jbd: jbd-debug and jbd2-debug should be writable
[linux-2.6.git] / fs / jbd2 / journal.c
1 /*
2  * linux/fs/jbd2/journal.c
3  *
4  * Written by Stephen C. Tweedie <sct@redhat.com>, 1998
5  *
6  * Copyright 1998 Red Hat corp --- All Rights Reserved
7  *
8  * This file is part of the Linux kernel and is made available under
9  * the terms of the GNU General Public License, version 2, or at your
10  * option, any later version, incorporated herein by reference.
11  *
12  * Generic filesystem journal-writing code; part of the ext2fs
13  * journaling system.
14  *
15  * This file manages journals: areas of disk reserved for logging
16  * transactional updates.  This includes the kernel journaling thread
17  * which is responsible for scheduling updates to the log.
18  *
19  * We do not actually manage the physical storage of the journal in this
20  * file: that is left to a per-journal policy function, which allows us
21  * to store the journal within a filesystem-specified area for ext2
22  * journaling (ext2 can use a reserved inode for storing the log).
23  */
24
25 #include <linux/module.h>
26 #include <linux/time.h>
27 #include <linux/fs.h>
28 #include <linux/jbd2.h>
29 #include <linux/errno.h>
30 #include <linux/slab.h>
31 #include <linux/init.h>
32 #include <linux/mm.h>
33 #include <linux/freezer.h>
34 #include <linux/pagemap.h>
35 #include <linux/kthread.h>
36 #include <linux/poison.h>
37 #include <linux/proc_fs.h>
38 #include <linux/debugfs.h>
39 #include <linux/seq_file.h>
40 #include <linux/math64.h>
41 #include <linux/hash.h>
42
43 #define CREATE_TRACE_POINTS
44 #include <trace/events/jbd2.h>
45
46 #include <asm/uaccess.h>
47 #include <asm/page.h>
48
49 EXPORT_SYMBOL(jbd2_journal_start);
50 EXPORT_SYMBOL(jbd2_journal_restart);
51 EXPORT_SYMBOL(jbd2_journal_extend);
52 EXPORT_SYMBOL(jbd2_journal_stop);
53 EXPORT_SYMBOL(jbd2_journal_lock_updates);
54 EXPORT_SYMBOL(jbd2_journal_unlock_updates);
55 EXPORT_SYMBOL(jbd2_journal_get_write_access);
56 EXPORT_SYMBOL(jbd2_journal_get_create_access);
57 EXPORT_SYMBOL(jbd2_journal_get_undo_access);
58 EXPORT_SYMBOL(jbd2_journal_set_triggers);
59 EXPORT_SYMBOL(jbd2_journal_dirty_metadata);
60 EXPORT_SYMBOL(jbd2_journal_release_buffer);
61 EXPORT_SYMBOL(jbd2_journal_forget);
62 #if 0
63 EXPORT_SYMBOL(journal_sync_buffer);
64 #endif
65 EXPORT_SYMBOL(jbd2_journal_flush);
66 EXPORT_SYMBOL(jbd2_journal_revoke);
67
68 EXPORT_SYMBOL(jbd2_journal_init_dev);
69 EXPORT_SYMBOL(jbd2_journal_init_inode);
70 EXPORT_SYMBOL(jbd2_journal_update_format);
71 EXPORT_SYMBOL(jbd2_journal_check_used_features);
72 EXPORT_SYMBOL(jbd2_journal_check_available_features);
73 EXPORT_SYMBOL(jbd2_journal_set_features);
74 EXPORT_SYMBOL(jbd2_journal_load);
75 EXPORT_SYMBOL(jbd2_journal_destroy);
76 EXPORT_SYMBOL(jbd2_journal_abort);
77 EXPORT_SYMBOL(jbd2_journal_errno);
78 EXPORT_SYMBOL(jbd2_journal_ack_err);
79 EXPORT_SYMBOL(jbd2_journal_clear_err);
80 EXPORT_SYMBOL(jbd2_log_wait_commit);
81 EXPORT_SYMBOL(jbd2_log_start_commit);
82 EXPORT_SYMBOL(jbd2_journal_start_commit);
83 EXPORT_SYMBOL(jbd2_journal_force_commit_nested);
84 EXPORT_SYMBOL(jbd2_journal_wipe);
85 EXPORT_SYMBOL(jbd2_journal_blocks_per_page);
86 EXPORT_SYMBOL(jbd2_journal_invalidatepage);
87 EXPORT_SYMBOL(jbd2_journal_try_to_free_buffers);
88 EXPORT_SYMBOL(jbd2_journal_force_commit);
89 EXPORT_SYMBOL(jbd2_journal_file_inode);
90 EXPORT_SYMBOL(jbd2_journal_init_jbd_inode);
91 EXPORT_SYMBOL(jbd2_journal_release_jbd_inode);
92 EXPORT_SYMBOL(jbd2_journal_begin_ordered_truncate);
93
94 static int journal_convert_superblock_v1(journal_t *, journal_superblock_t *);
95 static void __journal_abort_soft (journal_t *journal, int errno);
96
97 /*
98  * Helper function used to manage commit timeouts
99  */
100
101 static void commit_timeout(unsigned long __data)
102 {
103         struct task_struct * p = (struct task_struct *) __data;
104
105         wake_up_process(p);
106 }
107
108 /*
109  * kjournald2: The main thread function used to manage a logging device
110  * journal.
111  *
112  * This kernel thread is responsible for two things:
113  *
114  * 1) COMMIT:  Every so often we need to commit the current state of the
115  *    filesystem to disk.  The journal thread is responsible for writing
116  *    all of the metadata buffers to disk.
117  *
118  * 2) CHECKPOINT: We cannot reuse a used section of the log file until all
119  *    of the data in that part of the log has been rewritten elsewhere on
120  *    the disk.  Flushing these old buffers to reclaim space in the log is
121  *    known as checkpointing, and this thread is responsible for that job.
122  */
123
124 static int kjournald2(void *arg)
125 {
126         journal_t *journal = arg;
127         transaction_t *transaction;
128
129         /*
130          * Set up an interval timer which can be used to trigger a commit wakeup
131          * after the commit interval expires
132          */
133         setup_timer(&journal->j_commit_timer, commit_timeout,
134                         (unsigned long)current);
135
136         /* Record that the journal thread is running */
137         journal->j_task = current;
138         wake_up(&journal->j_wait_done_commit);
139
140         /*
141          * And now, wait forever for commit wakeup events.
142          */
143         spin_lock(&journal->j_state_lock);
144
145 loop:
146         if (journal->j_flags & JBD2_UNMOUNT)
147                 goto end_loop;
148
149         jbd_debug(1, "commit_sequence=%d, commit_request=%d\n",
150                 journal->j_commit_sequence, journal->j_commit_request);
151
152         if (journal->j_commit_sequence != journal->j_commit_request) {
153                 jbd_debug(1, "OK, requests differ\n");
154                 spin_unlock(&journal->j_state_lock);
155                 del_timer_sync(&journal->j_commit_timer);
156                 jbd2_journal_commit_transaction(journal);
157                 spin_lock(&journal->j_state_lock);
158                 goto loop;
159         }
160
161         wake_up(&journal->j_wait_done_commit);
162         if (freezing(current)) {
163                 /*
164                  * The simpler the better. Flushing journal isn't a
165                  * good idea, because that depends on threads that may
166                  * be already stopped.
167                  */
168                 jbd_debug(1, "Now suspending kjournald2\n");
169                 spin_unlock(&journal->j_state_lock);
170                 refrigerator();
171                 spin_lock(&journal->j_state_lock);
172         } else {
173                 /*
174                  * We assume on resume that commits are already there,
175                  * so we don't sleep
176                  */
177                 DEFINE_WAIT(wait);
178                 int should_sleep = 1;
179
180                 prepare_to_wait(&journal->j_wait_commit, &wait,
181                                 TASK_INTERRUPTIBLE);
182                 if (journal->j_commit_sequence != journal->j_commit_request)
183                         should_sleep = 0;
184                 transaction = journal->j_running_transaction;
185                 if (transaction && time_after_eq(jiffies,
186                                                 transaction->t_expires))
187                         should_sleep = 0;
188                 if (journal->j_flags & JBD2_UNMOUNT)
189                         should_sleep = 0;
190                 if (should_sleep) {
191                         spin_unlock(&journal->j_state_lock);
192                         schedule();
193                         spin_lock(&journal->j_state_lock);
194                 }
195                 finish_wait(&journal->j_wait_commit, &wait);
196         }
197
198         jbd_debug(1, "kjournald2 wakes\n");
199
200         /*
201          * Were we woken up by a commit wakeup event?
202          */
203         transaction = journal->j_running_transaction;
204         if (transaction && time_after_eq(jiffies, transaction->t_expires)) {
205                 journal->j_commit_request = transaction->t_tid;
206                 jbd_debug(1, "woke because of timeout\n");
207         }
208         goto loop;
209
210 end_loop:
211         spin_unlock(&journal->j_state_lock);
212         del_timer_sync(&journal->j_commit_timer);
213         journal->j_task = NULL;
214         wake_up(&journal->j_wait_done_commit);
215         jbd_debug(1, "Journal thread exiting.\n");
216         return 0;
217 }
218
219 static int jbd2_journal_start_thread(journal_t *journal)
220 {
221         struct task_struct *t;
222
223         t = kthread_run(kjournald2, journal, "jbd2/%s",
224                         journal->j_devname);
225         if (IS_ERR(t))
226                 return PTR_ERR(t);
227
228         wait_event(journal->j_wait_done_commit, journal->j_task != NULL);
229         return 0;
230 }
231
232 static void journal_kill_thread(journal_t *journal)
233 {
234         spin_lock(&journal->j_state_lock);
235         journal->j_flags |= JBD2_UNMOUNT;
236
237         while (journal->j_task) {
238                 wake_up(&journal->j_wait_commit);
239                 spin_unlock(&journal->j_state_lock);
240                 wait_event(journal->j_wait_done_commit, journal->j_task == NULL);
241                 spin_lock(&journal->j_state_lock);
242         }
243         spin_unlock(&journal->j_state_lock);
244 }
245
246 /*
247  * jbd2_journal_write_metadata_buffer: write a metadata buffer to the journal.
248  *
249  * Writes a metadata buffer to a given disk block.  The actual IO is not
250  * performed but a new buffer_head is constructed which labels the data
251  * to be written with the correct destination disk block.
252  *
253  * Any magic-number escaping which needs to be done will cause a
254  * copy-out here.  If the buffer happens to start with the
255  * JBD2_MAGIC_NUMBER, then we can't write it to the log directly: the
256  * magic number is only written to the log for descripter blocks.  In
257  * this case, we copy the data and replace the first word with 0, and we
258  * return a result code which indicates that this buffer needs to be
259  * marked as an escaped buffer in the corresponding log descriptor
260  * block.  The missing word can then be restored when the block is read
261  * during recovery.
