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 jbd_lock_bh_state(bh_in);
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 jbd_unlock_bh_state(bh_in);
408 tmp = jbd2_alloc(bh_in->b_size, GFP_NOFS);
413 jbd_lock_bh_state(bh_in);
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 jbd_unlock_bh_state(bh_in);
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 * jbd2_journal_force_commit_nested - Force and wait upon a commit if the
566 * calling process is not within transaction.
568 * @journal: journal to force
569 * Returns true if progress was made.
571 * This is used for forcing out undo-protected data which contains
572 * bitmaps, when the fs is running out of space.
574 int jbd2_journal_force_commit_nested(journal_t *journal)
578 ret = __jbd2_journal_force_commit(journal);
583 * jbd2_journal_force_commit() - force any uncommitted transactions
584 * @journal: journal to force
586 * Caller want unconditional commit. We can only force the running transaction
587 * if we don't have an active handle, otherwise, we will deadlock.
589 int jbd2_journal_force_commit(journal_t *journal)
593 J_ASSERT(!current->journal_info);
594 ret = __jbd2_journal_force_commit(journal);
601 * Start a commit of the current running transaction (if any). Returns true
602 * if a transaction is going to be committed (or is currently already
603 * committing), and fills its tid in at *ptid
605 int jbd2_journal_start_commit(journal_t *journal, tid_t *ptid)
609 write_lock(&journal->j_state_lock);
610 if (journal->j_running_transaction) {
611 tid_t tid = journal->j_running_transaction->t_tid;
613 __jbd2_log_start_commit(journal, tid);
614 /* There's a running transaction and we've just made sure
615 * it's commit has been scheduled. */
619 } else if (journal->j_committing_transaction) {
621 * If commit has been started, then we have to wait for
622 * completion of that transaction.
625 *ptid = journal->j_committing_transaction->t_tid;
628 write_unlock(&journal->j_state_lock);
633 * Return 1 if a given transaction has not yet sent barrier request
634 * connected with a transaction commit. If 0 is returned, transaction
635 * may or may not have sent the barrier. Used to avoid sending barrier
636 * twice in common cases.
638 int jbd2_trans_will_send_data_barrier(journal_t *journal, tid_t tid)
641 transaction_t *commit_trans;
643 if (!(journal->j_flags & JBD2_BARRIER))
645 read_lock(&journal->j_state_lock);
646 /* Transaction already committed? */
647 if (tid_geq(journal->j_commit_sequence, tid))
649 commit_trans = journal->j_committing_transaction;
650 if (!commit_trans || commit_trans->t_tid != tid) {
655 * Transaction is being committed and we already proceeded to
656 * submitting a flush to fs partition?
658 if (journal->j_fs_dev != journal->j_dev) {
659 if (!commit_trans->t_need_data_flush ||
660 commit_trans->t_state >= T_COMMIT_DFLUSH)
663 if (commit_trans->t_state >= T_COMMIT_JFLUSH)
668 read_unlock(&journal->j_state_lock);
671 EXPORT_SYMBOL(jbd2_trans_will_send_data_barrier);
674 * Wait for a specified commit to complete.
675 * The caller may not hold the journal lock.
677 int jbd2_log_wait_commit(journal_t *journal, tid_t tid)
681 read_lock(&journal->j_state_lock);
682 #ifdef CONFIG_PROVE_LOCKING
684 * Some callers make sure transaction is already committing and in that
685 * case we cannot block on open handles anymore. So don't warn in that
688 if (tid_gt(tid, journal->j_commit_sequence) &&
689 (!journal->j_committing_transaction ||
690 journal->j_committing_transaction->t_tid != tid)) {
691 read_unlock(&journal->j_state_lock);
692 jbd2_might_wait_for_commit(journal);
693 read_lock(&journal->j_state_lock);
696 #ifdef CONFIG_JBD2_DEBUG
697 if (!tid_geq(journal->j_commit_request, tid)) {
699 "%s: error: j_commit_request=%u, tid=%u\n",
700 __func__, journal->j_commit_request, tid);
703 while (tid_gt(tid, journal->j_commit_sequence)) {
704 jbd_debug(1, "JBD2: want %u, j_commit_sequence=%u\n",
705 tid, journal->j_commit_sequence);
706 read_unlock(&journal->j_state_lock);
707 wake_up(&journal->j_wait_commit);
708 wait_event(journal->j_wait_done_commit,
709 !tid_gt(tid, journal->j_commit_sequence));
710 read_lock(&journal->j_state_lock);
712 read_unlock(&journal->j_state_lock);
714 if (unlikely(is_journal_aborted(journal)))
719 /* Return 1 when transaction with given tid has already committed. */
720 int jbd2_transaction_committed(journal_t *journal, tid_t tid)
724 read_lock(&journal->j_state_lock);
725 if (journal->j_running_transaction &&
726 journal->j_running_transaction->t_tid == tid)
728 if (journal->j_committing_transaction &&
729 journal->j_committing_transaction->t_tid == tid)
731 read_unlock(&journal->j_state_lock);
734 EXPORT_SYMBOL(jbd2_transaction_committed);
737 * When this function returns the transaction corresponding to tid
738 * will be completed. If the transaction has currently running, start
739 * committing that transaction before waiting for it to complete. If
740 * the transaction id is stale, it is by definition already completed,
741 * so just return SUCCESS.
743 int jbd2_complete_transaction(journal_t *journal, tid_t tid)
745 int need_to_wait = 1;
747 read_lock(&journal->j_state_lock);
748 if (journal->j_running_transaction &&
749 journal->j_running_transaction->t_tid == tid) {
750 if (journal->j_commit_request != tid) {
751 /* transaction not yet started, so request it */
752 read_unlock(&journal->j_state_lock);
753 jbd2_log_start_commit(journal, tid);
756 } else if (!(journal->j_committing_transaction &&
757 journal->j_committing_transaction->t_tid == tid))
759 read_unlock(&journal->j_state_lock);
763 return jbd2_log_wait_commit(journal, tid);
765 EXPORT_SYMBOL(jbd2_complete_transaction);
768 * Log buffer allocation routines:
771 int jbd2_journal_next_log_block(journal_t *journal, unsigned long long *retp)
773 unsigned long blocknr;
775 write_lock(&journal->j_state_lock);
776 J_ASSERT(journal->j_free > 1);
778 blocknr = journal->j_head;
781 if (journal->j_head == journal->j_last)
782 journal->j_head = journal->j_first;
783 write_unlock(&journal->j_state_lock);
784 return jbd2_journal_bmap(journal, blocknr, retp);
788 * Conversion of logical to physical block numbers for the journal
790 * On external journals the journal blocks are identity-mapped, so
791 * this is a no-op. If needed, we can use j_blk_offset - everything is
794 int jbd2_journal_bmap(journal_t *journal, unsigned long blocknr,
795 unsigned long long *retp)
798 unsigned long long ret;
800 if (journal->j_inode) {
801 ret = bmap(journal->j_inode, blocknr);
805 printk(KERN_ALERT "%s: journal block not found "
806 "at offset %lu on %s\n",
807 __func__, blocknr, journal->j_devname);
809 jbd2_journal_abort(journal, err);
812 *retp = blocknr; /* +journal->j_blk_offset */
818 * We play buffer_head aliasing tricks to write data/metadata blocks to
819 * the journal without copying their contents, but for journal
820 * descriptor blocks we do need to generate bona fide buffers.
822 * After the caller of jbd2_journal_get_descriptor_buffer() has finished modifying
823 * the buffer's contents they really should run flush_dcache_page(bh->b_page).
824 * But we don't bother doing that, so there will be coherency problems with
825 * mmaps of blockdevs which hold live JBD-controlled filesystems.
828 jbd2_journal_get_descriptor_buffer(transaction_t *transaction, int type)
830 journal_t *journal = transaction->t_journal;
831 struct buffer_head *bh;
832 unsigned long long blocknr;
833 journal_header_t *header;
836 err = jbd2_journal_next_log_block(journal, &blocknr);
841 bh = __getblk(journal->j_dev, blocknr, journal->j_blocksize);
845 memset(bh->b_data, 0, journal->j_blocksize);
846 header = (journal_header_t *)bh->b_data;
847 header->h_magic = cpu_to_be32(JBD2_MAGIC_NUMBER);
848 header->h_blocktype = cpu_to_be32(type);
849 header->h_sequence = cpu_to_be32(transaction->t_tid);
850 set_buffer_uptodate(bh);
852 BUFFER_TRACE(bh, "return this buffer");
856 void jbd2_descriptor_block_csum_set(journal_t *j, struct buffer_head *bh)
858 struct jbd2_journal_block_tail *tail;
861 if (!jbd2_journal_has_csum_v2or3(j))
864 tail = (struct jbd2_journal_block_tail *)(bh->b_data + j->j_blocksize -
865 sizeof(struct jbd2_journal_block_tail));
866 tail->t_checksum = 0;
867 csum = jbd2_chksum(j, j->j_csum_seed, bh->b_data, j->j_blocksize);
868 tail->t_checksum = cpu_to_be32(csum);
872 * Return tid of the oldest transaction in the journal and block in the journal
873 * where the transaction starts.
