2 * linux/fs/jbd2/journal.c
4 * Written by Stephen C. Tweedie <sct@redhat.com>, 1998
6 * Copyright 1998 Red Hat corp --- All Rights Reserved
8 * This file is part of the Linux kernel and is made available under
9 * the terms of the GNU General Public License, version 2, or at your
10 * option, any later version, incorporated herein by reference.
12 * Generic filesystem journal-writing code; part of the ext2fs
15 * This file manages journals: areas of disk reserved for logging
16 * transactional updates. This includes the kernel journaling thread
17 * which is responsible for scheduling updates to the log.
19 * We do not actually manage the physical storage of the journal in this
20 * file: that is left to a per-journal policy function, which allows us
21 * to store the journal within a filesystem-specified area for ext2
22 * journaling (ext2 can use a reserved inode for storing the log).
25 #include <linux/module.h>
26 #include <linux/time.h>
28 #include <linux/jbd2.h>
29 #include <linux/errno.h>
30 #include <linux/slab.h>
31 #include <linux/init.h>
33 #include <linux/freezer.h>
34 #include <linux/pagemap.h>
35 #include <linux/kthread.h>
36 #include <linux/poison.h>
37 #include <linux/proc_fs.h>
38 #include <linux/seq_file.h>
39 #include <linux/math64.h>
40 #include <linux/hash.h>
41 #include <linux/log2.h>
42 #include <linux/vmalloc.h>
43 #include <linux/backing-dev.h>
44 #include <linux/bitops.h>
45 #include <linux/ratelimit.h>
47 #define CREATE_TRACE_POINTS
48 #include <trace/events/jbd2.h>
50 #include <asm/uaccess.h>
53 #ifdef CONFIG_JBD2_DEBUG
54 ushort jbd2_journal_enable_debug __read_mostly;
55 EXPORT_SYMBOL(jbd2_journal_enable_debug);
57 module_param_named(jbd2_debug, jbd2_journal_enable_debug, ushort, 0644);
58 MODULE_PARM_DESC(jbd2_debug, "Debugging level for jbd2");
61 EXPORT_SYMBOL(jbd2_journal_extend);
62 EXPORT_SYMBOL(jbd2_journal_stop);
63 EXPORT_SYMBOL(jbd2_journal_lock_updates);
64 EXPORT_SYMBOL(jbd2_journal_unlock_updates);
65 EXPORT_SYMBOL(jbd2_journal_get_write_access);
66 EXPORT_SYMBOL(jbd2_journal_get_create_access);
67 EXPORT_SYMBOL(jbd2_journal_get_undo_access);
68 EXPORT_SYMBOL(jbd2_journal_set_triggers);
69 EXPORT_SYMBOL(jbd2_journal_dirty_metadata);
70 EXPORT_SYMBOL(jbd2_journal_forget);
72 EXPORT_SYMBOL(journal_sync_buffer);
74 EXPORT_SYMBOL(jbd2_journal_flush);
75 EXPORT_SYMBOL(jbd2_journal_revoke);
77 EXPORT_SYMBOL(jbd2_journal_init_dev);
78 EXPORT_SYMBOL(jbd2_journal_init_inode);
79 EXPORT_SYMBOL(jbd2_journal_check_used_features);
80 EXPORT_SYMBOL(jbd2_journal_check_available_features);
81 EXPORT_SYMBOL(jbd2_journal_set_features);
82 EXPORT_SYMBOL(jbd2_journal_load);
83 EXPORT_SYMBOL(jbd2_journal_destroy);
84 EXPORT_SYMBOL(jbd2_journal_abort);
85 EXPORT_SYMBOL(jbd2_journal_errno);
86 EXPORT_SYMBOL(jbd2_journal_ack_err);
87 EXPORT_SYMBOL(jbd2_journal_clear_err);
88 EXPORT_SYMBOL(jbd2_log_wait_commit);
89 EXPORT_SYMBOL(jbd2_log_start_commit);
90 EXPORT_SYMBOL(jbd2_journal_start_commit);
91 EXPORT_SYMBOL(jbd2_journal_force_commit_nested);
92 EXPORT_SYMBOL(jbd2_journal_wipe);
93 EXPORT_SYMBOL(jbd2_journal_blocks_per_page);
94 EXPORT_SYMBOL(jbd2_journal_invalidatepage);
95 EXPORT_SYMBOL(jbd2_journal_try_to_free_buffers);
96 EXPORT_SYMBOL(jbd2_journal_force_commit);
97 EXPORT_SYMBOL(jbd2_journal_file_inode);
98 EXPORT_SYMBOL(jbd2_journal_init_jbd_inode);
99 EXPORT_SYMBOL(jbd2_journal_release_jbd_inode);
100 EXPORT_SYMBOL(jbd2_journal_begin_ordered_truncate);
101 EXPORT_SYMBOL(jbd2_inode_cache);
103 static void __journal_abort_soft (journal_t *journal, int errno);
104 static int jbd2_journal_create_slab(size_t slab_size);
106 /* Checksumming functions */
107 int jbd2_verify_csum_type(journal_t *j, journal_superblock_t *sb)
109 if (!JBD2_HAS_INCOMPAT_FEATURE(j, JBD2_FEATURE_INCOMPAT_CSUM_V2))
112 return sb->s_checksum_type == JBD2_CRC32C_CHKSUM;
115 static __u32 jbd2_superblock_csum(journal_t *j, journal_superblock_t *sb)
117 __u32 csum, old_csum;
119 old_csum = sb->s_checksum;
121 csum = jbd2_chksum(j, ~0, (char *)sb, sizeof(journal_superblock_t));
122 sb->s_checksum = old_csum;
124 return cpu_to_be32(csum);
127 int jbd2_superblock_csum_verify(journal_t *j, journal_superblock_t *sb)
129 if (!JBD2_HAS_INCOMPAT_FEATURE(j, JBD2_FEATURE_INCOMPAT_CSUM_V2))
132 return sb->s_checksum == jbd2_superblock_csum(j, sb);
135 void jbd2_superblock_csum_set(journal_t *j, journal_superblock_t *sb)
137 if (!JBD2_HAS_INCOMPAT_FEATURE(j, JBD2_FEATURE_INCOMPAT_CSUM_V2))
140 sb->s_checksum = jbd2_superblock_csum(j, sb);
144 * Helper function used to manage commit timeouts
147 static void commit_timeout(unsigned long __data)
149 struct task_struct * p = (struct task_struct *) __data;
155 * kjournald2: The main thread function used to manage a logging device
158 * This kernel thread is responsible for two things:
160 * 1) COMMIT: Every so often we need to commit the current state of the
161 * filesystem to disk. The journal thread is responsible for writing
162 * all of the metadata buffers to disk.
164 * 2) CHECKPOINT: We cannot reuse a used section of the log file until all
165 * of the data in that part of the log has been rewritten elsewhere on
166 * the disk. Flushing these old buffers to reclaim space in the log is
167 * known as checkpointing, and this thread is responsible for that job.
170 static int kjournald2(void *arg)
172 journal_t *journal = arg;
173 transaction_t *transaction;
176 * Set up an interval timer which can be used to trigger a commit wakeup
177 * after the commit interval expires
179 setup_timer(&journal->j_commit_timer, commit_timeout,
180 (unsigned long)current);
184 /* Record that the journal thread is running */
185 journal->j_task = current;
186 wake_up(&journal->j_wait_done_commit);
189 * And now, wait forever for commit wakeup events.
191 write_lock(&journal->j_state_lock);
194 if (journal->j_flags & JBD2_UNMOUNT)
197 jbd_debug(1, "commit_sequence=%d, commit_request=%d\n",
198 journal->j_commit_sequence, journal->j_commit_request);
200 if (journal->j_commit_sequence != journal->j_commit_request) {
201 jbd_debug(1, "OK, requests differ\n");
202 write_unlock(&journal->j_state_lock);
203 del_timer_sync(&journal->j_commit_timer);
204 jbd2_journal_commit_transaction(journal);
205 write_lock(&journal->j_state_lock);
209 wake_up(&journal->j_wait_done_commit);
210 if (freezing(current)) {
212 * The simpler the better. Flushing journal isn't a
213 * good idea, because that depends on threads that may
214 * be already stopped.
216 jbd_debug(1, "Now suspending kjournald2\n");
217 write_unlock(&journal->j_state_lock);
219 write_lock(&journal->j_state_lock);
222 * We assume on resume that commits are already there,
226 int should_sleep = 1;
228 prepare_to_wait(&journal->j_wait_commit, &wait,
230 if (journal->j_commit_sequence != journal->j_commit_request)
232 transaction = journal->j_running_transaction;
233 if (transaction && time_after_eq(jiffies,
234 transaction->t_expires))
236 if (journal->j_flags & JBD2_UNMOUNT)
239 write_unlock(&journal->j_state_lock);
241 write_lock(&journal->j_state_lock);
243 finish_wait(&journal->j_wait_commit, &wait);
246 jbd_debug(1, "kjournald2 wakes\n");
249 * Were we woken up by a commit wakeup event?
251 transaction = journal->j_running_transaction;
252 if (transaction && time_after_eq(jiffies, transaction->t_expires)) {
253 journal->j_commit_request = transaction->t_tid;
254 jbd_debug(1, "woke because of timeout\n");
259 write_unlock(&journal->j_state_lock);
260 del_timer_sync(&journal->j_commit_timer);
261 journal->j_task = NULL;
262 wake_up(&journal->j_wait_done_commit);
263 jbd_debug(1, "Journal thread exiting.\n");
267 static int jbd2_journal_start_thread(journal_t *journal)
269 struct task_struct *t;
271 t = kthread_run(kjournald2, journal, "jbd2/%s",
276 wait_event(journal->j_wait_done_commit, journal->j_task != NULL);
280 static void journal_kill_thread(journal_t *journal)
282 write_lock(&journal->j_state_lock);
283 journal->j_flags |= JBD2_UNMOUNT;
285 while (journal->j_task) {
286 wake_up(&journal->j_wait_commit);
287 write_unlock(&journal->j_state_lock);
288 wait_event(journal->j_wait_done_commit, journal->j_task == NULL);
289 write_lock(&journal->j_state_lock);
291 write_unlock(&journal->j_state_lock);
295 * jbd2_journal_write_metadata_buffer: write a metadata buffer to the journal.
297 * Writes a metadata buffer to a given disk block. The actual IO is not
298 * performed but a new buffer_head is constructed which labels the data
299 * to be written with the correct destination disk block.
