_transaction;
1561 __jbd2_log_start_commit(journal, transaction->t_tid);
1562 } else if (journal->j_committing_transaction)
1563 transaction = journal->j_committing_transaction;
1565 /* Wait for the log commit to complete... */
1567 tid_t tid = transaction->t_tid;
1569 write_unlock(&journal->j_state_lock);
1570 jbd2_log_wait_commit(journal, tid);
1572 write_unlock(&journal->j_state_lock);
1575 /* ...and flush everything in the log out to disk. */
1576 spin_lock(&journal->j_list_lock);
1577 while (!err && journal->j_checkpoint_transactions != NULL) {
1578 spin_unlock(&journal->j_list_lock);
1579 mutex_lock(&journal->j_checkpoint_mutex);
1580 err = jbd2_log_do_checkpoint(journal);
1581 mutex_unlock(&journal->j_checkpoint_mutex);
1582 spin_lock(&journal->j_list_lock);
1584 spin_unlock(&journal->j_list_lock);
1586 if (is_journal_aborted(journal))
1589 jbd2_cleanup_journal_tail(journal);
1591 /* Finally, mark the journal as really needing no recovery.
1592 * This sets s_start==0 in the underlying superblock, which is
1593 * the magic code for a fully-recovered superblock. Any future
1594 * commits of data to the journal will restore the current
1596 write_lock(&journal->j_state_lock);
1597 old_tail = journal->j_tail;
1598 journal->j_tail = 0;
1599 write_unlock(&journal->j_state_lock);
1600 jbd2_journal_update_superblock(journal, 1);
1601 write_lock(&journal->j_state_lock);
1602 journal->j_tail = old_tail;
1604 J_ASSERT(!journal->j_running_transaction);
1605 J_ASSERT(!journal->j_committing_transaction);
1606 J_ASSERT(!journal->j_checkpoint_transactions);
1607 J_ASSERT(journal->j_head == journal->j_tail);
1608 J_ASSERT(journal->j_tail_sequence == journal->j_transaction_sequence);
1609 write_unlock(&journal->j_state_lock);
1614 * int jbd2_journal_wipe() - Wipe journal contents
1615 * @journal: Journal to act on.
1616 * @write: flag (see below)
1618 * Wipe out all of the contents of a journal, safely. This will produce
1619 * a warning if the journal contains any valid recovery information.
1620 * Must be called between journal_init_*() and jbd2_journal_load().
1622 * If 'write' is non-zero, then we wipe out the journal on disk; otherwise
1623 * we merely suppress recovery.
1626 int jbd2_journal_wipe(journal_t *journal, int write)
1630 J_ASSERT (!(journal->j_flags & JBD2_LOADED));
1632 err = load_superblock(journal);
1636 if (!journal->j_tail)
1639 printk (KERN_WARNING "JBD: %s recovery information on journal\n",
1640 write ? "Clearing" : "Ignoring");
1642 err = jbd2_journal_skip_recovery(journal);
1644 jbd2_journal_update_superblock(journal, 1);
1651 * Journal abort has very specific semantics, which we describe
1652 * for journal abort.
1654 * Two internal functions, which provide abort to the jbd layer
1659 * Quick version for internal journal use (doesn't lock the journal).
1660 * Aborts hard --- we mark the abort as occurred, but do _nothing_ else,
1661 * and don't attempt to make any other journal updates.
1663 void __jbd2_journal_abort_hard(journal_t *journal)
1665 transaction_t *transaction;
1667 if (journal->j_flags & JBD2_ABORT)
1670 printk(KERN_ERR "Aborting journal on device %s.\n",
1671 journal->j_devname);
1673 write_lock(&journal->j_state_lock);
1674 journal->j_flags |= JBD2_ABORT;
1675 transaction = journal->j_running_transaction;
1677 __jbd2_log_start_commit(journal, transaction->t_tid);
1678 write_unlock(&journal->j_state_lock);
1681 /* Soft abort: record the abort error status in the journal superblock,
1682 * but don't do any other IO. */
1683 static void __journal_abort_soft (journal_t *journal, int errno)
1685 if (journal->j_flags & JBD2_ABORT)
1688 if (!journal->j_errno)
1689 journal->j_errno = errno;
1691 __jbd2_journal_abort_hard(journal);
1694 jbd2_journal_update_superblock(journal, 1);
1698 * void jbd2_journal_abort () - Shutdown the journal immediately.
1699 * @journal: the journal to shutdown.
