Merge git://git.kernel.org/pub/scm/linux/kernel/git/wim/linux-2.6-watchdog
[linux-2.6] / fs / jbd2 / journal.c
1 /*
2  * linux/fs/jbd2/journal.c
3  *
4  * Written by Stephen C. Tweedie <sct@redhat.com>, 1998
5  *
6  * Copyright 1998 Red Hat corp --- All Rights Reserved
7  *
8  * This file is part of the Linux kernel and is made available under
9  * the terms of the GNU General Public License, version 2, or at your
10  * option, any later version, incorporated herein by reference.
11  *
12  * Generic filesystem journal-writing code; part of the ext2fs
13  * journaling system.
14  *
15  * This file manages journals: areas of disk reserved for logging
16  * transactional updates.  This includes the kernel journaling thread
17  * which is responsible for scheduling updates to the log.
18  *
19  * We do not actually manage the physical storage of the journal in this
20  * file: that is left to a per-journal policy function, which allows us
21  * to store the journal within a filesystem-specified area for ext2
22  * journaling (ext2 can use a reserved inode for storing the log).
23  */
24
25 #include <linux/module.h>
26 #include <linux/time.h>
27 #include <linux/fs.h>
28 #include <linux/jbd2.h>
29 #include <linux/errno.h>
30 #include <linux/slab.h>
31 #include <linux/init.h>
32 #include <linux/mm.h>
33 #include <linux/freezer.h>
34 #include <linux/pagemap.h>
35 #include <linux/kthread.h>
36 #include <linux/poison.h>
37 #include <linux/proc_fs.h>
38 #include <linux/debugfs.h>
39 #include <linux/seq_file.h>
40 #include <linux/math64.h>
41
42 #include <asm/uaccess.h>
43 #include <asm/page.h>
44
45 EXPORT_SYMBOL(jbd2_journal_start);
46 EXPORT_SYMBOL(jbd2_journal_restart);
47 EXPORT_SYMBOL(jbd2_journal_extend);
48 EXPORT_SYMBOL(jbd2_journal_stop);
49 EXPORT_SYMBOL(jbd2_journal_lock_updates);
50 EXPORT_SYMBOL(jbd2_journal_unlock_updates);
51 EXPORT_SYMBOL(jbd2_journal_get_write_access);
52 EXPORT_SYMBOL(jbd2_journal_get_create_access);
53 EXPORT_SYMBOL(jbd2_journal_get_undo_access);
54 EXPORT_SYMBOL(jbd2_journal_set_triggers);
55 EXPORT_SYMBOL(jbd2_journal_dirty_metadata);
56 EXPORT_SYMBOL(jbd2_journal_release_buffer);
57 EXPORT_SYMBOL(jbd2_journal_forget);
58 #if 0
59 EXPORT_SYMBOL(journal_sync_buffer);
60 #endif
61 EXPORT_SYMBOL(jbd2_journal_flush);
62 EXPORT_SYMBOL(jbd2_journal_revoke);
63
64 EXPORT_SYMBOL(jbd2_journal_init_dev);
65 EXPORT_SYMBOL(jbd2_journal_init_inode);
66 EXPORT_SYMBOL(jbd2_journal_update_format);
67 EXPORT_SYMBOL(jbd2_journal_check_used_features);
68 EXPORT_SYMBOL(jbd2_journal_check_available_features);
69 EXPORT_SYMBOL(jbd2_journal_set_features);
70 EXPORT_SYMBOL(jbd2_journal_load);
71 EXPORT_SYMBOL(jbd2_journal_destroy);
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 EXPORT_SYMBOL(jbd2_journal_file_inode);
85 EXPORT_SYMBOL(jbd2_journal_init_jbd_inode);
86 EXPORT_SYMBOL(jbd2_journal_release_jbd_inode);
87 EXPORT_SYMBOL(jbd2_journal_begin_ordered_truncate);
88
89 static int journal_convert_superblock_v1(journal_t *, journal_superblock_t *);
90 static void __journal_abort_soft (journal_t *journal, int errno);
91
92 /*
93  * Helper function used to manage commit timeouts
94  */
95
96 static void commit_timeout(unsigned long __data)
97 {
98         struct task_struct * p = (struct task_struct *) __data;
99
100         wake_up_process(p);
101 }
102
103 /*
104  * kjournald2: The main thread function used to manage a logging device
105  * journal.
106  *
107  * This kernel thread is responsible for two things:
108  *
109  * 1) COMMIT:  Every so often we need to commit the current state of the
110  *    filesystem to disk.  The journal thread is responsible for writing
111  *    all of the metadata buffers to disk.
112  *
113  * 2) CHECKPOINT: We cannot reuse a used section of the log file until all
114  *    of the data in that part of the log has been rewritten elsewhere on
115  *    the disk.  Flushing these old buffers to reclaim space in the log is
116  *    known as checkpointing, and this thread is responsible for that job.
117  */
118
119 static int kjournald2(void *arg)
120 {
121         journal_t *journal = arg;
122         transaction_t *transaction;
123
124         /*
125          * Set up an interval timer which can be used to trigger a commit wakeup
126          * after the commit interval expires
127          */
128         setup_timer(&journal->j_commit_timer, commit_timeout,
129                         (unsigned long)current);
130
131         /* Record that the journal thread is running */
132         journal->j_task = current;
133         wake_up(&journal->j_wait_done_commit);
134
135         printk(KERN_INFO "kjournald2 starting: pid %d, dev %s, "
136                "commit interval %ld seconds\n", current->pid,
137                journal->j_devname, journal->j_commit_interval / HZ);
138
139         /*
140          * And now, wait forever for commit wakeup events.
141          */
142         spin_lock(&journal->j_state_lock);
143
144 loop:
145         if (journal->j_flags & JBD2_UNMOUNT)
146                 goto end_loop;
147
148         jbd_debug(1, "commit_sequence=%d, commit_request=%d\n",
149                 journal->j_commit_sequence, journal->j_commit_request);
150
151         if (journal->j_commit_sequence != journal->j_commit_request) {
152                 jbd_debug(1, "OK, requests differ\n");
153                 spin_unlock(&journal->j_state_lock);
154                 del_timer_sync(&journal->j_commit_timer);
155                 jbd2_journal_commit_transaction(journal);
156                 spin_lock(&journal->j_state_lock);
157                 goto loop;
158         }
159
160         wake_up(&journal->j_wait_done_commit);
161         if (freezing(current)) {
162                 /*
163                  * The simpler the better. Flushing journal isn't a
164                  * good idea, because that depends on threads that may
165                  * be already stopped.
166                  */
167                 jbd_debug(1, "Now suspending kjournald2\n");
168                 spin_unlock(&journal->j_state_lock);
169                 refrigerator();
170                 spin_lock(&journal->j_state_lock);
171         } else {
172                 /*
173                  * We assume on resume that commits are already there,
174                  * so we don't sleep
175                  */
176                 DEFINE_WAIT(wait);
177                 int should_sleep = 1;
178
179                 prepare_to_wait(&journal->j_wait_commit, &wait,
180                                 TASK_INTERRUPTIBLE);
181                 if (journal->j_commit_sequence != journal->j_commit_request)
182                         should_sleep = 0;
183                 transaction = journal->j_running_transaction;
184                 if (transaction && time_after_eq(jiffies,
185                                                 transaction->t_expires))
186                         should_sleep = 0;
187                 if (journal->j_flags & JBD2_UNMOUNT)
188                         should_sleep = 0;
189                 if (should_sleep) {
190                         spin_unlock(&journal->j_state_lock);
191                         schedule();
192                         spin_lock(&journal->j_state_lock);
193                 }
194                 finish_wait(&journal->j_wait_commit, &wait);
195         }
196
197         jbd_debug(1, "kjournald2 wakes\n");
198
199         /*
200          * Were we woken up by a commit wakeup event?
201          */
202         transaction = journal->j_running_transaction;
203         if (transaction && time_after_eq(jiffies, transaction->t_expires)) {
204                 journal->j_commit_request = transaction->t_tid;
205                 jbd_debug(1, "woke because of timeout\n");
206         }
207         goto loop;
208
209 end_loop:
210         spin_unlock(&journal->j_state_lock);
211         del_timer_sync(&journal->j_commit_timer);
212         journal->j_task = NULL;
213         wake_up(&journal->j_wait_done_commit);
214         jbd_debug(1, "Journal thread exiting.\n");
215         return 0;
216 }
217
218 static int jbd2_journal_start_thread(journal_t *journal)
219 {
220         struct task_struct *t;
221
222         t = kthread_run(kjournald2, journal, "kjournald2");
223         if (IS_ERR(t))
224                 return PTR_ERR(t);
225
226         wait_event(journal->j_wait_done_commit, journal->j_task != NULL);
227         return 0;
228 }
229
230 static void journal_kill_thread(journal_t *journal)
231 {
232         spin_lock(&journal->j_state_lock);
233         journal->j_flags |= JBD2_UNMOUNT;
234
235         while (journal->j_task) {
236                 wake_up(&journal->j_wait_commit);
237                 spin_unlock(&journal->j_state_lock);
238                 wait_event(journal->j_wait_done_commit, journal->j_task == NULL);
239                 spin_lock(&journal->j_state_lock);
240         }
241         spin_unlock(&journal->j_state_lock);
242 }
243
244 /*
245  * jbd2_journal_write_metadata_buffer: write a metadata buffer to the journal.
246  *
247  * Writes a metadata buffer to a given disk block.  The actual IO is not
248  * performed but a new buffer_head is constructed which labels the data
249  * to be written with the correct destination disk block.
250  *
251  * Any magic-number escaping which needs to be done will cause a
252  * copy-out here.  If the buffer happens to start with the
253  * JBD2_MAGIC_NUMBER, then we can't write it to the log directly: the
254  * magic number is only written to the log for descripter blocks.  In
255  * this case, we copy the data and replace the first word with 0, and we
256  * return a result code which indicates that this buffer needs to be
257  * marked as an escaped buffer in the corresponding log descriptor
258  * block.  The missing word can then be restored when the block is read
259  * during recovery.
260  *
261  * If the source buffer has already been modified by a new transaction
262  * since we took the last commit snapshot, we use the frozen copy of
263  * that data for IO.  If we end up using the existing buffer_head's data
264  * for the write, then we *have* to lock the buffer to prevent anyone
265  * else from using and possibly modifying it while the IO is in
266  * progress.
267  *
268  * The function returns a pointer to the buffer_heads to be used for IO.
269  *
270  * We assume that the journal has already been locked in this function.
271  *
272  * Return value:
273  *  <0: Error
274  * >=0: Finished OK
275  *
276  * On success:
277  * Bit 0 set == escape performed on the data
278  * Bit 1 set == buffer copy-out performed (kfree the data after IO)
279  */
280
281 int jbd2_journal_write_metadata_buffer(transaction_t *transaction,
282                                   struct journal_head  *jh_in,
283                                   struct journal_head **jh_out,
284                                   unsigned long long blocknr)
285 {
286         int need_copy_out = 0;
287         int done_copy_out = 0;
288         int do_escape = 0;
289         char *mapped_data;
290         struct buffer_head *new_bh;
291         struct journal_head *new_jh;
292         struct page *new_page;
293         unsigned int new_offset;
294         struct buffer_head *bh_in = jh2bh(jh_in);
295         struct jbd2_buffer_trigger_type *triggers;
296
297         /*
298          * The buffer really shouldn't be locked: only the current committing
299          * transaction is allowed to write it, so nobody else is allowed
300          * to do any IO.
