ARM64: DTS: Add rk3399-firefly uart4 device, node as /dev/ttyS1
[firefly-linux-kernel-4.4.55.git] / fs / jbd2 / journal.c
1 /*
2  * linux/fs/jbd2/journal.c
3  *
4  * Written by Stephen C. Tweedie <sct@redhat.com>, 1998
5  *
6  * Copyright 1998 Red Hat corp --- All Rights Reserved
7  *
8  * This file is part of the Linux kernel and is made available under
9  * the terms of the GNU General Public License, version 2, or at your
10  * option, any later version, incorporated herein by reference.
11  *
12  * Generic filesystem journal-writing code; part of the ext2fs
13  * journaling system.
14  *
15  * This file manages journals: areas of disk reserved for logging
16  * transactional updates.  This includes the kernel journaling thread
17  * which is responsible for scheduling updates to the log.
18  *
19  * We do not actually manage the physical storage of the journal in this
20  * file: that is left to a per-journal policy function, which allows us
21  * to store the journal within a filesystem-specified area for ext2
22  * journaling (ext2 can use a reserved inode for storing the log).
23  */
24
25 #include <linux/module.h>
26 #include <linux/time.h>
27 #include <linux/fs.h>
28 #include <linux/jbd2.h>
29 #include <linux/errno.h>
30 #include <linux/slab.h>
31 #include <linux/init.h>
32 #include <linux/mm.h>
33 #include <linux/freezer.h>
34 #include <linux/pagemap.h>
35 #include <linux/kthread.h>
36 #include <linux/poison.h>
37 #include <linux/proc_fs.h>
38 #include <linux/seq_file.h>
39 #include <linux/math64.h>
40 #include <linux/hash.h>
41 #include <linux/log2.h>
42 #include <linux/vmalloc.h>
43 #include <linux/backing-dev.h>
44 #include <linux/bitops.h>
45 #include <linux/ratelimit.h>
46
47 #define CREATE_TRACE_POINTS
48 #include <trace/events/jbd2.h>
49
50 #include <asm/uaccess.h>
51 #include <asm/page.h>
52
53 #ifdef CONFIG_JBD2_DEBUG
54 ushort jbd2_journal_enable_debug __read_mostly;
55 EXPORT_SYMBOL(jbd2_journal_enable_debug);
56
57 module_param_named(jbd2_debug, jbd2_journal_enable_debug, ushort, 0644);
58 MODULE_PARM_DESC(jbd2_debug, "Debugging level for jbd2");
59 #endif
60
61 EXPORT_SYMBOL(jbd2_journal_extend);
62 EXPORT_SYMBOL(jbd2_journal_stop);
63 EXPORT_SYMBOL(jbd2_journal_lock_updates);
64 EXPORT_SYMBOL(jbd2_journal_unlock_updates);
65 EXPORT_SYMBOL(jbd2_journal_get_write_access);
66 EXPORT_SYMBOL(jbd2_journal_get_create_access);
67 EXPORT_SYMBOL(jbd2_journal_get_undo_access);
68 EXPORT_SYMBOL(jbd2_journal_set_triggers);
69 EXPORT_SYMBOL(jbd2_journal_dirty_metadata);
70 EXPORT_SYMBOL(jbd2_journal_forget);
71 #if 0
72 EXPORT_SYMBOL(journal_sync_buffer);
73 #endif
74 EXPORT_SYMBOL(jbd2_journal_flush);
75 EXPORT_SYMBOL(jbd2_journal_revoke);
76
77 EXPORT_SYMBOL(jbd2_journal_init_dev);
78 EXPORT_SYMBOL(jbd2_journal_init_inode);
79 EXPORT_SYMBOL(jbd2_journal_check_used_features);
80 EXPORT_SYMBOL(jbd2_journal_check_available_features);
81 EXPORT_SYMBOL(jbd2_journal_set_features);
82 EXPORT_SYMBOL(jbd2_journal_load);
83 EXPORT_SYMBOL(jbd2_journal_destroy);
84 EXPORT_SYMBOL(jbd2_journal_abort);
85 EXPORT_SYMBOL(jbd2_journal_errno);
86 EXPORT_SYMBOL(jbd2_journal_ack_err);
87 EXPORT_SYMBOL(jbd2_journal_clear_err);
88 EXPORT_SYMBOL(jbd2_log_wait_commit);
89 EXPORT_SYMBOL(jbd2_log_start_commit);
90 EXPORT_SYMBOL(jbd2_journal_start_commit);
91 EXPORT_SYMBOL(jbd2_journal_force_commit_nested);
92 EXPORT_SYMBOL(jbd2_journal_wipe);
93 EXPORT_SYMBOL(jbd2_journal_blocks_per_page);
94 EXPORT_SYMBOL(jbd2_journal_invalidatepage);
95 EXPORT_SYMBOL(jbd2_journal_try_to_free_buffers);
96 EXPORT_SYMBOL(jbd2_journal_force_commit);
97 EXPORT_SYMBOL(jbd2_journal_file_inode);
98 EXPORT_SYMBOL(jbd2_journal_init_jbd_inode);
99 EXPORT_SYMBOL(jbd2_journal_release_jbd_inode);
100 EXPORT_SYMBOL(jbd2_journal_begin_ordered_truncate);
101 EXPORT_SYMBOL(jbd2_inode_cache);
102
103 static void __journal_abort_soft (journal_t *journal, int errno);
104 static int jbd2_journal_create_slab(size_t slab_size);
105
106 #ifdef CONFIG_JBD2_DEBUG
107 void __jbd2_debug(int level, const char *file, const char *func,
108                   unsigned int line, const char *fmt, ...)
109 {
110         struct va_format vaf;
111         va_list args;
112
113         if (level > jbd2_journal_enable_debug)
114                 return;
115         va_start(args, fmt);
116         vaf.fmt = fmt;
117         vaf.va = &args;
118         printk(KERN_DEBUG "%s: (%s, %u): %pV\n", file, func, line, &vaf);
119         va_end(args);
120 }
121 EXPORT_SYMBOL(__jbd2_debug);
122 #endif
123
124 /* Checksumming functions */
125 static int jbd2_verify_csum_type(journal_t *j, journal_superblock_t *sb)
126 {
127         if (!jbd2_journal_has_csum_v2or3_feature(j))
128                 return 1;
129
130         return sb->s_checksum_type == JBD2_CRC32C_CHKSUM;
131 }
132
133 static __be32 jbd2_superblock_csum(journal_t *j, journal_superblock_t *sb)
134 {
135         __u32 csum;
136         __be32 old_csum;
137
138         old_csum = sb->s_checksum;
139         sb->s_checksum = 0;
140         csum = jbd2_chksum(j, ~0, (char *)sb, sizeof(journal_superblock_t));
141         sb->s_checksum = old_csum;
142
143         return cpu_to_be32(csum);
144 }
145
146 static int jbd2_superblock_csum_verify(journal_t *j, journal_superblock_t *sb)
147 {
148         if (!jbd2_journal_has_csum_v2or3(j))
149                 return 1;
150
151         return sb->s_checksum == jbd2_superblock_csum(j, sb);
152 }
153
154 static void jbd2_superblock_csum_set(journal_t *j, journal_superblock_t *sb)
155 {
156         if (!jbd2_journal_has_csum_v2or3(j))
157                 return;
158
159         sb->s_checksum = jbd2_superblock_csum(j, sb);
160 }
161
162 /*
163  * Helper function used to manage commit timeouts
164  */
165
166 static void commit_timeout(unsigned long __data)
167 {
168         struct task_struct * p = (struct task_struct *) __data;
169
170         wake_up_process(p);
171 }
172
173 /*
174  * kjournald2: The main thread function used to manage a logging device
175  * journal.
176  *
177  * This kernel thread is responsible for two things:
178  *
179  * 1) COMMIT:  Every so often we need to commit the current state of the
180  *    filesystem to disk.  The journal thread is responsible for writing
181  *    all of the metadata buffers to disk.
182  *
183  * 2) CHECKPOINT: We cannot reuse a used section of the log file until all
184  *    of the data in that part of the log has been rewritten elsewhere on
185  *    the disk.  Flushing these old buffers to reclaim space in the log is
186  *    known as checkpointing, and this thread is responsible for that job.
187  */
188
189 static int kjournald2(void *arg)
190 {
191         journal_t *journal = arg;
192         transaction_t *transaction;
193
194         /*
195          * Set up an interval timer which can be used to trigger a commit wakeup
196          * after the commit interval expires
197          */
198         setup_timer(&journal->j_commit_timer, commit_timeout,
199                         (unsigned long)current);
200
201         set_freezable();
202
203         /* Record that the journal thread is running */
204         journal->j_task = current;
205         wake_up(&journal->j_wait_done_commit);
206
207         /*
208          * And now, wait forever for commit wakeup events.
209          */
210         write_lock(&journal->j_state_lock);
211
212 loop:
213         if (journal->j_flags & JBD2_UNMOUNT)
214                 goto end_loop;
215
216         jbd_debug(1, "commit_sequence=%d, commit_request=%d\n",
217                 journal->j_commit_sequence, journal->j_commit_request);
218
219         if (journal->j_commit_sequence != journal->j_commit_request) {
220                 jbd_debug(1, "OK, requests differ\n");
221                 write_unlock(&journal->j_state_lock);
222                 del_timer_sync(&journal->j_commit_timer);
223                 jbd2_journal_commit_transaction(journal);
224                 write_lock(&journal->j_state_lock);
225                 goto loop;
226         }
227
228         wake_up(&journal->j_wait_done_commit);
229         if (freezing(current)) {
230                 /*
231                  * The simpler the better. Flushing journal isn't a
232                  * good idea, because that depends on threads that may
233                  * be already stopped.
234                  */
235                 jbd_debug(1, "Now suspending kjournald2\n");
236                 write_unlock(&journal->j_state_lock);
237                 try_to_freeze();
238                 write_lock(&journal->j_state_lock);
239         } else {
240                 /*
241                  * We assume on resume that commits are already there,
242                  * so we don't sleep
243                  */
244                 DEFINE_WAIT(wait);
245                 int should_sleep = 1;
246
247                 prepare_to_wait(&journal->j_wait_commit, &wait,
248                                 TASK_INTERRUPTIBLE);
249                 if (journal->j_commit_sequence != journal->j_commit_request)
250                         should_sleep = 0;
251                 transaction = journal->j_running_transaction;
252                 if (transaction && time_after_eq(jiffies,
253                                                 transaction->t_expires))
254                         should_sleep = 0;
255                 if (journal->j_flags & JBD2_UNMOUNT)
256                         should_sleep = 0;
257                 if (should_sleep) {
258                         write_unlock(&journal->j_state_lock);
259                         schedule();
260                         write_lock(&journal->j_state_lock);
261                 }
262                 finish_wait(&journal->j_wait_commit, &wait);
263         }
264
265         jbd_debug(1, "kjournald2 wakes\n");
266
267         /*
268          * Were we woken up by a commit wakeup event?
269          */
270         transaction = journal->j_running_transaction;
271         if (transaction && time_after_eq(jiffies, transaction->t_expires)) {
272                 journal->j_commit_request = transaction->t_tid;
273                 jbd_debug(1, "woke because of timeout\n");
274         }
275         goto loop;
276
277 end_loop:
278         write_unlock(&journal->j_state_lock);
279         del_timer_sync(&journal->j_commit_timer);
280         journal->j_task = NULL;
281         wake_up(&journal->j_wait_done_commit);
282         jbd_debug(1, "Journal thread exiting.\n");
283         return 0;
284 }
285
286 static int jbd2_journal_start_thread(journal_t *journal)
287 {
288         struct task_struct *t;
289
290         t = kthread_run(kjournald2, journal, "jbd2/%s",
291                         journal->j_devname);
292         if (IS_ERR(t))
293                 return PTR_ERR(t);
294
295         wait_event(journal->j_wait_done_commit, journal->j_task != NULL);
296         return 0;
297 }
298
299 static void journal_kill_thread(journal_t *journal)
300 {
301         write_lock(&journal->j_state_lock);
302         journal->j_flags |= JBD2_UNMOUNT;
303
304         while (journal->j_task) {
305                 write_unlock(&journal->j_state_lock);
306                 wake_up(&journal->j_wait_commit);
307                 wait_event(journal->j_wait_done_commit, journal->j_task == NULL);
308                 write_lock(&journal->j_state_lock);
309         }
310         write_unlock(&journal->j_state_lock);
311 }
312
313 /*
314  * jbd2_journal_write_metadata_buffer: write a metadata buffer to the journal.
315  *
316  * Writes a metadata buffer to a given disk block.  The actual IO is not
317  * performed but a new buffer_head is constructed which labels the data
318  * to be written with the correct destination disk block.
319  *
320  * Any magic-number escaping which needs to be done will cause a
321  * copy-out here.  If the buffer happens to start with the
322  * JBD2_MAGIC_NUMBER, then we can't write it to the log directly: the
323  * magic number is only written to the log for descripter blocks.  In
324  * this case, we copy the data and replace the first word with 0, and we
325  * return a result code which indicates that this buffer needs to be
326  * marked as an escaped buffer in the corresponding log descriptor
327  * block.  The missing word can then be restored when the block is read
328  * during recovery.
