~ [ source navigation ] ~ [ diff markup ] ~ [ identifier search ] ~

TOMOYO Linux Cross Reference
Linux/fs/ocfs2/journal.c

Version: ~ [ linux-5.3-rc5 ] ~ [ linux-5.2.9 ] ~ [ linux-5.1.21 ] ~ [ linux-5.0.21 ] ~ [ linux-4.20.17 ] ~ [ linux-4.19.67 ] ~ [ linux-4.18.20 ] ~ [ linux-4.17.19 ] ~ [ linux-4.16.18 ] ~ [ linux-4.15.18 ] ~ [ linux-4.14.139 ] ~ [ linux-4.13.16 ] ~ [ linux-4.12.14 ] ~ [ linux-4.11.12 ] ~ [ linux-4.10.17 ] ~ [ linux-4.9.189 ] ~ [ linux-4.8.17 ] ~ [ linux-4.7.10 ] ~ [ linux-4.6.7 ] ~ [ linux-4.5.7 ] ~ [ linux-4.4.189 ] ~ [ linux-4.3.6 ] ~ [ linux-4.2.8 ] ~ [ linux-4.1.52 ] ~ [ linux-4.0.9 ] ~ [ linux-3.19.8 ] ~ [ linux-3.18.140 ] ~ [ linux-3.17.8 ] ~ [ linux-3.16.72 ] ~ [ linux-3.15.10 ] ~ [ linux-3.14.79 ] ~ [ linux-3.13.11 ] ~ [ linux-3.12.74 ] ~ [ linux-3.11.10 ] ~ [ linux-3.10.108 ] ~ [ linux-3.9.11 ] ~ [ linux-3.8.13 ] ~ [ linux-3.7.10 ] ~ [ linux-3.6.11 ] ~ [ linux-3.5.7 ] ~ [ linux-3.4.113 ] ~ [ linux-3.3.8 ] ~ [ linux-3.2.102 ] ~ [ linux-3.1.10 ] ~ [ linux-3.0.101 ] ~ [ linux-2.6.32.71 ] ~ [ linux-2.6.0 ] ~ [ linux-2.4.37.11 ] ~ [ unix-v6-master ] ~ [ ccs-tools-1.8.5 ] ~ [ policy-sample ] ~
Architecture: ~ [ i386 ] ~ [ alpha ] ~ [ m68k ] ~ [ mips ] ~ [ ppc ] ~ [ sparc ] ~ [ sparc64 ] ~

  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 
 34 #include <cluster/masklog.h>
 35 
 36 #include "ocfs2.h"
 37 
 38 #include "alloc.h"
 39 #include "blockcheck.h"
 40 #include "dir.h"
 41 #include "dlmglue.h"
 42 #include "extent_map.h"
 43 #include "heartbeat.h"
 44 #include "inode.h"
 45 #include "journal.h"
 46 #include "localalloc.h"
 47 #include "slot_map.h"
 48 #include "super.h"
 49 #include "sysfile.h"
 50 #include "uptodate.h"
 51 #include "quota.h"
 52 
 53 #include "buffer_head_io.h"
 54 #include "ocfs2_trace.h"
 55 
 56 DEFINE_SPINLOCK(trans_inc_lock);
 57 
 58 #define ORPHAN_SCAN_SCHEDULE_TIMEOUT 300000
 59 
 60 static int ocfs2_force_read_journal(struct inode *inode);
 61 static int ocfs2_recover_node(struct ocfs2_super *osb,
 62                               int node_num, int slot_num);
 63 static int __ocfs2_recovery_thread(void *arg);
 64 static int ocfs2_commit_cache(struct ocfs2_super *osb);
 65 static int __ocfs2_wait_on_mount(struct ocfs2_super *osb, int quota);
 66 static int ocfs2_journal_toggle_dirty(struct ocfs2_super *osb,
 67                                       int dirty, int replayed);
 68 static int ocfs2_trylock_journal(struct ocfs2_super *osb,
 69                                  int slot_num);
 70 static int ocfs2_recover_orphans(struct ocfs2_super *osb,
 71                                  int slot);
 72 static int ocfs2_commit_thread(void *arg);
 73 static void ocfs2_queue_recovery_completion(struct ocfs2_journal *journal,
 74                                             int slot_num,
 75                                             struct ocfs2_dinode *la_dinode,
 76                                             struct ocfs2_dinode *tl_dinode,
 77                                             struct ocfs2_quota_recovery *qrec);
 78 
 79 static inline int ocfs2_wait_on_mount(struct ocfs2_super *osb)
 80 {
 81         return __ocfs2_wait_on_mount(osb, 0);
 82 }
 83 
 84 static inline int ocfs2_wait_on_quotas(struct ocfs2_super *osb)
 85 {
 86         return __ocfs2_wait_on_mount(osb, 1);
 87 }
 88 
 89 /*
 90  * This replay_map is to track online/offline slots, so we could recover
 91  * offline slots during recovery and mount
 92  */
 93 
 94 enum ocfs2_replay_state {
 95         REPLAY_UNNEEDED = 0,    /* Replay is not needed, so ignore this map */
 96         REPLAY_NEEDED,          /* Replay slots marked in rm_replay_slots */
 97         REPLAY_DONE             /* Replay was already queued */
 98 };
 99 
100 struct ocfs2_replay_map {
101         unsigned int rm_slots;
102         enum ocfs2_replay_state rm_state;
103         unsigned char rm_replay_slots[0];
104 };
105 
106 void ocfs2_replay_map_set_state(struct ocfs2_super *osb, int state)
107 {
108         if (!osb->replay_map)
109                 return;
110 
111         /* If we've already queued the replay, we don't have any more to do */
112         if (osb->replay_map->rm_state == REPLAY_DONE)
113                 return;
114 
115         osb->replay_map->rm_state = state;
116 }
117 
118 int ocfs2_compute_replay_slots(struct ocfs2_super *osb)
119 {
120         struct ocfs2_replay_map *replay_map;
121         int i, node_num;
122 
123         /* If replay map is already set, we don't do it again */
124         if (osb->replay_map)
125                 return 0;
126 
127         replay_map = kzalloc(sizeof(struct ocfs2_replay_map) +
128                              (osb->max_slots * sizeof(char)), GFP_KERNEL);
129 
130         if (!replay_map) {
131                 mlog_errno(-ENOMEM);
132                 return -ENOMEM;
133         }
134 
135         spin_lock(&osb->osb_lock);
136 
137         replay_map->rm_slots = osb->max_slots;
138         replay_map->rm_state = REPLAY_UNNEEDED;
139 
140         /* set rm_replay_slots for offline slot(s) */
141         for (i = 0; i < replay_map->rm_slots; i++) {
142                 if (ocfs2_slot_to_node_num_locked(osb, i, &node_num) == -ENOENT)
143                         replay_map->rm_replay_slots[i] = 1;
144         }
145 
146         osb->replay_map = replay_map;
147         spin_unlock(&osb->osb_lock);
148         return 0;
149 }
150 
151 void ocfs2_queue_replay_slots(struct ocfs2_super *osb)
152 {
153         struct ocfs2_replay_map *replay_map = osb->replay_map;
154         int i;
155 
156         if (!replay_map)
157                 return;
158 
159         if (replay_map->rm_state != REPLAY_NEEDED)
160                 return;
161 
162         for (i = 0; i < replay_map->rm_slots; i++)
163                 if (replay_map->rm_replay_slots[i])
164                         ocfs2_queue_recovery_completion(osb->journal, i, NULL,
165                                                         NULL, NULL);
166         replay_map->rm_state = REPLAY_DONE;
167 }
168 
169 void ocfs2_free_replay_slots(struct ocfs2_super *osb)
170 {
171         struct ocfs2_replay_map *replay_map = osb->replay_map;
172 
173         if (!osb->replay_map)
174                 return;
175 
176         kfree(replay_map);
177         osb->replay_map = NULL;
178 }
179 
180 int ocfs2_recovery_init(struct ocfs2_super *osb)
181 {
182         struct ocfs2_recovery_map *rm;
183 
184         mutex_init(&osb->recovery_lock);
185         osb->disable_recovery = 0;
186         osb->recovery_thread_task = NULL;
187         init_waitqueue_head(&osb->recovery_event);
188 
189         rm = kzalloc(sizeof(struct ocfs2_recovery_map) +
190                      osb->max_slots * sizeof(unsigned int),
191                      GFP_KERNEL);
192         if (!rm) {
193                 mlog_errno(-ENOMEM);
194                 return -ENOMEM;
195         }
196 
197         rm->rm_entries = (unsigned int *)((char *)rm +
198                                           sizeof(struct ocfs2_recovery_map));
199         osb->recovery_map = rm;
200 
201         return 0;
202 }
203 
204 /* we can't grab the goofy sem lock from inside wait_event, so we use
205  * memory barriers to make sure that we'll see the null task before
206  * being woken up */
207 static int ocfs2_recovery_thread_running(struct ocfs2_super *osb)
208 {
209         mb();
210         return osb->recovery_thread_task != NULL;
211 }
212 
213 void ocfs2_recovery_exit(struct ocfs2_super *osb)
214 {
215         struct ocfs2_recovery_map *rm;
216 
217         /* disable any new recovery threads and wait for any currently
218          * running ones to exit. Do this before setting the vol_state. */
219         mutex_lock(&osb->recovery_lock);
220         osb->disable_recovery = 1;
221         mutex_unlock(&osb->recovery_lock);
222         wait_event(osb->recovery_event, !ocfs2_recovery_thread_running(osb));
223 
224         /* At this point, we know that no more recovery threads can be
225          * launched, so wait for any recovery completion work to
226          * complete. */
227         flush_workqueue(ocfs2_wq);
228 
229         /*
230          * Now that recovery is shut down, and the osb is about to be
231          * freed,  the osb_lock is not taken here.
