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_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);
96 EXPORT_SYMBOL(jbd2_inode_cache);
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 with j_state_lock locked for writing.
475 * Returns true if a transaction commit was started.
477 int __jbd2_log_start_commit(journal_t *journal, tid_t target)
480 * The only transaction we can possibly wait upon is the
481 * currently running transaction (if it exists). Otherwise,
482 * the target tid must be an old one.
484 if (journal->j_running_transaction &&
485 journal->j_running_transaction->t_tid == target) {
487 * We want a new commit: OK, mark the request and wakeup the
488 * commit thread. We do _not_ do the commit ourselves.
491 journal->j_commit_request = target;
492 jbd_debug(1, "JBD2: requesting commit %d/%d\n",
493 journal->j_commit_request,
494 journal->j_commit_sequence);
495 wake_up(&journal->j_wait_commit);
497 } else if (!tid_geq(journal->j_commit_request, target))
498 /* This should never happen, but if it does, preserve
499 the evidence before kjournald goes into a loop and
500 increments j_commit_sequence beyond all recognition. */
501 WARN_ONCE(1, "JBD2: bad log_start_commit: %u %u %u %u\n",
502 journal->j_commit_request,
503 journal->j_commit_sequence,
504 target, journal->j_running_transaction ?
505 journal->j_running_transaction->t_tid : 0);
509 int jbd2_log_start_commit(journal_t *journal, tid_t tid)
513 write_lock(&journal->j_state_lock);
514 ret = __jbd2_log_start_commit(journal, tid);
515 write_unlock(&journal->j_state_lock);
520 * Force and wait upon a commit if the calling process is not within
521 * transaction. This is used for forcing out undo-protected data which contains
522 * bitmaps, when the fs is running out of space.
524 * We can only force the running transaction if we don't have an active handle;
525 * otherwise, we will deadlock.
527 * Returns true if a transaction was started.
529 int jbd2_journal_force_commit_nested(journal_t *journal)
531 transaction_t *transaction = NULL;
533 int need_to_start = 0;
535 read_lock(&journal->j_state_lock);
536 if (journal->j_running_transaction && !current->journal_info) {
537 transaction = journal->j_running_transaction;
538 if (!tid_geq(journal->j_commit_request, transaction->t_tid))
540 } else if (journal->j_committing_transaction)
541 transaction = journal->j_committing_transaction;
544 read_unlock(&journal->j_state_lock);
545 return 0; /* Nothing to retry */
548 tid = transaction->t_tid;
549 read_unlock(&journal->j_state_lock);
551 jbd2_log_start_commit(journal, tid);
552 jbd2_log_wait_commit(journal, tid);
557 * Start a commit of the current running transaction (if any). Returns true
558 * if a transaction is going to be committed (or is currently already
559 * committing), and fills its tid in at *ptid
561 int jbd2_journal_start_commit(journal_t *journal, tid_t *ptid)
565 write_lock(&journal->j_state_lock);
566 if (journal->j_running_transaction) {
567 tid_t tid = journal->j_running_transaction->t_tid;
569 __jbd2_log_start_commit(journal, tid);
570 /* There's a running transaction and we've just made sure
571 * it's commit has been scheduled. */
575 } else if (journal->j_committing_transaction) {
577 * If ext3_write_super() recently started a commit, then we
578 * have to wait for completion of that transaction
581 *ptid = journal->j_committing_transaction->t_tid;
584 write_unlock(&journal->j_state_lock);
589 * Return 1 if a given transaction has not yet sent barrier request
590 * connected with a transaction commit. If 0 is returned, transaction
591 * may or may not have sent the barrier. Used to avoid sending barrier
592 * twice in common cases.
594 int jbd2_trans_will_send_data_barrier(journal_t *journal, tid_t tid)
597 transaction_t *commit_trans;
599 if (!(journal->j_flags & JBD2_BARRIER))
601 read_lock(&journal->j_state_lock);
602 /* Transaction already committed? */
603 if (tid_geq(journal->j_commit_sequence, tid))
605 commit_trans = journal->j_committing_transaction;
606 if (!commit_trans || commit_trans->t_tid != tid) {
611 * Transaction is being committed and we already proceeded to
612 * submitting a flush to fs partition?
614 if (journal->j_fs_dev != journal->j_dev) {
615 if (!commit_trans->t_need_data_flush ||
616 commit_trans->t_state >= T_COMMIT_DFLUSH)
619 if (commit_trans->t_state >= T_COMMIT_JFLUSH)
624 read_unlock(&journal->j_state_lock);
627 EXPORT_SYMBOL(jbd2_trans_will_send_data_barrier);
630 * Wait for a specified commit to complete.
631 * The caller may not hold the journal lock.
633 int jbd2_log_wait_commit(journal_t *journal, tid_t tid)
637 read_lock(&journal->j_state_lock);
638 #ifdef CONFIG_JBD2_DEBUG
639 if (!tid_geq(journal->j_commit_request, tid)) {
641 "%s: error: j_commit_request=%d, tid=%d\n",
642 __func__, journal->j_commit_request, tid);
645 while (tid_gt(tid, journal->j_commit_sequence)) {
646 jbd_debug(1, "JBD2: want %d, j_commit_sequence=%d\n",
647 tid, journal->j_commit_sequence);
648 wake_up(&journal->j_wait_commit);
649 read_unlock(&journal->j_state_lock);
650 wait_event(journal->j_wait_done_commit,
651 !tid_gt(tid, journal->j_commit_sequence));
652 read_lock(&journal->j_state_lock);
654 read_unlock(&journal->j_state_lock);
656 if (unlikely(is_journal_aborted(journal))) {
657 printk(KERN_EMERG "journal commit I/O error\n");
664 * Log buffer allocation routines:
667 int jbd2_journal_next_log_block(journal_t *journal, unsigned long long *retp)
669 unsigned long blocknr;
671 write_lock(&journal->j_state_lock);
672 J_ASSERT(journal->j_free > 1);
674 blocknr = journal->j_head;
677 if (journal->j_head == journal->j_last)
678 journal->j_head = journal->j_first;
679 write_unlock(&journal->j_state_lock);
680 return jbd2_journal_bmap(journal, blocknr, retp);
684 * Conversion of logical to physical block numbers for the journal
686 * On external journals the journal blocks are identity-mapped, so
687 * this is a no-op. If needed, we can use j_blk_offset - everything is
690 int jbd2_journal_bmap(journal_t *journal, unsigned long blocknr,
691 unsigned long long *retp)
694 unsigned long long ret;
696 if (journal->j_inode) {
697 ret = bmap(journal->j_inode, blocknr);
701 printk(KERN_ALERT "%s: journal block not found "
702 "at offset %lu on %s\n",
703 __func__, blocknr, journal->j_devname);
705 __journal_abort_soft(journal, err);
708 *retp = blocknr; /* +journal->j_blk_offset */
714 * We play buffer_head aliasing tricks to write data/metadata blocks to
715 * the journal without copying their contents, but for journal
716 * descriptor blocks we do need to generate bona fide buffers.
718 * After the caller of jbd2_journal_get_descriptor_buffer() has finished modifying
719 * the buffer's contents they really should run flush_dcache_page(bh->b_page).
720 * But we don't bother doing that, so there will be coherency problems with
721 * mmaps of blockdevs which hold live JBD-controlled filesystems.