301          *
302          * akpm: except if we're journalling data, and write() output is
303          * also part of a shared mapping, and another thread has
304          * decided to launch a writepage() against this buffer.
305          */
306         J_ASSERT_BH(bh_in, buffer_jbddirty(bh_in));
307
308         new_bh = alloc_buffer_head(GFP_NOFS|__GFP_NOFAIL);
309
310         /*
311          * If a new transaction has already done a buffer copy-out, then
312          * we use that version of the data for the commit.
313          */
314         jbd_lock_bh_state(bh_in);
315 repeat:
316         if (jh_in->b_frozen_data) {
317                 done_copy_out = 1;
318                 new_page = virt_to_page(jh_in->b_frozen_data);
319                 new_offset = offset_in_page(jh_in->b_frozen_data);
320                 triggers = jh_in->b_frozen_triggers;
321         } else {
322                 new_page = jh2bh(jh_in)->b_page;
323                 new_offset = offset_in_page(jh2bh(jh_in)->b_data);
324                 triggers = jh_in->b_triggers;
325         }
326
327         mapped_data = kmap_atomic(new_page, KM_USER0);
328         /*
329          * Fire any commit trigger.  Do this before checking for escaping,
330          * as the trigger may modify the magic offset.  If a copy-out
331          * happens afterwards, it will have the correct data in the buffer.
332          */
333         jbd2_buffer_commit_trigger(jh_in, mapped_data + new_offset,
334                                    triggers);
335
336         /*
337          * Check for escaping
338          */
339         if (*((__be32 *)(mapped_data + new_offset)) ==
340                                 cpu_to_be32(JBD2_MAGIC_NUMBER)) {
341                 need_copy_out = 1;
342                 do_escape = 1;
343         }
344         kunmap_atomic(mapped_data, KM_USER0);
345
346         /*
347          * Do we need to do a data copy?
348          */
349         if (need_copy_out && !done_copy_out) {
350                 char *tmp;
351
352                 jbd_unlock_bh_state(bh_in);
353                 tmp = jbd2_alloc(bh_in->b_size, GFP_NOFS);
354                 jbd_lock_bh_state(bh_in);
355                 if (jh_in->b_frozen_data) {
356                         jbd2_free(tmp, bh_in->b_size);
357                         goto repeat;
358                 }
359
360                 jh_in->b_frozen_data = tmp;
361                 mapped_data = kmap_atomic(new_page, KM_USER0);
362                 memcpy(tmp, mapped_data + new_offset, jh2bh(jh_in)->b_size);
363                 kunmap_atomic(mapped_data, KM_USER0);
364
365                 new_page = virt_to_page(tmp);
366                 new_offset = offset_in_page(tmp);
367                 done_copy_out = 1;
368
369                 /*
370                  * This isn't strictly necessary, as we're using frozen
371                  * data for the escaping, but it keeps consistency with
372                  * b_frozen_data usage.
373                  */
374                 jh_in->b_frozen_triggers = jh_in->b_triggers;
375         }
376
377         /*
378          * Did we need to do an escaping?  Now we've done all the
379          * copying, we can finally do so.
380          */
381         if (do_escape) {
382                 mapped_data = kmap_atomic(new_page, KM_USER0);
383                 *((unsigned int *)(mapped_data + new_offset)) = 0;
384                 kunmap_atomic(mapped_data, KM_USER0);
385         }
386
387         /* keep subsequent assertions sane */
388         new_bh->b_state = 0;
389         init_buffer(new_bh, NULL, NULL);
390         atomic_set(&new_bh->b_count, 1);
391         jbd_unlock_bh_state(bh_in);
392
393         new_jh = jbd2_journal_add_journal_head(new_bh); /* This sleeps */
394
395         set_bh_page(new_bh, new_page, new_offset);
396         new_jh->b_transaction = NULL;
397         new_bh->b_size = jh2bh(jh_in)->b_size;
398         new_bh->b_bdev = transaction->t_journal->j_dev;
399         new_bh->b_blocknr = blocknr;
400         set_buffer_mapped(new_bh);
401         set_buffer_dirty(new_bh);
402
403         *jh_out = new_jh;
404
405         /*
406          * The to-be-written buffer needs to get moved to the io queue,
407          * and the original buffer whose contents we are shadowing or
408          * copying is moved to the transaction's shadow queue.
409          */
410         JBUFFER_TRACE(jh_in, "file as BJ_Shadow");
411         jbd2_journal_file_buffer(jh_in, transaction, BJ_Shadow);
412         JBUFFER_TRACE(new_jh, "file as BJ_IO");
413         jbd2_journal_file_buffer(new_jh, transaction, BJ_IO);
414
415         return do_escape | (done_copy_out << 1);
416 }
417
418 /*
419  * Allocation code for the journal file.  Manage the space left in the
420  * journal, so that we can begin checkpointing when appropriate.
421  */
422
423 /*
424  * __jbd2_log_space_left: Return the number of free blocks left in the journal.
425  *
426  * Called with the journal already locked.
427  *
428  * Called under j_state_lock
429  */
430
431 int __jbd2_log_space_left(journal_t *journal)
432 {
433         int left = journal->j_free;
434
435         assert_spin_locked(&journal->j_state_lock);
436
437         /*
438          * Be pessimistic here about the number of those free blocks which
439          * might be required for log descriptor control blocks.
440          */
441
442 #define MIN_LOG_RESERVED_BLOCKS 32 /* Allow for rounding errors */
443
444         left -= MIN_LOG_RESERVED_BLOCKS;
445
446         if (left <= 0)
447                 return 0;
448         left -= (left >> 3);
449         return left;
450 }
451
452 /*
453  * Called under j_state_lock.  Returns true if a transaction commit was started.
454  */
455 int __jbd2_log_start_commit(journal_t *journal, tid_t target)
456 {
457         /*
458          * Are we already doing a recent enough commit?
459          */
460         if (!tid_geq(journal->j_commit_request, target)) {
461                 /*
462                  * We want a new commit: OK, mark the request and wakup the
463                  * commit thread.  We do _not_ do the commit ourselves.
464                  */
465
466                 journal->j_commit_request = target;
467                 jbd_debug(1, "JBD: requesting commit %d/%d\n",
468                           journal->j_commit_request,
469                           journal->j_commit_sequence);
470                 wake_up(&journal->j_wait_commit);
471                 return 1;
472         }
473         return 0;
474 }
475
476 int jbd2_log_start_commit(journal_t *journal, tid_t tid)
477 {
478         int ret;
479
480         spin_lock(&journal->j_state_lock);
481         ret = __jbd2_log_start_commit(journal, tid);
482         spin_unlock(&journal->j_state_lock);
483         return ret;
484 }
485
486 /*
487  * Force and wait upon a commit if the calling process is not within
488  * transaction.  This is used for forcing out undo-protected data which contains
489  * bitmaps, when the fs is running out of space.
490  *
491  * We can only force the running transaction if we don't have an active handle;
492  * otherwise, we will deadlock.
493  *
494  * Returns true if a transaction was started.
495  */
496 int jbd2_journal_force_commit_nested(journal_t *journal)
497 {
498         transaction_t *transaction = NULL;
499         tid_t tid;
500
501         spin_lock(&journal->j_state_lock);
502         if (journal->j_running_transaction && !current->journal_info) {
503                 transaction = journal->j_running_transaction;
504                 __jbd2_log_start_commit(journal, transaction->t_tid);
505         } else if (journal->j_committing_transaction)
506                 transaction = journal->j_committing_transaction;
507
508         if (!transaction) {
509                 spin_unlock(&journal->j_state_lock);
510                 return 0;       /* Nothing to retry */
511         }
512
513         tid = transaction->t_tid;
514         spin_unlock(&journal->j_state_lock);
515         jbd2_log_wait_commit(journal, tid);
516         return 1;
517 }
518
519 /*
520  * Start a commit of the current running transaction (if any).  Returns true
521  * if a transaction is going to be committed (or is currently already
522  * committing), and fills its tid in at *ptid
523  */
524 int jbd2_journal_start_commit(journal_t *journal, tid_t *ptid)
525 {
526         int ret = 0;
527
528         spin_lock(&journal->j_state_lock);
529         if (journal->j_running_transaction) {
530                 tid_t tid = journal->j_running_transaction->t_tid;
531
532                 __jbd2_log_start_commit(journal, tid);
533                 /* There's a running transaction and we've just made sure
534                  * it's commit has been scheduled. */
535                 if (ptid)
536                         *ptid = tid;
537                 ret = 1;
538         } else if (journal->j_committing_transaction) {
539                 /*
540                  * If ext3_write_super() recently started a commit, then we
541                  * have to wait for completion of that transaction
542                  */
543                 if (ptid)
544                         *ptid = journal->j_committing_transaction->t_tid;
545                 ret = 1;
546         }
547         spin_unlock(&journal->j_state_lock);
548         return ret;
549 }
550
551 /*
552  * Wait for a specified commit to complete.
553  * The caller may not hold the journal lock.
554  */
555 int jbd2_log_wait_commit(journal_t *journal, tid_t tid)
556 {
557         int err = 0;
558
559 #ifdef CONFIG_JBD2_DEBUG
560         spin_lock(&journal->j_state_lock);
561         if (!tid_geq(journal->j_commit_request, tid)) {
562                 printk(KERN_EMERG
563                        "%s: error: j_commit_request=%d, tid=%d\n",
564                        __func__, journal->j_commit_request, tid);
565         }
566         spin_unlock(&journal->j_state_lock);
567 #endif
568         spin_lock(&journal->j_state_lock);
569         while (tid_gt(tid, journal->j_commit_sequence)) {
570                 jbd_debug(1, "JBD: want %d, j_commit_sequence=%d\n",
571                                   tid, journal->j_commit_sequence);
572                 wake_up(&journal->j_wait_commit);
573                 spin_unlock(&journal->j_state_lock);
574                 wait_event(journal->j_wait_done_commit,
575                                 !tid_gt(tid, journal->j_commit_sequence));
576                 spin_lock(&journal->j_state_lock);
577         }
578         spin_unlock(&journal->j_state_lock);
579
580         if (unlikely(is_journal_aborted(journal))) {
581                 printk(KERN_EMERG "journal commit I/O error\n");
582                 err = -EIO;
583         }
584         return err;
585 }
586
587 /*
588  * Log buffer allocation routines:
589  */
590
591 int jbd2_journal_next_log_block(journal_t *journal, unsigned long long *retp)
592 {
593         unsigned long blocknr;
594
595         spin_lock(&journal->j_state_lock);
596         J_ASSERT(journal->j_free > 1);
597
598         blocknr = journal->j_head;
599         journal->j_head++;
600         journal->j_free--;
601         if (journal->j_head == journal->j_last)
602                 journal->j_head = journal->j_first;
603         spin_unlock(&journal->j_state_lock);
604         return jbd2_journal_bmap(journal, blocknr, retp);
605 }
606
607 /*
608  * Conversion of logical to physical block numbers for the journal
609  *
610  * On external journals the journal blocks are identity-mapped, so
611  * this is a no-op.  If needed, we can use j_blk_offset - everything is
612  * ready.
