jbd2: Add ENOMEM checking in and for jbd2_journal_write_metadata_buffer()
[linux-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                 if (!tmp) {
362                         jbd2_journal_put_journal_head(new_jh);
363                         return -ENOMEM;
364                 }
365                 jbd_lock_bh_state(bh_in);
366                 if (jh_in->b_frozen_data) {
367                         jbd2_free(tmp, bh_in->b_size);
368                         goto repeat;
369                 }
370
371                 jh_in->b_frozen_data = tmp;
372                 mapped_data = kmap_atomic(new_page, KM_USER0);
373                 memcpy(tmp, mapped_data + new_offset, jh2bh(jh_in)->b_size);
374                 kunmap_atomic(mapped_data, KM_USER0);
375
376                 new_page = virt_to_page(tmp);
377                 new_offset = offset_in_page(tmp);
378                 done_copy_out = 1;
379
380                 /*
381                  * This isn't strictly necessary, as we're using frozen
382                  * data for the escaping, but it keeps consistency with
383                  * b_frozen_data usage.
384                  */
385                 jh_in->b_frozen_triggers = jh_in->b_triggers;
386         }
387
388         /*
389          * Did we need to do an escaping?  Now we've done all the
390          * copying, we can finally do so.
391          */
392         if (do_escape) {
393                 mapped_data = kmap_atomic(new_page, KM_USER0);
394                 *((unsigned int *)(mapped_data + new_offset)) = 0;
395                 kunmap_atomic(mapped_data, KM_USER0);
396         }
397
398         set_bh_page(new_bh, new_page, new_offset);
399         new_jh->b_transaction = NULL;
400         new_bh->b_size = jh2bh(jh_in)->b_size;
401         new_bh->b_bdev = transaction->t_journal->j_dev;
402         new_bh->b_blocknr = blocknr;
403         set_buffer_mapped(new_bh);
404         set_buffer_dirty(new_bh);
405
406         *jh_out = new_jh;
407
408         /*
409          * The to-be-written buffer needs to get moved to the io queue,
410          * and the original buffer whose contents we are shadowing or
411          * copying is moved to the transaction's shadow queue.
412          */
413         JBUFFER_TRACE(jh_in, "file as BJ_Shadow");
414         spin_lock(&journal->j_list_lock);
415         __jbd2_journal_file_buffer(jh_in, transaction, BJ_Shadow);
416         spin_unlock(&journal->j_list_lock);
417         jbd_unlock_bh_state(bh_in);
418
419         JBUFFER_TRACE(new_jh, "file as BJ_IO");
420         jbd2_journal_file_buffer(new_jh, transaction, BJ_IO);
421
422         return do_escape | (done_copy_out << 1);
423 }
424
425 /*
426  * Allocation code for the journal file.  Manage the space left in the
427  * journal, so that we can begin checkpointing when appropriate.
428  */
429
430 /*
431  * __jbd2_log_space_left: Return the number of free blocks left in the journal.
432  *
433  * Called with the journal already locked.
434  *
435  * Called under j_state_lock
436  */
437
438 int __jbd2_log_space_left(journal_t *journal)
439 {
440         int left = journal->j_free;
441
442         assert_spin_locked(&journal->j_state_lock);
443
444         /*
445          * Be pessimistic here about the number of those free blocks which
446          * might be required for log descriptor control blocks.
447          */
448
449 #define MIN_LOG_RESERVED_BLOCKS 32 /* Allow for rounding errors */
450
451         left -= MIN_LOG_RESERVED_BLOCKS;
452
453         if (left <= 0)
454                 return 0;
455         left -= (left >> 3);
456         return left;
457 }
458
459 /*
460  * Called under j_state_lock.  Returns true if a transaction commit was started.
461  */
462 int __jbd2_log_start_commit(journal_t *journal, tid_t target)
463 {
464         /*
465          * Are we already doing a recent enough commit?
466          */
467         if (!tid_geq(journal->j_commit_request, target)) {
468                 /*
469                  * We want a new commit: OK, mark the request and wakup the
470                  * commit thread.  We do _not_ do the commit ourselves.
471                  */
472
473                 journal->j_commit_request = target;
474                 jbd_debug(1, "JBD: requesting commit %d/%d\n",
475                           journal->j_commit_request,
476                           journal->j_commit_sequence);
477                 wake_up(&journal->j_wait_commit);
478                 return 1;
479         }
480         return 0;
481 }
482
483 int jbd2_log_start_commit(journal_t *journal, tid_t tid)
484 {
485         int ret;
486
487         spin_lock(&journal->j_state_lock);
488         ret = __jbd2_log_start_commit(journal, tid);
489         spin_unlock(&journal->j_state_lock);
490         return ret;
491 }
492
493 /*
494  * Force and wait upon a commit if the calling process is not within
495  * transaction.  This is used for forcing out undo-protected data which contains
496  * bitmaps, when the fs is running out of space.
497  *
498  * We can only force the running transaction if we don't have an active handle;
499  * otherwise, we will deadlock.
500  *
501  * Returns true if a transaction was started.
502  */
503 int jbd2_journal_force_commit_nested(journal_t *journal)
504 {
505         transaction_t *transaction = NULL;
506         tid_t tid;
507
508         spin_lock(&journal->j_state_lock);
509         if (journal->j_running_transaction && !current->journal_info) {
510                 transaction = journal->j_running_transaction;
511                 __jbd2_log_start_commit(journal, transaction->t_tid);
512         } else if (journal->j_committing_transaction)
513                 transaction = journal->j_committing_transaction;
514
515         if (!transaction) {
516                 spin_unlock(&journal->j_state_lock);
517                 return 0;       /* Nothing to retry */
518         }
519
520         tid = transaction->t_tid;
521         spin_unlock(&journal->j_state_lock);
522         jbd2_log_wait_commit(journal, tid);
523         return 1;
524 }
525
526 /*
527  * Start a commit of the current running transaction (if any).  Returns true
528  * if a transaction is going to be committed (or is currently already
529  * committing), and fills its tid in at *ptid
530  */
531 int jbd2_journal_start_commit(journal_t *journal, tid_t *ptid)
532 {
533         int ret = 0;
534
535         spin_lock(&journal->j_state_lock);
536         if (journal->j_running_transaction) {
537                 tid_t tid = journal->j_running_transaction->t_tid;
538
539                 __jbd2_log_start_commit(journal, tid);
540                 /* There's a running transaction and we've just made sure
541                  * it's commit has been scheduled. */
542                 if (ptid)
543                         *ptid = tid;
544                 ret = 1;
545         } else if (journal->j_committing_transaction) {
546                 /*
547                  * If ext3_write_super() recently started a commit, then we
548                  * have to wait for completion of that transaction
549                  */
550                 if (ptid)
551                         *ptid = journal->j_committing_transaction->t_tid;
552                 ret = 1;
553         }
554         spin_unlock(&journal->j_state_lock);
555         return ret;
556 }
557
558 /*
559  * Wait for a specified commit to complete.
560  * The caller may not hold the journal lock.
561  */
562 int jbd2_log_wait_commit(journal_t *journal, tid_t tid)
563 {
564         int err = 0;
565
566 #ifdef CONFIG_JBD2_DEBUG
567         spin_lock(&journal->j_state_lock);
568         if (!tid_geq(journal->j_commit_request, tid)) {
569                 printk(KERN_EMERG
570                        "%s: error: j_commit_request=%d, tid=%d\n",
571                        __func__, journal->j_commit_request, tid);
572         }
573         spin_unlock(&journal->j_state_lock);
574 #endif
575         spin_lock(&journal->j_state_lock);
576         while (tid_gt(tid, journal->j_commit_sequence)) {
577                 jbd_debug(1, "JBD: want %d, j_commit_sequence=%d\n",
578                                   tid, journal->j_commit_sequence);
579                 wake_up(&journal->j_wait_commit);
580                 spin_unlock(&journal->j_state_lock);
581                 wait_event(journal->j_wait_done_commit,
582                                 !tid_gt(tid, journal->j_commit_sequence));
583                 spin_lock(&journal->j_state_lock);
584         }
585         spin_unlock(&journal->j_state_lock);
586
587         if (unlikely(is_journal_aborted(journal))) {
588                 printk(KERN_EMERG "journal commit I/O error\n");
589                 err = -EIO;
590         }
591         return err;
592 }
593
594 /*
595  * Log buffer allocation routines:
596  */
597
598 int jbd2_journal_next_log_block(journal_t *journal, unsigned long long *retp)
599 {
600         unsigned long blocknr;
601
602         spin_lock(&journal->j_state_lock);
603         J_ASSERT(journal->j_free > 1);
604
605         blocknr = journal->j_head;
606         journal->j_head++;
607         journal->j_free--;
608         if (journal->j_head == journal->j_last)
609                 journal->j_head = journal->j_first;
610         spin_unlock(&journal->j_state_lock);
611         return jbd2_journal_bmap(journal, blocknr, retp);
612 }
613
614 /*
615  * Conversion of logical to physical block numbers for the journal
616  *
617  * On external journals the journal blocks are identity-mapped, so
618  * this is a no-op.  If needed, we can use j_blk_offset - everything is
619  * ready.
