1 // SPDX-License-Identifier: GPL-2.0+
3 * linux/fs/jbd2/journal.c
5 * Written by Stephen C. Tweedie <sct@redhat.com>, 1998
7 * Copyright 1998 Red Hat corp --- All Rights Reserved
9 * Generic filesystem journal-writing code; part of the ext2fs
12 * This file manages journals: areas of disk reserved for logging
13 * transactional updates. This includes the kernel journaling thread
14 * which is responsible for scheduling updates to the log.
16 * We do not actually manage the physical storage of the journal in this
17 * file: that is left to a per-journal policy function, which allows us
18 * to store the journal within a filesystem-specified area for ext2
19 * journaling (ext2 can use a reserved inode for storing the log).
22 #include <linux/module.h>
23 #include <linux/time.h>
25 #include <linux/jbd2.h>
26 #include <linux/errno.h>
27 #include <linux/slab.h>
28 #include <linux/init.h>
30 #include <linux/freezer.h>
31 #include <linux/pagemap.h>
32 #include <linux/kthread.h>
33 #include <linux/poison.h>
34 #include <linux/proc_fs.h>
35 #include <linux/seq_file.h>
36 #include <linux/math64.h>
37 #include <linux/hash.h>
38 #include <linux/log2.h>
39 #include <linux/vmalloc.h>
40 #include <linux/backing-dev.h>
41 #include <linux/bitops.h>
42 #include <linux/ratelimit.h>
43 #include <linux/sched/mm.h>
45 #define CREATE_TRACE_POINTS
46 #include <trace/events/jbd2.h>
48 #include <linux/uaccess.h>
51 #ifdef CONFIG_JBD2_DEBUG
52 ushort jbd2_journal_enable_debug __read_mostly;
53 EXPORT_SYMBOL(jbd2_journal_enable_debug);
55 module_param_named(jbd2_debug, jbd2_journal_enable_debug, ushort, 0644);
56 MODULE_PARM_DESC(jbd2_debug, "Debugging level for jbd2");
59 EXPORT_SYMBOL(jbd2_journal_extend);
60 EXPORT_SYMBOL(jbd2_journal_stop);
61 EXPORT_SYMBOL(jbd2_journal_lock_updates);
62 EXPORT_SYMBOL(jbd2_journal_unlock_updates);
63 EXPORT_SYMBOL(jbd2_journal_get_write_access);
64 EXPORT_SYMBOL(jbd2_journal_get_create_access);
65 EXPORT_SYMBOL(jbd2_journal_get_undo_access);
66 EXPORT_SYMBOL(jbd2_journal_set_triggers);
67 EXPORT_SYMBOL(jbd2_journal_dirty_metadata);
68 EXPORT_SYMBOL(jbd2_journal_forget);
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_inode_ranged_write);
93 EXPORT_SYMBOL(jbd2_journal_inode_ranged_wait);
94 EXPORT_SYMBOL(jbd2_journal_submit_inode_data_buffers);
95 EXPORT_SYMBOL(jbd2_journal_finish_inode_data_buffers);
96 EXPORT_SYMBOL(jbd2_journal_init_jbd_inode);
97 EXPORT_SYMBOL(jbd2_journal_release_jbd_inode);
98 EXPORT_SYMBOL(jbd2_journal_begin_ordered_truncate);
99 EXPORT_SYMBOL(jbd2_inode_cache);
101 static int jbd2_journal_create_slab(size_t slab_size);
103 #ifdef CONFIG_JBD2_DEBUG
104 void __jbd2_debug(int level, const char *file, const char *func,
105 unsigned int line, const char *fmt, ...)
107 struct va_format vaf;
110 if (level > jbd2_journal_enable_debug)
115 printk(KERN_DEBUG "%s: (%s, %u): %pV", file, func, line, &vaf);
118 EXPORT_SYMBOL(__jbd2_debug);
121 /* Checksumming functions */
122 static int jbd2_verify_csum_type(journal_t *j, journal_superblock_t *sb)
124 if (!jbd2_journal_has_csum_v2or3_feature(j))
127 return sb->s_checksum_type == JBD2_CRC32C_CHKSUM;
130 static __be32 jbd2_superblock_csum(journal_t *j, journal_superblock_t *sb)
135 old_csum = sb->s_checksum;
137 csum = jbd2_chksum(j, ~0, (char *)sb, sizeof(journal_superblock_t));
138 sb->s_checksum = old_csum;
140 return cpu_to_be32(csum);
144 * Helper function used to manage commit timeouts
147 static void commit_timeout(struct timer_list *t)
149 journal_t *journal = from_timer(journal, t, j_commit_timer);
151 wake_up_process(journal->j_task);
155 * kjournald2: The main thread function used to manage a logging device
158 * This kernel thread is responsible for two things:
160 * 1) COMMIT: Every so often we need to commit the current state of the
161 * filesystem to disk. The journal thread is responsible for writing
162 * all of the metadata buffers to disk. If a fast commit is ongoing
163 * journal thread waits until it's done and then continues from
166 * 2) CHECKPOINT: We cannot reuse a used section of the log file until all
167 * of the data in that part of the log has been rewritten elsewhere on
168 * the disk. Flushing these old buffers to reclaim space in the log is
169 * known as checkpointing, and this thread is responsible for that job.
172 static int kjournald2(void *arg)
174 journal_t *journal = arg;
175 transaction_t *transaction;
178 * Set up an interval timer which can be used to trigger a commit wakeup
179 * after the commit interval expires
181 timer_setup(&journal->j_commit_timer, commit_timeout, 0);
185 /* Record that the journal thread is running */
186 journal->j_task = current;
187 wake_up(&journal->j_wait_done_commit);
190 * Make sure that no allocations from this kernel thread will ever
191 * recurse to the fs layer because we are responsible for the
192 * transaction commit and any fs involvement might get stuck waiting for
195 memalloc_nofs_save();
198 * And now, wait forever for commit wakeup events.
200 write_lock(&journal->j_state_lock);
203 if (journal->j_flags & JBD2_UNMOUNT)
206 jbd_debug(1, "commit_sequence=%u, commit_request=%u\n",
207 journal->j_commit_sequence, journal->j_commit_request);
209 if (journal->j_commit_sequence != journal->j_commit_request) {
210 jbd_debug(1, "OK, requests differ\n");
211 write_unlock(&journal->j_state_lock);
212 del_timer_sync(&journal->j_commit_timer);
213 jbd2_journal_commit_transaction(journal);
214 write_lock(&journal->j_state_lock);
218 wake_up(&journal->j_wait_done_commit);
219 if (freezing(current)) {
221 * The simpler the better. Flushing journal isn't a
222 * good idea, because that depends on threads that may
223 * be already stopped.
225 jbd_debug(1, "Now suspending kjournald2\n");
226 write_unlock(&journal->j_state_lock);
228 write_lock(&journal->j_state_lock);
231 * We assume on resume that commits are already there,
235 int should_sleep = 1;
237 prepare_to_wait(&journal->j_wait_commit, &wait,
239 if (journal->j_commit_sequence != journal->j_commit_request)
241 transaction = journal->j_running_transaction;
242 if (transaction && time_after_eq(jiffies,
243 transaction->t_expires))
245 if (journal->j_flags & JBD2_UNMOUNT)
248 write_unlock(&journal->j_state_lock);
250 write_lock(&journal->j_state_lock);
252 finish_wait(&journal->j_wait_commit, &wait);
255 jbd_debug(1, "kjournald2 wakes\n");
258 * Were we woken up by a commit wakeup event?
260 transaction = journal->j_running_transaction;
261 if (transaction && time_after_eq(jiffies, transaction->t_expires)) {
262 journal->j_commit_request = transaction->t_tid;
263 jbd_debug(1, "woke because of timeout\n");
268 del_timer_sync(&journal->j_commit_timer);
269 journal->j_task = NULL;
270 wake_up(&journal->j_wait_done_commit);
271 jbd_debug(1, "Journal thread exiting.\n");
272 write_unlock(&journal->j_state_lock);
276 static int jbd2_journal_start_thread(journal_t *journal)
278 struct task_struct *t;
280 t = kthread_run(kjournald2, journal, "jbd2/%s",
285 wait_event(journal->j_wait_done_commit, journal->j_task != NULL);
289 static void journal_kill_thread(journal_t *journal)
291 write_lock(&journal->j_state_lock);
292 journal->j_flags |= JBD2_UNMOUNT;
294 while (journal->j_task) {
295 write_unlock(&journal->j_state_lock);
296 wake_up(&journal->j_wait_commit);
297 wait_event(journal->j_wait_done_commit, journal->j_task == NULL);
298 write_lock(&journal->j_state_lock);
300 write_unlock(&journal->j_state_lock);
304 * jbd2_journal_write_metadata_buffer: write a metadata buffer to the journal.
306 * Writes a metadata buffer to a given disk block. The actual IO is not
307 * performed but a new buffer_head is constructed which labels the data
308 * to be written with the correct destination disk block.
310 * Any magic-number escaping which needs to be done will cause a
311 * copy-out here. If the buffer happens to start with the
312 * JBD2_MAGIC_NUMBER, then we can't write it to the log directly: the
313 * magic number is only written to the log for descripter blocks. In
314 * this case, we copy the data and replace the first word with 0, and we
315 * return a result code which indicates that this buffer needs to be
316 * marked as an escaped buffer in the corresponding log descriptor
317 * block. The missing word can then be restored when the block is read
320 * If the source buffer has already been modified by a new transaction
321 * since we took the last commit snapshot, we use the frozen copy of
322 * that data for IO. If we end up using the existing buffer_head's data
323 * for the write, then we have to make sure nobody modifies it while the
324 * IO is in progress. do_get_write_access() handles this.
326 * The function returns a pointer to the buffer_head to be used for IO.
334 * Bit 0 set == escape performed on the data
335 * Bit 1 set == buffer copy-out performed (kfree the data after IO)
338 int jbd2_journal_write_metadata_buffer(transaction_t *transaction,
339 struct journal_head *jh_in,
340 struct buffer_head **bh_out,
343 int need_copy_out = 0;
344 int done_copy_out = 0;
347 struct buffer_head *new_bh;
348 struct page *new_page;
349 unsigned int new_offset;
350 struct buffer_head *bh_in = jh2bh(jh_in);
351 journal_t *journal = transaction->t_journal;
354 * The buffer really shouldn't be locked: only the current committing
355 * transaction is allowed to write it, so nobody else is allowed
358 * akpm: except if we're journalling data, and write() output is
359 * also part of a shared mapping, and another thread has
360 * decided to launch a writepage() against this buffer.
362 J_ASSERT_BH(bh_in, buffer_jbddirty(bh_in));
364 new_bh = alloc_buffer_head(GFP_NOFS|__GFP_NOFAIL);
366 /* keep subsequent assertions sane */
367 atomic_set(&new_bh->b_count, 1);
369 spin_lock(&jh_in->b_state_lock);
372 * If a new transaction has already done a buffer copy-out, then
373 * we use that version of the data for the commit.
375 if (jh_in->b_frozen_data) {
377 new_page = virt_to_page(jh_in->b_frozen_data);
378 new_offset = offset_in_page(jh_in->b_frozen_data);
380 new_page = jh2bh(jh_in)->b_page;
381 new_offset = offset_in_page(jh2bh(jh_in)->b_data);
384 mapped_data = kmap_atomic(new_page);
386 * Fire data frozen trigger if data already wasn't frozen. Do this
387 * before checking for escaping, as the trigger may modify the magic
388 * offset. If a copy-out happens afterwards, it will have the correct
389 * data in the buffer.
392 jbd2_buffer_frozen_trigger(jh_in, mapped_data + new_offset,
398 if (*((__be32 *)(mapped_data + new_offset)) ==
399 cpu_to_be32(JBD2_MAGIC_NUMBER)) {
403 kunmap_atomic(mapped_data);
406 * Do we need to do a data copy?
