Merge branches 'release', 'acpica', 'bugzilla-10224', 'bugzilla-9772', 'bugzilla...
[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
41 #include <asm/uaccess.h>
42 #include <asm/page.h>
43
44 EXPORT_SYMBOL(jbd2_journal_start);
45 EXPORT_SYMBOL(jbd2_journal_restart);
46 EXPORT_SYMBOL(jbd2_journal_extend);
47 EXPORT_SYMBOL(jbd2_journal_stop);
48 EXPORT_SYMBOL(jbd2_journal_lock_updates);
49 EXPORT_SYMBOL(jbd2_journal_unlock_updates);
50 EXPORT_SYMBOL(jbd2_journal_get_write_access);
51 EXPORT_SYMBOL(jbd2_journal_get_create_access);
52 EXPORT_SYMBOL(jbd2_journal_get_undo_access);
53 EXPORT_SYMBOL(jbd2_journal_dirty_data);
54 EXPORT_SYMBOL(jbd2_journal_dirty_metadata);
55 EXPORT_SYMBOL(jbd2_journal_release_buffer);
56 EXPORT_SYMBOL(jbd2_journal_forget);
57 #if 0
58 EXPORT_SYMBOL(journal_sync_buffer);
59 #endif
60 EXPORT_SYMBOL(jbd2_journal_flush);
61 EXPORT_SYMBOL(jbd2_journal_revoke);
62
63 EXPORT_SYMBOL(jbd2_journal_init_dev);
64 EXPORT_SYMBOL(jbd2_journal_init_inode);
65 EXPORT_SYMBOL(jbd2_journal_update_format);
66 EXPORT_SYMBOL(jbd2_journal_check_used_features);
67 EXPORT_SYMBOL(jbd2_journal_check_available_features);
68 EXPORT_SYMBOL(jbd2_journal_set_features);
69 EXPORT_SYMBOL(jbd2_journal_create);
70 EXPORT_SYMBOL(jbd2_journal_load);
71 EXPORT_SYMBOL(jbd2_journal_destroy);
72 EXPORT_SYMBOL(jbd2_journal_update_superblock);
73 EXPORT_SYMBOL(jbd2_journal_abort);
74 EXPORT_SYMBOL(jbd2_journal_errno);
75 EXPORT_SYMBOL(jbd2_journal_ack_err);
76 EXPORT_SYMBOL(jbd2_journal_clear_err);
77 EXPORT_SYMBOL(jbd2_log_wait_commit);
78 EXPORT_SYMBOL(jbd2_journal_start_commit);
79 EXPORT_SYMBOL(jbd2_journal_force_commit_nested);
80 EXPORT_SYMBOL(jbd2_journal_wipe);
81 EXPORT_SYMBOL(jbd2_journal_blocks_per_page);
82 EXPORT_SYMBOL(jbd2_journal_invalidatepage);
83 EXPORT_SYMBOL(jbd2_journal_try_to_free_buffers);
84 EXPORT_SYMBOL(jbd2_journal_force_commit);
85
86 static int journal_convert_superblock_v1(journal_t *, journal_superblock_t *);
87 static void __journal_abort_soft (journal_t *journal, int errno);
88
89 /*
90  * Helper function used to manage commit timeouts
91  */
92
93 static void commit_timeout(unsigned long __data)
94 {
95         struct task_struct * p = (struct task_struct *) __data;
96
97         wake_up_process(p);
98 }
99
100 /*
101  * kjournald2: The main thread function used to manage a logging device
102  * journal.
103  *
104  * This kernel thread is responsible for two things:
105  *
106  * 1) COMMIT:  Every so often we need to commit the current state of the
107  *    filesystem to disk.  The journal thread is responsible for writing
108  *    all of the metadata buffers to disk.
109  *
110  * 2) CHECKPOINT: We cannot reuse a used section of the log file until all
111  *    of the data in that part of the log has been rewritten elsewhere on
112  *    the disk.  Flushing these old buffers to reclaim space in the log is
113  *    known as checkpointing, and this thread is responsible for that job.
114  */
115
116 static int kjournald2(void *arg)
117 {
118         journal_t *journal = arg;
119         transaction_t *transaction;
120
121         /*
122          * Set up an interval timer which can be used to trigger a commit wakeup
123          * after the commit interval expires
124          */
125         setup_timer(&journal->j_commit_timer, commit_timeout,
126                         (unsigned long)current);
127
128         /* Record that the journal thread is running */
129         journal->j_task = current;
130         wake_up(&journal->j_wait_done_commit);
131
132         printk(KERN_INFO "kjournald2 starting.  Commit interval %ld seconds\n",
133                         journal->j_commit_interval / HZ);
134
135         /*
136          * And now, wait forever for commit wakeup events.
137          */
138         spin_lock(&journal->j_state_lock);
139
140 loop:
141         if (journal->j_flags & JBD2_UNMOUNT)
142                 goto end_loop;
143
144         jbd_debug(1, "commit_sequence=%d, commit_request=%d\n",
145                 journal->j_commit_sequence, journal->j_commit_request);
146
147         if (journal->j_commit_sequence != journal->j_commit_request) {
148                 jbd_debug(1, "OK, requests differ\n");
149                 spin_unlock(&journal->j_state_lock);
150                 del_timer_sync(&journal->j_commit_timer);
151                 jbd2_journal_commit_transaction(journal);
152                 spin_lock(&journal->j_state_lock);
153                 goto loop;
154         }
155
156         wake_up(&journal->j_wait_done_commit);
157         if (freezing(current)) {
158                 /*
159                  * The simpler the better. Flushing journal isn't a
160                  * good idea, because that depends on threads that may
161                  * be already stopped.
162                  */
163                 jbd_debug(1, "Now suspending kjournald2\n");
164                 spin_unlock(&journal->j_state_lock);
165                 refrigerator();
166                 spin_lock(&journal->j_state_lock);
167         } else {
168                 /*
169                  * We assume on resume that commits are already there,
170                  * so we don't sleep
171                  */
172                 DEFINE_WAIT(wait);
173                 int should_sleep = 1;
174
175                 prepare_to_wait(&journal->j_wait_commit, &wait,
176                                 TASK_INTERRUPTIBLE);
177                 if (journal->j_commit_sequence != journal->j_commit_request)
178                         should_sleep = 0;
179                 transaction = journal->j_running_transaction;
180                 if (transaction && time_after_eq(jiffies,
181                                                 transaction->t_expires))
182                         should_sleep = 0;
183                 if (journal->j_flags & JBD2_UNMOUNT)
184                         should_sleep = 0;
185                 if (should_sleep) {
186                         spin_unlock(&journal->j_state_lock);
187                         schedule();
188                         spin_lock(&journal->j_state_lock);
189                 }
190                 finish_wait(&journal->j_wait_commit, &wait);
191         }
192
193         jbd_debug(1, "kjournald2 wakes\n");
194
195         /*
196          * Were we woken up by a commit wakeup event?
197          */
198         transaction = journal->j_running_transaction;
199         if (transaction && time_after_eq(jiffies, transaction->t_expires)) {
200                 journal->j_commit_request = transaction->t_tid;
201                 jbd_debug(1, "woke because of timeout\n");
202         }
203         goto loop;
204
205 end_loop:
206         spin_unlock(&journal->j_state_lock);
207         del_timer_sync(&journal->j_commit_timer);
208         journal->j_task = NULL;
209         wake_up(&journal->j_wait_done_commit);
210         jbd_debug(1, "Journal thread exiting.\n");
211         return 0;
212 }
213
214 static int jbd2_journal_start_thread(journal_t *journal)
215 {
216         struct task_struct *t;
217
218         t = kthread_run(kjournald2, journal, "kjournald2");
219         if (IS_ERR(t))
220                 return PTR_ERR(t);
221
222         wait_event(journal->j_wait_done_commit, journal->j_task != NULL);
223         return 0;
224 }
225
226 static void journal_kill_thread(journal_t *journal)
227 {
228         spin_lock(&journal->j_state_lock);
229         journal->j_flags |= JBD2_UNMOUNT;
230
231         while (journal->j_task) {
232                 wake_up(&journal->j_wait_commit);
233                 spin_unlock(&journal->j_state_lock);
234                 wait_event(journal->j_wait_done_commit, journal->j_task == NULL);
235                 spin_lock(&journal->j_state_lock);
236         }
237         spin_unlock(&journal->j_state_lock);
238 }
239
240 /*
241  * jbd2_journal_write_metadata_buffer: write a metadata buffer to the journal.
242  *
243  * Writes a metadata buffer to a given disk block.  The actual IO is not
244  * performed but a new buffer_head is constructed which labels the data
245  * to be written with the correct destination disk block.
246  *
247  * Any magic-number escaping which needs to be done will cause a
248  * copy-out here.  If the buffer happens to start with the
249  * JBD2_MAGIC_NUMBER, then we can't write it to the log directly: the
250  * magic number is only written to the log for descripter blocks.  In
251  * this case, we copy the data and replace the first word with 0, and we
252  * return a result code which indicates that this buffer needs to be
253  * marked as an escaped buffer in the corresponding log descriptor
254  * block.  The missing word can then be restored when the block is read
255  * during recovery.
256  *
257  * If the source buffer has already been modified by a new transaction
258  * since we took the last commit snapshot, we use the frozen copy of
259  * that data for IO.  If we end up using the existing buffer_head's data
260  * for the write, then we *have* to lock the buffer to prevent anyone
261  * else from using and possibly modifying it while the IO is in
262  * progress.
263  *
264  * The function returns a pointer to the buffer_heads to be used for IO.
265  *
266  * We assume that the journal has already been locked in this function.
267  *
268  * Return value:
269  *  <0: Error
270  * >=0: Finished OK
271  *
272  * On success:
273  * Bit 0 set == escape performed on the data
274  * Bit 1 set == buffer copy-out performed (kfree the data after IO)
275  */
276
277 int jbd2_journal_write_metadata_buffer(transaction_t *transaction,
278                                   struct journal_head  *jh_in,
279                                   struct journal_head **jh_out,
280                                   unsigned long long blocknr)
281 {
282         int need_copy_out = 0;
283         int done_copy_out = 0;
284         int do_escape = 0;
285         char *mapped_data;
286         struct buffer_head *new_bh;
287         struct journal_head *new_jh;
288         struct page *new_page;
289         unsigned int new_offset;
290         struct buffer_head *bh_in = jh2bh(jh_in);
291
292         /*
293          * The buffer really shouldn't be locked: only the current committing
294          * transaction is allowed to write it, so nobody else is allowed
295          * to do any IO.
296          *
297          * akpm: except if we're journalling data, and write() output is
298          * also part of a shared mapping, and another thread has
299          * decided to launch a writepage() against this buffer.
300          */
301         J_ASSERT_BH(bh_in, buffer_jbddirty(bh_in));
302
303         new_bh = alloc_buffer_head(GFP_NOFS|__GFP_NOFAIL);
304
305         /*
306          * If a new transaction has already done a buffer copy-out, then
307          * we use that version of the data for the commit.