613  */
614 int jbd2_journal_bmap(journal_t *journal, unsigned long blocknr,
615                  unsigned long long *retp)
616 {
617         int err = 0;
618         unsigned long long ret;
619
620         if (journal->j_inode) {
621                 ret = bmap(journal->j_inode, blocknr);
622                 if (ret)
623                         *retp = ret;
624                 else {
625                         printk(KERN_ALERT "%s: journal block not found "
626                                         "at offset %lu on %s\n",
627                                __func__, blocknr, journal->j_devname);
628                         err = -EIO;
629                         __journal_abort_soft(journal, err);
630                 }
631         } else {
632                 *retp = blocknr; /* +journal->j_blk_offset */
633         }
634         return err;
635 }
636
637 /*
638  * We play buffer_head aliasing tricks to write data/metadata blocks to
639  * the journal without copying their contents, but for journal
640  * descriptor blocks we do need to generate bona fide buffers.
641  *
642  * After the caller of jbd2_journal_get_descriptor_buffer() has finished modifying
643  * the buffer's contents they really should run flush_dcache_page(bh->b_page).
644  * But we don't bother doing that, so there will be coherency problems with
645  * mmaps of blockdevs which hold live JBD-controlled filesystems.
646  */
647 struct journal_head *jbd2_journal_get_descriptor_buffer(journal_t *journal)
648 {
649         struct buffer_head *bh;
650         unsigned long long blocknr;
651         int err;
652
653         err = jbd2_journal_next_log_block(journal, &blocknr);
654
655         if (err)
656                 return NULL;
657
658         bh = __getblk(journal->j_dev, blocknr, journal->j_blocksize);
659         if (!bh)
660                 return NULL;
661         lock_buffer(bh);
662         memset(bh->b_data, 0, journal->j_blocksize);
663         set_buffer_uptodate(bh);
664         unlock_buffer(bh);
665         BUFFER_TRACE(bh, "return this buffer");
666         return jbd2_journal_add_journal_head(bh);
667 }
668
669 struct jbd2_stats_proc_session {
670         journal_t *journal;
671         struct transaction_stats_s *stats;
672         int start;
673         int max;
674 };
675
676 static void *jbd2_history_skip_empty(struct jbd2_stats_proc_session *s,
677                                         struct transaction_stats_s *ts,
678                                         int first)
679 {
680         if (ts == s->stats + s->max)
681                 ts = s->stats;
682         if (!first && ts == s->stats + s->start)
683                 return NULL;
684         while (ts->ts_type == 0) {
685                 ts++;
686                 if (ts == s->stats + s->max)
687                         ts = s->stats;
688                 if (ts == s->stats + s->start)
689                         return NULL;
690         }
691         return ts;
692
693 }
694
695 static void *jbd2_seq_history_start(struct seq_file *seq, loff_t *pos)
696 {
697         struct jbd2_stats_proc_session *s = seq->private;
698         struct transaction_stats_s *ts;
699         int l = *pos;
700
701         if (l == 0)
702                 return SEQ_START_TOKEN;
703         ts = jbd2_history_skip_empty(s, s->stats + s->start, 1);
704         if (!ts)
705                 return NULL;
706         l--;
707         while (l) {
708                 ts = jbd2_history_skip_empty(s, ++ts, 0);
709                 if (!ts)
710                         break;
711                 l--;
712         }
713         return ts;
714 }
715
716 static void *jbd2_seq_history_next(struct seq_file *seq, void *v, loff_t *pos)
717 {
718         struct jbd2_stats_proc_session *s = seq->private;
719         struct transaction_stats_s *ts = v;
720
721         ++*pos;
722         if (v == SEQ_START_TOKEN)
723                 return jbd2_history_skip_empty(s, s->stats + s->start, 1);
724         else
725                 return jbd2_history_skip_empty(s, ++ts, 0);
726 }
727
728 static int jbd2_seq_history_show(struct seq_file *seq, void *v)
729 {
730         struct transaction_stats_s *ts = v;
731         if (v == SEQ_START_TOKEN) {
732                 seq_printf(seq, "%-4s %-5s %-5s %-5s %-5s %-5s %-5s %-6s %-5s "
733                                 "%-5s %-5s %-5s %-5s %-5s\n", "R/C", "tid",
734                                 "wait", "run", "lock", "flush", "log", "hndls",
735                                 "block", "inlog", "ctime", "write", "drop",
736                                 "close");
737                 return 0;
738         }
739         if (ts->ts_type == JBD2_STATS_RUN)
740                 seq_printf(seq, "%-4s %-5lu %-5u %-5u %-5u %-5u %-5u "
741                                 "%-6lu %-5lu %-5lu\n", "R", ts->ts_tid,
742                                 jiffies_to_msecs(ts->u.run.rs_wait),
743                                 jiffies_to_msecs(ts->u.run.rs_running),
744                                 jiffies_to_msecs(ts->u.run.rs_locked),
745                                 jiffies_to_msecs(ts->u.run.rs_flushing),
746                                 jiffies_to_msecs(ts->u.run.rs_logging),
747                                 ts->u.run.rs_handle_count,
748                                 ts->u.run.rs_blocks,
749                                 ts->u.run.rs_blocks_logged);
750         else if (ts->ts_type == JBD2_STATS_CHECKPOINT)
751                 seq_printf(seq, "%-4s %-5lu %48s %-5u %-5lu %-5lu %-5lu\n",
752                                 "C", ts->ts_tid, " ",
753                                 jiffies_to_msecs(ts->u.chp.cs_chp_time),
754                                 ts->u.chp.cs_written, ts->u.chp.cs_dropped,
755                                 ts->u.chp.cs_forced_to_close);
756         else
757                 J_ASSERT(0);
758         return 0;
759 }
760
761 static void jbd2_seq_history_stop(struct seq_file *seq, void *v)
762 {
763 }
764
765 static struct seq_operations jbd2_seq_history_ops = {
766         .start  = jbd2_seq_history_start,
767         .next   = jbd2_seq_history_next,
768         .stop   = jbd2_seq_history_stop,
769         .show   = jbd2_seq_history_show,
770 };
771
772 static int jbd2_seq_history_open(struct inode *inode, struct file *file)
773 {
774         journal_t *journal = PDE(inode)->data;
775         struct jbd2_stats_proc_session *s;
776         int rc, size;
777
778         s = kmalloc(sizeof(*s), GFP_KERNEL);
779         if (s == NULL)
780                 return -ENOMEM;
781         size = sizeof(struct transaction_stats_s) * journal->j_history_max;
782         s->stats = kmalloc(size, GFP_KERNEL);
783         if (s->stats == NULL) {
784                 kfree(s);
785                 return -ENOMEM;
786         }
787         spin_lock(&journal->j_history_lock);
788         memcpy(s->stats, journal->j_history, size);
789         s->max = journal->j_history_max;
790         s->start = journal->j_history_cur % s->max;
791         spin_unlock(&journal->j_history_lock);
792
793         rc = seq_open(file, &jbd2_seq_history_ops);
794         if (rc == 0) {
795                 struct seq_file *m = file->private_data;
796                 m->private = s;
797         } else {
798                 kfree(s->stats);
799                 kfree(s);
800         }
801         return rc;
802
803 }
804
805 static int jbd2_seq_history_release(struct inode *inode, struct file *file)
806 {
807         struct seq_file *seq = file->private_data;
808         struct jbd2_stats_proc_session *s = seq->private;
809
810         kfree(s->stats);
811         kfree(s);
812         return seq_release(inode, file);
813 }
814
815 static struct file_operations jbd2_seq_history_fops = {
816         .owner          = THIS_MODULE,
817         .open           = jbd2_seq_history_open,
818         .read           = seq_read,
819         .llseek         = seq_lseek,
820         .release        = jbd2_seq_history_release,
821 };
822
823 static void *jbd2_seq_info_start(struct seq_file *seq, loff_t *pos)
824 {
825         return *pos ? NULL : SEQ_START_TOKEN;
826 }
827
828 static void *jbd2_seq_info_next(struct seq_file *seq, void *v, loff_t *pos)
829 {
830         return NULL;
831 }
832
833 static int jbd2_seq_info_show(struct seq_file *seq, void *v)
834 {
835         struct jbd2_stats_proc_session *s = seq->private;
836
837         if (v != SEQ_START_TOKEN)
838                 return 0;
839         seq_printf(seq, "%lu transaction, each upto %u blocks\n",
840                         s->stats->ts_tid,
841                         s->journal->j_max_transaction_buffers);
842         if (s->stats->ts_tid == 0)
843                 return 0;
844         seq_printf(seq, "average: \n  %ums waiting for transaction\n",
845             jiffies_to_msecs(s->stats->u.run.rs_wait / s->stats->ts_tid));
846         seq_printf(seq, "  %ums running transaction\n",
847             jiffies_to_msecs(s->stats->u.run.rs_running / s->stats->ts_tid));
848         seq_printf(seq, "  %ums transaction was being locked\n",
849             jiffies_to_msecs(s->stats->u.run.rs_locked / s->stats->ts_tid));
850         seq_printf(seq, "  %ums flushing data (in ordered mode)\n",
851             jiffies_to_msecs(s->stats->u.run.rs_flushing / s->stats->ts_tid));
852         seq_printf(seq, "  %ums logging transaction\n",
853             jiffies_to_msecs(s->stats->u.run.rs_logging / s->stats->ts_tid));
854         seq_printf(seq, "  %lluus average transaction commit time\n",
855                    div_u64(s->journal->j_average_commit_time, 1000));
856         seq_printf(seq, "  %lu handles per transaction\n",
857             s->stats->u.run.rs_handle_count / s->stats->ts_tid);
858         seq_printf(seq, "  %lu blocks per transaction\n",
859             s->stats->u.run.rs_blocks / s->stats->ts_tid);
860         seq_printf(seq, "  %lu logged blocks per transaction\n",
861             s->stats->u.run.rs_blocks_logged / s->stats->ts_tid);
862         return 0;
863 }
864
865 static void jbd2_seq_info_stop(struct seq_file *seq, void *v)
866 {
867 }
868
869 static struct seq_operations jbd2_seq_info_ops = {
870         .start  = jbd2_seq_info_start,
871         .next   = jbd2_seq_info_next,
872         .stop   = jbd2_seq_info_stop,
873         .show   = jbd2_seq_info_show,
874 };
875
876 static int jbd2_seq_info_open(struct inode *inode, struct file *file)
877 {
878         journal_t *journal = PDE(inode)->data;
879         struct jbd2_stats_proc_session *s;
880         int rc, size;
881
882         s = kmalloc(sizeof(*s), GFP_KERNEL);
883         if (s == NULL)
884                 return -ENOMEM;
885         size = sizeof(struct transaction_stats_s);
886         s->stats = kmalloc(size, GFP_KERNEL);
887         if (s->stats == NULL) {
888                 kfree(s);
889                 return -ENOMEM;
890         }
891         spin_lock(&journal->j_history_lock);
892         memcpy(s->stats, &journal->j_stats, size);
893         s->journal = journal;
894         spin_unlock(&journal->j_history_lock);
895
896         rc = seq_open(file, &jbd2_seq_info_ops);
897         if (rc == 0) {
898                 struct seq_file *m = file->private_data;
899                 m->private = s;
900         } else {
901                 kfree(s->stats);
902                 kfree(s);
903         }
904         return rc;
905
906 }
907
908 static int jbd2_seq_info_release(struct inode *inode, struct file *file)
