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;
799 if (journal->j_inode) {
800 ret = bmap(journal->j_inode, blocknr);
804 printk(KERN_ALERT "%s: journal block not found "
805 "at offset %lu on %s\n",
806 __func__, blocknr, journal->j_devname);
808 __journal_abort_soft(journal, err);
811 *retp = blocknr; /* +journal->j_blk_offset */
817 * We play buffer_head aliasing tricks to write data/metadata blocks to
818 * the journal without copying their contents, but for journal
819 * descriptor blocks we do need to generate bona fide buffers.
821 * After the caller of jbd2_journal_get_descriptor_buffer() has finished modifying
822 * the buffer's contents they really should run flush_dcache_page(bh->b_page).
823 * But we don't bother doing that, so there will be coherency problems with
824 * mmaps of blockdevs which hold live JBD-controlled filesystems.
827 jbd2_journal_get_descriptor_buffer(transaction_t *transaction, int type)
829 journal_t *journal = transaction->t_journal;
830 struct buffer_head *bh;
831 unsigned long long blocknr;
832 journal_header_t *header;
835 err = jbd2_journal_next_log_block(journal, &blocknr);
840 bh = __getblk(journal->j_dev, blocknr, journal->j_blocksize);
844 memset(bh->b_data, 0, journal->j_blocksize);
845 header = (journal_header_t *)bh->b_data;
846 header->h_magic = cpu_to_be32(JBD2_MAGIC_NUMBER);
847 header->h_blocktype = cpu_to_be32(type);
848 header->h_sequence = cpu_to_be32(transaction->t_tid);
849 set_buffer_uptodate(bh);
851 BUFFER_TRACE(bh, "return this buffer");
855 void jbd2_descriptor_block_csum_set(journal_t *j, struct buffer_head *bh)
857 struct jbd2_journal_block_tail *tail;
860 if (!jbd2_journal_has_csum_v2or3(j))
863 tail = (struct jbd2_journal_block_tail *)(bh->b_data + j->j_blocksize -
864 sizeof(struct jbd2_journal_block_tail));
865 tail->t_checksum = 0;
866 csum = jbd2_chksum(j, j->j_csum_seed, bh->b_data, j->j_blocksize);
867 tail->t_checksum = cpu_to_be32(csum);
871 * Return tid of the oldest transaction in the journal and block in the journal
872 * where the transaction starts.
874 * If the journal is now empty, return which will be the next transaction ID
875 * we will write and where will that transaction start.
877 * The return value is 0 if journal tail cannot be pushed any further, 1 if
880 int jbd2_journal_get_log_tail(journal_t *journal, tid_t *tid,
881 unsigned long *block)
883 transaction_t *transaction;
886 read_lock(&journal->j_state_lock);
887 spin_lock(&journal->j_list_lock);
888 transaction = journal->j_checkpoint_transactions;
890 *tid = transaction->t_tid;
891 *block = transaction->t_log_start;
892 } else if ((transaction = journal->j_committing_transaction) != NULL) {
893 *tid = transaction->t_tid;
894 *block = transaction->t_log_start;
895 } else if ((transaction = journal->j_running_transaction) != NULL) {
896 *tid = transaction->t_tid;
897 *block = journal->j_head;
899 *tid = journal->j_transaction_sequence;
900 *block = journal->j_head;
902 ret = tid_gt(*tid, journal->j_tail_sequence);
903 spin_unlock(&journal->j_list_lock);
904 read_unlock(&journal->j_state_lock);
910 * Update information in journal structure and in on disk journal superblock
911 * about log tail. This function does not check whether information passed in
912 * really pushes log tail further. It's responsibility of the caller to make
913 * sure provided log tail information is valid (e.g. by holding
914 * j_checkpoint_mutex all the time between computing log tail and calling this
915 * function as is the case with jbd2_cleanup_journal_tail()).
917 * Requires j_checkpoint_mutex
919 int __jbd2_update_log_tail(journal_t *journal, tid_t tid, unsigned long block)
924 BUG_ON(!mutex_is_locked(&journal->j_checkpoint_mutex));
927 * We cannot afford for write to remain in drive's caches since as
928 * soon as we update j_tail, next transaction can start reusing journal
929 * space and if we lose sb update during power failure we'd replay
930 * old transaction with possibly newly overwritten data.
932 ret = jbd2_journal_update_sb_log_tail(journal, tid, block,
937 write_lock(&journal->j_state_lock);
938 freed = block - journal->j_tail;
939 if (block < journal->j_tail)
940 freed += journal->j_last - journal->j_first;
942 trace_jbd2_update_log_tail(journal, tid, block, freed);
944 "Cleaning journal tail from %u to %u (offset %lu), "
946 journal->j_tail_sequence, tid, block, freed);
948 journal->j_free += freed;
949 journal->j_tail_sequence = tid;
950 journal->j_tail = block;
951 write_unlock(&journal->j_state_lock);
958 * This is a variation of __jbd2_update_log_tail which checks for validity of
959 * provided log tail and locks j_checkpoint_mutex. So it is safe against races
960 * with other threads updating log tail.
962 void jbd2_update_log_tail(journal_t *journal, tid_t tid, unsigned long block)
964 mutex_lock_io(&journal->j_checkpoint_mutex);
965 if (tid_gt(tid, journal->j_tail_sequence))
966 __jbd2_update_log_tail(journal, tid, block);
967 mutex_unlock(&journal->j_checkpoint_mutex);
970 struct jbd2_stats_proc_session {
972 struct transaction_stats_s *stats;
977 static void *jbd2_seq_info_start(struct seq_file *seq, loff_t *pos)
979 return *pos ? NULL : SEQ_START_TOKEN;
982 static void *jbd2_seq_info_next(struct seq_file *seq, void *v, loff_t *pos)
987 static int jbd2_seq_info_show(struct seq_file *seq, void *v)
989 struct jbd2_stats_proc_session *s = seq->private;
991 if (v != SEQ_START_TOKEN)
993 seq_printf(seq, "%lu transactions (%lu requested), "
994 "each up to %u blocks\n",
995 s->stats->ts_tid, s->stats->ts_requested,
996 s->journal->j_max_transaction_buffers);
997 if (s->stats->ts_tid == 0)
999 seq_printf(seq, "average: \n %ums waiting for transaction\n",
1000 jiffies_to_msecs(s->stats->run.rs_wait / s->stats->ts_tid));
1001 seq_printf(seq, " %ums request delay\n",
1002 (s->stats->ts_requested == 0) ? 0 :
1003 jiffies_to_msecs(s->stats->run.rs_request_delay /
1004 s->stats->ts_requested));
1005 seq_printf(seq, " %ums running transaction\n",
1006 jiffies_to_msecs(s->stats->run.rs_running / s->stats->ts_tid));
1007 seq_printf(seq, " %ums transaction was being locked\n",
1008 jiffies_to_msecs(s->stats->run.rs_locked / s->stats->ts_tid));
1009 seq_printf(seq, " %ums flushing data (in ordered mode)\n",
1010 jiffies_to_msecs(s->stats->run.rs_flushing / s->stats->ts_tid));
1011 seq_printf(seq, " %ums logging transaction\n",
1012 jiffies_to_msecs(s->stats->run.rs_logging / s->stats->ts_tid));
1013 seq_printf(seq, " %lluus average transaction commit time\n",
1014 div_u64(s->journal->j_average_commit_time, 1000));
1015 seq_printf(seq, " %lu handles per transaction\n",
1016 s->stats->run.rs_handle_count / s->stats->ts_tid);
1017 seq_printf(seq, " %lu blocks per transaction\n",
1018 s->stats->run.rs_blocks / s->stats->ts_tid);
1019 seq_printf(seq, " %lu logged blocks per transaction\n",