408 if (need_copy_out && !done_copy_out) {
411 spin_unlock(&jh_in->b_state_lock);
412 tmp = jbd2_alloc(bh_in->b_size, GFP_NOFS);
417 spin_lock(&jh_in->b_state_lock);
418 if (jh_in->b_frozen_data) {
419 jbd2_free(tmp, bh_in->b_size);
423 jh_in->b_frozen_data = tmp;
424 mapped_data = kmap_atomic(new_page);
425 memcpy(tmp, mapped_data + new_offset, bh_in->b_size);
426 kunmap_atomic(mapped_data);
428 new_page = virt_to_page(tmp);
429 new_offset = offset_in_page(tmp);
433 * This isn't strictly necessary, as we're using frozen
434 * data for the escaping, but it keeps consistency with
435 * b_frozen_data usage.
437 jh_in->b_frozen_triggers = jh_in->b_triggers;
441 * Did we need to do an escaping? Now we've done all the
442 * copying, we can finally do so.
445 mapped_data = kmap_atomic(new_page);
446 *((unsigned int *)(mapped_data + new_offset)) = 0;
447 kunmap_atomic(mapped_data);
450 set_bh_page(new_bh, new_page, new_offset);
451 new_bh->b_size = bh_in->b_size;
452 new_bh->b_bdev = journal->j_dev;
453 new_bh->b_blocknr = blocknr;
454 new_bh->b_private = bh_in;
455 set_buffer_mapped(new_bh);
456 set_buffer_dirty(new_bh);
461 * The to-be-written buffer needs to get moved to the io queue,
462 * and the original buffer whose contents we are shadowing or
463 * copying is moved to the transaction's shadow queue.
465 JBUFFER_TRACE(jh_in, "file as BJ_Shadow");
466 spin_lock(&journal->j_list_lock);
467 __jbd2_journal_file_buffer(jh_in, transaction, BJ_Shadow);
468 spin_unlock(&journal->j_list_lock);
469 set_buffer_shadow(bh_in);
470 spin_unlock(&jh_in->b_state_lock);
472 return do_escape | (done_copy_out << 1);
476 * Allocation code for the journal file. Manage the space left in the
477 * journal, so that we can begin checkpointing when appropriate.
481 * Called with j_state_lock locked for writing.
482 * Returns true if a transaction commit was started.
484 int __jbd2_log_start_commit(journal_t *journal, tid_t target)
486 /* Return if the txn has already requested to be committed */
487 if (journal->j_commit_request == target)
491 * The only transaction we can possibly wait upon is the
492 * currently running transaction (if it exists). Otherwise,
493 * the target tid must be an old one.
495 if (journal->j_running_transaction &&
496 journal->j_running_transaction->t_tid == target) {
498 * We want a new commit: OK, mark the request and wakeup the
499 * commit thread. We do _not_ do the commit ourselves.
502 journal->j_commit_request = target;
503 jbd_debug(1, "JBD2: requesting commit %u/%u\n",
504 journal->j_commit_request,
505 journal->j_commit_sequence);
506 journal->j_running_transaction->t_requested = jiffies;
507 wake_up(&journal->j_wait_commit);
509 } else if (!tid_geq(journal->j_commit_request, target))
510 /* This should never happen, but if it does, preserve
511 the evidence before kjournald goes into a loop and
512 increments j_commit_sequence beyond all recognition. */
513 WARN_ONCE(1, "JBD2: bad log_start_commit: %u %u %u %u\n",
514 journal->j_commit_request,
515 journal->j_commit_sequence,
516 target, journal->j_running_transaction ?
517 journal->j_running_transaction->t_tid : 0);
521 int jbd2_log_start_commit(journal_t *journal, tid_t tid)
525 write_lock(&journal->j_state_lock);
526 ret = __jbd2_log_start_commit(journal, tid);
527 write_unlock(&journal->j_state_lock);
532 * Force and wait any uncommitted transactions. We can only force the running
533 * transaction if we don't have an active handle, otherwise, we will deadlock.
534 * Returns: <0 in case of error,
535 * 0 if nothing to commit,
536 * 1 if transaction was successfully committed.
538 static int __jbd2_journal_force_commit(journal_t *journal)
540 transaction_t *transaction = NULL;
542 int need_to_start = 0, ret = 0;
544 read_lock(&journal->j_state_lock);
545 if (journal->j_running_transaction && !current->journal_info) {
546 transaction = journal->j_running_transaction;
547 if (!tid_geq(journal->j_commit_request, transaction->t_tid))
549 } else if (journal->j_committing_transaction)
550 transaction = journal->j_committing_transaction;
553 /* Nothing to commit */
554 read_unlock(&journal->j_state_lock);
557 tid = transaction->t_tid;
558 read_unlock(&journal->j_state_lock);
560 jbd2_log_start_commit(journal, tid);
561 ret = jbd2_log_wait_commit(journal, tid);
569 * jbd2_journal_force_commit_nested - Force and wait upon a commit if the
570 * calling process is not within transaction.
572 * @journal: journal to force
573 * Returns true if progress was made.
575 * This is used for forcing out undo-protected data which contains
576 * bitmaps, when the fs is running out of space.
578 int jbd2_journal_force_commit_nested(journal_t *journal)
582 ret = __jbd2_journal_force_commit(journal);
587 * jbd2_journal_force_commit() - force any uncommitted transactions
588 * @journal: journal to force
590 * Caller want unconditional commit. We can only force the running transaction
591 * if we don't have an active handle, otherwise, we will deadlock.
593 int jbd2_journal_force_commit(journal_t *journal)
597 J_ASSERT(!current->journal_info);
598 ret = __jbd2_journal_force_commit(journal);
605 * Start a commit of the current running transaction (if any). Returns true
606 * if a transaction is going to be committed (or is currently already
607 * committing), and fills its tid in at *ptid
609 int jbd2_journal_start_commit(journal_t *journal, tid_t *ptid)
613 write_lock(&journal->j_state_lock);
614 if (journal->j_running_transaction) {
615 tid_t tid = journal->j_running_transaction->t_tid;
617 __jbd2_log_start_commit(journal, tid);
618 /* There's a running transaction and we've just made sure
619 * it's commit has been scheduled. */
623 } else if (journal->j_committing_transaction) {
625 * If commit has been started, then we have to wait for
626 * completion of that transaction.
629 *ptid = journal->j_committing_transaction->t_tid;
632 write_unlock(&journal->j_state_lock);
637 * Return 1 if a given transaction has not yet sent barrier request
638 * connected with a transaction commit. If 0 is returned, transaction
639 * may or may not have sent the barrier. Used to avoid sending barrier
640 * twice in common cases.
642 int jbd2_trans_will_send_data_barrier(journal_t *journal, tid_t tid)
645 transaction_t *commit_trans;
647 if (!(journal->j_flags & JBD2_BARRIER))
649 read_lock(&journal->j_state_lock);
650 /* Transaction already committed? */
651 if (tid_geq(journal->j_commit_sequence, tid))
653 commit_trans = journal->j_committing_transaction;
654 if (!commit_trans || commit_trans->t_tid != tid) {
659 * Transaction is being committed and we already proceeded to
660 * submitting a flush to fs partition?
662 if (journal->j_fs_dev != journal->j_dev) {
663 if (!commit_trans->t_need_data_flush ||
664 commit_trans->t_state >= T_COMMIT_DFLUSH)
667 if (commit_trans->t_state >= T_COMMIT_JFLUSH)
672 read_unlock(&journal->j_state_lock);
675 EXPORT_SYMBOL(jbd2_trans_will_send_data_barrier);
678 * Wait for a specified commit to complete.
679 * The caller may not hold the journal lock.
681 int jbd2_log_wait_commit(journal_t *journal, tid_t tid)
685 read_lock(&journal->j_state_lock);
686 #ifdef CONFIG_PROVE_LOCKING
688 * Some callers make sure transaction is already committing and in that
689 * case we cannot block on open handles anymore. So don't warn in that
692 if (tid_gt(tid, journal->j_commit_sequence) &&
693 (!journal->j_committing_transaction ||
694 journal->j_committing_transaction->t_tid != tid)) {
695 read_unlock(&journal->j_state_lock);
696 jbd2_might_wait_for_commit(journal);
697 read_lock(&journal->j_state_lock);
700 #ifdef CONFIG_JBD2_DEBUG
701 if (!tid_geq(journal->j_commit_request, tid)) {
703 "%s: error: j_commit_request=%u, tid=%u\n",
704 __func__, journal->j_commit_request, tid);
707 while (tid_gt(tid, journal->j_commit_sequence)) {
708 jbd_debug(1, "JBD2: want %u, j_commit_sequence=%u\n",
709 tid, journal->j_commit_sequence);
710 read_unlock(&journal->j_state_lock);
711 wake_up(&journal->j_wait_commit);
712 wait_event(journal->j_wait_done_commit,
713 !tid_gt(tid, journal->j_commit_sequence));
714 read_lock(&journal->j_state_lock);
716 read_unlock(&journal->j_state_lock);
718 if (unlikely(is_journal_aborted(journal)))
724 * Start a fast commit. If there's an ongoing fast or full commit wait for
725 * it to complete. Returns 0 if a new fast commit was started. Returns -EALREADY
726 * if a fast commit is not needed, either because there's an already a commit
727 * going on or this tid has already been committed. Returns -EINVAL if no jbd2
728 * commit has yet been performed.
730 int jbd2_fc_begin_commit(journal_t *journal, tid_t tid)
732 if (unlikely(is_journal_aborted(journal)))
735 * Fast commits only allowed if at least one full commit has
738 if (!journal->j_stats.ts_tid)
741 write_lock(&journal->j_state_lock);
742 if (tid <= journal->j_commit_sequence) {
743 write_unlock(&journal->j_state_lock);
747 if (journal->j_flags & JBD2_FULL_COMMIT_ONGOING ||
748 (journal->j_flags & JBD2_FAST_COMMIT_ONGOING)) {
751 prepare_to_wait(&journal->j_fc_wait, &wait,
752 TASK_UNINTERRUPTIBLE);
753 write_unlock(&journal->j_state_lock);
755 finish_wait(&journal->j_fc_wait, &wait);
758 journal->j_flags |= JBD2_FAST_COMMIT_ONGOING;
759 write_unlock(&journal->j_state_lock);
760 jbd2_journal_lock_updates(journal);
764 EXPORT_SYMBOL(jbd2_fc_begin_commit);
767 * Stop a fast commit. If fallback is set, this function starts commit of
768 * TID tid before any other fast commit can start.
770 static int __jbd2_fc_end_commit(journal_t *journal, tid_t tid, bool fallback)
772 jbd2_journal_unlock_updates(journal);
773 if (journal->j_fc_cleanup_callback)
774 journal->j_fc_cleanup_callback(journal, 0);
775 write_lock(&journal->j_state_lock);
776 journal->j_flags &= ~JBD2_FAST_COMMIT_ONGOING;
778 journal->j_flags |= JBD2_FULL_COMMIT_ONGOING;
779 write_unlock(&journal->j_state_lock);
780 wake_up(&journal->j_fc_wait);
782 return jbd2_complete_transaction(journal, tid);
786 int jbd2_fc_end_commit(journal_t *journal)
788 return __jbd2_fc_end_commit(journal, 0, false);
790 EXPORT_SYMBOL(jbd2_fc_end_commit);
792 int jbd2_fc_end_commit_fallback(journal_t *journal)
796 read_lock(&journal->j_state_lock);
797 tid = journal->j_running_transaction ?
798 journal->j_running_transaction->t_tid : 0;
799 read_unlock(&journal->j_state_lock);
800 return __jbd2_fc_end_commit(journal, tid, true);
802 EXPORT_SYMBOL(jbd2_fc_end_commit_fallback);
804 /* Return 1 when transaction with given tid has already committed. */
805 int jbd2_transaction_committed(journal_t *journal, tid_t tid)
809 read_lock(&journal->j_state_lock);
810 if (journal->j_running_transaction &&
811 journal->j_running_transaction->t_tid == tid)
813 if (journal->j_committing_transaction &&
814 journal->j_committing_transaction->t_tid == tid)
816 read_unlock(&journal->j_state_lock);
819 EXPORT_SYMBOL(jbd2_transaction_committed);
822 * When this function returns the transaction corresponding to tid
823 * will be completed. If the transaction has currently running, start
824 * committing that transaction before waiting for it to complete. If
825 * the transaction id is stale, it is by definition already completed,
826 * so just return SUCCESS.
