Merge branch 'for-linus' of master.kernel.org:/pub/scm/linux/kernel/git/roland/infiniband
[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
40 #include <asm/uaccess.h>
41 #include <asm/page.h>
42
43 EXPORT_SYMBOL(jbd2_journal_start);
44 EXPORT_SYMBOL(jbd2_journal_restart);
45 EXPORT_SYMBOL(jbd2_journal_extend);
46 EXPORT_SYMBOL(jbd2_journal_stop);
47 EXPORT_SYMBOL(jbd2_journal_lock_updates);
48 EXPORT_SYMBOL(jbd2_journal_unlock_updates);
49 EXPORT_SYMBOL(jbd2_journal_get_write_access);
50 EXPORT_SYMBOL(jbd2_journal_get_create_access);
51 EXPORT_SYMBOL(jbd2_journal_get_undo_access);
52 EXPORT_SYMBOL(jbd2_journal_dirty_data);
53 EXPORT_SYMBOL(jbd2_journal_dirty_metadata);
54 EXPORT_SYMBOL(jbd2_journal_release_buffer);
55 EXPORT_SYMBOL(jbd2_journal_forget);
56 #if 0
57 EXPORT_SYMBOL(journal_sync_buffer);
58 #endif
59 EXPORT_SYMBOL(jbd2_journal_flush);
60 EXPORT_SYMBOL(jbd2_journal_revoke);
61
62 EXPORT_SYMBOL(jbd2_journal_init_dev);
63 EXPORT_SYMBOL(jbd2_journal_init_inode);
64 EXPORT_SYMBOL(jbd2_journal_update_format);
65 EXPORT_SYMBOL(jbd2_journal_check_used_features);
66 EXPORT_SYMBOL(jbd2_journal_check_available_features);
67 EXPORT_SYMBOL(jbd2_journal_set_features);
68 EXPORT_SYMBOL(jbd2_journal_create);
69 EXPORT_SYMBOL(jbd2_journal_load);
70 EXPORT_SYMBOL(jbd2_journal_destroy);
71 EXPORT_SYMBOL(jbd2_journal_update_superblock);
72 EXPORT_SYMBOL(jbd2_journal_abort);
73 EXPORT_SYMBOL(jbd2_journal_errno);
74 EXPORT_SYMBOL(jbd2_journal_ack_err);
75 EXPORT_SYMBOL(jbd2_journal_clear_err);
76 EXPORT_SYMBOL(jbd2_log_wait_commit);
77 EXPORT_SYMBOL(jbd2_journal_start_commit);
78 EXPORT_SYMBOL(jbd2_journal_force_commit_nested);
79 EXPORT_SYMBOL(jbd2_journal_wipe);
80 EXPORT_SYMBOL(jbd2_journal_blocks_per_page);
81 EXPORT_SYMBOL(jbd2_journal_invalidatepage);
82 EXPORT_SYMBOL(jbd2_journal_try_to_free_buffers);
83 EXPORT_SYMBOL(jbd2_journal_force_commit);
84
85 static int journal_convert_superblock_v1(journal_t *, journal_superblock_t *);
86 static void __journal_abort_soft (journal_t *journal, int errno);
87 static int jbd2_journal_create_jbd_slab(size_t slab_size);
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 != 0);
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 == 0);
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_slab_alloc(bh_in->b_size, GFP_NOFS);
339                 jbd_lock_bh_state(bh_in);
340                 if (jh_in->b_frozen_data) {
341                         jbd2_slab_free(tmp, bh_in->b_size);
342                         goto repeat;
343                 }
344
345                 jh_in->b_frozen_data = tmp;
346                 mapped_data = kmap_atomic(new_page, KM_USER0);
347                 memcpy(tmp, mapped_data + new_offset, jh2bh(jh_in)->b_size);
348                 kunmap_atomic(mapped_data, KM_USER0);
349
350                 new_page = virt_to_page(tmp);
351                 new_offset = offset_in_page(tmp);
352                 done_copy_out = 1;
353         }
354
355         /*
356          * Did we need to do an escaping?  Now we've done all the
357          * copying, we can finally do so.
358          */
359         if (do_escape) {
360                 mapped_data = kmap_atomic(new_page, KM_USER0);
361                 *((unsigned int *)(mapped_data + new_offset)) = 0;
362                 kunmap_atomic(mapped_data, KM_USER0);
363         }
364
365         /* keep subsequent assertions sane */
366         new_bh->b_state = 0;
367         init_buffer(new_bh, NULL, NULL);
368         atomic_set(&new_bh->b_count, 1);
369         jbd_unlock_bh_state(bh_in);
370
371         new_jh = jbd2_journal_add_journal_head(new_bh); /* This sleeps */
372
373         set_bh_page(new_bh, new_page, new_offset);
374         new_jh->b_transaction = NULL;
375         new_bh->b_size = jh2bh(jh_in)->b_size;
376         new_bh->b_bdev = transaction->t_journal->j_dev;
377         new_bh->b_blocknr = blocknr;
378         set_buffer_mapped(new_bh);
379         set_buffer_dirty(new_bh);
380
381         *jh_out = new_jh;
382
383         /*
384          * The to-be-written buffer needs to get moved to the io queue,
385          * and the original buffer whose contents we are shadowing or
386          * copying is moved to the transaction's shadow queue.
387          */
388         JBUFFER_TRACE(jh_in, "file as BJ_Shadow");
389         jbd2_journal_file_buffer(jh_in, transaction, BJ_Shadow);
390         JBUFFER_TRACE(new_jh, "file as BJ_IO");
391         jbd2_journal_file_buffer(new_jh, transaction, BJ_IO);
392
393         return do_escape | (done_copy_out << 1);
394 }
395
396 /*
397  * Allocation code for the journal file.  Manage the space left in the
398  * journal, so that we can begin checkpointing when appropriate.
399  */
400
401 /*
402  * __jbd2_log_space_left: Return the number of free blocks left in the journal.
403  *
404  * Called with the journal already locked.
405  *
406  * Called under j_state_lock
407  */
408
409 int __jbd2_log_space_left(journal_t *journal)
410 {
411         int left = journal->j_free;
412
413         assert_spin_locked(&journal->j_state_lock);
414
415         /*
416          * Be pessimistic here about the number of those free blocks which
417          * might be required for log descriptor control blocks.
418          */
419
420 #define MIN_LOG_RESERVED_BLOCKS 32 /* Allow for rounding errors */
421
422         left -= MIN_LOG_RESERVED_BLOCKS;
423
424         if (left <= 0)
425                 return 0;
426         left -= (left >> 3);
427         return left;
428 }
429
430 /*
431  * Called under j_state_lock.  Returns true if a transaction was started.
432  */
433 int __jbd2_log_start_commit(journal_t *journal, tid_t target)
434 {
435         /*
436          * Are we already doing a recent enough commit?
437          */
438         if (!tid_geq(journal->j_commit_request, target)) {
439                 /*
440                  * We want a new commit: OK, mark the request and wakup the
441                  * commit thread.  We do _not_ do the commit ourselves.
442                  */
443
444                 journal->j_commit_request = target;
445                 jbd_debug(1, "JBD: requesting commit %d/%d\n",
446                           journal->j_commit_request,
447                           journal->j_commit_sequence);
448                 wake_up(&journal->j_wait_commit);
449                 return 1;
450         }
451         return 0;
452 }
453
454 int jbd2_log_start_commit(journal_t *journal, tid_t tid)
455 {
456         int ret;
457
458         spin_lock(&journal->j_state_lock);
459         ret = __jbd2_log_start_commit(journal, tid);
460         spin_unlock(&journal->j_state_lock);
461         return ret;
462 }
463
464 /*
465  * Force and wait upon a commit if the calling process is not within
466  * transaction.  This is used for forcing out undo-protected data which contains
467  * bitmaps, when the fs is running out of space.
468  *
469  * We can only force the running transaction if we don't have an active handle;
470  * otherwise, we will deadlock.
471  *
472  * Returns true if a transaction was started.