262  *
263  * If the source buffer has already been modified by a new transaction
264  * since we took the last commit snapshot, we use the frozen copy of
265  * that data for IO.  If we end up using the existing buffer_head's data
266  * for the write, then we *have* to lock the buffer to prevent anyone
267  * else from using and possibly modifying it while the IO is in
268  * progress.
269  *
270  * The function returns a pointer to the buffer_heads to be used for IO.
271  *
272  * We assume that the journal has already been locked in this function.
273  *
274  * Return value:
275  *  <0: Error
276  * >=0: Finished OK
277  *
278  * On success:
279  * Bit 0 set == escape performed on the data
280  * Bit 1 set == buffer copy-out performed (kfree the data after IO)
281  */
282
283 int jbd2_journal_write_metadata_buffer(transaction_t *transaction,
284                                   struct journal_head  *jh_in,
285                                   struct journal_head **jh_out,
286                                   unsigned long long blocknr)
287 {
288         int need_copy_out = 0;
289         int done_copy_out = 0;
290         int do_escape = 0;
291         char *mapped_data;
292         struct buffer_head *new_bh;
293         struct journal_head *new_jh;
294         struct page *new_page;
295         unsigned int new_offset;
296         struct buffer_head *bh_in = jh2bh(jh_in);
297         struct jbd2_buffer_trigger_type *triggers;
298         journal_t *journal = transaction->t_journal;
299
300         /*
301          * The buffer really shouldn't be locked: only the current committing
302          * transaction is allowed to write it, so nobody else is allowed
303          * to do any IO.
304          *
305          * akpm: except if we're journalling data, and write() output is
306          * also part of a shared mapping, and another thread has
307          * decided to launch a writepage() against this buffer.
308          */
309         J_ASSERT_BH(bh_in, buffer_jbddirty(bh_in));
310
311         new_bh = alloc_buffer_head(GFP_NOFS|__GFP_NOFAIL);
312         /* keep subsequent assertions sane */
313         new_bh->b_state = 0;
314         init_buffer(new_bh, NULL, NULL);
315         atomic_set(&new_bh->b_count, 1);
316         new_jh = jbd2_journal_add_journal_head(new_bh); /* This sleeps */
317
318         /*
319          * If a new transaction has already done a buffer copy-out, then
320          * we use that version of the data for the commit.
321          */
322         jbd_lock_bh_state(bh_in);
323 repeat:
324         if (jh_in->b_frozen_data) {
325                 done_copy_out = 1;
326                 new_page = virt_to_page(jh_in->b_frozen_data);
327                 new_offset = offset_in_page(jh_in->b_frozen_data);
328                 triggers = jh_in->b_frozen_triggers;
329         } else {
330                 new_page = jh2bh(jh_in)->b_page;
331                 new_offset = offset_in_page(jh2bh(jh_in)->b_data);
332                 triggers = jh_in->b_triggers;
333         }
334
335         mapped_data = kmap_atomic(new_page, KM_USER0);
336         /*
337          * Fire any commit trigger.  Do this before checking for escaping,
338          * as the trigger may modify the magic offset.  If a copy-out
339          * happens afterwards, it will have the correct data in the buffer.
340          */
341         jbd2_buffer_commit_trigger(jh_in, mapped_data + new_offset,
342                                    triggers);
343
344         /*
345          * Check for escaping
346          */
347         if (*((__be32 *)(mapped_data + new_offset)) ==
348                                 cpu_to_be32(JBD2_MAGIC_NUMBER)) {
349                 need_copy_out = 1;
350                 do_escape = 1;
351         }
352         kunmap_atomic(mapped_data, KM_USER0);
353
354         /*
355          * Do we need to do a data copy?
356          */
357         if (need_copy_out && !done_copy_out) {
358                 char *tmp;
359
360                 jbd_unlock_bh_state(bh_in);
361                 tmp = jbd2_alloc(bh_in->b_size, GFP_NOFS);
362                 if (!tmp) {
363                         jbd2_journal_put_journal_head(new_jh);
364                         return -ENOMEM;
365                 }
366                 jbd_lock_bh_state(bh_in);
367                 if (jh_in->b_frozen_data) {
368                         jbd2_free(tmp, bh_in->b_size);
369                         goto repeat;
370                 }
371
372                 jh_in->b_frozen_data = tmp;
373                 mapped_data = kmap_atomic(new_page, KM_USER0);
374                 memcpy(tmp, mapped_data + new_offset, jh2bh(jh_in)->b_size);
375                 kunmap_atomic(mapped_data, KM_USER0);
376
377                 new_page = virt_to_page(tmp);
378                 new_offset = offset_in_page(tmp);
379                 done_copy_out = 1;
380
381                 /*
382                  * This isn't strictly necessary, as we're using frozen
383                  * data for the escaping, but it keeps consistency with
384                  * b_frozen_data usage.
385                  */
386                 jh_in->b_frozen_triggers = jh_in->b_triggers;
387         }
388
389         /*
390          * Did we need to do an escaping?  Now we've done all the
391          * copying, we can finally do so.
392          */
393         if (do_escape) {
394                 mapped_data = kmap_atomic(new_page, KM_USER0);
395                 *((unsigned int *)(mapped_data + new_offset)) = 0;
396                 kunmap_atomic(mapped_data, KM_USER0);
397         }
398
399         set_bh_page(new_bh, new_page, new_offset);
400         new_jh->b_transaction = NULL;
401         new_bh->b_size = jh2bh(jh_in)->b_size;
402         new_bh->b_bdev = transaction->t_journal->j_dev;
403         new_bh->b_blocknr = blocknr;
404         set_buffer_mapped(new_bh);
405         set_buffer_dirty(new_bh);
406
407         *jh_out = new_jh;
408
409         /*
410          * The to-be-written buffer needs to get moved to the io queue,
411          * and the original buffer whose contents we are shadowing or
412          * copying is moved to the transaction's shadow queue.
413          */
414         JBUFFER_TRACE(jh_in, "file as BJ_Shadow");
415         spin_lock(&journal->j_list_lock);
416         __jbd2_journal_file_buffer(jh_in, transaction, BJ_Shadow);
417         spin_unlock(&journal->j_list_lock);
418         jbd_unlock_bh_state(bh_in);
419
420         JBUFFER_TRACE(new_jh, "file as BJ_IO");
421         jbd2_journal_file_buffer(new_jh, transaction, BJ_IO);
422
423         return do_escape | (done_copy_out << 1);
424 }
425
426 /*
427  * Allocation code for the journal file.  Manage the space left in the
428  * journal, so that we can begin checkpointing when appropriate.
429  */
430
431 /*
432  * __jbd2_log_space_left: Return the number of free blocks left in the journal.
433  *
434  * Called with the journal already locked.
435  *
436  * Called under j_state_lock
437  */
438
439 int __jbd2_log_space_left(journal_t *journal)
440 {
441         int left = journal->j_free;
442
443         assert_spin_locked(&journal->j_state_lock);
444
445         /*
446          * Be pessimistic here about the number of those free blocks which
447          * might be required for log descriptor control blocks.
448          */
449
450 #define MIN_LOG_RESERVED_BLOCKS 32 /* Allow for rounding errors */
451
452         left -= MIN_LOG_RESERVED_BLOCKS;
453
454         if (left <= 0)
455                 return 0;
456         left -= (left >> 3);
457         return left;
458 }
459
460 /*
461  * Called under j_state_lock.  Returns true if a transaction commit was started.
462  */
463 int __jbd2_log_start_commit(journal_t *journal, tid_t target)
464 {
465         /*
466          * Are we already doing a recent enough commit?
467          */
468         if (!tid_geq(journal->j_commit_request, target)) {
469                 /*
470                  * We want a new commit: OK, mark the request and wakup the
471                  * commit thread.  We do _not_ do the commit ourselves.
472                  */
473
474                 journal->j_commit_request = target;
475                 jbd_debug(1, "JBD: requesting commit %d/%d\n",
476                           journal->j_commit_request,
477                           journal->j_commit_sequence);
478                 wake_up(&journal->j_wait_commit);
479                 return 1;
480         }
481         return 0;
482 }
483
484 int jbd2_log_start_commit(journal_t *journal, tid_t tid)
485 {
486         int ret;
487
488         spin_lock(&journal->j_state_lock);
489         ret = __jbd2_log_start_commit(journal, tid);
490         spin_unlock(&journal->j_state_lock);
491         return ret;
492 }
493
494 /*
495  * Force and wait upon a commit if the calling process is not within
496  * transaction.  This is used for forcing out undo-protected data which contains
497  * bitmaps, when the fs is running out of space.
498  *
499  * We can only force the running transaction if we don't have an active handle;
500  * otherwise, we will deadlock.
501  *
502  * Returns true if a transaction was started.
503  */
504 int jbd2_journal_force_commit_nested(journal_t *journal)
505 {
506         transaction_t *transaction = NULL;
507         tid_t tid;
508
509         spin_lock(&journal->j_state_lock);
510         if (journal->j_running_transaction && !current->journal_info) {
511                 transaction = journal->j_running_transaction;
512                 __jbd2_log_start_commit(journal, transaction->t_tid);
513         } else if (journal->j_committing_transaction)
514                 transaction = journal->j_committing_transaction;
515
516         if (!transaction) {
517                 spin_unlock(&journal->j_state_lock);
518                 return 0;       /* Nothing to retry */
519         }
520
521         tid = transaction->t_tid;
522         spin_unlock(&journal->j_state_lock);
523         jbd2_log_wait_commit(journal, tid);
524         return 1;
525 }
526
527 /*
528  * Start a commit of the current running transaction (if any).  Returns true
529  * if a transaction is going to be committed (or is currently already
530  * committing), and fills its tid in at *ptid
531  */
532 int jbd2_journal_start_commit(journal_t *journal, tid_t *ptid)
533 {
534         int ret = 0;
535
536         spin_lock(&journal->j_state_lock);
537         if (journal->j_running_transaction) {
538                 tid_t tid = journal->j_running_transaction->t_tid;
539
540                 __jbd2_log_start_commit(journal, tid);
541                 /* There's a running transaction and we've just made sure
542                  * it's commit has been scheduled. */
543                 if (ptid)
544                         *ptid = tid;
545                 ret = 1;
546         } else if (journal->j_committing_transaction) {
547                 /*
548                  * If ext3_write_super() recently started a commit, then we
549                  * have to wait for completion of that transaction
550                  */
551                 if (ptid)
552                         *ptid = journal->j_committing_transaction->t_tid;
553                 ret = 1;
554         }
555         spin_unlock(&journal->j_state_lock);
556         return ret;
557 }
558
559 /*
560  * Wait for a specified commit to complete.
561  * The caller may not hold the journal lock.
