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
59DEFINE_SPINLOCK(trans_inc_lock);
60
61#define ORPHAN_SCAN_SCHEDULE_TIMEOUT 300000
62
63static int ocfs2_force_read_journal(struct inode *inode);
64static int ocfs2_recover_node(struct ocfs2_super *osb,
65			      int node_num, int slot_num);
66static int __ocfs2_recovery_thread(void *arg);
67static int ocfs2_commit_cache(struct ocfs2_super *osb);
68static int __ocfs2_wait_on_mount(struct ocfs2_super *osb, int quota);
69static int ocfs2_journal_toggle_dirty(struct ocfs2_super *osb,
70				      int dirty, int replayed);
71static int ocfs2_trylock_journal(struct ocfs2_super *osb,
72				 int slot_num);
73static int ocfs2_recover_orphans(struct ocfs2_super *osb,
74				 int slot,
75				 enum ocfs2_orphan_reco_type orphan_reco_type);
76static int ocfs2_commit_thread(void *arg);
77static 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
84static inline int ocfs2_wait_on_mount(struct ocfs2_super *osb)
85{
86	return __ocfs2_wait_on_mount(osb, 0);
87}
88
89static 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
99enum 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
105struct ocfs2_replay_map {
106	unsigned int rm_slots;
107	enum ocfs2_replay_state rm_state;
108	unsigned char rm_replay_slots[0];
109};
110
111static 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
123int 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
156static 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
176static 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
187int 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 */
214static int ocfs2_recovery_thread_running(struct ocfs2_super *osb)
215{
216	mb();
217	return osb->recovery_thread_task != NULL;
218}
219
220void ocfs2_recovery_exit(struct ocfs2_super *osb)
221{
222	struct ocfs2_recovery_map *rm;
223
224	/* disable any new recovery threads and wait for any currently
225	 * running ones to exit. Do this before setting the vol_state. */
226	mutex_lock(&osb->recovery_lock);
227	osb->disable_recovery = 1;
228	mutex_unlock(&osb->recovery_lock);
229	wait_event(osb->recovery_event, !ocfs2_recovery_thread_running(osb));
230
231	/* At this point, we know that no more recovery threads can be
232	 * launched, so wait for any recovery completion work to
233	 * complete. */
234	flush_workqueue(ocfs2_wq);
235
236	/*
237	 * Now that recovery is shut down, and the osb is about to be
238	 * freed,  the osb_lock is not taken here.
239	 */
240	rm = osb->recovery_map;
241	/* XXX: Should we bug if there are dirty entries? */
242
243	kfree(rm);
244}
245
246static 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 */
263static 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
284static 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
307static 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);
344finally:
345	return status;
346}
347
348handle_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
388int 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 */
426int 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;
461bail:
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 */
471int 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
495bail:
496	return status;
497}
498
499
500struct ocfs2_triggers {
501	struct jbd2_buffer_trigger_type	ot_triggers;
502	int				ot_offset;
503};
504
505static 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
510static 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 */
529static 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 */
549static 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
565static 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
578static 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
586static 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
594static 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
602static 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
610static struct ocfs2_triggers db_triggers = {
611	.ot_triggers = {
612		.t_frozen = ocfs2_db_frozen_trigger,
613		.t_abort = ocfs2_abort_trigger,
614	},
615};
616
617static 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
625static struct ocfs2_triggers dq_triggers = {
626	.ot_triggers = {
627		.t_frozen = ocfs2_dq_frozen_trigger,
628		.t_abort = ocfs2_abort_trigger,
629	},
630};
631
632static 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
640static 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
648static 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
724int 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
730int 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
736int 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
743int 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
749int 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
755int 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
761int 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
767int 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
773int 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
779int 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
785void 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
810void 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
827int 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;
906done:
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
920static void ocfs2_bump_recovery_generation(struct ocfs2_dinode *di)
921{
922	le32_add_cpu(&(di->id1.journal1.ij_recovery_generation), 1);
923}
924
925static u32 ocfs2_get_recovery_generation(struct ocfs2_dinode *di)
926{
927	return le32_to_cpu(di->id1.journal1.ij_recovery_generation);
928}
929
930static 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 */
968void 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);
1044done:
1045	if (inode)
1046		iput(inode);
1047}
1048
1049static void ocfs2_clear_journal_error(struct super_block *sb,
1050				      journal_t *journal,
1051				      int slot)
1052{
1053	int olderr;
1054
1055	olderr = jbd2_journal_errno(journal);
1056	if (olderr) {
1057		mlog(ML_ERROR, "File system error %d recorded in "
1058		     "journal %u.\n", olderr, slot);
1059		mlog(ML_ERROR, "File system on device %s needs checking.\n",
1060		     sb->s_id);
1061
1062		jbd2_journal_ack_err(journal);
1063		jbd2_journal_clear_err(journal);
1064	}
1065}
1066
1067int ocfs2_journal_load(struct ocfs2_journal *journal, int local, int replayed)
1068{
1069	int status = 0;
1070	struct ocfs2_super *osb;
1071
1072	BUG_ON(!journal);
1073
1074	osb = journal->j_osb;
1075
1076	status = jbd2_journal_load(journal->j_journal);