1020 s->stats->run.rs_blocks_logged / s->stats->ts_tid);
1024 static void jbd2_seq_info_stop(struct seq_file *seq, void *v)
1028 static const struct seq_operations jbd2_seq_info_ops = {
1029 .start = jbd2_seq_info_start,
1030 .next = jbd2_seq_info_next,
1031 .stop = jbd2_seq_info_stop,
1032 .show = jbd2_seq_info_show,
1035 static int jbd2_seq_info_open(struct inode *inode, struct file *file)
1037 journal_t *journal = PDE_DATA(inode);
1038 struct jbd2_stats_proc_session *s;
1041 s = kmalloc(sizeof(*s), GFP_KERNEL);
1044 size = sizeof(struct transaction_stats_s);
1045 s->stats = kmalloc(size, GFP_KERNEL);
1046 if (s->stats == NULL) {
1050 spin_lock(&journal->j_history_lock);
1051 memcpy(s->stats, &journal->j_stats, size);
1052 s->journal = journal;
1053 spin_unlock(&journal->j_history_lock);
1055 rc = seq_open(file, &jbd2_seq_info_ops);
1057 struct seq_file *m = file->private_data;
1067 static int jbd2_seq_info_release(struct inode *inode, struct file *file)
1069 struct seq_file *seq = file->private_data;
1070 struct jbd2_stats_proc_session *s = seq->private;
1073 return seq_release(inode, file);
1076 static const struct file_operations jbd2_seq_info_fops = {
1077 .owner = THIS_MODULE,
1078 .open = jbd2_seq_info_open,
1080 .llseek = seq_lseek,
1081 .release = jbd2_seq_info_release,
1084 static struct proc_dir_entry *proc_jbd2_stats;
1086 static void jbd2_stats_proc_init(journal_t *journal)
1088 journal->j_proc_entry = proc_mkdir(journal->j_devname, proc_jbd2_stats);
1089 if (journal->j_proc_entry) {
1090 proc_create_data("info", S_IRUGO, journal->j_proc_entry,
1091 &jbd2_seq_info_fops, journal);
1095 static void jbd2_stats_proc_exit(journal_t *journal)
1097 remove_proc_entry("info", journal->j_proc_entry);
1098 remove_proc_entry(journal->j_devname, proc_jbd2_stats);
1102 * Management for journal control blocks: functions to create and
1103 * destroy journal_t structures, and to initialise and read existing
1104 * journal blocks from disk. */
1106 /* First: create and setup a journal_t object in memory. We initialise
1107 * very few fields yet: that has to wait until we have created the
1108 * journal structures from from scratch, or loaded them from disk. */
1110 static journal_t *journal_init_common(struct block_device *bdev,
1111 struct block_device *fs_dev,
1112 unsigned long long start, int len, int blocksize)
1114 static struct lock_class_key jbd2_trans_commit_key;
1117 struct buffer_head *bh;
1120 journal = kzalloc(sizeof(*journal), GFP_KERNEL);
1124 init_waitqueue_head(&journal->j_wait_transaction_locked);
1125 init_waitqueue_head(&journal->j_wait_done_commit);
1126 init_waitqueue_head(&journal->j_wait_commit);
1127 init_waitqueue_head(&journal->j_wait_updates);
1128 init_waitqueue_head(&journal->j_wait_reserved);
1129 mutex_init(&journal->j_barrier);
1130 mutex_init(&journal->j_checkpoint_mutex);
1131 spin_lock_init(&journal->j_revoke_lock);
1132 spin_lock_init(&journal->j_list_lock);
1133 rwlock_init(&journal->j_state_lock);
1135 journal->j_commit_interval = (HZ * JBD2_DEFAULT_MAX_COMMIT_AGE);
1136 journal->j_min_batch_time = 0;
1137 journal->j_max_batch_time = 15000; /* 15ms */
1138 atomic_set(&journal->j_reserved_credits, 0);
1140 /* The journal is marked for error until we succeed with recovery! */
1141 journal->j_flags = JBD2_ABORT;
1143 /* Set up a default-sized revoke table for the new mount. */
1144 err = jbd2_journal_init_revoke(journal, JOURNAL_REVOKE_DEFAULT_HASH);
1148 spin_lock_init(&journal->j_history_lock);
1150 lockdep_init_map(&journal->j_trans_commit_map, "jbd2_handle",
1151 &jbd2_trans_commit_key, 0);
1153 /* journal descriptor can store up to n blocks -bzzz */
1154 journal->j_blocksize = blocksize;
1155 journal->j_dev = bdev;
1156 journal->j_fs_dev = fs_dev;
1157 journal->j_blk_offset = start;
1158 journal->j_maxlen = len;
1159 n = journal->j_blocksize / sizeof(journal_block_tag_t);
1160 journal->j_wbufsize = n;
1161 journal->j_wbuf = kmalloc_array(n, sizeof(struct buffer_head *),
1163 if (!journal->j_wbuf)
1166 bh = getblk_unmovable(journal->j_dev, start, journal->j_blocksize);
1168 pr_err("%s: Cannot get buffer for journal superblock\n",
1172 journal->j_sb_buffer = bh;
1173 journal->j_superblock = (journal_superblock_t *)bh->b_data;
1178 kfree(journal->j_wbuf);
1179 jbd2_journal_destroy_revoke(journal);
1184 /* jbd2_journal_init_dev and jbd2_journal_init_inode:
1186 * Create a journal structure assigned some fixed set of disk blocks to
1187 * the journal. We don't actually touch those disk blocks yet, but we
1188 * need to set up all of the mapping information to tell the journaling
1189 * system where the journal blocks are.
1194 * journal_t * jbd2_journal_init_dev() - creates and initialises a journal structure
1195 * @bdev: Block device on which to create the journal
1196 * @fs_dev: Device which hold journalled filesystem for this journal.
1197 * @start: Block nr Start of journal.
1198 * @len: Length of the journal in blocks.
1199 * @blocksize: blocksize of journalling device
1201 * Returns: a newly created journal_t *
1203 * jbd2_journal_init_dev creates a journal which maps a fixed contiguous
1204 * range of blocks on an arbitrary block device.
1207 journal_t *jbd2_journal_init_dev(struct block_device *bdev,
1208 struct block_device *fs_dev,
1209 unsigned long long start, int len, int blocksize)
1213 journal = journal_init_common(bdev, fs_dev, start, len, blocksize);
1217 bdevname(journal->j_dev, journal->j_devname);
1218 strreplace(journal->j_devname, '/', '!');
1219 jbd2_stats_proc_init(journal);
1225 * journal_t * jbd2_journal_init_inode () - creates a journal which maps to a inode.
1226 * @inode: An inode to create the journal in
1228 * jbd2_journal_init_inode creates a journal which maps an on-disk inode as
1229 * the journal. The inode must exist already, must support bmap() and
1230 * must have all data blocks preallocated.
1232 journal_t *jbd2_journal_init_inode(struct inode *inode)
1236 unsigned long long blocknr;
1238 blocknr = bmap(inode, 0);
1240 pr_err("%s: Cannot locate journal superblock\n",
1245 jbd_debug(1, "JBD2: inode %s/%ld, size %lld, bits %d, blksize %ld\n",
1246 inode->i_sb->s_id, inode->i_ino, (long long) inode->i_size,
1247 inode->i_sb->s_blocksize_bits, inode->i_sb->s_blocksize);
1249 journal = journal_init_common(inode->i_sb->s_bdev, inode->i_sb->s_bdev,
1250 blocknr, inode->i_size >> inode->i_sb->s_blocksize_bits,
1251 inode->i_sb->s_blocksize);
1255 journal->j_inode = inode;
1256 bdevname(journal->j_dev, journal->j_devname);
1257 p = strreplace(journal->j_devname, '/', '!');
1258 sprintf(p, "-%lu", journal->j_inode->i_ino);
1259 jbd2_stats_proc_init(journal);
1265 * If the journal init or create aborts, we need to mark the journal
1266 * superblock as being NULL to prevent the journal destroy from writing
1267 * back a bogus superblock.
