2 * linux/fs/jbd2/journal.c
4 * Written by Stephen C. Tweedie <sct@redhat.com>, 1998
6 * Copyright 1998 Red Hat corp --- All Rights Reserved
8 * This file is part of the Linux kernel and is made available under
9 * the terms of the GNU General Public License, version 2, or at your
10 * option, any later version, incorporated herein by reference.
12 * Generic filesystem journal-writing code; part of the ext2fs
15 * This file manages journals: areas of disk reserved for logging
16 * transactional updates. This includes the kernel journaling thread
17 * which is responsible for scheduling updates to the log.
19 * We do not actually manage the physical storage of the journal in this
20 * file: that is left to a per-journal policy function, which allows us
21 * to store the journal within a filesystem-specified area for ext2
22 * journaling (ext2 can use a reserved inode for storing the log).
25 #include <linux/module.h>
26 #include <linux/time.h>
28 #include <linux/jbd2.h>
29 #include <linux/errno.h>
30 #include <linux/slab.h>
31 #include <linux/init.h>
33 #include <linux/freezer.h>
34 #include <linux/pagemap.h>
35 #include <linux/kthread.h>
36 #include <linux/poison.h>
37 #include <linux/proc_fs.h>
38 #include <linux/seq_file.h>
39 #include <linux/math64.h>
40 #include <linux/hash.h>
41 #include <linux/log2.h>
42 #include <linux/vmalloc.h>
43 #include <linux/backing-dev.h>
44 #include <linux/bitops.h>
45 #include <linux/ratelimit.h>
47 #define CREATE_TRACE_POINTS
48 #include <trace/events/jbd2.h>
50 #include <asm/uaccess.h>
53 #ifdef CONFIG_JBD2_DEBUG
54 ushort jbd2_journal_enable_debug __read_mostly;
55 EXPORT_SYMBOL(jbd2_journal_enable_debug);
57 module_param_named(jbd2_debug, jbd2_journal_enable_debug, ushort, 0644);
58 MODULE_PARM_DESC(jbd2_debug, "Debugging level for jbd2");
61 EXPORT_SYMBOL(jbd2_journal_extend);
62 EXPORT_SYMBOL(jbd2_journal_stop);
63 EXPORT_SYMBOL(jbd2_journal_lock_updates);
64 EXPORT_SYMBOL(jbd2_journal_unlock_updates);
65 EXPORT_SYMBOL(jbd2_journal_get_write_access);
66 EXPORT_SYMBOL(jbd2_journal_get_create_access);
67 EXPORT_SYMBOL(jbd2_journal_get_undo_access);
68 EXPORT_SYMBOL(jbd2_journal_set_triggers);
69 EXPORT_SYMBOL(jbd2_journal_dirty_metadata);
70 EXPORT_SYMBOL(jbd2_journal_forget);
72 EXPORT_SYMBOL(journal_sync_buffer);
74 EXPORT_SYMBOL(jbd2_journal_flush);
75 EXPORT_SYMBOL(jbd2_journal_revoke);
77 EXPORT_SYMBOL(jbd2_journal_init_dev);
78 EXPORT_SYMBOL(jbd2_journal_init_inode);
79 EXPORT_SYMBOL(jbd2_journal_check_used_features);
80 EXPORT_SYMBOL(jbd2_journal_check_available_features);
81 EXPORT_SYMBOL(jbd2_journal_set_features);
82 EXPORT_SYMBOL(jbd2_journal_load);
83 EXPORT_SYMBOL(jbd2_journal_destroy);
84 EXPORT_SYMBOL(jbd2_journal_abort);
85 EXPORT_SYMBOL(jbd2_journal_errno);
86 EXPORT_SYMBOL(jbd2_journal_ack_err);
87 EXPORT_SYMBOL(jbd2_journal_clear_err);
88 EXPORT_SYMBOL(jbd2_log_wait_commit);
89 EXPORT_SYMBOL(jbd2_log_start_commit);
90 EXPORT_SYMBOL(jbd2_journal_start_commit);
91 EXPORT_SYMBOL(jbd2_journal_force_commit_nested);
92 EXPORT_SYMBOL(jbd2_journal_wipe);
93 EXPORT_SYMBOL(jbd2_journal_blocks_per_page);
94 EXPORT_SYMBOL(jbd2_journal_invalidatepage);
95 EXPORT_SYMBOL(jbd2_journal_try_to_free_buffers);
96 EXPORT_SYMBOL(jbd2_journal_force_commit);
97 EXPORT_SYMBOL(jbd2_journal_file_inode);
98 EXPORT_SYMBOL(jbd2_journal_init_jbd_inode);
99 EXPORT_SYMBOL(jbd2_journal_release_jbd_inode);
100 EXPORT_SYMBOL(jbd2_journal_begin_ordered_truncate);
101 EXPORT_SYMBOL(jbd2_inode_cache);
103 static void __journal_abort_soft (journal_t *journal, int errno);
104 static int jbd2_journal_create_slab(size_t slab_size);
106 #ifdef CONFIG_JBD2_DEBUG
107 void __jbd2_debug(int level, const char *file, const char *func,
108 unsigned int line, const char *fmt, ...)
110 struct va_format vaf;
113 if (level > jbd2_journal_enable_debug)
118 printk(KERN_DEBUG "%s: (%s, %u): %pV\n", file, func, line, &vaf);
121 EXPORT_SYMBOL(__jbd2_debug);
124 /* Checksumming functions */
125 static int jbd2_verify_csum_type(journal_t *j, journal_superblock_t *sb)
127 if (!jbd2_journal_has_csum_v2or3(j))
130 return sb->s_checksum_type == JBD2_CRC32C_CHKSUM;
133 static __be32 jbd2_superblock_csum(journal_t *j, journal_superblock_t *sb)
138 old_csum = sb->s_checksum;
140 csum = jbd2_chksum(j, ~0, (char *)sb, sizeof(journal_superblock_t));
141 sb->s_checksum = old_csum;
143 return cpu_to_be32(csum);
146 static int jbd2_superblock_csum_verify(journal_t *j, journal_superblock_t *sb)
148 if (!jbd2_journal_has_csum_v2or3(j))
151 return sb->s_checksum == jbd2_superblock_csum(j, sb);
154 static void jbd2_superblock_csum_set(journal_t *j, journal_superblock_t *sb)
156 if (!jbd2_journal_has_csum_v2or3(j))
159 sb->s_checksum = jbd2_superblock_csum(j, sb);
163 * Helper function used to manage commit timeouts
166 static void commit_timeout(unsigned long __data)
168 struct task_struct * p = (struct task_struct *) __data;
174 * kjournald2: The main thread function used to manage a logging device
177 * This kernel thread is responsible for two things:
179 * 1) COMMIT: Every so often we need to commit the current state of the
180 * filesystem to disk. The journal thread is responsible for writing
181 * all of the metadata buffers to disk.
183 * 2) CHECKPOINT: We cannot reuse a used section of the log file until all
184 * of the data in that part of the log has been rewritten elsewhere on
185 * the disk. Flushing these old buffers to reclaim space in the log is
186 * known as checkpointing, and this thread is responsible for that job.
189 static int kjournald2(void *arg)
191 journal_t *journal = arg;
192 transaction_t *transaction;
195 * Set up an interval timer which can be used to trigger a commit wakeup
196 * after the commit interval expires
198 setup_timer(&journal->j_commit_timer, commit_timeout,
199 (unsigned long)current);
203 /* Record that the journal thread is running */
204 journal->j_task = current;
205 wake_up(&journal->j_wait_done_commit);
208 * And now, wait forever for commit wakeup events.
210 write_lock(&journal->j_state_lock);
213 if (journal->j_flags & JBD2_UNMOUNT)
216 jbd_debug(1, "commit_sequence=%d, commit_request=%d\n",
217 journal->j_commit_sequence, journal->j_commit_request);
219 if (journal->j_commit_sequence != journal->j_commit_request) {
220 jbd_debug(1, "OK, requests differ\n");
221 write_unlock(&journal->j_state_lock);
222 del_timer_sync(&journal->j_commit_timer);
223 jbd2_journal_commit_transaction(journal);
224 write_lock(&journal->j_state_lock);
228 wake_up(&journal->j_wait_done_commit);
229 if (freezing(current)) {
231 * The simpler the better. Flushing journal isn't a
232 * good idea, because that depends on threads that may
233 * be already stopped.
235 jbd_debug(1, "Now suspending kjournald2\n");
236 write_unlock(&journal->j_state_lock);
238 write_lock(&journal->j_state_lock);
241 * We assume on resume that commits are already there,
245 int should_sleep = 1;
247 prepare_to_wait(&journal->j_wait_commit, &wait,
249 if (journal->j_commit_sequence != journal->j_commit_request)
251 transaction = journal->j_running_transaction;
252 if (transaction && time_after_eq(jiffies,
253 transaction->t_expires))
255 if (journal->j_flags & JBD2_UNMOUNT)
258 write_unlock(&journal->j_state_lock);
260 write_lock(&journal->j_state_lock);
262 finish_wait(&journal->j_wait_commit, &wait);
265 jbd_debug(1, "kjournald2 wakes\n");
268 * Were we woken up by a commit wakeup event?
270 transaction = journal->j_running_transaction;
271 if (transaction && time_after_eq(jiffies, transaction->t_expires)) {
272 journal->j_commit_request = transaction->t_tid;
273 jbd_debug(1, "woke because of timeout\n");
278 write_unlock(&journal->j_state_lock);
279 del_timer_sync(&journal->j_commit_timer);
280 journal->j_task = NULL;
281 wake_up(&journal->j_wait_done_commit);
282 jbd_debug(1, "Journal thread exiting.\n");
286 static int jbd2_journal_start_thread(journal_t *journal)
288 struct task_struct *t;
290 t = kthread_run(kjournald2, journal, "jbd2/%s",
295 wait_event(journal->j_wait_done_commit, journal->j_task != NULL);
299 static void journal_kill_thread(journal_t *journal)
301 write_lock(&journal->j_state_lock);
302 journal->j_flags |= JBD2_UNMOUNT;
304 while (journal->j_task) {
305 write_unlock(&journal->j_state_lock);
306 wake_up(&journal->j_wait_commit);
307 wait_event(journal->j_wait_done_commit, journal->j_task == NULL);
308 write_lock(&journal->j_state_lock);
310 write_unlock(&journal->j_state_lock);
314 * jbd2_journal_write_metadata_buffer: write a metadata buffer to the journal.
316 * Writes a metadata buffer to a given disk block. The actual IO is not
317 * performed but a new buffer_head is constructed which labels the data
318 * to be written with the correct destination disk block.
