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
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>
28 #include <linux/jbd2.h>
29 #include <linux/errno.h>
30 #include <linux/slab.h>
31 #include <linux/init.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/debugfs.h>
39 #include <linux/seq_file.h>
41 #include <asm/uaccess.h>
44 EXPORT_SYMBOL(jbd2_journal_start);
45 EXPORT_SYMBOL(jbd2_journal_restart);
46 EXPORT_SYMBOL(jbd2_journal_extend);
47 EXPORT_SYMBOL(jbd2_journal_stop);
48 EXPORT_SYMBOL(jbd2_journal_lock_updates);
49 EXPORT_SYMBOL(jbd2_journal_unlock_updates);
50 EXPORT_SYMBOL(jbd2_journal_get_write_access);
51 EXPORT_SYMBOL(jbd2_journal_get_create_access);
52 EXPORT_SYMBOL(jbd2_journal_get_undo_access);
53 EXPORT_SYMBOL(jbd2_journal_dirty_data);
54 EXPORT_SYMBOL(jbd2_journal_dirty_metadata);
55 EXPORT_SYMBOL(jbd2_journal_release_buffer);
56 EXPORT_SYMBOL(jbd2_journal_forget);
58 EXPORT_SYMBOL(journal_sync_buffer);
60 EXPORT_SYMBOL(jbd2_journal_flush);
61 EXPORT_SYMBOL(jbd2_journal_revoke);
63 EXPORT_SYMBOL(jbd2_journal_init_dev);
64 EXPORT_SYMBOL(jbd2_journal_init_inode);
65 EXPORT_SYMBOL(jbd2_journal_update_format);
66 EXPORT_SYMBOL(jbd2_journal_check_used_features);
67 EXPORT_SYMBOL(jbd2_journal_check_available_features);
68 EXPORT_SYMBOL(jbd2_journal_set_features);
69 EXPORT_SYMBOL(jbd2_journal_create);
70 EXPORT_SYMBOL(jbd2_journal_load);
71 EXPORT_SYMBOL(jbd2_journal_destroy);
72 EXPORT_SYMBOL(jbd2_journal_update_superblock);
73 EXPORT_SYMBOL(jbd2_journal_abort);
74 EXPORT_SYMBOL(jbd2_journal_errno);
75 EXPORT_SYMBOL(jbd2_journal_ack_err);
76 EXPORT_SYMBOL(jbd2_journal_clear_err);
77 EXPORT_SYMBOL(jbd2_log_wait_commit);
78 EXPORT_SYMBOL(jbd2_journal_start_commit);
79 EXPORT_SYMBOL(jbd2_journal_force_commit_nested);
80 EXPORT_SYMBOL(jbd2_journal_wipe);
81 EXPORT_SYMBOL(jbd2_journal_blocks_per_page);
82 EXPORT_SYMBOL(jbd2_journal_invalidatepage);
83 EXPORT_SYMBOL(jbd2_journal_try_to_free_buffers);
84 EXPORT_SYMBOL(jbd2_journal_force_commit);
85 EXPORT_SYMBOL(jbd2_journal_file_inode);
86 EXPORT_SYMBOL(jbd2_journal_init_jbd_inode);
87 EXPORT_SYMBOL(jbd2_journal_release_jbd_inode);
88 EXPORT_SYMBOL(jbd2_journal_begin_ordered_truncate);
90 static int journal_convert_superblock_v1(journal_t *, journal_superblock_t *);
91 static void __journal_abort_soft (journal_t *journal, int errno);
94 * Helper function used to manage commit timeouts
97 static void commit_timeout(unsigned long __data)
99 struct task_struct * p = (struct task_struct *) __data;
105 * kjournald2: The main thread function used to manage a logging device
108 * This kernel thread is responsible for two things:
110 * 1) COMMIT: Every so often we need to commit the current state of the
111 * filesystem to disk. The journal thread is responsible for writing
112 * all of the metadata buffers to disk.
114 * 2) CHECKPOINT: We cannot reuse a used section of the log file until all
115 * of the data in that part of the log has been rewritten elsewhere on
116 * the disk. Flushing these old buffers to reclaim space in the log is
117 * known as checkpointing, and this thread is responsible for that job.
120 static int kjournald2(void *arg)
122 journal_t *journal = arg;
123 transaction_t *transaction;
126 * Set up an interval timer which can be used to trigger a commit wakeup
127 * after the commit interval expires
129 setup_timer(&journal->j_commit_timer, commit_timeout,
130 (unsigned long)current);
132 /* Record that the journal thread is running */
133 journal->j_task = current;
134 wake_up(&journal->j_wait_done_commit);
136 printk(KERN_INFO "kjournald2 starting. Commit interval %ld seconds\n",
137 journal->j_commit_interval / HZ);
140 * And now, wait forever for commit wakeup events.
142 spin_lock(&journal->j_state_lock);
145 if (journal->j_flags & JBD2_UNMOUNT)
148 jbd_debug(1, "commit_sequence=%d, commit_request=%d\n",
149 journal->j_commit_sequence, journal->j_commit_request);
151 if (journal->j_commit_sequence != journal->j_commit_request) {
152 jbd_debug(1, "OK, requests differ\n");
153 spin_unlock(&journal->j_state_lock);
154 del_timer_sync(&journal->j_commit_timer);
155 jbd2_journal_commit_transaction(journal);
156 spin_lock(&journal->j_state_lock);
160 wake_up(&journal->j_wait_done_commit);
161 if (freezing(current)) {
163 * The simpler the better. Flushing journal isn't a
164 * good idea, because that depends on threads that may
165 * be already stopped.
167 jbd_debug(1, "Now suspending kjournald2\n");
168 spin_unlock(&journal->j_state_lock);
170 spin_lock(&journal->j_state_lock);
173 * We assume on resume that commits are already there,
177 int should_sleep = 1;
179 prepare_to_wait(&journal->j_wait_commit, &wait,
181 if (journal->j_commit_sequence != journal->j_commit_request)
183 transaction = journal->j_running_transaction;
184 if (transaction && time_after_eq(jiffies,
185 transaction->t_expires))
187 if (journal->j_flags & JBD2_UNMOUNT)
190 spin_unlock(&journal->j_state_lock);
192 spin_lock(&journal->j_state_lock);
194 finish_wait(&journal->j_wait_commit, &wait);
197 jbd_debug(1, "kjournald2 wakes\n");
200 * Were we woken up by a commit wakeup event?
202 transaction = journal->j_running_transaction;
203 if (transaction && time_after_eq(jiffies, transaction->t_expires)) {
204 journal->j_commit_request = transaction->t_tid;
205 jbd_debug(1, "woke because of timeout\n");
210 spin_unlock(&journal->j_state_lock);
211 del_timer_sync(&journal->j_commit_timer);
212 journal->j_task = NULL;
213 wake_up(&journal->j_wait_done_commit);
214 jbd_debug(1, "Journal thread exiting.\n");
218 static int jbd2_journal_start_thread(journal_t *journal)
220 struct task_struct *t;
222 t = kthread_run(kjournald2, journal, "kjournald2");
226 wait_event(journal->j_wait_done_commit, journal->j_task != NULL);
230 static void journal_kill_thread(journal_t *journal)
232 spin_lock(&journal->j_state_lock);
233 journal->j_flags |= JBD2_UNMOUNT;
235 while (journal->j_task) {
236 wake_up(&journal->j_wait_commit);
237 spin_unlock(&journal->j_state_lock);
238 wait_event(journal->j_wait_done_commit, journal->j_task == NULL);
239 spin_lock(&journal->j_state_lock);
241 spin_unlock(&journal->j_state_lock);
245 * jbd2_journal_write_metadata_buffer: write a metadata buffer to the journal.
247 * Writes a metadata buffer to a given disk block. The actual IO is not
248 * performed but a new buffer_head is constructed which labels the data
249 * to be written with the correct destination disk block.
251 * Any magic-number escaping which needs to be done will cause a
252 * copy-out here. If the buffer happens to start with the
253 * JBD2_MAGIC_NUMBER, then we can't write it to the log directly: the
254 * magic number is only written to the log for descripter blocks. In
255 * this case, we copy the data and replace the first word with 0, and we
256 * return a result code which indicates that this buffer needs to be
257 * marked as an escaped buffer in the corresponding log descriptor
258 * block. The missing word can then be restored when the block is read
261 * If the source buffer has already been modified by a new transaction
262 * since we took the last commit snapshot, we use the frozen copy of
263 * that data for IO. If we end up using the existing buffer_head's data
264 * for the write, then we *have* to lock the buffer to prevent anyone
265 * else from using and possibly modifying it while the IO is in
268 * The function returns a pointer to the buffer_heads to be used for IO.
270 * We assume that the journal has already been locked in this function.
277 * Bit 0 set == escape performed on the data
278 * Bit 1 set == buffer copy-out performed (kfree the data after IO)
281 int jbd2_journal_write_metadata_buffer(transaction_t *transaction,
282 struct journal_head *jh_in,
283 struct journal_head **jh_out,
284 unsigned long long blocknr)
286 int need_copy_out = 0;
287 int done_copy_out = 0;
290 struct buffer_head *new_bh;
291 struct journal_head *new_jh;
292 struct page *new_page;
293 unsigned int new_offset;
294 struct buffer_head *bh_in = jh2bh(jh_in);
297 * The buffer really shouldn't be locked: only the current committing
298 * transaction is allowed to write it, so nobody else is allowed
301 * akpm: except if we're journalling data, and write() output is
302 * also part of a shared mapping, and another thread has
303 * decided to launch a writepage() against this buffer.
