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/debugfs.h>
39 #include <linux/seq_file.h>
40 #include <linux/math64.h>
41 #include <linux/hash.h>
42 #include <linux/log2.h>
43 #include <linux/vmalloc.h>
44 #include <linux/backing-dev.h>
45 #include <linux/bitops.h>
46 #include <linux/ratelimit.h>
48 #define CREATE_TRACE_POINTS
49 #include <trace/events/jbd2.h>
51 #include <asm/uaccess.h>
54 EXPORT_SYMBOL(jbd2_journal_extend);
55 EXPORT_SYMBOL(jbd2_journal_stop);
56 EXPORT_SYMBOL(jbd2_journal_lock_updates);
57 EXPORT_SYMBOL(jbd2_journal_unlock_updates);
58 EXPORT_SYMBOL(jbd2_journal_get_write_access);
59 EXPORT_SYMBOL(jbd2_journal_get_create_access);
60 EXPORT_SYMBOL(jbd2_journal_get_undo_access);
61 EXPORT_SYMBOL(jbd2_journal_set_triggers);
62 EXPORT_SYMBOL(jbd2_journal_dirty_metadata);
63 EXPORT_SYMBOL(jbd2_journal_release_buffer);
64 EXPORT_SYMBOL(jbd2_journal_forget);
66 EXPORT_SYMBOL(journal_sync_buffer);
68 EXPORT_SYMBOL(jbd2_journal_flush);
69 EXPORT_SYMBOL(jbd2_journal_revoke);
71 EXPORT_SYMBOL(jbd2_journal_init_dev);
72 EXPORT_SYMBOL(jbd2_journal_init_inode);
73 EXPORT_SYMBOL(jbd2_journal_check_used_features);
74 EXPORT_SYMBOL(jbd2_journal_check_available_features);
75 EXPORT_SYMBOL(jbd2_journal_set_features);
76 EXPORT_SYMBOL(jbd2_journal_load);
77 EXPORT_SYMBOL(jbd2_journal_destroy);
78 EXPORT_SYMBOL(jbd2_journal_abort);
79 EXPORT_SYMBOL(jbd2_journal_errno);
80 EXPORT_SYMBOL(jbd2_journal_ack_err);
81 EXPORT_SYMBOL(jbd2_journal_clear_err);
82 EXPORT_SYMBOL(jbd2_log_wait_commit);
83 EXPORT_SYMBOL(jbd2_log_start_commit);
84 EXPORT_SYMBOL(jbd2_journal_start_commit);
85 EXPORT_SYMBOL(jbd2_journal_force_commit_nested);
86 EXPORT_SYMBOL(jbd2_journal_wipe);
87 EXPORT_SYMBOL(jbd2_journal_blocks_per_page);
88 EXPORT_SYMBOL(jbd2_journal_invalidatepage);
89 EXPORT_SYMBOL(jbd2_journal_try_to_free_buffers);
90 EXPORT_SYMBOL(jbd2_journal_force_commit);
91 EXPORT_SYMBOL(jbd2_journal_file_inode);
92 EXPORT_SYMBOL(jbd2_journal_init_jbd_inode);
93 EXPORT_SYMBOL(jbd2_journal_release_jbd_inode);
94 EXPORT_SYMBOL(jbd2_journal_begin_ordered_truncate);
95 EXPORT_SYMBOL(jbd2_inode_cache);
97 static void __journal_abort_soft (journal_t *journal, int errno);
98 static int jbd2_journal_create_slab(size_t slab_size);
100 /* Checksumming functions */
101 int jbd2_verify_csum_type(journal_t *j, journal_superblock_t *sb)
103 if (!JBD2_HAS_INCOMPAT_FEATURE(j, JBD2_FEATURE_INCOMPAT_CSUM_V2))
106 return sb->s_checksum_type == JBD2_CRC32C_CHKSUM;
110 * Helper function used to manage commit timeouts
113 static void commit_timeout(unsigned long __data)
115 struct task_struct * p = (struct task_struct *) __data;
121 * kjournald2: The main thread function used to manage a logging device
124 * This kernel thread is responsible for two things:
126 * 1) COMMIT: Every so often we need to commit the current state of the
127 * filesystem to disk. The journal thread is responsible for writing
128 * all of the metadata buffers to disk.
130 * 2) CHECKPOINT: We cannot reuse a used section of the log file until all
131 * of the data in that part of the log has been rewritten elsewhere on
132 * the disk. Flushing these old buffers to reclaim space in the log is
133 * known as checkpointing, and this thread is responsible for that job.
136 static int kjournald2(void *arg)
138 journal_t *journal = arg;
139 transaction_t *transaction;
142 * Set up an interval timer which can be used to trigger a commit wakeup
143 * after the commit interval expires
145 setup_timer(&journal->j_commit_timer, commit_timeout,
146 (unsigned long)current);
150 /* Record that the journal thread is running */
151 journal->j_task = current;
152 wake_up(&journal->j_wait_done_commit);
155 * And now, wait forever for commit wakeup events.
157 write_lock(&journal->j_state_lock);
160 if (journal->j_flags & JBD2_UNMOUNT)
163 jbd_debug(1, "commit_sequence=%d, commit_request=%d\n",
164 journal->j_commit_sequence, journal->j_commit_request);
166 if (journal->j_commit_sequence != journal->j_commit_request) {
167 jbd_debug(1, "OK, requests differ\n");
168 write_unlock(&journal->j_state_lock);
169 del_timer_sync(&journal->j_commit_timer);
170 jbd2_journal_commit_transaction(journal);
171 write_lock(&journal->j_state_lock);
175 wake_up(&journal->j_wait_done_commit);
176 if (freezing(current)) {
178 * The simpler the better. Flushing journal isn't a
179 * good idea, because that depends on threads that may
180 * be already stopped.
182 jbd_debug(1, "Now suspending kjournald2\n");
183 write_unlock(&journal->j_state_lock);
185 write_lock(&journal->j_state_lock);
188 * We assume on resume that commits are already there,
192 int should_sleep = 1;
194 prepare_to_wait(&journal->j_wait_commit, &wait,
196 if (journal->j_commit_sequence != journal->j_commit_request)
198 transaction = journal->j_running_transaction;
199 if (transaction && time_after_eq(jiffies,
200 transaction->t_expires))
202 if (journal->j_flags & JBD2_UNMOUNT)
205 write_unlock(&journal->j_state_lock);
207 write_lock(&journal->j_state_lock);
209 finish_wait(&journal->j_wait_commit, &wait);
212 jbd_debug(1, "kjournald2 wakes\n");
215 * Were we woken up by a commit wakeup event?
217 transaction = journal->j_running_transaction;
218 if (transaction && time_after_eq(jiffies, transaction->t_expires)) {
219 journal->j_commit_request = transaction->t_tid;
220 jbd_debug(1, "woke because of timeout\n");
225 write_unlock(&journal->j_state_lock);
226 del_timer_sync(&journal->j_commit_timer);
227 journal->j_task = NULL;
228 wake_up(&journal->j_wait_done_commit);
229 jbd_debug(1, "Journal thread exiting.\n");
233 static int jbd2_journal_start_thread(journal_t *journal)
235 struct task_struct *t;
237 t = kthread_run(kjournald2, journal, "jbd2/%s",
242 wait_event(journal->j_wait_done_commit, journal->j_task != NULL);
246 static void journal_kill_thread(journal_t *journal)
248 write_lock(&journal->j_state_lock);
249 journal->j_flags |= JBD2_UNMOUNT;
251 while (journal->j_task) {
252 wake_up(&journal->j_wait_commit);
253 write_unlock(&journal->j_state_lock);
254 wait_event(journal->j_wait_done_commit, journal->j_task == NULL);
255 write_lock(&journal->j_state_lock);
257 write_unlock(&journal->j_state_lock);
261 * jbd2_journal_write_metadata_buffer: write a metadata buffer to the journal.
263 * Writes a metadata buffer to a given disk block. The actual IO is not
264 * performed but a new buffer_head is constructed which labels the data
265 * to be written with the correct destination disk block.
267 * Any magic-number escaping which needs to be done will cause a
268 * copy-out here. If the buffer happens to start with the
269 * JBD2_MAGIC_NUMBER, then we can't write it to the log directly: the
270 * magic number is only written to the log for descripter blocks. In
271 * this case, we copy the data and replace the first word with 0, and we
272 * return a result code which indicates that this buffer needs to be
273 * marked as an escaped buffer in the corresponding log descriptor
274 * block. The missing word can then be restored when the block is read
277 * If the source buffer has already been modified by a new transaction
278 * since we took the last commit snapshot, we use the frozen copy of
279 * that data for IO. If we end up using the existing buffer_head's data
280 * for the write, then we *have* to lock the buffer to prevent anyone
281 * else from using and possibly modifying it while the IO is in
284 * The function returns a pointer to the buffer_heads to be used for IO.
286 * We assume that the journal has already been locked in this function.
293 * Bit 0 set == escape performed on the data
294 * Bit 1 set == buffer copy-out performed (kfree the data after IO)
297 int jbd2_journal_write_metadata_buffer(transaction_t *transaction,
298 struct journal_head *jh_in,
299 struct journal_head **jh_out,
300 unsigned long long blocknr)
302 int need_copy_out = 0;
303 int done_copy_out = 0;
306 struct buffer_head *new_bh;
307 struct journal_head *new_jh;
308 struct page *new_page;
309 unsigned int new_offset;
310 struct buffer_head *bh_in = jh2bh(jh_in);
311 journal_t *journal = transaction->t_journal;
314 * The buffer really shouldn't be locked: only the current committing
315 * transaction is allowed to write it, so nobody else is allowed
318 * akpm: except if we're journalling data, and write() output is
319 * also part of a shared mapping, and another thread has
320 * decided to launch a writepage() against this buffer.
322 J_ASSERT_BH(bh_in, buffer_jbddirty(bh_in));
325 new_bh = alloc_buffer_head(GFP_NOFS);
328 * Failure is not an option, but __GFP_NOFAIL is going
329 * away; so we retry ourselves here.