828 int jbd2_complete_transaction(journal_t *journal, tid_t tid)
830 int need_to_wait = 1;
832 read_lock(&journal->j_state_lock);
833 if (journal->j_running_transaction &&
834 journal->j_running_transaction->t_tid == tid) {
835 if (journal->j_commit_request != tid) {
836 /* transaction not yet started, so request it */
837 read_unlock(&journal->j_state_lock);
838 jbd2_log_start_commit(journal, tid);
841 } else if (!(journal->j_committing_transaction &&
842 journal->j_committing_transaction->t_tid == tid))
844 read_unlock(&journal->j_state_lock);
848 return jbd2_log_wait_commit(journal, tid);
850 EXPORT_SYMBOL(jbd2_complete_transaction);
853 * Log buffer allocation routines:
856 int jbd2_journal_next_log_block(journal_t *journal, unsigned long long *retp)
858 unsigned long blocknr;
860 write_lock(&journal->j_state_lock);
861 J_ASSERT(journal->j_free > 1);
863 blocknr = journal->j_head;
866 if (journal->j_head == journal->j_last)
867 journal->j_head = journal->j_first;
868 write_unlock(&journal->j_state_lock);
869 return jbd2_journal_bmap(journal, blocknr, retp);
872 /* Map one fast commit buffer for use by the file system */
873 int jbd2_fc_get_buf(journal_t *journal, struct buffer_head **bh_out)
875 unsigned long long pblock;
876 unsigned long blocknr;
878 struct buffer_head *bh;
883 if (journal->j_fc_off + journal->j_fc_first < journal->j_fc_last) {
884 fc_off = journal->j_fc_off;
885 blocknr = journal->j_fc_first + fc_off;
894 ret = jbd2_journal_bmap(journal, blocknr, &pblock);
898 bh = __getblk(journal->j_dev, pblock, journal->j_blocksize);
903 journal->j_fc_wbuf[fc_off] = bh;
909 EXPORT_SYMBOL(jbd2_fc_get_buf);
912 * Wait on fast commit buffers that were allocated by jbd2_fc_get_buf
915 int jbd2_fc_wait_bufs(journal_t *journal, int num_blks)
917 struct buffer_head *bh;
920 j_fc_off = journal->j_fc_off;
923 * Wait in reverse order to minimize chances of us being woken up before
924 * all IOs have completed
926 for (i = j_fc_off - 1; i >= j_fc_off - num_blks; i--) {
927 bh = journal->j_fc_wbuf[i];
930 * Update j_fc_off so jbd2_fc_release_bufs can release remain
933 if (unlikely(!buffer_uptodate(bh))) {
934 journal->j_fc_off = i + 1;
938 journal->j_fc_wbuf[i] = NULL;
943 EXPORT_SYMBOL(jbd2_fc_wait_bufs);
946 * Wait on fast commit buffers that were allocated by jbd2_fc_get_buf
949 int jbd2_fc_release_bufs(journal_t *journal)
951 struct buffer_head *bh;
954 j_fc_off = journal->j_fc_off;
957 * Wait in reverse order to minimize chances of us being woken up before
958 * all IOs have completed
960 for (i = j_fc_off - 1; i >= 0; i--) {
961 bh = journal->j_fc_wbuf[i];
965 journal->j_fc_wbuf[i] = NULL;
970 EXPORT_SYMBOL(jbd2_fc_release_bufs);
973 * Conversion of logical to physical block numbers for the journal
975 * On external journals the journal blocks are identity-mapped, so
976 * this is a no-op. If needed, we can use j_blk_offset - everything is
979 int jbd2_journal_bmap(journal_t *journal, unsigned long blocknr,
980 unsigned long long *retp)
983 unsigned long long ret;
986 if (journal->j_inode) {
988 ret = bmap(journal->j_inode, &block);
991 printk(KERN_ALERT "%s: journal block not found "
992 "at offset %lu on %s\n",
993 __func__, blocknr, journal->j_devname);
995 jbd2_journal_abort(journal, err);
1001 *retp = blocknr; /* +journal->j_blk_offset */
1007 * We play buffer_head aliasing tricks to write data/metadata blocks to
1008 * the journal without copying their contents, but for journal
1009 * descriptor blocks we do need to generate bona fide buffers.
1011 * After the caller of jbd2_journal_get_descriptor_buffer() has finished modifying
1012 * the buffer's contents they really should run flush_dcache_page(bh->b_page).
1013 * But we don't bother doing that, so there will be coherency problems with
1014 * mmaps of blockdevs which hold live JBD-controlled filesystems.
1016 struct buffer_head *
1017 jbd2_journal_get_descriptor_buffer(transaction_t *transaction, int type)
1019 journal_t *journal = transaction->t_journal;
1020 struct buffer_head *bh;
1021 unsigned long long blocknr;
1022 journal_header_t *header;
1025 err = jbd2_journal_next_log_block(journal, &blocknr);
1030 bh = __getblk(journal->j_dev, blocknr, journal->j_blocksize);
1033 atomic_dec(&transaction->t_outstanding_credits);
1035 memset(bh->b_data, 0, journal->j_blocksize);
1036 header = (journal_header_t *)bh->b_data;
1037 header->h_magic = cpu_to_be32(JBD2_MAGIC_NUMBER);
1038 header->h_blocktype = cpu_to_be32(type);
1039 header->h_sequence = cpu_to_be32(transaction->t_tid);
1040 set_buffer_uptodate(bh);
1042 BUFFER_TRACE(bh, "return this buffer");
1046 void jbd2_descriptor_block_csum_set(journal_t *j, struct buffer_head *bh)
1048 struct jbd2_journal_block_tail *tail;
1051 if (!jbd2_journal_has_csum_v2or3(j))
1054 tail = (struct jbd2_journal_block_tail *)(bh->b_data + j->j_blocksize -
1055 sizeof(struct jbd2_journal_block_tail));
1056 tail->t_checksum = 0;
1057 csum = jbd2_chksum(j, j->j_csum_seed, bh->b_data, j->j_blocksize);
1058 tail->t_checksum = cpu_to_be32(csum);
1062 * Return tid of the oldest transaction in the journal and block in the journal
1063 * where the transaction starts.
1065 * If the journal is now empty, return which will be the next transaction ID
1066 * we will write and where will that transaction start.
1068 * The return value is 0 if journal tail cannot be pushed any further, 1 if
1071 int jbd2_journal_get_log_tail(journal_t *journal, tid_t *tid,
1072 unsigned long *block)
1074 transaction_t *transaction;
1077 read_lock(&journal->j_state_lock);
1078 spin_lock(&journal->j_list_lock);
1079 transaction = journal->j_checkpoint_transactions;
1081 *tid = transaction->t_tid;
1082 *block = transaction->t_log_start;
1083 } else if ((transaction = journal->j_committing_transaction) != NULL) {
1084 *tid = transaction->t_tid;
1085 *block = transaction->t_log_start;
1086 } else if ((transaction = journal->j_running_transaction) != NULL) {
1087 *tid = transaction->t_tid;
1088 *block = journal->j_head;
1090 *tid = journal->j_transaction_sequence;
1091 *block = journal->j_head;
1093 ret = tid_gt(*tid, journal->j_tail_sequence);
1094 spin_unlock(&journal->j_list_lock);
1095 read_unlock(&journal->j_state_lock);
1101 * Update information in journal structure and in on disk journal superblock
1102 * about log tail. This function does not check whether information passed in
1103 * really pushes log tail further. It's responsibility of the caller to make
1104 * sure provided log tail information is valid (e.g. by holding
1105 * j_checkpoint_mutex all the time between computing log tail and calling this
1106 * function as is the case with jbd2_cleanup_journal_tail()).
1108 * Requires j_checkpoint_mutex
1110 int __jbd2_update_log_tail(journal_t *journal, tid_t tid, unsigned long block)
1112 unsigned long freed;
1115 BUG_ON(!mutex_is_locked(&journal->j_checkpoint_mutex));
1118 * We cannot afford for write to remain in drive's caches since as
1119 * soon as we update j_tail, next transaction can start reusing journal
1120 * space and if we lose sb update during power failure we'd replay
1121 * old transaction with possibly newly overwritten data.
1123 ret = jbd2_journal_update_sb_log_tail(journal, tid, block,
1124 REQ_SYNC | REQ_FUA);
1128 write_lock(&journal->j_state_lock);
1129 freed = block - journal->j_tail;
1130 if (block < journal->j_tail)
1131 freed += journal->j_last - journal->j_first;
1133 trace_jbd2_update_log_tail(journal, tid, block, freed);
1135 "Cleaning journal tail from %u to %u (offset %lu), "
1137 journal->j_tail_sequence, tid, block, freed);
1139 journal->j_free += freed;
1140 journal->j_tail_sequence = tid;
1141 journal->j_tail = block;
1142 write_unlock(&journal->j_state_lock);
1149 * This is a variation of __jbd2_update_log_tail which checks for validity of
1150 * provided log tail and locks j_checkpoint_mutex. So it is safe against races
1151 * with other threads updating log tail.