909 {
910         struct seq_file *seq = file->private_data;
911         struct jbd2_stats_proc_session *s = seq->private;
912         kfree(s->stats);
913         kfree(s);
914         return seq_release(inode, file);
915 }
916
917 static struct file_operations jbd2_seq_info_fops = {
918         .owner          = THIS_MODULE,
919         .open           = jbd2_seq_info_open,
920         .read           = seq_read,
921         .llseek         = seq_lseek,
922         .release        = jbd2_seq_info_release,
923 };
924
925 static struct proc_dir_entry *proc_jbd2_stats;
926
927 static void jbd2_stats_proc_init(journal_t *journal)
928 {
929         journal->j_proc_entry = proc_mkdir(journal->j_devname, proc_jbd2_stats);
930         if (journal->j_proc_entry) {
931                 proc_create_data("history", S_IRUGO, journal->j_proc_entry,
932                                  &jbd2_seq_history_fops, journal);
933                 proc_create_data("info", S_IRUGO, journal->j_proc_entry,
934                                  &jbd2_seq_info_fops, journal);
935         }
936 }
937
938 static void jbd2_stats_proc_exit(journal_t *journal)
939 {
940         remove_proc_entry("info", journal->j_proc_entry);
941         remove_proc_entry("history", journal->j_proc_entry);
942         remove_proc_entry(journal->j_devname, proc_jbd2_stats);
943 }
944
945 static void journal_init_stats(journal_t *journal)
946 {
947         int size;
948
949         if (!proc_jbd2_stats)
950                 return;
951
952         journal->j_history_max = 100;
953         size = sizeof(struct transaction_stats_s) * journal->j_history_max;
954         journal->j_history = kzalloc(size, GFP_KERNEL);
955         if (!journal->j_history) {
956                 journal->j_history_max = 0;
957                 return;
958         }
959         spin_lock_init(&journal->j_history_lock);
960 }
961
962 /*
963  * Management for journal control blocks: functions to create and
964  * destroy journal_t structures, and to initialise and read existing
965  * journal blocks from disk.  */
966
967 /* First: create and setup a journal_t object in memory.  We initialise
968  * very few fields yet: that has to wait until we have created the
969  * journal structures from from scratch, or loaded them from disk. */
970
971 static journal_t * journal_init_common (void)
972 {
973         journal_t *journal;
974         int err;
975
976         journal = kzalloc(sizeof(*journal), GFP_KERNEL|__GFP_NOFAIL);
977         if (!journal)
978                 goto fail;
979
980         init_waitqueue_head(&journal->j_wait_transaction_locked);
981         init_waitqueue_head(&journal->j_wait_logspace);
982         init_waitqueue_head(&journal->j_wait_done_commit);
983         init_waitqueue_head(&journal->j_wait_checkpoint);
984         init_waitqueue_head(&journal->j_wait_commit);
985         init_waitqueue_head(&journal->j_wait_updates);
986         mutex_init(&journal->j_barrier);
987         mutex_init(&journal->j_checkpoint_mutex);
988         spin_lock_init(&journal->j_revoke_lock);
989         spin_lock_init(&journal->j_list_lock);
990         spin_lock_init(&journal->j_state_lock);
991
992         journal->j_commit_interval = (HZ * JBD2_DEFAULT_MAX_COMMIT_AGE);
993         journal->j_min_batch_time = 0;
994         journal->j_max_batch_time = 15000; /* 15ms */
995
996         /* The journal is marked for error until we succeed with recovery! */
997         journal->j_flags = JBD2_ABORT;
998
999         /* Set up a default-sized revoke table for the new mount. */
1000         err = jbd2_journal_init_revoke(journal, JOURNAL_REVOKE_DEFAULT_HASH);
1001         if (err) {
1002                 kfree(journal);
1003                 goto fail;
1004         }
1005
1006         journal_init_stats(journal);
1007
1008         return journal;
1009 fail:
1010         return NULL;
1011 }
1012
1013 /* jbd2_journal_init_dev and jbd2_journal_init_inode:
1014  *
1015  * Create a journal structure assigned some fixed set of disk blocks to
1016  * the journal.  We don't actually touch those disk blocks yet, but we
1017  * need to set up all of the mapping information to tell the journaling
1018  * system where the journal blocks are.
1019  *
1020  */
1021
1022 /**
1023  *  journal_t * jbd2_journal_init_dev() - creates and initialises a journal structure
1024  *  @bdev: Block device on which to create the journal
1025  *  @fs_dev: Device which hold journalled filesystem for this journal.
1026  *  @start: Block nr Start of journal.
1027  *  @len:  Length of the journal in blocks.
1028  *  @blocksize: blocksize of journalling device
1029  *
1030  *  Returns: a newly created journal_t *
1031  *
1032  *  jbd2_journal_init_dev creates a journal which maps a fixed contiguous
1033  *  range of blocks on an arbitrary block device.
1034  *
1035  */
1036 journal_t * jbd2_journal_init_dev(struct block_device *bdev,
1037                         struct block_device *fs_dev,
1038                         unsigned long long start, int len, int blocksize)
1039 {
1040         journal_t *journal = journal_init_common();
1041         struct buffer_head *bh;
1042         char *p;
1043         int n;
1044
1045         if (!journal)
1046                 return NULL;
1047
1048         /* journal descriptor can store up to n blocks -bzzz */
1049         journal->j_blocksize = blocksize;
1050         jbd2_stats_proc_init(journal);
1051         n = journal->j_blocksize / sizeof(journal_block_tag_t);
1052         journal->j_wbufsize = n;
1053         journal->j_wbuf = kmalloc(n * sizeof(struct buffer_head*), GFP_KERNEL);
1054         if (!journal->j_wbuf) {
1055                 printk(KERN_ERR "%s: Cant allocate bhs for commit thread\n",
1056                         __func__);
1057                 goto out_err;
1058         }
1059         journal->j_dev = bdev;
1060         journal->j_fs_dev = fs_dev;
1061         journal->j_blk_offset = start;
1062         journal->j_maxlen = len;
1063         bdevname(journal->j_dev, journal->j_devname);
1064         p = journal->j_devname;
1065         while ((p = strchr(p, '/')))
1066                 *p = '!';
1067
1068         bh = __getblk(journal->j_dev, start, journal->j_blocksize);
1069         if (!bh) {
1070                 printk(KERN_ERR
1071                        "%s: Cannot get buffer for journal superblock\n",
1072                        __func__);
1073                 goto out_err;
1074         }
1075         journal->j_sb_buffer = bh;
1076         journal->j_superblock = (journal_superblock_t *)bh->b_data;
1077
1078         return journal;
1079 out_err:
1080         jbd2_stats_proc_exit(journal);
1081         kfree(journal);
1082         return NULL;
1083 }
1084
1085 /**
1086  *  journal_t * jbd2_journal_init_inode () - creates a journal which maps to a inode.
1087  *  @inode: An inode to create the journal in
1088  *
1089  * jbd2_journal_init_inode creates a journal which maps an on-disk inode as
1090  * the journal.  The inode must exist already, must support bmap() and
1091  * must have all data blocks preallocated.
1092  */
1093 journal_t * jbd2_journal_init_inode (struct inode *inode)
1094 {
1095         struct buffer_head *bh;
1096         journal_t *journal = journal_init_common();
1097         char *p;
1098         int err;
1099         int n;
1100         unsigned long long blocknr;
1101
1102         if (!journal)
1103                 return NULL;
1104
1105         journal->j_dev = journal->j_fs_dev = inode->i_sb->s_bdev;
1106         journal->j_inode = inode;
1107         bdevname(journal->j_dev, journal->j_devname);
1108         p = journal->j_devname;
1109         while ((p = strchr(p, '/')))
1110                 *p = '!';
1111         p = journal->j_devname + strlen(journal->j_devname);
1112         sprintf(p, ":%lu", journal->j_inode->i_ino);
1113         jbd_debug(1,
1114                   "journal %p: inode %s/%ld, size %Ld, bits %d, blksize %ld\n",
1115                   journal, inode->i_sb->s_id, inode->i_ino,
1116                   (long long) inode->i_size,
1117                   inode->i_sb->s_blocksize_bits, inode->i_sb->s_blocksize);
1118
1119         journal->j_maxlen = inode->i_size >> inode->i_sb->s_blocksize_bits;
1120         journal->j_blocksize = inode->i_sb->s_blocksize;
1121         jbd2_stats_proc_init(journal);
1122
1123         /* journal descriptor can store up to n blocks -bzzz */
1124         n = journal->j_blocksize / sizeof(journal_block_tag_t);
1125         journal->j_wbufsize = n;
1126         journal->j_wbuf = kmalloc(n * sizeof(struct buffer_head*), GFP_KERNEL);
1127         if (!journal->j_wbuf) {
1128                 printk(KERN_ERR "%s: Cant allocate bhs for commit thread\n",
1129                         __func__);
1130                 goto out_err;
1131         }
1132
1133         err = jbd2_journal_bmap(journal, 0, &blocknr);
1134         /* If that failed, give up */
1135         if (err) {
1136                 printk(KERN_ERR "%s: Cannnot locate journal superblock\n",
1137                        __func__);
1138                 goto out_err;
1139         }
1140
1141         bh = __getblk(journal->j_dev, blocknr, journal->j_blocksize);
1142         if (!bh) {
1143                 printk(KERN_ERR
1144                        "%s: Cannot get buffer for journal superblock\n",
1145                        __func__);
1146                 goto out_err;
1147         }
1148         journal->j_sb_buffer = bh;
1149         journal->j_superblock = (journal_superblock_t *)bh->b_data;
1150
1151         return journal;
1152 out_err:
1153         jbd2_stats_proc_exit(journal);
1154         kfree(journal);
1155         return NULL;
1156 }
1157
1158 /*
1159  * If the journal init or create aborts, we need to mark the journal
1160  * superblock as being NULL to prevent the journal destroy from writing
1161  * back a bogus superblock.
1162  */
1163 static void journal_fail_superblock (journal_t *journal)
1164 {
1165         struct buffer_head *bh = journal->j_sb_buffer;
1166         brelse(bh);
1167         journal->j_sb_buffer = NULL;
1168 }
1169
1170 /*
1171  * Given a journal_t structure, initialise the various fields for
1172  * startup of a new journaling session.  We use this both when creating
1173  * a journal, and after recovering an old journal to reset it for
1174  * subsequent use.
1175  */
1176
1177 static int journal_reset(journal_t *journal)
1178 {
1179         journal_superblock_t *sb = journal->j_superblock;
1180         unsigned long long first, last;
1181
1182         first = be32_to_cpu(sb->s_first);
1183         last = be32_to_cpu(sb->s_maxlen);
1184
1185         journal->j_first = first;
1186         journal->j_last = last;
1187
1188         journal->j_head = first;
1189         journal->j_tail = first;
1190         journal->j_free = last - first;
1191
1192         journal->j_tail_sequence = journal->j_transaction_sequence;
1193         journal->j_commit_sequence = journal->j_transaction_sequence - 1;
1194         journal->j_commit_request = journal->j_commit_sequence;
1195
1196         journal->j_max_transaction_buffers = journal->j_maxlen / 4;
1197
1198         /* Add the dynamic fields and write it to disk. */
1199         jbd2_journal_update_superblock(journal, 1);
1200         return jbd2_journal_start_thread(journal);
1201 }
1202
1203 /**
1204  * void jbd2_journal_update_superblock() - Update journal sb on disk.