331 congestion_wait(BLK_RW_ASYNC, HZ/50);
335 /* keep subsequent assertions sane */
337 init_buffer(new_bh, NULL, NULL);
338 atomic_set(&new_bh->b_count, 1);
339 new_jh = jbd2_journal_add_journal_head(new_bh); /* This sleeps */
342 * If a new transaction has already done a buffer copy-out, then
343 * we use that version of the data for the commit.
345 jbd_lock_bh_state(bh_in);
347 if (jh_in->b_frozen_data) {
349 new_page = virt_to_page(jh_in->b_frozen_data);
350 new_offset = offset_in_page(jh_in->b_frozen_data);
352 new_page = jh2bh(jh_in)->b_page;
353 new_offset = offset_in_page(jh2bh(jh_in)->b_data);
356 mapped_data = kmap_atomic(new_page);
358 * Fire data frozen trigger if data already wasn't frozen. Do this
359 * before checking for escaping, as the trigger may modify the magic
360 * offset. If a copy-out happens afterwards, it will have the correct
361 * data in the buffer.
364 jbd2_buffer_frozen_trigger(jh_in, mapped_data + new_offset,
370 if (*((__be32 *)(mapped_data + new_offset)) ==
371 cpu_to_be32(JBD2_MAGIC_NUMBER)) {
375 kunmap_atomic(mapped_data);
378 * Do we need to do a data copy?
380 if (need_copy_out && !done_copy_out) {
383 jbd_unlock_bh_state(bh_in);
384 tmp = jbd2_alloc(bh_in->b_size, GFP_NOFS);
386 jbd2_journal_put_journal_head(new_jh);
389 jbd_lock_bh_state(bh_in);
390 if (jh_in->b_frozen_data) {
391 jbd2_free(tmp, bh_in->b_size);
395 jh_in->b_frozen_data = tmp;
396 mapped_data = kmap_atomic(new_page);
397 memcpy(tmp, mapped_data + new_offset, jh2bh(jh_in)->b_size);
398 kunmap_atomic(mapped_data);
400 new_page = virt_to_page(tmp);
401 new_offset = offset_in_page(tmp);
405 * This isn't strictly necessary, as we're using frozen
406 * data for the escaping, but it keeps consistency with
407 * b_frozen_data usage.
409 jh_in->b_frozen_triggers = jh_in->b_triggers;
413 * Did we need to do an escaping? Now we've done all the
414 * copying, we can finally do so.
417 mapped_data = kmap_atomic(new_page);
418 *((unsigned int *)(mapped_data + new_offset)) = 0;
419 kunmap_atomic(mapped_data);
422 set_bh_page(new_bh, new_page, new_offset);
423 new_jh->b_transaction = NULL;
424 new_bh->b_size = jh2bh(jh_in)->b_size;
425 new_bh->b_bdev = transaction->t_journal->j_dev;
426 new_bh->b_blocknr = blocknr;
427 set_buffer_mapped(new_bh);
428 set_buffer_dirty(new_bh);
433 * The to-be-written buffer needs to get moved to the io queue,
434 * and the original buffer whose contents we are shadowing or
435 * copying is moved to the transaction's shadow queue.
437 JBUFFER_TRACE(jh_in, "file as BJ_Shadow");
438 spin_lock(&journal->j_list_lock);
439 __jbd2_journal_file_buffer(jh_in, transaction, BJ_Shadow);
440 spin_unlock(&journal->j_list_lock);
441 jbd_unlock_bh_state(bh_in);
443 JBUFFER_TRACE(new_jh, "file as BJ_IO");
444 jbd2_journal_file_buffer(new_jh, transaction, BJ_IO);
446 return do_escape | (done_copy_out << 1);
450 * Allocation code for the journal file. Manage the space left in the
451 * journal, so that we can begin checkpointing when appropriate.
455 * __jbd2_log_space_left: Return the number of free blocks left in the journal.
457 * Called with the journal already locked.
459 * Called under j_state_lock
462 int __jbd2_log_space_left(journal_t *journal)
464 int left = journal->j_free;
466 /* assert_spin_locked(&journal->j_state_lock); */
469 * Be pessimistic here about the number of those free blocks which
470 * might be required for log descriptor control blocks.
473 #define MIN_LOG_RESERVED_BLOCKS 32 /* Allow for rounding errors */
475 left -= MIN_LOG_RESERVED_BLOCKS;
484 * Called with j_state_lock locked for writing.
485 * Returns true if a transaction commit was started.
487 int __jbd2_log_start_commit(journal_t *journal, tid_t target)
490 * The only transaction we can possibly wait upon is the
491 * currently running transaction (if it exists). Otherwise,
492 * the target tid must be an old one.
494 if (journal->j_running_transaction &&
495 journal->j_running_transaction->t_tid == target) {
497 * We want a new commit: OK, mark the request and wakeup the
498 * commit thread. We do _not_ do the commit ourselves.
501 journal->j_commit_request = target;
502 jbd_debug(1, "JBD2: requesting commit %d/%d\n",
503 journal->j_commit_request,
504 journal->j_commit_sequence);
505 wake_up(&journal->j_wait_commit);
507 } else if (!tid_geq(journal->j_commit_request, target))
508 /* This should never happen, but if it does, preserve
509 the evidence before kjournald goes into a loop and
510 increments j_commit_sequence beyond all recognition. */
511 WARN_ONCE(1, "JBD2: bad log_start_commit: %u %u %u %u\n",
512 journal->j_commit_request,
513 journal->j_commit_sequence,
514 target, journal->j_running_transaction ?
515 journal->j_running_transaction->t_tid : 0);
519 int jbd2_log_start_commit(journal_t *journal, tid_t tid)
523 write_lock(&journal->j_state_lock);
524 ret = __jbd2_log_start_commit(journal, tid);
525 write_unlock(&journal->j_state_lock);
530 * Force and wait upon a commit if the calling process is not within
531 * transaction. This is used for forcing out undo-protected data which contains
532 * bitmaps, when the fs is running out of space.
534 * We can only force the running transaction if we don't have an active handle;
535 * otherwise, we will deadlock.
537 * Returns true if a transaction was started.
539 int jbd2_journal_force_commit_nested(journal_t *journal)
541 transaction_t *transaction = NULL;
543 int need_to_start = 0;
545 read_lock(&journal->j_state_lock);
546 if (journal->j_running_transaction && !current->journal_info) {
547 transaction = journal->j_running_transaction;
548 if (!tid_geq(journal->j_commit_request, transaction->t_tid))
550 } else if (journal->j_committing_transaction)
551 transaction = journal->j_committing_transaction;
554 read_unlock(&journal->j_state_lock);
555 return 0; /* Nothing to retry */
558 tid = transaction->t_tid;
559 read_unlock(&journal->j_state_lock);
561 jbd2_log_start_commit(journal, tid);
562 jbd2_log_wait_commit(journal, tid);
567 * Start a commit of the current running transaction (if any). Returns true
568 * if a transaction is going to be committed (or is currently already
569 * committing), and fills its tid in at *ptid
571 int jbd2_journal_start_commit(journal_t *journal, tid_t *ptid)
575 write_lock(&journal->j_state_lock);
576 if (journal->j_running_transaction) {
577 tid_t tid = journal->j_running_transaction->t_tid;
579 __jbd2_log_start_commit(journal, tid);
580 /* There's a running transaction and we've just made sure
581 * it's commit has been scheduled. */
585 } else if (journal->j_committing_transaction) {
587 * If ext3_write_super() recently started a commit, then we
588 * have to wait for completion of that transaction
591 *ptid = journal->j_committing_transaction->t_tid;
594 write_unlock(&journal->j_state_lock);
599 * Return 1 if a given transaction has not yet sent barrier request
600 * connected with a transaction commit. If 0 is returned, transaction
601 * may or may not have sent the barrier. Used to avoid sending barrier
602 * twice in common cases.
604 int jbd2_trans_will_send_data_barrier(journal_t *journal, tid_t tid)
607 transaction_t *commit_trans;
609 if (!(journal->j_flags & JBD2_BARRIER))
611 read_lock(&journal->j_state_lock);
612 /* Transaction already committed? */
613 if (tid_geq(journal->j_commit_sequence, tid))
615 commit_trans = journal->j_committing_transaction;
616 if (!commit_trans || commit_trans->t_tid != tid) {
621 * Transaction is being committed and we already proceeded to
622 * submitting a flush to fs partition?
624 if (journal->j_fs_dev != journal->j_dev) {
625 if (!commit_trans->t_need_data_flush ||
626 commit_trans->t_state >= T_COMMIT_DFLUSH)
629 if (commit_trans->t_state >= T_COMMIT_JFLUSH)
634 read_unlock(&journal->j_state_lock);
637 EXPORT_SYMBOL(jbd2_trans_will_send_data_barrier);
640 * Wait for a specified commit to complete.
641 * The caller may not hold the journal lock.
643 int jbd2_log_wait_commit(journal_t *journal, tid_t tid)
647 read_lock(&journal->j_state_lock);
648 #ifdef CONFIG_JBD2_DEBUG
649 if (!tid_geq(journal->j_commit_request, tid)) {
651 "%s: error: j_commit_request=%d, tid=%d\n",
652 __func__, journal->j_commit_request, tid);
655 while (tid_gt(tid, journal->j_commit_sequence)) {
656 jbd_debug(1, "JBD2: want %d, j_commit_sequence=%d\n",
657 tid, journal->j_commit_sequence);
658 wake_up(&journal->j_wait_commit);
659 read_unlock(&journal->j_state_lock);
660 wait_event(journal->j_wait_done_commit,
661 !tid_gt(tid, journal->j_commit_sequence));
662 read_lock(&journal->j_state_lock);
664 read_unlock(&journal->j_state_lock);
666 if (unlikely(is_journal_aborted(journal))) {
667 printk(KERN_EMERG "journal commit I/O error\n");
674 * Log buffer allocation routines:
677 int jbd2_journal_next_log_block(journal_t *journal, unsigned long long *retp)
679 unsigned long blocknr;
681 write_lock(&journal->j_state_lock);
682 J_ASSERT(journal->j_free > 1);
684 blocknr = journal->j_head;
687 if (journal->j_head == journal->j_last)
688 journal->j_head = journal->j_first;
689 write_unlock(&journal->j_state_lock);
690 return jbd2_journal_bmap(journal, blocknr, retp);
694 * Conversion of logical to physical block numbers for the journal
696 * On external journals the journal blocks are identity-mapped, so
697 * this is a no-op. If needed, we can use j_blk_offset - everything is
700 int jbd2_journal_bmap(journal_t *journal, unsigned long blocknr,
701 unsigned long long *retp)
704 unsigned long long ret;
706 if (journal->j_inode) {
707 ret = bmap(journal->j_inode, blocknr);
711 printk(KERN_ALERT "%s: journal block not found "
712 "at offset %lu on %s\n",
713 __func__, blocknr, journal->j_devname);
715 __journal_abort_soft(journal, err);
718 *retp = blocknr; /* +journal->j_blk_offset */
724 * We play buffer_head aliasing tricks to write data/metadata blocks to
725 * the journal without copying their contents, but for journal
726 * descriptor blocks we do need to generate bona fide buffers.
