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>
46 #include <linux/sched/mm.h>
48 #define CREATE_TRACE_POINTS
49 #include <trace/events/jbd2.h>
51 #include <linux/uaccess.h>
54 #ifdef CONFIG_JBD2_DEBUG
55 ushort jbd2_journal_enable_debug __read_mostly;
56 EXPORT_SYMBOL(jbd2_journal_enable_debug);
58 module_param_named(jbd2_debug, jbd2_journal_enable_debug, ushort, 0644);
59 MODULE_PARM_DESC(jbd2_debug, "Debugging level for jbd2");
62 EXPORT_SYMBOL(jbd2_journal_extend);
63 EXPORT_SYMBOL(jbd2_journal_stop);
64 EXPORT_SYMBOL(jbd2_journal_lock_updates);
65 EXPORT_SYMBOL(jbd2_journal_unlock_updates);
66 EXPORT_SYMBOL(jbd2_journal_get_write_access);
67 EXPORT_SYMBOL(jbd2_journal_get_create_access);
68 EXPORT_SYMBOL(jbd2_journal_get_undo_access);
69 EXPORT_SYMBOL(jbd2_journal_set_triggers);
70 EXPORT_SYMBOL(jbd2_journal_dirty_metadata);
71 EXPORT_SYMBOL(jbd2_journal_forget);
73 EXPORT_SYMBOL(journal_sync_buffer);
75 EXPORT_SYMBOL(jbd2_journal_flush);
76 EXPORT_SYMBOL(jbd2_journal_revoke);
78 EXPORT_SYMBOL(jbd2_journal_init_dev);
79 EXPORT_SYMBOL(jbd2_journal_init_inode);
80 EXPORT_SYMBOL(jbd2_journal_check_used_features);
81 EXPORT_SYMBOL(jbd2_journal_check_available_features);
82 EXPORT_SYMBOL(jbd2_journal_set_features);
83 EXPORT_SYMBOL(jbd2_journal_load);
84 EXPORT_SYMBOL(jbd2_journal_destroy);
85 EXPORT_SYMBOL(jbd2_journal_abort);
86 EXPORT_SYMBOL(jbd2_journal_errno);
87 EXPORT_SYMBOL(jbd2_journal_ack_err);
88 EXPORT_SYMBOL(jbd2_journal_clear_err);
89 EXPORT_SYMBOL(jbd2_log_wait_commit);
90 EXPORT_SYMBOL(jbd2_log_start_commit);
91 EXPORT_SYMBOL(jbd2_journal_start_commit);
92 EXPORT_SYMBOL(jbd2_journal_force_commit_nested);
93 EXPORT_SYMBOL(jbd2_journal_wipe);
94 EXPORT_SYMBOL(jbd2_journal_blocks_per_page);
95 EXPORT_SYMBOL(jbd2_journal_invalidatepage);
96 EXPORT_SYMBOL(jbd2_journal_try_to_free_buffers);
97 EXPORT_SYMBOL(jbd2_journal_force_commit);
98 EXPORT_SYMBOL(jbd2_journal_inode_add_write);
99 EXPORT_SYMBOL(jbd2_journal_inode_add_wait);
100 EXPORT_SYMBOL(jbd2_journal_init_jbd_inode);
101 EXPORT_SYMBOL(jbd2_journal_release_jbd_inode);
102 EXPORT_SYMBOL(jbd2_journal_begin_ordered_truncate);
103 EXPORT_SYMBOL(jbd2_inode_cache);
105 static void __journal_abort_soft (journal_t *journal, int errno);
106 static int jbd2_journal_create_slab(size_t slab_size);
108 #ifdef CONFIG_JBD2_DEBUG
109 void __jbd2_debug(int level, const char *file, const char *func,
110 unsigned int line, const char *fmt, ...)
112 struct va_format vaf;
115 if (level > jbd2_journal_enable_debug)
120 printk(KERN_DEBUG "%s: (%s, %u): %pV\n", file, func, line, &vaf);
123 EXPORT_SYMBOL(__jbd2_debug);
126 /* Checksumming functions */
127 static int jbd2_verify_csum_type(journal_t *j, journal_superblock_t *sb)
129 if (!jbd2_journal_has_csum_v2or3_feature(j))
132 return sb->s_checksum_type == JBD2_CRC32C_CHKSUM;
135 static __be32 jbd2_superblock_csum(journal_t *j, journal_superblock_t *sb)
140 old_csum = sb->s_checksum;
142 csum = jbd2_chksum(j, ~0, (char *)sb, sizeof(journal_superblock_t));
143 sb->s_checksum = old_csum;
145 return cpu_to_be32(csum);
148 static int jbd2_superblock_csum_verify(journal_t *j, journal_superblock_t *sb)
150 if (!jbd2_journal_has_csum_v2or3(j))
153 return sb->s_checksum == jbd2_superblock_csum(j, sb);
156 static void jbd2_superblock_csum_set(journal_t *j, journal_superblock_t *sb)
158 if (!jbd2_journal_has_csum_v2or3(j))
161 sb->s_checksum = jbd2_superblock_csum(j, sb);
165 * Helper function used to manage commit timeouts
168 static void commit_timeout(unsigned long __data)
170 struct task_struct * p = (struct task_struct *) __data;
176 * kjournald2: The main thread function used to manage a logging device
179 * This kernel thread is responsible for two things:
181 * 1) COMMIT: Every so often we need to commit the current state of the
182 * filesystem to disk. The journal thread is responsible for writing
183 * all of the metadata buffers to disk.
185 * 2) CHECKPOINT: We cannot reuse a used section of the log file until all
186 * of the data in that part of the log has been rewritten elsewhere on
187 * the disk. Flushing these old buffers to reclaim space in the log is
188 * known as checkpointing, and this thread is responsible for that job.
191 static int kjournald2(void *arg)
193 journal_t *journal = arg;
194 transaction_t *transaction;
197 * Set up an interval timer which can be used to trigger a commit wakeup
198 * after the commit interval expires
200 setup_timer(&journal->j_commit_timer, commit_timeout,
201 (unsigned long)current);
205 /* Record that the journal thread is running */
206 journal->j_task = current;
207 wake_up(&journal->j_wait_done_commit);
210 * Make sure that no allocations from this kernel thread will ever
211 * recurse to the fs layer because we are responsible for the
212 * transaction commit and any fs involvement might get stuck waiting for
215 memalloc_nofs_save();
218 * And now, wait forever for commit wakeup events.
220 write_lock(&journal->j_state_lock);
223 if (journal->j_flags & JBD2_UNMOUNT)
226 jbd_debug(1, "commit_sequence=%d, commit_request=%d\n",
227 journal->j_commit_sequence, journal->j_commit_request);
229 if (journal->j_commit_sequence != journal->j_commit_request) {
230 jbd_debug(1, "OK, requests differ\n");
231 write_unlock(&journal->j_state_lock);
232 del_timer_sync(&journal->j_commit_timer);
233 jbd2_journal_commit_transaction(journal);
234 write_lock(&journal->j_state_lock);
238 wake_up(&journal->j_wait_done_commit);
239 if (freezing(current)) {
241 * The simpler the better. Flushing journal isn't a
242 * good idea, because that depends on threads that may
243 * be already stopped.
245 jbd_debug(1, "Now suspending kjournald2\n");
246 write_unlock(&journal->j_state_lock);
248 write_lock(&journal->j_state_lock);
251 * We assume on resume that commits are already there,
255 int should_sleep = 1;
257 prepare_to_wait(&journal->j_wait_commit, &wait,
259 if (journal->j_commit_sequence != journal->j_commit_request)
261 transaction = journal->j_running_transaction;
262 if (transaction && time_after_eq(jiffies,
263 transaction->t_expires))
265 if (journal->j_flags & JBD2_UNMOUNT)
268 write_unlock(&journal->j_state_lock);
270 write_lock(&journal->j_state_lock);
272 finish_wait(&journal->j_wait_commit, &wait);
275 jbd_debug(1, "kjournald2 wakes\n");
278 * Were we woken up by a commit wakeup event?
280 transaction = journal->j_running_transaction;
281 if (transaction && time_after_eq(jiffies, transaction->t_expires)) {
282 journal->j_commit_request = transaction->t_tid;
283 jbd_debug(1, "woke because of timeout\n");
288 del_timer_sync(&journal->j_commit_timer);
289 journal->j_task = NULL;
290 wake_up(&journal->j_wait_done_commit);
291 jbd_debug(1, "Journal thread exiting.\n");
292 write_unlock(&journal->j_state_lock);
296 static int jbd2_journal_start_thread(journal_t *journal)
298 struct task_struct *t;
300 t = kthread_run(kjournald2, journal, "jbd2/%s",
305 wait_event(journal->j_wait_done_commit, journal->j_task != NULL);
309 static void journal_kill_thread(journal_t *journal)
311 write_lock(&journal->j_state_lock);
312 journal->j_flags |= JBD2_UNMOUNT;
314 while (journal->j_task) {
315 write_unlock(&journal->j_state_lock);
316 wake_up(&journal->j_wait_commit);
317 wait_event(journal->j_wait_done_commit, journal->j_task == NULL);
318 write_lock(&journal->j_state_lock);
320 write_unlock(&journal->j_state_lock);
324 * jbd2_journal_write_metadata_buffer: write a metadata buffer to the journal.
326 * Writes a metadata buffer to a given disk block. The actual IO is not
327 * performed but a new buffer_head is constructed which labels the data
328 * to be written with the correct destination disk block.
330 * Any magic-number escaping which needs to be done will cause a
331 * copy-out here. If the buffer happens to start with the
332 * JBD2_MAGIC_NUMBER, then we can't write it to the log directly: the
333 * magic number is only written to the log for descripter blocks. In
334 * this case, we copy the data and replace the first word with 0, and we
335 * return a result code which indicates that this buffer needs to be
336 * marked as an escaped buffer in the corresponding log descriptor
337 * block. The missing word can then be restored when the block is read
340 * If the source buffer has already been modified by a new transaction
341 * since we took the last commit snapshot, we use the frozen copy of
342 * that data for IO. If we end up using the existing buffer_head's data
343 * for the write, then we have to make sure nobody modifies it while the
344 * IO is in progress. do_get_write_access() handles this.
346 * The function returns a pointer to the buffer_head to be used for IO.
