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);
70 EXPORT_SYMBOL(journal_sync_buffer);
72 EXPORT_SYMBOL(jbd2_journal_flush);
73 EXPORT_SYMBOL(jbd2_journal_revoke);
75 EXPORT_SYMBOL(jbd2_journal_init_dev);
76 EXPORT_SYMBOL(jbd2_journal_init_inode);
77 EXPORT_SYMBOL(jbd2_journal_check_used_features);
78 EXPORT_SYMBOL(jbd2_journal_check_available_features);
79 EXPORT_SYMBOL(jbd2_journal_set_features);
80 EXPORT_SYMBOL(jbd2_journal_load);
81 EXPORT_SYMBOL(jbd2_journal_destroy);
82 EXPORT_SYMBOL(jbd2_journal_abort);
83 EXPORT_SYMBOL(jbd2_journal_errno);
84 EXPORT_SYMBOL(jbd2_journal_ack_err);
85 EXPORT_SYMBOL(jbd2_journal_clear_err);
86 EXPORT_SYMBOL(jbd2_log_wait_commit);
87 EXPORT_SYMBOL(jbd2_log_start_commit);
88 EXPORT_SYMBOL(jbd2_journal_start_commit);
89 EXPORT_SYMBOL(jbd2_journal_force_commit_nested);
90 EXPORT_SYMBOL(jbd2_journal_wipe);
91 EXPORT_SYMBOL(jbd2_journal_blocks_per_page);
92 EXPORT_SYMBOL(jbd2_journal_invalidatepage);
93 EXPORT_SYMBOL(jbd2_journal_try_to_free_buffers);
94 EXPORT_SYMBOL(jbd2_journal_force_commit);
95 EXPORT_SYMBOL(jbd2_journal_inode_add_write);
96 EXPORT_SYMBOL(jbd2_journal_inode_add_wait);
97 EXPORT_SYMBOL(jbd2_journal_inode_ranged_write);
98 EXPORT_SYMBOL(jbd2_journal_inode_ranged_wait);
99 EXPORT_SYMBOL(jbd2_journal_init_jbd_inode);
100 EXPORT_SYMBOL(jbd2_journal_release_jbd_inode);
101 EXPORT_SYMBOL(jbd2_journal_begin_ordered_truncate);
102 EXPORT_SYMBOL(jbd2_inode_cache);
104 static void __journal_abort_soft (journal_t *journal, int errno);
105 static int jbd2_journal_create_slab(size_t slab_size);
107 #ifdef CONFIG_JBD2_DEBUG
108 void __jbd2_debug(int level, const char *file, const char *func,
109 unsigned int line, const char *fmt, ...)
111 struct va_format vaf;
114 if (level > jbd2_journal_enable_debug)
119 printk(KERN_DEBUG "%s: (%s, %u): %pV", file, func, line, &vaf);
122 EXPORT_SYMBOL(__jbd2_debug);
125 /* Checksumming functions */
126 static int jbd2_verify_csum_type(journal_t *j, journal_superblock_t *sb)
128 if (!jbd2_journal_has_csum_v2or3_feature(j))
131 return sb->s_checksum_type == JBD2_CRC32C_CHKSUM;
134 static __be32 jbd2_superblock_csum(journal_t *j, journal_superblock_t *sb)
139 old_csum = sb->s_checksum;
141 csum = jbd2_chksum(j, ~0, (char *)sb, sizeof(journal_superblock_t));
142 sb->s_checksum = old_csum;
144 return cpu_to_be32(csum);
147 static int jbd2_superblock_csum_verify(journal_t *j, journal_superblock_t *sb)
149 if (!jbd2_journal_has_csum_v2or3(j))
152 return sb->s_checksum == jbd2_superblock_csum(j, sb);
155 static void jbd2_superblock_csum_set(journal_t *j, journal_superblock_t *sb)
157 if (!jbd2_journal_has_csum_v2or3(j))
160 sb->s_checksum = jbd2_superblock_csum(j, sb);
164 * Helper function used to manage commit timeouts
167 static void commit_timeout(struct timer_list *t)
169 journal_t *journal = from_timer(journal, t, j_commit_timer);
171 wake_up_process(journal->j_task);
175 * kjournald2: The main thread function used to manage a logging device
178 * This kernel thread is responsible for two things:
180 * 1) COMMIT: Every so often we need to commit the current state of the
181 * filesystem to disk. The journal thread is responsible for writing
182 * all of the metadata buffers to disk.
184 * 2) CHECKPOINT: We cannot reuse a used section of the log file until all
185 * of the data in that part of the log has been rewritten elsewhere on
186 * the disk. Flushing these old buffers to reclaim space in the log is
187 * known as checkpointing, and this thread is responsible for that job.
190 static int kjournald2(void *arg)
192 journal_t *journal = arg;
193 transaction_t *transaction;
196 * Set up an interval timer which can be used to trigger a commit wakeup
197 * after the commit interval expires
199 timer_setup(&journal->j_commit_timer, commit_timeout, 0);
203 /* Record that the journal thread is running */
204 journal->j_task = current;
205 wake_up(&journal->j_wait_done_commit);
208 * Make sure that no allocations from this kernel thread will ever
209 * recurse to the fs layer because we are responsible for the
210 * transaction commit and any fs involvement might get stuck waiting for
213 memalloc_nofs_save();
216 * And now, wait forever for commit wakeup events.
218 write_lock(&journal->j_state_lock);
221 if (journal->j_flags & JBD2_UNMOUNT)
224 jbd_debug(1, "commit_sequence=%d, commit_request=%d\n",
225 journal->j_commit_sequence, journal->j_commit_request);
227 if (journal->j_commit_sequence != journal->j_commit_request) {
228 jbd_debug(1, "OK, requests differ\n");
229 write_unlock(&journal->j_state_lock);
230 del_timer_sync(&journal->j_commit_timer);
231 jbd2_journal_commit_transaction(journal);
232 write_lock(&journal->j_state_lock);
236 wake_up(&journal->j_wait_done_commit);
237 if (freezing(current)) {
239 * The simpler the better. Flushing journal isn't a
240 * good idea, because that depends on threads that may
241 * be already stopped.
243 jbd_debug(1, "Now suspending kjournald2\n");
244 write_unlock(&journal->j_state_lock);
246 write_lock(&journal->j_state_lock);
249 * We assume on resume that commits are already there,
253 int should_sleep = 1;
255 prepare_to_wait(&journal->j_wait_commit, &wait,
257 if (journal->j_commit_sequence != journal->j_commit_request)
259 transaction = journal->j_running_transaction;
260 if (transaction && time_after_eq(jiffies,
261 transaction->t_expires))
263 if (journal->j_flags & JBD2_UNMOUNT)
266 write_unlock(&journal->j_state_lock);
268 write_lock(&journal->j_state_lock);
270 finish_wait(&journal->j_wait_commit, &wait);
273 jbd_debug(1, "kjournald2 wakes\n");
276 * Were we woken up by a commit wakeup event?
278 transaction = journal->j_running_transaction;
279 if (transaction && time_after_eq(jiffies, transaction->t_expires)) {
280 journal->j_commit_request = transaction->t_tid;
281 jbd_debug(1, "woke because of timeout\n");
286 del_timer_sync(&journal->j_commit_timer);
287 journal->j_task = NULL;
288 wake_up(&journal->j_wait_done_commit);
289 jbd_debug(1, "Journal thread exiting.\n");
290 write_unlock(&journal->j_state_lock);
294 static int jbd2_journal_start_thread(journal_t *journal)
296 struct task_struct *t;
298 t = kthread_run(kjournald2, journal, "jbd2/%s",
303 wait_event(journal->j_wait_done_commit, journal->j_task != NULL);
307 static void journal_kill_thread(journal_t *journal)
309 write_lock(&journal->j_state_lock);
310 journal->j_flags |= JBD2_UNMOUNT;
312 while (journal->j_task) {
313 write_unlock(&journal->j_state_lock);
314 wake_up(&journal->j_wait_commit);
315 wait_event(journal->j_wait_done_commit, journal->j_task == NULL);
316 write_lock(&journal->j_state_lock);
318 write_unlock(&journal->j_state_lock);
322 * jbd2_journal_write_metadata_buffer: write a metadata buffer to the journal.
324 * Writes a metadata buffer to a given disk block. The actual IO is not
325 * performed but a new buffer_head is constructed which labels the data
326 * to be written with the correct destination disk block.
328 * Any magic-number escaping which needs to be done will cause a
329 * copy-out here. If the buffer happens to start with the
330 * JBD2_MAGIC_NUMBER, then we can't write it to the log directly: the
331 * magic number is only written to the log for descripter blocks. In
332 * this case, we copy the data and replace the first word with 0, and we
333 * return a result code which indicates that this buffer needs to be
334 * marked as an escaped buffer in the corresponding log descriptor
335 * block. The missing word can then be restored when the block is read
338 * If the source buffer has already been modified by a new transaction
339 * since we took the last commit snapshot, we use the frozen copy of
340 * that data for IO. If we end up using the existing buffer_head's data
341 * for the write, then we have to make sure nobody modifies it while the
342 * IO is in progress. do_get_write_access() handles this.
344 * The function returns a pointer to the buffer_head to be used for IO.
352 * Bit 0 set == escape performed on the data
353 * Bit 1 set == buffer copy-out performed (kfree the data after IO)
356 int jbd2_journal_write_metadata_buffer(transaction_t *transaction,
357 struct journal_head *jh_in,
358 struct buffer_head **bh_out,
361 int need_copy_out = 0;
362 int done_copy_out = 0;
365 struct buffer_head *new_bh;
366 struct page *new_page;
367 unsigned int new_offset;
368 struct buffer_head *bh_in = jh2bh(jh_in);
369 journal_t *journal = transaction->t_journal;
372 * The buffer really shouldn't be locked: only the current committing
373 * transaction is allowed to write it, so nobody else is allowed
376 * akpm: except if we're journalling data, and write() output is
377 * also part of a shared mapping, and another thread has
378 * decided to launch a writepage() against this buffer.
380 J_ASSERT_BH(bh_in, buffer_jbddirty(bh_in));
382 new_bh = alloc_buffer_head(GFP_NOFS|__GFP_NOFAIL);
384 /* keep subsequent assertions sane */
385 atomic_set(&new_bh->b_count, 1);
387 jbd_lock_bh_state(bh_in);
390 * If a new transaction has already done a buffer copy-out, then
391 * we use that version of the data for the commit.