728 * After the caller of jbd2_journal_get_descriptor_buffer() has finished modifying
729 * the buffer's contents they really should run flush_dcache_page(bh->b_page).
730 * But we don't bother doing that, so there will be coherency problems with
731 * mmaps of blockdevs which hold live JBD-controlled filesystems.
733 struct journal_head *jbd2_journal_get_descriptor_buffer(journal_t *journal)
735 struct buffer_head *bh;
736 unsigned long long blocknr;
739 err = jbd2_journal_next_log_block(journal, &blocknr);
744 bh = __getblk(journal->j_dev, blocknr, journal->j_blocksize);
748 memset(bh->b_data, 0, journal->j_blocksize);
749 set_buffer_uptodate(bh);
751 BUFFER_TRACE(bh, "return this buffer");
752 return jbd2_journal_add_journal_head(bh);
756 * Return tid of the oldest transaction in the journal and block in the journal
757 * where the transaction starts.
759 * If the journal is now empty, return which will be the next transaction ID
760 * we will write and where will that transaction start.
762 * The return value is 0 if journal tail cannot be pushed any further, 1 if
765 int jbd2_journal_get_log_tail(journal_t *journal, tid_t *tid,
766 unsigned long *block)
768 transaction_t *transaction;
771 read_lock(&journal->j_state_lock);
772 spin_lock(&journal->j_list_lock);
773 transaction = journal->j_checkpoint_transactions;
775 *tid = transaction->t_tid;
776 *block = transaction->t_log_start;
777 } else if ((transaction = journal->j_committing_transaction) != NULL) {
778 *tid = transaction->t_tid;
779 *block = transaction->t_log_start;
780 } else if ((transaction = journal->j_running_transaction) != NULL) {
781 *tid = transaction->t_tid;
782 *block = journal->j_head;
784 *tid = journal->j_transaction_sequence;
785 *block = journal->j_head;
787 ret = tid_gt(*tid, journal->j_tail_sequence);
788 spin_unlock(&journal->j_list_lock);
789 read_unlock(&journal->j_state_lock);
795 * Update information in journal structure and in on disk journal superblock
796 * about log tail. This function does not check whether information passed in
797 * really pushes log tail further. It's responsibility of the caller to make
798 * sure provided log tail information is valid (e.g. by holding
799 * j_checkpoint_mutex all the time between computing log tail and calling this
800 * function as is the case with jbd2_cleanup_journal_tail()).
802 * Requires j_checkpoint_mutex
804 void __jbd2_update_log_tail(journal_t *journal, tid_t tid, unsigned long block)
808 BUG_ON(!mutex_is_locked(&journal->j_checkpoint_mutex));
811 * We cannot afford for write to remain in drive's caches since as
812 * soon as we update j_tail, next transaction can start reusing journal
813 * space and if we lose sb update during power failure we'd replay
814 * old transaction with possibly newly overwritten data.
816 jbd2_journal_update_sb_log_tail(journal, tid, block, WRITE_FUA);
817 write_lock(&journal->j_state_lock);
818 freed = block - journal->j_tail;
819 if (block < journal->j_tail)
820 freed += journal->j_last - journal->j_first;
822 trace_jbd2_update_log_tail(journal, tid, block, freed);
824 "Cleaning journal tail from %d to %d (offset %lu), "
826 journal->j_tail_sequence, tid, block, freed);
828 journal->j_free += freed;
829 journal->j_tail_sequence = tid;
830 journal->j_tail = block;
831 write_unlock(&journal->j_state_lock);
835 * This is a variaon of __jbd2_update_log_tail which checks for validity of
836 * provided log tail and locks j_checkpoint_mutex. So it is safe against races
837 * with other threads updating log tail.
839 void jbd2_update_log_tail(journal_t *journal, tid_t tid, unsigned long block)
841 mutex_lock(&journal->j_checkpoint_mutex);
842 if (tid_gt(tid, journal->j_tail_sequence))
843 __jbd2_update_log_tail(journal, tid, block);
844 mutex_unlock(&journal->j_checkpoint_mutex);
847 struct jbd2_stats_proc_session {
849 struct transaction_stats_s *stats;
854 static void *jbd2_seq_info_start(struct seq_file *seq, loff_t *pos)
856 return *pos ? NULL : SEQ_START_TOKEN;
859 static void *jbd2_seq_info_next(struct seq_file *seq, void *v, loff_t *pos)
864 static int jbd2_seq_info_show(struct seq_file *seq, void *v)
866 struct jbd2_stats_proc_session *s = seq->private;
868 if (v != SEQ_START_TOKEN)
870 seq_printf(seq, "%lu transaction, each up to %u blocks\n",
872 s->journal->j_max_transaction_buffers);
873 if (s->stats->ts_tid == 0)
875 seq_printf(seq, "average: \n %ums waiting for transaction\n",
876 jiffies_to_msecs(s->stats->run.rs_wait / s->stats->ts_tid));
877 seq_printf(seq, " %ums running transaction\n",
878 jiffies_to_msecs(s->stats->run.rs_running / s->stats->ts_tid));
879 seq_printf(seq, " %ums transaction was being locked\n",
880 jiffies_to_msecs(s->stats->run.rs_locked / s->stats->ts_tid));
881 seq_printf(seq, " %ums flushing data (in ordered mode)\n",
882 jiffies_to_msecs(s->stats->run.rs_flushing / s->stats->ts_tid));
883 seq_printf(seq, " %ums logging transaction\n",
884 jiffies_to_msecs(s->stats->run.rs_logging / s->stats->ts_tid));
885 seq_printf(seq, " %lluus average transaction commit time\n",
886 div_u64(s->journal->j_average_commit_time, 1000));
887 seq_printf(seq, " %lu handles per transaction\n",
888 s->stats->run.rs_handle_count / s->stats->ts_tid);
889 seq_printf(seq, " %lu blocks per transaction\n",
890 s->stats->run.rs_blocks / s->stats->ts_tid);
891 seq_printf(seq, " %lu logged blocks per transaction\n",
892 s->stats->run.rs_blocks_logged / s->stats->ts_tid);
896 static void jbd2_seq_info_stop(struct seq_file *seq, void *v)
900 static const struct seq_operations jbd2_seq_info_ops = {
901 .start = jbd2_seq_info_start,
902 .next = jbd2_seq_info_next,
903 .stop = jbd2_seq_info_stop,
904 .show = jbd2_seq_info_show,
907 static int jbd2_seq_info_open(struct inode *inode, struct file *file)
909 journal_t *journal = PDE(inode)->data;
910 struct jbd2_stats_proc_session *s;
913 s = kmalloc(sizeof(*s), GFP_KERNEL);
916 size = sizeof(struct transaction_stats_s);
917 s->stats = kmalloc(size, GFP_KERNEL);
918 if (s->stats == NULL) {
922 spin_lock(&journal->j_history_lock);
923 memcpy(s->stats, &journal->j_stats, size);
924 s->journal = journal;
925 spin_unlock(&journal->j_history_lock);
927 rc = seq_open(file, &jbd2_seq_info_ops);
929 struct seq_file *m = file->private_data;
939 static int jbd2_seq_info_release(struct inode *inode, struct file *file)
941 struct seq_file *seq = file->private_data;
942 struct jbd2_stats_proc_session *s = seq->private;
945 return seq_release(inode, file);
948 static const struct file_operations jbd2_seq_info_fops = {
949 .owner = THIS_MODULE,
950 .open = jbd2_seq_info_open,
953 .release = jbd2_seq_info_release,
956 static struct proc_dir_entry *proc_jbd2_stats;
958 static void jbd2_stats_proc_init(journal_t *journal)
960 journal->j_proc_entry = proc_mkdir(journal->j_devname, proc_jbd2_stats);
961 if (journal->j_proc_entry) {
962 proc_create_data("info", S_IRUGO, journal->j_proc_entry,
963 &jbd2_seq_info_fops, journal);
967 static void jbd2_stats_proc_exit(journal_t *journal)
969 remove_proc_entry("info", journal->j_proc_entry);
970 remove_proc_entry(journal->j_devname, proc_jbd2_stats);
974 * Management for journal control blocks: functions to create and
975 * destroy journal_t structures, and to initialise and read existing
976 * journal blocks from disk. */
978 /* First: create and setup a journal_t object in memory. We initialise
979 * very few fields yet: that has to wait until we have created the
980 * journal structures from from scratch, or loaded them from disk. */
982 static journal_t * journal_init_common (void)
987 journal = kzalloc(sizeof(*journal), GFP_KERNEL);
991 init_waitqueue_head(&journal->j_wait_transaction_locked);
992 init_waitqueue_head(&journal->j_wait_logspace);
993 init_waitqueue_head(&journal->j_wait_done_commit);
994 init_waitqueue_head(&journal->j_wait_checkpoint);
995 init_waitqueue_head(&journal->j_wait_commit);
996 init_waitqueue_head(&journal->j_wait_updates);
997 mutex_init(&journal->j_barrier);
998 mutex_init(&journal->j_checkpoint_mutex);
999 spin_lock_init(&journal->j_revoke_lock);
1000 spin_lock_init(&journal->j_list_lock);
1001 rwlock_init(&journal->j_state_lock);
1003 journal->j_commit_interval = (HZ * JBD2_DEFAULT_MAX_COMMIT_AGE);
1004 journal->j_min_batch_time = 0;
1005 journal->j_max_batch_time = 15000; /* 15ms */
1007 /* The journal is marked for error until we succeed with recovery! */
1008 journal->j_flags = JBD2_ABORT;
1010 /* Set up a default-sized revoke table for the new mount. */
1011 err = jbd2_journal_init_revoke(journal, JOURNAL_REVOKE_DEFAULT_HASH);
1017 spin_lock_init(&journal->j_history_lock);
1022 /* jbd2_journal_init_dev and jbd2_journal_init_inode:
1024 * Create a journal structure assigned some fixed set of disk blocks to
1025 * the journal. We don't actually touch those disk blocks yet, but we
1026 * need to set up all of the mapping information to tell the journaling
1027 * system where the journal blocks are.
