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);
843 atomic_dec(&transaction->t_outstanding_credits);
845 memset(bh->b_data, 0, journal->j_blocksize);
846 header = (journal_header_t *)bh->b_data;
847 header->h_magic = cpu_to_be32(JBD2_MAGIC_NUMBER);
848 header->h_blocktype = cpu_to_be32(type);
849 header->h_sequence = cpu_to_be32(transaction->t_tid);
850 set_buffer_uptodate(bh);
852 BUFFER_TRACE(bh, "return this buffer");
856 void jbd2_descriptor_block_csum_set(journal_t *j, struct buffer_head *bh)
858 struct jbd2_journal_block_tail *tail;
861 if (!jbd2_journal_has_csum_v2or3(j))
864 tail = (struct jbd2_journal_block_tail *)(bh->b_data + j->j_blocksize -
865 sizeof(struct jbd2_journal_block_tail));
866 tail->t_checksum = 0;
867 csum = jbd2_chksum(j, j->j_csum_seed, bh->b_data, j->j_blocksize);
868 tail->t_checksum = cpu_to_be32(csum);
872 * Return tid of the oldest transaction in the journal and block in the journal
873 * where the transaction starts.
875 * If the journal is now empty, return which will be the next transaction ID
876 * we will write and where will that transaction start.
878 * The return value is 0 if journal tail cannot be pushed any further, 1 if
881 int jbd2_journal_get_log_tail(journal_t *journal, tid_t *tid,
882 unsigned long *block)
884 transaction_t *transaction;
887 read_lock(&journal->j_state_lock);
888 spin_lock(&journal->j_list_lock);
889 transaction = journal->j_checkpoint_transactions;
891 *tid = transaction->t_tid;
892 *block = transaction->t_log_start;
893 } else if ((transaction = journal->j_committing_transaction) != NULL) {
894 *tid = transaction->t_tid;
895 *block = transaction->t_log_start;
896 } else if ((transaction = journal->j_running_transaction) != NULL) {
897 *tid = transaction->t_tid;
898 *block = journal->j_head;
900 *tid = journal->j_transaction_sequence;
901 *block = journal->j_head;
903 ret = tid_gt(*tid, journal->j_tail_sequence);
904 spin_unlock(&journal->j_list_lock);
905 read_unlock(&journal->j_state_lock);
911 * Update information in journal structure and in on disk journal superblock
912 * about log tail. This function does not check whether information passed in
913 * really pushes log tail further. It's responsibility of the caller to make
914 * sure provided log tail information is valid (e.g. by holding
915 * j_checkpoint_mutex all the time between computing log tail and calling this
916 * function as is the case with jbd2_cleanup_journal_tail()).
918 * Requires j_checkpoint_mutex
920 int __jbd2_update_log_tail(journal_t *journal, tid_t tid, unsigned long block)
925 BUG_ON(!mutex_is_locked(&journal->j_checkpoint_mutex));
928 * We cannot afford for write to remain in drive's caches since as
929 * soon as we update j_tail, next transaction can start reusing journal
930 * space and if we lose sb update during power failure we'd replay
931 * old transaction with possibly newly overwritten data.
933 ret = jbd2_journal_update_sb_log_tail(journal, tid, block,
938 write_lock(&journal->j_state_lock);
939 freed = block - journal->j_tail;
940 if (block < journal->j_tail)
941 freed += journal->j_last - journal->j_first;
943 trace_jbd2_update_log_tail(journal, tid, block, freed);
945 "Cleaning journal tail from %u to %u (offset %lu), "
947 journal->j_tail_sequence, tid, block, freed);
949 journal->j_free += freed;
950 journal->j_tail_sequence = tid;
951 journal->j_tail = block;
952 write_unlock(&journal->j_state_lock);
959 * This is a variation of __jbd2_update_log_tail which checks for validity of
960 * provided log tail and locks j_checkpoint_mutex. So it is safe against races
961 * with other threads updating log tail.
963 void jbd2_update_log_tail(journal_t *journal, tid_t tid, unsigned long block)
965 mutex_lock_io(&journal->j_checkpoint_mutex);
966 if (tid_gt(tid, journal->j_tail_sequence))
967 __jbd2_update_log_tail(journal, tid, block);
968 mutex_unlock(&journal->j_checkpoint_mutex);
971 struct jbd2_stats_proc_session {
973 struct transaction_stats_s *stats;
978 static void *jbd2_seq_info_start(struct seq_file *seq, loff_t *pos)
980 return *pos ? NULL : SEQ_START_TOKEN;
983 static void *jbd2_seq_info_next(struct seq_file *seq, void *v, loff_t *pos)
989 static int jbd2_seq_info_show(struct seq_file *seq, void *v)
991 struct jbd2_stats_proc_session *s = seq->private;
993 if (v != SEQ_START_TOKEN)
995 seq_printf(seq, "%lu transactions (%lu requested), "
996 "each up to %u blocks\n",
997 s->stats->ts_tid, s->stats->ts_requested,
998 s->journal->j_max_transaction_buffers);
999 if (s->stats->ts_tid == 0)
1001 seq_printf(seq, "average: \n %ums waiting for transaction\n",
1002 jiffies_to_msecs(s->stats->run.rs_wait / s->stats->ts_tid));
1003 seq_printf(seq, " %ums request delay\n",
1004 (s->stats->ts_requested == 0) ? 0 :
1005 jiffies_to_msecs(s->stats->run.rs_request_delay /
1006 s->stats->ts_requested));
1007 seq_printf(seq, " %ums running transaction\n",
1008 jiffies_to_msecs(s->stats->run.rs_running / s->stats->ts_tid));
1009 seq_printf(seq, " %ums transaction was being locked\n",
1010 jiffies_to_msecs(s->stats->run.rs_locked / s->stats->ts_tid));
1011 seq_printf(seq, " %ums flushing data (in ordered mode)\n",
1012 jiffies_to_msecs(s->stats->run.rs_flushing / s->stats->ts_tid));
1013 seq_printf(seq, " %ums logging transaction\n",
1014 jiffies_to_msecs(s->stats->run.rs_logging / s->stats->ts_tid));
1015 seq_printf(seq, " %lluus average transaction commit time\n",
1016 div_u64(s->journal->j_average_commit_time, 1000));
1017 seq_printf(seq, " %lu handles per transaction\n",
1018 s->stats->run.rs_handle_count / s->stats->ts_tid);
1019 seq_printf(seq, " %lu blocks per transaction\n",
1020 s->stats->run.rs_blocks / s->stats->ts_tid);
1021 seq_printf(seq, " %lu logged blocks per transaction\n",
1022 s->stats->run.rs_blocks_logged / s->stats->ts_tid);
1026 static void jbd2_seq_info_stop(struct seq_file *seq, void *v)
1030 static const struct seq_operations jbd2_seq_info_ops = {
1031 .start = jbd2_seq_info_start,
1032 .next = jbd2_seq_info_next,
1033 .stop = jbd2_seq_info_stop,
1034 .show = jbd2_seq_info_show,
1037 static int jbd2_seq_info_open(struct inode *inode, struct file *file)
1039 journal_t *journal = PDE_DATA(inode);
1040 struct jbd2_stats_proc_session *s;
1043 s = kmalloc(sizeof(*s), GFP_KERNEL);
1046 size = sizeof(struct transaction_stats_s);
1047 s->stats = kmalloc(size, GFP_KERNEL);
1048 if (s->stats == NULL) {
1052 spin_lock(&journal->j_history_lock);
1053 memcpy(s->stats, &journal->j_stats, size);
1054 s->journal = journal;
1055 spin_unlock(&journal->j_history_lock);
1057 rc = seq_open(file, &jbd2_seq_info_ops);
1059 struct seq_file *m = file->private_data;
1069 static int jbd2_seq_info_release(struct inode *inode, struct file *file)
1071 struct seq_file *seq = file->private_data;
1072 struct jbd2_stats_proc_session *s = seq->private;
1075 return seq_release(inode, file);
1078 static const struct file_operations jbd2_seq_info_fops = {
1079 .owner = THIS_MODULE,
1080 .open = jbd2_seq_info_open,
1082 .llseek = seq_lseek,
1083 .release = jbd2_seq_info_release,
1086 static struct proc_dir_entry *proc_jbd2_stats;
1088 static void jbd2_stats_proc_init(journal_t *journal)
