xref: /openbmc/linux/fs/ocfs2/journal.c (revision 54cbac81)
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 struct ocfs2_triggers {
459 	struct jbd2_buffer_trigger_type	ot_triggers;
460 	int				ot_offset;
461 };
462 
463 static inline struct ocfs2_triggers *to_ocfs2_trigger(struct jbd2_buffer_trigger_type *triggers)
464 {
465 	return container_of(triggers, struct ocfs2_triggers, ot_triggers);
466 }
467 
468 static void ocfs2_frozen_trigger(struct jbd2_buffer_trigger_type *triggers,
469 				 struct buffer_head *bh,
470 				 void *data, size_t size)
471 {
472 	struct ocfs2_triggers *ot = to_ocfs2_trigger(triggers);
473 
474 	/*
475 	 * We aren't guaranteed to have the superblock here, so we
476 	 * must unconditionally compute the ecc data.
477 	 * __ocfs2_journal_access() will only set the triggers if
478 	 * metaecc is enabled.
479 	 */
480 	ocfs2_block_check_compute(data, size, data + ot->ot_offset);
481 }
482 
483 /*
484  * Quota blocks have their own trigger because the struct ocfs2_block_check
485  * offset depends on the blocksize.
486  */
487 static void ocfs2_dq_frozen_trigger(struct jbd2_buffer_trigger_type *triggers,
488 				 struct buffer_head *bh,
489 				 void *data, size_t size)
490 {
491 	struct ocfs2_disk_dqtrailer *dqt =
492 		ocfs2_block_dqtrailer(size, data);
493 
494 	/*
495 	 * We aren't guaranteed to have the superblock here, so we
496 	 * must unconditionally compute the ecc data.
497 	 * __ocfs2_journal_access() will only set the triggers if
498 	 * metaecc is enabled.
499 	 */
500 	ocfs2_block_check_compute(data, size, &dqt->dq_check);
501 }
502 
503 /*
504  * Directory blocks also have their own trigger because the
505  * struct ocfs2_block_check offset depends on the blocksize.
506  */
507 static void ocfs2_db_frozen_trigger(struct jbd2_buffer_trigger_type *triggers,
508 				 struct buffer_head *bh,
509 				 void *data, size_t size)
510 {
511 	struct ocfs2_dir_block_trailer *trailer =
512 		ocfs2_dir_trailer_from_size(size, data);
513 
514 	/*
515 	 * We aren't guaranteed to have the superblock here, so we
516 	 * must unconditionally compute the ecc data.
517 	 * __ocfs2_journal_access() will only set the triggers if
518 	 * metaecc is enabled.
519 	 */
520 	ocfs2_block_check_compute(data, size, &trailer->db_check);
521 }
522 
523 static void ocfs2_abort_trigger(struct jbd2_buffer_trigger_type *triggers,
524 				struct buffer_head *bh)
525 {
526 	mlog(ML_ERROR,
527 	     "ocfs2_abort_trigger called by JBD2.  bh = 0x%lx, "
528 	     "bh->b_blocknr = %llu\n",
529 	     (unsigned long)bh,
530 	     (unsigned long long)bh->b_blocknr);
531 
532 	/* We aren't guaranteed to have the superblock here - but if we
533 	 * don't, it'll just crash. */
534 	ocfs2_error(bh->b_assoc_map->host->i_sb,
535 		    "JBD2 has aborted our journal, ocfs2 cannot continue\n");
536 }
537 
538 static struct ocfs2_triggers di_triggers = {
539 	.ot_triggers = {
540 		.t_frozen = ocfs2_frozen_trigger,
541 		.t_abort = ocfs2_abort_trigger,
542 	},
543 	.ot_offset	= offsetof(struct ocfs2_dinode, i_check),
544 };
545 
546 static struct ocfs2_triggers eb_triggers = {
547 	.ot_triggers = {
548 		.t_frozen = ocfs2_frozen_trigger,
549 		.t_abort = ocfs2_abort_trigger,
550 	},
551 	.ot_offset	= offsetof(struct ocfs2_extent_block, h_check),
552 };
553 
554 static struct ocfs2_triggers rb_triggers = {
555 	.ot_triggers = {
556 		.t_frozen = ocfs2_frozen_trigger,
557 		.t_abort = ocfs2_abort_trigger,
558 	},
559 	.ot_offset	= offsetof(struct ocfs2_refcount_block, rf_check),
560 };
561 
562 static struct ocfs2_triggers gd_triggers = {
563 	.ot_triggers = {
564 		.t_frozen = ocfs2_frozen_trigger,
565 		.t_abort = ocfs2_abort_trigger,
566 	},
567 	.ot_offset	= offsetof(struct ocfs2_group_desc, bg_check),
568 };
569 
570 static struct ocfs2_triggers db_triggers = {
571 	.ot_triggers = {
572 		.t_frozen = ocfs2_db_frozen_trigger,
573 		.t_abort = ocfs2_abort_trigger,
574 	},
575 };
576 
577 static struct ocfs2_triggers xb_triggers = {
578 	.ot_triggers = {
579 		.t_frozen = ocfs2_frozen_trigger,
580 		.t_abort = ocfs2_abort_trigger,
581 	},
582 	.ot_offset	= offsetof(struct ocfs2_xattr_block, xb_check),
583 };
584 
585 static struct ocfs2_triggers dq_triggers = {
586 	.ot_triggers = {
587 		.t_frozen = ocfs2_dq_frozen_trigger,
588 		.t_abort = ocfs2_abort_trigger,
589 	},
590 };
591 
592 static struct ocfs2_triggers dr_triggers = {
593 	.ot_triggers = {
594 		.t_frozen = ocfs2_frozen_trigger,
595 		.t_abort = ocfs2_abort_trigger,
596 	},
597 	.ot_offset	= offsetof(struct ocfs2_dx_root_block, dr_check),
598 };
599 
600 static struct ocfs2_triggers dl_triggers = {
601 	.ot_triggers = {
602 		.t_frozen = ocfs2_frozen_trigger,
603 		.t_abort = ocfs2_abort_trigger,
604 	},
605 	.ot_offset	= offsetof(struct ocfs2_dx_leaf, dl_check),
606 };
607 
608 static int __ocfs2_journal_access(handle_t *handle,
609 				  struct ocfs2_caching_info *ci,
610 				  struct buffer_head *bh,
611 				  struct ocfs2_triggers *triggers,
612 				  int type)
613 {
614 	int status;
615 	struct ocfs2_super *osb =
616 		OCFS2_SB(ocfs2_metadata_cache_get_super(ci));
617 
618 	BUG_ON(!ci || !ci->ci_ops);
619 	BUG_ON(!handle);
620 	BUG_ON(!bh);
621 
622 	trace_ocfs2_journal_access(
623 		(unsigned long long)ocfs2_metadata_cache_owner(ci),
624 		(unsigned long long)bh->b_blocknr, type, bh->b_size);
625 
626 	/* we can safely remove this assertion after testing. */
627 	if (!buffer_uptodate(bh)) {
628 		mlog(ML_ERROR, "giving me a buffer that's not uptodate!\n");
629 		mlog(ML_ERROR, "b_blocknr=%llu\n",
630 		     (unsigned long long)bh->b_blocknr);
631 		BUG();
632 	}
633 
634 	/* Set the current transaction information on the ci so
635 	 * that the locking code knows whether it can drop it's locks
636 	 * on this ci or not. We're protected from the commit
637 	 * thread updating the current transaction id until
638 	 * ocfs2_commit_trans() because ocfs2_start_trans() took
639 	 * j_trans_barrier for us. */
640 	ocfs2_set_ci_lock_trans(osb->journal, ci);
641 
642 	ocfs2_metadata_cache_io_lock(ci);
643 	switch (type) {
644 	case OCFS2_JOURNAL_ACCESS_CREATE:
645 	case OCFS2_JOURNAL_ACCESS_WRITE:
646 		status = jbd2_journal_get_write_access(handle, bh);
647 		break;
648 
649 	case OCFS2_JOURNAL_ACCESS_UNDO:
650 		status = jbd2_journal_get_undo_access(handle, bh);
651 		break;
652 
653 	default:
654 		status = -EINVAL;
655 		mlog(ML_ERROR, "Unknown access type!\n");
656 	}
657 	if (!status && ocfs2_meta_ecc(osb) && triggers)
658 		jbd2_journal_set_triggers(bh, &triggers->ot_triggers);
659 	ocfs2_metadata_cache_io_unlock(ci);
660 
661 	if (status < 0)
662 		mlog(ML_ERROR, "Error %d getting %d access to buffer!\n",
663 		     status, type);
664 
665 	return status;
666 }
667 
668 int ocfs2_journal_access_di(handle_t *handle, struct ocfs2_caching_info *ci,
669 			    struct buffer_head *bh, int type)
670 {
