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 int jbd2_journal_create_slab(size_t slab_size);
101 #ifdef CONFIG_JBD2_DEBUG
102 void __jbd2_debug(int level, const char *file, const char *func,
103 unsigned int line, const char *fmt, ...)
105 struct va_format vaf;
108 if (level > jbd2_journal_enable_debug)
113 printk(KERN_DEBUG "%s: (%s, %u): %pV", file, func, line, &vaf);
116 EXPORT_SYMBOL(__jbd2_debug);
119 /* Checksumming functions */
120 static int jbd2_verify_csum_type(journal_t *j, journal_superblock_t *sb)
122 if (!jbd2_journal_has_csum_v2or3_feature(j))
125 return sb->s_checksum_type == JBD2_CRC32C_CHKSUM;
128 static __be32 jbd2_superblock_csum(journal_t *j, journal_superblock_t *sb)
133 old_csum = sb->s_checksum;
135 csum = jbd2_chksum(j, ~0, (char *)sb, sizeof(journal_superblock_t));
136 sb->s_checksum = old_csum;
138 return cpu_to_be32(csum);
142 * Helper function used to manage commit timeouts
145 static void commit_timeout(struct timer_list *t)
147 journal_t *journal = from_timer(journal, t, j_commit_timer);
149 wake_up_process(journal->j_task);
153 * kjournald2: The main thread function used to manage a logging device
156 * This kernel thread is responsible for two things:
158 * 1) COMMIT: Every so often we need to commit the current state of the
159 * filesystem to disk. The journal thread is responsible for writing
160 * all of the metadata buffers to disk.
162 * 2) CHECKPOINT: We cannot reuse a used section of the log file until all
163 * of the data in that part of the log has been rewritten elsewhere on
164 * the disk. Flushing these old buffers to reclaim space in the log is
165 * known as checkpointing, and this thread is responsible for that job.
168 static int kjournald2(void *arg)
170 journal_t *journal = arg;
171 transaction_t *transaction;
174 * Set up an interval timer which can be used to trigger a commit wakeup
175 * after the commit interval expires
177 timer_setup(&journal->j_commit_timer, commit_timeout, 0);
181 /* Record that the journal thread is running */
182 journal->j_task = current;
183 wake_up(&journal->j_wait_done_commit);
186 * Make sure that no allocations from this kernel thread will ever
187 * recurse to the fs layer because we are responsible for the
188 * transaction commit and any fs involvement might get stuck waiting for
191 memalloc_nofs_save();
194 * And now, wait forever for commit wakeup events.
196 write_lock(&journal->j_state_lock);
199 if (journal->j_flags & JBD2_UNMOUNT)
202 jbd_debug(1, "commit_sequence=%u, commit_request=%u\n",
203 journal->j_commit_sequence, journal->j_commit_request);
205 if (journal->j_commit_sequence != journal->j_commit_request) {
206 jbd_debug(1, "OK, requests differ\n");
207 write_unlock(&journal->j_state_lock);
208 del_timer_sync(&journal->j_commit_timer);
209 jbd2_journal_commit_transaction(journal);
210 write_lock(&journal->j_state_lock);
214 wake_up(&journal->j_wait_done_commit);
215 if (freezing(current)) {
217 * The simpler the better. Flushing journal isn't a
218 * good idea, because that depends on threads that may
219 * be already stopped.
221 jbd_debug(1, "Now suspending kjournald2\n");
222 write_unlock(&journal->j_state_lock);
224 write_lock(&journal->j_state_lock);
227 * We assume on resume that commits are already there,
231 int should_sleep = 1;
233 prepare_to_wait(&journal->j_wait_commit, &wait,
235 if (journal->j_commit_sequence != journal->j_commit_request)
237 transaction = journal->j_running_transaction;
238 if (transaction && time_after_eq(jiffies,
239 transaction->t_expires))
241 if (journal->j_flags & JBD2_UNMOUNT)
244 write_unlock(&journal->j_state_lock);
246 write_lock(&journal->j_state_lock);
248 finish_wait(&journal->j_wait_commit, &wait);
251 jbd_debug(1, "kjournald2 wakes\n");
254 * Were we woken up by a commit wakeup event?
256 transaction = journal->j_running_transaction;
257 if (transaction && time_after_eq(jiffies, transaction->t_expires)) {
258 journal->j_commit_request = transaction->t_tid;
259 jbd_debug(1, "woke because of timeout\n");
264 del_timer_sync(&journal->j_commit_timer);
265 journal->j_task = NULL;
266 wake_up(&journal->j_wait_done_commit);
267 jbd_debug(1, "Journal thread exiting.\n");
268 write_unlock(&journal->j_state_lock);
272 static int jbd2_journal_start_thread(journal_t *journal)
274 struct task_struct *t;
276 t = kthread_run(kjournald2, journal, "jbd2/%s",
281 wait_event(journal->j_wait_done_commit, journal->j_task != NULL);
285 static void journal_kill_thread(journal_t *journal)
287 write_lock(&journal->j_state_lock);
288 journal->j_flags |= JBD2_UNMOUNT;
290 while (journal->j_task) {
291 write_unlock(&journal->j_state_lock);
292 wake_up(&journal->j_wait_commit);
293 wait_event(journal->j_wait_done_commit, journal->j_task == NULL);
294 write_lock(&journal->j_state_lock);
296 write_unlock(&journal->j_state_lock);
300 * jbd2_journal_write_metadata_buffer: write a metadata buffer to the journal.
302 * Writes a metadata buffer to a given disk block. The actual IO is not
303 * performed but a new buffer_head is constructed which labels the data
304 * to be written with the correct destination disk block.
306 * Any magic-number escaping which needs to be done will cause a
307 * copy-out here. If the buffer happens to start with the
308 * JBD2_MAGIC_NUMBER, then we can't write it to the log directly: the
309 * magic number is only written to the log for descripter blocks. In
310 * this case, we copy the data and replace the first word with 0, and we
311 * return a result code which indicates that this buffer needs to be
312 * marked as an escaped buffer in the corresponding log descriptor
313 * block. The missing word can then be restored when the block is read
316 * If the source buffer has already been modified by a new transaction
317 * since we took the last commit snapshot, we use the frozen copy of
318 * that data for IO. If we end up using the existing buffer_head's data
319 * for the write, then we have to make sure nobody modifies it while the
320 * IO is in progress. do_get_write_access() handles this.
322 * The function returns a pointer to the buffer_head to be used for IO.
330 * Bit 0 set == escape performed on the data
331 * Bit 1 set == buffer copy-out performed (kfree the data after IO)
334 int jbd2_journal_write_metadata_buffer(transaction_t *transaction,
335 struct journal_head *jh_in,
336 struct buffer_head **bh_out,
339 int need_copy_out = 0;
340 int done_copy_out = 0;
343 struct buffer_head *new_bh;
344 struct page *new_page;
345 unsigned int new_offset;
346 struct buffer_head *bh_in = jh2bh(jh_in);
347 journal_t *journal = transaction->t_journal;
350 * The buffer really shouldn't be locked: only the current committing
351 * transaction is allowed to write it, so nobody else is allowed
354 * akpm: except if we're journalling data, and write() output is
355 * also part of a shared mapping, and another thread has
356 * decided to launch a writepage() against this buffer.
358 J_ASSERT_BH(bh_in, buffer_jbddirty(bh_in));
360 new_bh = alloc_buffer_head(GFP_NOFS|__GFP_NOFAIL);
362 /* keep subsequent assertions sane */
363 atomic_set(&new_bh->b_count, 1);
365 spin_lock(&jh_in->b_state_lock);
368 * If a new transaction has already done a buffer copy-out, then
369 * we use that version of the data for the commit.
371 if (jh_in->b_frozen_data) {
373 new_page = virt_to_page(jh_in->b_frozen_data);
374 new_offset = offset_in_page(jh_in->b_frozen_data);
376 new_page = jh2bh(jh_in)->b_page;
377 new_offset = offset_in_page(jh2bh(jh_in)->b_data);
380 mapped_data = kmap_atomic(new_page);
382 * Fire data frozen trigger if data already wasn't frozen. Do this
383 * before checking for escaping, as the trigger may modify the magic
384 * offset. If a copy-out happens afterwards, it will have the correct
385 * data in the buffer.
388 jbd2_buffer_frozen_trigger(jh_in, mapped_data + new_offset,
394 if (*((__be32 *)(mapped_data + new_offset)) ==
395 cpu_to_be32(JBD2_MAGIC_NUMBER)) {
399 kunmap_atomic(mapped_data);
402 * Do we need to do a data copy?
404 if (need_copy_out && !done_copy_out) {
407 spin_unlock(&jh_in->b_state_lock);
408 tmp = jbd2_alloc(bh_in->b_size, GFP_NOFS);
413 spin_lock(&jh_in->b_state_lock);
414 if (jh_in->b_frozen_data) {
415 jbd2_free(tmp, bh_in->b_size);
419 jh_in->b_frozen_data = tmp;
420 mapped_data = kmap_atomic(new_page);
421 memcpy(tmp, mapped_data + new_offset, bh_in->b_size);
422 kunmap_atomic(mapped_data);
424 new_page = virt_to_page(tmp);
425 new_offset = offset_in_page(tmp);
429 * This isn't strictly necessary, as we're using frozen
430 * data for the escaping, but it keeps consistency with
431 * b_frozen_data usage.
433 jh_in->b_frozen_triggers = jh_in->b_triggers;
437 * Did we need to do an escaping? Now we've done all the
438 * copying, we can finally do so.
441 mapped_data = kmap_atomic(new_page);
442 *((unsigned int *)(mapped_data + new_offset)) = 0;
443 kunmap_atomic(mapped_data);
446 set_bh_page(new_bh, new_page, new_offset);
447 new_bh->b_size = bh_in->b_size;
448 new_bh->b_bdev = journal->j_dev;
449 new_bh->b_blocknr = blocknr;
450 new_bh->b_private = bh_in;
451 set_buffer_mapped(new_bh);
452 set_buffer_dirty(new_bh);
457 * The to-be-written buffer needs to get moved to the io queue,
458 * and the original buffer whose contents we are shadowing or
459 * copying is moved to the transaction's shadow queue.