354 * Bit 0 set == escape performed on the data
355 * Bit 1 set == buffer copy-out performed (kfree the data after IO)
358 int jbd2_journal_write_metadata_buffer(transaction_t *transaction,
359 struct journal_head *jh_in,
360 struct buffer_head **bh_out,
363 int need_copy_out = 0;
364 int done_copy_out = 0;
367 struct buffer_head *new_bh;
368 struct page *new_page;
369 unsigned int new_offset;
370 struct buffer_head *bh_in = jh2bh(jh_in);
371 journal_t *journal = transaction->t_journal;
374 * The buffer really shouldn't be locked: only the current committing
375 * transaction is allowed to write it, so nobody else is allowed
378 * akpm: except if we're journalling data, and write() output is
379 * also part of a shared mapping, and another thread has
380 * decided to launch a writepage() against this buffer.
382 J_ASSERT_BH(bh_in, buffer_jbddirty(bh_in));
384 new_bh = alloc_buffer_head(GFP_NOFS|__GFP_NOFAIL);
386 /* keep subsequent assertions sane */
387 atomic_set(&new_bh->b_count, 1);
389 jbd_lock_bh_state(bh_in);
392 * If a new transaction has already done a buffer copy-out, then
393 * we use that version of the data for the commit.
395 if (jh_in->b_frozen_data) {
397 new_page = virt_to_page(jh_in->b_frozen_data);
398 new_offset = offset_in_page(jh_in->b_frozen_data);
400 new_page = jh2bh(jh_in)->b_page;
401 new_offset = offset_in_page(jh2bh(jh_in)->b_data);
404 mapped_data = kmap_atomic(new_page);
406 * Fire data frozen trigger if data already wasn't frozen. Do this
407 * before checking for escaping, as the trigger may modify the magic
408 * offset. If a copy-out happens afterwards, it will have the correct
409 * data in the buffer.
412 jbd2_buffer_frozen_trigger(jh_in, mapped_data + new_offset,
418 if (*((__be32 *)(mapped_data + new_offset)) ==
419 cpu_to_be32(JBD2_MAGIC_NUMBER)) {
423 kunmap_atomic(mapped_data);
426 * Do we need to do a data copy?
428 if (need_copy_out && !done_copy_out) {
431 jbd_unlock_bh_state(bh_in);
432 tmp = jbd2_alloc(bh_in->b_size, GFP_NOFS);
437 jbd_lock_bh_state(bh_in);
438 if (jh_in->b_frozen_data) {
439 jbd2_free(tmp, bh_in->b_size);
443 jh_in->b_frozen_data = tmp;
444 mapped_data = kmap_atomic(new_page);
445 memcpy(tmp, mapped_data + new_offset, bh_in->b_size);
446 kunmap_atomic(mapped_data);
448 new_page = virt_to_page(tmp);
449 new_offset = offset_in_page(tmp);
453 * This isn't strictly necessary, as we're using frozen
454 * data for the escaping, but it keeps consistency with
455 * b_frozen_data usage.
457 jh_in->b_frozen_triggers = jh_in->b_triggers;
461 * Did we need to do an escaping? Now we've done all the
462 * copying, we can finally do so.
465 mapped_data = kmap_atomic(new_page);
466 *((unsigned int *)(mapped_data + new_offset)) = 0;
467 kunmap_atomic(mapped_data);
470 set_bh_page(new_bh, new_page, new_offset);
471 new_bh->b_size = bh_in->b_size;
472 new_bh->b_bdev = journal->j_dev;
473 new_bh->b_blocknr = blocknr;
474 new_bh->b_private = bh_in;
475 set_buffer_mapped(new_bh);
476 set_buffer_dirty(new_bh);
481 * The to-be-written buffer needs to get moved to the io queue,
482 * and the original buffer whose contents we are shadowing or
483 * copying is moved to the transaction's shadow queue.
485 JBUFFER_TRACE(jh_in, "file as BJ_Shadow");
486 spin_lock(&journal->j_list_lock);
487 __jbd2_journal_file_buffer(jh_in, transaction, BJ_Shadow);
488 spin_unlock(&journal->j_list_lock);
489 set_buffer_shadow(bh_in);
490 jbd_unlock_bh_state(bh_in);
492 return do_escape | (done_copy_out << 1);
496 * Allocation code for the journal file. Manage the space left in the
497 * journal, so that we can begin checkpointing when appropriate.
501 * Called with j_state_lock locked for writing.
502 * Returns true if a transaction commit was started.
504 int __jbd2_log_start_commit(journal_t *journal, tid_t target)
506 /* Return if the txn has already requested to be committed */
507 if (journal->j_commit_request == target)
511 * The only transaction we can possibly wait upon is the
512 * currently running transaction (if it exists). Otherwise,
513 * the target tid must be an old one.
515 if (journal->j_running_transaction &&
516 journal->j_running_transaction->t_tid == target) {
518 * We want a new commit: OK, mark the request and wakeup the
519 * commit thread. We do _not_ do the commit ourselves.
522 journal->j_commit_request = target;
523 jbd_debug(1, "JBD2: requesting commit %d/%d\n",
524 journal->j_commit_request,
525 journal->j_commit_sequence);
526 journal->j_running_transaction->t_requested = jiffies;
527 wake_up(&journal->j_wait_commit);
529 } else if (!tid_geq(journal->j_commit_request, target))
530 /* This should never happen, but if it does, preserve
531 the evidence before kjournald goes into a loop and
532 increments j_commit_sequence beyond all recognition. */
533 WARN_ONCE(1, "JBD2: bad log_start_commit: %u %u %u %u\n",
534 journal->j_commit_request,
535 journal->j_commit_sequence,
536 target, journal->j_running_transaction ?
537 journal->j_running_transaction->t_tid : 0);
541 int jbd2_log_start_commit(journal_t *journal, tid_t tid)
545 write_lock(&journal->j_state_lock);
546 ret = __jbd2_log_start_commit(journal, tid);
547 write_unlock(&journal->j_state_lock);
552 * Force and wait any uncommitted transactions. We can only force the running
553 * transaction if we don't have an active handle, otherwise, we will deadlock.
554 * Returns: <0 in case of error,
555 * 0 if nothing to commit,
556 * 1 if transaction was successfully committed.
558 static int __jbd2_journal_force_commit(journal_t *journal)
560 transaction_t *transaction = NULL;
562 int need_to_start = 0, ret = 0;
564 read_lock(&journal->j_state_lock);
565 if (journal->j_running_transaction && !current->journal_info) {
566 transaction = journal->j_running_transaction;
567 if (!tid_geq(journal->j_commit_request, transaction->t_tid))
569 } else if (journal->j_committing_transaction)
570 transaction = journal->j_committing_transaction;
573 /* Nothing to commit */
574 read_unlock(&journal->j_state_lock);
577 tid = transaction->t_tid;
578 read_unlock(&journal->j_state_lock);
580 jbd2_log_start_commit(journal, tid);
581 ret = jbd2_log_wait_commit(journal, tid);
589 * Force and wait upon a commit if the calling process is not within
590 * transaction. This is used for forcing out undo-protected data which contains
591 * bitmaps, when the fs is running out of space.
593 * @journal: journal to force
594 * Returns true if progress was made.
596 int jbd2_journal_force_commit_nested(journal_t *journal)
600 ret = __jbd2_journal_force_commit(journal);
605 * int journal_force_commit() - force any uncommitted transactions
606 * @journal: journal to force
608 * Caller want unconditional commit. We can only force the running transaction
609 * if we don't have an active handle, otherwise, we will deadlock.
611 int jbd2_journal_force_commit(journal_t *journal)
615 J_ASSERT(!current->journal_info);
616 ret = __jbd2_journal_force_commit(journal);
623 * Start a commit of the current running transaction (if any). Returns true
624 * if a transaction is going to be committed (or is currently already
625 * committing), and fills its tid in at *ptid
627 int jbd2_journal_start_commit(journal_t *journal, tid_t *ptid)
631 write_lock(&journal->j_state_lock);
632 if (journal->j_running_transaction) {
633 tid_t tid = journal->j_running_transaction->t_tid;
635 __jbd2_log_start_commit(journal, tid);
636 /* There's a running transaction and we've just made sure
637 * it's commit has been scheduled. */
641 } else if (journal->j_committing_transaction) {
643 * If commit has been started, then we have to wait for
644 * completion of that transaction.
647 *ptid = journal->j_committing_transaction->t_tid;
650 write_unlock(&journal->j_state_lock);
655 * Return 1 if a given transaction has not yet sent barrier request
656 * connected with a transaction commit. If 0 is returned, transaction
657 * may or may not have sent the barrier. Used to avoid sending barrier
658 * twice in common cases.
660 int jbd2_trans_will_send_data_barrier(journal_t *journal, tid_t tid)
663 transaction_t *commit_trans;
665 if (!(journal->j_flags & JBD2_BARRIER))
667 read_lock(&journal->j_state_lock);
668 /* Transaction already committed? */
669 if (tid_geq(journal->j_commit_sequence, tid))
671 commit_trans = journal->j_committing_transaction;
672 if (!commit_trans || commit_trans->t_tid != tid) {
677 * Transaction is being committed and we already proceeded to
678 * submitting a flush to fs partition?
680 if (journal->j_fs_dev != journal->j_dev) {
681 if (!commit_trans->t_need_data_flush ||
682 commit_trans->t_state >= T_COMMIT_DFLUSH)
685 if (commit_trans->t_state >= T_COMMIT_JFLUSH)
690 read_unlock(&journal->j_state_lock);
693 EXPORT_SYMBOL(jbd2_trans_will_send_data_barrier);
696 * Wait for a specified commit to complete.
697 * The caller may not hold the journal lock.
699 int jbd2_log_wait_commit(journal_t *journal, tid_t tid)
703 jbd2_might_wait_for_commit(journal);
704 read_lock(&journal->j_state_lock);
705 #ifdef CONFIG_JBD2_DEBUG
706 if (!tid_geq(journal->j_commit_request, tid)) {
708 "%s: error: j_commit_request=%d, tid=%d\n",
709 __func__, journal->j_commit_request, tid);
712 while (tid_gt(tid, journal->j_commit_sequence)) {
713 jbd_debug(1, "JBD2: want %d, j_commit_sequence=%d\n",
714 tid, journal->j_commit_sequence);
715 read_unlock(&journal->j_state_lock);
716 wake_up(&journal->j_wait_commit);
717 wait_event(journal->j_wait_done_commit,
718 !tid_gt(tid, journal->j_commit_sequence));
719 read_lock(&journal->j_state_lock);
721 read_unlock(&journal->j_state_lock);
723 if (unlikely(is_journal_aborted(journal)))
729 * When this function returns the transaction corresponding to tid
730 * will be completed. If the transaction has currently running, start
731 * committing that transaction before waiting for it to complete. If
732 * the transaction id is stale, it is by definition already completed,
733 * so just return SUCCESS.