875 * If the journal is now empty, return which will be the next transaction ID
876 * we will write and where will that transaction start.
878 * The return value is 0 if journal tail cannot be pushed any further, 1 if
881 int jbd2_journal_get_log_tail(journal_t *journal, tid_t *tid,
882 unsigned long *block)
884 transaction_t *transaction;
887 read_lock(&journal->j_state_lock);
888 spin_lock(&journal->j_list_lock);
889 transaction = journal->j_checkpoint_transactions;
891 *tid = transaction->t_tid;
892 *block = transaction->t_log_start;
893 } else if ((transaction = journal->j_committing_transaction) != NULL) {
894 *tid = transaction->t_tid;
895 *block = transaction->t_log_start;
896 } else if ((transaction = journal->j_running_transaction) != NULL) {
897 *tid = transaction->t_tid;
898 *block = journal->j_head;
900 *tid = journal->j_transaction_sequence;
901 *block = journal->j_head;
903 ret = tid_gt(*tid, journal->j_tail_sequence);
904 spin_unlock(&journal->j_list_lock);
905 read_unlock(&journal->j_state_lock);
911 * Update information in journal structure and in on disk journal superblock
912 * about log tail. This function does not check whether information passed in
913 * really pushes log tail further. It's responsibility of the caller to make
914 * sure provided log tail information is valid (e.g. by holding
915 * j_checkpoint_mutex all the time between computing log tail and calling this
916 * function as is the case with jbd2_cleanup_journal_tail()).
918 * Requires j_checkpoint_mutex
920 int __jbd2_update_log_tail(journal_t *journal, tid_t tid, unsigned long block)
925 BUG_ON(!mutex_is_locked(&journal->j_checkpoint_mutex));
928 * We cannot afford for write to remain in drive's caches since as
929 * soon as we update j_tail, next transaction can start reusing journal
930 * space and if we lose sb update during power failure we'd replay
931 * old transaction with possibly newly overwritten data.
933 ret = jbd2_journal_update_sb_log_tail(journal, tid, block,
938 write_lock(&journal->j_state_lock);
939 freed = block - journal->j_tail;
940 if (block < journal->j_tail)
941 freed += journal->j_last - journal->j_first;
943 trace_jbd2_update_log_tail(journal, tid, block, freed);
945 "Cleaning journal tail from %u to %u (offset %lu), "
947 journal->j_tail_sequence, tid, block, freed);
949 journal->j_free += freed;
950 journal->j_tail_sequence = tid;
951 journal->j_tail = block;
952 write_unlock(&journal->j_state_lock);
959 * This is a variation of __jbd2_update_log_tail which checks for validity of
960 * provided log tail and locks j_checkpoint_mutex. So it is safe against races
961 * with other threads updating log tail.
963 void jbd2_update_log_tail(journal_t *journal, tid_t tid, unsigned long block)
965 mutex_lock_io(&journal->j_checkpoint_mutex);
966 if (tid_gt(tid, journal->j_tail_sequence))
967 __jbd2_update_log_tail(journal, tid, block);
968 mutex_unlock(&journal->j_checkpoint_mutex);
971 struct jbd2_stats_proc_session {
973 struct transaction_stats_s *stats;
978 static void *jbd2_seq_info_start(struct seq_file *seq, loff_t *pos)
980 return *pos ? NULL : SEQ_START_TOKEN;
983 static void *jbd2_seq_info_next(struct seq_file *seq, void *v, loff_t *pos)
989 static int jbd2_seq_info_show(struct seq_file *seq, void *v)
991 struct jbd2_stats_proc_session *s = seq->private;
993 if (v != SEQ_START_TOKEN)
995 seq_printf(seq, "%lu transactions (%lu requested), "
996 "each up to %u blocks\n",
997 s->stats->ts_tid, s->stats->ts_requested,
998 s->journal->j_max_transaction_buffers);
999 if (s->stats->ts_tid == 0)
1001 seq_printf(seq, "average: \n %ums waiting for transaction\n",
1002 jiffies_to_msecs(s->stats->run.rs_wait / s->stats->ts_tid));
1003 seq_printf(seq, " %ums request delay\n",
1004 (s->stats->ts_requested == 0) ? 0 :
1005 jiffies_to_msecs(s->stats->run.rs_request_delay /
1006 s->stats->ts_requested));
1007 seq_printf(seq, " %ums running transaction\n",
1008 jiffies_to_msecs(s->stats->run.rs_running / s->stats->ts_tid));
1009 seq_printf(seq, " %ums transaction was being locked\n",
1010 jiffies_to_msecs(s->stats->run.rs_locked / s->stats->ts_tid));
1011 seq_printf(seq, " %ums flushing data (in ordered mode)\n",
1012 jiffies_to_msecs(s->stats->run.rs_flushing / s->stats->ts_tid));
1013 seq_printf(seq, " %ums logging transaction\n",
1014 jiffies_to_msecs(s->stats->run.rs_logging / s->stats->ts_tid));
1015 seq_printf(seq, " %lluus average transaction commit time\n",
1016 div_u64(s->journal->j_average_commit_time, 1000));
1017 seq_printf(seq, " %lu handles per transaction\n",
1018 s->stats->run.rs_handle_count / s->stats->ts_tid);
1019 seq_printf(seq, " %lu blocks per transaction\n",
1020 s->stats->run.rs_blocks / s->stats->ts_tid);
1021 seq_printf(seq, " %lu logged blocks per transaction\n",
1022 s->stats->run.rs_blocks_logged / s->stats->ts_tid);
1026 static void jbd2_seq_info_stop(struct seq_file *seq, void *v)
1030 static const struct seq_operations jbd2_seq_info_ops = {
1031 .start = jbd2_seq_info_start,
1032 .next = jbd2_seq_info_next,
1033 .stop = jbd2_seq_info_stop,
1034 .show = jbd2_seq_info_show,
1037 static int jbd2_seq_info_open(struct inode *inode, struct file *file)
1039 journal_t *journal = PDE_DATA(inode);
1040 struct jbd2_stats_proc_session *s;
1043 s = kmalloc(sizeof(*s), GFP_KERNEL);
1046 size = sizeof(struct transaction_stats_s);
1047 s->stats = kmalloc(size, GFP_KERNEL);
1048 if (s->stats == NULL) {
1052 spin_lock(&journal->j_history_lock);
1053 memcpy(s->stats, &journal->j_stats, size);
1054 s->journal = journal;
1055 spin_unlock(&journal->j_history_lock);
1057 rc = seq_open(file, &jbd2_seq_info_ops);
1059 struct seq_file *m = file->private_data;
1069 static int jbd2_seq_info_release(struct inode *inode, struct file *file)
1071 struct seq_file *seq = file->private_data;
1072 struct jbd2_stats_proc_session *s = seq->private;
1075 return seq_release(inode, file);
1078 static const struct file_operations jbd2_seq_info_fops = {
1079 .owner = THIS_MODULE,
1080 .open = jbd2_seq_info_open,
1082 .llseek = seq_lseek,
1083 .release = jbd2_seq_info_release,
1086 static struct proc_dir_entry *proc_jbd2_stats;
1088 static void jbd2_stats_proc_init(journal_t *journal)
1090 journal->j_proc_entry = proc_mkdir(journal->j_devname, proc_jbd2_stats);
1091 if (journal->j_proc_entry) {
1092 proc_create_data("info", S_IRUGO, journal->j_proc_entry,
1093 &jbd2_seq_info_fops, journal);
1097 static void jbd2_stats_proc_exit(journal_t *journal)
1099 remove_proc_entry("info", journal->j_proc_entry);
1100 remove_proc_entry(journal->j_devname, proc_jbd2_stats);
1104 * Management for journal control blocks: functions to create and
1105 * destroy journal_t structures, and to initialise and read existing
1106 * journal blocks from disk. */
1108 /* First: create and setup a journal_t object in memory. We initialise
1109 * very few fields yet: that has to wait until we have created the
1110 * journal structures from from scratch, or loaded them from disk. */
1112 static journal_t *journal_init_common(struct block_device *bdev,
1113 struct block_device *fs_dev,
1114 unsigned long long start, int len, int blocksize)
1116 static struct lock_class_key jbd2_trans_commit_key;
1119 struct buffer_head *bh;
1122 journal = kzalloc(sizeof(*journal), GFP_KERNEL);
1126 init_waitqueue_head(&journal->j_wait_transaction_locked);