1077	if (status < 0) {
1078		mlog(ML_ERROR, "Failed to load journal!\n");
1079		goto done;
1080	}
1081
1082	ocfs2_clear_journal_error(osb->sb, journal->j_journal, osb->slot_num);
1083
1084	status = ocfs2_journal_toggle_dirty(osb, 1, replayed);
1085	if (status < 0) {
1086		mlog_errno(status);
1087		goto done;
1088	}
1089
1090	/* Launch the commit thread */
1091	if (!local) {
1092		osb->commit_task = kthread_run(ocfs2_commit_thread, osb,
1093				"ocfs2cmt-%s", osb->uuid_str);
1094		if (IS_ERR(osb->commit_task)) {
1095			status = PTR_ERR(osb->commit_task);
1096			osb->commit_task = NULL;
1097			mlog(ML_ERROR, "unable to launch ocfs2commit thread, "
1098			     "error=%d", status);
1099			goto done;
1100		}
1101	} else
1102		osb->commit_task = NULL;
1103
1104done:
1105	return status;
1106}
1107
1108
1109/* 'full' flag tells us whether we clear out all blocks or if we just
1110 * mark the journal clean */
1111int ocfs2_journal_wipe(struct ocfs2_journal *journal, int full)
1112{
1113	int status;
1114
1115	BUG_ON(!journal);
1116
1117	status = jbd2_journal_wipe(journal->j_journal, full);
1118	if (status < 0) {
1119		mlog_errno(status);
1120		goto bail;
1121	}
1122
1123	status = ocfs2_journal_toggle_dirty(journal->j_osb, 0, 0);
1124	if (status < 0)
1125		mlog_errno(status);
1126
1127bail:
1128	return status;
1129}
1130
1131static int ocfs2_recovery_completed(struct ocfs2_super *osb)
1132{
1133	int empty;
1134	struct ocfs2_recovery_map *rm = osb->recovery_map;
1135
1136	spin_lock(&osb->osb_lock);
1137	empty = (rm->rm_used == 0);
1138	spin_unlock(&osb->osb_lock);
1139
1140	return empty;
1141}
1142
1143void ocfs2_wait_for_recovery(struct ocfs2_super *osb)
1144{
1145	wait_event(osb->recovery_event, ocfs2_recovery_completed(osb));
1146}
1147
1148/*
1149 * JBD Might read a cached version of another nodes journal file. We
1150 * don't want this as this file changes often and we get no
1151 * notification on those changes. The only way to be sure that we've
1152 * got the most up to date version of those blocks then is to force
1153 * read them off disk. Just searching through the buffer cache won't
1154 * work as there may be pages backing this file which are still marked
1155 * up to date. We know things can't change on this file underneath us
1156 * as we have the lock by now :)
1157 */
1158static int ocfs2_force_read_journal(struct inode *inode)
1159{
1160	int status = 0;
1161	int i;
1162	u64 v_blkno, p_blkno, p_blocks, num_blocks;
1163#define CONCURRENT_JOURNAL_FILL 32ULL
1164	struct buffer_head *bhs[CONCURRENT_JOURNAL_FILL];
1165
1166	memset(bhs, 0, sizeof(struct buffer_head *) * CONCURRENT_JOURNAL_FILL);
1167
1168	num_blocks = ocfs2_blocks_for_bytes(inode->i_sb, i_size_read(inode));
1169	v_blkno = 0;
1170	while (v_blkno < num_blocks) {
1171		status = ocfs2_extent_map_get_blocks(inode, v_blkno,
1172						     &p_blkno, &p_blocks, NULL);
1173		if (status < 0) {
1174			mlog_errno(status);
1175			goto bail;
1176		}
1177
1178		if (p_blocks > CONCURRENT_JOURNAL_FILL)
1179			p_blocks = CONCURRENT_JOURNAL_FILL;
1180
1181		/* We are reading journal data which should not
1182		 * be put in the uptodate cache */
1183		status = ocfs2_read_blocks_sync(OCFS2_SB(inode->i_sb),
1184						p_blkno, p_blocks, bhs);
1185		if (status < 0) {
1186			mlog_errno(status);
1187			goto bail;
1188		}
1189
1190		for(i = 0; i < p_blocks; i++) {
1191			brelse(bhs[i]);
1192			bhs[i] = NULL;
1193		}
1194
1195		v_blkno += p_blocks;
1196	}
1197
1198bail:
1199	for(i = 0; i < CONCURRENT_JOURNAL_FILL; i++)
1200		brelse(bhs[i]);
1201	return status;
1202}
1203
1204struct ocfs2_la_recovery_item {
1205	struct list_head	lri_list;
1206	int			lri_slot;
1207	struct ocfs2_dinode	*lri_la_dinode;
1208	struct ocfs2_dinode	*lri_tl_dinode;
1209	struct ocfs2_quota_recovery *lri_qrec;
1210	enum ocfs2_orphan_reco_type  lri_orphan_reco_type;
1211};
1212
1213/* Does the second half of the recovery process. By this point, the
1214 * node is marked clean and can actually be considered recovered,
1215 * hence it's no longer in the recovery map, but there's still some
1216 * cleanup we can do which shouldn't happen within the recovery thread
1217 * as locking in that context becomes very difficult if we are to take
1218 * recovering nodes into account.
1219 *
1220 * NOTE: This function can and will sleep on recovery of other nodes
1221 * during cluster locking, just like any other ocfs2 process.
1222 */
1223void ocfs2_complete_recovery(struct work_struct *work)
1224{
1225	int ret = 0;
1226	struct ocfs2_journal *journal =
1227		container_of(work, struct ocfs2_journal, j_recovery_work);
1228	struct ocfs2_super *osb = journal->j_osb;
1229	struct ocfs2_dinode *la_dinode, *tl_dinode;
1230	struct ocfs2_la_recovery_item *item, *n;
1231	struct ocfs2_quota_recovery *qrec;
1232	enum ocfs2_orphan_reco_type orphan_reco_type;
1233	LIST_HEAD(tmp_la_list);
1234
1235	trace_ocfs2_complete_recovery(
1236		(unsigned long long)OCFS2_I(journal->j_inode)->ip_blkno);
1237
1238	spin_lock(&journal->j_lock);
1239	list_splice_init(&journal->j_la_cleanups, &tmp_la_list);
1240	spin_unlock(&journal->j_lock);
1241
1242	list_for_each_entry_safe(item, n, &tmp_la_list, lri_list) {
1243		list_del_init(&item->lri_list);
1244
1245		ocfs2_wait_on_quotas(osb);
1246
1247		la_dinode = item->lri_la_dinode;
1248		tl_dinode = item->lri_tl_dinode;
1249		qrec = item->lri_qrec;
1250		orphan_reco_type = item->lri_orphan_reco_type;
1251
1252		trace_ocfs2_complete_recovery_slot(item->lri_slot,
1253			la_dinode ? le64_to_cpu(la_dinode->i_blkno) : 0,
1254			tl_dinode ? le64_to_cpu(tl_dinode->i_blkno) : 0,
1255			qrec);
1256
1257		if (la_dinode) {
1258			ret = ocfs2_complete_local_alloc_recovery(osb,
1259								  la_dinode);
1260			if (ret < 0)
1261				mlog_errno(ret);
1262
1263			kfree(la_dinode);
1264		}
1265
1266		if (tl_dinode) {
1267			ret = ocfs2_complete_truncate_log_recovery(osb,
1268								   tl_dinode);
1269			if (ret < 0)
1270				mlog_errno(ret);
1271
1272			kfree(tl_dinode);
1273		}
1274
1275		ret = ocfs2_recover_orphans(osb, item->lri_slot,
1276				orphan_reco_type);
1277		if (ret < 0)
1278			mlog_errno(ret);
1279
1280		if (qrec) {
1281			ret = ocfs2_finish_quota_recovery(osb, qrec,
1282							  item->lri_slot);
1283			if (ret < 0)
1284				mlog_errno(ret);
1285			/* Recovery info is already freed now */
1286		}
1287
1288		kfree(item);
1289	}
1290
1291	trace_ocfs2_complete_recovery_end(ret);
1292}
1293
1294/* NOTE: This function always eats your references to la_dinode and
1295 * tl_dinode, either manually on error, or by passing them to
1296 * ocfs2_complete_recovery */
1297static void ocfs2_queue_recovery_completion(struct ocfs2_journal *journal,