1090 journal->j_proc_entry = proc_mkdir(journal->j_devname, proc_jbd2_stats);
1091 if (journal->j_proc_entry) {
1092 proc_create_data("info", S_IRUGO, journal->j_proc_entry,
1093 &jbd2_seq_info_fops, journal);
1097 static void jbd2_stats_proc_exit(journal_t *journal)
1099 remove_proc_entry("info", journal->j_proc_entry);
1100 remove_proc_entry(journal->j_devname, proc_jbd2_stats);
1103 /* Minimum size of descriptor tag */
1104 static int jbd2_min_tag_size(void)
1107 * Tag with 32-bit block numbers does not use last four bytes of the
1110 return sizeof(journal_block_tag_t) - 4;
1114 * Management for journal control blocks: functions to create and
1115 * destroy journal_t structures, and to initialise and read existing
1116 * journal blocks from disk. */
1118 /* First: create and setup a journal_t object in memory. We initialise
1119 * very few fields yet: that has to wait until we have created the
1120 * journal structures from from scratch, or loaded them from disk. */
1122 static journal_t *journal_init_common(struct block_device *bdev,
1123 struct block_device *fs_dev,
1124 unsigned long long start, int len, int blocksize)
1126 static struct lock_class_key jbd2_trans_commit_key;
1129 struct buffer_head *bh;
1132 journal = kzalloc(sizeof(*journal), GFP_KERNEL);
1136 init_waitqueue_head(&journal->j_wait_transaction_locked);
1137 init_waitqueue_head(&journal->j_wait_done_commit);
1138 init_waitqueue_head(&journal->j_wait_commit);
1139 init_waitqueue_head(&journal->j_wait_updates);
1140 init_waitqueue_head(&journal->j_wait_reserved);
1141 mutex_init(&journal->j_barrier);
1142 mutex_init(&journal->j_checkpoint_mutex);
1143 spin_lock_init(&journal->j_revoke_lock);
1144 spin_lock_init(&journal->j_list_lock);
1145 rwlock_init(&journal->j_state_lock);
1147 journal->j_commit_interval = (HZ * JBD2_DEFAULT_MAX_COMMIT_AGE);
1148 journal->j_min_batch_time = 0;
1149 journal->j_max_batch_time = 15000; /* 15ms */
1150 atomic_set(&journal->j_reserved_credits, 0);
1152 /* The journal is marked for error until we succeed with recovery! */
1153 journal->j_flags = JBD2_ABORT;
1155 /* Set up a default-sized revoke table for the new mount. */
1156 err = jbd2_journal_init_revoke(journal, JOURNAL_REVOKE_DEFAULT_HASH);
1160 spin_lock_init(&journal->j_history_lock);
1162 lockdep_init_map(&journal->j_trans_commit_map, "jbd2_handle",
1163 &jbd2_trans_commit_key, 0);
1165 /* journal descriptor can store up to n blocks -bzzz */
1166 journal->j_blocksize = blocksize;
1167 journal->j_dev = bdev;
1168 journal->j_fs_dev = fs_dev;
1169 journal->j_blk_offset = start;
1170 journal->j_maxlen = len;
1171 /* We need enough buffers to write out full descriptor block. */
1172 n = journal->j_blocksize / jbd2_min_tag_size();
1173 journal->j_wbufsize = n;
1174 journal->j_wbuf = kmalloc_array(n, sizeof(struct buffer_head *),
1176 if (!journal->j_wbuf)
1179 bh = getblk_unmovable(journal->j_dev, start, journal->j_blocksize);
1181 pr_err("%s: Cannot get buffer for journal superblock\n",
1185 journal->j_sb_buffer = bh;
1186 journal->j_superblock = (journal_superblock_t *)bh->b_data;
1191 kfree(journal->j_wbuf);
1192 jbd2_journal_destroy_revoke(journal);
1197 /* jbd2_journal_init_dev and jbd2_journal_init_inode:
1199 * Create a journal structure assigned some fixed set of disk blocks to
1200 * the journal. We don't actually touch those disk blocks yet, but we
1201 * need to set up all of the mapping information to tell the journaling
1202 * system where the journal blocks are.
1207 * journal_t * jbd2_journal_init_dev() - creates and initialises a journal structure
1208 * @bdev: Block device on which to create the journal
1209 * @fs_dev: Device which hold journalled filesystem for this journal.
1210 * @start: Block nr Start of journal.
1211 * @len: Length of the journal in blocks.
1212 * @blocksize: blocksize of journalling device
1214 * Returns: a newly created journal_t *
1216 * jbd2_journal_init_dev creates a journal which maps a fixed contiguous
1217 * range of blocks on an arbitrary block device.
1220 journal_t *jbd2_journal_init_dev(struct block_device *bdev,
1221 struct block_device *fs_dev,
1222 unsigned long long start, int len, int blocksize)
1226 journal = journal_init_common(bdev, fs_dev, start, len, blocksize);
1230 bdevname(journal->j_dev, journal->j_devname);
1231 strreplace(journal->j_devname, '/', '!');
1232 jbd2_stats_proc_init(journal);
1238 * journal_t * jbd2_journal_init_inode () - creates a journal which maps to a inode.
1239 * @inode: An inode to create the journal in
1241 * jbd2_journal_init_inode creates a journal which maps an on-disk inode as
1242 * the journal. The inode must exist already, must support bmap() and
1243 * must have all data blocks preallocated.
1245 journal_t *jbd2_journal_init_inode(struct inode *inode)
1249 unsigned long long blocknr;
1251 blocknr = bmap(inode, 0);
1253 pr_err("%s: Cannot locate journal superblock\n",
1258 jbd_debug(1, "JBD2: inode %s/%ld, size %lld, bits %d, blksize %ld\n",
1259 inode->i_sb->s_id, inode->i_ino, (long long) inode->i_size,
1260 inode->i_sb->s_blocksize_bits, inode->i_sb->s_blocksize);
1262 journal = journal_init_common(inode->i_sb->s_bdev, inode->i_sb->s_bdev,
1263 blocknr, inode->i_size >> inode->i_sb->s_blocksize_bits,
1264 inode->i_sb->s_blocksize);
1268 journal->j_inode = inode;
1269 bdevname(journal->j_dev, journal->j_devname);
1270 p = strreplace(journal->j_devname, '/', '!');
1271 sprintf(p, "-%lu", journal->j_inode->i_ino);
1272 jbd2_stats_proc_init(journal);
1278 * If the journal init or create aborts, we need to mark the journal
1279 * superblock as being NULL to prevent the journal destroy from writing
1280 * back a bogus superblock.
1282 static void journal_fail_superblock (journal_t *journal)
1284 struct buffer_head *bh = journal->j_sb_buffer;
1286 journal->j_sb_buffer = NULL;
1290 * Given a journal_t structure, initialise the various fields for
1291 * startup of a new journaling session. We use this both when creating
1292 * a journal, and after recovering an old journal to reset it for
1296 static int journal_reset(journal_t *journal)
1298 journal_superblock_t *sb = journal->j_superblock;
1299 unsigned long long first, last;
1301 first = be32_to_cpu(sb->s_first);
1302 last = be32_to_cpu(sb->s_maxlen);
1303 if (first + JBD2_MIN_JOURNAL_BLOCKS > last + 1) {
1304 printk(KERN_ERR "JBD2: Journal too short (blocks %llu-%llu).\n",
1306 journal_fail_superblock(journal);
1310 journal->j_first = first;
1311 journal->j_last = last;
1313 journal->j_head = first;
1314 journal->j_tail = first;
1315 journal->j_free = last - first;
1317 journal->j_tail_sequence = journal->j_transaction_sequence;
1318 journal->j_commit_sequence = journal->j_transaction_sequence - 1;
1319 journal->j_commit_request = journal->j_commit_sequence;
1321 journal->j_max_transaction_buffers = journal->j_maxlen / 4;
1324 * As a special case, if the on-disk copy is already marked as needing
1325 * no recovery (s_start == 0), then we can safely defer the superblock
1326 * update until the next commit by setting JBD2_FLUSHED. This avoids
1327 * attempting a write to a potential-readonly device.
