Merge branch 'linus' into sched/core
[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_metadata);
54 EXPORT_SYMBOL(jbd2_journal_release_buffer);
55 EXPORT_SYMBOL(jbd2_journal_forget);
56 #if 0
57 EXPORT_SYMBOL(journal_sync_buffer);
58 #endif
59 EXPORT_SYMBOL(jbd2_journal_flush);
60 EXPORT_SYMBOL(jbd2_journal_revoke);
61
62 EXPORT_SYMBOL(jbd2_journal_init_dev);
63 EXPORT_SYMBOL(jbd2_journal_init_inode);
64 EXPORT_SYMBOL(jbd2_journal_update_format);
65 EXPORT_SYMBOL(jbd2_journal_check_used_features);
66 EXPORT_SYMBOL(jbd2_journal_check_available_features);
67 EXPORT_SYMBOL(jbd2_journal_set_features);
68 EXPORT_SYMBOL(jbd2_journal_create);
69 EXPORT_SYMBOL(jbd2_journal_load);
70 EXPORT_SYMBOL(jbd2_journal_destroy);
71 EXPORT_SYMBOL(jbd2_journal_abort);
72 EXPORT_SYMBOL(jbd2_journal_errno);
73 EXPORT_SYMBOL(jbd2_journal_ack_err);
74 EXPORT_SYMBOL(jbd2_journal_clear_err);
75 EXPORT_SYMBOL(jbd2_log_wait_commit);
76 EXPORT_SYMBOL(jbd2_journal_start_commit);
77 EXPORT_SYMBOL(jbd2_journal_force_commit_nested);
78 EXPORT_SYMBOL(jbd2_journal_wipe);
79 EXPORT_SYMBOL(jbd2_journal_blocks_per_page);
80 EXPORT_SYMBOL(jbd2_journal_invalidatepage);
81 EXPORT_SYMBOL(jbd2_journal_try_to_free_buffers);
82 EXPORT_SYMBOL(jbd2_journal_force_commit);
83 EXPORT_SYMBOL(jbd2_journal_file_inode);
84 EXPORT_SYMBOL(jbd2_journal_init_jbd_inode);
85 EXPORT_SYMBOL(jbd2_journal_release_jbd_inode);
86 EXPORT_SYMBOL(jbd2_journal_begin_ordered_truncate);
87
88 static int journal_convert_superblock_v1(journal_t *, journal_superblock_t *);
89 static void __journal_abort_soft (journal_t *journal, int errno);
90
91 /*
92  * Helper function used to manage commit timeouts
93  */
94
95 static void commit_timeout(unsigned long __data)
96 {
97         struct task_struct * p = (struct task_struct *) __data;
98
99         wake_up_process(p);
100 }
101
102 /*
103  * kjournald2: The main thread function used to manage a logging device
104  * journal.
105  *
106  * This kernel thread is responsible for two things:
107  *
108  * 1) COMMIT:  Every so often we need to commit the current state of the
109  *    filesystem to disk.  The journal thread is responsible for writing
110  *    all of the metadata buffers to disk.
111  *
112  * 2) CHECKPOINT: We cannot reuse a used section of the log file until all
113  *    of the data in that part of the log has been rewritten elsewhere on
114  *    the disk.  Flushing these old buffers to reclaim space in the log is
115  *    known as checkpointing, and this thread is responsible for that job.
116  */
117
118 static int kjournald2(void *arg)
119 {
120         journal_t *journal = arg;
121         transaction_t *transaction;
122
123         /*
124          * Set up an interval timer which can be used to trigger a commit wakeup
125          * after the commit interval expires
126          */
127         setup_timer(&journal->j_commit_timer, commit_timeout,
128                         (unsigned long)current);
129
130         /* Record that the journal thread is running */
131         journal->j_task = current;
132         wake_up(&journal->j_wait_done_commit);
133
134         printk(KERN_INFO "kjournald2 starting.  Commit interval %ld seconds\n",
135                         journal->j_commit_interval / HZ);
136
137         /*
138          * And now, wait forever for commit wakeup events.
139          */
140         spin_lock(&journal->j_state_lock);
141
142 loop:
143         if (journal->j_flags & JBD2_UNMOUNT)
144                 goto end_loop;
145
146         jbd_debug(1, "commit_sequence=%d, commit_request=%d\n",
147                 journal->j_commit_sequence, journal->j_commit_request);
148
149         if (journal->j_commit_sequence != journal->j_commit_request) {
150                 jbd_debug(1, "OK, requests differ\n");
151                 spin_unlock(&journal->j_state_lock);
152                 del_timer_sync(&journal->j_commit_timer);
153                 jbd2_journal_commit_transaction(journal);
154                 spin_lock(&journal->j_state_lock);
155                 goto loop;
156         }
157
158         wake_up(&journal->j_wait_done_commit);
159         if (freezing(current)) {
160                 /*
161                  * The simpler the better. Flushing journal isn't a
162                  * good idea, because that depends on threads that may
163                  * be already stopped.
164                  */
165                 jbd_debug(1, "Now suspending kjournald2\n");
166                 spin_unlock(&journal->j_state_lock);
167                 refrigerator();
168                 spin_lock(&journal->j_state_lock);
169         } else {
170                 /*
171                  * We assume on resume that commits are already there,
172                  * so we don't sleep
173                  */
174                 DEFINE_WAIT(wait);
175                 int should_sleep = 1;
176
177                 prepare_to_wait(&journal->j_wait_commit, &wait,
178                                 TASK_INTERRUPTIBLE);
179                 if (journal->j_commit_sequence != journal->j_commit_request)
180                         should_sleep = 0;
181                 transaction = journal->j_running_transaction;
182                 if (transaction && time_after_eq(jiffies,
183                                                 transaction->t_expires))
184                         should_sleep = 0;
185                 if (journal->j_flags & JBD2_UNMOUNT)
186                         should_sleep = 0;
187                 if (should_sleep) {
188                         spin_unlock(&journal->j_state_lock);
189                         schedule();
190                         spin_lock(&journal->j_state_lock);
191                 }
192                 finish_wait(&journal->j_wait_commit, &wait);
193         }
194
195         jbd_debug(1, "kjournald2 wakes\n");
196
197         /*
198          * Were we woken up by a commit wakeup event?
199          */
200         transaction = journal->j_running_transaction;
201         if (transaction && time_after_eq(jiffies, transaction->t_expires)) {
202                 journal->j_commit_request = transaction->t_tid;
203                 jbd_debug(1, "woke because of timeout\n");
204         }
205         goto loop;
206
207 end_loop:
208         spin_unlock(&journal->j_state_lock);
209         del_timer_sync(&journal->j_commit_timer);
210         journal->j_task = NULL;
211         wake_up(&journal->j_wait_done_commit);
212         jbd_debug(1, "Journal thread exiting.\n");
213         return 0;
214 }
215
216 static int jbd2_journal_start_thread(journal_t *journal)
217 {
218         struct task_struct *t;
219
220         t = kthread_run(kjournald2, journal, "kjournald2");
221         if (IS_ERR(t))
222                 return PTR_ERR(t);
223
224         wait_event(journal->j_wait_done_commit, journal->j_task != NULL);
225         return 0;
226 }
227
228 static void journal_kill_thread(journal_t *journal)
229 {
230         spin_lock(&journal->j_state_lock);
231         journal->j_flags |= JBD2_UNMOUNT;
232
233         while (journal->j_task) {
234                 wake_up(&journal->j_wait_commit);
235                 spin_unlock(&journal->j_state_lock);
236                 wait_event(journal->j_wait_done_commit, journal->j_task == NULL);
237                 spin_lock(&journal->j_state_lock);
238         }
239         spin_unlock(&journal->j_state_lock);
240 }
241
242 /*
243  * jbd2_journal_write_metadata_buffer: write a metadata buffer to the journal.
244  *
245  * Writes a metadata buffer to a given disk block.  The actual IO is not
246  * performed but a new buffer_head is constructed which labels the data
247  * to be written with the correct destination disk block.
248  *
249  * Any magic-number escaping which needs to be done will cause a
250  * copy-out here.  If the buffer happens to start with the
251  * JBD2_MAGIC_NUMBER, then we can't write it to the log directly: the
252  * magic number is only written to the log for descripter blocks.  In
253  * this case, we copy the data and replace the first word with 0, and we
254  * return a result code which indicates that this buffer needs to be
255  * marked as an escaped buffer in the corresponding log descriptor
256  * block.  The missing word can then be restored when the block is read
257  * during recovery.
258  *
259  * If the source buffer has already been modified by a new transaction
260  * since we took the last commit snapshot, we use the frozen copy of
261  * that data for IO.  If we end up using the existing buffer_head's data
262  * for the write, then we *have* to lock the buffer to prevent anyone
263  * else from using and possibly modifying it while the IO is in
264  * progress.
265  *
266  * The function returns a pointer to the buffer_heads to be used for IO.
267  *
268  * We assume that the journal has already been locked in this function.
269  *
270  * Return value:
271  *  <0: Error
272  * >=0: Finished OK
273  *
274  * On success:
275  * Bit 0 set == escape performed on the data
276  * Bit 1 set == buffer copy-out performed (kfree the data after IO)
277  */
278
279 int jbd2_journal_write_metadata_buffer(transaction_t *transaction,
280                                   struct journal_head  *jh_in,
281                                   struct journal_head **jh_out,
282                                   unsigned long long blocknr)
283 {
284         int need_copy_out = 0;
285         int done_copy_out = 0;
286         int do_escape = 0;
287         char *mapped_data;
288         struct buffer_head *new_bh;
289         struct journal_head *new_jh;
290         struct page *new_page;
291         unsigned int new_offset;
292         struct buffer_head *bh_in = jh2bh(jh_in);
293
294         /*
295          * The buffer really shouldn't be locked: only the current committing
296          * transaction is allowed to write it, so nobody else is allowed
297          * to do any IO.
298          *
299          * akpm: except if we're journalling data, and write() output is
300          * also part of a shared mapping, and another thread has
301          * decided to launch a writepage() against this buffer.
302          */
303         J_ASSERT_BH(bh_in, buffer_jbddirty(bh_in));
304
305         new_bh = alloc_buffer_head(GFP_NOFS|__GFP_NOFAIL);
306
307         /*
308          * If a new transaction has already done a buffer copy-out, then
309          * we use that version of the data for the commit.
310          */
311         jbd_lock_bh_state(bh_in);
312 repeat:
313         if (jh_in->b_frozen_data) {
314                 done_copy_out = 1;
315                 new_page = virt_to_page(jh_in->b_frozen_data);
316                 new_offset = offset_in_page(jh_in->b_frozen_data);
317         } else {
318                 new_page = jh2bh(jh_in)->b_page;
319                 new_offset = offset_in_page(jh2bh(jh_in)->b_data);
320         }
321
322         mapped_data = kmap_atomic(new_page, KM_USER0);
323         /*
324          * Check for escaping
325          */
326         if (*((__be32 *)(mapped_data + new_offset)) ==
327                                 cpu_to_be32(JBD2_MAGIC_NUMBER)) {
328                 need_copy_out = 1;
329                 do_escape = 1;
330         }
331         kunmap_atomic(mapped_data, KM_USER0);
332
333         /*
334          * Do we need to do a data copy?
