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)
988 static int jbd2_seq_info_show(struct seq_file *seq, void *v)
990 struct jbd2_stats_proc_session *s = seq->private;
992 if (v != SEQ_START_TOKEN)
994 seq_printf(seq, "%lu transactions (%lu requested), "
995 "each up to %u blocks\n",
996 s->stats->ts_tid, s->stats->ts_requested,
997 s->journal->j_max_transaction_buffers);
998 if (s->stats->ts_tid == 0)
1000 seq_printf(seq, "average: \n %ums waiting for transaction\n",
1001 jiffies_to_msecs(s->stats->run.rs_wait / s->stats->ts_tid));
1002 seq_printf(seq, " %ums request delay\n",
1003 (s->stats->ts_requested == 0) ? 0 :
1004 jiffies_to_msecs(s->stats->run.rs_request_delay /
1005 s->stats->ts_requested));
1006 seq_printf(seq, " %ums running transaction\n",
1007 jiffies_to_msecs(s->stats->run.rs_running / s->stats->ts_tid));
1008 seq_printf(seq, " %ums transaction was being locked\n",
1009 jiffies_to_msecs(s->stats->run.rs_locked / s->stats->ts_tid));
1010 seq_printf(seq, " %ums flushing data (in ordered mode)\n",
1011 jiffies_to_msecs(s->stats->run.rs_flushing / s->stats->ts_tid));
1012 seq_printf(seq, " %ums logging transaction\n",
1013 jiffies_to_msecs(s->stats->run.rs_logging / s->stats->ts_tid));
1014 seq_printf(seq, " %lluus average transaction commit time\n",
1015 div_u64(s->journal->j_average_commit_time, 1000));
1016 seq_printf(seq, " %lu handles per transaction\n",
1017 s->stats->run.rs_handle_count / s->stats->ts_tid);
1018 seq_printf(seq, " %lu blocks per transaction\n",
1019 s->stats->run.rs_blocks / s->stats->ts_tid);
1020 seq_printf(seq, " %lu logged blocks per transaction\n",
1021 s->stats->run.rs_blocks_logged / s->stats->ts_tid);
1025 static void jbd2_seq_info_stop(struct seq_file *seq, void *v)
1029 static const struct seq_operations jbd2_seq_info_ops = {
1030 .start = jbd2_seq_info_start,
1031 .next = jbd2_seq_info_next,
1032 .stop = jbd2_seq_info_stop,
1033 .show = jbd2_seq_info_show,
1036 static int jbd2_seq_info_open(struct inode *inode, struct file *file)
1038 journal_t *journal = PDE_DATA(inode);
1039 struct jbd2_stats_proc_session *s;
1042 s = kmalloc(sizeof(*s), GFP_KERNEL);
1045 size = sizeof(struct transaction_stats_s);
1046 s->stats = kmalloc(size, GFP_KERNEL);
1047 if (s->stats == NULL) {
1051 spin_lock(&journal->j_history_lock);
1052 memcpy(s->stats, &journal->j_stats, size);
1053 s->journal = journal;
1054 spin_unlock(&journal->j_history_lock);
1056 rc = seq_open(file, &jbd2_seq_info_ops);
1058 struct seq_file *m = file->private_data;
1068 static int jbd2_seq_info_release(struct inode *inode, struct file *file)
1070 struct seq_file *seq = file->private_data;
1071 struct jbd2_stats_proc_session *s = seq->private;
1074 return seq_release(inode, file);
1077 static const struct file_operations jbd2_seq_info_fops = {
1078 .owner = THIS_MODULE,
1079 .open = jbd2_seq_info_open,
1081 .llseek = seq_lseek,
1082 .release = jbd2_seq_info_release,
1085 static struct proc_dir_entry *proc_jbd2_stats;
1087 static void jbd2_stats_proc_init(journal_t *journal)
1089 journal->j_proc_entry = proc_mkdir(journal->j_devname, proc_jbd2_stats);
1090 if (journal->j_proc_entry) {
1091 proc_create_data("info", S_IRUGO, journal->j_proc_entry,
1092 &jbd2_seq_info_fops, journal);
1096 static void jbd2_stats_proc_exit(journal_t *journal)
1098 remove_proc_entry("info", journal->j_proc_entry);
1099 remove_proc_entry(journal->j_devname, proc_jbd2_stats);
1102 /* Minimum size of descriptor tag */
1103 static int jbd2_min_tag_size(void)
1106 * Tag with 32-bit block numbers does not use last four bytes of the
1109 return sizeof(journal_block_tag_t) - 4;
1113 * Management for journal control blocks: functions to create and
1114 * destroy journal_t structures, and to initialise and read existing
1115 * journal blocks from disk. */
1117 /* First: create and setup a journal_t object in memory. We initialise
1118 * very few fields yet: that has to wait until we have created the
1119 * journal structures from from scratch, or loaded them from disk. */
1121 static journal_t *journal_init_common(struct block_device *bdev,
1122 struct block_device *fs_dev,
1123 unsigned long long start, int len, int blocksize)
1125 static struct lock_class_key jbd2_trans_commit_key;
1128 struct buffer_head *bh;
1131 journal = kzalloc(sizeof(*journal), GFP_KERNEL);
1135 init_waitqueue_head(&journal->j_wait_transaction_locked);
1136 init_waitqueue_head(&journal->j_wait_done_commit);
1137 init_waitqueue_head(&journal->j_wait_commit);
1138 init_waitqueue_head(&journal->j_wait_updates);
1139 init_waitqueue_head(&journal->j_wait_reserved);
1140 mutex_init(&journal->j_barrier);
1141 mutex_init(&journal->j_checkpoint_mutex);
1142 spin_lock_init(&journal->j_revoke_lock);
1143 spin_lock_init(&journal->j_list_lock);
1144 rwlock_init(&journal->j_state_lock);
1146 journal->j_commit_interval = (HZ * JBD2_DEFAULT_MAX_COMMIT_AGE);
1147 journal->j_min_batch_time = 0;
1148 journal->j_max_batch_time = 15000; /* 15ms */
1149 atomic_set(&journal->j_reserved_credits, 0);
1151 /* The journal is marked for error until we succeed with recovery! */
1152 journal->j_flags = JBD2_ABORT;
1154 /* Set up a default-sized revoke table for the new mount. */
1155 err = jbd2_journal_init_revoke(journal, JOURNAL_REVOKE_DEFAULT_HASH);
1159 spin_lock_init(&journal->j_history_lock);
1161 lockdep_init_map(&journal->j_trans_commit_map, "jbd2_handle",
1162 &jbd2_trans_commit_key, 0);
1164 /* journal descriptor can store up to n blocks -bzzz */
1165 journal->j_blocksize = blocksize;
1166 journal->j_dev = bdev;
1167 journal->j_fs_dev = fs_dev;
1168 journal->j_blk_offset = start;
1169 journal->j_maxlen = len;
1170 /* We need enough buffers to write out full descriptor block. */
1171 n = journal->j_blocksize / jbd2_min_tag_size();
1172 journal->j_wbufsize = n;
1173 journal->j_wbuf = kmalloc_array(n, sizeof(struct buffer_head *),
1175 if (!journal->j_wbuf)
1178 bh = getblk_unmovable(journal->j_dev, start, journal->j_blocksize);
1180 pr_err("%s: Cannot get buffer for journal superblock\n",
1184 journal->j_sb_buffer = bh;
1185 journal->j_superblock = (journal_superblock_t *)bh->b_data;
1190 kfree(journal->j_wbuf);
1191 jbd2_journal_destroy_revoke(journal);
1196 /* jbd2_journal_init_dev and jbd2_journal_init_inode:
1198 * Create a journal structure assigned some fixed set of disk blocks to
1199 * the journal. We don't actually touch those disk blocks yet, but we
1200 * need to set up all of the mapping information to tell the journaling
1201 * system where the journal blocks are.
