HID: hiddev: Fix slab-out-of-bounds write in hiddev_ioctl_usage()
[linux/fpc-iii.git] / fs / jbd2 / journal.c
blobd62435897d0d0c58295c51494da28a9fc345f754
1 /*
2 * linux/fs/jbd2/journal.c
4 * Written by Stephen C. Tweedie <sct@redhat.com>, 1998
6 * Copyright 1998 Red Hat corp --- All Rights Reserved
8 * This file is part of the Linux kernel and is made available under
9 * the terms of the GNU General Public License, version 2, or at your
10 * option, any later version, incorporated herein by reference.
12 * Generic filesystem journal-writing code; part of the ext2fs
13 * journaling system.
15 * This file manages journals: areas of disk reserved for logging
16 * transactional updates. This includes the kernel journaling thread
17 * which is responsible for scheduling updates to the log.
19 * We do not actually manage the physical storage of the journal in this
20 * file: that is left to a per-journal policy function, which allows us
21 * to store the journal within a filesystem-specified area for ext2
22 * journaling (ext2 can use a reserved inode for storing the log).
25 #include <linux/module.h>
26 #include <linux/time.h>
27 #include <linux/fs.h>
28 #include <linux/jbd2.h>
29 #include <linux/errno.h>
30 #include <linux/slab.h>
31 #include <linux/init.h>
32 #include <linux/mm.h>
33 #include <linux/freezer.h>
34 #include <linux/pagemap.h>
35 #include <linux/kthread.h>
36 #include <linux/poison.h>
37 #include <linux/proc_fs.h>
38 #include <linux/seq_file.h>
39 #include <linux/math64.h>
40 #include <linux/hash.h>
41 #include <linux/log2.h>
42 #include <linux/vmalloc.h>
43 #include <linux/backing-dev.h>
44 #include <linux/bitops.h>
45 #include <linux/ratelimit.h>
47 #define CREATE_TRACE_POINTS
48 #include <trace/events/jbd2.h>
50 #include <asm/uaccess.h>
51 #include <asm/page.h>
53 #ifdef CONFIG_JBD2_DEBUG
54 ushort jbd2_journal_enable_debug __read_mostly;
55 EXPORT_SYMBOL(jbd2_journal_enable_debug);
57 module_param_named(jbd2_debug, jbd2_journal_enable_debug, ushort, 0644);
58 MODULE_PARM_DESC(jbd2_debug, "Debugging level for jbd2");
59 #endif
61 EXPORT_SYMBOL(jbd2_journal_extend);
62 EXPORT_SYMBOL(jbd2_journal_stop);
63 EXPORT_SYMBOL(jbd2_journal_lock_updates);
64 EXPORT_SYMBOL(jbd2_journal_unlock_updates);
65 EXPORT_SYMBOL(jbd2_journal_get_write_access);
66 EXPORT_SYMBOL(jbd2_journal_get_create_access);
67 EXPORT_SYMBOL(jbd2_journal_get_undo_access);
68 EXPORT_SYMBOL(jbd2_journal_set_triggers);
69 EXPORT_SYMBOL(jbd2_journal_dirty_metadata);
70 EXPORT_SYMBOL(jbd2_journal_forget);
71 #if 0
72 EXPORT_SYMBOL(journal_sync_buffer);
73 #endif
74 EXPORT_SYMBOL(jbd2_journal_flush);
75 EXPORT_SYMBOL(jbd2_journal_revoke);
77 EXPORT_SYMBOL(jbd2_journal_init_dev);
78 EXPORT_SYMBOL(jbd2_journal_init_inode);
79 EXPORT_SYMBOL(jbd2_journal_check_used_features);
80 EXPORT_SYMBOL(jbd2_journal_check_available_features);
81 EXPORT_SYMBOL(jbd2_journal_set_features);
82 EXPORT_SYMBOL(jbd2_journal_load);
83 EXPORT_SYMBOL(jbd2_journal_destroy);
84 EXPORT_SYMBOL(jbd2_journal_abort);
85 EXPORT_SYMBOL(jbd2_journal_errno);
86 EXPORT_SYMBOL(jbd2_journal_ack_err);
87 EXPORT_SYMBOL(jbd2_journal_clear_err);
88 EXPORT_SYMBOL(jbd2_log_wait_commit);
89 EXPORT_SYMBOL(jbd2_log_start_commit);
90 EXPORT_SYMBOL(jbd2_journal_start_commit);
91 EXPORT_SYMBOL(jbd2_journal_force_commit_nested);
92 EXPORT_SYMBOL(jbd2_journal_wipe);
93 EXPORT_SYMBOL(jbd2_journal_blocks_per_page);
94 EXPORT_SYMBOL(jbd2_journal_invalidatepage);
95 EXPORT_SYMBOL(jbd2_journal_try_to_free_buffers);
96 EXPORT_SYMBOL(jbd2_journal_force_commit);
97 EXPORT_SYMBOL(jbd2_journal_file_inode);
98 EXPORT_SYMBOL(jbd2_journal_init_jbd_inode);
99 EXPORT_SYMBOL(jbd2_journal_release_jbd_inode);
100 EXPORT_SYMBOL(jbd2_journal_begin_ordered_truncate);
101 EXPORT_SYMBOL(jbd2_inode_cache);
103 static void __journal_abort_soft (journal_t *journal, int errno);
104 static int jbd2_journal_create_slab(size_t slab_size);
106 #ifdef CONFIG_JBD2_DEBUG
107 void __jbd2_debug(int level, const char *file, const char *func,
108 unsigned int line, const char *fmt, ...)
110 struct va_format vaf;
111 va_list args;
113 if (level > jbd2_journal_enable_debug)
114 return;
115 va_start(args, fmt);
116 vaf.fmt = fmt;
117 vaf.va = &args;
118 printk(KERN_DEBUG "%s: (%s, %u): %pV\n", file, func, line, &vaf);
119 va_end(args);
121 EXPORT_SYMBOL(__jbd2_debug);
122 #endif
124 /* Checksumming functions */
125 static int jbd2_verify_csum_type(journal_t *j, journal_superblock_t *sb)
127 if (!jbd2_journal_has_csum_v2or3_feature(j))
128 return 1;
130 return sb->s_checksum_type == JBD2_CRC32C_CHKSUM;
133 static __be32 jbd2_superblock_csum(journal_t *j, journal_superblock_t *sb)
135 __u32 csum;
136 __be32 old_csum;
138 old_csum = sb->s_checksum;
139 sb->s_checksum = 0;
140 csum = jbd2_chksum(j, ~0, (char *)sb, sizeof(journal_superblock_t));
141 sb->s_checksum = old_csum;
143 return cpu_to_be32(csum);
146 static int jbd2_superblock_csum_verify(journal_t *j, journal_superblock_t *sb)
148 if (!jbd2_journal_has_csum_v2or3(j))
149 return 1;
151 return sb->s_checksum == jbd2_superblock_csum(j, sb);
154 static void jbd2_superblock_csum_set(journal_t *j, journal_superblock_t *sb)
156 if (!jbd2_journal_has_csum_v2or3(j))
157 return;
159 sb->s_checksum = jbd2_superblock_csum(j, sb);
163 * Helper function used to manage commit timeouts
166 static void commit_timeout(unsigned long __data)
168 struct task_struct * p = (struct task_struct *) __data;
170 wake_up_process(p);
174 * kjournald2: The main thread function used to manage a logging device
175 * journal.
177 * This kernel thread is responsible for two things:
179 * 1) COMMIT: Every so often we need to commit the current state of the
180 * filesystem to disk. The journal thread is responsible for writing
181 * all of the metadata buffers to disk.
183 * 2) CHECKPOINT: We cannot reuse a used section of the log file until all
184 * of the data in that part of the log has been rewritten elsewhere on
185 * the disk. Flushing these old buffers to reclaim space in the log is
186 * known as checkpointing, and this thread is responsible for that job.
189 static int kjournald2(void *arg)
191 journal_t *journal = arg;
192 transaction_t *transaction;
195 * Set up an interval timer which can be used to trigger a commit wakeup
196 * after the commit interval expires
198 setup_timer(&journal->j_commit_timer, commit_timeout,
199 (unsigned long)current);
201 set_freezable();
203 /* Record that the journal thread is running */
204 journal->j_task = current;
205 wake_up(&journal->j_wait_done_commit);
208 * And now, wait forever for commit wakeup events.
210 write_lock(&journal->j_state_lock);
212 loop:
213 if (journal->j_flags & JBD2_UNMOUNT)
214 goto end_loop;
216 jbd_debug(1, "commit_sequence=%d, commit_request=%d\n",
217 journal->j_commit_sequence, journal->j_commit_request);
219 if (journal->j_commit_sequence != journal->j_commit_request) {
220 jbd_debug(1, "OK, requests differ\n");
221 write_unlock(&journal->j_state_lock);
222 del_timer_sync(&journal->j_commit_timer);
223 jbd2_journal_commit_transaction(journal);
224 write_lock(&journal->j_state_lock);
225 goto loop;
228 wake_up(&journal->j_wait_done_commit);
229 if (freezing(current)) {
231 * The simpler the better. Flushing journal isn't a
232 * good idea, because that depends on threads that may
233 * be already stopped.
235 jbd_debug(1, "Now suspending kjournald2\n");
236 write_unlock(&journal->j_state_lock);
237 try_to_freeze();
238 write_lock(&journal->j_state_lock);
239 } else {
241 * We assume on resume that commits are already there,
242 * so we don't sleep
244 DEFINE_WAIT(wait);
245 int should_sleep = 1;
247 prepare_to_wait(&journal->j_wait_commit, &wait,
248 TASK_INTERRUPTIBLE);
249 if (journal->j_commit_sequence != journal->j_commit_request)
250 should_sleep = 0;
251 transaction = journal->j_running_transaction;
252 if (transaction && time_after_eq(jiffies,
253 transaction->t_expires))
254 should_sleep = 0;
255 if (journal->j_flags & JBD2_UNMOUNT)
256 should_sleep = 0;
257 if (should_sleep) {
258 write_unlock(&journal->j_state_lock);
259 schedule();
260 write_lock(&journal->j_state_lock);
262 finish_wait(&journal->j_wait_commit, &wait);
265 jbd_debug(1, "kjournald2 wakes\n");
268 * Were we woken up by a commit wakeup event?
270 transaction = journal->j_running_transaction;
271 if (transaction && time_after_eq(jiffies, transaction->t_expires)) {
272 journal->j_commit_request = transaction->t_tid;
273 jbd_debug(1, "woke because of timeout\n");
275 goto loop;
277 end_loop:
278 del_timer_sync(&journal->j_commit_timer);
279 journal->j_task = NULL;
280 wake_up(&journal->j_wait_done_commit);
281 jbd_debug(1, "Journal thread exiting.\n");
282 write_unlock(&journal->j_state_lock);
283 return 0;
286 static int jbd2_journal_start_thread(journal_t *journal)
288 struct task_struct *t;
290 t = kthread_run(kjournald2, journal, "jbd2/%s",
291 journal->j_devname);
292 if (IS_ERR(t))
293 return PTR_ERR(t);
295 wait_event(journal->j_wait_done_commit, journal->j_task != NULL);
296 return 0;
299 static void journal_kill_thread(journal_t *journal)
301 write_lock(&journal->j_state_lock);
302 journal->j_flags |= JBD2_UNMOUNT;
304 while (journal->j_task) {
305 write_unlock(&journal->j_state_lock);
306 wake_up(&journal->j_wait_commit);
307 wait_event(journal->j_wait_done_commit, journal->j_task == NULL);
308 write_lock(&journal->j_state_lock);
310 write_unlock(&journal->j_state_lock);
314 * jbd2_journal_write_metadata_buffer: write a metadata buffer to the journal.
316 * Writes a metadata buffer to a given disk block. The actual IO is not
317 * performed but a new buffer_head is constructed which labels the data
318 * to be written with the correct destination disk block.