735 int jbd2_complete_transaction(journal_t *journal, tid_t tid)
737 int need_to_wait = 1;
739 read_lock(&journal->j_state_lock);
740 if (journal->j_running_transaction &&
741 journal->j_running_transaction->t_tid == tid) {
742 if (journal->j_commit_request != tid) {
743 /* transaction not yet started, so request it */
744 read_unlock(&journal->j_state_lock);
745 jbd2_log_start_commit(journal, tid);
748 } else if (!(journal->j_committing_transaction &&
749 journal->j_committing_transaction->t_tid == tid))
751 read_unlock(&journal->j_state_lock);
755 return jbd2_log_wait_commit(journal, tid);
757 EXPORT_SYMBOL(jbd2_complete_transaction);
760 * Log buffer allocation routines:
763 int jbd2_journal_next_log_block(journal_t *journal, unsigned long long *retp)
765 unsigned long blocknr;
767 write_lock(&journal->j_state_lock);
768 J_ASSERT(journal->j_free > 1);
770 blocknr = journal->j_head;
773 if (journal->j_head == journal->j_last)
774 journal->j_head = journal->j_first;
775 write_unlock(&journal->j_state_lock);
776 return jbd2_journal_bmap(journal, blocknr, retp);
780 * Conversion of logical to physical block numbers for the journal
782 * On external journals the journal blocks are identity-mapped, so
783 * this is a no-op. If needed, we can use j_blk_offset - everything is
786 int jbd2_journal_bmap(journal_t *journal, unsigned long blocknr,
787 unsigned long long *retp)
790 unsigned long long ret;
792 if (journal->j_inode) {
793 ret = bmap(journal->j_inode, blocknr);
797 printk(KERN_ALERT "%s: journal block not found "
798 "at offset %lu on %s\n",
799 __func__, blocknr, journal->j_devname);
801 __journal_abort_soft(journal, err);
804 *retp = blocknr; /* +journal->j_blk_offset */
810 * We play buffer_head aliasing tricks to write data/metadata blocks to
811 * the journal without copying their contents, but for journal
812 * descriptor blocks we do need to generate bona fide buffers.
814 * After the caller of jbd2_journal_get_descriptor_buffer() has finished modifying
815 * the buffer's contents they really should run flush_dcache_page(bh->b_page).
816 * But we don't bother doing that, so there will be coherency problems with
817 * mmaps of blockdevs which hold live JBD-controlled filesystems.
820 jbd2_journal_get_descriptor_buffer(transaction_t *transaction, int type)
822 journal_t *journal = transaction->t_journal;
823 struct buffer_head *bh;
824 unsigned long long blocknr;
825 journal_header_t *header;
828 err = jbd2_journal_next_log_block(journal, &blocknr);
833 bh = __getblk(journal->j_dev, blocknr, journal->j_blocksize);
837 memset(bh->b_data, 0, journal->j_blocksize);
838 header = (journal_header_t *)bh->b_data;
839 header->h_magic = cpu_to_be32(JBD2_MAGIC_NUMBER);
840 header->h_blocktype = cpu_to_be32(type);
841 header->h_sequence = cpu_to_be32(transaction->t_tid);
842 set_buffer_uptodate(bh);
844 BUFFER_TRACE(bh, "return this buffer");
848 void jbd2_descriptor_block_csum_set(journal_t *j, struct buffer_head *bh)
850 struct jbd2_journal_block_tail *tail;
853 if (!jbd2_journal_has_csum_v2or3(j))
856 tail = (struct jbd2_journal_block_tail *)(bh->b_data + j->j_blocksize -
857 sizeof(struct jbd2_journal_block_tail));
858 tail->t_checksum = 0;
859 csum = jbd2_chksum(j, j->j_csum_seed, bh->b_data, j->j_blocksize);
860 tail->t_checksum = cpu_to_be32(csum);
864 * Return tid of the oldest transaction in the journal and block in the journal
865 * where the transaction starts.
867 * If the journal is now empty, return which will be the next transaction ID
868 * we will write and where will that transaction start.
870 * The return value is 0 if journal tail cannot be pushed any further, 1 if
873 int jbd2_journal_get_log_tail(journal_t *journal, tid_t *tid,
874 unsigned long *block)
876 transaction_t *transaction;
879 read_lock(&journal->j_state_lock);
880 spin_lock(&journal->j_list_lock);
881 transaction = journal->j_checkpoint_transactions;
883 *tid = transaction->t_tid;
884 *block = transaction->t_log_start;
885 } else if ((transaction = journal->j_committing_transaction) != NULL) {
886 *tid = transaction->t_tid;
887 *block = transaction->t_log_start;
888 } else if ((transaction = journal->j_running_transaction) != NULL) {
889 *tid = transaction->t_tid;
890 *block = journal->j_head;
892 *tid = journal->j_transaction_sequence;
893 *block = journal->j_head;
895 ret = tid_gt(*tid, journal->j_tail_sequence);
896 spin_unlock(&journal->j_list_lock);
897 read_unlock(&journal->j_state_lock);
903 * Update information in journal structure and in on disk journal superblock
904 * about log tail. This function does not check whether information passed in
905 * really pushes log tail further. It's responsibility of the caller to make
906 * sure provided log tail information is valid (e.g. by holding
907 * j_checkpoint_mutex all the time between computing log tail and calling this
908 * function as is the case with jbd2_cleanup_journal_tail()).
910 * Requires j_checkpoint_mutex
912 int __jbd2_update_log_tail(journal_t *journal, tid_t tid, unsigned long block)
917 BUG_ON(!mutex_is_locked(&journal->j_checkpoint_mutex));
920 * We cannot afford for write to remain in drive's caches since as
921 * soon as we update j_tail, next transaction can start reusing journal
922 * space and if we lose sb update during power failure we'd replay
923 * old transaction with possibly newly overwritten data.
925 ret = jbd2_journal_update_sb_log_tail(journal, tid, block, REQ_FUA);
929 write_lock(&journal->j_state_lock);
930 freed = block - journal->j_tail;
931 if (block < journal->j_tail)
932 freed += journal->j_last - journal->j_first;
934 trace_jbd2_update_log_tail(journal, tid, block, freed);
936 "Cleaning journal tail from %d to %d (offset %lu), "
938 journal->j_tail_sequence, tid, block, freed);
940 journal->j_free += freed;
941 journal->j_tail_sequence = tid;
942 journal->j_tail = block;
943 write_unlock(&journal->j_state_lock);
950 * This is a variaon of __jbd2_update_log_tail which checks for validity of
951 * provided log tail and locks j_checkpoint_mutex. So it is safe against races
952 * with other threads updating log tail.
954 void jbd2_update_log_tail(journal_t *journal, tid_t tid, unsigned long block)
956 mutex_lock_io(&journal->j_checkpoint_mutex);
957 if (tid_gt(tid, journal->j_tail_sequence))
958 __jbd2_update_log_tail(journal, tid, block);
959 mutex_unlock(&journal->j_checkpoint_mutex);
962 struct jbd2_stats_proc_session {
964 struct transaction_stats_s *stats;
969 static void *jbd2_seq_info_start(struct seq_file *seq, loff_t *pos)
971 return *pos ? NULL : SEQ_START_TOKEN;
974 static void *jbd2_seq_info_next(struct seq_file *seq, void *v, loff_t *pos)
979 static int jbd2_seq_info_show(struct seq_file *seq, void *v)
981 struct jbd2_stats_proc_session *s = seq->private;
983 if (v != SEQ_START_TOKEN)
985 seq_printf(seq, "%lu transactions (%lu requested), "
986 "each up to %u blocks\n",
987 s->stats->ts_tid, s->stats->ts_requested,
988 s->journal->j_max_transaction_buffers);
989 if (s->stats->ts_tid == 0)
991 seq_printf(seq, "average: \n %ums waiting for transaction\n",
992 jiffies_to_msecs(s->stats->run.rs_wait / s->stats->ts_tid));
993 seq_printf(seq, " %ums request delay\n",
994 (s->stats->ts_requested == 0) ? 0 :
995 jiffies_to_msecs(s->stats->run.rs_request_delay /
996 s->stats->ts_requested));
997 seq_printf(seq, " %ums running transaction\n",
998 jiffies_to_msecs(s->stats->run.rs_running / s->stats->ts_tid));
999 seq_printf(seq, " %ums transaction was being locked\n",
1000 jiffies_to_msecs(s->stats->run.rs_locked / s->stats->ts_tid));
1001 seq_printf(seq, " %ums flushing data (in ordered mode)\n",
1002 jiffies_to_msecs(s->stats->run.rs_flushing / s->stats->ts_tid));
1003 seq_printf(seq, " %ums logging transaction\n",
1004 jiffies_to_msecs(s->stats->run.rs_logging / s->stats->ts_tid));
1005 seq_printf(seq, " %lluus average transaction commit time\n",
1006 div_u64(s->journal->j_average_commit_time, 1000));
1007 seq_printf(seq, " %lu handles per transaction\n",
1008 s->stats->run.rs_handle_count / s->stats->ts_tid);
1009 seq_printf(seq, " %lu blocks per transaction\n",
1010 s->stats->run.rs_blocks / s->stats->ts_tid);
1011 seq_printf(seq, " %lu logged blocks per transaction\n",
1012 s->stats->run.rs_blocks_logged / s->stats->ts_tid);
1016 static void jbd2_seq_info_stop(struct seq_file *seq, void *v)
1020 static const struct seq_operations jbd2_seq_info_ops = {
1021 .start = jbd2_seq_info_start,
1022 .next = jbd2_seq_info_next,
1023 .stop = jbd2_seq_info_stop,
1024 .show = jbd2_seq_info_show,
1027 static int jbd2_seq_info_open(struct inode *inode, struct file *file)
1029 journal_t *journal = PDE_DATA(inode);
1030 struct jbd2_stats_proc_session *s;
1033 s = kmalloc(sizeof(*s), GFP_KERNEL);
1036 size = sizeof(struct transaction_stats_s);
1037 s->stats = kmalloc(size, GFP_KERNEL);
1038 if (s->stats == NULL) {
1042 spin_lock(&journal->j_history_lock);
1043 memcpy(s->stats, &journal->j_stats, size);
1044 s->journal = journal;
1045 spin_unlock(&journal->j_history_lock);
1047 rc = seq_open(file, &jbd2_seq_info_ops);
1049 struct seq_file *m = file->private_data;
1059 static int jbd2_seq_info_release(struct inode *inode, struct file *file)
1061 struct seq_file *seq = file->private_data;
1062 struct jbd2_stats_proc_session *s = seq->private;
