1 // SPDX-License-Identifier: GPL-2.0+
3 * linux/fs/jbd2/journal.c
5 * Written by Stephen C. Tweedie <sct@redhat.com>, 1998
7 * Copyright 1998 Red Hat corp --- All Rights Reserved
9 * Generic filesystem journal-writing code; part of the ext2fs
12 * This file manages journals: areas of disk reserved for logging
13 * transactional updates. This includes the kernel journaling thread
14 * which is responsible for scheduling updates to the log.
16 * We do not actually manage the physical storage of the journal in this
17 * file: that is left to a per-journal policy function, which allows us
18 * to store the journal within a filesystem-specified area for ext2
19 * journaling (ext2 can use a reserved inode for storing the log).
22 #include <linux/module.h>
23 #include <linux/time.h>
25 #include <linux/jbd2.h>
26 #include <linux/errno.h>
27 #include <linux/slab.h>
28 #include <linux/init.h>
30 #include <linux/freezer.h>
31 #include <linux/pagemap.h>
32 #include <linux/kthread.h>
33 #include <linux/poison.h>
34 #include <linux/proc_fs.h>
35 #include <linux/seq_file.h>
36 #include <linux/math64.h>
37 #include <linux/hash.h>
38 #include <linux/log2.h>
39 #include <linux/vmalloc.h>
40 #include <linux/backing-dev.h>
41 #include <linux/bitops.h>
42 #include <linux/ratelimit.h>
43 #include <linux/sched/mm.h>
45 #define CREATE_TRACE_POINTS
46 #include <trace/events/jbd2.h>
48 #include <linux/uaccess.h>
51 #ifdef CONFIG_JBD2_DEBUG
52 ushort jbd2_journal_enable_debug __read_mostly;
53 EXPORT_SYMBOL(jbd2_journal_enable_debug);
55 module_param_named(jbd2_debug, jbd2_journal_enable_debug, ushort, 0644);
56 MODULE_PARM_DESC(jbd2_debug, "Debugging level for jbd2");
59 EXPORT_SYMBOL(jbd2_journal_extend);
60 EXPORT_SYMBOL(jbd2_journal_stop);
61 EXPORT_SYMBOL(jbd2_journal_lock_updates);
62 EXPORT_SYMBOL(jbd2_journal_unlock_updates);
63 EXPORT_SYMBOL(jbd2_journal_get_write_access);
64 EXPORT_SYMBOL(jbd2_journal_get_create_access);
65 EXPORT_SYMBOL(jbd2_journal_get_undo_access);
66 EXPORT_SYMBOL(jbd2_journal_set_triggers);
67 EXPORT_SYMBOL(jbd2_journal_dirty_metadata);
68 EXPORT_SYMBOL(jbd2_journal_forget);
69 EXPORT_SYMBOL(jbd2_journal_flush);
70 EXPORT_SYMBOL(jbd2_journal_revoke);
72 EXPORT_SYMBOL(jbd2_journal_init_dev);
73 EXPORT_SYMBOL(jbd2_journal_init_inode);
74 EXPORT_SYMBOL(jbd2_journal_check_used_features);
75 EXPORT_SYMBOL(jbd2_journal_check_available_features);
76 EXPORT_SYMBOL(jbd2_journal_set_features);
77 EXPORT_SYMBOL(jbd2_journal_load);
78 EXPORT_SYMBOL(jbd2_journal_destroy);
79 EXPORT_SYMBOL(jbd2_journal_abort);
80 EXPORT_SYMBOL(jbd2_journal_errno);
81 EXPORT_SYMBOL(jbd2_journal_ack_err);
82 EXPORT_SYMBOL(jbd2_journal_clear_err);
83 EXPORT_SYMBOL(jbd2_log_wait_commit);
84 EXPORT_SYMBOL(jbd2_log_start_commit);
85 EXPORT_SYMBOL(jbd2_journal_start_commit);
86 EXPORT_SYMBOL(jbd2_journal_force_commit_nested);
87 EXPORT_SYMBOL(jbd2_journal_wipe);
88 EXPORT_SYMBOL(jbd2_journal_blocks_per_page);
89 EXPORT_SYMBOL(jbd2_journal_invalidatepage);
90 EXPORT_SYMBOL(jbd2_journal_try_to_free_buffers);
91 EXPORT_SYMBOL(jbd2_journal_force_commit);
92 EXPORT_SYMBOL(jbd2_journal_inode_ranged_write);
93 EXPORT_SYMBOL(jbd2_journal_inode_ranged_wait);
94 EXPORT_SYMBOL(jbd2_journal_init_jbd_inode);
95 EXPORT_SYMBOL(jbd2_journal_release_jbd_inode);
96 EXPORT_SYMBOL(jbd2_journal_begin_ordered_truncate);
97 EXPORT_SYMBOL(jbd2_inode_cache);
99 static void __journal_abort_soft (journal_t *journal, int errno);
100 static int jbd2_journal_create_slab(size_t slab_size);
102 #ifdef CONFIG_JBD2_DEBUG
103 void __jbd2_debug(int level, const char *file, const char *func,
104 unsigned int line, const char *fmt, ...)
106 struct va_format vaf;
109 if (level > jbd2_journal_enable_debug)
114 printk(KERN_DEBUG "%s: (%s, %u): %pV", file, func, line, &vaf);
117 EXPORT_SYMBOL(__jbd2_debug);
120 /* Checksumming functions */
121 static int jbd2_verify_csum_type(journal_t *j, journal_superblock_t *sb)
123 if (!jbd2_journal_has_csum_v2or3_feature(j))
126 return sb->s_checksum_type == JBD2_CRC32C_CHKSUM;
129 static __be32 jbd2_superblock_csum(journal_t *j, journal_superblock_t *sb)
134 old_csum = sb->s_checksum;
136 csum = jbd2_chksum(j, ~0, (char *)sb, sizeof(journal_superblock_t));
137 sb->s_checksum = old_csum;
139 return cpu_to_be32(csum);
143 * Helper function used to manage commit timeouts
146 static void commit_timeout(struct timer_list *t)
148 journal_t *journal = from_timer(journal, t, j_commit_timer);
150 wake_up_process(journal->j_task);
154 * kjournald2: The main thread function used to manage a logging device
157 * This kernel thread is responsible for two things:
159 * 1) COMMIT: Every so often we need to commit the current state of the
160 * filesystem to disk. The journal thread is responsible for writing
161 * all of the metadata buffers to disk.
163 * 2) CHECKPOINT: We cannot reuse a used section of the log file until all
164 * of the data in that part of the log has been rewritten elsewhere on
165 * the disk. Flushing these old buffers to reclaim space in the log is
166 * known as checkpointing, and this thread is responsible for that job.
169 static int kjournald2(void *arg)
171 journal_t *journal = arg;
172 transaction_t *transaction;
175 * Set up an interval timer which can be used to trigger a commit wakeup
176 * after the commit interval expires
178 timer_setup(&journal->j_commit_timer, commit_timeout, 0);
182 /* Record that the journal thread is running */
183 journal->j_task = current;
184 wake_up(&journal->j_wait_done_commit);
187 * Make sure that no allocations from this kernel thread will ever
188 * recurse to the fs layer because we are responsible for the
189 * transaction commit and any fs involvement might get stuck waiting for
192 memalloc_nofs_save();
195 * And now, wait forever for commit wakeup events.
197 write_lock(&journal->j_state_lock);
200 if (journal->j_flags & JBD2_UNMOUNT)
203 jbd_debug(1, "commit_sequence=%u, commit_request=%u\n",
204 journal->j_commit_sequence, journal->j_commit_request);
206 if (journal->j_commit_sequence != journal->j_commit_request) {
207 jbd_debug(1, "OK, requests differ\n");
208 write_unlock(&journal->j_state_lock);
209 del_timer_sync(&journal->j_commit_timer);
210 jbd2_journal_commit_transaction(journal);
211 write_lock(&journal->j_state_lock);
215 wake_up(&journal->j_wait_done_commit);
216 if (freezing(current)) {
218 * The simpler the better. Flushing journal isn't a
219 * good idea, because that depends on threads that may
220 * be already stopped.
222 jbd_debug(1, "Now suspending kjournald2\n");
223 write_unlock(&journal->j_state_lock);
225 write_lock(&journal->j_state_lock);
228 * We assume on resume that commits are already there,
232 int should_sleep = 1;
234 prepare_to_wait(&journal->j_wait_commit, &wait,
236 if (journal->j_commit_sequence != journal->j_commit_request)
238 transaction = journal->j_running_transaction;
239 if (transaction && time_after_eq(jiffies,
240 transaction->t_expires))
242 if (journal->j_flags & JBD2_UNMOUNT)
245 write_unlock(&journal->j_state_lock);
247 write_lock(&journal->j_state_lock);
249 finish_wait(&journal->j_wait_commit, &wait);
252 jbd_debug(1, "kjournald2 wakes\n");
255 * Were we woken up by a commit wakeup event?
257 transaction = journal->j_running_transaction;
258 if (transaction && time_after_eq(jiffies, transaction->t_expires)) {
259 journal->j_commit_request = transaction->t_tid;
260 jbd_debug(1, "woke because of timeout\n");
265 del_timer_sync(&journal->j_commit_timer);
266 journal->j_task = NULL;
267 wake_up(&journal->j_wait_done_commit);
268 jbd_debug(1, "Journal thread exiting.\n");
269 write_unlock(&journal->j_state_lock);
273 static int jbd2_journal_start_thread(journal_t *journal)
275 struct task_struct *t;
277 t = kthread_run(kjournald2, journal, "jbd2/%s",
282 wait_event(journal->j_wait_done_commit, journal->j_task != NULL);
286 static void journal_kill_thread(journal_t *journal)
288 write_lock(&journal->j_state_lock);
289 journal->j_flags |= JBD2_UNMOUNT;
291 while (journal->j_task) {
292 write_unlock(&journal->j_state_lock);
293 wake_up(&journal->j_wait_commit);
294 wait_event(journal->j_wait_done_commit, journal->j_task == NULL);
295 write_lock(&journal->j_state_lock);
297 write_unlock(&journal->j_state_lock);
301 * jbd2_journal_write_metadata_buffer: write a metadata buffer to the journal.
303 * Writes a metadata buffer to a given disk block. The actual IO is not
304 * performed but a new buffer_head is constructed which labels the data
305 * to be written with the correct destination disk block.
307 * Any magic-number escaping which needs to be done will cause a
308 * copy-out here. If the buffer happens to start with the
309 * JBD2_MAGIC_NUMBER, then we can't write it to the log directly: the
310 * magic number is only written to the log for descripter blocks. In
311 * this case, we copy the data and replace the first word with 0, and we
312 * return a result code which indicates that this buffer needs to be
313 * marked as an escaped buffer in the corresponding log descriptor
314 * block. The missing word can then be restored when the block is read
317 * If the source buffer has already been modified by a new transaction
318 * since we took the last commit snapshot, we use the frozen copy of
319 * that data for IO. If we end up using the existing buffer_head's data
320 * for the write, then we have to make sure nobody modifies it while the
321 * IO is in progress. do_get_write_access() handles this.
