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>
47 #define CREATE_TRACE_POINTS
48 #include <trace/events/jbd2.h>
50 #include <asm/uaccess.h>
52 #include <asm/system.h>
54 EXPORT_SYMBOL(jbd2_journal_extend);
55 EXPORT_SYMBOL(jbd2_journal_stop);
56 EXPORT_SYMBOL(jbd2_journal_lock_updates);
57 EXPORT_SYMBOL(jbd2_journal_unlock_updates);
58 EXPORT_SYMBOL(jbd2_journal_get_write_access);
59 EXPORT_SYMBOL(jbd2_journal_get_create_access);
60 EXPORT_SYMBOL(jbd2_journal_get_undo_access);
61 EXPORT_SYMBOL(jbd2_journal_set_triggers);
62 EXPORT_SYMBOL(jbd2_journal_dirty_metadata);
63 EXPORT_SYMBOL(jbd2_journal_release_buffer);
64 EXPORT_SYMBOL(jbd2_journal_forget);
66 EXPORT_SYMBOL(journal_sync_buffer);
68 EXPORT_SYMBOL(jbd2_journal_flush);
69 EXPORT_SYMBOL(jbd2_journal_revoke);
71 EXPORT_SYMBOL(jbd2_journal_init_dev);
72 EXPORT_SYMBOL(jbd2_journal_init_inode);
73 EXPORT_SYMBOL(jbd2_journal_update_format);
74 EXPORT_SYMBOL(jbd2_journal_check_used_features);
75 EXPORT_SYMBOL(jbd2_journal_check_available_features);
76 EXPORT_SYMBOL(jbd2_journal_set_features);
77 EXPORT_SYMBOL(jbd2_journal_load);
78 EXPORT_SYMBOL(jbd2_journal_destroy);
79 EXPORT_SYMBOL(jbd2_journal_abort);
80 EXPORT_SYMBOL(jbd2_journal_errno);
81 EXPORT_SYMBOL(jbd2_journal_ack_err);
82 EXPORT_SYMBOL(jbd2_journal_clear_err);
83 EXPORT_SYMBOL(jbd2_log_wait_commit);
84 EXPORT_SYMBOL(jbd2_log_start_commit);
85 EXPORT_SYMBOL(jbd2_journal_start_commit);
86 EXPORT_SYMBOL(jbd2_journal_force_commit_nested);
87 EXPORT_SYMBOL(jbd2_journal_wipe);
88 EXPORT_SYMBOL(jbd2_journal_blocks_per_page);
89 EXPORT_SYMBOL(jbd2_journal_invalidatepage);
90 EXPORT_SYMBOL(jbd2_journal_try_to_free_buffers);
91 EXPORT_SYMBOL(jbd2_journal_force_commit);
92 EXPORT_SYMBOL(jbd2_journal_file_inode);
93 EXPORT_SYMBOL(jbd2_journal_init_jbd_inode);
94 EXPORT_SYMBOL(jbd2_journal_release_jbd_inode);
95 EXPORT_SYMBOL(jbd2_journal_begin_ordered_truncate);
97 static int journal_convert_superblock_v1(journal_t *, journal_superblock_t *);
98 static void __journal_abort_soft (journal_t *journal, int errno);
99 static int jbd2_journal_create_slab(size_t slab_size);
102 * Helper function used to manage commit timeouts
105 static void commit_timeout(unsigned long __data)
107 struct task_struct * p = (struct task_struct *) __data;
113 * kjournald2: The main thread function used to manage a logging device
116 * This kernel thread is responsible for two things:
118 * 1) COMMIT: Every so often we need to commit the current state of the
119 * filesystem to disk. The journal thread is responsible for writing
120 * all of the metadata buffers to disk.
122 * 2) CHECKPOINT: We cannot reuse a used section of the log file until all
123 * of the data in that part of the log has been rewritten elsewhere on
124 * the disk. Flushing these old buffers to reclaim space in the log is
125 * known as checkpointing, and this thread is responsible for that job.
128 static int kjournald2(void *arg)
130 journal_t *journal = arg;
131 transaction_t *transaction;
134 * Set up an interval timer which can be used to trigger a commit wakeup
135 * after the commit interval expires
137 setup_timer(&journal->j_commit_timer, commit_timeout,
138 (unsigned long)current);
140 /* Record that the journal thread is running */
141 journal->j_task = current;
142 wake_up(&journal->j_wait_done_commit);
145 * And now, wait forever for commit wakeup events.
147 write_lock(&journal->j_state_lock);
150 if (journal->j_flags & JBD2_UNMOUNT)
153 jbd_debug(1, "commit_sequence=%d, commit_request=%d\n",
154 journal->j_commit_sequence, journal->j_commit_request);
156 if (journal->j_commit_sequence != journal->j_commit_request) {
157 jbd_debug(1, "OK, requests differ\n");
158 write_unlock(&journal->j_state_lock);
159 del_timer_sync(&journal->j_commit_timer);
160 jbd2_journal_commit_transaction(journal);
161 write_lock(&journal->j_state_lock);
165 wake_up(&journal->j_wait_done_commit);
166 if (freezing(current)) {
168 * The simpler the better. Flushing journal isn't a
169 * good idea, because that depends on threads that may
170 * be already stopped.
172 jbd_debug(1, "Now suspending kjournald2\n");
173 write_unlock(&journal->j_state_lock);
175 write_lock(&journal->j_state_lock);
178 * We assume on resume that commits are already there,
182 int should_sleep = 1;
184 prepare_to_wait(&journal->j_wait_commit, &wait,
186 if (journal->j_commit_sequence != journal->j_commit_request)
188 transaction = journal->j_running_transaction;
189 if (transaction && time_after_eq(jiffies,
190 transaction->t_expires))
192 if (journal->j_flags & JBD2_UNMOUNT)
195 write_unlock(&journal->j_state_lock);
197 write_lock(&journal->j_state_lock);
199 finish_wait(&journal->j_wait_commit, &wait);
202 jbd_debug(1, "kjournald2 wakes\n");
205 * Were we woken up by a commit wakeup event?
207 transaction = journal->j_running_transaction;
208 if (transaction && time_after_eq(jiffies, transaction->t_expires)) {
209 journal->j_commit_request = transaction->t_tid;
210 jbd_debug(1, "woke because of timeout\n");
215 write_unlock(&journal->j_state_lock);
216 del_timer_sync(&journal->j_commit_timer);
217 journal->j_task = NULL;
218 wake_up(&journal->j_wait_done_commit);
219 jbd_debug(1, "Journal thread exiting.\n");
223 static int jbd2_journal_start_thread(journal_t *journal)
225 struct task_struct *t;
227 t = kthread_run(kjournald2, journal, "jbd2/%s",
232 wait_event(journal->j_wait_done_commit, journal->j_task != NULL);
236 static void journal_kill_thread(journal_t *journal)
238 write_lock(&journal->j_state_lock);
239 journal->j_flags |= JBD2_UNMOUNT;
241 while (journal->j_task) {
242 wake_up(&journal->j_wait_commit);
243 write_unlock(&journal->j_state_lock);
244 wait_event(journal->j_wait_done_commit, journal->j_task == NULL);
245 write_lock(&journal->j_state_lock);
247 write_unlock(&journal->j_state_lock);
251 * jbd2_journal_write_metadata_buffer: write a metadata buffer to the journal.
253 * Writes a metadata buffer to a given disk block. The actual IO is not
254 * performed but a new buffer_head is constructed which labels the data
255 * to be written with the correct destination disk block.
257 * Any magic-number escaping which needs to be done will cause a
258 * copy-out here. If the buffer happens to start with the
259 * JBD2_MAGIC_NUMBER, then we can't write it to the log directly: the
260 * magic number is only written to the log for descripter blocks. In
261 * this case, we copy the data and replace the first word with 0, and we
262 * return a result code which indicates that this buffer needs to be
263 * marked as an escaped buffer in the corresponding log descriptor
264 * block. The missing word can then be restored when the block is read
267 * If the source buffer has already been modified by a new transaction
268 * since we took the last commit snapshot, we use the frozen copy of
269 * that data for IO. If we end up using the existing buffer_head's data
270 * for the write, then we *have* to lock the buffer to prevent anyone
271 * else from using and possibly modifying it while the IO is in
274 * The function returns a pointer to the buffer_heads to be used for IO.
276 * We assume that the journal has already been locked in this function.
283 * Bit 0 set == escape performed on the data
284 * Bit 1 set == buffer copy-out performed (kfree the data after IO)
287 int jbd2_journal_write_metadata_buffer(transaction_t *transaction,
288 struct journal_head *jh_in,
289 struct journal_head **jh_out,
290 unsigned long long blocknr)
292 int need_copy_out = 0;
293 int done_copy_out = 0;
296 struct buffer_head *new_bh;
297 struct journal_head *new_jh;
298 struct page *new_page;
299 unsigned int new_offset;
300 struct buffer_head *bh_in = jh2bh(jh_in);
301 journal_t *journal = transaction->t_journal;
304 * The buffer really shouldn't be locked: only the current committing
305 * transaction is allowed to write it, so nobody else is allowed
308 * akpm: except if we're journalling data, and write() output is
309 * also part of a shared mapping, and another thread has
310 * decided to launch a writepage() against this buffer.
