JBD2: debug code cleanup.
[linux-2.6.git] / fs / jbd2 / journal.c
1 /*
2  * linux/fs/jbd2/journal.c
3  *
4  * Written by Stephen C. Tweedie <sct@redhat.com>, 1998
5  *
6  * Copyright 1998 Red Hat corp --- All Rights Reserved
7  *
8  * This file is part of the Linux kernel and is made available under
9  * the terms of the GNU General Public License, version 2, or at your
10  * option, any later version, incorporated herein by reference.
11  *
12  * Generic filesystem journal-writing code; part of the ext2fs
13  * journaling system.
14  *
15  * This file manages journals: areas of disk reserved for logging
16  * transactional updates.  This includes the kernel journaling thread
17  * which is responsible for scheduling updates to the log.
18  *
19  * We do not actually manage the physical storage of the journal in this
20  * file: that is left to a per-journal policy function, which allows us
21  * to store the journal within a filesystem-specified area for ext2
22  * journaling (ext2 can use a reserved inode for storing the log).
23  */
24
25 #include <linux/module.h>
26 #include <linux/time.h>
27 #include <linux/fs.h>
28 #include <linux/jbd2.h>
29 #include <linux/errno.h>
30 #include <linux/slab.h>
31 #include <linux/init.h>
32 #include <linux/mm.h>
33 #include <linux/freezer.h>
34 #include <linux/pagemap.h>
35 #include <linux/kthread.h>
36 #include <linux/poison.h>
37 #include <linux/proc_fs.h>
38 #include <linux/debugfs.h>
39
40 #include <asm/uaccess.h>
41 #include <asm/page.h>
42
43 EXPORT_SYMBOL(jbd2_journal_start);
44 EXPORT_SYMBOL(jbd2_journal_restart);
45 EXPORT_SYMBOL(jbd2_journal_extend);
46 EXPORT_SYMBOL(jbd2_journal_stop);
47 EXPORT_SYMBOL(jbd2_journal_lock_updates);
48 EXPORT_SYMBOL(jbd2_journal_unlock_updates);
49 EXPORT_SYMBOL(jbd2_journal_get_write_access);
50 EXPORT_SYMBOL(jbd2_journal_get_create_access);
51 EXPORT_SYMBOL(jbd2_journal_get_undo_access);
52 EXPORT_SYMBOL(jbd2_journal_dirty_data);
53 EXPORT_SYMBOL(jbd2_journal_dirty_metadata);
54 EXPORT_SYMBOL(jbd2_journal_release_buffer);
55 EXPORT_SYMBOL(jbd2_journal_forget);
56 #if 0
57 EXPORT_SYMBOL(journal_sync_buffer);
58 #endif
59 EXPORT_SYMBOL(jbd2_journal_flush);
60 EXPORT_SYMBOL(jbd2_journal_revoke);
61
62 EXPORT_SYMBOL(jbd2_journal_init_dev);
63 EXPORT_SYMBOL(jbd2_journal_init_inode);
64 EXPORT_SYMBOL(jbd2_journal_update_format);
65 EXPORT_SYMBOL(jbd2_journal_check_used_features);
66 EXPORT_SYMBOL(jbd2_journal_check_available_features);
67 EXPORT_SYMBOL(jbd2_journal_set_features);
68 EXPORT_SYMBOL(jbd2_journal_create);
69 EXPORT_SYMBOL(jbd2_journal_load);
70 EXPORT_SYMBOL(jbd2_journal_destroy);
71 EXPORT_SYMBOL(jbd2_journal_update_superblock);
72 EXPORT_SYMBOL(jbd2_journal_abort);
73 EXPORT_SYMBOL(jbd2_journal_errno);
74 EXPORT_SYMBOL(jbd2_journal_ack_err);
75 EXPORT_SYMBOL(jbd2_journal_clear_err);
76 EXPORT_SYMBOL(jbd2_log_wait_commit);
77 EXPORT_SYMBOL(jbd2_journal_start_commit);
78 EXPORT_SYMBOL(jbd2_journal_force_commit_nested);
79 EXPORT_SYMBOL(jbd2_journal_wipe);
80 EXPORT_SYMBOL(jbd2_journal_blocks_per_page);
81 EXPORT_SYMBOL(jbd2_journal_invalidatepage);
82 EXPORT_SYMBOL(jbd2_journal_try_to_free_buffers);
83 EXPORT_SYMBOL(jbd2_journal_force_commit);
84
85 static int journal_convert_superblock_v1(journal_t *, journal_superblock_t *);
86 static void __journal_abort_soft (journal_t *journal, int errno);
87
88 /*
89  * Helper function used to manage commit timeouts
90  */
91
92 static void commit_timeout(unsigned long __data)
93 {
94         struct task_struct * p = (struct task_struct *) __data;
95
96         wake_up_process(p);
97 }
98
99 /*
100  * kjournald2: The main thread function used to manage a logging device
101  * journal.
102  *
103  * This kernel thread is responsible for two things:
104  *
105  * 1) COMMIT:  Every so often we need to commit the current state of the
106  *    filesystem to disk.  The journal thread is responsible for writing
107  *    all of the metadata buffers to disk.
108  *
109  * 2) CHECKPOINT: We cannot reuse a used section of the log file until all
110  *    of the data in that part of the log has been rewritten elsewhere on
111  *    the disk.  Flushing these old buffers to reclaim space in the log is
112  *    known as checkpointing, and this thread is responsible for that job.
113  */
114
115 static int kjournald2(void *arg)
116 {
117         journal_t *journal = arg;
118         transaction_t *transaction;
119
120         /*
121          * Set up an interval timer which can be used to trigger a commit wakeup
122          * after the commit interval expires
123          */
124         setup_timer(&journal->j_commit_timer, commit_timeout,
125                         (unsigned long)current);
126
127         /* Record that the journal thread is running */
128         journal->j_task = current;
129         wake_up(&journal->j_wait_done_commit);
130
131         printk(KERN_INFO "kjournald2 starting.  Commit interval %ld seconds\n",
132                         journal->j_commit_interval / HZ);
133
134         /*
135          * And now, wait forever for commit wakeup events.
136          */
137         spin_lock(&journal->j_state_lock);
138
139 loop:
140         if (journal->j_flags & JBD2_UNMOUNT)
141                 goto end_loop;
142
143         jbd_debug(1, "commit_sequence=%d, commit_request=%d\n",
144                 journal->j_commit_sequence, journal->j_commit_request);
145
146         if (journal->j_commit_sequence != journal->j_commit_request) {
147                 jbd_debug(1, "OK, requests differ\n");
148                 spin_unlock(&journal->j_state_lock);
149                 del_timer_sync(&journal->j_commit_timer);
150                 jbd2_journal_commit_transaction(journal);
151                 spin_lock(&journal->j_state_lock);
152                 goto loop;
153         }
154
155         wake_up(&journal->j_wait_done_commit);
156         if (freezing(current)) {
157                 /*
158                  * The simpler the better. Flushing journal isn't a
159                  * good idea, because that depends on threads that may
160                  * be already stopped.
161                  */
162                 jbd_debug(1, "Now suspending kjournald2\n");
163                 spin_unlock(&journal->j_state_lock);
164                 refrigerator();
165                 spin_lock(&journal->j_state_lock);
166         } else {
167                 /*
168                  * We assume on resume that commits are already there,
169                  * so we don't sleep
170                  */
171                 DEFINE_WAIT(wait);
172                 int should_sleep = 1;
173
174                 prepare_to_wait(&journal->j_wait_commit, &wait,
175                                 TASK_INTERRUPTIBLE);
176                 if (journal->j_commit_sequence != journal->j_commit_request)
177                         should_sleep = 0;
178                 transaction = journal->j_running_transaction;
179                 if (transaction && time_after_eq(jiffies,
180                                                 transaction->t_expires))
181                         should_sleep = 0;
182                 if (journal->j_flags & JBD2_UNMOUNT)
183                         should_sleep = 0;
184                 if (should_sleep) {
185                         spin_unlock(&journal->j_state_lock);
186                         schedule();
187                         spin_lock(&journal->j_state_lock);
188                 }
189                 finish_wait(&journal->j_wait_commit, &wait);
190         }
191
192         jbd_debug(1, "kjournald2 wakes\n");
193
194         /*
195          * Were we woken up by a commit wakeup event?
196          */
197         transaction = journal->j_running_transaction;
198         if (transaction && time_after_eq(jiffies, transaction->t_expires)) {
199                 journal->j_commit_request = transaction->t_tid;
200                 jbd_debug(1, "woke because of timeout\n");
201         }
202         goto loop;
203
204 end_loop:
205         spin_unlock(&journal->j_state_lock);
206         del_timer_sync(&journal->j_commit_timer);
207         journal->j_task = NULL;
208         wake_up(&journal->j_wait_done_commit);
209         jbd_debug(1, "Journal thread exiting.\n");
210         return 0;
211 }
212
213 static int jbd2_journal_start_thread(journal_t *journal)
214 {
215         struct task_struct *t;
216
217         t = kthread_run(kjournald2, journal, "kjournald2");
218         if (IS_ERR(t))
219                 return PTR_ERR(t);
220
221         wait_event(journal->j_wait_done_commit, journal->j_task != 0);
222         return 0;
223 }
224
225 static void journal_kill_thread(journal_t *journal)
226 {
227         spin_lock(&journal->j_state_lock);
228         journal->j_flags |= JBD2_UNMOUNT;
229
230         while (journal->j_task) {
231                 wake_up(&journal->j_wait_commit);
232                 spin_unlock(&journal->j_state_lock);
233                 wait_event(journal->j_wait_done_commit, journal->j_task == 0);
234                 spin_lock(&journal->j_state_lock);
235         }
236         spin_unlock(&journal->j_state_lock);
237 }
238
239 /*
240  * jbd2_journal_write_metadata_buffer: write a metadata buffer to the journal.