308          */
309         jbd_lock_bh_state(bh_in);
310 repeat:
311         if (jh_in->b_frozen_data) {
312                 done_copy_out = 1;
313                 new_page = virt_to_page(jh_in->b_frozen_data);
314                 new_offset = offset_in_page(jh_in->b_frozen_data);
315         } else {
316                 new_page = jh2bh(jh_in)->b_page;
317                 new_offset = offset_in_page(jh2bh(jh_in)->b_data);
318         }
319
320         mapped_data = kmap_atomic(new_page, KM_USER0);
321         /*
322          * Check for escaping
323          */
324         if (*((__be32 *)(mapped_data + new_offset)) ==
325                                 cpu_to_be32(JBD2_MAGIC_NUMBER)) {
326                 need_copy_out = 1;
327                 do_escape = 1;
328         }
329         kunmap_atomic(mapped_data, KM_USER0);
330
331         /*
332          * Do we need to do a data copy?
333          */
334         if (need_copy_out && !done_copy_out) {
335                 char *tmp;
336
337                 jbd_unlock_bh_state(bh_in);
338                 tmp = jbd2_alloc(bh_in->b_size, GFP_NOFS);
339                 jbd_lock_bh_state(bh_in);
340                 if (jh_in->b_frozen_data) {
341                         jbd2_free(tmp, bh_in->b_size);
342                         goto repeat;
343                 }
344
345                 jh_in->b_frozen_data = tmp;
346                 mapped_data = kmap_atomic(new_page, KM_USER0);
347                 memcpy(tmp, mapped_data + new_offset, jh2bh(jh_in)->b_size);
348                 kunmap_atomic(mapped_data, KM_USER0);
349
350                 new_page = virt_to_page(tmp);
351                 new_offset = offset_in_page(tmp);
352                 done_copy_out = 1;
353         }
354
355         /*
356          * Did we need to do an escaping?  Now we've done all the
357          * copying, we can finally do so.
358          */
359         if (do_escape) {
360                 mapped_data = kmap_atomic(new_page, KM_USER0);
361                 *((unsigned int *)(mapped_data + new_offset)) = 0;
362                 kunmap_atomic(mapped_data, KM_USER0);
363         }
364
365         /* keep subsequent assertions sane */
366         new_bh->b_state = 0;
367         init_buffer(new_bh, NULL, NULL);
368         atomic_set(&new_bh->b_count, 1);
369         jbd_unlock_bh_state(bh_in);
370
371         new_jh = jbd2_journal_add_journal_head(new_bh); /* This sleeps */
372
373         set_bh_page(new_bh, new_page, new_offset);
374         new_jh->b_transaction = NULL;
375         new_bh->b_size = jh2bh(jh_in)->b_size;
376         new_bh->b_bdev = transaction->t_journal->j_dev;
377         new_bh->b_blocknr = blocknr;
378         set_buffer_mapped(new_bh);
379         set_buffer_dirty(new_bh);
380
381         *jh_out = new_jh;
382
383         /*
384          * The to-be-written buffer needs to get moved to the io queue,
385          * and the original buffer whose contents we are shadowing or
386          * copying is moved to the transaction's shadow queue.
387          */
388         JBUFFER_TRACE(jh_in, "file as BJ_Shadow");
389         jbd2_journal_file_buffer(jh_in, transaction, BJ_Shadow);
390         JBUFFER_TRACE(new_jh, "file as BJ_IO");
391         jbd2_journal_file_buffer(new_jh, transaction, BJ_IO);
392
393         return do_escape | (done_copy_out << 1);
394 }
395
396 /*
397  * Allocation code for the journal file.  Manage the space left in the
398  * journal, so that we can begin checkpointing when appropriate.
399  */
400
401 /*
402  * __jbd2_log_space_left: Return the number of free blocks left in the journal.
403  *
404  * Called with the journal already locked.
405  *
406  * Called under j_state_lock
407  */
408
409 int __jbd2_log_space_left(journal_t *journal)
410 {
411         int left = journal->j_free;
412
413         assert_spin_locked(&journal->j_state_lock);
414
415         /*
416          * Be pessimistic here about the number of those free blocks which
417          * might be required for log descriptor control blocks.
418          */
419
420 #define MIN_LOG_RESERVED_BLOCKS 32 /* Allow for rounding errors */
421
422         left -= MIN_LOG_RESERVED_BLOCKS;
423
424         if (left <= 0)
425                 return 0;
426         left -= (left >> 3);
427         return left;
428 }
429
430 /*
431  * Called under j_state_lock.  Returns true if a transaction was started.
432  */
433 int __jbd2_log_start_commit(journal_t *journal, tid_t target)
434 {
435         /*
436          * Are we already doing a recent enough commit?
437          */
438         if (!tid_geq(journal->j_commit_request, target)) {
439                 /*
440                  * We want a new commit: OK, mark the request and wakup the
441                  * commit thread.  We do _not_ do the commit ourselves.
442                  */
443
444                 journal->j_commit_request = target;
445                 jbd_debug(1, "JBD: requesting commit %d/%d\n",
446                           journal->j_commit_request,
447                           journal->j_commit_sequence);
448                 wake_up(&journal->j_wait_commit);
449                 return 1;
450         }
451         return 0;
452 }
453
454 int jbd2_log_start_commit(journal_t *journal, tid_t tid)
455 {
456         int ret;
457
458         spin_lock(&journal->j_state_lock);
459         ret = __jbd2_log_start_commit(journal, tid);
460         spin_unlock(&journal->j_state_lock);
461         return ret;
462 }
463
464 /*
465  * Force and wait upon a commit if the calling process is not within
466  * transaction.  This is used for forcing out undo-protected data which contains
467  * bitmaps, when the fs is running out of space.
468  *
469  * We can only force the running transaction if we don't have an active handle;
470  * otherwise, we will deadlock.
471  *
472  * Returns true if a transaction was started.
473  */
474 int jbd2_journal_force_commit_nested(journal_t *journal)
475 {
476         transaction_t *transaction = NULL;
477         tid_t tid;
478
479         spin_lock(&journal->j_state_lock);
480         if (journal->j_running_transaction && !current->journal_info) {
481                 transaction = journal->j_running_transaction;
482                 __jbd2_log_start_commit(journal, transaction->t_tid);
483         } else if (journal->j_committing_transaction)
484                 transaction = journal->j_committing_transaction;
485
486         if (!transaction) {
487                 spin_unlock(&journal->j_state_lock);
488                 return 0;       /* Nothing to retry */
489         }
490
491         tid = transaction->t_tid;
492         spin_unlock(&journal->j_state_lock);
493         jbd2_log_wait_commit(journal, tid);
494         return 1;
495 }
496
497 /*
498  * Start a commit of the current running transaction (if any).  Returns true
499  * if a transaction was started, and fills its tid in at *ptid
500  */
501 int jbd2_journal_start_commit(journal_t *journal, tid_t *ptid)
502 {
503         int ret = 0;
504
505         spin_lock(&journal->j_state_lock);
506         if (journal->j_running_transaction) {
507                 tid_t tid = journal->j_running_transaction->t_tid;
508
509                 ret = __jbd2_log_start_commit(journal, tid);
510                 if (ret && ptid)
511                         *ptid = tid;
512         } else if (journal->j_committing_transaction && ptid) {
513                 /*
514                  * If ext3_write_super() recently started a commit, then we
515                  * have to wait for completion of that transaction
516                  */
517                 *ptid = journal->j_committing_transaction->t_tid;
518                 ret = 1;
519         }
520         spin_unlock(&journal->j_state_lock);
521         return ret;
522 }
523
524 /*
525  * Wait for a specified commit to complete.
526  * The caller may not hold the journal lock.
527  */
528 int jbd2_log_wait_commit(journal_t *journal, tid_t tid)
529 {
530         int err = 0;
531
532 #ifdef CONFIG_JBD2_DEBUG
533         spin_lock(&journal->j_state_lock);
534         if (!tid_geq(journal->j_commit_request, tid)) {
535                 printk(KERN_EMERG
536                        "%s: error: j_commit_request=%d, tid=%d\n",
537                        __FUNCTION__, journal->j_commit_request, tid);
538         }
539         spin_unlock(&journal->j_state_lock);
540 #endif
541         spin_lock(&journal->j_state_lock);
542         while (tid_gt(tid, journal->j_commit_sequence)) {
543                 jbd_debug(1, "JBD: want %d, j_commit_sequence=%d\n",
544                                   tid, journal->j_commit_sequence);
545                 wake_up(&journal->j_wait_commit);
546                 spin_unlock(&journal->j_state_lock);
547                 wait_event(journal->j_wait_done_commit,
548                                 !tid_gt(tid, journal->j_commit_sequence));
549                 spin_lock(&journal->j_state_lock);
550         }
551         spin_unlock(&journal->j_state_lock);
552
553         if (unlikely(is_journal_aborted(journal))) {
554                 printk(KERN_EMERG "journal commit I/O error\n");
555                 err = -EIO;
556         }
557         return err;
558 }
559
560 /*
561  * Log buffer allocation routines:
562  */
563
564 int jbd2_journal_next_log_block(journal_t *journal, unsigned long long *retp)
565 {
566         unsigned long blocknr;
567
568         spin_lock(&journal->j_state_lock);
569         J_ASSERT(journal->j_free > 1);
570
571         blocknr = journal->j_head;
572         journal->j_head++;
573         journal->j_free--;
574         if (journal->j_head == journal->j_last)
575                 journal->j_head = journal->j_first;
576         spin_unlock(&journal->j_state_lock);
577         return jbd2_journal_bmap(journal, blocknr, retp);
578 }
579
580 /*
581  * Conversion of logical to physical block numbers for the journal
582  *
583  * On external journals the journal blocks are identity-mapped, so
584  * this is a no-op.  If needed, we can use j_blk_offset - everything is
585  * ready.
586  */
587 int jbd2_journal_bmap(journal_t *journal, unsigned long blocknr,
588                  unsigned long long *retp)
589 {
590         int err = 0;
591         unsigned long long ret;
592
593         if (journal->j_inode) {
594                 ret = bmap(journal->j_inode, blocknr);
595                 if (ret)
596                         *retp = ret;
597                 else {
598                         char b[BDEVNAME_SIZE];
599
600                         printk(KERN_ALERT "%s: journal block not found "
601                                         "at offset %lu on %s\n",
602                                 __FUNCTION__,
603                                 blocknr,
604                                 bdevname(journal->j_dev, b));
605                         err = -EIO;
606                         __journal_abort_soft(journal, err);
607                 }
608         } else {
609                 *retp = blocknr; /* +journal->j_blk_offset */
610         }
611         return err;
612 }
613
614 /*
615  * We play buffer_head aliasing tricks to write data/metadata blocks to
616  * the journal without copying their contents, but for journal
617  * descriptor blocks we do need to generate bona fide buffers.
618  *
619  * After the caller of jbd2_journal_get_descriptor_buffer() has finished modifying
620  * the buffer's contents they really should run flush_dcache_page(bh->b_page).