1205  * @journal: The journal to update.
1206  * @wait: Set to '0' if you don't want to wait for IO completion.
1207  *
1208  * Update a journal's dynamic superblock fields and write it to disk,
1209  * optionally waiting for the IO to complete.
1210  */
1211 void jbd2_journal_update_superblock(journal_t *journal, int wait)
1212 {
1213         journal_superblock_t *sb = journal->j_superblock;
1214         struct buffer_head *bh = journal->j_sb_buffer;
1215
1216         /*
1217          * As a special case, if the on-disk copy is already marked as needing
1218          * no recovery (s_start == 0) and there are no outstanding transactions
1219          * in the filesystem, then we can safely defer the superblock update
1220          * until the next commit by setting JBD2_FLUSHED.  This avoids
1221          * attempting a write to a potential-readonly device.
1222          */
1223         if (sb->s_start == 0 && journal->j_tail_sequence ==
1224                                 journal->j_transaction_sequence) {
1225                 jbd_debug(1,"JBD: Skipping superblock update on recovered sb "
1226                         "(start %ld, seq %d, errno %d)\n",
1227                         journal->j_tail, journal->j_tail_sequence,
1228                         journal->j_errno);
1229                 goto out;
1230         }
1231
1232         if (buffer_write_io_error(bh)) {
1233                 /*
1234                  * Oh, dear.  A previous attempt to write the journal
1235                  * superblock failed.  This could happen because the
1236                  * USB device was yanked out.  Or it could happen to
1237                  * be a transient write error and maybe the block will
1238                  * be remapped.  Nothing we can do but to retry the
1239                  * write and hope for the best.
1240                  */
1241                 printk(KERN_ERR "JBD2: previous I/O error detected "
1242                        "for journal superblock update for %s.\n",
1243                        journal->j_devname);
1244                 clear_buffer_write_io_error(bh);
1245                 set_buffer_uptodate(bh);
1246         }
1247
1248         spin_lock(&journal->j_state_lock);
1249         jbd_debug(1,"JBD: updating superblock (start %ld, seq %d, errno %d)\n",
1250                   journal->j_tail, journal->j_tail_sequence, journal->j_errno);
1251
1252         sb->s_sequence = cpu_to_be32(journal->j_tail_sequence);
1253         sb->s_start    = cpu_to_be32(journal->j_tail);
1254         sb->s_errno    = cpu_to_be32(journal->j_errno);
1255         spin_unlock(&journal->j_state_lock);
1256
1257         BUFFER_TRACE(bh, "marking dirty");
1258         mark_buffer_dirty(bh);
1259         if (wait) {
1260                 sync_dirty_buffer(bh);
1261                 if (buffer_write_io_error(bh)) {
1262                         printk(KERN_ERR "JBD2: I/O error detected "
1263                                "when updating journal superblock for %s.\n",
1264                                journal->j_devname);
1265                         clear_buffer_write_io_error(bh);
1266                         set_buffer_uptodate(bh);
1267                 }
1268         } else
1269                 ll_rw_block(SWRITE, 1, &bh);
1270
1271 out:
1272         /* If we have just flushed the log (by marking s_start==0), then
1273          * any future commit will have to be careful to update the
1274          * superblock again to re-record the true start of the log. */
1275
1276         spin_lock(&journal->j_state_lock);
1277         if (sb->s_start)
1278                 journal->j_flags &= ~JBD2_FLUSHED;
1279         else
1280                 journal->j_flags |= JBD2_FLUSHED;
1281         spin_unlock(&journal->j_state_lock);
1282 }
1283
1284 /*
1285  * Read the superblock for a given journal, performing initial
1286  * validation of the format.
1287  */
1288
1289 static int journal_get_superblock(journal_t *journal)
1290 {
1291         struct buffer_head *bh;
1292         journal_superblock_t *sb;
1293         int err = -EIO;
1294
1295         bh = journal->j_sb_buffer;
1296
1297         J_ASSERT(bh != NULL);
1298         if (!buffer_uptodate(bh)) {
1299                 ll_rw_block(READ, 1, &bh);
1300                 wait_on_buffer(bh);
1301                 if (!buffer_uptodate(bh)) {
1302                         printk (KERN_ERR
1303                                 "JBD: IO error reading journal superblock\n");
1304                         goto out;
1305                 }
1306         }
1307
1308         sb = journal->j_superblock;
1309
1310         err = -EINVAL;
1311
1312         if (sb->s_header.h_magic != cpu_to_be32(JBD2_MAGIC_NUMBER) ||
1313             sb->s_blocksize != cpu_to_be32(journal->j_blocksize)) {
1314                 printk(KERN_WARNING "JBD: no valid journal superblock found\n");
1315                 goto out;
1316         }
1317
1318         switch(be32_to_cpu(sb->s_header.h_blocktype)) {
1319         case JBD2_SUPERBLOCK_V1:
1320                 journal->j_format_version = 1;
1321                 break;
1322         case JBD2_SUPERBLOCK_V2:
1323                 journal->j_format_version = 2;
1324                 break;
1325         default:
1326                 printk(KERN_WARNING "JBD: unrecognised superblock format ID\n");
1327                 goto out;
1328         }
1329
1330         if (be32_to_cpu(sb->s_maxlen) < journal->j_maxlen)
1331                 journal->j_maxlen = be32_to_cpu(sb->s_maxlen);
1332         else if (be32_to_cpu(sb->s_maxlen) > journal->j_maxlen) {
1333                 printk (KERN_WARNING "JBD: journal file too short\n");
1334                 goto out;
1335         }
1336
1337         return 0;
1338
1339 out:
1340         journal_fail_superblock(journal);
1341         return err;
1342 }
1343
1344 /*
1345  * Load the on-disk journal superblock and read the key fields into the
1346  * journal_t.
1347  */
1348
1349 static int load_superblock(journal_t *journal)
1350 {
1351         int err;
1352         journal_superblock_t *sb;
1353
1354         err = journal_get_superblock(journal);
1355         if (err)
1356                 return err;
1357
1358         sb = journal->j_superblock;
1359
1360         journal->j_tail_sequence = be32_to_cpu(sb->s_sequence);
1361         journal->j_tail = be32_to_cpu(sb->s_start);
1362         journal->j_first = be32_to_cpu(sb->s_first);
1363         journal->j_last = be32_to_cpu(sb->s_maxlen);
1364         journal->j_errno = be32_to_cpu(sb->s_errno);
1365
1366         return 0;
1367 }
1368
1369
1370 /**
1371  * int jbd2_journal_load() - Read journal from disk.
1372  * @journal: Journal to act on.
1373  *
1374  * Given a journal_t structure which tells us which disk blocks contain
1375  * a journal, read the journal from disk to initialise the in-memory
1376  * structures.
1377  */
1378 int jbd2_journal_load(journal_t *journal)
1379 {
1380         int err;
1381         journal_superblock_t *sb;
1382
1383         err = load_superblock(journal);
1384         if (err)
1385                 return err;
1386
1387         sb = journal->j_superblock;
1388         /* If this is a V2 superblock, then we have to check the
1389          * features flags on it. */
1390
1391         if (journal->j_format_version >= 2) {
1392                 if ((sb->s_feature_ro_compat &
1393                      ~cpu_to_be32(JBD2_KNOWN_ROCOMPAT_FEATURES)) ||
1394                     (sb->s_feature_incompat &
1395                      ~cpu_to_be32(JBD2_KNOWN_INCOMPAT_FEATURES))) {
1396                         printk (KERN_WARNING
1397                                 "JBD: Unrecognised features on journal\n");
1398                         return -EINVAL;
1399                 }
1400         }
1401
1402         /* Let the recovery code check whether it needs to recover any
1403          * data from the journal. */
1404         if (jbd2_journal_recover(journal))
1405                 goto recovery_error;
1406
1407         /* OK, we've finished with the dynamic journal bits:
1408          * reinitialise the dynamic contents of the superblock in memory
1409          * and reset them on disk. */
1410         if (journal_reset(journal))
1411                 goto recovery_error;
1412
1413         journal->j_flags &= ~JBD2_ABORT;
1414         journal->j_flags |= JBD2_LOADED;
1415         return 0;
1416
1417 recovery_error:
1418         printk (KERN_WARNING "JBD: recovery failed\n");
1419         return -EIO;
1420 }
1421
1422 /**
1423  * void jbd2_journal_destroy() - Release a journal_t structure.
1424  * @journal: Journal to act on.
1425  *
1426  * Release a journal_t structure once it is no longer in use by the
1427  * journaled object.
1428  * Return <0 if we couldn't clean up the journal.
1429  */
1430 int jbd2_journal_destroy(journal_t *journal)
1431 {
1432         int err = 0;
1433
1434         /* Wait for the commit thread to wake up and die. */
1435         journal_kill_thread(journal);
1436
1437         /* Force a final log commit */
1438         if (journal->j_running_transaction)
1439                 jbd2_journal_commit_transaction(journal);
1440
1441         /* Force any old transactions to disk */
1442
1443         /* Totally anal locking here... */
1444         spin_lock(&journal->j_list_lock);
1445         while (journal->j_checkpoint_transactions != NULL) {
1446                 spin_unlock(&journal->j_list_lock);
1447                 mutex_lock(&journal->j_checkpoint_mutex);
1448                 jbd2_log_do_checkpoint(journal);
1449                 mutex_unlock(&journal->j_checkpoint_mutex);
1450                 spin_lock(&journal->j_list_lock);
1451         }
1452
1453         J_ASSERT(journal->j_running_transaction == NULL);
1454         J_ASSERT(journal->j_committing_transaction == NULL);
1455         J_ASSERT(journal->j_checkpoint_transactions == NULL);
1456         spin_unlock(&journal->j_list_lock);
1457
1458         if (journal->j_sb_buffer) {
1459                 if (!is_journal_aborted(journal)) {
1460                         /* We can now mark the journal as empty. */
1461                         journal->j_tail = 0;
1462                         journal->j_tail_sequence =
1463                                 ++journal->j_transaction_sequence;
1464                         jbd2_journal_update_superblock(journal, 1);
1465                 } else {
1466                         err = -EIO;
1467                 }
1468                 brelse(journal->j_sb_buffer);
1469         }
1470
1471         if (journal->j_proc_entry)
1472                 jbd2_stats_proc_exit(journal);
1473         if (journal->j_inode)
1474                 iput(journal->j_inode);
1475         if (journal->j_revoke)
1476                 jbd2_journal_destroy_revoke(journal);
1477         kfree(journal->j_wbuf);
1478         kfree(journal);
1479
1480         return err;
1481 }
1482
1483
1484 /**
1485  *int jbd2_journal_check_used_features () - Check if features specified are used.
1486  * @journal: Journal to check.
1487  * @compat: bitmask of compatible features
1488  * @ro: bitmask of features that force read-only mount
1489  * @incompat: bitmask of incompatible features
1490  *
1491  * Check whether the journal uses all of a given set of
1492  * features.  Return true (non-zero) if it does.