393 if (jh_in->b_frozen_data) {
395 new_page = virt_to_page(jh_in->b_frozen_data);
396 new_offset = offset_in_page(jh_in->b_frozen_data);
398 new_page = jh2bh(jh_in)->b_page;
399 new_offset = offset_in_page(jh2bh(jh_in)->b_data);
402 mapped_data = kmap_atomic(new_page);
404 * Fire data frozen trigger if data already wasn't frozen. Do this
405 * before checking for escaping, as the trigger may modify the magic
406 * offset. If a copy-out happens afterwards, it will have the correct
407 * data in the buffer.
410 jbd2_buffer_frozen_trigger(jh_in, mapped_data + new_offset,
416 if (*((__be32 *)(mapped_data + new_offset)) ==
417 cpu_to_be32(JBD2_MAGIC_NUMBER)) {
421 kunmap_atomic(mapped_data);
424 * Do we need to do a data copy?
426 if (need_copy_out && !done_copy_out) {
429 jbd_unlock_bh_state(bh_in);
430 tmp = jbd2_alloc(bh_in->b_size, GFP_NOFS);
435 jbd_lock_bh_state(bh_in);
436 if (jh_in->b_frozen_data) {
437 jbd2_free(tmp, bh_in->b_size);
441 jh_in->b_frozen_data = tmp;
442 mapped_data = kmap_atomic(new_page);
443 memcpy(tmp, mapped_data + new_offset, bh_in->b_size);
444 kunmap_atomic(mapped_data);
446 new_page = virt_to_page(tmp);
447 new_offset = offset_in_page(tmp);
451 * This isn't strictly necessary, as we're using frozen
452 * data for the escaping, but it keeps consistency with
453 * b_frozen_data usage.
455 jh_in->b_frozen_triggers = jh_in->b_triggers;
459 * Did we need to do an escaping? Now we've done all the
460 * copying, we can finally do so.
463 mapped_data = kmap_atomic(new_page);
464 *((unsigned int *)(mapped_data + new_offset)) = 0;
465 kunmap_atomic(mapped_data);
468 set_bh_page(new_bh, new_page, new_offset);
469 new_bh->b_size = bh_in->b_size;
470 new_bh->b_bdev = journal->j_dev;
471 new_bh->b_blocknr = blocknr;
472 new_bh->b_private = bh_in;
473 set_buffer_mapped(new_bh);
474 set_buffer_dirty(new_bh);
479 * The to-be-written buffer needs to get moved to the io queue,
480 * and the original buffer whose contents we are shadowing or
481 * copying is moved to the transaction's shadow queue.
483 JBUFFER_TRACE(jh_in, "file as BJ_Shadow");
484 spin_lock(&journal->j_list_lock);
485 __jbd2_journal_file_buffer(jh_in, transaction, BJ_Shadow);
486 spin_unlock(&journal->j_list_lock);
487 set_buffer_shadow(bh_in);
488 jbd_unlock_bh_state(bh_in);
490 return do_escape | (done_copy_out << 1);
494 * Allocation code for the journal file. Manage the space left in the
495 * journal, so that we can begin checkpointing when appropriate.
499 * Called with j_state_lock locked for writing.
500 * Returns true if a transaction commit was started.
502 int __jbd2_log_start_commit(journal_t *journal, tid_t target)
504 /* Return if the txn has already requested to be committed */
505 if (journal->j_commit_request == target)
509 * The only transaction we can possibly wait upon is the
510 * currently running transaction (if it exists). Otherwise,
511 * the target tid must be an old one.
513 if (journal->j_running_transaction &&
514 journal->j_running_transaction->t_tid == target) {
516 * We want a new commit: OK, mark the request and wakeup the
517 * commit thread. We do _not_ do the commit ourselves.
520 journal->j_commit_request = target;
521 jbd_debug(1, "JBD2: requesting commit %d/%d\n",
522 journal->j_commit_request,
523 journal->j_commit_sequence);
524 journal->j_running_transaction->t_requested = jiffies;
525 wake_up(&journal->j_wait_commit);
527 } else if (!tid_geq(journal->j_commit_request, target))
528 /* This should never happen, but if it does, preserve
529 the evidence before kjournald goes into a loop and
530 increments j_commit_sequence beyond all recognition. */
531 WARN_ONCE(1, "JBD2: bad log_start_commit: %u %u %u %u\n",
532 journal->j_commit_request,
533 journal->j_commit_sequence,
534 target, journal->j_running_transaction ?
535 journal->j_running_transaction->t_tid : 0);
539 int jbd2_log_start_commit(journal_t *journal, tid_t tid)
543 write_lock(&journal->j_state_lock);
544 ret = __jbd2_log_start_commit(journal, tid);
545 write_unlock(&journal->j_state_lock);
550 * Force and wait any uncommitted transactions. We can only force the running
551 * transaction if we don't have an active handle, otherwise, we will deadlock.
552 * Returns: <0 in case of error,
553 * 0 if nothing to commit,
554 * 1 if transaction was successfully committed.
556 static int __jbd2_journal_force_commit(journal_t *journal)
558 transaction_t *transaction = NULL;
560 int need_to_start = 0, ret = 0;
562 read_lock(&journal->j_state_lock);
563 if (journal->j_running_transaction && !current->journal_info) {
564 transaction = journal->j_running_transaction;
565 if (!tid_geq(journal->j_commit_request, transaction->t_tid))
567 } else if (journal->j_committing_transaction)
568 transaction = journal->j_committing_transaction;
571 /* Nothing to commit */
572 read_unlock(&journal->j_state_lock);
575 tid = transaction->t_tid;
576 read_unlock(&journal->j_state_lock);
578 jbd2_log_start_commit(journal, tid);
579 ret = jbd2_log_wait_commit(journal, tid);
587 * Force and wait upon a commit if the calling process is not within
588 * transaction. This is used for forcing out undo-protected data which contains
589 * bitmaps, when the fs is running out of space.
591 * @journal: journal to force
592 * Returns true if progress was made.
594 int jbd2_journal_force_commit_nested(journal_t *journal)
598 ret = __jbd2_journal_force_commit(journal);
603 * int journal_force_commit() - force any uncommitted transactions
604 * @journal: journal to force
606 * Caller want unconditional commit. We can only force the running transaction
607 * if we don't have an active handle, otherwise, we will deadlock.
609 int jbd2_journal_force_commit(journal_t *journal)
613 J_ASSERT(!current->journal_info);
614 ret = __jbd2_journal_force_commit(journal);
621 * Start a commit of the current running transaction (if any). Returns true
622 * if a transaction is going to be committed (or is currently already
623 * committing), and fills its tid in at *ptid
625 int jbd2_journal_start_commit(journal_t *journal, tid_t *ptid)
629 write_lock(&journal->j_state_lock);
630 if (journal->j_running_transaction) {
631 tid_t tid = journal->j_running_transaction->t_tid;
633 __jbd2_log_start_commit(journal, tid);
634 /* There's a running transaction and we've just made sure
635 * it's commit has been scheduled. */
639 } else if (journal->j_committing_transaction) {
641 * If commit has been started, then we have to wait for
642 * completion of that transaction.
645 *ptid = journal->j_committing_transaction->t_tid;
648 write_unlock(&journal->j_state_lock);
653 * Return 1 if a given transaction has not yet sent barrier request
654 * connected with a transaction commit. If 0 is returned, transaction
655 * may or may not have sent the barrier. Used to avoid sending barrier
656 * twice in common cases.
658 int jbd2_trans_will_send_data_barrier(journal_t *journal, tid_t tid)
661 transaction_t *commit_trans;
663 if (!(journal->j_flags & JBD2_BARRIER))
665 read_lock(&journal->j_state_lock);
666 /* Transaction already committed? */
667 if (tid_geq(journal->j_commit_sequence, tid))
669 commit_trans = journal->j_committing_transaction;
670 if (!commit_trans || commit_trans->t_tid != tid) {
675 * Transaction is being committed and we already proceeded to
676 * submitting a flush to fs partition?
678 if (journal->j_fs_dev != journal->j_dev) {
679 if (!commit_trans->t_need_data_flush ||
680 commit_trans->t_state >= T_COMMIT_DFLUSH)
683 if (commit_trans->t_state >= T_COMMIT_JFLUSH)
688 read_unlock(&journal->j_state_lock);
691 EXPORT_SYMBOL(jbd2_trans_will_send_data_barrier);
694 * Wait for a specified commit to complete.
695 * The caller may not hold the journal lock.
697 int jbd2_log_wait_commit(journal_t *journal, tid_t tid)
701 read_lock(&journal->j_state_lock);
702 #ifdef CONFIG_PROVE_LOCKING
704 * Some callers make sure transaction is already committing and in that
705 * case we cannot block on open handles anymore. So don't warn in that
708 if (tid_gt(tid, journal->j_commit_sequence) &&
709 (!journal->j_committing_transaction ||
710 journal->j_committing_transaction->t_tid != tid)) {
711 read_unlock(&journal->j_state_lock);
712 jbd2_might_wait_for_commit(journal);
713 read_lock(&journal->j_state_lock);
716 #ifdef CONFIG_JBD2_DEBUG
717 if (!tid_geq(journal->j_commit_request, tid)) {
719 "%s: error: j_commit_request=%d, tid=%d\n",
720 __func__, journal->j_commit_request, tid);
723 while (tid_gt(tid, journal->j_commit_sequence)) {
724 jbd_debug(1, "JBD2: want %d, j_commit_sequence=%d\n",
725 tid, journal->j_commit_sequence);
726 read_unlock(&journal->j_state_lock);
727 wake_up(&journal->j_wait_commit);
728 wait_event(journal->j_wait_done_commit,
729 !tid_gt(tid, journal->j_commit_sequence));
730 read_lock(&journal->j_state_lock);
732 read_unlock(&journal->j_state_lock);
734 if (unlikely(is_journal_aborted(journal)))
739 /* Return 1 when transaction with given tid has already committed. */
740 int jbd2_transaction_committed(journal_t *journal, tid_t tid)
744 read_lock(&journal->j_state_lock);
745 if (journal->j_running_transaction &&
746 journal->j_running_transaction->t_tid == tid)
748 if (journal->j_committing_transaction &&
749 journal->j_committing_transaction->t_tid == tid)
751 read_unlock(&journal->j_state_lock);
754 EXPORT_SYMBOL(jbd2_transaction_committed);
757 * When this function returns the transaction corresponding to tid
758 * will be completed. If the transaction has currently running, start
759 * committing that transaction before waiting for it to complete. If
760 * the transaction id is stale, it is by definition already completed,
761 * so just return SUCCESS.