301 * Any magic-number escaping which needs to be done will cause a
302 * copy-out here. If the buffer happens to start with the
303 * JBD2_MAGIC_NUMBER, then we can't write it to the log directly: the
304 * magic number is only written to the log for descripter blocks. In
305 * this case, we copy the data and replace the first word with 0, and we
306 * return a result code which indicates that this buffer needs to be
307 * marked as an escaped buffer in the corresponding log descriptor
308 * block. The missing word can then be restored when the block is read
311 * If the source buffer has already been modified by a new transaction
312 * since we took the last commit snapshot, we use the frozen copy of
313 * that data for IO. If we end up using the existing buffer_head's data
314 * for the write, then we have to make sure nobody modifies it while the
315 * IO is in progress. do_get_write_access() handles this.
317 * The function returns a pointer to the buffer_head to be used for IO.
325 * Bit 0 set == escape performed on the data
326 * Bit 1 set == buffer copy-out performed (kfree the data after IO)
329 int jbd2_journal_write_metadata_buffer(transaction_t *transaction,
330 struct journal_head *jh_in,
331 struct buffer_head **bh_out,
334 int need_copy_out = 0;
335 int done_copy_out = 0;
338 struct buffer_head *new_bh;
339 struct page *new_page;
340 unsigned int new_offset;
341 struct buffer_head *bh_in = jh2bh(jh_in);
342 journal_t *journal = transaction->t_journal;
345 * The buffer really shouldn't be locked: only the current committing
346 * transaction is allowed to write it, so nobody else is allowed
349 * akpm: except if we're journalling data, and write() output is
350 * also part of a shared mapping, and another thread has
351 * decided to launch a writepage() against this buffer.
353 J_ASSERT_BH(bh_in, buffer_jbddirty(bh_in));
356 new_bh = alloc_buffer_head(GFP_NOFS);
359 * Failure is not an option, but __GFP_NOFAIL is going
360 * away; so we retry ourselves here.
362 congestion_wait(BLK_RW_ASYNC, HZ/50);
366 /* keep subsequent assertions sane */
367 atomic_set(&new_bh->b_count, 1);
369 jbd_lock_bh_state(bh_in);
372 * If a new transaction has already done a buffer copy-out, then
373 * we use that version of the data for the commit.
375 if (jh_in->b_frozen_data) {
377 new_page = virt_to_page(jh_in->b_frozen_data);
378 new_offset = offset_in_page(jh_in->b_frozen_data);
380 new_page = jh2bh(jh_in)->b_page;
381 new_offset = offset_in_page(jh2bh(jh_in)->b_data);
384 mapped_data = kmap_atomic(new_page);
386 * Fire data frozen trigger if data already wasn't frozen. Do this
387 * before checking for escaping, as the trigger may modify the magic
388 * offset. If a copy-out happens afterwards, it will have the correct
389 * data in the buffer.
392 jbd2_buffer_frozen_trigger(jh_in, mapped_data + new_offset,
398 if (*((__be32 *)(mapped_data + new_offset)) ==
399 cpu_to_be32(JBD2_MAGIC_NUMBER)) {
403 kunmap_atomic(mapped_data);
406 * Do we need to do a data copy?
408 if (need_copy_out && !done_copy_out) {
411 jbd_unlock_bh_state(bh_in);
412 tmp = jbd2_alloc(bh_in->b_size, GFP_NOFS);
417 jbd_lock_bh_state(bh_in);
418 if (jh_in->b_frozen_data) {
419 jbd2_free(tmp, bh_in->b_size);
423 jh_in->b_frozen_data = tmp;
424 mapped_data = kmap_atomic(new_page);
425 memcpy(tmp, mapped_data + new_offset, bh_in->b_size);
426 kunmap_atomic(mapped_data);
428 new_page = virt_to_page(tmp);
429 new_offset = offset_in_page(tmp);
433 * This isn't strictly necessary, as we're using frozen
434 * data for the escaping, but it keeps consistency with
435 * b_frozen_data usage.
437 jh_in->b_frozen_triggers = jh_in->b_triggers;
441 * Did we need to do an escaping? Now we've done all the
442 * copying, we can finally do so.
445 mapped_data = kmap_atomic(new_page);
446 *((unsigned int *)(mapped_data + new_offset)) = 0;
447 kunmap_atomic(mapped_data);
450 set_bh_page(new_bh, new_page, new_offset);
451 new_bh->b_size = bh_in->b_size;
452 new_bh->b_bdev = journal->j_dev;
453 new_bh->b_blocknr = blocknr;
454 new_bh->b_private = bh_in;
455 set_buffer_mapped(new_bh);
456 set_buffer_dirty(new_bh);
461 * The to-be-written buffer needs to get moved to the io queue,
462 * and the original buffer whose contents we are shadowing or
463 * copying is moved to the transaction's shadow queue.
465 JBUFFER_TRACE(jh_in, "file as BJ_Shadow");
466 spin_lock(&journal->j_list_lock);
467 __jbd2_journal_file_buffer(jh_in, transaction, BJ_Shadow);
468 spin_unlock(&journal->j_list_lock);
469 set_buffer_shadow(bh_in);
470 jbd_unlock_bh_state(bh_in);
472 return do_escape | (done_copy_out << 1);
476 * Allocation code for the journal file. Manage the space left in the
477 * journal, so that we can begin checkpointing when appropriate.
481 * Called with j_state_lock locked for writing.
482 * Returns true if a transaction commit was started.
484 int __jbd2_log_start_commit(journal_t *journal, tid_t target)
486 /* Return if the txn has already requested to be committed */
487 if (journal->j_commit_request == target)
491 * The only transaction we can possibly wait upon is the
492 * currently running transaction (if it exists). Otherwise,
493 * the target tid must be an old one.
495 if (journal->j_running_transaction &&
496 journal->j_running_transaction->t_tid == target) {
498 * We want a new commit: OK, mark the request and wakeup the
499 * commit thread. We do _not_ do the commit ourselves.
502 journal->j_commit_request = target;
503 jbd_debug(1, "JBD2: requesting commit %d/%d\n",
504 journal->j_commit_request,
505 journal->j_commit_sequence);
506 journal->j_running_transaction->t_requested = jiffies;
507 wake_up(&journal->j_wait_commit);
509 } else if (!tid_geq(journal->j_commit_request, target))
510 /* This should never happen, but if it does, preserve
511 the evidence before kjournald goes into a loop and
512 increments j_commit_sequence beyond all recognition. */
513 WARN_ONCE(1, "JBD2: bad log_start_commit: %u %u %u %u\n",
514 journal->j_commit_request,
515 journal->j_commit_sequence,
516 target, journal->j_running_transaction ?
517 journal->j_running_transaction->t_tid : 0);
521 int jbd2_log_start_commit(journal_t *journal, tid_t tid)
525 write_lock(&journal->j_state_lock);
526 ret = __jbd2_log_start_commit(journal, tid);
527 write_unlock(&journal->j_state_lock);
532 * Force and wait upon a commit if the calling process is not within
533 * transaction. This is used for forcing out undo-protected data which contains
534 * bitmaps, when the fs is running out of space.
536 * We can only force the running transaction if we don't have an active handle;
537 * otherwise, we will deadlock.
539 * Returns true if a transaction was started.
541 int jbd2_journal_force_commit_nested(journal_t *journal)
543 transaction_t *transaction = NULL;
545 int need_to_start = 0;
547 read_lock(&journal->j_state_lock);
548 if (journal->j_running_transaction && !current->journal_info) {
549 transaction = journal->j_running_transaction;
550 if (!tid_geq(journal->j_commit_request, transaction->t_tid))
552 } else if (journal->j_committing_transaction)
553 transaction = journal->j_committing_transaction;
556 read_unlock(&journal->j_state_lock);
557 return 0; /* Nothing to retry */
560 tid = transaction->t_tid;
561 read_unlock(&journal->j_state_lock);
563 jbd2_log_start_commit(journal, tid);
564 jbd2_log_wait_commit(journal, tid);
569 * Start a commit of the current running transaction (if any). Returns true
570 * if a transaction is going to be committed (or is currently already
571 * committing), and fills its tid in at *ptid
573 int jbd2_journal_start_commit(journal_t *journal, tid_t *ptid)
577 write_lock(&journal->j_state_lock);
578 if (journal->j_running_transaction) {
579 tid_t tid = journal->j_running_transaction->t_tid;
581 __jbd2_log_start_commit(journal, tid);
582 /* There's a running transaction and we've just made sure
583 * it's commit has been scheduled. */
587 } else if (journal->j_committing_transaction) {
589 * If commit has been started, then we have to wait for
590 * completion of that transaction.
593 *ptid = journal->j_committing_transaction->t_tid;
596 write_unlock(&journal->j_state_lock);
601 * Return 1 if a given transaction has not yet sent barrier request
602 * connected with a transaction commit. If 0 is returned, transaction
603 * may or may not have sent the barrier. Used to avoid sending barrier
604 * twice in common cases.
606 int jbd2_trans_will_send_data_barrier(journal_t *journal, tid_t tid)
609 transaction_t *commit_trans;
611 if (!(journal->j_flags & JBD2_BARRIER))
613 read_lock(&journal->j_state_lock);
614 /* Transaction already committed? */
615 if (tid_geq(journal->j_commit_sequence, tid))
617 commit_trans = journal->j_committing_transaction;
618 if (!commit_trans || commit_trans->t_tid != tid) {
623 * Transaction is being committed and we already proceeded to
624 * submitting a flush to fs partition?
626 if (journal->j_fs_dev != journal->j_dev) {
627 if (!commit_trans->t_need_data_flush ||
628 commit_trans->t_state >= T_COMMIT_DFLUSH)
631 if (commit_trans->t_state >= T_COMMIT_JFLUSH)
636 read_unlock(&journal->j_state_lock);
639 EXPORT_SYMBOL(jbd2_trans_will_send_data_barrier);
642 * Wait for a specified commit to complete.
643 * The caller may not hold the journal lock.
