orinoco: checkpatch cleanup
[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 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 was started, and fills its tid in at *ptid
522  */
523 int jbd2_journal_start_commit(journal_t *journal, tid_t *ptid)
524 {
525         int ret = 0;
526
527         spin_lock(&journal->j_state_lock);
528         if (journal->j_running_transaction) {
529                 tid_t tid = journal->j_running_transaction->t_tid;
530
531                 ret = __jbd2_log_start_commit(journal, tid);
532                 if (ret && ptid)
533                         *ptid = tid;
534         } else if (journal->j_committing_transaction && ptid) {
535                 /*
536                  * If ext3_write_super() recently started a commit, then we
537                  * have to wait for completion of that transaction
538                  */
539                 *ptid = journal->j_committing_transaction->t_tid;
540                 ret = 1;
541         }
542         spin_unlock(&journal->j_state_lock);
543         return ret;
544 }
545
546 /*
547  * Wait for a specified commit to complete.
548  * The caller may not hold the journal lock.
549  */
550 int jbd2_log_wait_commit(journal_t *journal, tid_t tid)
551 {
552         int err = 0;
553
554 #ifdef CONFIG_JBD2_DEBUG
555         spin_lock(&journal->j_state_lock);
556         if (!tid_geq(journal->j_commit_request, tid)) {
557                 printk(KERN_EMERG
558                        "%s: error: j_commit_request=%d, tid=%d\n",
559                        __func__, journal->j_commit_request, tid);
560         }
561         spin_unlock(&journal->j_state_lock);
562 #endif
563         spin_lock(&journal->j_state_lock);
564         while (tid_gt(tid, journal->j_commit_sequence)) {
565                 jbd_debug(1, "JBD: want %d, j_commit_sequence=%d\n",
566                                   tid, journal->j_commit_sequence);
567                 wake_up(&journal->j_wait_commit);
568                 spin_unlock(&journal->j_state_lock);
569                 wait_event(journal->j_wait_done_commit,
570                                 !tid_gt(tid, journal->j_commit_sequence));
571                 spin_lock(&journal->j_state_lock);
572         }
573         spin_unlock(&journal->j_state_lock);
574
575         if (unlikely(is_journal_aborted(journal))) {
576                 printk(KERN_EMERG "journal commit I/O error\n");
577                 err = -EIO;
578         }
579         return err;
580 }
581
582 /*
583  * Log buffer allocation routines:
584  */
585
586 int jbd2_journal_next_log_block(journal_t *journal, unsigned long long *retp)
587 {
588         unsigned long blocknr;
589
590         spin_lock(&journal->j_state_lock);
591         J_ASSERT(journal->j_free > 1);
592
593         blocknr = journal->j_head;
594         journal->j_head++;
595         journal->j_free--;
596         if (journal->j_head == journal->j_last)
597                 journal->j_head = journal->j_first;
598         spin_unlock(&journal->j_state_lock);
599         return jbd2_journal_bmap(journal, blocknr, retp);
600 }
601
602 /*
603  * Conversion of logical to physical block numbers for the journal
604  *
605  * On external journals the journal blocks are identity-mapped, so
606  * this is a no-op.  If needed, we can use j_blk_offset - everything is
607  * ready.
608  */
609 int jbd2_journal_bmap(journal_t *journal, unsigned long blocknr,
610                  unsigned long long *retp)
611 {
612         int err = 0;
613         unsigned long long ret;
614
615         if (journal->j_inode) {
616                 ret = bmap(journal->j_inode, blocknr);
617                 if (ret)
618                         *retp = ret;
619                 else {
620                         printk(KERN_ALERT "%s: journal block not found "
621                                         "at offset %lu on %s\n",
622                                __func__, blocknr, journal->j_devname);
623                         err = -EIO;
624                         __journal_abort_soft(journal, err);
625                 }
626         } else {
627                 *retp = blocknr; /* +journal->j_blk_offset */
628         }
629         return err;
630 }
631
632 /*
633  * We play buffer_head aliasing tricks to write data/metadata blocks to
634  * the journal without copying their contents, but for journal
635  * descriptor blocks we do need to generate bona fide buffers.
636  *
637  * After the caller of jbd2_journal_get_descriptor_buffer() has finished modifying
638  * the buffer's contents they really should run flush_dcache_page(bh->b_page).
639  * But we don't bother doing that, so there will be coherency problems with
640  * mmaps of blockdevs which hold live JBD-controlled filesystems.
641  */
642 struct journal_head *jbd2_journal_get_descriptor_buffer(journal_t *journal)
643 {
644         struct buffer_head *bh;
645         unsigned long long blocknr;
646         int err;
647
648         err = jbd2_journal_next_log_block(journal, &blocknr);
649
650         if (err)
651                 return NULL;
652
653         bh = __getblk(journal->j_dev, blocknr, journal->j_blocksize);
654         if (!bh)
655                 return NULL;
656         lock_buffer(bh);
657         memset(bh->b_data, 0, journal->j_blocksize);
658         set_buffer_uptodate(bh);
659         unlock_buffer(bh);
660         BUFFER_TRACE(bh, "return this buffer");
661         return jbd2_journal_add_journal_head(bh);
662 }
663
664 struct jbd2_stats_proc_session {
665         journal_t *journal;
666         struct transaction_stats_s *stats;
667         int start;
668         int max;
669 };
670
671 static void *jbd2_history_skip_empty(struct jbd2_stats_proc_session *s,
672                                         struct transaction_stats_s *ts,
673                                         int first)
674 {
675         if (ts == s->stats + s->max)
676                 ts = s->stats;
677         if (!first && ts == s->stats + s->start)
678                 return NULL;
679         while (ts->ts_type == 0) {
680                 ts++;
681                 if (ts == s->stats + s->max)
682                         ts = s->stats;
683                 if (ts == s->stats + s->start)
684                         return NULL;
685         }
686         return ts;
687
688 }
689
690 static void *jbd2_seq_history_start(struct seq_file *seq, loff_t *pos)
691 {
692         struct jbd2_stats_proc_session *s = seq->private;
693         struct transaction_stats_s *ts;
694         int l = *pos;
695
696         if (l == 0)
697                 return SEQ_START_TOKEN;
698         ts = jbd2_history_skip_empty(s, s->stats + s->start, 1);
699         if (!ts)
700                 return NULL;
701         l--;
702         while (l) {
703                 ts = jbd2_history_skip_empty(s, ++ts, 0);
704                 if (!ts)
705                         break;
706                 l--;
707         }
708         return ts;
709 }
710
711 static void *jbd2_seq_history_next(struct seq_file *seq, void *v, loff_t *pos)
712 {
713         struct jbd2_stats_proc_session *s = seq->private;
714         struct transaction_stats_s *ts = v;
715
716         ++*pos;
717         if (v == SEQ_START_TOKEN)
718                 return jbd2_history_skip_empty(s, s->stats + s->start, 1);
719         else
720                 return jbd2_history_skip_empty(s, ++ts, 0);
721 }
722
723 static int jbd2_seq_history_show(struct seq_file *seq, void *v)
724 {
725         struct transaction_stats_s *ts = v;
726         if (v == SEQ_START_TOKEN) {
727                 seq_printf(seq, "%-4s %-5s %-5s %-5s %-5s %-5s %-5s %-6s %-5s "
728                                 "%-5s %-5s %-5s %-5s %-5s\n", "R/C", "tid",
729                                 "wait", "run", "lock", "flush", "log", "hndls",
730                                 "block", "inlog", "ctime", "write", "drop",
731                                 "close");
732                 return 0;
733         }
734         if (ts->ts_type == JBD2_STATS_RUN)
735                 seq_printf(seq, "%-4s %-5lu %-5u %-5u %-5u %-5u %-5u "
736                                 "%-6lu %-5lu %-5lu\n", "R", ts->ts_tid,
737                                 jiffies_to_msecs(ts->u.run.rs_wait),
738                                 jiffies_to_msecs(ts->u.run.rs_running),
739                                 jiffies_to_msecs(ts->u.run.rs_locked),
740                                 jiffies_to_msecs(ts->u.run.rs_flushing),
741                                 jiffies_to_msecs(ts->u.run.rs_logging),
742                                 ts->u.run.rs_handle_count,
743                                 ts->u.run.rs_blocks,
744                                 ts->u.run.rs_blocks_logged);
745         else if (ts->ts_type == JBD2_STATS_CHECKPOINT)
746                 seq_printf(seq, "%-4s %-5lu %48s %-5u %-5lu %-5lu %-5lu\n",
747                                 "C", ts->ts_tid, " ",
748                                 jiffies_to_msecs(ts->u.chp.cs_chp_time),
749                                 ts->u.chp.cs_written, ts->u.chp.cs_dropped,
750                                 ts->u.chp.cs_forced_to_close);
751         else
752                 J_ASSERT(0);
753         return 0;
754 }
755
756 static void jbd2_seq_history_stop(struct seq_file *seq, void *v)
757 {
758 }
759
760 static struct seq_operations jbd2_seq_history_ops = {
761         .start  = jbd2_seq_history_start,
762         .next   = jbd2_seq_history_next,
763         .stop   = jbd2_seq_history_stop,