1127 init_waitqueue_head(&journal->j_wait_done_commit);
1128 init_waitqueue_head(&journal->j_wait_commit);
1129 init_waitqueue_head(&journal->j_wait_updates);
1130 init_waitqueue_head(&journal->j_wait_reserved);
1131 mutex_init(&journal->j_barrier);
1132 mutex_init(&journal->j_checkpoint_mutex);
1133 spin_lock_init(&journal->j_revoke_lock);
1134 spin_lock_init(&journal->j_list_lock);
1135 rwlock_init(&journal->j_state_lock);
1137 journal->j_commit_interval = (HZ * JBD2_DEFAULT_MAX_COMMIT_AGE);
1138 journal->j_min_batch_time = 0;
1139 journal->j_max_batch_time = 15000; /* 15ms */
1140 atomic_set(&journal->j_reserved_credits, 0);
1142 /* The journal is marked for error until we succeed with recovery! */
1143 journal->j_flags = JBD2_ABORT;
1145 /* Set up a default-sized revoke table for the new mount. */
1146 err = jbd2_journal_init_revoke(journal, JOURNAL_REVOKE_DEFAULT_HASH);
1150 spin_lock_init(&journal->j_history_lock);
1152 lockdep_init_map(&journal->j_trans_commit_map, "jbd2_handle",
1153 &jbd2_trans_commit_key, 0);
1155 /* journal descriptor can store up to n blocks -bzzz */
1156 journal->j_blocksize = blocksize;
1157 journal->j_dev = bdev;
1158 journal->j_fs_dev = fs_dev;
1159 journal->j_blk_offset = start;
1160 journal->j_maxlen = len;
1161 n = journal->j_blocksize / sizeof(journal_block_tag_t);
1162 journal->j_wbufsize = n;
1163 journal->j_wbuf = kmalloc_array(n, sizeof(struct buffer_head *),
1165 if (!journal->j_wbuf)
1168 bh = getblk_unmovable(journal->j_dev, start, journal->j_blocksize);
1170 pr_err("%s: Cannot get buffer for journal superblock\n",
1174 journal->j_sb_buffer = bh;
1175 journal->j_superblock = (journal_superblock_t *)bh->b_data;
1180 kfree(journal->j_wbuf);
1181 jbd2_journal_destroy_revoke(journal);
1186 /* jbd2_journal_init_dev and jbd2_journal_init_inode:
1188 * Create a journal structure assigned some fixed set of disk blocks to
1189 * the journal. We don't actually touch those disk blocks yet, but we
1190 * need to set up all of the mapping information to tell the journaling
1191 * system where the journal blocks are.
1196 * journal_t * jbd2_journal_init_dev() - creates and initialises a journal structure
1197 * @bdev: Block device on which to create the journal
1198 * @fs_dev: Device which hold journalled filesystem for this journal.
1199 * @start: Block nr Start of journal.
1200 * @len: Length of the journal in blocks.
1201 * @blocksize: blocksize of journalling device
1203 * Returns: a newly created journal_t *
1205 * jbd2_journal_init_dev creates a journal which maps a fixed contiguous
1206 * range of blocks on an arbitrary block device.
1209 journal_t *jbd2_journal_init_dev(struct block_device *bdev,
1210 struct block_device *fs_dev,
1211 unsigned long long start, int len, int blocksize)
1215 journal = journal_init_common(bdev, fs_dev, start, len, blocksize);
1219 bdevname(journal->j_dev, journal->j_devname);
1220 strreplace(journal->j_devname, '/', '!');
1221 jbd2_stats_proc_init(journal);
1227 * journal_t * jbd2_journal_init_inode () - creates a journal which maps to a inode.
1228 * @inode: An inode to create the journal in
1230 * jbd2_journal_init_inode creates a journal which maps an on-disk inode as
1231 * the journal. The inode must exist already, must support bmap() and
1232 * must have all data blocks preallocated.
1234 journal_t *jbd2_journal_init_inode(struct inode *inode)
1238 unsigned long long blocknr;
1240 blocknr = bmap(inode, 0);
1242 pr_err("%s: Cannot locate journal superblock\n",
1247 jbd_debug(1, "JBD2: inode %s/%ld, size %lld, bits %d, blksize %ld\n",
1248 inode->i_sb->s_id, inode->i_ino, (long long) inode->i_size,
1249 inode->i_sb->s_blocksize_bits, inode->i_sb->s_blocksize);
1251 journal = journal_init_common(inode->i_sb->s_bdev, inode->i_sb->s_bdev,
1252 blocknr, inode->i_size >> inode->i_sb->s_blocksize_bits,
1253 inode->i_sb->s_blocksize);
1257 journal->j_inode = inode;
1258 bdevname(journal->j_dev, journal->j_devname);
1259 p = strreplace(journal->j_devname, '/', '!');
1260 sprintf(p, "-%lu", journal->j_inode->i_ino);
1261 jbd2_stats_proc_init(journal);
1267 * If the journal init or create aborts, we need to mark the journal
1268 * superblock as being NULL to prevent the journal destroy from writing
1269 * back a bogus superblock.
1271 static void journal_fail_superblock(journal_t *journal)
1273 struct buffer_head *bh = journal->j_sb_buffer;
1275 journal->j_sb_buffer = NULL;
1279 * Given a journal_t structure, initialise the various fields for
1280 * startup of a new journaling session. We use this both when creating
1281 * a journal, and after recovering an old journal to reset it for
1285 static int journal_reset(journal_t *journal)
1287 journal_superblock_t *sb = journal->j_superblock;
1288 unsigned long long first, last;
1290 first = be32_to_cpu(sb->s_first);
1291 last = be32_to_cpu(sb->s_maxlen);
1292 if (first + JBD2_MIN_JOURNAL_BLOCKS > last + 1) {
1293 printk(KERN_ERR "JBD2: Journal too short (blocks %llu-%llu).\n",
1295 journal_fail_superblock(journal);
1299 journal->j_first = first;
1300 journal->j_last = last;
1302 journal->j_head = first;
1303 journal->j_tail = first;
1304 journal->j_free = last - first;
1306 journal->j_tail_sequence = journal->j_transaction_sequence;
1307 journal->j_commit_sequence = journal->j_transaction_sequence - 1;
1308 journal->j_commit_request = journal->j_commit_sequence;
1310 journal->j_max_transaction_buffers = journal->j_maxlen / 4;
1313 * As a special case, if the on-disk copy is already marked as needing
1314 * no recovery (s_start == 0), then we can safely defer the superblock
1315 * update until the next commit by setting JBD2_FLUSHED. This avoids
1316 * attempting a write to a potential-readonly device.
1318 if (sb->s_start == 0) {
1319 jbd_debug(1, "JBD2: Skipping superblock update on recovered sb "
1320 "(start %ld, seq %u, errno %d)\n",
1321 journal->j_tail, journal->j_tail_sequence,
1323 journal->j_flags |= JBD2_FLUSHED;
1325 /* Lock here to make assertions happy... */
1326 mutex_lock_io(&journal->j_checkpoint_mutex);
1328 * Update log tail information. We use REQ_FUA since new
1329 * transaction will start reusing journal space and so we
1330 * must make sure information about current log tail is on
1333 jbd2_journal_update_sb_log_tail(journal,
1334 journal->j_tail_sequence,
1336 REQ_SYNC | REQ_FUA);
1337 mutex_unlock(&journal->j_checkpoint_mutex);
1339 return jbd2_journal_start_thread(journal);
1343 * This function expects that the caller will have locked the journal
1344 * buffer head, and will return with it unlocked
1346 static int jbd2_write_superblock(journal_t *journal, int write_flags)
1348 struct buffer_head *bh = journal->j_sb_buffer;
1349 journal_superblock_t *sb = journal->j_superblock;
1352 /* Buffer got discarded which means block device got invalidated */
1353 if (!buffer_mapped(bh)) {
1358 trace_jbd2_write_superblock(journal, write_flags);
1359 if (!(journal->j_flags & JBD2_BARRIER))
1360 write_flags &= ~(REQ_FUA | REQ_PREFLUSH);
1361 if (buffer_write_io_error(bh)) {
1363 * Oh, dear. A previous attempt to write the journal
1364 * superblock failed. This could happen because the
1365 * USB device was yanked out. Or it could happen to
1366 * be a transient write error and maybe the block will
1367 * be remapped. Nothing we can do but to retry the
1368 * write and hope for the best.