232          */
233         rm = osb->recovery_map;
234         /* XXX: Should we bug if there are dirty entries? */
235 
236         kfree(rm);
237 }
238 
239 static int __ocfs2_recovery_map_test(struct ocfs2_super *osb,
240                                      unsigned int node_num)
241 {
242         int i;
243         struct ocfs2_recovery_map *rm = osb->recovery_map;
244 
245         assert_spin_locked(&osb->osb_lock);
246 
247         for (i = 0; i < rm->rm_used; i++) {
248                 if (rm->rm_entries[i] == node_num)
249                         return 1;
250         }
251 
252         return 0;
253 }
254 
255 /* Behaves like test-and-set.  Returns the previous value */
256 static int ocfs2_recovery_map_set(struct ocfs2_super *osb,
257                                   unsigned int node_num)
258 {
259         struct ocfs2_recovery_map *rm = osb->recovery_map;
260 
261         spin_lock(&osb->osb_lock);
262         if (__ocfs2_recovery_map_test(osb, node_num)) {
263                 spin_unlock(&osb->osb_lock);
264                 return 1;
265         }
266 
267         /* XXX: Can this be exploited? Not from o2dlm... */
268         BUG_ON(rm->rm_used >= osb->max_slots);
269 
270         rm->rm_entries[rm->rm_used] = node_num;
271         rm->rm_used++;
272         spin_unlock(&osb->osb_lock);
273 
274         return 0;
275 }
276 
277 static void ocfs2_recovery_map_clear(struct ocfs2_super *osb,
278                                      unsigned int node_num)
279 {
280         int i;
281         struct ocfs2_recovery_map *rm = osb->recovery_map;
282 
283         spin_lock(&osb->osb_lock);
284 
285         for (i = 0; i < rm->rm_used; i++) {
286                 if (rm->rm_entries[i] == node_num)
287                         break;
288         }
289 
290         if (i < rm->rm_used) {
291                 /* XXX: be careful with the pointer math */
292                 memmove(&(rm->rm_entries[i]), &(rm->rm_entries[i + 1]),
293                         (rm->rm_used - i - 1) * sizeof(unsigned int));
294                 rm->rm_used--;
295         }
296 
297         spin_unlock(&osb->osb_lock);
298 }
299 
300 static int ocfs2_commit_cache(struct ocfs2_super *osb)
301 {
302         int status = 0;
303         unsigned int flushed;
304         struct ocfs2_journal *journal = NULL;
305 
306         journal = osb->journal;
307 
308         /* Flush all pending commits and checkpoint the journal. */
309         down_write(&journal->j_trans_barrier);
310 
311         flushed = atomic_read(&journal->j_num_trans);
312         trace_ocfs2_commit_cache_begin(flushed);
313         if (flushed == 0) {
314                 up_write(&journal->j_trans_barrier);
315                 goto finally;
316         }
317 
318         jbd2_journal_lock_updates(journal->j_journal);
319         status = jbd2_journal_flush(journal->j_journal);
320         jbd2_journal_unlock_updates(journal->j_journal);
321         if (status < 0) {
322                 up_write(&journal->j_trans_barrier);
323                 mlog_errno(status);
324                 goto finally;
325         }
326 
327         ocfs2_inc_trans_id(journal);
328 
329         flushed = atomic_read(&journal->j_num_trans);
330         atomic_set(&journal->j_num_trans, 0);
331         up_write(&journal->j_trans_barrier);
332 
333         trace_ocfs2_commit_cache_end(journal->j_trans_id, flushed);
334 
335         ocfs2_wake_downconvert_thread(osb);
336         wake_up(&journal->j_checkpointed);
337 finally:
338         return status;
339 }
340 
341 handle_t *ocfs2_start_trans(struct ocfs2_super *osb, int max_buffs)
342 {
343         journal_t *journal = osb->journal->j_journal;
344         handle_t *handle;
345 
346         BUG_ON(!osb || !osb->journal->j_journal);
347 
348         if (ocfs2_is_hard_readonly(osb))
349                 return ERR_PTR(-EROFS);
350 
351         BUG_ON(osb->journal->j_state == OCFS2_JOURNAL_FREE);
352         BUG_ON(max_buffs <= 0);
353 
354         /* Nested transaction? Just return the handle... */
355         if (journal_current_handle())
356                 return jbd2_journal_start(journal, max_buffs);
357 
358         sb_start_intwrite(osb->sb);
359 
360         down_read(&osb->journal->j_trans_barrier);
361 
362         handle = jbd2_journal_start(journal, max_buffs);
363         if (IS_ERR(handle)) {
364                 up_read(&osb->journal->j_trans_barrier);
365                 sb_end_intwrite(osb->sb);
366 
367                 mlog_errno(PTR_ERR(handle));
368 
369                 if (is_journal_aborted(journal)) {
370                         ocfs2_abort(osb->sb, "Detected aborted journal");
371                         handle = ERR_PTR(-EROFS);
372                 }
373         } else {
374                 if (!ocfs2_mount_local(osb))
375                         atomic_inc(&(osb->journal->j_num_trans));
376         }
377 
378         return handle;
379 }
380 
381 int ocfs2_commit_trans(struct ocfs2_super *osb,
382                        handle_t *handle)
383 {
384         int ret, nested;
385         struct ocfs2_journal *journal = osb->journal;
386 
387         BUG_ON(!handle);
388 
389         nested = handle->h_ref > 1;
390         ret = jbd2_journal_stop(handle);
391         if (ret < 0)
392                 mlog_errno(ret);
393 
394         if (!nested) {
395                 up_read(&journal->j_trans_barrier);
396                 sb_end_intwrite(osb->sb);
397         }
398 
399         return ret;
400 }
401 
402 /*
403  * 'nblocks' is what you want to add to the current transaction.
404  *
405  * This might call jbd2_journal_restart() which will commit dirty buffers
406  * and then restart the transaction. Before calling
407  * ocfs2_extend_trans(), any changed blocks should have been
408  * dirtied. After calling it, all blocks which need to be changed must
409  * go through another set of journal_access/journal_dirty calls.
410  *
411  * WARNING: This will not release any semaphores or disk locks taken
412  * during the transaction, so make sure they were taken *before*
413  * start_trans or we'll have ordering deadlocks.
414  *
415  * WARNING2: Note that we do *not* drop j_trans_barrier here. This is
416  * good because transaction ids haven't yet been recorded on the
417  * cluster locks associated with this handle.
418  */
419 int ocfs2_extend_trans(handle_t *handle, int nblocks)
420 {
421         int status, old_nblocks;
422 
423         BUG_ON(!handle);
424         BUG_ON(nblocks < 0);
425 
426         if (!nblocks)
427                 return 0;
428 
429         old_nblocks = handle->h_buffer_credits;
430 
431         trace_ocfs2_extend_trans(old_nblocks, nblocks);
432 
433 #ifdef CONFIG_OCFS2_DEBUG_FS
434         status = 1;
435 #else
436         status = jbd2_journal_extend(handle, nblocks);
437         if (status < 0) {
438                 mlog_errno(status);
439                 goto bail;
440         }
441 #endif
442 
443         if (status > 0) {
444                 trace_ocfs2_extend_trans_restart(old_nblocks + nblocks);
445                 status = jbd2_journal_restart(handle,
446                                               old_nblocks + nblocks);
447                 if (status < 0) {
448                         mlog_errno(status);
449                         goto bail;
450                 }
451         }
452 
453         status = 0;
454 bail:
455         return status;
456 }
457 
458 /*
459  * If we have fewer than thresh credits, extend by OCFS2_MAX_TRANS_DATA.
460  * If that fails, restart the transaction & regain write access for the
461  * buffer head which is used for metadata modifications.
462  * Taken from Ext4: extend_or_restart_transaction()
463  */
464 int ocfs2_allocate_extend_trans(handle_t *handle, int thresh)
465 {
466         int status, old_nblks;
467 
468         BUG_ON(!handle);
469 
470         old_nblks = handle->h_buffer_credits;
471         trace_ocfs2_allocate_extend_trans(old_nblks, thresh);
472 
473         if (old_nblks < thresh)
474                 return 0;
475 
476         status = jbd2_journal_extend(handle, OCFS2_MAX_TRANS_DATA);
477         if (status < 0) {
478                 mlog_errno(status);
479                 goto bail;
480         }
481 
482         if (status > 0) {
483                 status = jbd2_journal_restart(handle, OCFS2_MAX_TRANS_DATA);
484                 if (status < 0)
485                         mlog_errno(status);
486         }
487 
488 bail:
489         return status;
490 }
491 
492 
493 struct ocfs2_triggers {
494         struct jbd2_buffer_trigger_type ot_triggers;
495         int                             ot_offset;
496 };
497 
498 static inline struct ocfs2_triggers *to_ocfs2_trigger(struct jbd2_buffer_trigger_type *triggers)
499 {
500         return container_of(triggers, struct ocfs2_triggers, ot_triggers);
501 }
502 
503 static void ocfs2_frozen_trigger(struct jbd2_buffer_trigger_type *triggers,
504                                  struct buffer_head *bh,
505                                  void *data, size_t size)
506 {
507         struct ocfs2_triggers *ot = to_ocfs2_trigger(triggers);
508 
509         /*
510          * We aren't guaranteed to have the superblock here, so we
511          * must unconditionally compute the ecc data.
512          * __ocfs2_journal_access() will only set the triggers if
513          * metaecc is enabled.
514          */
515         ocfs2_block_check_compute(data, size, data + ot->ot_offset);
516 }
517 
518 /*
519  * Quota blocks have their own trigger because the struct ocfs2_block_check
520  * offset depends on the blocksize.
521  */
522 static void ocfs2_dq_frozen_trigger(struct jbd2_buffer_trigger_type *triggers,
523                                  struct buffer_head *bh,
524                                  void *data, size_t size)
525 {
526         struct ocfs2_disk_dqtrailer *dqt =
527                 ocfs2_block_dqtrailer(size, data);
528 
529         /*
530          * We aren't guaranteed to have the superblock here, so we
531          * must unconditionally compute the ecc data.
532          * __ocfs2_journal_access() will only set the triggers if
533          * metaecc is enabled.
534          */
535         ocfs2_block_check_compute(data, size, &dqt->dq_check);
536 }
537 
538 /*
539  * Directory blocks also have their own trigger because the
540  * struct ocfs2_block_check offset depends on the blocksize.
541  */
542 static void ocfs2_db_frozen_trigger(struct jbd2_buffer_trigger_type *triggers,
543                                  struct buffer_head *bh,
544                                  void *data, size_t size)
545 {
546         struct ocfs2_dir_block_trailer *trailer =
547                 ocfs2_dir_trailer_from_size(size, data);
548 
549         /*
550          * We aren't guaranteed to have the superblock here, so we
551          * must unconditionally compute the ecc data.
552          * __ocfs2_journal_access() will only set the triggers if
553          * metaecc is enabled.
