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 jbd2_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 jbd2_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 jbd2_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 jbd2_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 jbd2_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 jbd2_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 jbd2_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 jbd2_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, tid);
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);
945 int jbd2_fc_release_bufs(journal_t *journal)
947 struct buffer_head *bh;
950 j_fc_off = journal->j_fc_off;
952 for (i = j_fc_off - 1; i >= 0; i--) {
953 bh = journal->j_fc_wbuf[i];
957 journal->j_fc_wbuf[i] = NULL;
962 EXPORT_SYMBOL(jbd2_fc_release_bufs);
965 * Conversion of logical to physical block numbers for the journal
967 * On external journals the journal blocks are identity-mapped, so
968 * this is a no-op. If needed, we can use j_blk_offset - everything is
971 int jbd2_journal_bmap(journal_t *journal, unsigned long blocknr,
972 unsigned long long *retp)
975 unsigned long long ret;
978 if (journal->j_inode) {
980 ret = bmap(journal->j_inode, &block);
983 printk(KERN_ALERT "%s: journal block not found "
984 "at offset %lu on %s\n",
985 __func__, blocknr, journal->j_devname);
987 jbd2_journal_abort(journal, err);
993 *retp = blocknr; /* +journal->j_blk_offset */
999 * We play buffer_head aliasing tricks to write data/metadata blocks to
1000 * the journal without copying their contents, but for journal
1001 * descriptor blocks we do need to generate bona fide buffers.
1003 * After the caller of jbd2_journal_get_descriptor_buffer() has finished modifying
1004 * the buffer's contents they really should run flush_dcache_page(bh->b_page).
1005 * But we don't bother doing that, so there will be coherency problems with
1006 * mmaps of blockdevs which hold live JBD-controlled filesystems.
1008 struct buffer_head *
1009 jbd2_journal_get_descriptor_buffer(transaction_t *transaction, int type)
1011 journal_t *journal = transaction->t_journal;
1012 struct buffer_head *bh;
1013 unsigned long long blocknr;
1014 journal_header_t *header;
1017 err = jbd2_journal_next_log_block(journal, &blocknr);
1022 bh = __getblk(journal->j_dev, blocknr, journal->j_blocksize);
1025 atomic_dec(&transaction->t_outstanding_credits);
1027 memset(bh->b_data, 0, journal->j_blocksize);
1028 header = (journal_header_t *)bh->b_data;
1029 header->h_magic = cpu_to_be32(JBD2_MAGIC_NUMBER);
1030 header->h_blocktype = cpu_to_be32(type);
1031 header->h_sequence = cpu_to_be32(transaction->t_tid);
1032 set_buffer_uptodate(bh);
1034 BUFFER_TRACE(bh, "return this buffer");
1038 void jbd2_descriptor_block_csum_set(journal_t *j, struct buffer_head *bh)
1040 struct jbd2_journal_block_tail *tail;
1043 if (!jbd2_journal_has_csum_v2or3(j))
1046 tail = (struct jbd2_journal_block_tail *)(bh->b_data + j->j_blocksize -
1047 sizeof(struct jbd2_journal_block_tail));
1048 tail->t_checksum = 0;
1049 csum = jbd2_chksum(j, j->j_csum_seed, bh->b_data, j->j_blocksize);
1050 tail->t_checksum = cpu_to_be32(csum);
1054 * Return tid of the oldest transaction in the journal and block in the journal
1055 * where the transaction starts.
1057 * If the journal is now empty, return which will be the next transaction ID
1058 * we will write and where will that transaction start.
1060 * The return value is 0 if journal tail cannot be pushed any further, 1 if
1063 int jbd2_journal_get_log_tail(journal_t *journal, tid_t *tid,
1064 unsigned long *block)
1066 transaction_t *transaction;
1069 read_lock(&journal->j_state_lock);
1070 spin_lock(&journal->j_list_lock);
1071 transaction = journal->j_checkpoint_transactions;
1073 *tid = transaction->t_tid;
1074 *block = transaction->t_log_start;
1075 } else if ((transaction = journal->j_committing_transaction) != NULL) {
1076 *tid = transaction->t_tid;
1077 *block = transaction->t_log_start;
1078 } else if ((transaction = journal->j_running_transaction) != NULL) {
1079 *tid = transaction->t_tid;
1080 *block = journal->j_head;
1082 *tid = journal->j_transaction_sequence;
1083 *block = journal->j_head;
1085 ret = tid_gt(*tid, journal->j_tail_sequence);
1086 spin_unlock(&journal->j_list_lock);
1087 read_unlock(&journal->j_state_lock);
1093 * Update information in journal structure and in on disk journal superblock
1094 * about log tail. This function does not check whether information passed in
1095 * really pushes log tail further. It's responsibility of the caller to make
1096 * sure provided log tail information is valid (e.g. by holding
1097 * j_checkpoint_mutex all the time between computing log tail and calling this
1098 * function as is the case with jbd2_cleanup_journal_tail()).
1100 * Requires j_checkpoint_mutex
1102 int __jbd2_update_log_tail(journal_t *journal, tid_t tid, unsigned long block)
1104 unsigned long freed;
1107 BUG_ON(!mutex_is_locked(&journal->j_checkpoint_mutex));
1110 * We cannot afford for write to remain in drive's caches since as
1111 * soon as we update j_tail, next transaction can start reusing journal
1112 * space and if we lose sb update during power failure we'd replay
1113 * old transaction with possibly newly overwritten data.
1115 ret = jbd2_journal_update_sb_log_tail(journal, tid, block,
1116 REQ_SYNC | REQ_FUA);
1120 write_lock(&journal->j_state_lock);
1121 freed = block - journal->j_tail;
1122 if (block < journal->j_tail)
1123 freed += journal->j_last - journal->j_first;
1125 trace_jbd2_update_log_tail(journal, tid, block, freed);
1127 "Cleaning journal tail from %u to %u (offset %lu), "
1129 journal->j_tail_sequence, tid, block, freed);
1131 journal->j_free += freed;
1132 journal->j_tail_sequence = tid;
1133 journal->j_tail = block;
1134 write_unlock(&journal->j_state_lock);
1141 * This is a variation of __jbd2_update_log_tail which checks for validity of
1142 * provided log tail and locks j_checkpoint_mutex. So it is safe against races
1143 * with other threads updating log tail.
1145 void jbd2_update_log_tail(journal_t *journal, tid_t tid, unsigned long block)
1147 mutex_lock_io(&journal->j_checkpoint_mutex);
1148 if (tid_gt(tid, journal->j_tail_sequence))
1149 __jbd2_update_log_tail(journal, tid, block);
1150 mutex_unlock(&journal->j_checkpoint_mutex);
1153 struct jbd2_stats_proc_session {
1155 struct transaction_stats_s *stats;
1160 static void *jbd2_seq_info_start(struct seq_file *seq, loff_t *pos)
1162 return *pos ? NULL : SEQ_START_TOKEN;
1165 static void *jbd2_seq_info_next(struct seq_file *seq, void *v, loff_t *pos)
1171 static int jbd2_seq_info_show(struct seq_file *seq, void *v)
1173 struct jbd2_stats_proc_session *s = seq->private;
1175 if (v != SEQ_START_TOKEN)
1177 seq_printf(seq, "%lu transactions (%lu requested), "
1178 "each up to %u blocks\n",
1179 s->stats->ts_tid, s->stats->ts_requested,
1180 s->journal->j_max_transaction_buffers);
1181 if (s->stats->ts_tid == 0)
1183 seq_printf(seq, "average: \n %ums waiting for transaction\n",
1184 jiffies_to_msecs(s->stats->run.rs_wait / s->stats->ts_tid));
1185 seq_printf(seq, " %ums request delay\n",
1186 (s->stats->ts_requested == 0) ? 0 :
1187 jiffies_to_msecs(s->stats->run.rs_request_delay /
1188 s->stats->ts_requested));
1189 seq_printf(seq, " %ums running transaction\n",
1190 jiffies_to_msecs(s->stats->run.rs_running / s->stats->ts_tid));
1191 seq_printf(seq, " %ums transaction was being locked\n",
1192 jiffies_to_msecs(s->stats->run.rs_locked / s->stats->ts_tid));
1193 seq_printf(seq, " %ums flushing data (in ordered mode)\n",
1194 jiffies_to_msecs(s->stats->run.rs_flushing / s->stats->ts_tid));
1195 seq_printf(seq, " %ums logging transaction\n",
1196 jiffies_to_msecs(s->stats->run.rs_logging / s->stats->ts_tid));
1197 seq_printf(seq, " %lluus average transaction commit time\n",
1198 div_u64(s->journal->j_average_commit_time, 1000));
1199 seq_printf(seq, " %lu handles per transaction\n",
1200 s->stats->run.rs_handle_count / s->stats->ts_tid);
1201 seq_printf(seq, " %lu blocks per transaction\n",
