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 void __jbd2_update_log_tail(journal_t *journal, tid_t tid, unsigned long block)
892 BUG_ON(!mutex_is_locked(&journal->j_checkpoint_mutex));
895 * We cannot afford for write to remain in drive's caches since as
896 * soon as we update j_tail, next transaction can start reusing journal
897 * space and if we lose sb update during power failure we'd replay
898 * old transaction with possibly newly overwritten data.
900 jbd2_journal_update_sb_log_tail(journal, tid, block, WRITE_FUA);
901 write_lock(&journal->j_state_lock);
902 freed = block - journal->j_tail;
903 if (block < journal->j_tail)
904 freed += journal->j_last - journal->j_first;
906 trace_jbd2_update_log_tail(journal, tid, block, freed);
908 "Cleaning journal tail from %d to %d (offset %lu), "
910 journal->j_tail_sequence, tid, block, freed);
912 journal->j_free += freed;
913 journal->j_tail_sequence = tid;
914 journal->j_tail = block;
915 write_unlock(&journal->j_state_lock);
919 * This is a variaon of __jbd2_update_log_tail which checks for validity of
920 * provided log tail and locks j_checkpoint_mutex. So it is safe against races
921 * with other threads updating log tail.
923 void jbd2_update_log_tail(journal_t *journal, tid_t tid, unsigned long block)
925 mutex_lock(&journal->j_checkpoint_mutex);
926 if (tid_gt(tid, journal->j_tail_sequence))
927 __jbd2_update_log_tail(journal, tid, block);
928 mutex_unlock(&journal->j_checkpoint_mutex);
931 struct jbd2_stats_proc_session {
933 struct transaction_stats_s *stats;
938 static void *jbd2_seq_info_start(struct seq_file *seq, loff_t *pos)
940 return *pos ? NULL : SEQ_START_TOKEN;
943 static void *jbd2_seq_info_next(struct seq_file *seq, void *v, loff_t *pos)
948 static int jbd2_seq_info_show(struct seq_file *seq, void *v)
950 struct jbd2_stats_proc_session *s = seq->private;
952 if (v != SEQ_START_TOKEN)
954 seq_printf(seq, "%lu transactions (%lu requested), "
955 "each up to %u blocks\n",
956 s->stats->ts_tid, s->stats->ts_requested,
957 s->journal->j_max_transaction_buffers);
958 if (s->stats->ts_tid == 0)
960 seq_printf(seq, "average: \n %ums waiting for transaction\n",
961 jiffies_to_msecs(s->stats->run.rs_wait / s->stats->ts_tid));
962 seq_printf(seq, " %ums request delay\n",
963 (s->stats->ts_requested == 0) ? 0 :
964 jiffies_to_msecs(s->stats->run.rs_request_delay /
965 s->stats->ts_requested));
966 seq_printf(seq, " %ums running transaction\n",
967 jiffies_to_msecs(s->stats->run.rs_running / s->stats->ts_tid));
968 seq_printf(seq, " %ums transaction was being locked\n",
969 jiffies_to_msecs(s->stats->run.rs_locked / s->stats->ts_tid));
970 seq_printf(seq, " %ums flushing data (in ordered mode)\n",
971 jiffies_to_msecs(s->stats->run.rs_flushing / s->stats->ts_tid));
972 seq_printf(seq, " %ums logging transaction\n",
973 jiffies_to_msecs(s->stats->run.rs_logging / s->stats->ts_tid));
974 seq_printf(seq, " %lluus average transaction commit time\n",
975 div_u64(s->journal->j_average_commit_time, 1000));
976 seq_printf(seq, " %lu handles per transaction\n",
977 s->stats->run.rs_handle_count / s->stats->ts_tid);
978 seq_printf(seq, " %lu blocks per transaction\n",
979 s->stats->run.rs_blocks / s->stats->ts_tid);
980 seq_printf(seq, " %lu logged blocks per transaction\n",
981 s->stats->run.rs_blocks_logged / s->stats->ts_tid);
985 static void jbd2_seq_info_stop(struct seq_file *seq, void *v)
989 static const struct seq_operations jbd2_seq_info_ops = {
990 .start = jbd2_seq_info_start,
991 .next = jbd2_seq_info_next,
992 .stop = jbd2_seq_info_stop,
993 .show = jbd2_seq_info_show,
996 static int jbd2_seq_info_open(struct inode *inode, struct file *file)
998 journal_t *journal = PDE_DATA(inode);
999 struct jbd2_stats_proc_session *s;
1002 s = kmalloc(sizeof(*s), GFP_KERNEL);
1005 size = sizeof(struct transaction_stats_s);
1006 s->stats = kmalloc(size, GFP_KERNEL);
1007 if (s->stats == NULL) {
1011 spin_lock(&journal->j_history_lock);
1012 memcpy(s->stats, &journal->j_stats, size);
1013 s->journal = journal;
1014 spin_unlock(&journal->j_history_lock);
1016 rc = seq_open(file, &jbd2_seq_info_ops);
1018 struct seq_file *m = file->private_data;
1028 static int jbd2_seq_info_release(struct inode *inode, struct file *file)
1030 struct seq_file *seq = file->private_data;
1031 struct jbd2_stats_proc_session *s = seq->private;
1034 return seq_release(inode, file);
1037 static const struct file_operations jbd2_seq_info_fops = {
1038 .owner = THIS_MODULE,
1039 .open = jbd2_seq_info_open,
1041 .llseek = seq_lseek,
1042 .release = jbd2_seq_info_release,
1045 static struct proc_dir_entry *proc_jbd2_stats;
1047 static void jbd2_stats_proc_init(journal_t *journal)
1049 journal->j_proc_entry = proc_mkdir(journal->j_devname, proc_jbd2_stats);
1050 if (journal->j_proc_entry) {
1051 proc_create_data("info", S_IRUGO, journal->j_proc_entry,
1052 &jbd2_seq_info_fops, journal);
1056 static void jbd2_stats_proc_exit(journal_t *journal)
1058 remove_proc_entry("info", journal->j_proc_entry);
1059 remove_proc_entry(journal->j_devname, proc_jbd2_stats);
1063 * Management for journal control blocks: functions to create and
1064 * destroy journal_t structures, and to initialise and read existing
1065 * journal blocks from disk. */
1067 /* First: create and setup a journal_t object in memory. We initialise
1068 * very few fields yet: that has to wait until we have created the
1069 * journal structures from from scratch, or loaded them from disk. */
1071 static journal_t * journal_init_common (void)
1076 journal = kzalloc(sizeof(*journal), GFP_KERNEL);
1080 init_waitqueue_head(&journal->j_wait_transaction_locked);
1081 init_waitqueue_head(&journal->j_wait_done_commit);
1082 init_waitqueue_head(&journal->j_wait_commit);
1083 init_waitqueue_head(&journal->j_wait_updates);
1084 init_waitqueue_head(&journal->j_wait_reserved);
1085 mutex_init(&journal->j_barrier);
1086 mutex_init(&journal->j_checkpoint_mutex);
1087 spin_lock_init(&journal->j_revoke_lock);
1088 spin_lock_init(&journal->j_list_lock);
1089 rwlock_init(&journal->j_state_lock);
1091 journal->j_commit_interval = (HZ * JBD2_DEFAULT_MAX_COMMIT_AGE);
1092 journal->j_min_batch_time = 0;
1093 journal->j_max_batch_time = 15000; /* 15ms */
1094 atomic_set(&journal->j_reserved_credits, 0);
1096 /* The journal is marked for error until we succeed with recovery! */
1097 journal->j_flags = JBD2_ABORT;
1099 /* Set up a default-sized revoke table for the new mount. */
1100 err = jbd2_journal_init_revoke(journal, JOURNAL_REVOKE_DEFAULT_HASH);
1106 spin_lock_init(&journal->j_history_lock);
1111 /* jbd2_journal_init_dev and jbd2_journal_init_inode:
1113 * Create a journal structure assigned some fixed set of disk blocks to
1114 * the journal. We don't actually touch those disk blocks yet, but we
1115 * need to set up all of the mapping information to tell the journaling
1116 * system where the journal blocks are.