305 J_ASSERT_BH(bh_in, buffer_jbddirty(bh_in));
307 new_bh = alloc_buffer_head(GFP_NOFS|__GFP_NOFAIL);
310 * If a new transaction has already done a buffer copy-out, then
311 * we use that version of the data for the commit.
313 jbd_lock_bh_state(bh_in);
315 if (jh_in->b_frozen_data) {
317 new_page = virt_to_page(jh_in->b_frozen_data);
318 new_offset = offset_in_page(jh_in->b_frozen_data);
320 new_page = jh2bh(jh_in)->b_page;
321 new_offset = offset_in_page(jh2bh(jh_in)->b_data);
324 mapped_data = kmap_atomic(new_page, KM_USER0);
328 if (*((__be32 *)(mapped_data + new_offset)) ==
329 cpu_to_be32(JBD2_MAGIC_NUMBER)) {
333 kunmap_atomic(mapped_data, KM_USER0);
336 * Do we need to do a data copy?
338 if (need_copy_out && !done_copy_out) {
341 jbd_unlock_bh_state(bh_in);
342 tmp = jbd2_alloc(bh_in->b_size, GFP_NOFS);
343 jbd_lock_bh_state(bh_in);
344 if (jh_in->b_frozen_data) {
345 jbd2_free(tmp, bh_in->b_size);
349 jh_in->b_frozen_data = tmp;
350 mapped_data = kmap_atomic(new_page, KM_USER0);
351 memcpy(tmp, mapped_data + new_offset, jh2bh(jh_in)->b_size);
352 kunmap_atomic(mapped_data, KM_USER0);
354 new_page = virt_to_page(tmp);
355 new_offset = offset_in_page(tmp);
360 * Did we need to do an escaping? Now we've done all the
361 * copying, we can finally do so.
364 mapped_data = kmap_atomic(new_page, KM_USER0);
365 *((unsigned int *)(mapped_data + new_offset)) = 0;
366 kunmap_atomic(mapped_data, KM_USER0);
369 /* keep subsequent assertions sane */
371 init_buffer(new_bh, NULL, NULL);
372 atomic_set(&new_bh->b_count, 1);
373 jbd_unlock_bh_state(bh_in);
375 new_jh = jbd2_journal_add_journal_head(new_bh); /* This sleeps */
377 set_bh_page(new_bh, new_page, new_offset);
378 new_jh->b_transaction = NULL;
379 new_bh->b_size = jh2bh(jh_in)->b_size;
380 new_bh->b_bdev = transaction->t_journal->j_dev;
381 new_bh->b_blocknr = blocknr;
382 set_buffer_mapped(new_bh);
383 set_buffer_dirty(new_bh);
388 * The to-be-written buffer needs to get moved to the io queue,
389 * and the original buffer whose contents we are shadowing or
390 * copying is moved to the transaction's shadow queue.
392 JBUFFER_TRACE(jh_in, "file as BJ_Shadow");
393 jbd2_journal_file_buffer(jh_in, transaction, BJ_Shadow);
394 JBUFFER_TRACE(new_jh, "file as BJ_IO");
395 jbd2_journal_file_buffer(new_jh, transaction, BJ_IO);
397 return do_escape | (done_copy_out << 1);
401 * Allocation code for the journal file. Manage the space left in the
402 * journal, so that we can begin checkpointing when appropriate.
406 * __jbd2_log_space_left: Return the number of free blocks left in the journal.
408 * Called with the journal already locked.
410 * Called under j_state_lock
413 int __jbd2_log_space_left(journal_t *journal)
415 int left = journal->j_free;
417 assert_spin_locked(&journal->j_state_lock);
420 * Be pessimistic here about the number of those free blocks which
421 * might be required for log descriptor control blocks.
424 #define MIN_LOG_RESERVED_BLOCKS 32 /* Allow for rounding errors */
426 left -= MIN_LOG_RESERVED_BLOCKS;
435 * Called under j_state_lock. Returns true if a transaction was started.
437 int __jbd2_log_start_commit(journal_t *journal, tid_t target)
440 * Are we already doing a recent enough commit?
442 if (!tid_geq(journal->j_commit_request, target)) {
444 * We want a new commit: OK, mark the request and wakup the
445 * commit thread. We do _not_ do the commit ourselves.
448 journal->j_commit_request = target;
449 jbd_debug(1, "JBD: requesting commit %d/%d\n",
450 journal->j_commit_request,
451 journal->j_commit_sequence);
452 wake_up(&journal->j_wait_commit);
458 int jbd2_log_start_commit(journal_t *journal, tid_t tid)
462 spin_lock(&journal->j_state_lock);
463 ret = __jbd2_log_start_commit(journal, tid);
464 spin_unlock(&journal->j_state_lock);
469 * Force and wait upon a commit if the calling process is not within
470 * transaction. This is used for forcing out undo-protected data which contains
471 * bitmaps, when the fs is running out of space.
473 * We can only force the running transaction if we don't have an active handle;
474 * otherwise, we will deadlock.
476 * Returns true if a transaction was started.
478 int jbd2_journal_force_commit_nested(journal_t *journal)
480 transaction_t *transaction = NULL;
483 spin_lock(&journal->j_state_lock);
484 if (journal->j_running_transaction && !current->journal_info) {
485 transaction = journal->j_running_transaction;
486 __jbd2_log_start_commit(journal, transaction->t_tid);
487 } else if (journal->j_committing_transaction)
488 transaction = journal->j_committing_transaction;
491 spin_unlock(&journal->j_state_lock);
492 return 0; /* Nothing to retry */
495 tid = transaction->t_tid;
496 spin_unlock(&journal->j_state_lock);
497 jbd2_log_wait_commit(journal, tid);
502 * Start a commit of the current running transaction (if any). Returns true
503 * if a transaction was started, and fills its tid in at *ptid
505 int jbd2_journal_start_commit(journal_t *journal, tid_t *ptid)
509 spin_lock(&journal->j_state_lock);
510 if (journal->j_running_transaction) {
511 tid_t tid = journal->j_running_transaction->t_tid;
513 ret = __jbd2_log_start_commit(journal, tid);
516 } else if (journal->j_committing_transaction && ptid) {
518 * If ext3_write_super() recently started a commit, then we
519 * have to wait for completion of that transaction
521 *ptid = journal->j_committing_transaction->t_tid;
524 spin_unlock(&journal->j_state_lock);
529 * Wait for a specified commit to complete.
530 * The caller may not hold the journal lock.
532 int jbd2_log_wait_commit(journal_t *journal, tid_t tid)
536 #ifdef CONFIG_JBD2_DEBUG
537 spin_lock(&journal->j_state_lock);
538 if (!tid_geq(journal->j_commit_request, tid)) {
540 "%s: error: j_commit_request=%d, tid=%d\n",
541 __func__, journal->j_commit_request, tid);
543 spin_unlock(&journal->j_state_lock);
545 spin_lock(&journal->j_state_lock);
546 while (tid_gt(tid, journal->j_commit_sequence)) {
547 jbd_debug(1, "JBD: want %d, j_commit_sequence=%d\n",
548 tid, journal->j_commit_sequence);
549 wake_up(&journal->j_wait_commit);
550 spin_unlock(&journal->j_state_lock);
551 wait_event(journal->j_wait_done_commit,
552 !tid_gt(tid, journal->j_commit_sequence));
553 spin_lock(&journal->j_state_lock);
555 spin_unlock(&journal->j_state_lock);
557 if (unlikely(is_journal_aborted(journal))) {
558 printk(KERN_EMERG "journal commit I/O error\n");
565 * Log buffer allocation routines:
568 int jbd2_journal_next_log_block(journal_t *journal, unsigned long long *retp)
570 unsigned long blocknr;
572 spin_lock(&journal->j_state_lock);
573 J_ASSERT(journal->j_free > 1);
575 blocknr = journal->j_head;
578 if (journal->j_head == journal->j_last)
579 journal->j_head = journal->j_first;
580 spin_unlock(&journal->j_state_lock);
581 return jbd2_journal_bmap(journal, blocknr, retp);
585 * Conversion of logical to physical block numbers for the journal
587 * On external journals the journal blocks are identity-mapped, so
588 * this is a no-op. If needed, we can use j_blk_offset - everything is
591 int jbd2_journal_bmap(journal_t *journal, unsigned long blocknr,
592 unsigned long long *retp)
595 unsigned long long ret;
597 if (journal->j_inode) {
598 ret = bmap(journal->j_inode, blocknr);
602 char b[BDEVNAME_SIZE];
604 printk(KERN_ALERT "%s: journal block not found "
605 "at offset %lu on %s\n",
608 bdevname(journal->j_dev, b));
610 __journal_abort_soft(journal, err);
613 *retp = blocknr; /* +journal->j_blk_offset */
619 * We play buffer_head aliasing tricks to write data/metadata blocks to
620 * the journal without copying their contents, but for journal
621 * descriptor blocks we do need to generate bona fide buffers.