764         .show   = jbd2_seq_history_show,
765 };
766
767 static int jbd2_seq_history_open(struct inode *inode, struct file *file)
768 {
769         journal_t *journal = PDE(inode)->data;
770         struct jbd2_stats_proc_session *s;
771         int rc, size;
772
773         s = kmalloc(sizeof(*s), GFP_KERNEL);
774         if (s == NULL)
775                 return -ENOMEM;
776         size = sizeof(struct transaction_stats_s) * journal->j_history_max;
777         s->stats = kmalloc(size, GFP_KERNEL);
778         if (s->stats == NULL) {
779                 kfree(s);
780                 return -ENOMEM;
781         }
782         spin_lock(&journal->j_history_lock);
783         memcpy(s->stats, journal->j_history, size);
784         s->max = journal->j_history_max;
785         s->start = journal->j_history_cur % s->max;
786         spin_unlock(&journal->j_history_lock);
787
788         rc = seq_open(file, &jbd2_seq_history_ops);
789         if (rc == 0) {
790                 struct seq_file *m = file->private_data;
791                 m->private = s;
792         } else {
793                 kfree(s->stats);
794                 kfree(s);
795         }
796         return rc;
797
798 }
799
800 static int jbd2_seq_history_release(struct inode *inode, struct file *file)
801 {
802         struct seq_file *seq = file->private_data;
803         struct jbd2_stats_proc_session *s = seq->private;
804
805         kfree(s->stats);
806         kfree(s);
807         return seq_release(inode, file);
808 }
809
810 static struct file_operations jbd2_seq_history_fops = {
811         .owner          = THIS_MODULE,
812         .open           = jbd2_seq_history_open,
813         .read           = seq_read,
814         .llseek         = seq_lseek,
815         .release        = jbd2_seq_history_release,
816 };
817
818 static void *jbd2_seq_info_start(struct seq_file *seq, loff_t *pos)
819 {
820         return *pos ? NULL : SEQ_START_TOKEN;
821 }
822
823 static void *jbd2_seq_info_next(struct seq_file *seq, void *v, loff_t *pos)
824 {
825         return NULL;
826 }
827
828 static int jbd2_seq_info_show(struct seq_file *seq, void *v)
829 {
830         struct jbd2_stats_proc_session *s = seq->private;
831
832         if (v != SEQ_START_TOKEN)
833                 return 0;
834         seq_printf(seq, "%lu transaction, each upto %u blocks\n",
835                         s->stats->ts_tid,
836                         s->journal->j_max_transaction_buffers);
837         if (s->stats->ts_tid == 0)
838                 return 0;
839         seq_printf(seq, "average: \n  %ums waiting for transaction\n",
840             jiffies_to_msecs(s->stats->u.run.rs_wait / s->stats->ts_tid));
841         seq_printf(seq, "  %ums running transaction\n",
842             jiffies_to_msecs(s->stats->u.run.rs_running / s->stats->ts_tid));
843         seq_printf(seq, "  %ums transaction was being locked\n",
844             jiffies_to_msecs(s->stats->u.run.rs_locked / s->stats->ts_tid));
845         seq_printf(seq, "  %ums flushing data (in ordered mode)\n",
846             jiffies_to_msecs(s->stats->u.run.rs_flushing / s->stats->ts_tid));
847         seq_printf(seq, "  %ums logging transaction\n",
848             jiffies_to_msecs(s->stats->u.run.rs_logging / s->stats->ts_tid));
849         seq_printf(seq, "  %lluus average transaction commit time\n",
850                    div_u64(s->journal->j_average_commit_time, 1000));
851         seq_printf(seq, "  %lu handles per transaction\n",
852             s->stats->u.run.rs_handle_count / s->stats->ts_tid);
853         seq_printf(seq, "  %lu blocks per transaction\n",
854             s->stats->u.run.rs_blocks / s->stats->ts_tid);
855         seq_printf(seq, "  %lu logged blocks per transaction\n",
856             s->stats->u.run.rs_blocks_logged / s->stats->ts_tid);
857         return 0;
858 }
859
860 static void jbd2_seq_info_stop(struct seq_file *seq, void *v)
861 {
862 }
863
864 static struct seq_operations jbd2_seq_info_ops = {
865         .start  = jbd2_seq_info_start,
866         .next   = jbd2_seq_info_next,
867         .stop   = jbd2_seq_info_stop,
868         .show   = jbd2_seq_info_show,
869 };
870
871 static int jbd2_seq_info_open(struct inode *inode, struct file *file)
872 {
873         journal_t *journal = PDE(inode)->data;
874         struct jbd2_stats_proc_session *s;
875         int rc, size;
876
877         s = kmalloc(sizeof(*s), GFP_KERNEL);
878         if (s == NULL)
879                 return -ENOMEM;
880         size = sizeof(struct transaction_stats_s);
881         s->stats = kmalloc(size, GFP_KERNEL);
882         if (s->stats == NULL) {
883                 kfree(s);
884                 return -ENOMEM;
885         }
886         spin_lock(&journal->j_history_lock);
887         memcpy(s->stats, &journal->j_stats, size);
888         s->journal = journal;
889         spin_unlock(&journal->j_history_lock);
890
891         rc = seq_open(file, &jbd2_seq_info_ops);
892         if (rc == 0) {
893                 struct seq_file *m = file->private_data;
894                 m->private = s;
895         } else {
896                 kfree(s->stats);
897                 kfree(s);
898         }
899         return rc;
900
901 }
902
903 static int jbd2_seq_info_release(struct inode *inode, struct file *file)
904 {
905         struct seq_file *seq = file->private_data;
906         struct jbd2_stats_proc_session *s = seq->private;
907         kfree(s->stats);
908         kfree(s);
909         return seq_release(inode, file);
910 }
911
912 static struct file_operations jbd2_seq_info_fops = {
913         .owner          = THIS_MODULE,
914         .open           = jbd2_seq_info_open,
915         .read           = seq_read,
916         .llseek         = seq_lseek,
917         .release        = jbd2_seq_info_release,
918 };
919
920 static struct proc_dir_entry *proc_jbd2_stats;
921
922 static void jbd2_stats_proc_init(journal_t *journal)
923 {
924         journal->j_proc_entry = proc_mkdir(journal->j_devname, proc_jbd2_stats);
925         if (journal->j_proc_entry) {
926                 proc_create_data("history", S_IRUGO, journal->j_proc_entry,
927                                  &jbd2_seq_history_fops, journal);
928                 proc_create_data("info", S_IRUGO, journal->j_proc_entry,
929                                  &jbd2_seq_info_fops, journal);
930         }
931 }
932
933 static void jbd2_stats_proc_exit(journal_t *journal)
934 {
935         remove_proc_entry("info", journal->j_proc_entry);
936         remove_proc_entry("history", journal->j_proc_entry);
937         remove_proc_entry(journal->j_devname, proc_jbd2_stats);
938 }
939
940 static void journal_init_stats(journal_t *journal)
941 {
942         int size;
943
944         if (!proc_jbd2_stats)
945                 return;
946
947         journal->j_history_max = 100;
948         size = sizeof(struct transaction_stats_s) * journal->j_history_max;
949         journal->j_history = kzalloc(size, GFP_KERNEL);
950         if (!journal->j_history) {
951                 journal->j_history_max = 0;
952                 return;
953         }
954         spin_lock_init(&journal->j_history_lock);
955 }
956
957 /*
958  * Management for journal control blocks: functions to create and
959  * destroy journal_t structures, and to initialise and read existing
960  * journal blocks from disk.  */
961
962 /* First: create and setup a journal_t object in memory.  We initialise
963  * very few fields yet: that has to wait until we have created the
964  * journal structures from from scratch, or loaded them from disk. */
965
966 static journal_t * journal_init_common (void)
967 {
968         journal_t *journal;
969         int err;
970
971         journal = kzalloc(sizeof(*journal), GFP_KERNEL|__GFP_NOFAIL);
972         if (!journal)
973                 goto fail;
974
975         init_waitqueue_head(&journal->j_wait_transaction_locked);
976         init_waitqueue_head(&journal->j_wait_logspace);
977         init_waitqueue_head(&journal->j_wait_done_commit);
978         init_waitqueue_head(&journal->j_wait_checkpoint);
979         init_waitqueue_head(&journal->j_wait_commit);
980         init_waitqueue_head(&journal->j_wait_updates);
981         mutex_init(&journal->j_barrier);
982         mutex_init(&journal->j_checkpoint_mutex);
983         spin_lock_init(&journal->j_revoke_lock);
984         spin_lock_init(&journal->j_list_lock);
985         spin_lock_init(&journal->j_state_lock);
986
987         journal->j_commit_interval = (HZ * JBD2_DEFAULT_MAX_COMMIT_AGE);
988         journal->j_min_batch_time = 0;
989         journal->j_max_batch_time = 15000; /* 15ms */
990
991         /* The journal is marked for error until we succeed with recovery! */
992         journal->j_flags = JBD2_ABORT;
993
994         /* Set up a default-sized revoke table for the new mount. */
995         err = jbd2_journal_init_revoke(journal, JOURNAL_REVOKE_DEFAULT_HASH);
996         if (err) {
997                 kfree(journal);
998                 goto fail;
999         }
1000
1001         journal_init_stats(journal);
1002
1003         return journal;
1004 fail:
1005         return NULL;
1006 }
1007
1008 /* jbd2_journal_init_dev and jbd2_journal_init_inode:
1009  *
1010  * Create a journal structure assigned some fixed set of disk blocks to
1011  * the journal.  We don't actually touch those disk blocks yet, but we
1012  * need to set up all of the mapping information to tell the journaling
1013  * system where the journal blocks are.