473  */
474 int jbd2_journal_force_commit_nested(journal_t *journal)
475 {
476         transaction_t *transaction = NULL;
477         tid_t tid;
478
479         spin_lock(&journal->j_state_lock);
480         if (journal->j_running_transaction && !current->journal_info) {
481                 transaction = journal->j_running_transaction;
482                 __jbd2_log_start_commit(journal, transaction->t_tid);
483         } else if (journal->j_committing_transaction)
484                 transaction = journal->j_committing_transaction;
485
486         if (!transaction) {
487                 spin_unlock(&journal->j_state_lock);
488                 return 0;       /* Nothing to retry */
489         }
490
491         tid = transaction->t_tid;
492         spin_unlock(&journal->j_state_lock);
493         jbd2_log_wait_commit(journal, tid);
494         return 1;
495 }
496
497 /*
498  * Start a commit of the current running transaction (if any).  Returns true
499  * if a transaction was started, and fills its tid in at *ptid
500  */
501 int jbd2_journal_start_commit(journal_t *journal, tid_t *ptid)
502 {
503         int ret = 0;
504
505         spin_lock(&journal->j_state_lock);
506         if (journal->j_running_transaction) {
507                 tid_t tid = journal->j_running_transaction->t_tid;
508
509                 ret = __jbd2_log_start_commit(journal, tid);
510                 if (ret && ptid)
511                         *ptid = tid;
512         } else if (journal->j_committing_transaction && ptid) {
513                 /*
514                  * If ext3_write_super() recently started a commit, then we
515                  * have to wait for completion of that transaction
516                  */
517                 *ptid = journal->j_committing_transaction->t_tid;
518                 ret = 1;
519         }
520         spin_unlock(&journal->j_state_lock);
521         return ret;
522 }
523
524 /*
525  * Wait for a specified commit to complete.
526  * The caller may not hold the journal lock.
527  */
528 int jbd2_log_wait_commit(journal_t *journal, tid_t tid)
529 {
530         int err = 0;
531
532 #ifdef CONFIG_JBD2_DEBUG
533         spin_lock(&journal->j_state_lock);
534         if (!tid_geq(journal->j_commit_request, tid)) {
535                 printk(KERN_EMERG
536                        "%s: error: j_commit_request=%d, tid=%d\n",
537                        __FUNCTION__, journal->j_commit_request, tid);
538         }
539         spin_unlock(&journal->j_state_lock);
540 #endif
541         spin_lock(&journal->j_state_lock);
542         while (tid_gt(tid, journal->j_commit_sequence)) {
543                 jbd_debug(1, "JBD: want %d, j_commit_sequence=%d\n",
544                                   tid, journal->j_commit_sequence);
545                 wake_up(&journal->j_wait_commit);
546                 spin_unlock(&journal->j_state_lock);
547                 wait_event(journal->j_wait_done_commit,
548                                 !tid_gt(tid, journal->j_commit_sequence));
549                 spin_lock(&journal->j_state_lock);
550         }
551         spin_unlock(&journal->j_state_lock);
552
553         if (unlikely(is_journal_aborted(journal))) {
554                 printk(KERN_EMERG "journal commit I/O error\n");
555                 err = -EIO;
556         }
557         return err;
558 }
559
560 /*
561  * Log buffer allocation routines:
562  */
563
564 int jbd2_journal_next_log_block(journal_t *journal, unsigned long long *retp)
565 {
566         unsigned long blocknr;
567
568         spin_lock(&journal->j_state_lock);
569         J_ASSERT(journal->j_free > 1);
570
571         blocknr = journal->j_head;
572         journal->j_head++;
573         journal->j_free--;
574         if (journal->j_head == journal->j_last)
575                 journal->j_head = journal->j_first;
576         spin_unlock(&journal->j_state_lock);
577         return jbd2_journal_bmap(journal, blocknr, retp);
578 }
579
580 /*
581  * Conversion of logical to physical block numbers for the journal
582  *
583  * On external journals the journal blocks are identity-mapped, so
584  * this is a no-op.  If needed, we can use j_blk_offset - everything is
585  * ready.
586  */
587 int jbd2_journal_bmap(journal_t *journal, unsigned long blocknr,
588                  unsigned long long *retp)
589 {
590         int err = 0;
591         unsigned long long ret;
592
593         if (journal->j_inode) {
594                 ret = bmap(journal->j_inode, blocknr);
595                 if (ret)
596                         *retp = ret;
597                 else {
598                         char b[BDEVNAME_SIZE];
599
600                         printk(KERN_ALERT "%s: journal block not found "
601                                         "at offset %lu on %s\n",
602                                 __FUNCTION__,
603                                 blocknr,
604                                 bdevname(journal->j_dev, b));
605                         err = -EIO;
606                         __journal_abort_soft(journal, err);
607                 }
608         } else {
609                 *retp = blocknr; /* +journal->j_blk_offset */
610         }
611         return err;
612 }
613
614 /*
615  * We play buffer_head aliasing tricks to write data/metadata blocks to
616  * the journal without copying their contents, but for journal
617  * descriptor blocks we do need to generate bona fide buffers.
618  *
619  * After the caller of jbd2_journal_get_descriptor_buffer() has finished modifying
620  * the buffer's contents they really should run flush_dcache_page(bh->b_page).
621  * But we don't bother doing that, so there will be coherency problems with
622  * mmaps of blockdevs which hold live JBD-controlled filesystems.
623  */
624 struct journal_head *jbd2_journal_get_descriptor_buffer(journal_t *journal)
625 {
626         struct buffer_head *bh;
627         unsigned long long blocknr;
628         int err;
629
630         err = jbd2_journal_next_log_block(journal, &blocknr);
631
632         if (err)
633                 return NULL;
634
635         bh = __getblk(journal->j_dev, blocknr, journal->j_blocksize);
636         lock_buffer(bh);
637         memset(bh->b_data, 0, journal->j_blocksize);
638         set_buffer_uptodate(bh);
639         unlock_buffer(bh);
640         BUFFER_TRACE(bh, "return this buffer");
641         return jbd2_journal_add_journal_head(bh);
642 }
643
644 /*
645  * Management for journal control blocks: functions to create and
646  * destroy journal_t structures, and to initialise and read existing
647  * journal blocks from disk.  */
648
649 /* First: create and setup a journal_t object in memory.  We initialise
650  * very few fields yet: that has to wait until we have created the
651  * journal structures from from scratch, or loaded them from disk. */
652
653 static journal_t * journal_init_common (void)
654 {
655         journal_t *journal;
656         int err;
657
658         journal = jbd_kmalloc(sizeof(*journal), GFP_KERNEL);
659         if (!journal)
660                 goto fail;
661         memset(journal, 0, sizeof(*journal));
662
663         init_waitqueue_head(&journal->j_wait_transaction_locked);
664         init_waitqueue_head(&journal->j_wait_logspace);
665         init_waitqueue_head(&journal->j_wait_done_commit);
666         init_waitqueue_head(&journal->j_wait_checkpoint);
667         init_waitqueue_head(&journal->j_wait_commit);
668         init_waitqueue_head(&journal->j_wait_updates);
669         mutex_init(&journal->j_barrier);
670         mutex_init(&journal->j_checkpoint_mutex);
671         spin_lock_init(&journal->j_revoke_lock);
672         spin_lock_init(&journal->j_list_lock);
673         spin_lock_init(&journal->j_state_lock);
674
675         journal->j_commit_interval = (HZ * JBD_DEFAULT_MAX_COMMIT_AGE);
676
677         /* The journal is marked for error until we succeed with recovery! */
678         journal->j_flags = JBD2_ABORT;
679
680         /* Set up a default-sized revoke table for the new mount. */
681         err = jbd2_journal_init_revoke(journal, JOURNAL_REVOKE_DEFAULT_HASH);
682         if (err) {
683                 kfree(journal);
684                 goto fail;
685         }
686         return journal;
687 fail:
688         return NULL;
689 }
690
691 /* jbd2_journal_init_dev and jbd2_journal_init_inode:
692  *
693  * Create a journal structure assigned some fixed set of disk blocks to
694  * the journal.  We don't actually touch those disk blocks yet, but we
695  * need to set up all of the mapping information to tell the journaling
696  * system where the journal blocks are.
697  *
698  */
699
700 /**
701  *  journal_t * jbd2_journal_init_dev() - creates an initialises a journal structure
702  *  @bdev: Block device on which to create the journal
703  *  @fs_dev: Device which hold journalled filesystem for this journal.
704  *  @start: Block nr Start of journal.
705  *  @len:  Length of the journal in blocks.
706  *  @blocksize: blocksize of journalling device
707  *  @returns: a newly created journal_t *
708  *
709  *  jbd2_journal_init_dev creates a journal which maps a fixed contiguous
710  *  range of blocks on an arbitrary block device.
