Merge branch 'stealer/ipvs/sync-daemon-cleanup-for-next' of git://git.stealer.net...
[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                        __func__, 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                                 __func__,
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         bdevname(journal->j_dev, name);
905         journal->j_proc_entry = proc_mkdir(name, proc_jbd2_stats);
906         if (journal->j_proc_entry) {
907                 proc_create_data("history", S_IRUGO, journal->j_proc_entry,
908                                  &jbd2_seq_history_fops, journal);
909                 proc_create_data("info", S_IRUGO, journal->j_proc_entry,
910                                  &jbd2_seq_info_fops, journal);
911         }
912 }
913
914 static void jbd2_stats_proc_exit(journal_t *journal)
915 {
916         char name[BDEVNAME_SIZE];
917
918         bdevname(journal->j_dev, name);
919         remove_proc_entry("info", journal->j_proc_entry);
920         remove_proc_entry("history", journal->j_proc_entry);
921         remove_proc_entry(name, proc_jbd2_stats);
922 }
923
924 static void journal_init_stats(journal_t *journal)
925 {
926         int size;
927
928         if (!proc_jbd2_stats)
929                 return;
930
931         journal->j_history_max = 100;
932         size = sizeof(struct transaction_stats_s) * journal->j_history_max;
933         journal->j_history = kzalloc(size, GFP_KERNEL);
934         if (!journal->j_history) {
935                 journal->j_history_max = 0;
936                 return;
937         }
938         spin_lock_init(&journal->j_history_lock);
939 }
940
941 /*
942  * Management for journal control blocks: functions to create and
943  * destroy journal_t structures, and to initialise and read existing
944  * journal blocks from disk.  */
945
946 /* First: create and setup a journal_t object in memory.  We initialise
947  * very few fields yet: that has to wait until we have created the
948  * journal structures from from scratch, or loaded them from disk. */
949
950 static journal_t * journal_init_common (void)
951 {
952         journal_t *journal;
953         int err;
954
955         journal = kzalloc(sizeof(*journal), GFP_KERNEL|__GFP_NOFAIL);
956         if (!journal)
957                 goto fail;
958
959         init_waitqueue_head(&journal->j_wait_transaction_locked);
960         init_waitqueue_head(&journal->j_wait_logspace);
961         init_waitqueue_head(&journal->j_wait_done_commit);
962         init_waitqueue_head(&journal->j_wait_checkpoint);
963         init_waitqueue_head(&journal->j_wait_commit);
964         init_waitqueue_head(&journal->j_wait_updates);
965         mutex_init(&journal->j_barrier);
966         mutex_init(&journal->j_checkpoint_mutex);
967         spin_lock_init(&journal->j_revoke_lock);
968         spin_lock_init(&journal->j_list_lock);
969         spin_lock_init(&journal->j_state_lock);
970
971         journal->j_commit_interval = (HZ * JBD2_DEFAULT_MAX_COMMIT_AGE);
972
973         /* The journal is marked for error until we succeed with recovery! */
974         journal->j_flags = JBD2_ABORT;
975
976         /* Set up a default-sized revoke table for the new mount. */
977         err = jbd2_journal_init_revoke(journal, JOURNAL_REVOKE_DEFAULT_HASH);
978         if (err) {
979                 kfree(journal);
980                 goto fail;
981         }
982
983         journal_init_stats(journal);
984
985         return journal;
986 fail:
987         return NULL;
988 }
989
990 /* jbd2_journal_init_dev and jbd2_journal_init_inode:
991  *
992  * Create a journal structure assigned some fixed set of disk blocks to
993  * the journal.  We don't actually touch those disk blocks yet, but we
994  * need to set up all of the mapping information to tell the journaling
995  * system where the journal blocks are.
996  *
997  */
998
999 /**
1000  *  journal_t * jbd2_journal_init_dev() - creates and initialises a journal structure
1001  *  @bdev: Block device on which to create the journal
1002  *  @fs_dev: Device which hold journalled filesystem for this journal.
1003  *  @start: Block nr Start of journal.
1004  *  @len:  Length of the journal in blocks.
1005  *  @blocksize: blocksize of journalling device
1006  *
1007  *  Returns: a newly created journal_t *
1008  *
1009  *  jbd2_journal_init_dev creates a journal which maps a fixed contiguous
1010  *  range of blocks on an arbitrary block device.
1011  *
1012  */
1013 journal_t * jbd2_journal_init_dev(struct block_device *bdev,
1014                         struct block_device *fs_dev,
1015                         unsigned long long start, int len, int blocksize)
1016 {
1017         journal_t *journal = journal_init_common();
1018         struct buffer_head *bh;
1019         int n;
1020
1021         if (!journal)
1022                 return NULL;
1023
1024         /* journal descriptor can store up to n blocks -bzzz */
1025         journal->j_blocksize = blocksize;
1026         n = journal->j_blocksize / sizeof(journal_block_tag_t);
1027         journal->j_wbufsize = n;
1028         journal->j_wbuf = kmalloc(n * sizeof(struct buffer_head*), GFP_KERNEL);
1029         if (!journal->j_wbuf) {
1030                 printk(KERN_ERR "%s: Cant allocate bhs for commit thread\n",
1031                         __func__);
1032                 kfree(journal);
1033                 journal = NULL;
1034                 goto out;
1035         }
1036         journal->j_dev = bdev;
1037         journal->j_fs_dev = fs_dev;
1038         journal->j_blk_offset = start;
1039         journal->j_maxlen = len;
1040         jbd2_stats_proc_init(journal);
1041
1042         bh = __getblk(journal->j_dev, start, journal->j_blocksize);
1043         J_ASSERT(bh != NULL);
1044         journal->j_sb_buffer = bh;
1045         journal->j_superblock = (journal_superblock_t *)bh->b_data;
1046 out:
1047         return journal;
1048 }
1049
1050 /**
1051  *  journal_t * jbd2_journal_init_inode () - creates a journal which maps to a inode.
1052  *  @inode: An inode to create the journal in
1053  *
1054  * jbd2_journal_init_inode creates a journal which maps an on-disk inode as
1055  * the journal.  The inode must exist already, must support bmap() and
1056  * must have all data blocks preallocated.