623 * After the caller of jbd2_journal_get_descriptor_buffer() has finished modifying
624 * the buffer's contents they really should run flush_dcache_page(bh->b_page).
625 * But we don't bother doing that, so there will be coherency problems with
626 * mmaps of blockdevs which hold live JBD-controlled filesystems.
628 struct journal_head *jbd2_journal_get_descriptor_buffer(journal_t *journal)
630 struct buffer_head *bh;
631 unsigned long long blocknr;
634 err = jbd2_journal_next_log_block(journal, &blocknr);
639 bh = __getblk(journal->j_dev, blocknr, journal->j_blocksize);
641 memset(bh->b_data, 0, journal->j_blocksize);
642 set_buffer_uptodate(bh);
644 BUFFER_TRACE(bh, "return this buffer");
645 return jbd2_journal_add_journal_head(bh);
648 struct jbd2_stats_proc_session {
650 struct transaction_stats_s *stats;
655 static void *jbd2_history_skip_empty(struct jbd2_stats_proc_session *s,
656 struct transaction_stats_s *ts,
659 if (ts == s->stats + s->max)
661 if (!first && ts == s->stats + s->start)
663 while (ts->ts_type == 0) {
665 if (ts == s->stats + s->max)
667 if (ts == s->stats + s->start)
674 static void *jbd2_seq_history_start(struct seq_file *seq, loff_t *pos)
676 struct jbd2_stats_proc_session *s = seq->private;
677 struct transaction_stats_s *ts;
681 return SEQ_START_TOKEN;
682 ts = jbd2_history_skip_empty(s, s->stats + s->start, 1);
687 ts = jbd2_history_skip_empty(s, ++ts, 0);
695 static void *jbd2_seq_history_next(struct seq_file *seq, void *v, loff_t *pos)
697 struct jbd2_stats_proc_session *s = seq->private;
698 struct transaction_stats_s *ts = v;
701 if (v == SEQ_START_TOKEN)
702 return jbd2_history_skip_empty(s, s->stats + s->start, 1);
704 return jbd2_history_skip_empty(s, ++ts, 0);
707 static int jbd2_seq_history_show(struct seq_file *seq, void *v)
709 struct transaction_stats_s *ts = v;
710 if (v == SEQ_START_TOKEN) {
711 seq_printf(seq, "%-4s %-5s %-5s %-5s %-5s %-5s %-5s %-6s %-5s "
712 "%-5s %-5s %-5s %-5s %-5s\n", "R/C", "tid",
713 "wait", "run", "lock", "flush", "log", "hndls",
714 "block", "inlog", "ctime", "write", "drop",
718 if (ts->ts_type == JBD2_STATS_RUN)
719 seq_printf(seq, "%-4s %-5lu %-5u %-5u %-5u %-5u %-5u "
720 "%-6lu %-5lu %-5lu\n", "R", ts->ts_tid,
721 jiffies_to_msecs(ts->u.run.rs_wait),
722 jiffies_to_msecs(ts->u.run.rs_running),
723 jiffies_to_msecs(ts->u.run.rs_locked),
724 jiffies_to_msecs(ts->u.run.rs_flushing),
725 jiffies_to_msecs(ts->u.run.rs_logging),
726 ts->u.run.rs_handle_count,
728 ts->u.run.rs_blocks_logged);
729 else if (ts->ts_type == JBD2_STATS_CHECKPOINT)
730 seq_printf(seq, "%-4s %-5lu %48s %-5u %-5lu %-5lu %-5lu\n",
731 "C", ts->ts_tid, " ",
732 jiffies_to_msecs(ts->u.chp.cs_chp_time),
733 ts->u.chp.cs_written, ts->u.chp.cs_dropped,
734 ts->u.chp.cs_forced_to_close);
740 static void jbd2_seq_history_stop(struct seq_file *seq, void *v)
744 static struct seq_operations jbd2_seq_history_ops = {
745 .start = jbd2_seq_history_start,
746 .next = jbd2_seq_history_next,
747 .stop = jbd2_seq_history_stop,
748 .show = jbd2_seq_history_show,
751 static int jbd2_seq_history_open(struct inode *inode, struct file *file)
753 journal_t *journal = PDE(inode)->data;
754 struct jbd2_stats_proc_session *s;
757 s = kmalloc(sizeof(*s), GFP_KERNEL);
760 size = sizeof(struct transaction_stats_s) * journal->j_history_max;
761 s->stats = kmalloc(size, GFP_KERNEL);
762 if (s->stats == NULL) {
766 spin_lock(&journal->j_history_lock);
767 memcpy(s->stats, journal->j_history, size);
768 s->max = journal->j_history_max;
769 s->start = journal->j_history_cur % s->max;
770 spin_unlock(&journal->j_history_lock);
772 rc = seq_open(file, &jbd2_seq_history_ops);
774 struct seq_file *m = file->private_data;
784 static int jbd2_seq_history_release(struct inode *inode, struct file *file)
786 struct seq_file *seq = file->private_data;
787 struct jbd2_stats_proc_session *s = seq->private;
791 return seq_release(inode, file);
794 static struct file_operations jbd2_seq_history_fops = {
795 .owner = THIS_MODULE,
796 .open = jbd2_seq_history_open,
799 .release = jbd2_seq_history_release,
802 static void *jbd2_seq_info_start(struct seq_file *seq, loff_t *pos)
804 return *pos ? NULL : SEQ_START_TOKEN;
807 static void *jbd2_seq_info_next(struct seq_file *seq, void *v, loff_t *pos)
812 static int jbd2_seq_info_show(struct seq_file *seq, void *v)
814 struct jbd2_stats_proc_session *s = seq->private;
816 if (v != SEQ_START_TOKEN)
818 seq_printf(seq, "%lu transaction, each upto %u blocks\n",
820 s->journal->j_max_transaction_buffers);
821 if (s->stats->ts_tid == 0)
823 seq_printf(seq, "average: \n %ums waiting for transaction\n",
824 jiffies_to_msecs(s->stats->u.run.rs_wait / s->stats->ts_tid));
825 seq_printf(seq, " %ums running transaction\n",
826 jiffies_to_msecs(s->stats->u.run.rs_running / s->stats->ts_tid));
827 seq_printf(seq, " %ums transaction was being locked\n",
828 jiffies_to_msecs(s->stats->u.run.rs_locked / s->stats->ts_tid));
829 seq_printf(seq, " %ums flushing data (in ordered mode)\n",
830 jiffies_to_msecs(s->stats->u.run.rs_flushing / s->stats->ts_tid));
831 seq_printf(seq, " %ums logging transaction\n",
832 jiffies_to_msecs(s->stats->u.run.rs_logging / s->stats->ts_tid));
833 seq_printf(seq, " %lu handles per transaction\n",
834 s->stats->u.run.rs_handle_count / s->stats->ts_tid);
835 seq_printf(seq, " %lu blocks per transaction\n",
836 s->stats->u.run.rs_blocks / s->stats->ts_tid);
837 seq_printf(seq, " %lu logged blocks per transaction\n",
838 s->stats->u.run.rs_blocks_logged / s->stats->ts_tid);
842 static void jbd2_seq_info_stop(struct seq_file *seq, void *v)
846 static struct seq_operations jbd2_seq_info_ops = {
847 .start = jbd2_seq_info_start,
848 .next = jbd2_seq_info_next,
849 .stop = jbd2_seq_info_stop,
850 .show = jbd2_seq_info_show,
853 static int jbd2_seq_info_open(struct inode *inode, struct file *file)
855 journal_t *journal = PDE(inode)->data;
856 struct jbd2_stats_proc_session *s;
859 s = kmalloc(sizeof(*s), GFP_KERNEL);
862 size = sizeof(struct transaction_stats_s);
863 s->stats = kmalloc(size, GFP_KERNEL);
864 if (s->stats == NULL) {
868 spin_lock(&journal->j_history_lock);
869 memcpy(s->stats, &journal->j_stats, size);
870 s->journal = journal;
871 spin_unlock(&journal->j_history_lock);
873 rc = seq_open(file, &jbd2_seq_info_ops);
875 struct seq_file *m = file->private_data;
885 static int jbd2_seq_info_release(struct inode *inode, struct file *file)
887 struct seq_file *seq = file->private_data;
888 struct jbd2_stats_proc_session *s = seq->private;
891 return seq_release(inode, file);
894 static struct file_operations jbd2_seq_info_fops = {
895 .owner = THIS_MODULE,
896 .open = jbd2_seq_info_open,
899 .release = jbd2_seq_info_release,
902 static struct proc_dir_entry *proc_jbd2_stats;
904 static void jbd2_stats_proc_init(journal_t *journal)
906 char name[BDEVNAME_SIZE];
908 bdevname(journal->j_dev, name);
909 journal->j_proc_entry = proc_mkdir(name, proc_jbd2_stats);
910 if (journal->j_proc_entry) {
911 proc_create_data("history", S_IRUGO, journal->j_proc_entry,
912 &jbd2_seq_history_fops, journal);
913 proc_create_data("info", S_IRUGO, journal->j_proc_entry,
914 &jbd2_seq_info_fops, journal);
918 static void jbd2_stats_proc_exit(journal_t *journal)
920 char name[BDEVNAME_SIZE];
922 bdevname(journal->j_dev, name);
923 remove_proc_entry("info", journal->j_proc_entry);
