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 static ushort jbd2_journal_enable_debug __read_mostly;
54 module_param_named(jbd2_debug, jbd2_journal_enable_debug, ushort, 0644);
55 MODULE_PARM_DESC(jbd2_debug, "Debugging level for jbd2");
58 EXPORT_SYMBOL(jbd2_journal_extend);
59 EXPORT_SYMBOL(jbd2_journal_stop);
60 EXPORT_SYMBOL(jbd2_journal_lock_updates);
61 EXPORT_SYMBOL(jbd2_journal_unlock_updates);
62 EXPORT_SYMBOL(jbd2_journal_get_write_access);
63 EXPORT_SYMBOL(jbd2_journal_get_create_access);
64 EXPORT_SYMBOL(jbd2_journal_get_undo_access);
65 EXPORT_SYMBOL(jbd2_journal_set_triggers);
66 EXPORT_SYMBOL(jbd2_journal_dirty_metadata);
67 EXPORT_SYMBOL(jbd2_journal_forget);
68 EXPORT_SYMBOL(jbd2_journal_flush);
69 EXPORT_SYMBOL(jbd2_journal_revoke);
71 EXPORT_SYMBOL(jbd2_journal_init_dev);
72 EXPORT_SYMBOL(jbd2_journal_init_inode);
73 EXPORT_SYMBOL(jbd2_journal_check_used_features);
74 EXPORT_SYMBOL(jbd2_journal_check_available_features);
75 EXPORT_SYMBOL(jbd2_journal_set_features);
76 EXPORT_SYMBOL(jbd2_journal_load);
77 EXPORT_SYMBOL(jbd2_journal_destroy);
78 EXPORT_SYMBOL(jbd2_journal_abort);
79 EXPORT_SYMBOL(jbd2_journal_errno);
80 EXPORT_SYMBOL(jbd2_journal_ack_err);
81 EXPORT_SYMBOL(jbd2_journal_clear_err);
82 EXPORT_SYMBOL(jbd2_log_wait_commit);
83 EXPORT_SYMBOL(jbd2_journal_start_commit);
84 EXPORT_SYMBOL(jbd2_journal_force_commit_nested);
85 EXPORT_SYMBOL(jbd2_journal_wipe);
86 EXPORT_SYMBOL(jbd2_journal_blocks_per_page);
87 EXPORT_SYMBOL(jbd2_journal_invalidate_folio);
88 EXPORT_SYMBOL(jbd2_journal_try_to_free_buffers);
89 EXPORT_SYMBOL(jbd2_journal_force_commit);
90 EXPORT_SYMBOL(jbd2_journal_inode_ranged_write);
91 EXPORT_SYMBOL(jbd2_journal_inode_ranged_wait);
92 EXPORT_SYMBOL(jbd2_journal_submit_inode_data_buffers);
93 EXPORT_SYMBOL(jbd2_journal_finish_inode_data_buffers);
94 EXPORT_SYMBOL(jbd2_journal_init_jbd_inode);
95 EXPORT_SYMBOL(jbd2_journal_release_jbd_inode);
96 EXPORT_SYMBOL(jbd2_journal_begin_ordered_truncate);
97 EXPORT_SYMBOL(jbd2_inode_cache);
99 static int jbd2_journal_create_slab(size_t slab_size);
101 #ifdef CONFIG_JBD2_DEBUG
102 void __jbd2_debug(int level, const char *file, const char *func,
103 unsigned int line, const char *fmt, ...)
105 struct va_format vaf;
108 if (level > jbd2_journal_enable_debug)
113 printk(KERN_DEBUG "%s: (%s, %u): %pV", file, func, line, &vaf);
118 /* Checksumming functions */
119 static int jbd2_verify_csum_type(journal_t *j, journal_superblock_t *sb)
121 if (!jbd2_journal_has_csum_v2or3_feature(j))
124 return sb->s_checksum_type == JBD2_CRC32C_CHKSUM;
127 static __be32 jbd2_superblock_csum(journal_t *j, journal_superblock_t *sb)
132 old_csum = sb->s_checksum;
134 csum = jbd2_chksum(j, ~0, (char *)sb, sizeof(journal_superblock_t));
135 sb->s_checksum = old_csum;
137 return cpu_to_be32(csum);
141 * Helper function used to manage commit timeouts
144 static void commit_timeout(struct timer_list *t)
146 journal_t *journal = from_timer(journal, t, j_commit_timer);
148 wake_up_process(journal->j_task);
152 * kjournald2: The main thread function used to manage a logging device
155 * This kernel thread is responsible for two things:
157 * 1) COMMIT: Every so often we need to commit the current state of the
158 * filesystem to disk. The journal thread is responsible for writing
159 * all of the metadata buffers to disk. If a fast commit is ongoing
160 * journal thread waits until it's done and then continues from
163 * 2) CHECKPOINT: We cannot reuse a used section of the log file until all
164 * of the data in that part of the log has been rewritten elsewhere on
165 * the disk. Flushing these old buffers to reclaim space in the log is
166 * known as checkpointing, and this thread is responsible for that job.
169 static int kjournald2(void *arg)
171 journal_t *journal = arg;
172 transaction_t *transaction;
175 * Set up an interval timer which can be used to trigger a commit wakeup
176 * after the commit interval expires
178 timer_setup(&journal->j_commit_timer, commit_timeout, 0);
182 /* Record that the journal thread is running */
183 journal->j_task = current;
184 wake_up(&journal->j_wait_done_commit);
187 * Make sure that no allocations from this kernel thread will ever
188 * recurse to the fs layer because we are responsible for the
189 * transaction commit and any fs involvement might get stuck waiting for
192 memalloc_nofs_save();
195 * And now, wait forever for commit wakeup events.
197 write_lock(&journal->j_state_lock);
200 if (journal->j_flags & JBD2_UNMOUNT)
203 jbd2_debug(1, "commit_sequence=%u, commit_request=%u\n",
204 journal->j_commit_sequence, journal->j_commit_request);
206 if (journal->j_commit_sequence != journal->j_commit_request) {
207 jbd2_debug(1, "OK, requests differ\n");
208 write_unlock(&journal->j_state_lock);
209 del_timer_sync(&journal->j_commit_timer);
210 jbd2_journal_commit_transaction(journal);
211 write_lock(&journal->j_state_lock);
215 wake_up(&journal->j_wait_done_commit);
216 if (freezing(current)) {
218 * The simpler the better. Flushing journal isn't a
219 * good idea, because that depends on threads that may
220 * be already stopped.
222 jbd2_debug(1, "Now suspending kjournald2\n");
223 write_unlock(&journal->j_state_lock);
225 write_lock(&journal->j_state_lock);
228 * We assume on resume that commits are already there,
232 int should_sleep = 1;
234 prepare_to_wait(&journal->j_wait_commit, &wait,
236 if (journal->j_commit_sequence != journal->j_commit_request)
238 transaction = journal->j_running_transaction;
239 if (transaction && time_after_eq(jiffies,
240 transaction->t_expires))
242 if (journal->j_flags & JBD2_UNMOUNT)
245 write_unlock(&journal->j_state_lock);
247 write_lock(&journal->j_state_lock);
249 finish_wait(&journal->j_wait_commit, &wait);
252 jbd2_debug(1, "kjournald2 wakes\n");
255 * Were we woken up by a commit wakeup event?
257 transaction = journal->j_running_transaction;
258 if (transaction && time_after_eq(jiffies, transaction->t_expires)) {
259 journal->j_commit_request = transaction->t_tid;
260 jbd2_debug(1, "woke because of timeout\n");
265 del_timer_sync(&journal->j_commit_timer);
266 journal->j_task = NULL;
267 wake_up(&journal->j_wait_done_commit);
268 jbd2_debug(1, "Journal thread exiting.\n");
269 write_unlock(&journal->j_state_lock);
273 static int jbd2_journal_start_thread(journal_t *journal)
275 struct task_struct *t;
277 t = kthread_run(kjournald2, journal, "jbd2/%s",
282 wait_event(journal->j_wait_done_commit, journal->j_task != NULL);
286 static void journal_kill_thread(journal_t *journal)
288 write_lock(&journal->j_state_lock);
289 journal->j_flags |= JBD2_UNMOUNT;
291 while (journal->j_task) {
292 write_unlock(&journal->j_state_lock);
293 wake_up(&journal->j_wait_commit);
294 wait_event(journal->j_wait_done_commit, journal->j_task == NULL);
295 write_lock(&journal->j_state_lock);
297 write_unlock(&journal->j_state_lock);
301 * jbd2_journal_write_metadata_buffer: write a metadata buffer to the journal.
303 * Writes a metadata buffer to a given disk block. The actual IO is not
304 * performed but a new buffer_head is constructed which labels the data
305 * to be written with the correct destination disk block.
307 * Any magic-number escaping which needs to be done will cause a
308 * copy-out here. If the buffer happens to start with the
309 * JBD2_MAGIC_NUMBER, then we can't write it to the log directly: the
310 * magic number is only written to the log for descripter blocks. In
311 * this case, we copy the data and replace the first word with 0, and we
312 * return a result code which indicates that this buffer needs to be
313 * marked as an escaped buffer in the corresponding log descriptor
314 * block. The missing word can then be restored when the block is read
317 * If the source buffer has already been modified by a new transaction
318 * since we took the last commit snapshot, we use the frozen copy of
319 * that data for IO. If we end up using the existing buffer_head's data
320 * for the write, then we have to make sure nobody modifies it while the
321 * IO is in progress. do_get_write_access() handles this.
323 * The function returns a pointer to the buffer_head to be used for IO.
331 * Bit 0 set == escape performed on the data
332 * Bit 1 set == buffer copy-out performed (kfree the data after IO)
335 int jbd2_journal_write_metadata_buffer(transaction_t *transaction,
336 struct journal_head *jh_in,
337 struct buffer_head **bh_out,
340 int need_copy_out = 0;
341 int done_copy_out = 0;
344 struct buffer_head *new_bh;
345 struct page *new_page;
346 unsigned int new_offset;
347 struct buffer_head *bh_in = jh2bh(jh_in);
348 journal_t *journal = transaction->t_journal;
351 * The buffer really shouldn't be locked: only the current committing
352 * transaction is allowed to write it, so nobody else is allowed
355 * akpm: except if we're journalling data, and write() output is
356 * also part of a shared mapping, and another thread has
357 * decided to launch a writepage() against this buffer.
359 J_ASSERT_BH(bh_in, buffer_jbddirty(bh_in));
361 new_bh = alloc_buffer_head(GFP_NOFS|__GFP_NOFAIL);
363 /* keep subsequent assertions sane */
364 atomic_set(&new_bh->b_count, 1);
366 spin_lock(&jh_in->b_state_lock);
369 * If a new transaction has already done a buffer copy-out, then
370 * we use that version of the data for the commit.
372 if (jh_in->b_frozen_data) {
374 new_page = virt_to_page(jh_in->b_frozen_data);
375 new_offset = offset_in_page(jh_in->b_frozen_data);
377 new_page = jh2bh(jh_in)->b_page;
378 new_offset = offset_in_page(jh2bh(jh_in)->b_data);
381 mapped_data = kmap_atomic(new_page);
383 * Fire data frozen trigger if data already wasn't frozen. Do this
384 * before checking for escaping, as the trigger may modify the magic
385 * offset. If a copy-out happens afterwards, it will have the correct
386 * data in the buffer.
389 jbd2_buffer_frozen_trigger(jh_in, mapped_data + new_offset,
395 if (*((__be32 *)(mapped_data + new_offset)) ==
396 cpu_to_be32(JBD2_MAGIC_NUMBER)) {
400 kunmap_atomic(mapped_data);
403 * Do we need to do a data copy?
405 if (need_copy_out && !done_copy_out) {
408 spin_unlock(&jh_in->b_state_lock);
409 tmp = jbd2_alloc(bh_in->b_size, GFP_NOFS);
414 spin_lock(&jh_in->b_state_lock);
415 if (jh_in->b_frozen_data) {
416 jbd2_free(tmp, bh_in->b_size);
420 jh_in->b_frozen_data = tmp;
421 mapped_data = kmap_atomic(new_page);
422 memcpy(tmp, mapped_data + new_offset, bh_in->b_size);
423 kunmap_atomic(mapped_data);
425 new_page = virt_to_page(tmp);
426 new_offset = offset_in_page(tmp);
430 * This isn't strictly necessary, as we're using frozen
431 * data for the escaping, but it keeps consistency with
432 * b_frozen_data usage.
434 jh_in->b_frozen_triggers = jh_in->b_triggers;
438 * Did we need to do an escaping? Now we've done all the
439 * copying, we can finally do so.