711  *
712  */
713 journal_t * jbd2_journal_init_dev(struct block_device *bdev,
714                         struct block_device *fs_dev,
715                         unsigned long long start, int len, int blocksize)
716 {
717         journal_t *journal = journal_init_common();
718         struct buffer_head *bh;
719         int n;
720
721         if (!journal)
722                 return NULL;
723
724         /* journal descriptor can store up to n blocks -bzzz */
725         journal->j_blocksize = blocksize;
726         n = journal->j_blocksize / sizeof(journal_block_tag_t);
727         journal->j_wbufsize = n;
728         journal->j_wbuf = kmalloc(n * sizeof(struct buffer_head*), GFP_KERNEL);
729         if (!journal->j_wbuf) {
730                 printk(KERN_ERR "%s: Cant allocate bhs for commit thread\n",
731                         __FUNCTION__);
732                 kfree(journal);
733                 journal = NULL;
734                 goto out;
735         }
736         journal->j_dev = bdev;
737         journal->j_fs_dev = fs_dev;
738         journal->j_blk_offset = start;
739         journal->j_maxlen = len;
740
741         bh = __getblk(journal->j_dev, start, journal->j_blocksize);
742         J_ASSERT(bh != NULL);
743         journal->j_sb_buffer = bh;
744         journal->j_superblock = (journal_superblock_t *)bh->b_data;
745 out:
746         return journal;
747 }
748
749 /**
750  *  journal_t * jbd2_journal_init_inode () - creates a journal which maps to a inode.
751  *  @inode: An inode to create the journal in
752  *
753  * jbd2_journal_init_inode creates a journal which maps an on-disk inode as
754  * the journal.  The inode must exist already, must support bmap() and
755  * must have all data blocks preallocated.
756  */
757 journal_t * jbd2_journal_init_inode (struct inode *inode)
758 {
759         struct buffer_head *bh;
760         journal_t *journal = journal_init_common();
761         int err;
762         int n;
763         unsigned long long blocknr;
764
765         if (!journal)
766                 return NULL;
767
768         journal->j_dev = journal->j_fs_dev = inode->i_sb->s_bdev;
769         journal->j_inode = inode;
770         jbd_debug(1,
771                   "journal %p: inode %s/%ld, size %Ld, bits %d, blksize %ld\n",
772                   journal, inode->i_sb->s_id, inode->i_ino,
773                   (long long) inode->i_size,
774                   inode->i_sb->s_blocksize_bits, inode->i_sb->s_blocksize);
775
776         journal->j_maxlen = inode->i_size >> inode->i_sb->s_blocksize_bits;
777         journal->j_blocksize = inode->i_sb->s_blocksize;
778
779         /* journal descriptor can store up to n blocks -bzzz */
780         n = journal->j_blocksize / sizeof(journal_block_tag_t);
781         journal->j_wbufsize = n;
782         journal->j_wbuf = kmalloc(n * sizeof(struct buffer_head*), GFP_KERNEL);
783         if (!journal->j_wbuf) {
784                 printk(KERN_ERR "%s: Cant allocate bhs for commit thread\n",
785                         __FUNCTION__);
786                 kfree(journal);
787                 return NULL;
788         }
789
790         err = jbd2_journal_bmap(journal, 0, &blocknr);
791         /* If that failed, give up */
792         if (err) {
793                 printk(KERN_ERR "%s: Cannnot locate journal superblock\n",
794                        __FUNCTION__);
795                 kfree(journal);
796                 return NULL;
797         }
798
799         bh = __getblk(journal->j_dev, blocknr, journal->j_blocksize);
800         J_ASSERT(bh != NULL);
801         journal->j_sb_buffer = bh;
802         journal->j_superblock = (journal_superblock_t *)bh->b_data;
803
804         return journal;
805 }
806
807 /*
808  * If the journal init or create aborts, we need to mark the journal
809  * superblock as being NULL to prevent the journal destroy from writing
810  * back a bogus superblock.
811  */
812 static void journal_fail_superblock (journal_t *journal)
813 {
814         struct buffer_head *bh = journal->j_sb_buffer;
815         brelse(bh);
816         journal->j_sb_buffer = NULL;
817 }
818
819 /*
820  * Given a journal_t structure, initialise the various fields for
821  * startup of a new journaling session.  We use this both when creating
822  * a journal, and after recovering an old journal to reset it for
823  * subsequent use.
824  */
825
826 static int journal_reset(journal_t *journal)
827 {
828         journal_superblock_t *sb = journal->j_superblock;
829         unsigned long long first, last;
830
831         first = be32_to_cpu(sb->s_first);
832         last = be32_to_cpu(sb->s_maxlen);
833
834         journal->j_first = first;
835         journal->j_last = last;
836
837         journal->j_head = first;
838         journal->j_tail = first;
839         journal->j_free = last - first;
840
841         journal->j_tail_sequence = journal->j_transaction_sequence;
842         journal->j_commit_sequence = journal->j_transaction_sequence - 1;
843         journal->j_commit_request = journal->j_commit_sequence;
844
845         journal->j_max_transaction_buffers = journal->j_maxlen / 4;
846
847         /* Add the dynamic fields and write it to disk. */
848         jbd2_journal_update_superblock(journal, 1);
849         return jbd2_journal_start_thread(journal);
850 }
851
852 /**
853  * int jbd2_journal_create() - Initialise the new journal file
854  * @journal: Journal to create. This structure must have been initialised
855  *
856  * Given a journal_t structure which tells us which disk blocks we can
857  * use, create a new journal superblock and initialise all of the
858  * journal fields from scratch.
859  **/
860 int jbd2_journal_create(journal_t *journal)
861 {
862         unsigned long long blocknr;
863         struct buffer_head *bh;
864         journal_superblock_t *sb;
865         int i, err;
866
867         if (journal->j_maxlen < JBD2_MIN_JOURNAL_BLOCKS) {
868                 printk (KERN_ERR "Journal length (%d blocks) too short.\n",
869                         journal->j_maxlen);
870                 journal_fail_superblock(journal);
871                 return -EINVAL;
872         }
873
874         if (journal->j_inode == NULL) {
875                 /*
876                  * We don't know what block to start at!
877                  */
878                 printk(KERN_EMERG
879                        "%s: creation of journal on external device!\n",
880                        __FUNCTION__);
881                 BUG();
882         }
883
884         /* Zero out the entire journal on disk.  We cannot afford to
885            have any blocks on disk beginning with JBD2_MAGIC_NUMBER. */
886         jbd_debug(1, "JBD: Zeroing out journal blocks...\n");
887         for (i = 0; i < journal->j_maxlen; i++) {
888                 err = jbd2_journal_bmap(journal, i, &blocknr);
889                 if (err)
890                         return err;
891                 bh = __getblk(journal->j_dev, blocknr, journal->j_blocksize);
892                 lock_buffer(bh);
893                 memset (bh->b_data, 0, journal->j_blocksize);
894                 BUFFER_TRACE(bh, "marking dirty");
895                 mark_buffer_dirty(bh);
896                 BUFFER_TRACE(bh, "marking uptodate");
897                 set_buffer_uptodate(bh);
898                 unlock_buffer(bh);
899                 __brelse(bh);
900         }
901
902         sync_blockdev(journal->j_dev);
903         jbd_debug(1, "JBD: journal cleared.\n");
904
905         /* OK, fill in the initial static fields in the new superblock */
906         sb = journal->j_superblock;
907
908         sb->s_header.h_magic     = cpu_to_be32(JBD2_MAGIC_NUMBER);
909         sb->s_header.h_blocktype = cpu_to_be32(JBD2_SUPERBLOCK_V2);
910
911         sb->s_blocksize = cpu_to_be32(journal->j_blocksize);
912         sb->s_maxlen    = cpu_to_be32(journal->j_maxlen);
913         sb->s_first     = cpu_to_be32(1);
914
915         journal->j_transaction_sequence = 1;
916
917         journal->j_flags &= ~JBD2_ABORT;
918         journal->j_format_version = 2;
919
920         return journal_reset(journal);
921 }
922
923 /**
924  * void jbd2_journal_update_superblock() - Update journal sb on disk.
925  * @journal: The journal to update.
926  * @wait: Set to '0' if you don't want to wait for IO completion.
927  *
928  * Update a journal's dynamic superblock fields and write it to disk,
929  * optionally waiting for the IO to complete.
930  */
931 void jbd2_journal_update_superblock(journal_t *journal, int wait)
932 {
933         journal_superblock_t *sb = journal->j_superblock;
934         struct buffer_head *bh = journal->j_sb_buffer;
935
936         /*
937          * As a special case, if the on-disk copy is already marked as needing
938          * no recovery (s_start == 0) and there are no outstanding transactions
939          * in the filesystem, then we can safely defer the superblock update
940          * until the next commit by setting JBD2_FLUSHED.  This avoids
941          * attempting a write to a potential-readonly device.