554          */
555         ocfs2_block_check_compute(data, size, &trailer->db_check);
556 }
557 
558 static void ocfs2_abort_trigger(struct jbd2_buffer_trigger_type *triggers,
559                                 struct buffer_head *bh)
560 {
561         mlog(ML_ERROR,
562              "ocfs2_abort_trigger called by JBD2.  bh = 0x%lx, "
563              "bh->b_blocknr = %llu\n",
564              (unsigned long)bh,
565              (unsigned long long)bh->b_blocknr);
566 
567         /* We aren't guaranteed to have the superblock here - but if we
568          * don't, it'll just crash. */
569         ocfs2_error(bh->b_assoc_map->host->i_sb,
570                     "JBD2 has aborted our journal, ocfs2 cannot continue\n");
571 }
572 
573 static struct ocfs2_triggers di_triggers = {
574         .ot_triggers = {
575                 .t_frozen = ocfs2_frozen_trigger,
576                 .t_abort = ocfs2_abort_trigger,
577         },
578         .ot_offset      = offsetof(struct ocfs2_dinode, i_check),
579 };
580 
581 static struct ocfs2_triggers eb_triggers = {
582         .ot_triggers = {
583                 .t_frozen = ocfs2_frozen_trigger,
584                 .t_abort = ocfs2_abort_trigger,
585         },
586         .ot_offset      = offsetof(struct ocfs2_extent_block, h_check),
587 };
588 
589 static struct ocfs2_triggers rb_triggers = {
590         .ot_triggers = {
591                 .t_frozen = ocfs2_frozen_trigger,
592                 .t_abort = ocfs2_abort_trigger,
593         },
594         .ot_offset      = offsetof(struct ocfs2_refcount_block, rf_check),
595 };
596 
597 static struct ocfs2_triggers gd_triggers = {
598         .ot_triggers = {
599                 .t_frozen = ocfs2_frozen_trigger,
600                 .t_abort = ocfs2_abort_trigger,
601         },
602         .ot_offset      = offsetof(struct ocfs2_group_desc, bg_check),
603 };
604 
605 static struct ocfs2_triggers db_triggers = {
606         .ot_triggers = {
607                 .t_frozen = ocfs2_db_frozen_trigger,
608                 .t_abort = ocfs2_abort_trigger,
609         },
610 };
611 
612 static struct ocfs2_triggers xb_triggers = {
613         .ot_triggers = {
614                 .t_frozen = ocfs2_frozen_trigger,
615                 .t_abort = ocfs2_abort_trigger,
616         },
617         .ot_offset      = offsetof(struct ocfs2_xattr_block, xb_check),
618 };
619 
620 static struct ocfs2_triggers dq_triggers = {
621         .ot_triggers = {
622                 .t_frozen = ocfs2_dq_frozen_trigger,
623                 .t_abort = ocfs2_abort_trigger,
624         },
625 };
626 
627 static struct ocfs2_triggers dr_triggers = {
628         .ot_triggers = {
629                 .t_frozen = ocfs2_frozen_trigger,
630                 .t_abort = ocfs2_abort_trigger,
631         },
632         .ot_offset      = offsetof(struct ocfs2_dx_root_block, dr_check),
633 };
634 
635 static struct ocfs2_triggers dl_triggers = {
636         .ot_triggers = {
637                 .t_frozen = ocfs2_frozen_trigger,
638                 .t_abort = ocfs2_abort_trigger,
639         },
640         .ot_offset      = offsetof(struct ocfs2_dx_leaf, dl_check),
641 };
642 
643 static int __ocfs2_journal_access(handle_t *handle,
644                                   struct ocfs2_caching_info *ci,
645                                   struct buffer_head *bh,
646                                   struct ocfs2_triggers *triggers,
647                                   int type)
648 {
649         int status;
650         struct ocfs2_super *osb =
651                 OCFS2_SB(ocfs2_metadata_cache_get_super(ci));
652 
653         BUG_ON(!ci || !ci->ci_ops);
654         BUG_ON(!handle);
655         BUG_ON(!bh);
656 
657         trace_ocfs2_journal_access(
658                 (unsigned long long)ocfs2_metadata_cache_owner(ci),
659                 (unsigned long long)bh->b_blocknr, type, bh->b_size);
660 
661         /* we can safely remove this assertion after testing. */
662         if (!buffer_uptodate(bh)) {
663                 mlog(ML_ERROR, "giving me a buffer that's not uptodate!\n");
664                 mlog(ML_ERROR, "b_blocknr=%llu\n",
665                      (unsigned long long)bh->b_blocknr);
666                 BUG();
667         }
668 
669         /* Set the current transaction information on the ci so
670          * that the locking code knows whether it can drop it's locks
671          * on this ci or not. We're protected from the commit
672          * thread updating the current transaction id until
673          * ocfs2_commit_trans() because ocfs2_start_trans() took
674          * j_trans_barrier for us. */
675         ocfs2_set_ci_lock_trans(osb->journal, ci);
676 
677         ocfs2_metadata_cache_io_lock(ci);
678         switch (type) {
679         case OCFS2_JOURNAL_ACCESS_CREATE:
680         case OCFS2_JOURNAL_ACCESS_WRITE:
681                 status = jbd2_journal_get_write_access(handle, bh);
682                 break;
683 
684         case OCFS2_JOURNAL_ACCESS_UNDO:
685                 status = jbd2_journal_get_undo_access(handle, bh);
686                 break;
687 
688         default:
689                 status = -EINVAL;
690                 mlog(ML_ERROR, "Unknown access type!\n");
691         }
692         if (!status && ocfs2_meta_ecc(osb) && triggers)
693                 jbd2_journal_set_triggers(bh, &triggers->ot_triggers);
694         ocfs2_metadata_cache_io_unlock(ci);
695 
696         if (status < 0)
697                 mlog(ML_ERROR, "Error %d getting %d access to buffer!\n",
698                      status, type);
699 
700         return status;
701 }
702 
703 int ocfs2_journal_access_di(handle_t *handle, struct ocfs2_caching_info *ci,
704                             struct buffer_head *bh, int type)
705 {
706         return __ocfs2_journal_access(handle, ci, bh, &di_triggers, type);
707 }
708 
709 int ocfs2_journal_access_eb(handle_t *handle, struct ocfs2_caching_info *ci,
710                             struct buffer_head *bh, int type)
711 {
712         return __ocfs2_journal_access(handle, ci, bh, &eb_triggers, type);
713 }
714 
715 int ocfs2_journal_access_rb(handle_t *handle, struct ocfs2_caching_info *ci,
716                             struct buffer_head *bh, int type)
717 {
718         return __ocfs2_journal_access(handle, ci, bh, &rb_triggers,
719                                       type);
720 }
721 
722 int ocfs2_journal_access_gd(handle_t *handle, struct ocfs2_caching_info *ci,
723                             struct buffer_head *bh, int type)
724 {
725         return __ocfs2_journal_access(handle, ci, bh, &gd_triggers, type);
726 }
727 
728 int ocfs2_journal_access_db(handle_t *handle, struct ocfs2_caching_info *ci,
729                             struct buffer_head *bh, int type)
730 {
731         return __ocfs2_journal_access(handle, ci, bh, &db_triggers, type);
732 }
733 
734 int ocfs2_journal_access_xb(handle_t *handle, struct ocfs2_caching_info *ci,
735                             struct buffer_head *bh, int type)
736 {
737         return __ocfs2_journal_access(handle, ci, bh, &xb_triggers, type);
738 }
739 
740 int ocfs2_journal_access_dq(handle_t *handle, struct ocfs2_caching_info *ci,
741                             struct buffer_head *bh, int type)
742 {
743         return __ocfs2_journal_access(handle, ci, bh, &dq_triggers, type);
744 }
745 
746 int ocfs2_journal_access_dr(handle_t *handle, struct ocfs2_caching_info *ci,
747                             struct buffer_head *bh, int type)
748 {
749         return __ocfs2_journal_access(handle, ci, bh, &dr_triggers, type);
750 }
751 
752 int ocfs2_journal_access_dl(handle_t *handle, struct ocfs2_caching_info *ci,
753                             struct buffer_head *bh, int type)
754 {
755         return __ocfs2_journal_access(handle, ci, bh, &dl_triggers, type);
756 }
757 
758 int ocfs2_journal_access(handle_t *handle, struct ocfs2_caching_info *ci,
759                          struct buffer_head *bh, int type)
760 {
761         return __ocfs2_journal_access(handle, ci, bh, NULL, type);
762 }
763 
764 void ocfs2_journal_dirty(handle_t *handle, struct buffer_head *bh)
765 {
766         int status;
767 
768         trace_ocfs2_journal_dirty((unsigned long long)bh->b_blocknr);
769 
770         status = jbd2_journal_dirty_metadata(handle, bh);
771         BUG_ON(status);
772 }
773 
774 #define OCFS2_DEFAULT_COMMIT_INTERVAL   (HZ * JBD2_DEFAULT_MAX_COMMIT_AGE)
775 
776 void ocfs2_set_journal_params(struct ocfs2_super *osb)
777 {
778         journal_t *journal = osb->journal->j_journal;
779         unsigned long commit_interval = OCFS2_DEFAULT_COMMIT_INTERVAL;
780 
781         if (osb->osb_commit_interval)
782                 commit_interval = osb->osb_commit_interval;
783 
784         write_lock(&journal->j_state_lock);
785         journal->j_commit_interval = commit_interval;
786         if (osb->s_mount_opt & OCFS2_MOUNT_BARRIER)
787                 journal->j_flags |= JBD2_BARRIER;
788         else
789                 journal->j_flags &= ~JBD2_BARRIER;
790         write_unlock(&journal->j_state_lock);
791 }
792 
793 int ocfs2_journal_init(struct ocfs2_journal *journal, int *dirty)
794 {
795         int status = -1;
796         struct inode *inode = NULL; /* the journal inode */
797         journal_t *j_journal = NULL;
798         struct ocfs2_dinode *di = NULL;
799         struct buffer_head *bh = NULL;
800         struct ocfs2_super *osb;
801         int inode_lock = 0;
802 
803         BUG_ON(!journal);
804 
805         osb = journal->j_osb;
806 
807         /* already have the inode for our journal */
808         inode = ocfs2_get_system_file_inode(osb, JOURNAL_SYSTEM_INODE,
809                                             osb->slot_num);
810         if (inode == NULL) {
811                 status = -EACCES;
812                 mlog_errno(status);
813                 goto done;
814         }
815         if (is_bad_inode(inode)) {
816                 mlog(ML_ERROR, "access error (bad inode)\n");
817                 iput(inode);
818                 inode = NULL;
819                 status = -EACCES;
820                 goto done;
821         }
822 
823         SET_INODE_JOURNAL(inode);
824         OCFS2_I(inode)->ip_open_count++;
825 
826         /* Skip recovery waits here - journal inode metadata never
827          * changes in a live cluster so it can be considered an
828          * exception to the rule. */
829         status = ocfs2_inode_lock_full(inode, &bh, 1, OCFS2_META_LOCK_RECOVERY);
830         if (status < 0) {
831                 if (status != -ERESTARTSYS)
832                         mlog(ML_ERROR, "Could not get lock on journal!\n");
833                 goto done;
834         }
835 
836         inode_lock = 1;
837         di = (struct ocfs2_dinode *)bh->b_data;
838 
839         if (i_size_read(inode) <  OCFS2_MIN_JOURNAL_SIZE) {
840                 mlog(ML_ERROR, "Journal file size (%lld) is too small!\n",
841                      i_size_read(inode));
842                 status = -EINVAL;
843                 goto done;
844         }
845 
846         trace_ocfs2_journal_init(i_size_read(inode),
847                                  (unsigned long long)inode->i_blocks,
848                                  OCFS2_I(inode)->ip_clusters);
849 
850         /* call the kernels journal init function now */
851         j_journal = jbd2_journal_init_inode(inode);
852         if (j_journal == NULL) {
853                 mlog(ML_ERROR, "Linux journal layer error\n");
854                 status = -EINVAL;
855                 goto done;
856         }
857 
858         trace_ocfs2_journal_init_maxlen(j_journal->j_maxlen);
859 
860         *dirty = (le32_to_cpu(di->id1.journal1.ij_flags) &
861                   OCFS2_JOURNAL_DIRTY_FL);
862 
863         journal->j_journal = j_journal;
864         journal->j_inode = inode;
865         journal->j_bh = bh;
866 
867         ocfs2_set_journal_params(osb);
868 
869         journal->j_state = OCFS2_JOURNAL_LOADED;
870 
871         status = 0;
872 done:
873         if (status < 0) {
874                 if (inode_lock)
875                         ocfs2_inode_unlock(inode, 1);
876                 brelse(bh);
877                 if (inode) {
878                         OCFS2_I(inode)->ip_open_count--;
879                         iput(inode);
880                 }
881         }
882 
883         return status;
884 }
885 
886 static void ocfs2_bump_recovery_generation(struct ocfs2_dinode *di)
887 {
888         le32_add_cpu(&(di->id1.journal1.ij_recovery_generation), 1);
889 }
890 
891 static u32 ocfs2_get_recovery_generation(struct ocfs2_dinode *di)
892 {
893         return le32_to_cpu(di->id1.journal1.ij_recovery_generation);
894 }
895 
896 static int ocfs2_journal_toggle_dirty(struct ocfs2_super *osb,
897                                       int dirty, int replayed)
898 {
899         int status;
900         unsigned int flags;
901         struct ocfs2_journal *journal = osb->journal;
902         struct buffer_head *bh = journal->j_bh;
903         struct ocfs2_dinode *fe;
904 
905         fe = (struct ocfs2_dinode *)bh->b_data;
906 
907         /* The journal bh on the osb always comes from ocfs2_journal_init()
908          * and was validated there inside ocfs2_inode_lock_full().  It's a
909          * code bug if we mess it up. */
910         BUG_ON(!OCFS2_IS_VALID_DINODE(fe));
911 
912         flags = le32_to_cpu(fe->id1.journal1.ij_flags);
913         if (dirty)
914                 flags |= OCFS2_JOURNAL_DIRTY_FL;
915         else
916                 flags &= ~OCFS2_JOURNAL_DIRTY_FL;
917         fe->id1.journal1.ij_flags = cpu_to_le32(flags);
918 
919         if (replayed)
920                 ocfs2_bump_recovery_generation(fe);
921 
922         ocfs2_compute_meta_ecc(osb->sb, bh->b_data, &fe->i_check);
923         status = ocfs2_write_block(osb, bh, INODE_CACHE(journal->j_inode));
924         if (status < 0)
925                 mlog_errno(status);
926 
927         return status;
928 }
929 
930 /*
931  * If the journal has been kmalloc'd it needs to be freed after this
932  * call.