329  *
330  * If the source buffer has already been modified by a new transaction
331  * since we took the last commit snapshot, we use the frozen copy of
332  * that data for IO. If we end up using the existing buffer_head's data
333  * for the write, then we have to make sure nobody modifies it while the
334  * IO is in progress. do_get_write_access() handles this.
335  *
336  * The function returns a pointer to the buffer_head to be used for IO.
337  * 
338  *
339  * Return value:
340  *  <0: Error
341  * >=0: Finished OK
342  *
343  * On success:
344  * Bit 0 set == escape performed on the data
345  * Bit 1 set == buffer copy-out performed (kfree the data after IO)
346  */
347
348 int jbd2_journal_write_metadata_buffer(transaction_t *transaction,
349                                   struct journal_head  *jh_in,
350                                   struct buffer_head **bh_out,
351                                   sector_t blocknr)
352 {
353         int need_copy_out = 0;
354         int done_copy_out = 0;
355         int do_escape = 0;
356         char *mapped_data;
357         struct buffer_head *new_bh;
358         struct page *new_page;
359         unsigned int new_offset;
360         struct buffer_head *bh_in = jh2bh(jh_in);
361         journal_t *journal = transaction->t_journal;
362
363         /*
364          * The buffer really shouldn't be locked: only the current committing
365          * transaction is allowed to write it, so nobody else is allowed
366          * to do any IO.
367          *
368          * akpm: except if we're journalling data, and write() output is
369          * also part of a shared mapping, and another thread has
370          * decided to launch a writepage() against this buffer.
371          */
372         J_ASSERT_BH(bh_in, buffer_jbddirty(bh_in));
373
374         new_bh = alloc_buffer_head(GFP_NOFS|__GFP_NOFAIL);
375
376         /* keep subsequent assertions sane */
377         atomic_set(&new_bh->b_count, 1);
378
379         jbd_lock_bh_state(bh_in);
380 repeat:
381         /*
382          * If a new transaction has already done a buffer copy-out, then
383          * we use that version of the data for the commit.
384          */
385         if (jh_in->b_frozen_data) {
386                 done_copy_out = 1;
387                 new_page = virt_to_page(jh_in->b_frozen_data);
388                 new_offset = offset_in_page(jh_in->b_frozen_data);
389         } else {
390                 new_page = jh2bh(jh_in)->b_page;
391                 new_offset = offset_in_page(jh2bh(jh_in)->b_data);
392         }
393
394         mapped_data = kmap_atomic(new_page);
395         /*
396          * Fire data frozen trigger if data already wasn't frozen.  Do this
397          * before checking for escaping, as the trigger may modify the magic
398          * offset.  If a copy-out happens afterwards, it will have the correct
399          * data in the buffer.
400          */
401         if (!done_copy_out)
402                 jbd2_buffer_frozen_trigger(jh_in, mapped_data + new_offset,
403                                            jh_in->b_triggers);
404
405         /*
406          * Check for escaping
407          */
408         if (*((__be32 *)(mapped_data + new_offset)) ==
409                                 cpu_to_be32(JBD2_MAGIC_NUMBER)) {
410                 need_copy_out = 1;
411                 do_escape = 1;
412         }
413         kunmap_atomic(mapped_data);
414
415         /*
416          * Do we need to do a data copy?
417          */
418         if (need_copy_out && !done_copy_out) {
419                 char *tmp;
420
421                 jbd_unlock_bh_state(bh_in);
422                 tmp = jbd2_alloc(bh_in->b_size, GFP_NOFS);
423                 if (!tmp) {
424                         brelse(new_bh);
425                         return -ENOMEM;
426                 }
427                 jbd_lock_bh_state(bh_in);
428                 if (jh_in->b_frozen_data) {
429                         jbd2_free(tmp, bh_in->b_size);
430                         goto repeat;
431                 }
432
433                 jh_in->b_frozen_data = tmp;
434                 mapped_data = kmap_atomic(new_page);
435                 memcpy(tmp, mapped_data + new_offset, bh_in->b_size);
436                 kunmap_atomic(mapped_data);
437
438                 new_page = virt_to_page(tmp);
439                 new_offset = offset_in_page(tmp);
440                 done_copy_out = 1;
441
442                 /*
443                  * This isn't strictly necessary, as we're using frozen
444                  * data for the escaping, but it keeps consistency with
445                  * b_frozen_data usage.
446                  */
447                 jh_in->b_frozen_triggers = jh_in->b_triggers;
448         }
449
450         /*
451          * Did we need to do an escaping?  Now we've done all the
452          * copying, we can finally do so.
453          */
454         if (do_escape) {
455                 mapped_data = kmap_atomic(new_page);
456                 *((unsigned int *)(mapped_data + new_offset)) = 0;
457                 kunmap_atomic(mapped_data);
458         }
459
460         set_bh_page(new_bh, new_page, new_offset);
461         new_bh->b_size = bh_in->b_size;
462         new_bh->b_bdev = journal->j_dev;
463         new_bh->b_blocknr = blocknr;
464         new_bh->b_private = bh_in;
465         set_buffer_mapped(new_bh);
466         set_buffer_dirty(new_bh);
467
468         *bh_out = new_bh;
469
470         /*
471          * The to-be-written buffer needs to get moved to the io queue,
472          * and the original buffer whose contents we are shadowing or
473          * copying is moved to the transaction's shadow queue.
474          */
475         JBUFFER_TRACE(jh_in, "file as BJ_Shadow");
476         spin_lock(&journal->j_list_lock);
477         __jbd2_journal_file_buffer(jh_in, transaction, BJ_Shadow);
478         spin_unlock(&journal->j_list_lock);
479         set_buffer_shadow(bh_in);
480         jbd_unlock_bh_state(bh_in);
481
482         return do_escape | (done_copy_out << 1);
483 }
484
485 /*
486  * Allocation code for the journal file.  Manage the space left in the
487  * journal, so that we can begin checkpointing when appropriate.
488  */
489
490 /*
491  * Called with j_state_lock locked for writing.
492  * Returns true if a transaction commit was started.
493  */
494 int __jbd2_log_start_commit(journal_t *journal, tid_t target)
495 {
496         /* Return if the txn has already requested to be committed */
497         if (journal->j_commit_request == target)
498                 return 0;
499
500         /*
501          * The only transaction we can possibly wait upon is the
502          * currently running transaction (if it exists).  Otherwise,
503          * the target tid must be an old one.
504          */
505         if (journal->j_running_transaction &&
506             journal->j_running_transaction->t_tid == target) {
507                 /*
508                  * We want a new commit: OK, mark the request and wakeup the
509                  * commit thread.  We do _not_ do the commit ourselves.
510                  */
511
512                 journal->j_commit_request = target;
513                 jbd_debug(1, "JBD2: requesting commit %d/%d\n",
514                           journal->j_commit_request,
515                           journal->j_commit_sequence);
516                 journal->j_running_transaction->t_requested = jiffies;
517                 wake_up(&journal->j_wait_commit);
518                 return 1;
519         } else if (!tid_geq(journal->j_commit_request, target))
520                 /* This should never happen, but if it does, preserve
521                    the evidence before kjournald goes into a loop and
522                    increments j_commit_sequence beyond all recognition. */
523                 WARN_ONCE(1, "JBD2: bad log_start_commit: %u %u %u %u\n",
524                           journal->j_commit_request,
525                           journal->j_commit_sequence,
526                           target, journal->j_running_transaction ? 
527                           journal->j_running_transaction->t_tid : 0);
528         return 0;
529 }
530
531 int jbd2_log_start_commit(journal_t *journal, tid_t tid)
532 {
533         int ret;
534
535         write_lock(&journal->j_state_lock);
536         ret = __jbd2_log_start_commit(journal, tid);
537         write_unlock(&journal->j_state_lock);
538         return ret;
539 }
540
541 /*
542  * Force and wait any uncommitted transactions.  We can only force the running
543  * transaction if we don't have an active handle, otherwise, we will deadlock.
544  * Returns: <0 in case of error,
545  *           0 if nothing to commit,
546  *           1 if transaction was successfully committed.
547  */
548 static int __jbd2_journal_force_commit(journal_t *journal)
549 {
550         transaction_t *transaction = NULL;
551         tid_t tid;
552         int need_to_start = 0, ret = 0;
553
554         read_lock(&journal->j_state_lock);
555         if (journal->j_running_transaction && !current->journal_info) {
556                 transaction = journal->j_running_transaction;
557                 if (!tid_geq(journal->j_commit_request, transaction->t_tid))
558                         need_to_start = 1;
559         } else if (journal->j_committing_transaction)
560                 transaction = journal->j_committing_transaction;
561
562         if (!transaction) {
563                 /* Nothing to commit */
564                 read_unlock(&journal->j_state_lock);
565                 return 0;
566         }
567         tid = transaction->t_tid;
568         read_unlock(&journal->j_state_lock);
569         if (need_to_start)
570                 jbd2_log_start_commit(journal, tid);
571         ret = jbd2_log_wait_commit(journal, tid);
572         if (!ret)
573                 ret = 1;
574
575         return ret;
576 }
577
578 /**
579  * Force and wait upon a commit if the calling process is not within
580  * transaction.  This is used for forcing out undo-protected data which contains
581  * bitmaps, when the fs is running out of space.
582  *
583  * @journal: journal to force
584  * Returns true if progress was made.
585  */
586 int jbd2_journal_force_commit_nested(journal_t *journal)
587 {
588         int ret;
589
590         ret = __jbd2_journal_force_commit(journal);
591         return ret > 0;
592 }
593
594 /**
595  * int journal_force_commit() - force any uncommitted transactions
596  * @journal: journal to force
597  *
598  * Caller want unconditional commit. We can only force the running transaction
599  * if we don't have an active handle, otherwise, we will deadlock.
600  */
601 int jbd2_journal_force_commit(journal_t *journal)
602 {
603         int ret;
604
605         J_ASSERT(!current->journal_info);
606         ret = __jbd2_journal_force_commit(journal);
607         if (ret > 0)
608                 ret = 0;
609         return ret;
610 }
611
612 /*
613  * Start a commit of the current running transaction (if any).  Returns true
614  * if a transaction is going to be committed (or is currently already
615  * committing), and fills its tid in at *ptid
616  */
617 int jbd2_journal_start_commit(journal_t *journal, tid_t *ptid)
618 {
619         int ret = 0;
620
621         write_lock(&journal->j_state_lock);
622         if (journal->j_running_transaction) {
623                 tid_t tid = journal->j_running_transaction->t_tid;
624
625                 __jbd2_log_start_commit(journal, tid);
626                 /* There's a running transaction and we've just made sure
627                  * it's commit has been scheduled. */
628                 if (ptid)
629                         *ptid = tid;
630                 ret = 1;
631         } else if (journal->j_committing_transaction) {
632                 /*
633                  * If commit has been started, then we have to wait for
634                  * completion of that transaction.
635                  */
636                 if (ptid)
637                         *ptid = journal->j_committing_transaction->t_tid;
638                 ret = 1;
639         }
640         write_unlock(&journal->j_state_lock);
641         return ret;
642 }
643
644 /*
645  * Return 1 if a given transaction has not yet sent barrier request
646  * connected with a transaction commit. If 0 is returned, transaction
647  * may or may not have sent the barrier. Used to avoid sending barrier
648  * twice in common cases.
649  */
650 int jbd2_trans_will_send_data_barrier(journal_t *journal, tid_t tid)
651 {
652         int ret = 0;
653         transaction_t *commit_trans;
654
655         if (!(journal->j_flags & JBD2_BARRIER))
656                 return 0;
657         read_lock(&journal->j_state_lock);
658         /* Transaction already committed? */
659         if (tid_geq(journal->j_commit_sequence, tid))
660                 goto out;
661         commit_trans = journal->j_committing_transaction;
662         if (!commit_trans || commit_trans->t_tid != tid) {
663                 ret = 1;
664                 goto out;
665         }
666         /*
667          * Transaction is being committed and we already proceeded to
668          * submitting a flush to fs partition?
669          */
670         if (journal->j_fs_dev != journal->j_dev) {
671                 if (!commit_trans->t_need_data_flush ||
672                     commit_trans->t_state >= T_COMMIT_DFLUSH)
673                         goto out;
674         } else {
675                 if (commit_trans->t_state >= T_COMMIT_JFLUSH)
676                         goto out;
677         }
678         ret = 1;
679 out:
680         read_unlock(&journal->j_state_lock);
681         return ret;
682 }
683 EXPORT_SYMBOL(jbd2_trans_will_send_data_barrier);
684
685 /*
686  * Wait for a specified commit to complete.
687  * The caller may not hold the journal lock.
