jbd: Fix comment to match the code in journal_start()
[linux-2.6.git] / fs / jbd / journal.c
1 /*
2  * linux/fs/jbd/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/jbd.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/ratelimit.h>
40
41 #include <asm/uaccess.h>
42 #include <asm/page.h>
43
44 EXPORT_SYMBOL(journal_start);
45 EXPORT_SYMBOL(journal_restart);
46 EXPORT_SYMBOL(journal_extend);
47 EXPORT_SYMBOL(journal_stop);
48 EXPORT_SYMBOL(journal_lock_updates);
49 EXPORT_SYMBOL(journal_unlock_updates);
50 EXPORT_SYMBOL(journal_get_write_access);
51 EXPORT_SYMBOL(journal_get_create_access);
52 EXPORT_SYMBOL(journal_get_undo_access);
53 EXPORT_SYMBOL(journal_dirty_data);
54 EXPORT_SYMBOL(journal_dirty_metadata);
55 EXPORT_SYMBOL(journal_release_buffer);
56 EXPORT_SYMBOL(journal_forget);
57 #if 0
58 EXPORT_SYMBOL(journal_sync_buffer);
59 #endif
60 EXPORT_SYMBOL(journal_flush);
61 EXPORT_SYMBOL(journal_revoke);
62
63 EXPORT_SYMBOL(journal_init_dev);
64 EXPORT_SYMBOL(journal_init_inode);
65 EXPORT_SYMBOL(journal_update_format);
66 EXPORT_SYMBOL(journal_check_used_features);
67 EXPORT_SYMBOL(journal_check_available_features);
68 EXPORT_SYMBOL(journal_set_features);
69 EXPORT_SYMBOL(journal_create);
70 EXPORT_SYMBOL(journal_load);
71 EXPORT_SYMBOL(journal_destroy);
72 EXPORT_SYMBOL(journal_abort);
73 EXPORT_SYMBOL(journal_errno);
74 EXPORT_SYMBOL(journal_ack_err);
75 EXPORT_SYMBOL(journal_clear_err);
76 EXPORT_SYMBOL(log_wait_commit);
77 EXPORT_SYMBOL(log_start_commit);
78 EXPORT_SYMBOL(journal_start_commit);
79 EXPORT_SYMBOL(journal_force_commit_nested);
80 EXPORT_SYMBOL(journal_wipe);
81 EXPORT_SYMBOL(journal_blocks_per_page);
82 EXPORT_SYMBOL(journal_invalidatepage);
83 EXPORT_SYMBOL(journal_try_to_free_buffers);
84 EXPORT_SYMBOL(journal_force_commit);
85
86 static int journal_convert_superblock_v1(journal_t *, journal_superblock_t *);
87 static void __journal_abort_soft (journal_t *journal, int errno);
88 static const char *journal_dev_name(journal_t *journal, char *buffer);
89
90 /*
91  * Helper function used to manage commit timeouts
92  */
93
94 static void commit_timeout(unsigned long __data)
95 {
96         struct task_struct * p = (struct task_struct *) __data;
97
98         wake_up_process(p);
99 }
100
101 /*
102  * kjournald: The main thread function used to manage a logging device
103  * journal.
104  *
105  * This kernel thread is responsible for two things:
106  *
107  * 1) COMMIT:  Every so often we need to commit the current state of the
108  *    filesystem to disk.  The journal thread is responsible for writing
109  *    all of the metadata buffers to disk.
110  *
111  * 2) CHECKPOINT: We cannot reuse a used section of the log file until all
112  *    of the data in that part of the log has been rewritten elsewhere on
113  *    the disk.  Flushing these old buffers to reclaim space in the log is
114  *    known as checkpointing, and this thread is responsible for that job.
115  */
116
117 static int kjournald(void *arg)
118 {
119         journal_t *journal = arg;
120         transaction_t *transaction;
121
122         /*
123          * Set up an interval timer which can be used to trigger a commit wakeup
124          * after the commit interval expires
125          */
126         setup_timer(&journal->j_commit_timer, commit_timeout,
127                         (unsigned long)current);
128
129         /* Record that the journal thread is running */
130         journal->j_task = current;
131         wake_up(&journal->j_wait_done_commit);
132
133         printk(KERN_INFO "kjournald starting.  Commit interval %ld seconds\n",
134                         journal->j_commit_interval / HZ);
135
136         /*
137          * And now, wait forever for commit wakeup events.
138          */
139         spin_lock(&journal->j_state_lock);
140
141 loop:
142         if (journal->j_flags & JFS_UNMOUNT)
143                 goto end_loop;
144
145         jbd_debug(1, "commit_sequence=%d, commit_request=%d\n",
146                 journal->j_commit_sequence, journal->j_commit_request);
147
148         if (journal->j_commit_sequence != journal->j_commit_request) {
149                 jbd_debug(1, "OK, requests differ\n");
150                 spin_unlock(&journal->j_state_lock);
151                 del_timer_sync(&journal->j_commit_timer);
152                 journal_commit_transaction(journal);
153                 spin_lock(&journal->j_state_lock);
154                 goto loop;
155         }
156
157         wake_up(&journal->j_wait_done_commit);
158         if (freezing(current)) {
159                 /*
160                  * The simpler the better. Flushing journal isn't a
161                  * good idea, because that depends on threads that may
162                  * be already stopped.
163                  */
164                 jbd_debug(1, "Now suspending kjournald\n");
165                 spin_unlock(&journal->j_state_lock);
166                 refrigerator();
167                 spin_lock(&journal->j_state_lock);
168         } else {
169                 /*
170                  * We assume on resume that commits are already there,
171                  * so we don't sleep
172                  */
173                 DEFINE_WAIT(wait);
174                 int should_sleep = 1;
175
176                 prepare_to_wait(&journal->j_wait_commit, &wait,
177                                 TASK_INTERRUPTIBLE);
178                 if (journal->j_commit_sequence != journal->j_commit_request)
179                         should_sleep = 0;
180                 transaction = journal->j_running_transaction;
181                 if (transaction && time_after_eq(jiffies,
182                                                 transaction->t_expires))
183                         should_sleep = 0;
184                 if (journal->j_flags & JFS_UNMOUNT)
185                         should_sleep = 0;
186                 if (should_sleep) {
187                         spin_unlock(&journal->j_state_lock);
188                         schedule();
189                         spin_lock(&journal->j_state_lock);
190                 }
191                 finish_wait(&journal->j_wait_commit, &wait);
192         }
193
194         jbd_debug(1, "kjournald wakes\n");
195
196         /*
197          * Were we woken up by a commit wakeup event?
198          */
199         transaction = journal->j_running_transaction;
200         if (transaction && time_after_eq(jiffies, transaction->t_expires)) {
201                 journal->j_commit_request = transaction->t_tid;
202                 jbd_debug(1, "woke because of timeout\n");
203         }
204         goto loop;
205
206 end_loop:
207         spin_unlock(&journal->j_state_lock);
208         del_timer_sync(&journal->j_commit_timer);
209         journal->j_task = NULL;
210         wake_up(&journal->j_wait_done_commit);
211         jbd_debug(1, "Journal thread exiting.\n");
212         return 0;
213 }
214
215 static int journal_start_thread(journal_t *journal)
216 {
217         struct task_struct *t;
218
219         t = kthread_run(kjournald, journal, "kjournald");
220         if (IS_ERR(t))
221                 return PTR_ERR(t);
222
223         wait_event(journal->j_wait_done_commit, journal->j_task != NULL);
224         return 0;
225 }
226
227 static void journal_kill_thread(journal_t *journal)
228 {
229         spin_lock(&journal->j_state_lock);
230         journal->j_flags |= JFS_UNMOUNT;
231
232         while (journal->j_task) {
233                 wake_up(&journal->j_wait_commit);
234                 spin_unlock(&journal->j_state_lock);
235                 wait_event(journal->j_wait_done_commit,
236                                 journal->j_task == NULL);
237                 spin_lock(&journal->j_state_lock);
238         }
239         spin_unlock(&journal->j_state_lock);
240 }
241
242 /*
243  * journal_write_metadata_buffer: write a metadata buffer to the journal.
244  *
245  * Writes a metadata buffer to a given disk block.  The actual IO is not
246  * performed but a new buffer_head is constructed which labels the data
247  * to be written with the correct destination disk block.
248  *
249  * Any magic-number escaping which needs to be done will cause a
250  * copy-out here.  If the buffer happens to start with the
251  * JFS_MAGIC_NUMBER, then we can't write it to the log directly: the
252  * magic number is only written to the log for descripter blocks.  In
253  * this case, we copy the data and replace the first word with 0, and we
254  * return a result code which indicates that this buffer needs to be
255  * marked as an escaped buffer in the corresponding log descriptor
256  * block.  The missing word can then be restored when the block is read
257  * during recovery.
258  *
259  * If the source buffer has already been modified by a new transaction
260  * since we took the last commit snapshot, we use the frozen copy of
261  * that data for IO.  If we end up using the existing buffer_head's data
262  * for the write, then we *have* to lock the buffer to prevent anyone
263  * else from using and possibly modifying it while the IO is in
264  * progress.
265  *
266  * The function returns a pointer to the buffer_heads to be used for IO.
267  *
268  * We assume that the journal has already been locked in this function.