924 remove_proc_entry("history", journal->j_proc_entry);
925 remove_proc_entry(name, proc_jbd2_stats);
928 static void journal_init_stats(journal_t *journal)
932 if (!proc_jbd2_stats)
935 journal->j_history_max = 100;
936 size = sizeof(struct transaction_stats_s) * journal->j_history_max;
937 journal->j_history = kzalloc(size, GFP_KERNEL);
938 if (!journal->j_history) {
939 journal->j_history_max = 0;
942 spin_lock_init(&journal->j_history_lock);
946 * Management for journal control blocks: functions to create and
947 * destroy journal_t structures, and to initialise and read existing
948 * journal blocks from disk. */
950 /* First: create and setup a journal_t object in memory. We initialise
951 * very few fields yet: that has to wait until we have created the
952 * journal structures from from scratch, or loaded them from disk. */
954 static journal_t * journal_init_common (void)
959 journal = kzalloc(sizeof(*journal), GFP_KERNEL|__GFP_NOFAIL);
963 init_waitqueue_head(&journal->j_wait_transaction_locked);
964 init_waitqueue_head(&journal->j_wait_logspace);
965 init_waitqueue_head(&journal->j_wait_done_commit);
966 init_waitqueue_head(&journal->j_wait_checkpoint);
967 init_waitqueue_head(&journal->j_wait_commit);
968 init_waitqueue_head(&journal->j_wait_updates);
969 mutex_init(&journal->j_barrier);
970 mutex_init(&journal->j_checkpoint_mutex);
971 spin_lock_init(&journal->j_revoke_lock);
972 spin_lock_init(&journal->j_list_lock);
973 spin_lock_init(&journal->j_state_lock);
975 journal->j_commit_interval = (HZ * JBD2_DEFAULT_MAX_COMMIT_AGE);
977 /* The journal is marked for error until we succeed with recovery! */
978 journal->j_flags = JBD2_ABORT;
980 /* Set up a default-sized revoke table for the new mount. */
981 err = jbd2_journal_init_revoke(journal, JOURNAL_REVOKE_DEFAULT_HASH);
987 journal_init_stats(journal);
994 /* jbd2_journal_init_dev and jbd2_journal_init_inode:
996 * Create a journal structure assigned some fixed set of disk blocks to
997 * the journal. We don't actually touch those disk blocks yet, but we
998 * need to set up all of the mapping information to tell the journaling
999 * system where the journal blocks are.
1004 * journal_t * jbd2_journal_init_dev() - creates and initialises a journal structure
1005 * @bdev: Block device on which to create the journal
1006 * @fs_dev: Device which hold journalled filesystem for this journal.
1007 * @start: Block nr Start of journal.
1008 * @len: Length of the journal in blocks.
1009 * @blocksize: blocksize of journalling device
1011 * Returns: a newly created journal_t *
1013 * jbd2_journal_init_dev creates a journal which maps a fixed contiguous
1014 * range of blocks on an arbitrary block device.
1017 journal_t * jbd2_journal_init_dev(struct block_device *bdev,
1018 struct block_device *fs_dev,
1019 unsigned long long start, int len, int blocksize)
1021 journal_t *journal = journal_init_common();
1022 struct buffer_head *bh;
1028 /* journal descriptor can store up to n blocks -bzzz */
1029 journal->j_blocksize = blocksize;
1030 n = journal->j_blocksize / sizeof(journal_block_tag_t);
1031 journal->j_wbufsize = n;
1032 journal->j_wbuf = kmalloc(n * sizeof(struct buffer_head*), GFP_KERNEL);
1033 if (!journal->j_wbuf) {
1034 printk(KERN_ERR "%s: Cant allocate bhs for commit thread\n",
1040 journal->j_dev = bdev;
1041 journal->j_fs_dev = fs_dev;
1042 journal->j_blk_offset = start;
1043 journal->j_maxlen = len;
1044 jbd2_stats_proc_init(journal);
1046 bh = __getblk(journal->j_dev, start, journal->j_blocksize);
1047 J_ASSERT(bh != NULL);
1048 journal->j_sb_buffer = bh;
1049 journal->j_superblock = (journal_superblock_t *)bh->b_data;
1055 * journal_t * jbd2_journal_init_inode () - creates a journal which maps to a inode.
1056 * @inode: An inode to create the journal in
1058 * jbd2_journal_init_inode creates a journal which maps an on-disk inode as
1059 * the journal. The inode must exist already, must support bmap() and
1060 * must have all data blocks preallocated.
1062 journal_t * jbd2_journal_init_inode (struct inode *inode)
1064 struct buffer_head *bh;
1065 journal_t *journal = journal_init_common();
1068 unsigned long long blocknr;
1073 journal->j_dev = journal->j_fs_dev = inode->i_sb->s_bdev;
1074 journal->j_inode = inode;
1076 "journal %p: inode %s/%ld, size %Ld, bits %d, blksize %ld\n",
1077 journal, inode->i_sb->s_id, inode->i_ino,
1078 (long long) inode->i_size,
1079 inode->i_sb->s_blocksize_bits, inode->i_sb->s_blocksize);
1081 journal->j_maxlen = inode->i_size >> inode->i_sb->s_blocksize_bits;
1082 journal->j_blocksize = inode->i_sb->s_blocksize;
1083 jbd2_stats_proc_init(journal);
1085 /* journal descriptor can store up to n blocks -bzzz */
1086 n = journal->j_blocksize / sizeof(journal_block_tag_t);
1087 journal->j_wbufsize = n;
1088 journal->j_wbuf = kmalloc(n * sizeof(struct buffer_head*), GFP_KERNEL);
1089 if (!journal->j_wbuf) {
1090 printk(KERN_ERR "%s: Cant allocate bhs for commit thread\n",
1096 err = jbd2_journal_bmap(journal, 0, &blocknr);
1097 /* If that failed, give up */
1099 printk(KERN_ERR "%s: Cannnot locate journal superblock\n",
1105 bh = __getblk(journal->j_dev, blocknr, journal->j_blocksize);
1106 J_ASSERT(bh != NULL);
1107 journal->j_sb_buffer = bh;
1108 journal->j_superblock = (journal_superblock_t *)bh->b_data;
1114 * If the journal init or create aborts, we need to mark the journal
1115 * superblock as being NULL to prevent the journal destroy from writing
1116 * back a bogus superblock.
1118 static void journal_fail_superblock (journal_t *journal)
1120 struct buffer_head *bh = journal->j_sb_buffer;
1122 journal->j_sb_buffer = NULL;
1126 * Given a journal_t structure, initialise the various fields for
1127 * startup of a new journaling session. We use this both when creating
1128 * a journal, and after recovering an old journal to reset it for
1132 static int journal_reset(journal_t *journal)
1134 journal_superblock_t *sb = journal->j_superblock;
1135 unsigned long long first, last;
1137 first = be32_to_cpu(sb->s_first);
1138 last = be32_to_cpu(sb->s_maxlen);
1140 journal->j_first = first;
1141 journal->j_last = last;
1143 journal->j_head = first;
1144 journal->j_tail = first;
1145 journal->j_free = last - first;
1147 journal->j_tail_sequence = journal->j_transaction_sequence;
1148 journal->j_commit_sequence = journal->j_transaction_sequence - 1;
1149 journal->j_commit_request = journal->j_commit_sequence;
1151 journal->j_max_transaction_buffers = journal->j_maxlen / 4;
1153 /* Add the dynamic fields and write it to disk. */
1154 jbd2_journal_update_superblock(journal, 1);
1155 return jbd2_journal_start_thread(journal);
1159 * int jbd2_journal_create() - Initialise the new journal file
1160 * @journal: Journal to create. This structure must have been initialised
1162 * Given a journal_t structure which tells us which disk blocks we can
1163 * use, create a new journal superblock and initialise all of the
1164 * journal fields from scratch.
1166 int jbd2_journal_create(journal_t *journal)
1168 unsigned long long blocknr;
1169 struct buffer_head *bh;
1170 journal_superblock_t *sb;
1173 if (journal->j_maxlen < JBD2_MIN_JOURNAL_BLOCKS) {
1174 printk (KERN_ERR "Journal length (%d blocks) too short.\n",
1176 journal_fail_superblock(journal);
1180 if (journal->j_inode == NULL) {
1182 * We don't know what block to start at!