323 * The function returns a pointer to the buffer_head to be used for IO.
331 * Bit 0 set == escape performed on the data
332 * Bit 1 set == buffer copy-out performed (kfree the data after IO)
335 int jbd2_journal_write_metadata_buffer(transaction_t *transaction,
336 struct journal_head *jh_in,
337 struct buffer_head **bh_out,
340 int need_copy_out = 0;
341 int done_copy_out = 0;
344 struct buffer_head *new_bh;
345 struct page *new_page;
346 unsigned int new_offset;
347 struct buffer_head *bh_in = jh2bh(jh_in);
348 journal_t *journal = transaction->t_journal;
351 * The buffer really shouldn't be locked: only the current committing
352 * transaction is allowed to write it, so nobody else is allowed
355 * akpm: except if we're journalling data, and write() output is
356 * also part of a shared mapping, and another thread has
357 * decided to launch a writepage() against this buffer.
359 J_ASSERT_BH(bh_in, buffer_jbddirty(bh_in));
361 new_bh = alloc_buffer_head(GFP_NOFS|__GFP_NOFAIL);
363 /* keep subsequent assertions sane */
364 atomic_set(&new_bh->b_count, 1);
366 spin_lock(&jh_in->b_state_lock);
369 * If a new transaction has already done a buffer copy-out, then
370 * we use that version of the data for the commit.
372 if (jh_in->b_frozen_data) {
374 new_page = virt_to_page(jh_in->b_frozen_data);
375 new_offset = offset_in_page(jh_in->b_frozen_data);
377 new_page = jh2bh(jh_in)->b_page;
378 new_offset = offset_in_page(jh2bh(jh_in)->b_data);
381 mapped_data = kmap_atomic(new_page);
383 * Fire data frozen trigger if data already wasn't frozen. Do this
384 * before checking for escaping, as the trigger may modify the magic
385 * offset. If a copy-out happens afterwards, it will have the correct
386 * data in the buffer.
389 jbd2_buffer_frozen_trigger(jh_in, mapped_data + new_offset,
395 if (*((__be32 *)(mapped_data + new_offset)) ==
396 cpu_to_be32(JBD2_MAGIC_NUMBER)) {
400 kunmap_atomic(mapped_data);
403 * Do we need to do a data copy?
405 if (need_copy_out && !done_copy_out) {
408 spin_unlock(&jh_in->b_state_lock);
409 tmp = jbd2_alloc(bh_in->b_size, GFP_NOFS);
414 spin_lock(&jh_in->b_state_lock);
415 if (jh_in->b_frozen_data) {
416 jbd2_free(tmp, bh_in->b_size);
420 jh_in->b_frozen_data = tmp;
421 mapped_data = kmap_atomic(new_page);
422 memcpy(tmp, mapped_data + new_offset, bh_in->b_size);
423 kunmap_atomic(mapped_data);
425 new_page = virt_to_page(tmp);
426 new_offset = offset_in_page(tmp);
430 * This isn't strictly necessary, as we're using frozen
431 * data for the escaping, but it keeps consistency with
432 * b_frozen_data usage.
434 jh_in->b_frozen_triggers = jh_in->b_triggers;
438 * Did we need to do an escaping? Now we've done all the
439 * copying, we can finally do so.
442 mapped_data = kmap_atomic(new_page);
443 *((unsigned int *)(mapped_data + new_offset)) = 0;
444 kunmap_atomic(mapped_data);
447 set_bh_page(new_bh, new_page, new_offset);
448 new_bh->b_size = bh_in->b_size;
449 new_bh->b_bdev = journal->j_dev;
450 new_bh->b_blocknr = blocknr;
451 new_bh->b_private = bh_in;
452 set_buffer_mapped(new_bh);
453 set_buffer_dirty(new_bh);
458 * The to-be-written buffer needs to get moved to the io queue,
459 * and the original buffer whose contents we are shadowing or
460 * copying is moved to the transaction's shadow queue.
462 JBUFFER_TRACE(jh_in, "file as BJ_Shadow");
463 spin_lock(&journal->j_list_lock);
464 __jbd2_journal_file_buffer(jh_in, transaction, BJ_Shadow);
465 spin_unlock(&journal->j_list_lock);
466 set_buffer_shadow(bh_in);
467 spin_unlock(&jh_in->b_state_lock);
469 return do_escape | (done_copy_out << 1);
473 * Allocation code for the journal file. Manage the space left in the
474 * journal, so that we can begin checkpointing when appropriate.
478 * Called with j_state_lock locked for writing.
479 * Returns true if a transaction commit was started.
481 int __jbd2_log_start_commit(journal_t *journal, tid_t target)
483 /* Return if the txn has already requested to be committed */
484 if (journal->j_commit_request == target)
488 * The only transaction we can possibly wait upon is the
489 * currently running transaction (if it exists). Otherwise,
490 * the target tid must be an old one.
492 if (journal->j_running_transaction &&
493 journal->j_running_transaction->t_tid == target) {
495 * We want a new commit: OK, mark the request and wakeup the
496 * commit thread. We do _not_ do the commit ourselves.
499 journal->j_commit_request = target;
500 jbd_debug(1, "JBD2: requesting commit %u/%u\n",
501 journal->j_commit_request,
502 journal->j_commit_sequence);
503 journal->j_running_transaction->t_requested = jiffies;
504 wake_up(&journal->j_wait_commit);
506 } else if (!tid_geq(journal->j_commit_request, target))
507 /* This should never happen, but if it does, preserve
508 the evidence before kjournald goes into a loop and
509 increments j_commit_sequence beyond all recognition. */
510 WARN_ONCE(1, "JBD2: bad log_start_commit: %u %u %u %u\n",
511 journal->j_commit_request,
512 journal->j_commit_sequence,
513 target, journal->j_running_transaction ?
514 journal->j_running_transaction->t_tid : 0);
518 int jbd2_log_start_commit(journal_t *journal, tid_t tid)
522 write_lock(&journal->j_state_lock);
523 ret = __jbd2_log_start_commit(journal, tid);
524 write_unlock(&journal->j_state_lock);
529 * Force and wait any uncommitted transactions. We can only force the running
530 * transaction if we don't have an active handle, otherwise, we will deadlock.
531 * Returns: <0 in case of error,
532 * 0 if nothing to commit,
533 * 1 if transaction was successfully committed.
535 static int __jbd2_journal_force_commit(journal_t *journal)
537 transaction_t *transaction = NULL;
539 int need_to_start = 0, ret = 0;
541 read_lock(&journal->j_state_lock);
542 if (journal->j_running_transaction && !current->journal_info) {
543 transaction = journal->j_running_transaction;
544 if (!tid_geq(journal->j_commit_request, transaction->t_tid))
546 } else if (journal->j_committing_transaction)
547 transaction = journal->j_committing_transaction;
550 /* Nothing to commit */
551 read_unlock(&journal->j_state_lock);
554 tid = transaction->t_tid;
555 read_unlock(&journal->j_state_lock);
557 jbd2_log_start_commit(journal, tid);
558 ret = jbd2_log_wait_commit(journal, tid);
566 * Force and wait upon a commit if the calling process is not within
567 * transaction. This is used for forcing out undo-protected data which contains
568 * bitmaps, when the fs is running out of space.
570 * @journal: journal to force
571 * Returns true if progress was made.
573 int jbd2_journal_force_commit_nested(journal_t *journal)
577 ret = __jbd2_journal_force_commit(journal);
582 * int journal_force_commit() - force any uncommitted transactions
583 * @journal: journal to force
585 * Caller want unconditional commit. We can only force the running transaction
586 * if we don't have an active handle, otherwise, we will deadlock.
588 int jbd2_journal_force_commit(journal_t *journal)
592 J_ASSERT(!current->journal_info);
593 ret = __jbd2_journal_force_commit(journal);
600 * Start a commit of the current running transaction (if any). Returns true
601 * if a transaction is going to be committed (or is currently already
602 * committing), and fills its tid in at *ptid
604 int jbd2_journal_start_commit(journal_t *journal, tid_t *ptid)
608 write_lock(&journal->j_state_lock);
609 if (journal->j_running_transaction) {
610 tid_t tid = journal->j_running_transaction->t_tid;
612 __jbd2_log_start_commit(journal, tid);
613 /* There's a running transaction and we've just made sure
614 * it's commit has been scheduled. */
618 } else if (journal->j_committing_transaction) {
620 * If commit has been started, then we have to wait for
621 * completion of that transaction.
624 *ptid = journal->j_committing_transaction->t_tid;
627 write_unlock(&journal->j_state_lock);
632 * Return 1 if a given transaction has not yet sent barrier request
633 * connected with a transaction commit. If 0 is returned, transaction
634 * may or may not have sent the barrier. Used to avoid sending barrier
635 * twice in common cases.
637 int jbd2_trans_will_send_data_barrier(journal_t *journal, tid_t tid)
640 transaction_t *commit_trans;
642 if (!(journal->j_flags & JBD2_BARRIER))
644 read_lock(&journal->j_state_lock);
645 /* Transaction already committed? */
646 if (tid_geq(journal->j_commit_sequence, tid))
648 commit_trans = journal->j_committing_transaction;
649 if (!commit_trans || commit_trans->t_tid != tid) {
654 * Transaction is being committed and we already proceeded to
655 * submitting a flush to fs partition?
657 if (journal->j_fs_dev != journal->j_dev) {
658 if (!commit_trans->t_need_data_flush ||
659 commit_trans->t_state >= T_COMMIT_DFLUSH)
662 if (commit_trans->t_state >= T_COMMIT_JFLUSH)
667 read_unlock(&journal->j_state_lock);
670 EXPORT_SYMBOL(jbd2_trans_will_send_data_barrier);
673 * Wait for a specified commit to complete.
674 * The caller may not hold the journal lock.
676 int jbd2_log_wait_commit(journal_t *journal, tid_t tid)
680 read_lock(&journal->j_state_lock);
681 #ifdef CONFIG_PROVE_LOCKING
683 * Some callers make sure transaction is already committing and in that
684 * case we cannot block on open handles anymore. So don't warn in that
687 if (tid_gt(tid, journal->j_commit_sequence) &&
688 (!journal->j_committing_transaction ||
689 journal->j_committing_transaction->t_tid != tid)) {
690 read_unlock(&journal->j_state_lock);
691 jbd2_might_wait_for_commit(journal);
692 read_lock(&journal->j_state_lock);
695 #ifdef CONFIG_JBD2_DEBUG
696 if (!tid_geq(journal->j_commit_request, tid)) {
698 "%s: error: j_commit_request=%u, tid=%u\n",
699 __func__, journal->j_commit_request, tid);
702 while (tid_gt(tid, journal->j_commit_sequence)) {
703 jbd_debug(1, "JBD2: want %u, j_commit_sequence=%u\n",
704 tid, journal->j_commit_sequence);
705 read_unlock(&journal->j_state_lock);
706 wake_up(&journal->j_wait_commit);
707 wait_event(journal->j_wait_done_commit,
708 !tid_gt(tid, journal->j_commit_sequence));
709 read_lock(&journal->j_state_lock);
711 read_unlock(&journal->j_state_lock);
713 if (unlikely(is_journal_aborted(journal)))
718 /* Return 1 when transaction with given tid has already committed. */
719 int jbd2_transaction_committed(journal_t *journal, tid_t tid)
723 read_lock(&journal->j_state_lock);
724 if (journal->j_running_transaction &&
725 journal->j_running_transaction->t_tid == tid)
727 if (journal->j_committing_transaction &&
728 journal->j_committing_transaction->t_tid == tid)
730 read_unlock(&journal->j_state_lock);
733 EXPORT_SYMBOL(jbd2_transaction_committed);
736 * When this function returns the transaction corresponding to tid
737 * will be completed. If the transaction has currently running, start
738 * committing that transaction before waiting for it to complete. If
739 * the transaction id is stale, it is by definition already completed,
740 * so just return SUCCESS.