1121 * journal_t * jbd2_journal_init_dev() - creates and initialises a journal structure
1122 * @bdev: Block device on which to create the journal
1123 * @fs_dev: Device which hold journalled filesystem for this journal.
1124 * @start: Block nr Start of journal.
1125 * @len: Length of the journal in blocks.
1126 * @blocksize: blocksize of journalling device
1128 * Returns: a newly created journal_t *
1130 * jbd2_journal_init_dev creates a journal which maps a fixed contiguous
1131 * range of blocks on an arbitrary block device.
1134 journal_t * jbd2_journal_init_dev(struct block_device *bdev,
1135 struct block_device *fs_dev,
1136 unsigned long long start, int len, int blocksize)
1138 journal_t *journal = journal_init_common();
1139 struct buffer_head *bh;
1145 /* journal descriptor can store up to n blocks -bzzz */
1146 journal->j_blocksize = blocksize;
1147 journal->j_dev = bdev;
1148 journal->j_fs_dev = fs_dev;
1149 journal->j_blk_offset = start;
1150 journal->j_maxlen = len;
1151 bdevname(journal->j_dev, journal->j_devname);
1152 strreplace(journal->j_devname, '/', '!');
1153 jbd2_stats_proc_init(journal);
1154 n = journal->j_blocksize / sizeof(journal_block_tag_t);
1155 journal->j_wbufsize = n;
1156 journal->j_wbuf = kmalloc(n * sizeof(struct buffer_head*), GFP_KERNEL);
1157 if (!journal->j_wbuf) {
1158 printk(KERN_ERR "%s: Can't allocate bhs for commit thread\n",
1163 bh = __getblk(journal->j_dev, start, journal->j_blocksize);
1166 "%s: Cannot get buffer for journal superblock\n",
1170 journal->j_sb_buffer = bh;
1171 journal->j_superblock = (journal_superblock_t *)bh->b_data;
1175 kfree(journal->j_wbuf);
1176 jbd2_stats_proc_exit(journal);
1182 * journal_t * jbd2_journal_init_inode () - creates a journal which maps to a inode.
1183 * @inode: An inode to create the journal in
1185 * jbd2_journal_init_inode creates a journal which maps an on-disk inode as
1186 * the journal. The inode must exist already, must support bmap() and
1187 * must have all data blocks preallocated.
1189 journal_t * jbd2_journal_init_inode (struct inode *inode)
1191 struct buffer_head *bh;
1192 journal_t *journal = journal_init_common();
1196 unsigned long long blocknr;
1201 journal->j_dev = journal->j_fs_dev = inode->i_sb->s_bdev;
1202 journal->j_inode = inode;
1203 bdevname(journal->j_dev, journal->j_devname);
1204 p = strreplace(journal->j_devname, '/', '!');
1205 sprintf(p, "-%lu", journal->j_inode->i_ino);
1207 "journal %p: inode %s/%ld, size %Ld, bits %d, blksize %ld\n",
1208 journal, inode->i_sb->s_id, inode->i_ino,
1209 (long long) inode->i_size,
1210 inode->i_sb->s_blocksize_bits, inode->i_sb->s_blocksize);
1212 journal->j_maxlen = inode->i_size >> inode->i_sb->s_blocksize_bits;
1213 journal->j_blocksize = inode->i_sb->s_blocksize;
1214 jbd2_stats_proc_init(journal);
1216 /* journal descriptor can store up to n blocks -bzzz */
1217 n = journal->j_blocksize / sizeof(journal_block_tag_t);
1218 journal->j_wbufsize = n;
1219 journal->j_wbuf = kmalloc(n * sizeof(struct buffer_head*), GFP_KERNEL);
1220 if (!journal->j_wbuf) {
1221 printk(KERN_ERR "%s: Can't allocate bhs for commit thread\n",
1226 err = jbd2_journal_bmap(journal, 0, &blocknr);
1227 /* If that failed, give up */
1229 printk(KERN_ERR "%s: Cannot locate journal superblock\n",
1234 bh = getblk_unmovable(journal->j_dev, blocknr, journal->j_blocksize);
1237 "%s: Cannot get buffer for journal superblock\n",
1241 journal->j_sb_buffer = bh;
1242 journal->j_superblock = (journal_superblock_t *)bh->b_data;
1246 kfree(journal->j_wbuf);
1247 jbd2_stats_proc_exit(journal);
1253 * If the journal init or create aborts, we need to mark the journal
1254 * superblock as being NULL to prevent the journal destroy from writing
1255 * back a bogus superblock.
1257 static void journal_fail_superblock (journal_t *journal)
1259 struct buffer_head *bh = journal->j_sb_buffer;
1261 journal->j_sb_buffer = NULL;
1265 * Given a journal_t structure, initialise the various fields for
1266 * startup of a new journaling session. We use this both when creating
1267 * a journal, and after recovering an old journal to reset it for
1271 static int journal_reset(journal_t *journal)
1273 journal_superblock_t *sb = journal->j_superblock;
1274 unsigned long long first, last;
1276 first = be32_to_cpu(sb->s_first);
1277 last = be32_to_cpu(sb->s_maxlen);
1278 if (first + JBD2_MIN_JOURNAL_BLOCKS > last + 1) {
1279 printk(KERN_ERR "JBD2: Journal too short (blocks %llu-%llu).\n",
1281 journal_fail_superblock(journal);
1285 journal->j_first = first;
1286 journal->j_last = last;
1288 journal->j_head = first;
1289 journal->j_tail = first;
1290 journal->j_free = last - first;
1292 journal->j_tail_sequence = journal->j_transaction_sequence;
1293 journal->j_commit_sequence = journal->j_transaction_sequence - 1;
1294 journal->j_commit_request = journal->j_commit_sequence;
1296 journal->j_max_transaction_buffers = journal->j_maxlen / 4;
1299 * As a special case, if the on-disk copy is already marked as needing
1300 * no recovery (s_start == 0), then we can safely defer the superblock
1301 * update until the next commit by setting JBD2_FLUSHED. This avoids
1302 * attempting a write to a potential-readonly device.
