2 * linux/fs/jbd2/journal.c
4 * Written by Stephen C. Tweedie <sct@redhat.com>, 1998
6 * Copyright 1998 Red Hat corp --- All Rights Reserved
8 * This file is part of the Linux kernel and is made available under
9 * the terms of the GNU General Public License, version 2, or at your
10 * option, any later version, incorporated herein by reference.
12 * Generic filesystem journal-writing code; part of the ext2fs
15 * This file manages journals: areas of disk reserved for logging
16 * transactional updates. This includes the kernel journaling thread
17 * which is responsible for scheduling updates to the log.
19 * We do not actually manage the physical storage of the journal in this
20 * file: that is left to a per-journal policy function, which allows us
21 * to store the journal within a filesystem-specified area for ext2
22 * journaling (ext2 can use a reserved inode for storing the log).
25 #include <linux/module.h>
26 #include <linux/time.h>
28 #include <linux/jbd2.h>
29 #include <linux/errno.h>
30 #include <linux/slab.h>
31 #include <linux/init.h>
33 #include <linux/freezer.h>
34 #include <linux/pagemap.h>
35 #include <linux/kthread.h>
36 #include <linux/poison.h>
37 #include <linux/proc_fs.h>
38 #include <linux/seq_file.h>
39 #include <linux/math64.h>
40 #include <linux/hash.h>
41 #include <linux/log2.h>
42 #include <linux/vmalloc.h>
43 #include <linux/backing-dev.h>
44 #include <linux/bitops.h>
45 #include <linux/ratelimit.h>
47 #define CREATE_TRACE_POINTS
48 #include <trace/events/jbd2.h>
50 #include <asm/uaccess.h>
53 #ifdef CONFIG_JBD2_DEBUG
54 ushort jbd2_journal_enable_debug __read_mostly;
55 EXPORT_SYMBOL(jbd2_journal_enable_debug);
57 module_param_named(jbd2_debug, jbd2_journal_enable_debug, ushort, 0644);
58 MODULE_PARM_DESC(jbd2_debug, "Debugging level for jbd2");
61 EXPORT_SYMBOL(jbd2_journal_extend);
62 EXPORT_SYMBOL(jbd2_journal_stop);
63 EXPORT_SYMBOL(jbd2_journal_lock_updates);
64 EXPORT_SYMBOL(jbd2_journal_unlock_updates);
65 EXPORT_SYMBOL(jbd2_journal_get_write_access);
66 EXPORT_SYMBOL(jbd2_journal_get_create_access);
67 EXPORT_SYMBOL(jbd2_journal_get_undo_access);
68 EXPORT_SYMBOL(jbd2_journal_set_triggers);
69 EXPORT_SYMBOL(jbd2_journal_dirty_metadata);
70 EXPORT_SYMBOL(jbd2_journal_forget);
72 EXPORT_SYMBOL(journal_sync_buffer);
74 EXPORT_SYMBOL(jbd2_journal_flush);
75 EXPORT_SYMBOL(jbd2_journal_revoke);
77 EXPORT_SYMBOL(jbd2_journal_init_dev);
78 EXPORT_SYMBOL(jbd2_journal_init_inode);
79 EXPORT_SYMBOL(jbd2_journal_check_used_features);
80 EXPORT_SYMBOL(jbd2_journal_check_available_features);
81 EXPORT_SYMBOL(jbd2_journal_set_features);
82 EXPORT_SYMBOL(jbd2_journal_load);
83 EXPORT_SYMBOL(jbd2_journal_destroy);
84 EXPORT_SYMBOL(jbd2_journal_abort);
85 EXPORT_SYMBOL(jbd2_journal_errno);
86 EXPORT_SYMBOL(jbd2_journal_ack_err);
87 EXPORT_SYMBOL(jbd2_journal_clear_err);
88 EXPORT_SYMBOL(jbd2_log_wait_commit);
89 EXPORT_SYMBOL(jbd2_log_start_commit);
90 EXPORT_SYMBOL(jbd2_journal_start_commit);
91 EXPORT_SYMBOL(jbd2_journal_force_commit_nested);
92 EXPORT_SYMBOL(jbd2_journal_wipe);
93 EXPORT_SYMBOL(jbd2_journal_blocks_per_page);
94 EXPORT_SYMBOL(jbd2_journal_invalidatepage);
95 EXPORT_SYMBOL(jbd2_journal_try_to_free_buffers);
96 EXPORT_SYMBOL(jbd2_journal_force_commit);
97 EXPORT_SYMBOL(jbd2_journal_file_inode);
98 EXPORT_SYMBOL(jbd2_journal_init_jbd_inode);
99 EXPORT_SYMBOL(jbd2_journal_release_jbd_inode);
100 EXPORT_SYMBOL(jbd2_journal_begin_ordered_truncate);
101 EXPORT_SYMBOL(jbd2_inode_cache);
103 static void __journal_abort_soft (journal_t *journal, int errno);
104 static int jbd2_journal_create_slab(size_t slab_size);
106 /* Checksumming functions */
107 int jbd2_verify_csum_type(journal_t *j, journal_superblock_t *sb)
109 if (!JBD2_HAS_INCOMPAT_FEATURE(j, JBD2_FEATURE_INCOMPAT_CSUM_V2))
112 return sb->s_checksum_type == JBD2_CRC32C_CHKSUM;
115 static __u32 jbd2_superblock_csum(journal_t *j, journal_superblock_t *sb)
117 __u32 csum, old_csum;
119 old_csum = sb->s_checksum;
121 csum = jbd2_chksum(j, ~0, (char *)sb, sizeof(journal_superblock_t));
122 sb->s_checksum = old_csum;
124 return cpu_to_be32(csum);
127 int jbd2_superblock_csum_verify(journal_t *j, journal_superblock_t *sb)
129 if (!JBD2_HAS_INCOMPAT_FEATURE(j, JBD2_FEATURE_INCOMPAT_CSUM_V2))
132 return sb->s_checksum == jbd2_superblock_csum(j, sb);
135 void jbd2_superblock_csum_set(journal_t *j, journal_superblock_t *sb)
137 if (!JBD2_HAS_INCOMPAT_FEATURE(j, JBD2_FEATURE_INCOMPAT_CSUM_V2))
140 sb->s_checksum = jbd2_superblock_csum(j, sb);
144 * Helper function used to manage commit timeouts
147 static void commit_timeout(unsigned long __data)
149 struct task_struct * p = (struct task_struct *) __data;
155 * kjournald2: The main thread function used to manage a logging device
158 * This kernel thread is responsible for two things:
160 * 1) COMMIT: Every so often we need to commit the current state of the
161 * filesystem to disk. The journal thread is responsible for writing
162 * all of the metadata buffers to disk.
164 * 2) CHECKPOINT: We cannot reuse a used section of the log file until all
165 * of the data in that part of the log has been rewritten elsewhere on
166 * the disk. Flushing these old buffers to reclaim space in the log is
167 * known as checkpointing, and this thread is responsible for that job.
170 static int kjournald2(void *arg)
172 journal_t *journal = arg;
173 transaction_t *transaction;
176 * Set up an interval timer which can be used to trigger a commit wakeup
177 * after the commit interval expires
179 setup_timer(&journal->j_commit_timer, commit_timeout,
180 (unsigned long)current);
184 /* Record that the journal thread is running */
185 journal->j_task = current;
186 wake_up(&journal->j_wait_done_commit);
189 * And now, wait forever for commit wakeup events.
191 write_lock(&journal->j_state_lock);
194 if (journal->j_flags & JBD2_UNMOUNT)
197 jbd_debug(1, "commit_sequence=%d, commit_request=%d\n",
198 journal->j_commit_sequence, journal->j_commit_request);
200 if (journal->j_commit_sequence != journal->j_commit_request) {
201 jbd_debug(1, "OK, requests differ\n");
202 write_unlock(&journal->j_state_lock);
203 del_timer_sync(&journal->j_commit_timer);
204 jbd2_journal_commit_transaction(journal);
205 write_lock(&journal->j_state_lock);
209 wake_up(&journal->j_wait_done_commit);
210 if (freezing(current)) {
212 * The simpler the better. Flushing journal isn't a
213 * good idea, because that depends on threads that may
214 * be already stopped.
216 jbd_debug(1, "Now suspending kjournald2\n");
217 write_unlock(&journal->j_state_lock);
219 write_lock(&journal->j_state_lock);
222 * We assume on resume that commits are already there,
226 int should_sleep = 1;
228 prepare_to_wait(&journal->j_wait_commit, &wait,
230 if (journal->j_commit_sequence != journal->j_commit_request)
232 transaction = journal->j_running_transaction;
233 if (transaction && time_after_eq(jiffies,
234 transaction->t_expires))
236 if (journal->j_flags & JBD2_UNMOUNT)
239 write_unlock(&journal->j_state_lock);
241 write_lock(&journal->j_state_lock);
243 finish_wait(&journal->j_wait_commit, &wait);
246 jbd_debug(1, "kjournald2 wakes\n");
249 * Were we woken up by a commit wakeup event?
251 transaction = journal->j_running_transaction;
252 if (transaction && time_after_eq(jiffies, transaction->t_expires)) {
253 journal->j_commit_request = transaction->t_tid;
254 jbd_debug(1, "woke because of timeout\n");
259 write_unlock(&journal->j_state_lock);
260 del_timer_sync(&journal->j_commit_timer);
261 journal->j_task = NULL;
262 wake_up(&journal->j_wait_done_commit);
263 jbd_debug(1, "Journal thread exiting.\n");
267 static int jbd2_journal_start_thread(journal_t *journal)
269 struct task_struct *t;
271 t = kthread_run(kjournald2, journal, "jbd2/%s",
276 wait_event(journal->j_wait_done_commit, journal->j_task != NULL);
280 static void journal_kill_thread(journal_t *journal)
282 write_lock(&journal->j_state_lock);
283 journal->j_flags |= JBD2_UNMOUNT;
285 while (journal->j_task) {
286 wake_up(&journal->j_wait_commit);
287 write_unlock(&journal->j_state_lock);
288 wait_event(journal->j_wait_done_commit, journal->j_task == NULL);
289 write_lock(&journal->j_state_lock);
291 write_unlock(&journal->j_state_lock);
295 * jbd2_journal_write_metadata_buffer: write a metadata buffer to the journal.
297 * Writes a metadata buffer to a given disk block. The actual IO is not
298 * performed but a new buffer_head is constructed which labels the data
299 * to be written with the correct destination disk block.
