]> bbs.cooldavid.org Git - net-next-2.6.git/blob - fs/jbd2/journal.c
jbd2: don't wipe the journal on a failed journal checksum
[net-next-2.6.git] / fs / jbd2 / journal.c
1 /*
2  * linux/fs/jbd2/journal.c
3  *
4  * Written by Stephen C. Tweedie <sct@redhat.com>, 1998
5  *
6  * Copyright 1998 Red Hat corp --- All Rights Reserved
7  *
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.
11  *
12  * Generic filesystem journal-writing code; part of the ext2fs
13  * journaling system.
14  *
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.
18  *
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).
23  */
24
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/debugfs.h>
39 #include <linux/seq_file.h>
40 #include <linux/math64.h>
41 #include <linux/hash.h>
42
43 #define CREATE_TRACE_POINTS
44 #include <trace/events/jbd2.h>
45
46 #include <asm/uaccess.h>
47 #include <asm/page.h>
48
49 EXPORT_SYMBOL(jbd2_journal_start);
50 EXPORT_SYMBOL(jbd2_journal_restart);
51 EXPORT_SYMBOL(jbd2_journal_extend);
52 EXPORT_SYMBOL(jbd2_journal_stop);
53 EXPORT_SYMBOL(jbd2_journal_lock_updates);
54 EXPORT_SYMBOL(jbd2_journal_unlock_updates);
55 EXPORT_SYMBOL(jbd2_journal_get_write_access);
56 EXPORT_SYMBOL(jbd2_journal_get_create_access);
57 EXPORT_SYMBOL(jbd2_journal_get_undo_access);
58 EXPORT_SYMBOL(jbd2_journal_set_triggers);
59 EXPORT_SYMBOL(jbd2_journal_dirty_metadata);
60 EXPORT_SYMBOL(jbd2_journal_release_buffer);
61 EXPORT_SYMBOL(jbd2_journal_forget);
62 #if 0
63 EXPORT_SYMBOL(journal_sync_buffer);
64 #endif
65 EXPORT_SYMBOL(jbd2_journal_flush);
66 EXPORT_SYMBOL(jbd2_journal_revoke);
67
68 EXPORT_SYMBOL(jbd2_journal_init_dev);
69 EXPORT_SYMBOL(jbd2_journal_init_inode);
70 EXPORT_SYMBOL(jbd2_journal_update_format);
71 EXPORT_SYMBOL(jbd2_journal_check_used_features);
72 EXPORT_SYMBOL(jbd2_journal_check_available_features);
73 EXPORT_SYMBOL(jbd2_journal_set_features);
74 EXPORT_SYMBOL(jbd2_journal_load);
75 EXPORT_SYMBOL(jbd2_journal_destroy);
76 EXPORT_SYMBOL(jbd2_journal_abort);
77 EXPORT_SYMBOL(jbd2_journal_errno);
78 EXPORT_SYMBOL(jbd2_journal_ack_err);
79 EXPORT_SYMBOL(jbd2_journal_clear_err);
80 EXPORT_SYMBOL(jbd2_log_wait_commit);
81 EXPORT_SYMBOL(jbd2_journal_start_commit);
82 EXPORT_SYMBOL(jbd2_journal_force_commit_nested);
83 EXPORT_SYMBOL(jbd2_journal_wipe);
84 EXPORT_SYMBOL(jbd2_journal_blocks_per_page);
85 EXPORT_SYMBOL(jbd2_journal_invalidatepage);
86 EXPORT_SYMBOL(jbd2_journal_try_to_free_buffers);
87 EXPORT_SYMBOL(jbd2_journal_force_commit);
88 EXPORT_SYMBOL(jbd2_journal_file_inode);
89 EXPORT_SYMBOL(jbd2_journal_init_jbd_inode);
90 EXPORT_SYMBOL(jbd2_journal_release_jbd_inode);
91 EXPORT_SYMBOL(jbd2_journal_begin_ordered_truncate);
92
93 static int journal_convert_superblock_v1(journal_t *, journal_superblock_t *);
94 static void __journal_abort_soft (journal_t *journal, int errno);
95
96 /*
97  * Helper function used to manage commit timeouts
98  */
99
100 static void commit_timeout(unsigned long __data)
101 {
102         struct task_struct * p = (struct task_struct *) __data;
103
104         wake_up_process(p);
105 }
106
107 /*
108  * kjournald2: The main thread function used to manage a logging device
109  * journal.
110  *
111  * This kernel thread is responsible for two things:
112  *
113  * 1) COMMIT:  Every so often we need to commit the current state of the
114  *    filesystem to disk.  The journal thread is responsible for writing
115  *    all of the metadata buffers to disk.
116  *
117  * 2) CHECKPOINT: We cannot reuse a used section of the log file until all
118  *    of the data in that part of the log has been rewritten elsewhere on
119  *    the disk.  Flushing these old buffers to reclaim space in the log is
120  *    known as checkpointing, and this thread is responsible for that job.
121  */
122
123 static int kjournald2(void *arg)
124 {
125         journal_t *journal = arg;
126         transaction_t *transaction;
127
128         /*
129          * Set up an interval timer which can be used to trigger a commit wakeup
130          * after the commit interval expires
131          */
132         setup_timer(&journal->j_commit_timer, commit_timeout,
133                         (unsigned long)current);
134
135         /* Record that the journal thread is running */
136         journal->j_task = current;
137         wake_up(&journal->j_wait_done_commit);
138
139         /*
140          * And now, wait forever for commit wakeup events.
141          */
142         spin_lock(&journal->j_state_lock);
143
144 loop:
145         if (journal->j_flags & JBD2_UNMOUNT)
146                 goto end_loop;
147
148         jbd_debug(1, "commit_sequence=%d, commit_request=%d\n",
149                 journal->j_commit_sequence, journal->j_commit_request);
150
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);
157                 goto loop;
158         }
159
160         wake_up(&journal->j_wait_done_commit);
161         if (freezing(current)) {
162                 /*
163                  * The simpler the better. Flushing journal isn't a
164                  * good idea, because that depends on threads that may
165                  * be already stopped.
166                  */
167                 jbd_debug(1, "Now suspending kjournald2\n");
168                 spin_unlock(&journal->j_state_lock);
169                 refrigerator();
170                 spin_lock(&journal->j_state_lock);
171         } else {
172                 /*
173                  * We assume on resume that commits are already there,
174                  * so we don't sleep
175                  */
176                 DEFINE_WAIT(wait);
177                 int should_sleep = 1;
178
179                 prepare_to_wait(&journal->j_wait_commit, &wait,
180                                 TASK_INTERRUPTIBLE);
181                 if (journal->j_commit_sequence != journal->j_commit_request)
182                         should_sleep = 0;
183                 transaction = journal->j_running_transaction;
184                 if (transaction && time_after_eq(jiffies,
185                                                 transaction->t_expires))
186                         should_sleep = 0;
187                 if (journal->j_flags & JBD2_UNMOUNT)
188                         should_sleep = 0;
189                 if (should_sleep) {
190                         spin_unlock(&journal->j_state_lock);
191                         schedule();
192                         spin_lock(&journal->j_state_lock);
193                 }
194                 finish_wait(&journal->j_wait_commit, &wait);
195         }
196
197         jbd_debug(1, "kjournald2 wakes\n");
198
199         /*
200          * Were we woken up by a commit wakeup event?
201          */
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");
206         }
207         goto loop;
208
209 end_loop:
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");
215         return 0;
216 }
217
218 static int jbd2_journal_start_thread(journal_t *journal)
219 {
220         struct task_struct *t;
221
222         t = kthread_run(kjournald2, journal, "jbd2/%s",
223                         journal->j_devname);
224         if (IS_ERR(t))
225                 return PTR_ERR(t);
226
227         wait_event(journal->j_wait_done_commit, journal->j_task != NULL);
228         return 0;
229 }
230
231 static void journal_kill_thread(journal_t *journal)
232 {
233         spin_lock(&journal->j_state_lock);
234         journal->j_flags |= JBD2_UNMOUNT;
235
236         while (journal->j_task) {
237                 wake_up(&journal->j_wait_commit);
238                 spin_unlock(&journal->j_state_lock);
239                 wait_event(journal->j_wait_done_commit, journal->j_task == NULL);
240                 spin_lock(&journal->j_state_lock);
241         }
242         spin_unlock(&journal->j_state_lock);
243 }
244
245 /*
246  * jbd2_journal_write_metadata_buffer: write a metadata buffer to the journal.
247  *
248  * Writes a metadata buffer to a given disk block.  The actual IO is not
249  * performed but a new buffer_head is constructed which labels the data
250  * to be written with the correct destination disk block.
251  *
252  * Any magic-number escaping which needs to be done will cause a
253  * copy-out here.  If the buffer happens to start with the
254  * JBD2_MAGIC_NUMBER, then we can't write it to the log directly: the
255  * magic number is only written to the log for descripter blocks.  In
256  * this case, we copy the data and replace the first word with 0, and we
257  * return a result code which indicates that this buffer needs to be
258  * marked as an escaped buffer in the corresponding log descriptor
259  * block.  The missing word can then be restored when the block is read
260  * during recovery.
261  *
262  * If the source buffer has already been modified by a new transaction
263  * since we took the last commit snapshot, we use the frozen copy of
264  * that data for IO.  If we end up using the existing buffer_head's data
265  * for the write, then we *have* to lock the buffer to prevent anyone
266  * else from using and possibly modifying it while the IO is in
267  * progress.
268  *
269  * The function returns a pointer to the buffer_heads to be used for IO.
270  *
271  * We assume that the journal has already been locked in this function.
272  *
273  * Return value:
274  *  <0: Error
275  * >=0: Finished OK
276  *
277  * On success:
278  * Bit 0 set == escape performed on the data
279  * Bit 1 set == buffer copy-out performed (kfree the data after IO)
280  */
281
282 int jbd2_journal_write_metadata_buffer(transaction_t *transaction,
283                                   struct journal_head  *jh_in,
284                                   struct journal_head **jh_out,
285                                   unsigned long long blocknr)
286 {
287         int need_copy_out = 0;
288         int done_copy_out = 0;
289         int do_escape = 0;
290         char *mapped_data;
291         struct buffer_head *new_bh;
292         struct journal_head *new_jh;
293         struct page *new_page;
294         unsigned int new_offset;
295         struct buffer_head *bh_in = jh2bh(jh_in);
296         struct jbd2_buffer_trigger_type *triggers;
297         journal_t *journal = transaction->t_journal;
298
299         /*
300          * The buffer really shouldn't be locked: only the current committing
301          * transaction is allowed to write it, so nobody else is allowed
302          * to do any IO.
303          *
304          * akpm: except if we're journalling data, and write() output is
305          * also part of a shared mapping, and another thread has
306          * decided to launch a writepage() against this buffer.