1304 if (sb->s_start == 0) {
1305 jbd_debug(1, "JBD2: Skipping superblock update on recovered sb "
1306 "(start %ld, seq %d, errno %d)\n",
1307 journal->j_tail, journal->j_tail_sequence,
1309 journal->j_flags |= JBD2_FLUSHED;
1311 /* Lock here to make assertions happy... */
1312 mutex_lock(&journal->j_checkpoint_mutex);
1314 * Update log tail information. We use WRITE_FUA since new
1315 * transaction will start reusing journal space and so we
1316 * must make sure information about current log tail is on
1319 jbd2_journal_update_sb_log_tail(journal,
1320 journal->j_tail_sequence,
1323 mutex_unlock(&journal->j_checkpoint_mutex);
1325 return jbd2_journal_start_thread(journal);
1328 static void jbd2_write_superblock(journal_t *journal, int write_op)
1330 struct buffer_head *bh = journal->j_sb_buffer;
1331 journal_superblock_t *sb = journal->j_superblock;
1334 trace_jbd2_write_superblock(journal, write_op);
1335 if (!(journal->j_flags & JBD2_BARRIER))
1336 write_op &= ~(REQ_FUA | REQ_FLUSH);
1338 if (buffer_write_io_error(bh)) {
1340 * Oh, dear. A previous attempt to write the journal
1341 * superblock failed. This could happen because the
1342 * USB device was yanked out. Or it could happen to
1343 * be a transient write error and maybe the block will
1344 * be remapped. Nothing we can do but to retry the
1345 * write and hope for the best.
1347 printk(KERN_ERR "JBD2: previous I/O error detected "
1348 "for journal superblock update for %s.\n",
1349 journal->j_devname);
1350 clear_buffer_write_io_error(bh);
1351 set_buffer_uptodate(bh);
1353 jbd2_superblock_csum_set(journal, sb);
1355 bh->b_end_io = end_buffer_write_sync;
1356 ret = submit_bh(write_op, bh);
1358 if (buffer_write_io_error(bh)) {
1359 clear_buffer_write_io_error(bh);
1360 set_buffer_uptodate(bh);
1364 printk(KERN_ERR "JBD2: Error %d detected when updating "
1365 "journal superblock for %s.\n", ret,
1366 journal->j_devname);
1371 * jbd2_journal_update_sb_log_tail() - Update log tail in journal sb on disk.
1372 * @journal: The journal to update.
1373 * @tail_tid: TID of the new transaction at the tail of the log
1374 * @tail_block: The first block of the transaction at the tail of the log
1375 * @write_op: With which operation should we write the journal sb
1377 * Update a journal's superblock information about log tail and write it to
1378 * disk, waiting for the IO to complete.
1380 void jbd2_journal_update_sb_log_tail(journal_t *journal, tid_t tail_tid,
1381 unsigned long tail_block, int write_op)
1383 journal_superblock_t *sb = journal->j_superblock;
1385 BUG_ON(!mutex_is_locked(&journal->j_checkpoint_mutex));
1386 jbd_debug(1, "JBD2: updating superblock (start %lu, seq %u)\n",
1387 tail_block, tail_tid);
1389 sb->s_sequence = cpu_to_be32(tail_tid);
1390 sb->s_start = cpu_to_be32(tail_block);
1392 jbd2_write_superblock(journal, write_op);
1394 /* Log is no longer empty */
1395 write_lock(&journal->j_state_lock);
1396 WARN_ON(!sb->s_sequence);
1397 journal->j_flags &= ~JBD2_FLUSHED;
1398 write_unlock(&journal->j_state_lock);
1402 * jbd2_mark_journal_empty() - Mark on disk journal as empty.
1403 * @journal: The journal to update.
1405 * Update a journal's dynamic superblock fields to show that journal is empty.
1406 * Write updated superblock to disk waiting for IO to complete.
1408 static void jbd2_mark_journal_empty(journal_t *journal)
1410 journal_superblock_t *sb = journal->j_superblock;
1412 BUG_ON(!mutex_is_locked(&journal->j_checkpoint_mutex));
1413 read_lock(&journal->j_state_lock);
1414 /* Is it already empty? */
1415 if (sb->s_start == 0) {
1416 read_unlock(&journal->j_state_lock);
1419 jbd_debug(1, "JBD2: Marking journal as empty (seq %d)\n",
1420 journal->j_tail_sequence);
1422 sb->s_sequence = cpu_to_be32(journal->j_tail_sequence);
1423 sb->s_start = cpu_to_be32(0);
1424 read_unlock(&journal->j_state_lock);
1426 jbd2_write_superblock(journal, WRITE_FUA);
1428 /* Log is no longer empty */
1429 write_lock(&journal->j_state_lock);
1430 journal->j_flags |= JBD2_FLUSHED;
1431 write_unlock(&journal->j_state_lock);
1436 * jbd2_journal_update_sb_errno() - Update error in the journal.
1437 * @journal: The journal to update.
1439 * Update a journal's errno. Write updated superblock to disk waiting for IO
1442 void jbd2_journal_update_sb_errno(journal_t *journal)
1444 journal_superblock_t *sb = journal->j_superblock;
1446 read_lock(&journal->j_state_lock);
1447 jbd_debug(1, "JBD2: updating superblock error (errno %d)\n",
1449 sb->s_errno = cpu_to_be32(journal->j_errno);
1450 read_unlock(&journal->j_state_lock);
1452 jbd2_write_superblock(journal, WRITE_SYNC);
1454 EXPORT_SYMBOL(jbd2_journal_update_sb_errno);
1457 * Read the superblock for a given journal, performing initial
1458 * validation of the format.
1460 static int journal_get_superblock(journal_t *journal)
1462 struct buffer_head *bh;
1463 journal_superblock_t *sb;
1466 bh = journal->j_sb_buffer;
1468 J_ASSERT(bh != NULL);
1469 if (!buffer_uptodate(bh)) {
1470 ll_rw_block(READ, 1, &bh);
1472 if (!buffer_uptodate(bh)) {
1474 "JBD2: IO error reading journal superblock\n");
1479 if (buffer_verified(bh))
1482 sb = journal->j_superblock;
1486 if (sb->s_header.h_magic != cpu_to_be32(JBD2_MAGIC_NUMBER) ||
1487 sb->s_blocksize != cpu_to_be32(journal->j_blocksize)) {
1488 printk(KERN_WARNING "JBD2: no valid journal superblock found\n");
1492 switch(be32_to_cpu(sb->s_header.h_blocktype)) {
1493 case JBD2_SUPERBLOCK_V1:
1494 journal->j_format_version = 1;
1496 case JBD2_SUPERBLOCK_V2:
1497 journal->j_format_version = 2;
1500 printk(KERN_WARNING "JBD2: unrecognised superblock format ID\n");
1504 if (be32_to_cpu(sb->s_maxlen) < journal->j_maxlen)
1505 journal->j_maxlen = be32_to_cpu(sb->s_maxlen);
1506 else if (be32_to_cpu(sb->s_maxlen) > journal->j_maxlen) {
1507 printk(KERN_WARNING "JBD2: journal file too short\n");
1511 if (be32_to_cpu(sb->s_first) == 0 ||
1512 be32_to_cpu(sb->s_first) >= journal->j_maxlen) {
1514 "JBD2: Invalid start block of journal: %u\n",
1515 be32_to_cpu(sb->s_first));
1519 if (JBD2_HAS_INCOMPAT_FEATURE(journal, JBD2_FEATURE_INCOMPAT_CSUM_V2) &&
1520 JBD2_HAS_INCOMPAT_FEATURE(journal, JBD2_FEATURE_INCOMPAT_CSUM_V3)) {
1521 /* Can't have checksum v2 and v3 at the same time! */
1522 printk(KERN_ERR "JBD2: Can't enable checksumming v2 and v3 "
1523 "at the same time!\n");
1527 if (jbd2_journal_has_csum_v2or3(journal) &&
1528 JBD2_HAS_COMPAT_FEATURE(journal, JBD2_FEATURE_COMPAT_CHECKSUM)) {
1529 /* Can't have checksum v1 and v2 on at the same time! */
1530 printk(KERN_ERR "JBD2: Can't enable checksumming v1 and v2/3 "
1531 "at the same time!\n");
1535 if (!jbd2_verify_csum_type(journal, sb)) {
1536 printk(KERN_ERR "JBD2: Unknown checksum type\n");
1540 /* Load the checksum driver */
1541 if (jbd2_journal_has_csum_v2or3(journal)) {
1542 journal->j_chksum_driver = crypto_alloc_shash("crc32c", 0, 0);
1543 if (IS_ERR(journal->j_chksum_driver)) {
1544 printk(KERN_ERR "JBD2: Cannot load crc32c driver.\n");
1545 err = PTR_ERR(journal->j_chksum_driver);
1546 journal->j_chksum_driver = NULL;
1551 /* Check superblock checksum */
1552 if (!jbd2_superblock_csum_verify(journal, sb)) {
1553 printk(KERN_ERR "JBD2: journal checksum error\n");
1557 /* Precompute checksum seed for all metadata */
1558 if (jbd2_journal_has_csum_v2or3(journal))
1559 journal->j_csum_seed = jbd2_chksum(journal, ~0, sb->s_uuid,
1560 sizeof(sb->s_uuid));
1562 set_buffer_verified(bh);
1567 journal_fail_superblock(journal);
1572 * Load the on-disk journal superblock and read the key fields into the
1576 static int load_superblock(journal_t *journal)
1579 journal_superblock_t *sb;
1581 err = journal_get_superblock(journal);
1585 sb = journal->j_superblock;
1587 journal->j_tail_sequence = be32_to_cpu(sb->s_sequence);
1588 journal->j_tail = be32_to_cpu(sb->s_start);
1589 journal->j_first = be32_to_cpu(sb->s_first);
1590 journal->j_last = be32_to_cpu(sb->s_maxlen);
1591 journal->j_errno = be32_to_cpu(sb->s_errno);
1598 * int jbd2_journal_load() - Read journal from disk.