1493  **/
1494
1495 int jbd2_journal_check_used_features (journal_t *journal, unsigned long compat,
1496                                  unsigned long ro, unsigned long incompat)
1497 {
1498         journal_superblock_t *sb;
1499
1500         if (!compat && !ro && !incompat)
1501                 return 1;
1502         if (journal->j_format_version == 1)
1503                 return 0;
1504
1505         sb = journal->j_superblock;
1506
1507         if (((be32_to_cpu(sb->s_feature_compat) & compat) == compat) &&
1508             ((be32_to_cpu(sb->s_feature_ro_compat) & ro) == ro) &&
1509             ((be32_to_cpu(sb->s_feature_incompat) & incompat) == incompat))
1510                 return 1;
1511
1512         return 0;
1513 }
1514
1515 /**
1516  * int jbd2_journal_check_available_features() - Check feature set in journalling layer
1517  * @journal: Journal to check.
1518  * @compat: bitmask of compatible features
1519  * @ro: bitmask of features that force read-only mount
1520  * @incompat: bitmask of incompatible features
1521  *
1522  * Check whether the journaling code supports the use of
1523  * all of a given set of features on this journal.  Return true
1524  * (non-zero) if it can. */
1525
1526 int jbd2_journal_check_available_features (journal_t *journal, unsigned long compat,
1527                                       unsigned long ro, unsigned long incompat)
1528 {
1529         journal_superblock_t *sb;
1530
1531         if (!compat && !ro && !incompat)
1532                 return 1;
1533
1534         sb = journal->j_superblock;
1535
1536         /* We can support any known requested features iff the
1537          * superblock is in version 2.  Otherwise we fail to support any
1538          * extended sb features. */
1539
1540         if (journal->j_format_version != 2)
1541                 return 0;
1542
1543         if ((compat   & JBD2_KNOWN_COMPAT_FEATURES) == compat &&
1544             (ro       & JBD2_KNOWN_ROCOMPAT_FEATURES) == ro &&
1545             (incompat & JBD2_KNOWN_INCOMPAT_FEATURES) == incompat)
1546                 return 1;
1547
1548         return 0;
1549 }
1550
1551 /**
1552  * int jbd2_journal_set_features () - Mark a given journal feature in the superblock
1553  * @journal: Journal to act on.
1554  * @compat: bitmask of compatible features
1555  * @ro: bitmask of features that force read-only mount
1556  * @incompat: bitmask of incompatible features
1557  *
1558  * Mark a given journal feature as present on the
1559  * superblock.  Returns true if the requested features could be set.
1560  *
1561  */
1562
1563 int jbd2_journal_set_features (journal_t *journal, unsigned long compat,
1564                           unsigned long ro, unsigned long incompat)
1565 {
1566         journal_superblock_t *sb;
1567
1568         if (jbd2_journal_check_used_features(journal, compat, ro, incompat))
1569                 return 1;
1570
1571         if (!jbd2_journal_check_available_features(journal, compat, ro, incompat))
1572                 return 0;
1573
1574         jbd_debug(1, "Setting new features 0x%lx/0x%lx/0x%lx\n",
1575                   compat, ro, incompat);
1576
1577         sb = journal->j_superblock;
1578
1579         sb->s_feature_compat    |= cpu_to_be32(compat);
1580         sb->s_feature_ro_compat |= cpu_to_be32(ro);
1581         sb->s_feature_incompat  |= cpu_to_be32(incompat);
1582
1583         return 1;
1584 }
1585
1586 /*
1587  * jbd2_journal_clear_features () - Clear a given journal feature in the
1588  *                                  superblock
1589  * @journal: Journal to act on.
1590  * @compat: bitmask of compatible features
1591  * @ro: bitmask of features that force read-only mount
1592  * @incompat: bitmask of incompatible features
1593  *
1594  * Clear a given journal feature as present on the
1595  * superblock.
1596  */
1597 void jbd2_journal_clear_features(journal_t *journal, unsigned long compat,
1598                                 unsigned long ro, unsigned long incompat)
1599 {
1600         journal_superblock_t *sb;
1601
1602         jbd_debug(1, "Clear features 0x%lx/0x%lx/0x%lx\n",
1603                   compat, ro, incompat);
1604
1605         sb = journal->j_superblock;
1606
1607         sb->s_feature_compat    &= ~cpu_to_be32(compat);
1608         sb->s_feature_ro_compat &= ~cpu_to_be32(ro);
1609         sb->s_feature_incompat  &= ~cpu_to_be32(incompat);
1610 }
1611 EXPORT_SYMBOL(jbd2_journal_clear_features);
1612
1613 /**
1614  * int jbd2_journal_update_format () - Update on-disk journal structure.
1615  * @journal: Journal to act on.
1616  *
1617  * Given an initialised but unloaded journal struct, poke about in the
1618  * on-disk structure to update it to the most recent supported version.
1619  */
1620 int jbd2_journal_update_format (journal_t *journal)
1621 {
1622         journal_superblock_t *sb;
1623         int err;
1624
1625         err = journal_get_superblock(journal);
1626         if (err)
1627                 return err;
1628
1629         sb = journal->j_superblock;
1630
1631         switch (be32_to_cpu(sb->s_header.h_blocktype)) {
1632         case JBD2_SUPERBLOCK_V2:
1633                 return 0;
1634         case JBD2_SUPERBLOCK_V1:
1635                 return journal_convert_superblock_v1(journal, sb);
1636         default:
1637                 break;
1638         }
1639         return -EINVAL;
1640 }
1641
1642 static int journal_convert_superblock_v1(journal_t *journal,
1643                                          journal_superblock_t *sb)
1644 {
1645         int offset, blocksize;
1646         struct buffer_head *bh;
1647
1648         printk(KERN_WARNING
1649                 "JBD: Converting superblock from version 1 to 2.\n");
1650
1651         /* Pre-initialise new fields to zero */
1652         offset = ((char *) &(sb->s_feature_compat)) - ((char *) sb);
1653         blocksize = be32_to_cpu(sb->s_blocksize);
1654         memset(&sb->s_feature_compat, 0, blocksize-offset);
1655
1656         sb->s_nr_users = cpu_to_be32(1);
1657         sb->s_header.h_blocktype = cpu_to_be32(JBD2_SUPERBLOCK_V2);
1658         journal->j_format_version = 2;
1659
1660         bh = journal->j_sb_buffer;
1661         BUFFER_TRACE(bh, "marking dirty");
1662         mark_buffer_dirty(bh);
1663         sync_dirty_buffer(bh);
1664         return 0;
1665 }
1666
1667
1668 /**
1669  * int jbd2_journal_flush () - Flush journal
1670  * @journal: Journal to act on.
1671  *
1672  * Flush all data for a given journal to disk and empty the journal.
1673  * Filesystems can use this when remounting readonly to ensure that
1674  * recovery does not need to happen on remount.
1675  */
1676
1677 int jbd2_journal_flush(journal_t *journal)
1678 {
1679         int err = 0;
1680         transaction_t *transaction = NULL;
1681         unsigned long old_tail;
1682
1683         spin_lock(&journal->j_state_lock);
1684
1685         /* Force everything buffered to the log... */
1686         if (journal->j_running_transaction) {
1687                 transaction = journal->j_running_transaction;
1688                 __jbd2_log_start_commit(journal, transaction->t_tid);
1689         } else if (journal->j_committing_transaction)
1690                 transaction = journal->j_committing_transaction;
1691
1692         /* Wait for the log commit to complete... */
1693         if (transaction) {
1694                 tid_t tid = transaction->t_tid;
1695
1696                 spin_unlock(&journal->j_state_lock);
1697                 jbd2_log_wait_commit(journal, tid);
1698         } else {
1699                 spin_unlock(&journal->j_state_lock);
1700         }
1701
1702         /* ...and flush everything in the log out to disk. */
1703         spin_lock(&journal->j_list_lock);
1704         while (!err && journal->j_checkpoint_transactions != NULL) {
1705                 spin_unlock(&journal->j_list_lock);
1706                 mutex_lock(&journal->j_checkpoint_mutex);
1707                 err = jbd2_log_do_checkpoint(journal);
1708                 mutex_unlock(&journal->j_checkpoint_mutex);
1709                 spin_lock(&journal->j_list_lock);
1710         }
1711         spin_unlock(&journal->j_list_lock);
1712
1713         if (is_journal_aborted(journal))
1714                 return -EIO;
1715
1716         jbd2_cleanup_journal_tail(journal);
1717
1718         /* Finally, mark the journal as really needing no recovery.
1719          * This sets s_start==0 in the underlying superblock, which is
1720          * the magic code for a fully-recovered superblock.  Any future
1721          * commits of data to the journal will restore the current
1722          * s_start value. */
1723         spin_lock(&journal->j_state_lock);
1724         old_tail = journal->j_tail;
1725         journal->j_tail = 0;
1726         spin_unlock(&journal->j_state_lock);
1727         jbd2_journal_update_superblock(journal, 1);
1728         spin_lock(&journal->j_state_lock);
1729         journal->j_tail = old_tail;
1730
1731         J_ASSERT(!journal->j_running_transaction);
1732         J_ASSERT(!journal->j_committing_transaction);
1733         J_ASSERT(!journal->j_checkpoint_transactions);
1734         J_ASSERT(journal->j_head == journal->j_tail);
1735         J_ASSERT(journal->j_tail_sequence == journal->j_transaction_sequence);
1736         spin_unlock(&journal->j_state_lock);
1737         return 0;
1738 }
1739
1740 /**
1741  * int jbd2_journal_wipe() - Wipe journal contents
1742  * @journal: Journal to act on.
1743  * @write: flag (see below)
1744  *
1745  * Wipe out all of the contents of a journal, safely.  This will produce
1746  * a warning if the journal contains any valid recovery information.
1747  * Must be called between journal_init_*() and jbd2_journal_load().
1748  *
1749  * If 'write' is non-zero, then we wipe out the journal on disk; otherwise
1750  * we merely suppress recovery.
1751  */
1752
1753 int jbd2_journal_wipe(journal_t *journal, int write)
1754 {
1755         journal_superblock_t *sb;
1756         int err = 0;
1757
1758         J_ASSERT (!(journal->j_flags & JBD2_LOADED));
1759
1760         err = load_superblock(journal);
1761         if (err)
1762                 return err;
1763
1764         sb = journal->j_superblock;
1765
1766         if (!journal->j_tail)
1767                 goto no_recovery;
1768
1769         printk (KERN_WARNING "JBD: %s recovery information on journal\n",
1770                 write ? "Clearing" : "Ignoring");
1771
1772         err = jbd2_journal_skip_recovery(journal);
1773         if (write)
1774                 jbd2_journal_update_superblock(journal, 1);
1775
1776  no_recovery:
1777         return err;
1778 }
1779
1780 /*
1781  * Journal abort has very specific semantics, which we describe
1782  * for journal abort.
1783  *
1784  * Two internal function, which provide abort to te jbd layer
1785  * itself are here.
1786  */
1787
1788 /*
1789  * Quick version for internal journal use (doesn't lock the journal).
1790  * Aborts hard --- we mark the abort as occurred, but do _nothing_ else,
1791  * and don't attempt to make any other journal updates.