671 	return __ocfs2_journal_access(handle, ci, bh, &di_triggers, type);
672 }
673 
674 int ocfs2_journal_access_eb(handle_t *handle, struct ocfs2_caching_info *ci,
675 			    struct buffer_head *bh, int type)
676 {
677 	return __ocfs2_journal_access(handle, ci, bh, &eb_triggers, type);
678 }
679 
680 int ocfs2_journal_access_rb(handle_t *handle, struct ocfs2_caching_info *ci,
681 			    struct buffer_head *bh, int type)
682 {
683 	return __ocfs2_journal_access(handle, ci, bh, &rb_triggers,
684 				      type);
685 }
686 
687 int ocfs2_journal_access_gd(handle_t *handle, struct ocfs2_caching_info *ci,
688 			    struct buffer_head *bh, int type)
689 {
690 	return __ocfs2_journal_access(handle, ci, bh, &gd_triggers, type);
691 }
692 
693 int ocfs2_journal_access_db(handle_t *handle, struct ocfs2_caching_info *ci,
694 			    struct buffer_head *bh, int type)
695 {
696 	return __ocfs2_journal_access(handle, ci, bh, &db_triggers, type);
697 }
698 
699 int ocfs2_journal_access_xb(handle_t *handle, struct ocfs2_caching_info *ci,
700 			    struct buffer_head *bh, int type)
701 {
702 	return __ocfs2_journal_access(handle, ci, bh, &xb_triggers, type);
703 }
704 
705 int ocfs2_journal_access_dq(handle_t *handle, struct ocfs2_caching_info *ci,
706 			    struct buffer_head *bh, int type)
707 {
708 	return __ocfs2_journal_access(handle, ci, bh, &dq_triggers, type);
709 }
710 
711 int ocfs2_journal_access_dr(handle_t *handle, struct ocfs2_caching_info *ci,
712 			    struct buffer_head *bh, int type)
713 {
714 	return __ocfs2_journal_access(handle, ci, bh, &dr_triggers, type);
715 }
716 
717 int ocfs2_journal_access_dl(handle_t *handle, struct ocfs2_caching_info *ci,
718 			    struct buffer_head *bh, int type)
719 {
720 	return __ocfs2_journal_access(handle, ci, bh, &dl_triggers, type);
721 }
722 
723 int ocfs2_journal_access(handle_t *handle, struct ocfs2_caching_info *ci,
724 			 struct buffer_head *bh, int type)
725 {
726 	return __ocfs2_journal_access(handle, ci, bh, NULL, type);
727 }
728 
729 void ocfs2_journal_dirty(handle_t *handle, struct buffer_head *bh)
730 {
731 	int status;
732 
733 	trace_ocfs2_journal_dirty((unsigned long long)bh->b_blocknr);
734 
735 	status = jbd2_journal_dirty_metadata(handle, bh);
736 	BUG_ON(status);
737 }
738 
739 #define OCFS2_DEFAULT_COMMIT_INTERVAL	(HZ * JBD2_DEFAULT_MAX_COMMIT_AGE)
740 
741 void ocfs2_set_journal_params(struct ocfs2_super *osb)
742 {
743 	journal_t *journal = osb->journal->j_journal;
744 	unsigned long commit_interval = OCFS2_DEFAULT_COMMIT_INTERVAL;
745 
746 	if (osb->osb_commit_interval)
747 		commit_interval = osb->osb_commit_interval;
748 
749 	write_lock(&journal->j_state_lock);
750 	journal->j_commit_interval = commit_interval;
751 	if (osb->s_mount_opt & OCFS2_MOUNT_BARRIER)
752 		journal->j_flags |= JBD2_BARRIER;
753 	else
754 		journal->j_flags &= ~JBD2_BARRIER;
755 	write_unlock(&journal->j_state_lock);
756 }
757 
758 int ocfs2_journal_init(struct ocfs2_journal *journal, int *dirty)
759 {
760 	int status = -1;
761 	struct inode *inode = NULL; /* the journal inode */
762 	journal_t *j_journal = NULL;
763 	struct ocfs2_dinode *di = NULL;
764 	struct buffer_head *bh = NULL;
765 	struct ocfs2_super *osb;
766 	int inode_lock = 0;
767 
768 	BUG_ON(!journal);
769 
770 	osb = journal->j_osb;
771 
772 	/* already have the inode for our journal */
773 	inode = ocfs2_get_system_file_inode(osb, JOURNAL_SYSTEM_INODE,
774 					    osb->slot_num);
775 	if (inode == NULL) {
776 		status = -EACCES;
777 		mlog_errno(status);
778 		goto done;
779 	}
780 	if (is_bad_inode(inode)) {
781 		mlog(ML_ERROR, "access error (bad inode)\n");
782 		iput(inode);
783 		inode = NULL;
784 		status = -EACCES;
785 		goto done;
786 	}
787 
788 	SET_INODE_JOURNAL(inode);
789 	OCFS2_I(inode)->ip_open_count++;
790 
791 	/* Skip recovery waits here - journal inode metadata never
792 	 * changes in a live cluster so it can be considered an
793 	 * exception to the rule. */
794 	status = ocfs2_inode_lock_full(inode, &bh, 1, OCFS2_META_LOCK_RECOVERY);
795 	if (status < 0) {
796 		if (status != -ERESTARTSYS)
797 			mlog(ML_ERROR, "Could not get lock on journal!\n");
798 		goto done;
799 	}
800 
801 	inode_lock = 1;
802 	di = (struct ocfs2_dinode *)bh->b_data;
803 
804 	if (inode->i_size <  OCFS2_MIN_JOURNAL_SIZE) {
805 		mlog(ML_ERROR, "Journal file size (%lld) is too small!\n",
806 		     inode->i_size);
807 		status = -EINVAL;
808 		goto done;
809 	}
810 
811 	trace_ocfs2_journal_init(inode->i_size,
812 				 (unsigned long long)inode->i_blocks,
813 				 OCFS2_I(inode)->ip_clusters);
814 
815 	/* call the kernels journal init function now */
816 	j_journal = jbd2_journal_init_inode(inode);
817 	if (j_journal == NULL) {
818 		mlog(ML_ERROR, "Linux journal layer error\n");
819 		status = -EINVAL;
820 		goto done;
821 	}
822 
823 	trace_ocfs2_journal_init_maxlen(j_journal->j_maxlen);
824 
825 	*dirty = (le32_to_cpu(di->id1.journal1.ij_flags) &
826 		  OCFS2_JOURNAL_DIRTY_FL);
827 
828 	journal->j_journal = j_journal;
829 	journal->j_inode = inode;
830 	journal->j_bh = bh;
831 
832 	ocfs2_set_journal_params(osb);
833 
834 	journal->j_state = OCFS2_JOURNAL_LOADED;
835 
836 	status = 0;
837 done:
838 	if (status < 0) {
839 		if (inode_lock)
840 			ocfs2_inode_unlock(inode, 1);
841 		brelse(bh);
842 		if (inode) {
843 			OCFS2_I(inode)->ip_open_count--;
844 			iput(inode);
845 		}
846 	}
847 
848 	return status;
849 }
850 
851 static void ocfs2_bump_recovery_generation(struct ocfs2_dinode *di)
852 {
853 	le32_add_cpu(&(di->id1.journal1.ij_recovery_generation), 1);
854 }
855 
856 static u32 ocfs2_get_recovery_generation(struct ocfs2_dinode *di)
857 {
858 	return le32_to_cpu(di->id1.journal1.ij_recovery_generation);
859 }
860 
861 static int ocfs2_journal_toggle_dirty(struct ocfs2_super *osb,
862 				      int dirty, int replayed)
863 {
864 	int status;
865 	unsigned int flags;
866 	struct ocfs2_journal *journal = osb->journal;
867 	struct buffer_head *bh = journal->j_bh;
868 	struct ocfs2_dinode *fe;
869 
870 	fe = (struct ocfs2_dinode *)bh->b_data;
871 
872 	/* The journal bh on the osb always comes from ocfs2_journal_init()
873 	 * and was validated there inside ocfs2_inode_lock_full().  It's a
874 	 * code bug if we mess it up. */
875 	BUG_ON(!OCFS2_IS_VALID_DINODE(fe));
876 
877 	flags = le32_to_cpu(fe->id1.journal1.ij_flags);
878 	if (dirty)
879 		flags |= OCFS2_JOURNAL_DIRTY_FL;
880 	else
881 		flags &= ~OCFS2_JOURNAL_DIRTY_FL;
882 	fe->id1.journal1.ij_flags = cpu_to_le32(flags);
883 
884 	if (replayed)
885 		ocfs2_bump_recovery_generation(fe);
886 
887 	ocfs2_compute_meta_ecc(osb->sb, bh->b_data, &fe->i_check);
888 	status = ocfs2_write_block(osb, bh, INODE_CACHE(journal->j_inode));
889 	if (status < 0)
890 		mlog_errno(status);
891 
892 	return status;
893 }
894 
895 /*
896  * If the journal has been kmalloc'd it needs to be freed after this
897  * call.