461 JBUFFER_TRACE(jh_in, "file as BJ_Shadow");
462 spin_lock(&journal->j_list_lock);
463 __jbd2_journal_file_buffer(jh_in, transaction, BJ_Shadow);
464 spin_unlock(&journal->j_list_lock);
465 set_buffer_shadow(bh_in);
466 spin_unlock(&jh_in->b_state_lock);
468 return do_escape | (done_copy_out << 1);
472 * Allocation code for the journal file. Manage the space left in the
473 * journal, so that we can begin checkpointing when appropriate.
477 * Called with j_state_lock locked for writing.
478 * Returns true if a transaction commit was started.
480 int __jbd2_log_start_commit(journal_t *journal, tid_t target)
482 /* Return if the txn has already requested to be committed */
483 if (journal->j_commit_request == target)
487 * The only transaction we can possibly wait upon is the
488 * currently running transaction (if it exists). Otherwise,
489 * the target tid must be an old one.
491 if (journal->j_running_transaction &&
492 journal->j_running_transaction->t_tid == target) {
494 * We want a new commit: OK, mark the request and wakeup the
495 * commit thread. We do _not_ do the commit ourselves.
498 journal->j_commit_request = target;
499 jbd_debug(1, "JBD2: requesting commit %u/%u\n",
500 journal->j_commit_request,
501 journal->j_commit_sequence);
502 journal->j_running_transaction->t_requested = jiffies;
503 wake_up(&journal->j_wait_commit);
505 } else if (!tid_geq(journal->j_commit_request, target))
506 /* This should never happen, but if it does, preserve
507 the evidence before kjournald goes into a loop and
508 increments j_commit_sequence beyond all recognition. */
509 WARN_ONCE(1, "JBD2: bad log_start_commit: %u %u %u %u\n",
510 journal->j_commit_request,
511 journal->j_commit_sequence,
512 target, journal->j_running_transaction ?
513 journal->j_running_transaction->t_tid : 0);
517 int jbd2_log_start_commit(journal_t *journal, tid_t tid)
521 write_lock(&journal->j_state_lock);
522 ret = __jbd2_log_start_commit(journal, tid);
523 write_unlock(&journal->j_state_lock);
528 * Force and wait any uncommitted transactions. We can only force the running
529 * transaction if we don't have an active handle, otherwise, we will deadlock.
530 * Returns: <0 in case of error,
531 * 0 if nothing to commit,
532 * 1 if transaction was successfully committed.
534 static int __jbd2_journal_force_commit(journal_t *journal)
536 transaction_t *transaction = NULL;
538 int need_to_start = 0, ret = 0;
540 read_lock(&journal->j_state_lock);
541 if (journal->j_running_transaction && !current->journal_info) {
542 transaction = journal->j_running_transaction;
543 if (!tid_geq(journal->j_commit_request, transaction->t_tid))
545 } else if (journal->j_committing_transaction)
546 transaction = journal->j_committing_transaction;
549 /* Nothing to commit */
550 read_unlock(&journal->j_state_lock);
553 tid = transaction->t_tid;
554 read_unlock(&journal->j_state_lock);
556 jbd2_log_start_commit(journal, tid);
557 ret = jbd2_log_wait_commit(journal, tid);
565 * Force and wait upon a commit if the calling process is not within
566 * transaction. This is used for forcing out undo-protected data which contains
567 * bitmaps, when the fs is running out of space.
569 * @journal: journal to force
570 * Returns true if progress was made.
572 int jbd2_journal_force_commit_nested(journal_t *journal)
576 ret = __jbd2_journal_force_commit(journal);
581 * int journal_force_commit() - force any uncommitted transactions
582 * @journal: journal to force
584 * Caller want unconditional commit. We can only force the running transaction
585 * if we don't have an active handle, otherwise, we will deadlock.
587 int jbd2_journal_force_commit(journal_t *journal)
591 J_ASSERT(!current->journal_info);
592 ret = __jbd2_journal_force_commit(journal);
599 * Start a commit of the current running transaction (if any). Returns true
600 * if a transaction is going to be committed (or is currently already
601 * committing), and fills its tid in at *ptid
603 int jbd2_journal_start_commit(journal_t *journal, tid_t *ptid)
607 write_lock(&journal->j_state_lock);
608 if (journal->j_running_transaction) {
609 tid_t tid = journal->j_running_transaction->t_tid;
611 __jbd2_log_start_commit(journal, tid);
612 /* There's a running transaction and we've just made sure
613 * it's commit has been scheduled. */
617 } else if (journal->j_committing_transaction) {
619 * If commit has been started, then we have to wait for
620 * completion of that transaction.
623 *ptid = journal->j_committing_transaction->t_tid;
626 write_unlock(&journal->j_state_lock);
631 * Return 1 if a given transaction has not yet sent barrier request
632 * connected with a transaction commit. If 0 is returned, transaction
633 * may or may not have sent the barrier. Used to avoid sending barrier
634 * twice in common cases.
636 int jbd2_trans_will_send_data_barrier(journal_t *journal, tid_t tid)
639 transaction_t *commit_trans;
641 if (!(journal->j_flags & JBD2_BARRIER))
643 read_lock(&journal->j_state_lock);
644 /* Transaction already committed? */
645 if (tid_geq(journal->j_commit_sequence, tid))
647 commit_trans = journal->j_committing_transaction;
648 if (!commit_trans || commit_trans->t_tid != tid) {
653 * Transaction is being committed and we already proceeded to
654 * submitting a flush to fs partition?
656 if (journal->j_fs_dev != journal->j_dev) {
657 if (!commit_trans->t_need_data_flush ||
658 commit_trans->t_state >= T_COMMIT_DFLUSH)
661 if (commit_trans->t_state >= T_COMMIT_JFLUSH)
666 read_unlock(&journal->j_state_lock);
669 EXPORT_SYMBOL(jbd2_trans_will_send_data_barrier);
672 * Wait for a specified commit to complete.
673 * The caller may not hold the journal lock.
675 int jbd2_log_wait_commit(journal_t *journal, tid_t tid)
679 read_lock(&journal->j_state_lock);
680 #ifdef CONFIG_PROVE_LOCKING
682 * Some callers make sure transaction is already committing and in that
683 * case we cannot block on open handles anymore. So don't warn in that
686 if (tid_gt(tid, journal->j_commit_sequence) &&
687 (!journal->j_committing_transaction ||
688 journal->j_committing_transaction->t_tid != tid)) {
689 read_unlock(&journal->j_state_lock);
690 jbd2_might_wait_for_commit(journal);
691 read_lock(&journal->j_state_lock);
694 #ifdef CONFIG_JBD2_DEBUG
695 if (!tid_geq(journal->j_commit_request, tid)) {
697 "%s: error: j_commit_request=%u, tid=%u\n",
698 __func__, journal->j_commit_request, tid);
701 while (tid_gt(tid, journal->j_commit_sequence)) {
702 jbd_debug(1, "JBD2: want %u, j_commit_sequence=%u\n",
703 tid, journal->j_commit_sequence);
704 read_unlock(&journal->j_state_lock);
705 wake_up(&journal->j_wait_commit);
706 wait_event(journal->j_wait_done_commit,
707 !tid_gt(tid, journal->j_commit_sequence));
708 read_lock(&journal->j_state_lock);
710 read_unlock(&journal->j_state_lock);
712 if (unlikely(is_journal_aborted(journal)))
717 /* Return 1 when transaction with given tid has already committed. */
718 int jbd2_transaction_committed(journal_t *journal, tid_t tid)
722 read_lock(&journal->j_state_lock);
723 if (journal->j_running_transaction &&
724 journal->j_running_transaction->t_tid == tid)
726 if (journal->j_committing_transaction &&
727 journal->j_committing_transaction->t_tid == tid)
729 read_unlock(&journal->j_state_lock);
732 EXPORT_SYMBOL(jbd2_transaction_committed);
735 * When this function returns the transaction corresponding to tid
736 * will be completed. If the transaction has currently running, start
737 * committing that transaction before waiting for it to complete. If
738 * the transaction id is stale, it is by definition already completed,
739 * so just return SUCCESS.
741 int jbd2_complete_transaction(journal_t *journal, tid_t tid)
743 int need_to_wait = 1;
745 read_lock(&journal->j_state_lock);
746 if (journal->j_running_transaction &&
747 journal->j_running_transaction->t_tid == tid) {
748 if (journal->j_commit_request != tid) {
749 /* transaction not yet started, so request it */
750 read_unlock(&journal->j_state_lock);
751 jbd2_log_start_commit(journal, tid);
754 } else if (!(journal->j_committing_transaction &&
755 journal->j_committing_transaction->t_tid == tid))
757 read_unlock(&journal->j_state_lock);
761 return jbd2_log_wait_commit(journal, tid);
763 EXPORT_SYMBOL(jbd2_complete_transaction);
766 * Log buffer allocation routines:
769 int jbd2_journal_next_log_block(journal_t *journal, unsigned long long *retp)
771 unsigned long blocknr;
773 write_lock(&journal->j_state_lock);
774 J_ASSERT(journal->j_free > 1);
776 blocknr = journal->j_head;
779 if (journal->j_head == journal->j_last)
780 journal->j_head = journal->j_first;
781 write_unlock(&journal->j_state_lock);
782 return jbd2_journal_bmap(journal, blocknr, retp);
786 * Conversion of logical to physical block numbers for the journal
788 * On external journals the journal blocks are identity-mapped, so
789 * this is a no-op. If needed, we can use j_blk_offset - everything is
792 int jbd2_journal_bmap(journal_t *journal, unsigned long blocknr,
793 unsigned long long *retp)
796 unsigned long long ret;
798 if (journal->j_inode) {
799 ret = bmap(journal->j_inode, blocknr);
803 printk(KERN_ALERT "%s: journal block not found "
804 "at offset %lu on %s\n",
805 __func__, blocknr, journal->j_devname);
807 jbd2_journal_abort(journal, err);
810 *retp = blocknr; /* +journal->j_blk_offset */
816 * We play buffer_head aliasing tricks to write data/metadata blocks to
817 * the journal without copying their contents, but for journal
818 * descriptor blocks we do need to generate bona fide buffers.