620  */
621 int jbd2_journal_bmap(journal_t *journal, unsigned long blocknr,
622                  unsigned long long *retp)
623 {
624         int err = 0;
625         unsigned long long ret;
626
627         if (journal->j_inode) {
628                 ret = bmap(journal->j_inode, blocknr);
629                 if (ret)
630                         *retp = ret;
631                 else {
632                         printk(KERN_ALERT "%s: journal block not found "
633                                         "at offset %lu on %s\n",
634                                __func__, blocknr, journal->j_devname);
635                         err = -EIO;
636                         __journal_abort_soft(journal, err);
637                 }
638         } else {
639                 *retp = blocknr; /* +journal->j_blk_offset */
640         }
641         return err;
642 }
643
644 /*
645  * We play buffer_head aliasing tricks to write data/metadata blocks to
646  * the journal without copying their contents, but for journal
647  * descriptor blocks we do need to generate bona fide buffers.
648  *
649  * After the caller of jbd2_journal_get_descriptor_buffer() has finished modifying
650  * the buffer's contents they really should run flush_dcache_page(bh->b_page).
651  * But we don't bother doing that, so there will be coherency problems with
652  * mmaps of blockdevs which hold live JBD-controlled filesystems.
653  */
654 struct journal_head *jbd2_journal_get_descriptor_buffer(journal_t *journal)
655 {
656         struct buffer_head *bh;
657         unsigned long long blocknr;
658         int err;
659
660         err = jbd2_journal_next_log_block(journal, &blocknr);
661
662         if (err)
663                 return NULL;
664
665         bh = __getblk(journal->j_dev, blocknr, journal->j_blocksize);
666         if (!bh)
667                 return NULL;
668         lock_buffer(bh);
669         memset(bh->b_data, 0, journal->j_blocksize);
670         set_buffer_uptodate(bh);
671         unlock_buffer(bh);
672         BUFFER_TRACE(bh, "return this buffer");
673         return jbd2_journal_add_journal_head(bh);
674 }
675
676 struct jbd2_stats_proc_session {
677         journal_t *journal;
678         struct transaction_stats_s *stats;
679         int start;
680         int max;
681 };
682
683 static void *jbd2_seq_info_start(struct seq_file *seq, loff_t *pos)
684 {
685         return *pos ? NULL : SEQ_START_TOKEN;
686 }
687
688 static void *jbd2_seq_info_next(struct seq_file *seq, void *v, loff_t *pos)
689 {
690         return NULL;
691 }
692
693 static int jbd2_seq_info_show(struct seq_file *seq, void *v)
694 {
695         struct jbd2_stats_proc_session *s = seq->private;
696
697         if (v != SEQ_START_TOKEN)
698                 return 0;
699         seq_printf(seq, "%lu transaction, each up to %u blocks\n",
700                         s->stats->ts_tid,
701                         s->journal->j_max_transaction_buffers);
702         if (s->stats->ts_tid == 0)
703                 return 0;
704         seq_printf(seq, "average: \n  %ums waiting for transaction\n",
705             jiffies_to_msecs(s->stats->run.rs_wait / s->stats->ts_tid));
706         seq_printf(seq, "  %ums running transaction\n",
707             jiffies_to_msecs(s->stats->run.rs_running / s->stats->ts_tid));
708         seq_printf(seq, "  %ums transaction was being locked\n",
709             jiffies_to_msecs(s->stats->run.rs_locked / s->stats->ts_tid));
710         seq_printf(seq, "  %ums flushing data (in ordered mode)\n",
711             jiffies_to_msecs(s->stats->run.rs_flushing / s->stats->ts_tid));
712         seq_printf(seq, "  %ums logging transaction\n",
713             jiffies_to_msecs(s->stats->run.rs_logging / s->stats->ts_tid));
714         seq_printf(seq, "  %lluus average transaction commit time\n",
715                    div_u64(s->journal->j_average_commit_time, 1000));
716         seq_printf(seq, "  %lu handles per transaction\n",
717             s->stats->run.rs_handle_count / s->stats->ts_tid);
718         seq_printf(seq, "  %lu blocks per transaction\n",
719             s->stats->run.rs_blocks / s->stats->ts_tid);
720         seq_printf(seq, "  %lu logged blocks per transaction\n",
721             s->stats->run.rs_blocks_logged / s->stats->ts_tid);
722         return 0;
723 }
724
725 static void jbd2_seq_info_stop(struct seq_file *seq, void *v)
726 {
727 }
728
729 static const struct seq_operations jbd2_seq_info_ops = {
730         .start  = jbd2_seq_info_start,
731         .next   = jbd2_seq_info_next,
732         .stop   = jbd2_seq_info_stop,
733         .show   = jbd2_seq_info_show,
734 };
735
736 static int jbd2_seq_info_open(struct inode *inode, struct file *file)
737 {
738         journal_t *journal = PDE(inode)->data;
739         struct jbd2_stats_proc_session *s;
740         int rc, size;
741
742         s = kmalloc(sizeof(*s), GFP_KERNEL);
743         if (s == NULL)
744                 return -ENOMEM;
745         size = sizeof(struct transaction_stats_s);
746         s->stats = kmalloc(size, GFP_KERNEL);
747         if (s->stats == NULL) {
748                 kfree(s);
749                 return -ENOMEM;
750         }
751         spin_lock(&journal->j_history_lock);
752         memcpy(s->stats, &journal->j_stats, size);
753         s->journal = journal;
754         spin_unlock(&journal->j_history_lock);
755
756         rc = seq_open(file, &jbd2_seq_info_ops);
757         if (rc == 0) {
758                 struct seq_file *m = file->private_data;
759                 m->private = s;
760         } else {
761                 kfree(s->stats);
762                 kfree(s);
763         }
764         return rc;
765
766 }
767
768 static int jbd2_seq_info_release(struct inode *inode, struct file *file)
769 {
770         struct seq_file *seq = file->private_data;
771         struct jbd2_stats_proc_session *s = seq->private;
772         kfree(s->stats);
773         kfree(s);
774         return seq_release(inode, file);
775 }
776
777 static const struct file_operations jbd2_seq_info_fops = {
778         .owner          = THIS_MODULE,
779         .open           = jbd2_seq_info_open,
780         .read           = seq_read,
781         .llseek         = seq_lseek,
782         .release        = jbd2_seq_info_release,
783 };
784
785 static struct proc_dir_entry *proc_jbd2_stats;
786
787 static void jbd2_stats_proc_init(journal_t *journal)
788 {
789         journal->j_proc_entry = proc_mkdir(journal->j_devname, proc_jbd2_stats);
790         if (journal->j_proc_entry) {
791                 proc_create_data("info", S_IRUGO, journal->j_proc_entry,
792                                  &jbd2_seq_info_fops, journal);
793         }
794 }
795
796 static void jbd2_stats_proc_exit(journal_t *journal)
797 {
798         remove_proc_entry("info", journal->j_proc_entry);
799         remove_proc_entry(journal->j_devname, proc_jbd2_stats);
800 }
801
802 /*
803  * Management for journal control blocks: functions to create and
804  * destroy journal_t structures, and to initialise and read existing
805  * journal blocks from disk.  */
806
807 /* First: create and setup a journal_t object in memory.  We initialise
808  * very few fields yet: that has to wait until we have created the
809  * journal structures from from scratch, or loaded them from disk. */
810
811 static journal_t * journal_init_common (void)
812 {
813         journal_t *journal;
814         int err;
815
816         journal = kzalloc(sizeof(*journal), GFP_KERNEL|__GFP_NOFAIL);
817         if (!journal)
818                 goto fail;
819
820         init_waitqueue_head(&journal->j_wait_transaction_locked);
821         init_waitqueue_head(&journal->j_wait_logspace);
822         init_waitqueue_head(&journal->j_wait_done_commit);
823         init_waitqueue_head(&journal->j_wait_checkpoint);
824         init_waitqueue_head(&journal->j_wait_commit);
825         init_waitqueue_head(&journal->j_wait_updates);
826         mutex_init(&journal->j_barrier);
827         mutex_init(&journal->j_checkpoint_mutex);
828         spin_lock_init(&journal->j_revoke_lock);
829         spin_lock_init(&journal->j_list_lock);
830         spin_lock_init(&journal->j_state_lock);
831
832         journal->j_commit_interval = (HZ * JBD2_DEFAULT_MAX_COMMIT_AGE);
833         journal->j_min_batch_time = 0;
834         journal->j_max_batch_time = 15000; /* 15ms */
835
836         /* The journal is marked for error until we succeed with recovery! */
837         journal->j_flags = JBD2_ABORT;
838
839         /* Set up a default-sized revoke table for the new mount. */
840         err = jbd2_journal_init_revoke(journal, JOURNAL_REVOKE_DEFAULT_HASH);
841         if (err) {
842                 kfree(journal);
843                 goto fail;
844         }
845
846         spin_lock_init(&journal->j_history_lock);
847
848         return journal;
849 fail:
850         return NULL;
851 }
852
853 /* jbd2_journal_init_dev and jbd2_journal_init_inode:
854  *
855  * Create a journal structure assigned some fixed set of disk blocks to
856  * the journal.  We don't actually touch those disk blocks yet, but we
857  * need to set up all of the mapping information to tell the journaling
858  * system where the journal blocks are.