898  */
899 void ocfs2_journal_shutdown(struct ocfs2_super *osb)
900 {
901 	struct ocfs2_journal *journal = NULL;
902 	int status = 0;
903 	struct inode *inode = NULL;
904 	int num_running_trans = 0;
905 
906 	BUG_ON(!osb);
907 
908 	journal = osb->journal;
909 	if (!journal)
910 		goto done;
911 
912 	inode = journal->j_inode;
913 
914 	if (journal->j_state != OCFS2_JOURNAL_LOADED)
915 		goto done;
916 
917 	/* need to inc inode use count - jbd2_journal_destroy will iput. */
918 	if (!igrab(inode))
919 		BUG();
920 
921 	num_running_trans = atomic_read(&(osb->journal->j_num_trans));
922 	trace_ocfs2_journal_shutdown(num_running_trans);
923 
924 	/* Do a commit_cache here. It will flush our journal, *and*
925 	 * release any locks that are still held.
926 	 * set the SHUTDOWN flag and release the trans lock.
927 	 * the commit thread will take the trans lock for us below. */
928 	journal->j_state = OCFS2_JOURNAL_IN_SHUTDOWN;
929 
930 	/* The OCFS2_JOURNAL_IN_SHUTDOWN will signal to commit_cache to not
931 	 * drop the trans_lock (which we want to hold until we
932 	 * completely destroy the journal. */
933 	if (osb->commit_task) {
934 		/* Wait for the commit thread */
935 		trace_ocfs2_journal_shutdown_wait(osb->commit_task);
936 		kthread_stop(osb->commit_task);
937 		osb->commit_task = NULL;
938 	}
939 
940 	BUG_ON(atomic_read(&(osb->journal->j_num_trans)) != 0);
941 
942 	if (ocfs2_mount_local(osb)) {
943 		jbd2_journal_lock_updates(journal->j_journal);
944 		status = jbd2_journal_flush(journal->j_journal);
945 		jbd2_journal_unlock_updates(journal->j_journal);
946 		if (status < 0)
947 			mlog_errno(status);
948 	}
949 
950 	if (status == 0) {
951 		/*
952 		 * Do not toggle if flush was unsuccessful otherwise
953 		 * will leave dirty metadata in a "clean" journal
954 		 */
955 		status = ocfs2_journal_toggle_dirty(osb, 0, 0);
956 		if (status < 0)
957 			mlog_errno(status);
958 	}
959 
960 	/* Shutdown the kernel journal system */
961 	jbd2_journal_destroy(journal->j_journal);
962 	journal->j_journal = NULL;
963 
964 	OCFS2_I(inode)->ip_open_count--;
965 
966 	/* unlock our journal */
967 	ocfs2_inode_unlock(inode, 1);
968 
969 	brelse(journal->j_bh);
970 	journal->j_bh = NULL;
971 
972 	journal->j_state = OCFS2_JOURNAL_FREE;
973 
974 //	up_write(&journal->j_trans_barrier);
975 done:
976 	if (inode)
977 		iput(inode);
978 }
979 
980 static void ocfs2_clear_journal_error(struct super_block *sb,
981 				      journal_t *journal,
982 				      int slot)
983 {
984 	int olderr;
985 
986 	olderr = jbd2_journal_errno(journal);
987 	if (olderr) {
988 		mlog(ML_ERROR, "File system error %d recorded in "
989 		     "journal %u.\n", olderr, slot);
990 		mlog(ML_ERROR, "File system on device %s needs checking.\n",
991 		     sb->s_id);
992 
993 		jbd2_journal_ack_err(journal);
994 		jbd2_journal_clear_err(journal);
995 	}
996 }
997 
998 int ocfs2_journal_load(struct ocfs2_journal *journal, int local, int replayed)
999 {
1000 	int status = 0;
1001 	struct ocfs2_super *osb;
1002 
1003 	BUG_ON(!journal);
1004 
1005 	osb = journal->j_osb;
1006 
1007 	status = jbd2_journal_load(journal->j_journal);
1008 	if (status < 0) {
1009 		mlog(ML_ERROR, "Failed to load journal!\n");
1010 		goto done;
1011 	}
1012 
1013 	ocfs2_clear_journal_error(osb->sb, journal->j_journal, osb->slot_num);
1014 
1015 	status = ocfs2_journal_toggle_dirty(osb, 1, replayed);
1016 	if (status < 0) {
1017 		mlog_errno(status);
1018 		goto done;
1019 	}
1020 
1021 	/* Launch the commit thread */
1022 	if (!local) {
1023 		osb->commit_task = kthread_run(ocfs2_commit_thread, osb,
1024 					       "ocfs2cmt");
1025 		if (IS_ERR(osb->commit_task)) {
1026 			status = PTR_ERR(osb->commit_task);
1027 			osb->commit_task = NULL;
1028 			mlog(ML_ERROR, "unable to launch ocfs2commit thread, "
1029 			     "error=%d", status);
1030 			goto done;
1031 		}
1032 	} else
1033 		osb->commit_task = NULL;
1034 
1035 done:
1036 	return status;
1037 }
1038 
1039 
1040 /* 'full' flag tells us whether we clear out all blocks or if we just
1041  * mark the journal clean */
1042 int ocfs2_journal_wipe(struct ocfs2_journal *journal, int full)
1043 {
1044 	int status;
1045 
1046 	BUG_ON(!journal);
1047 
1048 	status = jbd2_journal_wipe(journal->j_journal, full);
1049 	if (status < 0) {
1050 		mlog_errno(status);
1051 		goto bail;
1052 	}
1053 
1054 	status = ocfs2_journal_toggle_dirty(journal->j_osb, 0, 0);
1055 	if (status < 0)
1056 		mlog_errno(status);
1057 
1058 bail:
1059 	return status;
1060 }
1061 
1062 static int ocfs2_recovery_completed(struct ocfs2_super *osb)
1063 {
1064 	int empty;
1065 	struct ocfs2_recovery_map *rm = osb->recovery_map;
1066 
1067 	spin_lock(&osb->osb_lock);
1068 	empty = (rm->rm_used == 0);
1069 	spin_unlock(&osb->osb_lock);
1070 
1071 	return empty;
1072 }
1073 
1074 void ocfs2_wait_for_recovery(struct ocfs2_super *osb)
1075 {
1076 	wait_event(osb->recovery_event, ocfs2_recovery_completed(osb));
1077 }
1078 
1079 /*
1080  * JBD Might read a cached version of another nodes journal file. We
1081  * don't want this as this file changes often and we get no
1082  * notification on those changes. The only way to be sure that we've
1083  * got the most up to date version of those blocks then is to force
1084  * read them off disk. Just searching through the buffer cache won't
1085  * work as there may be pages backing this file which are still marked
1086  * up to date. We know things can't change on this file underneath us
1087  * as we have the lock by now :)
1088  */
1089 static int ocfs2_force_read_journal(struct inode *inode)
1090 {
1091 	int status = 0;
1092 	int i;
1093 	u64 v_blkno, p_blkno, p_blocks, num_blocks;
1094 #define CONCURRENT_JOURNAL_FILL 32ULL
1095 	struct buffer_head *bhs[CONCURRENT_JOURNAL_FILL];
1096 
1097 	memset(bhs, 0, sizeof(struct buffer_head *) * CONCURRENT_JOURNAL_FILL);
1098 
1099 	num_blocks = ocfs2_blocks_for_bytes(inode->i_sb, inode->i_size);
1100 	v_blkno = 0;
1101 	while (v_blkno < num_blocks) {
1102 		status = ocfs2_extent_map_get_blocks(inode, v_blkno,
1103 						     &p_blkno, &p_blocks, NULL);
1104 		if (status < 0) {
1105 			mlog_errno(status);
1106 			goto bail;
1107 		}
1108 
1109 		if (p_blocks > CONCURRENT_JOURNAL_FILL)
1110 			p_blocks = CONCURRENT_JOURNAL_FILL;
1111 
1112 		/* We are reading journal data which should not
1113 		 * be put in the uptodate cache */
1114 		status = ocfs2_read_blocks_sync(OCFS2_SB(inode->i_sb),
1115 						p_blkno, p_blocks, bhs);
1116 		if (status < 0) {
1117 			mlog_errno(status);
1118 			goto bail;
1119 		}
1120 
1121 		for(i = 0; i < p_blocks; i++) {
1122 			brelse(bhs[i]);
1123 			bhs[i] = NULL;
1124 		}
1125 
1126 		v_blkno += p_blocks;
1127 	}
1128 
1129 bail:
1130 	for(i = 0; i < CONCURRENT_JOURNAL_FILL; i++)
1131 		brelse(bhs[i]);
1132 	return status;
1133 }
1134 
1135 struct ocfs2_la_recovery_item {
1136 	struct list_head	lri_list;
1137 	int			lri_slot;
1138 	struct ocfs2_dinode	*lri_la_dinode;
1139 	struct ocfs2_dinode	*lri_tl_dinode;
1140 	struct ocfs2_quota_recovery *lri_qrec;
1141 };
1142 
1143 /* Does the second half of the recovery process. By this point, the
1144  * node is marked clean and can actually be considered recovered,
1145  * hence it's no longer in the recovery map, but there's still some
1146  * cleanup we can do which shouldn't happen within the recovery thread
1147  * as locking in that context becomes very difficult if we are to take
1148  * recovering nodes into account.