312 J_ASSERT_BH(bh_in, buffer_jbddirty(bh_in));
315 new_bh = alloc_buffer_head(GFP_NOFS);
318 * Failure is not an option, but __GFP_NOFAIL is going
319 * away; so we retry ourselves here.
321 congestion_wait(BLK_RW_ASYNC, HZ/50);
325 /* keep subsequent assertions sane */
327 init_buffer(new_bh, NULL, NULL);
328 atomic_set(&new_bh->b_count, 1);
329 new_jh = jbd2_journal_add_journal_head(new_bh); /* This sleeps */
332 * If a new transaction has already done a buffer copy-out, then
333 * we use that version of the data for the commit.
335 jbd_lock_bh_state(bh_in);
337 if (jh_in->b_frozen_data) {
339 new_page = virt_to_page(jh_in->b_frozen_data);
340 new_offset = offset_in_page(jh_in->b_frozen_data);
342 new_page = jh2bh(jh_in)->b_page;
343 new_offset = offset_in_page(jh2bh(jh_in)->b_data);
346 mapped_data = kmap_atomic(new_page, KM_USER0);
348 * Fire data frozen trigger if data already wasn't frozen. Do this
349 * before checking for escaping, as the trigger may modify the magic
350 * offset. If a copy-out happens afterwards, it will have the correct
351 * data in the buffer.
354 jbd2_buffer_frozen_trigger(jh_in, mapped_data + new_offset,
360 if (*((__be32 *)(mapped_data + new_offset)) ==
361 cpu_to_be32(JBD2_MAGIC_NUMBER)) {
365 kunmap_atomic(mapped_data, KM_USER0);
368 * Do we need to do a data copy?
370 if (need_copy_out && !done_copy_out) {
373 jbd_unlock_bh_state(bh_in);
374 tmp = jbd2_alloc(bh_in->b_size, GFP_NOFS);
376 jbd2_journal_put_journal_head(new_jh);
379 jbd_lock_bh_state(bh_in);
380 if (jh_in->b_frozen_data) {
381 jbd2_free(tmp, bh_in->b_size);
385 jh_in->b_frozen_data = tmp;
386 mapped_data = kmap_atomic(new_page, KM_USER0);
387 memcpy(tmp, mapped_data + new_offset, jh2bh(jh_in)->b_size);
388 kunmap_atomic(mapped_data, KM_USER0);
390 new_page = virt_to_page(tmp);
391 new_offset = offset_in_page(tmp);
395 * This isn't strictly necessary, as we're using frozen
396 * data for the escaping, but it keeps consistency with
397 * b_frozen_data usage.
399 jh_in->b_frozen_triggers = jh_in->b_triggers;
403 * Did we need to do an escaping? Now we've done all the
404 * copying, we can finally do so.
407 mapped_data = kmap_atomic(new_page, KM_USER0);
408 *((unsigned int *)(mapped_data + new_offset)) = 0;
409 kunmap_atomic(mapped_data, KM_USER0);
412 set_bh_page(new_bh, new_page, new_offset);
413 new_jh->b_transaction = NULL;
414 new_bh->b_size = jh2bh(jh_in)->b_size;
415 new_bh->b_bdev = transaction->t_journal->j_dev;
416 new_bh->b_blocknr = blocknr;
417 set_buffer_mapped(new_bh);
418 set_buffer_dirty(new_bh);
423 * The to-be-written buffer needs to get moved to the io queue,
424 * and the original buffer whose contents we are shadowing or
425 * copying is moved to the transaction's shadow queue.
427 JBUFFER_TRACE(jh_in, "file as BJ_Shadow");
428 spin_lock(&journal->j_list_lock);
429 __jbd2_journal_file_buffer(jh_in, transaction, BJ_Shadow);
430 spin_unlock(&journal->j_list_lock);
431 jbd_unlock_bh_state(bh_in);
433 JBUFFER_TRACE(new_jh, "file as BJ_IO");
434 jbd2_journal_file_buffer(new_jh, transaction, BJ_IO);
436 return do_escape | (done_copy_out << 1);
440 * Allocation code for the journal file. Manage the space left in the
441 * journal, so that we can begin checkpointing when appropriate.
445 * __jbd2_log_space_left: Return the number of free blocks left in the journal.
447 * Called with the journal already locked.
449 * Called under j_state_lock
452 int __jbd2_log_space_left(journal_t *journal)
454 int left = journal->j_free;
456 /* assert_spin_locked(&journal->j_state_lock); */
459 * Be pessimistic here about the number of those free blocks which
460 * might be required for log descriptor control blocks.
463 #define MIN_LOG_RESERVED_BLOCKS 32 /* Allow for rounding errors */
465 left -= MIN_LOG_RESERVED_BLOCKS;
474 * Called under j_state_lock. Returns true if a transaction commit was started.
476 int __jbd2_log_start_commit(journal_t *journal, tid_t target)
479 * Are we already doing a recent enough commit?
481 if (!tid_geq(journal->j_commit_request, target)) {
483 * We want a new commit: OK, mark the request and wakup the
484 * commit thread. We do _not_ do the commit ourselves.
487 journal->j_commit_request = target;
488 jbd_debug(1, "JBD: requesting commit %d/%d\n",
489 journal->j_commit_request,
490 journal->j_commit_sequence);
491 wake_up(&journal->j_wait_commit);
497 int jbd2_log_start_commit(journal_t *journal, tid_t tid)
501 write_lock(&journal->j_state_lock);
502 ret = __jbd2_log_start_commit(journal, tid);
503 write_unlock(&journal->j_state_lock);
508 * Force and wait upon a commit if the calling process is not within
509 * transaction. This is used for forcing out undo-protected data which contains
510 * bitmaps, when the fs is running out of space.
512 * We can only force the running transaction if we don't have an active handle;
513 * otherwise, we will deadlock.
515 * Returns true if a transaction was started.
517 int jbd2_journal_force_commit_nested(journal_t *journal)
519 transaction_t *transaction = NULL;
522 read_lock(&journal->j_state_lock);
523 if (journal->j_running_transaction && !current->journal_info) {
524 transaction = journal->j_running_transaction;
525 __jbd2_log_start_commit(journal, transaction->t_tid);
526 } else if (journal->j_committing_transaction)
527 transaction = journal->j_committing_transaction;
530 read_unlock(&journal->j_state_lock);
531 return 0; /* Nothing to retry */
534 tid = transaction->t_tid;
535 read_unlock(&journal->j_state_lock);
536 jbd2_log_wait_commit(journal, tid);
541 * Start a commit of the current running transaction (if any). Returns true
542 * if a transaction is going to be committed (or is currently already
543 * committing), and fills its tid in at *ptid
545 int jbd2_journal_start_commit(journal_t *journal, tid_t *ptid)
549 write_lock(&journal->j_state_lock);
550 if (journal->j_running_transaction) {
551 tid_t tid = journal->j_running_transaction->t_tid;
553 __jbd2_log_start_commit(journal, tid);
554 /* There's a running transaction and we've just made sure
555 * it's commit has been scheduled. */
559 } else if (journal->j_committing_transaction) {
561 * If ext3_write_super() recently started a commit, then we
562 * have to wait for completion of that transaction
565 *ptid = journal->j_committing_transaction->t_tid;
568 write_unlock(&journal->j_state_lock);
573 * Wait for a specified commit to complete.
574 * The caller may not hold the journal lock.
576 int jbd2_log_wait_commit(journal_t *journal, tid_t tid)
580 read_lock(&journal->j_state_lock);
581 #ifdef CONFIG_JBD2_DEBUG
582 if (!tid_geq(journal->j_commit_request, tid)) {
584 "%s: error: j_commit_request=%d, tid=%d\n",
585 __func__, journal->j_commit_request, tid);
588 while (tid_gt(tid, journal->j_commit_sequence)) {
589 jbd_debug(1, "JBD: want %d, j_commit_sequence=%d\n",
590 tid, journal->j_commit_sequence);
591 wake_up(&journal->j_wait_commit);
592 read_unlock(&journal->j_state_lock);
593 wait_event(journal->j_wait_done_commit,
594 !tid_gt(tid, journal->j_commit_sequence));
595 read_lock(&journal->j_state_lock);
597 read_unlock(&journal->j_state_lock);
599 if (unlikely(is_journal_aborted(journal))) {
600 printk(KERN_EMERG "journal commit I/O error\n");
607 * Log buffer allocation routines:
610 int jbd2_journal_next_log_block(journal_t *journal, unsigned long long *retp)
612 unsigned long blocknr;
614 write_lock(&journal->j_state_lock);
615 J_ASSERT(journal->j_free > 1);
617 blocknr = journal->j_head;
620 if (journal->j_head == journal->j_last)
621 journal->j_head = journal->j_first;
622 write_unlock(&journal->j_state_lock);
623 return jbd2_journal_bmap(journal, blocknr, retp);
627 * Conversion of logical to physical block numbers for the journal
629 * On external journals the journal blocks are identity-mapped, so
630 * this is a no-op. If needed, we can use j_blk_offset - everything is
633 int jbd2_journal_bmap(journal_t *journal, unsigned long blocknr,
634 unsigned long long *retp)
637 unsigned long long ret;
639 if (journal->j_inode) {
640 ret = bmap(journal->j_inode, blocknr);
644 printk(KERN_ALERT "%s: journal block not found "
645 "at offset %lu on %s\n",
646 __func__, blocknr, journal->j_devname);
648 __journal_abort_soft(journal, err);
651 *retp = blocknr; /* +journal->j_blk_offset */
657 * We play buffer_head aliasing tricks to write data/metadata blocks to
658 * the journal without copying their contents, but for journal
659 * descriptor blocks we do need to generate bona fide buffers.
