Add the rt linux 4.1.3-rt3 as base
[kvmfornfv.git] / kernel / fs / ocfs2 / journal.c
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 #include <linux/time.h>
32 #include <linux/random.h>
33 #include <linux/delay.h>
34
35 #include <cluster/masklog.h>
36
37 #include "ocfs2.h"
38
39 #include "alloc.h"
40 #include "blockcheck.h"
41 #include "dir.h"
42 #include "dlmglue.h"
43 #include "extent_map.h"
44 #include "heartbeat.h"
45 #include "inode.h"
46 #include "journal.h"
47 #include "localalloc.h"
48 #include "slot_map.h"
49 #include "super.h"
50 #include "sysfile.h"
51 #include "uptodate.h"
52 #include "quota.h"
53 #include "file.h"
54 #include "namei.h"
55
56 #include "buffer_head_io.h"
57 #include "ocfs2_trace.h"
58
59 DEFINE_SPINLOCK(trans_inc_lock);
60
61 #define ORPHAN_SCAN_SCHEDULE_TIMEOUT 300000
62
63 static int ocfs2_force_read_journal(struct inode *inode);
64 static int ocfs2_recover_node(struct ocfs2_super *osb,
65                               int node_num, int slot_num);
66 static int __ocfs2_recovery_thread(void *arg);
67 static int ocfs2_commit_cache(struct ocfs2_super *osb);
68 static int __ocfs2_wait_on_mount(struct ocfs2_super *osb, int quota);
69 static int ocfs2_journal_toggle_dirty(struct ocfs2_super *osb,
70                                       int dirty, int replayed);
71 static int ocfs2_trylock_journal(struct ocfs2_super *osb,
72                                  int slot_num);
73 static int ocfs2_recover_orphans(struct ocfs2_super *osb,
74                                  int slot,
75                                  enum ocfs2_orphan_reco_type orphan_reco_type);
76 static int ocfs2_commit_thread(void *arg);
77 static void ocfs2_queue_recovery_completion(struct ocfs2_journal *journal,
78                                             int slot_num,
79                                             struct ocfs2_dinode *la_dinode,
80                                             struct ocfs2_dinode *tl_dinode,
81                                             struct ocfs2_quota_recovery *qrec,
82                                             enum ocfs2_orphan_reco_type orphan_reco_type);
83
84 static inline int ocfs2_wait_on_mount(struct ocfs2_super *osb)
85 {
86         return __ocfs2_wait_on_mount(osb, 0);
87 }
88
89 static inline int ocfs2_wait_on_quotas(struct ocfs2_super *osb)
90 {
91         return __ocfs2_wait_on_mount(osb, 1);
92 }
93
94 /*
95  * This replay_map is to track online/offline slots, so we could recover
96  * offline slots during recovery and mount
97  */
98
99 enum ocfs2_replay_state {
100         REPLAY_UNNEEDED = 0,    /* Replay is not needed, so ignore this map */
101         REPLAY_NEEDED,          /* Replay slots marked in rm_replay_slots */
102         REPLAY_DONE             /* Replay was already queued */
103 };
104
105 struct ocfs2_replay_map {
106         unsigned int rm_slots;
107         enum ocfs2_replay_state rm_state;
108         unsigned char rm_replay_slots[0];
109 };
110
111 void ocfs2_replay_map_set_state(struct ocfs2_super *osb, int state)
112 {
113         if (!osb->replay_map)
114                 return;
115
116         /* If we've already queued the replay, we don't have any more to do */
117         if (osb->replay_map->rm_state == REPLAY_DONE)
118                 return;
119
120         osb->replay_map->rm_state = state;
121 }
122
123 int ocfs2_compute_replay_slots(struct ocfs2_super *osb)
124 {
125         struct ocfs2_replay_map *replay_map;
126         int i, node_num;
127
128         /* If replay map is already set, we don't do it again */
129         if (osb->replay_map)
130                 return 0;
131
132         replay_map = kzalloc(sizeof(struct ocfs2_replay_map) +
133                              (osb->max_slots * sizeof(char)), GFP_KERNEL);
134
135         if (!replay_map) {
136                 mlog_errno(-ENOMEM);
137                 return -ENOMEM;
138         }
139
140         spin_lock(&osb->osb_lock);
141
142         replay_map->rm_slots = osb->max_slots;
143         replay_map->rm_state = REPLAY_UNNEEDED;
144
145         /* set rm_replay_slots for offline slot(s) */
146         for (i = 0; i < replay_map->rm_slots; i++) {
147                 if (ocfs2_slot_to_node_num_locked(osb, i, &node_num) == -ENOENT)
148                         replay_map->rm_replay_slots[i] = 1;
149         }
150
151         osb->replay_map = replay_map;
152         spin_unlock(&osb->osb_lock);
153         return 0;
154 }
155
156 void ocfs2_queue_replay_slots(struct ocfs2_super *osb,
157                 enum ocfs2_orphan_reco_type orphan_reco_type)
158 {
159         struct ocfs2_replay_map *replay_map = osb->replay_map;
160         int i;
161
162         if (!replay_map)
163                 return;
164
165         if (replay_map->rm_state != REPLAY_NEEDED)
166                 return;
167
168         for (i = 0; i < replay_map->rm_slots; i++)
169                 if (replay_map->rm_replay_slots[i])
170                         ocfs2_queue_recovery_completion(osb->journal, i, NULL,
171                                                         NULL, NULL,
172                                                         orphan_reco_type);
173         replay_map->rm_state = REPLAY_DONE;
174 }
175
176 void ocfs2_free_replay_slots(struct ocfs2_super *osb)
177 {
178         struct ocfs2_replay_map *replay_map = osb->replay_map;
179
180         if (!osb->replay_map)
181                 return;
182
183         kfree(replay_map);
184         osb->replay_map = NULL;
185 }
186
187 int ocfs2_recovery_init(struct ocfs2_super *osb)
188 {
189         struct ocfs2_recovery_map *rm;
190
191         mutex_init(&osb->recovery_lock);
192         osb->disable_recovery = 0;
193         osb->recovery_thread_task = NULL;
194         init_waitqueue_head(&osb->recovery_event);
195
196         rm = kzalloc(sizeof(struct ocfs2_recovery_map) +
197                      osb->max_slots * sizeof(unsigned int),
198                      GFP_KERNEL);
199         if (!rm) {
200                 mlog_errno(-ENOMEM);
201                 return -ENOMEM;
202         }
203
204         rm->rm_entries = (unsigned int *)((char *)rm +
205                                           sizeof(struct ocfs2_recovery_map));
206         osb->recovery_map = rm;
207
208         return 0;
209 }
210
211 /* we can't grab the goofy sem lock from inside wait_event, so we use
212  * memory barriers to make sure that we'll see the null task before
213  * being woken up */
214 static int ocfs2_recovery_thread_running(struct ocfs2_super *osb)
215 {
216         mb();
217         return osb->recovery_thread_task != NULL;
218 }
219
220 void ocfs2_recovery_exit(struct ocfs2_super *osb)
221 {
222         struct ocfs2_recovery_map *rm;
223
224         /* disable any new recovery threads and wait for any currently
225          * running ones to exit. Do this before setting the vol_state. */
226         mutex_lock(&osb->recovery_lock);
227         osb->disable_recovery = 1;
228         mutex_unlock(&osb->recovery_lock);
229         wait_event(osb->recovery_event, !ocfs2_recovery_thread_running(osb));
230
231         /* At this point, we know that no more recovery threads can be
232          * launched, so wait for any recovery completion work to
233          * complete. */
234         flush_workqueue(ocfs2_wq);
235
236         /*
237          * Now that recovery is shut down, and the osb is about to be
238          * freed,  the osb_lock is not taken here.
239          */
240         rm = osb->recovery_map;
241         /* XXX: Should we bug if there are dirty entries? */
242
243         kfree(rm);
244 }
245
246 static int __ocfs2_recovery_map_test(struct ocfs2_super *osb,
247                                      unsigned int node_num)
248 {
249         int i;
250         struct ocfs2_recovery_map *rm = osb->recovery_map;
251
252         assert_spin_locked(&osb->osb_lock);
253
254         for (i = 0; i < rm->rm_used; i++) {
255                 if (rm->rm_entries[i] == node_num)
256                         return 1;
257         }
258
259         return 0;
260 }
261
262 /* Behaves like test-and-set.  Returns the previous value */
263 static int ocfs2_recovery_map_set(struct ocfs2_super *osb,
264                                   unsigned int node_num)
265 {
266         struct ocfs2_recovery_map *rm = osb->recovery_map;
267
268         spin_lock(&osb->osb_lock);
269         if (__ocfs2_recovery_map_test(osb, node_num)) {
270                 spin_unlock(&osb->osb_lock);
271                 return 1;
272         }
273
274         /* XXX: Can this be exploited? Not from o2dlm... */
275         BUG_ON(rm->rm_used >= osb->max_slots);
276
277         rm->rm_entries[rm->rm_used] = node_num;
278         rm->rm_used++;
279         spin_unlock(&osb->osb_lock);
280
281         return 0;
282 }
283
284 static void ocfs2_recovery_map_clear(struct ocfs2_super *osb,
285                                      unsigned int node_num)
286 {
287         int i;
288         struct ocfs2_recovery_map *rm = osb->recovery_map;
289
290         spin_lock(&osb->osb_lock);
291
292         for (i = 0; i < rm->rm_used; i++) {
293                 if (rm->rm_entries[i] == node_num)
294                         break;
295         }
296
297         if (i < rm->rm_used) {
298                 /* XXX: be careful with the pointer math */
299                 memmove(&(rm->rm_entries[i]), &(rm->rm_entries[i + 1]),
300                         (rm->rm_used - i - 1) * sizeof(unsigned int));
301                 rm->rm_used--;
302         }
303
304         spin_unlock(&osb->osb_lock);
305 }
306
307 static int ocfs2_commit_cache(struct ocfs2_super *osb)
308 {
309         int status = 0;
310         unsigned int flushed;
311         struct ocfs2_journal *journal = NULL;
312
313         journal = osb->journal;
314
315         /* Flush all pending commits and checkpoint the journal. */
316         down_write(&journal->j_trans_barrier);
317
318         flushed = atomic_read(&journal->j_num_trans);
319         trace_ocfs2_commit_cache_begin(flushed);
320         if (flushed == 0) {
321                 up_write(&journal->j_trans_barrier);
322                 goto finally;
323         }
324
325         jbd2_journal_lock_updates(journal->j_journal);
326         status = jbd2_journal_flush(journal->j_journal);
327         jbd2_journal_unlock_updates(journal->j_journal);
328         if (status < 0) {
329                 up_write(&journal->j_trans_barrier);
330                 mlog_errno(status);
331                 goto finally;
332         }
333
334         ocfs2_inc_trans_id(journal);
335
336         flushed = atomic_read(&journal->j_num_trans);
337         atomic_set(&journal->j_num_trans, 0);
338         up_write(&journal->j_trans_barrier);
339
340         trace_ocfs2_commit_cache_end(journal->j_trans_id, flushed);
341
342         ocfs2_wake_downconvert_thread(osb);
343         wake_up(&journal->j_checkpointed);
344 finally:
345         return status;
346 }
347
348 handle_t *ocfs2_start_trans(struct ocfs2_super *osb, int max_buffs)
349 {
350         journal_t *journal = osb->journal->j_journal;
351         handle_t *handle;
352
353         BUG_ON(!osb || !osb->journal->j_journal);
354
355         if (ocfs2_is_hard_readonly(osb))
356                 return ERR_PTR(-EROFS);
357
358         BUG_ON(osb->journal->j_state == OCFS2_JOURNAL_FREE);
359         BUG_ON(max_buffs <= 0);
360
361         /* Nested transaction? Just return the handle... */
362         if (journal_current_handle())
363                 return jbd2_journal_start(journal, max_buffs);
364
365         sb_start_intwrite(osb->sb);
366
367         down_read(&osb->journal->j_trans_barrier);
368
369         handle = jbd2_journal_start(journal, max_buffs);
370         if (IS_ERR(handle)) {
371                 up_read(&osb->journal->j_trans_barrier);
372                 sb_end_intwrite(osb->sb);
373
374                 mlog_errno(PTR_ERR(handle));
375
376                 if (is_journal_aborted(journal)) {
377                         ocfs2_abort(osb->sb, "Detected aborted journal");
378                         handle = ERR_PTR(-EROFS);
379                 }
380         } else {
381                 if (!ocfs2_mount_local(osb))
382                         atomic_inc(&(osb->journal->j_num_trans));
383         }
384
385         return handle;
386 }
387
388 int ocfs2_commit_trans(struct ocfs2_super *osb,
389                        handle_t *handle)
390 {
391         int ret, nested;
392         struct ocfs2_journal *journal = osb->journal;
393
394         BUG_ON(!handle);
395
396         nested = handle->h_ref > 1;
397         ret = jbd2_journal_stop(handle);
398         if (ret < 0)
399                 mlog_errno(ret);
400
401         if (!nested) {
402                 up_read(&journal->j_trans_barrier);
403                 sb_end_intwrite(osb->sb);
404         }
405
406         return ret;
407 }
408
409 /*
410  * 'nblocks' is what you want to add to the current transaction.