723 struct journal_head *jbd2_journal_get_descriptor_buffer(journal_t *journal)
725 struct buffer_head *bh;
726 unsigned long long blocknr;
729 err = jbd2_journal_next_log_block(journal, &blocknr);
734 bh = __getblk(journal->j_dev, blocknr, journal->j_blocksize);
738 memset(bh->b_data, 0, journal->j_blocksize);
739 set_buffer_uptodate(bh);
741 BUFFER_TRACE(bh, "return this buffer");
742 return jbd2_journal_add_journal_head(bh);
745 struct jbd2_stats_proc_session {
747 struct transaction_stats_s *stats;
752 static void *jbd2_seq_info_start(struct seq_file *seq, loff_t *pos)
754 return *pos ? NULL : SEQ_START_TOKEN;
757 static void *jbd2_seq_info_next(struct seq_file *seq, void *v, loff_t *pos)
762 static int jbd2_seq_info_show(struct seq_file *seq, void *v)
764 struct jbd2_stats_proc_session *s = seq->private;
766 if (v != SEQ_START_TOKEN)
768 seq_printf(seq, "%lu transaction, each up to %u blocks\n",
770 s->journal->j_max_transaction_buffers);
771 if (s->stats->ts_tid == 0)
773 seq_printf(seq, "average: \n %ums waiting for transaction\n",
774 jiffies_to_msecs(s->stats->run.rs_wait / s->stats->ts_tid));
775 seq_printf(seq, " %ums running transaction\n",
776 jiffies_to_msecs(s->stats->run.rs_running / s->stats->ts_tid));
777 seq_printf(seq, " %ums transaction was being locked\n",
778 jiffies_to_msecs(s->stats->run.rs_locked / s->stats->ts_tid));
779 seq_printf(seq, " %ums flushing data (in ordered mode)\n",
780 jiffies_to_msecs(s->stats->run.rs_flushing / s->stats->ts_tid));
781 seq_printf(seq, " %ums logging transaction\n",
782 jiffies_to_msecs(s->stats->run.rs_logging / s->stats->ts_tid));
783 seq_printf(seq, " %lluus average transaction commit time\n",
784 div_u64(s->journal->j_average_commit_time, 1000));
785 seq_printf(seq, " %lu handles per transaction\n",
786 s->stats->run.rs_handle_count / s->stats->ts_tid);
787 seq_printf(seq, " %lu blocks per transaction\n",
788 s->stats->run.rs_blocks / s->stats->ts_tid);
789 seq_printf(seq, " %lu logged blocks per transaction\n",
790 s->stats->run.rs_blocks_logged / s->stats->ts_tid);
794 static void jbd2_seq_info_stop(struct seq_file *seq, void *v)
798 static const struct seq_operations jbd2_seq_info_ops = {
799 .start = jbd2_seq_info_start,
800 .next = jbd2_seq_info_next,
801 .stop = jbd2_seq_info_stop,
802 .show = jbd2_seq_info_show,
805 static int jbd2_seq_info_open(struct inode *inode, struct file *file)
807 journal_t *journal = PDE(inode)->data;
808 struct jbd2_stats_proc_session *s;
811 s = kmalloc(sizeof(*s), GFP_KERNEL);
814 size = sizeof(struct transaction_stats_s);
815 s->stats = kmalloc(size, GFP_KERNEL);
816 if (s->stats == NULL) {
820 spin_lock(&journal->j_history_lock);
821 memcpy(s->stats, &journal->j_stats, size);
822 s->journal = journal;
823 spin_unlock(&journal->j_history_lock);
825 rc = seq_open(file, &jbd2_seq_info_ops);
827 struct seq_file *m = file->private_data;
837 static int jbd2_seq_info_release(struct inode *inode, struct file *file)
839 struct seq_file *seq = file->private_data;
840 struct jbd2_stats_proc_session *s = seq->private;
843 return seq_release(inode, file);
846 static const struct file_operations jbd2_seq_info_fops = {
847 .owner = THIS_MODULE,
848 .open = jbd2_seq_info_open,
851 .release = jbd2_seq_info_release,
854 static struct proc_dir_entry *proc_jbd2_stats;
856 static void jbd2_stats_proc_init(journal_t *journal)
858 journal->j_proc_entry = proc_mkdir(journal->j_devname, proc_jbd2_stats);
859 if (journal->j_proc_entry) {
860 proc_create_data("info", S_IRUGO, journal->j_proc_entry,
861 &jbd2_seq_info_fops, journal);
865 static void jbd2_stats_proc_exit(journal_t *journal)
867 remove_proc_entry("info", journal->j_proc_entry);
868 remove_proc_entry(journal->j_devname, proc_jbd2_stats);
872 * Management for journal control blocks: functions to create and
873 * destroy journal_t structures, and to initialise and read existing
874 * journal blocks from disk. */
876 /* First: create and setup a journal_t object in memory. We initialise
877 * very few fields yet: that has to wait until we have created the
878 * journal structures from from scratch, or loaded them from disk. */
880 static journal_t * journal_init_common (void)
885 journal = kzalloc(sizeof(*journal), GFP_KERNEL);
889 init_waitqueue_head(&journal->j_wait_transaction_locked);
890 init_waitqueue_head(&journal->j_wait_logspace);
891 init_waitqueue_head(&journal->j_wait_done_commit);
892 init_waitqueue_head(&journal->j_wait_checkpoint);
893 init_waitqueue_head(&journal->j_wait_commit);
894 init_waitqueue_head(&journal->j_wait_updates);
895 mutex_init(&journal->j_barrier);
896 mutex_init(&journal->j_checkpoint_mutex);
897 spin_lock_init(&journal->j_revoke_lock);
898 spin_lock_init(&journal->j_list_lock);
899 rwlock_init(&journal->j_state_lock);
901 journal->j_commit_interval = (HZ * JBD2_DEFAULT_MAX_COMMIT_AGE);
902 journal->j_min_batch_time = 0;
903 journal->j_max_batch_time = 15000; /* 15ms */
905 /* The journal is marked for error until we succeed with recovery! */
906 journal->j_flags = JBD2_ABORT;
908 /* Set up a default-sized revoke table for the new mount. */
909 err = jbd2_journal_init_revoke(journal, JOURNAL_REVOKE_DEFAULT_HASH);
915 spin_lock_init(&journal->j_history_lock);
920 /* jbd2_journal_init_dev and jbd2_journal_init_inode:
922 * Create a journal structure assigned some fixed set of disk blocks to
923 * the journal. We don't actually touch those disk blocks yet, but we
924 * need to set up all of the mapping information to tell the journaling
925 * system where the journal blocks are.
930 * journal_t * jbd2_journal_init_dev() - creates and initialises a journal structure
931 * @bdev: Block device on which to create the journal
932 * @fs_dev: Device which hold journalled filesystem for this journal.
933 * @start: Block nr Start of journal.
934 * @len: Length of the journal in blocks.
935 * @blocksize: blocksize of journalling device
937 * Returns: a newly created journal_t *
939 * jbd2_journal_init_dev creates a journal which maps a fixed contiguous
940 * range of blocks on an arbitrary block device.