661 * After the caller of jbd2_journal_get_descriptor_buffer() has finished modifying
662 * the buffer's contents they really should run flush_dcache_page(bh->b_page).
663 * But we don't bother doing that, so there will be coherency problems with
664 * mmaps of blockdevs which hold live JBD-controlled filesystems.
666 struct journal_head *jbd2_journal_get_descriptor_buffer(journal_t *journal)
668 struct buffer_head *bh;
669 unsigned long long blocknr;
672 err = jbd2_journal_next_log_block(journal, &blocknr);
677 bh = __getblk(journal->j_dev, blocknr, journal->j_blocksize);
681 memset(bh->b_data, 0, journal->j_blocksize);
682 set_buffer_uptodate(bh);
684 BUFFER_TRACE(bh, "return this buffer");
685 return jbd2_journal_add_journal_head(bh);
688 struct jbd2_stats_proc_session {
690 struct transaction_stats_s *stats;
695 static void *jbd2_seq_info_start(struct seq_file *seq, loff_t *pos)
697 return *pos ? NULL : SEQ_START_TOKEN;
700 static void *jbd2_seq_info_next(struct seq_file *seq, void *v, loff_t *pos)
705 static int jbd2_seq_info_show(struct seq_file *seq, void *v)
707 struct jbd2_stats_proc_session *s = seq->private;
709 if (v != SEQ_START_TOKEN)
711 seq_printf(seq, "%lu transaction, each up to %u blocks\n",
713 s->journal->j_max_transaction_buffers);
714 if (s->stats->ts_tid == 0)
716 seq_printf(seq, "average: \n %ums waiting for transaction\n",
717 jiffies_to_msecs(s->stats->run.rs_wait / s->stats->ts_tid));
718 seq_printf(seq, " %ums running transaction\n",
719 jiffies_to_msecs(s->stats->run.rs_running / s->stats->ts_tid));
720 seq_printf(seq, " %ums transaction was being locked\n",
721 jiffies_to_msecs(s->stats->run.rs_locked / s->stats->ts_tid));
722 seq_printf(seq, " %ums flushing data (in ordered mode)\n",
723 jiffies_to_msecs(s->stats->run.rs_flushing / s->stats->ts_tid));
724 seq_printf(seq, " %ums logging transaction\n",
725 jiffies_to_msecs(s->stats->run.rs_logging / s->stats->ts_tid));
726 seq_printf(seq, " %lluus average transaction commit time\n",
727 div_u64(s->journal->j_average_commit_time, 1000));
728 seq_printf(seq, " %lu handles per transaction\n",
729 s->stats->run.rs_handle_count / s->stats->ts_tid);
730 seq_printf(seq, " %lu blocks per transaction\n",
731 s->stats->run.rs_blocks / s->stats->ts_tid);
732 seq_printf(seq, " %lu logged blocks per transaction\n",
733 s->stats->run.rs_blocks_logged / s->stats->ts_tid);
737 static void jbd2_seq_info_stop(struct seq_file *seq, void *v)
741 static const struct seq_operations jbd2_seq_info_ops = {
742 .start = jbd2_seq_info_start,
743 .next = jbd2_seq_info_next,
744 .stop = jbd2_seq_info_stop,
745 .show = jbd2_seq_info_show,
748 static int jbd2_seq_info_open(struct inode *inode, struct file *file)
750 journal_t *journal = PDE(inode)->data;
751 struct jbd2_stats_proc_session *s;
754 s = kmalloc(sizeof(*s), GFP_KERNEL);
757 size = sizeof(struct transaction_stats_s);
758 s->stats = kmalloc(size, GFP_KERNEL);
759 if (s->stats == NULL) {
763 spin_lock(&journal->j_history_lock);
764 memcpy(s->stats, &journal->j_stats, size);
765 s->journal = journal;
766 spin_unlock(&journal->j_history_lock);
768 rc = seq_open(file, &jbd2_seq_info_ops);
770 struct seq_file *m = file->private_data;
780 static int jbd2_seq_info_release(struct inode *inode, struct file *file)
782 struct seq_file *seq = file->private_data;
783 struct jbd2_stats_proc_session *s = seq->private;
786 return seq_release(inode, file);
789 static const struct file_operations jbd2_seq_info_fops = {
790 .owner = THIS_MODULE,
791 .open = jbd2_seq_info_open,
794 .release = jbd2_seq_info_release,
797 static struct proc_dir_entry *proc_jbd2_stats;
799 static void jbd2_stats_proc_init(journal_t *journal)
801 journal->j_proc_entry = proc_mkdir(journal->j_devname, proc_jbd2_stats);
802 if (journal->j_proc_entry) {
803 proc_create_data("info", S_IRUGO, journal->j_proc_entry,
804 &jbd2_seq_info_fops, journal);
808 static void jbd2_stats_proc_exit(journal_t *journal)
810 remove_proc_entry("info", journal->j_proc_entry);
811 remove_proc_entry(journal->j_devname, proc_jbd2_stats);
815 * Management for journal control blocks: functions to create and
816 * destroy journal_t structures, and to initialise and read existing
817 * journal blocks from disk. */
819 /* First: create and setup a journal_t object in memory. We initialise
820 * very few fields yet: that has to wait until we have created the
821 * journal structures from from scratch, or loaded them from disk. */
823 static journal_t * journal_init_common (void)
828 journal = kzalloc(sizeof(*journal), GFP_KERNEL);
832 init_waitqueue_head(&journal->j_wait_transaction_locked);
833 init_waitqueue_head(&journal->j_wait_logspace);
834 init_waitqueue_head(&journal->j_wait_done_commit);
835 init_waitqueue_head(&journal->j_wait_checkpoint);
836 init_waitqueue_head(&journal->j_wait_commit);
837 init_waitqueue_head(&journal->j_wait_updates);
838 mutex_init(&journal->j_barrier);
839 mutex_init(&journal->j_checkpoint_mutex);
840 spin_lock_init(&journal->j_revoke_lock);
841 spin_lock_init(&journal->j_list_lock);
842 rwlock_init(&journal->j_state_lock);
844 journal->j_commit_interval = (HZ * JBD2_DEFAULT_MAX_COMMIT_AGE);
845 journal->j_min_batch_time = 0;
846 journal->j_max_batch_time = 15000; /* 15ms */
848 /* The journal is marked for error until we succeed with recovery! */
849 journal->j_flags = JBD2_ABORT;
851 /* Set up a default-sized revoke table for the new mount. */
852 err = jbd2_journal_init_revoke(journal, JOURNAL_REVOKE_DEFAULT_HASH);
858 spin_lock_init(&journal->j_history_lock);
865 /* jbd2_journal_init_dev and jbd2_journal_init_inode:
867 * Create a journal structure assigned some fixed set of disk blocks to
868 * the journal. We don't actually touch those disk blocks yet, but we
869 * need to set up all of the mapping information to tell the journaling
870 * system where the journal blocks are.
875 * journal_t * jbd2_journal_init_dev() - creates and initialises a journal structure
876 * @bdev: Block device on which to create the journal
877 * @fs_dev: Device which hold journalled filesystem for this journal.
878 * @start: Block nr Start of journal.
879 * @len: Length of the journal in blocks.
880 * @blocksize: blocksize of journalling device
882 * Returns: a newly created journal_t *
884 * jbd2_journal_init_dev creates a journal which maps a fixed contiguous
885 * range of blocks on an arbitrary block device.
888 journal_t * jbd2_journal_init_dev(struct block_device *bdev,
889 struct block_device *fs_dev,
890 unsigned long long start, int len, int blocksize)
892 journal_t *journal = journal_init_common();
893 struct buffer_head *bh;
900 /* journal descriptor can store up to n blocks -bzzz */
901 journal->j_blocksize = blocksize;
902 jbd2_stats_proc_init(journal);
903 n = journal->j_blocksize / sizeof(journal_block_tag_t);
904 journal->j_wbufsize = n;
905 journal->j_wbuf = kmalloc(n * sizeof(struct buffer_head*), GFP_KERNEL);
906 if (!journal->j_wbuf) {
907 printk(KERN_ERR "%s: Cant allocate bhs for commit thread\n",
911 journal->j_dev = bdev;
912 journal->j_fs_dev = fs_dev;
913 journal->j_blk_offset = start;
914 journal->j_maxlen = len;
915 bdevname(journal->j_dev, journal->j_devname);
916 p = journal->j_devname;
917 while ((p = strchr(p, '/')))
920 bh = __getblk(journal->j_dev, start, journal->j_blocksize);
923 "%s: Cannot get buffer for journal superblock\n",
927 journal->j_sb_buffer = bh;
928 journal->j_superblock = (journal_superblock_t *)bh->b_data;
932 kfree(journal->j_wbuf);
933 jbd2_stats_proc_exit(journal);
939 * journal_t * jbd2_journal_init_inode () - creates a journal which maps to a inode.