1370 printk(KERN_ERR "JBD2: previous I/O error detected "
1371 "for journal superblock update for %s.\n",
1372 journal->j_devname);
1373 clear_buffer_write_io_error(bh);
1374 set_buffer_uptodate(bh);
1376 if (jbd2_journal_has_csum_v2or3(journal))
1377 sb->s_checksum = jbd2_superblock_csum(journal, sb);
1379 bh->b_end_io = end_buffer_write_sync;
1380 ret = submit_bh(REQ_OP_WRITE, write_flags, bh);
1382 if (buffer_write_io_error(bh)) {
1383 clear_buffer_write_io_error(bh);
1384 set_buffer_uptodate(bh);
1388 printk(KERN_ERR "JBD2: Error %d detected when updating "
1389 "journal superblock for %s.\n", ret,
1390 journal->j_devname);
1391 jbd2_journal_abort(journal, ret);
1398 * jbd2_journal_update_sb_log_tail() - Update log tail in journal sb on disk.
1399 * @journal: The journal to update.
1400 * @tail_tid: TID of the new transaction at the tail of the log
1401 * @tail_block: The first block of the transaction at the tail of the log
1402 * @write_op: With which operation should we write the journal sb
1404 * Update a journal's superblock information about log tail and write it to
1405 * disk, waiting for the IO to complete.
1407 int jbd2_journal_update_sb_log_tail(journal_t *journal, tid_t tail_tid,
1408 unsigned long tail_block, int write_op)
1410 journal_superblock_t *sb = journal->j_superblock;
1413 if (is_journal_aborted(journal))
1416 BUG_ON(!mutex_is_locked(&journal->j_checkpoint_mutex));
1417 jbd_debug(1, "JBD2: updating superblock (start %lu, seq %u)\n",
1418 tail_block, tail_tid);
1420 lock_buffer(journal->j_sb_buffer);
1421 sb->s_sequence = cpu_to_be32(tail_tid);
1422 sb->s_start = cpu_to_be32(tail_block);
1424 ret = jbd2_write_superblock(journal, write_op);
1428 /* Log is no longer empty */
1429 write_lock(&journal->j_state_lock);
1430 WARN_ON(!sb->s_sequence);
1431 journal->j_flags &= ~JBD2_FLUSHED;
1432 write_unlock(&journal->j_state_lock);
1439 * jbd2_mark_journal_empty() - Mark on disk journal as empty.
1440 * @journal: The journal to update.
1441 * @write_op: With which operation should we write the journal sb
1443 * Update a journal's dynamic superblock fields to show that journal is empty.
1444 * Write updated superblock to disk waiting for IO to complete.
1446 static void jbd2_mark_journal_empty(journal_t *journal, int write_op)
1448 journal_superblock_t *sb = journal->j_superblock;
1450 BUG_ON(!mutex_is_locked(&journal->j_checkpoint_mutex));
1451 lock_buffer(journal->j_sb_buffer);
1452 if (sb->s_start == 0) { /* Is it already empty? */
1453 unlock_buffer(journal->j_sb_buffer);
1457 jbd_debug(1, "JBD2: Marking journal as empty (seq %u)\n",
1458 journal->j_tail_sequence);
1460 sb->s_sequence = cpu_to_be32(journal->j_tail_sequence);
1461 sb->s_start = cpu_to_be32(0);
1463 jbd2_write_superblock(journal, write_op);
1465 /* Log is no longer empty */
1466 write_lock(&journal->j_state_lock);
1467 journal->j_flags |= JBD2_FLUSHED;
1468 write_unlock(&journal->j_state_lock);
1473 * jbd2_journal_update_sb_errno() - Update error in the journal.
1474 * @journal: The journal to update.
1476 * Update a journal's errno. Write updated superblock to disk waiting for IO
1479 void jbd2_journal_update_sb_errno(journal_t *journal)
1481 journal_superblock_t *sb = journal->j_superblock;
1484 lock_buffer(journal->j_sb_buffer);
1485 errcode = journal->j_errno;
1486 if (errcode == -ESHUTDOWN)
1488 jbd_debug(1, "JBD2: updating superblock error (errno %d)\n", errcode);
1489 sb->s_errno = cpu_to_be32(errcode);
1491 jbd2_write_superblock(journal, REQ_SYNC | REQ_FUA);
1493 EXPORT_SYMBOL(jbd2_journal_update_sb_errno);
1496 * Read the superblock for a given journal, performing initial
1497 * validation of the format.
1499 static int journal_get_superblock(journal_t *journal)
1501 struct buffer_head *bh;
1502 journal_superblock_t *sb;
1505 bh = journal->j_sb_buffer;
1507 J_ASSERT(bh != NULL);
1508 if (!buffer_uptodate(bh)) {
1509 ll_rw_block(REQ_OP_READ, 0, 1, &bh);
1511 if (!buffer_uptodate(bh)) {
1513 "JBD2: IO error reading journal superblock\n");
1518 if (buffer_verified(bh))
1521 sb = journal->j_superblock;
1525 if (sb->s_header.h_magic != cpu_to_be32(JBD2_MAGIC_NUMBER) ||
1526 sb->s_blocksize != cpu_to_be32(journal->j_blocksize)) {
1527 printk(KERN_WARNING "JBD2: no valid journal superblock found\n");
1531 switch(be32_to_cpu(sb->s_header.h_blocktype)) {
1532 case JBD2_SUPERBLOCK_V1:
1533 journal->j_format_version = 1;
1535 case JBD2_SUPERBLOCK_V2:
1536 journal->j_format_version = 2;
1539 printk(KERN_WARNING "JBD2: unrecognised superblock format ID\n");
1543 if (be32_to_cpu(sb->s_maxlen) < journal->j_maxlen)
1544 journal->j_maxlen = be32_to_cpu(sb->s_maxlen);
1545 else if (be32_to_cpu(sb->s_maxlen) > journal->j_maxlen) {
1546 printk(KERN_WARNING "JBD2: journal file too short\n");
1550 if (be32_to_cpu(sb->s_first) == 0 ||
1551 be32_to_cpu(sb->s_first) >= journal->j_maxlen) {
1553 "JBD2: Invalid start block of journal: %u\n",
1554 be32_to_cpu(sb->s_first));
1558 if (jbd2_has_feature_csum2(journal) &&
1559 jbd2_has_feature_csum3(journal)) {
1560 /* Can't have checksum v2 and v3 at the same time! */
1561 printk(KERN_ERR "JBD2: Can't enable checksumming v2 and v3 "
1562 "at the same time!\n");
1566 if (jbd2_journal_has_csum_v2or3_feature(journal) &&
1567 jbd2_has_feature_checksum(journal)) {
1568 /* Can't have checksum v1 and v2 on at the same time! */
1569 printk(KERN_ERR "JBD2: Can't enable checksumming v1 and v2/3 "
1570 "at the same time!\n");
1574 if (!jbd2_verify_csum_type(journal, sb)) {
1575 printk(KERN_ERR "JBD2: Unknown checksum type\n");
1579 /* Load the checksum driver */
1580 if (jbd2_journal_has_csum_v2or3_feature(journal)) {
1581 journal->j_chksum_driver = crypto_alloc_shash("crc32c", 0, 0);
1582 if (IS_ERR(journal->j_chksum_driver)) {
1583 printk(KERN_ERR "JBD2: Cannot load crc32c driver.\n");
1584 err = PTR_ERR(journal->j_chksum_driver);
1585 journal->j_chksum_driver = NULL;
1590 if (jbd2_journal_has_csum_v2or3(journal)) {
1591 /* Check superblock checksum */
1592 if (sb->s_checksum != jbd2_superblock_csum(journal, sb)) {
1593 printk(KERN_ERR "JBD2: journal checksum error\n");
1598 /* Precompute checksum seed for all metadata */
1599 journal->j_csum_seed = jbd2_chksum(journal, ~0, sb->s_uuid,
1600 sizeof(sb->s_uuid));
1603 set_buffer_verified(bh);
1608 journal_fail_superblock(journal);
1613 * Load the on-disk journal superblock and read the key fields into the
1617 static int load_superblock(journal_t *journal)
1620 journal_superblock_t *sb;
1622 err = journal_get_superblock(journal);
1626 sb = journal->j_superblock;
1628 journal->j_tail_sequence = be32_to_cpu(sb->s_sequence);
1629 journal->j_tail = be32_to_cpu(sb->s_start);
1630 journal->j_first = be32_to_cpu(sb->s_first);
1631 journal->j_last = be32_to_cpu(sb->s_maxlen);
1632 journal->j_errno = be32_to_cpu(sb->s_errno);
1639 * jbd2_journal_load() - Read journal from disk.
1640 * @journal: Journal to act on.