1700 * @errno: an error number to record in the journal indicating
1701 * the reason for the shutdown.
1703 * Perform a complete, immediate shutdown of the ENTIRE
1704 * journal (not of a single transaction). This operation cannot be
1705 * undone without closing and reopening the journal.
1707 * The jbd2_journal_abort function is intended to support higher level error
1708 * recovery mechanisms such as the ext2/ext3 remount-readonly error
1711 * Journal abort has very specific semantics. Any existing dirty,
1712 * unjournaled buffers in the main filesystem will still be written to
1713 * disk by bdflush, but the journaling mechanism will be suspended
1714 * immediately and no further transaction commits will be honoured.
1716 * Any dirty, journaled buffers will be written back to disk without
1717 * hitting the journal. Atomicity cannot be guaranteed on an aborted
1718 * filesystem, but we _do_ attempt to leave as much data as possible
1719 * behind for fsck to use for cleanup.
1721 * Any attempt to get a new transaction handle on a journal which is in
1722 * ABORT state will just result in an -EROFS error return. A
1723 * jbd2_journal_stop on an existing handle will return -EIO if we have
1724 * entered abort state during the update.
1726 * Recursive transactions are not disturbed by journal abort until the
1727 * final jbd2_journal_stop, which will receive the -EIO error.
1729 * Finally, the jbd2_journal_abort call allows the caller to supply an errno
1730 * which will be recorded (if possible) in the journal superblock. This
1731 * allows a client to record failure conditions in the middle of a
1732 * transaction without having to complete the transaction to record the
1733 * failure to disk. ext3_error, for example, now uses this
1736 * Errors which originate from within the journaling layer will NOT
1737 * supply an errno; a null errno implies that absolutely no further
1738 * writes are done to the journal (unless there are any already in
1743 void jbd2_journal_abort(journal_t *journal, int errno)
1745 __journal_abort_soft(journal, errno);
1749 * int jbd2_journal_errno () - returns the journal's error state.
1750 * @journal: journal to examine.
1752 * This is the errno number set with jbd2_journal_abort(), the last
1753 * time the journal was mounted - if the journal was stopped
1754 * without calling abort this will be 0.
1756 * If the journal has been aborted on this mount time -EROFS will
1759 int jbd2_journal_errno(journal_t *journal)
1763 read_lock(&journal->j_state_lock);
1764 if (journal->j_flags & JBD2_ABORT)
1767 err = journal->j_errno;
1768 read_unlock(&journal->j_state_lock);
1773 * int jbd2_journal_clear_err () - clears the journal's error state
1774 * @journal: journal to act on.
1776 * An error must be cleared or acked to take a FS out of readonly
1779 int jbd2_journal_clear_err(journal_t *journal)
1783 write_lock(&journal->j_state_lock);
1784 if (journal->j_flags & JBD2_ABORT)
1787 journal->j_errno = 0;
1788 write_unlock(&journal->j_state_lock);
1793 * void jbd2_journal_ack_err() - Ack journal err.
1794 * @journal: journal to act on.
1796 * An error must be cleared or acked to take a FS out of readonly
1799 void jbd2_journal_ack_err(journal_t *journal)
1801 write_lock(&journal->j_state_lock);
1802 if (journal->j_errno)
1803 journal->j_flags |= JBD2_ACK_ERR;
1804 write_unlock(&journal->j_state_lock);
1807 int jbd2_journal_blocks_per_page(struct inode *inode)
1809 return 1 << (PAGE_CACHE_SHIFT - inode->i_sb->s_blocksize_bits);
1813 * helper functions to deal with 32 or 64bit block numbers.
1815 size_t journal_tag_bytes(journal_t *journal)
1817 if (JBD2_HAS_INCOMPAT_FEATURE(journal, JBD2_FEATURE_INCOMPAT_64BIT))
1818 return JBD2_TAG_SIZE64;
1820 return JBD2_TAG_SIZE32;
1824 * JBD memory management
1826 * These functions are used to allocate block-sized chunks of memory
1827 * used for making copies of buffer_head data. Very often it will be
1828 * page-sized chunks of data, but sometimes it will be in
1829 * sub-page-size chunks. (For example, 16k pages on Power systems
1830 * with a 4k block file system.) For blocks smaller than a page, we
1831 * use a SLAB allocator. There are slab caches for each block size,
1832 * which are allocated at mount time, if necessary, and we only free
1833 * (all of) the slab caches when/if the jbd2 module is unloaded. For
1834 * this reason we don't need to a mutex to protect access to
1835 * jbd2_slab[] allocating or releasing memory; only in
1836 * jbd2_journal_create_slab().