320 * Any magic-number escaping which needs to be done will cause a
321 * copy-out here. If the buffer happens to start with the
322 * JBD2_MAGIC_NUMBER, then we can't write it to the log directly: the
323 * magic number is only written to the log for descripter blocks. In
324 * this case, we copy the data and replace the first word with 0, and we
325 * return a result code which indicates that this buffer needs to be
326 * marked as an escaped buffer in the corresponding log descriptor
327 * block. The missing word can then be restored when the block is read
328 * during recovery.
330 * If the source buffer has already been modified by a new transaction
331 * since we took the last commit snapshot, we use the frozen copy of
332 * that data for IO. If we end up using the existing buffer_head's data
333 * for the write, then we have to make sure nobody modifies it while the
334 * IO is in progress. do_get_write_access() handles this.
336 * The function returns a pointer to the buffer_head to be used for IO.
339 * Return value:
340 * <0: Error
341 * >=0: Finished OK
343 * On success:
344 * Bit 0 set == escape performed on the data
345 * Bit 1 set == buffer copy-out performed (kfree the data after IO)
348 int jbd2_journal_write_metadata_buffer(transaction_t *transaction,
349 struct journal_head *jh_in,
350 struct buffer_head **bh_out,
351 sector_t blocknr)
353 int need_copy_out = 0;
354 int done_copy_out = 0;
355 int do_escape = 0;
356 char *mapped_data;
357 struct buffer_head *new_bh;
358 struct page *new_page;
359 unsigned int new_offset;
360 struct buffer_head *bh_in = jh2bh(jh_in);
361 journal_t *journal = transaction->t_journal;
364 * The buffer really shouldn't be locked: only the current committing
365 * transaction is allowed to write it, so nobody else is allowed
366 * to do any IO.
368 * akpm: except if we're journalling data, and write() output is
369 * also part of a shared mapping, and another thread has
370 * decided to launch a writepage() against this buffer.
372 J_ASSERT_BH(bh_in, buffer_jbddirty(bh_in));
374 new_bh = alloc_buffer_head(GFP_NOFS|__GFP_NOFAIL);
376 /* keep subsequent assertions sane */
377 atomic_set(&new_bh->b_count, 1);
379 jbd_lock_bh_state(bh_in);
380 repeat:
382 * If a new transaction has already done a buffer copy-out, then
383 * we use that version of the data for the commit.
385 if (jh_in->b_frozen_data) {
386 done_copy_out = 1;
387 new_page = virt_to_page(jh_in->b_frozen_data);
388 new_offset = offset_in_page(jh_in->b_frozen_data);
389 } else {
390 new_page = jh2bh(jh_in)->b_page;
391 new_offset = offset_in_page(jh2bh(jh_in)->b_data);
394 mapped_data = kmap_atomic(new_page);
396 * Fire data frozen trigger if data already wasn't frozen. Do this
397 * before checking for escaping, as the trigger may modify the magic
398 * offset. If a copy-out happens afterwards, it will have the correct
399 * data in the buffer.
401 if (!done_copy_out)
402 jbd2_buffer_frozen_trigger(jh_in, mapped_data + new_offset,
403 jh_in->b_triggers);
406 * Check for escaping
408 if (*((__be32 *)(mapped_data + new_offset)) ==
409 cpu_to_be32(JBD2_MAGIC_NUMBER)) {
410 need_copy_out = 1;
411 do_escape = 1;
413 kunmap_atomic(mapped_data);
416 * Do we need to do a data copy?
418 if (need_copy_out && !done_copy_out) {
419 char *tmp;
421 jbd_unlock_bh_state(bh_in);
422 tmp = jbd2_alloc(bh_in->b_size, GFP_NOFS);
423 if (!tmp) {
424 brelse(new_bh);
425 return -ENOMEM;
427 jbd_lock_bh_state(bh_in);
428 if (jh_in->b_frozen_data) {
429 jbd2_free(tmp, bh_in->b_size);
430 goto repeat;
433 jh_in->b_frozen_data = tmp;
434 mapped_data = kmap_atomic(new_page);
435 memcpy(tmp, mapped_data + new_offset, bh_in->b_size);
436 kunmap_atomic(mapped_data);
438 new_page = virt_to_page(tmp);
439 new_offset = offset_in_page(tmp);
440 done_copy_out = 1;
443 * This isn't strictly necessary, as we're using frozen
444 * data for the escaping, but it keeps consistency with
445 * b_frozen_data usage.
447 jh_in->b_frozen_triggers = jh_in->b_triggers;
451 * Did we need to do an escaping? Now we've done all the
452 * copying, we can finally do so.
454 if (do_escape) {
455 mapped_data = kmap_atomic(new_page);
456 *((unsigned int *)(mapped_data + new_offset)) = 0;
457 kunmap_atomic(mapped_data);
460 set_bh_page(new_bh, new_page, new_offset);
461 new_bh->b_size = bh_in->b_size;
462 new_bh->b_bdev = journal->j_dev;
463 new_bh->b_blocknr = blocknr;
464 new_bh->b_private = bh_in;
465 set_buffer_mapped(new_bh);
466 set_buffer_dirty(new_bh);
468 *bh_out = new_bh;
471 * The to-be-written buffer needs to get moved to the io queue,
472 * and the original buffer whose contents we are shadowing or
473 * copying is moved to the transaction's shadow queue.
475 JBUFFER_TRACE(jh_in, "file as BJ_Shadow");
476 spin_lock(&journal->j_list_lock);
477 __jbd2_journal_file_buffer(jh_in, transaction, BJ_Shadow);
478 spin_unlock(&journal->j_list_lock);
479 set_buffer_shadow(bh_in);
480 jbd_unlock_bh_state(bh_in);
482 return do_escape | (done_copy_out << 1);
486 * Allocation code for the journal file. Manage the space left in the
487 * journal, so that we can begin checkpointing when appropriate.
491 * Called with j_state_lock locked for writing.
492 * Returns true if a transaction commit was started.
494 int __jbd2_log_start_commit(journal_t *journal, tid_t target)
496 /* Return if the txn has already requested to be committed */
497 if (journal->j_commit_request == target)
498 return 0;
501 * The only transaction we can possibly wait upon is the
502 * currently running transaction (if it exists). Otherwise,
503 * the target tid must be an old one.
505 if (journal->j_running_transaction &&
506 journal->j_running_transaction->t_tid == target) {
508 * We want a new commit: OK, mark the request and wakeup the
509 * commit thread. We do _not_ do the commit ourselves.
512 journal->j_commit_request = target;
513 jbd_debug(1, "JBD2: requesting commit %d/%d\n",
514 journal->j_commit_request,
515 journal->j_commit_sequence);
516 journal->j_running_transaction->t_requested = jiffies;
517 wake_up(&journal->j_wait_commit);
518 return 1;
519 } else if (!tid_geq(journal->j_commit_request, target))
520 /* This should never happen, but if it does, preserve
521 the evidence before kjournald goes into a loop and
522 increments j_commit_sequence beyond all recognition. */
523 WARN_ONCE(1, "JBD2: bad log_start_commit: %u %u %u %u\n",
524 journal->j_commit_request,
525 journal->j_commit_sequence,
526 target, journal->j_running_transaction ?
527 journal->j_running_transaction->t_tid : 0);
528 return 0;
531 int jbd2_log_start_commit(journal_t *journal, tid_t tid)
533 int ret;
535 write_lock(&journal->j_state_lock);
536 ret = __jbd2_log_start_commit(journal, tid);
537 write_unlock(&journal->j_state_lock);
538 return ret;
542 * Force and wait any uncommitted transactions. We can only force the running
543 * transaction if we don't have an active handle, otherwise, we will deadlock.
544 * Returns: <0 in case of error,
545 * 0 if nothing to commit,
546 * 1 if transaction was successfully committed.
548 static int __jbd2_journal_force_commit(journal_t *journal)
550 transaction_t *transaction = NULL;
551 tid_t tid;
552 int need_to_start = 0, ret = 0;
554 read_lock(&journal->j_state_lock);
555 if (journal->j_running_transaction && !current->journal_info) {
556 transaction = journal->j_running_transaction;
557 if (!tid_geq(journal->j_commit_request, transaction->t_tid))
558 need_to_start = 1;
559 } else if (journal->j_committing_transaction)
560 transaction = journal->j_committing_transaction;
562 if (!transaction) {
563 /* Nothing to commit */
564 read_unlock(&journal->j_state_lock);
565 return 0;
567 tid = transaction->t_tid;
568 read_unlock(&journal->j_state_lock);
569 if (need_to_start)
570 jbd2_log_start_commit(journal, tid);
571 ret = jbd2_log_wait_commit(journal, tid);
572 if (!ret)
573 ret = 1;
575 return ret;
579 * Force and wait upon a commit if the calling process is not within
580 * transaction. This is used for forcing out undo-protected data which contains
581 * bitmaps, when the fs is running out of space.
583 * @journal: journal to force
584 * Returns true if progress was made.
586 int jbd2_journal_force_commit_nested(journal_t *journal)
588 int ret;
590 ret = __jbd2_journal_force_commit(journal);
591 return ret > 0;
595 * int journal_force_commit() - force any uncommitted transactions
596 * @journal: journal to force
598 * Caller want unconditional commit. We can only force the running transaction
599 * if we don't have an active handle, otherwise, we will deadlock.
601 int jbd2_journal_force_commit(journal_t *journal)
603 int ret;
605 J_ASSERT(!current->journal_info);
606 ret = __jbd2_journal_force_commit(journal);
607 if (ret > 0)
608 ret = 0;
609 return ret;
613 * Start a commit of the current running transaction (if any). Returns true
614 * if a transaction is going to be committed (or is currently already
615 * committing), and fills its tid in at *ptid
617 int jbd2_journal_start_commit(journal_t *journal, tid_t *ptid)
619 int ret = 0;
621 write_lock(&journal->j_state_lock);
622 if (journal->j_running_transaction) {
623 tid_t tid = journal->j_running_transaction->t_tid;
625 __jbd2_log_start_commit(journal, tid);
626 /* There's a running transaction and we've just made sure
627 * it's commit has been scheduled. */
628 if (ptid)
629 *ptid = tid;
630 ret = 1;
631 } else if (journal->j_committing_transaction) {
633 * If commit has been started, then we have to wait for
634 * completion of that transaction.
636 if (ptid)
637 *ptid = journal->j_committing_transaction->t_tid;
638 ret = 1;
640 write_unlock(&journal->j_state_lock);
641 return ret;
645 * Return 1 if a given transaction has not yet sent barrier request
646 * connected with a transaction commit. If 0 is returned, transaction
647 * may or may not have sent the barrier. Used to avoid sending barrier
648 * twice in common cases.
650 int jbd2_trans_will_send_data_barrier(journal_t *journal, tid_t tid)
652 int ret = 0;
653 transaction_t *commit_trans;
655 if (!(journal->j_flags & JBD2_BARRIER))
656 return 0;
657 read_lock(&journal->j_state_lock);
658 /* Transaction already committed? */
659 if (tid_geq(journal->j_commit_sequence, tid))
660 goto out;
661 commit_trans = journal->j_committing_transaction;
662 if (!commit_trans || commit_trans->t_tid != tid) {
663 ret = 1;
664 goto out;
667 * Transaction is being committed and we already proceeded to
668 * submitting a flush to fs partition?
670 if (journal->j_fs_dev != journal->j_dev) {
671 if (!commit_trans->t_need_data_flush ||
672 commit_trans->t_state >= T_COMMIT_DFLUSH)
673 goto out;
674 } else {
675 if (commit_trans->t_state >= T_COMMIT_JFLUSH)
676 goto out;
678 ret = 1;
679 out:
680 read_unlock(&journal->j_state_lock);
681 return ret;
683 EXPORT_SYMBOL(jbd2_trans_will_send_data_barrier);
686 * Wait for a specified commit to complete.