742 int jbd2_complete_transaction(journal_t *journal, tid_t tid)
744 int need_to_wait = 1;
746 read_lock(&journal->j_state_lock);
747 if (journal->j_running_transaction &&
748 journal->j_running_transaction->t_tid == tid) {
749 if (journal->j_commit_request != tid) {
750 /* transaction not yet started, so request it */
751 read_unlock(&journal->j_state_lock);
752 jbd2_log_start_commit(journal, tid);
755 } else if (!(journal->j_committing_transaction &&
756 journal->j_committing_transaction->t_tid == tid))
758 read_unlock(&journal->j_state_lock);
762 return jbd2_log_wait_commit(journal, tid);
764 EXPORT_SYMBOL(jbd2_complete_transaction);
767 * Log buffer allocation routines:
770 int jbd2_journal_next_log_block(journal_t *journal, unsigned long long *retp)
772 unsigned long blocknr;
774 write_lock(&journal->j_state_lock);
775 J_ASSERT(journal->j_free > 1);
777 blocknr = journal->j_head;
780 if (journal->j_head == journal->j_last)
781 journal->j_head = journal->j_first;
782 write_unlock(&journal->j_state_lock);
783 return jbd2_journal_bmap(journal, blocknr, retp);
787 * Conversion of logical to physical block numbers for the journal
789 * On external journals the journal blocks are identity-mapped, so
790 * this is a no-op. If needed, we can use j_blk_offset - everything is
793 int jbd2_journal_bmap(journal_t *journal, unsigned long blocknr,
794 unsigned long long *retp)
797 unsigned long long ret;
800 if (journal->j_inode) {
802 ret = bmap(journal->j_inode, &block);
805 printk(KERN_ALERT "%s: journal block not found "
806 "at offset %lu on %s\n",
807 __func__, blocknr, journal->j_devname);
809 __journal_abort_soft(journal, err);
816 *retp = blocknr; /* +journal->j_blk_offset */
822 * We play buffer_head aliasing tricks to write data/metadata blocks to
823 * the journal without copying their contents, but for journal
824 * descriptor blocks we do need to generate bona fide buffers.
826 * After the caller of jbd2_journal_get_descriptor_buffer() has finished modifying
827 * the buffer's contents they really should run flush_dcache_page(bh->b_page).
828 * But we don't bother doing that, so there will be coherency problems with
829 * mmaps of blockdevs which hold live JBD-controlled filesystems.
832 jbd2_journal_get_descriptor_buffer(transaction_t *transaction, int type)
834 journal_t *journal = transaction->t_journal;
835 struct buffer_head *bh;
836 unsigned long long blocknr;
837 journal_header_t *header;
840 err = jbd2_journal_next_log_block(journal, &blocknr);
845 bh = __getblk(journal->j_dev, blocknr, journal->j_blocksize);
848 atomic_dec(&transaction->t_outstanding_credits);
850 memset(bh->b_data, 0, journal->j_blocksize);
851 header = (journal_header_t *)bh->b_data;
852 header->h_magic = cpu_to_be32(JBD2_MAGIC_NUMBER);
853 header->h_blocktype = cpu_to_be32(type);
854 header->h_sequence = cpu_to_be32(transaction->t_tid);
855 set_buffer_uptodate(bh);
857 BUFFER_TRACE(bh, "return this buffer");
861 void jbd2_descriptor_block_csum_set(journal_t *j, struct buffer_head *bh)
863 struct jbd2_journal_block_tail *tail;
866 if (!jbd2_journal_has_csum_v2or3(j))
869 tail = (struct jbd2_journal_block_tail *)(bh->b_data + j->j_blocksize -
870 sizeof(struct jbd2_journal_block_tail));
871 tail->t_checksum = 0;
872 csum = jbd2_chksum(j, j->j_csum_seed, bh->b_data, j->j_blocksize);
873 tail->t_checksum = cpu_to_be32(csum);
877 * Return tid of the oldest transaction in the journal and block in the journal
878 * where the transaction starts.
880 * If the journal is now empty, return which will be the next transaction ID
881 * we will write and where will that transaction start.
883 * The return value is 0 if journal tail cannot be pushed any further, 1 if
886 int jbd2_journal_get_log_tail(journal_t *journal, tid_t *tid,
887 unsigned long *block)
889 transaction_t *transaction;
892 read_lock(&journal->j_state_lock);
893 spin_lock(&journal->j_list_lock);
894 transaction = journal->j_checkpoint_transactions;
896 *tid = transaction->t_tid;
897 *block = transaction->t_log_start;
898 } else if ((transaction = journal->j_committing_transaction) != NULL) {
899 *tid = transaction->t_tid;
900 *block = transaction->t_log_start;
901 } else if ((transaction = journal->j_running_transaction) != NULL) {
902 *tid = transaction->t_tid;
903 *block = journal->j_head;
905 *tid = journal->j_transaction_sequence;
906 *block = journal->j_head;
908 ret = tid_gt(*tid, journal->j_tail_sequence);
909 spin_unlock(&journal->j_list_lock);
910 read_unlock(&journal->j_state_lock);
916 * Update information in journal structure and in on disk journal superblock
917 * about log tail. This function does not check whether information passed in
918 * really pushes log tail further. It's responsibility of the caller to make
919 * sure provided log tail information is valid (e.g. by holding
920 * j_checkpoint_mutex all the time between computing log tail and calling this
921 * function as is the case with jbd2_cleanup_journal_tail()).
923 * Requires j_checkpoint_mutex
925 int __jbd2_update_log_tail(journal_t *journal, tid_t tid, unsigned long block)
930 BUG_ON(!mutex_is_locked(&journal->j_checkpoint_mutex));
933 * We cannot afford for write to remain in drive's caches since as
934 * soon as we update j_tail, next transaction can start reusing journal
935 * space and if we lose sb update during power failure we'd replay
936 * old transaction with possibly newly overwritten data.
938 ret = jbd2_journal_update_sb_log_tail(journal, tid, block,
943 write_lock(&journal->j_state_lock);
944 freed = block - journal->j_tail;
945 if (block < journal->j_tail)
946 freed += journal->j_last - journal->j_first;
948 trace_jbd2_update_log_tail(journal, tid, block, freed);
950 "Cleaning journal tail from %u to %u (offset %lu), "
952 journal->j_tail_sequence, tid, block, freed);
954 journal->j_free += freed;
955 journal->j_tail_sequence = tid;
956 journal->j_tail = block;
957 write_unlock(&journal->j_state_lock);
964 * This is a variation of __jbd2_update_log_tail which checks for validity of
965 * provided log tail and locks j_checkpoint_mutex. So it is safe against races
966 * with other threads updating log tail.
968 void jbd2_update_log_tail(journal_t *journal, tid_t tid, unsigned long block)
970 mutex_lock_io(&journal->j_checkpoint_mutex);
971 if (tid_gt(tid, journal->j_tail_sequence))
972 __jbd2_update_log_tail(journal, tid, block);
973 mutex_unlock(&journal->j_checkpoint_mutex);
976 struct jbd2_stats_proc_session {
978 struct transaction_stats_s *stats;
983 static void *jbd2_seq_info_start(struct seq_file *seq, loff_t *pos)
985 return *pos ? NULL : SEQ_START_TOKEN;
988 static void *jbd2_seq_info_next(struct seq_file *seq, void *v, loff_t *pos)
993 static int jbd2_seq_info_show(struct seq_file *seq, void *v)
995 struct jbd2_stats_proc_session *s = seq->private;
997 if (v != SEQ_START_TOKEN)
999 seq_printf(seq, "%lu transactions (%lu requested), "
1000 "each up to %u blocks\n",
1001 s->stats->ts_tid, s->stats->ts_requested,
1002 s->journal->j_max_transaction_buffers);
1003 if (s->stats->ts_tid == 0)
1005 seq_printf(seq, "average: \n %ums waiting for transaction\n",
1006 jiffies_to_msecs(s->stats->run.rs_wait / s->stats->ts_tid));
1007 seq_printf(seq, " %ums request delay\n",
1008 (s->stats->ts_requested == 0) ? 0 :
1009 jiffies_to_msecs(s->stats->run.rs_request_delay /
1010 s->stats->ts_requested));
1011 seq_printf(seq, " %ums running transaction\n",
1012 jiffies_to_msecs(s->stats->run.rs_running / s->stats->ts_tid));
1013 seq_printf(seq, " %ums transaction was being locked\n",
1014 jiffies_to_msecs(s->stats->run.rs_locked / s->stats->ts_tid));
1015 seq_printf(seq, " %ums flushing data (in ordered mode)\n",
1016 jiffies_to_msecs(s->stats->run.rs_flushing / s->stats->ts_tid));
1017 seq_printf(seq, " %ums logging transaction\n",
1018 jiffies_to_msecs(s->stats->run.rs_logging / s->stats->ts_tid));
1019 seq_printf(seq, " %lluus average transaction commit time\n",
1020 div_u64(s->journal->j_average_commit_time, 1000));
1021 seq_printf(seq, " %lu handles per transaction\n",
1022 s->stats->run.rs_handle_count / s->stats->ts_tid);
1023 seq_printf(seq, " %lu blocks per transaction\n",
1024 s->stats->run.rs_blocks / s->stats->ts_tid);
1025 seq_printf(seq, " %lu logged blocks per transaction\n",
1026 s->stats->run.rs_blocks_logged / s->stats->ts_tid);
1030 static void jbd2_seq_info_stop(struct seq_file *seq, void *v)
1034 static const struct seq_operations jbd2_seq_info_ops = {
1035 .start = jbd2_seq_info_start,
1036 .next = jbd2_seq_info_next,
1037 .stop = jbd2_seq_info_stop,
1038 .show = jbd2_seq_info_show,
1041 static int jbd2_seq_info_open(struct inode *inode, struct file *file)
1043 journal_t *journal = PDE_DATA(inode);
1044 struct jbd2_stats_proc_session *s;
1047 s = kmalloc(sizeof(*s), GFP_KERNEL);
1050 size = sizeof(struct transaction_stats_s);
1051 s->stats = kmalloc(size, GFP_KERNEL);
1052 if (s->stats == NULL) {
1056 spin_lock(&journal->j_history_lock);
1057 memcpy(s->stats, &journal->j_stats, size);
1058 s->journal = journal;
1059 spin_unlock(&journal->j_history_lock);
1061 rc = seq_open(file, &jbd2_seq_info_ops);
1063 struct seq_file *m = file->private_data;
1073 static int jbd2_seq_info_release(struct inode *inode, struct file *file)
1075 struct seq_file *seq = file->private_data;
1076 struct jbd2_stats_proc_session *s = seq->private;
1079 return seq_release(inode, file);
1082 static const struct file_operations jbd2_seq_info_fops = {
1083 .owner = THIS_MODULE,
1084 .open = jbd2_seq_info_open,
1086 .llseek = seq_lseek,
1087 .release = jbd2_seq_info_release,
