xref: /openbmc/linux/fs/ocfs2/journal.c (revision 4f205687)
1 /* -*- mode: c; c-basic-offset: 8; -*-
2  * vim: noexpandtab sw=8 ts=8 sts=0:
3  *
4  * journal.c
5  *
6  * Defines functions of journalling api
7  *
8  * Copyright (C) 2003, 2004 Oracle.  All rights reserved.
9  *
10  * This program is free software; you can redistribute it and/or
11  * modify it under the terms of the GNU General Public
12  * License as published by the Free Software Foundation; either
13  * version 2 of the License, or (at your option) any later version.
14  *
15  * This program is distributed in the hope that it will be useful,
16  * but WITHOUT ANY WARRANTY; without even the implied warranty of
17  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
18  * General Public License for more details.
19  *
20  * You should have received a copy of the GNU General Public
21  * License along with this program; if not, write to the
22  * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
23  * Boston, MA 021110-1307, USA.
24  */
25 
26 #include <linux/fs.h>
27 #include <linux/types.h>
28 #include <linux/slab.h>
29 #include <linux/highmem.h>
30 #include <linux/kthread.h>
31 #include <linux/time.h>
32 #include <linux/random.h>
33 #include <linux/delay.h>
34 
35 #include <cluster/masklog.h>
36 
37 #include "ocfs2.h"
38 
39 #include "alloc.h"
40 #include "blockcheck.h"
41 #include "dir.h"
42 #include "dlmglue.h"
43 #include "extent_map.h"
44 #include "heartbeat.h"
45 #include "inode.h"
46 #include "journal.h"
47 #include "localalloc.h"
48 #include "slot_map.h"
49 #include "super.h"
50 #include "sysfile.h"
51 #include "uptodate.h"
52 #include "quota.h"
53 #include "file.h"
54 #include "namei.h"
55 
56 #include "buffer_head_io.h"
57 #include "ocfs2_trace.h"
58 
59 DEFINE_SPINLOCK(trans_inc_lock);
60 
61 #define ORPHAN_SCAN_SCHEDULE_TIMEOUT 300000
62 
63 static int ocfs2_force_read_journal(struct inode *inode);
64 static int ocfs2_recover_node(struct ocfs2_super *osb,
65 			      int node_num, int slot_num);
66 static int __ocfs2_recovery_thread(void *arg);
67 static int ocfs2_commit_cache(struct ocfs2_super *osb);
68 static int __ocfs2_wait_on_mount(struct ocfs2_super *osb, int quota);
69 static int ocfs2_journal_toggle_dirty(struct ocfs2_super *osb,
70 				      int dirty, int replayed);
71 static int ocfs2_trylock_journal(struct ocfs2_super *osb,
72 				 int slot_num);
73 static int ocfs2_recover_orphans(struct ocfs2_super *osb,
74 				 int slot,
75 				 enum ocfs2_orphan_reco_type orphan_reco_type);
76 static int ocfs2_commit_thread(void *arg);
77 static void ocfs2_queue_recovery_completion(struct ocfs2_journal *journal,
78 					    int slot_num,
79 					    struct ocfs2_dinode *la_dinode,
80 					    struct ocfs2_dinode *tl_dinode,
81 					    struct ocfs2_quota_recovery *qrec,
82 					    enum ocfs2_orphan_reco_type orphan_reco_type);
83 
84 static inline int ocfs2_wait_on_mount(struct ocfs2_super *osb)
85 {
86 	return __ocfs2_wait_on_mount(osb, 0);
87 }
88 
89 static inline int ocfs2_wait_on_quotas(struct ocfs2_super *osb)
90 {
91 	return __ocfs2_wait_on_mount(osb, 1);
92 }
93 
94 /*
95  * This replay_map is to track online/offline slots, so we could recover
96  * offline slots during recovery and mount
97  */
98 
99 enum ocfs2_replay_state {
100 	REPLAY_UNNEEDED = 0,	/* Replay is not needed, so ignore this map */
101 	REPLAY_NEEDED, 		/* Replay slots marked in rm_replay_slots */
102 	REPLAY_DONE 		/* Replay was already queued */
103 };
104 
105 struct ocfs2_replay_map {
106 	unsigned int rm_slots;
107 	enum ocfs2_replay_state rm_state;
108 	unsigned char rm_replay_slots[0];
109 };
110 
111 static void ocfs2_replay_map_set_state(struct ocfs2_super *osb, int state)
112 {
113 	if (!osb->replay_map)
114 		return;
115 
116 	/* If we've already queued the replay, we don't have any more to do */
117 	if (osb->replay_map->rm_state == REPLAY_DONE)
118 		return;
119 
120 	osb->replay_map->rm_state = state;
121 }
122 
123 int ocfs2_compute_replay_slots(struct ocfs2_super *osb)
124 {
125 	struct ocfs2_replay_map *replay_map;
126 	int i, node_num;
127 
128 	/* If replay map is already set, we don't do it again */
129 	if (osb->replay_map)
130 		return 0;
131 
132 	replay_map = kzalloc(sizeof(struct ocfs2_replay_map) +
133 			     (osb->max_slots * sizeof(char)), GFP_KERNEL);
134 
135 	if (!replay_map) {
136 		mlog_errno(-ENOMEM);
137 		return -ENOMEM;
138 	}
139 
140 	spin_lock(&osb->osb_lock);
141 
142 	replay_map->rm_slots = osb->max_slots;
143 	replay_map->rm_state = REPLAY_UNNEEDED;
144 
145 	/* set rm_replay_slots for offline slot(s) */
146 	for (i = 0; i < replay_map->rm_slots; i++) {
147 		if (ocfs2_slot_to_node_num_locked(osb, i, &node_num) == -ENOENT)
148 			replay_map->rm_replay_slots[i] = 1;
149 	}
150 
151 	osb->replay_map = replay_map;
152 	spin_unlock(&osb->osb_lock);
153 	return 0;
154 }
155 
156 static void ocfs2_queue_replay_slots(struct ocfs2_super *osb,
157 		enum ocfs2_orphan_reco_type orphan_reco_type)
158 {
159 	struct ocfs2_replay_map *replay_map = osb->replay_map;
160 	int i;
161 
162 	if (!replay_map)
163 		return;
164 
165 	if (replay_map->rm_state != REPLAY_NEEDED)
166 		return;
167 
168 	for (i = 0; i < replay_map->rm_slots; i++)
169 		if (replay_map->rm_replay_slots[i])
170 			ocfs2_queue_recovery_completion(osb->journal, i, NULL,
171 							NULL, NULL,
172 							orphan_reco_type);
173 	replay_map->rm_state = REPLAY_DONE;
174 }
175 
176 static void ocfs2_free_replay_slots(struct ocfs2_super *osb)
177 {
178 	struct ocfs2_replay_map *replay_map = osb->replay_map;
179 
180 	if (!osb->replay_map)
181 		return;
182 
183 	kfree(replay_map);
184 	osb->replay_map = NULL;
185 }
186 
187 int ocfs2_recovery_init(struct ocfs2_super *osb)
188 {
189 	struct ocfs2_recovery_map *rm;
190 
191 	mutex_init(&osb->recovery_lock);
192 	osb->disable_recovery = 0;
193 	osb->recovery_thread_task = NULL;
194 	init_waitqueue_head(&osb->recovery_event);
195 
196 	rm = kzalloc(sizeof(struct ocfs2_recovery_map) +
197 		     osb->max_slots * sizeof(unsigned int),
198 		     GFP_KERNEL);
199 	if (!rm) {
200 		mlog_errno(-ENOMEM);
201 		return -ENOMEM;
202 	}
203 
204 	rm->rm_entries = (unsigned int *)((char *)rm +
205 					  sizeof(struct ocfs2_recovery_map));
206 	osb->recovery_map = rm;
207 
208 	return 0;
209 }
210 
211 /* we can't grab the goofy sem lock from inside wait_event, so we use
212  * memory barriers to make sure that we'll see the null task before
213  * being woken up */
214 static int ocfs2_recovery_thread_running(struct ocfs2_super *osb)
215 {
216 	mb();
217 	return osb->recovery_thread_task != NULL;
218 }
219 
220 void ocfs2_recovery_exit(struct ocfs2_super *osb)
221 {
222 	struct ocfs2_recovery_map *rm;
223 
224 	/* disable any new recovery threads and wait for any currently
225 	 * running ones to exit. Do this before setting the vol_state. */
226 	mutex_lock(&osb->recovery_lock);
227 	osb->disable_recovery = 1;
228 	mutex_unlock(&osb->recovery_lock);
229 	wait_event(osb->recovery_event, !ocfs2_recovery_thread_running(osb));
230 
231 	/* At this point, we know that no more recovery threads can be
232 	 * launched, so wait for any recovery completion work to
233 	 * complete. */
234 	flush_workqueue(osb->ocfs2_wq);
235 
236 	/*
237 	 * Now that recovery is shut down, and the osb is about to be
238 	 * freed,  the osb_lock is not taken here.