241  *
242  * Writes a metadata buffer to a given disk block.  The actual IO is not
243  * performed but a new buffer_head is constructed which labels the data
244  * to be written with the correct destination disk block.
245  *
246  * Any magic-number escaping which needs to be done will cause a
247  * copy-out here.  If the buffer happens to start with the
248  * JBD2_MAGIC_NUMBER, then we can't write it to the log directly: the
249  * magic number is only written to the log for descripter blocks.  In
250  * this case, we copy the data and replace the first word with 0, and we
251  * return a result code which indicates that this buffer needs to be
252  * marked as an escaped buffer in the corresponding log descriptor
253  * block.  The missing word can then be restored when the block is read
254  * during recovery.
255  *
256  * If the source buffer has already been modified by a new transaction
257  * since we took the last commit snapshot, we use the frozen copy of
258  * that data for IO.  If we end up using the existing buffer_head's data
259  * for the write, then we *have* to lock the buffer to prevent anyone
260  * else from using and possibly modifying it while the IO is in
261  * progress.
262  *
263  * The function returns a pointer to the buffer_heads to be used for IO.
264  *
265  * We assume that the journal has already been locked in this function.
266  *
267  * Return value:
268  *  <0: Error
269  * >=0: Finished OK
270  *
271  * On success:
272  * Bit 0 set == escape performed on the data
273  * Bit 1 set == buffer copy-out performed (kfree the data after IO)
274  */
275
276 int jbd2_journal_write_metadata_buffer(transaction_t *transaction,
277                                   struct journal_head  *jh_in,
278                                   struct journal_head **jh_out,
279                                   unsigned long long blocknr)
280 {
281         int need_copy_out = 0;
282         int done_copy_out = 0;
283         int do_escape = 0;
284         char *mapped_data;
285         struct buffer_head *new_bh;
286         struct journal_head *new_jh;
287         struct page *new_page;
288         unsigned int new_offset;
289         struct buffer_head *bh_in = jh2bh(jh_in);
290
291         /*
292          * The buffer really shouldn't be locked: only the current committing
293          * transaction is allowed to write it, so nobody else is allowed
294          * to do any IO.
295          *
296          * akpm: except if we're journalling data, and write() output is
297          * also part of a shared mapping, and another thread has
298          * decided to launch a writepage() against this buffer.
299          */
300         J_ASSERT_BH(bh_in, buffer_jbddirty(bh_in));
301
302         new_bh = alloc_buffer_head(GFP_NOFS|__GFP_NOFAIL);
303
304         /*
305          * If a new transaction has already done a buffer copy-out, then
306          * we use that version of the data for the commit.
307          */
308         jbd_lock_bh_state(bh_in);
309 repeat:
310         if (jh_in->b_frozen_data) {
311                 done_copy_out = 1;
312                 new_page = virt_to_page(jh_in->b_frozen_data);
313                 new_offset = offset_in_page(jh_in->b_frozen_data);
314         } else {
315                 new_page = jh2bh(jh_in)->b_page;
316                 new_offset = offset_in_page(jh2bh(jh_in)->b_data);
317         }
318
319         mapped_data = kmap_atomic(new_page, KM_USER0);
320         /*
321          * Check for escaping
322          */
323         if (*((__be32 *)(mapped_data + new_offset)) ==
324                                 cpu_to_be32(JBD2_MAGIC_NUMBER)) {
325                 need_copy_out = 1;
326                 do_escape = 1;
327         }
328         kunmap_atomic(mapped_data, KM_USER0);
329
330         /*
331          * Do we need to do a data copy?
332          */
333         if (need_copy_out && !done_copy_out) {
334                 char *tmp;
335
336                 jbd_unlock_bh_state(bh_in);
337                 tmp = jbd2_alloc(bh_in->b_size, GFP_NOFS);
338                 jbd_lock_bh_state(bh_in);
339                 if (jh_in->b_frozen_data) {
340                         jbd2_free(tmp, bh_in->b_size);
341                         goto repeat;
342                 }
343
344                 jh_in->b_frozen_data = tmp;
345                 mapped_data = kmap_atomic(new_page, KM_USER0);
346                 memcpy(tmp, mapped_data + new_offset, jh2bh(jh_in)->b_size);
347                 kunmap_atomic(mapped_data, KM_USER0);
348
349                 new_page = virt_to_page(tmp);
350                 new_offset = offset_in_page(tmp);
351                 done_copy_out = 1;
352         }
353
354         /*
355          * Did we need to do an escaping?  Now we've done all the
356          * copying, we can finally do so.
357          */
358         if (do_escape) {
359                 mapped_data = kmap_atomic(new_page, KM_USER0);
360                 *((unsigned int *)(mapped_data + new_offset)) = 0;
361                 kunmap_atomic(mapped_data, KM_USER0);
362         }
363
364         /* keep subsequent assertions sane */
365         new_bh->b_state = 0;
366         init_buffer(new_bh, NULL, NULL);
367         atomic_set(&new_bh->b_count, 1);
368         jbd_unlock_bh_state(bh_in);
369
370         new_jh = jbd2_journal_add_journal_head(new_bh); /* This sleeps */
371
372         set_bh_page(new_bh, new_page, new_offset);
373         new_jh->b_transaction = NULL;
374         new_bh->b_size = jh2bh(jh_in)->b_size;
375         new_bh->b_bdev = transaction->t_journal->j_dev;
376         new_bh->b_blocknr = blocknr;
377         set_buffer_mapped(new_bh);
378         set_buffer_dirty(new_bh);
379
380         *jh_out = new_jh;
381
382         /*
383          * The to-be-written buffer needs to get moved to the io queue,
384          * and the original buffer whose contents we are shadowing or
385          * copying is moved to the transaction's shadow queue.
386          */
387         JBUFFER_TRACE(jh_in, "file as BJ_Shadow");
388         jbd2_journal_file_buffer(jh_in, transaction, BJ_Shadow);
389         JBUFFER_TRACE(new_jh, "file as BJ_IO");
390         jbd2_journal_file_buffer(new_jh, transaction, BJ_IO);
391
392         return do_escape | (done_copy_out << 1);
393 }
394
395 /*
396  * Allocation code for the journal file.  Manage the space left in the
397  * journal, so that we can begin checkpointing when appropriate.
398  */
399
400 /*
401  * __jbd2_log_space_left: Return the number of free blocks left in the journal.
402  *
403  * Called with the journal already locked.
404  *
405  * Called under j_state_lock
406  */
407
408 int __jbd2_log_space_left(journal_t *journal)
409 {
410         int left = journal->j_free;
411
412         assert_spin_locked(&journal->j_state_lock);
413
414         /*
415          * Be pessimistic here about the number of those free blocks which
416          * might be required for log descriptor control blocks.
417          */
418
419 #define MIN_LOG_RESERVED_BLOCKS 32 /* Allow for rounding errors */
420
421         left -= MIN_LOG_RESERVED_BLOCKS;
422
423         if (left <= 0)
424                 return 0;
425         left -= (left >> 3);
426         return left;
427 }
428
429 /*
430  * Called under j_state_lock.  Returns true if a transaction was started.
431  */
432 int __jbd2_log_start_commit(journal_t *journal, tid_t target)
433 {
434         /*
435          * Are we already doing a recent enough commit?
436          */
437         if (!tid_geq(journal->j_commit_request, target)) {
438                 /*
439                  * We want a new commit: OK, mark the request and wakup the
440                  * commit thread.  We do _not_ do the commit ourselves.
441                  */
442
443                 journal->j_commit_request = target;
444                 jbd_debug(1, "JBD: requesting commit %d/%d\n",
445                           journal->j_commit_request,
446                           journal->j_commit_sequence);
447                 wake_up(&journal->j_wait_commit);
448                 return 1;
449         }
450         return 0;
451 }
452
453 int jbd2_log_start_commit(journal_t *journal, tid_t tid)
454 {
455         int ret;
456
457         spin_lock(&journal->j_state_lock);
458         ret = __jbd2_log_start_commit(journal, tid);
459         spin_unlock(&journal->j_state_lock);
460         return ret;
461 }
462
463 /*
464  * Force and wait upon a commit if the calling process is not within
465  * transaction.  This is used for forcing out undo-protected data which contains
466  * bitmaps, when the fs is running out of space.
467  *
468  * We can only force the running transaction if we don't have an active handle;
469  * otherwise, we will deadlock.
470  *
471  * Returns true if a transaction was started.