1090 static struct proc_dir_entry *proc_jbd2_stats;
1092 static void jbd2_stats_proc_init(journal_t *journal)
1094 journal->j_proc_entry = proc_mkdir(journal->j_devname, proc_jbd2_stats);
1095 if (journal->j_proc_entry) {
1096 proc_create_data("info", S_IRUGO, journal->j_proc_entry,
1097 &jbd2_seq_info_fops, journal);
1101 static void jbd2_stats_proc_exit(journal_t *journal)
1103 remove_proc_entry("info", journal->j_proc_entry);
1104 remove_proc_entry(journal->j_devname, proc_jbd2_stats);
1107 /* Minimum size of descriptor tag */
1108 static int jbd2_min_tag_size(void)
1111 * Tag with 32-bit block numbers does not use last four bytes of the
1114 return sizeof(journal_block_tag_t) - 4;
1118 * Management for journal control blocks: functions to create and
1119 * destroy journal_t structures, and to initialise and read existing
1120 * journal blocks from disk. */
1122 /* First: create and setup a journal_t object in memory. We initialise
1123 * very few fields yet: that has to wait until we have created the
1124 * journal structures from from scratch, or loaded them from disk. */
1126 static journal_t *journal_init_common(struct block_device *bdev,
1127 struct block_device *fs_dev,
1128 unsigned long long start, int len, int blocksize)
1130 static struct lock_class_key jbd2_trans_commit_key;
1133 struct buffer_head *bh;
1136 journal = kzalloc(sizeof(*journal), GFP_KERNEL);
1140 init_waitqueue_head(&journal->j_wait_transaction_locked);
1141 init_waitqueue_head(&journal->j_wait_done_commit);
1142 init_waitqueue_head(&journal->j_wait_commit);
1143 init_waitqueue_head(&journal->j_wait_updates);
1144 init_waitqueue_head(&journal->j_wait_reserved);
1145 mutex_init(&journal->j_barrier);
1146 mutex_init(&journal->j_checkpoint_mutex);
1147 spin_lock_init(&journal->j_revoke_lock);
1148 spin_lock_init(&journal->j_list_lock);
1149 rwlock_init(&journal->j_state_lock);
1151 journal->j_commit_interval = (HZ * JBD2_DEFAULT_MAX_COMMIT_AGE);
1152 journal->j_min_batch_time = 0;
1153 journal->j_max_batch_time = 15000; /* 15ms */
1154 atomic_set(&journal->j_reserved_credits, 0);
1156 /* The journal is marked for error until we succeed with recovery! */
1157 journal->j_flags = JBD2_ABORT;
1159 /* Set up a default-sized revoke table for the new mount. */
1160 err = jbd2_journal_init_revoke(journal, JOURNAL_REVOKE_DEFAULT_HASH);
1164 spin_lock_init(&journal->j_history_lock);
1166 lockdep_init_map(&journal->j_trans_commit_map, "jbd2_handle",
1167 &jbd2_trans_commit_key, 0);
1169 /* journal descriptor can store up to n blocks -bzzz */
1170 journal->j_blocksize = blocksize;
1171 journal->j_dev = bdev;
1172 journal->j_fs_dev = fs_dev;
1173 journal->j_blk_offset = start;
1174 journal->j_maxlen = len;
1175 /* We need enough buffers to write out full descriptor block. */
1176 n = journal->j_blocksize / jbd2_min_tag_size();
1177 journal->j_wbufsize = n;
1178 journal->j_wbuf = kmalloc_array(n, sizeof(struct buffer_head *),
1180 if (!journal->j_wbuf)
1183 bh = getblk_unmovable(journal->j_dev, start, journal->j_blocksize);
1185 pr_err("%s: Cannot get buffer for journal superblock\n",
1189 journal->j_sb_buffer = bh;
1190 journal->j_superblock = (journal_superblock_t *)bh->b_data;
1195 kfree(journal->j_wbuf);
1196 jbd2_journal_destroy_revoke(journal);
1201 /* jbd2_journal_init_dev and jbd2_journal_init_inode:
1203 * Create a journal structure assigned some fixed set of disk blocks to
1204 * the journal. We don't actually touch those disk blocks yet, but we
1205 * need to set up all of the mapping information to tell the journaling
1206 * system where the journal blocks are.
1211 * journal_t * jbd2_journal_init_dev() - creates and initialises a journal structure
1212 * @bdev: Block device on which to create the journal
1213 * @fs_dev: Device which hold journalled filesystem for this journal.
1214 * @start: Block nr Start of journal.
1215 * @len: Length of the journal in blocks.
1216 * @blocksize: blocksize of journalling device
1218 * Returns: a newly created journal_t *
1220 * jbd2_journal_init_dev creates a journal which maps a fixed contiguous
1221 * range of blocks on an arbitrary block device.
1224 journal_t *jbd2_journal_init_dev(struct block_device *bdev,
1225 struct block_device *fs_dev,
1226 unsigned long long start, int len, int blocksize)
1230 journal = journal_init_common(bdev, fs_dev, start, len, blocksize);
1234 bdevname(journal->j_dev, journal->j_devname);
1235 strreplace(journal->j_devname, '/', '!');
1236 jbd2_stats_proc_init(journal);
1242 * journal_t * jbd2_journal_init_inode () - creates a journal which maps to a inode.
1243 * @inode: An inode to create the journal in
1245 * jbd2_journal_init_inode creates a journal which maps an on-disk inode as
1246 * the journal. The inode must exist already, must support bmap() and
1247 * must have all data blocks preallocated.
1249 journal_t *jbd2_journal_init_inode(struct inode *inode)
1257 err = bmap(inode, &blocknr);
1259 if (err || !blocknr) {
1260 pr_err("%s: Cannot locate journal superblock\n",
1265 jbd_debug(1, "JBD2: inode %s/%ld, size %lld, bits %d, blksize %ld\n",
1266 inode->i_sb->s_id, inode->i_ino, (long long) inode->i_size,
1267 inode->i_sb->s_blocksize_bits, inode->i_sb->s_blocksize);
1269 journal = journal_init_common(inode->i_sb->s_bdev, inode->i_sb->s_bdev,
1270 blocknr, inode->i_size >> inode->i_sb->s_blocksize_bits,
1271 inode->i_sb->s_blocksize);
1275 journal->j_inode = inode;
1276 bdevname(journal->j_dev, journal->j_devname);
1277 p = strreplace(journal->j_devname, '/', '!');
1278 sprintf(p, "-%lu", journal->j_inode->i_ino);
1279 jbd2_stats_proc_init(journal);
1285 * If the journal init or create aborts, we need to mark the journal
1286 * superblock as being NULL to prevent the journal destroy from writing
1287 * back a bogus superblock.
1289 static void journal_fail_superblock (journal_t *journal)
1291 struct buffer_head *bh = journal->j_sb_buffer;
1293 journal->j_sb_buffer = NULL;
1297 * Given a journal_t structure, initialise the various fields for
1298 * startup of a new journaling session. We use this both when creating
1299 * a journal, and after recovering an old journal to reset it for
1303 static int journal_reset(journal_t *journal)
1305 journal_superblock_t *sb = journal->j_superblock;
1306 unsigned long long first, last;
1308 first = be32_to_cpu(sb->s_first);
1309 last = be32_to_cpu(sb->s_maxlen);
1310 if (first + JBD2_MIN_JOURNAL_BLOCKS > last + 1) {
1311 printk(KERN_ERR "JBD2: Journal too short (blocks %llu-%llu).\n",
1313 journal_fail_superblock(journal);
1317 journal->j_first = first;
1318 journal->j_last = last;
1320 journal->j_head = first;
1321 journal->j_tail = first;
1322 journal->j_free = last - first;
1324 journal->j_tail_sequence = journal->j_transaction_sequence;
1325 journal->j_commit_sequence = journal->j_transaction_sequence - 1;
1326 journal->j_commit_request = journal->j_commit_sequence;
1328 journal->j_max_transaction_buffers = journal->j_maxlen / 4;
1331 * As a special case, if the on-disk copy is already marked as needing
1332 * no recovery (s_start == 0), then we can safely defer the superblock
1333 * update until the next commit by setting JBD2_FLUSHED. This avoids
1334 * attempting a write to a potential-readonly device.
1336 if (sb->s_start == 0) {
1337 jbd_debug(1, "JBD2: Skipping superblock update on recovered sb "
1338 "(start %ld, seq %u, errno %d)\n",
1339 journal->j_tail, journal->j_tail_sequence,
1341 journal->j_flags |= JBD2_FLUSHED;
1343 /* Lock here to make assertions happy... */
1344 mutex_lock_io(&journal->j_checkpoint_mutex);
1346 * Update log tail information. We use REQ_FUA since new
1347 * transaction will start reusing journal space and so we
1348 * must make sure information about current log tail is on
1351 jbd2_journal_update_sb_log_tail(journal,
1352 journal->j_tail_sequence,
1354 REQ_SYNC | REQ_FUA);
1355 mutex_unlock(&journal->j_checkpoint_mutex);
1357 return jbd2_journal_start_thread(journal);
1361 * This function expects that the caller will have locked the journal
1362 * buffer head, and will return with it unlocked
1364 static int jbd2_write_superblock(journal_t *journal, int write_flags)
1366 struct buffer_head *bh = journal->j_sb_buffer;
1367 journal_superblock_t *sb = journal->j_superblock;
1370 /* Buffer got discarded which means block device got invalidated */
1371 if (!buffer_mapped(bh))
1374 trace_jbd2_write_superblock(journal, write_flags);
1375 if (!(journal->j_flags & JBD2_BARRIER))
1376 write_flags &= ~(REQ_FUA | REQ_PREFLUSH);
1377 if (buffer_write_io_error(bh)) {
1379 * Oh, dear. A previous attempt to write the journal
1380 * superblock failed. This could happen because the
1381 * USB device was yanked out. Or it could happen to
1382 * be a transient write error and maybe the block will
1383 * be remapped. Nothing we can do but to retry the
1384 * write and hope for the best.
1386 printk(KERN_ERR "JBD2: previous I/O error detected "
1387 "for journal superblock update for %s.\n",
1388 journal->j_devname);
1389 clear_buffer_write_io_error(bh);
1390 set_buffer_uptodate(bh);
1392 if (jbd2_journal_has_csum_v2or3(journal))
1393 sb->s_checksum = jbd2_superblock_csum(journal, sb);
1395 bh->b_end_io = end_buffer_write_sync;
1396 ret = submit_bh(REQ_OP_WRITE, write_flags, bh);
1398 if (buffer_write_io_error(bh)) {
1399 clear_buffer_write_io_error(bh);
1400 set_buffer_uptodate(bh);
1404 printk(KERN_ERR "JBD2: Error %d detected when updating "
1405 "journal superblock for %s.\n", ret,
1406 journal->j_devname);
1407 jbd2_journal_abort(journal, ret);
1414 * jbd2_journal_update_sb_log_tail() - Update log tail in journal sb on disk.