1202 s->stats->run.rs_blocks / s->stats->ts_tid);
1203 seq_printf(seq, " %lu logged blocks per transaction\n",
1204 s->stats->run.rs_blocks_logged / s->stats->ts_tid);
1208 static void jbd2_seq_info_stop(struct seq_file *seq, void *v)
1212 static const struct seq_operations jbd2_seq_info_ops = {
1213 .start = jbd2_seq_info_start,
1214 .next = jbd2_seq_info_next,
1215 .stop = jbd2_seq_info_stop,
1216 .show = jbd2_seq_info_show,
1219 static int jbd2_seq_info_open(struct inode *inode, struct file *file)
1221 journal_t *journal = PDE_DATA(inode);
1222 struct jbd2_stats_proc_session *s;
1225 s = kmalloc(sizeof(*s), GFP_KERNEL);
1228 size = sizeof(struct transaction_stats_s);
1229 s->stats = kmalloc(size, GFP_KERNEL);
1230 if (s->stats == NULL) {
1234 spin_lock(&journal->j_history_lock);
1235 memcpy(s->stats, &journal->j_stats, size);
1236 s->journal = journal;
1237 spin_unlock(&journal->j_history_lock);
1239 rc = seq_open(file, &jbd2_seq_info_ops);
1241 struct seq_file *m = file->private_data;
1251 static int jbd2_seq_info_release(struct inode *inode, struct file *file)
1253 struct seq_file *seq = file->private_data;
1254 struct jbd2_stats_proc_session *s = seq->private;
1257 return seq_release(inode, file);
1260 static const struct proc_ops jbd2_info_proc_ops = {
1261 .proc_open = jbd2_seq_info_open,
1262 .proc_read = seq_read,
1263 .proc_lseek = seq_lseek,
1264 .proc_release = jbd2_seq_info_release,
1267 static struct proc_dir_entry *proc_jbd2_stats;
1269 static void jbd2_stats_proc_init(journal_t *journal)
1271 journal->j_proc_entry = proc_mkdir(journal->j_devname, proc_jbd2_stats);
1272 if (journal->j_proc_entry) {
1273 proc_create_data("info", S_IRUGO, journal->j_proc_entry,
1274 &jbd2_info_proc_ops, journal);
1278 static void jbd2_stats_proc_exit(journal_t *journal)
1280 remove_proc_entry("info", journal->j_proc_entry);
1281 remove_proc_entry(journal->j_devname, proc_jbd2_stats);
1284 /* Minimum size of descriptor tag */
1285 static int jbd2_min_tag_size(void)
1288 * Tag with 32-bit block numbers does not use last four bytes of the
1291 return sizeof(journal_block_tag_t) - 4;
1295 * jbd2_journal_shrink_scan()
1297 * Scan the checkpointed buffer on the checkpoint list and release the
1300 static unsigned long jbd2_journal_shrink_scan(struct shrinker *shrink,
1301 struct shrink_control *sc)
1303 journal_t *journal = container_of(shrink, journal_t, j_shrinker);
1304 unsigned long nr_to_scan = sc->nr_to_scan;
1305 unsigned long nr_shrunk;
1306 unsigned long count;
1308 count = percpu_counter_read_positive(&journal->j_checkpoint_jh_count);
1309 trace_jbd2_shrink_scan_enter(journal, sc->nr_to_scan, count);
1311 nr_shrunk = jbd2_journal_shrink_checkpoint_list(journal, &nr_to_scan);
1313 count = percpu_counter_read_positive(&journal->j_checkpoint_jh_count);
1314 trace_jbd2_shrink_scan_exit(journal, nr_to_scan, nr_shrunk, count);
1320 * jbd2_journal_shrink_count()
1322 * Count the number of checkpoint buffers on the checkpoint list.
1324 static unsigned long jbd2_journal_shrink_count(struct shrinker *shrink,
1325 struct shrink_control *sc)
1327 journal_t *journal = container_of(shrink, journal_t, j_shrinker);
1328 unsigned long count;
1330 count = percpu_counter_read_positive(&journal->j_checkpoint_jh_count);
1331 trace_jbd2_shrink_count(journal, sc->nr_to_scan, count);
1337 * Management for journal control blocks: functions to create and
1338 * destroy journal_t structures, and to initialise and read existing
1339 * journal blocks from disk. */
1341 /* First: create and setup a journal_t object in memory. We initialise
1342 * very few fields yet: that has to wait until we have created the
1343 * journal structures from from scratch, or loaded them from disk. */
1345 static journal_t *journal_init_common(struct block_device *bdev,
1346 struct block_device *fs_dev,
1347 unsigned long long start, int len, int blocksize)
1349 static struct lock_class_key jbd2_trans_commit_key;
1352 struct buffer_head *bh;
1355 journal = kzalloc(sizeof(*journal), GFP_KERNEL);
1359 init_waitqueue_head(&journal->j_wait_transaction_locked);
1360 init_waitqueue_head(&journal->j_wait_done_commit);
1361 init_waitqueue_head(&journal->j_wait_commit);
1362 init_waitqueue_head(&journal->j_wait_updates);
1363 init_waitqueue_head(&journal->j_wait_reserved);
1364 init_waitqueue_head(&journal->j_fc_wait);
1365 mutex_init(&journal->j_abort_mutex);
1366 mutex_init(&journal->j_barrier);
1367 mutex_init(&journal->j_checkpoint_mutex);
1368 spin_lock_init(&journal->j_revoke_lock);
1369 spin_lock_init(&journal->j_list_lock);
1370 rwlock_init(&journal->j_state_lock);
1372 journal->j_commit_interval = (HZ * JBD2_DEFAULT_MAX_COMMIT_AGE);
1373 journal->j_min_batch_time = 0;
1374 journal->j_max_batch_time = 15000; /* 15ms */
1375 atomic_set(&journal->j_reserved_credits, 0);
1377 /* The journal is marked for error until we succeed with recovery! */
1378 journal->j_flags = JBD2_ABORT;
1380 /* Set up a default-sized revoke table for the new mount. */
1381 err = jbd2_journal_init_revoke(journal, JOURNAL_REVOKE_DEFAULT_HASH);
1385 spin_lock_init(&journal->j_history_lock);
1387 lockdep_init_map(&journal->j_trans_commit_map, "jbd2_handle",
1388 &jbd2_trans_commit_key, 0);
1390 /* journal descriptor can store up to n blocks -bzzz */
1391 journal->j_blocksize = blocksize;
1392 journal->j_dev = bdev;
1393 journal->j_fs_dev = fs_dev;
1394 journal->j_blk_offset = start;
1395 journal->j_total_len = len;
1396 /* We need enough buffers to write out full descriptor block. */
1397 n = journal->j_blocksize / jbd2_min_tag_size();
1398 journal->j_wbufsize = n;
1399 journal->j_fc_wbuf = NULL;
1400 journal->j_wbuf = kmalloc_array(n, sizeof(struct buffer_head *),
1402 if (!journal->j_wbuf)
1405 bh = getblk_unmovable(journal->j_dev, start, journal->j_blocksize);
1407 pr_err("%s: Cannot get buffer for journal superblock\n",
1411 journal->j_sb_buffer = bh;
1412 journal->j_superblock = (journal_superblock_t *)bh->b_data;
1414 journal->j_shrink_transaction = NULL;
1415 journal->j_shrinker.scan_objects = jbd2_journal_shrink_scan;
1416 journal->j_shrinker.count_objects = jbd2_journal_shrink_count;
1417 journal->j_shrinker.seeks = DEFAULT_SEEKS;
1418 journal->j_shrinker.batch = journal->j_max_transaction_buffers;
1420 if (percpu_counter_init(&journal->j_checkpoint_jh_count, 0, GFP_KERNEL))
1423 if (register_shrinker(&journal->j_shrinker)) {
1424 percpu_counter_destroy(&journal->j_checkpoint_jh_count);
1430 brelse(journal->j_sb_buffer);
1431 kfree(journal->j_wbuf);
1432 jbd2_journal_destroy_revoke(journal);
1437 /* jbd2_journal_init_dev and jbd2_journal_init_inode:
1439 * Create a journal structure assigned some fixed set of disk blocks to
1440 * the journal. We don't actually touch those disk blocks yet, but we
1441 * need to set up all of the mapping information to tell the journaling
1442 * system where the journal blocks are.
1447 * journal_t * jbd2_journal_init_dev() - creates and initialises a journal structure
1448 * @bdev: Block device on which to create the journal
1449 * @fs_dev: Device which hold journalled filesystem for this journal.
1450 * @start: Block nr Start of journal.
1451 * @len: Length of the journal in blocks.
1452 * @blocksize: blocksize of journalling device
1454 * Returns: a newly created journal_t *
1456 * jbd2_journal_init_dev creates a journal which maps a fixed contiguous
1457 * range of blocks on an arbitrary block device.
1460 journal_t *jbd2_journal_init_dev(struct block_device *bdev,
1461 struct block_device *fs_dev,
1462 unsigned long long start, int len, int blocksize)
1466 journal = journal_init_common(bdev, fs_dev, start, len, blocksize);
1470 bdevname(journal->j_dev, journal->j_devname);
1471 strreplace(journal->j_devname, '/', '!');
1472 jbd2_stats_proc_init(journal);
1478 * journal_t * jbd2_journal_init_inode () - creates a journal which maps to a inode.
1479 * @inode: An inode to create the journal in
1481 * jbd2_journal_init_inode creates a journal which maps an on-disk inode as
1482 * the journal. The inode must exist already, must support bmap() and
1483 * must have all data blocks preallocated.