1329 if (sb->s_start == 0) {
1330 jbd_debug(1, "JBD2: Skipping superblock update on recovered sb "
1331 "(start %ld, seq %u, errno %d)\n",
1332 journal->j_tail, journal->j_tail_sequence,
1334 journal->j_flags |= JBD2_FLUSHED;
1336 /* Lock here to make assertions happy... */
1337 mutex_lock_io(&journal->j_checkpoint_mutex);
1339 * Update log tail information. We use REQ_FUA since new
1340 * transaction will start reusing journal space and so we
1341 * must make sure information about current log tail is on
1344 jbd2_journal_update_sb_log_tail(journal,
1345 journal->j_tail_sequence,
1347 REQ_SYNC | REQ_FUA);
1348 mutex_unlock(&journal->j_checkpoint_mutex);
1350 return jbd2_journal_start_thread(journal);
1354 * This function expects that the caller will have locked the journal
1355 * buffer head, and will return with it unlocked
1357 static int jbd2_write_superblock(journal_t *journal, int write_flags)
1359 struct buffer_head *bh = journal->j_sb_buffer;
1360 journal_superblock_t *sb = journal->j_superblock;
1363 /* Buffer got discarded which means block device got invalidated */
1364 if (!buffer_mapped(bh))
1367 trace_jbd2_write_superblock(journal, write_flags);
1368 if (!(journal->j_flags & JBD2_BARRIER))
1369 write_flags &= ~(REQ_FUA | REQ_PREFLUSH);
1370 if (buffer_write_io_error(bh)) {
1372 * Oh, dear. A previous attempt to write the journal
1373 * superblock failed. This could happen because the
1374 * USB device was yanked out. Or it could happen to
1375 * be a transient write error and maybe the block will
1376 * be remapped. Nothing we can do but to retry the
1377 * write and hope for the best.
1379 printk(KERN_ERR "JBD2: previous I/O error detected "
1380 "for journal superblock update for %s.\n",
1381 journal->j_devname);
1382 clear_buffer_write_io_error(bh);
1383 set_buffer_uptodate(bh);
1385 if (jbd2_journal_has_csum_v2or3(journal))
1386 sb->s_checksum = jbd2_superblock_csum(journal, sb);
1388 bh->b_end_io = end_buffer_write_sync;
1389 ret = submit_bh(REQ_OP_WRITE, write_flags, bh);
1391 if (buffer_write_io_error(bh)) {
1392 clear_buffer_write_io_error(bh);
1393 set_buffer_uptodate(bh);
1397 printk(KERN_ERR "JBD2: Error %d detected when updating "
1398 "journal superblock for %s.\n", ret,
1399 journal->j_devname);
1400 jbd2_journal_abort(journal, ret);
1407 * jbd2_journal_update_sb_log_tail() - Update log tail in journal sb on disk.
1408 * @journal: The journal to update.
1409 * @tail_tid: TID of the new transaction at the tail of the log
1410 * @tail_block: The first block of the transaction at the tail of the log
1411 * @write_op: With which operation should we write the journal sb
1413 * Update a journal's superblock information about log tail and write it to
1414 * disk, waiting for the IO to complete.
1416 int jbd2_journal_update_sb_log_tail(journal_t *journal, tid_t tail_tid,
1417 unsigned long tail_block, int write_op)
1419 journal_superblock_t *sb = journal->j_superblock;
1422 if (is_journal_aborted(journal))
1425 BUG_ON(!mutex_is_locked(&journal->j_checkpoint_mutex));
1426 jbd_debug(1, "JBD2: updating superblock (start %lu, seq %u)\n",
1427 tail_block, tail_tid);
1429 lock_buffer(journal->j_sb_buffer);
1430 sb->s_sequence = cpu_to_be32(tail_tid);
1431 sb->s_start = cpu_to_be32(tail_block);
1433 ret = jbd2_write_superblock(journal, write_op);
1437 /* Log is no longer empty */
1438 write_lock(&journal->j_state_lock);
1439 WARN_ON(!sb->s_sequence);
1440 journal->j_flags &= ~JBD2_FLUSHED;
1441 write_unlock(&journal->j_state_lock);
1448 * jbd2_mark_journal_empty() - Mark on disk journal as empty.
1449 * @journal: The journal to update.
1450 * @write_op: With which operation should we write the journal sb
1452 * Update a journal's dynamic superblock fields to show that journal is empty.
1453 * Write updated superblock to disk waiting for IO to complete.
1455 static void jbd2_mark_journal_empty(journal_t *journal, int write_op)
1457 journal_superblock_t *sb = journal->j_superblock;
1459 BUG_ON(!mutex_is_locked(&journal->j_checkpoint_mutex));
1460 lock_buffer(journal->j_sb_buffer);
1461 if (sb->s_start == 0) { /* Is it already empty? */
1462 unlock_buffer(journal->j_sb_buffer);
1466 jbd_debug(1, "JBD2: Marking journal as empty (seq %u)\n",
1467 journal->j_tail_sequence);
1469 sb->s_sequence = cpu_to_be32(journal->j_tail_sequence);
1470 sb->s_start = cpu_to_be32(0);
1472 jbd2_write_superblock(journal, write_op);
1474 /* Log is no longer empty */
1475 write_lock(&journal->j_state_lock);
1476 journal->j_flags |= JBD2_FLUSHED;
1477 write_unlock(&journal->j_state_lock);
1482 * jbd2_journal_update_sb_errno() - Update error in the journal.
1483 * @journal: The journal to update.
1485 * Update a journal's errno. Write updated superblock to disk waiting for IO
1488 void jbd2_journal_update_sb_errno(journal_t *journal)
1490 journal_superblock_t *sb = journal->j_superblock;
1493 lock_buffer(journal->j_sb_buffer);
1494 errcode = journal->j_errno;
1495 if (errcode == -ESHUTDOWN)
1497 jbd_debug(1, "JBD2: updating superblock error (errno %d)\n", errcode);
1498 sb->s_errno = cpu_to_be32(errcode);
1500 jbd2_write_superblock(journal, REQ_SYNC | REQ_FUA);
1502 EXPORT_SYMBOL(jbd2_journal_update_sb_errno);
1504 static int journal_revoke_records_per_block(journal_t *journal)
1507 int space = journal->j_blocksize - sizeof(jbd2_journal_revoke_header_t);
1509 if (jbd2_has_feature_64bit(journal))
1514 if (jbd2_journal_has_csum_v2or3(journal))
1515 space -= sizeof(struct jbd2_journal_block_tail);
1516 return space / record_size;
1520 * Read the superblock for a given journal, performing initial
1521 * validation of the format.