562  */
563 int jbd2_log_wait_commit(journal_t *journal, tid_t tid)
564 {
565         int err = 0;
566
567 #ifdef CONFIG_JBD2_DEBUG
568         spin_lock(&journal->j_state_lock);
569         if (!tid_geq(journal->j_commit_request, tid)) {
570                 printk(KERN_EMERG
571                        "%s: error: j_commit_request=%d, tid=%d\n",
572                        __func__, journal->j_commit_request, tid);
573         }
574         spin_unlock(&journal->j_state_lock);
575 #endif
576         spin_lock(&journal->j_state_lock);
577         while (tid_gt(tid, journal->j_commit_sequence)) {
578                 jbd_debug(1, "JBD: want %d, j_commit_sequence=%d\n",
579                                   tid, journal->j_commit_sequence);
580                 wake_up(&journal->j_wait_commit);
581                 spin_unlock(&journal->j_state_lock);
582                 wait_event(journal->j_wait_done_commit,
583                                 !tid_gt(tid, journal->j_commit_sequence));
584                 spin_lock(&journal->j_state_lock);
585         }
586         spin_unlock(&journal->j_state_lock);
587
588         if (unlikely(is_journal_aborted(journal))) {
589                 printk(KERN_EMERG "journal commit I/O error\n");
590                 err = -EIO;
591         }
592         return err;
593 }
594
595 /*
596  * Log buffer allocation routines:
597  */
598
599 int jbd2_journal_next_log_block(journal_t *journal, unsigned long long *retp)
600 {
601         unsigned long blocknr;
602
603         spin_lock(&journal->j_state_lock);
604         J_ASSERT(journal->j_free > 1);
605
606         blocknr = journal->j_head;
607         journal->j_head++;
608         journal->j_free--;
609         if (journal->j_head == journal->j_last)
610                 journal->j_head = journal->j_first;
611         spin_unlock(&journal->j_state_lock);
612         return jbd2_journal_bmap(journal, blocknr, retp);
613 }
614
615 /*
616  * Conversion of logical to physical block numbers for the journal
617  *
618  * On external journals the journal blocks are identity-mapped, so
619  * this is a no-op.  If needed, we can use j_blk_offset - everything is
620  * ready.
621  */
622 int jbd2_journal_bmap(journal_t *journal, unsigned long blocknr,
623                  unsigned long long *retp)
624 {
625         int err = 0;
626         unsigned long long ret;
627
628         if (journal->j_inode) {
629                 ret = bmap(journal->j_inode, blocknr);
630                 if (ret)
631                         *retp = ret;
632                 else {
633                         printk(KERN_ALERT "%s: journal block not found "
634                                         "at offset %lu on %s\n",
635                                __func__, blocknr, journal->j_devname);
636                         err = -EIO;
637                         __journal_abort_soft(journal, err);
638                 }
639         } else {
640                 *retp = blocknr; /* +journal->j_blk_offset */
641         }
642         return err;
643 }
644
645 /*
646  * We play buffer_head aliasing tricks to write data/metadata blocks to
647  * the journal without copying their contents, but for journal
648  * descriptor blocks we do need to generate bona fide buffers.
649  *
650  * After the caller of jbd2_journal_get_descriptor_buffer() has finished modifying
651  * the buffer's contents they really should run flush_dcache_page(bh->b_page).
652  * But we don't bother doing that, so there will be coherency problems with
653  * mmaps of blockdevs which hold live JBD-controlled filesystems.
654  */
655 struct journal_head *jbd2_journal_get_descriptor_buffer(journal_t *journal)
656 {
657         struct buffer_head *bh;
658         unsigned long long blocknr;
659         int err;
660
661         err = jbd2_journal_next_log_block(journal, &blocknr);
662
663         if (err)
664                 return NULL;
665
666         bh = __getblk(journal->j_dev, blocknr, journal->j_blocksize);
667         if (!bh)
668                 return NULL;
669         lock_buffer(bh);
670         memset(bh->b_data, 0, journal->j_blocksize);
671         set_buffer_uptodate(bh);
672         unlock_buffer(bh);
673         BUFFER_TRACE(bh, "return this buffer");
674         return jbd2_journal_add_journal_head(bh);
675 }
676
677 struct jbd2_stats_proc_session {
678         journal_t *journal;
679         struct transaction_stats_s *stats;
680         int start;
681         int max;
682 };
683
684 static void *jbd2_seq_info_start(struct seq_file *seq, loff_t *pos)
685 {
686         return *pos ? NULL : SEQ_START_TOKEN;
687 }
688
689 static void *jbd2_seq_info_next(struct seq_file *seq, void *v, loff_t *pos)
690 {
691         return NULL;
692 }
693
694 static int jbd2_seq_info_show(struct seq_file *seq, void *v)
695 {
696         struct jbd2_stats_proc_session *s = seq->private;
697
698         if (v != SEQ_START_TOKEN)
699                 return 0;
700         seq_printf(seq, "%lu transaction, each up to %u blocks\n",
701                         s->stats->ts_tid,
702                         s->journal->j_max_transaction_buffers);
703         if (s->stats->ts_tid == 0)
704                 return 0;
705         seq_printf(seq, "average: \n  %ums waiting for transaction\n",
706             jiffies_to_msecs(s->stats->run.rs_wait / s->stats->ts_tid));
707         seq_printf(seq, "  %ums running transaction\n",
708             jiffies_to_msecs(s->stats->run.rs_running / s->stats->ts_tid));
709         seq_printf(seq, "  %ums transaction was being locked\n",
710             jiffies_to_msecs(s->stats->run.rs_locked / s->stats->ts_tid));
711         seq_printf(seq, "  %ums flushing data (in ordered mode)\n",
712             jiffies_to_msecs(s->stats->run.rs_flushing / s->stats->ts_tid));
713         seq_printf(seq, "  %ums logging transaction\n",
714             jiffies_to_msecs(s->stats->run.rs_logging / s->stats->ts_tid));
715         seq_printf(seq, "  %lluus average transaction commit time\n",
716                    div_u64(s->journal->j_average_commit_time, 1000));
717         seq_printf(seq, "  %lu handles per transaction\n",
718             s->stats->run.rs_handle_count / s->stats->ts_tid);
719         seq_printf(seq, "  %lu blocks per transaction\n",
720             s->stats->run.rs_blocks / s->stats->ts_tid);
721         seq_printf(seq, "  %lu logged blocks per transaction\n",
722             s->stats->run.rs_blocks_logged / s->stats->ts_tid);
723         return 0;
724 }
725
726 static void jbd2_seq_info_stop(struct seq_file *seq, void *v)
727 {
728 }
729
730 static const struct seq_operations jbd2_seq_info_ops = {
731         .start  = jbd2_seq_info_start,
732         .next   = jbd2_seq_info_next,
733         .stop   = jbd2_seq_info_stop,
734         .show   = jbd2_seq_info_show,
735 };
736
737 static int jbd2_seq_info_open(struct inode *inode, struct file *file)
738 {
739         journal_t *journal = PDE(inode)->data;
740         struct jbd2_stats_proc_session *s;
741         int rc, size;
742
743         s = kmalloc(sizeof(*s), GFP_KERNEL);
744         if (s == NULL)
745                 return -ENOMEM;
746         size = sizeof(struct transaction_stats_s);
747         s->stats = kmalloc(size, GFP_KERNEL);
748         if (s->stats == NULL) {
749                 kfree(s);
750                 return -ENOMEM;
751         }
752         spin_lock(&journal->j_history_lock);
753         memcpy(s->stats, &journal->j_stats, size);
754         s->journal = journal;
755         spin_unlock(&journal->j_history_lock);
756
757         rc = seq_open(file, &jbd2_seq_info_ops);
758         if (rc == 0) {
759                 struct seq_file *m = file->private_data;
760                 m->private = s;
761         } else {
762                 kfree(s->stats);
763                 kfree(s);
764         }
765         return rc;
766
767 }
768
769 static int jbd2_seq_info_release(struct inode *inode, struct file *file)
770 {
771         struct seq_file *seq = file->private_data;
772         struct jbd2_stats_proc_session *s = seq->private;
773         kfree(s->stats);
774         kfree(s);
775         return seq_release(inode, file);
776 }
777
778 static const struct file_operations jbd2_seq_info_fops = {
779         .owner          = THIS_MODULE,
780         .open           = jbd2_seq_info_open,
781         .read           = seq_read,
782         .llseek         = seq_lseek,
783         .release        = jbd2_seq_info_release,
784 };
785
786 static struct proc_dir_entry *proc_jbd2_stats;
787
788 static void jbd2_stats_proc_init(journal_t *journal)
789 {
790         journal->j_proc_entry = proc_mkdir(journal->j_devname, proc_jbd2_stats);
791         if (journal->j_proc_entry) {
792                 proc_create_data("info", S_IRUGO, journal->j_proc_entry,
793                                  &jbd2_seq_info_fops, journal);
794         }
795 }
796
797 static void jbd2_stats_proc_exit(journal_t *journal)
798 {
799         remove_proc_entry("info", journal->j_proc_entry);
800         remove_proc_entry(journal->j_devname, proc_jbd2_stats);
801 }
802
803 /*
804  * Management for journal control blocks: functions to create and
805  * destroy journal_t structures, and to initialise and read existing
806  * journal blocks from disk.  */
807
808 /* First: create and setup a journal_t object in memory.  We initialise
809  * very few fields yet: that has to wait until we have created the
810  * journal structures from from scratch, or loaded them from disk. */
811
812 static journal_t * journal_init_common (void)
813 {
814         journal_t *journal;
815         int err;
816
817         journal = kzalloc(sizeof(*journal), GFP_KERNEL|__GFP_NOFAIL);
818         if (!journal)
819                 goto fail;
820
821         init_waitqueue_head(&journal->j_wait_transaction_locked);
822         init_waitqueue_head(&journal->j_wait_logspace);
823         init_waitqueue_head(&journal->j_wait_done_commit);
824         init_waitqueue_head(&journal->j_wait_checkpoint);
825         init_waitqueue_head(&journal->j_wait_commit);
826         init_waitqueue_head(&journal->j_wait_updates);
827         mutex_init(&journal->j_barrier);
828         mutex_init(&journal->j_checkpoint_mutex);
829         spin_lock_init(&journal->j_revoke_lock);
830         spin_lock_init(&journal->j_list_lock);
831         spin_lock_init(&journal->j_state_lock);
832
833         journal->j_commit_interval = (HZ * JBD2_DEFAULT_MAX_COMMIT_AGE);
834         journal->j_min_batch_time = 0;
835         journal->j_max_batch_time = 15000; /* 15ms */
836
837         /* The journal is marked for error until we succeed with recovery! */
838         journal->j_flags = JBD2_ABORT;
839
840         /* Set up a default-sized revoke table for the new mount. */
841         err = jbd2_journal_init_revoke(journal, JOURNAL_REVOKE_DEFAULT_HASH);
842         if (err) {
843                 kfree(journal);
844                 goto fail;
845         }
846
847         spin_lock_init(&journal->j_history_lock);
848
849         return journal;
850 fail:
851         return NULL;
852 }
853
854 /* jbd2_journal_init_dev and jbd2_journal_init_inode:
855  *
856  * Create a journal structure assigned some fixed set of disk blocks to
857  * the journal.  We don't actually touch those disk blocks yet, but we
858  * need to set up all of the mapping information to tell the journaling
859  * system where the journal blocks are.