1792  */
1793 void __jbd2_journal_abort_hard(journal_t *journal)
1794 {
1795         transaction_t *transaction;
1796
1797         if (journal->j_flags & JBD2_ABORT)
1798                 return;
1799
1800         printk(KERN_ERR "Aborting journal on device %s.\n",
1801                journal->j_devname);
1802
1803         spin_lock(&journal->j_state_lock);
1804         journal->j_flags |= JBD2_ABORT;
1805         transaction = journal->j_running_transaction;
1806         if (transaction)
1807                 __jbd2_log_start_commit(journal, transaction->t_tid);
1808         spin_unlock(&journal->j_state_lock);
1809 }
1810
1811 /* Soft abort: record the abort error status in the journal superblock,
1812  * but don't do any other IO. */
1813 static void __journal_abort_soft (journal_t *journal, int errno)
1814 {
1815         if (journal->j_flags & JBD2_ABORT)
1816                 return;
1817
1818         if (!journal->j_errno)
1819                 journal->j_errno = errno;
1820
1821         __jbd2_journal_abort_hard(journal);
1822
1823         if (errno)
1824                 jbd2_journal_update_superblock(journal, 1);
1825 }
1826
1827 /**
1828  * void jbd2_journal_abort () - Shutdown the journal immediately.
1829  * @journal: the journal to shutdown.
1830  * @errno:   an error number to record in the journal indicating
1831  *           the reason for the shutdown.
1832  *
1833  * Perform a complete, immediate shutdown of the ENTIRE
1834  * journal (not of a single transaction).  This operation cannot be
1835  * undone without closing and reopening the journal.
1836  *
1837  * The jbd2_journal_abort function is intended to support higher level error
1838  * recovery mechanisms such as the ext2/ext3 remount-readonly error
1839  * mode.
1840  *
1841  * Journal abort has very specific semantics.  Any existing dirty,
1842  * unjournaled buffers in the main filesystem will still be written to
1843  * disk by bdflush, but the journaling mechanism will be suspended
1844  * immediately and no further transaction commits will be honoured.
1845  *
1846  * Any dirty, journaled buffers will be written back to disk without
1847  * hitting the journal.  Atomicity cannot be guaranteed on an aborted
1848  * filesystem, but we _do_ attempt to leave as much data as possible
1849  * behind for fsck to use for cleanup.
1850  *
1851  * Any attempt to get a new transaction handle on a journal which is in
1852  * ABORT state will just result in an -EROFS error return.  A
1853  * jbd2_journal_stop on an existing handle will return -EIO if we have
1854  * entered abort state during the update.
1855  *
1856  * Recursive transactions are not disturbed by journal abort until the
1857  * final jbd2_journal_stop, which will receive the -EIO error.
1858  *
1859  * Finally, the jbd2_journal_abort call allows the caller to supply an errno
1860  * which will be recorded (if possible) in the journal superblock.  This
1861  * allows a client to record failure conditions in the middle of a
1862  * transaction without having to complete the transaction to record the
1863  * failure to disk.  ext3_error, for example, now uses this
1864  * functionality.
1865  *
1866  * Errors which originate from within the journaling layer will NOT
1867  * supply an errno; a null errno implies that absolutely no further
1868  * writes are done to the journal (unless there are any already in
1869  * progress).
1870  *
1871  */
1872
1873 void jbd2_journal_abort(journal_t *journal, int errno)
1874 {
1875         __journal_abort_soft(journal, errno);
1876 }
1877
1878 /**
1879  * int jbd2_journal_errno () - returns the journal's error state.
1880  * @journal: journal to examine.
1881  *
1882  * This is the errno numbet set with jbd2_journal_abort(), the last
1883  * time the journal was mounted - if the journal was stopped
1884  * without calling abort this will be 0.
1885  *
1886  * If the journal has been aborted on this mount time -EROFS will
1887  * be returned.
1888  */
1889 int jbd2_journal_errno(journal_t *journal)
1890 {
1891         int err;
1892
1893         spin_lock(&journal->j_state_lock);
1894         if (journal->j_flags & JBD2_ABORT)
1895                 err = -EROFS;
1896         else
1897                 err = journal->j_errno;
1898         spin_unlock(&journal->j_state_lock);
1899         return err;
1900 }
1901
1902 /**
1903  * int jbd2_journal_clear_err () - clears the journal's error state
1904  * @journal: journal to act on.
1905  *
1906  * An error must be cleared or Acked to take a FS out of readonly
1907  * mode.
1908  */
1909 int jbd2_journal_clear_err(journal_t *journal)
1910 {
1911         int err = 0;
1912
1913         spin_lock(&journal->j_state_lock);
1914         if (journal->j_flags & JBD2_ABORT)
1915                 err = -EROFS;
1916         else
1917                 journal->j_errno = 0;
1918         spin_unlock(&journal->j_state_lock);
1919         return err;
1920 }
1921
1922 /**
1923  * void jbd2_journal_ack_err() - Ack journal err.
1924  * @journal: journal to act on.
1925  *
1926  * An error must be cleared or Acked to take a FS out of readonly
1927  * mode.
1928  */
1929 void jbd2_journal_ack_err(journal_t *journal)
1930 {
1931         spin_lock(&journal->j_state_lock);
1932         if (journal->j_errno)
1933                 journal->j_flags |= JBD2_ACK_ERR;
1934         spin_unlock(&journal->j_state_lock);
1935 }
1936
1937 int jbd2_journal_blocks_per_page(struct inode *inode)
1938 {
1939         return 1 << (PAGE_CACHE_SHIFT - inode->i_sb->s_blocksize_bits);
1940 }
1941
1942 /*
1943  * helper functions to deal with 32 or 64bit block numbers.
1944  */
1945 size_t journal_tag_bytes(journal_t *journal)
1946 {
1947         if (JBD2_HAS_INCOMPAT_FEATURE(journal, JBD2_FEATURE_INCOMPAT_64BIT))
1948                 return JBD2_TAG_SIZE64;
1949         else
1950                 return JBD2_TAG_SIZE32;
1951 }
1952
1953 /*
1954  * Journal_head storage management
1955  */
1956 static struct kmem_cache *jbd2_journal_head_cache;
1957 #ifdef CONFIG_JBD2_DEBUG
1958 static atomic_t nr_journal_heads = ATOMIC_INIT(0);
1959 #endif
1960
1961 static int journal_init_jbd2_journal_head_cache(void)
1962 {
1963         int retval;
1964
1965         J_ASSERT(jbd2_journal_head_cache == NULL);
1966         jbd2_journal_head_cache = kmem_cache_create("jbd2_journal_head",
1967                                 sizeof(struct journal_head),
1968                                 0,              /* offset */
1969                                 SLAB_TEMPORARY, /* flags */
1970                                 NULL);          /* ctor */
1971         retval = 0;
1972         if (!jbd2_journal_head_cache) {
1973                 retval = -ENOMEM;
1974                 printk(KERN_EMERG "JBD: no memory for journal_head cache\n");
1975         }
1976         return retval;
1977 }
1978
1979 static void jbd2_journal_destroy_jbd2_journal_head_cache(void)
1980 {
1981         if (jbd2_journal_head_cache) {
1982                 kmem_cache_destroy(jbd2_journal_head_cache);
1983                 jbd2_journal_head_cache = NULL;
1984         }
1985 }
1986
1987 /*
1988  * journal_head splicing and dicing
1989  */
1990 static struct journal_head *journal_alloc_journal_head(void)
1991 {
1992         struct journal_head *ret;
1993         static unsigned long last_warning;
1994
1995 #ifdef CONFIG_JBD2_DEBUG
1996         atomic_inc(&nr_journal_heads);
1997 #endif
1998         ret = kmem_cache_alloc(jbd2_journal_head_cache, GFP_NOFS);
1999         if (!ret) {
2000                 jbd_debug(1, "out of memory for journal_head\n");
2001                 if (time_after(jiffies, last_warning + 5*HZ)) {
2002                         printk(KERN_NOTICE "ENOMEM in %s, retrying.\n",
2003                                __func__);
2004                         last_warning = jiffies;
2005                 }
2006                 while (!ret) {
2007                         yield();
2008                         ret = kmem_cache_alloc(jbd2_journal_head_cache, GFP_NOFS);
2009                 }
2010         }
2011         return ret;
2012 }
2013
2014 static void journal_free_journal_head(struct journal_head *jh)
2015 {
2016 #ifdef CONFIG_JBD2_DEBUG
2017         atomic_dec(&nr_journal_heads);
2018         memset(jh, JBD2_POISON_FREE, sizeof(*jh));
2019 #endif
2020         kmem_cache_free(jbd2_journal_head_cache, jh);
2021 }
2022
2023 /*
2024  * A journal_head is attached to a buffer_head whenever JBD has an
2025  * interest in the buffer.
2026  *
2027  * Whenever a buffer has an attached journal_head, its ->b_state:BH_JBD bit
2028  * is set.  This bit is tested in core kernel code where we need to take
2029  * JBD-specific actions.  Testing the zeroness of ->b_private is not reliable
2030  * there.
2031  *
2032  * When a buffer has its BH_JBD bit set, its ->b_count is elevated by one.
2033  *
2034  * When a buffer has its BH_JBD bit set it is immune from being released by
2035  * core kernel code, mainly via ->b_count.
2036  *
2037  * A journal_head may be detached from its buffer_head when the journal_head's
2038  * b_transaction, b_cp_transaction and b_next_transaction pointers are NULL.
2039  * Various places in JBD call jbd2_journal_remove_journal_head() to indicate that the
2040  * journal_head can be dropped if needed.
2041  *
2042  * Various places in the kernel want to attach a journal_head to a buffer_head
2043  * _before_ attaching the journal_head to a transaction.  To protect the
2044  * journal_head in this situation, jbd2_journal_add_journal_head elevates the
2045  * journal_head's b_jcount refcount by one.  The caller must call
2046  * jbd2_journal_put_journal_head() to undo this.
2047  *
2048  * So the typical usage would be:
2049  *
2050  *      (Attach a journal_head if needed.  Increments b_jcount)
2051  *      struct journal_head *jh = jbd2_journal_add_journal_head(bh);
2052  *      ...
2053  *      jh->b_transaction = xxx;
2054  *      jbd2_journal_put_journal_head(jh);
2055  *
2056  * Now, the journal_head's b_jcount is zero, but it is safe from being released
2057  * because it has a non-zero b_transaction.
2058  */
2059
2060 /*
2061  * Give a buffer_head a journal_head.
2062  *
2063  * Doesn't need the journal lock.
2064  * May sleep.