687 * The caller may not hold the journal lock.
689 int jbd2_log_wait_commit(journal_t *journal, tid_t tid)
691 int err = 0;
693 read_lock(&journal->j_state_lock);
694 #ifdef CONFIG_JBD2_DEBUG
695 if (!tid_geq(journal->j_commit_request, tid)) {
696 printk(KERN_ERR
697 "%s: error: j_commit_request=%d, tid=%d\n",
698 __func__, journal->j_commit_request, tid);
700 #endif
701 while (tid_gt(tid, journal->j_commit_sequence)) {
702 jbd_debug(1, "JBD2: want %d, j_commit_sequence=%d\n",
703 tid, journal->j_commit_sequence);
704 read_unlock(&journal->j_state_lock);
705 wake_up(&journal->j_wait_commit);
706 wait_event(journal->j_wait_done_commit,
707 !tid_gt(tid, journal->j_commit_sequence));
708 read_lock(&journal->j_state_lock);
710 read_unlock(&journal->j_state_lock);
712 if (unlikely(is_journal_aborted(journal)))
713 err = -EIO;
714 return err;
718 * When this function returns the transaction corresponding to tid
719 * will be completed. If the transaction has currently running, start
720 * committing that transaction before waiting for it to complete. If
721 * the transaction id is stale, it is by definition already completed,
722 * so just return SUCCESS.
724 int jbd2_complete_transaction(journal_t *journal, tid_t tid)
726 int need_to_wait = 1;
728 read_lock(&journal->j_state_lock);
729 if (journal->j_running_transaction &&
730 journal->j_running_transaction->t_tid == tid) {
731 if (journal->j_commit_request != tid) {
732 /* transaction not yet started, so request it */
733 read_unlock(&journal->j_state_lock);
734 jbd2_log_start_commit(journal, tid);
735 goto wait_commit;
737 } else if (!(journal->j_committing_transaction &&
738 journal->j_committing_transaction->t_tid == tid))
739 need_to_wait = 0;
740 read_unlock(&journal->j_state_lock);
741 if (!need_to_wait)
742 return 0;
743 wait_commit:
744 return jbd2_log_wait_commit(journal, tid);
746 EXPORT_SYMBOL(jbd2_complete_transaction);
749 * Log buffer allocation routines:
752 int jbd2_journal_next_log_block(journal_t *journal, unsigned long long *retp)
754 unsigned long blocknr;
756 write_lock(&journal->j_state_lock);
757 J_ASSERT(journal->j_free > 1);
759 blocknr = journal->j_head;
760 journal->j_head++;
761 journal->j_free--;
762 if (journal->j_head == journal->j_last)
763 journal->j_head = journal->j_first;
764 write_unlock(&journal->j_state_lock);
765 return jbd2_journal_bmap(journal, blocknr, retp);
769 * Conversion of logical to physical block numbers for the journal
771 * On external journals the journal blocks are identity-mapped, so
772 * this is a no-op. If needed, we can use j_blk_offset - everything is
773 * ready.
775 int jbd2_journal_bmap(journal_t *journal, unsigned long blocknr,
776 unsigned long long *retp)
778 int err = 0;
779 unsigned long long ret;
781 if (journal->j_inode) {
782 ret = bmap(journal->j_inode, blocknr);
783 if (ret)
784 *retp = ret;
785 else {
786 printk(KERN_ALERT "%s: journal block not found "
787 "at offset %lu on %s\n",
788 __func__, blocknr, journal->j_devname);
789 err = -EIO;
790 __journal_abort_soft(journal, err);
792 } else {
793 *retp = blocknr; /* +journal->j_blk_offset */
795 return err;
799 * We play buffer_head aliasing tricks to write data/metadata blocks to
800 * the journal without copying their contents, but for journal
801 * descriptor blocks we do need to generate bona fide buffers.
803 * After the caller of jbd2_journal_get_descriptor_buffer() has finished modifying
804 * the buffer's contents they really should run flush_dcache_page(bh->b_page).
805 * But we don't bother doing that, so there will be coherency problems with
806 * mmaps of blockdevs which hold live JBD-controlled filesystems.
808 struct buffer_head *jbd2_journal_get_descriptor_buffer(journal_t *journal)
810 struct buffer_head *bh;
811 unsigned long long blocknr;
812 int err;
814 err = jbd2_journal_next_log_block(journal, &blocknr);
816 if (err)
817 return NULL;
819 bh = __getblk(journal->j_dev, blocknr, journal->j_blocksize);
820 if (!bh)
821 return NULL;
822 lock_buffer(bh);
823 memset(bh->b_data, 0, journal->j_blocksize);
824 set_buffer_uptodate(bh);
825 unlock_buffer(bh);
826 BUFFER_TRACE(bh, "return this buffer");
827 return bh;
831 * Return tid of the oldest transaction in the journal and block in the journal
832 * where the transaction starts.
834 * If the journal is now empty, return which will be the next transaction ID
835 * we will write and where will that transaction start.
837 * The return value is 0 if journal tail cannot be pushed any further, 1 if
838 * it can.
840 int jbd2_journal_get_log_tail(journal_t *journal, tid_t *tid,
841 unsigned long *block)
843 transaction_t *transaction;
844 int ret;
846 read_lock(&journal->j_state_lock);
847 spin_lock(&journal->j_list_lock);
848 transaction = journal->j_checkpoint_transactions;
849 if (transaction) {
850 *tid = transaction->t_tid;
851 *block = transaction->t_log_start;
852 } else if ((transaction = journal->j_committing_transaction) != NULL) {
853 *tid = transaction->t_tid;
854 *block = transaction->t_log_start;
855 } else if ((transaction = journal->j_running_transaction) != NULL) {
856 *tid = transaction->t_tid;
857 *block = journal->j_head;
858 } else {
859 *tid = journal->j_transaction_sequence;
860 *block = journal->j_head;
862 ret = tid_gt(*tid, journal->j_tail_sequence);
863 spin_unlock(&journal->j_list_lock);
864 read_unlock(&journal->j_state_lock);
866 return ret;
870 * Update information in journal structure and in on disk journal superblock
871 * about log tail. This function does not check whether information passed in
872 * really pushes log tail further. It's responsibility of the caller to make
873 * sure provided log tail information is valid (e.g. by holding
874 * j_checkpoint_mutex all the time between computing log tail and calling this
875 * function as is the case with jbd2_cleanup_journal_tail()).
877 * Requires j_checkpoint_mutex
879 int __jbd2_update_log_tail(journal_t *journal, tid_t tid, unsigned long block)
881 unsigned long freed;
882 int ret;
884 BUG_ON(!mutex_is_locked(&journal->j_checkpoint_mutex));
887 * We cannot afford for write to remain in drive's caches since as
888 * soon as we update j_tail, next transaction can start reusing journal
889 * space and if we lose sb update during power failure we'd replay
890 * old transaction with possibly newly overwritten data.
892 ret = jbd2_journal_update_sb_log_tail(journal, tid, block, WRITE_FUA);
893 if (ret)
894 goto out;
896 write_lock(&journal->j_state_lock);
897 freed = block - journal->j_tail;
898 if (block < journal->j_tail)
899 freed += journal->j_last - journal->j_first;
901 trace_jbd2_update_log_tail(journal, tid, block, freed);
902 jbd_debug(1,
903 "Cleaning journal tail from %d to %d (offset %lu), "
904 "freeing %lu\n",
905 journal->j_tail_sequence, tid, block, freed);
907 journal->j_free += freed;
908 journal->j_tail_sequence = tid;
909 journal->j_tail = block;
910 write_unlock(&journal->j_state_lock);
912 out:
913 return ret;
917 * This is a variation of __jbd2_update_log_tail which checks for validity of
918 * provided log tail and locks j_checkpoint_mutex. So it is safe against races
919 * with other threads updating log tail.
921 void jbd2_update_log_tail(journal_t *journal, tid_t tid, unsigned long block)
923 mutex_lock(&journal->j_checkpoint_mutex);
924 if (tid_gt(tid, journal->j_tail_sequence))
925 __jbd2_update_log_tail(journal, tid, block);
926 mutex_unlock(&journal->j_checkpoint_mutex);
929 struct jbd2_stats_proc_session {
930 journal_t *journal;
931 struct transaction_stats_s *stats;
932 int start;
933 int max;
936 static void *jbd2_seq_info_start(struct seq_file *seq, loff_t *pos)
938 return *pos ? NULL : SEQ_START_TOKEN;
941 static void *jbd2_seq_info_next(struct seq_file *seq, void *v, loff_t *pos)
943 return NULL;
946 static int jbd2_seq_info_show(struct seq_file *seq, void *v)
948 struct jbd2_stats_proc_session *s = seq->private;
950 if (v != SEQ_START_TOKEN)
951 return 0;
952 seq_printf(seq, "%lu transactions (%lu requested), "
953 "each up to %u blocks\n",
954 s->stats->ts_tid, s->stats->ts_requested,
955 s->journal->j_max_transaction_buffers);
956 if (s->stats->ts_tid == 0)
957 return 0;
958 seq_printf(seq, "average: \n %ums waiting for transaction\n",
959 jiffies_to_msecs(s->stats->run.rs_wait / s->stats->ts_tid));
960 seq_printf(seq, " %ums request delay\n",
961 (s->stats->ts_requested == 0) ? 0 :
962 jiffies_to_msecs(s->stats->run.rs_request_delay /
963 s->stats->ts_requested));
964 seq_printf(seq, " %ums running transaction\n",
965 jiffies_to_msecs(s->stats->run.rs_running / s->stats->ts_tid));
966 seq_printf(seq, " %ums transaction was being locked\n",
967 jiffies_to_msecs(s->stats->run.rs_locked / s->stats->ts_tid));
968 seq_printf(seq, " %ums flushing data (in ordered mode)\n",
969 jiffies_to_msecs(s->stats->run.rs_flushing / s->stats->ts_tid));
970 seq_printf(seq, " %ums logging transaction\n",
971 jiffies_to_msecs(s->stats->run.rs_logging / s->stats->ts_tid));
972 seq_printf(seq, " %lluus average transaction commit time\n",
973 div_u64(s->journal->j_average_commit_time, 1000));
974 seq_printf(seq, " %lu handles per transaction\n",
975 s->stats->run.rs_handle_count / s->stats->ts_tid);
976 seq_printf(seq, " %lu blocks per transaction\n",
977 s->stats->run.rs_blocks / s->stats->ts_tid);
978 seq_printf(seq, " %lu logged blocks per transaction\n",
979 s->stats->run.rs_blocks_logged / s->stats->ts_tid);
980 return 0;
983 static void jbd2_seq_info_stop(struct seq_file *seq, void *v)
987 static const struct seq_operations jbd2_seq_info_ops = {
988 .start = jbd2_seq_info_start,
989 .next = jbd2_seq_info_next,
990 .stop = jbd2_seq_info_stop,
991 .show = jbd2_seq_info_show,
994 static int jbd2_seq_info_open(struct inode *inode, struct file *file)
996 journal_t *journal = PDE_DATA(inode);
997 struct jbd2_stats_proc_session *s;
998 int rc, size;
1000 s = kmalloc(sizeof(*s), GFP_KERNEL);
1001 if (s == NULL)
1002 return -ENOMEM;
1003 size = sizeof(struct transaction_stats_s);
1004 s->stats = kmalloc(size, GFP_KERNEL);
1005 if (s->stats == NULL) {
1006 kfree(s);
1007 return -ENOMEM;
1009 spin_lock(&journal->j_history_lock);
1010 memcpy(s->stats, &journal->j_stats, size);
1011 s->journal = journal;
1012 spin_unlock(&journal->j_history_lock);
1014 rc = seq_open(file, &jbd2_seq_info_ops);
1015 if (rc == 0) {
1016 struct seq_file *m = file->private_data;
1017 m->private = s;
1018 } else {
1019 kfree(s->stats);
1020 kfree(s);
1022 return rc;
1026 static int jbd2_seq_info_release(struct inode *inode, struct file *file)
1028 struct seq_file *seq = file->private_data;
1029 struct jbd2_stats_proc_session *s = seq->private;
1030 kfree(s->stats);
1031 kfree(s);
1032 return seq_release(inode, file);
1035 static const struct file_operations jbd2_seq_info_fops = {
1036 .owner = THIS_MODULE,
1037 .open = jbd2_seq_info_open,
1038 .read = seq_read,
1039 .llseek = seq_lseek,
1040 .release = jbd2_seq_info_release,
1043 static struct proc_dir_entry *proc_jbd2_stats;
1045 static void jbd2_stats_proc_init(journal_t *journal)
1047 journal->j_proc_entry = proc_mkdir(journal->j_devname, proc_jbd2_stats);
1048 if (journal->j_proc_entry) {
1049 proc_create_data("info", S_IRUGO, journal->j_proc_entry,
1050 &jbd2_seq_info_fops, journal);
1054 static void jbd2_stats_proc_exit(journal_t *journal)
1056 remove_proc_entry("info", journal->j_proc_entry);
1057 remove_proc_entry(journal->j_devname, proc_jbd2_stats);
1061 * Management for journal control blocks: functions to create and
1062 * destroy journal_t structures, and to initialise and read existing
1063 * journal blocks from disk. */
1065 /* First: create and setup a journal_t object in memory. We initialise
1066 * very few fields yet: that has to wait until we have created the
1067 * journal structures from from scratch, or loaded them from disk. */
1069 static journal_t * journal_init_common (void)
1071 journal_t *journal;
1072 int err;
1074 journal = kzalloc(sizeof(*journal), GFP_KERNEL);
1075 if (!journal)
1076 return NULL;
1078 init_waitqueue_head(&journal->j_wait_transaction_locked);
1079 init_waitqueue_head(&journal->j_wait_done_commit);
1080 init_waitqueue_head(&journal->j_wait_commit);
1081 init_waitqueue_head(&journal->j_wait_updates);
1082 init_waitqueue_head(&journal->j_wait_reserved);
1083 mutex_init(&journal->j_barrier);
1084 mutex_init(&journal->j_checkpoint_mutex);
1085 spin_lock_init(&journal->j_revoke_lock);
1086 spin_lock_init(&journal->j_list_lock);
1087 rwlock_init(&journal->j_state_lock);
1089 journal->j_commit_interval = (HZ * JBD2_DEFAULT_MAX_COMMIT_AGE);
1090 journal->j_min_batch_time = 0;
1091 journal->j_max_batch_time = 15000; /* 15ms */
1092 atomic_set(&journal->j_reserved_credits, 0);
1094 /* The journal is marked for error until we succeed with recovery! */
1095 journal->j_flags = JBD2_ABORT;
1097 /* Set up a default-sized revoke table for the new mount. */
1098 err = jbd2_journal_init_revoke(journal, JOURNAL_REVOKE_DEFAULT_HASH);
1099 if (err) {
1100 kfree(journal);
1101 return NULL;
1104 spin_lock_init(&journal->j_history_lock);
1106 return journal;
1109 /* jbd2_journal_init_dev and jbd2_journal_init_inode:
1111 * Create a journal structure assigned some fixed set of disk blocks to
1112 * the journal. We don't actually touch those disk blocks yet, but we
1113 * need to set up all of the mapping information to tell the journaling
1114 * system where the journal blocks are.