1642 * Given a journal_t structure which tells us which disk blocks contain
1643 * a journal, read the journal from disk to initialise the in-memory
1646 int jbd2_journal_load(journal_t *journal)
1649 journal_superblock_t *sb;
1651 err = load_superblock(journal);
1655 sb = journal->j_superblock;
1656 /* If this is a V2 superblock, then we have to check the
1657 * features flags on it. */
1659 if (journal->j_format_version >= 2) {
1660 if ((sb->s_feature_ro_compat &
1661 ~cpu_to_be32(JBD2_KNOWN_ROCOMPAT_FEATURES)) ||
1662 (sb->s_feature_incompat &
1663 ~cpu_to_be32(JBD2_KNOWN_INCOMPAT_FEATURES))) {
1665 "JBD2: Unrecognised features on journal\n");
1671 * Create a slab for this blocksize
1673 err = jbd2_journal_create_slab(be32_to_cpu(sb->s_blocksize));
1677 /* Let the recovery code check whether it needs to recover any
1678 * data from the journal. */
1679 if (jbd2_journal_recover(journal))
1680 goto recovery_error;
1682 if (journal->j_failed_commit) {
1683 printk(KERN_ERR "JBD2: journal transaction %u on %s "
1684 "is corrupt.\n", journal->j_failed_commit,
1685 journal->j_devname);
1686 return -EFSCORRUPTED;
1689 * clear JBD2_ABORT flag initialized in journal_init_common
1690 * here to update log tail information with the newest seq.
1692 journal->j_flags &= ~JBD2_ABORT;
1694 /* OK, we've finished with the dynamic journal bits:
1695 * reinitialise the dynamic contents of the superblock in memory
1696 * and reset them on disk. */
1697 if (journal_reset(journal))
1698 goto recovery_error;
1700 journal->j_flags |= JBD2_LOADED;
1704 printk(KERN_WARNING "JBD2: recovery failed\n");
1709 * jbd2_journal_destroy() - Release a journal_t structure.
1710 * @journal: Journal to act on.
1712 * Release a journal_t structure once it is no longer in use by the
1714 * Return <0 if we couldn't clean up the journal.
1716 int jbd2_journal_destroy(journal_t *journal)
1720 /* Wait for the commit thread to wake up and die. */
1721 journal_kill_thread(journal);
1723 /* Force a final log commit */
1724 if (journal->j_running_transaction)
1725 jbd2_journal_commit_transaction(journal);
1727 /* Force any old transactions to disk */
1729 /* Totally anal locking here... */
1730 spin_lock(&journal->j_list_lock);
1731 while (journal->j_checkpoint_transactions != NULL) {
1732 spin_unlock(&journal->j_list_lock);
1733 mutex_lock_io(&journal->j_checkpoint_mutex);
1734 err = jbd2_log_do_checkpoint(journal);
1735 mutex_unlock(&journal->j_checkpoint_mutex);
1737 * If checkpointing failed, just free the buffers to avoid
1741 jbd2_journal_destroy_checkpoint(journal);
1742 spin_lock(&journal->j_list_lock);
1745 spin_lock(&journal->j_list_lock);
1748 J_ASSERT(journal->j_running_transaction == NULL);
1749 J_ASSERT(journal->j_committing_transaction == NULL);
1750 J_ASSERT(journal->j_checkpoint_transactions == NULL);
1751 spin_unlock(&journal->j_list_lock);
1753 if (journal->j_sb_buffer) {
1754 if (!is_journal_aborted(journal)) {
1755 mutex_lock_io(&journal->j_checkpoint_mutex);
1757 write_lock(&journal->j_state_lock);
1758 journal->j_tail_sequence =
1759 ++journal->j_transaction_sequence;
1760 write_unlock(&journal->j_state_lock);
1762 jbd2_mark_journal_empty(journal,
1763 REQ_SYNC | REQ_PREFLUSH | REQ_FUA);
1764 mutex_unlock(&journal->j_checkpoint_mutex);
1767 brelse(journal->j_sb_buffer);
1770 if (journal->j_proc_entry)
1771 jbd2_stats_proc_exit(journal);
1772 iput(journal->j_inode);
1773 if (journal->j_revoke)
1774 jbd2_journal_destroy_revoke(journal);
1775 if (journal->j_chksum_driver)
1776 crypto_free_shash(journal->j_chksum_driver);
1777 kfree(journal->j_wbuf);
1785 * jbd2_journal_check_used_features() - Check if features specified are used.
1786 * @journal: Journal to check.
1787 * @compat: bitmask of compatible features
1788 * @ro: bitmask of features that force read-only mount
1789 * @incompat: bitmask of incompatible features
1791 * Check whether the journal uses all of a given set of
1792 * features. Return true (non-zero) if it does.
1795 int jbd2_journal_check_used_features(journal_t *journal, unsigned long compat,
1796 unsigned long ro, unsigned long incompat)
1798 journal_superblock_t *sb;
1800 if (!compat && !ro && !incompat)
1802 /* Load journal superblock if it is not loaded yet. */
1803 if (journal->j_format_version == 0 &&
1804 journal_get_superblock(journal) != 0)
1806 if (journal->j_format_version == 1)
1809 sb = journal->j_superblock;
1811 if (((be32_to_cpu(sb->s_feature_compat) & compat) == compat) &&
1812 ((be32_to_cpu(sb->s_feature_ro_compat) & ro) == ro) &&
1813 ((be32_to_cpu(sb->s_feature_incompat) & incompat) == incompat))
1820 * jbd2_journal_check_available_features() - Check feature set in journalling layer
1821 * @journal: Journal to check.
1822 * @compat: bitmask of compatible features
1823 * @ro: bitmask of features that force read-only mount
1824 * @incompat: bitmask of incompatible features
1826 * Check whether the journaling code supports the use of
1827 * all of a given set of features on this journal. Return true
1828 * (non-zero) if it can. */
1830 int jbd2_journal_check_available_features(journal_t *journal, unsigned long compat,
1831 unsigned long ro, unsigned long incompat)
1833 if (!compat && !ro && !incompat)
1836 /* We can support any known requested features iff the
1837 * superblock is in version 2. Otherwise we fail to support any
1838 * extended sb features. */
1840 if (journal->j_format_version != 2)
1843 if ((compat & JBD2_KNOWN_COMPAT_FEATURES) == compat &&
1844 (ro & JBD2_KNOWN_ROCOMPAT_FEATURES) == ro &&
1845 (incompat & JBD2_KNOWN_INCOMPAT_FEATURES) == incompat)
1852 * jbd2_journal_set_features() - Mark a given journal feature in the superblock
1853 * @journal: Journal to act on.
1854 * @compat: bitmask of compatible features
1855 * @ro: bitmask of features that force read-only mount
1856 * @incompat: bitmask of incompatible features
1858 * Mark a given journal feature as present on the
1859 * superblock. Returns true if the requested features could be set.
1863 int jbd2_journal_set_features(journal_t *journal, unsigned long compat,
1864 unsigned long ro, unsigned long incompat)
1866 #define INCOMPAT_FEATURE_ON(f) \
1867 ((incompat & (f)) && !(sb->s_feature_incompat & cpu_to_be32(f)))
1868 #define COMPAT_FEATURE_ON(f) \
1869 ((compat & (f)) && !(sb->s_feature_compat & cpu_to_be32(f)))
1870 journal_superblock_t *sb;
1872 if (jbd2_journal_check_used_features(journal, compat, ro, incompat))
1875 if (!jbd2_journal_check_available_features(journal, compat, ro, incompat))
1878 /* If enabling v2 checksums, turn on v3 instead */
1879 if (incompat & JBD2_FEATURE_INCOMPAT_CSUM_V2) {
1880 incompat &= ~JBD2_FEATURE_INCOMPAT_CSUM_V2;
1881 incompat |= JBD2_FEATURE_INCOMPAT_CSUM_V3;
1884 /* Asking for checksumming v3 and v1? Only give them v3. */
1885 if (incompat & JBD2_FEATURE_INCOMPAT_CSUM_V3 &&
1886 compat & JBD2_FEATURE_COMPAT_CHECKSUM)
1887 compat &= ~JBD2_FEATURE_COMPAT_CHECKSUM;
1889 jbd_debug(1, "Setting new features 0x%lx/0x%lx/0x%lx\n",
1890 compat, ro, incompat);
1892 sb = journal->j_superblock;
1894 /* Load the checksum driver if necessary */
1895 if ((journal->j_chksum_driver == NULL) &&
1896 INCOMPAT_FEATURE_ON(JBD2_FEATURE_INCOMPAT_CSUM_V3)) {
1897 journal->j_chksum_driver = crypto_alloc_shash("crc32c", 0, 0);
1898 if (IS_ERR(journal->j_chksum_driver)) {
1899 printk(KERN_ERR "JBD2: Cannot load crc32c driver.\n");
1900 journal->j_chksum_driver = NULL;
1903 /* Precompute checksum seed for all metadata */
1904 journal->j_csum_seed = jbd2_chksum(journal, ~0, sb->s_uuid,
1905 sizeof(sb->s_uuid));
1908 lock_buffer(journal->j_sb_buffer);
1910 /* If enabling v3 checksums, update superblock */
1911 if (INCOMPAT_FEATURE_ON(JBD2_FEATURE_INCOMPAT_CSUM_V3)) {
1912 sb->s_checksum_type = JBD2_CRC32C_CHKSUM;
1913 sb->s_feature_compat &=
1914 ~cpu_to_be32(JBD2_FEATURE_COMPAT_CHECKSUM);
1917 /* If enabling v1 checksums, downgrade superblock */
1918 if (COMPAT_FEATURE_ON(JBD2_FEATURE_COMPAT_CHECKSUM))
1919 sb->s_feature_incompat &=
1920 ~cpu_to_be32(JBD2_FEATURE_INCOMPAT_CSUM_V2 |
1921 JBD2_FEATURE_INCOMPAT_CSUM_V3);
1923 sb->s_feature_compat |= cpu_to_be32(compat);
1924 sb->s_feature_ro_compat |= cpu_to_be32(ro);
1925 sb->s_feature_incompat |= cpu_to_be32(incompat);
1926 unlock_buffer(journal->j_sb_buffer);
1929 #undef COMPAT_FEATURE_ON
1930 #undef INCOMPAT_FEATURE_ON
1934 * jbd2_journal_clear_features() - Clear a given journal feature in the
1936 * @journal: Journal to act on.