1032 * journal_t * jbd2_journal_init_dev() - creates and initialises a journal structure
1033 * @bdev: Block device on which to create the journal
1034 * @fs_dev: Device which hold journalled filesystem for this journal.
1035 * @start: Block nr Start of journal.
1036 * @len: Length of the journal in blocks.
1037 * @blocksize: blocksize of journalling device
1039 * Returns: a newly created journal_t *
1041 * jbd2_journal_init_dev creates a journal which maps a fixed contiguous
1042 * range of blocks on an arbitrary block device.
1045 journal_t * jbd2_journal_init_dev(struct block_device *bdev,
1046 struct block_device *fs_dev,
1047 unsigned long long start, int len, int blocksize)
1049 journal_t *journal = journal_init_common();
1050 struct buffer_head *bh;
1057 /* journal descriptor can store up to n blocks -bzzz */
1058 journal->j_blocksize = blocksize;
1059 journal->j_dev = bdev;
1060 journal->j_fs_dev = fs_dev;
1061 journal->j_blk_offset = start;
1062 journal->j_maxlen = len;
1063 bdevname(journal->j_dev, journal->j_devname);
1064 p = journal->j_devname;
1065 while ((p = strchr(p, '/')))
1067 jbd2_stats_proc_init(journal);
1068 n = journal->j_blocksize / sizeof(journal_block_tag_t);
1069 journal->j_wbufsize = n;
1070 journal->j_wbuf = kmalloc(n * sizeof(struct buffer_head*), GFP_KERNEL);
1071 if (!journal->j_wbuf) {
1072 printk(KERN_ERR "%s: Can't allocate bhs for commit thread\n",
1077 bh = __getblk(journal->j_dev, start, journal->j_blocksize);
1080 "%s: Cannot get buffer for journal superblock\n",
1084 journal->j_sb_buffer = bh;
1085 journal->j_superblock = (journal_superblock_t *)bh->b_data;
1089 kfree(journal->j_wbuf);
1090 jbd2_stats_proc_exit(journal);
1096 * journal_t * jbd2_journal_init_inode () - creates a journal which maps to a inode.
1097 * @inode: An inode to create the journal in
1099 * jbd2_journal_init_inode creates a journal which maps an on-disk inode as
1100 * the journal. The inode must exist already, must support bmap() and
1101 * must have all data blocks preallocated.
1103 journal_t * jbd2_journal_init_inode (struct inode *inode)
1105 struct buffer_head *bh;
1106 journal_t *journal = journal_init_common();
1110 unsigned long long blocknr;
1115 journal->j_dev = journal->j_fs_dev = inode->i_sb->s_bdev;
1116 journal->j_inode = inode;
1117 bdevname(journal->j_dev, journal->j_devname);
1118 p = journal->j_devname;
1119 while ((p = strchr(p, '/')))
1121 p = journal->j_devname + strlen(journal->j_devname);
1122 sprintf(p, "-%lu", journal->j_inode->i_ino);
1124 "journal %p: inode %s/%ld, size %Ld, bits %d, blksize %ld\n",
1125 journal, inode->i_sb->s_id, inode->i_ino,
1126 (long long) inode->i_size,
1127 inode->i_sb->s_blocksize_bits, inode->i_sb->s_blocksize);
1129 journal->j_maxlen = inode->i_size >> inode->i_sb->s_blocksize_bits;
1130 journal->j_blocksize = inode->i_sb->s_blocksize;
1131 jbd2_stats_proc_init(journal);
1133 /* journal descriptor can store up to n blocks -bzzz */
1134 n = journal->j_blocksize / sizeof(journal_block_tag_t);
1135 journal->j_wbufsize = n;
1136 journal->j_wbuf = kmalloc(n * sizeof(struct buffer_head*), GFP_KERNEL);
1137 if (!journal->j_wbuf) {
1138 printk(KERN_ERR "%s: Can't allocate bhs for commit thread\n",
1143 err = jbd2_journal_bmap(journal, 0, &blocknr);
1144 /* If that failed, give up */
1146 printk(KERN_ERR "%s: Cannot locate journal superblock\n",
1151 bh = __getblk(journal->j_dev, blocknr, journal->j_blocksize);
1154 "%s: Cannot get buffer for journal superblock\n",
1158 journal->j_sb_buffer = bh;
1159 journal->j_superblock = (journal_superblock_t *)bh->b_data;
1163 kfree(journal->j_wbuf);
1164 jbd2_stats_proc_exit(journal);
1170 * If the journal init or create aborts, we need to mark the journal
1171 * superblock as being NULL to prevent the journal destroy from writing
1172 * back a bogus superblock.
1174 static void journal_fail_superblock (journal_t *journal)
1176 struct buffer_head *bh = journal->j_sb_buffer;
1178 journal->j_sb_buffer = NULL;
1182 * Given a journal_t structure, initialise the various fields for
1183 * startup of a new journaling session. We use this both when creating
1184 * a journal, and after recovering an old journal to reset it for
1188 static int journal_reset(journal_t *journal)
1190 journal_superblock_t *sb = journal->j_superblock;
1191 unsigned long long first, last;
1193 first = be32_to_cpu(sb->s_first);
1194 last = be32_to_cpu(sb->s_maxlen);
1195 if (first + JBD2_MIN_JOURNAL_BLOCKS > last + 1) {
1196 printk(KERN_ERR "JBD2: Journal too short (blocks %llu-%llu).\n",
1198 journal_fail_superblock(journal);
1202 journal->j_first = first;
1203 journal->j_last = last;
1205 journal->j_head = first;
1206 journal->j_tail = first;
1207 journal->j_free = last - first;
1209 journal->j_tail_sequence = journal->j_transaction_sequence;
1210 journal->j_commit_sequence = journal->j_transaction_sequence - 1;
1211 journal->j_commit_request = journal->j_commit_sequence;
1213 journal->j_max_transaction_buffers = journal->j_maxlen / 4;
1216 * As a special case, if the on-disk copy is already marked as needing
1217 * no recovery (s_start == 0), then we can safely defer the superblock
1218 * update until the next commit by setting JBD2_FLUSHED. This avoids
1219 * attempting a write to a potential-readonly device.
1221 if (sb->s_start == 0) {
1222 jbd_debug(1, "JBD2: Skipping superblock update on recovered sb "
1223 "(start %ld, seq %d, errno %d)\n",
1224 journal->j_tail, journal->j_tail_sequence,
1226 journal->j_flags |= JBD2_FLUSHED;
1228 /* Lock here to make assertions happy... */
1229 mutex_lock(&journal->j_checkpoint_mutex);
1231 * Update log tail information. We use WRITE_FUA since new
1232 * transaction will start reusing journal space and so we
1233 * must make sure information about current log tail is on
1236 jbd2_journal_update_sb_log_tail(journal,
1237 journal->j_tail_sequence,
1240 mutex_unlock(&journal->j_checkpoint_mutex);
1242 return jbd2_journal_start_thread(journal);
1245 static void jbd2_write_superblock(journal_t *journal, int write_op)
1247 struct buffer_head *bh = journal->j_sb_buffer;
1250 trace_jbd2_write_superblock(journal, write_op);
1251 if (!(journal->j_flags & JBD2_BARRIER))
1252 write_op &= ~(REQ_FUA | REQ_FLUSH);
1254 if (buffer_write_io_error(bh)) {
1256 * Oh, dear. A previous attempt to write the journal
1257 * superblock failed. This could happen because the
1258 * USB device was yanked out. Or it could happen to
1259 * be a transient write error and maybe the block will
1260 * be remapped. Nothing we can do but to retry the
1261 * write and hope for the best.
1263 printk(KERN_ERR "JBD2: previous I/O error detected "
1264 "for journal superblock update for %s.\n",
1265 journal->j_devname);
1266 clear_buffer_write_io_error(bh);
1267 set_buffer_uptodate(bh);
1270 bh->b_end_io = end_buffer_write_sync;
1271 ret = submit_bh(write_op, bh);
1273 if (buffer_write_io_error(bh)) {
1274 clear_buffer_write_io_error(bh);
1275 set_buffer_uptodate(bh);
1279 printk(KERN_ERR "JBD2: Error %d detected when updating "
1280 "journal superblock for %s.\n", ret,
1281 journal->j_devname);
1286 * jbd2_journal_update_sb_log_tail() - Update log tail in journal sb on disk.
1287 * @journal: The journal to update.
1288 * @tail_tid: TID of the new transaction at the tail of the log
1289 * @tail_block: The first block of the transaction at the tail of the log
1290 * @write_op: With which operation should we write the journal sb
1292 * Update a journal's superblock information about log tail and write it to
1293 * disk, waiting for the IO to complete.
1295 void jbd2_journal_update_sb_log_tail(journal_t *journal, tid_t tail_tid,
1296 unsigned long tail_block, int write_op)
1298 journal_superblock_t *sb = journal->j_superblock;
1300 BUG_ON(!mutex_is_locked(&journal->j_checkpoint_mutex));
1301 jbd_debug(1, "JBD2: updating superblock (start %lu, seq %u)\n",
1302 tail_block, tail_tid);
1304 sb->s_sequence = cpu_to_be32(tail_tid);
1305 sb->s_start = cpu_to_be32(tail_block);
1307 jbd2_write_superblock(journal, write_op);
1309 /* Log is no longer empty */
1310 write_lock(&journal->j_state_lock);
1311 WARN_ON(!sb->s_sequence);
1312 journal->j_flags &= ~JBD2_FLUSHED;
1313 write_unlock(&journal->j_state_lock);
1317 * jbd2_mark_journal_empty() - Mark on disk journal as empty.