943 journal_t * jbd2_journal_init_dev(struct block_device *bdev,
944 struct block_device *fs_dev,
945 unsigned long long start, int len, int blocksize)
947 journal_t *journal = journal_init_common();
948 struct buffer_head *bh;
955 /* journal descriptor can store up to n blocks -bzzz */
956 journal->j_blocksize = blocksize;
957 journal->j_dev = bdev;
958 journal->j_fs_dev = fs_dev;
959 journal->j_blk_offset = start;
960 journal->j_maxlen = len;
961 bdevname(journal->j_dev, journal->j_devname);
962 p = journal->j_devname;
963 while ((p = strchr(p, '/')))
965 jbd2_stats_proc_init(journal);
966 n = journal->j_blocksize / sizeof(journal_block_tag_t);
967 journal->j_wbufsize = n;
968 journal->j_wbuf = kmalloc(n * sizeof(struct buffer_head*), GFP_KERNEL);
969 if (!journal->j_wbuf) {
970 printk(KERN_ERR "%s: Can't allocate bhs for commit thread\n",
975 bh = __getblk(journal->j_dev, start, journal->j_blocksize);
978 "%s: Cannot get buffer for journal superblock\n",
982 journal->j_sb_buffer = bh;
983 journal->j_superblock = (journal_superblock_t *)bh->b_data;
987 kfree(journal->j_wbuf);
988 jbd2_stats_proc_exit(journal);
994 * journal_t * jbd2_journal_init_inode () - creates a journal which maps to a inode.
995 * @inode: An inode to create the journal in
997 * jbd2_journal_init_inode creates a journal which maps an on-disk inode as
998 * the journal. The inode must exist already, must support bmap() and
999 * must have all data blocks preallocated.
1001 journal_t * jbd2_journal_init_inode (struct inode *inode)
1003 struct buffer_head *bh;
1004 journal_t *journal = journal_init_common();
1008 unsigned long long blocknr;
1013 journal->j_dev = journal->j_fs_dev = inode->i_sb->s_bdev;
1014 journal->j_inode = inode;
1015 bdevname(journal->j_dev, journal->j_devname);
1016 p = journal->j_devname;
1017 while ((p = strchr(p, '/')))
1019 p = journal->j_devname + strlen(journal->j_devname);
1020 sprintf(p, "-%lu", journal->j_inode->i_ino);
1022 "journal %p: inode %s/%ld, size %Ld, bits %d, blksize %ld\n",
1023 journal, inode->i_sb->s_id, inode->i_ino,
1024 (long long) inode->i_size,
1025 inode->i_sb->s_blocksize_bits, inode->i_sb->s_blocksize);
1027 journal->j_maxlen = inode->i_size >> inode->i_sb->s_blocksize_bits;
1028 journal->j_blocksize = inode->i_sb->s_blocksize;
1029 jbd2_stats_proc_init(journal);
1031 /* journal descriptor can store up to n blocks -bzzz */
1032 n = journal->j_blocksize / sizeof(journal_block_tag_t);
1033 journal->j_wbufsize = n;
1034 journal->j_wbuf = kmalloc(n * sizeof(struct buffer_head*), GFP_KERNEL);
1035 if (!journal->j_wbuf) {
1036 printk(KERN_ERR "%s: Can't allocate bhs for commit thread\n",
1041 err = jbd2_journal_bmap(journal, 0, &blocknr);
1042 /* If that failed, give up */
1044 printk(KERN_ERR "%s: Cannot locate journal superblock\n",
1049 bh = __getblk(journal->j_dev, blocknr, journal->j_blocksize);
1052 "%s: Cannot get buffer for journal superblock\n",
1056 journal->j_sb_buffer = bh;
1057 journal->j_superblock = (journal_superblock_t *)bh->b_data;
1061 kfree(journal->j_wbuf);
1062 jbd2_stats_proc_exit(journal);
1068 * If the journal init or create aborts, we need to mark the journal
1069 * superblock as being NULL to prevent the journal destroy from writing
1070 * back a bogus superblock.
1072 static void journal_fail_superblock (journal_t *journal)
1074 struct buffer_head *bh = journal->j_sb_buffer;
1076 journal->j_sb_buffer = NULL;
1080 * Given a journal_t structure, initialise the various fields for
1081 * startup of a new journaling session. We use this both when creating
1082 * a journal, and after recovering an old journal to reset it for
1086 static int journal_reset(journal_t *journal)
1088 journal_superblock_t *sb = journal->j_superblock;
1089 unsigned long long first, last;
1091 first = be32_to_cpu(sb->s_first);
1092 last = be32_to_cpu(sb->s_maxlen);
1093 if (first + JBD2_MIN_JOURNAL_BLOCKS > last + 1) {
1094 printk(KERN_ERR "JBD2: Journal too short (blocks %llu-%llu).\n",
1096 journal_fail_superblock(journal);
1100 journal->j_first = first;
1101 journal->j_last = last;
1103 journal->j_head = first;
1104 journal->j_tail = first;
1105 journal->j_free = last - first;
1107 journal->j_tail_sequence = journal->j_transaction_sequence;
1108 journal->j_commit_sequence = journal->j_transaction_sequence - 1;
1109 journal->j_commit_request = journal->j_commit_sequence;
1111 journal->j_max_transaction_buffers = journal->j_maxlen / 4;
1114 * As a special case, if the on-disk copy is already marked as needing
1115 * no recovery (s_start == 0), then we can safely defer the superblock
1116 * update until the next commit by setting JBD2_FLUSHED. This avoids
1117 * attempting a write to a potential-readonly device.
1119 if (sb->s_start == 0) {
1120 jbd_debug(1, "JBD2: Skipping superblock update on recovered sb "
1121 "(start %ld, seq %d, errno %d)\n",
1122 journal->j_tail, journal->j_tail_sequence,
1124 journal->j_flags |= JBD2_FLUSHED;
1126 /* Add the dynamic fields and write it to disk. */
1127 jbd2_journal_update_sb_log_tail(journal);
1129 return jbd2_journal_start_thread(journal);
1132 static void jbd2_write_superblock(journal_t *journal)
1134 struct buffer_head *bh = journal->j_sb_buffer;
1136 trace_jbd2_write_superblock(journal);
1137 if (buffer_write_io_error(bh)) {
1139 * Oh, dear. A previous attempt to write the journal
1140 * superblock failed. This could happen because the
1141 * USB device was yanked out. Or it could happen to
1142 * be a transient write error and maybe the block will
1143 * be remapped. Nothing we can do but to retry the
1144 * write and hope for the best.
1146 printk(KERN_ERR "JBD2: previous I/O error detected "
1147 "for journal superblock update for %s.\n",
1148 journal->j_devname);
1149 clear_buffer_write_io_error(bh);
1150 set_buffer_uptodate(bh);
1153 BUFFER_TRACE(bh, "marking dirty");
1154 mark_buffer_dirty(bh);
1155 sync_dirty_buffer(bh);
1156 if (buffer_write_io_error(bh)) {
1157 printk(KERN_ERR "JBD2: I/O error detected "
1158 "when updating journal superblock for %s.\n",
1159 journal->j_devname);
1160 clear_buffer_write_io_error(bh);
1161 set_buffer_uptodate(bh);
1166 * jbd2_journal_update_sb_log_tail() - Update log tail in journal sb on disk.
1167 * @journal: The journal to update.
1169 * Update a journal's superblock information about log tail and write it to
1170 * disk, waiting for the IO to complete.