239 	 */
240 	rm = osb->recovery_map;
241 	/* XXX: Should we bug if there are dirty entries? */
242 
243 	kfree(rm);
244 }
245 
246 static int __ocfs2_recovery_map_test(struct ocfs2_super *osb,
247 				     unsigned int node_num)
248 {
249 	int i;
250 	struct ocfs2_recovery_map *rm = osb->recovery_map;
251 
252 	assert_spin_locked(&osb->osb_lock);
253 
254 	for (i = 0; i < rm->rm_used; i++) {
255 		if (rm->rm_entries[i] == node_num)
256 			return 1;
257 	}
258 
259 	return 0;
260 }
261 
262 /* Behaves like test-and-set.  Returns the previous value */
263 static int ocfs2_recovery_map_set(struct ocfs2_super *osb,
264 				  unsigned int node_num)
265 {
266 	struct ocfs2_recovery_map *rm = osb->recovery_map;
267 
268 	spin_lock(&osb->osb_lock);
269 	if (__ocfs2_recovery_map_test(osb, node_num)) {
270 		spin_unlock(&osb->osb_lock);
271 		return 1;
272 	}
273 
274 	/* XXX: Can this be exploited? Not from o2dlm... */
275 	BUG_ON(rm->rm_used >= osb->max_slots);
276 
277 	rm->rm_entries[rm->rm_used] = node_num;
278 	rm->rm_used++;
279 	spin_unlock(&osb->osb_lock);
280 
281 	return 0;
282 }
283 
284 static void ocfs2_recovery_map_clear(struct ocfs2_super *osb,
285 				     unsigned int node_num)
286 {
287 	int i;
288 	struct ocfs2_recovery_map *rm = osb->recovery_map;
289 
290 	spin_lock(&osb->osb_lock);
291 
292 	for (i = 0; i < rm->rm_used; i++) {
293 		if (rm->rm_entries[i] == node_num)
294 			break;
295 	}
296 
297 	if (i < rm->rm_used) {
298 		/* XXX: be careful with the pointer math */
299 		memmove(&(rm->rm_entries[i]), &(rm->rm_entries[i + 1]),
300 			(rm->rm_used - i - 1) * sizeof(unsigned int));
301 		rm->rm_used--;
302 	}
303 
304 	spin_unlock(&osb->osb_lock);
305 }
306 
307 static int ocfs2_commit_cache(struct ocfs2_super *osb)
308 {
309 	int status = 0;
310 	unsigned int flushed;
311 	struct ocfs2_journal *journal = NULL;
312 
313 	journal = osb->journal;
314 
315 	/* Flush all pending commits and checkpoint the journal. */
316 	down_write(&journal->j_trans_barrier);
317 
318 	flushed = atomic_read(&journal->j_num_trans);
319 	trace_ocfs2_commit_cache_begin(flushed);
320 	if (flushed == 0) {
321 		up_write(&journal->j_trans_barrier);
322 		goto finally;
323 	}
324 
325 	jbd2_journal_lock_updates(journal->j_journal);
326 	status = jbd2_journal_flush(journal->j_journal);
327 	jbd2_journal_unlock_updates(journal->j_journal);
328 	if (status < 0) {
329 		up_write(&journal->j_trans_barrier);
330 		mlog_errno(status);
331 		goto finally;
332 	}
333 
334 	ocfs2_inc_trans_id(journal);
335 
336 	flushed = atomic_read(&journal->j_num_trans);
337 	atomic_set(&journal->j_num_trans, 0);
338 	up_write(&journal->j_trans_barrier);
339 
340 	trace_ocfs2_commit_cache_end(journal->j_trans_id, flushed);
341 
342 	ocfs2_wake_downconvert_thread(osb);
343 	wake_up(&journal->j_checkpointed);
344 finally:
345 	return status;
346 }
347 
348 handle_t *ocfs2_start_trans(struct ocfs2_super *osb, int max_buffs)
349 {
350 	journal_t *journal = osb->journal->j_journal;
351 	handle_t *handle;
352 
353 	BUG_ON(!osb || !osb->journal->j_journal);
354 
355 	if (ocfs2_is_hard_readonly(osb))
356 		return ERR_PTR(-EROFS);
357 
358 	BUG_ON(osb->journal->j_state == OCFS2_JOURNAL_FREE);
359 	BUG_ON(max_buffs <= 0);
360 
361 	/* Nested transaction? Just return the handle... */
362 	if (journal_current_handle())
363 		return jbd2_journal_start(journal, max_buffs);
364 
365 	sb_start_intwrite(osb->sb);
366 
367 	down_read(&osb->journal->j_trans_barrier);
368 
369 	handle = jbd2_journal_start(journal, max_buffs);
370 	if (IS_ERR(handle)) {
371 		up_read(&osb->journal->j_trans_barrier);
372 		sb_end_intwrite(osb->sb);
373 
374 		mlog_errno(PTR_ERR(handle));
375 
376 		if (is_journal_aborted(journal)) {
377 			ocfs2_abort(osb->sb, "Detected aborted journal\n");
378 			handle = ERR_PTR(-EROFS);
379 		}
380 	} else {
381 		if (!ocfs2_mount_local(osb))
382 			atomic_inc(&(osb->journal->j_num_trans));
383 	}
384 
385 	return handle;
386 }
387 
388 int ocfs2_commit_trans(struct ocfs2_super *osb,
389 		       handle_t *handle)
390 {
391 	int ret, nested;
392 	struct ocfs2_journal *journal = osb->journal;
393 
394 	BUG_ON(!handle);
395 
396 	nested = handle->h_ref > 1;
397 	ret = jbd2_journal_stop(handle);
398 	if (ret < 0)
399 		mlog_errno(ret);
400 
401 	if (!nested) {
402 		up_read(&journal->j_trans_barrier);
403 		sb_end_intwrite(osb->sb);
404 	}
405 
406 	return ret;
407 }
408 
409 /*
410  * 'nblocks' is what you want to add to the current transaction.
411  *
412  * This might call jbd2_journal_restart() which will commit dirty buffers
413  * and then restart the transaction. Before calling
414  * ocfs2_extend_trans(), any changed blocks should have been
415  * dirtied. After calling it, all blocks which need to be changed must
416  * go through another set of journal_access/journal_dirty calls.
417  *
418  * WARNING: This will not release any semaphores or disk locks taken
419  * during the transaction, so make sure they were taken *before*
420  * start_trans or we'll have ordering deadlocks.
421  *
422  * WARNING2: Note that we do *not* drop j_trans_barrier here. This is
423  * good because transaction ids haven't yet been recorded on the
424  * cluster locks associated with this handle.
425  */
426 int ocfs2_extend_trans(handle_t *handle, int nblocks)
427 {
428 	int status, old_nblocks;
429 
430 	BUG_ON(!handle);
431 	BUG_ON(nblocks < 0);
432 
433 	if (!nblocks)
434 		return 0;
435 
436 	old_nblocks = handle->h_buffer_credits;
437 
438 	trace_ocfs2_extend_trans(old_nblocks, nblocks);
439 
440 #ifdef CONFIG_OCFS2_DEBUG_FS
441 	status = 1;
442 #else
443 	status = jbd2_journal_extend(handle, nblocks);
444 	if (status < 0) {
445 		mlog_errno(status);
446 		goto bail;
447 	}
448 #endif
449 
450 	if (status > 0) {
451 		trace_ocfs2_extend_trans_restart(old_nblocks + nblocks);
452 		status = jbd2_journal_restart(handle,
453 					      old_nblocks + nblocks);
454 		if (status < 0) {
455 			mlog_errno(status);
456 			goto bail;
457 		}
458 	}
459 
460 	status = 0;
461 bail:
462 	return status;
463 }
464 
465 /*
466  * If we have fewer than thresh credits, extend by OCFS2_MAX_TRANS_DATA.
467  * If that fails, restart the transaction & regain write access for the
468  * buffer head which is used for metadata modifications.
469  * Taken from Ext4: extend_or_restart_transaction()
470  */
471 int ocfs2_allocate_extend_trans(handle_t *handle, int thresh)
472 {
473 	int status, old_nblks;
474 
475 	BUG_ON(!handle);
476 
477 	old_nblks = handle->h_buffer_credits;
478 	trace_ocfs2_allocate_extend_trans(old_nblks, thresh);
479 
480 	if (old_nblks < thresh)
481 		return 0;
482 
483 	status = jbd2_journal_extend(handle, OCFS2_MAX_TRANS_DATA);
484 	if (status < 0) {
485 		mlog_errno(status);
486 		goto bail;
487 	}
488 
489 	if (status > 0) {
490 		status = jbd2_journal_restart(handle, OCFS2_MAX_TRANS_DATA);
491 		if (status < 0)
492 			mlog_errno(status);
493 	}
494 
495 bail:
496 	return status;
497 }
498 
499 
500 struct ocfs2_triggers {
501 	struct jbd2_buffer_trigger_type	ot_triggers;
502 	int				ot_offset;
503 };
504 
505 static inline struct ocfs2_triggers *to_ocfs2_trigger(struct jbd2_buffer_trigger_type *triggers)
506 {
507 	return container_of(triggers, struct ocfs2_triggers, ot_triggers);
508 }
509 
510 static void ocfs2_frozen_trigger(struct jbd2_buffer_trigger_type *triggers,
511 				 struct buffer_head *bh,
512 				 void *data, size_t size)
513 {
514 	struct ocfs2_triggers *ot = to_ocfs2_trigger(triggers);
515 
516 	/*
517 	 * We aren't guaranteed to have the superblock here, so we
518 	 * must unconditionally compute the ecc data.
519 	 * __ocfs2_journal_access() will only set the triggers if
520 	 * metaecc is enabled.
521 	 */
522 	ocfs2_block_check_compute(data, size, data + ot->ot_offset);
523 }
524 
525 /*
526  * Quota blocks have their own trigger because the struct ocfs2_block_check
527  * offset depends on the blocksize.
528  */
529 static void ocfs2_dq_frozen_trigger(struct jbd2_buffer_trigger_type *triggers,
530 				 struct buffer_head *bh,
531 				 void *data, size_t size)
532 {
533 	struct ocfs2_disk_dqtrailer *dqt =
534 		ocfs2_block_dqtrailer(size, data);
535 
536 	/*
537 	 * We aren't guaranteed to have the superblock here, so we
538 	 * must unconditionally compute the ecc data.
539 	 * __ocfs2_journal_access() will only set the triggers if
540 	 * metaecc is enabled.
541 	 */
542 	ocfs2_block_check_compute(data, size, &dqt->dq_check);
543 }
544 
545 /*
546  * Directory blocks also have their own trigger because the
547  * struct ocfs2_block_check offset depends on the blocksize.
548  */
549 static void ocfs2_db_frozen_trigger(struct jbd2_buffer_trigger_type *triggers,
550 				 struct buffer_head *bh,
551 				 void *data, size_t size)
552 {
553 	struct ocfs2_dir_block_trailer *trailer =
554 		ocfs2_dir_trailer_from_size(size, data);
555 
556 	/*
557 	 * We aren't guaranteed to have the superblock here, so we
558 	 * must unconditionally compute the ecc data.
559 	 * __ocfs2_journal_access() will only set the triggers if
560 	 * metaecc is enabled.
561 	 */
562 	ocfs2_block_check_compute(data, size, &trailer->db_check);
563 }
564 
565 static void ocfs2_abort_trigger(struct jbd2_buffer_trigger_type *triggers,
566 				struct buffer_head *bh)
567 {
568 	mlog(ML_ERROR,
569 	     "ocfs2_abort_trigger called by JBD2.  bh = 0x%lx, "
570 	     "bh->b_blocknr = %llu\n",
571 	     (unsigned long)bh,
572 	     (unsigned long long)bh->b_blocknr);
573 
574 	ocfs2_error(bh->b_bdev->bd_super,
575 		    "JBD2 has aborted our journal, ocfs2 cannot continue\n");
576 }
577 
578 static struct ocfs2_triggers di_triggers = {
579 	.ot_triggers = {
580 		.t_frozen = ocfs2_frozen_trigger,
581 		.t_abort = ocfs2_abort_trigger,
582 	},
583 	.ot_offset	= offsetof(struct ocfs2_dinode, i_check),
584 };
585 
586 static struct ocfs2_triggers eb_triggers = {
587 	.ot_triggers = {
588 		.t_frozen = ocfs2_frozen_trigger,
589 		.t_abort = ocfs2_abort_trigger,
590 	},
591 	.ot_offset	= offsetof(struct ocfs2_extent_block, h_check),
592 };
593 
594 static struct ocfs2_triggers rb_triggers = {
595 	.ot_triggers = {
596 		.t_frozen = ocfs2_frozen_trigger,
597 		.t_abort = ocfs2_abort_trigger,
598 	},
599 	.ot_offset	= offsetof(struct ocfs2_refcount_block, rf_check),
600 };
601 
602 static struct ocfs2_triggers gd_triggers = {
603 	.ot_triggers = {
604 		.t_frozen = ocfs2_frozen_trigger,
605 		.t_abort = ocfs2_abort_trigger,
606 	},
607 	.ot_offset	= offsetof(struct ocfs2_group_desc, bg_check),
608 };
609 
610 static struct ocfs2_triggers db_triggers = {
611 	.ot_triggers = {
612 		.t_frozen = ocfs2_db_frozen_trigger,
613 		.t_abort = ocfs2_abort_trigger,
614 	},
615 };
616 
617 static struct ocfs2_triggers xb_triggers = {
618 	.ot_triggers = {
619 		.t_frozen = ocfs2_frozen_trigger,
620 		.t_abort = ocfs2_abort_trigger,
621 	},
622 	.ot_offset	= offsetof(struct ocfs2_xattr_block, xb_check),
623 };
624 
625 static struct ocfs2_triggers dq_triggers = {
626 	.ot_triggers = {
627 		.t_frozen = ocfs2_dq_frozen_trigger,
628 		.t_abort = ocfs2_abort_trigger,
629 	},
630 };
631 
632 static struct ocfs2_triggers dr_triggers = {
633 	.ot_triggers = {
634 		.t_frozen = ocfs2_frozen_trigger,
635 		.t_abort = ocfs2_abort_trigger,
636 	},
637 	.ot_offset	= offsetof(struct ocfs2_dx_root_block, dr_check),
638 };
639 
640 static struct ocfs2_triggers dl_triggers = {
641 	.ot_triggers = {
642 		.t_frozen = ocfs2_frozen_trigger,
643 		.t_abort = ocfs2_abort_trigger,
644 	},
645 	.ot_offset	= offsetof(struct ocfs2_dx_leaf, dl_check),
646 };
647 
648 static int __ocfs2_journal_access(handle_t *handle,
649 				  struct ocfs2_caching_info *ci,
650 				  struct buffer_head *bh,
651 				  struct ocfs2_triggers *triggers,
652 				  int type)
653 {
654 	int status;
655 	struct ocfs2_super *osb =
656 		OCFS2_SB(ocfs2_metadata_cache_get_super(ci));
657 
658 	BUG_ON(!ci || !ci->ci_ops);
659 	BUG_ON(!handle);
660 	BUG_ON(!bh);
661 
662 	trace_ocfs2_journal_access(
663 		(unsigned long long)ocfs2_metadata_cache_owner(ci),
664 		(unsigned long long)bh->b_blocknr, type, bh->b_size);
665 
666 	/* we can safely remove this assertion after testing. */
667 	if (!buffer_uptodate(bh)) {
668 		mlog(ML_ERROR, "giving me a buffer that's not uptodate!\n");
669 		mlog(ML_ERROR, "b_blocknr=%llu\n",
670 		     (unsigned long long)bh->b_blocknr);
671 
672 		lock_buffer(bh);
673 		/*
674 		 * A previous attempt to write this buffer head failed.
