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 /* Checksumming functions */
107 int jbd2_verify_csum_type(journal_t *j, journal_superblock_t *sb)
109 if (!JBD2_HAS_INCOMPAT_FEATURE(j, JBD2_FEATURE_INCOMPAT_CSUM_V2))
112 return sb->s_checksum_type == JBD2_CRC32C_CHKSUM;
115 static __u32 jbd2_superblock_csum(journal_t *j, journal_superblock_t *sb)
117 __u32 csum, old_csum;
119 old_csum = sb->s_checksum;
121 csum = jbd2_chksum(j, ~0, (char *)sb, sizeof(journal_superblock_t));
122 sb->s_checksum = old_csum;
124 return cpu_to_be32(csum);
127 int jbd2_superblock_csum_verify(journal_t *j, journal_superblock_t *sb)
129 if (!JBD2_HAS_INCOMPAT_FEATURE(j, JBD2_FEATURE_INCOMPAT_CSUM_V2))
132 return sb->s_checksum == jbd2_superblock_csum(j, sb);
135 void jbd2_superblock_csum_set(journal_t *j, journal_superblock_t *sb)
137 if (!JBD2_HAS_INCOMPAT_FEATURE(j, JBD2_FEATURE_INCOMPAT_CSUM_V2))
140 sb->s_checksum = jbd2_superblock_csum(j, sb);
144 * Helper function used to manage commit timeouts
147 static void commit_timeout(unsigned long __data)
149 struct task_struct * p = (struct task_struct *) __data;
155 * kjournald2: The main thread function used to manage a logging device
158 * This kernel thread is responsible for two things:
160 * 1) COMMIT: Every so often we need to commit the current state of the
161 * filesystem to disk. The journal thread is responsible for writing
162 * all of the metadata buffers to disk.
164 * 2) CHECKPOINT: We cannot reuse a used section of the log file until all
165 * of the data in that part of the log has been rewritten elsewhere on
166 * the disk. Flushing these old buffers to reclaim space in the log is
167 * known as checkpointing, and this thread is responsible for that job.
170 static int kjournald2(void *arg)
172 journal_t *journal = arg;
173 transaction_t *transaction;
176 * Set up an interval timer which can be used to trigger a commit wakeup
177 * after the commit interval expires
179 setup_timer(&journal->j_commit_timer, commit_timeout,
180 (unsigned long)current);
184 /* Record that the journal thread is running */
185 journal->j_task = current;
186 wake_up(&journal->j_wait_done_commit);
189 * And now, wait forever for commit wakeup events.
191 write_lock(&journal->j_state_lock);
194 if (journal->j_flags & JBD2_UNMOUNT)
197 jbd_debug(1, "commit_sequence=%d, commit_request=%d\n",
198 journal->j_commit_sequence, journal->j_commit_request);
200 if (journal->j_commit_sequence != journal->j_commit_request) {
201 jbd_debug(1, "OK, requests differ\n");
202 write_unlock(&journal->j_state_lock);
203 del_timer_sync(&journal->j_commit_timer);
204 jbd2_journal_commit_transaction(journal);
205 write_lock(&journal->j_state_lock);
209 wake_up(&journal->j_wait_done_commit);
210 if (freezing(current)) {
212 * The simpler the better. Flushing journal isn't a
213 * good idea, because that depends on threads that may
214 * be already stopped.
216 jbd_debug(1, "Now suspending kjournald2\n");
217 write_unlock(&journal->j_state_lock);
219 write_lock(&journal->j_state_lock);
222 * We assume on resume that commits are already there,
226 int should_sleep = 1;
228 prepare_to_wait(&journal->j_wait_commit, &wait,
230 if (journal->j_commit_sequence != journal->j_commit_request)
232 transaction = journal->j_running_transaction;
233 if (transaction && time_after_eq(jiffies,
234 transaction->t_expires))
236 if (journal->j_flags & JBD2_UNMOUNT)
239 write_unlock(&journal->j_state_lock);
241 write_lock(&journal->j_state_lock);
243 finish_wait(&journal->j_wait_commit, &wait);
246 jbd_debug(1, "kjournald2 wakes\n");
249 * Were we woken up by a commit wakeup event?
251 transaction = journal->j_running_transaction;
252 if (transaction && time_after_eq(jiffies, transaction->t_expires)) {
253 journal->j_commit_request = transaction->t_tid;
254 jbd_debug(1, "woke because of timeout\n");
259 write_unlock(&journal->j_state_lock);
260 del_timer_sync(&journal->j_commit_timer);
261 journal->j_task = NULL;
262 wake_up(&journal->j_wait_done_commit);
263 jbd_debug(1, "Journal thread exiting.\n");
267 static int jbd2_journal_start_thread(journal_t *journal)
269 struct task_struct *t;
271 t = kthread_run(kjournald2, journal, "jbd2/%s",
276 wait_event(journal->j_wait_done_commit, journal->j_task != NULL);
280 static void journal_kill_thread(journal_t *journal)
282 write_lock(&journal->j_state_lock);
283 journal->j_flags |= JBD2_UNMOUNT;
285 while (journal->j_task) {
286 wake_up(&journal->j_wait_commit);
287 write_unlock(&journal->j_state_lock);
288 wait_event(journal->j_wait_done_commit, journal->j_task == NULL);
289 write_lock(&journal->j_state_lock);
291 write_unlock(&journal->j_state_lock);
295 * jbd2_journal_write_metadata_buffer: write a metadata buffer to the journal.
297 * Writes a metadata buffer to a given disk block. The actual IO is not
298 * performed but a new buffer_head is constructed which labels the data
299 * to be written with the correct destination disk block.
301 * Any magic-number escaping which needs to be done will cause a
302 * copy-out here. If the buffer happens to start with the
303 * JBD2_MAGIC_NUMBER, then we can't write it to the log directly: the
304 * magic number is only written to the log for descripter blocks. In
305 * this case, we copy the data and replace the first word with 0, and we
306 * return a result code which indicates that this buffer needs to be
307 * marked as an escaped buffer in the corresponding log descriptor
308 * block. The missing word can then be restored when the block is read
311 * If the source buffer has already been modified by a new transaction
312 * since we took the last commit snapshot, we use the frozen copy of
313 * that data for IO. If we end up using the existing buffer_head's data
314 * for the write, then we have to make sure nobody modifies it while the
315 * IO is in progress. do_get_write_access() handles this.
317 * The function returns a pointer to the buffer_head to be used for IO.
325 * Bit 0 set == escape performed on the data
326 * Bit 1 set == buffer copy-out performed (kfree the data after IO)
329 int jbd2_journal_write_metadata_buffer(transaction_t *transaction,
330 struct journal_head *jh_in,
331 struct buffer_head **bh_out,
334 int need_copy_out = 0;
335 int done_copy_out = 0;
338 struct buffer_head *new_bh;
339 struct page *new_page;
340 unsigned int new_offset;
341 struct buffer_head *bh_in = jh2bh(jh_in);
342 journal_t *journal = transaction->t_journal;
345 * The buffer really shouldn't be locked: only the current committing
346 * transaction is allowed to write it, so nobody else is allowed
349 * akpm: except if we're journalling data, and write() output is
350 * also part of a shared mapping, and another thread has
351 * decided to launch a writepage() against this buffer.
353 J_ASSERT_BH(bh_in, buffer_jbddirty(bh_in));
356 new_bh = alloc_buffer_head(GFP_NOFS);
359 * Failure is not an option, but __GFP_NOFAIL is going
360 * away; so we retry ourselves here.
362 congestion_wait(BLK_RW_ASYNC, HZ/50);
366 /* keep subsequent assertions sane */
367 atomic_set(&new_bh->b_count, 1);
369 jbd_lock_bh_state(bh_in);
372 * If a new transaction has already done a buffer copy-out, then
373 * we use that version of the data for the commit.
375 if (jh_in->b_frozen_data) {
377 new_page = virt_to_page(jh_in->b_frozen_data);
378 new_offset = offset_in_page(jh_in->b_frozen_data);
380 new_page = jh2bh(jh_in)->b_page;
381 new_offset = offset_in_page(jh2bh(jh_in)->b_data);
384 mapped_data = kmap_atomic(new_page);
386 * Fire data frozen trigger if data already wasn't frozen. Do this
387 * before checking for escaping, as the trigger may modify the magic
388 * offset. If a copy-out happens afterwards, it will have the correct
389 * data in the buffer.
392 jbd2_buffer_frozen_trigger(jh_in, mapped_data + new_offset,
398 if (*((__be32 *)(mapped_data + new_offset)) ==
399 cpu_to_be32(JBD2_MAGIC_NUMBER)) {
403 kunmap_atomic(mapped_data);
406 * Do we need to do a data copy?
408 if (need_copy_out && !done_copy_out) {
411 jbd_unlock_bh_state(bh_in);
412 tmp = jbd2_alloc(bh_in->b_size, GFP_NOFS);
417 jbd_lock_bh_state(bh_in);
418 if (jh_in->b_frozen_data) {
419 jbd2_free(tmp, bh_in->b_size);
423 jh_in->b_frozen_data = tmp;
424 mapped_data = kmap_atomic(new_page);
425 memcpy(tmp, mapped_data + new_offset, bh_in->b_size);
426 kunmap_atomic(mapped_data);
428 new_page = virt_to_page(tmp);
429 new_offset = offset_in_page(tmp);
433 * This isn't strictly necessary, as we're using frozen
434 * data for the escaping, but it keeps consistency with
435 * b_frozen_data usage.
437 jh_in->b_frozen_triggers = jh_in->b_triggers;
441 * Did we need to do an escaping? Now we've done all the
442 * copying, we can finally do so.