411  *
412  * This might call jbd2_journal_restart() which will commit dirty buffers
413  * and then restart the transaction. Before calling
414  * ocfs2_extend_trans(), any changed blocks should have been
415  * dirtied. After calling it, all blocks which need to be changed must
416  * go through another set of journal_access/journal_dirty calls.
417  *
418  * WARNING: This will not release any semaphores or disk locks taken
419  * during the transaction, so make sure they were taken *before*
420  * start_trans or we'll have ordering deadlocks.
421  *
422  * WARNING2: Note that we do *not* drop j_trans_barrier here. This is
423  * good because transaction ids haven't yet been recorded on the
424  * cluster locks associated with this handle.
425  */
426 int ocfs2_extend_trans(handle_t *handle, int nblocks)
427 {
428         int status, old_nblocks;
429
430         BUG_ON(!handle);
431         BUG_ON(nblocks < 0);
432
433         if (!nblocks)
434                 return 0;
435
436         old_nblocks = handle->h_buffer_credits;
437
438         trace_ocfs2_extend_trans(old_nblocks, nblocks);
439
440 #ifdef CONFIG_OCFS2_DEBUG_FS
441         status = 1;
442 #else
443         status = jbd2_journal_extend(handle, nblocks);
444         if (status < 0) {
445                 mlog_errno(status);
446                 goto bail;
447         }
448 #endif
449
450         if (status > 0) {
451                 trace_ocfs2_extend_trans_restart(old_nblocks + nblocks);
452                 status = jbd2_journal_restart(handle,
453                                               old_nblocks + nblocks);
454                 if (status < 0) {
455                         mlog_errno(status);
456                         goto bail;
457                 }
458         }
459
460         status = 0;
461 bail:
462         return status;
463 }
464
465 /*
466  * If we have fewer than thresh credits, extend by OCFS2_MAX_TRANS_DATA.
467  * If that fails, restart the transaction & regain write access for the
468  * buffer head which is used for metadata modifications.
469  * Taken from Ext4: extend_or_restart_transaction()
470  */
471 int ocfs2_allocate_extend_trans(handle_t *handle, int thresh)
472 {
473         int status, old_nblks;
474
475         BUG_ON(!handle);
476
477         old_nblks = handle->h_buffer_credits;
478         trace_ocfs2_allocate_extend_trans(old_nblks, thresh);
479
480         if (old_nblks < thresh)
481                 return 0;
482
483         status = jbd2_journal_extend(handle, OCFS2_MAX_TRANS_DATA);
484         if (status < 0) {
485                 mlog_errno(status);
486                 goto bail;
487         }
488
489         if (status > 0) {
490                 status = jbd2_journal_restart(handle, OCFS2_MAX_TRANS_DATA);
491                 if (status < 0)
492                         mlog_errno(status);
493         }
494
495 bail:
496         return status;
497 }
498
499
500 struct ocfs2_triggers {
501         struct jbd2_buffer_trigger_type ot_triggers;
502         int                             ot_offset;
503 };
504
505 static inline struct ocfs2_triggers *to_ocfs2_trigger(struct jbd2_buffer_trigger_type *triggers)
506 {
507         return container_of(triggers, struct ocfs2_triggers, ot_triggers);
508 }
509
510 static void ocfs2_frozen_trigger(struct jbd2_buffer_trigger_type *triggers,
511                                  struct buffer_head *bh,
512                                  void *data, size_t size)
513 {
514         struct ocfs2_triggers *ot = to_ocfs2_trigger(triggers);
515
516         /*
517          * We aren't guaranteed to have the superblock here, so we
518          * must unconditionally compute the ecc data.
519          * __ocfs2_journal_access() will only set the triggers if
520          * metaecc is enabled.
521          */
522         ocfs2_block_check_compute(data, size, data + ot->ot_offset);
523 }
524
525 /*
526  * Quota blocks have their own trigger because the struct ocfs2_block_check
527  * offset depends on the blocksize.
528  */
529 static void ocfs2_dq_frozen_trigger(struct jbd2_buffer_trigger_type *triggers,
530                                  struct buffer_head *bh,
531                                  void *data, size_t size)
532 {
533         struct ocfs2_disk_dqtrailer *dqt =
534                 ocfs2_block_dqtrailer(size, data);
535
536         /*
537          * We aren't guaranteed to have the superblock here, so we
538          * must unconditionally compute the ecc data.
539          * __ocfs2_journal_access() will only set the triggers if
540          * metaecc is enabled.
541          */
542         ocfs2_block_check_compute(data, size, &dqt->dq_check);
543 }
544
545 /*
546  * Directory blocks also have their own trigger because the
547  * struct ocfs2_block_check offset depends on the blocksize.
548  */
549 static void ocfs2_db_frozen_trigger(struct jbd2_buffer_trigger_type *triggers,
550                                  struct buffer_head *bh,
551                                  void *data, size_t size)
552 {
553         struct ocfs2_dir_block_trailer *trailer =
554                 ocfs2_dir_trailer_from_size(size, data);
555
556         /*
557          * We aren't guaranteed to have the superblock here, so we
558          * must unconditionally compute the ecc data.
559          * __ocfs2_journal_access() will only set the triggers if
560          * metaecc is enabled.
561          */
562         ocfs2_block_check_compute(data, size, &trailer->db_check);
563 }
564
565 static void ocfs2_abort_trigger(struct jbd2_buffer_trigger_type *triggers,
566                                 struct buffer_head *bh)
567 {
568         mlog(ML_ERROR,
569              "ocfs2_abort_trigger called by JBD2.  bh = 0x%lx, "
570              "bh->b_blocknr = %llu\n",
571              (unsigned long)bh,
572              (unsigned long long)bh->b_blocknr);
573
574         /* We aren't guaranteed to have the superblock here - but if we
575          * don't, it'll just crash. */
576         ocfs2_error(bh->b_assoc_map->host->i_sb,
577                     "JBD2 has aborted our journal, ocfs2 cannot continue\n");
578 }
579
580 static struct ocfs2_triggers di_triggers = {
581         .ot_triggers = {
582                 .t_frozen = ocfs2_frozen_trigger,
583                 .t_abort = ocfs2_abort_trigger,
584         },
585         .ot_offset      = offsetof(struct ocfs2_dinode, i_check),
586 };
587
588 static struct ocfs2_triggers eb_triggers = {
589         .ot_triggers = {
590                 .t_frozen = ocfs2_frozen_trigger,
591                 .t_abort = ocfs2_abort_trigger,
592         },
593         .ot_offset      = offsetof(struct ocfs2_extent_block, h_check),
594 };
595
596 static struct ocfs2_triggers rb_triggers = {
597         .ot_triggers = {
598                 .t_frozen = ocfs2_frozen_trigger,
599                 .t_abort = ocfs2_abort_trigger,
600         },
601         .ot_offset      = offsetof(struct ocfs2_refcount_block, rf_check),
602 };
603
604 static struct ocfs2_triggers gd_triggers = {
605         .ot_triggers = {
606                 .t_frozen = ocfs2_frozen_trigger,
607                 .t_abort = ocfs2_abort_trigger,
608         },
609         .ot_offset      = offsetof(struct ocfs2_group_desc, bg_check),
610 };
611
612 static struct ocfs2_triggers db_triggers = {
613         .ot_triggers = {
614                 .t_frozen = ocfs2_db_frozen_trigger,
615                 .t_abort = ocfs2_abort_trigger,
616         },
617 };
618
619 static struct ocfs2_triggers xb_triggers = {
620         .ot_triggers = {
621                 .t_frozen = ocfs2_frozen_trigger,
622                 .t_abort = ocfs2_abort_trigger,
623         },
624         .ot_offset      = offsetof(struct ocfs2_xattr_block, xb_check),
625 };
626
627 static struct ocfs2_triggers dq_triggers = {
628         .ot_triggers = {
629                 .t_frozen = ocfs2_dq_frozen_trigger,
630                 .t_abort = ocfs2_abort_trigger,
631         },
632 };
633
634 static struct ocfs2_triggers dr_triggers = {
635         .ot_triggers = {
636                 .t_frozen = ocfs2_frozen_trigger,
637                 .t_abort = ocfs2_abort_trigger,
638         },
639         .ot_offset      = offsetof(struct ocfs2_dx_root_block, dr_check),
640 };
641
642 static struct ocfs2_triggers dl_triggers = {
643         .ot_triggers = {
644                 .t_frozen = ocfs2_frozen_trigger,
645                 .t_abort = ocfs2_abort_trigger,
646         },
647         .ot_offset      = offsetof(struct ocfs2_dx_leaf, dl_check),
648 };
649
650 static int __ocfs2_journal_access(handle_t *handle,
651                                   struct ocfs2_caching_info *ci,
652                                   struct buffer_head *bh,
653                                   struct ocfs2_triggers *triggers,
654                                   int type)
655 {
656         int status;
657         struct ocfs2_super *osb =
658                 OCFS2_SB(ocfs2_metadata_cache_get_super(ci));
659
660         BUG_ON(!ci || !ci->ci_ops);
661         BUG_ON(!handle);
662         BUG_ON(!bh);
663
664         trace_ocfs2_journal_access(
665                 (unsigned long long)ocfs2_metadata_cache_owner(ci),
666                 (unsigned long long)bh->b_blocknr, type, bh->b_size);
667
668         /* we can safely remove this assertion after testing. */
669         if (!buffer_uptodate(bh)) {
670                 mlog(ML_ERROR, "giving me a buffer that's not uptodate!\n");
671                 mlog(ML_ERROR, "b_blocknr=%llu\n",
672                      (unsigned long long)bh->b_blocknr);
673                 BUG();
674         }
675
676         /* Set the current transaction information on the ci so