763 int jbd2_complete_transaction(journal_t *journal, tid_t tid)
765 int need_to_wait = 1;
767 read_lock(&journal->j_state_lock);
768 if (journal->j_running_transaction &&
769 journal->j_running_transaction->t_tid == tid) {
770 if (journal->j_commit_request != tid) {
771 /* transaction not yet started, so request it */
772 read_unlock(&journal->j_state_lock);
773 jbd2_log_start_commit(journal, tid);
776 } else if (!(journal->j_committing_transaction &&
777 journal->j_committing_transaction->t_tid == tid))
779 read_unlock(&journal->j_state_lock);
783 return jbd2_log_wait_commit(journal, tid);
785 EXPORT_SYMBOL(jbd2_complete_transaction);
788 * Log buffer allocation routines:
791 int jbd2_journal_next_log_block(journal_t *journal, unsigned long long *retp)
793 unsigned long blocknr;
795 write_lock(&journal->j_state_lock);
796 J_ASSERT(journal->j_free > 1);
798 blocknr = journal->j_head;
801 if (journal->j_head == journal->j_last)
802 journal->j_head = journal->j_first;
803 write_unlock(&journal->j_state_lock);
804 return jbd2_journal_bmap(journal, blocknr, retp);
808 * Conversion of logical to physical block numbers for the journal
810 * On external journals the journal blocks are identity-mapped, so
811 * this is a no-op. If needed, we can use j_blk_offset - everything is
814 int jbd2_journal_bmap(journal_t *journal, unsigned long blocknr,
815 unsigned long long *retp)
818 unsigned long long ret;
820 if (journal->j_inode) {
821 ret = bmap(journal->j_inode, blocknr);
825 printk(KERN_ALERT "%s: journal block not found "
826 "at offset %lu on %s\n",
827 __func__, blocknr, journal->j_devname);
829 __journal_abort_soft(journal, err);
832 *retp = blocknr; /* +journal->j_blk_offset */
838 * We play buffer_head aliasing tricks to write data/metadata blocks to
839 * the journal without copying their contents, but for journal
840 * descriptor blocks we do need to generate bona fide buffers.
842 * After the caller of jbd2_journal_get_descriptor_buffer() has finished modifying
843 * the buffer's contents they really should run flush_dcache_page(bh->b_page).
844 * But we don't bother doing that, so there will be coherency problems with
845 * mmaps of blockdevs which hold live JBD-controlled filesystems.
848 jbd2_journal_get_descriptor_buffer(transaction_t *transaction, int type)
850 journal_t *journal = transaction->t_journal;
851 struct buffer_head *bh;
852 unsigned long long blocknr;
853 journal_header_t *header;
856 err = jbd2_journal_next_log_block(journal, &blocknr);
861 bh = __getblk(journal->j_dev, blocknr, journal->j_blocksize);
865 memset(bh->b_data, 0, journal->j_blocksize);
866 header = (journal_header_t *)bh->b_data;
867 header->h_magic = cpu_to_be32(JBD2_MAGIC_NUMBER);
868 header->h_blocktype = cpu_to_be32(type);
869 header->h_sequence = cpu_to_be32(transaction->t_tid);
870 set_buffer_uptodate(bh);
872 BUFFER_TRACE(bh, "return this buffer");
876 void jbd2_descriptor_block_csum_set(journal_t *j, struct buffer_head *bh)
878 struct jbd2_journal_block_tail *tail;
881 if (!jbd2_journal_has_csum_v2or3(j))
884 tail = (struct jbd2_journal_block_tail *)(bh->b_data + j->j_blocksize -
885 sizeof(struct jbd2_journal_block_tail));
886 tail->t_checksum = 0;
887 csum = jbd2_chksum(j, j->j_csum_seed, bh->b_data, j->j_blocksize);
888 tail->t_checksum = cpu_to_be32(csum);
892 * Return tid of the oldest transaction in the journal and block in the journal
893 * where the transaction starts.
895 * If the journal is now empty, return which will be the next transaction ID
896 * we will write and where will that transaction start.
898 * The return value is 0 if journal tail cannot be pushed any further, 1 if
901 int jbd2_journal_get_log_tail(journal_t *journal, tid_t *tid,
902 unsigned long *block)
904 transaction_t *transaction;
907 read_lock(&journal->j_state_lock);
908 spin_lock(&journal->j_list_lock);
909 transaction = journal->j_checkpoint_transactions;
911 *tid = transaction->t_tid;
912 *block = transaction->t_log_start;
913 } else if ((transaction = journal->j_committing_transaction) != NULL) {
914 *tid = transaction->t_tid;
915 *block = transaction->t_log_start;
916 } else if ((transaction = journal->j_running_transaction) != NULL) {
917 *tid = transaction->t_tid;
918 *block = journal->j_head;
920 *tid = journal->j_transaction_sequence;
921 *block = journal->j_head;
923 ret = tid_gt(*tid, journal->j_tail_sequence);
924 spin_unlock(&journal->j_list_lock);
925 read_unlock(&journal->j_state_lock);
931 * Update information in journal structure and in on disk journal superblock
932 * about log tail. This function does not check whether information passed in
933 * really pushes log tail further. It's responsibility of the caller to make
934 * sure provided log tail information is valid (e.g. by holding
935 * j_checkpoint_mutex all the time between computing log tail and calling this
936 * function as is the case with jbd2_cleanup_journal_tail()).
938 * Requires j_checkpoint_mutex
940 int __jbd2_update_log_tail(journal_t *journal, tid_t tid, unsigned long block)
945 BUG_ON(!mutex_is_locked(&journal->j_checkpoint_mutex));
948 * We cannot afford for write to remain in drive's caches since as
949 * soon as we update j_tail, next transaction can start reusing journal
950 * space and if we lose sb update during power failure we'd replay
951 * old transaction with possibly newly overwritten data.
953 ret = jbd2_journal_update_sb_log_tail(journal, tid, block,
958 write_lock(&journal->j_state_lock);
959 freed = block - journal->j_tail;
960 if (block < journal->j_tail)
961 freed += journal->j_last - journal->j_first;
963 trace_jbd2_update_log_tail(journal, tid, block, freed);
965 "Cleaning journal tail from %d to %d (offset %lu), "
967 journal->j_tail_sequence, tid, block, freed);
969 journal->j_free += freed;
970 journal->j_tail_sequence = tid;
971 journal->j_tail = block;
972 write_unlock(&journal->j_state_lock);
979 * This is a variation of __jbd2_update_log_tail which checks for validity of
980 * provided log tail and locks j_checkpoint_mutex. So it is safe against races
981 * with other threads updating log tail.
983 void jbd2_update_log_tail(journal_t *journal, tid_t tid, unsigned long block)
985 mutex_lock_io(&journal->j_checkpoint_mutex);
986 if (tid_gt(tid, journal->j_tail_sequence))
987 __jbd2_update_log_tail(journal, tid, block);
988 mutex_unlock(&journal->j_checkpoint_mutex);
991 struct jbd2_stats_proc_session {
993 struct transaction_stats_s *stats;
998 static void *jbd2_seq_info_start(struct seq_file *seq, loff_t *pos)
1000 return *pos ? NULL : SEQ_START_TOKEN;
1003 static void *jbd2_seq_info_next(struct seq_file *seq, void *v, loff_t *pos)
1009 static int jbd2_seq_info_show(struct seq_file *seq, void *v)
1011 struct jbd2_stats_proc_session *s = seq->private;
1013 if (v != SEQ_START_TOKEN)
1015 seq_printf(seq, "%lu transactions (%lu requested), "
1016 "each up to %u blocks\n",
1017 s->stats->ts_tid, s->stats->ts_requested,
1018 s->journal->j_max_transaction_buffers);
1019 if (s->stats->ts_tid == 0)
1021 seq_printf(seq, "average: \n %ums waiting for transaction\n",
1022 jiffies_to_msecs(s->stats->run.rs_wait / s->stats->ts_tid));
1023 seq_printf(seq, " %ums request delay\n",
1024 (s->stats->ts_requested == 0) ? 0 :
1025 jiffies_to_msecs(s->stats->run.rs_request_delay /
1026 s->stats->ts_requested));
1027 seq_printf(seq, " %ums running transaction\n",
1028 jiffies_to_msecs(s->stats->run.rs_running / s->stats->ts_tid));
1029 seq_printf(seq, " %ums transaction was being locked\n",
1030 jiffies_to_msecs(s->stats->run.rs_locked / s->stats->ts_tid));
1031 seq_printf(seq, " %ums flushing data (in ordered mode)\n",
1032 jiffies_to_msecs(s->stats->run.rs_flushing / s->stats->ts_tid));
1033 seq_printf(seq, " %ums logging transaction\n",
1034 jiffies_to_msecs(s->stats->run.rs_logging / s->stats->ts_tid));
1035 seq_printf(seq, " %lluus average transaction commit time\n",
1036 div_u64(s->journal->j_average_commit_time, 1000));
1037 seq_printf(seq, " %lu handles per transaction\n",
1038 s->stats->run.rs_handle_count / s->stats->ts_tid);
1039 seq_printf(seq, " %lu blocks per transaction\n",
1040 s->stats->run.rs_blocks / s->stats->ts_tid);
1041 seq_printf(seq, " %lu logged blocks per transaction\n",
1042 s->stats->run.rs_blocks_logged / s->stats->ts_tid);
1046 static void jbd2_seq_info_stop(struct seq_file *seq, void *v)
1050 static const struct seq_operations jbd2_seq_info_ops = {
1051 .start = jbd2_seq_info_start,
1052 .next = jbd2_seq_info_next,
1053 .stop = jbd2_seq_info_stop,
1054 .show = jbd2_seq_info_show,
1057 static int jbd2_seq_info_open(struct inode *inode, struct file *file)
1059 journal_t *journal = PDE_DATA(inode);
1060 struct jbd2_stats_proc_session *s;
1063 s = kmalloc(sizeof(*s), GFP_KERNEL);
1066 size = sizeof(struct transaction_stats_s);
1067 s->stats = kmalloc(size, GFP_KERNEL);
1068 if (s->stats == NULL) {
1072 spin_lock(&journal->j_history_lock);
1073 memcpy(s->stats, &journal->j_stats, size);
1074 s->journal = journal;
1075 spin_unlock(&journal->j_history_lock);
1077 rc = seq_open(file, &jbd2_seq_info_ops);
1079 struct seq_file *m = file->private_data;
1089 static int jbd2_seq_info_release(struct inode *inode, struct file *file)
1091 struct seq_file *seq = file->private_data;
1092 struct jbd2_stats_proc_session *s = seq->private;
1095 return seq_release(inode, file);
1098 static const struct file_operations jbd2_seq_info_fops = {
1099 .owner = THIS_MODULE,
1100 .open = jbd2_seq_info_open,
1102 .llseek = seq_lseek,
1103 .release = jbd2_seq_info_release,
1106 static struct proc_dir_entry *proc_jbd2_stats;