1057  */
1058 journal_t * jbd2_journal_init_inode (struct inode *inode)
1059 {
1060         struct buffer_head *bh;
1061         journal_t *journal = journal_init_common();
1062         int err;
1063         int n;
1064         unsigned long long blocknr;
1065
1066         if (!journal)
1067                 return NULL;
1068
1069         journal->j_dev = journal->j_fs_dev = inode->i_sb->s_bdev;
1070         journal->j_inode = inode;
1071         jbd_debug(1,
1072                   "journal %p: inode %s/%ld, size %Ld, bits %d, blksize %ld\n",
1073                   journal, inode->i_sb->s_id, inode->i_ino,
1074                   (long long) inode->i_size,
1075                   inode->i_sb->s_blocksize_bits, inode->i_sb->s_blocksize);
1076
1077         journal->j_maxlen = inode->i_size >> inode->i_sb->s_blocksize_bits;
1078         journal->j_blocksize = inode->i_sb->s_blocksize;
1079         jbd2_stats_proc_init(journal);
1080
1081         /* journal descriptor can store up to n blocks -bzzz */
1082         n = journal->j_blocksize / sizeof(journal_block_tag_t);
1083         journal->j_wbufsize = n;
1084         journal->j_wbuf = kmalloc(n * sizeof(struct buffer_head*), GFP_KERNEL);
1085         if (!journal->j_wbuf) {
1086                 printk(KERN_ERR "%s: Cant allocate bhs for commit thread\n",
1087                         __func__);
1088                 kfree(journal);
1089                 return NULL;
1090         }
1091
1092         err = jbd2_journal_bmap(journal, 0, &blocknr);
1093         /* If that failed, give up */
1094         if (err) {
1095                 printk(KERN_ERR "%s: Cannnot locate journal superblock\n",
1096                        __func__);
1097                 kfree(journal);
1098                 return NULL;
1099         }
1100
1101         bh = __getblk(journal->j_dev, blocknr, journal->j_blocksize);
1102         J_ASSERT(bh != NULL);
1103         journal->j_sb_buffer = bh;
1104         journal->j_superblock = (journal_superblock_t *)bh->b_data;
1105
1106         return journal;
1107 }
1108
1109 /*
1110  * If the journal init or create aborts, we need to mark the journal
1111  * superblock as being NULL to prevent the journal destroy from writing
1112  * back a bogus superblock.
1113  */
1114 static void journal_fail_superblock (journal_t *journal)
1115 {
1116         struct buffer_head *bh = journal->j_sb_buffer;
1117         brelse(bh);
1118         journal->j_sb_buffer = NULL;
1119 }
1120
1121 /*
1122  * Given a journal_t structure, initialise the various fields for
1123  * startup of a new journaling session.  We use this both when creating
1124  * a journal, and after recovering an old journal to reset it for
1125  * subsequent use.
1126  */
1127
1128 static int journal_reset(journal_t *journal)
1129 {
1130         journal_superblock_t *sb = journal->j_superblock;
1131         unsigned long long first, last;
1132
1133         first = be32_to_cpu(sb->s_first);
1134         last = be32_to_cpu(sb->s_maxlen);
1135
1136         journal->j_first = first;
1137         journal->j_last = last;
1138
1139         journal->j_head = first;
1140         journal->j_tail = first;
1141         journal->j_free = last - first;
1142
1143         journal->j_tail_sequence = journal->j_transaction_sequence;
1144         journal->j_commit_sequence = journal->j_transaction_sequence - 1;
1145         journal->j_commit_request = journal->j_commit_sequence;
1146
1147         journal->j_max_transaction_buffers = journal->j_maxlen / 4;
1148
1149         /* Add the dynamic fields and write it to disk. */
1150         jbd2_journal_update_superblock(journal, 1);
1151         return jbd2_journal_start_thread(journal);
1152 }
1153
1154 /**
1155  * int jbd2_journal_create() - Initialise the new journal file
1156  * @journal: Journal to create. This structure must have been initialised
1157  *
1158  * Given a journal_t structure which tells us which disk blocks we can
1159  * use, create a new journal superblock and initialise all of the
1160  * journal fields from scratch.
1161  **/
1162 int jbd2_journal_create(journal_t *journal)
1163 {
1164         unsigned long long blocknr;
1165         struct buffer_head *bh;
1166         journal_superblock_t *sb;
1167         int i, err;
1168
1169         if (journal->j_maxlen < JBD2_MIN_JOURNAL_BLOCKS) {
1170                 printk (KERN_ERR "Journal length (%d blocks) too short.\n",
1171                         journal->j_maxlen);
1172                 journal_fail_superblock(journal);
1173                 return -EINVAL;
1174         }
1175
1176         if (journal->j_inode == NULL) {
1177                 /*
1178                  * We don't know what block to start at!
1179                  */
1180                 printk(KERN_EMERG
1181                        "%s: creation of journal on external device!\n",
1182                        __func__);
1183                 BUG();
1184         }
1185
1186         /* Zero out the entire journal on disk.  We cannot afford to
1187            have any blocks on disk beginning with JBD2_MAGIC_NUMBER. */
1188         jbd_debug(1, "JBD: Zeroing out journal blocks...\n");
1189         for (i = 0; i < journal->j_maxlen; i++) {
1190                 err = jbd2_journal_bmap(journal, i, &blocknr);
1191                 if (err)
1192                         return err;
1193                 bh = __getblk(journal->j_dev, blocknr, journal->j_blocksize);
1194                 lock_buffer(bh);
1195                 memset (bh->b_data, 0, journal->j_blocksize);
1196                 BUFFER_TRACE(bh, "marking dirty");
1197                 mark_buffer_dirty(bh);
1198                 BUFFER_TRACE(bh, "marking uptodate");
1199                 set_buffer_uptodate(bh);
1200                 unlock_buffer(bh);
1201                 __brelse(bh);
1202         }
1203
1204         sync_blockdev(journal->j_dev);
1205         jbd_debug(1, "JBD: journal cleared.\n");
1206
1207         /* OK, fill in the initial static fields in the new superblock */
1208         sb = journal->j_superblock;
1209
1210         sb->s_header.h_magic     = cpu_to_be32(JBD2_MAGIC_NUMBER);
1211         sb->s_header.h_blocktype = cpu_to_be32(JBD2_SUPERBLOCK_V2);
1212
1213         sb->s_blocksize = cpu_to_be32(journal->j_blocksize);
1214         sb->s_maxlen    = cpu_to_be32(journal->j_maxlen);
1215         sb->s_first     = cpu_to_be32(1);
1216
1217         journal->j_transaction_sequence = 1;
1218
1219         journal->j_flags &= ~JBD2_ABORT;
1220         journal->j_format_version = 2;
1221
1222         return journal_reset(journal);
1223 }
1224
1225 /**
1226  * void jbd2_journal_update_superblock() - Update journal sb on disk.
1227  * @journal: The journal to update.
1228  * @wait: Set to '0' if you don't want to wait for IO completion.
1229  *
1230  * Update a journal's dynamic superblock fields and write it to disk,
1231  * optionally waiting for the IO to complete.
1232  */
1233 void jbd2_journal_update_superblock(journal_t *journal, int wait)
1234 {
1235         journal_superblock_t *sb = journal->j_superblock;
1236         struct buffer_head *bh = journal->j_sb_buffer;
1237
1238         /*
1239          * As a special case, if the on-disk copy is already marked as needing
1240          * no recovery (s_start == 0) and there are no outstanding transactions
1241          * in the filesystem, then we can safely defer the superblock update
1242          * until the next commit by setting JBD2_FLUSHED.  This avoids
1243          * attempting a write to a potential-readonly device.
