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 * Update j_fc_off so jbd2_fc_release_bufs can release remain
930 if (unlikely(!buffer_uptodate(bh))) {
931 journal->j_fc_off = i + 1;
935 journal->j_fc_wbuf[i] = NULL;
940 EXPORT_SYMBOL(jbd2_fc_wait_bufs);
942 int jbd2_fc_release_bufs(journal_t *journal)
944 struct buffer_head *bh;
947 j_fc_off = journal->j_fc_off;
949 for (i = j_fc_off - 1; i >= 0; i--) {
950 bh = journal->j_fc_wbuf[i];
954 journal->j_fc_wbuf[i] = NULL;
959 EXPORT_SYMBOL(jbd2_fc_release_bufs);
962 * Conversion of logical to physical block numbers for the journal
964 * On external journals the journal blocks are identity-mapped, so
965 * this is a no-op. If needed, we can use j_blk_offset - everything is
968 int jbd2_journal_bmap(journal_t *journal, unsigned long blocknr,
969 unsigned long long *retp)
972 unsigned long long ret;
973 sector_t block = blocknr;
975 if (journal->j_bmap) {
976 err = journal->j_bmap(journal, &block);
979 } else 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()
1296 * @shrink: shrinker to work on
1297 * @sc: reclaim request to process
1299 * Scan the checkpointed buffer on the checkpoint list and release the
1302 static unsigned long jbd2_journal_shrink_scan(struct shrinker *shrink,
1303 struct shrink_control *sc)
1305 journal_t *journal = container_of(shrink, journal_t, j_shrinker);
1306 unsigned long nr_to_scan = sc->nr_to_scan;
1307 unsigned long nr_shrunk;
1308 unsigned long count;
1310 count = percpu_counter_read_positive(&journal->j_checkpoint_jh_count);
1311 trace_jbd2_shrink_scan_enter(journal, sc->nr_to_scan, count);
1313 nr_shrunk = jbd2_journal_shrink_checkpoint_list(journal, &nr_to_scan);
1315 count = percpu_counter_read_positive(&journal->j_checkpoint_jh_count);
1316 trace_jbd2_shrink_scan_exit(journal, nr_to_scan, nr_shrunk, count);
1322 * jbd2_journal_shrink_count()
1323 * @shrink: shrinker to work on
1324 * @sc: reclaim request to process
1326 * Count the number of checkpoint buffers on the checkpoint list.
1328 static unsigned long jbd2_journal_shrink_count(struct shrinker *shrink,
1329 struct shrink_control *sc)
1331 journal_t *journal = container_of(shrink, journal_t, j_shrinker);
1332 unsigned long count;
1334 count = percpu_counter_read_positive(&journal->j_checkpoint_jh_count);
1335 trace_jbd2_shrink_count(journal, sc->nr_to_scan, count);
1341 * Management for journal control blocks: functions to create and
1342 * destroy journal_t structures, and to initialise and read existing
1343 * journal blocks from disk. */
1345 /* First: create and setup a journal_t object in memory. We initialise
1346 * very few fields yet: that has to wait until we have created the
1347 * journal structures from from scratch, or loaded them from disk. */
1349 static journal_t *journal_init_common(struct block_device *bdev,
1350 struct block_device *fs_dev,
1351 unsigned long long start, int len, int blocksize)
1353 static struct lock_class_key jbd2_trans_commit_key;
1356 struct buffer_head *bh;
1359 journal = kzalloc(sizeof(*journal), GFP_KERNEL);
1363 init_waitqueue_head(&journal->j_wait_transaction_locked);
1364 init_waitqueue_head(&journal->j_wait_done_commit);
1365 init_waitqueue_head(&journal->j_wait_commit);
1366 init_waitqueue_head(&journal->j_wait_updates);
1367 init_waitqueue_head(&journal->j_wait_reserved);
1368 init_waitqueue_head(&journal->j_fc_wait);
1369 mutex_init(&journal->j_abort_mutex);
1370 mutex_init(&journal->j_barrier);
1371 mutex_init(&journal->j_checkpoint_mutex);
1372 spin_lock_init(&journal->j_revoke_lock);
1373 spin_lock_init(&journal->j_list_lock);
1374 rwlock_init(&journal->j_state_lock);
1376 journal->j_commit_interval = (HZ * JBD2_DEFAULT_MAX_COMMIT_AGE);
1377 journal->j_min_batch_time = 0;
1378 journal->j_max_batch_time = 15000; /* 15ms */
1379 atomic_set(&journal->j_reserved_credits, 0);
1381 /* The journal is marked for error until we succeed with recovery! */
1382 journal->j_flags = JBD2_ABORT;
1384 /* Set up a default-sized revoke table for the new mount. */
1385 err = jbd2_journal_init_revoke(journal, JOURNAL_REVOKE_DEFAULT_HASH);
1389 spin_lock_init(&journal->j_history_lock);
1391 lockdep_init_map(&journal->j_trans_commit_map, "jbd2_handle",
1392 &jbd2_trans_commit_key, 0);
1394 /* journal descriptor can store up to n blocks -bzzz */
1395 journal->j_blocksize = blocksize;
1396 journal->j_dev = bdev;
1397 journal->j_fs_dev = fs_dev;
1398 journal->j_blk_offset = start;
1399 journal->j_total_len = len;
1400 /* We need enough buffers to write out full descriptor block. */
1401 n = journal->j_blocksize / jbd2_min_tag_size();
1402 journal->j_wbufsize = n;
1403 journal->j_fc_wbuf = NULL;
1404 journal->j_wbuf = kmalloc_array(n, sizeof(struct buffer_head *),
1406 if (!journal->j_wbuf)
1409 bh = getblk_unmovable(journal->j_dev, start, journal->j_blocksize);
1411 pr_err("%s: Cannot get buffer for journal superblock\n",
1415 journal->j_sb_buffer = bh;
1416 journal->j_superblock = (journal_superblock_t *)bh->b_data;
1418 journal->j_shrink_transaction = NULL;
1419 journal->j_shrinker.scan_objects = jbd2_journal_shrink_scan;
1420 journal->j_shrinker.count_objects = jbd2_journal_shrink_count;
1421 journal->j_shrinker.seeks = DEFAULT_SEEKS;
1422 journal->j_shrinker.batch = journal->j_max_transaction_buffers;
1424 if (percpu_counter_init(&journal->j_checkpoint_jh_count, 0, GFP_KERNEL))
1427 if (register_shrinker(&journal->j_shrinker, "jbd2-journal:(%u:%u)",
1428 MAJOR(bdev->bd_dev), MINOR(bdev->bd_dev))) {
1429 percpu_counter_destroy(&journal->j_checkpoint_jh_count);
1435 brelse(journal->j_sb_buffer);
1436 kfree(journal->j_wbuf);
1437 jbd2_journal_destroy_revoke(journal);
1442 /* jbd2_journal_init_dev and jbd2_journal_init_inode:
1444 * Create a journal structure assigned some fixed set of disk blocks to
1445 * the journal. We don't actually touch those disk blocks yet, but we
1446 * need to set up all of the mapping information to tell the journaling
1447 * system where the journal blocks are.
1452 * journal_t * jbd2_journal_init_dev() - creates and initialises a journal structure
1453 * @bdev: Block device on which to create the journal
1454 * @fs_dev: Device which hold journalled filesystem for this journal.
1455 * @start: Block nr Start of journal.
1456 * @len: Length of the journal in blocks.
1457 * @blocksize: blocksize of journalling device
1459 * Returns: a newly created journal_t *
1461 * jbd2_journal_init_dev creates a journal which maps a fixed contiguous
1462 * range of blocks on an arbitrary block device.
1465 journal_t *jbd2_journal_init_dev(struct block_device *bdev,
1466 struct block_device *fs_dev,
1467 unsigned long long start, int len, int blocksize)
1471 journal = journal_init_common(bdev, fs_dev, start, len, blocksize);
1475 snprintf(journal->j_devname, sizeof(journal->j_devname),
1476 "%pg", journal->j_dev);
1477 strreplace(journal->j_devname, '/', '!');
1478 jbd2_stats_proc_init(journal);
1484 * journal_t * jbd2_journal_init_inode () - creates a journal which maps to a inode.