688  */
689 int jbd2_log_wait_commit(journal_t *journal, tid_t tid)
690 {
691         int err = 0;
692
693         read_lock(&journal->j_state_lock);
694 #ifdef CONFIG_JBD2_DEBUG
695         if (!tid_geq(journal->j_commit_request, tid)) {
696                 printk(KERN_ERR
697                        "%s: error: j_commit_request=%d, tid=%d\n",
698                        __func__, journal->j_commit_request, tid);
699         }
700 #endif
701         while (tid_gt(tid, journal->j_commit_sequence)) {
702                 jbd_debug(1, "JBD2: want %d, j_commit_sequence=%d\n",
703                                   tid, journal->j_commit_sequence);
704                 read_unlock(&journal->j_state_lock);
705                 wake_up(&journal->j_wait_commit);
706                 wait_event(journal->j_wait_done_commit,
707                                 !tid_gt(tid, journal->j_commit_sequence));
708                 read_lock(&journal->j_state_lock);
709         }
710         read_unlock(&journal->j_state_lock);
711
712         if (unlikely(is_journal_aborted(journal)))
713                 err = -EIO;
714         return err;
715 }
716
717 /*
718  * When this function returns the transaction corresponding to tid
719  * will be completed.  If the transaction has currently running, start
720  * committing that transaction before waiting for it to complete.  If
721  * the transaction id is stale, it is by definition already completed,
722  * so just return SUCCESS.
723  */
724 int jbd2_complete_transaction(journal_t *journal, tid_t tid)
725 {
726         int     need_to_wait = 1;
727
728         read_lock(&journal->j_state_lock);
729         if (journal->j_running_transaction &&
730             journal->j_running_transaction->t_tid == tid) {
731                 if (journal->j_commit_request != tid) {
732                         /* transaction not yet started, so request it */
733                         read_unlock(&journal->j_state_lock);
734                         jbd2_log_start_commit(journal, tid);
735                         goto wait_commit;
736                 }
737         } else if (!(journal->j_committing_transaction &&
738                      journal->j_committing_transaction->t_tid == tid))
739                 need_to_wait = 0;
740         read_unlock(&journal->j_state_lock);
741         if (!need_to_wait)
742                 return 0;
743 wait_commit:
744         return jbd2_log_wait_commit(journal, tid);
745 }
746 EXPORT_SYMBOL(jbd2_complete_transaction);
747
748 /*
749  * Log buffer allocation routines:
750  */
751
752 int jbd2_journal_next_log_block(journal_t *journal, unsigned long long *retp)
753 {
754         unsigned long blocknr;
755
756         write_lock(&journal->j_state_lock);
757         J_ASSERT(journal->j_free > 1);
758
759         blocknr = journal->j_head;
760         journal->j_head++;
761         journal->j_free--;
762         if (journal->j_head == journal->j_last)
763                 journal->j_head = journal->j_first;
764         write_unlock(&journal->j_state_lock);
765         return jbd2_journal_bmap(journal, blocknr, retp);
766 }
767
768 /*
769  * Conversion of logical to physical block numbers for the journal
770  *
771  * On external journals the journal blocks are identity-mapped, so
772  * this is a no-op.  If needed, we can use j_blk_offset - everything is
773  * ready.
774  */
775 int jbd2_journal_bmap(journal_t *journal, unsigned long blocknr,
776                  unsigned long long *retp)
777 {
778         int err = 0;
779         unsigned long long ret;
780
781         if (journal->j_inode) {
782                 ret = bmap(journal->j_inode, blocknr);
783                 if (ret)
784                         *retp = ret;
785                 else {
786                         printk(KERN_ALERT "%s: journal block not found "
787                                         "at offset %lu on %s\n",
788                                __func__, blocknr, journal->j_devname);
789                         err = -EIO;
790                         __journal_abort_soft(journal, err);
791                 }
792         } else {
793                 *retp = blocknr; /* +journal->j_blk_offset */
794         }
795         return err;
796 }
797
798 /*
799  * We play buffer_head aliasing tricks to write data/metadata blocks to
800  * the journal without copying their contents, but for journal
801  * descriptor blocks we do need to generate bona fide buffers.
802  *
803  * After the caller of jbd2_journal_get_descriptor_buffer() has finished modifying
804  * the buffer's contents they really should run flush_dcache_page(bh->b_page).
805  * But we don't bother doing that, so there will be coherency problems with
806  * mmaps of blockdevs which hold live JBD-controlled filesystems.
807  */
808 struct buffer_head *jbd2_journal_get_descriptor_buffer(journal_t *journal)
809 {
810         struct buffer_head *bh;
811         unsigned long long blocknr;
812         int err;
813
814         err = jbd2_journal_next_log_block(journal, &blocknr);
815
816         if (err)
817                 return NULL;
818
819         bh = __getblk(journal->j_dev, blocknr, journal->j_blocksize);
820         if (!bh)
821                 return NULL;
822         lock_buffer(bh);
823         memset(bh->b_data, 0, journal->j_blocksize);
824         set_buffer_uptodate(bh);
825         unlock_buffer(bh);
826         BUFFER_TRACE(bh, "return this buffer");
827         return bh;
828 }
829
830 /*
831  * Return tid of the oldest transaction in the journal and block in the journal
832  * where the transaction starts.
833  *
834  * If the journal is now empty, return which will be the next transaction ID
835  * we will write and where will that transaction start.
836  *
837  * The return value is 0 if journal tail cannot be pushed any further, 1 if
838  * it can.
839  */
840 int jbd2_journal_get_log_tail(journal_t *journal, tid_t *tid,
841                               unsigned long *block)
842 {
843         transaction_t *transaction;
844         int ret;
845
846         read_lock(&journal->j_state_lock);
847         spin_lock(&journal->j_list_lock);
848         transaction = journal->j_checkpoint_transactions;
849         if (transaction) {
850                 *tid = transaction->t_tid;
851                 *block = transaction->t_log_start;
852         } else if ((transaction = journal->j_committing_transaction) != NULL) {
853                 *tid = transaction->t_tid;
854                 *block = transaction->t_log_start;
855         } else if ((transaction = journal->j_running_transaction) != NULL) {
856                 *tid = transaction->t_tid;
857                 *block = journal->j_head;
858         } else {
859                 *tid = journal->j_transaction_sequence;
860                 *block = journal->j_head;
861         }
862         ret = tid_gt(*tid, journal->j_tail_sequence);
863         spin_unlock(&journal->j_list_lock);
864         read_unlock(&journal->j_state_lock);
865
866         return ret;
867 }
868
869 /*
870  * Update information in journal structure and in on disk journal superblock
871  * about log tail. This function does not check whether information passed in
872  * really pushes log tail further. It's responsibility of the caller to make
873  * sure provided log tail information is valid (e.g. by holding
874  * j_checkpoint_mutex all the time between computing log tail and calling this
875  * function as is the case with jbd2_cleanup_journal_tail()).
876  *
877  * Requires j_checkpoint_mutex
878  */
879 int __jbd2_update_log_tail(journal_t *journal, tid_t tid, unsigned long block)
880 {
881         unsigned long freed;
882         int ret;
883
884         BUG_ON(!mutex_is_locked(&journal->j_checkpoint_mutex));
885
886         /*
887          * We cannot afford for write to remain in drive's caches since as
888          * soon as we update j_tail, next transaction can start reusing journal
889          * space and if we lose sb update during power failure we'd replay
890          * old transaction with possibly newly overwritten data.
891          */
892         ret = jbd2_journal_update_sb_log_tail(journal, tid, block, WRITE_FUA);
893         if (ret)
894                 goto out;
895
896         write_lock(&journal->j_state_lock);
897         freed = block - journal->j_tail;
898         if (block < journal->j_tail)
899                 freed += journal->j_last - journal->j_first;
900
901         trace_jbd2_update_log_tail(journal, tid, block, freed);
902         jbd_debug(1,
903                   "Cleaning journal tail from %d to %d (offset %lu), "
904                   "freeing %lu\n",
905                   journal->j_tail_sequence, tid, block, freed);
906
907         journal->j_free += freed;
908         journal->j_tail_sequence = tid;
909         journal->j_tail = block;
910         write_unlock(&journal->j_state_lock);
911
912 out:
913         return ret;
914 }
915
916 /*
917  * This is a variaon of __jbd2_update_log_tail which checks for validity of
918  * provided log tail and locks j_checkpoint_mutex. So it is safe against races
919  * with other threads updating log tail.
920  */
921 void jbd2_update_log_tail(journal_t *journal, tid_t tid, unsigned long block)
922 {
923         mutex_lock(&journal->j_checkpoint_mutex);
924         if (tid_gt(tid, journal->j_tail_sequence))
925                 __jbd2_update_log_tail(journal, tid, block);
926         mutex_unlock(&journal->j_checkpoint_mutex);
927 }
928
929 struct jbd2_stats_proc_session {
930         journal_t *journal;
931         struct transaction_stats_s *stats;
932         int start;
933         int max;
934 };
935
936 static void *jbd2_seq_info_start(struct seq_file *seq, loff_t *pos)
937 {
938         return *pos ? NULL : SEQ_START_TOKEN;
939 }
940
941 static void *jbd2_seq_info_next(struct seq_file *seq, void *v, loff_t *pos)
942 {
943         return NULL;
944 }
945
946 static int jbd2_seq_info_show(struct seq_file *seq, void *v)
947 {
948         struct jbd2_stats_proc_session *s = seq->private;
949
950         if (v != SEQ_START_TOKEN)
951                 return 0;
952         seq_printf(seq, "%lu transactions (%lu requested), "
953                    "each up to %u blocks\n",
954                    s->stats->ts_tid, s->stats->ts_requested,
955                    s->journal->j_max_transaction_buffers);
956         if (s->stats->ts_tid == 0)
957                 return 0;
958         seq_printf(seq, "average: \n  %ums waiting for transaction\n",
959             jiffies_to_msecs(s->stats->run.rs_wait / s->stats->ts_tid));
960         seq_printf(seq, "  %ums request delay\n",
961             (s->stats->ts_requested == 0) ? 0 :
962             jiffies_to_msecs(s->stats->run.rs_request_delay /
963                              s->stats->ts_requested));
964         seq_printf(seq, "  %ums running transaction\n",
965             jiffies_to_msecs(s->stats->run.rs_running / s->stats->ts_tid));
966         seq_printf(seq, "  %ums transaction was being locked\n",
967             jiffies_to_msecs(s->stats->run.rs_locked / s->stats->ts_tid));
968         seq_printf(seq, "  %ums flushing data (in ordered mode)\n",
969             jiffies_to_msecs(s->stats->run.rs_flushing / s->stats->ts_tid));
970         seq_printf(seq, "  %ums logging transaction\n",
971             jiffies_to_msecs(s->stats->run.rs_logging / s->stats->ts_tid));
972         seq_printf(seq, "  %lluus average transaction commit time\n",
973                    div_u64(s->journal->j_average_commit_time, 1000));
974         seq_printf(seq, "  %lu handles per transaction\n",
975             s->stats->run.rs_handle_count / s->stats->ts_tid);
976         seq_printf(seq, "  %lu blocks per transaction\n",
977             s->stats->run.rs_blocks / s->stats->ts_tid);
978         seq_printf(seq, "  %lu logged blocks per transaction\n",
979             s->stats->run.rs_blocks_logged / s->stats->ts_tid);
980         return 0;
981 }
982
983 static void jbd2_seq_info_stop(struct seq_file *seq, void *v)
984 {
985 }
986
987 static const struct seq_operations jbd2_seq_info_ops = {
988         .start  = jbd2_seq_info_start,
989         .next   = jbd2_seq_info_next,
990         .stop   = jbd2_seq_info_stop,
991         .show   = jbd2_seq_info_show,
992 };
993
994 static int jbd2_seq_info_open(struct inode *inode, struct file *file)
995 {
996         journal_t *journal = PDE_DATA(inode);
997         struct jbd2_stats_proc_session *s;
998         int rc, size;
999
1000         s = kmalloc(sizeof(*s), GFP_KERNEL);
1001         if (s == NULL)
1002                 return -ENOMEM;
1003         size = sizeof(struct transaction_stats_s);
1004         s->stats = kmalloc(size, GFP_KERNEL);
1005         if (s->stats == NULL) {
1006                 kfree(s);
1007                 return -ENOMEM;
1008         }
1009         spin_lock(&journal->j_history_lock);
1010         memcpy(s->stats, &journal->j_stats, size);
1011         s->journal = journal;
1012         spin_unlock(&journal->j_history_lock);
1013
1014         rc = seq_open(file, &jbd2_seq_info_ops);
1015         if (rc == 0) {
1016                 struct seq_file *m = file->private_data;
1017                 m->private = s;
1018         } else {
1019                 kfree(s->stats);
1020                 kfree(s);
1021         }
1022         return rc;
1023
1024 }
1025
1026 static int jbd2_seq_info_release(struct inode *inode, struct file *file)
1027 {
1028         struct seq_file *seq = file->private_data;
1029         struct jbd2_stats_proc_session *s = seq->private;
1030         kfree(s->stats);
1031         kfree(s);
1032         return seq_release(inode, file);
1033 }
1034
1035 static const struct file_operations jbd2_seq_info_fops = {
1036         .owner          = THIS_MODULE,
1037         .open           = jbd2_seq_info_open,
1038         .read           = seq_read,
1039         .llseek         = seq_lseek,
1040         .release        = jbd2_seq_info_release,
1041 };
1042
1043 static struct proc_dir_entry *proc_jbd2_stats;
1044
1045 static void jbd2_stats_proc_init(journal_t *journal)
1046 {
1047         journal->j_proc_entry = proc_mkdir(journal->j_devname, proc_jbd2_stats);
1048         if (journal->j_proc_entry) {
1049                 proc_create_data("info", S_IRUGO, journal->j_proc_entry,
1050                                  &jbd2_seq_info_fops, journal);
1051         }
1052 }
1053
1054 static void jbd2_stats_proc_exit(journal_t *journal)
1055 {
1056         remove_proc_entry("info", journal->j_proc_entry);
1057         remove_proc_entry(journal->j_devname, proc_jbd2_stats);
1058 }
1059
1060 /*
1061  * Management for journal control blocks: functions to create and
1062  * destroy journal_t structures, and to initialise and read existing
1063  * journal blocks from disk.  */
1064
1065 /* First: create and setup a journal_t object in memory.  We initialise
1066  * very few fields yet: that has to wait until we have created the
1067  * journal structures from from scratch, or loaded them from disk. */
1068
1069 static journal_t * journal_init_common (void)
1070 {
1071         journal_t *journal;
1072         int err;
1073
1074         journal = kzalloc(sizeof(*journal), GFP_KERNEL);
1075         if (!journal)
1076                 return NULL;
1077
1078         init_waitqueue_head(&journal->j_wait_transaction_locked);
1079         init_waitqueue_head(&journal->j_wait_done_commit);
1080         init_waitqueue_head(&journal->j_wait_commit);
1081         init_waitqueue_head(&journal->j_wait_updates);
1082         init_waitqueue_head(&journal->j_wait_reserved);
1083         mutex_init(&journal->j_barrier);
1084         mutex_init(&journal->j_checkpoint_mutex);
1085         spin_lock_init(&journal->j_revoke_lock);
1086         spin_lock_init(&journal->j_list_lock);
1087         rwlock_init(&journal->j_state_lock);
1088
1089         journal->j_commit_interval = (HZ * JBD2_DEFAULT_MAX_COMMIT_AGE);
1090         journal->j_min_batch_time = 0;
1091         journal->j_max_batch_time = 15000; /* 15ms */
1092         atomic_set(&journal->j_reserved_credits, 0);
1093
1094         /* The journal is marked for error until we succeed with recovery! */
1095         journal->j_flags = JBD2_ABORT;
1096
1097         /* Set up a default-sized revoke table for the new mount. */
1098         err = jbd2_journal_init_revoke(journal, JOURNAL_REVOKE_DEFAULT_HASH);
1099         if (err) {
1100                 kfree(journal);
1101                 return NULL;
1102         }
1103
1104         spin_lock_init(&journal->j_history_lock);
1105
1106         return journal;
1107 }
1108
1109 /* jbd2_journal_init_dev and jbd2_journal_init_inode:
1110  *
1111  * Create a journal structure assigned some fixed set of disk blocks to
1112  * the journal.  We don't actually touch those disk blocks yet, but we
1113  * need to set up all of the mapping information to tell the journaling
1114  * system where the journal blocks are.