1523 static int journal_get_superblock(journal_t *journal)
1525 struct buffer_head *bh;
1526 journal_superblock_t *sb;
1529 bh = journal->j_sb_buffer;
1531 J_ASSERT(bh != NULL);
1532 if (!buffer_uptodate(bh)) {
1533 ll_rw_block(REQ_OP_READ, 0, 1, &bh);
1535 if (!buffer_uptodate(bh)) {
1537 "JBD2: IO error reading journal superblock\n");
1542 if (buffer_verified(bh))
1545 sb = journal->j_superblock;
1549 if (sb->s_header.h_magic != cpu_to_be32(JBD2_MAGIC_NUMBER) ||
1550 sb->s_blocksize != cpu_to_be32(journal->j_blocksize)) {
1551 printk(KERN_WARNING "JBD2: no valid journal superblock found\n");
1555 switch(be32_to_cpu(sb->s_header.h_blocktype)) {
1556 case JBD2_SUPERBLOCK_V1:
1557 journal->j_format_version = 1;
1559 case JBD2_SUPERBLOCK_V2:
1560 journal->j_format_version = 2;
1563 printk(KERN_WARNING "JBD2: unrecognised superblock format ID\n");
1567 if (be32_to_cpu(sb->s_maxlen) < journal->j_maxlen)
1568 journal->j_maxlen = be32_to_cpu(sb->s_maxlen);
1569 else if (be32_to_cpu(sb->s_maxlen) > journal->j_maxlen) {
1570 printk(KERN_WARNING "JBD2: journal file too short\n");
1574 if (be32_to_cpu(sb->s_first) == 0 ||
1575 be32_to_cpu(sb->s_first) >= journal->j_maxlen) {
1577 "JBD2: Invalid start block of journal: %u\n",
1578 be32_to_cpu(sb->s_first));
1582 if (jbd2_has_feature_csum2(journal) &&
1583 jbd2_has_feature_csum3(journal)) {
1584 /* Can't have checksum v2 and v3 at the same time! */
1585 printk(KERN_ERR "JBD2: Can't enable checksumming v2 and v3 "
1586 "at the same time!\n");
1590 if (jbd2_journal_has_csum_v2or3_feature(journal) &&
1591 jbd2_has_feature_checksum(journal)) {
1592 /* Can't have checksum v1 and v2 on at the same time! */
1593 printk(KERN_ERR "JBD2: Can't enable checksumming v1 and v2/3 "
1594 "at the same time!\n");
1598 if (!jbd2_verify_csum_type(journal, sb)) {
1599 printk(KERN_ERR "JBD2: Unknown checksum type\n");
1603 /* Load the checksum driver */
1604 if (jbd2_journal_has_csum_v2or3_feature(journal)) {
1605 journal->j_chksum_driver = crypto_alloc_shash("crc32c", 0, 0);
1606 if (IS_ERR(journal->j_chksum_driver)) {
1607 printk(KERN_ERR "JBD2: Cannot load crc32c driver.\n");
1608 err = PTR_ERR(journal->j_chksum_driver);
1609 journal->j_chksum_driver = NULL;
1614 if (jbd2_journal_has_csum_v2or3(journal)) {
1615 /* Check superblock checksum */
1616 if (sb->s_checksum != jbd2_superblock_csum(journal, sb)) {
1617 printk(KERN_ERR "JBD2: journal checksum error\n");
1622 /* Precompute checksum seed for all metadata */
1623 journal->j_csum_seed = jbd2_chksum(journal, ~0, sb->s_uuid,
1624 sizeof(sb->s_uuid));
1627 journal->j_revoke_records_per_block =
1628 journal_revoke_records_per_block(journal);
1629 set_buffer_verified(bh);
1634 journal_fail_superblock(journal);
1639 * Load the on-disk journal superblock and read the key fields into the
1643 static int load_superblock(journal_t *journal)
1646 journal_superblock_t *sb;
1648 err = journal_get_superblock(journal);
1652 sb = journal->j_superblock;
1654 journal->j_tail_sequence = be32_to_cpu(sb->s_sequence);
1655 journal->j_tail = be32_to_cpu(sb->s_start);
1656 journal->j_first = be32_to_cpu(sb->s_first);
1657 journal->j_last = be32_to_cpu(sb->s_maxlen);
1658 journal->j_errno = be32_to_cpu(sb->s_errno);
1665 * int jbd2_journal_load() - Read journal from disk.
1666 * @journal: Journal to act on.
1668 * Given a journal_t structure which tells us which disk blocks contain
1669 * a journal, read the journal from disk to initialise the in-memory
1672 int jbd2_journal_load(journal_t *journal)
1675 journal_superblock_t *sb;
1677 err = load_superblock(journal);
1681 sb = journal->j_superblock;
1682 /* If this is a V2 superblock, then we have to check the
1683 * features flags on it. */
1685 if (journal->j_format_version >= 2) {
1686 if ((sb->s_feature_ro_compat &
1687 ~cpu_to_be32(JBD2_KNOWN_ROCOMPAT_FEATURES)) ||
1688 (sb->s_feature_incompat &
1689 ~cpu_to_be32(JBD2_KNOWN_INCOMPAT_FEATURES))) {
1691 "JBD2: Unrecognised features on journal\n");
1697 * Create a slab for this blocksize
1699 err = jbd2_journal_create_slab(be32_to_cpu(sb->s_blocksize));
1703 /* Let the recovery code check whether it needs to recover any
1704 * data from the journal. */
1705 if (jbd2_journal_recover(journal))
1706 goto recovery_error;
1708 if (journal->j_failed_commit) {
1709 printk(KERN_ERR "JBD2: journal transaction %u on %s "
1710 "is corrupt.\n", journal->j_failed_commit,
1711 journal->j_devname);
1712 return -EFSCORRUPTED;
1715 * clear JBD2_ABORT flag initialized in journal_init_common
1716 * here to update log tail information with the newest seq.
1718 journal->j_flags &= ~JBD2_ABORT;
1720 /* OK, we've finished with the dynamic journal bits:
1721 * reinitialise the dynamic contents of the superblock in memory
1722 * and reset them on disk. */
1723 if (journal_reset(journal))
1724 goto recovery_error;
1726 journal->j_flags |= JBD2_LOADED;
1730 printk(KERN_WARNING "JBD2: recovery failed\n");
1735 * void jbd2_journal_destroy() - Release a journal_t structure.
1736 * @journal: Journal to act on.
1738 * Release a journal_t structure once it is no longer in use by the
1740 * Return <0 if we couldn't clean up the journal.
1742 int jbd2_journal_destroy(journal_t *journal)
1746 /* Wait for the commit thread to wake up and die. */
1747 journal_kill_thread(journal);
1749 /* Force a final log commit */
1750 if (journal->j_running_transaction)
1751 jbd2_journal_commit_transaction(journal);
1753 /* Force any old transactions to disk */
1755 /* Totally anal locking here... */
1756 spin_lock(&journal->j_list_lock);
1757 while (journal->j_checkpoint_transactions != NULL) {
1758 spin_unlock(&journal->j_list_lock);
1759 mutex_lock_io(&journal->j_checkpoint_mutex);
1760 err = jbd2_log_do_checkpoint(journal);
1761 mutex_unlock(&journal->j_checkpoint_mutex);
1763 * If checkpointing failed, just free the buffers to avoid
1767 jbd2_journal_destroy_checkpoint(journal);
1768 spin_lock(&journal->j_list_lock);
1771 spin_lock(&journal->j_list_lock);
1774 J_ASSERT(journal->j_running_transaction == NULL);
1775 J_ASSERT(journal->j_committing_transaction == NULL);
1776 J_ASSERT(journal->j_checkpoint_transactions == NULL);
1777 spin_unlock(&journal->j_list_lock);
1779 if (journal->j_sb_buffer) {
1780 if (!is_journal_aborted(journal)) {
1781 mutex_lock_io(&journal->j_checkpoint_mutex);
1783 write_lock(&journal->j_state_lock);
1784 journal->j_tail_sequence =
1785 ++journal->j_transaction_sequence;
1786 write_unlock(&journal->j_state_lock);
1788 jbd2_mark_journal_empty(journal,
1789 REQ_SYNC | REQ_PREFLUSH | REQ_FUA);
1790 mutex_unlock(&journal->j_checkpoint_mutex);
1793 brelse(journal->j_sb_buffer);
1796 if (journal->j_proc_entry)
1797 jbd2_stats_proc_exit(journal);
1798 iput(journal->j_inode);
1799 if (journal->j_revoke)
1800 jbd2_journal_destroy_revoke(journal);
1801 if (journal->j_chksum_driver)
1802 crypto_free_shash(journal->j_chksum_driver);
1803 kfree(journal->j_wbuf);
1811 *int jbd2_journal_check_used_features () - Check if features specified are used.