335          */
336         if (need_copy_out && !done_copy_out) {
337                 char *tmp;
338
339                 jbd_unlock_bh_state(bh_in);
340                 tmp = jbd2_alloc(bh_in->b_size, GFP_NOFS);
341                 jbd_lock_bh_state(bh_in);
342                 if (jh_in->b_frozen_data) {
343                         jbd2_free(tmp, bh_in->b_size);
344                         goto repeat;
345                 }
346
347                 jh_in->b_frozen_data = tmp;
348                 mapped_data = kmap_atomic(new_page, KM_USER0);
349                 memcpy(tmp, mapped_data + new_offset, jh2bh(jh_in)->b_size);
350                 kunmap_atomic(mapped_data, KM_USER0);
351
352                 new_page = virt_to_page(tmp);
353                 new_offset = offset_in_page(tmp);
354                 done_copy_out = 1;
355         }
356
357         /*
358          * Did we need to do an escaping?  Now we've done all the
359          * copying, we can finally do so.
360          */
361         if (do_escape) {
362                 mapped_data = kmap_atomic(new_page, KM_USER0);
363                 *((unsigned int *)(mapped_data + new_offset)) = 0;
364                 kunmap_atomic(mapped_data, KM_USER0);
365         }
366
367         /* keep subsequent assertions sane */
368         new_bh->b_state = 0;
369         init_buffer(new_bh, NULL, NULL);
370         atomic_set(&new_bh->b_count, 1);
371         jbd_unlock_bh_state(bh_in);
372
373         new_jh = jbd2_journal_add_journal_head(new_bh); /* This sleeps */
374
375         set_bh_page(new_bh, new_page, new_offset);
376         new_jh->b_transaction = NULL;
377         new_bh->b_size = jh2bh(jh_in)->b_size;
378         new_bh->b_bdev = transaction->t_journal->j_dev;
379         new_bh->b_blocknr = blocknr;
380         set_buffer_mapped(new_bh);
381         set_buffer_dirty(new_bh);
382
383         *jh_out = new_jh;
384
385         /*
386          * The to-be-written buffer needs to get moved to the io queue,
387          * and the original buffer whose contents we are shadowing or
388          * copying is moved to the transaction's shadow queue.
389          */
390         JBUFFER_TRACE(jh_in, "file as BJ_Shadow");
391         jbd2_journal_file_buffer(jh_in, transaction, BJ_Shadow);
392         JBUFFER_TRACE(new_jh, "file as BJ_IO");
393         jbd2_journal_file_buffer(new_jh, transaction, BJ_IO);
394
395         return do_escape | (done_copy_out << 1);
396 }
397
398 /*
399  * Allocation code for the journal file.  Manage the space left in the
400  * journal, so that we can begin checkpointing when appropriate.
401  */
402
403 /*
404  * __jbd2_log_space_left: Return the number of free blocks left in the journal.
405  *
406  * Called with the journal already locked.
407  *
408  * Called under j_state_lock
409  */
410
411 int __jbd2_log_space_left(journal_t *journal)
412 {
413         int left = journal->j_free;
414
415         assert_spin_locked(&journal->j_state_lock);
416
417         /*
418          * Be pessimistic here about the number of those free blocks which
419          * might be required for log descriptor control blocks.
420          */
421
422 #define MIN_LOG_RESERVED_BLOCKS 32 /* Allow for rounding errors */
423
424         left -= MIN_LOG_RESERVED_BLOCKS;
425
426         if (left <= 0)
427                 return 0;
428         left -= (left >> 3);
429         return left;
430 }
431
432 /*
433  * Called under j_state_lock.  Returns true if a transaction was started.
434  */
435 int __jbd2_log_start_commit(journal_t *journal, tid_t target)
436 {
437         /*
438          * Are we already doing a recent enough commit?
439          */
440         if (!tid_geq(journal->j_commit_request, target)) {
441                 /*
442                  * We want a new commit: OK, mark the request and wakup the
443                  * commit thread.  We do _not_ do the commit ourselves.
444                  */
445
446                 journal->j_commit_request = target;
447                 jbd_debug(1, "JBD: requesting commit %d/%d\n",
448                           journal->j_commit_request,
449                           journal->j_commit_sequence);
450                 wake_up(&journal->j_wait_commit);
451                 return 1;
452         }
453         return 0;
454 }
455
456 int jbd2_log_start_commit(journal_t *journal, tid_t tid)
457 {
458         int ret;
459
460         spin_lock(&journal->j_state_lock);
461         ret = __jbd2_log_start_commit(journal, tid);
462         spin_unlock(&journal->j_state_lock);
463         return ret;
464 }
465
466 /*
467  * Force and wait upon a commit if the calling process is not within
468  * transaction.  This is used for forcing out undo-protected data which contains
469  * bitmaps, when the fs is running out of space.
470  *
471  * We can only force the running transaction if we don't have an active handle;
472  * otherwise, we will deadlock.
473  *
474  * Returns true if a transaction was started.
475  */
476 int jbd2_journal_force_commit_nested(journal_t *journal)
477 {
478         transaction_t *transaction = NULL;
479         tid_t tid;
480
481         spin_lock(&journal->j_state_lock);
482         if (journal->j_running_transaction && !current->journal_info) {
483                 transaction = journal->j_running_transaction;
484                 __jbd2_log_start_commit(journal, transaction->t_tid);
485         } else if (journal->j_committing_transaction)
486                 transaction = journal->j_committing_transaction;
487
488         if (!transaction) {
489                 spin_unlock(&journal->j_state_lock);
490                 return 0;       /* Nothing to retry */
491         }
492
493         tid = transaction->t_tid;
494         spin_unlock(&journal->j_state_lock);
495         jbd2_log_wait_commit(journal, tid);
496         return 1;
497 }
498
499 /*
500  * Start a commit of the current running transaction (if any).  Returns true
501  * if a transaction was started, and fills its tid in at *ptid
502  */
503 int jbd2_journal_start_commit(journal_t *journal, tid_t *ptid)
504 {
505         int ret = 0;
506
507         spin_lock(&journal->j_state_lock);
508         if (journal->j_running_transaction) {
509                 tid_t tid = journal->j_running_transaction->t_tid;
510
511                 ret = __jbd2_log_start_commit(journal, tid);
512                 if (ret && ptid)
513                         *ptid = tid;
514         } else if (journal->j_committing_transaction && ptid) {
515                 /*
516                  * If ext3_write_super() recently started a commit, then we
517                  * have to wait for completion of that transaction
518                  */
519                 *ptid = journal->j_committing_transaction->t_tid;
520                 ret = 1;
521         }
522         spin_unlock(&journal->j_state_lock);
523         return ret;
524 }
525
526 /*
527  * Wait for a specified commit to complete.
528  * The caller may not hold the journal lock.
529  */
530 int jbd2_log_wait_commit(journal_t *journal, tid_t tid)
531 {
532         int err = 0;
533
534 #ifdef CONFIG_JBD2_DEBUG
535         spin_lock(&journal->j_state_lock);
536         if (!tid_geq(journal->j_commit_request, tid)) {
537                 printk(KERN_EMERG
538                        "%s: error: j_commit_request=%d, tid=%d\n",
539                        __func__, journal->j_commit_request, tid);
540         }
541         spin_unlock(&journal->j_state_lock);
542 #endif
543         spin_lock(&journal->j_state_lock);
544         while (tid_gt(tid, journal->j_commit_sequence)) {
545                 jbd_debug(1, "JBD: want %d, j_commit_sequence=%d\n",
546                                   tid, journal->j_commit_sequence);
547                 wake_up(&journal->j_wait_commit);
548                 spin_unlock(&journal->j_state_lock);
549                 wait_event(journal->j_wait_done_commit,
550                                 !tid_gt(tid, journal->j_commit_sequence));
551                 spin_lock(&journal->j_state_lock);
552         }
553         spin_unlock(&journal->j_state_lock);
554
555         if (unlikely(is_journal_aborted(journal))) {
556                 printk(KERN_EMERG "journal commit I/O error\n");
557                 err = -EIO;
558         }
559         return err;
560 }
561
562 /*
563  * Log buffer allocation routines:
564  */
565
566 int jbd2_journal_next_log_block(journal_t *journal, unsigned long long *retp)
567 {
568         unsigned long blocknr;
569
570         spin_lock(&journal->j_state_lock);
571         J_ASSERT(journal->j_free > 1);
572
573         blocknr = journal->j_head;
574         journal->j_head++;
575         journal->j_free--;
576         if (journal->j_head == journal->j_last)
577                 journal->j_head = journal->j_first;
578         spin_unlock(&journal->j_state_lock);
579         return jbd2_journal_bmap(journal, blocknr, retp);
580 }
581
582 /*
583  * Conversion of logical to physical block numbers for the journal
584  *
585  * On external journals the journal blocks are identity-mapped, so
586  * this is a no-op.  If needed, we can use j_blk_offset - everything is
587  * ready.
588  */
589 int jbd2_journal_bmap(journal_t *journal, unsigned long blocknr,
590                  unsigned long long *retp)
591 {
592         int err = 0;
593         unsigned long long ret;
594
595         if (journal->j_inode) {
596                 ret = bmap(journal->j_inode, blocknr);
597                 if (ret)
598                         *retp = ret;
599                 else {
600                         printk(KERN_ALERT "%s: journal block not found "
601                                         "at offset %lu on %s\n",
602                                __func__, blocknr, journal->j_devname);
603                         err = -EIO;
604                         __journal_abort_soft(journal, err);
605                 }
606         } else {
607                 *retp = blocknr; /* +journal->j_blk_offset */
608         }
609         return err;
610 }
611
612 /*
613  * We play buffer_head aliasing tricks to write data/metadata blocks to
614  * the journal without copying their contents, but for journal
615  * descriptor blocks we do need to generate bona fide buffers.
616  *
617  * After the caller of jbd2_journal_get_descriptor_buffer() has finished modifying
618  * the buffer's contents they really should run flush_dcache_page(bh->b_page).
619  * But we don't bother doing that, so there will be coherency problems with
620  * mmaps of blockdevs which hold live JBD-controlled filesystems.