645 int jbd2_log_wait_commit(journal_t *journal, tid_t tid)
649 read_lock(&journal->j_state_lock);
650 #ifdef CONFIG_JBD2_DEBUG
651 if (!tid_geq(journal->j_commit_request, tid)) {
653 "%s: error: j_commit_request=%d, tid=%d\n",
654 __func__, journal->j_commit_request, tid);
657 while (tid_gt(tid, journal->j_commit_sequence)) {
658 jbd_debug(1, "JBD2: want %d, j_commit_sequence=%d\n",
659 tid, journal->j_commit_sequence);
660 wake_up(&journal->j_wait_commit);
661 read_unlock(&journal->j_state_lock);
662 wait_event(journal->j_wait_done_commit,
663 !tid_gt(tid, journal->j_commit_sequence));
664 read_lock(&journal->j_state_lock);
666 read_unlock(&journal->j_state_lock);
668 if (unlikely(is_journal_aborted(journal))) {
669 printk(KERN_EMERG "journal commit I/O error\n");
676 * When this function returns the transaction corresponding to tid
677 * will be completed. If the transaction has currently running, start
678 * committing that transaction before waiting for it to complete. If
679 * the transaction id is stale, it is by definition already completed,
680 * so just return SUCCESS.
682 int jbd2_complete_transaction(journal_t *journal, tid_t tid)
684 int need_to_wait = 1;
686 read_lock(&journal->j_state_lock);
687 if (journal->j_running_transaction &&
688 journal->j_running_transaction->t_tid == tid) {
689 if (journal->j_commit_request != tid) {
690 /* transaction not yet started, so request it */
691 read_unlock(&journal->j_state_lock);
692 jbd2_log_start_commit(journal, tid);
695 } else if (!(journal->j_committing_transaction &&
696 journal->j_committing_transaction->t_tid == tid))
698 read_unlock(&journal->j_state_lock);
702 return jbd2_log_wait_commit(journal, tid);
704 EXPORT_SYMBOL(jbd2_complete_transaction);
707 * Log buffer allocation routines:
710 int jbd2_journal_next_log_block(journal_t *journal, unsigned long long *retp)
712 unsigned long blocknr;
714 write_lock(&journal->j_state_lock);
715 J_ASSERT(journal->j_free > 1);
717 blocknr = journal->j_head;
720 if (journal->j_head == journal->j_last)
721 journal->j_head = journal->j_first;
722 write_unlock(&journal->j_state_lock);
723 return jbd2_journal_bmap(journal, blocknr, retp);
727 * Conversion of logical to physical block numbers for the journal
729 * On external journals the journal blocks are identity-mapped, so
730 * this is a no-op. If needed, we can use j_blk_offset - everything is
733 int jbd2_journal_bmap(journal_t *journal, unsigned long blocknr,
734 unsigned long long *retp)
737 unsigned long long ret;
739 if (journal->j_inode) {
740 ret = bmap(journal->j_inode, blocknr);
744 printk(KERN_ALERT "%s: journal block not found "
745 "at offset %lu on %s\n",
746 __func__, blocknr, journal->j_devname);
748 __journal_abort_soft(journal, err);
751 *retp = blocknr; /* +journal->j_blk_offset */
757 * We play buffer_head aliasing tricks to write data/metadata blocks to
758 * the journal without copying their contents, but for journal
759 * descriptor blocks we do need to generate bona fide buffers.
761 * After the caller of jbd2_journal_get_descriptor_buffer() has finished modifying
762 * the buffer's contents they really should run flush_dcache_page(bh->b_page).
763 * But we don't bother doing that, so there will be coherency problems with
764 * mmaps of blockdevs which hold live JBD-controlled filesystems.
766 struct buffer_head *jbd2_journal_get_descriptor_buffer(journal_t *journal)
768 struct buffer_head *bh;
769 unsigned long long blocknr;
772 err = jbd2_journal_next_log_block(journal, &blocknr);
777 bh = __getblk(journal->j_dev, blocknr, journal->j_blocksize);
781 memset(bh->b_data, 0, journal->j_blocksize);
782 set_buffer_uptodate(bh);
784 BUFFER_TRACE(bh, "return this buffer");
789 * Return tid of the oldest transaction in the journal and block in the journal
790 * where the transaction starts.
792 * If the journal is now empty, return which will be the next transaction ID
793 * we will write and where will that transaction start.
795 * The return value is 0 if journal tail cannot be pushed any further, 1 if
798 int jbd2_journal_get_log_tail(journal_t *journal, tid_t *tid,
799 unsigned long *block)
801 transaction_t *transaction;
804 read_lock(&journal->j_state_lock);
805 spin_lock(&journal->j_list_lock);
806 transaction = journal->j_checkpoint_transactions;
808 *tid = transaction->t_tid;
809 *block = transaction->t_log_start;
810 } else if ((transaction = journal->j_committing_transaction) != NULL) {
811 *tid = transaction->t_tid;
812 *block = transaction->t_log_start;
813 } else if ((transaction = journal->j_running_transaction) != NULL) {
814 *tid = transaction->t_tid;
815 *block = journal->j_head;
817 *tid = journal->j_transaction_sequence;
818 *block = journal->j_head;
820 ret = tid_gt(*tid, journal->j_tail_sequence);
821 spin_unlock(&journal->j_list_lock);
822 read_unlock(&journal->j_state_lock);
828 * Update information in journal structure and in on disk journal superblock
829 * about log tail. This function does not check whether information passed in
830 * really pushes log tail further. It's responsibility of the caller to make
831 * sure provided log tail information is valid (e.g. by holding
832 * j_checkpoint_mutex all the time between computing log tail and calling this
833 * function as is the case with jbd2_cleanup_journal_tail()).
835 * Requires j_checkpoint_mutex
837 void __jbd2_update_log_tail(journal_t *journal, tid_t tid, unsigned long block)
841 BUG_ON(!mutex_is_locked(&journal->j_checkpoint_mutex));
844 * We cannot afford for write to remain in drive's caches since as
845 * soon as we update j_tail, next transaction can start reusing journal
846 * space and if we lose sb update during power failure we'd replay
847 * old transaction with possibly newly overwritten data.
849 jbd2_journal_update_sb_log_tail(journal, tid, block, WRITE_FUA);
850 write_lock(&journal->j_state_lock);
851 freed = block - journal->j_tail;
852 if (block < journal->j_tail)
853 freed += journal->j_last - journal->j_first;
855 trace_jbd2_update_log_tail(journal, tid, block, freed);
857 "Cleaning journal tail from %d to %d (offset %lu), "
859 journal->j_tail_sequence, tid, block, freed);
861 journal->j_free += freed;
862 journal->j_tail_sequence = tid;
863 journal->j_tail = block;
864 write_unlock(&journal->j_state_lock);
868 * This is a variaon of __jbd2_update_log_tail which checks for validity of
869 * provided log tail and locks j_checkpoint_mutex. So it is safe against races
870 * with other threads updating log tail.
872 void jbd2_update_log_tail(journal_t *journal, tid_t tid, unsigned long block)
874 mutex_lock(&journal->j_checkpoint_mutex);
875 if (tid_gt(tid, journal->j_tail_sequence))
876 __jbd2_update_log_tail(journal, tid, block);
877 mutex_unlock(&journal->j_checkpoint_mutex);
880 struct jbd2_stats_proc_session {
882 struct transaction_stats_s *stats;
887 static void *jbd2_seq_info_start(struct seq_file *seq, loff_t *pos)
889 return *pos ? NULL : SEQ_START_TOKEN;
892 static void *jbd2_seq_info_next(struct seq_file *seq, void *v, loff_t *pos)
897 static int jbd2_seq_info_show(struct seq_file *seq, void *v)
899 struct jbd2_stats_proc_session *s = seq->private;
901 if (v != SEQ_START_TOKEN)
903 seq_printf(seq, "%lu transactions (%lu requested), "
904 "each up to %u blocks\n",
905 s->stats->ts_tid, s->stats->ts_requested,
906 s->journal->j_max_transaction_buffers);
907 if (s->stats->ts_tid == 0)
909 seq_printf(seq, "average: \n %ums waiting for transaction\n",
910 jiffies_to_msecs(s->stats->run.rs_wait / s->stats->ts_tid));
911 seq_printf(seq, " %ums request delay\n",
912 (s->stats->ts_requested == 0) ? 0 :
913 jiffies_to_msecs(s->stats->run.rs_request_delay /
914 s->stats->ts_requested));
915 seq_printf(seq, " %ums running transaction\n",
916 jiffies_to_msecs(s->stats->run.rs_running / s->stats->ts_tid));
917 seq_printf(seq, " %ums transaction was being locked\n",
918 jiffies_to_msecs(s->stats->run.rs_locked / s->stats->ts_tid));
919 seq_printf(seq, " %ums flushing data (in ordered mode)\n",
920 jiffies_to_msecs(s->stats->run.rs_flushing / s->stats->ts_tid));
921 seq_printf(seq, " %ums logging transaction\n",
922 jiffies_to_msecs(s->stats->run.rs_logging / s->stats->ts_tid));
923 seq_printf(seq, " %lluus average transaction commit time\n",
924 div_u64(s->journal->j_average_commit_time, 1000));
925 seq_printf(seq, " %lu handles per transaction\n",
926 s->stats->run.rs_handle_count / s->stats->ts_tid);
927 seq_printf(seq, " %lu blocks per transaction\n",
928 s->stats->run.rs_blocks / s->stats->ts_tid);
929 seq_printf(seq, " %lu logged blocks per transaction\n",
930 s->stats->run.rs_blocks_logged / s->stats->ts_tid);
934 static void jbd2_seq_info_stop(struct seq_file *seq, void *v)
938 static const struct seq_operations jbd2_seq_info_ops = {
939 .start = jbd2_seq_info_start,
940 .next = jbd2_seq_info_next,
941 .stop = jbd2_seq_info_stop,
942 .show = jbd2_seq_info_show,
945 static int jbd2_seq_info_open(struct inode *inode, struct file *file)
947 journal_t *journal = PDE_DATA(inode);
948 struct jbd2_stats_proc_session *s;
951 s = kmalloc(sizeof(*s), GFP_KERNEL);
954 size = sizeof(struct transaction_stats_s);
955 s->stats = kmalloc(size, GFP_KERNEL);
956 if (s->stats == NULL) {
960 spin_lock(&journal->j_history_lock);
961 memcpy(s->stats, &journal->j_stats, size);
962 s->journal = journal;
963 spin_unlock(&journal->j_history_lock);
965 rc = seq_open(file, &jbd2_seq_info_ops);
967 struct seq_file *m = file->private_data;
977 static int jbd2_seq_info_release(struct inode *inode, struct file *file)
979 struct seq_file *seq = file->private_data;
980 struct jbd2_stats_proc_session *s = seq->private;
983 return seq_release(inode, file);
986 static const struct file_operations jbd2_seq_info_fops = {
987 .owner = THIS_MODULE,
988 .open = jbd2_seq_info_open,
991 .release = jbd2_seq_info_release,
994 static struct proc_dir_entry *proc_jbd2_stats;
996 static void jbd2_stats_proc_init(journal_t *journal)
998 journal->j_proc_entry = proc_mkdir(journal->j_devname, proc_jbd2_stats);
999 if (journal->j_proc_entry) {
1000 proc_create_data("info", S_IRUGO, journal->j_proc_entry,
1001 &jbd2_seq_info_fops, journal);
1005 static void jbd2_stats_proc_exit(journal_t *journal)
1007 remove_proc_entry("info", journal->j_proc_entry);
1008 remove_proc_entry(journal->j_devname, proc_jbd2_stats);
1012 * Management for journal control blocks: functions to create and
1013 * destroy journal_t structures, and to initialise and read existing
1014 * journal blocks from disk. */
1016 /* First: create and setup a journal_t object in memory. We initialise
1017 * very few fields yet: that has to wait until we have created the
1018 * journal structures from from scratch, or loaded them from disk. */
1020 static journal_t * journal_init_common (void)
1025 journal = kzalloc(sizeof(*journal), GFP_KERNEL);
1029 init_waitqueue_head(&journal->j_wait_transaction_locked);
1030 init_waitqueue_head(&journal->j_wait_logspace);
1031 init_waitqueue_head(&journal->j_wait_done_commit);
1032 init_waitqueue_head(&journal->j_wait_checkpoint);
1033 init_waitqueue_head(&journal->j_wait_commit);
1034 init_waitqueue_head(&journal->j_wait_updates);
1035 mutex_init(&journal->j_barrier);
1036 mutex_init(&journal->j_checkpoint_mutex);
1037 spin_lock_init(&journal->j_revoke_lock);
1038 spin_lock_init(&journal->j_list_lock);
1039 rwlock_init(&journal->j_state_lock);
1041 journal->j_commit_interval = (HZ * JBD2_DEFAULT_MAX_COMMIT_AGE);
1042 journal->j_min_batch_time = 0;
1043 journal->j_max_batch_time = 15000; /* 15ms */
1045 /* The journal is marked for error until we succeed with recovery! */
1046 journal->j_flags = JBD2_ABORT;
1048 /* Set up a default-sized revoke table for the new mount. */
1049 err = jbd2_journal_init_revoke(journal, JOURNAL_REVOKE_DEFAULT_HASH);
1055 spin_lock_init(&journal->j_history_lock);
1060 /* jbd2_journal_init_dev and jbd2_journal_init_inode:
1062 * Create a journal structure assigned some fixed set of disk blocks to
1063 * the journal. We don't actually touch those disk blocks yet, but we
1064 * need to set up all of the mapping information to tell the journaling
1065 * system where the journal blocks are.