307          */
308         J_ASSERT_BH(bh_in, buffer_jbddirty(bh_in));
309
310         new_bh = alloc_buffer_head(GFP_NOFS|__GFP_NOFAIL);
311         /* keep subsequent assertions sane */
312         new_bh->b_state = 0;
313         init_buffer(new_bh, NULL, NULL);
314         atomic_set(&new_bh->b_count, 1);
315         new_jh = jbd2_journal_add_journal_head(new_bh); /* This sleeps */
316
317         /*
318          * If a new transaction has already done a buffer copy-out, then
319          * we use that version of the data for the commit.
320          */
321         jbd_lock_bh_state(bh_in);
322 repeat:
323         if (jh_in->b_frozen_data) {
324                 done_copy_out = 1;
325                 new_page = virt_to_page(jh_in->b_frozen_data);
326                 new_offset = offset_in_page(jh_in->b_frozen_data);
327                 triggers = jh_in->b_frozen_triggers;
328         } else {
329                 new_page = jh2bh(jh_in)->b_page;
330                 new_offset = offset_in_page(jh2bh(jh_in)->b_data);
331                 triggers = jh_in->b_triggers;
332         }
333
334         mapped_data = kmap_atomic(new_page, KM_USER0);
335         /*
336          * Fire any commit trigger.  Do this before checking for escaping,
337          * as the trigger may modify the magic offset.  If a copy-out
338          * happens afterwards, it will have the correct data in the buffer.
339          */
340         jbd2_buffer_commit_trigger(jh_in, mapped_data + new_offset,
341                                    triggers);
342
343         /*
344          * Check for escaping
345          */
346         if (*((__be32 *)(mapped_data + new_offset)) ==
347                                 cpu_to_be32(JBD2_MAGIC_NUMBER)) {
348                 need_copy_out = 1;
349                 do_escape = 1;
350         }
351         kunmap_atomic(mapped_data, KM_USER0);
352
353         /*
354          * Do we need to do a data copy?
355          */
356         if (need_copy_out && !done_copy_out) {
357                 char *tmp;
358
359                 jbd_unlock_bh_state(bh_in);
360                 tmp = jbd2_alloc(bh_in->b_size, GFP_NOFS);
361                 jbd_lock_bh_state(bh_in);
362                 if (jh_in->b_frozen_data) {
363                         jbd2_free(tmp, bh_in->b_size);
364                         goto repeat;
365                 }
366
367                 jh_in->b_frozen_data = tmp;
368                 mapped_data = kmap_atomic(new_page, KM_USER0);
369                 memcpy(tmp, mapped_data + new_offset, jh2bh(jh_in)->b_size);
370                 kunmap_atomic(mapped_data, KM_USER0);
371
372                 new_page = virt_to_page(tmp);
373                 new_offset = offset_in_page(tmp);
374                 done_copy_out = 1;
375
376                 /*
377                  * This isn't strictly necessary, as we're using frozen
378                  * data for the escaping, but it keeps consistency with
379                  * b_frozen_data usage.
380                  */
381                 jh_in->b_frozen_triggers = jh_in->b_triggers;
382         }
383
384         /*
385          * Did we need to do an escaping?  Now we've done all the
386          * copying, we can finally do so.
387          */
388         if (do_escape) {
389                 mapped_data = kmap_atomic(new_page, KM_USER0);
390                 *((unsigned int *)(mapped_data + new_offset)) = 0;
391                 kunmap_atomic(mapped_data, KM_USER0);
392         }
393
394         set_bh_page(new_bh, new_page, new_offset);
395         new_jh->b_transaction = NULL;
396         new_bh->b_size = jh2bh(jh_in)->b_size;
397         new_bh->b_bdev = transaction->t_journal->j_dev;
398         new_bh->b_blocknr = blocknr;
399         set_buffer_mapped(new_bh);
400         set_buffer_dirty(new_bh);
401
402         *jh_out = new_jh;
403
404         /*
405          * The to-be-written buffer needs to get moved to the io queue,
406          * and the original buffer whose contents we are shadowing or
407          * copying is moved to the transaction's shadow queue.
408          */
409         JBUFFER_TRACE(jh_in, "file as BJ_Shadow");
410         spin_lock(&journal->j_list_lock);
411         __jbd2_journal_file_buffer(jh_in, transaction, BJ_Shadow);
412         spin_unlock(&journal->j_list_lock);
413         jbd_unlock_bh_state(bh_in);
414
415         JBUFFER_TRACE(new_jh, "file as BJ_IO");
416         jbd2_journal_file_buffer(new_jh, transaction, BJ_IO);
417
418         return do_escape | (done_copy_out << 1);
419 }
420
421 /*
422  * Allocation code for the journal file.  Manage the space left in the
423  * journal, so that we can begin checkpointing when appropriate.
424  */
425
426 /*
427  * __jbd2_log_space_left: Return the number of free blocks left in the journal.
428  *
429  * Called with the journal already locked.
430  *
431  * Called under j_state_lock
432  */
433
434 int __jbd2_log_space_left(journal_t *journal)
435 {
436         int left = journal->j_free;
437
438         assert_spin_locked(&journal->j_state_lock);
439
440         /*
441          * Be pessimistic here about the number of those free blocks which
442          * might be required for log descriptor control blocks.
443          */
444
445 #define MIN_LOG_RESERVED_BLOCKS 32 /* Allow for rounding errors */
446
447         left -= MIN_LOG_RESERVED_BLOCKS;
448
449         if (left <= 0)
450                 return 0;
451         left -= (left >> 3);
452         return left;
453 }
454
455 /*
456  * Called under j_state_lock.  Returns true if a transaction commit was started.
457  */
458 int __jbd2_log_start_commit(journal_t *journal, tid_t target)
459 {
460         /*
461          * Are we already doing a recent enough commit?
462          */
463         if (!tid_geq(journal->j_commit_request, target)) {
464                 /*
465                  * We want a new commit: OK, mark the request and wakup the
466                  * commit thread.  We do _not_ do the commit ourselves.
467                  */
468
469                 journal->j_commit_request = target;
470                 jbd_debug(1, "JBD: requesting commit %d/%d\n",
471                           journal->j_commit_request,
472                           journal->j_commit_sequence);
473                 wake_up(&journal->j_wait_commit);
474                 return 1;
475         }
476         return 0;
477 }
478
479 int jbd2_log_start_commit(journal_t *journal, tid_t tid)
480 {
481         int ret;
482
483         spin_lock(&journal->j_state_lock);
484         ret = __jbd2_log_start_commit(journal, tid);
485         spin_unlock(&journal->j_state_lock);
486         return ret;
487 }
488
489 /*
490  * Force and wait upon a commit if the calling process is not within
491  * transaction.  This is used for forcing out undo-protected data which contains
492  * bitmaps, when the fs is running out of space.
493  *
494  * We can only force the running transaction if we don't have an active handle;
495  * otherwise, we will deadlock.
496  *
497  * Returns true if a transaction was started.
498  */
499 int jbd2_journal_force_commit_nested(journal_t *journal)
500 {
501         transaction_t *transaction = NULL;
502         tid_t tid;
503
504         spin_lock(&journal->j_state_lock);
505         if (journal->j_running_transaction && !current->journal_info) {
506                 transaction = journal->j_running_transaction;
507                 __jbd2_log_start_commit(journal, transaction->t_tid);
508         } else if (journal->j_committing_transaction)
509                 transaction = journal->j_committing_transaction;
510
511         if (!transaction) {
512                 spin_unlock(&journal->j_state_lock);
513                 return 0;       /* Nothing to retry */
514         }
515
516         tid = transaction->t_tid;
517         spin_unlock(&journal->j_state_lock);
518         jbd2_log_wait_commit(journal, tid);
519         return 1;
520 }
521
522 /*
523  * Start a commit of the current running transaction (if any).  Returns true
524  * if a transaction is going to be committed (or is currently already
525  * committing), and fills its tid in at *ptid
526  */
527 int jbd2_journal_start_commit(journal_t *journal, tid_t *ptid)
528 {
529         int ret = 0;
530
531         spin_lock(&journal->j_state_lock);
532         if (journal->j_running_transaction) {
533                 tid_t tid = journal->j_running_transaction->t_tid;
534
535                 __jbd2_log_start_commit(journal, tid);
536                 /* There's a running transaction and we've just made sure
537                  * it's commit has been scheduled. */
538                 if (ptid)
539                         *ptid = tid;
540                 ret = 1;
541         } else if (journal->j_committing_transaction) {
542                 /*
543                  * If ext3_write_super() recently started a commit, then we
544                  * have to wait for completion of that transaction
545                  */
546                 if (ptid)
547                         *ptid = journal->j_committing_transaction->t_tid;
548                 ret = 1;
549         }
550         spin_unlock(&journal->j_state_lock);
551         return ret;
552 }
553
554 /*
555  * Wait for a specified commit to complete.
556  * The caller may not hold the journal lock.
557  */
558 int jbd2_log_wait_commit(journal_t *journal, tid_t tid)
559 {
560         int err = 0;
561
562 #ifdef CONFIG_JBD2_DEBUG
563         spin_lock(&journal->j_state_lock);
564         if (!tid_geq(journal->j_commit_request, tid)) {
565                 printk(KERN_EMERG
566                        "%s: error: j_commit_request=%d, tid=%d\n",
567                        __func__, journal->j_commit_request, tid);
568         }
569         spin_unlock(&journal->j_state_lock);
570 #endif
571         spin_lock(&journal->j_state_lock);
572         while (tid_gt(tid, journal->j_commit_sequence)) {
573                 jbd_debug(1, "JBD: want %d, j_commit_sequence=%d\n",
574                                   tid, journal->j_commit_sequence);
575                 wake_up(&journal->j_wait_commit);
576                 spin_unlock(&journal->j_state_lock);
577                 wait_event(journal->j_wait_done_commit,
578                                 !tid_gt(tid, journal->j_commit_sequence));
579                 spin_lock(&journal->j_state_lock);
580         }
581         spin_unlock(&journal->j_state_lock);
582
583         if (unlikely(is_journal_aborted(journal))) {
584                 printk(KERN_EMERG "journal commit I/O error\n");
585                 err = -EIO;
586         }
587         return err;
588 }
589
590 /*
591  * Log buffer allocation routines:
592  */
593
594 int jbd2_journal_next_log_block(journal_t *journal, unsigned long long *retp)
595 {
596         unsigned long blocknr;
597
598         spin_lock(&journal->j_state_lock);
599         J_ASSERT(journal->j_free > 1);
600
601         blocknr = journal->j_head;
602         journal->j_head++;
603         journal->j_free--;
604         if (journal->j_head == journal->j_last)
605                 journal->j_head = journal->j_first;
606         spin_unlock(&journal->j_state_lock);
607         return jbd2_journal_bmap(journal, blocknr, retp);
608 }
609
610 /*
611  * Conversion of logical to physical block numbers for the journal
612  *
613  * On external journals the journal blocks are identity-mapped, so
614  * this is a no-op.  If needed, we can use j_blk_offset - everything is
615  * ready.