1269 static void journal_fail_superblock (journal_t *journal)
1271 struct buffer_head *bh = journal->j_sb_buffer;
1273 journal->j_sb_buffer = NULL;
1277 * Given a journal_t structure, initialise the various fields for
1278 * startup of a new journaling session. We use this both when creating
1279 * a journal, and after recovering an old journal to reset it for
1283 static int journal_reset(journal_t *journal)
1285 journal_superblock_t *sb = journal->j_superblock;
1286 unsigned long long first, last;
1288 first = be32_to_cpu(sb->s_first);
1289 last = be32_to_cpu(sb->s_maxlen);
1290 if (first + JBD2_MIN_JOURNAL_BLOCKS > last + 1) {
1291 printk(KERN_ERR "JBD2: Journal too short (blocks %llu-%llu).\n",
1293 journal_fail_superblock(journal);
1297 journal->j_first = first;
1298 journal->j_last = last;
1300 journal->j_head = first;
1301 journal->j_tail = first;
1302 journal->j_free = last - first;
1304 journal->j_tail_sequence = journal->j_transaction_sequence;
1305 journal->j_commit_sequence = journal->j_transaction_sequence - 1;
1306 journal->j_commit_request = journal->j_commit_sequence;
1308 journal->j_max_transaction_buffers = journal->j_maxlen / 4;
1311 * As a special case, if the on-disk copy is already marked as needing
1312 * no recovery (s_start == 0), then we can safely defer the superblock
1313 * update until the next commit by setting JBD2_FLUSHED. This avoids
1314 * attempting a write to a potential-readonly device.
1316 if (sb->s_start == 0) {
1317 jbd_debug(1, "JBD2: Skipping superblock update on recovered sb "
1318 "(start %ld, seq %u, errno %d)\n",
1319 journal->j_tail, journal->j_tail_sequence,
1321 journal->j_flags |= JBD2_FLUSHED;
1323 /* Lock here to make assertions happy... */
1324 mutex_lock_io(&journal->j_checkpoint_mutex);
1326 * Update log tail information. We use REQ_FUA since new
1327 * transaction will start reusing journal space and so we
1328 * must make sure information about current log tail is on
1331 jbd2_journal_update_sb_log_tail(journal,
1332 journal->j_tail_sequence,
1334 REQ_SYNC | REQ_FUA);
1335 mutex_unlock(&journal->j_checkpoint_mutex);
1337 return jbd2_journal_start_thread(journal);
1341 * This function expects that the caller will have locked the journal
1342 * buffer head, and will return with it unlocked
1344 static int jbd2_write_superblock(journal_t *journal, int write_flags)
1346 struct buffer_head *bh = journal->j_sb_buffer;
1347 journal_superblock_t *sb = journal->j_superblock;
1350 /* Buffer got discarded which means block device got invalidated */
1351 if (!buffer_mapped(bh))
1354 trace_jbd2_write_superblock(journal, write_flags);
1355 if (!(journal->j_flags & JBD2_BARRIER))
1356 write_flags &= ~(REQ_FUA | REQ_PREFLUSH);
1357 if (buffer_write_io_error(bh)) {
1359 * Oh, dear. A previous attempt to write the journal
1360 * superblock failed. This could happen because the
1361 * USB device was yanked out. Or it could happen to
1362 * be a transient write error and maybe the block will
1363 * be remapped. Nothing we can do but to retry the
1364 * write and hope for the best.
1366 printk(KERN_ERR "JBD2: previous I/O error detected "
1367 "for journal superblock update for %s.\n",
1368 journal->j_devname);
1369 clear_buffer_write_io_error(bh);
1370 set_buffer_uptodate(bh);
1372 if (jbd2_journal_has_csum_v2or3(journal))
1373 sb->s_checksum = jbd2_superblock_csum(journal, sb);
1375 bh->b_end_io = end_buffer_write_sync;
1376 ret = submit_bh(REQ_OP_WRITE, write_flags, bh);
1378 if (buffer_write_io_error(bh)) {
1379 clear_buffer_write_io_error(bh);
1380 set_buffer_uptodate(bh);
1384 printk(KERN_ERR "JBD2: Error %d detected when updating "
1385 "journal superblock for %s.\n", ret,
1386 journal->j_devname);
1387 jbd2_journal_abort(journal, ret);
1394 * jbd2_journal_update_sb_log_tail() - Update log tail in journal sb on disk.
1395 * @journal: The journal to update.
1396 * @tail_tid: TID of the new transaction at the tail of the log
1397 * @tail_block: The first block of the transaction at the tail of the log
1398 * @write_op: With which operation should we write the journal sb
1400 * Update a journal's superblock information about log tail and write it to
1401 * disk, waiting for the IO to complete.
1403 int jbd2_journal_update_sb_log_tail(journal_t *journal, tid_t tail_tid,
1404 unsigned long tail_block, int write_op)
1406 journal_superblock_t *sb = journal->j_superblock;
1409 if (is_journal_aborted(journal))
1412 BUG_ON(!mutex_is_locked(&journal->j_checkpoint_mutex));
1413 jbd_debug(1, "JBD2: updating superblock (start %lu, seq %u)\n",
1414 tail_block, tail_tid);
1416 lock_buffer(journal->j_sb_buffer);
1417 sb->s_sequence = cpu_to_be32(tail_tid);
1418 sb->s_start = cpu_to_be32(tail_block);
1420 ret = jbd2_write_superblock(journal, write_op);
1424 /* Log is no longer empty */
1425 write_lock(&journal->j_state_lock);
1426 WARN_ON(!sb->s_sequence);
1427 journal->j_flags &= ~JBD2_FLUSHED;
1428 write_unlock(&journal->j_state_lock);
1435 * jbd2_mark_journal_empty() - Mark on disk journal as empty.
1436 * @journal: The journal to update.
1437 * @write_op: With which operation should we write the journal sb
1439 * Update a journal's dynamic superblock fields to show that journal is empty.
1440 * Write updated superblock to disk waiting for IO to complete.
1442 static void jbd2_mark_journal_empty(journal_t *journal, int write_op)
1444 journal_superblock_t *sb = journal->j_superblock;
1446 BUG_ON(!mutex_is_locked(&journal->j_checkpoint_mutex));
1447 lock_buffer(journal->j_sb_buffer);
1448 if (sb->s_start == 0) { /* Is it already empty? */
1449 unlock_buffer(journal->j_sb_buffer);
1453 jbd_debug(1, "JBD2: Marking journal as empty (seq %u)\n",
1454 journal->j_tail_sequence);
1456 sb->s_sequence = cpu_to_be32(journal->j_tail_sequence);
1457 sb->s_start = cpu_to_be32(0);
1459 jbd2_write_superblock(journal, write_op);
1461 /* Log is no longer empty */
1462 write_lock(&journal->j_state_lock);
1463 journal->j_flags |= JBD2_FLUSHED;
1464 write_unlock(&journal->j_state_lock);
1469 * jbd2_journal_update_sb_errno() - Update error in the journal.
1470 * @journal: The journal to update.
1472 * Update a journal's errno. Write updated superblock to disk waiting for IO
1475 void jbd2_journal_update_sb_errno(journal_t *journal)
1477 journal_superblock_t *sb = journal->j_superblock;
1480 lock_buffer(journal->j_sb_buffer);
1481 errcode = journal->j_errno;
1482 if (errcode == -ESHUTDOWN)
1484 jbd_debug(1, "JBD2: updating superblock error (errno %d)\n", errcode);
1485 sb->s_errno = cpu_to_be32(errcode);
1487 jbd2_write_superblock(journal, REQ_SYNC | REQ_FUA);
1489 EXPORT_SYMBOL(jbd2_journal_update_sb_errno);
1492 * Read the superblock for a given journal, performing initial
1493 * validation of the format.