933  */
934 void ocfs2_journal_shutdown(struct ocfs2_super *osb)
935 {
936         struct ocfs2_journal *journal = NULL;
937         int status = 0;
938         struct inode *inode = NULL;
939         int num_running_trans = 0;
940 
941         BUG_ON(!osb);
942 
943         journal = osb->journal;
944         if (!journal)
945                 goto done;
946 
947         inode = journal->j_inode;
948 
949         if (journal->j_state != OCFS2_JOURNAL_LOADED)
950                 goto done;
951 
952         /* need to inc inode use count - jbd2_journal_destroy will iput. */
953         if (!igrab(inode))
954                 BUG();
955 
956         num_running_trans = atomic_read(&(osb->journal->j_num_trans));
957         trace_ocfs2_journal_shutdown(num_running_trans);
958 
959         /* Do a commit_cache here. It will flush our journal, *and*
960          * release any locks that are still held.
961          * set the SHUTDOWN flag and release the trans lock.
962          * the commit thread will take the trans lock for us below. */
963         journal->j_state = OCFS2_JOURNAL_IN_SHUTDOWN;
964 
965         /* The OCFS2_JOURNAL_IN_SHUTDOWN will signal to commit_cache to not
966          * drop the trans_lock (which we want to hold until we
967          * completely destroy the journal. */
968         if (osb->commit_task) {
969                 /* Wait for the commit thread */
970                 trace_ocfs2_journal_shutdown_wait(osb->commit_task);
971                 kthread_stop(osb->commit_task);
972                 osb->commit_task = NULL;
973         }
974 
975         BUG_ON(atomic_read(&(osb->journal->j_num_trans)) != 0);
976 
977         if (ocfs2_mount_local(osb)) {
978                 jbd2_journal_lock_updates(journal->j_journal);
979                 status = jbd2_journal_flush(journal->j_journal);
980                 jbd2_journal_unlock_updates(journal->j_journal);
981                 if (status < 0)
982                         mlog_errno(status);
983         }
984 
985         if (status == 0) {
986                 /*
987                  * Do not toggle if flush was unsuccessful otherwise
988                  * will leave dirty metadata in a "clean" journal
989                  */
990                 status = ocfs2_journal_toggle_dirty(osb, 0, 0);
991                 if (status < 0)
992                         mlog_errno(status);
993         }
994 
995         /* Shutdown the kernel journal system */
996         jbd2_journal_destroy(journal->j_journal);
997         journal->j_journal = NULL;
998 
999         OCFS2_I(inode)->ip_open_count--;
1000 
1001         /* unlock our journal */
1002         ocfs2_inode_unlock(inode, 1);
1003 
1004         brelse(journal->j_bh);
1005         journal->j_bh = NULL;
1006 
1007         journal->j_state = OCFS2_JOURNAL_FREE;
1008 
1009 //      up_write(&journal->j_trans_barrier);
1010 done:
1011         if (inode)
1012                 iput(inode);
1013 }
1014 
1015 static void ocfs2_clear_journal_error(struct super_block *sb,
1016                                       journal_t *journal,
1017                                       int slot)
1018 {
1019         int olderr;
1020 
1021         olderr = jbd2_journal_errno(journal);
1022         if (olderr) {
1023                 mlog(ML_ERROR, "File system error %d recorded in "
1024                      "journal %u.\n", olderr, slot);
1025                 mlog(ML_ERROR, "File system on device %s needs checking.\n",
1026                      sb->s_id);
1027 
1028                 jbd2_journal_ack_err(journal);
1029                 jbd2_journal_clear_err(journal);
1030         }
1031 }
1032 
1033 int ocfs2_journal_load(struct ocfs2_journal *journal, int local, int replayed)
1034 {
1035         int status = 0;
1036         struct ocfs2_super *osb;
1037 
1038         BUG_ON(!journal);
1039 
1040         osb = journal->j_osb;
1041 
1042         status = jbd2_journal_load(journal->j_journal);
1043         if (status < 0) {
1044                 mlog(ML_ERROR, "Failed to load journal!\n");
1045                 goto done;
1046         }
1047 
1048         ocfs2_clear_journal_error(osb->sb, journal->j_journal, osb->slot_num);
1049 
1050         status = ocfs2_journal_toggle_dirty(osb, 1, replayed);
1051         if (status < 0) {
1052                 mlog_errno(status);
1053                 goto done;
1054         }
1055 
1056         /* Launch the commit thread */
1057         if (!local) {
1058                 osb->commit_task = kthread_run(ocfs2_commit_thread, osb,
1059                                                "ocfs2cmt");
1060                 if (IS_ERR(osb->commit_task)) {
1061                         status = PTR_ERR(osb->commit_task);
1062                         osb->commit_task = NULL;
1063                         mlog(ML_ERROR, "unable to launch ocfs2commit thread, "
1064                              "error=%d", status);
1065                         goto done;
1066                 }
1067         } else
1068                 osb->commit_task = NULL;
1069 
1070 done:
1071         return status;
1072 }
1073 
1074 
1075 /* 'full' flag tells us whether we clear out all blocks or if we just
1076  * mark the journal clean */
1077 int ocfs2_journal_wipe(struct ocfs2_journal *journal, int full)
1078 {
1079         int status;
1080 
1081         BUG_ON(!journal);
1082 
1083         status = jbd2_journal_wipe(journal->j_journal, full);
1084         if (status < 0) {
1085                 mlog_errno(status);
1086                 goto bail;
1087         }
1088 
1089         status = ocfs2_journal_toggle_dirty(journal->j_osb, 0, 0);
1090         if (status < 0)
1091                 mlog_errno(status);
1092 
1093 bail:
1094         return status;
1095 }
1096 
1097 static int ocfs2_recovery_completed(struct ocfs2_super *osb)
1098 {
1099         int empty;
1100         struct ocfs2_recovery_map *rm = osb->recovery_map;
1101 
1102         spin_lock(&osb->osb_lock);
1103         empty = (rm->rm_used == 0);
1104         spin_unlock(&osb->osb_lock);
1105 
1106         return empty;
1107 }
1108 
1109 void ocfs2_wait_for_recovery(struct ocfs2_super *osb)
1110 {
1111         wait_event(osb->recovery_event, ocfs2_recovery_completed(osb));
1112 }
1113 
1114 /*
1115  * JBD Might read a cached version of another nodes journal file. We
1116  * don't want this as this file changes often and we get no
1117  * notification on those changes. The only way to be sure that we've
1118  * got the most up to date version of those blocks then is to force
1119  * read them off disk. Just searching through the buffer cache won't
1120  * work as there may be pages backing this file which are still marked
1121  * up to date. We know things can't change on this file underneath us
1122  * as we have the lock by now :)
1123  */
1124 static int ocfs2_force_read_journal(struct inode *inode)
1125 {
1126         int status = 0;
1127         int i;
1128         u64 v_blkno, p_blkno, p_blocks, num_blocks;
1129 #define CONCURRENT_JOURNAL_FILL 32ULL
1130         struct buffer_head *bhs[CONCURRENT_JOURNAL_FILL];
1131 
1132         memset(bhs, 0, sizeof(struct buffer_head *) * CONCURRENT_JOURNAL_FILL);
1133 
1134         num_blocks = ocfs2_blocks_for_bytes(inode->i_sb, i_size_read(inode));
1135         v_blkno = 0;
1136         while (v_blkno < num_blocks) {
1137                 status = ocfs2_extent_map_get_blocks(inode, v_blkno,
1138                                                      &p_blkno, &p_blocks, NULL);
1139                 if (status < 0) {
1140                         mlog_errno(status);
1141                         goto bail;
1142                 }
1143 
1144                 if (p_blocks > CONCURRENT_JOURNAL_FILL)
1145                         p_blocks = CONCURRENT_JOURNAL_FILL;
1146 
1147                 /* We are reading journal data which should not
1148                  * be put in the uptodate cache */
1149                 status = ocfs2_read_blocks_sync(OCFS2_SB(inode->i_sb),
1150                                                 p_blkno, p_blocks, bhs);
1151                 if (status < 0) {
1152                         mlog_errno(status);
1153                         goto bail;
1154                 }
1155 
1156                 for(i = 0; i < p_blocks; i++) {
1157                         brelse(bhs[i]);
1158                         bhs[i] = NULL;
1159                 }
1160 
1161                 v_blkno += p_blocks;
1162         }
1163 
1164 bail:
1165         for(i = 0; i < CONCURRENT_JOURNAL_FILL; i++)
1166                 brelse(bhs[i]);
1167         return status;
1168 }
1169 
1170 struct ocfs2_la_recovery_item {
1171         struct list_head        lri_list;
1172         int                     lri_slot;
1173         struct ocfs2_dinode     *lri_la_dinode;
1174         struct ocfs2_dinode     *lri_tl_dinode;
1175         struct ocfs2_quota_recovery *lri_qrec;
1176 };
1177 
1178 /* Does the second half of the recovery process. By this point, the
1179  * node is marked clean and can actually be considered recovered,
1180  * hence it's no longer in the recovery map, but there's still some
1181  * cleanup we can do which shouldn't happen within the recovery thread
1182  * as locking in that context becomes very difficult if we are to take
1183  * recovering nodes into account.