1065 return seq_release(inode, file);
1068 static const struct file_operations jbd2_seq_info_fops = {
1069 .owner = THIS_MODULE,
1070 .open = jbd2_seq_info_open,
1072 .llseek = seq_lseek,
1073 .release = jbd2_seq_info_release,
1076 static struct proc_dir_entry *proc_jbd2_stats;
1078 static void jbd2_stats_proc_init(journal_t *journal)
1080 journal->j_proc_entry = proc_mkdir(journal->j_devname, proc_jbd2_stats);
1081 if (journal->j_proc_entry) {
1082 proc_create_data("info", S_IRUGO, journal->j_proc_entry,
1083 &jbd2_seq_info_fops, journal);
1087 static void jbd2_stats_proc_exit(journal_t *journal)
1089 remove_proc_entry("info", journal->j_proc_entry);
1090 remove_proc_entry(journal->j_devname, proc_jbd2_stats);
1094 * Management for journal control blocks: functions to create and
1095 * destroy journal_t structures, and to initialise and read existing
1096 * journal blocks from disk. */
1098 /* First: create and setup a journal_t object in memory. We initialise
1099 * very few fields yet: that has to wait until we have created the
1100 * journal structures from from scratch, or loaded them from disk. */
1102 static journal_t *journal_init_common(struct block_device *bdev,
1103 struct block_device *fs_dev,
1104 unsigned long long start, int len, int blocksize)
1106 static struct lock_class_key jbd2_trans_commit_key;
1109 struct buffer_head *bh;
1112 journal = kzalloc(sizeof(*journal), GFP_KERNEL);
1116 init_waitqueue_head(&journal->j_wait_transaction_locked);
1117 init_waitqueue_head(&journal->j_wait_done_commit);
1118 init_waitqueue_head(&journal->j_wait_commit);
1119 init_waitqueue_head(&journal->j_wait_updates);
1120 init_waitqueue_head(&journal->j_wait_reserved);
1121 mutex_init(&journal->j_barrier);
1122 mutex_init(&journal->j_checkpoint_mutex);
1123 spin_lock_init(&journal->j_revoke_lock);
1124 spin_lock_init(&journal->j_list_lock);
1125 rwlock_init(&journal->j_state_lock);
1127 journal->j_commit_interval = (HZ * JBD2_DEFAULT_MAX_COMMIT_AGE);
1128 journal->j_min_batch_time = 0;
1129 journal->j_max_batch_time = 15000; /* 15ms */
1130 atomic_set(&journal->j_reserved_credits, 0);
1132 /* The journal is marked for error until we succeed with recovery! */
1133 journal->j_flags = JBD2_ABORT;
1135 /* Set up a default-sized revoke table for the new mount. */
1136 err = jbd2_journal_init_revoke(journal, JOURNAL_REVOKE_DEFAULT_HASH);
1140 spin_lock_init(&journal->j_history_lock);
1142 lockdep_init_map(&journal->j_trans_commit_map, "jbd2_handle",
1143 &jbd2_trans_commit_key, 0);
1145 /* journal descriptor can store up to n blocks -bzzz */
1146 journal->j_blocksize = blocksize;
1147 journal->j_dev = bdev;
1148 journal->j_fs_dev = fs_dev;
1149 journal->j_blk_offset = start;
1150 journal->j_maxlen = len;
1151 n = journal->j_blocksize / sizeof(journal_block_tag_t);
1152 journal->j_wbufsize = n;
1153 journal->j_wbuf = kmalloc_array(n, sizeof(struct buffer_head *),
1155 if (!journal->j_wbuf)
1158 bh = getblk_unmovable(journal->j_dev, start, journal->j_blocksize);
1160 pr_err("%s: Cannot get buffer for journal superblock\n",
1164 journal->j_sb_buffer = bh;
1165 journal->j_superblock = (journal_superblock_t *)bh->b_data;
1170 kfree(journal->j_wbuf);
1171 jbd2_journal_destroy_revoke(journal);
1176 /* jbd2_journal_init_dev and jbd2_journal_init_inode:
1178 * Create a journal structure assigned some fixed set of disk blocks to
1179 * the journal. We don't actually touch those disk blocks yet, but we
1180 * need to set up all of the mapping information to tell the journaling
1181 * system where the journal blocks are.
1186 * journal_t * jbd2_journal_init_dev() - creates and initialises a journal structure
1187 * @bdev: Block device on which to create the journal
1188 * @fs_dev: Device which hold journalled filesystem for this journal.
1189 * @start: Block nr Start of journal.
1190 * @len: Length of the journal in blocks.
1191 * @blocksize: blocksize of journalling device
1193 * Returns: a newly created journal_t *
1195 * jbd2_journal_init_dev creates a journal which maps a fixed contiguous
1196 * range of blocks on an arbitrary block device.
1199 journal_t *jbd2_journal_init_dev(struct block_device *bdev,
1200 struct block_device *fs_dev,
1201 unsigned long long start, int len, int blocksize)
1205 journal = journal_init_common(bdev, fs_dev, start, len, blocksize);
1209 bdevname(journal->j_dev, journal->j_devname);
1210 strreplace(journal->j_devname, '/', '!');
1211 jbd2_stats_proc_init(journal);
1217 * journal_t * jbd2_journal_init_inode () - creates a journal which maps to a inode.
1218 * @inode: An inode to create the journal in
1220 * jbd2_journal_init_inode creates a journal which maps an on-disk inode as
1221 * the journal. The inode must exist already, must support bmap() and
1222 * must have all data blocks preallocated.
1224 journal_t *jbd2_journal_init_inode(struct inode *inode)
1228 unsigned long long blocknr;
1230 blocknr = bmap(inode, 0);
1232 pr_err("%s: Cannot locate journal superblock\n",
1237 jbd_debug(1, "JBD2: inode %s/%ld, size %lld, bits %d, blksize %ld\n",
1238 inode->i_sb->s_id, inode->i_ino, (long long) inode->i_size,
1239 inode->i_sb->s_blocksize_bits, inode->i_sb->s_blocksize);
1241 journal = journal_init_common(inode->i_sb->s_bdev, inode->i_sb->s_bdev,
1242 blocknr, inode->i_size >> inode->i_sb->s_blocksize_bits,
1243 inode->i_sb->s_blocksize);
1247 journal->j_inode = inode;
1248 bdevname(journal->j_dev, journal->j_devname);
1249 p = strreplace(journal->j_devname, '/', '!');
1250 sprintf(p, "-%lu", journal->j_inode->i_ino);
1251 jbd2_stats_proc_init(journal);
1257 * If the journal init or create aborts, we need to mark the journal
1258 * superblock as being NULL to prevent the journal destroy from writing
1259 * back a bogus superblock.
1261 static void journal_fail_superblock (journal_t *journal)
1263 struct buffer_head *bh = journal->j_sb_buffer;
1265 journal->j_sb_buffer = NULL;
1269 * Given a journal_t structure, initialise the various fields for
1270 * startup of a new journaling session. We use this both when creating
1271 * a journal, and after recovering an old journal to reset it for
1275 static int journal_reset(journal_t *journal)
1277 journal_superblock_t *sb = journal->j_superblock;
1278 unsigned long long first, last;
1280 first = be32_to_cpu(sb->s_first);
1281 last = be32_to_cpu(sb->s_maxlen);
1282 if (first + JBD2_MIN_JOURNAL_BLOCKS > last + 1) {
1283 printk(KERN_ERR "JBD2: Journal too short (blocks %llu-%llu).\n",
1285 journal_fail_superblock(journal);
1289 journal->j_first = first;
1290 journal->j_last = last;
1292 journal->j_head = first;
1293 journal->j_tail = first;
1294 journal->j_free = last - first;
1296 journal->j_tail_sequence = journal->j_transaction_sequence;
1297 journal->j_commit_sequence = journal->j_transaction_sequence - 1;
1298 journal->j_commit_request = journal->j_commit_sequence;
1300 journal->j_max_transaction_buffers = journal->j_maxlen / 4;
1303 * As a special case, if the on-disk copy is already marked as needing
1304 * no recovery (s_start == 0), then we can safely defer the superblock
1305 * update until the next commit by setting JBD2_FLUSHED. This avoids
1306 * attempting a write to a potential-readonly device.
1308 if (sb->s_start == 0) {
1309 jbd_debug(1, "JBD2: Skipping superblock update on recovered sb "
1310 "(start %ld, seq %d, errno %d)\n",
1311 journal->j_tail, journal->j_tail_sequence,
1313 journal->j_flags |= JBD2_FLUSHED;
1315 /* Lock here to make assertions happy... */
1316 mutex_lock_io(&journal->j_checkpoint_mutex);
1318 * Update log tail information. We use REQ_FUA since new
1319 * transaction will start reusing journal space and so we
1320 * must make sure information about current log tail is on
1323 jbd2_journal_update_sb_log_tail(journal,
1324 journal->j_tail_sequence,
1327 mutex_unlock(&journal->j_checkpoint_mutex);
1329 return jbd2_journal_start_thread(journal);
1332 static int jbd2_write_superblock(journal_t *journal, int write_flags)
1334 struct buffer_head *bh = journal->j_sb_buffer;
1335 journal_superblock_t *sb = journal->j_superblock;
1338 trace_jbd2_write_superblock(journal, write_flags);
1339 if (!(journal->j_flags & JBD2_BARRIER))
1340 write_flags &= ~(REQ_FUA | REQ_PREFLUSH);
1342 if (buffer_write_io_error(bh)) {
1344 * Oh, dear. A previous attempt to write the journal
1345 * superblock failed. This could happen because the
1346 * USB device was yanked out. Or it could happen to
1347 * be a transient write error and maybe the block will
1348 * be remapped. Nothing we can do but to retry the
1349 * write and hope for the best.
1351 printk(KERN_ERR "JBD2: previous I/O error detected "
1352 "for journal superblock update for %s.\n",
1353 journal->j_devname);
1354 clear_buffer_write_io_error(bh);
1355 set_buffer_uptodate(bh);
1357 jbd2_superblock_csum_set(journal, sb);
1359 bh->b_end_io = end_buffer_write_sync;
1360 ret = submit_bh(REQ_OP_WRITE, write_flags, bh);
1362 if (buffer_write_io_error(bh)) {
1363 clear_buffer_write_io_error(bh);
1364 set_buffer_uptodate(bh);
1368 printk(KERN_ERR "JBD2: Error %d detected when updating "
1369 "journal superblock for %s.\n", ret,
1370 journal->j_devname);
1371 jbd2_journal_abort(journal, ret);
1378 * jbd2_journal_update_sb_log_tail() - Update log tail in journal sb on disk.