621  * But we don't bother doing that, so there will be coherency problems with
622  * mmaps of blockdevs which hold live JBD-controlled filesystems.
623  */
624 struct journal_head *jbd2_journal_get_descriptor_buffer(journal_t *journal)
625 {
626         struct buffer_head *bh;
627         unsigned long long blocknr;
628         int err;
629
630         err = jbd2_journal_next_log_block(journal, &blocknr);
631
632         if (err)
633                 return NULL;
634
635         bh = __getblk(journal->j_dev, blocknr, journal->j_blocksize);
636         lock_buffer(bh);
637         memset(bh->b_data, 0, journal->j_blocksize);
638         set_buffer_uptodate(bh);
639         unlock_buffer(bh);
640         BUFFER_TRACE(bh, "return this buffer");
641         return jbd2_journal_add_journal_head(bh);
642 }
643
644 struct jbd2_stats_proc_session {
645         journal_t *journal;
646         struct transaction_stats_s *stats;
647         int start;
648         int max;
649 };
650
651 static void *jbd2_history_skip_empty(struct jbd2_stats_proc_session *s,
652                                         struct transaction_stats_s *ts,
653                                         int first)
654 {
655         if (ts == s->stats + s->max)
656                 ts = s->stats;
657         if (!first && ts == s->stats + s->start)
658                 return NULL;
659         while (ts->ts_type == 0) {
660                 ts++;
661                 if (ts == s->stats + s->max)
662                         ts = s->stats;
663                 if (ts == s->stats + s->start)
664                         return NULL;
665         }
666         return ts;
667
668 }
669
670 static void *jbd2_seq_history_start(struct seq_file *seq, loff_t *pos)
671 {
672         struct jbd2_stats_proc_session *s = seq->private;
673         struct transaction_stats_s *ts;
674         int l = *pos;
675
676         if (l == 0)
677                 return SEQ_START_TOKEN;
678         ts = jbd2_history_skip_empty(s, s->stats + s->start, 1);
679         if (!ts)
680                 return NULL;
681         l--;
682         while (l) {
683                 ts = jbd2_history_skip_empty(s, ++ts, 0);
684                 if (!ts)
685                         break;
686                 l--;
687         }
688         return ts;
689 }
690
691 static void *jbd2_seq_history_next(struct seq_file *seq, void *v, loff_t *pos)
692 {
693         struct jbd2_stats_proc_session *s = seq->private;
694         struct transaction_stats_s *ts = v;
695
696         ++*pos;
697         if (v == SEQ_START_TOKEN)
698                 return jbd2_history_skip_empty(s, s->stats + s->start, 1);
699         else
700                 return jbd2_history_skip_empty(s, ++ts, 0);
701 }
702
703 static int jbd2_seq_history_show(struct seq_file *seq, void *v)
704 {
705         struct transaction_stats_s *ts = v;
706         if (v == SEQ_START_TOKEN) {
707                 seq_printf(seq, "%-4s %-5s %-5s %-5s %-5s %-5s %-5s %-6s %-5s "
708                                 "%-5s %-5s %-5s %-5s %-5s\n", "R/C", "tid",
709                                 "wait", "run", "lock", "flush", "log", "hndls",
710                                 "block", "inlog", "ctime", "write", "drop",
711                                 "close");
712                 return 0;
713         }
714         if (ts->ts_type == JBD2_STATS_RUN)
715                 seq_printf(seq, "%-4s %-5lu %-5u %-5u %-5u %-5u %-5u "
716                                 "%-6lu %-5lu %-5lu\n", "R", ts->ts_tid,
717                                 jiffies_to_msecs(ts->u.run.rs_wait),
718                                 jiffies_to_msecs(ts->u.run.rs_running),
719                                 jiffies_to_msecs(ts->u.run.rs_locked),
720                                 jiffies_to_msecs(ts->u.run.rs_flushing),
721                                 jiffies_to_msecs(ts->u.run.rs_logging),
722                                 ts->u.run.rs_handle_count,
723                                 ts->u.run.rs_blocks,
724                                 ts->u.run.rs_blocks_logged);
725         else if (ts->ts_type == JBD2_STATS_CHECKPOINT)
726                 seq_printf(seq, "%-4s %-5lu %48s %-5u %-5lu %-5lu %-5lu\n",
727                                 "C", ts->ts_tid, " ",
728                                 jiffies_to_msecs(ts->u.chp.cs_chp_time),
729                                 ts->u.chp.cs_written, ts->u.chp.cs_dropped,
730                                 ts->u.chp.cs_forced_to_close);
731         else
732                 J_ASSERT(0);
733         return 0;
734 }
735
736 static void jbd2_seq_history_stop(struct seq_file *seq, void *v)
737 {
738 }
739
740 static struct seq_operations jbd2_seq_history_ops = {
741         .start  = jbd2_seq_history_start,
742         .next   = jbd2_seq_history_next,
743         .stop   = jbd2_seq_history_stop,
744         .show   = jbd2_seq_history_show,
745 };
746
747 static int jbd2_seq_history_open(struct inode *inode, struct file *file)
748 {
749         journal_t *journal = PDE(inode)->data;
750         struct jbd2_stats_proc_session *s;
751         int rc, size;
752
753         s = kmalloc(sizeof(*s), GFP_KERNEL);
754         if (s == NULL)
755                 return -ENOMEM;
756         size = sizeof(struct transaction_stats_s) * journal->j_history_max;
757         s->stats = kmalloc(size, GFP_KERNEL);
758         if (s->stats == NULL) {
759                 kfree(s);
760                 return -ENOMEM;
761         }
762         spin_lock(&journal->j_history_lock);
763         memcpy(s->stats, journal->j_history, size);
764         s->max = journal->j_history_max;
765         s->start = journal->j_history_cur % s->max;
766         spin_unlock(&journal->j_history_lock);
767
768         rc = seq_open(file, &jbd2_seq_history_ops);
769         if (rc == 0) {
770                 struct seq_file *m = file->private_data;
771                 m->private = s;
772         } else {
773                 kfree(s->stats);
774                 kfree(s);
775         }
776         return rc;
777
778 }
779
780 static int jbd2_seq_history_release(struct inode *inode, struct file *file)
781 {
782         struct seq_file *seq = file->private_data;
783         struct jbd2_stats_proc_session *s = seq->private;
784
785         kfree(s->stats);
786         kfree(s);
787         return seq_release(inode, file);
788 }
789
790 static struct file_operations jbd2_seq_history_fops = {
791         .owner          = THIS_MODULE,
792         .open           = jbd2_seq_history_open,
793         .read           = seq_read,
794         .llseek         = seq_lseek,
795         .release        = jbd2_seq_history_release,
796 };
797
798 static void *jbd2_seq_info_start(struct seq_file *seq, loff_t *pos)
799 {
800         return *pos ? NULL : SEQ_START_TOKEN;
801 }
802
803 static void *jbd2_seq_info_next(struct seq_file *seq, void *v, loff_t *pos)
804 {
805         return NULL;
806 }
807
808 static int jbd2_seq_info_show(struct seq_file *seq, void *v)
809 {
810         struct jbd2_stats_proc_session *s = seq->private;
811
812         if (v != SEQ_START_TOKEN)
813                 return 0;
814         seq_printf(seq, "%lu transaction, each upto %u blocks\n",
815                         s->stats->ts_tid,
816                         s->journal->j_max_transaction_buffers);
817         if (s->stats->ts_tid == 0)
818                 return 0;
819         seq_printf(seq, "average: \n  %ums waiting for transaction\n",
820             jiffies_to_msecs(s->stats->u.run.rs_wait / s->stats->ts_tid));
821         seq_printf(seq, "  %ums running transaction\n",
822             jiffies_to_msecs(s->stats->u.run.rs_running / s->stats->ts_tid));
823         seq_printf(seq, "  %ums transaction was being locked\n",
824             jiffies_to_msecs(s->stats->u.run.rs_locked / s->stats->ts_tid));
825         seq_printf(seq, "  %ums flushing data (in ordered mode)\n",
826             jiffies_to_msecs(s->stats->u.run.rs_flushing / s->stats->ts_tid));
827         seq_printf(seq, "  %ums logging transaction\n",
828             jiffies_to_msecs(s->stats->u.run.rs_logging / s->stats->ts_tid));
829         seq_printf(seq, "  %lu handles per transaction\n",
830             s->stats->u.run.rs_handle_count / s->stats->ts_tid);
831         seq_printf(seq, "  %lu blocks per transaction\n",
832             s->stats->u.run.rs_blocks / s->stats->ts_tid);
833         seq_printf(seq, "  %lu logged blocks per transaction\n",
834             s->stats->u.run.rs_blocks_logged / s->stats->ts_tid);
835         return 0;
836 }
837
838 static void jbd2_seq_info_stop(struct seq_file *seq, void *v)
839 {
840 }
841
842 static struct seq_operations jbd2_seq_info_ops = {
843         .start  = jbd2_seq_info_start,
844         .next   = jbd2_seq_info_next,
845         .stop   = jbd2_seq_info_stop,
846         .show   = jbd2_seq_info_show,
847 };
848
849 static int jbd2_seq_info_open(struct inode *inode, struct file *file)
850 {
851         journal_t *journal = PDE(inode)->data;
852         struct jbd2_stats_proc_session *s;
853         int rc, size;
854
855         s = kmalloc(sizeof(*s), GFP_KERNEL);
856         if (s == NULL)
857                 return -ENOMEM;
858         size = sizeof(struct transaction_stats_s);
859         s->stats = kmalloc(size, GFP_KERNEL);
860         if (s->stats == NULL) {
861                 kfree(s);
862                 return -ENOMEM;
863         }
864         spin_lock(&journal->j_history_lock);
865         memcpy(s->stats, &journal->j_stats, size);
866         s->journal = journal;
867         spin_unlock(&journal->j_history_lock);
868
869         rc = seq_open(file, &jbd2_seq_info_ops);
870         if (rc == 0) {
871                 struct seq_file *m = file->private_data;
872                 m->private = s;
873         } else {
874                 kfree(s->stats);
875                 kfree(s);
876         }
877         return rc;
878
879 }
880
881 static int jbd2_seq_info_release(struct inode *inode, struct file *file)
882 {
883         struct seq_file *seq = file->private_data;
884         struct jbd2_stats_proc_session *s = seq->private;
885         kfree(s->stats);
886         kfree(s);
887         return seq_release(inode, file);
888 }
889
890 static struct file_operations jbd2_seq_info_fops = {
891         .owner          = THIS_MODULE,
892         .open           = jbd2_seq_info_open,
893         .read           = seq_read,
894         .llseek         = seq_lseek,
895         .release        = jbd2_seq_info_release,
896 };
897
898 static struct proc_dir_entry *proc_jbd2_stats;
899
900 static void jbd2_stats_proc_init(journal_t *journal)
901 {
902         char name[BDEVNAME_SIZE];
903
904         snprintf(name, sizeof(name) - 1, "%s", bdevname(journal->j_dev, name));
905         journal->j_proc_entry = proc_mkdir(name, proc_jbd2_stats);
906         if (journal->j_proc_entry) {
907                 struct proc_dir_entry *p;