301 * Any magic-number escaping which needs to be done will cause a
302 * copy-out here. If the buffer happens to start with the
303 * JBD2_MAGIC_NUMBER, then we can't write it to the log directly: the
304 * magic number is only written to the log for descripter blocks. In
305 * this case, we copy the data and replace the first word with 0, and we
306 * return a result code which indicates that this buffer needs to be
307 * marked as an escaped buffer in the corresponding log descriptor
308 * block. The missing word can then be restored when the block is read
311 * If the source buffer has already been modified by a new transaction
312 * since we took the last commit snapshot, we use the frozen copy of
313 * that data for IO. If we end up using the existing buffer_head's data
314 * for the write, then we *have* to lock the buffer to prevent anyone
315 * else from using and possibly modifying it while the IO is in
318 * The function returns a pointer to the buffer_heads to be used for IO.
320 * We assume that the journal has already been locked in this function.
327 * Bit 0 set == escape performed on the data
328 * Bit 1 set == buffer copy-out performed (kfree the data after IO)
331 int jbd2_journal_write_metadata_buffer(transaction_t *transaction,
332 struct journal_head *jh_in,
333 struct journal_head **jh_out,
334 unsigned long long blocknr)
336 int need_copy_out = 0;
337 int done_copy_out = 0;
340 struct buffer_head *new_bh;
341 struct journal_head *new_jh;
342 struct page *new_page;
343 unsigned int new_offset;
344 struct buffer_head *bh_in = jh2bh(jh_in);
345 journal_t *journal = transaction->t_journal;
348 * The buffer really shouldn't be locked: only the current committing
349 * transaction is allowed to write it, so nobody else is allowed
352 * akpm: except if we're journalling data, and write() output is
353 * also part of a shared mapping, and another thread has
354 * decided to launch a writepage() against this buffer.
356 J_ASSERT_BH(bh_in, buffer_jbddirty(bh_in));
359 new_bh = alloc_buffer_head(GFP_NOFS);
362 * Failure is not an option, but __GFP_NOFAIL is going
363 * away; so we retry ourselves here.
365 congestion_wait(BLK_RW_ASYNC, HZ/50);
369 /* keep subsequent assertions sane */
370 atomic_set(&new_bh->b_count, 1);
371 new_jh = jbd2_journal_add_journal_head(new_bh); /* This sleeps */
374 * If a new transaction has already done a buffer copy-out, then
375 * we use that version of the data for the commit.
377 jbd_lock_bh_state(bh_in);
379 if (jh_in->b_frozen_data) {
381 new_page = virt_to_page(jh_in->b_frozen_data);
382 new_offset = offset_in_page(jh_in->b_frozen_data);
384 new_page = jh2bh(jh_in)->b_page;
385 new_offset = offset_in_page(jh2bh(jh_in)->b_data);
388 mapped_data = kmap_atomic(new_page);
390 * Fire data frozen trigger if data already wasn't frozen. Do this
391 * before checking for escaping, as the trigger may modify the magic
392 * offset. If a copy-out happens afterwards, it will have the correct
393 * data in the buffer.
396 jbd2_buffer_frozen_trigger(jh_in, mapped_data + new_offset,
402 if (*((__be32 *)(mapped_data + new_offset)) ==
403 cpu_to_be32(JBD2_MAGIC_NUMBER)) {
407 kunmap_atomic(mapped_data);
410 * Do we need to do a data copy?
412 if (need_copy_out && !done_copy_out) {
415 jbd_unlock_bh_state(bh_in);
416 tmp = jbd2_alloc(bh_in->b_size, GFP_NOFS);
418 jbd2_journal_put_journal_head(new_jh);
421 jbd_lock_bh_state(bh_in);
422 if (jh_in->b_frozen_data) {
423 jbd2_free(tmp, bh_in->b_size);
427 jh_in->b_frozen_data = tmp;
428 mapped_data = kmap_atomic(new_page);
429 memcpy(tmp, mapped_data + new_offset, jh2bh(jh_in)->b_size);
430 kunmap_atomic(mapped_data);
432 new_page = virt_to_page(tmp);
433 new_offset = offset_in_page(tmp);
437 * This isn't strictly necessary, as we're using frozen
438 * data for the escaping, but it keeps consistency with
439 * b_frozen_data usage.
441 jh_in->b_frozen_triggers = jh_in->b_triggers;
445 * Did we need to do an escaping? Now we've done all the
446 * copying, we can finally do so.
449 mapped_data = kmap_atomic(new_page);
450 *((unsigned int *)(mapped_data + new_offset)) = 0;
451 kunmap_atomic(mapped_data);
454 set_bh_page(new_bh, new_page, new_offset);
455 new_jh->b_transaction = NULL;
456 new_bh->b_size = jh2bh(jh_in)->b_size;
457 new_bh->b_bdev = transaction->t_journal->j_dev;
458 new_bh->b_blocknr = blocknr;
459 set_buffer_mapped(new_bh);
460 set_buffer_dirty(new_bh);
465 * The to-be-written buffer needs to get moved to the io queue,
466 * and the original buffer whose contents we are shadowing or
467 * copying is moved to the transaction's shadow queue.
469 JBUFFER_TRACE(jh_in, "file as BJ_Shadow");
470 spin_lock(&journal->j_list_lock);
471 __jbd2_journal_file_buffer(jh_in, transaction, BJ_Shadow);
472 spin_unlock(&journal->j_list_lock);
473 jbd_unlock_bh_state(bh_in);
475 JBUFFER_TRACE(new_jh, "file as BJ_IO");
476 jbd2_journal_file_buffer(new_jh, transaction, BJ_IO);
478 return do_escape | (done_copy_out << 1);
482 * Allocation code for the journal file. Manage the space left in the
483 * journal, so that we can begin checkpointing when appropriate.
487 * __jbd2_log_space_left: Return the number of free blocks left in the journal.
489 * Called with the journal already locked.
491 * Called under j_state_lock
494 int __jbd2_log_space_left(journal_t *journal)
496 int left = journal->j_free;
498 /* assert_spin_locked(&journal->j_state_lock); */
501 * Be pessimistic here about the number of those free blocks which
502 * might be required for log descriptor control blocks.
505 #define MIN_LOG_RESERVED_BLOCKS 32 /* Allow for rounding errors */
507 left -= MIN_LOG_RESERVED_BLOCKS;
516 * Called with j_state_lock locked for writing.
517 * Returns true if a transaction commit was started.
519 int __jbd2_log_start_commit(journal_t *journal, tid_t target)
521 /* Return if the txn has already requested to be committed */
522 if (journal->j_commit_request == target)
526 * The only transaction we can possibly wait upon is the
527 * currently running transaction (if it exists). Otherwise,
528 * the target tid must be an old one.
530 if (journal->j_running_transaction &&
531 journal->j_running_transaction->t_tid == target) {
533 * We want a new commit: OK, mark the request and wakeup the
534 * commit thread. We do _not_ do the commit ourselves.
537 journal->j_commit_request = target;
538 jbd_debug(1, "JBD2: requesting commit %d/%d\n",
539 journal->j_commit_request,
540 journal->j_commit_sequence);
541 journal->j_running_transaction->t_requested = jiffies;
542 wake_up(&journal->j_wait_commit);
544 } else if (!tid_geq(journal->j_commit_request, target))
545 /* This should never happen, but if it does, preserve
546 the evidence before kjournald goes into a loop and
547 increments j_commit_sequence beyond all recognition. */
548 WARN_ONCE(1, "JBD2: bad log_start_commit: %u %u %u %u\n",
549 journal->j_commit_request,
550 journal->j_commit_sequence,
551 target, journal->j_running_transaction ?
552 journal->j_running_transaction->t_tid : 0);
556 int jbd2_log_start_commit(journal_t *journal, tid_t tid)
560 write_lock(&journal->j_state_lock);
561 ret = __jbd2_log_start_commit(journal, tid);
562 write_unlock(&journal->j_state_lock);
567 * Force and wait upon a commit if the calling process is not within
568 * transaction. This is used for forcing out undo-protected data which contains
569 * bitmaps, when the fs is running out of space.
571 * We can only force the running transaction if we don't have an active handle;
572 * otherwise, we will deadlock.
574 * Returns true if a transaction was started.
576 int jbd2_journal_force_commit_nested(journal_t *journal)
578 transaction_t *transaction = NULL;
580 int need_to_start = 0;
582 read_lock(&journal->j_state_lock);
583 if (journal->j_running_transaction && !current->journal_info) {
584 transaction = journal->j_running_transaction;
585 if (!tid_geq(journal->j_commit_request, transaction->t_tid))
587 } else if (journal->j_committing_transaction)
588 transaction = journal->j_committing_transaction;
591 read_unlock(&journal->j_state_lock);
592 return 0; /* Nothing to retry */
595 tid = transaction->t_tid;
596 read_unlock(&journal->j_state_lock);
598 jbd2_log_start_commit(journal, tid);
599 jbd2_log_wait_commit(journal, tid);
604 * Start a commit of the current running transaction (if any). Returns true
605 * if a transaction is going to be committed (or is currently already
606 * committing), and fills its tid in at *ptid
608 int jbd2_journal_start_commit(journal_t *journal, tid_t *ptid)
612 write_lock(&journal->j_state_lock);
613 if (journal->j_running_transaction) {
614 tid_t tid = journal->j_running_transaction->t_tid;
616 __jbd2_log_start_commit(journal, tid);
617 /* There's a running transaction and we've just made sure
618 * it's commit has been scheduled. */
622 } else if (journal->j_committing_transaction) {
624 * If commit has been started, then we have to wait for
625 * completion of that transaction.
628 *ptid = journal->j_committing_transaction->t_tid;
631 write_unlock(&journal->j_state_lock);
636 * Return 1 if a given transaction has not yet sent barrier request
637 * connected with a transaction commit. If 0 is returned, transaction
638 * may or may not have sent the barrier. Used to avoid sending barrier
639 * twice in common cases.
641 int jbd2_trans_will_send_data_barrier(journal_t *journal, tid_t tid)
644 transaction_t *commit_trans;
646 if (!(journal->j_flags & JBD2_BARRIER))
648 read_lock(&journal->j_state_lock);
649 /* Transaction already committed? */
650 if (tid_geq(journal->j_commit_sequence, tid))
652 commit_trans = journal->j_committing_transaction;
653 if (!commit_trans || commit_trans->t_tid != tid) {
658 * Transaction is being committed and we already proceeded to
659 * submitting a flush to fs partition?
661 if (journal->j_fs_dev != journal->j_dev) {
662 if (!commit_trans->t_need_data_flush ||
663 commit_trans->t_state >= T_COMMIT_DFLUSH)
666 if (commit_trans->t_state >= T_COMMIT_JFLUSH)
671 read_unlock(&journal->j_state_lock);
674 EXPORT_SYMBOL(jbd2_trans_will_send_data_barrier);
677 * Wait for a specified commit to complete.