1244          */
1245         if (sb->s_start == 0 && journal->j_tail_sequence ==
1246                                 journal->j_transaction_sequence) {
1247                 jbd_debug(1,"JBD: Skipping superblock update on recovered sb "
1248                         "(start %ld, seq %d, errno %d)\n",
1249                         journal->j_tail, journal->j_tail_sequence,
1250                         journal->j_errno);
1251                 goto out;
1252         }
1253
1254         spin_lock(&journal->j_state_lock);
1255         jbd_debug(1,"JBD: updating superblock (start %ld, seq %d, errno %d)\n",
1256                   journal->j_tail, journal->j_tail_sequence, journal->j_errno);
1257
1258         sb->s_sequence = cpu_to_be32(journal->j_tail_sequence);
1259         sb->s_start    = cpu_to_be32(journal->j_tail);
1260         sb->s_errno    = cpu_to_be32(journal->j_errno);
1261         spin_unlock(&journal->j_state_lock);
1262
1263         BUFFER_TRACE(bh, "marking dirty");
1264         mark_buffer_dirty(bh);
1265         if (wait)
1266                 sync_dirty_buffer(bh);
1267         else
1268                 ll_rw_block(SWRITE, 1, &bh);
1269
1270 out:
1271         /* If we have just flushed the log (by marking s_start==0), then
1272          * any future commit will have to be careful to update the
1273          * superblock again to re-record the true start of the log. */
1274
1275         spin_lock(&journal->j_state_lock);
1276         if (sb->s_start)
1277                 journal->j_flags &= ~JBD2_FLUSHED;
1278         else
1279                 journal->j_flags |= JBD2_FLUSHED;
1280         spin_unlock(&journal->j_state_lock);
1281 }
1282
1283 /*
1284  * Read the superblock for a given journal, performing initial
1285  * validation of the format.
1286  */
1287
1288 static int journal_get_superblock(journal_t *journal)
1289 {
1290         struct buffer_head *bh;
1291         journal_superblock_t *sb;
1292         int err = -EIO;
1293
1294         bh = journal->j_sb_buffer;
1295
1296         J_ASSERT(bh != NULL);
1297         if (!buffer_uptodate(bh)) {
1298                 ll_rw_block(READ, 1, &bh);
1299                 wait_on_buffer(bh);
1300                 if (!buffer_uptodate(bh)) {
1301                         printk (KERN_ERR
1302                                 "JBD: IO error reading journal superblock\n");
1303                         goto out;
1304                 }
1305         }
1306
1307         sb = journal->j_superblock;
1308
1309         err = -EINVAL;
1310
1311         if (sb->s_header.h_magic != cpu_to_be32(JBD2_MAGIC_NUMBER) ||
1312             sb->s_blocksize != cpu_to_be32(journal->j_blocksize)) {
1313                 printk(KERN_WARNING "JBD: no valid journal superblock found\n");
1314                 goto out;
1315         }
1316
1317         switch(be32_to_cpu(sb->s_header.h_blocktype)) {
1318         case JBD2_SUPERBLOCK_V1:
1319                 journal->j_format_version = 1;
1320                 break;
1321         case JBD2_SUPERBLOCK_V2:
1322                 journal->j_format_version = 2;
1323                 break;
1324         default:
1325                 printk(KERN_WARNING "JBD: unrecognised superblock format ID\n");
1326                 goto out;
1327         }
1328
1329         if (be32_to_cpu(sb->s_maxlen) < journal->j_maxlen)
1330                 journal->j_maxlen = be32_to_cpu(sb->s_maxlen);
1331         else if (be32_to_cpu(sb->s_maxlen) > journal->j_maxlen) {
1332                 printk (KERN_WARNING "JBD: journal file too short\n");
1333                 goto out;
1334         }
1335
1336         return 0;
1337
1338 out:
1339         journal_fail_superblock(journal);
1340         return err;
1341 }
1342
1343 /*
1344  * Load the on-disk journal superblock and read the key fields into the
1345  * journal_t.
1346  */
1347
1348 static int load_superblock(journal_t *journal)
1349 {
1350         int err;
1351         journal_superblock_t *sb;
1352
1353         err = journal_get_superblock(journal);
1354         if (err)
1355                 return err;
1356
1357         sb = journal->j_superblock;
1358
1359         journal->j_tail_sequence = be32_to_cpu(sb->s_sequence);
1360         journal->j_tail = be32_to_cpu(sb->s_start);
1361         journal->j_first = be32_to_cpu(sb->s_first);
1362         journal->j_last = be32_to_cpu(sb->s_maxlen);
1363         journal->j_errno = be32_to_cpu(sb->s_errno);
1364
1365         return 0;
1366 }
1367
1368
1369 /**
1370  * int jbd2_journal_load() - Read journal from disk.
1371  * @journal: Journal to act on.
1372  *
1373  * Given a journal_t structure which tells us which disk blocks contain
1374  * a journal, read the journal from disk to initialise the in-memory
1375  * structures.
1376  */
1377 int jbd2_journal_load(journal_t *journal)
1378 {
1379         int err;
1380         journal_superblock_t *sb;
1381
1382         err = load_superblock(journal);
1383         if (err)
1384                 return err;
1385
1386         sb = journal->j_superblock;
1387         /* If this is a V2 superblock, then we have to check the
1388          * features flags on it. */
1389
1390         if (journal->j_format_version >= 2) {
1391                 if ((sb->s_feature_ro_compat &
1392                      ~cpu_to_be32(JBD2_KNOWN_ROCOMPAT_FEATURES)) ||
1393                     (sb->s_feature_incompat &
1394                      ~cpu_to_be32(JBD2_KNOWN_INCOMPAT_FEATURES))) {
1395                         printk (KERN_WARNING
1396                                 "JBD: Unrecognised features on journal\n");
1397                         return -EINVAL;
1398                 }
1399         }
1400
1401         /* Let the recovery code check whether it needs to recover any
1402          * data from the journal. */
1403         if (jbd2_journal_recover(journal))
1404                 goto recovery_error;
1405
1406         /* OK, we've finished with the dynamic journal bits:
1407          * reinitialise the dynamic contents of the superblock in memory
1408          * and reset them on disk. */
1409         if (journal_reset(journal))
1410                 goto recovery_error;
1411
1412         journal->j_flags &= ~JBD2_ABORT;
1413         journal->j_flags |= JBD2_LOADED;
1414         return 0;
1415
1416 recovery_error:
1417         printk (KERN_WARNING "JBD: recovery failed\n");
1418         return -EIO;
1419 }
1420
1421 /**
1422  * void jbd2_journal_destroy() - Release a journal_t structure.
1423  * @journal: Journal to act on.
1424  *
1425  * Release a journal_t structure once it is no longer in use by the
1426  * journaled object.