860  *
861  */
862
863 /**
864  *  journal_t * jbd2_journal_init_dev() - creates and initialises a journal structure
865  *  @bdev: Block device on which to create the journal
866  *  @fs_dev: Device which hold journalled filesystem for this journal.
867  *  @start: Block nr Start of journal.
868  *  @len:  Length of the journal in blocks.
869  *  @blocksize: blocksize of journalling device
870  *
871  *  Returns: a newly created journal_t *
872  *
873  *  jbd2_journal_init_dev creates a journal which maps a fixed contiguous
874  *  range of blocks on an arbitrary block device.
875  *
876  */
877 journal_t * jbd2_journal_init_dev(struct block_device *bdev,
878                         struct block_device *fs_dev,
879                         unsigned long long start, int len, int blocksize)
880 {
881         journal_t *journal = journal_init_common();
882         struct buffer_head *bh;
883         char *p;
884         int n;
885
886         if (!journal)
887                 return NULL;
888
889         /* journal descriptor can store up to n blocks -bzzz */
890         journal->j_blocksize = blocksize;
891         jbd2_stats_proc_init(journal);
892         n = journal->j_blocksize / sizeof(journal_block_tag_t);
893         journal->j_wbufsize = n;
894         journal->j_wbuf = kmalloc(n * sizeof(struct buffer_head*), GFP_KERNEL);
895         if (!journal->j_wbuf) {
896                 printk(KERN_ERR "%s: Cant allocate bhs for commit thread\n",
897                         __func__);
898                 goto out_err;
899         }
900         journal->j_dev = bdev;
901         journal->j_fs_dev = fs_dev;
902         journal->j_blk_offset = start;
903         journal->j_maxlen = len;
904         bdevname(journal->j_dev, journal->j_devname);
905         p = journal->j_devname;
906         while ((p = strchr(p, '/')))
907                 *p = '!';
908
909         bh = __getblk(journal->j_dev, start, journal->j_blocksize);
910         if (!bh) {
911                 printk(KERN_ERR
912                        "%s: Cannot get buffer for journal superblock\n",
913                        __func__);
914                 goto out_err;
915         }
916         journal->j_sb_buffer = bh;
917         journal->j_superblock = (journal_superblock_t *)bh->b_data;
918
919         return journal;
920 out_err:
921         kfree(journal->j_wbuf);
922         jbd2_stats_proc_exit(journal);
923         kfree(journal);
924         return NULL;
925 }
926
927 /**
928  *  journal_t * jbd2_journal_init_inode () - creates a journal which maps to a inode.
929  *  @inode: An inode to create the journal in
930  *
931  * jbd2_journal_init_inode creates a journal which maps an on-disk inode as
932  * the journal.  The inode must exist already, must support bmap() and
933  * must have all data blocks preallocated.
934  */
935 journal_t * jbd2_journal_init_inode (struct inode *inode)
936 {
937         struct buffer_head *bh;
938         journal_t *journal = journal_init_common();
939         char *p;
940         int err;
941         int n;
942         unsigned long long blocknr;
943
944         if (!journal)
945                 return NULL;
946
947         journal->j_dev = journal->j_fs_dev = inode->i_sb->s_bdev;
948         journal->j_inode = inode;
949         bdevname(journal->j_dev, journal->j_devname);
950         p = journal->j_devname;
951         while ((p = strchr(p, '/')))
952                 *p = '!';
953         p = journal->j_devname + strlen(journal->j_devname);
954         sprintf(p, "-%lu", journal->j_inode->i_ino);
955         jbd_debug(1,
956                   "journal %p: inode %s/%ld, size %Ld, bits %d, blksize %ld\n",
957                   journal, inode->i_sb->s_id, inode->i_ino,
958                   (long long) inode->i_size,
959                   inode->i_sb->s_blocksize_bits, inode->i_sb->s_blocksize);
960
961         journal->j_maxlen = inode->i_size >> inode->i_sb->s_blocksize_bits;
962         journal->j_blocksize = inode->i_sb->s_blocksize;
963         jbd2_stats_proc_init(journal);
964
965         /* journal descriptor can store up to n blocks -bzzz */
966         n = journal->j_blocksize / sizeof(journal_block_tag_t);
967         journal->j_wbufsize = n;
968         journal->j_wbuf = kmalloc(n * sizeof(struct buffer_head*), GFP_KERNEL);
969         if (!journal->j_wbuf) {
970                 printk(KERN_ERR "%s: Cant allocate bhs for commit thread\n",
971                         __func__);
972                 goto out_err;
973         }
974
975         err = jbd2_journal_bmap(journal, 0, &blocknr);
976         /* If that failed, give up */
977         if (err) {
978                 printk(KERN_ERR "%s: Cannnot locate journal superblock\n",
979                        __func__);
980                 goto out_err;
981         }
982
983         bh = __getblk(journal->j_dev, blocknr, journal->j_blocksize);
984         if (!bh) {
985                 printk(KERN_ERR
986                        "%s: Cannot get buffer for journal superblock\n",
987                        __func__);
988                 goto out_err;
989         }
990         journal->j_sb_buffer = bh;
991         journal->j_superblock = (journal_superblock_t *)bh->b_data;
992
993         return journal;
994 out_err:
995         kfree(journal->j_wbuf);
996         jbd2_stats_proc_exit(journal);
997         kfree(journal);
998         return NULL;
999 }
1000
1001 /*
1002  * If the journal init or create aborts, we need to mark the journal
1003  * superblock as being NULL to prevent the journal destroy from writing
1004  * back a bogus superblock.
1005  */
1006 static void journal_fail_superblock (journal_t *journal)
1007 {
1008         struct buffer_head *bh = journal->j_sb_buffer;
1009         brelse(bh);
1010         journal->j_sb_buffer = NULL;
1011 }
1012
1013 /*
1014  * Given a journal_t structure, initialise the various fields for
1015  * startup of a new journaling session.  We use this both when creating
1016  * a journal, and after recovering an old journal to reset it for
1017  * subsequent use.
1018  */
1019
1020 static int journal_reset(journal_t *journal)
1021 {
1022         journal_superblock_t *sb = journal->j_superblock;
1023         unsigned long long first, last;
1024
1025         first = be32_to_cpu(sb->s_first);
1026         last = be32_to_cpu(sb->s_maxlen);
1027         if (first + JBD2_MIN_JOURNAL_BLOCKS > last + 1) {
1028                 printk(KERN_ERR "JBD: Journal too short (blocks %llu-%llu).\n",
1029                        first, last);
1030                 journal_fail_superblock(journal);
1031                 return -EINVAL;
1032         }
1033
1034         journal->j_first = first;
1035         journal->j_last = last;
1036
1037         journal->j_head = first;
1038         journal->j_tail = first;
1039         journal->j_free = last - first;
1040
1041         journal->j_tail_sequence = journal->j_transaction_sequence;
1042         journal->j_commit_sequence = journal->j_transaction_sequence - 1;
1043         journal->j_commit_request = journal->j_commit_sequence;
1044
1045         journal->j_max_transaction_buffers = journal->j_maxlen / 4;
1046
1047         /* Add the dynamic fields and write it to disk. */
1048         jbd2_journal_update_superblock(journal, 1);
1049         return jbd2_journal_start_thread(journal);
1050 }
1051
1052 /**
1053  * void jbd2_journal_update_superblock() - Update journal sb on disk.
1054  * @journal: The journal to update.
1055  * @wait: Set to '0' if you don't want to wait for IO completion.
1056  *
1057  * Update a journal's dynamic superblock fields and write it to disk,
1058  * optionally waiting for the IO to complete.
1059  */
1060 void jbd2_journal_update_superblock(journal_t *journal, int wait)
1061 {
1062         journal_superblock_t *sb = journal->j_superblock;
1063         struct buffer_head *bh = journal->j_sb_buffer;
1064
1065         /*
1066          * As a special case, if the on-disk copy is already marked as needing
1067          * no recovery (s_start == 0) and there are no outstanding transactions
1068          * in the filesystem, then we can safely defer the superblock update
1069          * until the next commit by setting JBD2_FLUSHED.  This avoids
1070          * attempting a write to a potential-readonly device.
1071          */
1072         if (sb->s_start == 0 && journal->j_tail_sequence ==
1073                                 journal->j_transaction_sequence) {
1074                 jbd_debug(1,"JBD: Skipping superblock update on recovered sb "
1075                         "(start %ld, seq %d, errno %d)\n",
1076                         journal->j_tail, journal->j_tail_sequence,
1077                         journal->j_errno);
1078                 goto out;
1079         }
1080
1081         if (buffer_write_io_error(bh)) {
1082                 /*
1083                  * Oh, dear.  A previous attempt to write the journal
1084                  * superblock failed.  This could happen because the
1085                  * USB device was yanked out.  Or it could happen to
1086                  * be a transient write error and maybe the block will
1087                  * be remapped.  Nothing we can do but to retry the
1088                  * write and hope for the best.
1089                  */
1090                 printk(KERN_ERR "JBD2: previous I/O error detected "
1091                        "for journal superblock update for %s.\n",
1092                        journal->j_devname);
1093                 clear_buffer_write_io_error(bh);
1094                 set_buffer_uptodate(bh);
1095         }
1096
1097         spin_lock(&journal->j_state_lock);
1098         jbd_debug(1,"JBD: updating superblock (start %ld, seq %d, errno %d)\n",
1099                   journal->j_tail, journal->j_tail_sequence, journal->j_errno);
1100
1101         sb->s_sequence = cpu_to_be32(journal->j_tail_sequence);
1102         sb->s_start    = cpu_to_be32(journal->j_tail);
1103         sb->s_errno    = cpu_to_be32(journal->j_errno);
1104         spin_unlock(&journal->j_state_lock);
1105
1106         BUFFER_TRACE(bh, "marking dirty");
1107         mark_buffer_dirty(bh);
1108         if (wait) {
1109                 sync_dirty_buffer(bh);
1110                 if (buffer_write_io_error(bh)) {
1111                         printk(KERN_ERR "JBD2: I/O error detected "
1112                                "when updating journal superblock for %s.\n",
1113                                journal->j_devname);
1114                         clear_buffer_write_io_error(bh);
1115                         set_buffer_uptodate(bh);
1116                 }
1117         } else
1118                 ll_rw_block(SWRITE, 1, &bh);
1119
1120 out:
1121         /* If we have just flushed the log (by marking s_start==0), then
1122          * any future commit will have to be careful to update the
1123          * superblock again to re-record the true start of the log. */
1124
1125         spin_lock(&journal->j_state_lock);
1126         if (sb->s_start)
1127                 journal->j_flags &= ~JBD2_FLUSHED;
1128         else
1129                 journal->j_flags |= JBD2_FLUSHED;
1130         spin_unlock(&journal->j_state_lock);
1131 }
1132
1133 /*
1134  * Read the superblock for a given journal, performing initial
1135  * validation of the format.