1153 void jbd2_update_log_tail(journal_t *journal, tid_t tid, unsigned long block)
1155 mutex_lock_io(&journal->j_checkpoint_mutex);
1156 if (tid_gt(tid, journal->j_tail_sequence))
1157 __jbd2_update_log_tail(journal, tid, block);
1158 mutex_unlock(&journal->j_checkpoint_mutex);
1161 struct jbd2_stats_proc_session {
1163 struct transaction_stats_s *stats;
1168 static void *jbd2_seq_info_start(struct seq_file *seq, loff_t *pos)
1170 return *pos ? NULL : SEQ_START_TOKEN;
1173 static void *jbd2_seq_info_next(struct seq_file *seq, void *v, loff_t *pos)
1179 static int jbd2_seq_info_show(struct seq_file *seq, void *v)
1181 struct jbd2_stats_proc_session *s = seq->private;
1183 if (v != SEQ_START_TOKEN)
1185 seq_printf(seq, "%lu transactions (%lu requested), "
1186 "each up to %u blocks\n",
1187 s->stats->ts_tid, s->stats->ts_requested,
1188 s->journal->j_max_transaction_buffers);
1189 if (s->stats->ts_tid == 0)
1191 seq_printf(seq, "average: \n %ums waiting for transaction\n",
1192 jiffies_to_msecs(s->stats->run.rs_wait / s->stats->ts_tid));
1193 seq_printf(seq, " %ums request delay\n",
1194 (s->stats->ts_requested == 0) ? 0 :
1195 jiffies_to_msecs(s->stats->run.rs_request_delay /
1196 s->stats->ts_requested));
1197 seq_printf(seq, " %ums running transaction\n",
1198 jiffies_to_msecs(s->stats->run.rs_running / s->stats->ts_tid));
1199 seq_printf(seq, " %ums transaction was being locked\n",
1200 jiffies_to_msecs(s->stats->run.rs_locked / s->stats->ts_tid));
1201 seq_printf(seq, " %ums flushing data (in ordered mode)\n",
1202 jiffies_to_msecs(s->stats->run.rs_flushing / s->stats->ts_tid));
1203 seq_printf(seq, " %ums logging transaction\n",
1204 jiffies_to_msecs(s->stats->run.rs_logging / s->stats->ts_tid));
1205 seq_printf(seq, " %lluus average transaction commit time\n",
1206 div_u64(s->journal->j_average_commit_time, 1000));
1207 seq_printf(seq, " %lu handles per transaction\n",
1208 s->stats->run.rs_handle_count / s->stats->ts_tid);
1209 seq_printf(seq, " %lu blocks per transaction\n",
1210 s->stats->run.rs_blocks / s->stats->ts_tid);
1211 seq_printf(seq, " %lu logged blocks per transaction\n",
1212 s->stats->run.rs_blocks_logged / s->stats->ts_tid);
1216 static void jbd2_seq_info_stop(struct seq_file *seq, void *v)
1220 static const struct seq_operations jbd2_seq_info_ops = {
1221 .start = jbd2_seq_info_start,
1222 .next = jbd2_seq_info_next,
1223 .stop = jbd2_seq_info_stop,
1224 .show = jbd2_seq_info_show,
1227 static int jbd2_seq_info_open(struct inode *inode, struct file *file)
1229 journal_t *journal = PDE_DATA(inode);
1230 struct jbd2_stats_proc_session *s;
1233 s = kmalloc(sizeof(*s), GFP_KERNEL);
1236 size = sizeof(struct transaction_stats_s);
1237 s->stats = kmalloc(size, GFP_KERNEL);
1238 if (s->stats == NULL) {
1242 spin_lock(&journal->j_history_lock);
1243 memcpy(s->stats, &journal->j_stats, size);
1244 s->journal = journal;
1245 spin_unlock(&journal->j_history_lock);
1247 rc = seq_open(file, &jbd2_seq_info_ops);
1249 struct seq_file *m = file->private_data;
1259 static int jbd2_seq_info_release(struct inode *inode, struct file *file)
1261 struct seq_file *seq = file->private_data;
1262 struct jbd2_stats_proc_session *s = seq->private;
1265 return seq_release(inode, file);
1268 static const struct proc_ops jbd2_info_proc_ops = {
1269 .proc_open = jbd2_seq_info_open,
1270 .proc_read = seq_read,
1271 .proc_lseek = seq_lseek,
1272 .proc_release = jbd2_seq_info_release,
1275 static struct proc_dir_entry *proc_jbd2_stats;
1277 static void jbd2_stats_proc_init(journal_t *journal)
1279 journal->j_proc_entry = proc_mkdir(journal->j_devname, proc_jbd2_stats);
1280 if (journal->j_proc_entry) {
1281 proc_create_data("info", S_IRUGO, journal->j_proc_entry,
1282 &jbd2_info_proc_ops, journal);
1286 static void jbd2_stats_proc_exit(journal_t *journal)
1288 remove_proc_entry("info", journal->j_proc_entry);
1289 remove_proc_entry(journal->j_devname, proc_jbd2_stats);
1292 /* Minimum size of descriptor tag */
1293 static int jbd2_min_tag_size(void)
1296 * Tag with 32-bit block numbers does not use last four bytes of the
1299 return sizeof(journal_block_tag_t) - 4;
1303 * Management for journal control blocks: functions to create and
1304 * destroy journal_t structures, and to initialise and read existing
1305 * journal blocks from disk. */
1307 /* First: create and setup a journal_t object in memory. We initialise
1308 * very few fields yet: that has to wait until we have created the
1309 * journal structures from from scratch, or loaded them from disk. */
1311 static journal_t *journal_init_common(struct block_device *bdev,
1312 struct block_device *fs_dev,
1313 unsigned long long start, int len, int blocksize)
1315 static struct lock_class_key jbd2_trans_commit_key;
1318 struct buffer_head *bh;
1321 journal = kzalloc(sizeof(*journal), GFP_KERNEL);
1325 init_waitqueue_head(&journal->j_wait_transaction_locked);
1326 init_waitqueue_head(&journal->j_wait_done_commit);
1327 init_waitqueue_head(&journal->j_wait_commit);
1328 init_waitqueue_head(&journal->j_wait_updates);
1329 init_waitqueue_head(&journal->j_wait_reserved);
1330 init_waitqueue_head(&journal->j_fc_wait);
1331 mutex_init(&journal->j_abort_mutex);
1332 mutex_init(&journal->j_barrier);
1333 mutex_init(&journal->j_checkpoint_mutex);
1334 spin_lock_init(&journal->j_revoke_lock);
1335 spin_lock_init(&journal->j_list_lock);
1336 rwlock_init(&journal->j_state_lock);
1338 journal->j_commit_interval = (HZ * JBD2_DEFAULT_MAX_COMMIT_AGE);
1339 journal->j_min_batch_time = 0;
1340 journal->j_max_batch_time = 15000; /* 15ms */
1341 atomic_set(&journal->j_reserved_credits, 0);
1343 /* The journal is marked for error until we succeed with recovery! */
1344 journal->j_flags = JBD2_ABORT;
1346 /* Set up a default-sized revoke table for the new mount. */
1347 err = jbd2_journal_init_revoke(journal, JOURNAL_REVOKE_DEFAULT_HASH);
1351 spin_lock_init(&journal->j_history_lock);
1353 lockdep_init_map(&journal->j_trans_commit_map, "jbd2_handle",
1354 &jbd2_trans_commit_key, 0);
1356 /* journal descriptor can store up to n blocks -bzzz */
1357 journal->j_blocksize = blocksize;
1358 journal->j_dev = bdev;
1359 journal->j_fs_dev = fs_dev;
1360 journal->j_blk_offset = start;
1361 journal->j_total_len = len;
1362 /* We need enough buffers to write out full descriptor block. */
1363 n = journal->j_blocksize / jbd2_min_tag_size();
1364 journal->j_wbufsize = n;
1365 journal->j_fc_wbuf = NULL;
1366 journal->j_wbuf = kmalloc_array(n, sizeof(struct buffer_head *),
1368 if (!journal->j_wbuf)
1371 bh = getblk_unmovable(journal->j_dev, start, journal->j_blocksize);
1373 pr_err("%s: Cannot get buffer for journal superblock\n",
1377 journal->j_sb_buffer = bh;
1378 journal->j_superblock = (journal_superblock_t *)bh->b_data;
1383 kfree(journal->j_wbuf);
1384 jbd2_journal_destroy_revoke(journal);
1389 /* jbd2_journal_init_dev and jbd2_journal_init_inode:
1391 * Create a journal structure assigned some fixed set of disk blocks to
1392 * the journal. We don't actually touch those disk blocks yet, but we
1393 * need to set up all of the mapping information to tell the journaling
1394 * system where the journal blocks are.
1399 * journal_t * jbd2_journal_init_dev() - creates and initialises a journal structure
1400 * @bdev: Block device on which to create the journal
1401 * @fs_dev: Device which hold journalled filesystem for this journal.
1402 * @start: Block nr Start of journal.
1403 * @len: Length of the journal in blocks.
1404 * @blocksize: blocksize of journalling device
1406 * Returns: a newly created journal_t *
1408 * jbd2_journal_init_dev creates a journal which maps a fixed contiguous
1409 * range of blocks on an arbitrary block device.
1412 journal_t *jbd2_journal_init_dev(struct block_device *bdev,
1413 struct block_device *fs_dev,
1414 unsigned long long start, int len, int blocksize)
1418 journal = journal_init_common(bdev, fs_dev, start, len, blocksize);
1422 bdevname(journal->j_dev, journal->j_devname);
1423 strreplace(journal->j_devname, '/', '!');
1424 jbd2_stats_proc_init(journal);
1430 * journal_t * jbd2_journal_init_inode () - creates a journal which maps to a inode.
1431 * @inode: An inode to create the journal in
1433 * jbd2_journal_init_inode creates a journal which maps an on-disk inode as
1434 * the journal. The inode must exist already, must support bmap() and
1435 * must have all data blocks preallocated.
1437 journal_t *jbd2_journal_init_inode(struct inode *inode)
1445 err = bmap(inode, &blocknr);
1447 if (err || !blocknr) {
1448 pr_err("%s: Cannot locate journal superblock\n",
1453 jbd_debug(1, "JBD2: inode %s/%ld, size %lld, bits %d, blksize %ld\n",
1454 inode->i_sb->s_id, inode->i_ino, (long long) inode->i_size,
1455 inode->i_sb->s_blocksize_bits, inode->i_sb->s_blocksize);
1457 journal = journal_init_common(inode->i_sb->s_bdev, inode->i_sb->s_bdev,
1458 blocknr, inode->i_size >> inode->i_sb->s_blocksize_bits,
1459 inode->i_sb->s_blocksize);
1463 journal->j_inode = inode;
1464 bdevname(journal->j_dev, journal->j_devname);
1465 p = strreplace(journal->j_devname, '/', '!');
1466 sprintf(p, "-%lu", journal->j_inode->i_ino);
1467 jbd2_stats_proc_init(journal);
1473 * If the journal init or create aborts, we need to mark the journal
1474 * superblock as being NULL to prevent the journal destroy from writing
1475 * back a bogus superblock.
1477 static void journal_fail_superblock(journal_t *journal)
1479 struct buffer_head *bh = journal->j_sb_buffer;
1481 journal->j_sb_buffer = NULL;
1485 * Given a journal_t structure, initialise the various fields for
1486 * startup of a new journaling session. We use this both when creating
1487 * a journal, and after recovering an old journal to reset it for
1491 static int journal_reset(journal_t *journal)
1493 journal_superblock_t *sb = journal->j_superblock;
1494 unsigned long long first, last;
1496 first = be32_to_cpu(sb->s_first);
1497 last = be32_to_cpu(sb->s_maxlen);
1498 if (first + JBD2_MIN_JOURNAL_BLOCKS > last + 1) {
1499 printk(KERN_ERR "JBD2: Journal too short (blocks %llu-%llu).\n",
1501 journal_fail_superblock(journal);
1505 journal->j_first = first;
1506 journal->j_last = last;
1508 journal->j_head = journal->j_first;
1509 journal->j_tail = journal->j_first;
1510 journal->j_free = journal->j_last - journal->j_first;
1512 journal->j_tail_sequence = journal->j_transaction_sequence;
1513 journal->j_commit_sequence = journal->j_transaction_sequence - 1;
1514 journal->j_commit_request = journal->j_commit_sequence;
1516 journal->j_max_transaction_buffers = jbd2_journal_get_max_txn_bufs(journal);
1519 * Now that journal recovery is done, turn fast commits off here. This
1520 * way, if fast commit was enabled before the crash but if now FS has
1521 * disabled it, we don't enable fast commits.
1523 jbd2_clear_feature_fast_commit(journal);
1526 * As a special case, if the on-disk copy is already marked as needing
1527 * no recovery (s_start == 0), then we can safely defer the superblock
1528 * update until the next commit by setting JBD2_FLUSHED. This avoids
1529 * attempting a write to a potential-readonly device.
1531 if (sb->s_start == 0) {
1532 jbd_debug(1, "JBD2: Skipping superblock update on recovered sb "
1533 "(start %ld, seq %u, errno %d)\n",
1534 journal->j_tail, journal->j_tail_sequence,
1536 journal->j_flags |= JBD2_FLUSHED;
1538 /* Lock here to make assertions happy... */
1539 mutex_lock_io(&journal->j_checkpoint_mutex);
1541 * Update log tail information. We use REQ_FUA since new
1542 * transaction will start reusing journal space and so we
1543 * must make sure information about current log tail is on
1546 jbd2_journal_update_sb_log_tail(journal,
1547 journal->j_tail_sequence,
1549 REQ_SYNC | REQ_FUA);
1550 mutex_unlock(&journal->j_checkpoint_mutex);
1552 return jbd2_journal_start_thread(journal);
1556 * This function expects that the caller will have locked the journal
1557 * buffer head, and will return with it unlocked
1559 static int jbd2_write_superblock(journal_t *journal, int write_flags)
1561 struct buffer_head *bh = journal->j_sb_buffer;
1562 journal_superblock_t *sb = journal->j_superblock;
1565 /* Buffer got discarded which means block device got invalidated */
1566 if (!buffer_mapped(bh)) {
1571 if (!(journal->j_flags & JBD2_BARRIER))
1572 write_flags &= ~(REQ_FUA | REQ_PREFLUSH);
1574 trace_jbd2_write_superblock(journal, write_flags);
1576 if (buffer_write_io_error(bh)) {
1578 * Oh, dear. A previous attempt to write the journal
1579 * superblock failed. This could happen because the
1580 * USB device was yanked out. Or it could happen to
1581 * be a transient write error and maybe the block will
1582 * be remapped. Nothing we can do but to retry the
1583 * write and hope for the best.