1014  *
1015  */
1016
1017 /**
1018  *  journal_t * jbd2_journal_init_dev() - creates and initialises a journal structure
1019  *  @bdev: Block device on which to create the journal
1020  *  @fs_dev: Device which hold journalled filesystem for this journal.
1021  *  @start: Block nr Start of journal.
1022  *  @len:  Length of the journal in blocks.
1023  *  @blocksize: blocksize of journalling device
1024  *
1025  *  Returns: a newly created journal_t *
1026  *
1027  *  jbd2_journal_init_dev creates a journal which maps a fixed contiguous
1028  *  range of blocks on an arbitrary block device.
1029  *
1030  */
1031 journal_t * jbd2_journal_init_dev(struct block_device *bdev,
1032                         struct block_device *fs_dev,
1033                         unsigned long long start, int len, int blocksize)
1034 {
1035         journal_t *journal = journal_init_common();
1036         struct buffer_head *bh;
1037         char *p;
1038         int n;
1039
1040         if (!journal)
1041                 return NULL;
1042
1043         /* journal descriptor can store up to n blocks -bzzz */
1044         journal->j_blocksize = blocksize;
1045         jbd2_stats_proc_init(journal);
1046         n = journal->j_blocksize / sizeof(journal_block_tag_t);
1047         journal->j_wbufsize = n;
1048         journal->j_wbuf = kmalloc(n * sizeof(struct buffer_head*), GFP_KERNEL);
1049         if (!journal->j_wbuf) {
1050                 printk(KERN_ERR "%s: Cant allocate bhs for commit thread\n",
1051                         __func__);
1052                 goto out_err;
1053         }
1054         journal->j_dev = bdev;
1055         journal->j_fs_dev = fs_dev;
1056         journal->j_blk_offset = start;
1057         journal->j_maxlen = len;
1058         bdevname(journal->j_dev, journal->j_devname);
1059         p = journal->j_devname;
1060         while ((p = strchr(p, '/')))
1061                 *p = '!';
1062
1063         bh = __getblk(journal->j_dev, start, journal->j_blocksize);
1064         if (!bh) {
1065                 printk(KERN_ERR
1066                        "%s: Cannot get buffer for journal superblock\n",
1067                        __func__);
1068                 goto out_err;
1069         }
1070         journal->j_sb_buffer = bh;
1071         journal->j_superblock = (journal_superblock_t *)bh->b_data;
1072
1073         return journal;
1074 out_err:
1075         jbd2_stats_proc_exit(journal);
1076         kfree(journal);
1077         return NULL;
1078 }
1079
1080 /**
1081  *  journal_t * jbd2_journal_init_inode () - creates a journal which maps to a inode.
1082  *  @inode: An inode to create the journal in
1083  *
1084  * jbd2_journal_init_inode creates a journal which maps an on-disk inode as
1085  * the journal.  The inode must exist already, must support bmap() and
1086  * must have all data blocks preallocated.
1087  */
1088 journal_t * jbd2_journal_init_inode (struct inode *inode)
1089 {
1090         struct buffer_head *bh;
1091         journal_t *journal = journal_init_common();
1092         char *p;
1093         int err;
1094         int n;
1095         unsigned long long blocknr;
1096
1097         if (!journal)
1098                 return NULL;
1099
1100         journal->j_dev = journal->j_fs_dev = inode->i_sb->s_bdev;
1101         journal->j_inode = inode;
1102         bdevname(journal->j_dev, journal->j_devname);
1103         p = journal->j_devname;
1104         while ((p = strchr(p, '/')))
1105                 *p = '!';
1106         p = journal->j_devname + strlen(journal->j_devname);
1107         sprintf(p, ":%lu", journal->j_inode->i_ino);
1108         jbd_debug(1,
1109                   "journal %p: inode %s/%ld, size %Ld, bits %d, blksize %ld\n",
1110                   journal, inode->i_sb->s_id, inode->i_ino,
1111                   (long long) inode->i_size,
1112                   inode->i_sb->s_blocksize_bits, inode->i_sb->s_blocksize);
1113
1114         journal->j_maxlen = inode->i_size >> inode->i_sb->s_blocksize_bits;
1115         journal->j_blocksize = inode->i_sb->s_blocksize;
1116         jbd2_stats_proc_init(journal);
1117
1118         /* journal descriptor can store up to n blocks -bzzz */
1119         n = journal->j_blocksize / sizeof(journal_block_tag_t);
1120         journal->j_wbufsize = n;
1121         journal->j_wbuf = kmalloc(n * sizeof(struct buffer_head*), GFP_KERNEL);
1122         if (!journal->j_wbuf) {
1123                 printk(KERN_ERR "%s: Cant allocate bhs for commit thread\n",
1124                         __func__);
1125                 goto out_err;
1126         }
1127
1128         err = jbd2_journal_bmap(journal, 0, &blocknr);
1129         /* If that failed, give up */
1130         if (err) {
1131                 printk(KERN_ERR "%s: Cannnot locate journal superblock\n",
1132                        __func__);
1133                 goto out_err;
1134         }
1135
1136         bh = __getblk(journal->j_dev, blocknr, journal->j_blocksize);
1137         if (!bh) {
1138                 printk(KERN_ERR
1139                        "%s: Cannot get buffer for journal superblock\n",
1140                        __func__);
1141                 goto out_err;
1142         }
1143         journal->j_sb_buffer = bh;
1144         journal->j_superblock = (journal_superblock_t *)bh->b_data;
1145
1146         return journal;
1147 out_err:
1148         jbd2_stats_proc_exit(journal);
1149         kfree(journal);
1150         return NULL;
1151 }
1152
1153 /*
1154  * If the journal init or create aborts, we need to mark the journal
1155  * superblock as being NULL to prevent the journal destroy from writing
1156  * back a bogus superblock.
1157  */
1158 static void journal_fail_superblock (journal_t *journal)
1159 {
1160         struct buffer_head *bh = journal->j_sb_buffer;
1161         brelse(bh);
1162         journal->j_sb_buffer = NULL;
1163 }
1164
1165 /*
1166  * Given a journal_t structure, initialise the various fields for
1167  * startup of a new journaling session.  We use this both when creating
1168  * a journal, and after recovering an old journal to reset it for
1169  * subsequent use.
1170  */
1171
1172 static int journal_reset(journal_t *journal)
1173 {
1174         journal_superblock_t *sb = journal->j_superblock;
1175         unsigned long long first, last;
1176
1177         first = be32_to_cpu(sb->s_first);
1178         last = be32_to_cpu(sb->s_maxlen);
1179
1180         journal->j_first = first;
1181         journal->j_last = last;
1182
1183         journal->j_head = first;
1184         journal->j_tail = first;
1185         journal->j_free = last - first;
1186
1187         journal->j_tail_sequence = journal->j_transaction_sequence;
1188         journal->j_commit_sequence = journal->j_transaction_sequence - 1;
1189         journal->j_commit_request = journal->j_commit_sequence;
1190
1191         journal->j_max_transaction_buffers = journal->j_maxlen / 4;
1192
1193         /* Add the dynamic fields and write it to disk. */
1194         jbd2_journal_update_superblock(journal, 1);
1195         return jbd2_journal_start_thread(journal);
1196 }
1197
1198 /**
1199  * void jbd2_journal_update_superblock() - Update journal sb on disk.
1200  * @journal: The journal to update.
1201  * @wait: Set to '0' if you don't want to wait for IO completion.
1202  *
1203  * Update a journal's dynamic superblock fields and write it to disk,
1204  * optionally waiting for the IO to complete.
1205  */
1206 void jbd2_journal_update_superblock(journal_t *journal, int wait)
1207 {
1208         journal_superblock_t *sb = journal->j_superblock;
1209         struct buffer_head *bh = journal->j_sb_buffer;
1210
1211         /*
1212          * As a special case, if the on-disk copy is already marked as needing
1213          * no recovery (s_start == 0) and there are no outstanding transactions
1214          * in the filesystem, then we can safely defer the superblock update
1215          * until the next commit by setting JBD2_FLUSHED.  This avoids
1216          * attempting a write to a potential-readonly device.
