aboutsummaryrefslogtreecommitdiffstats
path: root/fs/ocfs2/journal.c
diff options
context:
space:
mode:
Diffstat (limited to 'fs/ocfs2/journal.c')
-rw-r--r--fs/ocfs2/journal.c1652
1 files changed, 1652 insertions, 0 deletions
diff --git a/fs/ocfs2/journal.c b/fs/ocfs2/journal.c
new file mode 100644
index 000000000000..04428042e5e5
--- /dev/null
+++ b/fs/ocfs2/journal.c
@@ -0,0 +1,1652 @@
1/* -*- mode: c; c-basic-offset: 8; -*-
2 * vim: noexpandtab sw=8 ts=8 sts=0:
3 *
4 * journal.c
5 *
6 * Defines functions of journalling api
7 *
8 * Copyright (C) 2003, 2004 Oracle. All rights reserved.
9 *
10 * This program is free software; you can redistribute it and/or
11 * modify it under the terms of the GNU General Public
12 * License as published by the Free Software Foundation; either
13 * version 2 of the License, or (at your option) any later version.
14 *
15 * This program is distributed in the hope that it will be useful,
16 * but WITHOUT ANY WARRANTY; without even the implied warranty of
17 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
18 * General Public License for more details.
19 *
20 * You should have received a copy of the GNU General Public
21 * License along with this program; if not, write to the
22 * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
23 * Boston, MA 021110-1307, USA.
24 */
25
26#include <linux/fs.h>
27#include <linux/types.h>
28#include <linux/slab.h>
29#include <linux/highmem.h>
30#include <linux/kthread.h>
31
32#define MLOG_MASK_PREFIX ML_JOURNAL
33#include <cluster/masklog.h>
34
35#include "ocfs2.h"
36
37#include "alloc.h"
38#include "dlmglue.h"
39#include "extent_map.h"
40#include "heartbeat.h"
41#include "inode.h"
42#include "journal.h"
43#include "localalloc.h"
44#include "namei.h"
45#include "slot_map.h"
46#include "super.h"
47#include "vote.h"
48#include "sysfile.h"
49
50#include "buffer_head_io.h"
51
52spinlock_t trans_inc_lock = SPIN_LOCK_UNLOCKED;
53
54static int ocfs2_force_read_journal(struct inode *inode);
55static int ocfs2_recover_node(struct ocfs2_super *osb,
56 int node_num);
57static int __ocfs2_recovery_thread(void *arg);
58static int ocfs2_commit_cache(struct ocfs2_super *osb);
59static int ocfs2_wait_on_mount(struct ocfs2_super *osb);
60static void ocfs2_handle_cleanup_locks(struct ocfs2_journal *journal,
61 struct ocfs2_journal_handle *handle);
62static void ocfs2_commit_unstarted_handle(struct ocfs2_journal_handle *handle);
63static int ocfs2_journal_toggle_dirty(struct ocfs2_super *osb,
64 int dirty);
65static int ocfs2_trylock_journal(struct ocfs2_super *osb,
66 int slot_num);
67static int ocfs2_recover_orphans(struct ocfs2_super *osb,
68 int slot);
69static int ocfs2_commit_thread(void *arg);
70
71static int ocfs2_commit_cache(struct ocfs2_super *osb)
72{
73 int status = 0;
74 unsigned int flushed;
75 unsigned long old_id;
76 struct ocfs2_journal *journal = NULL;
77
78 mlog_entry_void();
79
80 journal = osb->journal;
81
82 /* Flush all pending commits and checkpoint the journal. */
83 down_write(&journal->j_trans_barrier);
84
85 if (atomic_read(&journal->j_num_trans) == 0) {
86 up_write(&journal->j_trans_barrier);
87 mlog(0, "No transactions for me to flush!\n");
88 goto finally;
89 }
90
91 journal_lock_updates(journal->j_journal);
92 status = journal_flush(journal->j_journal);
93 journal_unlock_updates(journal->j_journal);
94 if (status < 0) {
95 up_write(&journal->j_trans_barrier);
96 mlog_errno(status);
97 goto finally;
98 }
99
100 old_id = ocfs2_inc_trans_id(journal);
101
102 flushed = atomic_read(&journal->j_num_trans);
103 atomic_set(&journal->j_num_trans, 0);
104 up_write(&journal->j_trans_barrier);
105
106 mlog(0, "commit_thread: flushed transaction %lu (%u handles)\n",
107 journal->j_trans_id, flushed);
108
109 ocfs2_kick_vote_thread(osb);
110 wake_up(&journal->j_checkpointed);
111finally:
112 mlog_exit(status);
113 return status;
114}
115
116struct ocfs2_journal_handle *ocfs2_alloc_handle(struct ocfs2_super *osb)
117{
118 struct ocfs2_journal_handle *retval = NULL;
119
120 retval = kcalloc(1, sizeof(*retval), GFP_KERNEL);
121 if (!retval) {
122 mlog(ML_ERROR, "Failed to allocate memory for journal "
123 "handle!\n");
124 return NULL;
125 }
126
127 retval->max_buffs = 0;
128 retval->num_locks = 0;
129 retval->k_handle = NULL;
130
131 INIT_LIST_HEAD(&retval->locks);
132 INIT_LIST_HEAD(&retval->inode_list);
133 retval->journal = osb->journal;
134
135 return retval;
136}
137
138/* pass it NULL and it will allocate a new handle object for you. If
139 * you pass it a handle however, it may still return error, in which
140 * case it has free'd the passed handle for you. */
141struct ocfs2_journal_handle *ocfs2_start_trans(struct ocfs2_super *osb,
142 struct ocfs2_journal_handle *handle,
143 int max_buffs)
144{
145 int ret;
146 journal_t *journal = osb->journal->j_journal;
147
148 mlog_entry("(max_buffs = %d)\n", max_buffs);
149
150 if (!osb || !osb->journal->j_journal)
151 BUG();
152
153 if (ocfs2_is_hard_readonly(osb)) {
154 ret = -EROFS;
155 goto done_free;
156 }
157
158 BUG_ON(osb->journal->j_state == OCFS2_JOURNAL_FREE);
159 BUG_ON(max_buffs <= 0);
160
161 /* JBD might support this, but our journalling code doesn't yet. */
162 if (journal_current_handle()) {
163 mlog(ML_ERROR, "Recursive transaction attempted!\n");
164 BUG();
165 }
166
167 if (!handle)
168 handle = ocfs2_alloc_handle(osb);
169 if (!handle) {
170 ret = -ENOMEM;
171 mlog(ML_ERROR, "Failed to allocate memory for journal "
172 "handle!\n");
173 goto done_free;
174 }
175
176 handle->max_buffs = max_buffs;
177
178 down_read(&osb->journal->j_trans_barrier);