1585 printk(KERN_ERR "JBD2: previous I/O error detected "
1586 "for journal superblock update for %s.\n",
1587 journal->j_devname);
1588 clear_buffer_write_io_error(bh);
1589 set_buffer_uptodate(bh);
1591 if (jbd2_journal_has_csum_v2or3(journal))
1592 sb->s_checksum = jbd2_superblock_csum(journal, sb);
1594 bh->b_end_io = end_buffer_write_sync;
1595 ret = submit_bh(REQ_OP_WRITE, write_flags, bh);
1597 if (buffer_write_io_error(bh)) {
1598 clear_buffer_write_io_error(bh);
1599 set_buffer_uptodate(bh);
1603 printk(KERN_ERR "JBD2: Error %d detected when updating "
1604 "journal superblock for %s.\n", ret,
1605 journal->j_devname);
1606 if (!is_journal_aborted(journal))
1607 jbd2_journal_abort(journal, ret);
1614 * jbd2_journal_update_sb_log_tail() - Update log tail in journal sb on disk.
1615 * @journal: The journal to update.
1616 * @tail_tid: TID of the new transaction at the tail of the log
1617 * @tail_block: The first block of the transaction at the tail of the log
1618 * @write_op: With which operation should we write the journal sb
1620 * Update a journal's superblock information about log tail and write it to
1621 * disk, waiting for the IO to complete.
1623 int jbd2_journal_update_sb_log_tail(journal_t *journal, tid_t tail_tid,
1624 unsigned long tail_block, int write_op)
1626 journal_superblock_t *sb = journal->j_superblock;
1629 if (is_journal_aborted(journal))
1632 BUG_ON(!mutex_is_locked(&journal->j_checkpoint_mutex));
1633 jbd_debug(1, "JBD2: updating superblock (start %lu, seq %u)\n",
1634 tail_block, tail_tid);
1636 lock_buffer(journal->j_sb_buffer);
1637 sb->s_sequence = cpu_to_be32(tail_tid);
1638 sb->s_start = cpu_to_be32(tail_block);
1640 ret = jbd2_write_superblock(journal, write_op);
1644 /* Log is no longer empty */
1645 write_lock(&journal->j_state_lock);
1646 WARN_ON(!sb->s_sequence);
1647 journal->j_flags &= ~JBD2_FLUSHED;
1648 write_unlock(&journal->j_state_lock);
1655 * jbd2_mark_journal_empty() - Mark on disk journal as empty.
1656 * @journal: The journal to update.
1657 * @write_op: With which operation should we write the journal sb
1659 * Update a journal's dynamic superblock fields to show that journal is empty.
1660 * Write updated superblock to disk waiting for IO to complete.
1662 static void jbd2_mark_journal_empty(journal_t *journal, int write_op)
1664 journal_superblock_t *sb = journal->j_superblock;
1665 bool had_fast_commit = false;
1667 BUG_ON(!mutex_is_locked(&journal->j_checkpoint_mutex));
1668 lock_buffer(journal->j_sb_buffer);
1669 if (sb->s_start == 0) { /* Is it already empty? */
1670 unlock_buffer(journal->j_sb_buffer);
1674 jbd_debug(1, "JBD2: Marking journal as empty (seq %u)\n",
1675 journal->j_tail_sequence);
1677 sb->s_sequence = cpu_to_be32(journal->j_tail_sequence);
1678 sb->s_start = cpu_to_be32(0);
1679 if (jbd2_has_feature_fast_commit(journal)) {
1681 * When journal is clean, no need to commit fast commit flag and
1682 * make file system incompatible with older kernels.
1684 jbd2_clear_feature_fast_commit(journal);
1685 had_fast_commit = true;
1688 jbd2_write_superblock(journal, write_op);
1690 if (had_fast_commit)
1691 jbd2_set_feature_fast_commit(journal);
1693 /* Log is no longer empty */
1694 write_lock(&journal->j_state_lock);
1695 journal->j_flags |= JBD2_FLUSHED;
1696 write_unlock(&journal->j_state_lock);
1701 * jbd2_journal_update_sb_errno() - Update error in the journal.
1702 * @journal: The journal to update.
1704 * Update a journal's errno. Write updated superblock to disk waiting for IO
1707 void jbd2_journal_update_sb_errno(journal_t *journal)
1709 journal_superblock_t *sb = journal->j_superblock;
1712 lock_buffer(journal->j_sb_buffer);
1713 errcode = journal->j_errno;
1714 if (errcode == -ESHUTDOWN)
1716 jbd_debug(1, "JBD2: updating superblock error (errno %d)\n", errcode);
1717 sb->s_errno = cpu_to_be32(errcode);
1719 jbd2_write_superblock(journal, REQ_SYNC | REQ_FUA);
1721 EXPORT_SYMBOL(jbd2_journal_update_sb_errno);
1723 static int journal_revoke_records_per_block(journal_t *journal)
1726 int space = journal->j_blocksize - sizeof(jbd2_journal_revoke_header_t);
1728 if (jbd2_has_feature_64bit(journal))
1733 if (jbd2_journal_has_csum_v2or3(journal))
1734 space -= sizeof(struct jbd2_journal_block_tail);
1735 return space / record_size;
1739 * Read the superblock for a given journal, performing initial
1740 * validation of the format.
1742 static int journal_get_superblock(journal_t *journal)
1744 struct buffer_head *bh;
1745 journal_superblock_t *sb;
1748 bh = journal->j_sb_buffer;
1750 J_ASSERT(bh != NULL);
1751 if (!buffer_uptodate(bh)) {
1752 ll_rw_block(REQ_OP_READ, 0, 1, &bh);
1754 if (!buffer_uptodate(bh)) {
1756 "JBD2: IO error reading journal superblock\n");
1761 if (buffer_verified(bh))
1764 sb = journal->j_superblock;
1768 if (sb->s_header.h_magic != cpu_to_be32(JBD2_MAGIC_NUMBER) ||
1769 sb->s_blocksize != cpu_to_be32(journal->j_blocksize)) {
1770 printk(KERN_WARNING "JBD2: no valid journal superblock found\n");
1774 switch(be32_to_cpu(sb->s_header.h_blocktype)) {
1775 case JBD2_SUPERBLOCK_V1:
1776 journal->j_format_version = 1;
1778 case JBD2_SUPERBLOCK_V2:
1779 journal->j_format_version = 2;
1782 printk(KERN_WARNING "JBD2: unrecognised superblock format ID\n");
1786 if (be32_to_cpu(sb->s_maxlen) < journal->j_total_len)
1787 journal->j_total_len = be32_to_cpu(sb->s_maxlen);
1788 else if (be32_to_cpu(sb->s_maxlen) > journal->j_total_len) {
1789 printk(KERN_WARNING "JBD2: journal file too short\n");
1793 if (be32_to_cpu(sb->s_first) == 0 ||
1794 be32_to_cpu(sb->s_first) >= journal->j_total_len) {
1796 "JBD2: Invalid start block of journal: %u\n",
1797 be32_to_cpu(sb->s_first));
1801 if (jbd2_has_feature_csum2(journal) &&
1802 jbd2_has_feature_csum3(journal)) {
1803 /* Can't have checksum v2 and v3 at the same time! */
1804 printk(KERN_ERR "JBD2: Can't enable checksumming v2 and v3 "
1805 "at the same time!\n");
1809 if (jbd2_journal_has_csum_v2or3_feature(journal) &&
1810 jbd2_has_feature_checksum(journal)) {
1811 /* Can't have checksum v1 and v2 on at the same time! */
1812 printk(KERN_ERR "JBD2: Can't enable checksumming v1 and v2/3 "
1813 "at the same time!\n");
1817 if (!jbd2_verify_csum_type(journal, sb)) {
1818 printk(KERN_ERR "JBD2: Unknown checksum type\n");
1822 /* Load the checksum driver */
1823 if (jbd2_journal_has_csum_v2or3_feature(journal)) {
1824 journal->j_chksum_driver = crypto_alloc_shash("crc32c", 0, 0);
1825 if (IS_ERR(journal->j_chksum_driver)) {
1826 printk(KERN_ERR "JBD2: Cannot load crc32c driver.\n");
1827 err = PTR_ERR(journal->j_chksum_driver);
1828 journal->j_chksum_driver = NULL;
1833 if (jbd2_journal_has_csum_v2or3(journal)) {
1834 /* Check superblock checksum */
1835 if (sb->s_checksum != jbd2_superblock_csum(journal, sb)) {
1836 printk(KERN_ERR "JBD2: journal checksum error\n");
1841 /* Precompute checksum seed for all metadata */
1842 journal->j_csum_seed = jbd2_chksum(journal, ~0, sb->s_uuid,
1843 sizeof(sb->s_uuid));
1846 journal->j_revoke_records_per_block =
1847 journal_revoke_records_per_block(journal);
1848 set_buffer_verified(bh);
1853 journal_fail_superblock(journal);
1858 * Load the on-disk journal superblock and read the key fields into the
1862 static int load_superblock(journal_t *journal)
1865 journal_superblock_t *sb;
1868 err = journal_get_superblock(journal);
1872 sb = journal->j_superblock;
1874 journal->j_tail_sequence = be32_to_cpu(sb->s_sequence);
1875 journal->j_tail = be32_to_cpu(sb->s_start);
1876 journal->j_first = be32_to_cpu(sb->s_first);
1877 journal->j_errno = be32_to_cpu(sb->s_errno);
1878 journal->j_last = be32_to_cpu(sb->s_maxlen);
1880 if (jbd2_has_feature_fast_commit(journal)) {
1881 journal->j_fc_last = be32_to_cpu(sb->s_maxlen);
1882 num_fc_blocks = be32_to_cpu(sb->s_num_fc_blks);
1884 num_fc_blocks = JBD2_MIN_FC_BLOCKS;
1885 if (journal->j_last - num_fc_blocks >= JBD2_MIN_JOURNAL_BLOCKS)
1886 journal->j_last = journal->j_fc_last - num_fc_blocks;
1887 journal->j_fc_first = journal->j_last + 1;
1888 journal->j_fc_off = 0;
1896 * jbd2_journal_load() - Read journal from disk.
1897 * @journal: Journal to act on.
1899 * Given a journal_t structure which tells us which disk blocks contain
1900 * a journal, read the journal from disk to initialise the in-memory
1903 int jbd2_journal_load(journal_t *journal)
1906 journal_superblock_t *sb;
1908 err = load_superblock(journal);
1912 sb = journal->j_superblock;
1913 /* If this is a V2 superblock, then we have to check the
1914 * features flags on it. */
1916 if (journal->j_format_version >= 2) {
1917 if ((sb->s_feature_ro_compat &
1918 ~cpu_to_be32(JBD2_KNOWN_ROCOMPAT_FEATURES)) ||
1919 (sb->s_feature_incompat &
1920 ~cpu_to_be32(JBD2_KNOWN_INCOMPAT_FEATURES))) {
1922 "JBD2: Unrecognised features on journal\n");
1928 * Create a slab for this blocksize
1930 err = jbd2_journal_create_slab(be32_to_cpu(sb->s_blocksize));
1934 /* Let the recovery code check whether it needs to recover any
1935 * data from the journal. */
1936 if (jbd2_journal_recover(journal))
1937 goto recovery_error;
1939 if (journal->j_failed_commit) {
1940 printk(KERN_ERR "JBD2: journal transaction %u on %s "
1941 "is corrupt.\n", journal->j_failed_commit,
1942 journal->j_devname);
1943 return -EFSCORRUPTED;
1946 * clear JBD2_ABORT flag initialized in journal_init_common
1947 * here to update log tail information with the newest seq.