1495 static int journal_get_superblock(journal_t *journal)
1497 struct buffer_head *bh;
1498 journal_superblock_t *sb;
1501 bh = journal->j_sb_buffer;
1503 J_ASSERT(bh != NULL);
1504 if (!buffer_uptodate(bh)) {
1505 ll_rw_block(REQ_OP_READ, 0, 1, &bh);
1507 if (!buffer_uptodate(bh)) {
1509 "JBD2: IO error reading journal superblock\n");
1514 if (buffer_verified(bh))
1517 sb = journal->j_superblock;
1521 if (sb->s_header.h_magic != cpu_to_be32(JBD2_MAGIC_NUMBER) ||
1522 sb->s_blocksize != cpu_to_be32(journal->j_blocksize)) {
1523 printk(KERN_WARNING "JBD2: no valid journal superblock found\n");
1527 switch(be32_to_cpu(sb->s_header.h_blocktype)) {
1528 case JBD2_SUPERBLOCK_V1:
1529 journal->j_format_version = 1;
1531 case JBD2_SUPERBLOCK_V2:
1532 journal->j_format_version = 2;
1535 printk(KERN_WARNING "JBD2: unrecognised superblock format ID\n");
1539 if (be32_to_cpu(sb->s_maxlen) < journal->j_maxlen)
1540 journal->j_maxlen = be32_to_cpu(sb->s_maxlen);
1541 else if (be32_to_cpu(sb->s_maxlen) > journal->j_maxlen) {
1542 printk(KERN_WARNING "JBD2: journal file too short\n");
1546 if (be32_to_cpu(sb->s_first) == 0 ||
1547 be32_to_cpu(sb->s_first) >= journal->j_maxlen) {
1549 "JBD2: Invalid start block of journal: %u\n",
1550 be32_to_cpu(sb->s_first));
1554 if (jbd2_has_feature_csum2(journal) &&
1555 jbd2_has_feature_csum3(journal)) {
1556 /* Can't have checksum v2 and v3 at the same time! */
1557 printk(KERN_ERR "JBD2: Can't enable checksumming v2 and v3 "
1558 "at the same time!\n");
1562 if (jbd2_journal_has_csum_v2or3_feature(journal) &&
1563 jbd2_has_feature_checksum(journal)) {
1564 /* Can't have checksum v1 and v2 on at the same time! */
1565 printk(KERN_ERR "JBD2: Can't enable checksumming v1 and v2/3 "
1566 "at the same time!\n");
1570 if (!jbd2_verify_csum_type(journal, sb)) {
1571 printk(KERN_ERR "JBD2: Unknown checksum type\n");
1575 /* Load the checksum driver */
1576 if (jbd2_journal_has_csum_v2or3_feature(journal)) {
1577 journal->j_chksum_driver = crypto_alloc_shash("crc32c", 0, 0);
1578 if (IS_ERR(journal->j_chksum_driver)) {
1579 printk(KERN_ERR "JBD2: Cannot load crc32c driver.\n");
1580 err = PTR_ERR(journal->j_chksum_driver);
1581 journal->j_chksum_driver = NULL;
1586 if (jbd2_journal_has_csum_v2or3(journal)) {
1587 /* Check superblock checksum */
1588 if (sb->s_checksum != jbd2_superblock_csum(journal, sb)) {
1589 printk(KERN_ERR "JBD2: journal checksum error\n");
1594 /* Precompute checksum seed for all metadata */
1595 journal->j_csum_seed = jbd2_chksum(journal, ~0, sb->s_uuid,
1596 sizeof(sb->s_uuid));
1599 set_buffer_verified(bh);
1604 journal_fail_superblock(journal);
1609 * Load the on-disk journal superblock and read the key fields into the
1613 static int load_superblock(journal_t *journal)
1616 journal_superblock_t *sb;
1618 err = journal_get_superblock(journal);
1622 sb = journal->j_superblock;
1624 journal->j_tail_sequence = be32_to_cpu(sb->s_sequence);
1625 journal->j_tail = be32_to_cpu(sb->s_start);
1626 journal->j_first = be32_to_cpu(sb->s_first);
1627 journal->j_last = be32_to_cpu(sb->s_maxlen);
1628 journal->j_errno = be32_to_cpu(sb->s_errno);
1635 * int jbd2_journal_load() - Read journal from disk.
1636 * @journal: Journal to act on.
1638 * Given a journal_t structure which tells us which disk blocks contain
1639 * a journal, read the journal from disk to initialise the in-memory
1642 int jbd2_journal_load(journal_t *journal)
1645 journal_superblock_t *sb;
1647 err = load_superblock(journal);
1651 sb = journal->j_superblock;
1652 /* If this is a V2 superblock, then we have to check the
1653 * features flags on it. */
1655 if (journal->j_format_version >= 2) {
1656 if ((sb->s_feature_ro_compat &
1657 ~cpu_to_be32(JBD2_KNOWN_ROCOMPAT_FEATURES)) ||
1658 (sb->s_feature_incompat &
1659 ~cpu_to_be32(JBD2_KNOWN_INCOMPAT_FEATURES))) {
1661 "JBD2: Unrecognised features on journal\n");
1667 * Create a slab for this blocksize
1669 err = jbd2_journal_create_slab(be32_to_cpu(sb->s_blocksize));
1673 /* Let the recovery code check whether it needs to recover any
1674 * data from the journal. */
1675 if (jbd2_journal_recover(journal))
1676 goto recovery_error;
1678 if (journal->j_failed_commit) {
1679 printk(KERN_ERR "JBD2: journal transaction %u on %s "
1680 "is corrupt.\n", journal->j_failed_commit,
1681 journal->j_devname);
1682 return -EFSCORRUPTED;
1685 /* OK, we've finished with the dynamic journal bits:
1686 * reinitialise the dynamic contents of the superblock in memory
1687 * and reset them on disk. */
1688 if (journal_reset(journal))
1689 goto recovery_error;
1691 journal->j_flags &= ~JBD2_ABORT;
1692 journal->j_flags |= JBD2_LOADED;
1696 printk(KERN_WARNING "JBD2: recovery failed\n");
1701 * void jbd2_journal_destroy() - Release a journal_t structure.
1702 * @journal: Journal to act on.
1704 * Release a journal_t structure once it is no longer in use by the
1706 * Return <0 if we couldn't clean up the journal.
1708 int jbd2_journal_destroy(journal_t *journal)
1712 /* Wait for the commit thread to wake up and die. */
1713 journal_kill_thread(journal);
1715 /* Force a final log commit */
1716 if (journal->j_running_transaction)
1717 jbd2_journal_commit_transaction(journal);
1719 /* Force any old transactions to disk */
1721 /* Totally anal locking here... */
1722 spin_lock(&journal->j_list_lock);
1723 while (journal->j_checkpoint_transactions != NULL) {
1724 spin_unlock(&journal->j_list_lock);
1725 mutex_lock_io(&journal->j_checkpoint_mutex);
1726 err = jbd2_log_do_checkpoint(journal);
1727 mutex_unlock(&journal->j_checkpoint_mutex);
1729 * If checkpointing failed, just free the buffers to avoid
1733 jbd2_journal_destroy_checkpoint(journal);
1734 spin_lock(&journal->j_list_lock);
1737 spin_lock(&journal->j_list_lock);
1740 J_ASSERT(journal->j_running_transaction == NULL);
1741 J_ASSERT(journal->j_committing_transaction == NULL);
1742 J_ASSERT(journal->j_checkpoint_transactions == NULL);
1743 spin_unlock(&journal->j_list_lock);
1745 if (journal->j_sb_buffer) {
1746 if (!is_journal_aborted(journal)) {
1747 mutex_lock_io(&journal->j_checkpoint_mutex);
1749 write_lock(&journal->j_state_lock);
1750 journal->j_tail_sequence =
1751 ++journal->j_transaction_sequence;
1752 write_unlock(&journal->j_state_lock);
1754 jbd2_mark_journal_empty(journal,
1755 REQ_SYNC | REQ_PREFLUSH | REQ_FUA);
1756 mutex_unlock(&journal->j_checkpoint_mutex);
1759 brelse(journal->j_sb_buffer);
1762 if (journal->j_proc_entry)
1763 jbd2_stats_proc_exit(journal);
1764 iput(journal->j_inode);
1765 if (journal->j_revoke)
1766 jbd2_journal_destroy_revoke(journal);
1767 if (journal->j_chksum_driver)
1768 crypto_free_shash(journal->j_chksum_driver);
1769 kfree(journal->j_wbuf);
1777 *int jbd2_journal_check_used_features () - Check if features specified are used.