942          */
943         if (sb->s_start == 0 && journal->j_tail_sequence ==
944                                 journal->j_transaction_sequence) {
945                 jbd_debug(1,"JBD: Skipping superblock update on recovered sb "
946                         "(start %ld, seq %d, errno %d)\n",
947                         journal->j_tail, journal->j_tail_sequence,
948                         journal->j_errno);
949                 goto out;
950         }
951
952         spin_lock(&journal->j_state_lock);
953         jbd_debug(1,"JBD: updating superblock (start %ld, seq %d, errno %d)\n",
954                   journal->j_tail, journal->j_tail_sequence, journal->j_errno);
955
956         sb->s_sequence = cpu_to_be32(journal->j_tail_sequence);
957         sb->s_start    = cpu_to_be32(journal->j_tail);
958         sb->s_errno    = cpu_to_be32(journal->j_errno);
959         spin_unlock(&journal->j_state_lock);
960
961         BUFFER_TRACE(bh, "marking dirty");
962         mark_buffer_dirty(bh);
963         if (wait)
964                 sync_dirty_buffer(bh);
965         else
966                 ll_rw_block(SWRITE, 1, &bh);
967
968 out:
969         /* If we have just flushed the log (by marking s_start==0), then
970          * any future commit will have to be careful to update the
971          * superblock again to re-record the true start of the log. */
972
973         spin_lock(&journal->j_state_lock);
974         if (sb->s_start)
975                 journal->j_flags &= ~JBD2_FLUSHED;
976         else
977                 journal->j_flags |= JBD2_FLUSHED;
978         spin_unlock(&journal->j_state_lock);
979 }
980
981 /*
982  * Read the superblock for a given journal, performing initial
983  * validation of the format.
984  */
985
986 static int journal_get_superblock(journal_t *journal)
987 {
988         struct buffer_head *bh;
989         journal_superblock_t *sb;
990         int err = -EIO;
991
992         bh = journal->j_sb_buffer;
993
994         J_ASSERT(bh != NULL);
995         if (!buffer_uptodate(bh)) {
996                 ll_rw_block(READ, 1, &bh);
997                 wait_on_buffer(bh);
998                 if (!buffer_uptodate(bh)) {
999                         printk (KERN_ERR
1000                                 "JBD: IO error reading journal superblock\n");
1001                         goto out;
1002                 }
1003         }
1004
1005         sb = journal->j_superblock;
1006
1007         err = -EINVAL;
1008
1009         if (sb->s_header.h_magic != cpu_to_be32(JBD2_MAGIC_NUMBER) ||
1010             sb->s_blocksize != cpu_to_be32(journal->j_blocksize)) {
1011                 printk(KERN_WARNING "JBD: no valid journal superblock found\n");
1012                 goto out;
1013         }
1014
1015         switch(be32_to_cpu(sb->s_header.h_blocktype)) {
1016         case JBD2_SUPERBLOCK_V1:
1017                 journal->j_format_version = 1;
1018                 break;
1019         case JBD2_SUPERBLOCK_V2:
1020                 journal->j_format_version = 2;
1021                 break;
1022         default:
1023                 printk(KERN_WARNING "JBD: unrecognised superblock format ID\n");
1024                 goto out;
1025         }
1026
1027         if (be32_to_cpu(sb->s_maxlen) < journal->j_maxlen)
1028                 journal->j_maxlen = be32_to_cpu(sb->s_maxlen);
1029         else if (be32_to_cpu(sb->s_maxlen) > journal->j_maxlen) {
1030                 printk (KERN_WARNING "JBD: journal file too short\n");
1031                 goto out;
1032         }
1033
1034         return 0;
1035
1036 out:
1037         journal_fail_superblock(journal);
1038         return err;
1039 }
1040
1041 /*
1042  * Load the on-disk journal superblock and read the key fields into the
1043  * journal_t.
1044  */
1045
1046 static int load_superblock(journal_t *journal)
1047 {
1048         int err;
1049         journal_superblock_t *sb;
1050
1051         err = journal_get_superblock(journal);
1052         if (err)
1053                 return err;
1054
1055         sb = journal->j_superblock;
1056
1057         journal->j_tail_sequence = be32_to_cpu(sb->s_sequence);
1058         journal->j_tail = be32_to_cpu(sb->s_start);
1059         journal->j_first = be32_to_cpu(sb->s_first);
1060         journal->j_last = be32_to_cpu(sb->s_maxlen);
1061         journal->j_errno = be32_to_cpu(sb->s_errno);
1062
1063         return 0;
1064 }
1065
1066
1067 /**
1068  * int jbd2_journal_load() - Read journal from disk.
1069  * @journal: Journal to act on.
1070  *
1071  * Given a journal_t structure which tells us which disk blocks contain
1072  * a journal, read the journal from disk to initialise the in-memory
1073  * structures.
1074  */
1075 int jbd2_journal_load(journal_t *journal)
1076 {
1077         int err;
1078         journal_superblock_t *sb;
1079
1080         err = load_superblock(journal);
1081         if (err)
1082                 return err;
1083
1084         sb = journal->j_superblock;
1085         /* If this is a V2 superblock, then we have to check the
1086          * features flags on it. */
1087
1088         if (journal->j_format_version >= 2) {
1089                 if ((sb->s_feature_ro_compat &
1090                      ~cpu_to_be32(JBD2_KNOWN_ROCOMPAT_FEATURES)) ||
1091                     (sb->s_feature_incompat &
1092                      ~cpu_to_be32(JBD2_KNOWN_INCOMPAT_FEATURES))) {
1093                         printk (KERN_WARNING
1094                                 "JBD: Unrecognised features on journal\n");
1095                         return -EINVAL;
1096                 }
1097         }
1098
1099         /*
1100          * Create a slab for this blocksize
1101          */
1102         err = jbd2_journal_create_jbd_slab(be32_to_cpu(sb->s_blocksize));
1103         if (err)
1104                 return err;
1105
1106         /* Let the recovery code check whether it needs to recover any
1107          * data from the journal. */
1108         if (jbd2_journal_recover(journal))
1109                 goto recovery_error;
1110
1111         /* OK, we've finished with the dynamic journal bits:
1112          * reinitialise the dynamic contents of the superblock in memory
1113          * and reset them on disk. */
1114         if (journal_reset(journal))
1115                 goto recovery_error;
1116
1117         journal->j_flags &= ~JBD2_ABORT;
1118         journal->j_flags |= JBD2_LOADED;
1119         return 0;
1120
1121 recovery_error:
1122         printk (KERN_WARNING "JBD: recovery failed\n");
1123         return -EIO;
1124 }
1125
1126 /**
1127  * void jbd2_journal_destroy() - Release a journal_t structure.
1128  * @journal: Journal to act on.
1129  *
1130  * Release a journal_t structure once it is no longer in use by the
1131  * journaled object.
1132  */
1133 void jbd2_journal_destroy(journal_t *journal)
1134 {
1135         /* Wait for the commit thread to wake up and die. */
1136         journal_kill_thread(journal);
1137
1138         /* Force a final log commit */
1139         if (journal->j_running_transaction)
1140                 jbd2_journal_commit_transaction(journal);
1141
1142         /* Force any old transactions to disk */
1143
1144         /* Totally anal locking here... */
1145         spin_lock(&journal->j_list_lock);
1146         while (journal->j_checkpoint_transactions != NULL) {
1147                 spin_unlock(&journal->j_list_lock);
1148                 jbd2_log_do_checkpoint(journal);
1149                 spin_lock(&journal->j_list_lock);
1150         }
1151
1152         J_ASSERT(journal->j_running_transaction == NULL);
1153         J_ASSERT(journal->j_committing_transaction == NULL);
1154         J_ASSERT(journal->j_checkpoint_transactions == NULL);
1155         spin_unlock(&journal->j_list_lock);
1156
1157         /* We can now mark the journal as empty. */
1158         journal->j_tail = 0;
1159         journal->j_tail_sequence = ++journal->j_transaction_sequence;
1160         if (journal->j_sb_buffer) {
1161                 jbd2_journal_update_superblock(journal, 1);
1162                 brelse(journal->j_sb_buffer);
1163         }
1164
1165         if (journal->j_inode)
1166                 iput(journal->j_inode);
1167         if (journal->j_revoke)
1168                 jbd2_journal_destroy_revoke(journal);
1169         kfree(journal->j_wbuf);
1170         kfree(journal);
1171 }
1172
1173
1174 /**
1175  *int jbd2_journal_check_used_features () - Check if features specified are used.