1149  *
1150  * NOTE: This function can and will sleep on recovery of other nodes
1151  * during cluster locking, just like any other ocfs2 process.
1152  */
1153 void ocfs2_complete_recovery(struct work_struct *work)
1154 {
1155 	int ret = 0;
1156 	struct ocfs2_journal *journal =
1157 		container_of(work, struct ocfs2_journal, j_recovery_work);
1158 	struct ocfs2_super *osb = journal->j_osb;
1159 	struct ocfs2_dinode *la_dinode, *tl_dinode;
1160 	struct ocfs2_la_recovery_item *item, *n;
1161 	struct ocfs2_quota_recovery *qrec;
1162 	LIST_HEAD(tmp_la_list);
1163 
1164 	trace_ocfs2_complete_recovery(
1165 		(unsigned long long)OCFS2_I(journal->j_inode)->ip_blkno);
1166 
1167 	spin_lock(&journal->j_lock);
1168 	list_splice_init(&journal->j_la_cleanups, &tmp_la_list);
1169 	spin_unlock(&journal->j_lock);
1170 
1171 	list_for_each_entry_safe(item, n, &tmp_la_list, lri_list) {
1172 		list_del_init(&item->lri_list);
1173 
1174 		ocfs2_wait_on_quotas(osb);
1175 
1176 		la_dinode = item->lri_la_dinode;
1177 		tl_dinode = item->lri_tl_dinode;
1178 		qrec = item->lri_qrec;
1179 
1180 		trace_ocfs2_complete_recovery_slot(item->lri_slot,
1181 			la_dinode ? le64_to_cpu(la_dinode->i_blkno) : 0,
1182 			tl_dinode ? le64_to_cpu(tl_dinode->i_blkno) : 0,
1183 			qrec);
1184 
1185 		if (la_dinode) {
1186 			ret = ocfs2_complete_local_alloc_recovery(osb,
1187 								  la_dinode);
1188 			if (ret < 0)
1189 				mlog_errno(ret);
1190 
1191 			kfree(la_dinode);
1192 		}
1193 
1194 		if (tl_dinode) {
1195 			ret = ocfs2_complete_truncate_log_recovery(osb,
1196 								   tl_dinode);
1197 			if (ret < 0)
1198 				mlog_errno(ret);
1199 
1200 			kfree(tl_dinode);
1201 		}
1202 
1203 		ret = ocfs2_recover_orphans(osb, item->lri_slot);
1204 		if (ret < 0)
1205 			mlog_errno(ret);
1206 
1207 		if (qrec) {
1208 			ret = ocfs2_finish_quota_recovery(osb, qrec,
1209 							  item->lri_slot);
1210 			if (ret < 0)
1211 				mlog_errno(ret);
1212 			/* Recovery info is already freed now */
1213 		}
1214 
1215 		kfree(item);
1216 	}
1217 
1218 	trace_ocfs2_complete_recovery_end(ret);
1219 }
1220 
1221 /* NOTE: This function always eats your references to la_dinode and
1222  * tl_dinode, either manually on error, or by passing them to
1223  * ocfs2_complete_recovery */
1224 static void ocfs2_queue_recovery_completion(struct ocfs2_journal *journal,
1225 					    int slot_num,
1226 					    struct ocfs2_dinode *la_dinode,
1227 					    struct ocfs2_dinode *tl_dinode,
1228 					    struct ocfs2_quota_recovery *qrec)
1229 {
1230 	struct ocfs2_la_recovery_item *item;
1231 
1232 	item = kmalloc(sizeof(struct ocfs2_la_recovery_item), GFP_NOFS);
1233 	if (!item) {
1234 		/* Though we wish to avoid it, we are in fact safe in
1235 		 * skipping local alloc cleanup as fsck.ocfs2 is more
1236 		 * than capable of reclaiming unused space. */
1237 		if (la_dinode)
1238 			kfree(la_dinode);
1239 
1240 		if (tl_dinode)
1241 			kfree(tl_dinode);
1242 
1243 		if (qrec)
1244 			ocfs2_free_quota_recovery(qrec);
1245 
1246 		mlog_errno(-ENOMEM);
1247 		return;
1248 	}
1249 
1250 	INIT_LIST_HEAD(&item->lri_list);
1251 	item->lri_la_dinode = la_dinode;
1252 	item->lri_slot = slot_num;
1253 	item->lri_tl_dinode = tl_dinode;
1254 	item->lri_qrec = qrec;
1255 
1256 	spin_lock(&journal->j_lock);
1257 	list_add_tail(&item->lri_list, &journal->j_la_cleanups);
1258 	queue_work(ocfs2_wq, &journal->j_recovery_work);
1259 	spin_unlock(&journal->j_lock);
1260 }
1261 
1262 /* Called by the mount code to queue recovery the last part of
1263  * recovery for it's own and offline slot(s). */
1264 void ocfs2_complete_mount_recovery(struct ocfs2_super *osb)
1265 {
1266 	struct ocfs2_journal *journal = osb->journal;
1267 
1268 	if (ocfs2_is_hard_readonly(osb))
1269 		return;
1270 
1271 	/* No need to queue up our truncate_log as regular cleanup will catch
1272 	 * that */
1273 	ocfs2_queue_recovery_completion(journal, osb->slot_num,
1274 					osb->local_alloc_copy, NULL, NULL);
1275 	ocfs2_schedule_truncate_log_flush(osb, 0);
1276 
1277 	osb->local_alloc_copy = NULL;
1278 	osb->dirty = 0;
1279 
1280 	/* queue to recover orphan slots for all offline slots */
1281 	ocfs2_replay_map_set_state(osb, REPLAY_NEEDED);
1282 	ocfs2_queue_replay_slots(osb);
1283 	ocfs2_free_replay_slots(osb);
1284 }
1285 
1286 void ocfs2_complete_quota_recovery(struct ocfs2_super *osb)
1287 {
1288 	if (osb->quota_rec) {
1289 		ocfs2_queue_recovery_completion(osb->journal,
1290 						osb->slot_num,
1291 						NULL,
1292 						NULL,
1293 						osb->quota_rec);
1294 		osb->quota_rec = NULL;
1295 	}
1296 }
1297 
1298 static int __ocfs2_recovery_thread(void *arg)
1299 {
1300 	int status, node_num, slot_num;
1301 	struct ocfs2_super *osb = arg;
1302 	struct ocfs2_recovery_map *rm = osb->recovery_map;
1303 	int *rm_quota = NULL;
1304 	int rm_quota_used = 0, i;
1305 	struct ocfs2_quota_recovery *qrec;
1306 
1307 	status = ocfs2_wait_on_mount(osb);
1308 	if (status < 0) {
1309 		goto bail;
1310 	}
1311 
1312 	rm_quota = kzalloc(osb->max_slots * sizeof(int), GFP_NOFS);
1313 	if (!rm_quota) {
1314 		status = -ENOMEM;
1315 		goto bail;
1316 	}
1317 restart:
1318 	status = ocfs2_super_lock(osb, 1);
1319 	if (status < 0) {
1320 		mlog_errno(status);
1321 		goto bail;
1322 	}
1323 
1324 	status = ocfs2_compute_replay_slots(osb);
1325 	if (status < 0)
1326 		mlog_errno(status);
1327 
1328 	/* queue recovery for our own slot */
1329 	ocfs2_queue_recovery_completion(osb->journal, osb->slot_num, NULL,
1330 					NULL, NULL);
1331 
1332 	spin_lock(&osb->osb_lock);
1333 	while (rm->rm_used) {
1334 		/* It's always safe to remove entry zero, as we won't
1335 		 * clear it until ocfs2_recover_node() has succeeded. */
1336 		node_num = rm->rm_entries[0];
1337 		spin_unlock(&osb->osb_lock);