472  */
473 int jbd2_journal_force_commit_nested(journal_t *journal)
474 {
475         transaction_t *transaction = NULL;
476         tid_t tid;
477
478         spin_lock(&journal->j_state_lock);
479         if (journal->j_running_transaction && !current->journal_info) {
480                 transaction = journal->j_running_transaction;
481                 __jbd2_log_start_commit(journal, transaction->t_tid);
482         } else if (journal->j_committing_transaction)
483                 transaction = journal->j_committing_transaction;
484
485         if (!transaction) {
486                 spin_unlock(&journal->j_state_lock);
487                 return 0;       /* Nothing to retry */
488         }
489
490         tid = transaction->t_tid;
491         spin_unlock(&journal->j_state_lock);
492         jbd2_log_wait_commit(journal, tid);
493         return 1;
494 }
495
496 /*
497  * Start a commit of the current running transaction (if any).  Returns true
498  * if a transaction was started, and fills its tid in at *ptid
499  */
500 int jbd2_journal_start_commit(journal_t *journal, tid_t *ptid)
501 {
502         int ret = 0;
503
504         spin_lock(&journal->j_state_lock);
505         if (journal->j_running_transaction) {
506                 tid_t tid = journal->j_running_transaction->t_tid;
507
508                 ret = __jbd2_log_start_commit(journal, tid);
509                 if (ret && ptid)
510                         *ptid = tid;
511         } else if (journal->j_committing_transaction && ptid) {
512                 /*
513                  * If ext3_write_super() recently started a commit, then we
514                  * have to wait for completion of that transaction
515                  */
516                 *ptid = journal->j_committing_transaction->t_tid;
517                 ret = 1;
518         }
519         spin_unlock(&journal->j_state_lock);
520         return ret;
521 }
522
523 /*
524  * Wait for a specified commit to complete.
525  * The caller may not hold the journal lock.
526  */
527 int jbd2_log_wait_commit(journal_t *journal, tid_t tid)
528 {
529         int err = 0;
530
531 #ifdef CONFIG_JBD2_DEBUG
532         spin_lock(&journal->j_state_lock);
533         if (!tid_geq(journal->j_commit_request, tid)) {
534                 printk(KERN_EMERG
535                        "%s: error: j_commit_request=%d, tid=%d\n",
536                        __FUNCTION__, journal->j_commit_request, tid);
537         }
538         spin_unlock(&journal->j_state_lock);
539 #endif
540         spin_lock(&journal->j_state_lock);
541         while (tid_gt(tid, journal->j_commit_sequence)) {
542                 jbd_debug(1, "JBD: want %d, j_commit_sequence=%d\n",
543                                   tid, journal->j_commit_sequence);
544                 wake_up(&journal->j_wait_commit);
545                 spin_unlock(&journal->j_state_lock);
546                 wait_event(journal->j_wait_done_commit,
547                                 !tid_gt(tid, journal->j_commit_sequence));
548                 spin_lock(&journal->j_state_lock);
549         }
550         spin_unlock(&journal->j_state_lock);
551
552         if (unlikely(is_journal_aborted(journal))) {
553                 printk(KERN_EMERG "journal commit I/O error\n");
554                 err = -EIO;
555         }
556         return err;
557 }
558
559 /*
560  * Log buffer allocation routines:
561  */
562
563 int jbd2_journal_next_log_block(journal_t *journal, unsigned long long *retp)
564 {
565         unsigned long blocknr;
566
567         spin_lock(&journal->j_state_lock);
568         J_ASSERT(journal->j_free > 1);
569
570         blocknr = journal->j_head;
571         journal->j_head++;
572         journal->j_free--;
573         if (journal->j_head == journal->j_last)
574                 journal->j_head = journal->j_first;
575         spin_unlock(&journal->j_state_lock);
576         return jbd2_journal_bmap(journal, blocknr, retp);
577 }
578
579 /*
580  * Conversion of logical to physical block numbers for the journal
581  *
582  * On external journals the journal blocks are identity-mapped, so
583  * this is a no-op.  If needed, we can use j_blk_offset - everything is
584  * ready.
585  */
586 int jbd2_journal_bmap(journal_t *journal, unsigned long blocknr,
587                  unsigned long long *retp)
588 {
589         int err = 0;
590         unsigned long long ret;
591
592         if (journal->j_inode) {
593                 ret = bmap(journal->j_inode, blocknr);
594                 if (ret)
595                         *retp = ret;
596                 else {
597                         char b[BDEVNAME_SIZE];
598
599                         printk(KERN_ALERT "%s: journal block not found "
600                                         "at offset %lu on %s\n",
601                                 __FUNCTION__,
602                                 blocknr,
603                                 bdevname(journal->j_dev, b));
604                         err = -EIO;
605                         __journal_abort_soft(journal, err);
606                 }
607         } else {
608                 *retp = blocknr; /* +journal->j_blk_offset */
609         }
610         return err;
611 }
612
613 /*
614  * We play buffer_head aliasing tricks to write data/metadata blocks to
615  * the journal without copying their contents, but for journal
616  * descriptor blocks we do need to generate bona fide buffers.
617  *
618  * After the caller of jbd2_journal_get_descriptor_buffer() has finished modifying
619  * the buffer's contents they really should run flush_dcache_page(bh->b_page).
620  * But we don't bother doing that, so there will be coherency problems with
621  * mmaps of blockdevs which hold live JBD-controlled filesystems.
622  */
623 struct journal_head *jbd2_journal_get_descriptor_buffer(journal_t *journal)
624 {
625         struct buffer_head *bh;
626         unsigned long long blocknr;
627         int err;
628
629         err = jbd2_journal_next_log_block(journal, &blocknr);
630
631         if (err)
632                 return NULL;
633
634         bh = __getblk(journal->j_dev, blocknr, journal->j_blocksize);
635         lock_buffer(bh);
636         memset(bh->b_data, 0, journal->j_blocksize);
637         set_buffer_uptodate(bh);
638         unlock_buffer(bh);
639         BUFFER_TRACE(bh, "return this buffer");
640         return jbd2_journal_add_journal_head(bh);
641 }
642
643 /*
644  * Management for journal control blocks: functions to create and
645  * destroy journal_t structures, and to initialise and read existing
646  * journal blocks from disk.  */
647
648 /* First: create and setup a journal_t object in memory.  We initialise
649  * very few fields yet: that has to wait until we have created the
650  * journal structures from from scratch, or loaded them from disk. */
651
652 static journal_t * journal_init_common (void)
653 {
654         journal_t *journal;
655         int err;
656
657         journal = kzalloc(sizeof(*journal), GFP_KERNEL|__GFP_NOFAIL);
658         if (!journal)
659                 goto fail;
660
661         init_waitqueue_head(&journal->j_wait_transaction_locked);
662         init_waitqueue_head(&journal->j_wait_logspace);
663         init_waitqueue_head(&journal->j_wait_done_commit);
664         init_waitqueue_head(&journal->j_wait_checkpoint);
665         init_waitqueue_head(&journal->j_wait_commit);
666         init_waitqueue_head(&journal->j_wait_updates);
667         mutex_init(&journal->j_barrier);
668         mutex_init(&journal->j_checkpoint_mutex);
669         spin_lock_init(&journal->j_revoke_lock);
670         spin_lock_init(&journal->j_list_lock);
671         spin_lock_init(&journal->j_state_lock);
672
673         journal->j_commit_interval = (HZ * JBD2_DEFAULT_MAX_COMMIT_AGE);
674
675         /* The journal is marked for error until we succeed with recovery! */
676         journal->j_flags = JBD2_ABORT;
677
678         /* Set up a default-sized revoke table for the new mount. */
679         err = jbd2_journal_init_revoke(journal, JOURNAL_REVOKE_DEFAULT_HASH);
680         if (err) {
681                 kfree(journal);
682                 goto fail;
683         }
684         return journal;
685 fail:
686         return NULL;
687 }
688
689 /* jbd2_journal_init_dev and jbd2_journal_init_inode:
690  *
691  * Create a journal structure assigned some fixed set of disk blocks to
692  * the journal.  We don't actually touch those disk blocks yet, but we
693  * need to set up all of the mapping information to tell the journaling
694  * system where the journal blocks are.
695  *
696  */
697
698 /**
699  *  journal_t * jbd2_journal_init_dev() - creates an initialises a journal structure
700  *  @bdev: Block device on which to create the journal
701  *  @fs_dev: Device which hold journalled filesystem for this journal.
702  *  @start: Block nr Start of journal.
703  *  @len:  Length of the journal in blocks.
704  *  @blocksize: blocksize of journalling device
705  *  @returns: a newly created journal_t *
706  *
707  *  jbd2_journal_init_dev creates a journal which maps a fixed contiguous
708  *  range of blocks on an arbitrary block device.
709  *
710  */
711 journal_t * jbd2_journal_init_dev(struct block_device *bdev,
712                         struct block_device *fs_dev,
713                         unsigned long long start, int len, int blocksize)
714 {
715         journal_t *journal = journal_init_common();
716         struct buffer_head *bh;
717         int n;
718
719         if (!journal)
720                 return NULL;
721
722         /* journal descriptor can store up to n blocks -bzzz */
723         journal->j_blocksize = blocksize;
724         n = journal->j_blocksize / sizeof(journal_block_tag_t);
725         journal->j_wbufsize = n;
726         journal->j_wbuf = kmalloc(n * sizeof(struct buffer_head*), GFP_KERNEL);
727         if (!journal->j_wbuf) {
728                 printk(KERN_ERR "%s: Cant allocate bhs for commit thread\n",
729                         __FUNCTION__);
730                 kfree(journal);
731                 journal = NULL;
732                 goto out;
733         }
734         journal->j_dev = bdev;
735         journal->j_fs_dev = fs_dev;
736         journal->j_blk_offset = start;
737         journal->j_maxlen = len;
738
739         bh = __getblk(journal->j_dev, start, journal->j_blocksize);
740         J_ASSERT(bh != NULL);
741         journal->j_sb_buffer = bh;
742         journal->j_superblock = (journal_superblock_t *)bh->b_data;
743 out:
744         return journal;
745 }
746
747 /**
748  *  journal_t * jbd2_journal_init_inode () - creates a journal which maps to a inode.