1115  *
1116  */
1117
1118 /**
1119  *  journal_t * jbd2_journal_init_dev() - creates and initialises a journal structure
1120  *  @bdev: Block device on which to create the journal
1121  *  @fs_dev: Device which hold journalled filesystem for this journal.
1122  *  @start: Block nr Start of journal.
1123  *  @len:  Length of the journal in blocks.
1124  *  @blocksize: blocksize of journalling device
1125  *
1126  *  Returns: a newly created journal_t *
1127  *
1128  *  jbd2_journal_init_dev creates a journal which maps a fixed contiguous
1129  *  range of blocks on an arbitrary block device.
1130  *
1131  */
1132 journal_t * jbd2_journal_init_dev(struct block_device *bdev,
1133                         struct block_device *fs_dev,
1134                         unsigned long long start, int len, int blocksize)
1135 {
1136         journal_t *journal = journal_init_common();
1137         struct buffer_head *bh;
1138         int n;
1139
1140         if (!journal)
1141                 return NULL;
1142
1143         /* journal descriptor can store up to n blocks -bzzz */
1144         journal->j_blocksize = blocksize;
1145         journal->j_dev = bdev;
1146         journal->j_fs_dev = fs_dev;
1147         journal->j_blk_offset = start;
1148         journal->j_maxlen = len;
1149         bdevname(journal->j_dev, journal->j_devname);
1150         strreplace(journal->j_devname, '/', '!');
1151         jbd2_stats_proc_init(journal);
1152         n = journal->j_blocksize / sizeof(journal_block_tag_t);
1153         journal->j_wbufsize = n;
1154         journal->j_wbuf = kmalloc(n * sizeof(struct buffer_head*), GFP_KERNEL);
1155         if (!journal->j_wbuf) {
1156                 printk(KERN_ERR "%s: Can't allocate bhs for commit thread\n",
1157                         __func__);
1158                 goto out_err;
1159         }
1160
1161         bh = __getblk(journal->j_dev, start, journal->j_blocksize);
1162         if (!bh) {
1163                 printk(KERN_ERR
1164                        "%s: Cannot get buffer for journal superblock\n",
1165                        __func__);
1166                 goto out_err;
1167         }
1168         journal->j_sb_buffer = bh;
1169         journal->j_superblock = (journal_superblock_t *)bh->b_data;
1170
1171         return journal;
1172 out_err:
1173         kfree(journal->j_wbuf);
1174         jbd2_stats_proc_exit(journal);
1175         kfree(journal);
1176         return NULL;
1177 }
1178
1179 /**
1180  *  journal_t * jbd2_journal_init_inode () - creates a journal which maps to a inode.
1181  *  @inode: An inode to create the journal in
1182  *
1183  * jbd2_journal_init_inode creates a journal which maps an on-disk inode as
1184  * the journal.  The inode must exist already, must support bmap() and
1185  * must have all data blocks preallocated.
1186  */
1187 journal_t * jbd2_journal_init_inode (struct inode *inode)
1188 {
1189         struct buffer_head *bh;
1190         journal_t *journal = journal_init_common();
1191         char *p;
1192         int err;
1193         int n;
1194         unsigned long long blocknr;
1195
1196         if (!journal)
1197                 return NULL;
1198
1199         journal->j_dev = journal->j_fs_dev = inode->i_sb->s_bdev;
1200         journal->j_inode = inode;
1201         bdevname(journal->j_dev, journal->j_devname);
1202         p = strreplace(journal->j_devname, '/', '!');
1203         sprintf(p, "-%lu", journal->j_inode->i_ino);
1204         jbd_debug(1,
1205                   "journal %p: inode %s/%ld, size %Ld, bits %d, blksize %ld\n",
1206                   journal, inode->i_sb->s_id, inode->i_ino,
1207                   (long long) inode->i_size,
1208                   inode->i_sb->s_blocksize_bits, inode->i_sb->s_blocksize);
1209
1210         journal->j_maxlen = inode->i_size >> inode->i_sb->s_blocksize_bits;
1211         journal->j_blocksize = inode->i_sb->s_blocksize;
1212         jbd2_stats_proc_init(journal);
1213
1214         /* journal descriptor can store up to n blocks -bzzz */
1215         n = journal->j_blocksize / sizeof(journal_block_tag_t);
1216         journal->j_wbufsize = n;
1217         journal->j_wbuf = kmalloc(n * sizeof(struct buffer_head*), GFP_KERNEL);
1218         if (!journal->j_wbuf) {
1219                 printk(KERN_ERR "%s: Can't allocate bhs for commit thread\n",
1220                         __func__);
1221                 goto out_err;
1222         }
1223
1224         err = jbd2_journal_bmap(journal, 0, &blocknr);
1225         /* If that failed, give up */
1226         if (err) {
1227                 printk(KERN_ERR "%s: Cannot locate journal superblock\n",
1228                        __func__);
1229                 goto out_err;
1230         }
1231
1232         bh = getblk_unmovable(journal->j_dev, blocknr, journal->j_blocksize);
1233         if (!bh) {
1234                 printk(KERN_ERR
1235                        "%s: Cannot get buffer for journal superblock\n",
1236                        __func__);
1237                 goto out_err;
1238         }
1239         journal->j_sb_buffer = bh;
1240         journal->j_superblock = (journal_superblock_t *)bh->b_data;
1241
1242         return journal;
1243 out_err:
1244         kfree(journal->j_wbuf);
1245         jbd2_stats_proc_exit(journal);
1246         kfree(journal);
1247         return NULL;
1248 }
1249
1250 /*
1251  * If the journal init or create aborts, we need to mark the journal
1252  * superblock as being NULL to prevent the journal destroy from writing
1253  * back a bogus superblock.
1254  */
1255 static void journal_fail_superblock (journal_t *journal)
1256 {
1257         struct buffer_head *bh = journal->j_sb_buffer;
1258         brelse(bh);
1259         journal->j_sb_buffer = NULL;
1260 }
1261
1262 /*
1263  * Given a journal_t structure, initialise the various fields for
1264  * startup of a new journaling session.  We use this both when creating
1265  * a journal, and after recovering an old journal to reset it for
1266  * subsequent use.
1267  */
1268
1269 static int journal_reset(journal_t *journal)
1270 {
1271         journal_superblock_t *sb = journal->j_superblock;
1272         unsigned long long first, last;
1273
1274         first = be32_to_cpu(sb->s_first);
1275         last = be32_to_cpu(sb->s_maxlen);
1276         if (first + JBD2_MIN_JOURNAL_BLOCKS > last + 1) {
1277                 printk(KERN_ERR "JBD2: Journal too short (blocks %llu-%llu).\n",
1278                        first, last);
1279                 journal_fail_superblock(journal);
1280                 return -EINVAL;
1281         }
1282
1283         journal->j_first = first;
1284         journal->j_last = last;
1285
1286         journal->j_head = first;
1287         journal->j_tail = first;
1288         journal->j_free = last - first;
1289
1290         journal->j_tail_sequence = journal->j_transaction_sequence;
1291         journal->j_commit_sequence = journal->j_transaction_sequence - 1;
1292         journal->j_commit_request = journal->j_commit_sequence;
1293
1294         journal->j_max_transaction_buffers = journal->j_maxlen / 4;
1295
1296         /*
1297          * As a special case, if the on-disk copy is already marked as needing
1298          * no recovery (s_start == 0), then we can safely defer the superblock
1299          * update until the next commit by setting JBD2_FLUSHED.  This avoids
1300          * attempting a write to a potential-readonly device.
1301          */
1302         if (sb->s_start == 0) {
1303                 jbd_debug(1, "JBD2: Skipping superblock update on recovered sb "
1304                         "(start %ld, seq %d, errno %d)\n",
1305                         journal->j_tail, journal->j_tail_sequence,
1306                         journal->j_errno);
1307                 journal->j_flags |= JBD2_FLUSHED;
1308         } else {
1309                 /* Lock here to make assertions happy... */
1310                 mutex_lock(&journal->j_checkpoint_mutex);
1311                 /*
1312                  * Update log tail information. We use WRITE_FUA since new
1313                  * transaction will start reusing journal space and so we
1314                  * must make sure information about current log tail is on
1315                  * disk before that.
1316                  */
1317                 jbd2_journal_update_sb_log_tail(journal,
1318                                                 journal->j_tail_sequence,
1319                                                 journal->j_tail,
1320                                                 WRITE_FUA);
1321                 mutex_unlock(&journal->j_checkpoint_mutex);
1322         }
1323         return jbd2_journal_start_thread(journal);
1324 }
1325
1326 static int jbd2_write_superblock(journal_t *journal, int write_op)
1327 {
1328         struct buffer_head *bh = journal->j_sb_buffer;
1329         journal_superblock_t *sb = journal->j_superblock;
1330         int ret;
1331
1332         trace_jbd2_write_superblock(journal, write_op);
1333         if (!(journal->j_flags & JBD2_BARRIER))
1334                 write_op &= ~(REQ_FUA | REQ_FLUSH);
1335         lock_buffer(bh);
1336         if (buffer_write_io_error(bh)) {
1337                 /*
1338                  * Oh, dear.  A previous attempt to write the journal
1339                  * superblock failed.  This could happen because the
1340                  * USB device was yanked out.  Or it could happen to
1341                  * be a transient write error and maybe the block will
1342                  * be remapped.  Nothing we can do but to retry the
1343                  * write and hope for the best.
1344                  */
1345                 printk(KERN_ERR "JBD2: previous I/O error detected "
1346                        "for journal superblock update for %s.\n",
1347                        journal->j_devname);
1348                 clear_buffer_write_io_error(bh);
1349                 set_buffer_uptodate(bh);
1350         }
1351         jbd2_superblock_csum_set(journal, sb);
1352         get_bh(bh);
1353         bh->b_end_io = end_buffer_write_sync;
1354         ret = submit_bh(write_op, bh);
1355         wait_on_buffer(bh);
1356         if (buffer_write_io_error(bh)) {
1357                 clear_buffer_write_io_error(bh);
1358                 set_buffer_uptodate(bh);
1359                 ret = -EIO;
1360         }
1361         if (ret) {
1362                 printk(KERN_ERR "JBD2: Error %d detected when updating "
1363                        "journal superblock for %s.\n", ret,
1364                        journal->j_devname);
1365                 jbd2_journal_abort(journal, ret);
1366         }
1367
1368         return ret;
1369 }
1370
1371 /**
1372  * jbd2_journal_update_sb_log_tail() - Update log tail in journal sb on disk.