1379 * @journal: The journal to update.
1380 * @tail_tid: TID of the new transaction at the tail of the log
1381 * @tail_block: The first block of the transaction at the tail of the log
1382 * @write_op: With which operation should we write the journal sb
1384 * Update a journal's superblock information about log tail and write it to
1385 * disk, waiting for the IO to complete.
1387 int jbd2_journal_update_sb_log_tail(journal_t *journal, tid_t tail_tid,
1388 unsigned long tail_block, int write_op)
1390 journal_superblock_t *sb = journal->j_superblock;
1393 BUG_ON(!mutex_is_locked(&journal->j_checkpoint_mutex));
1394 jbd_debug(1, "JBD2: updating superblock (start %lu, seq %u)\n",
1395 tail_block, tail_tid);
1397 sb->s_sequence = cpu_to_be32(tail_tid);
1398 sb->s_start = cpu_to_be32(tail_block);
1400 ret = jbd2_write_superblock(journal, write_op);
1404 /* Log is no longer empty */
1405 write_lock(&journal->j_state_lock);
1406 WARN_ON(!sb->s_sequence);
1407 journal->j_flags &= ~JBD2_FLUSHED;
1408 write_unlock(&journal->j_state_lock);
1415 * jbd2_mark_journal_empty() - Mark on disk journal as empty.
1416 * @journal: The journal to update.
1417 * @write_op: With which operation should we write the journal sb
1419 * Update a journal's dynamic superblock fields to show that journal is empty.
1420 * Write updated superblock to disk waiting for IO to complete.
1422 static void jbd2_mark_journal_empty(journal_t *journal, int write_op)
1424 journal_superblock_t *sb = journal->j_superblock;
1426 BUG_ON(!mutex_is_locked(&journal->j_checkpoint_mutex));
1427 read_lock(&journal->j_state_lock);
1428 /* Is it already empty? */
1429 if (sb->s_start == 0) {
1430 read_unlock(&journal->j_state_lock);
1433 jbd_debug(1, "JBD2: Marking journal as empty (seq %d)\n",
1434 journal->j_tail_sequence);
1436 sb->s_sequence = cpu_to_be32(journal->j_tail_sequence);
1437 sb->s_start = cpu_to_be32(0);
1438 read_unlock(&journal->j_state_lock);
1440 jbd2_write_superblock(journal, write_op);
1442 /* Log is no longer empty */
1443 write_lock(&journal->j_state_lock);
1444 journal->j_flags |= JBD2_FLUSHED;
1445 write_unlock(&journal->j_state_lock);
1450 * jbd2_journal_update_sb_errno() - Update error in the journal.
1451 * @journal: The journal to update.
1453 * Update a journal's errno. Write updated superblock to disk waiting for IO
1456 void jbd2_journal_update_sb_errno(journal_t *journal)
1458 journal_superblock_t *sb = journal->j_superblock;
1460 read_lock(&journal->j_state_lock);
1461 jbd_debug(1, "JBD2: updating superblock error (errno %d)\n",
1463 sb->s_errno = cpu_to_be32(journal->j_errno);
1464 read_unlock(&journal->j_state_lock);
1466 jbd2_write_superblock(journal, REQ_FUA);
1468 EXPORT_SYMBOL(jbd2_journal_update_sb_errno);
1471 * Read the superblock for a given journal, performing initial
1472 * validation of the format.
1474 static int journal_get_superblock(journal_t *journal)
1476 struct buffer_head *bh;
1477 journal_superblock_t *sb;
1480 bh = journal->j_sb_buffer;
1482 J_ASSERT(bh != NULL);
1483 if (!buffer_uptodate(bh)) {
1484 ll_rw_block(REQ_OP_READ, 0, 1, &bh);
1486 if (!buffer_uptodate(bh)) {
1488 "JBD2: IO error reading journal superblock\n");
1493 if (buffer_verified(bh))
1496 sb = journal->j_superblock;
1500 if (sb->s_header.h_magic != cpu_to_be32(JBD2_MAGIC_NUMBER) ||
1501 sb->s_blocksize != cpu_to_be32(journal->j_blocksize)) {
1502 printk(KERN_WARNING "JBD2: no valid journal superblock found\n");
1506 switch(be32_to_cpu(sb->s_header.h_blocktype)) {
1507 case JBD2_SUPERBLOCK_V1:
1508 journal->j_format_version = 1;
1510 case JBD2_SUPERBLOCK_V2:
1511 journal->j_format_version = 2;
1514 printk(KERN_WARNING "JBD2: unrecognised superblock format ID\n");
1518 if (be32_to_cpu(sb->s_maxlen) < journal->j_maxlen)
1519 journal->j_maxlen = be32_to_cpu(sb->s_maxlen);
1520 else if (be32_to_cpu(sb->s_maxlen) > journal->j_maxlen) {
1521 printk(KERN_WARNING "JBD2: journal file too short\n");
1525 if (be32_to_cpu(sb->s_first) == 0 ||
1526 be32_to_cpu(sb->s_first) >= journal->j_maxlen) {
1528 "JBD2: Invalid start block of journal: %u\n",
1529 be32_to_cpu(sb->s_first));
1533 if (jbd2_has_feature_csum2(journal) &&
1534 jbd2_has_feature_csum3(journal)) {
1535 /* Can't have checksum v2 and v3 at the same time! */
1536 printk(KERN_ERR "JBD2: Can't enable checksumming v2 and v3 "
1537 "at the same time!\n");
1541 if (jbd2_journal_has_csum_v2or3_feature(journal) &&
1542 jbd2_has_feature_checksum(journal)) {
1543 /* Can't have checksum v1 and v2 on at the same time! */
1544 printk(KERN_ERR "JBD2: Can't enable checksumming v1 and v2/3 "
1545 "at the same time!\n");
1549 if (!jbd2_verify_csum_type(journal, sb)) {
1550 printk(KERN_ERR "JBD2: Unknown checksum type\n");
1554 /* Load the checksum driver */
1555 if (jbd2_journal_has_csum_v2or3_feature(journal)) {
1556 journal->j_chksum_driver = crypto_alloc_shash("crc32c", 0, 0);
1557 if (IS_ERR(journal->j_chksum_driver)) {
1558 printk(KERN_ERR "JBD2: Cannot load crc32c driver.\n");
1559 err = PTR_ERR(journal->j_chksum_driver);
1560 journal->j_chksum_driver = NULL;
1565 /* Check superblock checksum */
1566 if (!jbd2_superblock_csum_verify(journal, sb)) {
1567 printk(KERN_ERR "JBD2: journal checksum error\n");
1572 /* Precompute checksum seed for all metadata */
1573 if (jbd2_journal_has_csum_v2or3(journal))
1574 journal->j_csum_seed = jbd2_chksum(journal, ~0, sb->s_uuid,
1575 sizeof(sb->s_uuid));
1577 set_buffer_verified(bh);
1582 journal_fail_superblock(journal);
1587 * Load the on-disk journal superblock and read the key fields into the
1591 static int load_superblock(journal_t *journal)
1594 journal_superblock_t *sb;
1596 err = journal_get_superblock(journal);
1600 sb = journal->j_superblock;
1602 journal->j_tail_sequence = be32_to_cpu(sb->s_sequence);
1603 journal->j_tail = be32_to_cpu(sb->s_start);
1604 journal->j_first = be32_to_cpu(sb->s_first);
1605 journal->j_last = be32_to_cpu(sb->s_maxlen);
1606 journal->j_errno = be32_to_cpu(sb->s_errno);
1613 * int jbd2_journal_load() - Read journal from disk.
1614 * @journal: Journal to act on.
1616 * Given a journal_t structure which tells us which disk blocks contain
1617 * a journal, read the journal from disk to initialise the in-memory
1620 int jbd2_journal_load(journal_t *journal)
1623 journal_superblock_t *sb;
1625 err = load_superblock(journal);
1629 sb = journal->j_superblock;
1630 /* If this is a V2 superblock, then we have to check the
1631 * features flags on it. */
1633 if (journal->j_format_version >= 2) {
1634 if ((sb->s_feature_ro_compat &
1635 ~cpu_to_be32(JBD2_KNOWN_ROCOMPAT_FEATURES)) ||
1636 (sb->s_feature_incompat &
1637 ~cpu_to_be32(JBD2_KNOWN_INCOMPAT_FEATURES))) {
1639 "JBD2: Unrecognised features on journal\n");
1645 * Create a slab for this blocksize
1647 err = jbd2_journal_create_slab(be32_to_cpu(sb->s_blocksize));
1651 /* Let the recovery code check whether it needs to recover any
1652 * data from the journal. */
1653 if (jbd2_journal_recover(journal))
1654 goto recovery_error;
1656 if (journal->j_failed_commit) {
1657 printk(KERN_ERR "JBD2: journal transaction %u on %s "
1658 "is corrupt.\n", journal->j_failed_commit,
1659 journal->j_devname);
1660 return -EFSCORRUPTED;
1663 /* OK, we've finished with the dynamic journal bits:
1664 * reinitialise the dynamic contents of the superblock in memory
1665 * and reset them on disk. */
1666 if (journal_reset(journal))
1667 goto recovery_error;
1669 journal->j_flags &= ~JBD2_ABORT;
1670 journal->j_flags |= JBD2_LOADED;
1674 printk(KERN_WARNING "JBD2: recovery failed\n");
1679 * void jbd2_journal_destroy() - Release a journal_t structure.
1680 * @journal: Journal to act on.
1682 * Release a journal_t structure once it is no longer in use by the
1684 * Return <0 if we couldn't clean up the journal.