940 * @inode: An inode to create the journal in
942 * jbd2_journal_init_inode creates a journal which maps an on-disk inode as
943 * the journal. The inode must exist already, must support bmap() and
944 * must have all data blocks preallocated.
946 journal_t * jbd2_journal_init_inode (struct inode *inode)
948 struct buffer_head *bh;
949 journal_t *journal = journal_init_common();
953 unsigned long long blocknr;
958 journal->j_dev = journal->j_fs_dev = inode->i_sb->s_bdev;
959 journal->j_inode = inode;
960 bdevname(journal->j_dev, journal->j_devname);
961 p = journal->j_devname;
962 while ((p = strchr(p, '/')))
964 p = journal->j_devname + strlen(journal->j_devname);
965 sprintf(p, "-%lu", journal->j_inode->i_ino);
967 "journal %p: inode %s/%ld, size %Ld, bits %d, blksize %ld\n",
968 journal, inode->i_sb->s_id, inode->i_ino,
969 (long long) inode->i_size,
970 inode->i_sb->s_blocksize_bits, inode->i_sb->s_blocksize);
972 journal->j_maxlen = inode->i_size >> inode->i_sb->s_blocksize_bits;
973 journal->j_blocksize = inode->i_sb->s_blocksize;
974 jbd2_stats_proc_init(journal);
976 /* journal descriptor can store up to n blocks -bzzz */
977 n = journal->j_blocksize / sizeof(journal_block_tag_t);
978 journal->j_wbufsize = n;
979 journal->j_wbuf = kmalloc(n * sizeof(struct buffer_head*), GFP_KERNEL);
980 if (!journal->j_wbuf) {
981 printk(KERN_ERR "%s: Cant allocate bhs for commit thread\n",
986 err = jbd2_journal_bmap(journal, 0, &blocknr);
987 /* If that failed, give up */
989 printk(KERN_ERR "%s: Cannnot locate journal superblock\n",
994 bh = __getblk(journal->j_dev, blocknr, journal->j_blocksize);
997 "%s: Cannot get buffer for journal superblock\n",
1001 journal->j_sb_buffer = bh;
1002 journal->j_superblock = (journal_superblock_t *)bh->b_data;
1006 kfree(journal->j_wbuf);
1007 jbd2_stats_proc_exit(journal);
1013 * If the journal init or create aborts, we need to mark the journal
1014 * superblock as being NULL to prevent the journal destroy from writing
1015 * back a bogus superblock.
1017 static void journal_fail_superblock (journal_t *journal)
1019 struct buffer_head *bh = journal->j_sb_buffer;
1021 journal->j_sb_buffer = NULL;
1025 * Given a journal_t structure, initialise the various fields for
1026 * startup of a new journaling session. We use this both when creating
1027 * a journal, and after recovering an old journal to reset it for
1031 static int journal_reset(journal_t *journal)
1033 journal_superblock_t *sb = journal->j_superblock;
1034 unsigned long long first, last;
1036 first = be32_to_cpu(sb->s_first);
1037 last = be32_to_cpu(sb->s_maxlen);
1038 if (first + JBD2_MIN_JOURNAL_BLOCKS > last + 1) {
1039 printk(KERN_ERR "JBD: Journal too short (blocks %llu-%llu).\n",
1041 journal_fail_superblock(journal);
1045 journal->j_first = first;
1046 journal->j_last = last;
1048 journal->j_head = first;
1049 journal->j_tail = first;
1050 journal->j_free = last - first;
1052 journal->j_tail_sequence = journal->j_transaction_sequence;
1053 journal->j_commit_sequence = journal->j_transaction_sequence - 1;
1054 journal->j_commit_request = journal->j_commit_sequence;
1056 journal->j_max_transaction_buffers = journal->j_maxlen / 4;
1058 /* Add the dynamic fields and write it to disk. */
1059 jbd2_journal_update_superblock(journal, 1);
1060 return jbd2_journal_start_thread(journal);
1064 * void jbd2_journal_update_superblock() - Update journal sb on disk.
1065 * @journal: The journal to update.
1066 * @wait: Set to '0' if you don't want to wait for IO completion.
1068 * Update a journal's dynamic superblock fields and write it to disk,
1069 * optionally waiting for the IO to complete.
1071 void jbd2_journal_update_superblock(journal_t *journal, int wait)
1073 journal_superblock_t *sb = journal->j_superblock;
1074 struct buffer_head *bh = journal->j_sb_buffer;
1077 * As a special case, if the on-disk copy is already marked as needing
1078 * no recovery (s_start == 0) and there are no outstanding transactions
1079 * in the filesystem, then we can safely defer the superblock update
1080 * until the next commit by setting JBD2_FLUSHED. This avoids
1081 * attempting a write to a potential-readonly device.
1083 if (sb->s_start == 0 && journal->j_tail_sequence ==
1084 journal->j_transaction_sequence) {
1085 jbd_debug(1,"JBD: Skipping superblock update on recovered sb "
1086 "(start %ld, seq %d, errno %d)\n",
1087 journal->j_tail, journal->j_tail_sequence,
1092 if (buffer_write_io_error(bh)) {
1094 * Oh, dear. A previous attempt to write the journal
1095 * superblock failed. This could happen because the
1096 * USB device was yanked out. Or it could happen to
1097 * be a transient write error and maybe the block will
1098 * be remapped. Nothing we can do but to retry the
1099 * write and hope for the best.
1101 printk(KERN_ERR "JBD2: previous I/O error detected "
1102 "for journal superblock update for %s.\n",
1103 journal->j_devname);
1104 clear_buffer_write_io_error(bh);
1105 set_buffer_uptodate(bh);
1108 read_lock(&journal->j_state_lock);
1109 jbd_debug(1,"JBD: updating superblock (start %ld, seq %d, errno %d)\n",
1110 journal->j_tail, journal->j_tail_sequence, journal->j_errno);
1112 sb->s_sequence = cpu_to_be32(journal->j_tail_sequence);
1113 sb->s_start = cpu_to_be32(journal->j_tail);
1114 sb->s_errno = cpu_to_be32(journal->j_errno);
1115 read_unlock(&journal->j_state_lock);
1117 BUFFER_TRACE(bh, "marking dirty");
1118 mark_buffer_dirty(bh);
1120 sync_dirty_buffer(bh);
1121 if (buffer_write_io_error(bh)) {
1122 printk(KERN_ERR "JBD2: I/O error detected "
1123 "when updating journal superblock for %s.\n",
1124 journal->j_devname);
1125 clear_buffer_write_io_error(bh);
1126 set_buffer_uptodate(bh);
1129 write_dirty_buffer(bh, WRITE);
1132 /* If we have just flushed the log (by marking s_start==0), then
1133 * any future commit will have to be careful to update the
1134 * superblock again to re-record the true start of the log. */
1136 write_lock(&journal->j_state_lock);
1138 journal->j_flags &= ~JBD2_FLUSHED;
1140 journal->j_flags |= JBD2_FLUSHED;
1141 write_unlock(&journal->j_state_lock);
1145 * Read the superblock for a given journal, performing initial
1146 * validation of the format.
1149 static int journal_get_superblock(journal_t *journal)
1151 struct buffer_head *bh;
1152 journal_superblock_t *sb;
1155 bh = journal->j_sb_buffer;
1157 J_ASSERT(bh != NULL);
1158 if (!buffer_uptodate(bh)) {
1159 ll_rw_block(READ, 1, &bh);
1161 if (!buffer_uptodate(bh)) {
1163 "JBD: IO error reading journal superblock\n");
1168 sb = journal->j_superblock;
1172 if (sb->s_header.h_magic != cpu_to_be32(JBD2_MAGIC_NUMBER) ||
1173 sb->s_blocksize != cpu_to_be32(journal->j_blocksize)) {
1174 printk(KERN_WARNING "JBD: no valid journal superblock found\n");
1178 switch(be32_to_cpu(sb->s_header.h_blocktype)) {
1179 case JBD2_SUPERBLOCK_V1:
1180 journal->j_format_version = 1;
1182 case JBD2_SUPERBLOCK_V2:
1183 journal->j_format_version = 2;
1186 printk(KERN_WARNING "JBD: unrecognised superblock format ID\n");
1190 if (be32_to_cpu(sb->s_maxlen) < journal->j_maxlen)
1191 journal->j_maxlen = be32_to_cpu(sb->s_maxlen);
1192 else if (be32_to_cpu(sb->s_maxlen) > journal->j_maxlen) {
1193 printk (KERN_WARNING "JBD: journal file too short\n");
1200 journal_fail_superblock(journal);
1205 * Load the on-disk journal superblock and read the key fields into the
1209 static int load_superblock(journal_t *journal)
1212 journal_superblock_t *sb;
1214 err = journal_get_superblock(journal);
1218 sb = journal->j_superblock;
1220 journal->j_tail_sequence = be32_to_cpu(sb->s_sequence);
1221 journal->j_tail = be32_to_cpu(sb->s_start);
1222 journal->j_first = be32_to_cpu(sb->s_first);
1223 journal->j_last = be32_to_cpu(sb->s_maxlen);
1224 journal->j_errno = be32_to_cpu(sb->s_errno);
1231 * int jbd2_journal_load() - Read journal from disk.