1599 * @journal: Journal to act on.
1601 * Given a journal_t structure which tells us which disk blocks contain
1602 * a journal, read the journal from disk to initialise the in-memory
1605 int jbd2_journal_load(journal_t *journal)
1608 journal_superblock_t *sb;
1610 err = load_superblock(journal);
1614 sb = journal->j_superblock;
1615 /* If this is a V2 superblock, then we have to check the
1616 * features flags on it. */
1618 if (journal->j_format_version >= 2) {
1619 if ((sb->s_feature_ro_compat &
1620 ~cpu_to_be32(JBD2_KNOWN_ROCOMPAT_FEATURES)) ||
1621 (sb->s_feature_incompat &
1622 ~cpu_to_be32(JBD2_KNOWN_INCOMPAT_FEATURES))) {
1624 "JBD2: Unrecognised features on journal\n");
1630 * Create a slab for this blocksize
1632 err = jbd2_journal_create_slab(be32_to_cpu(sb->s_blocksize));
1636 /* Let the recovery code check whether it needs to recover any
1637 * data from the journal. */
1638 if (jbd2_journal_recover(journal))
1639 goto recovery_error;
1641 if (journal->j_failed_commit) {
1642 printk(KERN_ERR "JBD2: journal transaction %u on %s "
1643 "is corrupt.\n", journal->j_failed_commit,
1644 journal->j_devname);
1648 /* OK, we've finished with the dynamic journal bits:
1649 * reinitialise the dynamic contents of the superblock in memory
1650 * and reset them on disk. */
1651 if (journal_reset(journal))
1652 goto recovery_error;
1654 journal->j_flags &= ~JBD2_ABORT;
1655 journal->j_flags |= JBD2_LOADED;
1659 printk(KERN_WARNING "JBD2: recovery failed\n");
1664 * void jbd2_journal_destroy() - Release a journal_t structure.
1665 * @journal: Journal to act on.
1667 * Release a journal_t structure once it is no longer in use by the
1669 * Return <0 if we couldn't clean up the journal.
1671 int jbd2_journal_destroy(journal_t *journal)
1675 /* Wait for the commit thread to wake up and die. */
1676 journal_kill_thread(journal);
1678 /* Force a final log commit */
1679 if (journal->j_running_transaction)
1680 jbd2_journal_commit_transaction(journal);
1682 /* Force any old transactions to disk */
1684 /* Totally anal locking here... */
1685 spin_lock(&journal->j_list_lock);
1686 while (journal->j_checkpoint_transactions != NULL) {
1687 spin_unlock(&journal->j_list_lock);
1688 mutex_lock(&journal->j_checkpoint_mutex);
1689 jbd2_log_do_checkpoint(journal);
1690 mutex_unlock(&journal->j_checkpoint_mutex);
1691 spin_lock(&journal->j_list_lock);
1694 J_ASSERT(journal->j_running_transaction == NULL);
1695 J_ASSERT(journal->j_committing_transaction == NULL);
1696 J_ASSERT(journal->j_checkpoint_transactions == NULL);
1697 spin_unlock(&journal->j_list_lock);
1699 if (journal->j_sb_buffer) {
1700 if (!is_journal_aborted(journal)) {
1701 mutex_lock(&journal->j_checkpoint_mutex);
1702 jbd2_mark_journal_empty(journal);
1703 mutex_unlock(&journal->j_checkpoint_mutex);
1706 brelse(journal->j_sb_buffer);
1709 if (journal->j_proc_entry)
1710 jbd2_stats_proc_exit(journal);
1711 iput(journal->j_inode);
1712 if (journal->j_revoke)
1713 jbd2_journal_destroy_revoke(journal);
1714 if (journal->j_chksum_driver)
1715 crypto_free_shash(journal->j_chksum_driver);
1716 kfree(journal->j_wbuf);
1724 *int jbd2_journal_check_used_features () - Check if features specified are used.
1725 * @journal: Journal to check.
1726 * @compat: bitmask of compatible features
1727 * @ro: bitmask of features that force read-only mount
1728 * @incompat: bitmask of incompatible features
1730 * Check whether the journal uses all of a given set of
1731 * features. Return true (non-zero) if it does.
1734 int jbd2_journal_check_used_features (journal_t *journal, unsigned long compat,
1735 unsigned long ro, unsigned long incompat)
1737 journal_superblock_t *sb;
1739 if (!compat && !ro && !incompat)
1741 /* Load journal superblock if it is not loaded yet. */
1742 if (journal->j_format_version == 0 &&
1743 journal_get_superblock(journal) != 0)
1745 if (journal->j_format_version == 1)
1748 sb = journal->j_superblock;
1750 if (((be32_to_cpu(sb->s_feature_compat) & compat) == compat) &&
1751 ((be32_to_cpu(sb->s_feature_ro_compat) & ro) == ro) &&
1752 ((be32_to_cpu(sb->s_feature_incompat) & incompat) == incompat))
1759 * int jbd2_journal_check_available_features() - Check feature set in journalling layer
1760 * @journal: Journal to check.