677          * that the locking code knows whether it can drop it's locks
678          * on this ci or not. We're protected from the commit
679          * thread updating the current transaction id until
680          * ocfs2_commit_trans() because ocfs2_start_trans() took
681          * j_trans_barrier for us. */
682         ocfs2_set_ci_lock_trans(osb->journal, ci);
683
684         ocfs2_metadata_cache_io_lock(ci);
685         switch (type) {
686         case OCFS2_JOURNAL_ACCESS_CREATE:
687         case OCFS2_JOURNAL_ACCESS_WRITE:
688                 status = jbd2_journal_get_write_access(handle, bh);
689                 break;
690
691         case OCFS2_JOURNAL_ACCESS_UNDO:
692                 status = jbd2_journal_get_undo_access(handle, bh);
693                 break;
694
695         default:
696                 status = -EINVAL;
697                 mlog(ML_ERROR, "Unknown access type!\n");
698         }
699         if (!status && ocfs2_meta_ecc(osb) && triggers)
700                 jbd2_journal_set_triggers(bh, &triggers->ot_triggers);
701         ocfs2_metadata_cache_io_unlock(ci);
702
703         if (status < 0)
704                 mlog(ML_ERROR, "Error %d getting %d access to buffer!\n",
705                      status, type);
706
707         return status;
708 }
709
710 int ocfs2_journal_access_di(handle_t *handle, struct ocfs2_caching_info *ci,
711                             struct buffer_head *bh, int type)
712 {
713         return __ocfs2_journal_access(handle, ci, bh, &di_triggers, type);
714 }
715
716 int ocfs2_journal_access_eb(handle_t *handle, struct ocfs2_caching_info *ci,
717                             struct buffer_head *bh, int type)
718 {
719         return __ocfs2_journal_access(handle, ci, bh, &eb_triggers, type);
720 }
721
722 int ocfs2_journal_access_rb(handle_t *handle, struct ocfs2_caching_info *ci,
723                             struct buffer_head *bh, int type)
724 {
725         return __ocfs2_journal_access(handle, ci, bh, &rb_triggers,
726                                       type);
727 }
728
729 int ocfs2_journal_access_gd(handle_t *handle, struct ocfs2_caching_info *ci,
730                             struct buffer_head *bh, int type)
731 {
732         return __ocfs2_journal_access(handle, ci, bh, &gd_triggers, type);
733 }
734
735 int ocfs2_journal_access_db(handle_t *handle, struct ocfs2_caching_info *ci,
736                             struct buffer_head *bh, int type)
737 {
738         return __ocfs2_journal_access(handle, ci, bh, &db_triggers, type);
739 }
740
741 int ocfs2_journal_access_xb(handle_t *handle, struct ocfs2_caching_info *ci,
742                             struct buffer_head *bh, int type)
743 {
744         return __ocfs2_journal_access(handle, ci, bh, &xb_triggers, type);
745 }
746
747 int ocfs2_journal_access_dq(handle_t *handle, struct ocfs2_caching_info *ci,
748                             struct buffer_head *bh, int type)
749 {
750         return __ocfs2_journal_access(handle, ci, bh, &dq_triggers, type);
751 }
752
753 int ocfs2_journal_access_dr(handle_t *handle, struct ocfs2_caching_info *ci,
754                             struct buffer_head *bh, int type)
755 {
756         return __ocfs2_journal_access(handle, ci, bh, &dr_triggers, type);
757 }
758
759 int ocfs2_journal_access_dl(handle_t *handle, struct ocfs2_caching_info *ci,
760                             struct buffer_head *bh, int type)
761 {
762         return __ocfs2_journal_access(handle, ci, bh, &dl_triggers, type);
763 }
764
765 int ocfs2_journal_access(handle_t *handle, struct ocfs2_caching_info *ci,
766                          struct buffer_head *bh, int type)
767 {
768         return __ocfs2_journal_access(handle, ci, bh, NULL, type);
769 }
770
771 void ocfs2_journal_dirty(handle_t *handle, struct buffer_head *bh)
772 {
773         int status;
774
775         trace_ocfs2_journal_dirty((unsigned long long)bh->b_blocknr);
776
777         status = jbd2_journal_dirty_metadata(handle, bh);
778         BUG_ON(status);
779 }
780
781 #define OCFS2_DEFAULT_COMMIT_INTERVAL   (HZ * JBD2_DEFAULT_MAX_COMMIT_AGE)
782
783 void ocfs2_set_journal_params(struct ocfs2_super *osb)
784 {
785         journal_t *journal = osb->journal->j_journal;
786         unsigned long commit_interval = OCFS2_DEFAULT_COMMIT_INTERVAL;
787
788         if (osb->osb_commit_interval)
789                 commit_interval = osb->osb_commit_interval;
790
791         write_lock(&journal->j_state_lock);
792         journal->j_commit_interval = commit_interval;
793         if (osb->s_mount_opt & OCFS2_MOUNT_BARRIER)
794                 journal->j_flags |= JBD2_BARRIER;
795         else
796                 journal->j_flags &= ~JBD2_BARRIER;
797         write_unlock(&journal->j_state_lock);
798 }
799
800 int ocfs2_journal_init(struct ocfs2_journal *journal, int *dirty)
801 {
802         int status = -1;
803         struct inode *inode = NULL; /* the journal inode */
804         journal_t *j_journal = NULL;
805         struct ocfs2_dinode *di = NULL;
806         struct buffer_head *bh = NULL;
807         struct ocfs2_super *osb;
808         int inode_lock = 0;
809
810         BUG_ON(!journal);
811
812         osb = journal->j_osb;
813
814         /* already have the inode for our journal */
815         inode = ocfs2_get_system_file_inode(osb, JOURNAL_SYSTEM_INODE,
816                                             osb->slot_num);
817         if (inode == NULL) {
818                 status = -EACCES;
819                 mlog_errno(status);
820                 goto done;
821         }
822         if (is_bad_inode(inode)) {
823                 mlog(ML_ERROR, "access error (bad inode)\n");
824                 iput(inode);
825                 inode = NULL;
826                 status = -EACCES;
827                 goto done;
828         }
829
830         SET_INODE_JOURNAL(inode);
831         OCFS2_I(inode)->ip_open_count++;
832
833         /* Skip recovery waits here - journal inode metadata never
834          * changes in a live cluster so it can be considered an
835          * exception to the rule. */
836         status = ocfs2_inode_lock_full(inode, &bh, 1, OCFS2_META_LOCK_RECOVERY);
837         if (status < 0) {
838                 if (status != -ERESTARTSYS)
839                         mlog(ML_ERROR, "Could not get lock on journal!\n");
840                 goto done;
841         }
842
843         inode_lock = 1;
844         di = (struct ocfs2_dinode *)bh->b_data;
845
846         if (i_size_read(inode) <  OCFS2_MIN_JOURNAL_SIZE) {
847                 mlog(ML_ERROR, "Journal file size (%lld) is too small!\n",
848                      i_size_read(inode));
849                 status = -EINVAL;
850                 goto done;
851         }
852
853         trace_ocfs2_journal_init(i_size_read(inode),
854                                  (unsigned long long)inode->i_blocks,
855                                  OCFS2_I(inode)->ip_clusters);
856
857         /* call the kernels journal init function now */
858         j_journal = jbd2_journal_init_inode(inode);
859         if (j_journal == NULL) {
860                 mlog(ML_ERROR, "Linux journal layer error\n");
861                 status = -EINVAL;
862                 goto done;
863         }
864
865         trace_ocfs2_journal_init_maxlen(j_journal->j_maxlen);
866
867         *dirty = (le32_to_cpu(di->id1.journal1.ij_flags) &
868                   OCFS2_JOURNAL_DIRTY_FL);
869
870         journal->j_journal = j_journal;
871         journal->j_inode = inode;
872         journal->j_bh = bh;
873
874         ocfs2_set_journal_params(osb);
875
876         journal->j_state = OCFS2_JOURNAL_LOADED;
877
878         status = 0;
879 done:
880         if (status < 0) {
881                 if (inode_lock)
882                         ocfs2_inode_unlock(inode, 1);
883                 brelse(bh);
884                 if (inode) {
885                         OCFS2_I(inode)->ip_open_count--;
886                         iput(inode);
887                 }
888         }
889
890         return status;
891 }
892
893 static void ocfs2_bump_recovery_generation(struct ocfs2_dinode *di)
894 {
895         le32_add_cpu(&(di->id1.journal1.ij_recovery_generation), 1);
896 }
897
898 static u32 ocfs2_get_recovery_generation(struct ocfs2_dinode *di)
899 {
900         return le32_to_cpu(di->id1.journal1.ij_recovery_generation);
901 }
902
903 static int ocfs2_journal_toggle_dirty(struct ocfs2_super *osb,
904                                       int dirty, int replayed)
905 {
906         int status;
907         unsigned int flags;
908         struct ocfs2_journal *journal = osb->journal;
909         struct buffer_head *bh = journal->j_bh;
910         struct ocfs2_dinode *fe;
911
912         fe = (struct ocfs2_dinode *)bh->b_data;
913
914         /* The journal bh on the osb always comes from ocfs2_journal_init()
915          * and was validated there inside ocfs2_inode_lock_full().  It's a
916          * code bug if we mess it up. */
917         BUG_ON(!OCFS2_IS_VALID_DINODE(fe));
918
919         flags = le32_to_cpu(fe->id1.journal1.ij_flags);
920         if (dirty)
921                 flags |= OCFS2_JOURNAL_DIRTY_FL;
922         else
923                 flags &= ~OCFS2_JOURNAL_DIRTY_FL;
924         fe->id1.journal1.ij_flags = cpu_to_le32(flags);
925
926         if (replayed)
927                 ocfs2_bump_recovery_generation(fe);
928
929         ocfs2_compute_meta_ecc(osb->sb, bh->b_data, &fe->i_check);
930         status = ocfs2_write_block(osb, bh, INODE_CACHE(journal->j_inode));
931         if (status < 0)
932                 mlog_errno(status);
933
934         return status;
935 }
936
937 /*
938  * If the journal has been kmalloc'd it needs to be freed after this
939  * call.