1136  */
1137
1138 static int journal_get_superblock(journal_t *journal)
1139 {
1140         struct buffer_head *bh;
1141         journal_superblock_t *sb;
1142         int err = -EIO;
1143
1144         bh = journal->j_sb_buffer;
1145
1146         J_ASSERT(bh != NULL);
1147         if (!buffer_uptodate(bh)) {
1148                 ll_rw_block(READ, 1, &bh);
1149                 wait_on_buffer(bh);
1150                 if (!buffer_uptodate(bh)) {
1151                         printk (KERN_ERR
1152                                 "JBD: IO error reading journal superblock\n");
1153                         goto out;
1154                 }
1155         }
1156
1157         sb = journal->j_superblock;
1158
1159         err = -EINVAL;
1160
1161         if (sb->s_header.h_magic != cpu_to_be32(JBD2_MAGIC_NUMBER) ||
1162             sb->s_blocksize != cpu_to_be32(journal->j_blocksize)) {
1163                 printk(KERN_WARNING "JBD: no valid journal superblock found\n");
1164                 goto out;
1165         }
1166
1167         switch(be32_to_cpu(sb->s_header.h_blocktype)) {
1168         case JBD2_SUPERBLOCK_V1:
1169                 journal->j_format_version = 1;
1170                 break;
1171         case JBD2_SUPERBLOCK_V2:
1172                 journal->j_format_version = 2;
1173                 break;
1174         default:
1175                 printk(KERN_WARNING "JBD: unrecognised superblock format ID\n");
1176                 goto out;
1177         }
1178
1179         if (be32_to_cpu(sb->s_maxlen) < journal->j_maxlen)
1180                 journal->j_maxlen = be32_to_cpu(sb->s_maxlen);
1181         else if (be32_to_cpu(sb->s_maxlen) > journal->j_maxlen) {
1182                 printk (KERN_WARNING "JBD: journal file too short\n");
1183                 goto out;
1184         }
1185
1186         return 0;
1187
1188 out:
1189         journal_fail_superblock(journal);
1190         return err;
1191 }
1192
1193 /*
1194  * Load the on-disk journal superblock and read the key fields into the
1195  * journal_t.
1196  */
1197
1198 static int load_superblock(journal_t *journal)
1199 {
1200         int err;
1201         journal_superblock_t *sb;
1202
1203         err = journal_get_superblock(journal);
1204         if (err)
1205                 return err;
1206
1207         sb = journal->j_superblock;
1208
1209         journal->j_tail_sequence = be32_to_cpu(sb->s_sequence);
1210         journal->j_tail = be32_to_cpu(sb->s_start);
1211         journal->j_first = be32_to_cpu(sb->s_first);
1212         journal->j_last = be32_to_cpu(sb->s_maxlen);
1213         journal->j_errno = be32_to_cpu(sb->s_errno);
1214
1215         return 0;
1216 }
1217
1218
1219 /**
1220  * int jbd2_journal_load() - Read journal from disk.
1221  * @journal: Journal to act on.
1222  *
1223  * Given a journal_t structure which tells us which disk blocks contain
1224  * a journal, read the journal from disk to initialise the in-memory
1225  * structures.
1226  */
1227 int jbd2_journal_load(journal_t *journal)
1228 {
1229         int err;
1230         journal_superblock_t *sb;
1231
1232         err = load_superblock(journal);
1233         if (err)
1234                 return err;
1235
1236         sb = journal->j_superblock;
1237         /* If this is a V2 superblock, then we have to check the
1238          * features flags on it. */
1239
1240         if (journal->j_format_version >= 2) {
1241                 if ((sb->s_feature_ro_compat &
1242                      ~cpu_to_be32(JBD2_KNOWN_ROCOMPAT_FEATURES)) ||
1243                     (sb->s_feature_incompat &
1244                      ~cpu_to_be32(JBD2_KNOWN_INCOMPAT_FEATURES))) {
1245                         printk (KERN_WARNING
1246                                 "JBD: Unrecognised features on journal\n");
1247                         return -EINVAL;
1248                 }
1249         }
1250
1251         /* Let the recovery code check whether it needs to recover any
1252          * data from the journal. */
1253         if (jbd2_journal_recover(journal))
1254                 goto recovery_error;
1255
1256         if (journal->j_failed_commit) {
1257                 printk(KERN_ERR "JBD2: journal transaction %u on %s "
1258                        "is corrupt.\n", journal->j_failed_commit,
1259                        journal->j_devname);
1260                 return -EIO;
1261         }
1262
1263         /* OK, we've finished with the dynamic journal bits:
1264          * reinitialise the dynamic contents of the superblock in memory
1265          * and reset them on disk. */
1266         if (journal_reset(journal))
1267                 goto recovery_error;
1268
1269         journal->j_flags &= ~JBD2_ABORT;
1270         journal->j_flags |= JBD2_LOADED;
1271         return 0;
1272
1273 recovery_error:
1274         printk (KERN_WARNING "JBD: recovery failed\n");
1275         return -EIO;
1276 }
1277
1278 /**
1279  * void jbd2_journal_destroy() - Release a journal_t structure.
1280  * @journal: Journal to act on.
1281  *
1282  * Release a journal_t structure once it is no longer in use by the
1283  * journaled object.
1284  * Return <0 if we couldn't clean up the journal.
1285  */
1286 int jbd2_journal_destroy(journal_t *journal)
1287 {
1288         int err = 0;
1289
1290         /* Wait for the commit thread to wake up and die. */
1291         journal_kill_thread(journal);
1292
1293         /* Force a final log commit */
1294         if (journal->j_running_transaction)
1295                 jbd2_journal_commit_transaction(journal);
1296
1297         /* Force any old transactions to disk */
1298
1299         /* Totally anal locking here... */
1300         spin_lock(&journal->j_list_lock);
1301         while (journal->j_checkpoint_transactions != NULL) {
1302                 spin_unlock(&journal->j_list_lock);
1303                 mutex_lock(&journal->j_checkpoint_mutex);
1304                 jbd2_log_do_checkpoint(journal);
1305                 mutex_unlock(&journal->j_checkpoint_mutex);
1306                 spin_lock(&journal->j_list_lock);
1307         }
1308
1309         J_ASSERT(journal->j_running_transaction == NULL);
1310         J_ASSERT(journal->j_committing_transaction == NULL);
1311         J_ASSERT(journal->j_checkpoint_transactions == NULL);
1312         spin_unlock(&journal->j_list_lock);
1313
1314         if (journal->j_sb_buffer) {
1315                 if (!is_journal_aborted(journal)) {
1316                         /* We can now mark the journal as empty. */
1317                         journal->j_tail = 0;
1318                         journal->j_tail_sequence =
1319                                 ++journal->j_transaction_sequence;
1320                         jbd2_journal_update_superblock(journal, 1);
1321                 } else {
1322                         err = -EIO;
1323                 }
1324                 brelse(journal->j_sb_buffer);
1325         }
1326
1327         if (journal->j_proc_entry)
1328                 jbd2_stats_proc_exit(journal);
1329         if (journal->j_inode)
1330                 iput(journal->j_inode);
1331         if (journal->j_revoke)
1332                 jbd2_journal_destroy_revoke(journal);
1333         kfree(journal->j_wbuf);
1334         kfree(journal);
1335
1336         return err;
1337 }
1338
1339
1340 /**
1341  *int jbd2_journal_check_used_features () - Check if features specified are used.
1342  * @journal: Journal to check.
1343  * @compat: bitmask of compatible features
1344  * @ro: bitmask of features that force read-only mount
1345  * @incompat: bitmask of incompatible features
1346  *
1347  * Check whether the journal uses all of a given set of
1348  * features.  Return true (non-zero) if it does.
1349  **/
1350
1351 int jbd2_journal_check_used_features (journal_t *journal, unsigned long compat,
1352                                  unsigned long ro, unsigned long incompat)
1353 {
1354         journal_superblock_t *sb;
1355
1356         if (!compat && !ro && !incompat)
1357                 return 1;
1358         if (journal->j_format_version == 1)
1359                 return 0;
1360
1361         sb = journal->j_superblock;
1362
1363         if (((be32_to_cpu(sb->s_feature_compat) & compat) == compat) &&
1364             ((be32_to_cpu(sb->s_feature_ro_compat) & ro) == ro) &&
1365             ((be32_to_cpu(sb->s_feature_incompat) & incompat) == incompat))
1366                 return 1;
1367
1368         return 0;
1369 }
1370
1371 /**
1372  * int jbd2_journal_check_available_features() - Check feature set in journalling layer
1373  * @journal: Journal to check.
1374  * @compat: bitmask of compatible features
1375  * @ro: bitmask of features that force read-only mount
1376  * @incompat: bitmask of incompatible features
1377  *
1378  * Check whether the journaling code supports the use of
1379  * all of a given set of features on this journal.  Return true
1380  * (non-zero) if it can. */
1381
1382 int jbd2_journal_check_available_features (journal_t *journal, unsigned long compat,
1383                                       unsigned long ro, unsigned long incompat)
1384 {
1385         journal_superblock_t *sb;
1386
1387         if (!compat && !ro && !incompat)
1388                 return 1;
1389
1390         sb = journal->j_superblock;
1391
1392         /* We can support any known requested features iff the
1393          * superblock is in version 2.  Otherwise we fail to support any
1394          * extended sb features. */
1395
1396         if (journal->j_format_version != 2)
1397                 return 0;
1398
1399         if ((compat   & JBD2_KNOWN_COMPAT_FEATURES) == compat &&
1400             (ro       & JBD2_KNOWN_ROCOMPAT_FEATURES) == ro &&
1401             (incompat & JBD2_KNOWN_INCOMPAT_FEATURES) == incompat)
1402                 return 1;
1403
1404         return 0;
1405 }
1406
1407 /**
1408  * int jbd2_journal_set_features () - Mark a given journal feature in the superblock
1409  * @journal: Journal to act on.
1410  * @compat: bitmask of compatible features
1411  * @ro: bitmask of features that force read-only mount
1412  * @incompat: bitmask of incompatible features
1413  *
1414  * Mark a given journal feature as present on the
1415  * superblock.  Returns true if the requested features could be set.
1416  *
1417  */
1418
1419 int jbd2_journal_set_features (journal_t *journal, unsigned long compat,
1420                           unsigned long ro, unsigned long incompat)
1421 {
1422         journal_superblock_t *sb;
1423
1424         if (jbd2_journal_check_used_features(journal, compat, ro, incompat))
1425                 return 1;
1426
1427         if (!jbd2_journal_check_available_features(journal, compat, ro, incompat))
1428                 return 0;
1429
1430         jbd_debug(1, "Setting new features 0x%lx/0x%lx/0x%lx\n",
1431                   compat, ro, incompat);
1432
1433         sb = journal->j_superblock;
1434
1435         sb->s_feature_compat    |= cpu_to_be32(compat);
1436         sb->s_feature_ro_compat |= cpu_to_be32(ro);
1437         sb->s_feature_incompat  |= cpu_to_be32(incompat);
1438
1439         return 1;
1440 }
1441
1442 /*
1443  * jbd2_journal_clear_features () - Clear a given journal feature in the
1444  *                                  superblock
1445  * @journal: Journal to act on.