1119 * journal_t * jbd2_journal_init_dev() - creates and initialises a journal structure
1120 * @bdev: Block device on which to create the journal
1121 * @fs_dev: Device which hold journalled filesystem for this journal.
1122 * @start: Block nr Start of journal.
1123 * @len: Length of the journal in blocks.
1124 * @blocksize: blocksize of journalling device
1126 * Returns: a newly created journal_t *
1128 * jbd2_journal_init_dev creates a journal which maps a fixed contiguous
1129 * range of blocks on an arbitrary block device.
1132 journal_t * jbd2_journal_init_dev(struct block_device *bdev,
1133 struct block_device *fs_dev,
1134 unsigned long long start, int len, int blocksize)
1136 journal_t *journal = journal_init_common();
1137 struct buffer_head *bh;
1138 int n;
1140 if (!journal)
1141 return NULL;
1143 /* journal descriptor can store up to n blocks -bzzz */
1144 journal->j_blocksize = blocksize;
1145 journal->j_dev = bdev;
1146 journal->j_fs_dev = fs_dev;
1147 journal->j_blk_offset = start;
1148 journal->j_maxlen = len;
1149 bdevname(journal->j_dev, journal->j_devname);
1150 strreplace(journal->j_devname, '/', '!');
1151 jbd2_stats_proc_init(journal);
1152 n = journal->j_blocksize / sizeof(journal_block_tag_t);
1153 journal->j_wbufsize = n;
1154 journal->j_wbuf = kmalloc(n * sizeof(struct buffer_head*), GFP_KERNEL);
1155 if (!journal->j_wbuf) {
1156 printk(KERN_ERR "%s: Can't allocate bhs for commit thread\n",
1157 __func__);
1158 goto out_err;
1161 bh = __getblk(journal->j_dev, start, journal->j_blocksize);
1162 if (!bh) {
1163 printk(KERN_ERR
1164 "%s: Cannot get buffer for journal superblock\n",
1165 __func__);
1166 goto out_err;
1168 journal->j_sb_buffer = bh;
1169 journal->j_superblock = (journal_superblock_t *)bh->b_data;
1171 return journal;
1172 out_err:
1173 kfree(journal->j_wbuf);
1174 jbd2_stats_proc_exit(journal);
1175 kfree(journal);
1176 return NULL;
1180 * journal_t * jbd2_journal_init_inode () - creates a journal which maps to a inode.
1181 * @inode: An inode to create the journal in
1183 * jbd2_journal_init_inode creates a journal which maps an on-disk inode as
1184 * the journal. The inode must exist already, must support bmap() and
1185 * must have all data blocks preallocated.
1187 journal_t * jbd2_journal_init_inode (struct inode *inode)
1189 struct buffer_head *bh;
1190 journal_t *journal = journal_init_common();
1191 char *p;
1192 int err;
1193 int n;
1194 unsigned long long blocknr;
1196 if (!journal)
1197 return NULL;
1199 journal->j_dev = journal->j_fs_dev = inode->i_sb->s_bdev;
1200 journal->j_inode = inode;
1201 bdevname(journal->j_dev, journal->j_devname);
1202 p = strreplace(journal->j_devname, '/', '!');
1203 sprintf(p, "-%lu", journal->j_inode->i_ino);
1204 jbd_debug(1,
1205 "journal %p: inode %s/%ld, size %Ld, bits %d, blksize %ld\n",
1206 journal, inode->i_sb->s_id, inode->i_ino,
1207 (long long) inode->i_size,
1208 inode->i_sb->s_blocksize_bits, inode->i_sb->s_blocksize);
1210 journal->j_maxlen = inode->i_size >> inode->i_sb->s_blocksize_bits;
1211 journal->j_blocksize = inode->i_sb->s_blocksize;
1212 jbd2_stats_proc_init(journal);
1214 /* journal descriptor can store up to n blocks -bzzz */
1215 n = journal->j_blocksize / sizeof(journal_block_tag_t);
1216 journal->j_wbufsize = n;
1217 journal->j_wbuf = kmalloc(n * sizeof(struct buffer_head*), GFP_KERNEL);
1218 if (!journal->j_wbuf) {
1219 printk(KERN_ERR "%s: Can't allocate bhs for commit thread\n",
1220 __func__);
1221 goto out_err;
1224 err = jbd2_journal_bmap(journal, 0, &blocknr);
1225 /* If that failed, give up */
1226 if (err) {
1227 printk(KERN_ERR "%s: Cannot locate journal superblock\n",
1228 __func__);
1229 goto out_err;
1232 bh = getblk_unmovable(journal->j_dev, blocknr, journal->j_blocksize);
1233 if (!bh) {
1234 printk(KERN_ERR
1235 "%s: Cannot get buffer for journal superblock\n",
1236 __func__);
1237 goto out_err;
1239 journal->j_sb_buffer = bh;
1240 journal->j_superblock = (journal_superblock_t *)bh->b_data;
1242 return journal;
1243 out_err:
1244 kfree(journal->j_wbuf);
1245 jbd2_stats_proc_exit(journal);
1246 kfree(journal);
1247 return NULL;
1251 * If the journal init or create aborts, we need to mark the journal
1252 * superblock as being NULL to prevent the journal destroy from writing
1253 * back a bogus superblock.
1255 static void journal_fail_superblock (journal_t *journal)
1257 struct buffer_head *bh = journal->j_sb_buffer;
1258 brelse(bh);
1259 journal->j_sb_buffer = NULL;
1263 * Given a journal_t structure, initialise the various fields for
1264 * startup of a new journaling session. We use this both when creating
1265 * a journal, and after recovering an old journal to reset it for
1266 * subsequent use.
1269 static int journal_reset(journal_t *journal)
1271 journal_superblock_t *sb = journal->j_superblock;
1272 unsigned long long first, last;
1274 first = be32_to_cpu(sb->s_first);
1275 last = be32_to_cpu(sb->s_maxlen);
1276 if (first + JBD2_MIN_JOURNAL_BLOCKS > last + 1) {
1277 printk(KERN_ERR "JBD2: Journal too short (blocks %llu-%llu).\n",
1278 first, last);
1279 journal_fail_superblock(journal);
1280 return -EINVAL;
1283 journal->j_first = first;
1284 journal->j_last = last;
1286 journal->j_head = first;
1287 journal->j_tail = first;
1288 journal->j_free = last - first;
1290 journal->j_tail_sequence = journal->j_transaction_sequence;
1291 journal->j_commit_sequence = journal->j_transaction_sequence - 1;
1292 journal->j_commit_request = journal->j_commit_sequence;
1294 journal->j_max_transaction_buffers = journal->j_maxlen / 4;
1297 * As a special case, if the on-disk copy is already marked as needing
1298 * no recovery (s_start == 0), then we can safely defer the superblock
1299 * update until the next commit by setting JBD2_FLUSHED. This avoids
1300 * attempting a write to a potential-readonly device.
1302 if (sb->s_start == 0) {
1303 jbd_debug(1, "JBD2: Skipping superblock update on recovered sb "
1304 "(start %ld, seq %d, errno %d)\n",
1305 journal->j_tail, journal->j_tail_sequence,
1306 journal->j_errno);
1307 journal->j_flags |= JBD2_FLUSHED;
1308 } else {
1309 /* Lock here to make assertions happy... */
1310 mutex_lock(&journal->j_checkpoint_mutex);
1312 * Update log tail information. We use WRITE_FUA since new
1313 * transaction will start reusing journal space and so we
1314 * must make sure information about current log tail is on
1315 * disk before that.
1317 jbd2_journal_update_sb_log_tail(journal,
1318 journal->j_tail_sequence,
1319 journal->j_tail,
1320 WRITE_FUA);
1321 mutex_unlock(&journal->j_checkpoint_mutex);
1323 return jbd2_journal_start_thread(journal);
1326 static int jbd2_write_superblock(journal_t *journal, int write_op)
1328 struct buffer_head *bh = journal->j_sb_buffer;
1329 journal_superblock_t *sb = journal->j_superblock;
1330 int ret;
1332 trace_jbd2_write_superblock(journal, write_op);
1333 if (!(journal->j_flags & JBD2_BARRIER))
1334 write_op &= ~(REQ_FUA | REQ_FLUSH);
1335 lock_buffer(bh);
1336 if (buffer_write_io_error(bh)) {
1338 * Oh, dear. A previous attempt to write the journal
1339 * superblock failed. This could happen because the
1340 * USB device was yanked out. Or it could happen to
1341 * be a transient write error and maybe the block will
1342 * be remapped. Nothing we can do but to retry the
1343 * write and hope for the best.