1206 * journal_t * jbd2_journal_init_dev() - creates and initialises a journal structure
1207 * @bdev: Block device on which to create the journal
1208 * @fs_dev: Device which hold journalled filesystem for this journal.
1209 * @start: Block nr Start of journal.
1210 * @len: Length of the journal in blocks.
1211 * @blocksize: blocksize of journalling device
1213 * Returns: a newly created journal_t *
1215 * jbd2_journal_init_dev creates a journal which maps a fixed contiguous
1216 * range of blocks on an arbitrary block device.
1219 journal_t *jbd2_journal_init_dev(struct block_device *bdev,
1220 struct block_device *fs_dev,
1221 unsigned long long start, int len, int blocksize)
1225 journal = journal_init_common(bdev, fs_dev, start, len, blocksize);
1229 bdevname(journal->j_dev, journal->j_devname);
1230 strreplace(journal->j_devname, '/', '!');
1231 jbd2_stats_proc_init(journal);
1237 * journal_t * jbd2_journal_init_inode () - creates a journal which maps to a inode.
1238 * @inode: An inode to create the journal in
1240 * jbd2_journal_init_inode creates a journal which maps an on-disk inode as
1241 * the journal. The inode must exist already, must support bmap() and
1242 * must have all data blocks preallocated.
1244 journal_t *jbd2_journal_init_inode(struct inode *inode)
1248 unsigned long long blocknr;
1250 blocknr = bmap(inode, 0);
1252 pr_err("%s: Cannot locate journal superblock\n",
1257 jbd_debug(1, "JBD2: inode %s/%ld, size %lld, bits %d, blksize %ld\n",
1258 inode->i_sb->s_id, inode->i_ino, (long long) inode->i_size,
1259 inode->i_sb->s_blocksize_bits, inode->i_sb->s_blocksize);
1261 journal = journal_init_common(inode->i_sb->s_bdev, inode->i_sb->s_bdev,
1262 blocknr, inode->i_size >> inode->i_sb->s_blocksize_bits,
1263 inode->i_sb->s_blocksize);
1267 journal->j_inode = inode;
1268 bdevname(journal->j_dev, journal->j_devname);
1269 p = strreplace(journal->j_devname, '/', '!');
1270 sprintf(p, "-%lu", journal->j_inode->i_ino);
1271 jbd2_stats_proc_init(journal);
1277 * If the journal init or create aborts, we need to mark the journal
1278 * superblock as being NULL to prevent the journal destroy from writing
1279 * back a bogus superblock.
1281 static void journal_fail_superblock (journal_t *journal)
1283 struct buffer_head *bh = journal->j_sb_buffer;
1285 journal->j_sb_buffer = NULL;
1289 * Given a journal_t structure, initialise the various fields for
1290 * startup of a new journaling session. We use this both when creating
1291 * a journal, and after recovering an old journal to reset it for
1295 static int journal_reset(journal_t *journal)
1297 journal_superblock_t *sb = journal->j_superblock;
1298 unsigned long long first, last;
1300 first = be32_to_cpu(sb->s_first);
1301 last = be32_to_cpu(sb->s_maxlen);
1302 if (first + JBD2_MIN_JOURNAL_BLOCKS > last + 1) {
1303 printk(KERN_ERR "JBD2: Journal too short (blocks %llu-%llu).\n",
1305 journal_fail_superblock(journal);
1309 journal->j_first = first;
1310 journal->j_last = last;
1312 journal->j_head = first;
1313 journal->j_tail = first;
1314 journal->j_free = last - first;
1316 journal->j_tail_sequence = journal->j_transaction_sequence;
1317 journal->j_commit_sequence = journal->j_transaction_sequence - 1;
1318 journal->j_commit_request = journal->j_commit_sequence;
1320 journal->j_max_transaction_buffers = journal->j_maxlen / 4;
1323 * As a special case, if the on-disk copy is already marked as needing
1324 * no recovery (s_start == 0), then we can safely defer the superblock
1325 * update until the next commit by setting JBD2_FLUSHED. This avoids
1326 * attempting a write to a potential-readonly device.
1328 if (sb->s_start == 0) {
1329 jbd_debug(1, "JBD2: Skipping superblock update on recovered sb "
1330 "(start %ld, seq %u, errno %d)\n",
1331 journal->j_tail, journal->j_tail_sequence,
1333 journal->j_flags |= JBD2_FLUSHED;
1335 /* Lock here to make assertions happy... */
1336 mutex_lock_io(&journal->j_checkpoint_mutex);
1338 * Update log tail information. We use REQ_FUA since new
1339 * transaction will start reusing journal space and so we
1340 * must make sure information about current log tail is on
1343 jbd2_journal_update_sb_log_tail(journal,
1344 journal->j_tail_sequence,
1346 REQ_SYNC | REQ_FUA);
1347 mutex_unlock(&journal->j_checkpoint_mutex);
1349 return jbd2_journal_start_thread(journal);
1353 * This function expects that the caller will have locked the journal
1354 * buffer head, and will return with it unlocked
1356 static int jbd2_write_superblock(journal_t *journal, int write_flags)
1358 struct buffer_head *bh = journal->j_sb_buffer;
1359 journal_superblock_t *sb = journal->j_superblock;
1362 /* Buffer got discarded which means block device got invalidated */
1363 if (!buffer_mapped(bh))
1366 trace_jbd2_write_superblock(journal, write_flags);
1367 if (!(journal->j_flags & JBD2_BARRIER))
1368 write_flags &= ~(REQ_FUA | REQ_PREFLUSH);
1369 if (buffer_write_io_error(bh)) {
1371 * Oh, dear. A previous attempt to write the journal
1372 * superblock failed. This could happen because the
1373 * USB device was yanked out. Or it could happen to
1374 * be a transient write error and maybe the block will
1375 * be remapped. Nothing we can do but to retry the
1376 * write and hope for the best.
1378 printk(KERN_ERR "JBD2: previous I/O error detected "
1379 "for journal superblock update for %s.\n",
1380 journal->j_devname);
1381 clear_buffer_write_io_error(bh);
1382 set_buffer_uptodate(bh);
1384 if (jbd2_journal_has_csum_v2or3(journal))
1385 sb->s_checksum = jbd2_superblock_csum(journal, sb);
1387 bh->b_end_io = end_buffer_write_sync;
1388 ret = submit_bh(REQ_OP_WRITE, write_flags, bh);
1390 if (buffer_write_io_error(bh)) {
1391 clear_buffer_write_io_error(bh);
1392 set_buffer_uptodate(bh);
1396 printk(KERN_ERR "JBD2: Error %d detected when updating "
1397 "journal superblock for %s.\n", ret,
1398 journal->j_devname);
1399 jbd2_journal_abort(journal, ret);
1406 * jbd2_journal_update_sb_log_tail() - Update log tail in journal sb on disk.
1407 * @journal: The journal to update.