1812 * @journal: Journal to check.
1813 * @compat: bitmask of compatible features
1814 * @ro: bitmask of features that force read-only mount
1815 * @incompat: bitmask of incompatible features
1817 * Check whether the journal uses all of a given set of
1818 * features. Return true (non-zero) if it does.
1821 int jbd2_journal_check_used_features (journal_t *journal, unsigned long compat,
1822 unsigned long ro, unsigned long incompat)
1824 journal_superblock_t *sb;
1826 if (!compat && !ro && !incompat)
1828 /* Load journal superblock if it is not loaded yet. */
1829 if (journal->j_format_version == 0 &&
1830 journal_get_superblock(journal) != 0)
1832 if (journal->j_format_version == 1)
1835 sb = journal->j_superblock;
1837 if (((be32_to_cpu(sb->s_feature_compat) & compat) == compat) &&
1838 ((be32_to_cpu(sb->s_feature_ro_compat) & ro) == ro) &&
1839 ((be32_to_cpu(sb->s_feature_incompat) & incompat) == incompat))
1846 * int jbd2_journal_check_available_features() - Check feature set in journalling layer
1847 * @journal: Journal to check.
1848 * @compat: bitmask of compatible features
1849 * @ro: bitmask of features that force read-only mount
1850 * @incompat: bitmask of incompatible features
1852 * Check whether the journaling code supports the use of
1853 * all of a given set of features on this journal. Return true
1854 * (non-zero) if it can. */
1856 int jbd2_journal_check_available_features (journal_t *journal, unsigned long compat,
1857 unsigned long ro, unsigned long incompat)
1859 if (!compat && !ro && !incompat)
1862 /* We can support any known requested features iff the
1863 * superblock is in version 2. Otherwise we fail to support any
1864 * extended sb features. */
1866 if (journal->j_format_version != 2)
1869 if ((compat & JBD2_KNOWN_COMPAT_FEATURES) == compat &&
1870 (ro & JBD2_KNOWN_ROCOMPAT_FEATURES) == ro &&
1871 (incompat & JBD2_KNOWN_INCOMPAT_FEATURES) == incompat)
1878 * int jbd2_journal_set_features () - Mark a given journal feature in the superblock
1879 * @journal: Journal to act on.
1880 * @compat: bitmask of compatible features
1881 * @ro: bitmask of features that force read-only mount
1882 * @incompat: bitmask of incompatible features
1884 * Mark a given journal feature as present on the
1885 * superblock. Returns true if the requested features could be set.
1889 int jbd2_journal_set_features (journal_t *journal, unsigned long compat,
1890 unsigned long ro, unsigned long incompat)
1892 #define INCOMPAT_FEATURE_ON(f) \
1893 ((incompat & (f)) && !(sb->s_feature_incompat & cpu_to_be32(f)))
1894 #define COMPAT_FEATURE_ON(f) \
1895 ((compat & (f)) && !(sb->s_feature_compat & cpu_to_be32(f)))
1896 journal_superblock_t *sb;
1898 if (jbd2_journal_check_used_features(journal, compat, ro, incompat))
1901 if (!jbd2_journal_check_available_features(journal, compat, ro, incompat))
1904 /* If enabling v2 checksums, turn on v3 instead */
1905 if (incompat & JBD2_FEATURE_INCOMPAT_CSUM_V2) {
1906 incompat &= ~JBD2_FEATURE_INCOMPAT_CSUM_V2;
1907 incompat |= JBD2_FEATURE_INCOMPAT_CSUM_V3;
1910 /* Asking for checksumming v3 and v1? Only give them v3. */
1911 if (incompat & JBD2_FEATURE_INCOMPAT_CSUM_V3 &&
1912 compat & JBD2_FEATURE_COMPAT_CHECKSUM)
1913 compat &= ~JBD2_FEATURE_COMPAT_CHECKSUM;
1915 jbd_debug(1, "Setting new features 0x%lx/0x%lx/0x%lx\n",
1916 compat, ro, incompat);
1918 sb = journal->j_superblock;
1920 /* Load the checksum driver if necessary */
1921 if ((journal->j_chksum_driver == NULL) &&
1922 INCOMPAT_FEATURE_ON(JBD2_FEATURE_INCOMPAT_CSUM_V3)) {
1923 journal->j_chksum_driver = crypto_alloc_shash("crc32c", 0, 0);
1924 if (IS_ERR(journal->j_chksum_driver)) {
1925 printk(KERN_ERR "JBD2: Cannot load crc32c driver.\n");
1926 journal->j_chksum_driver = NULL;
1929 /* Precompute checksum seed for all metadata */
1930 journal->j_csum_seed = jbd2_chksum(journal, ~0, sb->s_uuid,
1931 sizeof(sb->s_uuid));
1934 lock_buffer(journal->j_sb_buffer);
1936 /* If enabling v3 checksums, update superblock */
1937 if (INCOMPAT_FEATURE_ON(JBD2_FEATURE_INCOMPAT_CSUM_V3)) {
1938 sb->s_checksum_type = JBD2_CRC32C_CHKSUM;
1939 sb->s_feature_compat &=
1940 ~cpu_to_be32(JBD2_FEATURE_COMPAT_CHECKSUM);
1943 /* If enabling v1 checksums, downgrade superblock */
1944 if (COMPAT_FEATURE_ON(JBD2_FEATURE_COMPAT_CHECKSUM))
1945 sb->s_feature_incompat &=
1946 ~cpu_to_be32(JBD2_FEATURE_INCOMPAT_CSUM_V2 |
1947 JBD2_FEATURE_INCOMPAT_CSUM_V3);
1949 sb->s_feature_compat |= cpu_to_be32(compat);
1950 sb->s_feature_ro_compat |= cpu_to_be32(ro);
1951 sb->s_feature_incompat |= cpu_to_be32(incompat);
1952 unlock_buffer(journal->j_sb_buffer);
1953 journal->j_revoke_records_per_block =
1954 journal_revoke_records_per_block(journal);
1957 #undef COMPAT_FEATURE_ON
1958 #undef INCOMPAT_FEATURE_ON
1962 * jbd2_journal_clear_features () - Clear a given journal feature in the
1964 * @journal: Journal to act on.
1965 * @compat: bitmask of compatible features
1966 * @ro: bitmask of features that force read-only mount
1967 * @incompat: bitmask of incompatible features
1969 * Clear a given journal feature as present on the
1972 void jbd2_journal_clear_features(journal_t *journal, unsigned long compat,
1973 unsigned long ro, unsigned long incompat)
1975 journal_superblock_t *sb;
1977 jbd_debug(1, "Clear features 0x%lx/0x%lx/0x%lx\n",
1978 compat, ro, incompat);
1980 sb = journal->j_superblock;
1982 sb->s_feature_compat &= ~cpu_to_be32(compat);
1983 sb->s_feature_ro_compat &= ~cpu_to_be32(ro);
1984 sb->s_feature_incompat &= ~cpu_to_be32(incompat);
1985 journal->j_revoke_records_per_block =
1986 journal_revoke_records_per_block(journal);
1988 EXPORT_SYMBOL(jbd2_journal_clear_features);
1991 * int jbd2_journal_flush () - Flush journal
1992 * @journal: Journal to act on.
1994 * Flush all data for a given journal to disk and empty the journal.
1995 * Filesystems can use this when remounting readonly to ensure that
1996 * recovery does not need to happen on remount.