675 		 * Nothing we can do but to retry the write and hope for
676 		 * the best.
677 		 */
678 		if (buffer_write_io_error(bh) && !buffer_uptodate(bh)) {
679 			clear_buffer_write_io_error(bh);
680 			set_buffer_uptodate(bh);
681 		}
682 
683 		if (!buffer_uptodate(bh)) {
684 			unlock_buffer(bh);
685 			return -EIO;
686 		}
687 		unlock_buffer(bh);
688 	}
689 
690 	/* Set the current transaction information on the ci so
691 	 * that the locking code knows whether it can drop it's locks
692 	 * on this ci or not. We're protected from the commit
693 	 * thread updating the current transaction id until
694 	 * ocfs2_commit_trans() because ocfs2_start_trans() took
695 	 * j_trans_barrier for us. */
696 	ocfs2_set_ci_lock_trans(osb->journal, ci);
697 
698 	ocfs2_metadata_cache_io_lock(ci);
699 	switch (type) {
700 	case OCFS2_JOURNAL_ACCESS_CREATE:
701 	case OCFS2_JOURNAL_ACCESS_WRITE:
702 		status = jbd2_journal_get_write_access(handle, bh);
703 		break;
704 
705 	case OCFS2_JOURNAL_ACCESS_UNDO:
706 		status = jbd2_journal_get_undo_access(handle, bh);
707 		break;
708 
709 	default:
710 		status = -EINVAL;
711 		mlog(ML_ERROR, "Unknown access type!\n");
712 	}
713 	if (!status && ocfs2_meta_ecc(osb) && triggers)
714 		jbd2_journal_set_triggers(bh, &triggers->ot_triggers);
715 	ocfs2_metadata_cache_io_unlock(ci);
716 
717 	if (status < 0)
718 		mlog(ML_ERROR, "Error %d getting %d access to buffer!\n",
719 		     status, type);
720 
721 	return status;
722 }
723 
724 int ocfs2_journal_access_di(handle_t *handle, struct ocfs2_caching_info *ci,
725 			    struct buffer_head *bh, int type)
726 {
727 	return __ocfs2_journal_access(handle, ci, bh, &di_triggers, type);
728 }
729 
730 int ocfs2_journal_access_eb(handle_t *handle, struct ocfs2_caching_info *ci,
731 			    struct buffer_head *bh, int type)
732 {
733 	return __ocfs2_journal_access(handle, ci, bh, &eb_triggers, type);
734 }
735 
736 int ocfs2_journal_access_rb(handle_t *handle, struct ocfs2_caching_info *ci,
737 			    struct buffer_head *bh, int type)
738 {
739 	return __ocfs2_journal_access(handle, ci, bh, &rb_triggers,
740 				      type);
741 }
742 
743 int ocfs2_journal_access_gd(handle_t *handle, struct ocfs2_caching_info *ci,
744 			    struct buffer_head *bh, int type)
745 {
746 	return __ocfs2_journal_access(handle, ci, bh, &gd_triggers, type);
747 }
748 
749 int ocfs2_journal_access_db(handle_t *handle, struct ocfs2_caching_info *ci,
750 			    struct buffer_head *bh, int type)
751 {
752 	return __ocfs2_journal_access(handle, ci, bh, &db_triggers, type);
753 }
754 
755 int ocfs2_journal_access_xb(handle_t *handle, struct ocfs2_caching_info *ci,
756 			    struct buffer_head *bh, int type)
757 {
758 	return __ocfs2_journal_access(handle, ci, bh, &xb_triggers, type);
759 }
760 
761 int ocfs2_journal_access_dq(handle_t *handle, struct ocfs2_caching_info *ci,
762 			    struct buffer_head *bh, int type)
763 {
764 	return __ocfs2_journal_access(handle, ci, bh, &dq_triggers, type);
765 }
766 
767 int ocfs2_journal_access_dr(handle_t *handle, struct ocfs2_caching_info *ci,
768 			    struct buffer_head *bh, int type)
769 {
770 	return __ocfs2_journal_access(handle, ci, bh, &dr_triggers, type);
771 }
772 
773 int ocfs2_journal_access_dl(handle_t *handle, struct ocfs2_caching_info *ci,
774 			    struct buffer_head *bh, int type)
775 {
776 	return __ocfs2_journal_access(handle, ci, bh, &dl_triggers, type);
777 }
778 
779 int ocfs2_journal_access(handle_t *handle, struct ocfs2_caching_info *ci,
780 			 struct buffer_head *bh, int type)
781 {
782 	return __ocfs2_journal_access(handle, ci, bh, NULL, type);
783 }
784 
785 void ocfs2_journal_dirty(handle_t *handle, struct buffer_head *bh)
786 {
787 	int status;
788 
789 	trace_ocfs2_journal_dirty((unsigned long long)bh->b_blocknr);
790 
791 	status = jbd2_journal_dirty_metadata(handle, bh);
792 	if (status) {
793 		mlog_errno(status);
794 		if (!is_handle_aborted(handle)) {
795 			journal_t *journal = handle->h_transaction->t_journal;
796 			struct super_block *sb = bh->b_bdev->bd_super;
797 
798 			mlog(ML_ERROR, "jbd2_journal_dirty_metadata failed. "
799 					"Aborting transaction and journal.\n");
800 			handle->h_err = status;
801 			jbd2_journal_abort_handle(handle);
802 			jbd2_journal_abort(journal, status);
803 			ocfs2_abort(sb, "Journal already aborted.\n");
804 		}
805 	}
806 }
807 
808 #define OCFS2_DEFAULT_COMMIT_INTERVAL	(HZ * JBD2_DEFAULT_MAX_COMMIT_AGE)
809 
810 void ocfs2_set_journal_params(struct ocfs2_super *osb)
811 {
812 	journal_t *journal = osb->journal->j_journal;
813 	unsigned long commit_interval = OCFS2_DEFAULT_COMMIT_INTERVAL;
814 
815 	if (osb->osb_commit_interval)
816 		commit_interval = osb->osb_commit_interval;
817 
818 	write_lock(&journal->j_state_lock);
819 	journal->j_commit_interval = commit_interval;
820 	if (osb->s_mount_opt & OCFS2_MOUNT_BARRIER)
821 		journal->j_flags |= JBD2_BARRIER;
822 	else
823 		journal->j_flags &= ~JBD2_BARRIER;
824 	write_unlock(&journal->j_state_lock);
825 }
826 
827 int ocfs2_journal_init(struct ocfs2_journal *journal, int *dirty)
828 {
829 	int status = -1;
830 	struct inode *inode = NULL; /* the journal inode */
831 	journal_t *j_journal = NULL;
832 	struct ocfs2_dinode *di = NULL;
833 	struct buffer_head *bh = NULL;
834 	struct ocfs2_super *osb;
835 	int inode_lock = 0;
836 
837 	BUG_ON(!journal);
838 
839 	osb = journal->j_osb;
840 
841 	/* already have the inode for our journal */
842 	inode = ocfs2_get_system_file_inode(osb, JOURNAL_SYSTEM_INODE,
843 					    osb->slot_num);
844 	if (inode == NULL) {
845 		status = -EACCES;
846 		mlog_errno(status);
847 		goto done;
848 	}
849 	if (is_bad_inode(inode)) {
850 		mlog(ML_ERROR, "access error (bad inode)\n");
851 		iput(inode);
852 		inode = NULL;
853 		status = -EACCES;
854 		goto done;
855 	}
856 
857 	SET_INODE_JOURNAL(inode);
858 	OCFS2_I(inode)->ip_open_count++;
859 
860 	/* Skip recovery waits here - journal inode metadata never
861 	 * changes in a live cluster so it can be considered an
862 	 * exception to the rule. */
863 	status = ocfs2_inode_lock_full(inode, &bh, 1, OCFS2_META_LOCK_RECOVERY);
864 	if (status < 0) {
865 		if (status != -ERESTARTSYS)
866 			mlog(ML_ERROR, "Could not get lock on journal!\n");
867 		goto done;
868 	}
869 
870 	inode_lock = 1;
871 	di = (struct ocfs2_dinode *)bh->b_data;
872 
873 	if (i_size_read(inode) <  OCFS2_MIN_JOURNAL_SIZE) {
874 		mlog(ML_ERROR, "Journal file size (%lld) is too small!\n",
875 		     i_size_read(inode));
876 		status = -EINVAL;
877 		goto done;
878 	}
879 
880 	trace_ocfs2_journal_init(i_size_read(inode),
881 				 (unsigned long long)inode->i_blocks,
882 				 OCFS2_I(inode)->ip_clusters);
883 
884 	/* call the kernels journal init function now */
885 	j_journal = jbd2_journal_init_inode(inode);
886 	if (j_journal == NULL) {
887 		mlog(ML_ERROR, "Linux journal layer error\n");
888 		status = -EINVAL;
889 		goto done;
890 	}
891 
892 	trace_ocfs2_journal_init_maxlen(j_journal->j_maxlen);
893 
894 	*dirty = (le32_to_cpu(di->id1.journal1.ij_flags) &
895 		  OCFS2_JOURNAL_DIRTY_FL);
896 
897 	journal->j_journal = j_journal;
898 	journal->j_inode = inode;
899 	journal->j_bh = bh;
900 
901 	ocfs2_set_journal_params(osb);
902 
903 	journal->j_state = OCFS2_JOURNAL_LOADED;
904 
905 	status = 0;
906 done:
907 	if (status < 0) {
908 		if (inode_lock)
909 			ocfs2_inode_unlock(inode, 1);
910 		brelse(bh);
911 		if (inode) {
912 			OCFS2_I(inode)->ip_open_count--;
913 			iput(inode);
914 		}
915 	}
916 
917 	return status;
918 }
919 
920 static void ocfs2_bump_recovery_generation(struct ocfs2_dinode *di)
921 {
922 	le32_add_cpu(&(di->id1.journal1.ij_recovery_generation), 1);
923 }
924 
925 static u32 ocfs2_get_recovery_generation(struct ocfs2_dinode *di)
926 {
927 	return le32_to_cpu(di->id1.journal1.ij_recovery_generation);
928 }
929 
930 static int ocfs2_journal_toggle_dirty(struct ocfs2_super *osb,
931 				      int dirty, int replayed)
932 {
933 	int status;
934 	unsigned int flags;
935 	struct ocfs2_journal *journal = osb->journal;
936 	struct buffer_head *bh = journal->j_bh;
937 	struct ocfs2_dinode *fe;
938 
939 	fe = (struct ocfs2_dinode *)bh->b_data;
940 
941 	/* The journal bh on the osb always comes from ocfs2_journal_init()
942 	 * and was validated there inside ocfs2_inode_lock_full().  It's a
943 	 * code bug if we mess it up. */
944 	BUG_ON(!OCFS2_IS_VALID_DINODE(fe));
945 
946 	flags = le32_to_cpu(fe->id1.journal1.ij_flags);
947 	if (dirty)
948 		flags |= OCFS2_JOURNAL_DIRTY_FL;
949 	else
950 		flags &= ~OCFS2_JOURNAL_DIRTY_FL;
951 	fe->id1.journal1.ij_flags = cpu_to_le32(flags);
952 
953 	if (replayed)
954 		ocfs2_bump_recovery_generation(fe);
955 
956 	ocfs2_compute_meta_ecc(osb->sb, bh->b_data, &fe->i_check);
957 	status = ocfs2_write_block(osb, bh, INODE_CACHE(journal->j_inode));
958 	if (status < 0)
959 		mlog_errno(status);
960 
961 	return status;
962 }
963 
964 /*
965  * If the journal has been kmalloc'd it needs to be freed after this
966  * call.