1838 #define JBD2_MAX_SLABS 8
1839 static struct kmem_cache *jbd2_slab[JBD2_MAX_SLABS];
1841 static const char *jbd2_slab_names[JBD2_MAX_SLABS] = {
1842 "jbd2_1k", "jbd2_2k", "jbd2_4k", "jbd2_8k",
1843 "jbd2_16k", "jbd2_32k", "jbd2_64k", "jbd2_128k"
1847 static void jbd2_journal_destroy_slabs(void)
1851 for (i = 0; i < JBD2_MAX_SLABS; i++) {
1853 kmem_cache_destroy(jbd2_slab[i]);
1854 jbd2_slab[i] = NULL;
1858 static int jbd2_journal_create_slab(size_t size)
1860 static DEFINE_MUTEX(jbd2_slab_create_mutex);
1861 int i = order_base_2(size) - 10;
1864 if (size == PAGE_SIZE)
1867 if (i >= JBD2_MAX_SLABS)
1870 if (unlikely(i < 0))
1872 mutex_lock(&jbd2_slab_create_mutex);
1874 mutex_unlock(&jbd2_slab_create_mutex);
1875 return 0; /* Already created */
1878 slab_size = 1 << (i+10);
1879 jbd2_slab[i] = kmem_cache_create(jbd2_slab_names[i], slab_size,
1880 slab_size, 0, NULL);
1881 mutex_unlock(&jbd2_slab_create_mutex);
1882 if (!jbd2_slab[i]) {
1883 printk(KERN_EMERG "JBD2: no memory for jbd2_slab cache\n");
1889 static struct kmem_cache *get_slab(size_t size)
1891 int i = order_base_2(size) - 10;
1893 BUG_ON(i >= JBD2_MAX_SLABS);
1894 if (unlikely(i < 0))
1896 BUG_ON(jbd2_slab[i] == NULL);
1897 return jbd2_slab[i];
1900 void *jbd2_alloc(size_t size, gfp_t flags)
1904 BUG_ON(size & (size-1)); /* Must be a power of 2 */
1906 flags |= __GFP_REPEAT;
1907 if (size == PAGE_SIZE)
1908 ptr = (void *)__get_free_pages(flags, 0);
1909 else if (size > PAGE_SIZE) {
1910 int order = get_order(size);
1913 ptr = (void *)__get_free_pages(flags, order);
1915 ptr = vmalloc(size);
1917 ptr = kmem_cache_alloc(get_slab(size), flags);
1919 /* Check alignment; SLUB has gotten this wrong in the past,
1920 * and this can lead to user data corruption! */
1921 BUG_ON(((unsigned long) ptr) & (size-1));
1926 void jbd2_free(void *ptr, size_t size)
1928 if (size == PAGE_SIZE) {
1929 free_pages((unsigned long)ptr, 0);
1932 if (size > PAGE_SIZE) {
1933 int order = get_order(size);
1936 free_pages((unsigned long)ptr, order);
1941 kmem_cache_free(get_slab(size), ptr);
1945 * Journal_head storage management
1947 static struct kmem_cache *jbd2_journal_head_cache;
1948 #ifdef CONFIG_JBD2_DEBUG
1949 static atomic_t nr_journal_heads = ATOMIC_INIT(0);
1952 static int journal_init_jbd2_journal_head_cache(void)
1956 J_ASSERT(jbd2_journal_head_cache == NULL);
1957 jbd2_journal_head_cache = kmem_cache_create("jbd2_journal_head",
1958 sizeof(struct journal_head),
1960 SLAB_TEMPORARY, /* flags */
1963 if (!jbd2_journal_head_cache) {
1965 printk(KERN_EMERG "JBD: no memory for journal_head cache\n");
1970 static void jbd2_journal_destroy_jbd2_journal_head_cache(void)
1972 if (jbd2_journal_head_cache) {
1973 kmem_cache_destroy(jbd2_journal_head_cache);
1974 jbd2_journal_head_cache = NULL;
1979 * journal_head splicing and dicing
1981 static struct journal_head *journal_alloc_journal_head(void)
1983 struct journal_head *ret;
1985 #ifdef CONFIG_JBD2_DEBUG
1986 atomic_inc(&nr_journal_heads);
1988 ret = kmem_cache_alloc(jbd2_journal_head_cache, GFP_NOFS);
1990 jbd_debug(1, "out of memory for journal_head\n");
1991 pr_notice_ratelimited("ENOMEM in %s, retrying.\n", __func__);
1994 ret = kmem_cache_alloc(jbd2_journal_head_cache, GFP_NOFS);
2000 static void journal_free_journal_head(struct journal_head *jh)
2002 #ifdef CONFIG_JBD2_DEBUG
2003 atomic_dec(&nr_journal_heads);
2004 memset(jh, JBD2_POISON_FREE, sizeof(*jh));
2006 kmem_cache_free(jbd2_journal_head_cache, jh);
2010 * A journal_head is attached to a buffer_head whenever JBD has an
2011 * interest in the buffer.