1185 "%s: creation of journal on external device!\n",
1190 /* Zero out the entire journal on disk. We cannot afford to
1191 have any blocks on disk beginning with JBD2_MAGIC_NUMBER. */
1192 jbd_debug(1, "JBD: Zeroing out journal blocks...\n");
1193 for (i = 0; i < journal->j_maxlen; i++) {
1194 err = jbd2_journal_bmap(journal, i, &blocknr);
1197 bh = __getblk(journal->j_dev, blocknr, journal->j_blocksize);
1199 memset (bh->b_data, 0, journal->j_blocksize);
1200 BUFFER_TRACE(bh, "marking dirty");
1201 mark_buffer_dirty(bh);
1202 BUFFER_TRACE(bh, "marking uptodate");
1203 set_buffer_uptodate(bh);
1208 sync_blockdev(journal->j_dev);
1209 jbd_debug(1, "JBD: journal cleared.\n");
1211 /* OK, fill in the initial static fields in the new superblock */
1212 sb = journal->j_superblock;
1214 sb->s_header.h_magic = cpu_to_be32(JBD2_MAGIC_NUMBER);
1215 sb->s_header.h_blocktype = cpu_to_be32(JBD2_SUPERBLOCK_V2);
1217 sb->s_blocksize = cpu_to_be32(journal->j_blocksize);
1218 sb->s_maxlen = cpu_to_be32(journal->j_maxlen);
1219 sb->s_first = cpu_to_be32(1);
1221 journal->j_transaction_sequence = 1;
1223 journal->j_flags &= ~JBD2_ABORT;
1224 journal->j_format_version = 2;
1226 return journal_reset(journal);
1230 * void jbd2_journal_update_superblock() - Update journal sb on disk.
1231 * @journal: The journal to update.
1232 * @wait: Set to '0' if you don't want to wait for IO completion.
1234 * Update a journal's dynamic superblock fields and write it to disk,
1235 * optionally waiting for the IO to complete.
1237 void jbd2_journal_update_superblock(journal_t *journal, int wait)
1239 journal_superblock_t *sb = journal->j_superblock;
1240 struct buffer_head *bh = journal->j_sb_buffer;
1243 * As a special case, if the on-disk copy is already marked as needing
1244 * no recovery (s_start == 0) and there are no outstanding transactions
1245 * in the filesystem, then we can safely defer the superblock update
1246 * until the next commit by setting JBD2_FLUSHED. This avoids
1247 * attempting a write to a potential-readonly device.
1249 if (sb->s_start == 0 && journal->j_tail_sequence ==
1250 journal->j_transaction_sequence) {
1251 jbd_debug(1,"JBD: Skipping superblock update on recovered sb "
1252 "(start %ld, seq %d, errno %d)\n",
1253 journal->j_tail, journal->j_tail_sequence,
1258 spin_lock(&journal->j_state_lock);
1259 jbd_debug(1,"JBD: updating superblock (start %ld, seq %d, errno %d)\n",
1260 journal->j_tail, journal->j_tail_sequence, journal->j_errno);
1262 sb->s_sequence = cpu_to_be32(journal->j_tail_sequence);
1263 sb->s_start = cpu_to_be32(journal->j_tail);
1264 sb->s_errno = cpu_to_be32(journal->j_errno);
1265 spin_unlock(&journal->j_state_lock);
1267 BUFFER_TRACE(bh, "marking dirty");
1268 mark_buffer_dirty(bh);
1270 sync_dirty_buffer(bh);
1272 ll_rw_block(SWRITE, 1, &bh);
1275 /* If we have just flushed the log (by marking s_start==0), then
1276 * any future commit will have to be careful to update the
1277 * superblock again to re-record the true start of the log. */
1279 spin_lock(&journal->j_state_lock);
1281 journal->j_flags &= ~JBD2_FLUSHED;
1283 journal->j_flags |= JBD2_FLUSHED;
1284 spin_unlock(&journal->j_state_lock);
1288 * Read the superblock for a given journal, performing initial
1289 * validation of the format.
1292 static int journal_get_superblock(journal_t *journal)
1294 struct buffer_head *bh;
1295 journal_superblock_t *sb;
1298 bh = journal->j_sb_buffer;
1300 J_ASSERT(bh != NULL);
1301 if (!buffer_uptodate(bh)) {
1302 ll_rw_block(READ, 1, &bh);
1304 if (!buffer_uptodate(bh)) {
1306 "JBD: IO error reading journal superblock\n");
1311 sb = journal->j_superblock;
1315 if (sb->s_header.h_magic != cpu_to_be32(JBD2_MAGIC_NUMBER) ||
1316 sb->s_blocksize != cpu_to_be32(journal->j_blocksize)) {
1317 printk(KERN_WARNING "JBD: no valid journal superblock found\n");
1321 switch(be32_to_cpu(sb->s_header.h_blocktype)) {
1322 case JBD2_SUPERBLOCK_V1:
1323 journal->j_format_version = 1;
1325 case JBD2_SUPERBLOCK_V2:
1326 journal->j_format_version = 2;
1329 printk(KERN_WARNING "JBD: unrecognised superblock format ID\n");
1333 if (be32_to_cpu(sb->s_maxlen) < journal->j_maxlen)
1334 journal->j_maxlen = be32_to_cpu(sb->s_maxlen);
1335 else if (be32_to_cpu(sb->s_maxlen) > journal->j_maxlen) {
1336 printk (KERN_WARNING "JBD: journal file too short\n");
1343 journal_fail_superblock(journal);
1348 * Load the on-disk journal superblock and read the key fields into the
1352 static int load_superblock(journal_t *journal)
1355 journal_superblock_t *sb;
1357 err = journal_get_superblock(journal);
1361 sb = journal->j_superblock;
1363 journal->j_tail_sequence = be32_to_cpu(sb->s_sequence);
1364 journal->j_tail = be32_to_cpu(sb->s_start);
1365 journal->j_first = be32_to_cpu(sb->s_first);
1366 journal->j_last = be32_to_cpu(sb->s_maxlen);
1367 journal->j_errno = be32_to_cpu(sb->s_errno);
1374 * int jbd2_journal_load() - Read journal from disk.
1375 * @journal: Journal to act on.
1377 * Given a journal_t structure which tells us which disk blocks contain
1378 * a journal, read the journal from disk to initialise the in-memory
1381 int jbd2_journal_load(journal_t *journal)
1384 journal_superblock_t *sb;
1386 err = load_superblock(journal);
1390 sb = journal->j_superblock;
1391 /* If this is a V2 superblock, then we have to check the
1392 * features flags on it. */
1394 if (journal->j_format_version >= 2) {
1395 if ((sb->s_feature_ro_compat &
1396 ~cpu_to_be32(JBD2_KNOWN_ROCOMPAT_FEATURES)) ||
1397 (sb->s_feature_incompat &
1398 ~cpu_to_be32(JBD2_KNOWN_INCOMPAT_FEATURES))) {
1399 printk (KERN_WARNING
1400 "JBD: Unrecognised features on journal\n");
1405 /* Let the recovery code check whether it needs to recover any
1406 * data from the journal. */
1407 if (jbd2_journal_recover(journal))
1408 goto recovery_error;
1410 /* OK, we've finished with the dynamic journal bits:
1411 * reinitialise the dynamic contents of the superblock in memory
1412 * and reset them on disk. */
1413 if (journal_reset(journal))
1414 goto recovery_error;
1416 journal->j_flags &= ~JBD2_ABORT;
1417 journal->j_flags |= JBD2_LOADED;
1421 printk (KERN_WARNING "JBD: recovery failed\n");
1426 * void jbd2_journal_destroy() - Release a journal_t structure.
1427 * @journal: Journal to act on.
1429 * Release a journal_t structure once it is no longer in use by the
1432 void jbd2_journal_destroy(journal_t *journal)
1434 /* Wait for the commit thread to wake up and die. */
1435 journal_kill_thread(journal);
1437 /* Force a final log commit */
1438 if (journal->j_running_transaction)
1439 jbd2_journal_commit_transaction(journal);
1441 /* Force any old transactions to disk */
1443 /* Totally anal locking here... */
1444 spin_lock(&journal->j_list_lock);
1445 while (journal->j_checkpoint_transactions != NULL) {
1446 spin_unlock(&journal->j_list_lock);
1447 jbd2_log_do_checkpoint(journal);
1448 spin_lock(&journal->j_list_lock);
1451 J_ASSERT(journal->j_running_transaction == NULL);
1452 J_ASSERT(journal->j_committing_transaction == NULL);
1453 J_ASSERT(journal->j_checkpoint_transactions == NULL);
1454 spin_unlock(&journal->j_list_lock);
1456 /* We can now mark the journal as empty. */
1457 journal->j_tail = 0;
1458 journal->j_tail_sequence = ++journal->j_transaction_sequence;
1459 if (journal->j_sb_buffer) {
1460 jbd2_journal_update_superblock(journal, 1);
1461 brelse(journal->j_sb_buffer);
1464 if (journal->j_proc_entry)
1465 jbd2_stats_proc_exit(journal);
1466 if (journal->j_inode)
1467 iput(journal->j_inode);
1468 if (journal->j_revoke)
1469 jbd2_journal_destroy_revoke(journal);
1470 kfree(journal->j_wbuf);
1476 *int jbd2_journal_check_used_features () - Check if features specified are used.
1477 * @journal: Journal to check.
1478 * @compat: bitmask of compatible features
1479 * @ro: bitmask of features that force read-only mount
1480 * @incompat: bitmask of incompatible features
1482 * Check whether the journal uses all of a given set of
1483 * features. Return true (non-zero) if it does.