1318 * @journal: The journal to update.
1320 * Update a journal's dynamic superblock fields to show that journal is empty.
1321 * Write updated superblock to disk waiting for IO to complete.
1323 static void jbd2_mark_journal_empty(journal_t *journal)
1325 journal_superblock_t *sb = journal->j_superblock;
1327 BUG_ON(!mutex_is_locked(&journal->j_checkpoint_mutex));
1328 read_lock(&journal->j_state_lock);
1329 jbd_debug(1, "JBD2: Marking journal as empty (seq %d)\n",
1330 journal->j_tail_sequence);
1332 sb->s_sequence = cpu_to_be32(journal->j_tail_sequence);
1333 sb->s_start = cpu_to_be32(0);
1334 read_unlock(&journal->j_state_lock);
1336 jbd2_write_superblock(journal, WRITE_FUA);
1338 /* Log is no longer empty */
1339 write_lock(&journal->j_state_lock);
1340 journal->j_flags |= JBD2_FLUSHED;
1341 write_unlock(&journal->j_state_lock);
1346 * jbd2_journal_update_sb_errno() - Update error in the journal.
1347 * @journal: The journal to update.
1349 * Update a journal's errno. Write updated superblock to disk waiting for IO
1352 static void jbd2_journal_update_sb_errno(journal_t *journal)
1354 journal_superblock_t *sb = journal->j_superblock;
1356 read_lock(&journal->j_state_lock);
1357 jbd_debug(1, "JBD2: updating superblock error (errno %d)\n",
1359 sb->s_errno = cpu_to_be32(journal->j_errno);
1360 read_unlock(&journal->j_state_lock);
1362 jbd2_write_superblock(journal, WRITE_SYNC);
1366 * Read the superblock for a given journal, performing initial
1367 * validation of the format.
1369 static int journal_get_superblock(journal_t *journal)
1371 struct buffer_head *bh;
1372 journal_superblock_t *sb;
1375 bh = journal->j_sb_buffer;
1377 J_ASSERT(bh != NULL);
1378 if (!buffer_uptodate(bh)) {
1379 ll_rw_block(READ, 1, &bh);
1381 if (!buffer_uptodate(bh)) {
1383 "JBD2: IO error reading journal superblock\n");
1388 if (buffer_verified(bh))
1391 sb = journal->j_superblock;
1395 if (sb->s_header.h_magic != cpu_to_be32(JBD2_MAGIC_NUMBER) ||
1396 sb->s_blocksize != cpu_to_be32(journal->j_blocksize)) {
1397 printk(KERN_WARNING "JBD2: no valid journal superblock found\n");
1401 switch(be32_to_cpu(sb->s_header.h_blocktype)) {
1402 case JBD2_SUPERBLOCK_V1:
1403 journal->j_format_version = 1;
1405 case JBD2_SUPERBLOCK_V2:
1406 journal->j_format_version = 2;
1409 printk(KERN_WARNING "JBD2: unrecognised superblock format ID\n");
1413 if (be32_to_cpu(sb->s_maxlen) < journal->j_maxlen)
1414 journal->j_maxlen = be32_to_cpu(sb->s_maxlen);
1415 else if (be32_to_cpu(sb->s_maxlen) > journal->j_maxlen) {
1416 printk(KERN_WARNING "JBD2: journal file too short\n");
1420 if (be32_to_cpu(sb->s_first) == 0 ||
1421 be32_to_cpu(sb->s_first) >= journal->j_maxlen) {
1423 "JBD2: Invalid start block of journal: %u\n",
1424 be32_to_cpu(sb->s_first));
1428 if (JBD2_HAS_COMPAT_FEATURE(journal, JBD2_FEATURE_COMPAT_CHECKSUM) &&
1429 JBD2_HAS_INCOMPAT_FEATURE(journal, JBD2_FEATURE_INCOMPAT_CSUM_V2)) {
1430 /* Can't have checksum v1 and v2 on at the same time! */
1431 printk(KERN_ERR "JBD: Can't enable checksumming v1 and v2 "
1432 "at the same time!\n");
1436 if (!jbd2_verify_csum_type(journal, sb)) {
1437 printk(KERN_ERR "JBD: Unknown checksum type\n");
1441 /* Load the checksum driver */
1442 if (JBD2_HAS_INCOMPAT_FEATURE(journal, JBD2_FEATURE_INCOMPAT_CSUM_V2)) {
1443 journal->j_chksum_driver = crypto_alloc_shash("crc32c", 0, 0);
1444 if (IS_ERR(journal->j_chksum_driver)) {
1445 printk(KERN_ERR "JBD: Cannot load crc32c driver.\n");
1446 err = PTR_ERR(journal->j_chksum_driver);
1447 journal->j_chksum_driver = NULL;
1452 set_buffer_verified(bh);
1457 journal_fail_superblock(journal);
1462 * Load the on-disk journal superblock and read the key fields into the
1466 static int load_superblock(journal_t *journal)
1469 journal_superblock_t *sb;
1471 err = journal_get_superblock(journal);
1475 sb = journal->j_superblock;
1477 journal->j_tail_sequence = be32_to_cpu(sb->s_sequence);
1478 journal->j_tail = be32_to_cpu(sb->s_start);
1479 journal->j_first = be32_to_cpu(sb->s_first);
1480 journal->j_last = be32_to_cpu(sb->s_maxlen);
1481 journal->j_errno = be32_to_cpu(sb->s_errno);
1488 * int jbd2_journal_load() - Read journal from disk.
1489 * @journal: Journal to act on.
1491 * Given a journal_t structure which tells us which disk blocks contain
1492 * a journal, read the journal from disk to initialise the in-memory
1495 int jbd2_journal_load(journal_t *journal)
1498 journal_superblock_t *sb;
1500 err = load_superblock(journal);
1504 sb = journal->j_superblock;
1505 /* If this is a V2 superblock, then we have to check the
1506 * features flags on it. */
1508 if (journal->j_format_version >= 2) {
1509 if ((sb->s_feature_ro_compat &
1510 ~cpu_to_be32(JBD2_KNOWN_ROCOMPAT_FEATURES)) ||
1511 (sb->s_feature_incompat &
1512 ~cpu_to_be32(JBD2_KNOWN_INCOMPAT_FEATURES))) {
1514 "JBD2: Unrecognised features on journal\n");
1520 * Create a slab for this blocksize
1522 err = jbd2_journal_create_slab(be32_to_cpu(sb->s_blocksize));
1526 /* Let the recovery code check whether it needs to recover any
1527 * data from the journal. */
1528 if (jbd2_journal_recover(journal))
1529 goto recovery_error;
1531 if (journal->j_failed_commit) {
1532 printk(KERN_ERR "JBD2: journal transaction %u on %s "
1533 "is corrupt.\n", journal->j_failed_commit,
1534 journal->j_devname);
1538 /* OK, we've finished with the dynamic journal bits:
1539 * reinitialise the dynamic contents of the superblock in memory
1540 * and reset them on disk. */
1541 if (journal_reset(journal))
1542 goto recovery_error;
1544 journal->j_flags &= ~JBD2_ABORT;
1545 journal->j_flags |= JBD2_LOADED;
1549 printk(KERN_WARNING "JBD2: recovery failed\n");
1554 * void jbd2_journal_destroy() - Release a journal_t structure.
1555 * @journal: Journal to act on.
1557 * Release a journal_t structure once it is no longer in use by the
1559 * Return <0 if we couldn't clean up the journal.
1561 int jbd2_journal_destroy(journal_t *journal)
1565 /* Wait for the commit thread to wake up and die. */
1566 journal_kill_thread(journal);
1568 /* Force a final log commit */
1569 if (journal->j_running_transaction)
1570 jbd2_journal_commit_transaction(journal);
1572 /* Force any old transactions to disk */
1574 /* Totally anal locking here... */
1575 spin_lock(&journal->j_list_lock);
1576 while (journal->j_checkpoint_transactions != NULL) {
1577 spin_unlock(&journal->j_list_lock);
1578 mutex_lock(&journal->j_checkpoint_mutex);
1579 jbd2_log_do_checkpoint(journal);
1580 mutex_unlock(&journal->j_checkpoint_mutex);
1581 spin_lock(&journal->j_list_lock);
1584 J_ASSERT(journal->j_running_transaction == NULL);
1585 J_ASSERT(journal->j_committing_transaction == NULL);
1586 J_ASSERT(journal->j_checkpoint_transactions == NULL);
1587 spin_unlock(&journal->j_list_lock);
1589 if (journal->j_sb_buffer) {
1590 if (!is_journal_aborted(journal)) {
1591 mutex_lock(&journal->j_checkpoint_mutex);
1592 jbd2_mark_journal_empty(journal);
1593 mutex_unlock(&journal->j_checkpoint_mutex);
1596 brelse(journal->j_sb_buffer);
1599 if (journal->j_proc_entry)
1600 jbd2_stats_proc_exit(journal);
1601 if (journal->j_inode)
1602 iput(journal->j_inode);
1603 if (journal->j_revoke)
1604 jbd2_journal_destroy_revoke(journal);
1605 if (journal->j_chksum_driver)
1606 crypto_free_shash(journal->j_chksum_driver);
1607 kfree(journal->j_wbuf);
1615 *int jbd2_journal_check_used_features () - Check if features specified are used.
1616 * @journal: Journal to check.
1617 * @compat: bitmask of compatible features
1618 * @ro: bitmask of features that force read-only mount
1619 * @incompat: bitmask of incompatible features
1621 * Check whether the journal uses all of a given set of
1622 * features. Return true (non-zero) if it does.