1686 int jbd2_journal_destroy(journal_t *journal)
1690 /* Wait for the commit thread to wake up and die. */
1691 journal_kill_thread(journal);
1693 /* Force a final log commit */
1694 if (journal->j_running_transaction)
1695 jbd2_journal_commit_transaction(journal);
1697 /* Force any old transactions to disk */
1699 /* Totally anal locking here... */
1700 spin_lock(&journal->j_list_lock);
1701 while (journal->j_checkpoint_transactions != NULL) {
1702 spin_unlock(&journal->j_list_lock);
1703 mutex_lock_io(&journal->j_checkpoint_mutex);
1704 err = jbd2_log_do_checkpoint(journal);
1705 mutex_unlock(&journal->j_checkpoint_mutex);
1707 * If checkpointing failed, just free the buffers to avoid
1711 jbd2_journal_destroy_checkpoint(journal);
1712 spin_lock(&journal->j_list_lock);
1715 spin_lock(&journal->j_list_lock);
1718 J_ASSERT(journal->j_running_transaction == NULL);
1719 J_ASSERT(journal->j_committing_transaction == NULL);
1720 J_ASSERT(journal->j_checkpoint_transactions == NULL);
1721 spin_unlock(&journal->j_list_lock);
1723 if (journal->j_sb_buffer) {
1724 if (!is_journal_aborted(journal)) {
1725 mutex_lock_io(&journal->j_checkpoint_mutex);
1727 write_lock(&journal->j_state_lock);
1728 journal->j_tail_sequence =
1729 ++journal->j_transaction_sequence;
1730 write_unlock(&journal->j_state_lock);
1732 jbd2_mark_journal_empty(journal,
1733 REQ_PREFLUSH | REQ_FUA);
1734 mutex_unlock(&journal->j_checkpoint_mutex);
1737 brelse(journal->j_sb_buffer);
1740 if (journal->j_proc_entry)
1741 jbd2_stats_proc_exit(journal);
1742 iput(journal->j_inode);
1743 if (journal->j_revoke)
1744 jbd2_journal_destroy_revoke(journal);
1745 if (journal->j_chksum_driver)
1746 crypto_free_shash(journal->j_chksum_driver);
1747 kfree(journal->j_wbuf);
1755 *int jbd2_journal_check_used_features () - Check if features specified are used.
1756 * @journal: Journal to check.
1757 * @compat: bitmask of compatible features
1758 * @ro: bitmask of features that force read-only mount
1759 * @incompat: bitmask of incompatible features
1761 * Check whether the journal uses all of a given set of
1762 * features. Return true (non-zero) if it does.
1765 int jbd2_journal_check_used_features (journal_t *journal, unsigned long compat,
1766 unsigned long ro, unsigned long incompat)
1768 journal_superblock_t *sb;
1770 if (!compat && !ro && !incompat)
1772 /* Load journal superblock if it is not loaded yet. */
1773 if (journal->j_format_version == 0 &&
1774 journal_get_superblock(journal) != 0)
1776 if (journal->j_format_version == 1)
1779 sb = journal->j_superblock;
1781 if (((be32_to_cpu(sb->s_feature_compat) & compat) == compat) &&
1782 ((be32_to_cpu(sb->s_feature_ro_compat) & ro) == ro) &&
1783 ((be32_to_cpu(sb->s_feature_incompat) & incompat) == incompat))
1790 * int jbd2_journal_check_available_features() - Check feature set in journalling layer
1791 * @journal: Journal to check.
1792 * @compat: bitmask of compatible features
1793 * @ro: bitmask of features that force read-only mount
1794 * @incompat: bitmask of incompatible features
1796 * Check whether the journaling code supports the use of
1797 * all of a given set of features on this journal. Return true
1798 * (non-zero) if it can. */
1800 int jbd2_journal_check_available_features (journal_t *journal, unsigned long compat,
1801 unsigned long ro, unsigned long incompat)
1803 if (!compat && !ro && !incompat)
1806 /* We can support any known requested features iff the
1807 * superblock is in version 2. Otherwise we fail to support any
1808 * extended sb features. */
1810 if (journal->j_format_version != 2)
1813 if ((compat & JBD2_KNOWN_COMPAT_FEATURES) == compat &&
1814 (ro & JBD2_KNOWN_ROCOMPAT_FEATURES) == ro &&
1815 (incompat & JBD2_KNOWN_INCOMPAT_FEATURES) == incompat)
1822 * int jbd2_journal_set_features () - Mark a given journal feature in the superblock
1823 * @journal: Journal to act on.
1824 * @compat: bitmask of compatible features
1825 * @ro: bitmask of features that force read-only mount
1826 * @incompat: bitmask of incompatible features
1828 * Mark a given journal feature as present on the
1829 * superblock. Returns true if the requested features could be set.
1833 int jbd2_journal_set_features (journal_t *journal, unsigned long compat,
1834 unsigned long ro, unsigned long incompat)
1836 #define INCOMPAT_FEATURE_ON(f) \
1837 ((incompat & (f)) && !(sb->s_feature_incompat & cpu_to_be32(f)))
1838 #define COMPAT_FEATURE_ON(f) \
1839 ((compat & (f)) && !(sb->s_feature_compat & cpu_to_be32(f)))
1840 journal_superblock_t *sb;
1842 if (jbd2_journal_check_used_features(journal, compat, ro, incompat))
1845 if (!jbd2_journal_check_available_features(journal, compat, ro, incompat))
1848 /* If enabling v2 checksums, turn on v3 instead */
1849 if (incompat & JBD2_FEATURE_INCOMPAT_CSUM_V2) {
1850 incompat &= ~JBD2_FEATURE_INCOMPAT_CSUM_V2;
1851 incompat |= JBD2_FEATURE_INCOMPAT_CSUM_V3;
1854 /* Asking for checksumming v3 and v1? Only give them v3. */
1855 if (incompat & JBD2_FEATURE_INCOMPAT_CSUM_V3 &&
1856 compat & JBD2_FEATURE_COMPAT_CHECKSUM)
1857 compat &= ~JBD2_FEATURE_COMPAT_CHECKSUM;
1859 jbd_debug(1, "Setting new features 0x%lx/0x%lx/0x%lx\n",
1860 compat, ro, incompat);
1862 sb = journal->j_superblock;
1864 /* If enabling v3 checksums, update superblock */
1865 if (INCOMPAT_FEATURE_ON(JBD2_FEATURE_INCOMPAT_CSUM_V3)) {
1866 sb->s_checksum_type = JBD2_CRC32C_CHKSUM;
1867 sb->s_feature_compat &=
1868 ~cpu_to_be32(JBD2_FEATURE_COMPAT_CHECKSUM);
1870 /* Load the checksum driver */
1871 if (journal->j_chksum_driver == NULL) {
1872 journal->j_chksum_driver = crypto_alloc_shash("crc32c",
1874 if (IS_ERR(journal->j_chksum_driver)) {
1875 printk(KERN_ERR "JBD2: Cannot load crc32c "
1877 journal->j_chksum_driver = NULL;
1881 /* Precompute checksum seed for all metadata */
1882 journal->j_csum_seed = jbd2_chksum(journal, ~0,
1884 sizeof(sb->s_uuid));
1888 /* If enabling v1 checksums, downgrade superblock */
1889 if (COMPAT_FEATURE_ON(JBD2_FEATURE_COMPAT_CHECKSUM))
1890 sb->s_feature_incompat &=
1891 ~cpu_to_be32(JBD2_FEATURE_INCOMPAT_CSUM_V2 |
1892 JBD2_FEATURE_INCOMPAT_CSUM_V3);
1894 sb->s_feature_compat |= cpu_to_be32(compat);
1895 sb->s_feature_ro_compat |= cpu_to_be32(ro);
1896 sb->s_feature_incompat |= cpu_to_be32(incompat);
1899 #undef COMPAT_FEATURE_ON
1900 #undef INCOMPAT_FEATURE_ON
1904 * jbd2_journal_clear_features () - Clear a given journal feature in the
1906 * @journal: Journal to act on.
1907 * @compat: bitmask of compatible features
1908 * @ro: bitmask of features that force read-only mount
1909 * @incompat: bitmask of incompatible features
1911 * Clear a given journal feature as present on the
1914 void jbd2_journal_clear_features(journal_t *journal, unsigned long compat,
1915 unsigned long ro, unsigned long incompat)
1917 journal_superblock_t *sb;
1919 jbd_debug(1, "Clear features 0x%lx/0x%lx/0x%lx\n",
1920 compat, ro, incompat);
1922 sb = journal->j_superblock;
1924 sb->s_feature_compat &= ~cpu_to_be32(compat);
1925 sb->s_feature_ro_compat &= ~cpu_to_be32(ro);
1926 sb->s_feature_incompat &= ~cpu_to_be32(incompat);
1928 EXPORT_SYMBOL(jbd2_journal_clear_features);
1931 * int jbd2_journal_flush () - Flush journal
1932 * @journal: Journal to act on.
1934 * Flush all data for a given journal to disk and empty the journal.
1935 * Filesystems can use this when remounting readonly to ensure that
1936 * recovery does not need to happen on remount.
1939 int jbd2_journal_flush(journal_t *journal)
1942 transaction_t *transaction = NULL;
1944 write_lock(&journal->j_state_lock);
1946 /* Force everything buffered to the log... */
1947 if (journal->j_running_transaction) {
1948 transaction = journal->j_running_transaction;
1949 __jbd2_log_start_commit(journal, transaction->t_tid);
1950 } else if (journal->j_committing_transaction)
1951 transaction = journal->j_committing_transaction;
1953 /* Wait for the log commit to complete... */
1955 tid_t tid = transaction->t_tid;
1957 write_unlock(&journal->j_state_lock);
1958 jbd2_log_wait_commit(journal, tid);
1960 write_unlock(&journal->j_state_lock);
1963 /* ...and flush everything in the log out to disk. */
1964 spin_lock(&journal->j_list_lock);
1965 while (!err && journal->j_checkpoint_transactions != NULL) {
1966 spin_unlock(&journal->j_list_lock);
1967 mutex_lock_io(&journal->j_checkpoint_mutex);
1968 err = jbd2_log_do_checkpoint(journal);
1969 mutex_unlock(&journal->j_checkpoint_mutex);
1970 spin_lock(&journal->j_list_lock);
1972 spin_unlock(&journal->j_list_lock);
1974 if (is_journal_aborted(journal))
1977 mutex_lock_io(&journal->j_checkpoint_mutex);
1979 err = jbd2_cleanup_journal_tail(journal);
1981 mutex_unlock(&journal->j_checkpoint_mutex);
1987 /* Finally, mark the journal as really needing no recovery.
1988 * This sets s_start==0 in the underlying superblock, which is
1989 * the magic code for a fully-recovered superblock. Any future
1990 * commits of data to the journal will restore the current
1992 jbd2_mark_journal_empty(journal, REQ_FUA);
1993 mutex_unlock(&journal->j_checkpoint_mutex);
1994 write_lock(&journal->j_state_lock);
1995 J_ASSERT(!journal->j_running_transaction);
1996 J_ASSERT(!journal->j_committing_transaction);
1997 J_ASSERT(!journal->j_checkpoint_transactions);
1998 J_ASSERT(journal->j_head == journal->j_tail);
1999 J_ASSERT(journal->j_tail_sequence == journal->j_transaction_sequence);
2000 write_unlock(&journal->j_state_lock);
2006 * int jbd2_journal_wipe() - Wipe journal contents
2007 * @journal: Journal to act on.