1446  * @compat: bitmask of compatible features
1447  * @ro: bitmask of features that force read-only mount
1448  * @incompat: bitmask of incompatible features
1449  *
1450  * Clear a given journal feature as present on the
1451  * superblock.
1452  */
1453 void jbd2_journal_clear_features(journal_t *journal, unsigned long compat,
1454                                 unsigned long ro, unsigned long incompat)
1455 {
1456         journal_superblock_t *sb;
1457
1458         jbd_debug(1, "Clear features 0x%lx/0x%lx/0x%lx\n",
1459                   compat, ro, incompat);
1460
1461         sb = journal->j_superblock;
1462
1463         sb->s_feature_compat    &= ~cpu_to_be32(compat);
1464         sb->s_feature_ro_compat &= ~cpu_to_be32(ro);
1465         sb->s_feature_incompat  &= ~cpu_to_be32(incompat);
1466 }
1467 EXPORT_SYMBOL(jbd2_journal_clear_features);
1468
1469 /**
1470  * int jbd2_journal_update_format () - Update on-disk journal structure.
1471  * @journal: Journal to act on.
1472  *
1473  * Given an initialised but unloaded journal struct, poke about in the
1474  * on-disk structure to update it to the most recent supported version.
1475  */
1476 int jbd2_journal_update_format (journal_t *journal)
1477 {
1478         journal_superblock_t *sb;
1479         int err;
1480
1481         err = journal_get_superblock(journal);
1482         if (err)
1483                 return err;
1484
1485         sb = journal->j_superblock;
1486
1487         switch (be32_to_cpu(sb->s_header.h_blocktype)) {
1488         case JBD2_SUPERBLOCK_V2:
1489                 return 0;
1490         case JBD2_SUPERBLOCK_V1:
1491                 return journal_convert_superblock_v1(journal, sb);
1492         default:
1493                 break;
1494         }
1495         return -EINVAL;
1496 }
1497
1498 static int journal_convert_superblock_v1(journal_t *journal,
1499                                          journal_superblock_t *sb)
1500 {
1501         int offset, blocksize;
1502         struct buffer_head *bh;
1503
1504         printk(KERN_WARNING
1505                 "JBD: Converting superblock from version 1 to 2.\n");
1506
1507         /* Pre-initialise new fields to zero */
1508         offset = ((char *) &(sb->s_feature_compat)) - ((char *) sb);
1509         blocksize = be32_to_cpu(sb->s_blocksize);
1510         memset(&sb->s_feature_compat, 0, blocksize-offset);
1511
1512         sb->s_nr_users = cpu_to_be32(1);
1513         sb->s_header.h_blocktype = cpu_to_be32(JBD2_SUPERBLOCK_V2);
1514         journal->j_format_version = 2;
1515
1516         bh = journal->j_sb_buffer;
1517         BUFFER_TRACE(bh, "marking dirty");
1518         mark_buffer_dirty(bh);
1519         sync_dirty_buffer(bh);
1520         return 0;
1521 }
1522
1523
1524 /**
1525  * int jbd2_journal_flush () - Flush journal
1526  * @journal: Journal to act on.
1527  *
1528  * Flush all data for a given journal to disk and empty the journal.
1529  * Filesystems can use this when remounting readonly to ensure that
1530  * recovery does not need to happen on remount.
1531  */
1532
1533 int jbd2_journal_flush(journal_t *journal)
1534 {
1535         int err = 0;
1536         transaction_t *transaction = NULL;
1537         unsigned long old_tail;
1538
1539         spin_lock(&journal->j_state_lock);
1540
1541         /* Force everything buffered to the log... */
1542         if (journal->j_running_transaction) {
1543                 transaction = journal->j_running_transaction;
1544                 __jbd2_log_start_commit(journal, transaction->t_tid);
1545         } else if (journal->j_committing_transaction)
1546                 transaction = journal->j_committing_transaction;
1547
1548         /* Wait for the log commit to complete... */
1549         if (transaction) {
1550                 tid_t tid = transaction->t_tid;
1551
1552                 spin_unlock(&journal->j_state_lock);
1553                 jbd2_log_wait_commit(journal, tid);
1554         } else {
1555                 spin_unlock(&journal->j_state_lock);
1556         }
1557
1558         /* ...and flush everything in the log out to disk. */
1559         spin_lock(&journal->j_list_lock);
1560         while (!err && journal->j_checkpoint_transactions != NULL) {
1561                 spin_unlock(&journal->j_list_lock);
1562                 mutex_lock(&journal->j_checkpoint_mutex);
1563                 err = jbd2_log_do_checkpoint(journal);
1564                 mutex_unlock(&journal->j_checkpoint_mutex);
1565                 spin_lock(&journal->j_list_lock);
1566         }
1567         spin_unlock(&journal->j_list_lock);
1568
1569         if (is_journal_aborted(journal))
1570                 return -EIO;
1571
1572         jbd2_cleanup_journal_tail(journal);
1573
1574         /* Finally, mark the journal as really needing no recovery.
1575          * This sets s_start==0 in the underlying superblock, which is
1576          * the magic code for a fully-recovered superblock.  Any future
1577          * commits of data to the journal will restore the current
1578          * s_start value. */
1579         spin_lock(&journal->j_state_lock);
1580         old_tail = journal->j_tail;
1581         journal->j_tail = 0;
1582         spin_unlock(&journal->j_state_lock);
1583         jbd2_journal_update_superblock(journal, 1);
1584         spin_lock(&journal->j_state_lock);
1585         journal->j_tail = old_tail;
1586
1587         J_ASSERT(!journal->j_running_transaction);
1588         J_ASSERT(!journal->j_committing_transaction);
1589         J_ASSERT(!journal->j_checkpoint_transactions);
1590         J_ASSERT(journal->j_head == journal->j_tail);
1591         J_ASSERT(journal->j_tail_sequence == journal->j_transaction_sequence);
1592         spin_unlock(&journal->j_state_lock);
1593         return 0;
1594 }
1595
1596 /**
1597  * int jbd2_journal_wipe() - Wipe journal contents
1598  * @journal: Journal to act on.
1599  * @write: flag (see below)
1600  *
1601  * Wipe out all of the contents of a journal, safely.  This will produce
1602  * a warning if the journal contains any valid recovery information.
1603  * Must be called between journal_init_*() and jbd2_journal_load().
1604  *
1605  * If 'write' is non-zero, then we wipe out the journal on disk; otherwise
1606  * we merely suppress recovery.
1607  */
1608
1609 int jbd2_journal_wipe(journal_t *journal, int write)
1610 {
1611         journal_superblock_t *sb;
1612         int err = 0;
1613
1614         J_ASSERT (!(journal->j_flags & JBD2_LOADED));
1615
1616         err = load_superblock(journal);
1617         if (err)
1618                 return err;
1619
1620         sb = journal->j_superblock;
1621
1622         if (!journal->j_tail)
1623                 goto no_recovery;
1624
1625         printk (KERN_WARNING "JBD: %s recovery information on journal\n",
1626                 write ? "Clearing" : "Ignoring");
1627
1628         err = jbd2_journal_skip_recovery(journal);
1629         if (write)
1630                 jbd2_journal_update_superblock(journal, 1);
1631
1632  no_recovery:
1633         return err;
1634 }
1635
1636 /*
1637  * Journal abort has very specific semantics, which we describe
1638  * for journal abort.
1639  *
1640  * Two internal functions, which provide abort to the jbd layer
1641  * itself are here.
1642  */
1643
1644 /*
1645  * Quick version for internal journal use (doesn't lock the journal).
1646  * Aborts hard --- we mark the abort as occurred, but do _nothing_ else,
1647  * and don't attempt to make any other journal updates.
1648  */
1649 void __jbd2_journal_abort_hard(journal_t *journal)
1650 {
1651         transaction_t *transaction;
1652
1653         if (journal->j_flags & JBD2_ABORT)
1654                 return;
1655
1656         printk(KERN_ERR "Aborting journal on device %s.\n",
1657                journal->j_devname);
1658
1659         spin_lock(&journal->j_state_lock);
1660         journal->j_flags |= JBD2_ABORT;
1661         transaction = journal->j_running_transaction;
1662         if (transaction)
1663                 __jbd2_log_start_commit(journal, transaction->t_tid);
1664         spin_unlock(&journal->j_state_lock);
1665 }
1666
1667 /* Soft abort: record the abort error status in the journal superblock,
1668  * but don't do any other IO. */
1669 static void __journal_abort_soft (journal_t *journal, int errno)
1670 {
1671         if (journal->j_flags & JBD2_ABORT)
1672                 return;
1673
1674         if (!journal->j_errno)
1675                 journal->j_errno = errno;
1676
1677         __jbd2_journal_abort_hard(journal);
1678
1679         if (errno)
1680                 jbd2_journal_update_superblock(journal, 1);
1681 }
1682
1683 /**
1684  * void jbd2_journal_abort () - Shutdown the journal immediately.
1685  * @journal: the journal to shutdown.
1686  * @errno:   an error number to record in the journal indicating
1687  *           the reason for the shutdown.
1688  *
1689  * Perform a complete, immediate shutdown of the ENTIRE
1690  * journal (not of a single transaction).  This operation cannot be
1691  * undone without closing and reopening the journal.
1692  *
1693  * The jbd2_journal_abort function is intended to support higher level error
1694  * recovery mechanisms such as the ext2/ext3 remount-readonly error
1695  * mode.
1696  *
1697  * Journal abort has very specific semantics.  Any existing dirty,
1698  * unjournaled buffers in the main filesystem will still be written to
1699  * disk by bdflush, but the journaling mechanism will be suspended
1700  * immediately and no further transaction commits will be honoured.
1701  *
1702  * Any dirty, journaled buffers will be written back to disk without
1703  * hitting the journal.  Atomicity cannot be guaranteed on an aborted
1704  * filesystem, but we _do_ attempt to leave as much data as possible
1705  * behind for fsck to use for cleanup.
1706  *
1707  * Any attempt to get a new transaction handle on a journal which is in
1708  * ABORT state will just result in an -EROFS error return.  A
1709  * jbd2_journal_stop on an existing handle will return -EIO if we have
1710  * entered abort state during the update.
1711  *
1712  * Recursive transactions are not disturbed by journal abort until the
1713  * final jbd2_journal_stop, which will receive the -EIO error.
1714  *
1715  * Finally, the jbd2_journal_abort call allows the caller to supply an errno
1716  * which will be recorded (if possible) in the journal superblock.  This
1717  * allows a client to record failure conditions in the middle of a
1718  * transaction without having to complete the transaction to record the
1719  * failure to disk.  ext3_error, for example, now uses this
1720  * functionality.
1721  *
1722  * Errors which originate from within the journaling layer will NOT
1723  * supply an errno; a null errno implies that absolutely no further
1724  * writes are done to the journal (unless there are any already in
1725  * progress).