616  */
617 int jbd2_journal_bmap(journal_t *journal, unsigned long blocknr,
618                  unsigned long long *retp)
619 {
620         int err = 0;
621         unsigned long long ret;
622
623         if (journal->j_inode) {
624                 ret = bmap(journal->j_inode, blocknr);
625                 if (ret)
626                         *retp = ret;
627                 else {
628                         printk(KERN_ALERT "%s: journal block not found "
629                                         "at offset %lu on %s\n",
630                                __func__, blocknr, journal->j_devname);
631                         err = -EIO;
632                         __journal_abort_soft(journal, err);
633                 }
634         } else {
635                 *retp = blocknr; /* +journal->j_blk_offset */
636         }
637         return err;
638 }
639
640 /*
641  * We play buffer_head aliasing tricks to write data/metadata blocks to
642  * the journal without copying their contents, but for journal
643  * descriptor blocks we do need to generate bona fide buffers.
644  *
645  * After the caller of jbd2_journal_get_descriptor_buffer() has finished modifying
646  * the buffer's contents they really should run flush_dcache_page(bh->b_page).
647  * But we don't bother doing that, so there will be coherency problems with
648  * mmaps of blockdevs which hold live JBD-controlled filesystems.
649  */
650 struct journal_head *jbd2_journal_get_descriptor_buffer(journal_t *journal)
651 {
652         struct buffer_head *bh;
653         unsigned long long blocknr;
654         int err;
655
656         err = jbd2_journal_next_log_block(journal, &blocknr);
657
658         if (err)
659                 return NULL;
660
661         bh = __getblk(journal->j_dev, blocknr, journal->j_blocksize);
662         if (!bh)
663                 return NULL;
664         lock_buffer(bh);
665         memset(bh->b_data, 0, journal->j_blocksize);
666         set_buffer_uptodate(bh);
667         unlock_buffer(bh);
668         BUFFER_TRACE(bh, "return this buffer");
669         return jbd2_journal_add_journal_head(bh);
670 }
671
672 struct jbd2_stats_proc_session {
673         journal_t *journal;
674         struct transaction_stats_s *stats;
675         int start;
676         int max;
677 };
678
679 static void *jbd2_seq_info_start(struct seq_file *seq, loff_t *pos)
680 {
681         return *pos ? NULL : SEQ_START_TOKEN;
682 }
683
684 static void *jbd2_seq_info_next(struct seq_file *seq, void *v, loff_t *pos)
685 {
686         return NULL;
687 }
688
689 static int jbd2_seq_info_show(struct seq_file *seq, void *v)
690 {
691         struct jbd2_stats_proc_session *s = seq->private;
692
693         if (v != SEQ_START_TOKEN)
694                 return 0;
695         seq_printf(seq, "%lu transaction, each up to %u blocks\n",
696                         s->stats->ts_tid,
697                         s->journal->j_max_transaction_buffers);
698         if (s->stats->ts_tid == 0)
699                 return 0;
700         seq_printf(seq, "average: \n  %ums waiting for transaction\n",
701             jiffies_to_msecs(s->stats->run.rs_wait / s->stats->ts_tid));
702         seq_printf(seq, "  %ums running transaction\n",
703             jiffies_to_msecs(s->stats->run.rs_running / s->stats->ts_tid));
704         seq_printf(seq, "  %ums transaction was being locked\n",
705             jiffies_to_msecs(s->stats->run.rs_locked / s->stats->ts_tid));
706         seq_printf(seq, "  %ums flushing data (in ordered mode)\n",
707             jiffies_to_msecs(s->stats->run.rs_flushing / s->stats->ts_tid));
708         seq_printf(seq, "  %ums logging transaction\n",
709             jiffies_to_msecs(s->stats->run.rs_logging / s->stats->ts_tid));
710         seq_printf(seq, "  %lluus average transaction commit time\n",
711                    div_u64(s->journal->j_average_commit_time, 1000));
712         seq_printf(seq, "  %lu handles per transaction\n",
713             s->stats->run.rs_handle_count / s->stats->ts_tid);
714         seq_printf(seq, "  %lu blocks per transaction\n",
715             s->stats->run.rs_blocks / s->stats->ts_tid);
716         seq_printf(seq, "  %lu logged blocks per transaction\n",
717             s->stats->run.rs_blocks_logged / s->stats->ts_tid);
718         return 0;
719 }
720
721 static void jbd2_seq_info_stop(struct seq_file *seq, void *v)
722 {
723 }
724
725 static const struct seq_operations jbd2_seq_info_ops = {
726         .start  = jbd2_seq_info_start,
727         .next   = jbd2_seq_info_next,
728         .stop   = jbd2_seq_info_stop,
729         .show   = jbd2_seq_info_show,
730 };
731
732 static int jbd2_seq_info_open(struct inode *inode, struct file *file)
733 {
734         journal_t *journal = PDE(inode)->data;
735         struct jbd2_stats_proc_session *s;
736         int rc, size;
737
738         s = kmalloc(sizeof(*s), GFP_KERNEL);
739         if (s == NULL)
740                 return -ENOMEM;
741         size = sizeof(struct transaction_stats_s);
742         s->stats = kmalloc(size, GFP_KERNEL);
743         if (s->stats == NULL) {
744                 kfree(s);
745                 return -ENOMEM;
746         }
747         spin_lock(&journal->j_history_lock);
748         memcpy(s->stats, &journal->j_stats, size);
749         s->journal = journal;
750         spin_unlock(&journal->j_history_lock);
751
752         rc = seq_open(file, &jbd2_seq_info_ops);
753         if (rc == 0) {
754                 struct seq_file *m = file->private_data;
755                 m->private = s;
756         } else {
757                 kfree(s->stats);
758                 kfree(s);
759         }
760         return rc;
761
762 }
763
764 static int jbd2_seq_info_release(struct inode *inode, struct file *file)
765 {
766         struct seq_file *seq = file->private_data;
767         struct jbd2_stats_proc_session *s = seq->private;
768         kfree(s->stats);
769         kfree(s);
770         return seq_release(inode, file);
771 }
772
773 static const struct file_operations jbd2_seq_info_fops = {
774         .owner          = THIS_MODULE,
775         .open           = jbd2_seq_info_open,
776         .read           = seq_read,
777         .llseek         = seq_lseek,
778         .release        = jbd2_seq_info_release,
779 };
780
781 static struct proc_dir_entry *proc_jbd2_stats;
782
783 static void jbd2_stats_proc_init(journal_t *journal)
784 {
785         journal->j_proc_entry = proc_mkdir(journal->j_devname, proc_jbd2_stats);
786         if (journal->j_proc_entry) {
787                 proc_create_data("info", S_IRUGO, journal->j_proc_entry,
788                                  &jbd2_seq_info_fops, journal);
789         }
790 }
791
792 static void jbd2_stats_proc_exit(journal_t *journal)
793 {
794         remove_proc_entry("info", journal->j_proc_entry);
795         remove_proc_entry(journal->j_devname, proc_jbd2_stats);
796 }
797
798 /*
799  * Management for journal control blocks: functions to create and
800  * destroy journal_t structures, and to initialise and read existing
801  * journal blocks from disk.  */
802
803 /* First: create and setup a journal_t object in memory.  We initialise
804  * very few fields yet: that has to wait until we have created the
805  * journal structures from from scratch, or loaded them from disk. */
806
807 static journal_t * journal_init_common (void)
808 {
809         journal_t *journal;
810         int err;
811
812         journal = kzalloc(sizeof(*journal), GFP_KERNEL|__GFP_NOFAIL);
813         if (!journal)
814                 goto fail;
815
816         init_waitqueue_head(&journal->j_wait_transaction_locked);
817         init_waitqueue_head(&journal->j_wait_logspace);
818         init_waitqueue_head(&journal->j_wait_done_commit);
819         init_waitqueue_head(&journal->j_wait_checkpoint);
820         init_waitqueue_head(&journal->j_wait_commit);
821         init_waitqueue_head(&journal->j_wait_updates);
822         mutex_init(&journal->j_barrier);
823         mutex_init(&journal->j_checkpoint_mutex);
824         spin_lock_init(&journal->j_revoke_lock);
825         spin_lock_init(&journal->j_list_lock);
826         spin_lock_init(&journal->j_state_lock);
827
828         journal->j_commit_interval = (HZ * JBD2_DEFAULT_MAX_COMMIT_AGE);
829         journal->j_min_batch_time = 0;
830         journal->j_max_batch_time = 15000; /* 15ms */
831
832         /* The journal is marked for error until we succeed with recovery! */
833         journal->j_flags = JBD2_ABORT;
834
835         /* Set up a default-sized revoke table for the new mount. */
836         err = jbd2_journal_init_revoke(journal, JOURNAL_REVOKE_DEFAULT_HASH);
837         if (err) {
838                 kfree(journal);
839                 goto fail;
840         }
841
842         spin_lock_init(&journal->j_history_lock);
843
844         return journal;
845 fail:
846         return NULL;
847 }
848
849 /* jbd2_journal_init_dev and jbd2_journal_init_inode:
850  *
851  * Create a journal structure assigned some fixed set of disk blocks to
852  * the journal.  We don't actually touch those disk blocks yet, but we
853  * need to set up all of the mapping information to tell the journaling
854  * system where the journal blocks are.
855  *
856  */
857
858 /**
859  *  journal_t * jbd2_journal_init_dev() - creates and initialises a journal structure
860  *  @bdev: Block device on which to create the journal
861  *  @fs_dev: Device which hold journalled filesystem for this journal.
862  *  @start: Block nr Start of journal.
863  *  @len:  Length of the journal in blocks.
864  *  @blocksize: blocksize of journalling device
865  *
866  *  Returns: a newly created journal_t *
867  *
868  *  jbd2_journal_init_dev creates a journal which maps a fixed contiguous
869  *  range of blocks on an arbitrary block device.