1108 static void jbd2_stats_proc_init(journal_t *journal)
1110 journal->j_proc_entry = proc_mkdir(journal->j_devname, proc_jbd2_stats);
1111 if (journal->j_proc_entry) {
1112 proc_create_data("info", S_IRUGO, journal->j_proc_entry,
1113 &jbd2_seq_info_fops, journal);
1117 static void jbd2_stats_proc_exit(journal_t *journal)
1119 remove_proc_entry("info", journal->j_proc_entry);
1120 remove_proc_entry(journal->j_devname, proc_jbd2_stats);
1124 * Management for journal control blocks: functions to create and
1125 * destroy journal_t structures, and to initialise and read existing
1126 * journal blocks from disk. */
1128 /* First: create and setup a journal_t object in memory. We initialise
1129 * very few fields yet: that has to wait until we have created the
1130 * journal structures from from scratch, or loaded them from disk. */
1132 static journal_t *journal_init_common(struct block_device *bdev,
1133 struct block_device *fs_dev,
1134 unsigned long long start, int len, int blocksize)
1136 static struct lock_class_key jbd2_trans_commit_key;
1139 struct buffer_head *bh;
1142 journal = kzalloc(sizeof(*journal), GFP_KERNEL);
1146 init_waitqueue_head(&journal->j_wait_transaction_locked);
1147 init_waitqueue_head(&journal->j_wait_done_commit);
1148 init_waitqueue_head(&journal->j_wait_commit);
1149 init_waitqueue_head(&journal->j_wait_updates);
1150 init_waitqueue_head(&journal->j_wait_reserved);
1151 mutex_init(&journal->j_barrier);
1152 mutex_init(&journal->j_checkpoint_mutex);
1153 spin_lock_init(&journal->j_revoke_lock);
1154 spin_lock_init(&journal->j_list_lock);
1155 rwlock_init(&journal->j_state_lock);
1157 journal->j_commit_interval = (HZ * JBD2_DEFAULT_MAX_COMMIT_AGE);
1158 journal->j_min_batch_time = 0;
1159 journal->j_max_batch_time = 15000; /* 15ms */
1160 atomic_set(&journal->j_reserved_credits, 0);
1162 /* The journal is marked for error until we succeed with recovery! */
1163 journal->j_flags = JBD2_ABORT;
1165 /* Set up a default-sized revoke table for the new mount. */
1166 err = jbd2_journal_init_revoke(journal, JOURNAL_REVOKE_DEFAULT_HASH);
1170 spin_lock_init(&journal->j_history_lock);
1172 lockdep_init_map(&journal->j_trans_commit_map, "jbd2_handle",
1173 &jbd2_trans_commit_key, 0);
1175 /* journal descriptor can store up to n blocks -bzzz */
1176 journal->j_blocksize = blocksize;
1177 journal->j_dev = bdev;
1178 journal->j_fs_dev = fs_dev;
1179 journal->j_blk_offset = start;
1180 journal->j_maxlen = len;
1181 n = journal->j_blocksize / sizeof(journal_block_tag_t);
1182 journal->j_wbufsize = n;
1183 journal->j_wbuf = kmalloc_array(n, sizeof(struct buffer_head *),
1185 if (!journal->j_wbuf)
1188 bh = getblk_unmovable(journal->j_dev, start, journal->j_blocksize);
1190 pr_err("%s: Cannot get buffer for journal superblock\n",
1194 journal->j_sb_buffer = bh;
1195 journal->j_superblock = (journal_superblock_t *)bh->b_data;
1200 kfree(journal->j_wbuf);
1201 jbd2_journal_destroy_revoke(journal);
1206 /* jbd2_journal_init_dev and jbd2_journal_init_inode:
1208 * Create a journal structure assigned some fixed set of disk blocks to
1209 * the journal. We don't actually touch those disk blocks yet, but we
1210 * need to set up all of the mapping information to tell the journaling
1211 * system where the journal blocks are.
1216 * journal_t * jbd2_journal_init_dev() - creates and initialises a journal structure
1217 * @bdev: Block device on which to create the journal
1218 * @fs_dev: Device which hold journalled filesystem for this journal.
1219 * @start: Block nr Start of journal.
1220 * @len: Length of the journal in blocks.
1221 * @blocksize: blocksize of journalling device
1223 * Returns: a newly created journal_t *
1225 * jbd2_journal_init_dev creates a journal which maps a fixed contiguous
1226 * range of blocks on an arbitrary block device.
1229 journal_t *jbd2_journal_init_dev(struct block_device *bdev,
1230 struct block_device *fs_dev,
1231 unsigned long long start, int len, int blocksize)
1235 journal = journal_init_common(bdev, fs_dev, start, len, blocksize);
1239 bdevname(journal->j_dev, journal->j_devname);
1240 strreplace(journal->j_devname, '/', '!');
1241 jbd2_stats_proc_init(journal);
1247 * journal_t * jbd2_journal_init_inode () - creates a journal which maps to a inode.
1248 * @inode: An inode to create the journal in
1250 * jbd2_journal_init_inode creates a journal which maps an on-disk inode as
1251 * the journal. The inode must exist already, must support bmap() and
1252 * must have all data blocks preallocated.
1254 journal_t *jbd2_journal_init_inode(struct inode *inode)
1258 unsigned long long blocknr;
1260 blocknr = bmap(inode, 0);
1262 pr_err("%s: Cannot locate journal superblock\n",
1267 jbd_debug(1, "JBD2: inode %s/%ld, size %lld, bits %d, blksize %ld\n",
1268 inode->i_sb->s_id, inode->i_ino, (long long) inode->i_size,
1269 inode->i_sb->s_blocksize_bits, inode->i_sb->s_blocksize);
1271 journal = journal_init_common(inode->i_sb->s_bdev, inode->i_sb->s_bdev,
1272 blocknr, inode->i_size >> inode->i_sb->s_blocksize_bits,
1273 inode->i_sb->s_blocksize);
1277 journal->j_inode = inode;
1278 bdevname(journal->j_dev, journal->j_devname);
1279 p = strreplace(journal->j_devname, '/', '!');
1280 sprintf(p, "-%lu", journal->j_inode->i_ino);
1281 jbd2_stats_proc_init(journal);
1287 * If the journal init or create aborts, we need to mark the journal
1288 * superblock as being NULL to prevent the journal destroy from writing
1289 * back a bogus superblock.
1291 static void journal_fail_superblock (journal_t *journal)
1293 struct buffer_head *bh = journal->j_sb_buffer;
1295 journal->j_sb_buffer = NULL;
1299 * Given a journal_t structure, initialise the various fields for
1300 * startup of a new journaling session. We use this both when creating
1301 * a journal, and after recovering an old journal to reset it for
1305 static int journal_reset(journal_t *journal)
1307 journal_superblock_t *sb = journal->j_superblock;
1308 unsigned long long first, last;
1310 first = be32_to_cpu(sb->s_first);
1311 last = be32_to_cpu(sb->s_maxlen);
1312 if (first + JBD2_MIN_JOURNAL_BLOCKS > last + 1) {
1313 printk(KERN_ERR "JBD2: Journal too short (blocks %llu-%llu).\n",
1315 journal_fail_superblock(journal);
1319 journal->j_first = first;
1320 journal->j_last = last;
1322 journal->j_head = first;
1323 journal->j_tail = first;
1324 journal->j_free = last - first;
1326 journal->j_tail_sequence = journal->j_transaction_sequence;
1327 journal->j_commit_sequence = journal->j_transaction_sequence - 1;
1328 journal->j_commit_request = journal->j_commit_sequence;
1330 journal->j_max_transaction_buffers = journal->j_maxlen / 4;
1333 * As a special case, if the on-disk copy is already marked as needing
1334 * no recovery (s_start == 0), then we can safely defer the superblock
1335 * update until the next commit by setting JBD2_FLUSHED. This avoids
1336 * attempting a write to a potential-readonly device.
1338 if (sb->s_start == 0) {
1339 jbd_debug(1, "JBD2: Skipping superblock update on recovered sb "
1340 "(start %ld, seq %d, errno %d)\n",
1341 journal->j_tail, journal->j_tail_sequence,
1343 journal->j_flags |= JBD2_FLUSHED;
1345 /* Lock here to make assertions happy... */
1346 mutex_lock_io(&journal->j_checkpoint_mutex);
1348 * Update log tail information. We use REQ_FUA since new
1349 * transaction will start reusing journal space and so we
1350 * must make sure information about current log tail is on
1353 jbd2_journal_update_sb_log_tail(journal,
1354 journal->j_tail_sequence,
1356 REQ_SYNC | REQ_FUA);
1357 mutex_unlock(&journal->j_checkpoint_mutex);
1359 return jbd2_journal_start_thread(journal);
1363 * This function expects that the caller will have locked the journal
1364 * buffer head, and will return with it unlocked
1366 static int jbd2_write_superblock(journal_t *journal, int write_flags)
1368 struct buffer_head *bh = journal->j_sb_buffer;
1369 journal_superblock_t *sb = journal->j_superblock;
1372 /* Buffer got discarded which means block device got invalidated */
1373 if (!buffer_mapped(bh)) {
1378 trace_jbd2_write_superblock(journal, write_flags);
1379 if (!(journal->j_flags & JBD2_BARRIER))
1380 write_flags &= ~(REQ_FUA | REQ_PREFLUSH);
1381 if (buffer_write_io_error(bh)) {
1383 * Oh, dear. A previous attempt to write the journal
1384 * superblock failed. This could happen because the
1385 * USB device was yanked out. Or it could happen to
1386 * be a transient write error and maybe the block will
1387 * be remapped. Nothing we can do but to retry the
1388 * write and hope for the best.
1390 printk(KERN_ERR "JBD2: previous I/O error detected "
1391 "for journal superblock update for %s.\n",
1392 journal->j_devname);
1393 clear_buffer_write_io_error(bh);
1394 set_buffer_uptodate(bh);
1396 jbd2_superblock_csum_set(journal, sb);
1398 bh->b_end_io = end_buffer_write_sync;
1399 ret = submit_bh(REQ_OP_WRITE, write_flags, bh);
1401 if (buffer_write_io_error(bh)) {
1402 clear_buffer_write_io_error(bh);
1403 set_buffer_uptodate(bh);
1407 printk(KERN_ERR "JBD2: Error %d detected when updating "
1408 "journal superblock for %s.\n", ret,
1409 journal->j_devname);
1410 jbd2_journal_abort(journal, ret);
1417 * jbd2_journal_update_sb_log_tail() - Update log tail in journal sb on disk.
1418 * @journal: The journal to update.
1419 * @tail_tid: TID of the new transaction at the tail of the log
1420 * @tail_block: The first block of the transaction at the tail of the log
1421 * @write_op: With which operation should we write the journal sb
1423 * Update a journal's superblock information about log tail and write it to
1424 * disk, waiting for the IO to complete.