1070 * journal_t * jbd2_journal_init_dev() - creates and initialises a journal structure
1071 * @bdev: Block device on which to create the journal
1072 * @fs_dev: Device which hold journalled filesystem for this journal.
1073 * @start: Block nr Start of journal.
1074 * @len: Length of the journal in blocks.
1075 * @blocksize: blocksize of journalling device
1077 * Returns: a newly created journal_t *
1079 * jbd2_journal_init_dev creates a journal which maps a fixed contiguous
1080 * range of blocks on an arbitrary block device.
1083 journal_t * jbd2_journal_init_dev(struct block_device *bdev,
1084 struct block_device *fs_dev,
1085 unsigned long long start, int len, int blocksize)
1087 journal_t *journal = journal_init_common();
1088 struct buffer_head *bh;
1095 /* journal descriptor can store up to n blocks -bzzz */
1096 journal->j_blocksize = blocksize;
1097 journal->j_dev = bdev;
1098 journal->j_fs_dev = fs_dev;
1099 journal->j_blk_offset = start;
1100 journal->j_maxlen = len;
1101 bdevname(journal->j_dev, journal->j_devname);
1102 p = journal->j_devname;
1103 while ((p = strchr(p, '/')))
1105 jbd2_stats_proc_init(journal);
1106 n = journal->j_blocksize / sizeof(journal_block_tag_t);
1107 journal->j_wbufsize = n;
1108 journal->j_wbuf = kmalloc(n * sizeof(struct buffer_head*), GFP_KERNEL);
1109 if (!journal->j_wbuf) {
1110 printk(KERN_ERR "%s: Can't allocate bhs for commit thread\n",
1115 bh = __getblk(journal->j_dev, start, journal->j_blocksize);
1118 "%s: Cannot get buffer for journal superblock\n",
1122 journal->j_sb_buffer = bh;
1123 journal->j_superblock = (journal_superblock_t *)bh->b_data;
1127 kfree(journal->j_wbuf);
1128 jbd2_stats_proc_exit(journal);
1134 * journal_t * jbd2_journal_init_inode () - creates a journal which maps to a inode.
1135 * @inode: An inode to create the journal in
1137 * jbd2_journal_init_inode creates a journal which maps an on-disk inode as
1138 * the journal. The inode must exist already, must support bmap() and
1139 * must have all data blocks preallocated.
1141 journal_t * jbd2_journal_init_inode (struct inode *inode)
1143 struct buffer_head *bh;
1144 journal_t *journal = journal_init_common();
1148 unsigned long long blocknr;
1153 journal->j_dev = journal->j_fs_dev = inode->i_sb->s_bdev;
1154 journal->j_inode = inode;
1155 bdevname(journal->j_dev, journal->j_devname);
1156 p = journal->j_devname;
1157 while ((p = strchr(p, '/')))
1159 p = journal->j_devname + strlen(journal->j_devname);
1160 sprintf(p, "-%lu", journal->j_inode->i_ino);
1162 "journal %p: inode %s/%ld, size %Ld, bits %d, blksize %ld\n",
1163 journal, inode->i_sb->s_id, inode->i_ino,
1164 (long long) inode->i_size,
1165 inode->i_sb->s_blocksize_bits, inode->i_sb->s_blocksize);
1167 journal->j_maxlen = inode->i_size >> inode->i_sb->s_blocksize_bits;
1168 journal->j_blocksize = inode->i_sb->s_blocksize;
1169 jbd2_stats_proc_init(journal);
1171 /* journal descriptor can store up to n blocks -bzzz */
1172 n = journal->j_blocksize / sizeof(journal_block_tag_t);
1173 journal->j_wbufsize = n;
1174 journal->j_wbuf = kmalloc(n * sizeof(struct buffer_head*), GFP_KERNEL);
1175 if (!journal->j_wbuf) {
1176 printk(KERN_ERR "%s: Can't allocate bhs for commit thread\n",
1181 err = jbd2_journal_bmap(journal, 0, &blocknr);
1182 /* If that failed, give up */
1184 printk(KERN_ERR "%s: Cannot locate journal superblock\n",
1189 bh = __getblk(journal->j_dev, blocknr, journal->j_blocksize);
1192 "%s: Cannot get buffer for journal superblock\n",
1196 journal->j_sb_buffer = bh;
1197 journal->j_superblock = (journal_superblock_t *)bh->b_data;
1201 kfree(journal->j_wbuf);
1202 jbd2_stats_proc_exit(journal);
1208 * If the journal init or create aborts, we need to mark the journal
1209 * superblock as being NULL to prevent the journal destroy from writing
1210 * back a bogus superblock.
1212 static void journal_fail_superblock (journal_t *journal)
1214 struct buffer_head *bh = journal->j_sb_buffer;
1216 journal->j_sb_buffer = NULL;
1220 * Given a journal_t structure, initialise the various fields for
1221 * startup of a new journaling session. We use this both when creating
1222 * a journal, and after recovering an old journal to reset it for
1226 static int journal_reset(journal_t *journal)
1228 journal_superblock_t *sb = journal->j_superblock;
1229 unsigned long long first, last;
1231 first = be32_to_cpu(sb->s_first);
1232 last = be32_to_cpu(sb->s_maxlen);
1233 if (first + JBD2_MIN_JOURNAL_BLOCKS > last + 1) {
1234 printk(KERN_ERR "JBD2: Journal too short (blocks %llu-%llu).\n",
1236 journal_fail_superblock(journal);
1240 journal->j_first = first;
1241 journal->j_last = last;
1243 journal->j_head = first;
1244 journal->j_tail = first;
1245 journal->j_free = last - first;
1247 journal->j_tail_sequence = journal->j_transaction_sequence;
1248 journal->j_commit_sequence = journal->j_transaction_sequence - 1;
1249 journal->j_commit_request = journal->j_commit_sequence;
1251 journal->j_max_transaction_buffers = journal->j_maxlen / 4;
1254 * As a special case, if the on-disk copy is already marked as needing
1255 * no recovery (s_start == 0), then we can safely defer the superblock
1256 * update until the next commit by setting JBD2_FLUSHED. This avoids
1257 * attempting a write to a potential-readonly device.
1259 if (sb->s_start == 0) {
1260 jbd_debug(1, "JBD2: Skipping superblock update on recovered sb "
1261 "(start %ld, seq %d, errno %d)\n",
1262 journal->j_tail, journal->j_tail_sequence,
1264 journal->j_flags |= JBD2_FLUSHED;
1266 /* Lock here to make assertions happy... */
1267 mutex_lock(&journal->j_checkpoint_mutex);
1269 * Update log tail information. We use WRITE_FUA since new
1270 * transaction will start reusing journal space and so we
1271 * must make sure information about current log tail is on
1274 jbd2_journal_update_sb_log_tail(journal,
1275 journal->j_tail_sequence,
1278 mutex_unlock(&journal->j_checkpoint_mutex);
1280 return jbd2_journal_start_thread(journal);
1283 static void jbd2_write_superblock(journal_t *journal, int write_op)
1285 struct buffer_head *bh = journal->j_sb_buffer;
1288 trace_jbd2_write_superblock(journal, write_op);
1289 if (!(journal->j_flags & JBD2_BARRIER))
1290 write_op &= ~(REQ_FUA | REQ_FLUSH);
1292 if (buffer_write_io_error(bh)) {
1294 * Oh, dear. A previous attempt to write the journal
1295 * superblock failed. This could happen because the
1296 * USB device was yanked out. Or it could happen to
1297 * be a transient write error and maybe the block will
1298 * be remapped. Nothing we can do but to retry the
1299 * write and hope for the best.