870  *
871  */
872 journal_t * jbd2_journal_init_dev(struct block_device *bdev,
873                         struct block_device *fs_dev,
874                         unsigned long long start, int len, int blocksize)
875 {
876         journal_t *journal = journal_init_common();
877         struct buffer_head *bh;
878         char *p;
879         int n;
880
881         if (!journal)
882                 return NULL;
883
884         /* journal descriptor can store up to n blocks -bzzz */
885         journal->j_blocksize = blocksize;
886         jbd2_stats_proc_init(journal);
887         n = journal->j_blocksize / sizeof(journal_block_tag_t);
888         journal->j_wbufsize = n;
889         journal->j_wbuf = kmalloc(n * sizeof(struct buffer_head*), GFP_KERNEL);
890         if (!journal->j_wbuf) {
891                 printk(KERN_ERR "%s: Cant allocate bhs for commit thread\n",
892                         __func__);
893                 goto out_err;
894         }
895         journal->j_dev = bdev;
896         journal->j_fs_dev = fs_dev;
897         journal->j_blk_offset = start;
898         journal->j_maxlen = len;
899         bdevname(journal->j_dev, journal->j_devname);
900         p = journal->j_devname;
901         while ((p = strchr(p, '/')))
902                 *p = '!';
903
904         bh = __getblk(journal->j_dev, start, journal->j_blocksize);
905         if (!bh) {
906                 printk(KERN_ERR
907                        "%s: Cannot get buffer for journal superblock\n",
908                        __func__);
909                 goto out_err;
910         }
911         journal->j_sb_buffer = bh;
912         journal->j_superblock = (journal_superblock_t *)bh->b_data;
913
914         return journal;
915 out_err:
916         kfree(journal->j_wbuf);
917         jbd2_stats_proc_exit(journal);
918         kfree(journal);
919         return NULL;
920 }
921
922 /**
923  *  journal_t * jbd2_journal_init_inode () - creates a journal which maps to a inode.
924  *  @inode: An inode to create the journal in
925  *
926  * jbd2_journal_init_inode creates a journal which maps an on-disk inode as
927  * the journal.  The inode must exist already, must support bmap() and
928  * must have all data blocks preallocated.
929  */
930 journal_t * jbd2_journal_init_inode (struct inode *inode)
931 {
932         struct buffer_head *bh;
933         journal_t *journal = journal_init_common();
934         char *p;
935         int err;
936         int n;
937         unsigned long long blocknr;
938
939         if (!journal)
940                 return NULL;
941
942         journal->j_dev = journal->j_fs_dev = inode->i_sb->s_bdev;
943         journal->j_inode = inode;
944         bdevname(journal->j_dev, journal->j_devname);
945         p = journal->j_devname;
946         while ((p = strchr(p, '/')))
947                 *p = '!';
948         p = journal->j_devname + strlen(journal->j_devname);
949         sprintf(p, "-%lu", journal->j_inode->i_ino);
950         jbd_debug(1,
951                   "journal %p: inode %s/%ld, size %Ld, bits %d, blksize %ld\n",
952                   journal, inode->i_sb->s_id, inode->i_ino,
953                   (long long) inode->i_size,
954                   inode->i_sb->s_blocksize_bits, inode->i_sb->s_blocksize);
955
956         journal->j_maxlen = inode->i_size >> inode->i_sb->s_blocksize_bits;
957         journal->j_blocksize = inode->i_sb->s_blocksize;
958         jbd2_stats_proc_init(journal);
959
960         /* journal descriptor can store up to n blocks -bzzz */
961         n = journal->j_blocksize / sizeof(journal_block_tag_t);
962         journal->j_wbufsize = n;
963         journal->j_wbuf = kmalloc(n * sizeof(struct buffer_head*), GFP_KERNEL);
964         if (!journal->j_wbuf) {
965                 printk(KERN_ERR "%s: Cant allocate bhs for commit thread\n",
966                         __func__);
967                 goto out_err;
968         }
969
970         err = jbd2_journal_bmap(journal, 0, &blocknr);
971         /* If that failed, give up */
972         if (err) {
973                 printk(KERN_ERR "%s: Cannnot locate journal superblock\n",
974                        __func__);
975                 goto out_err;
976         }
977
978         bh = __getblk(journal->j_dev, blocknr, journal->j_blocksize);
979         if (!bh) {
980                 printk(KERN_ERR
981                        "%s: Cannot get buffer for journal superblock\n",
982                        __func__);
983                 goto out_err;
984         }
985         journal->j_sb_buffer = bh;
986         journal->j_superblock = (journal_superblock_t *)bh->b_data;
987
988         return journal;
989 out_err:
990         kfree(journal->j_wbuf);
991         jbd2_stats_proc_exit(journal);
992         kfree(journal);
993         return NULL;
994 }
995
996 /*
997  * If the journal init or create aborts, we need to mark the journal
998  * superblock as being NULL to prevent the journal destroy from writing
999  * back a bogus superblock.
1000  */
1001 static void journal_fail_superblock (journal_t *journal)
1002 {
1003         struct buffer_head *bh = journal->j_sb_buffer;
1004         brelse(bh);
1005         journal->j_sb_buffer = NULL;
1006 }
1007
1008 /*
1009  * Given a journal_t structure, initialise the various fields for
1010  * startup of a new journaling session.  We use this both when creating
1011  * a journal, and after recovering an old journal to reset it for
1012  * subsequent use.
1013  */
1014
1015 static int journal_reset(journal_t *journal)
1016 {
1017         journal_superblock_t *sb = journal->j_superblock;
1018         unsigned long long first, last;
1019
1020         first = be32_to_cpu(sb->s_first);
1021         last = be32_to_cpu(sb->s_maxlen);
1022         if (first + JBD2_MIN_JOURNAL_BLOCKS > last + 1) {
1023                 printk(KERN_ERR "JBD: Journal too short (blocks %llu-%llu).\n",
1024                        first, last);
1025                 journal_fail_superblock(journal);
1026                 return -EINVAL;
1027         }
1028
1029         journal->j_first = first;
1030         journal->j_last = last;
1031
1032         journal->j_head = first;
1033         journal->j_tail = first;
1034         journal->j_free = last - first;
1035
1036         journal->j_tail_sequence = journal->j_transaction_sequence;
1037         journal->j_commit_sequence = journal->j_transaction_sequence - 1;
1038         journal->j_commit_request = journal->j_commit_sequence;
1039
1040         journal->j_max_transaction_buffers = journal->j_maxlen / 4;
1041
1042         /* Add the dynamic fields and write it to disk. */
1043         jbd2_journal_update_superblock(journal, 1);
1044         return jbd2_journal_start_thread(journal);
1045 }
1046
1047 /**
1048  * void jbd2_journal_update_superblock() - Update journal sb on disk.
1049  * @journal: The journal to update.
1050  * @wait: Set to '0' if you don't want to wait for IO completion.
1051  *
1052  * Update a journal's dynamic superblock fields and write it to disk,
1053  * optionally waiting for the IO to complete.
1054  */
1055 void jbd2_journal_update_superblock(journal_t *journal, int wait)
1056 {
1057         journal_superblock_t *sb = journal->j_superblock;
1058         struct buffer_head *bh = journal->j_sb_buffer;
1059
1060         /*
1061          * As a special case, if the on-disk copy is already marked as needing
1062          * no recovery (s_start == 0) and there are no outstanding transactions
1063          * in the filesystem, then we can safely defer the superblock update
1064          * until the next commit by setting JBD2_FLUSHED.  This avoids
1065          * attempting a write to a potential-readonly device.
1066          */
1067         if (sb->s_start == 0 && journal->j_tail_sequence ==
1068                                 journal->j_transaction_sequence) {
1069                 jbd_debug(1,"JBD: Skipping superblock update on recovered sb "
1070                         "(start %ld, seq %d, errno %d)\n",
1071                         journal->j_tail, journal->j_tail_sequence,
1072                         journal->j_errno);
1073                 goto out;
1074         }
1075
1076         if (buffer_write_io_error(bh)) {
1077                 /*
1078                  * Oh, dear.  A previous attempt to write the journal
1079                  * superblock failed.  This could happen because the
1080                  * USB device was yanked out.  Or it could happen to
1081                  * be a transient write error and maybe the block will
1082                  * be remapped.  Nothing we can do but to retry the
1083                  * write and hope for the best.
1084                  */
1085                 printk(KERN_ERR "JBD2: previous I/O error detected "
1086                        "for journal superblock update for %s.\n",
1087                        journal->j_devname);
1088                 clear_buffer_write_io_error(bh);
1089                 set_buffer_uptodate(bh);
1090         }
1091
1092         spin_lock(&journal->j_state_lock);
1093         jbd_debug(1,"JBD: updating superblock (start %ld, seq %d, errno %d)\n",
1094                   journal->j_tail, journal->j_tail_sequence, journal->j_errno);
1095
1096         sb->s_sequence = cpu_to_be32(journal->j_tail_sequence);
1097         sb->s_start    = cpu_to_be32(journal->j_tail);
1098         sb->s_errno    = cpu_to_be32(journal->j_errno);
1099         spin_unlock(&journal->j_state_lock);
1100
1101         BUFFER_TRACE(bh, "marking dirty");
1102         mark_buffer_dirty(bh);
1103         if (wait) {
1104                 sync_dirty_buffer(bh);
1105                 if (buffer_write_io_error(bh)) {
1106                         printk(KERN_ERR "JBD2: I/O error detected "
1107                                "when updating journal superblock for %s.\n",
1108                                journal->j_devname);
1109                         clear_buffer_write_io_error(bh);
1110                         set_buffer_uptodate(bh);
1111                 }
1112         } else
1113                 ll_rw_block(SWRITE, 1, &bh);
1114
1115 out:
1116         /* If we have just flushed the log (by marking s_start==0), then
1117          * any future commit will have to be careful to update the
1118          * superblock again to re-record the true start of the log. */
1119
1120         spin_lock(&journal->j_state_lock);
1121         if (sb->s_start)
1122                 journal->j_flags &= ~JBD2_FLUSHED;
1123         else
1124                 journal->j_flags |= JBD2_FLUSHED;
1125         spin_unlock(&journal->j_state_lock);
1126 }
1127
1128 /*
1129  * Read the superblock for a given journal, performing initial
1130  * validation of the format.