179
180 /* actually start the transaction now */
181 handle->k_handle = journal_start(journal, max_buffs);
182 if (IS_ERR(handle->k_handle)) {
183 up_read(&osb->journal->j_trans_barrier);
184
185 ret = PTR_ERR(handle->k_handle);
186 handle->k_handle = NULL;
187 mlog_errno(ret);
188
189 if (is_journal_aborted(journal)) {
190 ocfs2_abort(osb->sb, "Detected aborted journal");
191 ret = -EROFS;
192 }
193 goto done_free;
194 }
195
196 atomic_inc(&(osb->journal->j_num_trans));
197 handle->flags |= OCFS2_HANDLE_STARTED;
198
199 mlog_exit_ptr(handle);
200 return handle;
201
202done_free:
203 if (handle)
204 ocfs2_commit_unstarted_handle(handle); /* will kfree handle */
205
206 mlog_exit(ret);
207 return ERR_PTR(ret);
208}
209
210void ocfs2_handle_add_inode(struct ocfs2_journal_handle *handle,
211 struct inode *inode)
212{
213 BUG_ON(!handle);
214 BUG_ON(!inode);
215
216 atomic_inc(&inode->i_count);
217
218 /* we're obviously changing it... */
219 down(&inode->i_sem);
220
221 /* sanity check */
222 BUG_ON(OCFS2_I(inode)->ip_handle);
223 BUG_ON(!list_empty(&OCFS2_I(inode)->ip_handle_list));
224
225 OCFS2_I(inode)->ip_handle = handle;
226 list_del(&(OCFS2_I(inode)->ip_handle_list));
227 list_add_tail(&(OCFS2_I(inode)->ip_handle_list), &(handle->inode_list));
228}
229
230static void ocfs2_handle_unlock_inodes(struct ocfs2_journal_handle *handle)
231{
232 struct list_head *p, *n;
233 struct inode *inode;
234 struct ocfs2_inode_info *oi;
235
236 list_for_each_safe(p, n, &handle->inode_list) {
237 oi = list_entry(p, struct ocfs2_inode_info,
238 ip_handle_list);
239 inode = &oi->vfs_inode;
240
241 OCFS2_I(inode)->ip_handle = NULL;
242 list_del_init(&OCFS2_I(inode)->ip_handle_list);
243
244 up(&inode->i_sem);
245 iput(inode);
246 }
247}
248
249/* This is trivial so we do it out of the main commit
250 * paths. Beware, it can be called from start_trans too! */
251static void ocfs2_commit_unstarted_handle(struct ocfs2_journal_handle *handle)
252{
253 mlog_entry_void();
254
255 BUG_ON(handle->flags & OCFS2_HANDLE_STARTED);
256
257 ocfs2_handle_unlock_inodes(handle);
258 /* You are allowed to add journal locks before the transaction
259 * has started. */
260 ocfs2_handle_cleanup_locks(handle->journal, handle);
261
262 kfree(handle);
263
264 mlog_exit_void();
265}
266
267void ocfs2_commit_trans(struct ocfs2_journal_handle *handle)
268{
269 handle_t *jbd_handle;
270 int retval;
271 struct ocfs2_journal *journal = handle->journal;
272
273 mlog_entry_void();
274
275 BUG_ON(!handle);
276
277 if (!(handle->flags & OCFS2_HANDLE_STARTED)) {
278 ocfs2_commit_unstarted_handle(handle);
279 mlog_exit_void();
280 return;
281 }
282
283 /* release inode semaphores we took during this transaction */
284 ocfs2_handle_unlock_inodes(handle);
285
286 /* ocfs2_extend_trans may have had to call journal_restart
287 * which will always commit the transaction, but may return
288 * error for any number of reasons. If this is the case, we
289 * clear k_handle as it's not valid any more. */
290 if (handle->k_handle) {
291 jbd_handle = handle->k_handle;
292
293 if (handle->flags & OCFS2_HANDLE_SYNC)
294 jbd_handle->h_sync = 1;
295 else
296 jbd_handle->h_sync = 0;
297
298 /* actually stop the transaction. if we've set h_sync,
299 * it'll have been committed when we return */
300 retval = journal_stop(jbd_handle);
301 if (retval < 0) {
302 mlog_errno(retval);
303 mlog(ML_ERROR, "Could not commit transaction\n");
304 BUG();
305 }
306
307 handle->k_handle = NULL; /* it's been free'd in journal_stop */
308 }
309
310 ocfs2_handle_cleanup_locks(journal, handle);
311
312 up_read(&journal->j_trans_barrier);
313
314 kfree(handle);
315 mlog_exit_void();
316}
317
318/*
319 * 'nblocks' is what you want to add to the current
320 * transaction. extend_trans will either extend the current handle by
321 * nblocks, or commit it and start a new one with nblocks credits.
322 *
323 * WARNING: This will not release any semaphores or disk locks taken
324 * during the transaction, so make sure they were taken *before*
325 * start_trans or we'll have ordering deadlocks.
326 *
327 * WARNING2: Note that we do *not* drop j_trans_barrier here. This is
328 * good because transaction ids haven't yet been recorded on the
329 * cluster locks associated with this handle.
330 */
331int ocfs2_extend_trans(struct ocfs2_journal_handle *handle,
332 int nblocks)
333{
334 int status;
335
336 BUG_ON(!handle);
337 BUG_ON(!(handle->flags & OCFS2_HANDLE_STARTED));
338 BUG_ON(!nblocks);
339
340 mlog_entry_void();
341
342 mlog(0, "Trying to extend transaction by %d blocks\n", nblocks);
343
344 status = journal_extend(handle->k_handle, nblocks);
345 if (status < 0) {
346 mlog_errno(status);
347 goto bail;
348 }
349
350 if (status > 0) {
351 mlog(0, "journal_extend failed, trying journal_restart\n");
352 status = journal_restart(handle->k_handle, nblocks);
353 if (status < 0) {
354 handle->k_handle = NULL;
355 mlog_errno(status);
356 goto bail;
357 }
358 handle->max_buffs = nblocks;
359 } else
360 handle->max_buffs += nblocks;
361
362 status = 0;
363bail:
364
365 mlog_exit(status);
366 return status;
367}
368
369int ocfs2_journal_access(struct ocfs2_journal_handle *handle,
370 struct inode *inode,
371 struct buffer_head *bh,
372 int type)
373{
374 int status;
375
376 BUG_ON(!inode);
377 BUG_ON(!handle);
378 BUG_ON(!bh);
379 BUG_ON(!(handle->flags & OCFS2_HANDLE_STARTED));
380
381 mlog_entry("bh->b_blocknr=%llu, type=%d (\"%s\"), bh->b_size = %hu\n",
382 (unsigned long long)bh->b_blocknr, type,
383 (type == OCFS2_JOURNAL_ACCESS_CREATE) ?