269  *
270  * Return value:
271  *  <0: Error
272  * >=0: Finished OK
273  *
274  * On success:
275  * Bit 0 set == escape performed on the data
276  * Bit 1 set == buffer copy-out performed (kfree the data after IO)
277  */
278
279 int journal_write_metadata_buffer(transaction_t *transaction,
280                                   struct journal_head  *jh_in,
281                                   struct journal_head **jh_out,
282                                   unsigned int blocknr)
283 {
284         int need_copy_out = 0;
285         int done_copy_out = 0;
286         int do_escape = 0;
287         char *mapped_data;
288         struct buffer_head *new_bh;
289         struct journal_head *new_jh;
290         struct page *new_page;
291         unsigned int new_offset;
292         struct buffer_head *bh_in = jh2bh(jh_in);
293         journal_t *journal = transaction->t_journal;
294
295         /*
296          * The buffer really shouldn't be locked: only the current committing
297          * transaction is allowed to write it, so nobody else is allowed
298          * to do any IO.
299          *
300          * akpm: except if we're journalling data, and write() output is
301          * also part of a shared mapping, and another thread has
302          * decided to launch a writepage() against this buffer.
303          */
304         J_ASSERT_BH(bh_in, buffer_jbddirty(bh_in));
305
306         new_bh = alloc_buffer_head(GFP_NOFS|__GFP_NOFAIL);
307         /* keep subsequent assertions sane */
308         new_bh->b_state = 0;
309         init_buffer(new_bh, NULL, NULL);
310         atomic_set(&new_bh->b_count, 1);
311         new_jh = journal_add_journal_head(new_bh);      /* This sleeps */
312
313         /*
314          * If a new transaction has already done a buffer copy-out, then
315          * we use that version of the data for the commit.
316          */
317         jbd_lock_bh_state(bh_in);
318 repeat:
319         if (jh_in->b_frozen_data) {
320                 done_copy_out = 1;
321                 new_page = virt_to_page(jh_in->b_frozen_data);
322                 new_offset = offset_in_page(jh_in->b_frozen_data);
323         } else {
324                 new_page = jh2bh(jh_in)->b_page;
325                 new_offset = offset_in_page(jh2bh(jh_in)->b_data);
326         }
327
328         mapped_data = kmap_atomic(new_page, KM_USER0);
329         /*
330          * Check for escaping
331          */
332         if (*((__be32 *)(mapped_data + new_offset)) ==
333                                 cpu_to_be32(JFS_MAGIC_NUMBER)) {
334                 need_copy_out = 1;
335                 do_escape = 1;
336         }
337         kunmap_atomic(mapped_data, KM_USER0);
338
339         /*
340          * Do we need to do a data copy?
341          */
342         if (need_copy_out && !done_copy_out) {
343                 char *tmp;
344
345                 jbd_unlock_bh_state(bh_in);
346                 tmp = jbd_alloc(bh_in->b_size, GFP_NOFS);
347                 jbd_lock_bh_state(bh_in);
348                 if (jh_in->b_frozen_data) {
349                         jbd_free(tmp, bh_in->b_size);
350                         goto repeat;
351                 }
352
353                 jh_in->b_frozen_data = tmp;
354                 mapped_data = kmap_atomic(new_page, KM_USER0);
355                 memcpy(tmp, mapped_data + new_offset, jh2bh(jh_in)->b_size);
356                 kunmap_atomic(mapped_data, KM_USER0);
357
358                 new_page = virt_to_page(tmp);
359                 new_offset = offset_in_page(tmp);
360                 done_copy_out = 1;
361         }
362
363         /*
364          * Did we need to do an escaping?  Now we've done all the
365          * copying, we can finally do so.
366          */
367         if (do_escape) {
368                 mapped_data = kmap_atomic(new_page, KM_USER0);
369                 *((unsigned int *)(mapped_data + new_offset)) = 0;
370                 kunmap_atomic(mapped_data, KM_USER0);
371         }
372
373         set_bh_page(new_bh, new_page, new_offset);
374         new_jh->b_transaction = NULL;
375         new_bh->b_size = jh2bh(jh_in)->b_size;
376         new_bh->b_bdev = transaction->t_journal->j_dev;
377         new_bh->b_blocknr = blocknr;
378         set_buffer_mapped(new_bh);
379         set_buffer_dirty(new_bh);
380
381         *jh_out = new_jh;
382
383         /*
384          * The to-be-written buffer needs to get moved to the io queue,
385          * and the original buffer whose contents we are shadowing or
386          * copying is moved to the transaction's shadow queue.
387          */
388         JBUFFER_TRACE(jh_in, "file as BJ_Shadow");
389         spin_lock(&journal->j_list_lock);
390         __journal_file_buffer(jh_in, transaction, BJ_Shadow);
391         spin_unlock(&journal->j_list_lock);
392         jbd_unlock_bh_state(bh_in);
393
394         JBUFFER_TRACE(new_jh, "file as BJ_IO");
395         journal_file_buffer(new_jh, transaction, BJ_IO);
396
397         return do_escape | (done_copy_out << 1);
398 }
399
400 /*
401  * Allocation code for the journal file.  Manage the space left in the
402  * journal, so that we can begin checkpointing when appropriate.
403  */
404
405 /*
406  * __log_space_left: Return the number of free blocks left in the journal.
407  *
408  * Called with the journal already locked.
409  *
410  * Called under j_state_lock
411  */
412
413 int __log_space_left(journal_t *journal)
414 {
415         int left = journal->j_free;
416
417         assert_spin_locked(&journal->j_state_lock);
418
419         /*
420          * Be pessimistic here about the number of those free blocks which
421          * might be required for log descriptor control blocks.
422          */
423
424 #define MIN_LOG_RESERVED_BLOCKS 32 /* Allow for rounding errors */
425
426         left -= MIN_LOG_RESERVED_BLOCKS;
427
428         if (left <= 0)
429                 return 0;
430         left -= (left >> 3);
431         return left;
432 }
433
434 /*
435  * Called under j_state_lock.  Returns true if a transaction commit was started.
436  */
437 int __log_start_commit(journal_t *journal, tid_t target)
438 {
439         /*
440          * The only transaction we can possibly wait upon is the
441          * currently running transaction (if it exists).  Otherwise,
442          * the target tid must be an old one.
443          */
444         if (journal->j_running_transaction &&
445             journal->j_running_transaction->t_tid == target) {
446                 /*
447                  * We want a new commit: OK, mark the request and wakeup the
448                  * commit thread.  We do _not_ do the commit ourselves.
449                  */
450
451                 journal->j_commit_request = target;
452                 jbd_debug(1, "JBD: requesting commit %d/%d\n",
453                           journal->j_commit_request,
454                           journal->j_commit_sequence);
455                 wake_up(&journal->j_wait_commit);
456                 return 1;
457         } else if (!tid_geq(journal->j_commit_request, target))
458                 /* This should never happen, but if it does, preserve
459                    the evidence before kjournald goes into a loop and
460                    increments j_commit_sequence beyond all recognition. */
461                 WARN_ONCE(1, "jbd: bad log_start_commit: %u %u %u %u\n",
462                     journal->j_commit_request, journal->j_commit_sequence,
463                     target, journal->j_running_transaction ?
464                     journal->j_running_transaction->t_tid : 0);
465         return 0;
466 }
467
468 int log_start_commit(journal_t *journal, tid_t tid)
469 {
470         int ret;
471
472         spin_lock(&journal->j_state_lock);
473         ret = __log_start_commit(journal, tid);
474         spin_unlock(&journal->j_state_lock);
475         return ret;
476 }
477
478 /*
479  * Force and wait upon a commit if the calling process is not within
480  * transaction.  This is used for forcing out undo-protected data which contains
481  * bitmaps, when the fs is running out of space.
482  *
483  * We can only force the running transaction if we don't have an active handle;
484  * otherwise, we will deadlock.
485  *
486  * Returns true if a transaction was started.
487  */
488 int journal_force_commit_nested(journal_t *journal)
489 {
490         transaction_t *transaction = NULL;
491         tid_t tid;
492
493         spin_lock(&journal->j_state_lock);
494         if (journal->j_running_transaction && !current->journal_info) {
495                 transaction = journal->j_running_transaction;
496                 __log_start_commit(journal, transaction->t_tid);
497         } else if (journal->j_committing_transaction)
498                 transaction = journal->j_committing_transaction;
499
500         if (!transaction) {
501                 spin_unlock(&journal->j_state_lock);
502                 return 0;       /* Nothing to retry */
503         }
504
505         tid = transaction->t_tid;
506         spin_unlock(&journal->j_state_lock);
507         log_wait_commit(journal, tid);
508         return 1;
509 }
510
511 /*
512  * Start a commit of the current running transaction (if any).  Returns true
513  * if a transaction is going to be committed (or is currently already
514  * committing), and fills its tid in at *ptid
515  */
516 int journal_start_commit(journal_t *journal, tid_t *ptid)
517 {
518         int ret = 0;
519
520         spin_lock(&journal->j_state_lock);
521         if (journal->j_running_transaction) {
522                 tid_t tid = journal->j_running_transaction->t_tid;
523
524                 __log_start_commit(journal, tid);
525                 /* There's a running transaction and we've just made sure
526                  * it's commit has been scheduled. */
527                 if (ptid)
528                         *ptid = tid;
529                 ret = 1;
530         } else if (journal->j_committing_transaction) {
531                 /*
532                  * If ext3_write_super() recently started a commit, then we
533                  * have to wait for completion of that transaction
534                  */
535                 if (ptid)
536                         *ptid = journal->j_committing_transaction->t_tid;
537                 ret = 1;
538         }
539         spin_unlock(&journal->j_state_lock);
540         return ret;
541 }
542
543 /*
544  * Wait for a specified commit to complete.