1217          */
1218         if (sb->s_start == 0 && journal->j_tail_sequence ==
1219                                 journal->j_transaction_sequence) {
1220                 jbd_debug(1,"JBD: Skipping superblock update on recovered sb "
1221                         "(start %ld, seq %d, errno %d)\n",
1222                         journal->j_tail, journal->j_tail_sequence,
1223                         journal->j_errno);
1224                 goto out;
1225         }
1226
1227         if (buffer_write_io_error(bh)) {
1228                 /*
1229                  * Oh, dear.  A previous attempt to write the journal
1230                  * superblock failed.  This could happen because the
1231                  * USB device was yanked out.  Or it could happen to
1232                  * be a transient write error and maybe the block will
1233                  * be remapped.  Nothing we can do but to retry the
1234                  * write and hope for the best.
1235                  */
1236                 printk(KERN_ERR "JBD2: previous I/O error detected "
1237                        "for journal superblock update for %s.\n",
1238                        journal->j_devname);
1239                 clear_buffer_write_io_error(bh);
1240                 set_buffer_uptodate(bh);
1241         }
1242
1243         spin_lock(&journal->j_state_lock);
1244         jbd_debug(1,"JBD: updating superblock (start %ld, seq %d, errno %d)\n",
1245                   journal->j_tail, journal->j_tail_sequence, journal->j_errno);
1246
1247         sb->s_sequence = cpu_to_be32(journal->j_tail_sequence);
1248         sb->s_start    = cpu_to_be32(journal->j_tail);
1249         sb->s_errno    = cpu_to_be32(journal->j_errno);
1250         spin_unlock(&journal->j_state_lock);
1251
1252         BUFFER_TRACE(bh, "marking dirty");
1253         mark_buffer_dirty(bh);
1254         if (wait) {
1255                 sync_dirty_buffer(bh);
1256                 if (buffer_write_io_error(bh)) {
1257                         printk(KERN_ERR "JBD2: I/O error detected "
1258                                "when updating journal superblock for %s.\n",
1259                                journal->j_devname);
1260                         clear_buffer_write_io_error(bh);
1261                         set_buffer_uptodate(bh);
1262                 }
1263         } else
1264                 ll_rw_block(SWRITE, 1, &bh);
1265
1266 out:
1267         /* If we have just flushed the log (by marking s_start==0), then
1268          * any future commit will have to be careful to update the
1269          * superblock again to re-record the true start of the log. */
1270
1271         spin_lock(&journal->j_state_lock);
1272         if (sb->s_start)
1273                 journal->j_flags &= ~JBD2_FLUSHED;
1274         else
1275                 journal->j_flags |= JBD2_FLUSHED;
1276         spin_unlock(&journal->j_state_lock);
1277 }
1278
1279 /*
1280  * Read the superblock for a given journal, performing initial
1281  * validation of the format.
1282  */
1283
1284 static int journal_get_superblock(journal_t *journal)
1285 {
1286         struct buffer_head *bh;
1287         journal_superblock_t *sb;
1288         int err = -EIO;
1289
1290         bh = journal->j_sb_buffer;
1291
1292         J_ASSERT(bh != NULL);
1293         if (!buffer_uptodate(bh)) {
1294                 ll_rw_block(READ, 1, &bh);
1295                 wait_on_buffer(bh);
1296                 if (!buffer_uptodate(bh)) {
1297                         printk (KERN_ERR
1298                                 "JBD: IO error reading journal superblock\n");
1299                         goto out;
1300                 }
1301         }
1302
1303         sb = journal->j_superblock;
1304
1305         err = -EINVAL;
1306
1307         if (sb->s_header.h_magic != cpu_to_be32(JBD2_MAGIC_NUMBER) ||
1308             sb->s_blocksize != cpu_to_be32(journal->j_blocksize)) {
1309                 printk(KERN_WARNING "JBD: no valid journal superblock found\n");
1310                 goto out;
1311         }
1312
1313         switch(be32_to_cpu(sb->s_header.h_blocktype)) {
1314         case JBD2_SUPERBLOCK_V1:
1315                 journal->j_format_version = 1;
1316                 break;
1317         case JBD2_SUPERBLOCK_V2:
1318                 journal->j_format_version = 2;
1319                 break;
1320         default:
1321                 printk(KERN_WARNING "JBD: unrecognised superblock format ID\n");
1322                 goto out;
1323         }
1324
1325         if (be32_to_cpu(sb->s_maxlen) < journal->j_maxlen)
1326                 journal->j_maxlen = be32_to_cpu(sb->s_maxlen);
1327         else if (be32_to_cpu(sb->s_maxlen) > journal->j_maxlen) {
1328                 printk (KERN_WARNING "JBD: journal file too short\n");
1329                 goto out;
1330         }
1331
1332         return 0;
1333
1334 out:
1335         journal_fail_superblock(journal);
1336         return err;
1337 }
1338
1339 /*
1340  * Load the on-disk journal superblock and read the key fields into the
1341  * journal_t.
1342  */
1343
1344 static int load_superblock(journal_t *journal)
1345 {
1346         int err;
1347         journal_superblock_t *sb;
1348
1349         err = journal_get_superblock(journal);
1350         if (err)
1351                 return err;
1352
1353         sb = journal->j_superblock;
1354
1355         journal->j_tail_sequence = be32_to_cpu(sb->s_sequence);
1356         journal->j_tail = be32_to_cpu(sb->s_start);
1357         journal->j_first = be32_to_cpu(sb->s_first);
1358         journal->j_last = be32_to_cpu(sb->s_maxlen);
1359         journal->j_errno = be32_to_cpu(sb->s_errno);
1360
1361         return 0;
1362 }
1363
1364
1365 /**
1366  * int jbd2_journal_load() - Read journal from disk.
1367  * @journal: Journal to act on.
1368  *
1369  * Given a journal_t structure which tells us which disk blocks contain
1370  * a journal, read the journal from disk to initialise the in-memory
1371  * structures.
1372  */
1373 int jbd2_journal_load(journal_t *journal)
1374 {
1375         int err;
1376         journal_superblock_t *sb;
1377
1378         err = load_superblock(journal);
1379         if (err)
1380                 return err;
1381
1382         sb = journal->j_superblock;
1383         /* If this is a V2 superblock, then we have to check the
1384          * features flags on it. */
1385
1386         if (journal->j_format_version >= 2) {
1387                 if ((sb->s_feature_ro_compat &
1388                      ~cpu_to_be32(JBD2_KNOWN_ROCOMPAT_FEATURES)) ||
1389                     (sb->s_feature_incompat &
1390                      ~cpu_to_be32(JBD2_KNOWN_INCOMPAT_FEATURES))) {
1391                         printk (KERN_WARNING
1392                                 "JBD: Unrecognised features on journal\n");
1393                         return -EINVAL;
1394                 }
1395         }
1396
1397         /* Let the recovery code check whether it needs to recover any
1398          * data from the journal. */
1399         if (jbd2_journal_recover(journal))
1400                 goto recovery_error;
1401
1402         /* OK, we've finished with the dynamic journal bits:
1403          * reinitialise the dynamic contents of the superblock in memory
1404          * and reset them on disk. */
1405         if (journal_reset(journal))
1406                 goto recovery_error;
1407
1408         journal->j_flags &= ~JBD2_ABORT;
1409         journal->j_flags |= JBD2_LOADED;
1410         return 0;
1411
1412 recovery_error:
1413         printk (KERN_WARNING "JBD: recovery failed\n");
1414         return -EIO;
1415 }
1416
1417 /**
1418  * void jbd2_journal_destroy() - Release a journal_t structure.
1419  * @journal: Journal to act on.
1420  *
1421  * Release a journal_t structure once it is no longer in use by the
1422  * journaled object.
1423  * Return <0 if we couldn't clean up the journal.