1761 * @compat: bitmask of compatible features
1762 * @ro: bitmask of features that force read-only mount
1763 * @incompat: bitmask of incompatible features
1765 * Check whether the journaling code supports the use of
1766 * all of a given set of features on this journal. Return true
1767 * (non-zero) if it can. */
1769 int jbd2_journal_check_available_features (journal_t *journal, unsigned long compat,
1770 unsigned long ro, unsigned long incompat)
1772 if (!compat && !ro && !incompat)
1775 /* We can support any known requested features iff the
1776 * superblock is in version 2. Otherwise we fail to support any
1777 * extended sb features. */
1779 if (journal->j_format_version != 2)
1782 if ((compat & JBD2_KNOWN_COMPAT_FEATURES) == compat &&
1783 (ro & JBD2_KNOWN_ROCOMPAT_FEATURES) == ro &&
1784 (incompat & JBD2_KNOWN_INCOMPAT_FEATURES) == incompat)
1791 * int jbd2_journal_set_features () - Mark a given journal feature in the superblock
1792 * @journal: Journal to act on.
1793 * @compat: bitmask of compatible features
1794 * @ro: bitmask of features that force read-only mount
1795 * @incompat: bitmask of incompatible features
1797 * Mark a given journal feature as present on the
1798 * superblock. Returns true if the requested features could be set.
1802 int jbd2_journal_set_features (journal_t *journal, unsigned long compat,
1803 unsigned long ro, unsigned long incompat)
1805 #define INCOMPAT_FEATURE_ON(f) \
1806 ((incompat & (f)) && !(sb->s_feature_incompat & cpu_to_be32(f)))
1807 #define COMPAT_FEATURE_ON(f) \
1808 ((compat & (f)) && !(sb->s_feature_compat & cpu_to_be32(f)))
1809 journal_superblock_t *sb;
1811 if (jbd2_journal_check_used_features(journal, compat, ro, incompat))
1814 if (!jbd2_journal_check_available_features(journal, compat, ro, incompat))
1817 /* If enabling v2 checksums, turn on v3 instead */
1818 if (incompat & JBD2_FEATURE_INCOMPAT_CSUM_V2) {
1819 incompat &= ~JBD2_FEATURE_INCOMPAT_CSUM_V2;
1820 incompat |= JBD2_FEATURE_INCOMPAT_CSUM_V3;
1823 /* Asking for checksumming v3 and v1? Only give them v3. */
1824 if (incompat & JBD2_FEATURE_INCOMPAT_CSUM_V3 &&
1825 compat & JBD2_FEATURE_COMPAT_CHECKSUM)
1826 compat &= ~JBD2_FEATURE_COMPAT_CHECKSUM;
1828 jbd_debug(1, "Setting new features 0x%lx/0x%lx/0x%lx\n",
1829 compat, ro, incompat);
1831 sb = journal->j_superblock;
1833 /* If enabling v3 checksums, update superblock */
1834 if (INCOMPAT_FEATURE_ON(JBD2_FEATURE_INCOMPAT_CSUM_V3)) {
1835 sb->s_checksum_type = JBD2_CRC32C_CHKSUM;
1836 sb->s_feature_compat &=
1837 ~cpu_to_be32(JBD2_FEATURE_COMPAT_CHECKSUM);
1839 /* Load the checksum driver */
1840 if (journal->j_chksum_driver == NULL) {
1841 journal->j_chksum_driver = crypto_alloc_shash("crc32c",
1843 if (IS_ERR(journal->j_chksum_driver)) {
1844 printk(KERN_ERR "JBD2: Cannot load crc32c "
1846 journal->j_chksum_driver = NULL;
1850 /* Precompute checksum seed for all metadata */
1851 journal->j_csum_seed = jbd2_chksum(journal, ~0,
1853 sizeof(sb->s_uuid));
1857 /* If enabling v1 checksums, downgrade superblock */
1858 if (COMPAT_FEATURE_ON(JBD2_FEATURE_COMPAT_CHECKSUM))
1859 sb->s_feature_incompat &=
1860 ~cpu_to_be32(JBD2_FEATURE_INCOMPAT_CSUM_V2 |
1861 JBD2_FEATURE_INCOMPAT_CSUM_V3);
1863 sb->s_feature_compat |= cpu_to_be32(compat);
1864 sb->s_feature_ro_compat |= cpu_to_be32(ro);
1865 sb->s_feature_incompat |= cpu_to_be32(incompat);
1868 #undef COMPAT_FEATURE_ON
1869 #undef INCOMPAT_FEATURE_ON
1873 * jbd2_journal_clear_features () - Clear a given journal feature in the
1875 * @journal: Journal to act on.
1876 * @compat: bitmask of compatible features
1877 * @ro: bitmask of features that force read-only mount
1878 * @incompat: bitmask of incompatible features
1880 * Clear a given journal feature as present on the
1883 void jbd2_journal_clear_features(journal_t *journal, unsigned long compat,
1884 unsigned long ro, unsigned long incompat)
1886 journal_superblock_t *sb;
1888 jbd_debug(1, "Clear features 0x%lx/0x%lx/0x%lx\n",
1889 compat, ro, incompat);
1891 sb = journal->j_superblock;
1893 sb->s_feature_compat &= ~cpu_to_be32(compat);
1894 sb->s_feature_ro_compat &= ~cpu_to_be32(ro);
1895 sb->s_feature_incompat &= ~cpu_to_be32(incompat);
1897 EXPORT_SYMBOL(jbd2_journal_clear_features);
1900 * int jbd2_journal_flush () - Flush journal
1901 * @journal: Journal to act on.
1903 * Flush all data for a given journal to disk and empty the journal.
1904 * Filesystems can use this when remounting readonly to ensure that
1905 * recovery does not need to happen on remount.
1908 int jbd2_journal_flush(journal_t *journal)
1911 transaction_t *transaction = NULL;
1913 write_lock(&journal->j_state_lock);
1915 /* Force everything buffered to the log... */
1916 if (journal->j_running_transaction) {
1917 transaction = journal->j_running_transaction;
1918 __jbd2_log_start_commit(journal, transaction->t_tid);
1919 } else if (journal->j_committing_transaction)
1920 transaction = journal->j_committing_transaction;
1922 /* Wait for the log commit to complete... */
1924 tid_t tid = transaction->t_tid;
1926 write_unlock(&journal->j_state_lock);
1927 jbd2_log_wait_commit(journal, tid);
1929 write_unlock(&journal->j_state_lock);
1932 /* ...and flush everything in the log out to disk. */
1933 spin_lock(&journal->j_list_lock);
1934 while (!err && journal->j_checkpoint_transactions != NULL) {
1935 spin_unlock(&journal->j_list_lock);
1936 mutex_lock(&journal->j_checkpoint_mutex);
1937 err = jbd2_log_do_checkpoint(journal);
1938 mutex_unlock(&journal->j_checkpoint_mutex);
1939 spin_lock(&journal->j_list_lock);
1941 spin_unlock(&journal->j_list_lock);
1943 if (is_journal_aborted(journal))
1946 mutex_lock(&journal->j_checkpoint_mutex);
1947 jbd2_cleanup_journal_tail(journal);
1949 /* Finally, mark the journal as really needing no recovery.