940  */
941 void ocfs2_journal_shutdown(struct ocfs2_super *osb)
942 {
943         struct ocfs2_journal *journal = NULL;
944         int status = 0;
945         struct inode *inode = NULL;
946         int num_running_trans = 0;
947
948         BUG_ON(!osb);
949
950         journal = osb->journal;
951         if (!journal)
952                 goto done;
953
954         inode = journal->j_inode;
955
956         if (journal->j_state != OCFS2_JOURNAL_LOADED)
957                 goto done;
958
959         /* need to inc inode use count - jbd2_journal_destroy will iput. */
960         if (!igrab(inode))
961                 BUG();
962
963         num_running_trans = atomic_read(&(osb->journal->j_num_trans));
964         trace_ocfs2_journal_shutdown(num_running_trans);
965
966         /* Do a commit_cache here. It will flush our journal, *and*
967          * release any locks that are still held.
968          * set the SHUTDOWN flag and release the trans lock.
969          * the commit thread will take the trans lock for us below. */
970         journal->j_state = OCFS2_JOURNAL_IN_SHUTDOWN;
971
972         /* The OCFS2_JOURNAL_IN_SHUTDOWN will signal to commit_cache to not
973          * drop the trans_lock (which we want to hold until we
974          * completely destroy the journal. */
975         if (osb->commit_task) {
976                 /* Wait for the commit thread */
977                 trace_ocfs2_journal_shutdown_wait(osb->commit_task);
978                 kthread_stop(osb->commit_task);
979                 osb->commit_task = NULL;
980         }
981
982         BUG_ON(atomic_read(&(osb->journal->j_num_trans)) != 0);
983
984         if (ocfs2_mount_local(osb)) {
985                 jbd2_journal_lock_updates(journal->j_journal);
986                 status = jbd2_journal_flush(journal->j_journal);
987                 jbd2_journal_unlock_updates(journal->j_journal);
988                 if (status < 0)
989                         mlog_errno(status);
990         }
991
992         if (status == 0) {
993                 /*
994                  * Do not toggle if flush was unsuccessful otherwise
995                  * will leave dirty metadata in a "clean" journal
996                  */
997                 status = ocfs2_journal_toggle_dirty(osb, 0, 0);
998                 if (status < 0)
999                         mlog_errno(status);
1000         }
1001
1002         /* Shutdown the kernel journal system */
1003         jbd2_journal_destroy(journal->j_journal);
1004         journal->j_journal = NULL;
1005
1006         OCFS2_I(inode)->ip_open_count--;
1007
1008         /* unlock our journal */
1009         ocfs2_inode_unlock(inode, 1);
1010
1011         brelse(journal->j_bh);
1012         journal->j_bh = NULL;
1013
1014         journal->j_state = OCFS2_JOURNAL_FREE;
1015
1016 //      up_write(&journal->j_trans_barrier);
1017 done:
1018         if (inode)
1019                 iput(inode);
1020 }
1021
1022 static void ocfs2_clear_journal_error(struct super_block *sb,
1023                                       journal_t *journal,
1024                                       int slot)
1025 {
1026         int olderr;
1027
1028         olderr = jbd2_journal_errno(journal);
1029         if (olderr) {
1030                 mlog(ML_ERROR, "File system error %d recorded in "
1031                      "journal %u.\n", olderr, slot);
1032                 mlog(ML_ERROR, "File system on device %s needs checking.\n",
1033                      sb->s_id);
1034
1035                 jbd2_journal_ack_err(journal);
1036                 jbd2_journal_clear_err(journal);
1037         }
1038 }
1039
1040 int ocfs2_journal_load(struct ocfs2_journal *journal, int local, int replayed)
1041 {
1042         int status = 0;
1043         struct ocfs2_super *osb;
1044
1045         BUG_ON(!journal);
1046
1047         osb = journal->j_osb;
1048
1049         status = jbd2_journal_load(journal->j_journal);
1050         if (status < 0) {
1051                 mlog(ML_ERROR, "Failed to load journal!\n");
1052                 goto done;
1053         }
1054
1055         ocfs2_clear_journal_error(osb->sb, journal->j_journal, osb->slot_num);
1056
1057         status = ocfs2_journal_toggle_dirty(osb, 1, replayed);
1058         if (status < 0) {
1059                 mlog_errno(status);
1060                 goto done;
1061         }
1062
1063         /* Launch the commit thread */
1064         if (!local) {
1065                 osb->commit_task = kthread_run(ocfs2_commit_thread, osb,
1066                                                "ocfs2cmt");
1067                 if (IS_ERR(osb->commit_task)) {
1068                         status = PTR_ERR(osb->commit_task);
1069                         osb->commit_task = NULL;
1070                         mlog(ML_ERROR, "unable to launch ocfs2commit thread, "
1071                              "error=%d", status);
1072                         goto done;
1073                 }
1074         } else
1075                 osb->commit_task = NULL;
1076
1077 done:
1078         return status;
1079 }
1080
1081
1082 /* 'full' flag tells us whether we clear out all blocks or if we just
1083  * mark the journal clean */
1084 int ocfs2_journal_wipe(struct ocfs2_journal *journal, int full)
1085 {
1086         int status;
1087
1088         BUG_ON(!journal);
1089
1090         status = jbd2_journal_wipe(journal->j_journal, full);
1091         if (status < 0) {
1092                 mlog_errno(status);
1093                 goto bail;
1094         }
1095
1096         status = ocfs2_journal_toggle_dirty(journal->j_osb, 0, 0);
1097         if (status < 0)
1098                 mlog_errno(status);
1099
1100 bail:
1101         return status;
1102 }
1103
1104 static int ocfs2_recovery_completed(struct ocfs2_super *osb)
1105 {
1106         int empty;
1107         struct ocfs2_recovery_map *rm = osb->recovery_map;
1108
1109         spin_lock(&osb->osb_lock);
1110         empty = (rm->rm_used == 0);
1111         spin_unlock(&osb->osb_lock);
1112
1113         return empty;
1114 }
1115
1116 void ocfs2_wait_for_recovery(struct ocfs2_super *osb)
1117 {
1118         wait_event(osb->recovery_event, ocfs2_recovery_completed(osb));
1119 }
1120
1121 /*
1122  * JBD Might read a cached version of another nodes journal file. We
1123  * don't want this as this file changes often and we get no
1124  * notification on those changes. The only way to be sure that we've
1125  * got the most up to date version of those blocks then is to force
1126  * read them off disk. Just searching through the buffer cache won't
1127  * work as there may be pages backing this file which are still marked
1128  * up to date. We know things can't change on this file underneath us
1129  * as we have the lock by now :)
1130  */
1131 static int ocfs2_force_read_journal(struct inode *inode)
1132 {
1133         int status = 0;
1134         int i;
1135         u64 v_blkno, p_blkno, p_blocks, num_blocks;
1136 #define CONCURRENT_JOURNAL_FILL 32ULL
1137         struct buffer_head *bhs[CONCURRENT_JOURNAL_FILL];
1138
1139         memset(bhs, 0, sizeof(struct buffer_head *) * CONCURRENT_JOURNAL_FILL);
1140
1141         num_blocks = ocfs2_blocks_for_bytes(inode->i_sb, i_size_read(inode));
1142         v_blkno = 0;
1143         while (v_blkno < num_blocks) {
1144                 status = ocfs2_extent_map_get_blocks(inode, v_blkno,
1145                                                      &p_blkno, &p_blocks, NULL);
1146                 if (status < 0) {
1147                         mlog_errno(status);
1148                         goto bail;
1149                 }
1150
1151                 if (p_blocks > CONCURRENT_JOURNAL_FILL)
1152                         p_blocks = CONCURRENT_JOURNAL_FILL;
1153
1154                 /* We are reading journal data which should not
1155                  * be put in the uptodate cache */
1156                 status = ocfs2_read_blocks_sync(OCFS2_SB(inode->i_sb),
1157                                                 p_blkno, p_blocks, bhs);
1158                 if (status < 0) {
1159                         mlog_errno(status);
1160                         goto bail;
1161                 }
1162
1163                 for(i = 0; i < p_blocks; i++) {
1164                         brelse(bhs[i]);
1165                         bhs[i] = NULL;
1166                 }
1167
1168                 v_blkno += p_blocks;
1169         }
1170
1171 bail:
1172         for(i = 0; i < CONCURRENT_JOURNAL_FILL; i++)
1173                 brelse(bhs[i]);
1174         return status;
1175 }
1176
1177 struct ocfs2_la_recovery_item {
1178         struct list_head        lri_list;
1179         int                     lri_slot;
1180         struct ocfs2_dinode     *lri_la_dinode;
1181         struct ocfs2_dinode     *lri_tl_dinode;
1182         struct ocfs2_quota_recovery *lri_qrec;
1183         enum ocfs2_orphan_reco_type  lri_orphan_reco_type;
1184 };
1185
1186 /* Does the second half of the recovery process. By this point, the
1187  * node is marked clean and can actually be considered recovered,
1188  * hence it's no longer in the recovery map, but there's still some
1189  * cleanup we can do which shouldn't happen within the recovery thread
1190  * as locking in that context becomes very difficult if we are to take
1191  * recovering nodes into account.
1192  *
1193  * NOTE: This function can and will sleep on recovery of other nodes
1194  * during cluster locking, just like any other ocfs2 process.