1408 * @tail_tid: TID of the new transaction at the tail of the log
1409 * @tail_block: The first block of the transaction at the tail of the log
1410 * @write_op: With which operation should we write the journal sb
1412 * Update a journal's superblock information about log tail and write it to
1413 * disk, waiting for the IO to complete.
1415 int jbd2_journal_update_sb_log_tail(journal_t *journal, tid_t tail_tid,
1416 unsigned long tail_block, int write_op)
1418 journal_superblock_t *sb = journal->j_superblock;
1421 if (is_journal_aborted(journal))
1424 BUG_ON(!mutex_is_locked(&journal->j_checkpoint_mutex));
1425 jbd_debug(1, "JBD2: updating superblock (start %lu, seq %u)\n",
1426 tail_block, tail_tid);
1428 lock_buffer(journal->j_sb_buffer);
1429 sb->s_sequence = cpu_to_be32(tail_tid);
1430 sb->s_start = cpu_to_be32(tail_block);
1432 ret = jbd2_write_superblock(journal, write_op);
1436 /* Log is no longer empty */
1437 write_lock(&journal->j_state_lock);
1438 WARN_ON(!sb->s_sequence);
1439 journal->j_flags &= ~JBD2_FLUSHED;
1440 write_unlock(&journal->j_state_lock);
1447 * jbd2_mark_journal_empty() - Mark on disk journal as empty.
1448 * @journal: The journal to update.
1449 * @write_op: With which operation should we write the journal sb
1451 * Update a journal's dynamic superblock fields to show that journal is empty.
1452 * Write updated superblock to disk waiting for IO to complete.
1454 static void jbd2_mark_journal_empty(journal_t *journal, int write_op)
1456 journal_superblock_t *sb = journal->j_superblock;
1458 BUG_ON(!mutex_is_locked(&journal->j_checkpoint_mutex));
1459 lock_buffer(journal->j_sb_buffer);
1460 if (sb->s_start == 0) { /* Is it already empty? */
1461 unlock_buffer(journal->j_sb_buffer);
1465 jbd_debug(1, "JBD2: Marking journal as empty (seq %u)\n",
1466 journal->j_tail_sequence);
1468 sb->s_sequence = cpu_to_be32(journal->j_tail_sequence);
1469 sb->s_start = cpu_to_be32(0);
1471 jbd2_write_superblock(journal, write_op);
1473 /* Log is no longer empty */
1474 write_lock(&journal->j_state_lock);
1475 journal->j_flags |= JBD2_FLUSHED;
1476 write_unlock(&journal->j_state_lock);
1481 * jbd2_journal_update_sb_errno() - Update error in the journal.
1482 * @journal: The journal to update.
1484 * Update a journal's errno. Write updated superblock to disk waiting for IO
1487 void jbd2_journal_update_sb_errno(journal_t *journal)
1489 journal_superblock_t *sb = journal->j_superblock;
1492 lock_buffer(journal->j_sb_buffer);
1493 errcode = journal->j_errno;
1494 if (errcode == -ESHUTDOWN)
1496 jbd_debug(1, "JBD2: updating superblock error (errno %d)\n", errcode);
1497 sb->s_errno = cpu_to_be32(errcode);
1499 jbd2_write_superblock(journal, REQ_SYNC | REQ_FUA);
1501 EXPORT_SYMBOL(jbd2_journal_update_sb_errno);
1503 static int journal_revoke_records_per_block(journal_t *journal)
1506 int space = journal->j_blocksize - sizeof(jbd2_journal_revoke_header_t);
1508 if (jbd2_has_feature_64bit(journal))
1513 if (jbd2_journal_has_csum_v2or3(journal))
1514 space -= sizeof(struct jbd2_journal_block_tail);
1515 return space / record_size;
1519 * Read the superblock for a given journal, performing initial
1520 * validation of the format.
1522 static int journal_get_superblock(journal_t *journal)
1524 struct buffer_head *bh;
1525 journal_superblock_t *sb;
1528 bh = journal->j_sb_buffer;
1530 J_ASSERT(bh != NULL);
1531 if (!buffer_uptodate(bh)) {
1532 ll_rw_block(REQ_OP_READ, 0, 1, &bh);
1534 if (!buffer_uptodate(bh)) {
1536 "JBD2: IO error reading journal superblock\n");
1541 if (buffer_verified(bh))
1544 sb = journal->j_superblock;
1548 if (sb->s_header.h_magic != cpu_to_be32(JBD2_MAGIC_NUMBER) ||
1549 sb->s_blocksize != cpu_to_be32(journal->j_blocksize)) {
1550 printk(KERN_WARNING "JBD2: no valid journal superblock found\n");
1554 switch(be32_to_cpu(sb->s_header.h_blocktype)) {
1555 case JBD2_SUPERBLOCK_V1:
1556 journal->j_format_version = 1;
1558 case JBD2_SUPERBLOCK_V2:
1559 journal->j_format_version = 2;
1562 printk(KERN_WARNING "JBD2: unrecognised superblock format ID\n");
1566 if (be32_to_cpu(sb->s_maxlen) < journal->j_maxlen)
1567 journal->j_maxlen = be32_to_cpu(sb->s_maxlen);
1568 else if (be32_to_cpu(sb->s_maxlen) > journal->j_maxlen) {
1569 printk(KERN_WARNING "JBD2: journal file too short\n");
1573 if (be32_to_cpu(sb->s_first) == 0 ||
1574 be32_to_cpu(sb->s_first) >= journal->j_maxlen) {
1576 "JBD2: Invalid start block of journal: %u\n",
1577 be32_to_cpu(sb->s_first));
1581 if (jbd2_has_feature_csum2(journal) &&
1582 jbd2_has_feature_csum3(journal)) {
1583 /* Can't have checksum v2 and v3 at the same time! */
1584 printk(KERN_ERR "JBD2: Can't enable checksumming v2 and v3 "
1585 "at the same time!\n");
1589 if (jbd2_journal_has_csum_v2or3_feature(journal) &&
1590 jbd2_has_feature_checksum(journal)) {
1591 /* Can't have checksum v1 and v2 on at the same time! */
1592 printk(KERN_ERR "JBD2: Can't enable checksumming v1 and v2/3 "
1593 "at the same time!\n");
1597 if (!jbd2_verify_csum_type(journal, sb)) {
1598 printk(KERN_ERR "JBD2: Unknown checksum type\n");
1602 /* Load the checksum driver */
1603 if (jbd2_journal_has_csum_v2or3_feature(journal)) {
1604 journal->j_chksum_driver = crypto_alloc_shash("crc32c", 0, 0);
1605 if (IS_ERR(journal->j_chksum_driver)) {
1606 printk(KERN_ERR "JBD2: Cannot load crc32c driver.\n");
1607 err = PTR_ERR(journal->j_chksum_driver);
1608 journal->j_chksum_driver = NULL;
1613 if (jbd2_journal_has_csum_v2or3(journal)) {
1614 /* Check superblock checksum */
1615 if (sb->s_checksum != jbd2_superblock_csum(journal, sb)) {
1616 printk(KERN_ERR "JBD2: journal checksum error\n");
1621 /* Precompute checksum seed for all metadata */
1622 journal->j_csum_seed = jbd2_chksum(journal, ~0, sb->s_uuid,
1623 sizeof(sb->s_uuid));
1626 journal->j_revoke_records_per_block =
1627 journal_revoke_records_per_block(journal);
1628 set_buffer_verified(bh);
1633 journal_fail_superblock(journal);
1638 * Load the on-disk journal superblock and read the key fields into the
1642 static int load_superblock(journal_t *journal)
1645 journal_superblock_t *sb;
1647 err = journal_get_superblock(journal);
1651 sb = journal->j_superblock;
1653 journal->j_tail_sequence = be32_to_cpu(sb->s_sequence);
1654 journal->j_tail = be32_to_cpu(sb->s_start);
1655 journal->j_first = be32_to_cpu(sb->s_first);
1656 journal->j_last = be32_to_cpu(sb->s_maxlen);
1657 journal->j_errno = be32_to_cpu(sb->s_errno);
1664 * int jbd2_journal_load() - Read journal from disk.