1338 		slot_num = ocfs2_node_num_to_slot(osb, node_num);
1339 		trace_ocfs2_recovery_thread_node(node_num, slot_num);
1340 		if (slot_num == -ENOENT) {
1341 			status = 0;
1342 			goto skip_recovery;
1343 		}
1344 
1345 		/* It is a bit subtle with quota recovery. We cannot do it
1346 		 * immediately because we have to obtain cluster locks from
1347 		 * quota files and we also don't want to just skip it because
1348 		 * then quota usage would be out of sync until some node takes
1349 		 * the slot. So we remember which nodes need quota recovery
1350 		 * and when everything else is done, we recover quotas. */
1351 		for (i = 0; i < rm_quota_used && rm_quota[i] != slot_num; i++);
1352 		if (i == rm_quota_used)
1353 			rm_quota[rm_quota_used++] = slot_num;
1354 
1355 		status = ocfs2_recover_node(osb, node_num, slot_num);
1356 skip_recovery:
1357 		if (!status) {
1358 			ocfs2_recovery_map_clear(osb, node_num);
1359 		} else {
1360 			mlog(ML_ERROR,
1361 			     "Error %d recovering node %d on device (%u,%u)!\n",
1362 			     status, node_num,
1363 			     MAJOR(osb->sb->s_dev), MINOR(osb->sb->s_dev));
1364 			mlog(ML_ERROR, "Volume requires unmount.\n");
1365 		}
1366 
1367 		spin_lock(&osb->osb_lock);
1368 	}
1369 	spin_unlock(&osb->osb_lock);
1370 	trace_ocfs2_recovery_thread_end(status);
1371 
1372 	/* Refresh all journal recovery generations from disk */
1373 	status = ocfs2_check_journals_nolocks(osb);
1374 	status = (status == -EROFS) ? 0 : status;
1375 	if (status < 0)
1376 		mlog_errno(status);
1377 
1378 	/* Now it is right time to recover quotas... We have to do this under
1379 	 * superblock lock so that no one can start using the slot (and crash)
1380 	 * before we recover it */
1381 	for (i = 0; i < rm_quota_used; i++) {
1382 		qrec = ocfs2_begin_quota_recovery(osb, rm_quota[i]);
1383 		if (IS_ERR(qrec)) {
1384 			status = PTR_ERR(qrec);
1385 			mlog_errno(status);
1386 			continue;
1387 		}
1388 		ocfs2_queue_recovery_completion(osb->journal, rm_quota[i],
1389 						NULL, NULL, qrec);
1390 	}
1391 
1392 	ocfs2_super_unlock(osb, 1);
1393 
1394 	/* queue recovery for offline slots */
1395 	ocfs2_queue_replay_slots(osb);
1396 
1397 bail:
1398 	mutex_lock(&osb->recovery_lock);
1399 	if (!status && !ocfs2_recovery_completed(osb)) {
1400 		mutex_unlock(&osb->recovery_lock);
1401 		goto restart;
1402 	}
1403 
1404 	ocfs2_free_replay_slots(osb);
1405 	osb->recovery_thread_task = NULL;
1406 	mb(); /* sync with ocfs2_recovery_thread_running */
1407 	wake_up(&osb->recovery_event);
1408 
1409 	mutex_unlock(&osb->recovery_lock);
1410 
1411 	if (rm_quota)
1412 		kfree(rm_quota);
1413 
1414 	/* no one is callint kthread_stop() for us so the kthread() api
1415 	 * requires that we call do_exit().  And it isn't exported, but
1416 	 * complete_and_exit() seems to be a minimal wrapper around it. */
1417 	complete_and_exit(NULL, status);
1418 	return status;
1419 }
1420 
1421 void ocfs2_recovery_thread(struct ocfs2_super *osb, int node_num)
1422 {
1423 	mutex_lock(&osb->recovery_lock);
1424 
1425 	trace_ocfs2_recovery_thread(node_num, osb->node_num,
1426 		osb->disable_recovery, osb->recovery_thread_task,
1427 		osb->disable_recovery ?
1428 		-1 : ocfs2_recovery_map_set(osb, node_num));
1429 
1430 	if (osb->disable_recovery)
1431 		goto out;
1432 
1433 	if (osb->recovery_thread_task)
1434 		goto out;
1435 
1436 	osb->recovery_thread_task =  kthread_run(__ocfs2_recovery_thread, osb,
1437 						 "ocfs2rec");
1438 	if (IS_ERR(osb->recovery_thread_task)) {
1439 		mlog_errno((int)PTR_ERR(osb->recovery_thread_task));
1440 		osb->recovery_thread_task = NULL;
1441 	}
1442 
1443 out:
1444 	mutex_unlock(&osb->recovery_lock);
1445 	wake_up(&osb->recovery_event);
1446 }
1447 
1448 static int ocfs2_read_journal_inode(struct ocfs2_super *osb,
1449 				    int slot_num,
1450 				    struct buffer_head **bh,
1451 				    struct inode **ret_inode)
1452 {
1453 	int status = -EACCES;
1454 	struct inode *inode = NULL;
1455 
1456 	BUG_ON(slot_num >= osb->max_slots);
1457 
1458 	inode = ocfs2_get_system_file_inode(osb, JOURNAL_SYSTEM_INODE,
1459 					    slot_num);
1460 	if (!inode || is_bad_inode(inode)) {
1461 		mlog_errno(status);
1462 		goto bail;
1463 	}
1464 	SET_INODE_JOURNAL(inode);
1465 
1466 	status = ocfs2_read_inode_block_full(inode, bh, OCFS2_BH_IGNORE_CACHE);
1467 	if (status < 0) {
1468 		mlog_errno(status);
1469 		goto bail;
1470 	}
1471 
1472 	status = 0;
1473 
1474 bail:
1475 	if (inode) {
1476 		if (status || !ret_inode)
1477 			iput(inode);
1478 		else
1479 			*ret_inode = inode;
1480 	}
1481 	return status;
1482 }
1483 
1484 /* Does the actual journal replay and marks the journal inode as
1485  * clean. Will only replay if the journal inode is marked dirty. */
1486 static int ocfs2_replay_journal(struct ocfs2_super *osb,
1487 				int node_num,
1488 				int slot_num)
1489 {
1490 	int status;
1491 	int got_lock = 0;
1492 	unsigned int flags;
1493 	struct inode *inode = NULL;
1494 	struct ocfs2_dinode *fe;
1495 	journal_t *journal = NULL;
1496 	struct buffer_head *bh = NULL;
1497 	u32 slot_reco_gen;
1498 
1499 	status = ocfs2_read_journal_inode(osb, slot_num, &bh, &inode);
1500 	if (status) {
1501 		mlog_errno(status);
1502 		goto done;
1503 	}
1504 
1505 	fe = (struct ocfs2_dinode *)bh->b_data;
1506 	slot_reco_gen = ocfs2_get_recovery_generation(fe);
1507 	brelse(bh);
1508 	bh = NULL;
1509 
1510 	/*
1511 	 * As the fs recovery is asynchronous, there is a small chance that
1512 	 * another node mounted (and recovered) the slot before the recovery
1513 	 * thread could get the lock. To handle that, we dirty read the journal
1514 	 * inode for that slot to get the recovery generation. If it is
1515 	 * different than what we expected, the slot has been recovered.
1516 	 * If not, it needs recovery.