1485 * @inode: An inode to create the journal in
1487 * jbd2_journal_init_inode creates a journal which maps an on-disk inode as
1488 * the journal. The inode must exist already, must support bmap() and
1489 * must have all data blocks preallocated.
1491 journal_t *jbd2_journal_init_inode(struct inode *inode)
1499 err = bmap(inode, &blocknr);
1501 if (err || !blocknr) {
1502 pr_err("%s: Cannot locate journal superblock\n",
1507 jbd2_debug(1, "JBD2: inode %s/%ld, size %lld, bits %d, blksize %ld\n",
1508 inode->i_sb->s_id, inode->i_ino, (long long) inode->i_size,
1509 inode->i_sb->s_blocksize_bits, inode->i_sb->s_blocksize);
1511 journal = journal_init_common(inode->i_sb->s_bdev, inode->i_sb->s_bdev,
1512 blocknr, inode->i_size >> inode->i_sb->s_blocksize_bits,
1513 inode->i_sb->s_blocksize);
1517 journal->j_inode = inode;
1518 snprintf(journal->j_devname, sizeof(journal->j_devname),
1519 "%pg", journal->j_dev);
1520 p = strreplace(journal->j_devname, '/', '!');
1521 sprintf(p, "-%lu", journal->j_inode->i_ino);
1522 jbd2_stats_proc_init(journal);
1528 * If the journal init or create aborts, we need to mark the journal
1529 * superblock as being NULL to prevent the journal destroy from writing
1530 * back a bogus superblock.
1532 static void journal_fail_superblock(journal_t *journal)
1534 struct buffer_head *bh = journal->j_sb_buffer;
1536 journal->j_sb_buffer = NULL;
1540 * Given a journal_t structure, initialise the various fields for
1541 * startup of a new journaling session. We use this both when creating
1542 * a journal, and after recovering an old journal to reset it for
1546 static int journal_reset(journal_t *journal)
1548 journal_superblock_t *sb = journal->j_superblock;
1549 unsigned long long first, last;
1551 first = be32_to_cpu(sb->s_first);
1552 last = be32_to_cpu(sb->s_maxlen);
1553 if (first + JBD2_MIN_JOURNAL_BLOCKS > last + 1) {
1554 printk(KERN_ERR "JBD2: Journal too short (blocks %llu-%llu).\n",
1556 journal_fail_superblock(journal);
1560 journal->j_first = first;
1561 journal->j_last = last;
1563 journal->j_head = journal->j_first;
1564 journal->j_tail = journal->j_first;
1565 journal->j_free = journal->j_last - journal->j_first;
1567 journal->j_tail_sequence = journal->j_transaction_sequence;
1568 journal->j_commit_sequence = journal->j_transaction_sequence - 1;
1569 journal->j_commit_request = journal->j_commit_sequence;
1571 journal->j_max_transaction_buffers = jbd2_journal_get_max_txn_bufs(journal);
1574 * Now that journal recovery is done, turn fast commits off here. This
1575 * way, if fast commit was enabled before the crash but if now FS has
1576 * disabled it, we don't enable fast commits.
1578 jbd2_clear_feature_fast_commit(journal);
1581 * As a special case, if the on-disk copy is already marked as needing
1582 * no recovery (s_start == 0), then we can safely defer the superblock
1583 * update until the next commit by setting JBD2_FLUSHED. This avoids
1584 * attempting a write to a potential-readonly device.
1586 if (sb->s_start == 0) {
1587 jbd2_debug(1, "JBD2: Skipping superblock update on recovered sb "
1588 "(start %ld, seq %u, errno %d)\n",
1589 journal->j_tail, journal->j_tail_sequence,
1591 journal->j_flags |= JBD2_FLUSHED;
1593 /* Lock here to make assertions happy... */
1594 mutex_lock_io(&journal->j_checkpoint_mutex);
1596 * Update log tail information. We use REQ_FUA since new
1597 * transaction will start reusing journal space and so we
1598 * must make sure information about current log tail is on
1601 jbd2_journal_update_sb_log_tail(journal,
1602 journal->j_tail_sequence,
1604 REQ_SYNC | REQ_FUA);
1605 mutex_unlock(&journal->j_checkpoint_mutex);
1607 return jbd2_journal_start_thread(journal);
1611 * This function expects that the caller will have locked the journal
1612 * buffer head, and will return with it unlocked
1614 static int jbd2_write_superblock(journal_t *journal, blk_opf_t write_flags)
1616 struct buffer_head *bh = journal->j_sb_buffer;
1617 journal_superblock_t *sb = journal->j_superblock;
1620 /* Buffer got discarded which means block device got invalidated */
1621 if (!buffer_mapped(bh)) {
1626 trace_jbd2_write_superblock(journal, write_flags);
1627 if (!(journal->j_flags & JBD2_BARRIER))
1628 write_flags &= ~(REQ_FUA | REQ_PREFLUSH);
1629 if (buffer_write_io_error(bh)) {
1631 * Oh, dear. A previous attempt to write the journal
1632 * superblock failed. This could happen because the
1633 * USB device was yanked out. Or it could happen to
1634 * be a transient write error and maybe the block will
1635 * be remapped. Nothing we can do but to retry the
1636 * write and hope for the best.
1638 printk(KERN_ERR "JBD2: previous I/O error detected "
1639 "for journal superblock update for %s.\n",
1640 journal->j_devname);
1641 clear_buffer_write_io_error(bh);
1642 set_buffer_uptodate(bh);
1644 if (jbd2_journal_has_csum_v2or3(journal))
1645 sb->s_checksum = jbd2_superblock_csum(journal, sb);
1647 bh->b_end_io = end_buffer_write_sync;
1648 submit_bh(REQ_OP_WRITE | write_flags, bh);
1650 if (buffer_write_io_error(bh)) {
1651 clear_buffer_write_io_error(bh);
1652 set_buffer_uptodate(bh);
1656 printk(KERN_ERR "JBD2: I/O error when updating journal superblock for %s.\n",
1657 journal->j_devname);
1658 if (!is_journal_aborted(journal))
1659 jbd2_journal_abort(journal, ret);
1666 * jbd2_journal_update_sb_log_tail() - Update log tail in journal sb on disk.
1667 * @journal: The journal to update.
1668 * @tail_tid: TID of the new transaction at the tail of the log
1669 * @tail_block: The first block of the transaction at the tail of the log
1670 * @write_flags: Flags for the journal sb write operation
1672 * Update a journal's superblock information about log tail and write it to
1673 * disk, waiting for the IO to complete.
1675 int jbd2_journal_update_sb_log_tail(journal_t *journal, tid_t tail_tid,
1676 unsigned long tail_block,
1677 blk_opf_t write_flags)
1679 journal_superblock_t *sb = journal->j_superblock;
1682 if (is_journal_aborted(journal))
1684 if (test_bit(JBD2_CHECKPOINT_IO_ERROR, &journal->j_atomic_flags)) {
1685 jbd2_journal_abort(journal, -EIO);
1689 BUG_ON(!mutex_is_locked(&journal->j_checkpoint_mutex));
1690 jbd2_debug(1, "JBD2: updating superblock (start %lu, seq %u)\n",
1691 tail_block, tail_tid);
1693 lock_buffer(journal->j_sb_buffer);
1694 sb->s_sequence = cpu_to_be32(tail_tid);
1695 sb->s_start = cpu_to_be32(tail_block);
1697 ret = jbd2_write_superblock(journal, write_flags);
1701 /* Log is no longer empty */
1702 write_lock(&journal->j_state_lock);
1703 WARN_ON(!sb->s_sequence);
1704 journal->j_flags &= ~JBD2_FLUSHED;
1705 write_unlock(&journal->j_state_lock);
1712 * jbd2_mark_journal_empty() - Mark on disk journal as empty.
1713 * @journal: The journal to update.
1714 * @write_flags: Flags for the journal sb write operation
1716 * Update a journal's dynamic superblock fields to show that journal is empty.
1717 * Write updated superblock to disk waiting for IO to complete.