908                 p = create_proc_entry("history", S_IRUGO,
909                                 journal->j_proc_entry);
910                 if (p) {
911                         p->proc_fops = &jbd2_seq_history_fops;
912                         p->data = journal;
913                         p = create_proc_entry("info", S_IRUGO,
914                                                 journal->j_proc_entry);
915                         if (p) {
916                                 p->proc_fops = &jbd2_seq_info_fops;
917                                 p->data = journal;
918                         }
919                 }
920         }
921 }
922
923 static void jbd2_stats_proc_exit(journal_t *journal)
924 {
925         char name[BDEVNAME_SIZE];
926
927         snprintf(name, sizeof(name) - 1, "%s", bdevname(journal->j_dev, name));
928         remove_proc_entry("info", journal->j_proc_entry);
929         remove_proc_entry("history", journal->j_proc_entry);
930         remove_proc_entry(name, proc_jbd2_stats);
931 }
932
933 static void journal_init_stats(journal_t *journal)
934 {
935         int size;
936
937         if (!proc_jbd2_stats)
938                 return;
939
940         journal->j_history_max = 100;
941         size = sizeof(struct transaction_stats_s) * journal->j_history_max;
942         journal->j_history = kzalloc(size, GFP_KERNEL);
943         if (!journal->j_history) {
944                 journal->j_history_max = 0;
945                 return;
946         }
947         spin_lock_init(&journal->j_history_lock);
948 }
949
950 /*
951  * Management for journal control blocks: functions to create and
952  * destroy journal_t structures, and to initialise and read existing
953  * journal blocks from disk.  */
954
955 /* First: create and setup a journal_t object in memory.  We initialise
956  * very few fields yet: that has to wait until we have created the
957  * journal structures from from scratch, or loaded them from disk. */
958
959 static journal_t * journal_init_common (void)
960 {
961         journal_t *journal;
962         int err;
963
964         journal = kzalloc(sizeof(*journal), GFP_KERNEL|__GFP_NOFAIL);
965         if (!journal)
966                 goto fail;
967
968         init_waitqueue_head(&journal->j_wait_transaction_locked);
969         init_waitqueue_head(&journal->j_wait_logspace);
970         init_waitqueue_head(&journal->j_wait_done_commit);
971         init_waitqueue_head(&journal->j_wait_checkpoint);
972         init_waitqueue_head(&journal->j_wait_commit);
973         init_waitqueue_head(&journal->j_wait_updates);
974         mutex_init(&journal->j_barrier);
975         mutex_init(&journal->j_checkpoint_mutex);
976         spin_lock_init(&journal->j_revoke_lock);
977         spin_lock_init(&journal->j_list_lock);
978         spin_lock_init(&journal->j_state_lock);
979
980         journal->j_commit_interval = (HZ * JBD2_DEFAULT_MAX_COMMIT_AGE);
981
982         /* The journal is marked for error until we succeed with recovery! */
983         journal->j_flags = JBD2_ABORT;
984
985         /* Set up a default-sized revoke table for the new mount. */
986         err = jbd2_journal_init_revoke(journal, JOURNAL_REVOKE_DEFAULT_HASH);
987         if (err) {
988                 kfree(journal);
989                 goto fail;
990         }
991
992         journal_init_stats(journal);
993
994         return journal;
995 fail:
996         return NULL;
997 }
998
999 /* jbd2_journal_init_dev and jbd2_journal_init_inode:
1000  *
1001  * Create a journal structure assigned some fixed set of disk blocks to
1002  * the journal.  We don't actually touch those disk blocks yet, but we
1003  * need to set up all of the mapping information to tell the journaling
1004  * system where the journal blocks are.
1005  *
1006  */
1007
1008 /**
1009  *  journal_t * jbd2_journal_init_dev() - creates an initialises a journal structure
1010  *  @bdev: Block device on which to create the journal
1011  *  @fs_dev: Device which hold journalled filesystem for this journal.
1012  *  @start: Block nr Start of journal.
1013  *  @len:  Length of the journal in blocks.
1014  *  @blocksize: blocksize of journalling device
1015  *  @returns: a newly created journal_t *
1016  *
1017  *  jbd2_journal_init_dev creates a journal which maps a fixed contiguous
1018  *  range of blocks on an arbitrary block device.
1019  *
1020  */
1021 journal_t * jbd2_journal_init_dev(struct block_device *bdev,
1022                         struct block_device *fs_dev,
1023                         unsigned long long start, int len, int blocksize)
1024 {
1025         journal_t *journal = journal_init_common();
1026         struct buffer_head *bh;
1027         int n;
1028
1029         if (!journal)
1030                 return NULL;
1031
1032         /* journal descriptor can store up to n blocks -bzzz */
1033         journal->j_blocksize = blocksize;
1034         n = journal->j_blocksize / sizeof(journal_block_tag_t);
1035         journal->j_wbufsize = n;
1036         journal->j_wbuf = kmalloc(n * sizeof(struct buffer_head*), GFP_KERNEL);
1037         if (!journal->j_wbuf) {
1038                 printk(KERN_ERR "%s: Cant allocate bhs for commit thread\n",
1039                         __FUNCTION__);
1040                 kfree(journal);
1041                 journal = NULL;
1042                 goto out;
1043         }
1044         journal->j_dev = bdev;
1045         journal->j_fs_dev = fs_dev;
1046         journal->j_blk_offset = start;
1047         journal->j_maxlen = len;
1048         jbd2_stats_proc_init(journal);
1049
1050         bh = __getblk(journal->j_dev, start, journal->j_blocksize);
1051         J_ASSERT(bh != NULL);
1052         journal->j_sb_buffer = bh;
1053         journal->j_superblock = (journal_superblock_t *)bh->b_data;
1054 out:
1055         return journal;
1056 }
1057
1058 /**
1059  *  journal_t * jbd2_journal_init_inode () - creates a journal which maps to a inode.
1060  *  @inode: An inode to create the journal in
1061  *
1062  * jbd2_journal_init_inode creates a journal which maps an on-disk inode as
1063  * the journal.  The inode must exist already, must support bmap() and
1064  * must have all data blocks preallocated.
1065  */
1066 journal_t * jbd2_journal_init_inode (struct inode *inode)
1067 {
1068         struct buffer_head *bh;
1069         journal_t *journal = journal_init_common();
1070         int err;
1071         int n;
1072         unsigned long long blocknr;
1073
1074         if (!journal)
1075                 return NULL;
1076
1077         journal->j_dev = journal->j_fs_dev = inode->i_sb->s_bdev;
1078         journal->j_inode = inode;
1079         jbd_debug(1,
1080                   "journal %p: inode %s/%ld, size %Ld, bits %d, blksize %ld\n",
1081                   journal, inode->i_sb->s_id, inode->i_ino,
1082                   (long long) inode->i_size,
1083                   inode->i_sb->s_blocksize_bits, inode->i_sb->s_blocksize);
1084
1085         journal->j_maxlen = inode->i_size >> inode->i_sb->s_blocksize_bits;
1086         journal->j_blocksize = inode->i_sb->s_blocksize;
1087         jbd2_stats_proc_init(journal);
1088
1089         /* journal descriptor can store up to n blocks -bzzz */
1090         n = journal->j_blocksize / sizeof(journal_block_tag_t);
1091         journal->j_wbufsize = n;
1092         journal->j_wbuf = kmalloc(n * sizeof(struct buffer_head*), GFP_KERNEL);
1093         if (!journal->j_wbuf) {
1094                 printk(KERN_ERR "%s: Cant allocate bhs for commit thread\n",
1095                         __FUNCTION__);
1096                 kfree(journal);
1097                 return NULL;
1098         }
1099
1100         err = jbd2_journal_bmap(journal, 0, &blocknr);
1101         /* If that failed, give up */
1102         if (err) {
1103                 printk(KERN_ERR "%s: Cannnot locate journal superblock\n",
1104                        __FUNCTION__);
1105                 kfree(journal);
1106                 return NULL;
1107         }
1108
1109         bh = __getblk(journal->j_dev, blocknr, journal->j_blocksize);
1110         J_ASSERT(bh != NULL);
1111         journal->j_sb_buffer = bh;
1112         journal->j_superblock = (journal_superblock_t *)bh->b_data;
1113
1114         return journal;
1115 }
1116
1117 /*
1118  * If the journal init or create aborts, we need to mark the journal
1119  * superblock as being NULL to prevent the journal destroy from writing
1120  * back a bogus superblock.
1121  */
1122 static void journal_fail_superblock (journal_t *journal)
1123 {
1124         struct buffer_head *bh = journal->j_sb_buffer;
1125         brelse(bh);
1126         journal->j_sb_buffer = NULL;
1127 }
1128
1129 /*
1130  * Given a journal_t structure, initialise the various fields for
1131  * startup of a new journaling session.  We use this both when creating
1132  * a journal, and after recovering an old journal to reset it for
1133  * subsequent use.
1134  */
1135
1136 static int journal_reset(journal_t *journal)
1137 {
1138         journal_superblock_t *sb = journal->j_superblock;
1139         unsigned long long first, last;
1140
1141         first = be32_to_cpu(sb->s_first);
1142         last = be32_to_cpu(sb->s_maxlen);
1143
1144         journal->j_first = first;
1145         journal->j_last = last;
1146
1147         journal->j_head = first;
1148         journal->j_tail = first;
1149         journal->j_free = last - first;
1150
1151         journal->j_tail_sequence = journal->j_transaction_sequence;
1152         journal->j_commit_sequence = journal->j_transaction_sequence - 1;
1153         journal->j_commit_request = journal->j_commit_sequence;
1154
1155         journal->j_max_transaction_buffers = journal->j_maxlen / 4;
1156
1157         /* Add the dynamic fields and write it to disk. */
1158         jbd2_journal_update_superblock(journal, 1);
1159         return jbd2_journal_start_thread(journal);
1160 }
1161
1162 /**
1163  * int jbd2_journal_create() - Initialise the new journal file
1164  * @journal: Journal to create. This structure must have been initialised
1165  *
1166  * Given a journal_t structure which tells us which disk blocks we can
1167  * use, create a new journal superblock and initialise all of the
1168  * journal fields from scratch.
1169  **/
1170 int jbd2_journal_create(journal_t *journal)
1171 {
1172         unsigned long long blocknr;
1173         struct buffer_head *bh;
1174         journal_superblock_t *sb;
1175         int i, err;
1176
1177         if (journal->j_maxlen < JBD2_MIN_JOURNAL_BLOCKS) {
1178                 printk (KERN_ERR "Journal length (%d blocks) too short.\n",
1179                         journal->j_maxlen);
1180                 journal_fail_superblock(journal);
1181                 return -EINVAL;
1182         }
1183
1184         if (journal->j_inode == NULL) {
1185                 /*
1186                  * We don't know what block to start at!