1625 int jbd2_journal_check_used_features (journal_t *journal, unsigned long compat,
1626 unsigned long ro, unsigned long incompat)
1628 journal_superblock_t *sb;
1630 if (!compat && !ro && !incompat)
1632 /* Load journal superblock if it is not loaded yet. */
1633 if (journal->j_format_version == 0 &&
1634 journal_get_superblock(journal) != 0)
1636 if (journal->j_format_version == 1)
1639 sb = journal->j_superblock;
1641 if (((be32_to_cpu(sb->s_feature_compat) & compat) == compat) &&
1642 ((be32_to_cpu(sb->s_feature_ro_compat) & ro) == ro) &&
1643 ((be32_to_cpu(sb->s_feature_incompat) & incompat) == incompat))
1650 * int jbd2_journal_check_available_features() - Check feature set in journalling layer
1651 * @journal: Journal to check.
1652 * @compat: bitmask of compatible features
1653 * @ro: bitmask of features that force read-only mount
1654 * @incompat: bitmask of incompatible features
1656 * Check whether the journaling code supports the use of
1657 * all of a given set of features on this journal. Return true
1658 * (non-zero) if it can. */
1660 int jbd2_journal_check_available_features (journal_t *journal, unsigned long compat,
1661 unsigned long ro, unsigned long incompat)
1663 if (!compat && !ro && !incompat)
1666 /* We can support any known requested features iff the
1667 * superblock is in version 2. Otherwise we fail to support any
1668 * extended sb features. */
1670 if (journal->j_format_version != 2)
1673 if ((compat & JBD2_KNOWN_COMPAT_FEATURES) == compat &&
1674 (ro & JBD2_KNOWN_ROCOMPAT_FEATURES) == ro &&
1675 (incompat & JBD2_KNOWN_INCOMPAT_FEATURES) == incompat)
1682 * int jbd2_journal_set_features () - Mark a given journal feature in the superblock
1683 * @journal: Journal to act on.
1684 * @compat: bitmask of compatible features
1685 * @ro: bitmask of features that force read-only mount
1686 * @incompat: bitmask of incompatible features
1688 * Mark a given journal feature as present on the
1689 * superblock. Returns true if the requested features could be set.
1693 int jbd2_journal_set_features (journal_t *journal, unsigned long compat,
1694 unsigned long ro, unsigned long incompat)
1696 #define INCOMPAT_FEATURE_ON(f) \
1697 ((incompat & (f)) && !(sb->s_feature_incompat & cpu_to_be32(f)))
1698 #define COMPAT_FEATURE_ON(f) \
1699 ((compat & (f)) && !(sb->s_feature_compat & cpu_to_be32(f)))
1700 journal_superblock_t *sb;
1702 if (jbd2_journal_check_used_features(journal, compat, ro, incompat))
1705 if (!jbd2_journal_check_available_features(journal, compat, ro, incompat))
1708 /* Asking for checksumming v2 and v1? Only give them v2. */
1709 if (incompat & JBD2_FEATURE_INCOMPAT_CSUM_V2 &&
1710 compat & JBD2_FEATURE_COMPAT_CHECKSUM)
1711 compat &= ~JBD2_FEATURE_COMPAT_CHECKSUM;
1713 jbd_debug(1, "Setting new features 0x%lx/0x%lx/0x%lx\n",
1714 compat, ro, incompat);
1716 sb = journal->j_superblock;
1718 /* If enabling v2 checksums, update superblock */
1719 if (INCOMPAT_FEATURE_ON(JBD2_FEATURE_INCOMPAT_CSUM_V2)) {
1720 sb->s_checksum_type = JBD2_CRC32C_CHKSUM;
1721 sb->s_feature_compat &=
1722 ~cpu_to_be32(JBD2_FEATURE_COMPAT_CHECKSUM);
1724 /* Load the checksum driver */
1725 if (journal->j_chksum_driver == NULL) {
1726 journal->j_chksum_driver = crypto_alloc_shash("crc32c",
1728 if (IS_ERR(journal->j_chksum_driver)) {
1729 printk(KERN_ERR "JBD: Cannot load crc32c "
1731 journal->j_chksum_driver = NULL;
1737 /* If enabling v1 checksums, downgrade superblock */
1738 if (COMPAT_FEATURE_ON(JBD2_FEATURE_COMPAT_CHECKSUM))
1739 sb->s_feature_incompat &=
1740 ~cpu_to_be32(JBD2_FEATURE_INCOMPAT_CSUM_V2);
1742 sb->s_feature_compat |= cpu_to_be32(compat);
1743 sb->s_feature_ro_compat |= cpu_to_be32(ro);
1744 sb->s_feature_incompat |= cpu_to_be32(incompat);
1747 #undef COMPAT_FEATURE_ON
1748 #undef INCOMPAT_FEATURE_ON
1752 * jbd2_journal_clear_features () - Clear a given journal feature in the
1754 * @journal: Journal to act on.
1755 * @compat: bitmask of compatible features
1756 * @ro: bitmask of features that force read-only mount
1757 * @incompat: bitmask of incompatible features
1759 * Clear a given journal feature as present on the
1762 void jbd2_journal_clear_features(journal_t *journal, unsigned long compat,
1763 unsigned long ro, unsigned long incompat)
1765 journal_superblock_t *sb;
1767 jbd_debug(1, "Clear features 0x%lx/0x%lx/0x%lx\n",
1768 compat, ro, incompat);
1770 sb = journal->j_superblock;
1772 sb->s_feature_compat &= ~cpu_to_be32(compat);
1773 sb->s_feature_ro_compat &= ~cpu_to_be32(ro);
1774 sb->s_feature_incompat &= ~cpu_to_be32(incompat);
1776 EXPORT_SYMBOL(jbd2_journal_clear_features);
1779 * int jbd2_journal_flush () - Flush journal
1780 * @journal: Journal to act on.
1782 * Flush all data for a given journal to disk and empty the journal.
1783 * Filesystems can use this when remounting readonly to ensure that
1784 * recovery does not need to happen on remount.
1787 int jbd2_journal_flush(journal_t *journal)
1790 transaction_t *transaction = NULL;
1792 write_lock(&journal->j_state_lock);
1794 /* Force everything buffered to the log... */
1795 if (journal->j_running_transaction) {
1796 transaction = journal->j_running_transaction;
1797 __jbd2_log_start_commit(journal, transaction->t_tid);
1798 } else if (journal->j_committing_transaction)
1799 transaction = journal->j_committing_transaction;
1801 /* Wait for the log commit to complete... */
1803 tid_t tid = transaction->t_tid;
1805 write_unlock(&journal->j_state_lock);
1806 jbd2_log_wait_commit(journal, tid);
1808 write_unlock(&journal->j_state_lock);
1811 /* ...and flush everything in the log out to disk. */
1812 spin_lock(&journal->j_list_lock);
1813 while (!err && journal->j_checkpoint_transactions != NULL) {
1814 spin_unlock(&journal->j_list_lock);
1815 mutex_lock(&journal->j_checkpoint_mutex);
1816 err = jbd2_log_do_checkpoint(journal);
1817 mutex_unlock(&journal->j_checkpoint_mutex);
1818 spin_lock(&journal->j_list_lock);
1820 spin_unlock(&journal->j_list_lock);
1822 if (is_journal_aborted(journal))
1825 mutex_lock(&journal->j_checkpoint_mutex);
1826 jbd2_cleanup_journal_tail(journal);
1828 /* Finally, mark the journal as really needing no recovery.
1829 * This sets s_start==0 in the underlying superblock, which is
1830 * the magic code for a fully-recovered superblock. Any future
1831 * commits of data to the journal will restore the current
1833 jbd2_mark_journal_empty(journal);
1834 mutex_unlock(&journal->j_checkpoint_mutex);
1835 write_lock(&journal->j_state_lock);
1836 J_ASSERT(!journal->j_running_transaction);
1837 J_ASSERT(!journal->j_committing_transaction);
1838 J_ASSERT(!journal->j_checkpoint_transactions);
1839 J_ASSERT(journal->j_head == journal->j_tail);
1840 J_ASSERT(journal->j_tail_sequence == journal->j_transaction_sequence);
1841 write_unlock(&journal->j_state_lock);
1846 * int jbd2_journal_wipe() - Wipe journal contents
1847 * @journal: Journal to act on.
1848 * @write: flag (see below)
1850 * Wipe out all of the contents of a journal, safely. This will produce
1851 * a warning if the journal contains any valid recovery information.
1852 * Must be called between journal_init_*() and jbd2_journal_load().
1854 * If 'write' is non-zero, then we wipe out the journal on disk; otherwise
1855 * we merely suppress recovery.
1858 int jbd2_journal_wipe(journal_t *journal, int write)
1862 J_ASSERT (!(journal->j_flags & JBD2_LOADED));
1864 err = load_superblock(journal);
1868 if (!journal->j_tail)
1871 printk(KERN_WARNING "JBD2: %s recovery information on journal\n",
1872 write ? "Clearing" : "Ignoring");
1874 err = jbd2_journal_skip_recovery(journal);
1876 /* Lock to make assertions happy... */
1877 mutex_lock(&journal->j_checkpoint_mutex);
1878 jbd2_mark_journal_empty(journal);
1879 mutex_unlock(&journal->j_checkpoint_mutex);
1887 * Journal abort has very specific semantics, which we describe
1888 * for journal abort.
1890 * Two internal functions, which provide abort to the jbd layer
1895 * Quick version for internal journal use (doesn't lock the journal).
1896 * Aborts hard --- we mark the abort as occurred, but do _nothing_ else,
1897 * and don't attempt to make any other journal updates.
1899 void __jbd2_journal_abort_hard(journal_t *journal)
1901 transaction_t *transaction;
1903 if (journal->j_flags & JBD2_ABORT)
1906 printk(KERN_ERR "Aborting journal on device %s.\n",
1907 journal->j_devname);
1909 write_lock(&journal->j_state_lock);
1910 journal->j_flags |= JBD2_ABORT;
1911 transaction = journal->j_running_transaction;
1913 __jbd2_log_start_commit(journal, transaction->t_tid);
1914 write_unlock(&journal->j_state_lock);
1917 /* Soft abort: record the abort error status in the journal superblock,
1918 * but don't do any other IO. */
1919 static void __journal_abort_soft (journal_t *journal, int errno)
1921 if (journal->j_flags & JBD2_ABORT)
1924 if (!journal->j_errno)
1925 journal->j_errno = errno;
1927 __jbd2_journal_abort_hard(journal);
1930 jbd2_journal_update_sb_errno(journal);
1934 * void jbd2_journal_abort () - Shutdown the journal immediately.