1485 journal_t *jbd2_journal_init_inode(struct inode *inode)
1493 err = bmap(inode, &blocknr);
1495 if (err || !blocknr) {
1496 pr_err("%s: Cannot locate journal superblock\n",
1501 jbd2_debug(1, "JBD2: inode %s/%ld, size %lld, bits %d, blksize %ld\n",
1502 inode->i_sb->s_id, inode->i_ino, (long long) inode->i_size,
1503 inode->i_sb->s_blocksize_bits, inode->i_sb->s_blocksize);
1505 journal = journal_init_common(inode->i_sb->s_bdev, inode->i_sb->s_bdev,
1506 blocknr, inode->i_size >> inode->i_sb->s_blocksize_bits,
1507 inode->i_sb->s_blocksize);
1511 journal->j_inode = inode;
1512 bdevname(journal->j_dev, journal->j_devname);
1513 p = strreplace(journal->j_devname, '/', '!');
1514 sprintf(p, "-%lu", journal->j_inode->i_ino);
1515 jbd2_stats_proc_init(journal);
1521 * If the journal init or create aborts, we need to mark the journal
1522 * superblock as being NULL to prevent the journal destroy from writing
1523 * back a bogus superblock.
1525 static void journal_fail_superblock(journal_t *journal)
1527 struct buffer_head *bh = journal->j_sb_buffer;
1529 journal->j_sb_buffer = NULL;
1533 * Given a journal_t structure, initialise the various fields for
1534 * startup of a new journaling session. We use this both when creating
1535 * a journal, and after recovering an old journal to reset it for
1539 static int journal_reset(journal_t *journal)
1541 journal_superblock_t *sb = journal->j_superblock;
1542 unsigned long long first, last;
1544 first = be32_to_cpu(sb->s_first);
1545 last = be32_to_cpu(sb->s_maxlen);
1546 if (first + JBD2_MIN_JOURNAL_BLOCKS > last + 1) {
1547 printk(KERN_ERR "JBD2: Journal too short (blocks %llu-%llu).\n",
1549 journal_fail_superblock(journal);
1553 journal->j_first = first;
1554 journal->j_last = last;
1556 journal->j_head = journal->j_first;
1557 journal->j_tail = journal->j_first;
1558 journal->j_free = journal->j_last - journal->j_first;
1560 journal->j_tail_sequence = journal->j_transaction_sequence;
1561 journal->j_commit_sequence = journal->j_transaction_sequence - 1;
1562 journal->j_commit_request = journal->j_commit_sequence;
1564 journal->j_max_transaction_buffers = jbd2_journal_get_max_txn_bufs(journal);
1567 * Now that journal recovery is done, turn fast commits off here. This
1568 * way, if fast commit was enabled before the crash but if now FS has
1569 * disabled it, we don't enable fast commits.
1571 jbd2_clear_feature_fast_commit(journal);
1574 * As a special case, if the on-disk copy is already marked as needing
1575 * no recovery (s_start == 0), then we can safely defer the superblock
1576 * update until the next commit by setting JBD2_FLUSHED. This avoids
1577 * attempting a write to a potential-readonly device.
1579 if (sb->s_start == 0) {
1580 jbd2_debug(1, "JBD2: Skipping superblock update on recovered sb "
1581 "(start %ld, seq %u, errno %d)\n",
1582 journal->j_tail, journal->j_tail_sequence,
1584 journal->j_flags |= JBD2_FLUSHED;
1586 /* Lock here to make assertions happy... */
1587 mutex_lock_io(&journal->j_checkpoint_mutex);
1589 * Update log tail information. We use REQ_FUA since new
1590 * transaction will start reusing journal space and so we
1591 * must make sure information about current log tail is on
1594 jbd2_journal_update_sb_log_tail(journal,
1595 journal->j_tail_sequence,
1597 REQ_SYNC | REQ_FUA);
1598 mutex_unlock(&journal->j_checkpoint_mutex);
1600 return jbd2_journal_start_thread(journal);
1604 * This function expects that the caller will have locked the journal
1605 * buffer head, and will return with it unlocked
1607 static int jbd2_write_superblock(journal_t *journal, int write_flags)
1609 struct buffer_head *bh = journal->j_sb_buffer;
1610 journal_superblock_t *sb = journal->j_superblock;
1613 /* Buffer got discarded which means block device got invalidated */
1614 if (!buffer_mapped(bh)) {
1619 trace_jbd2_write_superblock(journal, write_flags);
1620 if (!(journal->j_flags & JBD2_BARRIER))
1621 write_flags &= ~(REQ_FUA | REQ_PREFLUSH);
1622 if (buffer_write_io_error(bh)) {
1624 * Oh, dear. A previous attempt to write the journal
1625 * superblock failed. This could happen because the
1626 * USB device was yanked out. Or it could happen to
1627 * be a transient write error and maybe the block will
1628 * be remapped. Nothing we can do but to retry the
1629 * write and hope for the best.
1631 printk(KERN_ERR "JBD2: previous I/O error detected "
1632 "for journal superblock update for %s.\n",
1633 journal->j_devname);
1634 clear_buffer_write_io_error(bh);
1635 set_buffer_uptodate(bh);
1637 if (jbd2_journal_has_csum_v2or3(journal))
1638 sb->s_checksum = jbd2_superblock_csum(journal, sb);
1640 bh->b_end_io = end_buffer_write_sync;
1641 ret = submit_bh(REQ_OP_WRITE, write_flags, bh);
1643 if (buffer_write_io_error(bh)) {
1644 clear_buffer_write_io_error(bh);
1645 set_buffer_uptodate(bh);
1649 printk(KERN_ERR "JBD2: Error %d detected when updating "
1650 "journal superblock for %s.\n", ret,
1651 journal->j_devname);
1652 if (!is_journal_aborted(journal))
1653 jbd2_journal_abort(journal, ret);
1660 * jbd2_journal_update_sb_log_tail() - Update log tail in journal sb on disk.
1661 * @journal: The journal to update.
1662 * @tail_tid: TID of the new transaction at the tail of the log
1663 * @tail_block: The first block of the transaction at the tail of the log
1664 * @write_op: With which operation should we write the journal sb
1666 * Update a journal's superblock information about log tail and write it to
1667 * disk, waiting for the IO to complete.
1669 int jbd2_journal_update_sb_log_tail(journal_t *journal, tid_t tail_tid,
1670 unsigned long tail_block, int write_op)
1672 journal_superblock_t *sb = journal->j_superblock;
1675 if (is_journal_aborted(journal))
1677 if (test_bit(JBD2_CHECKPOINT_IO_ERROR, &journal->j_atomic_flags)) {
1678 jbd2_journal_abort(journal, -EIO);
1682 BUG_ON(!mutex_is_locked(&journal->j_checkpoint_mutex));
1683 jbd2_debug(1, "JBD2: updating superblock (start %lu, seq %u)\n",
1684 tail_block, tail_tid);
1686 lock_buffer(journal->j_sb_buffer);
1687 sb->s_sequence = cpu_to_be32(tail_tid);
1688 sb->s_start = cpu_to_be32(tail_block);
1690 ret = jbd2_write_superblock(journal, write_op);
1694 /* Log is no longer empty */
1695 write_lock(&journal->j_state_lock);
1696 WARN_ON(!sb->s_sequence);
1697 journal->j_flags &= ~JBD2_FLUSHED;
1698 write_unlock(&journal->j_state_lock);
1705 * jbd2_mark_journal_empty() - Mark on disk journal as empty.
1706 * @journal: The journal to update.
1707 * @write_op: With which operation should we write the journal sb
1709 * Update a journal's dynamic superblock fields to show that journal is empty.
1710 * Write updated superblock to disk waiting for IO to complete.
1712 static void jbd2_mark_journal_empty(journal_t *journal, int write_op)
1714 journal_superblock_t *sb = journal->j_superblock;
1715 bool had_fast_commit = false;
1717 BUG_ON(!mutex_is_locked(&journal->j_checkpoint_mutex));
1718 lock_buffer(journal->j_sb_buffer);
1719 if (sb->s_start == 0) { /* Is it already empty? */
1720 unlock_buffer(journal->j_sb_buffer);
1724 jbd2_debug(1, "JBD2: Marking journal as empty (seq %u)\n",
1725 journal->j_tail_sequence);
1727 sb->s_sequence = cpu_to_be32(journal->j_tail_sequence);
1728 sb->s_start = cpu_to_be32(0);
1729 if (jbd2_has_feature_fast_commit(journal)) {
1731 * When journal is clean, no need to commit fast commit flag and
1732 * make file system incompatible with older kernels.
1734 jbd2_clear_feature_fast_commit(journal);
1735 had_fast_commit = true;
1738 jbd2_write_superblock(journal, write_op);
1740 if (had_fast_commit)
1741 jbd2_set_feature_fast_commit(journal);
1743 /* Log is no longer empty */
1744 write_lock(&journal->j_state_lock);
1745 journal->j_flags |= JBD2_FLUSHED;
1746 write_unlock(&journal->j_state_lock);
1750 * __jbd2_journal_erase() - Discard or zeroout journal blocks (excluding superblock)
1751 * @journal: The journal to erase.
1752 * @flags: A discard/zeroout request is sent for each physically contigous
1753 * region of the journal. Either JBD2_JOURNAL_FLUSH_DISCARD or
1754 * JBD2_JOURNAL_FLUSH_ZEROOUT must be set to determine which operation
1757 * Note: JBD2_JOURNAL_FLUSH_ZEROOUT attempts to use hardware offload. Zeroes
1758 * will be explicitly written if no hardware offload is available, see
1759 * blkdev_issue_zeroout for more details.