967  */
968 void ocfs2_journal_shutdown(struct ocfs2_super *osb)
969 {
970 	struct ocfs2_journal *journal = NULL;
971 	int status = 0;
972 	struct inode *inode = NULL;
973 	int num_running_trans = 0;
974 
975 	BUG_ON(!osb);
976 
977 	journal = osb->journal;
978 	if (!journal)
979 		goto done;
980 
981 	inode = journal->j_inode;
982 
983 	if (journal->j_state != OCFS2_JOURNAL_LOADED)
984 		goto done;
985 
986 	/* need to inc inode use count - jbd2_journal_destroy will iput. */
987 	if (!igrab(inode))
988 		BUG();
989 
990 	num_running_trans = atomic_read(&(osb->journal->j_num_trans));
991 	trace_ocfs2_journal_shutdown(num_running_trans);
992 
993 	/* Do a commit_cache here. It will flush our journal, *and*
994 	 * release any locks that are still held.
995 	 * set the SHUTDOWN flag and release the trans lock.
996 	 * the commit thread will take the trans lock for us below. */
997 	journal->j_state = OCFS2_JOURNAL_IN_SHUTDOWN;
998 
999 	/* The OCFS2_JOURNAL_IN_SHUTDOWN will signal to commit_cache to not
1000 	 * drop the trans_lock (which we want to hold until we
1001 	 * completely destroy the journal. */
1002 	if (osb->commit_task) {
1003 		/* Wait for the commit thread */
1004 		trace_ocfs2_journal_shutdown_wait(osb->commit_task);
1005 		kthread_stop(osb->commit_task);
1006 		osb->commit_task = NULL;
1007 	}
1008 
1009 	BUG_ON(atomic_read(&(osb->journal->j_num_trans)) != 0);
1010 
1011 	if (ocfs2_mount_local(osb)) {
1012 		jbd2_journal_lock_updates(journal->j_journal);
1013 		status = jbd2_journal_flush(journal->j_journal);
1014 		jbd2_journal_unlock_updates(journal->j_journal);
1015 		if (status < 0)
1016 			mlog_errno(status);
1017 	}
1018 
1019 	if (status == 0) {
1020 		/*
1021 		 * Do not toggle if flush was unsuccessful otherwise
1022 		 * will leave dirty metadata in a "clean" journal
1023 		 */
1024 		status = ocfs2_journal_toggle_dirty(osb, 0, 0);
1025 		if (status < 0)
1026 			mlog_errno(status);
1027 	}
1028 
1029 	/* Shutdown the kernel journal system */
1030 	jbd2_journal_destroy(journal->j_journal);
1031 	journal->j_journal = NULL;
1032 
1033 	OCFS2_I(inode)->ip_open_count--;
1034 
1035 	/* unlock our journal */
1036 	ocfs2_inode_unlock(inode, 1);
1037 
1038 	brelse(journal->j_bh);
1039 	journal->j_bh = NULL;
1040 
1041 	journal->j_state = OCFS2_JOURNAL_FREE;
1042 
1043 //	up_write(&journal->j_trans_barrier);
1044 done:
1045 	iput(inode);
1046 }
1047 
1048 static void ocfs2_clear_journal_error(struct super_block *sb,
1049 				      journal_t *journal,
1050 				      int slot)
1051 {
1052 	int olderr;
1053 
1054 	olderr = jbd2_journal_errno(journal);
1055 	if (olderr) {
1056 		mlog(ML_ERROR, "File system error %d recorded in "
1057 		     "journal %u.\n", olderr, slot);
1058 		mlog(ML_ERROR, "File system on device %s needs checking.\n",
1059 		     sb->s_id);
1060 
1061 		jbd2_journal_ack_err(journal);
1062 		jbd2_journal_clear_err(journal);
1063 	}
1064 }
1065 
1066 int ocfs2_journal_load(struct ocfs2_journal *journal, int local, int replayed)
1067 {
1068 	int status = 0;
1069 	struct ocfs2_super *osb;
1070 
1071 	BUG_ON(!journal);
1072 
1073 	osb = journal->j_osb;
1074 
1075 	status = jbd2_journal_load(journal->j_journal);
1076 	if (status < 0) {
1077 		mlog(ML_ERROR, "Failed to load journal!\n");
1078 		goto done;
1079 	}
1080 
1081 	ocfs2_clear_journal_error(osb->sb, journal->j_journal, osb->slot_num);
1082 
1083 	status = ocfs2_journal_toggle_dirty(osb, 1, replayed);
1084 	if (status < 0) {
1085 		mlog_errno(status);
1086 		goto done;
1087 	}
1088 
1089 	/* Launch the commit thread */
1090 	if (!local) {
1091 		osb->commit_task = kthread_run(ocfs2_commit_thread, osb,
1092 				"ocfs2cmt-%s", osb->uuid_str);
1093 		if (IS_ERR(osb->commit_task)) {
1094 			status = PTR_ERR(osb->commit_task);
1095 			osb->commit_task = NULL;
1096 			mlog(ML_ERROR, "unable to launch ocfs2commit thread, "
1097 			     "error=%d", status);
1098 			goto done;
1099 		}
1100 	} else
1101 		osb->commit_task = NULL;
1102 
1103 done:
1104 	return status;
1105 }
1106 
1107 
1108 /* 'full' flag tells us whether we clear out all blocks or if we just
1109  * mark the journal clean */
1110 int ocfs2_journal_wipe(struct ocfs2_journal *journal, int full)
1111 {
1112 	int status;
1113 
1114 	BUG_ON(!journal);
1115 
1116 	status = jbd2_journal_wipe(journal->j_journal, full);
1117 	if (status < 0) {
1118 		mlog_errno(status);
1119 		goto bail;
1120 	}
1121 
1122 	status = ocfs2_journal_toggle_dirty(journal->j_osb, 0, 0);
1123 	if (status < 0)
1124 		mlog_errno(status);
1125 
1126 bail:
1127 	return status;
1128 }
1129 
1130 static int ocfs2_recovery_completed(struct ocfs2_super *osb)
1131 {
1132 	int empty;
1133 	struct ocfs2_recovery_map *rm = osb->recovery_map;
1134 
1135 	spin_lock(&osb->osb_lock);
1136 	empty = (rm->rm_used == 0);
1137 	spin_unlock(&osb->osb_lock);
1138 
1139 	return empty;
1140 }
1141 
1142 void ocfs2_wait_for_recovery(struct ocfs2_super *osb)
1143 {
1144 	wait_event(osb->recovery_event, ocfs2_recovery_completed(osb));
1145 }
1146 
1147 /*
1148  * JBD Might read a cached version of another nodes journal file. We
1149  * don't want this as this file changes often and we get no
1150  * notification on those changes. The only way to be sure that we've
1151  * got the most up to date version of those blocks then is to force
1152  * read them off disk. Just searching through the buffer cache won't
1153  * work as there may be pages backing this file which are still marked
1154  * up to date. We know things can't change on this file underneath us
1155  * as we have the lock by now :)
1156  */
1157 static int ocfs2_force_read_journal(struct inode *inode)
1158 {
1159 	int status = 0;
1160 	int i;
1161 	u64 v_blkno, p_blkno, p_blocks, num_blocks;
1162 #define CONCURRENT_JOURNAL_FILL 32ULL
1163 	struct buffer_head *bhs[CONCURRENT_JOURNAL_FILL];
1164 
1165 	memset(bhs, 0, sizeof(struct buffer_head *) * CONCURRENT_JOURNAL_FILL);
1166 
1167 	num_blocks = ocfs2_blocks_for_bytes(inode->i_sb, i_size_read(inode));
1168 	v_blkno = 0;
1169 	while (v_blkno < num_blocks) {
1170 		status = ocfs2_extent_map_get_blocks(inode, v_blkno,
1171 						     &p_blkno, &p_blocks, NULL);
1172 		if (status < 0) {
1173 			mlog_errno(status);
1174 			goto bail;
1175 		}
1176 
1177 		if (p_blocks > CONCURRENT_JOURNAL_FILL)
1178 			p_blocks = CONCURRENT_JOURNAL_FILL;
1179 
1180 		/* We are reading journal data which should not
1181 		 * be put in the uptodate cache */
1182 		status = ocfs2_read_blocks_sync(OCFS2_SB(inode->i_sb),
1183 						p_blkno, p_blocks, bhs);
1184 		if (status < 0) {
1185 			mlog_errno(status);
1186 			goto bail;
1187 		}
1188 
1189 		for(i = 0; i < p_blocks; i++) {
1190 			brelse(bhs[i]);
1191 			bhs[i] = NULL;
1192 		}
1193 
1194 		v_blkno += p_blocks;
1195 	}
1196 
1197 bail:
1198 	for(i = 0; i < CONCURRENT_JOURNAL_FILL; i++)
1199 		brelse(bhs[i]);
1200 	return status;
1201 }
1202 
1203 struct ocfs2_la_recovery_item {
1204 	struct list_head	lri_list;
1205 	int			lri_slot;
1206 	struct ocfs2_dinode	*lri_la_dinode;
1207 	struct ocfs2_dinode	*lri_tl_dinode;
1208 	struct ocfs2_quota_recovery *lri_qrec;
1209 	enum ocfs2_orphan_reco_type  lri_orphan_reco_type;
1210 };
1211 
1212 /* Does the second half of the recovery process. By this point, the
1213  * node is marked clean and can actually be considered recovered,
1214  * hence it's no longer in the recovery map, but there's still some
1215  * cleanup we can do which shouldn't happen within the recovery thread
1216  * as locking in that context becomes very difficult if we are to take
1217  * recovering nodes into account.
1218  *
1219  * NOTE: This function can and will sleep on recovery of other nodes
1220  * during cluster locking, just like any other ocfs2 process.