621  */
622 struct journal_head *jbd2_journal_get_descriptor_buffer(journal_t *journal)
623 {
624         struct buffer_head *bh;
625         unsigned long long blocknr;
626         int err;
627
628         err = jbd2_journal_next_log_block(journal, &blocknr);
629
630         if (err)
631                 return NULL;
632
633         bh = __getblk(journal->j_dev, blocknr, journal->j_blocksize);
634         lock_buffer(bh);
635         memset(bh->b_data, 0, journal->j_blocksize);
636         set_buffer_uptodate(bh);
637         unlock_buffer(bh);
638         BUFFER_TRACE(bh, "return this buffer");
639         return jbd2_journal_add_journal_head(bh);
640 }
641
642 struct jbd2_stats_proc_session {
643         journal_t *journal;
644         struct transaction_stats_s *stats;
645         int start;
646         int max;
647 };
648
649 static void *jbd2_history_skip_empty(struct jbd2_stats_proc_session *s,
650                                         struct transaction_stats_s *ts,
651                                         int first)
652 {
653         if (ts == s->stats + s->max)
654                 ts = s->stats;
655         if (!first && ts == s->stats + s->start)
656                 return NULL;
657         while (ts->ts_type == 0) {
658                 ts++;
659                 if (ts == s->stats + s->max)
660                         ts = s->stats;
661                 if (ts == s->stats + s->start)
662                         return NULL;
663         }
664         return ts;
665
666 }
667
668 static void *jbd2_seq_history_start(struct seq_file *seq, loff_t *pos)
669 {
670         struct jbd2_stats_proc_session *s = seq->private;
671         struct transaction_stats_s *ts;
672         int l = *pos;
673
674         if (l == 0)
675                 return SEQ_START_TOKEN;
676         ts = jbd2_history_skip_empty(s, s->stats + s->start, 1);
677         if (!ts)
678                 return NULL;
679         l--;
680         while (l) {
681                 ts = jbd2_history_skip_empty(s, ++ts, 0);
682                 if (!ts)
683                         break;
684                 l--;
685         }
686         return ts;
687 }
688
689 static void *jbd2_seq_history_next(struct seq_file *seq, void *v, loff_t *pos)
690 {
691         struct jbd2_stats_proc_session *s = seq->private;
692         struct transaction_stats_s *ts = v;
693
694         ++*pos;
695         if (v == SEQ_START_TOKEN)
696                 return jbd2_history_skip_empty(s, s->stats + s->start, 1);
697         else
698                 return jbd2_history_skip_empty(s, ++ts, 0);
699 }
700
701 static int jbd2_seq_history_show(struct seq_file *seq, void *v)
702 {
703         struct transaction_stats_s *ts = v;
704         if (v == SEQ_START_TOKEN) {
705                 seq_printf(seq, "%-4s %-5s %-5s %-5s %-5s %-5s %-5s %-6s %-5s "
706                                 "%-5s %-5s %-5s %-5s %-5s\n", "R/C", "tid",
707                                 "wait", "run", "lock", "flush", "log", "hndls",
708                                 "block", "inlog", "ctime", "write", "drop",
709                                 "close");
710                 return 0;
711         }
712         if (ts->ts_type == JBD2_STATS_RUN)
713                 seq_printf(seq, "%-4s %-5lu %-5u %-5u %-5u %-5u %-5u "
714                                 "%-6lu %-5lu %-5lu\n", "R", ts->ts_tid,
715                                 jiffies_to_msecs(ts->u.run.rs_wait),
716                                 jiffies_to_msecs(ts->u.run.rs_running),
717                                 jiffies_to_msecs(ts->u.run.rs_locked),
718                                 jiffies_to_msecs(ts->u.run.rs_flushing),
719                                 jiffies_to_msecs(ts->u.run.rs_logging),
720                                 ts->u.run.rs_handle_count,
721                                 ts->u.run.rs_blocks,
722                                 ts->u.run.rs_blocks_logged);
723         else if (ts->ts_type == JBD2_STATS_CHECKPOINT)
724                 seq_printf(seq, "%-4s %-5lu %48s %-5u %-5lu %-5lu %-5lu\n",
725                                 "C", ts->ts_tid, " ",
726                                 jiffies_to_msecs(ts->u.chp.cs_chp_time),
727                                 ts->u.chp.cs_written, ts->u.chp.cs_dropped,
728                                 ts->u.chp.cs_forced_to_close);
729         else
730                 J_ASSERT(0);
731         return 0;
732 }
733
734 static void jbd2_seq_history_stop(struct seq_file *seq, void *v)
735 {
736 }
737
738 static struct seq_operations jbd2_seq_history_ops = {
739         .start  = jbd2_seq_history_start,
740         .next   = jbd2_seq_history_next,
741         .stop   = jbd2_seq_history_stop,
742         .show   = jbd2_seq_history_show,
743 };
744
745 static int jbd2_seq_history_open(struct inode *inode, struct file *file)
746 {
747         journal_t *journal = PDE(inode)->data;
748         struct jbd2_stats_proc_session *s;
749         int rc, size;
750
751         s = kmalloc(sizeof(*s), GFP_KERNEL);
752         if (s == NULL)
753                 return -ENOMEM;
754         size = sizeof(struct transaction_stats_s) * journal->j_history_max;
755         s->stats = kmalloc(size, GFP_KERNEL);
756         if (s->stats == NULL) {
757                 kfree(s);
758                 return -ENOMEM;
759         }
760         spin_lock(&journal->j_history_lock);
761         memcpy(s->stats, journal->j_history, size);
762         s->max = journal->j_history_max;
763         s->start = journal->j_history_cur % s->max;
764         spin_unlock(&journal->j_history_lock);
765
766         rc = seq_open(file, &jbd2_seq_history_ops);
767         if (rc == 0) {
768                 struct seq_file *m = file->private_data;
769                 m->private = s;
770         } else {
771                 kfree(s->stats);
772                 kfree(s);
773         }
774         return rc;
775
776 }
777
778 static int jbd2_seq_history_release(struct inode *inode, struct file *file)
779 {
780         struct seq_file *seq = file->private_data;
781         struct jbd2_stats_proc_session *s = seq->private;
782
783         kfree(s->stats);
784         kfree(s);
785         return seq_release(inode, file);
786 }
787
788 static struct file_operations jbd2_seq_history_fops = {
789         .owner          = THIS_MODULE,
790         .open           = jbd2_seq_history_open,
791         .read           = seq_read,
792         .llseek         = seq_lseek,
793         .release        = jbd2_seq_history_release,
794 };
795
796 static void *jbd2_seq_info_start(struct seq_file *seq, loff_t *pos)
797 {
798         return *pos ? NULL : SEQ_START_TOKEN;
799 }
800
801 static void *jbd2_seq_info_next(struct seq_file *seq, void *v, loff_t *pos)
802 {
803         return NULL;
804 }
805
806 static int jbd2_seq_info_show(struct seq_file *seq, void *v)
807 {
808         struct jbd2_stats_proc_session *s = seq->private;
809
810         if (v != SEQ_START_TOKEN)
811                 return 0;
812         seq_printf(seq, "%lu transaction, each upto %u blocks\n",
813                         s->stats->ts_tid,
814                         s->journal->j_max_transaction_buffers);
815         if (s->stats->ts_tid == 0)
816                 return 0;
817         seq_printf(seq, "average: \n  %ums waiting for transaction\n",
818             jiffies_to_msecs(s->stats->u.run.rs_wait / s->stats->ts_tid));
819         seq_printf(seq, "  %ums running transaction\n",
820             jiffies_to_msecs(s->stats->u.run.rs_running / s->stats->ts_tid));
821         seq_printf(seq, "  %ums transaction was being locked\n",
822             jiffies_to_msecs(s->stats->u.run.rs_locked / s->stats->ts_tid));
823         seq_printf(seq, "  %ums flushing data (in ordered mode)\n",
824             jiffies_to_msecs(s->stats->u.run.rs_flushing / s->stats->ts_tid));
825         seq_printf(seq, "  %ums logging transaction\n",
826             jiffies_to_msecs(s->stats->u.run.rs_logging / s->stats->ts_tid));
827         seq_printf(seq, "  %lu handles per transaction\n",
828             s->stats->u.run.rs_handle_count / s->stats->ts_tid);
829         seq_printf(seq, "  %lu blocks per transaction\n",
830             s->stats->u.run.rs_blocks / s->stats->ts_tid);
831         seq_printf(seq, "  %lu logged blocks per transaction\n",
832             s->stats->u.run.rs_blocks_logged / s->stats->ts_tid);
833         return 0;
834 }
835
836 static void jbd2_seq_info_stop(struct seq_file *seq, void *v)
837 {
838 }
839
840 static struct seq_operations jbd2_seq_info_ops = {
841         .start  = jbd2_seq_info_start,
842         .next   = jbd2_seq_info_next,
843         .stop   = jbd2_seq_info_stop,
844         .show   = jbd2_seq_info_show,
845 };
846
847 static int jbd2_seq_info_open(struct inode *inode, struct file *file)
848 {
849         journal_t *journal = PDE(inode)->data;
850         struct jbd2_stats_proc_session *s;
851         int rc, size;
852
853         s = kmalloc(sizeof(*s), GFP_KERNEL);
854         if (s == NULL)
855                 return -ENOMEM;
856         size = sizeof(struct transaction_stats_s);
857         s->stats = kmalloc(size, GFP_KERNEL);
858         if (s->stats == NULL) {
859                 kfree(s);
860                 return -ENOMEM;
861         }
862         spin_lock(&journal->j_history_lock);
863         memcpy(s->stats, &journal->j_stats, size);
864         s->journal = journal;
865         spin_unlock(&journal->j_history_lock);
866
867         rc = seq_open(file, &jbd2_seq_info_ops);
868         if (rc == 0) {
869                 struct seq_file *m = file->private_data;
870                 m->private = s;
871         } else {
872                 kfree(s->stats);
873                 kfree(s);
874         }
875         return rc;
876
877 }
878
879 static int jbd2_seq_info_release(struct inode *inode, struct file *file)
880 {
881         struct seq_file *seq = file->private_data;
882         struct jbd2_stats_proc_session *s = seq->private;
883         kfree(s->stats);
884         kfree(s);
885         return seq_release(inode, file);
886 }
887
888 static struct file_operations jbd2_seq_info_fops = {
889         .owner          = THIS_MODULE,
890         .open           = jbd2_seq_info_open,
891         .read           = seq_read,
892         .llseek         = seq_lseek,
893         .release        = jbd2_seq_info_release,
894 };
895
896 static struct proc_dir_entry *proc_jbd2_stats;
897
898 static void jbd2_stats_proc_init(journal_t *journal)
899 {
900         journal->j_proc_entry = proc_mkdir(journal->j_devname, proc_jbd2_stats);
901         if (journal->j_proc_entry) {
902                 proc_create_data("history", S_IRUGO, journal->j_proc_entry,
903                                  &jbd2_seq_history_fops, journal);
904                 proc_create_data("info", S_IRUGO, journal->j_proc_entry,
905                                  &jbd2_seq_info_fops, journal);
906         }
907 }
908
909 static void jbd2_stats_proc_exit(journal_t *journal)
910 {
911         remove_proc_entry("info", journal->j_proc_entry);
912         remove_proc_entry("history", journal->j_proc_entry);
913         remove_proc_entry(journal->j_devname, proc_jbd2_stats);
914 }
915
916 static void journal_init_stats(journal_t *journal)
917 {
918         int size;
919
920         if (!proc_jbd2_stats)
921                 return;
922
923         journal->j_history_max = 100;
924         size = sizeof(struct transaction_stats_s) * journal->j_history_max;
925         journal->j_history = kzalloc(size, GFP_KERNEL);
926         if (!journal->j_history) {
927                 journal->j_history_max = 0;
928                 return;
929         }
930         spin_lock_init(&journal->j_history_lock);
931 }
932
933 /*
934  * Management for journal control blocks: functions to create and
935  * destroy journal_t structures, and to initialise and read existing
936  * journal blocks from disk.  */
937
938 /* First: create and setup a journal_t object in memory.  We initialise
939  * very few fields yet: that has to wait until we have created the
940  * journal structures from from scratch, or loaded them from disk. */
941
942 static journal_t * journal_init_common (void)
943 {
944         journal_t *journal;
945         int err;
946
947         journal = kzalloc(sizeof(*journal), GFP_KERNEL|__GFP_NOFAIL);
948         if (!journal)
949                 goto fail;
950
951         init_waitqueue_head(&journal->j_wait_transaction_locked);
952         init_waitqueue_head(&journal->j_wait_logspace);
953         init_waitqueue_head(&journal->j_wait_done_commit);
954         init_waitqueue_head(&journal->j_wait_checkpoint);
955         init_waitqueue_head(&journal->j_wait_commit);
956         init_waitqueue_head(&journal->j_wait_updates);
957         mutex_init(&journal->j_barrier);
958         mutex_init(&journal->j_checkpoint_mutex);
959         spin_lock_init(&journal->j_revoke_lock);
960         spin_lock_init(&journal->j_list_lock);
961         spin_lock_init(&journal->j_state_lock);
962
963         journal->j_commit_interval = (HZ * JBD2_DEFAULT_MAX_COMMIT_AGE);
964
965         /* The journal is marked for error until we succeed with recovery! */
966         journal->j_flags = JBD2_ABORT;
967
968         /* Set up a default-sized revoke table for the new mount. */
969         err = jbd2_journal_init_revoke(journal, JOURNAL_REVOKE_DEFAULT_HASH);
970         if (err) {
971                 kfree(journal);
972                 goto fail;
973         }
974
975         journal_init_stats(journal);
976
977         return journal;
978 fail:
979         return NULL;
980 }
981
982 /* jbd2_journal_init_dev and jbd2_journal_init_inode:
983  *
984  * Create a journal structure assigned some fixed set of disk blocks to
985  * the journal.  We don't actually touch those disk blocks yet, but we
986  * need to set up all of the mapping information to tell the journaling
987  * system where the journal blocks are.