1172 void jbd2_journal_update_sb_log_tail(journal_t *journal)
1174 journal_superblock_t *sb = journal->j_superblock;
1176 read_lock(&journal->j_state_lock);
1177 jbd_debug(1, "JBD2: updating superblock (start %ld, seq %d)\n",
1178 journal->j_tail, journal->j_tail_sequence);
1180 sb->s_sequence = cpu_to_be32(journal->j_tail_sequence);
1181 sb->s_start = cpu_to_be32(journal->j_tail);
1182 read_unlock(&journal->j_state_lock);
1184 jbd2_write_superblock(journal);
1186 /* Log is no longer empty */
1187 write_lock(&journal->j_state_lock);
1188 WARN_ON(!sb->s_sequence);
1189 journal->j_flags &= ~JBD2_FLUSHED;
1190 write_unlock(&journal->j_state_lock);
1194 * jbd2_mark_journal_empty() - Mark on disk journal as empty.
1195 * @journal: The journal to update.
1197 * Update a journal's dynamic superblock fields to show that journal is empty.
1198 * Write updated superblock to disk waiting for IO to complete.
1200 static void jbd2_mark_journal_empty(journal_t *journal)
1202 journal_superblock_t *sb = journal->j_superblock;
1204 read_lock(&journal->j_state_lock);
1205 jbd_debug(1, "JBD2: Marking journal as empty (seq %d)\n",
1206 journal->j_tail_sequence);
1208 sb->s_sequence = cpu_to_be32(journal->j_tail_sequence);
1209 sb->s_start = cpu_to_be32(0);
1210 read_unlock(&journal->j_state_lock);
1212 jbd2_write_superblock(journal);
1214 /* Log is no longer empty */
1215 write_lock(&journal->j_state_lock);
1216 journal->j_flags |= JBD2_FLUSHED;
1217 write_unlock(&journal->j_state_lock);
1222 * jbd2_journal_update_sb_errno() - Update error in the journal.
1223 * @journal: The journal to update.
1225 * Update a journal's errno. Write updated superblock to disk waiting for IO
1228 static void jbd2_journal_update_sb_errno(journal_t *journal)
1230 journal_superblock_t *sb = journal->j_superblock;
1232 read_lock(&journal->j_state_lock);
1233 jbd_debug(1, "JBD2: updating superblock error (errno %d)\n",
1235 sb->s_errno = cpu_to_be32(journal->j_errno);
1236 read_unlock(&journal->j_state_lock);
1238 jbd2_write_superblock(journal);
1242 * Read the superblock for a given journal, performing initial
1243 * validation of the format.
1245 static int journal_get_superblock(journal_t *journal)
1247 struct buffer_head *bh;
1248 journal_superblock_t *sb;
1251 bh = journal->j_sb_buffer;
1253 J_ASSERT(bh != NULL);
1254 if (!buffer_uptodate(bh)) {
1255 ll_rw_block(READ, 1, &bh);
1257 if (!buffer_uptodate(bh)) {
1259 "JBD2: IO error reading journal superblock\n");
1264 sb = journal->j_superblock;
1268 if (sb->s_header.h_magic != cpu_to_be32(JBD2_MAGIC_NUMBER) ||
1269 sb->s_blocksize != cpu_to_be32(journal->j_blocksize)) {
1270 printk(KERN_WARNING "JBD2: no valid journal superblock found\n");
1274 switch(be32_to_cpu(sb->s_header.h_blocktype)) {
1275 case JBD2_SUPERBLOCK_V1:
1276 journal->j_format_version = 1;
1278 case JBD2_SUPERBLOCK_V2:
1279 journal->j_format_version = 2;
1282 printk(KERN_WARNING "JBD2: unrecognised superblock format ID\n");
1286 if (be32_to_cpu(sb->s_maxlen) < journal->j_maxlen)
1287 journal->j_maxlen = be32_to_cpu(sb->s_maxlen);
1288 else if (be32_to_cpu(sb->s_maxlen) > journal->j_maxlen) {
1289 printk(KERN_WARNING "JBD2: journal file too short\n");
1293 if (be32_to_cpu(sb->s_first) == 0 ||
1294 be32_to_cpu(sb->s_first) >= journal->j_maxlen) {
1296 "JBD2: Invalid start block of journal: %u\n",
1297 be32_to_cpu(sb->s_first));
1304 journal_fail_superblock(journal);
1309 * Load the on-disk journal superblock and read the key fields into the
1313 static int load_superblock(journal_t *journal)
1316 journal_superblock_t *sb;
1318 err = journal_get_superblock(journal);
1322 sb = journal->j_superblock;
1324 journal->j_tail_sequence = be32_to_cpu(sb->s_sequence);
1325 journal->j_tail = be32_to_cpu(sb->s_start);
1326 journal->j_first = be32_to_cpu(sb->s_first);
1327 journal->j_last = be32_to_cpu(sb->s_maxlen);
1328 journal->j_errno = be32_to_cpu(sb->s_errno);
1335 * int jbd2_journal_load() - Read journal from disk.
1336 * @journal: Journal to act on.
1338 * Given a journal_t structure which tells us which disk blocks contain
1339 * a journal, read the journal from disk to initialise the in-memory
1342 int jbd2_journal_load(journal_t *journal)
1345 journal_superblock_t *sb;
1347 err = load_superblock(journal);
1351 sb = journal->j_superblock;
1352 /* If this is a V2 superblock, then we have to check the
1353 * features flags on it. */
1355 if (journal->j_format_version >= 2) {
1356 if ((sb->s_feature_ro_compat &
1357 ~cpu_to_be32(JBD2_KNOWN_ROCOMPAT_FEATURES)) ||
1358 (sb->s_feature_incompat &
1359 ~cpu_to_be32(JBD2_KNOWN_INCOMPAT_FEATURES))) {
1361 "JBD2: Unrecognised features on journal\n");
1367 * Create a slab for this blocksize
1369 err = jbd2_journal_create_slab(be32_to_cpu(sb->s_blocksize));
1373 /* Let the recovery code check whether it needs to recover any
1374 * data from the journal. */
1375 if (jbd2_journal_recover(journal))
1376 goto recovery_error;
1378 if (journal->j_failed_commit) {
1379 printk(KERN_ERR "JBD2: journal transaction %u on %s "
1380 "is corrupt.\n", journal->j_failed_commit,
1381 journal->j_devname);
1385 /* OK, we've finished with the dynamic journal bits:
1386 * reinitialise the dynamic contents of the superblock in memory
1387 * and reset them on disk. */
1388 if (journal_reset(journal))
1389 goto recovery_error;
1391 journal->j_flags &= ~JBD2_ABORT;
1392 journal->j_flags |= JBD2_LOADED;
1396 printk(KERN_WARNING "JBD2: recovery failed\n");
1401 * void jbd2_journal_destroy() - Release a journal_t structure.
1402 * @journal: Journal to act on.
1404 * Release a journal_t structure once it is no longer in use by the
1406 * Return <0 if we couldn't clean up the journal.