1937 * @compat: bitmask of compatible features
1938 * @ro: bitmask of features that force read-only mount
1939 * @incompat: bitmask of incompatible features
1941 * Clear a given journal feature as present on the
1944 void jbd2_journal_clear_features(journal_t *journal, unsigned long compat,
1945 unsigned long ro, unsigned long incompat)
1947 journal_superblock_t *sb;
1949 jbd_debug(1, "Clear features 0x%lx/0x%lx/0x%lx\n",
1950 compat, ro, incompat);
1952 sb = journal->j_superblock;
1954 sb->s_feature_compat &= ~cpu_to_be32(compat);
1955 sb->s_feature_ro_compat &= ~cpu_to_be32(ro);
1956 sb->s_feature_incompat &= ~cpu_to_be32(incompat);
1958 EXPORT_SYMBOL(jbd2_journal_clear_features);
1961 * jbd2_journal_flush() - Flush journal
1962 * @journal: Journal to act on.
1964 * Flush all data for a given journal to disk and empty the journal.
1965 * Filesystems can use this when remounting readonly to ensure that
1966 * recovery does not need to happen on remount.
1969 int jbd2_journal_flush(journal_t *journal)
1972 transaction_t *transaction = NULL;
1974 write_lock(&journal->j_state_lock);
1976 /* Force everything buffered to the log... */
1977 if (journal->j_running_transaction) {
1978 transaction = journal->j_running_transaction;
1979 __jbd2_log_start_commit(journal, transaction->t_tid);
1980 } else if (journal->j_committing_transaction)
1981 transaction = journal->j_committing_transaction;
1983 /* Wait for the log commit to complete... */
1985 tid_t tid = transaction->t_tid;
1987 write_unlock(&journal->j_state_lock);
1988 jbd2_log_wait_commit(journal, tid);
1990 write_unlock(&journal->j_state_lock);
1993 /* ...and flush everything in the log out to disk. */
1994 spin_lock(&journal->j_list_lock);
1995 while (!err && journal->j_checkpoint_transactions != NULL) {
1996 spin_unlock(&journal->j_list_lock);
1997 mutex_lock_io(&journal->j_checkpoint_mutex);
1998 err = jbd2_log_do_checkpoint(journal);
1999 mutex_unlock(&journal->j_checkpoint_mutex);
2000 spin_lock(&journal->j_list_lock);
2002 spin_unlock(&journal->j_list_lock);
2004 if (is_journal_aborted(journal))
2007 mutex_lock_io(&journal->j_checkpoint_mutex);
2009 err = jbd2_cleanup_journal_tail(journal);
2011 mutex_unlock(&journal->j_checkpoint_mutex);
2017 /* Finally, mark the journal as really needing no recovery.
2018 * This sets s_start==0 in the underlying superblock, which is
2019 * the magic code for a fully-recovered superblock. Any future
2020 * commits of data to the journal will restore the current
2022 jbd2_mark_journal_empty(journal, REQ_SYNC | REQ_FUA);
2023 mutex_unlock(&journal->j_checkpoint_mutex);
2024 write_lock(&journal->j_state_lock);
2025 J_ASSERT(!journal->j_running_transaction);
2026 J_ASSERT(!journal->j_committing_transaction);
2027 J_ASSERT(!journal->j_checkpoint_transactions);
2028 J_ASSERT(journal->j_head == journal->j_tail);
2029 J_ASSERT(journal->j_tail_sequence == journal->j_transaction_sequence);
2030 write_unlock(&journal->j_state_lock);
2036 * jbd2_journal_wipe() - Wipe journal contents
2037 * @journal: Journal to act on.
2038 * @write: flag (see below)
2040 * Wipe out all of the contents of a journal, safely. This will produce
2041 * a warning if the journal contains any valid recovery information.
2042 * Must be called between journal_init_*() and jbd2_journal_load().
2044 * If 'write' is non-zero, then we wipe out the journal on disk; otherwise
2045 * we merely suppress recovery.
2048 int jbd2_journal_wipe(journal_t *journal, int write)
2052 J_ASSERT (!(journal->j_flags & JBD2_LOADED));
2054 err = load_superblock(journal);
2058 if (!journal->j_tail)
2061 printk(KERN_WARNING "JBD2: %s recovery information on journal\n",
2062 write ? "Clearing" : "Ignoring");
2064 err = jbd2_journal_skip_recovery(journal);
2066 /* Lock to make assertions happy... */
2067 mutex_lock_io(&journal->j_checkpoint_mutex);
2068 jbd2_mark_journal_empty(journal, REQ_SYNC | REQ_FUA);
2069 mutex_unlock(&journal->j_checkpoint_mutex);
2077 * jbd2_journal_abort () - Shutdown the journal immediately.
2078 * @journal: the journal to shutdown.
2079 * @errno: an error number to record in the journal indicating
2080 * the reason for the shutdown.
2082 * Perform a complete, immediate shutdown of the ENTIRE
2083 * journal (not of a single transaction). This operation cannot be
2084 * undone without closing and reopening the journal.
2086 * The jbd2_journal_abort function is intended to support higher level error
2087 * recovery mechanisms such as the ext2/ext3 remount-readonly error
2090 * Journal abort has very specific semantics. Any existing dirty,
2091 * unjournaled buffers in the main filesystem will still be written to
2092 * disk by bdflush, but the journaling mechanism will be suspended
2093 * immediately and no further transaction commits will be honoured.
2095 * Any dirty, journaled buffers will be written back to disk without
2096 * hitting the journal. Atomicity cannot be guaranteed on an aborted
2097 * filesystem, but we _do_ attempt to leave as much data as possible
2098 * behind for fsck to use for cleanup.
2100 * Any attempt to get a new transaction handle on a journal which is in
2101 * ABORT state will just result in an -EROFS error return. A
2102 * jbd2_journal_stop on an existing handle will return -EIO if we have
2103 * entered abort state during the update.
2105 * Recursive transactions are not disturbed by journal abort until the
2106 * final jbd2_journal_stop, which will receive the -EIO error.
2108 * Finally, the jbd2_journal_abort call allows the caller to supply an errno
2109 * which will be recorded (if possible) in the journal superblock. This
2110 * allows a client to record failure conditions in the middle of a
2111 * transaction without having to complete the transaction to record the
2112 * failure to disk. ext3_error, for example, now uses this
2117 void jbd2_journal_abort(journal_t *journal, int errno)
2119 transaction_t *transaction;
2122 * ESHUTDOWN always takes precedence because a file system check
2123 * caused by any other journal abort error is not required after
2124 * a shutdown triggered.
2126 write_lock(&journal->j_state_lock);
2127 if (journal->j_flags & JBD2_ABORT) {
2128 int old_errno = journal->j_errno;
2130 write_unlock(&journal->j_state_lock);
2131 if (old_errno != -ESHUTDOWN && errno == -ESHUTDOWN) {
2132 journal->j_errno = errno;
2133 jbd2_journal_update_sb_errno(journal);
2139 * Mark the abort as occurred and start current running transaction
2140 * to release all journaled buffer.
2142 pr_err("Aborting journal on device %s.\n", journal->j_devname);
2144 journal->j_flags |= JBD2_ABORT;
2145 journal->j_errno = errno;
2146 transaction = journal->j_running_transaction;
2148 __jbd2_log_start_commit(journal, transaction->t_tid);
2149 write_unlock(&journal->j_state_lock);
2152 * Record errno to the journal super block, so that fsck and jbd2
2153 * layer could realise that a filesystem check is needed.