2013 * Whenever a buffer has an attached journal_head, its ->b_state:BH_JBD bit
2014 * is set. This bit is tested in core kernel code where we need to take
2015 * JBD-specific actions. Testing the zeroness of ->b_private is not reliable
2018 * When a buffer has its BH_JBD bit set, its ->b_count is elevated by one.
2020 * When a buffer has its BH_JBD bit set it is immune from being released by
2021 * core kernel code, mainly via ->b_count.
2023 * A journal_head may be detached from its buffer_head when the journal_head's
2024 * b_transaction, b_cp_transaction and b_next_transaction pointers are NULL.
2025 * Various places in JBD call jbd2_journal_remove_journal_head() to indicate that the
2026 * journal_head can be dropped if needed.
2028 * Various places in the kernel want to attach a journal_head to a buffer_head
2029 * _before_ attaching the journal_head to a transaction. To protect the
2030 * journal_head in this situation, jbd2_journal_add_journal_head elevates the
2031 * journal_head's b_jcount refcount by one. The caller must call
2032 * jbd2_journal_put_journal_head() to undo this.
2034 * So the typical usage would be:
2036 * (Attach a journal_head if needed. Increments b_jcount)
2037 * struct journal_head *jh = jbd2_journal_add_journal_head(bh);
2039 * jh->b_transaction = xxx;
2040 * jbd2_journal_put_journal_head(jh);
2042 * Now, the journal_head's b_jcount is zero, but it is safe from being released
2043 * because it has a non-zero b_transaction.
2047 * Give a buffer_head a journal_head.
2049 * Doesn't need the journal lock.
2052 struct journal_head *jbd2_journal_add_journal_head(struct buffer_head *bh)
2054 struct journal_head *jh;
2055 struct journal_head *new_jh = NULL;
2058 if (!buffer_jbd(bh)) {
2059 new_jh = journal_alloc_journal_head();
2060 memset(new_jh, 0, sizeof(*new_jh));
2063 jbd_lock_bh_journal_head(bh);
2064 if (buffer_jbd(bh)) {
2068 (atomic_read(&bh->b_count) > 0) ||
2069 (bh->b_page && bh->b_page->mapping));
2072 jbd_unlock_bh_journal_head(bh);
2077 new_jh = NULL; /* We consumed it */
2082 BUFFER_TRACE(bh, "added journal_head");
2085 jbd_unlock_bh_journal_head(bh);
2087 journal_free_journal_head(new_jh);
2088 return bh->b_private;
2092 * Grab a ref against this buffer_head's journal_head. If it ended up not
2093 * having a journal_head, return NULL
2095 struct journal_head *jbd2_journal_grab_journal_head(struct buffer_head *bh)
2097 struct journal_head *jh = NULL;
2099 jbd_lock_bh_journal_head(bh);
2100 if (buffer_jbd(bh)) {
2104 jbd_unlock_bh_journal_head(bh);
2108 static void __journal_remove_journal_head(struct buffer_head *bh)
2110 struct journal_head *jh = bh2jh(bh);
2112 J_ASSERT_JH(jh, jh->b_jcount >= 0);
2115 if (jh->b_jcount == 0) {
2116 if (jh->b_transaction == NULL &&
2117 jh->b_next_transaction == NULL &&
2118 jh->b_cp_transaction == NULL) {
2119 J_ASSERT_JH(jh, jh->b_jlist == BJ_None);
2120 J_ASSERT_BH(bh, buffer_jbd(bh));
2121 J_ASSERT_BH(bh, jh2bh(jh) == bh);
2122 BUFFER_TRACE(bh, "remove journal_head");
2123 if (jh->b_frozen_data) {
2124 printk(KERN_WARNING "%s: freeing "
2127 jbd2_free(jh->b_frozen_data, bh->b_size);
2129 if (jh->b_committed_data) {
2130 printk(KERN_WARNING "%s: freeing "
2131 "b_committed_data\n",
2133 jbd2_free(jh->b_committed_data, bh->b_size);
2135 bh->b_private = NULL;
2136 jh->b_bh = NULL; /* debug, really */
2137 clear_buffer_jbd(bh);
2139 journal_free_journal_head(jh);
2141 BUFFER_TRACE(bh, "journal_head was locked");
2147 * jbd2_journal_remove_journal_head(): if the buffer isn't attached to a transaction
2148 * and has a zero b_jcount then remove and release its journal_head. If we did
2149 * see that the buffer is not used by any transaction we also "logically"
2150 * decrement ->b_count.