1221  */
1222 void ocfs2_complete_recovery(struct work_struct *work)
1223 {
1224 	int ret = 0;
1225 	struct ocfs2_journal *journal =
1226 		container_of(work, struct ocfs2_journal, j_recovery_work);
1227 	struct ocfs2_super *osb = journal->j_osb;
1228 	struct ocfs2_dinode *la_dinode, *tl_dinode;
1229 	struct ocfs2_la_recovery_item *item, *n;
1230 	struct ocfs2_quota_recovery *qrec;
1231 	enum ocfs2_orphan_reco_type orphan_reco_type;
1232 	LIST_HEAD(tmp_la_list);
1233 
1234 	trace_ocfs2_complete_recovery(
1235 		(unsigned long long)OCFS2_I(journal->j_inode)->ip_blkno);
1236 
1237 	spin_lock(&journal->j_lock);
1238 	list_splice_init(&journal->j_la_cleanups, &tmp_la_list);
1239 	spin_unlock(&journal->j_lock);
1240 
1241 	list_for_each_entry_safe(item, n, &tmp_la_list, lri_list) {
1242 		list_del_init(&item->lri_list);
1243 
1244 		ocfs2_wait_on_quotas(osb);
1245 
1246 		la_dinode = item->lri_la_dinode;
1247 		tl_dinode = item->lri_tl_dinode;
1248 		qrec = item->lri_qrec;
1249 		orphan_reco_type = item->lri_orphan_reco_type;
1250 
1251 		trace_ocfs2_complete_recovery_slot(item->lri_slot,
1252 			la_dinode ? le64_to_cpu(la_dinode->i_blkno) : 0,
1253 			tl_dinode ? le64_to_cpu(tl_dinode->i_blkno) : 0,
1254 			qrec);
1255 
1256 		if (la_dinode) {
1257 			ret = ocfs2_complete_local_alloc_recovery(osb,
1258 								  la_dinode);
1259 			if (ret < 0)
1260 				mlog_errno(ret);
1261 
1262 			kfree(la_dinode);
1263 		}
1264 
1265 		if (tl_dinode) {
1266 			ret = ocfs2_complete_truncate_log_recovery(osb,
1267 								   tl_dinode);
1268 			if (ret < 0)
1269 				mlog_errno(ret);
1270 
1271 			kfree(tl_dinode);
1272 		}
1273 
1274 		ret = ocfs2_recover_orphans(osb, item->lri_slot,
1275 				orphan_reco_type);
1276 		if (ret < 0)
1277 			mlog_errno(ret);
1278 
1279 		if (qrec) {
1280 			ret = ocfs2_finish_quota_recovery(osb, qrec,
1281 							  item->lri_slot);
1282 			if (ret < 0)
1283 				mlog_errno(ret);
1284 			/* Recovery info is already freed now */
1285 		}
1286 
1287 		kfree(item);
1288 	}
1289 
1290 	trace_ocfs2_complete_recovery_end(ret);
1291 }
1292 
1293 /* NOTE: This function always eats your references to la_dinode and
1294  * tl_dinode, either manually on error, or by passing them to
1295  * ocfs2_complete_recovery */
1296 static void ocfs2_queue_recovery_completion(struct ocfs2_journal *journal,
1297 					    int slot_num,
1298 					    struct ocfs2_dinode *la_dinode,
1299 					    struct ocfs2_dinode *tl_dinode,
1300 					    struct ocfs2_quota_recovery *qrec,
1301 					    enum ocfs2_orphan_reco_type orphan_reco_type)
1302 {
1303 	struct ocfs2_la_recovery_item *item;
1304 
1305 	item = kmalloc(sizeof(struct ocfs2_la_recovery_item), GFP_NOFS);
1306 	if (!item) {
1307 		/* Though we wish to avoid it, we are in fact safe in
1308 		 * skipping local alloc cleanup as fsck.ocfs2 is more
1309 		 * than capable of reclaiming unused space. */
1310 		kfree(la_dinode);
1311 		kfree(tl_dinode);
1312 
1313 		if (qrec)
1314 			ocfs2_free_quota_recovery(qrec);
1315 
1316 		mlog_errno(-ENOMEM);
1317 		return;
1318 	}
1319 
1320 	INIT_LIST_HEAD(&item->lri_list);
1321 	item->lri_la_dinode = la_dinode;
1322 	item->lri_slot = slot_num;
1323 	item->lri_tl_dinode = tl_dinode;
1324 	item->lri_qrec = qrec;
1325 	item->lri_orphan_reco_type = orphan_reco_type;
1326 
1327 	spin_lock(&journal->j_lock);
1328 	list_add_tail(&item->lri_list, &journal->j_la_cleanups);
1329 	queue_work(journal->j_osb->ocfs2_wq, &journal->j_recovery_work);
1330 	spin_unlock(&journal->j_lock);
1331 }
1332 
1333 /* Called by the mount code to queue recovery the last part of
1334  * recovery for it's own and offline slot(s). */
1335 void ocfs2_complete_mount_recovery(struct ocfs2_super *osb)
1336 {
1337 	struct ocfs2_journal *journal = osb->journal;
1338 
1339 	if (ocfs2_is_hard_readonly(osb))
1340 		return;
1341 
1342 	/* No need to queue up our truncate_log as regular cleanup will catch
1343 	 * that */
1344 	ocfs2_queue_recovery_completion(journal, osb->slot_num,
1345 					osb->local_alloc_copy, NULL, NULL,
1346 					ORPHAN_NEED_TRUNCATE);
1347 	ocfs2_schedule_truncate_log_flush(osb, 0);
1348 
1349 	osb->local_alloc_copy = NULL;
1350 	osb->dirty = 0;
1351 
1352 	/* queue to recover orphan slots for all offline slots */
1353 	ocfs2_replay_map_set_state(osb, REPLAY_NEEDED);
1354 	ocfs2_queue_replay_slots(osb, ORPHAN_NEED_TRUNCATE);
1355 	ocfs2_free_replay_slots(osb);
1356 }
1357 
1358 void ocfs2_complete_quota_recovery(struct ocfs2_super *osb)
1359 {
1360 	if (osb->quota_rec) {
1361 		ocfs2_queue_recovery_completion(osb->journal,
1362 						osb->slot_num,
1363 						NULL,
1364 						NULL,
1365 						osb->quota_rec,
1366 						ORPHAN_NEED_TRUNCATE);
1367 		osb->quota_rec = NULL;
1368 	}
1369 }
1370 
1371 static int __ocfs2_recovery_thread(void *arg)
1372 {
1373 	int status, node_num, slot_num;
1374 	struct ocfs2_super *osb = arg;
1375 	struct ocfs2_recovery_map *rm = osb->recovery_map;
1376 	int *rm_quota = NULL;
1377 	int rm_quota_used = 0, i;
1378 	struct ocfs2_quota_recovery *qrec;
1379 
1380 	status = ocfs2_wait_on_mount(osb);
1381 	if (status < 0) {
1382 		goto bail;
1383 	}
1384 
1385 	rm_quota = kzalloc(osb->max_slots * sizeof(int), GFP_NOFS);
1386 	if (!rm_quota) {
1387 		status = -ENOMEM;
1388 		goto bail;
1389 	}
1390 restart:
1391 	status = ocfs2_super_lock(osb, 1);
1392 	if (status < 0) {
1393 		mlog_errno(status);
1394 		goto bail;
1395 	}
1396 
1397 	status = ocfs2_compute_replay_slots(osb);
1398 	if (status < 0)
1399 		mlog_errno(status);
1400 
1401 	/* queue recovery for our own slot */
1402 	ocfs2_queue_recovery_completion(osb->journal, osb->slot_num, NULL,
1403 					NULL, NULL, ORPHAN_NO_NEED_TRUNCATE);
1404 
1405 	spin_lock(&osb->osb_lock);
1406 	while (rm->rm_used) {
1407 		/* It's always safe to remove entry zero, as we won't
1408 		 * clear it until ocfs2_recover_node() has succeeded. */
1409 		node_num = rm->rm_entries[0];
1410 		spin_unlock(&osb->osb_lock);
1411 		slot_num = ocfs2_node_num_to_slot(osb, node_num);
1412 		trace_ocfs2_recovery_thread_node(node_num, slot_num);
1413 		if (slot_num == -ENOENT) {
1414 			status = 0;
1415 			goto skip_recovery;
1416 		}
1417 
1418 		/* It is a bit subtle with quota recovery. We cannot do it
1419 		 * immediately because we have to obtain cluster locks from
1420 		 * quota files and we also don't want to just skip it because
1421 		 * then quota usage would be out of sync until some node takes
1422 		 * the slot. So we remember which nodes need quota recovery
1423 		 * and when everything else is done, we recover quotas. */
1424 		for (i = 0; i < rm_quota_used && rm_quota[i] != slot_num; i++);
1425 		if (i == rm_quota_used)
1426 			rm_quota[rm_quota_used++] = slot_num;
1427 
1428 		status = ocfs2_recover_node(osb, node_num, slot_num);
1429 skip_recovery:
1430 		if (!status) {
1431 			ocfs2_recovery_map_clear(osb, node_num);
1432 		} else {
1433 			mlog(ML_ERROR,
1434 			     "Error %d recovering node %d on device (%u,%u)!\n",
1435 			     status, node_num,
1436 			     MAJOR(osb->sb->s_dev), MINOR(osb->sb->s_dev));
1437 			mlog(ML_ERROR, "Volume requires unmount.\n");
1438 		}
1439 
1440 		spin_lock(&osb->osb_lock);
1441 	}
1442 	spin_unlock(&osb->osb_lock);
1443 	trace_ocfs2_recovery_thread_end(status);
1444 
1445 	/* Refresh all journal recovery generations from disk */
1446 	status = ocfs2_check_journals_nolocks(osb);
1447 	status = (status == -EROFS) ? 0 : status;
1448 	if (status < 0)
1449 		mlog_errno(status);
1450 
1451 	/* Now it is right time to recover quotas... We have to do this under
1452 	 * superblock lock so that no one can start using the slot (and crash)
1453 	 * before we recover it */
1454 	for (i = 0; i < rm_quota_used; i++) {
1455 		qrec = ocfs2_begin_quota_recovery(osb, rm_quota[i]);
1456 		if (IS_ERR(qrec)) {
1457 			status = PTR_ERR(qrec);
1458 			mlog_errno(status);
1459 			continue;
1460 		}
1461 		ocfs2_queue_recovery_completion(osb->journal, rm_quota[i],
1462 						NULL, NULL, qrec,
1463 						ORPHAN_NEED_TRUNCATE);
1464 	}
1465 
1466 	ocfs2_super_unlock(osb, 1);
1467 
1468 	/* queue recovery for offline slots */
1469 	ocfs2_queue_replay_slots(osb, ORPHAN_NEED_TRUNCATE);
1470 
1471 bail:
1472 	mutex_lock(&osb->recovery_lock);
1473 	if (!status && !ocfs2_recovery_completed(osb)) {
1474 		mutex_unlock(&osb->recovery_lock);
1475 		goto restart;
1476 	}
1477 
1478 	ocfs2_free_replay_slots(osb);
1479 	osb->recovery_thread_task = NULL;
1480 	mb(); /* sync with ocfs2_recovery_thread_running */
1481 	wake_up(&osb->recovery_event);
1482 
1483 	mutex_unlock(&osb->recovery_lock);
1484 
1485 	kfree(rm_quota);
1486 
1487 	/* no one is callint kthread_stop() for us so the kthread() api
1488 	 * requires that we call do_exit().  And it isn't exported, but
1489 	 * complete_and_exit() seems to be a minimal wrapper around it. */
1490 	complete_and_exit(NULL, status);
1491 }
1492 
1493 void ocfs2_recovery_thread(struct ocfs2_super *osb, int node_num)
1494 {
1495 	mutex_lock(&osb->recovery_lock);
1496 
1497 	trace_ocfs2_recovery_thread(node_num, osb->node_num,
1498 		osb->disable_recovery, osb->recovery_thread_task,
1499 		osb->disable_recovery ?