1761 static int __jbd2_journal_erase(journal_t *journal, unsigned int flags)
1764 unsigned long block, log_offset; /* logical */
1765 unsigned long long phys_block, block_start, block_stop; /* physical */
1766 loff_t byte_start, byte_stop, byte_count;
1767 struct request_queue *q = bdev_get_queue(journal->j_dev);
1769 /* flags must be set to either discard or zeroout */
1770 if ((flags & ~JBD2_JOURNAL_FLUSH_VALID) || !flags ||
1771 ((flags & JBD2_JOURNAL_FLUSH_DISCARD) &&
1772 (flags & JBD2_JOURNAL_FLUSH_ZEROOUT)))
1778 if ((flags & JBD2_JOURNAL_FLUSH_DISCARD) && !blk_queue_discard(q))
1782 * lookup block mapping and issue discard/zeroout for each
1785 log_offset = be32_to_cpu(journal->j_superblock->s_first);
1786 block_start = ~0ULL;
1787 for (block = log_offset; block < journal->j_total_len; block++) {
1788 err = jbd2_journal_bmap(journal, block, &phys_block);
1790 pr_err("JBD2: bad block at offset %lu", block);
1794 if (block_start == ~0ULL) {
1795 block_start = phys_block;
1796 block_stop = block_start - 1;
1800 * last block not contiguous with current block,
1801 * process last contiguous region and return to this block on
1804 if (phys_block != block_stop + 1) {
1809 * if this isn't the last block of journal,
1810 * no need to process now because next block may also
1811 * be part of this contiguous region
1813 if (block != journal->j_total_len - 1)
1818 * end of contiguous region or this is last block of journal,
1819 * take care of the region
1821 byte_start = block_start * journal->j_blocksize;
1822 byte_stop = block_stop * journal->j_blocksize;
1823 byte_count = (block_stop - block_start + 1) *
1824 journal->j_blocksize;
1826 truncate_inode_pages_range(journal->j_dev->bd_inode->i_mapping,
1827 byte_start, byte_stop);
1829 if (flags & JBD2_JOURNAL_FLUSH_DISCARD) {
1830 err = blkdev_issue_discard(journal->j_dev,
1831 byte_start >> SECTOR_SHIFT,
1832 byte_count >> SECTOR_SHIFT,
1834 } else if (flags & JBD2_JOURNAL_FLUSH_ZEROOUT) {
1835 err = blkdev_issue_zeroout(journal->j_dev,
1836 byte_start >> SECTOR_SHIFT,
1837 byte_count >> SECTOR_SHIFT,
1841 if (unlikely(err != 0)) {
1842 pr_err("JBD2: (error %d) unable to wipe journal at physical blocks %llu - %llu",
1843 err, block_start, block_stop);
1847 /* reset start and stop after processing a region */
1848 block_start = ~0ULL;
1851 return blkdev_issue_flush(journal->j_dev);
1855 * jbd2_journal_update_sb_errno() - Update error in the journal.
1856 * @journal: The journal to update.
1858 * Update a journal's errno. Write updated superblock to disk waiting for IO
1861 void jbd2_journal_update_sb_errno(journal_t *journal)
1863 journal_superblock_t *sb = journal->j_superblock;
1866 lock_buffer(journal->j_sb_buffer);
1867 errcode = journal->j_errno;
1868 if (errcode == -ESHUTDOWN)
1870 jbd2_debug(1, "JBD2: updating superblock error (errno %d)\n", errcode);
1871 sb->s_errno = cpu_to_be32(errcode);
1873 jbd2_write_superblock(journal, REQ_SYNC | REQ_FUA);
1875 EXPORT_SYMBOL(jbd2_journal_update_sb_errno);
1877 static int journal_revoke_records_per_block(journal_t *journal)
1880 int space = journal->j_blocksize - sizeof(jbd2_journal_revoke_header_t);
1882 if (jbd2_has_feature_64bit(journal))
1887 if (jbd2_journal_has_csum_v2or3(journal))
1888 space -= sizeof(struct jbd2_journal_block_tail);
1889 return space / record_size;
1893 * Read the superblock for a given journal, performing initial
1894 * validation of the format.
1896 static int journal_get_superblock(journal_t *journal)
1898 struct buffer_head *bh;
1899 journal_superblock_t *sb;
1902 bh = journal->j_sb_buffer;
1904 J_ASSERT(bh != NULL);
1905 if (!buffer_uptodate(bh)) {
1906 ll_rw_block(REQ_OP_READ, 0, 1, &bh);
1908 if (!buffer_uptodate(bh)) {
1910 "JBD2: IO error reading journal superblock\n");
1915 if (buffer_verified(bh))
1918 sb = journal->j_superblock;
1922 if (sb->s_header.h_magic != cpu_to_be32(JBD2_MAGIC_NUMBER) ||
1923 sb->s_blocksize != cpu_to_be32(journal->j_blocksize)) {
1924 printk(KERN_WARNING "JBD2: no valid journal superblock found\n");
1928 switch(be32_to_cpu(sb->s_header.h_blocktype)) {
1929 case JBD2_SUPERBLOCK_V1:
1930 journal->j_format_version = 1;
1932 case JBD2_SUPERBLOCK_V2:
1933 journal->j_format_version = 2;
1936 printk(KERN_WARNING "JBD2: unrecognised superblock format ID\n");
1940 if (be32_to_cpu(sb->s_maxlen) < journal->j_total_len)
1941 journal->j_total_len = be32_to_cpu(sb->s_maxlen);
1942 else if (be32_to_cpu(sb->s_maxlen) > journal->j_total_len) {
1943 printk(KERN_WARNING "JBD2: journal file too short\n");
1947 if (be32_to_cpu(sb->s_first) == 0 ||
1948 be32_to_cpu(sb->s_first) >= journal->j_total_len) {
1950 "JBD2: Invalid start block of journal: %u\n",
1951 be32_to_cpu(sb->s_first));
1955 if (jbd2_has_feature_csum2(journal) &&
1956 jbd2_has_feature_csum3(journal)) {
1957 /* Can't have checksum v2 and v3 at the same time! */
1958 printk(KERN_ERR "JBD2: Can't enable checksumming v2 and v3 "
1959 "at the same time!\n");
1963 if (jbd2_journal_has_csum_v2or3_feature(journal) &&
1964 jbd2_has_feature_checksum(journal)) {
1965 /* Can't have checksum v1 and v2 on at the same time! */
1966 printk(KERN_ERR "JBD2: Can't enable checksumming v1 and v2/3 "
1967 "at the same time!\n");
1971 if (!jbd2_verify_csum_type(journal, sb)) {
1972 printk(KERN_ERR "JBD2: Unknown checksum type\n");
1976 /* Load the checksum driver */
1977 if (jbd2_journal_has_csum_v2or3_feature(journal)) {
1978 journal->j_chksum_driver = crypto_alloc_shash("crc32c", 0, 0);
1979 if (IS_ERR(journal->j_chksum_driver)) {
1980 printk(KERN_ERR "JBD2: Cannot load crc32c driver.\n");
1981 err = PTR_ERR(journal->j_chksum_driver);
1982 journal->j_chksum_driver = NULL;
1987 if (jbd2_journal_has_csum_v2or3(journal)) {
1988 /* Check superblock checksum */
1989 if (sb->s_checksum != jbd2_superblock_csum(journal, sb)) {
1990 printk(KERN_ERR "JBD2: journal checksum error\n");
1995 /* Precompute checksum seed for all metadata */
1996 journal->j_csum_seed = jbd2_chksum(journal, ~0, sb->s_uuid,
1997 sizeof(sb->s_uuid));
2000 journal->j_revoke_records_per_block =
2001 journal_revoke_records_per_block(journal);
2002 set_buffer_verified(bh);
2007 journal_fail_superblock(journal);
2012 * Load the on-disk journal superblock and read the key fields into the
2016 static int load_superblock(journal_t *journal)
2019 journal_superblock_t *sb;
2022 err = journal_get_superblock(journal);
2026 sb = journal->j_superblock;
2028 journal->j_tail_sequence = be32_to_cpu(sb->s_sequence);
2029 journal->j_tail = be32_to_cpu(sb->s_start);
2030 journal->j_first = be32_to_cpu(sb->s_first);
2031 journal->j_errno = be32_to_cpu(sb->s_errno);
2032 journal->j_last = be32_to_cpu(sb->s_maxlen);
2034 if (jbd2_has_feature_fast_commit(journal)) {
2035 journal->j_fc_last = be32_to_cpu(sb->s_maxlen);
2036 num_fc_blocks = jbd2_journal_get_num_fc_blks(sb);
2037 if (journal->j_last - num_fc_blocks >= JBD2_MIN_JOURNAL_BLOCKS)
2038 journal->j_last = journal->j_fc_last - num_fc_blocks;
2039 journal->j_fc_first = journal->j_last + 1;
2040 journal->j_fc_off = 0;
2048 * jbd2_journal_load() - Read journal from disk.
2049 * @journal: Journal to act on.