442 mapped_data = kmap_atomic(new_page);
443 *((unsigned int *)(mapped_data + new_offset)) = 0;
444 kunmap_atomic(mapped_data);
447 set_bh_page(new_bh, new_page, new_offset);
448 new_bh->b_size = bh_in->b_size;
449 new_bh->b_bdev = journal->j_dev;
450 new_bh->b_blocknr = blocknr;
451 new_bh->b_private = bh_in;
452 set_buffer_mapped(new_bh);
453 set_buffer_dirty(new_bh);
458 * The to-be-written buffer needs to get moved to the io queue,
459 * and the original buffer whose contents we are shadowing or
460 * copying is moved to the transaction's shadow queue.
462 JBUFFER_TRACE(jh_in, "file as BJ_Shadow");
463 spin_lock(&journal->j_list_lock);
464 __jbd2_journal_file_buffer(jh_in, transaction, BJ_Shadow);
465 spin_unlock(&journal->j_list_lock);
466 set_buffer_shadow(bh_in);
467 spin_unlock(&jh_in->b_state_lock);
469 return do_escape | (done_copy_out << 1);
473 * Allocation code for the journal file. Manage the space left in the
474 * journal, so that we can begin checkpointing when appropriate.
478 * Called with j_state_lock locked for writing.
479 * Returns true if a transaction commit was started.
481 static int __jbd2_log_start_commit(journal_t *journal, tid_t target)
483 /* Return if the txn has already requested to be committed */
484 if (journal->j_commit_request == target)
488 * The only transaction we can possibly wait upon is the
489 * currently running transaction (if it exists). Otherwise,
490 * the target tid must be an old one.
492 if (journal->j_running_transaction &&
493 journal->j_running_transaction->t_tid == target) {
495 * We want a new commit: OK, mark the request and wakeup the
496 * commit thread. We do _not_ do the commit ourselves.
499 journal->j_commit_request = target;
500 jbd2_debug(1, "JBD2: requesting commit %u/%u\n",
501 journal->j_commit_request,
502 journal->j_commit_sequence);
503 journal->j_running_transaction->t_requested = jiffies;
504 wake_up(&journal->j_wait_commit);
506 } else if (!tid_geq(journal->j_commit_request, target))
507 /* This should never happen, but if it does, preserve
508 the evidence before kjournald goes into a loop and
509 increments j_commit_sequence beyond all recognition. */
510 WARN_ONCE(1, "JBD2: bad log_start_commit: %u %u %u %u\n",
511 journal->j_commit_request,
512 journal->j_commit_sequence,
513 target, journal->j_running_transaction ?
514 journal->j_running_transaction->t_tid : 0);
518 int jbd2_log_start_commit(journal_t *journal, tid_t tid)
522 write_lock(&journal->j_state_lock);
523 ret = __jbd2_log_start_commit(journal, tid);
524 write_unlock(&journal->j_state_lock);
529 * Force and wait any uncommitted transactions. We can only force the running
530 * transaction if we don't have an active handle, otherwise, we will deadlock.
531 * Returns: <0 in case of error,
532 * 0 if nothing to commit,
533 * 1 if transaction was successfully committed.
535 static int __jbd2_journal_force_commit(journal_t *journal)
537 transaction_t *transaction = NULL;
539 int need_to_start = 0, ret = 0;
541 read_lock(&journal->j_state_lock);
542 if (journal->j_running_transaction && !current->journal_info) {
543 transaction = journal->j_running_transaction;
544 if (!tid_geq(journal->j_commit_request, transaction->t_tid))
546 } else if (journal->j_committing_transaction)
547 transaction = journal->j_committing_transaction;
550 /* Nothing to commit */
551 read_unlock(&journal->j_state_lock);
554 tid = transaction->t_tid;
555 read_unlock(&journal->j_state_lock);
557 jbd2_log_start_commit(journal, tid);
558 ret = jbd2_log_wait_commit(journal, tid);
566 * jbd2_journal_force_commit_nested - Force and wait upon a commit if the
567 * calling process is not within transaction.
569 * @journal: journal to force
570 * Returns true if progress was made.
572 * This is used for forcing out undo-protected data which contains
573 * bitmaps, when the fs is running out of space.
575 int jbd2_journal_force_commit_nested(journal_t *journal)
579 ret = __jbd2_journal_force_commit(journal);
584 * jbd2_journal_force_commit() - force any uncommitted transactions
585 * @journal: journal to force
587 * Caller want unconditional commit. We can only force the running transaction
588 * if we don't have an active handle, otherwise, we will deadlock.
590 int jbd2_journal_force_commit(journal_t *journal)
594 J_ASSERT(!current->journal_info);
595 ret = __jbd2_journal_force_commit(journal);
602 * Start a commit of the current running transaction (if any). Returns true
603 * if a transaction is going to be committed (or is currently already
604 * committing), and fills its tid in at *ptid
606 int jbd2_journal_start_commit(journal_t *journal, tid_t *ptid)
610 write_lock(&journal->j_state_lock);
611 if (journal->j_running_transaction) {
612 tid_t tid = journal->j_running_transaction->t_tid;
614 __jbd2_log_start_commit(journal, tid);
615 /* There's a running transaction and we've just made sure
616 * it's commit has been scheduled. */
620 } else if (journal->j_committing_transaction) {
622 * If commit has been started, then we have to wait for
623 * completion of that transaction.
626 *ptid = journal->j_committing_transaction->t_tid;
629 write_unlock(&journal->j_state_lock);
634 * Return 1 if a given transaction has not yet sent barrier request
635 * connected with a transaction commit. If 0 is returned, transaction
636 * may or may not have sent the barrier. Used to avoid sending barrier
637 * twice in common cases.
639 int jbd2_trans_will_send_data_barrier(journal_t *journal, tid_t tid)
642 transaction_t *commit_trans;
644 if (!(journal->j_flags & JBD2_BARRIER))
646 read_lock(&journal->j_state_lock);
647 /* Transaction already committed? */
648 if (tid_geq(journal->j_commit_sequence, tid))
650 commit_trans = journal->j_committing_transaction;
651 if (!commit_trans || commit_trans->t_tid != tid) {
656 * Transaction is being committed and we already proceeded to
657 * submitting a flush to fs partition?
659 if (journal->j_fs_dev != journal->j_dev) {
660 if (!commit_trans->t_need_data_flush ||
661 commit_trans->t_state >= T_COMMIT_DFLUSH)
664 if (commit_trans->t_state >= T_COMMIT_JFLUSH)
669 read_unlock(&journal->j_state_lock);
672 EXPORT_SYMBOL(jbd2_trans_will_send_data_barrier);
675 * Wait for a specified commit to complete.
676 * The caller may not hold the journal lock.
678 int jbd2_log_wait_commit(journal_t *journal, tid_t tid)
682 read_lock(&journal->j_state_lock);
683 #ifdef CONFIG_PROVE_LOCKING
685 * Some callers make sure transaction is already committing and in that
686 * case we cannot block on open handles anymore. So don't warn in that
689 if (tid_gt(tid, journal->j_commit_sequence) &&
690 (!journal->j_committing_transaction ||
691 journal->j_committing_transaction->t_tid != tid)) {
692 read_unlock(&journal->j_state_lock);
693 jbd2_might_wait_for_commit(journal);
694 read_lock(&journal->j_state_lock);
697 #ifdef CONFIG_JBD2_DEBUG
698 if (!tid_geq(journal->j_commit_request, tid)) {
700 "%s: error: j_commit_request=%u, tid=%u\n",
701 __func__, journal->j_commit_request, tid);
704 while (tid_gt(tid, journal->j_commit_sequence)) {
705 jbd2_debug(1, "JBD2: want %u, j_commit_sequence=%u\n",
706 tid, journal->j_commit_sequence);
707 read_unlock(&journal->j_state_lock);
708 wake_up(&journal->j_wait_commit);
709 wait_event(journal->j_wait_done_commit,
710 !tid_gt(tid, journal->j_commit_sequence));
711 read_lock(&journal->j_state_lock);
713 read_unlock(&journal->j_state_lock);
715 if (unlikely(is_journal_aborted(journal)))
721 * Start a fast commit. If there's an ongoing fast or full commit wait for
722 * it to complete. Returns 0 if a new fast commit was started. Returns -EALREADY
723 * if a fast commit is not needed, either because there's an already a commit
724 * going on or this tid has already been committed. Returns -EINVAL if no jbd2
725 * commit has yet been performed.
727 int jbd2_fc_begin_commit(journal_t *journal, tid_t tid)
729 if (unlikely(is_journal_aborted(journal)))
732 * Fast commits only allowed if at least one full commit has
735 if (!journal->j_stats.ts_tid)
738 write_lock(&journal->j_state_lock);
739 if (tid <= journal->j_commit_sequence) {
740 write_unlock(&journal->j_state_lock);
744 if (journal->j_flags & JBD2_FULL_COMMIT_ONGOING ||
745 (journal->j_flags & JBD2_FAST_COMMIT_ONGOING)) {
748 prepare_to_wait(&journal->j_fc_wait, &wait,
749 TASK_UNINTERRUPTIBLE);
750 write_unlock(&journal->j_state_lock);
752 finish_wait(&journal->j_fc_wait, &wait);
755 journal->j_flags |= JBD2_FAST_COMMIT_ONGOING;
756 write_unlock(&journal->j_state_lock);
757 jbd2_journal_lock_updates(journal);
761 EXPORT_SYMBOL(jbd2_fc_begin_commit);
764 * Stop a fast commit. If fallback is set, this function starts commit of
765 * TID tid before any other fast commit can start.
767 static int __jbd2_fc_end_commit(journal_t *journal, tid_t tid, bool fallback)
769 jbd2_journal_unlock_updates(journal);
770 if (journal->j_fc_cleanup_callback)
771 journal->j_fc_cleanup_callback(journal, 0, tid);
772 write_lock(&journal->j_state_lock);
773 journal->j_flags &= ~JBD2_FAST_COMMIT_ONGOING;
775 journal->j_flags |= JBD2_FULL_COMMIT_ONGOING;
776 write_unlock(&journal->j_state_lock);
777 wake_up(&journal->j_fc_wait);
779 return jbd2_complete_transaction(journal, tid);
783 int jbd2_fc_end_commit(journal_t *journal)
785 return __jbd2_fc_end_commit(journal, 0, false);
787 EXPORT_SYMBOL(jbd2_fc_end_commit);
789 int jbd2_fc_end_commit_fallback(journal_t *journal)
793 read_lock(&journal->j_state_lock);
794 tid = journal->j_running_transaction ?
795 journal->j_running_transaction->t_tid : 0;
796 read_unlock(&journal->j_state_lock);
797 return __jbd2_fc_end_commit(journal, tid, true);
799 EXPORT_SYMBOL(jbd2_fc_end_commit_fallback);
801 /* Return 1 when transaction with given tid has already committed. */
802 int jbd2_transaction_committed(journal_t *journal, tid_t tid)
806 read_lock(&journal->j_state_lock);
807 if (journal->j_running_transaction &&
808 journal->j_running_transaction->t_tid == tid)
810 if (journal->j_committing_transaction &&
811 journal->j_committing_transaction->t_tid == tid)
813 read_unlock(&journal->j_state_lock);
816 EXPORT_SYMBOL(jbd2_transaction_committed);
819 * When this function returns the transaction corresponding to tid
820 * will be completed. If the transaction has currently running, start
821 * committing that transaction before waiting for it to complete. If
822 * the transaction id is stale, it is by definition already completed,
823 * so just return SUCCESS.