1665 * @journal: Journal to act on.
1667 * Given a journal_t structure which tells us which disk blocks contain
1668 * a journal, read the journal from disk to initialise the in-memory
1671 int jbd2_journal_load(journal_t *journal)
1674 journal_superblock_t *sb;
1676 err = load_superblock(journal);
1680 sb = journal->j_superblock;
1681 /* If this is a V2 superblock, then we have to check the
1682 * features flags on it. */
1684 if (journal->j_format_version >= 2) {
1685 if ((sb->s_feature_ro_compat &
1686 ~cpu_to_be32(JBD2_KNOWN_ROCOMPAT_FEATURES)) ||
1687 (sb->s_feature_incompat &
1688 ~cpu_to_be32(JBD2_KNOWN_INCOMPAT_FEATURES))) {
1690 "JBD2: Unrecognised features on journal\n");
1696 * Create a slab for this blocksize
1698 err = jbd2_journal_create_slab(be32_to_cpu(sb->s_blocksize));
1702 /* Let the recovery code check whether it needs to recover any
1703 * data from the journal. */
1704 if (jbd2_journal_recover(journal))
1705 goto recovery_error;
1707 if (journal->j_failed_commit) {
1708 printk(KERN_ERR "JBD2: journal transaction %u on %s "
1709 "is corrupt.\n", journal->j_failed_commit,
1710 journal->j_devname);
1711 return -EFSCORRUPTED;
1714 /* OK, we've finished with the dynamic journal bits:
1715 * reinitialise the dynamic contents of the superblock in memory
1716 * and reset them on disk. */
1717 if (journal_reset(journal))
1718 goto recovery_error;
1720 journal->j_flags &= ~JBD2_ABORT;
1721 journal->j_flags |= JBD2_LOADED;
1725 printk(KERN_WARNING "JBD2: recovery failed\n");
1730 * void jbd2_journal_destroy() - Release a journal_t structure.
1731 * @journal: Journal to act on.
1733 * Release a journal_t structure once it is no longer in use by the
1735 * Return <0 if we couldn't clean up the journal.
1737 int jbd2_journal_destroy(journal_t *journal)
1741 /* Wait for the commit thread to wake up and die. */
1742 journal_kill_thread(journal);
1744 /* Force a final log commit */
1745 if (journal->j_running_transaction)
1746 jbd2_journal_commit_transaction(journal);
1748 /* Force any old transactions to disk */
1750 /* Totally anal locking here... */
1751 spin_lock(&journal->j_list_lock);
1752 while (journal->j_checkpoint_transactions != NULL) {
1753 spin_unlock(&journal->j_list_lock);
1754 mutex_lock_io(&journal->j_checkpoint_mutex);
1755 err = jbd2_log_do_checkpoint(journal);
1756 mutex_unlock(&journal->j_checkpoint_mutex);
1758 * If checkpointing failed, just free the buffers to avoid
1762 jbd2_journal_destroy_checkpoint(journal);
1763 spin_lock(&journal->j_list_lock);
1766 spin_lock(&journal->j_list_lock);
1769 J_ASSERT(journal->j_running_transaction == NULL);
1770 J_ASSERT(journal->j_committing_transaction == NULL);
1771 J_ASSERT(journal->j_checkpoint_transactions == NULL);
1772 spin_unlock(&journal->j_list_lock);
1774 if (journal->j_sb_buffer) {
1775 if (!is_journal_aborted(journal)) {
1776 mutex_lock_io(&journal->j_checkpoint_mutex);
1778 write_lock(&journal->j_state_lock);
1779 journal->j_tail_sequence =
1780 ++journal->j_transaction_sequence;
1781 write_unlock(&journal->j_state_lock);
1783 jbd2_mark_journal_empty(journal,
1784 REQ_SYNC | REQ_PREFLUSH | REQ_FUA);
1785 mutex_unlock(&journal->j_checkpoint_mutex);
1788 brelse(journal->j_sb_buffer);
1791 if (journal->j_proc_entry)
1792 jbd2_stats_proc_exit(journal);
1793 iput(journal->j_inode);
1794 if (journal->j_revoke)
1795 jbd2_journal_destroy_revoke(journal);
1796 if (journal->j_chksum_driver)
1797 crypto_free_shash(journal->j_chksum_driver);
1798 kfree(journal->j_wbuf);
1806 *int jbd2_journal_check_used_features () - Check if features specified are used.
1807 * @journal: Journal to check.
1808 * @compat: bitmask of compatible features
1809 * @ro: bitmask of features that force read-only mount
1810 * @incompat: bitmask of incompatible features
1812 * Check whether the journal uses all of a given set of
1813 * features. Return true (non-zero) if it does.
1816 int jbd2_journal_check_used_features (journal_t *journal, unsigned long compat,
1817 unsigned long ro, unsigned long incompat)
1819 journal_superblock_t *sb;
1821 if (!compat && !ro && !incompat)
1823 /* Load journal superblock if it is not loaded yet. */
1824 if (journal->j_format_version == 0 &&
1825 journal_get_superblock(journal) != 0)
1827 if (journal->j_format_version == 1)
1830 sb = journal->j_superblock;
1832 if (((be32_to_cpu(sb->s_feature_compat) & compat) == compat) &&
1833 ((be32_to_cpu(sb->s_feature_ro_compat) & ro) == ro) &&
1834 ((be32_to_cpu(sb->s_feature_incompat) & incompat) == incompat))
1841 * int jbd2_journal_check_available_features() - Check feature set in journalling layer
1842 * @journal: Journal to check.
1843 * @compat: bitmask of compatible features
1844 * @ro: bitmask of features that force read-only mount
1845 * @incompat: bitmask of incompatible features
1847 * Check whether the journaling code supports the use of
1848 * all of a given set of features on this journal. Return true
1849 * (non-zero) if it can. */
1851 int jbd2_journal_check_available_features (journal_t *journal, unsigned long compat,
1852 unsigned long ro, unsigned long incompat)
1854 if (!compat && !ro && !incompat)
1857 /* We can support any known requested features iff the
1858 * superblock is in version 2. Otherwise we fail to support any
1859 * extended sb features. */
1861 if (journal->j_format_version != 2)
1864 if ((compat & JBD2_KNOWN_COMPAT_FEATURES) == compat &&
1865 (ro & JBD2_KNOWN_ROCOMPAT_FEATURES) == ro &&
1866 (incompat & JBD2_KNOWN_INCOMPAT_FEATURES) == incompat)
1873 * int jbd2_journal_set_features () - Mark a given journal feature in the superblock
1874 * @journal: Journal to act on.
1875 * @compat: bitmask of compatible features
1876 * @ro: bitmask of features that force read-only mount
1877 * @incompat: bitmask of incompatible features
1879 * Mark a given journal feature as present on the
1880 * superblock. Returns true if the requested features could be set.