1373  * @journal: The journal to update.
1374  * @tail_tid: TID of the new transaction at the tail of the log
1375  * @tail_block: The first block of the transaction at the tail of the log
1376  * @write_op: With which operation should we write the journal sb
1377  *
1378  * Update a journal's superblock information about log tail and write it to
1379  * disk, waiting for the IO to complete.
1380  */
1381 int jbd2_journal_update_sb_log_tail(journal_t *journal, tid_t tail_tid,
1382                                      unsigned long tail_block, int write_op)
1383 {
1384         journal_superblock_t *sb = journal->j_superblock;
1385         int ret;
1386
1387         BUG_ON(!mutex_is_locked(&journal->j_checkpoint_mutex));
1388         jbd_debug(1, "JBD2: updating superblock (start %lu, seq %u)\n",
1389                   tail_block, tail_tid);
1390
1391         sb->s_sequence = cpu_to_be32(tail_tid);
1392         sb->s_start    = cpu_to_be32(tail_block);
1393
1394         ret = jbd2_write_superblock(journal, write_op);
1395         if (ret)
1396                 goto out;
1397
1398         /* Log is no longer empty */
1399         write_lock(&journal->j_state_lock);
1400         WARN_ON(!sb->s_sequence);
1401         journal->j_flags &= ~JBD2_FLUSHED;
1402         write_unlock(&journal->j_state_lock);
1403
1404 out:
1405         return ret;
1406 }
1407
1408 /**
1409  * jbd2_mark_journal_empty() - Mark on disk journal as empty.
1410  * @journal: The journal to update.
1411  * @write_op: With which operation should we write the journal sb
1412  *
1413  * Update a journal's dynamic superblock fields to show that journal is empty.
1414  * Write updated superblock to disk waiting for IO to complete.
1415  */
1416 static void jbd2_mark_journal_empty(journal_t *journal, int write_op)
1417 {
1418         journal_superblock_t *sb = journal->j_superblock;
1419
1420         BUG_ON(!mutex_is_locked(&journal->j_checkpoint_mutex));
1421         read_lock(&journal->j_state_lock);
1422         /* Is it already empty? */
1423         if (sb->s_start == 0) {
1424                 read_unlock(&journal->j_state_lock);
1425                 return;
1426         }
1427         jbd_debug(1, "JBD2: Marking journal as empty (seq %d)\n",
1428                   journal->j_tail_sequence);
1429
1430         sb->s_sequence = cpu_to_be32(journal->j_tail_sequence);
1431         sb->s_start    = cpu_to_be32(0);
1432         read_unlock(&journal->j_state_lock);
1433
1434         jbd2_write_superblock(journal, write_op);
1435
1436         /* Log is no longer empty */
1437         write_lock(&journal->j_state_lock);
1438         journal->j_flags |= JBD2_FLUSHED;
1439         write_unlock(&journal->j_state_lock);
1440 }
1441
1442
1443 /**
1444  * jbd2_journal_update_sb_errno() - Update error in the journal.
1445  * @journal: The journal to update.
1446  *
1447  * Update a journal's errno.  Write updated superblock to disk waiting for IO
1448  * to complete.
1449  */
1450 void jbd2_journal_update_sb_errno(journal_t *journal)
1451 {
1452         journal_superblock_t *sb = journal->j_superblock;
1453
1454         read_lock(&journal->j_state_lock);
1455         jbd_debug(1, "JBD2: updating superblock error (errno %d)\n",
1456                   journal->j_errno);
1457         sb->s_errno    = cpu_to_be32(journal->j_errno);
1458         read_unlock(&journal->j_state_lock);
1459
1460         jbd2_write_superblock(journal, WRITE_FUA);
1461 }
1462 EXPORT_SYMBOL(jbd2_journal_update_sb_errno);
1463
1464 /*
1465  * Read the superblock for a given journal, performing initial
1466  * validation of the format.
1467  */
1468 static int journal_get_superblock(journal_t *journal)
1469 {
1470         struct buffer_head *bh;
1471         journal_superblock_t *sb;
1472         int err = -EIO;
1473
1474         bh = journal->j_sb_buffer;
1475
1476         J_ASSERT(bh != NULL);
1477         if (!buffer_uptodate(bh)) {
1478                 ll_rw_block(READ, 1, &bh);
1479                 wait_on_buffer(bh);
1480                 if (!buffer_uptodate(bh)) {
1481                         printk(KERN_ERR
1482                                 "JBD2: IO error reading journal superblock\n");
1483                         goto out;
1484                 }
1485         }
1486
1487         if (buffer_verified(bh))
1488                 return 0;
1489
1490         sb = journal->j_superblock;
1491
1492         err = -EINVAL;
1493
1494         if (sb->s_header.h_magic != cpu_to_be32(JBD2_MAGIC_NUMBER) ||
1495             sb->s_blocksize != cpu_to_be32(journal->j_blocksize)) {
1496                 printk(KERN_WARNING "JBD2: no valid journal superblock found\n");
1497                 goto out;
1498         }
1499
1500         switch(be32_to_cpu(sb->s_header.h_blocktype)) {
1501         case JBD2_SUPERBLOCK_V1:
1502                 journal->j_format_version = 1;
1503                 break;
1504         case JBD2_SUPERBLOCK_V2:
1505                 journal->j_format_version = 2;
1506                 break;
1507         default:
1508                 printk(KERN_WARNING "JBD2: unrecognised superblock format ID\n");
1509                 goto out;
1510         }
1511
1512         if (be32_to_cpu(sb->s_maxlen) < journal->j_maxlen)
1513                 journal->j_maxlen = be32_to_cpu(sb->s_maxlen);
1514         else if (be32_to_cpu(sb->s_maxlen) > journal->j_maxlen) {
1515                 printk(KERN_WARNING "JBD2: journal file too short\n");
1516                 goto out;
1517         }
1518
1519         if (be32_to_cpu(sb->s_first) == 0 ||
1520             be32_to_cpu(sb->s_first) >= journal->j_maxlen) {
1521                 printk(KERN_WARNING
1522                         "JBD2: Invalid start block of journal: %u\n",
1523                         be32_to_cpu(sb->s_first));
1524                 goto out;
1525         }
1526
1527         if (jbd2_has_feature_csum2(journal) &&
1528             jbd2_has_feature_csum3(journal)) {
1529                 /* Can't have checksum v2 and v3 at the same time! */
1530                 printk(KERN_ERR "JBD2: Can't enable checksumming v2 and v3 "
1531                        "at the same time!\n");
1532                 goto out;
1533         }
1534
1535         if (jbd2_journal_has_csum_v2or3_feature(journal) &&
1536             jbd2_has_feature_checksum(journal)) {
1537                 /* Can't have checksum v1 and v2 on at the same time! */
1538                 printk(KERN_ERR "JBD2: Can't enable checksumming v1 and v2/3 "
1539                        "at the same time!\n");
1540                 goto out;
1541         }
1542
1543         if (!jbd2_verify_csum_type(journal, sb)) {
1544                 printk(KERN_ERR "JBD2: Unknown checksum type\n");
1545                 goto out;
1546         }
1547
1548         /* Load the checksum driver */
1549         if (jbd2_journal_has_csum_v2or3_feature(journal)) {
1550                 journal->j_chksum_driver = crypto_alloc_shash("crc32c", 0, 0);
1551                 if (IS_ERR(journal->j_chksum_driver)) {
1552                         printk(KERN_ERR "JBD2: Cannot load crc32c driver.\n");
1553                         err = PTR_ERR(journal->j_chksum_driver);
1554                         journal->j_chksum_driver = NULL;
1555                         goto out;
1556                 }
1557         }
1558
1559         /* Check superblock checksum */
1560         if (!jbd2_superblock_csum_verify(journal, sb)) {
1561                 printk(KERN_ERR "JBD2: journal checksum error\n");
1562                 err = -EFSBADCRC;
1563                 goto out;
1564         }
1565
1566         /* Precompute checksum seed for all metadata */
1567         if (jbd2_journal_has_csum_v2or3(journal))
1568                 journal->j_csum_seed = jbd2_chksum(journal, ~0, sb->s_uuid,
1569                                                    sizeof(sb->s_uuid));
1570
1571         set_buffer_verified(bh);
1572
1573         return 0;
1574
1575 out:
1576         journal_fail_superblock(journal);
1577         return err;
1578 }
1579
1580 /*
1581  * Load the on-disk journal superblock and read the key fields into the
1582  * journal_t.
1583  */
1584
1585 static int load_superblock(journal_t *journal)
1586 {
1587         int err;
1588         journal_superblock_t *sb;
1589
1590         err = journal_get_superblock(journal);
1591         if (err)
1592                 return err;
1593
1594         sb = journal->j_superblock;
1595
1596         journal->j_tail_sequence = be32_to_cpu(sb->s_sequence);
1597         journal->j_tail = be32_to_cpu(sb->s_start);
1598         journal->j_first = be32_to_cpu(sb->s_first);
1599         journal->j_last = be32_to_cpu(sb->s_maxlen);
1600         journal->j_errno = be32_to_cpu(sb->s_errno);
1601
1602         return 0;
1603 }
1604
1605
1606 /**
1607  * int jbd2_journal_load() - Read journal from disk.
1608  * @journal: Journal to act on.
1609  *
1610  * Given a journal_t structure which tells us which disk blocks contain
1611  * a journal, read the journal from disk to initialise the in-memory
1612  * structures.
1613  */
1614 int jbd2_journal_load(journal_t *journal)
1615 {
1616         int err;
1617         journal_superblock_t *sb;
1618
1619         err = load_superblock(journal);
1620         if (err)
1621                 return err;
1622
1623         sb = journal->j_superblock;
1624         /* If this is a V2 superblock, then we have to check the
1625          * features flags on it. */
1626
1627         if (journal->j_format_version >= 2) {
1628                 if ((sb->s_feature_ro_compat &
1629                      ~cpu_to_be32(JBD2_KNOWN_ROCOMPAT_FEATURES)) ||
1630                     (sb->s_feature_incompat &
1631                      ~cpu_to_be32(JBD2_KNOWN_INCOMPAT_FEATURES))) {
1632                         printk(KERN_WARNING
1633                                 "JBD2: Unrecognised features on journal\n");
1634                         return -EINVAL;
1635                 }
1636         }
1637
1638         /*
1639          * Create a slab for this blocksize
1640          */
1641         err = jbd2_journal_create_slab(be32_to_cpu(sb->s_blocksize));
1642         if (err)
1643                 return err;
1644
1645         /* Let the recovery code check whether it needs to recover any
1646          * data from the journal. */
1647         if (jbd2_journal_recover(journal))
1648                 goto recovery_error;
1649
1650         if (journal->j_failed_commit) {
1651                 printk(KERN_ERR "JBD2: journal transaction %u on %s "
1652                        "is corrupt.\n", journal->j_failed_commit,
1653                        journal->j_devname);
1654                 return -EFSCORRUPTED;
1655         }
1656
1657         /* OK, we've finished with the dynamic journal bits:
1658          * reinitialise the dynamic contents of the superblock in memory
1659          * and reset them on disk. */
1660         if (journal_reset(journal))
1661                 goto recovery_error;
1662
1663         journal->j_flags &= ~JBD2_ABORT;
1664         journal->j_flags |= JBD2_LOADED;
1665         return 0;
1666
1667 recovery_error:
1668         printk(KERN_WARNING "JBD2: recovery failed\n");
1669         return -EIO;
1670 }
1671
1672 /**
1673  * void jbd2_journal_destroy() - Release a journal_t structure.
1674  * @journal: Journal to act on.
1675  *
1676  * Release a journal_t structure once it is no longer in use by the
1677  * journaled object.
1678  * Return <0 if we couldn't clean up the journal.