445 mapped_data = kmap_atomic(new_page);
446 *((unsigned int *)(mapped_data + new_offset)) = 0;
447 kunmap_atomic(mapped_data);
450 set_bh_page(new_bh, new_page, new_offset);
451 new_bh->b_size = bh_in->b_size;
452 new_bh->b_bdev = journal->j_dev;
453 new_bh->b_blocknr = blocknr;
454 set_buffer_mapped(new_bh);
455 set_buffer_dirty(new_bh);
460 * The to-be-written buffer needs to get moved to the io queue,
461 * and the original buffer whose contents we are shadowing or
462 * copying is moved to the transaction's shadow queue.
464 JBUFFER_TRACE(jh_in, "file as BJ_Shadow");
465 spin_lock(&journal->j_list_lock);
466 __jbd2_journal_file_buffer(jh_in, transaction, BJ_Shadow);
467 spin_unlock(&journal->j_list_lock);
468 jbd_unlock_bh_state(bh_in);
470 return do_escape | (done_copy_out << 1);
474 * Allocation code for the journal file. Manage the space left in the
475 * journal, so that we can begin checkpointing when appropriate.
479 * __jbd2_log_space_left: Return the number of free blocks left in the journal.
481 * Called with the journal already locked.
483 * Called under j_state_lock
486 int __jbd2_log_space_left(journal_t *journal)
488 int left = journal->j_free;
490 /* assert_spin_locked(&journal->j_state_lock); */
493 * Be pessimistic here about the number of those free blocks which
494 * might be required for log descriptor control blocks.
497 #define MIN_LOG_RESERVED_BLOCKS 32 /* Allow for rounding errors */
499 left -= MIN_LOG_RESERVED_BLOCKS;
508 * Called with j_state_lock locked for writing.
509 * Returns true if a transaction commit was started.
511 int __jbd2_log_start_commit(journal_t *journal, tid_t target)
513 /* Return if the txn has already requested to be committed */
514 if (journal->j_commit_request == target)
518 * The only transaction we can possibly wait upon is the
519 * currently running transaction (if it exists). Otherwise,
520 * the target tid must be an old one.
522 if (journal->j_running_transaction &&
523 journal->j_running_transaction->t_tid == target) {
525 * We want a new commit: OK, mark the request and wakeup the
526 * commit thread. We do _not_ do the commit ourselves.
529 journal->j_commit_request = target;
530 jbd_debug(1, "JBD2: requesting commit %d/%d\n",
531 journal->j_commit_request,
532 journal->j_commit_sequence);
533 journal->j_running_transaction->t_requested = jiffies;
534 wake_up(&journal->j_wait_commit);
536 } else if (!tid_geq(journal->j_commit_request, target))
537 /* This should never happen, but if it does, preserve
538 the evidence before kjournald goes into a loop and
539 increments j_commit_sequence beyond all recognition. */
540 WARN_ONCE(1, "JBD2: bad log_start_commit: %u %u %u %u\n",
541 journal->j_commit_request,
542 journal->j_commit_sequence,
543 target, journal->j_running_transaction ?
544 journal->j_running_transaction->t_tid : 0);
548 int jbd2_log_start_commit(journal_t *journal, tid_t tid)
552 write_lock(&journal->j_state_lock);
553 ret = __jbd2_log_start_commit(journal, tid);
554 write_unlock(&journal->j_state_lock);
559 * Force and wait upon a commit if the calling process is not within
560 * transaction. This is used for forcing out undo-protected data which contains
561 * bitmaps, when the fs is running out of space.
563 * We can only force the running transaction if we don't have an active handle;
564 * otherwise, we will deadlock.
566 * Returns true if a transaction was started.
568 int jbd2_journal_force_commit_nested(journal_t *journal)
570 transaction_t *transaction = NULL;
572 int need_to_start = 0;
574 read_lock(&journal->j_state_lock);
575 if (journal->j_running_transaction && !current->journal_info) {
576 transaction = journal->j_running_transaction;
577 if (!tid_geq(journal->j_commit_request, transaction->t_tid))
579 } else if (journal->j_committing_transaction)
580 transaction = journal->j_committing_transaction;
583 read_unlock(&journal->j_state_lock);
584 return 0; /* Nothing to retry */
587 tid = transaction->t_tid;
588 read_unlock(&journal->j_state_lock);
590 jbd2_log_start_commit(journal, tid);
591 jbd2_log_wait_commit(journal, tid);
596 * Start a commit of the current running transaction (if any). Returns true
597 * if a transaction is going to be committed (or is currently already
598 * committing), and fills its tid in at *ptid
600 int jbd2_journal_start_commit(journal_t *journal, tid_t *ptid)
604 write_lock(&journal->j_state_lock);
605 if (journal->j_running_transaction) {
606 tid_t tid = journal->j_running_transaction->t_tid;
608 __jbd2_log_start_commit(journal, tid);
609 /* There's a running transaction and we've just made sure
610 * it's commit has been scheduled. */
614 } else if (journal->j_committing_transaction) {
616 * If commit has been started, then we have to wait for
617 * completion of that transaction.
620 *ptid = journal->j_committing_transaction->t_tid;
623 write_unlock(&journal->j_state_lock);
628 * Return 1 if a given transaction has not yet sent barrier request
629 * connected with a transaction commit. If 0 is returned, transaction
630 * may or may not have sent the barrier. Used to avoid sending barrier
631 * twice in common cases.
633 int jbd2_trans_will_send_data_barrier(journal_t *journal, tid_t tid)
636 transaction_t *commit_trans;
638 if (!(journal->j_flags & JBD2_BARRIER))
640 read_lock(&journal->j_state_lock);
641 /* Transaction already committed? */
642 if (tid_geq(journal->j_commit_sequence, tid))
644 commit_trans = journal->j_committing_transaction;
645 if (!commit_trans || commit_trans->t_tid != tid) {
650 * Transaction is being committed and we already proceeded to
651 * submitting a flush to fs partition?
653 if (journal->j_fs_dev != journal->j_dev) {
654 if (!commit_trans->t_need_data_flush ||
655 commit_trans->t_state >= T_COMMIT_DFLUSH)
658 if (commit_trans->t_state >= T_COMMIT_JFLUSH)
663 read_unlock(&journal->j_state_lock);
666 EXPORT_SYMBOL(jbd2_trans_will_send_data_barrier);
669 * Wait for a specified commit to complete.
670 * The caller may not hold the journal lock.
672 int jbd2_log_wait_commit(journal_t *journal, tid_t tid)
676 read_lock(&journal->j_state_lock);
677 #ifdef CONFIG_JBD2_DEBUG
678 if (!tid_geq(journal->j_commit_request, tid)) {
680 "%s: error: j_commit_request=%d, tid=%d\n",
681 __func__, journal->j_commit_request, tid);
684 while (tid_gt(tid, journal->j_commit_sequence)) {
685 jbd_debug(1, "JBD2: want %d, j_commit_sequence=%d\n",
686 tid, journal->j_commit_sequence);
687 wake_up(&journal->j_wait_commit);
688 read_unlock(&journal->j_state_lock);
689 wait_event(journal->j_wait_done_commit,
690 !tid_gt(tid, journal->j_commit_sequence));
691 read_lock(&journal->j_state_lock);
693 read_unlock(&journal->j_state_lock);
695 if (unlikely(is_journal_aborted(journal))) {
696 printk(KERN_EMERG "journal commit I/O error\n");
703 * When this function returns the transaction corresponding to tid
704 * will be completed. If the transaction has currently running, start
705 * committing that transaction before waiting for it to complete. If
706 * the transaction id is stale, it is by definition already completed,
707 * so just return SUCCESS.
709 int jbd2_complete_transaction(journal_t *journal, tid_t tid)
711 int need_to_wait = 1;
713 read_lock(&journal->j_state_lock);
714 if (journal->j_running_transaction &&
715 journal->j_running_transaction->t_tid == tid) {
716 if (journal->j_commit_request != tid) {
717 /* transaction not yet started, so request it */
718 read_unlock(&journal->j_state_lock);
719 jbd2_log_start_commit(journal, tid);
722 } else if (!(journal->j_committing_transaction &&
723 journal->j_committing_transaction->t_tid == tid))
725 read_unlock(&journal->j_state_lock);
729 return jbd2_log_wait_commit(journal, tid);
731 EXPORT_SYMBOL(jbd2_complete_transaction);
734 * Log buffer allocation routines:
737 int jbd2_journal_next_log_block(journal_t *journal, unsigned long long *retp)
739 unsigned long blocknr;
741 write_lock(&journal->j_state_lock);
742 J_ASSERT(journal->j_free > 1);
744 blocknr = journal->j_head;
747 if (journal->j_head == journal->j_last)
748 journal->j_head = journal->j_first;
749 write_unlock(&journal->j_state_lock);
750 return jbd2_journal_bmap(journal, blocknr, retp);
754 * Conversion of logical to physical block numbers for the journal
756 * On external journals the journal blocks are identity-mapped, so
757 * this is a no-op. If needed, we can use j_blk_offset - everything is
760 int jbd2_journal_bmap(journal_t *journal, unsigned long blocknr,
761 unsigned long long *retp)
764 unsigned long long ret;
766 if (journal->j_inode) {
767 ret = bmap(journal->j_inode, blocknr);
771 printk(KERN_ALERT "%s: journal block not found "
772 "at offset %lu on %s\n",
773 __func__, blocknr, journal->j_devname);
775 __journal_abort_soft(journal, err);
778 *retp = blocknr; /* +journal->j_blk_offset */
784 * We play buffer_head aliasing tricks to write data/metadata blocks to
785 * the journal without copying their contents, but for journal
786 * descriptor blocks we do need to generate bona fide buffers.
788 * After the caller of jbd2_journal_get_descriptor_buffer() has finished modifying
789 * the buffer's contents they really should run flush_dcache_page(bh->b_page).
790 * But we don't bother doing that, so there will be coherency problems with
791 * mmaps of blockdevs which hold live JBD-controlled filesystems.