1301 printk(KERN_ERR "JBD2: previous I/O error detected "
1302 "for journal superblock update for %s.\n",
1303 journal->j_devname);
1304 clear_buffer_write_io_error(bh);
1305 set_buffer_uptodate(bh);
1308 bh->b_end_io = end_buffer_write_sync;
1309 ret = submit_bh(write_op, bh);
1311 if (buffer_write_io_error(bh)) {
1312 clear_buffer_write_io_error(bh);
1313 set_buffer_uptodate(bh);
1317 printk(KERN_ERR "JBD2: Error %d detected when updating "
1318 "journal superblock for %s.\n", ret,
1319 journal->j_devname);
1324 * jbd2_journal_update_sb_log_tail() - Update log tail in journal sb on disk.
1325 * @journal: The journal to update.
1326 * @tail_tid: TID of the new transaction at the tail of the log
1327 * @tail_block: The first block of the transaction at the tail of the log
1328 * @write_op: With which operation should we write the journal sb
1330 * Update a journal's superblock information about log tail and write it to
1331 * disk, waiting for the IO to complete.
1333 void jbd2_journal_update_sb_log_tail(journal_t *journal, tid_t tail_tid,
1334 unsigned long tail_block, int write_op)
1336 journal_superblock_t *sb = journal->j_superblock;
1338 BUG_ON(!mutex_is_locked(&journal->j_checkpoint_mutex));
1339 jbd_debug(1, "JBD2: updating superblock (start %lu, seq %u)\n",
1340 tail_block, tail_tid);
1342 sb->s_sequence = cpu_to_be32(tail_tid);
1343 sb->s_start = cpu_to_be32(tail_block);
1345 jbd2_write_superblock(journal, write_op);
1347 /* Log is no longer empty */
1348 write_lock(&journal->j_state_lock);
1349 WARN_ON(!sb->s_sequence);
1350 journal->j_flags &= ~JBD2_FLUSHED;
1351 write_unlock(&journal->j_state_lock);
1355 * jbd2_mark_journal_empty() - Mark on disk journal as empty.
1356 * @journal: The journal to update.
1358 * Update a journal's dynamic superblock fields to show that journal is empty.
1359 * Write updated superblock to disk waiting for IO to complete.
1361 static void jbd2_mark_journal_empty(journal_t *journal)
1363 journal_superblock_t *sb = journal->j_superblock;
1365 BUG_ON(!mutex_is_locked(&journal->j_checkpoint_mutex));
1366 read_lock(&journal->j_state_lock);
1367 /* Is it already empty? */
1368 if (sb->s_start == 0) {
1369 read_unlock(&journal->j_state_lock);
1372 jbd_debug(1, "JBD2: Marking journal as empty (seq %d)\n",
1373 journal->j_tail_sequence);
1375 sb->s_sequence = cpu_to_be32(journal->j_tail_sequence);
1376 sb->s_start = cpu_to_be32(0);
1377 read_unlock(&journal->j_state_lock);
1379 jbd2_write_superblock(journal, WRITE_FUA);
1381 /* Log is no longer empty */
1382 write_lock(&journal->j_state_lock);
1383 journal->j_flags |= JBD2_FLUSHED;
1384 write_unlock(&journal->j_state_lock);
1389 * jbd2_journal_update_sb_errno() - Update error in the journal.
1390 * @journal: The journal to update.
1392 * Update a journal's errno. Write updated superblock to disk waiting for IO
1395 void jbd2_journal_update_sb_errno(journal_t *journal)
1397 journal_superblock_t *sb = journal->j_superblock;
1399 read_lock(&journal->j_state_lock);
1400 jbd_debug(1, "JBD2: updating superblock error (errno %d)\n",
1402 sb->s_errno = cpu_to_be32(journal->j_errno);
1403 jbd2_superblock_csum_set(journal, sb);
1404 read_unlock(&journal->j_state_lock);
1406 jbd2_write_superblock(journal, WRITE_SYNC);
1408 EXPORT_SYMBOL(jbd2_journal_update_sb_errno);
1411 * Read the superblock for a given journal, performing initial
1412 * validation of the format.
1414 static int journal_get_superblock(journal_t *journal)
1416 struct buffer_head *bh;
1417 journal_superblock_t *sb;
1420 bh = journal->j_sb_buffer;
1422 J_ASSERT(bh != NULL);
1423 if (!buffer_uptodate(bh)) {
1424 ll_rw_block(READ, 1, &bh);
1426 if (!buffer_uptodate(bh)) {
1428 "JBD2: IO error reading journal superblock\n");
1433 if (buffer_verified(bh))
1436 sb = journal->j_superblock;
1440 if (sb->s_header.h_magic != cpu_to_be32(JBD2_MAGIC_NUMBER) ||
1441 sb->s_blocksize != cpu_to_be32(journal->j_blocksize)) {
1442 printk(KERN_WARNING "JBD2: no valid journal superblock found\n");
1446 switch(be32_to_cpu(sb->s_header.h_blocktype)) {
1447 case JBD2_SUPERBLOCK_V1:
1448 journal->j_format_version = 1;
1450 case JBD2_SUPERBLOCK_V2:
1451 journal->j_format_version = 2;
1454 printk(KERN_WARNING "JBD2: unrecognised superblock format ID\n");
1458 if (be32_to_cpu(sb->s_maxlen) < journal->j_maxlen)
1459 journal->j_maxlen = be32_to_cpu(sb->s_maxlen);
1460 else if (be32_to_cpu(sb->s_maxlen) > journal->j_maxlen) {
1461 printk(KERN_WARNING "JBD2: journal file too short\n");
1465 if (be32_to_cpu(sb->s_first) == 0 ||
1466 be32_to_cpu(sb->s_first) >= journal->j_maxlen) {
1468 "JBD2: Invalid start block of journal: %u\n",
1469 be32_to_cpu(sb->s_first));
1473 if (JBD2_HAS_COMPAT_FEATURE(journal, JBD2_FEATURE_COMPAT_CHECKSUM) &&
1474 JBD2_HAS_INCOMPAT_FEATURE(journal, JBD2_FEATURE_INCOMPAT_CSUM_V2)) {
1475 /* Can't have checksum v1 and v2 on at the same time! */
1476 printk(KERN_ERR "JBD: Can't enable checksumming v1 and v2 "
1477 "at the same time!\n");
1481 if (!jbd2_verify_csum_type(journal, sb)) {
1482 printk(KERN_ERR "JBD: Unknown checksum type\n");
1486 /* Load the checksum driver */
1487 if (JBD2_HAS_INCOMPAT_FEATURE(journal, JBD2_FEATURE_INCOMPAT_CSUM_V2)) {
1488 journal->j_chksum_driver = crypto_alloc_shash("crc32c", 0, 0);
1489 if (IS_ERR(journal->j_chksum_driver)) {
1490 printk(KERN_ERR "JBD: Cannot load crc32c driver.\n");
1491 err = PTR_ERR(journal->j_chksum_driver);
1492 journal->j_chksum_driver = NULL;
1497 /* Check superblock checksum */
1498 if (!jbd2_superblock_csum_verify(journal, sb)) {
1499 printk(KERN_ERR "JBD: journal checksum error\n");
1503 /* Precompute checksum seed for all metadata */
1504 if (JBD2_HAS_INCOMPAT_FEATURE(journal, JBD2_FEATURE_INCOMPAT_CSUM_V2))
1505 journal->j_csum_seed = jbd2_chksum(journal, ~0, sb->s_uuid,
1506 sizeof(sb->s_uuid));
1508 set_buffer_verified(bh);
1513 journal_fail_superblock(journal);
1518 * Load the on-disk journal superblock and read the key fields into the
1522 static int load_superblock(journal_t *journal)
1525 journal_superblock_t *sb;
1527 err = journal_get_superblock(journal);
1531 sb = journal->j_superblock;
1533 journal->j_tail_sequence = be32_to_cpu(sb->s_sequence);
1534 journal->j_tail = be32_to_cpu(sb->s_start);
1535 journal->j_first = be32_to_cpu(sb->s_first);
1536 journal->j_last = be32_to_cpu(sb->s_maxlen);
1537 journal->j_errno = be32_to_cpu(sb->s_errno);
1544 * int jbd2_journal_load() - Read journal from disk.
1545 * @journal: Journal to act on.
1547 * Given a journal_t structure which tells us which disk blocks contain
1548 * a journal, read the journal from disk to initialise the in-memory
1551 int jbd2_journal_load(journal_t *journal)
1554 journal_superblock_t *sb;
1556 err = load_superblock(journal);
1560 sb = journal->j_superblock;
1561 /* If this is a V2 superblock, then we have to check the
1562 * features flags on it. */
1564 if (journal->j_format_version >= 2) {
1565 if ((sb->s_feature_ro_compat &
1566 ~cpu_to_be32(JBD2_KNOWN_ROCOMPAT_FEATURES)) ||
1567 (sb->s_feature_incompat &
1568 ~cpu_to_be32(JBD2_KNOWN_INCOMPAT_FEATURES))) {
1570 "JBD2: Unrecognised features on journal\n");
1576 * Create a slab for this blocksize
1578 err = jbd2_journal_create_slab(be32_to_cpu(sb->s_blocksize));
1582 /* Let the recovery code check whether it needs to recover any
1583 * data from the journal. */
1584 if (jbd2_journal_recover(journal))
1585 goto recovery_error;
1587 if (journal->j_failed_commit) {
1588 printk(KERN_ERR "JBD2: journal transaction %u on %s "
1589 "is corrupt.\n", journal->j_failed_commit,
1590 journal->j_devname);
1594 /* OK, we've finished with the dynamic journal bits:
1595 * reinitialise the dynamic contents of the superblock in memory
1596 * and reset them on disk. */
1597 if (journal_reset(journal))
1598 goto recovery_error;
1600 journal->j_flags &= ~JBD2_ABORT;
1601 journal->j_flags |= JBD2_LOADED;
1605 printk(KERN_WARNING "JBD2: recovery failed\n");
1610 * void jbd2_journal_destroy() - Release a journal_t structure.
1611 * @journal: Journal to act on.
1613 * Release a journal_t structure once it is no longer in use by the
1615 * Return <0 if we couldn't clean up the journal.