825 int jbd2_complete_transaction(journal_t *journal, tid_t tid)
827 int need_to_wait = 1;
829 read_lock(&journal->j_state_lock);
830 if (journal->j_running_transaction &&
831 journal->j_running_transaction->t_tid == tid) {
832 if (journal->j_commit_request != tid) {
833 /* transaction not yet started, so request it */
834 read_unlock(&journal->j_state_lock);
835 jbd2_log_start_commit(journal, tid);
838 } else if (!(journal->j_committing_transaction &&
839 journal->j_committing_transaction->t_tid == tid))
841 read_unlock(&journal->j_state_lock);
845 return jbd2_log_wait_commit(journal, tid);
847 EXPORT_SYMBOL(jbd2_complete_transaction);
850 * Log buffer allocation routines:
853 int jbd2_journal_next_log_block(journal_t *journal, unsigned long long *retp)
855 unsigned long blocknr;
857 write_lock(&journal->j_state_lock);
858 J_ASSERT(journal->j_free > 1);
860 blocknr = journal->j_head;
863 if (journal->j_head == journal->j_last)
864 journal->j_head = journal->j_first;
865 write_unlock(&journal->j_state_lock);
866 return jbd2_journal_bmap(journal, blocknr, retp);
869 /* Map one fast commit buffer for use by the file system */
870 int jbd2_fc_get_buf(journal_t *journal, struct buffer_head **bh_out)
872 unsigned long long pblock;
873 unsigned long blocknr;
875 struct buffer_head *bh;
880 if (journal->j_fc_off + journal->j_fc_first < journal->j_fc_last) {
881 fc_off = journal->j_fc_off;
882 blocknr = journal->j_fc_first + fc_off;
891 ret = jbd2_journal_bmap(journal, blocknr, &pblock);
895 bh = __getblk(journal->j_dev, pblock, journal->j_blocksize);
900 journal->j_fc_wbuf[fc_off] = bh;
906 EXPORT_SYMBOL(jbd2_fc_get_buf);
909 * Wait on fast commit buffers that were allocated by jbd2_fc_get_buf
912 int jbd2_fc_wait_bufs(journal_t *journal, int num_blks)
914 struct buffer_head *bh;
917 j_fc_off = journal->j_fc_off;
920 * Wait in reverse order to minimize chances of us being woken up before
921 * all IOs have completed
923 for (i = j_fc_off - 1; i >= j_fc_off - num_blks; i--) {
924 bh = journal->j_fc_wbuf[i];
927 journal->j_fc_wbuf[i] = NULL;
928 if (unlikely(!buffer_uptodate(bh)))
934 EXPORT_SYMBOL(jbd2_fc_wait_bufs);
936 int jbd2_fc_release_bufs(journal_t *journal)
938 struct buffer_head *bh;
941 j_fc_off = journal->j_fc_off;
943 for (i = j_fc_off - 1; i >= 0; i--) {
944 bh = journal->j_fc_wbuf[i];
948 journal->j_fc_wbuf[i] = NULL;
953 EXPORT_SYMBOL(jbd2_fc_release_bufs);
956 * Conversion of logical to physical block numbers for the journal
958 * On external journals the journal blocks are identity-mapped, so
959 * this is a no-op. If needed, we can use j_blk_offset - everything is
962 int jbd2_journal_bmap(journal_t *journal, unsigned long blocknr,
963 unsigned long long *retp)
966 unsigned long long ret;
969 if (journal->j_inode) {
971 ret = bmap(journal->j_inode, &block);
974 printk(KERN_ALERT "%s: journal block not found "
975 "at offset %lu on %s\n",
976 __func__, blocknr, journal->j_devname);
978 jbd2_journal_abort(journal, err);
984 *retp = blocknr; /* +journal->j_blk_offset */
990 * We play buffer_head aliasing tricks to write data/metadata blocks to
991 * the journal without copying their contents, but for journal
992 * descriptor blocks we do need to generate bona fide buffers.
994 * After the caller of jbd2_journal_get_descriptor_buffer() has finished modifying
995 * the buffer's contents they really should run flush_dcache_page(bh->b_page).
996 * But we don't bother doing that, so there will be coherency problems with
997 * mmaps of blockdevs which hold live JBD-controlled filesystems.
1000 jbd2_journal_get_descriptor_buffer(transaction_t *transaction, int type)
1002 journal_t *journal = transaction->t_journal;
1003 struct buffer_head *bh;
1004 unsigned long long blocknr;
1005 journal_header_t *header;
1008 err = jbd2_journal_next_log_block(journal, &blocknr);
1013 bh = __getblk(journal->j_dev, blocknr, journal->j_blocksize);
1016 atomic_dec(&transaction->t_outstanding_credits);
1018 memset(bh->b_data, 0, journal->j_blocksize);
1019 header = (journal_header_t *)bh->b_data;
1020 header->h_magic = cpu_to_be32(JBD2_MAGIC_NUMBER);
1021 header->h_blocktype = cpu_to_be32(type);
1022 header->h_sequence = cpu_to_be32(transaction->t_tid);
1023 set_buffer_uptodate(bh);
1025 BUFFER_TRACE(bh, "return this buffer");
1029 void jbd2_descriptor_block_csum_set(journal_t *j, struct buffer_head *bh)
1031 struct jbd2_journal_block_tail *tail;
1034 if (!jbd2_journal_has_csum_v2or3(j))
1037 tail = (struct jbd2_journal_block_tail *)(bh->b_data + j->j_blocksize -
1038 sizeof(struct jbd2_journal_block_tail));
1039 tail->t_checksum = 0;
1040 csum = jbd2_chksum(j, j->j_csum_seed, bh->b_data, j->j_blocksize);
1041 tail->t_checksum = cpu_to_be32(csum);
1045 * Return tid of the oldest transaction in the journal and block in the journal
1046 * where the transaction starts.
1048 * If the journal is now empty, return which will be the next transaction ID
1049 * we will write and where will that transaction start.
1051 * The return value is 0 if journal tail cannot be pushed any further, 1 if
1054 int jbd2_journal_get_log_tail(journal_t *journal, tid_t *tid,
1055 unsigned long *block)
1057 transaction_t *transaction;
1060 read_lock(&journal->j_state_lock);
1061 spin_lock(&journal->j_list_lock);
1062 transaction = journal->j_checkpoint_transactions;
1064 *tid = transaction->t_tid;
1065 *block = transaction->t_log_start;
1066 } else if ((transaction = journal->j_committing_transaction) != NULL) {
1067 *tid = transaction->t_tid;
1068 *block = transaction->t_log_start;
1069 } else if ((transaction = journal->j_running_transaction) != NULL) {
1070 *tid = transaction->t_tid;
1071 *block = journal->j_head;
1073 *tid = journal->j_transaction_sequence;
1074 *block = journal->j_head;
1076 ret = tid_gt(*tid, journal->j_tail_sequence);
1077 spin_unlock(&journal->j_list_lock);
1078 read_unlock(&journal->j_state_lock);
1084 * Update information in journal structure and in on disk journal superblock
1085 * about log tail. This function does not check whether information passed in
1086 * really pushes log tail further. It's responsibility of the caller to make
1087 * sure provided log tail information is valid (e.g. by holding
1088 * j_checkpoint_mutex all the time between computing log tail and calling this
1089 * function as is the case with jbd2_cleanup_journal_tail()).
1091 * Requires j_checkpoint_mutex
1093 int __jbd2_update_log_tail(journal_t *journal, tid_t tid, unsigned long block)
1095 unsigned long freed;
1098 BUG_ON(!mutex_is_locked(&journal->j_checkpoint_mutex));
1101 * We cannot afford for write to remain in drive's caches since as
1102 * soon as we update j_tail, next transaction can start reusing journal
1103 * space and if we lose sb update during power failure we'd replay
1104 * old transaction with possibly newly overwritten data.
1106 ret = jbd2_journal_update_sb_log_tail(journal, tid, block,
1107 REQ_SYNC | REQ_FUA);
1111 write_lock(&journal->j_state_lock);
1112 freed = block - journal->j_tail;
1113 if (block < journal->j_tail)
1114 freed += journal->j_last - journal->j_first;
1116 trace_jbd2_update_log_tail(journal, tid, block, freed);
1118 "Cleaning journal tail from %u to %u (offset %lu), "
1120 journal->j_tail_sequence, tid, block, freed);
1122 journal->j_free += freed;
1123 journal->j_tail_sequence = tid;
1124 journal->j_tail = block;
1125 write_unlock(&journal->j_state_lock);
1132 * This is a variation of __jbd2_update_log_tail which checks for validity of
1133 * provided log tail and locks j_checkpoint_mutex. So it is safe against races
1134 * with other threads updating log tail.
1136 void jbd2_update_log_tail(journal_t *journal, tid_t tid, unsigned long block)
1138 mutex_lock_io(&journal->j_checkpoint_mutex);
1139 if (tid_gt(tid, journal->j_tail_sequence))
1140 __jbd2_update_log_tail(journal, tid, block);
1141 mutex_unlock(&journal->j_checkpoint_mutex);
1144 struct jbd2_stats_proc_session {
1146 struct transaction_stats_s *stats;
1151 static void *jbd2_seq_info_start(struct seq_file *seq, loff_t *pos)
1153 return *pos ? NULL : SEQ_START_TOKEN;
1156 static void *jbd2_seq_info_next(struct seq_file *seq, void *v, loff_t *pos)
1162 static int jbd2_seq_info_show(struct seq_file *seq, void *v)
1164 struct jbd2_stats_proc_session *s = seq->private;
1166 if (v != SEQ_START_TOKEN)
1168 seq_printf(seq, "%lu transactions (%lu requested), "
1169 "each up to %u blocks\n",
1170 s->stats->ts_tid, s->stats->ts_requested,
1171 s->journal->j_max_transaction_buffers);
1172 if (s->stats->ts_tid == 0)
1174 seq_printf(seq, "average: \n %ums waiting for transaction\n",
1175 jiffies_to_msecs(s->stats->run.rs_wait / s->stats->ts_tid));
1176 seq_printf(seq, " %ums request delay\n",
1177 (s->stats->ts_requested == 0) ? 0 :
1178 jiffies_to_msecs(s->stats->run.rs_request_delay /
1179 s->stats->ts_requested));
1180 seq_printf(seq, " %ums running transaction\n",
1181 jiffies_to_msecs(s->stats->run.rs_running / s->stats->ts_tid));
1182 seq_printf(seq, " %ums transaction was being locked\n",
1183 jiffies_to_msecs(s->stats->run.rs_locked / s->stats->ts_tid));
1184 seq_printf(seq, " %ums flushing data (in ordered mode)\n",
1185 jiffies_to_msecs(s->stats->run.rs_flushing / s->stats->ts_tid));
1186 seq_printf(seq, " %ums logging transaction\n",
1187 jiffies_to_msecs(s->stats->run.rs_logging / s->stats->ts_tid));
1188 seq_printf(seq, " %lluus average transaction commit time\n",
1189 div_u64(s->journal->j_average_commit_time, 1000));
1190 seq_printf(seq, " %lu handles per transaction\n",
1191 s->stats->run.rs_handle_count / s->stats->ts_tid);
1192 seq_printf(seq, " %lu blocks per transaction\n",
1193 s->stats->run.rs_blocks / s->stats->ts_tid);
1194 seq_printf(seq, " %lu logged blocks per transaction\n",
1195 s->stats->run.rs_blocks_logged / s->stats->ts_tid);
1199 static void jbd2_seq_info_stop(struct seq_file *seq, void *v)
1203 static const struct seq_operations jbd2_seq_info_ops = {
1204 .start = jbd2_seq_info_start,
1205 .next = jbd2_seq_info_next,
1206 .stop = jbd2_seq_info_stop,
1207 .show = jbd2_seq_info_show,
1210 static int jbd2_seq_info_open(struct inode *inode, struct file *file)
1212 journal_t *journal = pde_data(inode);
1213 struct jbd2_stats_proc_session *s;
1216 s = kmalloc(sizeof(*s), GFP_KERNEL);
1219 size = sizeof(struct transaction_stats_s);
1220 s->stats = kmalloc(size, GFP_KERNEL);
1221 if (s->stats == NULL) {
1225 spin_lock(&journal->j_history_lock);
1226 memcpy(s->stats, &journal->j_stats, size);
1227 s->journal = journal;
1228 spin_unlock(&journal->j_history_lock);
1230 rc = seq_open(file, &jbd2_seq_info_ops);
1232 struct seq_file *m = file->private_data;
1242 static int jbd2_seq_info_release(struct inode *inode, struct file *file)
1244 struct seq_file *seq = file->private_data;
1245 struct jbd2_stats_proc_session *s = seq->private;
1248 return seq_release(inode, file);
1251 static const struct proc_ops jbd2_info_proc_ops = {
1252 .proc_open = jbd2_seq_info_open,
1253 .proc_read = seq_read,
1254 .proc_lseek = seq_lseek,
1255 .proc_release = jbd2_seq_info_release,
1258 static struct proc_dir_entry *proc_jbd2_stats;
1260 static void jbd2_stats_proc_init(journal_t *journal)
1262 journal->j_proc_entry = proc_mkdir(journal->j_devname, proc_jbd2_stats);
1263 if (journal->j_proc_entry) {
1264 proc_create_data("info", S_IRUGO, journal->j_proc_entry,
1265 &jbd2_info_proc_ops, journal);
1269 static void jbd2_stats_proc_exit(journal_t *journal)
1271 remove_proc_entry("info", journal->j_proc_entry);
1272 remove_proc_entry(journal->j_devname, proc_jbd2_stats);
1275 /* Minimum size of descriptor tag */
1276 static int jbd2_min_tag_size(void)
1279 * Tag with 32-bit block numbers does not use last four bytes of the
1282 return sizeof(journal_block_tag_t) - 4;
1286 * jbd2_journal_shrink_scan()
1287 * @shrink: shrinker to work on
1288 * @sc: reclaim request to process
1290 * Scan the checkpointed buffer on the checkpoint list and release the
1293 static unsigned long jbd2_journal_shrink_scan(struct shrinker *shrink,
1294 struct shrink_control *sc)
1296 journal_t *journal = container_of(shrink, journal_t, j_shrinker);
1297 unsigned long nr_to_scan = sc->nr_to_scan;
1298 unsigned long nr_shrunk;
1299 unsigned long count;
1301 count = percpu_counter_read_positive(&journal->j_checkpoint_jh_count);
1302 trace_jbd2_shrink_scan_enter(journal, sc->nr_to_scan, count);
1304 nr_shrunk = jbd2_journal_shrink_checkpoint_list(journal, &nr_to_scan);
1306 count = percpu_counter_read_positive(&journal->j_checkpoint_jh_count);
1307 trace_jbd2_shrink_scan_exit(journal, nr_to_scan, nr_shrunk, count);
1313 * jbd2_journal_shrink_count()
1314 * @shrink: shrinker to work on
1315 * @sc: reclaim request to process
1317 * Count the number of checkpoint buffers on the checkpoint list.