859  *
860  */
861
862 /**
863  *  journal_t * jbd2_journal_init_dev() - creates and initialises a journal structure
864  *  @bdev: Block device on which to create the journal
865  *  @fs_dev: Device which hold journalled filesystem for this journal.
866  *  @start: Block nr Start of journal.
867  *  @len:  Length of the journal in blocks.
868  *  @blocksize: blocksize of journalling device
869  *
870  *  Returns: a newly created journal_t *
871  *
872  *  jbd2_journal_init_dev creates a journal which maps a fixed contiguous
873  *  range of blocks on an arbitrary block device.
874  *
875  */
876 journal_t * jbd2_journal_init_dev(struct block_device *bdev,
877                         struct block_device *fs_dev,
878                         unsigned long long start, int len, int blocksize)
879 {
880         journal_t *journal = journal_init_common();
881         struct buffer_head *bh;
882         char *p;
883         int n;
884
885         if (!journal)
886                 return NULL;
887
888         /* journal descriptor can store up to n blocks -bzzz */
889         journal->j_blocksize = blocksize;
890         jbd2_stats_proc_init(journal);
891         n = journal->j_blocksize / sizeof(journal_block_tag_t);
892         journal->j_wbufsize = n;
893         journal->j_wbuf = kmalloc(n * sizeof(struct buffer_head*), GFP_KERNEL);
894         if (!journal->j_wbuf) {
895                 printk(KERN_ERR "%s: Cant allocate bhs for commit thread\n",
896                         __func__);
897                 goto out_err;
898         }
899         journal->j_dev = bdev;
900         journal->j_fs_dev = fs_dev;
901         journal->j_blk_offset = start;
902         journal->j_maxlen = len;
903         bdevname(journal->j_dev, journal->j_devname);
904         p = journal->j_devname;
905         while ((p = strchr(p, '/')))
906                 *p = '!';
907
908         bh = __getblk(journal->j_dev, start, journal->j_blocksize);
909         if (!bh) {
910                 printk(KERN_ERR
911                        "%s: Cannot get buffer for journal superblock\n",
912                        __func__);
913                 goto out_err;
914         }
915         journal->j_sb_buffer = bh;
916         journal->j_superblock = (journal_superblock_t *)bh->b_data;
917
918         return journal;
919 out_err:
920         kfree(journal->j_wbuf);
921         jbd2_stats_proc_exit(journal);
922         kfree(journal);
923         return NULL;
924 }
925
926 /**
927  *  journal_t * jbd2_journal_init_inode () - creates a journal which maps to a inode.
928  *  @inode: An inode to create the journal in
929  *
930  * jbd2_journal_init_inode creates a journal which maps an on-disk inode as
931  * the journal.  The inode must exist already, must support bmap() and
932  * must have all data blocks preallocated.
933  */
934 journal_t * jbd2_journal_init_inode (struct inode *inode)
935 {
936         struct buffer_head *bh;
937         journal_t *journal = journal_init_common();
938         char *p;
939         int err;
940         int n;
941         unsigned long long blocknr;
942
943         if (!journal)
944                 return NULL;
945
946         journal->j_dev = journal->j_fs_dev = inode->i_sb->s_bdev;
947         journal->j_inode = inode;
948         bdevname(journal->j_dev, journal->j_devname);
949         p = journal->j_devname;
950         while ((p = strchr(p, '/')))
951                 *p = '!';
952         p = journal->j_devname + strlen(journal->j_devname);
953         sprintf(p, "-%lu", journal->j_inode->i_ino);
954         jbd_debug(1,
955                   "journal %p: inode %s/%ld, size %Ld, bits %d, blksize %ld\n",
956                   journal, inode->i_sb->s_id, inode->i_ino,
957                   (long long) inode->i_size,
958                   inode->i_sb->s_blocksize_bits, inode->i_sb->s_blocksize);
959
960         journal->j_maxlen = inode->i_size >> inode->i_sb->s_blocksize_bits;
961         journal->j_blocksize = inode->i_sb->s_blocksize;
962         jbd2_stats_proc_init(journal);
963
964         /* journal descriptor can store up to n blocks -bzzz */
965         n = journal->j_blocksize / sizeof(journal_block_tag_t);
966         journal->j_wbufsize = n;
967         journal->j_wbuf = kmalloc(n * sizeof(struct buffer_head*), GFP_KERNEL);
968         if (!journal->j_wbuf) {
969                 printk(KERN_ERR "%s: Cant allocate bhs for commit thread\n",
970                         __func__);
971                 goto out_err;
972         }
973
974         err = jbd2_journal_bmap(journal, 0, &blocknr);
975         /* If that failed, give up */
976         if (err) {
977                 printk(KERN_ERR "%s: Cannnot locate journal superblock\n",
978                        __func__);
979                 goto out_err;
980         }
981
982         bh = __getblk(journal->j_dev, blocknr, journal->j_blocksize);
983         if (!bh) {
984                 printk(KERN_ERR
985                        "%s: Cannot get buffer for journal superblock\n",
986                        __func__);
987                 goto out_err;
988         }
989         journal->j_sb_buffer = bh;
990         journal->j_superblock = (journal_superblock_t *)bh->b_data;
991
992         return journal;
993 out_err:
994         kfree(journal->j_wbuf);
995         jbd2_stats_proc_exit(journal);
996         kfree(journal);
997         return NULL;
998 }
999
1000 /*
1001  * If the journal init or create aborts, we need to mark the journal
1002  * superblock as being NULL to prevent the journal destroy from writing
1003  * back a bogus superblock.
1004  */
1005 static void journal_fail_superblock (journal_t *journal)
1006 {
1007         struct buffer_head *bh = journal->j_sb_buffer;
1008         brelse(bh);
1009         journal->j_sb_buffer = NULL;
1010 }
1011
1012 /*
1013  * Given a journal_t structure, initialise the various fields for
1014  * startup of a new journaling session.  We use this both when creating
1015  * a journal, and after recovering an old journal to reset it for
1016  * subsequent use.
1017  */
1018
1019 static int journal_reset(journal_t *journal)
1020 {
1021         journal_superblock_t *sb = journal->j_superblock;
1022         unsigned long long first, last;
1023
1024         first = be32_to_cpu(sb->s_first);
1025         last = be32_to_cpu(sb->s_maxlen);
1026         if (first + JBD2_MIN_JOURNAL_BLOCKS > last + 1) {
1027                 printk(KERN_ERR "JBD: Journal too short (blocks %llu-%llu).\n",
1028                        first, last);
1029                 journal_fail_superblock(journal);
1030                 return -EINVAL;
1031         }
1032
1033         journal->j_first = first;
1034         journal->j_last = last;
1035
1036         journal->j_head = first;
1037         journal->j_tail = first;
1038         journal->j_free = last - first;
1039
1040         journal->j_tail_sequence = journal->j_transaction_sequence;
1041         journal->j_commit_sequence = journal->j_transaction_sequence - 1;
1042         journal->j_commit_request = journal->j_commit_sequence;
1043
1044         journal->j_max_transaction_buffers = journal->j_maxlen / 4;
1045
1046         /* Add the dynamic fields and write it to disk. */
1047         jbd2_journal_update_superblock(journal, 1);
1048         return jbd2_journal_start_thread(journal);
1049 }
1050
1051 /**
1052  * void jbd2_journal_update_superblock() - Update journal sb on disk.
1053  * @journal: The journal to update.
1054  * @wait: Set to '0' if you don't want to wait for IO completion.
1055  *
1056  * Update a journal's dynamic superblock fields and write it to disk,
1057  * optionally waiting for the IO to complete.
1058  */
1059 void jbd2_journal_update_superblock(journal_t *journal, int wait)
1060 {
1061         journal_superblock_t *sb = journal->j_superblock;
1062         struct buffer_head *bh = journal->j_sb_buffer;
1063
1064         /*
1065          * As a special case, if the on-disk copy is already marked as needing
1066          * no recovery (s_start == 0) and there are no outstanding transactions
1067          * in the filesystem, then we can safely defer the superblock update
1068          * until the next commit by setting JBD2_FLUSHED.  This avoids
1069          * attempting a write to a potential-readonly device.
1070          */
1071         if (sb->s_start == 0 && journal->j_tail_sequence ==
1072                                 journal->j_transaction_sequence) {
1073                 jbd_debug(1,"JBD: Skipping superblock update on recovered sb "
1074                         "(start %ld, seq %d, errno %d)\n",
1075                         journal->j_tail, journal->j_tail_sequence,
1076                         journal->j_errno);
1077                 goto out;
1078         }
1079
1080         if (buffer_write_io_error(bh)) {
1081                 /*
1082                  * Oh, dear.  A previous attempt to write the journal
1083                  * superblock failed.  This could happen because the
1084                  * USB device was yanked out.  Or it could happen to
1085                  * be a transient write error and maybe the block will
1086                  * be remapped.  Nothing we can do but to retry the
1087                  * write and hope for the best.