2065  */
2066 struct journal_head *jbd2_journal_add_journal_head(struct buffer_head *bh)
2067 {
2068         struct journal_head *jh;
2069         struct journal_head *new_jh = NULL;
2070
2071 repeat:
2072         if (!buffer_jbd(bh)) {
2073                 new_jh = journal_alloc_journal_head();
2074                 memset(new_jh, 0, sizeof(*new_jh));
2075         }
2076
2077         jbd_lock_bh_journal_head(bh);
2078         if (buffer_jbd(bh)) {
2079                 jh = bh2jh(bh);
2080         } else {
2081                 J_ASSERT_BH(bh,
2082                         (atomic_read(&bh->b_count) > 0) ||
2083                         (bh->b_page && bh->b_page->mapping));
2084
2085                 if (!new_jh) {
2086                         jbd_unlock_bh_journal_head(bh);
2087                         goto repeat;
2088                 }
2089
2090                 jh = new_jh;
2091                 new_jh = NULL;          /* We consumed it */
2092                 set_buffer_jbd(bh);
2093                 bh->b_private = jh;
2094                 jh->b_bh = bh;
2095                 get_bh(bh);
2096                 BUFFER_TRACE(bh, "added journal_head");
2097         }
2098         jh->b_jcount++;
2099         jbd_unlock_bh_journal_head(bh);
2100         if (new_jh)
2101                 journal_free_journal_head(new_jh);
2102         return bh->b_private;
2103 }
2104
2105 /*
2106  * Grab a ref against this buffer_head's journal_head.  If it ended up not
2107  * having a journal_head, return NULL
2108  */
2109 struct journal_head *jbd2_journal_grab_journal_head(struct buffer_head *bh)
2110 {
2111         struct journal_head *jh = NULL;
2112
2113         jbd_lock_bh_journal_head(bh);
2114         if (buffer_jbd(bh)) {
2115                 jh = bh2jh(bh);
2116                 jh->b_jcount++;
2117         }
2118         jbd_unlock_bh_journal_head(bh);
2119         return jh;
2120 }
2121
2122 static void __journal_remove_journal_head(struct buffer_head *bh)
2123 {
2124         struct journal_head *jh = bh2jh(bh);
2125
2126         J_ASSERT_JH(jh, jh->b_jcount >= 0);
2127
2128         get_bh(bh);
2129         if (jh->b_jcount == 0) {
2130                 if (jh->b_transaction == NULL &&
2131                                 jh->b_next_transaction == NULL &&
2132                                 jh->b_cp_transaction == NULL) {
2133                         J_ASSERT_JH(jh, jh->b_jlist == BJ_None);
2134                         J_ASSERT_BH(bh, buffer_jbd(bh));
2135                         J_ASSERT_BH(bh, jh2bh(jh) == bh);
2136                         BUFFER_TRACE(bh, "remove journal_head");
2137                         if (jh->b_frozen_data) {
2138                                 printk(KERN_WARNING "%s: freeing "
2139                                                 "b_frozen_data\n",
2140                                                 __func__);
2141                                 jbd2_free(jh->b_frozen_data, bh->b_size);
2142                         }
2143                         if (jh->b_committed_data) {
2144                                 printk(KERN_WARNING "%s: freeing "
2145                                                 "b_committed_data\n",
2146                                                 __func__);
2147                                 jbd2_free(jh->b_committed_data, bh->b_size);
2148                         }
2149                         bh->b_private = NULL;
2150                         jh->b_bh = NULL;        /* debug, really */
2151                         clear_buffer_jbd(bh);
2152                         __brelse(bh);
2153                         journal_free_journal_head(jh);
2154                 } else {
2155                         BUFFER_TRACE(bh, "journal_head was locked");
2156                 }
2157         }
2158 }
2159
2160 /*
2161  * jbd2_journal_remove_journal_head(): if the buffer isn't attached to a transaction
2162  * and has a zero b_jcount then remove and release its journal_head.   If we did
2163  * see that the buffer is not used by any transaction we also "logically"
2164  * decrement ->b_count.
2165  *
2166  * We in fact take an additional increment on ->b_count as a convenience,
2167  * because the caller usually wants to do additional things with the bh
2168  * after calling here.
2169  * The caller of jbd2_journal_remove_journal_head() *must* run __brelse(bh) at some
2170  * time.  Once the caller has run __brelse(), the buffer is eligible for
2171  * reaping by try_to_free_buffers().
2172  */
2173 void jbd2_journal_remove_journal_head(struct buffer_head *bh)
2174 {
2175         jbd_lock_bh_journal_head(bh);
2176         __journal_remove_journal_head(bh);
2177         jbd_unlock_bh_journal_head(bh);
2178 }
2179
2180 /*
2181  * Drop a reference on the passed journal_head.  If it fell to zero then try to
2182  * release the journal_head from the buffer_head.
2183  */
2184 void jbd2_journal_put_journal_head(struct journal_head *jh)
2185 {
2186         struct buffer_head *bh = jh2bh(jh);
2187
2188         jbd_lock_bh_journal_head(bh);
2189         J_ASSERT_JH(jh, jh->b_jcount > 0);
2190         --jh->b_jcount;
2191         if (!jh->b_jcount && !jh->b_transaction) {
2192                 __journal_remove_journal_head(bh);
2193                 __brelse(bh);
2194         }
2195         jbd_unlock_bh_journal_head(bh);
2196 }
2197
2198 /*
2199  * Initialize jbd inode head
2200  */
2201 void jbd2_journal_init_jbd_inode(struct jbd2_inode *jinode, struct inode *inode)
2202 {
2203         jinode->i_transaction = NULL;
2204         jinode->i_next_transaction = NULL;
2205         jinode->i_vfs_inode = inode;
2206         jinode->i_flags = 0;
2207         INIT_LIST_HEAD(&jinode->i_list);
2208 }
2209
2210 /*
2211  * Function to be called before we start removing inode from memory (i.e.,
2212  * clear_inode() is a fine place to be called from). It removes inode from
2213  * transaction's lists.
2214  */
2215 void jbd2_journal_release_jbd_inode(journal_t *journal,
2216                                     struct jbd2_inode *jinode)
2217 {
2218         int writeout = 0;
2219
2220         if (!journal)
2221                 return;
2222 restart:
2223         spin_lock(&journal->j_list_lock);
2224         /* Is commit writing out inode - we have to wait */
2225         if (jinode->i_flags & JI_COMMIT_RUNNING) {
2226                 wait_queue_head_t *wq;
2227                 DEFINE_WAIT_BIT(wait, &jinode->i_flags, __JI_COMMIT_RUNNING);
2228                 wq = bit_waitqueue(&jinode->i_flags, __JI_COMMIT_RUNNING);
2229                 prepare_to_wait(wq, &wait.wait, TASK_UNINTERRUPTIBLE);
2230                 spin_unlock(&journal->j_list_lock);
2231                 schedule();
2232                 finish_wait(wq, &wait.wait);
2233                 goto restart;
2234         }
2235
2236         /* Do we need to wait for data writeback? */
2237         if (journal->j_committing_transaction == jinode->i_transaction)
2238                 writeout = 1;
2239         if (jinode->i_transaction) {
2240                 list_del(&jinode->i_list);
2241                 jinode->i_transaction = NULL;
2242         }
2243         spin_unlock(&journal->j_list_lock);
2244 }
2245
2246 /*
2247  * debugfs tunables
2248  */
2249 #ifdef CONFIG_JBD2_DEBUG
2250 u8 jbd2_journal_enable_debug __read_mostly;
2251 EXPORT_SYMBOL(jbd2_journal_enable_debug);
2252
2253 #define JBD2_DEBUG_NAME "jbd2-debug"
2254
2255 static struct dentry *jbd2_debugfs_dir;
2256 static struct dentry *jbd2_debug;
2257
2258 static void __init jbd2_create_debugfs_entry(void)
2259 {
2260         jbd2_debugfs_dir = debugfs_create_dir("jbd2", NULL);
2261         if (jbd2_debugfs_dir)
2262                 jbd2_debug = debugfs_create_u8(JBD2_DEBUG_NAME, S_IRUGO,
2263                                                jbd2_debugfs_dir,
2264                                                &jbd2_journal_enable_debug);
2265 }
2266
2267 static void __exit jbd2_remove_debugfs_entry(void)
2268 {
2269         debugfs_remove(jbd2_debug);
2270         debugfs_remove(jbd2_debugfs_dir);
2271 }
2272
2273 #else
2274
2275 static void __init jbd2_create_debugfs_entry(void)
2276 {
2277 }
2278
2279 static void __exit jbd2_remove_debugfs_entry(void)
2280 {
2281 }
2282
2283 #endif
2284
2285 #ifdef CONFIG_PROC_FS
2286
2287 #define JBD2_STATS_PROC_NAME "fs/jbd2"
2288
2289 static void __init jbd2_create_jbd_stats_proc_entry(void)
2290 {
2291         proc_jbd2_stats = proc_mkdir(JBD2_STATS_PROC_NAME, NULL);
2292 }
2293
2294 static void __exit jbd2_remove_jbd_stats_proc_entry(void)
2295 {
2296         if (proc_jbd2_stats)
2297                 remove_proc_entry(JBD2_STATS_PROC_NAME, NULL);
2298 }
2299
2300 #else
2301
2302 #define jbd2_create_jbd_stats_proc_entry() do {} while (0)
2303 #define jbd2_remove_jbd_stats_proc_entry() do {} while (0)
2304
2305 #endif
2306
2307 struct kmem_cache *jbd2_handle_cache;
2308
2309 static int __init journal_init_handle_cache(void)
2310 {
2311         jbd2_handle_cache = kmem_cache_create("jbd2_journal_handle",
2312                                 sizeof(handle_t),
2313                                 0,              /* offset */
2314                                 SLAB_TEMPORARY, /* flags */
2315                                 NULL);          /* ctor */
2316         if (jbd2_handle_cache == NULL) {
2317                 printk(KERN_EMERG "JBD: failed to create handle cache\n");
2318                 return -ENOMEM;
2319         }
2320         return 0;
2321 }
2322
2323 static void jbd2_journal_destroy_handle_cache(void)
2324 {
2325         if (jbd2_handle_cache)
2326                 kmem_cache_destroy(jbd2_handle_cache);
2327 }
2328
2329 /*
2330  * Module startup and shutdown
2331  */
2332
2333 static int __init journal_init_caches(void)
2334 {
2335         int ret;
2336
2337         ret = jbd2_journal_init_revoke_caches();
2338         if (ret == 0)
2339                 ret = journal_init_jbd2_journal_head_cache();
2340         if (ret == 0)
2341                 ret = journal_init_handle_cache();
2342         return ret;
2343 }
2344
2345 static void jbd2_journal_destroy_caches(void)
2346 {
2347         jbd2_journal_destroy_revoke_caches();
2348         jbd2_journal_destroy_jbd2_journal_head_cache();
2349         jbd2_journal_destroy_handle_cache();
2350 }
2351
2352 static int __init journal_init(void)
2353 {
2354         int ret;
2355
2356         BUILD_BUG_ON(sizeof(struct journal_superblock_s) != 1024);
2357
2358         ret = journal_init_caches();
2359         if (ret == 0) {
2360                 jbd2_create_debugfs_entry();
2361                 jbd2_create_jbd_stats_proc_entry();
2362         } else {
2363                 jbd2_journal_destroy_caches();
2364         }
2365         return ret;
2366 }
2367
2368 static void __exit journal_exit(void)
2369 {
2370 #ifdef CONFIG_JBD2_DEBUG
2371         int n = atomic_read(&nr_journal_heads);
2372         if (n)
2373                 printk(KERN_EMERG "JBD: leaked %d journal_heads!\n", n);
2374 #endif
2375         jbd2_remove_debugfs_entry();
2376         jbd2_remove_jbd_stats_proc_entry();
2377         jbd2_journal_destroy_caches();
2378 }
2379
2380 MODULE_LICENSE("GPL");
2381 module_init(journal_init);
2382 module_exit(journal_exit);
2383