1298					    int slot_num,
1299					    struct ocfs2_dinode *la_dinode,
1300					    struct ocfs2_dinode *tl_dinode,
1301					    struct ocfs2_quota_recovery *qrec,
1302					    enum ocfs2_orphan_reco_type orphan_reco_type)
1303{
1304	struct ocfs2_la_recovery_item *item;
1305
1306	item = kmalloc(sizeof(struct ocfs2_la_recovery_item), GFP_NOFS);
1307	if (!item) {
1308		/* Though we wish to avoid it, we are in fact safe in
1309		 * skipping local alloc cleanup as fsck.ocfs2 is more
1310		 * than capable of reclaiming unused space. */
1311		kfree(la_dinode);
1312		kfree(tl_dinode);
1313
1314		if (qrec)
1315			ocfs2_free_quota_recovery(qrec);
1316
1317		mlog_errno(-ENOMEM);
1318		return;
1319	}
1320
1321	INIT_LIST_HEAD(&item->lri_list);
1322	item->lri_la_dinode = la_dinode;
1323	item->lri_slot = slot_num;
1324	item->lri_tl_dinode = tl_dinode;
1325	item->lri_qrec = qrec;
1326	item->lri_orphan_reco_type = orphan_reco_type;
1327
1328	spin_lock(&journal->j_lock);
1329	list_add_tail(&item->lri_list, &journal->j_la_cleanups);
1330	queue_work(ocfs2_wq, &journal->j_recovery_work);
1331	spin_unlock(&journal->j_lock);
1332}
1333
1334/* Called by the mount code to queue recovery the last part of
1335 * recovery for it's own and offline slot(s). */
1336void ocfs2_complete_mount_recovery(struct ocfs2_super *osb)
1337{
1338	struct ocfs2_journal *journal = osb->journal;
1339
1340	if (ocfs2_is_hard_readonly(osb))
1341		return;
1342
1343	/* No need to queue up our truncate_log as regular cleanup will catch
1344	 * that */
1345	ocfs2_queue_recovery_completion(journal, osb->slot_num,
1346					osb->local_alloc_copy, NULL, NULL,
1347					ORPHAN_NEED_TRUNCATE);
1348	ocfs2_schedule_truncate_log_flush(osb, 0);
1349
1350	osb->local_alloc_copy = NULL;
1351	osb->dirty = 0;
1352
1353	/* queue to recover orphan slots for all offline slots */
1354	ocfs2_replay_map_set_state(osb, REPLAY_NEEDED);
1355	ocfs2_queue_replay_slots(osb, ORPHAN_NEED_TRUNCATE);
1356	ocfs2_free_replay_slots(osb);
1357}
1358
1359void ocfs2_complete_quota_recovery(struct ocfs2_super *osb)
1360{
1361	if (osb->quota_rec) {
1362		ocfs2_queue_recovery_completion(osb->journal,
1363						osb->slot_num,
1364						NULL,
1365						NULL,
1366						osb->quota_rec,
1367						ORPHAN_NEED_TRUNCATE);
1368		osb->quota_rec = NULL;
1369	}
1370}
1371
1372static int __ocfs2_recovery_thread(void *arg)
1373{
1374	int status, node_num, slot_num;
1375	struct ocfs2_super *osb = arg;
1376	struct ocfs2_recovery_map *rm = osb->recovery_map;
1377	int *rm_quota = NULL;
1378	int rm_quota_used = 0, i;
1379	struct ocfs2_quota_recovery *qrec;
1380
1381	status = ocfs2_wait_on_mount(osb);
1382	if (status < 0) {
1383		goto bail;
1384	}
1385
1386	rm_quota = kzalloc(osb->max_slots * sizeof(int), GFP_NOFS);
1387	if (!rm_quota) {
1388		status = -ENOMEM;
1389		goto bail;
1390	}
1391restart:
1392	status = ocfs2_super_lock(osb, 1);
1393	if (status < 0) {
1394		mlog_errno(status);
1395		goto bail;
1396	}
1397
1398	status = ocfs2_compute_replay_slots(osb);
1399	if (status < 0)
1400		mlog_errno(status);
1401
1402	/* queue recovery for our own slot */
1403	ocfs2_queue_recovery_completion(osb->journal, osb->slot_num, NULL,
1404					NULL, NULL, ORPHAN_NO_NEED_TRUNCATE);
1405
1406	spin_lock(&osb->osb_lock);
1407	while (rm->rm_used) {
1408		/* It's always safe to remove entry zero, as we won't
1409		 * clear it until ocfs2_recover_node() has succeeded. */
1410		node_num = rm->rm_entries[0];
1411		spin_unlock(&osb->osb_lock);
1412		slot_num = ocfs2_node_num_to_slot(osb, node_num);
1413		trace_ocfs2_recovery_thread_node(node_num, slot_num);
1414		if (slot_num == -ENOENT) {
1415			status = 0;
1416			goto skip_recovery;
1417		}
1418
1419		/* It is a bit subtle with quota recovery. We cannot do it
1420		 * immediately because we have to obtain cluster locks from
1421		 * quota files and we also don't want to just skip it because
1422		 * then quota usage would be out of sync until some node takes
1423		 * the slot. So we remember which nodes need quota recovery
1424		 * and when everything else is done, we recover quotas. */
1425		for (i = 0; i < rm_quota_used && rm_quota[i] != slot_num; i++);
1426		if (i == rm_quota_used)
1427			rm_quota[rm_quota_used++] = slot_num;
1428
1429		status = ocfs2_recover_node(osb, node_num, slot_num);
1430skip_recovery:
1431		if (!status) {
1432			ocfs2_recovery_map_clear(osb, node_num);
1433		} else {
1434			mlog(ML_ERROR,
1435			     "Error %d recovering node %d on device (%u,%u)!\n",
1436			     status, node_num,
1437			     MAJOR(osb->sb->s_dev), MINOR(osb->sb->s_dev));
1438			mlog(ML_ERROR, "Volume requires unmount.\n");
1439		}
1440
1441		spin_lock(&osb->osb_lock);
1442	}
1443	spin_unlock(&osb->osb_lock);
1444	trace_ocfs2_recovery_thread_end(status);
1445
1446	/* Refresh all journal recovery generations from disk */
1447	status = ocfs2_check_journals_nolocks(osb);
1448	status = (status == -EROFS) ? 0 : status;
1449	if (status < 0)
1450		mlog_errno(status);
1451
1452	/* Now it is right time to recover quotas... We have to do this under
1453	 * superblock lock so that no one can start using the slot (and crash)
1454	 * before we recover it */
1455	for (i = 0; i < rm_quota_used; i++) {
1456		qrec = ocfs2_begin_quota_recovery(osb, rm_quota[i]);
1457		if (IS_ERR(qrec)) {
1458			status = PTR_ERR(qrec);
1459			mlog_errno(status);
1460			continue;
1461		}
1462		ocfs2_queue_recovery_completion(osb->journal, rm_quota[i],
1463						NULL, NULL, qrec,
1464						ORPHAN_NEED_TRUNCATE);
1465	}
1466
1467	ocfs2_super_unlock(osb, 1);
1468
1469	/* queue recovery for offline slots */
1470	ocfs2_queue_replay_slots(osb, ORPHAN_NEED_TRUNCATE);
1471
1472bail:
1473	mutex_lock(&osb->recovery_lock);
1474	if (!status && !ocfs2_recovery_completed(osb)) {
1475		mutex_unlock(&osb->recovery_lock);
1476		goto restart;
1477	}
1478
1479	ocfs2_free_replay_slots(osb);
1480	osb->recovery_thread_task = NULL;
1481	mb(); /* sync with ocfs2_recovery_thread_running */
1482	wake_up(&osb->recovery_event);
1483
1484	mutex_unlock(&osb->recovery_lock);
1485
1486	kfree(rm_quota);
1487
1488	/* no one is callint kthread_stop() for us so the kthread() api
1489	 * requires that we call do_exit().  And it isn't exported, but
1490	 * complete_and_exit() seems to be a minimal wrapper around it. */
1491	complete_and_exit(NULL, status);
1492}
1493
1494void ocfs2_recovery_thread(struct ocfs2_super *osb, int node_num)
1495{
1496	mutex_lock(&osb->recovery_lock);
1497
1498	trace_ocfs2_recovery_thread(node_num, osb->node_num,
1499		osb->disable_recovery, osb->recovery_thread_task,
1500		osb->disable_recovery ?