1424  */
1425 int jbd2_journal_destroy(journal_t *journal)
1426 {
1427         int err = 0;
1428
1429         /* Wait for the commit thread to wake up and die. */
1430         journal_kill_thread(journal);
1431
1432         /* Force a final log commit */
1433         if (journal->j_running_transaction)
1434                 jbd2_journal_commit_transaction(journal);
1435
1436         /* Force any old transactions to disk */
1437
1438         /* Totally anal locking here... */
1439         spin_lock(&journal->j_list_lock);
1440         while (journal->j_checkpoint_transactions != NULL) {
1441                 spin_unlock(&journal->j_list_lock);
1442                 mutex_lock(&journal->j_checkpoint_mutex);
1443                 jbd2_log_do_checkpoint(journal);
1444                 mutex_unlock(&journal->j_checkpoint_mutex);
1445                 spin_lock(&journal->j_list_lock);
1446         }
1447
1448         J_ASSERT(journal->j_running_transaction == NULL);
1449         J_ASSERT(journal->j_committing_transaction == NULL);
1450         J_ASSERT(journal->j_checkpoint_transactions == NULL);
1451         spin_unlock(&journal->j_list_lock);
1452
1453         if (journal->j_sb_buffer) {
1454                 if (!is_journal_aborted(journal)) {
1455                         /* We can now mark the journal as empty. */
1456                         journal->j_tail = 0;
1457                         journal->j_tail_sequence =
1458                                 ++journal->j_transaction_sequence;
1459                         jbd2_journal_update_superblock(journal, 1);
1460                 } else {
1461                         err = -EIO;
1462                 }
1463                 brelse(journal->j_sb_buffer);
1464         }
1465
1466         if (journal->j_proc_entry)
1467                 jbd2_stats_proc_exit(journal);
1468         if (journal->j_inode)
1469                 iput(journal->j_inode);
1470         if (journal->j_revoke)
1471                 jbd2_journal_destroy_revoke(journal);
1472         kfree(journal->j_wbuf);
1473         kfree(journal);
1474
1475         return err;
1476 }
1477
1478
1479 /**
1480  *int jbd2_journal_check_used_features () - Check if features specified are used.
1481  * @journal: Journal to check.
1482  * @compat: bitmask of compatible features
1483  * @ro: bitmask of features that force read-only mount
1484  * @incompat: bitmask of incompatible features
1485  *
1486  * Check whether the journal uses all of a given set of
1487  * features.  Return true (non-zero) if it does.
1488  **/
1489
1490 int jbd2_journal_check_used_features (journal_t *journal, unsigned long compat,
1491                                  unsigned long ro, unsigned long incompat)
1492 {
1493         journal_superblock_t *sb;
1494
1495         if (!compat && !ro && !incompat)
1496                 return 1;
1497         if (journal->j_format_version == 1)
1498                 return 0;
1499
1500         sb = journal->j_superblock;
1501
1502         if (((be32_to_cpu(sb->s_feature_compat) & compat) == compat) &&
1503             ((be32_to_cpu(sb->s_feature_ro_compat) & ro) == ro) &&
1504             ((be32_to_cpu(sb->s_feature_incompat) & incompat) == incompat))
1505                 return 1;
1506
1507         return 0;
1508 }
1509
1510 /**
1511  * int jbd2_journal_check_available_features() - Check feature set in journalling layer
1512  * @journal: Journal to check.
1513  * @compat: bitmask of compatible features
1514  * @ro: bitmask of features that force read-only mount
1515  * @incompat: bitmask of incompatible features
1516  *
1517  * Check whether the journaling code supports the use of
1518  * all of a given set of features on this journal.  Return true
1519  * (non-zero) if it can. */
1520
1521 int jbd2_journal_check_available_features (journal_t *journal, unsigned long compat,
1522                                       unsigned long ro, unsigned long incompat)
1523 {
1524         journal_superblock_t *sb;
1525
1526         if (!compat && !ro && !incompat)
1527                 return 1;
1528
1529         sb = journal->j_superblock;
1530
1531         /* We can support any known requested features iff the
1532          * superblock is in version 2.  Otherwise we fail to support any
1533          * extended sb features. */
1534
1535         if (journal->j_format_version != 2)
1536                 return 0;
1537
1538         if ((compat   & JBD2_KNOWN_COMPAT_FEATURES) == compat &&
1539             (ro       & JBD2_KNOWN_ROCOMPAT_FEATURES) == ro &&
1540             (incompat & JBD2_KNOWN_INCOMPAT_FEATURES) == incompat)
1541                 return 1;
1542
1543         return 0;
1544 }
1545
1546 /**
1547  * int jbd2_journal_set_features () - Mark a given journal feature in the superblock
1548  * @journal: Journal to act on.
1549  * @compat: bitmask of compatible features
1550  * @ro: bitmask of features that force read-only mount
1551  * @incompat: bitmask of incompatible features
1552  *
1553  * Mark a given journal feature as present on the
1554  * superblock.  Returns true if the requested features could be set.
1555  *
1556  */
1557
1558 int jbd2_journal_set_features (journal_t *journal, unsigned long compat,
1559                           unsigned long ro, unsigned long incompat)
1560 {
1561         journal_superblock_t *sb;
1562
1563         if (jbd2_journal_check_used_features(journal, compat, ro, incompat))
1564                 return 1;
1565
1566         if (!jbd2_journal_check_available_features(journal, compat, ro, incompat))
1567                 return 0;
1568
1569         jbd_debug(1, "Setting new features 0x%lx/0x%lx/0x%lx\n",
1570                   compat, ro, incompat);
1571
1572         sb = journal->j_superblock;
1573
1574         sb->s_feature_compat    |= cpu_to_be32(compat);
1575         sb->s_feature_ro_compat |= cpu_to_be32(ro);
1576         sb->s_feature_incompat  |= cpu_to_be32(incompat);
1577
1578         return 1;
1579 }
1580
1581 /*
1582  * jbd2_journal_clear_features () - Clear a given journal feature in the
1583  *                                  superblock
1584  * @journal: Journal to act on.
1585  * @compat: bitmask of compatible features
1586  * @ro: bitmask of features that force read-only mount
1587  * @incompat: bitmask of incompatible features
1588  *
1589  * Clear a given journal feature as present on the
1590  * superblock.
1591  */
1592 void jbd2_journal_clear_features(journal_t *journal, unsigned long compat,
1593                                 unsigned long ro, unsigned long incompat)
1594 {
1595         journal_superblock_t *sb;
1596
1597         jbd_debug(1, "Clear features 0x%lx/0x%lx/0x%lx\n",
1598                   compat, ro, incompat);
1599
1600         sb = journal->j_superblock;
1601
1602         sb->s_feature_compat    &= ~cpu_to_be32(compat);
1603         sb->s_feature_ro_compat &= ~cpu_to_be32(ro);
1604         sb->s_feature_incompat  &= ~cpu_to_be32(incompat);
1605 }
1606 EXPORT_SYMBOL(jbd2_journal_clear_features);
1607
1608 /**
1609  * int jbd2_journal_update_format () - Update on-disk journal structure.
1610  * @journal: Journal to act on.
1611  *
1612  * Given an initialised but unloaded journal struct, poke about in the
1613  * on-disk structure to update it to the most recent supported version.
1614  */
1615 int jbd2_journal_update_format (journal_t *journal)
1616 {
1617         journal_superblock_t *sb;
1618         int err;
1619
1620         err = journal_get_superblock(journal);
1621         if (err)
1622                 return err;
1623
1624         sb = journal->j_superblock;
1625
1626         switch (be32_to_cpu(sb->s_header.h_blocktype)) {
1627         case JBD2_SUPERBLOCK_V2:
1628                 return 0;
1629         case JBD2_SUPERBLOCK_V1:
1630                 return journal_convert_superblock_v1(journal, sb);
1631         default:
1632                 break;
1633         }
1634         return -EINVAL;
1635 }
1636
1637 static int journal_convert_superblock_v1(journal_t *journal,
1638                                          journal_superblock_t *sb)
1639 {
1640         int offset, blocksize;
1641         struct buffer_head *bh;
1642
1643         printk(KERN_WARNING
1644                 "JBD: Converting superblock from version 1 to 2.\n");
1645
1646         /* Pre-initialise new fields to zero */
1647         offset = ((char *) &(sb->s_feature_compat)) - ((char *) sb);
1648         blocksize = be32_to_cpu(sb->s_blocksize);
1649         memset(&sb->s_feature_compat, 0, blocksize-offset);
1650
1651         sb->s_nr_users = cpu_to_be32(1);
1652         sb->s_header.h_blocktype = cpu_to_be32(JBD2_SUPERBLOCK_V2);
1653         journal->j_format_version = 2;
1654
1655         bh = journal->j_sb_buffer;
1656         BUFFER_TRACE(bh, "marking dirty");
1657         mark_buffer_dirty(bh);
1658         sync_dirty_buffer(bh);
1659         return 0;
1660 }
1661
1662
1663 /**
1664  * int jbd2_journal_flush () - Flush journal
1665  * @journal: Journal to act on.
1666  *
1667  * Flush all data for a given journal to disk and empty the journal.
1668  * Filesystems can use this when remounting readonly to ensure that
1669  * recovery does not need to happen on remount.