384 "OCFS2_JOURNAL_ACCESS_CREATE" :
385 "OCFS2_JOURNAL_ACCESS_WRITE",
386 bh->b_size);
387
388 /* we can safely remove this assertion after testing. */
389 if (!buffer_uptodate(bh)) {
390 mlog(ML_ERROR, "giving me a buffer that's not uptodate!\n");
391 mlog(ML_ERROR, "b_blocknr=%llu\n",
392 (unsigned long long)bh->b_blocknr);
393 BUG();
394 }
395
396 /* Set the current transaction information on the inode so
397 * that the locking code knows whether it can drop it's locks
398 * on this inode or not. We're protected from the commit
399 * thread updating the current transaction id until
400 * ocfs2_commit_trans() because ocfs2_start_trans() took
401 * j_trans_barrier for us. */
402 ocfs2_set_inode_lock_trans(OCFS2_SB(inode->i_sb)->journal, inode);
403
404 down(&OCFS2_I(inode)->ip_io_sem);
405 switch (type) {
406 case OCFS2_JOURNAL_ACCESS_CREATE:
407 case OCFS2_JOURNAL_ACCESS_WRITE:
408 status = journal_get_write_access(handle->k_handle, bh);
409 break;
410
411 case OCFS2_JOURNAL_ACCESS_UNDO:
412 status = journal_get_undo_access(handle->k_handle, bh);
413 break;
414
415 default:
416 status = -EINVAL;
417 mlog(ML_ERROR, "Uknown access type!\n");
418 }
419 up(&OCFS2_I(inode)->ip_io_sem);
420
421 if (status < 0)
422 mlog(ML_ERROR, "Error %d getting %d access to buffer!\n",
423 status, type);
424
425 mlog_exit(status);
426 return status;
427}
428
429int ocfs2_journal_dirty(struct ocfs2_journal_handle *handle,
430 struct buffer_head *bh)
431{
432 int status;
433
434 BUG_ON(!(handle->flags & OCFS2_HANDLE_STARTED));
435
436 mlog_entry("(bh->b_blocknr=%llu)\n",
437 (unsigned long long)bh->b_blocknr);
438
439 status = journal_dirty_metadata(handle->k_handle, bh);
440 if (status < 0)
441 mlog(ML_ERROR, "Could not dirty metadata buffer. "
442 "(bh->b_blocknr=%llu)\n",
443 (unsigned long long)bh->b_blocknr);
444
445 mlog_exit(status);
446 return status;
447}
448
449int ocfs2_journal_dirty_data(handle_t *handle,
450 struct buffer_head *bh)
451{
452 int err = journal_dirty_data(handle, bh);
453 if (err)
454 mlog_errno(err);
455 /* TODO: When we can handle it, abort the handle and go RO on
456 * error here. */
457
458 return err;
459}
460
461/* We always assume you're adding a metadata lock at level 'ex' */
462int ocfs2_handle_add_lock(struct ocfs2_journal_handle *handle,
463 struct inode *inode)
464{
465 int status;
466 struct ocfs2_journal_lock *lock;
467
468 BUG_ON(!inode);
469
470 lock = kmem_cache_alloc(ocfs2_lock_cache, GFP_NOFS);
471 if (!lock) {
472 status = -ENOMEM;
473 mlog_errno(-ENOMEM);
474 goto bail;
475 }
476
477 if (!igrab(inode))
478 BUG();
479 lock->jl_inode = inode;
480
481 list_add_tail(&(lock->jl_lock_list), &(handle->locks));
482 handle->num_locks++;
483
484 status = 0;
485bail:
486 mlog_exit(status);
487 return status;
488}
489
490static void ocfs2_handle_cleanup_locks(struct ocfs2_journal *journal,
491 struct ocfs2_journal_handle *handle)
492{
493 struct list_head *p, *n;
494 struct ocfs2_journal_lock *lock;
495 struct inode *inode;
496
497 list_for_each_safe(p, n, &(handle->locks)) {
498 lock = list_entry(p, struct ocfs2_journal_lock,
499 jl_lock_list);
500 list_del(&lock->jl_lock_list);
501 handle->num_locks--;
502
503 inode = lock->jl_inode;
504 ocfs2_meta_unlock(inode, 1);
505 if (atomic_read(&inode->i_count) == 1)
506 mlog(ML_ERROR,
507 "Inode %"MLFu64", I'm doing a last iput for!",
508 OCFS2_I(inode)->ip_blkno);
509 iput(inode);
510 kmem_cache_free(ocfs2_lock_cache, lock);
511 }
512}
513
514#define OCFS2_DEFAULT_COMMIT_INTERVAL (HZ * 5)
515
516void ocfs2_set_journal_params(struct ocfs2_super *osb)
517{
518 journal_t *journal = osb->journal->j_journal;
519
520 spin_lock(&journal->j_state_lock);
521 journal->j_commit_interval = OCFS2_DEFAULT_COMMIT_INTERVAL;
522 if (osb->s_mount_opt & OCFS2_MOUNT_BARRIER)
523 journal->j_flags |= JFS_BARRIER;
524 else
525 journal->j_flags &= ~JFS_BARRIER;
526 spin_unlock(&journal->j_state_lock);
527}
528
529int ocfs2_journal_init(struct ocfs2_journal *journal, int *dirty)
530{
531 int status = -1;
532 struct inode *inode = NULL; /* the journal inode */
533 journal_t *j_journal = NULL;
534 struct ocfs2_dinode *di = NULL;
535 struct buffer_head *bh = NULL;
536 struct ocfs2_super *osb;
537 int meta_lock = 0;
538
539 mlog_entry_void();
540
541 BUG_ON(!journal);
542
543 osb = journal->j_osb;
544
545 /* already have the inode for our journal */
546 inode = ocfs2_get_system_file_inode(osb, JOURNAL_SYSTEM_INODE,
547 osb->slot_num);
548 if (inode == NULL) {
549 status = -EACCES;
550 mlog_errno(status);
551 goto done;
552 }
553 if (is_bad_inode(inode)) {
554 mlog(ML_ERROR, "access error (bad inode)\n");
555 iput(inode);
556 inode = NULL;
557 status = -EACCES;
558 goto done;
559 }
560
561 SET_INODE_JOURNAL(inode);
562 OCFS2_I(inode)->ip_open_count++;
563
564 status = ocfs2_meta_lock(inode, NULL, &bh, 1);
565 if (status < 0) {
566 if (status != -ERESTARTSYS)
567 mlog(ML_ERROR, "Could not get lock on journal!\n");
568 goto done;
569 }
570
571 meta_lock = 1;
572 di = (struct ocfs2_dinode *)bh->b_data;
573
574 if (inode->i_size < OCFS2_MIN_JOURNAL_SIZE) {
575 mlog(ML_ERROR, "Journal file size (%lld) is too small!\n",
576 inode->i_size);
577 status = -EINVAL;
578 goto done;
579 }
580
581 mlog(0, "inode->i_size = %lld\n", inode->i_size);
582 mlog(0, "inode->i_blocks = %lu\n", inode->i_blocks);
583 mlog(0, "inode->ip_clusters = %u\n", OCFS2_I(inode)->ip_clusters);
584
585 /* call the kernels journal init function now */
586 j_journal = journal_init_inode(inode);
587 if (j_journal == NULL) {
588 mlog(ML_ERROR, "Linux journal layer error\n");
589 status = -EINVAL;
590 goto done;
591 }
592
593 mlog(0, "Returned from journal_init_inode\n");
594 mlog(0, "j_journal->j_maxlen = %u\n", j_journal->j_maxlen);
595
596 *dirty = (le32_to_cpu(di->id1.journal1.ij_flags) &
597 OCFS2_JOURNAL_DIRTY_FL);
598
599 journal->j_journal = j_journal;
600 journal->j_inode = inode;
601 journal->j_bh = bh;
602
603 ocfs2_set_journal_params(osb);
604
605 journal->j_state = OCFS2_JOURNAL_LOADED;
606
607 status = 0;
608done:
609 if (status < 0) {
610 if (meta_lock)
611 ocfs2_meta_unlock(inode, 1);
612 if (bh != NULL)
613 brelse(bh);
614 if (inode) {
615 OCFS2_I(inode)->ip_open_count--;
616 iput(inode);
617 }
618 }
619
620 mlog_exit(status);
621 return status;
622}
623
624static int ocfs2_journal_toggle_dirty(struct ocfs2_super *osb,
625 int dirty)
626{
627 int status;
628 unsigned int flags;
629 struct ocfs2_journal *journal = osb->journal;
630 struct buffer_head *bh = journal->j_bh;
631 struct ocfs2_dinode *fe;
632
633 mlog_entry_void();
634
635 fe = (struct ocfs2_dinode *)bh->b_data;
636 if (!OCFS2_IS_VALID_DINODE(fe)) {
637 /* This is called from startup/shutdown which will
638 * handle the errors in a specific manner, so no need
639 * to call ocfs2_error() here. */
640 mlog(ML_ERROR, "Journal dinode %"MLFu64" has invalid "
641 "signature: %.*s", fe->i_blkno, 7, fe->i_signature);
642 status = -EIO;
643 goto out;
644 }
645
646 flags = le32_to_cpu(fe->id1.journal1.ij_flags);
647 if (dirty)
648 flags |= OCFS2_JOURNAL_DIRTY_FL;
649 else
650 flags &= ~OCFS2_JOURNAL_DIRTY_FL;
651 fe->id1.journal1.ij_flags = cpu_to_le32(flags);
652
653 status = ocfs2_write_block(osb, bh, journal->j_inode);
654 if (status < 0)
655 mlog_errno(status);
656
657out:
658 mlog_exit(status);
659 return status;
660}
661
662/*
663 * If the journal has been kmalloc'd it needs to be freed after this
664 * call.