1500 		-1 : ocfs2_recovery_map_set(osb, node_num));
1501 
1502 	if (osb->disable_recovery)
1503 		goto out;
1504 
1505 	if (osb->recovery_thread_task)
1506 		goto out;
1507 
1508 	osb->recovery_thread_task =  kthread_run(__ocfs2_recovery_thread, osb,
1509 			"ocfs2rec-%s", osb->uuid_str);
1510 	if (IS_ERR(osb->recovery_thread_task)) {
1511 		mlog_errno((int)PTR_ERR(osb->recovery_thread_task));
1512 		osb->recovery_thread_task = NULL;
1513 	}
1514 
1515 out:
1516 	mutex_unlock(&osb->recovery_lock);
1517 	wake_up(&osb->recovery_event);
1518 }
1519 
1520 static int ocfs2_read_journal_inode(struct ocfs2_super *osb,
1521 				    int slot_num,
1522 				    struct buffer_head **bh,
1523 				    struct inode **ret_inode)
1524 {
1525 	int status = -EACCES;
1526 	struct inode *inode = NULL;
1527 
1528 	BUG_ON(slot_num >= osb->max_slots);
1529 
1530 	inode = ocfs2_get_system_file_inode(osb, JOURNAL_SYSTEM_INODE,
1531 					    slot_num);
1532 	if (!inode || is_bad_inode(inode)) {
1533 		mlog_errno(status);
1534 		goto bail;
1535 	}
1536 	SET_INODE_JOURNAL(inode);
1537 
1538 	status = ocfs2_read_inode_block_full(inode, bh, OCFS2_BH_IGNORE_CACHE);
1539 	if (status < 0) {
1540 		mlog_errno(status);
1541 		goto bail;
1542 	}
1543 
1544 	status = 0;
1545 
1546 bail:
1547 	if (inode) {
1548 		if (status || !ret_inode)
1549 			iput(inode);
1550 		else
1551 			*ret_inode = inode;
1552 	}
1553 	return status;
1554 }
1555 
1556 /* Does the actual journal replay and marks the journal inode as
1557  * clean. Will only replay if the journal inode is marked dirty. */
1558 static int ocfs2_replay_journal(struct ocfs2_super *osb,
1559 				int node_num,
1560 				int slot_num)
1561 {
1562 	int status;
1563 	int got_lock = 0;
1564 	unsigned int flags;
1565 	struct inode *inode = NULL;
1566 	struct ocfs2_dinode *fe;
1567 	journal_t *journal = NULL;
1568 	struct buffer_head *bh = NULL;
1569 	u32 slot_reco_gen;
1570 
1571 	status = ocfs2_read_journal_inode(osb, slot_num, &bh, &inode);
1572 	if (status) {
1573 		mlog_errno(status);
1574 		goto done;
1575 	}
1576 
1577 	fe = (struct ocfs2_dinode *)bh->b_data;
1578 	slot_reco_gen = ocfs2_get_recovery_generation(fe);
1579 	brelse(bh);
1580 	bh = NULL;
1581 
1582 	/*
1583 	 * As the fs recovery is asynchronous, there is a small chance that
1584 	 * another node mounted (and recovered) the slot before the recovery
1585 	 * thread could get the lock. To handle that, we dirty read the journal
1586 	 * inode for that slot to get the recovery generation. If it is
1587 	 * different than what we expected, the slot has been recovered.
1588 	 * If not, it needs recovery.
1589 	 */
1590 	if (osb->slot_recovery_generations[slot_num] != slot_reco_gen) {
1591 		trace_ocfs2_replay_journal_recovered(slot_num,
1592 		     osb->slot_recovery_generations[slot_num], slot_reco_gen);
1593 		osb->slot_recovery_generations[slot_num] = slot_reco_gen;
1594 		status = -EBUSY;
1595 		goto done;
1596 	}
1597 
1598 	/* Continue with recovery as the journal has not yet been recovered */
1599 
1600 	status = ocfs2_inode_lock_full(inode, &bh, 1, OCFS2_META_LOCK_RECOVERY);
1601 	if (status < 0) {
1602 		trace_ocfs2_replay_journal_lock_err(status);
1603 		if (status != -ERESTARTSYS)
1604 			mlog(ML_ERROR, "Could not lock journal!\n");
1605 		goto done;
1606 	}
1607 	got_lock = 1;
1608 
1609 	fe = (struct ocfs2_dinode *) bh->b_data;
1610 
1611 	flags = le32_to_cpu(fe->id1.journal1.ij_flags);
1612 	slot_reco_gen = ocfs2_get_recovery_generation(fe);
1613 
1614 	if (!(flags & OCFS2_JOURNAL_DIRTY_FL)) {
1615 		trace_ocfs2_replay_journal_skip(node_num);
1616 		/* Refresh recovery generation for the slot */
1617 		osb->slot_recovery_generations[slot_num] = slot_reco_gen;
1618 		goto done;
1619 	}
1620 
1621 	/* we need to run complete recovery for offline orphan slots */
1622 	ocfs2_replay_map_set_state(osb, REPLAY_NEEDED);
1623 
1624 	printk(KERN_NOTICE "ocfs2: Begin replay journal (node %d, slot %d) on "\
1625 	       "device (%u,%u)\n", node_num, slot_num, MAJOR(osb->sb->s_dev),
1626 	       MINOR(osb->sb->s_dev));
1627 
1628 	OCFS2_I(inode)->ip_clusters = le32_to_cpu(fe->i_clusters);
1629 
1630 	status = ocfs2_force_read_journal(inode);
1631 	if (status < 0) {
1632 		mlog_errno(status);
1633 		goto done;
1634 	}
1635 
1636 	journal = jbd2_journal_init_inode(inode);
1637 	if (journal == NULL) {
1638 		mlog(ML_ERROR, "Linux journal layer error\n");
1639 		status = -EIO;
1640 		goto done;
1641 	}
1642 
1643 	status = jbd2_journal_load(journal);
1644 	if (status < 0) {
1645 		mlog_errno(status);
1646 		if (!igrab(inode))
1647 			BUG();
1648 		jbd2_journal_destroy(journal);
1649 		goto done;
1650 	}
1651 
1652 	ocfs2_clear_journal_error(osb->sb, journal, slot_num);
1653 
1654 	/* wipe the journal */
1655 	jbd2_journal_lock_updates(journal);
1656 	status = jbd2_journal_flush(journal);
1657 	jbd2_journal_unlock_updates(journal);
1658 	if (status < 0)
1659 		mlog_errno(status);
1660 
1661 	/* This will mark the node clean */
1662 	flags = le32_to_cpu(fe->id1.journal1.ij_flags);
1663 	flags &= ~OCFS2_JOURNAL_DIRTY_FL;
1664 	fe->id1.journal1.ij_flags = cpu_to_le32(flags);
1665 
1666 	/* Increment recovery generation to indicate successful recovery */
1667 	ocfs2_bump_recovery_generation(fe);
1668 	osb->slot_recovery_generations[slot_num] =
1669 					ocfs2_get_recovery_generation(fe);
1670 
1671 	ocfs2_compute_meta_ecc(osb->sb, bh->b_data, &fe->i_check);
1672 	status = ocfs2_write_block(osb, bh, INODE_CACHE(inode));
1673 	if (status < 0)
1674 		mlog_errno(status);
1675 
1676 	if (!igrab(inode))
1677 		BUG();
1678 
1679 	jbd2_journal_destroy(journal);
1680 
1681 	printk(KERN_NOTICE "ocfs2: End replay journal (node %d, slot %d) on "\
1682 	       "device (%u,%u)\n", node_num, slot_num, MAJOR(osb->sb->s_dev),
1683 	       MINOR(osb->sb->s_dev));
1684 done:
1685 	/* drop the lock on this nodes journal */
1686 	if (got_lock)
1687 		ocfs2_inode_unlock(inode, 1);
1688 
1689 	iput(inode);
1690 	brelse(bh);
1691 
1692 	return status;
1693 }
1694 
1695 /*
1696  * Do the most important parts of node recovery:
1697  *  - Replay it's journal
1698  *  - Stamp a clean local allocator file
1699  *  - Stamp a clean truncate log
1700  *  - Mark the node clean
1701  *
1702  * If this function completes without error, a node in OCFS2 can be
1703  * said to have been safely recovered. As a result, failure during the
1704  * second part of a nodes recovery process (local alloc recovery) is
1705  * far less concerning.
1706  */
1707 static int ocfs2_recover_node(struct ocfs2_super *osb,
1708 			      int node_num, int slot_num)
1709 {
1710 	int status = 0;
1711 	struct ocfs2_dinode *la_copy = NULL;
1712 	struct ocfs2_dinode *tl_copy = NULL;
1713 
1714 	trace_ocfs2_recover_node(node_num, slot_num, osb->node_num);
1715 
1716 	/* Should not ever be called to recover ourselves -- in that
1717 	 * case we should've called ocfs2_journal_load instead. */
1718 	BUG_ON(osb->node_num == node_num);
1719 
1720 	status = ocfs2_replay_journal(osb, node_num, slot_num);
1721 	if (status < 0) {
1722 		if (status == -EBUSY) {
1723 			trace_ocfs2_recover_node_skip(slot_num, node_num);
1724 			status = 0;
1725 			goto done;
1726 		}
1727 		mlog_errno(status);
1728 		goto done;
1729 	}
1730 
1731 	/* Stamp a clean local alloc file AFTER recovering the journal... */
1732 	status = ocfs2_begin_local_alloc_recovery(osb, slot_num, &la_copy);
1733 	if (status < 0) {
1734 		mlog_errno(status);
1735 		goto done;
1736 	}
1737 
1738 	/* An error from begin_truncate_log_recovery is not
1739 	 * serious enough to warrant halting the rest of
1740 	 * recovery. */
1741 	status = ocfs2_begin_truncate_log_recovery(osb, slot_num, &tl_copy);
1742 	if (status < 0)
1743 		mlog_errno(status);
1744 
1745 	/* Likewise, this would be a strange but ultimately not so
1746 	 * harmful place to get an error... */
1747 	status = ocfs2_clear_slot(osb, slot_num);
1748 	if (status < 0)
1749 		mlog_errno(status);
1750 
1751 	/* This will kfree the memory pointed to by la_copy and tl_copy */
1752 	ocfs2_queue_recovery_completion(osb->journal, slot_num, la_copy,
1753 					tl_copy, NULL, ORPHAN_NEED_TRUNCATE);
1754 
1755 	status = 0;
1756 done:
1757 
1758 	return status;
1759 }
1760 
1761 /* Test node liveness by trylocking his journal. If we get the lock,
1762  * we drop it here. Return 0 if we got the lock, -EAGAIN if node is
1763  * still alive (we couldn't get the lock) and < 0 on error. */
1764 static int ocfs2_trylock_journal(struct ocfs2_super *osb,
1765 				 int slot_num)
1766 {
1767 	int status, flags;
1768 	struct inode *inode = NULL;
1769 
1770 	inode = ocfs2_get_system_file_inode(osb, JOURNAL_SYSTEM_INODE,
1771 					    slot_num);
1772 	if (inode == NULL) {
1773 		mlog(ML_ERROR, "access error\n");
1774 		status = -EACCES;
1775 		goto bail;
1776 	}
1777 	if (is_bad_inode(inode)) {
1778 		mlog(ML_ERROR, "access error (bad inode)\n");
1779 		iput(inode);
1780 		inode = NULL;
1781 		status = -EACCES;
1782 		goto bail;
1783 	}
1784 	SET_INODE_JOURNAL(inode);
1785 
1786 	flags = OCFS2_META_LOCK_RECOVERY | OCFS2_META_LOCK_NOQUEUE;
1787 	status = ocfs2_inode_lock_full(inode, NULL, 1, flags);
1788 	if (status < 0) {
1789 		if (status != -EAGAIN)
1790 			mlog_errno(status);
1791 		goto bail;
1792 	}
1793 
1794 	ocfs2_inode_unlock(inode, 1);
1795 bail:
1796 	iput(inode);
1797 
1798 	return status;
1799 }
1800 
1801 /* Call this underneath ocfs2_super_lock. It also assumes that the
1802  * slot info struct has been updated from disk. */
1803 int ocfs2_mark_dead_nodes(struct ocfs2_super *osb)
1804 {
1805 	unsigned int node_num;
1806 	int status, i;
1807 	u32 gen;
1808 	struct buffer_head *bh = NULL;
1809 	struct ocfs2_dinode *di;
1810 
1811 	/* This is called with the super block cluster lock, so we
1812 	 * know that the slot map can't change underneath us. */
1813 
1814 	for (i = 0; i < osb->max_slots; i++) {
1815 		/* Read journal inode to get the recovery generation */
1816 		status = ocfs2_read_journal_inode(osb, i, &bh, NULL);
1817 		if (status) {
1818 			mlog_errno(status);
1819 			goto bail;
1820 		}
1821 		di = (struct ocfs2_dinode *)bh->b_data;
1822 		gen = ocfs2_get_recovery_generation(di);
1823 		brelse(bh);
1824 		bh = NULL;
1825 
1826 		spin_lock(&osb->osb_lock);
1827 		osb->slot_recovery_generations[i] = gen;
1828 
1829 		trace_ocfs2_mark_dead_nodes(i,
1830 					    osb->slot_recovery_generations[i]);
1831 
1832 		if (i == osb->slot_num) {
1833 			spin_unlock(&osb->osb_lock);
1834 			continue;
1835 		}
1836 
1837 		status = ocfs2_slot_to_node_num_locked(osb, i, &node_num);
1838 		if (status == -ENOENT) {
1839 			spin_unlock(&osb->osb_lock);
1840 			continue;
1841 		}
1842 
1843 		if (__ocfs2_recovery_map_test(osb, node_num)) {
1844 			spin_unlock(&osb->osb_lock);
1845 			continue;
1846 		}
1847 		spin_unlock(&osb->osb_lock);
1848 
1849 		/* Ok, we have a slot occupied by another node which
1850 		 * is not in the recovery map. We trylock his journal
1851 		 * file here to test if he's alive. */
1852 		status = ocfs2_trylock_journal(osb, i);
1853 		if (!status) {
1854 			/* Since we're called from mount, we know that
1855 			 * the recovery thread can't race us on
1856 			 * setting / checking the recovery bits. */
1857 			ocfs2_recovery_thread(osb, node_num);
1858 		} else if ((status < 0) && (status != -EAGAIN)) {
1859 			mlog_errno(status);
1860 			goto bail;
1861 		}
1862 	}
1863 
1864 	status = 0;
1865 bail:
1866 	return status;
1867 }
1868 
1869 /*
1870  * Scan timer should get fired every ORPHAN_SCAN_SCHEDULE_TIMEOUT. Add some
1871  * randomness to the timeout to minimize multple nodes firing the timer at the
1872  * same time.