2008 * @write: flag (see below)
2010 * Wipe out all of the contents of a journal, safely. This will produce
2011 * a warning if the journal contains any valid recovery information.
2012 * Must be called between journal_init_*() and jbd2_journal_load().
2014 * If 'write' is non-zero, then we wipe out the journal on disk; otherwise
2015 * we merely suppress recovery.
2018 int jbd2_journal_wipe(journal_t *journal, int write)
2022 J_ASSERT (!(journal->j_flags & JBD2_LOADED));
2024 err = load_superblock(journal);
2028 if (!journal->j_tail)
2031 printk(KERN_WARNING "JBD2: %s recovery information on journal\n",
2032 write ? "Clearing" : "Ignoring");
2034 err = jbd2_journal_skip_recovery(journal);
2036 /* Lock to make assertions happy... */
2037 mutex_lock(&journal->j_checkpoint_mutex);
2038 jbd2_mark_journal_empty(journal, REQ_FUA);
2039 mutex_unlock(&journal->j_checkpoint_mutex);
2047 * Journal abort has very specific semantics, which we describe
2048 * for journal abort.
2050 * Two internal functions, which provide abort to the jbd layer
2055 * Quick version for internal journal use (doesn't lock the journal).
2056 * Aborts hard --- we mark the abort as occurred, but do _nothing_ else,
2057 * and don't attempt to make any other journal updates.
2059 void __jbd2_journal_abort_hard(journal_t *journal)
2061 transaction_t *transaction;
2063 if (journal->j_flags & JBD2_ABORT)
2066 printk(KERN_ERR "Aborting journal on device %s.\n",
2067 journal->j_devname);
2069 write_lock(&journal->j_state_lock);
2070 journal->j_flags |= JBD2_ABORT;
2071 transaction = journal->j_running_transaction;
2073 __jbd2_log_start_commit(journal, transaction->t_tid);
2074 write_unlock(&journal->j_state_lock);
2077 /* Soft abort: record the abort error status in the journal superblock,
2078 * but don't do any other IO. */
2079 static void __journal_abort_soft (journal_t *journal, int errno)
2081 if (journal->j_flags & JBD2_ABORT)
2084 if (!journal->j_errno)
2085 journal->j_errno = errno;
2087 __jbd2_journal_abort_hard(journal);
2090 jbd2_journal_update_sb_errno(journal);
2091 write_lock(&journal->j_state_lock);
2092 journal->j_flags |= JBD2_REC_ERR;
2093 write_unlock(&journal->j_state_lock);
2098 * void jbd2_journal_abort () - Shutdown the journal immediately.
2099 * @journal: the journal to shutdown.
2100 * @errno: an error number to record in the journal indicating
2101 * the reason for the shutdown.
2103 * Perform a complete, immediate shutdown of the ENTIRE
2104 * journal (not of a single transaction). This operation cannot be
2105 * undone without closing and reopening the journal.
2107 * The jbd2_journal_abort function is intended to support higher level error
2108 * recovery mechanisms such as the ext2/ext3 remount-readonly error
2111 * Journal abort has very specific semantics. Any existing dirty,
2112 * unjournaled buffers in the main filesystem will still be written to
2113 * disk by bdflush, but the journaling mechanism will be suspended
2114 * immediately and no further transaction commits will be honoured.
2116 * Any dirty, journaled buffers will be written back to disk without
2117 * hitting the journal. Atomicity cannot be guaranteed on an aborted
2118 * filesystem, but we _do_ attempt to leave as much data as possible
2119 * behind for fsck to use for cleanup.
2121 * Any attempt to get a new transaction handle on a journal which is in
2122 * ABORT state will just result in an -EROFS error return. A
2123 * jbd2_journal_stop on an existing handle will return -EIO if we have
2124 * entered abort state during the update.
2126 * Recursive transactions are not disturbed by journal abort until the
2127 * final jbd2_journal_stop, which will receive the -EIO error.
2129 * Finally, the jbd2_journal_abort call allows the caller to supply an errno
2130 * which will be recorded (if possible) in the journal superblock. This
2131 * allows a client to record failure conditions in the middle of a
2132 * transaction without having to complete the transaction to record the
2133 * failure to disk. ext3_error, for example, now uses this
2136 * Errors which originate from within the journaling layer will NOT
2137 * supply an errno; a null errno implies that absolutely no further
2138 * writes are done to the journal (unless there are any already in
2143 void jbd2_journal_abort(journal_t *journal, int errno)
2145 __journal_abort_soft(journal, errno);
2149 * int jbd2_journal_errno () - returns the journal's error state.
2150 * @journal: journal to examine.
2152 * This is the errno number set with jbd2_journal_abort(), the last
2153 * time the journal was mounted - if the journal was stopped
2154 * without calling abort this will be 0.
2156 * If the journal has been aborted on this mount time -EROFS will
2159 int jbd2_journal_errno(journal_t *journal)
2163 read_lock(&journal->j_state_lock);
2164 if (journal->j_flags & JBD2_ABORT)
2167 err = journal->j_errno;
2168 read_unlock(&journal->j_state_lock);
2173 * int jbd2_journal_clear_err () - clears the journal's error state
2174 * @journal: journal to act on.
2176 * An error must be cleared or acked to take a FS out of readonly
2179 int jbd2_journal_clear_err(journal_t *journal)
2183 write_lock(&journal->j_state_lock);
2184 if (journal->j_flags & JBD2_ABORT)
2187 journal->j_errno = 0;
2188 write_unlock(&journal->j_state_lock);
2193 * void jbd2_journal_ack_err() - Ack journal err.
2194 * @journal: journal to act on.
2196 * An error must be cleared or acked to take a FS out of readonly
2199 void jbd2_journal_ack_err(journal_t *journal)
2201 write_lock(&journal->j_state_lock);
2202 if (journal->j_errno)
2203 journal->j_flags |= JBD2_ACK_ERR;
2204 write_unlock(&journal->j_state_lock);
2207 int jbd2_journal_blocks_per_page(struct inode *inode)
2209 return 1 << (PAGE_SHIFT - inode->i_sb->s_blocksize_bits);
2213 * helper functions to deal with 32 or 64bit block numbers.
2215 size_t journal_tag_bytes(journal_t *journal)
2219 if (jbd2_has_feature_csum3(journal))
2220 return sizeof(journal_block_tag3_t);
2222 sz = sizeof(journal_block_tag_t);
2224 if (jbd2_has_feature_csum2(journal))
2225 sz += sizeof(__u16);
2227 if (jbd2_has_feature_64bit(journal))
2230 return sz - sizeof(__u32);
2234 * JBD memory management
2236 * These functions are used to allocate block-sized chunks of memory
2237 * used for making copies of buffer_head data. Very often it will be
2238 * page-sized chunks of data, but sometimes it will be in
2239 * sub-page-size chunks. (For example, 16k pages on Power systems
2240 * with a 4k block file system.) For blocks smaller than a page, we
2241 * use a SLAB allocator. There are slab caches for each block size,
2242 * which are allocated at mount time, if necessary, and we only free
2243 * (all of) the slab caches when/if the jbd2 module is unloaded. For
2244 * this reason we don't need to a mutex to protect access to
2245 * jbd2_slab[] allocating or releasing memory; only in
2246 * jbd2_journal_create_slab().
2248 #define JBD2_MAX_SLABS 8
2249 static struct kmem_cache *jbd2_slab[JBD2_MAX_SLABS];
2251 static const char *jbd2_slab_names[JBD2_MAX_SLABS] = {
2252 "jbd2_1k", "jbd2_2k", "jbd2_4k", "jbd2_8k",
2253 "jbd2_16k", "jbd2_32k", "jbd2_64k", "jbd2_128k"
2257 static void jbd2_journal_destroy_slabs(void)
2261 for (i = 0; i < JBD2_MAX_SLABS; i++) {
2263 kmem_cache_destroy(jbd2_slab[i]);
2264 jbd2_slab[i] = NULL;
2268 static int jbd2_journal_create_slab(size_t size)
2270 static DEFINE_MUTEX(jbd2_slab_create_mutex);
2271 int i = order_base_2(size) - 10;
2274 if (size == PAGE_SIZE)
2277 if (i >= JBD2_MAX_SLABS)
2280 if (unlikely(i < 0))
2282 mutex_lock(&jbd2_slab_create_mutex);
2284 mutex_unlock(&jbd2_slab_create_mutex);
2285 return 0; /* Already created */
2288 slab_size = 1 << (i+10);
2289 jbd2_slab[i] = kmem_cache_create(jbd2_slab_names[i], slab_size,
2290 slab_size, 0, NULL);
2291 mutex_unlock(&jbd2_slab_create_mutex);
2292 if (!jbd2_slab[i]) {
2293 printk(KERN_EMERG "JBD2: no memory for jbd2_slab cache\n");
2299 static struct kmem_cache *get_slab(size_t size)
2301 int i = order_base_2(size) - 10;
2303 BUG_ON(i >= JBD2_MAX_SLABS);
2304 if (unlikely(i < 0))
2306 BUG_ON(jbd2_slab[i] == NULL);
2307 return jbd2_slab[i];
2310 void *jbd2_alloc(size_t size, gfp_t flags)
2314 BUG_ON(size & (size-1)); /* Must be a power of 2 */
2316 if (size < PAGE_SIZE)
2317 ptr = kmem_cache_alloc(get_slab(size), flags);
2319 ptr = (void *)__get_free_pages(flags, get_order(size));
2321 /* Check alignment; SLUB has gotten this wrong in the past,
2322 * and this can lead to user data corruption! */
2323 BUG_ON(((unsigned long) ptr) & (size-1));
2328 void jbd2_free(void *ptr, size_t size)
2330 if (size < PAGE_SIZE)
2331 kmem_cache_free(get_slab(size), ptr);
2333 free_pages((unsigned long)ptr, get_order(size));
2337 * Journal_head storage management
2339 static struct kmem_cache *jbd2_journal_head_cache;
2340 #ifdef CONFIG_JBD2_DEBUG
2341 static atomic_t nr_journal_heads = ATOMIC_INIT(0);
2344 static int jbd2_journal_init_journal_head_cache(void)
2348 J_ASSERT(jbd2_journal_head_cache == NULL);
2349 jbd2_journal_head_cache = kmem_cache_create("jbd2_journal_head",
2350 sizeof(struct journal_head),
2352 SLAB_TEMPORARY | SLAB_DESTROY_BY_RCU,
2355 if (!jbd2_journal_head_cache) {
2357 printk(KERN_EMERG "JBD2: no memory for journal_head cache\n");
2362 static void jbd2_journal_destroy_journal_head_cache(void)
2364 if (jbd2_journal_head_cache) {
2365 kmem_cache_destroy(jbd2_journal_head_cache);
2366 jbd2_journal_head_cache = NULL;
2371 * journal_head splicing and dicing
2373 static struct journal_head *journal_alloc_journal_head(void)
2375 struct journal_head *ret;
2377 #ifdef CONFIG_JBD2_DEBUG
2378 atomic_inc(&nr_journal_heads);
2380 ret = kmem_cache_zalloc(jbd2_journal_head_cache, GFP_NOFS);
2382 jbd_debug(1, "out of memory for journal_head\n");
2383 pr_notice_ratelimited("ENOMEM in %s, retrying.\n", __func__);
2384 ret = kmem_cache_zalloc(jbd2_journal_head_cache,
2385 GFP_NOFS | __GFP_NOFAIL);
2390 static void journal_free_journal_head(struct journal_head *jh)
2392 #ifdef CONFIG_JBD2_DEBUG
2393 atomic_dec(&nr_journal_heads);
2394 memset(jh, JBD2_POISON_FREE, sizeof(*jh));
2396 kmem_cache_free(jbd2_journal_head_cache, jh);
2400 * A journal_head is attached to a buffer_head whenever JBD has an
2401 * interest in the buffer.