665 */
666void ocfs2_journal_shutdown(struct ocfs2_super *osb)
667{
668 struct ocfs2_journal *journal = NULL;
669 int status = 0;
670 struct inode *inode = NULL;
671 int num_running_trans = 0;
672
673 mlog_entry_void();
674
675 if (!osb)
676 BUG();
677
678 journal = osb->journal;
679 if (!journal)
680 goto done;
681
682 inode = journal->j_inode;
683
684 if (journal->j_state != OCFS2_JOURNAL_LOADED)
685 goto done;
686
687 /* need to inc inode use count as journal_destroy will iput. */
688 if (!igrab(inode))
689 BUG();
690
691 num_running_trans = atomic_read(&(osb->journal->j_num_trans));
692 if (num_running_trans > 0)
693 mlog(0, "Shutting down journal: must wait on %d "
694 "running transactions!\n",
695 num_running_trans);
696
697 /* Do a commit_cache here. It will flush our journal, *and*
698 * release any locks that are still held.
699 * set the SHUTDOWN flag and release the trans lock.
700 * the commit thread will take the trans lock for us below. */
701 journal->j_state = OCFS2_JOURNAL_IN_SHUTDOWN;
702
703 /* The OCFS2_JOURNAL_IN_SHUTDOWN will signal to commit_cache to not
704 * drop the trans_lock (which we want to hold until we
705 * completely destroy the journal. */
706 if (osb->commit_task) {
707 /* Wait for the commit thread */
708 mlog(0, "Waiting for ocfs2commit to exit....\n");
709 kthread_stop(osb->commit_task);
710 osb->commit_task = NULL;
711 }
712
713 BUG_ON(atomic_read(&(osb->journal->j_num_trans)) != 0);
714
715 status = ocfs2_journal_toggle_dirty(osb, 0);
716 if (status < 0)
717 mlog_errno(status);
718
719 /* Shutdown the kernel journal system */
720 journal_destroy(journal->j_journal);
721
722 OCFS2_I(inode)->ip_open_count--;
723
724 /* unlock our journal */
725 ocfs2_meta_unlock(inode, 1);
726
727 brelse(journal->j_bh);
728 journal->j_bh = NULL;
729
730 journal->j_state = OCFS2_JOURNAL_FREE;
731
732// up_write(&journal->j_trans_barrier);
733done:
734 if (inode)
735 iput(inode);
736 mlog_exit_void();
737}
738
739static void ocfs2_clear_journal_error(struct super_block *sb,
740 journal_t *journal,
741 int slot)
742{
743 int olderr;
744
745 olderr = journal_errno(journal);
746 if (olderr) {
747 mlog(ML_ERROR, "File system error %d recorded in "
748 "journal %u.\n", olderr, slot);
749 mlog(ML_ERROR, "File system on device %s needs checking.\n",
750 sb->s_id);
751
752 journal_ack_err(journal);
753 journal_clear_err(journal);
754 }
755}
756
757int ocfs2_journal_load(struct ocfs2_journal *journal)
758{
759 int status = 0;
760 struct ocfs2_super *osb;
761
762 mlog_entry_void();
763
764 if (!journal)
765 BUG();
766
767 osb = journal->j_osb;
768
769 status = journal_load(journal->j_journal);
770 if (status < 0) {
771 mlog(ML_ERROR, "Failed to load journal!\n");
772 goto done;
773 }
774
775 ocfs2_clear_journal_error(osb->sb, journal->j_journal, osb->slot_num);
776
777 status = ocfs2_journal_toggle_dirty(osb, 1);
778 if (status < 0) {
779 mlog_errno(status);
780 goto done;
781 }
782
783 /* Launch the commit thread */
784 osb->commit_task = kthread_run(ocfs2_commit_thread, osb, "ocfs2cmt-%d",
785 osb->osb_id);
786 if (IS_ERR(osb->commit_task)) {
787 status = PTR_ERR(osb->commit_task);
788 osb->commit_task = NULL;
789 mlog(ML_ERROR, "unable to launch ocfs2commit thread, error=%d",
790 status);
791 goto done;
792 }
793
794done:
795 mlog_exit(status);
796 return status;
797}
798
799
800/* 'full' flag tells us whether we clear out all blocks or if we just
801 * mark the journal clean */
802int ocfs2_journal_wipe(struct ocfs2_journal *journal, int full)
803{
804 int status;
805
806 mlog_entry_void();
807
808 if (!journal)
809 BUG();
810
811 status = journal_wipe(journal->j_journal, full);
812 if (status < 0) {
813 mlog_errno(status);
814 goto bail;
815 }
816
817 status = ocfs2_journal_toggle_dirty(journal->j_osb, 0);
818 if (status < 0)
819 mlog_errno(status);
820
821bail:
822 mlog_exit(status);
823 return status;
824}
825
826/*
827 * JBD Might read a cached version of another nodes journal file. We
828 * don't want this as this file changes often and we get no
829 * notification on those changes. The only way to be sure that we've
830 * got the most up to date version of those blocks then is to force
831 * read them off disk. Just searching through the buffer cache won't
832 * work as there may be pages backing this file which are still marked
833 * up to date. We know things can't change on this file underneath us
834 * as we have the lock by now :)
835 */
836static int ocfs2_force_read_journal(struct inode *inode)
837{
838 int status = 0;
839 int i, p_blocks;
840 u64 v_blkno, p_blkno;
841#define CONCURRENT_JOURNAL_FILL 32
842 struct buffer_head *bhs[CONCURRENT_JOURNAL_FILL];
843
844 mlog_entry_void();
845
846 BUG_ON(inode->i_blocks !=
847 ocfs2_align_bytes_to_sectors(i_size_read(inode)));
848
849 memset(bhs, 0, sizeof(struct buffer_head *) * CONCURRENT_JOURNAL_FILL);
850
851 mlog(0, "Force reading %lu blocks\n",
852 (inode->i_blocks >> (inode->i_sb->s_blocksize_bits - 9)));
853
854 v_blkno = 0;
855 while (v_blkno <
856 (inode->i_blocks >> (inode->i_sb->s_blocksize_bits - 9))) {
857
858 status = ocfs2_extent_map_get_blocks(inode, v_blkno,
859 1, &p_blkno,
860 &p_blocks);
861 if (status < 0) {
862 mlog_errno(status);
863 goto bail;
864 }
865
866 if (p_blocks > CONCURRENT_JOURNAL_FILL)
867 p_blocks = CONCURRENT_JOURNAL_FILL;
868
869 status = ocfs2_read_blocks(OCFS2_SB(inode->i_sb),
870 p_blkno, p_blocks, bhs, 0,
871 inode);
872 if (status < 0) {
873 mlog_errno(status);
874 goto bail;
875 }
876
877 for(i = 0; i < p_blocks; i++) {
878 brelse(bhs[i]);
879 bhs[i] = NULL;
880 }
881
882 v_blkno += p_blocks;
883 }
884
885bail:
886 for(i = 0; i < CONCURRENT_JOURNAL_FILL; i++)
887 if (bhs[i])
888 brelse(bhs[i]);
889 mlog_exit(status);
890 return status;
891}
892
893struct ocfs2_la_recovery_item {
894 struct list_head lri_list;
895 int lri_slot;
896 struct ocfs2_dinode *lri_la_dinode;
897 struct ocfs2_dinode *lri_tl_dinode;
898};
899
900/* Does the second half of the recovery process. By this point, the
901 * node is marked clean and can actually be considered recovered,
902 * hence it's no longer in the recovery map, but there's still some
903 * cleanup we can do which shouldn't happen within the recovery thread
904 * as locking in that context becomes very difficult if we are to take
905 * recovering nodes into account.