320 * Any magic-number escaping which needs to be done will cause a
321 * copy-out here. If the buffer happens to start with the
322 * JBD2_MAGIC_NUMBER, then we can't write it to the log directly: the
323 * magic number is only written to the log for descripter blocks. In
324 * this case, we copy the data and replace the first word with 0, and we
325 * return a result code which indicates that this buffer needs to be
326 * marked as an escaped buffer in the corresponding log descriptor
327 * block. The missing word can then be restored when the block is read
330 * If the source buffer has already been modified by a new transaction
331 * since we took the last commit snapshot, we use the frozen copy of
332 * that data for IO. If we end up using the existing buffer_head's data
333 * for the write, then we have to make sure nobody modifies it while the
334 * IO is in progress. do_get_write_access() handles this.
336 * The function returns a pointer to the buffer_head to be used for IO.
344 * Bit 0 set == escape performed on the data
345 * Bit 1 set == buffer copy-out performed (kfree the data after IO)
348 int jbd2_journal_write_metadata_buffer(transaction_t *transaction,
349 struct journal_head *jh_in,
350 struct buffer_head **bh_out,
353 int need_copy_out = 0;
354 int done_copy_out = 0;
357 struct buffer_head *new_bh;
358 struct page *new_page;
359 unsigned int new_offset;
360 struct buffer_head *bh_in = jh2bh(jh_in);
361 journal_t *journal = transaction->t_journal;
364 * The buffer really shouldn't be locked: only the current committing
365 * transaction is allowed to write it, so nobody else is allowed
368 * akpm: except if we're journalling data, and write() output is
369 * also part of a shared mapping, and another thread has
370 * decided to launch a writepage() against this buffer.
372 J_ASSERT_BH(bh_in, buffer_jbddirty(bh_in));
375 new_bh = alloc_buffer_head(GFP_NOFS);
378 * Failure is not an option, but __GFP_NOFAIL is going
379 * away; so we retry ourselves here.
381 congestion_wait(BLK_RW_ASYNC, HZ/50);
385 /* keep subsequent assertions sane */
386 atomic_set(&new_bh->b_count, 1);
388 jbd_lock_bh_state(bh_in);
391 * If a new transaction has already done a buffer copy-out, then
392 * we use that version of the data for the commit.
394 if (jh_in->b_frozen_data) {
396 new_page = virt_to_page(jh_in->b_frozen_data);
397 new_offset = offset_in_page(jh_in->b_frozen_data);
399 new_page = jh2bh(jh_in)->b_page;
400 new_offset = offset_in_page(jh2bh(jh_in)->b_data);
403 mapped_data = kmap_atomic(new_page);
405 * Fire data frozen trigger if data already wasn't frozen. Do this
406 * before checking for escaping, as the trigger may modify the magic
407 * offset. If a copy-out happens afterwards, it will have the correct
408 * data in the buffer.
411 jbd2_buffer_frozen_trigger(jh_in, mapped_data + new_offset,
417 if (*((__be32 *)(mapped_data + new_offset)) ==
418 cpu_to_be32(JBD2_MAGIC_NUMBER)) {
422 kunmap_atomic(mapped_data);
425 * Do we need to do a data copy?
427 if (need_copy_out && !done_copy_out) {
430 jbd_unlock_bh_state(bh_in);
431 tmp = jbd2_alloc(bh_in->b_size, GFP_NOFS);
436 jbd_lock_bh_state(bh_in);
437 if (jh_in->b_frozen_data) {
438 jbd2_free(tmp, bh_in->b_size);
442 jh_in->b_frozen_data = tmp;
443 mapped_data = kmap_atomic(new_page);
444 memcpy(tmp, mapped_data + new_offset, bh_in->b_size);
445 kunmap_atomic(mapped_data);
447 new_page = virt_to_page(tmp);
448 new_offset = offset_in_page(tmp);
452 * This isn't strictly necessary, as we're using frozen
453 * data for the escaping, but it keeps consistency with
454 * b_frozen_data usage.
456 jh_in->b_frozen_triggers = jh_in->b_triggers;
460 * Did we need to do an escaping? Now we've done all the
461 * copying, we can finally do so.
464 mapped_data = kmap_atomic(new_page);
465 *((unsigned int *)(mapped_data + new_offset)) = 0;
466 kunmap_atomic(mapped_data);
469 set_bh_page(new_bh, new_page, new_offset);
470 new_bh->b_size = bh_in->b_size;
471 new_bh->b_bdev = journal->j_dev;
472 new_bh->b_blocknr = blocknr;
473 new_bh->b_private = bh_in;
474 set_buffer_mapped(new_bh);
475 set_buffer_dirty(new_bh);
480 * The to-be-written buffer needs to get moved to the io queue,
481 * and the original buffer whose contents we are shadowing or
482 * copying is moved to the transaction's shadow queue.
484 JBUFFER_TRACE(jh_in, "file as BJ_Shadow");
485 spin_lock(&journal->j_list_lock);
486 __jbd2_journal_file_buffer(jh_in, transaction, BJ_Shadow);
487 spin_unlock(&journal->j_list_lock);
488 set_buffer_shadow(bh_in);
489 jbd_unlock_bh_state(bh_in);
491 return do_escape | (done_copy_out << 1);
495 * Allocation code for the journal file. Manage the space left in the
496 * journal, so that we can begin checkpointing when appropriate.
500 * Called with j_state_lock locked for writing.
501 * Returns true if a transaction commit was started.
503 int __jbd2_log_start_commit(journal_t *journal, tid_t target)
505 /* Return if the txn has already requested to be committed */
506 if (journal->j_commit_request == target)
510 * The only transaction we can possibly wait upon is the
511 * currently running transaction (if it exists). Otherwise,
512 * the target tid must be an old one.
514 if (journal->j_running_transaction &&
515 journal->j_running_transaction->t_tid == target) {
517 * We want a new commit: OK, mark the request and wakeup the
518 * commit thread. We do _not_ do the commit ourselves.
521 journal->j_commit_request = target;
522 jbd_debug(1, "JBD2: requesting commit %d/%d\n",
523 journal->j_commit_request,
524 journal->j_commit_sequence);
525 journal->j_running_transaction->t_requested = jiffies;
526 wake_up(&journal->j_wait_commit);
528 } else if (!tid_geq(journal->j_commit_request, target))
529 /* This should never happen, but if it does, preserve
530 the evidence before kjournald goes into a loop and
531 increments j_commit_sequence beyond all recognition. */
532 WARN_ONCE(1, "JBD2: bad log_start_commit: %u %u %u %u\n",
533 journal->j_commit_request,
534 journal->j_commit_sequence,
535 target, journal->j_running_transaction ?
536 journal->j_running_transaction->t_tid : 0);
540 int jbd2_log_start_commit(journal_t *journal, tid_t tid)
544 write_lock(&journal->j_state_lock);
545 ret = __jbd2_log_start_commit(journal, tid);
546 write_unlock(&journal->j_state_lock);
551 * Force and wait any uncommitted transactions. We can only force the running
552 * transaction if we don't have an active handle, otherwise, we will deadlock.
553 * Returns: <0 in case of error,
554 * 0 if nothing to commit,
555 * 1 if transaction was successfully committed.
557 static int __jbd2_journal_force_commit(journal_t *journal)
559 transaction_t *transaction = NULL;
561 int need_to_start = 0, ret = 0;
563 read_lock(&journal->j_state_lock);
564 if (journal->j_running_transaction && !current->journal_info) {
565 transaction = journal->j_running_transaction;
566 if (!tid_geq(journal->j_commit_request, transaction->t_tid))
568 } else if (journal->j_committing_transaction)
569 transaction = journal->j_committing_transaction;
572 /* Nothing to commit */
573 read_unlock(&journal->j_state_lock);
576 tid = transaction->t_tid;
577 read_unlock(&journal->j_state_lock);
579 jbd2_log_start_commit(journal, tid);
580 ret = jbd2_log_wait_commit(journal, tid);
588 * Force and wait upon a commit if the calling process is not within
589 * transaction. This is used for forcing out undo-protected data which contains
590 * bitmaps, when the fs is running out of space.
592 * @journal: journal to force
593 * Returns true if progress was made.
595 int jbd2_journal_force_commit_nested(journal_t *journal)
599 ret = __jbd2_journal_force_commit(journal);
604 * int journal_force_commit() - force any uncommitted transactions
605 * @journal: journal to force
607 * Caller want unconditional commit. We can only force the running transaction
608 * if we don't have an active handle, otherwise, we will deadlock.
610 int jbd2_journal_force_commit(journal_t *journal)
614 J_ASSERT(!current->journal_info);
615 ret = __jbd2_journal_force_commit(journal);
622 * Start a commit of the current running transaction (if any). Returns true
623 * if a transaction is going to be committed (or is currently already
624 * committing), and fills its tid in at *ptid
626 int jbd2_journal_start_commit(journal_t *journal, tid_t *ptid)
630 write_lock(&journal->j_state_lock);
631 if (journal->j_running_transaction) {
632 tid_t tid = journal->j_running_transaction->t_tid;
634 __jbd2_log_start_commit(journal, tid);
635 /* There's a running transaction and we've just made sure
636 * it's commit has been scheduled. */
640 } else if (journal->j_committing_transaction) {
642 * If commit has been started, then we have to wait for
643 * completion of that transaction.
646 *ptid = journal->j_committing_transaction->t_tid;
649 write_unlock(&journal->j_state_lock);
654 * Return 1 if a given transaction has not yet sent barrier request
655 * connected with a transaction commit. If 0 is returned, transaction
656 * may or may not have sent the barrier. Used to avoid sending barrier
657 * twice in common cases.
659 int jbd2_trans_will_send_data_barrier(journal_t *journal, tid_t tid)
662 transaction_t *commit_trans;
664 if (!(journal->j_flags & JBD2_BARRIER))
666 read_lock(&journal->j_state_lock);
667 /* Transaction already committed? */
668 if (tid_geq(journal->j_commit_sequence, tid))
670 commit_trans = journal->j_committing_transaction;
671 if (!commit_trans || commit_trans->t_tid != tid) {
676 * Transaction is being committed and we already proceeded to
677 * submitting a flush to fs partition?
679 if (journal->j_fs_dev != journal->j_dev) {
680 if (!commit_trans->t_need_data_flush ||
681 commit_trans->t_state >= T_COMMIT_DFLUSH)
684 if (commit_trans->t_state >= T_COMMIT_JFLUSH)
689 read_unlock(&journal->j_state_lock);
692 EXPORT_SYMBOL(jbd2_trans_will_send_data_barrier);
695 * Wait for a specified commit to complete.
696 * The caller may not hold the journal lock.