2403 * Whenever a buffer has an attached journal_head, its ->b_state:BH_JBD bit
2404 * is set. This bit is tested in core kernel code where we need to take
2405 * JBD-specific actions. Testing the zeroness of ->b_private is not reliable
2408 * When a buffer has its BH_JBD bit set, its ->b_count is elevated by one.
2410 * When a buffer has its BH_JBD bit set it is immune from being released by
2411 * core kernel code, mainly via ->b_count.
2413 * A journal_head is detached from its buffer_head when the journal_head's
2414 * b_jcount reaches zero. Running transaction (b_transaction) and checkpoint
2415 * transaction (b_cp_transaction) hold their references to b_jcount.
2417 * Various places in the kernel want to attach a journal_head to a buffer_head
2418 * _before_ attaching the journal_head to a transaction. To protect the
2419 * journal_head in this situation, jbd2_journal_add_journal_head elevates the
2420 * journal_head's b_jcount refcount by one. The caller must call
2421 * jbd2_journal_put_journal_head() to undo this.
2423 * So the typical usage would be:
2425 * (Attach a journal_head if needed. Increments b_jcount)
2426 * struct journal_head *jh = jbd2_journal_add_journal_head(bh);
2428 * (Get another reference for transaction)
2429 * jbd2_journal_grab_journal_head(bh);
2430 * jh->b_transaction = xxx;
2431 * (Put original reference)
2432 * jbd2_journal_put_journal_head(jh);
2436 * Give a buffer_head a journal_head.
2440 struct journal_head *jbd2_journal_add_journal_head(struct buffer_head *bh)
2442 struct journal_head *jh;
2443 struct journal_head *new_jh = NULL;
2446 if (!buffer_jbd(bh))
2447 new_jh = journal_alloc_journal_head();
2449 jbd_lock_bh_journal_head(bh);
2450 if (buffer_jbd(bh)) {
2454 (atomic_read(&bh->b_count) > 0) ||
2455 (bh->b_page && bh->b_page->mapping));
2458 jbd_unlock_bh_journal_head(bh);
2463 new_jh = NULL; /* We consumed it */
2468 BUFFER_TRACE(bh, "added journal_head");
2471 jbd_unlock_bh_journal_head(bh);
2473 journal_free_journal_head(new_jh);
2474 return bh->b_private;
2478 * Grab a ref against this buffer_head's journal_head. If it ended up not
2479 * having a journal_head, return NULL
2481 struct journal_head *jbd2_journal_grab_journal_head(struct buffer_head *bh)
2483 struct journal_head *jh = NULL;
2485 jbd_lock_bh_journal_head(bh);
2486 if (buffer_jbd(bh)) {
2490 jbd_unlock_bh_journal_head(bh);
2494 static void __journal_remove_journal_head(struct buffer_head *bh)
2496 struct journal_head *jh = bh2jh(bh);
2498 J_ASSERT_JH(jh, jh->b_jcount >= 0);
2499 J_ASSERT_JH(jh, jh->b_transaction == NULL);
2500 J_ASSERT_JH(jh, jh->b_next_transaction == NULL);
2501 J_ASSERT_JH(jh, jh->b_cp_transaction == NULL);
2502 J_ASSERT_JH(jh, jh->b_jlist == BJ_None);
2503 J_ASSERT_BH(bh, buffer_jbd(bh));
2504 J_ASSERT_BH(bh, jh2bh(jh) == bh);
2505 BUFFER_TRACE(bh, "remove journal_head");
2506 if (jh->b_frozen_data) {
2507 printk(KERN_WARNING "%s: freeing b_frozen_data\n", __func__);
2508 jbd2_free(jh->b_frozen_data, bh->b_size);
2510 if (jh->b_committed_data) {
2511 printk(KERN_WARNING "%s: freeing b_committed_data\n", __func__);
2512 jbd2_free(jh->b_committed_data, bh->b_size);
2514 bh->b_private = NULL;
2515 jh->b_bh = NULL; /* debug, really */
2516 clear_buffer_jbd(bh);
2517 journal_free_journal_head(jh);
2521 * Drop a reference on the passed journal_head. If it fell to zero then
2522 * release the journal_head from the buffer_head.
2524 void jbd2_journal_put_journal_head(struct journal_head *jh)
2526 struct buffer_head *bh = jh2bh(jh);
2528 jbd_lock_bh_journal_head(bh);
2529 J_ASSERT_JH(jh, jh->b_jcount > 0);
2531 if (!jh->b_jcount) {
2532 __journal_remove_journal_head(bh);
2533 jbd_unlock_bh_journal_head(bh);
2536 jbd_unlock_bh_journal_head(bh);
2540 * Initialize jbd inode head
2542 void jbd2_journal_init_jbd_inode(struct jbd2_inode *jinode, struct inode *inode)
2544 jinode->i_transaction = NULL;
2545 jinode->i_next_transaction = NULL;
2546 jinode->i_vfs_inode = inode;
2547 jinode->i_flags = 0;
2548 INIT_LIST_HEAD(&jinode->i_list);
2552 * Function to be called before we start removing inode from memory (i.e.,
2553 * clear_inode() is a fine place to be called from). It removes inode from
2554 * transaction's lists.
2556 void jbd2_journal_release_jbd_inode(journal_t *journal,
2557 struct jbd2_inode *jinode)
2562 spin_lock(&journal->j_list_lock);
2563 /* Is commit writing out inode - we have to wait */
2564 if (jinode->i_flags & JI_COMMIT_RUNNING) {
2565 wait_queue_head_t *wq;
2566 DEFINE_WAIT_BIT(wait, &jinode->i_flags, __JI_COMMIT_RUNNING);
2567 wq = bit_waitqueue(&jinode->i_flags, __JI_COMMIT_RUNNING);
2568 prepare_to_wait(wq, &wait.wait, TASK_UNINTERRUPTIBLE);
2569 spin_unlock(&journal->j_list_lock);
2571 finish_wait(wq, &wait.wait);
2575 if (jinode->i_transaction) {
2576 list_del(&jinode->i_list);
2577 jinode->i_transaction = NULL;
2579 spin_unlock(&journal->j_list_lock);
2583 #ifdef CONFIG_PROC_FS
2585 #define JBD2_STATS_PROC_NAME "fs/jbd2"
2587 static void __init jbd2_create_jbd_stats_proc_entry(void)
2589 proc_jbd2_stats = proc_mkdir(JBD2_STATS_PROC_NAME, NULL);
2592 static void __exit jbd2_remove_jbd_stats_proc_entry(void)
2594 if (proc_jbd2_stats)
2595 remove_proc_entry(JBD2_STATS_PROC_NAME, NULL);
2600 #define jbd2_create_jbd_stats_proc_entry() do {} while (0)
2601 #define jbd2_remove_jbd_stats_proc_entry() do {} while (0)
2605 struct kmem_cache *jbd2_handle_cache, *jbd2_inode_cache;
2607 static int __init jbd2_journal_init_handle_cache(void)
2609 jbd2_handle_cache = KMEM_CACHE(jbd2_journal_handle, SLAB_TEMPORARY);
2610 if (jbd2_handle_cache == NULL) {
2611 printk(KERN_EMERG "JBD2: failed to create handle cache\n");
2614 jbd2_inode_cache = KMEM_CACHE(jbd2_inode, 0);
2615 if (jbd2_inode_cache == NULL) {
2616 printk(KERN_EMERG "JBD2: failed to create inode cache\n");
2617 kmem_cache_destroy(jbd2_handle_cache);
2623 static void jbd2_journal_destroy_handle_cache(void)
2625 if (jbd2_handle_cache)
2626 kmem_cache_destroy(jbd2_handle_cache);
2627 if (jbd2_inode_cache)
2628 kmem_cache_destroy(jbd2_inode_cache);
2633 * Module startup and shutdown
2636 static int __init journal_init_caches(void)
2640 ret = jbd2_journal_init_revoke_caches();
2642 ret = jbd2_journal_init_journal_head_cache();
2644 ret = jbd2_journal_init_handle_cache();
2646 ret = jbd2_journal_init_transaction_cache();
2650 static void jbd2_journal_destroy_caches(void)
2652 jbd2_journal_destroy_revoke_caches();
2653 jbd2_journal_destroy_journal_head_cache();
2654 jbd2_journal_destroy_handle_cache();
2655 jbd2_journal_destroy_transaction_cache();
2656 jbd2_journal_destroy_slabs();
2659 static int __init journal_init(void)
2663 BUILD_BUG_ON(sizeof(struct journal_superblock_s) != 1024);
2665 ret = journal_init_caches();
2667 jbd2_create_jbd_stats_proc_entry();
2669 jbd2_journal_destroy_caches();
2674 static void __exit journal_exit(void)
2676 #ifdef CONFIG_JBD2_DEBUG
2677 int n = atomic_read(&nr_journal_heads);
2679 printk(KERN_ERR "JBD2: leaked %d journal_heads!\n", n);
2681 jbd2_remove_jbd_stats_proc_entry();
2682 jbd2_journal_destroy_caches();
2685 MODULE_LICENSE("GPL");
2686 module_init(journal_init);
2687 module_exit(journal_exit);