1719 static void jbd2_mark_journal_empty(journal_t *journal, blk_opf_t write_flags)
1721 journal_superblock_t *sb = journal->j_superblock;
1722 bool had_fast_commit = false;
1724 BUG_ON(!mutex_is_locked(&journal->j_checkpoint_mutex));
1725 lock_buffer(journal->j_sb_buffer);
1726 if (sb->s_start == 0) { /* Is it already empty? */
1727 unlock_buffer(journal->j_sb_buffer);
1731 jbd2_debug(1, "JBD2: Marking journal as empty (seq %u)\n",
1732 journal->j_tail_sequence);
1734 sb->s_sequence = cpu_to_be32(journal->j_tail_sequence);
1735 sb->s_start = cpu_to_be32(0);
1736 if (jbd2_has_feature_fast_commit(journal)) {
1738 * When journal is clean, no need to commit fast commit flag and
1739 * make file system incompatible with older kernels.
1741 jbd2_clear_feature_fast_commit(journal);
1742 had_fast_commit = true;
1745 jbd2_write_superblock(journal, write_flags);
1747 if (had_fast_commit)
1748 jbd2_set_feature_fast_commit(journal);
1750 /* Log is no longer empty */
1751 write_lock(&journal->j_state_lock);
1752 journal->j_flags |= JBD2_FLUSHED;
1753 write_unlock(&journal->j_state_lock);
1757 * __jbd2_journal_erase() - Discard or zeroout journal blocks (excluding superblock)
1758 * @journal: The journal to erase.
1759 * @flags: A discard/zeroout request is sent for each physically contigous
1760 * region of the journal. Either JBD2_JOURNAL_FLUSH_DISCARD or
1761 * JBD2_JOURNAL_FLUSH_ZEROOUT must be set to determine which operation
1764 * Note: JBD2_JOURNAL_FLUSH_ZEROOUT attempts to use hardware offload. Zeroes
1765 * will be explicitly written if no hardware offload is available, see
1766 * blkdev_issue_zeroout for more details.
1768 static int __jbd2_journal_erase(journal_t *journal, unsigned int flags)
1771 unsigned long block, log_offset; /* logical */
1772 unsigned long long phys_block, block_start, block_stop; /* physical */
1773 loff_t byte_start, byte_stop, byte_count;
1775 /* flags must be set to either discard or zeroout */
1776 if ((flags & ~JBD2_JOURNAL_FLUSH_VALID) || !flags ||
1777 ((flags & JBD2_JOURNAL_FLUSH_DISCARD) &&
1778 (flags & JBD2_JOURNAL_FLUSH_ZEROOUT)))
1781 if ((flags & JBD2_JOURNAL_FLUSH_DISCARD) &&
1782 !bdev_max_discard_sectors(journal->j_dev))
1786 * lookup block mapping and issue discard/zeroout for each
1789 log_offset = be32_to_cpu(journal->j_superblock->s_first);
1790 block_start = ~0ULL;
1791 for (block = log_offset; block < journal->j_total_len; block++) {
1792 err = jbd2_journal_bmap(journal, block, &phys_block);
1794 pr_err("JBD2: bad block at offset %lu", block);
1798 if (block_start == ~0ULL) {
1799 block_start = phys_block;
1800 block_stop = block_start - 1;
1804 * last block not contiguous with current block,
1805 * process last contiguous region and return to this block on
1808 if (phys_block != block_stop + 1) {
1813 * if this isn't the last block of journal,
1814 * no need to process now because next block may also
1815 * be part of this contiguous region
1817 if (block != journal->j_total_len - 1)
1822 * end of contiguous region or this is last block of journal,
1823 * take care of the region
1825 byte_start = block_start * journal->j_blocksize;
1826 byte_stop = block_stop * journal->j_blocksize;
1827 byte_count = (block_stop - block_start + 1) *
1828 journal->j_blocksize;
1830 truncate_inode_pages_range(journal->j_dev->bd_inode->i_mapping,
1831 byte_start, byte_stop);
1833 if (flags & JBD2_JOURNAL_FLUSH_DISCARD) {
1834 err = blkdev_issue_discard(journal->j_dev,
1835 byte_start >> SECTOR_SHIFT,
1836 byte_count >> SECTOR_SHIFT,
1838 } else if (flags & JBD2_JOURNAL_FLUSH_ZEROOUT) {
1839 err = blkdev_issue_zeroout(journal->j_dev,
1840 byte_start >> SECTOR_SHIFT,
1841 byte_count >> SECTOR_SHIFT,
1845 if (unlikely(err != 0)) {
1846 pr_err("JBD2: (error %d) unable to wipe journal at physical blocks %llu - %llu",
1847 err, block_start, block_stop);
1851 /* reset start and stop after processing a region */
1852 block_start = ~0ULL;
1855 return blkdev_issue_flush(journal->j_dev);
1859 * jbd2_journal_update_sb_errno() - Update error in the journal.
1860 * @journal: The journal to update.
1862 * Update a journal's errno. Write updated superblock to disk waiting for IO
1865 void jbd2_journal_update_sb_errno(journal_t *journal)
1867 journal_superblock_t *sb = journal->j_superblock;
1870 lock_buffer(journal->j_sb_buffer);
1871 errcode = journal->j_errno;
1872 if (errcode == -ESHUTDOWN)
1874 jbd2_debug(1, "JBD2: updating superblock error (errno %d)\n", errcode);
1875 sb->s_errno = cpu_to_be32(errcode);
1877 jbd2_write_superblock(journal, REQ_SYNC | REQ_FUA);
1879 EXPORT_SYMBOL(jbd2_journal_update_sb_errno);
1881 static int journal_revoke_records_per_block(journal_t *journal)
1884 int space = journal->j_blocksize - sizeof(jbd2_journal_revoke_header_t);
1886 if (jbd2_has_feature_64bit(journal))
1891 if (jbd2_journal_has_csum_v2or3(journal))
1892 space -= sizeof(struct jbd2_journal_block_tail);
1893 return space / record_size;
1897 * Read the superblock for a given journal, performing initial
1898 * validation of the format.