678 * The caller may not hold the journal lock.
680 int jbd2_log_wait_commit(journal_t *journal, tid_t tid)
684 read_lock(&journal->j_state_lock);
685 #ifdef CONFIG_JBD2_DEBUG
686 if (!tid_geq(journal->j_commit_request, tid)) {
688 "%s: error: j_commit_request=%d, tid=%d\n",
689 __func__, journal->j_commit_request, tid);
692 while (tid_gt(tid, journal->j_commit_sequence)) {
693 jbd_debug(1, "JBD2: want %d, j_commit_sequence=%d\n",
694 tid, journal->j_commit_sequence);
695 wake_up(&journal->j_wait_commit);
696 read_unlock(&journal->j_state_lock);
697 wait_event(journal->j_wait_done_commit,
698 !tid_gt(tid, journal->j_commit_sequence));
699 read_lock(&journal->j_state_lock);
701 read_unlock(&journal->j_state_lock);
703 if (unlikely(is_journal_aborted(journal))) {
704 printk(KERN_EMERG "journal commit I/O error\n");
711 * When this function returns the transaction corresponding to tid
712 * will be completed. If the transaction has currently running, start
713 * committing that transaction before waiting for it to complete. If
714 * the transaction id is stale, it is by definition already completed,
715 * so just return SUCCESS.
717 int jbd2_complete_transaction(journal_t *journal, tid_t tid)
719 int need_to_wait = 1;
721 read_lock(&journal->j_state_lock);
722 if (journal->j_running_transaction &&
723 journal->j_running_transaction->t_tid == tid) {
724 if (journal->j_commit_request != tid) {
725 /* transaction not yet started, so request it */
726 read_unlock(&journal->j_state_lock);
727 jbd2_log_start_commit(journal, tid);
730 } else if (!(journal->j_committing_transaction &&
731 journal->j_committing_transaction->t_tid == tid))
733 read_unlock(&journal->j_state_lock);
737 return jbd2_log_wait_commit(journal, tid);
739 EXPORT_SYMBOL(jbd2_complete_transaction);
742 * Log buffer allocation routines:
745 int jbd2_journal_next_log_block(journal_t *journal, unsigned long long *retp)
747 unsigned long blocknr;
749 write_lock(&journal->j_state_lock);
750 J_ASSERT(journal->j_free > 1);
752 blocknr = journal->j_head;
755 if (journal->j_head == journal->j_last)
756 journal->j_head = journal->j_first;
757 write_unlock(&journal->j_state_lock);
758 return jbd2_journal_bmap(journal, blocknr, retp);
762 * Conversion of logical to physical block numbers for the journal
764 * On external journals the journal blocks are identity-mapped, so
765 * this is a no-op. If needed, we can use j_blk_offset - everything is
768 int jbd2_journal_bmap(journal_t *journal, unsigned long blocknr,
769 unsigned long long *retp)
772 unsigned long long ret;
774 if (journal->j_inode) {
775 ret = bmap(journal->j_inode, blocknr);
779 printk(KERN_ALERT "%s: journal block not found "
780 "at offset %lu on %s\n",
781 __func__, blocknr, journal->j_devname);
783 __journal_abort_soft(journal, err);
786 *retp = blocknr; /* +journal->j_blk_offset */
792 * We play buffer_head aliasing tricks to write data/metadata blocks to
793 * the journal without copying their contents, but for journal
794 * descriptor blocks we do need to generate bona fide buffers.
796 * After the caller of jbd2_journal_get_descriptor_buffer() has finished modifying
797 * the buffer's contents they really should run flush_dcache_page(bh->b_page).
798 * But we don't bother doing that, so there will be coherency problems with
799 * mmaps of blockdevs which hold live JBD-controlled filesystems.
801 struct journal_head *jbd2_journal_get_descriptor_buffer(journal_t *journal)
803 struct buffer_head *bh;
804 unsigned long long blocknr;
807 err = jbd2_journal_next_log_block(journal, &blocknr);
812 bh = __getblk(journal->j_dev, blocknr, journal->j_blocksize);
816 memset(bh->b_data, 0, journal->j_blocksize);
817 set_buffer_uptodate(bh);
819 BUFFER_TRACE(bh, "return this buffer");
820 return jbd2_journal_add_journal_head(bh);
824 * Return tid of the oldest transaction in the journal and block in the journal
825 * where the transaction starts.
827 * If the journal is now empty, return which will be the next transaction ID
828 * we will write and where will that transaction start.
830 * The return value is 0 if journal tail cannot be pushed any further, 1 if
833 int jbd2_journal_get_log_tail(journal_t *journal, tid_t *tid,
834 unsigned long *block)
836 transaction_t *transaction;
839 read_lock(&journal->j_state_lock);
840 spin_lock(&journal->j_list_lock);
841 transaction = journal->j_checkpoint_transactions;
843 *tid = transaction->t_tid;
844 *block = transaction->t_log_start;
845 } else if ((transaction = journal->j_committing_transaction) != NULL) {
846 *tid = transaction->t_tid;
847 *block = transaction->t_log_start;
848 } else if ((transaction = journal->j_running_transaction) != NULL) {
849 *tid = transaction->t_tid;
850 *block = journal->j_head;
852 *tid = journal->j_transaction_sequence;
853 *block = journal->j_head;
855 ret = tid_gt(*tid, journal->j_tail_sequence);
856 spin_unlock(&journal->j_list_lock);
857 read_unlock(&journal->j_state_lock);
863 * Update information in journal structure and in on disk journal superblock
864 * about log tail. This function does not check whether information passed in
865 * really pushes log tail further. It's responsibility of the caller to make
866 * sure provided log tail information is valid (e.g. by holding
867 * j_checkpoint_mutex all the time between computing log tail and calling this
868 * function as is the case with jbd2_cleanup_journal_tail()).
870 * Requires j_checkpoint_mutex
872 void __jbd2_update_log_tail(journal_t *journal, tid_t tid, unsigned long block)
876 BUG_ON(!mutex_is_locked(&journal->j_checkpoint_mutex));
879 * We cannot afford for write to remain in drive's caches since as
880 * soon as we update j_tail, next transaction can start reusing journal
881 * space and if we lose sb update during power failure we'd replay
882 * old transaction with possibly newly overwritten data.
884 jbd2_journal_update_sb_log_tail(journal, tid, block, WRITE_FUA);
885 write_lock(&journal->j_state_lock);
886 freed = block - journal->j_tail;
887 if (block < journal->j_tail)
888 freed += journal->j_last - journal->j_first;
890 trace_jbd2_update_log_tail(journal, tid, block, freed);
892 "Cleaning journal tail from %d to %d (offset %lu), "
894 journal->j_tail_sequence, tid, block, freed);
896 journal->j_free += freed;
897 journal->j_tail_sequence = tid;
898 journal->j_tail = block;
899 write_unlock(&journal->j_state_lock);
903 * This is a variaon of __jbd2_update_log_tail which checks for validity of
904 * provided log tail and locks j_checkpoint_mutex. So it is safe against races
905 * with other threads updating log tail.
907 void jbd2_update_log_tail(journal_t *journal, tid_t tid, unsigned long block)
909 mutex_lock(&journal->j_checkpoint_mutex);
910 if (tid_gt(tid, journal->j_tail_sequence))
911 __jbd2_update_log_tail(journal, tid, block);
912 mutex_unlock(&journal->j_checkpoint_mutex);
915 struct jbd2_stats_proc_session {
917 struct transaction_stats_s *stats;
922 static void *jbd2_seq_info_start(struct seq_file *seq, loff_t *pos)
924 return *pos ? NULL : SEQ_START_TOKEN;
927 static void *jbd2_seq_info_next(struct seq_file *seq, void *v, loff_t *pos)
932 static int jbd2_seq_info_show(struct seq_file *seq, void *v)
934 struct jbd2_stats_proc_session *s = seq->private;
936 if (v != SEQ_START_TOKEN)
938 seq_printf(seq, "%lu transactions (%lu requested), "
939 "each up to %u blocks\n",
940 s->stats->ts_tid, s->stats->ts_requested,
941 s->journal->j_max_transaction_buffers);
942 if (s->stats->ts_tid == 0)
944 seq_printf(seq, "average: \n %ums waiting for transaction\n",
945 jiffies_to_msecs(s->stats->run.rs_wait / s->stats->ts_tid));
946 seq_printf(seq, " %ums request delay\n",
947 (s->stats->ts_requested == 0) ? 0 :
948 jiffies_to_msecs(s->stats->run.rs_request_delay /
949 s->stats->ts_requested));
950 seq_printf(seq, " %ums running transaction\n",
951 jiffies_to_msecs(s->stats->run.rs_running / s->stats->ts_tid));
952 seq_printf(seq, " %ums transaction was being locked\n",
953 jiffies_to_msecs(s->stats->run.rs_locked / s->stats->ts_tid));
954 seq_printf(seq, " %ums flushing data (in ordered mode)\n",
955 jiffies_to_msecs(s->stats->run.rs_flushing / s->stats->ts_tid));
956 seq_printf(seq, " %ums logging transaction\n",
957 jiffies_to_msecs(s->stats->run.rs_logging / s->stats->ts_tid));
958 seq_printf(seq, " %lluus average transaction commit time\n",
959 div_u64(s->journal->j_average_commit_time, 1000));
960 seq_printf(seq, " %lu handles per transaction\n",
961 s->stats->run.rs_handle_count / s->stats->ts_tid);
962 seq_printf(seq, " %lu blocks per transaction\n",
963 s->stats->run.rs_blocks / s->stats->ts_tid);
964 seq_printf(seq, " %lu logged blocks per transaction\n",
965 s->stats->run.rs_blocks_logged / s->stats->ts_tid);
969 static void jbd2_seq_info_stop(struct seq_file *seq, void *v)
973 static const struct seq_operations jbd2_seq_info_ops = {
974 .start = jbd2_seq_info_start,
975 .next = jbd2_seq_info_next,
976 .stop = jbd2_seq_info_stop,
977 .show = jbd2_seq_info_show,
980 static int jbd2_seq_info_open(struct inode *inode, struct file *file)
982 journal_t *journal = PDE_DATA(inode);
983 struct jbd2_stats_proc_session *s;
986 s = kmalloc(sizeof(*s), GFP_KERNEL);
989 size = sizeof(struct transaction_stats_s);
990 s->stats = kmalloc(size, GFP_KERNEL);
991 if (s->stats == NULL) {
995 spin_lock(&journal->j_history_lock);
996 memcpy(s->stats, &journal->j_stats, size);
997 s->journal = journal;
998 spin_unlock(&journal->j_history_lock);