820 * After the caller of jbd2_journal_get_descriptor_buffer() has finished modifying
821 * the buffer's contents they really should run flush_dcache_page(bh->b_page).
822 * But we don't bother doing that, so there will be coherency problems with
823 * mmaps of blockdevs which hold live JBD-controlled filesystems.
826 jbd2_journal_get_descriptor_buffer(transaction_t *transaction, int type)
828 journal_t *journal = transaction->t_journal;
829 struct buffer_head *bh;
830 unsigned long long blocknr;
831 journal_header_t *header;
834 err = jbd2_journal_next_log_block(journal, &blocknr);
839 bh = __getblk(journal->j_dev, blocknr, journal->j_blocksize);
842 atomic_dec(&transaction->t_outstanding_credits);
844 memset(bh->b_data, 0, journal->j_blocksize);
845 header = (journal_header_t *)bh->b_data;
846 header->h_magic = cpu_to_be32(JBD2_MAGIC_NUMBER);
847 header->h_blocktype = cpu_to_be32(type);
848 header->h_sequence = cpu_to_be32(transaction->t_tid);
849 set_buffer_uptodate(bh);
851 BUFFER_TRACE(bh, "return this buffer");
855 void jbd2_descriptor_block_csum_set(journal_t *j, struct buffer_head *bh)
857 struct jbd2_journal_block_tail *tail;
860 if (!jbd2_journal_has_csum_v2or3(j))
863 tail = (struct jbd2_journal_block_tail *)(bh->b_data + j->j_blocksize -
864 sizeof(struct jbd2_journal_block_tail));
865 tail->t_checksum = 0;
866 csum = jbd2_chksum(j, j->j_csum_seed, bh->b_data, j->j_blocksize);
867 tail->t_checksum = cpu_to_be32(csum);
871 * Return tid of the oldest transaction in the journal and block in the journal
872 * where the transaction starts.
874 * If the journal is now empty, return which will be the next transaction ID
875 * we will write and where will that transaction start.
877 * The return value is 0 if journal tail cannot be pushed any further, 1 if
880 int jbd2_journal_get_log_tail(journal_t *journal, tid_t *tid,
881 unsigned long *block)
883 transaction_t *transaction;
886 read_lock(&journal->j_state_lock);
887 spin_lock(&journal->j_list_lock);
888 transaction = journal->j_checkpoint_transactions;
890 *tid = transaction->t_tid;
891 *block = transaction->t_log_start;
892 } else if ((transaction = journal->j_committing_transaction) != NULL) {
893 *tid = transaction->t_tid;
894 *block = transaction->t_log_start;
895 } else if ((transaction = journal->j_running_transaction) != NULL) {
896 *tid = transaction->t_tid;
897 *block = journal->j_head;
899 *tid = journal->j_transaction_sequence;
900 *block = journal->j_head;
902 ret = tid_gt(*tid, journal->j_tail_sequence);
903 spin_unlock(&journal->j_list_lock);
904 read_unlock(&journal->j_state_lock);
910 * Update information in journal structure and in on disk journal superblock
911 * about log tail. This function does not check whether information passed in
912 * really pushes log tail further. It's responsibility of the caller to make
913 * sure provided log tail information is valid (e.g. by holding
914 * j_checkpoint_mutex all the time between computing log tail and calling this
915 * function as is the case with jbd2_cleanup_journal_tail()).
917 * Requires j_checkpoint_mutex
919 int __jbd2_update_log_tail(journal_t *journal, tid_t tid, unsigned long block)
924 BUG_ON(!mutex_is_locked(&journal->j_checkpoint_mutex));
927 * We cannot afford for write to remain in drive's caches since as
928 * soon as we update j_tail, next transaction can start reusing journal
929 * space and if we lose sb update during power failure we'd replay
930 * old transaction with possibly newly overwritten data.
932 ret = jbd2_journal_update_sb_log_tail(journal, tid, block,
937 write_lock(&journal->j_state_lock);
938 freed = block - journal->j_tail;
939 if (block < journal->j_tail)
940 freed += journal->j_last - journal->j_first;
942 trace_jbd2_update_log_tail(journal, tid, block, freed);
944 "Cleaning journal tail from %u to %u (offset %lu), "
946 journal->j_tail_sequence, tid, block, freed);
948 journal->j_free += freed;
949 journal->j_tail_sequence = tid;
950 journal->j_tail = block;
951 write_unlock(&journal->j_state_lock);
958 * This is a variation of __jbd2_update_log_tail which checks for validity of
959 * provided log tail and locks j_checkpoint_mutex. So it is safe against races
960 * with other threads updating log tail.
962 void jbd2_update_log_tail(journal_t *journal, tid_t tid, unsigned long block)
964 mutex_lock_io(&journal->j_checkpoint_mutex);
965 if (tid_gt(tid, journal->j_tail_sequence))
966 __jbd2_update_log_tail(journal, tid, block);
967 mutex_unlock(&journal->j_checkpoint_mutex);
970 struct jbd2_stats_proc_session {
972 struct transaction_stats_s *stats;
977 static void *jbd2_seq_info_start(struct seq_file *seq, loff_t *pos)
979 return *pos ? NULL : SEQ_START_TOKEN;
982 static void *jbd2_seq_info_next(struct seq_file *seq, void *v, loff_t *pos)
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 proc_ops jbd2_info_proc_ops = {
1078 .proc_open = jbd2_seq_info_open,
1079 .proc_read = seq_read,
1080 .proc_lseek = seq_lseek,
1081 .proc_release = jbd2_seq_info_release,
1084 static struct proc_dir_entry *proc_jbd2_stats;
1086 static void jbd2_stats_proc_init(journal_t *journal)
1088 journal->j_proc_entry = proc_mkdir(journal->j_devname, proc_jbd2_stats);
1089 if (journal->j_proc_entry) {
1090 proc_create_data("info", S_IRUGO, journal->j_proc_entry,
1091 &jbd2_info_proc_ops, journal);
1095 static void jbd2_stats_proc_exit(journal_t *journal)
1097 remove_proc_entry("info", journal->j_proc_entry);
1098 remove_proc_entry(journal->j_devname, proc_jbd2_stats);
1101 /* Minimum size of descriptor tag */
1102 static int jbd2_min_tag_size(void)
1105 * Tag with 32-bit block numbers does not use last four bytes of the
1108 return sizeof(journal_block_tag_t) - 4;
1112 * Management for journal control blocks: functions to create and
1113 * destroy journal_t structures, and to initialise and read existing
1114 * journal blocks from disk. */
1116 /* First: create and setup a journal_t object in memory. We initialise
1117 * very few fields yet: that has to wait until we have created the
1118 * journal structures from from scratch, or loaded them from disk. */
1120 static journal_t *journal_init_common(struct block_device *bdev,
1121 struct block_device *fs_dev,
1122 unsigned long long start, int len, int blocksize)
1124 static struct lock_class_key jbd2_trans_commit_key;
1127 struct buffer_head *bh;
1130 journal = kzalloc(sizeof(*journal), GFP_KERNEL);
1134 init_waitqueue_head(&journal->j_wait_transaction_locked);
1135 init_waitqueue_head(&journal->j_wait_done_commit);
1136 init_waitqueue_head(&journal->j_wait_commit);
1137 init_waitqueue_head(&journal->j_wait_updates);
1138 init_waitqueue_head(&journal->j_wait_reserved);
1139 mutex_init(&journal->j_barrier);
1140 mutex_init(&journal->j_checkpoint_mutex);
1141 spin_lock_init(&journal->j_revoke_lock);
1142 spin_lock_init(&journal->j_list_lock);
1143 rwlock_init(&journal->j_state_lock);
1145 journal->j_commit_interval = (HZ * JBD2_DEFAULT_MAX_COMMIT_AGE);
1146 journal->j_min_batch_time = 0;
1147 journal->j_max_batch_time = 15000; /* 15ms */
1148 atomic_set(&journal->j_reserved_credits, 0);
1150 /* The journal is marked for error until we succeed with recovery! */
1151 journal->j_flags = JBD2_ABORT;
1153 /* Set up a default-sized revoke table for the new mount. */
1154 err = jbd2_journal_init_revoke(journal, JOURNAL_REVOKE_DEFAULT_HASH);
1158 spin_lock_init(&journal->j_history_lock);
1160 lockdep_init_map(&journal->j_trans_commit_map, "jbd2_handle",
1161 &jbd2_trans_commit_key, 0);
1163 /* journal descriptor can store up to n blocks -bzzz */
1164 journal->j_blocksize = blocksize;
1165 journal->j_dev = bdev;
1166 journal->j_fs_dev = fs_dev;
1167 journal->j_blk_offset = start;
1168 journal->j_maxlen = len;
1169 /* We need enough buffers to write out full descriptor block. */
1170 n = journal->j_blocksize / jbd2_min_tag_size();
1171 journal->j_wbufsize = n;
1172 journal->j_wbuf = kmalloc_array(n, sizeof(struct buffer_head *),
1174 if (!journal->j_wbuf)
1177 bh = getblk_unmovable(journal->j_dev, start, journal->j_blocksize);
1179 pr_err("%s: Cannot get buffer for journal superblock\n",
1183 journal->j_sb_buffer = bh;
1184 journal->j_superblock = (journal_superblock_t *)bh->b_data;
1189 kfree(journal->j_wbuf);
1190 jbd2_journal_destroy_revoke(journal);
1195 /* jbd2_journal_init_dev and jbd2_journal_init_inode:
1197 * Create a journal structure assigned some fixed set of disk blocks to
1198 * the journal. We don't actually touch those disk blocks yet, but we
1199 * need to set up all of the mapping information to tell the journaling
1200 * system where the journal blocks are.
1205 * journal_t * jbd2_journal_init_dev() - creates and initialises a journal structure
1206 * @bdev: Block device on which to create the journal
1207 * @fs_dev: Device which hold journalled filesystem for this journal.