1778 * @journal: Journal to check.
1779 * @compat: bitmask of compatible features
1780 * @ro: bitmask of features that force read-only mount
1781 * @incompat: bitmask of incompatible features
1783 * Check whether the journal uses all of a given set of
1784 * features. Return true (non-zero) if it does.
1787 int jbd2_journal_check_used_features (journal_t *journal, unsigned long compat,
1788 unsigned long ro, unsigned long incompat)
1790 journal_superblock_t *sb;
1792 if (!compat && !ro && !incompat)
1794 /* Load journal superblock if it is not loaded yet. */
1795 if (journal->j_format_version == 0 &&
1796 journal_get_superblock(journal) != 0)
1798 if (journal->j_format_version == 1)
1801 sb = journal->j_superblock;
1803 if (((be32_to_cpu(sb->s_feature_compat) & compat) == compat) &&
1804 ((be32_to_cpu(sb->s_feature_ro_compat) & ro) == ro) &&
1805 ((be32_to_cpu(sb->s_feature_incompat) & incompat) == incompat))
1812 * int jbd2_journal_check_available_features() - Check feature set in journalling layer
1813 * @journal: Journal to check.
1814 * @compat: bitmask of compatible features
1815 * @ro: bitmask of features that force read-only mount
1816 * @incompat: bitmask of incompatible features
1818 * Check whether the journaling code supports the use of
1819 * all of a given set of features on this journal. Return true
1820 * (non-zero) if it can. */
1822 int jbd2_journal_check_available_features (journal_t *journal, unsigned long compat,
1823 unsigned long ro, unsigned long incompat)
1825 if (!compat && !ro && !incompat)
1828 /* We can support any known requested features iff the
1829 * superblock is in version 2. Otherwise we fail to support any
1830 * extended sb features. */
1832 if (journal->j_format_version != 2)
1835 if ((compat & JBD2_KNOWN_COMPAT_FEATURES) == compat &&
1836 (ro & JBD2_KNOWN_ROCOMPAT_FEATURES) == ro &&
1837 (incompat & JBD2_KNOWN_INCOMPAT_FEATURES) == incompat)
1844 * int jbd2_journal_set_features () - Mark a given journal feature in the superblock
1845 * @journal: Journal to act on.
1846 * @compat: bitmask of compatible features
1847 * @ro: bitmask of features that force read-only mount
1848 * @incompat: bitmask of incompatible features
1850 * Mark a given journal feature as present on the
1851 * superblock. Returns true if the requested features could be set.
1855 int jbd2_journal_set_features (journal_t *journal, unsigned long compat,
1856 unsigned long ro, unsigned long incompat)
1858 #define INCOMPAT_FEATURE_ON(f) \
1859 ((incompat & (f)) && !(sb->s_feature_incompat & cpu_to_be32(f)))
1860 #define COMPAT_FEATURE_ON(f) \
1861 ((compat & (f)) && !(sb->s_feature_compat & cpu_to_be32(f)))
1862 journal_superblock_t *sb;
1864 if (jbd2_journal_check_used_features(journal, compat, ro, incompat))
1867 if (!jbd2_journal_check_available_features(journal, compat, ro, incompat))
1870 /* If enabling v2 checksums, turn on v3 instead */
1871 if (incompat & JBD2_FEATURE_INCOMPAT_CSUM_V2) {
1872 incompat &= ~JBD2_FEATURE_INCOMPAT_CSUM_V2;
1873 incompat |= JBD2_FEATURE_INCOMPAT_CSUM_V3;
1876 /* Asking for checksumming v3 and v1? Only give them v3. */
1877 if (incompat & JBD2_FEATURE_INCOMPAT_CSUM_V3 &&
1878 compat & JBD2_FEATURE_COMPAT_CHECKSUM)
1879 compat &= ~JBD2_FEATURE_COMPAT_CHECKSUM;
1881 jbd_debug(1, "Setting new features 0x%lx/0x%lx/0x%lx\n",
1882 compat, ro, incompat);
1884 sb = journal->j_superblock;
1886 /* Load the checksum driver if necessary */
1887 if ((journal->j_chksum_driver == NULL) &&
1888 INCOMPAT_FEATURE_ON(JBD2_FEATURE_INCOMPAT_CSUM_V3)) {
1889 journal->j_chksum_driver = crypto_alloc_shash("crc32c", 0, 0);
1890 if (IS_ERR(journal->j_chksum_driver)) {
1891 printk(KERN_ERR "JBD2: Cannot load crc32c driver.\n");
1892 journal->j_chksum_driver = NULL;
1895 /* Precompute checksum seed for all metadata */
1896 journal->j_csum_seed = jbd2_chksum(journal, ~0, sb->s_uuid,
1897 sizeof(sb->s_uuid));
1900 lock_buffer(journal->j_sb_buffer);
1902 /* If enabling v3 checksums, update superblock */
1903 if (INCOMPAT_FEATURE_ON(JBD2_FEATURE_INCOMPAT_CSUM_V3)) {
1904 sb->s_checksum_type = JBD2_CRC32C_CHKSUM;
1905 sb->s_feature_compat &=
1906 ~cpu_to_be32(JBD2_FEATURE_COMPAT_CHECKSUM);
1909 /* If enabling v1 checksums, downgrade superblock */
1910 if (COMPAT_FEATURE_ON(JBD2_FEATURE_COMPAT_CHECKSUM))
1911 sb->s_feature_incompat &=
1912 ~cpu_to_be32(JBD2_FEATURE_INCOMPAT_CSUM_V2 |
1913 JBD2_FEATURE_INCOMPAT_CSUM_V3);
1915 sb->s_feature_compat |= cpu_to_be32(compat);
1916 sb->s_feature_ro_compat |= cpu_to_be32(ro);
1917 sb->s_feature_incompat |= cpu_to_be32(incompat);
1918 unlock_buffer(journal->j_sb_buffer);
1921 #undef COMPAT_FEATURE_ON
1922 #undef INCOMPAT_FEATURE_ON
1926 * jbd2_journal_clear_features () - Clear a given journal feature in the
1928 * @journal: Journal to act on.
1929 * @compat: bitmask of compatible features
1930 * @ro: bitmask of features that force read-only mount
1931 * @incompat: bitmask of incompatible features
1933 * Clear a given journal feature as present on the
1936 void jbd2_journal_clear_features(journal_t *journal, unsigned long compat,
1937 unsigned long ro, unsigned long incompat)
1939 journal_superblock_t *sb;
1941 jbd_debug(1, "Clear features 0x%lx/0x%lx/0x%lx\n",
1942 compat, ro, incompat);
1944 sb = journal->j_superblock;
1946 sb->s_feature_compat &= ~cpu_to_be32(compat);
1947 sb->s_feature_ro_compat &= ~cpu_to_be32(ro);
1948 sb->s_feature_incompat &= ~cpu_to_be32(incompat);
1950 EXPORT_SYMBOL(jbd2_journal_clear_features);
1953 * int jbd2_journal_flush () - Flush journal
1954 * @journal: Journal to act on.
1956 * Flush all data for a given journal to disk and empty the journal.
1957 * Filesystems can use this when remounting readonly to ensure that
1958 * recovery does not need to happen on remount.