1088                  */
1089                 printk(KERN_ERR "JBD2: previous I/O error detected "
1090                        "for journal superblock update for %s.\n",
1091                        journal->j_devname);
1092                 clear_buffer_write_io_error(bh);
1093                 set_buffer_uptodate(bh);
1094         }
1095
1096         spin_lock(&journal->j_state_lock);
1097         jbd_debug(1,"JBD: updating superblock (start %ld, seq %d, errno %d)\n",
1098                   journal->j_tail, journal->j_tail_sequence, journal->j_errno);
1099
1100         sb->s_sequence = cpu_to_be32(journal->j_tail_sequence);
1101         sb->s_start    = cpu_to_be32(journal->j_tail);
1102         sb->s_errno    = cpu_to_be32(journal->j_errno);
1103         spin_unlock(&journal->j_state_lock);
1104
1105         BUFFER_TRACE(bh, "marking dirty");
1106         mark_buffer_dirty(bh);
1107         if (wait) {
1108                 sync_dirty_buffer(bh);
1109                 if (buffer_write_io_error(bh)) {
1110                         printk(KERN_ERR "JBD2: I/O error detected "
1111                                "when updating journal superblock for %s.\n",
1112                                journal->j_devname);
1113                         clear_buffer_write_io_error(bh);
1114                         set_buffer_uptodate(bh);
1115                 }
1116         } else
1117                 ll_rw_block(SWRITE, 1, &bh);
1118
1119 out:
1120         /* If we have just flushed the log (by marking s_start==0), then
1121          * any future commit will have to be careful to update the
1122          * superblock again to re-record the true start of the log. */
1123
1124         spin_lock(&journal->j_state_lock);
1125         if (sb->s_start)
1126                 journal->j_flags &= ~JBD2_FLUSHED;
1127         else
1128                 journal->j_flags |= JBD2_FLUSHED;
1129         spin_unlock(&journal->j_state_lock);
1130 }
1131
1132 /*
1133  * Read the superblock for a given journal, performing initial
1134  * validation of the format.
1135  */
1136
1137 static int journal_get_superblock(journal_t *journal)
1138 {
1139         struct buffer_head *bh;
1140         journal_superblock_t *sb;
1141         int err = -EIO;
1142
1143         bh = journal->j_sb_buffer;
1144
1145         J_ASSERT(bh != NULL);
1146         if (!buffer_uptodate(bh)) {
1147                 ll_rw_block(READ, 1, &bh);
1148                 wait_on_buffer(bh);
1149                 if (!buffer_uptodate(bh)) {
1150                         printk (KERN_ERR
1151                                 "JBD: IO error reading journal superblock\n");
1152                         goto out;
1153                 }
1154         }
1155
1156         sb = journal->j_superblock;
1157
1158         err = -EINVAL;
1159
1160         if (sb->s_header.h_magic != cpu_to_be32(JBD2_MAGIC_NUMBER) ||
1161             sb->s_blocksize != cpu_to_be32(journal->j_blocksize)) {
1162                 printk(KERN_WARNING "JBD: no valid journal superblock found\n");
1163                 goto out;
1164         }
1165
1166         switch(be32_to_cpu(sb->s_header.h_blocktype)) {
1167         case JBD2_SUPERBLOCK_V1:
1168                 journal->j_format_version = 1;
1169                 break;
1170         case JBD2_SUPERBLOCK_V2:
1171                 journal->j_format_version = 2;
1172                 break;
1173         default:
1174                 printk(KERN_WARNING "JBD: unrecognised superblock format ID\n");
1175                 goto out;
1176         }
1177
1178         if (be32_to_cpu(sb->s_maxlen) < journal->j_maxlen)
1179                 journal->j_maxlen = be32_to_cpu(sb->s_maxlen);
1180         else if (be32_to_cpu(sb->s_maxlen) > journal->j_maxlen) {
1181                 printk (KERN_WARNING "JBD: journal file too short\n");
1182                 goto out;
1183         }
1184
1185         return 0;
1186
1187 out:
1188         journal_fail_superblock(journal);
1189         return err;
1190 }
1191
1192 /*
1193  * Load the on-disk journal superblock and read the key fields into the
1194  * journal_t.
1195  */
1196
1197 static int load_superblock(journal_t *journal)
1198 {
1199         int err;
1200         journal_superblock_t *sb;
1201
1202         err = journal_get_superblock(journal);
1203         if (err)
1204                 return err;
1205
1206         sb = journal->j_superblock;
1207
1208         journal->j_tail_sequence = be32_to_cpu(sb->s_sequence);
1209         journal->j_tail = be32_to_cpu(sb->s_start);
1210         journal->j_first = be32_to_cpu(sb->s_first);
1211         journal->j_last = be32_to_cpu(sb->s_maxlen);
1212         journal->j_errno = be32_to_cpu(sb->s_errno);
1213
1214         return 0;
1215 }
1216
1217
1218 /**
1219  * int jbd2_journal_load() - Read journal from disk.
1220  * @journal: Journal to act on.
1221  *
1222  * Given a journal_t structure which tells us which disk blocks contain
1223  * a journal, read the journal from disk to initialise the in-memory
1224  * structures.
1225  */
1226 int jbd2_journal_load(journal_t *journal)
1227 {
1228         int err;
1229         journal_superblock_t *sb;
1230
1231         err = load_superblock(journal);
1232         if (err)
1233                 return err;
1234
1235         sb = journal->j_superblock;
1236         /* If this is a V2 superblock, then we have to check the
1237          * features flags on it. */
1238
1239         if (journal->j_format_version >= 2) {
1240                 if ((sb->s_feature_ro_compat &
1241                      ~cpu_to_be32(JBD2_KNOWN_ROCOMPAT_FEATURES)) ||
1242                     (sb->s_feature_incompat &
1243                      ~cpu_to_be32(JBD2_KNOWN_INCOMPAT_FEATURES))) {
1244                         printk (KERN_WARNING
1245                                 "JBD: Unrecognised features on journal\n");
1246                         return -EINVAL;
1247                 }
1248         }
1249
1250         /* Let the recovery code check whether it needs to recover any
1251          * data from the journal. */
1252         if (jbd2_journal_recover(journal))
1253                 goto recovery_error;
1254
1255         if (journal->j_failed_commit) {
1256                 printk(KERN_ERR "JBD2: journal transaction %u on %s "
1257                        "is corrupt.\n", journal->j_failed_commit,
1258                        journal->j_devname);
1259                 return -EIO;
1260         }
1261
1262         /* OK, we've finished with the dynamic journal bits:
1263          * reinitialise the dynamic contents of the superblock in memory
1264          * and reset them on disk. */
1265         if (journal_reset(journal))
1266                 goto recovery_error;
1267
1268         journal->j_flags &= ~JBD2_ABORT;
1269         journal->j_flags |= JBD2_LOADED;
1270         return 0;
1271
1272 recovery_error:
1273         printk (KERN_WARNING "JBD: recovery failed\n");
1274         return -EIO;
1275 }
1276
1277 /**
1278  * void jbd2_journal_destroy() - Release a journal_t structure.
1279  * @journal: Journal to act on.
1280  *
1281  * Release a journal_t structure once it is no longer in use by the
1282  * journaled object.
1283  * Return <0 if we couldn't clean up the journal.
1284  */
1285 int jbd2_journal_destroy(journal_t *journal)
1286 {
1287         int err = 0;
1288
1289         /* Wait for the commit thread to wake up and die. */
1290         journal_kill_thread(journal);
1291
1292         /* Force a final log commit */
1293         if (journal->j_running_transaction)
1294                 jbd2_journal_commit_transaction(journal);
1295
1296         /* Force any old transactions to disk */
1297
1298         /* Totally anal locking here... */
1299         spin_lock(&journal->j_list_lock);
1300         while (journal->j_checkpoint_transactions != NULL) {
1301                 spin_unlock(&journal->j_list_lock);
1302                 mutex_lock(&journal->j_checkpoint_mutex);
1303                 jbd2_log_do_checkpoint(journal);
1304                 mutex_unlock(&journal->j_checkpoint_mutex);
1305                 spin_lock(&journal->j_list_lock);
1306         }
1307
1308         J_ASSERT(journal->j_running_transaction == NULL);
1309         J_ASSERT(journal->j_committing_transaction == NULL);
1310         J_ASSERT(journal->j_checkpoint_transactions == NULL);
1311         spin_unlock(&journal->j_list_lock);
1312
1313         if (journal->j_sb_buffer) {
1314                 if (!is_journal_aborted(journal)) {
1315                         /* We can now mark the journal as empty. */
1316                         journal->j_tail = 0;
1317                         journal->j_tail_sequence =
1318                                 ++journal->j_transaction_sequence;
1319                         jbd2_journal_update_superblock(journal, 1);
1320                 } else {
1321                         err = -EIO;
1322                 }
1323                 brelse(journal->j_sb_buffer);
1324         }
1325
1326         if (journal->j_proc_entry)
1327                 jbd2_stats_proc_exit(journal);
1328         if (journal->j_inode)
1329                 iput(journal->j_inode);
1330         if (journal->j_revoke)
1331                 jbd2_journal_destroy_revoke(journal);
1332         kfree(journal->j_wbuf);
1333         kfree(journal);
1334
1335         return err;
1336 }
1337
1338
1339 /**
1340  *int jbd2_journal_check_used_features () - Check if features specified are used.