1900 static int journal_get_superblock(journal_t *journal)
1902 struct buffer_head *bh;
1903 journal_superblock_t *sb;
1906 bh = journal->j_sb_buffer;
1908 J_ASSERT(bh != NULL);
1909 err = bh_read(bh, 0);
1912 "JBD2: IO error reading journal superblock\n");
1916 if (buffer_verified(bh))
1919 sb = journal->j_superblock;
1923 if (sb->s_header.h_magic != cpu_to_be32(JBD2_MAGIC_NUMBER) ||
1924 sb->s_blocksize != cpu_to_be32(journal->j_blocksize)) {
1925 printk(KERN_WARNING "JBD2: no valid journal superblock found\n");
1929 switch(be32_to_cpu(sb->s_header.h_blocktype)) {
1930 case JBD2_SUPERBLOCK_V1:
1931 journal->j_format_version = 1;
1933 case JBD2_SUPERBLOCK_V2:
1934 journal->j_format_version = 2;
1937 printk(KERN_WARNING "JBD2: unrecognised superblock format ID\n");
1941 if (be32_to_cpu(sb->s_maxlen) < journal->j_total_len)
1942 journal->j_total_len = be32_to_cpu(sb->s_maxlen);
1943 else if (be32_to_cpu(sb->s_maxlen) > journal->j_total_len) {
1944 printk(KERN_WARNING "JBD2: journal file too short\n");
1948 if (be32_to_cpu(sb->s_first) == 0 ||
1949 be32_to_cpu(sb->s_first) >= journal->j_total_len) {
1951 "JBD2: Invalid start block of journal: %u\n",
1952 be32_to_cpu(sb->s_first));
1956 if (jbd2_has_feature_csum2(journal) &&
1957 jbd2_has_feature_csum3(journal)) {
1958 /* Can't have checksum v2 and v3 at the same time! */
1959 printk(KERN_ERR "JBD2: Can't enable checksumming v2 and v3 "
1960 "at the same time!\n");
1964 if (jbd2_journal_has_csum_v2or3_feature(journal) &&
1965 jbd2_has_feature_checksum(journal)) {
1966 /* Can't have checksum v1 and v2 on at the same time! */
1967 printk(KERN_ERR "JBD2: Can't enable checksumming v1 and v2/3 "
1968 "at the same time!\n");
1972 if (!jbd2_verify_csum_type(journal, sb)) {
1973 printk(KERN_ERR "JBD2: Unknown checksum type\n");
1977 /* Load the checksum driver */
1978 if (jbd2_journal_has_csum_v2or3_feature(journal)) {
1979 journal->j_chksum_driver = crypto_alloc_shash("crc32c", 0, 0);
1980 if (IS_ERR(journal->j_chksum_driver)) {
1981 printk(KERN_ERR "JBD2: Cannot load crc32c driver.\n");
1982 err = PTR_ERR(journal->j_chksum_driver);
1983 journal->j_chksum_driver = NULL;
1988 if (jbd2_journal_has_csum_v2or3(journal)) {
1989 /* Check superblock checksum */
1990 if (sb->s_checksum != jbd2_superblock_csum(journal, sb)) {
1991 printk(KERN_ERR "JBD2: journal checksum error\n");
1996 /* Precompute checksum seed for all metadata */
1997 journal->j_csum_seed = jbd2_chksum(journal, ~0, sb->s_uuid,
1998 sizeof(sb->s_uuid));
2001 journal->j_revoke_records_per_block =
2002 journal_revoke_records_per_block(journal);
2003 set_buffer_verified(bh);
2008 journal_fail_superblock(journal);
2013 * Load the on-disk journal superblock and read the key fields into the
2017 static int load_superblock(journal_t *journal)
2020 journal_superblock_t *sb;
2023 err = journal_get_superblock(journal);
2027 sb = journal->j_superblock;
2029 journal->j_tail_sequence = be32_to_cpu(sb->s_sequence);
2030 journal->j_tail = be32_to_cpu(sb->s_start);
2031 journal->j_first = be32_to_cpu(sb->s_first);
2032 journal->j_errno = be32_to_cpu(sb->s_errno);
2033 journal->j_last = be32_to_cpu(sb->s_maxlen);
2035 if (jbd2_has_feature_fast_commit(journal)) {
2036 journal->j_fc_last = be32_to_cpu(sb->s_maxlen);
2037 num_fc_blocks = jbd2_journal_get_num_fc_blks(sb);
2038 if (journal->j_last - num_fc_blocks >= JBD2_MIN_JOURNAL_BLOCKS)
2039 journal->j_last = journal->j_fc_last - num_fc_blocks;
2040 journal->j_fc_first = journal->j_last + 1;
2041 journal->j_fc_off = 0;
2049 * jbd2_journal_load() - Read journal from disk.
2050 * @journal: Journal to act on.
2052 * Given a journal_t structure which tells us which disk blocks contain
2053 * a journal, read the journal from disk to initialise the in-memory
2056 int jbd2_journal_load(journal_t *journal)
2059 journal_superblock_t *sb;
2061 err = load_superblock(journal);
2065 sb = journal->j_superblock;
2066 /* If this is a V2 superblock, then we have to check the
2067 * features flags on it. */
2069 if (journal->j_format_version >= 2) {
2070 if ((sb->s_feature_ro_compat &
2071 ~cpu_to_be32(JBD2_KNOWN_ROCOMPAT_FEATURES)) ||
2072 (sb->s_feature_incompat &
2073 ~cpu_to_be32(JBD2_KNOWN_INCOMPAT_FEATURES))) {
2075 "JBD2: Unrecognised features on journal\n");
2081 * Create a slab for this blocksize
2083 err = jbd2_journal_create_slab(be32_to_cpu(sb->s_blocksize));
2087 /* Let the recovery code check whether it needs to recover any
2088 * data from the journal. */
2089 if (jbd2_journal_recover(journal))
2090 goto recovery_error;
2092 if (journal->j_failed_commit) {
2093 printk(KERN_ERR "JBD2: journal transaction %u on %s "
2094 "is corrupt.\n", journal->j_failed_commit,
2095 journal->j_devname);
2096 return -EFSCORRUPTED;
2099 * clear JBD2_ABORT flag initialized in journal_init_common
2100 * here to update log tail information with the newest seq.
2102 journal->j_flags &= ~JBD2_ABORT;
2104 /* OK, we've finished with the dynamic journal bits:
2105 * reinitialise the dynamic contents of the superblock in memory
2106 * and reset them on disk. */
2107 if (journal_reset(journal))
2108 goto recovery_error;
2110 journal->j_flags |= JBD2_LOADED;
2114 printk(KERN_WARNING "JBD2: recovery failed\n");
2119 * jbd2_journal_destroy() - Release a journal_t structure.
2120 * @journal: Journal to act on.
2122 * Release a journal_t structure once it is no longer in use by the
2124 * Return <0 if we couldn't clean up the journal.
2126 int jbd2_journal_destroy(journal_t *journal)
2130 /* Wait for the commit thread to wake up and die. */
2131 journal_kill_thread(journal);
2133 /* Force a final log commit */
2134 if (journal->j_running_transaction)
2135 jbd2_journal_commit_transaction(journal);
2137 /* Force any old transactions to disk */
2139 /* Totally anal locking here... */
2140 spin_lock(&journal->j_list_lock);
2141 while (journal->j_checkpoint_transactions != NULL) {
2142 spin_unlock(&journal->j_list_lock);
2143 mutex_lock_io(&journal->j_checkpoint_mutex);
2144 err = jbd2_log_do_checkpoint(journal);
2145 mutex_unlock(&journal->j_checkpoint_mutex);
2147 * If checkpointing failed, just free the buffers to avoid
2151 jbd2_journal_destroy_checkpoint(journal);
2152 spin_lock(&journal->j_list_lock);
2155 spin_lock(&journal->j_list_lock);
2158 J_ASSERT(journal->j_running_transaction == NULL);
2159 J_ASSERT(journal->j_committing_transaction == NULL);
2160 J_ASSERT(journal->j_checkpoint_transactions == NULL);
2161 spin_unlock(&journal->j_list_lock);
2164 * OK, all checkpoint transactions have been checked, now check the
2165 * write out io error flag and abort the journal if some buffer failed
2166 * to write back to the original location, otherwise the filesystem
2167 * may become inconsistent.
2169 if (!is_journal_aborted(journal) &&
2170 test_bit(JBD2_CHECKPOINT_IO_ERROR, &journal->j_atomic_flags))
2171 jbd2_journal_abort(journal, -EIO);
2173 if (journal->j_sb_buffer) {
2174 if (!is_journal_aborted(journal)) {
2175 mutex_lock_io(&journal->j_checkpoint_mutex);
2177 write_lock(&journal->j_state_lock);
2178 journal->j_tail_sequence =
2179 ++journal->j_transaction_sequence;
2180 write_unlock(&journal->j_state_lock);
2182 jbd2_mark_journal_empty(journal,
2183 REQ_SYNC | REQ_PREFLUSH | REQ_FUA);
2184 mutex_unlock(&journal->j_checkpoint_mutex);
2187 brelse(journal->j_sb_buffer);
2190 if (journal->j_shrinker.flags & SHRINKER_REGISTERED) {
2191 percpu_counter_destroy(&journal->j_checkpoint_jh_count);
2192 unregister_shrinker(&journal->j_shrinker);
2194 if (journal->j_proc_entry)
2195 jbd2_stats_proc_exit(journal);
2196 iput(journal->j_inode);
2197 if (journal->j_revoke)
2198 jbd2_journal_destroy_revoke(journal);
2199 if (journal->j_chksum_driver)
2200 crypto_free_shash(journal->j_chksum_driver);
2201 kfree(journal->j_fc_wbuf);
2202 kfree(journal->j_wbuf);
2210 * jbd2_journal_check_used_features() - Check if features specified are used.
2211 * @journal: Journal to check.
2212 * @compat: bitmask of compatible features
2213 * @ro: bitmask of features that force read-only mount
2214 * @incompat: bitmask of incompatible features
2216 * Check whether the journal uses all of a given set of
2217 * features. Return true (non-zero) if it does.