1679  */
1680 int jbd2_journal_destroy(journal_t *journal)
1681 {
1682         int err = 0;
1683
1684         /* Wait for the commit thread to wake up and die. */
1685         journal_kill_thread(journal);
1686
1687         /* Force a final log commit */
1688         if (journal->j_running_transaction)
1689                 jbd2_journal_commit_transaction(journal);
1690
1691         /* Force any old transactions to disk */
1692
1693         /* Totally anal locking here... */
1694         spin_lock(&journal->j_list_lock);
1695         while (journal->j_checkpoint_transactions != NULL) {
1696                 spin_unlock(&journal->j_list_lock);
1697                 mutex_lock(&journal->j_checkpoint_mutex);
1698                 err = jbd2_log_do_checkpoint(journal);
1699                 mutex_unlock(&journal->j_checkpoint_mutex);
1700                 /*
1701                  * If checkpointing failed, just free the buffers to avoid
1702                  * looping forever
1703                  */
1704                 if (err) {
1705                         jbd2_journal_destroy_checkpoint(journal);
1706                         spin_lock(&journal->j_list_lock);
1707                         break;
1708                 }
1709                 spin_lock(&journal->j_list_lock);
1710         }
1711
1712         J_ASSERT(journal->j_running_transaction == NULL);
1713         J_ASSERT(journal->j_committing_transaction == NULL);
1714         J_ASSERT(journal->j_checkpoint_transactions == NULL);
1715         spin_unlock(&journal->j_list_lock);
1716
1717         if (journal->j_sb_buffer) {
1718                 if (!is_journal_aborted(journal)) {
1719                         mutex_lock(&journal->j_checkpoint_mutex);
1720
1721                         write_lock(&journal->j_state_lock);
1722                         journal->j_tail_sequence =
1723                                 ++journal->j_transaction_sequence;
1724                         write_unlock(&journal->j_state_lock);
1725
1726                         jbd2_mark_journal_empty(journal, WRITE_FLUSH_FUA);
1727                         mutex_unlock(&journal->j_checkpoint_mutex);
1728                 } else
1729                         err = -EIO;
1730                 brelse(journal->j_sb_buffer);
1731         }
1732
1733         if (journal->j_proc_entry)
1734                 jbd2_stats_proc_exit(journal);
1735         iput(journal->j_inode);
1736         if (journal->j_revoke)
1737                 jbd2_journal_destroy_revoke(journal);
1738         if (journal->j_chksum_driver)
1739                 crypto_free_shash(journal->j_chksum_driver);
1740         kfree(journal->j_wbuf);
1741         kfree(journal);
1742
1743         return err;
1744 }
1745
1746
1747 /**
1748  *int jbd2_journal_check_used_features () - Check if features specified are used.
1749  * @journal: Journal to check.
1750  * @compat: bitmask of compatible features
1751  * @ro: bitmask of features that force read-only mount
1752  * @incompat: bitmask of incompatible features
1753  *
1754  * Check whether the journal uses all of a given set of
1755  * features.  Return true (non-zero) if it does.
1756  **/
1757
1758 int jbd2_journal_check_used_features (journal_t *journal, unsigned long compat,
1759                                  unsigned long ro, unsigned long incompat)
1760 {
1761         journal_superblock_t *sb;
1762
1763         if (!compat && !ro && !incompat)
1764                 return 1;
1765         /* Load journal superblock if it is not loaded yet. */
1766         if (journal->j_format_version == 0 &&
1767             journal_get_superblock(journal) != 0)
1768                 return 0;
1769         if (journal->j_format_version == 1)
1770                 return 0;
1771
1772         sb = journal->j_superblock;
1773
1774         if (((be32_to_cpu(sb->s_feature_compat) & compat) == compat) &&
1775             ((be32_to_cpu(sb->s_feature_ro_compat) & ro) == ro) &&
1776             ((be32_to_cpu(sb->s_feature_incompat) & incompat) == incompat))
1777                 return 1;
1778
1779         return 0;
1780 }
1781
1782 /**
1783  * int jbd2_journal_check_available_features() - Check feature set in journalling layer
1784  * @journal: Journal to check.
1785  * @compat: bitmask of compatible features
1786  * @ro: bitmask of features that force read-only mount
1787  * @incompat: bitmask of incompatible features
1788  *
1789  * Check whether the journaling code supports the use of
1790  * all of a given set of features on this journal.  Return true
1791  * (non-zero) if it can. */
1792
1793 int jbd2_journal_check_available_features (journal_t *journal, unsigned long compat,
1794                                       unsigned long ro, unsigned long incompat)
1795 {
1796         if (!compat && !ro && !incompat)
1797                 return 1;
1798
1799         /* We can support any known requested features iff the
1800          * superblock is in version 2.  Otherwise we fail to support any
1801          * extended sb features. */
1802
1803         if (journal->j_format_version != 2)
1804                 return 0;
1805
1806         if ((compat   & JBD2_KNOWN_COMPAT_FEATURES) == compat &&
1807             (ro       & JBD2_KNOWN_ROCOMPAT_FEATURES) == ro &&
1808             (incompat & JBD2_KNOWN_INCOMPAT_FEATURES) == incompat)
1809                 return 1;
1810
1811         return 0;
1812 }
1813
1814 /**
1815  * int jbd2_journal_set_features () - Mark a given journal feature in the superblock
1816  * @journal: Journal to act on.
1817  * @compat: bitmask of compatible features
1818  * @ro: bitmask of features that force read-only mount
1819  * @incompat: bitmask of incompatible features
1820  *
1821  * Mark a given journal feature as present on the
1822  * superblock.  Returns true if the requested features could be set.
1823  *
1824  */
1825
1826 int jbd2_journal_set_features (journal_t *journal, unsigned long compat,
1827                           unsigned long ro, unsigned long incompat)
1828 {
1829 #define INCOMPAT_FEATURE_ON(f) \
1830                 ((incompat & (f)) && !(sb->s_feature_incompat & cpu_to_be32(f)))
1831 #define COMPAT_FEATURE_ON(f) \
1832                 ((compat & (f)) && !(sb->s_feature_compat & cpu_to_be32(f)))
1833         journal_superblock_t *sb;
1834
1835         if (jbd2_journal_check_used_features(journal, compat, ro, incompat))
1836                 return 1;
1837
1838         if (!jbd2_journal_check_available_features(journal, compat, ro, incompat))
1839                 return 0;
1840
1841         /* If enabling v2 checksums, turn on v3 instead */
1842         if (incompat & JBD2_FEATURE_INCOMPAT_CSUM_V2) {
1843                 incompat &= ~JBD2_FEATURE_INCOMPAT_CSUM_V2;
1844                 incompat |= JBD2_FEATURE_INCOMPAT_CSUM_V3;
1845         }
1846
1847         /* Asking for checksumming v3 and v1?  Only give them v3. */
1848         if (incompat & JBD2_FEATURE_INCOMPAT_CSUM_V3 &&
1849             compat & JBD2_FEATURE_COMPAT_CHECKSUM)
1850                 compat &= ~JBD2_FEATURE_COMPAT_CHECKSUM;
1851
1852         jbd_debug(1, "Setting new features 0x%lx/0x%lx/0x%lx\n",
1853                   compat, ro, incompat);
1854
1855         sb = journal->j_superblock;
1856
1857         /* If enabling v3 checksums, update superblock */
1858         if (INCOMPAT_FEATURE_ON(JBD2_FEATURE_INCOMPAT_CSUM_V3)) {
1859                 sb->s_checksum_type = JBD2_CRC32C_CHKSUM;
1860                 sb->s_feature_compat &=
1861                         ~cpu_to_be32(JBD2_FEATURE_COMPAT_CHECKSUM);
1862
1863                 /* Load the checksum driver */
1864                 if (journal->j_chksum_driver == NULL) {
1865                         journal->j_chksum_driver = crypto_alloc_shash("crc32c",
1866                                                                       0, 0);
1867                         if (IS_ERR(journal->j_chksum_driver)) {
1868                                 printk(KERN_ERR "JBD2: Cannot load crc32c "
1869                                        "driver.\n");
1870                                 journal->j_chksum_driver = NULL;
1871                                 return 0;
1872                         }
1873
1874                         /* Precompute checksum seed for all metadata */
1875                         journal->j_csum_seed = jbd2_chksum(journal, ~0,
1876                                                            sb->s_uuid,
1877                                                            sizeof(sb->s_uuid));
1878                 }
1879         }
1880
1881         /* If enabling v1 checksums, downgrade superblock */
1882         if (COMPAT_FEATURE_ON(JBD2_FEATURE_COMPAT_CHECKSUM))
1883                 sb->s_feature_incompat &=
1884                         ~cpu_to_be32(JBD2_FEATURE_INCOMPAT_CSUM_V2 |
1885                                      JBD2_FEATURE_INCOMPAT_CSUM_V3);
1886
1887         sb->s_feature_compat    |= cpu_to_be32(compat);
1888         sb->s_feature_ro_compat |= cpu_to_be32(ro);
1889         sb->s_feature_incompat  |= cpu_to_be32(incompat);
1890
1891         return 1;
1892 #undef COMPAT_FEATURE_ON
1893 #undef INCOMPAT_FEATURE_ON
1894 }
1895
1896 /*
1897  * jbd2_journal_clear_features () - Clear a given journal feature in the
1898  *                                  superblock
1899  * @journal: Journal to act on.
1900  * @compat: bitmask of compatible features
1901  * @ro: bitmask of features that force read-only mount
1902  * @incompat: bitmask of incompatible features
1903  *
1904  * Clear a given journal feature as present on the
1905  * superblock.
1906  */
1907 void jbd2_journal_clear_features(journal_t *journal, unsigned long compat,
1908                                 unsigned long ro, unsigned long incompat)
1909 {
1910         journal_superblock_t *sb;
1911
1912         jbd_debug(1, "Clear features 0x%lx/0x%lx/0x%lx\n",
1913                   compat, ro, incompat);
1914
1915         sb = journal->j_superblock;
1916
1917         sb->s_feature_compat    &= ~cpu_to_be32(compat);
1918         sb->s_feature_ro_compat &= ~cpu_to_be32(ro);
1919         sb->s_feature_incompat  &= ~cpu_to_be32(incompat);
1920 }
1921 EXPORT_SYMBOL(jbd2_journal_clear_features);
1922
1923 /**
1924  * int jbd2_journal_flush () - Flush journal
1925  * @journal: Journal to act on.
1926  *
1927  * Flush all data for a given journal to disk and empty the journal.
1928  * Filesystems can use this when remounting readonly to ensure that
1929  * recovery does not need to happen on remount.
1930  */
1931
1932 int jbd2_journal_flush(journal_t *journal)
1933 {
1934         int err = 0;
1935         transaction_t *transaction = NULL;
1936
1937         write_lock(&journal->j_state_lock);
1938
1939         /* Force everything buffered to the log... */
1940         if (journal->j_running_transaction) {
1941                 transaction = journal->j_running_transaction;
1942                 __jbd2_log_start_commit(journal, transaction->t_tid);
1943         } else if (journal->j_committing_transaction)
1944                 transaction = journal->j_committing_transaction;
1945
1946         /* Wait for the log commit to complete... */
1947         if (transaction) {
1948                 tid_t tid = transaction->t_tid;
1949
1950                 write_unlock(&journal->j_state_lock);
1951                 jbd2_log_wait_commit(journal, tid);
1952         } else {
1953                 write_unlock(&journal->j_state_lock);
1954         }
1955
1956         /* ...and flush everything in the log out to disk. */
1957         spin_lock(&journal->j_list_lock);
1958         while (!err && journal->j_checkpoint_transactions != NULL) {
1959                 spin_unlock(&journal->j_list_lock);
1960                 mutex_lock(&journal->j_checkpoint_mutex);
1961                 err = jbd2_log_do_checkpoint(journal);
1962                 mutex_unlock(&journal->j_checkpoint_mutex);
1963                 spin_lock(&journal->j_list_lock);
1964         }
1965         spin_unlock(&journal->j_list_lock);
1966
1967         if (is_journal_aborted(journal))
1968                 return -EIO;
1969
1970         mutex_lock(&journal->j_checkpoint_mutex);
1971         if (!err) {
1972                 err = jbd2_cleanup_journal_tail(journal);
1973                 if (err < 0) {
1974                         mutex_unlock(&journal->j_checkpoint_mutex);
1975                         goto out;
1976                 }
1977                 err = 0;
1978         }
1979
1980         /* Finally, mark the journal as really needing no recovery.
1981          * This sets s_start==0 in the underlying superblock, which is
1982          * the magic code for a fully-recovered superblock.  Any future
1983          * commits of data to the journal will restore the current
1984          * s_start value. */
1985         jbd2_mark_journal_empty(journal, WRITE_FUA);
1986         mutex_unlock(&journal->j_checkpoint_mutex);
1987         write_lock(&journal->j_state_lock);
1988         J_ASSERT(!journal->j_running_transaction);
1989         J_ASSERT(!journal->j_committing_transaction);
1990         J_ASSERT(!journal->j_checkpoint_transactions);
1991         J_ASSERT(journal->j_head == journal->j_tail);
1992         J_ASSERT(journal->j_tail_sequence == journal->j_transaction_sequence);
1993         write_unlock(&journal->j_state_lock);
1994 out:
1995         return err;
1996 }
1997
1998 /**
1999  * int jbd2_journal_wipe() - Wipe journal contents
2000  * @journal: Journal to act on.
2001  * @write: flag (see below)
2002  *
2003  * Wipe out all of the contents of a journal, safely.  This will produce
2004  * a warning if the journal contains any valid recovery information.
2005  * Must be called between journal_init_*() and jbd2_journal_load().
2006  *
2007  * If 'write' is non-zero, then we wipe out the journal on disk; otherwise
2008  * we merely suppress recovery.