545  * The caller may not hold the journal lock.
546  */
547 int log_wait_commit(journal_t *journal, tid_t tid)
548 {
549         int err = 0;
550
551 #ifdef CONFIG_JBD_DEBUG
552         spin_lock(&journal->j_state_lock);
553         if (!tid_geq(journal->j_commit_request, tid)) {
554                 printk(KERN_EMERG
555                        "%s: error: j_commit_request=%d, tid=%d\n",
556                        __func__, journal->j_commit_request, tid);
557         }
558         spin_unlock(&journal->j_state_lock);
559 #endif
560         spin_lock(&journal->j_state_lock);
561         while (tid_gt(tid, journal->j_commit_sequence)) {
562                 jbd_debug(1, "JBD: want %d, j_commit_sequence=%d\n",
563                                   tid, journal->j_commit_sequence);
564                 wake_up(&journal->j_wait_commit);
565                 spin_unlock(&journal->j_state_lock);
566                 wait_event(journal->j_wait_done_commit,
567                                 !tid_gt(tid, journal->j_commit_sequence));
568                 spin_lock(&journal->j_state_lock);
569         }
570         spin_unlock(&journal->j_state_lock);
571
572         if (unlikely(is_journal_aborted(journal))) {
573                 printk(KERN_EMERG "journal commit I/O error\n");
574                 err = -EIO;
575         }
576         return err;
577 }
578
579 /*
580  * Return 1 if a given transaction has not yet sent barrier request
581  * connected with a transaction commit. If 0 is returned, transaction
582  * may or may not have sent the barrier. Used to avoid sending barrier
583  * twice in common cases.
584  */
585 int journal_trans_will_send_data_barrier(journal_t *journal, tid_t tid)
586 {
587         int ret = 0;
588         transaction_t *commit_trans;
589
590         if (!(journal->j_flags & JFS_BARRIER))
591                 return 0;
592         spin_lock(&journal->j_state_lock);
593         /* Transaction already committed? */
594         if (tid_geq(journal->j_commit_sequence, tid))
595                 goto out;
596         /*
597          * Transaction is being committed and we already proceeded to
598          * writing commit record?
599          */
600         commit_trans = journal->j_committing_transaction;
601         if (commit_trans && commit_trans->t_tid == tid &&
602             commit_trans->t_state >= T_COMMIT_RECORD)
603                 goto out;
604         ret = 1;
605 out:
606         spin_unlock(&journal->j_state_lock);
607         return ret;
608 }
609 EXPORT_SYMBOL(journal_trans_will_send_data_barrier);
610
611 /*
612  * Log buffer allocation routines:
613  */
614
615 int journal_next_log_block(journal_t *journal, unsigned int *retp)
616 {
617         unsigned int blocknr;
618
619         spin_lock(&journal->j_state_lock);
620         J_ASSERT(journal->j_free > 1);
621
622         blocknr = journal->j_head;
623         journal->j_head++;
624         journal->j_free--;
625         if (journal->j_head == journal->j_last)
626                 journal->j_head = journal->j_first;
627         spin_unlock(&journal->j_state_lock);
628         return journal_bmap(journal, blocknr, retp);
629 }
630
631 /*
632  * Conversion of logical to physical block numbers for the journal
633  *
634  * On external journals the journal blocks are identity-mapped, so
635  * this is a no-op.  If needed, we can use j_blk_offset - everything is
636  * ready.
637  */
638 int journal_bmap(journal_t *journal, unsigned int blocknr,
639                  unsigned int *retp)
640 {
641         int err = 0;
642         unsigned int ret;
643
644         if (journal->j_inode) {
645                 ret = bmap(journal->j_inode, blocknr);
646                 if (ret)
647                         *retp = ret;
648                 else {
649                         char b[BDEVNAME_SIZE];
650
651                         printk(KERN_ALERT "%s: journal block not found "
652                                         "at offset %u on %s\n",
653                                 __func__,
654                                 blocknr,
655                                 bdevname(journal->j_dev, b));
656                         err = -EIO;
657                         __journal_abort_soft(journal, err);
658                 }
659         } else {
660                 *retp = blocknr; /* +journal->j_blk_offset */
661         }
662         return err;
663 }
664
665 /*
666  * We play buffer_head aliasing tricks to write data/metadata blocks to
667  * the journal without copying their contents, but for journal
668  * descriptor blocks we do need to generate bona fide buffers.
669  *
670  * After the caller of journal_get_descriptor_buffer() has finished modifying
671  * the buffer's contents they really should run flush_dcache_page(bh->b_page).
672  * But we don't bother doing that, so there will be coherency problems with
673  * mmaps of blockdevs which hold live JBD-controlled filesystems.
674  */
675 struct journal_head *journal_get_descriptor_buffer(journal_t *journal)
676 {
677         struct buffer_head *bh;
678         unsigned int blocknr;
679         int err;
680
681         err = journal_next_log_block(journal, &blocknr);
682
683         if (err)
684                 return NULL;
685
686         bh = __getblk(journal->j_dev, blocknr, journal->j_blocksize);
687         if (!bh)
688                 return NULL;
689         lock_buffer(bh);
690         memset(bh->b_data, 0, journal->j_blocksize);
691         set_buffer_uptodate(bh);
692         unlock_buffer(bh);
693         BUFFER_TRACE(bh, "return this buffer");
694         return journal_add_journal_head(bh);
695 }
696
697 /*
698  * Management for journal control blocks: functions to create and
699  * destroy journal_t structures, and to initialise and read existing
700  * journal blocks from disk.  */
701
702 /* First: create and setup a journal_t object in memory.  We initialise
703  * very few fields yet: that has to wait until we have created the
704  * journal structures from from scratch, or loaded them from disk. */
705
706 static journal_t * journal_init_common (void)
707 {
708         journal_t *journal;
709         int err;
710
711         journal = kzalloc(sizeof(*journal), GFP_KERNEL);
712         if (!journal)
713                 goto fail;
714
715         init_waitqueue_head(&journal->j_wait_transaction_locked);
716         init_waitqueue_head(&journal->j_wait_logspace);
717         init_waitqueue_head(&journal->j_wait_done_commit);
718         init_waitqueue_head(&journal->j_wait_checkpoint);
719         init_waitqueue_head(&journal->j_wait_commit);
720         init_waitqueue_head(&journal->j_wait_updates);
721         mutex_init(&journal->j_barrier);
722         mutex_init(&journal->j_checkpoint_mutex);
723         spin_lock_init(&journal->j_revoke_lock);
724         spin_lock_init(&journal->j_list_lock);
725         spin_lock_init(&journal->j_state_lock);
726
727         journal->j_commit_interval = (HZ * JBD_DEFAULT_MAX_COMMIT_AGE);
728
729         /* The journal is marked for error until we succeed with recovery! */
730         journal->j_flags = JFS_ABORT;
731
732         /* Set up a default-sized revoke table for the new mount. */
733         err = journal_init_revoke(journal, JOURNAL_REVOKE_DEFAULT_HASH);
734         if (err) {
735                 kfree(journal);
736                 goto fail;
737         }
738         return journal;
739 fail:
740         return NULL;
741 }
742
743 /* journal_init_dev and journal_init_inode:
744  *
745  * Create a journal structure assigned some fixed set of disk blocks to
746  * the journal.  We don't actually touch those disk blocks yet, but we
747  * need to set up all of the mapping information to tell the journaling
748  * system where the journal blocks are.
749  *
750  */
751
752 /**
753  *  journal_t * journal_init_dev() - creates and initialises a journal structure
754  *  @bdev: Block device on which to create the journal
755  *  @fs_dev: Device which hold journalled filesystem for this journal.
756  *  @start: Block nr Start of journal.
757  *  @len:  Length of the journal in blocks.
758  *  @blocksize: blocksize of journalling device
759  *
760  *  Returns: a newly created journal_t *
761  *
762  *  journal_init_dev creates a journal which maps a fixed contiguous
763  *  range of blocks on an arbitrary block device.
764  *
765  */
766 journal_t * journal_init_dev(struct block_device *bdev,
767                         struct block_device *fs_dev,
768                         int start, int len, int blocksize)
769 {
770         journal_t *journal = journal_init_common();
771         struct buffer_head *bh;
772         int n;
773
774         if (!journal)
775                 return NULL;
776
777         /* journal descriptor can store up to n blocks -bzzz */
778         journal->j_blocksize = blocksize;
779         n = journal->j_blocksize / sizeof(journal_block_tag_t);
780         journal->j_wbufsize = n;
781         journal->j_wbuf = kmalloc(n * sizeof(struct buffer_head*), GFP_KERNEL);
782         if (!journal->j_wbuf) {
783                 printk(KERN_ERR "%s: Can't allocate bhs for commit thread\n",
784                         __func__);
785                 goto out_err;
786         }
787         journal->j_dev = bdev;
788         journal->j_fs_dev = fs_dev;
789         journal->j_blk_offset = start;
790         journal->j_maxlen = len;
791
792         bh = __getblk(journal->j_dev, start, journal->j_blocksize);
793         if (!bh) {
794                 printk(KERN_ERR
795                        "%s: Cannot get buffer for journal superblock\n",
796                        __func__);
797                 goto out_err;
798         }
799         journal->j_sb_buffer = bh;
800         journal->j_superblock = (journal_superblock_t *)bh->b_data;
801
802         return journal;
803 out_err:
804         kfree(journal->j_wbuf);
805         kfree(journal);
806         return NULL;
807 }
808
809 /**
810  *  journal_t * journal_init_inode () - creates a journal which maps to a inode.