749  *  @inode: An inode to create the journal in
750  *
751  * jbd2_journal_init_inode creates a journal which maps an on-disk inode as
752  * the journal.  The inode must exist already, must support bmap() and
753  * must have all data blocks preallocated.
754  */
755 journal_t * jbd2_journal_init_inode (struct inode *inode)
756 {
757         struct buffer_head *bh;
758         journal_t *journal = journal_init_common();
759         int err;
760         int n;
761         unsigned long long blocknr;
762
763         if (!journal)
764                 return NULL;
765
766         journal->j_dev = journal->j_fs_dev = inode->i_sb->s_bdev;
767         journal->j_inode = inode;
768         jbd_debug(1,
769                   "journal %p: inode %s/%ld, size %Ld, bits %d, blksize %ld\n",
770                   journal, inode->i_sb->s_id, inode->i_ino,
771                   (long long) inode->i_size,
772                   inode->i_sb->s_blocksize_bits, inode->i_sb->s_blocksize);
773
774         journal->j_maxlen = inode->i_size >> inode->i_sb->s_blocksize_bits;
775         journal->j_blocksize = inode->i_sb->s_blocksize;
776
777         /* journal descriptor can store up to n blocks -bzzz */
778         n = journal->j_blocksize / sizeof(journal_block_tag_t);
779         journal->j_wbufsize = n;
780         journal->j_wbuf = kmalloc(n * sizeof(struct buffer_head*), GFP_KERNEL);
781         if (!journal->j_wbuf) {
782                 printk(KERN_ERR "%s: Cant allocate bhs for commit thread\n",
783                         __FUNCTION__);
784                 kfree(journal);
785                 return NULL;
786         }
787
788         err = jbd2_journal_bmap(journal, 0, &blocknr);
789         /* If that failed, give up */
790         if (err) {
791                 printk(KERN_ERR "%s: Cannnot locate journal superblock\n",
792                        __FUNCTION__);
793                 kfree(journal);
794                 return NULL;
795         }
796
797         bh = __getblk(journal->j_dev, blocknr, journal->j_blocksize);
798         J_ASSERT(bh != NULL);
799         journal->j_sb_buffer = bh;
800         journal->j_superblock = (journal_superblock_t *)bh->b_data;
801
802         return journal;
803 }
804
805 /*
806  * If the journal init or create aborts, we need to mark the journal
807  * superblock as being NULL to prevent the journal destroy from writing
808  * back a bogus superblock.
809  */
810 static void journal_fail_superblock (journal_t *journal)
811 {
812         struct buffer_head *bh = journal->j_sb_buffer;
813         brelse(bh);
814         journal->j_sb_buffer = NULL;
815 }
816
817 /*
818  * Given a journal_t structure, initialise the various fields for
819  * startup of a new journaling session.  We use this both when creating
820  * a journal, and after recovering an old journal to reset it for
821  * subsequent use.
822  */
823
824 static int journal_reset(journal_t *journal)
825 {
826         journal_superblock_t *sb = journal->j_superblock;
827         unsigned long long first, last;
828
829         first = be32_to_cpu(sb->s_first);
830         last = be32_to_cpu(sb->s_maxlen);
831
832         journal->j_first = first;
833         journal->j_last = last;
834
835         journal->j_head = first;
836         journal->j_tail = first;
837         journal->j_free = last - first;
838
839         journal->j_tail_sequence = journal->j_transaction_sequence;
840         journal->j_commit_sequence = journal->j_transaction_sequence - 1;
841         journal->j_commit_request = journal->j_commit_sequence;
842
843         journal->j_max_transaction_buffers = journal->j_maxlen / 4;
844
845         /* Add the dynamic fields and write it to disk. */
846         jbd2_journal_update_superblock(journal, 1);
847         return jbd2_journal_start_thread(journal);
848 }
849
850 /**
851  * int jbd2_journal_create() - Initialise the new journal file
852  * @journal: Journal to create. This structure must have been initialised
853  *
854  * Given a journal_t structure which tells us which disk blocks we can
855  * use, create a new journal superblock and initialise all of the
856  * journal fields from scratch.
857  **/
858 int jbd2_journal_create(journal_t *journal)
859 {
860         unsigned long long blocknr;
861         struct buffer_head *bh;
862         journal_superblock_t *sb;
863         int i, err;
864
865         if (journal->j_maxlen < JBD2_MIN_JOURNAL_BLOCKS) {
866                 printk (KERN_ERR "Journal length (%d blocks) too short.\n",
867                         journal->j_maxlen);
868                 journal_fail_superblock(journal);
869                 return -EINVAL;
870         }
871
872         if (journal->j_inode == NULL) {
873                 /*
874                  * We don't know what block to start at!
875                  */
876                 printk(KERN_EMERG
877                        "%s: creation of journal on external device!\n",
878                        __FUNCTION__);
879                 BUG();
880         }
881
882         /* Zero out the entire journal on disk.  We cannot afford to
883            have any blocks on disk beginning with JBD2_MAGIC_NUMBER. */
884         jbd_debug(1, "JBD: Zeroing out journal blocks...\n");
885         for (i = 0; i < journal->j_maxlen; i++) {
886                 err = jbd2_journal_bmap(journal, i, &blocknr);
887                 if (err)
888                         return err;
889                 bh = __getblk(journal->j_dev, blocknr, journal->j_blocksize);
890                 lock_buffer(bh);
891                 memset (bh->b_data, 0, journal->j_blocksize);
892                 BUFFER_TRACE(bh, "marking dirty");
893                 mark_buffer_dirty(bh);
894                 BUFFER_TRACE(bh, "marking uptodate");
895                 set_buffer_uptodate(bh);
896                 unlock_buffer(bh);
897                 __brelse(bh);
898         }
899
900         sync_blockdev(journal->j_dev);
901         jbd_debug(1, "JBD: journal cleared.\n");
902
903         /* OK, fill in the initial static fields in the new superblock */
904         sb = journal->j_superblock;
905
906         sb->s_header.h_magic     = cpu_to_be32(JBD2_MAGIC_NUMBER);
907         sb->s_header.h_blocktype = cpu_to_be32(JBD2_SUPERBLOCK_V2);
908
909         sb->s_blocksize = cpu_to_be32(journal->j_blocksize);
910         sb->s_maxlen    = cpu_to_be32(journal->j_maxlen);
911         sb->s_first     = cpu_to_be32(1);
912
913         journal->j_transaction_sequence = 1;
914
915         journal->j_flags &= ~JBD2_ABORT;
916         journal->j_format_version = 2;
917
918         return journal_reset(journal);
919 }
920
921 /**
922  * void jbd2_journal_update_superblock() - Update journal sb on disk.
923  * @journal: The journal to update.
924  * @wait: Set to '0' if you don't want to wait for IO completion.
925  *
926  * Update a journal's dynamic superblock fields and write it to disk,
927  * optionally waiting for the IO to complete.
928  */
929 void jbd2_journal_update_superblock(journal_t *journal, int wait)
930 {
931         journal_superblock_t *sb = journal->j_superblock;
932         struct buffer_head *bh = journal->j_sb_buffer;
933
934         /*
935          * As a special case, if the on-disk copy is already marked as needing
936          * no recovery (s_start == 0) and there are no outstanding transactions
937          * in the filesystem, then we can safely defer the superblock update
938          * until the next commit by setting JBD2_FLUSHED.  This avoids
939          * attempting a write to a potential-readonly device.
940          */
941         if (sb->s_start == 0 && journal->j_tail_sequence ==
942                                 journal->j_transaction_sequence) {
943                 jbd_debug(1,"JBD: Skipping superblock update on recovered sb "
944                         "(start %ld, seq %d, errno %d)\n",
945                         journal->j_tail, journal->j_tail_sequence,
946                         journal->j_errno);
947                 goto out;
948         }
949
950         spin_lock(&journal->j_state_lock);
951         jbd_debug(1,"JBD: updating superblock (start %ld, seq %d, errno %d)\n",
952                   journal->j_tail, journal->j_tail_sequence, journal->j_errno);
953
954         sb->s_sequence = cpu_to_be32(journal->j_tail_sequence);
955         sb->s_start    = cpu_to_be32(journal->j_tail);
956         sb->s_errno    = cpu_to_be32(journal->j_errno);
957         spin_unlock(&journal->j_state_lock);
958
959         BUFFER_TRACE(bh, "marking dirty");
960         mark_buffer_dirty(bh);
961         if (wait)
962                 sync_dirty_buffer(bh);
963         else
964                 ll_rw_block(SWRITE, 1, &bh);
965
966 out:
967         /* If we have just flushed the log (by marking s_start==0), then
968          * any future commit will have to be careful to update the
969          * superblock again to re-record the true start of the log. */
970
971         spin_lock(&journal->j_state_lock);
972         if (sb->s_start)
973                 journal->j_flags &= ~JBD2_FLUSHED;
974         else
975                 journal->j_flags |= JBD2_FLUSHED;
976         spin_unlock(&journal->j_state_lock);
977 }
978
979 /*
980  * Read the superblock for a given journal, performing initial
981  * validation of the format.