906 *
907 * NOTE: This function can and will sleep on recovery of other nodes
908 * during cluster locking, just like any other ocfs2 process.
909 */
910void ocfs2_complete_recovery(void *data)
911{
912 int ret;
913 struct ocfs2_super *osb = data;
914 struct ocfs2_journal *journal = osb->journal;
915 struct ocfs2_dinode *la_dinode, *tl_dinode;
916 struct ocfs2_la_recovery_item *item;
917 struct list_head *p, *n;
918 LIST_HEAD(tmp_la_list);
919
920 mlog_entry_void();
921
922 mlog(0, "completing recovery from keventd\n");
923
924 spin_lock(&journal->j_lock);
925 list_splice_init(&journal->j_la_cleanups, &tmp_la_list);
926 spin_unlock(&journal->j_lock);
927
928 list_for_each_safe(p, n, &tmp_la_list) {
929 item = list_entry(p, struct ocfs2_la_recovery_item, lri_list);
930 list_del_init(&item->lri_list);
931
932 mlog(0, "Complete recovery for slot %d\n", item->lri_slot);
933
934 la_dinode = item->lri_la_dinode;
935 if (la_dinode) {
936 mlog(0, "Clean up local alloc %"MLFu64"\n",
937 la_dinode->i_blkno);
938
939 ret = ocfs2_complete_local_alloc_recovery(osb,
940 la_dinode);
941 if (ret < 0)
942 mlog_errno(ret);
943
944 kfree(la_dinode);
945 }
946
947 tl_dinode = item->lri_tl_dinode;
948 if (tl_dinode) {
949 mlog(0, "Clean up truncate log %"MLFu64"\n",
950 tl_dinode->i_blkno);
951
952 ret = ocfs2_complete_truncate_log_recovery(osb,
953 tl_dinode);
954 if (ret < 0)
955 mlog_errno(ret);
956
957 kfree(tl_dinode);
958 }
959
960 ret = ocfs2_recover_orphans(osb, item->lri_slot);
961 if (ret < 0)
962 mlog_errno(ret);
963
964 kfree(item);
965 }
966
967 mlog(0, "Recovery completion\n");
968 mlog_exit_void();
969}
970
971/* NOTE: This function always eats your references to la_dinode and
972 * tl_dinode, either manually on error, or by passing them to
973 * ocfs2_complete_recovery */
974static void ocfs2_queue_recovery_completion(struct ocfs2_journal *journal,
975 int slot_num,
976 struct ocfs2_dinode *la_dinode,
977 struct ocfs2_dinode *tl_dinode)
978{
979 struct ocfs2_la_recovery_item *item;
980
981 item = kmalloc(sizeof(struct ocfs2_la_recovery_item), GFP_KERNEL);
982 if (!item) {
983 /* Though we wish to avoid it, we are in fact safe in
984 * skipping local alloc cleanup as fsck.ocfs2 is more
985 * than capable of reclaiming unused space. */
986 if (la_dinode)
987 kfree(la_dinode);
988
989 if (tl_dinode)
990 kfree(tl_dinode);
991
992 mlog_errno(-ENOMEM);
993 return;
994 }
995
996 INIT_LIST_HEAD(&item->lri_list);
997 item->lri_la_dinode = la_dinode;
998 item->lri_slot = slot_num;
999 item->lri_tl_dinode = tl_dinode;
1000
1001 spin_lock(&journal->j_lock);
1002 list_add_tail(&item->lri_list, &journal->j_la_cleanups);
1003 queue_work(ocfs2_wq, &journal->j_recovery_work);
1004 spin_unlock(&journal->j_lock);
1005}
1006
1007/* Called by the mount code to queue recovery the last part of
1008 * recovery for it's own slot. */
1009void ocfs2_complete_mount_recovery(struct ocfs2_super *osb)
1010{
1011 struct ocfs2_journal *journal = osb->journal;
1012
1013 if (osb->dirty) {
1014 /* No need to queue up our truncate_log as regular
1015 * cleanup will catch that. */
1016 ocfs2_queue_recovery_completion(journal,
1017 osb->slot_num,
1018 osb->local_alloc_copy,
1019 NULL);
1020 ocfs2_schedule_truncate_log_flush(osb, 0);
1021
1022 osb->local_alloc_copy = NULL;
1023 osb->dirty = 0;
1024 }
1025}
1026
1027static int __ocfs2_recovery_thread(void *arg)
1028{
1029 int status, node_num;
1030 struct ocfs2_super *osb = arg;
1031
1032 mlog_entry_void();
1033
1034 status = ocfs2_wait_on_mount(osb);
1035 if (status < 0) {
1036 goto bail;
1037 }
1038
1039restart:
1040 status = ocfs2_super_lock(osb, 1);
1041 if (status < 0) {
1042 mlog_errno(status);
1043 goto bail;
1044 }
1045
1046 while(!ocfs2_node_map_is_empty(osb, &osb->recovery_map)) {
1047 node_num = ocfs2_node_map_first_set_bit(osb,
1048 &osb->recovery_map);
1049 if (node_num == O2NM_INVALID_NODE_NUM) {
1050 mlog(0, "Out of nodes to recover.\n");
1051 break;
1052 }
1053
1054 status = ocfs2_recover_node(osb, node_num);
1055 if (status < 0) {
1056 mlog(ML_ERROR,
1057 "Error %d recovering node %d on device (%u,%u)!\n",
1058 status, node_num,
1059 MAJOR(osb->sb->s_dev), MINOR(osb->sb->s_dev));
1060 mlog(ML_ERROR, "Volume requires unmount.\n");
1061 continue;
1062 }
1063
1064 ocfs2_recovery_map_clear(osb, node_num);
1065 }
1066 ocfs2_super_unlock(osb, 1);
1067
1068 /* We always run recovery on our own orphan dir - the dead
1069 * node(s) may have voted "no" on an inode delete earlier. A
1070 * revote is therefore required. */
1071 ocfs2_queue_recovery_completion(osb->journal, osb->slot_num, NULL,
1072 NULL);
1073
1074bail:
1075 down(&osb->recovery_lock);
1076 if (!status &&
1077 !ocfs2_node_map_is_empty(osb, &osb->recovery_map)) {
1078 up(&osb->recovery_lock);
1079 goto restart;
1080 }
1081
1082 osb->recovery_thread_task = NULL;