1517 	 */
1518 	if (osb->slot_recovery_generations[slot_num] != slot_reco_gen) {
1519 		trace_ocfs2_replay_journal_recovered(slot_num,
1520 		     osb->slot_recovery_generations[slot_num], slot_reco_gen);
1521 		osb->slot_recovery_generations[slot_num] = slot_reco_gen;
1522 		status = -EBUSY;
1523 		goto done;
1524 	}
1525 
1526 	/* Continue with recovery as the journal has not yet been recovered */
1527 
1528 	status = ocfs2_inode_lock_full(inode, &bh, 1, OCFS2_META_LOCK_RECOVERY);
1529 	if (status < 0) {
1530 		trace_ocfs2_replay_journal_lock_err(status);
1531 		if (status != -ERESTARTSYS)
1532 			mlog(ML_ERROR, "Could not lock journal!\n");
1533 		goto done;
1534 	}
1535 	got_lock = 1;
1536 
1537 	fe = (struct ocfs2_dinode *) bh->b_data;
1538 
1539 	flags = le32_to_cpu(fe->id1.journal1.ij_flags);
1540 	slot_reco_gen = ocfs2_get_recovery_generation(fe);
1541 
1542 	if (!(flags & OCFS2_JOURNAL_DIRTY_FL)) {
1543 		trace_ocfs2_replay_journal_skip(node_num);
1544 		/* Refresh recovery generation for the slot */
1545 		osb->slot_recovery_generations[slot_num] = slot_reco_gen;
1546 		goto done;
1547 	}
1548 
1549 	/* we need to run complete recovery for offline orphan slots */
1550 	ocfs2_replay_map_set_state(osb, REPLAY_NEEDED);
1551 
1552 	printk(KERN_NOTICE "ocfs2: Begin replay journal (node %d, slot %d) on "\
1553 	       "device (%u,%u)\n", node_num, slot_num, MAJOR(osb->sb->s_dev),
1554 	       MINOR(osb->sb->s_dev));
1555 
1556 	OCFS2_I(inode)->ip_clusters = le32_to_cpu(fe->i_clusters);
1557 
1558 	status = ocfs2_force_read_journal(inode);
1559 	if (status < 0) {
1560 		mlog_errno(status);
1561 		goto done;
1562 	}
1563 
1564 	journal = jbd2_journal_init_inode(inode);
1565 	if (journal == NULL) {
1566 		mlog(ML_ERROR, "Linux journal layer error\n");
1567 		status = -EIO;
1568 		goto done;
1569 	}
1570 
1571 	status = jbd2_journal_load(journal);
1572 	if (status < 0) {
1573 		mlog_errno(status);
1574 		if (!igrab(inode))
1575 			BUG();
1576 		jbd2_journal_destroy(journal);
1577 		goto done;
1578 	}
1579 
1580 	ocfs2_clear_journal_error(osb->sb, journal, slot_num);
1581 
1582 	/* wipe the journal */
1583 	jbd2_journal_lock_updates(journal);
1584 	status = jbd2_journal_flush(journal);
1585 	jbd2_journal_unlock_updates(journal);
1586 	if (status < 0)
1587 		mlog_errno(status);
1588 
1589 	/* This will mark the node clean */
1590 	flags = le32_to_cpu(fe->id1.journal1.ij_flags);
1591 	flags &= ~OCFS2_JOURNAL_DIRTY_FL;
1592 	fe->id1.journal1.ij_flags = cpu_to_le32(flags);
1593 
1594 	/* Increment recovery generation to indicate successful recovery */
1595 	ocfs2_bump_recovery_generation(fe);
1596 	osb->slot_recovery_generations[slot_num] =
1597 					ocfs2_get_recovery_generation(fe);
1598 
1599 	ocfs2_compute_meta_ecc(osb->sb, bh->b_data, &fe->i_check);
1600 	status = ocfs2_write_block(osb, bh, INODE_CACHE(inode));
1601 	if (status < 0)
1602 		mlog_errno(status);
1603 
1604 	if (!igrab(inode))
1605 		BUG();
1606 
1607 	jbd2_journal_destroy(journal);
1608 
1609 	printk(KERN_NOTICE "ocfs2: End replay journal (node %d, slot %d) on "\
1610 	       "device (%u,%u)\n", node_num, slot_num, MAJOR(osb->sb->s_dev),
1611 	       MINOR(osb->sb->s_dev));
1612 done:
1613 	/* drop the lock on this nodes journal */
1614 	if (got_lock)
1615 		ocfs2_inode_unlock(inode, 1);
1616 
1617 	if (inode)
1618 		iput(inode);
1619 
1620 	brelse(bh);
1621 
1622 	return status;
1623 }
1624 
1625 /*
1626  * Do the most important parts of node recovery:
1627  *  - Replay it's journal
1628  *  - Stamp a clean local allocator file
1629  *  - Stamp a clean truncate log
1630  *  - Mark the node clean
1631  *
1632  * If this function completes without error, a node in OCFS2 can be
1633  * said to have been safely recovered. As a result, failure during the
1634  * second part of a nodes recovery process (local alloc recovery) is
1635  * far less concerning.
1636  */
1637 static int ocfs2_recover_node(struct ocfs2_super *osb,
1638 			      int node_num, int slot_num)
1639 {
1640 	int status = 0;
1641 	struct ocfs2_dinode *la_copy = NULL;
1642 	struct ocfs2_dinode *tl_copy = NULL;
1643 
1644 	trace_ocfs2_recover_node(node_num, slot_num, osb->node_num);
1645 
1646 	/* Should not ever be called to recover ourselves -- in that
1647 	 * case we should've called ocfs2_journal_load instead. */
1648 	BUG_ON(osb->node_num == node_num);
1649 
1650 	status = ocfs2_replay_journal(osb, node_num, slot_num);
1651 	if (status < 0) {
1652 		if (status == -EBUSY) {
1653 			trace_ocfs2_recover_node_skip(slot_num, node_num);
1654 			status = 0;
1655 			goto done;
1656 		}
1657 		mlog_errno(status);
1658 		goto done;
1659 	}
1660 
1661 	/* Stamp a clean local alloc file AFTER recovering the journal... */
1662 	status = ocfs2_begin_local_alloc_recovery(osb, slot_num, &la_copy);
1663 	if (status < 0) {
1664 		mlog_errno(status);
1665 		goto done;
1666 	}
1667 
1668 	/* An error from begin_truncate_log_recovery is not
1669 	 * serious enough to warrant halting the rest of
1670 	 * recovery. */
1671 	status = ocfs2_begin_truncate_log_recovery(osb, slot_num, &tl_copy);
1672 	if (status < 0)
1673 		mlog_errno(status);
1674 
1675 	/* Likewise, this would be a strange but ultimately not so
1676 	 * harmful place to get an error... */
1677 	status = ocfs2_clear_slot(osb, slot_num);
1678 	if (status < 0)
1679 		mlog_errno(status);
1680 
1681 	/* This will kfree the memory pointed to by la_copy and tl_copy */
1682 	ocfs2_queue_recovery_completion(osb->journal, slot_num, la_copy,
1683 					tl_copy, NULL);
1684 
1685 	status = 0;
1686 done:
1687 
1688 	return status;
1689 }
1690 
1691 /* Test node liveness by trylocking his journal. If we get the lock,
1692  * we drop it here. Return 0 if we got the lock, -EAGAIN if node is
1693  * still alive (we couldn't get the lock) and < 0 on error. */
1694 static int ocfs2_trylock_journal(struct ocfs2_super *osb,
1695 				 int slot_num)
1696 {
1697 	int status, flags;
1698 	struct inode *inode = NULL;
1699 
1700 	inode = ocfs2_get_system_file_inode(osb, JOURNAL_SYSTEM_INODE,
1701 					    slot_num);
1702 	if (inode == NULL) {
1703 		mlog(ML_ERROR, "access error\n");
1704 		status = -EACCES;
1705 		goto bail;
1706 	}
1707 	if (is_bad_inode(inode)) {
1708 		mlog(ML_ERROR, "access error (bad inode)\n");
1709 		iput(inode);
1710 		inode = NULL;
1711 		status = -EACCES;
1712 		goto bail;
1713 	}
1714 	SET_INODE_JOURNAL(inode);
1715 
1716 	flags = OCFS2_META_LOCK_RECOVERY | OCFS2_META_LOCK_NOQUEUE;
1717 	status = ocfs2_inode_lock_full(inode, NULL, 1, flags);
1718 	if (status < 0) {
1719 		if (status != -EAGAIN)
1720 			mlog_errno(status);
1721 		goto bail;
1722 	}
1723 
1724 	ocfs2_inode_unlock(inode, 1);
1725 bail:
1726 	if (inode)
1727 		iput(inode);
1728 
1729 	return status;
1730 }
1731 
1732 /* Call this underneath ocfs2_super_lock. It also assumes that the
1733  * slot info struct has been updated from disk. */
1734 int ocfs2_mark_dead_nodes(struct ocfs2_super *osb)
1735 {
1736 	unsigned int node_num;
1737 	int status, i;
1738 	u32 gen;
1739 	struct buffer_head *bh = NULL;
1740 	struct ocfs2_dinode *di;
1741 
1742 	/* This is called with the super block cluster lock, so we
1743 	 * know that the slot map can't change underneath us. */
1744 
1745 	for (i = 0; i < osb->max_slots; i++) {
1746 		/* Read journal inode to get the recovery generation */
1747 		status = ocfs2_read_journal_inode(osb, i, &bh, NULL);
1748 		if (status) {
1749 			mlog_errno(status);
1750 			goto bail;
1751 		}
1752 		di = (struct ocfs2_dinode *)bh->b_data;
1753 		gen = ocfs2_get_recovery_generation(di);
1754 		brelse(bh);
1755 		bh = NULL;
1756 
1757 		spin_lock(&osb->osb_lock);
1758 		osb->slot_recovery_generations[i] = gen;
1759 
1760 		trace_ocfs2_mark_dead_nodes(i,
1761 					    osb->slot_recovery_generations[i]);
1762 
1763 		if (i == osb->slot_num) {
1764 			spin_unlock(&osb->osb_lock);
1765 			continue;
1766 		}
1767 
1768 		status = ocfs2_slot_to_node_num_locked(osb, i, &node_num);
1769 		if (status == -ENOENT) {
1770 			spin_unlock(&osb->osb_lock);
1771 			continue;
1772 		}
1773 
1774 		if (__ocfs2_recovery_map_test(osb, node_num)) {
1775 			spin_unlock(&osb->osb_lock);
1776 			continue;
1777 		}
1778 		spin_unlock(&osb->osb_lock);
1779 
1780 		/* Ok, we have a slot occupied by another node which
1781 		 * is not in the recovery map. We trylock his journal
1782 		 * file here to test if he's alive. */
1783 		status = ocfs2_trylock_journal(osb, i);
1784 		if (!status) {
1785 			/* Since we're called from mount, we know that
1786 			 * the recovery thread can't race us on
1787 			 * setting / checking the recovery bits. */
1788 			ocfs2_recovery_thread(osb, node_num);
1789 		} else if ((status < 0) && (status != -EAGAIN)) {
1790 			mlog_errno(status);
1791 			goto bail;
1792 		}
1793 	}
1794 
1795 	status = 0;
1796 bail:
1797 	return status;
1798 }
1799 
1800 /*
1801  * Scan timer should get fired every ORPHAN_SCAN_SCHEDULE_TIMEOUT. Add some
1802  * randomness to the timeout to minimize multple nodes firing the timer at the
1803  * same time.