1345 printk(KERN_ERR "JBD2: previous I/O error detected "
1346 "for journal superblock update for %s.\n",
1347 journal->j_devname);
1348 clear_buffer_write_io_error(bh);
1349 set_buffer_uptodate(bh);
1351 jbd2_superblock_csum_set(journal, sb);
1352 get_bh(bh);
1353 bh->b_end_io = end_buffer_write_sync;
1354 ret = submit_bh(write_op, bh);
1355 wait_on_buffer(bh);
1356 if (buffer_write_io_error(bh)) {
1357 clear_buffer_write_io_error(bh);
1358 set_buffer_uptodate(bh);
1359 ret = -EIO;
1361 if (ret) {
1362 printk(KERN_ERR "JBD2: Error %d detected when updating "
1363 "journal superblock for %s.\n", ret,
1364 journal->j_devname);
1365 jbd2_journal_abort(journal, ret);
1368 return ret;
1372 * jbd2_journal_update_sb_log_tail() - Update log tail in journal sb on disk.
1373 * @journal: The journal to update.
1374 * @tail_tid: TID of the new transaction at the tail of the log
1375 * @tail_block: The first block of the transaction at the tail of the log
1376 * @write_op: With which operation should we write the journal sb
1378 * Update a journal's superblock information about log tail and write it to
1379 * disk, waiting for the IO to complete.
1381 int jbd2_journal_update_sb_log_tail(journal_t *journal, tid_t tail_tid,
1382 unsigned long tail_block, int write_op)
1384 journal_superblock_t *sb = journal->j_superblock;
1385 int ret;
1387 if (is_journal_aborted(journal))
1388 return -EIO;
1390 BUG_ON(!mutex_is_locked(&journal->j_checkpoint_mutex));
1391 jbd_debug(1, "JBD2: updating superblock (start %lu, seq %u)\n",
1392 tail_block, tail_tid);
1394 sb->s_sequence = cpu_to_be32(tail_tid);
1395 sb->s_start = cpu_to_be32(tail_block);
1397 ret = jbd2_write_superblock(journal, write_op);
1398 if (ret)
1399 goto out;
1401 /* Log is no longer empty */
1402 write_lock(&journal->j_state_lock);
1403 WARN_ON(!sb->s_sequence);
1404 journal->j_flags &= ~JBD2_FLUSHED;
1405 write_unlock(&journal->j_state_lock);
1407 out:
1408 return ret;
1412 * jbd2_mark_journal_empty() - Mark on disk journal as empty.
1413 * @journal: The journal to update.
1414 * @write_op: With which operation should we write the journal sb
1416 * Update a journal's dynamic superblock fields to show that journal is empty.
1417 * Write updated superblock to disk waiting for IO to complete.
1419 static void jbd2_mark_journal_empty(journal_t *journal, int write_op)
1421 journal_superblock_t *sb = journal->j_superblock;
1423 BUG_ON(!mutex_is_locked(&journal->j_checkpoint_mutex));
1424 read_lock(&journal->j_state_lock);
1425 /* Is it already empty? */
1426 if (sb->s_start == 0) {
1427 read_unlock(&journal->j_state_lock);
1428 return;
1430 jbd_debug(1, "JBD2: Marking journal as empty (seq %d)\n",
1431 journal->j_tail_sequence);
1433 sb->s_sequence = cpu_to_be32(journal->j_tail_sequence);
1434 sb->s_start = cpu_to_be32(0);
1435 read_unlock(&journal->j_state_lock);
1437 jbd2_write_superblock(journal, write_op);
1439 /* Log is no longer empty */
1440 write_lock(&journal->j_state_lock);
1441 journal->j_flags |= JBD2_FLUSHED;
1442 write_unlock(&journal->j_state_lock);
1447 * jbd2_journal_update_sb_errno() - Update error in the journal.
1448 * @journal: The journal to update.
1450 * Update a journal's errno. Write updated superblock to disk waiting for IO
1451 * to complete.
1453 void jbd2_journal_update_sb_errno(journal_t *journal)
1455 journal_superblock_t *sb = journal->j_superblock;
1457 read_lock(&journal->j_state_lock);
1458 jbd_debug(1, "JBD2: updating superblock error (errno %d)\n",
1459 journal->j_errno);
1460 sb->s_errno = cpu_to_be32(journal->j_errno);
1461 read_unlock(&journal->j_state_lock);
1463 jbd2_write_superblock(journal, WRITE_FUA);
1465 EXPORT_SYMBOL(jbd2_journal_update_sb_errno);
1468 * Read the superblock for a given journal, performing initial
1469 * validation of the format.
1471 static int journal_get_superblock(journal_t *journal)
1473 struct buffer_head *bh;
1474 journal_superblock_t *sb;
1475 int err = -EIO;
1477 bh = journal->j_sb_buffer;
1479 J_ASSERT(bh != NULL);
1480 if (!buffer_uptodate(bh)) {
1481 ll_rw_block(READ, 1, &bh);
1482 wait_on_buffer(bh);
1483 if (!buffer_uptodate(bh)) {
1484 printk(KERN_ERR
1485 "JBD2: IO error reading journal superblock\n");
1486 goto out;
1490 if (buffer_verified(bh))
1491 return 0;
1493 sb = journal->j_superblock;
1495 err = -EINVAL;
1497 if (sb->s_header.h_magic != cpu_to_be32(JBD2_MAGIC_NUMBER) ||
1498 sb->s_blocksize != cpu_to_be32(journal->j_blocksize)) {
1499 printk(KERN_WARNING "JBD2: no valid journal superblock found\n");
1500 goto out;
1503 switch(be32_to_cpu(sb->s_header.h_blocktype)) {
1504 case JBD2_SUPERBLOCK_V1:
1505 journal->j_format_version = 1;
1506 break;
1507 case JBD2_SUPERBLOCK_V2:
1508 journal->j_format_version = 2;
1509 break;
1510 default:
1511 printk(KERN_WARNING "JBD2: unrecognised superblock format ID\n");
1512 goto out;
1515 if (be32_to_cpu(sb->s_maxlen) < journal->j_maxlen)
1516 journal->j_maxlen = be32_to_cpu(sb->s_maxlen);
1517 else if (be32_to_cpu(sb->s_maxlen) > journal->j_maxlen) {
1518 printk(KERN_WARNING "JBD2: journal file too short\n");
1519 goto out;
1522 if (be32_to_cpu(sb->s_first) == 0 ||
1523 be32_to_cpu(sb->s_first) >= journal->j_maxlen) {
1524 printk(KERN_WARNING
1525 "JBD2: Invalid start block of journal: %u\n",
1526 be32_to_cpu(sb->s_first));
1527 goto out;
1530 if (jbd2_has_feature_csum2(journal) &&
1531 jbd2_has_feature_csum3(journal)) {
1532 /* Can't have checksum v2 and v3 at the same time! */
1533 printk(KERN_ERR "JBD2: Can't enable checksumming v2 and v3 "
1534 "at the same time!\n");
1535 goto out;
1538 if (jbd2_journal_has_csum_v2or3_feature(journal) &&
1539 jbd2_has_feature_checksum(journal)) {
1540 /* Can't have checksum v1 and v2 on at the same time! */
1541 printk(KERN_ERR "JBD2: Can't enable checksumming v1 and v2/3 "
1542 "at the same time!\n");
1543 goto out;
1546 if (!jbd2_verify_csum_type(journal, sb)) {
1547 printk(KERN_ERR "JBD2: Unknown checksum type\n");
1548 goto out;
1551 /* Load the checksum driver */
1552 if (jbd2_journal_has_csum_v2or3_feature(journal)) {
1553 journal->j_chksum_driver = crypto_alloc_shash("crc32c", 0, 0);
1554 if (IS_ERR(journal->j_chksum_driver)) {
1555 printk(KERN_ERR "JBD2: Cannot load crc32c driver.\n");
1556 err = PTR_ERR(journal->j_chksum_driver);
1557 journal->j_chksum_driver = NULL;
1558 goto out;
1562 /* Check superblock checksum */
1563 if (!jbd2_superblock_csum_verify(journal, sb)) {
1564 printk(KERN_ERR "JBD2: journal checksum error\n");
1565 err = -EFSBADCRC;
1566 goto out;
1569 /* Precompute checksum seed for all metadata */
1570 if (jbd2_journal_has_csum_v2or3(journal))
1571 journal->j_csum_seed = jbd2_chksum(journal, ~0, sb->s_uuid,
1572 sizeof(sb->s_uuid));
1574 set_buffer_verified(bh);
1576 return 0;
1578 out:
1579 journal_fail_superblock(journal);
1580 return err;
1584 * Load the on-disk journal superblock and read the key fields into the
1585 * journal_t.
1588 static int load_superblock(journal_t *journal)
1590 int err;
1591 journal_superblock_t *sb;
1593 err = journal_get_superblock(journal);
1594 if (err)
1595 return err;
1597 sb = journal->j_superblock;
1599 journal->j_tail_sequence = be32_to_cpu(sb->s_sequence);
1600 journal->j_tail = be32_to_cpu(sb->s_start);
1601 journal->j_first = be32_to_cpu(sb->s_first);
1602 journal->j_last = be32_to_cpu(sb->s_maxlen);
1603 journal->j_errno = be32_to_cpu(sb->s_errno);
1605 return 0;
1610 * int jbd2_journal_load() - Read journal from disk.
1611 * @journal: Journal to act on.
1613 * Given a journal_t structure which tells us which disk blocks contain
1614 * a journal, read the journal from disk to initialise the in-memory
1615 * structures.
1617 int jbd2_journal_load(journal_t *journal)
1619 int err;
1620 journal_superblock_t *sb;
1622 err = load_superblock(journal);
1623 if (err)
1624 return err;
1626 sb = journal->j_superblock;
1627 /* If this is a V2 superblock, then we have to check the
1628 * features flags on it. */
1630 if (journal->j_format_version >= 2) {
1631 if ((sb->s_feature_ro_compat &
1632 ~cpu_to_be32(JBD2_KNOWN_ROCOMPAT_FEATURES)) ||
1633 (sb->s_feature_incompat &
1634 ~cpu_to_be32(JBD2_KNOWN_INCOMPAT_FEATURES))) {
1635 printk(KERN_WARNING
1636 "JBD2: Unrecognised features on journal\n");
1637 return -EINVAL;
1642 * Create a slab for this blocksize
1644 err = jbd2_journal_create_slab(be32_to_cpu(sb->s_blocksize));
1645 if (err)
1646 return err;
1648 /* Let the recovery code check whether it needs to recover any
1649 * data from the journal. */
1650 if (jbd2_journal_recover(journal))
1651 goto recovery_error;
1653 if (journal->j_failed_commit) {
1654 printk(KERN_ERR "JBD2: journal transaction %u on %s "
1655 "is corrupt.\n", journal->j_failed_commit,
1656 journal->j_devname);
1657 return -EFSCORRUPTED;
1660 * clear JBD2_ABORT flag initialized in journal_init_common
1661 * here to update log tail information with the newest seq.
1663 journal->j_flags &= ~JBD2_ABORT;
1665 /* OK, we've finished with the dynamic journal bits:
1666 * reinitialise the dynamic contents of the superblock in memory
1667 * and reset them on disk. */
1668 if (journal_reset(journal))
1669 goto recovery_error;
1671 journal->j_flags |= JBD2_LOADED;
1672 return 0;
1674 recovery_error:
1675 printk(KERN_WARNING "JBD2: recovery failed\n");
1676 return -EIO;
1680 * void jbd2_journal_destroy() - Release a journal_t structure.
1681 * @journal: Journal to act on.
1683 * Release a journal_t structure once it is no longer in use by the
1684 * journaled object.
1685 * Return <0 if we couldn't clean up the journal.