1670  */
1671
1672 int jbd2_journal_flush(journal_t *journal)
1673 {
1674         int err = 0;
1675         transaction_t *transaction = NULL;
1676         unsigned long old_tail;
1677
1678         spin_lock(&journal->j_state_lock);
1679
1680         /* Force everything buffered to the log... */
1681         if (journal->j_running_transaction) {
1682                 transaction = journal->j_running_transaction;
1683                 __jbd2_log_start_commit(journal, transaction->t_tid);
1684         } else if (journal->j_committing_transaction)
1685                 transaction = journal->j_committing_transaction;
1686
1687         /* Wait for the log commit to complete... */
1688         if (transaction) {
1689                 tid_t tid = transaction->t_tid;
1690
1691                 spin_unlock(&journal->j_state_lock);
1692                 jbd2_log_wait_commit(journal, tid);
1693         } else {
1694                 spin_unlock(&journal->j_state_lock);
1695         }
1696
1697         /* ...and flush everything in the log out to disk. */
1698         spin_lock(&journal->j_list_lock);
1699         while (!err && journal->j_checkpoint_transactions != NULL) {
1700                 spin_unlock(&journal->j_list_lock);
1701                 mutex_lock(&journal->j_checkpoint_mutex);
1702                 err = jbd2_log_do_checkpoint(journal);
1703                 mutex_unlock(&journal->j_checkpoint_mutex);
1704                 spin_lock(&journal->j_list_lock);
1705         }
1706         spin_unlock(&journal->j_list_lock);
1707
1708         if (is_journal_aborted(journal))
1709                 return -EIO;
1710
1711         jbd2_cleanup_journal_tail(journal);
1712
1713         /* Finally, mark the journal as really needing no recovery.
1714          * This sets s_start==0 in the underlying superblock, which is
1715          * the magic code for a fully-recovered superblock.  Any future
1716          * commits of data to the journal will restore the current
1717          * s_start value. */
1718         spin_lock(&journal->j_state_lock);
1719         old_tail = journal->j_tail;
1720         journal->j_tail = 0;
1721         spin_unlock(&journal->j_state_lock);
1722         jbd2_journal_update_superblock(journal, 1);
1723         spin_lock(&journal->j_state_lock);
1724         journal->j_tail = old_tail;
1725
1726         J_ASSERT(!journal->j_running_transaction);
1727         J_ASSERT(!journal->j_committing_transaction);
1728         J_ASSERT(!journal->j_checkpoint_transactions);
1729         J_ASSERT(journal->j_head == journal->j_tail);
1730         J_ASSERT(journal->j_tail_sequence == journal->j_transaction_sequence);
1731         spin_unlock(&journal->j_state_lock);
1732         return 0;
1733 }
1734
1735 /**
1736  * int jbd2_journal_wipe() - Wipe journal contents
1737  * @journal: Journal to act on.
1738  * @write: flag (see below)
1739  *
1740  * Wipe out all of the contents of a journal, safely.  This will produce
1741  * a warning if the journal contains any valid recovery information.
1742  * Must be called between journal_init_*() and jbd2_journal_load().
1743  *
1744  * If 'write' is non-zero, then we wipe out the journal on disk; otherwise
1745  * we merely suppress recovery.
1746  */
1747
1748 int jbd2_journal_wipe(journal_t *journal, int write)
1749 {
1750         journal_superblock_t *sb;
1751         int err = 0;
1752
1753         J_ASSERT (!(journal->j_flags & JBD2_LOADED));
1754
1755         err = load_superblock(journal);
1756         if (err)
1757                 return err;
1758
1759         sb = journal->j_superblock;
1760
1761         if (!journal->j_tail)
1762                 goto no_recovery;
1763
1764         printk (KERN_WARNING "JBD: %s recovery information on journal\n",
1765                 write ? "Clearing" : "Ignoring");
1766
1767         err = jbd2_journal_skip_recovery(journal);
1768         if (write)
1769                 jbd2_journal_update_superblock(journal, 1);
1770
1771  no_recovery:
1772         return err;
1773 }
1774
1775 /*
1776  * Journal abort has very specific semantics, which we describe
1777  * for journal abort.
1778  *
1779  * Two internal function, which provide abort to te jbd layer
1780  * itself are here.
1781  */
1782
1783 /*
1784  * Quick version for internal journal use (doesn't lock the journal).
1785  * Aborts hard --- we mark the abort as occurred, but do _nothing_ else,
1786  * and don't attempt to make any other journal updates.
1787  */
1788 void __jbd2_journal_abort_hard(journal_t *journal)
1789 {
1790         transaction_t *transaction;
1791
1792         if (journal->j_flags & JBD2_ABORT)
1793                 return;
1794
1795         printk(KERN_ERR "Aborting journal on device %s.\n",
1796                journal->j_devname);
1797
1798         spin_lock(&journal->j_state_lock);
1799         journal->j_flags |= JBD2_ABORT;
1800         transaction = journal->j_running_transaction;
1801         if (transaction)
1802                 __jbd2_log_start_commit(journal, transaction->t_tid);
1803         spin_unlock(&journal->j_state_lock);
1804 }
1805
1806 /* Soft abort: record the abort error status in the journal superblock,
1807  * but don't do any other IO. */
1808 static void __journal_abort_soft (journal_t *journal, int errno)
1809 {
1810         if (journal->j_flags & JBD2_ABORT)
1811                 return;
1812
1813         if (!journal->j_errno)
1814                 journal->j_errno = errno;
1815
1816         __jbd2_journal_abort_hard(journal);
1817
1818         if (errno)
1819                 jbd2_journal_update_superblock(journal, 1);
1820 }
1821
1822 /**
1823  * void jbd2_journal_abort () - Shutdown the journal immediately.
1824  * @journal: the journal to shutdown.
1825  * @errno:   an error number to record in the journal indicating
1826  *           the reason for the shutdown.
1827  *
1828  * Perform a complete, immediate shutdown of the ENTIRE
1829  * journal (not of a single transaction).  This operation cannot be
1830  * undone without closing and reopening the journal.
1831  *
1832  * The jbd2_journal_abort function is intended to support higher level error
1833  * recovery mechanisms such as the ext2/ext3 remount-readonly error
1834  * mode.
1835  *
1836  * Journal abort has very specific semantics.  Any existing dirty,
1837  * unjournaled buffers in the main filesystem will still be written to
1838  * disk by bdflush, but the journaling mechanism will be suspended
1839  * immediately and no further transaction commits will be honoured.
1840  *
1841  * Any dirty, journaled buffers will be written back to disk without
1842  * hitting the journal.  Atomicity cannot be guaranteed on an aborted
1843  * filesystem, but we _do_ attempt to leave as much data as possible
1844  * behind for fsck to use for cleanup.
1845  *
1846  * Any attempt to get a new transaction handle on a journal which is in
1847  * ABORT state will just result in an -EROFS error return.  A
1848  * jbd2_journal_stop on an existing handle will return -EIO if we have
1849  * entered abort state during the update.
1850  *
1851  * Recursive transactions are not disturbed by journal abort until the
1852  * final jbd2_journal_stop, which will receive the -EIO error.
1853  *
1854  * Finally, the jbd2_journal_abort call allows the caller to supply an errno
1855  * which will be recorded (if possible) in the journal superblock.  This
1856  * allows a client to record failure conditions in the middle of a
1857  * transaction without having to complete the transaction to record the
1858  * failure to disk.  ext3_error, for example, now uses this
1859  * functionality.
1860  *
1861  * Errors which originate from within the journaling layer will NOT
1862  * supply an errno; a null errno implies that absolutely no further
1863  * writes are done to the journal (unless there are any already in
1864  * progress).
1865  *
1866  */
1867
1868 void jbd2_journal_abort(journal_t *journal, int errno)
1869 {
1870         __journal_abort_soft(journal, errno);
1871 }
1872
1873 /**
1874  * int jbd2_journal_errno () - returns the journal's error state.
1875  * @journal: journal to examine.
1876  *
1877  * This is the errno numbet set with jbd2_journal_abort(), the last
1878  * time the journal was mounted - if the journal was stopped
1879  * without calling abort this will be 0.
1880  *
1881  * If the journal has been aborted on this mount time -EROFS will
1882  * be returned.
1883  */
1884 int jbd2_journal_errno(journal_t *journal)
1885 {
1886         int err;
1887
1888         spin_lock(&journal->j_state_lock);
1889         if (journal->j_flags & JBD2_ABORT)
1890                 err = -EROFS;
1891         else
1892                 err = journal->j_errno;
1893         spin_unlock(&journal->j_state_lock);
1894         return err;
1895 }
1896
1897 /**
1898  * int jbd2_journal_clear_err () - clears the journal's error state
1899  * @journal: journal to act on.
1900  *
1901  * An error must be cleared or Acked to take a FS out of readonly
1902  * mode.
1903  */
1904 int jbd2_journal_clear_err(journal_t *journal)
1905 {
1906         int err = 0;
1907
1908         spin_lock(&journal->j_state_lock);
1909         if (journal->j_flags & JBD2_ABORT)
1910                 err = -EROFS;
1911         else
1912                 journal->j_errno = 0;
1913         spin_unlock(&journal->j_state_lock);
1914         return err;
1915 }
1916
1917 /**
1918  * void jbd2_journal_ack_err() - Ack journal err.