1341  * @journal: Journal to check.
1342  * @compat: bitmask of compatible features
1343  * @ro: bitmask of features that force read-only mount
1344  * @incompat: bitmask of incompatible features
1345  *
1346  * Check whether the journal uses all of a given set of
1347  * features.  Return true (non-zero) if it does.
1348  **/
1349
1350 int jbd2_journal_check_used_features (journal_t *journal, unsigned long compat,
1351                                  unsigned long ro, unsigned long incompat)
1352 {
1353         journal_superblock_t *sb;
1354
1355         if (!compat && !ro && !incompat)
1356                 return 1;
1357         if (journal->j_format_version == 1)
1358                 return 0;
1359
1360         sb = journal->j_superblock;
1361
1362         if (((be32_to_cpu(sb->s_feature_compat) & compat) == compat) &&
1363             ((be32_to_cpu(sb->s_feature_ro_compat) & ro) == ro) &&
1364             ((be32_to_cpu(sb->s_feature_incompat) & incompat) == incompat))
1365                 return 1;
1366
1367         return 0;
1368 }
1369
1370 /**
1371  * int jbd2_journal_check_available_features() - Check feature set in journalling layer
1372  * @journal: Journal to check.
1373  * @compat: bitmask of compatible features
1374  * @ro: bitmask of features that force read-only mount
1375  * @incompat: bitmask of incompatible features
1376  *
1377  * Check whether the journaling code supports the use of
1378  * all of a given set of features on this journal.  Return true
1379  * (non-zero) if it can. */
1380
1381 int jbd2_journal_check_available_features (journal_t *journal, unsigned long compat,
1382                                       unsigned long ro, unsigned long incompat)
1383 {
1384         journal_superblock_t *sb;
1385
1386         if (!compat && !ro && !incompat)
1387                 return 1;
1388
1389         sb = journal->j_superblock;
1390
1391         /* We can support any known requested features iff the
1392          * superblock is in version 2.  Otherwise we fail to support any
1393          * extended sb features. */
1394
1395         if (journal->j_format_version != 2)
1396                 return 0;
1397
1398         if ((compat   & JBD2_KNOWN_COMPAT_FEATURES) == compat &&
1399             (ro       & JBD2_KNOWN_ROCOMPAT_FEATURES) == ro &&
1400             (incompat & JBD2_KNOWN_INCOMPAT_FEATURES) == incompat)
1401                 return 1;
1402
1403         return 0;
1404 }
1405
1406 /**
1407  * int jbd2_journal_set_features () - Mark a given journal feature in the superblock
1408  * @journal: Journal to act on.
1409  * @compat: bitmask of compatible features
1410  * @ro: bitmask of features that force read-only mount
1411  * @incompat: bitmask of incompatible features
1412  *
1413  * Mark a given journal feature as present on the
1414  * superblock.  Returns true if the requested features could be set.
1415  *
1416  */
1417
1418 int jbd2_journal_set_features (journal_t *journal, unsigned long compat,
1419                           unsigned long ro, unsigned long incompat)
1420 {
1421         journal_superblock_t *sb;
1422
1423         if (jbd2_journal_check_used_features(journal, compat, ro, incompat))
1424                 return 1;
1425
1426         if (!jbd2_journal_check_available_features(journal, compat, ro, incompat))
1427                 return 0;
1428
1429         jbd_debug(1, "Setting new features 0x%lx/0x%lx/0x%lx\n",
1430                   compat, ro, incompat);
1431
1432         sb = journal->j_superblock;
1433
1434         sb->s_feature_compat    |= cpu_to_be32(compat);
1435         sb->s_feature_ro_compat |= cpu_to_be32(ro);
1436         sb->s_feature_incompat  |= cpu_to_be32(incompat);
1437
1438         return 1;
1439 }
1440
1441 /*
1442  * jbd2_journal_clear_features () - Clear a given journal feature in the
1443  *                                  superblock
1444  * @journal: Journal to act on.
1445  * @compat: bitmask of compatible features
1446  * @ro: bitmask of features that force read-only mount
1447  * @incompat: bitmask of incompatible features
1448  *
1449  * Clear a given journal feature as present on the
1450  * superblock.
1451  */
1452 void jbd2_journal_clear_features(journal_t *journal, unsigned long compat,
1453                                 unsigned long ro, unsigned long incompat)
1454 {
1455         journal_superblock_t *sb;
1456
1457         jbd_debug(1, "Clear features 0x%lx/0x%lx/0x%lx\n",
1458                   compat, ro, incompat);
1459
1460         sb = journal->j_superblock;
1461
1462         sb->s_feature_compat    &= ~cpu_to_be32(compat);
1463         sb->s_feature_ro_compat &= ~cpu_to_be32(ro);
1464         sb->s_feature_incompat  &= ~cpu_to_be32(incompat);
1465 }
1466 EXPORT_SYMBOL(jbd2_journal_clear_features);
1467
1468 /**
1469  * int jbd2_journal_update_format () - Update on-disk journal structure.
1470  * @journal: Journal to act on.
1471  *
1472  * Given an initialised but unloaded journal struct, poke about in the
1473  * on-disk structure to update it to the most recent supported version.
1474  */
1475 int jbd2_journal_update_format (journal_t *journal)
1476 {
1477         journal_superblock_t *sb;
1478         int err;
1479
1480         err = journal_get_superblock(journal);
1481         if (err)
1482                 return err;
1483
1484         sb = journal->j_superblock;
1485
1486         switch (be32_to_cpu(sb->s_header.h_blocktype)) {
1487         case JBD2_SUPERBLOCK_V2:
1488                 return 0;
1489         case JBD2_SUPERBLOCK_V1:
1490                 return journal_convert_superblock_v1(journal, sb);
1491         default:
1492                 break;
1493         }
1494         return -EINVAL;
1495 }
1496
1497 static int journal_convert_superblock_v1(journal_t *journal,
1498                                          journal_superblock_t *sb)
1499 {
1500         int offset, blocksize;
1501         struct buffer_head *bh;
1502
1503         printk(KERN_WARNING
1504                 "JBD: Converting superblock from version 1 to 2.\n");
1505
1506         /* Pre-initialise new fields to zero */
1507         offset = ((char *) &(sb->s_feature_compat)) - ((char *) sb);
1508         blocksize = be32_to_cpu(sb->s_blocksize);
1509         memset(&sb->s_feature_compat, 0, blocksize-offset);
1510
1511         sb->s_nr_users = cpu_to_be32(1);
1512         sb->s_header.h_blocktype = cpu_to_be32(JBD2_SUPERBLOCK_V2);
1513         journal->j_format_version = 2;
1514
1515         bh = journal->j_sb_buffer;
1516         BUFFER_TRACE(bh, "marking dirty");
1517         mark_buffer_dirty(bh);
1518         sync_dirty_buffer(bh);
1519         return 0;
1520 }
1521
1522
1523 /**
1524  * int jbd2_journal_flush () - Flush journal
1525  * @journal: Journal to act on.
1526  *
1527  * Flush all data for a given journal to disk and empty the journal.
1528  * Filesystems can use this when remounting readonly to ensure that
1529  * recovery does not need to happen on remount.
1530  */
1531
1532 int jbd2_journal_flush(journal_t *journal)
1533 {
1534         int err = 0;
1535         transaction_t *transaction = NULL;
1536         unsigned long old_tail;
1537
1538         spin_lock(&journal->j_state_lock);
1539
1540         /* Force everything buffered to the log... */
1541         if (journal->j_running_transaction) {
1542                 transaction = journal->j_running_transaction;
1543                 __jbd2_log_start_commit(journal, transaction->t_tid);
1544         } else if (journal->j_committing_transaction)
1545                 transaction = journal->j_committing_transaction;
1546
1547         /* Wait for the log commit to complete... */
1548         if (transaction) {
1549                 tid_t tid = transaction->t_tid;
1550
1551                 spin_unlock(&journal->j_state_lock);
1552                 jbd2_log_wait_commit(journal, tid);
1553         } else {
1554                 spin_unlock(&journal->j_state_lock);
1555         }
1556
1557         /* ...and flush everything in the log out to disk. */
1558         spin_lock(&journal->j_list_lock);
1559         while (!err && journal->j_checkpoint_transactions != NULL) {
1560                 spin_unlock(&journal->j_list_lock);
1561                 mutex_lock(&journal->j_checkpoint_mutex);
1562                 err = jbd2_log_do_checkpoint(journal);
1563                 mutex_unlock(&journal->j_checkpoint_mutex);
1564                 spin_lock(&journal->j_list_lock);
1565         }
1566         spin_unlock(&journal->j_list_lock);
1567
1568         if (is_journal_aborted(journal))
1569                 return -EIO;
1570
1571         jbd2_cleanup_journal_tail(journal);
1572
1573         /* Finally, mark the journal as really needing no recovery.