1184  *
1185  * NOTE: This function can and will sleep on recovery of other nodes
1186  * during cluster locking, just like any other ocfs2 process.
1187  */
1188 void ocfs2_complete_recovery(struct work_struct *work)
1189 {
1190         int ret = 0;
1191         struct ocfs2_journal *journal =
1192                 container_of(work, struct ocfs2_journal, j_recovery_work);
1193         struct ocfs2_super *osb = journal->j_osb;
1194         struct ocfs2_dinode *la_dinode, *tl_dinode;
1195         struct ocfs2_la_recovery_item *item, *n;
1196         struct ocfs2_quota_recovery *qrec;
1197         LIST_HEAD(tmp_la_list);
1198 
1199         trace_ocfs2_complete_recovery(
1200                 (unsigned long long)OCFS2_I(journal->j_inode)->ip_blkno);
1201 
1202         spin_lock(&journal->j_lock);
1203         list_splice_init(&journal->j_la_cleanups, &tmp_la_list);
1204         spin_unlock(&journal->j_lock);
1205 
1206         list_for_each_entry_safe(item, n, &tmp_la_list, lri_list) {
1207                 list_del_init(&item->lri_list);
1208 
1209                 ocfs2_wait_on_quotas(osb);
1210 
1211                 la_dinode = item->lri_la_dinode;
1212                 tl_dinode = item->lri_tl_dinode;
1213                 qrec = item->lri_qrec;
1214 
1215                 trace_ocfs2_complete_recovery_slot(item->lri_slot,
1216                         la_dinode ? le64_to_cpu(la_dinode->i_blkno) : 0,
1217                         tl_dinode ? le64_to_cpu(tl_dinode->i_blkno) : 0,
1218                         qrec);
1219 
1220                 if (la_dinode) {
1221                         ret = ocfs2_complete_local_alloc_recovery(osb,
1222                                                                   la_dinode);
1223                         if (ret < 0)
1224                                 mlog_errno(ret);
1225 
1226                         kfree(la_dinode);
1227                 }
1228 
1229                 if (tl_dinode) {
1230                         ret = ocfs2_complete_truncate_log_recovery(osb,
1231                                                                    tl_dinode);
1232                         if (ret < 0)
1233                                 mlog_errno(ret);
1234 
1235                         kfree(tl_dinode);
1236                 }
1237 
1238                 ret = ocfs2_recover_orphans(osb, item->lri_slot);
1239                 if (ret < 0)
1240                         mlog_errno(ret);
1241 
1242                 if (qrec) {
1243                         ret = ocfs2_finish_quota_recovery(osb, qrec,
1244                                                           item->lri_slot);
1245                         if (ret < 0)
1246                                 mlog_errno(ret);
1247                         /* Recovery info is already freed now */
1248                 }
1249 
1250                 kfree(item);
1251         }
1252 
1253         trace_ocfs2_complete_recovery_end(ret);
1254 }
1255 
1256 /* NOTE: This function always eats your references to la_dinode and
1257  * tl_dinode, either manually on error, or by passing them to
1258  * ocfs2_complete_recovery */
1259 static void ocfs2_queue_recovery_completion(struct ocfs2_journal *journal,
1260                                             int slot_num,
1261                                             struct ocfs2_dinode *la_dinode,
1262                                             struct ocfs2_dinode *tl_dinode,
1263                                             struct ocfs2_quota_recovery *qrec)
1264 {
1265         struct ocfs2_la_recovery_item *item;
1266 
1267         item = kmalloc(sizeof(struct ocfs2_la_recovery_item), GFP_NOFS);
1268         if (!item) {
1269                 /* Though we wish to avoid it, we are in fact safe in
1270                  * skipping local alloc cleanup as fsck.ocfs2 is more
1271                  * than capable of reclaiming unused space. */
1272                 kfree(la_dinode);
1273                 kfree(tl_dinode);
1274 
1275                 if (qrec)
1276                         ocfs2_free_quota_recovery(qrec);
1277 
1278                 mlog_errno(-ENOMEM);
1279                 return;
1280         }
1281 
1282         INIT_LIST_HEAD(&item->lri_list);
1283         item->lri_la_dinode = la_dinode;
1284         item->lri_slot = slot_num;
1285         item->lri_tl_dinode = tl_dinode;
1286         item->lri_qrec = qrec;
1287 
1288         spin_lock(&journal->j_lock);
1289         list_add_tail(&item->lri_list, &journal->j_la_cleanups);
1290         queue_work(ocfs2_wq, &journal->j_recovery_work);
1291         spin_unlock(&journal->j_lock);
1292 }
1293 
1294 /* Called by the mount code to queue recovery the last part of
1295  * recovery for it's own and offline slot(s). */
1296 void ocfs2_complete_mount_recovery(struct ocfs2_super *osb)
1297 {
1298         struct ocfs2_journal *journal = osb->journal;
1299 
1300         if (ocfs2_is_hard_readonly(osb))
1301                 return;
1302 
1303         /* No need to queue up our truncate_log as regular cleanup will catch
1304          * that */
1305         ocfs2_queue_recovery_completion(journal, osb->slot_num,
1306                                         osb->local_alloc_copy, NULL, NULL);
1307         ocfs2_schedule_truncate_log_flush(osb, 0);
1308 
1309         osb->local_alloc_copy = NULL;
1310         osb->dirty = 0;
1311 
1312         /* queue to recover orphan slots for all offline slots */
1313         ocfs2_replay_map_set_state(osb, REPLAY_NEEDED);
1314         ocfs2_queue_replay_slots(osb);
1315         ocfs2_free_replay_slots(osb);
1316 }
1317 
1318 void ocfs2_complete_quota_recovery(struct ocfs2_super *osb)
1319 {
1320         if (osb->quota_rec) {
1321                 ocfs2_queue_recovery_completion(osb->journal,
1322                                                 osb->slot_num,
1323                                                 NULL,
1324                                                 NULL,
1325                                                 osb->quota_rec);
1326                 osb->quota_rec = NULL;
1327         }
1328 }
1329 
1330 static int __ocfs2_recovery_thread(void *arg)
1331 {
1332         int status, node_num, slot_num;
1333         struct ocfs2_super *osb = arg;
1334         struct ocfs2_recovery_map *rm = osb->recovery_map;
1335         int *rm_quota = NULL;
1336         int rm_quota_used = 0, i;
1337         struct ocfs2_quota_recovery *qrec;
1338 
1339         status = ocfs2_wait_on_mount(osb);
1340         if (status < 0) {
1341                 goto bail;
1342         }
1343 
1344         rm_quota = kzalloc(osb->max_slots * sizeof(int), GFP_NOFS);
1345         if (!rm_quota) {
1346                 status = -ENOMEM;
1347                 goto bail;
1348         }
1349 restart:
1350         status = ocfs2_super_lock(osb, 1);
1351         if (status < 0) {
1352                 mlog_errno(status);
1353                 goto bail;
1354         }
1355 
1356         status = ocfs2_compute_replay_slots(osb);
1357         if (status < 0)
1358                 mlog_errno(status);
1359 
1360         /* queue recovery for our own slot */
1361         ocfs2_queue_recovery_completion(osb->journal, osb->slot_num, NULL,
1362                                         NULL, NULL);
1363 
1364         spin_lock(&osb->osb_lock);
1365         while (rm->rm_used) {
1366                 /* It's always safe to remove entry zero, as we won't
1367                  * clear it until ocfs2_recover_node() has succeeded. */
1368                 node_num = rm->rm_entries[0];
1369                 spin_unlock(&osb->osb_lock);
1370                 slot_num = ocfs2_node_num_to_slot(osb, node_num);
1371                 trace_ocfs2_recovery_thread_node(node_num, slot_num);
1372                 if (slot_num == -ENOENT) {
1373                         status = 0;
1374                         goto skip_recovery;
1375                 }
1376 
1377                 /* It is a bit subtle with quota recovery. We cannot do it
1378                  * immediately because we have to obtain cluster locks from
1379                  * quota files and we also don't want to just skip it because
1380                  * then quota usage would be out of sync until some node takes
1381                  * the slot. So we remember which nodes need quota recovery
1382                  * and when everything else is done, we recover quotas. */
1383                 for (i = 0; i < rm_quota_used && rm_quota[i] != slot_num; i++);
1384                 if (i == rm_quota_used)
1385                         rm_quota[rm_quota_used++] = slot_num;
1386 
1387                 status = ocfs2_recover_node(osb, node_num, slot_num);
1388 skip_recovery:
1389                 if (!status) {
1390                         ocfs2_recovery_map_clear(osb, node_num);
1391                 } else {
1392                         mlog(ML_ERROR,
1393                              "Error %d recovering node %d on device (%u,%u)!\n",
1394                              status, node_num,
1395                              MAJOR(osb->sb->s_dev), MINOR(osb->sb->s_dev));
1396                         mlog(ML_ERROR, "Volume requires unmount.\n");
1397                 }
1398 
1399                 spin_lock(&osb->osb_lock);
1400         }
1401         spin_unlock(&osb->osb_lock);
1402         trace_ocfs2_recovery_thread_end(status);
1403 
1404         /* Refresh all journal recovery generations from disk */
1405         status = ocfs2_check_journals_nolocks(osb);
1406         status = (status == -EROFS) ? 0 : status;
1407         if (status < 0)
1408                 mlog_errno(status);
1409 
1410         /* Now it is right time to recover quotas... We have to do this under
1411          * superblock lock so that no one can start using the slot (and crash)
1412          * before we recover it */
1413         for (i = 0; i < rm_quota_used; i++) {
1414                 qrec = ocfs2_begin_quota_recovery(osb, rm_quota[i]);
1415                 if (IS_ERR(qrec)) {
1416                         status = PTR_ERR(qrec);
1417                         mlog_errno(status);
1418                         continue;
1419                 }
1420                 ocfs2_queue_recovery_completion(osb->journal, rm_quota[i],
1421                                                 NULL, NULL, qrec);
1422         }
1423 
1424         ocfs2_super_unlock(osb, 1);
1425 
1426         /* queue recovery for offline slots */
1427         ocfs2_queue_replay_slots(osb);
1428 
1429 bail:
1430         mutex_lock(&osb->recovery_lock);
1431         if (!status && !ocfs2_recovery_completed(osb)) {
1432                 mutex_unlock(&osb->recovery_lock);
1433                 goto restart;
1434         }
1435 
1436         ocfs2_free_replay_slots(osb);
1437         osb->recovery_thread_task = NULL;
1438         mb(); /* sync with ocfs2_recovery_thread_running */
1439         wake_up(&osb->recovery_event);
1440 
1441         mutex_unlock(&osb->recovery_lock);
1442 
1443         kfree(rm_quota);
1444 
1445         /* no one is callint kthread_stop() for us so the kthread() api
1446          * requires that we call do_exit().  And it isn't exported, but
1447          * complete_and_exit() seems to be a minimal wrapper around it. */
1448         complete_and_exit(NULL, status);
1449         return status;
1450 }
1451 
1452 void ocfs2_recovery_thread(struct ocfs2_super *osb, int node_num)
1453 {
1454         mutex_lock(&osb->recovery_lock);
1455 
1456         trace_ocfs2_recovery_thread(node_num, osb->node_num,
1457                 osb->disable_recovery, osb->recovery_thread_task,
1458                 osb->disable_recovery ?