988  *
989  */
990
991 /**
992  *  journal_t * jbd2_journal_init_dev() - creates and initialises a journal structure
993  *  @bdev: Block device on which to create the journal
994  *  @fs_dev: Device which hold journalled filesystem for this journal.
995  *  @start: Block nr Start of journal.
996  *  @len:  Length of the journal in blocks.
997  *  @blocksize: blocksize of journalling device
998  *
999  *  Returns: a newly created journal_t *
1000  *
1001  *  jbd2_journal_init_dev creates a journal which maps a fixed contiguous
1002  *  range of blocks on an arbitrary block device.
1003  *
1004  */
1005 journal_t * jbd2_journal_init_dev(struct block_device *bdev,
1006                         struct block_device *fs_dev,
1007                         unsigned long long start, int len, int blocksize)
1008 {
1009         journal_t *journal = journal_init_common();
1010         struct buffer_head *bh;
1011         char *p;
1012         int n;
1013
1014         if (!journal)
1015                 return NULL;
1016
1017         /* journal descriptor can store up to n blocks -bzzz */
1018         journal->j_blocksize = blocksize;
1019         n = journal->j_blocksize / sizeof(journal_block_tag_t);
1020         journal->j_wbufsize = n;
1021         journal->j_wbuf = kmalloc(n * sizeof(struct buffer_head*), GFP_KERNEL);
1022         if (!journal->j_wbuf) {
1023                 printk(KERN_ERR "%s: Cant allocate bhs for commit thread\n",
1024                         __func__);
1025                 kfree(journal);
1026                 journal = NULL;
1027                 goto out;
1028         }
1029         journal->j_dev = bdev;
1030         journal->j_fs_dev = fs_dev;
1031         journal->j_blk_offset = start;
1032         journal->j_maxlen = len;
1033         bdevname(journal->j_dev, journal->j_devname);
1034         p = journal->j_devname;
1035         while ((p = strchr(p, '/')))
1036                 *p = '!';
1037         jbd2_stats_proc_init(journal);
1038
1039         bh = __getblk(journal->j_dev, start, journal->j_blocksize);
1040         J_ASSERT(bh != NULL);
1041         journal->j_sb_buffer = bh;
1042         journal->j_superblock = (journal_superblock_t *)bh->b_data;
1043 out:
1044         return journal;
1045 }
1046
1047 /**
1048  *  journal_t * jbd2_journal_init_inode () - creates a journal which maps to a inode.
1049  *  @inode: An inode to create the journal in
1050  *
1051  * jbd2_journal_init_inode creates a journal which maps an on-disk inode as
1052  * the journal.  The inode must exist already, must support bmap() and
1053  * must have all data blocks preallocated.
1054  */
1055 journal_t * jbd2_journal_init_inode (struct inode *inode)
1056 {
1057         struct buffer_head *bh;
1058         journal_t *journal = journal_init_common();
1059         char *p;
1060         int err;
1061         int n;
1062         unsigned long long blocknr;
1063
1064         if (!journal)
1065                 return NULL;
1066
1067         journal->j_dev = journal->j_fs_dev = inode->i_sb->s_bdev;
1068         journal->j_inode = inode;
1069         bdevname(journal->j_dev, journal->j_devname);
1070         p = journal->j_devname;
1071         while ((p = strchr(p, '/')))
1072                 *p = '!';
1073         p = journal->j_devname + strlen(journal->j_devname);
1074         sprintf(p, ":%lu", journal->j_inode->i_ino);
1075         jbd_debug(1,
1076                   "journal %p: inode %s/%ld, size %Ld, bits %d, blksize %ld\n",
1077                   journal, inode->i_sb->s_id, inode->i_ino,
1078                   (long long) inode->i_size,
1079                   inode->i_sb->s_blocksize_bits, inode->i_sb->s_blocksize);
1080
1081         journal->j_maxlen = inode->i_size >> inode->i_sb->s_blocksize_bits;
1082         journal->j_blocksize = inode->i_sb->s_blocksize;
1083         jbd2_stats_proc_init(journal);
1084
1085         /* journal descriptor can store up to n blocks -bzzz */
1086         n = journal->j_blocksize / sizeof(journal_block_tag_t);
1087         journal->j_wbufsize = n;
1088         journal->j_wbuf = kmalloc(n * sizeof(struct buffer_head*), GFP_KERNEL);
1089         if (!journal->j_wbuf) {
1090                 printk(KERN_ERR "%s: Cant allocate bhs for commit thread\n",
1091                         __func__);
1092                 jbd2_stats_proc_exit(journal);
1093                 kfree(journal);
1094                 return NULL;
1095         }
1096
1097         err = jbd2_journal_bmap(journal, 0, &blocknr);
1098         /* If that failed, give up */
1099         if (err) {
1100                 printk(KERN_ERR "%s: Cannnot locate journal superblock\n",
1101                        __func__);
1102                 jbd2_stats_proc_exit(journal);
1103                 kfree(journal);
1104                 return NULL;
1105         }
1106
1107         bh = __getblk(journal->j_dev, blocknr, journal->j_blocksize);
1108         J_ASSERT(bh != NULL);
1109         journal->j_sb_buffer = bh;
1110         journal->j_superblock = (journal_superblock_t *)bh->b_data;
1111
1112         return journal;
1113 }
1114
1115 /*
1116  * If the journal init or create aborts, we need to mark the journal
1117  * superblock as being NULL to prevent the journal destroy from writing
1118  * back a bogus superblock.
1119  */
1120 static void journal_fail_superblock (journal_t *journal)
1121 {
1122         struct buffer_head *bh = journal->j_sb_buffer;
1123         brelse(bh);
1124         journal->j_sb_buffer = NULL;
1125 }
1126
1127 /*
1128  * Given a journal_t structure, initialise the various fields for
1129  * startup of a new journaling session.  We use this both when creating
1130  * a journal, and after recovering an old journal to reset it for
1131  * subsequent use.
1132  */
1133
1134 static int journal_reset(journal_t *journal)
1135 {
1136         journal_superblock_t *sb = journal->j_superblock;
1137         unsigned long long first, last;
1138
1139         first = be32_to_cpu(sb->s_first);
1140         last = be32_to_cpu(sb->s_maxlen);
1141
1142         journal->j_first = first;
1143         journal->j_last = last;
1144
1145         journal->j_head = first;
1146         journal->j_tail = first;
1147         journal->j_free = last - first;
1148
1149         journal->j_tail_sequence = journal->j_transaction_sequence;
1150         journal->j_commit_sequence = journal->j_transaction_sequence - 1;
1151         journal->j_commit_request = journal->j_commit_sequence;
1152
1153         journal->j_max_transaction_buffers = journal->j_maxlen / 4;
1154
1155         /* Add the dynamic fields and write it to disk. */
1156         jbd2_journal_update_superblock(journal, 1);
1157         return jbd2_journal_start_thread(journal);
1158 }
1159
1160 /**
1161  * int jbd2_journal_create() - Initialise the new journal file
1162  * @journal: Journal to create. This structure must have been initialised
1163  *
1164  * Given a journal_t structure which tells us which disk blocks we can
1165  * use, create a new journal superblock and initialise all of the
1166  * journal fields from scratch.
1167  **/
1168 int jbd2_journal_create(journal_t *journal)
1169 {
1170         unsigned long long blocknr;
1171         struct buffer_head *bh;
1172         journal_superblock_t *sb;
1173         int i, err;
1174
1175         if (journal->j_maxlen < JBD2_MIN_JOURNAL_BLOCKS) {
1176                 printk (KERN_ERR "Journal length (%d blocks) too short.\n",
1177                         journal->j_maxlen);
1178                 journal_fail_superblock(journal);
1179                 return -EINVAL;
1180         }
1181
1182         if (journal->j_inode == NULL) {
1183                 /*
1184                  * We don't know what block to start at!
1185                  */
1186                 printk(KERN_EMERG
1187                        "%s: creation of journal on external device!\n",
1188                        __func__);
1189                 BUG();
1190         }
1191
1192         /* Zero out the entire journal on disk.  We cannot afford to
1193            have any blocks on disk beginning with JBD2_MAGIC_NUMBER. */
1194         jbd_debug(1, "JBD: Zeroing out journal blocks...\n");
1195         for (i = 0; i < journal->j_maxlen; i++) {
1196                 err = jbd2_journal_bmap(journal, i, &blocknr);
1197                 if (err)
1198                         return err;
1199                 bh = __getblk(journal->j_dev, blocknr, journal->j_blocksize);
1200                 lock_buffer(bh);
1201                 memset (bh->b_data, 0, journal->j_blocksize);
1202                 BUFFER_TRACE(bh, "marking dirty");
1203                 mark_buffer_dirty(bh);
1204                 BUFFER_TRACE(bh, "marking uptodate");
1205                 set_buffer_uptodate(bh);
1206                 unlock_buffer(bh);
1207                 __brelse(bh);
1208         }
1209
1210         sync_blockdev(journal->j_dev);
1211         jbd_debug(1, "JBD: journal cleared.\n");
1212
1213         /* OK, fill in the initial static fields in the new superblock */
1214         sb = journal->j_superblock;
1215
1216         sb->s_header.h_magic     = cpu_to_be32(JBD2_MAGIC_NUMBER);
1217         sb->s_header.h_blocktype = cpu_to_be32(JBD2_SUPERBLOCK_V2);
1218
1219         sb->s_blocksize = cpu_to_be32(journal->j_blocksize);
1220         sb->s_maxlen    = cpu_to_be32(journal->j_maxlen);
1221         sb->s_first     = cpu_to_be32(1);
1222
1223         journal->j_transaction_sequence = 1;
1224
1225         journal->j_flags &= ~JBD2_ABORT;
1226         journal->j_format_version = 2;
1227
1228         return journal_reset(journal);
1229 }
1230
1231 /**
1232  * void jbd2_journal_update_superblock() - Update journal sb on disk.