1687 int jbd2_journal_destroy(journal_t *journal)
1689 int err = 0;
1691 /* Wait for the commit thread to wake up and die. */
1692 journal_kill_thread(journal);
1694 /* Force a final log commit */
1695 if (journal->j_running_transaction)
1696 jbd2_journal_commit_transaction(journal);
1698 /* Force any old transactions to disk */
1700 /* Totally anal locking here... */
1701 spin_lock(&journal->j_list_lock);
1702 while (journal->j_checkpoint_transactions != NULL) {
1703 spin_unlock(&journal->j_list_lock);
1704 mutex_lock(&journal->j_checkpoint_mutex);
1705 err = jbd2_log_do_checkpoint(journal);
1706 mutex_unlock(&journal->j_checkpoint_mutex);
1708 * If checkpointing failed, just free the buffers to avoid
1709 * looping forever
1711 if (err) {
1712 jbd2_journal_destroy_checkpoint(journal);
1713 spin_lock(&journal->j_list_lock);
1714 break;
1716 spin_lock(&journal->j_list_lock);
1719 J_ASSERT(journal->j_running_transaction == NULL);
1720 J_ASSERT(journal->j_committing_transaction == NULL);
1721 J_ASSERT(journal->j_checkpoint_transactions == NULL);
1722 spin_unlock(&journal->j_list_lock);
1724 if (journal->j_sb_buffer) {
1725 if (!is_journal_aborted(journal)) {
1726 mutex_lock(&journal->j_checkpoint_mutex);
1728 write_lock(&journal->j_state_lock);
1729 journal->j_tail_sequence =
1730 ++journal->j_transaction_sequence;
1731 write_unlock(&journal->j_state_lock);
1733 jbd2_mark_journal_empty(journal, WRITE_FLUSH_FUA);
1734 mutex_unlock(&journal->j_checkpoint_mutex);
1735 } else
1736 err = -EIO;
1737 brelse(journal->j_sb_buffer);
1740 if (journal->j_proc_entry)
1741 jbd2_stats_proc_exit(journal);
1742 iput(journal->j_inode);
1743 if (journal->j_revoke)
1744 jbd2_journal_destroy_revoke(journal);
1745 if (journal->j_chksum_driver)
1746 crypto_free_shash(journal->j_chksum_driver);
1747 kfree(journal->j_wbuf);
1748 kfree(journal);
1750 return err;
1755 *int jbd2_journal_check_used_features () - Check if features specified are used.
1756 * @journal: Journal to check.
1757 * @compat: bitmask of compatible features
1758 * @ro: bitmask of features that force read-only mount
1759 * @incompat: bitmask of incompatible features
1761 * Check whether the journal uses all of a given set of
1762 * features. Return true (non-zero) if it does.
1765 int jbd2_journal_check_used_features (journal_t *journal, unsigned long compat,
1766 unsigned long ro, unsigned long incompat)
1768 journal_superblock_t *sb;
1770 if (!compat && !ro && !incompat)
1771 return 1;
1772 /* Load journal superblock if it is not loaded yet. */
1773 if (journal->j_format_version == 0 &&
1774 journal_get_superblock(journal) != 0)
1775 return 0;
1776 if (journal->j_format_version == 1)
1777 return 0;
1779 sb = journal->j_superblock;
1781 if (((be32_to_cpu(sb->s_feature_compat) & compat) == compat) &&
1782 ((be32_to_cpu(sb->s_feature_ro_compat) & ro) == ro) &&
1783 ((be32_to_cpu(sb->s_feature_incompat) & incompat) == incompat))
1784 return 1;
1786 return 0;
1790 * int jbd2_journal_check_available_features() - Check feature set in journalling layer
1791 * @journal: Journal to check.
1792 * @compat: bitmask of compatible features
1793 * @ro: bitmask of features that force read-only mount
1794 * @incompat: bitmask of incompatible features
1796 * Check whether the journaling code supports the use of
1797 * all of a given set of features on this journal. Return true
1798 * (non-zero) if it can. */
1800 int jbd2_journal_check_available_features (journal_t *journal, unsigned long compat,
1801 unsigned long ro, unsigned long incompat)
1803 if (!compat && !ro && !incompat)
1804 return 1;
1806 /* We can support any known requested features iff the
1807 * superblock is in version 2. Otherwise we fail to support any
1808 * extended sb features. */
1810 if (journal->j_format_version != 2)
1811 return 0;
1813 if ((compat & JBD2_KNOWN_COMPAT_FEATURES) == compat &&
1814 (ro & JBD2_KNOWN_ROCOMPAT_FEATURES) == ro &&
1815 (incompat & JBD2_KNOWN_INCOMPAT_FEATURES) == incompat)
1816 return 1;
1818 return 0;
1822 * int jbd2_journal_set_features () - Mark a given journal feature in the superblock
1823 * @journal: Journal to act on.
1824 * @compat: bitmask of compatible features
1825 * @ro: bitmask of features that force read-only mount
1826 * @incompat: bitmask of incompatible features
1828 * Mark a given journal feature as present on the
1829 * superblock. Returns true if the requested features could be set.
1833 int jbd2_journal_set_features (journal_t *journal, unsigned long compat,
1834 unsigned long ro, unsigned long incompat)
1836 #define INCOMPAT_FEATURE_ON(f) \
1837 ((incompat & (f)) && !(sb->s_feature_incompat & cpu_to_be32(f)))
1838 #define COMPAT_FEATURE_ON(f) \
1839 ((compat & (f)) && !(sb->s_feature_compat & cpu_to_be32(f)))
1840 journal_superblock_t *sb;
1842 if (jbd2_journal_check_used_features(journal, compat, ro, incompat))
1843 return 1;
1845 if (!jbd2_journal_check_available_features(journal, compat, ro, incompat))
1846 return 0;
1848 /* If enabling v2 checksums, turn on v3 instead */
1849 if (incompat & JBD2_FEATURE_INCOMPAT_CSUM_V2) {
1850 incompat &= ~JBD2_FEATURE_INCOMPAT_CSUM_V2;
1851 incompat |= JBD2_FEATURE_INCOMPAT_CSUM_V3;
1854 /* Asking for checksumming v3 and v1? Only give them v3. */
1855 if (incompat & JBD2_FEATURE_INCOMPAT_CSUM_V3 &&
1856 compat & JBD2_FEATURE_COMPAT_CHECKSUM)
1857 compat &= ~JBD2_FEATURE_COMPAT_CHECKSUM;
1859 jbd_debug(1, "Setting new features 0x%lx/0x%lx/0x%lx\n",
1860 compat, ro, incompat);
1862 sb = journal->j_superblock;
1864 /* If enabling v3 checksums, update superblock */
1865 if (INCOMPAT_FEATURE_ON(JBD2_FEATURE_INCOMPAT_CSUM_V3)) {
1866 sb->s_checksum_type = JBD2_CRC32C_CHKSUM;
1867 sb->s_feature_compat &=
1868 ~cpu_to_be32(JBD2_FEATURE_COMPAT_CHECKSUM);
1870 /* Load the checksum driver */
1871 if (journal->j_chksum_driver == NULL) {
1872 journal->j_chksum_driver = crypto_alloc_shash("crc32c",
1873 0, 0);
1874 if (IS_ERR(journal->j_chksum_driver)) {
1875 printk(KERN_ERR "JBD2: Cannot load crc32c "
1876 "driver.\n");
1877 journal->j_chksum_driver = NULL;
1878 return 0;
1881 /* Precompute checksum seed for all metadata */
1882 journal->j_csum_seed = jbd2_chksum(journal, ~0,
1883 sb->s_uuid,
1884 sizeof(sb->s_uuid));
1888 /* If enabling v1 checksums, downgrade superblock */
1889 if (COMPAT_FEATURE_ON(JBD2_FEATURE_COMPAT_CHECKSUM))
1890 sb->s_feature_incompat &=
1891 ~cpu_to_be32(JBD2_FEATURE_INCOMPAT_CSUM_V2 |
1892 JBD2_FEATURE_INCOMPAT_CSUM_V3);
1894 sb->s_feature_compat |= cpu_to_be32(compat);
1895 sb->s_feature_ro_compat |= cpu_to_be32(ro);
1896 sb->s_feature_incompat |= cpu_to_be32(incompat);
1898 return 1;
1899 #undef COMPAT_FEATURE_ON
1900 #undef INCOMPAT_FEATURE_ON
1904 * jbd2_journal_clear_features () - Clear a given journal feature in the
1905 * superblock
1906 * @journal: Journal to act on.
1907 * @compat: bitmask of compatible features
1908 * @ro: bitmask of features that force read-only mount
1909 * @incompat: bitmask of incompatible features
1911 * Clear a given journal feature as present on the
1912 * superblock.
1914 void jbd2_journal_clear_features(journal_t *journal, unsigned long compat,
1915 unsigned long ro, unsigned long incompat)
1917 journal_superblock_t *sb;
1919 jbd_debug(1, "Clear features 0x%lx/0x%lx/0x%lx\n",
1920 compat, ro, incompat);
1922 sb = journal->j_superblock;
1924 sb->s_feature_compat &= ~cpu_to_be32(compat);
1925 sb->s_feature_ro_compat &= ~cpu_to_be32(ro);
1926 sb->s_feature_incompat &= ~cpu_to_be32(incompat);
1928 EXPORT_SYMBOL(jbd2_journal_clear_features);
1931 * int jbd2_journal_flush () - Flush journal
1932 * @journal: Journal to act on.
1934 * Flush all data for a given journal to disk and empty the journal.
1935 * Filesystems can use this when remounting readonly to ensure that
1936 * recovery does not need to happen on remount.
1939 int jbd2_journal_flush(journal_t *journal)
1941 int err = 0;
1942 transaction_t *transaction = NULL;
1944 write_lock(&journal->j_state_lock);
1946 /* Force everything buffered to the log... */
1947 if (journal->j_running_transaction) {
1948 transaction = journal->j_running_transaction;
1949 __jbd2_log_start_commit(journal, transaction->t_tid);
1950 } else if (journal->j_committing_transaction)
1951 transaction = journal->j_committing_transaction;
1953 /* Wait for the log commit to complete... */
1954 if (transaction) {
1955 tid_t tid = transaction->t_tid;
1957 write_unlock(&journal->j_state_lock);
1958 jbd2_log_wait_commit(journal, tid);
1959 } else {
1960 write_unlock(&journal->j_state_lock);
1963 /* ...and flush everything in the log out to disk. */
1964 spin_lock(&journal->j_list_lock);
1965 while (!err && journal->j_checkpoint_transactions != NULL) {
1966 spin_unlock(&journal->j_list_lock);
1967 mutex_lock(&journal->j_checkpoint_mutex);
1968 err = jbd2_log_do_checkpoint(journal);
1969 mutex_unlock(&journal->j_checkpoint_mutex);
1970 spin_lock(&journal->j_list_lock);
1972 spin_unlock(&journal->j_list_lock);
1974 if (is_journal_aborted(journal))
1975 return -EIO;
1977 mutex_lock(&journal->j_checkpoint_mutex);
1978 if (!err) {
1979 err = jbd2_cleanup_journal_tail(journal);
1980 if (err < 0) {
1981 mutex_unlock(&journal->j_checkpoint_mutex);
1982 goto out;
1984 err = 0;
1987 /* Finally, mark the journal as really needing no recovery.
1988 * This sets s_start==0 in the underlying superblock, which is
1989 * the magic code for a fully-recovered superblock. Any future
1990 * commits of data to the journal will restore the current
1991 * s_start value. */
1992 jbd2_mark_journal_empty(journal, WRITE_FUA);
1993 mutex_unlock(&journal->j_checkpoint_mutex);
1994 write_lock(&journal->j_state_lock);
1995 J_ASSERT(!journal->j_running_transaction);
1996 J_ASSERT(!journal->j_committing_transaction);
1997 J_ASSERT(!journal->j_checkpoint_transactions);
1998 J_ASSERT(journal->j_head == journal->j_tail);
1999 J_ASSERT(journal->j_tail_sequence == journal->j_transaction_sequence);
2000 write_unlock(&journal->j_state_lock);
2001 out:
2002 return err;
2006 * int jbd2_journal_wipe() - Wipe journal contents
2007 * @journal: Journal to act on.