1426 int jbd2_journal_update_sb_log_tail(journal_t *journal, tid_t tail_tid,
1427 unsigned long tail_block, int write_op)
1429 journal_superblock_t *sb = journal->j_superblock;
1432 if (is_journal_aborted(journal))
1435 BUG_ON(!mutex_is_locked(&journal->j_checkpoint_mutex));
1436 jbd_debug(1, "JBD2: updating superblock (start %lu, seq %u)\n",
1437 tail_block, tail_tid);
1439 lock_buffer(journal->j_sb_buffer);
1440 sb->s_sequence = cpu_to_be32(tail_tid);
1441 sb->s_start = cpu_to_be32(tail_block);
1443 ret = jbd2_write_superblock(journal, write_op);
1447 /* Log is no longer empty */
1448 write_lock(&journal->j_state_lock);
1449 WARN_ON(!sb->s_sequence);
1450 journal->j_flags &= ~JBD2_FLUSHED;
1451 write_unlock(&journal->j_state_lock);
1458 * jbd2_mark_journal_empty() - Mark on disk journal as empty.
1459 * @journal: The journal to update.
1460 * @write_op: With which operation should we write the journal sb
1462 * Update a journal's dynamic superblock fields to show that journal is empty.
1463 * Write updated superblock to disk waiting for IO to complete.
1465 static void jbd2_mark_journal_empty(journal_t *journal, int write_op)
1467 journal_superblock_t *sb = journal->j_superblock;
1469 BUG_ON(!mutex_is_locked(&journal->j_checkpoint_mutex));
1470 lock_buffer(journal->j_sb_buffer);
1471 if (sb->s_start == 0) { /* Is it already empty? */
1472 unlock_buffer(journal->j_sb_buffer);
1476 jbd_debug(1, "JBD2: Marking journal as empty (seq %d)\n",
1477 journal->j_tail_sequence);
1479 sb->s_sequence = cpu_to_be32(journal->j_tail_sequence);
1480 sb->s_start = cpu_to_be32(0);
1482 jbd2_write_superblock(journal, write_op);
1484 /* Log is no longer empty */
1485 write_lock(&journal->j_state_lock);
1486 journal->j_flags |= JBD2_FLUSHED;
1487 write_unlock(&journal->j_state_lock);
1492 * jbd2_journal_update_sb_errno() - Update error in the journal.
1493 * @journal: The journal to update.
1495 * Update a journal's errno. Write updated superblock to disk waiting for IO
1498 void jbd2_journal_update_sb_errno(journal_t *journal)
1500 journal_superblock_t *sb = journal->j_superblock;
1503 lock_buffer(journal->j_sb_buffer);
1504 errcode = journal->j_errno;
1505 if (errcode == -ESHUTDOWN)
1507 jbd_debug(1, "JBD2: updating superblock error (errno %d)\n", errcode);
1508 sb->s_errno = cpu_to_be32(errcode);
1510 jbd2_write_superblock(journal, REQ_SYNC | REQ_FUA);
1512 EXPORT_SYMBOL(jbd2_journal_update_sb_errno);
1515 * Read the superblock for a given journal, performing initial
1516 * validation of the format.
1518 static int journal_get_superblock(journal_t *journal)
1520 struct buffer_head *bh;
1521 journal_superblock_t *sb;
1524 bh = journal->j_sb_buffer;
1526 J_ASSERT(bh != NULL);
1527 if (!buffer_uptodate(bh)) {
1528 ll_rw_block(REQ_OP_READ, 0, 1, &bh);
1530 if (!buffer_uptodate(bh)) {
1532 "JBD2: IO error reading journal superblock\n");
1537 if (buffer_verified(bh))
1540 sb = journal->j_superblock;
1544 if (sb->s_header.h_magic != cpu_to_be32(JBD2_MAGIC_NUMBER) ||
1545 sb->s_blocksize != cpu_to_be32(journal->j_blocksize)) {
1546 printk(KERN_WARNING "JBD2: no valid journal superblock found\n");
1550 switch(be32_to_cpu(sb->s_header.h_blocktype)) {
1551 case JBD2_SUPERBLOCK_V1:
1552 journal->j_format_version = 1;
1554 case JBD2_SUPERBLOCK_V2:
1555 journal->j_format_version = 2;
1558 printk(KERN_WARNING "JBD2: unrecognised superblock format ID\n");
1562 if (be32_to_cpu(sb->s_maxlen) < journal->j_maxlen)
1563 journal->j_maxlen = be32_to_cpu(sb->s_maxlen);
1564 else if (be32_to_cpu(sb->s_maxlen) > journal->j_maxlen) {
1565 printk(KERN_WARNING "JBD2: journal file too short\n");
1569 if (be32_to_cpu(sb->s_first) == 0 ||
1570 be32_to_cpu(sb->s_first) >= journal->j_maxlen) {
1572 "JBD2: Invalid start block of journal: %u\n",
1573 be32_to_cpu(sb->s_first));
1577 if (jbd2_has_feature_csum2(journal) &&
1578 jbd2_has_feature_csum3(journal)) {
1579 /* Can't have checksum v2 and v3 at the same time! */
1580 printk(KERN_ERR "JBD2: Can't enable checksumming v2 and v3 "
1581 "at the same time!\n");
1585 if (jbd2_journal_has_csum_v2or3_feature(journal) &&
1586 jbd2_has_feature_checksum(journal)) {
1587 /* Can't have checksum v1 and v2 on at the same time! */
1588 printk(KERN_ERR "JBD2: Can't enable checksumming v1 and v2/3 "
1589 "at the same time!\n");
1593 if (!jbd2_verify_csum_type(journal, sb)) {
1594 printk(KERN_ERR "JBD2: Unknown checksum type\n");
1598 /* Load the checksum driver */
1599 if (jbd2_journal_has_csum_v2or3_feature(journal)) {
1600 journal->j_chksum_driver = crypto_alloc_shash("crc32c", 0, 0);
1601 if (IS_ERR(journal->j_chksum_driver)) {
1602 printk(KERN_ERR "JBD2: Cannot load crc32c driver.\n");
1603 err = PTR_ERR(journal->j_chksum_driver);
1604 journal->j_chksum_driver = NULL;
1609 /* Check superblock checksum */
1610 if (!jbd2_superblock_csum_verify(journal, sb)) {
1611 printk(KERN_ERR "JBD2: journal checksum error\n");
1616 /* Precompute checksum seed for all metadata */
1617 if (jbd2_journal_has_csum_v2or3(journal))
1618 journal->j_csum_seed = jbd2_chksum(journal, ~0, sb->s_uuid,
1619 sizeof(sb->s_uuid));
1621 set_buffer_verified(bh);
1626 journal_fail_superblock(journal);
1631 * Load the on-disk journal superblock and read the key fields into the
1635 static int load_superblock(journal_t *journal)
1638 journal_superblock_t *sb;
1640 err = journal_get_superblock(journal);
1644 sb = journal->j_superblock;
1646 journal->j_tail_sequence = be32_to_cpu(sb->s_sequence);
1647 journal->j_tail = be32_to_cpu(sb->s_start);
1648 journal->j_first = be32_to_cpu(sb->s_first);
1649 journal->j_last = be32_to_cpu(sb->s_maxlen);
1650 journal->j_errno = be32_to_cpu(sb->s_errno);
1657 * int jbd2_journal_load() - Read journal from disk.
1658 * @journal: Journal to act on.
1660 * Given a journal_t structure which tells us which disk blocks contain
1661 * a journal, read the journal from disk to initialise the in-memory
1664 int jbd2_journal_load(journal_t *journal)
1667 journal_superblock_t *sb;
1669 err = load_superblock(journal);
1673 sb = journal->j_superblock;
1674 /* If this is a V2 superblock, then we have to check the
1675 * features flags on it. */
1677 if (journal->j_format_version >= 2) {
1678 if ((sb->s_feature_ro_compat &
1679 ~cpu_to_be32(JBD2_KNOWN_ROCOMPAT_FEATURES)) ||
1680 (sb->s_feature_incompat &
1681 ~cpu_to_be32(JBD2_KNOWN_INCOMPAT_FEATURES))) {
1683 "JBD2: Unrecognised features on journal\n");
1689 * Create a slab for this blocksize
1691 err = jbd2_journal_create_slab(be32_to_cpu(sb->s_blocksize));
1695 /* Let the recovery code check whether it needs to recover any
1696 * data from the journal. */
1697 if (jbd2_journal_recover(journal))
1698 goto recovery_error;
1700 if (journal->j_failed_commit) {
1701 printk(KERN_ERR "JBD2: journal transaction %u on %s "
1702 "is corrupt.\n", journal->j_failed_commit,
1703 journal->j_devname);
1704 return -EFSCORRUPTED;
1707 * clear JBD2_ABORT flag initialized in journal_init_common
1708 * here to update log tail information with the newest seq.
1710 journal->j_flags &= ~JBD2_ABORT;
1712 /* OK, we've finished with the dynamic journal bits:
1713 * reinitialise the dynamic contents of the superblock in memory
1714 * and reset them on disk. */
1715 if (journal_reset(journal))
1716 goto recovery_error;
1718 journal->j_flags |= JBD2_LOADED;
1722 printk(KERN_WARNING "JBD2: recovery failed\n");
1727 * void jbd2_journal_destroy() - Release a journal_t structure.
1728 * @journal: Journal to act on.
1730 * Release a journal_t structure once it is no longer in use by the
1732 * Return <0 if we couldn't clean up the journal.
1734 int jbd2_journal_destroy(journal_t *journal)
1738 /* Wait for the commit thread to wake up and die. */
1739 journal_kill_thread(journal);
1741 /* Force a final log commit */
1742 if (journal->j_running_transaction)
1743 jbd2_journal_commit_transaction(journal);
1745 /* Force any old transactions to disk */
1747 /* Totally anal locking here... */
1748 spin_lock(&journal->j_list_lock);
1749 while (journal->j_checkpoint_transactions != NULL) {
1750 spin_unlock(&journal->j_list_lock);
1751 mutex_lock_io(&journal->j_checkpoint_mutex);
1752 err = jbd2_log_do_checkpoint(journal);
1753 mutex_unlock(&journal->j_checkpoint_mutex);
1755 * If checkpointing failed, just free the buffers to avoid
1759 jbd2_journal_destroy_checkpoint(journal);
1760 spin_lock(&journal->j_list_lock);
1763 spin_lock(&journal->j_list_lock);
1766 J_ASSERT(journal->j_running_transaction == NULL);
1767 J_ASSERT(journal->j_committing_transaction == NULL);
1768 J_ASSERT(journal->j_checkpoint_transactions == NULL);
1769 spin_unlock(&journal->j_list_lock);
1771 if (journal->j_sb_buffer) {
1772 if (!is_journal_aborted(journal)) {
1773 mutex_lock_io(&journal->j_checkpoint_mutex);
1775 write_lock(&journal->j_state_lock);
1776 journal->j_tail_sequence =
1777 ++journal->j_transaction_sequence;
1778 write_unlock(&journal->j_state_lock);
1780 jbd2_mark_journal_empty(journal,
1781 REQ_SYNC | REQ_PREFLUSH | REQ_FUA);
1782 mutex_unlock(&journal->j_checkpoint_mutex);
1785 brelse(journal->j_sb_buffer);
1788 if (journal->j_proc_entry)
1789 jbd2_stats_proc_exit(journal);
1790 iput(journal->j_inode);
1791 if (journal->j_revoke)
1792 jbd2_journal_destroy_revoke(journal);
1793 if (journal->j_chksum_driver)
1794 crypto_free_shash(journal->j_chksum_driver);
1795 kfree(journal->j_wbuf);
1803 *int jbd2_journal_check_used_features () - Check if features specified are used.