1501		-1 : ocfs2_recovery_map_set(osb, node_num));
1502
1503	if (osb->disable_recovery)
1504		goto out;
1505
1506	if (osb->recovery_thread_task)
1507		goto out;
1508
1509	osb->recovery_thread_task =  kthread_run(__ocfs2_recovery_thread, osb,
1510			"ocfs2rec-%s", osb->uuid_str);
1511	if (IS_ERR(osb->recovery_thread_task)) {
1512		mlog_errno((int)PTR_ERR(osb->recovery_thread_task));
1513		osb->recovery_thread_task = NULL;
1514	}
1515
1516out:
1517	mutex_unlock(&osb->recovery_lock);
1518	wake_up(&osb->recovery_event);
1519}
1520
1521static int ocfs2_read_journal_inode(struct ocfs2_super *osb,
1522				    int slot_num,
1523				    struct buffer_head **bh,
1524				    struct inode **ret_inode)
1525{
1526	int status = -EACCES;
1527	struct inode *inode = NULL;
1528
1529	BUG_ON(slot_num >= osb->max_slots);
1530
1531	inode = ocfs2_get_system_file_inode(osb, JOURNAL_SYSTEM_INODE,
1532					    slot_num);
1533	if (!inode || is_bad_inode(inode)) {
1534		mlog_errno(status);
1535		goto bail;
1536	}
1537	SET_INODE_JOURNAL(inode);
1538
1539	status = ocfs2_read_inode_block_full(inode, bh, OCFS2_BH_IGNORE_CACHE);
1540	if (status < 0) {
1541		mlog_errno(status);
1542		goto bail;
1543	}
1544
1545	status = 0;
1546
1547bail:
1548	if (inode) {
1549		if (status || !ret_inode)
1550			iput(inode);
1551		else
1552			*ret_inode = inode;
1553	}
1554	return status;
1555}
1556
1557/* Does the actual journal replay and marks the journal inode as
1558 * clean. Will only replay if the journal inode is marked dirty. */
1559static int ocfs2_replay_journal(struct ocfs2_super *osb,
1560				int node_num,
1561				int slot_num)
1562{
1563	int status;
1564	int got_lock = 0;
1565	unsigned int flags;
1566	struct inode *inode = NULL;
1567	struct ocfs2_dinode *fe;
1568	journal_t *journal = NULL;
1569	struct buffer_head *bh = NULL;
1570	u32 slot_reco_gen;
1571
1572	status = ocfs2_read_journal_inode(osb, slot_num, &bh, &inode);
1573	if (status) {
1574		mlog_errno(status);
1575		goto done;
1576	}
1577
1578	fe = (struct ocfs2_dinode *)bh->b_data;
1579	slot_reco_gen = ocfs2_get_recovery_generation(fe);
1580	brelse(bh);
1581	bh = NULL;
1582
1583	/*
1584	 * As the fs recovery is asynchronous, there is a small chance that
1585	 * another node mounted (and recovered) the slot before the recovery
1586	 * thread could get the lock. To handle that, we dirty read the journal
1587	 * inode for that slot to get the recovery generation. If it is
1588	 * different than what we expected, the slot has been recovered.
1589	 * If not, it needs recovery.
1590	 */
1591	if (osb->slot_recovery_generations[slot_num] != slot_reco_gen) {
1592		trace_ocfs2_replay_journal_recovered(slot_num,
1593		     osb->slot_recovery_generations[slot_num], slot_reco_gen);
1594		osb->slot_recovery_generations[slot_num] = slot_reco_gen;
1595		status = -EBUSY;
1596		goto done;
1597	}
1598
1599	/* Continue with recovery as the journal has not yet been recovered */
1600
1601	status = ocfs2_inode_lock_full(inode, &bh, 1, OCFS2_META_LOCK_RECOVERY);
1602	if (status < 0) {
1603		trace_ocfs2_replay_journal_lock_err(status);
1604		if (status != -ERESTARTSYS)
1605			mlog(ML_ERROR, "Could not lock journal!\n");
1606		goto done;
1607	}
1608	got_lock = 1;
1609
1610	fe = (struct ocfs2_dinode *) bh->b_data;
1611
1612	flags = le32_to_cpu(fe->id1.journal1.ij_flags);
1613	slot_reco_gen = ocfs2_get_recovery_generation(fe);
1614
1615	if (!(flags & OCFS2_JOURNAL_DIRTY_FL)) {
1616		trace_ocfs2_replay_journal_skip(node_num);
1617		/* Refresh recovery generation for the slot */
1618		osb->slot_recovery_generations[slot_num] = slot_reco_gen;
1619		goto done;
1620	}
1621
1622	/* we need to run complete recovery for offline orphan slots */
1623	ocfs2_replay_map_set_state(osb, REPLAY_NEEDED);
1624
1625	printk(KERN_NOTICE "ocfs2: Begin replay journal (node %d, slot %d) on "\
1626	       "device (%u,%u)\n", node_num, slot_num, MAJOR(osb->sb->s_dev),
1627	       MINOR(osb->sb->s_dev));
1628
1629	OCFS2_I(inode)->ip_clusters = le32_to_cpu(fe->i_clusters);
1630
1631	status = ocfs2_force_read_journal(inode);
1632	if (status < 0) {
1633		mlog_errno(status);
1634		goto done;
1635	}
1636
1637	journal = jbd2_journal_init_inode(inode);
1638	if (journal == NULL) {
1639		mlog(ML_ERROR, "Linux journal layer error\n");
1640		status = -EIO;
1641		goto done;
1642	}
1643
1644	status = jbd2_journal_load(journal);
1645	if (status < 0) {
1646		mlog_errno(status);
1647		if (!igrab(inode))
1648			BUG();
1649		jbd2_journal_destroy(journal);
1650		goto done;
1651	}
1652
1653	ocfs2_clear_journal_error(osb->sb, journal, slot_num);
1654
1655	/* wipe the journal */
1656	jbd2_journal_lock_updates(journal);
1657	status = jbd2_journal_flush(journal);
1658	jbd2_journal_unlock_updates(journal);
1659	if (status < 0)
1660		mlog_errno(status);
1661
1662	/* This will mark the node clean */
1663	flags = le32_to_cpu(fe->id1.journal1.ij_flags);
1664	flags &= ~OCFS2_JOURNAL_DIRTY_FL;
1665	fe->id1.journal1.ij_flags = cpu_to_le32(flags);
1666
1667	/* Increment recovery generation to indicate successful recovery */
1668	ocfs2_bump_recovery_generation(fe);
1669	osb->slot_recovery_generations[slot_num] =
1670					ocfs2_get_recovery_generation(fe);
1671
1672	ocfs2_compute_meta_ecc(osb->sb, bh->b_data, &fe->i_check);
1673	status = ocfs2_write_block(osb, bh, INODE_CACHE(inode));
1674	if (status < 0)
1675		mlog_errno(status);
1676
1677	if (!igrab(inode))
1678		BUG();
1679
1680	jbd2_journal_destroy(journal);
1681
1682	printk(KERN_NOTICE "ocfs2: End replay journal (node %d, slot %d) on "\
1683	       "device (%u,%u)\n", node_num, slot_num, MAJOR(osb->sb->s_dev),
1684	       MINOR(osb->sb->s_dev));
1685done:
1686	/* drop the lock on this nodes journal */
1687	if (got_lock)
1688		ocfs2_inode_unlock(inode, 1);
1689
1690	if (inode)
1691		iput(inode);
1692
1693	brelse(bh);
1694
1695	return status;
1696}
1697
1698/*
1699 * Do the most important parts of node recovery:
1700 *  - Replay it's journal
1701 *  - Stamp a clean local allocator file
1702 *  - Stamp a clean truncate log
1703 *  - Mark the node clean
1704 *
1705 * If this function completes without error, a node in OCFS2 can be
1706 * said to have been safely recovered. As a result, failure during the
1707 * second part of a nodes recovery process (local alloc recovery) is
1708 * far less concerning.