1459                 -1 : ocfs2_recovery_map_set(osb, node_num));
1460 
1461         if (osb->disable_recovery)
1462                 goto out;
1463 
1464         if (osb->recovery_thread_task)
1465                 goto out;
1466 
1467         osb->recovery_thread_task =  kthread_run(__ocfs2_recovery_thread, osb,
1468                                                  "ocfs2rec");
1469         if (IS_ERR(osb->recovery_thread_task)) {
1470                 mlog_errno((int)PTR_ERR(osb->recovery_thread_task));
1471                 osb->recovery_thread_task = NULL;
1472         }
1473 
1474 out:
1475         mutex_unlock(&osb->recovery_lock);
1476         wake_up(&osb->recovery_event);
1477 }
1478 
1479 static int ocfs2_read_journal_inode(struct ocfs2_super *osb,
1480                                     int slot_num,
1481                                     struct buffer_head **bh,
1482                                     struct inode **ret_inode)
1483 {
1484         int status = -EACCES;
1485         struct inode *inode = NULL;
1486 
1487         BUG_ON(slot_num >= osb->max_slots);
1488 
1489         inode = ocfs2_get_system_file_inode(osb, JOURNAL_SYSTEM_INODE,
1490                                             slot_num);
1491         if (!inode || is_bad_inode(inode)) {
1492                 mlog_errno(status);
1493                 goto bail;
1494         }
1495         SET_INODE_JOURNAL(inode);
1496 
1497         status = ocfs2_read_inode_block_full(inode, bh, OCFS2_BH_IGNORE_CACHE);
1498         if (status < 0) {
1499                 mlog_errno(status);
1500                 goto bail;
1501         }
1502 
1503         status = 0;
1504 
1505 bail:
1506         if (inode) {
1507                 if (status || !ret_inode)
1508                         iput(inode);
1509                 else
1510                         *ret_inode = inode;
1511         }
1512         return status;
1513 }
1514 
1515 /* Does the actual journal replay and marks the journal inode as
1516  * clean. Will only replay if the journal inode is marked dirty. */
1517 static int ocfs2_replay_journal(struct ocfs2_super *osb,
1518                                 int node_num,
1519                                 int slot_num)
1520 {
1521         int status;
1522         int got_lock = 0;
1523         unsigned int flags;
1524         struct inode *inode = NULL;
1525         struct ocfs2_dinode *fe;
1526         journal_t *journal = NULL;
1527         struct buffer_head *bh = NULL;
1528         u32 slot_reco_gen;
1529 
1530         status = ocfs2_read_journal_inode(osb, slot_num, &bh, &inode);
1531         if (status) {
1532                 mlog_errno(status);
1533                 goto done;
1534         }
1535 
1536         fe = (struct ocfs2_dinode *)bh->b_data;
1537         slot_reco_gen = ocfs2_get_recovery_generation(fe);
1538         brelse(bh);
1539         bh = NULL;
1540 
1541         /*
1542          * As the fs recovery is asynchronous, there is a small chance that
1543          * another node mounted (and recovered) the slot before the recovery
1544          * thread could get the lock. To handle that, we dirty read the journal
1545          * inode for that slot to get the recovery generation. If it is
1546          * different than what we expected, the slot has been recovered.
1547          * If not, it needs recovery.
1548          */
1549         if (osb->slot_recovery_generations[slot_num] != slot_reco_gen) {
1550                 trace_ocfs2_replay_journal_recovered(slot_num,
1551                      osb->slot_recovery_generations[slot_num], slot_reco_gen);
1552                 osb->slot_recovery_generations[slot_num] = slot_reco_gen;
1553                 status = -EBUSY;
1554                 goto done;
1555         }
1556 
1557         /* Continue with recovery as the journal has not yet been recovered */
1558 
1559         status = ocfs2_inode_lock_full(inode, &bh, 1, OCFS2_META_LOCK_RECOVERY);
1560         if (status < 0) {
1561                 trace_ocfs2_replay_journal_lock_err(status);
1562                 if (status != -ERESTARTSYS)
1563                         mlog(ML_ERROR, "Could not lock journal!\n");
1564                 goto done;
1565         }
1566         got_lock = 1;
1567 
1568         fe = (struct ocfs2_dinode *) bh->b_data;
1569 
1570         flags = le32_to_cpu(fe->id1.journal1.ij_flags);
1571         slot_reco_gen = ocfs2_get_recovery_generation(fe);
1572 
1573         if (!(flags & OCFS2_JOURNAL_DIRTY_FL)) {
1574                 trace_ocfs2_replay_journal_skip(node_num);
1575                 /* Refresh recovery generation for the slot */
1576                 osb->slot_recovery_generations[slot_num] = slot_reco_gen;
1577                 goto done;
1578         }
1579 
1580         /* we need to run complete recovery for offline orphan slots */
1581         ocfs2_replay_map_set_state(osb, REPLAY_NEEDED);
1582 
1583         printk(KERN_NOTICE "ocfs2: Begin replay journal (node %d, slot %d) on "\
1584                "device (%u,%u)\n", node_num, slot_num, MAJOR(osb->sb->s_dev),
1585                MINOR(osb->sb->s_dev));
1586 
1587         OCFS2_I(inode)->ip_clusters = le32_to_cpu(fe->i_clusters);
1588 
1589         status = ocfs2_force_read_journal(inode);
1590         if (status < 0) {
1591                 mlog_errno(status);
1592                 goto done;
1593         }
1594 
1595         journal = jbd2_journal_init_inode(inode);
1596         if (journal == NULL) {
1597                 mlog(ML_ERROR, "Linux journal layer error\n");
1598                 status = -EIO;
1599                 goto done;
1600         }
1601 
1602         status = jbd2_journal_load(journal);
1603         if (status < 0) {
1604                 mlog_errno(status);
1605                 if (!igrab(inode))
1606                         BUG();
1607                 jbd2_journal_destroy(journal);
1608                 goto done;
1609         }
1610 
1611         ocfs2_clear_journal_error(osb->sb, journal, slot_num);
1612 
1613         /* wipe the journal */
1614         jbd2_journal_lock_updates(journal);
1615         status = jbd2_journal_flush(journal);
1616         jbd2_journal_unlock_updates(journal);
1617         if (status < 0)
1618                 mlog_errno(status);
1619 
1620         /* This will mark the node clean */
1621         flags = le32_to_cpu(fe->id1.journal1.ij_flags);
1622         flags &= ~OCFS2_JOURNAL_DIRTY_FL;
1623         fe->id1.journal1.ij_flags = cpu_to_le32(flags);
1624 
1625         /* Increment recovery generation to indicate successful recovery */
1626         ocfs2_bump_recovery_generation(fe);
1627         osb->slot_recovery_generations[slot_num] =
1628                                         ocfs2_get_recovery_generation(fe);
1629 
1630         ocfs2_compute_meta_ecc(osb->sb, bh->b_data, &fe->i_check);
1631         status = ocfs2_write_block(osb, bh, INODE_CACHE(inode));
1632         if (status < 0)
1633                 mlog_errno(status);
1634 
1635         if (!igrab(inode))
1636                 BUG();
1637 
1638         jbd2_journal_destroy(journal);
1639 
1640         printk(KERN_NOTICE "ocfs2: End replay journal (node %d, slot %d) on "\
1641                "device (%u,%u)\n", node_num, slot_num, MAJOR(osb->sb->s_dev),
1642                MINOR(osb->sb->s_dev));
1643 done:
1644         /* drop the lock on this nodes journal */
1645         if (got_lock)
1646                 ocfs2_inode_unlock(inode, 1);
1647 
1648         if (inode)
1649                 iput(inode);
1650 
1651         brelse(bh);
1652 
1653         return status;
1654 }
1655 
1656 /*
1657  * Do the most important parts of node recovery:
1658  *  - Replay it's journal
1659  *  - Stamp a clean local allocator file
1660  *  - Stamp a clean truncate log
1661  *  - Mark the node clean
1662  *
1663  * If this function completes without error, a node in OCFS2 can be
1664  * said to have been safely recovered. As a result, failure during the
1665  * second part of a nodes recovery process (local alloc recovery) is
1666  * far less concerning.