1949 journal->j_flags &= ~JBD2_ABORT;
1951 /* OK, we've finished with the dynamic journal bits:
1952 * reinitialise the dynamic contents of the superblock in memory
1953 * and reset them on disk. */
1954 if (journal_reset(journal))
1955 goto recovery_error;
1957 journal->j_flags |= JBD2_LOADED;
1961 printk(KERN_WARNING "JBD2: recovery failed\n");
1966 * jbd2_journal_destroy() - Release a journal_t structure.
1967 * @journal: Journal to act on.
1969 * Release a journal_t structure once it is no longer in use by the
1971 * Return <0 if we couldn't clean up the journal.
1973 int jbd2_journal_destroy(journal_t *journal)
1977 /* Wait for the commit thread to wake up and die. */
1978 journal_kill_thread(journal);
1980 /* Force a final log commit */
1981 if (journal->j_running_transaction)
1982 jbd2_journal_commit_transaction(journal);
1984 /* Force any old transactions to disk */
1986 /* Totally anal locking here... */
1987 spin_lock(&journal->j_list_lock);
1988 while (journal->j_checkpoint_transactions != NULL) {
1989 spin_unlock(&journal->j_list_lock);
1990 mutex_lock_io(&journal->j_checkpoint_mutex);
1991 err = jbd2_log_do_checkpoint(journal);
1992 mutex_unlock(&journal->j_checkpoint_mutex);
1994 * If checkpointing failed, just free the buffers to avoid
1998 jbd2_journal_destroy_checkpoint(journal);
1999 spin_lock(&journal->j_list_lock);
2002 spin_lock(&journal->j_list_lock);
2005 J_ASSERT(journal->j_running_transaction == NULL);
2006 J_ASSERT(journal->j_committing_transaction == NULL);
2007 J_ASSERT(journal->j_checkpoint_transactions == NULL);
2008 spin_unlock(&journal->j_list_lock);
2010 if (journal->j_sb_buffer) {
2011 if (!is_journal_aborted(journal)) {
2012 mutex_lock_io(&journal->j_checkpoint_mutex);
2014 write_lock(&journal->j_state_lock);
2015 journal->j_tail_sequence =
2016 ++journal->j_transaction_sequence;
2017 write_unlock(&journal->j_state_lock);
2019 jbd2_mark_journal_empty(journal,
2020 REQ_SYNC | REQ_PREFLUSH | REQ_FUA);
2021 mutex_unlock(&journal->j_checkpoint_mutex);
2024 brelse(journal->j_sb_buffer);
2027 if (journal->j_proc_entry)
2028 jbd2_stats_proc_exit(journal);
2029 iput(journal->j_inode);
2030 if (journal->j_revoke)
2031 jbd2_journal_destroy_revoke(journal);
2032 if (journal->j_chksum_driver)
2033 crypto_free_shash(journal->j_chksum_driver);
2034 kfree(journal->j_fc_wbuf);
2035 kfree(journal->j_wbuf);
2043 * jbd2_journal_check_used_features() - Check if features specified are used.
2044 * @journal: Journal to check.
2045 * @compat: bitmask of compatible features
2046 * @ro: bitmask of features that force read-only mount
2047 * @incompat: bitmask of incompatible features
2049 * Check whether the journal uses all of a given set of
2050 * features. Return true (non-zero) if it does.
2053 int jbd2_journal_check_used_features(journal_t *journal, unsigned long compat,
2054 unsigned long ro, unsigned long incompat)
2056 journal_superblock_t *sb;
2058 if (!compat && !ro && !incompat)
2060 /* Load journal superblock if it is not loaded yet. */
2061 if (journal->j_format_version == 0 &&
2062 journal_get_superblock(journal) != 0)
2064 if (journal->j_format_version == 1)
2067 sb = journal->j_superblock;
2069 if (((be32_to_cpu(sb->s_feature_compat) & compat) == compat) &&
2070 ((be32_to_cpu(sb->s_feature_ro_compat) & ro) == ro) &&
2071 ((be32_to_cpu(sb->s_feature_incompat) & incompat) == incompat))
2078 * jbd2_journal_check_available_features() - Check feature set in journalling layer
2079 * @journal: Journal to check.
2080 * @compat: bitmask of compatible features
2081 * @ro: bitmask of features that force read-only mount
2082 * @incompat: bitmask of incompatible features
2084 * Check whether the journaling code supports the use of
2085 * all of a given set of features on this journal. Return true
2086 * (non-zero) if it can. */
2088 int jbd2_journal_check_available_features(journal_t *journal, unsigned long compat,
2089 unsigned long ro, unsigned long incompat)
2091 if (!compat && !ro && !incompat)
2094 /* We can support any known requested features iff the
2095 * superblock is in version 2. Otherwise we fail to support any
2096 * extended sb features. */
2098 if (journal->j_format_version != 2)
2101 if ((compat & JBD2_KNOWN_COMPAT_FEATURES) == compat &&
2102 (ro & JBD2_KNOWN_ROCOMPAT_FEATURES) == ro &&
2103 (incompat & JBD2_KNOWN_INCOMPAT_FEATURES) == incompat)
2110 jbd2_journal_initialize_fast_commit(journal_t *journal)
2112 journal_superblock_t *sb = journal->j_superblock;
2113 unsigned long long num_fc_blks;
2115 num_fc_blks = be32_to_cpu(sb->s_num_fc_blks);
2116 if (num_fc_blks == 0)
2117 num_fc_blks = JBD2_MIN_FC_BLOCKS;
2118 if (journal->j_last - num_fc_blks < JBD2_MIN_JOURNAL_BLOCKS)
2121 /* Are we called twice? */
2122 WARN_ON(journal->j_fc_wbuf != NULL);
2123 journal->j_fc_wbuf = kmalloc_array(num_fc_blks,
2124 sizeof(struct buffer_head *), GFP_KERNEL);
2125 if (!journal->j_fc_wbuf)
2128 journal->j_fc_wbufsize = num_fc_blks;
2129 journal->j_fc_last = journal->j_last;
2130 journal->j_last = journal->j_fc_last - num_fc_blks;
2131 journal->j_fc_first = journal->j_last + 1;
2132 journal->j_fc_off = 0;
2133 journal->j_free = journal->j_last - journal->j_first;
2134 journal->j_max_transaction_buffers =
2135 jbd2_journal_get_max_txn_bufs(journal);
2141 * jbd2_journal_set_features() - Mark a given journal feature in the superblock
2142 * @journal: Journal to act on.
2143 * @compat: bitmask of compatible features
2144 * @ro: bitmask of features that force read-only mount
2145 * @incompat: bitmask of incompatible features
2147 * Mark a given journal feature as present on the
2148 * superblock. Returns true if the requested features could be set.
2152 int jbd2_journal_set_features(journal_t *journal, unsigned long compat,
2153 unsigned long ro, unsigned long incompat)
2155 #define INCOMPAT_FEATURE_ON(f) \
2156 ((incompat & (f)) && !(sb->s_feature_incompat & cpu_to_be32(f)))
2157 #define COMPAT_FEATURE_ON(f) \
2158 ((compat & (f)) && !(sb->s_feature_compat & cpu_to_be32(f)))
2159 journal_superblock_t *sb;
2161 if (jbd2_journal_check_used_features(journal, compat, ro, incompat))
2164 if (!jbd2_journal_check_available_features(journal, compat, ro, incompat))
2167 /* If enabling v2 checksums, turn on v3 instead */
2168 if (incompat & JBD2_FEATURE_INCOMPAT_CSUM_V2) {
2169 incompat &= ~JBD2_FEATURE_INCOMPAT_CSUM_V2;
2170 incompat |= JBD2_FEATURE_INCOMPAT_CSUM_V3;
2173 /* Asking for checksumming v3 and v1? Only give them v3. */
2174 if (incompat & JBD2_FEATURE_INCOMPAT_CSUM_V3 &&
2175 compat & JBD2_FEATURE_COMPAT_CHECKSUM)
2176 compat &= ~JBD2_FEATURE_COMPAT_CHECKSUM;
2178 jbd_debug(1, "Setting new features 0x%lx/0x%lx/0x%lx\n",
2179 compat, ro, incompat);
2181 sb = journal->j_superblock;
2183 if (incompat & JBD2_FEATURE_INCOMPAT_FAST_COMMIT) {
2184 if (jbd2_journal_initialize_fast_commit(journal)) {
2185 pr_err("JBD2: Cannot enable fast commits.\n");
2190 /* Load the checksum driver if necessary */
2191 if ((journal->j_chksum_driver == NULL) &&
2192 INCOMPAT_FEATURE_ON(JBD2_FEATURE_INCOMPAT_CSUM_V3)) {
2193 journal->j_chksum_driver = crypto_alloc_shash("crc32c", 0, 0);
2194 if (IS_ERR(journal->j_chksum_driver)) {
2195 printk(KERN_ERR "JBD2: Cannot load crc32c driver.\n");
2196 journal->j_chksum_driver = NULL;
2199 /* Precompute checksum seed for all metadata */
2200 journal->j_csum_seed = jbd2_chksum(journal, ~0, sb->s_uuid,
2201 sizeof(sb->s_uuid));
2204 lock_buffer(journal->j_sb_buffer);
2206 /* If enabling v3 checksums, update superblock */
2207 if (INCOMPAT_FEATURE_ON(JBD2_FEATURE_INCOMPAT_CSUM_V3)) {
2208 sb->s_checksum_type = JBD2_CRC32C_CHKSUM;
2209 sb->s_feature_compat &=
2210 ~cpu_to_be32(JBD2_FEATURE_COMPAT_CHECKSUM);
2213 /* If enabling v1 checksums, downgrade superblock */
2214 if (COMPAT_FEATURE_ON(JBD2_FEATURE_COMPAT_CHECKSUM))
2215 sb->s_feature_incompat &=
2216 ~cpu_to_be32(JBD2_FEATURE_INCOMPAT_CSUM_V2 |
2217 JBD2_FEATURE_INCOMPAT_CSUM_V3);
2219 sb->s_feature_compat |= cpu_to_be32(compat);
2220 sb->s_feature_ro_compat |= cpu_to_be32(ro);
2221 sb->s_feature_incompat |= cpu_to_be32(incompat);
2222 unlock_buffer(journal->j_sb_buffer);
2223 journal->j_revoke_records_per_block =
2224 journal_revoke_records_per_block(journal);
2227 #undef COMPAT_FEATURE_ON
2228 #undef INCOMPAT_FEATURE_ON
2232 * jbd2_journal_clear_features() - Clear a given journal feature in the
2234 * @journal: Journal to act on.
2235 * @compat: bitmask of compatible features
2236 * @ro: bitmask of features that force read-only mount
2237 * @incompat: bitmask of incompatible features
2239 * Clear a given journal feature as present on the
2242 void jbd2_journal_clear_features(journal_t *journal, unsigned long compat,
2243 unsigned long ro, unsigned long incompat)
2245 journal_superblock_t *sb;
2247 jbd_debug(1, "Clear features 0x%lx/0x%lx/0x%lx\n",
2248 compat, ro, incompat);
2250 sb = journal->j_superblock;
2252 sb->s_feature_compat &= ~cpu_to_be32(compat);
2253 sb->s_feature_ro_compat &= ~cpu_to_be32(ro);
2254 sb->s_feature_incompat &= ~cpu_to_be32(incompat);
2255 journal->j_revoke_records_per_block =
2256 journal_revoke_records_per_block(journal);
2258 EXPORT_SYMBOL(jbd2_journal_clear_features);
2261 * jbd2_journal_flush() - Flush journal
2262 * @journal: Journal to act on.
2264 * Flush all data for a given journal to disk and empty the journal.
2265 * Filesystems can use this when remounting readonly to ensure that
2266 * recovery does not need to happen on remount.