1709 */
1710static int ocfs2_recover_node(struct ocfs2_super *osb,
1711			      int node_num, int slot_num)
1712{
1713	int status = 0;
1714	struct ocfs2_dinode *la_copy = NULL;
1715	struct ocfs2_dinode *tl_copy = NULL;
1716
1717	trace_ocfs2_recover_node(node_num, slot_num, osb->node_num);
1718
1719	/* Should not ever be called to recover ourselves -- in that
1720	 * case we should've called ocfs2_journal_load instead. */
1721	BUG_ON(osb->node_num == node_num);
1722
1723	status = ocfs2_replay_journal(osb, node_num, slot_num);
1724	if (status < 0) {
1725		if (status == -EBUSY) {
1726			trace_ocfs2_recover_node_skip(slot_num, node_num);
1727			status = 0;
1728			goto done;
1729		}
1730		mlog_errno(status);
1731		goto done;
1732	}
1733
1734	/* Stamp a clean local alloc file AFTER recovering the journal... */
1735	status = ocfs2_begin_local_alloc_recovery(osb, slot_num, &la_copy);
1736	if (status < 0) {
1737		mlog_errno(status);
1738		goto done;
1739	}
1740
1741	/* An error from begin_truncate_log_recovery is not
1742	 * serious enough to warrant halting the rest of
1743	 * recovery. */
1744	status = ocfs2_begin_truncate_log_recovery(osb, slot_num, &tl_copy);
1745	if (status < 0)
1746		mlog_errno(status);
1747
1748	/* Likewise, this would be a strange but ultimately not so
1749	 * harmful place to get an error... */
1750	status = ocfs2_clear_slot(osb, slot_num);
1751	if (status < 0)
1752		mlog_errno(status);
1753
1754	/* This will kfree the memory pointed to by la_copy and tl_copy */
1755	ocfs2_queue_recovery_completion(osb->journal, slot_num, la_copy,
1756					tl_copy, NULL, ORPHAN_NEED_TRUNCATE);
1757
1758	status = 0;
1759done:
1760
1761	return status;
1762}
1763
1764/* Test node liveness by trylocking his journal. If we get the lock,
1765 * we drop it here. Return 0 if we got the lock, -EAGAIN if node is
1766 * still alive (we couldn't get the lock) and < 0 on error. */
1767static int ocfs2_trylock_journal(struct ocfs2_super *osb,
1768				 int slot_num)
1769{
1770	int status, flags;
1771	struct inode *inode = NULL;
1772
1773	inode = ocfs2_get_system_file_inode(osb, JOURNAL_SYSTEM_INODE,
1774					    slot_num);
1775	if (inode == NULL) {
1776		mlog(ML_ERROR, "access error\n");
1777		status = -EACCES;
1778		goto bail;
1779	}
1780	if (is_bad_inode(inode)) {
1781		mlog(ML_ERROR, "access error (bad inode)\n");
1782		iput(inode);
1783		inode = NULL;
1784		status = -EACCES;
1785		goto bail;
1786	}
1787	SET_INODE_JOURNAL(inode);
1788
1789	flags = OCFS2_META_LOCK_RECOVERY | OCFS2_META_LOCK_NOQUEUE;
1790	status = ocfs2_inode_lock_full(inode, NULL, 1, flags);
1791	if (status < 0) {
1792		if (status != -EAGAIN)
1793			mlog_errno(status);
1794		goto bail;
1795	}
1796
1797	ocfs2_inode_unlock(inode, 1);
1798bail:
1799	if (inode)
1800		iput(inode);
1801
1802	return status;
1803}
1804
1805/* Call this underneath ocfs2_super_lock. It also assumes that the
1806 * slot info struct has been updated from disk. */
1807int ocfs2_mark_dead_nodes(struct ocfs2_super *osb)
1808{
1809	unsigned int node_num;
1810	int status, i;
1811	u32 gen;
1812	struct buffer_head *bh = NULL;
1813	struct ocfs2_dinode *di;
1814
1815	/* This is called with the super block cluster lock, so we
1816	 * know that the slot map can't change underneath us. */
1817
1818	for (i = 0; i < osb->max_slots; i++) {
1819		/* Read journal inode to get the recovery generation */
1820		status = ocfs2_read_journal_inode(osb, i, &bh, NULL);
1821		if (status) {
1822			mlog_errno(status);
1823			goto bail;
1824		}
1825		di = (struct ocfs2_dinode *)bh->b_data;
1826		gen = ocfs2_get_recovery_generation(di);
1827		brelse(bh);
1828		bh = NULL;
1829
1830		spin_lock(&osb->osb_lock);
1831		osb->slot_recovery_generations[i] = gen;
1832
1833		trace_ocfs2_mark_dead_nodes(i,
1834					    osb->slot_recovery_generations[i]);
1835
1836		if (i == osb->slot_num) {
1837			spin_unlock(&osb->osb_lock);
1838			continue;
1839		}
1840
1841		status = ocfs2_slot_to_node_num_locked(osb, i, &node_num);
1842		if (status == -ENOENT) {
1843			spin_unlock(&osb->osb_lock);
1844			continue;
1845		}
1846
1847		if (__ocfs2_recovery_map_test(osb, node_num)) {
1848			spin_unlock(&osb->osb_lock);
1849			continue;
1850		}
1851		spin_unlock(&osb->osb_lock);
1852
1853		/* Ok, we have a slot occupied by another node which
1854		 * is not in the recovery map. We trylock his journal
1855		 * file here to test if he's alive. */
1856		status = ocfs2_trylock_journal(osb, i);
1857		if (!status) {
1858			/* Since we're called from mount, we know that
1859			 * the recovery thread can't race us on
1860			 * setting / checking the recovery bits. */
1861			ocfs2_recovery_thread(osb, node_num);
1862		} else if ((status < 0) && (status != -EAGAIN)) {
1863			mlog_errno(status);
1864			goto bail;
1865		}
1866	}
1867
1868	status = 0;
1869bail:
1870	return status;
1871}
1872
1873/*
1874 * Scan timer should get fired every ORPHAN_SCAN_SCHEDULE_TIMEOUT. Add some
1875 * randomness to the timeout to minimize multple nodes firing the timer at the
1876 * same time.