1961 int jbd2_journal_flush(journal_t *journal)
1964 transaction_t *transaction = NULL;
1966 write_lock(&journal->j_state_lock);
1968 /* Force everything buffered to the log... */
1969 if (journal->j_running_transaction) {
1970 transaction = journal->j_running_transaction;
1971 __jbd2_log_start_commit(journal, transaction->t_tid);
1972 } else if (journal->j_committing_transaction)
1973 transaction = journal->j_committing_transaction;
1975 /* Wait for the log commit to complete... */
1977 tid_t tid = transaction->t_tid;
1979 write_unlock(&journal->j_state_lock);
1980 jbd2_log_wait_commit(journal, tid);
1982 write_unlock(&journal->j_state_lock);
1985 /* ...and flush everything in the log out to disk. */
1986 spin_lock(&journal->j_list_lock);
1987 while (!err && journal->j_checkpoint_transactions != NULL) {
1988 spin_unlock(&journal->j_list_lock);
1989 mutex_lock_io(&journal->j_checkpoint_mutex);
1990 err = jbd2_log_do_checkpoint(journal);
1991 mutex_unlock(&journal->j_checkpoint_mutex);
1992 spin_lock(&journal->j_list_lock);
1994 spin_unlock(&journal->j_list_lock);
1996 if (is_journal_aborted(journal))
1999 mutex_lock_io(&journal->j_checkpoint_mutex);
2001 err = jbd2_cleanup_journal_tail(journal);
2003 mutex_unlock(&journal->j_checkpoint_mutex);
2009 /* Finally, mark the journal as really needing no recovery.
2010 * This sets s_start==0 in the underlying superblock, which is
2011 * the magic code for a fully-recovered superblock. Any future
2012 * commits of data to the journal will restore the current
2014 jbd2_mark_journal_empty(journal, REQ_SYNC | REQ_FUA);
2015 mutex_unlock(&journal->j_checkpoint_mutex);
2016 write_lock(&journal->j_state_lock);
2017 J_ASSERT(!journal->j_running_transaction);
2018 J_ASSERT(!journal->j_committing_transaction);
2019 J_ASSERT(!journal->j_checkpoint_transactions);
2020 J_ASSERT(journal->j_head == journal->j_tail);
2021 J_ASSERT(journal->j_tail_sequence == journal->j_transaction_sequence);
2022 write_unlock(&journal->j_state_lock);
2028 * int jbd2_journal_wipe() - Wipe journal contents
2029 * @journal: Journal to act on.
2030 * @write: flag (see below)
2032 * Wipe out all of the contents of a journal, safely. This will produce
2033 * a warning if the journal contains any valid recovery information.
2034 * Must be called between journal_init_*() and jbd2_journal_load().
2036 * If 'write' is non-zero, then we wipe out the journal on disk; otherwise
2037 * we merely suppress recovery.
2040 int jbd2_journal_wipe(journal_t *journal, int write)
2044 J_ASSERT (!(journal->j_flags & JBD2_LOADED));
2046 err = load_superblock(journal);
2050 if (!journal->j_tail)
2053 printk(KERN_WARNING "JBD2: %s recovery information on journal\n",
2054 write ? "Clearing" : "Ignoring");
2056 err = jbd2_journal_skip_recovery(journal);
2058 /* Lock to make assertions happy... */
2059 mutex_lock_io(&journal->j_checkpoint_mutex);
2060 jbd2_mark_journal_empty(journal, REQ_SYNC | REQ_FUA);
2061 mutex_unlock(&journal->j_checkpoint_mutex);
2069 * Journal abort has very specific semantics, which we describe
2070 * for journal abort.
2072 * Two internal functions, which provide abort to the jbd layer
2077 * Quick version for internal journal use (doesn't lock the journal).
2078 * Aborts hard --- we mark the abort as occurred, but do _nothing_ else,
2079 * and don't attempt to make any other journal updates.
2081 void __jbd2_journal_abort_hard(journal_t *journal)
2083 transaction_t *transaction;
2085 if (journal->j_flags & JBD2_ABORT)
2088 printk(KERN_ERR "Aborting journal on device %s.\n",
2089 journal->j_devname);
2091 write_lock(&journal->j_state_lock);
2092 journal->j_flags |= JBD2_ABORT;
2093 transaction = journal->j_running_transaction;
2095 __jbd2_log_start_commit(journal, transaction->t_tid);
2096 write_unlock(&journal->j_state_lock);
2099 /* Soft abort: record the abort error status in the journal superblock,
2100 * but don't do any other IO. */
2101 static void __journal_abort_soft (journal_t *journal, int errno)
2105 write_lock(&journal->j_state_lock);
2106 old_errno = journal->j_errno;
2107 if (!journal->j_errno || errno == -ESHUTDOWN)
2108 journal->j_errno = errno;
2110 if (journal->j_flags & JBD2_ABORT) {
2111 write_unlock(&journal->j_state_lock);
2112 if (!old_errno && old_errno != -ESHUTDOWN &&
2113 errno == -ESHUTDOWN)
2114 jbd2_journal_update_sb_errno(journal);
2117 write_unlock(&journal->j_state_lock);
2119 __jbd2_journal_abort_hard(journal);
2122 jbd2_journal_update_sb_errno(journal);
2123 write_lock(&journal->j_state_lock);
2124 journal->j_flags |= JBD2_REC_ERR;
2125 write_unlock(&journal->j_state_lock);
2130 * void jbd2_journal_abort () - Shutdown the journal immediately.
2131 * @journal: the journal to shutdown.
2132 * @errno: an error number to record in the journal indicating
2133 * the reason for the shutdown.
2135 * Perform a complete, immediate shutdown of the ENTIRE
2136 * journal (not of a single transaction). This operation cannot be
2137 * undone without closing and reopening the journal.
2139 * The jbd2_journal_abort function is intended to support higher level error
2140 * recovery mechanisms such as the ext2/ext3 remount-readonly error
2143 * Journal abort has very specific semantics. Any existing dirty,
2144 * unjournaled buffers in the main filesystem will still be written to
2145 * disk by bdflush, but the journaling mechanism will be suspended
2146 * immediately and no further transaction commits will be honoured.
2148 * Any dirty, journaled buffers will be written back to disk without
2149 * hitting the journal. Atomicity cannot be guaranteed on an aborted
2150 * filesystem, but we _do_ attempt to leave as much data as possible
2151 * behind for fsck to use for cleanup.
2153 * Any attempt to get a new transaction handle on a journal which is in
2154 * ABORT state will just result in an -EROFS error return. A
2155 * jbd2_journal_stop on an existing handle will return -EIO if we have
2156 * entered abort state during the update.
2158 * Recursive transactions are not disturbed by journal abort until the
2159 * final jbd2_journal_stop, which will receive the -EIO error.
2161 * Finally, the jbd2_journal_abort call allows the caller to supply an errno
2162 * which will be recorded (if possible) in the journal superblock. This
2163 * allows a client to record failure conditions in the middle of a
2164 * transaction without having to complete the transaction to record the
2165 * failure to disk. ext3_error, for example, now uses this
2168 * Errors which originate from within the journaling layer will NOT
2169 * supply an errno; a null errno implies that absolutely no further
2170 * writes are done to the journal (unless there are any already in
2175 void jbd2_journal_abort(journal_t *journal, int errno)
2177 __journal_abort_soft(journal, errno);
2181 * int jbd2_journal_errno () - returns the journal's error state.
2182 * @journal: journal to examine.
2184 * This is the errno number set with jbd2_journal_abort(), the last
2185 * time the journal was mounted - if the journal was stopped
2186 * without calling abort this will be 0.
2188 * If the journal has been aborted on this mount time -EROFS will
2191 int jbd2_journal_errno(journal_t *journal)
2195 read_lock(&journal->j_state_lock);
2196 if (journal->j_flags & JBD2_ABORT)
2199 err = journal->j_errno;
2200 read_unlock(&journal->j_state_lock);
2205 * int jbd2_journal_clear_err () - clears the journal's error state
2206 * @journal: journal to act on.
2208 * An error must be cleared or acked to take a FS out of readonly
2211 int jbd2_journal_clear_err(journal_t *journal)
2215 write_lock(&journal->j_state_lock);
2216 if (journal->j_flags & JBD2_ABORT)
2219 journal->j_errno = 0;
2220 write_unlock(&journal->j_state_lock);
2225 * void jbd2_journal_ack_err() - Ack journal err.
2226 * @journal: journal to act on.