1176  * @journal: Journal to check.
1177  * @compat: bitmask of compatible features
1178  * @ro: bitmask of features that force read-only mount
1179  * @incompat: bitmask of incompatible features
1180  *
1181  * Check whether the journal uses all of a given set of
1182  * features.  Return true (non-zero) if it does.
1183  **/
1184
1185 int jbd2_journal_check_used_features (journal_t *journal, unsigned long compat,
1186                                  unsigned long ro, unsigned long incompat)
1187 {
1188         journal_superblock_t *sb;
1189
1190         if (!compat && !ro && !incompat)
1191                 return 1;
1192         if (journal->j_format_version == 1)
1193                 return 0;
1194
1195         sb = journal->j_superblock;
1196
1197         if (((be32_to_cpu(sb->s_feature_compat) & compat) == compat) &&
1198             ((be32_to_cpu(sb->s_feature_ro_compat) & ro) == ro) &&
1199             ((be32_to_cpu(sb->s_feature_incompat) & incompat) == incompat))
1200                 return 1;
1201
1202         return 0;
1203 }
1204
1205 /**
1206  * int jbd2_journal_check_available_features() - Check feature set in journalling layer
1207  * @journal: Journal to check.
1208  * @compat: bitmask of compatible features
1209  * @ro: bitmask of features that force read-only mount
1210  * @incompat: bitmask of incompatible features
1211  *
1212  * Check whether the journaling code supports the use of
1213  * all of a given set of features on this journal.  Return true
1214  * (non-zero) if it can. */
1215
1216 int jbd2_journal_check_available_features (journal_t *journal, unsigned long compat,
1217                                       unsigned long ro, unsigned long incompat)
1218 {
1219         journal_superblock_t *sb;
1220
1221         if (!compat && !ro && !incompat)
1222                 return 1;
1223
1224         sb = journal->j_superblock;
1225
1226         /* We can support any known requested features iff the
1227          * superblock is in version 2.  Otherwise we fail to support any
1228          * extended sb features. */
1229
1230         if (journal->j_format_version != 2)
1231                 return 0;
1232
1233         if ((compat   & JBD2_KNOWN_COMPAT_FEATURES) == compat &&
1234             (ro       & JBD2_KNOWN_ROCOMPAT_FEATURES) == ro &&
1235             (incompat & JBD2_KNOWN_INCOMPAT_FEATURES) == incompat)
1236                 return 1;
1237
1238         return 0;
1239 }
1240
1241 /**
1242  * int jbd2_journal_set_features () - Mark a given journal feature in the superblock
1243  * @journal: Journal to act on.
1244  * @compat: bitmask of compatible features
1245  * @ro: bitmask of features that force read-only mount
1246  * @incompat: bitmask of incompatible features
1247  *
1248  * Mark a given journal feature as present on the
1249  * superblock.  Returns true if the requested features could be set.
1250  *
1251  */
1252
1253 int jbd2_journal_set_features (journal_t *journal, unsigned long compat,
1254                           unsigned long ro, unsigned long incompat)
1255 {
1256         journal_superblock_t *sb;
1257
1258         if (jbd2_journal_check_used_features(journal, compat, ro, incompat))
1259                 return 1;
1260
1261         if (!jbd2_journal_check_available_features(journal, compat, ro, incompat))
1262                 return 0;
1263
1264         jbd_debug(1, "Setting new features 0x%lx/0x%lx/0x%lx\n",
1265                   compat, ro, incompat);
1266
1267         sb = journal->j_superblock;
1268
1269         sb->s_feature_compat    |= cpu_to_be32(compat);
1270         sb->s_feature_ro_compat |= cpu_to_be32(ro);
1271         sb->s_feature_incompat  |= cpu_to_be32(incompat);
1272
1273         return 1;
1274 }
1275
1276
1277 /**
1278  * int jbd2_journal_update_format () - Update on-disk journal structure.
1279  * @journal: Journal to act on.
1280  *
1281  * Given an initialised but unloaded journal struct, poke about in the
1282  * on-disk structure to update it to the most recent supported version.
1283  */
1284 int jbd2_journal_update_format (journal_t *journal)
1285 {
1286         journal_superblock_t *sb;
1287         int err;
1288
1289         err = journal_get_superblock(journal);
1290         if (err)
1291                 return err;
1292
1293         sb = journal->j_superblock;
1294
1295         switch (be32_to_cpu(sb->s_header.h_blocktype)) {
1296         case JBD2_SUPERBLOCK_V2:
1297                 return 0;
1298         case JBD2_SUPERBLOCK_V1:
1299                 return journal_convert_superblock_v1(journal, sb);
1300         default:
1301                 break;
1302         }
1303         return -EINVAL;
1304 }
1305
1306 static int journal_convert_superblock_v1(journal_t *journal,
1307                                          journal_superblock_t *sb)
1308 {
1309         int offset, blocksize;
1310         struct buffer_head *bh;
1311
1312         printk(KERN_WARNING
1313                 "JBD: Converting superblock from version 1 to 2.\n");
1314
1315         /* Pre-initialise new fields to zero */
1316         offset = ((char *) &(sb->s_feature_compat)) - ((char *) sb);
1317         blocksize = be32_to_cpu(sb->s_blocksize);
1318         memset(&sb->s_feature_compat, 0, blocksize-offset);
1319
1320         sb->s_nr_users = cpu_to_be32(1);
1321         sb->s_header.h_blocktype = cpu_to_be32(JBD2_SUPERBLOCK_V2);
1322         journal->j_format_version = 2;
1323
1324         bh = journal->j_sb_buffer;
1325         BUFFER_TRACE(bh, "marking dirty");
1326         mark_buffer_dirty(bh);
1327         sync_dirty_buffer(bh);
1328         return 0;
1329 }
1330
1331
1332 /**
1333  * int jbd2_journal_flush () - Flush journal
1334  * @journal: Journal to act on.
1335  *
1336  * Flush all data for a given journal to disk and empty the journal.
1337  * Filesystems can use this when remounting readonly to ensure that
1338  * recovery does not need to happen on remount.
1339  */
1340
1341 int jbd2_journal_flush(journal_t *journal)
1342 {
1343         int err = 0;
1344         transaction_t *transaction = NULL;
1345         unsigned long old_tail;
1346
1347         spin_lock(&journal->j_state_lock);
1348
1349         /* Force everything buffered to the log... */
1350         if (journal->j_running_transaction) {
1351                 transaction = journal->j_running_transaction;
1352                 __jbd2_log_start_commit(journal, transaction->t_tid);
1353         } else if (journal->j_committing_transaction)
1354                 transaction = journal->j_committing_transaction;
1355
1356         /* Wait for the log commit to complete... */
1357         if (transaction) {
1358                 tid_t tid = transaction->t_tid;
1359
1360                 spin_unlock(&journal->j_state_lock);
1361                 jbd2_log_wait_commit(journal, tid);
1362         } else {
1363                 spin_unlock(&journal->j_state_lock);
1364         }
1365
1366         /* ...and flush everything in the log out to disk. */
1367         spin_lock(&journal->j_list_lock);
1368         while (!err && journal->j_checkpoint_transactions != NULL) {
1369                 spin_unlock(&journal->j_list_lock);
1370                 err = jbd2_log_do_checkpoint(journal);
1371                 spin_lock(&journal->j_list_lock);
1372         }
1373         spin_unlock(&journal->j_list_lock);
1374         jbd2_cleanup_journal_tail(journal);
1375
1376         /* Finally, mark the journal as really needing no recovery.
1377          * This sets s_start==0 in the underlying superblock, which is
1378          * the magic code for a fully-recovered superblock.  Any future
1379          * commits of data to the journal will restore the current
1380          * s_start value. */
1381         spin_lock(&journal->j_state_lock);
1382         old_tail = journal->j_tail;
1383         journal->j_tail = 0;
1384         spin_unlock(&journal->j_state_lock);
1385         jbd2_journal_update_superblock(journal, 1);
1386         spin_lock(&journal->j_state_lock);
1387         journal->j_tail = old_tail;
1388
1389         J_ASSERT(!journal->j_running_transaction);
1390         J_ASSERT(!journal->j_committing_transaction);
1391         J_ASSERT(!journal->j_checkpoint_transactions);
1392         J_ASSERT(journal->j_head == journal->j_tail);
1393         J_ASSERT(journal->j_tail_sequence == journal->j_transaction_sequence);
1394         spin_unlock(&journal->j_state_lock);
1395         return err;
1396 }
1397
1398 /**
1399  * int jbd2_journal_wipe() - Wipe journal contents
1400  * @journal: Journal to act on.