1000 rc = seq_open(file, &jbd2_seq_info_ops);
1002 struct seq_file *m = file->private_data;
1012 static int jbd2_seq_info_release(struct inode *inode, struct file *file)
1014 struct seq_file *seq = file->private_data;
1015 struct jbd2_stats_proc_session *s = seq->private;
1018 return seq_release(inode, file);
1021 static const struct file_operations jbd2_seq_info_fops = {
1022 .owner = THIS_MODULE,
1023 .open = jbd2_seq_info_open,
1025 .llseek = seq_lseek,
1026 .release = jbd2_seq_info_release,
1029 static struct proc_dir_entry *proc_jbd2_stats;
1031 static void jbd2_stats_proc_init(journal_t *journal)
1033 journal->j_proc_entry = proc_mkdir(journal->j_devname, proc_jbd2_stats);
1034 if (journal->j_proc_entry) {
1035 proc_create_data("info", S_IRUGO, journal->j_proc_entry,
1036 &jbd2_seq_info_fops, journal);
1040 static void jbd2_stats_proc_exit(journal_t *journal)
1042 remove_proc_entry("info", journal->j_proc_entry);
1043 remove_proc_entry(journal->j_devname, proc_jbd2_stats);
1047 * Management for journal control blocks: functions to create and
1048 * destroy journal_t structures, and to initialise and read existing
1049 * journal blocks from disk. */
1051 /* First: create and setup a journal_t object in memory. We initialise
1052 * very few fields yet: that has to wait until we have created the
1053 * journal structures from from scratch, or loaded them from disk. */
1055 static journal_t * journal_init_common (void)
1060 journal = kzalloc(sizeof(*journal), GFP_KERNEL);
1064 init_waitqueue_head(&journal->j_wait_transaction_locked);
1065 init_waitqueue_head(&journal->j_wait_logspace);
1066 init_waitqueue_head(&journal->j_wait_done_commit);
1067 init_waitqueue_head(&journal->j_wait_checkpoint);
1068 init_waitqueue_head(&journal->j_wait_commit);
1069 init_waitqueue_head(&journal->j_wait_updates);
1070 mutex_init(&journal->j_barrier);
1071 mutex_init(&journal->j_checkpoint_mutex);
1072 spin_lock_init(&journal->j_revoke_lock);
1073 spin_lock_init(&journal->j_list_lock);
1074 rwlock_init(&journal->j_state_lock);
1076 journal->j_commit_interval = (HZ * JBD2_DEFAULT_MAX_COMMIT_AGE);
1077 journal->j_min_batch_time = 0;
1078 journal->j_max_batch_time = 15000; /* 15ms */
1080 /* The journal is marked for error until we succeed with recovery! */
1081 journal->j_flags = JBD2_ABORT;
1083 /* Set up a default-sized revoke table for the new mount. */
1084 err = jbd2_journal_init_revoke(journal, JOURNAL_REVOKE_DEFAULT_HASH);
1090 spin_lock_init(&journal->j_history_lock);
1095 /* jbd2_journal_init_dev and jbd2_journal_init_inode:
1097 * Create a journal structure assigned some fixed set of disk blocks to
1098 * the journal. We don't actually touch those disk blocks yet, but we
1099 * need to set up all of the mapping information to tell the journaling
1100 * system where the journal blocks are.
1105 * journal_t * jbd2_journal_init_dev() - creates and initialises a journal structure
1106 * @bdev: Block device on which to create the journal
1107 * @fs_dev: Device which hold journalled filesystem for this journal.
1108 * @start: Block nr Start of journal.
1109 * @len: Length of the journal in blocks.
1110 * @blocksize: blocksize of journalling device
1112 * Returns: a newly created journal_t *
1114 * jbd2_journal_init_dev creates a journal which maps a fixed contiguous
1115 * range of blocks on an arbitrary block device.
1118 journal_t * jbd2_journal_init_dev(struct block_device *bdev,
1119 struct block_device *fs_dev,
1120 unsigned long long start, int len, int blocksize)
1122 journal_t *journal = journal_init_common();
1123 struct buffer_head *bh;
1130 /* journal descriptor can store up to n blocks -bzzz */
1131 journal->j_blocksize = blocksize;
1132 journal->j_dev = bdev;
1133 journal->j_fs_dev = fs_dev;
1134 journal->j_blk_offset = start;
1135 journal->j_maxlen = len;
1136 bdevname(journal->j_dev, journal->j_devname);
1137 p = journal->j_devname;
1138 while ((p = strchr(p, '/')))
1140 jbd2_stats_proc_init(journal);
1141 n = journal->j_blocksize / sizeof(journal_block_tag_t);
1142 journal->j_wbufsize = n;
1143 journal->j_wbuf = kmalloc(n * sizeof(struct buffer_head*), GFP_KERNEL);
1144 if (!journal->j_wbuf) {
1145 printk(KERN_ERR "%s: Can't allocate bhs for commit thread\n",
1150 bh = __getblk(journal->j_dev, start, journal->j_blocksize);
1153 "%s: Cannot get buffer for journal superblock\n",
1157 journal->j_sb_buffer = bh;
1158 journal->j_superblock = (journal_superblock_t *)bh->b_data;
1162 kfree(journal->j_wbuf);
1163 jbd2_stats_proc_exit(journal);
1169 * journal_t * jbd2_journal_init_inode () - creates a journal which maps to a inode.
1170 * @inode: An inode to create the journal in
1172 * jbd2_journal_init_inode creates a journal which maps an on-disk inode as
1173 * the journal. The inode must exist already, must support bmap() and
1174 * must have all data blocks preallocated.
1176 journal_t * jbd2_journal_init_inode (struct inode *inode)
1178 struct buffer_head *bh;
1179 journal_t *journal = journal_init_common();
1183 unsigned long long blocknr;
1188 journal->j_dev = journal->j_fs_dev = inode->i_sb->s_bdev;
1189 journal->j_inode = inode;
1190 bdevname(journal->j_dev, journal->j_devname);
1191 p = journal->j_devname;
1192 while ((p = strchr(p, '/')))
1194 p = journal->j_devname + strlen(journal->j_devname);
1195 sprintf(p, "-%lu", journal->j_inode->i_ino);
1197 "journal %p: inode %s/%ld, size %Ld, bits %d, blksize %ld\n",
1198 journal, inode->i_sb->s_id, inode->i_ino,
1199 (long long) inode->i_size,
1200 inode->i_sb->s_blocksize_bits, inode->i_sb->s_blocksize);
1202 journal->j_maxlen = inode->i_size >> inode->i_sb->s_blocksize_bits;
1203 journal->j_blocksize = inode->i_sb->s_blocksize;
1204 jbd2_stats_proc_init(journal);
1206 /* journal descriptor can store up to n blocks -bzzz */
1207 n = journal->j_blocksize / sizeof(journal_block_tag_t);
1208 journal->j_wbufsize = n;
1209 journal->j_wbuf = kmalloc(n * sizeof(struct buffer_head*), GFP_KERNEL);
1210 if (!journal->j_wbuf) {
1211 printk(KERN_ERR "%s: Can't allocate bhs for commit thread\n",
1216 err = jbd2_journal_bmap(journal, 0, &blocknr);
1217 /* If that failed, give up */
1219 printk(KERN_ERR "%s: Cannot locate journal superblock\n",
1224 bh = __getblk(journal->j_dev, blocknr, journal->j_blocksize);
1227 "%s: Cannot get buffer for journal superblock\n",
1231 journal->j_sb_buffer = bh;
1232 journal->j_superblock = (journal_superblock_t *)bh->b_data;
1236 kfree(journal->j_wbuf);
1237 jbd2_stats_proc_exit(journal);
1243 * If the journal init or create aborts, we need to mark the journal
1244 * superblock as being NULL to prevent the journal destroy from writing
1245 * back a bogus superblock.
1247 static void journal_fail_superblock (journal_t *journal)
1249 struct buffer_head *bh = journal->j_sb_buffer;
1251 journal->j_sb_buffer = NULL;
1255 * Given a journal_t structure, initialise the various fields for
1256 * startup of a new journaling session. We use this both when creating
1257 * a journal, and after recovering an old journal to reset it for
1261 static int journal_reset(journal_t *journal)
1263 journal_superblock_t *sb = journal->j_superblock;
1264 unsigned long long first, last;
1266 first = be32_to_cpu(sb->s_first);
1267 last = be32_to_cpu(sb->s_maxlen);
1268 if (first + JBD2_MIN_JOURNAL_BLOCKS > last + 1) {
1269 printk(KERN_ERR "JBD2: Journal too short (blocks %llu-%llu).\n",
1271 journal_fail_superblock(journal);
1275 journal->j_first = first;
1276 journal->j_last = last;
1278 journal->j_head = first;
1279 journal->j_tail = first;
1280 journal->j_free = last - first;
1282 journal->j_tail_sequence = journal->j_transaction_sequence;
1283 journal->j_commit_sequence = journal->j_transaction_sequence - 1;
1284 journal->j_commit_request = journal->j_commit_sequence;
1286 journal->j_max_transaction_buffers = journal->j_maxlen / 4;
1289 * As a special case, if the on-disk copy is already marked as needing
1290 * no recovery (s_start == 0), then we can safely defer the superblock
1291 * update until the next commit by setting JBD2_FLUSHED. This avoids
1292 * attempting a write to a potential-readonly device.
1294 if (sb->s_start == 0) {
1295 jbd_debug(1, "JBD2: Skipping superblock update on recovered sb "
1296 "(start %ld, seq %d, errno %d)\n",
1297 journal->j_tail, journal->j_tail_sequence,
1299 journal->j_flags |= JBD2_FLUSHED;
1301 /* Lock here to make assertions happy... */
1302 mutex_lock(&journal->j_checkpoint_mutex);
1304 * Update log tail information. We use WRITE_FUA since new
1305 * transaction will start reusing journal space and so we
1306 * must make sure information about current log tail is on
1309 jbd2_journal_update_sb_log_tail(journal,
1310 journal->j_tail_sequence,
1313 mutex_unlock(&journal->j_checkpoint_mutex);
1315 return jbd2_journal_start_thread(journal);
1318 static void jbd2_write_superblock(journal_t *journal, int write_op)
1320 struct buffer_head *bh = journal->j_sb_buffer;
1321 journal_superblock_t *sb = journal->j_superblock;
1324 trace_jbd2_write_superblock(journal, write_op);
1325 if (!(journal->j_flags & JBD2_BARRIER))
1326 write_op &= ~(REQ_FUA | REQ_FLUSH);
1328 if (buffer_write_io_error(bh)) {
1330 * Oh, dear. A previous attempt to write the journal
1331 * superblock failed. This could happen because the
1332 * USB device was yanked out. Or it could happen to
1333 * be a transient write error and maybe the block will
1334 * be remapped. Nothing we can do but to retry the
1335 * write and hope for the best.