1877 */
1878static inline unsigned long ocfs2_orphan_scan_timeout(void)
1879{
1880	unsigned long time;
1881
1882	get_random_bytes(&time, sizeof(time));
1883	time = ORPHAN_SCAN_SCHEDULE_TIMEOUT + (time % 5000);
1884	return msecs_to_jiffies(time);
1885}
1886
1887/*
1888 * ocfs2_queue_orphan_scan calls ocfs2_queue_recovery_completion for
1889 * every slot, queuing a recovery of the slot on the ocfs2_wq thread. This
1890 * is done to catch any orphans that are left over in orphan directories.
1891 *
1892 * It scans all slots, even ones that are in use. It does so to handle the
1893 * case described below:
1894 *
1895 *   Node 1 has an inode it was using. The dentry went away due to memory
1896 *   pressure.  Node 1 closes the inode, but it's on the free list. The node
1897 *   has the open lock.
1898 *   Node 2 unlinks the inode. It grabs the dentry lock to notify others,
1899 *   but node 1 has no dentry and doesn't get the message. It trylocks the
1900 *   open lock, sees that another node has a PR, and does nothing.
1901 *   Later node 2 runs its orphan dir. It igets the inode, trylocks the
1902 *   open lock, sees the PR still, and does nothing.
1903 *   Basically, we have to trigger an orphan iput on node 1. The only way
1904 *   for this to happen is if node 1 runs node 2's orphan dir.
1905 *
1906 * ocfs2_queue_orphan_scan gets called every ORPHAN_SCAN_SCHEDULE_TIMEOUT
1907 * seconds.  It gets an EX lock on os_lockres and checks sequence number
1908 * stored in LVB. If the sequence number has changed, it means some other
1909 * node has done the scan.  This node skips the scan and tracks the
1910 * sequence number.  If the sequence number didn't change, it means a scan
1911 * hasn't happened.  The node queues a scan and increments the
1912 * sequence number in the LVB.
1913 */
1914static void ocfs2_queue_orphan_scan(struct ocfs2_super *osb)
1915{
1916	struct ocfs2_orphan_scan *os;
1917	int status, i;
1918	u32 seqno = 0;
1919
1920	os = &osb->osb_orphan_scan;
1921
1922	if (atomic_read(&os->os_state) == ORPHAN_SCAN_INACTIVE)
1923		goto out;
1924
1925	trace_ocfs2_queue_orphan_scan_begin(os->os_count, os->os_seqno,
1926					    atomic_read(&os->os_state));
1927
1928	status = ocfs2_orphan_scan_lock(osb, &seqno);
1929	if (status < 0) {
1930		if (status != -EAGAIN)
1931			mlog_errno(status);
1932		goto out;
1933	}
1934
1935	/* Do no queue the tasks if the volume is being umounted */
1936	if (atomic_read(&os->os_state) == ORPHAN_SCAN_INACTIVE)
1937		goto unlock;
1938
1939	if (os->os_seqno != seqno) {
1940		os->os_seqno = seqno;
1941		goto unlock;
1942	}
1943
1944	for (i = 0; i < osb->max_slots; i++)
1945		ocfs2_queue_recovery_completion(osb->journal, i, NULL, NULL,
1946						NULL, ORPHAN_NO_NEED_TRUNCATE);
1947	/*
1948	 * We queued a recovery on orphan slots, increment the sequence
1949	 * number and update LVB so other node will skip the scan for a while
1950	 */
1951	seqno++;
1952	os->os_count++;
1953	os->os_scantime = CURRENT_TIME;
1954unlock:
1955	ocfs2_orphan_scan_unlock(osb, seqno);
1956out:
1957	trace_ocfs2_queue_orphan_scan_end(os->os_count, os->os_seqno,
1958					  atomic_read(&os->os_state));
1959	return;
1960}
1961
1962/* Worker task that gets fired every ORPHAN_SCAN_SCHEDULE_TIMEOUT millsec */
1963static void ocfs2_orphan_scan_work(struct work_struct *work)
1964{
1965	struct ocfs2_orphan_scan *os;
1966	struct ocfs2_super *osb;
1967
1968	os = container_of(work, struct ocfs2_orphan_scan,
1969			  os_orphan_scan_work.work);
1970	osb = os->os_osb;
1971
1972	mutex_lock(&os->os_lock);
1973	ocfs2_queue_orphan_scan(osb);
1974	if (atomic_read(&os->os_state) == ORPHAN_SCAN_ACTIVE)
1975		queue_delayed_work(ocfs2_wq, &os->os_orphan_scan_work,
1976				      ocfs2_orphan_scan_timeout());
1977	mutex_unlock(&os->os_lock);
1978}
1979
1980void ocfs2_orphan_scan_stop(struct ocfs2_super *osb)
1981{
1982	struct ocfs2_orphan_scan *os;
1983
1984	os = &osb->osb_orphan_scan;
1985	if (atomic_read(&os->os_state) == ORPHAN_SCAN_ACTIVE) {
1986		atomic_set(&os->os_state, ORPHAN_SCAN_INACTIVE);
1987		mutex_lock(&os->os_lock);
1988		cancel_delayed_work(&os->os_orphan_scan_work);
1989		mutex_unlock(&os->os_lock);
1990	}
1991}
1992
1993void ocfs2_orphan_scan_init(struct ocfs2_super *osb)
1994{
1995	struct ocfs2_orphan_scan *os;
1996
1997	os = &osb->osb_orphan_scan;
1998	os->os_osb = osb;
1999	os->os_count = 0;
2000	os->os_seqno = 0;
2001	mutex_init(&os->os_lock);
2002	INIT_DELAYED_WORK(&os->os_orphan_scan_work, ocfs2_orphan_scan_work);
2003}
2004
2005void ocfs2_orphan_scan_start(struct ocfs2_super *osb)
2006{
2007	struct ocfs2_orphan_scan *os;
2008
2009	os = &osb->osb_orphan_scan;
2010	os->os_scantime = CURRENT_TIME;
2011	if (ocfs2_is_hard_readonly(osb) || ocfs2_mount_local(osb))
2012		atomic_set(&os->os_state, ORPHAN_SCAN_INACTIVE);
2013	else {
2014		atomic_set(&os->os_state, ORPHAN_SCAN_ACTIVE);
2015		queue_delayed_work(ocfs2_wq, &os->os_orphan_scan_work,
2016				   ocfs2_orphan_scan_timeout());
2017	}
2018}