2155 jbd2_journal_update_sb_errno(journal);
2157 write_lock(&journal->j_state_lock);
2158 journal->j_flags |= JBD2_REC_ERR;
2159 write_unlock(&journal->j_state_lock);
2163 * jbd2_journal_errno() - returns the journal's error state.
2164 * @journal: journal to examine.
2166 * This is the errno number set with jbd2_journal_abort(), the last
2167 * time the journal was mounted - if the journal was stopped
2168 * without calling abort this will be 0.
2170 * If the journal has been aborted on this mount time -EROFS will
2173 int jbd2_journal_errno(journal_t *journal)
2177 read_lock(&journal->j_state_lock);
2178 if (journal->j_flags & JBD2_ABORT)
2181 err = journal->j_errno;
2182 read_unlock(&journal->j_state_lock);
2187 * jbd2_journal_clear_err() - clears the journal's error state
2188 * @journal: journal to act on.
2190 * An error must be cleared or acked to take a FS out of readonly
2193 int jbd2_journal_clear_err(journal_t *journal)
2197 write_lock(&journal->j_state_lock);
2198 if (journal->j_flags & JBD2_ABORT)
2201 journal->j_errno = 0;
2202 write_unlock(&journal->j_state_lock);
2207 * jbd2_journal_ack_err() - Ack journal err.
2208 * @journal: journal to act on.
2210 * An error must be cleared or acked to take a FS out of readonly
2213 void jbd2_journal_ack_err(journal_t *journal)
2215 write_lock(&journal->j_state_lock);
2216 if (journal->j_errno)
2217 journal->j_flags |= JBD2_ACK_ERR;
2218 write_unlock(&journal->j_state_lock);
2221 int jbd2_journal_blocks_per_page(struct inode *inode)
2223 return 1 << (PAGE_SHIFT - inode->i_sb->s_blocksize_bits);
2227 * helper functions to deal with 32 or 64bit block numbers.
2229 size_t journal_tag_bytes(journal_t *journal)
2233 if (jbd2_has_feature_csum3(journal))
2234 return sizeof(journal_block_tag3_t);
2236 sz = sizeof(journal_block_tag_t);
2238 if (jbd2_has_feature_csum2(journal))
2239 sz += sizeof(__u16);
2241 if (jbd2_has_feature_64bit(journal))
2244 return sz - sizeof(__u32);
2248 * JBD memory management
2250 * These functions are used to allocate block-sized chunks of memory
2251 * used for making copies of buffer_head data. Very often it will be
2252 * page-sized chunks of data, but sometimes it will be in
2253 * sub-page-size chunks. (For example, 16k pages on Power systems
2254 * with a 4k block file system.) For blocks smaller than a page, we
2255 * use a SLAB allocator. There are slab caches for each block size,
2256 * which are allocated at mount time, if necessary, and we only free
2257 * (all of) the slab caches when/if the jbd2 module is unloaded. For
2258 * this reason we don't need to a mutex to protect access to
2259 * jbd2_slab[] allocating or releasing memory; only in
2260 * jbd2_journal_create_slab().
2262 #define JBD2_MAX_SLABS 8
2263 static struct kmem_cache *jbd2_slab[JBD2_MAX_SLABS];
2265 static const char *jbd2_slab_names[JBD2_MAX_SLABS] = {
2266 "jbd2_1k", "jbd2_2k", "jbd2_4k", "jbd2_8k",
2267 "jbd2_16k", "jbd2_32k", "jbd2_64k", "jbd2_128k"
2271 static void jbd2_journal_destroy_slabs(void)
2275 for (i = 0; i < JBD2_MAX_SLABS; i++) {
2276 kmem_cache_destroy(jbd2_slab[i]);
2277 jbd2_slab[i] = NULL;
2281 static int jbd2_journal_create_slab(size_t size)
2283 static DEFINE_MUTEX(jbd2_slab_create_mutex);
2284 int i = order_base_2(size) - 10;
2287 if (size == PAGE_SIZE)
2290 if (i >= JBD2_MAX_SLABS)
2293 if (unlikely(i < 0))
2295 mutex_lock(&jbd2_slab_create_mutex);
2297 mutex_unlock(&jbd2_slab_create_mutex);
2298 return 0; /* Already created */
2301 slab_size = 1 << (i+10);
2302 jbd2_slab[i] = kmem_cache_create(jbd2_slab_names[i], slab_size,
2303 slab_size, 0, NULL);
2304 mutex_unlock(&jbd2_slab_create_mutex);
2305 if (!jbd2_slab[i]) {
2306 printk(KERN_EMERG "JBD2: no memory for jbd2_slab cache\n");
2312 static struct kmem_cache *get_slab(size_t size)
2314 int i = order_base_2(size) - 10;
2316 BUG_ON(i >= JBD2_MAX_SLABS);
2317 if (unlikely(i < 0))
2319 BUG_ON(jbd2_slab[i] == NULL);
2320 return jbd2_slab[i];
2323 void *jbd2_alloc(size_t size, gfp_t flags)
2327 BUG_ON(size & (size-1)); /* Must be a power of 2 */
2329 if (size < PAGE_SIZE)
2330 ptr = kmem_cache_alloc(get_slab(size), flags);
2332 ptr = (void *)__get_free_pages(flags, get_order(size));
2334 /* Check alignment; SLUB has gotten this wrong in the past,
2335 * and this can lead to user data corruption! */
2336 BUG_ON(((unsigned long) ptr) & (size-1));
2341 void jbd2_free(void *ptr, size_t size)
2343 if (size < PAGE_SIZE)
2344 kmem_cache_free(get_slab(size), ptr);
2346 free_pages((unsigned long)ptr, get_order(size));
2350 * Journal_head storage management
2352 static struct kmem_cache *jbd2_journal_head_cache;
2353 #ifdef CONFIG_JBD2_DEBUG
2354 static atomic_t nr_journal_heads = ATOMIC_INIT(0);
2357 static int __init jbd2_journal_init_journal_head_cache(void)
2359 J_ASSERT(!jbd2_journal_head_cache);
2360 jbd2_journal_head_cache = kmem_cache_create("jbd2_journal_head",
2361 sizeof(struct journal_head),
2363 SLAB_TEMPORARY | SLAB_TYPESAFE_BY_RCU,
2365 if (!jbd2_journal_head_cache) {
2366 printk(KERN_EMERG "JBD2: no memory for journal_head cache\n");
2372 static void jbd2_journal_destroy_journal_head_cache(void)
2374 kmem_cache_destroy(jbd2_journal_head_cache);
2375 jbd2_journal_head_cache = NULL;
2379 * journal_head splicing and dicing
2381 static struct journal_head *journal_alloc_journal_head(void)
2383 struct journal_head *ret;
2385 #ifdef CONFIG_JBD2_DEBUG
2386 atomic_inc(&nr_journal_heads);
2388 ret = kmem_cache_zalloc(jbd2_journal_head_cache, GFP_NOFS);
2390 jbd_debug(1, "out of memory for journal_head\n");
2391 pr_notice_ratelimited("ENOMEM in %s, retrying.\n", __func__);
2392 ret = kmem_cache_zalloc(jbd2_journal_head_cache,
2393 GFP_NOFS | __GFP_NOFAIL);
2398 static void journal_free_journal_head(struct journal_head *jh)
2400 #ifdef CONFIG_JBD2_DEBUG
2401 atomic_dec(&nr_journal_heads);
2402 memset(jh, JBD2_POISON_FREE, sizeof(*jh));
2404 kmem_cache_free(jbd2_journal_head_cache, jh);
2408 * A journal_head is attached to a buffer_head whenever JBD has an
2409 * interest in the buffer.
2411 * Whenever a buffer has an attached journal_head, its ->b_state:BH_JBD bit
2412 * is set. This bit is tested in core kernel code where we need to take
2413 * JBD-specific actions. Testing the zeroness of ->b_private is not reliable
2416 * When a buffer has its BH_JBD bit set, its ->b_count is elevated by one.
2418 * When a buffer has its BH_JBD bit set it is immune from being released by
2419 * core kernel code, mainly via ->b_count.
2421 * A journal_head is detached from its buffer_head when the journal_head's
2422 * b_jcount reaches zero. Running transaction (b_transaction) and checkpoint
2423 * transaction (b_cp_transaction) hold their references to b_jcount.
2425 * Various places in the kernel want to attach a journal_head to a buffer_head
2426 * _before_ attaching the journal_head to a transaction. To protect the
2427 * journal_head in this situation, jbd2_journal_add_journal_head elevates the
2428 * journal_head's b_jcount refcount by one. The caller must call
2429 * jbd2_journal_put_journal_head() to undo this.