1415 * @journal: The journal to update.
1416 * @tail_tid: TID of the new transaction at the tail of the log
1417 * @tail_block: The first block of the transaction at the tail of the log
1418 * @write_op: With which operation should we write the journal sb
1420 * Update a journal's superblock information about log tail and write it to
1421 * disk, waiting for the IO to complete.
1423 int jbd2_journal_update_sb_log_tail(journal_t *journal, tid_t tail_tid,
1424 unsigned long tail_block, int write_op)
1426 journal_superblock_t *sb = journal->j_superblock;
1429 if (is_journal_aborted(journal))
1432 BUG_ON(!mutex_is_locked(&journal->j_checkpoint_mutex));
1433 jbd_debug(1, "JBD2: updating superblock (start %lu, seq %u)\n",
1434 tail_block, tail_tid);
1436 lock_buffer(journal->j_sb_buffer);
1437 sb->s_sequence = cpu_to_be32(tail_tid);
1438 sb->s_start = cpu_to_be32(tail_block);
1440 ret = jbd2_write_superblock(journal, write_op);
1444 /* Log is no longer empty */
1445 write_lock(&journal->j_state_lock);
1446 WARN_ON(!sb->s_sequence);
1447 journal->j_flags &= ~JBD2_FLUSHED;
1448 write_unlock(&journal->j_state_lock);
1455 * jbd2_mark_journal_empty() - Mark on disk journal as empty.
1456 * @journal: The journal to update.
1457 * @write_op: With which operation should we write the journal sb
1459 * Update a journal's dynamic superblock fields to show that journal is empty.
1460 * Write updated superblock to disk waiting for IO to complete.
1462 static void jbd2_mark_journal_empty(journal_t *journal, int write_op)
1464 journal_superblock_t *sb = journal->j_superblock;
1466 BUG_ON(!mutex_is_locked(&journal->j_checkpoint_mutex));
1467 lock_buffer(journal->j_sb_buffer);
1468 if (sb->s_start == 0) { /* Is it already empty? */
1469 unlock_buffer(journal->j_sb_buffer);
1473 jbd_debug(1, "JBD2: Marking journal as empty (seq %u)\n",
1474 journal->j_tail_sequence);
1476 sb->s_sequence = cpu_to_be32(journal->j_tail_sequence);
1477 sb->s_start = cpu_to_be32(0);
1479 jbd2_write_superblock(journal, write_op);
1481 /* Log is no longer empty */
1482 write_lock(&journal->j_state_lock);
1483 journal->j_flags |= JBD2_FLUSHED;
1484 write_unlock(&journal->j_state_lock);
1489 * jbd2_journal_update_sb_errno() - Update error in the journal.
1490 * @journal: The journal to update.
1492 * Update a journal's errno. Write updated superblock to disk waiting for IO
1495 void jbd2_journal_update_sb_errno(journal_t *journal)
1497 journal_superblock_t *sb = journal->j_superblock;
1500 lock_buffer(journal->j_sb_buffer);
1501 errcode = journal->j_errno;
1502 if (errcode == -ESHUTDOWN)
1504 jbd_debug(1, "JBD2: updating superblock error (errno %d)\n", errcode);
1505 sb->s_errno = cpu_to_be32(errcode);
1507 jbd2_write_superblock(journal, REQ_SYNC | REQ_FUA);
1509 EXPORT_SYMBOL(jbd2_journal_update_sb_errno);
1511 static int journal_revoke_records_per_block(journal_t *journal)
1514 int space = journal->j_blocksize - sizeof(jbd2_journal_revoke_header_t);
1516 if (jbd2_has_feature_64bit(journal))
1521 if (jbd2_journal_has_csum_v2or3(journal))
1522 space -= sizeof(struct jbd2_journal_block_tail);
1523 return space / record_size;
1527 * Read the superblock for a given journal, performing initial
1528 * validation of the format.
1530 static int journal_get_superblock(journal_t *journal)
1532 struct buffer_head *bh;
1533 journal_superblock_t *sb;
1536 bh = journal->j_sb_buffer;
1538 J_ASSERT(bh != NULL);
1539 if (!buffer_uptodate(bh)) {
1540 ll_rw_block(REQ_OP_READ, 0, 1, &bh);
1542 if (!buffer_uptodate(bh)) {
1544 "JBD2: IO error reading journal superblock\n");
1549 if (buffer_verified(bh))
1552 sb = journal->j_superblock;
1556 if (sb->s_header.h_magic != cpu_to_be32(JBD2_MAGIC_NUMBER) ||
1557 sb->s_blocksize != cpu_to_be32(journal->j_blocksize)) {
1558 printk(KERN_WARNING "JBD2: no valid journal superblock found\n");
1562 switch(be32_to_cpu(sb->s_header.h_blocktype)) {
1563 case JBD2_SUPERBLOCK_V1:
1564 journal->j_format_version = 1;
1566 case JBD2_SUPERBLOCK_V2:
1567 journal->j_format_version = 2;
1570 printk(KERN_WARNING "JBD2: unrecognised superblock format ID\n");
1574 if (be32_to_cpu(sb->s_maxlen) < journal->j_maxlen)
1575 journal->j_maxlen = be32_to_cpu(sb->s_maxlen);
1576 else if (be32_to_cpu(sb->s_maxlen) > journal->j_maxlen) {
1577 printk(KERN_WARNING "JBD2: journal file too short\n");
1581 if (be32_to_cpu(sb->s_first) == 0 ||
1582 be32_to_cpu(sb->s_first) >= journal->j_maxlen) {
1584 "JBD2: Invalid start block of journal: %u\n",
1585 be32_to_cpu(sb->s_first));
1589 if (jbd2_has_feature_csum2(journal) &&
1590 jbd2_has_feature_csum3(journal)) {
1591 /* Can't have checksum v2 and v3 at the same time! */
1592 printk(KERN_ERR "JBD2: Can't enable checksumming v2 and v3 "
1593 "at the same time!\n");
1597 if (jbd2_journal_has_csum_v2or3_feature(journal) &&
1598 jbd2_has_feature_checksum(journal)) {
1599 /* Can't have checksum v1 and v2 on at the same time! */
1600 printk(KERN_ERR "JBD2: Can't enable checksumming v1 and v2/3 "
1601 "at the same time!\n");
1605 if (!jbd2_verify_csum_type(journal, sb)) {
1606 printk(KERN_ERR "JBD2: Unknown checksum type\n");
1610 /* Load the checksum driver */
1611 if (jbd2_journal_has_csum_v2or3_feature(journal)) {
1612 journal->j_chksum_driver = crypto_alloc_shash("crc32c", 0, 0);
1613 if (IS_ERR(journal->j_chksum_driver)) {
1614 printk(KERN_ERR "JBD2: Cannot load crc32c driver.\n");
1615 err = PTR_ERR(journal->j_chksum_driver);
1616 journal->j_chksum_driver = NULL;
1621 if (jbd2_journal_has_csum_v2or3(journal)) {
1622 /* Check superblock checksum */
1623 if (sb->s_checksum != jbd2_superblock_csum(journal, sb)) {
1624 printk(KERN_ERR "JBD2: journal checksum error\n");
1629 /* Precompute checksum seed for all metadata */
1630 journal->j_csum_seed = jbd2_chksum(journal, ~0, sb->s_uuid,
1631 sizeof(sb->s_uuid));
1634 journal->j_revoke_records_per_block =
1635 journal_revoke_records_per_block(journal);
1636 set_buffer_verified(bh);
1641 journal_fail_superblock(journal);
1646 * Load the on-disk journal superblock and read the key fields into the
1650 static int load_superblock(journal_t *journal)
1653 journal_superblock_t *sb;
1655 err = journal_get_superblock(journal);
1659 sb = journal->j_superblock;
1661 journal->j_tail_sequence = be32_to_cpu(sb->s_sequence);
1662 journal->j_tail = be32_to_cpu(sb->s_start);
1663 journal->j_first = be32_to_cpu(sb->s_first);
1664 journal->j_last = be32_to_cpu(sb->s_maxlen);
1665 journal->j_errno = be32_to_cpu(sb->s_errno);
1672 * int jbd2_journal_load() - Read journal from disk.
1673 * @journal: Journal to act on.
1675 * Given a journal_t structure which tells us which disk blocks contain
1676 * a journal, read the journal from disk to initialise the in-memory
1679 int jbd2_journal_load(journal_t *journal)
1682 journal_superblock_t *sb;
1684 err = load_superblock(journal);
1688 sb = journal->j_superblock;
1689 /* If this is a V2 superblock, then we have to check the
1690 * features flags on it. */
1692 if (journal->j_format_version >= 2) {
1693 if ((sb->s_feature_ro_compat &
1694 ~cpu_to_be32(JBD2_KNOWN_ROCOMPAT_FEATURES)) ||
1695 (sb->s_feature_incompat &
1696 ~cpu_to_be32(JBD2_KNOWN_INCOMPAT_FEATURES))) {
1698 "JBD2: Unrecognised features on journal\n");
1704 * Create a slab for this blocksize
1706 err = jbd2_journal_create_slab(be32_to_cpu(sb->s_blocksize));
1710 /* Let the recovery code check whether it needs to recover any
1711 * data from the journal. */
1712 if (jbd2_journal_recover(journal))
1713 goto recovery_error;
1715 if (journal->j_failed_commit) {
1716 printk(KERN_ERR "JBD2: journal transaction %u on %s "
1717 "is corrupt.\n", journal->j_failed_commit,
1718 journal->j_devname);
1719 return -EFSCORRUPTED;
1722 /* OK, we've finished with the dynamic journal bits:
1723 * reinitialise the dynamic contents of the superblock in memory
1724 * and reset them on disk. */
1725 if (journal_reset(journal))
1726 goto recovery_error;
1728 journal->j_flags &= ~JBD2_ABORT;
1729 journal->j_flags |= JBD2_LOADED;
1733 printk(KERN_WARNING "JBD2: recovery failed\n");
1738 * void jbd2_journal_destroy() - Release a journal_t structure.
1739 * @journal: Journal to act on.
1741 * Release a journal_t structure once it is no longer in use by the
1743 * Return <0 if we couldn't clean up the journal.