2051 * Given a journal_t structure which tells us which disk blocks contain
2052 * a journal, read the journal from disk to initialise the in-memory
2055 int jbd2_journal_load(journal_t *journal)
2058 journal_superblock_t *sb;
2060 err = load_superblock(journal);
2064 sb = journal->j_superblock;
2065 /* If this is a V2 superblock, then we have to check the
2066 * features flags on it. */
2068 if (journal->j_format_version >= 2) {
2069 if ((sb->s_feature_ro_compat &
2070 ~cpu_to_be32(JBD2_KNOWN_ROCOMPAT_FEATURES)) ||
2071 (sb->s_feature_incompat &
2072 ~cpu_to_be32(JBD2_KNOWN_INCOMPAT_FEATURES))) {
2074 "JBD2: Unrecognised features on journal\n");
2080 * Create a slab for this blocksize
2082 err = jbd2_journal_create_slab(be32_to_cpu(sb->s_blocksize));
2086 /* Let the recovery code check whether it needs to recover any
2087 * data from the journal. */
2088 if (jbd2_journal_recover(journal))
2089 goto recovery_error;
2091 if (journal->j_failed_commit) {
2092 printk(KERN_ERR "JBD2: journal transaction %u on %s "
2093 "is corrupt.\n", journal->j_failed_commit,
2094 journal->j_devname);
2095 return -EFSCORRUPTED;
2098 * clear JBD2_ABORT flag initialized in journal_init_common
2099 * here to update log tail information with the newest seq.
2101 journal->j_flags &= ~JBD2_ABORT;
2103 /* OK, we've finished with the dynamic journal bits:
2104 * reinitialise the dynamic contents of the superblock in memory
2105 * and reset them on disk. */
2106 if (journal_reset(journal))
2107 goto recovery_error;
2109 journal->j_flags |= JBD2_LOADED;
2113 printk(KERN_WARNING "JBD2: recovery failed\n");
2118 * jbd2_journal_destroy() - Release a journal_t structure.
2119 * @journal: Journal to act on.
2121 * Release a journal_t structure once it is no longer in use by the
2123 * Return <0 if we couldn't clean up the journal.
2125 int jbd2_journal_destroy(journal_t *journal)
2129 /* Wait for the commit thread to wake up and die. */
2130 journal_kill_thread(journal);
2132 /* Force a final log commit */
2133 if (journal->j_running_transaction)
2134 jbd2_journal_commit_transaction(journal);
2136 /* Force any old transactions to disk */
2138 /* Totally anal locking here... */
2139 spin_lock(&journal->j_list_lock);
2140 while (journal->j_checkpoint_transactions != NULL) {
2141 spin_unlock(&journal->j_list_lock);
2142 mutex_lock_io(&journal->j_checkpoint_mutex);
2143 err = jbd2_log_do_checkpoint(journal);
2144 mutex_unlock(&journal->j_checkpoint_mutex);
2146 * If checkpointing failed, just free the buffers to avoid
2150 jbd2_journal_destroy_checkpoint(journal);
2151 spin_lock(&journal->j_list_lock);
2154 spin_lock(&journal->j_list_lock);
2157 J_ASSERT(journal->j_running_transaction == NULL);
2158 J_ASSERT(journal->j_committing_transaction == NULL);
2159 J_ASSERT(journal->j_checkpoint_transactions == NULL);
2160 spin_unlock(&journal->j_list_lock);
2163 * OK, all checkpoint transactions have been checked, now check the
2164 * write out io error flag and abort the journal if some buffer failed
2165 * to write back to the original location, otherwise the filesystem
2166 * may become inconsistent.
2168 if (!is_journal_aborted(journal) &&
2169 test_bit(JBD2_CHECKPOINT_IO_ERROR, &journal->j_atomic_flags))
2170 jbd2_journal_abort(journal, -EIO);
2172 if (journal->j_sb_buffer) {
2173 if (!is_journal_aborted(journal)) {
2174 mutex_lock_io(&journal->j_checkpoint_mutex);
2176 write_lock(&journal->j_state_lock);
2177 journal->j_tail_sequence =
2178 ++journal->j_transaction_sequence;
2179 write_unlock(&journal->j_state_lock);
2181 jbd2_mark_journal_empty(journal,
2182 REQ_SYNC | REQ_PREFLUSH | REQ_FUA);
2183 mutex_unlock(&journal->j_checkpoint_mutex);
2186 brelse(journal->j_sb_buffer);
2189 if (journal->j_shrinker.flags & SHRINKER_REGISTERED) {
2190 percpu_counter_destroy(&journal->j_checkpoint_jh_count);
2191 unregister_shrinker(&journal->j_shrinker);
2193 if (journal->j_proc_entry)
2194 jbd2_stats_proc_exit(journal);
2195 iput(journal->j_inode);
2196 if (journal->j_revoke)
2197 jbd2_journal_destroy_revoke(journal);
2198 if (journal->j_chksum_driver)
2199 crypto_free_shash(journal->j_chksum_driver);
2200 kfree(journal->j_fc_wbuf);
2201 kfree(journal->j_wbuf);
2209 * jbd2_journal_check_used_features() - Check if features specified are used.
2210 * @journal: Journal to check.
2211 * @compat: bitmask of compatible features
2212 * @ro: bitmask of features that force read-only mount
2213 * @incompat: bitmask of incompatible features
2215 * Check whether the journal uses all of a given set of
2216 * features. Return true (non-zero) if it does.
2219 int jbd2_journal_check_used_features(journal_t *journal, unsigned long compat,
2220 unsigned long ro, unsigned long incompat)
2222 journal_superblock_t *sb;
2224 if (!compat && !ro && !incompat)
2226 /* Load journal superblock if it is not loaded yet. */
2227 if (journal->j_format_version == 0 &&
2228 journal_get_superblock(journal) != 0)
2230 if (journal->j_format_version == 1)
2233 sb = journal->j_superblock;
2235 if (((be32_to_cpu(sb->s_feature_compat) & compat) == compat) &&
2236 ((be32_to_cpu(sb->s_feature_ro_compat) & ro) == ro) &&
2237 ((be32_to_cpu(sb->s_feature_incompat) & incompat) == incompat))
2244 * jbd2_journal_check_available_features() - Check feature set in journalling layer
2245 * @journal: Journal to check.
2246 * @compat: bitmask of compatible features
2247 * @ro: bitmask of features that force read-only mount
2248 * @incompat: bitmask of incompatible features
2250 * Check whether the journaling code supports the use of
2251 * all of a given set of features on this journal. Return true
2252 * (non-zero) if it can. */
2254 int jbd2_journal_check_available_features(journal_t *journal, unsigned long compat,
2255 unsigned long ro, unsigned long incompat)
2257 if (!compat && !ro && !incompat)
2260 /* We can support any known requested features iff the
2261 * superblock is in version 2. Otherwise we fail to support any
2262 * extended sb features. */
2264 if (journal->j_format_version != 2)
2267 if ((compat & JBD2_KNOWN_COMPAT_FEATURES) == compat &&
2268 (ro & JBD2_KNOWN_ROCOMPAT_FEATURES) == ro &&
2269 (incompat & JBD2_KNOWN_INCOMPAT_FEATURES) == incompat)
2276 jbd2_journal_initialize_fast_commit(journal_t *journal)
2278 journal_superblock_t *sb = journal->j_superblock;
2279 unsigned long long num_fc_blks;
2281 num_fc_blks = jbd2_journal_get_num_fc_blks(sb);
2282 if (journal->j_last - num_fc_blks < JBD2_MIN_JOURNAL_BLOCKS)
2285 /* Are we called twice? */
2286 WARN_ON(journal->j_fc_wbuf != NULL);
2287 journal->j_fc_wbuf = kmalloc_array(num_fc_blks,
2288 sizeof(struct buffer_head *), GFP_KERNEL);
2289 if (!journal->j_fc_wbuf)
2292 journal->j_fc_wbufsize = num_fc_blks;
2293 journal->j_fc_last = journal->j_last;
2294 journal->j_last = journal->j_fc_last - num_fc_blks;
2295 journal->j_fc_first = journal->j_last + 1;
2296 journal->j_fc_off = 0;
2297 journal->j_free = journal->j_last - journal->j_first;
2298 journal->j_max_transaction_buffers =
2299 jbd2_journal_get_max_txn_bufs(journal);
2305 * jbd2_journal_set_features() - Mark a given journal feature in the superblock
2306 * @journal: Journal to act on.
2307 * @compat: bitmask of compatible features
2308 * @ro: bitmask of features that force read-only mount
2309 * @incompat: bitmask of incompatible features
2311 * Mark a given journal feature as present on the
2312 * superblock. Returns true if the requested features could be set.