1486 int jbd2_journal_check_used_features (journal_t *journal, unsigned long compat,
1487 unsigned long ro, unsigned long incompat)
1489 journal_superblock_t *sb;
1491 if (!compat && !ro && !incompat)
1493 if (journal->j_format_version == 1)
1496 sb = journal->j_superblock;
1498 if (((be32_to_cpu(sb->s_feature_compat) & compat) == compat) &&
1499 ((be32_to_cpu(sb->s_feature_ro_compat) & ro) == ro) &&
1500 ((be32_to_cpu(sb->s_feature_incompat) & incompat) == incompat))
1507 * int jbd2_journal_check_available_features() - Check feature set in journalling layer
1508 * @journal: Journal to check.
1509 * @compat: bitmask of compatible features
1510 * @ro: bitmask of features that force read-only mount
1511 * @incompat: bitmask of incompatible features
1513 * Check whether the journaling code supports the use of
1514 * all of a given set of features on this journal. Return true
1515 * (non-zero) if it can. */
1517 int jbd2_journal_check_available_features (journal_t *journal, unsigned long compat,
1518 unsigned long ro, unsigned long incompat)
1520 journal_superblock_t *sb;
1522 if (!compat && !ro && !incompat)
1525 sb = journal->j_superblock;
1527 /* We can support any known requested features iff the
1528 * superblock is in version 2. Otherwise we fail to support any
1529 * extended sb features. */
1531 if (journal->j_format_version != 2)
1534 if ((compat & JBD2_KNOWN_COMPAT_FEATURES) == compat &&
1535 (ro & JBD2_KNOWN_ROCOMPAT_FEATURES) == ro &&
1536 (incompat & JBD2_KNOWN_INCOMPAT_FEATURES) == incompat)
1543 * int jbd2_journal_set_features () - Mark a given journal feature in the superblock
1544 * @journal: Journal to act on.
1545 * @compat: bitmask of compatible features
1546 * @ro: bitmask of features that force read-only mount
1547 * @incompat: bitmask of incompatible features
1549 * Mark a given journal feature as present on the
1550 * superblock. Returns true if the requested features could be set.
1554 int jbd2_journal_set_features (journal_t *journal, unsigned long compat,
1555 unsigned long ro, unsigned long incompat)
1557 journal_superblock_t *sb;
1559 if (jbd2_journal_check_used_features(journal, compat, ro, incompat))
1562 if (!jbd2_journal_check_available_features(journal, compat, ro, incompat))
1565 jbd_debug(1, "Setting new features 0x%lx/0x%lx/0x%lx\n",
1566 compat, ro, incompat);
1568 sb = journal->j_superblock;
1570 sb->s_feature_compat |= cpu_to_be32(compat);
1571 sb->s_feature_ro_compat |= cpu_to_be32(ro);
1572 sb->s_feature_incompat |= cpu_to_be32(incompat);
1578 * jbd2_journal_clear_features () - Clear a given journal feature in the
1580 * @journal: Journal to act on.
1581 * @compat: bitmask of compatible features
1582 * @ro: bitmask of features that force read-only mount
1583 * @incompat: bitmask of incompatible features
1585 * Clear a given journal feature as present on the
1588 void jbd2_journal_clear_features(journal_t *journal, unsigned long compat,
1589 unsigned long ro, unsigned long incompat)
1591 journal_superblock_t *sb;
1593 jbd_debug(1, "Clear features 0x%lx/0x%lx/0x%lx\n",
1594 compat, ro, incompat);
1596 sb = journal->j_superblock;
1598 sb->s_feature_compat &= ~cpu_to_be32(compat);
1599 sb->s_feature_ro_compat &= ~cpu_to_be32(ro);
1600 sb->s_feature_incompat &= ~cpu_to_be32(incompat);
1602 EXPORT_SYMBOL(jbd2_journal_clear_features);
1605 * int jbd2_journal_update_format () - Update on-disk journal structure.
1606 * @journal: Journal to act on.
1608 * Given an initialised but unloaded journal struct, poke about in the
1609 * on-disk structure to update it to the most recent supported version.
1611 int jbd2_journal_update_format (journal_t *journal)
1613 journal_superblock_t *sb;
1616 err = journal_get_superblock(journal);
1620 sb = journal->j_superblock;
1622 switch (be32_to_cpu(sb->s_header.h_blocktype)) {
1623 case JBD2_SUPERBLOCK_V2:
1625 case JBD2_SUPERBLOCK_V1:
1626 return journal_convert_superblock_v1(journal, sb);
1633 static int journal_convert_superblock_v1(journal_t *journal,
1634 journal_superblock_t *sb)
1636 int offset, blocksize;
1637 struct buffer_head *bh;
1640 "JBD: Converting superblock from version 1 to 2.\n");
1642 /* Pre-initialise new fields to zero */
1643 offset = ((char *) &(sb->s_feature_compat)) - ((char *) sb);
1644 blocksize = be32_to_cpu(sb->s_blocksize);
1645 memset(&sb->s_feature_compat, 0, blocksize-offset);
1647 sb->s_nr_users = cpu_to_be32(1);
1648 sb->s_header.h_blocktype = cpu_to_be32(JBD2_SUPERBLOCK_V2);
1649 journal->j_format_version = 2;
1651 bh = journal->j_sb_buffer;
1652 BUFFER_TRACE(bh, "marking dirty");
1653 mark_buffer_dirty(bh);
1654 sync_dirty_buffer(bh);
1660 * int jbd2_journal_flush () - Flush journal
1661 * @journal: Journal to act on.
1663 * Flush all data for a given journal to disk and empty the journal.
1664 * Filesystems can use this when remounting readonly to ensure that
1665 * recovery does not need to happen on remount.
1668 int jbd2_journal_flush(journal_t *journal)
1671 transaction_t *transaction = NULL;
1672 unsigned long old_tail;
1674 spin_lock(&journal->j_state_lock);
1676 /* Force everything buffered to the log... */
1677 if (journal->j_running_transaction) {
1678 transaction = journal->j_running_transaction;
1679 __jbd2_log_start_commit(journal, transaction->t_tid);
1680 } else if (journal->j_committing_transaction)
1681 transaction = journal->j_committing_transaction;
1683 /* Wait for the log commit to complete... */
1685 tid_t tid = transaction->t_tid;
1687 spin_unlock(&journal->j_state_lock);
1688 jbd2_log_wait_commit(journal, tid);
1690 spin_unlock(&journal->j_state_lock);
1693 /* ...and flush everything in the log out to disk. */
1694 spin_lock(&journal->j_list_lock);
1695 while (!err && journal->j_checkpoint_transactions != NULL) {
1696 spin_unlock(&journal->j_list_lock);
1697 err = jbd2_log_do_checkpoint(journal);
1698 spin_lock(&journal->j_list_lock);
1700 spin_unlock(&journal->j_list_lock);
1701 jbd2_cleanup_journal_tail(journal);
1703 /* Finally, mark the journal as really needing no recovery.
1704 * This sets s_start==0 in the underlying superblock, which is
1705 * the magic code for a fully-recovered superblock. Any future
1706 * commits of data to the journal will restore the current
1708 spin_lock(&journal->j_state_lock);
1709 old_tail = journal->j_tail;
1710 journal->j_tail = 0;
1711 spin_unlock(&journal->j_state_lock);
1712 jbd2_journal_update_superblock(journal, 1);
1713 spin_lock(&journal->j_state_lock);
1714 journal->j_tail = old_tail;
1716 J_ASSERT(!journal->j_running_transaction);
1717 J_ASSERT(!journal->j_committing_transaction);
1718 J_ASSERT(!journal->j_checkpoint_transactions);
1719 J_ASSERT(journal->j_head == journal->j_tail);
1720 J_ASSERT(journal->j_tail_sequence == journal->j_transaction_sequence);
1721 spin_unlock(&journal->j_state_lock);
1726 * int jbd2_journal_wipe() - Wipe journal contents
1727 * @journal: Journal to act on.
1728 * @write: flag (see below)
1730 * Wipe out all of the contents of a journal, safely. This will produce
1731 * a warning if the journal contains any valid recovery information.
1732 * Must be called between journal_init_*() and jbd2_journal_load().
1734 * If 'write' is non-zero, then we wipe out the journal on disk; otherwise
1735 * we merely suppress recovery.
1738 int jbd2_journal_wipe(journal_t *journal, int write)
1740 journal_superblock_t *sb;
1743 J_ASSERT (!(journal->j_flags & JBD2_LOADED));
1745 err = load_superblock(journal);
1749 sb = journal->j_superblock;
1751 if (!journal->j_tail)
1754 printk (KERN_WARNING "JBD: %s recovery information on journal\n",
1755 write ? "Clearing" : "Ignoring");
1757 err = jbd2_journal_skip_recovery(journal);
1759 jbd2_journal_update_superblock(journal, 1);
1766 * journal_dev_name: format a character string to describe on what
1767 * device this journal is present.
1770 static const char *journal_dev_name(journal_t *journal, char *buffer)
1772 struct block_device *bdev;
1774 if (journal->j_inode)
1775 bdev = journal->j_inode->i_sb->s_bdev;
1777 bdev = journal->j_dev;
1779 return bdevname(bdev, buffer);
1783 * Journal abort has very specific semantics, which we describe
1784 * for journal abort.