1617 int jbd2_journal_destroy(journal_t *journal)
1621 /* Wait for the commit thread to wake up and die. */
1622 journal_kill_thread(journal);
1624 /* Force a final log commit */
1625 if (journal->j_running_transaction)
1626 jbd2_journal_commit_transaction(journal);
1628 /* Force any old transactions to disk */
1630 /* Totally anal locking here... */
1631 spin_lock(&journal->j_list_lock);
1632 while (journal->j_checkpoint_transactions != NULL) {
1633 spin_unlock(&journal->j_list_lock);
1634 mutex_lock(&journal->j_checkpoint_mutex);
1635 jbd2_log_do_checkpoint(journal);
1636 mutex_unlock(&journal->j_checkpoint_mutex);
1637 spin_lock(&journal->j_list_lock);
1640 J_ASSERT(journal->j_running_transaction == NULL);
1641 J_ASSERT(journal->j_committing_transaction == NULL);
1642 J_ASSERT(journal->j_checkpoint_transactions == NULL);
1643 spin_unlock(&journal->j_list_lock);
1645 if (journal->j_sb_buffer) {
1646 if (!is_journal_aborted(journal)) {
1647 mutex_lock(&journal->j_checkpoint_mutex);
1648 jbd2_mark_journal_empty(journal);
1649 mutex_unlock(&journal->j_checkpoint_mutex);
1652 brelse(journal->j_sb_buffer);
1655 if (journal->j_proc_entry)
1656 jbd2_stats_proc_exit(journal);
1657 if (journal->j_inode)
1658 iput(journal->j_inode);
1659 if (journal->j_revoke)
1660 jbd2_journal_destroy_revoke(journal);
1661 if (journal->j_chksum_driver)
1662 crypto_free_shash(journal->j_chksum_driver);
1663 kfree(journal->j_wbuf);
1671 *int jbd2_journal_check_used_features () - Check if features specified are used.
1672 * @journal: Journal to check.
1673 * @compat: bitmask of compatible features
1674 * @ro: bitmask of features that force read-only mount
1675 * @incompat: bitmask of incompatible features
1677 * Check whether the journal uses all of a given set of
1678 * features. Return true (non-zero) if it does.
1681 int jbd2_journal_check_used_features (journal_t *journal, unsigned long compat,
1682 unsigned long ro, unsigned long incompat)
1684 journal_superblock_t *sb;
1686 if (!compat && !ro && !incompat)
1688 /* Load journal superblock if it is not loaded yet. */
1689 if (journal->j_format_version == 0 &&
1690 journal_get_superblock(journal) != 0)
1692 if (journal->j_format_version == 1)
1695 sb = journal->j_superblock;
1697 if (((be32_to_cpu(sb->s_feature_compat) & compat) == compat) &&
1698 ((be32_to_cpu(sb->s_feature_ro_compat) & ro) == ro) &&
1699 ((be32_to_cpu(sb->s_feature_incompat) & incompat) == incompat))
1706 * int jbd2_journal_check_available_features() - Check feature set in journalling layer
1707 * @journal: Journal to check.
1708 * @compat: bitmask of compatible features
1709 * @ro: bitmask of features that force read-only mount
1710 * @incompat: bitmask of incompatible features
1712 * Check whether the journaling code supports the use of
1713 * all of a given set of features on this journal. Return true
1714 * (non-zero) if it can. */
1716 int jbd2_journal_check_available_features (journal_t *journal, unsigned long compat,
1717 unsigned long ro, unsigned long incompat)
1719 if (!compat && !ro && !incompat)
1722 /* We can support any known requested features iff the
1723 * superblock is in version 2. Otherwise we fail to support any
1724 * extended sb features. */
1726 if (journal->j_format_version != 2)
1729 if ((compat & JBD2_KNOWN_COMPAT_FEATURES) == compat &&
1730 (ro & JBD2_KNOWN_ROCOMPAT_FEATURES) == ro &&
1731 (incompat & JBD2_KNOWN_INCOMPAT_FEATURES) == incompat)
1738 * int jbd2_journal_set_features () - Mark a given journal feature in the superblock
1739 * @journal: Journal to act on.
1740 * @compat: bitmask of compatible features
1741 * @ro: bitmask of features that force read-only mount
1742 * @incompat: bitmask of incompatible features
1744 * Mark a given journal feature as present on the
1745 * superblock. Returns true if the requested features could be set.
1749 int jbd2_journal_set_features (journal_t *journal, unsigned long compat,
1750 unsigned long ro, unsigned long incompat)
1752 #define INCOMPAT_FEATURE_ON(f) \
1753 ((incompat & (f)) && !(sb->s_feature_incompat & cpu_to_be32(f)))
1754 #define COMPAT_FEATURE_ON(f) \
1755 ((compat & (f)) && !(sb->s_feature_compat & cpu_to_be32(f)))
1756 journal_superblock_t *sb;
1758 if (jbd2_journal_check_used_features(journal, compat, ro, incompat))
1761 if (!jbd2_journal_check_available_features(journal, compat, ro, incompat))
1764 /* Asking for checksumming v2 and v1? Only give them v2. */
1765 if (incompat & JBD2_FEATURE_INCOMPAT_CSUM_V2 &&
1766 compat & JBD2_FEATURE_COMPAT_CHECKSUM)
1767 compat &= ~JBD2_FEATURE_COMPAT_CHECKSUM;
1769 jbd_debug(1, "Setting new features 0x%lx/0x%lx/0x%lx\n",
1770 compat, ro, incompat);
1772 sb = journal->j_superblock;
1774 /* If enabling v2 checksums, update superblock */
1775 if (INCOMPAT_FEATURE_ON(JBD2_FEATURE_INCOMPAT_CSUM_V2)) {
1776 sb->s_checksum_type = JBD2_CRC32C_CHKSUM;
1777 sb->s_feature_compat &=
1778 ~cpu_to_be32(JBD2_FEATURE_COMPAT_CHECKSUM);
1780 /* Load the checksum driver */
1781 if (journal->j_chksum_driver == NULL) {
1782 journal->j_chksum_driver = crypto_alloc_shash("crc32c",
1784 if (IS_ERR(journal->j_chksum_driver)) {
1785 printk(KERN_ERR "JBD: Cannot load crc32c "
1787 journal->j_chksum_driver = NULL;
1792 /* Precompute checksum seed for all metadata */
1793 if (JBD2_HAS_INCOMPAT_FEATURE(journal,
1794 JBD2_FEATURE_INCOMPAT_CSUM_V2))
1795 journal->j_csum_seed = jbd2_chksum(journal, ~0,
1797 sizeof(sb->s_uuid));
1800 /* If enabling v1 checksums, downgrade superblock */
1801 if (COMPAT_FEATURE_ON(JBD2_FEATURE_COMPAT_CHECKSUM))
1802 sb->s_feature_incompat &=
1803 ~cpu_to_be32(JBD2_FEATURE_INCOMPAT_CSUM_V2);
1805 sb->s_feature_compat |= cpu_to_be32(compat);
1806 sb->s_feature_ro_compat |= cpu_to_be32(ro);
1807 sb->s_feature_incompat |= cpu_to_be32(incompat);
1810 #undef COMPAT_FEATURE_ON
1811 #undef INCOMPAT_FEATURE_ON
1815 * jbd2_journal_clear_features () - Clear a given journal feature in the
1817 * @journal: Journal to act on.
1818 * @compat: bitmask of compatible features
1819 * @ro: bitmask of features that force read-only mount
1820 * @incompat: bitmask of incompatible features
1822 * Clear a given journal feature as present on the
1825 void jbd2_journal_clear_features(journal_t *journal, unsigned long compat,
1826 unsigned long ro, unsigned long incompat)
1828 journal_superblock_t *sb;
1830 jbd_debug(1, "Clear features 0x%lx/0x%lx/0x%lx\n",
1831 compat, ro, incompat);
1833 sb = journal->j_superblock;
1835 sb->s_feature_compat &= ~cpu_to_be32(compat);
1836 sb->s_feature_ro_compat &= ~cpu_to_be32(ro);
1837 sb->s_feature_incompat &= ~cpu_to_be32(incompat);
1839 EXPORT_SYMBOL(jbd2_journal_clear_features);
1842 * int jbd2_journal_flush () - Flush journal
1843 * @journal: Journal to act on.
1845 * Flush all data for a given journal to disk and empty the journal.
1846 * Filesystems can use this when remounting readonly to ensure that
1847 * recovery does not need to happen on remount.
1850 int jbd2_journal_flush(journal_t *journal)
1853 transaction_t *transaction = NULL;
1855 write_lock(&journal->j_state_lock);
1857 /* Force everything buffered to the log... */
1858 if (journal->j_running_transaction) {
1859 transaction = journal->j_running_transaction;
1860 __jbd2_log_start_commit(journal, transaction->t_tid);
1861 } else if (journal->j_committing_transaction)
1862 transaction = journal->j_committing_transaction;
1864 /* Wait for the log commit to complete... */
1866 tid_t tid = transaction->t_tid;
1868 write_unlock(&journal->j_state_lock);
1869 jbd2_log_wait_commit(journal, tid);
1871 write_unlock(&journal->j_state_lock);
1874 /* ...and flush everything in the log out to disk. */
1875 spin_lock(&journal->j_list_lock);
1876 while (!err && journal->j_checkpoint_transactions != NULL) {
1877 spin_unlock(&journal->j_list_lock);
1878 mutex_lock(&journal->j_checkpoint_mutex);
1879 err = jbd2_log_do_checkpoint(journal);
1880 mutex_unlock(&journal->j_checkpoint_mutex);
1881 spin_lock(&journal->j_list_lock);
1883 spin_unlock(&journal->j_list_lock);
1885 if (is_journal_aborted(journal))
1888 mutex_lock(&journal->j_checkpoint_mutex);
1889 jbd2_cleanup_journal_tail(journal);
1891 /* Finally, mark the journal as really needing no recovery.
1892 * This sets s_start==0 in the underlying superblock, which is
1893 * the magic code for a fully-recovered superblock. Any future
1894 * commits of data to the journal will restore the current
1896 jbd2_mark_journal_empty(journal);
1897 mutex_unlock(&journal->j_checkpoint_mutex);
1898 write_lock(&journal->j_state_lock);
1899 J_ASSERT(!journal->j_running_transaction);
1900 J_ASSERT(!journal->j_committing_transaction);
1901 J_ASSERT(!journal->j_checkpoint_transactions);
1902 J_ASSERT(journal->j_head == journal->j_tail);
1903 J_ASSERT(journal->j_tail_sequence == journal->j_transaction_sequence);
1904 write_unlock(&journal->j_state_lock);
1909 * int jbd2_journal_wipe() - Wipe journal contents
1910 * @journal: Journal to act on.
1911 * @write: flag (see below)
1913 * Wipe out all of the contents of a journal, safely. This will produce
1914 * a warning if the journal contains any valid recovery information.
1915 * Must be called between journal_init_*() and jbd2_journal_load().