1083 mb(); /* sync with ocfs2_recovery_thread_running */
1084 wake_up(&osb->recovery_event);
1085
1086 up(&osb->recovery_lock);
1087
1088 mlog_exit(status);
1089 /* no one is callint kthread_stop() for us so the kthread() api
1090 * requires that we call do_exit(). And it isn't exported, but
1091 * complete_and_exit() seems to be a minimal wrapper around it. */
1092 complete_and_exit(NULL, status);
1093 return status;
1094}
1095
1096void ocfs2_recovery_thread(struct ocfs2_super *osb, int node_num)
1097{
1098 mlog_entry("(node_num=%d, osb->node_num = %d)\n",
1099 node_num, osb->node_num);
1100
1101 down(&osb->recovery_lock);
1102 if (osb->disable_recovery)
1103 goto out;
1104
1105 /* People waiting on recovery will wait on
1106 * the recovery map to empty. */
1107 if (!ocfs2_recovery_map_set(osb, node_num))
1108 mlog(0, "node %d already be in recovery.\n", node_num);
1109
1110 mlog(0, "starting recovery thread...\n");
1111
1112 if (osb->recovery_thread_task)
1113 goto out;
1114
1115 osb->recovery_thread_task = kthread_run(__ocfs2_recovery_thread, osb,
1116 "ocfs2rec-%d", osb->osb_id);
1117 if (IS_ERR(osb->recovery_thread_task)) {
1118 mlog_errno((int)PTR_ERR(osb->recovery_thread_task));
1119 osb->recovery_thread_task = NULL;
1120 }
1121
1122out:
1123 up(&osb->recovery_lock);
1124 wake_up(&osb->recovery_event);
1125
1126 mlog_exit_void();
1127}
1128
1129/* Does the actual journal replay and marks the journal inode as
1130 * clean. Will only replay if the journal inode is marked dirty. */
1131static int ocfs2_replay_journal(struct ocfs2_super *osb,
1132 int node_num,
1133 int slot_num)
1134{
1135 int status;
1136 int got_lock = 0;
1137 unsigned int flags;
1138 struct inode *inode = NULL;
1139 struct ocfs2_dinode *fe;
1140 journal_t *journal = NULL;
1141 struct buffer_head *bh = NULL;
1142
1143 inode = ocfs2_get_system_file_inode(osb, JOURNAL_SYSTEM_INODE,
1144 slot_num);
1145 if (inode == NULL) {
1146 status = -EACCES;
1147 mlog_errno(status);
1148 goto done;
1149 }
1150 if (is_bad_inode(inode)) {
1151 status = -EACCES;
1152 iput(inode);
1153 inode = NULL;
1154 mlog_errno(status);
1155 goto done;
1156 }
1157 SET_INODE_JOURNAL(inode);
1158
1159 status = ocfs2_meta_lock_full(inode, NULL, &bh, 1,
1160 OCFS2_META_LOCK_RECOVERY);
1161 if (status < 0) {
1162 mlog(0, "status returned from ocfs2_meta_lock=%d\n", status);
1163 if (status != -ERESTARTSYS)
1164 mlog(ML_ERROR, "Could not lock journal!\n");
1165 goto done;
1166 }
1167 got_lock = 1;
1168
1169 fe = (struct ocfs2_dinode *) bh->b_data;
1170
1171 flags = le32_to_cpu(fe->id1.journal1.ij_flags);
1172
1173 if (!(flags & OCFS2_JOURNAL_DIRTY_FL)) {
1174 mlog(0, "No recovery required for node %d\n", node_num);
1175 goto done;
1176 }
1177
1178 mlog(ML_NOTICE, "Recovering node %d from slot %d on device (%u,%u)\n",
1179 node_num, slot_num,
1180 MAJOR(osb->sb->s_dev), MINOR(osb->sb->s_dev));
1181
1182 OCFS2_I(inode)->ip_clusters = le32_to_cpu(fe->i_clusters);
1183
1184 status = ocfs2_force_read_journal(inode);
1185 if (status < 0) {
1186 mlog_errno(status);
1187 goto done;
1188 }
1189
1190 mlog(0, "calling journal_init_inode\n");
1191 journal = journal_init_inode(inode);
1192 if (journal == NULL) {
1193 mlog(ML_ERROR, "Linux journal layer error\n");
1194 status = -EIO;
1195 goto done;
1196 }
1197
1198 status = journal_load(journal);
1199 if (status < 0) {
1200 mlog_errno(status);
1201 if (!igrab(inode))
1202 BUG();
1203 journal_destroy(journal);
1204 goto done;
1205 }
1206
1207 ocfs2_clear_journal_error(osb->sb, journal, slot_num);
1208
1209 /* wipe the journal */
1210 mlog(0, "flushing the journal.\n");
1211 journal_lock_updates(journal);
1212 status = journal_flush(journal);
1213 journal_unlock_updates(journal);
1214 if (status < 0)
1215 mlog_errno(status);
1216
1217 /* This will mark the node clean */
1218 flags = le32_to_cpu(fe->id1.journal1.ij_flags);
1219 flags &= ~OCFS2_JOURNAL_DIRTY_FL;
1220 fe->id1.journal1.ij_flags = cpu_to_le32(flags);
1221
1222 status = ocfs2_write_block(osb, bh, inode);
1223 if (status < 0)
1224 mlog_errno(status);
1225
1226 if (!igrab(inode))
1227 BUG();
1228
1229 journal_destroy(journal);
1230
1231done:
1232 /* drop the lock on this nodes journal */
1233 if (got_lock)
1234 ocfs2_meta_unlock(inode, 1);
1235
1236 if (inode)
1237 iput(inode);
1238
1239 if (bh)
1240 brelse(bh);
1241
1242 mlog_exit(status);
1243 return status;
1244}
1245
1246/*
1247 * Do the most important parts of node recovery:
1248 * - Replay it's journal
1249 * - Stamp a clean local allocator file
1250 * - Stamp a clean truncate log
1251 * - Mark the node clean
1252 *
1253 * If this function completes without error, a node in OCFS2 can be
1254 * said to have been safely recovered. As a result, failure during the
1255 * second part of a nodes recovery process (local alloc recovery) is
1256 * far less concerning.