1999 int jbd2_journal_flush(journal_t *journal)
2002 transaction_t *transaction = NULL;
2004 write_lock(&journal->j_state_lock);
2006 /* Force everything buffered to the log... */
2007 if (journal->j_running_transaction) {
2008 transaction = journal->j_running_transaction;
2009 __jbd2_log_start_commit(journal, transaction->t_tid);
2010 } else if (journal->j_committing_transaction)
2011 transaction = journal->j_committing_transaction;
2013 /* Wait for the log commit to complete... */
2015 tid_t tid = transaction->t_tid;
2017 write_unlock(&journal->j_state_lock);
2018 jbd2_log_wait_commit(journal, tid);
2020 write_unlock(&journal->j_state_lock);
2023 /* ...and flush everything in the log out to disk. */
2024 spin_lock(&journal->j_list_lock);
2025 while (!err && journal->j_checkpoint_transactions != NULL) {
2026 spin_unlock(&journal->j_list_lock);
2027 mutex_lock_io(&journal->j_checkpoint_mutex);
2028 err = jbd2_log_do_checkpoint(journal);
2029 mutex_unlock(&journal->j_checkpoint_mutex);
2030 spin_lock(&journal->j_list_lock);
2032 spin_unlock(&journal->j_list_lock);
2034 if (is_journal_aborted(journal))
2037 mutex_lock_io(&journal->j_checkpoint_mutex);
2039 err = jbd2_cleanup_journal_tail(journal);
2041 mutex_unlock(&journal->j_checkpoint_mutex);
2047 /* Finally, mark the journal as really needing no recovery.
2048 * This sets s_start==0 in the underlying superblock, which is
2049 * the magic code for a fully-recovered superblock. Any future
2050 * commits of data to the journal will restore the current
2052 jbd2_mark_journal_empty(journal, REQ_SYNC | REQ_FUA);
2053 mutex_unlock(&journal->j_checkpoint_mutex);
2054 write_lock(&journal->j_state_lock);
2055 J_ASSERT(!journal->j_running_transaction);
2056 J_ASSERT(!journal->j_committing_transaction);
2057 J_ASSERT(!journal->j_checkpoint_transactions);
2058 J_ASSERT(journal->j_head == journal->j_tail);
2059 J_ASSERT(journal->j_tail_sequence == journal->j_transaction_sequence);
2060 write_unlock(&journal->j_state_lock);
2066 * int jbd2_journal_wipe() - Wipe journal contents
2067 * @journal: Journal to act on.
2068 * @write: flag (see below)
2070 * Wipe out all of the contents of a journal, safely. This will produce
2071 * a warning if the journal contains any valid recovery information.
2072 * Must be called between journal_init_*() and jbd2_journal_load().
2074 * If 'write' is non-zero, then we wipe out the journal on disk; otherwise
2075 * we merely suppress recovery.
2078 int jbd2_journal_wipe(journal_t *journal, int write)
2082 J_ASSERT (!(journal->j_flags & JBD2_LOADED));
2084 err = load_superblock(journal);
2088 if (!journal->j_tail)
2091 printk(KERN_WARNING "JBD2: %s recovery information on journal\n",
2092 write ? "Clearing" : "Ignoring");
2094 err = jbd2_journal_skip_recovery(journal);
2096 /* Lock to make assertions happy... */
2097 mutex_lock_io(&journal->j_checkpoint_mutex);
2098 jbd2_mark_journal_empty(journal, REQ_SYNC | REQ_FUA);
2099 mutex_unlock(&journal->j_checkpoint_mutex);
2107 * Journal abort has very specific semantics, which we describe
2108 * for journal abort.
2110 * Two internal functions, which provide abort to the jbd layer
2115 * Quick version for internal journal use (doesn't lock the journal).
2116 * Aborts hard --- we mark the abort as occurred, but do _nothing_ else,
2117 * and don't attempt to make any other journal updates.
2119 void __jbd2_journal_abort_hard(journal_t *journal)
2121 transaction_t *transaction;
2123 if (journal->j_flags & JBD2_ABORT)
2126 printk(KERN_ERR "Aborting journal on device %s.\n",
2127 journal->j_devname);
2129 write_lock(&journal->j_state_lock);
2130 journal->j_flags |= JBD2_ABORT;
2131 transaction = journal->j_running_transaction;
2133 __jbd2_log_start_commit(journal, transaction->t_tid);
2134 write_unlock(&journal->j_state_lock);
2137 /* Soft abort: record the abort error status in the journal superblock,
2138 * but don't do any other IO. */
2139 static void __journal_abort_soft (journal_t *journal, int errno)
2143 write_lock(&journal->j_state_lock);
2144 old_errno = journal->j_errno;
2145 if (!journal->j_errno || errno == -ESHUTDOWN)
2146 journal->j_errno = errno;
2148 if (journal->j_flags & JBD2_ABORT) {
2149 write_unlock(&journal->j_state_lock);
2150 if (!old_errno && old_errno != -ESHUTDOWN &&
2151 errno == -ESHUTDOWN)
2152 jbd2_journal_update_sb_errno(journal);
2155 write_unlock(&journal->j_state_lock);
2157 __jbd2_journal_abort_hard(journal);
2159 jbd2_journal_update_sb_errno(journal);
2160 write_lock(&journal->j_state_lock);
2161 journal->j_flags |= JBD2_REC_ERR;
2162 write_unlock(&journal->j_state_lock);
2166 * void jbd2_journal_abort () - Shutdown the journal immediately.
2167 * @journal: the journal to shutdown.
2168 * @errno: an error number to record in the journal indicating
2169 * the reason for the shutdown.
2171 * Perform a complete, immediate shutdown of the ENTIRE
2172 * journal (not of a single transaction). This operation cannot be
2173 * undone without closing and reopening the journal.
2175 * The jbd2_journal_abort function is intended to support higher level error
2176 * recovery mechanisms such as the ext2/ext3 remount-readonly error
2179 * Journal abort has very specific semantics. Any existing dirty,
2180 * unjournaled buffers in the main filesystem will still be written to
2181 * disk by bdflush, but the journaling mechanism will be suspended
2182 * immediately and no further transaction commits will be honoured.
2184 * Any dirty, journaled buffers will be written back to disk without
2185 * hitting the journal. Atomicity cannot be guaranteed on an aborted
2186 * filesystem, but we _do_ attempt to leave as much data as possible
2187 * behind for fsck to use for cleanup.
2189 * Any attempt to get a new transaction handle on a journal which is in
2190 * ABORT state will just result in an -EROFS error return. A
2191 * jbd2_journal_stop on an existing handle will return -EIO if we have
2192 * entered abort state during the update.
2194 * Recursive transactions are not disturbed by journal abort until the
2195 * final jbd2_journal_stop, which will receive the -EIO error.
2197 * Finally, the jbd2_journal_abort call allows the caller to supply an errno
2198 * which will be recorded (if possible) in the journal superblock. This
2199 * allows a client to record failure conditions in the middle of a
2200 * transaction without having to complete the transaction to record the
2201 * failure to disk. ext3_error, for example, now uses this
2206 void jbd2_journal_abort(journal_t *journal, int errno)
2208 __journal_abort_soft(journal, errno);
2212 * int jbd2_journal_errno () - returns the journal's error state.
2213 * @journal: journal to examine.
2215 * This is the errno number set with jbd2_journal_abort(), the last
2216 * time the journal was mounted - if the journal was stopped
2217 * without calling abort this will be 0.
2219 * If the journal has been aborted on this mount time -EROFS will
2222 int jbd2_journal_errno(journal_t *journal)
2226 read_lock(&journal->j_state_lock);
2227 if (journal->j_flags & JBD2_ABORT)
2230 err = journal->j_errno;
2231 read_unlock(&journal->j_state_lock);
2236 * int jbd2_journal_clear_err () - clears the journal's error state
2237 * @journal: journal to act on.
2239 * An error must be cleared or acked to take a FS out of readonly
2242 int jbd2_journal_clear_err(journal_t *journal)
2246 write_lock(&journal->j_state_lock);
2247 if (journal->j_flags & JBD2_ABORT)
2250 journal->j_errno = 0;
2251 write_unlock(&journal->j_state_lock);
2256 * void jbd2_journal_ack_err() - Ack journal err.
2257 * @journal: journal to act on.