982  */
983
984 static int journal_get_superblock(journal_t *journal)
985 {
986         struct buffer_head *bh;
987         journal_superblock_t *sb;
988         int err = -EIO;
989
990         bh = journal->j_sb_buffer;
991
992         J_ASSERT(bh != NULL);
993         if (!buffer_uptodate(bh)) {
994                 ll_rw_block(READ, 1, &bh);
995                 wait_on_buffer(bh);
996                 if (!buffer_uptodate(bh)) {
997                         printk (KERN_ERR
998                                 "JBD: IO error reading journal superblock\n");
999                         goto out;
1000                 }
1001         }
1002
1003         sb = journal->j_superblock;
1004
1005         err = -EINVAL;
1006
1007         if (sb->s_header.h_magic != cpu_to_be32(JBD2_MAGIC_NUMBER) ||
1008             sb->s_blocksize != cpu_to_be32(journal->j_blocksize)) {
1009                 printk(KERN_WARNING "JBD: no valid journal superblock found\n");
1010                 goto out;
1011         }
1012
1013         switch(be32_to_cpu(sb->s_header.h_blocktype)) {
1014         case JBD2_SUPERBLOCK_V1:
1015                 journal->j_format_version = 1;
1016                 break;
1017         case JBD2_SUPERBLOCK_V2:
1018                 journal->j_format_version = 2;
1019                 break;
1020         default:
1021                 printk(KERN_WARNING "JBD: unrecognised superblock format ID\n");
1022                 goto out;
1023         }
1024
1025         if (be32_to_cpu(sb->s_maxlen) < journal->j_maxlen)
1026                 journal->j_maxlen = be32_to_cpu(sb->s_maxlen);
1027         else if (be32_to_cpu(sb->s_maxlen) > journal->j_maxlen) {
1028                 printk (KERN_WARNING "JBD: journal file too short\n");
1029                 goto out;
1030         }
1031
1032         return 0;
1033
1034 out:
1035         journal_fail_superblock(journal);
1036         return err;
1037 }
1038
1039 /*
1040  * Load the on-disk journal superblock and read the key fields into the
1041  * journal_t.
1042  */
1043
1044 static int load_superblock(journal_t *journal)
1045 {
1046         int err;
1047         journal_superblock_t *sb;
1048
1049         err = journal_get_superblock(journal);
1050         if (err)
1051                 return err;
1052
1053         sb = journal->j_superblock;
1054
1055         journal->j_tail_sequence = be32_to_cpu(sb->s_sequence);
1056         journal->j_tail = be32_to_cpu(sb->s_start);
1057         journal->j_first = be32_to_cpu(sb->s_first);
1058         journal->j_last = be32_to_cpu(sb->s_maxlen);
1059         journal->j_errno = be32_to_cpu(sb->s_errno);
1060
1061         return 0;
1062 }
1063
1064
1065 /**
1066  * int jbd2_journal_load() - Read journal from disk.
1067  * @journal: Journal to act on.
1068  *
1069  * Given a journal_t structure which tells us which disk blocks contain
1070  * a journal, read the journal from disk to initialise the in-memory
1071  * structures.
1072  */
1073 int jbd2_journal_load(journal_t *journal)
1074 {
1075         int err;
1076         journal_superblock_t *sb;
1077
1078         err = load_superblock(journal);
1079         if (err)
1080                 return err;
1081
1082         sb = journal->j_superblock;
1083         /* If this is a V2 superblock, then we have to check the
1084          * features flags on it. */
1085
1086         if (journal->j_format_version >= 2) {
1087                 if ((sb->s_feature_ro_compat &
1088                      ~cpu_to_be32(JBD2_KNOWN_ROCOMPAT_FEATURES)) ||
1089                     (sb->s_feature_incompat &
1090                      ~cpu_to_be32(JBD2_KNOWN_INCOMPAT_FEATURES))) {
1091                         printk (KERN_WARNING
1092                                 "JBD: Unrecognised features on journal\n");
1093                         return -EINVAL;
1094                 }
1095         }
1096
1097         /* Let the recovery code check whether it needs to recover any
1098          * data from the journal. */
1099         if (jbd2_journal_recover(journal))
1100                 goto recovery_error;
1101
1102         /* OK, we've finished with the dynamic journal bits:
1103          * reinitialise the dynamic contents of the superblock in memory
1104          * and reset them on disk. */
1105         if (journal_reset(journal))
1106                 goto recovery_error;
1107
1108         journal->j_flags &= ~JBD2_ABORT;
1109         journal->j_flags |= JBD2_LOADED;
1110         return 0;
1111
1112 recovery_error:
1113         printk (KERN_WARNING "JBD: recovery failed\n");
1114         return -EIO;
1115 }
1116
1117 /**
1118  * void jbd2_journal_destroy() - Release a journal_t structure.
1119  * @journal: Journal to act on.
1120  *
1121  * Release a journal_t structure once it is no longer in use by the
1122  * journaled object.
1123  */
1124 void jbd2_journal_destroy(journal_t *journal)
1125 {
1126         /* Wait for the commit thread to wake up and die. */
1127         journal_kill_thread(journal);
1128
1129         /* Force a final log commit */
1130         if (journal->j_running_transaction)
1131                 jbd2_journal_commit_transaction(journal);
1132
1133         /* Force any old transactions to disk */
1134
1135         /* Totally anal locking here... */
1136         spin_lock(&journal->j_list_lock);
1137         while (journal->j_checkpoint_transactions != NULL) {
1138                 spin_unlock(&journal->j_list_lock);
1139                 jbd2_log_do_checkpoint(journal);
1140                 spin_lock(&journal->j_list_lock);
1141         }
1142
1143         J_ASSERT(journal->j_running_transaction == NULL);
1144         J_ASSERT(journal->j_committing_transaction == NULL);
1145         J_ASSERT(journal->j_checkpoint_transactions == NULL);
1146         spin_unlock(&journal->j_list_lock);
1147
1148         /* We can now mark the journal as empty. */
1149         journal->j_tail = 0;
1150         journal->j_tail_sequence = ++journal->j_transaction_sequence;
1151         if (journal->j_sb_buffer) {
1152                 jbd2_journal_update_superblock(journal, 1);
1153                 brelse(journal->j_sb_buffer);
1154         }
1155
1156         if (journal->j_inode)
1157                 iput(journal->j_inode);
1158         if (journal->j_revoke)
1159                 jbd2_journal_destroy_revoke(journal);
1160         kfree(journal->j_wbuf);
1161         kfree(journal);
1162 }
1163
1164
1165 /**
1166  *int jbd2_journal_check_used_features () - Check if features specified are used.
1167  * @journal: Journal to check.
1168  * @compat: bitmask of compatible features
1169  * @ro: bitmask of features that force read-only mount
1170  * @incompat: bitmask of incompatible features
1171  *
1172  * Check whether the journal uses all of a given set of
1173  * features.  Return true (non-zero) if it does.
1174  **/
1175
1176 int jbd2_journal_check_used_features (journal_t *journal, unsigned long compat,
1177                                  unsigned long ro, unsigned long incompat)
1178 {
1179         journal_superblock_t *sb;
1180
1181         if (!compat && !ro && !incompat)
1182                 return 1;
1183         if (journal->j_format_version == 1)
1184                 return 0;
1185
1186         sb = journal->j_superblock;
1187
1188         if (((be32_to_cpu(sb->s_feature_compat) & compat) == compat) &&
1189             ((be32_to_cpu(sb->s_feature_ro_compat) & ro) == ro) &&
1190             ((be32_to_cpu(sb->s_feature_incompat) & incompat) == incompat))
1191                 return 1;
1192
1193         return 0;
1194 }
1195
1196 /**
1197  * int jbd2_journal_check_available_features() - Check feature set in journalling layer
1198  * @journal: Journal to check.
1199  * @compat: bitmask of compatible features
1200  * @ro: bitmask of features that force read-only mount
1201  * @incompat: bitmask of incompatible features
1202  *
1203  * Check whether the journaling code supports the use of
1204  * all of a given set of features on this journal.  Return true
1205  * (non-zero) if it can. */
1206
1207 int jbd2_journal_check_available_features (journal_t *journal, unsigned long compat,
1208                                       unsigned long ro, unsigned long incompat)
1209 {
1210         journal_superblock_t *sb;
1211
1212         if (!compat && !ro && !incompat)
1213                 return 1;
1214
1215         sb = journal->j_superblock;
1216
1217         /* We can support any known requested features iff the
1218          * superblock is in version 2.  Otherwise we fail to support any
1219          * extended sb features. */
1220
1221         if (journal->j_format_version != 2)
1222                 return 0;
1223
1224         if ((compat   & JBD2_KNOWN_COMPAT_FEATURES) == compat &&
1225             (ro       & JBD2_KNOWN_ROCOMPAT_FEATURES) == ro &&
1226             (incompat & JBD2_KNOWN_INCOMPAT_FEATURES) == incompat)
1227                 return 1;
1228
1229         return 0;
1230 }
1231
1232 /**
1233  * int jbd2_journal_set_features () - Mark a given journal feature in the superblock
1234  * @journal: Journal to act on.