1257 */
1258static int ocfs2_recover_node(struct ocfs2_super *osb,
1259 int node_num)
1260{
1261 int status = 0;
1262 int slot_num;
1263 struct ocfs2_slot_info *si = osb->slot_info;
1264 struct ocfs2_dinode *la_copy = NULL;
1265 struct ocfs2_dinode *tl_copy = NULL;
1266
1267 mlog_entry("(node_num=%d, osb->node_num = %d)\n",
1268 node_num, osb->node_num);
1269
1270 mlog(0, "checking node %d\n", node_num);
1271
1272 /* Should not ever be called to recover ourselves -- in that
1273 * case we should've called ocfs2_journal_load instead. */
1274 if (osb->node_num == node_num)
1275 BUG();
1276
1277 slot_num = ocfs2_node_num_to_slot(si, node_num);
1278 if (slot_num == OCFS2_INVALID_SLOT) {
1279 status = 0;
1280 mlog(0, "no slot for this node, so no recovery required.\n");
1281 goto done;
1282 }
1283
1284 mlog(0, "node %d was using slot %d\n", node_num, slot_num);
1285
1286 status = ocfs2_replay_journal(osb, node_num, slot_num);
1287 if (status < 0) {
1288 mlog_errno(status);
1289 goto done;
1290 }
1291
1292 /* Stamp a clean local alloc file AFTER recovering the journal... */
1293 status = ocfs2_begin_local_alloc_recovery(osb, slot_num, &la_copy);
1294 if (status < 0) {
1295 mlog_errno(status);
1296 goto done;
1297 }
1298
1299 /* An error from begin_truncate_log_recovery is not
1300 * serious enough to warrant halting the rest of
1301 * recovery. */
1302 status = ocfs2_begin_truncate_log_recovery(osb, slot_num, &tl_copy);
1303 if (status < 0)
1304 mlog_errno(status);
1305
1306 /* Likewise, this would be a strange but ultimately not so
1307 * harmful place to get an error... */
1308 ocfs2_clear_slot(si, slot_num);
1309 status = ocfs2_update_disk_slots(osb, si);
1310 if (status < 0)
1311 mlog_errno(status);
1312
1313 /* This will kfree the memory pointed to by la_copy and tl_copy */
1314 ocfs2_queue_recovery_completion(osb->journal, slot_num, la_copy,
1315 tl_copy);
1316
1317 status = 0;
1318done:
1319
1320 mlog_exit(status);
1321 return status;
1322}
1323
1324/* Test node liveness by trylocking his journal. If we get the lock,
1325 * we drop it here. Return 0 if we got the lock, -EAGAIN if node is
1326 * still alive (we couldn't get the lock) and < 0 on error. */
1327static int ocfs2_trylock_journal(struct ocfs2_super *osb,
1328 int slot_num)
1329{
1330 int status, flags;
1331 struct inode *inode = NULL;
1332
1333 inode = ocfs2_get_system_file_inode(osb, JOURNAL_SYSTEM_INODE,
1334 slot_num);
1335 if (inode == NULL) {
1336 mlog(ML_ERROR, "access error\n");
1337 status = -EACCES;
1338 goto bail;
1339 }
1340 if (is_bad_inode(inode)) {
1341 mlog(ML_ERROR, "access error (bad inode)\n");
1342 iput(inode);
1343 inode = NULL;
1344 status = -EACCES;
1345 goto bail;
1346 }
1347 SET_INODE_JOURNAL(inode);
1348
1349 flags = OCFS2_META_LOCK_RECOVERY | OCFS2_META_LOCK_NOQUEUE;
1350 status = ocfs2_meta_lock_full(inode, NULL, NULL, 1, flags);
1351 if (status < 0) {
1352 if (status != -EAGAIN)
1353 mlog_errno(status);
1354 goto bail;
1355 }
1356
1357 ocfs2_meta_unlock(inode, 1);
1358bail:
1359 if (inode)
1360 iput(inode);
1361
1362 return status;
1363}
1364
1365/* Call this underneath ocfs2_super_lock. It also assumes that the
1366 * slot info struct has been updated from disk. */
1367int ocfs2_mark_dead_nodes(struct ocfs2_super *osb)
1368{
1369 int status, i, node_num;
1370 struct ocfs2_slot_info *si = osb->slot_info;
1371
1372 /* This is called with the super block cluster lock, so we
1373 * know that the slot map can't change underneath us. */
1374
1375 spin_lock(&si->si_lock);
1376 for(i = 0; i < si->si_num_slots; i++) {
1377 if (i == osb->slot_num)
1378 continue;
1379 if (ocfs2_is_empty_slot(si, i))
1380 continue;
1381
1382 node_num = si->si_global_node_nums[i];
1383 if (ocfs2_node_map_test_bit(osb, &osb->recovery_map, node_num))
1384 continue;
1385 spin_unlock(&si->si_lock);
1386
1387 /* Ok, we have a slot occupied by another node which
1388 * is not in the recovery map. We trylock his journal
1389 * file here to test if he's alive. */
1390 status = ocfs2_trylock_journal(osb, i);
1391 if (!status) {
1392 /* Since we're called from mount, we know that
1393 * the recovery thread can't race us on
1394 * setting / checking the recovery bits. */
1395 ocfs2_recovery_thread(osb, node_num);
1396 } else if ((status < 0) && (status != -EAGAIN)) {
1397 mlog_errno(status);
1398 goto bail;
1399 }
1400
1401 spin_lock(&si->si_lock);
1402 }
1403 spin_unlock(&si->si_lock);
1404
1405 status = 0;
1406bail:
1407 mlog_exit(status);
1408 return status;
1409}
1410
1411static int ocfs2_recover_orphans(struct ocfs2_super *osb,
1412 int slot)
1413{
1414 int status = 0;
1415 int have_disk_lock = 0;
1416 struct inode *inode = NULL;
1417 struct inode *iter;
1418 struct inode *orphan_dir_inode = NULL;
1419 unsigned long offset, blk, local;
1420 struct buffer_head *bh = NULL;
1421 struct ocfs2_dir_entry *de;
1422 struct super_block *sb = osb->sb;
1423 struct ocfs2_inode_info *oi;
1424
1425 mlog(0, "Recover inodes from orphan dir in slot %d\n", slot);
1426
1427 orphan_dir_inode = ocfs2_get_system_file_inode(osb,
1428 ORPHAN_DIR_SYSTEM_INODE,
1429 slot);
1430 if (!orphan_dir_inode) {
1431 status = -ENOENT;
1432 mlog_errno(status);
1433 goto out;
1434 }
1435
1436 down(&orphan_dir_inode->i_sem);
1437 status = ocfs2_meta_lock(orphan_dir_inode, NULL, NULL, 0);
1438 if (status < 0) {
1439 up(&orphan_dir_inode->i_sem);
1440 mlog_errno(status);
1441 goto out;
1442 }
1443 have_disk_lock = 1;
1444
1445 offset = 0;
1446 iter = NULL;
1447 while(offset < i_size_read(orphan_dir_inode)) {