1884 int jbd2_journal_set_features (journal_t *journal, unsigned long compat,
1885 unsigned long ro, unsigned long incompat)
1887 #define INCOMPAT_FEATURE_ON(f) \
1888 ((incompat & (f)) && !(sb->s_feature_incompat & cpu_to_be32(f)))
1889 #define COMPAT_FEATURE_ON(f) \
1890 ((compat & (f)) && !(sb->s_feature_compat & cpu_to_be32(f)))
1891 journal_superblock_t *sb;
1893 if (jbd2_journal_check_used_features(journal, compat, ro, incompat))
1896 if (!jbd2_journal_check_available_features(journal, compat, ro, incompat))
1899 /* If enabling v2 checksums, turn on v3 instead */
1900 if (incompat & JBD2_FEATURE_INCOMPAT_CSUM_V2) {
1901 incompat &= ~JBD2_FEATURE_INCOMPAT_CSUM_V2;
1902 incompat |= JBD2_FEATURE_INCOMPAT_CSUM_V3;
1905 /* Asking for checksumming v3 and v1? Only give them v3. */
1906 if (incompat & JBD2_FEATURE_INCOMPAT_CSUM_V3 &&
1907 compat & JBD2_FEATURE_COMPAT_CHECKSUM)
1908 compat &= ~JBD2_FEATURE_COMPAT_CHECKSUM;
1910 jbd_debug(1, "Setting new features 0x%lx/0x%lx/0x%lx\n",
1911 compat, ro, incompat);
1913 sb = journal->j_superblock;
1915 /* Load the checksum driver if necessary */
1916 if ((journal->j_chksum_driver == NULL) &&
1917 INCOMPAT_FEATURE_ON(JBD2_FEATURE_INCOMPAT_CSUM_V3)) {
1918 journal->j_chksum_driver = crypto_alloc_shash("crc32c", 0, 0);
1919 if (IS_ERR(journal->j_chksum_driver)) {
1920 printk(KERN_ERR "JBD2: Cannot load crc32c driver.\n");
1921 journal->j_chksum_driver = NULL;
1924 /* Precompute checksum seed for all metadata */
1925 journal->j_csum_seed = jbd2_chksum(journal, ~0, sb->s_uuid,
1926 sizeof(sb->s_uuid));
1929 lock_buffer(journal->j_sb_buffer);
1931 /* If enabling v3 checksums, update superblock */
1932 if (INCOMPAT_FEATURE_ON(JBD2_FEATURE_INCOMPAT_CSUM_V3)) {
1933 sb->s_checksum_type = JBD2_CRC32C_CHKSUM;
1934 sb->s_feature_compat &=
1935 ~cpu_to_be32(JBD2_FEATURE_COMPAT_CHECKSUM);
1938 /* If enabling v1 checksums, downgrade superblock */
1939 if (COMPAT_FEATURE_ON(JBD2_FEATURE_COMPAT_CHECKSUM))
1940 sb->s_feature_incompat &=
1941 ~cpu_to_be32(JBD2_FEATURE_INCOMPAT_CSUM_V2 |
1942 JBD2_FEATURE_INCOMPAT_CSUM_V3);
1944 sb->s_feature_compat |= cpu_to_be32(compat);
1945 sb->s_feature_ro_compat |= cpu_to_be32(ro);
1946 sb->s_feature_incompat |= cpu_to_be32(incompat);
1947 unlock_buffer(journal->j_sb_buffer);
1948 journal->j_revoke_records_per_block =
1949 journal_revoke_records_per_block(journal);
1952 #undef COMPAT_FEATURE_ON
1953 #undef INCOMPAT_FEATURE_ON
1957 * jbd2_journal_clear_features () - Clear a given journal feature in the
1959 * @journal: Journal to act on.
1960 * @compat: bitmask of compatible features
1961 * @ro: bitmask of features that force read-only mount
1962 * @incompat: bitmask of incompatible features
1964 * Clear a given journal feature as present on the
1967 void jbd2_journal_clear_features(journal_t *journal, unsigned long compat,
1968 unsigned long ro, unsigned long incompat)
1970 journal_superblock_t *sb;
1972 jbd_debug(1, "Clear features 0x%lx/0x%lx/0x%lx\n",
1973 compat, ro, incompat);
1975 sb = journal->j_superblock;
1977 sb->s_feature_compat &= ~cpu_to_be32(compat);
1978 sb->s_feature_ro_compat &= ~cpu_to_be32(ro);
1979 sb->s_feature_incompat &= ~cpu_to_be32(incompat);
1980 journal->j_revoke_records_per_block =
1981 journal_revoke_records_per_block(journal);
1983 EXPORT_SYMBOL(jbd2_journal_clear_features);
1986 * int jbd2_journal_flush () - Flush journal
1987 * @journal: Journal to act on.
1989 * Flush all data for a given journal to disk and empty the journal.
1990 * Filesystems can use this when remounting readonly to ensure that
1991 * recovery does not need to happen on remount.
1994 int jbd2_journal_flush(journal_t *journal)
1997 transaction_t *transaction = NULL;
1999 write_lock(&journal->j_state_lock);
2001 /* Force everything buffered to the log... */
2002 if (journal->j_running_transaction) {
2003 transaction = journal->j_running_transaction;
2004 __jbd2_log_start_commit(journal, transaction->t_tid);
2005 } else if (journal->j_committing_transaction)
2006 transaction = journal->j_committing_transaction;
2008 /* Wait for the log commit to complete... */
2010 tid_t tid = transaction->t_tid;
2012 write_unlock(&journal->j_state_lock);
2013 jbd2_log_wait_commit(journal, tid);
2015 write_unlock(&journal->j_state_lock);
2018 /* ...and flush everything in the log out to disk. */
2019 spin_lock(&journal->j_list_lock);
2020 while (!err && journal->j_checkpoint_transactions != NULL) {
2021 spin_unlock(&journal->j_list_lock);
2022 mutex_lock_io(&journal->j_checkpoint_mutex);
2023 err = jbd2_log_do_checkpoint(journal);
2024 mutex_unlock(&journal->j_checkpoint_mutex);
2025 spin_lock(&journal->j_list_lock);
2027 spin_unlock(&journal->j_list_lock);
2029 if (is_journal_aborted(journal))
2032 mutex_lock_io(&journal->j_checkpoint_mutex);
2034 err = jbd2_cleanup_journal_tail(journal);
2036 mutex_unlock(&journal->j_checkpoint_mutex);
2042 /* Finally, mark the journal as really needing no recovery.
2043 * This sets s_start==0 in the underlying superblock, which is
2044 * the magic code for a fully-recovered superblock. Any future
2045 * commits of data to the journal will restore the current
2047 jbd2_mark_journal_empty(journal, REQ_SYNC | REQ_FUA);
2048 mutex_unlock(&journal->j_checkpoint_mutex);
2049 write_lock(&journal->j_state_lock);
2050 J_ASSERT(!journal->j_running_transaction);
2051 J_ASSERT(!journal->j_committing_transaction);
2052 J_ASSERT(!journal->j_checkpoint_transactions);
2053 J_ASSERT(journal->j_head == journal->j_tail);
2054 J_ASSERT(journal->j_tail_sequence == journal->j_transaction_sequence);
2055 write_unlock(&journal->j_state_lock);
2061 * int jbd2_journal_wipe() - Wipe journal contents
2062 * @journal: Journal to act on.