1804 * @journal: Journal to check.
1805 * @compat: bitmask of compatible features
1806 * @ro: bitmask of features that force read-only mount
1807 * @incompat: bitmask of incompatible features
1809 * Check whether the journal uses all of a given set of
1810 * features. Return true (non-zero) if it does.
1813 int jbd2_journal_check_used_features (journal_t *journal, unsigned long compat,
1814 unsigned long ro, unsigned long incompat)
1816 journal_superblock_t *sb;
1818 if (!compat && !ro && !incompat)
1820 /* Load journal superblock if it is not loaded yet. */
1821 if (journal->j_format_version == 0 &&
1822 journal_get_superblock(journal) != 0)
1824 if (journal->j_format_version == 1)
1827 sb = journal->j_superblock;
1829 if (((be32_to_cpu(sb->s_feature_compat) & compat) == compat) &&
1830 ((be32_to_cpu(sb->s_feature_ro_compat) & ro) == ro) &&
1831 ((be32_to_cpu(sb->s_feature_incompat) & incompat) == incompat))
1838 * int jbd2_journal_check_available_features() - Check feature set in journalling layer
1839 * @journal: Journal to check.
1840 * @compat: bitmask of compatible features
1841 * @ro: bitmask of features that force read-only mount
1842 * @incompat: bitmask of incompatible features
1844 * Check whether the journaling code supports the use of
1845 * all of a given set of features on this journal. Return true
1846 * (non-zero) if it can. */
1848 int jbd2_journal_check_available_features (journal_t *journal, unsigned long compat,
1849 unsigned long ro, unsigned long incompat)
1851 if (!compat && !ro && !incompat)
1854 /* We can support any known requested features iff the
1855 * superblock is in version 2. Otherwise we fail to support any
1856 * extended sb features. */
1858 if (journal->j_format_version != 2)
1861 if ((compat & JBD2_KNOWN_COMPAT_FEATURES) == compat &&
1862 (ro & JBD2_KNOWN_ROCOMPAT_FEATURES) == ro &&
1863 (incompat & JBD2_KNOWN_INCOMPAT_FEATURES) == incompat)
1870 * int jbd2_journal_set_features () - Mark a given journal feature in the superblock
1871 * @journal: Journal to act on.
1872 * @compat: bitmask of compatible features
1873 * @ro: bitmask of features that force read-only mount
1874 * @incompat: bitmask of incompatible features
1876 * Mark a given journal feature as present on the
1877 * superblock. Returns true if the requested features could be set.
1881 int jbd2_journal_set_features (journal_t *journal, unsigned long compat,
1882 unsigned long ro, unsigned long incompat)
1884 #define INCOMPAT_FEATURE_ON(f) \
1885 ((incompat & (f)) && !(sb->s_feature_incompat & cpu_to_be32(f)))
1886 #define COMPAT_FEATURE_ON(f) \
1887 ((compat & (f)) && !(sb->s_feature_compat & cpu_to_be32(f)))
1888 journal_superblock_t *sb;
1890 if (jbd2_journal_check_used_features(journal, compat, ro, incompat))
1893 if (!jbd2_journal_check_available_features(journal, compat, ro, incompat))
1896 /* If enabling v2 checksums, turn on v3 instead */
1897 if (incompat & JBD2_FEATURE_INCOMPAT_CSUM_V2) {
1898 incompat &= ~JBD2_FEATURE_INCOMPAT_CSUM_V2;
1899 incompat |= JBD2_FEATURE_INCOMPAT_CSUM_V3;
1902 /* Asking for checksumming v3 and v1? Only give them v3. */
1903 if (incompat & JBD2_FEATURE_INCOMPAT_CSUM_V3 &&
1904 compat & JBD2_FEATURE_COMPAT_CHECKSUM)
1905 compat &= ~JBD2_FEATURE_COMPAT_CHECKSUM;
1907 jbd_debug(1, "Setting new features 0x%lx/0x%lx/0x%lx\n",
1908 compat, ro, incompat);
1910 sb = journal->j_superblock;
1912 /* Load the checksum driver if necessary */
1913 if ((journal->j_chksum_driver == NULL) &&
1914 INCOMPAT_FEATURE_ON(JBD2_FEATURE_INCOMPAT_CSUM_V3)) {
1915 journal->j_chksum_driver = crypto_alloc_shash("crc32c", 0, 0);
1916 if (IS_ERR(journal->j_chksum_driver)) {
1917 printk(KERN_ERR "JBD2: Cannot load crc32c driver.\n");
1918 journal->j_chksum_driver = NULL;
1921 /* Precompute checksum seed for all metadata */
1922 journal->j_csum_seed = jbd2_chksum(journal, ~0, sb->s_uuid,
1923 sizeof(sb->s_uuid));
1926 lock_buffer(journal->j_sb_buffer);
1928 /* If enabling v3 checksums, update superblock */
1929 if (INCOMPAT_FEATURE_ON(JBD2_FEATURE_INCOMPAT_CSUM_V3)) {
1930 sb->s_checksum_type = JBD2_CRC32C_CHKSUM;
1931 sb->s_feature_compat &=
1932 ~cpu_to_be32(JBD2_FEATURE_COMPAT_CHECKSUM);
1935 /* If enabling v1 checksums, downgrade superblock */
1936 if (COMPAT_FEATURE_ON(JBD2_FEATURE_COMPAT_CHECKSUM))
1937 sb->s_feature_incompat &=
1938 ~cpu_to_be32(JBD2_FEATURE_INCOMPAT_CSUM_V2 |
1939 JBD2_FEATURE_INCOMPAT_CSUM_V3);
1941 sb->s_feature_compat |= cpu_to_be32(compat);
1942 sb->s_feature_ro_compat |= cpu_to_be32(ro);
1943 sb->s_feature_incompat |= cpu_to_be32(incompat);
1944 unlock_buffer(journal->j_sb_buffer);
1947 #undef COMPAT_FEATURE_ON
1948 #undef INCOMPAT_FEATURE_ON
1952 * jbd2_journal_clear_features () - Clear a given journal feature in the
1954 * @journal: Journal to act on.
1955 * @compat: bitmask of compatible features
1956 * @ro: bitmask of features that force read-only mount
1957 * @incompat: bitmask of incompatible features
1959 * Clear a given journal feature as present on the
1962 void jbd2_journal_clear_features(journal_t *journal, unsigned long compat,
1963 unsigned long ro, unsigned long incompat)
1965 journal_superblock_t *sb;
1967 jbd_debug(1, "Clear features 0x%lx/0x%lx/0x%lx\n",
1968 compat, ro, incompat);
1970 sb = journal->j_superblock;
1972 sb->s_feature_compat &= ~cpu_to_be32(compat);
1973 sb->s_feature_ro_compat &= ~cpu_to_be32(ro);
1974 sb->s_feature_incompat &= ~cpu_to_be32(incompat);
1976 EXPORT_SYMBOL(jbd2_journal_clear_features);
1979 * int jbd2_journal_flush () - Flush journal
1980 * @journal: Journal to act on.
1982 * Flush all data for a given journal to disk and empty the journal.
1983 * Filesystems can use this when remounting readonly to ensure that
1984 * recovery does not need to happen on remount.
1987 int jbd2_journal_flush(journal_t *journal)
1990 transaction_t *transaction = NULL;
1992 write_lock(&journal->j_state_lock);
1994 /* Force everything buffered to the log... */
1995 if (journal->j_running_transaction) {
1996 transaction = journal->j_running_transaction;
1997 __jbd2_log_start_commit(journal, transaction->t_tid);
1998 } else if (journal->j_committing_transaction)
1999 transaction = journal->j_committing_transaction;
2001 /* Wait for the log commit to complete... */
2003 tid_t tid = transaction->t_tid;
2005 write_unlock(&journal->j_state_lock);
2006 jbd2_log_wait_commit(journal, tid);
2008 write_unlock(&journal->j_state_lock);
2011 /* ...and flush everything in the log out to disk. */
2012 spin_lock(&journal->j_list_lock);
2013 while (!err && journal->j_checkpoint_transactions != NULL) {
2014 spin_unlock(&journal->j_list_lock);
2015 mutex_lock_io(&journal->j_checkpoint_mutex);
2016 err = jbd2_log_do_checkpoint(journal);
2017 mutex_unlock(&journal->j_checkpoint_mutex);
2018 spin_lock(&journal->j_list_lock);
2020 spin_unlock(&journal->j_list_lock);
2022 if (is_journal_aborted(journal))
2025 mutex_lock_io(&journal->j_checkpoint_mutex);
2027 err = jbd2_cleanup_journal_tail(journal);
2029 mutex_unlock(&journal->j_checkpoint_mutex);
2035 /* Finally, mark the journal as really needing no recovery.
2036 * This sets s_start==0 in the underlying superblock, which is
2037 * the magic code for a fully-recovered superblock. Any future
2038 * commits of data to the journal will restore the current
2040 jbd2_mark_journal_empty(journal, REQ_SYNC | REQ_FUA);
2041 mutex_unlock(&journal->j_checkpoint_mutex);
2042 write_lock(&journal->j_state_lock);
2043 J_ASSERT(!journal->j_running_transaction);
2044 J_ASSERT(!journal->j_committing_transaction);
2045 J_ASSERT(!journal->j_checkpoint_transactions);
2046 J_ASSERT(journal->j_head == journal->j_tail);
2047 J_ASSERT(journal->j_tail_sequence == journal->j_transaction_sequence);
2048 write_unlock(&journal->j_state_lock);
2054 * int jbd2_journal_wipe() - Wipe journal contents
2055 * @journal: Journal to act on.
2056 * @write: flag (see below)
2058 * Wipe out all of the contents of a journal, safely. This will produce
2059 * a warning if the journal contains any valid recovery information.
2060 * Must be called between journal_init_*() and jbd2_journal_load().