1319 static unsigned long jbd2_journal_shrink_count(struct shrinker *shrink,
1320 struct shrink_control *sc)
1322 journal_t *journal = container_of(shrink, journal_t, j_shrinker);
1323 unsigned long count;
1325 count = percpu_counter_read_positive(&journal->j_checkpoint_jh_count);
1326 trace_jbd2_shrink_count(journal, sc->nr_to_scan, count);
1332 * Management for journal control blocks: functions to create and
1333 * destroy journal_t structures, and to initialise and read existing
1334 * journal blocks from disk. */
1336 /* First: create and setup a journal_t object in memory. We initialise
1337 * very few fields yet: that has to wait until we have created the
1338 * journal structures from from scratch, or loaded them from disk. */
1340 static journal_t *journal_init_common(struct block_device *bdev,
1341 struct block_device *fs_dev,
1342 unsigned long long start, int len, int blocksize)
1344 static struct lock_class_key jbd2_trans_commit_key;
1347 struct buffer_head *bh;
1350 journal = kzalloc(sizeof(*journal), GFP_KERNEL);
1354 init_waitqueue_head(&journal->j_wait_transaction_locked);
1355 init_waitqueue_head(&journal->j_wait_done_commit);
1356 init_waitqueue_head(&journal->j_wait_commit);
1357 init_waitqueue_head(&journal->j_wait_updates);
1358 init_waitqueue_head(&journal->j_wait_reserved);
1359 init_waitqueue_head(&journal->j_fc_wait);
1360 mutex_init(&journal->j_abort_mutex);
1361 mutex_init(&journal->j_barrier);
1362 mutex_init(&journal->j_checkpoint_mutex);
1363 spin_lock_init(&journal->j_revoke_lock);
1364 spin_lock_init(&journal->j_list_lock);
1365 rwlock_init(&journal->j_state_lock);
1367 journal->j_commit_interval = (HZ * JBD2_DEFAULT_MAX_COMMIT_AGE);
1368 journal->j_min_batch_time = 0;
1369 journal->j_max_batch_time = 15000; /* 15ms */
1370 atomic_set(&journal->j_reserved_credits, 0);
1372 /* The journal is marked for error until we succeed with recovery! */
1373 journal->j_flags = JBD2_ABORT;
1375 /* Set up a default-sized revoke table for the new mount. */
1376 err = jbd2_journal_init_revoke(journal, JOURNAL_REVOKE_DEFAULT_HASH);
1380 spin_lock_init(&journal->j_history_lock);
1382 lockdep_init_map(&journal->j_trans_commit_map, "jbd2_handle",
1383 &jbd2_trans_commit_key, 0);
1385 /* journal descriptor can store up to n blocks -bzzz */
1386 journal->j_blocksize = blocksize;
1387 journal->j_dev = bdev;
1388 journal->j_fs_dev = fs_dev;
1389 journal->j_blk_offset = start;
1390 journal->j_total_len = len;
1391 /* We need enough buffers to write out full descriptor block. */
1392 n = journal->j_blocksize / jbd2_min_tag_size();
1393 journal->j_wbufsize = n;
1394 journal->j_fc_wbuf = NULL;
1395 journal->j_wbuf = kmalloc_array(n, sizeof(struct buffer_head *),
1397 if (!journal->j_wbuf)
1400 bh = getblk_unmovable(journal->j_dev, start, journal->j_blocksize);
1402 pr_err("%s: Cannot get buffer for journal superblock\n",
1406 journal->j_sb_buffer = bh;
1407 journal->j_superblock = (journal_superblock_t *)bh->b_data;
1409 journal->j_shrink_transaction = NULL;
1410 journal->j_shrinker.scan_objects = jbd2_journal_shrink_scan;
1411 journal->j_shrinker.count_objects = jbd2_journal_shrink_count;
1412 journal->j_shrinker.seeks = DEFAULT_SEEKS;
1413 journal->j_shrinker.batch = journal->j_max_transaction_buffers;
1415 if (percpu_counter_init(&journal->j_checkpoint_jh_count, 0, GFP_KERNEL))
1418 if (register_shrinker(&journal->j_shrinker, "jbd2-journal:(%u:%u)",
1419 MAJOR(bdev->bd_dev), MINOR(bdev->bd_dev))) {
1420 percpu_counter_destroy(&journal->j_checkpoint_jh_count);
1426 brelse(journal->j_sb_buffer);
1427 kfree(journal->j_wbuf);
1428 jbd2_journal_destroy_revoke(journal);
1433 /* jbd2_journal_init_dev and jbd2_journal_init_inode:
1435 * Create a journal structure assigned some fixed set of disk blocks to
1436 * the journal. We don't actually touch those disk blocks yet, but we
1437 * need to set up all of the mapping information to tell the journaling
1438 * system where the journal blocks are.
1443 * journal_t * jbd2_journal_init_dev() - creates and initialises a journal structure
1444 * @bdev: Block device on which to create the journal
1445 * @fs_dev: Device which hold journalled filesystem for this journal.
1446 * @start: Block nr Start of journal.
1447 * @len: Length of the journal in blocks.
1448 * @blocksize: blocksize of journalling device
1450 * Returns: a newly created journal_t *
1452 * jbd2_journal_init_dev creates a journal which maps a fixed contiguous
1453 * range of blocks on an arbitrary block device.
1456 journal_t *jbd2_journal_init_dev(struct block_device *bdev,
1457 struct block_device *fs_dev,
1458 unsigned long long start, int len, int blocksize)
1462 journal = journal_init_common(bdev, fs_dev, start, len, blocksize);
1466 snprintf(journal->j_devname, sizeof(journal->j_devname),
1467 "%pg", journal->j_dev);
1468 strreplace(journal->j_devname, '/', '!');
1469 jbd2_stats_proc_init(journal);
1475 * journal_t * jbd2_journal_init_inode () - creates a journal which maps to a inode.
1476 * @inode: An inode to create the journal in
1478 * jbd2_journal_init_inode creates a journal which maps an on-disk inode as
1479 * the journal. The inode must exist already, must support bmap() and
1480 * must have all data blocks preallocated.
1482 journal_t *jbd2_journal_init_inode(struct inode *inode)
1490 err = bmap(inode, &blocknr);
1492 if (err || !blocknr) {
1493 pr_err("%s: Cannot locate journal superblock\n",
1498 jbd2_debug(1, "JBD2: inode %s/%ld, size %lld, bits %d, blksize %ld\n",
1499 inode->i_sb->s_id, inode->i_ino, (long long) inode->i_size,
1500 inode->i_sb->s_blocksize_bits, inode->i_sb->s_blocksize);
1502 journal = journal_init_common(inode->i_sb->s_bdev, inode->i_sb->s_bdev,
1503 blocknr, inode->i_size >> inode->i_sb->s_blocksize_bits,
1504 inode->i_sb->s_blocksize);
1508 journal->j_inode = inode;
1509 snprintf(journal->j_devname, sizeof(journal->j_devname),
1510 "%pg", journal->j_dev);
1511 p = strreplace(journal->j_devname, '/', '!');
1512 sprintf(p, "-%lu", journal->j_inode->i_ino);
1513 jbd2_stats_proc_init(journal);
1519 * If the journal init or create aborts, we need to mark the journal
1520 * superblock as being NULL to prevent the journal destroy from writing
1521 * back a bogus superblock.
1523 static void journal_fail_superblock(journal_t *journal)
1525 struct buffer_head *bh = journal->j_sb_buffer;
1527 journal->j_sb_buffer = NULL;
1531 * Given a journal_t structure, initialise the various fields for
1532 * startup of a new journaling session. We use this both when creating
1533 * a journal, and after recovering an old journal to reset it for
1537 static int journal_reset(journal_t *journal)
1539 journal_superblock_t *sb = journal->j_superblock;
1540 unsigned long long first, last;
1542 first = be32_to_cpu(sb->s_first);
1543 last = be32_to_cpu(sb->s_maxlen);
1544 if (first + JBD2_MIN_JOURNAL_BLOCKS > last + 1) {
1545 printk(KERN_ERR "JBD2: Journal too short (blocks %llu-%llu).\n",
1547 journal_fail_superblock(journal);
1551 journal->j_first = first;
1552 journal->j_last = last;
1554 journal->j_head = journal->j_first;
1555 journal->j_tail = journal->j_first;
1556 journal->j_free = journal->j_last - journal->j_first;
1558 journal->j_tail_sequence = journal->j_transaction_sequence;
1559 journal->j_commit_sequence = journal->j_transaction_sequence - 1;
1560 journal->j_commit_request = journal->j_commit_sequence;
1562 journal->j_max_transaction_buffers = jbd2_journal_get_max_txn_bufs(journal);
1565 * Now that journal recovery is done, turn fast commits off here. This
1566 * way, if fast commit was enabled before the crash but if now FS has
1567 * disabled it, we don't enable fast commits.
1569 jbd2_clear_feature_fast_commit(journal);
1572 * As a special case, if the on-disk copy is already marked as needing
1573 * no recovery (s_start == 0), then we can safely defer the superblock
1574 * update until the next commit by setting JBD2_FLUSHED. This avoids
1575 * attempting a write to a potential-readonly device.
1577 if (sb->s_start == 0) {
1578 jbd2_debug(1, "JBD2: Skipping superblock update on recovered sb "
1579 "(start %ld, seq %u, errno %d)\n",
1580 journal->j_tail, journal->j_tail_sequence,
1582 journal->j_flags |= JBD2_FLUSHED;
1584 /* Lock here to make assertions happy... */
1585 mutex_lock_io(&journal->j_checkpoint_mutex);
1587 * Update log tail information. We use REQ_FUA since new
1588 * transaction will start reusing journal space and so we
1589 * must make sure information about current log tail is on
1592 jbd2_journal_update_sb_log_tail(journal,
1593 journal->j_tail_sequence,
1595 REQ_SYNC | REQ_FUA);
1596 mutex_unlock(&journal->j_checkpoint_mutex);
1598 return jbd2_journal_start_thread(journal);
1602 * This function expects that the caller will have locked the journal
1603 * buffer head, and will return with it unlocked
1605 static int jbd2_write_superblock(journal_t *journal, blk_opf_t write_flags)
1607 struct buffer_head *bh = journal->j_sb_buffer;
1608 journal_superblock_t *sb = journal->j_superblock;
1611 /* Buffer got discarded which means block device got invalidated */
1612 if (!buffer_mapped(bh)) {
1617 trace_jbd2_write_superblock(journal, write_flags);
1618 if (!(journal->j_flags & JBD2_BARRIER))
1619 write_flags &= ~(REQ_FUA | REQ_PREFLUSH);
1620 if (buffer_write_io_error(bh)) {
1622 * Oh, dear. A previous attempt to write the journal
1623 * superblock failed. This could happen because the
1624 * USB device was yanked out. Or it could happen to
1625 * be a transient write error and maybe the block will
1626 * be remapped. Nothing we can do but to retry the
1627 * write and hope for the best.
1629 printk(KERN_ERR "JBD2: previous I/O error detected "
1630 "for journal superblock update for %s.\n",
1631 journal->j_devname);
1632 clear_buffer_write_io_error(bh);
1633 set_buffer_uptodate(bh);
1635 if (jbd2_journal_has_csum_v2or3(journal))
1636 sb->s_checksum = jbd2_superblock_csum(journal, sb);
1638 bh->b_end_io = end_buffer_write_sync;
1639 ret = submit_bh(REQ_OP_WRITE | write_flags, bh);
1641 if (buffer_write_io_error(bh)) {
1642 clear_buffer_write_io_error(bh);
1643 set_buffer_uptodate(bh);
1647 printk(KERN_ERR "JBD2: Error %d detected when updating "
1648 "journal superblock for %s.\n", ret,
1649 journal->j_devname);
1650 if (!is_journal_aborted(journal))
1651 jbd2_journal_abort(journal, ret);
1658 * jbd2_journal_update_sb_log_tail() - Update log tail in journal sb on disk.