698 int jbd2_log_wait_commit(journal_t *journal, tid_t tid)
702 read_lock(&journal->j_state_lock);
703 #ifdef CONFIG_JBD2_DEBUG
704 if (!tid_geq(journal->j_commit_request, tid)) {
706 "%s: error: j_commit_request=%d, tid=%d\n",
707 __func__, journal->j_commit_request, tid);
710 while (tid_gt(tid, journal->j_commit_sequence)) {
711 jbd_debug(1, "JBD2: want %d, j_commit_sequence=%d\n",
712 tid, journal->j_commit_sequence);
713 read_unlock(&journal->j_state_lock);
714 wake_up(&journal->j_wait_commit);
715 wait_event(journal->j_wait_done_commit,
716 !tid_gt(tid, journal->j_commit_sequence));
717 read_lock(&journal->j_state_lock);
719 read_unlock(&journal->j_state_lock);
721 if (unlikely(is_journal_aborted(journal)))
727 * When this function returns the transaction corresponding to tid
728 * will be completed. If the transaction has currently running, start
729 * committing that transaction before waiting for it to complete. If
730 * the transaction id is stale, it is by definition already completed,
731 * so just return SUCCESS.
733 int jbd2_complete_transaction(journal_t *journal, tid_t tid)
735 int need_to_wait = 1;
737 read_lock(&journal->j_state_lock);
738 if (journal->j_running_transaction &&
739 journal->j_running_transaction->t_tid == tid) {
740 if (journal->j_commit_request != tid) {
741 /* transaction not yet started, so request it */
742 read_unlock(&journal->j_state_lock);
743 jbd2_log_start_commit(journal, tid);
746 } else if (!(journal->j_committing_transaction &&
747 journal->j_committing_transaction->t_tid == tid))
749 read_unlock(&journal->j_state_lock);
753 return jbd2_log_wait_commit(journal, tid);
755 EXPORT_SYMBOL(jbd2_complete_transaction);
758 * Log buffer allocation routines:
761 int jbd2_journal_next_log_block(journal_t *journal, unsigned long long *retp)
763 unsigned long blocknr;
765 write_lock(&journal->j_state_lock);
766 J_ASSERT(journal->j_free > 1);
768 blocknr = journal->j_head;
771 if (journal->j_head == journal->j_last)
772 journal->j_head = journal->j_first;
773 write_unlock(&journal->j_state_lock);
774 return jbd2_journal_bmap(journal, blocknr, retp);
778 * Conversion of logical to physical block numbers for the journal
780 * On external journals the journal blocks are identity-mapped, so
781 * this is a no-op. If needed, we can use j_blk_offset - everything is
784 int jbd2_journal_bmap(journal_t *journal, unsigned long blocknr,
785 unsigned long long *retp)
788 unsigned long long ret;
790 if (journal->j_inode) {
791 ret = bmap(journal->j_inode, blocknr);
795 printk(KERN_ALERT "%s: journal block not found "
796 "at offset %lu on %s\n",
797 __func__, blocknr, journal->j_devname);
799 __journal_abort_soft(journal, err);
802 *retp = blocknr; /* +journal->j_blk_offset */
808 * We play buffer_head aliasing tricks to write data/metadata blocks to
809 * the journal without copying their contents, but for journal
810 * descriptor blocks we do need to generate bona fide buffers.
812 * After the caller of jbd2_journal_get_descriptor_buffer() has finished modifying
813 * the buffer's contents they really should run flush_dcache_page(bh->b_page).
814 * But we don't bother doing that, so there will be coherency problems with
815 * mmaps of blockdevs which hold live JBD-controlled filesystems.
817 struct buffer_head *jbd2_journal_get_descriptor_buffer(journal_t *journal)
819 struct buffer_head *bh;
820 unsigned long long blocknr;
823 err = jbd2_journal_next_log_block(journal, &blocknr);
828 bh = __getblk(journal->j_dev, blocknr, journal->j_blocksize);
832 memset(bh->b_data, 0, journal->j_blocksize);
833 set_buffer_uptodate(bh);
835 BUFFER_TRACE(bh, "return this buffer");
840 * Return tid of the oldest transaction in the journal and block in the journal
841 * where the transaction starts.
843 * If the journal is now empty, return which will be the next transaction ID
844 * we will write and where will that transaction start.
846 * The return value is 0 if journal tail cannot be pushed any further, 1 if
849 int jbd2_journal_get_log_tail(journal_t *journal, tid_t *tid,
850 unsigned long *block)
852 transaction_t *transaction;
855 read_lock(&journal->j_state_lock);
856 spin_lock(&journal->j_list_lock);
857 transaction = journal->j_checkpoint_transactions;
859 *tid = transaction->t_tid;
860 *block = transaction->t_log_start;
861 } else if ((transaction = journal->j_committing_transaction) != NULL) {
862 *tid = transaction->t_tid;
863 *block = transaction->t_log_start;
864 } else if ((transaction = journal->j_running_transaction) != NULL) {
865 *tid = transaction->t_tid;
866 *block = journal->j_head;
868 *tid = journal->j_transaction_sequence;
869 *block = journal->j_head;
871 ret = tid_gt(*tid, journal->j_tail_sequence);
872 spin_unlock(&journal->j_list_lock);
873 read_unlock(&journal->j_state_lock);
879 * Update information in journal structure and in on disk journal superblock
880 * about log tail. This function does not check whether information passed in
881 * really pushes log tail further. It's responsibility of the caller to make
882 * sure provided log tail information is valid (e.g. by holding
883 * j_checkpoint_mutex all the time between computing log tail and calling this
884 * function as is the case with jbd2_cleanup_journal_tail()).
886 * Requires j_checkpoint_mutex
888 int __jbd2_update_log_tail(journal_t *journal, tid_t tid, unsigned long block)
893 BUG_ON(!mutex_is_locked(&journal->j_checkpoint_mutex));
896 * We cannot afford for write to remain in drive's caches since as
897 * soon as we update j_tail, next transaction can start reusing journal
898 * space and if we lose sb update during power failure we'd replay
899 * old transaction with possibly newly overwritten data.
901 ret = jbd2_journal_update_sb_log_tail(journal, tid, block, WRITE_FUA);
905 write_lock(&journal->j_state_lock);
906 freed = block - journal->j_tail;
907 if (block < journal->j_tail)
908 freed += journal->j_last - journal->j_first;
910 trace_jbd2_update_log_tail(journal, tid, block, freed);
912 "Cleaning journal tail from %d to %d (offset %lu), "
914 journal->j_tail_sequence, tid, block, freed);
916 journal->j_free += freed;
917 journal->j_tail_sequence = tid;
918 journal->j_tail = block;
919 write_unlock(&journal->j_state_lock);
926 * This is a variaon of __jbd2_update_log_tail which checks for validity of
927 * provided log tail and locks j_checkpoint_mutex. So it is safe against races
928 * with other threads updating log tail.
930 void jbd2_update_log_tail(journal_t *journal, tid_t tid, unsigned long block)
932 mutex_lock(&journal->j_checkpoint_mutex);
933 if (tid_gt(tid, journal->j_tail_sequence))
934 __jbd2_update_log_tail(journal, tid, block);
935 mutex_unlock(&journal->j_checkpoint_mutex);
938 struct jbd2_stats_proc_session {
940 struct transaction_stats_s *stats;
945 static void *jbd2_seq_info_start(struct seq_file *seq, loff_t *pos)
947 return *pos ? NULL : SEQ_START_TOKEN;
950 static void *jbd2_seq_info_next(struct seq_file *seq, void *v, loff_t *pos)
955 static int jbd2_seq_info_show(struct seq_file *seq, void *v)
957 struct jbd2_stats_proc_session *s = seq->private;
959 if (v != SEQ_START_TOKEN)
961 seq_printf(seq, "%lu transactions (%lu requested), "
962 "each up to %u blocks\n",
963 s->stats->ts_tid, s->stats->ts_requested,
964 s->journal->j_max_transaction_buffers);
965 if (s->stats->ts_tid == 0)
967 seq_printf(seq, "average: \n %ums waiting for transaction\n",
968 jiffies_to_msecs(s->stats->run.rs_wait / s->stats->ts_tid));
969 seq_printf(seq, " %ums request delay\n",
970 (s->stats->ts_requested == 0) ? 0 :
971 jiffies_to_msecs(s->stats->run.rs_request_delay /
972 s->stats->ts_requested));
973 seq_printf(seq, " %ums running transaction\n",
974 jiffies_to_msecs(s->stats->run.rs_running / s->stats->ts_tid));
975 seq_printf(seq, " %ums transaction was being locked\n",
976 jiffies_to_msecs(s->stats->run.rs_locked / s->stats->ts_tid));
977 seq_printf(seq, " %ums flushing data (in ordered mode)\n",
978 jiffies_to_msecs(s->stats->run.rs_flushing / s->stats->ts_tid));
979 seq_printf(seq, " %ums logging transaction\n",
980 jiffies_to_msecs(s->stats->run.rs_logging / s->stats->ts_tid));
981 seq_printf(seq, " %lluus average transaction commit time\n",
982 div_u64(s->journal->j_average_commit_time, 1000));
983 seq_printf(seq, " %lu handles per transaction\n",
984 s->stats->run.rs_handle_count / s->stats->ts_tid);
985 seq_printf(seq, " %lu blocks per transaction\n",
986 s->stats->run.rs_blocks / s->stats->ts_tid);
987 seq_printf(seq, " %lu logged blocks per transaction\n",
988 s->stats->run.rs_blocks_logged / s->stats->ts_tid);
992 static void jbd2_seq_info_stop(struct seq_file *seq, void *v)
996 static const struct seq_operations jbd2_seq_info_ops = {
997 .start = jbd2_seq_info_start,
998 .next = jbd2_seq_info_next,
999 .stop = jbd2_seq_info_stop,
1000 .show = jbd2_seq_info_show,
1003 static int jbd2_seq_info_open(struct inode *inode, struct file *file)
1005 journal_t *journal = PDE_DATA(inode);
1006 struct jbd2_stats_proc_session *s;
1009 s = kmalloc(sizeof(*s), GFP_KERNEL);
1012 size = sizeof(struct transaction_stats_s);
1013 s->stats = kmalloc(size, GFP_KERNEL);
1014 if (s->stats == NULL) {
1018 spin_lock(&journal->j_history_lock);
1019 memcpy(s->stats, &journal->j_stats, size);
1020 s->journal = journal;
1021 spin_unlock(&journal->j_history_lock);
1023 rc = seq_open(file, &jbd2_seq_info_ops);
1025 struct seq_file *m = file->private_data;
1035 static int jbd2_seq_info_release(struct inode *inode, struct file *file)
1037 struct seq_file *seq = file->private_data;
1038 struct jbd2_stats_proc_session *s = seq->private;
1041 return seq_release(inode, file);
1044 static const struct file_operations jbd2_seq_info_fops = {
1045 .owner = THIS_MODULE,
1046 .open = jbd2_seq_info_open,
1048 .llseek = seq_lseek,
1049 .release = jbd2_seq_info_release,
1052 static struct proc_dir_entry *proc_jbd2_stats;
1054 static void jbd2_stats_proc_init(journal_t *journal)
1056 journal->j_proc_entry = proc_mkdir(journal->j_devname, proc_jbd2_stats);
1057 if (journal->j_proc_entry) {
1058 proc_create_data("info", S_IRUGO, journal->j_proc_entry,
1059 &jbd2_seq_info_fops, journal);
1063 static void jbd2_stats_proc_exit(journal_t *journal)
1065 remove_proc_entry("info", journal->j_proc_entry);
1066 remove_proc_entry(journal->j_devname, proc_jbd2_stats);
1070 * Management for journal control blocks: functions to create and
1071 * destroy journal_t structures, and to initialise and read existing
1072 * journal blocks from disk. */
1074 /* First: create and setup a journal_t object in memory. We initialise
1075 * very few fields yet: that has to wait until we have created the
1076 * journal structures from from scratch, or loaded them from disk. */
1078 static journal_t * journal_init_common (void)
1083 journal = kzalloc(sizeof(*journal), GFP_KERNEL);
1087 init_waitqueue_head(&journal->j_wait_transaction_locked);
1088 init_waitqueue_head(&journal->j_wait_done_commit);
1089 init_waitqueue_head(&journal->j_wait_commit);
1090 init_waitqueue_head(&journal->j_wait_updates);
1091 init_waitqueue_head(&journal->j_wait_reserved);
1092 mutex_init(&journal->j_barrier);
1093 mutex_init(&journal->j_checkpoint_mutex);
1094 spin_lock_init(&journal->j_revoke_lock);
1095 spin_lock_init(&journal->j_list_lock);
1096 rwlock_init(&journal->j_state_lock);
1098 journal->j_commit_interval = (HZ * JBD2_DEFAULT_MAX_COMMIT_AGE);
1099 journal->j_min_batch_time = 0;
1100 journal->j_max_batch_time = 15000; /* 15ms */
1101 atomic_set(&journal->j_reserved_credits, 0);
1103 /* The journal is marked for error until we succeed with recovery! */
1104 journal->j_flags = JBD2_ABORT;
1106 /* Set up a default-sized revoke table for the new mount. */
1107 err = jbd2_journal_init_revoke(journal, JOURNAL_REVOKE_DEFAULT_HASH);
1113 spin_lock_init(&journal->j_history_lock);
1118 /* jbd2_journal_init_dev and jbd2_journal_init_inode:
1120 * Create a journal structure assigned some fixed set of disk blocks to
1121 * the journal. We don't actually touch those disk blocks yet, but we
1122 * need to set up all of the mapping information to tell the journaling
1123 * system where the journal blocks are.