1187                  */
1188                 printk(KERN_EMERG
1189                        "%s: creation of journal on external device!\n",
1190                        __FUNCTION__);
1191                 BUG();
1192         }
1193
1194         /* Zero out the entire journal on disk.  We cannot afford to
1195            have any blocks on disk beginning with JBD2_MAGIC_NUMBER. */
1196         jbd_debug(1, "JBD: Zeroing out journal blocks...\n");
1197         for (i = 0; i < journal->j_maxlen; i++) {
1198                 err = jbd2_journal_bmap(journal, i, &blocknr);
1199                 if (err)
1200                         return err;
1201                 bh = __getblk(journal->j_dev, blocknr, journal->j_blocksize);
1202                 lock_buffer(bh);
1203                 memset (bh->b_data, 0, journal->j_blocksize);
1204                 BUFFER_TRACE(bh, "marking dirty");
1205                 mark_buffer_dirty(bh);
1206                 BUFFER_TRACE(bh, "marking uptodate");
1207                 set_buffer_uptodate(bh);
1208                 unlock_buffer(bh);
1209                 __brelse(bh);
1210         }
1211
1212         sync_blockdev(journal->j_dev);
1213         jbd_debug(1, "JBD: journal cleared.\n");
1214
1215         /* OK, fill in the initial static fields in the new superblock */
1216         sb = journal->j_superblock;
1217
1218         sb->s_header.h_magic     = cpu_to_be32(JBD2_MAGIC_NUMBER);
1219         sb->s_header.h_blocktype = cpu_to_be32(JBD2_SUPERBLOCK_V2);
1220
1221         sb->s_blocksize = cpu_to_be32(journal->j_blocksize);
1222         sb->s_maxlen    = cpu_to_be32(journal->j_maxlen);
1223         sb->s_first     = cpu_to_be32(1);
1224
1225         journal->j_transaction_sequence = 1;
1226
1227         journal->j_flags &= ~JBD2_ABORT;
1228         journal->j_format_version = 2;
1229
1230         return journal_reset(journal);
1231 }
1232
1233 /**
1234  * void jbd2_journal_update_superblock() - Update journal sb on disk.
1235  * @journal: The journal to update.
1236  * @wait: Set to '0' if you don't want to wait for IO completion.
1237  *
1238  * Update a journal's dynamic superblock fields and write it to disk,
1239  * optionally waiting for the IO to complete.
1240  */
1241 void jbd2_journal_update_superblock(journal_t *journal, int wait)
1242 {
1243         journal_superblock_t *sb = journal->j_superblock;
1244         struct buffer_head *bh = journal->j_sb_buffer;
1245
1246         /*
1247          * As a special case, if the on-disk copy is already marked as needing
1248          * no recovery (s_start == 0) and there are no outstanding transactions
1249          * in the filesystem, then we can safely defer the superblock update
1250          * until the next commit by setting JBD2_FLUSHED.  This avoids
1251          * attempting a write to a potential-readonly device.
1252          */
1253         if (sb->s_start == 0 && journal->j_tail_sequence ==
1254                                 journal->j_transaction_sequence) {
1255                 jbd_debug(1,"JBD: Skipping superblock update on recovered sb "
1256                         "(start %ld, seq %d, errno %d)\n",
1257                         journal->j_tail, journal->j_tail_sequence,
1258                         journal->j_errno);
1259                 goto out;
1260         }
1261
1262         spin_lock(&journal->j_state_lock);
1263         jbd_debug(1,"JBD: updating superblock (start %ld, seq %d, errno %d)\n",
1264                   journal->j_tail, journal->j_tail_sequence, journal->j_errno);
1265
1266         sb->s_sequence = cpu_to_be32(journal->j_tail_sequence);
1267         sb->s_start    = cpu_to_be32(journal->j_tail);
1268         sb->s_errno    = cpu_to_be32(journal->j_errno);
1269         spin_unlock(&journal->j_state_lock);
1270
1271         BUFFER_TRACE(bh, "marking dirty");
1272         mark_buffer_dirty(bh);
1273         if (wait)
1274                 sync_dirty_buffer(bh);
1275         else
1276                 ll_rw_block(SWRITE, 1, &bh);
1277
1278 out:
1279         /* If we have just flushed the log (by marking s_start==0), then
1280          * any future commit will have to be careful to update the
1281          * superblock again to re-record the true start of the log. */
1282
1283         spin_lock(&journal->j_state_lock);
1284         if (sb->s_start)
1285                 journal->j_flags &= ~JBD2_FLUSHED;
1286         else
1287                 journal->j_flags |= JBD2_FLUSHED;
1288         spin_unlock(&journal->j_state_lock);
1289 }
1290
1291 /*
1292  * Read the superblock for a given journal, performing initial
1293  * validation of the format.
1294  */
1295
1296 static int journal_get_superblock(journal_t *journal)
1297 {
1298         struct buffer_head *bh;
1299         journal_superblock_t *sb;
1300         int err = -EIO;
1301
1302         bh = journal->j_sb_buffer;
1303
1304         J_ASSERT(bh != NULL);
1305         if (!buffer_uptodate(bh)) {
1306                 ll_rw_block(READ, 1, &bh);
1307                 wait_on_buffer(bh);
1308                 if (!buffer_uptodate(bh)) {
1309                         printk (KERN_ERR
1310                                 "JBD: IO error reading journal superblock\n");
1311                         goto out;
1312                 }
1313         }
1314
1315         sb = journal->j_superblock;
1316
1317         err = -EINVAL;
1318
1319         if (sb->s_header.h_magic != cpu_to_be32(JBD2_MAGIC_NUMBER) ||
1320             sb->s_blocksize != cpu_to_be32(journal->j_blocksize)) {
1321                 printk(KERN_WARNING "JBD: no valid journal superblock found\n");
1322                 goto out;
1323         }
1324
1325         switch(be32_to_cpu(sb->s_header.h_blocktype)) {
1326         case JBD2_SUPERBLOCK_V1:
1327                 journal->j_format_version = 1;
1328                 break;
1329         case JBD2_SUPERBLOCK_V2:
1330                 journal->j_format_version = 2;
1331                 break;
1332         default:
1333                 printk(KERN_WARNING "JBD: unrecognised superblock format ID\n");
1334                 goto out;
1335         }
1336
1337         if (be32_to_cpu(sb->s_maxlen) < journal->j_maxlen)
1338                 journal->j_maxlen = be32_to_cpu(sb->s_maxlen);
1339         else if (be32_to_cpu(sb->s_maxlen) > journal->j_maxlen) {
1340                 printk (KERN_WARNING "JBD: journal file too short\n");
1341                 goto out;
1342         }
1343
1344         return 0;
1345
1346 out:
1347         journal_fail_superblock(journal);
1348         return err;
1349 }
1350
1351 /*
1352  * Load the on-disk journal superblock and read the key fields into the
1353  * journal_t.
1354  */
1355
1356 static int load_superblock(journal_t *journal)
1357 {
1358         int err;
1359         journal_superblock_t *sb;
1360
1361         err = journal_get_superblock(journal);
1362         if (err)
1363                 return err;
1364
1365         sb = journal->j_superblock;
1366
1367         journal->j_tail_sequence = be32_to_cpu(sb->s_sequence);
1368         journal->j_tail = be32_to_cpu(sb->s_start);
1369         journal->j_first = be32_to_cpu(sb->s_first);
1370         journal->j_last = be32_to_cpu(sb->s_maxlen);
1371         journal->j_errno = be32_to_cpu(sb->s_errno);
1372
1373         return 0;
1374 }
1375
1376
1377 /**
1378  * int jbd2_journal_load() - Read journal from disk.
1379  * @journal: Journal to act on.
1380  *
1381  * Given a journal_t structure which tells us which disk blocks contain
1382  * a journal, read the journal from disk to initialise the in-memory
1383  * structures.
1384  */
1385 int jbd2_journal_load(journal_t *journal)
1386 {
1387         int err;
1388         journal_superblock_t *sb;
1389
1390         err = load_superblock(journal);
1391         if (err)
1392                 return err;
1393
1394         sb = journal->j_superblock;
1395         /* If this is a V2 superblock, then we have to check the
1396          * features flags on it. */
1397
1398         if (journal->j_format_version >= 2) {
1399                 if ((sb->s_feature_ro_compat &
1400                      ~cpu_to_be32(JBD2_KNOWN_ROCOMPAT_FEATURES)) ||
1401                     (sb->s_feature_incompat &
1402                      ~cpu_to_be32(JBD2_KNOWN_INCOMPAT_FEATURES))) {
1403                         printk (KERN_WARNING
1404                                 "JBD: Unrecognised features on journal\n");
1405                         return -EINVAL;
1406                 }
1407         }
1408
1409         /* Let the recovery code check whether it needs to recover any
1410          * data from the journal. */
1411         if (jbd2_journal_recover(journal))
1412                 goto recovery_error;
1413
1414         /* OK, we've finished with the dynamic journal bits:
1415          * reinitialise the dynamic contents of the superblock in memory
1416          * and reset them on disk. */
1417         if (journal_reset(journal))
1418                 goto recovery_error;
1419
1420         journal->j_flags &= ~JBD2_ABORT;
1421         journal->j_flags |= JBD2_LOADED;
1422         return 0;
1423
1424 recovery_error:
1425         printk (KERN_WARNING "JBD: recovery failed\n");
1426         return -EIO;
1427 }
1428
1429 /**
1430  * void jbd2_journal_destroy() - Release a journal_t structure.
1431  * @journal: Journal to act on.
1432  *
1433  * Release a journal_t structure once it is no longer in use by the
1434  * journaled object.
1435  */
1436 void jbd2_journal_destroy(journal_t *journal)
1437 {
1438         /* Wait for the commit thread to wake up and die. */
1439         journal_kill_thread(journal);
1440
1441         /* Force a final log commit */
1442         if (journal->j_running_transaction)
1443                 jbd2_journal_commit_transaction(journal);
1444
1445         /* Force any old transactions to disk */
1446
1447         /* Totally anal locking here... */
1448         spin_lock(&journal->j_list_lock);
1449         while (journal->j_checkpoint_transactions != NULL) {
1450                 spin_unlock(&journal->j_list_lock);
1451                 jbd2_log_do_checkpoint(journal);
1452                 spin_lock(&journal->j_list_lock);
1453         }
1454
1455         J_ASSERT(journal->j_running_transaction == NULL);
1456         J_ASSERT(journal->j_committing_transaction == NULL);
1457         J_ASSERT(journal->j_checkpoint_transactions == NULL);
1458         spin_unlock(&journal->j_list_lock);
1459
1460         /* We can now mark the journal as empty. */
1461         journal->j_tail = 0;
1462         journal->j_tail_sequence = ++journal->j_transaction_sequence;
1463         if (journal->j_sb_buffer) {
1464                 jbd2_journal_update_superblock(journal, 1);
1465                 brelse(journal->j_sb_buffer);
1466         }
1467
1468         if (journal->j_proc_entry)
1469                 jbd2_stats_proc_exit(journal);
1470         if (journal->j_inode)
1471                 iput(journal->j_inode);
1472         if (journal->j_revoke)
1473                 jbd2_journal_destroy_revoke(journal);
1474         kfree(journal->j_wbuf);
1475         kfree(journal);
1476 }
1477
1478
1479 /**
1480  *int jbd2_journal_check_used_features () - Check if features specified are used.