1131  */
1132
1133 static int journal_get_superblock(journal_t *journal)
1134 {
1135         struct buffer_head *bh;
1136         journal_superblock_t *sb;
1137         int err = -EIO;
1138
1139         bh = journal->j_sb_buffer;
1140
1141         J_ASSERT(bh != NULL);
1142         if (!buffer_uptodate(bh)) {
1143                 ll_rw_block(READ, 1, &bh);
1144                 wait_on_buffer(bh);
1145                 if (!buffer_uptodate(bh)) {
1146                         printk (KERN_ERR
1147                                 "JBD: IO error reading journal superblock\n");
1148                         goto out;
1149                 }
1150         }
1151
1152         sb = journal->j_superblock;
1153
1154         err = -EINVAL;
1155
1156         if (sb->s_header.h_magic != cpu_to_be32(JBD2_MAGIC_NUMBER) ||
1157             sb->s_blocksize != cpu_to_be32(journal->j_blocksize)) {
1158                 printk(KERN_WARNING "JBD: no valid journal superblock found\n");
1159                 goto out;
1160         }
1161
1162         switch(be32_to_cpu(sb->s_header.h_blocktype)) {
1163         case JBD2_SUPERBLOCK_V1:
1164                 journal->j_format_version = 1;
1165                 break;
1166         case JBD2_SUPERBLOCK_V2:
1167                 journal->j_format_version = 2;
1168                 break;
1169         default:
1170                 printk(KERN_WARNING "JBD: unrecognised superblock format ID\n");
1171                 goto out;
1172         }
1173
1174         if (be32_to_cpu(sb->s_maxlen) < journal->j_maxlen)
1175                 journal->j_maxlen = be32_to_cpu(sb->s_maxlen);
1176         else if (be32_to_cpu(sb->s_maxlen) > journal->j_maxlen) {
1177                 printk (KERN_WARNING "JBD: journal file too short\n");
1178                 goto out;
1179         }
1180
1181         return 0;
1182
1183 out:
1184         journal_fail_superblock(journal);
1185         return err;
1186 }
1187
1188 /*
1189  * Load the on-disk journal superblock and read the key fields into the
1190  * journal_t.
1191  */
1192
1193 static int load_superblock(journal_t *journal)
1194 {
1195         int err;
1196         journal_superblock_t *sb;
1197
1198         err = journal_get_superblock(journal);
1199         if (err)
1200                 return err;
1201
1202         sb = journal->j_superblock;
1203
1204         journal->j_tail_sequence = be32_to_cpu(sb->s_sequence);
1205         journal->j_tail = be32_to_cpu(sb->s_start);
1206         journal->j_first = be32_to_cpu(sb->s_first);
1207         journal->j_last = be32_to_cpu(sb->s_maxlen);
1208         journal->j_errno = be32_to_cpu(sb->s_errno);
1209
1210         return 0;
1211 }
1212
1213
1214 /**
1215  * int jbd2_journal_load() - Read journal from disk.
1216  * @journal: Journal to act on.
1217  *
1218  * Given a journal_t structure which tells us which disk blocks contain
1219  * a journal, read the journal from disk to initialise the in-memory
1220  * structures.
1221  */
1222 int jbd2_journal_load(journal_t *journal)
1223 {
1224         int err;
1225         journal_superblock_t *sb;
1226
1227         err = load_superblock(journal);
1228         if (err)
1229                 return err;
1230
1231         sb = journal->j_superblock;
1232         /* If this is a V2 superblock, then we have to check the
1233          * features flags on it. */
1234
1235         if (journal->j_format_version >= 2) {
1236                 if ((sb->s_feature_ro_compat &
1237                      ~cpu_to_be32(JBD2_KNOWN_ROCOMPAT_FEATURES)) ||
1238                     (sb->s_feature_incompat &
1239                      ~cpu_to_be32(JBD2_KNOWN_INCOMPAT_FEATURES))) {
1240                         printk (KERN_WARNING
1241                                 "JBD: Unrecognised features on journal\n");
1242                         return -EINVAL;
1243                 }
1244         }
1245
1246         /* Let the recovery code check whether it needs to recover any
1247          * data from the journal. */
1248         if (jbd2_journal_recover(journal))
1249                 goto recovery_error;
1250
1251         if (journal->j_failed_commit) {
1252                 printk(KERN_ERR "JBD2: journal transaction %u on %s "
1253                        "is corrupt.\n", journal->j_failed_commit,
1254                        journal->j_devname);
1255                 return -EIO;
1256         }
1257
1258         /* OK, we've finished with the dynamic journal bits:
1259          * reinitialise the dynamic contents of the superblock in memory
1260          * and reset them on disk. */
1261         if (journal_reset(journal))
1262                 goto recovery_error;
1263
1264         journal->j_flags &= ~JBD2_ABORT;
1265         journal->j_flags |= JBD2_LOADED;
1266         return 0;
1267
1268 recovery_error:
1269         printk (KERN_WARNING "JBD: recovery failed\n");
1270         return -EIO;
1271 }
1272
1273 /**
1274  * void jbd2_journal_destroy() - Release a journal_t structure.
1275  * @journal: Journal to act on.
1276  *
1277  * Release a journal_t structure once it is no longer in use by the
1278  * journaled object.
1279  * Return <0 if we couldn't clean up the journal.
1280  */
1281 int jbd2_journal_destroy(journal_t *journal)
1282 {
1283         int err = 0;
1284
1285         /* Wait for the commit thread to wake up and die. */
1286         journal_kill_thread(journal);
1287
1288         /* Force a final log commit */
1289         if (journal->j_running_transaction)
1290                 jbd2_journal_commit_transaction(journal);
1291
1292         /* Force any old transactions to disk */
1293
1294         /* Totally anal locking here... */
1295         spin_lock(&journal->j_list_lock);
1296         while (journal->j_checkpoint_transactions != NULL) {
1297                 spin_unlock(&journal->j_list_lock);
1298                 mutex_lock(&journal->j_checkpoint_mutex);
1299                 jbd2_log_do_checkpoint(journal);
1300                 mutex_unlock(&journal->j_checkpoint_mutex);
1301                 spin_lock(&journal->j_list_lock);
1302         }
1303
1304         J_ASSERT(journal->j_running_transaction == NULL);
1305         J_ASSERT(journal->j_committing_transaction == NULL);
1306         J_ASSERT(journal->j_checkpoint_transactions == NULL);
1307         spin_unlock(&journal->j_list_lock);
1308
1309         if (journal->j_sb_buffer) {
1310                 if (!is_journal_aborted(journal)) {
1311                         /* We can now mark the journal as empty. */
1312                         journal->j_tail = 0;
1313                         journal->j_tail_sequence =
1314                                 ++journal->j_transaction_sequence;
1315                         jbd2_journal_update_superblock(journal, 1);
1316                 } else {
1317                         err = -EIO;
1318                 }
1319                 brelse(journal->j_sb_buffer);
1320         }
1321
1322         if (journal->j_proc_entry)
1323                 jbd2_stats_proc_exit(journal);
1324         if (journal->j_inode)
1325                 iput(journal->j_inode);
1326         if (journal->j_revoke)
1327                 jbd2_journal_destroy_revoke(journal);
1328         kfree(journal->j_wbuf);
1329         kfree(journal);
1330
1331         return err;
1332 }
1333
1334
1335 /**
1336  *int jbd2_journal_check_used_features () - Check if features specified are used.
1337  * @journal: Journal to check.
1338  * @compat: bitmask of compatible features
1339  * @ro: bitmask of features that force read-only mount
1340  * @incompat: bitmask of incompatible features
1341  *
1342  * Check whether the journal uses all of a given set of
1343  * features.  Return true (non-zero) if it does.
1344  **/
1345
1346 int jbd2_journal_check_used_features (journal_t *journal, unsigned long compat,
1347                                  unsigned long ro, unsigned long incompat)
1348 {
1349         journal_superblock_t *sb;
1350
1351         if (!compat && !ro && !incompat)
1352                 return 1;
1353         if (journal->j_format_version == 1)
1354                 return 0;
1355
1356         sb = journal->j_superblock;
1357
1358         if (((be32_to_cpu(sb->s_feature_compat) & compat) == compat) &&
1359             ((be32_to_cpu(sb->s_feature_ro_compat) & ro) == ro) &&
1360             ((be32_to_cpu(sb->s_feature_incompat) & incompat) == incompat))
1361                 return 1;
1362
1363         return 0;
1364 }
1365
1366 /**
1367  * int jbd2_journal_check_available_features() - Check feature set in journalling layer
1368  * @journal: Journal to check.
1369  * @compat: bitmask of compatible features
1370  * @ro: bitmask of features that force read-only mount
1371  * @incompat: bitmask of incompatible features
1372  *
1373  * Check whether the journaling code supports the use of
1374  * all of a given set of features on this journal.  Return true
1375  * (non-zero) if it can. */
1376
1377 int jbd2_journal_check_available_features (journal_t *journal, unsigned long compat,
1378                                       unsigned long ro, unsigned long incompat)
1379 {
1380         journal_superblock_t *sb;
1381
1382         if (!compat && !ro && !incompat)
1383                 return 1;
1384
1385         sb = journal->j_superblock;
1386
1387         /* We can support any known requested features iff the
1388          * superblock is in version 2.  Otherwise we fail to support any
1389          * extended sb features. */
1390
1391         if (journal->j_format_version != 2)
1392                 return 0;
1393
1394         if ((compat   & JBD2_KNOWN_COMPAT_FEATURES) == compat &&
1395             (ro       & JBD2_KNOWN_ROCOMPAT_FEATURES) == ro &&
1396             (incompat & JBD2_KNOWN_INCOMPAT_FEATURES) == incompat)
1397                 return 1;
1398
1399         return 0;
1400 }
1401
1402 /**
1403  * int jbd2_journal_set_features () - Mark a given journal feature in the superblock
1404  * @journal: Journal to act on.
1405  * @compat: bitmask of compatible features
1406  * @ro: bitmask of features that force read-only mount
1407  * @incompat: bitmask of incompatible features
1408  *
1409  * Mark a given journal feature as present on the
1410  * superblock.  Returns true if the requested features could be set.
1411  *
1412  */
1413
1414 int jbd2_journal_set_features (journal_t *journal, unsigned long compat,
1415                           unsigned long ro, unsigned long incompat)
1416 {
1417         journal_superblock_t *sb;
1418
1419         if (jbd2_journal_check_used_features(journal, compat, ro, incompat))
1420                 return 1;
1421
1422         if (!jbd2_journal_check_available_features(journal, compat, ro, incompat))
1423                 return 0;
1424
1425         jbd_debug(1, "Setting new features 0x%lx/0x%lx/0x%lx\n",
1426                   compat, ro, incompat);
1427
1428         sb = journal->j_superblock;
1429
1430         sb->s_feature_compat    |= cpu_to_be32(compat);
1431         sb->s_feature_ro_compat |= cpu_to_be32(ro);
1432         sb->s_feature_incompat  |= cpu_to_be32(incompat);
1433
1434         return 1;
1435 }
1436
1437 /*
1438  * jbd2_journal_clear_features () - Clear a given journal feature in the
1439  *                                  superblock
1440  * @journal: Journal to act on.
1441  * @compat: bitmask of compatible features
1442  * @ro: bitmask of features that force read-only mount
1443  * @incompat: bitmask of incompatible features
1444  *
1445  * Clear a given journal feature as present on the
1446  * superblock.