1128 * journal_t * jbd2_journal_init_dev() - creates and initialises a journal structure
1129 * @bdev: Block device on which to create the journal
1130 * @fs_dev: Device which hold journalled filesystem for this journal.
1131 * @start: Block nr Start of journal.
1132 * @len: Length of the journal in blocks.
1133 * @blocksize: blocksize of journalling device
1135 * Returns: a newly created journal_t *
1137 * jbd2_journal_init_dev creates a journal which maps a fixed contiguous
1138 * range of blocks on an arbitrary block device.
1141 journal_t * jbd2_journal_init_dev(struct block_device *bdev,
1142 struct block_device *fs_dev,
1143 unsigned long long start, int len, int blocksize)
1145 journal_t *journal = journal_init_common();
1146 struct buffer_head *bh;
1153 /* journal descriptor can store up to n blocks -bzzz */
1154 journal->j_blocksize = blocksize;
1155 journal->j_dev = bdev;
1156 journal->j_fs_dev = fs_dev;
1157 journal->j_blk_offset = start;
1158 journal->j_maxlen = len;
1159 bdevname(journal->j_dev, journal->j_devname);
1160 p = journal->j_devname;
1161 while ((p = strchr(p, '/')))
1163 jbd2_stats_proc_init(journal);
1164 n = journal->j_blocksize / sizeof(journal_block_tag_t);
1165 journal->j_wbufsize = n;
1166 journal->j_wbuf = kmalloc(n * sizeof(struct buffer_head*), GFP_KERNEL);
1167 if (!journal->j_wbuf) {
1168 printk(KERN_ERR "%s: Can't allocate bhs for commit thread\n",
1173 bh = __getblk(journal->j_dev, start, journal->j_blocksize);
1176 "%s: Cannot get buffer for journal superblock\n",
1180 journal->j_sb_buffer = bh;
1181 journal->j_superblock = (journal_superblock_t *)bh->b_data;
1185 kfree(journal->j_wbuf);
1186 jbd2_stats_proc_exit(journal);
1192 * journal_t * jbd2_journal_init_inode () - creates a journal which maps to a inode.
1193 * @inode: An inode to create the journal in
1195 * jbd2_journal_init_inode creates a journal which maps an on-disk inode as
1196 * the journal. The inode must exist already, must support bmap() and
1197 * must have all data blocks preallocated.
1199 journal_t * jbd2_journal_init_inode (struct inode *inode)
1201 struct buffer_head *bh;
1202 journal_t *journal = journal_init_common();
1206 unsigned long long blocknr;
1211 journal->j_dev = journal->j_fs_dev = inode->i_sb->s_bdev;
1212 journal->j_inode = inode;
1213 bdevname(journal->j_dev, journal->j_devname);
1214 p = journal->j_devname;
1215 while ((p = strchr(p, '/')))
1217 p = journal->j_devname + strlen(journal->j_devname);
1218 sprintf(p, "-%lu", journal->j_inode->i_ino);
1220 "journal %p: inode %s/%ld, size %Ld, bits %d, blksize %ld\n",
1221 journal, inode->i_sb->s_id, inode->i_ino,
1222 (long long) inode->i_size,
1223 inode->i_sb->s_blocksize_bits, inode->i_sb->s_blocksize);
1225 journal->j_maxlen = inode->i_size >> inode->i_sb->s_blocksize_bits;
1226 journal->j_blocksize = inode->i_sb->s_blocksize;
1227 jbd2_stats_proc_init(journal);
1229 /* journal descriptor can store up to n blocks -bzzz */
1230 n = journal->j_blocksize / sizeof(journal_block_tag_t);
1231 journal->j_wbufsize = n;
1232 journal->j_wbuf = kmalloc(n * sizeof(struct buffer_head*), GFP_KERNEL);
1233 if (!journal->j_wbuf) {
1234 printk(KERN_ERR "%s: Can't allocate bhs for commit thread\n",
1239 err = jbd2_journal_bmap(journal, 0, &blocknr);
1240 /* If that failed, give up */
1242 printk(KERN_ERR "%s: Cannot locate journal superblock\n",
1247 bh = getblk_unmovable(journal->j_dev, blocknr, journal->j_blocksize);
1250 "%s: Cannot get buffer for journal superblock\n",
1254 journal->j_sb_buffer = bh;
1255 journal->j_superblock = (journal_superblock_t *)bh->b_data;
1259 kfree(journal->j_wbuf);
1260 jbd2_stats_proc_exit(journal);
1266 * If the journal init or create aborts, we need to mark the journal
1267 * superblock as being NULL to prevent the journal destroy from writing
1268 * back a bogus superblock.
1270 static void journal_fail_superblock (journal_t *journal)
1272 struct buffer_head *bh = journal->j_sb_buffer;
1274 journal->j_sb_buffer = NULL;
1278 * Given a journal_t structure, initialise the various fields for
1279 * startup of a new journaling session. We use this both when creating
1280 * a journal, and after recovering an old journal to reset it for
1284 static int journal_reset(journal_t *journal)
1286 journal_superblock_t *sb = journal->j_superblock;
1287 unsigned long long first, last;
1289 first = be32_to_cpu(sb->s_first);
1290 last = be32_to_cpu(sb->s_maxlen);
1291 if (first + JBD2_MIN_JOURNAL_BLOCKS > last + 1) {
1292 printk(KERN_ERR "JBD2: Journal too short (blocks %llu-%llu).\n",
1294 journal_fail_superblock(journal);
1298 journal->j_first = first;
1299 journal->j_last = last;
1301 journal->j_head = first;
1302 journal->j_tail = first;
1303 journal->j_free = last - first;
1305 journal->j_tail_sequence = journal->j_transaction_sequence;
1306 journal->j_commit_sequence = journal->j_transaction_sequence - 1;
1307 journal->j_commit_request = journal->j_commit_sequence;
1309 journal->j_max_transaction_buffers = journal->j_maxlen / 4;
1312 * As a special case, if the on-disk copy is already marked as needing
1313 * no recovery (s_start == 0), then we can safely defer the superblock
1314 * update until the next commit by setting JBD2_FLUSHED. This avoids
1315 * attempting a write to a potential-readonly device.
1317 if (sb->s_start == 0) {
1318 jbd_debug(1, "JBD2: Skipping superblock update on recovered sb "
1319 "(start %ld, seq %d, errno %d)\n",
1320 journal->j_tail, journal->j_tail_sequence,
1322 journal->j_flags |= JBD2_FLUSHED;
1324 /* Lock here to make assertions happy... */
1325 mutex_lock(&journal->j_checkpoint_mutex);
1327 * Update log tail information. We use WRITE_FUA since new
1328 * transaction will start reusing journal space and so we
1329 * must make sure information about current log tail is on
1332 jbd2_journal_update_sb_log_tail(journal,
1333 journal->j_tail_sequence,
1336 mutex_unlock(&journal->j_checkpoint_mutex);
1338 return jbd2_journal_start_thread(journal);
1341 static int jbd2_write_superblock(journal_t *journal, int write_op)
1343 struct buffer_head *bh = journal->j_sb_buffer;
1344 journal_superblock_t *sb = journal->j_superblock;
1347 trace_jbd2_write_superblock(journal, write_op);
1348 if (!(journal->j_flags & JBD2_BARRIER))
1349 write_op &= ~(REQ_FUA | REQ_FLUSH);
1351 if (buffer_write_io_error(bh)) {
1353 * Oh, dear. A previous attempt to write the journal
1354 * superblock failed. This could happen because the
1355 * USB device was yanked out. Or it could happen to
1356 * be a transient write error and maybe the block will
1357 * be remapped. Nothing we can do but to retry the
1358 * write and hope for the best.