1667  */
1668 static int ocfs2_recover_node(struct ocfs2_super *osb,
1669                               int node_num, int slot_num)
1670 {
1671         int status = 0;
1672         struct ocfs2_dinode *la_copy = NULL;
1673         struct ocfs2_dinode *tl_copy = NULL;
1674 
1675         trace_ocfs2_recover_node(node_num, slot_num, osb->node_num);
1676 
1677         /* Should not ever be called to recover ourselves -- in that
1678          * case we should've called ocfs2_journal_load instead. */
1679         BUG_ON(osb->node_num == node_num);
1680 
1681         status = ocfs2_replay_journal(osb, node_num, slot_num);
1682         if (status < 0) {
1683                 if (status == -EBUSY) {
1684                         trace_ocfs2_recover_node_skip(slot_num, node_num);
1685                         status = 0;
1686                         goto done;
1687                 }
1688                 mlog_errno(status);
1689                 goto done;
1690         }
1691 
1692         /* Stamp a clean local alloc file AFTER recovering the journal... */
1693         status = ocfs2_begin_local_alloc_recovery(osb, slot_num, &la_copy);
1694         if (status < 0) {
1695                 mlog_errno(status);
1696                 goto done;
1697         }
1698 
1699         /* An error from begin_truncate_log_recovery is not
1700          * serious enough to warrant halting the rest of
1701          * recovery. */
1702         status = ocfs2_begin_truncate_log_recovery(osb, slot_num, &tl_copy);
1703         if (status < 0)
1704                 mlog_errno(status);
1705 
1706         /* Likewise, this would be a strange but ultimately not so
1707          * harmful place to get an error... */
1708         status = ocfs2_clear_slot(osb, slot_num);
1709         if (status < 0)
1710                 mlog_errno(status);
1711 
1712         /* This will kfree the memory pointed to by la_copy and tl_copy */
1713         ocfs2_queue_recovery_completion(osb->journal, slot_num, la_copy,
1714                                         tl_copy, NULL);
1715 
1716         status = 0;
1717 done:
1718 
1719         return status;
1720 }
1721 
1722 /* Test node liveness by trylocking his journal. If we get the lock,
1723  * we drop it here. Return 0 if we got the lock, -EAGAIN if node is
1724  * still alive (we couldn't get the lock) and < 0 on error. */
1725 static int ocfs2_trylock_journal(struct ocfs2_super *osb,
1726                                  int slot_num)
1727 {
1728         int status, flags;
1729         struct inode *inode = NULL;
1730 
1731         inode = ocfs2_get_system_file_inode(osb, JOURNAL_SYSTEM_INODE,
1732                                             slot_num);
1733         if (inode == NULL) {
1734                 mlog(ML_ERROR, "access error\n");
1735                 status = -EACCES;
1736                 goto bail;
1737         }
1738         if (is_bad_inode(inode)) {
1739                 mlog(ML_ERROR, "access error (bad inode)\n");
1740                 iput(inode);
1741                 inode = NULL;
1742                 status = -EACCES;
1743                 goto bail;
1744         }
1745         SET_INODE_JOURNAL(inode);
1746 
1747         flags = OCFS2_META_LOCK_RECOVERY | OCFS2_META_LOCK_NOQUEUE;
1748         status = ocfs2_inode_lock_full(inode, NULL, 1, flags);
1749         if (status < 0) {
1750                 if (status != -EAGAIN)
1751                         mlog_errno(status);
1752                 goto bail;
1753         }
1754 
1755         ocfs2_inode_unlock(inode, 1);
1756 bail:
1757         if (inode)
1758                 iput(inode);
1759 
1760         return status;
1761 }
1762 
1763 /* Call this underneath ocfs2_super_lock. It also assumes that the
1764  * slot info struct has been updated from disk. */
1765 int ocfs2_mark_dead_nodes(struct ocfs2_super *osb)
1766 {
1767         unsigned int node_num;
1768         int status, i;
1769         u32 gen;
1770         struct buffer_head *bh = NULL;
1771         struct ocfs2_dinode *di;
1772 
1773         /* This is called with the super block cluster lock, so we
1774          * know that the slot map can't change underneath us. */
1775 
1776         for (i = 0; i < osb->max_slots; i++) {
1777                 /* Read journal inode to get the recovery generation */
1778                 status = ocfs2_read_journal_inode(osb, i, &bh, NULL);
1779                 if (status) {
1780                         mlog_errno(status);
1781                         goto bail;
1782                 }
1783                 di = (struct ocfs2_dinode *)bh->b_data;
1784                 gen = ocfs2_get_recovery_generation(di);
1785                 brelse(bh);
1786                 bh = NULL;
1787 
1788                 spin_lock(&osb->osb_lock);
1789                 osb->slot_recovery_generations[i] = gen;
1790 
1791                 trace_ocfs2_mark_dead_nodes(i,
1792                                             osb->slot_recovery_generations[i]);
1793 
1794                 if (i == osb->slot_num) {
1795                         spin_unlock(&osb->osb_lock);
1796                         continue;
1797                 }
1798 
1799                 status = ocfs2_slot_to_node_num_locked(osb, i, &node_num);
1800                 if (status == -ENOENT) {
1801                         spin_unlock(&osb->osb_lock);
1802                         continue;
1803                 }
1804 
1805                 if (__ocfs2_recovery_map_test(osb, node_num)) {
1806                         spin_unlock(&osb->osb_lock);
1807                         continue;
1808                 }
1809                 spin_unlock(&osb->osb_lock);
1810 
1811                 /* Ok, we have a slot occupied by another node which
1812                  * is not in the recovery map. We trylock his journal
1813                  * file here to test if he's alive. */
1814                 status = ocfs2_trylock_journal(osb, i);
1815                 if (!status) {
1816                         /* Since we're called from mount, we know that
1817                          * the recovery thread can't race us on
1818                          * setting / checking the recovery bits. */
1819                         ocfs2_recovery_thread(osb, node_num);
1820                 } else if ((status < 0) && (status != -EAGAIN)) {
1821                         mlog_errno(status);
1822                         goto bail;
1823                 }
1824         }
1825 
1826         status = 0;
1827 bail:
1828         return status;
1829 }
1830 
1831 /*
1832  * Scan timer should get fired every ORPHAN_SCAN_SCHEDULE_TIMEOUT. Add some
1833  * randomness to the timeout to minimize multple nodes firing the timer at the
1834  * same time.
1835  */
1836 static inline unsigned long ocfs2_orphan_scan_timeout(void)
1837 {
1838         unsigned long time;
1839 
1840         get_random_bytes(&time, sizeof(time));
1841         time = ORPHAN_SCAN_SCHEDULE_TIMEOUT + (time % 5000);
1842         return msecs_to_jiffies(time);
1843 }
1844 
1845 /*
1846  * ocfs2_queue_orphan_scan calls ocfs2_queue_recovery_completion for
1847  * every slot, queuing a recovery of the slot on the ocfs2_wq thread. This
1848  * is done to catch any orphans that are left over in orphan directories.
1849  *
1850  * It scans all slots, even ones that are in use. It does so to handle the
1851  * case described below:
1852  *
1853  *   Node 1 has an inode it was using. The dentry went away due to memory
1854  *   pressure.  Node 1 closes the inode, but it's on the free list. The node
1855  *   has the open lock.
1856  *   Node 2 unlinks the inode. It grabs the dentry lock to notify others,
1857  *   but node 1 has no dentry and doesn't get the message. It trylocks the
1858  *   open lock, sees that another node has a PR, and does nothing.
1859  *   Later node 2 runs its orphan dir. It igets the inode, trylocks the
1860  *   open lock, sees the PR still, and does nothing.
1861  *   Basically, we have to trigger an orphan iput on node 1. The only way
1862  *   for this to happen is if node 1 runs node 2's orphan dir.
1863  *
1864  * ocfs2_queue_orphan_scan gets called every ORPHAN_SCAN_SCHEDULE_TIMEOUT
1865  * seconds.  It gets an EX lock on os_lockres and checks sequence number
1866  * stored in LVB. If the sequence number has changed, it means some other
1867  * node has done the scan.  This node skips the scan and tracks the
1868  * sequence number.  If the sequence number didn't change, it means a scan
1869  * hasn't happened.  The node queues a scan and increments the
1870  * sequence number in the LVB.
1871  */
1872 void ocfs2_queue_orphan_scan(struct ocfs2_super *osb)
1873 {
1874         struct ocfs2_orphan_scan *os;
1875         int status, i;
1876         u32 seqno = 0;
1877 
1878         os = &osb->osb_orphan_scan;
1879 
1880         if (atomic_read(&os->os_state) == ORPHAN_SCAN_INACTIVE)
1881                 goto out;
1882 
1883         trace_ocfs2_queue_orphan_scan_begin(os->os_count, os->os_seqno,
1884                                             atomic_read(&os->os_state));
1885 
1886         status = ocfs2_orphan_scan_lock(osb, &seqno);
1887         if (status < 0) {
1888                 if (status != -EAGAIN)
1889                         mlog_errno(status);
1890                 goto out;
1891         }
1892 
1893         /* Do no queue the tasks if the volume is being umounted */
1894         if (atomic_read(&os->os_state) == ORPHAN_SCAN_INACTIVE)
1895                 goto unlock;
1896 
1897         if (os->os_seqno != seqno) {
1898                 os->os_seqno = seqno;
1899                 goto unlock;
1900         }
1901 
1902         for (i = 0; i < osb->max_slots; i++)
1903                 ocfs2_queue_recovery_completion(osb->journal, i, NULL, NULL,
1904                                                 NULL);
1905         /*
1906          * We queued a recovery on orphan slots, increment the sequence
1907          * number and update LVB so other node will skip the scan for a while
1908          */
1909         seqno++;
1910         os->os_count++;
1911         os->os_scantime = CURRENT_TIME;
1912 unlock:
1913         ocfs2_orphan_scan_unlock(osb, seqno);
1914 out:
1915         trace_ocfs2_queue_orphan_scan_end(os->os_count, os->os_seqno,
1916                                           atomic_read(&os->os_state));
1917         return;
1918 }
1919 
1920 /* Worker task that gets fired every ORPHAN_SCAN_SCHEDULE_TIMEOUT millsec */
1921 void ocfs2_orphan_scan_work(struct work_struct *work)
1922 {
1923         struct ocfs2_orphan_scan *os;
1924         struct ocfs2_super *osb;
1925 
1926         os = container_of(work, struct ocfs2_orphan_scan,
1927                           os_orphan_scan_work.work);
1928         osb = os->os_osb;
1929 
1930         mutex_lock(&os->os_lock);
1931         ocfs2_queue_orphan_scan(osb);
1932         if (atomic_read(&os->os_state) == ORPHAN_SCAN_ACTIVE)
1933                 queue_delayed_work(ocfs2_wq, &os->os_orphan_scan_work,
1934                                       ocfs2_orphan_scan_timeout());
1935         mutex_unlock(&os->os_lock);
1936 }
1937 
1938 void ocfs2_orphan_scan_stop(struct ocfs2_super *osb)
1939 {
1940         struct ocfs2_orphan_scan *os;
1941 
1942         os = &osb->osb_orphan_scan;
1943         if (atomic_read(&os->os_state) == ORPHAN_SCAN_ACTIVE) {
1944                 atomic_set(&os->os_state, ORPHAN_SCAN_INACTIVE);
1945                 mutex_lock(&os->os_lock);
1946                 cancel_delayed_work(&os->os_orphan_scan_work);
1947                 mutex_unlock(&os->os_lock);
1948         }
1949 }
1950 
1951 void ocfs2_orphan_scan_init(struct ocfs2_super *osb)
1952 {
1953         struct ocfs2_orphan_scan *os;
1954 
1955         os = &osb->osb_orphan_scan;
1956         os->os_osb = osb;
1957         os->os_count = 0;
1958         os->os_seqno = 0;
1959         mutex_init(&os->os_lock);