1401  * @write: flag (see below)
1402  *
1403  * Wipe out all of the contents of a journal, safely.  This will produce
1404  * a warning if the journal contains any valid recovery information.
1405  * Must be called between journal_init_*() and jbd2_journal_load().
1406  *
1407  * If 'write' is non-zero, then we wipe out the journal on disk; otherwise
1408  * we merely suppress recovery.
1409  */
1410
1411 int jbd2_journal_wipe(journal_t *journal, int write)
1412 {
1413         journal_superblock_t *sb;
1414         int err = 0;
1415
1416         J_ASSERT (!(journal->j_flags & JBD2_LOADED));
1417
1418         err = load_superblock(journal);
1419         if (err)
1420                 return err;
1421
1422         sb = journal->j_superblock;
1423
1424         if (!journal->j_tail)
1425                 goto no_recovery;
1426
1427         printk (KERN_WARNING "JBD: %s recovery information on journal\n",
1428                 write ? "Clearing" : "Ignoring");
1429
1430         err = jbd2_journal_skip_recovery(journal);
1431         if (write)
1432                 jbd2_journal_update_superblock(journal, 1);
1433
1434  no_recovery:
1435         return err;
1436 }
1437
1438 /*
1439  * journal_dev_name: format a character string to describe on what
1440  * device this journal is present.
1441  */
1442
1443 static const char *journal_dev_name(journal_t *journal, char *buffer)
1444 {
1445         struct block_device *bdev;
1446
1447         if (journal->j_inode)
1448                 bdev = journal->j_inode->i_sb->s_bdev;
1449         else
1450                 bdev = journal->j_dev;
1451
1452         return bdevname(bdev, buffer);
1453 }
1454
1455 /*
1456  * Journal abort has very specific semantics, which we describe
1457  * for journal abort.
1458  *
1459  * Two internal function, which provide abort to te jbd layer
1460  * itself are here.
1461  */
1462
1463 /*
1464  * Quick version for internal journal use (doesn't lock the journal).
1465  * Aborts hard --- we mark the abort as occurred, but do _nothing_ else,
1466  * and don't attempt to make any other journal updates.
1467  */
1468 void __jbd2_journal_abort_hard(journal_t *journal)
1469 {
1470         transaction_t *transaction;
1471         char b[BDEVNAME_SIZE];
1472
1473         if (journal->j_flags & JBD2_ABORT)
1474                 return;
1475
1476         printk(KERN_ERR "Aborting journal on device %s.\n",
1477                 journal_dev_name(journal, b));
1478
1479         spin_lock(&journal->j_state_lock);
1480         journal->j_flags |= JBD2_ABORT;
1481         transaction = journal->j_running_transaction;
1482         if (transaction)
1483                 __jbd2_log_start_commit(journal, transaction->t_tid);
1484         spin_unlock(&journal->j_state_lock);
1485 }
1486
1487 /* Soft abort: record the abort error status in the journal superblock,
1488  * but don't do any other IO. */
1489 static void __journal_abort_soft (journal_t *journal, int errno)
1490 {
1491         if (journal->j_flags & JBD2_ABORT)
1492                 return;
1493
1494         if (!journal->j_errno)
1495                 journal->j_errno = errno;
1496
1497         __jbd2_journal_abort_hard(journal);
1498
1499         if (errno)
1500                 jbd2_journal_update_superblock(journal, 1);
1501 }
1502
1503 /**
1504  * void jbd2_journal_abort () - Shutdown the journal immediately.
1505  * @journal: the journal to shutdown.
1506  * @errno:   an error number to record in the journal indicating
1507  *           the reason for the shutdown.
1508  *
1509  * Perform a complete, immediate shutdown of the ENTIRE
1510  * journal (not of a single transaction).  This operation cannot be
1511  * undone without closing and reopening the journal.
1512  *
1513  * The jbd2_journal_abort function is intended to support higher level error
1514  * recovery mechanisms such as the ext2/ext3 remount-readonly error
1515  * mode.
1516  *
1517  * Journal abort has very specific semantics.  Any existing dirty,
1518  * unjournaled buffers in the main filesystem will still be written to
1519  * disk by bdflush, but the journaling mechanism will be suspended
1520  * immediately and no further transaction commits will be honoured.
1521  *
1522  * Any dirty, journaled buffers will be written back to disk without
1523  * hitting the journal.  Atomicity cannot be guaranteed on an aborted
1524  * filesystem, but we _do_ attempt to leave as much data as possible
1525  * behind for fsck to use for cleanup.
1526  *
1527  * Any attempt to get a new transaction handle on a journal which is in
1528  * ABORT state will just result in an -EROFS error return.  A
1529  * jbd2_journal_stop on an existing handle will return -EIO if we have
1530  * entered abort state during the update.
1531  *
1532  * Recursive transactions are not disturbed by journal abort until the
1533  * final jbd2_journal_stop, which will receive the -EIO error.
1534  *
1535  * Finally, the jbd2_journal_abort call allows the caller to supply an errno
1536  * which will be recorded (if possible) in the journal superblock.  This
1537  * allows a client to record failure conditions in the middle of a
1538  * transaction without having to complete the transaction to record the
1539  * failure to disk.  ext3_error, for example, now uses this
1540  * functionality.
1541  *
1542  * Errors which originate from within the journaling layer will NOT
1543  * supply an errno; a null errno implies that absolutely no further
1544  * writes are done to the journal (unless there are any already in
1545  * progress).
1546  *
1547  */
1548
1549 void jbd2_journal_abort(journal_t *journal, int errno)
1550 {
1551         __journal_abort_soft(journal, errno);
1552 }
1553
1554 /**
1555  * int jbd2_journal_errno () - returns the journal's error state.
1556  * @journal: journal to examine.
1557  *
1558  * This is the errno numbet set with jbd2_journal_abort(), the last
1559  * time the journal was mounted - if the journal was stopped
1560  * without calling abort this will be 0.
1561  *
1562  * If the journal has been aborted on this mount time -EROFS will
1563  * be returned.
1564  */
1565 int jbd2_journal_errno(journal_t *journal)
1566 {
1567         int err;
1568
1569         spin_lock(&journal->j_state_lock);
1570         if (journal->j_flags & JBD2_ABORT)
1571                 err = -EROFS;
1572         else
1573                 err = journal->j_errno;
1574         spin_unlock(&journal->j_state_lock);
1575         return err;
1576 }
1577
1578 /**
1579  * int jbd2_journal_clear_err () - clears the journal's error state
1580  * @journal: journal to act on.
1581  *
1582  * An error must be cleared or Acked to take a FS out of readonly
1583  * mode.
1584  */
1585 int jbd2_journal_clear_err(journal_t *journal)
1586 {
1587         int err = 0;
1588
1589         spin_lock(&journal->j_state_lock);
1590         if (journal->j_flags & JBD2_ABORT)
1591                 err = -EROFS;
1592         else
1593                 journal->j_errno = 0;
1594         spin_unlock(&journal->j_state_lock);
1595         return err;
1596 }
1597
1598 /**
1599  * void jbd2_journal_ack_err() - Ack journal err.
1600  * @journal: journal to act on.
1601  *
1602  * An error must be cleared or Acked to take a FS out of readonly
1603  * mode.
1604  */
1605 void jbd2_journal_ack_err(journal_t *journal)
1606 {
1607         spin_lock(&journal->j_state_lock);
1608         if (journal->j_errno)
1609                 journal->j_flags |= JBD2_ACK_ERR;
1610         spin_unlock(&journal->j_state_lock);
1611 }
1612
1613 int jbd2_journal_blocks_per_page(struct inode *inode)
1614 {
1615         return 1 << (PAGE_CACHE_SHIFT - inode->i_sb->s_blocksize_bits);
1616 }
1617
1618 /*
1619  * helper functions to deal with 32 or 64bit block numbers.
1620  */
1621 size_t journal_tag_bytes(journal_t *journal)
1622 {
1623         if (JBD2_HAS_INCOMPAT_FEATURE(journal, JBD2_FEATURE_INCOMPAT_64BIT))
1624                 return JBD_TAG_SIZE64;
1625         else
1626                 return JBD_TAG_SIZE32;
1627 }
1628
1629 /*
1630  * Simple support for retrying memory allocations.  Introduced to help to
1631  * debug different VM deadlock avoidance strategies.