1659 * @journal: The journal to update.
1660 * @tail_tid: TID of the new transaction at the tail of the log
1661 * @tail_block: The first block of the transaction at the tail of the log
1662 * @write_flags: Flags for the journal sb write operation
1664 * Update a journal's superblock information about log tail and write it to
1665 * disk, waiting for the IO to complete.
1667 int jbd2_journal_update_sb_log_tail(journal_t *journal, tid_t tail_tid,
1668 unsigned long tail_block,
1669 blk_opf_t write_flags)
1671 journal_superblock_t *sb = journal->j_superblock;
1674 if (is_journal_aborted(journal))
1676 if (test_bit(JBD2_CHECKPOINT_IO_ERROR, &journal->j_atomic_flags)) {
1677 jbd2_journal_abort(journal, -EIO);
1681 BUG_ON(!mutex_is_locked(&journal->j_checkpoint_mutex));
1682 jbd2_debug(1, "JBD2: updating superblock (start %lu, seq %u)\n",
1683 tail_block, tail_tid);
1685 lock_buffer(journal->j_sb_buffer);
1686 sb->s_sequence = cpu_to_be32(tail_tid);
1687 sb->s_start = cpu_to_be32(tail_block);
1689 ret = jbd2_write_superblock(journal, write_flags);
1693 /* Log is no longer empty */
1694 write_lock(&journal->j_state_lock);
1695 WARN_ON(!sb->s_sequence);
1696 journal->j_flags &= ~JBD2_FLUSHED;
1697 write_unlock(&journal->j_state_lock);
1704 * jbd2_mark_journal_empty() - Mark on disk journal as empty.
1705 * @journal: The journal to update.
1706 * @write_flags: Flags for the journal sb write operation
1708 * Update a journal's dynamic superblock fields to show that journal is empty.
1709 * Write updated superblock to disk waiting for IO to complete.
1711 static void jbd2_mark_journal_empty(journal_t *journal, blk_opf_t write_flags)
1713 journal_superblock_t *sb = journal->j_superblock;
1714 bool had_fast_commit = false;
1716 BUG_ON(!mutex_is_locked(&journal->j_checkpoint_mutex));
1717 lock_buffer(journal->j_sb_buffer);
1718 if (sb->s_start == 0) { /* Is it already empty? */
1719 unlock_buffer(journal->j_sb_buffer);
1723 jbd2_debug(1, "JBD2: Marking journal as empty (seq %u)\n",
1724 journal->j_tail_sequence);
1726 sb->s_sequence = cpu_to_be32(journal->j_tail_sequence);
1727 sb->s_start = cpu_to_be32(0);
1728 if (jbd2_has_feature_fast_commit(journal)) {
1730 * When journal is clean, no need to commit fast commit flag and
1731 * make file system incompatible with older kernels.
1733 jbd2_clear_feature_fast_commit(journal);
1734 had_fast_commit = true;
1737 jbd2_write_superblock(journal, write_flags);
1739 if (had_fast_commit)
1740 jbd2_set_feature_fast_commit(journal);
1742 /* Log is no longer empty */
1743 write_lock(&journal->j_state_lock);
1744 journal->j_flags |= JBD2_FLUSHED;
1745 write_unlock(&journal->j_state_lock);
1749 * __jbd2_journal_erase() - Discard or zeroout journal blocks (excluding superblock)
1750 * @journal: The journal to erase.
1751 * @flags: A discard/zeroout request is sent for each physically contigous
1752 * region of the journal. Either JBD2_JOURNAL_FLUSH_DISCARD or
1753 * JBD2_JOURNAL_FLUSH_ZEROOUT must be set to determine which operation
1756 * Note: JBD2_JOURNAL_FLUSH_ZEROOUT attempts to use hardware offload. Zeroes
1757 * will be explicitly written if no hardware offload is available, see
1758 * blkdev_issue_zeroout for more details.
1760 static int __jbd2_journal_erase(journal_t *journal, unsigned int flags)
1763 unsigned long block, log_offset; /* logical */
1764 unsigned long long phys_block, block_start, block_stop; /* physical */
1765 loff_t byte_start, byte_stop, byte_count;
1767 /* flags must be set to either discard or zeroout */
1768 if ((flags & ~JBD2_JOURNAL_FLUSH_VALID) || !flags ||
1769 ((flags & JBD2_JOURNAL_FLUSH_DISCARD) &&
1770 (flags & JBD2_JOURNAL_FLUSH_ZEROOUT)))
1773 if ((flags & JBD2_JOURNAL_FLUSH_DISCARD) &&
1774 !bdev_max_discard_sectors(journal->j_dev))
1778 * lookup block mapping and issue discard/zeroout for each
1781 log_offset = be32_to_cpu(journal->j_superblock->s_first);
1782 block_start = ~0ULL;
1783 for (block = log_offset; block < journal->j_total_len; block++) {
1784 err = jbd2_journal_bmap(journal, block, &phys_block);
1786 pr_err("JBD2: bad block at offset %lu", block);
1790 if (block_start == ~0ULL) {
1791 block_start = phys_block;
1792 block_stop = block_start - 1;
1796 * last block not contiguous with current block,
1797 * process last contiguous region and return to this block on
1800 if (phys_block != block_stop + 1) {
1805 * if this isn't the last block of journal,
1806 * no need to process now because next block may also
1807 * be part of this contiguous region
1809 if (block != journal->j_total_len - 1)
1814 * end of contiguous region or this is last block of journal,
1815 * take care of the region
1817 byte_start = block_start * journal->j_blocksize;
1818 byte_stop = block_stop * journal->j_blocksize;
1819 byte_count = (block_stop - block_start + 1) *
1820 journal->j_blocksize;
1822 truncate_inode_pages_range(journal->j_dev->bd_inode->i_mapping,
1823 byte_start, byte_stop);
1825 if (flags & JBD2_JOURNAL_FLUSH_DISCARD) {
1826 err = blkdev_issue_discard(journal->j_dev,
1827 byte_start >> SECTOR_SHIFT,
1828 byte_count >> SECTOR_SHIFT,
1830 } else if (flags & JBD2_JOURNAL_FLUSH_ZEROOUT) {
1831 err = blkdev_issue_zeroout(journal->j_dev,
1832 byte_start >> SECTOR_SHIFT,
1833 byte_count >> SECTOR_SHIFT,
1837 if (unlikely(err != 0)) {
1838 pr_err("JBD2: (error %d) unable to wipe journal at physical blocks %llu - %llu",
1839 err, block_start, block_stop);
1843 /* reset start and stop after processing a region */
1844 block_start = ~0ULL;
1847 return blkdev_issue_flush(journal->j_dev);
1851 * jbd2_journal_update_sb_errno() - Update error in the journal.
1852 * @journal: The journal to update.
1854 * Update a journal's errno. Write updated superblock to disk waiting for IO
1857 void jbd2_journal_update_sb_errno(journal_t *journal)
1859 journal_superblock_t *sb = journal->j_superblock;
1862 lock_buffer(journal->j_sb_buffer);
1863 errcode = journal->j_errno;
1864 if (errcode == -ESHUTDOWN)
1866 jbd2_debug(1, "JBD2: updating superblock error (errno %d)\n", errcode);
1867 sb->s_errno = cpu_to_be32(errcode);
1869 jbd2_write_superblock(journal, REQ_SYNC | REQ_FUA);
1871 EXPORT_SYMBOL(jbd2_journal_update_sb_errno);
1873 static int journal_revoke_records_per_block(journal_t *journal)
1876 int space = journal->j_blocksize - sizeof(jbd2_journal_revoke_header_t);
1878 if (jbd2_has_feature_64bit(journal))
1883 if (jbd2_journal_has_csum_v2or3(journal))
1884 space -= sizeof(struct jbd2_journal_block_tail);
1885 return space / record_size;
1889 * Read the superblock for a given journal, performing initial
1890 * validation of the format.
1892 static int journal_get_superblock(journal_t *journal)
1894 struct buffer_head *bh;
1895 journal_superblock_t *sb;
1898 bh = journal->j_sb_buffer;
1900 J_ASSERT(bh != NULL);
1901 if (!buffer_uptodate(bh)) {
1902 ll_rw_block(REQ_OP_READ, 1, &bh);
1904 if (!buffer_uptodate(bh)) {
1906 "JBD2: IO error reading journal superblock\n");
1911 if (buffer_verified(bh))
1914 sb = journal->j_superblock;
1918 if (sb->s_header.h_magic != cpu_to_be32(JBD2_MAGIC_NUMBER) ||
1919 sb->s_blocksize != cpu_to_be32(journal->j_blocksize)) {
1920 printk(KERN_WARNING "JBD2: no valid journal superblock found\n");
1924 switch(be32_to_cpu(sb->s_header.h_blocktype)) {
1925 case JBD2_SUPERBLOCK_V1:
1926 journal->j_format_version = 1;
1928 case JBD2_SUPERBLOCK_V2:
1929 journal->j_format_version = 2;
1932 printk(KERN_WARNING "JBD2: unrecognised superblock format ID\n");
1936 if (be32_to_cpu(sb->s_maxlen) < journal->j_total_len)
1937 journal->j_total_len = be32_to_cpu(sb->s_maxlen);
1938 else if (be32_to_cpu(sb->s_maxlen) > journal->j_total_len) {
1939 printk(KERN_WARNING "JBD2: journal file too short\n");
1943 if (be32_to_cpu(sb->s_first) == 0 ||
1944 be32_to_cpu(sb->s_first) >= journal->j_total_len) {
1946 "JBD2: Invalid start block of journal: %u\n",
1947 be32_to_cpu(sb->s_first));
1951 if (jbd2_has_feature_csum2(journal) &&
1952 jbd2_has_feature_csum3(journal)) {
1953 /* Can't have checksum v2 and v3 at the same time! */
1954 printk(KERN_ERR "JBD2: Can't enable checksumming v2 and v3 "
1955 "at the same time!\n");
1959 if (jbd2_journal_has_csum_v2or3_feature(journal) &&
1960 jbd2_has_feature_checksum(journal)) {
1961 /* Can't have checksum v1 and v2 on at the same time! */
1962 printk(KERN_ERR "JBD2: Can't enable checksumming v1 and v2/3 "
1963 "at the same time!\n");
1967 if (!jbd2_verify_csum_type(journal, sb)) {
1968 printk(KERN_ERR "JBD2: Unknown checksum type\n");
1972 /* Load the checksum driver */
1973 if (jbd2_journal_has_csum_v2or3_feature(journal)) {
1974 journal->j_chksum_driver = crypto_alloc_shash("crc32c", 0, 0);
1975 if (IS_ERR(journal->j_chksum_driver)) {
1976 printk(KERN_ERR "JBD2: Cannot load crc32c driver.\n");
1977 err = PTR_ERR(journal->j_chksum_driver);
1978 journal->j_chksum_driver = NULL;
1983 if (jbd2_journal_has_csum_v2or3(journal)) {
1984 /* Check superblock checksum */
1985 if (sb->s_checksum != jbd2_superblock_csum(journal, sb)) {
1986 printk(KERN_ERR "JBD2: journal checksum error\n");
1991 /* Precompute checksum seed for all metadata */
1992 journal->j_csum_seed = jbd2_chksum(journal, ~0, sb->s_uuid,
1993 sizeof(sb->s_uuid));
1996 journal->j_revoke_records_per_block =
1997 journal_revoke_records_per_block(journal);
1998 set_buffer_verified(bh);
2003 journal_fail_superblock(journal);
2008 * Load the on-disk journal superblock and read the key fields into the
2012 static int load_superblock(journal_t *journal)
2015 journal_superblock_t *sb;
2018 err = journal_get_superblock(journal);
2022 sb = journal->j_superblock;
2024 journal->j_tail_sequence = be32_to_cpu(sb->s_sequence);
2025 journal->j_tail = be32_to_cpu(sb->s_start);
2026 journal->j_first = be32_to_cpu(sb->s_first);
2027 journal->j_errno = be32_to_cpu(sb->s_errno);
2028 journal->j_last = be32_to_cpu(sb->s_maxlen);
2030 if (jbd2_has_feature_fast_commit(journal)) {
2031 journal->j_fc_last = be32_to_cpu(sb->s_maxlen);
2032 num_fc_blocks = jbd2_journal_get_num_fc_blks(sb);
2033 if (journal->j_last - num_fc_blocks >= JBD2_MIN_JOURNAL_BLOCKS)
2034 journal->j_last = journal->j_fc_last - num_fc_blocks;
2035 journal->j_fc_first = journal->j_last + 1;
2036 journal->j_fc_off = 0;
2044 * jbd2_journal_load() - Read journal from disk.