1360 printk(KERN_ERR "JBD2: previous I/O error detected "
1361 "for journal superblock update for %s.\n",
1362 journal->j_devname);
1363 clear_buffer_write_io_error(bh);
1364 set_buffer_uptodate(bh);
1366 jbd2_superblock_csum_set(journal, sb);
1368 bh->b_end_io = end_buffer_write_sync;
1369 ret = submit_bh(write_op, bh);
1371 if (buffer_write_io_error(bh)) {
1372 clear_buffer_write_io_error(bh);
1373 set_buffer_uptodate(bh);
1377 printk(KERN_ERR "JBD2: Error %d detected when updating "
1378 "journal superblock for %s.\n", ret,
1379 journal->j_devname);
1380 jbd2_journal_abort(journal, ret);
1387 * jbd2_journal_update_sb_log_tail() - Update log tail in journal sb on disk.
1388 * @journal: The journal to update.
1389 * @tail_tid: TID of the new transaction at the tail of the log
1390 * @tail_block: The first block of the transaction at the tail of the log
1391 * @write_op: With which operation should we write the journal sb
1393 * Update a journal's superblock information about log tail and write it to
1394 * disk, waiting for the IO to complete.
1396 int jbd2_journal_update_sb_log_tail(journal_t *journal, tid_t tail_tid,
1397 unsigned long tail_block, int write_op)
1399 journal_superblock_t *sb = journal->j_superblock;
1402 BUG_ON(!mutex_is_locked(&journal->j_checkpoint_mutex));
1403 jbd_debug(1, "JBD2: updating superblock (start %lu, seq %u)\n",
1404 tail_block, tail_tid);
1406 sb->s_sequence = cpu_to_be32(tail_tid);
1407 sb->s_start = cpu_to_be32(tail_block);
1409 ret = jbd2_write_superblock(journal, write_op);
1413 /* Log is no longer empty */
1414 write_lock(&journal->j_state_lock);
1415 WARN_ON(!sb->s_sequence);
1416 journal->j_flags &= ~JBD2_FLUSHED;
1417 write_unlock(&journal->j_state_lock);
1424 * jbd2_mark_journal_empty() - Mark on disk journal as empty.
1425 * @journal: The journal to update.
1427 * Update a journal's dynamic superblock fields to show that journal is empty.
1428 * Write updated superblock to disk waiting for IO to complete.
1430 static void jbd2_mark_journal_empty(journal_t *journal)
1432 journal_superblock_t *sb = journal->j_superblock;
1434 BUG_ON(!mutex_is_locked(&journal->j_checkpoint_mutex));
1435 read_lock(&journal->j_state_lock);
1436 /* Is it already empty? */
1437 if (sb->s_start == 0) {
1438 read_unlock(&journal->j_state_lock);
1441 jbd_debug(1, "JBD2: Marking journal as empty (seq %d)\n",
1442 journal->j_tail_sequence);
1444 sb->s_sequence = cpu_to_be32(journal->j_tail_sequence);
1445 sb->s_start = cpu_to_be32(0);
1446 read_unlock(&journal->j_state_lock);
1448 jbd2_write_superblock(journal, WRITE_FUA);
1450 /* Log is no longer empty */
1451 write_lock(&journal->j_state_lock);
1452 journal->j_flags |= JBD2_FLUSHED;
1453 write_unlock(&journal->j_state_lock);
1458 * jbd2_journal_update_sb_errno() - Update error in the journal.
1459 * @journal: The journal to update.
1461 * Update a journal's errno. Write updated superblock to disk waiting for IO
1464 void jbd2_journal_update_sb_errno(journal_t *journal)
1466 journal_superblock_t *sb = journal->j_superblock;
1468 read_lock(&journal->j_state_lock);
1469 jbd_debug(1, "JBD2: updating superblock error (errno %d)\n",
1471 sb->s_errno = cpu_to_be32(journal->j_errno);
1472 read_unlock(&journal->j_state_lock);
1474 jbd2_write_superblock(journal, WRITE_SYNC);
1476 EXPORT_SYMBOL(jbd2_journal_update_sb_errno);
1479 * Read the superblock for a given journal, performing initial
1480 * validation of the format.
1482 static int journal_get_superblock(journal_t *journal)
1484 struct buffer_head *bh;
1485 journal_superblock_t *sb;
1488 bh = journal->j_sb_buffer;
1490 J_ASSERT(bh != NULL);
1491 if (!buffer_uptodate(bh)) {
1492 ll_rw_block(READ, 1, &bh);
1494 if (!buffer_uptodate(bh)) {
1496 "JBD2: IO error reading journal superblock\n");
1501 if (buffer_verified(bh))
1504 sb = journal->j_superblock;
1508 if (sb->s_header.h_magic != cpu_to_be32(JBD2_MAGIC_NUMBER) ||
1509 sb->s_blocksize != cpu_to_be32(journal->j_blocksize)) {
1510 printk(KERN_WARNING "JBD2: no valid journal superblock found\n");
1514 switch(be32_to_cpu(sb->s_header.h_blocktype)) {
1515 case JBD2_SUPERBLOCK_V1:
1516 journal->j_format_version = 1;
1518 case JBD2_SUPERBLOCK_V2:
1519 journal->j_format_version = 2;
1522 printk(KERN_WARNING "JBD2: unrecognised superblock format ID\n");
1526 if (be32_to_cpu(sb->s_maxlen) < journal->j_maxlen)
1527 journal->j_maxlen = be32_to_cpu(sb->s_maxlen);
1528 else if (be32_to_cpu(sb->s_maxlen) > journal->j_maxlen) {
1529 printk(KERN_WARNING "JBD2: journal file too short\n");
1533 if (be32_to_cpu(sb->s_first) == 0 ||
1534 be32_to_cpu(sb->s_first) >= journal->j_maxlen) {
1536 "JBD2: Invalid start block of journal: %u\n",
1537 be32_to_cpu(sb->s_first));
1541 if (JBD2_HAS_INCOMPAT_FEATURE(journal, JBD2_FEATURE_INCOMPAT_CSUM_V2) &&
1542 JBD2_HAS_INCOMPAT_FEATURE(journal, JBD2_FEATURE_INCOMPAT_CSUM_V3)) {
1543 /* Can't have checksum v2 and v3 at the same time! */
1544 printk(KERN_ERR "JBD2: Can't enable checksumming v2 and v3 "
1545 "at the same time!\n");
1549 if (jbd2_journal_has_csum_v2or3(journal) &&
1550 JBD2_HAS_COMPAT_FEATURE(journal, JBD2_FEATURE_COMPAT_CHECKSUM)) {
1551 /* Can't have checksum v1 and v2 on at the same time! */
1552 printk(KERN_ERR "JBD2: Can't enable checksumming v1 and v2/3 "
1553 "at the same time!\n");
1557 if (!jbd2_verify_csum_type(journal, sb)) {
1558 printk(KERN_ERR "JBD2: Unknown checksum type\n");
1562 /* Load the checksum driver */
1563 if (jbd2_journal_has_csum_v2or3(journal)) {
1564 journal->j_chksum_driver = crypto_alloc_shash("crc32c", 0, 0);
1565 if (IS_ERR(journal->j_chksum_driver)) {
1566 printk(KERN_ERR "JBD2: Cannot load crc32c driver.\n");
1567 err = PTR_ERR(journal->j_chksum_driver);
1568 journal->j_chksum_driver = NULL;
1573 /* Check superblock checksum */
1574 if (!jbd2_superblock_csum_verify(journal, sb)) {
1575 printk(KERN_ERR "JBD2: journal checksum error\n");
1579 /* Precompute checksum seed for all metadata */
1580 if (jbd2_journal_has_csum_v2or3(journal))
1581 journal->j_csum_seed = jbd2_chksum(journal, ~0, sb->s_uuid,
1582 sizeof(sb->s_uuid));
1584 set_buffer_verified(bh);
1589 journal_fail_superblock(journal);
1594 * Load the on-disk journal superblock and read the key fields into the
1598 static int load_superblock(journal_t *journal)
1601 journal_superblock_t *sb;
1603 err = journal_get_superblock(journal);
1607 sb = journal->j_superblock;
1609 journal->j_tail_sequence = be32_to_cpu(sb->s_sequence);
1610 journal->j_tail = be32_to_cpu(sb->s_start);
1611 journal->j_first = be32_to_cpu(sb->s_first);
1612 journal->j_last = be32_to_cpu(sb->s_maxlen);
1613 journal->j_errno = be32_to_cpu(sb->s_errno);
1620 * int jbd2_journal_load() - Read journal from disk.
1621 * @journal: Journal to act on.
1623 * Given a journal_t structure which tells us which disk blocks contain
1624 * a journal, read the journal from disk to initialise the in-memory
1627 int jbd2_journal_load(journal_t *journal)
1630 journal_superblock_t *sb;
1632 err = load_superblock(journal);
1636 sb = journal->j_superblock;
1637 /* If this is a V2 superblock, then we have to check the
1638 * features flags on it. */
1640 if (journal->j_format_version >= 2) {
1641 if ((sb->s_feature_ro_compat &
1642 ~cpu_to_be32(JBD2_KNOWN_ROCOMPAT_FEATURES)) ||
1643 (sb->s_feature_incompat &
1644 ~cpu_to_be32(JBD2_KNOWN_INCOMPAT_FEATURES))) {
1646 "JBD2: Unrecognised features on journal\n");
1652 * Create a slab for this blocksize
1654 err = jbd2_journal_create_slab(be32_to_cpu(sb->s_blocksize));
1658 /* Let the recovery code check whether it needs to recover any
1659 * data from the journal. */
1660 if (jbd2_journal_recover(journal))
1661 goto recovery_error;
1663 if (journal->j_failed_commit) {
1664 printk(KERN_ERR "JBD2: journal transaction %u on %s "
1665 "is corrupt.\n", journal->j_failed_commit,
1666 journal->j_devname);
1670 /* OK, we've finished with the dynamic journal bits:
1671 * reinitialise the dynamic contents of the superblock in memory
1672 * and reset them on disk. */
1673 if (journal_reset(journal))
1674 goto recovery_error;
1676 journal->j_flags &= ~JBD2_ABORT;
1677 journal->j_flags |= JBD2_LOADED;
1681 printk(KERN_WARNING "JBD2: recovery failed\n");
1686 * void jbd2_journal_destroy() - Release a journal_t structure.
1687 * @journal: Journal to act on.
1689 * Release a journal_t structure once it is no longer in use by the
1691 * Return <0 if we couldn't clean up the journal.