1235  * @compat: bitmask of compatible features
1236  * @ro: bitmask of features that force read-only mount
1237  * @incompat: bitmask of incompatible features
1238  *
1239  * Mark a given journal feature as present on the
1240  * superblock.  Returns true if the requested features could be set.
1241  *
1242  */
1243
1244 int jbd2_journal_set_features (journal_t *journal, unsigned long compat,
1245                           unsigned long ro, unsigned long incompat)
1246 {
1247         journal_superblock_t *sb;
1248
1249         if (jbd2_journal_check_used_features(journal, compat, ro, incompat))
1250                 return 1;
1251
1252         if (!jbd2_journal_check_available_features(journal, compat, ro, incompat))
1253                 return 0;
1254
1255         jbd_debug(1, "Setting new features 0x%lx/0x%lx/0x%lx\n",
1256                   compat, ro, incompat);
1257
1258         sb = journal->j_superblock;
1259
1260         sb->s_feature_compat    |= cpu_to_be32(compat);
1261         sb->s_feature_ro_compat |= cpu_to_be32(ro);
1262         sb->s_feature_incompat  |= cpu_to_be32(incompat);
1263
1264         return 1;
1265 }
1266
1267
1268 /**
1269  * int jbd2_journal_update_format () - Update on-disk journal structure.
1270  * @journal: Journal to act on.
1271  *
1272  * Given an initialised but unloaded journal struct, poke about in the
1273  * on-disk structure to update it to the most recent supported version.
1274  */
1275 int jbd2_journal_update_format (journal_t *journal)
1276 {
1277         journal_superblock_t *sb;
1278         int err;
1279
1280         err = journal_get_superblock(journal);
1281         if (err)
1282                 return err;
1283
1284         sb = journal->j_superblock;
1285
1286         switch (be32_to_cpu(sb->s_header.h_blocktype)) {
1287         case JBD2_SUPERBLOCK_V2:
1288                 return 0;
1289         case JBD2_SUPERBLOCK_V1:
1290                 return journal_convert_superblock_v1(journal, sb);
1291         default:
1292                 break;
1293         }
1294         return -EINVAL;
1295 }
1296
1297 static int journal_convert_superblock_v1(journal_t *journal,
1298                                          journal_superblock_t *sb)
1299 {
1300         int offset, blocksize;
1301         struct buffer_head *bh;
1302
1303         printk(KERN_WARNING
1304                 "JBD: Converting superblock from version 1 to 2.\n");
1305
1306         /* Pre-initialise new fields to zero */
1307         offset = ((char *) &(sb->s_feature_compat)) - ((char *) sb);
1308         blocksize = be32_to_cpu(sb->s_blocksize);
1309         memset(&sb->s_feature_compat, 0, blocksize-offset);
1310
1311         sb->s_nr_users = cpu_to_be32(1);
1312         sb->s_header.h_blocktype = cpu_to_be32(JBD2_SUPERBLOCK_V2);
1313         journal->j_format_version = 2;
1314
1315         bh = journal->j_sb_buffer;
1316         BUFFER_TRACE(bh, "marking dirty");
1317         mark_buffer_dirty(bh);
1318         sync_dirty_buffer(bh);
1319         return 0;
1320 }
1321
1322
1323 /**
1324  * int jbd2_journal_flush () - Flush journal
1325  * @journal: Journal to act on.
1326  *
1327  * Flush all data for a given journal to disk and empty the journal.
1328  * Filesystems can use this when remounting readonly to ensure that
1329  * recovery does not need to happen on remount.
1330  */
1331
1332 int jbd2_journal_flush(journal_t *journal)
1333 {
1334         int err = 0;
1335         transaction_t *transaction = NULL;
1336         unsigned long old_tail;
1337
1338         spin_lock(&journal->j_state_lock);
1339
1340         /* Force everything buffered to the log... */
1341         if (journal->j_running_transaction) {
1342                 transaction = journal->j_running_transaction;
1343                 __jbd2_log_start_commit(journal, transaction->t_tid);
1344         } else if (journal->j_committing_transaction)
1345                 transaction = journal->j_committing_transaction;
1346
1347         /* Wait for the log commit to complete... */
1348         if (transaction) {
1349                 tid_t tid = transaction->t_tid;
1350
1351                 spin_unlock(&journal->j_state_lock);
1352                 jbd2_log_wait_commit(journal, tid);
1353         } else {
1354                 spin_unlock(&journal->j_state_lock);
1355         }
1356
1357         /* ...and flush everything in the log out to disk. */
1358         spin_lock(&journal->j_list_lock);
1359         while (!err && journal->j_checkpoint_transactions != NULL) {
1360                 spin_unlock(&journal->j_list_lock);
1361                 err = jbd2_log_do_checkpoint(journal);
1362                 spin_lock(&journal->j_list_lock);
1363         }
1364         spin_unlock(&journal->j_list_lock);
1365         jbd2_cleanup_journal_tail(journal);
1366
1367         /* Finally, mark the journal as really needing no recovery.
1368          * This sets s_start==0 in the underlying superblock, which is
1369          * the magic code for a fully-recovered superblock.  Any future
1370          * commits of data to the journal will restore the current
1371          * s_start value. */
1372         spin_lock(&journal->j_state_lock);
1373         old_tail = journal->j_tail;
1374         journal->j_tail = 0;
1375         spin_unlock(&journal->j_state_lock);
1376         jbd2_journal_update_superblock(journal, 1);
1377         spin_lock(&journal->j_state_lock);
1378         journal->j_tail = old_tail;
1379
1380         J_ASSERT(!journal->j_running_transaction);
1381         J_ASSERT(!journal->j_committing_transaction);
1382         J_ASSERT(!journal->j_checkpoint_transactions);
1383         J_ASSERT(journal->j_head == journal->j_tail);
1384         J_ASSERT(journal->j_tail_sequence == journal->j_transaction_sequence);
1385         spin_unlock(&journal->j_state_lock);
1386         return err;
1387 }
1388
1389 /**
1390  * int jbd2_journal_wipe() - Wipe journal contents
1391  * @journal: Journal to act on.
1392  * @write: flag (see below)
1393  *
1394  * Wipe out all of the contents of a journal, safely.  This will produce
1395  * a warning if the journal contains any valid recovery information.
1396  * Must be called between journal_init_*() and jbd2_journal_load().
1397  *
1398  * If 'write' is non-zero, then we wipe out the journal on disk; otherwise
1399  * we merely suppress recovery.
1400  */
1401
1402 int jbd2_journal_wipe(journal_t *journal, int write)
1403 {
1404         journal_superblock_t *sb;
1405         int err = 0;
1406
1407         J_ASSERT (!(journal->j_flags & JBD2_LOADED));
1408
1409         err = load_superblock(journal);
1410         if (err)
1411                 return err;
1412
1413         sb = journal->j_superblock;
1414
1415         if (!journal->j_tail)
1416                 goto no_recovery;
1417
1418         printk (KERN_WARNING "JBD: %s recovery information on journal\n",
1419                 write ? "Clearing" : "Ignoring");
1420
1421         err = jbd2_journal_skip_recovery(journal);
1422         if (write)
1423                 jbd2_journal_update_superblock(journal, 1);
1424
1425  no_recovery:
1426         return err;
1427 }
1428
1429 /*
1430  * journal_dev_name: format a character string to describe on what
1431  * device this journal is present.
1432  */
1433
1434 static const char *journal_dev_name(journal_t *journal, char *buffer)
1435 {
1436         struct block_device *bdev;
1437
1438         if (journal->j_inode)
1439                 bdev = journal->j_inode->i_sb->s_bdev;
1440         else
1441                 bdev = journal->j_dev;
1442
1443         return bdevname(bdev, buffer);
1444 }
1445
1446 /*
1447  * Journal abort has very specific semantics, which we describe
1448  * for journal abort.
1449  *
1450  * Two internal function, which provide abort to te jbd layer
1451  * itself are here.
1452  */
1453
1454 /*
1455  * Quick version for internal journal use (doesn't lock the journal).
1456  * Aborts hard --- we mark the abort as occurred, but do _nothing_ else,
1457  * and don't attempt to make any other journal updates.
1458  */
1459 void __jbd2_journal_abort_hard(journal_t *journal)
1460 {
1461         transaction_t *transaction;
1462         char b[BDEVNAME_SIZE];
1463
1464         if (journal->j_flags & JBD2_ABORT)
1465                 return;
1466
1467         printk(KERN_ERR "Aborting journal on device %s.\n",
1468                 journal_dev_name(journal, b));
1469
1470         spin_lock(&journal->j_state_lock);
1471         journal->j_flags |= JBD2_ABORT;
1472         transaction = journal->j_running_transaction;
1473         if (transaction)
1474                 __jbd2_log_start_commit(journal, transaction->t_tid);
1475         spin_unlock(&journal->j_state_lock);
1476 }
1477
1478 /* Soft abort: record the abort error status in the journal superblock,
1479  * but don't do any other IO. */
1480 static void __journal_abort_soft (journal_t *journal, int errno)
1481 {
1482         if (journal->j_flags & JBD2_ABORT)
1483                 return;
1484
1485         if (!journal->j_errno)
1486                 journal->j_errno = errno;
1487
1488         __jbd2_journal_abort_hard(journal);
1489
1490         if (errno)
1491                 jbd2_journal_update_superblock(journal, 1);
1492 }
1493
1494 /**
1495  * void jbd2_journal_abort () - Shutdown the journal immediately.