1233  * @journal: The journal to update.
1234  * @wait: Set to '0' if you don't want to wait for IO completion.
1235  *
1236  * Update a journal's dynamic superblock fields and write it to disk,
1237  * optionally waiting for the IO to complete.
1238  */
1239 void jbd2_journal_update_superblock(journal_t *journal, int wait)
1240 {
1241         journal_superblock_t *sb = journal->j_superblock;
1242         struct buffer_head *bh = journal->j_sb_buffer;
1243
1244         /*
1245          * As a special case, if the on-disk copy is already marked as needing
1246          * no recovery (s_start == 0) and there are no outstanding transactions
1247          * in the filesystem, then we can safely defer the superblock update
1248          * until the next commit by setting JBD2_FLUSHED.  This avoids
1249          * attempting a write to a potential-readonly device.
1250          */
1251         if (sb->s_start == 0 && journal->j_tail_sequence ==
1252                                 journal->j_transaction_sequence) {
1253                 jbd_debug(1,"JBD: Skipping superblock update on recovered sb "
1254                         "(start %ld, seq %d, errno %d)\n",
1255                         journal->j_tail, journal->j_tail_sequence,
1256                         journal->j_errno);
1257                 goto out;
1258         }
1259
1260         if (buffer_write_io_error(bh)) {
1261                 /*
1262                  * Oh, dear.  A previous attempt to write the journal
1263                  * superblock failed.  This could happen because the
1264                  * USB device was yanked out.  Or it could happen to
1265                  * be a transient write error and maybe the block will
1266                  * be remapped.  Nothing we can do but to retry the
1267                  * write and hope for the best.
1268                  */
1269                 printk(KERN_ERR "JBD2: previous I/O error detected "
1270                        "for journal superblock update for %s.\n",
1271                        journal->j_devname);
1272                 clear_buffer_write_io_error(bh);
1273                 set_buffer_uptodate(bh);
1274         }
1275
1276         spin_lock(&journal->j_state_lock);
1277         jbd_debug(1,"JBD: updating superblock (start %ld, seq %d, errno %d)\n",
1278                   journal->j_tail, journal->j_tail_sequence, journal->j_errno);
1279
1280         sb->s_sequence = cpu_to_be32(journal->j_tail_sequence);
1281         sb->s_start    = cpu_to_be32(journal->j_tail);
1282         sb->s_errno    = cpu_to_be32(journal->j_errno);
1283         spin_unlock(&journal->j_state_lock);
1284
1285         BUFFER_TRACE(bh, "marking dirty");
1286         mark_buffer_dirty(bh);
1287         if (wait) {
1288                 sync_dirty_buffer(bh);
1289                 if (buffer_write_io_error(bh)) {
1290                         printk(KERN_ERR "JBD2: I/O error detected "
1291                                "when updating journal superblock for %s.\n",
1292                                journal->j_devname);
1293                         clear_buffer_write_io_error(bh);
1294                         set_buffer_uptodate(bh);
1295                 }
1296         } else
1297                 ll_rw_block(SWRITE, 1, &bh);
1298
1299 out:
1300         /* If we have just flushed the log (by marking s_start==0), then
1301          * any future commit will have to be careful to update the
1302          * superblock again to re-record the true start of the log. */
1303
1304         spin_lock(&journal->j_state_lock);
1305         if (sb->s_start)
1306                 journal->j_flags &= ~JBD2_FLUSHED;
1307         else
1308                 journal->j_flags |= JBD2_FLUSHED;
1309         spin_unlock(&journal->j_state_lock);
1310 }
1311
1312 /*
1313  * Read the superblock for a given journal, performing initial
1314  * validation of the format.
1315  */
1316
1317 static int journal_get_superblock(journal_t *journal)
1318 {
1319         struct buffer_head *bh;
1320         journal_superblock_t *sb;
1321         int err = -EIO;
1322
1323         bh = journal->j_sb_buffer;
1324
1325         J_ASSERT(bh != NULL);
1326         if (!buffer_uptodate(bh)) {
1327                 ll_rw_block(READ, 1, &bh);
1328                 wait_on_buffer(bh);
1329                 if (!buffer_uptodate(bh)) {
1330                         printk (KERN_ERR
1331                                 "JBD: IO error reading journal superblock\n");
1332                         goto out;
1333                 }
1334         }
1335
1336         sb = journal->j_superblock;
1337
1338         err = -EINVAL;
1339
1340         if (sb->s_header.h_magic != cpu_to_be32(JBD2_MAGIC_NUMBER) ||
1341             sb->s_blocksize != cpu_to_be32(journal->j_blocksize)) {
1342                 printk(KERN_WARNING "JBD: no valid journal superblock found\n");
1343                 goto out;
1344         }
1345
1346         switch(be32_to_cpu(sb->s_header.h_blocktype)) {
1347         case JBD2_SUPERBLOCK_V1:
1348                 journal->j_format_version = 1;
1349                 break;
1350         case JBD2_SUPERBLOCK_V2:
1351                 journal->j_format_version = 2;
1352                 break;
1353         default:
1354                 printk(KERN_WARNING "JBD: unrecognised superblock format ID\n");
1355                 goto out;
1356         }
1357
1358         if (be32_to_cpu(sb->s_maxlen) < journal->j_maxlen)
1359                 journal->j_maxlen = be32_to_cpu(sb->s_maxlen);
1360         else if (be32_to_cpu(sb->s_maxlen) > journal->j_maxlen) {
1361                 printk (KERN_WARNING "JBD: journal file too short\n");
1362                 goto out;
1363         }
1364
1365         return 0;
1366
1367 out:
1368         journal_fail_superblock(journal);
1369         return err;
1370 }
1371
1372 /*
1373  * Load the on-disk journal superblock and read the key fields into the
1374  * journal_t.
1375  */
1376
1377 static int load_superblock(journal_t *journal)
1378 {
1379         int err;
1380         journal_superblock_t *sb;
1381
1382         err = journal_get_superblock(journal);
1383         if (err)
1384                 return err;
1385
1386         sb = journal->j_superblock;
1387
1388         journal->j_tail_sequence = be32_to_cpu(sb->s_sequence);
1389         journal->j_tail = be32_to_cpu(sb->s_start);
1390         journal->j_first = be32_to_cpu(sb->s_first);
1391         journal->j_last = be32_to_cpu(sb->s_maxlen);
1392         journal->j_errno = be32_to_cpu(sb->s_errno);
1393
1394         return 0;
1395 }
1396
1397
1398 /**
1399  * int jbd2_journal_load() - Read journal from disk.
1400  * @journal: Journal to act on.
1401  *
1402  * Given a journal_t structure which tells us which disk blocks contain
1403  * a journal, read the journal from disk to initialise the in-memory
1404  * structures.
1405  */
1406 int jbd2_journal_load(journal_t *journal)
1407 {
1408         int err;
1409         journal_superblock_t *sb;
1410
1411         err = load_superblock(journal);
1412         if (err)
1413                 return err;
1414
1415         sb = journal->j_superblock;
1416         /* If this is a V2 superblock, then we have to check the
1417          * features flags on it. */
1418
1419         if (journal->j_format_version >= 2) {
1420                 if ((sb->s_feature_ro_compat &
1421                      ~cpu_to_be32(JBD2_KNOWN_ROCOMPAT_FEATURES)) ||
1422                     (sb->s_feature_incompat &
1423                      ~cpu_to_be32(JBD2_KNOWN_INCOMPAT_FEATURES))) {
1424                         printk (KERN_WARNING
1425                                 "JBD: Unrecognised features on journal\n");
1426                         return -EINVAL;
1427                 }
1428         }
1429
1430         /* Let the recovery code check whether it needs to recover any
1431          * data from the journal. */
1432         if (jbd2_journal_recover(journal))
1433                 goto recovery_error;
1434
1435         /* OK, we've finished with the dynamic journal bits:
1436          * reinitialise the dynamic contents of the superblock in memory
1437          * and reset them on disk. */
1438         if (journal_reset(journal))
1439                 goto recovery_error;
1440
1441         journal->j_flags &= ~JBD2_ABORT;
1442         journal->j_flags |= JBD2_LOADED;
1443         return 0;
1444
1445 recovery_error:
1446         printk (KERN_WARNING "JBD: recovery failed\n");
1447         return -EIO;
1448 }
1449
1450 /**
1451  * void jbd2_journal_destroy() - Release a journal_t structure.
1452  * @journal: Journal to act on.
1453  *
1454  * Release a journal_t structure once it is no longer in use by the
1455  * journaled object.
1456  * Return <0 if we couldn't clean up the journal.
1457  */
1458 int jbd2_journal_destroy(journal_t *journal)
1459 {
1460         int err = 0;
1461
1462         /* Wait for the commit thread to wake up and die. */
1463         journal_kill_thread(journal);
1464
1465         /* Force a final log commit */
1466         if (journal->j_running_transaction)
1467                 jbd2_journal_commit_transaction(journal);
1468
1469         /* Force any old transactions to disk */
1470
1471         /* Totally anal locking here... */
1472         spin_lock(&journal->j_list_lock);
1473         while (journal->j_checkpoint_transactions != NULL) {
1474                 spin_unlock(&journal->j_list_lock);
1475                 jbd2_log_do_checkpoint(journal);
1476                 spin_lock(&journal->j_list_lock);
1477         }
1478
1479         J_ASSERT(journal->j_running_transaction == NULL);
1480         J_ASSERT(journal->j_committing_transaction == NULL);
1481         J_ASSERT(journal->j_checkpoint_transactions == NULL);
1482         spin_unlock(&journal->j_list_lock);
1483
1484         if (journal->j_sb_buffer) {
1485                 if (!is_journal_aborted(journal)) {
1486                         /* We can now mark the journal as empty. */
1487                         journal->j_tail = 0;
1488                         journal->j_tail_sequence =
1489                                 ++journal->j_transaction_sequence;
1490                         jbd2_journal_update_superblock(journal, 1);
1491                 } else {
1492                         err = -EIO;
1493                 }
1494                 brelse(journal->j_sb_buffer);
1495         }
1496
1497         if (journal->j_proc_entry)
1498                 jbd2_stats_proc_exit(journal);
1499         if (journal->j_inode)
1500                 iput(journal->j_inode);
1501         if (journal->j_revoke)
1502                 jbd2_journal_destroy_revoke(journal);
1503         kfree(journal->j_wbuf);
1504         kfree(journal);
1505
1506         return err;
1507 }
1508
1509
1510 /**
1511  *int jbd2_journal_check_used_features () - Check if features specified are used.
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 journal uses all of a given set of
1518  * features.  Return true (non-zero) if it does.