1447  */
1448 void jbd2_journal_clear_features(journal_t *journal, unsigned long compat,
1449                                 unsigned long ro, unsigned long incompat)
1450 {
1451         journal_superblock_t *sb;
1452
1453         jbd_debug(1, "Clear features 0x%lx/0x%lx/0x%lx\n",
1454                   compat, ro, incompat);
1455
1456         sb = journal->j_superblock;
1457
1458         sb->s_feature_compat    &= ~cpu_to_be32(compat);
1459         sb->s_feature_ro_compat &= ~cpu_to_be32(ro);
1460         sb->s_feature_incompat  &= ~cpu_to_be32(incompat);
1461 }
1462 EXPORT_SYMBOL(jbd2_journal_clear_features);
1463
1464 /**
1465  * int jbd2_journal_update_format () - Update on-disk journal structure.
1466  * @journal: Journal to act on.
1467  *
1468  * Given an initialised but unloaded journal struct, poke about in the
1469  * on-disk structure to update it to the most recent supported version.
1470  */
1471 int jbd2_journal_update_format (journal_t *journal)
1472 {
1473         journal_superblock_t *sb;
1474         int err;
1475
1476         err = journal_get_superblock(journal);
1477         if (err)
1478                 return err;
1479
1480         sb = journal->j_superblock;
1481
1482         switch (be32_to_cpu(sb->s_header.h_blocktype)) {
1483         case JBD2_SUPERBLOCK_V2:
1484                 return 0;
1485         case JBD2_SUPERBLOCK_V1:
1486                 return journal_convert_superblock_v1(journal, sb);
1487         default:
1488                 break;
1489         }
1490         return -EINVAL;
1491 }
1492
1493 static int journal_convert_superblock_v1(journal_t *journal,
1494                                          journal_superblock_t *sb)
1495 {
1496         int offset, blocksize;
1497         struct buffer_head *bh;
1498
1499         printk(KERN_WARNING
1500                 "JBD: Converting superblock from version 1 to 2.\n");
1501
1502         /* Pre-initialise new fields to zero */
1503         offset = ((char *) &(sb->s_feature_compat)) - ((char *) sb);
1504         blocksize = be32_to_cpu(sb->s_blocksize);
1505         memset(&sb->s_feature_compat, 0, blocksize-offset);
1506
1507         sb->s_nr_users = cpu_to_be32(1);
1508         sb->s_header.h_blocktype = cpu_to_be32(JBD2_SUPERBLOCK_V2);
1509         journal->j_format_version = 2;
1510
1511         bh = journal->j_sb_buffer;
1512         BUFFER_TRACE(bh, "marking dirty");
1513         mark_buffer_dirty(bh);
1514         sync_dirty_buffer(bh);
1515         return 0;
1516 }
1517
1518
1519 /**
1520  * int jbd2_journal_flush () - Flush journal
1521  * @journal: Journal to act on.
1522  *
1523  * Flush all data for a given journal to disk and empty the journal.
1524  * Filesystems can use this when remounting readonly to ensure that
1525  * recovery does not need to happen on remount.
1526  */
1527
1528 int jbd2_journal_flush(journal_t *journal)
1529 {
1530         int err = 0;
1531         transaction_t *transaction = NULL;
1532         unsigned long old_tail;
1533
1534         spin_lock(&journal->j_state_lock);
1535
1536         /* Force everything buffered to the log... */
1537         if (journal->j_running_transaction) {
1538                 transaction = journal->j_running_transaction;
1539                 __jbd2_log_start_commit(journal, transaction->t_tid);
1540         } else if (journal->j_committing_transaction)
1541                 transaction = journal->j_committing_transaction;
1542
1543         /* Wait for the log commit to complete... */
1544         if (transaction) {
1545                 tid_t tid = transaction->t_tid;
1546
1547                 spin_unlock(&journal->j_state_lock);
1548                 jbd2_log_wait_commit(journal, tid);
1549         } else {
1550                 spin_unlock(&journal->j_state_lock);
1551         }
1552
1553         /* ...and flush everything in the log out to disk. */
1554         spin_lock(&journal->j_list_lock);
1555         while (!err && journal->j_checkpoint_transactions != NULL) {
1556                 spin_unlock(&journal->j_list_lock);
1557                 mutex_lock(&journal->j_checkpoint_mutex);
1558                 err = jbd2_log_do_checkpoint(journal);
1559                 mutex_unlock(&journal->j_checkpoint_mutex);
1560                 spin_lock(&journal->j_list_lock);
1561         }
1562         spin_unlock(&journal->j_list_lock);
1563
1564         if (is_journal_aborted(journal))
1565                 return -EIO;
1566
1567         jbd2_cleanup_journal_tail(journal);
1568
1569         /* Finally, mark the journal as really needing no recovery.
1570          * This sets s_start==0 in the underlying superblock, which is
1571          * the magic code for a fully-recovered superblock.  Any future
1572          * commits of data to the journal will restore the current
1573          * s_start value. */
1574         spin_lock(&journal->j_state_lock);
1575         old_tail = journal->j_tail;
1576         journal->j_tail = 0;
1577         spin_unlock(&journal->j_state_lock);
1578         jbd2_journal_update_superblock(journal, 1);
1579         spin_lock(&journal->j_state_lock);
1580         journal->j_tail = old_tail;
1581
1582         J_ASSERT(!journal->j_running_transaction);
1583         J_ASSERT(!journal->j_committing_transaction);
1584         J_ASSERT(!journal->j_checkpoint_transactions);
1585         J_ASSERT(journal->j_head == journal->j_tail);
1586         J_ASSERT(journal->j_tail_sequence == journal->j_transaction_sequence);
1587         spin_unlock(&journal->j_state_lock);
1588         return 0;
1589 }
1590
1591 /**
1592  * int jbd2_journal_wipe() - Wipe journal contents
1593  * @journal: Journal to act on.
1594  * @write: flag (see below)
1595  *
1596  * Wipe out all of the contents of a journal, safely.  This will produce
1597  * a warning if the journal contains any valid recovery information.
1598  * Must be called between journal_init_*() and jbd2_journal_load().
1599  *
1600  * If 'write' is non-zero, then we wipe out the journal on disk; otherwise
1601  * we merely suppress recovery.
1602  */
1603
1604 int jbd2_journal_wipe(journal_t *journal, int write)
1605 {
1606         journal_superblock_t *sb;
1607         int err = 0;
1608
1609         J_ASSERT (!(journal->j_flags & JBD2_LOADED));
1610
1611         err = load_superblock(journal);
1612         if (err)
1613                 return err;
1614
1615         sb = journal->j_superblock;
1616
1617         if (!journal->j_tail)
1618                 goto no_recovery;
1619
1620         printk (KERN_WARNING "JBD: %s recovery information on journal\n",
1621                 write ? "Clearing" : "Ignoring");
1622
1623         err = jbd2_journal_skip_recovery(journal);
1624         if (write)
1625                 jbd2_journal_update_superblock(journal, 1);
1626
1627  no_recovery:
1628         return err;
1629 }
1630
1631 /*
1632  * Journal abort has very specific semantics, which we describe
1633  * for journal abort.
1634  *
1635  * Two internal functions, which provide abort to the jbd layer
1636  * itself are here.
1637  */
1638
1639 /*
1640  * Quick version for internal journal use (doesn't lock the journal).
1641  * Aborts hard --- we mark the abort as occurred, but do _nothing_ else,
1642  * and don't attempt to make any other journal updates.
1643  */
1644 void __jbd2_journal_abort_hard(journal_t *journal)
1645 {
1646         transaction_t *transaction;
1647
1648         if (journal->j_flags & JBD2_ABORT)
1649                 return;
1650
1651         printk(KERN_ERR "Aborting journal on device %s.\n",
1652                journal->j_devname);
1653
1654         spin_lock(&journal->j_state_lock);
1655         journal->j_flags |= JBD2_ABORT;
1656         transaction = journal->j_running_transaction;
1657         if (transaction)
1658                 __jbd2_log_start_commit(journal, transaction->t_tid);
1659         spin_unlock(&journal->j_state_lock);
1660 }
1661
1662 /* Soft abort: record the abort error status in the journal superblock,
1663  * but don't do any other IO. */
1664 static void __journal_abort_soft (journal_t *journal, int errno)
1665 {
1666         if (journal->j_flags & JBD2_ABORT)
1667                 return;
1668
1669         if (!journal->j_errno)
1670                 journal->j_errno = errno;
1671
1672         __jbd2_journal_abort_hard(journal);
1673
1674         if (errno)
1675                 jbd2_journal_update_superblock(journal, 1);
1676 }
1677
1678 /**
1679  * void jbd2_journal_abort () - Shutdown the journal immediately.
1680  * @journal: the journal to shutdown.
1681  * @errno:   an error number to record in the journal indicating
1682  *           the reason for the shutdown.
1683  *
1684  * Perform a complete, immediate shutdown of the ENTIRE
1685  * journal (not of a single transaction).  This operation cannot be
1686  * undone without closing and reopening the journal.
1687  *
1688  * The jbd2_journal_abort function is intended to support higher level error
1689  * recovery mechanisms such as the ext2/ext3 remount-readonly error
1690  * mode.
1691  *
1692  * Journal abort has very specific semantics.  Any existing dirty,
1693  * unjournaled buffers in the main filesystem will still be written to
1694  * disk by bdflush, but the journaling mechanism will be suspended
1695  * immediately and no further transaction commits will be honoured.
1696  *
1697  * Any dirty, journaled buffers will be written back to disk without
1698  * hitting the journal.  Atomicity cannot be guaranteed on an aborted
1699  * filesystem, but we _do_ attempt to leave as much data as possible
1700  * behind for fsck to use for cleanup.
1701  *
1702  * Any attempt to get a new transaction handle on a journal which is in
1703  * ABORT state will just result in an -EROFS error return.  A
1704  * jbd2_journal_stop on an existing handle will return -EIO if we have
1705  * entered abort state during the update.
1706  *
1707  * Recursive transactions are not disturbed by journal abort until the
1708  * final jbd2_journal_stop, which will receive the -EIO error.
1709  *
1710  * Finally, the jbd2_journal_abort call allows the caller to supply an errno
1711  * which will be recorded (if possible) in the journal superblock.  This
1712  * allows a client to record failure conditions in the middle of a
1713  * transaction without having to complete the transaction to record the
1714  * failure to disk.  ext3_error, for example, now uses this
1715  * functionality.
1716  *
1717  * Errors which originate from within the journaling layer will NOT
1718  * supply an errno; a null errno implies that absolutely no further
1719  * writes are done to the journal (unless there are any already in
1720  * progress).
1721  *
1722  */
1723
1724 void jbd2_journal_abort(journal_t *journal, int errno)
1725 {
1726         __journal_abort_soft(journal, errno);
1727 }
1728
1729 /**
1730  * int jbd2_journal_errno () - returns the journal's error state.
1731  * @journal: journal to examine.
1732  *
1733  * This is the errno number set with jbd2_journal_abort(), the last
1734  * time the journal was mounted - if the journal was stopped
1735  * without calling abort this will be 0.
