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