2220 int jbd2_journal_check_used_features(journal_t *journal, unsigned long compat,
2221 unsigned long ro, unsigned long incompat)
2223 journal_superblock_t *sb;
2225 if (!compat && !ro && !incompat)
2227 /* Load journal superblock if it is not loaded yet. */
2228 if (journal->j_format_version == 0 &&
2229 journal_get_superblock(journal) != 0)
2231 if (journal->j_format_version == 1)
2234 sb = journal->j_superblock;
2236 if (((be32_to_cpu(sb->s_feature_compat) & compat) == compat) &&
2237 ((be32_to_cpu(sb->s_feature_ro_compat) & ro) == ro) &&
2238 ((be32_to_cpu(sb->s_feature_incompat) & incompat) == incompat))
2245 * jbd2_journal_check_available_features() - Check feature set in journalling layer
2246 * @journal: Journal to check.
2247 * @compat: bitmask of compatible features
2248 * @ro: bitmask of features that force read-only mount
2249 * @incompat: bitmask of incompatible features
2251 * Check whether the journaling code supports the use of
2252 * all of a given set of features on this journal. Return true
2253 * (non-zero) if it can. */
2255 int jbd2_journal_check_available_features(journal_t *journal, unsigned long compat,
2256 unsigned long ro, unsigned long incompat)
2258 if (!compat && !ro && !incompat)
2261 /* We can support any known requested features iff the
2262 * superblock is in version 2. Otherwise we fail to support any
2263 * extended sb features. */
2265 if (journal->j_format_version != 2)
2268 if ((compat & JBD2_KNOWN_COMPAT_FEATURES) == compat &&
2269 (ro & JBD2_KNOWN_ROCOMPAT_FEATURES) == ro &&
2270 (incompat & JBD2_KNOWN_INCOMPAT_FEATURES) == incompat)
2277 jbd2_journal_initialize_fast_commit(journal_t *journal)
2279 journal_superblock_t *sb = journal->j_superblock;
2280 unsigned long long num_fc_blks;
2282 num_fc_blks = jbd2_journal_get_num_fc_blks(sb);
2283 if (journal->j_last - num_fc_blks < JBD2_MIN_JOURNAL_BLOCKS)
2286 /* Are we called twice? */
2287 WARN_ON(journal->j_fc_wbuf != NULL);
2288 journal->j_fc_wbuf = kmalloc_array(num_fc_blks,
2289 sizeof(struct buffer_head *), GFP_KERNEL);
2290 if (!journal->j_fc_wbuf)
2293 journal->j_fc_wbufsize = num_fc_blks;
2294 journal->j_fc_last = journal->j_last;
2295 journal->j_last = journal->j_fc_last - num_fc_blks;
2296 journal->j_fc_first = journal->j_last + 1;
2297 journal->j_fc_off = 0;
2298 journal->j_free = journal->j_last - journal->j_first;
2299 journal->j_max_transaction_buffers =
2300 jbd2_journal_get_max_txn_bufs(journal);
2306 * jbd2_journal_set_features() - Mark a given journal feature in the superblock
2307 * @journal: Journal to act on.
2308 * @compat: bitmask of compatible features
2309 * @ro: bitmask of features that force read-only mount
2310 * @incompat: bitmask of incompatible features
2312 * Mark a given journal feature as present on the
2313 * superblock. Returns true if the requested features could be set.
2317 int jbd2_journal_set_features(journal_t *journal, unsigned long compat,
2318 unsigned long ro, unsigned long incompat)
2320 #define INCOMPAT_FEATURE_ON(f) \
2321 ((incompat & (f)) && !(sb->s_feature_incompat & cpu_to_be32(f)))
2322 #define COMPAT_FEATURE_ON(f) \
2323 ((compat & (f)) && !(sb->s_feature_compat & cpu_to_be32(f)))
2324 journal_superblock_t *sb;
2326 if (jbd2_journal_check_used_features(journal, compat, ro, incompat))
2329 if (!jbd2_journal_check_available_features(journal, compat, ro, incompat))
2332 /* If enabling v2 checksums, turn on v3 instead */
2333 if (incompat & JBD2_FEATURE_INCOMPAT_CSUM_V2) {
2334 incompat &= ~JBD2_FEATURE_INCOMPAT_CSUM_V2;
2335 incompat |= JBD2_FEATURE_INCOMPAT_CSUM_V3;
2338 /* Asking for checksumming v3 and v1? Only give them v3. */
2339 if (incompat & JBD2_FEATURE_INCOMPAT_CSUM_V3 &&
2340 compat & JBD2_FEATURE_COMPAT_CHECKSUM)
2341 compat &= ~JBD2_FEATURE_COMPAT_CHECKSUM;
2343 jbd2_debug(1, "Setting new features 0x%lx/0x%lx/0x%lx\n",
2344 compat, ro, incompat);
2346 sb = journal->j_superblock;
2348 if (incompat & JBD2_FEATURE_INCOMPAT_FAST_COMMIT) {
2349 if (jbd2_journal_initialize_fast_commit(journal)) {
2350 pr_err("JBD2: Cannot enable fast commits.\n");
2355 /* Load the checksum driver if necessary */
2356 if ((journal->j_chksum_driver == NULL) &&
2357 INCOMPAT_FEATURE_ON(JBD2_FEATURE_INCOMPAT_CSUM_V3)) {
2358 journal->j_chksum_driver = crypto_alloc_shash("crc32c", 0, 0);
2359 if (IS_ERR(journal->j_chksum_driver)) {
2360 printk(KERN_ERR "JBD2: Cannot load crc32c driver.\n");
2361 journal->j_chksum_driver = NULL;
2364 /* Precompute checksum seed for all metadata */
2365 journal->j_csum_seed = jbd2_chksum(journal, ~0, sb->s_uuid,
2366 sizeof(sb->s_uuid));
2369 lock_buffer(journal->j_sb_buffer);
2371 /* If enabling v3 checksums, update superblock */
2372 if (INCOMPAT_FEATURE_ON(JBD2_FEATURE_INCOMPAT_CSUM_V3)) {
2373 sb->s_checksum_type = JBD2_CRC32C_CHKSUM;
2374 sb->s_feature_compat &=
2375 ~cpu_to_be32(JBD2_FEATURE_COMPAT_CHECKSUM);
2378 /* If enabling v1 checksums, downgrade superblock */
2379 if (COMPAT_FEATURE_ON(JBD2_FEATURE_COMPAT_CHECKSUM))
2380 sb->s_feature_incompat &=
2381 ~cpu_to_be32(JBD2_FEATURE_INCOMPAT_CSUM_V2 |
2382 JBD2_FEATURE_INCOMPAT_CSUM_V3);
2384 sb->s_feature_compat |= cpu_to_be32(compat);
2385 sb->s_feature_ro_compat |= cpu_to_be32(ro);
2386 sb->s_feature_incompat |= cpu_to_be32(incompat);
2387 unlock_buffer(journal->j_sb_buffer);
2388 journal->j_revoke_records_per_block =
2389 journal_revoke_records_per_block(journal);
2392 #undef COMPAT_FEATURE_ON
2393 #undef INCOMPAT_FEATURE_ON
2397 * jbd2_journal_clear_features() - Clear a given journal feature in the
2399 * @journal: Journal to act on.
2400 * @compat: bitmask of compatible features
2401 * @ro: bitmask of features that force read-only mount
2402 * @incompat: bitmask of incompatible features
2404 * Clear a given journal feature as present on the
2407 void jbd2_journal_clear_features(journal_t *journal, unsigned long compat,
2408 unsigned long ro, unsigned long incompat)
2410 journal_superblock_t *sb;
2412 jbd2_debug(1, "Clear features 0x%lx/0x%lx/0x%lx\n",
2413 compat, ro, incompat);
2415 sb = journal->j_superblock;
2417 sb->s_feature_compat &= ~cpu_to_be32(compat);
2418 sb->s_feature_ro_compat &= ~cpu_to_be32(ro);
2419 sb->s_feature_incompat &= ~cpu_to_be32(incompat);
2420 journal->j_revoke_records_per_block =
2421 journal_revoke_records_per_block(journal);
2423 EXPORT_SYMBOL(jbd2_journal_clear_features);
2426 * jbd2_journal_flush() - Flush journal
2427 * @journal: Journal to act on.
2428 * @flags: optional operation on the journal blocks after the flush (see below)
2430 * Flush all data for a given journal to disk and empty the journal.
2431 * Filesystems can use this when remounting readonly to ensure that
2432 * recovery does not need to happen on remount. Optionally, a discard or zeroout
2433 * can be issued on the journal blocks after flushing.