793 struct journal_head *jbd2_journal_get_descriptor_buffer(journal_t *journal)
795 struct buffer_head *bh;
796 unsigned long long blocknr;
799 err = jbd2_journal_next_log_block(journal, &blocknr);
804 bh = __getblk(journal->j_dev, blocknr, journal->j_blocksize);
808 memset(bh->b_data, 0, journal->j_blocksize);
809 set_buffer_uptodate(bh);
811 BUFFER_TRACE(bh, "return this buffer");
812 return jbd2_journal_add_journal_head(bh);
816 * Return tid of the oldest transaction in the journal and block in the journal
817 * where the transaction starts.
819 * If the journal is now empty, return which will be the next transaction ID
820 * we will write and where will that transaction start.
822 * The return value is 0 if journal tail cannot be pushed any further, 1 if
825 int jbd2_journal_get_log_tail(journal_t *journal, tid_t *tid,
826 unsigned long *block)
828 transaction_t *transaction;
831 read_lock(&journal->j_state_lock);
832 spin_lock(&journal->j_list_lock);
833 transaction = journal->j_checkpoint_transactions;
835 *tid = transaction->t_tid;
836 *block = transaction->t_log_start;
837 } else if ((transaction = journal->j_committing_transaction) != NULL) {
838 *tid = transaction->t_tid;
839 *block = transaction->t_log_start;
840 } else if ((transaction = journal->j_running_transaction) != NULL) {
841 *tid = transaction->t_tid;
842 *block = journal->j_head;
844 *tid = journal->j_transaction_sequence;
845 *block = journal->j_head;
847 ret = tid_gt(*tid, journal->j_tail_sequence);
848 spin_unlock(&journal->j_list_lock);
849 read_unlock(&journal->j_state_lock);
855 * Update information in journal structure and in on disk journal superblock
856 * about log tail. This function does not check whether information passed in
857 * really pushes log tail further. It's responsibility of the caller to make
858 * sure provided log tail information is valid (e.g. by holding
859 * j_checkpoint_mutex all the time between computing log tail and calling this
860 * function as is the case with jbd2_cleanup_journal_tail()).
862 * Requires j_checkpoint_mutex
864 void __jbd2_update_log_tail(journal_t *journal, tid_t tid, unsigned long block)
868 BUG_ON(!mutex_is_locked(&journal->j_checkpoint_mutex));
871 * We cannot afford for write to remain in drive's caches since as
872 * soon as we update j_tail, next transaction can start reusing journal
873 * space and if we lose sb update during power failure we'd replay
874 * old transaction with possibly newly overwritten data.
876 jbd2_journal_update_sb_log_tail(journal, tid, block, WRITE_FUA);
877 write_lock(&journal->j_state_lock);
878 freed = block - journal->j_tail;
879 if (block < journal->j_tail)
880 freed += journal->j_last - journal->j_first;
882 trace_jbd2_update_log_tail(journal, tid, block, freed);
884 "Cleaning journal tail from %d to %d (offset %lu), "
886 journal->j_tail_sequence, tid, block, freed);
888 journal->j_free += freed;
889 journal->j_tail_sequence = tid;
890 journal->j_tail = block;
891 write_unlock(&journal->j_state_lock);
895 * This is a variaon of __jbd2_update_log_tail which checks for validity of
896 * provided log tail and locks j_checkpoint_mutex. So it is safe against races
897 * with other threads updating log tail.
899 void jbd2_update_log_tail(journal_t *journal, tid_t tid, unsigned long block)
901 mutex_lock(&journal->j_checkpoint_mutex);
902 if (tid_gt(tid, journal->j_tail_sequence))
903 __jbd2_update_log_tail(journal, tid, block);
904 mutex_unlock(&journal->j_checkpoint_mutex);
907 struct jbd2_stats_proc_session {
909 struct transaction_stats_s *stats;
914 static void *jbd2_seq_info_start(struct seq_file *seq, loff_t *pos)
916 return *pos ? NULL : SEQ_START_TOKEN;
919 static void *jbd2_seq_info_next(struct seq_file *seq, void *v, loff_t *pos)
924 static int jbd2_seq_info_show(struct seq_file *seq, void *v)
926 struct jbd2_stats_proc_session *s = seq->private;
928 if (v != SEQ_START_TOKEN)
930 seq_printf(seq, "%lu transactions (%lu requested), "
931 "each up to %u blocks\n",
932 s->stats->ts_tid, s->stats->ts_requested,
933 s->journal->j_max_transaction_buffers);
934 if (s->stats->ts_tid == 0)
936 seq_printf(seq, "average: \n %ums waiting for transaction\n",
937 jiffies_to_msecs(s->stats->run.rs_wait / s->stats->ts_tid));
938 seq_printf(seq, " %ums request delay\n",
939 (s->stats->ts_requested == 0) ? 0 :
940 jiffies_to_msecs(s->stats->run.rs_request_delay /
941 s->stats->ts_requested));
942 seq_printf(seq, " %ums running transaction\n",
943 jiffies_to_msecs(s->stats->run.rs_running / s->stats->ts_tid));
944 seq_printf(seq, " %ums transaction was being locked\n",
945 jiffies_to_msecs(s->stats->run.rs_locked / s->stats->ts_tid));
946 seq_printf(seq, " %ums flushing data (in ordered mode)\n",
947 jiffies_to_msecs(s->stats->run.rs_flushing / s->stats->ts_tid));
948 seq_printf(seq, " %ums logging transaction\n",
949 jiffies_to_msecs(s->stats->run.rs_logging / s->stats->ts_tid));
950 seq_printf(seq, " %lluus average transaction commit time\n",
951 div_u64(s->journal->j_average_commit_time, 1000));
952 seq_printf(seq, " %lu handles per transaction\n",
953 s->stats->run.rs_handle_count / s->stats->ts_tid);
954 seq_printf(seq, " %lu blocks per transaction\n",
955 s->stats->run.rs_blocks / s->stats->ts_tid);
956 seq_printf(seq, " %lu logged blocks per transaction\n",
957 s->stats->run.rs_blocks_logged / s->stats->ts_tid);
961 static void jbd2_seq_info_stop(struct seq_file *seq, void *v)
965 static const struct seq_operations jbd2_seq_info_ops = {
966 .start = jbd2_seq_info_start,
967 .next = jbd2_seq_info_next,
968 .stop = jbd2_seq_info_stop,
969 .show = jbd2_seq_info_show,
972 static int jbd2_seq_info_open(struct inode *inode, struct file *file)
974 journal_t *journal = PDE_DATA(inode);
975 struct jbd2_stats_proc_session *s;
978 s = kmalloc(sizeof(*s), GFP_KERNEL);
981 size = sizeof(struct transaction_stats_s);
982 s->stats = kmalloc(size, GFP_KERNEL);
983 if (s->stats == NULL) {
987 spin_lock(&journal->j_history_lock);
988 memcpy(s->stats, &journal->j_stats, size);
989 s->journal = journal;
990 spin_unlock(&journal->j_history_lock);
992 rc = seq_open(file, &jbd2_seq_info_ops);
994 struct seq_file *m = file->private_data;
1004 static int jbd2_seq_info_release(struct inode *inode, struct file *file)
1006 struct seq_file *seq = file->private_data;
1007 struct jbd2_stats_proc_session *s = seq->private;
1010 return seq_release(inode, file);
1013 static const struct file_operations jbd2_seq_info_fops = {
1014 .owner = THIS_MODULE,
1015 .open = jbd2_seq_info_open,
1017 .llseek = seq_lseek,
1018 .release = jbd2_seq_info_release,
1021 static struct proc_dir_entry *proc_jbd2_stats;
1023 static void jbd2_stats_proc_init(journal_t *journal)
1025 journal->j_proc_entry = proc_mkdir(journal->j_devname, proc_jbd2_stats);
1026 if (journal->j_proc_entry) {
1027 proc_create_data("info", S_IRUGO, journal->j_proc_entry,
1028 &jbd2_seq_info_fops, journal);
1032 static void jbd2_stats_proc_exit(journal_t *journal)
1034 remove_proc_entry("info", journal->j_proc_entry);
1035 remove_proc_entry(journal->j_devname, proc_jbd2_stats);
1039 * Management for journal control blocks: functions to create and
1040 * destroy journal_t structures, and to initialise and read existing
1041 * journal blocks from disk. */
1043 /* First: create and setup a journal_t object in memory. We initialise
1044 * very few fields yet: that has to wait until we have created the
1045 * journal structures from from scratch, or loaded them from disk. */
1047 static journal_t * journal_init_common (void)
1052 journal = kzalloc(sizeof(*journal), GFP_KERNEL);
1056 init_waitqueue_head(&journal->j_wait_transaction_locked);
1057 init_waitqueue_head(&journal->j_wait_logspace);
1058 init_waitqueue_head(&journal->j_wait_done_commit);
1059 init_waitqueue_head(&journal->j_wait_checkpoint);
1060 init_waitqueue_head(&journal->j_wait_commit);
1061 init_waitqueue_head(&journal->j_wait_updates);
1062 mutex_init(&journal->j_barrier);
1063 mutex_init(&journal->j_checkpoint_mutex);
1064 spin_lock_init(&journal->j_revoke_lock);
1065 spin_lock_init(&journal->j_list_lock);
1066 rwlock_init(&journal->j_state_lock);
1068 journal->j_commit_interval = (HZ * JBD2_DEFAULT_MAX_COMMIT_AGE);
1069 journal->j_min_batch_time = 0;
1070 journal->j_max_batch_time = 15000; /* 15ms */
1072 /* The journal is marked for error until we succeed with recovery! */
1073 journal->j_flags = JBD2_ABORT;
1075 /* Set up a default-sized revoke table for the new mount. */
1076 err = jbd2_journal_init_revoke(journal, JOURNAL_REVOKE_DEFAULT_HASH);
1082 spin_lock_init(&journal->j_history_lock);
1087 /* jbd2_journal_init_dev and jbd2_journal_init_inode:
1089 * Create a journal structure assigned some fixed set of disk blocks to
1090 * the journal. We don't actually touch those disk blocks yet, but we
1091 * need to set up all of the mapping information to tell the journaling
1092 * system where the journal blocks are.