1804  */
1805 static inline unsigned long ocfs2_orphan_scan_timeout(void)
1806 {
1807 	unsigned long time;
1808 
1809 	get_random_bytes(&time, sizeof(time));
1810 	time = ORPHAN_SCAN_SCHEDULE_TIMEOUT + (time % 5000);
1811 	return msecs_to_jiffies(time);
1812 }
1813 
1814 /*
1815  * ocfs2_queue_orphan_scan calls ocfs2_queue_recovery_completion for
1816  * every slot, queuing a recovery of the slot on the ocfs2_wq thread. This
1817  * is done to catch any orphans that are left over in orphan directories.
1818  *
1819  * It scans all slots, even ones that are in use. It does so to handle the
1820  * case described below:
1821  *
1822  *   Node 1 has an inode it was using. The dentry went away due to memory
1823  *   pressure.  Node 1 closes the inode, but it's on the free list. The node
1824  *   has the open lock.
1825  *   Node 2 unlinks the inode. It grabs the dentry lock to notify others,
1826  *   but node 1 has no dentry and doesn't get the message. It trylocks the
1827  *   open lock, sees that another node has a PR, and does nothing.
1828  *   Later node 2 runs its orphan dir. It igets the inode, trylocks the
1829  *   open lock, sees the PR still, and does nothing.
1830  *   Basically, we have to trigger an orphan iput on node 1. The only way
1831  *   for this to happen is if node 1 runs node 2's orphan dir.
1832  *
1833  * ocfs2_queue_orphan_scan gets called every ORPHAN_SCAN_SCHEDULE_TIMEOUT
1834  * seconds.  It gets an EX lock on os_lockres and checks sequence number
1835  * stored in LVB. If the sequence number has changed, it means some other
1836  * node has done the scan.  This node skips the scan and tracks the
1837  * sequence number.  If the sequence number didn't change, it means a scan
1838  * hasn't happened.  The node queues a scan and increments the
1839  * sequence number in the LVB.
1840  */
1841 void ocfs2_queue_orphan_scan(struct ocfs2_super *osb)
1842 {
1843 	struct ocfs2_orphan_scan *os;
1844 	int status, i;
1845 	u32 seqno = 0;
1846 
1847 	os = &osb->osb_orphan_scan;
1848 
1849 	if (atomic_read(&os->os_state) == ORPHAN_SCAN_INACTIVE)
1850 		goto out;
1851 
1852 	trace_ocfs2_queue_orphan_scan_begin(os->os_count, os->os_seqno,
1853 					    atomic_read(&os->os_state));
1854 
1855 	status = ocfs2_orphan_scan_lock(osb, &seqno);
1856 	if (status < 0) {
1857 		if (status != -EAGAIN)
1858 			mlog_errno(status);
1859 		goto out;
1860 	}
1861 
1862 	/* Do no queue the tasks if the volume is being umounted */
1863 	if (atomic_read(&os->os_state) == ORPHAN_SCAN_INACTIVE)
1864 		goto unlock;
1865 
1866 	if (os->os_seqno != seqno) {
1867 		os->os_seqno = seqno;
1868 		goto unlock;
1869 	}
1870 
1871 	for (i = 0; i < osb->max_slots; i++)
1872 		ocfs2_queue_recovery_completion(osb->journal, i, NULL, NULL,
1873 						NULL);
1874 	/*
1875 	 * We queued a recovery on orphan slots, increment the sequence
1876 	 * number and update LVB so other node will skip the scan for a while
1877 	 */
1878 	seqno++;
1879 	os->os_count++;
1880 	os->os_scantime = CURRENT_TIME;
1881 unlock:
1882 	ocfs2_orphan_scan_unlock(osb, seqno);
1883 out:
1884 	trace_ocfs2_queue_orphan_scan_end(os->os_count, os->os_seqno,
1885 					  atomic_read(&os->os_state));
1886 	return;
1887 }
1888 
1889 /* Worker task that gets fired every ORPHAN_SCAN_SCHEDULE_TIMEOUT millsec */
1890 void ocfs2_orphan_scan_work(struct work_struct *work)
1891 {
1892 	struct ocfs2_orphan_scan *os;
1893 	struct ocfs2_super *osb;
1894 
1895 	os = container_of(work, struct ocfs2_orphan_scan,
1896 			  os_orphan_scan_work.work);
1897 	osb = os->os_osb;
1898 
1899 	mutex_lock(&os->os_lock);
1900 	ocfs2_queue_orphan_scan(osb);
1901 	if (atomic_read(&os->os_state) == ORPHAN_SCAN_ACTIVE)
1902 		queue_delayed_work(ocfs2_wq, &os->os_orphan_scan_work,
1903 				      ocfs2_orphan_scan_timeout());
1904 	mutex_unlock(&os->os_lock);
1905 }
1906 
1907 void ocfs2_orphan_scan_stop(struct ocfs2_super *osb)
1908 {
1909 	struct ocfs2_orphan_scan *os;
1910 
1911 	os = &osb->osb_orphan_scan;
1912 	if (atomic_read(&os->os_state) == ORPHAN_SCAN_ACTIVE) {
1913 		atomic_set(&os->os_state, ORPHAN_SCAN_INACTIVE);
1914 		mutex_lock(&os->os_lock);
1915 		cancel_delayed_work(&os->os_orphan_scan_work);
1916 		mutex_unlock(&os->os_lock);
1917 	}
1918 }
1919 
1920 void ocfs2_orphan_scan_init(struct ocfs2_super *osb)
1921 {
1922 	struct ocfs2_orphan_scan *os;
1923 
1924 	os = &osb->osb_orphan_scan;
1925 	os->os_osb = osb;
1926 	os->os_count = 0;
1927 	os->os_seqno = 0;
1928 	mutex_init(&os->os_lock);
1929 	INIT_DELAYED_WORK(&os->os_orphan_scan_work, ocfs2_orphan_scan_work);
1930 }
1931 
1932 void ocfs2_orphan_scan_start(struct ocfs2_super *osb)
1933 {
1934 	struct ocfs2_orphan_scan *os;
1935 
1936 	os = &osb->osb_orphan_scan;
1937 	os->os_scantime = CURRENT_TIME;
1938 	if (ocfs2_is_hard_readonly(osb) || ocfs2_mount_local(osb))
1939 		atomic_set(&os->os_state, ORPHAN_SCAN_INACTIVE);
1940 	else {
1941 		atomic_set(&os->os_state, ORPHAN_SCAN_ACTIVE);
1942 		queue_delayed_work(ocfs2_wq, &os->os_orphan_scan_work,
1943 				   ocfs2_orphan_scan_timeout());
1944 	}
1945 }
1946 
1947 struct ocfs2_orphan_filldir_priv {
1948 	struct inode		*head;
1949 	struct ocfs2_super	*osb;
1950 };
1951 
1952 static int ocfs2_orphan_filldir(void *priv, const char *name, int name_len,
1953 				loff_t pos, u64 ino, unsigned type)
1954 {
1955 	struct ocfs2_orphan_filldir_priv *p = priv;
1956 	struct inode *iter;
1957 
1958 	if (name_len == 1 && !strncmp(".", name, 1))
1959 		return 0;
1960 	if (name_len == 2 && !strncmp("..", name, 2))
1961 		return 0;
1962 
1963 	/* Skip bad inodes so that recovery can continue */
1964 	iter = ocfs2_iget(p->osb, ino,
1965 			  OCFS2_FI_FLAG_ORPHAN_RECOVERY, 0);
1966 	if (IS_ERR(iter))
1967 		return 0;
1968 
1969 	trace_ocfs2_orphan_filldir((unsigned long long)OCFS2_I(iter)->ip_blkno);
1970 	/* No locking is required for the next_orphan queue as there
1971 	 * is only ever a single process doing orphan recovery. */
1972 	OCFS2_I(iter)->ip_next_orphan = p->head;
1973 	p->head = iter;
1974 
1975 	return 0;
1976 }
1977 
1978 static int ocfs2_queue_orphans(struct ocfs2_super *osb,
1979 			       int slot,
1980 			       struct inode **head)
1981 {
1982 	int status;
1983 	struct inode *orphan_dir_inode = NULL;
1984 	struct ocfs2_orphan_filldir_priv priv;
1985 	loff_t pos = 0;
1986 
1987 	priv.osb = osb;
1988 	priv.head = *head;
1989 
1990 	orphan_dir_inode = ocfs2_get_system_file_inode(osb,
1991 						       ORPHAN_DIR_SYSTEM_INODE,
1992 						       slot);
1993 	if  (!orphan_dir_inode) {
1994 		status = -ENOENT;
1995 		mlog_errno(status);
1996 		return status;
1997 	}
1998 
1999 	mutex_lock(&orphan_dir_inode->i_mutex);
2000 	status = ocfs2_inode_lock(orphan_dir_inode, NULL, 0);
2001 	if (status < 0) {
2002 		mlog_errno(status);
2003 		goto out;
2004 	}
2005 
2006 	status = ocfs2_dir_foreach(orphan_dir_inode, &pos, &priv,
2007 				   ocfs2_orphan_filldir);
2008 	if (status) {
2009 		mlog_errno(status);
2010 		goto out_cluster;
2011 	}
2012 
2013 	*head = priv.head;
2014 
2015 out_cluster:
2016 	ocfs2_inode_unlock(orphan_dir_inode, 0);
2017 out:
2018 	mutex_unlock(&orphan_dir_inode->i_mutex);
2019 	iput(orphan_dir_inode);
2020 	return status;
2021 }
2022 
2023 static int ocfs2_orphan_recovery_can_continue(struct ocfs2_super *osb,
2024 					      int slot)
2025 {
2026 	int ret;
2027 
2028 	spin_lock(&osb->osb_lock);
2029 	ret = !osb->osb_orphan_wipes[slot];
2030 	spin_unlock(&osb->osb_lock);
2031 	return ret;
2032 }
2033 
2034 static void ocfs2_mark_recovering_orphan_dir(struct ocfs2_super *osb,
2035 					     int slot)
2036 {
2037 	spin_lock(&osb->osb_lock);
2038 	/* Mark ourselves such that new processes in delete_inode()
2039 	 * know to quit early. */
2040 	ocfs2_node_map_set_bit(osb, &osb->osb_recovering_orphan_dirs, slot);
2041 	while (osb->osb_orphan_wipes[slot]) {
2042 		/* If any processes are already in the middle of an
2043 		 * orphan wipe on this dir, then we need to wait for
2044 		 * them. */
2045 		spin_unlock(&osb->osb_lock);
2046 		wait_event_interruptible(osb->osb_wipe_event,
2047 					 ocfs2_orphan_recovery_can_continue(osb, slot));
2048 		spin_lock(&osb->osb_lock);
2049 	}
2050 	spin_unlock(&osb->osb_lock);
2051 }
2052 
2053 static void ocfs2_clear_recovering_orphan_dir(struct ocfs2_super *osb,
2054 					      int slot)
2055 {
2056 	ocfs2_node_map_clear_bit(osb, &osb->osb_recovering_orphan_dirs, slot);
2057 }
2058 
2059 /*
2060  * Orphan recovery. Each mounted node has it's own orphan dir which we
2061  * must run during recovery. Our strategy here is to build a list of
2062  * the inodes in the orphan dir and iget/iput them. The VFS does
2063  * (most) of the rest of the work.