1786 * Two internal function, which provide abort to te jbd layer
1791 * Quick version for internal journal use (doesn't lock the journal).
1792 * Aborts hard --- we mark the abort as occurred, but do _nothing_ else,
1793 * and don't attempt to make any other journal updates.
1795 void __jbd2_journal_abort_hard(journal_t *journal)
1797 transaction_t *transaction;
1798 char b[BDEVNAME_SIZE];
1800 if (journal->j_flags & JBD2_ABORT)
1803 printk(KERN_ERR "Aborting journal on device %s.\n",
1804 journal_dev_name(journal, b));
1806 spin_lock(&journal->j_state_lock);
1807 journal->j_flags |= JBD2_ABORT;
1808 transaction = journal->j_running_transaction;
1810 __jbd2_log_start_commit(journal, transaction->t_tid);
1811 spin_unlock(&journal->j_state_lock);
1814 /* Soft abort: record the abort error status in the journal superblock,
1815 * but don't do any other IO. */
1816 static void __journal_abort_soft (journal_t *journal, int errno)
1818 if (journal->j_flags & JBD2_ABORT)
1821 if (!journal->j_errno)
1822 journal->j_errno = errno;
1824 __jbd2_journal_abort_hard(journal);
1827 jbd2_journal_update_superblock(journal, 1);
1831 * void jbd2_journal_abort () - Shutdown the journal immediately.
1832 * @journal: the journal to shutdown.
1833 * @errno: an error number to record in the journal indicating
1834 * the reason for the shutdown.
1836 * Perform a complete, immediate shutdown of the ENTIRE
1837 * journal (not of a single transaction). This operation cannot be
1838 * undone without closing and reopening the journal.
1840 * The jbd2_journal_abort function is intended to support higher level error
1841 * recovery mechanisms such as the ext2/ext3 remount-readonly error
1844 * Journal abort has very specific semantics. Any existing dirty,
1845 * unjournaled buffers in the main filesystem will still be written to
1846 * disk by bdflush, but the journaling mechanism will be suspended
1847 * immediately and no further transaction commits will be honoured.
1849 * Any dirty, journaled buffers will be written back to disk without
1850 * hitting the journal. Atomicity cannot be guaranteed on an aborted
1851 * filesystem, but we _do_ attempt to leave as much data as possible
1852 * behind for fsck to use for cleanup.
1854 * Any attempt to get a new transaction handle on a journal which is in
1855 * ABORT state will just result in an -EROFS error return. A
1856 * jbd2_journal_stop on an existing handle will return -EIO if we have
1857 * entered abort state during the update.
1859 * Recursive transactions are not disturbed by journal abort until the
1860 * final jbd2_journal_stop, which will receive the -EIO error.
1862 * Finally, the jbd2_journal_abort call allows the caller to supply an errno
1863 * which will be recorded (if possible) in the journal superblock. This
1864 * allows a client to record failure conditions in the middle of a
1865 * transaction without having to complete the transaction to record the
1866 * failure to disk. ext3_error, for example, now uses this
1869 * Errors which originate from within the journaling layer will NOT
1870 * supply an errno; a null errno implies that absolutely no further
1871 * writes are done to the journal (unless there are any already in
1876 void jbd2_journal_abort(journal_t *journal, int errno)
1878 __journal_abort_soft(journal, errno);
1882 * int jbd2_journal_errno () - returns the journal's error state.
1883 * @journal: journal to examine.
1885 * This is the errno numbet set with jbd2_journal_abort(), the last
1886 * time the journal was mounted - if the journal was stopped
1887 * without calling abort this will be 0.
1889 * If the journal has been aborted on this mount time -EROFS will
1892 int jbd2_journal_errno(journal_t *journal)
1896 spin_lock(&journal->j_state_lock);
1897 if (journal->j_flags & JBD2_ABORT)
1900 err = journal->j_errno;
1901 spin_unlock(&journal->j_state_lock);
1906 * int jbd2_journal_clear_err () - clears the journal's error state
1907 * @journal: journal to act on.
1909 * An error must be cleared or Acked to take a FS out of readonly
1912 int jbd2_journal_clear_err(journal_t *journal)
1916 spin_lock(&journal->j_state_lock);
1917 if (journal->j_flags & JBD2_ABORT)
1920 journal->j_errno = 0;
1921 spin_unlock(&journal->j_state_lock);
1926 * void jbd2_journal_ack_err() - Ack journal err.
1927 * @journal: journal to act on.
1929 * An error must be cleared or Acked to take a FS out of readonly
1932 void jbd2_journal_ack_err(journal_t *journal)
1934 spin_lock(&journal->j_state_lock);
1935 if (journal->j_errno)
1936 journal->j_flags |= JBD2_ACK_ERR;
1937 spin_unlock(&journal->j_state_lock);
1940 int jbd2_journal_blocks_per_page(struct inode *inode)
1942 return 1 << (PAGE_CACHE_SHIFT - inode->i_sb->s_blocksize_bits);
1946 * helper functions to deal with 32 or 64bit block numbers.
1948 size_t journal_tag_bytes(journal_t *journal)
1950 if (JBD2_HAS_INCOMPAT_FEATURE(journal, JBD2_FEATURE_INCOMPAT_64BIT))
1951 return JBD2_TAG_SIZE64;
1953 return JBD2_TAG_SIZE32;
1957 * Journal_head storage management
1959 static struct kmem_cache *jbd2_journal_head_cache;
1960 #ifdef CONFIG_JBD2_DEBUG
1961 static atomic_t nr_journal_heads = ATOMIC_INIT(0);
1964 static int journal_init_jbd2_journal_head_cache(void)
1968 J_ASSERT(jbd2_journal_head_cache == NULL);
1969 jbd2_journal_head_cache = kmem_cache_create("jbd2_journal_head",
1970 sizeof(struct journal_head),
1972 SLAB_TEMPORARY, /* flags */
1975 if (!jbd2_journal_head_cache) {
1977 printk(KERN_EMERG "JBD: no memory for journal_head cache\n");
1982 static void jbd2_journal_destroy_jbd2_journal_head_cache(void)
1984 if (jbd2_journal_head_cache) {
1985 kmem_cache_destroy(jbd2_journal_head_cache);
1986 jbd2_journal_head_cache = NULL;
1991 * journal_head splicing and dicing
1993 static struct journal_head *journal_alloc_journal_head(void)
1995 struct journal_head *ret;
1996 static unsigned long last_warning;
1998 #ifdef CONFIG_JBD2_DEBUG
1999 atomic_inc(&nr_journal_heads);
2001 ret = kmem_cache_alloc(jbd2_journal_head_cache, GFP_NOFS);
2003 jbd_debug(1, "out of memory for journal_head\n");
2004 if (time_after(jiffies, last_warning + 5*HZ)) {
2005 printk(KERN_NOTICE "ENOMEM in %s, retrying.\n",
2007 last_warning = jiffies;
2011 ret = kmem_cache_alloc(jbd2_journal_head_cache, GFP_NOFS);
2017 static void journal_free_journal_head(struct journal_head *jh)
2019 #ifdef CONFIG_JBD2_DEBUG
2020 atomic_dec(&nr_journal_heads);
2021 memset(jh, JBD2_POISON_FREE, sizeof(*jh));
2023 kmem_cache_free(jbd2_journal_head_cache, jh);
2027 * A journal_head is attached to a buffer_head whenever JBD has an
2028 * interest in the buffer.
2030 * Whenever a buffer has an attached journal_head, its ->b_state:BH_JBD bit
2031 * is set. This bit is tested in core kernel code where we need to take
2032 * JBD-specific actions. Testing the zeroness of ->b_private is not reliable
2035 * When a buffer has its BH_JBD bit set, its ->b_count is elevated by one.
2037 * When a buffer has its BH_JBD bit set it is immune from being released by
2038 * core kernel code, mainly via ->b_count.
2040 * A journal_head may be detached from its buffer_head when the journal_head's
2041 * b_transaction, b_cp_transaction and b_next_transaction pointers are NULL.
2042 * Various places in JBD call jbd2_journal_remove_journal_head() to indicate that the
2043 * journal_head can be dropped if needed.
2045 * Various places in the kernel want to attach a journal_head to a buffer_head
2046 * _before_ attaching the journal_head to a transaction. To protect the
2047 * journal_head in this situation, jbd2_journal_add_journal_head elevates the
2048 * journal_head's b_jcount refcount by one. The caller must call
2049 * jbd2_journal_put_journal_head() to undo this.
2051 * So the typical usage would be:
2053 * (Attach a journal_head if needed. Increments b_jcount)
2054 * struct journal_head *jh = jbd2_journal_add_journal_head(bh);
2056 * jh->b_transaction = xxx;
2057 * jbd2_journal_put_journal_head(jh);
2059 * Now, the journal_head's b_jcount is zero, but it is safe from being released
2060 * because it has a non-zero b_transaction.
2064 * Give a buffer_head a journal_head.
2066 * Doesn't need the journal lock.