1481  * @journal: Journal to check.
1482  * @compat: bitmask of compatible features
1483  * @ro: bitmask of features that force read-only mount
1484  * @incompat: bitmask of incompatible features
1485  *
1486  * Check whether the journal uses all of a given set of
1487  * features.  Return true (non-zero) if it does.
1488  **/
1489
1490 int jbd2_journal_check_used_features (journal_t *journal, unsigned long compat,
1491                                  unsigned long ro, unsigned long incompat)
1492 {
1493         journal_superblock_t *sb;
1494
1495         if (!compat && !ro && !incompat)
1496                 return 1;
1497         if (journal->j_format_version == 1)
1498                 return 0;
1499
1500         sb = journal->j_superblock;
1501
1502         if (((be32_to_cpu(sb->s_feature_compat) & compat) == compat) &&
1503             ((be32_to_cpu(sb->s_feature_ro_compat) & ro) == ro) &&
1504             ((be32_to_cpu(sb->s_feature_incompat) & incompat) == incompat))
1505                 return 1;
1506
1507         return 0;
1508 }
1509
1510 /**
1511  * int jbd2_journal_check_available_features() - Check feature set in journalling layer
1512  * @journal: Journal to check.
1513  * @compat: bitmask of compatible features
1514  * @ro: bitmask of features that force read-only mount
1515  * @incompat: bitmask of incompatible features
1516  *
1517  * Check whether the journaling code supports the use of
1518  * all of a given set of features on this journal.  Return true
1519  * (non-zero) if it can. */
1520
1521 int jbd2_journal_check_available_features (journal_t *journal, unsigned long compat,
1522                                       unsigned long ro, unsigned long incompat)
1523 {
1524         journal_superblock_t *sb;
1525
1526         if (!compat && !ro && !incompat)
1527                 return 1;
1528
1529         sb = journal->j_superblock;
1530
1531         /* We can support any known requested features iff the
1532          * superblock is in version 2.  Otherwise we fail to support any
1533          * extended sb features. */
1534
1535         if (journal->j_format_version != 2)
1536                 return 0;
1537
1538         if ((compat   & JBD2_KNOWN_COMPAT_FEATURES) == compat &&
1539             (ro       & JBD2_KNOWN_ROCOMPAT_FEATURES) == ro &&
1540             (incompat & JBD2_KNOWN_INCOMPAT_FEATURES) == incompat)
1541                 return 1;
1542
1543         return 0;
1544 }
1545
1546 /**
1547  * int jbd2_journal_set_features () - Mark a given journal feature in the superblock
1548  * @journal: Journal to act on.
1549  * @compat: bitmask of compatible features
1550  * @ro: bitmask of features that force read-only mount
1551  * @incompat: bitmask of incompatible features
1552  *
1553  * Mark a given journal feature as present on the
1554  * superblock.  Returns true if the requested features could be set.
1555  *
1556  */
1557
1558 int jbd2_journal_set_features (journal_t *journal, unsigned long compat,
1559                           unsigned long ro, unsigned long incompat)
1560 {
1561         journal_superblock_t *sb;
1562
1563         if (jbd2_journal_check_used_features(journal, compat, ro, incompat))
1564                 return 1;
1565
1566         if (!jbd2_journal_check_available_features(journal, compat, ro, incompat))
1567                 return 0;
1568
1569         jbd_debug(1, "Setting new features 0x%lx/0x%lx/0x%lx\n",
1570                   compat, ro, incompat);
1571
1572         sb = journal->j_superblock;
1573
1574         sb->s_feature_compat    |= cpu_to_be32(compat);
1575         sb->s_feature_ro_compat |= cpu_to_be32(ro);
1576         sb->s_feature_incompat  |= cpu_to_be32(incompat);
1577
1578         return 1;
1579 }
1580
1581 /*
1582  * jbd2_journal_clear_features () - Clear a given journal feature in the
1583  *                                  superblock
1584  * @journal: Journal to act on.
1585  * @compat: bitmask of compatible features
1586  * @ro: bitmask of features that force read-only mount
1587  * @incompat: bitmask of incompatible features
1588  *
1589  * Clear a given journal feature as present on the
1590  * superblock.
1591  */
1592 void jbd2_journal_clear_features(journal_t *journal, unsigned long compat,
1593                                 unsigned long ro, unsigned long incompat)
1594 {
1595         journal_superblock_t *sb;
1596
1597         jbd_debug(1, "Clear features 0x%lx/0x%lx/0x%lx\n",
1598                   compat, ro, incompat);
1599
1600         sb = journal->j_superblock;
1601
1602         sb->s_feature_compat    &= ~cpu_to_be32(compat);
1603         sb->s_feature_ro_compat &= ~cpu_to_be32(ro);
1604         sb->s_feature_incompat  &= ~cpu_to_be32(incompat);
1605 }
1606 EXPORT_SYMBOL(jbd2_journal_clear_features);
1607
1608 /**
1609  * int jbd2_journal_update_format () - Update on-disk journal structure.
1610  * @journal: Journal to act on.
1611  *
1612  * Given an initialised but unloaded journal struct, poke about in the
1613  * on-disk structure to update it to the most recent supported version.
1614  */
1615 int jbd2_journal_update_format (journal_t *journal)
1616 {
1617         journal_superblock_t *sb;
1618         int err;
1619
1620         err = journal_get_superblock(journal);
1621         if (err)
1622                 return err;
1623
1624         sb = journal->j_superblock;
1625
1626         switch (be32_to_cpu(sb->s_header.h_blocktype)) {
1627         case JBD2_SUPERBLOCK_V2:
1628                 return 0;
1629         case JBD2_SUPERBLOCK_V1:
1630                 return journal_convert_superblock_v1(journal, sb);
1631         default:
1632                 break;
1633         }
1634         return -EINVAL;
1635 }
1636
1637 static int journal_convert_superblock_v1(journal_t *journal,
1638                                          journal_superblock_t *sb)
1639 {
1640         int offset, blocksize;
1641         struct buffer_head *bh;
1642
1643         printk(KERN_WARNING
1644                 "JBD: Converting superblock from version 1 to 2.\n");
1645
1646         /* Pre-initialise new fields to zero */
1647         offset = ((char *) &(sb->s_feature_compat)) - ((char *) sb);
1648         blocksize = be32_to_cpu(sb->s_blocksize);
1649         memset(&sb->s_feature_compat, 0, blocksize-offset);
1650
1651         sb->s_nr_users = cpu_to_be32(1);
1652         sb->s_header.h_blocktype = cpu_to_be32(JBD2_SUPERBLOCK_V2);
1653         journal->j_format_version = 2;
1654
1655         bh = journal->j_sb_buffer;
1656         BUFFER_TRACE(bh, "marking dirty");
1657         mark_buffer_dirty(bh);
1658         sync_dirty_buffer(bh);
1659         return 0;
1660 }
1661
1662
1663 /**
1664  * int jbd2_journal_flush () - Flush journal
1665  * @journal: Journal to act on.
1666  *
1667  * Flush all data for a given journal to disk and empty the journal.
1668  * Filesystems can use this when remounting readonly to ensure that
1669  * recovery does not need to happen on remount.
1670  */
1671
1672 int jbd2_journal_flush(journal_t *journal)
1673 {
1674         int err = 0;
1675         transaction_t *transaction = NULL;
1676         unsigned long old_tail;
1677
1678         spin_lock(&journal->j_state_lock);
1679
1680         /* Force everything buffered to the log... */
1681         if (journal->j_running_transaction) {
1682                 transaction = journal->j_running_transaction;
1683                 __jbd2_log_start_commit(journal, transaction->t_tid);
1684         } else if (journal->j_committing_transaction)
1685                 transaction = journal->j_committing_transaction;
1686
1687         /* Wait for the log commit to complete... */
1688         if (transaction) {
1689                 tid_t tid = transaction->t_tid;
1690
1691                 spin_unlock(&journal->j_state_lock);
1692                 jbd2_log_wait_commit(journal, tid);
1693         } else {
1694                 spin_unlock(&journal->j_state_lock);
1695         }
1696
1697         /* ...and flush everything in the log out to disk. */
1698         spin_lock(&journal->j_list_lock);
1699         while (!err && journal->j_checkpoint_transactions != NULL) {
1700                 spin_unlock(&journal->j_list_lock);
1701                 err = jbd2_log_do_checkpoint(journal);
1702                 spin_lock(&journal->j_list_lock);
1703         }
1704         spin_unlock(&journal->j_list_lock);
1705         jbd2_cleanup_journal_tail(journal);
1706
1707         /* Finally, mark the journal as really needing no recovery.
1708          * This sets s_start==0 in the underlying superblock, which is
1709          * the magic code for a fully-recovered superblock.  Any future
1710          * commits of data to the journal will restore the current
1711          * s_start value. */
1712         spin_lock(&journal->j_state_lock);
1713         old_tail = journal->j_tail;
1714         journal->j_tail = 0;
1715         spin_unlock(&journal->j_state_lock);
1716         jbd2_journal_update_superblock(journal, 1);
1717         spin_lock(&journal->j_state_lock);
1718         journal->j_tail = old_tail;
1719
1720         J_ASSERT(!journal->j_running_transaction);
1721         J_ASSERT(!journal->j_committing_transaction);
1722         J_ASSERT(!journal->j_checkpoint_transactions);
1723         J_ASSERT(journal->j_head == journal->j_tail);
1724         J_ASSERT(journal->j_tail_sequence == journal->j_transaction_sequence);
1725         spin_unlock(&journal->j_state_lock);
1726         return err;
1727 }
1728
1729 /**
1730  * int jbd2_journal_wipe() - Wipe journal contents
1731  * @journal: Journal to act on.
1732  * @write: flag (see below)
1733  *
1734  * Wipe out all of the contents of a journal, safely.  This will produce
1735  * a warning if the journal contains any valid recovery information.
1736  * Must be called between journal_init_*() and jbd2_journal_load().
1737  *
1738  * If 'write' is non-zero, then we wipe out the journal on disk; otherwise
1739  * we merely suppress recovery.
1740  */
1741
1742 int jbd2_journal_wipe(journal_t *journal, int write)
1743 {
1744         journal_superblock_t *sb;
1745         int err = 0;
1746
1747         J_ASSERT (!(journal->j_flags & JBD2_LOADED));
1748
1749         err = load_superblock(journal);
1750         if (err)
1751                 return err;
1752
1753         sb = journal->j_superblock;
1754
1755         if (!journal->j_tail)
1756                 goto no_recovery;
1757
1758         printk (KERN_WARNING "JBD: %s recovery information on journal\n",
1759                 write ? "Clearing" : "Ignoring");
1760
1761         err = jbd2_journal_skip_recovery(journal);
1762         if (write)
1763                 jbd2_journal_update_superblock(journal, 1);
1764
1765  no_recovery:
1766         return err;
1767 }
1768
1769 /*
1770  * journal_dev_name: format a character string to describe on what
1771  * device this journal is present.