811  *  @inode: An inode to create the journal in
812  *
813  * journal_init_inode creates a journal which maps an on-disk inode as
814  * the journal.  The inode must exist already, must support bmap() and
815  * must have all data blocks preallocated.
816  */
817 journal_t * journal_init_inode (struct inode *inode)
818 {
819         struct buffer_head *bh;
820         journal_t *journal = journal_init_common();
821         int err;
822         int n;
823         unsigned int blocknr;
824
825         if (!journal)
826                 return NULL;
827
828         journal->j_dev = journal->j_fs_dev = inode->i_sb->s_bdev;
829         journal->j_inode = inode;
830         jbd_debug(1,
831                   "journal %p: inode %s/%ld, size %Ld, bits %d, blksize %ld\n",
832                   journal, inode->i_sb->s_id, inode->i_ino,
833                   (long long) inode->i_size,
834                   inode->i_sb->s_blocksize_bits, inode->i_sb->s_blocksize);
835
836         journal->j_maxlen = inode->i_size >> inode->i_sb->s_blocksize_bits;
837         journal->j_blocksize = inode->i_sb->s_blocksize;
838
839         /* journal descriptor can store up to n blocks -bzzz */
840         n = journal->j_blocksize / sizeof(journal_block_tag_t);
841         journal->j_wbufsize = n;
842         journal->j_wbuf = kmalloc(n * sizeof(struct buffer_head*), GFP_KERNEL);
843         if (!journal->j_wbuf) {
844                 printk(KERN_ERR "%s: Can't allocate bhs for commit thread\n",
845                         __func__);
846                 goto out_err;
847         }
848
849         err = journal_bmap(journal, 0, &blocknr);
850         /* If that failed, give up */
851         if (err) {
852                 printk(KERN_ERR "%s: Cannot locate journal superblock\n",
853                        __func__);
854                 goto out_err;
855         }
856
857         bh = __getblk(journal->j_dev, blocknr, journal->j_blocksize);
858         if (!bh) {
859                 printk(KERN_ERR
860                        "%s: Cannot get buffer for journal superblock\n",
861                        __func__);
862                 goto out_err;
863         }
864         journal->j_sb_buffer = bh;
865         journal->j_superblock = (journal_superblock_t *)bh->b_data;
866
867         return journal;
868 out_err:
869         kfree(journal->j_wbuf);
870         kfree(journal);
871         return NULL;
872 }
873
874 /*
875  * If the journal init or create aborts, we need to mark the journal
876  * superblock as being NULL to prevent the journal destroy from writing
877  * back a bogus superblock.
878  */
879 static void journal_fail_superblock (journal_t *journal)
880 {
881         struct buffer_head *bh = journal->j_sb_buffer;
882         brelse(bh);
883         journal->j_sb_buffer = NULL;
884 }
885
886 /*
887  * Given a journal_t structure, initialise the various fields for
888  * startup of a new journaling session.  We use this both when creating
889  * a journal, and after recovering an old journal to reset it for
890  * subsequent use.
891  */
892
893 static int journal_reset(journal_t *journal)
894 {
895         journal_superblock_t *sb = journal->j_superblock;
896         unsigned int first, last;
897
898         first = be32_to_cpu(sb->s_first);
899         last = be32_to_cpu(sb->s_maxlen);
900         if (first + JFS_MIN_JOURNAL_BLOCKS > last + 1) {
901                 printk(KERN_ERR "JBD: Journal too short (blocks %u-%u).\n",
902                        first, last);
903                 journal_fail_superblock(journal);
904                 return -EINVAL;
905         }
906
907         journal->j_first = first;
908         journal->j_last = last;
909
910         journal->j_head = first;
911         journal->j_tail = first;
912         journal->j_free = last - first;
913
914         journal->j_tail_sequence = journal->j_transaction_sequence;
915         journal->j_commit_sequence = journal->j_transaction_sequence - 1;
916         journal->j_commit_request = journal->j_commit_sequence;
917
918         journal->j_max_transaction_buffers = journal->j_maxlen / 4;
919
920         /* Add the dynamic fields and write it to disk. */
921         journal_update_superblock(journal, 1);
922         return journal_start_thread(journal);
923 }
924
925 /**
926  * int journal_create() - Initialise the new journal file
927  * @journal: Journal to create. This structure must have been initialised
928  *
929  * Given a journal_t structure which tells us which disk blocks we can
930  * use, create a new journal superblock and initialise all of the
931  * journal fields from scratch.
932  **/
933 int journal_create(journal_t *journal)
934 {
935         unsigned int blocknr;
936         struct buffer_head *bh;
937         journal_superblock_t *sb;
938         int i, err;
939
940         if (journal->j_maxlen < JFS_MIN_JOURNAL_BLOCKS) {
941                 printk (KERN_ERR "Journal length (%d blocks) too short.\n",
942                         journal->j_maxlen);
943                 journal_fail_superblock(journal);
944                 return -EINVAL;
945         }
946
947         if (journal->j_inode == NULL) {
948                 /*
949                  * We don't know what block to start at!
950                  */
951                 printk(KERN_EMERG
952                        "%s: creation of journal on external device!\n",
953                        __func__);
954                 BUG();
955         }
956
957         /* Zero out the entire journal on disk.  We cannot afford to
958            have any blocks on disk beginning with JFS_MAGIC_NUMBER. */
959         jbd_debug(1, "JBD: Zeroing out journal blocks...\n");
960         for (i = 0; i < journal->j_maxlen; i++) {
961                 err = journal_bmap(journal, i, &blocknr);
962                 if (err)
963                         return err;
964                 bh = __getblk(journal->j_dev, blocknr, journal->j_blocksize);
965                 if (unlikely(!bh))
966                         return -ENOMEM;
967                 lock_buffer(bh);
968                 memset (bh->b_data, 0, journal->j_blocksize);
969                 BUFFER_TRACE(bh, "marking dirty");
970                 mark_buffer_dirty(bh);
971                 BUFFER_TRACE(bh, "marking uptodate");
972                 set_buffer_uptodate(bh);
973                 unlock_buffer(bh);
974                 __brelse(bh);
975         }
976
977         sync_blockdev(journal->j_dev);
978         jbd_debug(1, "JBD: journal cleared.\n");
979
980         /* OK, fill in the initial static fields in the new superblock */
981         sb = journal->j_superblock;
982
983         sb->s_header.h_magic     = cpu_to_be32(JFS_MAGIC_NUMBER);
984         sb->s_header.h_blocktype = cpu_to_be32(JFS_SUPERBLOCK_V2);
985
986         sb->s_blocksize = cpu_to_be32(journal->j_blocksize);
987         sb->s_maxlen    = cpu_to_be32(journal->j_maxlen);
988         sb->s_first     = cpu_to_be32(1);
989
990         journal->j_transaction_sequence = 1;
991
992         journal->j_flags &= ~JFS_ABORT;
993         journal->j_format_version = 2;
994
995         return journal_reset(journal);
996 }
997
998 /**
999  * void journal_update_superblock() - Update journal sb on disk.
1000  * @journal: The journal to update.
1001  * @wait: Set to '0' if you don't want to wait for IO completion.
1002  *
1003  * Update a journal's dynamic superblock fields and write it to disk,
1004  * optionally waiting for the IO to complete.
1005  */
1006 void journal_update_superblock(journal_t *journal, int wait)
1007 {
1008         journal_superblock_t *sb = journal->j_superblock;
1009         struct buffer_head *bh = journal->j_sb_buffer;
1010
1011         /*
1012          * As a special case, if the on-disk copy is already marked as needing
1013          * no recovery (s_start == 0) and there are no outstanding transactions
1014          * in the filesystem, then we can safely defer the superblock update
1015          * until the next commit by setting JFS_FLUSHED.  This avoids
1016          * attempting a write to a potential-readonly device.
1017          */
1018         if (sb->s_start == 0 && journal->j_tail_sequence ==
1019                                 journal->j_transaction_sequence) {
1020                 jbd_debug(1,"JBD: Skipping superblock update on recovered sb "
1021                         "(start %u, seq %d, errno %d)\n",
1022                         journal->j_tail, journal->j_tail_sequence,
1023                         journal->j_errno);
1024                 goto out;
1025         }
1026
1027         if (buffer_write_io_error(bh)) {
1028                 char b[BDEVNAME_SIZE];
1029                 /*
1030                  * Oh, dear.  A previous attempt to write the journal
1031                  * superblock failed.  This could happen because the
1032                  * USB device was yanked out.  Or it could happen to
1033                  * be a transient write error and maybe the block will
1034                  * be remapped.  Nothing we can do but to retry the
1035                  * write and hope for the best.