2008 * @write: flag (see below)
2010 * Wipe out all of the contents of a journal, safely. This will produce
2011 * a warning if the journal contains any valid recovery information.
2012 * Must be called between journal_init_*() and jbd2_journal_load().
2014 * If 'write' is non-zero, then we wipe out the journal on disk; otherwise
2015 * we merely suppress recovery.
2018 int jbd2_journal_wipe(journal_t *journal, int write)
2020 int err = 0;
2022 J_ASSERT (!(journal->j_flags & JBD2_LOADED));
2024 err = load_superblock(journal);
2025 if (err)
2026 return err;
2028 if (!journal->j_tail)
2029 goto no_recovery;
2031 printk(KERN_WARNING "JBD2: %s recovery information on journal\n",
2032 write ? "Clearing" : "Ignoring");
2034 err = jbd2_journal_skip_recovery(journal);
2035 if (write) {
2036 /* Lock to make assertions happy... */
2037 mutex_lock(&journal->j_checkpoint_mutex);
2038 jbd2_mark_journal_empty(journal, WRITE_FUA);
2039 mutex_unlock(&journal->j_checkpoint_mutex);
2042 no_recovery:
2043 return err;
2047 * Journal abort has very specific semantics, which we describe
2048 * for journal abort.
2050 * Two internal functions, which provide abort to the jbd layer
2051 * itself are here.
2055 * Quick version for internal journal use (doesn't lock the journal).
2056 * Aborts hard --- we mark the abort as occurred, but do _nothing_ else,
2057 * and don't attempt to make any other journal updates.
2059 void __jbd2_journal_abort_hard(journal_t *journal)
2061 transaction_t *transaction;
2063 if (journal->j_flags & JBD2_ABORT)
2064 return;
2066 printk(KERN_ERR "Aborting journal on device %s.\n",
2067 journal->j_devname);
2069 write_lock(&journal->j_state_lock);
2070 journal->j_flags |= JBD2_ABORT;
2071 transaction = journal->j_running_transaction;
2072 if (transaction)
2073 __jbd2_log_start_commit(journal, transaction->t_tid);
2074 write_unlock(&journal->j_state_lock);
2077 /* Soft abort: record the abort error status in the journal superblock,
2078 * but don't do any other IO. */
2079 static void __journal_abort_soft (journal_t *journal, int errno)
2081 if (journal->j_flags & JBD2_ABORT)
2082 return;
2084 if (!journal->j_errno)
2085 journal->j_errno = errno;
2087 __jbd2_journal_abort_hard(journal);
2089 jbd2_journal_update_sb_errno(journal);
2090 write_lock(&journal->j_state_lock);
2091 journal->j_flags |= JBD2_REC_ERR;
2092 write_unlock(&journal->j_state_lock);
2096 * void jbd2_journal_abort () - Shutdown the journal immediately.
2097 * @journal: the journal to shutdown.
2098 * @errno: an error number to record in the journal indicating
2099 * the reason for the shutdown.
2101 * Perform a complete, immediate shutdown of the ENTIRE
2102 * journal (not of a single transaction). This operation cannot be
2103 * undone without closing and reopening the journal.
2105 * The jbd2_journal_abort function is intended to support higher level error
2106 * recovery mechanisms such as the ext2/ext3 remount-readonly error
2107 * mode.
2109 * Journal abort has very specific semantics. Any existing dirty,
2110 * unjournaled buffers in the main filesystem will still be written to
2111 * disk by bdflush, but the journaling mechanism will be suspended
2112 * immediately and no further transaction commits will be honoured.
2114 * Any dirty, journaled buffers will be written back to disk without
2115 * hitting the journal. Atomicity cannot be guaranteed on an aborted
2116 * filesystem, but we _do_ attempt to leave as much data as possible
2117 * behind for fsck to use for cleanup.
2119 * Any attempt to get a new transaction handle on a journal which is in
2120 * ABORT state will just result in an -EROFS error return. A
2121 * jbd2_journal_stop on an existing handle will return -EIO if we have
2122 * entered abort state during the update.
2124 * Recursive transactions are not disturbed by journal abort until the
2125 * final jbd2_journal_stop, which will receive the -EIO error.
2127 * Finally, the jbd2_journal_abort call allows the caller to supply an errno
2128 * which will be recorded (if possible) in the journal superblock. This
2129 * allows a client to record failure conditions in the middle of a
2130 * transaction without having to complete the transaction to record the
2131 * failure to disk. ext3_error, for example, now uses this
2132 * functionality.
2136 void jbd2_journal_abort(journal_t *journal, int errno)
2138 __journal_abort_soft(journal, errno);
2142 * int jbd2_journal_errno () - returns the journal's error state.
2143 * @journal: journal to examine.
2145 * This is the errno number set with jbd2_journal_abort(), the last
2146 * time the journal was mounted - if the journal was stopped
2147 * without calling abort this will be 0.
2149 * If the journal has been aborted on this mount time -EROFS will
2150 * be returned.
2152 int jbd2_journal_errno(journal_t *journal)
2154 int err;
2156 read_lock(&journal->j_state_lock);
2157 if (journal->j_flags & JBD2_ABORT)
2158 err = -EROFS;
2159 else
2160 err = journal->j_errno;
2161 read_unlock(&journal->j_state_lock);
2162 return err;
2166 * int jbd2_journal_clear_err () - clears the journal's error state
2167 * @journal: journal to act on.
2169 * An error must be cleared or acked to take a FS out of readonly
2170 * mode.
2172 int jbd2_journal_clear_err(journal_t *journal)
2174 int err = 0;
2176 write_lock(&journal->j_state_lock);
2177 if (journal->j_flags & JBD2_ABORT)
2178 err = -EROFS;
2179 else
2180 journal->j_errno = 0;
2181 write_unlock(&journal->j_state_lock);
2182 return err;
2186 * void jbd2_journal_ack_err() - Ack journal err.
2187 * @journal: journal to act on.
2189 * An error must be cleared or acked to take a FS out of readonly
2190 * mode.
2192 void jbd2_journal_ack_err(journal_t *journal)
2194 write_lock(&journal->j_state_lock);
2195 if (journal->j_errno)
2196 journal->j_flags |= JBD2_ACK_ERR;
2197 write_unlock(&journal->j_state_lock);
2200 int jbd2_journal_blocks_per_page(struct inode *inode)
2202 return 1 << (PAGE_CACHE_SHIFT - inode->i_sb->s_blocksize_bits);
2206 * helper functions to deal with 32 or 64bit block numbers.
2208 size_t journal_tag_bytes(journal_t *journal)
2210 size_t sz;
2212 if (jbd2_has_feature_csum3(journal))
2213 return sizeof(journal_block_tag3_t);
2215 sz = sizeof(journal_block_tag_t);
2217 if (jbd2_has_feature_csum2(journal))
2218 sz += sizeof(__u16);
2220 if (jbd2_has_feature_64bit(journal))
2221 return sz;
2222 else
2223 return sz - sizeof(__u32);
2227 * JBD memory management
2229 * These functions are used to allocate block-sized chunks of memory
2230 * used for making copies of buffer_head data. Very often it will be
2231 * page-sized chunks of data, but sometimes it will be in
2232 * sub-page-size chunks. (For example, 16k pages on Power systems
2233 * with a 4k block file system.) For blocks smaller than a page, we
2234 * use a SLAB allocator. There are slab caches for each block size,
2235 * which are allocated at mount time, if necessary, and we only free
2236 * (all of) the slab caches when/if the jbd2 module is unloaded. For
2237 * this reason we don't need to a mutex to protect access to
2238 * jbd2_slab[] allocating or releasing memory; only in
2239 * jbd2_journal_create_slab().
2241 #define JBD2_MAX_SLABS 8
2242 static struct kmem_cache *jbd2_slab[JBD2_MAX_SLABS];
2244 static const char *jbd2_slab_names[JBD2_MAX_SLABS] = {
2245 "jbd2_1k", "jbd2_2k", "jbd2_4k", "jbd2_8k",
2246 "jbd2_16k", "jbd2_32k", "jbd2_64k", "jbd2_128k"
2250 static void jbd2_journal_destroy_slabs(void)
2252 int i;
2254 for (i = 0; i < JBD2_MAX_SLABS; i++) {
2255 if (jbd2_slab[i])
2256 kmem_cache_destroy(jbd2_slab[i]);
2257 jbd2_slab[i] = NULL;
2261 static int jbd2_journal_create_slab(size_t size)
2263 static DEFINE_MUTEX(jbd2_slab_create_mutex);
2264 int i = order_base_2(size) - 10;
2265 size_t slab_size;
2267 if (size == PAGE_SIZE)
2268 return 0;
2270 if (i >= JBD2_MAX_SLABS)
2271 return -EINVAL;
2273 if (unlikely(i < 0))
2274 i = 0;
2275 mutex_lock(&jbd2_slab_create_mutex);
2276 if (jbd2_slab[i]) {
2277 mutex_unlock(&jbd2_slab_create_mutex);
2278 return 0; /* Already created */
2281 slab_size = 1 << (i+10);
2282 jbd2_slab[i] = kmem_cache_create(jbd2_slab_names[i], slab_size,
2283 slab_size, 0, NULL);
2284 mutex_unlock(&jbd2_slab_create_mutex);
2285 if (!jbd2_slab[i]) {
2286 printk(KERN_EMERG "JBD2: no memory for jbd2_slab cache\n");
2287 return -ENOMEM;
2289 return 0;
2292 static struct kmem_cache *get_slab(size_t size)
2294 int i = order_base_2(size) - 10;
2296 BUG_ON(i >= JBD2_MAX_SLABS);
2297 if (unlikely(i < 0))
2298 i = 0;
2299 BUG_ON(jbd2_slab[i] == NULL);
2300 return jbd2_slab[i];
2303 void *jbd2_alloc(size_t size, gfp_t flags)
2305 void *ptr;
2307 BUG_ON(size & (size-1)); /* Must be a power of 2 */
2309 flags |= __GFP_REPEAT;
2310 if (size == PAGE_SIZE)
2311 ptr = (void *)__get_free_pages(flags, 0);
2312 else if (size > PAGE_SIZE) {
2313 int order = get_order(size);
2315 if (order < 3)
2316 ptr = (void *)__get_free_pages(flags, order);
2317 else
2318 ptr = vmalloc(size);
2319 } else
2320 ptr = kmem_cache_alloc(get_slab(size), flags);
2322 /* Check alignment; SLUB has gotten this wrong in the past,
2323 * and this can lead to user data corruption! */
2324 BUG_ON(((unsigned long) ptr) & (size-1));
2326 return ptr;
2329 void jbd2_free(void *ptr, size_t size)
2331 if (size == PAGE_SIZE) {
2332 free_pages((unsigned long)ptr, 0);
2333 return;
2335 if (size > PAGE_SIZE) {
2336 int order = get_order(size);
2338 if (order < 3)
2339 free_pages((unsigned long)ptr, order);
2340 else
2341 vfree(ptr);
2342 return;
2344 kmem_cache_free(get_slab(size), ptr);
2348 * Journal_head storage management
2350 static struct kmem_cache *jbd2_journal_head_cache;
2351 #ifdef CONFIG_JBD2_DEBUG
2352 static atomic_t nr_journal_heads = ATOMIC_INIT(0);
2353 #endif
2355 static int jbd2_journal_init_journal_head_cache(void)
2357 int retval;
2359 J_ASSERT(jbd2_journal_head_cache == NULL);
2360 jbd2_journal_head_cache = kmem_cache_create("jbd2_journal_head",
2361 sizeof(struct journal_head),
2362 0, /* offset */
2363 SLAB_TEMPORARY | SLAB_DESTROY_BY_RCU,
2364 NULL); /* ctor */
2365 retval = 0;
2366 if (!jbd2_journal_head_cache) {
2367 retval = -ENOMEM;
2368 printk(KERN_EMERG "JBD2: no memory for journal_head cache\n");
2370 return retval;
2373 static void jbd2_journal_destroy_journal_head_cache(void)
2375 if (jbd2_journal_head_cache) {
2376 kmem_cache_destroy(jbd2_journal_head_cache);
2377 jbd2_journal_head_cache = NULL;
2382 * journal_head splicing and dicing
2384 static struct journal_head *journal_alloc_journal_head(void)
2386 struct journal_head *ret;
2388 #ifdef CONFIG_JBD2_DEBUG
2389 atomic_inc(&nr_journal_heads);
2390 #endif
2391 ret = kmem_cache_zalloc(jbd2_journal_head_cache, GFP_NOFS);
2392 if (!ret) {
2393 jbd_debug(1, "out of memory for journal_head\n");
2394 pr_notice_ratelimited("ENOMEM in %s, retrying.\n", __func__);
2395 ret = kmem_cache_zalloc(jbd2_journal_head_cache,
2396 GFP_NOFS | __GFP_NOFAIL);
2398 return ret;
2401 static void journal_free_journal_head(struct journal_head *jh)
2403 #ifdef CONFIG_JBD2_DEBUG
2404 atomic_dec(&nr_journal_heads);
2405 memset(jh, JBD2_POISON_FREE, sizeof(*jh));
2406 #endif
2407 kmem_cache_free(jbd2_journal_head_cache, jh);
2411 * A journal_head is attached to a buffer_head whenever JBD has an
2412 * interest in the buffer.