1195  */
1196 void ocfs2_complete_recovery(struct work_struct *work)
1197 {
1198         int ret = 0;
1199         struct ocfs2_journal *journal =
1200                 container_of(work, struct ocfs2_journal, j_recovery_work);
1201         struct ocfs2_super *osb = journal->j_osb;
1202         struct ocfs2_dinode *la_dinode, *tl_dinode;
1203         struct ocfs2_la_recovery_item *item, *n;
1204         struct ocfs2_quota_recovery *qrec;
1205         enum ocfs2_orphan_reco_type orphan_reco_type;
1206         LIST_HEAD(tmp_la_list);
1207
1208         trace_ocfs2_complete_recovery(
1209                 (unsigned long long)OCFS2_I(journal->j_inode)->ip_blkno);
1210
1211         spin_lock(&journal->j_lock);
1212         list_splice_init(&journal->j_la_cleanups, &tmp_la_list);
1213         spin_unlock(&journal->j_lock);
1214
1215         list_for_each_entry_safe(item, n, &tmp_la_list, lri_list) {
1216                 list_del_init(&item->lri_list);
1217
1218                 ocfs2_wait_on_quotas(osb);
1219
1220                 la_dinode = item->lri_la_dinode;
1221                 tl_dinode = item->lri_tl_dinode;
1222                 qrec = item->lri_qrec;
1223                 orphan_reco_type = item->lri_orphan_reco_type;
1224
1225                 trace_ocfs2_complete_recovery_slot(item->lri_slot,
1226                         la_dinode ? le64_to_cpu(la_dinode->i_blkno) : 0,
1227                         tl_dinode ? le64_to_cpu(tl_dinode->i_blkno) : 0,
1228                         qrec);
1229
1230                 if (la_dinode) {
1231                         ret = ocfs2_complete_local_alloc_recovery(osb,
1232                                                                   la_dinode);
1233                         if (ret < 0)
1234                                 mlog_errno(ret);
1235
1236                         kfree(la_dinode);
1237                 }
1238
1239                 if (tl_dinode) {
1240                         ret = ocfs2_complete_truncate_log_recovery(osb,
1241                                                                    tl_dinode);
1242                         if (ret < 0)
1243                                 mlog_errno(ret);
1244
1245                         kfree(tl_dinode);
1246                 }
1247
1248                 ret = ocfs2_recover_orphans(osb, item->lri_slot,
1249                                 orphan_reco_type);
1250                 if (ret < 0)
1251                         mlog_errno(ret);
1252
1253                 if (qrec) {
1254                         ret = ocfs2_finish_quota_recovery(osb, qrec,
1255                                                           item->lri_slot);
1256                         if (ret < 0)
1257                                 mlog_errno(ret);
1258                         /* Recovery info is already freed now */
1259                 }
1260
1261                 kfree(item);
1262         }
1263
1264         trace_ocfs2_complete_recovery_end(ret);
1265 }
1266
1267 /* NOTE: This function always eats your references to la_dinode and
1268  * tl_dinode, either manually on error, or by passing them to
1269  * ocfs2_complete_recovery */
1270 static void ocfs2_queue_recovery_completion(struct ocfs2_journal *journal,
1271                                             int slot_num,
1272                                             struct ocfs2_dinode *la_dinode,
1273                                             struct ocfs2_dinode *tl_dinode,
1274                                             struct ocfs2_quota_recovery *qrec,
1275                                             enum ocfs2_orphan_reco_type orphan_reco_type)
1276 {
1277         struct ocfs2_la_recovery_item *item;
1278
1279         item = kmalloc(sizeof(struct ocfs2_la_recovery_item), GFP_NOFS);
1280         if (!item) {
1281                 /* Though we wish to avoid it, we are in fact safe in
1282                  * skipping local alloc cleanup as fsck.ocfs2 is more
1283                  * than capable of reclaiming unused space. */
1284                 kfree(la_dinode);
1285                 kfree(tl_dinode);
1286
1287                 if (qrec)
1288                         ocfs2_free_quota_recovery(qrec);
1289
1290                 mlog_errno(-ENOMEM);
1291                 return;
1292         }
1293
1294         INIT_LIST_HEAD(&item->lri_list);
1295         item->lri_la_dinode = la_dinode;
1296         item->lri_slot = slot_num;
1297         item->lri_tl_dinode = tl_dinode;
1298         item->lri_qrec = qrec;
1299         item->lri_orphan_reco_type = orphan_reco_type;
1300
1301         spin_lock(&journal->j_lock);
1302         list_add_tail(&item->lri_list, &journal->j_la_cleanups);
1303         queue_work(ocfs2_wq, &journal->j_recovery_work);
1304         spin_unlock(&journal->j_lock);
1305 }
1306
1307 /* Called by the mount code to queue recovery the last part of
1308  * recovery for it's own and offline slot(s). */
1309 void ocfs2_complete_mount_recovery(struct ocfs2_super *osb)
1310 {
1311         struct ocfs2_journal *journal = osb->journal;
1312
1313         if (ocfs2_is_hard_readonly(osb))
1314                 return;
1315
1316         /* No need to queue up our truncate_log as regular cleanup will catch
1317          * that */
1318         ocfs2_queue_recovery_completion(journal, osb->slot_num,
1319                                         osb->local_alloc_copy, NULL, NULL,
1320                                         ORPHAN_NEED_TRUNCATE);
1321         ocfs2_schedule_truncate_log_flush(osb, 0);
1322
1323         osb->local_alloc_copy = NULL;
1324         osb->dirty = 0;
1325
1326         /* queue to recover orphan slots for all offline slots */
1327         ocfs2_replay_map_set_state(osb, REPLAY_NEEDED);
1328         ocfs2_queue_replay_slots(osb, ORPHAN_NEED_TRUNCATE);
1329         ocfs2_free_replay_slots(osb);
1330 }
1331
1332 void ocfs2_complete_quota_recovery(struct ocfs2_super *osb)
1333 {
1334         if (osb->quota_rec) {
1335                 ocfs2_queue_recovery_completion(osb->journal,
1336                                                 osb->slot_num,
1337                                                 NULL,
1338                                                 NULL,
1339                                                 osb->quota_rec,
1340                                                 ORPHAN_NEED_TRUNCATE);
1341                 osb->quota_rec = NULL;
1342         }
1343 }
1344
1345 static int __ocfs2_recovery_thread(void *arg)
1346 {
1347         int status, node_num, slot_num;
1348         struct ocfs2_super *osb = arg;
1349         struct ocfs2_recovery_map *rm = osb->recovery_map;
1350         int *rm_quota = NULL;
1351         int rm_quota_used = 0, i;
1352         struct ocfs2_quota_recovery *qrec;
1353
1354         status = ocfs2_wait_on_mount(osb);
1355         if (status < 0) {
1356                 goto bail;
1357         }
1358
1359         rm_quota = kzalloc(osb->max_slots * sizeof(int), GFP_NOFS);
1360         if (!rm_quota) {
1361                 status = -ENOMEM;
1362                 goto bail;
1363         }
1364 restart:
1365         status = ocfs2_super_lock(osb, 1);
1366         if (status < 0) {
1367                 mlog_errno(status);
1368                 goto bail;
1369         }
1370
1371         status = ocfs2_compute_replay_slots(osb);
1372         if (status < 0)
1373                 mlog_errno(status);
1374
1375         /* queue recovery for our own slot */
1376         ocfs2_queue_recovery_completion(osb->journal, osb->slot_num, NULL,
1377                                         NULL, NULL, ORPHAN_NO_NEED_TRUNCATE);
1378
1379         spin_lock(&osb->osb_lock);
1380         while (rm->rm_used) {
1381                 /* It's always safe to remove entry zero, as we won't
1382                  * clear it until ocfs2_recover_node() has succeeded. */
1383                 node_num = rm->rm_entries[0];
1384                 spin_unlock(&osb->osb_lock);
1385                 slot_num = ocfs2_node_num_to_slot(osb, node_num);
1386                 trace_ocfs2_recovery_thread_node(node_num, slot_num);
1387                 if (slot_num == -ENOENT) {
1388                         status = 0;
1389                         goto skip_recovery;
1390                 }
1391
1392                 /* It is a bit subtle with quota recovery. We cannot do it
1393                  * immediately because we have to obtain cluster locks from
1394                  * quota files and we also don't want to just skip it because
1395                  * then quota usage would be out of sync until some node takes
1396                  * the slot. So we remember which nodes need quota recovery
1397                  * and when everything else is done, we recover quotas. */
1398                 for (i = 0; i < rm_quota_used && rm_quota[i] != slot_num; i++);
1399                 if (i == rm_quota_used)
1400                         rm_quota[rm_quota_used++] = slot_num;
1401
1402                 status = ocfs2_recover_node(osb, node_num, slot_num);
1403 skip_recovery:
1404                 if (!status) {
1405                         ocfs2_recovery_map_clear(osb, node_num);
1406                 } else {
1407                         mlog(ML_ERROR,
1408                              "Error %d recovering node %d on device (%u,%u)!\n",
1409                              status, node_num,
1410                              MAJOR(osb->sb->s_dev), MINOR(osb->sb->s_dev));
1411                         mlog(ML_ERROR, "Volume requires unmount.\n");
1412                 }
1413
1414                 spin_lock(&osb->osb_lock);
1415         }
1416         spin_unlock(&osb->osb_lock);
1417         trace_ocfs2_recovery_thread_end(status);
1418
1419         /* Refresh all journal recovery generations from disk */
1420         status = ocfs2_check_journals_nolocks(osb);
1421         status = (status == -EROFS) ? 0 : status;
1422         if (status < 0)
1423                 mlog_errno(status);
1424
1425         /* Now it is right time to recover quotas... We have to do this under
1426          * superblock lock so that no one can start using the slot (and crash)
1427          * before we recover it */
1428         for (i = 0; i < rm_quota_used; i++) {
1429                 qrec = ocfs2_begin_quota_recovery(osb, rm_quota[i]);
1430                 if (IS_ERR(qrec)) {
1431                         status = PTR_ERR(qrec);
1432                         mlog_errno(status);
1433                         continue;
1434                 }
1435                 ocfs2_queue_recovery_completion(osb->journal, rm_quota[i],
1436                                                 NULL, NULL, qrec,
1437                                                 ORPHAN_NEED_TRUNCATE);
1438         }
1439
1440         ocfs2_super_unlock(osb, 1);
1441
1442         /* queue recovery for offline slots */
1443         ocfs2_queue_replay_slots(osb, ORPHAN_NEED_TRUNCATE);
1444
1445 bail:
1446         mutex_lock(&osb->recovery_lock);
1447         if (!status && !ocfs2_recovery_completed(osb)) {
1448                 mutex_unlock(&osb->recovery_lock);
1449                 goto restart;
1450         }
1451
1452         ocfs2_free_replay_slots(osb);
1453         osb->recovery_thread_task = NULL;
1454         mb(); /* sync with ocfs2_recovery_thread_running */
1455         wake_up(&osb->recovery_event);
1456
1457         mutex_unlock(&osb->recovery_lock);
1458
1459         kfree(rm_quota);
1460
1461         /* no one is callint kthread_stop() for us so the kthread() api
1462          * requires that we call do_exit().  And it isn't exported, but
1463          * complete_and_exit() seems to be a minimal wrapper around it. */
1464         complete_and_exit(NULL, status);
1465 }
1466
1467 void ocfs2_recovery_thread(struct ocfs2_super *osb, int node_num)
1468 {
1469         mutex_lock(&osb->recovery_lock);
1470
1471         trace_ocfs2_recovery_thread(node_num, osb->node_num,
1472                 osb->disable_recovery, osb->recovery_thread_task,
1473                 osb->disable_recovery ?