2009  */
2010
2011 int jbd2_journal_wipe(journal_t *journal, int write)
2012 {
2013         int err = 0;
2014
2015         J_ASSERT (!(journal->j_flags & JBD2_LOADED));
2016
2017         err = load_superblock(journal);
2018         if (err)
2019                 return err;
2020
2021         if (!journal->j_tail)
2022                 goto no_recovery;
2023
2024         printk(KERN_WARNING "JBD2: %s recovery information on journal\n",
2025                 write ? "Clearing" : "Ignoring");
2026
2027         err = jbd2_journal_skip_recovery(journal);
2028         if (write) {
2029                 /* Lock to make assertions happy... */
2030                 mutex_lock(&journal->j_checkpoint_mutex);
2031                 jbd2_mark_journal_empty(journal, WRITE_FUA);
2032                 mutex_unlock(&journal->j_checkpoint_mutex);
2033         }
2034
2035  no_recovery:
2036         return err;
2037 }
2038
2039 /*
2040  * Journal abort has very specific semantics, which we describe
2041  * for journal abort.
2042  *
2043  * Two internal functions, which provide abort to the jbd layer
2044  * itself are here.
2045  */
2046
2047 /*
2048  * Quick version for internal journal use (doesn't lock the journal).
2049  * Aborts hard --- we mark the abort as occurred, but do _nothing_ else,
2050  * and don't attempt to make any other journal updates.
2051  */
2052 void __jbd2_journal_abort_hard(journal_t *journal)
2053 {
2054         transaction_t *transaction;
2055
2056         if (journal->j_flags & JBD2_ABORT)
2057                 return;
2058
2059         printk(KERN_ERR "Aborting journal on device %s.\n",
2060                journal->j_devname);
2061
2062         write_lock(&journal->j_state_lock);
2063         journal->j_flags |= JBD2_ABORT;
2064         transaction = journal->j_running_transaction;
2065         if (transaction)
2066                 __jbd2_log_start_commit(journal, transaction->t_tid);
2067         write_unlock(&journal->j_state_lock);
2068 }
2069
2070 /* Soft abort: record the abort error status in the journal superblock,
2071  * but don't do any other IO. */
2072 static void __journal_abort_soft (journal_t *journal, int errno)
2073 {
2074         if (journal->j_flags & JBD2_ABORT)
2075                 return;
2076
2077         if (!journal->j_errno)
2078                 journal->j_errno = errno;
2079
2080         __jbd2_journal_abort_hard(journal);
2081
2082         if (errno) {
2083                 jbd2_journal_update_sb_errno(journal);
2084                 write_lock(&journal->j_state_lock);
2085                 journal->j_flags |= JBD2_REC_ERR;
2086                 write_unlock(&journal->j_state_lock);
2087         }
2088 }
2089
2090 /**
2091  * void jbd2_journal_abort () - Shutdown the journal immediately.
2092  * @journal: the journal to shutdown.
2093  * @errno:   an error number to record in the journal indicating
2094  *           the reason for the shutdown.
2095  *
2096  * Perform a complete, immediate shutdown of the ENTIRE
2097  * journal (not of a single transaction).  This operation cannot be
2098  * undone without closing and reopening the journal.
2099  *
2100  * The jbd2_journal_abort function is intended to support higher level error
2101  * recovery mechanisms such as the ext2/ext3 remount-readonly error
2102  * mode.
2103  *
2104  * Journal abort has very specific semantics.  Any existing dirty,
2105  * unjournaled buffers in the main filesystem will still be written to
2106  * disk by bdflush, but the journaling mechanism will be suspended
2107  * immediately and no further transaction commits will be honoured.
2108  *
2109  * Any dirty, journaled buffers will be written back to disk without
2110  * hitting the journal.  Atomicity cannot be guaranteed on an aborted
2111  * filesystem, but we _do_ attempt to leave as much data as possible
2112  * behind for fsck to use for cleanup.
2113  *
2114  * Any attempt to get a new transaction handle on a journal which is in
2115  * ABORT state will just result in an -EROFS error return.  A
2116  * jbd2_journal_stop on an existing handle will return -EIO if we have
2117  * entered abort state during the update.
2118  *
2119  * Recursive transactions are not disturbed by journal abort until the
2120  * final jbd2_journal_stop, which will receive the -EIO error.
2121  *
2122  * Finally, the jbd2_journal_abort call allows the caller to supply an errno
2123  * which will be recorded (if possible) in the journal superblock.  This
2124  * allows a client to record failure conditions in the middle of a
2125  * transaction without having to complete the transaction to record the
2126  * failure to disk.  ext3_error, for example, now uses this
2127  * functionality.
2128  *
2129  * Errors which originate from within the journaling layer will NOT
2130  * supply an errno; a null errno implies that absolutely no further
2131  * writes are done to the journal (unless there are any already in
2132  * progress).
2133  *
2134  */
2135
2136 void jbd2_journal_abort(journal_t *journal, int errno)
2137 {
2138         __journal_abort_soft(journal, errno);
2139 }
2140
2141 /**
2142  * int jbd2_journal_errno () - returns the journal's error state.
2143  * @journal: journal to examine.
2144  *
2145  * This is the errno number set with jbd2_journal_abort(), the last
2146  * time the journal was mounted - if the journal was stopped
2147  * without calling abort this will be 0.
2148  *
2149  * If the journal has been aborted on this mount time -EROFS will
2150  * be returned.
2151  */
2152 int jbd2_journal_errno(journal_t *journal)
2153 {
2154         int err;
2155
2156         read_lock(&journal->j_state_lock);
2157         if (journal->j_flags & JBD2_ABORT)
2158                 err = -EROFS;
2159         else
2160                 err = journal->j_errno;
2161         read_unlock(&journal->j_state_lock);
2162         return err;
2163 }
2164
2165 /**
2166  * int jbd2_journal_clear_err () - clears the journal's error state
2167  * @journal: journal to act on.
2168  *
2169  * An error must be cleared or acked to take a FS out of readonly
2170  * mode.
2171  */
2172 int jbd2_journal_clear_err(journal_t *journal)
2173 {
2174         int err = 0;
2175
2176         write_lock(&journal->j_state_lock);
2177         if (journal->j_flags & JBD2_ABORT)
2178                 err = -EROFS;
2179         else
2180                 journal->j_errno = 0;
2181         write_unlock(&journal->j_state_lock);
2182         return err;
2183 }
2184
2185 /**
2186  * void jbd2_journal_ack_err() - Ack journal err.
2187  * @journal: journal to act on.
2188  *
2189  * An error must be cleared or acked to take a FS out of readonly
2190  * mode.
2191  */
2192 void jbd2_journal_ack_err(journal_t *journal)
2193 {
2194         write_lock(&journal->j_state_lock);
2195         if (journal->j_errno)
2196                 journal->j_flags |= JBD2_ACK_ERR;
2197         write_unlock(&journal->j_state_lock);
2198 }
2199
2200 int jbd2_journal_blocks_per_page(struct inode *inode)
2201 {
2202         return 1 << (PAGE_CACHE_SHIFT - inode->i_sb->s_blocksize_bits);
2203 }
2204
2205 /*
2206  * helper functions to deal with 32 or 64bit block numbers.
2207  */
2208 size_t journal_tag_bytes(journal_t *journal)
2209 {
2210         size_t sz;
2211
2212         if (jbd2_has_feature_csum3(journal))
2213                 return sizeof(journal_block_tag3_t);
2214
2215         sz = sizeof(journal_block_tag_t);
2216
2217         if (jbd2_has_feature_csum2(journal))
2218                 sz += sizeof(__u16);
2219
2220         if (jbd2_has_feature_64bit(journal))
2221                 return sz;
2222         else
2223                 return sz - sizeof(__u32);
2224 }
2225
2226 /*
2227  * JBD memory management
2228  *
2229  * These functions are used to allocate block-sized chunks of memory
2230  * used for making copies of buffer_head data.  Very often it will be
2231  * page-sized chunks of data, but sometimes it will be in
2232  * sub-page-size chunks.  (For example, 16k pages on Power systems
2233  * with a 4k block file system.)  For blocks smaller than a page, we
2234  * use a SLAB allocator.  There are slab caches for each block size,
2235  * which are allocated at mount time, if necessary, and we only free
2236  * (all of) the slab caches when/if the jbd2 module is unloaded.  For
2237  * this reason we don't need to a mutex to protect access to
2238  * jbd2_slab[] allocating or releasing memory; only in
2239  * jbd2_journal_create_slab().
2240  */
2241 #define JBD2_MAX_SLABS 8
2242 static struct kmem_cache *jbd2_slab[JBD2_MAX_SLABS];
2243
2244 static const char *jbd2_slab_names[JBD2_MAX_SLABS] = {
2245         "jbd2_1k", "jbd2_2k", "jbd2_4k", "jbd2_8k",
2246         "jbd2_16k", "jbd2_32k", "jbd2_64k", "jbd2_128k"
2247 };
2248
2249
2250 static void jbd2_journal_destroy_slabs(void)
2251 {
2252         int i;
2253
2254         for (i = 0; i < JBD2_MAX_SLABS; i++) {
2255                 if (jbd2_slab[i])
2256                         kmem_cache_destroy(jbd2_slab[i]);
2257                 jbd2_slab[i] = NULL;
2258         }
2259 }
2260
2261 static int jbd2_journal_create_slab(size_t size)
2262 {
2263         static DEFINE_MUTEX(jbd2_slab_create_mutex);
2264         int i = order_base_2(size) - 10;
2265         size_t slab_size;
2266
2267         if (size == PAGE_SIZE)
2268                 return 0;
2269
2270         if (i >= JBD2_MAX_SLABS)
2271                 return -EINVAL;
2272
2273         if (unlikely(i < 0))
2274                 i = 0;
2275         mutex_lock(&jbd2_slab_create_mutex);
2276         if (jbd2_slab[i]) {
2277                 mutex_unlock(&jbd2_slab_create_mutex);
2278                 return 0;       /* Already created */
2279         }
2280
2281         slab_size = 1 << (i+10);
2282         jbd2_slab[i] = kmem_cache_create(jbd2_slab_names[i], slab_size,
2283                                          slab_size, 0, NULL);
2284         mutex_unlock(&jbd2_slab_create_mutex);
2285         if (!jbd2_slab[i]) {
2286                 printk(KERN_EMERG "JBD2: no memory for jbd2_slab cache\n");
2287                 return -ENOMEM;
2288         }
2289         return 0;
2290 }
2291
2292 static struct kmem_cache *get_slab(size_t size)
2293 {
2294         int i = order_base_2(size) - 10;
2295
2296         BUG_ON(i >= JBD2_MAX_SLABS);
2297         if (unlikely(i < 0))
2298                 i = 0;
2299         BUG_ON(jbd2_slab[i] == NULL);
2300         return jbd2_slab[i];
2301 }
2302
2303 void *jbd2_alloc(size_t size, gfp_t flags)
2304 {
2305         void *ptr;
2306
2307         BUG_ON(size & (size-1)); /* Must be a power of 2 */
2308
2309         flags |= __GFP_REPEAT;
2310         if (size == PAGE_SIZE)
2311                 ptr = (void *)__get_free_pages(flags, 0);
2312         else if (size > PAGE_SIZE) {
2313                 int order = get_order(size);
2314
2315                 if (order < 3)
2316                         ptr = (void *)__get_free_pages(flags, order);
2317                 else
2318                         ptr = vmalloc(size);
2319         } else
2320                 ptr = kmem_cache_alloc(get_slab(size), flags);
2321
2322         /* Check alignment; SLUB has gotten this wrong in the past,
2323          * and this can lead to user data corruption! */
2324         BUG_ON(((unsigned long) ptr) & (size-1));
2325
2326         return ptr;
2327 }
2328
2329 void jbd2_free(void *ptr, size_t size)
2330 {
2331         if (size == PAGE_SIZE) {
2332                 free_pages((unsigned long)ptr, 0);
2333                 return;
2334         }
2335         if (size > PAGE_SIZE) {
2336                 int order = get_order(size);
2337
2338                 if (order < 3)
2339                         free_pages((unsigned long)ptr, order);
2340                 else
2341                         vfree(ptr);
2342                 return;
2343         }
2344         kmem_cache_free(get_slab(size), ptr);
2345 };
2346
2347 /*
2348  * Journal_head storage management
2349  */
2350 static struct kmem_cache *jbd2_journal_head_cache;
2351 #ifdef CONFIG_JBD2_DEBUG
2352 static atomic_t nr_journal_heads = ATOMIC_INIT(0);
2353 #endif
2354
2355 static int jbd2_journal_init_journal_head_cache(void)
2356 {
2357         int retval;
2358
2359         J_ASSERT(jbd2_journal_head_cache == NULL);
2360         jbd2_journal_head_cache = kmem_cache_create("jbd2_journal_head",
2361                                 sizeof(struct journal_head),
2362                                 0,              /* offset */
2363                                 SLAB_TEMPORARY | SLAB_DESTROY_BY_RCU,
2364                                 NULL);          /* ctor */
2365         retval = 0;
2366         if (!jbd2_journal_head_cache) {
2367                 retval = -ENOMEM;
2368                 printk(KERN_EMERG "JBD2: no memory for journal_head cache\n");
2369         }
2370         return retval;
2371 }
2372
2373 static void jbd2_journal_destroy_journal_head_cache(void)
2374 {
2375         if (jbd2_journal_head_cache) {
2376                 kmem_cache_destroy(jbd2_journal_head_cache);
2377                 jbd2_journal_head_cache = NULL;
2378         }
2379 }
2380
2381 /*
2382  * journal_head splicing and dicing
2383  */
2384 static struct journal_head *journal_alloc_journal_head(void)
2385 {
2386         struct journal_head *ret;
2387
2388 #ifdef CONFIG_JBD2_DEBUG
2389         atomic_inc(&nr_journal_heads);
2390 #endif
2391         ret = kmem_cache_zalloc(jbd2_journal_head_cache, GFP_NOFS);
2392         if (!ret) {
2393                 jbd_debug(1, "out of memory for journal_head\n");
2394                 pr_notice_ratelimited("ENOMEM in %s, retrying.\n", __func__);
2395                 ret = kmem_cache_zalloc(jbd2_journal_head_cache,
2396                                 GFP_NOFS | __GFP_NOFAIL);
2397         }
2398         return ret;
2399 }
2400
2401 static void journal_free_journal_head(struct journal_head *jh)
2402 {
2403 #ifdef CONFIG_JBD2_DEBUG
2404         atomic_dec(&nr_journal_heads);
2405         memset(jh, JBD2_POISON_FREE, sizeof(*jh));
2406 #endif
2407         kmem_cache_free(jbd2_journal_head_cache, jh);
2408 }
2409
2410 /*
2411  * A journal_head is attached to a buffer_head whenever JBD has an
2412  * interest in the buffer.