2414 * Whenever a buffer has an attached journal_head, its ->b_state:BH_JBD bit
2415 * is set. This bit is tested in core kernel code where we need to take
2416 * JBD-specific actions. Testing the zeroness of ->b_private is not reliable
2417 * there.
2419 * When a buffer has its BH_JBD bit set, its ->b_count is elevated by one.
2421 * When a buffer has its BH_JBD bit set it is immune from being released by
2422 * core kernel code, mainly via ->b_count.
2424 * A journal_head is detached from its buffer_head when the journal_head's
2425 * b_jcount reaches zero. Running transaction (b_transaction) and checkpoint
2426 * transaction (b_cp_transaction) hold their references to b_jcount.
2428 * Various places in the kernel want to attach a journal_head to a buffer_head
2429 * _before_ attaching the journal_head to a transaction. To protect the
2430 * journal_head in this situation, jbd2_journal_add_journal_head elevates the
2431 * journal_head's b_jcount refcount by one. The caller must call
2432 * jbd2_journal_put_journal_head() to undo this.
2434 * So the typical usage would be:
2436 * (Attach a journal_head if needed. Increments b_jcount)
2437 * struct journal_head *jh = jbd2_journal_add_journal_head(bh);
2438 * ...
2439 * (Get another reference for transaction)
2440 * jbd2_journal_grab_journal_head(bh);
2441 * jh->b_transaction = xxx;
2442 * (Put original reference)
2443 * jbd2_journal_put_journal_head(jh);
2447 * Give a buffer_head a journal_head.
2449 * May sleep.
2451 struct journal_head *jbd2_journal_add_journal_head(struct buffer_head *bh)
2453 struct journal_head *jh;
2454 struct journal_head *new_jh = NULL;
2456 repeat:
2457 if (!buffer_jbd(bh))
2458 new_jh = journal_alloc_journal_head();
2460 jbd_lock_bh_journal_head(bh);
2461 if (buffer_jbd(bh)) {
2462 jh = bh2jh(bh);
2463 } else {
2464 J_ASSERT_BH(bh,
2465 (atomic_read(&bh->b_count) > 0) ||
2466 (bh->b_page && bh->b_page->mapping));
2468 if (!new_jh) {
2469 jbd_unlock_bh_journal_head(bh);
2470 goto repeat;
2473 jh = new_jh;
2474 new_jh = NULL; /* We consumed it */
2475 set_buffer_jbd(bh);
2476 bh->b_private = jh;
2477 jh->b_bh = bh;
2478 get_bh(bh);
2479 BUFFER_TRACE(bh, "added journal_head");
2481 jh->b_jcount++;
2482 jbd_unlock_bh_journal_head(bh);
2483 if (new_jh)
2484 journal_free_journal_head(new_jh);
2485 return bh->b_private;
2489 * Grab a ref against this buffer_head's journal_head. If it ended up not
2490 * having a journal_head, return NULL
2492 struct journal_head *jbd2_journal_grab_journal_head(struct buffer_head *bh)
2494 struct journal_head *jh = NULL;
2496 jbd_lock_bh_journal_head(bh);
2497 if (buffer_jbd(bh)) {
2498 jh = bh2jh(bh);
2499 jh->b_jcount++;
2501 jbd_unlock_bh_journal_head(bh);
2502 return jh;
2505 static void __journal_remove_journal_head(struct buffer_head *bh)
2507 struct journal_head *jh = bh2jh(bh);
2509 J_ASSERT_JH(jh, jh->b_jcount >= 0);
2510 J_ASSERT_JH(jh, jh->b_transaction == NULL);
2511 J_ASSERT_JH(jh, jh->b_next_transaction == NULL);
2512 J_ASSERT_JH(jh, jh->b_cp_transaction == NULL);
2513 J_ASSERT_JH(jh, jh->b_jlist == BJ_None);
2514 J_ASSERT_BH(bh, buffer_jbd(bh));
2515 J_ASSERT_BH(bh, jh2bh(jh) == bh);
2516 BUFFER_TRACE(bh, "remove journal_head");
2517 if (jh->b_frozen_data) {
2518 printk(KERN_WARNING "%s: freeing b_frozen_data\n", __func__);
2519 jbd2_free(jh->b_frozen_data, bh->b_size);
2521 if (jh->b_committed_data) {
2522 printk(KERN_WARNING "%s: freeing b_committed_data\n", __func__);
2523 jbd2_free(jh->b_committed_data, bh->b_size);
2525 bh->b_private = NULL;
2526 jh->b_bh = NULL; /* debug, really */
2527 clear_buffer_jbd(bh);
2528 journal_free_journal_head(jh);
2532 * Drop a reference on the passed journal_head. If it fell to zero then
2533 * release the journal_head from the buffer_head.
2535 void jbd2_journal_put_journal_head(struct journal_head *jh)
2537 struct buffer_head *bh = jh2bh(jh);
2539 jbd_lock_bh_journal_head(bh);
2540 J_ASSERT_JH(jh, jh->b_jcount > 0);
2541 --jh->b_jcount;
2542 if (!jh->b_jcount) {
2543 __journal_remove_journal_head(bh);
2544 jbd_unlock_bh_journal_head(bh);
2545 __brelse(bh);
2546 } else
2547 jbd_unlock_bh_journal_head(bh);
2551 * Initialize jbd inode head
2553 void jbd2_journal_init_jbd_inode(struct jbd2_inode *jinode, struct inode *inode)
2555 jinode->i_transaction = NULL;
2556 jinode->i_next_transaction = NULL;
2557 jinode->i_vfs_inode = inode;
2558 jinode->i_flags = 0;
2559 INIT_LIST_HEAD(&jinode->i_list);
2563 * Function to be called before we start removing inode from memory (i.e.,
2564 * clear_inode() is a fine place to be called from). It removes inode from
2565 * transaction's lists.
2567 void jbd2_journal_release_jbd_inode(journal_t *journal,
2568 struct jbd2_inode *jinode)
2570 if (!journal)
2571 return;
2572 restart:
2573 spin_lock(&journal->j_list_lock);
2574 /* Is commit writing out inode - we have to wait */
2575 if (test_bit(__JI_COMMIT_RUNNING, &jinode->i_flags)) {
2576 wait_queue_head_t *wq;
2577 DEFINE_WAIT_BIT(wait, &jinode->i_flags, __JI_COMMIT_RUNNING);
2578 wq = bit_waitqueue(&jinode->i_flags, __JI_COMMIT_RUNNING);
2579 prepare_to_wait(wq, &wait.wait, TASK_UNINTERRUPTIBLE);
2580 spin_unlock(&journal->j_list_lock);
2581 schedule();
2582 finish_wait(wq, &wait.wait);
2583 goto restart;
2586 if (jinode->i_transaction) {
2587 list_del(&jinode->i_list);
2588 jinode->i_transaction = NULL;
2590 spin_unlock(&journal->j_list_lock);
2594 #ifdef CONFIG_PROC_FS
2596 #define JBD2_STATS_PROC_NAME "fs/jbd2"
2598 static void __init jbd2_create_jbd_stats_proc_entry(void)
2600 proc_jbd2_stats = proc_mkdir(JBD2_STATS_PROC_NAME, NULL);
2603 static void __exit jbd2_remove_jbd_stats_proc_entry(void)
2605 if (proc_jbd2_stats)
2606 remove_proc_entry(JBD2_STATS_PROC_NAME, NULL);
2609 #else
2611 #define jbd2_create_jbd_stats_proc_entry() do {} while (0)
2612 #define jbd2_remove_jbd_stats_proc_entry() do {} while (0)
2614 #endif
2616 struct kmem_cache *jbd2_handle_cache, *jbd2_inode_cache;
2618 static int __init jbd2_journal_init_handle_cache(void)
2620 jbd2_handle_cache = KMEM_CACHE(jbd2_journal_handle, SLAB_TEMPORARY);
2621 if (jbd2_handle_cache == NULL) {
2622 printk(KERN_EMERG "JBD2: failed to create handle cache\n");
2623 return -ENOMEM;
2625 jbd2_inode_cache = KMEM_CACHE(jbd2_inode, 0);
2626 if (jbd2_inode_cache == NULL) {
2627 printk(KERN_EMERG "JBD2: failed to create inode cache\n");
2628 kmem_cache_destroy(jbd2_handle_cache);
2629 return -ENOMEM;
2631 return 0;
2634 static void jbd2_journal_destroy_handle_cache(void)
2636 if (jbd2_handle_cache)
2637 kmem_cache_destroy(jbd2_handle_cache);
2638 if (jbd2_inode_cache)
2639 kmem_cache_destroy(jbd2_inode_cache);
2644 * Module startup and shutdown
2647 static int __init journal_init_caches(void)
2649 int ret;
2651 ret = jbd2_journal_init_revoke_caches();
2652 if (ret == 0)
2653 ret = jbd2_journal_init_journal_head_cache();
2654 if (ret == 0)
2655 ret = jbd2_journal_init_handle_cache();
2656 if (ret == 0)
2657 ret = jbd2_journal_init_transaction_cache();
2658 return ret;
2661 static void jbd2_journal_destroy_caches(void)
2663 jbd2_journal_destroy_revoke_caches();
2664 jbd2_journal_destroy_journal_head_cache();
2665 jbd2_journal_destroy_handle_cache();
2666 jbd2_journal_destroy_transaction_cache();
2667 jbd2_journal_destroy_slabs();
2670 static int __init journal_init(void)
2672 int ret;
2674 BUILD_BUG_ON(sizeof(struct journal_superblock_s) != 1024);
2676 ret = journal_init_caches();
2677 if (ret == 0) {
2678 jbd2_create_jbd_stats_proc_entry();
2679 } else {
2680 jbd2_journal_destroy_caches();
2682 return ret;
2685 static void __exit journal_exit(void)
2687 #ifdef CONFIG_JBD2_DEBUG
2688 int n = atomic_read(&nr_journal_heads);
2689 if (n)
2690 printk(KERN_ERR "JBD2: leaked %d journal_heads!\n", n);
2691 #endif
2692 jbd2_remove_jbd_stats_proc_entry();
2693 jbd2_journal_destroy_caches();
2696 MODULE_LICENSE("GPL");
2697 module_init(journal_init);
2698 module_exit(journal_exit);