1408 int jbd2_journal_destroy(journal_t *journal)
1412 /* Wait for the commit thread to wake up and die. */
1413 journal_kill_thread(journal);
1415 /* Force a final log commit */
1416 if (journal->j_running_transaction)
1417 jbd2_journal_commit_transaction(journal);
1419 /* Force any old transactions to disk */
1421 /* Totally anal locking here... */
1422 spin_lock(&journal->j_list_lock);
1423 while (journal->j_checkpoint_transactions != NULL) {
1424 spin_unlock(&journal->j_list_lock);
1425 mutex_lock(&journal->j_checkpoint_mutex);
1426 jbd2_log_do_checkpoint(journal);
1427 mutex_unlock(&journal->j_checkpoint_mutex);
1428 spin_lock(&journal->j_list_lock);
1431 J_ASSERT(journal->j_running_transaction == NULL);
1432 J_ASSERT(journal->j_committing_transaction == NULL);
1433 J_ASSERT(journal->j_checkpoint_transactions == NULL);
1434 spin_unlock(&journal->j_list_lock);
1436 if (journal->j_sb_buffer) {
1437 if (!is_journal_aborted(journal))
1438 jbd2_mark_journal_empty(journal);
1441 brelse(journal->j_sb_buffer);
1444 if (journal->j_proc_entry)
1445 jbd2_stats_proc_exit(journal);
1446 if (journal->j_inode)
1447 iput(journal->j_inode);
1448 if (journal->j_revoke)
1449 jbd2_journal_destroy_revoke(journal);
1450 kfree(journal->j_wbuf);
1458 *int jbd2_journal_check_used_features () - Check if features specified are used.
1459 * @journal: Journal to check.
1460 * @compat: bitmask of compatible features
1461 * @ro: bitmask of features that force read-only mount
1462 * @incompat: bitmask of incompatible features
1464 * Check whether the journal uses all of a given set of
1465 * features. Return true (non-zero) if it does.
1468 int jbd2_journal_check_used_features (journal_t *journal, unsigned long compat,
1469 unsigned long ro, unsigned long incompat)
1471 journal_superblock_t *sb;
1473 if (!compat && !ro && !incompat)
1475 /* Load journal superblock if it is not loaded yet. */
1476 if (journal->j_format_version == 0 &&
1477 journal_get_superblock(journal) != 0)
1479 if (journal->j_format_version == 1)
1482 sb = journal->j_superblock;
1484 if (((be32_to_cpu(sb->s_feature_compat) & compat) == compat) &&
1485 ((be32_to_cpu(sb->s_feature_ro_compat) & ro) == ro) &&
1486 ((be32_to_cpu(sb->s_feature_incompat) & incompat) == incompat))
1493 * int jbd2_journal_check_available_features() - Check feature set in journalling layer
1494 * @journal: Journal to check.
1495 * @compat: bitmask of compatible features
1496 * @ro: bitmask of features that force read-only mount
1497 * @incompat: bitmask of incompatible features
1499 * Check whether the journaling code supports the use of
1500 * all of a given set of features on this journal. Return true
1501 * (non-zero) if it can. */
1503 int jbd2_journal_check_available_features (journal_t *journal, unsigned long compat,
1504 unsigned long ro, unsigned long incompat)
1506 if (!compat && !ro && !incompat)
1509 /* We can support any known requested features iff the
1510 * superblock is in version 2. Otherwise we fail to support any
1511 * extended sb features. */
1513 if (journal->j_format_version != 2)
1516 if ((compat & JBD2_KNOWN_COMPAT_FEATURES) == compat &&
1517 (ro & JBD2_KNOWN_ROCOMPAT_FEATURES) == ro &&
1518 (incompat & JBD2_KNOWN_INCOMPAT_FEATURES) == incompat)
1525 * int jbd2_journal_set_features () - Mark a given journal feature in the superblock
1526 * @journal: Journal to act on.
1527 * @compat: bitmask of compatible features
1528 * @ro: bitmask of features that force read-only mount
1529 * @incompat: bitmask of incompatible features
1531 * Mark a given journal feature as present on the
1532 * superblock. Returns true if the requested features could be set.
1536 int jbd2_journal_set_features (journal_t *journal, unsigned long compat,
1537 unsigned long ro, unsigned long incompat)
1539 journal_superblock_t *sb;
1541 if (jbd2_journal_check_used_features(journal, compat, ro, incompat))
1544 if (!jbd2_journal_check_available_features(journal, compat, ro, incompat))
1547 jbd_debug(1, "Setting new features 0x%lx/0x%lx/0x%lx\n",
1548 compat, ro, incompat);
1550 sb = journal->j_superblock;
1552 sb->s_feature_compat |= cpu_to_be32(compat);
1553 sb->s_feature_ro_compat |= cpu_to_be32(ro);
1554 sb->s_feature_incompat |= cpu_to_be32(incompat);
1560 * jbd2_journal_clear_features () - Clear a given journal feature in the
1562 * @journal: Journal to act on.
1563 * @compat: bitmask of compatible features
1564 * @ro: bitmask of features that force read-only mount
1565 * @incompat: bitmask of incompatible features
1567 * Clear a given journal feature as present on the
1570 void jbd2_journal_clear_features(journal_t *journal, unsigned long compat,
1571 unsigned long ro, unsigned long incompat)
1573 journal_superblock_t *sb;
1575 jbd_debug(1, "Clear features 0x%lx/0x%lx/0x%lx\n",
1576 compat, ro, incompat);
1578 sb = journal->j_superblock;
1580 sb->s_feature_compat &= ~cpu_to_be32(compat);
1581 sb->s_feature_ro_compat &= ~cpu_to_be32(ro);
1582 sb->s_feature_incompat &= ~cpu_to_be32(incompat);
1584 EXPORT_SYMBOL(jbd2_journal_clear_features);
1587 * int jbd2_journal_flush () - Flush journal
1588 * @journal: Journal to act on.
1590 * Flush all data for a given journal to disk and empty the journal.
1591 * Filesystems can use this when remounting readonly to ensure that
1592 * recovery does not need to happen on remount.
1595 int jbd2_journal_flush(journal_t *journal)
1598 transaction_t *transaction = NULL;
1600 write_lock(&journal->j_state_lock);
1602 /* Force everything buffered to the log... */
1603 if (journal->j_running_transaction) {
1604 transaction = journal->j_running_transaction;
1605 __jbd2_log_start_commit(journal, transaction->t_tid);
1606 } else if (journal->j_committing_transaction)
1607 transaction = journal->j_committing_transaction;
1609 /* Wait for the log commit to complete... */
1611 tid_t tid = transaction->t_tid;
1613 write_unlock(&journal->j_state_lock);
1614 jbd2_log_wait_commit(journal, tid);
1616 write_unlock(&journal->j_state_lock);
1619 /* ...and flush everything in the log out to disk. */
1620 spin_lock(&journal->j_list_lock);
1621 while (!err && journal->j_checkpoint_transactions != NULL) {
1622 spin_unlock(&journal->j_list_lock);
1623 mutex_lock(&journal->j_checkpoint_mutex);
1624 err = jbd2_log_do_checkpoint(journal);
1625 mutex_unlock(&journal->j_checkpoint_mutex);
1626 spin_lock(&journal->j_list_lock);
1628 spin_unlock(&journal->j_list_lock);
1630 if (is_journal_aborted(journal))
1633 jbd2_cleanup_journal_tail(journal);
1635 /* Finally, mark the journal as really needing no recovery.
1636 * This sets s_start==0 in the underlying superblock, which is
1637 * the magic code for a fully-recovered superblock. Any future
1638 * commits of data to the journal will restore the current
1640 jbd2_mark_journal_empty(journal);
1641 write_lock(&journal->j_state_lock);
1642 J_ASSERT(!journal->j_running_transaction);
1643 J_ASSERT(!journal->j_committing_transaction);
1644 J_ASSERT(!journal->j_checkpoint_transactions);
1645 J_ASSERT(journal->j_head == journal->j_tail);
1646 J_ASSERT(journal->j_tail_sequence == journal->j_transaction_sequence);
1647 write_unlock(&journal->j_state_lock);
1652 * int jbd2_journal_wipe() - Wipe journal contents
1653 * @journal: Journal to act on.