2259 * An error must be cleared or acked to take a FS out of readonly
2262 void jbd2_journal_ack_err(journal_t *journal)
2264 write_lock(&journal->j_state_lock);
2265 if (journal->j_errno)
2266 journal->j_flags |= JBD2_ACK_ERR;
2267 write_unlock(&journal->j_state_lock);
2270 int jbd2_journal_blocks_per_page(struct inode *inode)
2272 return 1 << (PAGE_SHIFT - inode->i_sb->s_blocksize_bits);
2276 * helper functions to deal with 32 or 64bit block numbers.
2278 size_t journal_tag_bytes(journal_t *journal)
2282 if (jbd2_has_feature_csum3(journal))
2283 return sizeof(journal_block_tag3_t);
2285 sz = sizeof(journal_block_tag_t);
2287 if (jbd2_has_feature_csum2(journal))
2288 sz += sizeof(__u16);
2290 if (jbd2_has_feature_64bit(journal))
2293 return sz - sizeof(__u32);
2297 * JBD memory management
2299 * These functions are used to allocate block-sized chunks of memory
2300 * used for making copies of buffer_head data. Very often it will be
2301 * page-sized chunks of data, but sometimes it will be in
2302 * sub-page-size chunks. (For example, 16k pages on Power systems
2303 * with a 4k block file system.) For blocks smaller than a page, we
2304 * use a SLAB allocator. There are slab caches for each block size,
2305 * which are allocated at mount time, if necessary, and we only free
2306 * (all of) the slab caches when/if the jbd2 module is unloaded. For
2307 * this reason we don't need to a mutex to protect access to
2308 * jbd2_slab[] allocating or releasing memory; only in
2309 * jbd2_journal_create_slab().
2311 #define JBD2_MAX_SLABS 8
2312 static struct kmem_cache *jbd2_slab[JBD2_MAX_SLABS];
2314 static const char *jbd2_slab_names[JBD2_MAX_SLABS] = {
2315 "jbd2_1k", "jbd2_2k", "jbd2_4k", "jbd2_8k",
2316 "jbd2_16k", "jbd2_32k", "jbd2_64k", "jbd2_128k"
2320 static void jbd2_journal_destroy_slabs(void)
2324 for (i = 0; i < JBD2_MAX_SLABS; i++) {
2325 kmem_cache_destroy(jbd2_slab[i]);
2326 jbd2_slab[i] = NULL;
2330 static int jbd2_journal_create_slab(size_t size)
2332 static DEFINE_MUTEX(jbd2_slab_create_mutex);
2333 int i = order_base_2(size) - 10;
2336 if (size == PAGE_SIZE)
2339 if (i >= JBD2_MAX_SLABS)
2342 if (unlikely(i < 0))
2344 mutex_lock(&jbd2_slab_create_mutex);
2346 mutex_unlock(&jbd2_slab_create_mutex);
2347 return 0; /* Already created */
2350 slab_size = 1 << (i+10);
2351 jbd2_slab[i] = kmem_cache_create(jbd2_slab_names[i], slab_size,
2352 slab_size, 0, NULL);
2353 mutex_unlock(&jbd2_slab_create_mutex);
2354 if (!jbd2_slab[i]) {
2355 printk(KERN_EMERG "JBD2: no memory for jbd2_slab cache\n");
2361 static struct kmem_cache *get_slab(size_t size)
2363 int i = order_base_2(size) - 10;
2365 BUG_ON(i >= JBD2_MAX_SLABS);
2366 if (unlikely(i < 0))
2368 BUG_ON(jbd2_slab[i] == NULL);
2369 return jbd2_slab[i];
2372 void *jbd2_alloc(size_t size, gfp_t flags)
2376 BUG_ON(size & (size-1)); /* Must be a power of 2 */
2378 if (size < PAGE_SIZE)
2379 ptr = kmem_cache_alloc(get_slab(size), flags);
2381 ptr = (void *)__get_free_pages(flags, get_order(size));
2383 /* Check alignment; SLUB has gotten this wrong in the past,
2384 * and this can lead to user data corruption! */
2385 BUG_ON(((unsigned long) ptr) & (size-1));
2390 void jbd2_free(void *ptr, size_t size)
2392 if (size < PAGE_SIZE)
2393 kmem_cache_free(get_slab(size), ptr);
2395 free_pages((unsigned long)ptr, get_order(size));
2399 * Journal_head storage management
2401 static struct kmem_cache *jbd2_journal_head_cache;
2402 #ifdef CONFIG_JBD2_DEBUG
2403 static atomic_t nr_journal_heads = ATOMIC_INIT(0);
2406 static int __init jbd2_journal_init_journal_head_cache(void)
2408 J_ASSERT(!jbd2_journal_head_cache);
2409 jbd2_journal_head_cache = kmem_cache_create("jbd2_journal_head",
2410 sizeof(struct journal_head),
2412 SLAB_TEMPORARY | SLAB_TYPESAFE_BY_RCU,
2414 if (!jbd2_journal_head_cache) {
2415 printk(KERN_EMERG "JBD2: no memory for journal_head cache\n");
2421 static void jbd2_journal_destroy_journal_head_cache(void)
2423 kmem_cache_destroy(jbd2_journal_head_cache);
2424 jbd2_journal_head_cache = NULL;
2428 * journal_head splicing and dicing
2430 static struct journal_head *journal_alloc_journal_head(void)
2432 struct journal_head *ret;
2434 #ifdef CONFIG_JBD2_DEBUG
2435 atomic_inc(&nr_journal_heads);
2437 ret = kmem_cache_zalloc(jbd2_journal_head_cache, GFP_NOFS);
2439 jbd_debug(1, "out of memory for journal_head\n");
2440 pr_notice_ratelimited("ENOMEM in %s, retrying.\n", __func__);
2441 ret = kmem_cache_zalloc(jbd2_journal_head_cache,
2442 GFP_NOFS | __GFP_NOFAIL);
2445 spin_lock_init(&ret->b_state_lock);
2449 static void journal_free_journal_head(struct journal_head *jh)
2451 #ifdef CONFIG_JBD2_DEBUG
2452 atomic_dec(&nr_journal_heads);
2453 memset(jh, JBD2_POISON_FREE, sizeof(*jh));
2455 kmem_cache_free(jbd2_journal_head_cache, jh);
2459 * A journal_head is attached to a buffer_head whenever JBD has an
2460 * interest in the buffer.
2462 * Whenever a buffer has an attached journal_head, its ->b_state:BH_JBD bit
2463 * is set. This bit is tested in core kernel code where we need to take
2464 * JBD-specific actions. Testing the zeroness of ->b_private is not reliable
2467 * When a buffer has its BH_JBD bit set, its ->b_count is elevated by one.
2469 * When a buffer has its BH_JBD bit set it is immune from being released by
2470 * core kernel code, mainly via ->b_count.
2472 * A journal_head is detached from its buffer_head when the journal_head's
2473 * b_jcount reaches zero. Running transaction (b_transaction) and checkpoint
2474 * transaction (b_cp_transaction) hold their references to b_jcount.
2476 * Various places in the kernel want to attach a journal_head to a buffer_head
2477 * _before_ attaching the journal_head to a transaction. To protect the
2478 * journal_head in this situation, jbd2_journal_add_journal_head elevates the
2479 * journal_head's b_jcount refcount by one. The caller must call
2480 * jbd2_journal_put_journal_head() to undo this.
2482 * So the typical usage would be:
2484 * (Attach a journal_head if needed. Increments b_jcount)
2485 * struct journal_head *jh = jbd2_journal_add_journal_head(bh);
2487 * (Get another reference for transaction)
2488 * jbd2_journal_grab_journal_head(bh);
2489 * jh->b_transaction = xxx;
2490 * (Put original reference)
2491 * jbd2_journal_put_journal_head(jh);
2495 * Give a buffer_head a journal_head.