1097 * journal_t * jbd2_journal_init_dev() - creates and initialises a journal structure
1098 * @bdev: Block device on which to create the journal
1099 * @fs_dev: Device which hold journalled filesystem for this journal.
1100 * @start: Block nr Start of journal.
1101 * @len: Length of the journal in blocks.
1102 * @blocksize: blocksize of journalling device
1104 * Returns: a newly created journal_t *
1106 * jbd2_journal_init_dev creates a journal which maps a fixed contiguous
1107 * range of blocks on an arbitrary block device.
1110 journal_t * jbd2_journal_init_dev(struct block_device *bdev,
1111 struct block_device *fs_dev,
1112 unsigned long long start, int len, int blocksize)
1114 journal_t *journal = journal_init_common();
1115 struct buffer_head *bh;
1122 /* journal descriptor can store up to n blocks -bzzz */
1123 journal->j_blocksize = blocksize;
1124 journal->j_dev = bdev;
1125 journal->j_fs_dev = fs_dev;
1126 journal->j_blk_offset = start;
1127 journal->j_maxlen = len;
1128 bdevname(journal->j_dev, journal->j_devname);
1129 p = journal->j_devname;
1130 while ((p = strchr(p, '/')))
1132 jbd2_stats_proc_init(journal);
1133 n = journal->j_blocksize / sizeof(journal_block_tag_t);
1134 journal->j_wbufsize = n;
1135 journal->j_wbuf = kmalloc(n * sizeof(struct buffer_head*), GFP_KERNEL);
1136 if (!journal->j_wbuf) {
1137 printk(KERN_ERR "%s: Can't allocate bhs for commit thread\n",
1142 bh = __getblk(journal->j_dev, start, journal->j_blocksize);
1145 "%s: Cannot get buffer for journal superblock\n",
1149 journal->j_sb_buffer = bh;
1150 journal->j_superblock = (journal_superblock_t *)bh->b_data;
1154 kfree(journal->j_wbuf);
1155 jbd2_stats_proc_exit(journal);
1161 * journal_t * jbd2_journal_init_inode () - creates a journal which maps to a inode.
1162 * @inode: An inode to create the journal in
1164 * jbd2_journal_init_inode creates a journal which maps an on-disk inode as
1165 * the journal. The inode must exist already, must support bmap() and
1166 * must have all data blocks preallocated.
1168 journal_t * jbd2_journal_init_inode (struct inode *inode)
1170 struct buffer_head *bh;
1171 journal_t *journal = journal_init_common();
1175 unsigned long long blocknr;
1180 journal->j_dev = journal->j_fs_dev = inode->i_sb->s_bdev;
1181 journal->j_inode = inode;
1182 bdevname(journal->j_dev, journal->j_devname);
1183 p = journal->j_devname;
1184 while ((p = strchr(p, '/')))
1186 p = journal->j_devname + strlen(journal->j_devname);
1187 sprintf(p, "-%lu", journal->j_inode->i_ino);
1189 "journal %p: inode %s/%ld, size %Ld, bits %d, blksize %ld\n",
1190 journal, inode->i_sb->s_id, inode->i_ino,
1191 (long long) inode->i_size,
1192 inode->i_sb->s_blocksize_bits, inode->i_sb->s_blocksize);
1194 journal->j_maxlen = inode->i_size >> inode->i_sb->s_blocksize_bits;
1195 journal->j_blocksize = inode->i_sb->s_blocksize;
1196 jbd2_stats_proc_init(journal);
1198 /* journal descriptor can store up to n blocks -bzzz */
1199 n = journal->j_blocksize / sizeof(journal_block_tag_t);
1200 journal->j_wbufsize = n;
1201 journal->j_wbuf = kmalloc(n * sizeof(struct buffer_head*), GFP_KERNEL);
1202 if (!journal->j_wbuf) {
1203 printk(KERN_ERR "%s: Can't allocate bhs for commit thread\n",
1208 err = jbd2_journal_bmap(journal, 0, &blocknr);
1209 /* If that failed, give up */
1211 printk(KERN_ERR "%s: Cannot locate journal superblock\n",
1216 bh = __getblk(journal->j_dev, blocknr, journal->j_blocksize);
1219 "%s: Cannot get buffer for journal superblock\n",
1223 journal->j_sb_buffer = bh;
1224 journal->j_superblock = (journal_superblock_t *)bh->b_data;
1228 kfree(journal->j_wbuf);
1229 jbd2_stats_proc_exit(journal);
1235 * If the journal init or create aborts, we need to mark the journal
1236 * superblock as being NULL to prevent the journal destroy from writing
1237 * back a bogus superblock.
1239 static void journal_fail_superblock (journal_t *journal)
1241 struct buffer_head *bh = journal->j_sb_buffer;
1243 journal->j_sb_buffer = NULL;
1247 * Given a journal_t structure, initialise the various fields for
1248 * startup of a new journaling session. We use this both when creating
1249 * a journal, and after recovering an old journal to reset it for
1253 static int journal_reset(journal_t *journal)
1255 journal_superblock_t *sb = journal->j_superblock;
1256 unsigned long long first, last;
1258 first = be32_to_cpu(sb->s_first);
1259 last = be32_to_cpu(sb->s_maxlen);
1260 if (first + JBD2_MIN_JOURNAL_BLOCKS > last + 1) {
1261 printk(KERN_ERR "JBD2: Journal too short (blocks %llu-%llu).\n",
1263 journal_fail_superblock(journal);
1267 journal->j_first = first;
1268 journal->j_last = last;
1270 journal->j_head = first;
1271 journal->j_tail = first;
1272 journal->j_free = last - first;
1274 journal->j_tail_sequence = journal->j_transaction_sequence;
1275 journal->j_commit_sequence = journal->j_transaction_sequence - 1;
1276 journal->j_commit_request = journal->j_commit_sequence;
1278 journal->j_max_transaction_buffers = journal->j_maxlen / 4;
1281 * As a special case, if the on-disk copy is already marked as needing
1282 * no recovery (s_start == 0), then we can safely defer the superblock
1283 * update until the next commit by setting JBD2_FLUSHED. This avoids
1284 * attempting a write to a potential-readonly device.
1286 if (sb->s_start == 0) {
1287 jbd_debug(1, "JBD2: Skipping superblock update on recovered sb "
1288 "(start %ld, seq %d, errno %d)\n",
1289 journal->j_tail, journal->j_tail_sequence,
1291 journal->j_flags |= JBD2_FLUSHED;
1293 /* Lock here to make assertions happy... */
1294 mutex_lock(&journal->j_checkpoint_mutex);
1296 * Update log tail information. We use WRITE_FUA since new
1297 * transaction will start reusing journal space and so we
1298 * must make sure information about current log tail is on
1301 jbd2_journal_update_sb_log_tail(journal,
1302 journal->j_tail_sequence,
1305 mutex_unlock(&journal->j_checkpoint_mutex);
1307 return jbd2_journal_start_thread(journal);
1310 static void jbd2_write_superblock(journal_t *journal, int write_op)
1312 struct buffer_head *bh = journal->j_sb_buffer;
1315 trace_jbd2_write_superblock(journal, write_op);
1316 if (!(journal->j_flags & JBD2_BARRIER))
1317 write_op &= ~(REQ_FUA | REQ_FLUSH);
1319 if (buffer_write_io_error(bh)) {
1321 * Oh, dear. A previous attempt to write the journal
1322 * superblock failed. This could happen because the
1323 * USB device was yanked out. Or it could happen to
1324 * be a transient write error and maybe the block will
1325 * be remapped. Nothing we can do but to retry the
1326 * write and hope for the best.
1328 printk(KERN_ERR "JBD2: previous I/O error detected "
1329 "for journal superblock update for %s.\n",
1330 journal->j_devname);
1331 clear_buffer_write_io_error(bh);
1332 set_buffer_uptodate(bh);
1335 bh->b_end_io = end_buffer_write_sync;
1336 ret = submit_bh(write_op, bh);
1338 if (buffer_write_io_error(bh)) {
1339 clear_buffer_write_io_error(bh);
1340 set_buffer_uptodate(bh);
1344 printk(KERN_ERR "JBD2: Error %d detected when updating "
1345 "journal superblock for %s.\n", ret,
1346 journal->j_devname);
1351 * jbd2_journal_update_sb_log_tail() - Update log tail in journal sb on disk.
1352 * @journal: The journal to update.
1353 * @tail_tid: TID of the new transaction at the tail of the log
1354 * @tail_block: The first block of the transaction at the tail of the log
1355 * @write_op: With which operation should we write the journal sb
1357 * Update a journal's superblock information about log tail and write it to
1358 * disk, waiting for the IO to complete.
1360 void jbd2_journal_update_sb_log_tail(journal_t *journal, tid_t tail_tid,
1361 unsigned long tail_block, int write_op)
1363 journal_superblock_t *sb = journal->j_superblock;
1365 BUG_ON(!mutex_is_locked(&journal->j_checkpoint_mutex));
1366 jbd_debug(1, "JBD2: updating superblock (start %lu, seq %u)\n",
1367 tail_block, tail_tid);
1369 sb->s_sequence = cpu_to_be32(tail_tid);
1370 sb->s_start = cpu_to_be32(tail_block);
1372 jbd2_write_superblock(journal, write_op);
1374 /* Log is no longer empty */
1375 write_lock(&journal->j_state_lock);
1376 WARN_ON(!sb->s_sequence);
1377 journal->j_flags &= ~JBD2_FLUSHED;
1378 write_unlock(&journal->j_state_lock);
1382 * jbd2_mark_journal_empty() - Mark on disk journal as empty.
1383 * @journal: The journal to update.
1385 * Update a journal's dynamic superblock fields to show that journal is empty.
1386 * Write updated superblock to disk waiting for IO to complete.