1232 * @journal: Journal to act on.
1234 * Given a journal_t structure which tells us which disk blocks contain
1235 * a journal, read the journal from disk to initialise the in-memory
1238 int jbd2_journal_load(journal_t *journal)
1241 journal_superblock_t *sb;
1243 err = load_superblock(journal);
1247 sb = journal->j_superblock;
1248 /* If this is a V2 superblock, then we have to check the
1249 * features flags on it. */
1251 if (journal->j_format_version >= 2) {
1252 if ((sb->s_feature_ro_compat &
1253 ~cpu_to_be32(JBD2_KNOWN_ROCOMPAT_FEATURES)) ||
1254 (sb->s_feature_incompat &
1255 ~cpu_to_be32(JBD2_KNOWN_INCOMPAT_FEATURES))) {
1256 printk (KERN_WARNING
1257 "JBD: Unrecognised features on journal\n");
1263 * Create a slab for this blocksize
1265 err = jbd2_journal_create_slab(be32_to_cpu(sb->s_blocksize));
1269 /* Let the recovery code check whether it needs to recover any
1270 * data from the journal. */
1271 if (jbd2_journal_recover(journal))
1272 goto recovery_error;
1274 if (journal->j_failed_commit) {
1275 printk(KERN_ERR "JBD2: journal transaction %u on %s "
1276 "is corrupt.\n", journal->j_failed_commit,
1277 journal->j_devname);
1281 /* OK, we've finished with the dynamic journal bits:
1282 * reinitialise the dynamic contents of the superblock in memory
1283 * and reset them on disk. */
1284 if (journal_reset(journal))
1285 goto recovery_error;
1287 journal->j_flags &= ~JBD2_ABORT;
1288 journal->j_flags |= JBD2_LOADED;
1292 printk (KERN_WARNING "JBD: recovery failed\n");
1297 * void jbd2_journal_destroy() - Release a journal_t structure.
1298 * @journal: Journal to act on.
1300 * Release a journal_t structure once it is no longer in use by the
1302 * Return <0 if we couldn't clean up the journal.
1304 int jbd2_journal_destroy(journal_t *journal)
1308 /* Wait for the commit thread to wake up and die. */
1309 journal_kill_thread(journal);
1311 /* Force a final log commit */
1312 if (journal->j_running_transaction)
1313 jbd2_journal_commit_transaction(journal);
1315 /* Force any old transactions to disk */
1317 /* Totally anal locking here... */
1318 spin_lock(&journal->j_list_lock);
1319 while (journal->j_checkpoint_transactions != NULL) {
1320 spin_unlock(&journal->j_list_lock);
1321 mutex_lock(&journal->j_checkpoint_mutex);
1322 jbd2_log_do_checkpoint(journal);
1323 mutex_unlock(&journal->j_checkpoint_mutex);
1324 spin_lock(&journal->j_list_lock);
1327 J_ASSERT(journal->j_running_transaction == NULL);
1328 J_ASSERT(journal->j_committing_transaction == NULL);
1329 J_ASSERT(journal->j_checkpoint_transactions == NULL);
1330 spin_unlock(&journal->j_list_lock);
1332 if (journal->j_sb_buffer) {
1333 if (!is_journal_aborted(journal)) {
1334 /* We can now mark the journal as empty. */
1335 journal->j_tail = 0;
1336 journal->j_tail_sequence =
1337 ++journal->j_transaction_sequence;
1338 jbd2_journal_update_superblock(journal, 1);
1342 brelse(journal->j_sb_buffer);
1345 if (journal->j_proc_entry)
1346 jbd2_stats_proc_exit(journal);
1347 if (journal->j_inode)
1348 iput(journal->j_inode);
1349 if (journal->j_revoke)
1350 jbd2_journal_destroy_revoke(journal);
1351 kfree(journal->j_wbuf);
1359 *int jbd2_journal_check_used_features () - Check if features specified are used.
1360 * @journal: Journal to check.
1361 * @compat: bitmask of compatible features
1362 * @ro: bitmask of features that force read-only mount
1363 * @incompat: bitmask of incompatible features
1365 * Check whether the journal uses all of a given set of
1366 * features. Return true (non-zero) if it does.
1369 int jbd2_journal_check_used_features (journal_t *journal, unsigned long compat,
1370 unsigned long ro, unsigned long incompat)
1372 journal_superblock_t *sb;
1374 if (!compat && !ro && !incompat)
1376 if (journal->j_format_version == 1)
1379 sb = journal->j_superblock;
1381 if (((be32_to_cpu(sb->s_feature_compat) & compat) == compat) &&
1382 ((be32_to_cpu(sb->s_feature_ro_compat) & ro) == ro) &&
1383 ((be32_to_cpu(sb->s_feature_incompat) & incompat) == incompat))
1390 * int jbd2_journal_check_available_features() - Check feature set in journalling layer
1391 * @journal: Journal to check.
1392 * @compat: bitmask of compatible features
1393 * @ro: bitmask of features that force read-only mount
1394 * @incompat: bitmask of incompatible features
1396 * Check whether the journaling code supports the use of
1397 * all of a given set of features on this journal. Return true
1398 * (non-zero) if it can. */
1400 int jbd2_journal_check_available_features (journal_t *journal, unsigned long compat,
1401 unsigned long ro, unsigned long incompat)
1403 if (!compat && !ro && !incompat)
1406 /* We can support any known requested features iff the
1407 * superblock is in version 2. Otherwise we fail to support any
1408 * extended sb features. */
1410 if (journal->j_format_version != 2)
1413 if ((compat & JBD2_KNOWN_COMPAT_FEATURES) == compat &&
1414 (ro & JBD2_KNOWN_ROCOMPAT_FEATURES) == ro &&
1415 (incompat & JBD2_KNOWN_INCOMPAT_FEATURES) == incompat)
1422 * int jbd2_journal_set_features () - Mark a given journal feature in the superblock
1423 * @journal: Journal to act on.
1424 * @compat: bitmask of compatible features
1425 * @ro: bitmask of features that force read-only mount
1426 * @incompat: bitmask of incompatible features
1428 * Mark a given journal feature as present on the
1429 * superblock. Returns true if the requested features could be set.
1433 int jbd2_journal_set_features (journal_t *journal, unsigned long compat,
1434 unsigned long ro, unsigned long incompat)
1436 journal_superblock_t *sb;
1438 if (jbd2_journal_check_used_features(journal, compat, ro, incompat))
1441 if (!jbd2_journal_check_available_features(journal, compat, ro, incompat))
1444 jbd_debug(1, "Setting new features 0x%lx/0x%lx/0x%lx\n",
1445 compat, ro, incompat);
1447 sb = journal->j_superblock;
1449 sb->s_feature_compat |= cpu_to_be32(compat);
1450 sb->s_feature_ro_compat |= cpu_to_be32(ro);
1451 sb->s_feature_incompat |= cpu_to_be32(incompat);
1457 * jbd2_journal_clear_features () - Clear a given journal feature in the
1459 * @journal: Journal to act on.
1460 * @compat: bitmask of compatible features
1461 * @ro: bitmask of features that force read-only mount
1462 * @incompat: bitmask of incompatible features
1464 * Clear a given journal feature as present on the
1467 void jbd2_journal_clear_features(journal_t *journal, unsigned long compat,
1468 unsigned long ro, unsigned long incompat)
1470 journal_superblock_t *sb;
1472 jbd_debug(1, "Clear features 0x%lx/0x%lx/0x%lx\n",
1473 compat, ro, incompat);
1475 sb = journal->j_superblock;
1477 sb->s_feature_compat &= ~cpu_to_be32(compat);
1478 sb->s_feature_ro_compat &= ~cpu_to_be32(ro);
1479 sb->s_feature_incompat &= ~cpu_to_be32(incompat);
1481 EXPORT_SYMBOL(jbd2_journal_clear_features);
1484 * int jbd2_journal_update_format () - Update on-disk journal structure.
1485 * @journal: Journal to act on.
1487 * Given an initialised but unloaded journal struct, poke about in the
1488 * on-disk structure to update it to the most recent supported version.
1490 int jbd2_journal_update_format (journal_t *journal)
1492 journal_superblock_t *sb;
1495 err = journal_get_superblock(journal);
1499 sb = journal->j_superblock;
1501 switch (be32_to_cpu(sb->s_header.h_blocktype)) {
1502 case JBD2_SUPERBLOCK_V2:
1504 case JBD2_SUPERBLOCK_V1:
1505 return journal_convert_superblock_v1(journal, sb);
1512 static int journal_convert_superblock_v1(journal_t *journal,
1513 journal_superblock_t *sb)
1515 int offset, blocksize;
1516 struct buffer_head *bh;
1519 "JBD: Converting superblock from version 1 to 2.\n");
1521 /* Pre-initialise new fields to zero */
1522 offset = ((char *) &(sb->s_feature_compat)) - ((char *) sb);
1523 blocksize = be32_to_cpu(sb->s_blocksize);
1524 memset(&sb->s_feature_compat, 0, blocksize-offset);
1526 sb->s_nr_users = cpu_to_be32(1);
1527 sb->s_header.h_blocktype = cpu_to_be32(JBD2_SUPERBLOCK_V2);
1528 journal->j_format_version = 2;
1530 bh = journal->j_sb_buffer;
1531 BUFFER_TRACE(bh, "marking dirty");
1532 mark_buffer_dirty(bh);
1533 sync_dirty_buffer(bh);
1539 * int jbd2_journal_flush () - Flush journal
1540 * @journal: Journal to act on.