2019
2020struct ocfs2_orphan_filldir_priv {
2021	struct dir_context	ctx;
2022	struct inode		*head;
2023	struct ocfs2_super	*osb;
2024	enum ocfs2_orphan_reco_type orphan_reco_type;
2025};
2026
2027static int ocfs2_orphan_filldir(struct dir_context *ctx, const char *name,
2028				int name_len, loff_t pos, u64 ino,
2029				unsigned type)
2030{
2031	struct ocfs2_orphan_filldir_priv *p =
2032		container_of(ctx, struct ocfs2_orphan_filldir_priv, ctx);
2033	struct inode *iter;
2034
2035	if (name_len == 1 && !strncmp(".", name, 1))
2036		return 0;
2037	if (name_len == 2 && !strncmp("..", name, 2))
2038		return 0;
2039
2040	/* do not include dio entry in case of orphan scan */
2041	if ((p->orphan_reco_type == ORPHAN_NO_NEED_TRUNCATE) &&
2042			(!strncmp(name, OCFS2_DIO_ORPHAN_PREFIX,
2043			OCFS2_DIO_ORPHAN_PREFIX_LEN)))
2044		return 0;
2045
2046	/* Skip bad inodes so that recovery can continue */
2047	iter = ocfs2_iget(p->osb, ino,
2048			  OCFS2_FI_FLAG_ORPHAN_RECOVERY, 0);
2049	if (IS_ERR(iter))
2050		return 0;
2051
2052	if (!strncmp(name, OCFS2_DIO_ORPHAN_PREFIX,
2053			OCFS2_DIO_ORPHAN_PREFIX_LEN))
2054		OCFS2_I(iter)->ip_flags |= OCFS2_INODE_DIO_ORPHAN_ENTRY;
2055
2056	/* Skip inodes which are already added to recover list, since dio may
2057	 * happen concurrently with unlink/rename */
2058	if (OCFS2_I(iter)->ip_next_orphan) {
2059		iput(iter);
2060		return 0;
2061	}
2062
2063	trace_ocfs2_orphan_filldir((unsigned long long)OCFS2_I(iter)->ip_blkno);
2064	/* No locking is required for the next_orphan queue as there
2065	 * is only ever a single process doing orphan recovery. */
2066	OCFS2_I(iter)->ip_next_orphan = p->head;
2067	p->head = iter;
2068
2069	return 0;
2070}
2071
2072static int ocfs2_queue_orphans(struct ocfs2_super *osb,
2073			       int slot,
2074			       struct inode **head,
2075			       enum ocfs2_orphan_reco_type orphan_reco_type)
2076{
2077	int status;
2078	struct inode *orphan_dir_inode = NULL;
2079	struct ocfs2_orphan_filldir_priv priv = {
2080		.ctx.actor = ocfs2_orphan_filldir,
2081		.osb = osb,
2082		.head = *head,
2083		.orphan_reco_type = orphan_reco_type
2084	};
2085
2086	orphan_dir_inode = ocfs2_get_system_file_inode(osb,
2087						       ORPHAN_DIR_SYSTEM_INODE,
2088						       slot);
2089	if  (!orphan_dir_inode) {
2090		status = -ENOENT;
2091		mlog_errno(status);
2092		return status;
2093	}
2094
2095	mutex_lock(&orphan_dir_inode->i_mutex);
2096	status = ocfs2_inode_lock(orphan_dir_inode, NULL, 0);
2097	if (status < 0) {
2098		mlog_errno(status);
2099		goto out;
2100	}
2101
2102	status = ocfs2_dir_foreach(orphan_dir_inode, &priv.ctx);
2103	if (status) {
2104		mlog_errno(status);
2105		goto out_cluster;
2106	}
2107
2108	*head = priv.head;
2109
2110out_cluster:
2111	ocfs2_inode_unlock(orphan_dir_inode, 0);
2112out:
2113	mutex_unlock(&orphan_dir_inode->i_mutex);
2114	iput(orphan_dir_inode);
2115	return status;
2116}
2117
2118static int ocfs2_orphan_recovery_can_continue(struct ocfs2_super *osb,
2119					      int slot)
2120{
2121	int ret;
2122
2123	spin_lock(&osb->osb_lock);
2124	ret = !osb->osb_orphan_wipes[slot];
2125	spin_unlock(&osb->osb_lock);
2126	return ret;
2127}
2128
2129static void ocfs2_mark_recovering_orphan_dir(struct ocfs2_super *osb,
2130					     int slot)
2131{
2132	spin_lock(&osb->osb_lock);
2133	/* Mark ourselves such that new processes in delete_inode()
2134	 * know to quit early. */
2135	ocfs2_node_map_set_bit(osb, &osb->osb_recovering_orphan_dirs, slot);
2136	while (osb->osb_orphan_wipes[slot]) {
2137		/* If any processes are already in the middle of an
2138		 * orphan wipe on this dir, then we need to wait for
2139		 * them. */
2140		spin_unlock(&osb->osb_lock);
2141		wait_event_interruptible(osb->osb_wipe_event,
2142					 ocfs2_orphan_recovery_can_continue(osb, slot));
2143		spin_lock(&osb->osb_lock);
2144	}
2145	spin_unlock(&osb->osb_lock);
2146}
2147
2148static void ocfs2_clear_recovering_orphan_dir(struct ocfs2_super *osb,
2149					      int slot)
2150{
2151	ocfs2_node_map_clear_bit(osb, &osb->osb_recovering_orphan_dirs, slot);
2152}
2153
2154/*
2155 * Orphan recovery. Each mounted node has it's own orphan dir which we
2156 * must run during recovery. Our strategy here is to build a list of
2157 * the inodes in the orphan dir and iget/iput them. The VFS does
2158 * (most) of the rest of the work.
2159 *
2160 * Orphan recovery can happen at any time, not just mount so we have a
2161 * couple of extra considerations.
2162 *
2163 * - We grab as many inodes as we can under the orphan dir lock -
2164 *   doing iget() outside the orphan dir risks getting a reference on
2165 *   an invalid inode.
2166 * - We must be sure not to deadlock with other processes on the
2167 *   system wanting to run delete_inode(). This can happen when they go
2168 *   to lock the orphan dir and the orphan recovery process attempts to
2169 *   iget() inside the orphan dir lock. This can be avoided by
2170 *   advertising our state to ocfs2_delete_inode().