1745 int jbd2_journal_destroy(journal_t *journal)
1749 /* Wait for the commit thread to wake up and die. */
1750 journal_kill_thread(journal);
1752 /* Force a final log commit */
1753 if (journal->j_running_transaction)
1754 jbd2_journal_commit_transaction(journal);
1756 /* Force any old transactions to disk */
1758 /* Totally anal locking here... */
1759 spin_lock(&journal->j_list_lock);
1760 while (journal->j_checkpoint_transactions != NULL) {
1761 spin_unlock(&journal->j_list_lock);
1762 mutex_lock_io(&journal->j_checkpoint_mutex);
1763 err = jbd2_log_do_checkpoint(journal);
1764 mutex_unlock(&journal->j_checkpoint_mutex);
1766 * If checkpointing failed, just free the buffers to avoid
1770 jbd2_journal_destroy_checkpoint(journal);
1771 spin_lock(&journal->j_list_lock);
1774 spin_lock(&journal->j_list_lock);
1777 J_ASSERT(journal->j_running_transaction == NULL);
1778 J_ASSERT(journal->j_committing_transaction == NULL);
1779 J_ASSERT(journal->j_checkpoint_transactions == NULL);
1780 spin_unlock(&journal->j_list_lock);
1782 if (journal->j_sb_buffer) {
1783 if (!is_journal_aborted(journal)) {
1784 mutex_lock_io(&journal->j_checkpoint_mutex);
1786 write_lock(&journal->j_state_lock);
1787 journal->j_tail_sequence =
1788 ++journal->j_transaction_sequence;
1789 write_unlock(&journal->j_state_lock);
1791 jbd2_mark_journal_empty(journal,
1792 REQ_SYNC | REQ_PREFLUSH | REQ_FUA);
1793 mutex_unlock(&journal->j_checkpoint_mutex);
1796 brelse(journal->j_sb_buffer);
1799 if (journal->j_proc_entry)
1800 jbd2_stats_proc_exit(journal);
1801 iput(journal->j_inode);
1802 if (journal->j_revoke)
1803 jbd2_journal_destroy_revoke(journal);
1804 if (journal->j_chksum_driver)
1805 crypto_free_shash(journal->j_chksum_driver);
1806 kfree(journal->j_wbuf);
1814 *int jbd2_journal_check_used_features () - Check if features specified are used.
1815 * @journal: Journal to check.
1816 * @compat: bitmask of compatible features
1817 * @ro: bitmask of features that force read-only mount
1818 * @incompat: bitmask of incompatible features
1820 * Check whether the journal uses all of a given set of
1821 * features. Return true (non-zero) if it does.
1824 int jbd2_journal_check_used_features (journal_t *journal, unsigned long compat,
1825 unsigned long ro, unsigned long incompat)
1827 journal_superblock_t *sb;
1829 if (!compat && !ro && !incompat)
1831 /* Load journal superblock if it is not loaded yet. */
1832 if (journal->j_format_version == 0 &&
1833 journal_get_superblock(journal) != 0)
1835 if (journal->j_format_version == 1)
1838 sb = journal->j_superblock;
1840 if (((be32_to_cpu(sb->s_feature_compat) & compat) == compat) &&
1841 ((be32_to_cpu(sb->s_feature_ro_compat) & ro) == ro) &&
1842 ((be32_to_cpu(sb->s_feature_incompat) & incompat) == incompat))
1849 * int jbd2_journal_check_available_features() - Check feature set in journalling layer
1850 * @journal: Journal to check.
1851 * @compat: bitmask of compatible features
1852 * @ro: bitmask of features that force read-only mount
1853 * @incompat: bitmask of incompatible features
1855 * Check whether the journaling code supports the use of
1856 * all of a given set of features on this journal. Return true
1857 * (non-zero) if it can. */
1859 int jbd2_journal_check_available_features (journal_t *journal, unsigned long compat,
1860 unsigned long ro, unsigned long incompat)
1862 if (!compat && !ro && !incompat)
1865 /* We can support any known requested features iff the
1866 * superblock is in version 2. Otherwise we fail to support any
1867 * extended sb features. */
1869 if (journal->j_format_version != 2)
1872 if ((compat & JBD2_KNOWN_COMPAT_FEATURES) == compat &&
1873 (ro & JBD2_KNOWN_ROCOMPAT_FEATURES) == ro &&
1874 (incompat & JBD2_KNOWN_INCOMPAT_FEATURES) == incompat)
1881 * int jbd2_journal_set_features () - Mark a given journal feature in the superblock
1882 * @journal: Journal to act on.
1883 * @compat: bitmask of compatible features
1884 * @ro: bitmask of features that force read-only mount
1885 * @incompat: bitmask of incompatible features
1887 * Mark a given journal feature as present on the
1888 * superblock. Returns true if the requested features could be set.
1892 int jbd2_journal_set_features (journal_t *journal, unsigned long compat,
1893 unsigned long ro, unsigned long incompat)
1895 #define INCOMPAT_FEATURE_ON(f) \
1896 ((incompat & (f)) && !(sb->s_feature_incompat & cpu_to_be32(f)))
1897 #define COMPAT_FEATURE_ON(f) \
1898 ((compat & (f)) && !(sb->s_feature_compat & cpu_to_be32(f)))
1899 journal_superblock_t *sb;
1901 if (jbd2_journal_check_used_features(journal, compat, ro, incompat))
1904 if (!jbd2_journal_check_available_features(journal, compat, ro, incompat))
1907 /* If enabling v2 checksums, turn on v3 instead */
1908 if (incompat & JBD2_FEATURE_INCOMPAT_CSUM_V2) {
1909 incompat &= ~JBD2_FEATURE_INCOMPAT_CSUM_V2;
1910 incompat |= JBD2_FEATURE_INCOMPAT_CSUM_V3;
1913 /* Asking for checksumming v3 and v1? Only give them v3. */
1914 if (incompat & JBD2_FEATURE_INCOMPAT_CSUM_V3 &&
1915 compat & JBD2_FEATURE_COMPAT_CHECKSUM)
1916 compat &= ~JBD2_FEATURE_COMPAT_CHECKSUM;
1918 jbd_debug(1, "Setting new features 0x%lx/0x%lx/0x%lx\n",
1919 compat, ro, incompat);
1921 sb = journal->j_superblock;
1923 /* Load the checksum driver if necessary */
1924 if ((journal->j_chksum_driver == NULL) &&
1925 INCOMPAT_FEATURE_ON(JBD2_FEATURE_INCOMPAT_CSUM_V3)) {
1926 journal->j_chksum_driver = crypto_alloc_shash("crc32c", 0, 0);
1927 if (IS_ERR(journal->j_chksum_driver)) {
1928 printk(KERN_ERR "JBD2: Cannot load crc32c driver.\n");
1929 journal->j_chksum_driver = NULL;
1932 /* Precompute checksum seed for all metadata */
1933 journal->j_csum_seed = jbd2_chksum(journal, ~0, sb->s_uuid,
1934 sizeof(sb->s_uuid));
1937 lock_buffer(journal->j_sb_buffer);
1939 /* If enabling v3 checksums, update superblock */
1940 if (INCOMPAT_FEATURE_ON(JBD2_FEATURE_INCOMPAT_CSUM_V3)) {
1941 sb->s_checksum_type = JBD2_CRC32C_CHKSUM;
1942 sb->s_feature_compat &=
1943 ~cpu_to_be32(JBD2_FEATURE_COMPAT_CHECKSUM);
1946 /* If enabling v1 checksums, downgrade superblock */
1947 if (COMPAT_FEATURE_ON(JBD2_FEATURE_COMPAT_CHECKSUM))
1948 sb->s_feature_incompat &=
1949 ~cpu_to_be32(JBD2_FEATURE_INCOMPAT_CSUM_V2 |
1950 JBD2_FEATURE_INCOMPAT_CSUM_V3);
1952 sb->s_feature_compat |= cpu_to_be32(compat);
1953 sb->s_feature_ro_compat |= cpu_to_be32(ro);
1954 sb->s_feature_incompat |= cpu_to_be32(incompat);
1955 unlock_buffer(journal->j_sb_buffer);
1956 journal->j_revoke_records_per_block =
1957 journal_revoke_records_per_block(journal);
1960 #undef COMPAT_FEATURE_ON
1961 #undef INCOMPAT_FEATURE_ON
1965 * jbd2_journal_clear_features () - Clear a given journal feature in the
1967 * @journal: Journal to act on.
1968 * @compat: bitmask of compatible features
1969 * @ro: bitmask of features that force read-only mount
1970 * @incompat: bitmask of incompatible features
1972 * Clear a given journal feature as present on the
1975 void jbd2_journal_clear_features(journal_t *journal, unsigned long compat,
1976 unsigned long ro, unsigned long incompat)
1978 journal_superblock_t *sb;
1980 jbd_debug(1, "Clear features 0x%lx/0x%lx/0x%lx\n",
1981 compat, ro, incompat);
1983 sb = journal->j_superblock;
1985 sb->s_feature_compat &= ~cpu_to_be32(compat);
1986 sb->s_feature_ro_compat &= ~cpu_to_be32(ro);
1987 sb->s_feature_incompat &= ~cpu_to_be32(incompat);
1988 journal->j_revoke_records_per_block =
1989 journal_revoke_records_per_block(journal);
1991 EXPORT_SYMBOL(jbd2_journal_clear_features);
1994 * int jbd2_journal_flush () - Flush journal
1995 * @journal: Journal to act on.
1997 * Flush all data for a given journal to disk and empty the journal.
1998 * Filesystems can use this when remounting readonly to ensure that
1999 * recovery does not need to happen on remount.
2002 int jbd2_journal_flush(journal_t *journal)
2005 transaction_t *transaction = NULL;
2007 write_lock(&journal->j_state_lock);
2009 /* Force everything buffered to the log... */
2010 if (journal->j_running_transaction) {
2011 transaction = journal->j_running_transaction;
2012 __jbd2_log_start_commit(journal, transaction->t_tid);
2013 } else if (journal->j_committing_transaction)
2014 transaction = journal->j_committing_transaction;
2016 /* Wait for the log commit to complete... */
2018 tid_t tid = transaction->t_tid;
2020 write_unlock(&journal->j_state_lock);
2021 jbd2_log_wait_commit(journal, tid);
2023 write_unlock(&journal->j_state_lock);
2026 /* ...and flush everything in the log out to disk. */
2027 spin_lock(&journal->j_list_lock);
2028 while (!err && journal->j_checkpoint_transactions != NULL) {
2029 spin_unlock(&journal->j_list_lock);
2030 mutex_lock_io(&journal->j_checkpoint_mutex);
2031 err = jbd2_log_do_checkpoint(journal);
2032 mutex_unlock(&journal->j_checkpoint_mutex);
2033 spin_lock(&journal->j_list_lock);
2035 spin_unlock(&journal->j_list_lock);
2037 if (is_journal_aborted(journal))
2040 mutex_lock_io(&journal->j_checkpoint_mutex);
2042 err = jbd2_cleanup_journal_tail(journal);
2044 mutex_unlock(&journal->j_checkpoint_mutex);
2050 /* Finally, mark the journal as really needing no recovery.
2051 * This sets s_start==0 in the underlying superblock, which is
2052 * the magic code for a fully-recovered superblock. Any future
2053 * commits of data to the journal will restore the current
2055 jbd2_mark_journal_empty(journal, REQ_SYNC | REQ_FUA);
2056 mutex_unlock(&journal->j_checkpoint_mutex);
2057 write_lock(&journal->j_state_lock);
2058 J_ASSERT(!journal->j_running_transaction);
2059 J_ASSERT(!journal->j_committing_transaction);
2060 J_ASSERT(!journal->j_checkpoint_transactions);
2061 J_ASSERT(journal->j_head == journal->j_tail);
2062 J_ASSERT(journal->j_tail_sequence == journal->j_transaction_sequence);
2063 write_unlock(&journal->j_state_lock);
2069 * int jbd2_journal_wipe() - Wipe journal contents
2070 * @journal: Journal to act on.