1736  *
1737  * If the journal has been aborted on this mount time -EROFS will
1738  * be returned.
1739  */
1740 int jbd2_journal_errno(journal_t *journal)
1741 {
1742         int err;
1743
1744         spin_lock(&journal->j_state_lock);
1745         if (journal->j_flags & JBD2_ABORT)
1746                 err = -EROFS;
1747         else
1748                 err = journal->j_errno;
1749         spin_unlock(&journal->j_state_lock);
1750         return err;
1751 }
1752
1753 /**
1754  * int jbd2_journal_clear_err () - clears the journal's error state
1755  * @journal: journal to act on.
1756  *
1757  * An error must be cleared or acked to take a FS out of readonly
1758  * mode.
1759  */
1760 int jbd2_journal_clear_err(journal_t *journal)
1761 {
1762         int err = 0;
1763
1764         spin_lock(&journal->j_state_lock);
1765         if (journal->j_flags & JBD2_ABORT)
1766                 err = -EROFS;
1767         else
1768                 journal->j_errno = 0;
1769         spin_unlock(&journal->j_state_lock);
1770         return err;
1771 }
1772
1773 /**
1774  * void jbd2_journal_ack_err() - Ack journal err.
1775  * @journal: journal to act on.
1776  *
1777  * An error must be cleared or acked to take a FS out of readonly
1778  * mode.
1779  */
1780 void jbd2_journal_ack_err(journal_t *journal)
1781 {
1782         spin_lock(&journal->j_state_lock);
1783         if (journal->j_errno)
1784                 journal->j_flags |= JBD2_ACK_ERR;
1785         spin_unlock(&journal->j_state_lock);
1786 }
1787
1788 int jbd2_journal_blocks_per_page(struct inode *inode)
1789 {
1790         return 1 << (PAGE_CACHE_SHIFT - inode->i_sb->s_blocksize_bits);
1791 }
1792
1793 /*
1794  * helper functions to deal with 32 or 64bit block numbers.
1795  */
1796 size_t journal_tag_bytes(journal_t *journal)
1797 {
1798         if (JBD2_HAS_INCOMPAT_FEATURE(journal, JBD2_FEATURE_INCOMPAT_64BIT))
1799                 return JBD2_TAG_SIZE64;
1800         else
1801                 return JBD2_TAG_SIZE32;
1802 }
1803
1804 /*
1805  * Journal_head storage management
1806  */
1807 static struct kmem_cache *jbd2_journal_head_cache;
1808 #ifdef CONFIG_JBD2_DEBUG
1809 static atomic_t nr_journal_heads = ATOMIC_INIT(0);
1810 #endif
1811
1812 static int journal_init_jbd2_journal_head_cache(void)
1813 {
1814         int retval;
1815
1816         J_ASSERT(jbd2_journal_head_cache == NULL);
1817         jbd2_journal_head_cache = kmem_cache_create("jbd2_journal_head",
1818                                 sizeof(struct journal_head),
1819                                 0,              /* offset */
1820                                 SLAB_TEMPORARY, /* flags */
1821                                 NULL);          /* ctor */
1822         retval = 0;
1823         if (!jbd2_journal_head_cache) {
1824                 retval = -ENOMEM;
1825                 printk(KERN_EMERG "JBD: no memory for journal_head cache\n");
1826         }
1827         return retval;
1828 }
1829
1830 static void jbd2_journal_destroy_jbd2_journal_head_cache(void)
1831 {
1832         if (jbd2_journal_head_cache) {
1833                 kmem_cache_destroy(jbd2_journal_head_cache);
1834                 jbd2_journal_head_cache = NULL;
1835         }
1836 }
1837
1838 /*
1839  * journal_head splicing and dicing
1840  */
1841 static struct journal_head *journal_alloc_journal_head(void)
1842 {
1843         struct journal_head *ret;
1844         static unsigned long last_warning;
1845
1846 #ifdef CONFIG_JBD2_DEBUG
1847         atomic_inc(&nr_journal_heads);
1848 #endif
1849         ret = kmem_cache_alloc(jbd2_journal_head_cache, GFP_NOFS);
1850         if (!ret) {
1851                 jbd_debug(1, "out of memory for journal_head\n");
1852                 if (time_after(jiffies, last_warning + 5*HZ)) {
1853                         printk(KERN_NOTICE "ENOMEM in %s, retrying.\n",
1854                                __func__);
1855                         last_warning = jiffies;
1856                 }
1857                 while (!ret) {
1858                         yield();
1859                         ret = kmem_cache_alloc(jbd2_journal_head_cache, GFP_NOFS);
1860                 }
1861         }
1862         return ret;
1863 }
1864
1865 static void journal_free_journal_head(struct journal_head *jh)
1866 {
1867 #ifdef CONFIG_JBD2_DEBUG
1868         atomic_dec(&nr_journal_heads);
1869         memset(jh, JBD2_POISON_FREE, sizeof(*jh));
1870 #endif
1871         kmem_cache_free(jbd2_journal_head_cache, jh);
1872 }
1873
1874 /*
1875  * A journal_head is attached to a buffer_head whenever JBD has an
1876  * interest in the buffer.
1877  *
1878  * Whenever a buffer has an attached journal_head, its ->b_state:BH_JBD bit
1879  * is set.  This bit is tested in core kernel code where we need to take
1880  * JBD-specific actions.  Testing the zeroness of ->b_private is not reliable
1881  * there.
1882  *
1883  * When a buffer has its BH_JBD bit set, its ->b_count is elevated by one.
1884  *
1885  * When a buffer has its BH_JBD bit set it is immune from being released by
1886  * core kernel code, mainly via ->b_count.
1887  *
1888  * A journal_head may be detached from its buffer_head when the journal_head's
1889  * b_transaction, b_cp_transaction and b_next_transaction pointers are NULL.
1890  * Various places in JBD call jbd2_journal_remove_journal_head() to indicate that the
1891  * journal_head can be dropped if needed.
1892  *
1893  * Various places in the kernel want to attach a journal_head to a buffer_head
1894  * _before_ attaching the journal_head to a transaction.  To protect the
1895  * journal_head in this situation, jbd2_journal_add_journal_head elevates the
1896  * journal_head's b_jcount refcount by one.  The caller must call
1897  * jbd2_journal_put_journal_head() to undo this.
1898  *
1899  * So the typical usage would be:
1900  *
1901  *      (Attach a journal_head if needed.  Increments b_jcount)
1902  *      struct journal_head *jh = jbd2_journal_add_journal_head(bh);
1903  *      ...
1904  *      jh->b_transaction = xxx;
1905  *      jbd2_journal_put_journal_head(jh);
1906  *
1907  * Now, the journal_head's b_jcount is zero, but it is safe from being released
1908  * because it has a non-zero b_transaction.
1909  */
1910
1911 /*
1912  * Give a buffer_head a journal_head.
1913  *
1914  * Doesn't need the journal lock.
1915  * May sleep.
1916  */
1917 struct journal_head *jbd2_journal_add_journal_head(struct buffer_head *bh)
1918 {
1919         struct journal_head *jh;
1920         struct journal_head *new_jh = NULL;
1921
1922 repeat:
1923         if (!buffer_jbd(bh)) {
1924                 new_jh = journal_alloc_journal_head();
1925                 memset(new_jh, 0, sizeof(*new_jh));
1926         }
1927
1928         jbd_lock_bh_journal_head(bh);
1929         if (buffer_jbd(bh)) {
1930                 jh = bh2jh(bh);
1931         } else {
1932                 J_ASSERT_BH(bh,
1933                         (atomic_read(&bh->b_count) > 0) ||
1934                         (bh->b_page && bh->b_page->mapping));
1935
1936                 if (!new_jh) {
1937                         jbd_unlock_bh_journal_head(bh);
1938                         goto repeat;
1939                 }
1940
1941                 jh = new_jh;
1942                 new_jh = NULL;          /* We consumed it */
1943                 set_buffer_jbd(bh);
1944                 bh->b_private = jh;
1945                 jh->b_bh = bh;
1946                 get_bh(bh);
1947                 BUFFER_TRACE(bh, "added journal_head");
1948         }
1949         jh->b_jcount++;
1950         jbd_unlock_bh_journal_head(bh);
1951         if (new_jh)
1952                 journal_free_journal_head(new_jh);
1953         return bh->b_private;
1954 }
1955
1956 /*
1957  * Grab a ref against this buffer_head's journal_head.  If it ended up not
1958  * having a journal_head, return NULL
1959  */
1960 struct journal_head *jbd2_journal_grab_journal_head(struct buffer_head *bh)
1961 {
1962         struct journal_head *jh = NULL;
1963
1964         jbd_lock_bh_journal_head(bh);
1965         if (buffer_jbd(bh)) {
1966                 jh = bh2jh(bh);
1967                 jh->b_jcount++;
1968         }
1969         jbd_unlock_bh_journal_head(bh);
1970         return jh;
1971 }
1972
1973 static void __journal_remove_journal_head(struct buffer_head *bh)
1974 {
1975         struct journal_head *jh = bh2jh(bh);
1976
1977         J_ASSERT_JH(jh, jh->b_jcount >= 0);
1978
1979         get_bh(bh);
1980         if (jh->b_jcount == 0) {
1981                 if (jh->b_transaction == NULL &&
1982                                 jh->b_next_transaction == NULL &&
1983                                 jh->b_cp_transaction == NULL) {
1984                         J_ASSERT_JH(jh, jh->b_jlist == BJ_None);
1985                         J_ASSERT_BH(bh, buffer_jbd(bh));
1986                         J_ASSERT_BH(bh, jh2bh(jh) == bh);
1987                         BUFFER_TRACE(bh, "remove journal_head");
1988                         if (jh->b_frozen_data) {
1989                                 printk(KERN_WARNING "%s: freeing "
1990                                                 "b_frozen_data\n",
1991                                                 __func__);
1992                                 jbd2_free(jh->b_frozen_data, bh->b_size);
1993                         }
1994                         if (jh->b_committed_data) {
1995                                 printk(KERN_WARNING "%s: freeing "
1996                                                 "b_committed_data\n",
1997                                                 __func__);
1998                                 jbd2_free(jh->b_committed_data, bh->b_size);
1999                         }
2000                         bh->b_private = NULL;
2001                         jh->b_bh = NULL;        /* debug, really */
2002                         clear_buffer_jbd(bh);
2003                         __brelse(bh);
2004                         journal_free_journal_head(jh);
2005                 } else {
2006                         BUFFER_TRACE(bh, "journal_head was locked");
2007                 }
2008         }
2009 }
2010
2011 /*
2012  * jbd2_journal_remove_journal_head(): if the buffer isn't attached to a transaction
2013  * and has a zero b_jcount then remove and release its journal_head.   If we did
2014  * see that the buffer is not used by any transaction we also "logically"
2015  * decrement ->b_count.