1693 int jbd2_journal_destroy(journal_t *journal)
1697 /* Wait for the commit thread to wake up and die. */
1698 journal_kill_thread(journal);
1700 /* Force a final log commit */
1701 if (journal->j_running_transaction)
1702 jbd2_journal_commit_transaction(journal);
1704 /* Force any old transactions to disk */
1706 /* Totally anal locking here... */
1707 spin_lock(&journal->j_list_lock);
1708 while (journal->j_checkpoint_transactions != NULL) {
1709 spin_unlock(&journal->j_list_lock);
1710 mutex_lock(&journal->j_checkpoint_mutex);
1711 jbd2_log_do_checkpoint(journal);
1712 mutex_unlock(&journal->j_checkpoint_mutex);
1713 spin_lock(&journal->j_list_lock);
1716 J_ASSERT(journal->j_running_transaction == NULL);
1717 J_ASSERT(journal->j_committing_transaction == NULL);
1718 J_ASSERT(journal->j_checkpoint_transactions == NULL);
1719 spin_unlock(&journal->j_list_lock);
1721 if (journal->j_sb_buffer) {
1722 if (!is_journal_aborted(journal)) {
1723 mutex_lock(&journal->j_checkpoint_mutex);
1724 jbd2_mark_journal_empty(journal);
1725 mutex_unlock(&journal->j_checkpoint_mutex);
1728 brelse(journal->j_sb_buffer);
1731 if (journal->j_proc_entry)
1732 jbd2_stats_proc_exit(journal);
1733 iput(journal->j_inode);
1734 if (journal->j_revoke)
1735 jbd2_journal_destroy_revoke(journal);
1736 if (journal->j_chksum_driver)
1737 crypto_free_shash(journal->j_chksum_driver);
1738 kfree(journal->j_wbuf);
1746 *int jbd2_journal_check_used_features () - Check if features specified are used.
1747 * @journal: Journal to check.
1748 * @compat: bitmask of compatible features
1749 * @ro: bitmask of features that force read-only mount
1750 * @incompat: bitmask of incompatible features
1752 * Check whether the journal uses all of a given set of
1753 * features. Return true (non-zero) if it does.
1756 int jbd2_journal_check_used_features (journal_t *journal, unsigned long compat,
1757 unsigned long ro, unsigned long incompat)
1759 journal_superblock_t *sb;
1761 if (!compat && !ro && !incompat)
1763 /* Load journal superblock if it is not loaded yet. */
1764 if (journal->j_format_version == 0 &&
1765 journal_get_superblock(journal) != 0)
1767 if (journal->j_format_version == 1)
1770 sb = journal->j_superblock;
1772 if (((be32_to_cpu(sb->s_feature_compat) & compat) == compat) &&
1773 ((be32_to_cpu(sb->s_feature_ro_compat) & ro) == ro) &&
1774 ((be32_to_cpu(sb->s_feature_incompat) & incompat) == incompat))
1781 * int jbd2_journal_check_available_features() - Check feature set in journalling layer
1782 * @journal: Journal to check.
1783 * @compat: bitmask of compatible features
1784 * @ro: bitmask of features that force read-only mount
1785 * @incompat: bitmask of incompatible features
1787 * Check whether the journaling code supports the use of
1788 * all of a given set of features on this journal. Return true
1789 * (non-zero) if it can. */
1791 int jbd2_journal_check_available_features (journal_t *journal, unsigned long compat,
1792 unsigned long ro, unsigned long incompat)
1794 if (!compat && !ro && !incompat)
1797 /* We can support any known requested features iff the
1798 * superblock is in version 2. Otherwise we fail to support any
1799 * extended sb features. */
1801 if (journal->j_format_version != 2)
1804 if ((compat & JBD2_KNOWN_COMPAT_FEATURES) == compat &&
1805 (ro & JBD2_KNOWN_ROCOMPAT_FEATURES) == ro &&
1806 (incompat & JBD2_KNOWN_INCOMPAT_FEATURES) == incompat)
1813 * int jbd2_journal_set_features () - Mark a given journal feature in the superblock
1814 * @journal: Journal to act on.
1815 * @compat: bitmask of compatible features
1816 * @ro: bitmask of features that force read-only mount
1817 * @incompat: bitmask of incompatible features
1819 * Mark a given journal feature as present on the
1820 * superblock. Returns true if the requested features could be set.
1824 int jbd2_journal_set_features (journal_t *journal, unsigned long compat,
1825 unsigned long ro, unsigned long incompat)
1827 #define INCOMPAT_FEATURE_ON(f) \
1828 ((incompat & (f)) && !(sb->s_feature_incompat & cpu_to_be32(f)))
1829 #define COMPAT_FEATURE_ON(f) \
1830 ((compat & (f)) && !(sb->s_feature_compat & cpu_to_be32(f)))
1831 journal_superblock_t *sb;
1833 if (jbd2_journal_check_used_features(journal, compat, ro, incompat))
1836 if (!jbd2_journal_check_available_features(journal, compat, ro, incompat))
1839 /* If enabling v2 checksums, turn on v3 instead */
1840 if (incompat & JBD2_FEATURE_INCOMPAT_CSUM_V2) {
1841 incompat &= ~JBD2_FEATURE_INCOMPAT_CSUM_V2;
1842 incompat |= JBD2_FEATURE_INCOMPAT_CSUM_V3;
1845 /* Asking for checksumming v3 and v1? Only give them v3. */
1846 if (incompat & JBD2_FEATURE_INCOMPAT_CSUM_V3 &&
1847 compat & JBD2_FEATURE_COMPAT_CHECKSUM)
1848 compat &= ~JBD2_FEATURE_COMPAT_CHECKSUM;
1850 jbd_debug(1, "Setting new features 0x%lx/0x%lx/0x%lx\n",
1851 compat, ro, incompat);
1853 sb = journal->j_superblock;
1855 /* If enabling v3 checksums, update superblock */
1856 if (INCOMPAT_FEATURE_ON(JBD2_FEATURE_INCOMPAT_CSUM_V3)) {
1857 sb->s_checksum_type = JBD2_CRC32C_CHKSUM;
1858 sb->s_feature_compat &=
1859 ~cpu_to_be32(JBD2_FEATURE_COMPAT_CHECKSUM);
1861 /* Load the checksum driver */
1862 if (journal->j_chksum_driver == NULL) {
1863 journal->j_chksum_driver = crypto_alloc_shash("crc32c",
1865 if (IS_ERR(journal->j_chksum_driver)) {
1866 printk(KERN_ERR "JBD2: Cannot load crc32c "
1868 journal->j_chksum_driver = NULL;
1872 /* Precompute checksum seed for all metadata */
1873 journal->j_csum_seed = jbd2_chksum(journal, ~0,
1875 sizeof(sb->s_uuid));
1879 /* If enabling v1 checksums, downgrade superblock */
1880 if (COMPAT_FEATURE_ON(JBD2_FEATURE_COMPAT_CHECKSUM))
1881 sb->s_feature_incompat &=
1882 ~cpu_to_be32(JBD2_FEATURE_INCOMPAT_CSUM_V2 |
1883 JBD2_FEATURE_INCOMPAT_CSUM_V3);
1885 sb->s_feature_compat |= cpu_to_be32(compat);
1886 sb->s_feature_ro_compat |= cpu_to_be32(ro);
1887 sb->s_feature_incompat |= cpu_to_be32(incompat);
1890 #undef COMPAT_FEATURE_ON
1891 #undef INCOMPAT_FEATURE_ON
1895 * jbd2_journal_clear_features () - Clear a given journal feature in the
1897 * @journal: Journal to act on.
1898 * @compat: bitmask of compatible features
1899 * @ro: bitmask of features that force read-only mount
1900 * @incompat: bitmask of incompatible features
1902 * Clear a given journal feature as present on the
1905 void jbd2_journal_clear_features(journal_t *journal, unsigned long compat,
1906 unsigned long ro, unsigned long incompat)
1908 journal_superblock_t *sb;
1910 jbd_debug(1, "Clear features 0x%lx/0x%lx/0x%lx\n",
1911 compat, ro, incompat);
1913 sb = journal->j_superblock;
1915 sb->s_feature_compat &= ~cpu_to_be32(compat);
1916 sb->s_feature_ro_compat &= ~cpu_to_be32(ro);
1917 sb->s_feature_incompat &= ~cpu_to_be32(incompat);
1919 EXPORT_SYMBOL(jbd2_journal_clear_features);
1922 * int jbd2_journal_flush () - Flush journal
1923 * @journal: Journal to act on.
1925 * Flush all data for a given journal to disk and empty the journal.
1926 * Filesystems can use this when remounting readonly to ensure that
1927 * recovery does not need to happen on remount.
1930 int jbd2_journal_flush(journal_t *journal)
1933 transaction_t *transaction = NULL;
1935 write_lock(&journal->j_state_lock);
1937 /* Force everything buffered to the log... */
1938 if (journal->j_running_transaction) {
1939 transaction = journal->j_running_transaction;
1940 __jbd2_log_start_commit(journal, transaction->t_tid);
1941 } else if (journal->j_committing_transaction)
1942 transaction = journal->j_committing_transaction;
1944 /* Wait for the log commit to complete... */
1946 tid_t tid = transaction->t_tid;
1948 write_unlock(&journal->j_state_lock);
1949 jbd2_log_wait_commit(journal, tid);
1951 write_unlock(&journal->j_state_lock);
1954 /* ...and flush everything in the log out to disk. */
1955 spin_lock(&journal->j_list_lock);
1956 while (!err && journal->j_checkpoint_transactions != NULL) {
1957 spin_unlock(&journal->j_list_lock);
1958 mutex_lock(&journal->j_checkpoint_mutex);
1959 err = jbd2_log_do_checkpoint(journal);
1960 mutex_unlock(&journal->j_checkpoint_mutex);
1961 spin_lock(&journal->j_list_lock);
1963 spin_unlock(&journal->j_list_lock);
1965 if (is_journal_aborted(journal))
1968 mutex_lock(&journal->j_checkpoint_mutex);
1970 err = jbd2_cleanup_journal_tail(journal);
1972 mutex_unlock(&journal->j_checkpoint_mutex);
1978 /* Finally, mark the journal as really needing no recovery.
1979 * This sets s_start==0 in the underlying superblock, which is
1980 * the magic code for a fully-recovered superblock. Any future
1981 * commits of data to the journal will restore the current
1983 jbd2_mark_journal_empty(journal);
1984 mutex_unlock(&journal->j_checkpoint_mutex);
1985 write_lock(&journal->j_state_lock);
1986 J_ASSERT(!journal->j_running_transaction);
1987 J_ASSERT(!journal->j_committing_transaction);
1988 J_ASSERT(!journal->j_checkpoint_transactions);
1989 J_ASSERT(journal->j_head == journal->j_tail);
1990 J_ASSERT(journal->j_tail_sequence == journal->j_transaction_sequence);
1991 write_unlock(&journal->j_state_lock);
1997 * int jbd2_journal_wipe() - Wipe journal contents
1998 * @journal: Journal to act on.