1935 * @journal: the journal to shutdown.
1936 * @errno: an error number to record in the journal indicating
1937 * the reason for the shutdown.
1939 * Perform a complete, immediate shutdown of the ENTIRE
1940 * journal (not of a single transaction). This operation cannot be
1941 * undone without closing and reopening the journal.
1943 * The jbd2_journal_abort function is intended to support higher level error
1944 * recovery mechanisms such as the ext2/ext3 remount-readonly error
1947 * Journal abort has very specific semantics. Any existing dirty,
1948 * unjournaled buffers in the main filesystem will still be written to
1949 * disk by bdflush, but the journaling mechanism will be suspended
1950 * immediately and no further transaction commits will be honoured.
1952 * Any dirty, journaled buffers will be written back to disk without
1953 * hitting the journal. Atomicity cannot be guaranteed on an aborted
1954 * filesystem, but we _do_ attempt to leave as much data as possible
1955 * behind for fsck to use for cleanup.
1957 * Any attempt to get a new transaction handle on a journal which is in
1958 * ABORT state will just result in an -EROFS error return. A
1959 * jbd2_journal_stop on an existing handle will return -EIO if we have
1960 * entered abort state during the update.
1962 * Recursive transactions are not disturbed by journal abort until the
1963 * final jbd2_journal_stop, which will receive the -EIO error.
1965 * Finally, the jbd2_journal_abort call allows the caller to supply an errno
1966 * which will be recorded (if possible) in the journal superblock. This
1967 * allows a client to record failure conditions in the middle of a
1968 * transaction without having to complete the transaction to record the
1969 * failure to disk. ext3_error, for example, now uses this
1972 * Errors which originate from within the journaling layer will NOT
1973 * supply an errno; a null errno implies that absolutely no further
1974 * writes are done to the journal (unless there are any already in
1979 void jbd2_journal_abort(journal_t *journal, int errno)
1981 __journal_abort_soft(journal, errno);
1985 * int jbd2_journal_errno () - returns the journal's error state.
1986 * @journal: journal to examine.
1988 * This is the errno number set with jbd2_journal_abort(), the last
1989 * time the journal was mounted - if the journal was stopped
1990 * without calling abort this will be 0.
1992 * If the journal has been aborted on this mount time -EROFS will
1995 int jbd2_journal_errno(journal_t *journal)
1999 read_lock(&journal->j_state_lock);
2000 if (journal->j_flags & JBD2_ABORT)
2003 err = journal->j_errno;
2004 read_unlock(&journal->j_state_lock);
2009 * int jbd2_journal_clear_err () - clears the journal's error state
2010 * @journal: journal to act on.
2012 * An error must be cleared or acked to take a FS out of readonly
2015 int jbd2_journal_clear_err(journal_t *journal)
2019 write_lock(&journal->j_state_lock);
2020 if (journal->j_flags & JBD2_ABORT)
2023 journal->j_errno = 0;
2024 write_unlock(&journal->j_state_lock);
2029 * void jbd2_journal_ack_err() - Ack journal err.
2030 * @journal: journal to act on.
2032 * An error must be cleared or acked to take a FS out of readonly
2035 void jbd2_journal_ack_err(journal_t *journal)
2037 write_lock(&journal->j_state_lock);
2038 if (journal->j_errno)
2039 journal->j_flags |= JBD2_ACK_ERR;
2040 write_unlock(&journal->j_state_lock);
2043 int jbd2_journal_blocks_per_page(struct inode *inode)
2045 return 1 << (PAGE_CACHE_SHIFT - inode->i_sb->s_blocksize_bits);
2049 * helper functions to deal with 32 or 64bit block numbers.
2051 size_t journal_tag_bytes(journal_t *journal)
2053 if (JBD2_HAS_INCOMPAT_FEATURE(journal, JBD2_FEATURE_INCOMPAT_64BIT))
2054 return JBD2_TAG_SIZE64;
2056 return JBD2_TAG_SIZE32;
2060 * JBD memory management
2062 * These functions are used to allocate block-sized chunks of memory
2063 * used for making copies of buffer_head data. Very often it will be
2064 * page-sized chunks of data, but sometimes it will be in
2065 * sub-page-size chunks. (For example, 16k pages on Power systems
2066 * with a 4k block file system.) For blocks smaller than a page, we
2067 * use a SLAB allocator. There are slab caches for each block size,
2068 * which are allocated at mount time, if necessary, and we only free
2069 * (all of) the slab caches when/if the jbd2 module is unloaded. For
2070 * this reason we don't need to a mutex to protect access to
2071 * jbd2_slab[] allocating or releasing memory; only in
2072 * jbd2_journal_create_slab().
2074 #define JBD2_MAX_SLABS 8
2075 static struct kmem_cache *jbd2_slab[JBD2_MAX_SLABS];
2077 static const char *jbd2_slab_names[JBD2_MAX_SLABS] = {
2078 "jbd2_1k", "jbd2_2k", "jbd2_4k", "jbd2_8k",
2079 "jbd2_16k", "jbd2_32k", "jbd2_64k", "jbd2_128k"
2083 static void jbd2_journal_destroy_slabs(void)
2087 for (i = 0; i < JBD2_MAX_SLABS; i++) {
2089 kmem_cache_destroy(jbd2_slab[i]);
2090 jbd2_slab[i] = NULL;
2094 static int jbd2_journal_create_slab(size_t size)
2096 static DEFINE_MUTEX(jbd2_slab_create_mutex);
2097 int i = order_base_2(size) - 10;
2100 if (size == PAGE_SIZE)
2103 if (i >= JBD2_MAX_SLABS)
2106 if (unlikely(i < 0))
2108 mutex_lock(&jbd2_slab_create_mutex);
2110 mutex_unlock(&jbd2_slab_create_mutex);
2111 return 0; /* Already created */
2114 slab_size = 1 << (i+10);
2115 jbd2_slab[i] = kmem_cache_create(jbd2_slab_names[i], slab_size,
2116 slab_size, 0, NULL);
2117 mutex_unlock(&jbd2_slab_create_mutex);
2118 if (!jbd2_slab[i]) {
2119 printk(KERN_EMERG "JBD2: no memory for jbd2_slab cache\n");
2125 static struct kmem_cache *get_slab(size_t size)
2127 int i = order_base_2(size) - 10;
2129 BUG_ON(i >= JBD2_MAX_SLABS);
2130 if (unlikely(i < 0))
2132 BUG_ON(jbd2_slab[i] == NULL);
2133 return jbd2_slab[i];
2136 void *jbd2_alloc(size_t size, gfp_t flags)
2140 BUG_ON(size & (size-1)); /* Must be a power of 2 */
2142 flags |= __GFP_REPEAT;
2143 if (size == PAGE_SIZE)
2144 ptr = (void *)__get_free_pages(flags, 0);
2145 else if (size > PAGE_SIZE) {
2146 int order = get_order(size);
2149 ptr = (void *)__get_free_pages(flags, order);
2151 ptr = vmalloc(size);
2153 ptr = kmem_cache_alloc(get_slab(size), flags);
2155 /* Check alignment; SLUB has gotten this wrong in the past,
2156 * and this can lead to user data corruption! */
2157 BUG_ON(((unsigned long) ptr) & (size-1));
2162 void jbd2_free(void *ptr, size_t size)
2164 if (size == PAGE_SIZE) {
2165 free_pages((unsigned long)ptr, 0);
2168 if (size > PAGE_SIZE) {
2169 int order = get_order(size);
2172 free_pages((unsigned long)ptr, order);
2177 kmem_cache_free(get_slab(size), ptr);
2181 * Journal_head storage management
2183 static struct kmem_cache *jbd2_journal_head_cache;
2184 #ifdef CONFIG_JBD2_DEBUG
2185 static atomic_t nr_journal_heads = ATOMIC_INIT(0);
2188 static int jbd2_journal_init_journal_head_cache(void)
2192 J_ASSERT(jbd2_journal_head_cache == NULL);
2193 jbd2_journal_head_cache = kmem_cache_create("jbd2_journal_head",
2194 sizeof(struct journal_head),
2196 SLAB_TEMPORARY, /* flags */
2199 if (!jbd2_journal_head_cache) {
2201 printk(KERN_EMERG "JBD2: no memory for journal_head cache\n");
2206 static void jbd2_journal_destroy_journal_head_cache(void)
2208 if (jbd2_journal_head_cache) {
2209 kmem_cache_destroy(jbd2_journal_head_cache);
2210 jbd2_journal_head_cache = NULL;
2215 * journal_head splicing and dicing
2217 static struct journal_head *journal_alloc_journal_head(void)
2219 struct journal_head *ret;
2221 #ifdef CONFIG_JBD2_DEBUG
2222 atomic_inc(&nr_journal_heads);
2224 ret = kmem_cache_alloc(jbd2_journal_head_cache, GFP_NOFS);
2226 jbd_debug(1, "out of memory for journal_head\n");
2227 pr_notice_ratelimited("ENOMEM in %s, retrying.\n", __func__);
2230 ret = kmem_cache_alloc(jbd2_journal_head_cache, GFP_NOFS);
2236 static void journal_free_journal_head(struct journal_head *jh)
2238 #ifdef CONFIG_JBD2_DEBUG
2239 atomic_dec(&nr_journal_heads);
2240 memset(jh, JBD2_POISON_FREE, sizeof(*jh));
2242 kmem_cache_free(jbd2_journal_head_cache, jh);
2246 * A journal_head is attached to a buffer_head whenever JBD has an
2247 * interest in the buffer.
2249 * Whenever a buffer has an attached journal_head, its ->b_state:BH_JBD bit
2250 * is set. This bit is tested in core kernel code where we need to take
2251 * JBD-specific actions. Testing the zeroness of ->b_private is not reliable
2254 * When a buffer has its BH_JBD bit set, its ->b_count is elevated by one.