1388 static void jbd2_mark_journal_empty(journal_t *journal)
1390 journal_superblock_t *sb = journal->j_superblock;
1392 BUG_ON(!mutex_is_locked(&journal->j_checkpoint_mutex));
1393 read_lock(&journal->j_state_lock);
1394 /* Is it already empty? */
1395 if (sb->s_start == 0) {
1396 read_unlock(&journal->j_state_lock);
1399 jbd_debug(1, "JBD2: Marking journal as empty (seq %d)\n",
1400 journal->j_tail_sequence);
1402 sb->s_sequence = cpu_to_be32(journal->j_tail_sequence);
1403 sb->s_start = cpu_to_be32(0);
1404 read_unlock(&journal->j_state_lock);
1406 jbd2_write_superblock(journal, WRITE_FUA);
1408 /* Log is no longer empty */
1409 write_lock(&journal->j_state_lock);
1410 journal->j_flags |= JBD2_FLUSHED;
1411 write_unlock(&journal->j_state_lock);
1416 * jbd2_journal_update_sb_errno() - Update error in the journal.
1417 * @journal: The journal to update.
1419 * Update a journal's errno. Write updated superblock to disk waiting for IO
1422 void jbd2_journal_update_sb_errno(journal_t *journal)
1424 journal_superblock_t *sb = journal->j_superblock;
1426 read_lock(&journal->j_state_lock);
1427 jbd_debug(1, "JBD2: updating superblock error (errno %d)\n",
1429 sb->s_errno = cpu_to_be32(journal->j_errno);
1430 jbd2_superblock_csum_set(journal, sb);
1431 read_unlock(&journal->j_state_lock);
1433 jbd2_write_superblock(journal, WRITE_SYNC);
1435 EXPORT_SYMBOL(jbd2_journal_update_sb_errno);
1438 * Read the superblock for a given journal, performing initial
1439 * validation of the format.
1441 static int journal_get_superblock(journal_t *journal)
1443 struct buffer_head *bh;
1444 journal_superblock_t *sb;
1447 bh = journal->j_sb_buffer;
1449 J_ASSERT(bh != NULL);
1450 if (!buffer_uptodate(bh)) {
1451 ll_rw_block(READ, 1, &bh);
1453 if (!buffer_uptodate(bh)) {
1455 "JBD2: IO error reading journal superblock\n");
1460 if (buffer_verified(bh))
1463 sb = journal->j_superblock;
1467 if (sb->s_header.h_magic != cpu_to_be32(JBD2_MAGIC_NUMBER) ||
1468 sb->s_blocksize != cpu_to_be32(journal->j_blocksize)) {
1469 printk(KERN_WARNING "JBD2: no valid journal superblock found\n");
1473 switch(be32_to_cpu(sb->s_header.h_blocktype)) {
1474 case JBD2_SUPERBLOCK_V1:
1475 journal->j_format_version = 1;
1477 case JBD2_SUPERBLOCK_V2:
1478 journal->j_format_version = 2;
1481 printk(KERN_WARNING "JBD2: unrecognised superblock format ID\n");
1485 if (be32_to_cpu(sb->s_maxlen) < journal->j_maxlen)
1486 journal->j_maxlen = be32_to_cpu(sb->s_maxlen);
1487 else if (be32_to_cpu(sb->s_maxlen) > journal->j_maxlen) {
1488 printk(KERN_WARNING "JBD2: journal file too short\n");
1492 if (be32_to_cpu(sb->s_first) == 0 ||
1493 be32_to_cpu(sb->s_first) >= journal->j_maxlen) {
1495 "JBD2: Invalid start block of journal: %u\n",
1496 be32_to_cpu(sb->s_first));
1500 if (JBD2_HAS_COMPAT_FEATURE(journal, JBD2_FEATURE_COMPAT_CHECKSUM) &&
1501 JBD2_HAS_INCOMPAT_FEATURE(journal, JBD2_FEATURE_INCOMPAT_CSUM_V2)) {
1502 /* Can't have checksum v1 and v2 on at the same time! */
1503 printk(KERN_ERR "JBD: Can't enable checksumming v1 and v2 "
1504 "at the same time!\n");
1508 if (!jbd2_verify_csum_type(journal, sb)) {
1509 printk(KERN_ERR "JBD: Unknown checksum type\n");
1513 /* Load the checksum driver */
1514 if (JBD2_HAS_INCOMPAT_FEATURE(journal, JBD2_FEATURE_INCOMPAT_CSUM_V2)) {
1515 journal->j_chksum_driver = crypto_alloc_shash("crc32c", 0, 0);
1516 if (IS_ERR(journal->j_chksum_driver)) {
1517 printk(KERN_ERR "JBD: Cannot load crc32c driver.\n");
1518 err = PTR_ERR(journal->j_chksum_driver);
1519 journal->j_chksum_driver = NULL;
1524 /* Check superblock checksum */
1525 if (!jbd2_superblock_csum_verify(journal, sb)) {
1526 printk(KERN_ERR "JBD: journal checksum error\n");
1530 /* Precompute checksum seed for all metadata */
1531 if (JBD2_HAS_INCOMPAT_FEATURE(journal, JBD2_FEATURE_INCOMPAT_CSUM_V2))
1532 journal->j_csum_seed = jbd2_chksum(journal, ~0, sb->s_uuid,
1533 sizeof(sb->s_uuid));
1535 set_buffer_verified(bh);
1540 journal_fail_superblock(journal);
1545 * Load the on-disk journal superblock and read the key fields into the
1549 static int load_superblock(journal_t *journal)
1552 journal_superblock_t *sb;
1554 err = journal_get_superblock(journal);
1558 sb = journal->j_superblock;
1560 journal->j_tail_sequence = be32_to_cpu(sb->s_sequence);
1561 journal->j_tail = be32_to_cpu(sb->s_start);
1562 journal->j_first = be32_to_cpu(sb->s_first);
1563 journal->j_last = be32_to_cpu(sb->s_maxlen);
1564 journal->j_errno = be32_to_cpu(sb->s_errno);
1571 * int jbd2_journal_load() - Read journal from disk.
1572 * @journal: Journal to act on.
1574 * Given a journal_t structure which tells us which disk blocks contain
1575 * a journal, read the journal from disk to initialise the in-memory
1578 int jbd2_journal_load(journal_t *journal)
1581 journal_superblock_t *sb;
1583 err = load_superblock(journal);
1587 sb = journal->j_superblock;
1588 /* If this is a V2 superblock, then we have to check the
1589 * features flags on it. */
1591 if (journal->j_format_version >= 2) {
1592 if ((sb->s_feature_ro_compat &
1593 ~cpu_to_be32(JBD2_KNOWN_ROCOMPAT_FEATURES)) ||
1594 (sb->s_feature_incompat &
1595 ~cpu_to_be32(JBD2_KNOWN_INCOMPAT_FEATURES))) {
1597 "JBD2: Unrecognised features on journal\n");
1603 * Create a slab for this blocksize
1605 err = jbd2_journal_create_slab(be32_to_cpu(sb->s_blocksize));
1609 /* Let the recovery code check whether it needs to recover any
1610 * data from the journal. */
1611 if (jbd2_journal_recover(journal))
1612 goto recovery_error;
1614 if (journal->j_failed_commit) {
1615 printk(KERN_ERR "JBD2: journal transaction %u on %s "
1616 "is corrupt.\n", journal->j_failed_commit,
1617 journal->j_devname);
1621 /* OK, we've finished with the dynamic journal bits:
1622 * reinitialise the dynamic contents of the superblock in memory
1623 * and reset them on disk. */
1624 if (journal_reset(journal))
1625 goto recovery_error;
1627 journal->j_flags &= ~JBD2_ABORT;
1628 journal->j_flags |= JBD2_LOADED;
1632 printk(KERN_WARNING "JBD2: recovery failed\n");
1637 * void jbd2_journal_destroy() - Release a journal_t structure.
1638 * @journal: Journal to act on.
1640 * Release a journal_t structure once it is no longer in use by the
1642 * Return <0 if we couldn't clean up the journal.
1644 int jbd2_journal_destroy(journal_t *journal)
1648 /* Wait for the commit thread to wake up and die. */
1649 journal_kill_thread(journal);
1651 /* Force a final log commit */
1652 if (journal->j_running_transaction)
1653 jbd2_journal_commit_transaction(journal);
1655 /* Force any old transactions to disk */
1657 /* Totally anal locking here... */
1658 spin_lock(&journal->j_list_lock);
1659 while (journal->j_checkpoint_transactions != NULL) {
1660 spin_unlock(&journal->j_list_lock);
1661 mutex_lock(&journal->j_checkpoint_mutex);
1662 jbd2_log_do_checkpoint(journal);
1663 mutex_unlock(&journal->j_checkpoint_mutex);
1664 spin_lock(&journal->j_list_lock);
1667 J_ASSERT(journal->j_running_transaction == NULL);
1668 J_ASSERT(journal->j_committing_transaction == NULL);
1669 J_ASSERT(journal->j_checkpoint_transactions == NULL);
1670 spin_unlock(&journal->j_list_lock);
1672 if (journal->j_sb_buffer) {
1673 if (!is_journal_aborted(journal)) {
1674 mutex_lock(&journal->j_checkpoint_mutex);
1675 jbd2_mark_journal_empty(journal);
1676 mutex_unlock(&journal->j_checkpoint_mutex);
1679 brelse(journal->j_sb_buffer);
1682 if (journal->j_proc_entry)
1683 jbd2_stats_proc_exit(journal);
1684 if (journal->j_inode)
1685 iput(journal->j_inode);
1686 if (journal->j_revoke)
1687 jbd2_journal_destroy_revoke(journal);
1688 if (journal->j_chksum_driver)
1689 crypto_free_shash(journal->j_chksum_driver);
1690 kfree(journal->j_wbuf);
1698 *int jbd2_journal_check_used_features () - Check if features specified are used.
1699 * @journal: Journal to check.
1700 * @compat: bitmask of compatible features
1701 * @ro: bitmask of features that force read-only mount
1702 * @incompat: bitmask of incompatible features
1704 * Check whether the journal uses all of a given set of
1705 * features. Return true (non-zero) if it does.