1496  * @journal: the journal to shutdown.
1497  * @errno:   an error number to record in the journal indicating
1498  *           the reason for the shutdown.
1499  *
1500  * Perform a complete, immediate shutdown of the ENTIRE
1501  * journal (not of a single transaction).  This operation cannot be
1502  * undone without closing and reopening the journal.
1503  *
1504  * The jbd2_journal_abort function is intended to support higher level error
1505  * recovery mechanisms such as the ext2/ext3 remount-readonly error
1506  * mode.
1507  *
1508  * Journal abort has very specific semantics.  Any existing dirty,
1509  * unjournaled buffers in the main filesystem will still be written to
1510  * disk by bdflush, but the journaling mechanism will be suspended
1511  * immediately and no further transaction commits will be honoured.
1512  *
1513  * Any dirty, journaled buffers will be written back to disk without
1514  * hitting the journal.  Atomicity cannot be guaranteed on an aborted
1515  * filesystem, but we _do_ attempt to leave as much data as possible
1516  * behind for fsck to use for cleanup.
1517  *
1518  * Any attempt to get a new transaction handle on a journal which is in
1519  * ABORT state will just result in an -EROFS error return.  A
1520  * jbd2_journal_stop on an existing handle will return -EIO if we have
1521  * entered abort state during the update.
1522  *
1523  * Recursive transactions are not disturbed by journal abort until the
1524  * final jbd2_journal_stop, which will receive the -EIO error.
1525  *
1526  * Finally, the jbd2_journal_abort call allows the caller to supply an errno
1527  * which will be recorded (if possible) in the journal superblock.  This
1528  * allows a client to record failure conditions in the middle of a
1529  * transaction without having to complete the transaction to record the
1530  * failure to disk.  ext3_error, for example, now uses this
1531  * functionality.
1532  *
1533  * Errors which originate from within the journaling layer will NOT
1534  * supply an errno; a null errno implies that absolutely no further
1535  * writes are done to the journal (unless there are any already in
1536  * progress).
1537  *
1538  */
1539
1540 void jbd2_journal_abort(journal_t *journal, int errno)
1541 {
1542         __journal_abort_soft(journal, errno);
1543 }
1544
1545 /**
1546  * int jbd2_journal_errno () - returns the journal's error state.
1547  * @journal: journal to examine.
1548  *
1549  * This is the errno numbet set with jbd2_journal_abort(), the last
1550  * time the journal was mounted - if the journal was stopped
1551  * without calling abort this will be 0.
1552  *
1553  * If the journal has been aborted on this mount time -EROFS will
1554  * be returned.
1555  */
1556 int jbd2_journal_errno(journal_t *journal)
1557 {
1558         int err;
1559
1560         spin_lock(&journal->j_state_lock);
1561         if (journal->j_flags & JBD2_ABORT)
1562                 err = -EROFS;
1563         else
1564                 err = journal->j_errno;
1565         spin_unlock(&journal->j_state_lock);
1566         return err;
1567 }
1568
1569 /**
1570  * int jbd2_journal_clear_err () - clears the journal's error state
1571  * @journal: journal to act on.
1572  *
1573  * An error must be cleared or Acked to take a FS out of readonly
1574  * mode.
1575  */
1576 int jbd2_journal_clear_err(journal_t *journal)
1577 {
1578         int err = 0;
1579
1580         spin_lock(&journal->j_state_lock);
1581         if (journal->j_flags & JBD2_ABORT)
1582                 err = -EROFS;
1583         else
1584                 journal->j_errno = 0;
1585         spin_unlock(&journal->j_state_lock);
1586         return err;
1587 }
1588
1589 /**
1590  * void jbd2_journal_ack_err() - Ack journal err.
1591  * @journal: journal to act on.
1592  *
1593  * An error must be cleared or Acked to take a FS out of readonly
1594  * mode.
1595  */
1596 void jbd2_journal_ack_err(journal_t *journal)
1597 {
1598         spin_lock(&journal->j_state_lock);
1599         if (journal->j_errno)
1600                 journal->j_flags |= JBD2_ACK_ERR;
1601         spin_unlock(&journal->j_state_lock);
1602 }
1603
1604 int jbd2_journal_blocks_per_page(struct inode *inode)
1605 {
1606         return 1 << (PAGE_CACHE_SHIFT - inode->i_sb->s_blocksize_bits);
1607 }
1608
1609 /*
1610  * helper functions to deal with 32 or 64bit block numbers.
1611  */
1612 size_t journal_tag_bytes(journal_t *journal)
1613 {
1614         if (JBD2_HAS_INCOMPAT_FEATURE(journal, JBD2_FEATURE_INCOMPAT_64BIT))
1615                 return JBD2_TAG_SIZE64;
1616         else
1617                 return JBD2_TAG_SIZE32;
1618 }
1619
1620 /*
1621  * Journal_head storage management
1622  */
1623 static struct kmem_cache *jbd2_journal_head_cache;
1624 #ifdef CONFIG_JBD2_DEBUG
1625 static atomic_t nr_journal_heads = ATOMIC_INIT(0);
1626 #endif
1627
1628 static int journal_init_jbd2_journal_head_cache(void)
1629 {
1630         int retval;
1631
1632         J_ASSERT(jbd2_journal_head_cache == 0);
1633         jbd2_journal_head_cache = kmem_cache_create("jbd2_journal_head",
1634                                 sizeof(struct journal_head),
1635                                 0,              /* offset */
1636                                 0,              /* flags */
1637                                 NULL);          /* ctor */
1638         retval = 0;
1639         if (jbd2_journal_head_cache == 0) {
1640                 retval = -ENOMEM;
1641                 printk(KERN_EMERG "JBD: no memory for journal_head cache\n");
1642         }
1643         return retval;
1644 }
1645
1646 static void jbd2_journal_destroy_jbd2_journal_head_cache(void)
1647 {
1648         J_ASSERT(jbd2_journal_head_cache != NULL);
1649         kmem_cache_destroy(jbd2_journal_head_cache);
1650         jbd2_journal_head_cache = NULL;
1651 }
1652
1653 /*
1654  * journal_head splicing and dicing
1655  */
1656 static struct journal_head *journal_alloc_journal_head(void)
1657 {
1658         struct journal_head *ret;
1659         static unsigned long last_warning;
1660
1661 #ifdef CONFIG_JBD2_DEBUG
1662         atomic_inc(&nr_journal_heads);
1663 #endif
1664         ret = kmem_cache_alloc(jbd2_journal_head_cache, GFP_NOFS);
1665         if (ret == 0) {
1666                 jbd_debug(1, "out of memory for journal_head\n");
1667                 if (time_after(jiffies, last_warning + 5*HZ)) {
1668                         printk(KERN_NOTICE "ENOMEM in %s, retrying.\n",
1669                                __FUNCTION__);
1670                         last_warning = jiffies;
1671                 }
1672                 while (ret == 0) {
1673                         yield();
1674                         ret = kmem_cache_alloc(jbd2_journal_head_cache, GFP_NOFS);
1675                 }
1676         }
1677         return ret;
1678 }
1679
1680 static void journal_free_journal_head(struct journal_head *jh)
1681 {
1682 #ifdef CONFIG_JBD2_DEBUG
1683         atomic_dec(&nr_journal_heads);
1684         memset(jh, JBD2_POISON_FREE, sizeof(*jh));
1685 #endif
1686         kmem_cache_free(jbd2_journal_head_cache, jh);
1687 }
1688
1689 /*
1690  * A journal_head is attached to a buffer_head whenever JBD has an
1691  * interest in the buffer.
1692  *
1693  * Whenever a buffer has an attached journal_head, its ->b_state:BH_JBD bit
1694  * is set.  This bit is tested in core kernel code where we need to take
1695  * JBD-specific actions.  Testing the zeroness of ->b_private is not reliable
1696  * there.
1697  *
1698  * When a buffer has its BH_JBD bit set, its ->b_count is elevated by one.
1699  *
1700  * When a buffer has its BH_JBD bit set it is immune from being released by
1701  * core kernel code, mainly via ->b_count.
1702  *
1703  * A journal_head may be detached from its buffer_head when the journal_head's
1704  * b_transaction, b_cp_transaction and b_next_transaction pointers are NULL.
1705  * Various places in JBD call jbd2_journal_remove_journal_head() to indicate that the
1706  * journal_head can be dropped if needed.
1707  *
1708  * Various places in the kernel want to attach a journal_head to a buffer_head
1709  * _before_ attaching the journal_head to a transaction.  To protect the
1710  * journal_head in this situation, jbd2_journal_add_journal_head elevates the
1711  * journal_head's b_jcount refcount by one.  The caller must call
1712  * jbd2_journal_put_journal_head() to undo this.
1713  *
1714  * So the typical usage would be:
1715  *
1716  *      (Attach a journal_head if needed.  Increments b_jcount)
1717  *      struct journal_head *jh = jbd2_journal_add_journal_head(bh);
1718  *      ...
1719  *      jh->b_transaction = xxx;
1720  *      jbd2_journal_put_journal_head(jh);
1721  *
1722  * Now, the journal_head's b_jcount is zero, but it is safe from being released
1723  * because it has a non-zero b_transaction.