2063 * @write: flag (see below)
2065 * Wipe out all of the contents of a journal, safely. This will produce
2066 * a warning if the journal contains any valid recovery information.
2067 * Must be called between journal_init_*() and jbd2_journal_load().
2069 * If 'write' is non-zero, then we wipe out the journal on disk; otherwise
2070 * we merely suppress recovery.
2073 int jbd2_journal_wipe(journal_t *journal, int write)
2077 J_ASSERT (!(journal->j_flags & JBD2_LOADED));
2079 err = load_superblock(journal);
2083 if (!journal->j_tail)
2086 printk(KERN_WARNING "JBD2: %s recovery information on journal\n",
2087 write ? "Clearing" : "Ignoring");
2089 err = jbd2_journal_skip_recovery(journal);
2091 /* Lock to make assertions happy... */
2092 mutex_lock_io(&journal->j_checkpoint_mutex);
2093 jbd2_mark_journal_empty(journal, REQ_SYNC | REQ_FUA);
2094 mutex_unlock(&journal->j_checkpoint_mutex);
2102 * Journal abort has very specific semantics, which we describe
2103 * for journal abort.
2105 * Two internal functions, which provide abort to the jbd layer
2110 * Quick version for internal journal use (doesn't lock the journal).
2111 * Aborts hard --- we mark the abort as occurred, but do _nothing_ else,
2112 * and don't attempt to make any other journal updates.
2114 void __jbd2_journal_abort_hard(journal_t *journal)
2116 transaction_t *transaction;
2118 if (journal->j_flags & JBD2_ABORT)
2121 printk(KERN_ERR "Aborting journal on device %s.\n",
2122 journal->j_devname);
2124 write_lock(&journal->j_state_lock);
2125 journal->j_flags |= JBD2_ABORT;
2126 transaction = journal->j_running_transaction;
2128 __jbd2_log_start_commit(journal, transaction->t_tid);
2129 write_unlock(&journal->j_state_lock);
2132 /* Soft abort: record the abort error status in the journal superblock,
2133 * but don't do any other IO. */
2134 static void __journal_abort_soft (journal_t *journal, int errno)
2138 write_lock(&journal->j_state_lock);
2139 old_errno = journal->j_errno;
2140 if (!journal->j_errno || errno == -ESHUTDOWN)
2141 journal->j_errno = errno;
2143 if (journal->j_flags & JBD2_ABORT) {
2144 write_unlock(&journal->j_state_lock);
2145 if (!old_errno && old_errno != -ESHUTDOWN &&
2146 errno == -ESHUTDOWN)
2147 jbd2_journal_update_sb_errno(journal);
2150 write_unlock(&journal->j_state_lock);
2152 __jbd2_journal_abort_hard(journal);
2155 jbd2_journal_update_sb_errno(journal);
2156 write_lock(&journal->j_state_lock);
2157 journal->j_flags |= JBD2_REC_ERR;
2158 write_unlock(&journal->j_state_lock);
2163 * void jbd2_journal_abort () - Shutdown the journal immediately.
2164 * @journal: the journal to shutdown.
2165 * @errno: an error number to record in the journal indicating
2166 * the reason for the shutdown.
2168 * Perform a complete, immediate shutdown of the ENTIRE
2169 * journal (not of a single transaction). This operation cannot be
2170 * undone without closing and reopening the journal.
2172 * The jbd2_journal_abort function is intended to support higher level error
2173 * recovery mechanisms such as the ext2/ext3 remount-readonly error
2176 * Journal abort has very specific semantics. Any existing dirty,
2177 * unjournaled buffers in the main filesystem will still be written to
2178 * disk by bdflush, but the journaling mechanism will be suspended
2179 * immediately and no further transaction commits will be honoured.
2181 * Any dirty, journaled buffers will be written back to disk without
2182 * hitting the journal. Atomicity cannot be guaranteed on an aborted
2183 * filesystem, but we _do_ attempt to leave as much data as possible
2184 * behind for fsck to use for cleanup.
2186 * Any attempt to get a new transaction handle on a journal which is in
2187 * ABORT state will just result in an -EROFS error return. A
2188 * jbd2_journal_stop on an existing handle will return -EIO if we have
2189 * entered abort state during the update.
2191 * Recursive transactions are not disturbed by journal abort until the
2192 * final jbd2_journal_stop, which will receive the -EIO error.
2194 * Finally, the jbd2_journal_abort call allows the caller to supply an errno
2195 * which will be recorded (if possible) in the journal superblock. This
2196 * allows a client to record failure conditions in the middle of a
2197 * transaction without having to complete the transaction to record the
2198 * failure to disk. ext3_error, for example, now uses this
2201 * Errors which originate from within the journaling layer will NOT
2202 * supply an errno; a null errno implies that absolutely no further
2203 * writes are done to the journal (unless there are any already in
2208 void jbd2_journal_abort(journal_t *journal, int errno)
2210 __journal_abort_soft(journal, errno);
2214 * int jbd2_journal_errno () - returns the journal's error state.
2215 * @journal: journal to examine.
2217 * This is the errno number set with jbd2_journal_abort(), the last
2218 * time the journal was mounted - if the journal was stopped
2219 * without calling abort this will be 0.
2221 * If the journal has been aborted on this mount time -EROFS will
2224 int jbd2_journal_errno(journal_t *journal)
2228 read_lock(&journal->j_state_lock);
2229 if (journal->j_flags & JBD2_ABORT)
2232 err = journal->j_errno;
2233 read_unlock(&journal->j_state_lock);
2238 * int jbd2_journal_clear_err () - clears the journal's error state
2239 * @journal: journal to act on.
2241 * An error must be cleared or acked to take a FS out of readonly
2244 int jbd2_journal_clear_err(journal_t *journal)
2248 write_lock(&journal->j_state_lock);
2249 if (journal->j_flags & JBD2_ABORT)
2252 journal->j_errno = 0;
2253 write_unlock(&journal->j_state_lock);
2258 * void jbd2_journal_ack_err() - Ack journal err.
2259 * @journal: journal to act on.
2261 * An error must be cleared or acked to take a FS out of readonly
2264 void jbd2_journal_ack_err(journal_t *journal)
2266 write_lock(&journal->j_state_lock);
2267 if (journal->j_errno)
2268 journal->j_flags |= JBD2_ACK_ERR;
2269 write_unlock(&journal->j_state_lock);
2272 int jbd2_journal_blocks_per_page(struct inode *inode)
2274 return 1 << (PAGE_SHIFT - inode->i_sb->s_blocksize_bits);
2278 * helper functions to deal with 32 or 64bit block numbers.
2280 size_t journal_tag_bytes(journal_t *journal)
2284 if (jbd2_has_feature_csum3(journal))
2285 return sizeof(journal_block_tag3_t);
2287 sz = sizeof(journal_block_tag_t);
2289 if (jbd2_has_feature_csum2(journal))
2290 sz += sizeof(__u16);
2292 if (jbd2_has_feature_64bit(journal))
2295 return sz - sizeof(__u32);
2299 * JBD memory management
2301 * These functions are used to allocate block-sized chunks of memory
2302 * used for making copies of buffer_head data. Very often it will be
2303 * page-sized chunks of data, but sometimes it will be in
2304 * sub-page-size chunks. (For example, 16k pages on Power systems
2305 * with a 4k block file system.) For blocks smaller than a page, we
2306 * use a SLAB allocator. There are slab caches for each block size,
2307 * which are allocated at mount time, if necessary, and we only free
2308 * (all of) the slab caches when/if the jbd2 module is unloaded. For
2309 * this reason we don't need to a mutex to protect access to
2310 * jbd2_slab[] allocating or releasing memory; only in
2311 * jbd2_journal_create_slab().