1950 * This sets s_start==0 in the underlying superblock, which is
1951 * the magic code for a fully-recovered superblock. Any future
1952 * commits of data to the journal will restore the current
1954 jbd2_mark_journal_empty(journal);
1955 mutex_unlock(&journal->j_checkpoint_mutex);
1956 write_lock(&journal->j_state_lock);
1957 J_ASSERT(!journal->j_running_transaction);
1958 J_ASSERT(!journal->j_committing_transaction);
1959 J_ASSERT(!journal->j_checkpoint_transactions);
1960 J_ASSERT(journal->j_head == journal->j_tail);
1961 J_ASSERT(journal->j_tail_sequence == journal->j_transaction_sequence);
1962 write_unlock(&journal->j_state_lock);
1967 * int jbd2_journal_wipe() - Wipe journal contents
1968 * @journal: Journal to act on.
1969 * @write: flag (see below)
1971 * Wipe out all of the contents of a journal, safely. This will produce
1972 * a warning if the journal contains any valid recovery information.
1973 * Must be called between journal_init_*() and jbd2_journal_load().
1975 * If 'write' is non-zero, then we wipe out the journal on disk; otherwise
1976 * we merely suppress recovery.
1979 int jbd2_journal_wipe(journal_t *journal, int write)
1983 J_ASSERT (!(journal->j_flags & JBD2_LOADED));
1985 err = load_superblock(journal);
1989 if (!journal->j_tail)
1992 printk(KERN_WARNING "JBD2: %s recovery information on journal\n",
1993 write ? "Clearing" : "Ignoring");
1995 err = jbd2_journal_skip_recovery(journal);
1997 /* Lock to make assertions happy... */
1998 mutex_lock(&journal->j_checkpoint_mutex);
1999 jbd2_mark_journal_empty(journal);
2000 mutex_unlock(&journal->j_checkpoint_mutex);
2008 * Journal abort has very specific semantics, which we describe
2009 * for journal abort.
2011 * Two internal functions, which provide abort to the jbd layer
2016 * Quick version for internal journal use (doesn't lock the journal).
2017 * Aborts hard --- we mark the abort as occurred, but do _nothing_ else,
2018 * and don't attempt to make any other journal updates.
2020 void __jbd2_journal_abort_hard(journal_t *journal)
2022 transaction_t *transaction;
2024 if (journal->j_flags & JBD2_ABORT)
2027 printk(KERN_ERR "Aborting journal on device %s.\n",
2028 journal->j_devname);
2030 write_lock(&journal->j_state_lock);
2031 journal->j_flags |= JBD2_ABORT;
2032 transaction = journal->j_running_transaction;
2034 __jbd2_log_start_commit(journal, transaction->t_tid);
2035 write_unlock(&journal->j_state_lock);
2038 /* Soft abort: record the abort error status in the journal superblock,
2039 * but don't do any other IO. */
2040 static void __journal_abort_soft (journal_t *journal, int errno)
2042 if (journal->j_flags & JBD2_ABORT)
2045 if (!journal->j_errno)
2046 journal->j_errno = errno;
2048 __jbd2_journal_abort_hard(journal);
2051 jbd2_journal_update_sb_errno(journal);
2055 * void jbd2_journal_abort () - Shutdown the journal immediately.
2056 * @journal: the journal to shutdown.
2057 * @errno: an error number to record in the journal indicating
2058 * the reason for the shutdown.
2060 * Perform a complete, immediate shutdown of the ENTIRE
2061 * journal (not of a single transaction). This operation cannot be
2062 * undone without closing and reopening the journal.
2064 * The jbd2_journal_abort function is intended to support higher level error
2065 * recovery mechanisms such as the ext2/ext3 remount-readonly error
2068 * Journal abort has very specific semantics. Any existing dirty,
2069 * unjournaled buffers in the main filesystem will still be written to
2070 * disk by bdflush, but the journaling mechanism will be suspended
2071 * immediately and no further transaction commits will be honoured.
2073 * Any dirty, journaled buffers will be written back to disk without
2074 * hitting the journal. Atomicity cannot be guaranteed on an aborted
2075 * filesystem, but we _do_ attempt to leave as much data as possible
2076 * behind for fsck to use for cleanup.
2078 * Any attempt to get a new transaction handle on a journal which is in
2079 * ABORT state will just result in an -EROFS error return. A
2080 * jbd2_journal_stop on an existing handle will return -EIO if we have
2081 * entered abort state during the update.
2083 * Recursive transactions are not disturbed by journal abort until the
2084 * final jbd2_journal_stop, which will receive the -EIO error.
2086 * Finally, the jbd2_journal_abort call allows the caller to supply an errno
2087 * which will be recorded (if possible) in the journal superblock. This
2088 * allows a client to record failure conditions in the middle of a
2089 * transaction without having to complete the transaction to record the
2090 * failure to disk. ext3_error, for example, now uses this
2093 * Errors which originate from within the journaling layer will NOT
2094 * supply an errno; a null errno implies that absolutely no further
2095 * writes are done to the journal (unless there are any already in
2100 void jbd2_journal_abort(journal_t *journal, int errno)
2102 __journal_abort_soft(journal, errno);
2106 * int jbd2_journal_errno () - returns the journal's error state.
2107 * @journal: journal to examine.
2109 * This is the errno number set with jbd2_journal_abort(), the last
2110 * time the journal was mounted - if the journal was stopped
2111 * without calling abort this will be 0.
2113 * If the journal has been aborted on this mount time -EROFS will
2116 int jbd2_journal_errno(journal_t *journal)
2120 read_lock(&journal->j_state_lock);
2121 if (journal->j_flags & JBD2_ABORT)
2124 err = journal->j_errno;
2125 read_unlock(&journal->j_state_lock);
2130 * int jbd2_journal_clear_err () - clears the journal's error state
2131 * @journal: journal to act on.
2133 * An error must be cleared or acked to take a FS out of readonly
2136 int jbd2_journal_clear_err(journal_t *journal)
2140 write_lock(&journal->j_state_lock);
2141 if (journal->j_flags & JBD2_ABORT)
2144 journal->j_errno = 0;
2145 write_unlock(&journal->j_state_lock);
2150 * void jbd2_journal_ack_err() - Ack journal err.
2151 * @journal: journal to act on.
2153 * An error must be cleared or acked to take a FS out of readonly
2156 void jbd2_journal_ack_err(journal_t *journal)
2158 write_lock(&journal->j_state_lock);
2159 if (journal->j_errno)
2160 journal->j_flags |= JBD2_ACK_ERR;
2161 write_unlock(&journal->j_state_lock);
2164 int jbd2_journal_blocks_per_page(struct inode *inode)
2166 return 1 << (PAGE_CACHE_SHIFT - inode->i_sb->s_blocksize_bits);
2170 * helper functions to deal with 32 or 64bit block numbers.
2172 size_t journal_tag_bytes(journal_t *journal)
2176 if (JBD2_HAS_INCOMPAT_FEATURE(journal, JBD2_FEATURE_INCOMPAT_CSUM_V3))
2177 return sizeof(journal_block_tag3_t);
2179 sz = sizeof(journal_block_tag_t);
2181 if (JBD2_HAS_INCOMPAT_FEATURE(journal, JBD2_FEATURE_INCOMPAT_CSUM_V2))
2182 sz += sizeof(__u16);
2184 if (JBD2_HAS_INCOMPAT_FEATURE(journal, JBD2_FEATURE_INCOMPAT_64BIT))
2187 return sz - sizeof(__u32);
2191 * JBD memory management
2193 * These functions are used to allocate block-sized chunks of memory
2194 * used for making copies of buffer_head data. Very often it will be
2195 * page-sized chunks of data, but sometimes it will be in
2196 * sub-page-size chunks. (For example, 16k pages on Power systems
2197 * with a 4k block file system.) For blocks smaller than a page, we
2198 * use a SLAB allocator. There are slab caches for each block size,
2199 * which are allocated at mount time, if necessary, and we only free
2200 * (all of) the slab caches when/if the jbd2 module is unloaded. For
2201 * this reason we don't need to a mutex to protect access to
2202 * jbd2_slab[] allocating or releasing memory; only in
2203 * jbd2_journal_create_slab().