2436 * JBD2_JOURNAL_FLUSH_DISCARD: issues discards for the journal blocks
2437 * JBD2_JOURNAL_FLUSH_ZEROOUT: issues zeroouts for the journal blocks
2439 int jbd2_journal_flush(journal_t *journal, unsigned int flags)
2442 transaction_t *transaction = NULL;
2444 write_lock(&journal->j_state_lock);
2446 /* Force everything buffered to the log... */
2447 if (journal->j_running_transaction) {
2448 transaction = journal->j_running_transaction;
2449 __jbd2_log_start_commit(journal, transaction->t_tid);
2450 } else if (journal->j_committing_transaction)
2451 transaction = journal->j_committing_transaction;
2453 /* Wait for the log commit to complete... */
2455 tid_t tid = transaction->t_tid;
2457 write_unlock(&journal->j_state_lock);
2458 jbd2_log_wait_commit(journal, tid);
2460 write_unlock(&journal->j_state_lock);
2463 /* ...and flush everything in the log out to disk. */
2464 spin_lock(&journal->j_list_lock);
2465 while (!err && journal->j_checkpoint_transactions != NULL) {
2466 spin_unlock(&journal->j_list_lock);
2467 mutex_lock_io(&journal->j_checkpoint_mutex);
2468 err = jbd2_log_do_checkpoint(journal);
2469 mutex_unlock(&journal->j_checkpoint_mutex);
2470 spin_lock(&journal->j_list_lock);
2472 spin_unlock(&journal->j_list_lock);
2474 if (is_journal_aborted(journal))
2477 mutex_lock_io(&journal->j_checkpoint_mutex);
2479 err = jbd2_cleanup_journal_tail(journal);
2481 mutex_unlock(&journal->j_checkpoint_mutex);
2487 /* Finally, mark the journal as really needing no recovery.
2488 * This sets s_start==0 in the underlying superblock, which is
2489 * the magic code for a fully-recovered superblock. Any future
2490 * commits of data to the journal will restore the current
2492 jbd2_mark_journal_empty(journal, REQ_SYNC | REQ_FUA);
2495 err = __jbd2_journal_erase(journal, flags);
2497 mutex_unlock(&journal->j_checkpoint_mutex);
2498 write_lock(&journal->j_state_lock);
2499 J_ASSERT(!journal->j_running_transaction);
2500 J_ASSERT(!journal->j_committing_transaction);
2501 J_ASSERT(!journal->j_checkpoint_transactions);
2502 J_ASSERT(journal->j_head == journal->j_tail);
2503 J_ASSERT(journal->j_tail_sequence == journal->j_transaction_sequence);
2504 write_unlock(&journal->j_state_lock);
2510 * jbd2_journal_wipe() - Wipe journal contents
2511 * @journal: Journal to act on.
2512 * @write: flag (see below)
2514 * Wipe out all of the contents of a journal, safely. This will produce
2515 * a warning if the journal contains any valid recovery information.
2516 * Must be called between journal_init_*() and jbd2_journal_load().
2518 * If 'write' is non-zero, then we wipe out the journal on disk; otherwise
2519 * we merely suppress recovery.
2522 int jbd2_journal_wipe(journal_t *journal, int write)
2526 J_ASSERT (!(journal->j_flags & JBD2_LOADED));
2528 err = load_superblock(journal);
2532 if (!journal->j_tail)
2535 printk(KERN_WARNING "JBD2: %s recovery information on journal\n",
2536 write ? "Clearing" : "Ignoring");
2538 err = jbd2_journal_skip_recovery(journal);
2540 /* Lock to make assertions happy... */
2541 mutex_lock_io(&journal->j_checkpoint_mutex);
2542 jbd2_mark_journal_empty(journal, REQ_SYNC | REQ_FUA);
2543 mutex_unlock(&journal->j_checkpoint_mutex);
2551 * jbd2_journal_abort () - Shutdown the journal immediately.
2552 * @journal: the journal to shutdown.
2553 * @errno: an error number to record in the journal indicating
2554 * the reason for the shutdown.
2556 * Perform a complete, immediate shutdown of the ENTIRE
2557 * journal (not of a single transaction). This operation cannot be
2558 * undone without closing and reopening the journal.
2560 * The jbd2_journal_abort function is intended to support higher level error
2561 * recovery mechanisms such as the ext2/ext3 remount-readonly error
2564 * Journal abort has very specific semantics. Any existing dirty,
2565 * unjournaled buffers in the main filesystem will still be written to
2566 * disk by bdflush, but the journaling mechanism will be suspended
2567 * immediately and no further transaction commits will be honoured.
2569 * Any dirty, journaled buffers will be written back to disk without
2570 * hitting the journal. Atomicity cannot be guaranteed on an aborted
2571 * filesystem, but we _do_ attempt to leave as much data as possible
2572 * behind for fsck to use for cleanup.
2574 * Any attempt to get a new transaction handle on a journal which is in
2575 * ABORT state will just result in an -EROFS error return. A
2576 * jbd2_journal_stop on an existing handle will return -EIO if we have
2577 * entered abort state during the update.
2579 * Recursive transactions are not disturbed by journal abort until the
2580 * final jbd2_journal_stop, which will receive the -EIO error.
2582 * Finally, the jbd2_journal_abort call allows the caller to supply an errno
2583 * which will be recorded (if possible) in the journal superblock. This
2584 * allows a client to record failure conditions in the middle of a
2585 * transaction without having to complete the transaction to record the
2586 * failure to disk. ext3_error, for example, now uses this
2591 void jbd2_journal_abort(journal_t *journal, int errno)
2593 transaction_t *transaction;
2596 * Lock the aborting procedure until everything is done, this avoid
2597 * races between filesystem's error handling flow (e.g. ext4_abort()),
2598 * ensure panic after the error info is written into journal's
2601 mutex_lock(&journal->j_abort_mutex);
2603 * ESHUTDOWN always takes precedence because a file system check
2604 * caused by any other journal abort error is not required after
2605 * a shutdown triggered.
2607 write_lock(&journal->j_state_lock);
2608 if (journal->j_flags & JBD2_ABORT) {
2609 int old_errno = journal->j_errno;
2611 write_unlock(&journal->j_state_lock);
2612 if (old_errno != -ESHUTDOWN && errno == -ESHUTDOWN) {
2613 journal->j_errno = errno;
2614 jbd2_journal_update_sb_errno(journal);
2616 mutex_unlock(&journal->j_abort_mutex);
2621 * Mark the abort as occurred and start current running transaction
2622 * to release all journaled buffer.
2624 pr_err("Aborting journal on device %s.\n", journal->j_devname);
2626 journal->j_flags |= JBD2_ABORT;
2627 journal->j_errno = errno;
2628 transaction = journal->j_running_transaction;
2630 __jbd2_log_start_commit(journal, transaction->t_tid);
2631 write_unlock(&journal->j_state_lock);
2634 * Record errno to the journal super block, so that fsck and jbd2
2635 * layer could realise that a filesystem check is needed.
2637 jbd2_journal_update_sb_errno(journal);
2638 mutex_unlock(&journal->j_abort_mutex);
2642 * jbd2_journal_errno() - returns the journal's error state.
2643 * @journal: journal to examine.
2645 * This is the errno number set with jbd2_journal_abort(), the last
2646 * time the journal was mounted - if the journal was stopped
2647 * without calling abort this will be 0.
2649 * If the journal has been aborted on this mount time -EROFS will
2652 int jbd2_journal_errno(journal_t *journal)
2656 read_lock(&journal->j_state_lock);
2657 if (journal->j_flags & JBD2_ABORT)
2660 err = journal->j_errno;
2661 read_unlock(&journal->j_state_lock);
2666 * jbd2_journal_clear_err() - clears the journal's error state
2667 * @journal: journal to act on.
2669 * An error must be cleared or acked to take a FS out of readonly
2672 int jbd2_journal_clear_err(journal_t *journal)
2676 write_lock(&journal->j_state_lock);
2677 if (journal->j_flags & JBD2_ABORT)
2680 journal->j_errno = 0;
2681 write_unlock(&journal->j_state_lock);
2686 * jbd2_journal_ack_err() - Ack journal err.