1036                  */
1037                 printk(KERN_ERR "JBD: previous I/O error detected "
1038                        "for journal superblock update for %s.\n",
1039                        journal_dev_name(journal, b));
1040                 clear_buffer_write_io_error(bh);
1041                 set_buffer_uptodate(bh);
1042         }
1043
1044         spin_lock(&journal->j_state_lock);
1045         jbd_debug(1,"JBD: updating superblock (start %u, seq %d, errno %d)\n",
1046                   journal->j_tail, journal->j_tail_sequence, journal->j_errno);
1047
1048         sb->s_sequence = cpu_to_be32(journal->j_tail_sequence);
1049         sb->s_start    = cpu_to_be32(journal->j_tail);
1050         sb->s_errno    = cpu_to_be32(journal->j_errno);
1051         spin_unlock(&journal->j_state_lock);
1052
1053         BUFFER_TRACE(bh, "marking dirty");
1054         mark_buffer_dirty(bh);
1055         if (wait) {
1056                 sync_dirty_buffer(bh);
1057                 if (buffer_write_io_error(bh)) {
1058                         char b[BDEVNAME_SIZE];
1059                         printk(KERN_ERR "JBD: I/O error detected "
1060                                "when updating journal superblock for %s.\n",
1061                                journal_dev_name(journal, b));
1062                         clear_buffer_write_io_error(bh);
1063                         set_buffer_uptodate(bh);
1064                 }
1065         } else
1066                 write_dirty_buffer(bh, WRITE);
1067
1068 out:
1069         /* If we have just flushed the log (by marking s_start==0), then
1070          * any future commit will have to be careful to update the
1071          * superblock again to re-record the true start of the log. */
1072
1073         spin_lock(&journal->j_state_lock);
1074         if (sb->s_start)
1075                 journal->j_flags &= ~JFS_FLUSHED;
1076         else
1077                 journal->j_flags |= JFS_FLUSHED;
1078         spin_unlock(&journal->j_state_lock);
1079 }
1080
1081 /*
1082  * Read the superblock for a given journal, performing initial
1083  * validation of the format.
1084  */
1085
1086 static int journal_get_superblock(journal_t *journal)
1087 {
1088         struct buffer_head *bh;
1089         journal_superblock_t *sb;
1090         int err = -EIO;
1091
1092         bh = journal->j_sb_buffer;
1093
1094         J_ASSERT(bh != NULL);
1095         if (!buffer_uptodate(bh)) {
1096                 ll_rw_block(READ, 1, &bh);
1097                 wait_on_buffer(bh);
1098                 if (!buffer_uptodate(bh)) {
1099                         printk (KERN_ERR
1100                                 "JBD: IO error reading journal superblock\n");
1101                         goto out;
1102                 }
1103         }
1104
1105         sb = journal->j_superblock;
1106
1107         err = -EINVAL;
1108
1109         if (sb->s_header.h_magic != cpu_to_be32(JFS_MAGIC_NUMBER) ||
1110             sb->s_blocksize != cpu_to_be32(journal->j_blocksize)) {
1111                 printk(KERN_WARNING "JBD: no valid journal superblock found\n");
1112                 goto out;
1113         }
1114
1115         switch(be32_to_cpu(sb->s_header.h_blocktype)) {
1116         case JFS_SUPERBLOCK_V1:
1117                 journal->j_format_version = 1;
1118                 break;
1119         case JFS_SUPERBLOCK_V2:
1120                 journal->j_format_version = 2;
1121                 break;
1122         default:
1123                 printk(KERN_WARNING "JBD: unrecognised superblock format ID\n");
1124                 goto out;
1125         }
1126
1127         if (be32_to_cpu(sb->s_maxlen) < journal->j_maxlen)
1128                 journal->j_maxlen = be32_to_cpu(sb->s_maxlen);
1129         else if (be32_to_cpu(sb->s_maxlen) > journal->j_maxlen) {
1130                 printk (KERN_WARNING "JBD: journal file too short\n");
1131                 goto out;
1132         }
1133
1134         return 0;
1135
1136 out:
1137         journal_fail_superblock(journal);
1138         return err;
1139 }
1140
1141 /*
1142  * Load the on-disk journal superblock and read the key fields into the
1143  * journal_t.
1144  */
1145
1146 static int load_superblock(journal_t *journal)
1147 {
1148         int err;
1149         journal_superblock_t *sb;
1150
1151         err = journal_get_superblock(journal);
1152         if (err)
1153                 return err;
1154
1155         sb = journal->j_superblock;
1156
1157         journal->j_tail_sequence = be32_to_cpu(sb->s_sequence);
1158         journal->j_tail = be32_to_cpu(sb->s_start);
1159         journal->j_first = be32_to_cpu(sb->s_first);
1160         journal->j_last = be32_to_cpu(sb->s_maxlen);
1161         journal->j_errno = be32_to_cpu(sb->s_errno);
1162
1163         return 0;
1164 }
1165
1166
1167 /**
1168  * int journal_load() - Read journal from disk.
1169  * @journal: Journal to act on.
1170  *
1171  * Given a journal_t structure which tells us which disk blocks contain
1172  * a journal, read the journal from disk to initialise the in-memory
1173  * structures.
1174  */
1175 int journal_load(journal_t *journal)
1176 {
1177         int err;
1178         journal_superblock_t *sb;
1179
1180         err = load_superblock(journal);
1181         if (err)
1182                 return err;
1183
1184         sb = journal->j_superblock;
1185         /* If this is a V2 superblock, then we have to check the
1186          * features flags on it. */
1187
1188         if (journal->j_format_version >= 2) {
1189                 if ((sb->s_feature_ro_compat &
1190                      ~cpu_to_be32(JFS_KNOWN_ROCOMPAT_FEATURES)) ||
1191                     (sb->s_feature_incompat &
1192                      ~cpu_to_be32(JFS_KNOWN_INCOMPAT_FEATURES))) {
1193                         printk (KERN_WARNING
1194                                 "JBD: Unrecognised features on journal\n");
1195                         return -EINVAL;
1196                 }
1197         }
1198
1199         /* Let the recovery code check whether it needs to recover any
1200          * data from the journal. */
1201         if (journal_recover(journal))
1202                 goto recovery_error;
1203
1204         /* OK, we've finished with the dynamic journal bits:
1205          * reinitialise the dynamic contents of the superblock in memory
1206          * and reset them on disk. */
1207         if (journal_reset(journal))
1208                 goto recovery_error;
1209
1210         journal->j_flags &= ~JFS_ABORT;
1211         journal->j_flags |= JFS_LOADED;
1212         return 0;
1213
1214 recovery_error:
1215         printk (KERN_WARNING "JBD: recovery failed\n");
1216         return -EIO;
1217 }
1218
1219 /**
1220  * void journal_destroy() - Release a journal_t structure.
1221  * @journal: Journal to act on.
1222  *
1223  * Release a journal_t structure once it is no longer in use by the
1224  * journaled object.
1225  * Return <0 if we couldn't clean up the journal.
1226  */
1227 int journal_destroy(journal_t *journal)
1228 {
1229         int err = 0;
1230
1231         
1232         /* Wait for the commit thread to wake up and die. */
1233         journal_kill_thread(journal);
1234
1235         /* Force a final log commit */
1236         if (journal->j_running_transaction)
1237                 journal_commit_transaction(journal);
1238
1239         /* Force any old transactions to disk */
1240
1241         /* Totally anal locking here... */
1242         spin_lock(&journal->j_list_lock);
1243         while (journal->j_checkpoint_transactions != NULL) {
1244                 spin_unlock(&journal->j_list_lock);
1245                 log_do_checkpoint(journal);
1246                 spin_lock(&journal->j_list_lock);
1247         }
1248
1249         J_ASSERT(journal->j_running_transaction == NULL);
1250         J_ASSERT(journal->j_committing_transaction == NULL);
1251         J_ASSERT(journal->j_checkpoint_transactions == NULL);
1252         spin_unlock(&journal->j_list_lock);
1253
1254         if (journal->j_sb_buffer) {
1255                 if (!is_journal_aborted(journal)) {
1256                         /* We can now mark the journal as empty. */
1257                         journal->j_tail = 0;
1258                         journal->j_tail_sequence =
1259                                 ++journal->j_transaction_sequence;
1260                         journal_update_superblock(journal, 1);
1261                 } else {
1262                         err = -EIO;
1263                 }
1264                 brelse(journal->j_sb_buffer);
1265         }
1266
1267         if (journal->j_inode)
1268                 iput(journal->j_inode);
1269         if (journal->j_revoke)
1270                 journal_destroy_revoke(journal);
1271         kfree(journal->j_wbuf);
1272         kfree(journal);
1273
1274         return err;
1275 }
1276
1277
1278 /**
1279  *int journal_check_used_features () - Check if features specified are used.
1280  * @journal: Journal to check.
1281  * @compat: bitmask of compatible features
1282  * @ro: bitmask of features that force read-only mount
1283  * @incompat: bitmask of incompatible features
1284  *
1285  * Check whether the journal uses all of a given set of
1286  * features.  Return true (non-zero) if it does.
1287  **/
1288
1289 int journal_check_used_features (journal_t *journal, unsigned long compat,
1290                                  unsigned long ro, unsigned long incompat)
1291 {
1292         journal_superblock_t *sb;
1293
1294         if (!compat && !ro && !incompat)
1295                 return 1;
1296         if (journal->j_format_version == 1)
1297                 return 0;
1298
1299         sb = journal->j_superblock;
1300
1301         if (((be32_to_cpu(sb->s_feature_compat) & compat) == compat) &&
1302             ((be32_to_cpu(sb->s_feature_ro_compat) & ro) == ro) &&
1303             ((be32_to_cpu(sb->s_feature_incompat) & incompat) == incompat))
1304                 return 1;
1305
1306         return 0;
1307 }
1308
1309 /**
1310  * int journal_check_available_features() - Check feature set in journalling layer
1311  * @journal: Journal to check.