1208 * @start: Block nr Start of journal.
1209 * @len: Length of the journal in blocks.
1210 * @blocksize: blocksize of journalling device
1212 * Returns: a newly created journal_t *
1214 * jbd2_journal_init_dev creates a journal which maps a fixed contiguous
1215 * range of blocks on an arbitrary block device.
1218 journal_t *jbd2_journal_init_dev(struct block_device *bdev,
1219 struct block_device *fs_dev,
1220 unsigned long long start, int len, int blocksize)
1224 journal = journal_init_common(bdev, fs_dev, start, len, blocksize);
1228 bdevname(journal->j_dev, journal->j_devname);
1229 strreplace(journal->j_devname, '/', '!');
1230 jbd2_stats_proc_init(journal);
1236 * journal_t * jbd2_journal_init_inode () - creates a journal which maps to a inode.
1237 * @inode: An inode to create the journal in
1239 * jbd2_journal_init_inode creates a journal which maps an on-disk inode as
1240 * the journal. The inode must exist already, must support bmap() and
1241 * must have all data blocks preallocated.
1243 journal_t *jbd2_journal_init_inode(struct inode *inode)
1247 unsigned long long blocknr;
1249 blocknr = bmap(inode, 0);
1251 pr_err("%s: Cannot locate journal superblock\n",
1256 jbd_debug(1, "JBD2: inode %s/%ld, size %lld, bits %d, blksize %ld\n",
1257 inode->i_sb->s_id, inode->i_ino, (long long) inode->i_size,
1258 inode->i_sb->s_blocksize_bits, inode->i_sb->s_blocksize);
1260 journal = journal_init_common(inode->i_sb->s_bdev, inode->i_sb->s_bdev,
1261 blocknr, inode->i_size >> inode->i_sb->s_blocksize_bits,
1262 inode->i_sb->s_blocksize);
1266 journal->j_inode = inode;
1267 bdevname(journal->j_dev, journal->j_devname);
1268 p = strreplace(journal->j_devname, '/', '!');
1269 sprintf(p, "-%lu", journal->j_inode->i_ino);
1270 jbd2_stats_proc_init(journal);
1276 * If the journal init or create aborts, we need to mark the journal
1277 * superblock as being NULL to prevent the journal destroy from writing
1278 * back a bogus superblock.
1280 static void journal_fail_superblock (journal_t *journal)
1282 struct buffer_head *bh = journal->j_sb_buffer;
1284 journal->j_sb_buffer = NULL;
1288 * Given a journal_t structure, initialise the various fields for
1289 * startup of a new journaling session. We use this both when creating
1290 * a journal, and after recovering an old journal to reset it for
1294 static int journal_reset(journal_t *journal)
1296 journal_superblock_t *sb = journal->j_superblock;
1297 unsigned long long first, last;
1299 first = be32_to_cpu(sb->s_first);
1300 last = be32_to_cpu(sb->s_maxlen);
1301 if (first + JBD2_MIN_JOURNAL_BLOCKS > last + 1) {
1302 printk(KERN_ERR "JBD2: Journal too short (blocks %llu-%llu).\n",
1304 journal_fail_superblock(journal);
1308 journal->j_first = first;
1309 journal->j_last = last;
1311 journal->j_head = first;
1312 journal->j_tail = first;
1313 journal->j_free = last - first;
1315 journal->j_tail_sequence = journal->j_transaction_sequence;
1316 journal->j_commit_sequence = journal->j_transaction_sequence - 1;
1317 journal->j_commit_request = journal->j_commit_sequence;
1319 journal->j_max_transaction_buffers = journal->j_maxlen / 4;
1322 * As a special case, if the on-disk copy is already marked as needing
1323 * no recovery (s_start == 0), then we can safely defer the superblock
1324 * update until the next commit by setting JBD2_FLUSHED. This avoids
1325 * attempting a write to a potential-readonly device.
1327 if (sb->s_start == 0) {
1328 jbd_debug(1, "JBD2: Skipping superblock update on recovered sb "
1329 "(start %ld, seq %u, errno %d)\n",
1330 journal->j_tail, journal->j_tail_sequence,
1332 journal->j_flags |= JBD2_FLUSHED;
1334 /* Lock here to make assertions happy... */
1335 mutex_lock_io(&journal->j_checkpoint_mutex);
1337 * Update log tail information. We use REQ_FUA since new
1338 * transaction will start reusing journal space and so we
1339 * must make sure information about current log tail is on
1342 jbd2_journal_update_sb_log_tail(journal,
1343 journal->j_tail_sequence,
1345 REQ_SYNC | REQ_FUA);
1346 mutex_unlock(&journal->j_checkpoint_mutex);
1348 return jbd2_journal_start_thread(journal);
1352 * This function expects that the caller will have locked the journal
1353 * buffer head, and will return with it unlocked
1355 static int jbd2_write_superblock(journal_t *journal, int write_flags)
1357 struct buffer_head *bh = journal->j_sb_buffer;
1358 journal_superblock_t *sb = journal->j_superblock;
1361 /* Buffer got discarded which means block device got invalidated */
1362 if (!buffer_mapped(bh))
1365 trace_jbd2_write_superblock(journal, write_flags);
1366 if (!(journal->j_flags & JBD2_BARRIER))
1367 write_flags &= ~(REQ_FUA | REQ_PREFLUSH);
1368 if (buffer_write_io_error(bh)) {
1370 * Oh, dear. A previous attempt to write the journal
1371 * superblock failed. This could happen because the
1372 * USB device was yanked out. Or it could happen to
1373 * be a transient write error and maybe the block will
1374 * be remapped. Nothing we can do but to retry the
1375 * write and hope for the best.
1377 printk(KERN_ERR "JBD2: previous I/O error detected "
1378 "for journal superblock update for %s.\n",
1379 journal->j_devname);
1380 clear_buffer_write_io_error(bh);
1381 set_buffer_uptodate(bh);
1383 if (jbd2_journal_has_csum_v2or3(journal))
1384 sb->s_checksum = jbd2_superblock_csum(journal, sb);
1386 bh->b_end_io = end_buffer_write_sync;
1387 ret = submit_bh(REQ_OP_WRITE, write_flags, bh);
1389 if (buffer_write_io_error(bh)) {
1390 clear_buffer_write_io_error(bh);
1391 set_buffer_uptodate(bh);
1395 printk(KERN_ERR "JBD2: Error %d detected when updating "
1396 "journal superblock for %s.\n", ret,
1397 journal->j_devname);
1398 jbd2_journal_abort(journal, ret);
1405 * jbd2_journal_update_sb_log_tail() - Update log tail in journal sb on disk.
1406 * @journal: The journal to update.
1407 * @tail_tid: TID of the new transaction at the tail of the log
1408 * @tail_block: The first block of the transaction at the tail of the log
1409 * @write_op: With which operation should we write the journal sb
1411 * Update a journal's superblock information about log tail and write it to
1412 * disk, waiting for the IO to complete.
1414 int jbd2_journal_update_sb_log_tail(journal_t *journal, tid_t tail_tid,
1415 unsigned long tail_block, int write_op)
1417 journal_superblock_t *sb = journal->j_superblock;
1420 if (is_journal_aborted(journal))
1423 BUG_ON(!mutex_is_locked(&journal->j_checkpoint_mutex));
1424 jbd_debug(1, "JBD2: updating superblock (start %lu, seq %u)\n",
1425 tail_block, tail_tid);
1427 lock_buffer(journal->j_sb_buffer);
1428 sb->s_sequence = cpu_to_be32(tail_tid);
1429 sb->s_start = cpu_to_be32(tail_block);
1431 ret = jbd2_write_superblock(journal, write_op);
1435 /* Log is no longer empty */
1436 write_lock(&journal->j_state_lock);
1437 WARN_ON(!sb->s_sequence);
1438 journal->j_flags &= ~JBD2_FLUSHED;
1439 write_unlock(&journal->j_state_lock);
1446 * jbd2_mark_journal_empty() - Mark on disk journal as empty.
1447 * @journal: The journal to update.
1448 * @write_op: With which operation should we write the journal sb
1450 * Update a journal's dynamic superblock fields to show that journal is empty.
1451 * Write updated superblock to disk waiting for IO to complete.
1453 static void jbd2_mark_journal_empty(journal_t *journal, int write_op)
1455 journal_superblock_t *sb = journal->j_superblock;
1457 BUG_ON(!mutex_is_locked(&journal->j_checkpoint_mutex));
1458 lock_buffer(journal->j_sb_buffer);
1459 if (sb->s_start == 0) { /* Is it already empty? */
1460 unlock_buffer(journal->j_sb_buffer);
1464 jbd_debug(1, "JBD2: Marking journal as empty (seq %u)\n",
1465 journal->j_tail_sequence);
1467 sb->s_sequence = cpu_to_be32(journal->j_tail_sequence);
1468 sb->s_start = cpu_to_be32(0);
1470 jbd2_write_superblock(journal, write_op);
1472 /* Log is no longer empty */
1473 write_lock(&journal->j_state_lock);
1474 journal->j_flags |= JBD2_FLUSHED;
1475 write_unlock(&journal->j_state_lock);
1480 * jbd2_journal_update_sb_errno() - Update error in the journal.
1481 * @journal: The journal to update.
1483 * Update a journal's errno. Write updated superblock to disk waiting for IO
1486 void jbd2_journal_update_sb_errno(journal_t *journal)
1488 journal_superblock_t *sb = journal->j_superblock;
1491 lock_buffer(journal->j_sb_buffer);
1492 errcode = journal->j_errno;
1493 if (errcode == -ESHUTDOWN)
1495 jbd_debug(1, "JBD2: updating superblock error (errno %d)\n", errcode);
1496 sb->s_errno = cpu_to_be32(errcode);
1498 jbd2_write_superblock(journal, REQ_SYNC | REQ_FUA);
1500 EXPORT_SYMBOL(jbd2_journal_update_sb_errno);
1502 static int journal_revoke_records_per_block(journal_t *journal)
1505 int space = journal->j_blocksize - sizeof(jbd2_journal_revoke_header_t);
1507 if (jbd2_has_feature_64bit(journal))
1512 if (jbd2_journal_has_csum_v2or3(journal))
1513 space -= sizeof(struct jbd2_journal_block_tail);
1514 return space / record_size;
1518 * Read the superblock for a given journal, performing initial
1519 * validation of the format.