2064  *
2065  * Orphan recovery can happen at any time, not just mount so we have a
2066  * couple of extra considerations.
2067  *
2068  * - We grab as many inodes as we can under the orphan dir lock -
2069  *   doing iget() outside the orphan dir risks getting a reference on
2070  *   an invalid inode.
2071  * - We must be sure not to deadlock with other processes on the
2072  *   system wanting to run delete_inode(). This can happen when they go
2073  *   to lock the orphan dir and the orphan recovery process attempts to
2074  *   iget() inside the orphan dir lock. This can be avoided by
2075  *   advertising our state to ocfs2_delete_inode().
2076  */
2077 static int ocfs2_recover_orphans(struct ocfs2_super *osb,
2078 				 int slot)
2079 {
2080 	int ret = 0;
2081 	struct inode *inode = NULL;
2082 	struct inode *iter;
2083 	struct ocfs2_inode_info *oi;
2084 
2085 	trace_ocfs2_recover_orphans(slot);
2086 
2087 	ocfs2_mark_recovering_orphan_dir(osb, slot);
2088 	ret = ocfs2_queue_orphans(osb, slot, &inode);
2089 	ocfs2_clear_recovering_orphan_dir(osb, slot);
2090 
2091 	/* Error here should be noted, but we want to continue with as
2092 	 * many queued inodes as we've got. */
2093 	if (ret)
2094 		mlog_errno(ret);
2095 
2096 	while (inode) {
2097 		oi = OCFS2_I(inode);
2098 		trace_ocfs2_recover_orphans_iput(
2099 					(unsigned long long)oi->ip_blkno);
2100 
2101 		iter = oi->ip_next_orphan;
2102 
2103 		spin_lock(&oi->ip_lock);
2104 		/* The remote delete code may have set these on the
2105 		 * assumption that the other node would wipe them
2106 		 * successfully.  If they are still in the node's
2107 		 * orphan dir, we need to reset that state. */
2108 		oi->ip_flags &= ~(OCFS2_INODE_DELETED|OCFS2_INODE_SKIP_DELETE);
2109 
2110 		/* Set the proper information to get us going into
2111 		 * ocfs2_delete_inode. */
2112 		oi->ip_flags |= OCFS2_INODE_MAYBE_ORPHANED;
2113 		spin_unlock(&oi->ip_lock);
2114 
2115 		iput(inode);
2116 
2117 		inode = iter;
2118 	}
2119 
2120 	return ret;
2121 }
2122 
2123 static int __ocfs2_wait_on_mount(struct ocfs2_super *osb, int quota)
2124 {
2125 	/* This check is good because ocfs2 will wait on our recovery
2126 	 * thread before changing it to something other than MOUNTED
2127 	 * or DISABLED. */
2128 	wait_event(osb->osb_mount_event,
2129 		  (!quota && atomic_read(&osb->vol_state) == VOLUME_MOUNTED) ||
2130 		   atomic_read(&osb->vol_state) == VOLUME_MOUNTED_QUOTAS ||
2131 		   atomic_read(&osb->vol_state) == VOLUME_DISABLED);
2132 
2133 	/* If there's an error on mount, then we may never get to the
2134 	 * MOUNTED flag, but this is set right before
2135 	 * dismount_volume() so we can trust it. */
2136 	if (atomic_read(&osb->vol_state) == VOLUME_DISABLED) {
2137 		trace_ocfs2_wait_on_mount(VOLUME_DISABLED);
2138 		mlog(0, "mount error, exiting!\n");
2139 		return -EBUSY;
2140 	}
2141 
2142 	return 0;
2143 }
2144 
2145 static int ocfs2_commit_thread(void *arg)
2146 {
2147 	int status;
2148 	struct ocfs2_super *osb = arg;
2149 	struct ocfs2_journal *journal = osb->journal;
2150 
2151 	/* we can trust j_num_trans here because _should_stop() is only set in
2152 	 * shutdown and nobody other than ourselves should be able to start
2153 	 * transactions.  committing on shutdown might take a few iterations
2154 	 * as final transactions put deleted inodes on the list */
2155 	while (!(kthread_should_stop() &&
2156 		 atomic_read(&journal->j_num_trans) == 0)) {
2157 
2158 		wait_event_interruptible(osb->checkpoint_event,
2159 					 atomic_read(&journal->j_num_trans)
2160 					 || kthread_should_stop());
2161 
2162 		status = ocfs2_commit_cache(osb);
2163 		if (status < 0)
2164 			mlog_errno(status);
2165 
2166 		if (kthread_should_stop() && atomic_read(&journal->j_num_trans)){
2167 			mlog(ML_KTHREAD,
2168 			     "commit_thread: %u transactions pending on "
2169 			     "shutdown\n",
2170 			     atomic_read(&journal->j_num_trans));
2171 		}
2172 	}
2173 
2174 	return 0;
2175 }
2176 
2177 /* Reads all the journal inodes without taking any cluster locks. Used
2178  * for hard readonly access to determine whether any journal requires
2179  * recovery. Also used to refresh the recovery generation numbers after
2180  * a journal has been recovered by another node.
2181  */
2182 int ocfs2_check_journals_nolocks(struct ocfs2_super *osb)
2183 {
2184 	int ret = 0;
2185 	unsigned int slot;
2186 	struct buffer_head *di_bh = NULL;
2187 	struct ocfs2_dinode *di;
2188 	int journal_dirty = 0;
2189 
2190 	for(slot = 0; slot < osb->max_slots; slot++) {
2191 		ret = ocfs2_read_journal_inode(osb, slot, &di_bh, NULL);
2192 		if (ret) {
2193 			mlog_errno(ret);
2194 			goto out;
2195 		}
2196 
2197 		di = (struct ocfs2_dinode *) di_bh->b_data;
2198 
2199 		osb->slot_recovery_generations[slot] =
2200 					ocfs2_get_recovery_generation(di);
2201 
2202 		if (le32_to_cpu(di->id1.journal1.ij_flags) &
2203 		    OCFS2_JOURNAL_DIRTY_FL)
2204 			journal_dirty = 1;
2205 
2206 		brelse(di_bh);
2207 		di_bh = NULL;
2208 	}
2209 
2210 out:
2211 	if (journal_dirty)
2212 		ret = -EROFS;
2213 	return ret;
2214 }
2215