1519  **/
1520
1521 int jbd2_journal_check_used_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         if (journal->j_format_version == 1)
1529                 return 0;
1530
1531         sb = journal->j_superblock;
1532
1533         if (((be32_to_cpu(sb->s_feature_compat) & compat) == compat) &&
1534             ((be32_to_cpu(sb->s_feature_ro_compat) & ro) == ro) &&
1535             ((be32_to_cpu(sb->s_feature_incompat) & incompat) == incompat))
1536                 return 1;
1537
1538         return 0;
1539 }
1540
1541 /**
1542  * int jbd2_journal_check_available_features() - Check feature set in journalling layer
1543  * @journal: Journal to check.
1544  * @compat: bitmask of compatible features
1545  * @ro: bitmask of features that force read-only mount
1546  * @incompat: bitmask of incompatible features
1547  *
1548  * Check whether the journaling code supports the use of
1549  * all of a given set of features on this journal.  Return true
1550  * (non-zero) if it can. */
1551
1552 int jbd2_journal_check_available_features (journal_t *journal, unsigned long compat,
1553                                       unsigned long ro, unsigned long incompat)
1554 {
1555         journal_superblock_t *sb;
1556
1557         if (!compat && !ro && !incompat)
1558                 return 1;
1559
1560         sb = journal->j_superblock;
1561
1562         /* We can support any known requested features iff the
1563          * superblock is in version 2.  Otherwise we fail to support any
1564          * extended sb features. */
1565
1566         if (journal->j_format_version != 2)
1567                 return 0;
1568
1569         if ((compat   & JBD2_KNOWN_COMPAT_FEATURES) == compat &&
1570             (ro       & JBD2_KNOWN_ROCOMPAT_FEATURES) == ro &&
1571             (incompat & JBD2_KNOWN_INCOMPAT_FEATURES) == incompat)
1572                 return 1;
1573
1574         return 0;
1575 }
1576
1577 /**
1578  * int jbd2_journal_set_features () - Mark a given journal feature in the superblock
1579  * @journal: Journal to act on.
1580  * @compat: bitmask of compatible features
1581  * @ro: bitmask of features that force read-only mount
1582  * @incompat: bitmask of incompatible features
1583  *
1584  * Mark a given journal feature as present on the
1585  * superblock.  Returns true if the requested features could be set.
1586  *
1587  */
1588
1589 int jbd2_journal_set_features (journal_t *journal, unsigned long compat,
1590                           unsigned long ro, unsigned long incompat)
1591 {
1592         journal_superblock_t *sb;
1593
1594         if (jbd2_journal_check_used_features(journal, compat, ro, incompat))
1595                 return 1;
1596
1597         if (!jbd2_journal_check_available_features(journal, compat, ro, incompat))
1598                 return 0;
1599
1600         jbd_debug(1, "Setting new features 0x%lx/0x%lx/0x%lx\n",
1601                   compat, ro, incompat);
1602
1603         sb = journal->j_superblock;
1604
1605         sb->s_feature_compat    |= cpu_to_be32(compat);
1606         sb->s_feature_ro_compat |= cpu_to_be32(ro);
1607         sb->s_feature_incompat  |= cpu_to_be32(incompat);
1608
1609         return 1;
1610 }
1611
1612 /*
1613  * jbd2_journal_clear_features () - Clear a given journal feature in the
1614  *                                  superblock
1615  * @journal: Journal to act on.
1616  * @compat: bitmask of compatible features
1617  * @ro: bitmask of features that force read-only mount
1618  * @incompat: bitmask of incompatible features
1619  *
1620  * Clear a given journal feature as present on the
1621  * superblock.
1622  */
1623 void jbd2_journal_clear_features(journal_t *journal, unsigned long compat,
1624                                 unsigned long ro, unsigned long incompat)
1625 {
1626         journal_superblock_t *sb;
1627
1628         jbd_debug(1, "Clear features 0x%lx/0x%lx/0x%lx\n",
1629                   compat, ro, incompat);
1630
1631         sb = journal->j_superblock;
1632
1633         sb->s_feature_compat    &= ~cpu_to_be32(compat);
1634         sb->s_feature_ro_compat &= ~cpu_to_be32(ro);
1635         sb->s_feature_incompat  &= ~cpu_to_be32(incompat);
1636 }
1637 EXPORT_SYMBOL(jbd2_journal_clear_features);
1638
1639 /**
1640  * int jbd2_journal_update_format () - Update on-disk journal structure.
1641  * @journal: Journal to act on.
1642  *
1643  * Given an initialised but unloaded journal struct, poke about in the
1644  * on-disk structure to update it to the most recent supported version.
1645  */
1646 int jbd2_journal_update_format (journal_t *journal)
1647 {
1648         journal_superblock_t *sb;
1649         int err;
1650
1651         err = journal_get_superblock(journal);
1652         if (err)
1653                 return err;
1654
1655         sb = journal->j_superblock;
1656
1657         switch (be32_to_cpu(sb->s_header.h_blocktype)) {
1658         case JBD2_SUPERBLOCK_V2:
1659                 return 0;
1660         case JBD2_SUPERBLOCK_V1:
1661                 return journal_convert_superblock_v1(journal, sb);
1662         default:
1663                 break;
1664         }
1665         return -EINVAL;
1666 }
1667
1668 static int journal_convert_superblock_v1(journal_t *journal,
1669                                          journal_superblock_t *sb)
1670 {
1671         int offset, blocksize;
1672         struct buffer_head *bh;
1673
1674         printk(KERN_WARNING
1675                 "JBD: Converting superblock from version 1 to 2.\n");
1676
1677         /* Pre-initialise new fields to zero */
1678         offset = ((char *) &(sb->s_feature_compat)) - ((char *) sb);
1679         blocksize = be32_to_cpu(sb->s_blocksize);
1680         memset(&sb->s_feature_compat, 0, blocksize-offset);
1681
1682         sb->s_nr_users = cpu_to_be32(1);
1683         sb->s_header.h_blocktype = cpu_to_be32(JBD2_SUPERBLOCK_V2);
1684         journal->j_format_version = 2;
1685
1686         bh = journal->j_sb_buffer;
1687         BUFFER_TRACE(bh, "marking dirty");
1688         mark_buffer_dirty(bh);
1689         sync_dirty_buffer(bh);
1690         return 0;
1691 }
1692
1693
1694 /**
1695  * int jbd2_journal_flush () - Flush journal
1696  * @journal: Journal to act on.
1697  *
1698  * Flush all data for a given journal to disk and empty the journal.
1699  * Filesystems can use this when remounting readonly to ensure that
1700  * recovery does not need to happen on remount.
1701  */
1702
1703 int jbd2_journal_flush(journal_t *journal)
1704 {
1705         int err = 0;
1706         transaction_t *transaction = NULL;
1707         unsigned long old_tail;
1708
1709         spin_lock(&journal->j_state_lock);
1710
1711         /* Force everything buffered to the log... */
1712         if (journal->j_running_transaction) {
1713                 transaction = journal->j_running_transaction;
1714                 __jbd2_log_start_commit(journal, transaction->t_tid);
1715         } else if (journal->j_committing_transaction)
1716                 transaction = journal->j_committing_transaction;
1717
1718         /* Wait for the log commit to complete... */
1719         if (transaction) {
1720                 tid_t tid = transaction->t_tid;
1721
1722                 spin_unlock(&journal->j_state_lock);
1723                 jbd2_log_wait_commit(journal, tid);
1724         } else {
1725                 spin_unlock(&journal->j_state_lock);
1726         }
1727
1728         /* ...and flush everything in the log out to disk. */
1729         spin_lock(&journal->j_list_lock);
1730         while (!err && journal->j_checkpoint_transactions != NULL) {
1731                 spin_unlock(&journal->j_list_lock);
1732                 mutex_lock(&journal->j_checkpoint_mutex);
1733                 err = jbd2_log_do_checkpoint(journal);
1734                 mutex_unlock(&journal->j_checkpoint_mutex);
1735                 spin_lock(&journal->j_list_lock);
1736         }
1737         spin_unlock(&journal->j_list_lock);
1738
1739         if (is_journal_aborted(journal))
1740                 return -EIO;
1741
1742         jbd2_cleanup_journal_tail(journal);
1743
1744         /* Finally, mark the journal as really needing no recovery.
1745          * This sets s_start==0 in the underlying superblock, which is
1746          * the magic code for a fully-recovered superblock.  Any future
1747          * commits of data to the journal will restore the current
1748          * s_start value. */
1749         spin_lock(&journal->j_state_lock);
1750         old_tail = journal->j_tail;
1751         journal->j_tail = 0;
1752         spin_unlock(&journal->j_state_lock);
1753         jbd2_journal_update_superblock(journal, 1);
1754         spin_lock(&journal->j_state_lock);
1755         journal->j_tail = old_tail;
1756
1757         J_ASSERT(!journal->j_running_transaction);
1758         J_ASSERT(!journal->j_committing_transaction);
1759         J_ASSERT(!journal->j_checkpoint_transactions);
1760         J_ASSERT(journal->j_head == journal->j_tail);
1761         J_ASSERT(journal->j_tail_sequence == journal->j_transaction_sequence);
1762         spin_unlock(&journal->j_state_lock);
1763         return 0;
1764 }
1765
1766 /**
1767  * int jbd2_journal_wipe() - Wipe journal contents
1768  * @journal: Journal to act on.
1769  * @write: flag (see below)
1770  *
1771  * Wipe out all of the contents of a journal, safely.  This will produce
1772  * a warning if the journal contains any valid recovery information.
1773  * Must be called between journal_init_*() and jbd2_journal_load().
1774  *
1775  * If 'write' is non-zero, then we wipe out the journal on disk; otherwise
1776  * we merely suppress recovery.
1777  */
1778
1779 int jbd2_journal_wipe(journal_t *journal, int write)
1780 {
1781         journal_superblock_t *sb;
1782         int err = 0;
1783
1784         J_ASSERT (!(journal->j_flags & JBD2_LOADED));
1785
1786         err = load_superblock(journal);
1787         if (err)
1788                 return err;
1789
1790         sb = journal->j_superblock;
1791
1792         if (!journal->j_tail)
1793                 goto no_recovery;
1794
1795         printk (KERN_WARNING "JBD: %s recovery information on journal\n",
1796                 write ? "Clearing" : "Ignoring");
1797
1798         err = jbd2_journal_skip_recovery(journal);
1799         if (write)
1800                 jbd2_journal_update_superblock(journal, 1);
1801
1802  no_recovery:
1803         return err;
1804 }
1805
1806 /*
1807  * Journal abort has very specific semantics, which we describe
1808  * for journal abort.
1809  *
1810  * Two internal function, which provide abort to te jbd layer
1811  * itself are here.
1812  */
1813
1814 /*
1815  * Quick version for internal journal use (doesn't lock the journal).
1816  * Aborts hard --- we mark the abort as occurred, but do _nothing_ else,
1817  * and don't attempt to make any other journal updates.