2045 * @journal: Journal to act on.
2047 * Given a journal_t structure which tells us which disk blocks contain
2048 * a journal, read the journal from disk to initialise the in-memory
2051 int jbd2_journal_load(journal_t *journal)
2054 journal_superblock_t *sb;
2056 err = load_superblock(journal);
2060 sb = journal->j_superblock;
2061 /* If this is a V2 superblock, then we have to check the
2062 * features flags on it. */
2064 if (journal->j_format_version >= 2) {
2065 if ((sb->s_feature_ro_compat &
2066 ~cpu_to_be32(JBD2_KNOWN_ROCOMPAT_FEATURES)) ||
2067 (sb->s_feature_incompat &
2068 ~cpu_to_be32(JBD2_KNOWN_INCOMPAT_FEATURES))) {
2070 "JBD2: Unrecognised features on journal\n");
2076 * Create a slab for this blocksize
2078 err = jbd2_journal_create_slab(be32_to_cpu(sb->s_blocksize));
2082 /* Let the recovery code check whether it needs to recover any
2083 * data from the journal. */
2084 if (jbd2_journal_recover(journal))
2085 goto recovery_error;
2087 if (journal->j_failed_commit) {
2088 printk(KERN_ERR "JBD2: journal transaction %u on %s "
2089 "is corrupt.\n", journal->j_failed_commit,
2090 journal->j_devname);
2091 return -EFSCORRUPTED;
2094 * clear JBD2_ABORT flag initialized in journal_init_common
2095 * here to update log tail information with the newest seq.
2097 journal->j_flags &= ~JBD2_ABORT;
2099 /* OK, we've finished with the dynamic journal bits:
2100 * reinitialise the dynamic contents of the superblock in memory
2101 * and reset them on disk. */
2102 if (journal_reset(journal))
2103 goto recovery_error;
2105 journal->j_flags |= JBD2_LOADED;
2109 printk(KERN_WARNING "JBD2: recovery failed\n");
2114 * jbd2_journal_destroy() - Release a journal_t structure.
2115 * @journal: Journal to act on.
2117 * Release a journal_t structure once it is no longer in use by the
2119 * Return <0 if we couldn't clean up the journal.
2121 int jbd2_journal_destroy(journal_t *journal)
2125 /* Wait for the commit thread to wake up and die. */
2126 journal_kill_thread(journal);
2128 /* Force a final log commit */
2129 if (journal->j_running_transaction)
2130 jbd2_journal_commit_transaction(journal);
2132 /* Force any old transactions to disk */
2134 /* Totally anal locking here... */
2135 spin_lock(&journal->j_list_lock);
2136 while (journal->j_checkpoint_transactions != NULL) {
2137 spin_unlock(&journal->j_list_lock);
2138 mutex_lock_io(&journal->j_checkpoint_mutex);
2139 err = jbd2_log_do_checkpoint(journal);
2140 mutex_unlock(&journal->j_checkpoint_mutex);
2142 * If checkpointing failed, just free the buffers to avoid
2146 jbd2_journal_destroy_checkpoint(journal);
2147 spin_lock(&journal->j_list_lock);
2150 spin_lock(&journal->j_list_lock);
2153 J_ASSERT(journal->j_running_transaction == NULL);
2154 J_ASSERT(journal->j_committing_transaction == NULL);
2155 J_ASSERT(journal->j_checkpoint_transactions == NULL);
2156 spin_unlock(&journal->j_list_lock);
2159 * OK, all checkpoint transactions have been checked, now check the
2160 * write out io error flag and abort the journal if some buffer failed
2161 * to write back to the original location, otherwise the filesystem
2162 * may become inconsistent.
2164 if (!is_journal_aborted(journal) &&
2165 test_bit(JBD2_CHECKPOINT_IO_ERROR, &journal->j_atomic_flags))
2166 jbd2_journal_abort(journal, -EIO);
2168 if (journal->j_sb_buffer) {
2169 if (!is_journal_aborted(journal)) {
2170 mutex_lock_io(&journal->j_checkpoint_mutex);
2172 write_lock(&journal->j_state_lock);
2173 journal->j_tail_sequence =
2174 ++journal->j_transaction_sequence;
2175 write_unlock(&journal->j_state_lock);
2177 jbd2_mark_journal_empty(journal,
2178 REQ_SYNC | REQ_PREFLUSH | REQ_FUA);
2179 mutex_unlock(&journal->j_checkpoint_mutex);
2182 brelse(journal->j_sb_buffer);
2185 if (journal->j_shrinker.flags & SHRINKER_REGISTERED) {
2186 percpu_counter_destroy(&journal->j_checkpoint_jh_count);
2187 unregister_shrinker(&journal->j_shrinker);
2189 if (journal->j_proc_entry)
2190 jbd2_stats_proc_exit(journal);
2191 iput(journal->j_inode);
2192 if (journal->j_revoke)
2193 jbd2_journal_destroy_revoke(journal);
2194 if (journal->j_chksum_driver)
2195 crypto_free_shash(journal->j_chksum_driver);
2196 kfree(journal->j_fc_wbuf);
2197 kfree(journal->j_wbuf);
2205 * jbd2_journal_check_used_features() - Check if features specified are used.
2206 * @journal: Journal to check.
2207 * @compat: bitmask of compatible features
2208 * @ro: bitmask of features that force read-only mount
2209 * @incompat: bitmask of incompatible features
2211 * Check whether the journal uses all of a given set of
2212 * features. Return true (non-zero) if it does.
2215 int jbd2_journal_check_used_features(journal_t *journal, unsigned long compat,
2216 unsigned long ro, unsigned long incompat)
2218 journal_superblock_t *sb;
2220 if (!compat && !ro && !incompat)
2222 /* Load journal superblock if it is not loaded yet. */
2223 if (journal->j_format_version == 0 &&
2224 journal_get_superblock(journal) != 0)
2226 if (journal->j_format_version == 1)
2229 sb = journal->j_superblock;
2231 if (((be32_to_cpu(sb->s_feature_compat) & compat) == compat) &&
2232 ((be32_to_cpu(sb->s_feature_ro_compat) & ro) == ro) &&
2233 ((be32_to_cpu(sb->s_feature_incompat) & incompat) == incompat))
2240 * jbd2_journal_check_available_features() - Check feature set in journalling layer
2241 * @journal: Journal to check.
2242 * @compat: bitmask of compatible features
2243 * @ro: bitmask of features that force read-only mount
2244 * @incompat: bitmask of incompatible features
2246 * Check whether the journaling code supports the use of
2247 * all of a given set of features on this journal. Return true
2248 * (non-zero) if it can. */
2250 int jbd2_journal_check_available_features(journal_t *journal, unsigned long compat,
2251 unsigned long ro, unsigned long incompat)
2253 if (!compat && !ro && !incompat)
2256 /* We can support any known requested features iff the
2257 * superblock is in version 2. Otherwise we fail to support any
2258 * extended sb features. */
2260 if (journal->j_format_version != 2)
2263 if ((compat & JBD2_KNOWN_COMPAT_FEATURES) == compat &&
2264 (ro & JBD2_KNOWN_ROCOMPAT_FEATURES) == ro &&
2265 (incompat & JBD2_KNOWN_INCOMPAT_FEATURES) == incompat)
2272 jbd2_journal_initialize_fast_commit(journal_t *journal)
2274 journal_superblock_t *sb = journal->j_superblock;
2275 unsigned long long num_fc_blks;
2277 num_fc_blks = jbd2_journal_get_num_fc_blks(sb);
2278 if (journal->j_last - num_fc_blks < JBD2_MIN_JOURNAL_BLOCKS)
2281 /* Are we called twice? */
2282 WARN_ON(journal->j_fc_wbuf != NULL);
2283 journal->j_fc_wbuf = kmalloc_array(num_fc_blks,
2284 sizeof(struct buffer_head *), GFP_KERNEL);
2285 if (!journal->j_fc_wbuf)
2288 journal->j_fc_wbufsize = num_fc_blks;
2289 journal->j_fc_last = journal->j_last;
2290 journal->j_last = journal->j_fc_last - num_fc_blks;
2291 journal->j_fc_first = journal->j_last + 1;
2292 journal->j_fc_off = 0;
2293 journal->j_free = journal->j_last - journal->j_first;
2294 journal->j_max_transaction_buffers =
2295 jbd2_journal_get_max_txn_bufs(journal);
2301 * jbd2_journal_set_features() - Mark a given journal feature in the superblock
2302 * @journal: Journal to act on.
2303 * @compat: bitmask of compatible features
2304 * @ro: bitmask of features that force read-only mount
2305 * @incompat: bitmask of incompatible features
2307 * Mark a given journal feature as present on the
2308 * superblock. Returns true if the requested features could be set.
2312 int jbd2_journal_set_features(journal_t *journal, unsigned long compat,
2313 unsigned long ro, unsigned long incompat)
2315 #define INCOMPAT_FEATURE_ON(f) \
2316 ((incompat & (f)) && !(sb->s_feature_incompat & cpu_to_be32(f)))
2317 #define COMPAT_FEATURE_ON(f) \
2318 ((compat & (f)) && !(sb->s_feature_compat & cpu_to_be32(f)))
2319 journal_superblock_t *sb;
2321 if (jbd2_journal_check_used_features(journal, compat, ro, incompat))
2324 if (!jbd2_journal_check_available_features(journal, compat, ro, incompat))
2327 /* If enabling v2 checksums, turn on v3 instead */
2328 if (incompat & JBD2_FEATURE_INCOMPAT_CSUM_V2) {
2329 incompat &= ~JBD2_FEATURE_INCOMPAT_CSUM_V2;
2330 incompat |= JBD2_FEATURE_INCOMPAT_CSUM_V3;
2333 /* Asking for checksumming v3 and v1? Only give them v3. */
2334 if (incompat & JBD2_FEATURE_INCOMPAT_CSUM_V3 &&
2335 compat & JBD2_FEATURE_COMPAT_CHECKSUM)
2336 compat &= ~JBD2_FEATURE_COMPAT_CHECKSUM;
2338 jbd2_debug(1, "Setting new features 0x%lx/0x%lx/0x%lx\n",
2339 compat, ro, incompat);
2341 sb = journal->j_superblock;
2343 if (incompat & JBD2_FEATURE_INCOMPAT_FAST_COMMIT) {
2344 if (jbd2_journal_initialize_fast_commit(journal)) {
2345 pr_err("JBD2: Cannot enable fast commits.\n");
2350 /* Load the checksum driver if necessary */
2351 if ((journal->j_chksum_driver == NULL) &&
2352 INCOMPAT_FEATURE_ON(JBD2_FEATURE_INCOMPAT_CSUM_V3)) {
2353 journal->j_chksum_driver = crypto_alloc_shash("crc32c", 0, 0);
2354 if (IS_ERR(journal->j_chksum_driver)) {
2355 printk(KERN_ERR "JBD2: Cannot load crc32c driver.\n");
2356 journal->j_chksum_driver = NULL;
2359 /* Precompute checksum seed for all metadata */
2360 journal->j_csum_seed = jbd2_chksum(journal, ~0, sb->s_uuid,
2361 sizeof(sb->s_uuid));
2364 lock_buffer(journal->j_sb_buffer);
2366 /* If enabling v3 checksums, update superblock */
2367 if (INCOMPAT_FEATURE_ON(JBD2_FEATURE_INCOMPAT_CSUM_V3)) {
2368 sb->s_checksum_type = JBD2_CRC32C_CHKSUM;
2369 sb->s_feature_compat &=
2370 ~cpu_to_be32(JBD2_FEATURE_COMPAT_CHECKSUM);
2373 /* If enabling v1 checksums, downgrade superblock */
2374 if (COMPAT_FEATURE_ON(JBD2_FEATURE_COMPAT_CHECKSUM))
2375 sb->s_feature_incompat &=
2376 ~cpu_to_be32(JBD2_FEATURE_INCOMPAT_CSUM_V2 |
2377 JBD2_FEATURE_INCOMPAT_CSUM_V3);
2379 sb->s_feature_compat |= cpu_to_be32(compat);
2380 sb->s_feature_ro_compat |= cpu_to_be32(ro);
2381 sb->s_feature_incompat |= cpu_to_be32(incompat);
2382 unlock_buffer(journal->j_sb_buffer);
2383 journal->j_revoke_records_per_block =
2384 journal_revoke_records_per_block(journal);
2387 #undef COMPAT_FEATURE_ON
2388 #undef INCOMPAT_FEATURE_ON
2392 * jbd2_journal_clear_features() - Clear a given journal feature in the
2394 * @journal: Journal to act on.