2687 * @journal: journal to act on.
2689 * An error must be cleared or acked to take a FS out of readonly
2692 void jbd2_journal_ack_err(journal_t *journal)
2694 write_lock(&journal->j_state_lock);
2695 if (journal->j_errno)
2696 journal->j_flags |= JBD2_ACK_ERR;
2697 write_unlock(&journal->j_state_lock);
2700 int jbd2_journal_blocks_per_page(struct inode *inode)
2702 return 1 << (PAGE_SHIFT - inode->i_sb->s_blocksize_bits);
2706 * helper functions to deal with 32 or 64bit block numbers.
2708 size_t journal_tag_bytes(journal_t *journal)
2712 if (jbd2_has_feature_csum3(journal))
2713 return sizeof(journal_block_tag3_t);
2715 sz = sizeof(journal_block_tag_t);
2717 if (jbd2_has_feature_csum2(journal))
2718 sz += sizeof(__u16);
2720 if (jbd2_has_feature_64bit(journal))
2723 return sz - sizeof(__u32);
2727 * JBD memory management
2729 * These functions are used to allocate block-sized chunks of memory
2730 * used for making copies of buffer_head data. Very often it will be
2731 * page-sized chunks of data, but sometimes it will be in
2732 * sub-page-size chunks. (For example, 16k pages on Power systems
2733 * with a 4k block file system.) For blocks smaller than a page, we
2734 * use a SLAB allocator. There are slab caches for each block size,
2735 * which are allocated at mount time, if necessary, and we only free
2736 * (all of) the slab caches when/if the jbd2 module is unloaded. For
2737 * this reason we don't need to a mutex to protect access to
2738 * jbd2_slab[] allocating or releasing memory; only in
2739 * jbd2_journal_create_slab().
2741 #define JBD2_MAX_SLABS 8
2742 static struct kmem_cache *jbd2_slab[JBD2_MAX_SLABS];
2744 static const char *jbd2_slab_names[JBD2_MAX_SLABS] = {
2745 "jbd2_1k", "jbd2_2k", "jbd2_4k", "jbd2_8k",
2746 "jbd2_16k", "jbd2_32k", "jbd2_64k", "jbd2_128k"
2750 static void jbd2_journal_destroy_slabs(void)
2754 for (i = 0; i < JBD2_MAX_SLABS; i++) {
2755 kmem_cache_destroy(jbd2_slab[i]);
2756 jbd2_slab[i] = NULL;
2760 static int jbd2_journal_create_slab(size_t size)
2762 static DEFINE_MUTEX(jbd2_slab_create_mutex);
2763 int i = order_base_2(size) - 10;
2766 if (size == PAGE_SIZE)
2769 if (i >= JBD2_MAX_SLABS)
2772 if (unlikely(i < 0))
2774 mutex_lock(&jbd2_slab_create_mutex);
2776 mutex_unlock(&jbd2_slab_create_mutex);
2777 return 0; /* Already created */
2780 slab_size = 1 << (i+10);
2781 jbd2_slab[i] = kmem_cache_create(jbd2_slab_names[i], slab_size,
2782 slab_size, 0, NULL);
2783 mutex_unlock(&jbd2_slab_create_mutex);
2784 if (!jbd2_slab[i]) {
2785 printk(KERN_EMERG "JBD2: no memory for jbd2_slab cache\n");
2791 static struct kmem_cache *get_slab(size_t size)
2793 int i = order_base_2(size) - 10;
2795 BUG_ON(i >= JBD2_MAX_SLABS);
2796 if (unlikely(i < 0))
2798 BUG_ON(jbd2_slab[i] == NULL);
2799 return jbd2_slab[i];
2802 void *jbd2_alloc(size_t size, gfp_t flags)
2806 BUG_ON(size & (size-1)); /* Must be a power of 2 */
2808 if (size < PAGE_SIZE)
2809 ptr = kmem_cache_alloc(get_slab(size), flags);
2811 ptr = (void *)__get_free_pages(flags, get_order(size));
2813 /* Check alignment; SLUB has gotten this wrong in the past,
2814 * and this can lead to user data corruption! */
2815 BUG_ON(((unsigned long) ptr) & (size-1));
2820 void jbd2_free(void *ptr, size_t size)
2822 if (size < PAGE_SIZE)
2823 kmem_cache_free(get_slab(size), ptr);
2825 free_pages((unsigned long)ptr, get_order(size));
2829 * Journal_head storage management
2831 static struct kmem_cache *jbd2_journal_head_cache;
2832 #ifdef CONFIG_JBD2_DEBUG
2833 static atomic_t nr_journal_heads = ATOMIC_INIT(0);
2836 static int __init jbd2_journal_init_journal_head_cache(void)
2838 J_ASSERT(!jbd2_journal_head_cache);
2839 jbd2_journal_head_cache = kmem_cache_create("jbd2_journal_head",
2840 sizeof(struct journal_head),
2842 SLAB_TEMPORARY | SLAB_TYPESAFE_BY_RCU,
2844 if (!jbd2_journal_head_cache) {
2845 printk(KERN_EMERG "JBD2: no memory for journal_head cache\n");
2851 static void jbd2_journal_destroy_journal_head_cache(void)
2853 kmem_cache_destroy(jbd2_journal_head_cache);
2854 jbd2_journal_head_cache = NULL;
2858 * journal_head splicing and dicing
2860 static struct journal_head *journal_alloc_journal_head(void)
2862 struct journal_head *ret;
2864 #ifdef CONFIG_JBD2_DEBUG
2865 atomic_inc(&nr_journal_heads);
2867 ret = kmem_cache_zalloc(jbd2_journal_head_cache, GFP_NOFS);
2869 jbd2_debug(1, "out of memory for journal_head\n");
2870 pr_notice_ratelimited("ENOMEM in %s, retrying.\n", __func__);
2871 ret = kmem_cache_zalloc(jbd2_journal_head_cache,
2872 GFP_NOFS | __GFP_NOFAIL);
2875 spin_lock_init(&ret->b_state_lock);
2879 static void journal_free_journal_head(struct journal_head *jh)
2881 #ifdef CONFIG_JBD2_DEBUG
2882 atomic_dec(&nr_journal_heads);
2883 memset(jh, JBD2_POISON_FREE, sizeof(*jh));
2885 kmem_cache_free(jbd2_journal_head_cache, jh);
2889 * A journal_head is attached to a buffer_head whenever JBD has an
2890 * interest in the buffer.
2892 * Whenever a buffer has an attached journal_head, its ->b_state:BH_JBD bit
2893 * is set. This bit is tested in core kernel code where we need to take
2894 * JBD-specific actions. Testing the zeroness of ->b_private is not reliable
2897 * When a buffer has its BH_JBD bit set, its ->b_count is elevated by one.
2899 * When a buffer has its BH_JBD bit set it is immune from being released by
2900 * core kernel code, mainly via ->b_count.
2902 * A journal_head is detached from its buffer_head when the journal_head's
2903 * b_jcount reaches zero. Running transaction (b_transaction) and checkpoint
2904 * transaction (b_cp_transaction) hold their references to b_jcount.
2906 * Various places in the kernel want to attach a journal_head to a buffer_head
2907 * _before_ attaching the journal_head to a transaction. To protect the
2908 * journal_head in this situation, jbd2_journal_add_journal_head elevates the
2909 * journal_head's b_jcount refcount by one. The caller must call
2910 * jbd2_journal_put_journal_head() to undo this.
2912 * So the typical usage would be:
2914 * (Attach a journal_head if needed. Increments b_jcount)
2915 * struct journal_head *jh = jbd2_journal_add_journal_head(bh);
2917 * (Get another reference for transaction)
2918 * jbd2_journal_grab_journal_head(bh);
2919 * jh->b_transaction = xxx;
2920 * (Put original reference)
2921 * jbd2_journal_put_journal_head(jh);
2925 * Give a buffer_head a journal_head.