1999 * @write: flag (see below)
2001 * Wipe out all of the contents of a journal, safely. This will produce
2002 * a warning if the journal contains any valid recovery information.
2003 * Must be called between journal_init_*() and jbd2_journal_load().
2005 * If 'write' is non-zero, then we wipe out the journal on disk; otherwise
2006 * we merely suppress recovery.
2009 int jbd2_journal_wipe(journal_t *journal, int write)
2013 J_ASSERT (!(journal->j_flags & JBD2_LOADED));
2015 err = load_superblock(journal);
2019 if (!journal->j_tail)
2022 printk(KERN_WARNING "JBD2: %s recovery information on journal\n",
2023 write ? "Clearing" : "Ignoring");
2025 err = jbd2_journal_skip_recovery(journal);
2027 /* Lock to make assertions happy... */
2028 mutex_lock(&journal->j_checkpoint_mutex);
2029 jbd2_mark_journal_empty(journal);
2030 mutex_unlock(&journal->j_checkpoint_mutex);
2038 * Journal abort has very specific semantics, which we describe
2039 * for journal abort.
2041 * Two internal functions, which provide abort to the jbd layer
2046 * Quick version for internal journal use (doesn't lock the journal).
2047 * Aborts hard --- we mark the abort as occurred, but do _nothing_ else,
2048 * and don't attempt to make any other journal updates.
2050 void __jbd2_journal_abort_hard(journal_t *journal)
2052 transaction_t *transaction;
2054 if (journal->j_flags & JBD2_ABORT)
2057 printk(KERN_ERR "Aborting journal on device %s.\n",
2058 journal->j_devname);
2060 write_lock(&journal->j_state_lock);
2061 journal->j_flags |= JBD2_ABORT;
2062 transaction = journal->j_running_transaction;
2064 __jbd2_log_start_commit(journal, transaction->t_tid);
2065 write_unlock(&journal->j_state_lock);
2068 /* Soft abort: record the abort error status in the journal superblock,
2069 * but don't do any other IO. */
2070 static void __journal_abort_soft (journal_t *journal, int errno)
2072 if (journal->j_flags & JBD2_ABORT)
2075 if (!journal->j_errno)
2076 journal->j_errno = errno;
2078 __jbd2_journal_abort_hard(journal);
2081 jbd2_journal_update_sb_errno(journal);
2085 * void jbd2_journal_abort () - Shutdown the journal immediately.
2086 * @journal: the journal to shutdown.
2087 * @errno: an error number to record in the journal indicating
2088 * the reason for the shutdown.
2090 * Perform a complete, immediate shutdown of the ENTIRE
2091 * journal (not of a single transaction). This operation cannot be
2092 * undone without closing and reopening the journal.
2094 * The jbd2_journal_abort function is intended to support higher level error
2095 * recovery mechanisms such as the ext2/ext3 remount-readonly error
2098 * Journal abort has very specific semantics. Any existing dirty,
2099 * unjournaled buffers in the main filesystem will still be written to
2100 * disk by bdflush, but the journaling mechanism will be suspended
2101 * immediately and no further transaction commits will be honoured.
2103 * Any dirty, journaled buffers will be written back to disk without
2104 * hitting the journal. Atomicity cannot be guaranteed on an aborted
2105 * filesystem, but we _do_ attempt to leave as much data as possible
2106 * behind for fsck to use for cleanup.
2108 * Any attempt to get a new transaction handle on a journal which is in
2109 * ABORT state will just result in an -EROFS error return. A
2110 * jbd2_journal_stop on an existing handle will return -EIO if we have
2111 * entered abort state during the update.
2113 * Recursive transactions are not disturbed by journal abort until the
2114 * final jbd2_journal_stop, which will receive the -EIO error.
2116 * Finally, the jbd2_journal_abort call allows the caller to supply an errno
2117 * which will be recorded (if possible) in the journal superblock. This
2118 * allows a client to record failure conditions in the middle of a
2119 * transaction without having to complete the transaction to record the
2120 * failure to disk. ext3_error, for example, now uses this
2123 * Errors which originate from within the journaling layer will NOT
2124 * supply an errno; a null errno implies that absolutely no further
2125 * writes are done to the journal (unless there are any already in
2130 void jbd2_journal_abort(journal_t *journal, int errno)
2132 __journal_abort_soft(journal, errno);
2136 * int jbd2_journal_errno () - returns the journal's error state.
2137 * @journal: journal to examine.
2139 * This is the errno number set with jbd2_journal_abort(), the last
2140 * time the journal was mounted - if the journal was stopped
2141 * without calling abort this will be 0.
2143 * If the journal has been aborted on this mount time -EROFS will
2146 int jbd2_journal_errno(journal_t *journal)
2150 read_lock(&journal->j_state_lock);
2151 if (journal->j_flags & JBD2_ABORT)
2154 err = journal->j_errno;
2155 read_unlock(&journal->j_state_lock);
2160 * int jbd2_journal_clear_err () - clears the journal's error state
2161 * @journal: journal to act on.
2163 * An error must be cleared or acked to take a FS out of readonly
2166 int jbd2_journal_clear_err(journal_t *journal)
2170 write_lock(&journal->j_state_lock);
2171 if (journal->j_flags & JBD2_ABORT)
2174 journal->j_errno = 0;
2175 write_unlock(&journal->j_state_lock);
2180 * void jbd2_journal_ack_err() - Ack journal err.
2181 * @journal: journal to act on.
2183 * An error must be cleared or acked to take a FS out of readonly
2186 void jbd2_journal_ack_err(journal_t *journal)
2188 write_lock(&journal->j_state_lock);
2189 if (journal->j_errno)
2190 journal->j_flags |= JBD2_ACK_ERR;
2191 write_unlock(&journal->j_state_lock);
2194 int jbd2_journal_blocks_per_page(struct inode *inode)
2196 return 1 << (PAGE_CACHE_SHIFT - inode->i_sb->s_blocksize_bits);
2200 * helper functions to deal with 32 or 64bit block numbers.
2202 size_t journal_tag_bytes(journal_t *journal)
2206 if (JBD2_HAS_INCOMPAT_FEATURE(journal, JBD2_FEATURE_INCOMPAT_CSUM_V3))
2207 return sizeof(journal_block_tag3_t);
2209 sz = sizeof(journal_block_tag_t);
2211 if (JBD2_HAS_INCOMPAT_FEATURE(journal, JBD2_FEATURE_INCOMPAT_CSUM_V2))
2212 sz += sizeof(__u16);
2214 if (JBD2_HAS_INCOMPAT_FEATURE(journal, JBD2_FEATURE_INCOMPAT_64BIT))
2217 return sz - sizeof(__u32);
2221 * JBD memory management
2223 * These functions are used to allocate block-sized chunks of memory
2224 * used for making copies of buffer_head data. Very often it will be
2225 * page-sized chunks of data, but sometimes it will be in
2226 * sub-page-size chunks. (For example, 16k pages on Power systems
2227 * with a 4k block file system.) For blocks smaller than a page, we
2228 * use a SLAB allocator. There are slab caches for each block size,
2229 * which are allocated at mount time, if necessary, and we only free
2230 * (all of) the slab caches when/if the jbd2 module is unloaded. For
2231 * this reason we don't need to a mutex to protect access to
2232 * jbd2_slab[] allocating or releasing memory; only in
2233 * jbd2_journal_create_slab().
2235 #define JBD2_MAX_SLABS 8
2236 static struct kmem_cache *jbd2_slab[JBD2_MAX_SLABS];
2238 static const char *jbd2_slab_names[JBD2_MAX_SLABS] = {
2239 "jbd2_1k", "jbd2_2k", "jbd2_4k", "jbd2_8k",
2240 "jbd2_16k", "jbd2_32k", "jbd2_64k", "jbd2_128k"
2244 static void jbd2_journal_destroy_slabs(void)
2248 for (i = 0; i < JBD2_MAX_SLABS; i++) {
2250 kmem_cache_destroy(jbd2_slab[i]);
2251 jbd2_slab[i] = NULL;
2255 static int jbd2_journal_create_slab(size_t size)
2257 static DEFINE_MUTEX(jbd2_slab_create_mutex);
2258 int i = order_base_2(size) - 10;
2261 if (size == PAGE_SIZE)
2264 if (i >= JBD2_MAX_SLABS)
2267 if (unlikely(i < 0))
2269 mutex_lock(&jbd2_slab_create_mutex);
2271 mutex_unlock(&jbd2_slab_create_mutex);
2272 return 0; /* Already created */
2275 slab_size = 1 << (i+10);
2276 jbd2_slab[i] = kmem_cache_create(jbd2_slab_names[i], slab_size,
2277 slab_size, 0, NULL);
2278 mutex_unlock(&jbd2_slab_create_mutex);
2279 if (!jbd2_slab[i]) {
2280 printk(KERN_EMERG "JBD2: no memory for jbd2_slab cache\n");
2286 static struct kmem_cache *get_slab(size_t size)
2288 int i = order_base_2(size) - 10;
2290 BUG_ON(i >= JBD2_MAX_SLABS);
2291 if (unlikely(i < 0))
2293 BUG_ON(jbd2_slab[i] == NULL);
2294 return jbd2_slab[i];
2297 void *jbd2_alloc(size_t size, gfp_t flags)
2301 BUG_ON(size & (size-1)); /* Must be a power of 2 */
2303 flags |= __GFP_REPEAT;
2304 if (size == PAGE_SIZE)
2305 ptr = (void *)__get_free_pages(flags, 0);
2306 else if (size > PAGE_SIZE) {
2307 int order = get_order(size);
2310 ptr = (void *)__get_free_pages(flags, order);
2312 ptr = vmalloc(size);
2314 ptr = kmem_cache_alloc(get_slab(size), flags);
2316 /* Check alignment; SLUB has gotten this wrong in the past,
2317 * and this can lead to user data corruption! */
2318 BUG_ON(((unsigned long) ptr) & (size-1));
2323 void jbd2_free(void *ptr, size_t size)
2325 if (size == PAGE_SIZE) {
2326 free_pages((unsigned long)ptr, 0);
2329 if (size > PAGE_SIZE) {
2330 int order = get_order(size);
2333 free_pages((unsigned long)ptr, order);
2338 kmem_cache_free(get_slab(size), ptr);
2342 * Journal_head storage management
2344 static struct kmem_cache *jbd2_journal_head_cache;
2345 #ifdef CONFIG_JBD2_DEBUG
2346 static atomic_t nr_journal_heads = ATOMIC_INIT(0);
2349 static int jbd2_journal_init_journal_head_cache(void)
2353 J_ASSERT(jbd2_journal_head_cache == NULL);
2354 jbd2_journal_head_cache = kmem_cache_create("jbd2_journal_head",
2355 sizeof(struct journal_head),
2357 SLAB_TEMPORARY, /* flags */
2360 if (!jbd2_journal_head_cache) {
2362 printk(KERN_EMERG "JBD2: no memory for journal_head cache\n");
2367 static void jbd2_journal_destroy_journal_head_cache(void)
2369 if (jbd2_journal_head_cache) {
2370 kmem_cache_destroy(jbd2_journal_head_cache);
2371 jbd2_journal_head_cache = NULL;
2376 * journal_head splicing and dicing
2378 static struct journal_head *journal_alloc_journal_head(void)
2380 struct journal_head *ret;
2382 #ifdef CONFIG_JBD2_DEBUG
2383 atomic_inc(&nr_journal_heads);
2385 ret = kmem_cache_zalloc(jbd2_journal_head_cache, GFP_NOFS);
2387 jbd_debug(1, "out of memory for journal_head\n");
2388 pr_notice_ratelimited("ENOMEM in %s, retrying.\n", __func__);
2391 ret = kmem_cache_zalloc(jbd2_journal_head_cache, GFP_NOFS);
2397 static void journal_free_journal_head(struct journal_head *jh)
2399 #ifdef CONFIG_JBD2_DEBUG
2400 atomic_dec(&nr_journal_heads);
2401 memset(jh, JBD2_POISON_FREE, sizeof(*jh));
2403 kmem_cache_free(jbd2_journal_head_cache, jh);
2407 * A journal_head is attached to a buffer_head whenever JBD has an
2408 * interest in the buffer.