2071 * @write: flag (see below)
2073 * Wipe out all of the contents of a journal, safely. This will produce
2074 * a warning if the journal contains any valid recovery information.
2075 * Must be called between journal_init_*() and jbd2_journal_load().
2077 * If 'write' is non-zero, then we wipe out the journal on disk; otherwise
2078 * we merely suppress recovery.
2081 int jbd2_journal_wipe(journal_t *journal, int write)
2085 J_ASSERT (!(journal->j_flags & JBD2_LOADED));
2087 err = load_superblock(journal);
2091 if (!journal->j_tail)
2094 printk(KERN_WARNING "JBD2: %s recovery information on journal\n",
2095 write ? "Clearing" : "Ignoring");
2097 err = jbd2_journal_skip_recovery(journal);
2099 /* Lock to make assertions happy... */
2100 mutex_lock_io(&journal->j_checkpoint_mutex);
2101 jbd2_mark_journal_empty(journal, REQ_SYNC | REQ_FUA);
2102 mutex_unlock(&journal->j_checkpoint_mutex);
2110 * Journal abort has very specific semantics, which we describe
2111 * for journal abort.
2113 * Two internal functions, which provide abort to the jbd layer
2118 * Quick version for internal journal use (doesn't lock the journal).
2119 * Aborts hard --- we mark the abort as occurred, but do _nothing_ else,
2120 * and don't attempt to make any other journal updates.
2122 void __jbd2_journal_abort_hard(journal_t *journal)
2124 transaction_t *transaction;
2126 if (journal->j_flags & JBD2_ABORT)
2129 printk(KERN_ERR "Aborting journal on device %s.\n",
2130 journal->j_devname);
2132 write_lock(&journal->j_state_lock);
2133 journal->j_flags |= JBD2_ABORT;
2134 transaction = journal->j_running_transaction;
2136 __jbd2_log_start_commit(journal, transaction->t_tid);
2137 write_unlock(&journal->j_state_lock);
2140 /* Soft abort: record the abort error status in the journal superblock,
2141 * but don't do any other IO. */
2142 static void __journal_abort_soft (journal_t *journal, int errno)
2146 write_lock(&journal->j_state_lock);
2147 old_errno = journal->j_errno;
2148 if (!journal->j_errno || errno == -ESHUTDOWN)
2149 journal->j_errno = errno;
2151 if (journal->j_flags & JBD2_ABORT) {
2152 write_unlock(&journal->j_state_lock);
2153 if (!old_errno && old_errno != -ESHUTDOWN &&
2154 errno == -ESHUTDOWN)
2155 jbd2_journal_update_sb_errno(journal);
2158 write_unlock(&journal->j_state_lock);
2160 __jbd2_journal_abort_hard(journal);
2163 jbd2_journal_update_sb_errno(journal);
2164 write_lock(&journal->j_state_lock);
2165 journal->j_flags |= JBD2_REC_ERR;
2166 write_unlock(&journal->j_state_lock);
2171 * void jbd2_journal_abort () - Shutdown the journal immediately.
2172 * @journal: the journal to shutdown.
2173 * @errno: an error number to record in the journal indicating
2174 * the reason for the shutdown.
2176 * Perform a complete, immediate shutdown of the ENTIRE
2177 * journal (not of a single transaction). This operation cannot be
2178 * undone without closing and reopening the journal.
2180 * The jbd2_journal_abort function is intended to support higher level error
2181 * recovery mechanisms such as the ext2/ext3 remount-readonly error
2184 * Journal abort has very specific semantics. Any existing dirty,
2185 * unjournaled buffers in the main filesystem will still be written to
2186 * disk by bdflush, but the journaling mechanism will be suspended
2187 * immediately and no further transaction commits will be honoured.
2189 * Any dirty, journaled buffers will be written back to disk without
2190 * hitting the journal. Atomicity cannot be guaranteed on an aborted
2191 * filesystem, but we _do_ attempt to leave as much data as possible
2192 * behind for fsck to use for cleanup.
2194 * Any attempt to get a new transaction handle on a journal which is in
2195 * ABORT state will just result in an -EROFS error return. A
2196 * jbd2_journal_stop on an existing handle will return -EIO if we have
2197 * entered abort state during the update.
2199 * Recursive transactions are not disturbed by journal abort until the
2200 * final jbd2_journal_stop, which will receive the -EIO error.
2202 * Finally, the jbd2_journal_abort call allows the caller to supply an errno
2203 * which will be recorded (if possible) in the journal superblock. This
2204 * allows a client to record failure conditions in the middle of a
2205 * transaction without having to complete the transaction to record the
2206 * failure to disk. ext3_error, for example, now uses this
2209 * Errors which originate from within the journaling layer will NOT
2210 * supply an errno; a null errno implies that absolutely no further
2211 * writes are done to the journal (unless there are any already in
2216 void jbd2_journal_abort(journal_t *journal, int errno)
2218 __journal_abort_soft(journal, errno);
2222 * int jbd2_journal_errno () - returns the journal's error state.
2223 * @journal: journal to examine.
2225 * This is the errno number set with jbd2_journal_abort(), the last
2226 * time the journal was mounted - if the journal was stopped
2227 * without calling abort this will be 0.
2229 * If the journal has been aborted on this mount time -EROFS will
2232 int jbd2_journal_errno(journal_t *journal)
2236 read_lock(&journal->j_state_lock);
2237 if (journal->j_flags & JBD2_ABORT)
2240 err = journal->j_errno;
2241 read_unlock(&journal->j_state_lock);
2246 * int jbd2_journal_clear_err () - clears the journal's error state
2247 * @journal: journal to act on.
2249 * An error must be cleared or acked to take a FS out of readonly
2252 int jbd2_journal_clear_err(journal_t *journal)
2256 write_lock(&journal->j_state_lock);
2257 if (journal->j_flags & JBD2_ABORT)
2260 journal->j_errno = 0;
2261 write_unlock(&journal->j_state_lock);
2266 * void jbd2_journal_ack_err() - Ack journal err.
2267 * @journal: journal to act on.
2269 * An error must be cleared or acked to take a FS out of readonly
2272 void jbd2_journal_ack_err(journal_t *journal)
2274 write_lock(&journal->j_state_lock);
2275 if (journal->j_errno)
2276 journal->j_flags |= JBD2_ACK_ERR;
2277 write_unlock(&journal->j_state_lock);
2280 int jbd2_journal_blocks_per_page(struct inode *inode)
2282 return 1 << (PAGE_SHIFT - inode->i_sb->s_blocksize_bits);
2286 * helper functions to deal with 32 or 64bit block numbers.
2288 size_t journal_tag_bytes(journal_t *journal)
2292 if (jbd2_has_feature_csum3(journal))
2293 return sizeof(journal_block_tag3_t);
2295 sz = sizeof(journal_block_tag_t);
2297 if (jbd2_has_feature_csum2(journal))
2298 sz += sizeof(__u16);
2300 if (jbd2_has_feature_64bit(journal))
2303 return sz - sizeof(__u32);
2307 * JBD memory management
2309 * These functions are used to allocate block-sized chunks of memory
2310 * used for making copies of buffer_head data. Very often it will be
2311 * page-sized chunks of data, but sometimes it will be in
2312 * sub-page-size chunks. (For example, 16k pages on Power systems
2313 * with a 4k block file system.) For blocks smaller than a page, we
2314 * use a SLAB allocator. There are slab caches for each block size,
2315 * which are allocated at mount time, if necessary, and we only free
2316 * (all of) the slab caches when/if the jbd2 module is unloaded. For
2317 * this reason we don't need to a mutex to protect access to
2318 * jbd2_slab[] allocating or releasing memory; only in
2319 * jbd2_journal_create_slab().
2321 #define JBD2_MAX_SLABS 8
2322 static struct kmem_cache *jbd2_slab[JBD2_MAX_SLABS];
2324 static const char *jbd2_slab_names[JBD2_MAX_SLABS] = {
2325 "jbd2_1k", "jbd2_2k", "jbd2_4k", "jbd2_8k",
2326 "jbd2_16k", "jbd2_32k", "jbd2_64k", "jbd2_128k"
2330 static void jbd2_journal_destroy_slabs(void)
2334 for (i = 0; i < JBD2_MAX_SLABS; i++) {
2335 kmem_cache_destroy(jbd2_slab[i]);
2336 jbd2_slab[i] = NULL;
2340 static int jbd2_journal_create_slab(size_t size)
2342 static DEFINE_MUTEX(jbd2_slab_create_mutex);
2343 int i = order_base_2(size) - 10;
2346 if (size == PAGE_SIZE)
2349 if (i >= JBD2_MAX_SLABS)
2352 if (unlikely(i < 0))
2354 mutex_lock(&jbd2_slab_create_mutex);
2356 mutex_unlock(&jbd2_slab_create_mutex);
2357 return 0; /* Already created */
2360 slab_size = 1 << (i+10);
2361 jbd2_slab[i] = kmem_cache_create(jbd2_slab_names[i], slab_size,
2362 slab_size, 0, NULL);
2363 mutex_unlock(&jbd2_slab_create_mutex);
2364 if (!jbd2_slab[i]) {
2365 printk(KERN_EMERG "JBD2: no memory for jbd2_slab cache\n");
2371 static struct kmem_cache *get_slab(size_t size)
2373 int i = order_base_2(size) - 10;
2375 BUG_ON(i >= JBD2_MAX_SLABS);
2376 if (unlikely(i < 0))
2378 BUG_ON(jbd2_slab[i] == NULL);
2379 return jbd2_slab[i];
2382 void *jbd2_alloc(size_t size, gfp_t flags)
2386 BUG_ON(size & (size-1)); /* Must be a power of 2 */
2388 if (size < PAGE_SIZE)
2389 ptr = kmem_cache_alloc(get_slab(size), flags);
2391 ptr = (void *)__get_free_pages(flags, get_order(size));
2393 /* Check alignment; SLUB has gotten this wrong in the past,
2394 * and this can lead to user data corruption! */
2395 BUG_ON(((unsigned long) ptr) & (size-1));
2400 void jbd2_free(void *ptr, size_t size)
2402 if (size < PAGE_SIZE)
2403 kmem_cache_free(get_slab(size), ptr);
2405 free_pages((unsigned long)ptr, get_order(size));
2409 * Journal_head storage management
2411 static struct kmem_cache *jbd2_journal_head_cache;
2412 #ifdef CONFIG_JBD2_DEBUG
2413 static atomic_t nr_journal_heads = ATOMIC_INIT(0);
2416 static int __init jbd2_journal_init_journal_head_cache(void)
2418 J_ASSERT(!jbd2_journal_head_cache);
2419 jbd2_journal_head_cache = kmem_cache_create("jbd2_journal_head",
2420 sizeof(struct journal_head),
2422 SLAB_TEMPORARY | SLAB_TYPESAFE_BY_RCU,
2424 if (!jbd2_journal_head_cache) {
2425 printk(KERN_EMERG "JBD2: no memory for journal_head cache\n");
2431 static void jbd2_journal_destroy_journal_head_cache(void)
2433 kmem_cache_destroy(jbd2_journal_head_cache);
2434 jbd2_journal_head_cache = NULL;
2438 * journal_head splicing and dicing
2440 static struct journal_head *journal_alloc_journal_head(void)
2442 struct journal_head *ret;
2444 #ifdef CONFIG_JBD2_DEBUG
2445 atomic_inc(&nr_journal_heads);
2447 ret = kmem_cache_zalloc(jbd2_journal_head_cache, GFP_NOFS);
2449 jbd_debug(1, "out of memory for journal_head\n");
2450 pr_notice_ratelimited("ENOMEM in %s, retrying.\n", __func__);
2451 ret = kmem_cache_zalloc(jbd2_journal_head_cache,
2452 GFP_NOFS | __GFP_NOFAIL);
2455 spin_lock_init(&ret->b_state_lock);
2459 static void journal_free_journal_head(struct journal_head *jh)
2461 #ifdef CONFIG_JBD2_DEBUG
2462 atomic_dec(&nr_journal_heads);
2463 memset(jh, JBD2_POISON_FREE, sizeof(*jh));
2465 kmem_cache_free(jbd2_journal_head_cache, jh);
2469 * A journal_head is attached to a buffer_head whenever JBD has an
2470 * interest in the buffer.