2929 struct journal_head *jbd2_journal_add_journal_head(struct buffer_head *bh)
2931 struct journal_head *jh;
2932 struct journal_head *new_jh = NULL;
2935 if (!buffer_jbd(bh))
2936 new_jh = journal_alloc_journal_head();
2938 jbd_lock_bh_journal_head(bh);
2939 if (buffer_jbd(bh)) {
2943 (atomic_read(&bh->b_count) > 0) ||
2944 (bh->b_page && bh->b_page->mapping));
2947 jbd_unlock_bh_journal_head(bh);
2952 new_jh = NULL; /* We consumed it */
2957 BUFFER_TRACE(bh, "added journal_head");
2960 jbd_unlock_bh_journal_head(bh);
2962 journal_free_journal_head(new_jh);
2963 return bh->b_private;
2967 * Grab a ref against this buffer_head's journal_head. If it ended up not
2968 * having a journal_head, return NULL
2970 struct journal_head *jbd2_journal_grab_journal_head(struct buffer_head *bh)
2972 struct journal_head *jh = NULL;
2974 jbd_lock_bh_journal_head(bh);
2975 if (buffer_jbd(bh)) {
2979 jbd_unlock_bh_journal_head(bh);
2982 EXPORT_SYMBOL(jbd2_journal_grab_journal_head);
2984 static void __journal_remove_journal_head(struct buffer_head *bh)
2986 struct journal_head *jh = bh2jh(bh);
2988 J_ASSERT_JH(jh, jh->b_transaction == NULL);
2989 J_ASSERT_JH(jh, jh->b_next_transaction == NULL);
2990 J_ASSERT_JH(jh, jh->b_cp_transaction == NULL);
2991 J_ASSERT_JH(jh, jh->b_jlist == BJ_None);
2992 J_ASSERT_BH(bh, buffer_jbd(bh));
2993 J_ASSERT_BH(bh, jh2bh(jh) == bh);
2994 BUFFER_TRACE(bh, "remove journal_head");
2996 /* Unlink before dropping the lock */
2997 bh->b_private = NULL;
2998 jh->b_bh = NULL; /* debug, really */
2999 clear_buffer_jbd(bh);
3002 static void journal_release_journal_head(struct journal_head *jh, size_t b_size)
3004 if (jh->b_frozen_data) {
3005 printk(KERN_WARNING "%s: freeing b_frozen_data\n", __func__);
3006 jbd2_free(jh->b_frozen_data, b_size);
3008 if (jh->b_committed_data) {
3009 printk(KERN_WARNING "%s: freeing b_committed_data\n", __func__);
3010 jbd2_free(jh->b_committed_data, b_size);
3012 journal_free_journal_head(jh);
3016 * Drop a reference on the passed journal_head. If it fell to zero then
3017 * release the journal_head from the buffer_head.
3019 void jbd2_journal_put_journal_head(struct journal_head *jh)
3021 struct buffer_head *bh = jh2bh(jh);
3023 jbd_lock_bh_journal_head(bh);
3024 J_ASSERT_JH(jh, jh->b_jcount > 0);
3026 if (!jh->b_jcount) {
3027 __journal_remove_journal_head(bh);
3028 jbd_unlock_bh_journal_head(bh);
3029 journal_release_journal_head(jh, bh->b_size);
3032 jbd_unlock_bh_journal_head(bh);
3035 EXPORT_SYMBOL(jbd2_journal_put_journal_head);
3038 * Initialize jbd inode head
3040 void jbd2_journal_init_jbd_inode(struct jbd2_inode *jinode, struct inode *inode)
3042 jinode->i_transaction = NULL;
3043 jinode->i_next_transaction = NULL;
3044 jinode->i_vfs_inode = inode;
3045 jinode->i_flags = 0;
3046 jinode->i_dirty_start = 0;
3047 jinode->i_dirty_end = 0;
3048 INIT_LIST_HEAD(&jinode->i_list);
3052 * Function to be called before we start removing inode from memory (i.e.,
3053 * clear_inode() is a fine place to be called from). It removes inode from
3054 * transaction's lists.
3056 void jbd2_journal_release_jbd_inode(journal_t *journal,
3057 struct jbd2_inode *jinode)
3062 spin_lock(&journal->j_list_lock);
3063 /* Is commit writing out inode - we have to wait */
3064 if (jinode->i_flags & JI_COMMIT_RUNNING) {
3065 wait_queue_head_t *wq;
3066 DEFINE_WAIT_BIT(wait, &jinode->i_flags, __JI_COMMIT_RUNNING);
3067 wq = bit_waitqueue(&jinode->i_flags, __JI_COMMIT_RUNNING);
3068 prepare_to_wait(wq, &wait.wq_entry, TASK_UNINTERRUPTIBLE);
3069 spin_unlock(&journal->j_list_lock);
3071 finish_wait(wq, &wait.wq_entry);
3075 if (jinode->i_transaction) {
3076 list_del(&jinode->i_list);
3077 jinode->i_transaction = NULL;
3079 spin_unlock(&journal->j_list_lock);
3083 #ifdef CONFIG_PROC_FS
3085 #define JBD2_STATS_PROC_NAME "fs/jbd2"
3087 static void __init jbd2_create_jbd_stats_proc_entry(void)
3089 proc_jbd2_stats = proc_mkdir(JBD2_STATS_PROC_NAME, NULL);
3092 static void __exit jbd2_remove_jbd_stats_proc_entry(void)
3094 if (proc_jbd2_stats)
3095 remove_proc_entry(JBD2_STATS_PROC_NAME, NULL);
3100 #define jbd2_create_jbd_stats_proc_entry() do {} while (0)
3101 #define jbd2_remove_jbd_stats_proc_entry() do {} while (0)
3105 struct kmem_cache *jbd2_handle_cache, *jbd2_inode_cache;
3107 static int __init jbd2_journal_init_inode_cache(void)
3109 J_ASSERT(!jbd2_inode_cache);
3110 jbd2_inode_cache = KMEM_CACHE(jbd2_inode, 0);
3111 if (!jbd2_inode_cache) {
3112 pr_emerg("JBD2: failed to create inode cache\n");
3118 static int __init jbd2_journal_init_handle_cache(void)
3120 J_ASSERT(!jbd2_handle_cache);
3121 jbd2_handle_cache = KMEM_CACHE(jbd2_journal_handle, SLAB_TEMPORARY);
3122 if (!jbd2_handle_cache) {
3123 printk(KERN_EMERG "JBD2: failed to create handle cache\n");
3129 static void jbd2_journal_destroy_inode_cache(void)
3131 kmem_cache_destroy(jbd2_inode_cache);
3132 jbd2_inode_cache = NULL;
3135 static void jbd2_journal_destroy_handle_cache(void)
3137 kmem_cache_destroy(jbd2_handle_cache);
3138 jbd2_handle_cache = NULL;
3142 * Module startup and shutdown
3145 static int __init journal_init_caches(void)
3149 ret = jbd2_journal_init_revoke_record_cache();
3151 ret = jbd2_journal_init_revoke_table_cache();
3153 ret = jbd2_journal_init_journal_head_cache();
3155 ret = jbd2_journal_init_handle_cache();
3157 ret = jbd2_journal_init_inode_cache();
3159 ret = jbd2_journal_init_transaction_cache();
3163 static void jbd2_journal_destroy_caches(void)
3165 jbd2_journal_destroy_revoke_record_cache();
3166 jbd2_journal_destroy_revoke_table_cache();
3167 jbd2_journal_destroy_journal_head_cache();
3168 jbd2_journal_destroy_handle_cache();
3169 jbd2_journal_destroy_inode_cache();
3170 jbd2_journal_destroy_transaction_cache();
3171 jbd2_journal_destroy_slabs();
3174 static int __init journal_init(void)
3178 BUILD_BUG_ON(sizeof(struct journal_superblock_s) != 1024);
3180 ret = journal_init_caches();
3182 jbd2_create_jbd_stats_proc_entry();
3184 jbd2_journal_destroy_caches();
3189 static void __exit journal_exit(void)
3191 #ifdef CONFIG_JBD2_DEBUG
3192 int n = atomic_read(&nr_journal_heads);
3194 printk(KERN_ERR "JBD2: leaked %d journal_heads!\n", n);
3196 jbd2_remove_jbd_stats_proc_entry();
3197 jbd2_journal_destroy_caches();
3200 MODULE_LICENSE("GPL");
3201 module_init(journal_init);
3202 module_exit(journal_exit);