1873  */
1874 static inline unsigned long ocfs2_orphan_scan_timeout(void)
1875 {
1876 	unsigned long time;
1877 
1878 	get_random_bytes(&time, sizeof(time));
1879 	time = ORPHAN_SCAN_SCHEDULE_TIMEOUT + (time % 5000);
1880 	return msecs_to_jiffies(time);
1881 }
1882 
1883 /*
1884  * ocfs2_queue_orphan_scan calls ocfs2_queue_recovery_completion for
1885  * every slot, queuing a recovery of the slot on the ocfs2_wq thread. This
1886  * is done to catch any orphans that are left over in orphan directories.
1887  *
1888  * It scans all slots, even ones that are in use. It does so to handle the
1889  * case described below:
1890  *
1891  *   Node 1 has an inode it was using. The dentry went away due to memory
1892  *   pressure.  Node 1 closes the inode, but it's on the free list. The node
1893  *   has the open lock.
1894  *   Node 2 unlinks the inode. It grabs the dentry lock to notify others,
1895  *   but node 1 has no dentry and doesn't get the message. It trylocks the
1896  *   open lock, sees that another node has a PR, and does nothing.
1897  *   Later node 2 runs its orphan dir. It igets the inode, trylocks the
1898  *   open lock, sees the PR still, and does nothing.
1899  *   Basically, we have to trigger an orphan iput on node 1. The only way
1900  *   for this to happen is if node 1 runs node 2's orphan dir.
1901  *
1902  * ocfs2_queue_orphan_scan gets called every ORPHAN_SCAN_SCHEDULE_TIMEOUT
1903  * seconds.  It gets an EX lock on os_lockres and checks sequence number
1904  * stored in LVB. If the sequence number has changed, it means some other
1905  * node has done the scan.  This node skips the scan and tracks the
1906  * sequence number.  If the sequence number didn't change, it means a scan
1907  * hasn't happened.  The node queues a scan and increments the
1908  * sequence number in the LVB.
1909  */
1910 static void ocfs2_queue_orphan_scan(struct ocfs2_super *osb)
1911 {
1912 	struct ocfs2_orphan_scan *os;
1913 	int status, i;
1914 	u32 seqno = 0;
1915 
1916 	os = &osb->osb_orphan_scan;
1917 
1918 	if (atomic_read(&os->os_state) == ORPHAN_SCAN_INACTIVE)
1919 		goto out;
1920 
1921 	trace_ocfs2_queue_orphan_scan_begin(os->os_count, os->os_seqno,
1922 					    atomic_read(&os->os_state));
1923 
1924 	status = ocfs2_orphan_scan_lock(osb, &seqno);
1925 	if (status < 0) {
1926 		if (status != -EAGAIN)
1927 			mlog_errno(status);
1928 		goto out;
1929 	}
1930 
1931 	/* Do no queue the tasks if the volume is being umounted */
1932 	if (atomic_read(&os->os_state) == ORPHAN_SCAN_INACTIVE)
1933 		goto unlock;
1934 
1935 	if (os->os_seqno != seqno) {
1936 		os->os_seqno = seqno;
1937 		goto unlock;
1938 	}
1939 
1940 	for (i = 0; i < osb->max_slots; i++)
1941 		ocfs2_queue_recovery_completion(osb->journal, i, NULL, NULL,
1942 						NULL, ORPHAN_NO_NEED_TRUNCATE);
1943 	/*
1944 	 * We queued a recovery on orphan slots, increment the sequence
1945 	 * number and update LVB so other node will skip the scan for a while
1946 	 */
1947 	seqno++;
1948 	os->os_count++;
1949 	os->os_scantime = CURRENT_TIME;
1950 unlock:
1951 	ocfs2_orphan_scan_unlock(osb, seqno);
1952 out:
1953 	trace_ocfs2_queue_orphan_scan_end(os->os_count, os->os_seqno,
1954 					  atomic_read(&os->os_state));
1955 	return;
1956 }
1957 
1958 /* Worker task that gets fired every ORPHAN_SCAN_SCHEDULE_TIMEOUT millsec */
1959 static void ocfs2_orphan_scan_work(struct work_struct *work)
1960 {
1961 	struct ocfs2_orphan_scan *os;
1962 	struct ocfs2_super *osb;
1963 
1964 	os = container_of(work, struct ocfs2_orphan_scan,
1965 			  os_orphan_scan_work.work);
1966 	osb = os->os_osb;
1967 
1968 	mutex_lock(&os->os_lock);
1969 	ocfs2_queue_orphan_scan(osb);
1970 	if (atomic_read(&os->os_state) == ORPHAN_SCAN_ACTIVE)
1971 		queue_delayed_work(osb->ocfs2_wq, &os->os_orphan_scan_work,
1972 				      ocfs2_orphan_scan_timeout());
1973 	mutex_unlock(&os->os_lock);
1974 }
1975 
1976 void ocfs2_orphan_scan_stop(struct ocfs2_super *osb)
1977 {
1978 	struct ocfs2_orphan_scan *os;
1979 
1980 	os = &osb->osb_orphan_scan;
1981 	if (atomic_read(&os->os_state) == ORPHAN_SCAN_ACTIVE) {
1982 		atomic_set(&os->os_state, ORPHAN_SCAN_INACTIVE);
1983 		mutex_lock(&os->os_lock);
1984 		cancel_delayed_work(&os->os_orphan_scan_work);
1985 		mutex_unlock(&os->os_lock);
1986 	}
1987 }
1988 
1989 void ocfs2_orphan_scan_init(struct ocfs2_super *osb)
1990 {
1991 	struct ocfs2_orphan_scan *os;
1992 
1993 	os = &osb->osb_orphan_scan;
1994 	os->os_osb = osb;
1995 	os->os_count = 0;
1996 	os->os_seqno = 0;
1997 	mutex_init(&os->os_lock);
1998 	INIT_DELAYED_WORK(&os->os_orphan_scan_work, ocfs2_orphan_scan_work);
1999 }
2000 
2001 void ocfs2_orphan_scan_start(struct ocfs2_super *osb)
2002 {
2003 	struct ocfs2_orphan_scan *os;
2004 
2005 	os = &osb->osb_orphan_scan;
2006 	os->os_scantime = CURRENT_TIME;
2007 	if (ocfs2_is_hard_readonly(osb) || ocfs2_mount_local(osb))
2008 		atomic_set(&os->os_state, ORPHAN_SCAN_INACTIVE);
2009 	else {
2010 		atomic_set(&os->os_state, ORPHAN_SCAN_ACTIVE);
2011 		queue_delayed_work(osb->ocfs2_wq, &os->os_orphan_scan_work,
2012 				   ocfs2_orphan_scan_timeout());
2013 	}
2014 }
2015 
2016 struct ocfs2_orphan_filldir_priv {
2017 	struct dir_context	ctx;
2018 	struct inode		*head;
2019 	struct ocfs2_super	*osb;
2020 	enum ocfs2_orphan_reco_type orphan_reco_type;
2021 };
2022 
2023 static int ocfs2_orphan_filldir(struct dir_context *ctx, const char *name,
2024 				int name_len, loff_t pos, u64 ino,
2025 				unsigned type)
2026 {
2027 	struct ocfs2_orphan_filldir_priv *p =
2028 		container_of(ctx, struct ocfs2_orphan_filldir_priv, ctx);
2029 	struct inode *iter;
2030 
2031 	if (name_len == 1 && !strncmp(".", name, 1))
2032 		return 0;
2033 	if (name_len == 2 && !strncmp("..", name, 2))
2034 		return 0;
2035 
2036 	/* do not include dio entry in case of orphan scan */
2037 	if ((p->orphan_reco_type == ORPHAN_NO_NEED_TRUNCATE) &&
2038 			(!strncmp(name, OCFS2_DIO_ORPHAN_PREFIX,
2039 			OCFS2_DIO_ORPHAN_PREFIX_LEN)))
2040 		return 0;
2041 
2042 	/* Skip bad inodes so that recovery can continue */
2043 	iter = ocfs2_iget(p->osb, ino,
2044 			  OCFS2_FI_FLAG_ORPHAN_RECOVERY, 0);
2045 	if (IS_ERR(iter))
2046 		return 0;
2047 
2048 	if (!strncmp(name, OCFS2_DIO_ORPHAN_PREFIX,
2049 			OCFS2_DIO_ORPHAN_PREFIX_LEN))
2050 		OCFS2_I(iter)->ip_flags |= OCFS2_INODE_DIO_ORPHAN_ENTRY;
2051 
2052 	/* Skip inodes which are already added to recover list, since dio may
2053 	 * happen concurrently with unlink/rename */
2054 	if (OCFS2_I(iter)->ip_next_orphan) {
2055 		iput(iter);
2056 		return 0;
2057 	}
2058 
2059 	trace_ocfs2_orphan_filldir((unsigned long long)OCFS2_I(iter)->ip_blkno);
2060 	/* No locking is required for the next_orphan queue as there
2061 	 * is only ever a single process doing orphan recovery. */
2062 	OCFS2_I(iter)->ip_next_orphan = p->head;
2063 	p->head = iter;
2064 
2065 	return 0;
2066 }
2067 
2068 static int ocfs2_queue_orphans(struct ocfs2_super *osb,
2069 			       int slot,
2070 			       struct inode **head,
2071 			       enum ocfs2_orphan_reco_type orphan_reco_type)
2072 {
2073 	int status;
2074 	struct inode *orphan_dir_inode = NULL;
2075 	struct ocfs2_orphan_filldir_priv priv = {
2076 		.ctx.actor = ocfs2_orphan_filldir,
2077 		.osb = osb,
2078 		.head = *head,
2079 		.orphan_reco_type = orphan_reco_type
2080 	};
2081 
2082 	orphan_dir_inode = ocfs2_get_system_file_inode(osb,
2083 						       ORPHAN_DIR_SYSTEM_INODE,
2084 						       slot);
2085 	if  (!orphan_dir_inode) {
2086 		status = -ENOENT;
2087 		mlog_errno(status);
2088 		return status;
2089 	}
2090 
2091 	inode_lock(orphan_dir_inode);
2092 	status = ocfs2_inode_lock(orphan_dir_inode, NULL, 0);
2093 	if (status < 0) {
2094 		mlog_errno(status);
2095 		goto out;
2096 	}
2097 
2098 	status = ocfs2_dir_foreach(orphan_dir_inode, &priv.ctx);
2099 	if (status) {
2100 		mlog_errno(status);
2101 		goto out_cluster;
2102 	}
2103 
2104 	*head = priv.head;
2105 
2106 out_cluster:
2107 	ocfs2_inode_unlock(orphan_dir_inode, 0);
2108 out:
2109 	inode_unlock(orphan_dir_inode);
2110 	iput(orphan_dir_inode);
2111 	return status;
2112 }
2113 
2114 static int ocfs2_orphan_recovery_can_continue(struct ocfs2_super *osb,
2115 					      int slot)
2116 {
2117 	int ret;
2118 
2119 	spin_lock(&osb->osb_lock);
2120 	ret = !osb->osb_orphan_wipes[slot];
2121 	spin_unlock(&osb->osb_lock);
2122 	return ret;
2123 }
2124 
2125 static void ocfs2_mark_recovering_orphan_dir(struct ocfs2_super *osb,
2126 					     int slot)
2127 {
2128 	spin_lock(&osb->osb_lock);
2129 	/* Mark ourselves such that new processes in delete_inode()
2130 	 * know to quit early. */
2131 	ocfs2_node_map_set_bit(osb, &osb->osb_recovering_orphan_dirs, slot);
2132 	while (osb->osb_orphan_wipes[slot]) {
2133 		/* If any processes are already in the middle of an
2134 		 * orphan wipe on this dir, then we need to wait for
2135 		 * them. */
2136 		spin_unlock(&osb->osb_lock);
2137 		wait_event_interruptible(osb->osb_wipe_event,
2138 					 ocfs2_orphan_recovery_can_continue(osb, slot));
2139 		spin_lock(&osb->osb_lock);
2140 	}
2141 	spin_unlock(&osb->osb_lock);
2142 }
2143 
2144 static void ocfs2_clear_recovering_orphan_dir(struct ocfs2_super *osb,
2145 					      int slot)
2146 {
2147 	ocfs2_node_map_clear_bit(osb, &osb->osb_recovering_orphan_dirs, slot);
2148 }
2149 
2150 /*
2151  * Orphan recovery. Each mounted node has it's own orphan dir which we
2152  * must run during recovery. Our strategy here is to build a list of
2153  * the inodes in the orphan dir and iget/iput them. The VFS does
2154  * (most) of the rest of the work.