2313 #define JBD2_MAX_SLABS 8
2314 static struct kmem_cache *jbd2_slab[JBD2_MAX_SLABS];
2316 static const char *jbd2_slab_names[JBD2_MAX_SLABS] = {
2317 "jbd2_1k", "jbd2_2k", "jbd2_4k", "jbd2_8k",
2318 "jbd2_16k", "jbd2_32k", "jbd2_64k", "jbd2_128k"
2322 static void jbd2_journal_destroy_slabs(void)
2326 for (i = 0; i < JBD2_MAX_SLABS; i++) {
2327 kmem_cache_destroy(jbd2_slab[i]);
2328 jbd2_slab[i] = NULL;
2332 static int jbd2_journal_create_slab(size_t size)
2334 static DEFINE_MUTEX(jbd2_slab_create_mutex);
2335 int i = order_base_2(size) - 10;
2338 if (size == PAGE_SIZE)
2341 if (i >= JBD2_MAX_SLABS)
2344 if (unlikely(i < 0))
2346 mutex_lock(&jbd2_slab_create_mutex);
2348 mutex_unlock(&jbd2_slab_create_mutex);
2349 return 0; /* Already created */
2352 slab_size = 1 << (i+10);
2353 jbd2_slab[i] = kmem_cache_create(jbd2_slab_names[i], slab_size,
2354 slab_size, 0, NULL);
2355 mutex_unlock(&jbd2_slab_create_mutex);
2356 if (!jbd2_slab[i]) {
2357 printk(KERN_EMERG "JBD2: no memory for jbd2_slab cache\n");
2363 static struct kmem_cache *get_slab(size_t size)
2365 int i = order_base_2(size) - 10;
2367 BUG_ON(i >= JBD2_MAX_SLABS);
2368 if (unlikely(i < 0))
2370 BUG_ON(jbd2_slab[i] == NULL);
2371 return jbd2_slab[i];
2374 void *jbd2_alloc(size_t size, gfp_t flags)
2378 BUG_ON(size & (size-1)); /* Must be a power of 2 */
2380 if (size < PAGE_SIZE)
2381 ptr = kmem_cache_alloc(get_slab(size), flags);
2383 ptr = (void *)__get_free_pages(flags, get_order(size));
2385 /* Check alignment; SLUB has gotten this wrong in the past,
2386 * and this can lead to user data corruption! */
2387 BUG_ON(((unsigned long) ptr) & (size-1));
2392 void jbd2_free(void *ptr, size_t size)
2394 if (size < PAGE_SIZE)
2395 kmem_cache_free(get_slab(size), ptr);
2397 free_pages((unsigned long)ptr, get_order(size));
2401 * Journal_head storage management
2403 static struct kmem_cache *jbd2_journal_head_cache;
2404 #ifdef CONFIG_JBD2_DEBUG
2405 static atomic_t nr_journal_heads = ATOMIC_INIT(0);
2408 static int __init jbd2_journal_init_journal_head_cache(void)
2410 J_ASSERT(!jbd2_journal_head_cache);
2411 jbd2_journal_head_cache = kmem_cache_create("jbd2_journal_head",
2412 sizeof(struct journal_head),
2414 SLAB_TEMPORARY | SLAB_TYPESAFE_BY_RCU,
2416 if (!jbd2_journal_head_cache) {
2417 printk(KERN_EMERG "JBD2: no memory for journal_head cache\n");
2423 static void jbd2_journal_destroy_journal_head_cache(void)
2425 kmem_cache_destroy(jbd2_journal_head_cache);
2426 jbd2_journal_head_cache = NULL;
2430 * journal_head splicing and dicing
2432 static struct journal_head *journal_alloc_journal_head(void)
2434 struct journal_head *ret;
2436 #ifdef CONFIG_JBD2_DEBUG
2437 atomic_inc(&nr_journal_heads);
2439 ret = kmem_cache_zalloc(jbd2_journal_head_cache, GFP_NOFS);
2441 jbd_debug(1, "out of memory for journal_head\n");
2442 pr_notice_ratelimited("ENOMEM in %s, retrying.\n", __func__);
2443 ret = kmem_cache_zalloc(jbd2_journal_head_cache,
2444 GFP_NOFS | __GFP_NOFAIL);
2447 spin_lock_init(&ret->b_state_lock);
2451 static void journal_free_journal_head(struct journal_head *jh)
2453 #ifdef CONFIG_JBD2_DEBUG
2454 atomic_dec(&nr_journal_heads);
2455 memset(jh, JBD2_POISON_FREE, sizeof(*jh));
2457 kmem_cache_free(jbd2_journal_head_cache, jh);
2461 * A journal_head is attached to a buffer_head whenever JBD has an
2462 * interest in the buffer.
2464 * Whenever a buffer has an attached journal_head, its ->b_state:BH_JBD bit
2465 * is set. This bit is tested in core kernel code where we need to take
2466 * JBD-specific actions. Testing the zeroness of ->b_private is not reliable
2469 * When a buffer has its BH_JBD bit set, its ->b_count is elevated by one.
2471 * When a buffer has its BH_JBD bit set it is immune from being released by
2472 * core kernel code, mainly via ->b_count.
2474 * A journal_head is detached from its buffer_head when the journal_head's
2475 * b_jcount reaches zero. Running transaction (b_transaction) and checkpoint
2476 * transaction (b_cp_transaction) hold their references to b_jcount.
2478 * Various places in the kernel want to attach a journal_head to a buffer_head
2479 * _before_ attaching the journal_head to a transaction. To protect the
2480 * journal_head in this situation, jbd2_journal_add_journal_head elevates the
2481 * journal_head's b_jcount refcount by one. The caller must call
2482 * jbd2_journal_put_journal_head() to undo this.
2484 * So the typical usage would be:
2486 * (Attach a journal_head if needed. Increments b_jcount)
2487 * struct journal_head *jh = jbd2_journal_add_journal_head(bh);
2489 * (Get another reference for transaction)
2490 * jbd2_journal_grab_journal_head(bh);
2491 * jh->b_transaction = xxx;
2492 * (Put original reference)
2493 * jbd2_journal_put_journal_head(jh);
2497 * Give a buffer_head a journal_head.