2152 * We in fact take an additional increment on ->b_count as a convenience,
2153 * because the caller usually wants to do additional things with the bh
2154 * after calling here.
2155 * The caller of jbd2_journal_remove_journal_head() *must* run __brelse(bh) at some
2156 * time. Once the caller has run __brelse(), the buffer is eligible for
2157 * reaping by try_to_free_buffers().
2159 void jbd2_journal_remove_journal_head(struct buffer_head *bh)
2161 jbd_lock_bh_journal_head(bh);
2162 __journal_remove_journal_head(bh);
2163 jbd_unlock_bh_journal_head(bh);
2167 * Drop a reference on the passed journal_head. If it fell to zero then try to
2168 * release the journal_head from the buffer_head.
2170 void jbd2_journal_put_journal_head(struct journal_head *jh)
2172 struct buffer_head *bh = jh2bh(jh);
2174 jbd_lock_bh_journal_head(bh);
2175 J_ASSERT_JH(jh, jh->b_jcount > 0);
2177 if (!jh->b_jcount && !jh->b_transaction) {
2178 __journal_remove_journal_head(bh);
2181 jbd_unlock_bh_journal_head(bh);
2185 * Initialize jbd inode head
2187 void jbd2_journal_init_jbd_inode(struct jbd2_inode *jinode, struct inode *inode)
2189 jinode->i_transaction = NULL;
2190 jinode->i_next_transaction = NULL;
2191 jinode->i_vfs_inode = inode;
2192 jinode->i_flags = 0;
2193 INIT_LIST_HEAD(&jinode->i_list);
2197 * Function to be called before we start removing inode from memory (i.e.,
2198 * clear_inode() is a fine place to be called from). It removes inode from
2199 * transaction's lists.
2201 void jbd2_journal_release_jbd_inode(journal_t *journal,
2202 struct jbd2_inode *jinode)
2207 spin_lock(&journal->j_list_lock);
2208 /* Is commit writing out inode - we have to wait */
2209 if (test_bit(__JI_COMMIT_RUNNING, &jinode->i_flags)) {
2210 wait_queue_head_t *wq;
2211 DEFINE_WAIT_BIT(wait, &jinode->i_flags, __JI_COMMIT_RUNNING);
2212 wq = bit_waitqueue(&jinode->i_flags, __JI_COMMIT_RUNNING);
2213 prepare_to_wait(wq, &wait.wait, TASK_UNINTERRUPTIBLE);
2214 spin_unlock(&journal->j_list_lock);
2216 finish_wait(wq, &wait.wait);
2220 if (jinode->i_transaction) {
2221 list_del(&jinode->i_list);
2222 jinode->i_transaction = NULL;
2224 spin_unlock(&journal->j_list_lock);
2230 #ifdef CONFIG_JBD2_DEBUG
2231 u8 jbd2_journal_enable_debug __read_mostly;
2232 EXPORT_SYMBOL(jbd2_journal_enable_debug);
2234 #define JBD2_DEBUG_NAME "jbd2-debug"
2236 static struct dentry *jbd2_debugfs_dir;
2237 static struct dentry *jbd2_debug;
2239 static void __init jbd2_create_debugfs_entry(void)
2241 jbd2_debugfs_dir = debugfs_create_dir("jbd2", NULL);
2242 if (jbd2_debugfs_dir)
2243 jbd2_debug = debugfs_create_u8(JBD2_DEBUG_NAME,
2246 &jbd2_journal_enable_debug);
2249 static void __exit jbd2_remove_debugfs_entry(void)
2251 debugfs_remove(jbd2_debug);
2252 debugfs_remove(jbd2_debugfs_dir);
2257 static void __init jbd2_create_debugfs_entry(void)
2261 static void __exit jbd2_remove_debugfs_entry(void)
2267 #ifdef CONFIG_PROC_FS
2269 #define JBD2_STATS_PROC_NAME "fs/jbd2"
2271 static void __init jbd2_create_jbd_stats_proc_entry(void)
2273 proc_jbd2_stats = proc_mkdir(JBD2_STATS_PROC_NAME, NULL);
2276 static void __exit jbd2_remove_jbd_stats_proc_entry(void)
2278 if (proc_jbd2_stats)
2279 remove_proc_entry(JBD2_STATS_PROC_NAME, NULL);