2069 struct journal_head *jbd2_journal_add_journal_head(struct buffer_head *bh)
2071 struct journal_head *jh;
2072 struct journal_head *new_jh = NULL;
2075 if (!buffer_jbd(bh)) {
2076 new_jh = journal_alloc_journal_head();
2077 memset(new_jh, 0, sizeof(*new_jh));
2080 jbd_lock_bh_journal_head(bh);
2081 if (buffer_jbd(bh)) {
2085 (atomic_read(&bh->b_count) > 0) ||
2086 (bh->b_page && bh->b_page->mapping));
2089 jbd_unlock_bh_journal_head(bh);
2094 new_jh = NULL; /* We consumed it */
2099 BUFFER_TRACE(bh, "added journal_head");
2102 jbd_unlock_bh_journal_head(bh);
2104 journal_free_journal_head(new_jh);
2105 return bh->b_private;
2109 * Grab a ref against this buffer_head's journal_head. If it ended up not
2110 * having a journal_head, return NULL
2112 struct journal_head *jbd2_journal_grab_journal_head(struct buffer_head *bh)
2114 struct journal_head *jh = NULL;
2116 jbd_lock_bh_journal_head(bh);
2117 if (buffer_jbd(bh)) {
2121 jbd_unlock_bh_journal_head(bh);
2125 static void __journal_remove_journal_head(struct buffer_head *bh)
2127 struct journal_head *jh = bh2jh(bh);
2129 J_ASSERT_JH(jh, jh->b_jcount >= 0);
2132 if (jh->b_jcount == 0) {
2133 if (jh->b_transaction == NULL &&
2134 jh->b_next_transaction == NULL &&
2135 jh->b_cp_transaction == NULL) {
2136 J_ASSERT_JH(jh, jh->b_jlist == BJ_None);
2137 J_ASSERT_BH(bh, buffer_jbd(bh));
2138 J_ASSERT_BH(bh, jh2bh(jh) == bh);
2139 BUFFER_TRACE(bh, "remove journal_head");
2140 if (jh->b_frozen_data) {
2141 printk(KERN_WARNING "%s: freeing "
2144 jbd2_free(jh->b_frozen_data, bh->b_size);
2146 if (jh->b_committed_data) {
2147 printk(KERN_WARNING "%s: freeing "
2148 "b_committed_data\n",
2150 jbd2_free(jh->b_committed_data, bh->b_size);
2152 bh->b_private = NULL;
2153 jh->b_bh = NULL; /* debug, really */
2154 clear_buffer_jbd(bh);
2156 journal_free_journal_head(jh);
2158 BUFFER_TRACE(bh, "journal_head was locked");
2164 * jbd2_journal_remove_journal_head(): if the buffer isn't attached to a transaction
2165 * and has a zero b_jcount then remove and release its journal_head. If we did
2166 * see that the buffer is not used by any transaction we also "logically"
2167 * decrement ->b_count.
2169 * We in fact take an additional increment on ->b_count as a convenience,
2170 * because the caller usually wants to do additional things with the bh
2171 * after calling here.
2172 * The caller of jbd2_journal_remove_journal_head() *must* run __brelse(bh) at some
2173 * time. Once the caller has run __brelse(), the buffer is eligible for
2174 * reaping by try_to_free_buffers().
2176 void jbd2_journal_remove_journal_head(struct buffer_head *bh)
2178 jbd_lock_bh_journal_head(bh);
2179 __journal_remove_journal_head(bh);
2180 jbd_unlock_bh_journal_head(bh);
2184 * Drop a reference on the passed journal_head. If it fell to zero then try to
2185 * release the journal_head from the buffer_head.
2187 void jbd2_journal_put_journal_head(struct journal_head *jh)
2189 struct buffer_head *bh = jh2bh(jh);
2191 jbd_lock_bh_journal_head(bh);
2192 J_ASSERT_JH(jh, jh->b_jcount > 0);
2194 if (!jh->b_jcount && !jh->b_transaction) {
2195 __journal_remove_journal_head(bh);
2198 jbd_unlock_bh_journal_head(bh);
2202 * Initialize jbd inode head
2204 void jbd2_journal_init_jbd_inode(struct jbd2_inode *jinode, struct inode *inode)
2206 jinode->i_transaction = NULL;
2207 jinode->i_next_transaction = NULL;
2208 jinode->i_vfs_inode = inode;
2209 jinode->i_flags = 0;
2210 INIT_LIST_HEAD(&jinode->i_list);
2214 * Function to be called before we start removing inode from memory (i.e.,
2215 * clear_inode() is a fine place to be called from). It removes inode from
2216 * transaction's lists.
2218 void jbd2_journal_release_jbd_inode(journal_t *journal,
2219 struct jbd2_inode *jinode)
2226 spin_lock(&journal->j_list_lock);
2227 /* Is commit writing out inode - we have to wait */
2228 if (jinode->i_flags & JI_COMMIT_RUNNING) {
2229 wait_queue_head_t *wq;
2230 DEFINE_WAIT_BIT(wait, &jinode->i_flags, __JI_COMMIT_RUNNING);
2231 wq = bit_waitqueue(&jinode->i_flags, __JI_COMMIT_RUNNING);
2232 prepare_to_wait(wq, &wait.wait, TASK_UNINTERRUPTIBLE);
2233 spin_unlock(&journal->j_list_lock);
2235 finish_wait(wq, &wait.wait);
2239 /* Do we need to wait for data writeback? */
2240 if (journal->j_committing_transaction == jinode->i_transaction)
2242 if (jinode->i_transaction) {
2243 list_del(&jinode->i_list);
2244 jinode->i_transaction = NULL;
2246 spin_unlock(&journal->j_list_lock);
2252 #ifdef CONFIG_JBD2_DEBUG
2253 u8 jbd2_journal_enable_debug __read_mostly;
2254 EXPORT_SYMBOL(jbd2_journal_enable_debug);
2256 #define JBD2_DEBUG_NAME "jbd2-debug"
2258 static struct dentry *jbd2_debugfs_dir;
2259 static struct dentry *jbd2_debug;
2261 static void __init jbd2_create_debugfs_entry(void)
2263 jbd2_debugfs_dir = debugfs_create_dir("jbd2", NULL);
2264 if (jbd2_debugfs_dir)
2265 jbd2_debug = debugfs_create_u8(JBD2_DEBUG_NAME, S_IRUGO,
2267 &jbd2_journal_enable_debug);
2270 static void __exit jbd2_remove_debugfs_entry(void)
2272 debugfs_remove(jbd2_debug);
2273 debugfs_remove(jbd2_debugfs_dir);
2278 static void __init jbd2_create_debugfs_entry(void)
2282 static void __exit jbd2_remove_debugfs_entry(void)
2288 #ifdef CONFIG_PROC_FS
2290 #define JBD2_STATS_PROC_NAME "fs/jbd2"
2292 static void __init jbd2_create_jbd_stats_proc_entry(void)
2294 proc_jbd2_stats = proc_mkdir(JBD2_STATS_PROC_NAME, NULL);
2297 static void __exit jbd2_remove_jbd_stats_proc_entry(void)
2299 if (proc_jbd2_stats)
2300 remove_proc_entry(JBD2_STATS_PROC_NAME, NULL);
2305 #define jbd2_create_jbd_stats_proc_entry() do {} while (0)
2306 #define jbd2_remove_jbd_stats_proc_entry() do {} while (0)
2310 struct kmem_cache *jbd2_handle_cache;
2312 static int __init journal_init_handle_cache(void)
2314 jbd2_handle_cache = kmem_cache_create("jbd2_journal_handle",
2317 SLAB_TEMPORARY, /* flags */
2319 if (jbd2_handle_cache == NULL) {
2320 printk(KERN_EMERG "JBD: failed to create handle cache\n");
2326 static void jbd2_journal_destroy_handle_cache(void)
2328 if (jbd2_handle_cache)
2329 kmem_cache_destroy(jbd2_handle_cache);
2333 * Module startup and shutdown
2336 static int __init journal_init_caches(void)
2340 ret = jbd2_journal_init_revoke_caches();
2342 ret = journal_init_jbd2_journal_head_cache();
2344 ret = journal_init_handle_cache();
2348 static void jbd2_journal_destroy_caches(void)
2350 jbd2_journal_destroy_revoke_caches();
2351 jbd2_journal_destroy_jbd2_journal_head_cache();
2352 jbd2_journal_destroy_handle_cache();
2355 static int __init journal_init(void)
2359 BUILD_BUG_ON(sizeof(struct journal_superblock_s) != 1024);
2361 ret = journal_init_caches();
2363 jbd2_create_debugfs_entry();
2364 jbd2_create_jbd_stats_proc_entry();
2366 jbd2_journal_destroy_caches();
2371 static void __exit journal_exit(void)
2373 #ifdef CONFIG_JBD2_DEBUG
2374 int n = atomic_read(&nr_journal_heads);
2376 printk(KERN_EMERG "JBD: leaked %d journal_heads!\n", n);
2378 jbd2_remove_debugfs_entry();
2379 jbd2_remove_jbd_stats_proc_entry();
2380 jbd2_journal_destroy_caches();
2383 MODULE_LICENSE("GPL");
2384 module_init(journal_init);
2385 module_exit(journal_exit);