2501 struct journal_head *jbd2_journal_add_journal_head(struct buffer_head *bh)
2503 struct journal_head *jh;
2504 struct journal_head *new_jh = NULL;
2507 if (!buffer_jbd(bh))
2508 new_jh = journal_alloc_journal_head();
2510 jbd_lock_bh_journal_head(bh);
2511 if (buffer_jbd(bh)) {
2515 (atomic_read(&bh->b_count) > 0) ||
2516 (bh->b_page && bh->b_page->mapping));
2519 jbd_unlock_bh_journal_head(bh);
2524 new_jh = NULL; /* We consumed it */
2529 BUFFER_TRACE(bh, "added journal_head");
2532 jbd_unlock_bh_journal_head(bh);
2534 journal_free_journal_head(new_jh);
2535 return bh->b_private;
2539 * Grab a ref against this buffer_head's journal_head. If it ended up not
2540 * having a journal_head, return NULL
2542 struct journal_head *jbd2_journal_grab_journal_head(struct buffer_head *bh)
2544 struct journal_head *jh = NULL;
2546 jbd_lock_bh_journal_head(bh);
2547 if (buffer_jbd(bh)) {
2551 jbd_unlock_bh_journal_head(bh);
2555 static void __journal_remove_journal_head(struct buffer_head *bh)
2557 struct journal_head *jh = bh2jh(bh);
2559 J_ASSERT_JH(jh, jh->b_jcount >= 0);
2560 J_ASSERT_JH(jh, jh->b_transaction == NULL);
2561 J_ASSERT_JH(jh, jh->b_next_transaction == NULL);
2562 J_ASSERT_JH(jh, jh->b_cp_transaction == NULL);
2563 J_ASSERT_JH(jh, jh->b_jlist == BJ_None);
2564 J_ASSERT_BH(bh, buffer_jbd(bh));
2565 J_ASSERT_BH(bh, jh2bh(jh) == bh);
2566 BUFFER_TRACE(bh, "remove journal_head");
2568 /* Unlink before dropping the lock */
2569 bh->b_private = NULL;
2570 jh->b_bh = NULL; /* debug, really */
2571 clear_buffer_jbd(bh);
2574 static void journal_release_journal_head(struct journal_head *jh, size_t b_size)
2576 if (jh->b_frozen_data) {
2577 printk(KERN_WARNING "%s: freeing b_frozen_data\n", __func__);
2578 jbd2_free(jh->b_frozen_data, b_size);
2580 if (jh->b_committed_data) {
2581 printk(KERN_WARNING "%s: freeing b_committed_data\n", __func__);
2582 jbd2_free(jh->b_committed_data, b_size);
2584 journal_free_journal_head(jh);
2588 * Drop a reference on the passed journal_head. If it fell to zero then
2589 * release the journal_head from the buffer_head.
2591 void jbd2_journal_put_journal_head(struct journal_head *jh)
2593 struct buffer_head *bh = jh2bh(jh);
2595 jbd_lock_bh_journal_head(bh);
2596 J_ASSERT_JH(jh, jh->b_jcount > 0);
2598 if (!jh->b_jcount) {
2599 __journal_remove_journal_head(bh);
2600 jbd_unlock_bh_journal_head(bh);
2601 journal_release_journal_head(jh, bh->b_size);
2604 jbd_unlock_bh_journal_head(bh);
2609 * Initialize jbd inode head
2611 void jbd2_journal_init_jbd_inode(struct jbd2_inode *jinode, struct inode *inode)
2613 jinode->i_transaction = NULL;
2614 jinode->i_next_transaction = NULL;
2615 jinode->i_vfs_inode = inode;
2616 jinode->i_flags = 0;
2617 jinode->i_dirty_start = 0;
2618 jinode->i_dirty_end = 0;
2619 INIT_LIST_HEAD(&jinode->i_list);
2623 * Function to be called before we start removing inode from memory (i.e.,
2624 * clear_inode() is a fine place to be called from). It removes inode from
2625 * transaction's lists.
2627 void jbd2_journal_release_jbd_inode(journal_t *journal,
2628 struct jbd2_inode *jinode)
2633 spin_lock(&journal->j_list_lock);
2634 /* Is commit writing out inode - we have to wait */
2635 if (jinode->i_flags & JI_COMMIT_RUNNING) {
2636 wait_queue_head_t *wq;
2637 DEFINE_WAIT_BIT(wait, &jinode->i_flags, __JI_COMMIT_RUNNING);
2638 wq = bit_waitqueue(&jinode->i_flags, __JI_COMMIT_RUNNING);
2639 prepare_to_wait(wq, &wait.wq_entry, TASK_UNINTERRUPTIBLE);
2640 spin_unlock(&journal->j_list_lock);
2642 finish_wait(wq, &wait.wq_entry);
2646 if (jinode->i_transaction) {
2647 list_del(&jinode->i_list);
2648 jinode->i_transaction = NULL;
2650 spin_unlock(&journal->j_list_lock);
2654 #ifdef CONFIG_PROC_FS
2656 #define JBD2_STATS_PROC_NAME "fs/jbd2"
2658 static void __init jbd2_create_jbd_stats_proc_entry(void)
2660 proc_jbd2_stats = proc_mkdir(JBD2_STATS_PROC_NAME, NULL);
2663 static void __exit jbd2_remove_jbd_stats_proc_entry(void)
2665 if (proc_jbd2_stats)
2666 remove_proc_entry(JBD2_STATS_PROC_NAME, NULL);
2671 #define jbd2_create_jbd_stats_proc_entry() do {} while (0)
2672 #define jbd2_remove_jbd_stats_proc_entry() do {} while (0)
2676 struct kmem_cache *jbd2_handle_cache, *jbd2_inode_cache;
2678 static int __init jbd2_journal_init_inode_cache(void)
2680 J_ASSERT(!jbd2_inode_cache);
2681 jbd2_inode_cache = KMEM_CACHE(jbd2_inode, 0);
2682 if (!jbd2_inode_cache) {
2683 pr_emerg("JBD2: failed to create inode cache\n");
2689 static int __init jbd2_journal_init_handle_cache(void)
2691 J_ASSERT(!jbd2_handle_cache);
2692 jbd2_handle_cache = KMEM_CACHE(jbd2_journal_handle, SLAB_TEMPORARY);
2693 if (!jbd2_handle_cache) {
2694 printk(KERN_EMERG "JBD2: failed to create handle cache\n");
2700 static void jbd2_journal_destroy_inode_cache(void)
2702 kmem_cache_destroy(jbd2_inode_cache);
2703 jbd2_inode_cache = NULL;
2706 static void jbd2_journal_destroy_handle_cache(void)
2708 kmem_cache_destroy(jbd2_handle_cache);
2709 jbd2_handle_cache = NULL;
2713 * Module startup and shutdown
2716 static int __init journal_init_caches(void)
2720 ret = jbd2_journal_init_revoke_record_cache();
2722 ret = jbd2_journal_init_revoke_table_cache();
2724 ret = jbd2_journal_init_journal_head_cache();
2726 ret = jbd2_journal_init_handle_cache();
2728 ret = jbd2_journal_init_inode_cache();
2730 ret = jbd2_journal_init_transaction_cache();
2734 static void jbd2_journal_destroy_caches(void)
2736 jbd2_journal_destroy_revoke_record_cache();
2737 jbd2_journal_destroy_revoke_table_cache();
2738 jbd2_journal_destroy_journal_head_cache();
2739 jbd2_journal_destroy_handle_cache();
2740 jbd2_journal_destroy_inode_cache();
2741 jbd2_journal_destroy_transaction_cache();
2742 jbd2_journal_destroy_slabs();
2745 static int __init journal_init(void)
2749 BUILD_BUG_ON(sizeof(struct journal_superblock_s) != 1024);
2751 ret = journal_init_caches();
2753 jbd2_create_jbd_stats_proc_entry();
2755 jbd2_journal_destroy_caches();
2760 static void __exit journal_exit(void)
2762 #ifdef CONFIG_JBD2_DEBUG
2763 int n = atomic_read(&nr_journal_heads);
2765 printk(KERN_ERR "JBD2: leaked %d journal_heads!\n", n);
2767 jbd2_remove_jbd_stats_proc_entry();
2768 jbd2_journal_destroy_caches();
2771 MODULE_LICENSE("GPL");
2772 module_init(journal_init);
2773 module_exit(journal_exit);