1474                 -1 : ocfs2_recovery_map_set(osb, node_num));
1475
1476         if (osb->disable_recovery)
1477                 goto out;
1478
1479         if (osb->recovery_thread_task)
1480                 goto out;
1481
1482         osb->recovery_thread_task =  kthread_run(__ocfs2_recovery_thread, osb,
1483                                                  "ocfs2rec");
1484         if (IS_ERR(osb->recovery_thread_task)) {
1485                 mlog_errno((int)PTR_ERR(osb->recovery_thread_task));
1486                 osb->recovery_thread_task = NULL;
1487         }
1488
1489 out:
1490         mutex_unlock(&osb->recovery_lock);
1491         wake_up(&osb->recovery_event);
1492 }
1493
1494 static int ocfs2_read_journal_inode(struct ocfs2_super *osb,
1495                                     int slot_num,
1496                                     struct buffer_head **bh,
1497                                     struct inode **ret_inode)
1498 {
1499         int status = -EACCES;
1500         struct inode *inode = NULL;
1501
1502         BUG_ON(slot_num >= osb->max_slots);
1503
1504         inode = ocfs2_get_system_file_inode(osb, JOURNAL_SYSTEM_INODE,
1505                                             slot_num);
1506         if (!inode || is_bad_inode(inode)) {
1507                 mlog_errno(status);
1508                 goto bail;
1509         }
1510         SET_INODE_JOURNAL(inode);
1511
1512         status = ocfs2_read_inode_block_full(inode, bh, OCFS2_BH_IGNORE_CACHE);
1513         if (status < 0) {
1514                 mlog_errno(status);
1515                 goto bail;
1516         }
1517
1518         status = 0;
1519
1520 bail:
1521         if (inode) {
1522                 if (status || !ret_inode)
1523                         iput(inode);
1524                 else
1525                         *ret_inode = inode;
1526         }
1527         return status;
1528 }
1529
1530 /* Does the actual journal replay and marks the journal inode as
1531  * clean. Will only replay if the journal inode is marked dirty. */
1532 static int ocfs2_replay_journal(struct ocfs2_super *osb,
1533                                 int node_num,
1534                                 int slot_num)
1535 {
1536         int status;
1537         int got_lock = 0;
1538         unsigned int flags;
1539         struct inode *inode = NULL;
1540         struct ocfs2_dinode *fe;
1541         journal_t *journal = NULL;
1542         struct buffer_head *bh = NULL;
1543         u32 slot_reco_gen;
1544
1545         status = ocfs2_read_journal_inode(osb, slot_num, &bh, &inode);
1546         if (status) {
1547                 mlog_errno(status);
1548                 goto done;
1549         }
1550
1551         fe = (struct ocfs2_dinode *)bh->b_data;
1552         slot_reco_gen = ocfs2_get_recovery_generation(fe);
1553         brelse(bh);
1554         bh = NULL;
1555
1556         /*
1557          * As the fs recovery is asynchronous, there is a small chance that
1558          * another node mounted (and recovered) the slot before the recovery
1559          * thread could get the lock. To handle that, we dirty read the journal
1560          * inode for that slot to get the recovery generation. If it is
1561          * different than what we expected, the slot has been recovered.
1562          * If not, it needs recovery.
1563          */
1564         if (osb->slot_recovery_generations[slot_num] != slot_reco_gen) {
1565                 trace_ocfs2_replay_journal_recovered(slot_num,
1566                      osb->slot_recovery_generations[slot_num], slot_reco_gen);
1567                 osb->slot_recovery_generations[slot_num] = slot_reco_gen;
1568                 status = -EBUSY;
1569                 goto done;
1570         }
1571
1572         /* Continue with recovery as the journal has not yet been recovered */
1573
1574         status = ocfs2_inode_lock_full(inode, &bh, 1, OCFS2_META_LOCK_RECOVERY);
1575         if (status < 0) {
1576                 trace_ocfs2_replay_journal_lock_err(status);
1577                 if (status != -ERESTARTSYS)
1578                         mlog(ML_ERROR, "Could not lock journal!\n");
1579                 goto done;
1580         }
1581         got_lock = 1;
1582
1583         fe = (struct ocfs2_dinode *) bh->b_data;
1584
1585         flags = le32_to_cpu(fe->id1.journal1.ij_flags);
1586         slot_reco_gen = ocfs2_get_recovery_generation(fe);
1587
1588         if (!(flags & OCFS2_JOURNAL_DIRTY_FL)) {
1589                 trace_ocfs2_replay_journal_skip(node_num);
1590                 /* Refresh recovery generation for the slot */
1591                 osb->slot_recovery_generations[slot_num] = slot_reco_gen;
1592                 goto done;
1593         }
1594
1595         /* we need to run complete recovery for offline orphan slots */
1596         ocfs2_replay_map_set_state(osb, REPLAY_NEEDED);
1597
1598         printk(KERN_NOTICE "ocfs2: Begin replay journal (node %d, slot %d) on "\
1599                "device (%u,%u)\n", node_num, slot_num, MAJOR(osb->sb->s_dev),
1600                MINOR(osb->sb->s_dev));
1601
1602         OCFS2_I(inode)->ip_clusters = le32_to_cpu(fe->i_clusters);
1603
1604         status = ocfs2_force_read_journal(inode);
1605         if (status < 0) {
1606                 mlog_errno(status);
1607                 goto done;
1608         }
1609
1610         journal = jbd2_journal_init_inode(inode);
1611         if (journal == NULL) {
1612                 mlog(ML_ERROR, "Linux journal layer error\n");
1613                 status = -EIO;
1614                 goto done;
1615         }
1616
1617         status = jbd2_journal_load(journal);
1618         if (status < 0) {
1619                 mlog_errno(status);
1620                 if (!igrab(inode))
1621                         BUG();
1622                 jbd2_journal_destroy(journal);
1623                 goto done;
1624         }
1625
1626         ocfs2_clear_journal_error(osb->sb, journal, slot_num);
1627
1628         /* wipe the journal */
1629         jbd2_journal_lock_updates(journal);
1630         status = jbd2_journal_flush(journal);
1631         jbd2_journal_unlock_updates(journal);
1632         if (status < 0)
1633                 mlog_errno(status);
1634
1635         /* This will mark the node clean */
1636         flags = le32_to_cpu(fe->id1.journal1.ij_flags);
1637         flags &= ~OCFS2_JOURNAL_DIRTY_FL;
1638         fe->id1.journal1.ij_flags = cpu_to_le32(flags);
1639
1640         /* Increment recovery generation to indicate successful recovery */
1641         ocfs2_bump_recovery_generation(fe);
1642         osb->slot_recovery_generations[slot_num] =
1643                                         ocfs2_get_recovery_generation(fe);
1644
1645         ocfs2_compute_meta_ecc(osb->sb, bh->b_data, &fe->i_check);
1646         status = ocfs2_write_block(osb, bh, INODE_CACHE(inode));
1647         if (status < 0)
1648                 mlog_errno(status);
1649
1650         if (!igrab(inode))
1651                 BUG();
1652
1653         jbd2_journal_destroy(journal);
1654
1655         printk(KERN_NOTICE "ocfs2: End replay journal (node %d, slot %d) on "\
1656                "device (%u,%u)\n", node_num, slot_num, MAJOR(osb->sb->s_dev),
1657                MINOR(osb->sb->s_dev));
1658 done:
1659         /* drop the lock on this nodes journal */
1660         if (got_lock)
1661                 ocfs2_inode_unlock(inode, 1);
1662
1663         if (inode)
1664                 iput(inode);
1665
1666         brelse(bh);
1667
1668         return status;
1669 }
1670
1671 /*
1672  * Do the most important parts of node recovery:
1673  *  - Replay it's journal
1674  *  - Stamp a clean local allocator file
1675  *  - Stamp a clean truncate log
1676  *  - Mark the node clean
1677  *
1678  * If this function completes without error, a node in OCFS2 can be
1679  * said to have been safely recovered. As a result, failure during the
1680  * second part of a nodes recovery process (local alloc recovery) is
1681  * far less concerning.
1682  */
1683 static int ocfs2_recover_node(struct ocfs2_super *osb,
1684                               int node_num, int slot_num)
1685 {
1686         int status = 0;
1687         struct ocfs2_dinode *la_copy = NULL;
1688         struct ocfs2_dinode *tl_copy = NULL;
1689
1690         trace_ocfs2_recover_node(node_num, slot_num, osb->node_num);
1691
1692         /* Should not ever be called to recover ourselves -- in that
1693          * case we should've called ocfs2_journal_load instead. */
1694         BUG_ON(osb->node_num == node_num);
1695
1696         status = ocfs2_replay_journal(osb, node_num, slot_num);
1697         if (status < 0) {
1698                 if (status == -EBUSY) {
1699                         trace_ocfs2_recover_node_skip(slot_num, node_num);
1700                         status = 0;
1701                         goto done;
1702                 }
1703                 mlog_errno(status);
1704                 goto done;
1705         }
1706
1707         /* Stamp a clean local alloc file AFTER recovering the journal... */
1708         status = ocfs2_begin_local_alloc_recovery(osb, slot_num, &la_copy);
1709         if (status < 0) {
1710                 mlog_errno(status);
1711                 goto done;
1712         }
1713
1714         /* An error from begin_truncate_log_recovery is not
1715          * serious enough to warrant halting the rest of
1716          * recovery. */
1717         status = ocfs2_begin_truncate_log_recovery(osb, slot_num, &tl_copy);
1718         if (status < 0)
1719                 mlog_errno(status);
1720
1721         /* Likewise, this would be a strange but ultimately not so
1722          * harmful place to get an error... */
1723         status = ocfs2_clear_slot(osb, slot_num);
1724         if (status < 0)
1725                 mlog_errno(status);
1726
1727         /* This will kfree the memory pointed to by la_copy and tl_copy */
1728         ocfs2_queue_recovery_completion(osb->journal, slot_num, la_copy,
1729                                         tl_copy, NULL, ORPHAN_NEED_TRUNCATE);
1730
1731         status = 0;
1732 done:
1733
1734         return status;
1735 }
1736
1737 /* Test node liveness by trylocking his journal. If we get the lock,
1738  * we drop it here. Return 0 if we got the lock, -EAGAIN if node is
1739  * still alive (we couldn't get the lock) and < 0 on error. */
1740 static int ocfs2_trylock_journal(struct ocfs2_super *osb,
1741                                  int slot_num)
1742 {
1743         int status, flags;
1744         struct inode *inode = NULL;
1745
1746         inode = ocfs2_get_system_file_inode(osb, JOURNAL_SYSTEM_INODE,
1747                                             slot_num);
1748         if (inode == NULL) {
1749                 mlog(ML_ERROR, "access error\n");
1750                 status = -EACCES;
1751                 goto bail;
1752         }
1753         if (is_bad_inode(inode)) {
1754                 mlog(ML_ERROR, "access error (bad inode)\n");
1755                 iput(inode);
1756                 inode = NULL;
1757                 status = -EACCES;
1758                 goto bail;
1759         }
1760         SET_INODE_JOURNAL(inode);
1761
1762         flags = OCFS2_META_LOCK_RECOVERY | OCFS2_META_LOCK_NOQUEUE;
1763         status = ocfs2_inode_lock_full(inode, NULL, 1, flags);
1764         if (status < 0) {
1765                 if (status != -EAGAIN)
1766                         mlog_errno(status);
1767                 goto bail;
1768         }
1769
1770         ocfs2_inode_unlock(inode, 1);
1771 bail:
1772         if (inode)
1773                 iput(inode);
1774
1775         return status;
1776 }
1777
1778 /* Call this underneath ocfs2_super_lock. It also assumes that the
1779  * slot info struct has been updated from disk. */
1780 int ocfs2_mark_dead_nodes(struct ocfs2_super *osb)
1781 {
1782         unsigned int node_num;
1783         int status, i;
1784         u32 gen;
1785         struct buffer_head *bh = NULL;
1786         struct ocfs2_dinode *di;
1787
1788         /* This is called with the super block cluster lock, so we
1789          * know that the slot map can't change underneath us. */
1790
1791         for (i = 0; i < osb->max_slots; i++) {
1792                 /* Read journal inode to get the recovery generation */
1793                 status = ocfs2_read_journal_inode(osb, i, &bh, NULL);
1794                 if (status) {
1795                         mlog_errno(status);
1796                         goto bail;
1797                 }
1798                 di = (struct ocfs2_dinode *)bh->b_data;
1799                 gen = ocfs2_get_recovery_generation(di);
1800                 brelse(bh);
1801                 bh = NULL;
1802
1803                 spin_lock(&osb->osb_lock);
1804                 osb->slot_recovery_generations[i] = gen;
1805
1806                 trace_ocfs2_mark_dead_nodes(i,
1807                                             osb->slot_recovery_generations[i]);
1808
1809                 if (i == osb->slot_num) {
1810                         spin_unlock(&osb->osb_lock);
1811                         continue;
1812                 }
1813
1814                 status = ocfs2_slot_to_node_num_locked(osb, i, &node_num);
1815                 if (status == -ENOENT) {
1816                         spin_unlock(&osb->osb_lock);
1817                         continue;
1818                 }
1819
1820                 if (__ocfs2_recovery_map_test(osb, node_num)) {
1821                         spin_unlock(&osb->osb_lock);
1822                         continue;
1823                 }
1824                 spin_unlock(&osb->osb_lock);
1825
1826                 /* Ok, we have a slot occupied by another node which
1827                  * is not in the recovery map. We trylock his journal
1828                  * file here to test if he's alive. */
1829                 status = ocfs2_trylock_journal(osb, i);
1830                 if (!status) {
1831                         /* Since we're called from mount, we know that
1832                          * the recovery thread can't race us on
1833                          * setting / checking the recovery bits. */
1834                         ocfs2_recovery_thread(osb, node_num);
1835                 } else if ((status < 0) && (status != -EAGAIN)) {
1836                         mlog_errno(status);
1837                         goto bail;
1838                 }
1839         }
1840
1841         status = 0;
1842 bail:
1843         return status;
1844 }
1845
1846 /*
1847  * Scan timer should get fired every ORPHAN_SCAN_SCHEDULE_TIMEOUT. Add some
1848  * randomness to the timeout to minimize multple nodes firing the timer at the
1849  * same time.