2155  *
2156  * Orphan recovery can happen at any time, not just mount so we have a
2157  * couple of extra considerations.
2158  *
2159  * - We grab as many inodes as we can under the orphan dir lock -
2160  *   doing iget() outside the orphan dir risks getting a reference on
2161  *   an invalid inode.
2162  * - We must be sure not to deadlock with other processes on the
2163  *   system wanting to run delete_inode(). This can happen when they go
2164  *   to lock the orphan dir and the orphan recovery process attempts to
2165  *   iget() inside the orphan dir lock. This can be avoided by
2166  *   advertising our state to ocfs2_delete_inode().
2167  */
2168 static int ocfs2_recover_orphans(struct ocfs2_super *osb,
2169 				 int slot,
2170 				 enum ocfs2_orphan_reco_type orphan_reco_type)
2171 {
2172 	int ret = 0;
2173 	struct inode *inode = NULL;
2174 	struct inode *iter;
2175 	struct ocfs2_inode_info *oi;
2176 	struct buffer_head *di_bh = NULL;
2177 	struct ocfs2_dinode *di = NULL;
2178 
2179 	trace_ocfs2_recover_orphans(slot);
2180 
2181 	ocfs2_mark_recovering_orphan_dir(osb, slot);
2182 	ret = ocfs2_queue_orphans(osb, slot, &inode, orphan_reco_type);
2183 	ocfs2_clear_recovering_orphan_dir(osb, slot);
2184 
2185 	/* Error here should be noted, but we want to continue with as
2186 	 * many queued inodes as we've got. */
2187 	if (ret)
2188 		mlog_errno(ret);
2189 
2190 	while (inode) {
2191 		oi = OCFS2_I(inode);
2192 		trace_ocfs2_recover_orphans_iput(
2193 					(unsigned long long)oi->ip_blkno);
2194 
2195 		iter = oi->ip_next_orphan;
2196 		oi->ip_next_orphan = NULL;
2197 
2198 		if (oi->ip_flags & OCFS2_INODE_DIO_ORPHAN_ENTRY) {
2199 			inode_lock(inode);
2200 			ret = ocfs2_rw_lock(inode, 1);
2201 			if (ret < 0) {
2202 				mlog_errno(ret);
2203 				goto unlock_mutex;
2204 			}
2205 			/*
2206 			 * We need to take and drop the inode lock to
2207 			 * force read inode from disk.
2208 			 */
2209 			ret = ocfs2_inode_lock(inode, &di_bh, 1);
2210 			if (ret) {
2211 				mlog_errno(ret);
2212 				goto unlock_rw;
2213 			}
2214 
2215 			di = (struct ocfs2_dinode *)di_bh->b_data;
2216 
2217 			if (di->i_flags & cpu_to_le32(OCFS2_DIO_ORPHANED_FL)) {
2218 				ret = ocfs2_truncate_file(inode, di_bh,
2219 						i_size_read(inode));
2220 				if (ret < 0) {
2221 					if (ret != -ENOSPC)
2222 						mlog_errno(ret);
2223 					goto unlock_inode;
2224 				}
2225 
2226 				ret = ocfs2_del_inode_from_orphan(osb, inode,
2227 						di_bh, 0, 0);
2228 				if (ret)
2229 					mlog_errno(ret);
2230 			}
2231 unlock_inode:
2232 			ocfs2_inode_unlock(inode, 1);
2233 			brelse(di_bh);
2234 			di_bh = NULL;
2235 unlock_rw:
2236 			ocfs2_rw_unlock(inode, 1);
2237 unlock_mutex:
2238 			inode_unlock(inode);
2239 
2240 			/* clear dio flag in ocfs2_inode_info */
2241 			oi->ip_flags &= ~OCFS2_INODE_DIO_ORPHAN_ENTRY;
2242 		} else {
2243 			spin_lock(&oi->ip_lock);
2244 			/* Set the proper information to get us going into
2245 			 * ocfs2_delete_inode. */
2246 			oi->ip_flags |= OCFS2_INODE_MAYBE_ORPHANED;
2247 			spin_unlock(&oi->ip_lock);
2248 		}
2249 
2250 		iput(inode);
2251 		inode = iter;
2252 	}
2253 
2254 	return ret;
2255 }
2256 
2257 static int __ocfs2_wait_on_mount(struct ocfs2_super *osb, int quota)
2258 {
2259 	/* This check is good because ocfs2 will wait on our recovery
2260 	 * thread before changing it to something other than MOUNTED
2261 	 * or DISABLED. */
2262 	wait_event(osb->osb_mount_event,
2263 		  (!quota && atomic_read(&osb->vol_state) == VOLUME_MOUNTED) ||
2264 		   atomic_read(&osb->vol_state) == VOLUME_MOUNTED_QUOTAS ||
2265 		   atomic_read(&osb->vol_state) == VOLUME_DISABLED);
2266 
2267 	/* If there's an error on mount, then we may never get to the
2268 	 * MOUNTED flag, but this is set right before
2269 	 * dismount_volume() so we can trust it. */
2270 	if (atomic_read(&osb->vol_state) == VOLUME_DISABLED) {
2271 		trace_ocfs2_wait_on_mount(VOLUME_DISABLED);
2272 		mlog(0, "mount error, exiting!\n");
2273 		return -EBUSY;
2274 	}
2275 
2276 	return 0;
2277 }
2278 
2279 static int ocfs2_commit_thread(void *arg)
2280 {
2281 	int status;
2282 	struct ocfs2_super *osb = arg;
2283 	struct ocfs2_journal *journal = osb->journal;
2284 
2285 	/* we can trust j_num_trans here because _should_stop() is only set in
2286 	 * shutdown and nobody other than ourselves should be able to start
2287 	 * transactions.  committing on shutdown might take a few iterations
2288 	 * as final transactions put deleted inodes on the list */
2289 	while (!(kthread_should_stop() &&
2290 		 atomic_read(&journal->j_num_trans) == 0)) {
2291 
2292 		wait_event_interruptible(osb->checkpoint_event,
2293 					 atomic_read(&journal->j_num_trans)
2294 					 || kthread_should_stop());
2295 
2296 		status = ocfs2_commit_cache(osb);
2297 		if (status < 0) {
2298 			static unsigned long abort_warn_time;
2299 
2300 			/* Warn about this once per minute */
2301 			if (printk_timed_ratelimit(&abort_warn_time, 60*HZ))
2302 				mlog(ML_ERROR, "status = %d, journal is "
2303 						"already aborted.\n", status);
2304 			/*
2305 			 * After ocfs2_commit_cache() fails, j_num_trans has a
2306 			 * non-zero value.  Sleep here to avoid a busy-wait
2307 			 * loop.
2308 			 */
2309 			msleep_interruptible(1000);
2310 		}
2311 
2312 		if (kthread_should_stop() && atomic_read(&journal->j_num_trans)){
2313 			mlog(ML_KTHREAD,
2314 			     "commit_thread: %u transactions pending on "
2315 			     "shutdown\n",
2316 			     atomic_read(&journal->j_num_trans));
2317 		}
2318 	}
2319 
2320 	return 0;
2321 }
2322 
2323 /* Reads all the journal inodes without taking any cluster locks. Used
2324  * for hard readonly access to determine whether any journal requires
2325  * recovery. Also used to refresh the recovery generation numbers after
2326  * a journal has been recovered by another node.
2327  */
2328 int ocfs2_check_journals_nolocks(struct ocfs2_super *osb)
2329 {
2330 	int ret = 0;
2331 	unsigned int slot;
2332 	struct buffer_head *di_bh = NULL;
2333 	struct ocfs2_dinode *di;
2334 	int journal_dirty = 0;
2335 
2336 	for(slot = 0; slot < osb->max_slots; slot++) {
2337 		ret = ocfs2_read_journal_inode(osb, slot, &di_bh, NULL);
2338 		if (ret) {
2339 			mlog_errno(ret);
2340 			goto out;
2341 		}
2342 
2343 		di = (struct ocfs2_dinode *) di_bh->b_data;
2344 
2345 		osb->slot_recovery_generations[slot] =
2346 					ocfs2_get_recovery_generation(di);
2347 
2348 		if (le32_to_cpu(di->id1.journal1.ij_flags) &
2349 		    OCFS2_JOURNAL_DIRTY_FL)
2350 			journal_dirty = 1;
2351 
2352 		brelse(di_bh);
2353 		di_bh = NULL;
2354 	}
2355 
2356 out:
2357 	if (journal_dirty)
2358 		ret = -EROFS;
2359 	return ret;
2360 }
2361