1427  */
1428 void jbd2_journal_destroy(journal_t *journal)
1429 {
1430         /* Wait for the commit thread to wake up and die. */
1431         journal_kill_thread(journal);
1432
1433         /* Force a final log commit */
1434         if (journal->j_running_transaction)
1435                 jbd2_journal_commit_transaction(journal);
1436
1437         /* Force any old transactions to disk */
1438
1439         /* Totally anal locking here... */
1440         spin_lock(&journal->j_list_lock);
1441         while (journal->j_checkpoint_transactions != NULL) {
1442                 spin_unlock(&journal->j_list_lock);
1443                 jbd2_log_do_checkpoint(journal);
1444                 spin_lock(&journal->j_list_lock);
1445         }
1446
1447         J_ASSERT(journal->j_running_transaction == NULL);
1448         J_ASSERT(journal->j_committing_transaction == NULL);
1449         J_ASSERT(journal->j_checkpoint_transactions == NULL);
1450         spin_unlock(&journal->j_list_lock);
1451
1452         /* We can now mark the journal as empty. */
1453         journal->j_tail = 0;
1454         journal->j_tail_sequence = ++journal->j_transaction_sequence;
1455         if (journal->j_sb_buffer) {
1456                 jbd2_journal_update_superblock(journal, 1);
1457                 brelse(journal->j_sb_buffer);
1458         }
1459
1460         if (journal->j_proc_entry)
1461                 jbd2_stats_proc_exit(journal);
1462         if (journal->j_inode)
1463                 iput(journal->j_inode);
1464         if (journal->j_revoke)
1465                 jbd2_journal_destroy_revoke(journal);
1466         kfree(journal->j_wbuf);
1467         kfree(journal);
1468 }
1469
1470
1471 /**
1472  *int jbd2_journal_check_used_features () - Check if features specified are used.
1473  * @journal: Journal to check.
1474  * @compat: bitmask of compatible features
1475  * @ro: bitmask of features that force read-only mount
1476  * @incompat: bitmask of incompatible features
1477  *
1478  * Check whether the journal uses all of a given set of
1479  * features.  Return true (non-zero) if it does.
1480  **/
1481
1482 int jbd2_journal_check_used_features (journal_t *journal, unsigned long compat,
1483                                  unsigned long ro, unsigned long incompat)
1484 {
1485         journal_superblock_t *sb;
1486
1487         if (!compat && !ro && !incompat)
1488                 return 1;
1489         if (journal->j_format_version == 1)
1490                 return 0;
1491
1492         sb = journal->j_superblock;
1493
1494         if (((be32_to_cpu(sb->s_feature_compat) & compat) == compat) &&
1495             ((be32_to_cpu(sb->s_feature_ro_compat) & ro) == ro) &&
1496             ((be32_to_cpu(sb->s_feature_incompat) & incompat) == incompat))
1497                 return 1;
1498
1499         return 0;
1500 }
1501
1502 /**
1503  * int jbd2_journal_check_available_features() - Check feature set in journalling layer
1504  * @journal: Journal to check.
1505  * @compat: bitmask of compatible features
1506  * @ro: bitmask of features that force read-only mount
1507  * @incompat: bitmask of incompatible features
1508  *
1509  * Check whether the journaling code supports the use of
1510  * all of a given set of features on this journal.  Return true
1511  * (non-zero) if it can. */
1512
1513 int jbd2_journal_check_available_features (journal_t *journal, unsigned long compat,
1514                                       unsigned long ro, unsigned long incompat)
1515 {
1516         journal_superblock_t *sb;
1517
1518         if (!compat && !ro && !incompat)
1519                 return 1;
1520
1521         sb = journal->j_superblock;
1522
1523         /* We can support any known requested features iff the
1524          * superblock is in version 2.  Otherwise we fail to support any
1525          * extended sb features. */
1526
1527         if (journal->j_format_version != 2)
1528                 return 0;
1529
1530         if ((compat   & JBD2_KNOWN_COMPAT_FEATURES) == compat &&
1531             (ro       & JBD2_KNOWN_ROCOMPAT_FEATURES) == ro &&
1532             (incompat & JBD2_KNOWN_INCOMPAT_FEATURES) == incompat)
1533                 return 1;
1534
1535         return 0;
1536 }
1537
1538 /**
1539  * int jbd2_journal_set_features () - Mark a given journal feature in the superblock
1540  * @journal: Journal to act on.
1541  * @compat: bitmask of compatible features
1542  * @ro: bitmask of features that force read-only mount
1543  * @incompat: bitmask of incompatible features
1544  *
1545  * Mark a given journal feature as present on the
1546  * superblock.  Returns true if the requested features could be set.
1547  *
1548  */
1549
1550 int jbd2_journal_set_features (journal_t *journal, unsigned long compat,
1551                           unsigned long ro, unsigned long incompat)
1552 {
1553         journal_superblock_t *sb;
1554
1555         if (jbd2_journal_check_used_features(journal, compat, ro, incompat))
1556                 return 1;
1557
1558         if (!jbd2_journal_check_available_features(journal, compat, ro, incompat))
1559                 return 0;
1560
1561         jbd_debug(1, "Setting new features 0x%lx/0x%lx/0x%lx\n",
1562                   compat, ro, incompat);
1563
1564         sb = journal->j_superblock;
1565
1566         sb->s_feature_compat    |= cpu_to_be32(compat);
1567         sb->s_feature_ro_compat |= cpu_to_be32(ro);
1568         sb->s_feature_incompat  |= cpu_to_be32(incompat);
1569
1570         return 1;
1571 }
1572
1573 /*
1574  * jbd2_journal_clear_features () - Clear a given journal feature in the
1575  *                                  superblock
1576  * @journal: Journal to act on.
1577  * @compat: bitmask of compatible features
1578  * @ro: bitmask of features that force read-only mount
1579  * @incompat: bitmask of incompatible features
1580  *
1581  * Clear a given journal feature as present on the
1582  * superblock.
1583  */
1584 void jbd2_journal_clear_features(journal_t *journal, unsigned long compat,
1585                                 unsigned long ro, unsigned long incompat)
1586 {
1587         journal_superblock_t *sb;
1588
1589         jbd_debug(1, "Clear features 0x%lx/0x%lx/0x%lx\n",
1590                   compat, ro, incompat);
1591
1592         sb = journal->j_superblock;
1593
1594         sb->s_feature_compat    &= ~cpu_to_be32(compat);
1595         sb->s_feature_ro_compat &= ~cpu_to_be32(ro);
1596         sb->s_feature_incompat  &= ~cpu_to_be32(incompat);
1597 }
1598 EXPORT_SYMBOL(jbd2_journal_clear_features);
1599
1600 /**
1601  * int jbd2_journal_update_format () - Update on-disk journal structure.
1602  * @journal: Journal to act on.
1603  *
1604  * Given an initialised but unloaded journal struct, poke about in the
1605  * on-disk structure to update it to the most recent supported version.