2228 * An error must be cleared or acked to take a FS out of readonly
2231 void jbd2_journal_ack_err(journal_t *journal)
2233 write_lock(&journal->j_state_lock);
2234 if (journal->j_errno)
2235 journal->j_flags |= JBD2_ACK_ERR;
2236 write_unlock(&journal->j_state_lock);
2239 int jbd2_journal_blocks_per_page(struct inode *inode)
2241 return 1 << (PAGE_SHIFT - inode->i_sb->s_blocksize_bits);
2245 * helper functions to deal with 32 or 64bit block numbers.
2247 size_t journal_tag_bytes(journal_t *journal)
2251 if (jbd2_has_feature_csum3(journal))
2252 return sizeof(journal_block_tag3_t);
2254 sz = sizeof(journal_block_tag_t);
2256 if (jbd2_has_feature_csum2(journal))
2257 sz += sizeof(__u16);
2259 if (jbd2_has_feature_64bit(journal))
2262 return sz - sizeof(__u32);
2266 * JBD memory management
2268 * These functions are used to allocate block-sized chunks of memory
2269 * used for making copies of buffer_head data. Very often it will be
2270 * page-sized chunks of data, but sometimes it will be in
2271 * sub-page-size chunks. (For example, 16k pages on Power systems
2272 * with a 4k block file system.) For blocks smaller than a page, we
2273 * use a SLAB allocator. There are slab caches for each block size,
2274 * which are allocated at mount time, if necessary, and we only free
2275 * (all of) the slab caches when/if the jbd2 module is unloaded. For
2276 * this reason we don't need to a mutex to protect access to
2277 * jbd2_slab[] allocating or releasing memory; only in
2278 * jbd2_journal_create_slab().
2280 #define JBD2_MAX_SLABS 8
2281 static struct kmem_cache *jbd2_slab[JBD2_MAX_SLABS];
2283 static const char *jbd2_slab_names[JBD2_MAX_SLABS] = {
2284 "jbd2_1k", "jbd2_2k", "jbd2_4k", "jbd2_8k",
2285 "jbd2_16k", "jbd2_32k", "jbd2_64k", "jbd2_128k"
2289 static void jbd2_journal_destroy_slabs(void)
2293 for (i = 0; i < JBD2_MAX_SLABS; i++) {
2294 kmem_cache_destroy(jbd2_slab[i]);
2295 jbd2_slab[i] = NULL;
2299 static int jbd2_journal_create_slab(size_t size)
2301 static DEFINE_MUTEX(jbd2_slab_create_mutex);
2302 int i = order_base_2(size) - 10;
2305 if (size == PAGE_SIZE)
2308 if (i >= JBD2_MAX_SLABS)
2311 if (unlikely(i < 0))
2313 mutex_lock(&jbd2_slab_create_mutex);
2315 mutex_unlock(&jbd2_slab_create_mutex);
2316 return 0; /* Already created */
2319 slab_size = 1 << (i+10);
2320 jbd2_slab[i] = kmem_cache_create(jbd2_slab_names[i], slab_size,
2321 slab_size, 0, NULL);
2322 mutex_unlock(&jbd2_slab_create_mutex);
2323 if (!jbd2_slab[i]) {
2324 printk(KERN_EMERG "JBD2: no memory for jbd2_slab cache\n");
2330 static struct kmem_cache *get_slab(size_t size)
2332 int i = order_base_2(size) - 10;
2334 BUG_ON(i >= JBD2_MAX_SLABS);
2335 if (unlikely(i < 0))
2337 BUG_ON(jbd2_slab[i] == NULL);
2338 return jbd2_slab[i];
2341 void *jbd2_alloc(size_t size, gfp_t flags)
2345 BUG_ON(size & (size-1)); /* Must be a power of 2 */
2347 if (size < PAGE_SIZE)
2348 ptr = kmem_cache_alloc(get_slab(size), flags);
2350 ptr = (void *)__get_free_pages(flags, get_order(size));
2352 /* Check alignment; SLUB has gotten this wrong in the past,
2353 * and this can lead to user data corruption! */
2354 BUG_ON(((unsigned long) ptr) & (size-1));
2359 void jbd2_free(void *ptr, size_t size)
2361 if (size < PAGE_SIZE)
2362 kmem_cache_free(get_slab(size), ptr);
2364 free_pages((unsigned long)ptr, get_order(size));
2368 * Journal_head storage management
2370 static struct kmem_cache *jbd2_journal_head_cache;
2371 #ifdef CONFIG_JBD2_DEBUG
2372 static atomic_t nr_journal_heads = ATOMIC_INIT(0);
2375 static int __init jbd2_journal_init_journal_head_cache(void)
2377 J_ASSERT(!jbd2_journal_head_cache);
2378 jbd2_journal_head_cache = kmem_cache_create("jbd2_journal_head",
2379 sizeof(struct journal_head),
2381 SLAB_TEMPORARY | SLAB_TYPESAFE_BY_RCU,
2383 if (!jbd2_journal_head_cache) {
2384 printk(KERN_EMERG "JBD2: no memory for journal_head cache\n");
2390 static void jbd2_journal_destroy_journal_head_cache(void)
2392 kmem_cache_destroy(jbd2_journal_head_cache);
2393 jbd2_journal_head_cache = NULL;
2397 * journal_head splicing and dicing
2399 static struct journal_head *journal_alloc_journal_head(void)
2401 struct journal_head *ret;
2403 #ifdef CONFIG_JBD2_DEBUG
2404 atomic_inc(&nr_journal_heads);
2406 ret = kmem_cache_zalloc(jbd2_journal_head_cache, GFP_NOFS);
2408 jbd_debug(1, "out of memory for journal_head\n");
2409 pr_notice_ratelimited("ENOMEM in %s, retrying.\n", __func__);
2410 ret = kmem_cache_zalloc(jbd2_journal_head_cache,
2411 GFP_NOFS | __GFP_NOFAIL);
2414 spin_lock_init(&ret->b_state_lock);
2418 static void journal_free_journal_head(struct journal_head *jh)
2420 #ifdef CONFIG_JBD2_DEBUG
2421 atomic_dec(&nr_journal_heads);
2422 memset(jh, JBD2_POISON_FREE, sizeof(*jh));
2424 kmem_cache_free(jbd2_journal_head_cache, jh);
2428 * A journal_head is attached to a buffer_head whenever JBD has an
2429 * interest in the buffer.
2431 * Whenever a buffer has an attached journal_head, its ->b_state:BH_JBD bit
2432 * is set. This bit is tested in core kernel code where we need to take
2433 * JBD-specific actions. Testing the zeroness of ->b_private is not reliable
2436 * When a buffer has its BH_JBD bit set, its ->b_count is elevated by one.
2438 * When a buffer has its BH_JBD bit set it is immune from being released by
2439 * core kernel code, mainly via ->b_count.
2441 * A journal_head is detached from its buffer_head when the journal_head's
2442 * b_jcount reaches zero. Running transaction (b_transaction) and checkpoint
2443 * transaction (b_cp_transaction) hold their references to b_jcount.
2445 * Various places in the kernel want to attach a journal_head to a buffer_head
2446 * _before_ attaching the journal_head to a transaction. To protect the
2447 * journal_head in this situation, jbd2_journal_add_journal_head elevates the
2448 * journal_head's b_jcount refcount by one. The caller must call
2449 * jbd2_journal_put_journal_head() to undo this.
2451 * So the typical usage would be:
2453 * (Attach a journal_head if needed. Increments b_jcount)
2454 * struct journal_head *jh = jbd2_journal_add_journal_head(bh);
2456 * (Get another reference for transaction)
2457 * jbd2_journal_grab_journal_head(bh);
2458 * jh->b_transaction = xxx;
2459 * (Put original reference)
2460 * jbd2_journal_put_journal_head(jh);
2464 * Give a buffer_head a journal_head.