2171 */
2172static int ocfs2_recover_orphans(struct ocfs2_super *osb,
2173				 int slot,
2174				 enum ocfs2_orphan_reco_type orphan_reco_type)
2175{
2176	int ret = 0;
2177	struct inode *inode = NULL;
2178	struct inode *iter;
2179	struct ocfs2_inode_info *oi;
2180	struct buffer_head *di_bh = NULL;
2181	struct ocfs2_dinode *di = NULL;
2182
2183	trace_ocfs2_recover_orphans(slot);
2184
2185	ocfs2_mark_recovering_orphan_dir(osb, slot);
2186	ret = ocfs2_queue_orphans(osb, slot, &inode, orphan_reco_type);
2187	ocfs2_clear_recovering_orphan_dir(osb, slot);
2188
2189	/* Error here should be noted, but we want to continue with as
2190	 * many queued inodes as we've got. */
2191	if (ret)
2192		mlog_errno(ret);
2193
2194	while (inode) {
2195		oi = OCFS2_I(inode);
2196		trace_ocfs2_recover_orphans_iput(
2197					(unsigned long long)oi->ip_blkno);
2198
2199		iter = oi->ip_next_orphan;
2200		oi->ip_next_orphan = NULL;
2201
2202		if (oi->ip_flags & OCFS2_INODE_DIO_ORPHAN_ENTRY) {
2203			mutex_lock(&inode->i_mutex);
2204			ret = ocfs2_rw_lock(inode, 1);
2205			if (ret < 0) {
2206				mlog_errno(ret);
2207				goto unlock_mutex;
2208			}
2209			/*
2210			 * We need to take and drop the inode lock to
2211			 * force read inode from disk.
2212			 */
2213			ret = ocfs2_inode_lock(inode, &di_bh, 1);
2214			if (ret) {
2215				mlog_errno(ret);
2216				goto unlock_rw;
2217			}
2218
2219			di = (struct ocfs2_dinode *)di_bh->b_data;
2220
2221			if (di->i_flags & cpu_to_le32(OCFS2_DIO_ORPHANED_FL)) {
2222				ret = ocfs2_truncate_file(inode, di_bh,
2223						i_size_read(inode));
2224				if (ret < 0) {
2225					if (ret != -ENOSPC)
2226						mlog_errno(ret);
2227					goto unlock_inode;
2228				}
2229
2230				ret = ocfs2_del_inode_from_orphan(osb, inode,
2231						di_bh, 0, 0);
2232				if (ret)
2233					mlog_errno(ret);
2234			}
2235unlock_inode:
2236			ocfs2_inode_unlock(inode, 1);
2237			brelse(di_bh);
2238			di_bh = NULL;
2239unlock_rw:
2240			ocfs2_rw_unlock(inode, 1);
2241unlock_mutex:
2242			mutex_unlock(&inode->i_mutex);
2243
2244			/* clear dio flag in ocfs2_inode_info */
2245			oi->ip_flags &= ~OCFS2_INODE_DIO_ORPHAN_ENTRY;
2246		} else {
2247			spin_lock(&oi->ip_lock);
2248			/* Set the proper information to get us going into
2249			 * ocfs2_delete_inode. */
2250			oi->ip_flags |= OCFS2_INODE_MAYBE_ORPHANED;
2251			spin_unlock(&oi->ip_lock);
2252		}
2253
2254		iput(inode);
2255		inode = iter;
2256	}
2257
2258	return ret;
2259}
2260
2261static int __ocfs2_wait_on_mount(struct ocfs2_super *osb, int quota)
2262{
2263	/* This check is good because ocfs2 will wait on our recovery
2264	 * thread before changing it to something other than MOUNTED
2265	 * or DISABLED. */
2266	wait_event(osb->osb_mount_event,
2267		  (!quota && atomic_read(&osb->vol_state) == VOLUME_MOUNTED) ||
2268		   atomic_read(&osb->vol_state) == VOLUME_MOUNTED_QUOTAS ||
2269		   atomic_read(&osb->vol_state) == VOLUME_DISABLED);
2270
2271	/* If there's an error on mount, then we may never get to the
2272	 * MOUNTED flag, but this is set right before
2273	 * dismount_volume() so we can trust it. */
2274	if (atomic_read(&osb->vol_state) == VOLUME_DISABLED) {
2275		trace_ocfs2_wait_on_mount(VOLUME_DISABLED);
2276		mlog(0, "mount error, exiting!\n");
2277		return -EBUSY;
2278	}
2279
2280	return 0;
2281}
2282
2283static int ocfs2_commit_thread(void *arg)
2284{
2285	int status;
2286	struct ocfs2_super *osb = arg;
2287	struct ocfs2_journal *journal = osb->journal;
2288
2289	/* we can trust j_num_trans here because _should_stop() is only set in
2290	 * shutdown and nobody other than ourselves should be able to start
2291	 * transactions.  committing on shutdown might take a few iterations
2292	 * as final transactions put deleted inodes on the list */
2293	while (!(kthread_should_stop() &&
2294		 atomic_read(&journal->j_num_trans) == 0)) {
2295
2296		wait_event_interruptible(osb->checkpoint_event,
2297					 atomic_read(&journal->j_num_trans)
2298					 || kthread_should_stop());
2299
2300		status = ocfs2_commit_cache(osb);
2301		if (status < 0) {
2302			static unsigned long abort_warn_time;
2303
2304			/* Warn about this once per minute */
2305			if (printk_timed_ratelimit(&abort_warn_time, 60*HZ))
2306				mlog(ML_ERROR, "status = %d, journal is "
2307						"already aborted.\n", status);
2308			/*
2309			 * After ocfs2_commit_cache() fails, j_num_trans has a
2310			 * non-zero value.  Sleep here to avoid a busy-wait
2311			 * loop.
2312			 */
2313			msleep_interruptible(1000);
2314		}
2315
2316		if (kthread_should_stop() && atomic_read(&journal->j_num_trans)){
2317			mlog(ML_KTHREAD,
2318			     "commit_thread: %u transactions pending on "
2319			     "shutdown\n",
2320			     atomic_read(&journal->j_num_trans));
2321		}
2322	}
2323
2324	return 0;
2325}
2326
2327/* Reads all the journal inodes without taking any cluster locks. Used
2328 * for hard readonly access to determine whether any journal requires
2329 * recovery. Also used to refresh the recovery generation numbers after
2330 * a journal has been recovered by another node.
2331 */
2332int ocfs2_check_journals_nolocks(struct ocfs2_super *osb)
2333{
2334	int ret = 0;
2335	unsigned int slot;
2336	struct buffer_head *di_bh = NULL;
2337	struct ocfs2_dinode *di;
2338	int journal_dirty = 0;
2339
2340	for(slot = 0; slot < osb->max_slots; slot++) {
2341		ret = ocfs2_read_journal_inode(osb, slot, &di_bh, NULL);
2342		if (ret) {
2343			mlog_errno(ret);
2344			goto out;
2345		}
2346
2347		di = (struct ocfs2_dinode *) di_bh->b_data;
2348
2349		osb->slot_recovery_generations[slot] =
2350					ocfs2_get_recovery_generation(di);
2351
2352		if (le32_to_cpu(di->id1.journal1.ij_flags) &
2353		    OCFS2_JOURNAL_DIRTY_FL)
2354			journal_dirty = 1;
2355
2356		brelse(di_bh);
2357		di_bh = NULL;
2358	}
2359
2360out:
2361	if (journal_dirty)
2362		ret = -EROFS;
2363	return ret;
2364}
2365