1521 static int journal_get_superblock(journal_t *journal)
1523 struct buffer_head *bh;
1524 journal_superblock_t *sb;
1527 bh = journal->j_sb_buffer;
1529 J_ASSERT(bh != NULL);
1530 if (!buffer_uptodate(bh)) {
1531 ll_rw_block(REQ_OP_READ, 0, 1, &bh);
1533 if (!buffer_uptodate(bh)) {
1535 "JBD2: IO error reading journal superblock\n");
1540 if (buffer_verified(bh))
1543 sb = journal->j_superblock;
1547 if (sb->s_header.h_magic != cpu_to_be32(JBD2_MAGIC_NUMBER) ||
1548 sb->s_blocksize != cpu_to_be32(journal->j_blocksize)) {
1549 printk(KERN_WARNING "JBD2: no valid journal superblock found\n");
1553 switch(be32_to_cpu(sb->s_header.h_blocktype)) {
1554 case JBD2_SUPERBLOCK_V1:
1555 journal->j_format_version = 1;
1557 case JBD2_SUPERBLOCK_V2:
1558 journal->j_format_version = 2;
1561 printk(KERN_WARNING "JBD2: unrecognised superblock format ID\n");
1565 if (be32_to_cpu(sb->s_maxlen) < journal->j_maxlen)
1566 journal->j_maxlen = be32_to_cpu(sb->s_maxlen);
1567 else if (be32_to_cpu(sb->s_maxlen) > journal->j_maxlen) {
1568 printk(KERN_WARNING "JBD2: journal file too short\n");
1572 if (be32_to_cpu(sb->s_first) == 0 ||
1573 be32_to_cpu(sb->s_first) >= journal->j_maxlen) {
1575 "JBD2: Invalid start block of journal: %u\n",
1576 be32_to_cpu(sb->s_first));
1580 if (jbd2_has_feature_csum2(journal) &&
1581 jbd2_has_feature_csum3(journal)) {
1582 /* Can't have checksum v2 and v3 at the same time! */
1583 printk(KERN_ERR "JBD2: Can't enable checksumming v2 and v3 "
1584 "at the same time!\n");
1588 if (jbd2_journal_has_csum_v2or3_feature(journal) &&
1589 jbd2_has_feature_checksum(journal)) {
1590 /* Can't have checksum v1 and v2 on at the same time! */
1591 printk(KERN_ERR "JBD2: Can't enable checksumming v1 and v2/3 "
1592 "at the same time!\n");
1596 if (!jbd2_verify_csum_type(journal, sb)) {
1597 printk(KERN_ERR "JBD2: Unknown checksum type\n");
1601 /* Load the checksum driver */
1602 if (jbd2_journal_has_csum_v2or3_feature(journal)) {
1603 journal->j_chksum_driver = crypto_alloc_shash("crc32c", 0, 0);
1604 if (IS_ERR(journal->j_chksum_driver)) {
1605 printk(KERN_ERR "JBD2: Cannot load crc32c driver.\n");
1606 err = PTR_ERR(journal->j_chksum_driver);
1607 journal->j_chksum_driver = NULL;
1612 if (jbd2_journal_has_csum_v2or3(journal)) {
1613 /* Check superblock checksum */
1614 if (sb->s_checksum != jbd2_superblock_csum(journal, sb)) {
1615 printk(KERN_ERR "JBD2: journal checksum error\n");
1620 /* Precompute checksum seed for all metadata */
1621 journal->j_csum_seed = jbd2_chksum(journal, ~0, sb->s_uuid,
1622 sizeof(sb->s_uuid));
1625 journal->j_revoke_records_per_block =
1626 journal_revoke_records_per_block(journal);
1627 set_buffer_verified(bh);
1632 journal_fail_superblock(journal);
1637 * Load the on-disk journal superblock and read the key fields into the
1641 static int load_superblock(journal_t *journal)
1644 journal_superblock_t *sb;
1646 err = journal_get_superblock(journal);
1650 sb = journal->j_superblock;
1652 journal->j_tail_sequence = be32_to_cpu(sb->s_sequence);
1653 journal->j_tail = be32_to_cpu(sb->s_start);
1654 journal->j_first = be32_to_cpu(sb->s_first);
1655 journal->j_last = be32_to_cpu(sb->s_maxlen);
1656 journal->j_errno = be32_to_cpu(sb->s_errno);
1663 * int jbd2_journal_load() - Read journal from disk.
1664 * @journal: Journal to act on.
1666 * Given a journal_t structure which tells us which disk blocks contain
1667 * a journal, read the journal from disk to initialise the in-memory
1670 int jbd2_journal_load(journal_t *journal)
1673 journal_superblock_t *sb;
1675 err = load_superblock(journal);
1679 sb = journal->j_superblock;
1680 /* If this is a V2 superblock, then we have to check the
1681 * features flags on it. */
1683 if (journal->j_format_version >= 2) {
1684 if ((sb->s_feature_ro_compat &
1685 ~cpu_to_be32(JBD2_KNOWN_ROCOMPAT_FEATURES)) ||
1686 (sb->s_feature_incompat &
1687 ~cpu_to_be32(JBD2_KNOWN_INCOMPAT_FEATURES))) {
1689 "JBD2: Unrecognised features on journal\n");
1695 * Create a slab for this blocksize
1697 err = jbd2_journal_create_slab(be32_to_cpu(sb->s_blocksize));
1701 /* Let the recovery code check whether it needs to recover any
1702 * data from the journal. */
1703 if (jbd2_journal_recover(journal))
1704 goto recovery_error;
1706 if (journal->j_failed_commit) {
1707 printk(KERN_ERR "JBD2: journal transaction %u on %s "
1708 "is corrupt.\n", journal->j_failed_commit,
1709 journal->j_devname);
1710 return -EFSCORRUPTED;
1713 * clear JBD2_ABORT flag initialized in journal_init_common
1714 * here to update log tail information with the newest seq.
1716 journal->j_flags &= ~JBD2_ABORT;
1718 /* OK, we've finished with the dynamic journal bits:
1719 * reinitialise the dynamic contents of the superblock in memory
1720 * and reset them on disk. */
1721 if (journal_reset(journal))
1722 goto recovery_error;
1724 journal->j_flags |= JBD2_LOADED;
1728 printk(KERN_WARNING "JBD2: recovery failed\n");
1733 * void jbd2_journal_destroy() - Release a journal_t structure.
1734 * @journal: Journal to act on.
1736 * Release a journal_t structure once it is no longer in use by the
1738 * Return <0 if we couldn't clean up the journal.
1740 int jbd2_journal_destroy(journal_t *journal)
1744 /* Wait for the commit thread to wake up and die. */
1745 journal_kill_thread(journal);
1747 /* Force a final log commit */
1748 if (journal->j_running_transaction)
1749 jbd2_journal_commit_transaction(journal);
1751 /* Force any old transactions to disk */
1753 /* Totally anal locking here... */
1754 spin_lock(&journal->j_list_lock);
1755 while (journal->j_checkpoint_transactions != NULL) {
1756 spin_unlock(&journal->j_list_lock);
1757 mutex_lock_io(&journal->j_checkpoint_mutex);
1758 err = jbd2_log_do_checkpoint(journal);
1759 mutex_unlock(&journal->j_checkpoint_mutex);
1761 * If checkpointing failed, just free the buffers to avoid
1765 jbd2_journal_destroy_checkpoint(journal);
1766 spin_lock(&journal->j_list_lock);
1769 spin_lock(&journal->j_list_lock);
1772 J_ASSERT(journal->j_running_transaction == NULL);
1773 J_ASSERT(journal->j_committing_transaction == NULL);
1774 J_ASSERT(journal->j_checkpoint_transactions == NULL);
1775 spin_unlock(&journal->j_list_lock);
1777 if (journal->j_sb_buffer) {
1778 if (!is_journal_aborted(journal)) {
1779 mutex_lock_io(&journal->j_checkpoint_mutex);
1781 write_lock(&journal->j_state_lock);
1782 journal->j_tail_sequence =
1783 ++journal->j_transaction_sequence;
1784 write_unlock(&journal->j_state_lock);
1786 jbd2_mark_journal_empty(journal,
1787 REQ_SYNC | REQ_PREFLUSH | REQ_FUA);
1788 mutex_unlock(&journal->j_checkpoint_mutex);
1791 brelse(journal->j_sb_buffer);
1794 if (journal->j_proc_entry)
1795 jbd2_stats_proc_exit(journal);
1796 iput(journal->j_inode);
1797 if (journal->j_revoke)
1798 jbd2_journal_destroy_revoke(journal);
1799 if (journal->j_chksum_driver)
1800 crypto_free_shash(journal->j_chksum_driver);
1801 kfree(journal->j_wbuf);
1809 *int jbd2_journal_check_used_features () - Check if features specified are used.
1810 * @journal: Journal to check.
1811 * @compat: bitmask of compatible features
1812 * @ro: bitmask of features that force read-only mount
1813 * @incompat: bitmask of incompatible features
1815 * Check whether the journal uses all of a given set of
1816 * features. Return true (non-zero) if it does.
1819 int jbd2_journal_check_used_features (journal_t *journal, unsigned long compat,
1820 unsigned long ro, unsigned long incompat)
1822 journal_superblock_t *sb;
1824 if (!compat && !ro && !incompat)
1826 /* Load journal superblock if it is not loaded yet. */
1827 if (journal->j_format_version == 0 &&
1828 journal_get_superblock(journal) != 0)
1830 if (journal->j_format_version == 1)
1833 sb = journal->j_superblock;
1835 if (((be32_to_cpu(sb->s_feature_compat) & compat) == compat) &&
1836 ((be32_to_cpu(sb->s_feature_ro_compat) & ro) == ro) &&
1837 ((be32_to_cpu(sb->s_feature_incompat) & incompat) == incompat))
1844 * int jbd2_journal_check_available_features() - Check feature set in journalling layer
1845 * @journal: Journal to check.