2499 struct journal_head *jbd2_journal_add_journal_head(struct buffer_head *bh)
2501 struct journal_head *jh;
2502 struct journal_head *new_jh = NULL;
2505 if (!buffer_jbd(bh))
2506 new_jh = journal_alloc_journal_head();
2508 jbd_lock_bh_journal_head(bh);
2509 if (buffer_jbd(bh)) {
2513 (atomic_read(&bh->b_count) > 0) ||
2514 (bh->b_page && bh->b_page->mapping));
2517 jbd_unlock_bh_journal_head(bh);
2522 new_jh = NULL; /* We consumed it */
2527 BUFFER_TRACE(bh, "added journal_head");
2530 jbd_unlock_bh_journal_head(bh);
2532 journal_free_journal_head(new_jh);
2533 return bh->b_private;
2537 * Grab a ref against this buffer_head's journal_head. If it ended up not
2538 * having a journal_head, return NULL
2540 struct journal_head *jbd2_journal_grab_journal_head(struct buffer_head *bh)
2542 struct journal_head *jh = NULL;
2544 jbd_lock_bh_journal_head(bh);
2545 if (buffer_jbd(bh)) {
2549 jbd_unlock_bh_journal_head(bh);
2553 static void __journal_remove_journal_head(struct buffer_head *bh)
2555 struct journal_head *jh = bh2jh(bh);
2557 J_ASSERT_JH(jh, jh->b_transaction == NULL);
2558 J_ASSERT_JH(jh, jh->b_next_transaction == NULL);
2559 J_ASSERT_JH(jh, jh->b_cp_transaction == NULL);
2560 J_ASSERT_JH(jh, jh->b_jlist == BJ_None);
2561 J_ASSERT_BH(bh, buffer_jbd(bh));
2562 J_ASSERT_BH(bh, jh2bh(jh) == bh);
2563 BUFFER_TRACE(bh, "remove journal_head");
2565 /* Unlink before dropping the lock */
2566 bh->b_private = NULL;
2567 jh->b_bh = NULL; /* debug, really */
2568 clear_buffer_jbd(bh);
2571 static void journal_release_journal_head(struct journal_head *jh, size_t b_size)
2573 if (jh->b_frozen_data) {
2574 printk(KERN_WARNING "%s: freeing b_frozen_data\n", __func__);
2575 jbd2_free(jh->b_frozen_data, b_size);
2577 if (jh->b_committed_data) {
2578 printk(KERN_WARNING "%s: freeing b_committed_data\n", __func__);
2579 jbd2_free(jh->b_committed_data, b_size);
2581 journal_free_journal_head(jh);
2585 * Drop a reference on the passed journal_head. If it fell to zero then
2586 * release the journal_head from the buffer_head.
2588 void jbd2_journal_put_journal_head(struct journal_head *jh)
2590 struct buffer_head *bh = jh2bh(jh);
2592 jbd_lock_bh_journal_head(bh);
2593 J_ASSERT_JH(jh, jh->b_jcount > 0);
2595 if (!jh->b_jcount) {
2596 __journal_remove_journal_head(bh);
2597 jbd_unlock_bh_journal_head(bh);
2598 journal_release_journal_head(jh, bh->b_size);
2601 jbd_unlock_bh_journal_head(bh);
2606 * Initialize jbd inode head
2608 void jbd2_journal_init_jbd_inode(struct jbd2_inode *jinode, struct inode *inode)
2610 jinode->i_transaction = NULL;
2611 jinode->i_next_transaction = NULL;
2612 jinode->i_vfs_inode = inode;
2613 jinode->i_flags = 0;
2614 jinode->i_dirty_start = 0;
2615 jinode->i_dirty_end = 0;
2616 INIT_LIST_HEAD(&jinode->i_list);
2620 * Function to be called before we start removing inode from memory (i.e.,
2621 * clear_inode() is a fine place to be called from). It removes inode from
2622 * transaction's lists.
2624 void jbd2_journal_release_jbd_inode(journal_t *journal,
2625 struct jbd2_inode *jinode)
2630 spin_lock(&journal->j_list_lock);
2631 /* Is commit writing out inode - we have to wait */
2632 if (jinode->i_flags & JI_COMMIT_RUNNING) {
2633 wait_queue_head_t *wq;
2634 DEFINE_WAIT_BIT(wait, &jinode->i_flags, __JI_COMMIT_RUNNING);
2635 wq = bit_waitqueue(&jinode->i_flags, __JI_COMMIT_RUNNING);
2636 prepare_to_wait(wq, &wait.wq_entry, TASK_UNINTERRUPTIBLE);
2637 spin_unlock(&journal->j_list_lock);
2639 finish_wait(wq, &wait.wq_entry);
2643 if (jinode->i_transaction) {
2644 list_del(&jinode->i_list);
2645 jinode->i_transaction = NULL;
2647 spin_unlock(&journal->j_list_lock);
2651 #ifdef CONFIG_PROC_FS
2653 #define JBD2_STATS_PROC_NAME "fs/jbd2"
2655 static void __init jbd2_create_jbd_stats_proc_entry(void)
2657 proc_jbd2_stats = proc_mkdir(JBD2_STATS_PROC_NAME, NULL);
2660 static void __exit jbd2_remove_jbd_stats_proc_entry(void)
2662 if (proc_jbd2_stats)
2663 remove_proc_entry(JBD2_STATS_PROC_NAME, NULL);
2668 #define jbd2_create_jbd_stats_proc_entry() do {} while (0)
2669 #define jbd2_remove_jbd_stats_proc_entry() do {} while (0)
2673 struct kmem_cache *jbd2_handle_cache, *jbd2_inode_cache;
2675 static int __init jbd2_journal_init_inode_cache(void)
2677 J_ASSERT(!jbd2_inode_cache);
2678 jbd2_inode_cache = KMEM_CACHE(jbd2_inode, 0);
2679 if (!jbd2_inode_cache) {
2680 pr_emerg("JBD2: failed to create inode cache\n");
2686 static int __init jbd2_journal_init_handle_cache(void)
2688 J_ASSERT(!jbd2_handle_cache);
2689 jbd2_handle_cache = KMEM_CACHE(jbd2_journal_handle, SLAB_TEMPORARY);
2690 if (!jbd2_handle_cache) {
2691 printk(KERN_EMERG "JBD2: failed to create handle cache\n");
2697 static void jbd2_journal_destroy_inode_cache(void)
2699 kmem_cache_destroy(jbd2_inode_cache);
2700 jbd2_inode_cache = NULL;
2703 static void jbd2_journal_destroy_handle_cache(void)
2705 kmem_cache_destroy(jbd2_handle_cache);
2706 jbd2_handle_cache = NULL;
2710 * Module startup and shutdown
2713 static int __init journal_init_caches(void)
2717 ret = jbd2_journal_init_revoke_record_cache();
2719 ret = jbd2_journal_init_revoke_table_cache();
2721 ret = jbd2_journal_init_journal_head_cache();
2723 ret = jbd2_journal_init_handle_cache();
2725 ret = jbd2_journal_init_inode_cache();
2727 ret = jbd2_journal_init_transaction_cache();
2731 static void jbd2_journal_destroy_caches(void)
2733 jbd2_journal_destroy_revoke_record_cache();
2734 jbd2_journal_destroy_revoke_table_cache();
2735 jbd2_journal_destroy_journal_head_cache();
2736 jbd2_journal_destroy_handle_cache();
2737 jbd2_journal_destroy_inode_cache();
2738 jbd2_journal_destroy_transaction_cache();
2739 jbd2_journal_destroy_slabs();
2742 static int __init journal_init(void)
2746 BUILD_BUG_ON(sizeof(struct journal_superblock_s) != 1024);
2748 ret = journal_init_caches();
2750 jbd2_create_jbd_stats_proc_entry();
2752 jbd2_journal_destroy_caches();
2757 static void __exit journal_exit(void)
2759 #ifdef CONFIG_JBD2_DEBUG
2760 int n = atomic_read(&nr_journal_heads);
2762 printk(KERN_ERR "JBD2: leaked %d journal_heads!\n", n);
2764 jbd2_remove_jbd_stats_proc_entry();
2765 jbd2_journal_destroy_caches();
2768 MODULE_LICENSE("GPL");
2769 module_init(journal_init);
2770 module_exit(journal_exit);