1919  * @journal: journal to act on.
1920  *
1921  * An error must be cleared or Acked to take a FS out of readonly
1922  * mode.
1923  */
1924 void jbd2_journal_ack_err(journal_t *journal)
1925 {
1926         spin_lock(&journal->j_state_lock);
1927         if (journal->j_errno)
1928                 journal->j_flags |= JBD2_ACK_ERR;
1929         spin_unlock(&journal->j_state_lock);
1930 }
1931
1932 int jbd2_journal_blocks_per_page(struct inode *inode)
1933 {
1934         return 1 << (PAGE_CACHE_SHIFT - inode->i_sb->s_blocksize_bits);
1935 }
1936
1937 /*
1938  * helper functions to deal with 32 or 64bit block numbers.
1939  */
1940 size_t journal_tag_bytes(journal_t *journal)
1941 {
1942         if (JBD2_HAS_INCOMPAT_FEATURE(journal, JBD2_FEATURE_INCOMPAT_64BIT))
1943                 return JBD2_TAG_SIZE64;
1944         else
1945                 return JBD2_TAG_SIZE32;
1946 }
1947
1948 /*
1949  * Journal_head storage management
1950  */
1951 static struct kmem_cache *jbd2_journal_head_cache;
1952 #ifdef CONFIG_JBD2_DEBUG
1953 static atomic_t nr_journal_heads = ATOMIC_INIT(0);
1954 #endif
1955
1956 static int journal_init_jbd2_journal_head_cache(void)
1957 {
1958         int retval;
1959
1960         J_ASSERT(jbd2_journal_head_cache == NULL);
1961         jbd2_journal_head_cache = kmem_cache_create("jbd2_journal_head",
1962                                 sizeof(struct journal_head),
1963                                 0,              /* offset */
1964                                 SLAB_TEMPORARY, /* flags */
1965                                 NULL);          /* ctor */
1966         retval = 0;
1967         if (!jbd2_journal_head_cache) {
1968                 retval = -ENOMEM;
1969                 printk(KERN_EMERG "JBD: no memory for journal_head cache\n");
1970         }
1971         return retval;
1972 }
1973
1974 static void jbd2_journal_destroy_jbd2_journal_head_cache(void)
1975 {
1976         if (jbd2_journal_head_cache) {
1977                 kmem_cache_destroy(jbd2_journal_head_cache);
1978                 jbd2_journal_head_cache = NULL;
1979         }
1980 }
1981
1982 /*
1983  * journal_head splicing and dicing
1984  */
1985 static struct journal_head *journal_alloc_journal_head(void)
1986 {
1987         struct journal_head *ret;
1988         static unsigned long last_warning;
1989
1990 #ifdef CONFIG_JBD2_DEBUG
1991         atomic_inc(&nr_journal_heads);
1992 #endif
1993         ret = kmem_cache_alloc(jbd2_journal_head_cache, GFP_NOFS);
1994         if (!ret) {
1995                 jbd_debug(1, "out of memory for journal_head\n");
1996                 if (time_after(jiffies, last_warning + 5*HZ)) {
1997                         printk(KERN_NOTICE "ENOMEM in %s, retrying.\n",
1998                                __func__);
1999                         last_warning = jiffies;
2000                 }
2001                 while (!ret) {
2002                         yield();
2003                         ret = kmem_cache_alloc(jbd2_journal_head_cache, GFP_NOFS);
2004                 }
2005         }
2006         return ret;
2007 }
2008
2009 static void journal_free_journal_head(struct journal_head *jh)
2010 {
2011 #ifdef CONFIG_JBD2_DEBUG
2012         atomic_dec(&nr_journal_heads);
2013         memset(jh, JBD2_POISON_FREE, sizeof(*jh));
2014 #endif
2015         kmem_cache_free(jbd2_journal_head_cache, jh);
2016 }
2017
2018 /*
2019  * A journal_head is attached to a buffer_head whenever JBD has an
2020  * interest in the buffer.
2021  *
2022  * Whenever a buffer has an attached journal_head, its ->b_state:BH_JBD bit
2023  * is set.  This bit is tested in core kernel code where we need to take
2024  * JBD-specific actions.  Testing the zeroness of ->b_private is not reliable
2025  * there.
2026  *
2027  * When a buffer has its BH_JBD bit set, its ->b_count is elevated by one.
2028  *
2029  * When a buffer has its BH_JBD bit set it is immune from being released by
2030  * core kernel code, mainly via ->b_count.
2031  *
2032  * A journal_head may be detached from its buffer_head when the journal_head's
2033  * b_transaction, b_cp_transaction and b_next_transaction pointers are NULL.
2034  * Various places in JBD call jbd2_journal_remove_journal_head() to indicate that the
2035  * journal_head can be dropped if needed.
2036  *
2037  * Various places in the kernel want to attach a journal_head to a buffer_head
2038  * _before_ attaching the journal_head to a transaction.  To protect the
2039  * journal_head in this situation, jbd2_journal_add_journal_head elevates the
2040  * journal_head's b_jcount refcount by one.  The caller must call
2041  * jbd2_journal_put_journal_head() to undo this.
2042  *
2043  * So the typical usage would be:
2044  *
2045  *      (Attach a journal_head if needed.  Increments b_jcount)
2046  *      struct journal_head *jh = jbd2_journal_add_journal_head(bh);
2047  *      ...
2048  *      jh->b_transaction = xxx;
2049  *      jbd2_journal_put_journal_head(jh);
2050  *
2051  * Now, the journal_head's b_jcount is zero, but it is safe from being released
2052  * because it has a non-zero b_transaction.
2053  */
2054
2055 /*
2056  * Give a buffer_head a journal_head.
2057  *
2058  * Doesn't need the journal lock.
2059  * May sleep.
2060  */
2061 struct journal_head *jbd2_journal_add_journal_head(struct buffer_head *bh)
2062 {
2063         struct journal_head *jh;
2064         struct journal_head *new_jh = NULL;
2065
2066 repeat:
2067         if (!buffer_jbd(bh)) {
2068                 new_jh = journal_alloc_journal_head();
2069                 memset(new_jh, 0, sizeof(*new_jh));
2070         }
2071
2072         jbd_lock_bh_journal_head(bh);
2073         if (buffer_jbd(bh)) {
2074                 jh = bh2jh(bh);
2075         } else {
2076                 J_ASSERT_BH(bh,
2077                         (atomic_read(&bh->b_count) > 0) ||
2078                         (bh->b_page && bh->b_page->mapping));
2079
2080                 if (!new_jh) {
2081                         jbd_unlock_bh_journal_head(bh);
2082                         goto repeat;
2083                 }
2084
2085                 jh = new_jh;
2086                 new_jh = NULL;          /* We consumed it */
2087                 set_buffer_jbd(bh);
2088                 bh->b_private = jh;
2089                 jh->b_bh = bh;
2090                 get_bh(bh);
2091                 BUFFER_TRACE(bh, "added journal_head");
2092         }
2093         jh->b_jcount++;
2094         jbd_unlock_bh_journal_head(bh);
2095         if (new_jh)
2096                 journal_free_journal_head(new_jh);
2097         return bh->b_private;
2098 }
2099
2100 /*
2101  * Grab a ref against this buffer_head's journal_head.  If it ended up not
2102  * having a journal_head, return NULL
2103  */
2104 struct journal_head *jbd2_journal_grab_journal_head(struct buffer_head *bh)
2105 {
2106         struct journal_head *jh = NULL;
2107
2108         jbd_lock_bh_journal_head(bh);
2109         if (buffer_jbd(bh)) {
2110                 jh = bh2jh(bh);
2111                 jh->b_jcount++;
2112         }
2113         jbd_unlock_bh_journal_head(bh);
2114         return jh;
2115 }
2116
2117 static void __journal_remove_journal_head(struct buffer_head *bh)
2118 {
2119         struct journal_head *jh = bh2jh(bh);
2120
2121         J_ASSERT_JH(jh, jh->b_jcount >= 0);
2122
2123         get_bh(bh);
2124         if (jh->b_jcount == 0) {
2125                 if (jh->b_transaction == NULL &&
2126                                 jh->b_next_transaction == NULL &&
2127                                 jh->b_cp_transaction == NULL) {
2128                         J_ASSERT_JH(jh, jh->b_jlist == BJ_None);
2129                         J_ASSERT_BH(bh, buffer_jbd(bh));
2130                         J_ASSERT_BH(bh, jh2bh(jh) == bh);
2131                         BUFFER_TRACE(bh, "remove journal_head");
2132                         if (jh->b_frozen_data) {
2133                                 printk(KERN_WARNING "%s: freeing "
2134                                                 "b_frozen_data\n",
2135                                                 __func__);
2136                                 jbd2_free(jh->b_frozen_data, bh->b_size);
2137                         }
2138                         if (jh->b_committed_data) {
2139                                 printk(KERN_WARNING "%s: freeing "
2140                                                 "b_committed_data\n",
2141                                                 __func__);
2142                                 jbd2_free(jh->b_committed_data, bh->b_size);
2143                         }
2144                         bh->b_private = NULL;
2145                         jh->b_bh = NULL;        /* debug, really */
2146                         clear_buffer_jbd(bh);
2147                         __brelse(bh);
2148                         journal_free_journal_head(jh);
2149                 } else {
2150                         BUFFER_TRACE(bh, "journal_head was locked");
2151                 }
2152         }
2153 }
2154
2155 /*
2156  * jbd2_journal_remove_journal_head(): if the buffer isn't attached to a transaction
2157  * and has a zero b_jcount then remove and release its journal_head.   If we did
2158  * see that the buffer is not used by any transaction we also "logically"
2159  * decrement ->b_count.