1772  */
1773
1774 static const char *journal_dev_name(journal_t *journal, char *buffer)
1775 {
1776         struct block_device *bdev;
1777
1778         if (journal->j_inode)
1779                 bdev = journal->j_inode->i_sb->s_bdev;
1780         else
1781                 bdev = journal->j_dev;
1782
1783         return bdevname(bdev, buffer);
1784 }
1785
1786 /*
1787  * Journal abort has very specific semantics, which we describe
1788  * for journal abort.
1789  *
1790  * Two internal function, which provide abort to te jbd layer
1791  * itself are here.
1792  */
1793
1794 /*
1795  * Quick version for internal journal use (doesn't lock the journal).
1796  * Aborts hard --- we mark the abort as occurred, but do _nothing_ else,
1797  * and don't attempt to make any other journal updates.
1798  */
1799 void __jbd2_journal_abort_hard(journal_t *journal)
1800 {
1801         transaction_t *transaction;
1802         char b[BDEVNAME_SIZE];
1803
1804         if (journal->j_flags & JBD2_ABORT)
1805                 return;
1806
1807         printk(KERN_ERR "Aborting journal on device %s.\n",
1808                 journal_dev_name(journal, b));
1809
1810         spin_lock(&journal->j_state_lock);
1811         journal->j_flags |= JBD2_ABORT;
1812         transaction = journal->j_running_transaction;
1813         if (transaction)
1814                 __jbd2_log_start_commit(journal, transaction->t_tid);
1815         spin_unlock(&journal->j_state_lock);
1816 }
1817
1818 /* Soft abort: record the abort error status in the journal superblock,
1819  * but don't do any other IO. */
1820 static void __journal_abort_soft (journal_t *journal, int errno)
1821 {
1822         if (journal->j_flags & JBD2_ABORT)
1823                 return;
1824
1825         if (!journal->j_errno)
1826                 journal->j_errno = errno;
1827
1828         __jbd2_journal_abort_hard(journal);
1829
1830         if (errno)
1831                 jbd2_journal_update_superblock(journal, 1);
1832 }
1833
1834 /**
1835  * void jbd2_journal_abort () - Shutdown the journal immediately.
1836  * @journal: the journal to shutdown.
1837  * @errno:   an error number to record in the journal indicating
1838  *           the reason for the shutdown.
1839  *
1840  * Perform a complete, immediate shutdown of the ENTIRE
1841  * journal (not of a single transaction).  This operation cannot be
1842  * undone without closing and reopening the journal.
1843  *
1844  * The jbd2_journal_abort function is intended to support higher level error
1845  * recovery mechanisms such as the ext2/ext3 remount-readonly error
1846  * mode.
1847  *
1848  * Journal abort has very specific semantics.  Any existing dirty,
1849  * unjournaled buffers in the main filesystem will still be written to
1850  * disk by bdflush, but the journaling mechanism will be suspended
1851  * immediately and no further transaction commits will be honoured.
1852  *
1853  * Any dirty, journaled buffers will be written back to disk without
1854  * hitting the journal.  Atomicity cannot be guaranteed on an aborted
1855  * filesystem, but we _do_ attempt to leave as much data as possible
1856  * behind for fsck to use for cleanup.
1857  *
1858  * Any attempt to get a new transaction handle on a journal which is in
1859  * ABORT state will just result in an -EROFS error return.  A
1860  * jbd2_journal_stop on an existing handle will return -EIO if we have
1861  * entered abort state during the update.
1862  *
1863  * Recursive transactions are not disturbed by journal abort until the
1864  * final jbd2_journal_stop, which will receive the -EIO error.
1865  *
1866  * Finally, the jbd2_journal_abort call allows the caller to supply an errno
1867  * which will be recorded (if possible) in the journal superblock.  This
1868  * allows a client to record failure conditions in the middle of a
1869  * transaction without having to complete the transaction to record the
1870  * failure to disk.  ext3_error, for example, now uses this
1871  * functionality.
1872  *
1873  * Errors which originate from within the journaling layer will NOT
1874  * supply an errno; a null errno implies that absolutely no further
1875  * writes are done to the journal (unless there are any already in
1876  * progress).
1877  *
1878  */
1879
1880 void jbd2_journal_abort(journal_t *journal, int errno)
1881 {
1882         __journal_abort_soft(journal, errno);
1883 }
1884
1885 /**
1886  * int jbd2_journal_errno () - returns the journal's error state.
1887  * @journal: journal to examine.
1888  *
1889  * This is the errno numbet set with jbd2_journal_abort(), the last
1890  * time the journal was mounted - if the journal was stopped
1891  * without calling abort this will be 0.
1892  *
1893  * If the journal has been aborted on this mount time -EROFS will
1894  * be returned.
1895  */
1896 int jbd2_journal_errno(journal_t *journal)
1897 {
1898         int err;
1899
1900         spin_lock(&journal->j_state_lock);
1901         if (journal->j_flags & JBD2_ABORT)
1902                 err = -EROFS;
1903         else
1904                 err = journal->j_errno;
1905         spin_unlock(&journal->j_state_lock);
1906         return err;
1907 }
1908
1909 /**
1910  * int jbd2_journal_clear_err () - clears the journal's error state
1911  * @journal: journal to act on.
1912  *
1913  * An error must be cleared or Acked to take a FS out of readonly
1914  * mode.
1915  */
1916 int jbd2_journal_clear_err(journal_t *journal)
1917 {
1918         int err = 0;
1919
1920         spin_lock(&journal->j_state_lock);
1921         if (journal->j_flags & JBD2_ABORT)
1922                 err = -EROFS;
1923         else
1924                 journal->j_errno = 0;
1925         spin_unlock(&journal->j_state_lock);
1926         return err;
1927 }
1928
1929 /**
1930  * void jbd2_journal_ack_err() - Ack journal err.
1931  * @journal: journal to act on.
1932  *
1933  * An error must be cleared or Acked to take a FS out of readonly
1934  * mode.
1935  */
1936 void jbd2_journal_ack_err(journal_t *journal)
1937 {
1938         spin_lock(&journal->j_state_lock);
1939         if (journal->j_errno)
1940                 journal->j_flags |= JBD2_ACK_ERR;
1941         spin_unlock(&journal->j_state_lock);
1942 }
1943
1944 int jbd2_journal_blocks_per_page(struct inode *inode)
1945 {
1946         return 1 << (PAGE_CACHE_SHIFT - inode->i_sb->s_blocksize_bits);
1947 }
1948
1949 /*
1950  * helper functions to deal with 32 or 64bit block numbers.
1951  */
1952 size_t journal_tag_bytes(journal_t *journal)
1953 {
1954         if (JBD2_HAS_INCOMPAT_FEATURE(journal, JBD2_FEATURE_INCOMPAT_64BIT))
1955                 return JBD2_TAG_SIZE64;
1956         else
1957                 return JBD2_TAG_SIZE32;
1958 }
1959
1960 /*
1961  * Journal_head storage management
1962  */
1963 static struct kmem_cache *jbd2_journal_head_cache;
1964 #ifdef CONFIG_JBD2_DEBUG
1965 static atomic_t nr_journal_heads = ATOMIC_INIT(0);
1966 #endif
1967
1968 static int journal_init_jbd2_journal_head_cache(void)
1969 {
1970         int retval;
1971
1972         J_ASSERT(jbd2_journal_head_cache == NULL);
1973         jbd2_journal_head_cache = kmem_cache_create("jbd2_journal_head",
1974                                 sizeof(struct journal_head),
1975                                 0,              /* offset */
1976                                 SLAB_TEMPORARY, /* flags */
1977                                 NULL);          /* ctor */
1978         retval = 0;
1979         if (!jbd2_journal_head_cache) {
1980                 retval = -ENOMEM;
1981                 printk(KERN_EMERG "JBD: no memory for journal_head cache\n");
1982         }
1983         return retval;
1984 }
1985
1986 static void jbd2_journal_destroy_jbd2_journal_head_cache(void)
1987 {
1988         J_ASSERT(jbd2_journal_head_cache != NULL);
1989         kmem_cache_destroy(jbd2_journal_head_cache);
1990         jbd2_journal_head_cache = NULL;
1991 }
1992
1993 /*
1994  * journal_head splicing and dicing
1995  */
1996 static struct journal_head *journal_alloc_journal_head(void)
1997 {
1998         struct journal_head *ret;
1999         static unsigned long last_warning;
2000
2001 #ifdef CONFIG_JBD2_DEBUG
2002         atomic_inc(&nr_journal_heads);
2003 #endif
2004         ret = kmem_cache_alloc(jbd2_journal_head_cache, GFP_NOFS);
2005         if (!ret) {
2006                 jbd_debug(1, "out of memory for journal_head\n");
2007                 if (time_after(jiffies, last_warning + 5*HZ)) {
2008                         printk(KERN_NOTICE "ENOMEM in %s, retrying.\n",
2009                                __FUNCTION__);
2010                         last_warning = jiffies;
2011                 }
2012                 while (!ret) {
2013                         yield();
2014                         ret = kmem_cache_alloc(jbd2_journal_head_cache, GFP_NOFS);
2015                 }
2016         }
2017         return ret;
2018 }
2019
2020 static void journal_free_journal_head(struct journal_head *jh)
2021 {
2022 #ifdef CONFIG_JBD2_DEBUG
2023         atomic_dec(&nr_journal_heads);
2024         memset(jh, JBD2_POISON_FREE, sizeof(*jh));
2025 #endif
2026         kmem_cache_free(jbd2_journal_head_cache, jh);
2027 }
2028
2029 /*
2030  * A journal_head is attached to a buffer_head whenever JBD has an
2031  * interest in the buffer.
2032  *
2033  * Whenever a buffer has an attached journal_head, its ->b_state:BH_JBD bit
2034  * is set.  This bit is tested in core kernel code where we need to take
2035  * JBD-specific actions.  Testing the zeroness of ->b_private is not reliable
2036  * there.
2037  *
2038  * When a buffer has its BH_JBD bit set, its ->b_count is elevated by one.
2039  *
2040  * When a buffer has its BH_JBD bit set it is immune from being released by
2041  * core kernel code, mainly via ->b_count.
2042  *
2043  * A journal_head may be detached from its buffer_head when the journal_head's
2044  * b_transaction, b_cp_transaction and b_next_transaction pointers are NULL.
2045  * Various places in JBD call jbd2_journal_remove_journal_head() to indicate that the
2046  * journal_head can be dropped if needed.