2016  *
2017  * We in fact take an additional increment on ->b_count as a convenience,
2018  * because the caller usually wants to do additional things with the bh
2019  * after calling here.
2020  * The caller of jbd2_journal_remove_journal_head() *must* run __brelse(bh) at some
2021  * time.  Once the caller has run __brelse(), the buffer is eligible for
2022  * reaping by try_to_free_buffers().
2023  */
2024 void jbd2_journal_remove_journal_head(struct buffer_head *bh)
2025 {
2026         jbd_lock_bh_journal_head(bh);
2027         __journal_remove_journal_head(bh);
2028         jbd_unlock_bh_journal_head(bh);
2029 }
2030
2031 /*
2032  * Drop a reference on the passed journal_head.  If it fell to zero then try to
2033  * release the journal_head from the buffer_head.
2034  */
2035 void jbd2_journal_put_journal_head(struct journal_head *jh)
2036 {
2037         struct buffer_head *bh = jh2bh(jh);
2038
2039         jbd_lock_bh_journal_head(bh);
2040         J_ASSERT_JH(jh, jh->b_jcount > 0);
2041         --jh->b_jcount;
2042         if (!jh->b_jcount && !jh->b_transaction) {
2043                 __journal_remove_journal_head(bh);
2044                 __brelse(bh);
2045         }
2046         jbd_unlock_bh_journal_head(bh);
2047 }
2048
2049 /*
2050  * Initialize jbd inode head
2051  */
2052 void jbd2_journal_init_jbd_inode(struct jbd2_inode *jinode, struct inode *inode)
2053 {
2054         jinode->i_transaction = NULL;
2055         jinode->i_next_transaction = NULL;
2056         jinode->i_vfs_inode = inode;
2057         jinode->i_flags = 0;
2058         INIT_LIST_HEAD(&jinode->i_list);
2059 }
2060
2061 /*
2062  * Function to be called before we start removing inode from memory (i.e.,
2063  * clear_inode() is a fine place to be called from). It removes inode from
2064  * transaction's lists.
2065  */
2066 void jbd2_journal_release_jbd_inode(journal_t *journal,
2067                                     struct jbd2_inode *jinode)
2068 {
2069         int writeout = 0;
2070
2071         if (!journal)
2072                 return;
2073 restart:
2074         spin_lock(&journal->j_list_lock);
2075         /* Is commit writing out inode - we have to wait */
2076         if (jinode->i_flags & JI_COMMIT_RUNNING) {
2077                 wait_queue_head_t *wq;
2078                 DEFINE_WAIT_BIT(wait, &jinode->i_flags, __JI_COMMIT_RUNNING);
2079                 wq = bit_waitqueue(&jinode->i_flags, __JI_COMMIT_RUNNING);
2080                 prepare_to_wait(wq, &wait.wait, TASK_UNINTERRUPTIBLE);
2081                 spin_unlock(&journal->j_list_lock);
2082                 schedule();
2083                 finish_wait(wq, &wait.wait);
2084                 goto restart;
2085         }
2086
2087         /* Do we need to wait for data writeback? */
2088         if (journal->j_committing_transaction == jinode->i_transaction)
2089                 writeout = 1;
2090         if (jinode->i_transaction) {
2091                 list_del(&jinode->i_list);
2092                 jinode->i_transaction = NULL;
2093         }
2094         spin_unlock(&journal->j_list_lock);
2095 }
2096
2097 /*
2098  * debugfs tunables
2099  */
2100 #ifdef CONFIG_JBD2_DEBUG
2101 u8 jbd2_journal_enable_debug __read_mostly;
2102 EXPORT_SYMBOL(jbd2_journal_enable_debug);
2103
2104 #define JBD2_DEBUG_NAME "jbd2-debug"
2105
2106 static struct dentry *jbd2_debugfs_dir;
2107 static struct dentry *jbd2_debug;
2108
2109 static void __init jbd2_create_debugfs_entry(void)
2110 {
2111         jbd2_debugfs_dir = debugfs_create_dir("jbd2", NULL);
2112         if (jbd2_debugfs_dir)
2113                 jbd2_debug = debugfs_create_u8(JBD2_DEBUG_NAME, S_IRUGO,
2114                                                jbd2_debugfs_dir,
2115                                                &jbd2_journal_enable_debug);
2116 }
2117
2118 static void __exit jbd2_remove_debugfs_entry(void)
2119 {
2120         debugfs_remove(jbd2_debug);
2121         debugfs_remove(jbd2_debugfs_dir);
2122 }
2123
2124 #else
2125
2126 static void __init jbd2_create_debugfs_entry(void)
2127 {
2128 }
2129
2130 static void __exit jbd2_remove_debugfs_entry(void)
2131 {
2132 }
2133
2134 #endif
2135
2136 #ifdef CONFIG_PROC_FS
2137
2138 #define JBD2_STATS_PROC_NAME "fs/jbd2"
2139
2140 static void __init jbd2_create_jbd_stats_proc_entry(void)
2141 {
2142         proc_jbd2_stats = proc_mkdir(JBD2_STATS_PROC_NAME, NULL);
2143 }
2144
2145 static void __exit jbd2_remove_jbd_stats_proc_entry(void)
2146 {
2147         if (proc_jbd2_stats)
2148                 remove_proc_entry(JBD2_STATS_PROC_NAME, NULL);
2149 }
2150
2151 #else
2152
2153 #define jbd2_create_jbd_stats_proc_entry() do {} while (0)
2154 #define jbd2_remove_jbd_stats_proc_entry() do {} while (0)
2155
2156 #endif
2157
2158 struct kmem_cache *jbd2_handle_cache;
2159
2160 static int __init journal_init_handle_cache(void)
2161 {
2162         jbd2_handle_cache = kmem_cache_create("jbd2_journal_handle",
2163                                 sizeof(handle_t),
2164                                 0,              /* offset */
2165                                 SLAB_TEMPORARY, /* flags */
2166                                 NULL);          /* ctor */
2167         if (jbd2_handle_cache == NULL) {
2168                 printk(KERN_EMERG "JBD: failed to create handle cache\n");
2169                 return -ENOMEM;
2170         }
2171         return 0;
2172 }
2173
2174 static void jbd2_journal_destroy_handle_cache(void)
2175 {
2176         if (jbd2_handle_cache)
2177                 kmem_cache_destroy(jbd2_handle_cache);
2178 }
2179
2180 /*
2181  * Module startup and shutdown
2182  */
2183
2184 static int __init journal_init_caches(void)
2185 {
2186         int ret;
2187
2188         ret = jbd2_journal_init_revoke_caches();
2189         if (ret == 0)
2190                 ret = journal_init_jbd2_journal_head_cache();
2191         if (ret == 0)
2192                 ret = journal_init_handle_cache();
2193         return ret;
2194 }
2195
2196 static void jbd2_journal_destroy_caches(void)
2197 {
2198         jbd2_journal_destroy_revoke_caches();
2199         jbd2_journal_destroy_jbd2_journal_head_cache();
2200         jbd2_journal_destroy_handle_cache();
2201 }
2202
2203 static int __init journal_init(void)
2204 {
2205         int ret;
2206
2207         BUILD_BUG_ON(sizeof(struct journal_superblock_s) != 1024);
2208
2209         ret = journal_init_caches();
2210         if (ret == 0) {
2211                 jbd2_create_debugfs_entry();
2212                 jbd2_create_jbd_stats_proc_entry();
2213         } else {
2214                 jbd2_journal_destroy_caches();
2215         }
2216         return ret;
2217 }
2218
2219 static void __exit journal_exit(void)
2220 {
2221 #ifdef CONFIG_JBD2_DEBUG
2222         int n = atomic_read(&nr_journal_heads);
2223         if (n)
2224                 printk(KERN_EMERG "JBD: leaked %d journal_heads!\n", n);
2225 #endif
2226         jbd2_remove_debugfs_entry();
2227         jbd2_remove_jbd_stats_proc_entry();
2228         jbd2_journal_destroy_caches();
2229 }
2230
2231 /* 
2232  * jbd2_dev_to_name is a utility function used by the jbd2 and ext4 
2233  * tracing infrastructure to map a dev_t to a device name.
2234  *
2235  * The caller should use rcu_read_lock() in order to make sure the
2236  * device name stays valid until its done with it.  We use
2237  * rcu_read_lock() as well to make sure we're safe in case the caller
2238  * gets sloppy, and because rcu_read_lock() is cheap and can be safely
2239  * nested.
2240  */
2241 struct devname_cache {
2242         struct rcu_head rcu;
2243         dev_t           device;
2244         char            devname[BDEVNAME_SIZE];
2245 };
2246 #define CACHE_SIZE_BITS 6
2247 static struct devname_cache *devcache[1 << CACHE_SIZE_BITS];
2248 static DEFINE_SPINLOCK(devname_cache_lock);
2249
2250 static void free_devcache(struct rcu_head *rcu)
2251 {
2252         kfree(rcu);
2253 }
2254
2255 const char *jbd2_dev_to_name(dev_t device)
2256 {
2257         int     i = hash_32(device, CACHE_SIZE_BITS);
2258         char    *ret;
2259         struct block_device *bd;
2260         static struct devname_cache *new_dev;
2261
2262         rcu_read_lock();
2263         if (devcache[i] && devcache[i]->device == device) {
2264                 ret = devcache[i]->devname;
2265                 rcu_read_unlock();
2266                 return ret;
2267         }
2268         rcu_read_unlock();
2269
2270         new_dev = kmalloc(sizeof(struct devname_cache), GFP_KERNEL);
2271         if (!new_dev)
2272                 return "NODEV-ALLOCFAILURE"; /* Something non-NULL */
2273         spin_lock(&devname_cache_lock);
2274         if (devcache[i]) {
2275                 if (devcache[i]->device == device) {
2276                         kfree(new_dev);
2277                         ret = devcache[i]->devname;
2278                         spin_unlock(&devname_cache_lock);
2279                         return ret;
2280                 }
2281                 call_rcu(&devcache[i]->rcu, free_devcache);
2282         }
2283         devcache[i] = new_dev;
2284         devcache[i]->device = device;
2285         bd = bdget(device);
2286         if (bd) {
2287                 bdevname(bd, devcache[i]->devname);
2288                 bdput(bd);
2289         } else
2290                 __bdevname(device, devcache[i]->devname);
2291         ret = devcache[i]->devname;
2292         spin_unlock(&devname_cache_lock);
2293         return ret;
2294 }
2295 EXPORT_SYMBOL(jbd2_dev_to_name);
2296
2297 MODULE_LICENSE("GPL");
2298 module_init(journal_init);
2299 module_exit(journal_exit);
2300