1708 int jbd2_journal_check_used_features (journal_t *journal, unsigned long compat,
1709 unsigned long ro, unsigned long incompat)
1711 journal_superblock_t *sb;
1713 if (!compat && !ro && !incompat)
1715 /* Load journal superblock if it is not loaded yet. */
1716 if (journal->j_format_version == 0 &&
1717 journal_get_superblock(journal) != 0)
1719 if (journal->j_format_version == 1)
1722 sb = journal->j_superblock;
1724 if (((be32_to_cpu(sb->s_feature_compat) & compat) == compat) &&
1725 ((be32_to_cpu(sb->s_feature_ro_compat) & ro) == ro) &&
1726 ((be32_to_cpu(sb->s_feature_incompat) & incompat) == incompat))
1733 * int jbd2_journal_check_available_features() - Check feature set in journalling layer
1734 * @journal: Journal to check.
1735 * @compat: bitmask of compatible features
1736 * @ro: bitmask of features that force read-only mount
1737 * @incompat: bitmask of incompatible features
1739 * Check whether the journaling code supports the use of
1740 * all of a given set of features on this journal. Return true
1741 * (non-zero) if it can. */
1743 int jbd2_journal_check_available_features (journal_t *journal, unsigned long compat,
1744 unsigned long ro, unsigned long incompat)
1746 if (!compat && !ro && !incompat)
1749 /* We can support any known requested features iff the
1750 * superblock is in version 2. Otherwise we fail to support any
1751 * extended sb features. */
1753 if (journal->j_format_version != 2)
1756 if ((compat & JBD2_KNOWN_COMPAT_FEATURES) == compat &&
1757 (ro & JBD2_KNOWN_ROCOMPAT_FEATURES) == ro &&
1758 (incompat & JBD2_KNOWN_INCOMPAT_FEATURES) == incompat)
1765 * int jbd2_journal_set_features () - Mark a given journal feature in the superblock
1766 * @journal: Journal to act on.
1767 * @compat: bitmask of compatible features
1768 * @ro: bitmask of features that force read-only mount
1769 * @incompat: bitmask of incompatible features
1771 * Mark a given journal feature as present on the
1772 * superblock. Returns true if the requested features could be set.
1776 int jbd2_journal_set_features (journal_t *journal, unsigned long compat,
1777 unsigned long ro, unsigned long incompat)
1779 #define INCOMPAT_FEATURE_ON(f) \
1780 ((incompat & (f)) && !(sb->s_feature_incompat & cpu_to_be32(f)))
1781 #define COMPAT_FEATURE_ON(f) \
1782 ((compat & (f)) && !(sb->s_feature_compat & cpu_to_be32(f)))
1783 journal_superblock_t *sb;
1785 if (jbd2_journal_check_used_features(journal, compat, ro, incompat))
1788 if (!jbd2_journal_check_available_features(journal, compat, ro, incompat))
1791 /* Asking for checksumming v2 and v1? Only give them v2. */
1792 if (incompat & JBD2_FEATURE_INCOMPAT_CSUM_V2 &&
1793 compat & JBD2_FEATURE_COMPAT_CHECKSUM)
1794 compat &= ~JBD2_FEATURE_COMPAT_CHECKSUM;
1796 jbd_debug(1, "Setting new features 0x%lx/0x%lx/0x%lx\n",
1797 compat, ro, incompat);
1799 sb = journal->j_superblock;
1801 /* If enabling v2 checksums, update superblock */
1802 if (INCOMPAT_FEATURE_ON(JBD2_FEATURE_INCOMPAT_CSUM_V2)) {
1803 sb->s_checksum_type = JBD2_CRC32C_CHKSUM;
1804 sb->s_feature_compat &=
1805 ~cpu_to_be32(JBD2_FEATURE_COMPAT_CHECKSUM);
1807 /* Load the checksum driver */
1808 if (journal->j_chksum_driver == NULL) {
1809 journal->j_chksum_driver = crypto_alloc_shash("crc32c",
1811 if (IS_ERR(journal->j_chksum_driver)) {
1812 printk(KERN_ERR "JBD: Cannot load crc32c "
1814 journal->j_chksum_driver = NULL;
1819 /* Precompute checksum seed for all metadata */
1820 if (JBD2_HAS_INCOMPAT_FEATURE(journal,
1821 JBD2_FEATURE_INCOMPAT_CSUM_V2))
1822 journal->j_csum_seed = jbd2_chksum(journal, ~0,
1824 sizeof(sb->s_uuid));
1827 /* If enabling v1 checksums, downgrade superblock */
1828 if (COMPAT_FEATURE_ON(JBD2_FEATURE_COMPAT_CHECKSUM))
1829 sb->s_feature_incompat &=
1830 ~cpu_to_be32(JBD2_FEATURE_INCOMPAT_CSUM_V2);
1832 sb->s_feature_compat |= cpu_to_be32(compat);
1833 sb->s_feature_ro_compat |= cpu_to_be32(ro);
1834 sb->s_feature_incompat |= cpu_to_be32(incompat);
1837 #undef COMPAT_FEATURE_ON
1838 #undef INCOMPAT_FEATURE_ON
1842 * jbd2_journal_clear_features () - Clear a given journal feature in the
1844 * @journal: Journal to act on.
1845 * @compat: bitmask of compatible features
1846 * @ro: bitmask of features that force read-only mount
1847 * @incompat: bitmask of incompatible features
1849 * Clear a given journal feature as present on the
1852 void jbd2_journal_clear_features(journal_t *journal, unsigned long compat,
1853 unsigned long ro, unsigned long incompat)
1855 journal_superblock_t *sb;
1857 jbd_debug(1, "Clear features 0x%lx/0x%lx/0x%lx\n",
1858 compat, ro, incompat);
1860 sb = journal->j_superblock;
1862 sb->s_feature_compat &= ~cpu_to_be32(compat);
1863 sb->s_feature_ro_compat &= ~cpu_to_be32(ro);
1864 sb->s_feature_incompat &= ~cpu_to_be32(incompat);
1866 EXPORT_SYMBOL(jbd2_journal_clear_features);
1869 * int jbd2_journal_flush () - Flush journal
1870 * @journal: Journal to act on.
1872 * Flush all data for a given journal to disk and empty the journal.
1873 * Filesystems can use this when remounting readonly to ensure that
1874 * recovery does not need to happen on remount.
1877 int jbd2_journal_flush(journal_t *journal)
1880 transaction_t *transaction = NULL;
1882 write_lock(&journal->j_state_lock);
1884 /* Force everything buffered to the log... */
1885 if (journal->j_running_transaction) {
1886 transaction = journal->j_running_transaction;
1887 __jbd2_log_start_commit(journal, transaction->t_tid);
1888 } else if (journal->j_committing_transaction)
1889 transaction = journal->j_committing_transaction;
1891 /* Wait for the log commit to complete... */
1893 tid_t tid = transaction->t_tid;
1895 write_unlock(&journal->j_state_lock);
1896 jbd2_log_wait_commit(journal, tid);
1898 write_unlock(&journal->j_state_lock);
1901 /* ...and flush everything in the log out to disk. */
1902 spin_lock(&journal->j_list_lock);
1903 while (!err && journal->j_checkpoint_transactions != NULL) {
1904 spin_unlock(&journal->j_list_lock);
1905 mutex_lock(&journal->j_checkpoint_mutex);
1906 err = jbd2_log_do_checkpoint(journal);
1907 mutex_unlock(&journal->j_checkpoint_mutex);
1908 spin_lock(&journal->j_list_lock);
1910 spin_unlock(&journal->j_list_lock);
1912 if (is_journal_aborted(journal))
1915 mutex_lock(&journal->j_checkpoint_mutex);
1916 jbd2_cleanup_journal_tail(journal);
1918 /* Finally, mark the journal as really needing no recovery.
1919 * This sets s_start==0 in the underlying superblock, which is
1920 * the magic code for a fully-recovered superblock. Any future
1921 * commits of data to the journal will restore the current
1923 jbd2_mark_journal_empty(journal);
1924 mutex_unlock(&journal->j_checkpoint_mutex);
1925 write_lock(&journal->j_state_lock);
1926 J_ASSERT(!journal->j_running_transaction);
1927 J_ASSERT(!journal->j_committing_transaction);
1928 J_ASSERT(!journal->j_checkpoint_transactions);
1929 J_ASSERT(journal->j_head == journal->j_tail);
1930 J_ASSERT(journal->j_tail_sequence == journal->j_transaction_sequence);
1931 write_unlock(&journal->j_state_lock);
1936 * int jbd2_journal_wipe() - Wipe journal contents
1937 * @journal: Journal to act on.
1938 * @write: flag (see below)
1940 * Wipe out all of the contents of a journal, safely. This will produce
1941 * a warning if the journal contains any valid recovery information.
1942 * Must be called between journal_init_*() and jbd2_journal_load().
1944 * If 'write' is non-zero, then we wipe out the journal on disk; otherwise
1945 * we merely suppress recovery.
1948 int jbd2_journal_wipe(journal_t *journal, int write)
1952 J_ASSERT (!(journal->j_flags & JBD2_LOADED));
1954 err = load_superblock(journal);
1958 if (!journal->j_tail)
1961 printk(KERN_WARNING "JBD2: %s recovery information on journal\n",
1962 write ? "Clearing" : "Ignoring");
1964 err = jbd2_journal_skip_recovery(journal);
1966 /* Lock to make assertions happy... */
1967 mutex_lock(&journal->j_checkpoint_mutex);
1968 jbd2_mark_journal_empty(journal);
1969 mutex_unlock(&journal->j_checkpoint_mutex);
1977 * Journal abort has very specific semantics, which we describe
1978 * for journal abort.
1980 * Two internal functions, which provide abort to the jbd layer
1985 * Quick version for internal journal use (doesn't lock the journal).
1986 * Aborts hard --- we mark the abort as occurred, but do _nothing_ else,
1987 * and don't attempt to make any other journal updates.