2269 int jbd2_journal_flush(journal_t *journal)
2272 transaction_t *transaction = NULL;
2274 write_lock(&journal->j_state_lock);
2276 /* Force everything buffered to the log... */
2277 if (journal->j_running_transaction) {
2278 transaction = journal->j_running_transaction;
2279 __jbd2_log_start_commit(journal, transaction->t_tid);
2280 } else if (journal->j_committing_transaction)
2281 transaction = journal->j_committing_transaction;
2283 /* Wait for the log commit to complete... */
2285 tid_t tid = transaction->t_tid;
2287 write_unlock(&journal->j_state_lock);
2288 jbd2_log_wait_commit(journal, tid);
2290 write_unlock(&journal->j_state_lock);
2293 /* ...and flush everything in the log out to disk. */
2294 spin_lock(&journal->j_list_lock);
2295 while (!err && journal->j_checkpoint_transactions != NULL) {
2296 spin_unlock(&journal->j_list_lock);
2297 mutex_lock_io(&journal->j_checkpoint_mutex);
2298 err = jbd2_log_do_checkpoint(journal);
2299 mutex_unlock(&journal->j_checkpoint_mutex);
2300 spin_lock(&journal->j_list_lock);
2302 spin_unlock(&journal->j_list_lock);
2304 if (is_journal_aborted(journal))
2307 mutex_lock_io(&journal->j_checkpoint_mutex);
2309 err = jbd2_cleanup_journal_tail(journal);
2311 mutex_unlock(&journal->j_checkpoint_mutex);
2317 /* Finally, mark the journal as really needing no recovery.
2318 * This sets s_start==0 in the underlying superblock, which is
2319 * the magic code for a fully-recovered superblock. Any future
2320 * commits of data to the journal will restore the current
2322 jbd2_mark_journal_empty(journal, REQ_SYNC | REQ_FUA);
2323 mutex_unlock(&journal->j_checkpoint_mutex);
2324 write_lock(&journal->j_state_lock);
2325 J_ASSERT(!journal->j_running_transaction);
2326 J_ASSERT(!journal->j_committing_transaction);
2327 J_ASSERT(!journal->j_checkpoint_transactions);
2328 J_ASSERT(journal->j_head == journal->j_tail);
2329 J_ASSERT(journal->j_tail_sequence == journal->j_transaction_sequence);
2330 write_unlock(&journal->j_state_lock);
2336 * jbd2_journal_wipe() - Wipe journal contents
2337 * @journal: Journal to act on.
2338 * @write: flag (see below)
2340 * Wipe out all of the contents of a journal, safely. This will produce
2341 * a warning if the journal contains any valid recovery information.
2342 * Must be called between journal_init_*() and jbd2_journal_load().
2344 * If 'write' is non-zero, then we wipe out the journal on disk; otherwise
2345 * we merely suppress recovery.
2348 int jbd2_journal_wipe(journal_t *journal, int write)
2352 J_ASSERT (!(journal->j_flags & JBD2_LOADED));
2354 err = load_superblock(journal);
2358 if (!journal->j_tail)
2361 printk(KERN_WARNING "JBD2: %s recovery information on journal\n",
2362 write ? "Clearing" : "Ignoring");
2364 err = jbd2_journal_skip_recovery(journal);
2366 /* Lock to make assertions happy... */
2367 mutex_lock_io(&journal->j_checkpoint_mutex);
2368 jbd2_mark_journal_empty(journal, REQ_SYNC | REQ_FUA);
2369 mutex_unlock(&journal->j_checkpoint_mutex);
2377 * jbd2_journal_abort () - Shutdown the journal immediately.
2378 * @journal: the journal to shutdown.
2379 * @errno: an error number to record in the journal indicating
2380 * the reason for the shutdown.
2382 * Perform a complete, immediate shutdown of the ENTIRE
2383 * journal (not of a single transaction). This operation cannot be
2384 * undone without closing and reopening the journal.
2386 * The jbd2_journal_abort function is intended to support higher level error
2387 * recovery mechanisms such as the ext2/ext3 remount-readonly error
2390 * Journal abort has very specific semantics. Any existing dirty,
2391 * unjournaled buffers in the main filesystem will still be written to
2392 * disk by bdflush, but the journaling mechanism will be suspended
2393 * immediately and no further transaction commits will be honoured.
2395 * Any dirty, journaled buffers will be written back to disk without
2396 * hitting the journal. Atomicity cannot be guaranteed on an aborted
2397 * filesystem, but we _do_ attempt to leave as much data as possible
2398 * behind for fsck to use for cleanup.
2400 * Any attempt to get a new transaction handle on a journal which is in
2401 * ABORT state will just result in an -EROFS error return. A
2402 * jbd2_journal_stop on an existing handle will return -EIO if we have
2403 * entered abort state during the update.
2405 * Recursive transactions are not disturbed by journal abort until the
2406 * final jbd2_journal_stop, which will receive the -EIO error.
2408 * Finally, the jbd2_journal_abort call allows the caller to supply an errno
2409 * which will be recorded (if possible) in the journal superblock. This
2410 * allows a client to record failure conditions in the middle of a
2411 * transaction without having to complete the transaction to record the
2412 * failure to disk. ext3_error, for example, now uses this
2417 void jbd2_journal_abort(journal_t *journal, int errno)
2419 transaction_t *transaction;
2422 * Lock the aborting procedure until everything is done, this avoid
2423 * races between filesystem's error handling flow (e.g. ext4_abort()),
2424 * ensure panic after the error info is written into journal's
2427 mutex_lock(&journal->j_abort_mutex);
2429 * ESHUTDOWN always takes precedence because a file system check
2430 * caused by any other journal abort error is not required after
2431 * a shutdown triggered.
2433 write_lock(&journal->j_state_lock);
2434 if (journal->j_flags & JBD2_ABORT) {
2435 int old_errno = journal->j_errno;
2437 write_unlock(&journal->j_state_lock);
2438 if (old_errno != -ESHUTDOWN && errno == -ESHUTDOWN) {
2439 journal->j_errno = errno;
2440 jbd2_journal_update_sb_errno(journal);
2442 mutex_unlock(&journal->j_abort_mutex);
2447 * Mark the abort as occurred and start current running transaction
2448 * to release all journaled buffer.
2450 pr_err("Aborting journal on device %s.\n", journal->j_devname);
2452 journal->j_flags |= JBD2_ABORT;
2453 journal->j_errno = errno;
2454 transaction = journal->j_running_transaction;
2456 __jbd2_log_start_commit(journal, transaction->t_tid);
2457 write_unlock(&journal->j_state_lock);
2460 * Record errno to the journal super block, so that fsck and jbd2
2461 * layer could realise that a filesystem check is needed.
2463 jbd2_journal_update_sb_errno(journal);
2464 mutex_unlock(&journal->j_abort_mutex);
2468 * jbd2_journal_errno() - returns the journal's error state.
2469 * @journal: journal to examine.
2471 * This is the errno number set with jbd2_journal_abort(), the last
2472 * time the journal was mounted - if the journal was stopped
2473 * without calling abort this will be 0.
2475 * If the journal has been aborted on this mount time -EROFS will
2478 int jbd2_journal_errno(journal_t *journal)
2482 read_lock(&journal->j_state_lock);
2483 if (journal->j_flags & JBD2_ABORT)
2486 err = journal->j_errno;
2487 read_unlock(&journal->j_state_lock);
2492 * jbd2_journal_clear_err() - clears the journal's error state
2493 * @journal: journal to act on.
2495 * An error must be cleared or acked to take a FS out of readonly
2498 int jbd2_journal_clear_err(journal_t *journal)
2502 write_lock(&journal->j_state_lock);
2503 if (journal->j_flags & JBD2_ABORT)
2506 journal->j_errno = 0;
2507 write_unlock(&journal->j_state_lock);
2512 * jbd2_journal_ack_err() - Ack journal err.
2513 * @journal: journal to act on.
2515 * An error must be cleared or acked to take a FS out of readonly
2518 void jbd2_journal_ack_err(journal_t *journal)
2520 write_lock(&journal->j_state_lock);
2521 if (journal->j_errno)
2522 journal->j_flags |= JBD2_ACK_ERR;
2523 write_unlock(&journal->j_state_lock);
2526 int jbd2_journal_blocks_per_page(struct inode *inode)
2528 return 1 << (PAGE_SHIFT - inode->i_sb->s_blocksize_bits);
2532 * helper functions to deal with 32 or 64bit block numbers.
2534 size_t journal_tag_bytes(journal_t *journal)
2538 if (jbd2_has_feature_csum3(journal))
2539 return sizeof(journal_block_tag3_t);
2541 sz = sizeof(journal_block_tag_t);
2543 if (jbd2_has_feature_csum2(journal))
2544 sz += sizeof(__u16);
2546 if (jbd2_has_feature_64bit(journal))
2549 return sz - sizeof(__u32);
2553 * JBD memory management
2555 * These functions are used to allocate block-sized chunks of memory
2556 * used for making copies of buffer_head data. Very often it will be
2557 * page-sized chunks of data, but sometimes it will be in
2558 * sub-page-size chunks. (For example, 16k pages on Power systems
2559 * with a 4k block file system.) For blocks smaller than a page, we
2560 * use a SLAB allocator. There are slab caches for each block size,
2561 * which are allocated at mount time, if necessary, and we only free
2562 * (all of) the slab caches when/if the jbd2 module is unloaded. For
2563 * this reason we don't need to a mutex to protect access to
2564 * jbd2_slab[] allocating or releasing memory; only in
2565 * jbd2_journal_create_slab().
2567 #define JBD2_MAX_SLABS 8
2568 static struct kmem_cache *jbd2_slab[JBD2_MAX_SLABS];
2570 static const char *jbd2_slab_names[JBD2_MAX_SLABS] = {
2571 "jbd2_1k", "jbd2_2k", "jbd2_4k", "jbd2_8k",
2572 "jbd2_16k", "jbd2_32k", "jbd2_64k", "jbd2_128k"
2576 static void jbd2_journal_destroy_slabs(void)
2580 for (i = 0; i < JBD2_MAX_SLABS; i++) {
2581 kmem_cache_destroy(jbd2_slab[i]);
2582 jbd2_slab[i] = NULL;
2586 static int jbd2_journal_create_slab(size_t size)
2588 static DEFINE_MUTEX(jbd2_slab_create_mutex);
2589 int i = order_base_2(size) - 10;
2592 if (size == PAGE_SIZE)
2595 if (i >= JBD2_MAX_SLABS)
2598 if (unlikely(i < 0))
2600 mutex_lock(&jbd2_slab_create_mutex);
2602 mutex_unlock(&jbd2_slab_create_mutex);
2603 return 0; /* Already created */
2606 slab_size = 1 << (i+10);
2607 jbd2_slab[i] = kmem_cache_create(jbd2_slab_names[i], slab_size,
2608 slab_size, 0, NULL);
2609 mutex_unlock(&jbd2_slab_create_mutex);
2610 if (!jbd2_slab[i]) {
2611 printk(KERN_EMERG "JBD2: no memory for jbd2_slab cache\n");
2617 static struct kmem_cache *get_slab(size_t size)
2619 int i = order_base_2(size) - 10;
2621 BUG_ON(i >= JBD2_MAX_SLABS);
2622 if (unlikely(i < 0))
2624 BUG_ON(jbd2_slab[i] == NULL);
2625 return jbd2_slab[i];
2628 void *jbd2_alloc(size_t size, gfp_t flags)
2632 BUG_ON(size & (size-1)); /* Must be a power of 2 */
2634 if (size < PAGE_SIZE)
2635 ptr = kmem_cache_alloc(get_slab(size), flags);
2637 ptr = (void *)__get_free_pages(flags, get_order(size));
2639 /* Check alignment; SLUB has gotten this wrong in the past,
2640 * and this can lead to user data corruption! */
2641 BUG_ON(((unsigned long) ptr) & (size-1));
2646 void jbd2_free(void *ptr, size_t size)
2648 if (size < PAGE_SIZE)
2649 kmem_cache_free(get_slab(size), ptr);
2651 free_pages((unsigned long)ptr, get_order(size));
2655 * Journal_head storage management
2657 static struct kmem_cache *jbd2_journal_head_cache;
2658 #ifdef CONFIG_JBD2_DEBUG
2659 static atomic_t nr_journal_heads = ATOMIC_INIT(0);
2662 static int __init jbd2_journal_init_journal_head_cache(void)
2664 J_ASSERT(!jbd2_journal_head_cache);
2665 jbd2_journal_head_cache = kmem_cache_create("jbd2_journal_head",
2666 sizeof(struct journal_head),
2668 SLAB_TEMPORARY | SLAB_TYPESAFE_BY_RCU,
2670 if (!jbd2_journal_head_cache) {
2671 printk(KERN_EMERG "JBD2: no memory for journal_head cache\n");
2677 static void jbd2_journal_destroy_journal_head_cache(void)
2679 kmem_cache_destroy(jbd2_journal_head_cache);
2680 jbd2_journal_head_cache = NULL;
2684 * journal_head splicing and dicing
2686 static struct journal_head *journal_alloc_journal_head(void)
2688 struct journal_head *ret;
2690 #ifdef CONFIG_JBD2_DEBUG
2691 atomic_inc(&nr_journal_heads);
2693 ret = kmem_cache_zalloc(jbd2_journal_head_cache, GFP_NOFS);
2695 jbd_debug(1, "out of memory for journal_head\n");
2696 pr_notice_ratelimited("ENOMEM in %s, retrying.\n", __func__);
2697 ret = kmem_cache_zalloc(jbd2_journal_head_cache,
2698 GFP_NOFS | __GFP_NOFAIL);
2701 spin_lock_init(&ret->b_state_lock);
2705 static void journal_free_journal_head(struct journal_head *jh)
2707 #ifdef CONFIG_JBD2_DEBUG
2708 atomic_dec(&nr_journal_heads);
2709 memset(jh, JBD2_POISON_FREE, sizeof(*jh));
2711 kmem_cache_free(jbd2_journal_head_cache, jh);
2715 * A journal_head is attached to a buffer_head whenever JBD has an
2716 * interest in the buffer.