1818  */
1819 void __jbd2_journal_abort_hard(journal_t *journal)
1820 {
1821         transaction_t *transaction;
1822
1823         if (journal->j_flags & JBD2_ABORT)
1824                 return;
1825
1826         printk(KERN_ERR "Aborting journal on device %s.\n",
1827                journal->j_devname);
1828
1829         spin_lock(&journal->j_state_lock);
1830         journal->j_flags |= JBD2_ABORT;
1831         transaction = journal->j_running_transaction;
1832         if (transaction)
1833                 __jbd2_log_start_commit(journal, transaction->t_tid);
1834         spin_unlock(&journal->j_state_lock);
1835 }
1836
1837 /* Soft abort: record the abort error status in the journal superblock,
1838  * but don't do any other IO. */
1839 static void __journal_abort_soft (journal_t *journal, int errno)
1840 {
1841         if (journal->j_flags & JBD2_ABORT)
1842                 return;
1843
1844         if (!journal->j_errno)
1845                 journal->j_errno = errno;
1846
1847         __jbd2_journal_abort_hard(journal);
1848
1849         if (errno)
1850                 jbd2_journal_update_superblock(journal, 1);
1851 }
1852
1853 /**
1854  * void jbd2_journal_abort () - Shutdown the journal immediately.
1855  * @journal: the journal to shutdown.
1856  * @errno:   an error number to record in the journal indicating
1857  *           the reason for the shutdown.
1858  *
1859  * Perform a complete, immediate shutdown of the ENTIRE
1860  * journal (not of a single transaction).  This operation cannot be
1861  * undone without closing and reopening the journal.
1862  *
1863  * The jbd2_journal_abort function is intended to support higher level error
1864  * recovery mechanisms such as the ext2/ext3 remount-readonly error
1865  * mode.
1866  *
1867  * Journal abort has very specific semantics.  Any existing dirty,
1868  * unjournaled buffers in the main filesystem will still be written to
1869  * disk by bdflush, but the journaling mechanism will be suspended
1870  * immediately and no further transaction commits will be honoured.
1871  *
1872  * Any dirty, journaled buffers will be written back to disk without
1873  * hitting the journal.  Atomicity cannot be guaranteed on an aborted
1874  * filesystem, but we _do_ attempt to leave as much data as possible
1875  * behind for fsck to use for cleanup.
1876  *
1877  * Any attempt to get a new transaction handle on a journal which is in
1878  * ABORT state will just result in an -EROFS error return.  A
1879  * jbd2_journal_stop on an existing handle will return -EIO if we have
1880  * entered abort state during the update.
1881  *
1882  * Recursive transactions are not disturbed by journal abort until the
1883  * final jbd2_journal_stop, which will receive the -EIO error.
1884  *
1885  * Finally, the jbd2_journal_abort call allows the caller to supply an errno
1886  * which will be recorded (if possible) in the journal superblock.  This
1887  * allows a client to record failure conditions in the middle of a
1888  * transaction without having to complete the transaction to record the
1889  * failure to disk.  ext3_error, for example, now uses this
1890  * functionality.
1891  *
1892  * Errors which originate from within the journaling layer will NOT
1893  * supply an errno; a null errno implies that absolutely no further
1894  * writes are done to the journal (unless there are any already in
1895  * progress).
1896  *
1897  */
1898
1899 void jbd2_journal_abort(journal_t *journal, int errno)
1900 {
1901         __journal_abort_soft(journal, errno);
1902 }
1903
1904 /**
1905  * int jbd2_journal_errno () - returns the journal's error state.
1906  * @journal: journal to examine.
1907  *
1908  * This is the errno numbet set with jbd2_journal_abort(), the last
1909  * time the journal was mounted - if the journal was stopped
1910  * without calling abort this will be 0.
1911  *
1912  * If the journal has been aborted on this mount time -EROFS will
1913  * be returned.
1914  */
1915 int jbd2_journal_errno(journal_t *journal)
1916 {
1917         int err;
1918
1919         spin_lock(&journal->j_state_lock);
1920         if (journal->j_flags & JBD2_ABORT)
1921                 err = -EROFS;
1922         else
1923                 err = journal->j_errno;
1924         spin_unlock(&journal->j_state_lock);
1925         return err;
1926 }
1927
1928 /**
1929  * int jbd2_journal_clear_err () - clears the journal's error state
1930  * @journal: journal to act on.
1931  *
1932  * An error must be cleared or Acked to take a FS out of readonly
1933  * mode.
1934  */
1935 int jbd2_journal_clear_err(journal_t *journal)
1936 {
1937         int err = 0;
1938
1939         spin_lock(&journal->j_state_lock);
1940         if (journal->j_flags & JBD2_ABORT)
1941                 err = -EROFS;
1942         else
1943                 journal->j_errno = 0;
1944         spin_unlock(&journal->j_state_lock);
1945         return err;
1946 }
1947
1948 /**
1949  * void jbd2_journal_ack_err() - Ack journal err.
1950  * @journal: journal to act on.
1951  *
1952  * An error must be cleared or Acked to take a FS out of readonly
1953  * mode.
1954  */
1955 void jbd2_journal_ack_err(journal_t *journal)
1956 {
1957         spin_lock(&journal->j_state_lock);
1958         if (journal->j_errno)
1959                 journal->j_flags |= JBD2_ACK_ERR;
1960         spin_unlock(&journal->j_state_lock);
1961 }
1962
1963 int jbd2_journal_blocks_per_page(struct inode *inode)
1964 {
1965         return 1 << (PAGE_CACHE_SHIFT - inode->i_sb->s_blocksize_bits);
1966 }
1967
1968 /*
1969  * helper functions to deal with 32 or 64bit block numbers.
1970  */
1971 size_t journal_tag_bytes(journal_t *journal)
1972 {
1973         if (JBD2_HAS_INCOMPAT_FEATURE(journal, JBD2_FEATURE_INCOMPAT_64BIT))
1974                 return JBD2_TAG_SIZE64;
1975         else
1976                 return JBD2_TAG_SIZE32;
1977 }
1978
1979 /*
1980  * Journal_head storage management
1981  */
1982 static struct kmem_cache *jbd2_journal_head_cache;
1983 #ifdef CONFIG_JBD2_DEBUG
1984 static atomic_t nr_journal_heads = ATOMIC_INIT(0);
1985 #endif
1986
1987 static int journal_init_jbd2_journal_head_cache(void)
1988 {
1989         int retval;
1990
1991         J_ASSERT(jbd2_journal_head_cache == NULL);
1992         jbd2_journal_head_cache = kmem_cache_create("jbd2_journal_head",
1993                                 sizeof(struct journal_head),
1994                                 0,              /* offset */
1995                                 SLAB_TEMPORARY, /* flags */
1996                                 NULL);          /* ctor */
1997         retval = 0;
1998         if (!jbd2_journal_head_cache) {
1999                 retval = -ENOMEM;
2000                 printk(KERN_EMERG "JBD: no memory for journal_head cache\n");
2001         }
2002         return retval;
2003 }
2004
2005 static void jbd2_journal_destroy_jbd2_journal_head_cache(void)
2006 {
2007         if (jbd2_journal_head_cache) {
2008                 kmem_cache_destroy(jbd2_journal_head_cache);
2009                 jbd2_journal_head_cache = NULL;
2010         }
2011 }
2012
2013 /*
2014  * journal_head splicing and dicing
2015  */
2016 static struct journal_head *journal_alloc_journal_head(void)
2017 {
2018         struct journal_head *ret;
2019         static unsigned long last_warning;
2020
2021 #ifdef CONFIG_JBD2_DEBUG
2022         atomic_inc(&nr_journal_heads);
2023 #endif
2024         ret = kmem_cache_alloc(jbd2_journal_head_cache, GFP_NOFS);
2025         if (!ret) {
2026                 jbd_debug(1, "out of memory for journal_head\n");
2027                 if (time_after(jiffies, last_warning + 5*HZ)) {
2028                         printk(KERN_NOTICE "ENOMEM in %s, retrying.\n",
2029                                __func__);
2030                         last_warning = jiffies;
2031                 }
2032                 while (!ret) {
2033                         yield();
2034                         ret = kmem_cache_alloc(jbd2_journal_head_cache, GFP_NOFS);
2035                 }
2036         }
2037         return ret;
2038 }
2039
2040 static void journal_free_journal_head(struct journal_head *jh)
2041 {
2042 #ifdef CONFIG_JBD2_DEBUG
2043         atomic_dec(&nr_journal_heads);
2044         memset(jh, JBD2_POISON_FREE, sizeof(*jh));
2045 #endif
2046         kmem_cache_free(jbd2_journal_head_cache, jh);
2047 }
2048
2049 /*
2050  * A journal_head is attached to a buffer_head whenever JBD has an
2051  * interest in the buffer.
2052  *
2053  * Whenever a buffer has an attached journal_head, its ->b_state:BH_JBD bit
2054  * is set.  This bit is tested in core kernel code where we need to take
2055  * JBD-specific actions.  Testing the zeroness of ->b_private is not reliable
2056  * there.
2057  *
2058  * When a buffer has its BH_JBD bit set, its ->b_count is elevated by one.
2059  *
2060  * When a buffer has its BH_JBD bit set it is immune from being released by
2061  * core kernel code, mainly via ->b_count.
2062  *
2063  * A journal_head may be detached from its buffer_head when the journal_head's
2064  * b_transaction, b_cp_transaction and b_next_transaction pointers are NULL.
2065  * Various places in JBD call jbd2_journal_remove_journal_head() to indicate that the
2066  * journal_head can be dropped if needed.
2067  *
2068  * Various places in the kernel want to attach a journal_head to a buffer_head
2069  * _before_ attaching the journal_head to a transaction.  To protect the
2070  * journal_head in this situation, jbd2_journal_add_journal_head elevates the
2071  * journal_head's b_jcount refcount by one.  The caller must call
2072  * jbd2_journal_put_journal_head() to undo this.
2073  *
2074  * So the typical usage would be:
2075  *
2076  *      (Attach a journal_head if needed.  Increments b_jcount)
2077  *      struct journal_head *jh = jbd2_journal_add_journal_head(bh);
2078  *      ...
2079  *      jh->b_transaction = xxx;
2080  *      jbd2_journal_put_journal_head(jh);
2081  *
2082  * Now, the journal_head's b_jcount is zero, but it is safe from being released
2083  * because it has a non-zero b_transaction.
2084  */
2085
2086 /*
2087  * Give a buffer_head a journal_head.
2088  *
2089  * Doesn't need the journal lock.
2090  * May sleep.