1846 * @compat: bitmask of compatible features
1847 * @ro: bitmask of features that force read-only mount
1848 * @incompat: bitmask of incompatible features
1850 * Check whether the journaling code supports the use of
1851 * all of a given set of features on this journal. Return true
1852 * (non-zero) if it can. */
1854 int jbd2_journal_check_available_features (journal_t *journal, unsigned long compat,
1855 unsigned long ro, unsigned long incompat)
1857 if (!compat && !ro && !incompat)
1860 /* We can support any known requested features iff the
1861 * superblock is in version 2. Otherwise we fail to support any
1862 * extended sb features. */
1864 if (journal->j_format_version != 2)
1867 if ((compat & JBD2_KNOWN_COMPAT_FEATURES) == compat &&
1868 (ro & JBD2_KNOWN_ROCOMPAT_FEATURES) == ro &&
1869 (incompat & JBD2_KNOWN_INCOMPAT_FEATURES) == incompat)
1876 * int jbd2_journal_set_features () - Mark a given journal feature in the superblock
1877 * @journal: Journal to act on.
1878 * @compat: bitmask of compatible features
1879 * @ro: bitmask of features that force read-only mount
1880 * @incompat: bitmask of incompatible features
1882 * Mark a given journal feature as present on the
1883 * superblock. Returns true if the requested features could be set.
1887 int jbd2_journal_set_features (journal_t *journal, unsigned long compat,
1888 unsigned long ro, unsigned long incompat)
1890 #define INCOMPAT_FEATURE_ON(f) \
1891 ((incompat & (f)) && !(sb->s_feature_incompat & cpu_to_be32(f)))
1892 #define COMPAT_FEATURE_ON(f) \
1893 ((compat & (f)) && !(sb->s_feature_compat & cpu_to_be32(f)))
1894 journal_superblock_t *sb;
1896 if (jbd2_journal_check_used_features(journal, compat, ro, incompat))
1899 if (!jbd2_journal_check_available_features(journal, compat, ro, incompat))
1902 /* If enabling v2 checksums, turn on v3 instead */
1903 if (incompat & JBD2_FEATURE_INCOMPAT_CSUM_V2) {
1904 incompat &= ~JBD2_FEATURE_INCOMPAT_CSUM_V2;
1905 incompat |= JBD2_FEATURE_INCOMPAT_CSUM_V3;
1908 /* Asking for checksumming v3 and v1? Only give them v3. */
1909 if (incompat & JBD2_FEATURE_INCOMPAT_CSUM_V3 &&
1910 compat & JBD2_FEATURE_COMPAT_CHECKSUM)
1911 compat &= ~JBD2_FEATURE_COMPAT_CHECKSUM;
1913 jbd_debug(1, "Setting new features 0x%lx/0x%lx/0x%lx\n",
1914 compat, ro, incompat);
1916 sb = journal->j_superblock;
1918 /* Load the checksum driver if necessary */
1919 if ((journal->j_chksum_driver == NULL) &&
1920 INCOMPAT_FEATURE_ON(JBD2_FEATURE_INCOMPAT_CSUM_V3)) {
1921 journal->j_chksum_driver = crypto_alloc_shash("crc32c", 0, 0);
1922 if (IS_ERR(journal->j_chksum_driver)) {
1923 printk(KERN_ERR "JBD2: Cannot load crc32c driver.\n");
1924 journal->j_chksum_driver = NULL;
1927 /* Precompute checksum seed for all metadata */
1928 journal->j_csum_seed = jbd2_chksum(journal, ~0, sb->s_uuid,
1929 sizeof(sb->s_uuid));
1932 lock_buffer(journal->j_sb_buffer);
1934 /* If enabling v3 checksums, update superblock */
1935 if (INCOMPAT_FEATURE_ON(JBD2_FEATURE_INCOMPAT_CSUM_V3)) {
1936 sb->s_checksum_type = JBD2_CRC32C_CHKSUM;
1937 sb->s_feature_compat &=
1938 ~cpu_to_be32(JBD2_FEATURE_COMPAT_CHECKSUM);
1941 /* If enabling v1 checksums, downgrade superblock */
1942 if (COMPAT_FEATURE_ON(JBD2_FEATURE_COMPAT_CHECKSUM))
1943 sb->s_feature_incompat &=
1944 ~cpu_to_be32(JBD2_FEATURE_INCOMPAT_CSUM_V2 |
1945 JBD2_FEATURE_INCOMPAT_CSUM_V3);
1947 sb->s_feature_compat |= cpu_to_be32(compat);
1948 sb->s_feature_ro_compat |= cpu_to_be32(ro);
1949 sb->s_feature_incompat |= cpu_to_be32(incompat);
1950 unlock_buffer(journal->j_sb_buffer);
1951 journal->j_revoke_records_per_block =
1952 journal_revoke_records_per_block(journal);
1955 #undef COMPAT_FEATURE_ON
1956 #undef INCOMPAT_FEATURE_ON
1960 * jbd2_journal_clear_features () - Clear a given journal feature in the
1962 * @journal: Journal to act on.
1963 * @compat: bitmask of compatible features
1964 * @ro: bitmask of features that force read-only mount
1965 * @incompat: bitmask of incompatible features
1967 * Clear a given journal feature as present on the
1970 void jbd2_journal_clear_features(journal_t *journal, unsigned long compat,
1971 unsigned long ro, unsigned long incompat)
1973 journal_superblock_t *sb;
1975 jbd_debug(1, "Clear features 0x%lx/0x%lx/0x%lx\n",
1976 compat, ro, incompat);
1978 sb = journal->j_superblock;
1980 sb->s_feature_compat &= ~cpu_to_be32(compat);
1981 sb->s_feature_ro_compat &= ~cpu_to_be32(ro);
1982 sb->s_feature_incompat &= ~cpu_to_be32(incompat);
1983 journal->j_revoke_records_per_block =
1984 journal_revoke_records_per_block(journal);
1986 EXPORT_SYMBOL(jbd2_journal_clear_features);
1989 * int jbd2_journal_flush () - Flush journal
1990 * @journal: Journal to act on.
1992 * Flush all data for a given journal to disk and empty the journal.
1993 * Filesystems can use this when remounting readonly to ensure that
1994 * recovery does not need to happen on remount.
1997 int jbd2_journal_flush(journal_t *journal)
2000 transaction_t *transaction = NULL;
2002 write_lock(&journal->j_state_lock);
2004 /* Force everything buffered to the log... */
2005 if (journal->j_running_transaction) {
2006 transaction = journal->j_running_transaction;
2007 __jbd2_log_start_commit(journal, transaction->t_tid);
2008 } else if (journal->j_committing_transaction)
2009 transaction = journal->j_committing_transaction;
2011 /* Wait for the log commit to complete... */
2013 tid_t tid = transaction->t_tid;
2015 write_unlock(&journal->j_state_lock);
2016 jbd2_log_wait_commit(journal, tid);
2018 write_unlock(&journal->j_state_lock);
2021 /* ...and flush everything in the log out to disk. */
2022 spin_lock(&journal->j_list_lock);
2023 while (!err && journal->j_checkpoint_transactions != NULL) {
2024 spin_unlock(&journal->j_list_lock);
2025 mutex_lock_io(&journal->j_checkpoint_mutex);
2026 err = jbd2_log_do_checkpoint(journal);
2027 mutex_unlock(&journal->j_checkpoint_mutex);
2028 spin_lock(&journal->j_list_lock);
2030 spin_unlock(&journal->j_list_lock);
2032 if (is_journal_aborted(journal))
2035 mutex_lock_io(&journal->j_checkpoint_mutex);
2037 err = jbd2_cleanup_journal_tail(journal);
2039 mutex_unlock(&journal->j_checkpoint_mutex);
2045 /* Finally, mark the journal as really needing no recovery.
2046 * This sets s_start==0 in the underlying superblock, which is
2047 * the magic code for a fully-recovered superblock. Any future
2048 * commits of data to the journal will restore the current
2050 jbd2_mark_journal_empty(journal, REQ_SYNC | REQ_FUA);
2051 mutex_unlock(&journal->j_checkpoint_mutex);
2052 write_lock(&journal->j_state_lock);
2053 J_ASSERT(!journal->j_running_transaction);
2054 J_ASSERT(!journal->j_committing_transaction);
2055 J_ASSERT(!journal->j_checkpoint_transactions);
2056 J_ASSERT(journal->j_head == journal->j_tail);
2057 J_ASSERT(journal->j_tail_sequence == journal->j_transaction_sequence);
2058 write_unlock(&journal->j_state_lock);
2064 * int jbd2_journal_wipe() - Wipe journal contents
2065 * @journal: Journal to act on.
2066 * @write: flag (see below)
2068 * Wipe out all of the contents of a journal, safely. This will produce
2069 * a warning if the journal contains any valid recovery information.
2070 * Must be called between journal_init_*() and jbd2_journal_load().
2072 * If 'write' is non-zero, then we wipe out the journal on disk; otherwise
2073 * we merely suppress recovery.
2076 int jbd2_journal_wipe(journal_t *journal, int write)
2080 J_ASSERT (!(journal->j_flags & JBD2_LOADED));
2082 err = load_superblock(journal);
2086 if (!journal->j_tail)
2089 printk(KERN_WARNING "JBD2: %s recovery information on journal\n",
2090 write ? "Clearing" : "Ignoring");
2092 err = jbd2_journal_skip_recovery(journal);
2094 /* Lock to make assertions happy... */
2095 mutex_lock_io(&journal->j_checkpoint_mutex);
2096 jbd2_mark_journal_empty(journal, REQ_SYNC | REQ_FUA);
2097 mutex_unlock(&journal->j_checkpoint_mutex);
2105 * void jbd2_journal_abort () - Shutdown the journal immediately.
2106 * @journal: the journal to shutdown.
2107 * @errno: an error number to record in the journal indicating
2108 * the reason for the shutdown.