1850  */
1851 static inline unsigned long ocfs2_orphan_scan_timeout(void)
1852 {
1853         unsigned long time;
1854
1855         get_random_bytes(&time, sizeof(time));
1856         time = ORPHAN_SCAN_SCHEDULE_TIMEOUT + (time % 5000);
1857         return msecs_to_jiffies(time);
1858 }
1859
1860 /*
1861  * ocfs2_queue_orphan_scan calls ocfs2_queue_recovery_completion for
1862  * every slot, queuing a recovery of the slot on the ocfs2_wq thread. This
1863  * is done to catch any orphans that are left over in orphan directories.
1864  *
1865  * It scans all slots, even ones that are in use. It does so to handle the
1866  * case described below:
1867  *
1868  *   Node 1 has an inode it was using. The dentry went away due to memory
1869  *   pressure.  Node 1 closes the inode, but it's on the free list. The node
1870  *   has the open lock.
1871  *   Node 2 unlinks the inode. It grabs the dentry lock to notify others,
1872  *   but node 1 has no dentry and doesn't get the message. It trylocks the
1873  *   open lock, sees that another node has a PR, and does nothing.
1874  *   Later node 2 runs its orphan dir. It igets the inode, trylocks the
1875  *   open lock, sees the PR still, and does nothing.
1876  *   Basically, we have to trigger an orphan iput on node 1. The only way
1877  *   for this to happen is if node 1 runs node 2's orphan dir.
1878  *
1879  * ocfs2_queue_orphan_scan gets called every ORPHAN_SCAN_SCHEDULE_TIMEOUT
1880  * seconds.  It gets an EX lock on os_lockres and checks sequence number
1881  * stored in LVB. If the sequence number has changed, it means some other
1882  * node has done the scan.  This node skips the scan and tracks the
1883  * sequence number.  If the sequence number didn't change, it means a scan
1884  * hasn't happened.  The node queues a scan and increments the
1885  * sequence number in the LVB.
1886  */
1887 void ocfs2_queue_orphan_scan(struct ocfs2_super *osb)
1888 {
1889         struct ocfs2_orphan_scan *os;
1890         int status, i;
1891         u32 seqno = 0;
1892
1893         os = &osb->osb_orphan_scan;
1894
1895         if (atomic_read(&os->os_state) == ORPHAN_SCAN_INACTIVE)
1896                 goto out;
1897
1898         trace_ocfs2_queue_orphan_scan_begin(os->os_count, os->os_seqno,
1899                                             atomic_read(&os->os_state));
1900
1901         status = ocfs2_orphan_scan_lock(osb, &seqno);
1902         if (status < 0) {
1903                 if (status != -EAGAIN)
1904                         mlog_errno(status);
1905                 goto out;
1906         }
1907
1908         /* Do no queue the tasks if the volume is being umounted */
1909         if (atomic_read(&os->os_state) == ORPHAN_SCAN_INACTIVE)
1910                 goto unlock;
1911
1912         if (os->os_seqno != seqno) {
1913                 os->os_seqno = seqno;
1914                 goto unlock;
1915         }
1916
1917         for (i = 0; i < osb->max_slots; i++)
1918                 ocfs2_queue_recovery_completion(osb->journal, i, NULL, NULL,
1919                                                 NULL, ORPHAN_NO_NEED_TRUNCATE);
1920         /*
1921          * We queued a recovery on orphan slots, increment the sequence
1922          * number and update LVB so other node will skip the scan for a while
1923          */
1924         seqno++;
1925         os->os_count++;
1926         os->os_scantime = CURRENT_TIME;
1927 unlock:
1928         ocfs2_orphan_scan_unlock(osb, seqno);
1929 out:
1930         trace_ocfs2_queue_orphan_scan_end(os->os_count, os->os_seqno,
1931                                           atomic_read(&os->os_state));
1932         return;
1933 }
1934
1935 /* Worker task that gets fired every ORPHAN_SCAN_SCHEDULE_TIMEOUT millsec */
1936 void ocfs2_orphan_scan_work(struct work_struct *work)
1937 {
1938         struct ocfs2_orphan_scan *os;
1939         struct ocfs2_super *osb;
1940
1941         os = container_of(work, struct ocfs2_orphan_scan,
1942                           os_orphan_scan_work.work);
1943         osb = os->os_osb;
1944
1945         mutex_lock(&os->os_lock);
1946         ocfs2_queue_orphan_scan(osb);
1947         if (atomic_read(&os->os_state) == ORPHAN_SCAN_ACTIVE)
1948                 queue_delayed_work(ocfs2_wq, &os->os_orphan_scan_work,
1949                                       ocfs2_orphan_scan_timeout());
1950         mutex_unlock(&os->os_lock);
1951 }
1952
1953 void ocfs2_orphan_scan_stop(struct ocfs2_super *osb)
1954 {
1955         struct ocfs2_orphan_scan *os;
1956
1957         os = &osb->osb_orphan_scan;
1958         if (atomic_read(&os->os_state) == ORPHAN_SCAN_ACTIVE) {
1959                 atomic_set(&os->os_state, ORPHAN_SCAN_INACTIVE);
1960                 mutex_lock(&os->os_lock);
1961                 cancel_delayed_work(&os->os_orphan_scan_work);
1962                 mutex_unlock(&os->os_lock);
1963         }
1964 }
1965
1966 void ocfs2_orphan_scan_init(struct ocfs2_super *osb)
1967 {
1968         struct ocfs2_orphan_scan *os;
1969
1970         os = &osb->osb_orphan_scan;
1971         os->os_osb = osb;
1972         os->os_count = 0;
1973         os->os_seqno = 0;
1974         mutex_init(&os->os_lock);
1975         INIT_DELAYED_WORK(&os->os_orphan_scan_work, ocfs2_orphan_scan_work);
1976 }
1977
1978 void ocfs2_orphan_scan_start(struct ocfs2_super *osb)
1979 {
1980         struct ocfs2_orphan_scan *os;
1981
1982         os = &osb->osb_orphan_scan;
1983         os->os_scantime = CURRENT_TIME;
1984         if (ocfs2_is_hard_readonly(osb) || ocfs2_mount_local(osb))
1985                 atomic_set(&os->os_state, ORPHAN_SCAN_INACTIVE);
1986         else {
1987                 atomic_set(&os->os_state, ORPHAN_SCAN_ACTIVE);
1988                 queue_delayed_work(ocfs2_wq, &os->os_orphan_scan_work,
1989                                    ocfs2_orphan_scan_timeout());
1990         }
1991 }
1992
1993 struct ocfs2_orphan_filldir_priv {
1994         struct dir_context      ctx;
1995         struct inode            *head;
1996         struct ocfs2_super      *osb;
1997 };
1998
1999 static int ocfs2_orphan_filldir(struct dir_context *ctx, const char *name,
2000                                 int name_len, loff_t pos, u64 ino,
2001                                 unsigned type)
2002 {
2003         struct ocfs2_orphan_filldir_priv *p =
2004                 container_of(ctx, struct ocfs2_orphan_filldir_priv, ctx);
2005         struct inode *iter;
2006
2007         if (name_len == 1 && !strncmp(".", name, 1))
2008                 return 0;
2009         if (name_len == 2 && !strncmp("..", name, 2))
2010                 return 0;
2011
2012         /* Skip bad inodes so that recovery can continue */
2013         iter = ocfs2_iget(p->osb, ino,
2014                           OCFS2_FI_FLAG_ORPHAN_RECOVERY, 0);
2015         if (IS_ERR(iter))
2016                 return 0;
2017
2018         /* Skip inodes which are already added to recover list, since dio may
2019          * happen concurrently with unlink/rename */
2020         if (OCFS2_I(iter)->ip_next_orphan) {
2021                 iput(iter);
2022                 return 0;
2023         }
2024
2025         trace_ocfs2_orphan_filldir((unsigned long long)OCFS2_I(iter)->ip_blkno);
2026         /* No locking is required for the next_orphan queue as there
2027          * is only ever a single process doing orphan recovery. */
2028         OCFS2_I(iter)->ip_next_orphan = p->head;
2029         p->head = iter;
2030
2031         return 0;
2032 }
2033
2034 static int ocfs2_queue_orphans(struct ocfs2_super *osb,
2035                                int slot,
2036                                struct inode **head)
2037 {
2038         int status;
2039         struct inode *orphan_dir_inode = NULL;
2040         struct ocfs2_orphan_filldir_priv priv = {
2041                 .ctx.actor = ocfs2_orphan_filldir,
2042                 .osb = osb,
2043                 .head = *head
2044         };
2045
2046         orphan_dir_inode = ocfs2_get_system_file_inode(osb,
2047                                                        ORPHAN_DIR_SYSTEM_INODE,
2048                                                        slot);
2049         if  (!orphan_dir_inode) {
2050                 status = -ENOENT;
2051                 mlog_errno(status);
2052                 return status;
2053         }
2054
2055         mutex_lock(&orphan_dir_inode->i_mutex);
2056         status = ocfs2_inode_lock(orphan_dir_inode, NULL, 0);
2057         if (status < 0) {
2058                 mlog_errno(status);
2059                 goto out;
2060         }
2061
2062         status = ocfs2_dir_foreach(orphan_dir_inode, &priv.ctx);
2063         if (status) {
2064                 mlog_errno(status);
2065                 goto out_cluster;
2066         }
2067
2068         *head = priv.head;
2069
2070 out_cluster:
2071         ocfs2_inode_unlock(orphan_dir_inode, 0);
2072 out:
2073         mutex_unlock(&orphan_dir_inode->i_mutex);
2074         iput(orphan_dir_inode);
2075         return status;
2076 }
2077
2078 static int ocfs2_orphan_recovery_can_continue(struct ocfs2_super *osb,
2079                                               int slot)
2080 {
2081         int ret;
2082
2083         spin_lock(&osb->osb_lock);
2084         ret = !osb->osb_orphan_wipes[slot];
2085         spin_unlock(&osb->osb_lock);
2086         return ret;
2087 }
2088
2089 static void ocfs2_mark_recovering_orphan_dir(struct ocfs2_super *osb,
2090                                              int slot)
2091 {
2092         spin_lock(&osb->osb_lock);
2093         /* Mark ourselves such that new processes in delete_inode()
2094          * know to quit early. */
2095         ocfs2_node_map_set_bit(osb, &osb->osb_recovering_orphan_dirs, slot);
2096         while (osb->osb_orphan_wipes[slot]) {
2097                 /* If any processes are already in the middle of an
2098                  * orphan wipe on this dir, then we need to wait for
2099                  * them. */
2100                 spin_unlock(&osb->osb_lock);
2101                 wait_event_interruptible(osb->osb_wipe_event,
2102                                          ocfs2_orphan_recovery_can_continue(osb, slot));
2103                 spin_lock(&osb->osb_lock);
2104         }
2105         spin_unlock(&osb->osb_lock);
2106 }
2107
2108 static void ocfs2_clear_recovering_orphan_dir(struct ocfs2_super *osb,
2109                                               int slot)
2110 {
2111         ocfs2_node_map_clear_bit(osb, &osb->osb_recovering_orphan_dirs, slot);
2112 }
2113
2114 /*
2115  * Orphan recovery. Each mounted node has it's own orphan dir which we
2116  * must run during recovery. Our strategy here is to build a list of
2117  * the inodes in the orphan dir and iget/iput them. The VFS does
2118  * (most) of the rest of the work.