2160  *
2161  * We in fact take an additional increment on ->b_count as a convenience,
2162  * because the caller usually wants to do additional things with the bh
2163  * after calling here.
2164  * The caller of jbd2_journal_remove_journal_head() *must* run __brelse(bh) at some
2165  * time.  Once the caller has run __brelse(), the buffer is eligible for
2166  * reaping by try_to_free_buffers().
2167  */
2168 void jbd2_journal_remove_journal_head(struct buffer_head *bh)
2169 {
2170         jbd_lock_bh_journal_head(bh);
2171         __journal_remove_journal_head(bh);
2172         jbd_unlock_bh_journal_head(bh);
2173 }
2174
2175 /*
2176  * Drop a reference on the passed journal_head.  If it fell to zero then try to
2177  * release the journal_head from the buffer_head.
2178  */
2179 void jbd2_journal_put_journal_head(struct journal_head *jh)
2180 {
2181         struct buffer_head *bh = jh2bh(jh);
2182
2183         jbd_lock_bh_journal_head(bh);
2184         J_ASSERT_JH(jh, jh->b_jcount > 0);
2185         --jh->b_jcount;
2186         if (!jh->b_jcount && !jh->b_transaction) {
2187                 __journal_remove_journal_head(bh);
2188                 __brelse(bh);
2189         }
2190         jbd_unlock_bh_journal_head(bh);
2191 }
2192
2193 /*
2194  * Initialize jbd inode head
2195  */
2196 void jbd2_journal_init_jbd_inode(struct jbd2_inode *jinode, struct inode *inode)
2197 {
2198         jinode->i_transaction = NULL;
2199         jinode->i_next_transaction = NULL;
2200         jinode->i_vfs_inode = inode;
2201         jinode->i_flags = 0;
2202         INIT_LIST_HEAD(&jinode->i_list);
2203 }
2204
2205 /*
2206  * Function to be called before we start removing inode from memory (i.e.,
2207  * clear_inode() is a fine place to be called from). It removes inode from
2208  * transaction's lists.
2209  */
2210 void jbd2_journal_release_jbd_inode(journal_t *journal,
2211                                     struct jbd2_inode *jinode)
2212 {
2213         int writeout = 0;
2214
2215         if (!journal)
2216                 return;
2217 restart:
2218         spin_lock(&journal->j_list_lock);
2219         /* Is commit writing out inode - we have to wait */
2220         if (jinode->i_flags & JI_COMMIT_RUNNING) {
2221                 wait_queue_head_t *wq;
2222                 DEFINE_WAIT_BIT(wait, &jinode->i_flags, __JI_COMMIT_RUNNING);
2223                 wq = bit_waitqueue(&jinode->i_flags, __JI_COMMIT_RUNNING);
2224                 prepare_to_wait(wq, &wait.wait, TASK_UNINTERRUPTIBLE);
2225                 spin_unlock(&journal->j_list_lock);
2226                 schedule();
2227                 finish_wait(wq, &wait.wait);
2228                 goto restart;
2229         }
2230
2231         /* Do we need to wait for data writeback? */
2232         if (journal->j_committing_transaction == jinode->i_transaction)
2233                 writeout = 1;
2234         if (jinode->i_transaction) {
2235                 list_del(&jinode->i_list);
2236                 jinode->i_transaction = NULL;
2237         }
2238         spin_unlock(&journal->j_list_lock);
2239 }
2240
2241 /*
2242  * debugfs tunables
2243  */
2244 #ifdef CONFIG_JBD2_DEBUG
2245 u8 jbd2_journal_enable_debug __read_mostly;
2246 EXPORT_SYMBOL(jbd2_journal_enable_debug);
2247
2248 #define JBD2_DEBUG_NAME "jbd2-debug"
2249
2250 static struct dentry *jbd2_debugfs_dir;
2251 static struct dentry *jbd2_debug;
2252
2253 static void __init jbd2_create_debugfs_entry(void)
2254 {
2255         jbd2_debugfs_dir = debugfs_create_dir("jbd2", NULL);
2256         if (jbd2_debugfs_dir)
2257                 jbd2_debug = debugfs_create_u8(JBD2_DEBUG_NAME, S_IRUGO,
2258                                                jbd2_debugfs_dir,
2259                                                &jbd2_journal_enable_debug);
2260 }
2261
2262 static void __exit jbd2_remove_debugfs_entry(void)
2263 {
2264         debugfs_remove(jbd2_debug);
2265         debugfs_remove(jbd2_debugfs_dir);
2266 }
2267
2268 #else
2269
2270 static void __init jbd2_create_debugfs_entry(void)
2271 {
2272 }
2273
2274 static void __exit jbd2_remove_debugfs_entry(void)
2275 {
2276 }
2277
2278 #endif
2279
2280 #ifdef CONFIG_PROC_FS
2281
2282 #define JBD2_STATS_PROC_NAME "fs/jbd2"
2283
2284 static void __init jbd2_create_jbd_stats_proc_entry(void)
2285 {
2286         proc_jbd2_stats = proc_mkdir(JBD2_STATS_PROC_NAME, NULL);
2287 }
2288
2289 static void __exit jbd2_remove_jbd_stats_proc_entry(void)
2290 {
2291         if (proc_jbd2_stats)
2292                 remove_proc_entry(JBD2_STATS_PROC_NAME, NULL);
2293 }
2294
2295 #else
2296
2297 #define jbd2_create_jbd_stats_proc_entry() do {} while (0)
2298 #define jbd2_remove_jbd_stats_proc_entry() do {} while (0)
2299
2300 #endif
2301
2302 struct kmem_cache *jbd2_handle_cache;
2303
2304 static int __init journal_init_handle_cache(void)
2305 {
2306         jbd2_handle_cache = kmem_cache_create("jbd2_journal_handle",
2307                                 sizeof(handle_t),
2308                                 0,              /* offset */
2309                                 SLAB_TEMPORARY, /* flags */
2310                                 NULL);          /* ctor */
2311         if (jbd2_handle_cache == NULL) {
2312                 printk(KERN_EMERG "JBD: failed to create handle cache\n");
2313                 return -ENOMEM;
2314         }
2315         return 0;
2316 }
2317
2318 static void jbd2_journal_destroy_handle_cache(void)
2319 {
2320         if (jbd2_handle_cache)
2321                 kmem_cache_destroy(jbd2_handle_cache);
2322 }
2323
2324 /*
2325  * Module startup and shutdown
2326  */
2327
2328 static int __init journal_init_caches(void)
2329 {
2330         int ret;
2331
2332         ret = jbd2_journal_init_revoke_caches();
2333         if (ret == 0)
2334                 ret = journal_init_jbd2_journal_head_cache();
2335         if (ret == 0)
2336                 ret = journal_init_handle_cache();
2337         return ret;
2338 }
2339
2340 static void jbd2_journal_destroy_caches(void)
2341 {
2342         jbd2_journal_destroy_revoke_caches();
2343         jbd2_journal_destroy_jbd2_journal_head_cache();
2344         jbd2_journal_destroy_handle_cache();
2345 }
2346
2347 static int __init journal_init(void)
2348 {
2349         int ret;
2350
2351         BUILD_BUG_ON(sizeof(struct journal_superblock_s) != 1024);
2352
2353         ret = journal_init_caches();
2354         if (ret == 0) {
2355                 jbd2_create_debugfs_entry();
2356                 jbd2_create_jbd_stats_proc_entry();
2357         } else {
2358                 jbd2_journal_destroy_caches();
2359         }
2360         return ret;
2361 }
2362
2363 static void __exit journal_exit(void)
2364 {
2365 #ifdef CONFIG_JBD2_DEBUG
2366         int n = atomic_read(&nr_journal_heads);
2367         if (n)
2368                 printk(KERN_EMERG "JBD: leaked %d journal_heads!\n", n);
2369 #endif
2370         jbd2_remove_debugfs_entry();
2371         jbd2_remove_jbd_stats_proc_entry();
2372         jbd2_journal_destroy_caches();
2373 }
2374
2375 MODULE_LICENSE("GPL");
2376 module_init(journal_init);
2377 module_exit(journal_exit);
2378