1917 * If 'write' is non-zero, then we wipe out the journal on disk; otherwise
1918 * we merely suppress recovery.
1921 int jbd2_journal_wipe(journal_t *journal, int write)
1925 J_ASSERT (!(journal->j_flags & JBD2_LOADED));
1927 err = load_superblock(journal);
1931 if (!journal->j_tail)
1934 printk(KERN_WARNING "JBD2: %s recovery information on journal\n",
1935 write ? "Clearing" : "Ignoring");
1937 err = jbd2_journal_skip_recovery(journal);
1939 /* Lock to make assertions happy... */
1940 mutex_lock(&journal->j_checkpoint_mutex);
1941 jbd2_mark_journal_empty(journal);
1942 mutex_unlock(&journal->j_checkpoint_mutex);
1950 * Journal abort has very specific semantics, which we describe
1951 * for journal abort.
1953 * Two internal functions, which provide abort to the jbd layer
1958 * Quick version for internal journal use (doesn't lock the journal).
1959 * Aborts hard --- we mark the abort as occurred, but do _nothing_ else,
1960 * and don't attempt to make any other journal updates.
1962 void __jbd2_journal_abort_hard(journal_t *journal)
1964 transaction_t *transaction;
1966 if (journal->j_flags & JBD2_ABORT)
1969 printk(KERN_ERR "Aborting journal on device %s.\n",
1970 journal->j_devname);
1972 write_lock(&journal->j_state_lock);
1973 journal->j_flags |= JBD2_ABORT;
1974 transaction = journal->j_running_transaction;
1976 __jbd2_log_start_commit(journal, transaction->t_tid);
1977 write_unlock(&journal->j_state_lock);
1980 /* Soft abort: record the abort error status in the journal superblock,
1981 * but don't do any other IO. */
1982 static void __journal_abort_soft (journal_t *journal, int errno)
1984 if (journal->j_flags & JBD2_ABORT)
1987 if (!journal->j_errno)
1988 journal->j_errno = errno;
1990 __jbd2_journal_abort_hard(journal);
1993 jbd2_journal_update_sb_errno(journal);
1997 * void jbd2_journal_abort () - Shutdown the journal immediately.
1998 * @journal: the journal to shutdown.
1999 * @errno: an error number to record in the journal indicating
2000 * the reason for the shutdown.
2002 * Perform a complete, immediate shutdown of the ENTIRE
2003 * journal (not of a single transaction). This operation cannot be
2004 * undone without closing and reopening the journal.
2006 * The jbd2_journal_abort function is intended to support higher level error
2007 * recovery mechanisms such as the ext2/ext3 remount-readonly error
2010 * Journal abort has very specific semantics. Any existing dirty,
2011 * unjournaled buffers in the main filesystem will still be written to
2012 * disk by bdflush, but the journaling mechanism will be suspended
2013 * immediately and no further transaction commits will be honoured.
2015 * Any dirty, journaled buffers will be written back to disk without
2016 * hitting the journal. Atomicity cannot be guaranteed on an aborted
2017 * filesystem, but we _do_ attempt to leave as much data as possible
2018 * behind for fsck to use for cleanup.
2020 * Any attempt to get a new transaction handle on a journal which is in
2021 * ABORT state will just result in an -EROFS error return. A
2022 * jbd2_journal_stop on an existing handle will return -EIO if we have
2023 * entered abort state during the update.
2025 * Recursive transactions are not disturbed by journal abort until the
2026 * final jbd2_journal_stop, which will receive the -EIO error.
2028 * Finally, the jbd2_journal_abort call allows the caller to supply an errno
2029 * which will be recorded (if possible) in the journal superblock. This
2030 * allows a client to record failure conditions in the middle of a
2031 * transaction without having to complete the transaction to record the
2032 * failure to disk. ext3_error, for example, now uses this
2035 * Errors which originate from within the journaling layer will NOT
2036 * supply an errno; a null errno implies that absolutely no further
2037 * writes are done to the journal (unless there are any already in
2042 void jbd2_journal_abort(journal_t *journal, int errno)
2044 __journal_abort_soft(journal, errno);
2048 * int jbd2_journal_errno () - returns the journal's error state.
2049 * @journal: journal to examine.
2051 * This is the errno number set with jbd2_journal_abort(), the last
2052 * time the journal was mounted - if the journal was stopped
2053 * without calling abort this will be 0.
2055 * If the journal has been aborted on this mount time -EROFS will
2058 int jbd2_journal_errno(journal_t *journal)
2062 read_lock(&journal->j_state_lock);
2063 if (journal->j_flags & JBD2_ABORT)
2066 err = journal->j_errno;
2067 read_unlock(&journal->j_state_lock);
2072 * int jbd2_journal_clear_err () - clears the journal's error state
2073 * @journal: journal to act on.
2075 * An error must be cleared or acked to take a FS out of readonly
2078 int jbd2_journal_clear_err(journal_t *journal)
2082 write_lock(&journal->j_state_lock);
2083 if (journal->j_flags & JBD2_ABORT)
2086 journal->j_errno = 0;
2087 write_unlock(&journal->j_state_lock);
2092 * void jbd2_journal_ack_err() - Ack journal err.
2093 * @journal: journal to act on.
2095 * An error must be cleared or acked to take a FS out of readonly
2098 void jbd2_journal_ack_err(journal_t *journal)
2100 write_lock(&journal->j_state_lock);
2101 if (journal->j_errno)
2102 journal->j_flags |= JBD2_ACK_ERR;
2103 write_unlock(&journal->j_state_lock);
2106 int jbd2_journal_blocks_per_page(struct inode *inode)
2108 return 1 << (PAGE_CACHE_SHIFT - inode->i_sb->s_blocksize_bits);
2112 * helper functions to deal with 32 or 64bit block numbers.
2114 size_t journal_tag_bytes(journal_t *journal)
2116 journal_block_tag_t tag;
2119 if (JBD2_HAS_INCOMPAT_FEATURE(journal, JBD2_FEATURE_INCOMPAT_CSUM_V2))
2120 x += sizeof(tag.t_checksum);
2122 if (JBD2_HAS_INCOMPAT_FEATURE(journal, JBD2_FEATURE_INCOMPAT_64BIT))
2123 return x + JBD2_TAG_SIZE64;
2125 return x + JBD2_TAG_SIZE32;
2129 * JBD memory management
2131 * These functions are used to allocate block-sized chunks of memory
2132 * used for making copies of buffer_head data. Very often it will be
2133 * page-sized chunks of data, but sometimes it will be in
2134 * sub-page-size chunks. (For example, 16k pages on Power systems
2135 * with a 4k block file system.) For blocks smaller than a page, we
2136 * use a SLAB allocator. There are slab caches for each block size,
2137 * which are allocated at mount time, if necessary, and we only free
2138 * (all of) the slab caches when/if the jbd2 module is unloaded. For
2139 * this reason we don't need to a mutex to protect access to
2140 * jbd2_slab[] allocating or releasing memory; only in
2141 * jbd2_journal_create_slab().
2143 #define JBD2_MAX_SLABS 8
2144 static struct kmem_cache *jbd2_slab[JBD2_MAX_SLABS];
2146 static const char *jbd2_slab_names[JBD2_MAX_SLABS] = {
2147 "jbd2_1k", "jbd2_2k", "jbd2_4k", "jbd2_8k",
2148 "jbd2_16k", "jbd2_32k", "jbd2_64k", "jbd2_128k"
2152 static void jbd2_journal_destroy_slabs(void)
2156 for (i = 0; i < JBD2_MAX_SLABS; i++) {
2158 kmem_cache_destroy(jbd2_slab[i]);
2159 jbd2_slab[i] = NULL;
2163 static int jbd2_journal_create_slab(size_t size)
2165 static DEFINE_MUTEX(jbd2_slab_create_mutex);
2166 int i = order_base_2(size) - 10;
2169 if (size == PAGE_SIZE)
2172 if (i >= JBD2_MAX_SLABS)
2175 if (unlikely(i < 0))
2177 mutex_lock(&jbd2_slab_create_mutex);
2179 mutex_unlock(&jbd2_slab_create_mutex);
2180 return 0; /* Already created */
2183 slab_size = 1 << (i+10);
2184 jbd2_slab[i] = kmem_cache_create(jbd2_slab_names[i], slab_size,
2185 slab_size, 0, NULL);
2186 mutex_unlock(&jbd2_slab_create_mutex);
2187 if (!jbd2_slab[i]) {
2188 printk(KERN_EMERG "JBD2: no memory for jbd2_slab cache\n");
2194 static struct kmem_cache *get_slab(size_t size)
2196 int i = order_base_2(size) - 10;
2198 BUG_ON(i >= JBD2_MAX_SLABS);
2199 if (unlikely(i < 0))
2201 BUG_ON(jbd2_slab[i] == NULL);
2202 return jbd2_slab[i];
2205 void *jbd2_alloc(size_t size, gfp_t flags)
2209 BUG_ON(size & (size-1)); /* Must be a power of 2 */
2211 flags |= __GFP_REPEAT;
2212 if (size == PAGE_SIZE)
2213 ptr = (void *)__get_free_pages(flags, 0);
2214 else if (size > PAGE_SIZE) {
2215 int order = get_order(size);
2218 ptr = (void *)__get_free_pages(flags, order);
2220 ptr = vmalloc(size);
2222 ptr = kmem_cache_alloc(get_slab(size), flags);
2224 /* Check alignment; SLUB has gotten this wrong in the past,
2225 * and this can lead to user data corruption! */
2226 BUG_ON(((unsigned long) ptr) & (size-1));
2231 void jbd2_free(void *ptr, size_t size)
2233 if (size == PAGE_SIZE) {
2234 free_pages((unsigned long)ptr, 0);
2237 if (size > PAGE_SIZE) {
2238 int order = get_order(size);
2241 free_pages((unsigned long)ptr, order);
2246 kmem_cache_free(get_slab(size), ptr);
2250 * Journal_head storage management
2252 static struct kmem_cache *jbd2_journal_head_cache;
2253 #ifdef CONFIG_JBD2_DEBUG
2254 static atomic_t nr_journal_heads = ATOMIC_INIT(0);
2257 static int jbd2_journal_init_journal_head_cache(void)
2261 J_ASSERT(jbd2_journal_head_cache == NULL);
2262 jbd2_journal_head_cache = kmem_cache_create("jbd2_journal_head",
2263 sizeof(struct journal_head),
2265 SLAB_TEMPORARY, /* flags */
2268 if (!jbd2_journal_head_cache) {
2270 printk(KERN_EMERG "JBD2: no memory for journal_head cache\n");
2275 static void jbd2_journal_destroy_journal_head_cache(void)
2277 if (jbd2_journal_head_cache) {
2278 kmem_cache_destroy(jbd2_journal_head_cache);
2279 jbd2_journal_head_cache = NULL;
2284 * journal_head splicing and dicing
2286 static struct journal_head *journal_alloc_journal_head(void)
2288 struct journal_head *ret;
2290 #ifdef CONFIG_JBD2_DEBUG
2291 atomic_inc(&nr_journal_heads);
2293 ret = kmem_cache_zalloc(jbd2_journal_head_cache, GFP_NOFS);
2295 jbd_debug(1, "out of memory for journal_head\n");
2296 pr_notice_ratelimited("ENOMEM in %s, retrying.\n", __func__);
2299 ret = kmem_cache_zalloc(jbd2_journal_head_cache, GFP_NOFS);
2305 static void journal_free_journal_head(struct journal_head *jh)
2307 #ifdef CONFIG_JBD2_DEBUG
2308 atomic_dec(&nr_journal_heads);
2309 memset(jh, JBD2_POISON_FREE, sizeof(*jh));
2311 kmem_cache_free(jbd2_journal_head_cache, jh);
2315 * A journal_head is attached to a buffer_head whenever JBD has an
2316 * interest in the buffer.