2316 int jbd2_journal_set_features(journal_t *journal, unsigned long compat,
2317 unsigned long ro, unsigned long incompat)
2319 #define INCOMPAT_FEATURE_ON(f) \
2320 ((incompat & (f)) && !(sb->s_feature_incompat & cpu_to_be32(f)))
2321 #define COMPAT_FEATURE_ON(f) \
2322 ((compat & (f)) && !(sb->s_feature_compat & cpu_to_be32(f)))
2323 journal_superblock_t *sb;
2325 if (jbd2_journal_check_used_features(journal, compat, ro, incompat))
2328 if (!jbd2_journal_check_available_features(journal, compat, ro, incompat))
2331 /* If enabling v2 checksums, turn on v3 instead */
2332 if (incompat & JBD2_FEATURE_INCOMPAT_CSUM_V2) {
2333 incompat &= ~JBD2_FEATURE_INCOMPAT_CSUM_V2;
2334 incompat |= JBD2_FEATURE_INCOMPAT_CSUM_V3;
2337 /* Asking for checksumming v3 and v1? Only give them v3. */
2338 if (incompat & JBD2_FEATURE_INCOMPAT_CSUM_V3 &&
2339 compat & JBD2_FEATURE_COMPAT_CHECKSUM)
2340 compat &= ~JBD2_FEATURE_COMPAT_CHECKSUM;
2342 jbd2_debug(1, "Setting new features 0x%lx/0x%lx/0x%lx\n",
2343 compat, ro, incompat);
2345 sb = journal->j_superblock;
2347 if (incompat & JBD2_FEATURE_INCOMPAT_FAST_COMMIT) {
2348 if (jbd2_journal_initialize_fast_commit(journal)) {
2349 pr_err("JBD2: Cannot enable fast commits.\n");
2354 /* Load the checksum driver if necessary */
2355 if ((journal->j_chksum_driver == NULL) &&
2356 INCOMPAT_FEATURE_ON(JBD2_FEATURE_INCOMPAT_CSUM_V3)) {
2357 journal->j_chksum_driver = crypto_alloc_shash("crc32c", 0, 0);
2358 if (IS_ERR(journal->j_chksum_driver)) {
2359 printk(KERN_ERR "JBD2: Cannot load crc32c driver.\n");
2360 journal->j_chksum_driver = NULL;
2363 /* Precompute checksum seed for all metadata */
2364 journal->j_csum_seed = jbd2_chksum(journal, ~0, sb->s_uuid,
2365 sizeof(sb->s_uuid));
2368 lock_buffer(journal->j_sb_buffer);
2370 /* If enabling v3 checksums, update superblock */
2371 if (INCOMPAT_FEATURE_ON(JBD2_FEATURE_INCOMPAT_CSUM_V3)) {
2372 sb->s_checksum_type = JBD2_CRC32C_CHKSUM;
2373 sb->s_feature_compat &=
2374 ~cpu_to_be32(JBD2_FEATURE_COMPAT_CHECKSUM);
2377 /* If enabling v1 checksums, downgrade superblock */
2378 if (COMPAT_FEATURE_ON(JBD2_FEATURE_COMPAT_CHECKSUM))
2379 sb->s_feature_incompat &=
2380 ~cpu_to_be32(JBD2_FEATURE_INCOMPAT_CSUM_V2 |
2381 JBD2_FEATURE_INCOMPAT_CSUM_V3);
2383 sb->s_feature_compat |= cpu_to_be32(compat);
2384 sb->s_feature_ro_compat |= cpu_to_be32(ro);
2385 sb->s_feature_incompat |= cpu_to_be32(incompat);
2386 unlock_buffer(journal->j_sb_buffer);
2387 journal->j_revoke_records_per_block =
2388 journal_revoke_records_per_block(journal);
2391 #undef COMPAT_FEATURE_ON
2392 #undef INCOMPAT_FEATURE_ON
2396 * jbd2_journal_clear_features() - Clear a given journal feature in the
2398 * @journal: Journal to act on.
2399 * @compat: bitmask of compatible features
2400 * @ro: bitmask of features that force read-only mount
2401 * @incompat: bitmask of incompatible features
2403 * Clear a given journal feature as present on the
2406 void jbd2_journal_clear_features(journal_t *journal, unsigned long compat,
2407 unsigned long ro, unsigned long incompat)
2409 journal_superblock_t *sb;
2411 jbd2_debug(1, "Clear features 0x%lx/0x%lx/0x%lx\n",
2412 compat, ro, incompat);
2414 sb = journal->j_superblock;
2416 sb->s_feature_compat &= ~cpu_to_be32(compat);
2417 sb->s_feature_ro_compat &= ~cpu_to_be32(ro);
2418 sb->s_feature_incompat &= ~cpu_to_be32(incompat);
2419 journal->j_revoke_records_per_block =
2420 journal_revoke_records_per_block(journal);
2422 EXPORT_SYMBOL(jbd2_journal_clear_features);
2425 * jbd2_journal_flush() - Flush journal
2426 * @journal: Journal to act on.
2427 * @flags: optional operation on the journal blocks after the flush (see below)
2429 * Flush all data for a given journal to disk and empty the journal.
2430 * Filesystems can use this when remounting readonly to ensure that
2431 * recovery does not need to happen on remount. Optionally, a discard or zeroout
2432 * can be issued on the journal blocks after flushing.
2435 * JBD2_JOURNAL_FLUSH_DISCARD: issues discards for the journal blocks
2436 * JBD2_JOURNAL_FLUSH_ZEROOUT: issues zeroouts for the journal blocks
2438 int jbd2_journal_flush(journal_t *journal, unsigned int flags)
2441 transaction_t *transaction = NULL;
2443 write_lock(&journal->j_state_lock);
2445 /* Force everything buffered to the log... */
2446 if (journal->j_running_transaction) {
2447 transaction = journal->j_running_transaction;
2448 __jbd2_log_start_commit(journal, transaction->t_tid);
2449 } else if (journal->j_committing_transaction)
2450 transaction = journal->j_committing_transaction;
2452 /* Wait for the log commit to complete... */
2454 tid_t tid = transaction->t_tid;
2456 write_unlock(&journal->j_state_lock);
2457 jbd2_log_wait_commit(journal, tid);
2459 write_unlock(&journal->j_state_lock);
2462 /* ...and flush everything in the log out to disk. */
2463 spin_lock(&journal->j_list_lock);
2464 while (!err && journal->j_checkpoint_transactions != NULL) {
2465 spin_unlock(&journal->j_list_lock);
2466 mutex_lock_io(&journal->j_checkpoint_mutex);
2467 err = jbd2_log_do_checkpoint(journal);
2468 mutex_unlock(&journal->j_checkpoint_mutex);
2469 spin_lock(&journal->j_list_lock);
2471 spin_unlock(&journal->j_list_lock);
2473 if (is_journal_aborted(journal))
2476 mutex_lock_io(&journal->j_checkpoint_mutex);
2478 err = jbd2_cleanup_journal_tail(journal);
2480 mutex_unlock(&journal->j_checkpoint_mutex);
2486 /* Finally, mark the journal as really needing no recovery.
2487 * This sets s_start==0 in the underlying superblock, which is
2488 * the magic code for a fully-recovered superblock. Any future
2489 * commits of data to the journal will restore the current
2491 jbd2_mark_journal_empty(journal, REQ_SYNC | REQ_FUA);
2494 err = __jbd2_journal_erase(journal, flags);
2496 mutex_unlock(&journal->j_checkpoint_mutex);
2497 write_lock(&journal->j_state_lock);
2498 J_ASSERT(!journal->j_running_transaction);
2499 J_ASSERT(!journal->j_committing_transaction);
2500 J_ASSERT(!journal->j_checkpoint_transactions);
2501 J_ASSERT(journal->j_head == journal->j_tail);
2502 J_ASSERT(journal->j_tail_sequence == journal->j_transaction_sequence);
2503 write_unlock(&journal->j_state_lock);
2509 * jbd2_journal_wipe() - Wipe journal contents
2510 * @journal: Journal to act on.
2511 * @write: flag (see below)
2513 * Wipe out all of the contents of a journal, safely. This will produce
2514 * a warning if the journal contains any valid recovery information.
2515 * Must be called between journal_init_*() and jbd2_journal_load().
2517 * If 'write' is non-zero, then we wipe out the journal on disk; otherwise
2518 * we merely suppress recovery.
2521 int jbd2_journal_wipe(journal_t *journal, int write)
2525 J_ASSERT (!(journal->j_flags & JBD2_LOADED));
2527 err = load_superblock(journal);
2531 if (!journal->j_tail)
2534 printk(KERN_WARNING "JBD2: %s recovery information on journal\n",
2535 write ? "Clearing" : "Ignoring");
2537 err = jbd2_journal_skip_recovery(journal);
2539 /* Lock to make assertions happy... */
2540 mutex_lock_io(&journal->j_checkpoint_mutex);
2541 jbd2_mark_journal_empty(journal, REQ_SYNC | REQ_FUA);
2542 mutex_unlock(&journal->j_checkpoint_mutex);
2550 * jbd2_journal_abort () - Shutdown the journal immediately.
2551 * @journal: the journal to shutdown.
2552 * @errno: an error number to record in the journal indicating
2553 * the reason for the shutdown.
2555 * Perform a complete, immediate shutdown of the ENTIRE
2556 * journal (not of a single transaction). This operation cannot be
2557 * undone without closing and reopening the journal.
2559 * The jbd2_journal_abort function is intended to support higher level error
2560 * recovery mechanisms such as the ext2/ext3 remount-readonly error
2563 * Journal abort has very specific semantics. Any existing dirty,
2564 * unjournaled buffers in the main filesystem will still be written to
2565 * disk by bdflush, but the journaling mechanism will be suspended
2566 * immediately and no further transaction commits will be honoured.
2568 * Any dirty, journaled buffers will be written back to disk without
2569 * hitting the journal. Atomicity cannot be guaranteed on an aborted
2570 * filesystem, but we _do_ attempt to leave as much data as possible
2571 * behind for fsck to use for cleanup.
2573 * Any attempt to get a new transaction handle on a journal which is in
2574 * ABORT state will just result in an -EROFS error return. A
2575 * jbd2_journal_stop on an existing handle will return -EIO if we have
2576 * entered abort state during the update.
2578 * Recursive transactions are not disturbed by journal abort until the
2579 * final jbd2_journal_stop, which will receive the -EIO error.
2581 * Finally, the jbd2_journal_abort call allows the caller to supply an errno
2582 * which will be recorded (if possible) in the journal superblock. This
2583 * allows a client to record failure conditions in the middle of a
2584 * transaction without having to complete the transaction to record the
2585 * failure to disk. ext3_error, for example, now uses this
2590 void jbd2_journal_abort(journal_t *journal, int errno)
2592 transaction_t *transaction;
2595 * Lock the aborting procedure until everything is done, this avoid
2596 * races between filesystem's error handling flow (e.g. ext4_abort()),
2597 * ensure panic after the error info is written into journal's
2600 mutex_lock(&journal->j_abort_mutex);
2602 * ESHUTDOWN always takes precedence because a file system check
2603 * caused by any other journal abort error is not required after
2604 * a shutdown triggered.