1726  *
1727  */
1728
1729 void jbd2_journal_abort(journal_t *journal, int errno)
1730 {
1731         __journal_abort_soft(journal, errno);
1732 }
1733
1734 /**
1735  * int jbd2_journal_errno () - returns the journal's error state.
1736  * @journal: journal to examine.
1737  *
1738  * This is the errno number set with jbd2_journal_abort(), the last
1739  * time the journal was mounted - if the journal was stopped
1740  * without calling abort this will be 0.
1741  *
1742  * If the journal has been aborted on this mount time -EROFS will
1743  * be returned.
1744  */
1745 int jbd2_journal_errno(journal_t *journal)
1746 {
1747         int err;
1748
1749         spin_lock(&journal->j_state_lock);
1750         if (journal->j_flags & JBD2_ABORT)
1751                 err = -EROFS;
1752         else
1753                 err = journal->j_errno;
1754         spin_unlock(&journal->j_state_lock);
1755         return err;
1756 }
1757
1758 /**
1759  * int jbd2_journal_clear_err () - clears the journal's error state
1760  * @journal: journal to act on.
1761  *
1762  * An error must be cleared or acked to take a FS out of readonly
1763  * mode.
1764  */
1765 int jbd2_journal_clear_err(journal_t *journal)
1766 {
1767         int err = 0;
1768
1769         spin_lock(&journal->j_state_lock);
1770         if (journal->j_flags & JBD2_ABORT)
1771                 err = -EROFS;
1772         else
1773                 journal->j_errno = 0;
1774         spin_unlock(&journal->j_state_lock);
1775         return err;
1776 }
1777
1778 /**
1779  * void jbd2_journal_ack_err() - Ack journal err.
1780  * @journal: journal to act on.
1781  *
1782  * An error must be cleared or acked to take a FS out of readonly
1783  * mode.
1784  */
1785 void jbd2_journal_ack_err(journal_t *journal)
1786 {
1787         spin_lock(&journal->j_state_lock);
1788         if (journal->j_errno)
1789                 journal->j_flags |= JBD2_ACK_ERR;
1790         spin_unlock(&journal->j_state_lock);
1791 }
1792
1793 int jbd2_journal_blocks_per_page(struct inode *inode)
1794 {
1795         return 1 << (PAGE_CACHE_SHIFT - inode->i_sb->s_blocksize_bits);
1796 }
1797
1798 /*
1799  * helper functions to deal with 32 or 64bit block numbers.
1800  */
1801 size_t journal_tag_bytes(journal_t *journal)
1802 {
1803         if (JBD2_HAS_INCOMPAT_FEATURE(journal, JBD2_FEATURE_INCOMPAT_64BIT))
1804                 return JBD2_TAG_SIZE64;
1805         else
1806                 return JBD2_TAG_SIZE32;
1807 }
1808
1809 /*
1810  * Journal_head storage management
1811  */
1812 static struct kmem_cache *jbd2_journal_head_cache;
1813 #ifdef CONFIG_JBD2_DEBUG
1814 static atomic_t nr_journal_heads = ATOMIC_INIT(0);
1815 #endif
1816
1817 static int journal_init_jbd2_journal_head_cache(void)
1818 {
1819         int retval;
1820
1821         J_ASSERT(jbd2_journal_head_cache == NULL);
1822         jbd2_journal_head_cache = kmem_cache_create("jbd2_journal_head",
1823                                 sizeof(struct journal_head),
1824                                 0,              /* offset */
1825                                 SLAB_TEMPORARY, /* flags */
1826                                 NULL);          /* ctor */
1827         retval = 0;
1828         if (!jbd2_journal_head_cache) {
1829                 retval = -ENOMEM;
1830                 printk(KERN_EMERG "JBD: no memory for journal_head cache\n");
1831         }
1832         return retval;
1833 }
1834
1835 static void jbd2_journal_destroy_jbd2_journal_head_cache(void)
1836 {
1837         if (jbd2_journal_head_cache) {
1838                 kmem_cache_destroy(jbd2_journal_head_cache);
1839                 jbd2_journal_head_cache = NULL;
1840         }
1841 }
1842
1843 /*
1844  * journal_head splicing and dicing
1845  */
1846 static struct journal_head *journal_alloc_journal_head(void)
1847 {
1848         struct journal_head *ret;
1849         static unsigned long last_warning;
1850
1851 #ifdef CONFIG_JBD2_DEBUG
1852         atomic_inc(&nr_journal_heads);
1853 #endif
1854         ret = kmem_cache_alloc(jbd2_journal_head_cache, GFP_NOFS);
1855         if (!ret) {
1856                 jbd_debug(1, "out of memory for journal_head\n");
1857                 if (time_after(jiffies, last_warning + 5*HZ)) {
1858                         printk(KERN_NOTICE "ENOMEM in %s, retrying.\n",
1859                                __func__);
1860                         last_warning = jiffies;
1861                 }
1862                 while (!ret) {
1863                         yield();
1864                         ret = kmem_cache_alloc(jbd2_journal_head_cache, GFP_NOFS);
1865                 }
1866         }
1867         return ret;
1868 }
1869
1870 static void journal_free_journal_head(struct journal_head *jh)
1871 {
1872 #ifdef CONFIG_JBD2_DEBUG
1873         atomic_dec(&nr_journal_heads);
1874         memset(jh, JBD2_POISON_FREE, sizeof(*jh));
1875 #endif
1876         kmem_cache_free(jbd2_journal_head_cache, jh);
1877 }
1878
1879 /*
1880  * A journal_head is attached to a buffer_head whenever JBD has an
1881  * interest in the buffer.
1882  *
1883  * Whenever a buffer has an attached journal_head, its ->b_state:BH_JBD bit
1884  * is set.  This bit is tested in core kernel code where we need to take
1885  * JBD-specific actions.  Testing the zeroness of ->b_private is not reliable
1886  * there.
1887  *
1888  * When a buffer has its BH_JBD bit set, its ->b_count is elevated by one.
1889  *
1890  * When a buffer has its BH_JBD bit set it is immune from being released by
1891  * core kernel code, mainly via ->b_count.
1892  *
1893  * A journal_head may be detached from its buffer_head when the journal_head's
1894  * b_transaction, b_cp_transaction and b_next_transaction pointers are NULL.
1895  * Various places in JBD call jbd2_journal_remove_journal_head() to indicate that the
1896  * journal_head can be dropped if needed.
1897  *
1898  * Various places in the kernel want to attach a journal_head to a buffer_head
1899  * _before_ attaching the journal_head to a transaction.  To protect the
1900  * journal_head in this situation, jbd2_journal_add_journal_head elevates the
1901  * journal_head's b_jcount refcount by one.  The caller must call
1902  * jbd2_journal_put_journal_head() to undo this.
1903  *
1904  * So the typical usage would be:
1905  *
1906  *      (Attach a journal_head if needed.  Increments b_jcount)
1907  *      struct journal_head *jh = jbd2_journal_add_journal_head(bh);
1908  *      ...
1909  *      jh->b_transaction = xxx;
1910  *      jbd2_journal_put_journal_head(jh);
1911  *
1912  * Now, the journal_head's b_jcount is zero, but it is safe from being released
1913  * because it has a non-zero b_transaction.
1914  */
1915
1916 /*
1917  * Give a buffer_head a journal_head.
1918  *
1919  * Doesn't need the journal lock.
1920  * May sleep.
1921  */
1922 struct journal_head *jbd2_journal_add_journal_head(struct buffer_head *bh)
1923 {
1924         struct journal_head *jh;
1925         struct journal_head *new_jh = NULL;
1926
1927 repeat:
1928         if (!buffer_jbd(bh)) {
1929                 new_jh = journal_alloc_journal_head();
1930                 memset(new_jh, 0, sizeof(*new_jh));
1931         }
1932
1933         jbd_lock_bh_journal_head(bh);
1934         if (buffer_jbd(bh)) {
1935                 jh = bh2jh(bh);
1936         } else {
1937                 J_ASSERT_BH(bh,
1938                         (atomic_read(&bh->b_count) > 0) ||
1939                         (bh->b_page && bh->b_page->mapping));
1940
1941                 if (!new_jh) {
1942                         jbd_unlock_bh_journal_head(bh);
1943                         goto repeat;
1944                 }
1945
1946                 jh = new_jh;
1947                 new_jh = NULL;          /* We consumed it */
1948                 set_buffer_jbd(bh);
1949                 bh->b_private = jh;
1950                 jh->b_bh = bh;
1951                 get_bh(bh);
1952                 BUFFER_TRACE(bh, "added journal_head");
1953         }
1954         jh->b_jcount++;
1955         jbd_unlock_bh_journal_head(bh);
1956         if (new_jh)
1957                 journal_free_journal_head(new_jh);
1958         return bh->b_private;
1959 }
1960
1961 /*
1962  * Grab a ref against this buffer_head's journal_head.  If it ended up not
1963  * having a journal_head, return NULL
1964  */
1965 struct journal_head *jbd2_journal_grab_journal_head(struct buffer_head *bh)
1966 {
1967         struct journal_head *jh = NULL;
1968
1969         jbd_lock_bh_journal_head(bh);
1970         if (buffer_jbd(bh)) {
1971                 jh = bh2jh(bh);
1972                 jh->b_jcount++;
1973         }
1974         jbd_unlock_bh_journal_head(bh);
1975         return jh;
1976 }
1977
1978 static void __journal_remove_journal_head(struct buffer_head *bh)
1979 {
1980         struct journal_head *jh = bh2jh(bh);
1981
1982         J_ASSERT_JH(jh, jh->b_jcount >= 0);
1983
1984         get_bh(bh);
1985         if (jh->b_jcount == 0) {
1986                 if (jh->b_transaction == NULL &&
1987                                 jh->b_next_transaction == NULL &&
1988                                 jh->b_cp_transaction == NULL) {
1989                         J_ASSERT_JH(jh, jh->b_jlist == BJ_None);
1990                         J_ASSERT_BH(bh, buffer_jbd(bh));
1991                         J_ASSERT_BH(bh, jh2bh(jh) == bh);
1992                         BUFFER_TRACE(bh, "remove journal_head");
1993                         if (jh->b_frozen_data) {
1994                                 printk(KERN_WARNING "%s: freeing "
1995                                                 "b_frozen_data\n",
1996                                                 __func__);
1997                                 jbd2_free(jh->b_frozen_data, bh->b_size);
1998                         }
1999                         if (jh->b_committed_data) {
2000                                 printk(KERN_WARNING "%s: freeing "
2001                                                 "b_committed_data\n",
2002                                                 __func__);
2003                                 jbd2_free(jh->b_committed_data, bh->b_size);
2004                         }
2005                         bh->b_private = NULL;
2006                         jh->b_bh = NULL;        /* debug, really */
2007                         clear_buffer_jbd(bh);
2008                         __brelse(bh);
2009                         journal_free_journal_head(jh);
2010                 } else {
2011                         BUFFER_TRACE(bh, "journal_head was locked");
2012                 }
2013         }
2014 }
2015
2016 /*
2017  * jbd2_journal_remove_journal_head(): if the buffer isn't attached to a transaction
2018  * and has a zero b_jcount then remove and release its journal_head.   If we did
2019  * see that the buffer is not used by any transaction we also "logically"
2020  * decrement ->b_count.