2110 * Perform a complete, immediate shutdown of the ENTIRE
2111 * journal (not of a single transaction). This operation cannot be
2112 * undone without closing and reopening the journal.
2114 * The jbd2_journal_abort function is intended to support higher level error
2115 * recovery mechanisms such as the ext2/ext3 remount-readonly error
2118 * Journal abort has very specific semantics. Any existing dirty,
2119 * unjournaled buffers in the main filesystem will still be written to
2120 * disk by bdflush, but the journaling mechanism will be suspended
2121 * immediately and no further transaction commits will be honoured.
2123 * Any dirty, journaled buffers will be written back to disk without
2124 * hitting the journal. Atomicity cannot be guaranteed on an aborted
2125 * filesystem, but we _do_ attempt to leave as much data as possible
2126 * behind for fsck to use for cleanup.
2128 * Any attempt to get a new transaction handle on a journal which is in
2129 * ABORT state will just result in an -EROFS error return. A
2130 * jbd2_journal_stop on an existing handle will return -EIO if we have
2131 * entered abort state during the update.
2133 * Recursive transactions are not disturbed by journal abort until the
2134 * final jbd2_journal_stop, which will receive the -EIO error.
2136 * Finally, the jbd2_journal_abort call allows the caller to supply an errno
2137 * which will be recorded (if possible) in the journal superblock. This
2138 * allows a client to record failure conditions in the middle of a
2139 * transaction without having to complete the transaction to record the
2140 * failure to disk. ext3_error, for example, now uses this
2145 void jbd2_journal_abort(journal_t *journal, int errno)
2147 transaction_t *transaction;
2150 * ESHUTDOWN always takes precedence because a file system check
2151 * caused by any other journal abort error is not required after
2152 * a shutdown triggered.
2154 write_lock(&journal->j_state_lock);
2155 if (journal->j_flags & JBD2_ABORT) {
2156 int old_errno = journal->j_errno;
2158 write_unlock(&journal->j_state_lock);
2159 if (old_errno != -ESHUTDOWN && errno == -ESHUTDOWN) {
2160 journal->j_errno = errno;
2161 jbd2_journal_update_sb_errno(journal);
2167 * Mark the abort as occurred and start current running transaction
2168 * to release all journaled buffer.
2170 pr_err("Aborting journal on device %s.\n", journal->j_devname);
2172 journal->j_flags |= JBD2_ABORT;
2173 journal->j_errno = errno;
2174 transaction = journal->j_running_transaction;
2176 __jbd2_log_start_commit(journal, transaction->t_tid);
2177 write_unlock(&journal->j_state_lock);
2180 * Record errno to the journal super block, so that fsck and jbd2
2181 * layer could realise that a filesystem check is needed.
2183 jbd2_journal_update_sb_errno(journal);
2185 write_lock(&journal->j_state_lock);
2186 journal->j_flags |= JBD2_REC_ERR;
2187 write_unlock(&journal->j_state_lock);
2191 * int jbd2_journal_errno () - returns the journal's error state.
2192 * @journal: journal to examine.
2194 * This is the errno number set with jbd2_journal_abort(), the last
2195 * time the journal was mounted - if the journal was stopped
2196 * without calling abort this will be 0.
2198 * If the journal has been aborted on this mount time -EROFS will
2201 int jbd2_journal_errno(journal_t *journal)
2205 read_lock(&journal->j_state_lock);
2206 if (journal->j_flags & JBD2_ABORT)
2209 err = journal->j_errno;
2210 read_unlock(&journal->j_state_lock);
2215 * int jbd2_journal_clear_err () - clears the journal's error state
2216 * @journal: journal to act on.
2218 * An error must be cleared or acked to take a FS out of readonly
2221 int jbd2_journal_clear_err(journal_t *journal)
2225 write_lock(&journal->j_state_lock);
2226 if (journal->j_flags & JBD2_ABORT)
2229 journal->j_errno = 0;
2230 write_unlock(&journal->j_state_lock);
2235 * void jbd2_journal_ack_err() - Ack journal err.
2236 * @journal: journal to act on.
2238 * An error must be cleared or acked to take a FS out of readonly
2241 void jbd2_journal_ack_err(journal_t *journal)
2243 write_lock(&journal->j_state_lock);
2244 if (journal->j_errno)
2245 journal->j_flags |= JBD2_ACK_ERR;
2246 write_unlock(&journal->j_state_lock);
2249 int jbd2_journal_blocks_per_page(struct inode *inode)
2251 return 1 << (PAGE_SHIFT - inode->i_sb->s_blocksize_bits);
2255 * helper functions to deal with 32 or 64bit block numbers.
2257 size_t journal_tag_bytes(journal_t *journal)
2261 if (jbd2_has_feature_csum3(journal))
2262 return sizeof(journal_block_tag3_t);
2264 sz = sizeof(journal_block_tag_t);
2266 if (jbd2_has_feature_csum2(journal))
2267 sz += sizeof(__u16);
2269 if (jbd2_has_feature_64bit(journal))
2272 return sz - sizeof(__u32);
2276 * JBD memory management
2278 * These functions are used to allocate block-sized chunks of memory
2279 * used for making copies of buffer_head data. Very often it will be
2280 * page-sized chunks of data, but sometimes it will be in
2281 * sub-page-size chunks. (For example, 16k pages on Power systems
2282 * with a 4k block file system.) For blocks smaller than a page, we
2283 * use a SLAB allocator. There are slab caches for each block size,
2284 * which are allocated at mount time, if necessary, and we only free
2285 * (all of) the slab caches when/if the jbd2 module is unloaded. For
2286 * this reason we don't need to a mutex to protect access to
2287 * jbd2_slab[] allocating or releasing memory; only in
2288 * jbd2_journal_create_slab().
2290 #define JBD2_MAX_SLABS 8
2291 static struct kmem_cache *jbd2_slab[JBD2_MAX_SLABS];
2293 static const char *jbd2_slab_names[JBD2_MAX_SLABS] = {
2294 "jbd2_1k", "jbd2_2k", "jbd2_4k", "jbd2_8k",
2295 "jbd2_16k", "jbd2_32k", "jbd2_64k", "jbd2_128k"
2299 static void jbd2_journal_destroy_slabs(void)
2303 for (i = 0; i < JBD2_MAX_SLABS; i++) {
2304 kmem_cache_destroy(jbd2_slab[i]);
2305 jbd2_slab[i] = NULL;
2309 static int jbd2_journal_create_slab(size_t size)
2311 static DEFINE_MUTEX(jbd2_slab_create_mutex);
2312 int i = order_base_2(size) - 10;
2315 if (size == PAGE_SIZE)
2318 if (i >= JBD2_MAX_SLABS)
2321 if (unlikely(i < 0))
2323 mutex_lock(&jbd2_slab_create_mutex);
2325 mutex_unlock(&jbd2_slab_create_mutex);
2326 return 0; /* Already created */
2329 slab_size = 1 << (i+10);
2330 jbd2_slab[i] = kmem_cache_create(jbd2_slab_names[i], slab_size,
2331 slab_size, 0, NULL);
2332 mutex_unlock(&jbd2_slab_create_mutex);
2333 if (!jbd2_slab[i]) {
2334 printk(KERN_EMERG "JBD2: no memory for jbd2_slab cache\n");
2340 static struct kmem_cache *get_slab(size_t size)
2342 int i = order_base_2(size) - 10;
2344 BUG_ON(i >= JBD2_MAX_SLABS);
2345 if (unlikely(i < 0))
2347 BUG_ON(jbd2_slab[i] == NULL);
2348 return jbd2_slab[i];
2351 void *jbd2_alloc(size_t size, gfp_t flags)
2355 BUG_ON(size & (size-1)); /* Must be a power of 2 */
2357 if (size < PAGE_SIZE)
2358 ptr = kmem_cache_alloc(get_slab(size), flags);
2360 ptr = (void *)__get_free_pages(flags, get_order(size));
2362 /* Check alignment; SLUB has gotten this wrong in the past,
2363 * and this can lead to user data corruption! */
2364 BUG_ON(((unsigned long) ptr) & (size-1));
2369 void jbd2_free(void *ptr, size_t size)
2371 if (size < PAGE_SIZE)
2372 kmem_cache_free(get_slab(size), ptr);
2374 free_pages((unsigned long)ptr, get_order(size));
2378 * Journal_head storage management
2380 static struct kmem_cache *jbd2_journal_head_cache;
2381 #ifdef CONFIG_JBD2_DEBUG
2382 static atomic_t nr_journal_heads = ATOMIC_INIT(0);
2385 static int __init jbd2_journal_init_journal_head_cache(void)
2387 J_ASSERT(!jbd2_journal_head_cache);
2388 jbd2_journal_head_cache = kmem_cache_create("jbd2_journal_head",
2389 sizeof(struct journal_head),
2391 SLAB_TEMPORARY | SLAB_TYPESAFE_BY_RCU,
2393 if (!jbd2_journal_head_cache) {
2394 printk(KERN_EMERG "JBD2: no memory for journal_head cache\n");
2400 static void jbd2_journal_destroy_journal_head_cache(void)
2402 kmem_cache_destroy(jbd2_journal_head_cache);
2403 jbd2_journal_head_cache = NULL;
2407 * journal_head splicing and dicing
2409 static struct journal_head *journal_alloc_journal_head(void)
2411 struct journal_head *ret;
2413 #ifdef CONFIG_JBD2_DEBUG
2414 atomic_inc(&nr_journal_heads);
2416 ret = kmem_cache_zalloc(jbd2_journal_head_cache, GFP_NOFS);
2418 jbd_debug(1, "out of memory for journal_head\n");
2419 pr_notice_ratelimited("ENOMEM in %s, retrying.\n", __func__);
2420 ret = kmem_cache_zalloc(jbd2_journal_head_cache,
2421 GFP_NOFS | __GFP_NOFAIL);
2424 spin_lock_init(&ret->b_state_lock);
2428 static void journal_free_journal_head(struct journal_head *jh)
2430 #ifdef CONFIG_JBD2_DEBUG
2431 atomic_dec(&nr_journal_heads);
2432 memset(jh, JBD2_POISON_FREE, sizeof(*jh));
2434 kmem_cache_free(jbd2_journal_head_cache, jh);
2438 * A journal_head is attached to a buffer_head whenever JBD has an
2439 * interest in the buffer.