2318 * Whenever a buffer has an attached journal_head, its ->b_state:BH_JBD bit
2319 * is set. This bit is tested in core kernel code where we need to take
2320 * JBD-specific actions. Testing the zeroness of ->b_private is not reliable
2323 * When a buffer has its BH_JBD bit set, its ->b_count is elevated by one.
2325 * When a buffer has its BH_JBD bit set it is immune from being released by
2326 * core kernel code, mainly via ->b_count.
2328 * A journal_head is detached from its buffer_head when the journal_head's
2329 * b_jcount reaches zero. Running transaction (b_transaction) and checkpoint
2330 * transaction (b_cp_transaction) hold their references to b_jcount.
2332 * Various places in the kernel want to attach a journal_head to a buffer_head
2333 * _before_ attaching the journal_head to a transaction. To protect the
2334 * journal_head in this situation, jbd2_journal_add_journal_head elevates the
2335 * journal_head's b_jcount refcount by one. The caller must call
2336 * jbd2_journal_put_journal_head() to undo this.
2338 * So the typical usage would be:
2340 * (Attach a journal_head if needed. Increments b_jcount)
2341 * struct journal_head *jh = jbd2_journal_add_journal_head(bh);
2343 * (Get another reference for transaction)
2344 * jbd2_journal_grab_journal_head(bh);
2345 * jh->b_transaction = xxx;
2346 * (Put original reference)
2347 * jbd2_journal_put_journal_head(jh);
2351 * Give a buffer_head a journal_head.
2355 struct journal_head *jbd2_journal_add_journal_head(struct buffer_head *bh)
2357 struct journal_head *jh;
2358 struct journal_head *new_jh = NULL;
2361 if (!buffer_jbd(bh))
2362 new_jh = journal_alloc_journal_head();
2364 jbd_lock_bh_journal_head(bh);
2365 if (buffer_jbd(bh)) {
2369 (atomic_read(&bh->b_count) > 0) ||
2370 (bh->b_page && bh->b_page->mapping));
2373 jbd_unlock_bh_journal_head(bh);
2378 new_jh = NULL; /* We consumed it */
2383 BUFFER_TRACE(bh, "added journal_head");
2386 jbd_unlock_bh_journal_head(bh);
2388 journal_free_journal_head(new_jh);
2389 return bh->b_private;
2393 * Grab a ref against this buffer_head's journal_head. If it ended up not
2394 * having a journal_head, return NULL
2396 struct journal_head *jbd2_journal_grab_journal_head(struct buffer_head *bh)
2398 struct journal_head *jh = NULL;
2400 jbd_lock_bh_journal_head(bh);
2401 if (buffer_jbd(bh)) {
2405 jbd_unlock_bh_journal_head(bh);
2409 static void __journal_remove_journal_head(struct buffer_head *bh)
2411 struct journal_head *jh = bh2jh(bh);
2413 J_ASSERT_JH(jh, jh->b_jcount >= 0);
2414 J_ASSERT_JH(jh, jh->b_transaction == NULL);
2415 J_ASSERT_JH(jh, jh->b_next_transaction == NULL);
2416 J_ASSERT_JH(jh, jh->b_cp_transaction == NULL);
2417 J_ASSERT_JH(jh, jh->b_jlist == BJ_None);
2418 J_ASSERT_BH(bh, buffer_jbd(bh));
2419 J_ASSERT_BH(bh, jh2bh(jh) == bh);
2420 BUFFER_TRACE(bh, "remove journal_head");
2421 if (jh->b_frozen_data) {
2422 printk(KERN_WARNING "%s: freeing b_frozen_data\n", __func__);
2423 jbd2_free(jh->b_frozen_data, bh->b_size);
2425 if (jh->b_committed_data) {
2426 printk(KERN_WARNING "%s: freeing b_committed_data\n", __func__);
2427 jbd2_free(jh->b_committed_data, bh->b_size);
2429 bh->b_private = NULL;
2430 jh->b_bh = NULL; /* debug, really */
2431 clear_buffer_jbd(bh);
2432 journal_free_journal_head(jh);
2436 * Drop a reference on the passed journal_head. If it fell to zero then
2437 * release the journal_head from the buffer_head.
2439 void jbd2_journal_put_journal_head(struct journal_head *jh)
2441 struct buffer_head *bh = jh2bh(jh);
2443 jbd_lock_bh_journal_head(bh);
2444 J_ASSERT_JH(jh, jh->b_jcount > 0);
2446 if (!jh->b_jcount) {
2447 __journal_remove_journal_head(bh);
2448 jbd_unlock_bh_journal_head(bh);
2451 jbd_unlock_bh_journal_head(bh);
2455 * Initialize jbd inode head
2457 void jbd2_journal_init_jbd_inode(struct jbd2_inode *jinode, struct inode *inode)
2459 jinode->i_transaction = NULL;
2460 jinode->i_next_transaction = NULL;
2461 jinode->i_vfs_inode = inode;
2462 jinode->i_flags = 0;
2463 INIT_LIST_HEAD(&jinode->i_list);
2467 * Function to be called before we start removing inode from memory (i.e.,
2468 * clear_inode() is a fine place to be called from). It removes inode from
2469 * transaction's lists.
2471 void jbd2_journal_release_jbd_inode(journal_t *journal,
2472 struct jbd2_inode *jinode)
2477 spin_lock(&journal->j_list_lock);
2478 /* Is commit writing out inode - we have to wait */
2479 if (test_bit(__JI_COMMIT_RUNNING, &jinode->i_flags)) {
2480 wait_queue_head_t *wq;
2481 DEFINE_WAIT_BIT(wait, &jinode->i_flags, __JI_COMMIT_RUNNING);
2482 wq = bit_waitqueue(&jinode->i_flags, __JI_COMMIT_RUNNING);
2483 prepare_to_wait(wq, &wait.wait, TASK_UNINTERRUPTIBLE);
2484 spin_unlock(&journal->j_list_lock);
2486 finish_wait(wq, &wait.wait);
2490 if (jinode->i_transaction) {
2491 list_del(&jinode->i_list);
2492 jinode->i_transaction = NULL;
2494 spin_unlock(&journal->j_list_lock);
2498 #ifdef CONFIG_PROC_FS
2500 #define JBD2_STATS_PROC_NAME "fs/jbd2"
2502 static void __init jbd2_create_jbd_stats_proc_entry(void)
2504 proc_jbd2_stats = proc_mkdir(JBD2_STATS_PROC_NAME, NULL);
2507 static void __exit jbd2_remove_jbd_stats_proc_entry(void)
2509 if (proc_jbd2_stats)
2510 remove_proc_entry(JBD2_STATS_PROC_NAME, NULL);
2515 #define jbd2_create_jbd_stats_proc_entry() do {} while (0)
2516 #define jbd2_remove_jbd_stats_proc_entry() do {} while (0)
2520 struct kmem_cache *jbd2_handle_cache, *jbd2_inode_cache;
2522 static int __init jbd2_journal_init_handle_cache(void)
2524 jbd2_handle_cache = KMEM_CACHE(jbd2_journal_handle, SLAB_TEMPORARY);
2525 if (jbd2_handle_cache == NULL) {
2526 printk(KERN_EMERG "JBD2: failed to create handle cache\n");
2529 jbd2_inode_cache = KMEM_CACHE(jbd2_inode, 0);
2530 if (jbd2_inode_cache == NULL) {
2531 printk(KERN_EMERG "JBD2: failed to create inode cache\n");
2532 kmem_cache_destroy(jbd2_handle_cache);
2538 static void jbd2_journal_destroy_handle_cache(void)
2540 if (jbd2_handle_cache)
2541 kmem_cache_destroy(jbd2_handle_cache);
2542 if (jbd2_inode_cache)
2543 kmem_cache_destroy(jbd2_inode_cache);
2548 * Module startup and shutdown
2551 static int __init journal_init_caches(void)
2555 ret = jbd2_journal_init_revoke_caches();
2557 ret = jbd2_journal_init_journal_head_cache();
2559 ret = jbd2_journal_init_handle_cache();
2561 ret = jbd2_journal_init_transaction_cache();
2565 static void jbd2_journal_destroy_caches(void)
2567 jbd2_journal_destroy_revoke_caches();
2568 jbd2_journal_destroy_journal_head_cache();
2569 jbd2_journal_destroy_handle_cache();
2570 jbd2_journal_destroy_transaction_cache();
2571 jbd2_journal_destroy_slabs();
2574 static int __init journal_init(void)
2578 BUILD_BUG_ON(sizeof(struct journal_superblock_s) != 1024);
2580 ret = journal_init_caches();
2582 jbd2_create_jbd_stats_proc_entry();
2584 jbd2_journal_destroy_caches();
2589 static void __exit journal_exit(void)
2591 #ifdef CONFIG_JBD2_DEBUG
2592 int n = atomic_read(&nr_journal_heads);
2594 printk(KERN_EMERG "JBD2: leaked %d journal_heads!\n", n);
2596 jbd2_remove_jbd_stats_proc_entry();
2597 jbd2_journal_destroy_caches();
2600 MODULE_LICENSE("GPL");
2601 module_init(journal_init);
2602 module_exit(journal_exit);