2395 * @compat: bitmask of compatible features
2396 * @ro: bitmask of features that force read-only mount
2397 * @incompat: bitmask of incompatible features
2399 * Clear a given journal feature as present on the
2402 void jbd2_journal_clear_features(journal_t *journal, unsigned long compat,
2403 unsigned long ro, unsigned long incompat)
2405 journal_superblock_t *sb;
2407 jbd2_debug(1, "Clear features 0x%lx/0x%lx/0x%lx\n",
2408 compat, ro, incompat);
2410 sb = journal->j_superblock;
2412 sb->s_feature_compat &= ~cpu_to_be32(compat);
2413 sb->s_feature_ro_compat &= ~cpu_to_be32(ro);
2414 sb->s_feature_incompat &= ~cpu_to_be32(incompat);
2415 journal->j_revoke_records_per_block =
2416 journal_revoke_records_per_block(journal);
2418 EXPORT_SYMBOL(jbd2_journal_clear_features);
2421 * jbd2_journal_flush() - Flush journal
2422 * @journal: Journal to act on.
2423 * @flags: optional operation on the journal blocks after the flush (see below)
2425 * Flush all data for a given journal to disk and empty the journal.
2426 * Filesystems can use this when remounting readonly to ensure that
2427 * recovery does not need to happen on remount. Optionally, a discard or zeroout
2428 * can be issued on the journal blocks after flushing.
2431 * JBD2_JOURNAL_FLUSH_DISCARD: issues discards for the journal blocks
2432 * JBD2_JOURNAL_FLUSH_ZEROOUT: issues zeroouts for the journal blocks
2434 int jbd2_journal_flush(journal_t *journal, unsigned int flags)
2437 transaction_t *transaction = NULL;
2439 write_lock(&journal->j_state_lock);
2441 /* Force everything buffered to the log... */
2442 if (journal->j_running_transaction) {
2443 transaction = journal->j_running_transaction;
2444 __jbd2_log_start_commit(journal, transaction->t_tid);
2445 } else if (journal->j_committing_transaction)
2446 transaction = journal->j_committing_transaction;
2448 /* Wait for the log commit to complete... */
2450 tid_t tid = transaction->t_tid;
2452 write_unlock(&journal->j_state_lock);
2453 jbd2_log_wait_commit(journal, tid);
2455 write_unlock(&journal->j_state_lock);
2458 /* ...and flush everything in the log out to disk. */
2459 spin_lock(&journal->j_list_lock);
2460 while (!err && journal->j_checkpoint_transactions != NULL) {
2461 spin_unlock(&journal->j_list_lock);
2462 mutex_lock_io(&journal->j_checkpoint_mutex);
2463 err = jbd2_log_do_checkpoint(journal);
2464 mutex_unlock(&journal->j_checkpoint_mutex);
2465 spin_lock(&journal->j_list_lock);
2467 spin_unlock(&journal->j_list_lock);
2469 if (is_journal_aborted(journal))
2472 mutex_lock_io(&journal->j_checkpoint_mutex);
2474 err = jbd2_cleanup_journal_tail(journal);
2476 mutex_unlock(&journal->j_checkpoint_mutex);
2482 /* Finally, mark the journal as really needing no recovery.
2483 * This sets s_start==0 in the underlying superblock, which is
2484 * the magic code for a fully-recovered superblock. Any future
2485 * commits of data to the journal will restore the current
2487 jbd2_mark_journal_empty(journal, REQ_SYNC | REQ_FUA);
2490 err = __jbd2_journal_erase(journal, flags);
2492 mutex_unlock(&journal->j_checkpoint_mutex);
2493 write_lock(&journal->j_state_lock);
2494 J_ASSERT(!journal->j_running_transaction);
2495 J_ASSERT(!journal->j_committing_transaction);
2496 J_ASSERT(!journal->j_checkpoint_transactions);
2497 J_ASSERT(journal->j_head == journal->j_tail);
2498 J_ASSERT(journal->j_tail_sequence == journal->j_transaction_sequence);
2499 write_unlock(&journal->j_state_lock);
2505 * jbd2_journal_wipe() - Wipe journal contents
2506 * @journal: Journal to act on.
2507 * @write: flag (see below)
2509 * Wipe out all of the contents of a journal, safely. This will produce
2510 * a warning if the journal contains any valid recovery information.
2511 * Must be called between journal_init_*() and jbd2_journal_load().
2513 * If 'write' is non-zero, then we wipe out the journal on disk; otherwise
2514 * we merely suppress recovery.
2517 int jbd2_journal_wipe(journal_t *journal, int write)
2521 J_ASSERT (!(journal->j_flags & JBD2_LOADED));
2523 err = load_superblock(journal);
2527 if (!journal->j_tail)
2530 printk(KERN_WARNING "JBD2: %s recovery information on journal\n",
2531 write ? "Clearing" : "Ignoring");
2533 err = jbd2_journal_skip_recovery(journal);
2535 /* Lock to make assertions happy... */
2536 mutex_lock_io(&journal->j_checkpoint_mutex);
2537 jbd2_mark_journal_empty(journal, REQ_SYNC | REQ_FUA);
2538 mutex_unlock(&journal->j_checkpoint_mutex);
2546 * jbd2_journal_abort () - Shutdown the journal immediately.
2547 * @journal: the journal to shutdown.
2548 * @errno: an error number to record in the journal indicating
2549 * the reason for the shutdown.
2551 * Perform a complete, immediate shutdown of the ENTIRE
2552 * journal (not of a single transaction). This operation cannot be
2553 * undone without closing and reopening the journal.
2555 * The jbd2_journal_abort function is intended to support higher level error
2556 * recovery mechanisms such as the ext2/ext3 remount-readonly error
2559 * Journal abort has very specific semantics. Any existing dirty,
2560 * unjournaled buffers in the main filesystem will still be written to
2561 * disk by bdflush, but the journaling mechanism will be suspended
2562 * immediately and no further transaction commits will be honoured.
2564 * Any dirty, journaled buffers will be written back to disk without
2565 * hitting the journal. Atomicity cannot be guaranteed on an aborted
2566 * filesystem, but we _do_ attempt to leave as much data as possible
2567 * behind for fsck to use for cleanup.
2569 * Any attempt to get a new transaction handle on a journal which is in
2570 * ABORT state will just result in an -EROFS error return. A
2571 * jbd2_journal_stop on an existing handle will return -EIO if we have
2572 * entered abort state during the update.
2574 * Recursive transactions are not disturbed by journal abort until the
2575 * final jbd2_journal_stop, which will receive the -EIO error.
2577 * Finally, the jbd2_journal_abort call allows the caller to supply an errno
2578 * which will be recorded (if possible) in the journal superblock. This
2579 * allows a client to record failure conditions in the middle of a
2580 * transaction without having to complete the transaction to record the
2581 * failure to disk. ext3_error, for example, now uses this
2586 void jbd2_journal_abort(journal_t *journal, int errno)
2588 transaction_t *transaction;
2591 * Lock the aborting procedure until everything is done, this avoid
2592 * races between filesystem's error handling flow (e.g. ext4_abort()),
2593 * ensure panic after the error info is written into journal's
2596 mutex_lock(&journal->j_abort_mutex);
2598 * ESHUTDOWN always takes precedence because a file system check
2599 * caused by any other journal abort error is not required after
2600 * a shutdown triggered.
2602 write_lock(&journal->j_state_lock);
2603 if (journal->j_flags & JBD2_ABORT) {
2604 int old_errno = journal->j_errno;
2606 write_unlock(&journal->j_state_lock);
2607 if (old_errno != -ESHUTDOWN && errno == -ESHUTDOWN) {
2608 journal->j_errno = errno;
2609 jbd2_journal_update_sb_errno(journal);
2611 mutex_unlock(&journal->j_abort_mutex);
2616 * Mark the abort as occurred and start current running transaction
2617 * to release all journaled buffer.
2619 pr_err("Aborting journal on device %s.\n", journal->j_devname);
2621 journal->j_flags |= JBD2_ABORT;
2622 journal->j_errno = errno;
2623 transaction = journal->j_running_transaction;
2625 __jbd2_log_start_commit(journal, transaction->t_tid);
2626 write_unlock(&journal->j_state_lock);
2629 * Record errno to the journal super block, so that fsck and jbd2
2630 * layer could realise that a filesystem check is needed.
2632 jbd2_journal_update_sb_errno(journal);
2633 mutex_unlock(&journal->j_abort_mutex);
2637 * jbd2_journal_errno() - returns the journal's error state.
2638 * @journal: journal to examine.
2640 * This is the errno number set with jbd2_journal_abort(), the last
2641 * time the journal was mounted - if the journal was stopped
2642 * without calling abort this will be 0.
2644 * If the journal has been aborted on this mount time -EROFS will
2647 int jbd2_journal_errno(journal_t *journal)
2651 read_lock(&journal->j_state_lock);
2652 if (journal->j_flags & JBD2_ABORT)
2655 err = journal->j_errno;
2656 read_unlock(&journal->j_state_lock);
2661 * jbd2_journal_clear_err() - clears the journal's error state
2662 * @journal: journal to act on.
2664 * An error must be cleared or acked to take a FS out of readonly
2667 int jbd2_journal_clear_err(journal_t *journal)
2671 write_lock(&journal->j_state_lock);
2672 if (journal->j_flags & JBD2_ABORT)
2675 journal->j_errno = 0;
2676 write_unlock(&journal->j_state_lock);
2681 * jbd2_journal_ack_err() - Ack journal err.
2682 * @journal: journal to act on.
2684 * An error must be cleared or acked to take a FS out of readonly
2687 void jbd2_journal_ack_err(journal_t *journal)
2689 write_lock(&journal->j_state_lock);
2690 if (journal->j_errno)
2691 journal->j_flags |= JBD2_ACK_ERR;
2692 write_unlock(&journal->j_state_lock);
2695 int jbd2_journal_blocks_per_page(struct inode *inode)
2697 return 1 << (PAGE_SHIFT - inode->i_sb->s_blocksize_bits);
2701 * helper functions to deal with 32 or 64bit block numbers.
2703 size_t journal_tag_bytes(journal_t *journal)
2707 if (jbd2_has_feature_csum3(journal))
2708 return sizeof(journal_block_tag3_t);
2710 sz = sizeof(journal_block_tag_t);
2712 if (jbd2_has_feature_csum2(journal))
2713 sz += sizeof(__u16);
2715 if (jbd2_has_feature_64bit(journal))
2718 return sz - sizeof(__u32);
2722 * JBD memory management
2724 * These functions are used to allocate block-sized chunks of memory
2725 * used for making copies of buffer_head data. Very often it will be
2726 * page-sized chunks of data, but sometimes it will be in
2727 * sub-page-size chunks. (For example, 16k pages on Power systems
2728 * with a 4k block file system.) For blocks smaller than a page, we
2729 * use a SLAB allocator. There are slab caches for each block size,
2730 * which are allocated at mount time, if necessary, and we only free
2731 * (all of) the slab caches when/if the jbd2 module is unloaded. For
2732 * this reason we don't need to a mutex to protect access to
2733 * jbd2_slab[] allocating or releasing memory; only in
2734 * jbd2_journal_create_slab().