2441 * Whenever a buffer has an attached journal_head, its ->b_state:BH_JBD bit
2442 * is set. This bit is tested in core kernel code where we need to take
2443 * JBD-specific actions. Testing the zeroness of ->b_private is not reliable
2446 * When a buffer has its BH_JBD bit set, its ->b_count is elevated by one.
2448 * When a buffer has its BH_JBD bit set it is immune from being released by
2449 * core kernel code, mainly via ->b_count.
2451 * A journal_head is detached from its buffer_head when the journal_head's
2452 * b_jcount reaches zero. Running transaction (b_transaction) and checkpoint
2453 * transaction (b_cp_transaction) hold their references to b_jcount.
2455 * Various places in the kernel want to attach a journal_head to a buffer_head
2456 * _before_ attaching the journal_head to a transaction. To protect the
2457 * journal_head in this situation, jbd2_journal_add_journal_head elevates the
2458 * journal_head's b_jcount refcount by one. The caller must call
2459 * jbd2_journal_put_journal_head() to undo this.
2461 * So the typical usage would be:
2463 * (Attach a journal_head if needed. Increments b_jcount)
2464 * struct journal_head *jh = jbd2_journal_add_journal_head(bh);
2466 * (Get another reference for transaction)
2467 * jbd2_journal_grab_journal_head(bh);
2468 * jh->b_transaction = xxx;
2469 * (Put original reference)
2470 * jbd2_journal_put_journal_head(jh);
2474 * Give a buffer_head a journal_head.
2478 struct journal_head *jbd2_journal_add_journal_head(struct buffer_head *bh)
2480 struct journal_head *jh;
2481 struct journal_head *new_jh = NULL;
2484 if (!buffer_jbd(bh))
2485 new_jh = journal_alloc_journal_head();
2487 jbd_lock_bh_journal_head(bh);
2488 if (buffer_jbd(bh)) {
2492 (atomic_read(&bh->b_count) > 0) ||
2493 (bh->b_page && bh->b_page->mapping));
2496 jbd_unlock_bh_journal_head(bh);
2501 new_jh = NULL; /* We consumed it */
2506 BUFFER_TRACE(bh, "added journal_head");
2509 jbd_unlock_bh_journal_head(bh);
2511 journal_free_journal_head(new_jh);
2512 return bh->b_private;
2516 * Grab a ref against this buffer_head's journal_head. If it ended up not
2517 * having a journal_head, return NULL
2519 struct journal_head *jbd2_journal_grab_journal_head(struct buffer_head *bh)
2521 struct journal_head *jh = NULL;
2523 jbd_lock_bh_journal_head(bh);
2524 if (buffer_jbd(bh)) {
2528 jbd_unlock_bh_journal_head(bh);
2532 static void __journal_remove_journal_head(struct buffer_head *bh)
2534 struct journal_head *jh = bh2jh(bh);
2536 J_ASSERT_JH(jh, jh->b_transaction == NULL);
2537 J_ASSERT_JH(jh, jh->b_next_transaction == NULL);
2538 J_ASSERT_JH(jh, jh->b_cp_transaction == NULL);
2539 J_ASSERT_JH(jh, jh->b_jlist == BJ_None);
2540 J_ASSERT_BH(bh, buffer_jbd(bh));
2541 J_ASSERT_BH(bh, jh2bh(jh) == bh);
2542 BUFFER_TRACE(bh, "remove journal_head");
2544 /* Unlink before dropping the lock */
2545 bh->b_private = NULL;
2546 jh->b_bh = NULL; /* debug, really */
2547 clear_buffer_jbd(bh);
2550 static void journal_release_journal_head(struct journal_head *jh, size_t b_size)
2552 if (jh->b_frozen_data) {
2553 printk(KERN_WARNING "%s: freeing b_frozen_data\n", __func__);
2554 jbd2_free(jh->b_frozen_data, b_size);
2556 if (jh->b_committed_data) {
2557 printk(KERN_WARNING "%s: freeing b_committed_data\n", __func__);
2558 jbd2_free(jh->b_committed_data, b_size);
2560 journal_free_journal_head(jh);
2564 * Drop a reference on the passed journal_head. If it fell to zero then
2565 * release the journal_head from the buffer_head.
2567 void jbd2_journal_put_journal_head(struct journal_head *jh)
2569 struct buffer_head *bh = jh2bh(jh);
2571 jbd_lock_bh_journal_head(bh);
2572 J_ASSERT_JH(jh, jh->b_jcount > 0);
2574 if (!jh->b_jcount) {
2575 __journal_remove_journal_head(bh);
2576 jbd_unlock_bh_journal_head(bh);
2577 journal_release_journal_head(jh, bh->b_size);
2580 jbd_unlock_bh_journal_head(bh);
2585 * Initialize jbd inode head
2587 void jbd2_journal_init_jbd_inode(struct jbd2_inode *jinode, struct inode *inode)
2589 jinode->i_transaction = NULL;
2590 jinode->i_next_transaction = NULL;
2591 jinode->i_vfs_inode = inode;
2592 jinode->i_flags = 0;
2593 jinode->i_dirty_start = 0;
2594 jinode->i_dirty_end = 0;
2595 INIT_LIST_HEAD(&jinode->i_list);
2599 * Function to be called before we start removing inode from memory (i.e.,
2600 * clear_inode() is a fine place to be called from). It removes inode from
2601 * transaction's lists.
2603 void jbd2_journal_release_jbd_inode(journal_t *journal,
2604 struct jbd2_inode *jinode)
2609 spin_lock(&journal->j_list_lock);
2610 /* Is commit writing out inode - we have to wait */
2611 if (jinode->i_flags & JI_COMMIT_RUNNING) {
2612 wait_queue_head_t *wq;
2613 DEFINE_WAIT_BIT(wait, &jinode->i_flags, __JI_COMMIT_RUNNING);
2614 wq = bit_waitqueue(&jinode->i_flags, __JI_COMMIT_RUNNING);
2615 prepare_to_wait(wq, &wait.wq_entry, TASK_UNINTERRUPTIBLE);
2616 spin_unlock(&journal->j_list_lock);
2618 finish_wait(wq, &wait.wq_entry);
2622 if (jinode->i_transaction) {
2623 list_del(&jinode->i_list);
2624 jinode->i_transaction = NULL;
2626 spin_unlock(&journal->j_list_lock);
2630 #ifdef CONFIG_PROC_FS
2632 #define JBD2_STATS_PROC_NAME "fs/jbd2"
2634 static void __init jbd2_create_jbd_stats_proc_entry(void)
2636 proc_jbd2_stats = proc_mkdir(JBD2_STATS_PROC_NAME, NULL);
2639 static void __exit jbd2_remove_jbd_stats_proc_entry(void)
2641 if (proc_jbd2_stats)
2642 remove_proc_entry(JBD2_STATS_PROC_NAME, NULL);
2647 #define jbd2_create_jbd_stats_proc_entry() do {} while (0)
2648 #define jbd2_remove_jbd_stats_proc_entry() do {} while (0)
2652 struct kmem_cache *jbd2_handle_cache, *jbd2_inode_cache;
2654 static int __init jbd2_journal_init_inode_cache(void)
2656 J_ASSERT(!jbd2_inode_cache);
2657 jbd2_inode_cache = KMEM_CACHE(jbd2_inode, 0);
2658 if (!jbd2_inode_cache) {
2659 pr_emerg("JBD2: failed to create inode cache\n");
2665 static int __init jbd2_journal_init_handle_cache(void)
2667 J_ASSERT(!jbd2_handle_cache);
2668 jbd2_handle_cache = KMEM_CACHE(jbd2_journal_handle, SLAB_TEMPORARY);
2669 if (!jbd2_handle_cache) {
2670 printk(KERN_EMERG "JBD2: failed to create handle cache\n");
2676 static void jbd2_journal_destroy_inode_cache(void)
2678 kmem_cache_destroy(jbd2_inode_cache);
2679 jbd2_inode_cache = NULL;
2682 static void jbd2_journal_destroy_handle_cache(void)
2684 kmem_cache_destroy(jbd2_handle_cache);
2685 jbd2_handle_cache = NULL;
2689 * Module startup and shutdown
2692 static int __init journal_init_caches(void)
2696 ret = jbd2_journal_init_revoke_record_cache();
2698 ret = jbd2_journal_init_revoke_table_cache();
2700 ret = jbd2_journal_init_journal_head_cache();
2702 ret = jbd2_journal_init_handle_cache();
2704 ret = jbd2_journal_init_inode_cache();
2706 ret = jbd2_journal_init_transaction_cache();
2710 static void jbd2_journal_destroy_caches(void)
2712 jbd2_journal_destroy_revoke_record_cache();
2713 jbd2_journal_destroy_revoke_table_cache();
2714 jbd2_journal_destroy_journal_head_cache();
2715 jbd2_journal_destroy_handle_cache();
2716 jbd2_journal_destroy_inode_cache();
2717 jbd2_journal_destroy_transaction_cache();
2718 jbd2_journal_destroy_slabs();
2721 static int __init journal_init(void)
2725 BUILD_BUG_ON(sizeof(struct journal_superblock_s) != 1024);
2727 ret = journal_init_caches();
2729 jbd2_create_jbd_stats_proc_entry();
2731 jbd2_journal_destroy_caches();
2736 static void __exit journal_exit(void)
2738 #ifdef CONFIG_JBD2_DEBUG
2739 int n = atomic_read(&nr_journal_heads);
2741 printk(KERN_ERR "JBD2: leaked %d journal_heads!\n", n);
2743 jbd2_remove_jbd_stats_proc_entry();
2744 jbd2_journal_destroy_caches();
2747 MODULE_LICENSE("GPL");
2748 module_init(journal_init);
2749 module_exit(journal_exit);