1312  * @compat: bitmask of compatible features
1313  * @ro: bitmask of features that force read-only mount
1314  * @incompat: bitmask of incompatible features
1315  *
1316  * Check whether the journaling code supports the use of
1317  * all of a given set of features on this journal.  Return true
1318  * (non-zero) if it can. */
1319
1320 int journal_check_available_features (journal_t *journal, unsigned long compat,
1321                                       unsigned long ro, unsigned long incompat)
1322 {
1323         if (!compat && !ro && !incompat)
1324                 return 1;
1325
1326         /* We can support any known requested features iff the
1327          * superblock is in version 2.  Otherwise we fail to support any
1328          * extended sb features. */
1329
1330         if (journal->j_format_version != 2)
1331                 return 0;
1332
1333         if ((compat   & JFS_KNOWN_COMPAT_FEATURES) == compat &&
1334             (ro       & JFS_KNOWN_ROCOMPAT_FEATURES) == ro &&
1335             (incompat & JFS_KNOWN_INCOMPAT_FEATURES) == incompat)
1336                 return 1;
1337
1338         return 0;
1339 }
1340
1341 /**
1342  * int journal_set_features () - Mark a given journal feature in the superblock
1343  * @journal: Journal to act on.
1344  * @compat: bitmask of compatible features
1345  * @ro: bitmask of features that force read-only mount
1346  * @incompat: bitmask of incompatible features
1347  *
1348  * Mark a given journal feature as present on the
1349  * superblock.  Returns true if the requested features could be set.
1350  *
1351  */
1352
1353 int journal_set_features (journal_t *journal, unsigned long compat,
1354                           unsigned long ro, unsigned long incompat)
1355 {
1356         journal_superblock_t *sb;
1357
1358         if (journal_check_used_features(journal, compat, ro, incompat))
1359                 return 1;
1360
1361         if (!journal_check_available_features(journal, compat, ro, incompat))
1362                 return 0;
1363
1364         jbd_debug(1, "Setting new features 0x%lx/0x%lx/0x%lx\n",
1365                   compat, ro, incompat);
1366
1367         sb = journal->j_superblock;
1368
1369         sb->s_feature_compat    |= cpu_to_be32(compat);
1370         sb->s_feature_ro_compat |= cpu_to_be32(ro);
1371         sb->s_feature_incompat  |= cpu_to_be32(incompat);
1372
1373         return 1;
1374 }
1375
1376
1377 /**
1378  * int journal_update_format () - Update on-disk journal structure.
1379  * @journal: Journal to act on.
1380  *
1381  * Given an initialised but unloaded journal struct, poke about in the
1382  * on-disk structure to update it to the most recent supported version.
1383  */
1384 int journal_update_format (journal_t *journal)
1385 {
1386         journal_superblock_t *sb;
1387         int err;
1388
1389         err = journal_get_superblock(journal);
1390         if (err)
1391                 return err;
1392
1393         sb = journal->j_superblock;
1394
1395         switch (be32_to_cpu(sb->s_header.h_blocktype)) {
1396         case JFS_SUPERBLOCK_V2:
1397                 return 0;
1398         case JFS_SUPERBLOCK_V1:
1399                 return journal_convert_superblock_v1(journal, sb);
1400         default:
1401                 break;
1402         }
1403         return -EINVAL;
1404 }
1405
1406 static int journal_convert_superblock_v1(journal_t *journal,
1407                                          journal_superblock_t *sb)
1408 {
1409         int offset, blocksize;
1410         struct buffer_head *bh;
1411
1412         printk(KERN_WARNING
1413                 "JBD: Converting superblock from version 1 to 2.\n");
1414
1415         /* Pre-initialise new fields to zero */
1416         offset = ((char *) &(sb->s_feature_compat)) - ((char *) sb);
1417         blocksize = be32_to_cpu(sb->s_blocksize);
1418         memset(&sb->s_feature_compat, 0, blocksize-offset);
1419
1420         sb->s_nr_users = cpu_to_be32(1);
1421         sb->s_header.h_blocktype = cpu_to_be32(JFS_SUPERBLOCK_V2);
1422         journal->j_format_version = 2;
1423
1424         bh = journal->j_sb_buffer;
1425         BUFFER_TRACE(bh, "marking dirty");
1426         mark_buffer_dirty(bh);
1427         sync_dirty_buffer(bh);
1428         return 0;
1429 }
1430
1431
1432 /**
1433  * int journal_flush () - Flush journal
1434  * @journal: Journal to act on.
1435  *
1436  * Flush all data for a given journal to disk and empty the journal.
1437  * Filesystems can use this when remounting readonly to ensure that
1438  * recovery does not need to happen on remount.
1439  */
1440
1441 int journal_flush(journal_t *journal)
1442 {
1443         int err = 0;
1444         transaction_t *transaction = NULL;
1445         unsigned int old_tail;
1446
1447         spin_lock(&journal->j_state_lock);
1448
1449         /* Force everything buffered to the log... */
1450         if (journal->j_running_transaction) {
1451                 transaction = journal->j_running_transaction;
1452                 __log_start_commit(journal, transaction->t_tid);
1453         } else if (journal->j_committing_transaction)
1454                 transaction = journal->j_committing_transaction;
1455
1456         /* Wait for the log commit to complete... */
1457         if (transaction) {
1458                 tid_t tid = transaction->t_tid;
1459
1460                 spin_unlock(&journal->j_state_lock);
1461                 log_wait_commit(journal, tid);
1462         } else {
1463                 spin_unlock(&journal->j_state_lock);
1464         }
1465
1466         /* ...and flush everything in the log out to disk. */
1467         spin_lock(&journal->j_list_lock);
1468         while (!err && journal->j_checkpoint_transactions != NULL) {
1469                 spin_unlock(&journal->j_list_lock);
1470                 mutex_lock(&journal->j_checkpoint_mutex);
1471                 err = log_do_checkpoint(journal);
1472                 mutex_unlock(&journal->j_checkpoint_mutex);
1473                 spin_lock(&journal->j_list_lock);
1474         }
1475         spin_unlock(&journal->j_list_lock);
1476
1477         if (is_journal_aborted(journal))
1478                 return -EIO;
1479
1480         cleanup_journal_tail(journal);
1481
1482         /* Finally, mark the journal as really needing no recovery.
1483          * This sets s_start==0 in the underlying superblock, which is
1484          * the magic code for a fully-recovered superblock.  Any future
1485          * commits of data to the journal will restore the current
1486          * s_start value. */
1487         spin_lock(&journal->j_state_lock);
1488         old_tail = journal->j_tail;
1489         journal->j_tail = 0;
1490         spin_unlock(&journal->j_state_lock);
1491         journal_update_superblock(journal, 1);
1492         spin_lock(&journal->j_state_lock);
1493         journal->j_tail = old_tail;
1494
1495         J_ASSERT(!journal->j_running_transaction);
1496         J_ASSERT(!journal->j_committing_transaction);
1497         J_ASSERT(!journal->j_checkpoint_transactions);
1498         J_ASSERT(journal->j_head == journal->j_tail);
1499         J_ASSERT(journal->j_tail_sequence == journal->j_transaction_sequence);
1500         spin_unlock(&journal->j_state_lock);
1501         return 0;
1502 }
1503
1504 /**
1505  * int journal_wipe() - Wipe journal contents
1506  * @journal: Journal to act on.
1507  * @write: flag (see below)
1508  *
1509  * Wipe out all of the contents of a journal, safely.  This will produce
1510  * a warning if the journal contains any valid recovery information.
1511  * Must be called between journal_init_*() and journal_load().
1512  *
1513  * If 'write' is non-zero, then we wipe out the journal on disk; otherwise
1514  * we merely suppress recovery.
1515  */
1516
1517 int journal_wipe(journal_t *journal, int write)
1518 {
1519         int err = 0;
1520
1521         J_ASSERT (!(journal->j_flags & JFS_LOADED));
1522
1523         err = load_superblock(journal);
1524         if (err)
1525                 return err;
1526
1527         if (!journal->j_tail)
1528                 goto no_recovery;
1529
1530         printk (KERN_WARNING "JBD: %s recovery information on journal\n",
1531                 write ? "Clearing" : "Ignoring");
1532
1533         err = journal_skip_recovery(journal);
1534         if (write)
1535                 journal_update_superblock(journal, 1);
1536
1537  no_recovery:
1538         return err;
1539 }
1540
1541 /*
1542  * journal_dev_name: format a character string to describe on what
1543  * device this journal is present.
1544  */
1545
1546 static const char *journal_dev_name(journal_t *journal, char *buffer)
1547 {
1548         struct block_device *bdev;
1549
1550         if (journal->j_inode)
1551                 bdev = journal->j_inode->i_sb->s_bdev;
1552         else
1553                 bdev = journal->j_dev;
1554
1555         return bdevname(bdev, buffer);
1556 }
1557
1558 /*
1559  * Journal abort has very specific semantics, which we describe
1560  * for journal abort.