1542 * Flush all data for a given journal to disk and empty the journal.
1543 * Filesystems can use this when remounting readonly to ensure that
1544 * recovery does not need to happen on remount.
1547 int jbd2_journal_flush(journal_t *journal)
1550 transaction_t *transaction = NULL;
1551 unsigned long old_tail;
1553 write_lock(&journal->j_state_lock);
1555 /* Force everything buffered to the log... */
1556 if (journal->j_running_transaction) {
1557 transaction = journal->j_running_transaction;
1558 __jbd2_log_start_commit(journal, transaction->t_tid);
1559 } else if (journal->j_committing_transaction)
1560 transaction = journal->j_committing_transaction;
1562 /* Wait for the log commit to complete... */
1564 tid_t tid = transaction->t_tid;
1566 write_unlock(&journal->j_state_lock);
1567 jbd2_log_wait_commit(journal, tid);
1569 write_unlock(&journal->j_state_lock);
1572 /* ...and flush everything in the log out to disk. */
1573 spin_lock(&journal->j_list_lock);
1574 while (!err && journal->j_checkpoint_transactions != NULL) {
1575 spin_unlock(&journal->j_list_lock);
1576 mutex_lock(&journal->j_checkpoint_mutex);
1577 err = jbd2_log_do_checkpoint(journal);
1578 mutex_unlock(&journal->j_checkpoint_mutex);
1579 spin_lock(&journal->j_list_lock);
1581 spin_unlock(&journal->j_list_lock);
1583 if (is_journal_aborted(journal))
1586 jbd2_cleanup_journal_tail(journal);
1588 /* Finally, mark the journal as really needing no recovery.
1589 * This sets s_start==0 in the underlying superblock, which is
1590 * the magic code for a fully-recovered superblock. Any future
1591 * commits of data to the journal will restore the current
1593 write_lock(&journal->j_state_lock);
1594 old_tail = journal->j_tail;
1595 journal->j_tail = 0;
1596 write_unlock(&journal->j_state_lock);
1597 jbd2_journal_update_superblock(journal, 1);
1598 write_lock(&journal->j_state_lock);
1599 journal->j_tail = old_tail;
1601 J_ASSERT(!journal->j_running_transaction);
1602 J_ASSERT(!journal->j_committing_transaction);
1603 J_ASSERT(!journal->j_checkpoint_transactions);
1604 J_ASSERT(journal->j_head == journal->j_tail);
1605 J_ASSERT(journal->j_tail_sequence == journal->j_transaction_sequence);
1606 write_unlock(&journal->j_state_lock);
1611 * int jbd2_journal_wipe() - Wipe journal contents
1612 * @journal: Journal to act on.
1613 * @write: flag (see below)
1615 * Wipe out all of the contents of a journal, safely. This will produce
1616 * a warning if the journal contains any valid recovery information.
1617 * Must be called between journal_init_*() and jbd2_journal_load().
1619 * If 'write' is non-zero, then we wipe out the journal on disk; otherwise
1620 * we merely suppress recovery.
1623 int jbd2_journal_wipe(journal_t *journal, int write)
1627 J_ASSERT (!(journal->j_flags & JBD2_LOADED));
1629 err = load_superblock(journal);
1633 if (!journal->j_tail)
1636 printk (KERN_WARNING "JBD: %s recovery information on journal\n",
1637 write ? "Clearing" : "Ignoring");
1639 err = jbd2_journal_skip_recovery(journal);
1641 jbd2_journal_update_superblock(journal, 1);
1648 * Journal abort has very specific semantics, which we describe
1649 * for journal abort.
1651 * Two internal functions, which provide abort to the jbd layer
1656 * Quick version for internal journal use (doesn't lock the journal).
1657 * Aborts hard --- we mark the abort as occurred, but do _nothing_ else,
1658 * and don't attempt to make any other journal updates.
1660 void __jbd2_journal_abort_hard(journal_t *journal)
1662 transaction_t *transaction;
1664 if (journal->j_flags & JBD2_ABORT)
1667 printk(KERN_ERR "Aborting journal on device %s.\n",
1668 journal->j_devname);
1670 write_lock(&journal->j_state_lock);
1671 journal->j_flags |= JBD2_ABORT;
1672 transaction = journal->j_running_transaction;
1674 __jbd2_log_start_commit(journal, transaction->t_tid);
1675 write_unlock(&journal->j_state_lock);
1678 /* Soft abort: record the abort error status in the journal superblock,
1679 * but don't do any other IO. */
1680 static void __journal_abort_soft (journal_t *journal, int errno)
1682 if (journal->j_flags & JBD2_ABORT)
1685 if (!journal->j_errno)
1686 journal->j_errno = errno;
1688 __jbd2_journal_abort_hard(journal);
1691 jbd2_journal_update_superblock(journal, 1);
1695 * void jbd2_journal_abort () - Shutdown the journal immediately.
1696 * @journal: the journal to shutdown.
1697 * @errno: an error number to record in the journal indicating
1698 * the reason for the shutdown.
1700 * Perform a complete, immediate shutdown of the ENTIRE
1701 * journal (not of a single transaction). This operation cannot be
1702 * undone without closing and reopening the journal.
1704 * The jbd2_journal_abort function is intended to support higher level error
1705 * recovery mechanisms such as the ext2/ext3 remount-readonly error
1708 * Journal abort has very specific semantics. Any existing dirty,
1709 * unjournaled buffers in the main filesystem will still be written to
1710 * disk by bdflush, but the journaling mechanism will be suspended
1711 * immediately and no further transaction commits will be honoured.
1713 * Any dirty, journaled buffers will be written back to disk without
1714 * hitting the journal. Atomicity cannot be guaranteed on an aborted
1715 * filesystem, but we _do_ attempt to leave as much data as possible
1716 * behind for fsck to use for cleanup.
1718 * Any attempt to get a new transaction handle on a journal which is in
1719 * ABORT state will just result in an -EROFS error return. A
1720 * jbd2_journal_stop on an existing handle will return -EIO if we have
1721 * entered abort state during the update.
1723 * Recursive transactions are not disturbed by journal abort until the
1724 * final jbd2_journal_stop, which will receive the -EIO error.
1726 * Finally, the jbd2_journal_abort call allows the caller to supply an errno
1727 * which will be recorded (if possible) in the journal superblock. This
1728 * allows a client to record failure conditions in the middle of a
1729 * transaction without having to complete the transaction to record the
1730 * failure to disk. ext3_error, for example, now uses this
1733 * Errors which originate from within the journaling layer will NOT
1734 * supply an errno; a null errno implies that absolutely no further
1735 * writes are done to the journal (unless there are any already in
1740 void jbd2_journal_abort(journal_t *journal, int errno)
1742 __journal_abort_soft(journal, errno);
1746 * int jbd2_journal_errno () - returns the journal's error state.
1747 * @journal: journal to examine.
1749 * This is the errno number set with jbd2_journal_abort(), the last
1750 * time the journal was mounted - if the journal was stopped
1751 * without calling abort this will be 0.
1753 * If the journal has been aborted on this mount time -EROFS will
1756 int jbd2_journal_errno(journal_t *journal)
1760 read_lock(&journal->j_state_lock);
1761 if (journal->j_flags & JBD2_ABORT)
1764 err = journal->j_errno;
1765 read_unlock(&journal->j_state_lock);
1770 * int jbd2_journal_clear_err () - clears the journal's error state
1771 * @journal: journal to act on.
1773 * An error must be cleared or acked to take a FS out of readonly
1776 int jbd2_journal_clear_err(journal_t *journal)
1780 write_lock(&journal->j_state_lock);
1781 if (journal->j_flags & JBD2_ABORT)
1784 journal->j_errno = 0;
1785 write_unlock(&journal->j_state_lock);
1790 * void jbd2_journal_ack_err() - Ack journal err.
1791 * @journal: journal to act on.
1793 * An error must be cleared or acked to take a FS out of readonly
1796 void jbd2_journal_ack_err(journal_t *journal)
1798 write_lock(&journal->j_state_lock);
1799 if (journal->j_errno)
1800 journal->j_flags |= JBD2_ACK_ERR;
1801 write_unlock(&journal->j_state_lock);
1804 int jbd2_journal_blocks_per_page(struct inode *inode)
1806 return 1 << (PAGE_CACHE_SHIFT - inode->i_sb->s_blocksize_bits);
1810 * helper functions to deal with 32 or 64bit block numbers.