1606  */
1607 int jbd2_journal_update_format (journal_t *journal)
1608 {
1609         journal_superblock_t *sb;
1610         int err;
1611
1612         err = journal_get_superblock(journal);
1613         if (err)
1614                 return err;
1615
1616         sb = journal->j_superblock;
1617
1618         switch (be32_to_cpu(sb->s_header.h_blocktype)) {
1619         case JBD2_SUPERBLOCK_V2:
1620                 return 0;
1621         case JBD2_SUPERBLOCK_V1:
1622                 return journal_convert_superblock_v1(journal, sb);
1623         default:
1624                 break;
1625         }
1626         return -EINVAL;
1627 }
1628
1629 static int journal_convert_superblock_v1(journal_t *journal,
1630                                          journal_superblock_t *sb)
1631 {
1632         int offset, blocksize;
1633         struct buffer_head *bh;
1634
1635         printk(KERN_WARNING
1636                 "JBD: Converting superblock from version 1 to 2.\n");
1637
1638         /* Pre-initialise new fields to zero */
1639         offset = ((char *) &(sb->s_feature_compat)) - ((char *) sb);
1640         blocksize = be32_to_cpu(sb->s_blocksize);
1641         memset(&sb->s_feature_compat, 0, blocksize-offset);
1642
1643         sb->s_nr_users = cpu_to_be32(1);
1644         sb->s_header.h_blocktype = cpu_to_be32(JBD2_SUPERBLOCK_V2);
1645         journal->j_format_version = 2;
1646
1647         bh = journal->j_sb_buffer;
1648         BUFFER_TRACE(bh, "marking dirty");
1649         mark_buffer_dirty(bh);
1650         sync_dirty_buffer(bh);
1651         return 0;
1652 }
1653
1654
1655 /**
1656  * int jbd2_journal_flush () - Flush journal
1657  * @journal: Journal to act on.
1658  *
1659  * Flush all data for a given journal to disk and empty the journal.
1660  * Filesystems can use this when remounting readonly to ensure that
1661  * recovery does not need to happen on remount.
1662  */
1663
1664 int jbd2_journal_flush(journal_t *journal)
1665 {
1666         int err = 0;
1667         transaction_t *transaction = NULL;
1668         unsigned long old_tail;
1669
1670         spin_lock(&journal->j_state_lock);
1671
1672         /* Force everything buffered to the log... */
1673         if (journal->j_running_transaction) {
1674                 transaction = journal->j_running_transaction;
1675                 __jbd2_log_start_commit(journal, transaction->t_tid);
1676         } else if (journal->j_committing_transaction)
1677                 transaction = journal->j_committing_transaction;
1678
1679         /* Wait for the log commit to complete... */
1680         if (transaction) {
1681                 tid_t tid = transaction->t_tid;
1682
1683                 spin_unlock(&journal->j_state_lock);
1684                 jbd2_log_wait_commit(journal, tid);
1685         } else {
1686                 spin_unlock(&journal->j_state_lock);
1687         }
1688
1689         /* ...and flush everything in the log out to disk. */
1690         spin_lock(&journal->j_list_lock);
1691         while (!err && journal->j_checkpoint_transactions != NULL) {
1692                 spin_unlock(&journal->j_list_lock);
1693                 err = jbd2_log_do_checkpoint(journal);
1694                 spin_lock(&journal->j_list_lock);
1695         }
1696         spin_unlock(&journal->j_list_lock);
1697         jbd2_cleanup_journal_tail(journal);
1698
1699         /* Finally, mark the journal as really needing no recovery.
1700          * This sets s_start==0 in the underlying superblock, which is
1701          * the magic code for a fully-recovered superblock.  Any future
1702          * commits of data to the journal will restore the current
1703          * s_start value. */
1704         spin_lock(&journal->j_state_lock);
1705         old_tail = journal->j_tail;
1706         journal->j_tail = 0;
1707         spin_unlock(&journal->j_state_lock);
1708         jbd2_journal_update_superblock(journal, 1);
1709         spin_lock(&journal->j_state_lock);
1710         journal->j_tail = old_tail;
1711
1712         J_ASSERT(!journal->j_running_transaction);
1713         J_ASSERT(!journal->j_committing_transaction);
1714         J_ASSERT(!journal->j_checkpoint_transactions);
1715         J_ASSERT(journal->j_head == journal->j_tail);
1716         J_ASSERT(journal->j_tail_sequence == journal->j_transaction_sequence);
1717         spin_unlock(&journal->j_state_lock);
1718         return err;
1719 }
1720
1721 /**
1722  * int jbd2_journal_wipe() - Wipe journal contents
1723  * @journal: Journal to act on.
1724  * @write: flag (see below)
1725  *
1726  * Wipe out all of the contents of a journal, safely.  This will produce
1727  * a warning if the journal contains any valid recovery information.
1728  * Must be called between journal_init_*() and jbd2_journal_load().
1729  *
1730  * If 'write' is non-zero, then we wipe out the journal on disk; otherwise
1731  * we merely suppress recovery.
1732  */
1733
1734 int jbd2_journal_wipe(journal_t *journal, int write)
1735 {
1736         journal_superblock_t *sb;
1737         int err = 0;
1738
1739         J_ASSERT (!(journal->j_flags & JBD2_LOADED));
1740
1741         err = load_superblock(journal);
1742         if (err)
1743                 return err;
1744
1745         sb = journal->j_superblock;
1746
1747         if (!journal->j_tail)
1748                 goto no_recovery;
1749
1750         printk (KERN_WARNING "JBD: %s recovery information on journal\n",
1751                 write ? "Clearing" : "Ignoring");
1752
1753         err = jbd2_journal_skip_recovery(journal);
1754         if (write)
1755                 jbd2_journal_update_superblock(journal, 1);
1756
1757  no_recovery:
1758         return err;
1759 }
1760
1761 /*
1762  * journal_dev_name: format a character string to describe on what
1763  * device this journal is present.
1764  */
1765
1766 static const char *journal_dev_name(journal_t *journal, char *buffer)
1767 {
1768         struct block_device *bdev;
1769
1770         if (journal->j_inode)
1771                 bdev = journal->j_inode->i_sb->s_bdev;
1772         else
1773                 bdev = journal->j_dev;
1774
1775         return bdevname(bdev, buffer);
1776 }
1777
1778 /*
1779  * Journal abort has very specific semantics, which we describe
1780  * for journal abort.
1781  *
1782  * Two internal function, which provide abort to te jbd layer
1783  * itself are here.
1784  */
1785
1786 /*
1787  * Quick version for internal journal use (doesn't lock the journal).
1788  * Aborts hard --- we mark the abort as occurred, but do _nothing_ else,
1789  * and don't attempt to make any other journal updates.