1561  *
1562  * Two internal function, which provide abort to te jbd layer
1563  * itself are here.
1564  */
1565
1566 /*
1567  * Quick version for internal journal use (doesn't lock the journal).
1568  * Aborts hard --- we mark the abort as occurred, but do _nothing_ else,
1569  * and don't attempt to make any other journal updates.
1570  */
1571 static void __journal_abort_hard(journal_t *journal)
1572 {
1573         transaction_t *transaction;
1574         char b[BDEVNAME_SIZE];
1575
1576         if (journal->j_flags & JFS_ABORT)
1577                 return;
1578
1579         printk(KERN_ERR "Aborting journal on device %s.\n",
1580                 journal_dev_name(journal, b));
1581
1582         spin_lock(&journal->j_state_lock);
1583         journal->j_flags |= JFS_ABORT;
1584         transaction = journal->j_running_transaction;
1585         if (transaction)
1586                 __log_start_commit(journal, transaction->t_tid);
1587         spin_unlock(&journal->j_state_lock);
1588 }
1589
1590 /* Soft abort: record the abort error status in the journal superblock,
1591  * but don't do any other IO. */
1592 static void __journal_abort_soft (journal_t *journal, int errno)
1593 {
1594         if (journal->j_flags & JFS_ABORT)
1595                 return;
1596
1597         if (!journal->j_errno)
1598                 journal->j_errno = errno;
1599
1600         __journal_abort_hard(journal);
1601
1602         if (errno)
1603                 journal_update_superblock(journal, 1);
1604 }
1605
1606 /**
1607  * void journal_abort () - Shutdown the journal immediately.
1608  * @journal: the journal to shutdown.
1609  * @errno:   an error number to record in the journal indicating
1610  *           the reason for the shutdown.
1611  *
1612  * Perform a complete, immediate shutdown of the ENTIRE
1613  * journal (not of a single transaction).  This operation cannot be
1614  * undone without closing and reopening the journal.
1615  *
1616  * The journal_abort function is intended to support higher level error
1617  * recovery mechanisms such as the ext2/ext3 remount-readonly error
1618  * mode.
1619  *
1620  * Journal abort has very specific semantics.  Any existing dirty,
1621  * unjournaled buffers in the main filesystem will still be written to
1622  * disk by bdflush, but the journaling mechanism will be suspended
1623  * immediately and no further transaction commits will be honoured.
1624  *
1625  * Any dirty, journaled buffers will be written back to disk without
1626  * hitting the journal.  Atomicity cannot be guaranteed on an aborted
1627  * filesystem, but we _do_ attempt to leave as much data as possible
1628  * behind for fsck to use for cleanup.
1629  *
1630  * Any attempt to get a new transaction handle on a journal which is in
1631  * ABORT state will just result in an -EROFS error return.  A
1632  * journal_stop on an existing handle will return -EIO if we have
1633  * entered abort state during the update.
1634  *
1635  * Recursive transactions are not disturbed by journal abort until the
1636  * final journal_stop, which will receive the -EIO error.
1637  *
1638  * Finally, the journal_abort call allows the caller to supply an errno
1639  * which will be recorded (if possible) in the journal superblock.  This
1640  * allows a client to record failure conditions in the middle of a
1641  * transaction without having to complete the transaction to record the
1642  * failure to disk.  ext3_error, for example, now uses this
1643  * functionality.
1644  *
1645  * Errors which originate from within the journaling layer will NOT
1646  * supply an errno; a null errno implies that absolutely no further
1647  * writes are done to the journal (unless there are any already in
1648  * progress).
1649  *
1650  */
1651
1652 void journal_abort(journal_t *journal, int errno)
1653 {
1654         __journal_abort_soft(journal, errno);
1655 }
1656
1657 /**
1658  * int journal_errno () - returns the journal's error state.
1659  * @journal: journal to examine.
1660  *
1661  * This is the errno numbet set with journal_abort(), the last
1662  * time the journal was mounted - if the journal was stopped
1663  * without calling abort this will be 0.
1664  *
1665  * If the journal has been aborted on this mount time -EROFS will
1666  * be returned.
1667  */
1668 int journal_errno(journal_t *journal)
1669 {
1670         int err;
1671
1672         spin_lock(&journal->j_state_lock);
1673         if (journal->j_flags & JFS_ABORT)
1674                 err = -EROFS;
1675         else
1676                 err = journal->j_errno;
1677         spin_unlock(&journal->j_state_lock);
1678         return err;
1679 }
1680
1681 /**
1682  * int journal_clear_err () - clears the journal's error state
1683  * @journal: journal to act on.
1684  *
1685  * An error must be cleared or Acked to take a FS out of readonly
1686  * mode.
1687  */
1688 int journal_clear_err(journal_t *journal)
1689 {
1690         int err = 0;
1691
1692         spin_lock(&journal->j_state_lock);
1693         if (journal->j_flags & JFS_ABORT)
1694                 err = -EROFS;
1695         else
1696                 journal->j_errno = 0;
1697         spin_unlock(&journal->j_state_lock);
1698         return err;
1699 }
1700
1701 /**
1702  * void journal_ack_err() - Ack journal err.
1703  * @journal: journal to act on.
1704  *
1705  * An error must be cleared or Acked to take a FS out of readonly
1706  * mode.
1707  */
1708 void journal_ack_err(journal_t *journal)
1709 {
1710         spin_lock(&journal->j_state_lock);
1711         if (journal->j_errno)
1712                 journal->j_flags |= JFS_ACK_ERR;
1713         spin_unlock(&journal->j_state_lock);
1714 }
1715
1716 int journal_blocks_per_page(struct inode *inode)
1717 {
1718         return 1 << (PAGE_CACHE_SHIFT - inode->i_sb->s_blocksize_bits);
1719 }
1720
1721 /*
1722  * Journal_head storage management
1723  */
1724 static struct kmem_cache *journal_head_cache;
1725 #ifdef CONFIG_JBD_DEBUG
1726 static atomic_t nr_journal_heads = ATOMIC_INIT(0);
1727 #endif
1728
1729 static int journal_init_journal_head_cache(void)
1730 {
1731         int retval;
1732
1733         J_ASSERT(journal_head_cache == NULL);
1734         journal_head_cache = kmem_cache_create("journal_head",
1735                                 sizeof(struct journal_head),
1736                                 0,              /* offset */
1737                                 SLAB_TEMPORARY, /* flags */
1738                                 NULL);          /* ctor */
1739         retval = 0;
1740         if (!journal_head_cache) {
1741                 retval = -ENOMEM;
1742                 printk(KERN_EMERG "JBD: no memory for journal_head cache\n");
1743         }
1744         return retval;
1745 }
1746
1747 static void journal_destroy_journal_head_cache(void)
1748 {
1749         if (journal_head_cache) {
1750                 kmem_cache_destroy(journal_head_cache);
1751                 journal_head_cache = NULL;
1752         }
1753 }
1754
1755 /*
1756  * journal_head splicing and dicing
1757  */
1758 static struct journal_head *journal_alloc_journal_head(void)
1759 {
1760         struct journal_head *ret;
1761
1762 #ifdef CONFIG_JBD_DEBUG
1763         atomic_inc(&nr_journal_heads);
1764 #endif
1765         ret = kmem_cache_alloc(journal_head_cache, GFP_NOFS);
1766         if (ret == NULL) {
1767                 jbd_debug(1, "out of memory for journal_head\n");
1768                 printk_ratelimited(KERN_NOTICE "ENOMEM in %s, retrying.\n",
1769                                    __func__);
1770
1771                 while (ret == NULL) {
1772                         yield();
1773                         ret = kmem_cache_alloc(journal_head_cache, GFP_NOFS);
1774                 }
1775         }
1776         return ret;
1777 }
1778
1779 static void journal_free_journal_head(struct journal_head *jh)
1780 {
1781 #ifdef CONFIG_JBD_DEBUG
1782         atomic_dec(&nr_journal_heads);
1783         memset(jh, JBD_POISON_FREE, sizeof(*jh));
1784 #endif
1785         kmem_cache_free(journal_head_cache, jh);
1786 }
1787
1788 /*
1789  * A journal_head is attached to a buffer_head whenever JBD has an
1790  * interest in the buffer.
1791  *
1792  * Whenever a buffer has an attached journal_head, its ->b_state:BH_JBD bit
1793  * is set.  This bit is tested in core kernel code where we need to take
1794  * JBD-specific actions.  Testing the zeroness of ->b_private is not reliable
1795  * there.
1796  *
1797  * When a buffer has its BH_JBD bit set, its ->b_count is elevated by one.
1798  *
1799  * When a buffer has its BH_JBD bit set it is immune from being released by
1800  * core kernel code, mainly via ->b_count.
1801  *
1802  * A journal_head may be detached from its buffer_head when the journal_head's
1803  * b_transaction, b_cp_transaction and b_next_transaction pointers are NULL.
1804  * Various places in JBD call journal_remove_journal_head() to indicate that the
1805  * journal_head can be dropped if needed.
1806  *
1807  * Various places in the kernel want to attach a journal_head to a buffer_head
1808  * _before_ attaching the journal_head to a transaction.  To protect the
1809  * journal_head in this situation, journal_add_journal_head elevates the
1810  * journal_head's b_jcount refcount by one.  The caller must call
1811  * journal_put_journal_head() to undo this.