1989 void __jbd2_journal_abort_hard(journal_t *journal)
1991 transaction_t *transaction;
1993 if (journal->j_flags & JBD2_ABORT)
1996 printk(KERN_ERR "Aborting journal on device %s.\n",
1997 journal->j_devname);
1999 write_lock(&journal->j_state_lock);
2000 journal->j_flags |= JBD2_ABORT;
2001 transaction = journal->j_running_transaction;
2003 __jbd2_log_start_commit(journal, transaction->t_tid);
2004 write_unlock(&journal->j_state_lock);
2007 /* Soft abort: record the abort error status in the journal superblock,
2008 * but don't do any other IO. */
2009 static void __journal_abort_soft (journal_t *journal, int errno)
2011 if (journal->j_flags & JBD2_ABORT)
2014 if (!journal->j_errno)
2015 journal->j_errno = errno;
2017 __jbd2_journal_abort_hard(journal);
2020 jbd2_journal_update_sb_errno(journal);
2024 * void jbd2_journal_abort () - Shutdown the journal immediately.
2025 * @journal: the journal to shutdown.
2026 * @errno: an error number to record in the journal indicating
2027 * the reason for the shutdown.
2029 * Perform a complete, immediate shutdown of the ENTIRE
2030 * journal (not of a single transaction). This operation cannot be
2031 * undone without closing and reopening the journal.
2033 * The jbd2_journal_abort function is intended to support higher level error
2034 * recovery mechanisms such as the ext2/ext3 remount-readonly error
2037 * Journal abort has very specific semantics. Any existing dirty,
2038 * unjournaled buffers in the main filesystem will still be written to
2039 * disk by bdflush, but the journaling mechanism will be suspended
2040 * immediately and no further transaction commits will be honoured.
2042 * Any dirty, journaled buffers will be written back to disk without
2043 * hitting the journal. Atomicity cannot be guaranteed on an aborted
2044 * filesystem, but we _do_ attempt to leave as much data as possible
2045 * behind for fsck to use for cleanup.
2047 * Any attempt to get a new transaction handle on a journal which is in
2048 * ABORT state will just result in an -EROFS error return. A
2049 * jbd2_journal_stop on an existing handle will return -EIO if we have
2050 * entered abort state during the update.
2052 * Recursive transactions are not disturbed by journal abort until the
2053 * final jbd2_journal_stop, which will receive the -EIO error.
2055 * Finally, the jbd2_journal_abort call allows the caller to supply an errno
2056 * which will be recorded (if possible) in the journal superblock. This
2057 * allows a client to record failure conditions in the middle of a
2058 * transaction without having to complete the transaction to record the
2059 * failure to disk. ext3_error, for example, now uses this
2062 * Errors which originate from within the journaling layer will NOT
2063 * supply an errno; a null errno implies that absolutely no further
2064 * writes are done to the journal (unless there are any already in
2069 void jbd2_journal_abort(journal_t *journal, int errno)
2071 __journal_abort_soft(journal, errno);
2075 * int jbd2_journal_errno () - returns the journal's error state.
2076 * @journal: journal to examine.
2078 * This is the errno number set with jbd2_journal_abort(), the last
2079 * time the journal was mounted - if the journal was stopped
2080 * without calling abort this will be 0.
2082 * If the journal has been aborted on this mount time -EROFS will
2085 int jbd2_journal_errno(journal_t *journal)
2089 read_lock(&journal->j_state_lock);
2090 if (journal->j_flags & JBD2_ABORT)
2093 err = journal->j_errno;
2094 read_unlock(&journal->j_state_lock);
2099 * int jbd2_journal_clear_err () - clears the journal's error state
2100 * @journal: journal to act on.
2102 * An error must be cleared or acked to take a FS out of readonly
2105 int jbd2_journal_clear_err(journal_t *journal)
2109 write_lock(&journal->j_state_lock);
2110 if (journal->j_flags & JBD2_ABORT)
2113 journal->j_errno = 0;
2114 write_unlock(&journal->j_state_lock);
2119 * void jbd2_journal_ack_err() - Ack journal err.
2120 * @journal: journal to act on.
2122 * An error must be cleared or acked to take a FS out of readonly
2125 void jbd2_journal_ack_err(journal_t *journal)
2127 write_lock(&journal->j_state_lock);
2128 if (journal->j_errno)
2129 journal->j_flags |= JBD2_ACK_ERR;
2130 write_unlock(&journal->j_state_lock);
2133 int jbd2_journal_blocks_per_page(struct inode *inode)
2135 return 1 << (PAGE_CACHE_SHIFT - inode->i_sb->s_blocksize_bits);
2139 * helper functions to deal with 32 or 64bit block numbers.
2141 size_t journal_tag_bytes(journal_t *journal)
2143 journal_block_tag_t tag;
2146 if (JBD2_HAS_INCOMPAT_FEATURE(journal, JBD2_FEATURE_INCOMPAT_CSUM_V2))
2147 x += sizeof(tag.t_checksum);
2149 if (JBD2_HAS_INCOMPAT_FEATURE(journal, JBD2_FEATURE_INCOMPAT_64BIT))
2150 return x + JBD2_TAG_SIZE64;
2152 return x + JBD2_TAG_SIZE32;
2156 * JBD memory management
2158 * These functions are used to allocate block-sized chunks of memory
2159 * used for making copies of buffer_head data. Very often it will be
2160 * page-sized chunks of data, but sometimes it will be in
2161 * sub-page-size chunks. (For example, 16k pages on Power systems
2162 * with a 4k block file system.) For blocks smaller than a page, we
2163 * use a SLAB allocator. There are slab caches for each block size,
2164 * which are allocated at mount time, if necessary, and we only free
2165 * (all of) the slab caches when/if the jbd2 module is unloaded. For
2166 * this reason we don't need to a mutex to protect access to
2167 * jbd2_slab[] allocating or releasing memory; only in
2168 * jbd2_journal_create_slab().
2170 #define JBD2_MAX_SLABS 8
2171 static struct kmem_cache *jbd2_slab[JBD2_MAX_SLABS];
2173 static const char *jbd2_slab_names[JBD2_MAX_SLABS] = {
2174 "jbd2_1k", "jbd2_2k", "jbd2_4k", "jbd2_8k",
2175 "jbd2_16k", "jbd2_32k", "jbd2_64k", "jbd2_128k"
2179 static void jbd2_journal_destroy_slabs(void)
2183 for (i = 0; i < JBD2_MAX_SLABS; i++) {
2185 kmem_cache_destroy(jbd2_slab[i]);
2186 jbd2_slab[i] = NULL;
2190 static int jbd2_journal_create_slab(size_t size)
2192 static DEFINE_MUTEX(jbd2_slab_create_mutex);
2193 int i = order_base_2(size) - 10;
2196 if (size == PAGE_SIZE)
2199 if (i >= JBD2_MAX_SLABS)
2202 if (unlikely(i < 0))
2204 mutex_lock(&jbd2_slab_create_mutex);
2206 mutex_unlock(&jbd2_slab_create_mutex);
2207 return 0; /* Already created */
2210 slab_size = 1 << (i+10);
2211 jbd2_slab[i] = kmem_cache_create(jbd2_slab_names[i], slab_size,
2212 slab_size, 0, NULL);
2213 mutex_unlock(&jbd2_slab_create_mutex);
2214 if (!jbd2_slab[i]) {
2215 printk(KERN_EMERG "JBD2: no memory for jbd2_slab cache\n");
2221 static struct kmem_cache *get_slab(size_t size)
2223 int i = order_base_2(size) - 10;
2225 BUG_ON(i >= JBD2_MAX_SLABS);
2226 if (unlikely(i < 0))
2228 BUG_ON(jbd2_slab[i] == NULL);
2229 return jbd2_slab[i];
2232 void *jbd2_alloc(size_t size, gfp_t flags)
2236 BUG_ON(size & (size-1)); /* Must be a power of 2 */
2238 flags |= __GFP_REPEAT;
2239 if (size == PAGE_SIZE)
2240 ptr = (void *)__get_free_pages(flags, 0);
2241 else if (size > PAGE_SIZE) {
2242 int order = get_order(size);
2245 ptr = (void *)__get_free_pages(flags, order);
2247 ptr = vmalloc(size);
2249 ptr = kmem_cache_alloc(get_slab(size), flags);
2251 /* Check alignment; SLUB has gotten this wrong in the past,
2252 * and this can lead to user data corruption! */
2253 BUG_ON(((unsigned long) ptr) & (size-1));
2258 void jbd2_free(void *ptr, size_t size)
2260 if (size == PAGE_SIZE) {
2261 free_pages((unsigned long)ptr, 0);
2264 if (size > PAGE_SIZE) {
2265 int order = get_order(size);
2268 free_pages((unsigned long)ptr, order);
2273 kmem_cache_free(get_slab(size), ptr);
2277 * Journal_head storage management
2279 static struct kmem_cache *jbd2_journal_head_cache;
2280 #ifdef CONFIG_JBD2_DEBUG
2281 static atomic_t nr_journal_heads = ATOMIC_INIT(0);
2284 static int jbd2_journal_init_journal_head_cache(void)
2288 J_ASSERT(jbd2_journal_head_cache == NULL);
2289 jbd2_journal_head_cache = kmem_cache_create("jbd2_journal_head",
2290 sizeof(struct journal_head),
2292 SLAB_TEMPORARY, /* flags */
2295 if (!jbd2_journal_head_cache) {
2297 printk(KERN_EMERG "JBD2: no memory for journal_head cache\n");
2302 static void jbd2_journal_destroy_journal_head_cache(void)
2304 if (jbd2_journal_head_cache) {
2305 kmem_cache_destroy(jbd2_journal_head_cache);
2306 jbd2_journal_head_cache = NULL;
2311 * journal_head splicing and dicing
2313 static struct journal_head *journal_alloc_journal_head(void)
2315 struct journal_head *ret;
2317 #ifdef CONFIG_JBD2_DEBUG
2318 atomic_inc(&nr_journal_heads);
2320 ret = kmem_cache_zalloc(jbd2_journal_head_cache, GFP_NOFS);
2322 jbd_debug(1, "out of memory for journal_head\n");
2323 pr_notice_ratelimited("ENOMEM in %s, retrying.\n", __func__);
2326 ret = kmem_cache_zalloc(jbd2_journal_head_cache, GFP_NOFS);
2332 static void journal_free_journal_head(struct journal_head *jh)
2334 #ifdef CONFIG_JBD2_DEBUG
2335 atomic_dec(&nr_journal_heads);
2336 memset(jh, JBD2_POISON_FREE, sizeof(*jh));
2338 kmem_cache_free(jbd2_journal_head_cache, jh);
2342 * A journal_head is attached to a buffer_head whenever JBD has an
2343 * interest in the buffer.