1337 printk(KERN_ERR "JBD2: previous I/O error detected "
1338 "for journal superblock update for %s.\n",
1339 journal->j_devname);
1340 clear_buffer_write_io_error(bh);
1341 set_buffer_uptodate(bh);
1343 jbd2_superblock_csum_set(journal, sb);
1345 bh->b_end_io = end_buffer_write_sync;
1346 ret = submit_bh(write_op, bh);
1348 if (buffer_write_io_error(bh)) {
1349 clear_buffer_write_io_error(bh);
1350 set_buffer_uptodate(bh);
1354 printk(KERN_ERR "JBD2: Error %d detected when updating "
1355 "journal superblock for %s.\n", ret,
1356 journal->j_devname);
1361 * jbd2_journal_update_sb_log_tail() - Update log tail in journal sb on disk.
1362 * @journal: The journal to update.
1363 * @tail_tid: TID of the new transaction at the tail of the log
1364 * @tail_block: The first block of the transaction at the tail of the log
1365 * @write_op: With which operation should we write the journal sb
1367 * Update a journal's superblock information about log tail and write it to
1368 * disk, waiting for the IO to complete.
1370 void jbd2_journal_update_sb_log_tail(journal_t *journal, tid_t tail_tid,
1371 unsigned long tail_block, int write_op)
1373 journal_superblock_t *sb = journal->j_superblock;
1375 BUG_ON(!mutex_is_locked(&journal->j_checkpoint_mutex));
1376 jbd_debug(1, "JBD2: updating superblock (start %lu, seq %u)\n",
1377 tail_block, tail_tid);
1379 sb->s_sequence = cpu_to_be32(tail_tid);
1380 sb->s_start = cpu_to_be32(tail_block);
1382 jbd2_write_superblock(journal, write_op);
1384 /* Log is no longer empty */
1385 write_lock(&journal->j_state_lock);
1386 WARN_ON(!sb->s_sequence);
1387 journal->j_flags &= ~JBD2_FLUSHED;
1388 write_unlock(&journal->j_state_lock);
1392 * jbd2_mark_journal_empty() - Mark on disk journal as empty.
1393 * @journal: The journal to update.
1395 * Update a journal's dynamic superblock fields to show that journal is empty.
1396 * Write updated superblock to disk waiting for IO to complete.
1398 static void jbd2_mark_journal_empty(journal_t *journal)
1400 journal_superblock_t *sb = journal->j_superblock;
1402 BUG_ON(!mutex_is_locked(&journal->j_checkpoint_mutex));
1403 read_lock(&journal->j_state_lock);
1404 /* Is it already empty? */
1405 if (sb->s_start == 0) {
1406 read_unlock(&journal->j_state_lock);
1409 jbd_debug(1, "JBD2: Marking journal as empty (seq %d)\n",
1410 journal->j_tail_sequence);
1412 sb->s_sequence = cpu_to_be32(journal->j_tail_sequence);
1413 sb->s_start = cpu_to_be32(0);
1414 read_unlock(&journal->j_state_lock);
1416 jbd2_write_superblock(journal, WRITE_FUA);
1418 /* Log is no longer empty */
1419 write_lock(&journal->j_state_lock);
1420 journal->j_flags |= JBD2_FLUSHED;
1421 write_unlock(&journal->j_state_lock);
1426 * jbd2_journal_update_sb_errno() - Update error in the journal.
1427 * @journal: The journal to update.
1429 * Update a journal's errno. Write updated superblock to disk waiting for IO
1432 void jbd2_journal_update_sb_errno(journal_t *journal)
1434 journal_superblock_t *sb = journal->j_superblock;
1436 read_lock(&journal->j_state_lock);
1437 jbd_debug(1, "JBD2: updating superblock error (errno %d)\n",
1439 sb->s_errno = cpu_to_be32(journal->j_errno);
1440 read_unlock(&journal->j_state_lock);
1442 jbd2_write_superblock(journal, WRITE_SYNC);
1444 EXPORT_SYMBOL(jbd2_journal_update_sb_errno);
1447 * Read the superblock for a given journal, performing initial
1448 * validation of the format.
1450 static int journal_get_superblock(journal_t *journal)
1452 struct buffer_head *bh;
1453 journal_superblock_t *sb;
1456 bh = journal->j_sb_buffer;
1458 J_ASSERT(bh != NULL);
1459 if (!buffer_uptodate(bh)) {
1460 ll_rw_block(READ, 1, &bh);
1462 if (!buffer_uptodate(bh)) {
1464 "JBD2: IO error reading journal superblock\n");
1469 if (buffer_verified(bh))
1472 sb = journal->j_superblock;
1476 if (sb->s_header.h_magic != cpu_to_be32(JBD2_MAGIC_NUMBER) ||
1477 sb->s_blocksize != cpu_to_be32(journal->j_blocksize)) {
1478 printk(KERN_WARNING "JBD2: no valid journal superblock found\n");
1482 switch(be32_to_cpu(sb->s_header.h_blocktype)) {
1483 case JBD2_SUPERBLOCK_V1:
1484 journal->j_format_version = 1;
1486 case JBD2_SUPERBLOCK_V2:
1487 journal->j_format_version = 2;
1490 printk(KERN_WARNING "JBD2: unrecognised superblock format ID\n");
1494 if (be32_to_cpu(sb->s_maxlen) < journal->j_maxlen)
1495 journal->j_maxlen = be32_to_cpu(sb->s_maxlen);
1496 else if (be32_to_cpu(sb->s_maxlen) > journal->j_maxlen) {
1497 printk(KERN_WARNING "JBD2: journal file too short\n");
1501 if (be32_to_cpu(sb->s_first) == 0 ||
1502 be32_to_cpu(sb->s_first) >= journal->j_maxlen) {
1504 "JBD2: Invalid start block of journal: %u\n",
1505 be32_to_cpu(sb->s_first));
1509 if (JBD2_HAS_COMPAT_FEATURE(journal, JBD2_FEATURE_COMPAT_CHECKSUM) &&
1510 JBD2_HAS_INCOMPAT_FEATURE(journal, JBD2_FEATURE_INCOMPAT_CSUM_V2)) {
1511 /* Can't have checksum v1 and v2 on at the same time! */
1512 printk(KERN_ERR "JBD: Can't enable checksumming v1 and v2 "
1513 "at the same time!\n");
1517 if (!jbd2_verify_csum_type(journal, sb)) {
1518 printk(KERN_ERR "JBD: Unknown checksum type\n");
1522 /* Load the checksum driver */
1523 if (JBD2_HAS_INCOMPAT_FEATURE(journal, JBD2_FEATURE_INCOMPAT_CSUM_V2)) {
1524 journal->j_chksum_driver = crypto_alloc_shash("crc32c", 0, 0);
1525 if (IS_ERR(journal->j_chksum_driver)) {
1526 printk(KERN_ERR "JBD: Cannot load crc32c driver.\n");
1527 err = PTR_ERR(journal->j_chksum_driver);
1528 journal->j_chksum_driver = NULL;
1533 /* Check superblock checksum */
1534 if (!jbd2_superblock_csum_verify(journal, sb)) {
1535 printk(KERN_ERR "JBD: journal checksum error\n");
1539 /* Precompute checksum seed for all metadata */
1540 if (JBD2_HAS_INCOMPAT_FEATURE(journal, JBD2_FEATURE_INCOMPAT_CSUM_V2))
1541 journal->j_csum_seed = jbd2_chksum(journal, ~0, sb->s_uuid,
1542 sizeof(sb->s_uuid));
1544 set_buffer_verified(bh);
1549 journal_fail_superblock(journal);
1554 * Load the on-disk journal superblock and read the key fields into the
1558 static int load_superblock(journal_t *journal)
1561 journal_superblock_t *sb;
1563 err = journal_get_superblock(journal);
1567 sb = journal->j_superblock;
1569 journal->j_tail_sequence = be32_to_cpu(sb->s_sequence);
1570 journal->j_tail = be32_to_cpu(sb->s_start);
1571 journal->j_first = be32_to_cpu(sb->s_first);
1572 journal->j_last = be32_to_cpu(sb->s_maxlen);
1573 journal->j_errno = be32_to_cpu(sb->s_errno);
1580 * int jbd2_journal_load() - Read journal from disk.
1581 * @journal: Journal to act on.
1583 * Given a journal_t structure which tells us which disk blocks contain
1584 * a journal, read the journal from disk to initialise the in-memory
1587 int jbd2_journal_load(journal_t *journal)
1590 journal_superblock_t *sb;
1592 err = load_superblock(journal);
1596 sb = journal->j_superblock;
1597 /* If this is a V2 superblock, then we have to check the
1598 * features flags on it. */
1600 if (journal->j_format_version >= 2) {
1601 if ((sb->s_feature_ro_compat &
1602 ~cpu_to_be32(JBD2_KNOWN_ROCOMPAT_FEATURES)) ||
1603 (sb->s_feature_incompat &
1604 ~cpu_to_be32(JBD2_KNOWN_INCOMPAT_FEATURES))) {
1606 "JBD2: Unrecognised features on journal\n");
1612 * Create a slab for this blocksize
1614 err = jbd2_journal_create_slab(be32_to_cpu(sb->s_blocksize));
1618 /* Let the recovery code check whether it needs to recover any
1619 * data from the journal. */
1620 if (jbd2_journal_recover(journal))
1621 goto recovery_error;
1623 if (journal->j_failed_commit) {
1624 printk(KERN_ERR "JBD2: journal transaction %u on %s "
1625 "is corrupt.\n", journal->j_failed_commit,
1626 journal->j_devname);
1630 /* OK, we've finished with the dynamic journal bits:
1631 * reinitialise the dynamic contents of the superblock in memory
1632 * and reset them on disk. */
1633 if (journal_reset(journal))
1634 goto recovery_error;
1636 journal->j_flags &= ~JBD2_ABORT;
1637 journal->j_flags |= JBD2_LOADED;
1641 printk(KERN_WARNING "JBD2: recovery failed\n");
1646 * void jbd2_journal_destroy() - Release a journal_t structure.
1647 * @journal: Journal to act on.
1649 * Release a journal_t structure once it is no longer in use by the
1651 * Return <0 if we couldn't clean up the journal.