2021  *
2022  * We in fact take an additional increment on ->b_count as a convenience,
2023  * because the caller usually wants to do additional things with the bh
2024  * after calling here.
2025  * The caller of jbd2_journal_remove_journal_head() *must* run __brelse(bh) at some
2026  * time.  Once the caller has run __brelse(), the buffer is eligible for
2027  * reaping by try_to_free_buffers().
2028  */
2029 void jbd2_journal_remove_journal_head(struct buffer_head *bh)
2030 {
2031         jbd_lock_bh_journal_head(bh);
2032         __journal_remove_journal_head(bh);
2033         jbd_unlock_bh_journal_head(bh);
2034 }
2035
2036 /*
2037  * Drop a reference on the passed journal_head.  If it fell to zero then try to
2038  * release the journal_head from the buffer_head.
2039  */
2040 void jbd2_journal_put_journal_head(struct journal_head *jh)
2041 {
2042         struct buffer_head *bh = jh2bh(jh);
2043
2044         jbd_lock_bh_journal_head(bh);
2045         J_ASSERT_JH(jh, jh->b_jcount > 0);
2046         --jh->b_jcount;
2047         if (!jh->b_jcount && !jh->b_transaction) {
2048                 __journal_remove_journal_head(bh);
2049                 __brelse(bh);
2050         }
2051         jbd_unlock_bh_journal_head(bh);
2052 }
2053
2054 /*
2055  * Initialize jbd inode head
2056  */
2057 void jbd2_journal_init_jbd_inode(struct jbd2_inode *jinode, struct inode *inode)
2058 {
2059         jinode->i_transaction = NULL;
2060         jinode->i_next_transaction = NULL;
2061         jinode->i_vfs_inode = inode;
2062         jinode->i_flags = 0;
2063         INIT_LIST_HEAD(&jinode->i_list);
2064 }
2065
2066 /*
2067  * Function to be called before we start removing inode from memory (i.e.,
2068  * clear_inode() is a fine place to be called from). It removes inode from
2069  * transaction's lists.
2070  */
2071 void jbd2_journal_release_jbd_inode(journal_t *journal,
2072                                     struct jbd2_inode *jinode)
2073 {
2074         int writeout = 0;
2075
2076         if (!journal)
2077                 return;
2078 restart:
2079         spin_lock(&journal->j_list_lock);
2080         /* Is commit writing out inode - we have to wait */
2081         if (jinode->i_flags & JI_COMMIT_RUNNING) {
2082                 wait_queue_head_t *wq;
2083                 DEFINE_WAIT_BIT(wait, &jinode->i_flags, __JI_COMMIT_RUNNING);
2084                 wq = bit_waitqueue(&jinode->i_flags, __JI_COMMIT_RUNNING);
2085                 prepare_to_wait(wq, &wait.wait, TASK_UNINTERRUPTIBLE);
2086                 spin_unlock(&journal->j_list_lock);
2087                 schedule();
2088                 finish_wait(wq, &wait.wait);
2089                 goto restart;
2090         }
2091
2092         /* Do we need to wait for data writeback? */
2093         if (journal->j_committing_transaction == jinode->i_transaction)
2094                 writeout = 1;
2095         if (jinode->i_transaction) {
2096                 list_del(&jinode->i_list);
2097                 jinode->i_transaction = NULL;
2098         }
2099         spin_unlock(&journal->j_list_lock);
2100 }
2101
2102 /*
2103  * debugfs tunables
2104  */
2105 #ifdef CONFIG_JBD2_DEBUG
2106 u8 jbd2_journal_enable_debug __read_mostly;
2107 EXPORT_SYMBOL(jbd2_journal_enable_debug);
2108
2109 #define JBD2_DEBUG_NAME "jbd2-debug"
2110
2111 static struct dentry *jbd2_debugfs_dir;
2112 static struct dentry *jbd2_debug;
2113
2114 static void __init jbd2_create_debugfs_entry(void)
2115 {
2116         jbd2_debugfs_dir = debugfs_create_dir("jbd2", NULL);
2117         if (jbd2_debugfs_dir)
2118                 jbd2_debug = debugfs_create_u8(JBD2_DEBUG_NAME,
2119                                                S_IRUGO | S_IWUSR,
2120                                                jbd2_debugfs_dir,
2121                                                &jbd2_journal_enable_debug);
2122 }
2123
2124 static void __exit jbd2_remove_debugfs_entry(void)
2125 {
2126         debugfs_remove(jbd2_debug);
2127         debugfs_remove(jbd2_debugfs_dir);
2128 }
2129
2130 #else
2131
2132 static void __init jbd2_create_debugfs_entry(void)
2133 {
2134 }
2135
2136 static void __exit jbd2_remove_debugfs_entry(void)
2137 {
2138 }
2139
2140 #endif
2141
2142 #ifdef CONFIG_PROC_FS
2143
2144 #define JBD2_STATS_PROC_NAME "fs/jbd2"
2145
2146 static void __init jbd2_create_jbd_stats_proc_entry(void)
2147 {
2148         proc_jbd2_stats = proc_mkdir(JBD2_STATS_PROC_NAME, NULL);
2149 }
2150
2151 static void __exit jbd2_remove_jbd_stats_proc_entry(void)
2152 {
2153         if (proc_jbd2_stats)
2154                 remove_proc_entry(JBD2_STATS_PROC_NAME, NULL);
2155 }
2156
2157 #else
2158
2159 #define jbd2_create_jbd_stats_proc_entry() do {} while (0)
2160 #define jbd2_remove_jbd_stats_proc_entry() do {} while (0)
2161
2162 #endif
2163
2164 struct kmem_cache *jbd2_handle_cache;
2165
2166 static int __init journal_init_handle_cache(void)
2167 {
2168         jbd2_handle_cache = kmem_cache_create("jbd2_journal_handle",
2169                                 sizeof(handle_t),
2170                                 0,              /* offset */
2171                                 SLAB_TEMPORARY, /* flags */
2172                                 NULL);          /* ctor */
2173         if (jbd2_handle_cache == NULL) {
2174                 printk(KERN_EMERG "JBD: failed to create handle cache\n");
2175                 return -ENOMEM;
2176         }
2177         return 0;
2178 }
2179
2180 static void jbd2_journal_destroy_handle_cache(void)
2181 {
2182         if (jbd2_handle_cache)
2183                 kmem_cache_destroy(jbd2_handle_cache);
2184 }
2185
2186 /*
2187  * Module startup and shutdown
2188  */
2189
2190 static int __init journal_init_caches(void)
2191 {
2192         int ret;
2193
2194         ret = jbd2_journal_init_revoke_caches();
2195         if (ret == 0)
2196                 ret = journal_init_jbd2_journal_head_cache();
2197         if (ret == 0)
2198                 ret = journal_init_handle_cache();
2199         return ret;
2200 }
2201
2202 static void jbd2_journal_destroy_caches(void)
2203 {
2204         jbd2_journal_destroy_revoke_caches();
2205         jbd2_journal_destroy_jbd2_journal_head_cache();
2206         jbd2_journal_destroy_handle_cache();
2207 }
2208
2209 static int __init journal_init(void)
2210 {
2211         int ret;
2212
2213         BUILD_BUG_ON(sizeof(struct journal_superblock_s) != 1024);
2214
2215         ret = journal_init_caches();
2216         if (ret == 0) {
2217                 jbd2_create_debugfs_entry();
2218                 jbd2_create_jbd_stats_proc_entry();
2219         } else {
2220                 jbd2_journal_destroy_caches();
2221         }
2222         return ret;
2223 }
2224
2225 static void __exit journal_exit(void)
2226 {
2227 #ifdef CONFIG_JBD2_DEBUG
2228         int n = atomic_read(&nr_journal_heads);
2229         if (n)
2230                 printk(KERN_EMERG "JBD: leaked %d journal_heads!\n", n);
2231 #endif
2232         jbd2_remove_debugfs_entry();
2233         jbd2_remove_jbd_stats_proc_entry();
2234         jbd2_journal_destroy_caches();
2235 }
2236
2237 /* 
2238  * jbd2_dev_to_name is a utility function used by the jbd2 and ext4 
2239  * tracing infrastructure to map a dev_t to a device name.
2240  *
2241  * The caller should use rcu_read_lock() in order to make sure the
2242  * device name stays valid until its done with it.  We use
2243  * rcu_read_lock() as well to make sure we're safe in case the caller
2244  * gets sloppy, and because rcu_read_lock() is cheap and can be safely
2245  * nested.
2246  */
2247 struct devname_cache {
2248         struct rcu_head rcu;
2249         dev_t           device;
2250         char            devname[BDEVNAME_SIZE];
2251 };
2252 #define CACHE_SIZE_BITS 6
2253 static struct devname_cache *devcache[1 << CACHE_SIZE_BITS];
2254 static DEFINE_SPINLOCK(devname_cache_lock);
2255
2256 static void free_devcache(struct rcu_head *rcu)
2257 {
2258         kfree(rcu);
2259 }
2260
2261 const char *jbd2_dev_to_name(dev_t device)
2262 {
2263         int     i = hash_32(device, CACHE_SIZE_BITS);
2264         char    *ret;
2265         struct block_device *bd;
2266         static struct devname_cache *new_dev;
2267
2268         rcu_read_lock();
2269         if (devcache[i] && devcache[i]->device == device) {
2270                 ret = devcache[i]->devname;
2271                 rcu_read_unlock();
2272                 return ret;
2273         }
2274         rcu_read_unlock();
2275
2276         new_dev = kmalloc(sizeof(struct devname_cache), GFP_KERNEL);
2277         if (!new_dev)
2278                 return "NODEV-ALLOCFAILURE"; /* Something non-NULL */
2279         spin_lock(&devname_cache_lock);
2280         if (devcache[i]) {
2281                 if (devcache[i]->device == device) {
2282                         kfree(new_dev);
2283                         ret = devcache[i]->devname;
2284                         spin_unlock(&devname_cache_lock);
2285                         return ret;
2286                 }
2287                 call_rcu(&devcache[i]->rcu, free_devcache);
2288         }
2289         devcache[i] = new_dev;
2290         devcache[i]->device = device;
2291         bd = bdget(device);
2292         if (bd) {
2293                 bdevname(bd, devcache[i]->devname);
2294                 bdput(bd);
2295         } else
2296                 __bdevname(device, devcache[i]->devname);
2297         ret = devcache[i]->devname;
2298         spin_unlock(&devname_cache_lock);
2299         return ret;
2300 }
2301 EXPORT_SYMBOL(jbd2_dev_to_name);
2302
2303 MODULE_LICENSE("GPL");
2304 module_init(journal_init);
2305 module_exit(journal_exit);
2306