1654 * @write: flag (see below)
1656 * Wipe out all of the contents of a journal, safely. This will produce
1657 * a warning if the journal contains any valid recovery information.
1658 * Must be called between journal_init_*() and jbd2_journal_load().
1660 * If 'write' is non-zero, then we wipe out the journal on disk; otherwise
1661 * we merely suppress recovery.
1664 int jbd2_journal_wipe(journal_t *journal, int write)
1668 J_ASSERT (!(journal->j_flags & JBD2_LOADED));
1670 err = load_superblock(journal);
1674 if (!journal->j_tail)
1677 printk(KERN_WARNING "JBD2: %s recovery information on journal\n",
1678 write ? "Clearing" : "Ignoring");
1680 err = jbd2_journal_skip_recovery(journal);
1682 jbd2_mark_journal_empty(journal);
1689 * Journal abort has very specific semantics, which we describe
1690 * for journal abort.
1692 * Two internal functions, which provide abort to the jbd layer
1697 * Quick version for internal journal use (doesn't lock the journal).
1698 * Aborts hard --- we mark the abort as occurred, but do _nothing_ else,
1699 * and don't attempt to make any other journal updates.
1701 void __jbd2_journal_abort_hard(journal_t *journal)
1703 transaction_t *transaction;
1705 if (journal->j_flags & JBD2_ABORT)
1708 printk(KERN_ERR "Aborting journal on device %s.\n",
1709 journal->j_devname);
1711 write_lock(&journal->j_state_lock);
1712 journal->j_flags |= JBD2_ABORT;
1713 transaction = journal->j_running_transaction;
1715 __jbd2_log_start_commit(journal, transaction->t_tid);
1716 write_unlock(&journal->j_state_lock);
1719 /* Soft abort: record the abort error status in the journal superblock,
1720 * but don't do any other IO. */
1721 static void __journal_abort_soft (journal_t *journal, int errno)
1723 if (journal->j_flags & JBD2_ABORT)
1726 if (!journal->j_errno)
1727 journal->j_errno = errno;
1729 __jbd2_journal_abort_hard(journal);
1732 jbd2_journal_update_sb_errno(journal);
1736 * void jbd2_journal_abort () - Shutdown the journal immediately.
1737 * @journal: the journal to shutdown.
1738 * @errno: an error number to record in the journal indicating
1739 * the reason for the shutdown.
1741 * Perform a complete, immediate shutdown of the ENTIRE
1742 * journal (not of a single transaction). This operation cannot be
1743 * undone without closing and reopening the journal.
1745 * The jbd2_journal_abort function is intended to support higher level error
1746 * recovery mechanisms such as the ext2/ext3 remount-readonly error
1749 * Journal abort has very specific semantics. Any existing dirty,
1750 * unjournaled buffers in the main filesystem will still be written to
1751 * disk by bdflush, but the journaling mechanism will be suspended
1752 * immediately and no further transaction commits will be honoured.
1754 * Any dirty, journaled buffers will be written back to disk without
1755 * hitting the journal. Atomicity cannot be guaranteed on an aborted
1756 * filesystem, but we _do_ attempt to leave as much data as possible
1757 * behind for fsck to use for cleanup.
1759 * Any attempt to get a new transaction handle on a journal which is in
1760 * ABORT state will just result in an -EROFS error return. A
1761 * jbd2_journal_stop on an existing handle will return -EIO if we have
1762 * entered abort state during the update.
1764 * Recursive transactions are not disturbed by journal abort until the
1765 * final jbd2_journal_stop, which will receive the -EIO error.
1767 * Finally, the jbd2_journal_abort call allows the caller to supply an errno
1768 * which will be recorded (if possible) in the journal superblock. This
1769 * allows a client to record failure conditions in the middle of a
1770 * transaction without having to complete the transaction to record the
1771 * failure to disk. ext3_error, for example, now uses this
1774 * Errors which originate from within the journaling layer will NOT
1775 * supply an errno; a null errno implies that absolutely no further
1776 * writes are done to the journal (unless there are any already in
1781 void jbd2_journal_abort(journal_t *journal, int errno)
1783 __journal_abort_soft(journal, errno);
1787 * int jbd2_journal_errno () - returns the journal's error state.
1788 * @journal: journal to examine.
1790 * This is the errno number set with jbd2_journal_abort(), the last
1791 * time the journal was mounted - if the journal was stopped
1792 * without calling abort this will be 0.
1794 * If the journal has been aborted on this mount time -EROFS will
1797 int jbd2_journal_errno(journal_t *journal)
1801 read_lock(&journal->j_state_lock);
1802 if (journal->j_flags & JBD2_ABORT)
1805 err = journal->j_errno;
1806 read_unlock(&journal->j_state_lock);
1811 * int jbd2_journal_clear_err () - clears the journal's error state
1812 * @journal: journal to act on.
1814 * An error must be cleared or acked to take a FS out of readonly
1817 int jbd2_journal_clear_err(journal_t *journal)
1821 write_lock(&journal->j_state_lock);
1822 if (journal->j_flags & JBD2_ABORT)
1825 journal->j_errno = 0;
1826 write_unlock(&journal->j_state_lock);
1831 * void jbd2_journal_ack_err() - Ack journal err.
1832 * @journal: journal to act on.
1834 * An error must be cleared or acked to take a FS out of readonly
1837 void jbd2_journal_ack_err(journal_t *journal)
1839 write_lock(&journal->j_state_lock);
1840 if (journal->j_errno)
1841 journal->j_flags |= JBD2_ACK_ERR;
1842 write_unlock(&journal->j_state_lock);
1845 int jbd2_journal_blocks_per_page(struct inode *inode)
1847 return 1 << (PAGE_CACHE_SHIFT - inode->i_sb->s_blocksize_bits);
1851 * helper functions to deal with 32 or 64bit block numbers.
1853 size_t journal_tag_bytes(journal_t *journal)
1855 if (JBD2_HAS_INCOMPAT_FEATURE(journal, JBD2_FEATURE_INCOMPAT_64BIT))
1856 return JBD2_TAG_SIZE64;
1858 return JBD2_TAG_SIZE32;
1862 * JBD memory management
1864 * These functions are used to allocate block-sized chunks of memory
1865 * used for making copies of buffer_head data. Very often it will be
1866 * page-sized chunks of data, but sometimes it will be in
1867 * sub-page-size chunks. (For example, 16k pages on Power systems
1868 * with a 4k block file system.) For blocks smaller than a page, we
1869 * use a SLAB allocator. There are slab caches for each block size,
1870 * which are allocated at mount time, if necessary, and we only free
1871 * (all of) the slab caches when/if the jbd2 module is unloaded. For
1872 * this reason we don't need to a mutex to protect access to
1873 * jbd2_slab[] allocating or releasing memory; only in
1874 * jbd2_journal_create_slab().