1790  */
1791 void __jbd2_journal_abort_hard(journal_t *journal)
1792 {
1793         transaction_t *transaction;
1794         char b[BDEVNAME_SIZE];
1795
1796         if (journal->j_flags & JBD2_ABORT)
1797                 return;
1798
1799         printk(KERN_ERR "Aborting journal on device %s.\n",
1800                 journal_dev_name(journal, b));
1801
1802         spin_lock(&journal->j_state_lock);
1803         journal->j_flags |= JBD2_ABORT;
1804         transaction = journal->j_running_transaction;
1805         if (transaction)
1806                 __jbd2_log_start_commit(journal, transaction->t_tid);
1807         spin_unlock(&journal->j_state_lock);
1808 }
1809
1810 /* Soft abort: record the abort error status in the journal superblock,
1811  * but don't do any other IO. */
1812 static void __journal_abort_soft (journal_t *journal, int errno)
1813 {
1814         if (journal->j_flags & JBD2_ABORT)
1815                 return;
1816
1817         if (!journal->j_errno)
1818                 journal->j_errno = errno;
1819
1820         __jbd2_journal_abort_hard(journal);
1821
1822         if (errno)
1823                 jbd2_journal_update_superblock(journal, 1);
1824 }
1825
1826 /**
1827  * void jbd2_journal_abort () - Shutdown the journal immediately.
1828  * @journal: the journal to shutdown.
1829  * @errno:   an error number to record in the journal indicating
1830  *           the reason for the shutdown.
1831  *
1832  * Perform a complete, immediate shutdown of the ENTIRE
1833  * journal (not of a single transaction).  This operation cannot be
1834  * undone without closing and reopening the journal.
1835  *
1836  * The jbd2_journal_abort function is intended to support higher level error
1837  * recovery mechanisms such as the ext2/ext3 remount-readonly error
1838  * mode.
1839  *
1840  * Journal abort has very specific semantics.  Any existing dirty,
1841  * unjournaled buffers in the main filesystem will still be written to
1842  * disk by bdflush, but the journaling mechanism will be suspended
1843  * immediately and no further transaction commits will be honoured.
1844  *
1845  * Any dirty, journaled buffers will be written back to disk without
1846  * hitting the journal.  Atomicity cannot be guaranteed on an aborted
1847  * filesystem, but we _do_ attempt to leave as much data as possible
1848  * behind for fsck to use for cleanup.
1849  *
1850  * Any attempt to get a new transaction handle on a journal which is in
1851  * ABORT state will just result in an -EROFS error return.  A
1852  * jbd2_journal_stop on an existing handle will return -EIO if we have
1853  * entered abort state during the update.
1854  *
1855  * Recursive transactions are not disturbed by journal abort until the
1856  * final jbd2_journal_stop, which will receive the -EIO error.
1857  *
1858  * Finally, the jbd2_journal_abort call allows the caller to supply an errno
1859  * which will be recorded (if possible) in the journal superblock.  This
1860  * allows a client to record failure conditions in the middle of a
1861  * transaction without having to complete the transaction to record the
1862  * failure to disk.  ext3_error, for example, now uses this
1863  * functionality.
1864  *
1865  * Errors which originate from within the journaling layer will NOT
1866  * supply an errno; a null errno implies that absolutely no further
1867  * writes are done to the journal (unless there are any already in
1868  * progress).
1869  *
1870  */
1871
1872 void jbd2_journal_abort(journal_t *journal, int errno)
1873 {
1874         __journal_abort_soft(journal, errno);
1875 }
1876
1877 /**
1878  * int jbd2_journal_errno () - returns the journal's error state.
1879  * @journal: journal to examine.
1880  *
1881  * This is the errno numbet set with jbd2_journal_abort(), the last
1882  * time the journal was mounted - if the journal was stopped
1883  * without calling abort this will be 0.
1884  *
1885  * If the journal has been aborted on this mount time -EROFS will
1886  * be returned.
1887  */
1888 int jbd2_journal_errno(journal_t *journal)
1889 {
1890         int err;
1891
1892         spin_lock(&journal->j_state_lock);
1893         if (journal->j_flags & JBD2_ABORT)
1894                 err = -EROFS;
1895         else
1896                 err = journal->j_errno;
1897         spin_unlock(&journal->j_state_lock);
1898         return err;
1899 }
1900
1901 /**
1902  * int jbd2_journal_clear_err () - clears the journal's error state
1903  * @journal: journal to act on.
1904  *
1905  * An error must be cleared or Acked to take a FS out of readonly
1906  * mode.
1907  */
1908 int jbd2_journal_clear_err(journal_t *journal)
1909 {
1910         int err = 0;
1911
1912         spin_lock(&journal->j_state_lock);
1913         if (journal->j_flags & JBD2_ABORT)
1914                 err = -EROFS;
1915         else
1916                 journal->j_errno = 0;
1917         spin_unlock(&journal->j_state_lock);
1918         return err;
1919 }
1920
1921 /**
1922  * void jbd2_journal_ack_err() - Ack journal err.
1923  * @journal: journal to act on.
1924  *
1925  * An error must be cleared or Acked to take a FS out of readonly
1926  * mode.
1927  */
1928 void jbd2_journal_ack_err(journal_t *journal)
1929 {
1930         spin_lock(&journal->j_state_lock);
1931         if (journal->j_errno)
1932                 journal->j_flags |= JBD2_ACK_ERR;
1933         spin_unlock(&journal->j_state_lock);
1934 }
1935
1936 int jbd2_journal_blocks_per_page(struct inode *inode)
1937 {
1938         return 1 << (PAGE_CACHE_SHIFT - inode->i_sb->s_blocksize_bits);
1939 }
1940
1941 /*
1942  * helper functions to deal with 32 or 64bit block numbers.