1960         INIT_DELAYED_WORK(&os->os_orphan_scan_work, ocfs2_orphan_scan_work);
1961 }
1962 
1963 void ocfs2_orphan_scan_start(struct ocfs2_super *osb)
1964 {
1965         struct ocfs2_orphan_scan *os;
1966 
1967         os = &osb->osb_orphan_scan;
1968         os->os_scantime = CURRENT_TIME;
1969         if (ocfs2_is_hard_readonly(osb) || ocfs2_mount_local(osb))
1970                 atomic_set(&os->os_state, ORPHAN_SCAN_INACTIVE);
1971         else {
1972                 atomic_set(&os->os_state, ORPHAN_SCAN_ACTIVE);
1973                 queue_delayed_work(ocfs2_wq, &os->os_orphan_scan_work,
1974                                    ocfs2_orphan_scan_timeout());
1975         }
1976 }
1977 
1978 struct ocfs2_orphan_filldir_priv {
1979         struct dir_context      ctx;
1980         struct inode            *head;
1981         struct ocfs2_super      *osb;
1982 };
1983 
1984 static int ocfs2_orphan_filldir(void *priv, const char *name, int name_len,
1985                                 loff_t pos, u64 ino, unsigned type)
1986 {
1987         struct ocfs2_orphan_filldir_priv *p = priv;
1988         struct inode *iter;
1989 
1990         if (name_len == 1 && !strncmp(".", name, 1))
1991                 return 0;
1992         if (name_len == 2 && !strncmp("..", name, 2))
1993                 return 0;
1994 
1995         /* Skip bad inodes so that recovery can continue */
1996         iter = ocfs2_iget(p->osb, ino,
1997                           OCFS2_FI_FLAG_ORPHAN_RECOVERY, 0);
1998         if (IS_ERR(iter))
1999                 return 0;
2000 
2001         trace_ocfs2_orphan_filldir((unsigned long long)OCFS2_I(iter)->ip_blkno);
2002         /* No locking is required for the next_orphan queue as there
2003          * is only ever a single process doing orphan recovery. */
2004         OCFS2_I(iter)->ip_next_orphan = p->head;
2005         p->head = iter;
2006 
2007         return 0;
2008 }
2009 
2010 static int ocfs2_queue_orphans(struct ocfs2_super *osb,
2011                                int slot,
2012                                struct inode **head)
2013 {
2014         int status;
2015         struct inode *orphan_dir_inode = NULL;
2016         struct ocfs2_orphan_filldir_priv priv = {
2017                 .ctx.actor = ocfs2_orphan_filldir,
2018                 .osb = osb,
2019                 .head = *head
2020         };
2021 
2022         orphan_dir_inode = ocfs2_get_system_file_inode(osb,
2023                                                        ORPHAN_DIR_SYSTEM_INODE,
2024                                                        slot);
2025         if  (!orphan_dir_inode) {
2026                 status = -ENOENT;
2027                 mlog_errno(status);
2028                 return status;
2029         }
2030 
2031         mutex_lock(&orphan_dir_inode->i_mutex);
2032         status = ocfs2_inode_lock(orphan_dir_inode, NULL, 0);
2033         if (status < 0) {
2034                 mlog_errno(status);
2035                 goto out;
2036         }
2037 
2038         status = ocfs2_dir_foreach(orphan_dir_inode, &priv.ctx);
2039         if (status) {
2040                 mlog_errno(status);
2041                 goto out_cluster;
2042         }
2043 
2044         *head = priv.head;
2045 
2046 out_cluster:
2047         ocfs2_inode_unlock(orphan_dir_inode, 0);
2048 out:
2049         mutex_unlock(&orphan_dir_inode->i_mutex);
2050         iput(orphan_dir_inode);
2051         return status;
2052 }
2053 
2054 static int ocfs2_orphan_recovery_can_continue(struct ocfs2_super *osb,
2055                                               int slot)
2056 {
2057         int ret;
2058 
2059         spin_lock(&osb->osb_lock);
2060         ret = !osb->osb_orphan_wipes[slot];
2061         spin_unlock(&osb->osb_lock);
2062         return ret;
2063 }
2064 
2065 static void ocfs2_mark_recovering_orphan_dir(struct ocfs2_super *osb,
2066                                              int slot)
2067 {
2068         spin_lock(&osb->osb_lock);
2069         /* Mark ourselves such that new processes in delete_inode()
2070          * know to quit early. */
2071         ocfs2_node_map_set_bit(osb, &osb->osb_recovering_orphan_dirs, slot);
2072         while (osb->osb_orphan_wipes[slot]) {
2073                 /* If any processes are already in the middle of an
2074                  * orphan wipe on this dir, then we need to wait for
2075                  * them. */
2076                 spin_unlock(&osb->osb_lock);
2077                 wait_event_interruptible(osb->osb_wipe_event,
2078                                          ocfs2_orphan_recovery_can_continue(osb, slot));
2079                 spin_lock(&osb->osb_lock);
2080         }
2081         spin_unlock(&osb->osb_lock);
2082 }
2083 
2084 static void ocfs2_clear_recovering_orphan_dir(struct ocfs2_super *osb,
2085                                               int slot)
2086 {
2087         ocfs2_node_map_clear_bit(osb, &osb->osb_recovering_orphan_dirs, slot);
2088 }
2089 
2090 /*
2091  * Orphan recovery. Each mounted node has it's own orphan dir which we
2092  * must run during recovery. Our strategy here is to build a list of
2093  * the inodes in the orphan dir and iget/iput them. The VFS does
2094  * (most) of the rest of the work.
2095  *
2096  * Orphan recovery can happen at any time, not just mount so we have a
2097  * couple of extra considerations.
2098  *
2099  * - We grab as many inodes as we can under the orphan dir lock -
2100  *   doing iget() outside the orphan dir risks getting a reference on
2101  *   an invalid inode.
2102  * - We must be sure not to deadlock with other processes on the
2103  *   system wanting to run delete_inode(). This can happen when they go
2104  *   to lock the orphan dir and the orphan recovery process attempts to
2105  *   iget() inside the orphan dir lock. This can be avoided by
2106  *   advertising our state to ocfs2_delete_inode().
2107  */
2108 static int ocfs2_recover_orphans(struct ocfs2_super *osb,
2109                                  int slot)
2110 {
2111         int ret = 0;
2112         struct inode *inode = NULL;
2113         struct inode *iter;
2114         struct ocfs2_inode_info *oi;
2115 
2116         trace_ocfs2_recover_orphans(slot);
2117 
2118         ocfs2_mark_recovering_orphan_dir(osb, slot);
2119         ret = ocfs2_queue_orphans(osb, slot, &inode);
2120         ocfs2_clear_recovering_orphan_dir(osb, slot);
2121 
2122         /* Error here should be noted, but we want to continue with as
2123          * many queued inodes as we've got. */
2124         if (ret)
2125                 mlog_errno(ret);
2126 
2127         while (inode) {
2128                 oi = OCFS2_I(inode);
2129                 trace_ocfs2_recover_orphans_iput(
2130                                         (unsigned long long)oi->ip_blkno);
2131 
2132                 iter = oi->ip_next_orphan;
2133 
2134                 spin_lock(&oi->ip_lock);
2135                 /* The remote delete code may have set these on the
2136                  * assumption that the other node would wipe them
2137                  * successfully.  If they are still in the node's
2138                  * orphan dir, we need to reset that state. */
2139                 oi->ip_flags &= ~(OCFS2_INODE_DELETED|OCFS2_INODE_SKIP_DELETE);
2140 
2141                 /* Set the proper information to get us going into
2142                  * ocfs2_delete_inode. */
2143                 oi->ip_flags |= OCFS2_INODE_MAYBE_ORPHANED;
2144                 spin_unlock(&oi->ip_lock);
2145 
2146                 iput(inode);
2147 
2148                 inode = iter;
2149         }
2150 
2151         return ret;
2152 }
2153 
2154 static int __ocfs2_wait_on_mount(struct ocfs2_super *osb, int quota)
2155 {
2156         /* This check is good because ocfs2 will wait on our recovery
2157          * thread before changing it to something other than MOUNTED
2158          * or DISABLED. */
2159         wait_event(osb->osb_mount_event,
2160                   (!quota && atomic_read(&osb->vol_state) == VOLUME_MOUNTED) ||
2161                    atomic_read(&osb->vol_state) == VOLUME_MOUNTED_QUOTAS ||
2162                    atomic_read(&osb->vol_state) == VOLUME_DISABLED);
2163 
2164         /* If there's an error on mount, then we may never get to the
2165          * MOUNTED flag, but this is set right before
2166          * dismount_volume() so we can trust it. */
2167         if (atomic_read(&osb->vol_state) == VOLUME_DISABLED) {
2168                 trace_ocfs2_wait_on_mount(VOLUME_DISABLED);
2169                 mlog(0, "mount error, exiting!\n");
2170                 return -EBUSY;
2171         }
2172 
2173         return 0;
2174 }
2175 
2176 static int ocfs2_commit_thread(void *arg)
2177 {
2178         int status;
2179         struct ocfs2_super *osb = arg;
2180         struct ocfs2_journal *journal = osb->journal;
2181 
2182         /* we can trust j_num_trans here because _should_stop() is only set in
2183          * shutdown and nobody other than ourselves should be able to start
2184          * transactions.  committing on shutdown might take a few iterations
2185          * as final transactions put deleted inodes on the list */
2186         while (!(kthread_should_stop() &&
2187                  atomic_read(&journal->j_num_trans) == 0)) {
2188 
2189                 wait_event_interruptible(osb->checkpoint_event,
2190                                          atomic_read(&journal->j_num_trans)
2191                                          || kthread_should_stop());
2192 
2193                 status = ocfs2_commit_cache(osb);
2194                 if (status < 0)
2195                         mlog_errno(status);
2196 
2197                 if (kthread_should_stop() && atomic_read(&journal->j_num_trans)){
2198                         mlog(ML_KTHREAD,
2199                              "commit_thread: %u transactions pending on "
2200                              "shutdown\n",
2201                              atomic_read(&journal->j_num_trans));
2202                 }
2203         }
2204 
2205         return 0;
2206 }
2207 
2208 /* Reads all the journal inodes without taking any cluster locks. Used
2209  * for hard readonly access to determine whether any journal requires
2210  * recovery. Also used to refresh the recovery generation numbers after
2211  * a journal has been recovered by another node.
2212  */
2213 int ocfs2_check_journals_nolocks(struct ocfs2_super *osb)
2214 {
2215         int ret = 0;
2216         unsigned int slot;
2217         struct buffer_head *di_bh = NULL;
2218         struct ocfs2_dinode *di;
2219         int journal_dirty = 0;
2220 
2221         for(slot = 0; slot < osb->max_slots; slot++) {
2222                 ret = ocfs2_read_journal_inode(osb, slot, &di_bh, NULL);
2223                 if (ret) {
2224                         mlog_errno(ret);
2225                         goto out;
2226                 }
2227 
2228                 di = (struct ocfs2_dinode *) di_bh->b_data;
2229 
2230                 osb->slot_recovery_generations[slot] =
2231                                         ocfs2_get_recovery_generation(di);
2232 
2233                 if (le32_to_cpu(di->id1.journal1.ij_flags) &
2234                     OCFS2_JOURNAL_DIRTY_FL)
2235                         journal_dirty = 1;
2236 
2237                 brelse(di_bh);
2238                 di_bh = NULL;
2239         }
2240 
2241 out:
2242         if (journal_dirty)
2243                 ret = -EROFS;
2244         return ret;
2245 }
2246 

~ [ source navigation ] ~ [ diff markup ] ~ [ identifier search ] ~

kernel.org | git.kernel.org | LWN.net | Project Home | Wiki (Japanese) | Wiki (English) | SVN repository | Mail admin

Linux® is a registered trademark of Linus Torvalds in the United States and other countries.
TOMOYO® is a registered trademark of NTT DATA CORPORATION.

osdn.jp