1653 int jbd2_journal_destroy(journal_t *journal)
1657 /* Wait for the commit thread to wake up and die. */
1658 journal_kill_thread(journal);
1660 /* Force a final log commit */
1661 if (journal->j_running_transaction)
1662 jbd2_journal_commit_transaction(journal);
1664 /* Force any old transactions to disk */
1666 /* Totally anal locking here... */
1667 spin_lock(&journal->j_list_lock);
1668 while (journal->j_checkpoint_transactions != NULL) {
1669 spin_unlock(&journal->j_list_lock);
1670 mutex_lock(&journal->j_checkpoint_mutex);
1671 jbd2_log_do_checkpoint(journal);
1672 mutex_unlock(&journal->j_checkpoint_mutex);
1673 spin_lock(&journal->j_list_lock);
1676 J_ASSERT(journal->j_running_transaction == NULL);
1677 J_ASSERT(journal->j_committing_transaction == NULL);
1678 J_ASSERT(journal->j_checkpoint_transactions == NULL);
1679 spin_unlock(&journal->j_list_lock);
1681 if (journal->j_sb_buffer) {
1682 if (!is_journal_aborted(journal)) {
1683 mutex_lock(&journal->j_checkpoint_mutex);
1684 jbd2_mark_journal_empty(journal);
1685 mutex_unlock(&journal->j_checkpoint_mutex);
1688 brelse(journal->j_sb_buffer);
1691 if (journal->j_proc_entry)
1692 jbd2_stats_proc_exit(journal);
1693 if (journal->j_inode)
1694 iput(journal->j_inode);
1695 if (journal->j_revoke)
1696 jbd2_journal_destroy_revoke(journal);
1697 if (journal->j_chksum_driver)
1698 crypto_free_shash(journal->j_chksum_driver);
1699 kfree(journal->j_wbuf);
1707 *int jbd2_journal_check_used_features () - Check if features specified are used.
1708 * @journal: Journal to check.
1709 * @compat: bitmask of compatible features
1710 * @ro: bitmask of features that force read-only mount
1711 * @incompat: bitmask of incompatible features
1713 * Check whether the journal uses all of a given set of
1714 * features. Return true (non-zero) if it does.
1717 int jbd2_journal_check_used_features (journal_t *journal, unsigned long compat,
1718 unsigned long ro, unsigned long incompat)
1720 journal_superblock_t *sb;
1722 if (!compat && !ro && !incompat)
1724 /* Load journal superblock if it is not loaded yet. */
1725 if (journal->j_format_version == 0 &&
1726 journal_get_superblock(journal) != 0)
1728 if (journal->j_format_version == 1)
1731 sb = journal->j_superblock;
1733 if (((be32_to_cpu(sb->s_feature_compat) & compat) == compat) &&
1734 ((be32_to_cpu(sb->s_feature_ro_compat) & ro) == ro) &&
1735 ((be32_to_cpu(sb->s_feature_incompat) & incompat) == incompat))
1742 * int jbd2_journal_check_available_features() - Check feature set in journalling layer
1743 * @journal: Journal to check.
1744 * @compat: bitmask of compatible features
1745 * @ro: bitmask of features that force read-only mount
1746 * @incompat: bitmask of incompatible features
1748 * Check whether the journaling code supports the use of
1749 * all of a given set of features on this journal. Return true
1750 * (non-zero) if it can. */
1752 int jbd2_journal_check_available_features (journal_t *journal, unsigned long compat,
1753 unsigned long ro, unsigned long incompat)
1755 if (!compat && !ro && !incompat)
1758 /* We can support any known requested features iff the
1759 * superblock is in version 2. Otherwise we fail to support any
1760 * extended sb features. */
1762 if (journal->j_format_version != 2)
1765 if ((compat & JBD2_KNOWN_COMPAT_FEATURES) == compat &&
1766 (ro & JBD2_KNOWN_ROCOMPAT_FEATURES) == ro &&
1767 (incompat & JBD2_KNOWN_INCOMPAT_FEATURES) == incompat)
1774 * int jbd2_journal_set_features () - Mark a given journal feature in the superblock
1775 * @journal: Journal to act on.
1776 * @compat: bitmask of compatible features
1777 * @ro: bitmask of features that force read-only mount
1778 * @incompat: bitmask of incompatible features
1780 * Mark a given journal feature as present on the
1781 * superblock. Returns true if the requested features could be set.
1785 int jbd2_journal_set_features (journal_t *journal, unsigned long compat,
1786 unsigned long ro, unsigned long incompat)
1788 #define INCOMPAT_FEATURE_ON(f) \
1789 ((incompat & (f)) && !(sb->s_feature_incompat & cpu_to_be32(f)))
1790 #define COMPAT_FEATURE_ON(f) \
1791 ((compat & (f)) && !(sb->s_feature_compat & cpu_to_be32(f)))
1792 journal_superblock_t *sb;
1794 if (jbd2_journal_check_used_features(journal, compat, ro, incompat))
1797 if (!jbd2_journal_check_available_features(journal, compat, ro, incompat))
1800 /* Asking for checksumming v2 and v1? Only give them v2. */
1801 if (incompat & JBD2_FEATURE_INCOMPAT_CSUM_V2 &&
1802 compat & JBD2_FEATURE_COMPAT_CHECKSUM)
1803 compat &= ~JBD2_FEATURE_COMPAT_CHECKSUM;
1805 jbd_debug(1, "Setting new features 0x%lx/0x%lx/0x%lx\n",
1806 compat, ro, incompat);
1808 sb = journal->j_superblock;
1810 /* If enabling v2 checksums, update superblock */
1811 if (INCOMPAT_FEATURE_ON(JBD2_FEATURE_INCOMPAT_CSUM_V2)) {
1812 sb->s_checksum_type = JBD2_CRC32C_CHKSUM;
1813 sb->s_feature_compat &=
1814 ~cpu_to_be32(JBD2_FEATURE_COMPAT_CHECKSUM);
1816 /* Load the checksum driver */
1817 if (journal->j_chksum_driver == NULL) {
1818 journal->j_chksum_driver = crypto_alloc_shash("crc32c",
1820 if (IS_ERR(journal->j_chksum_driver)) {
1821 printk(KERN_ERR "JBD: Cannot load crc32c "
1823 journal->j_chksum_driver = NULL;
1828 /* Precompute checksum seed for all metadata */
1829 if (JBD2_HAS_INCOMPAT_FEATURE(journal,
1830 JBD2_FEATURE_INCOMPAT_CSUM_V2))
1831 journal->j_csum_seed = jbd2_chksum(journal, ~0,
1833 sizeof(sb->s_uuid));
1836 /* If enabling v1 checksums, downgrade superblock */
1837 if (COMPAT_FEATURE_ON(JBD2_FEATURE_COMPAT_CHECKSUM))
1838 sb->s_feature_incompat &=
1839 ~cpu_to_be32(JBD2_FEATURE_INCOMPAT_CSUM_V2);
1841 sb->s_feature_compat |= cpu_to_be32(compat);
1842 sb->s_feature_ro_compat |= cpu_to_be32(ro);
1843 sb->s_feature_incompat |= cpu_to_be32(incompat);
1846 #undef COMPAT_FEATURE_ON
1847 #undef INCOMPAT_FEATURE_ON
1851 * jbd2_journal_clear_features () - Clear a given journal feature in the
1853 * @journal: Journal to act on.
1854 * @compat: bitmask of compatible features
1855 * @ro: bitmask of features that force read-only mount
1856 * @incompat: bitmask of incompatible features
1858 * Clear a given journal feature as present on the
1861 void jbd2_journal_clear_features(journal_t *journal, unsigned long compat,
1862 unsigned long ro, unsigned long incompat)
1864 journal_superblock_t *sb;
1866 jbd_debug(1, "Clear features 0x%lx/0x%lx/0x%lx\n",
1867 compat, ro, incompat);
1869 sb = journal->j_superblock;
1871 sb->s_feature_compat &= ~cpu_to_be32(compat);
1872 sb->s_feature_ro_compat &= ~cpu_to_be32(ro);
1873 sb->s_feature_incompat &= ~cpu_to_be32(incompat);
1875 EXPORT_SYMBOL(jbd2_journal_clear_features);
1878 * int jbd2_journal_flush () - Flush journal
1879 * @journal: Journal to act on.
1881 * Flush all data for a given journal to disk and empty the journal.
1882 * Filesystems can use this when remounting readonly to ensure that
1883 * recovery does not need to happen on remount.
1886 int jbd2_journal_flush(journal_t *journal)
1889 transaction_t *transaction = NULL;
1891 write_lock(&journal->j_state_lock);
1893 /* Force everything buffered to the log... */
1894 if (journal->j_running_transaction) {
1895 transaction = journal->j_running_transaction;
1896 __jbd2_log_start_commit(journal, transaction->t_tid);
1897 } else if (journal->j_committing_transaction)
1898 transaction = journal->j_committing_transaction;
1900 /* Wait for the log commit to complete... */
1902 tid_t tid = transaction->t_tid;
1904 write_unlock(&journal->j_state_lock);
1905 jbd2_log_wait_commit(journal, tid);
1907 write_unlock(&journal->j_state_lock);
1910 /* ...and flush everything in the log out to disk. */
1911 spin_lock(&journal->j_list_lock);
1912 while (!err && journal->j_checkpoint_transactions != NULL) {
1913 spin_unlock(&journal->j_list_lock);
1914 mutex_lock(&journal->j_checkpoint_mutex);
1915 err = jbd2_log_do_checkpoint(journal);
1916 mutex_unlock(&journal->j_checkpoint_mutex);
1917 spin_lock(&journal->j_list_lock);
1919 spin_unlock(&journal->j_list_lock);
1921 if (is_journal_aborted(journal))
1924 mutex_lock(&journal->j_checkpoint_mutex);
1925 jbd2_cleanup_journal_tail(journal);
1927 /* Finally, mark the journal as really needing no recovery.
1928 * This sets s_start==0 in the underlying superblock, which is
1929 * the magic code for a fully-recovered superblock. Any future
1930 * commits of data to the journal will restore the current
1932 jbd2_mark_journal_empty(journal);
1933 mutex_unlock(&journal->j_checkpoint_mutex);
1934 write_lock(&journal->j_state_lock);
1935 J_ASSERT(!journal->j_running_transaction);
1936 J_ASSERT(!journal->j_committing_transaction);
1937 J_ASSERT(!journal->j_checkpoint_transactions);
1938 J_ASSERT(journal->j_head == journal->j_tail);
1939 J_ASSERT(journal->j_tail_sequence == journal->j_transaction_sequence);
1940 write_unlock(&journal->j_state_lock);
1945 * int jbd2_journal_wipe() - Wipe journal contents
1946 * @journal: Journal to act on.