2256 * When a buffer has its BH_JBD bit set it is immune from being released by
2257 * core kernel code, mainly via ->b_count.
2259 * A journal_head is detached from its buffer_head when the journal_head's
2260 * b_jcount reaches zero. Running transaction (b_transaction) and checkpoint
2261 * transaction (b_cp_transaction) hold their references to b_jcount.
2263 * Various places in the kernel want to attach a journal_head to a buffer_head
2264 * _before_ attaching the journal_head to a transaction. To protect the
2265 * journal_head in this situation, jbd2_journal_add_journal_head elevates the
2266 * journal_head's b_jcount refcount by one. The caller must call
2267 * jbd2_journal_put_journal_head() to undo this.
2269 * So the typical usage would be:
2271 * (Attach a journal_head if needed. Increments b_jcount)
2272 * struct journal_head *jh = jbd2_journal_add_journal_head(bh);
2274 * (Get another reference for transaction)
2275 * jbd2_journal_grab_journal_head(bh);
2276 * jh->b_transaction = xxx;
2277 * (Put original reference)
2278 * jbd2_journal_put_journal_head(jh);
2282 * Give a buffer_head a journal_head.
2286 struct journal_head *jbd2_journal_add_journal_head(struct buffer_head *bh)
2288 struct journal_head *jh;
2289 struct journal_head *new_jh = NULL;
2292 if (!buffer_jbd(bh)) {
2293 new_jh = journal_alloc_journal_head();
2294 memset(new_jh, 0, sizeof(*new_jh));
2297 jbd_lock_bh_journal_head(bh);
2298 if (buffer_jbd(bh)) {
2302 (atomic_read(&bh->b_count) > 0) ||
2303 (bh->b_page && bh->b_page->mapping));
2306 jbd_unlock_bh_journal_head(bh);
2311 new_jh = NULL; /* We consumed it */
2316 BUFFER_TRACE(bh, "added journal_head");
2319 jbd_unlock_bh_journal_head(bh);
2321 journal_free_journal_head(new_jh);
2322 return bh->b_private;
2326 * Grab a ref against this buffer_head's journal_head. If it ended up not
2327 * having a journal_head, return NULL
2329 struct journal_head *jbd2_journal_grab_journal_head(struct buffer_head *bh)
2331 struct journal_head *jh = NULL;
2333 jbd_lock_bh_journal_head(bh);
2334 if (buffer_jbd(bh)) {
2338 jbd_unlock_bh_journal_head(bh);
2342 static void __journal_remove_journal_head(struct buffer_head *bh)
2344 struct journal_head *jh = bh2jh(bh);
2346 J_ASSERT_JH(jh, jh->b_jcount >= 0);
2347 J_ASSERT_JH(jh, jh->b_transaction == NULL);
2348 J_ASSERT_JH(jh, jh->b_next_transaction == NULL);
2349 J_ASSERT_JH(jh, jh->b_cp_transaction == NULL);
2350 J_ASSERT_JH(jh, jh->b_jlist == BJ_None);
2351 J_ASSERT_BH(bh, buffer_jbd(bh));
2352 J_ASSERT_BH(bh, jh2bh(jh) == bh);
2353 BUFFER_TRACE(bh, "remove journal_head");
2354 if (jh->b_frozen_data) {
2355 printk(KERN_WARNING "%s: freeing b_frozen_data\n", __func__);
2356 jbd2_free(jh->b_frozen_data, bh->b_size);
2358 if (jh->b_committed_data) {
2359 printk(KERN_WARNING "%s: freeing b_committed_data\n", __func__);
2360 jbd2_free(jh->b_committed_data, bh->b_size);
2362 bh->b_private = NULL;
2363 jh->b_bh = NULL; /* debug, really */
2364 clear_buffer_jbd(bh);
2365 journal_free_journal_head(jh);
2369 * Drop a reference on the passed journal_head. If it fell to zero then
2370 * release the journal_head from the buffer_head.
2372 void jbd2_journal_put_journal_head(struct journal_head *jh)
2374 struct buffer_head *bh = jh2bh(jh);
2376 jbd_lock_bh_journal_head(bh);
2377 J_ASSERT_JH(jh, jh->b_jcount > 0);
2379 if (!jh->b_jcount) {
2380 __journal_remove_journal_head(bh);
2381 jbd_unlock_bh_journal_head(bh);
2384 jbd_unlock_bh_journal_head(bh);
2388 * Initialize jbd inode head
2390 void jbd2_journal_init_jbd_inode(struct jbd2_inode *jinode, struct inode *inode)
2392 jinode->i_transaction = NULL;
2393 jinode->i_next_transaction = NULL;
2394 jinode->i_vfs_inode = inode;
2395 jinode->i_flags = 0;
2396 INIT_LIST_HEAD(&jinode->i_list);
2400 * Function to be called before we start removing inode from memory (i.e.,
2401 * clear_inode() is a fine place to be called from). It removes inode from
2402 * transaction's lists.
2404 void jbd2_journal_release_jbd_inode(journal_t *journal,
2405 struct jbd2_inode *jinode)
2410 spin_lock(&journal->j_list_lock);
2411 /* Is commit writing out inode - we have to wait */
2412 if (test_bit(__JI_COMMIT_RUNNING, &jinode->i_flags)) {
2413 wait_queue_head_t *wq;
2414 DEFINE_WAIT_BIT(wait, &jinode->i_flags, __JI_COMMIT_RUNNING);
2415 wq = bit_waitqueue(&jinode->i_flags, __JI_COMMIT_RUNNING);
2416 prepare_to_wait(wq, &wait.wait, TASK_UNINTERRUPTIBLE);
2417 spin_unlock(&journal->j_list_lock);
2419 finish_wait(wq, &wait.wait);
2423 if (jinode->i_transaction) {
2424 list_del(&jinode->i_list);
2425 jinode->i_transaction = NULL;
2427 spin_unlock(&journal->j_list_lock);
2433 #ifdef CONFIG_JBD2_DEBUG
2434 u8 jbd2_journal_enable_debug __read_mostly;
2435 EXPORT_SYMBOL(jbd2_journal_enable_debug);
2437 #define JBD2_DEBUG_NAME "jbd2-debug"
2439 static struct dentry *jbd2_debugfs_dir;
2440 static struct dentry *jbd2_debug;
2442 static void __init jbd2_create_debugfs_entry(void)
2444 jbd2_debugfs_dir = debugfs_create_dir("jbd2", NULL);
2445 if (jbd2_debugfs_dir)
2446 jbd2_debug = debugfs_create_u8(JBD2_DEBUG_NAME,
2449 &jbd2_journal_enable_debug);
2452 static void __exit jbd2_remove_debugfs_entry(void)
2454 debugfs_remove(jbd2_debug);
2455 debugfs_remove(jbd2_debugfs_dir);
2460 static void __init jbd2_create_debugfs_entry(void)
2464 static void __exit jbd2_remove_debugfs_entry(void)
2470 #ifdef CONFIG_PROC_FS
2472 #define JBD2_STATS_PROC_NAME "fs/jbd2"
2474 static void __init jbd2_create_jbd_stats_proc_entry(void)
2476 proc_jbd2_stats = proc_mkdir(JBD2_STATS_PROC_NAME, NULL);
2479 static void __exit jbd2_remove_jbd_stats_proc_entry(void)
2481 if (proc_jbd2_stats)
2482 remove_proc_entry(JBD2_STATS_PROC_NAME, NULL);
2487 #define jbd2_create_jbd_stats_proc_entry() do {} while (0)
2488 #define jbd2_remove_jbd_stats_proc_entry() do {} while (0)
2492 struct kmem_cache *jbd2_handle_cache, *jbd2_inode_cache;
2494 static int __init jbd2_journal_init_handle_cache(void)
2496 jbd2_handle_cache = KMEM_CACHE(jbd2_journal_handle, SLAB_TEMPORARY);
2497 if (jbd2_handle_cache == NULL) {
2498 printk(KERN_EMERG "JBD2: failed to create handle cache\n");
2501 jbd2_inode_cache = KMEM_CACHE(jbd2_inode, 0);
2502 if (jbd2_inode_cache == NULL) {
2503 printk(KERN_EMERG "JBD2: failed to create inode cache\n");
2504 kmem_cache_destroy(jbd2_handle_cache);
2510 static void jbd2_journal_destroy_handle_cache(void)
2512 if (jbd2_handle_cache)
2513 kmem_cache_destroy(jbd2_handle_cache);
2514 if (jbd2_inode_cache)
2515 kmem_cache_destroy(jbd2_inode_cache);
2520 * Module startup and shutdown
2523 static int __init journal_init_caches(void)
2527 ret = jbd2_journal_init_revoke_caches();
2529 ret = jbd2_journal_init_journal_head_cache();
2531 ret = jbd2_journal_init_handle_cache();
2533 ret = jbd2_journal_init_transaction_cache();
2537 static void jbd2_journal_destroy_caches(void)
2539 jbd2_journal_destroy_revoke_caches();
2540 jbd2_journal_destroy_journal_head_cache();
2541 jbd2_journal_destroy_handle_cache();
2542 jbd2_journal_destroy_transaction_cache();
2543 jbd2_journal_destroy_slabs();
2546 static int __init journal_init(void)
2550 BUILD_BUG_ON(sizeof(struct journal_superblock_s) != 1024);
2552 ret = journal_init_caches();
2554 jbd2_create_debugfs_entry();
2555 jbd2_create_jbd_stats_proc_entry();
2557 jbd2_journal_destroy_caches();
2562 static void __exit journal_exit(void)
2564 #ifdef CONFIG_JBD2_DEBUG
2565 int n = atomic_read(&nr_journal_heads);
2567 printk(KERN_EMERG "JBD2: leaked %d journal_heads!\n", n);
2569 jbd2_remove_debugfs_entry();
2570 jbd2_remove_jbd_stats_proc_entry();
2571 jbd2_journal_destroy_caches();
2574 MODULE_LICENSE("GPL");
2575 module_init(journal_init);
2576 module_exit(journal_exit);