1943  */
1944 size_t journal_tag_bytes(journal_t *journal)
1945 {
1946         if (JBD2_HAS_INCOMPAT_FEATURE(journal, JBD2_FEATURE_INCOMPAT_64BIT))
1947                 return JBD2_TAG_SIZE64;
1948         else
1949                 return JBD2_TAG_SIZE32;
1950 }
1951
1952 /*
1953  * Journal_head storage management
1954  */
1955 static struct kmem_cache *jbd2_journal_head_cache;
1956 #ifdef CONFIG_JBD2_DEBUG
1957 static atomic_t nr_journal_heads = ATOMIC_INIT(0);
1958 #endif
1959
1960 static int journal_init_jbd2_journal_head_cache(void)
1961 {
1962         int retval;
1963
1964         J_ASSERT(jbd2_journal_head_cache == NULL);
1965         jbd2_journal_head_cache = kmem_cache_create("jbd2_journal_head",
1966                                 sizeof(struct journal_head),
1967                                 0,              /* offset */
1968                                 SLAB_TEMPORARY, /* flags */
1969                                 NULL);          /* ctor */
1970         retval = 0;
1971         if (!jbd2_journal_head_cache) {
1972                 retval = -ENOMEM;
1973                 printk(KERN_EMERG "JBD: no memory for journal_head cache\n");
1974         }
1975         return retval;
1976 }
1977
1978 static void jbd2_journal_destroy_jbd2_journal_head_cache(void)
1979 {
1980         if (jbd2_journal_head_cache) {
1981                 kmem_cache_destroy(jbd2_journal_head_cache);
1982                 jbd2_journal_head_cache = NULL;
1983         }
1984 }
1985
1986 /*
1987  * journal_head splicing and dicing
1988  */
1989 static struct journal_head *journal_alloc_journal_head(void)
1990 {
1991         struct journal_head *ret;
1992         static unsigned long last_warning;
1993
1994 #ifdef CONFIG_JBD2_DEBUG
1995         atomic_inc(&nr_journal_heads);
1996 #endif
1997         ret = kmem_cache_alloc(jbd2_journal_head_cache, GFP_NOFS);
1998         if (!ret) {
1999                 jbd_debug(1, "out of memory for journal_head\n");
2000                 if (time_after(jiffies, last_warning + 5*HZ)) {
2001                         printk(KERN_NOTICE "ENOMEM in %s, retrying.\n",
2002                                __func__);
2003                         last_warning = jiffies;
2004                 }
2005                 while (!ret) {
2006                         yield();
2007                         ret = kmem_cache_alloc(jbd2_journal_head_cache, GFP_NOFS);
2008                 }
2009         }
2010         return ret;
2011 }
2012
2013 static void journal_free_journal_head(struct journal_head *jh)
2014 {
2015 #ifdef CONFIG_JBD2_DEBUG
2016         atomic_dec(&nr_journal_heads);
2017         memset(jh, JBD2_POISON_FREE, sizeof(*jh));
2018 #endif
2019         kmem_cache_free(jbd2_journal_head_cache, jh);
2020 }
2021
2022 /*
2023  * A journal_head is attached to a buffer_head whenever JBD has an
2024  * interest in the buffer.
2025  *
2026  * Whenever a buffer has an attached journal_head, its ->b_state:BH_JBD bit
2027  * is set.  This bit is tested in core kernel code where we need to take
2028  * JBD-specific actions.  Testing the zeroness of ->b_private is not reliable
2029  * there.
2030  *
2031  * When a buffer has its BH_JBD bit set, its ->b_count is elevated by one.
2032  *
2033  * When a buffer has its BH_JBD bit set it is immune from being released by
2034  * core kernel code, mainly via ->b_count.
2035  *
2036  * A journal_head may be detached from its buffer_head when the journal_head's
2037  * b_transaction, b_cp_transaction and b_next_transaction pointers are NULL.
2038  * Various places in JBD call jbd2_journal_remove_journal_head() to indicate that the
2039  * journal_head can be dropped if needed.
2040  *
2041  * Various places in the kernel want to attach a journal_head to a buffer_head
2042  * _before_ attaching the journal_head to a transaction.  To protect the
2043  * journal_head in this situation, jbd2_journal_add_journal_head elevates the
2044  * journal_head's b_jcount refcount by one.  The caller must call
2045  * jbd2_journal_put_journal_head() to undo this.
2046  *
2047  * So the typical usage would be:
2048  *
2049  *      (Attach a journal_head if needed.  Increments b_jcount)
2050  *      struct journal_head *jh = jbd2_journal_add_journal_head(bh);
2051  *      ...
2052  *      jh->b_transaction = xxx;
2053  *      jbd2_journal_put_journal_head(jh);
2054  *
2055  * Now, the journal_head's b_jcount is zero, but it is safe from being released
2056  * because it has a non-zero b_transaction.
2057  */
2058
2059 /*
2060  * Give a buffer_head a journal_head.
2061  *
2062  * Doesn't need the journal lock.
2063  * May sleep.
2064  */
2065 struct journal_head *jbd2_journal_add_journal_head(struct buffer_head *bh)
2066 {
2067         struct journal_head *jh;
2068         struct journal_head *new_jh = NULL;
2069
2070 repeat:
2071         if (!buffer_jbd(bh)) {
2072                 new_jh = journal_alloc_journal_head();
2073                 memset(new_jh, 0, sizeof(*new_jh));
2074         }
2075
2076         jbd_lock_bh_journal_head(bh);
2077         if (buffer_jbd(bh)) {
2078                 jh = bh2jh(bh);
2079         } else {
2080                 J_ASSERT_BH(bh,
2081                         (atomic_read(&bh->b_count) > 0) ||
2082                         (bh->b_page && bh->b_page->mapping));
2083
2084                 if (!new_jh) {
2085                         jbd_unlock_bh_journal_head(bh);
2086                         goto repeat;
2087                 }
2088
2089                 jh = new_jh;
2090                 new_jh = NULL;          /* We consumed it */
2091                 set_buffer_jbd(bh);
2092                 bh->b_private = jh;
2093                 jh->b_bh = bh;
2094                 get_bh(bh);
2095                 BUFFER_TRACE(bh, "added journal_head");
2096         }
2097         jh->b_jcount++;
2098         jbd_unlock_bh_journal_head(bh);
2099         if (new_jh)
2100                 journal_free_journal_head(new_jh);
2101         return bh->b_private;
2102 }
2103
2104 /*
2105  * Grab a ref against this buffer_head's journal_head.  If it ended up not
2106  * having a journal_head, return NULL
2107  */
2108 struct journal_head *jbd2_journal_grab_journal_head(struct buffer_head *bh)
2109 {
2110         struct journal_head *jh = NULL;
2111
2112         jbd_lock_bh_journal_head(bh);
2113         if (buffer_jbd(bh)) {
2114                 jh = bh2jh(bh);
2115                 jh->b_jcount++;
2116         }
2117         jbd_unlock_bh_journal_head(bh);
2118         return jh;
2119 }
2120
2121 static void __journal_remove_journal_head(struct buffer_head *bh)
2122 {
2123         struct journal_head *jh = bh2jh(bh);
2124
2125         J_ASSERT_JH(jh, jh->b_jcount >= 0);
2126
2127         get_bh(bh);
2128         if (jh->b_jcount == 0) {
2129                 if (jh->b_transaction == NULL &&
2130                                 jh->b_next_transaction == NULL &&
2131                                 jh->b_cp_transaction == NULL) {
2132                         J_ASSERT_JH(jh, jh->b_jlist == BJ_None);
2133                         J_ASSERT_BH(bh, buffer_jbd(bh));
2134                         J_ASSERT_BH(bh, jh2bh(jh) == bh);
2135                         BUFFER_TRACE(bh, "remove journal_head");
2136                         if (jh->b_frozen_data) {
2137                                 printk(KERN_WARNING "%s: freeing "
2138                                                 "b_frozen_data\n",
2139                                                 __func__);
2140                                 jbd2_free(jh->b_frozen_data, bh->b_size);
2141                         }
2142                         if (jh->b_committed_data) {
2143                                 printk(KERN_WARNING "%s: freeing "
2144                                                 "b_committed_data\n",
2145                                                 __func__);
2146                                 jbd2_free(jh->b_committed_data, bh->b_size);
2147                         }
2148                         bh->b_private = NULL;
2149                         jh->b_bh = NULL;        /* debug, really */
2150                         clear_buffer_jbd(bh);
2151                         __brelse(bh);
2152                         journal_free_journal_head(jh);
2153                 } else {
2154                         BUFFER_TRACE(bh, "journal_head was locked");
2155                 }
2156         }
2157 }
2158
2159 /*
2160  * jbd2_journal_remove_journal_head(): if the buffer isn't attached to a transaction
2161  * and has a zero b_jcount then remove and release its journal_head.   If we did
2162  * see that the buffer is not used by any transaction we also "logically"
2163  * decrement ->b_count.