2119  *
2120  * Orphan recovery can happen at any time, not just mount so we have a
2121  * couple of extra considerations.
2122  *
2123  * - We grab as many inodes as we can under the orphan dir lock -
2124  *   doing iget() outside the orphan dir risks getting a reference on
2125  *   an invalid inode.
2126  * - We must be sure not to deadlock with other processes on the
2127  *   system wanting to run delete_inode(). This can happen when they go
2128  *   to lock the orphan dir and the orphan recovery process attempts to
2129  *   iget() inside the orphan dir lock. This can be avoided by
2130  *   advertising our state to ocfs2_delete_inode().
2131  */
2132 static int ocfs2_recover_orphans(struct ocfs2_super *osb,
2133                                  int slot,
2134                                  enum ocfs2_orphan_reco_type orphan_reco_type)
2135 {
2136         int ret = 0;
2137         struct inode *inode = NULL;
2138         struct inode *iter;
2139         struct ocfs2_inode_info *oi;
2140
2141         trace_ocfs2_recover_orphans(slot);
2142
2143         ocfs2_mark_recovering_orphan_dir(osb, slot);
2144         ret = ocfs2_queue_orphans(osb, slot, &inode);
2145         ocfs2_clear_recovering_orphan_dir(osb, slot);
2146
2147         /* Error here should be noted, but we want to continue with as
2148          * many queued inodes as we've got. */
2149         if (ret)
2150                 mlog_errno(ret);
2151
2152         while (inode) {
2153                 oi = OCFS2_I(inode);
2154                 trace_ocfs2_recover_orphans_iput(
2155                                         (unsigned long long)oi->ip_blkno);
2156
2157                 iter = oi->ip_next_orphan;
2158                 oi->ip_next_orphan = NULL;
2159
2160                 /*
2161                  * We need to take and drop the inode lock to
2162                  * force read inode from disk.
2163                  */
2164                 ret = ocfs2_inode_lock(inode, NULL, 0);
2165                 if (ret) {
2166                         mlog_errno(ret);
2167                         goto next;
2168                 }
2169                 ocfs2_inode_unlock(inode, 0);
2170
2171                 if (inode->i_nlink == 0) {
2172                         spin_lock(&oi->ip_lock);
2173                         /* Set the proper information to get us going into
2174                          * ocfs2_delete_inode. */
2175                         oi->ip_flags |= OCFS2_INODE_MAYBE_ORPHANED;
2176                         spin_unlock(&oi->ip_lock);
2177                 } else if (orphan_reco_type == ORPHAN_NEED_TRUNCATE) {
2178                         struct buffer_head *di_bh = NULL;
2179
2180                         ret = ocfs2_rw_lock(inode, 1);
2181                         if (ret) {
2182                                 mlog_errno(ret);
2183                                 goto next;
2184                         }
2185
2186                         ret = ocfs2_inode_lock(inode, &di_bh, 1);
2187                         if (ret < 0) {
2188                                 ocfs2_rw_unlock(inode, 1);
2189                                 mlog_errno(ret);
2190                                 goto next;
2191                         }
2192
2193                         ret = ocfs2_truncate_file(inode, di_bh,
2194                                         i_size_read(inode));
2195                         ocfs2_inode_unlock(inode, 1);
2196                         ocfs2_rw_unlock(inode, 1);
2197                         brelse(di_bh);
2198                         if (ret < 0) {
2199                                 if (ret != -ENOSPC)
2200                                         mlog_errno(ret);
2201                                 goto next;
2202                         }
2203
2204                         ret = ocfs2_del_inode_from_orphan(osb, inode, 0, 0);
2205                         if (ret)
2206                                 mlog_errno(ret);
2207
2208                         wake_up(&OCFS2_I(inode)->append_dio_wq);
2209                 } /* else if ORPHAN_NO_NEED_TRUNCATE, do nothing */
2210
2211 next:
2212                 iput(inode);
2213
2214                 inode = iter;
2215         }
2216
2217         return ret;
2218 }
2219
2220 static int __ocfs2_wait_on_mount(struct ocfs2_super *osb, int quota)
2221 {
2222         /* This check is good because ocfs2 will wait on our recovery
2223          * thread before changing it to something other than MOUNTED
2224          * or DISABLED. */
2225         wait_event(osb->osb_mount_event,
2226                   (!quota && atomic_read(&osb->vol_state) == VOLUME_MOUNTED) ||
2227                    atomic_read(&osb->vol_state) == VOLUME_MOUNTED_QUOTAS ||
2228                    atomic_read(&osb->vol_state) == VOLUME_DISABLED);
2229
2230         /* If there's an error on mount, then we may never get to the
2231          * MOUNTED flag, but this is set right before
2232          * dismount_volume() so we can trust it. */
2233         if (atomic_read(&osb->vol_state) == VOLUME_DISABLED) {
2234                 trace_ocfs2_wait_on_mount(VOLUME_DISABLED);
2235                 mlog(0, "mount error, exiting!\n");
2236                 return -EBUSY;
2237         }
2238
2239         return 0;
2240 }
2241
2242 static int ocfs2_commit_thread(void *arg)
2243 {
2244         int status;
2245         struct ocfs2_super *osb = arg;
2246         struct ocfs2_journal *journal = osb->journal;
2247
2248         /* we can trust j_num_trans here because _should_stop() is only set in
2249          * shutdown and nobody other than ourselves should be able to start
2250          * transactions.  committing on shutdown might take a few iterations
2251          * as final transactions put deleted inodes on the list */
2252         while (!(kthread_should_stop() &&
2253                  atomic_read(&journal->j_num_trans) == 0)) {
2254
2255                 wait_event_interruptible(osb->checkpoint_event,
2256                                          atomic_read(&journal->j_num_trans)
2257                                          || kthread_should_stop());
2258
2259                 status = ocfs2_commit_cache(osb);
2260                 if (status < 0) {
2261                         static unsigned long abort_warn_time;
2262
2263                         /* Warn about this once per minute */
2264                         if (printk_timed_ratelimit(&abort_warn_time, 60*HZ))
2265                                 mlog(ML_ERROR, "status = %d, journal is "
2266                                                 "already aborted.\n", status);
2267                         /*
2268                          * After ocfs2_commit_cache() fails, j_num_trans has a
2269                          * non-zero value.  Sleep here to avoid a busy-wait
2270                          * loop.
2271                          */
2272                         msleep_interruptible(1000);
2273                 }
2274
2275                 if (kthread_should_stop() && atomic_read(&journal->j_num_trans)){
2276                         mlog(ML_KTHREAD,
2277                              "commit_thread: %u transactions pending on "
2278                              "shutdown\n",
2279                              atomic_read(&journal->j_num_trans));
2280                 }
2281         }
2282
2283         return 0;
2284 }
2285
2286 /* Reads all the journal inodes without taking any cluster locks. Used
2287  * for hard readonly access to determine whether any journal requires
2288  * recovery. Also used to refresh the recovery generation numbers after
2289  * a journal has been recovered by another node.
2290  */
2291 int ocfs2_check_journals_nolocks(struct ocfs2_super *osb)
2292 {
2293         int ret = 0;
2294         unsigned int slot;
2295         struct buffer_head *di_bh = NULL;
2296         struct ocfs2_dinode *di;
2297         int journal_dirty = 0;
2298
2299         for(slot = 0; slot < osb->max_slots; slot++) {
2300                 ret = ocfs2_read_journal_inode(osb, slot, &di_bh, NULL);
2301                 if (ret) {
2302                         mlog_errno(ret);
2303                         goto out;
2304                 }
2305
2306                 di = (struct ocfs2_dinode *) di_bh->b_data;
2307
2308                 osb->slot_recovery_generations[slot] =
2309                                         ocfs2_get_recovery_generation(di);
2310
2311                 if (le32_to_cpu(di->id1.journal1.ij_flags) &
2312                     OCFS2_JOURNAL_DIRTY_FL)
2313                         journal_dirty = 1;
2314
2315                 brelse(di_bh);
2316                 di_bh = NULL;
2317         }
2318
2319 out:
2320         if (journal_dirty)
2321                 ret = -EROFS;
2322         return ret;
2323 }