1947 * @write: flag (see below)
1949 * Wipe out all of the contents of a journal, safely. This will produce
1950 * a warning if the journal contains any valid recovery information.
1951 * Must be called between journal_init_*() and jbd2_journal_load().
1953 * If 'write' is non-zero, then we wipe out the journal on disk; otherwise
1954 * we merely suppress recovery.
1957 int jbd2_journal_wipe(journal_t *journal, int write)
1961 J_ASSERT (!(journal->j_flags & JBD2_LOADED));
1963 err = load_superblock(journal);
1967 if (!journal->j_tail)
1970 printk(KERN_WARNING "JBD2: %s recovery information on journal\n",
1971 write ? "Clearing" : "Ignoring");
1973 err = jbd2_journal_skip_recovery(journal);
1975 /* Lock to make assertions happy... */
1976 mutex_lock(&journal->j_checkpoint_mutex);
1977 jbd2_mark_journal_empty(journal);
1978 mutex_unlock(&journal->j_checkpoint_mutex);
1986 * Journal abort has very specific semantics, which we describe
1987 * for journal abort.
1989 * Two internal functions, which provide abort to the jbd layer
1994 * Quick version for internal journal use (doesn't lock the journal).
1995 * Aborts hard --- we mark the abort as occurred, but do _nothing_ else,
1996 * and don't attempt to make any other journal updates.
1998 void __jbd2_journal_abort_hard(journal_t *journal)
2000 transaction_t *transaction;
2002 if (journal->j_flags & JBD2_ABORT)
2005 printk(KERN_ERR "Aborting journal on device %s.\n",
2006 journal->j_devname);
2008 write_lock(&journal->j_state_lock);
2009 journal->j_flags |= JBD2_ABORT;
2010 transaction = journal->j_running_transaction;
2012 __jbd2_log_start_commit(journal, transaction->t_tid);
2013 write_unlock(&journal->j_state_lock);
2016 /* Soft abort: record the abort error status in the journal superblock,
2017 * but don't do any other IO. */
2018 static void __journal_abort_soft (journal_t *journal, int errno)
2020 if (journal->j_flags & JBD2_ABORT)
2023 if (!journal->j_errno)
2024 journal->j_errno = errno;
2026 __jbd2_journal_abort_hard(journal);
2029 jbd2_journal_update_sb_errno(journal);
2033 * void jbd2_journal_abort () - Shutdown the journal immediately.
2034 * @journal: the journal to shutdown.
2035 * @errno: an error number to record in the journal indicating
2036 * the reason for the shutdown.
2038 * Perform a complete, immediate shutdown of the ENTIRE
2039 * journal (not of a single transaction). This operation cannot be
2040 * undone without closing and reopening the journal.
2042 * The jbd2_journal_abort function is intended to support higher level error
2043 * recovery mechanisms such as the ext2/ext3 remount-readonly error
2046 * Journal abort has very specific semantics. Any existing dirty,
2047 * unjournaled buffers in the main filesystem will still be written to
2048 * disk by bdflush, but the journaling mechanism will be suspended
2049 * immediately and no further transaction commits will be honoured.
2051 * Any dirty, journaled buffers will be written back to disk without
2052 * hitting the journal. Atomicity cannot be guaranteed on an aborted
2053 * filesystem, but we _do_ attempt to leave as much data as possible
2054 * behind for fsck to use for cleanup.
2056 * Any attempt to get a new transaction handle on a journal which is in
2057 * ABORT state will just result in an -EROFS error return. A
2058 * jbd2_journal_stop on an existing handle will return -EIO if we have
2059 * entered abort state during the update.
2061 * Recursive transactions are not disturbed by journal abort until the
2062 * final jbd2_journal_stop, which will receive the -EIO error.
2064 * Finally, the jbd2_journal_abort call allows the caller to supply an errno
2065 * which will be recorded (if possible) in the journal superblock. This
2066 * allows a client to record failure conditions in the middle of a
2067 * transaction without having to complete the transaction to record the
2068 * failure to disk. ext3_error, for example, now uses this
2071 * Errors which originate from within the journaling layer will NOT
2072 * supply an errno; a null errno implies that absolutely no further
2073 * writes are done to the journal (unless there are any already in
2078 void jbd2_journal_abort(journal_t *journal, int errno)
2080 __journal_abort_soft(journal, errno);
2084 * int jbd2_journal_errno () - returns the journal's error state.
2085 * @journal: journal to examine.
2087 * This is the errno number set with jbd2_journal_abort(), the last
2088 * time the journal was mounted - if the journal was stopped
2089 * without calling abort this will be 0.
2091 * If the journal has been aborted on this mount time -EROFS will
2094 int jbd2_journal_errno(journal_t *journal)
2098 read_lock(&journal->j_state_lock);
2099 if (journal->j_flags & JBD2_ABORT)
2102 err = journal->j_errno;
2103 read_unlock(&journal->j_state_lock);
2108 * int jbd2_journal_clear_err () - clears the journal's error state
2109 * @journal: journal to act on.
2111 * An error must be cleared or acked to take a FS out of readonly
2114 int jbd2_journal_clear_err(journal_t *journal)
2118 write_lock(&journal->j_state_lock);
2119 if (journal->j_flags & JBD2_ABORT)
2122 journal->j_errno = 0;
2123 write_unlock(&journal->j_state_lock);
2128 * void jbd2_journal_ack_err() - Ack journal err.
2129 * @journal: journal to act on.
2131 * An error must be cleared or acked to take a FS out of readonly
2134 void jbd2_journal_ack_err(journal_t *journal)
2136 write_lock(&journal->j_state_lock);
2137 if (journal->j_errno)
2138 journal->j_flags |= JBD2_ACK_ERR;
2139 write_unlock(&journal->j_state_lock);
2142 int jbd2_journal_blocks_per_page(struct inode *inode)
2144 return 1 << (PAGE_CACHE_SHIFT - inode->i_sb->s_blocksize_bits);
2148 * helper functions to deal with 32 or 64bit block numbers.
2150 size_t journal_tag_bytes(journal_t *journal)
2152 journal_block_tag_t tag;
2155 if (JBD2_HAS_INCOMPAT_FEATURE(journal, JBD2_FEATURE_INCOMPAT_CSUM_V2))
2156 x += sizeof(tag.t_checksum);
2158 if (JBD2_HAS_INCOMPAT_FEATURE(journal, JBD2_FEATURE_INCOMPAT_64BIT))
2159 return x + JBD2_TAG_SIZE64;
2161 return x + JBD2_TAG_SIZE32;
2165 * JBD memory management
2167 * These functions are used to allocate block-sized chunks of memory
2168 * used for making copies of buffer_head data. Very often it will be
2169 * page-sized chunks of data, but sometimes it will be in
2170 * sub-page-size chunks. (For example, 16k pages on Power systems
2171 * with a 4k block file system.) For blocks smaller than a page, we
2172 * use a SLAB allocator. There are slab caches for each block size,
2173 * which are allocated at mount time, if necessary, and we only free
2174 * (all of) the slab caches when/if the jbd2 module is unloaded. For
2175 * this reason we don't need to a mutex to protect access to
2176 * jbd2_slab[] allocating or releasing memory; only in
2177 * jbd2_journal_create_slab().
2179 #define JBD2_MAX_SLABS 8
2180 static struct kmem_cache *jbd2_slab[JBD2_MAX_SLABS];
2182 static const char *jbd2_slab_names[JBD2_MAX_SLABS] = {
2183 "jbd2_1k", "jbd2_2k", "jbd2_4k", "jbd2_8k",
2184 "jbd2_16k", "jbd2_32k", "jbd2_64k", "jbd2_128k"
2188 static void jbd2_journal_destroy_slabs(void)
2192 for (i = 0; i < JBD2_MAX_SLABS; i++) {
2194 kmem_cache_destroy(jbd2_slab[i]);
2195 jbd2_slab[i] = NULL;
2199 static int jbd2_journal_create_slab(size_t size)
2201 static DEFINE_MUTEX(jbd2_slab_create_mutex);
2202 int i = order_base_2(size) - 10;
2205 if (size == PAGE_SIZE)
2208 if (i >= JBD2_MAX_SLABS)
2211 if (unlikely(i < 0))
2213 mutex_lock(&jbd2_slab_create_mutex);
2215 mutex_unlock(&jbd2_slab_create_mutex);
2216 return 0; /* Already created */
2219 slab_size = 1 << (i+10);
2220 jbd2_slab[i] = kmem_cache_create(jbd2_slab_names[i], slab_size,
2221 slab_size, 0, NULL);
2222 mutex_unlock(&jbd2_slab_create_mutex);
2223 if (!jbd2_slab[i]) {
2224 printk(KERN_EMERG "JBD2: no memory for jbd2_slab cache\n");
2230 static struct kmem_cache *get_slab(size_t size)
2232 int i = order_base_2(size) - 10;
2234 BUG_ON(i >= JBD2_MAX_SLABS);
2235 if (unlikely(i < 0))
2237 BUG_ON(jbd2_slab[i] == NULL);
2238 return jbd2_slab[i];
2241 void *jbd2_alloc(size_t size, gfp_t flags)
2245 BUG_ON(size & (size-1)); /* Must be a power of 2 */
2247 flags |= __GFP_REPEAT;
2248 if (size == PAGE_SIZE)
2249 ptr = (void *)__get_free_pages(flags, 0);
2250 else if (size > PAGE_SIZE) {
2251 int order = get_order(size);
2254 ptr = (void *)__get_free_pages(flags, order);
2256 ptr = vmalloc(size);
2258 ptr = kmem_cache_alloc(get_slab(size), flags);
2260 /* Check alignment; SLUB has gotten this wrong in the past,
2261 * and this can lead to user data corruption! */
2262 BUG_ON(((unsigned long) ptr) & (size-1));
2267 void jbd2_free(void *ptr, size_t size)
2269 if (size == PAGE_SIZE) {
2270 free_pages((unsigned long)ptr, 0);
2273 if (size > PAGE_SIZE) {
2274 int order = get_order(size);
2277 free_pages((unsigned long)ptr, order);
2282 kmem_cache_free(get_slab(size), ptr);
2286 * Journal_head storage management
2288 static struct kmem_cache *jbd2_journal_head_cache;
2289 #ifdef CONFIG_JBD2_DEBUG
2290 static atomic_t nr_journal_heads = ATOMIC_INIT(0);
2293 static int jbd2_journal_init_journal_head_cache(void)
2297 J_ASSERT(jbd2_journal_head_cache == NULL);
2298 jbd2_journal_head_cache = kmem_cache_create("jbd2_journal_head",
2299 sizeof(struct journal_head),
2301 SLAB_TEMPORARY, /* flags */
2304 if (!jbd2_journal_head_cache) {
2306 printk(KERN_EMERG "JBD2: no memory for journal_head cache\n");
2311 static void jbd2_journal_destroy_journal_head_cache(void)
2313 if (jbd2_journal_head_cache) {
2314 kmem_cache_destroy(jbd2_journal_head_cache);
2315 jbd2_journal_head_cache = NULL;
2320 * journal_head splicing and dicing
2322 static struct journal_head *journal_alloc_journal_head(void)
2324 struct journal_head *ret;
2326 #ifdef CONFIG_JBD2_DEBUG
2327 atomic_inc(&nr_journal_heads);
2329 ret = kmem_cache_alloc(jbd2_journal_head_cache, GFP_NOFS);
2331 jbd_debug(1, "out of memory for journal_head\n");
2332 pr_notice_ratelimited("ENOMEM in %s, retrying.\n", __func__);
2335 ret = kmem_cache_alloc(jbd2_journal_head_cache, GFP_NOFS);
2341 static void journal_free_journal_head(struct journal_head *jh)
2343 #ifdef CONFIG_JBD2_DEBUG
2344 atomic_dec(&nr_journal_heads);
2345 memset(jh, JBD2_POISON_FREE, sizeof(*jh));
2347 kmem_cache_free(jbd2_journal_head_cache, jh);
2351 * A journal_head is attached to a buffer_head whenever JBD has an
2352 * interest in the buffer.