1448 blk = offset >> sb->s_blocksize_bits;
1449
1450 bh = ocfs2_bread(orphan_dir_inode, blk, &status, 0);
1451 if (!bh)
1452 status = -EINVAL;
1453 if (status < 0) {
1454 up(&orphan_dir_inode->i_sem);
1455 if (bh)
1456 brelse(bh);
1457 mlog_errno(status);
1458 goto out;
1459 }
1460
1461 local = 0;
1462 while(offset < i_size_read(orphan_dir_inode)
1463 && local < sb->s_blocksize) {
1464 de = (struct ocfs2_dir_entry *) (bh->b_data + local);
1465
1466 if (!ocfs2_check_dir_entry(orphan_dir_inode,
1467 de, bh, local)) {
1468 up(&orphan_dir_inode->i_sem);
1469 status = -EINVAL;
1470 mlog_errno(status);
1471 brelse(bh);
1472 goto out;
1473 }
1474
1475 local += le16_to_cpu(de->rec_len);
1476 offset += le16_to_cpu(de->rec_len);
1477
1478 /* I guess we silently fail on no inode? */
1479 if (!le64_to_cpu(de->inode))
1480 continue;
1481 if (de->file_type > OCFS2_FT_MAX) {
1482 mlog(ML_ERROR,
1483 "block %llu contains invalid de: "
1484 "inode = %"MLFu64", rec_len = %u, "
1485 "name_len = %u, file_type = %u, "
1486 "name='%.*s'\n",
1487 (unsigned long long)bh->b_blocknr,
1488 le64_to_cpu(de->inode),
1489 le16_to_cpu(de->rec_len),
1490 de->name_len,
1491 de->file_type,
1492 de->name_len,
1493 de->name);
1494 continue;
1495 }
1496 if (de->name_len == 1 && !strncmp(".", de->name, 1))
1497 continue;
1498 if (de->name_len == 2 && !strncmp("..", de->name, 2))
1499 continue;
1500
1501 iter = ocfs2_iget(osb, le64_to_cpu(de->inode));
1502 if (IS_ERR(iter))
1503 continue;
1504
1505 mlog(0, "queue orphan %"MLFu64"\n",
1506 OCFS2_I(iter)->ip_blkno);
1507 OCFS2_I(iter)->ip_next_orphan = inode;
1508 inode = iter;
1509 }
1510 brelse(bh);
1511 }
1512 up(&orphan_dir_inode->i_sem);
1513
1514 ocfs2_meta_unlock(orphan_dir_inode, 0);
1515 have_disk_lock = 0;
1516
1517 iput(orphan_dir_inode);
1518 orphan_dir_inode = NULL;
1519
1520 while (inode) {
1521 oi = OCFS2_I(inode);
1522 mlog(0, "iput orphan %"MLFu64"\n", oi->ip_blkno);
1523
1524 iter = oi->ip_next_orphan;
1525
1526 spin_lock(&oi->ip_lock);
1527 /* Delete voting may have set these on the assumption
1528 * that the other node would wipe them successfully.
1529 * If they are still in the node's orphan dir, we need
1530 * to reset that state. */
1531 oi->ip_flags &= ~(OCFS2_INODE_DELETED|OCFS2_INODE_SKIP_DELETE);
1532
1533 /* Set the proper information to get us going into
1534 * ocfs2_delete_inode. */
1535 oi->ip_flags |= OCFS2_INODE_MAYBE_ORPHANED;
1536 oi->ip_orphaned_slot = slot;
1537 spin_unlock(&oi->ip_lock);
1538
1539 iput(inode);
1540
1541 inode = iter;
1542 }
1543
1544out:
1545 if (have_disk_lock)
1546 ocfs2_meta_unlock(orphan_dir_inode, 0);
1547
1548 if (orphan_dir_inode)
1549 iput(orphan_dir_inode);
1550
1551 return status;
1552}
1553
1554static int ocfs2_wait_on_mount(struct ocfs2_super *osb)
1555{
1556 /* This check is good because ocfs2 will wait on our recovery
1557 * thread before changing it to something other than MOUNTED
1558 * or DISABLED. */
1559 wait_event(osb->osb_mount_event,
1560 atomic_read(&osb->vol_state) == VOLUME_MOUNTED ||
1561 atomic_read(&osb->vol_state) == VOLUME_DISABLED);
1562
1563 /* If there's an error on mount, then we may never get to the
1564 * MOUNTED flag, but this is set right before
1565 * dismount_volume() so we can trust it. */
1566 if (atomic_read(&osb->vol_state) == VOLUME_DISABLED) {
1567 mlog(0, "mount error, exiting!\n");
1568 return -EBUSY;
1569 }
1570
1571 return 0;
1572}
1573
1574static int ocfs2_commit_thread(void *arg)
1575{
1576 int status;
1577 struct ocfs2_super *osb = arg;
1578 struct ocfs2_journal *journal = osb->journal;
1579
1580 /* we can trust j_num_trans here because _should_stop() is only set in
1581 * shutdown and nobody other than ourselves should be able to start
1582 * transactions. committing on shutdown might take a few iterations
1583 * as final transactions put deleted inodes on the list */
1584 while (!(kthread_should_stop() &&
1585 atomic_read(&journal->j_num_trans) == 0)) {
1586
1587 wait_event_interruptible_timeout(osb->checkpoint_event,
1588 atomic_read(&journal->j_num_trans)
1589 || kthread_should_stop(),
1590 OCFS2_CHECKPOINT_INTERVAL);
1591
1592 status = ocfs2_commit_cache(osb);
1593 if (status < 0)
1594 mlog_errno(status);
1595
1596 if (kthread_should_stop() && atomic_read(&journal->j_num_trans)){
1597 mlog(ML_KTHREAD,
1598 "commit_thread: %u transactions pending on "
1599 "shutdown\n",
1600 atomic_read(&journal->j_num_trans));
1601 }
1602 }
1603
1604 return 0;
1605}
1606
1607/* Look for a dirty journal without taking any cluster locks. Used for
1608 * hard readonly access to determine whether the file system journals
1609 * require recovery. */
1610int ocfs2_check_journals_nolocks(struct ocfs2_super *osb)
1611{
1612 int ret = 0;
1613 unsigned int slot;
1614 struct buffer_head *di_bh;
1615 struct ocfs2_dinode *di;
1616 struct inode *journal = NULL;
1617
1618 for(slot = 0; slot < osb->max_slots; slot++) {
1619 journal = ocfs2_get_system_file_inode(osb,
1620 JOURNAL_SYSTEM_INODE,
1621 slot);
1622 if (!journal || is_bad_inode(journal)) {
1623 ret = -EACCES;
1624 mlog_errno(ret);
1625 goto out;
1626 }
1627
1628 di_bh = NULL;
1629 ret = ocfs2_read_block(osb, OCFS2_I(journal)->ip_blkno, &di_bh,
1630 0, journal);
1631 if (ret < 0) {
1632 mlog_errno(ret);
1633 goto out;
1634 }
1635
1636 di = (struct ocfs2_dinode *) di_bh->b_data;
1637
1638 if (le32_to_cpu(di->id1.journal1.ij_flags) &
1639 OCFS2_JOURNAL_DIRTY_FL)
1640 ret = -EROFS;
1641
1642 brelse(di_bh);
1643 if (ret)
1644 break;
1645 }
1646
1647out:
1648 if (journal)
1649 iput(journal);
1650
1651 return ret;
1652}