2718 * Whenever a buffer has an attached journal_head, its ->b_state:BH_JBD bit
2719 * is set. This bit is tested in core kernel code where we need to take
2720 * JBD-specific actions. Testing the zeroness of ->b_private is not reliable
2723 * When a buffer has its BH_JBD bit set, its ->b_count is elevated by one.
2725 * When a buffer has its BH_JBD bit set it is immune from being released by
2726 * core kernel code, mainly via ->b_count.
2728 * A journal_head is detached from its buffer_head when the journal_head's
2729 * b_jcount reaches zero. Running transaction (b_transaction) and checkpoint
2730 * transaction (b_cp_transaction) hold their references to b_jcount.
2732 * Various places in the kernel want to attach a journal_head to a buffer_head
2733 * _before_ attaching the journal_head to a transaction. To protect the
2734 * journal_head in this situation, jbd2_journal_add_journal_head elevates the
2735 * journal_head's b_jcount refcount by one. The caller must call
2736 * jbd2_journal_put_journal_head() to undo this.
2738 * So the typical usage would be:
2740 * (Attach a journal_head if needed. Increments b_jcount)
2741 * struct journal_head *jh = jbd2_journal_add_journal_head(bh);
2743 * (Get another reference for transaction)
2744 * jbd2_journal_grab_journal_head(bh);
2745 * jh->b_transaction = xxx;
2746 * (Put original reference)
2747 * jbd2_journal_put_journal_head(jh);
2751 * Give a buffer_head a journal_head.
2755 struct journal_head *jbd2_journal_add_journal_head(struct buffer_head *bh)
2757 struct journal_head *jh;
2758 struct journal_head *new_jh = NULL;
2761 if (!buffer_jbd(bh))
2762 new_jh = journal_alloc_journal_head();
2764 jbd_lock_bh_journal_head(bh);
2765 if (buffer_jbd(bh)) {
2769 (atomic_read(&bh->b_count) > 0) ||
2770 (bh->b_page && bh->b_page->mapping));
2773 jbd_unlock_bh_journal_head(bh);
2778 new_jh = NULL; /* We consumed it */
2783 BUFFER_TRACE(bh, "added journal_head");
2786 jbd_unlock_bh_journal_head(bh);
2788 journal_free_journal_head(new_jh);
2789 return bh->b_private;
2793 * Grab a ref against this buffer_head's journal_head. If it ended up not
2794 * having a journal_head, return NULL
2796 struct journal_head *jbd2_journal_grab_journal_head(struct buffer_head *bh)
2798 struct journal_head *jh = NULL;
2800 jbd_lock_bh_journal_head(bh);
2801 if (buffer_jbd(bh)) {
2805 jbd_unlock_bh_journal_head(bh);
2808 EXPORT_SYMBOL(jbd2_journal_grab_journal_head);
2810 static void __journal_remove_journal_head(struct buffer_head *bh)
2812 struct journal_head *jh = bh2jh(bh);
2814 J_ASSERT_JH(jh, jh->b_transaction == NULL);
2815 J_ASSERT_JH(jh, jh->b_next_transaction == NULL);
2816 J_ASSERT_JH(jh, jh->b_cp_transaction == NULL);
2817 J_ASSERT_JH(jh, jh->b_jlist == BJ_None);
2818 J_ASSERT_BH(bh, buffer_jbd(bh));
2819 J_ASSERT_BH(bh, jh2bh(jh) == bh);
2820 BUFFER_TRACE(bh, "remove journal_head");
2822 /* Unlink before dropping the lock */
2823 bh->b_private = NULL;
2824 jh->b_bh = NULL; /* debug, really */
2825 clear_buffer_jbd(bh);
2828 static void journal_release_journal_head(struct journal_head *jh, size_t b_size)
2830 if (jh->b_frozen_data) {
2831 printk(KERN_WARNING "%s: freeing b_frozen_data\n", __func__);
2832 jbd2_free(jh->b_frozen_data, b_size);
2834 if (jh->b_committed_data) {
2835 printk(KERN_WARNING "%s: freeing b_committed_data\n", __func__);
2836 jbd2_free(jh->b_committed_data, b_size);
2838 journal_free_journal_head(jh);
2842 * Drop a reference on the passed journal_head. If it fell to zero then
2843 * release the journal_head from the buffer_head.
2845 void jbd2_journal_put_journal_head(struct journal_head *jh)
2847 struct buffer_head *bh = jh2bh(jh);
2849 jbd_lock_bh_journal_head(bh);
2850 J_ASSERT_JH(jh, jh->b_jcount > 0);
2852 if (!jh->b_jcount) {
2853 __journal_remove_journal_head(bh);
2854 jbd_unlock_bh_journal_head(bh);
2855 journal_release_journal_head(jh, bh->b_size);
2858 jbd_unlock_bh_journal_head(bh);
2861 EXPORT_SYMBOL(jbd2_journal_put_journal_head);
2864 * Initialize jbd inode head
2866 void jbd2_journal_init_jbd_inode(struct jbd2_inode *jinode, struct inode *inode)
2868 jinode->i_transaction = NULL;
2869 jinode->i_next_transaction = NULL;
2870 jinode->i_vfs_inode = inode;
2871 jinode->i_flags = 0;
2872 jinode->i_dirty_start = 0;
2873 jinode->i_dirty_end = 0;
2874 INIT_LIST_HEAD(&jinode->i_list);
2878 * Function to be called before we start removing inode from memory (i.e.,
2879 * clear_inode() is a fine place to be called from). It removes inode from
2880 * transaction's lists.
2882 void jbd2_journal_release_jbd_inode(journal_t *journal,
2883 struct jbd2_inode *jinode)
2888 spin_lock(&journal->j_list_lock);
2889 /* Is commit writing out inode - we have to wait */
2890 if (jinode->i_flags & JI_COMMIT_RUNNING) {
2891 wait_queue_head_t *wq;
2892 DEFINE_WAIT_BIT(wait, &jinode->i_flags, __JI_COMMIT_RUNNING);
2893 wq = bit_waitqueue(&jinode->i_flags, __JI_COMMIT_RUNNING);
2894 prepare_to_wait(wq, &wait.wq_entry, TASK_UNINTERRUPTIBLE);
2895 spin_unlock(&journal->j_list_lock);
2897 finish_wait(wq, &wait.wq_entry);
2901 if (jinode->i_transaction) {
2902 list_del(&jinode->i_list);
2903 jinode->i_transaction = NULL;
2905 spin_unlock(&journal->j_list_lock);
2909 #ifdef CONFIG_PROC_FS
2911 #define JBD2_STATS_PROC_NAME "fs/jbd2"
2913 static void __init jbd2_create_jbd_stats_proc_entry(void)
2915 proc_jbd2_stats = proc_mkdir(JBD2_STATS_PROC_NAME, NULL);
2918 static void __exit jbd2_remove_jbd_stats_proc_entry(void)
2920 if (proc_jbd2_stats)
2921 remove_proc_entry(JBD2_STATS_PROC_NAME, NULL);
2926 #define jbd2_create_jbd_stats_proc_entry() do {} while (0)
2927 #define jbd2_remove_jbd_stats_proc_entry() do {} while (0)
2931 struct kmem_cache *jbd2_handle_cache, *jbd2_inode_cache;
2933 static int __init jbd2_journal_init_inode_cache(void)
2935 J_ASSERT(!jbd2_inode_cache);
2936 jbd2_inode_cache = KMEM_CACHE(jbd2_inode, 0);
2937 if (!jbd2_inode_cache) {
2938 pr_emerg("JBD2: failed to create inode cache\n");
2944 static int __init jbd2_journal_init_handle_cache(void)
2946 J_ASSERT(!jbd2_handle_cache);
2947 jbd2_handle_cache = KMEM_CACHE(jbd2_journal_handle, SLAB_TEMPORARY);
2948 if (!jbd2_handle_cache) {
2949 printk(KERN_EMERG "JBD2: failed to create handle cache\n");
2955 static void jbd2_journal_destroy_inode_cache(void)
2957 kmem_cache_destroy(jbd2_inode_cache);
2958 jbd2_inode_cache = NULL;
2961 static void jbd2_journal_destroy_handle_cache(void)
2963 kmem_cache_destroy(jbd2_handle_cache);
2964 jbd2_handle_cache = NULL;
2968 * Module startup and shutdown
2971 static int __init journal_init_caches(void)
2975 ret = jbd2_journal_init_revoke_record_cache();
2977 ret = jbd2_journal_init_revoke_table_cache();
2979 ret = jbd2_journal_init_journal_head_cache();
2981 ret = jbd2_journal_init_handle_cache();
2983 ret = jbd2_journal_init_inode_cache();
2985 ret = jbd2_journal_init_transaction_cache();
2989 static void jbd2_journal_destroy_caches(void)
2991 jbd2_journal_destroy_revoke_record_cache();
2992 jbd2_journal_destroy_revoke_table_cache();
2993 jbd2_journal_destroy_journal_head_cache();
2994 jbd2_journal_destroy_handle_cache();
2995 jbd2_journal_destroy_inode_cache();
2996 jbd2_journal_destroy_transaction_cache();
2997 jbd2_journal_destroy_slabs();
3000 static int __init journal_init(void)
3004 BUILD_BUG_ON(sizeof(struct journal_superblock_s) != 1024);
3006 ret = journal_init_caches();
3008 jbd2_create_jbd_stats_proc_entry();
3010 jbd2_journal_destroy_caches();
3015 static void __exit journal_exit(void)
3017 #ifdef CONFIG_JBD2_DEBUG
3018 int n = atomic_read(&nr_journal_heads);
3020 printk(KERN_ERR "JBD2: leaked %d journal_heads!\n", n);
3022 jbd2_remove_jbd_stats_proc_entry();
3023 jbd2_journal_destroy_caches();
3026 MODULE_LICENSE("GPL");
3027 module_init(journal_init);
3028 module_exit(journal_exit);