1574          * This sets s_start==0 in the underlying superblock, which is
1575          * the magic code for a fully-recovered superblock.  Any future
1576          * commits of data to the journal will restore the current
1577          * s_start value. */
1578         spin_lock(&journal->j_state_lock);
1579         old_tail = journal->j_tail;
1580         journal->j_tail = 0;
1581         spin_unlock(&journal->j_state_lock);
1582         jbd2_journal_update_superblock(journal, 1);
1583         spin_lock(&journal->j_state_lock);
1584         journal->j_tail = old_tail;
1585
1586         J_ASSERT(!journal->j_running_transaction);
1587         J_ASSERT(!journal->j_committing_transaction);
1588         J_ASSERT(!journal->j_checkpoint_transactions);
1589         J_ASSERT(journal->j_head == journal->j_tail);
1590         J_ASSERT(journal->j_tail_sequence == journal->j_transaction_sequence);
1591         spin_unlock(&journal->j_state_lock);
1592         return 0;
1593 }
1594
1595 /**
1596  * int jbd2_journal_wipe() - Wipe journal contents
1597  * @journal: Journal to act on.
1598  * @write: flag (see below)
1599  *
1600  * Wipe out all of the contents of a journal, safely.  This will produce
1601  * a warning if the journal contains any valid recovery information.
1602  * Must be called between journal_init_*() and jbd2_journal_load().
1603  *
1604  * If 'write' is non-zero, then we wipe out the journal on disk; otherwise
1605  * we merely suppress recovery.
1606  */
1607
1608 int jbd2_journal_wipe(journal_t *journal, int write)
1609 {
1610         journal_superblock_t *sb;
1611         int err = 0;
1612
1613         J_ASSERT (!(journal->j_flags & JBD2_LOADED));
1614
1615         err = load_superblock(journal);
1616         if (err)
1617                 return err;
1618
1619         sb = journal->j_superblock;
1620
1621         if (!journal->j_tail)
1622                 goto no_recovery;
1623
1624         printk (KERN_WARNING "JBD: %s recovery information on journal\n",
1625                 write ? "Clearing" : "Ignoring");
1626
1627         err = jbd2_journal_skip_recovery(journal);
1628         if (write)
1629                 jbd2_journal_update_superblock(journal, 1);
1630
1631  no_recovery:
1632         return err;
1633 }
1634
1635 /*
1636  * Journal abort has very specific semantics, which we describe
1637  * for journal abort.
1638  *
1639  * Two internal functions, which provide abort to the jbd layer
1640  * itself are here.
1641  */
1642
1643 /*
1644  * Quick version for internal journal use (doesn't lock the journal).
1645  * Aborts hard --- we mark the abort as occurred, but do _nothing_ else,
1646  * and don't attempt to make any other journal updates.
1647  */
1648 void __jbd2_journal_abort_hard(journal_t *journal)
1649 {
1650         transaction_t *transaction;
1651
1652         if (journal->j_flags & JBD2_ABORT)
1653                 return;
1654
1655         printk(KERN_ERR "Aborting journal on device %s.\n",
1656                journal->j_devname);
1657
1658         spin_lock(&journal->j_state_lock);
1659         journal->j_flags |= JBD2_ABORT;
1660         transaction = journal->j_running_transaction;
1661         if (transaction)
1662                 __jbd2_log_start_commit(journal, transaction->t_tid);
1663         spin_unlock(&journal->j_state_lock);
1664 }
1665
1666 /* Soft abort: record the abort error status in the journal superblock,
1667  * but don't do any other IO. */
1668 static void __journal_abort_soft (journal_t *journal, int errno)
1669 {
1670         if (journal->j_flags & JBD2_ABORT)
1671                 return;
1672
1673         if (!journal->j_errno)
1674                 journal->j_errno = errno;
1675
1676         __jbd2_journal_abort_hard(journal);
1677
1678         if (errno)
1679                 jbd2_journal_update_superblock(journal, 1);
1680 }
1681
1682 /**
1683  * void jbd2_journal_abort () - Shutdown the journal immediately.
1684  * @journal: the journal to shutdown.
1685  * @errno:   an error number to record in the journal indicating
1686  *           the reason for the shutdown.
1687  *
1688  * Perform a complete, immediate shutdown of the ENTIRE
1689  * journal (not of a single transaction).  This operation cannot be
1690  * undone without closing and reopening the journal.
1691  *
1692  * The jbd2_journal_abort function is intended to support higher level error
1693  * recovery mechanisms such as the ext2/ext3 remount-readonly error
1694  * mode.
1695  *
1696  * Journal abort has very specific semantics.  Any existing dirty,
1697  * unjournaled buffers in the main filesystem will still be written to
1698  * disk by bdflush, but the journaling mechanism will be suspended
1699  * immediately and no further transaction commits will be honoured.
1700  *
1701  * Any dirty, journaled buffers will be written back to disk without
1702  * hitting the journal.  Atomicity cannot be guaranteed on an aborted
1703  * filesystem, but we _do_ attempt to leave as much data as possible
1704  * behind for fsck to use for cleanup.
1705  *
1706  * Any attempt to get a new transaction handle on a journal which is in
1707  * ABORT state will just result in an -EROFS error return.  A
1708  * jbd2_journal_stop on an existing handle will return -EIO if we have
1709  * entered abort state during the update.
1710  *
1711  * Recursive transactions are not disturbed by journal abort until the
1712  * final jbd2_journal_stop, which will receive the -EIO error.
1713  *
1714  * Finally, the jbd2_journal_abort call allows the caller to supply an errno
1715  * which will be recorded (if possible) in the journal superblock.  This
1716  * allows a client to record failure conditions in the middle of a
1717  * transaction without having to complete the transaction to record the
1718  * failure to disk.  ext3_error, for example, now uses this
1719  * functionality.
1720  *
1721  * Errors which originate from within the journaling layer will NOT
1722  * supply an errno; a null errno implies that absolutely no further
1723  * writes are done to the journal (unless there are any already in
1724  * progress).
1725  *
1726  */
1727
1728 void jbd2_journal_abort(journal_t *journal, int errno)
1729 {
1730         __journal_abort_soft(journal, errno);
1731 }
1732
1733 /**
1734  * int jbd2_journal_errno () - returns the journal's error state.
1735  * @journal: journal to examine.
1736  *
1737  * This is the errno number set with jbd2_journal_abort(), the last
1738  * time the journal was mounted - if the journal was stopped
1739  * without calling abort this will be 0.
1740  *
1741  * If the journal has been aborted on this mount time -EROFS will
1742  * be returned.
1743  */
1744 int jbd2_journal_errno(journal_t *journal)
1745 {
1746         int err;
1747
1748         spin_lock(&journal->j_state_lock);
1749         if (journal->j_flags & JBD2_ABORT)
1750                 err = -EROFS;
1751         else
1752                 err = journal->j_errno;
1753         spin_unlock(&journal->j_state_lock);
1754         return err;
1755 }
1756
1757 /**
1758  * int jbd2_journal_clear_err () - clears the journal's error state
1759  * @journal: journal to act on.
1760  *
1761  * An error must be cleared or acked to take a FS out of readonly
1762  * mode.
1763  */
1764 int jbd2_journal_clear_err(journal_t *journal)
1765 {
1766         int err = 0;
1767
1768         spin_lock(&journal->j_state_lock);
1769         if (journal->j_flags & JBD2_ABORT)
1770                 err = -EROFS;
1771         else
1772                 journal->j_errno = 0;
1773         spin_unlock(&journal->j_state_lock);
1774         return err;
1775 }
1776
1777 /**
1778  * void jbd2_journal_ack_err() - Ack journal err.
1779  * @journal: journal to act on.
1780  *
1781  * An error must be cleared or acked to take a FS out of readonly
1782  * mode.
1783  */
1784 void jbd2_journal_ack_err(journal_t *journal)
1785 {
1786         spin_lock(&journal->j_state_lock);
1787         if (journal->j_errno)
1788                 journal->j_flags |= JBD2_ACK_ERR;
1789         spin_unlock(&journal->j_state_lock);
1790 }
1791
1792 int jbd2_journal_blocks_per_page(struct inode *inode)
1793 {
1794         return 1 << (PAGE_CACHE_SHIFT - inode->i_sb->s_blocksize_bits);
1795 }
1796
1797 /*
1798  * helper functions to deal with 32 or 64bit block numbers.