2164  *
2165  * We in fact take an additional increment on ->b_count as a convenience,
2166  * because the caller usually wants to do additional things with the bh
2167  * after calling here.
2168  * The caller of jbd2_journal_remove_journal_head() *must* run __brelse(bh) at some
2169  * time.  Once the caller has run __brelse(), the buffer is eligible for
2170  * reaping by try_to_free_buffers().
2171  */
2172 void jbd2_journal_remove_journal_head(struct buffer_head *bh)
2173 {
2174         jbd_lock_bh_journal_head(bh);
2175         __journal_remove_journal_head(bh);
2176         jbd_unlock_bh_journal_head(bh);
2177 }
2178
2179 /*
2180  * Drop a reference on the passed journal_head.  If it fell to zero then try to
2181  * release the journal_head from the buffer_head.
2182  */
2183 void jbd2_journal_put_journal_head(struct journal_head *jh)
2184 {
2185         struct buffer_head *bh = jh2bh(jh);
2186
2187         jbd_lock_bh_journal_head(bh);
2188         J_ASSERT_JH(jh, jh->b_jcount > 0);
2189         --jh->b_jcount;
2190         if (!jh->b_jcount && !jh->b_transaction) {
2191                 __journal_remove_journal_head(bh);
2192                 __brelse(bh);
2193         }
2194         jbd_unlock_bh_journal_head(bh);
2195 }
2196
2197 /*
2198  * debugfs tunables
2199  */
2200 #ifdef CONFIG_JBD2_DEBUG
2201 u8 jbd2_journal_enable_debug __read_mostly;
2202 EXPORT_SYMBOL(jbd2_journal_enable_debug);
2203
2204 #define JBD2_DEBUG_NAME "jbd2-debug"
2205
2206 static struct dentry *jbd2_debugfs_dir;
2207 static struct dentry *jbd2_debug;
2208
2209 static void __init jbd2_create_debugfs_entry(void)
2210 {
2211         jbd2_debugfs_dir = debugfs_create_dir("jbd2", NULL);
2212         if (jbd2_debugfs_dir)
2213                 jbd2_debug = debugfs_create_u8(JBD2_DEBUG_NAME, S_IRUGO,
2214                                                jbd2_debugfs_dir,
2215                                                &jbd2_journal_enable_debug);
2216 }
2217
2218 static void __exit jbd2_remove_debugfs_entry(void)
2219 {
2220         debugfs_remove(jbd2_debug);
2221         debugfs_remove(jbd2_debugfs_dir);
2222 }
2223
2224 #else
2225
2226 static void __init jbd2_create_debugfs_entry(void)
2227 {
2228 }
2229
2230 static void __exit jbd2_remove_debugfs_entry(void)
2231 {
2232 }
2233
2234 #endif
2235
2236 #ifdef CONFIG_PROC_FS
2237
2238 #define JBD2_STATS_PROC_NAME "fs/jbd2"
2239
2240 static void __init jbd2_create_jbd_stats_proc_entry(void)
2241 {
2242         proc_jbd2_stats = proc_mkdir(JBD2_STATS_PROC_NAME, NULL);
2243 }
2244
2245 static void __exit jbd2_remove_jbd_stats_proc_entry(void)
2246 {
2247         if (proc_jbd2_stats)
2248                 remove_proc_entry(JBD2_STATS_PROC_NAME, NULL);
2249 }
2250
2251 #else
2252
2253 #define jbd2_create_jbd_stats_proc_entry() do {} while (0)
2254 #define jbd2_remove_jbd_stats_proc_entry() do {} while (0)
2255
2256 #endif
2257
2258 struct kmem_cache *jbd2_handle_cache;
2259
2260 static int __init journal_init_handle_cache(void)
2261 {
2262         jbd2_handle_cache = kmem_cache_create("jbd2_journal_handle",
2263                                 sizeof(handle_t),
2264                                 0,              /* offset */
2265                                 SLAB_TEMPORARY, /* flags */
2266                                 NULL);          /* ctor */
2267         if (jbd2_handle_cache == NULL) {
2268                 printk(KERN_EMERG "JBD: failed to create handle cache\n");
2269                 return -ENOMEM;
2270         }
2271         return 0;
2272 }
2273
2274 static void jbd2_journal_destroy_handle_cache(void)
2275 {
2276         if (jbd2_handle_cache)
2277                 kmem_cache_destroy(jbd2_handle_cache);
2278 }
2279
2280 /*
2281  * Module startup and shutdown
2282  */
2283
2284 static int __init journal_init_caches(void)
2285 {
2286         int ret;
2287
2288         ret = jbd2_journal_init_revoke_caches();
2289         if (ret == 0)
2290                 ret = journal_init_jbd2_journal_head_cache();
2291         if (ret == 0)
2292                 ret = journal_init_handle_cache();
2293         return ret;
2294 }
2295
2296 static void jbd2_journal_destroy_caches(void)
2297 {
2298         jbd2_journal_destroy_revoke_caches();
2299         jbd2_journal_destroy_jbd2_journal_head_cache();
2300         jbd2_journal_destroy_handle_cache();
2301 }
2302
2303 static int __init journal_init(void)
2304 {
2305         int ret;
2306
2307         BUILD_BUG_ON(sizeof(struct journal_superblock_s) != 1024);
2308
2309         ret = journal_init_caches();
2310         if (ret == 0) {
2311                 jbd2_create_debugfs_entry();
2312                 jbd2_create_jbd_stats_proc_entry();
2313         } else {
2314                 jbd2_journal_destroy_caches();
2315         }
2316         return ret;
2317 }
2318
2319 static void __exit journal_exit(void)
2320 {
2321 #ifdef CONFIG_JBD2_DEBUG
2322         int n = atomic_read(&nr_journal_heads);
2323         if (n)
2324                 printk(KERN_EMERG "JBD: leaked %d journal_heads!\n", n);
2325 #endif
2326         jbd2_remove_debugfs_entry();
2327         jbd2_remove_jbd_stats_proc_entry();
2328         jbd2_journal_destroy_caches();
2329 }
2330
2331 MODULE_LICENSE("GPL");
2332 module_init(journal_init);
2333 module_exit(journal_exit);
2334