2736 #define JBD2_MAX_SLABS 8
2737 static struct kmem_cache *jbd2_slab[JBD2_MAX_SLABS];
2739 static const char *jbd2_slab_names[JBD2_MAX_SLABS] = {
2740 "jbd2_1k", "jbd2_2k", "jbd2_4k", "jbd2_8k",
2741 "jbd2_16k", "jbd2_32k", "jbd2_64k", "jbd2_128k"
2745 static void jbd2_journal_destroy_slabs(void)
2749 for (i = 0; i < JBD2_MAX_SLABS; i++) {
2750 kmem_cache_destroy(jbd2_slab[i]);
2751 jbd2_slab[i] = NULL;
2755 static int jbd2_journal_create_slab(size_t size)
2757 static DEFINE_MUTEX(jbd2_slab_create_mutex);
2758 int i = order_base_2(size) - 10;
2761 if (size == PAGE_SIZE)
2764 if (i >= JBD2_MAX_SLABS)
2767 if (unlikely(i < 0))
2769 mutex_lock(&jbd2_slab_create_mutex);
2771 mutex_unlock(&jbd2_slab_create_mutex);
2772 return 0; /* Already created */
2775 slab_size = 1 << (i+10);
2776 jbd2_slab[i] = kmem_cache_create(jbd2_slab_names[i], slab_size,
2777 slab_size, 0, NULL);
2778 mutex_unlock(&jbd2_slab_create_mutex);
2779 if (!jbd2_slab[i]) {
2780 printk(KERN_EMERG "JBD2: no memory for jbd2_slab cache\n");
2786 static struct kmem_cache *get_slab(size_t size)
2788 int i = order_base_2(size) - 10;
2790 BUG_ON(i >= JBD2_MAX_SLABS);
2791 if (unlikely(i < 0))
2793 BUG_ON(jbd2_slab[i] == NULL);
2794 return jbd2_slab[i];
2797 void *jbd2_alloc(size_t size, gfp_t flags)
2801 BUG_ON(size & (size-1)); /* Must be a power of 2 */
2803 if (size < PAGE_SIZE)
2804 ptr = kmem_cache_alloc(get_slab(size), flags);
2806 ptr = (void *)__get_free_pages(flags, get_order(size));
2808 /* Check alignment; SLUB has gotten this wrong in the past,
2809 * and this can lead to user data corruption! */
2810 BUG_ON(((unsigned long) ptr) & (size-1));
2815 void jbd2_free(void *ptr, size_t size)
2817 if (size < PAGE_SIZE)
2818 kmem_cache_free(get_slab(size), ptr);
2820 free_pages((unsigned long)ptr, get_order(size));
2824 * Journal_head storage management
2826 static struct kmem_cache *jbd2_journal_head_cache;
2827 #ifdef CONFIG_JBD2_DEBUG
2828 static atomic_t nr_journal_heads = ATOMIC_INIT(0);
2831 static int __init jbd2_journal_init_journal_head_cache(void)
2833 J_ASSERT(!jbd2_journal_head_cache);
2834 jbd2_journal_head_cache = kmem_cache_create("jbd2_journal_head",
2835 sizeof(struct journal_head),
2837 SLAB_TEMPORARY | SLAB_TYPESAFE_BY_RCU,
2839 if (!jbd2_journal_head_cache) {
2840 printk(KERN_EMERG "JBD2: no memory for journal_head cache\n");
2846 static void jbd2_journal_destroy_journal_head_cache(void)
2848 kmem_cache_destroy(jbd2_journal_head_cache);
2849 jbd2_journal_head_cache = NULL;
2853 * journal_head splicing and dicing
2855 static struct journal_head *journal_alloc_journal_head(void)
2857 struct journal_head *ret;
2859 #ifdef CONFIG_JBD2_DEBUG
2860 atomic_inc(&nr_journal_heads);
2862 ret = kmem_cache_zalloc(jbd2_journal_head_cache, GFP_NOFS);
2864 jbd2_debug(1, "out of memory for journal_head\n");
2865 pr_notice_ratelimited("ENOMEM in %s, retrying.\n", __func__);
2866 ret = kmem_cache_zalloc(jbd2_journal_head_cache,
2867 GFP_NOFS | __GFP_NOFAIL);
2870 spin_lock_init(&ret->b_state_lock);
2874 static void journal_free_journal_head(struct journal_head *jh)
2876 #ifdef CONFIG_JBD2_DEBUG
2877 atomic_dec(&nr_journal_heads);
2878 memset(jh, JBD2_POISON_FREE, sizeof(*jh));
2880 kmem_cache_free(jbd2_journal_head_cache, jh);
2884 * A journal_head is attached to a buffer_head whenever JBD has an
2885 * interest in the buffer.
2887 * Whenever a buffer has an attached journal_head, its ->b_state:BH_JBD bit
2888 * is set. This bit is tested in core kernel code where we need to take
2889 * JBD-specific actions. Testing the zeroness of ->b_private is not reliable
2892 * When a buffer has its BH_JBD bit set, its ->b_count is elevated by one.
2894 * When a buffer has its BH_JBD bit set it is immune from being released by
2895 * core kernel code, mainly via ->b_count.
2897 * A journal_head is detached from its buffer_head when the journal_head's
2898 * b_jcount reaches zero. Running transaction (b_transaction) and checkpoint
2899 * transaction (b_cp_transaction) hold their references to b_jcount.
2901 * Various places in the kernel want to attach a journal_head to a buffer_head
2902 * _before_ attaching the journal_head to a transaction. To protect the
2903 * journal_head in this situation, jbd2_journal_add_journal_head elevates the
2904 * journal_head's b_jcount refcount by one. The caller must call
2905 * jbd2_journal_put_journal_head() to undo this.
2907 * So the typical usage would be:
2909 * (Attach a journal_head if needed. Increments b_jcount)
2910 * struct journal_head *jh = jbd2_journal_add_journal_head(bh);
2912 * (Get another reference for transaction)
2913 * jbd2_journal_grab_journal_head(bh);
2914 * jh->b_transaction = xxx;
2915 * (Put original reference)
2916 * jbd2_journal_put_journal_head(jh);
2920 * Give a buffer_head a journal_head.
2924 struct journal_head *jbd2_journal_add_journal_head(struct buffer_head *bh)
2926 struct journal_head *jh;
2927 struct journal_head *new_jh = NULL;
2930 if (!buffer_jbd(bh))
2931 new_jh = journal_alloc_journal_head();
2933 jbd_lock_bh_journal_head(bh);
2934 if (buffer_jbd(bh)) {
2938 (atomic_read(&bh->b_count) > 0) ||
2939 (bh->b_page && bh->b_page->mapping));
2942 jbd_unlock_bh_journal_head(bh);
2947 new_jh = NULL; /* We consumed it */
2952 BUFFER_TRACE(bh, "added journal_head");
2955 jbd_unlock_bh_journal_head(bh);
2957 journal_free_journal_head(new_jh);
2958 return bh->b_private;
2962 * Grab a ref against this buffer_head's journal_head. If it ended up not
2963 * having a journal_head, return NULL
2965 struct journal_head *jbd2_journal_grab_journal_head(struct buffer_head *bh)
2967 struct journal_head *jh = NULL;
2969 jbd_lock_bh_journal_head(bh);
2970 if (buffer_jbd(bh)) {
2974 jbd_unlock_bh_journal_head(bh);
2977 EXPORT_SYMBOL(jbd2_journal_grab_journal_head);
2979 static void __journal_remove_journal_head(struct buffer_head *bh)
2981 struct journal_head *jh = bh2jh(bh);
2983 J_ASSERT_JH(jh, jh->b_transaction == NULL);
2984 J_ASSERT_JH(jh, jh->b_next_transaction == NULL);
2985 J_ASSERT_JH(jh, jh->b_cp_transaction == NULL);
2986 J_ASSERT_JH(jh, jh->b_jlist == BJ_None);
2987 J_ASSERT_BH(bh, buffer_jbd(bh));
2988 J_ASSERT_BH(bh, jh2bh(jh) == bh);
2989 BUFFER_TRACE(bh, "remove journal_head");
2991 /* Unlink before dropping the lock */
2992 bh->b_private = NULL;
2993 jh->b_bh = NULL; /* debug, really */
2994 clear_buffer_jbd(bh);
2997 static void journal_release_journal_head(struct journal_head *jh, size_t b_size)
2999 if (jh->b_frozen_data) {
3000 printk(KERN_WARNING "%s: freeing b_frozen_data\n", __func__);
3001 jbd2_free(jh->b_frozen_data, b_size);
3003 if (jh->b_committed_data) {
3004 printk(KERN_WARNING "%s: freeing b_committed_data\n", __func__);
3005 jbd2_free(jh->b_committed_data, b_size);
3007 journal_free_journal_head(jh);
3011 * Drop a reference on the passed journal_head. If it fell to zero then
3012 * release the journal_head from the buffer_head.
3014 void jbd2_journal_put_journal_head(struct journal_head *jh)
3016 struct buffer_head *bh = jh2bh(jh);
3018 jbd_lock_bh_journal_head(bh);
3019 J_ASSERT_JH(jh, jh->b_jcount > 0);
3021 if (!jh->b_jcount) {
3022 __journal_remove_journal_head(bh);
3023 jbd_unlock_bh_journal_head(bh);
3024 journal_release_journal_head(jh, bh->b_size);
3027 jbd_unlock_bh_journal_head(bh);
3030 EXPORT_SYMBOL(jbd2_journal_put_journal_head);
3033 * Initialize jbd inode head
3035 void jbd2_journal_init_jbd_inode(struct jbd2_inode *jinode, struct inode *inode)
3037 jinode->i_transaction = NULL;
3038 jinode->i_next_transaction = NULL;
3039 jinode->i_vfs_inode = inode;
3040 jinode->i_flags = 0;
3041 jinode->i_dirty_start = 0;
3042 jinode->i_dirty_end = 0;
3043 INIT_LIST_HEAD(&jinode->i_list);
3047 * Function to be called before we start removing inode from memory (i.e.,
3048 * clear_inode() is a fine place to be called from). It removes inode from
3049 * transaction's lists.
3051 void jbd2_journal_release_jbd_inode(journal_t *journal,
3052 struct jbd2_inode *jinode)
3057 spin_lock(&journal->j_list_lock);
3058 /* Is commit writing out inode - we have to wait */
3059 if (jinode->i_flags & JI_COMMIT_RUNNING) {
3060 wait_queue_head_t *wq;
3061 DEFINE_WAIT_BIT(wait, &jinode->i_flags, __JI_COMMIT_RUNNING);
3062 wq = bit_waitqueue(&jinode->i_flags, __JI_COMMIT_RUNNING);
3063 prepare_to_wait(wq, &wait.wq_entry, TASK_UNINTERRUPTIBLE);
3064 spin_unlock(&journal->j_list_lock);
3066 finish_wait(wq, &wait.wq_entry);
3070 if (jinode->i_transaction) {
3071 list_del(&jinode->i_list);
3072 jinode->i_transaction = NULL;
3074 spin_unlock(&journal->j_list_lock);
3078 #ifdef CONFIG_PROC_FS
3080 #define JBD2_STATS_PROC_NAME "fs/jbd2"
3082 static void __init jbd2_create_jbd_stats_proc_entry(void)
3084 proc_jbd2_stats = proc_mkdir(JBD2_STATS_PROC_NAME, NULL);
3087 static void __exit jbd2_remove_jbd_stats_proc_entry(void)
3089 if (proc_jbd2_stats)
3090 remove_proc_entry(JBD2_STATS_PROC_NAME, NULL);
3095 #define jbd2_create_jbd_stats_proc_entry() do {} while (0)
3096 #define jbd2_remove_jbd_stats_proc_entry() do {} while (0)
3100 struct kmem_cache *jbd2_handle_cache, *jbd2_inode_cache;
3102 static int __init jbd2_journal_init_inode_cache(void)
3104 J_ASSERT(!jbd2_inode_cache);
3105 jbd2_inode_cache = KMEM_CACHE(jbd2_inode, 0);
3106 if (!jbd2_inode_cache) {
3107 pr_emerg("JBD2: failed to create inode cache\n");
3113 static int __init jbd2_journal_init_handle_cache(void)
3115 J_ASSERT(!jbd2_handle_cache);
3116 jbd2_handle_cache = KMEM_CACHE(jbd2_journal_handle, SLAB_TEMPORARY);
3117 if (!jbd2_handle_cache) {
3118 printk(KERN_EMERG "JBD2: failed to create handle cache\n");
3124 static void jbd2_journal_destroy_inode_cache(void)
3126 kmem_cache_destroy(jbd2_inode_cache);
3127 jbd2_inode_cache = NULL;
3130 static void jbd2_journal_destroy_handle_cache(void)
3132 kmem_cache_destroy(jbd2_handle_cache);
3133 jbd2_handle_cache = NULL;
3137 * Module startup and shutdown
3140 static int __init journal_init_caches(void)
3144 ret = jbd2_journal_init_revoke_record_cache();
3146 ret = jbd2_journal_init_revoke_table_cache();
3148 ret = jbd2_journal_init_journal_head_cache();
3150 ret = jbd2_journal_init_handle_cache();
3152 ret = jbd2_journal_init_inode_cache();
3154 ret = jbd2_journal_init_transaction_cache();
3158 static void jbd2_journal_destroy_caches(void)
3160 jbd2_journal_destroy_revoke_record_cache();
3161 jbd2_journal_destroy_revoke_table_cache();
3162 jbd2_journal_destroy_journal_head_cache();
3163 jbd2_journal_destroy_handle_cache();
3164 jbd2_journal_destroy_inode_cache();
3165 jbd2_journal_destroy_transaction_cache();
3166 jbd2_journal_destroy_slabs();
3169 static int __init journal_init(void)
3173 BUILD_BUG_ON(sizeof(struct journal_superblock_s) != 1024);
3175 ret = journal_init_caches();
3177 jbd2_create_jbd_stats_proc_entry();
3179 jbd2_journal_destroy_caches();
3184 static void __exit journal_exit(void)
3186 #ifdef CONFIG_JBD2_DEBUG
3187 int n = atomic_read(&nr_journal_heads);
3189 printk(KERN_ERR "JBD2: leaked %d journal_heads!\n", n);
3191 jbd2_remove_jbd_stats_proc_entry();
3192 jbd2_journal_destroy_caches();
3195 MODULE_LICENSE("GPL");
3196 module_init(journal_init);
3197 module_exit(journal_exit);