2062 * If 'write' is non-zero, then we wipe out the journal on disk; otherwise
2063 * we merely suppress recovery.
2066 int jbd2_journal_wipe(journal_t *journal, int write)
2070 J_ASSERT (!(journal->j_flags & JBD2_LOADED));
2072 err = load_superblock(journal);
2076 if (!journal->j_tail)
2079 printk(KERN_WARNING "JBD2: %s recovery information on journal\n",
2080 write ? "Clearing" : "Ignoring");
2082 err = jbd2_journal_skip_recovery(journal);
2084 /* Lock to make assertions happy... */
2085 mutex_lock(&journal->j_checkpoint_mutex);
2086 jbd2_mark_journal_empty(journal, REQ_SYNC | REQ_FUA);
2087 mutex_unlock(&journal->j_checkpoint_mutex);
2095 * Journal abort has very specific semantics, which we describe
2096 * for journal abort.
2098 * Two internal functions, which provide abort to the jbd layer
2103 * Quick version for internal journal use (doesn't lock the journal).
2104 * Aborts hard --- we mark the abort as occurred, but do _nothing_ else,
2105 * and don't attempt to make any other journal updates.
2107 void __jbd2_journal_abort_hard(journal_t *journal)
2109 transaction_t *transaction;
2111 if (journal->j_flags & JBD2_ABORT)
2114 printk(KERN_ERR "Aborting journal on device %s.\n",
2115 journal->j_devname);
2117 write_lock(&journal->j_state_lock);
2118 journal->j_flags |= JBD2_ABORT;
2119 transaction = journal->j_running_transaction;
2121 __jbd2_log_start_commit(journal, transaction->t_tid);
2122 write_unlock(&journal->j_state_lock);
2125 /* Soft abort: record the abort error status in the journal superblock,
2126 * but don't do any other IO. */
2127 static void __journal_abort_soft (journal_t *journal, int errno)
2131 write_lock(&journal->j_state_lock);
2132 old_errno = journal->j_errno;
2133 if (!journal->j_errno || errno == -ESHUTDOWN)
2134 journal->j_errno = errno;
2136 if (journal->j_flags & JBD2_ABORT) {
2137 write_unlock(&journal->j_state_lock);
2138 if (old_errno != -ESHUTDOWN && errno == -ESHUTDOWN)
2139 jbd2_journal_update_sb_errno(journal);
2142 write_unlock(&journal->j_state_lock);
2144 __jbd2_journal_abort_hard(journal);
2146 jbd2_journal_update_sb_errno(journal);
2147 write_lock(&journal->j_state_lock);
2148 journal->j_flags |= JBD2_REC_ERR;
2149 write_unlock(&journal->j_state_lock);
2153 * void jbd2_journal_abort () - Shutdown the journal immediately.
2154 * @journal: the journal to shutdown.
2155 * @errno: an error number to record in the journal indicating
2156 * the reason for the shutdown.
2158 * Perform a complete, immediate shutdown of the ENTIRE
2159 * journal (not of a single transaction). This operation cannot be
2160 * undone without closing and reopening the journal.
2162 * The jbd2_journal_abort function is intended to support higher level error
2163 * recovery mechanisms such as the ext2/ext3 remount-readonly error
2166 * Journal abort has very specific semantics. Any existing dirty,
2167 * unjournaled buffers in the main filesystem will still be written to
2168 * disk by bdflush, but the journaling mechanism will be suspended
2169 * immediately and no further transaction commits will be honoured.
2171 * Any dirty, journaled buffers will be written back to disk without
2172 * hitting the journal. Atomicity cannot be guaranteed on an aborted
2173 * filesystem, but we _do_ attempt to leave as much data as possible
2174 * behind for fsck to use for cleanup.
2176 * Any attempt to get a new transaction handle on a journal which is in
2177 * ABORT state will just result in an -EROFS error return. A
2178 * jbd2_journal_stop on an existing handle will return -EIO if we have
2179 * entered abort state during the update.
2181 * Recursive transactions are not disturbed by journal abort until the
2182 * final jbd2_journal_stop, which will receive the -EIO error.
2184 * Finally, the jbd2_journal_abort call allows the caller to supply an errno
2185 * which will be recorded (if possible) in the journal superblock. This
2186 * allows a client to record failure conditions in the middle of a
2187 * transaction without having to complete the transaction to record the
2188 * failure to disk. ext3_error, for example, now uses this
2193 void jbd2_journal_abort(journal_t *journal, int errno)
2195 __journal_abort_soft(journal, errno);
2199 * int jbd2_journal_errno () - returns the journal's error state.
2200 * @journal: journal to examine.
2202 * This is the errno number set with jbd2_journal_abort(), the last
2203 * time the journal was mounted - if the journal was stopped
2204 * without calling abort this will be 0.
2206 * If the journal has been aborted on this mount time -EROFS will
2209 int jbd2_journal_errno(journal_t *journal)
2213 read_lock(&journal->j_state_lock);
2214 if (journal->j_flags & JBD2_ABORT)
2217 err = journal->j_errno;
2218 read_unlock(&journal->j_state_lock);
2223 * int jbd2_journal_clear_err () - clears the journal's error state
2224 * @journal: journal to act on.
2226 * An error must be cleared or acked to take a FS out of readonly
2229 int jbd2_journal_clear_err(journal_t *journal)
2233 write_lock(&journal->j_state_lock);
2234 if (journal->j_flags & JBD2_ABORT)
2237 journal->j_errno = 0;
2238 write_unlock(&journal->j_state_lock);
2243 * void jbd2_journal_ack_err() - Ack journal err.
2244 * @journal: journal to act on.
2246 * An error must be cleared or acked to take a FS out of readonly
2249 void jbd2_journal_ack_err(journal_t *journal)
2251 write_lock(&journal->j_state_lock);
2252 if (journal->j_errno)
2253 journal->j_flags |= JBD2_ACK_ERR;
2254 write_unlock(&journal->j_state_lock);
2257 int jbd2_journal_blocks_per_page(struct inode *inode)
2259 return 1 << (PAGE_SHIFT - inode->i_sb->s_blocksize_bits);
2263 * helper functions to deal with 32 or 64bit block numbers.
2265 size_t journal_tag_bytes(journal_t *journal)
2269 if (jbd2_has_feature_csum3(journal))
2270 return sizeof(journal_block_tag3_t);
2272 sz = sizeof(journal_block_tag_t);
2274 if (jbd2_has_feature_csum2(journal))
2275 sz += sizeof(__u16);
2277 if (jbd2_has_feature_64bit(journal))
2280 return sz - sizeof(__u32);
2284 * JBD memory management
2286 * These functions are used to allocate block-sized chunks of memory
2287 * used for making copies of buffer_head data. Very often it will be
2288 * page-sized chunks of data, but sometimes it will be in
2289 * sub-page-size chunks. (For example, 16k pages on Power systems
2290 * with a 4k block file system.) For blocks smaller than a page, we
2291 * use a SLAB allocator. There are slab caches for each block size,
2292 * which are allocated at mount time, if necessary, and we only free
2293 * (all of) the slab caches when/if the jbd2 module is unloaded. For
2294 * this reason we don't need to a mutex to protect access to
2295 * jbd2_slab[] allocating or releasing memory; only in
2296 * jbd2_journal_create_slab().
2298 #define JBD2_MAX_SLABS 8
2299 static struct kmem_cache *jbd2_slab[JBD2_MAX_SLABS];
2301 static const char *jbd2_slab_names[JBD2_MAX_SLABS] = {
2302 "jbd2_1k", "jbd2_2k", "jbd2_4k", "jbd2_8k",
2303 "jbd2_16k", "jbd2_32k", "jbd2_64k", "jbd2_128k"
2307 static void jbd2_journal_destroy_slabs(void)
2311 for (i = 0; i < JBD2_MAX_SLABS; i++) {
2312 kmem_cache_destroy(jbd2_slab[i]);
2313 jbd2_slab[i] = NULL;
2317 static int jbd2_journal_create_slab(size_t size)
2319 static DEFINE_MUTEX(jbd2_slab_create_mutex);
2320 int i = order_base_2(size) - 10;
2323 if (size == PAGE_SIZE)
2326 if (i >= JBD2_MAX_SLABS)
2329 if (unlikely(i < 0))
2331 mutex_lock(&jbd2_slab_create_mutex);
2333 mutex_unlock(&jbd2_slab_create_mutex);
2334 return 0; /* Already created */
2337 slab_size = 1 << (i+10);
2338 jbd2_slab[i] = kmem_cache_create(jbd2_slab_names[i], slab_size,
2339 slab_size, 0, NULL);
2340 mutex_unlock(&jbd2_slab_create_mutex);
2341 if (!jbd2_slab[i]) {
2342 printk(KERN_EMERG "JBD2: no memory for jbd2_slab cache\n");
2348 static struct kmem_cache *get_slab(size_t size)
2350 int i = order_base_2(size) - 10;
2352 BUG_ON(i >= JBD2_MAX_SLABS);
2353 if (unlikely(i < 0))
2355 BUG_ON(jbd2_slab[i] == NULL);
2356 return jbd2_slab[i];
2359 void *jbd2_alloc(size_t size, gfp_t flags)
2363 BUG_ON(size & (size-1)); /* Must be a power of 2 */
2365 if (size < PAGE_SIZE)
2366 ptr = kmem_cache_alloc(get_slab(size), flags);
2368 ptr = (void *)__get_free_pages(flags, get_order(size));
2370 /* Check alignment; SLUB has gotten this wrong in the past,
2371 * and this can lead to user data corruption! */
2372 BUG_ON(((unsigned long) ptr) & (size-1));
2377 void jbd2_free(void *ptr, size_t size)
2379 if (size < PAGE_SIZE)
2380 kmem_cache_free(get_slab(size), ptr);
2382 free_pages((unsigned long)ptr, get_order(size));
2386 * Journal_head storage management
2388 static struct kmem_cache *jbd2_journal_head_cache;
2389 #ifdef CONFIG_JBD2_DEBUG
2390 static atomic_t nr_journal_heads = ATOMIC_INIT(0);
2393 static int __init jbd2_journal_init_journal_head_cache(void)
2395 J_ASSERT(!jbd2_journal_head_cache);
2396 jbd2_journal_head_cache = kmem_cache_create("jbd2_journal_head",
2397 sizeof(struct journal_head),
2399 SLAB_TEMPORARY | SLAB_TYPESAFE_BY_RCU,
2401 if (!jbd2_journal_head_cache) {
2402 printk(KERN_EMERG "JBD2: no memory for journal_head cache\n");
2408 static void jbd2_journal_destroy_journal_head_cache(void)
2410 kmem_cache_destroy(jbd2_journal_head_cache);
2411 jbd2_journal_head_cache = NULL;
2415 * journal_head splicing and dicing
2417 static struct journal_head *journal_alloc_journal_head(void)
2419 struct journal_head *ret;
2421 #ifdef CONFIG_JBD2_DEBUG
2422 atomic_inc(&nr_journal_heads);
2424 ret = kmem_cache_zalloc(jbd2_journal_head_cache, GFP_NOFS);
2426 jbd_debug(1, "out of memory for journal_head\n");
2427 pr_notice_ratelimited("ENOMEM in %s, retrying.\n", __func__);
2428 ret = kmem_cache_zalloc(jbd2_journal_head_cache,
2429 GFP_NOFS | __GFP_NOFAIL);
2434 static void journal_free_journal_head(struct journal_head *jh)
2436 #ifdef CONFIG_JBD2_DEBUG
2437 atomic_dec(&nr_journal_heads);
2438 memset(jh, JBD2_POISON_FREE, sizeof(*jh));
2440 kmem_cache_free(jbd2_journal_head_cache, jh);
2444 * A journal_head is attached to a buffer_head whenever JBD has an
2445 * interest in the buffer.