2284 #define jbd2_create_jbd_stats_proc_entry() do {} while (0)
2285 #define jbd2_remove_jbd_stats_proc_entry() do {} while (0)
2289 struct kmem_cache *jbd2_handle_cache;
2291 static int __init journal_init_handle_cache(void)
2293 jbd2_handle_cache = kmem_cache_create("jbd2_journal_handle",
2296 SLAB_TEMPORARY, /* flags */
2298 if (jbd2_handle_cache == NULL) {
2299 printk(KERN_EMERG "JBD: failed to create handle cache\n");
2305 static void jbd2_journal_destroy_handle_cache(void)
2307 if (jbd2_handle_cache)
2308 kmem_cache_destroy(jbd2_handle_cache);
2312 * Module startup and shutdown
2315 static int __init journal_init_caches(void)
2319 ret = jbd2_journal_init_revoke_caches();
2321 ret = journal_init_jbd2_journal_head_cache();
2323 ret = journal_init_handle_cache();
2327 static void jbd2_journal_destroy_caches(void)
2329 jbd2_journal_destroy_revoke_caches();
2330 jbd2_journal_destroy_jbd2_journal_head_cache();
2331 jbd2_journal_destroy_handle_cache();
2332 jbd2_journal_destroy_slabs();
2335 static int __init journal_init(void)
2339 BUILD_BUG_ON(sizeof(struct journal_superblock_s) != 1024);
2341 ret = journal_init_caches();
2343 jbd2_create_debugfs_entry();
2344 jbd2_create_jbd_stats_proc_entry();
2346 jbd2_journal_destroy_caches();
2351 static void __exit journal_exit(void)
2353 #ifdef CONFIG_JBD2_DEBUG
2354 int n = atomic_read(&nr_journal_heads);
2356 printk(KERN_EMERG "JBD: leaked %d journal_heads!\n", n);
2358 jbd2_remove_debugfs_entry();
2359 jbd2_remove_jbd_stats_proc_entry();
2360 jbd2_journal_destroy_caches();
2364 * jbd2_dev_to_name is a utility function used by the jbd2 and ext4
2365 * tracing infrastructure to map a dev_t to a device name.
2367 * The caller should use rcu_read_lock() in order to make sure the
2368 * device name stays valid until its done with it. We use
2369 * rcu_read_lock() as well to make sure we're safe in case the caller
2370 * gets sloppy, and because rcu_read_lock() is cheap and can be safely
2373 struct devname_cache {
2374 struct rcu_head rcu;
2376 char devname[BDEVNAME_SIZE];
2378 #define CACHE_SIZE_BITS 6
2379 static struct devname_cache *devcache[1 << CACHE_SIZE_BITS];
2380 static DEFINE_SPINLOCK(devname_cache_lock);
2382 static void free_devcache(struct rcu_head *rcu)
2387 const char *jbd2_dev_to_name(dev_t device)
2389 int i = hash_32(device, CACHE_SIZE_BITS);
2391 struct block_device *bd;
2392 static struct devname_cache *new_dev;
2395 if (devcache[i] && devcache[i]->device == device) {
2396 ret = devcache[i]->devname;
2402 new_dev = kmalloc(sizeof(struct devname_cache), GFP_KERNEL);
2404 return "NODEV-ALLOCFAILURE"; /* Something non-NULL */
2405 spin_lock(&devname_cache_lock);
2407 if (devcache[i]->device == device) {
2409 ret = devcache[i]->devname;
2410 spin_unlock(&devname_cache_lock);
2413 call_rcu(&devcache[i]->rcu, free_devcache);
2415 devcache[i] = new_dev;
2416 devcache[i]->device = device;
2419 bdevname(bd, devcache[i]->devname);
2422 __bdevname(device, devcache[i]->devname);
2423 ret = devcache[i]->devname;
2424 spin_unlock(&devname_cache_lock);
2427 EXPORT_SYMBOL(jbd2_dev_to_name);
2429 MODULE_LICENSE("GPL");
2430 module_init(journal_init);
2431 module_exit(journal_exit);