1632  */
1633 void * __jbd2_kmalloc (const char *where, size_t size, gfp_t flags, int retry)
1634 {
1635         return kmalloc(size, flags | (retry ? __GFP_NOFAIL : 0));
1636 }
1637
1638 /*
1639  * jbd slab management: create 1k, 2k, 4k, 8k slabs as needed
1640  * and allocate frozen and commit buffers from these slabs.
1641  *
1642  * Reason for doing this is to avoid, SLAB_DEBUG - since it could
1643  * cause bh to cross page boundary.
1644  */
1645
1646 #define JBD_MAX_SLABS 5
1647 #define JBD_SLAB_INDEX(size)  (size >> 11)
1648
1649 static struct kmem_cache *jbd_slab[JBD_MAX_SLABS];
1650 static const char *jbd_slab_names[JBD_MAX_SLABS] = {
1651         "jbd2_1k", "jbd2_2k", "jbd2_4k", NULL, "jbd2_8k"
1652 };
1653
1654 static void jbd2_journal_destroy_jbd_slabs(void)
1655 {
1656         int i;
1657
1658         for (i = 0; i < JBD_MAX_SLABS; i++) {
1659                 if (jbd_slab[i])
1660                         kmem_cache_destroy(jbd_slab[i]);
1661                 jbd_slab[i] = NULL;
1662         }
1663 }
1664
1665 static int jbd2_journal_create_jbd_slab(size_t slab_size)
1666 {
1667         int i = JBD_SLAB_INDEX(slab_size);
1668
1669         BUG_ON(i >= JBD_MAX_SLABS);
1670
1671         /*
1672          * Check if we already have a slab created for this size
1673          */
1674         if (jbd_slab[i])
1675                 return 0;
1676
1677         /*
1678          * Create a slab and force alignment to be same as slabsize -
1679          * this will make sure that allocations won't cross the page
1680          * boundary.
1681          */
1682         jbd_slab[i] = kmem_cache_create(jbd_slab_names[i],
1683                                 slab_size, slab_size, 0, NULL, NULL);
1684         if (!jbd_slab[i]) {
1685                 printk(KERN_EMERG "JBD: no memory for jbd_slab cache\n");
1686                 return -ENOMEM;
1687         }
1688         return 0;
1689 }
1690
1691 void * jbd2_slab_alloc(size_t size, gfp_t flags)
1692 {
1693         int idx;
1694
1695         idx = JBD_SLAB_INDEX(size);
1696         BUG_ON(jbd_slab[idx] == NULL);
1697         return kmem_cache_alloc(jbd_slab[idx], flags | __GFP_NOFAIL);
1698 }
1699
1700 void jbd2_slab_free(void *ptr,  size_t size)
1701 {
1702         int idx;
1703
1704         idx = JBD_SLAB_INDEX(size);
1705         BUG_ON(jbd_slab[idx] == NULL);
1706         kmem_cache_free(jbd_slab[idx], ptr);
1707 }
1708
1709 /*
1710  * Journal_head storage management
1711  */
1712 static struct kmem_cache *jbd2_journal_head_cache;
1713 #ifdef CONFIG_JBD2_DEBUG
1714 static atomic_t nr_journal_heads = ATOMIC_INIT(0);
1715 #endif
1716
1717 static int journal_init_jbd2_journal_head_cache(void)
1718 {
1719         int retval;
1720
1721         J_ASSERT(jbd2_journal_head_cache == 0);
1722         jbd2_journal_head_cache = kmem_cache_create("jbd2_journal_head",
1723                                 sizeof(struct journal_head),
1724                                 0,              /* offset */
1725                                 0,              /* flags */
1726                                 NULL,           /* ctor */
1727                                 NULL);          /* dtor */
1728         retval = 0;
1729         if (jbd2_journal_head_cache == 0) {
1730                 retval = -ENOMEM;
1731                 printk(KERN_EMERG "JBD: no memory for journal_head cache\n");
1732         }
1733         return retval;
1734 }
1735
1736 static void jbd2_journal_destroy_jbd2_journal_head_cache(void)
1737 {
1738         J_ASSERT(jbd2_journal_head_cache != NULL);
1739         kmem_cache_destroy(jbd2_journal_head_cache);
1740         jbd2_journal_head_cache = NULL;
1741 }
1742
1743 /*
1744  * journal_head splicing and dicing
1745  */
1746 static struct journal_head *journal_alloc_journal_head(void)
1747 {
1748         struct journal_head *ret;
1749         static unsigned long last_warning;
1750
1751 #ifdef CONFIG_JBD2_DEBUG
1752         atomic_inc(&nr_journal_heads);
1753 #endif
1754         ret = kmem_cache_alloc(jbd2_journal_head_cache, GFP_NOFS);
1755         if (ret == 0) {
1756                 jbd_debug(1, "out of memory for journal_head\n");
1757                 if (time_after(jiffies, last_warning + 5*HZ)) {
1758                         printk(KERN_NOTICE "ENOMEM in %s, retrying.\n",
1759                                __FUNCTION__);
1760                         last_warning = jiffies;
1761                 }
1762                 while (ret == 0) {
1763                         yield();
1764                         ret = kmem_cache_alloc(jbd2_journal_head_cache, GFP_NOFS);
1765                 }
1766         }
1767         return ret;
1768 }
1769
1770 static void journal_free_journal_head(struct journal_head *jh)
1771 {
1772 #ifdef CONFIG_JBD2_DEBUG
1773         atomic_dec(&nr_journal_heads);
1774         memset(jh, JBD_POISON_FREE, sizeof(*jh));
1775 #endif
1776         kmem_cache_free(jbd2_journal_head_cache, jh);
1777 }
1778
1779 /*
1780  * A journal_head is attached to a buffer_head whenever JBD has an
1781  * interest in the buffer.
1782  *
1783  * Whenever a buffer has an attached journal_head, its ->b_state:BH_JBD bit
1784  * is set.  This bit is tested in core kernel code where we need to take
1785  * JBD-specific actions.  Testing the zeroness of ->b_private is not reliable
1786  * there.
1787  *
1788  * When a buffer has its BH_JBD bit set, its ->b_count is elevated by one.
1789  *
1790  * When a buffer has its BH_JBD bit set it is immune from being released by
1791  * core kernel code, mainly via ->b_count.
1792  *
1793  * A journal_head may be detached from its buffer_head when the journal_head's
1794  * b_transaction, b_cp_transaction and b_next_transaction pointers are NULL.
1795  * Various places in JBD call jbd2_journal_remove_journal_head() to indicate that the
1796  * journal_head can be dropped if needed.
1797  *
1798  * Various places in the kernel want to attach a journal_head to a buffer_head
1799  * _before_ attaching the journal_head to a transaction.  To protect the
1800  * journal_head in this situation, jbd2_journal_add_journal_head elevates the
1801  * journal_head's b_jcount refcount by one.  The caller must call
1802  * jbd2_journal_put_journal_head() to undo this.
1803  *
1804  * So the typical usage would be:
1805  *
1806  *      (Attach a journal_head if needed.  Increments b_jcount)
1807  *      struct journal_head *jh = jbd2_journal_add_journal_head(bh);
1808  *      ...
1809  *      jh->b_transaction = xxx;
1810  *      jbd2_journal_put_journal_head(jh);
1811  *
1812  * Now, the journal_head's b_jcount is zero, but it is safe from being released
1813  * because it has a non-zero b_transaction.
1814  */
1815
1816 /*
1817  * Give a buffer_head a journal_head.
1818  *
1819  * Doesn't need the journal lock.
1820  * May sleep.