2606 write_lock(&journal->j_state_lock);
2607 if (journal->j_flags & JBD2_ABORT) {
2608 int old_errno = journal->j_errno;
2610 write_unlock(&journal->j_state_lock);
2611 if (old_errno != -ESHUTDOWN && errno == -ESHUTDOWN) {
2612 journal->j_errno = errno;
2613 jbd2_journal_update_sb_errno(journal);
2615 mutex_unlock(&journal->j_abort_mutex);
2620 * Mark the abort as occurred and start current running transaction
2621 * to release all journaled buffer.
2623 pr_err("Aborting journal on device %s.\n", journal->j_devname);
2625 journal->j_flags |= JBD2_ABORT;
2626 journal->j_errno = errno;
2627 transaction = journal->j_running_transaction;
2629 __jbd2_log_start_commit(journal, transaction->t_tid);
2630 write_unlock(&journal->j_state_lock);
2633 * Record errno to the journal super block, so that fsck and jbd2
2634 * layer could realise that a filesystem check is needed.
2636 jbd2_journal_update_sb_errno(journal);
2637 mutex_unlock(&journal->j_abort_mutex);
2641 * jbd2_journal_errno() - returns the journal's error state.
2642 * @journal: journal to examine.
2644 * This is the errno number set with jbd2_journal_abort(), the last
2645 * time the journal was mounted - if the journal was stopped
2646 * without calling abort this will be 0.
2648 * If the journal has been aborted on this mount time -EROFS will
2651 int jbd2_journal_errno(journal_t *journal)
2655 read_lock(&journal->j_state_lock);
2656 if (journal->j_flags & JBD2_ABORT)
2659 err = journal->j_errno;
2660 read_unlock(&journal->j_state_lock);
2665 * jbd2_journal_clear_err() - clears the journal's error state
2666 * @journal: journal to act on.
2668 * An error must be cleared or acked to take a FS out of readonly
2671 int jbd2_journal_clear_err(journal_t *journal)
2675 write_lock(&journal->j_state_lock);
2676 if (journal->j_flags & JBD2_ABORT)
2679 journal->j_errno = 0;
2680 write_unlock(&journal->j_state_lock);
2685 * jbd2_journal_ack_err() - Ack journal err.
2686 * @journal: journal to act on.
2688 * An error must be cleared or acked to take a FS out of readonly
2691 void jbd2_journal_ack_err(journal_t *journal)
2693 write_lock(&journal->j_state_lock);
2694 if (journal->j_errno)
2695 journal->j_flags |= JBD2_ACK_ERR;
2696 write_unlock(&journal->j_state_lock);
2699 int jbd2_journal_blocks_per_page(struct inode *inode)
2701 return 1 << (PAGE_SHIFT - inode->i_sb->s_blocksize_bits);
2705 * helper functions to deal with 32 or 64bit block numbers.
2707 size_t journal_tag_bytes(journal_t *journal)
2711 if (jbd2_has_feature_csum3(journal))
2712 return sizeof(journal_block_tag3_t);
2714 sz = sizeof(journal_block_tag_t);
2716 if (jbd2_has_feature_csum2(journal))
2717 sz += sizeof(__u16);
2719 if (jbd2_has_feature_64bit(journal))
2722 return sz - sizeof(__u32);
2726 * JBD memory management
2728 * These functions are used to allocate block-sized chunks of memory
2729 * used for making copies of buffer_head data. Very often it will be
2730 * page-sized chunks of data, but sometimes it will be in
2731 * sub-page-size chunks. (For example, 16k pages on Power systems
2732 * with a 4k block file system.) For blocks smaller than a page, we
2733 * use a SLAB allocator. There are slab caches for each block size,
2734 * which are allocated at mount time, if necessary, and we only free
2735 * (all of) the slab caches when/if the jbd2 module is unloaded. For
2736 * this reason we don't need to a mutex to protect access to
2737 * jbd2_slab[] allocating or releasing memory; only in
2738 * jbd2_journal_create_slab().
2740 #define JBD2_MAX_SLABS 8
2741 static struct kmem_cache *jbd2_slab[JBD2_MAX_SLABS];
2743 static const char *jbd2_slab_names[JBD2_MAX_SLABS] = {
2744 "jbd2_1k", "jbd2_2k", "jbd2_4k", "jbd2_8k",
2745 "jbd2_16k", "jbd2_32k", "jbd2_64k", "jbd2_128k"
2749 static void jbd2_journal_destroy_slabs(void)
2753 for (i = 0; i < JBD2_MAX_SLABS; i++) {
2754 kmem_cache_destroy(jbd2_slab[i]);
2755 jbd2_slab[i] = NULL;
2759 static int jbd2_journal_create_slab(size_t size)
2761 static DEFINE_MUTEX(jbd2_slab_create_mutex);
2762 int i = order_base_2(size) - 10;
2765 if (size == PAGE_SIZE)
2768 if (i >= JBD2_MAX_SLABS)
2771 if (unlikely(i < 0))
2773 mutex_lock(&jbd2_slab_create_mutex);
2775 mutex_unlock(&jbd2_slab_create_mutex);
2776 return 0; /* Already created */
2779 slab_size = 1 << (i+10);
2780 jbd2_slab[i] = kmem_cache_create(jbd2_slab_names[i], slab_size,
2781 slab_size, 0, NULL);
2782 mutex_unlock(&jbd2_slab_create_mutex);
2783 if (!jbd2_slab[i]) {
2784 printk(KERN_EMERG "JBD2: no memory for jbd2_slab cache\n");
2790 static struct kmem_cache *get_slab(size_t size)
2792 int i = order_base_2(size) - 10;
2794 BUG_ON(i >= JBD2_MAX_SLABS);
2795 if (unlikely(i < 0))
2797 BUG_ON(jbd2_slab[i] == NULL);
2798 return jbd2_slab[i];
2801 void *jbd2_alloc(size_t size, gfp_t flags)
2805 BUG_ON(size & (size-1)); /* Must be a power of 2 */
2807 if (size < PAGE_SIZE)
2808 ptr = kmem_cache_alloc(get_slab(size), flags);
2810 ptr = (void *)__get_free_pages(flags, get_order(size));
2812 /* Check alignment; SLUB has gotten this wrong in the past,
2813 * and this can lead to user data corruption! */
2814 BUG_ON(((unsigned long) ptr) & (size-1));
2819 void jbd2_free(void *ptr, size_t size)
2821 if (size < PAGE_SIZE)
2822 kmem_cache_free(get_slab(size), ptr);
2824 free_pages((unsigned long)ptr, get_order(size));
2828 * Journal_head storage management
2830 static struct kmem_cache *jbd2_journal_head_cache;
2831 #ifdef CONFIG_JBD2_DEBUG
2832 static atomic_t nr_journal_heads = ATOMIC_INIT(0);
2835 static int __init jbd2_journal_init_journal_head_cache(void)
2837 J_ASSERT(!jbd2_journal_head_cache);
2838 jbd2_journal_head_cache = kmem_cache_create("jbd2_journal_head",
2839 sizeof(struct journal_head),
2841 SLAB_TEMPORARY | SLAB_TYPESAFE_BY_RCU,
2843 if (!jbd2_journal_head_cache) {
2844 printk(KERN_EMERG "JBD2: no memory for journal_head cache\n");
2850 static void jbd2_journal_destroy_journal_head_cache(void)
2852 kmem_cache_destroy(jbd2_journal_head_cache);
2853 jbd2_journal_head_cache = NULL;
2857 * journal_head splicing and dicing
2859 static struct journal_head *journal_alloc_journal_head(void)
2861 struct journal_head *ret;
2863 #ifdef CONFIG_JBD2_DEBUG
2864 atomic_inc(&nr_journal_heads);
2866 ret = kmem_cache_zalloc(jbd2_journal_head_cache, GFP_NOFS);
2868 jbd2_debug(1, "out of memory for journal_head\n");
2869 pr_notice_ratelimited("ENOMEM in %s, retrying.\n", __func__);
2870 ret = kmem_cache_zalloc(jbd2_journal_head_cache,
2871 GFP_NOFS | __GFP_NOFAIL);
2874 spin_lock_init(&ret->b_state_lock);
2878 static void journal_free_journal_head(struct journal_head *jh)
2880 #ifdef CONFIG_JBD2_DEBUG
2881 atomic_dec(&nr_journal_heads);
2882 memset(jh, JBD2_POISON_FREE, sizeof(*jh));
2884 kmem_cache_free(jbd2_journal_head_cache, jh);
2888 * A journal_head is attached to a buffer_head whenever JBD has an
2889 * interest in the buffer.
2891 * Whenever a buffer has an attached journal_head, its ->b_state:BH_JBD bit
2892 * is set. This bit is tested in core kernel code where we need to take
2893 * JBD-specific actions. Testing the zeroness of ->b_private is not reliable
2896 * When a buffer has its BH_JBD bit set, its ->b_count is elevated by one.
2898 * When a buffer has its BH_JBD bit set it is immune from being released by
2899 * core kernel code, mainly via ->b_count.