1799  */
1800 size_t journal_tag_bytes(journal_t *journal)
1801 {
1802         if (JBD2_HAS_INCOMPAT_FEATURE(journal, JBD2_FEATURE_INCOMPAT_64BIT))
1803                 return JBD2_TAG_SIZE64;
1804         else
1805                 return JBD2_TAG_SIZE32;
1806 }
1807
1808 /*
1809  * Journal_head storage management
1810  */
1811 static struct kmem_cache *jbd2_journal_head_cache;
1812 #ifdef CONFIG_JBD2_DEBUG
1813 static atomic_t nr_journal_heads = ATOMIC_INIT(0);
1814 #endif
1815
1816 static int journal_init_jbd2_journal_head_cache(void)
1817 {
1818         int retval;
1819
1820         J_ASSERT(jbd2_journal_head_cache == NULL);
1821         jbd2_journal_head_cache = kmem_cache_create("jbd2_journal_head",
1822                                 sizeof(struct journal_head),
1823                                 0,              /* offset */
1824                                 SLAB_TEMPORARY, /* flags */
1825                                 NULL);          /* ctor */
1826         retval = 0;
1827         if (!jbd2_journal_head_cache) {
1828                 retval = -ENOMEM;
1829                 printk(KERN_EMERG "JBD: no memory for journal_head cache\n");
1830         }
1831         return retval;
1832 }
1833
1834 static void jbd2_journal_destroy_jbd2_journal_head_cache(void)
1835 {
1836         if (jbd2_journal_head_cache) {
1837                 kmem_cache_destroy(jbd2_journal_head_cache);
1838                 jbd2_journal_head_cache = NULL;
1839         }
1840 }
1841
1842 /*
1843  * journal_head splicing and dicing
1844  */
1845 static struct journal_head *journal_alloc_journal_head(void)
1846 {
1847         struct journal_head *ret;
1848         static unsigned long last_warning;
1849
1850 #ifdef CONFIG_JBD2_DEBUG
1851         atomic_inc(&nr_journal_heads);
1852 #endif
1853         ret = kmem_cache_alloc(jbd2_journal_head_cache, GFP_NOFS);
1854         if (!ret) {
1855                 jbd_debug(1, "out of memory for journal_head\n");
1856                 if (time_after(jiffies, last_warning + 5*HZ)) {
1857                         printk(KERN_NOTICE "ENOMEM in %s, retrying.\n",
1858                                __func__);
1859                         last_warning = jiffies;
1860                 }
1861                 while (!ret) {
1862                         yield();
1863                         ret = kmem_cache_alloc(jbd2_journal_head_cache, GFP_NOFS);
1864                 }
1865         }
1866         return ret;
1867 }
1868
1869 static void journal_free_journal_head(struct journal_head *jh)
1870 {
1871 #ifdef CONFIG_JBD2_DEBUG
1872         atomic_dec(&nr_journal_heads);
1873         memset(jh, JBD2_POISON_FREE, sizeof(*jh));
1874 #endif
1875         kmem_cache_free(jbd2_journal_head_cache, jh);
1876 }
1877
1878 /*
1879  * A journal_head is attached to a buffer_head whenever JBD has an
1880  * interest in the buffer.
1881  *
1882  * Whenever a buffer has an attached journal_head, its ->b_state:BH_JBD bit
1883  * is set.  This bit is tested in core kernel code where we need to take
1884  * JBD-specific actions.  Testing the zeroness of ->b_private is not reliable
1885  * there.
1886  *
1887  * When a buffer has its BH_JBD bit set, its ->b_count is elevated by one.
1888  *
1889  * When a buffer has its BH_JBD bit set it is immune from being released by
1890  * core kernel code, mainly via ->b_count.
1891  *
1892  * A journal_head may be detached from its buffer_head when the journal_head's
1893  * b_transaction, b_cp_transaction and b_next_transaction pointers are NULL.
1894  * Various places in JBD call jbd2_journal_remove_journal_head() to indicate that the
1895  * journal_head can be dropped if needed.
1896  *
1897  * Various places in the kernel want to attach a journal_head to a buffer_head
1898  * _before_ attaching the journal_head to a transaction.  To protect the
1899  * journal_head in this situation, jbd2_journal_add_journal_head elevates the
1900  * journal_head's b_jcount refcount by one.  The caller must call
1901  * jbd2_journal_put_journal_head() to undo this.
1902  *
1903  * So the typical usage would be:
1904  *
1905  *      (Attach a journal_head if needed.  Increments b_jcount)
1906  *      struct journal_head *jh = jbd2_journal_add_journal_head(bh);
1907  *      ...
1908  *      jh->b_transaction = xxx;
1909  *      jbd2_journal_put_journal_head(jh);
1910  *
1911  * Now, the journal_head's b_jcount is zero, but it is safe from being released
1912  * because it has a non-zero b_transaction.
1913  */
1914
1915 /*
1916  * Give a buffer_head a journal_head.
1917  *
1918  * Doesn't need the journal lock.
1919  * May sleep.
1920  */
1921 struct journal_head *jbd2_journal_add_journal_head(struct buffer_head *bh)
1922 {
1923         struct journal_head *jh;
1924         struct journal_head *new_jh = NULL;
1925
1926 repeat:
1927         if (!buffer_jbd(bh)) {
1928                 new_jh = journal_alloc_journal_head();
1929                 memset(new_jh, 0, sizeof(*new_jh));
1930         }
1931
1932         jbd_lock_bh_journal_head(bh);
1933         if (buffer_jbd(bh)) {
1934                 jh = bh2jh(bh);
1935         } else {
1936                 J_ASSERT_BH(bh,
1937                         (atomic_read(&bh->b_count) > 0) ||
1938                         (bh->b_page && bh->b_page->mapping));
1939
1940                 if (!new_jh) {
1941                         jbd_unlock_bh_journal_head(bh);
1942                         goto repeat;
1943                 }
1944
1945                 jh = new_jh;
1946                 new_jh = NULL;          /* We consumed it */
1947                 set_buffer_jbd(bh);
1948                 bh->b_private = jh;
1949                 jh->b_bh = bh;
1950                 get_bh(bh);
1951                 BUFFER_TRACE(bh, "added journal_head");
1952         }
1953         jh->b_jcount++;
1954         jbd_unlock_bh_journal_head(bh);
1955         if (new_jh)
1956                 journal_free_journal_head(new_jh);
1957         return bh->b_private;
1958 }
1959
1960 /*
1961  * Grab a ref against this buffer_head's journal_head.  If it ended up not
1962  * having a journal_head, return NULL
1963  */
1964 struct journal_head *jbd2_journal_grab_journal_head(struct buffer_head *bh)
1965 {
1966         struct journal_head *jh = NULL;
1967
1968         jbd_lock_bh_journal_head(bh);
1969         if (buffer_jbd(bh)) {
1970                 jh = bh2jh(bh);
1971                 jh->b_jcount++;
1972         }
1973         jbd_unlock_bh_journal_head(bh);
1974         return jh;
1975 }
1976
1977 static void __journal_remove_journal_head(struct buffer_head *bh)
1978 {
1979         struct journal_head *jh = bh2jh(bh);
1980
1981         J_ASSERT_JH(jh, jh->b_jcount >= 0);
1982
1983         get_bh(bh);
1984         if (jh->b_jcount == 0) {
1985                 if (jh->b_transaction == NULL &&
1986                                 jh->b_next_transaction == NULL &&
1987                                 jh->b_cp_transaction == NULL) {
1988                         J_ASSERT_JH(jh, jh->b_jlist == BJ_None);
1989                         J_ASSERT_BH(bh, buffer_jbd(bh));
1990                         J_ASSERT_BH(bh, jh2bh(jh) == bh);
1991                         BUFFER_TRACE(bh, "remove journal_head");
1992                         if (jh->b_frozen_data) {
1993                                 printk(KERN_WARNING "%s: freeing "
1994                                                 "b_frozen_data\n",
1995                                                 __func__);
1996                                 jbd2_free(jh->b_frozen_data, bh->b_size);
1997                         }
1998                         if (jh->b_committed_data) {
1999                                 printk(KERN_WARNING "%s: freeing "
2000                                                 "b_committed_data\n",
2001                                                 __func__);
2002                                 jbd2_free(jh->b_committed_data, bh->b_size);
2003                         }
2004                         bh->b_private = NULL;
2005                         jh->b_bh = NULL;        /* debug, really */
2006                         clear_buffer_jbd(bh);
2007                         __brelse(bh);
2008                         journal_free_journal_head(jh);
2009                 } else {
2010                         BUFFER_TRACE(bh, "journal_head was locked");
2011                 }
2012         }
2013 }
2014
2015 /*
2016  * jbd2_journal_remove_journal_head(): if the buffer isn't attached to a transaction
2017  * and has a zero b_jcount then remove and release its journal_head.   If we did
2018  * see that the buffer is not used by any transaction we also "logically"
2019  * decrement ->b_count.
2020  *
2021  * We in fact take an additional increment on ->b_count as a convenience,
2022  * because the caller usually wants to do additional things with the bh
2023  * after calling here.
2024  * The caller of jbd2_journal_remove_journal_head() *must* run __brelse(bh) at some
2025  * time.  Once the caller has run __brelse(), the buffer is eligible for
2026  * reaping by try_to_free_buffers().
2027  */
2028 void jbd2_journal_remove_journal_head(struct buffer_head *bh)
2029 {
2030         jbd_lock_bh_journal_head(bh);
2031         __journal_remove_journal_head(bh);
2032         jbd_unlock_bh_journal_head(bh);
2033 }
2034
2035 /*
2036  * Drop a reference on the passed journal_head.  If it fell to zero then try to
2037  * release the journal_head from the buffer_head.
2038  */
2039 void jbd2_journal_put_journal_head(struct journal_head *jh)
2040 {
2041         struct buffer_head *bh = jh2bh(jh);
2042
2043         jbd_lock_bh_journal_head(bh);
2044         J_ASSERT_JH(jh, jh->b_jcount > 0);
2045         --jh->b_jcount;
2046         if (!jh->b_jcount && !jh->b_transaction) {
2047                 __journal_remove_journal_head(bh);
2048                 __brelse(bh);
2049         }
2050         jbd_unlock_bh_journal_head(bh);
2051 }
2052
2053 /*
2054  * Initialize jbd inode head
2055  */
2056 void jbd2_journal_init_jbd_inode(struct jbd2_inode *jinode, struct inode *inode)
2057 {
2058         jinode->i_transaction = NULL;
2059         jinode->i_next_transaction = NULL;
2060         jinode->i_vfs_inode = inode;
2061         jinode->i_flags = 0;
2062         INIT_LIST_HEAD(&jinode->i_list);
2063 }
2064
2065 /*
2066  * Function to be called before we start removing inode from memory (i.e.,
2067  * clear_inode() is a fine place to be called from). It removes inode from
2068  * transaction's lists.