1724  */
1725
1726 /*
1727  * Give a buffer_head a journal_head.
1728  *
1729  * Doesn't need the journal lock.
1730  * May sleep.
1731  */
1732 struct journal_head *jbd2_journal_add_journal_head(struct buffer_head *bh)
1733 {
1734         struct journal_head *jh;
1735         struct journal_head *new_jh = NULL;
1736
1737 repeat:
1738         if (!buffer_jbd(bh)) {
1739                 new_jh = journal_alloc_journal_head();
1740                 memset(new_jh, 0, sizeof(*new_jh));
1741         }
1742
1743         jbd_lock_bh_journal_head(bh);
1744         if (buffer_jbd(bh)) {
1745                 jh = bh2jh(bh);
1746         } else {
1747                 J_ASSERT_BH(bh,
1748                         (atomic_read(&bh->b_count) > 0) ||
1749                         (bh->b_page && bh->b_page->mapping));
1750
1751                 if (!new_jh) {
1752                         jbd_unlock_bh_journal_head(bh);
1753                         goto repeat;
1754                 }
1755
1756                 jh = new_jh;
1757                 new_jh = NULL;          /* We consumed it */
1758                 set_buffer_jbd(bh);
1759                 bh->b_private = jh;
1760                 jh->b_bh = bh;
1761                 get_bh(bh);
1762                 BUFFER_TRACE(bh, "added journal_head");
1763         }
1764         jh->b_jcount++;
1765         jbd_unlock_bh_journal_head(bh);
1766         if (new_jh)
1767                 journal_free_journal_head(new_jh);
1768         return bh->b_private;
1769 }
1770
1771 /*
1772  * Grab a ref against this buffer_head's journal_head.  If it ended up not
1773  * having a journal_head, return NULL
1774  */
1775 struct journal_head *jbd2_journal_grab_journal_head(struct buffer_head *bh)
1776 {
1777         struct journal_head *jh = NULL;
1778
1779         jbd_lock_bh_journal_head(bh);
1780         if (buffer_jbd(bh)) {
1781                 jh = bh2jh(bh);
1782                 jh->b_jcount++;
1783         }
1784         jbd_unlock_bh_journal_head(bh);
1785         return jh;
1786 }
1787
1788 static void __journal_remove_journal_head(struct buffer_head *bh)
1789 {
1790         struct journal_head *jh = bh2jh(bh);
1791
1792         J_ASSERT_JH(jh, jh->b_jcount >= 0);
1793
1794         get_bh(bh);
1795         if (jh->b_jcount == 0) {
1796                 if (jh->b_transaction == NULL &&
1797                                 jh->b_next_transaction == NULL &&
1798                                 jh->b_cp_transaction == NULL) {
1799                         J_ASSERT_JH(jh, jh->b_jlist == BJ_None);
1800                         J_ASSERT_BH(bh, buffer_jbd(bh));
1801                         J_ASSERT_BH(bh, jh2bh(jh) == bh);
1802                         BUFFER_TRACE(bh, "remove journal_head");
1803                         if (jh->b_frozen_data) {
1804                                 printk(KERN_WARNING "%s: freeing "
1805                                                 "b_frozen_data\n",
1806                                                 __FUNCTION__);
1807                                 jbd2_free(jh->b_frozen_data, bh->b_size);
1808                         }
1809                         if (jh->b_committed_data) {
1810                                 printk(KERN_WARNING "%s: freeing "
1811                                                 "b_committed_data\n",
1812                                                 __FUNCTION__);
1813                                 jbd2_free(jh->b_committed_data, bh->b_size);
1814                         }
1815                         bh->b_private = NULL;
1816                         jh->b_bh = NULL;        /* debug, really */
1817                         clear_buffer_jbd(bh);
1818                         __brelse(bh);
1819                         journal_free_journal_head(jh);
1820                 } else {
1821                         BUFFER_TRACE(bh, "journal_head was locked");
1822                 }
1823         }
1824 }
1825
1826 /*
1827  * jbd2_journal_remove_journal_head(): if the buffer isn't attached to a transaction
1828  * and has a zero b_jcount then remove and release its journal_head.   If we did
1829  * see that the buffer is not used by any transaction we also "logically"
1830  * decrement ->b_count.
1831  *
1832  * We in fact take an additional increment on ->b_count as a convenience,
1833  * because the caller usually wants to do additional things with the bh
1834  * after calling here.
1835  * The caller of jbd2_journal_remove_journal_head() *must* run __brelse(bh) at some
1836  * time.  Once the caller has run __brelse(), the buffer is eligible for
1837  * reaping by try_to_free_buffers().
1838  */
1839 void jbd2_journal_remove_journal_head(struct buffer_head *bh)
1840 {
1841         jbd_lock_bh_journal_head(bh);
1842         __journal_remove_journal_head(bh);
1843         jbd_unlock_bh_journal_head(bh);
1844 }
1845
1846 /*
1847  * Drop a reference on the passed journal_head.  If it fell to zero then try to
1848  * release the journal_head from the buffer_head.
1849  */
1850 void jbd2_journal_put_journal_head(struct journal_head *jh)
1851 {
1852         struct buffer_head *bh = jh2bh(jh);
1853
1854         jbd_lock_bh_journal_head(bh);
1855         J_ASSERT_JH(jh, jh->b_jcount > 0);
1856         --jh->b_jcount;
1857         if (!jh->b_jcount && !jh->b_transaction) {
1858                 __journal_remove_journal_head(bh);
1859                 __brelse(bh);
1860         }
1861         jbd_unlock_bh_journal_head(bh);
1862 }
1863
1864 /*
1865  * debugfs tunables
1866  */
1867 #ifdef CONFIG_JBD2_DEBUG
1868 u8 jbd2_journal_enable_debug __read_mostly;
1869 EXPORT_SYMBOL(jbd2_journal_enable_debug);
1870
1871 #define JBD2_DEBUG_NAME "jbd2-debug"
1872
1873 static struct dentry *jbd2_debugfs_dir;
1874 static struct dentry *jbd2_debug;
1875
1876 static void __init jbd2_create_debugfs_entry(void)
1877 {
1878         jbd2_debugfs_dir = debugfs_create_dir("jbd2", NULL);
1879         if (jbd2_debugfs_dir)
1880                 jbd2_debug = debugfs_create_u8(JBD2_DEBUG_NAME, S_IRUGO,
1881                                                jbd2_debugfs_dir,
1882                                                &jbd2_journal_enable_debug);
1883 }
1884
1885 static void __exit jbd2_remove_debugfs_entry(void)
1886 {
1887         debugfs_remove(jbd2_debug);
1888         debugfs_remove(jbd2_debugfs_dir);
1889 }
1890
1891 #else
1892
1893 static void __init jbd2_create_debugfs_entry(void)
1894 {
1895 }
1896
1897 static void __exit jbd2_remove_debugfs_entry(void)
1898 {
1899 }
1900
1901 #endif
1902
1903 struct kmem_cache *jbd2_handle_cache;
1904
1905 static int __init journal_init_handle_cache(void)
1906 {
1907         jbd2_handle_cache = kmem_cache_create("jbd2_journal_handle",
1908                                 sizeof(handle_t),
1909                                 0,              /* offset */
1910                                 0,              /* flags */
1911                                 NULL);          /* ctor */
1912         if (jbd2_handle_cache == NULL) {
1913                 printk(KERN_EMERG "JBD: failed to create handle cache\n");
1914                 return -ENOMEM;
1915         }
1916         return 0;
1917 }
1918
1919 static void jbd2_journal_destroy_handle_cache(void)
1920 {
1921         if (jbd2_handle_cache)
1922                 kmem_cache_destroy(jbd2_handle_cache);
1923 }
1924
1925 /*
1926  * Module startup and shutdown
1927  */
1928
1929 static int __init journal_init_caches(void)
1930 {
1931         int ret;
1932
1933         ret = jbd2_journal_init_revoke_caches();
1934         if (ret == 0)
1935                 ret = journal_init_jbd2_journal_head_cache();
1936         if (ret == 0)
1937                 ret = journal_init_handle_cache();
1938         return ret;
1939 }
1940
1941 static void jbd2_journal_destroy_caches(void)
1942 {
1943         jbd2_journal_destroy_revoke_caches();
1944         jbd2_journal_destroy_jbd2_journal_head_cache();
1945         jbd2_journal_destroy_handle_cache();
1946 }
1947
1948 static int __init journal_init(void)
1949 {
1950         int ret;
1951
1952         BUILD_BUG_ON(sizeof(struct journal_superblock_s) != 1024);
1953
1954         ret = journal_init_caches();
1955         if (ret != 0)
1956                 jbd2_journal_destroy_caches();
1957         jbd2_create_debugfs_entry();
1958         return ret;
1959 }
1960
1961 static void __exit journal_exit(void)
1962 {
1963 #ifdef CONFIG_JBD2_DEBUG
1964         int n = atomic_read(&nr_journal_heads);
1965         if (n)
1966                 printk(KERN_EMERG "JBD: leaked %d journal_heads!\n", n);
1967 #endif
1968         jbd2_remove_debugfs_entry();
1969         jbd2_journal_destroy_caches();
1970 }
1971
1972 MODULE_LICENSE("GPL");
1973 module_init(journal_init);
1974 module_exit(journal_exit);
1975