1812 size_t journal_tag_bytes(journal_t *journal)
1814 if (JBD2_HAS_INCOMPAT_FEATURE(journal, JBD2_FEATURE_INCOMPAT_64BIT))
1815 return JBD2_TAG_SIZE64;
1817 return JBD2_TAG_SIZE32;
1821 * JBD memory management
1823 * These functions are used to allocate block-sized chunks of memory
1824 * used for making copies of buffer_head data. Very often it will be
1825 * page-sized chunks of data, but sometimes it will be in
1826 * sub-page-size chunks. (For example, 16k pages on Power systems
1827 * with a 4k block file system.) For blocks smaller than a page, we
1828 * use a SLAB allocator. There are slab caches for each block size,
1829 * which are allocated at mount time, if necessary, and we only free
1830 * (all of) the slab caches when/if the jbd2 module is unloaded. For
1831 * this reason we don't need to a mutex to protect access to
1832 * jbd2_slab[] allocating or releasing memory; only in
1833 * jbd2_journal_create_slab().
1835 #define JBD2_MAX_SLABS 8
1836 static struct kmem_cache *jbd2_slab[JBD2_MAX_SLABS];
1837 static DECLARE_MUTEX(jbd2_slab_create_sem);
1839 static const char *jbd2_slab_names[JBD2_MAX_SLABS] = {
1840 "jbd2_1k", "jbd2_2k", "jbd2_4k", "jbd2_8k",
1841 "jbd2_16k", "jbd2_32k", "jbd2_64k", "jbd2_128k"
1845 static void jbd2_journal_destroy_slabs(void)
1849 for (i = 0; i < JBD2_MAX_SLABS; i++) {
1851 kmem_cache_destroy(jbd2_slab[i]);
1852 jbd2_slab[i] = NULL;
1856 static int jbd2_journal_create_slab(size_t size)
1858 int i = order_base_2(size) - 10;
1861 if (size == PAGE_SIZE)
1864 if (i >= JBD2_MAX_SLABS)
1867 if (unlikely(i < 0))
1869 down(&jbd2_slab_create_sem);
1871 up(&jbd2_slab_create_sem);
1872 return 0; /* Already created */
1875 slab_size = 1 << (i+10);
1876 jbd2_slab[i] = kmem_cache_create(jbd2_slab_names[i], slab_size,
1877 slab_size, 0, NULL);
1878 up(&jbd2_slab_create_sem);
1879 if (!jbd2_slab[i]) {
1880 printk(KERN_EMERG "JBD2: no memory for jbd2_slab cache\n");
1886 static struct kmem_cache *get_slab(size_t size)
1888 int i = order_base_2(size) - 10;
1890 BUG_ON(i >= JBD2_MAX_SLABS);
1891 if (unlikely(i < 0))
1893 BUG_ON(jbd2_slab[i] == NULL);
1894 return jbd2_slab[i];
1897 void *jbd2_alloc(size_t size, gfp_t flags)
1901 BUG_ON(size & (size-1)); /* Must be a power of 2 */
1903 flags |= __GFP_REPEAT;
1904 if (size == PAGE_SIZE)
1905 ptr = (void *)__get_free_pages(flags, 0);
1906 else if (size > PAGE_SIZE) {
1907 int order = get_order(size);
1910 ptr = (void *)__get_free_pages(flags, order);
1912 ptr = vmalloc(size);
1914 ptr = kmem_cache_alloc(get_slab(size), flags);
1916 /* Check alignment; SLUB has gotten this wrong in the past,
1917 * and this can lead to user data corruption! */
1918 BUG_ON(((unsigned long) ptr) & (size-1));
1923 void jbd2_free(void *ptr, size_t size)
1925 if (size == PAGE_SIZE) {
1926 free_pages((unsigned long)ptr, 0);
1929 if (size > PAGE_SIZE) {
1930 int order = get_order(size);
1933 free_pages((unsigned long)ptr, order);
1938 kmem_cache_free(get_slab(size), ptr);
1942 * Journal_head storage management
1944 static struct kmem_cache *jbd2_journal_head_cache;
1945 #ifdef CONFIG_JBD2_DEBUG
1946 static atomic_t nr_journal_heads = ATOMIC_INIT(0);
1949 static int journal_init_jbd2_journal_head_cache(void)
1953 J_ASSERT(jbd2_journal_head_cache == NULL);
1954 jbd2_journal_head_cache = kmem_cache_create("jbd2_journal_head",
1955 sizeof(struct journal_head),
1957 SLAB_TEMPORARY, /* flags */
1960 if (!jbd2_journal_head_cache) {
1962 printk(KERN_EMERG "JBD: no memory for journal_head cache\n");
1967 static void jbd2_journal_destroy_jbd2_journal_head_cache(void)
1969 if (jbd2_journal_head_cache) {
1970 kmem_cache_destroy(jbd2_journal_head_cache);
1971 jbd2_journal_head_cache = NULL;
1976 * journal_head splicing and dicing
1978 static struct journal_head *journal_alloc_journal_head(void)
1980 struct journal_head *ret;
1981 static unsigned long last_warning;
1983 #ifdef CONFIG_JBD2_DEBUG
1984 atomic_inc(&nr_journal_heads);
1986 ret = kmem_cache_alloc(jbd2_journal_head_cache, GFP_NOFS);
1988 jbd_debug(1, "out of memory for journal_head\n");
1989 if (time_after(jiffies, last_warning + 5*HZ)) {
1990 printk(KERN_NOTICE "ENOMEM in %s, retrying.\n",
1992 last_warning = jiffies;
1996 ret = kmem_cache_alloc(jbd2_journal_head_cache, GFP_NOFS);
2002 static void journal_free_journal_head(struct journal_head *jh)
2004 #ifdef CONFIG_JBD2_DEBUG
2005 atomic_dec(&nr_journal_heads);
2006 memset(jh, JBD2_POISON_FREE, sizeof(*jh));
2008 kmem_cache_free(jbd2_journal_head_cache, jh);
2012 * A journal_head is attached to a buffer_head whenever JBD has an
2013 * interest in the buffer.
2015 * Whenever a buffer has an attached journal_head, its ->b_state:BH_JBD bit
2016 * is set. This bit is tested in core kernel code where we need to take
2017 * JBD-specific actions. Testing the zeroness of ->b_private is not reliable
2020 * When a buffer has its BH_JBD bit set, its ->b_count is elevated by one.
2022 * When a buffer has its BH_JBD bit set it is immune from being released by
2023 * core kernel code, mainly via ->b_count.
2025 * A journal_head may be detached from its buffer_head when the journal_head's
2026 * b_transaction, b_cp_transaction and b_next_transaction pointers are NULL.
2027 * Various places in JBD call jbd2_journal_remove_journal_head() to indicate that the
2028 * journal_head can be dropped if needed.
2030 * Various places in the kernel want to attach a journal_head to a buffer_head
2031 * _before_ attaching the journal_head to a transaction. To protect the
2032 * journal_head in this situation, jbd2_journal_add_journal_head elevates the
2033 * journal_head's b_jcount refcount by one. The caller must call
2034 * jbd2_journal_put_journal_head() to undo this.
2036 * So the typical usage would be:
2038 * (Attach a journal_head if needed. Increments b_jcount)
2039 * struct journal_head *jh = jbd2_journal_add_journal_head(bh);
2041 * jh->b_transaction = xxx;
2042 * jbd2_journal_put_journal_head(jh);
2044 * Now, the journal_head's b_jcount is zero, but it is safe from being released
2045 * because it has a non-zero b_transaction.
2049 * Give a buffer_head a journal_head.
2051 * Doesn't need the journal lock.
2054 struct journal_head *jbd2_journal_add_journal_head(struct buffer_head *bh)
2056 struct journal_head *jh;
2057 struct journal_head *new_jh = NULL;
2060 if (!buffer_jbd(bh)) {
2061 new_jh = journal_alloc_journal_head();
2062 memset(new_jh, 0, sizeof(*new_jh));
2065 jbd_lock_bh_journal_head(bh);
2066 if (buffer_jbd(bh)) {
2070 (atomic_read(&bh->b_count) > 0) ||
2071 (bh->b_page && bh->b_page->mapping));
2074 jbd_unlock_bh_journal_head(bh);
2079 new_jh = NULL; /* We consumed it */
2084 BUFFER_TRACE(bh, "added journal_head");
2087 jbd_unlock_bh_journal_head(bh);
2089 journal_free_journal_head(new_jh);
2090 return bh->b_private;
2094 * Grab a ref against this buffer_head's journal_head. If it ended up not
2095 * having a journal_head, return NULL
2097 struct journal_head *jbd2_journal_grab_journal_head(struct buffer_head *bh)
2099 struct journal_head *jh = NULL;
2101 jbd_lock_bh_journal_head(bh);
2102 if (buffer_jbd(bh)) {
2106 jbd_unlock_bh_journal_head(bh);
2110 static void __journal_remove_journal_head(struct buffer_head *bh)
2112 struct journal_head *jh = bh2jh(bh);
2114 J_ASSERT_JH(jh, jh->b_jcount >= 0);
2117 if (jh->b_jcount == 0) {
2118 if (jh->b_transaction == NULL &&
2119 jh->b_next_transaction == NULL &&
2120 jh->b_cp_transaction == NULL) {
2121 J_ASSERT_JH(jh, jh->b_jlist == BJ_None);
2122 J_ASSERT_BH(bh, buffer_jbd(bh));
2123 J_ASSERT_BH(bh, jh2bh(jh) == bh);
2124 BUFFER_TRACE(bh, "remove journal_head");
2125 if (jh->b_frozen_data) {
2126 printk(KERN_WARNING "%s: freeing "
2129 jbd2_free(jh->b_frozen_data, bh->b_size);
2131 if (jh->b_committed_data) {
2132 printk(KERN_WARNING "%s: freeing "
2133 "b_committed_data\n",
2135 jbd2_free(jh->b_committed_data, bh->b_size);
2137 bh->b_private = NULL;
2138 jh->b_bh = NULL; /* debug, really */
2139 clear_buffer_jbd(bh);
2141 journal_free_journal_head(jh);
2143 BUFFER_TRACE(bh, "journal_head was locked");
2149 * jbd2_journal_remove_journal_head(): if the buffer isn't attached to a transaction
2150 * and has a zero b_jcount then remove and release its journal_head. If we did
2151 * see that the buffer is not used by any transaction we also "logically"
2152 * decrement ->b_count.