2410 * Whenever a buffer has an attached journal_head, its ->b_state:BH_JBD bit
2411 * is set. This bit is tested in core kernel code where we need to take
2412 * JBD-specific actions. Testing the zeroness of ->b_private is not reliable
2415 * When a buffer has its BH_JBD bit set, its ->b_count is elevated by one.
2417 * When a buffer has its BH_JBD bit set it is immune from being released by
2418 * core kernel code, mainly via ->b_count.
2420 * A journal_head is detached from its buffer_head when the journal_head's
2421 * b_jcount reaches zero. Running transaction (b_transaction) and checkpoint
2422 * transaction (b_cp_transaction) hold their references to b_jcount.
2424 * Various places in the kernel want to attach a journal_head to a buffer_head
2425 * _before_ attaching the journal_head to a transaction. To protect the
2426 * journal_head in this situation, jbd2_journal_add_journal_head elevates the
2427 * journal_head's b_jcount refcount by one. The caller must call
2428 * jbd2_journal_put_journal_head() to undo this.
2430 * So the typical usage would be:
2432 * (Attach a journal_head if needed. Increments b_jcount)
2433 * struct journal_head *jh = jbd2_journal_add_journal_head(bh);
2435 * (Get another reference for transaction)
2436 * jbd2_journal_grab_journal_head(bh);
2437 * jh->b_transaction = xxx;
2438 * (Put original reference)
2439 * jbd2_journal_put_journal_head(jh);
2443 * Give a buffer_head a journal_head.
2447 struct journal_head *jbd2_journal_add_journal_head(struct buffer_head *bh)
2449 struct journal_head *jh;
2450 struct journal_head *new_jh = NULL;
2453 if (!buffer_jbd(bh))
2454 new_jh = journal_alloc_journal_head();
2456 jbd_lock_bh_journal_head(bh);
2457 if (buffer_jbd(bh)) {
2461 (atomic_read(&bh->b_count) > 0) ||
2462 (bh->b_page && bh->b_page->mapping));
2465 jbd_unlock_bh_journal_head(bh);
2470 new_jh = NULL; /* We consumed it */
2475 BUFFER_TRACE(bh, "added journal_head");
2478 jbd_unlock_bh_journal_head(bh);
2480 journal_free_journal_head(new_jh);
2481 return bh->b_private;
2485 * Grab a ref against this buffer_head's journal_head. If it ended up not
2486 * having a journal_head, return NULL
2488 struct journal_head *jbd2_journal_grab_journal_head(struct buffer_head *bh)
2490 struct journal_head *jh = NULL;
2492 jbd_lock_bh_journal_head(bh);
2493 if (buffer_jbd(bh)) {
2497 jbd_unlock_bh_journal_head(bh);
2501 static void __journal_remove_journal_head(struct buffer_head *bh)
2503 struct journal_head *jh = bh2jh(bh);
2505 J_ASSERT_JH(jh, jh->b_jcount >= 0);
2506 J_ASSERT_JH(jh, jh->b_transaction == NULL);
2507 J_ASSERT_JH(jh, jh->b_next_transaction == NULL);
2508 J_ASSERT_JH(jh, jh->b_cp_transaction == NULL);
2509 J_ASSERT_JH(jh, jh->b_jlist == BJ_None);
2510 J_ASSERT_BH(bh, buffer_jbd(bh));
2511 J_ASSERT_BH(bh, jh2bh(jh) == bh);
2512 BUFFER_TRACE(bh, "remove journal_head");
2513 if (jh->b_frozen_data) {
2514 printk(KERN_WARNING "%s: freeing b_frozen_data\n", __func__);
2515 jbd2_free(jh->b_frozen_data, bh->b_size);
2517 if (jh->b_committed_data) {
2518 printk(KERN_WARNING "%s: freeing b_committed_data\n", __func__);
2519 jbd2_free(jh->b_committed_data, bh->b_size);
2521 bh->b_private = NULL;
2522 jh->b_bh = NULL; /* debug, really */
2523 clear_buffer_jbd(bh);
2524 journal_free_journal_head(jh);
2528 * Drop a reference on the passed journal_head. If it fell to zero then
2529 * release the journal_head from the buffer_head.
2531 void jbd2_journal_put_journal_head(struct journal_head *jh)
2533 struct buffer_head *bh = jh2bh(jh);
2535 jbd_lock_bh_journal_head(bh);
2536 J_ASSERT_JH(jh, jh->b_jcount > 0);
2538 if (!jh->b_jcount) {
2539 __journal_remove_journal_head(bh);
2540 jbd_unlock_bh_journal_head(bh);
2543 jbd_unlock_bh_journal_head(bh);
2547 * Initialize jbd inode head
2549 void jbd2_journal_init_jbd_inode(struct jbd2_inode *jinode, struct inode *inode)
2551 jinode->i_transaction = NULL;
2552 jinode->i_next_transaction = NULL;
2553 jinode->i_vfs_inode = inode;
2554 jinode->i_flags = 0;
2555 INIT_LIST_HEAD(&jinode->i_list);
2559 * Function to be called before we start removing inode from memory (i.e.,
2560 * clear_inode() is a fine place to be called from). It removes inode from
2561 * transaction's lists.
2563 void jbd2_journal_release_jbd_inode(journal_t *journal,
2564 struct jbd2_inode *jinode)
2569 spin_lock(&journal->j_list_lock);
2570 /* Is commit writing out inode - we have to wait */
2571 if (test_bit(__JI_COMMIT_RUNNING, &jinode->i_flags)) {
2572 wait_queue_head_t *wq;
2573 DEFINE_WAIT_BIT(wait, &jinode->i_flags, __JI_COMMIT_RUNNING);
2574 wq = bit_waitqueue(&jinode->i_flags, __JI_COMMIT_RUNNING);
2575 prepare_to_wait(wq, &wait.wait, TASK_UNINTERRUPTIBLE);
2576 spin_unlock(&journal->j_list_lock);
2578 finish_wait(wq, &wait.wait);
2582 if (jinode->i_transaction) {
2583 list_del(&jinode->i_list);
2584 jinode->i_transaction = NULL;
2586 spin_unlock(&journal->j_list_lock);
2590 #ifdef CONFIG_PROC_FS
2592 #define JBD2_STATS_PROC_NAME "fs/jbd2"
2594 static void __init jbd2_create_jbd_stats_proc_entry(void)
2596 proc_jbd2_stats = proc_mkdir(JBD2_STATS_PROC_NAME, NULL);
2599 static void __exit jbd2_remove_jbd_stats_proc_entry(void)
2601 if (proc_jbd2_stats)
2602 remove_proc_entry(JBD2_STATS_PROC_NAME, NULL);
2607 #define jbd2_create_jbd_stats_proc_entry() do {} while (0)
2608 #define jbd2_remove_jbd_stats_proc_entry() do {} while (0)
2612 struct kmem_cache *jbd2_handle_cache, *jbd2_inode_cache;
2614 static int __init jbd2_journal_init_handle_cache(void)
2616 jbd2_handle_cache = KMEM_CACHE(jbd2_journal_handle, SLAB_TEMPORARY);
2617 if (jbd2_handle_cache == NULL) {
2618 printk(KERN_EMERG "JBD2: failed to create handle cache\n");
2621 jbd2_inode_cache = KMEM_CACHE(jbd2_inode, 0);
2622 if (jbd2_inode_cache == NULL) {
2623 printk(KERN_EMERG "JBD2: failed to create inode cache\n");
2624 kmem_cache_destroy(jbd2_handle_cache);
2630 static void jbd2_journal_destroy_handle_cache(void)
2632 if (jbd2_handle_cache)
2633 kmem_cache_destroy(jbd2_handle_cache);
2634 if (jbd2_inode_cache)
2635 kmem_cache_destroy(jbd2_inode_cache);
2640 * Module startup and shutdown
2643 static int __init journal_init_caches(void)
2647 ret = jbd2_journal_init_revoke_caches();
2649 ret = jbd2_journal_init_journal_head_cache();
2651 ret = jbd2_journal_init_handle_cache();
2653 ret = jbd2_journal_init_transaction_cache();
2657 static void jbd2_journal_destroy_caches(void)
2659 jbd2_journal_destroy_revoke_caches();
2660 jbd2_journal_destroy_journal_head_cache();
2661 jbd2_journal_destroy_handle_cache();
2662 jbd2_journal_destroy_transaction_cache();
2663 jbd2_journal_destroy_slabs();
2666 static int __init journal_init(void)
2670 BUILD_BUG_ON(sizeof(struct journal_superblock_s) != 1024);
2672 ret = journal_init_caches();
2674 jbd2_create_jbd_stats_proc_entry();
2676 jbd2_journal_destroy_caches();
2681 static void __exit journal_exit(void)
2683 #ifdef CONFIG_JBD2_DEBUG
2684 int n = atomic_read(&nr_journal_heads);
2686 printk(KERN_ERR "JBD2: leaked %d journal_heads!\n", n);
2688 jbd2_remove_jbd_stats_proc_entry();
2689 jbd2_journal_destroy_caches();
2692 MODULE_LICENSE("GPL");
2693 module_init(journal_init);
2694 module_exit(journal_exit);