2354 * Whenever a buffer has an attached journal_head, its ->b_state:BH_JBD bit
2355 * is set. This bit is tested in core kernel code where we need to take
2356 * JBD-specific actions. Testing the zeroness of ->b_private is not reliable
2359 * When a buffer has its BH_JBD bit set, its ->b_count is elevated by one.
2361 * When a buffer has its BH_JBD bit set it is immune from being released by
2362 * core kernel code, mainly via ->b_count.
2364 * A journal_head is detached from its buffer_head when the journal_head's
2365 * b_jcount reaches zero. Running transaction (b_transaction) and checkpoint
2366 * transaction (b_cp_transaction) hold their references to b_jcount.
2368 * Various places in the kernel want to attach a journal_head to a buffer_head
2369 * _before_ attaching the journal_head to a transaction. To protect the
2370 * journal_head in this situation, jbd2_journal_add_journal_head elevates the
2371 * journal_head's b_jcount refcount by one. The caller must call
2372 * jbd2_journal_put_journal_head() to undo this.
2374 * So the typical usage would be:
2376 * (Attach a journal_head if needed. Increments b_jcount)
2377 * struct journal_head *jh = jbd2_journal_add_journal_head(bh);
2379 * (Get another reference for transaction)
2380 * jbd2_journal_grab_journal_head(bh);
2381 * jh->b_transaction = xxx;
2382 * (Put original reference)
2383 * jbd2_journal_put_journal_head(jh);
2387 * Give a buffer_head a journal_head.
2391 struct journal_head *jbd2_journal_add_journal_head(struct buffer_head *bh)
2393 struct journal_head *jh;
2394 struct journal_head *new_jh = NULL;
2397 if (!buffer_jbd(bh)) {
2398 new_jh = journal_alloc_journal_head();
2399 memset(new_jh, 0, sizeof(*new_jh));
2402 jbd_lock_bh_journal_head(bh);
2403 if (buffer_jbd(bh)) {
2407 (atomic_read(&bh->b_count) > 0) ||
2408 (bh->b_page && bh->b_page->mapping));
2411 jbd_unlock_bh_journal_head(bh);
2416 new_jh = NULL; /* We consumed it */
2421 BUFFER_TRACE(bh, "added journal_head");
2424 jbd_unlock_bh_journal_head(bh);
2426 journal_free_journal_head(new_jh);
2427 return bh->b_private;
2431 * Grab a ref against this buffer_head's journal_head. If it ended up not
2432 * having a journal_head, return NULL
2434 struct journal_head *jbd2_journal_grab_journal_head(struct buffer_head *bh)
2436 struct journal_head *jh = NULL;
2438 jbd_lock_bh_journal_head(bh);
2439 if (buffer_jbd(bh)) {
2443 jbd_unlock_bh_journal_head(bh);
2447 static void __journal_remove_journal_head(struct buffer_head *bh)
2449 struct journal_head *jh = bh2jh(bh);
2451 J_ASSERT_JH(jh, jh->b_jcount >= 0);
2452 J_ASSERT_JH(jh, jh->b_transaction == NULL);
2453 J_ASSERT_JH(jh, jh->b_next_transaction == NULL);
2454 J_ASSERT_JH(jh, jh->b_cp_transaction == NULL);
2455 J_ASSERT_JH(jh, jh->b_jlist == BJ_None);
2456 J_ASSERT_BH(bh, buffer_jbd(bh));
2457 J_ASSERT_BH(bh, jh2bh(jh) == bh);
2458 BUFFER_TRACE(bh, "remove journal_head");
2459 if (jh->b_frozen_data) {
2460 printk(KERN_WARNING "%s: freeing b_frozen_data\n", __func__);
2461 jbd2_free(jh->b_frozen_data, bh->b_size);
2463 if (jh->b_committed_data) {
2464 printk(KERN_WARNING "%s: freeing b_committed_data\n", __func__);
2465 jbd2_free(jh->b_committed_data, bh->b_size);
2467 bh->b_private = NULL;
2468 jh->b_bh = NULL; /* debug, really */
2469 clear_buffer_jbd(bh);
2470 journal_free_journal_head(jh);
2474 * Drop a reference on the passed journal_head. If it fell to zero then
2475 * release the journal_head from the buffer_head.
2477 void jbd2_journal_put_journal_head(struct journal_head *jh)
2479 struct buffer_head *bh = jh2bh(jh);
2481 jbd_lock_bh_journal_head(bh);
2482 J_ASSERT_JH(jh, jh->b_jcount > 0);
2484 if (!jh->b_jcount) {
2485 __journal_remove_journal_head(bh);
2486 jbd_unlock_bh_journal_head(bh);
2489 jbd_unlock_bh_journal_head(bh);
2493 * Initialize jbd inode head
2495 void jbd2_journal_init_jbd_inode(struct jbd2_inode *jinode, struct inode *inode)
2497 jinode->i_transaction = NULL;
2498 jinode->i_next_transaction = NULL;
2499 jinode->i_vfs_inode = inode;
2500 jinode->i_flags = 0;
2501 INIT_LIST_HEAD(&jinode->i_list);
2505 * Function to be called before we start removing inode from memory (i.e.,
2506 * clear_inode() is a fine place to be called from). It removes inode from
2507 * transaction's lists.
2509 void jbd2_journal_release_jbd_inode(journal_t *journal,
2510 struct jbd2_inode *jinode)
2515 spin_lock(&journal->j_list_lock);
2516 /* Is commit writing out inode - we have to wait */
2517 if (test_bit(__JI_COMMIT_RUNNING, &jinode->i_flags)) {
2518 wait_queue_head_t *wq;
2519 DEFINE_WAIT_BIT(wait, &jinode->i_flags, __JI_COMMIT_RUNNING);
2520 wq = bit_waitqueue(&jinode->i_flags, __JI_COMMIT_RUNNING);
2521 prepare_to_wait(wq, &wait.wait, TASK_UNINTERRUPTIBLE);
2522 spin_unlock(&journal->j_list_lock);
2524 finish_wait(wq, &wait.wait);
2528 if (jinode->i_transaction) {
2529 list_del(&jinode->i_list);
2530 jinode->i_transaction = NULL;
2532 spin_unlock(&journal->j_list_lock);
2536 #ifdef CONFIG_PROC_FS
2538 #define JBD2_STATS_PROC_NAME "fs/jbd2"
2540 static void __init jbd2_create_jbd_stats_proc_entry(void)
2542 proc_jbd2_stats = proc_mkdir(JBD2_STATS_PROC_NAME, NULL);
2545 static void __exit jbd2_remove_jbd_stats_proc_entry(void)
2547 if (proc_jbd2_stats)
2548 remove_proc_entry(JBD2_STATS_PROC_NAME, NULL);
2553 #define jbd2_create_jbd_stats_proc_entry() do {} while (0)
2554 #define jbd2_remove_jbd_stats_proc_entry() do {} while (0)
2558 struct kmem_cache *jbd2_handle_cache, *jbd2_inode_cache;
2560 static int __init jbd2_journal_init_handle_cache(void)
2562 jbd2_handle_cache = KMEM_CACHE(jbd2_journal_handle, SLAB_TEMPORARY);
2563 if (jbd2_handle_cache == NULL) {
2564 printk(KERN_EMERG "JBD2: failed to create handle cache\n");
2567 jbd2_inode_cache = KMEM_CACHE(jbd2_inode, 0);
2568 if (jbd2_inode_cache == NULL) {
2569 printk(KERN_EMERG "JBD2: failed to create inode cache\n");
2570 kmem_cache_destroy(jbd2_handle_cache);
2576 static void jbd2_journal_destroy_handle_cache(void)
2578 if (jbd2_handle_cache)
2579 kmem_cache_destroy(jbd2_handle_cache);
2580 if (jbd2_inode_cache)
2581 kmem_cache_destroy(jbd2_inode_cache);
2586 * Module startup and shutdown
2589 static int __init journal_init_caches(void)
2593 ret = jbd2_journal_init_revoke_caches();
2595 ret = jbd2_journal_init_journal_head_cache();
2597 ret = jbd2_journal_init_handle_cache();
2599 ret = jbd2_journal_init_transaction_cache();
2603 static void jbd2_journal_destroy_caches(void)
2605 jbd2_journal_destroy_revoke_caches();
2606 jbd2_journal_destroy_journal_head_cache();
2607 jbd2_journal_destroy_handle_cache();
2608 jbd2_journal_destroy_transaction_cache();
2609 jbd2_journal_destroy_slabs();
2612 static int __init journal_init(void)
2616 BUILD_BUG_ON(sizeof(struct journal_superblock_s) != 1024);
2618 ret = journal_init_caches();
2620 jbd2_create_jbd_stats_proc_entry();
2622 jbd2_journal_destroy_caches();
2627 static void __exit journal_exit(void)
2629 #ifdef CONFIG_JBD2_DEBUG
2630 int n = atomic_read(&nr_journal_heads);
2632 printk(KERN_EMERG "JBD2: leaked %d journal_heads!\n", n);
2634 jbd2_remove_jbd_stats_proc_entry();
2635 jbd2_journal_destroy_caches();
2638 MODULE_LICENSE("GPL");
2639 module_init(journal_init);
2640 module_exit(journal_exit);