2091  */
2092 struct journal_head *jbd2_journal_add_journal_head(struct buffer_head *bh)
2093 {
2094         struct journal_head *jh;
2095         struct journal_head *new_jh = NULL;
2096
2097 repeat:
2098         if (!buffer_jbd(bh)) {
2099                 new_jh = journal_alloc_journal_head();
2100                 memset(new_jh, 0, sizeof(*new_jh));
2101         }
2102
2103         jbd_lock_bh_journal_head(bh);
2104         if (buffer_jbd(bh)) {
2105                 jh = bh2jh(bh);
2106         } else {
2107                 J_ASSERT_BH(bh,
2108                         (atomic_read(&bh->b_count) > 0) ||
2109                         (bh->b_page && bh->b_page->mapping));
2110
2111                 if (!new_jh) {
2112                         jbd_unlock_bh_journal_head(bh);
2113                         goto repeat;
2114                 }
2115
2116                 jh = new_jh;
2117                 new_jh = NULL;          /* We consumed it */
2118                 set_buffer_jbd(bh);
2119                 bh->b_private = jh;
2120                 jh->b_bh = bh;
2121                 get_bh(bh);
2122                 BUFFER_TRACE(bh, "added journal_head");
2123         }
2124         jh->b_jcount++;
2125         jbd_unlock_bh_journal_head(bh);
2126         if (new_jh)
2127                 journal_free_journal_head(new_jh);
2128         return bh->b_private;
2129 }
2130
2131 /*
2132  * Grab a ref against this buffer_head's journal_head.  If it ended up not
2133  * having a journal_head, return NULL
2134  */
2135 struct journal_head *jbd2_journal_grab_journal_head(struct buffer_head *bh)
2136 {
2137         struct journal_head *jh = NULL;
2138
2139         jbd_lock_bh_journal_head(bh);
2140         if (buffer_jbd(bh)) {
2141                 jh = bh2jh(bh);
2142                 jh->b_jcount++;
2143         }
2144         jbd_unlock_bh_journal_head(bh);
2145         return jh;
2146 }
2147
2148 static void __journal_remove_journal_head(struct buffer_head *bh)
2149 {
2150         struct journal_head *jh = bh2jh(bh);
2151
2152         J_ASSERT_JH(jh, jh->b_jcount >= 0);
2153
2154         get_bh(bh);
2155         if (jh->b_jcount == 0) {
2156                 if (jh->b_transaction == NULL &&
2157                                 jh->b_next_transaction == NULL &&
2158                                 jh->b_cp_transaction == NULL) {
2159                         J_ASSERT_JH(jh, jh->b_jlist == BJ_None);
2160                         J_ASSERT_BH(bh, buffer_jbd(bh));
2161                         J_ASSERT_BH(bh, jh2bh(jh) == bh);
2162                         BUFFER_TRACE(bh, "remove journal_head");
2163                         if (jh->b_frozen_data) {
2164                                 printk(KERN_WARNING "%s: freeing "
2165                                                 "b_frozen_data\n",
2166                                                 __func__);
2167                                 jbd2_free(jh->b_frozen_data, bh->b_size);
2168                         }
2169                         if (jh->b_committed_data) {
2170                                 printk(KERN_WARNING "%s: freeing "
2171                                                 "b_committed_data\n",
2172                                                 __func__);
2173                                 jbd2_free(jh->b_committed_data, bh->b_size);
2174                         }
2175                         bh->b_private = NULL;
2176                         jh->b_bh = NULL;        /* debug, really */
2177                         clear_buffer_jbd(bh);
2178                         __brelse(bh);
2179                         journal_free_journal_head(jh);
2180                 } else {
2181                         BUFFER_TRACE(bh, "journal_head was locked");
2182                 }
2183         }
2184 }
2185
2186 /*
2187  * jbd2_journal_remove_journal_head(): if the buffer isn't attached to a transaction
2188  * and has a zero b_jcount then remove and release its journal_head.   If we did
2189  * see that the buffer is not used by any transaction we also "logically"
2190  * decrement ->b_count.
2191  *
2192  * We in fact take an additional increment on ->b_count as a convenience,
2193  * because the caller usually wants to do additional things with the bh
2194  * after calling here.
2195  * The caller of jbd2_journal_remove_journal_head() *must* run __brelse(bh) at some
2196  * time.  Once the caller has run __brelse(), the buffer is eligible for
2197  * reaping by try_to_free_buffers().
2198  */
2199 void jbd2_journal_remove_journal_head(struct buffer_head *bh)
2200 {
2201         jbd_lock_bh_journal_head(bh);
2202         __journal_remove_journal_head(bh);
2203         jbd_unlock_bh_journal_head(bh);
2204 }
2205
2206 /*
2207  * Drop a reference on the passed journal_head.  If it fell to zero then try to
2208  * release the journal_head from the buffer_head.
2209  */
2210 void jbd2_journal_put_journal_head(struct journal_head *jh)
2211 {
2212         struct buffer_head *bh = jh2bh(jh);
2213
2214         jbd_lock_bh_journal_head(bh);
2215         J_ASSERT_JH(jh, jh->b_jcount > 0);
2216         --jh->b_jcount;
2217         if (!jh->b_jcount && !jh->b_transaction) {
2218                 __journal_remove_journal_head(bh);
2219                 __brelse(bh);
2220         }
2221         jbd_unlock_bh_journal_head(bh);
2222 }
2223
2224 /*
2225  * Initialize jbd inode head
2226  */
2227 void jbd2_journal_init_jbd_inode(struct jbd2_inode *jinode, struct inode *inode)
2228 {
2229         jinode->i_transaction = NULL;
2230         jinode->i_next_transaction = NULL;
2231         jinode->i_vfs_inode = inode;
2232         jinode->i_flags = 0;
2233         INIT_LIST_HEAD(&jinode->i_list);
2234 }
2235
2236 /*
2237  * Function to be called before we start removing inode from memory (i.e.,
2238  * clear_inode() is a fine place to be called from). It removes inode from
2239  * transaction's lists.
2240  */
2241 void jbd2_journal_release_jbd_inode(journal_t *journal,
2242                                     struct jbd2_inode *jinode)
2243 {
2244         int writeout = 0;
2245
2246         if (!journal)
2247                 return;
2248 restart:
2249         spin_lock(&journal->j_list_lock);
2250         /* Is commit writing out inode - we have to wait */
2251         if (jinode->i_flags & JI_COMMIT_RUNNING) {
2252                 wait_queue_head_t *wq;
2253                 DEFINE_WAIT_BIT(wait, &jinode->i_flags, __JI_COMMIT_RUNNING);
2254                 wq = bit_waitqueue(&jinode->i_flags, __JI_COMMIT_RUNNING);
2255                 prepare_to_wait(wq, &wait.wait, TASK_UNINTERRUPTIBLE);
2256                 spin_unlock(&journal->j_list_lock);
2257                 schedule();
2258                 finish_wait(wq, &wait.wait);
2259                 goto restart;
2260         }
2261
2262         /* Do we need to wait for data writeback? */
2263         if (journal->j_committing_transaction == jinode->i_transaction)
2264                 writeout = 1;
2265         if (jinode->i_transaction) {
2266                 list_del(&jinode->i_list);
2267                 jinode->i_transaction = NULL;
2268         }
2269         spin_unlock(&journal->j_list_lock);
2270 }
2271
2272 /*
2273  * debugfs tunables
2274  */
2275 #ifdef CONFIG_JBD2_DEBUG
2276 u8 jbd2_journal_enable_debug __read_mostly;
2277 EXPORT_SYMBOL(jbd2_journal_enable_debug);
2278
2279 #define JBD2_DEBUG_NAME "jbd2-debug"
2280
2281 static struct dentry *jbd2_debugfs_dir;
2282 static struct dentry *jbd2_debug;
2283
2284 static void __init jbd2_create_debugfs_entry(void)
2285 {
2286         jbd2_debugfs_dir = debugfs_create_dir("jbd2", NULL);
2287         if (jbd2_debugfs_dir)
2288                 jbd2_debug = debugfs_create_u8(JBD2_DEBUG_NAME, S_IRUGO,
2289                                                jbd2_debugfs_dir,
2290                                                &jbd2_journal_enable_debug);
2291 }
2292
2293 static void __exit jbd2_remove_debugfs_entry(void)
2294 {
2295         debugfs_remove(jbd2_debug);
2296         debugfs_remove(jbd2_debugfs_dir);
2297 }
2298
2299 #else
2300
2301 static void __init jbd2_create_debugfs_entry(void)
2302 {
2303 }
2304
2305 static void __exit jbd2_remove_debugfs_entry(void)
2306 {
2307 }
2308
2309 #endif
2310
2311 #ifdef CONFIG_PROC_FS
2312
2313 #define JBD2_STATS_PROC_NAME "fs/jbd2"
2314
2315 static void __init jbd2_create_jbd_stats_proc_entry(void)
2316 {
2317         proc_jbd2_stats = proc_mkdir(JBD2_STATS_PROC_NAME, NULL);
2318 }
2319
2320 static void __exit jbd2_remove_jbd_stats_proc_entry(void)
2321 {
2322         if (proc_jbd2_stats)
2323                 remove_proc_entry(JBD2_STATS_PROC_NAME, NULL);
2324 }
2325
2326 #else
2327
2328 #define jbd2_create_jbd_stats_proc_entry() do {} while (0)
2329 #define jbd2_remove_jbd_stats_proc_entry() do {} while (0)
2330
2331 #endif
2332
2333 struct kmem_cache *jbd2_handle_cache;
2334
2335 static int __init journal_init_handle_cache(void)
2336 {
2337         jbd2_handle_cache = kmem_cache_create("jbd2_journal_handle",
2338                                 sizeof(handle_t),
2339                                 0,              /* offset */
2340                                 SLAB_TEMPORARY, /* flags */
2341                                 NULL);          /* ctor */
2342         if (jbd2_handle_cache == NULL) {
2343                 printk(KERN_EMERG "JBD: failed to create handle cache\n");
2344                 return -ENOMEM;
2345         }
2346         return 0;
2347 }
2348
2349 static void jbd2_journal_destroy_handle_cache(void)
2350 {
2351         if (jbd2_handle_cache)
2352                 kmem_cache_destroy(jbd2_handle_cache);
2353 }
2354
2355 /*
2356  * Module startup and shutdown
2357  */
2358
2359 static int __init journal_init_caches(void)
2360 {
2361         int ret;
2362
2363         ret = jbd2_journal_init_revoke_caches();
2364         if (ret == 0)
2365                 ret = journal_init_jbd2_journal_head_cache();
2366         if (ret == 0)
2367                 ret = journal_init_handle_cache();
2368         return ret;
2369 }
2370
2371 static void jbd2_journal_destroy_caches(void)
2372 {
2373         jbd2_journal_destroy_revoke_caches();
2374         jbd2_journal_destroy_jbd2_journal_head_cache();
2375         jbd2_journal_destroy_handle_cache();
2376 }
2377
2378 static int __init journal_init(void)
2379 {
2380         int ret;
2381
2382         BUILD_BUG_ON(sizeof(struct journal_superblock_s) != 1024);
2383
2384         ret = journal_init_caches();
2385         if (ret == 0) {
2386                 jbd2_create_debugfs_entry();
2387                 jbd2_create_jbd_stats_proc_entry();
2388         } else {
2389                 jbd2_journal_destroy_caches();
2390         }
2391         return ret;
2392 }
2393
2394 static void __exit journal_exit(void)
2395 {
2396 #ifdef CONFIG_JBD2_DEBUG
2397         int n = atomic_read(&nr_journal_heads);
2398         if (n)
2399                 printk(KERN_EMERG "JBD: leaked %d journal_heads!\n", n);
2400 #endif
2401         jbd2_remove_debugfs_entry();
2402         jbd2_remove_jbd_stats_proc_entry();
2403         jbd2_journal_destroy_caches();
2404 }
2405
2406 MODULE_LICENSE("GPL");
2407 module_init(journal_init);
2408 module_exit(journal_exit);
2409