2447 * Whenever a buffer has an attached journal_head, its ->b_state:BH_JBD bit
2448 * is set. This bit is tested in core kernel code where we need to take
2449 * JBD-specific actions. Testing the zeroness of ->b_private is not reliable
2452 * When a buffer has its BH_JBD bit set, its ->b_count is elevated by one.
2454 * When a buffer has its BH_JBD bit set it is immune from being released by
2455 * core kernel code, mainly via ->b_count.
2457 * A journal_head is detached from its buffer_head when the journal_head's
2458 * b_jcount reaches zero. Running transaction (b_transaction) and checkpoint
2459 * transaction (b_cp_transaction) hold their references to b_jcount.
2461 * Various places in the kernel want to attach a journal_head to a buffer_head
2462 * _before_ attaching the journal_head to a transaction. To protect the
2463 * journal_head in this situation, jbd2_journal_add_journal_head elevates the
2464 * journal_head's b_jcount refcount by one. The caller must call
2465 * jbd2_journal_put_journal_head() to undo this.
2467 * So the typical usage would be:
2469 * (Attach a journal_head if needed. Increments b_jcount)
2470 * struct journal_head *jh = jbd2_journal_add_journal_head(bh);
2472 * (Get another reference for transaction)
2473 * jbd2_journal_grab_journal_head(bh);
2474 * jh->b_transaction = xxx;
2475 * (Put original reference)
2476 * jbd2_journal_put_journal_head(jh);
2480 * Give a buffer_head a journal_head.
2484 struct journal_head *jbd2_journal_add_journal_head(struct buffer_head *bh)
2486 struct journal_head *jh;
2487 struct journal_head *new_jh = NULL;
2490 if (!buffer_jbd(bh))
2491 new_jh = journal_alloc_journal_head();
2493 jbd_lock_bh_journal_head(bh);
2494 if (buffer_jbd(bh)) {
2498 (atomic_read(&bh->b_count) > 0) ||
2499 (bh->b_page && bh->b_page->mapping));
2502 jbd_unlock_bh_journal_head(bh);
2507 new_jh = NULL; /* We consumed it */
2512 BUFFER_TRACE(bh, "added journal_head");
2515 jbd_unlock_bh_journal_head(bh);
2517 journal_free_journal_head(new_jh);
2518 return bh->b_private;
2522 * Grab a ref against this buffer_head's journal_head. If it ended up not
2523 * having a journal_head, return NULL
2525 struct journal_head *jbd2_journal_grab_journal_head(struct buffer_head *bh)
2527 struct journal_head *jh = NULL;
2529 jbd_lock_bh_journal_head(bh);
2530 if (buffer_jbd(bh)) {
2534 jbd_unlock_bh_journal_head(bh);
2538 static void __journal_remove_journal_head(struct buffer_head *bh)
2540 struct journal_head *jh = bh2jh(bh);
2542 J_ASSERT_JH(jh, jh->b_jcount >= 0);
2543 J_ASSERT_JH(jh, jh->b_transaction == NULL);
2544 J_ASSERT_JH(jh, jh->b_next_transaction == NULL);
2545 J_ASSERT_JH(jh, jh->b_cp_transaction == NULL);
2546 J_ASSERT_JH(jh, jh->b_jlist == BJ_None);
2547 J_ASSERT_BH(bh, buffer_jbd(bh));
2548 J_ASSERT_BH(bh, jh2bh(jh) == bh);
2549 BUFFER_TRACE(bh, "remove journal_head");
2550 if (jh->b_frozen_data) {
2551 printk(KERN_WARNING "%s: freeing b_frozen_data\n", __func__);
2552 jbd2_free(jh->b_frozen_data, bh->b_size);
2554 if (jh->b_committed_data) {
2555 printk(KERN_WARNING "%s: freeing b_committed_data\n", __func__);
2556 jbd2_free(jh->b_committed_data, bh->b_size);
2558 bh->b_private = NULL;
2559 jh->b_bh = NULL; /* debug, really */
2560 clear_buffer_jbd(bh);
2561 journal_free_journal_head(jh);
2565 * Drop a reference on the passed journal_head. If it fell to zero then
2566 * release the journal_head from the buffer_head.
2568 void jbd2_journal_put_journal_head(struct journal_head *jh)
2570 struct buffer_head *bh = jh2bh(jh);
2572 jbd_lock_bh_journal_head(bh);
2573 J_ASSERT_JH(jh, jh->b_jcount > 0);
2575 if (!jh->b_jcount) {
2576 __journal_remove_journal_head(bh);
2577 jbd_unlock_bh_journal_head(bh);
2580 jbd_unlock_bh_journal_head(bh);
2584 * Initialize jbd inode head
2586 void jbd2_journal_init_jbd_inode(struct jbd2_inode *jinode, struct inode *inode)
2588 jinode->i_transaction = NULL;
2589 jinode->i_next_transaction = NULL;
2590 jinode->i_vfs_inode = inode;
2591 jinode->i_flags = 0;
2592 jinode->i_dirty_start = 0;
2593 jinode->i_dirty_end = 0;
2594 INIT_LIST_HEAD(&jinode->i_list);
2598 * Function to be called before we start removing inode from memory (i.e.,
2599 * clear_inode() is a fine place to be called from). It removes inode from
2600 * transaction's lists.
2602 void jbd2_journal_release_jbd_inode(journal_t *journal,
2603 struct jbd2_inode *jinode)
2608 spin_lock(&journal->j_list_lock);
2609 /* Is commit writing out inode - we have to wait */
2610 if (jinode->i_flags & JI_COMMIT_RUNNING) {
2611 wait_queue_head_t *wq;
2612 DEFINE_WAIT_BIT(wait, &jinode->i_flags, __JI_COMMIT_RUNNING);
2613 wq = bit_waitqueue(&jinode->i_flags, __JI_COMMIT_RUNNING);
2614 prepare_to_wait(wq, &wait.wq_entry, TASK_UNINTERRUPTIBLE);
2615 spin_unlock(&journal->j_list_lock);
2617 finish_wait(wq, &wait.wq_entry);
2621 if (jinode->i_transaction) {
2622 list_del(&jinode->i_list);
2623 jinode->i_transaction = NULL;
2625 spin_unlock(&journal->j_list_lock);
2629 #ifdef CONFIG_PROC_FS
2631 #define JBD2_STATS_PROC_NAME "fs/jbd2"
2633 static void __init jbd2_create_jbd_stats_proc_entry(void)
2635 proc_jbd2_stats = proc_mkdir(JBD2_STATS_PROC_NAME, NULL);
2638 static void __exit jbd2_remove_jbd_stats_proc_entry(void)
2640 if (proc_jbd2_stats)
2641 remove_proc_entry(JBD2_STATS_PROC_NAME, NULL);
2646 #define jbd2_create_jbd_stats_proc_entry() do {} while (0)
2647 #define jbd2_remove_jbd_stats_proc_entry() do {} while (0)
2651 struct kmem_cache *jbd2_handle_cache, *jbd2_inode_cache;
2653 static int __init jbd2_journal_init_inode_cache(void)
2655 J_ASSERT(!jbd2_inode_cache);
2656 jbd2_inode_cache = KMEM_CACHE(jbd2_inode, 0);
2657 if (!jbd2_inode_cache) {
2658 pr_emerg("JBD2: failed to create inode cache\n");
2664 static int __init jbd2_journal_init_handle_cache(void)
2666 J_ASSERT(!jbd2_handle_cache);
2667 jbd2_handle_cache = KMEM_CACHE(jbd2_journal_handle, SLAB_TEMPORARY);
2668 if (!jbd2_handle_cache) {
2669 printk(KERN_EMERG "JBD2: failed to create handle cache\n");
2675 static void jbd2_journal_destroy_inode_cache(void)
2677 kmem_cache_destroy(jbd2_inode_cache);
2678 jbd2_inode_cache = NULL;
2681 static void jbd2_journal_destroy_handle_cache(void)
2683 kmem_cache_destroy(jbd2_handle_cache);
2684 jbd2_handle_cache = NULL;
2688 * Module startup and shutdown
2691 static int __init journal_init_caches(void)
2695 ret = jbd2_journal_init_revoke_record_cache();
2697 ret = jbd2_journal_init_revoke_table_cache();
2699 ret = jbd2_journal_init_journal_head_cache();
2701 ret = jbd2_journal_init_handle_cache();
2703 ret = jbd2_journal_init_inode_cache();
2705 ret = jbd2_journal_init_transaction_cache();
2709 static void jbd2_journal_destroy_caches(void)
2711 jbd2_journal_destroy_revoke_record_cache();
2712 jbd2_journal_destroy_revoke_table_cache();
2713 jbd2_journal_destroy_journal_head_cache();
2714 jbd2_journal_destroy_handle_cache();
2715 jbd2_journal_destroy_inode_cache();
2716 jbd2_journal_destroy_transaction_cache();
2717 jbd2_journal_destroy_slabs();
2720 static int __init journal_init(void)
2724 BUILD_BUG_ON(sizeof(struct journal_superblock_s) != 1024);
2726 ret = journal_init_caches();
2728 jbd2_create_jbd_stats_proc_entry();
2730 jbd2_journal_destroy_caches();
2735 static void __exit journal_exit(void)
2737 #ifdef CONFIG_JBD2_DEBUG
2738 int n = atomic_read(&nr_journal_heads);
2740 printk(KERN_ERR "JBD2: leaked %d journal_heads!\n", n);
2742 jbd2_remove_jbd_stats_proc_entry();
2743 jbd2_journal_destroy_caches();
2746 MODULE_LICENSE("GPL");
2747 module_init(journal_init);
2748 module_exit(journal_exit);