1821  */
1822 struct journal_head *jbd2_journal_add_journal_head(struct buffer_head *bh)
1823 {
1824         struct journal_head *jh;
1825         struct journal_head *new_jh = NULL;
1826
1827 repeat:
1828         if (!buffer_jbd(bh)) {
1829                 new_jh = journal_alloc_journal_head();
1830                 memset(new_jh, 0, sizeof(*new_jh));
1831         }
1832
1833         jbd_lock_bh_journal_head(bh);
1834         if (buffer_jbd(bh)) {
1835                 jh = bh2jh(bh);
1836         } else {
1837                 J_ASSERT_BH(bh,
1838                         (atomic_read(&bh->b_count) > 0) ||
1839                         (bh->b_page && bh->b_page->mapping));
1840
1841                 if (!new_jh) {
1842                         jbd_unlock_bh_journal_head(bh);
1843                         goto repeat;
1844                 }
1845
1846                 jh = new_jh;
1847                 new_jh = NULL;          /* We consumed it */
1848                 set_buffer_jbd(bh);
1849                 bh->b_private = jh;
1850                 jh->b_bh = bh;
1851                 get_bh(bh);
1852                 BUFFER_TRACE(bh, "added journal_head");
1853         }
1854         jh->b_jcount++;
1855         jbd_unlock_bh_journal_head(bh);
1856         if (new_jh)
1857                 journal_free_journal_head(new_jh);
1858         return bh->b_private;
1859 }
1860
1861 /*
1862  * Grab a ref against this buffer_head's journal_head.  If it ended up not
1863  * having a journal_head, return NULL
1864  */
1865 struct journal_head *jbd2_journal_grab_journal_head(struct buffer_head *bh)
1866 {
1867         struct journal_head *jh = NULL;
1868
1869         jbd_lock_bh_journal_head(bh);
1870         if (buffer_jbd(bh)) {
1871                 jh = bh2jh(bh);
1872                 jh->b_jcount++;
1873         }
1874         jbd_unlock_bh_journal_head(bh);
1875         return jh;
1876 }
1877
1878 static void __journal_remove_journal_head(struct buffer_head *bh)
1879 {
1880         struct journal_head *jh = bh2jh(bh);
1881
1882         J_ASSERT_JH(jh, jh->b_jcount >= 0);
1883
1884         get_bh(bh);
1885         if (jh->b_jcount == 0) {
1886                 if (jh->b_transaction == NULL &&
1887                                 jh->b_next_transaction == NULL &&
1888                                 jh->b_cp_transaction == NULL) {
1889                         J_ASSERT_JH(jh, jh->b_jlist == BJ_None);
1890                         J_ASSERT_BH(bh, buffer_jbd(bh));
1891                         J_ASSERT_BH(bh, jh2bh(jh) == bh);
1892                         BUFFER_TRACE(bh, "remove journal_head");
1893                         if (jh->b_frozen_data) {
1894                                 printk(KERN_WARNING "%s: freeing "
1895                                                 "b_frozen_data\n",
1896                                                 __FUNCTION__);
1897                                 jbd2_slab_free(jh->b_frozen_data, bh->b_size);
1898                         }
1899                         if (jh->b_committed_data) {
1900                                 printk(KERN_WARNING "%s: freeing "
1901                                                 "b_committed_data\n",
1902                                                 __FUNCTION__);
1903                                 jbd2_slab_free(jh->b_committed_data, bh->b_size);
1904                         }
1905                         bh->b_private = NULL;
1906                         jh->b_bh = NULL;        /* debug, really */
1907                         clear_buffer_jbd(bh);
1908                         __brelse(bh);
1909                         journal_free_journal_head(jh);
1910                 } else {
1911                         BUFFER_TRACE(bh, "journal_head was locked");
1912                 }
1913         }
1914 }
1915
1916 /*
1917  * jbd2_journal_remove_journal_head(): if the buffer isn't attached to a transaction
1918  * and has a zero b_jcount then remove and release its journal_head.   If we did
1919  * see that the buffer is not used by any transaction we also "logically"
1920  * decrement ->b_count.
1921  *
1922  * We in fact take an additional increment on ->b_count as a convenience,
1923  * because the caller usually wants to do additional things with the bh
1924  * after calling here.
1925  * The caller of jbd2_journal_remove_journal_head() *must* run __brelse(bh) at some
1926  * time.  Once the caller has run __brelse(), the buffer is eligible for
1927  * reaping by try_to_free_buffers().
1928  */
1929 void jbd2_journal_remove_journal_head(struct buffer_head *bh)
1930 {
1931         jbd_lock_bh_journal_head(bh);
1932         __journal_remove_journal_head(bh);
1933         jbd_unlock_bh_journal_head(bh);
1934 }
1935
1936 /*
1937  * Drop a reference on the passed journal_head.  If it fell to zero then try to
1938  * release the journal_head from the buffer_head.
1939  */
1940 void jbd2_journal_put_journal_head(struct journal_head *jh)
1941 {
1942         struct buffer_head *bh = jh2bh(jh);
1943
1944         jbd_lock_bh_journal_head(bh);
1945         J_ASSERT_JH(jh, jh->b_jcount > 0);
1946         --jh->b_jcount;
1947         if (!jh->b_jcount && !jh->b_transaction) {
1948                 __journal_remove_journal_head(bh);
1949                 __brelse(bh);
1950         }
1951         jbd_unlock_bh_journal_head(bh);
1952 }
1953
1954 /*
1955  * debugfs tunables
1956  */
1957 #if defined(CONFIG_JBD2_DEBUG)
1958 u8 jbd2_journal_enable_debug;
1959 EXPORT_SYMBOL(jbd2_journal_enable_debug);
1960 #endif
1961
1962 #if defined(CONFIG_JBD2_DEBUG) && defined(CONFIG_DEBUG_FS)
1963
1964 #define JBD2_DEBUG_NAME "jbd2-debug"
1965
1966 struct dentry *jbd2_debugfs_dir, *jbd2_debug;
1967
1968 static void __init jbd2_create_debugfs_entry(void)
1969 {
1970         jbd2_debugfs_dir = debugfs_create_dir("jbd2", NULL);
1971         if (jbd2_debugfs_dir)
1972                 jbd2_debug = debugfs_create_u8(JBD2_DEBUG_NAME, S_IRUGO,
1973                                                jbd2_debugfs_dir,
1974                                                &jbd2_journal_enable_debug);
1975 }
1976
1977 static void __exit jbd2_remove_debugfs_entry(void)
1978 {
1979         if (jbd2_debug)
1980                 debugfs_remove(jbd2_debug);
1981         if (jbd2_debugfs_dir)
1982                 debugfs_remove(jbd2_debugfs_dir);
1983 }
1984
1985 #else
1986
1987 static void __init jbd2_create_debugfs_entry(void)
1988 {
1989         do {
1990         } while (0);
1991 }
1992
1993 static void __exit jbd2_remove_debugfs_entry(void)
1994 {
1995         do {
1996         } while (0);
1997 }
1998
1999 #endif
2000
2001 struct kmem_cache *jbd2_handle_cache;
2002
2003 static int __init journal_init_handle_cache(void)
2004 {
2005         jbd2_handle_cache = kmem_cache_create("jbd2_journal_handle",
2006                                 sizeof(handle_t),
2007                                 0,              /* offset */
2008                                 0,              /* flags */
2009                                 NULL,           /* ctor */
2010                                 NULL);          /* dtor */
2011         if (jbd2_handle_cache == NULL) {
2012                 printk(KERN_EMERG "JBD: failed to create handle cache\n");
2013                 return -ENOMEM;
2014         }
2015         return 0;
2016 }
2017
2018 static void jbd2_journal_destroy_handle_cache(void)
2019 {
2020         if (jbd2_handle_cache)
2021                 kmem_cache_destroy(jbd2_handle_cache);
2022 }
2023
2024 /*
2025  * Module startup and shutdown
2026  */
2027
2028 static int __init journal_init_caches(void)
2029 {
2030         int ret;
2031
2032         ret = jbd2_journal_init_revoke_caches();
2033         if (ret == 0)
2034                 ret = journal_init_jbd2_journal_head_cache();
2035         if (ret == 0)
2036                 ret = journal_init_handle_cache();
2037         return ret;
2038 }
2039
2040 static void jbd2_journal_destroy_caches(void)
2041 {
2042         jbd2_journal_destroy_revoke_caches();
2043         jbd2_journal_destroy_jbd2_journal_head_cache();
2044         jbd2_journal_destroy_handle_cache();
2045         jbd2_journal_destroy_jbd_slabs();
2046 }
2047
2048 static int __init journal_init(void)
2049 {
2050         int ret;
2051
2052         BUILD_BUG_ON(sizeof(struct journal_superblock_s) != 1024);
2053
2054         ret = journal_init_caches();
2055         if (ret != 0)
2056                 jbd2_journal_destroy_caches();
2057         jbd2_create_debugfs_entry();
2058         return ret;
2059 }
2060
2061 static void __exit journal_exit(void)
2062 {
2063 #ifdef CONFIG_JBD2_DEBUG
2064         int n = atomic_read(&nr_journal_heads);
2065         if (n)
2066                 printk(KERN_EMERG "JBD: leaked %d journal_heads!\n", n);
2067 #endif
2068         jbd2_remove_debugfs_entry();
2069         jbd2_journal_destroy_caches();
2070 }
2071
2072 MODULE_LICENSE("GPL");
2073 module_init(journal_init);
2074 module_exit(journal_exit);
2075