2413  *
2414  * Whenever a buffer has an attached journal_head, its ->b_state:BH_JBD bit
2415  * is set.  This bit is tested in core kernel code where we need to take
2416  * JBD-specific actions.  Testing the zeroness of ->b_private is not reliable
2417  * there.
2418  *
2419  * When a buffer has its BH_JBD bit set, its ->b_count is elevated by one.
2420  *
2421  * When a buffer has its BH_JBD bit set it is immune from being released by
2422  * core kernel code, mainly via ->b_count.
2423  *
2424  * A journal_head is detached from its buffer_head when the journal_head's
2425  * b_jcount reaches zero. Running transaction (b_transaction) and checkpoint
2426  * transaction (b_cp_transaction) hold their references to b_jcount.
2427  *
2428  * Various places in the kernel want to attach a journal_head to a buffer_head
2429  * _before_ attaching the journal_head to a transaction.  To protect the
2430  * journal_head in this situation, jbd2_journal_add_journal_head elevates the
2431  * journal_head's b_jcount refcount by one.  The caller must call
2432  * jbd2_journal_put_journal_head() to undo this.
2433  *
2434  * So the typical usage would be:
2435  *
2436  *      (Attach a journal_head if needed.  Increments b_jcount)
2437  *      struct journal_head *jh = jbd2_journal_add_journal_head(bh);
2438  *      ...
2439  *      (Get another reference for transaction)
2440  *      jbd2_journal_grab_journal_head(bh);
2441  *      jh->b_transaction = xxx;
2442  *      (Put original reference)
2443  *      jbd2_journal_put_journal_head(jh);
2444  */
2445
2446 /*
2447  * Give a buffer_head a journal_head.
2448  *
2449  * May sleep.
2450  */
2451 struct journal_head *jbd2_journal_add_journal_head(struct buffer_head *bh)
2452 {
2453         struct journal_head *jh;
2454         struct journal_head *new_jh = NULL;
2455
2456 repeat:
2457         if (!buffer_jbd(bh))
2458                 new_jh = journal_alloc_journal_head();
2459
2460         jbd_lock_bh_journal_head(bh);
2461         if (buffer_jbd(bh)) {
2462                 jh = bh2jh(bh);
2463         } else {
2464                 J_ASSERT_BH(bh,
2465                         (atomic_read(&bh->b_count) > 0) ||
2466                         (bh->b_page && bh->b_page->mapping));
2467
2468                 if (!new_jh) {
2469                         jbd_unlock_bh_journal_head(bh);
2470                         goto repeat;
2471                 }
2472
2473                 jh = new_jh;
2474                 new_jh = NULL;          /* We consumed it */
2475                 set_buffer_jbd(bh);
2476                 bh->b_private = jh;
2477                 jh->b_bh = bh;
2478                 get_bh(bh);
2479                 BUFFER_TRACE(bh, "added journal_head");
2480         }
2481         jh->b_jcount++;
2482         jbd_unlock_bh_journal_head(bh);
2483         if (new_jh)
2484                 journal_free_journal_head(new_jh);
2485         return bh->b_private;
2486 }
2487
2488 /*
2489  * Grab a ref against this buffer_head's journal_head.  If it ended up not
2490  * having a journal_head, return NULL
2491  */
2492 struct journal_head *jbd2_journal_grab_journal_head(struct buffer_head *bh)
2493 {
2494         struct journal_head *jh = NULL;
2495
2496         jbd_lock_bh_journal_head(bh);
2497         if (buffer_jbd(bh)) {
2498                 jh = bh2jh(bh);
2499                 jh->b_jcount++;
2500         }
2501         jbd_unlock_bh_journal_head(bh);
2502         return jh;
2503 }
2504
2505 static void __journal_remove_journal_head(struct buffer_head *bh)
2506 {
2507         struct journal_head *jh = bh2jh(bh);
2508
2509         J_ASSERT_JH(jh, jh->b_jcount >= 0);
2510         J_ASSERT_JH(jh, jh->b_transaction == NULL);
2511         J_ASSERT_JH(jh, jh->b_next_transaction == NULL);
2512         J_ASSERT_JH(jh, jh->b_cp_transaction == NULL);
2513         J_ASSERT_JH(jh, jh->b_jlist == BJ_None);
2514         J_ASSERT_BH(bh, buffer_jbd(bh));
2515         J_ASSERT_BH(bh, jh2bh(jh) == bh);
2516         BUFFER_TRACE(bh, "remove journal_head");
2517         if (jh->b_frozen_data) {
2518                 printk(KERN_WARNING "%s: freeing b_frozen_data\n", __func__);
2519                 jbd2_free(jh->b_frozen_data, bh->b_size);
2520         }
2521         if (jh->b_committed_data) {
2522                 printk(KERN_WARNING "%s: freeing b_committed_data\n", __func__);
2523                 jbd2_free(jh->b_committed_data, bh->b_size);
2524         }
2525         bh->b_private = NULL;
2526         jh->b_bh = NULL;        /* debug, really */
2527         clear_buffer_jbd(bh);
2528         journal_free_journal_head(jh);
2529 }
2530
2531 /*
2532  * Drop a reference on the passed journal_head.  If it fell to zero then
2533  * release the journal_head from the buffer_head.
2534  */
2535 void jbd2_journal_put_journal_head(struct journal_head *jh)
2536 {
2537         struct buffer_head *bh = jh2bh(jh);
2538
2539         jbd_lock_bh_journal_head(bh);
2540         J_ASSERT_JH(jh, jh->b_jcount > 0);
2541         --jh->b_jcount;
2542         if (!jh->b_jcount) {
2543                 __journal_remove_journal_head(bh);
2544                 jbd_unlock_bh_journal_head(bh);
2545                 __brelse(bh);
2546         } else
2547                 jbd_unlock_bh_journal_head(bh);
2548 }
2549
2550 /*
2551  * Initialize jbd inode head
2552  */
2553 void jbd2_journal_init_jbd_inode(struct jbd2_inode *jinode, struct inode *inode)
2554 {
2555         jinode->i_transaction = NULL;
2556         jinode->i_next_transaction = NULL;
2557         jinode->i_vfs_inode = inode;
2558         jinode->i_flags = 0;
2559         INIT_LIST_HEAD(&jinode->i_list);
2560 }
2561
2562 /*
2563  * Function to be called before we start removing inode from memory (i.e.,
2564  * clear_inode() is a fine place to be called from). It removes inode from
2565  * transaction's lists.
2566  */
2567 void jbd2_journal_release_jbd_inode(journal_t *journal,
2568                                     struct jbd2_inode *jinode)
2569 {
2570         if (!journal)
2571                 return;
2572 restart:
2573         spin_lock(&journal->j_list_lock);
2574         /* Is commit writing out inode - we have to wait */
2575         if (test_bit(__JI_COMMIT_RUNNING, &jinode->i_flags)) {
2576                 wait_queue_head_t *wq;
2577                 DEFINE_WAIT_BIT(wait, &jinode->i_flags, __JI_COMMIT_RUNNING);
2578                 wq = bit_waitqueue(&jinode->i_flags, __JI_COMMIT_RUNNING);
2579                 prepare_to_wait(wq, &wait.wait, TASK_UNINTERRUPTIBLE);
2580                 spin_unlock(&journal->j_list_lock);
2581                 schedule();
2582                 finish_wait(wq, &wait.wait);
2583                 goto restart;
2584         }
2585
2586         if (jinode->i_transaction) {
2587                 list_del(&jinode->i_list);
2588                 jinode->i_transaction = NULL;
2589         }
2590         spin_unlock(&journal->j_list_lock);
2591 }
2592
2593
2594 #ifdef CONFIG_PROC_FS
2595
2596 #define JBD2_STATS_PROC_NAME "fs/jbd2"
2597
2598 static void __init jbd2_create_jbd_stats_proc_entry(void)
2599 {
2600         proc_jbd2_stats = proc_mkdir(JBD2_STATS_PROC_NAME, NULL);
2601 }
2602
2603 static void __exit jbd2_remove_jbd_stats_proc_entry(void)
2604 {
2605         if (proc_jbd2_stats)
2606                 remove_proc_entry(JBD2_STATS_PROC_NAME, NULL);
2607 }
2608
2609 #else
2610
2611 #define jbd2_create_jbd_stats_proc_entry() do {} while (0)
2612 #define jbd2_remove_jbd_stats_proc_entry() do {} while (0)
2613
2614 #endif
2615
2616 struct kmem_cache *jbd2_handle_cache, *jbd2_inode_cache;
2617
2618 static int __init jbd2_journal_init_handle_cache(void)
2619 {
2620         jbd2_handle_cache = KMEM_CACHE(jbd2_journal_handle, SLAB_TEMPORARY);
2621         if (jbd2_handle_cache == NULL) {
2622                 printk(KERN_EMERG "JBD2: failed to create handle cache\n");
2623                 return -ENOMEM;
2624         }
2625         jbd2_inode_cache = KMEM_CACHE(jbd2_inode, 0);
2626         if (jbd2_inode_cache == NULL) {
2627                 printk(KERN_EMERG "JBD2: failed to create inode cache\n");
2628                 kmem_cache_destroy(jbd2_handle_cache);
2629                 return -ENOMEM;
2630         }
2631         return 0;
2632 }
2633
2634 static void jbd2_journal_destroy_handle_cache(void)
2635 {
2636         if (jbd2_handle_cache)
2637                 kmem_cache_destroy(jbd2_handle_cache);
2638         if (jbd2_inode_cache)
2639                 kmem_cache_destroy(jbd2_inode_cache);
2640
2641 }
2642
2643 /*
2644  * Module startup and shutdown
2645  */
2646
2647 static int __init journal_init_caches(void)
2648 {
2649         int ret;
2650
2651         ret = jbd2_journal_init_revoke_caches();
2652         if (ret == 0)
2653                 ret = jbd2_journal_init_journal_head_cache();
2654         if (ret == 0)
2655                 ret = jbd2_journal_init_handle_cache();
2656         if (ret == 0)
2657                 ret = jbd2_journal_init_transaction_cache();
2658         return ret;
2659 }
2660
2661 static void jbd2_journal_destroy_caches(void)
2662 {
2663         jbd2_journal_destroy_revoke_caches();
2664         jbd2_journal_destroy_journal_head_cache();
2665         jbd2_journal_destroy_handle_cache();
2666         jbd2_journal_destroy_transaction_cache();
2667         jbd2_journal_destroy_slabs();
2668 }
2669
2670 static int __init journal_init(void)
2671 {
2672         int ret;
2673
2674         BUILD_BUG_ON(sizeof(struct journal_superblock_s) != 1024);
2675
2676         ret = journal_init_caches();
2677         if (ret == 0) {
2678                 jbd2_create_jbd_stats_proc_entry();
2679         } else {
2680                 jbd2_journal_destroy_caches();
2681         }
2682         return ret;
2683 }
2684
2685 static void __exit journal_exit(void)
2686 {
2687 #ifdef CONFIG_JBD2_DEBUG
2688         int n = atomic_read(&nr_journal_heads);
2689         if (n)
2690                 printk(KERN_ERR "JBD2: leaked %d journal_heads!\n", n);
2691 #endif
2692         jbd2_remove_jbd_stats_proc_entry();
2693         jbd2_journal_destroy_caches();
2694 }
2695
2696 MODULE_LICENSE("GPL");
2697 module_init(journal_init);
2698 module_exit(journal_exit);
2699