1812  *
1813  * So the typical usage would be:
1814  *
1815  *      (Attach a journal_head if needed.  Increments b_jcount)
1816  *      struct journal_head *jh = journal_add_journal_head(bh);
1817  *      ...
1818  *      jh->b_transaction = xxx;
1819  *      journal_put_journal_head(jh);
1820  *
1821  * Now, the journal_head's b_jcount is zero, but it is safe from being released
1822  * because it has a non-zero b_transaction.
1823  */
1824
1825 /*
1826  * Give a buffer_head a journal_head.
1827  *
1828  * Doesn't need the journal lock.
1829  * May sleep.
1830  */
1831 struct journal_head *journal_add_journal_head(struct buffer_head *bh)
1832 {
1833         struct journal_head *jh;
1834         struct journal_head *new_jh = NULL;
1835
1836 repeat:
1837         if (!buffer_jbd(bh)) {
1838                 new_jh = journal_alloc_journal_head();
1839                 memset(new_jh, 0, sizeof(*new_jh));
1840         }
1841
1842         jbd_lock_bh_journal_head(bh);
1843         if (buffer_jbd(bh)) {
1844                 jh = bh2jh(bh);
1845         } else {
1846                 J_ASSERT_BH(bh,
1847                         (atomic_read(&bh->b_count) > 0) ||
1848                         (bh->b_page && bh->b_page->mapping));
1849
1850                 if (!new_jh) {
1851                         jbd_unlock_bh_journal_head(bh);
1852                         goto repeat;
1853                 }
1854
1855                 jh = new_jh;
1856                 new_jh = NULL;          /* We consumed it */
1857                 set_buffer_jbd(bh);
1858                 bh->b_private = jh;
1859                 jh->b_bh = bh;
1860                 get_bh(bh);
1861                 BUFFER_TRACE(bh, "added journal_head");
1862         }
1863         jh->b_jcount++;
1864         jbd_unlock_bh_journal_head(bh);
1865         if (new_jh)
1866                 journal_free_journal_head(new_jh);
1867         return bh->b_private;
1868 }
1869
1870 /*
1871  * Grab a ref against this buffer_head's journal_head.  If it ended up not
1872  * having a journal_head, return NULL
1873  */
1874 struct journal_head *journal_grab_journal_head(struct buffer_head *bh)
1875 {
1876         struct journal_head *jh = NULL;
1877
1878         jbd_lock_bh_journal_head(bh);
1879         if (buffer_jbd(bh)) {
1880                 jh = bh2jh(bh);
1881                 jh->b_jcount++;
1882         }
1883         jbd_unlock_bh_journal_head(bh);
1884         return jh;
1885 }
1886
1887 static void __journal_remove_journal_head(struct buffer_head *bh)
1888 {
1889         struct journal_head *jh = bh2jh(bh);
1890
1891         J_ASSERT_JH(jh, jh->b_jcount >= 0);
1892
1893         get_bh(bh);
1894         if (jh->b_jcount == 0) {
1895                 if (jh->b_transaction == NULL &&
1896                                 jh->b_next_transaction == NULL &&
1897                                 jh->b_cp_transaction == NULL) {
1898                         J_ASSERT_JH(jh, jh->b_jlist == BJ_None);
1899                         J_ASSERT_BH(bh, buffer_jbd(bh));
1900                         J_ASSERT_BH(bh, jh2bh(jh) == bh);
1901                         BUFFER_TRACE(bh, "remove journal_head");
1902                         if (jh->b_frozen_data) {
1903                                 printk(KERN_WARNING "%s: freeing "
1904                                                 "b_frozen_data\n",
1905                                                 __func__);
1906                                 jbd_free(jh->b_frozen_data, bh->b_size);
1907                         }
1908                         if (jh->b_committed_data) {
1909                                 printk(KERN_WARNING "%s: freeing "
1910                                                 "b_committed_data\n",
1911                                                 __func__);
1912                                 jbd_free(jh->b_committed_data, bh->b_size);
1913                         }
1914                         bh->b_private = NULL;
1915                         jh->b_bh = NULL;        /* debug, really */
1916                         clear_buffer_jbd(bh);
1917                         __brelse(bh);
1918                         journal_free_journal_head(jh);
1919                 } else {
1920                         BUFFER_TRACE(bh, "journal_head was locked");
1921                 }
1922         }
1923 }
1924
1925 /*
1926  * journal_remove_journal_head(): if the buffer isn't attached to a transaction
1927  * and has a zero b_jcount then remove and release its journal_head.   If we did
1928  * see that the buffer is not used by any transaction we also "logically"
1929  * decrement ->b_count.
1930  *
1931  * We in fact take an additional increment on ->b_count as a convenience,
1932  * because the caller usually wants to do additional things with the bh
1933  * after calling here.
1934  * The caller of journal_remove_journal_head() *must* run __brelse(bh) at some
1935  * time.  Once the caller has run __brelse(), the buffer is eligible for
1936  * reaping by try_to_free_buffers().
1937  */
1938 void journal_remove_journal_head(struct buffer_head *bh)
1939 {
1940         jbd_lock_bh_journal_head(bh);
1941         __journal_remove_journal_head(bh);
1942         jbd_unlock_bh_journal_head(bh);
1943 }
1944
1945 /*
1946  * Drop a reference on the passed journal_head.  If it fell to zero then try to
1947  * release the journal_head from the buffer_head.
1948  */
1949 void journal_put_journal_head(struct journal_head *jh)
1950 {
1951         struct buffer_head *bh = jh2bh(jh);
1952
1953         jbd_lock_bh_journal_head(bh);
1954         J_ASSERT_JH(jh, jh->b_jcount > 0);
1955         --jh->b_jcount;
1956         if (!jh->b_jcount && !jh->b_transaction) {
1957                 __journal_remove_journal_head(bh);
1958                 __brelse(bh);
1959         }
1960         jbd_unlock_bh_journal_head(bh);
1961 }
1962
1963 /*
1964  * debugfs tunables
1965  */
1966 #ifdef CONFIG_JBD_DEBUG
1967
1968 u8 journal_enable_debug __read_mostly;
1969 EXPORT_SYMBOL(journal_enable_debug);
1970
1971 static struct dentry *jbd_debugfs_dir;
1972 static struct dentry *jbd_debug;
1973
1974 static void __init jbd_create_debugfs_entry(void)
1975 {
1976         jbd_debugfs_dir = debugfs_create_dir("jbd", NULL);
1977         if (jbd_debugfs_dir)
1978                 jbd_debug = debugfs_create_u8("jbd-debug", S_IRUGO | S_IWUSR,
1979                                                jbd_debugfs_dir,
1980                                                &journal_enable_debug);
1981 }
1982
1983 static void __exit jbd_remove_debugfs_entry(void)
1984 {
1985         debugfs_remove(jbd_debug);
1986         debugfs_remove(jbd_debugfs_dir);
1987 }
1988
1989 #else
1990
1991 static inline void jbd_create_debugfs_entry(void)
1992 {
1993 }
1994
1995 static inline void jbd_remove_debugfs_entry(void)
1996 {
1997 }
1998
1999 #endif
2000
2001 struct kmem_cache *jbd_handle_cache;
2002
2003 static int __init journal_init_handle_cache(void)
2004 {
2005         jbd_handle_cache = kmem_cache_create("journal_handle",
2006                                 sizeof(handle_t),
2007                                 0,              /* offset */
2008                                 SLAB_TEMPORARY, /* flags */
2009                                 NULL);          /* ctor */
2010         if (jbd_handle_cache == NULL) {
2011                 printk(KERN_EMERG "JBD: failed to create handle cache\n");
2012                 return -ENOMEM;
2013         }
2014         return 0;
2015 }
2016
2017 static void journal_destroy_handle_cache(void)
2018 {
2019         if (jbd_handle_cache)
2020                 kmem_cache_destroy(jbd_handle_cache);
2021 }
2022
2023 /*
2024  * Module startup and shutdown
2025  */
2026
2027 static int __init journal_init_caches(void)
2028 {
2029         int ret;
2030
2031         ret = journal_init_revoke_caches();
2032         if (ret == 0)
2033                 ret = journal_init_journal_head_cache();
2034         if (ret == 0)
2035                 ret = journal_init_handle_cache();
2036         return ret;
2037 }
2038
2039 static void journal_destroy_caches(void)
2040 {
2041         journal_destroy_revoke_caches();
2042         journal_destroy_journal_head_cache();
2043         journal_destroy_handle_cache();
2044 }
2045
2046 static int __init journal_init(void)
2047 {
2048         int ret;
2049
2050         BUILD_BUG_ON(sizeof(struct journal_superblock_s) != 1024);
2051
2052         ret = journal_init_caches();
2053         if (ret != 0)
2054                 journal_destroy_caches();
2055         jbd_create_debugfs_entry();
2056         return ret;
2057 }
2058
2059 static void __exit journal_exit(void)
2060 {
2061 #ifdef CONFIG_JBD_DEBUG
2062         int n = atomic_read(&nr_journal_heads);
2063         if (n)
2064                 printk(KERN_EMERG "JBD: leaked %d journal_heads!\n", n);
2065 #endif
2066         jbd_remove_debugfs_entry();
2067         journal_destroy_caches();
2068 }
2069
2070 MODULE_LICENSE("GPL");
2071 module_init(journal_init);
2072 module_exit(journal_exit);
2073