2345 * Whenever a buffer has an attached journal_head, its ->b_state:BH_JBD bit
2346 * is set. This bit is tested in core kernel code where we need to take
2347 * JBD-specific actions. Testing the zeroness of ->b_private is not reliable
2350 * When a buffer has its BH_JBD bit set, its ->b_count is elevated by one.
2352 * When a buffer has its BH_JBD bit set it is immune from being released by
2353 * core kernel code, mainly via ->b_count.
2355 * A journal_head is detached from its buffer_head when the journal_head's
2356 * b_jcount reaches zero. Running transaction (b_transaction) and checkpoint
2357 * transaction (b_cp_transaction) hold their references to b_jcount.
2359 * Various places in the kernel want to attach a journal_head to a buffer_head
2360 * _before_ attaching the journal_head to a transaction. To protect the
2361 * journal_head in this situation, jbd2_journal_add_journal_head elevates the
2362 * journal_head's b_jcount refcount by one. The caller must call
2363 * jbd2_journal_put_journal_head() to undo this.
2365 * So the typical usage would be:
2367 * (Attach a journal_head if needed. Increments b_jcount)
2368 * struct journal_head *jh = jbd2_journal_add_journal_head(bh);
2370 * (Get another reference for transaction)
2371 * jbd2_journal_grab_journal_head(bh);
2372 * jh->b_transaction = xxx;
2373 * (Put original reference)
2374 * jbd2_journal_put_journal_head(jh);
2378 * Give a buffer_head a journal_head.
2382 struct journal_head *jbd2_journal_add_journal_head(struct buffer_head *bh)
2384 struct journal_head *jh;
2385 struct journal_head *new_jh = NULL;
2388 if (!buffer_jbd(bh))
2389 new_jh = journal_alloc_journal_head();
2391 jbd_lock_bh_journal_head(bh);
2392 if (buffer_jbd(bh)) {
2396 (atomic_read(&bh->b_count) > 0) ||
2397 (bh->b_page && bh->b_page->mapping));
2400 jbd_unlock_bh_journal_head(bh);
2405 new_jh = NULL; /* We consumed it */
2410 BUFFER_TRACE(bh, "added journal_head");
2413 jbd_unlock_bh_journal_head(bh);
2415 journal_free_journal_head(new_jh);
2416 return bh->b_private;
2420 * Grab a ref against this buffer_head's journal_head. If it ended up not
2421 * having a journal_head, return NULL
2423 struct journal_head *jbd2_journal_grab_journal_head(struct buffer_head *bh)
2425 struct journal_head *jh = NULL;
2427 jbd_lock_bh_journal_head(bh);
2428 if (buffer_jbd(bh)) {
2432 jbd_unlock_bh_journal_head(bh);
2436 static void __journal_remove_journal_head(struct buffer_head *bh)
2438 struct journal_head *jh = bh2jh(bh);
2440 J_ASSERT_JH(jh, jh->b_jcount >= 0);
2441 J_ASSERT_JH(jh, jh->b_transaction == NULL);
2442 J_ASSERT_JH(jh, jh->b_next_transaction == NULL);
2443 J_ASSERT_JH(jh, jh->b_cp_transaction == NULL);
2444 J_ASSERT_JH(jh, jh->b_jlist == BJ_None);
2445 J_ASSERT_BH(bh, buffer_jbd(bh));
2446 J_ASSERT_BH(bh, jh2bh(jh) == bh);
2447 BUFFER_TRACE(bh, "remove journal_head");
2448 if (jh->b_frozen_data) {
2449 printk(KERN_WARNING "%s: freeing b_frozen_data\n", __func__);
2450 jbd2_free(jh->b_frozen_data, bh->b_size);
2452 if (jh->b_committed_data) {
2453 printk(KERN_WARNING "%s: freeing b_committed_data\n", __func__);
2454 jbd2_free(jh->b_committed_data, bh->b_size);
2456 bh->b_private = NULL;
2457 jh->b_bh = NULL; /* debug, really */
2458 clear_buffer_jbd(bh);
2459 journal_free_journal_head(jh);
2463 * Drop a reference on the passed journal_head. If it fell to zero then
2464 * release the journal_head from the buffer_head.
2466 void jbd2_journal_put_journal_head(struct journal_head *jh)
2468 struct buffer_head *bh = jh2bh(jh);
2470 jbd_lock_bh_journal_head(bh);
2471 J_ASSERT_JH(jh, jh->b_jcount > 0);
2473 if (!jh->b_jcount) {
2474 __journal_remove_journal_head(bh);
2475 jbd_unlock_bh_journal_head(bh);
2478 jbd_unlock_bh_journal_head(bh);
2482 * Initialize jbd inode head
2484 void jbd2_journal_init_jbd_inode(struct jbd2_inode *jinode, struct inode *inode)
2486 jinode->i_transaction = NULL;
2487 jinode->i_next_transaction = NULL;
2488 jinode->i_vfs_inode = inode;
2489 jinode->i_flags = 0;
2490 INIT_LIST_HEAD(&jinode->i_list);
2494 * Function to be called before we start removing inode from memory (i.e.,
2495 * clear_inode() is a fine place to be called from). It removes inode from
2496 * transaction's lists.
2498 void jbd2_journal_release_jbd_inode(journal_t *journal,
2499 struct jbd2_inode *jinode)
2504 spin_lock(&journal->j_list_lock);
2505 /* Is commit writing out inode - we have to wait */
2506 if (test_bit(__JI_COMMIT_RUNNING, &jinode->i_flags)) {
2507 wait_queue_head_t *wq;
2508 DEFINE_WAIT_BIT(wait, &jinode->i_flags, __JI_COMMIT_RUNNING);
2509 wq = bit_waitqueue(&jinode->i_flags, __JI_COMMIT_RUNNING);
2510 prepare_to_wait(wq, &wait.wait, TASK_UNINTERRUPTIBLE);
2511 spin_unlock(&journal->j_list_lock);
2513 finish_wait(wq, &wait.wait);
2517 if (jinode->i_transaction) {
2518 list_del(&jinode->i_list);
2519 jinode->i_transaction = NULL;
2521 spin_unlock(&journal->j_list_lock);
2525 #ifdef CONFIG_PROC_FS
2527 #define JBD2_STATS_PROC_NAME "fs/jbd2"
2529 static void __init jbd2_create_jbd_stats_proc_entry(void)
2531 proc_jbd2_stats = proc_mkdir(JBD2_STATS_PROC_NAME, NULL);
2534 static void __exit jbd2_remove_jbd_stats_proc_entry(void)
2536 if (proc_jbd2_stats)
2537 remove_proc_entry(JBD2_STATS_PROC_NAME, NULL);
2542 #define jbd2_create_jbd_stats_proc_entry() do {} while (0)
2543 #define jbd2_remove_jbd_stats_proc_entry() do {} while (0)
2547 struct kmem_cache *jbd2_handle_cache, *jbd2_inode_cache;
2549 static int __init jbd2_journal_init_handle_cache(void)
2551 jbd2_handle_cache = KMEM_CACHE(jbd2_journal_handle, SLAB_TEMPORARY);
2552 if (jbd2_handle_cache == NULL) {
2553 printk(KERN_EMERG "JBD2: failed to create handle cache\n");
2556 jbd2_inode_cache = KMEM_CACHE(jbd2_inode, 0);
2557 if (jbd2_inode_cache == NULL) {
2558 printk(KERN_EMERG "JBD2: failed to create inode cache\n");
2559 kmem_cache_destroy(jbd2_handle_cache);
2565 static void jbd2_journal_destroy_handle_cache(void)
2567 if (jbd2_handle_cache)
2568 kmem_cache_destroy(jbd2_handle_cache);
2569 if (jbd2_inode_cache)
2570 kmem_cache_destroy(jbd2_inode_cache);
2575 * Module startup and shutdown
2578 static int __init journal_init_caches(void)
2582 ret = jbd2_journal_init_revoke_caches();
2584 ret = jbd2_journal_init_journal_head_cache();
2586 ret = jbd2_journal_init_handle_cache();
2588 ret = jbd2_journal_init_transaction_cache();
2592 static void jbd2_journal_destroy_caches(void)
2594 jbd2_journal_destroy_revoke_caches();
2595 jbd2_journal_destroy_journal_head_cache();
2596 jbd2_journal_destroy_handle_cache();
2597 jbd2_journal_destroy_transaction_cache();
2598 jbd2_journal_destroy_slabs();
2601 static int __init journal_init(void)
2605 BUILD_BUG_ON(sizeof(struct journal_superblock_s) != 1024);
2607 ret = journal_init_caches();
2609 jbd2_create_jbd_stats_proc_entry();
2611 jbd2_journal_destroy_caches();
2616 static void __exit journal_exit(void)
2618 #ifdef CONFIG_JBD2_DEBUG
2619 int n = atomic_read(&nr_journal_heads);
2621 printk(KERN_EMERG "JBD2: leaked %d journal_heads!\n", n);
2623 jbd2_remove_jbd_stats_proc_entry();
2624 jbd2_journal_destroy_caches();
2627 MODULE_LICENSE("GPL");
2628 module_init(journal_init);
2629 module_exit(journal_exit);