2069  */
2070 void jbd2_journal_release_jbd_inode(journal_t *journal,
2071                                     struct jbd2_inode *jinode)
2072 {
2073         int writeout = 0;
2074
2075         if (!journal)
2076                 return;
2077 restart:
2078         spin_lock(&journal->j_list_lock);
2079         /* Is commit writing out inode - we have to wait */
2080         if (jinode->i_flags & JI_COMMIT_RUNNING) {
2081                 wait_queue_head_t *wq;
2082                 DEFINE_WAIT_BIT(wait, &jinode->i_flags, __JI_COMMIT_RUNNING);
2083                 wq = bit_waitqueue(&jinode->i_flags, __JI_COMMIT_RUNNING);
2084                 prepare_to_wait(wq, &wait.wait, TASK_UNINTERRUPTIBLE);
2085                 spin_unlock(&journal->j_list_lock);
2086                 schedule();
2087                 finish_wait(wq, &wait.wait);
2088                 goto restart;
2089         }
2090
2091         /* Do we need to wait for data writeback? */
2092         if (journal->j_committing_transaction == jinode->i_transaction)
2093                 writeout = 1;
2094         if (jinode->i_transaction) {
2095                 list_del(&jinode->i_list);
2096                 jinode->i_transaction = NULL;
2097         }
2098         spin_unlock(&journal->j_list_lock);
2099 }
2100
2101 /*
2102  * debugfs tunables
2103  */
2104 #ifdef CONFIG_JBD2_DEBUG
2105 u8 jbd2_journal_enable_debug __read_mostly;
2106 EXPORT_SYMBOL(jbd2_journal_enable_debug);
2107
2108 #define JBD2_DEBUG_NAME "jbd2-debug"
2109
2110 static struct dentry *jbd2_debugfs_dir;
2111 static struct dentry *jbd2_debug;
2112
2113 static void __init jbd2_create_debugfs_entry(void)
2114 {
2115         jbd2_debugfs_dir = debugfs_create_dir("jbd2", NULL);
2116         if (jbd2_debugfs_dir)
2117                 jbd2_debug = debugfs_create_u8(JBD2_DEBUG_NAME, S_IRUGO,
2118                                                jbd2_debugfs_dir,
2119                                                &jbd2_journal_enable_debug);
2120 }
2121
2122 static void __exit jbd2_remove_debugfs_entry(void)
2123 {
2124         debugfs_remove(jbd2_debug);
2125         debugfs_remove(jbd2_debugfs_dir);
2126 }
2127
2128 #else
2129
2130 static void __init jbd2_create_debugfs_entry(void)
2131 {
2132 }
2133
2134 static void __exit jbd2_remove_debugfs_entry(void)
2135 {
2136 }
2137
2138 #endif
2139
2140 #ifdef CONFIG_PROC_FS
2141
2142 #define JBD2_STATS_PROC_NAME "fs/jbd2"
2143
2144 static void __init jbd2_create_jbd_stats_proc_entry(void)
2145 {
2146         proc_jbd2_stats = proc_mkdir(JBD2_STATS_PROC_NAME, NULL);
2147 }
2148
2149 static void __exit jbd2_remove_jbd_stats_proc_entry(void)
2150 {
2151         if (proc_jbd2_stats)
2152                 remove_proc_entry(JBD2_STATS_PROC_NAME, NULL);
2153 }
2154
2155 #else
2156
2157 #define jbd2_create_jbd_stats_proc_entry() do {} while (0)
2158 #define jbd2_remove_jbd_stats_proc_entry() do {} while (0)
2159
2160 #endif
2161
2162 struct kmem_cache *jbd2_handle_cache;
2163
2164 static int __init journal_init_handle_cache(void)
2165 {
2166         jbd2_handle_cache = kmem_cache_create("jbd2_journal_handle",
2167                                 sizeof(handle_t),
2168                                 0,              /* offset */
2169                                 SLAB_TEMPORARY, /* flags */
2170                                 NULL);          /* ctor */
2171         if (jbd2_handle_cache == NULL) {
2172                 printk(KERN_EMERG "JBD: failed to create handle cache\n");
2173                 return -ENOMEM;
2174         }
2175         return 0;
2176 }
2177
2178 static void jbd2_journal_destroy_handle_cache(void)
2179 {
2180         if (jbd2_handle_cache)
2181                 kmem_cache_destroy(jbd2_handle_cache);
2182 }
2183
2184 /*
2185  * Module startup and shutdown
2186  */
2187
2188 static int __init journal_init_caches(void)
2189 {
2190         int ret;
2191
2192         ret = jbd2_journal_init_revoke_caches();
2193         if (ret == 0)
2194                 ret = journal_init_jbd2_journal_head_cache();
2195         if (ret == 0)
2196                 ret = journal_init_handle_cache();
2197         return ret;
2198 }
2199
2200 static void jbd2_journal_destroy_caches(void)
2201 {
2202         jbd2_journal_destroy_revoke_caches();
2203         jbd2_journal_destroy_jbd2_journal_head_cache();
2204         jbd2_journal_destroy_handle_cache();
2205 }
2206
2207 static int __init journal_init(void)
2208 {
2209         int ret;
2210
2211         BUILD_BUG_ON(sizeof(struct journal_superblock_s) != 1024);
2212
2213         ret = journal_init_caches();
2214         if (ret == 0) {
2215                 jbd2_create_debugfs_entry();
2216                 jbd2_create_jbd_stats_proc_entry();
2217         } else {
2218                 jbd2_journal_destroy_caches();
2219         }
2220         return ret;
2221 }
2222
2223 static void __exit journal_exit(void)
2224 {
2225 #ifdef CONFIG_JBD2_DEBUG
2226         int n = atomic_read(&nr_journal_heads);
2227         if (n)
2228                 printk(KERN_EMERG "JBD: leaked %d journal_heads!\n", n);
2229 #endif
2230         jbd2_remove_debugfs_entry();
2231         jbd2_remove_jbd_stats_proc_entry();
2232         jbd2_journal_destroy_caches();
2233 }
2234
2235 /* 
2236  * jbd2_dev_to_name is a utility function used by the jbd2 and ext4 
2237  * tracing infrastructure to map a dev_t to a device name.
2238  *
2239  * The caller should use rcu_read_lock() in order to make sure the
2240  * device name stays valid until its done with it.  We use
2241  * rcu_read_lock() as well to make sure we're safe in case the caller
2242  * gets sloppy, and because rcu_read_lock() is cheap and can be safely
2243  * nested.
2244  */
2245 struct devname_cache {
2246         struct rcu_head rcu;
2247         dev_t           device;
2248         char            devname[BDEVNAME_SIZE];
2249 };
2250 #define CACHE_SIZE_BITS 6
2251 static struct devname_cache *devcache[1 << CACHE_SIZE_BITS];
2252 static DEFINE_SPINLOCK(devname_cache_lock);
2253
2254 static void free_devcache(struct rcu_head *rcu)
2255 {
2256         kfree(rcu);
2257 }
2258
2259 const char *jbd2_dev_to_name(dev_t device)
2260 {
2261         int     i = hash_32(device, CACHE_SIZE_BITS);
2262         char    *ret;
2263         struct block_device *bd;
2264         static struct devname_cache *new_dev;
2265
2266         rcu_read_lock();
2267         if (devcache[i] && devcache[i]->device == device) {
2268                 ret = devcache[i]->devname;
2269                 rcu_read_unlock();
2270                 return ret;
2271         }
2272         rcu_read_unlock();
2273
2274         new_dev = kmalloc(sizeof(struct devname_cache), GFP_KERNEL);
2275         if (!new_dev)
2276                 return "NODEV-ALLOCFAILURE"; /* Something non-NULL */
2277         spin_lock(&devname_cache_lock);
2278         if (devcache[i]) {
2279                 if (devcache[i]->device == device) {
2280                         kfree(new_dev);
2281                         ret = devcache[i]->devname;
2282                         spin_unlock(&devname_cache_lock);
2283                         return ret;
2284                 }
2285                 call_rcu(&devcache[i]->rcu, free_devcache);
2286         }
2287         devcache[i] = new_dev;
2288         devcache[i]->device = device;
2289         bd = bdget(device);
2290         if (bd) {
2291                 bdevname(bd, devcache[i]->devname);
2292                 bdput(bd);
2293         } else
2294                 __bdevname(device, devcache[i]->devname);
2295         ret = devcache[i]->devname;
2296         spin_unlock(&devname_cache_lock);
2297         return ret;
2298 }
2299 EXPORT_SYMBOL(jbd2_dev_to_name);
2300
2301 MODULE_LICENSE("GPL");
2302 module_init(journal_init);
2303 module_exit(journal_exit);
2304