1876 #define JBD2_MAX_SLABS 8
1877 static struct kmem_cache *jbd2_slab[JBD2_MAX_SLABS];
1879 static const char *jbd2_slab_names[JBD2_MAX_SLABS] = {
1880 "jbd2_1k", "jbd2_2k", "jbd2_4k", "jbd2_8k",
1881 "jbd2_16k", "jbd2_32k", "jbd2_64k", "jbd2_128k"
1885 static void jbd2_journal_destroy_slabs(void)
1889 for (i = 0; i < JBD2_MAX_SLABS; i++) {
1891 kmem_cache_destroy(jbd2_slab[i]);
1892 jbd2_slab[i] = NULL;
1896 static int jbd2_journal_create_slab(size_t size)
1898 static DEFINE_MUTEX(jbd2_slab_create_mutex);
1899 int i = order_base_2(size) - 10;
1902 if (size == PAGE_SIZE)
1905 if (i >= JBD2_MAX_SLABS)
1908 if (unlikely(i < 0))
1910 mutex_lock(&jbd2_slab_create_mutex);
1912 mutex_unlock(&jbd2_slab_create_mutex);
1913 return 0; /* Already created */
1916 slab_size = 1 << (i+10);
1917 jbd2_slab[i] = kmem_cache_create(jbd2_slab_names[i], slab_size,
1918 slab_size, 0, NULL);
1919 mutex_unlock(&jbd2_slab_create_mutex);
1920 if (!jbd2_slab[i]) {
1921 printk(KERN_EMERG "JBD2: no memory for jbd2_slab cache\n");
1927 static struct kmem_cache *get_slab(size_t size)
1929 int i = order_base_2(size) - 10;
1931 BUG_ON(i >= JBD2_MAX_SLABS);
1932 if (unlikely(i < 0))
1934 BUG_ON(jbd2_slab[i] == NULL);
1935 return jbd2_slab[i];
1938 void *jbd2_alloc(size_t size, gfp_t flags)
1942 BUG_ON(size & (size-1)); /* Must be a power of 2 */
1944 flags |= __GFP_REPEAT;
1945 if (size == PAGE_SIZE)
1946 ptr = (void *)__get_free_pages(flags, 0);
1947 else if (size > PAGE_SIZE) {
1948 int order = get_order(size);
1951 ptr = (void *)__get_free_pages(flags, order);
1953 ptr = vmalloc(size);
1955 ptr = kmem_cache_alloc(get_slab(size), flags);
1957 /* Check alignment; SLUB has gotten this wrong in the past,
1958 * and this can lead to user data corruption! */
1959 BUG_ON(((unsigned long) ptr) & (size-1));
1964 void jbd2_free(void *ptr, size_t size)
1966 if (size == PAGE_SIZE) {
1967 free_pages((unsigned long)ptr, 0);
1970 if (size > PAGE_SIZE) {
1971 int order = get_order(size);
1974 free_pages((unsigned long)ptr, order);
1979 kmem_cache_free(get_slab(size), ptr);
1983 * Journal_head storage management
1985 static struct kmem_cache *jbd2_journal_head_cache;
1986 #ifdef CONFIG_JBD2_DEBUG
1987 static atomic_t nr_journal_heads = ATOMIC_INIT(0);
1990 static int jbd2_journal_init_journal_head_cache(void)
1994 J_ASSERT(jbd2_journal_head_cache == NULL);
1995 jbd2_journal_head_cache = kmem_cache_create("jbd2_journal_head",
1996 sizeof(struct journal_head),
1998 SLAB_TEMPORARY, /* flags */
2001 if (!jbd2_journal_head_cache) {
2003 printk(KERN_EMERG "JBD2: no memory for journal_head cache\n");
2008 static void jbd2_journal_destroy_journal_head_cache(void)
2010 if (jbd2_journal_head_cache) {
2011 kmem_cache_destroy(jbd2_journal_head_cache);
2012 jbd2_journal_head_cache = NULL;
2017 * journal_head splicing and dicing
2019 static struct journal_head *journal_alloc_journal_head(void)
2021 struct journal_head *ret;
2023 #ifdef CONFIG_JBD2_DEBUG
2024 atomic_inc(&nr_journal_heads);
2026 ret = kmem_cache_alloc(jbd2_journal_head_cache, GFP_NOFS);
2028 jbd_debug(1, "out of memory for journal_head\n");
2029 pr_notice_ratelimited("ENOMEM in %s, retrying.\n", __func__);
2032 ret = kmem_cache_alloc(jbd2_journal_head_cache, GFP_NOFS);
2038 static void journal_free_journal_head(struct journal_head *jh)
2040 #ifdef CONFIG_JBD2_DEBUG
2041 atomic_dec(&nr_journal_heads);
2042 memset(jh, JBD2_POISON_FREE, sizeof(*jh));
2044 kmem_cache_free(jbd2_journal_head_cache, jh);
2048 * A journal_head is attached to a buffer_head whenever JBD has an
2049 * interest in the buffer.
2051 * Whenever a buffer has an attached journal_head, its ->b_state:BH_JBD bit
2052 * is set. This bit is tested in core kernel code where we need to take
2053 * JBD-specific actions. Testing the zeroness of ->b_private is not reliable
2056 * When a buffer has its BH_JBD bit set, its ->b_count is elevated by one.
2058 * When a buffer has its BH_JBD bit set it is immune from being released by
2059 * core kernel code, mainly via ->b_count.
2061 * A journal_head is detached from its buffer_head when the journal_head's
2062 * b_jcount reaches zero. Running transaction (b_transaction) and checkpoint
2063 * transaction (b_cp_transaction) hold their references to b_jcount.
2065 * Various places in the kernel want to attach a journal_head to a buffer_head
2066 * _before_ attaching the journal_head to a transaction. To protect the
2067 * journal_head in this situation, jbd2_journal_add_journal_head elevates the
2068 * journal_head's b_jcount refcount by one. The caller must call
2069 * jbd2_journal_put_journal_head() to undo this.
2071 * So the typical usage would be:
2073 * (Attach a journal_head if needed. Increments b_jcount)
2074 * struct journal_head *jh = jbd2_journal_add_journal_head(bh);
2076 * (Get another reference for transaction)
2077 * jbd2_journal_grab_journal_head(bh);
2078 * jh->b_transaction = xxx;
2079 * (Put original reference)
2080 * jbd2_journal_put_journal_head(jh);
2084 * Give a buffer_head a journal_head.
2088 struct journal_head *jbd2_journal_add_journal_head(struct buffer_head *bh)
2090 struct journal_head *jh;
2091 struct journal_head *new_jh = NULL;
2094 if (!buffer_jbd(bh)) {
2095 new_jh = journal_alloc_journal_head();
2096 memset(new_jh, 0, sizeof(*new_jh));
2099 jbd_lock_bh_journal_head(bh);
2100 if (buffer_jbd(bh)) {
2104 (atomic_read(&bh->b_count) > 0) ||
2105 (bh->b_page && bh->b_page->mapping));
2108 jbd_unlock_bh_journal_head(bh);
2113 new_jh = NULL; /* We consumed it */
2118 BUFFER_TRACE(bh, "added journal_head");
2121 jbd_unlock_bh_journal_head(bh);
2123 journal_free_journal_head(new_jh);
2124 return bh->b_private;
2128 * Grab a ref against this buffer_head's journal_head. If it ended up not
2129 * having a journal_head, return NULL
2131 struct journal_head *jbd2_journal_grab_journal_head(struct buffer_head *bh)
2133 struct journal_head *jh = NULL;
2135 jbd_lock_bh_journal_head(bh);
2136 if (buffer_jbd(bh)) {
2140 jbd_unlock_bh_journal_head(bh);
2144 static void __journal_remove_journal_head(struct buffer_head *bh)
2146 struct journal_head *jh = bh2jh(bh);
2148 J_ASSERT_JH(jh, jh->b_jcount >= 0);
2149 J_ASSERT_JH(jh, jh->b_transaction == NULL);
2150 J_ASSERT_JH(jh, jh->b_next_transaction == NULL);
2151 J_ASSERT_JH(jh, jh->b_cp_transaction == NULL);
2152 J_ASSERT_JH(jh, jh->b_jlist == BJ_None);
2153 J_ASSERT_BH(bh, buffer_jbd(bh));
2154 J_ASSERT_BH(bh, jh2bh(jh) == bh);
2155 BUFFER_TRACE(bh, "remove journal_head");
2156 if (jh->b_frozen_data) {
2157 printk(KERN_WARNING "%s: freeing b_frozen_data\n", __func__);
2158 jbd2_free(jh->b_frozen_data, bh->b_size);
2160 if (jh->b_committed_data) {
2161 printk(KERN_WARNING "%s: freeing b_committed_data\n", __func__);
2162 jbd2_free(jh->b_committed_data, bh->b_size);
2164 bh->b_private = NULL;
2165 jh->b_bh = NULL; /* debug, really */
2166 clear_buffer_jbd(bh);
2167 journal_free_journal_head(jh);
2171 * Drop a reference on the passed journal_head. If it fell to zero then
2172 * release the journal_head from the buffer_head.