2154 * We in fact take an additional increment on ->b_count as a convenience,
2155 * because the caller usually wants to do additional things with the bh
2156 * after calling here.
2157 * The caller of jbd2_journal_remove_journal_head() *must* run __brelse(bh) at some
2158 * time. Once the caller has run __brelse(), the buffer is eligible for
2159 * reaping by try_to_free_buffers().
2161 void jbd2_journal_remove_journal_head(struct buffer_head *bh)
2163 jbd_lock_bh_journal_head(bh);
2164 __journal_remove_journal_head(bh);
2165 jbd_unlock_bh_journal_head(bh);
2169 * Drop a reference on the passed journal_head. If it fell to zero then try to
2170 * release the journal_head from the buffer_head.
2172 void jbd2_journal_put_journal_head(struct journal_head *jh)
2174 struct buffer_head *bh = jh2bh(jh);
2176 jbd_lock_bh_journal_head(bh);
2177 J_ASSERT_JH(jh, jh->b_jcount > 0);
2179 if (!jh->b_jcount && !jh->b_transaction) {
2180 __journal_remove_journal_head(bh);
2183 jbd_unlock_bh_journal_head(bh);
2187 * Initialize jbd inode head
2189 void jbd2_journal_init_jbd_inode(struct jbd2_inode *jinode, struct inode *inode)
2191 jinode->i_transaction = NULL;
2192 jinode->i_next_transaction = NULL;
2193 jinode->i_vfs_inode = inode;
2194 jinode->i_flags = 0;
2195 INIT_LIST_HEAD(&jinode->i_list);
2199 * Function to be called before we start removing inode from memory (i.e.,
2200 * clear_inode() is a fine place to be called from). It removes inode from
2201 * transaction's lists.
2203 void jbd2_journal_release_jbd_inode(journal_t *journal,
2204 struct jbd2_inode *jinode)
2209 spin_lock(&journal->j_list_lock);
2210 /* Is commit writing out inode - we have to wait */
2211 if (test_bit(__JI_COMMIT_RUNNING, &jinode->i_flags)) {
2212 wait_queue_head_t *wq;
2213 DEFINE_WAIT_BIT(wait, &jinode->i_flags, __JI_COMMIT_RUNNING);
2214 wq = bit_waitqueue(&jinode->i_flags, __JI_COMMIT_RUNNING);
2215 prepare_to_wait(wq, &wait.wait, TASK_UNINTERRUPTIBLE);
2216 spin_unlock(&journal->j_list_lock);
2218 finish_wait(wq, &wait.wait);
2222 if (jinode->i_transaction) {
2223 list_del(&jinode->i_list);
2224 jinode->i_transaction = NULL;
2226 spin_unlock(&journal->j_list_lock);
2232 #ifdef CONFIG_JBD2_DEBUG
2233 u8 jbd2_journal_enable_debug __read_mostly;
2234 EXPORT_SYMBOL(jbd2_journal_enable_debug);
2236 #define JBD2_DEBUG_NAME "jbd2-debug"
2238 static struct dentry *jbd2_debugfs_dir;
2239 static struct dentry *jbd2_debug;
2241 static void __init jbd2_create_debugfs_entry(void)
2243 jbd2_debugfs_dir = debugfs_create_dir("jbd2", NULL);
2244 if (jbd2_debugfs_dir)
2245 jbd2_debug = debugfs_create_u8(JBD2_DEBUG_NAME,
2248 &jbd2_journal_enable_debug);
2251 static void __exit jbd2_remove_debugfs_entry(void)
2253 debugfs_remove(jbd2_debug);
2254 debugfs_remove(jbd2_debugfs_dir);
2259 static void __init jbd2_create_debugfs_entry(void)
2263 static void __exit jbd2_remove_debugfs_entry(void)
2269 #ifdef CONFIG_PROC_FS
2271 #define JBD2_STATS_PROC_NAME "fs/jbd2"
2273 static void __init jbd2_create_jbd_stats_proc_entry(void)
2275 proc_jbd2_stats = proc_mkdir(JBD2_STATS_PROC_NAME, NULL);
2278 static void __exit jbd2_remove_jbd_stats_proc_entry(void)
2280 if (proc_jbd2_stats)
2281 remove_proc_entry(JBD2_STATS_PROC_NAME, NULL);
2286 #define jbd2_create_jbd_stats_proc_entry() do {} while (0)
2287 #define jbd2_remove_jbd_stats_proc_entry() do {} while (0)
2291 struct kmem_cache *jbd2_handle_cache;
2293 static int __init journal_init_handle_cache(void)
2295 jbd2_handle_cache = kmem_cache_create("jbd2_journal_handle",
2298 SLAB_TEMPORARY, /* flags */
2300 if (jbd2_handle_cache == NULL) {
2301 printk(KERN_EMERG "JBD: failed to create handle cache\n");
2307 static void jbd2_journal_destroy_handle_cache(void)
2309 if (jbd2_handle_cache)
2310 kmem_cache_destroy(jbd2_handle_cache);
2314 * Module startup and shutdown
2317 static int __init journal_init_caches(void)
2321 ret = jbd2_journal_init_revoke_caches();
2323 ret = journal_init_jbd2_journal_head_cache();
2325 ret = journal_init_handle_cache();
2329 static void jbd2_journal_destroy_caches(void)
2331 jbd2_journal_destroy_revoke_caches();
2332 jbd2_journal_destroy_jbd2_journal_head_cache();
2333 jbd2_journal_destroy_handle_cache();
2334 jbd2_journal_destroy_slabs();
2337 static int __init journal_init(void)
2341 BUILD_BUG_ON(sizeof(struct journal_superblock_s) != 1024);
2343 ret = journal_init_caches();
2345 jbd2_create_debugfs_entry();
2346 jbd2_create_jbd_stats_proc_entry();
2348 jbd2_journal_destroy_caches();
2353 static void __exit journal_exit(void)
2355 #ifdef CONFIG_JBD2_DEBUG
2356 int n = atomic_read(&nr_journal_heads);
2358 printk(KERN_EMERG "JBD: leaked %d journal_heads!\n", n);
2360 jbd2_remove_debugfs_entry();
2361 jbd2_remove_jbd_stats_proc_entry();
2362 jbd2_journal_destroy_caches();
2366 * jbd2_dev_to_name is a utility function used by the jbd2 and ext4
2367 * tracing infrastructure to map a dev_t to a device name.
2369 * The caller should use rcu_read_lock() in order to make sure the
2370 * device name stays valid until its done with it. We use
2371 * rcu_read_lock() as well to make sure we're safe in case the caller
2372 * gets sloppy, and because rcu_read_lock() is cheap and can be safely
2375 struct devname_cache {
2376 struct rcu_head rcu;
2378 char devname[BDEVNAME_SIZE];
2380 #define CACHE_SIZE_BITS 6
2381 static struct devname_cache *devcache[1 << CACHE_SIZE_BITS];
2382 static DEFINE_SPINLOCK(devname_cache_lock);
2384 static void free_devcache(struct rcu_head *rcu)
2389 const char *jbd2_dev_to_name(dev_t device)
2391 int i = hash_32(device, CACHE_SIZE_BITS);
2393 struct block_device *bd;
2394 static struct devname_cache *new_dev;
2397 if (devcache[i] && devcache[i]->device == device) {
2398 ret = devcache[i]->devname;
2404 new_dev = kmalloc(sizeof(struct devname_cache), GFP_KERNEL);
2406 return "NODEV-ALLOCFAILURE"; /* Something non-NULL */
2407 spin_lock(&devname_cache_lock);
2409 if (devcache[i]->device == device) {
2411 ret = devcache[i]->devname;
2412 spin_unlock(&devname_cache_lock);
2415 call_rcu(&devcache[i]->rcu, free_devcache);
2417 devcache[i] = new_dev;
2418 devcache[i]->device = device;
2421 bdevname(bd, devcache[i]->devname);
2424 __bdevname(device, devcache[i]->devname);
2425 ret = devcache[i]->devname;
2426 spin_unlock(&devname_cache_lock);
2429 EXPORT_SYMBOL(jbd2_dev_to_name);
2431 MODULE_LICENSE("GPL");
2432 module_init(journal_init);
2433 module_exit(journal_exit);