2472 * Whenever a buffer has an attached journal_head, its ->b_state:BH_JBD bit
2473 * is set. This bit is tested in core kernel code where we need to take
2474 * JBD-specific actions. Testing the zeroness of ->b_private is not reliable
2477 * When a buffer has its BH_JBD bit set, its ->b_count is elevated by one.
2479 * When a buffer has its BH_JBD bit set it is immune from being released by
2480 * core kernel code, mainly via ->b_count.
2482 * A journal_head is detached from its buffer_head when the journal_head's
2483 * b_jcount reaches zero. Running transaction (b_transaction) and checkpoint
2484 * transaction (b_cp_transaction) hold their references to b_jcount.
2486 * Various places in the kernel want to attach a journal_head to a buffer_head
2487 * _before_ attaching the journal_head to a transaction. To protect the
2488 * journal_head in this situation, jbd2_journal_add_journal_head elevates the
2489 * journal_head's b_jcount refcount by one. The caller must call
2490 * jbd2_journal_put_journal_head() to undo this.
2492 * So the typical usage would be:
2494 * (Attach a journal_head if needed. Increments b_jcount)
2495 * struct journal_head *jh = jbd2_journal_add_journal_head(bh);
2497 * (Get another reference for transaction)
2498 * jbd2_journal_grab_journal_head(bh);
2499 * jh->b_transaction = xxx;
2500 * (Put original reference)
2501 * jbd2_journal_put_journal_head(jh);
2505 * Give a buffer_head a journal_head.
2509 struct journal_head *jbd2_journal_add_journal_head(struct buffer_head *bh)
2511 struct journal_head *jh;
2512 struct journal_head *new_jh = NULL;
2515 if (!buffer_jbd(bh))
2516 new_jh = journal_alloc_journal_head();
2518 jbd_lock_bh_journal_head(bh);
2519 if (buffer_jbd(bh)) {
2523 (atomic_read(&bh->b_count) > 0) ||
2524 (bh->b_page && bh->b_page->mapping));
2527 jbd_unlock_bh_journal_head(bh);
2532 new_jh = NULL; /* We consumed it */
2537 BUFFER_TRACE(bh, "added journal_head");
2540 jbd_unlock_bh_journal_head(bh);
2542 journal_free_journal_head(new_jh);
2543 return bh->b_private;
2547 * Grab a ref against this buffer_head's journal_head. If it ended up not
2548 * having a journal_head, return NULL
2550 struct journal_head *jbd2_journal_grab_journal_head(struct buffer_head *bh)
2552 struct journal_head *jh = NULL;
2554 jbd_lock_bh_journal_head(bh);
2555 if (buffer_jbd(bh)) {
2559 jbd_unlock_bh_journal_head(bh);
2563 static void __journal_remove_journal_head(struct buffer_head *bh)
2565 struct journal_head *jh = bh2jh(bh);
2567 J_ASSERT_JH(jh, jh->b_jcount >= 0);
2568 J_ASSERT_JH(jh, jh->b_transaction == NULL);
2569 J_ASSERT_JH(jh, jh->b_next_transaction == NULL);
2570 J_ASSERT_JH(jh, jh->b_cp_transaction == NULL);
2571 J_ASSERT_JH(jh, jh->b_jlist == BJ_None);
2572 J_ASSERT_BH(bh, buffer_jbd(bh));
2573 J_ASSERT_BH(bh, jh2bh(jh) == bh);
2574 BUFFER_TRACE(bh, "remove journal_head");
2576 /* Unlink before dropping the lock */
2577 bh->b_private = NULL;
2578 jh->b_bh = NULL; /* debug, really */
2579 clear_buffer_jbd(bh);
2582 static void journal_release_journal_head(struct journal_head *jh, size_t b_size)
2584 if (jh->b_frozen_data) {
2585 printk(KERN_WARNING "%s: freeing b_frozen_data\n", __func__);
2586 jbd2_free(jh->b_frozen_data, b_size);
2588 if (jh->b_committed_data) {
2589 printk(KERN_WARNING "%s: freeing b_committed_data\n", __func__);
2590 jbd2_free(jh->b_committed_data, b_size);
2592 journal_free_journal_head(jh);
2596 * Drop a reference on the passed journal_head. If it fell to zero then
2597 * release the journal_head from the buffer_head.
2599 void jbd2_journal_put_journal_head(struct journal_head *jh)
2601 struct buffer_head *bh = jh2bh(jh);
2603 jbd_lock_bh_journal_head(bh);
2604 J_ASSERT_JH(jh, jh->b_jcount > 0);
2606 if (!jh->b_jcount) {
2607 __journal_remove_journal_head(bh);
2608 jbd_unlock_bh_journal_head(bh);
2609 journal_release_journal_head(jh, bh->b_size);
2612 jbd_unlock_bh_journal_head(bh);
2617 * Initialize jbd inode head
2619 void jbd2_journal_init_jbd_inode(struct jbd2_inode *jinode, struct inode *inode)
2621 jinode->i_transaction = NULL;
2622 jinode->i_next_transaction = NULL;
2623 jinode->i_vfs_inode = inode;
2624 jinode->i_flags = 0;
2625 jinode->i_dirty_start = 0;
2626 jinode->i_dirty_end = 0;
2627 INIT_LIST_HEAD(&jinode->i_list);
2631 * Function to be called before we start removing inode from memory (i.e.,
2632 * clear_inode() is a fine place to be called from). It removes inode from
2633 * transaction's lists.
2635 void jbd2_journal_release_jbd_inode(journal_t *journal,
2636 struct jbd2_inode *jinode)
2641 spin_lock(&journal->j_list_lock);
2642 /* Is commit writing out inode - we have to wait */
2643 if (jinode->i_flags & JI_COMMIT_RUNNING) {
2644 wait_queue_head_t *wq;
2645 DEFINE_WAIT_BIT(wait, &jinode->i_flags, __JI_COMMIT_RUNNING);
2646 wq = bit_waitqueue(&jinode->i_flags, __JI_COMMIT_RUNNING);
2647 prepare_to_wait(wq, &wait.wq_entry, TASK_UNINTERRUPTIBLE);
2648 spin_unlock(&journal->j_list_lock);
2650 finish_wait(wq, &wait.wq_entry);
2654 if (jinode->i_transaction) {
2655 list_del(&jinode->i_list);
2656 jinode->i_transaction = NULL;
2658 spin_unlock(&journal->j_list_lock);
2662 #ifdef CONFIG_PROC_FS
2664 #define JBD2_STATS_PROC_NAME "fs/jbd2"
2666 static void __init jbd2_create_jbd_stats_proc_entry(void)
2668 proc_jbd2_stats = proc_mkdir(JBD2_STATS_PROC_NAME, NULL);
2671 static void __exit jbd2_remove_jbd_stats_proc_entry(void)
2673 if (proc_jbd2_stats)
2674 remove_proc_entry(JBD2_STATS_PROC_NAME, NULL);
2679 #define jbd2_create_jbd_stats_proc_entry() do {} while (0)
2680 #define jbd2_remove_jbd_stats_proc_entry() do {} while (0)
2684 struct kmem_cache *jbd2_handle_cache, *jbd2_inode_cache;
2686 static int __init jbd2_journal_init_inode_cache(void)
2688 J_ASSERT(!jbd2_inode_cache);
2689 jbd2_inode_cache = KMEM_CACHE(jbd2_inode, 0);
2690 if (!jbd2_inode_cache) {
2691 pr_emerg("JBD2: failed to create inode cache\n");
2697 static int __init jbd2_journal_init_handle_cache(void)
2699 J_ASSERT(!jbd2_handle_cache);
2700 jbd2_handle_cache = KMEM_CACHE(jbd2_journal_handle, SLAB_TEMPORARY);
2701 if (!jbd2_handle_cache) {
2702 printk(KERN_EMERG "JBD2: failed to create handle cache\n");
2708 static void jbd2_journal_destroy_inode_cache(void)
2710 kmem_cache_destroy(jbd2_inode_cache);
2711 jbd2_inode_cache = NULL;
2714 static void jbd2_journal_destroy_handle_cache(void)
2716 kmem_cache_destroy(jbd2_handle_cache);
2717 jbd2_handle_cache = NULL;
2721 * Module startup and shutdown
2724 static int __init journal_init_caches(void)
2728 ret = jbd2_journal_init_revoke_record_cache();
2730 ret = jbd2_journal_init_revoke_table_cache();
2732 ret = jbd2_journal_init_journal_head_cache();
2734 ret = jbd2_journal_init_handle_cache();
2736 ret = jbd2_journal_init_inode_cache();
2738 ret = jbd2_journal_init_transaction_cache();
2742 static void jbd2_journal_destroy_caches(void)
2744 jbd2_journal_destroy_revoke_record_cache();
2745 jbd2_journal_destroy_revoke_table_cache();
2746 jbd2_journal_destroy_journal_head_cache();
2747 jbd2_journal_destroy_handle_cache();
2748 jbd2_journal_destroy_inode_cache();
2749 jbd2_journal_destroy_transaction_cache();
2750 jbd2_journal_destroy_slabs();
2753 static int __init journal_init(void)
2757 BUILD_BUG_ON(sizeof(struct journal_superblock_s) != 1024);
2759 ret = journal_init_caches();
2761 jbd2_create_jbd_stats_proc_entry();
2763 jbd2_journal_destroy_caches();
2768 static void __exit journal_exit(void)
2770 #ifdef CONFIG_JBD2_DEBUG
2771 int n = atomic_read(&nr_journal_heads);
2773 printk(KERN_ERR "JBD2: leaked %d journal_heads!\n", n);
2775 jbd2_remove_jbd_stats_proc_entry();
2776 jbd2_journal_destroy_caches();
2779 MODULE_LICENSE("GPL");
2780 module_init(journal_init);
2781 module_exit(journal_exit);