2205 #define JBD2_MAX_SLABS 8
2206 static struct kmem_cache *jbd2_slab[JBD2_MAX_SLABS];
2208 static const char *jbd2_slab_names[JBD2_MAX_SLABS] = {
2209 "jbd2_1k", "jbd2_2k", "jbd2_4k", "jbd2_8k",
2210 "jbd2_16k", "jbd2_32k", "jbd2_64k", "jbd2_128k"
2214 static void jbd2_journal_destroy_slabs(void)
2218 for (i = 0; i < JBD2_MAX_SLABS; i++) {
2220 kmem_cache_destroy(jbd2_slab[i]);
2221 jbd2_slab[i] = NULL;
2225 static int jbd2_journal_create_slab(size_t size)
2227 static DEFINE_MUTEX(jbd2_slab_create_mutex);
2228 int i = order_base_2(size) - 10;
2231 if (size == PAGE_SIZE)
2234 if (i >= JBD2_MAX_SLABS)
2237 if (unlikely(i < 0))
2239 mutex_lock(&jbd2_slab_create_mutex);
2241 mutex_unlock(&jbd2_slab_create_mutex);
2242 return 0; /* Already created */
2245 slab_size = 1 << (i+10);
2246 jbd2_slab[i] = kmem_cache_create(jbd2_slab_names[i], slab_size,
2247 slab_size, 0, NULL);
2248 mutex_unlock(&jbd2_slab_create_mutex);
2249 if (!jbd2_slab[i]) {
2250 printk(KERN_EMERG "JBD2: no memory for jbd2_slab cache\n");
2256 static struct kmem_cache *get_slab(size_t size)
2258 int i = order_base_2(size) - 10;
2260 BUG_ON(i >= JBD2_MAX_SLABS);
2261 if (unlikely(i < 0))
2263 BUG_ON(jbd2_slab[i] == NULL);
2264 return jbd2_slab[i];
2267 void *jbd2_alloc(size_t size, gfp_t flags)
2271 BUG_ON(size & (size-1)); /* Must be a power of 2 */
2273 flags |= __GFP_REPEAT;
2274 if (size == PAGE_SIZE)
2275 ptr = (void *)__get_free_pages(flags, 0);
2276 else if (size > PAGE_SIZE) {
2277 int order = get_order(size);
2280 ptr = (void *)__get_free_pages(flags, order);
2282 ptr = vmalloc(size);
2284 ptr = kmem_cache_alloc(get_slab(size), flags);
2286 /* Check alignment; SLUB has gotten this wrong in the past,
2287 * and this can lead to user data corruption! */
2288 BUG_ON(((unsigned long) ptr) & (size-1));
2293 void jbd2_free(void *ptr, size_t size)
2295 if (size == PAGE_SIZE) {
2296 free_pages((unsigned long)ptr, 0);
2299 if (size > PAGE_SIZE) {
2300 int order = get_order(size);
2303 free_pages((unsigned long)ptr, order);
2308 kmem_cache_free(get_slab(size), ptr);
2312 * Journal_head storage management
2314 static struct kmem_cache *jbd2_journal_head_cache;
2315 #ifdef CONFIG_JBD2_DEBUG
2316 static atomic_t nr_journal_heads = ATOMIC_INIT(0);
2319 static int jbd2_journal_init_journal_head_cache(void)
2323 J_ASSERT(jbd2_journal_head_cache == NULL);
2324 jbd2_journal_head_cache = kmem_cache_create("jbd2_journal_head",
2325 sizeof(struct journal_head),
2327 SLAB_TEMPORARY, /* flags */
2330 if (!jbd2_journal_head_cache) {
2332 printk(KERN_EMERG "JBD2: no memory for journal_head cache\n");
2337 static void jbd2_journal_destroy_journal_head_cache(void)
2339 if (jbd2_journal_head_cache) {
2340 kmem_cache_destroy(jbd2_journal_head_cache);
2341 jbd2_journal_head_cache = NULL;
2346 * journal_head splicing and dicing
2348 static struct journal_head *journal_alloc_journal_head(void)
2350 struct journal_head *ret;
2352 #ifdef CONFIG_JBD2_DEBUG
2353 atomic_inc(&nr_journal_heads);
2355 ret = kmem_cache_zalloc(jbd2_journal_head_cache, GFP_NOFS);
2357 jbd_debug(1, "out of memory for journal_head\n");
2358 pr_notice_ratelimited("ENOMEM in %s, retrying.\n", __func__);
2361 ret = kmem_cache_zalloc(jbd2_journal_head_cache, GFP_NOFS);
2367 static void journal_free_journal_head(struct journal_head *jh)
2369 #ifdef CONFIG_JBD2_DEBUG
2370 atomic_dec(&nr_journal_heads);
2371 memset(jh, JBD2_POISON_FREE, sizeof(*jh));
2373 kmem_cache_free(jbd2_journal_head_cache, jh);
2377 * A journal_head is attached to a buffer_head whenever JBD has an
2378 * interest in the buffer.
2380 * Whenever a buffer has an attached journal_head, its ->b_state:BH_JBD bit
2381 * is set. This bit is tested in core kernel code where we need to take
2382 * JBD-specific actions. Testing the zeroness of ->b_private is not reliable
2385 * When a buffer has its BH_JBD bit set, its ->b_count is elevated by one.
2387 * When a buffer has its BH_JBD bit set it is immune from being released by
2388 * core kernel code, mainly via ->b_count.
2390 * A journal_head is detached from its buffer_head when the journal_head's
2391 * b_jcount reaches zero. Running transaction (b_transaction) and checkpoint
2392 * transaction (b_cp_transaction) hold their references to b_jcount.
2394 * Various places in the kernel want to attach a journal_head to a buffer_head
2395 * _before_ attaching the journal_head to a transaction. To protect the
2396 * journal_head in this situation, jbd2_journal_add_journal_head elevates the
2397 * journal_head's b_jcount refcount by one. The caller must call
2398 * jbd2_journal_put_journal_head() to undo this.
2400 * So the typical usage would be:
2402 * (Attach a journal_head if needed. Increments b_jcount)
2403 * struct journal_head *jh = jbd2_journal_add_journal_head(bh);
2405 * (Get another reference for transaction)
2406 * jbd2_journal_grab_journal_head(bh);
2407 * jh->b_transaction = xxx;
2408 * (Put original reference)
2409 * jbd2_journal_put_journal_head(jh);
2413 * Give a buffer_head a journal_head.