2431 * So the typical usage would be:
2433 * (Attach a journal_head if needed. Increments b_jcount)
2434 * struct journal_head *jh = jbd2_journal_add_journal_head(bh);
2436 * (Get another reference for transaction)
2437 * jbd2_journal_grab_journal_head(bh);
2438 * jh->b_transaction = xxx;
2439 * (Put original reference)
2440 * jbd2_journal_put_journal_head(jh);
2444 * Give a buffer_head a journal_head.
2448 struct journal_head *jbd2_journal_add_journal_head(struct buffer_head *bh)
2450 struct journal_head *jh;
2451 struct journal_head *new_jh = NULL;
2454 if (!buffer_jbd(bh))
2455 new_jh = journal_alloc_journal_head();
2457 jbd_lock_bh_journal_head(bh);
2458 if (buffer_jbd(bh)) {
2462 (atomic_read(&bh->b_count) > 0) ||
2463 (bh->b_page && bh->b_page->mapping));
2466 jbd_unlock_bh_journal_head(bh);
2471 new_jh = NULL; /* We consumed it */
2476 BUFFER_TRACE(bh, "added journal_head");
2479 jbd_unlock_bh_journal_head(bh);
2481 journal_free_journal_head(new_jh);
2482 return bh->b_private;
2486 * Grab a ref against this buffer_head's journal_head. If it ended up not
2487 * having a journal_head, return NULL
2489 struct journal_head *jbd2_journal_grab_journal_head(struct buffer_head *bh)
2491 struct journal_head *jh = NULL;
2493 jbd_lock_bh_journal_head(bh);
2494 if (buffer_jbd(bh)) {
2498 jbd_unlock_bh_journal_head(bh);
2502 static void __journal_remove_journal_head(struct buffer_head *bh)
2504 struct journal_head *jh = bh2jh(bh);
2506 J_ASSERT_JH(jh, jh->b_jcount >= 0);
2507 J_ASSERT_JH(jh, jh->b_transaction == NULL);
2508 J_ASSERT_JH(jh, jh->b_next_transaction == NULL);
2509 J_ASSERT_JH(jh, jh->b_cp_transaction == NULL);
2510 J_ASSERT_JH(jh, jh->b_jlist == BJ_None);
2511 J_ASSERT_BH(bh, buffer_jbd(bh));
2512 J_ASSERT_BH(bh, jh2bh(jh) == bh);
2513 BUFFER_TRACE(bh, "remove journal_head");
2514 if (jh->b_frozen_data) {
2515 printk(KERN_WARNING "%s: freeing b_frozen_data\n", __func__);
2516 jbd2_free(jh->b_frozen_data, bh->b_size);
2518 if (jh->b_committed_data) {
2519 printk(KERN_WARNING "%s: freeing b_committed_data\n", __func__);
2520 jbd2_free(jh->b_committed_data, bh->b_size);
2522 bh->b_private = NULL;
2523 jh->b_bh = NULL; /* debug, really */
2524 clear_buffer_jbd(bh);
2525 journal_free_journal_head(jh);
2529 * Drop a reference on the passed journal_head. If it fell to zero then
2530 * release the journal_head from the buffer_head.
2532 void jbd2_journal_put_journal_head(struct journal_head *jh)
2534 struct buffer_head *bh = jh2bh(jh);
2536 jbd_lock_bh_journal_head(bh);
2537 J_ASSERT_JH(jh, jh->b_jcount > 0);
2539 if (!jh->b_jcount) {
2540 __journal_remove_journal_head(bh);
2541 jbd_unlock_bh_journal_head(bh);
2544 jbd_unlock_bh_journal_head(bh);
2548 * Initialize jbd inode head
2550 void jbd2_journal_init_jbd_inode(struct jbd2_inode *jinode, struct inode *inode)
2552 jinode->i_transaction = NULL;
2553 jinode->i_next_transaction = NULL;
2554 jinode->i_vfs_inode = inode;
2555 jinode->i_flags = 0;
2556 jinode->i_dirty_start = 0;
2557 jinode->i_dirty_end = 0;
2558 INIT_LIST_HEAD(&jinode->i_list);
2562 * Function to be called before we start removing inode from memory (i.e.,
2563 * clear_inode() is a fine place to be called from). It removes inode from
2564 * transaction's lists.
2566 void jbd2_journal_release_jbd_inode(journal_t *journal,
2567 struct jbd2_inode *jinode)
2572 spin_lock(&journal->j_list_lock);
2573 /* Is commit writing out inode - we have to wait */
2574 if (jinode->i_flags & JI_COMMIT_RUNNING) {
2575 wait_queue_head_t *wq;
2576 DEFINE_WAIT_BIT(wait, &jinode->i_flags, __JI_COMMIT_RUNNING);
2577 wq = bit_waitqueue(&jinode->i_flags, __JI_COMMIT_RUNNING);
2578 prepare_to_wait(wq, &wait.wq_entry, TASK_UNINTERRUPTIBLE);
2579 spin_unlock(&journal->j_list_lock);
2581 finish_wait(wq, &wait.wq_entry);
2585 if (jinode->i_transaction) {
2586 list_del(&jinode->i_list);
2587 jinode->i_transaction = NULL;
2589 spin_unlock(&journal->j_list_lock);
2593 #ifdef CONFIG_PROC_FS
2595 #define JBD2_STATS_PROC_NAME "fs/jbd2"
2597 static void __init jbd2_create_jbd_stats_proc_entry(void)
2599 proc_jbd2_stats = proc_mkdir(JBD2_STATS_PROC_NAME, NULL);
2602 static void __exit jbd2_remove_jbd_stats_proc_entry(void)
2604 if (proc_jbd2_stats)
2605 remove_proc_entry(JBD2_STATS_PROC_NAME, NULL);
2610 #define jbd2_create_jbd_stats_proc_entry() do {} while (0)
2611 #define jbd2_remove_jbd_stats_proc_entry() do {} while (0)
2615 struct kmem_cache *jbd2_handle_cache, *jbd2_inode_cache;
2617 static int __init jbd2_journal_init_inode_cache(void)
2619 J_ASSERT(!jbd2_inode_cache);
2620 jbd2_inode_cache = KMEM_CACHE(jbd2_inode, 0);
2621 if (!jbd2_inode_cache) {
2622 pr_emerg("JBD2: failed to create inode cache\n");
2628 static int __init jbd2_journal_init_handle_cache(void)
2630 J_ASSERT(!jbd2_handle_cache);
2631 jbd2_handle_cache = KMEM_CACHE(jbd2_journal_handle, SLAB_TEMPORARY);
2632 if (!jbd2_handle_cache) {
2633 printk(KERN_EMERG "JBD2: failed to create handle cache\n");
2639 static void jbd2_journal_destroy_inode_cache(void)
2641 kmem_cache_destroy(jbd2_inode_cache);
2642 jbd2_inode_cache = NULL;
2645 static void jbd2_journal_destroy_handle_cache(void)
2647 kmem_cache_destroy(jbd2_handle_cache);
2648 jbd2_handle_cache = NULL;
2652 * Module startup and shutdown
2655 static int __init journal_init_caches(void)
2659 ret = jbd2_journal_init_revoke_record_cache();
2661 ret = jbd2_journal_init_revoke_table_cache();
2663 ret = jbd2_journal_init_journal_head_cache();
2665 ret = jbd2_journal_init_handle_cache();
2667 ret = jbd2_journal_init_inode_cache();
2669 ret = jbd2_journal_init_transaction_cache();
2673 static void jbd2_journal_destroy_caches(void)
2675 jbd2_journal_destroy_revoke_record_cache();
2676 jbd2_journal_destroy_revoke_table_cache();
2677 jbd2_journal_destroy_journal_head_cache();
2678 jbd2_journal_destroy_handle_cache();
2679 jbd2_journal_destroy_inode_cache();
2680 jbd2_journal_destroy_transaction_cache();
2681 jbd2_journal_destroy_slabs();
2684 static int __init journal_init(void)
2688 BUILD_BUG_ON(sizeof(struct journal_superblock_s) != 1024);
2690 ret = journal_init_caches();
2692 jbd2_create_jbd_stats_proc_entry();
2694 jbd2_journal_destroy_caches();
2699 static void __exit journal_exit(void)
2701 #ifdef CONFIG_JBD2_DEBUG
2702 int n = atomic_read(&nr_journal_heads);
2704 printk(KERN_ERR "JBD2: leaked %d journal_heads!\n", n);
2706 jbd2_remove_jbd_stats_proc_entry();
2707 jbd2_journal_destroy_caches();
2710 MODULE_LICENSE("GPL");
2711 module_init(journal_init);
2712 module_exit(journal_exit);