2047  *
2048  * Various places in the kernel want to attach a journal_head to a buffer_head
2049  * _before_ attaching the journal_head to a transaction.  To protect the
2050  * journal_head in this situation, jbd2_journal_add_journal_head elevates the
2051  * journal_head's b_jcount refcount by one.  The caller must call
2052  * jbd2_journal_put_journal_head() to undo this.
2053  *
2054  * So the typical usage would be:
2055  *
2056  *      (Attach a journal_head if needed.  Increments b_jcount)
2057  *      struct journal_head *jh = jbd2_journal_add_journal_head(bh);
2058  *      ...
2059  *      jh->b_transaction = xxx;
2060  *      jbd2_journal_put_journal_head(jh);
2061  *
2062  * Now, the journal_head's b_jcount is zero, but it is safe from being released
2063  * because it has a non-zero b_transaction.
2064  */
2065
2066 /*
2067  * Give a buffer_head a journal_head.
2068  *
2069  * Doesn't need the journal lock.
2070  * May sleep.
2071  */
2072 struct journal_head *jbd2_journal_add_journal_head(struct buffer_head *bh)
2073 {
2074         struct journal_head *jh;
2075         struct journal_head *new_jh = NULL;
2076
2077 repeat:
2078         if (!buffer_jbd(bh)) {
2079                 new_jh = journal_alloc_journal_head();
2080                 memset(new_jh, 0, sizeof(*new_jh));
2081         }
2082
2083         jbd_lock_bh_journal_head(bh);
2084         if (buffer_jbd(bh)) {
2085                 jh = bh2jh(bh);
2086         } else {
2087                 J_ASSERT_BH(bh,
2088                         (atomic_read(&bh->b_count) > 0) ||
2089                         (bh->b_page && bh->b_page->mapping));
2090
2091                 if (!new_jh) {
2092                         jbd_unlock_bh_journal_head(bh);
2093                         goto repeat;
2094                 }
2095
2096                 jh = new_jh;
2097                 new_jh = NULL;          /* We consumed it */
2098                 set_buffer_jbd(bh);
2099                 bh->b_private = jh;
2100                 jh->b_bh = bh;
2101                 get_bh(bh);
2102                 BUFFER_TRACE(bh, "added journal_head");
2103         }
2104         jh->b_jcount++;
2105         jbd_unlock_bh_journal_head(bh);
2106         if (new_jh)
2107                 journal_free_journal_head(new_jh);
2108         return bh->b_private;
2109 }
2110
2111 /*
2112  * Grab a ref against this buffer_head's journal_head.  If it ended up not
2113  * having a journal_head, return NULL
2114  */
2115 struct journal_head *jbd2_journal_grab_journal_head(struct buffer_head *bh)
2116 {
2117         struct journal_head *jh = NULL;
2118
2119         jbd_lock_bh_journal_head(bh);
2120         if (buffer_jbd(bh)) {
2121                 jh = bh2jh(bh);
2122                 jh->b_jcount++;
2123         }
2124         jbd_unlock_bh_journal_head(bh);
2125         return jh;
2126 }
2127
2128 static void __journal_remove_journal_head(struct buffer_head *bh)
2129 {
2130         struct journal_head *jh = bh2jh(bh);
2131
2132         J_ASSERT_JH(jh, jh->b_jcount >= 0);
2133
2134         get_bh(bh);
2135         if (jh->b_jcount == 0) {
2136                 if (jh->b_transaction == NULL &&
2137                                 jh->b_next_transaction == NULL &&
2138                                 jh->b_cp_transaction == NULL) {
2139                         J_ASSERT_JH(jh, jh->b_jlist == BJ_None);
2140                         J_ASSERT_BH(bh, buffer_jbd(bh));
2141                         J_ASSERT_BH(bh, jh2bh(jh) == bh);
2142                         BUFFER_TRACE(bh, "remove journal_head");
2143                         if (jh->b_frozen_data) {
2144                                 printk(KERN_WARNING "%s: freeing "
2145                                                 "b_frozen_data\n",
2146                                                 __FUNCTION__);
2147                                 jbd2_free(jh->b_frozen_data, bh->b_size);
2148                         }
2149                         if (jh->b_committed_data) {
2150                                 printk(KERN_WARNING "%s: freeing "
2151                                                 "b_committed_data\n",
2152                                                 __FUNCTION__);
2153                                 jbd2_free(jh->b_committed_data, bh->b_size);
2154                         }
2155                         bh->b_private = NULL;
2156                         jh->b_bh = NULL;        /* debug, really */
2157                         clear_buffer_jbd(bh);
2158                         __brelse(bh);
2159                         journal_free_journal_head(jh);
2160                 } else {
2161                         BUFFER_TRACE(bh, "journal_head was locked");
2162                 }
2163         }
2164 }
2165
2166 /*
2167  * jbd2_journal_remove_journal_head(): if the buffer isn't attached to a transaction
2168  * and has a zero b_jcount then remove and release its journal_head.   If we did
2169  * see that the buffer is not used by any transaction we also "logically"
2170  * decrement ->b_count.
2171  *
2172  * We in fact take an additional increment on ->b_count as a convenience,
2173  * because the caller usually wants to do additional things with the bh
2174  * after calling here.
2175  * The caller of jbd2_journal_remove_journal_head() *must* run __brelse(bh) at some
2176  * time.  Once the caller has run __brelse(), the buffer is eligible for
2177  * reaping by try_to_free_buffers().
2178  */
2179 void jbd2_journal_remove_journal_head(struct buffer_head *bh)
2180 {
2181         jbd_lock_bh_journal_head(bh);
2182         __journal_remove_journal_head(bh);
2183         jbd_unlock_bh_journal_head(bh);
2184 }
2185
2186 /*
2187  * Drop a reference on the passed journal_head.  If it fell to zero then try to
2188  * release the journal_head from the buffer_head.
2189  */
2190 void jbd2_journal_put_journal_head(struct journal_head *jh)
2191 {
2192         struct buffer_head *bh = jh2bh(jh);
2193
2194         jbd_lock_bh_journal_head(bh);
2195         J_ASSERT_JH(jh, jh->b_jcount > 0);
2196         --jh->b_jcount;
2197         if (!jh->b_jcount && !jh->b_transaction) {
2198                 __journal_remove_journal_head(bh);
2199                 __brelse(bh);
2200         }
2201         jbd_unlock_bh_journal_head(bh);
2202 }
2203
2204 /*
2205  * debugfs tunables
2206  */
2207 #ifdef CONFIG_JBD2_DEBUG
2208 u8 jbd2_journal_enable_debug __read_mostly;
2209 EXPORT_SYMBOL(jbd2_journal_enable_debug);
2210
2211 #define JBD2_DEBUG_NAME "jbd2-debug"
2212
2213 static struct dentry *jbd2_debugfs_dir;
2214 static struct dentry *jbd2_debug;
2215
2216 static void __init jbd2_create_debugfs_entry(void)
2217 {
2218         jbd2_debugfs_dir = debugfs_create_dir("jbd2", NULL);
2219         if (jbd2_debugfs_dir)
2220                 jbd2_debug = debugfs_create_u8(JBD2_DEBUG_NAME, S_IRUGO,
2221                                                jbd2_debugfs_dir,
2222                                                &jbd2_journal_enable_debug);
2223 }
2224
2225 static void __exit jbd2_remove_debugfs_entry(void)
2226 {
2227         debugfs_remove(jbd2_debug);
2228         debugfs_remove(jbd2_debugfs_dir);
2229 }
2230
2231 #else
2232
2233 static void __init jbd2_create_debugfs_entry(void)
2234 {
2235 }
2236
2237 static void __exit jbd2_remove_debugfs_entry(void)
2238 {
2239 }
2240
2241 #endif
2242
2243 #ifdef CONFIG_PROC_FS
2244
2245 #define JBD2_STATS_PROC_NAME "fs/jbd2"
2246
2247 static void __init jbd2_create_jbd_stats_proc_entry(void)
2248 {
2249         proc_jbd2_stats = proc_mkdir(JBD2_STATS_PROC_NAME, NULL);
2250 }
2251
2252 static void __exit jbd2_remove_jbd_stats_proc_entry(void)
2253 {
2254         if (proc_jbd2_stats)
2255                 remove_proc_entry(JBD2_STATS_PROC_NAME, NULL);
2256 }
2257
2258 #else
2259
2260 #define jbd2_create_jbd_stats_proc_entry() do {} while (0)
2261 #define jbd2_remove_jbd_stats_proc_entry() do {} while (0)
2262
2263 #endif
2264
2265 struct kmem_cache *jbd2_handle_cache;
2266
2267 static int __init journal_init_handle_cache(void)
2268 {
2269         jbd2_handle_cache = kmem_cache_create("jbd2_journal_handle",
2270                                 sizeof(handle_t),
2271                                 0,              /* offset */
2272                                 SLAB_TEMPORARY, /* flags */
2273                                 NULL);          /* ctor */
2274         if (jbd2_handle_cache == NULL) {
2275                 printk(KERN_EMERG "JBD: failed to create handle cache\n");
2276                 return -ENOMEM;
2277         }
2278         return 0;
2279 }
2280
2281 static void jbd2_journal_destroy_handle_cache(void)
2282 {
2283         if (jbd2_handle_cache)
2284                 kmem_cache_destroy(jbd2_handle_cache);
2285 }
2286
2287 /*
2288  * Module startup and shutdown
2289  */
2290
2291 static int __init journal_init_caches(void)
2292 {
2293         int ret;
2294
2295         ret = jbd2_journal_init_revoke_caches();
2296         if (ret == 0)
2297                 ret = journal_init_jbd2_journal_head_cache();
2298         if (ret == 0)
2299                 ret = journal_init_handle_cache();
2300         return ret;
2301 }
2302
2303 static void jbd2_journal_destroy_caches(void)
2304 {
2305         jbd2_journal_destroy_revoke_caches();
2306         jbd2_journal_destroy_jbd2_journal_head_cache();
2307         jbd2_journal_destroy_handle_cache();
2308 }
2309
2310 static int __init journal_init(void)
2311 {
2312         int ret;
2313
2314         BUILD_BUG_ON(sizeof(struct journal_superblock_s) != 1024);
2315
2316         ret = journal_init_caches();
2317         if (ret != 0)
2318                 jbd2_journal_destroy_caches();
2319         jbd2_create_debugfs_entry();
2320         jbd2_create_jbd_stats_proc_entry();
2321         return ret;
2322 }
2323
2324 static void __exit journal_exit(void)
2325 {
2326 #ifdef CONFIG_JBD2_DEBUG
2327         int n = atomic_read(&nr_journal_heads);
2328         if (n)
2329                 printk(KERN_EMERG "JBD: leaked %d journal_heads!\n", n);
2330 #endif
2331         jbd2_remove_debugfs_entry();
2332         jbd2_remove_jbd_stats_proc_entry();
2333         jbd2_journal_destroy_caches();
2334 }
2335
2336 MODULE_LICENSE("GPL");
2337 module_init(journal_init);
2338 module_exit(journal_exit);
2339