2417 struct journal_head *jbd2_journal_add_journal_head(struct buffer_head *bh)
2419 struct journal_head *jh;
2420 struct journal_head *new_jh = NULL;
2423 if (!buffer_jbd(bh))
2424 new_jh = journal_alloc_journal_head();
2426 jbd_lock_bh_journal_head(bh);
2427 if (buffer_jbd(bh)) {
2431 (atomic_read(&bh->b_count) > 0) ||
2432 (bh->b_page && bh->b_page->mapping));
2435 jbd_unlock_bh_journal_head(bh);
2440 new_jh = NULL; /* We consumed it */
2445 BUFFER_TRACE(bh, "added journal_head");
2448 jbd_unlock_bh_journal_head(bh);
2450 journal_free_journal_head(new_jh);
2451 return bh->b_private;
2455 * Grab a ref against this buffer_head's journal_head. If it ended up not
2456 * having a journal_head, return NULL
2458 struct journal_head *jbd2_journal_grab_journal_head(struct buffer_head *bh)
2460 struct journal_head *jh = NULL;
2462 jbd_lock_bh_journal_head(bh);
2463 if (buffer_jbd(bh)) {
2467 jbd_unlock_bh_journal_head(bh);
2471 static void __journal_remove_journal_head(struct buffer_head *bh)
2473 struct journal_head *jh = bh2jh(bh);
2475 J_ASSERT_JH(jh, jh->b_jcount >= 0);
2476 J_ASSERT_JH(jh, jh->b_transaction == NULL);
2477 J_ASSERT_JH(jh, jh->b_next_transaction == NULL);
2478 J_ASSERT_JH(jh, jh->b_cp_transaction == NULL);
2479 J_ASSERT_JH(jh, jh->b_jlist == BJ_None);
2480 J_ASSERT_BH(bh, buffer_jbd(bh));
2481 J_ASSERT_BH(bh, jh2bh(jh) == bh);
2482 BUFFER_TRACE(bh, "remove journal_head");
2483 if (jh->b_frozen_data) {
2484 printk(KERN_WARNING "%s: freeing b_frozen_data\n", __func__);
2485 jbd2_free(jh->b_frozen_data, bh->b_size);
2487 if (jh->b_committed_data) {
2488 printk(KERN_WARNING "%s: freeing b_committed_data\n", __func__);
2489 jbd2_free(jh->b_committed_data, bh->b_size);
2491 bh->b_private = NULL;
2492 jh->b_bh = NULL; /* debug, really */
2493 clear_buffer_jbd(bh);
2494 journal_free_journal_head(jh);
2498 * Drop a reference on the passed journal_head. If it fell to zero then
2499 * release the journal_head from the buffer_head.
2501 void jbd2_journal_put_journal_head(struct journal_head *jh)
2503 struct buffer_head *bh = jh2bh(jh);
2505 jbd_lock_bh_journal_head(bh);
2506 J_ASSERT_JH(jh, jh->b_jcount > 0);
2508 if (!jh->b_jcount) {
2509 __journal_remove_journal_head(bh);
2510 jbd_unlock_bh_journal_head(bh);
2513 jbd_unlock_bh_journal_head(bh);
2517 * Initialize jbd inode head
2519 void jbd2_journal_init_jbd_inode(struct jbd2_inode *jinode, struct inode *inode)
2521 jinode->i_transaction = NULL;
2522 jinode->i_next_transaction = NULL;
2523 jinode->i_vfs_inode = inode;
2524 jinode->i_flags = 0;
2525 INIT_LIST_HEAD(&jinode->i_list);
2529 * Function to be called before we start removing inode from memory (i.e.,
2530 * clear_inode() is a fine place to be called from). It removes inode from
2531 * transaction's lists.
2533 void jbd2_journal_release_jbd_inode(journal_t *journal,
2534 struct jbd2_inode *jinode)
2539 spin_lock(&journal->j_list_lock);
2540 /* Is commit writing out inode - we have to wait */
2541 if (test_bit(__JI_COMMIT_RUNNING, &jinode->i_flags)) {
2542 wait_queue_head_t *wq;
2543 DEFINE_WAIT_BIT(wait, &jinode->i_flags, __JI_COMMIT_RUNNING);
2544 wq = bit_waitqueue(&jinode->i_flags, __JI_COMMIT_RUNNING);
2545 prepare_to_wait(wq, &wait.wait, TASK_UNINTERRUPTIBLE);
2546 spin_unlock(&journal->j_list_lock);
2548 finish_wait(wq, &wait.wait);
2552 if (jinode->i_transaction) {
2553 list_del(&jinode->i_list);
2554 jinode->i_transaction = NULL;
2556 spin_unlock(&journal->j_list_lock);
2560 #ifdef CONFIG_PROC_FS
2562 #define JBD2_STATS_PROC_NAME "fs/jbd2"
2564 static void __init jbd2_create_jbd_stats_proc_entry(void)
2566 proc_jbd2_stats = proc_mkdir(JBD2_STATS_PROC_NAME, NULL);
2569 static void __exit jbd2_remove_jbd_stats_proc_entry(void)
2571 if (proc_jbd2_stats)
2572 remove_proc_entry(JBD2_STATS_PROC_NAME, NULL);
2577 #define jbd2_create_jbd_stats_proc_entry() do {} while (0)
2578 #define jbd2_remove_jbd_stats_proc_entry() do {} while (0)
2582 struct kmem_cache *jbd2_handle_cache, *jbd2_inode_cache;
2584 static int __init jbd2_journal_init_handle_cache(void)
2586 jbd2_handle_cache = KMEM_CACHE(jbd2_journal_handle, SLAB_TEMPORARY);
2587 if (jbd2_handle_cache == NULL) {
2588 printk(KERN_EMERG "JBD2: failed to create handle cache\n");
2591 jbd2_inode_cache = KMEM_CACHE(jbd2_inode, 0);
2592 if (jbd2_inode_cache == NULL) {
2593 printk(KERN_EMERG "JBD2: failed to create inode cache\n");
2594 kmem_cache_destroy(jbd2_handle_cache);
2600 static void jbd2_journal_destroy_handle_cache(void)
2602 if (jbd2_handle_cache)
2603 kmem_cache_destroy(jbd2_handle_cache);
2604 if (jbd2_inode_cache)
2605 kmem_cache_destroy(jbd2_inode_cache);
2610 * Module startup and shutdown
2613 static int __init journal_init_caches(void)
2617 ret = jbd2_journal_init_revoke_caches();
2619 ret = jbd2_journal_init_journal_head_cache();
2621 ret = jbd2_journal_init_handle_cache();
2623 ret = jbd2_journal_init_transaction_cache();
2627 static void jbd2_journal_destroy_caches(void)
2629 jbd2_journal_destroy_revoke_caches();
2630 jbd2_journal_destroy_journal_head_cache();
2631 jbd2_journal_destroy_handle_cache();
2632 jbd2_journal_destroy_transaction_cache();
2633 jbd2_journal_destroy_slabs();
2636 static int __init journal_init(void)
2640 BUILD_BUG_ON(sizeof(struct journal_superblock_s) != 1024);
2642 ret = journal_init_caches();
2644 jbd2_create_jbd_stats_proc_entry();
2646 jbd2_journal_destroy_caches();
2651 static void __exit journal_exit(void)
2653 #ifdef CONFIG_JBD2_DEBUG
2654 int n = atomic_read(&nr_journal_heads);
2656 printk(KERN_ERR "JBD2: leaked %d journal_heads!\n", n);
2658 jbd2_remove_jbd_stats_proc_entry();
2659 jbd2_journal_destroy_caches();
2662 MODULE_LICENSE("GPL");
2663 module_init(journal_init);
2664 module_exit(journal_exit);