2468 struct journal_head *jbd2_journal_add_journal_head(struct buffer_head *bh)
2470 struct journal_head *jh;
2471 struct journal_head *new_jh = NULL;
2474 if (!buffer_jbd(bh))
2475 new_jh = journal_alloc_journal_head();
2477 jbd_lock_bh_journal_head(bh);
2478 if (buffer_jbd(bh)) {
2482 (atomic_read(&bh->b_count) > 0) ||
2483 (bh->b_page && bh->b_page->mapping));
2486 jbd_unlock_bh_journal_head(bh);
2491 new_jh = NULL; /* We consumed it */
2496 BUFFER_TRACE(bh, "added journal_head");
2499 jbd_unlock_bh_journal_head(bh);
2501 journal_free_journal_head(new_jh);
2502 return bh->b_private;
2506 * Grab a ref against this buffer_head's journal_head. If it ended up not
2507 * having a journal_head, return NULL
2509 struct journal_head *jbd2_journal_grab_journal_head(struct buffer_head *bh)
2511 struct journal_head *jh = NULL;
2513 jbd_lock_bh_journal_head(bh);
2514 if (buffer_jbd(bh)) {
2518 jbd_unlock_bh_journal_head(bh);
2522 static void __journal_remove_journal_head(struct buffer_head *bh)
2524 struct journal_head *jh = bh2jh(bh);
2526 J_ASSERT_JH(jh, jh->b_jcount >= 0);
2527 J_ASSERT_JH(jh, jh->b_transaction == NULL);
2528 J_ASSERT_JH(jh, jh->b_next_transaction == NULL);
2529 J_ASSERT_JH(jh, jh->b_cp_transaction == NULL);
2530 J_ASSERT_JH(jh, jh->b_jlist == BJ_None);
2531 J_ASSERT_BH(bh, buffer_jbd(bh));
2532 J_ASSERT_BH(bh, jh2bh(jh) == bh);
2533 BUFFER_TRACE(bh, "remove journal_head");
2535 /* Unlink before dropping the lock */
2536 bh->b_private = NULL;
2537 jh->b_bh = NULL; /* debug, really */
2538 clear_buffer_jbd(bh);
2541 static void journal_release_journal_head(struct journal_head *jh, size_t b_size)
2543 if (jh->b_frozen_data) {
2544 printk(KERN_WARNING "%s: freeing b_frozen_data\n", __func__);
2545 jbd2_free(jh->b_frozen_data, b_size);
2547 if (jh->b_committed_data) {
2548 printk(KERN_WARNING "%s: freeing b_committed_data\n", __func__);
2549 jbd2_free(jh->b_committed_data, b_size);
2551 journal_free_journal_head(jh);
2555 * Drop a reference on the passed journal_head. If it fell to zero then
2556 * release the journal_head from the buffer_head.
2558 void jbd2_journal_put_journal_head(struct journal_head *jh)
2560 struct buffer_head *bh = jh2bh(jh);
2562 jbd_lock_bh_journal_head(bh);
2563 J_ASSERT_JH(jh, jh->b_jcount > 0);
2565 if (!jh->b_jcount) {
2566 __journal_remove_journal_head(bh);
2567 jbd_unlock_bh_journal_head(bh);
2568 journal_release_journal_head(jh, bh->b_size);
2571 jbd_unlock_bh_journal_head(bh);
2576 * Initialize jbd inode head
2578 void jbd2_journal_init_jbd_inode(struct jbd2_inode *jinode, struct inode *inode)
2580 jinode->i_transaction = NULL;
2581 jinode->i_next_transaction = NULL;
2582 jinode->i_vfs_inode = inode;
2583 jinode->i_flags = 0;
2584 jinode->i_dirty_start = 0;
2585 jinode->i_dirty_end = 0;
2586 INIT_LIST_HEAD(&jinode->i_list);
2590 * Function to be called before we start removing inode from memory (i.e.,
2591 * clear_inode() is a fine place to be called from). It removes inode from
2592 * transaction's lists.
2594 void jbd2_journal_release_jbd_inode(journal_t *journal,
2595 struct jbd2_inode *jinode)
2600 spin_lock(&journal->j_list_lock);
2601 /* Is commit writing out inode - we have to wait */
2602 if (jinode->i_flags & JI_COMMIT_RUNNING) {
2603 wait_queue_head_t *wq;
2604 DEFINE_WAIT_BIT(wait, &jinode->i_flags, __JI_COMMIT_RUNNING);
2605 wq = bit_waitqueue(&jinode->i_flags, __JI_COMMIT_RUNNING);
2606 prepare_to_wait(wq, &wait.wq_entry, TASK_UNINTERRUPTIBLE);
2607 spin_unlock(&journal->j_list_lock);
2609 finish_wait(wq, &wait.wq_entry);
2613 if (jinode->i_transaction) {
2614 list_del(&jinode->i_list);
2615 jinode->i_transaction = NULL;
2617 spin_unlock(&journal->j_list_lock);
2621 #ifdef CONFIG_PROC_FS
2623 #define JBD2_STATS_PROC_NAME "fs/jbd2"
2625 static void __init jbd2_create_jbd_stats_proc_entry(void)
2627 proc_jbd2_stats = proc_mkdir(JBD2_STATS_PROC_NAME, NULL);
2630 static void __exit jbd2_remove_jbd_stats_proc_entry(void)
2632 if (proc_jbd2_stats)
2633 remove_proc_entry(JBD2_STATS_PROC_NAME, NULL);
2638 #define jbd2_create_jbd_stats_proc_entry() do {} while (0)
2639 #define jbd2_remove_jbd_stats_proc_entry() do {} while (0)
2643 struct kmem_cache *jbd2_handle_cache, *jbd2_inode_cache;
2645 static int __init jbd2_journal_init_inode_cache(void)
2647 J_ASSERT(!jbd2_inode_cache);
2648 jbd2_inode_cache = KMEM_CACHE(jbd2_inode, 0);
2649 if (!jbd2_inode_cache) {
2650 pr_emerg("JBD2: failed to create inode cache\n");
2656 static int __init jbd2_journal_init_handle_cache(void)
2658 J_ASSERT(!jbd2_handle_cache);
2659 jbd2_handle_cache = KMEM_CACHE(jbd2_journal_handle, SLAB_TEMPORARY);
2660 if (!jbd2_handle_cache) {
2661 printk(KERN_EMERG "JBD2: failed to create handle cache\n");
2667 static void jbd2_journal_destroy_inode_cache(void)
2669 kmem_cache_destroy(jbd2_inode_cache);
2670 jbd2_inode_cache = NULL;
2673 static void jbd2_journal_destroy_handle_cache(void)
2675 kmem_cache_destroy(jbd2_handle_cache);
2676 jbd2_handle_cache = NULL;
2680 * Module startup and shutdown
2683 static int __init journal_init_caches(void)
2687 ret = jbd2_journal_init_revoke_record_cache();
2689 ret = jbd2_journal_init_revoke_table_cache();
2691 ret = jbd2_journal_init_journal_head_cache();
2693 ret = jbd2_journal_init_handle_cache();
2695 ret = jbd2_journal_init_inode_cache();
2697 ret = jbd2_journal_init_transaction_cache();
2701 static void jbd2_journal_destroy_caches(void)
2703 jbd2_journal_destroy_revoke_record_cache();
2704 jbd2_journal_destroy_revoke_table_cache();
2705 jbd2_journal_destroy_journal_head_cache();
2706 jbd2_journal_destroy_handle_cache();
2707 jbd2_journal_destroy_inode_cache();
2708 jbd2_journal_destroy_transaction_cache();
2709 jbd2_journal_destroy_slabs();
2712 static int __init journal_init(void)
2716 BUILD_BUG_ON(sizeof(struct journal_superblock_s) != 1024);
2718 ret = journal_init_caches();
2720 jbd2_create_jbd_stats_proc_entry();
2722 jbd2_journal_destroy_caches();
2727 static void __exit journal_exit(void)
2729 #ifdef CONFIG_JBD2_DEBUG
2730 int n = atomic_read(&nr_journal_heads);
2732 printk(KERN_ERR "JBD2: leaked %d journal_heads!\n", n);
2734 jbd2_remove_jbd_stats_proc_entry();
2735 jbd2_journal_destroy_caches();
2738 MODULE_LICENSE("GPL");
2739 module_init(journal_init);
2740 module_exit(journal_exit);