2174 void jbd2_journal_put_journal_head(struct journal_head *jh)
2176 struct buffer_head *bh = jh2bh(jh);
2178 jbd_lock_bh_journal_head(bh);
2179 J_ASSERT_JH(jh, jh->b_jcount > 0);
2181 if (!jh->b_jcount) {
2182 __journal_remove_journal_head(bh);
2183 jbd_unlock_bh_journal_head(bh);
2186 jbd_unlock_bh_journal_head(bh);
2190 * Initialize jbd inode head
2192 void jbd2_journal_init_jbd_inode(struct jbd2_inode *jinode, struct inode *inode)
2194 jinode->i_transaction = NULL;
2195 jinode->i_next_transaction = NULL;
2196 jinode->i_vfs_inode = inode;
2197 jinode->i_flags = 0;
2198 INIT_LIST_HEAD(&jinode->i_list);
2202 * Function to be called before we start removing inode from memory (i.e.,
2203 * clear_inode() is a fine place to be called from). It removes inode from
2204 * transaction's lists.
2206 void jbd2_journal_release_jbd_inode(journal_t *journal,
2207 struct jbd2_inode *jinode)
2212 spin_lock(&journal->j_list_lock);
2213 /* Is commit writing out inode - we have to wait */
2214 if (test_bit(__JI_COMMIT_RUNNING, &jinode->i_flags)) {
2215 wait_queue_head_t *wq;
2216 DEFINE_WAIT_BIT(wait, &jinode->i_flags, __JI_COMMIT_RUNNING);
2217 wq = bit_waitqueue(&jinode->i_flags, __JI_COMMIT_RUNNING);
2218 prepare_to_wait(wq, &wait.wait, TASK_UNINTERRUPTIBLE);
2219 spin_unlock(&journal->j_list_lock);
2221 finish_wait(wq, &wait.wait);
2225 if (jinode->i_transaction) {
2226 list_del(&jinode->i_list);
2227 jinode->i_transaction = NULL;
2229 spin_unlock(&journal->j_list_lock);
2235 #ifdef CONFIG_JBD2_DEBUG
2236 u8 jbd2_journal_enable_debug __read_mostly;
2237 EXPORT_SYMBOL(jbd2_journal_enable_debug);
2239 #define JBD2_DEBUG_NAME "jbd2-debug"
2241 static struct dentry *jbd2_debugfs_dir;
2242 static struct dentry *jbd2_debug;
2244 static void __init jbd2_create_debugfs_entry(void)
2246 jbd2_debugfs_dir = debugfs_create_dir("jbd2", NULL);
2247 if (jbd2_debugfs_dir)
2248 jbd2_debug = debugfs_create_u8(JBD2_DEBUG_NAME,
2251 &jbd2_journal_enable_debug);
2254 static void __exit jbd2_remove_debugfs_entry(void)
2256 debugfs_remove(jbd2_debug);
2257 debugfs_remove(jbd2_debugfs_dir);
2262 static void __init jbd2_create_debugfs_entry(void)
2266 static void __exit jbd2_remove_debugfs_entry(void)
2272 #ifdef CONFIG_PROC_FS
2274 #define JBD2_STATS_PROC_NAME "fs/jbd2"
2276 static void __init jbd2_create_jbd_stats_proc_entry(void)
2278 proc_jbd2_stats = proc_mkdir(JBD2_STATS_PROC_NAME, NULL);
2281 static void __exit jbd2_remove_jbd_stats_proc_entry(void)
2283 if (proc_jbd2_stats)
2284 remove_proc_entry(JBD2_STATS_PROC_NAME, NULL);
2289 #define jbd2_create_jbd_stats_proc_entry() do {} while (0)
2290 #define jbd2_remove_jbd_stats_proc_entry() do {} while (0)
2294 struct kmem_cache *jbd2_handle_cache, *jbd2_inode_cache;
2296 static int __init jbd2_journal_init_handle_cache(void)
2298 jbd2_handle_cache = KMEM_CACHE(jbd2_journal_handle, SLAB_TEMPORARY);
2299 if (jbd2_handle_cache == NULL) {
2300 printk(KERN_EMERG "JBD2: failed to create handle cache\n");
2303 jbd2_inode_cache = KMEM_CACHE(jbd2_inode, 0);
2304 if (jbd2_inode_cache == NULL) {
2305 printk(KERN_EMERG "JBD2: failed to create inode cache\n");
2306 kmem_cache_destroy(jbd2_handle_cache);
2312 static void jbd2_journal_destroy_handle_cache(void)
2314 if (jbd2_handle_cache)
2315 kmem_cache_destroy(jbd2_handle_cache);
2316 if (jbd2_inode_cache)
2317 kmem_cache_destroy(jbd2_inode_cache);
2322 * Module startup and shutdown
2325 static int __init journal_init_caches(void)
2329 ret = jbd2_journal_init_revoke_caches();
2331 ret = jbd2_journal_init_journal_head_cache();
2333 ret = jbd2_journal_init_handle_cache();
2335 ret = jbd2_journal_init_transaction_cache();
2339 static void jbd2_journal_destroy_caches(void)
2341 jbd2_journal_destroy_revoke_caches();
2342 jbd2_journal_destroy_journal_head_cache();
2343 jbd2_journal_destroy_handle_cache();
2344 jbd2_journal_destroy_transaction_cache();
2345 jbd2_journal_destroy_slabs();
2348 static int __init journal_init(void)
2352 BUILD_BUG_ON(sizeof(struct journal_superblock_s) != 1024);
2354 ret = journal_init_caches();
2356 jbd2_create_debugfs_entry();
2357 jbd2_create_jbd_stats_proc_entry();
2359 jbd2_journal_destroy_caches();
2364 static void __exit journal_exit(void)
2366 #ifdef CONFIG_JBD2_DEBUG
2367 int n = atomic_read(&nr_journal_heads);
2369 printk(KERN_EMERG "JBD2: leaked %d journal_heads!\n", n);
2371 jbd2_remove_debugfs_entry();
2372 jbd2_remove_jbd_stats_proc_entry();
2373 jbd2_journal_destroy_caches();
2376 MODULE_LICENSE("GPL");
2377 module_init(journal_init);
2378 module_exit(journal_exit);