2901 * A journal_head is detached from its buffer_head when the journal_head's
2902 * b_jcount reaches zero. Running transaction (b_transaction) and checkpoint
2903 * transaction (b_cp_transaction) hold their references to b_jcount.
2905 * Various places in the kernel want to attach a journal_head to a buffer_head
2906 * _before_ attaching the journal_head to a transaction. To protect the
2907 * journal_head in this situation, jbd2_journal_add_journal_head elevates the
2908 * journal_head's b_jcount refcount by one. The caller must call
2909 * jbd2_journal_put_journal_head() to undo this.
2911 * So the typical usage would be:
2913 * (Attach a journal_head if needed. Increments b_jcount)
2914 * struct journal_head *jh = jbd2_journal_add_journal_head(bh);
2916 * (Get another reference for transaction)
2917 * jbd2_journal_grab_journal_head(bh);
2918 * jh->b_transaction = xxx;
2919 * (Put original reference)
2920 * jbd2_journal_put_journal_head(jh);
2924 * Give a buffer_head a journal_head.
2928 struct journal_head *jbd2_journal_add_journal_head(struct buffer_head *bh)
2930 struct journal_head *jh;
2931 struct journal_head *new_jh = NULL;
2934 if (!buffer_jbd(bh))
2935 new_jh = journal_alloc_journal_head();
2937 jbd_lock_bh_journal_head(bh);
2938 if (buffer_jbd(bh)) {
2942 (atomic_read(&bh->b_count) > 0) ||
2943 (bh->b_page && bh->b_page->mapping));
2946 jbd_unlock_bh_journal_head(bh);
2951 new_jh = NULL; /* We consumed it */
2956 BUFFER_TRACE(bh, "added journal_head");
2959 jbd_unlock_bh_journal_head(bh);
2961 journal_free_journal_head(new_jh);
2962 return bh->b_private;
2966 * Grab a ref against this buffer_head's journal_head. If it ended up not
2967 * having a journal_head, return NULL
2969 struct journal_head *jbd2_journal_grab_journal_head(struct buffer_head *bh)
2971 struct journal_head *jh = NULL;
2973 jbd_lock_bh_journal_head(bh);
2974 if (buffer_jbd(bh)) {
2978 jbd_unlock_bh_journal_head(bh);
2981 EXPORT_SYMBOL(jbd2_journal_grab_journal_head);
2983 static void __journal_remove_journal_head(struct buffer_head *bh)
2985 struct journal_head *jh = bh2jh(bh);
2987 J_ASSERT_JH(jh, jh->b_transaction == NULL);
2988 J_ASSERT_JH(jh, jh->b_next_transaction == NULL);
2989 J_ASSERT_JH(jh, jh->b_cp_transaction == NULL);
2990 J_ASSERT_JH(jh, jh->b_jlist == BJ_None);
2991 J_ASSERT_BH(bh, buffer_jbd(bh));
2992 J_ASSERT_BH(bh, jh2bh(jh) == bh);
2993 BUFFER_TRACE(bh, "remove journal_head");
2995 /* Unlink before dropping the lock */
2996 bh->b_private = NULL;
2997 jh->b_bh = NULL; /* debug, really */
2998 clear_buffer_jbd(bh);
3001 static void journal_release_journal_head(struct journal_head *jh, size_t b_size)
3003 if (jh->b_frozen_data) {
3004 printk(KERN_WARNING "%s: freeing b_frozen_data\n", __func__);
3005 jbd2_free(jh->b_frozen_data, b_size);
3007 if (jh->b_committed_data) {
3008 printk(KERN_WARNING "%s: freeing b_committed_data\n", __func__);
3009 jbd2_free(jh->b_committed_data, b_size);
3011 journal_free_journal_head(jh);
3015 * Drop a reference on the passed journal_head. If it fell to zero then
3016 * release the journal_head from the buffer_head.
3018 void jbd2_journal_put_journal_head(struct journal_head *jh)
3020 struct buffer_head *bh = jh2bh(jh);
3022 jbd_lock_bh_journal_head(bh);
3023 J_ASSERT_JH(jh, jh->b_jcount > 0);
3025 if (!jh->b_jcount) {
3026 __journal_remove_journal_head(bh);
3027 jbd_unlock_bh_journal_head(bh);
3028 journal_release_journal_head(jh, bh->b_size);
3031 jbd_unlock_bh_journal_head(bh);
3034 EXPORT_SYMBOL(jbd2_journal_put_journal_head);
3037 * Initialize jbd inode head
3039 void jbd2_journal_init_jbd_inode(struct jbd2_inode *jinode, struct inode *inode)
3041 jinode->i_transaction = NULL;
3042 jinode->i_next_transaction = NULL;
3043 jinode->i_vfs_inode = inode;
3044 jinode->i_flags = 0;
3045 jinode->i_dirty_start = 0;
3046 jinode->i_dirty_end = 0;
3047 INIT_LIST_HEAD(&jinode->i_list);
3051 * Function to be called before we start removing inode from memory (i.e.,
3052 * clear_inode() is a fine place to be called from). It removes inode from
3053 * transaction's lists.
3055 void jbd2_journal_release_jbd_inode(journal_t *journal,
3056 struct jbd2_inode *jinode)
3061 spin_lock(&journal->j_list_lock);
3062 /* Is commit writing out inode - we have to wait */
3063 if (jinode->i_flags & JI_COMMIT_RUNNING) {
3064 wait_queue_head_t *wq;
3065 DEFINE_WAIT_BIT(wait, &jinode->i_flags, __JI_COMMIT_RUNNING);
3066 wq = bit_waitqueue(&jinode->i_flags, __JI_COMMIT_RUNNING);
3067 prepare_to_wait(wq, &wait.wq_entry, TASK_UNINTERRUPTIBLE);
3068 spin_unlock(&journal->j_list_lock);
3070 finish_wait(wq, &wait.wq_entry);
3074 if (jinode->i_transaction) {
3075 list_del(&jinode->i_list);
3076 jinode->i_transaction = NULL;
3078 spin_unlock(&journal->j_list_lock);
3082 #ifdef CONFIG_PROC_FS
3084 #define JBD2_STATS_PROC_NAME "fs/jbd2"
3086 static void __init jbd2_create_jbd_stats_proc_entry(void)
3088 proc_jbd2_stats = proc_mkdir(JBD2_STATS_PROC_NAME, NULL);
3091 static void __exit jbd2_remove_jbd_stats_proc_entry(void)
3093 if (proc_jbd2_stats)
3094 remove_proc_entry(JBD2_STATS_PROC_NAME, NULL);
3099 #define jbd2_create_jbd_stats_proc_entry() do {} while (0)
3100 #define jbd2_remove_jbd_stats_proc_entry() do {} while (0)
3104 struct kmem_cache *jbd2_handle_cache, *jbd2_inode_cache;
3106 static int __init jbd2_journal_init_inode_cache(void)
3108 J_ASSERT(!jbd2_inode_cache);
3109 jbd2_inode_cache = KMEM_CACHE(jbd2_inode, 0);
3110 if (!jbd2_inode_cache) {
3111 pr_emerg("JBD2: failed to create inode cache\n");
3117 static int __init jbd2_journal_init_handle_cache(void)
3119 J_ASSERT(!jbd2_handle_cache);
3120 jbd2_handle_cache = KMEM_CACHE(jbd2_journal_handle, SLAB_TEMPORARY);
3121 if (!jbd2_handle_cache) {
3122 printk(KERN_EMERG "JBD2: failed to create handle cache\n");
3128 static void jbd2_journal_destroy_inode_cache(void)
3130 kmem_cache_destroy(jbd2_inode_cache);
3131 jbd2_inode_cache = NULL;
3134 static void jbd2_journal_destroy_handle_cache(void)
3136 kmem_cache_destroy(jbd2_handle_cache);
3137 jbd2_handle_cache = NULL;
3141 * Module startup and shutdown
3144 static int __init journal_init_caches(void)
3148 ret = jbd2_journal_init_revoke_record_cache();
3150 ret = jbd2_journal_init_revoke_table_cache();
3152 ret = jbd2_journal_init_journal_head_cache();
3154 ret = jbd2_journal_init_handle_cache();
3156 ret = jbd2_journal_init_inode_cache();
3158 ret = jbd2_journal_init_transaction_cache();
3162 static void jbd2_journal_destroy_caches(void)
3164 jbd2_journal_destroy_revoke_record_cache();
3165 jbd2_journal_destroy_revoke_table_cache();
3166 jbd2_journal_destroy_journal_head_cache();
3167 jbd2_journal_destroy_handle_cache();
3168 jbd2_journal_destroy_inode_cache();
3169 jbd2_journal_destroy_transaction_cache();
3170 jbd2_journal_destroy_slabs();
3173 static int __init journal_init(void)
3177 BUILD_BUG_ON(sizeof(struct journal_superblock_s) != 1024);
3179 ret = journal_init_caches();
3181 jbd2_create_jbd_stats_proc_entry();
3183 jbd2_journal_destroy_caches();
3188 static void __exit journal_exit(void)
3190 #ifdef CONFIG_JBD2_DEBUG
3191 int n = atomic_read(&nr_journal_heads);
3193 printk(KERN_ERR "JBD2: leaked %d journal_heads!\n", n);
3195 jbd2_remove_jbd_stats_proc_entry();
3196 jbd2_journal_destroy_caches();
3199 MODULE_LICENSE("GPL");
3200 module_init(journal_init);
3201 module_exit(journal_exit);