summaryrefslogtreecommitdiff
path: root/fs/bcachefs/journal.c
blob: a348e360c4fd4d80b0380b77a1bcd8e361230bf8 (plain)
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
389
390
391
392
393
394
395
396
397
398
399
400
401
402
403
404
405
406
407
408
409
410
411
412
413
414
415
416
417
418
419
420
421
422
423
424
425
426
427
428
429
430
431
432
433
434
435
436
437
438
439
440
441
442
443
444
445
446
447
448
449
450
451
452
453
454
455
456
457
458
459
460
461
462
463
464
465
466
467
468
469
470
471
472
473
474
475
476
477
478
479
480
481
482
483
484
485
486
487
488
489
490
491
492
493
494
495
496
497
498
499
500
501
502
503
504
505
506
507
508
509
510
511
512
513
514
515
516
517
518
519
520
521
522
523
524
525
526
527
528
529
530
531
532
533
534
535
536
537
538
539
540
541
542
543
544
545
546
547
548
549
550
551
552
553
554
555
556
557
558
559
560
561
562
563
564
565
566
567
568
569
570
571
572
573
574
575
576
577
578
579
580
581
582
583
584
585
586
587
588
589
590
591
592
593
594
595
596
597
598
599
600
601
602
603
604
605
606
607
608
609
610
611
612
613
614
615
616
617
618
619
620
621
622
623
624
625
626
627
628
629
630
631
632
633
634
635
636
637
638
639
640
641
642
643
644
645
646
647
648
649
650
651
652
653
654
655
656
657
658
659
660
661
662
663
664
665
666
667
668
669
670
671
672
673
674
675
676
677
678
679
680
681
682
683
684
685
686
687
688
689
690
691
692
693
694
695
696
697
698
699
700
701
702
703
704
705
706
707
708
709
710
711
712
713
714
715
716
717
718
719
720
721
722
723
724
725
726
727
728
729
730
731
732
733
734
735
736
737
738
739
740
741
742
743
744
745
746
747
748
749
750
751
752
753
754
755
756
757
758
759
760
761
762
763
764
765
766
767
768
769
770
771
772
773
774
775
776
777
778
779
780
781
782
783
784
785
786
787
788
789
790
791
792
793
794
795
796
797
798
799
800
801
802
803
804
805
806
807
808
809
810
811
812
813
814
815
816
817
818
819
820
821
822
823
824
825
826
827
828
829
830
831
832
833
834
835
836
837
838
839
840
841
842
843
844
845
846
847
848
849
850
851
852
853
854
855
856
857
858
859
860
861
862
863
864
865
866
867
868
869
870
871
872
873
874
875
876
877
878
879
880
881
882
883
884
885
886
887
888
889
890
891
892
893
894
895
896
897
898
899
900
901
902
903
904
905
906
907
908
909
910
911
912
913
914
915
916
917
918
919
920
921
922
923
924
925
926
927
928
929
930
931
932
933
934
935
936
937
938
939
940
941
942
943
944
945
946
947
948
949
950
951
952
953
954
955
956
957
958
959
960
961
962
963
964
965
966
967
968
969
970
971
972
973
974
975
976
977
978
979
980
981
982
983
984
985
986
987
988
989
990
991
992
993
994
995
996
997
998
999
1000
1001
1002
1003
1004
1005
1006
1007
1008
1009
1010
1011
1012
1013
1014
1015
1016
1017
1018
1019
1020
1021
1022
1023
1024
1025
1026
1027
1028
1029
1030
1031
1032
1033
1034
1035
1036
1037
1038
1039
1040
1041
1042
1043
1044
1045
1046
1047
1048
1049
1050
1051
1052
1053
1054
1055
1056
1057
1058
1059
1060
1061
1062
1063
1064
1065
1066
1067
1068
1069
1070
1071
1072
1073
1074
1075
1076
1077
1078
1079
1080
1081
1082
1083
1084
1085
1086
1087
1088
1089
1090
1091
1092
1093
1094
1095
1096
1097
1098
1099
1100
1101
1102
1103
1104
1105
1106
1107
1108
1109
1110
1111
1112
1113
1114
1115
1116
1117
1118
1119
1120
1121
1122
1123
1124
1125
1126
1127
1128
1129
1130
1131
1132
1133
1134
1135
1136
1137
1138
1139
1140
1141
1142
1143
1144
1145
1146
1147
1148
1149
1150
1151
1152
1153
1154
1155
1156
1157
1158
1159
1160
1161
1162
1163
1164
1165
1166
1167
1168
1169
1170
1171
1172
1173
1174
1175
1176
1177
1178
1179
1180
1181
1182
1183
1184
1185
1186
1187
1188
1189
1190
1191
1192
1193
1194
1195
1196
1197
1198
1199
1200
1201
1202
1203
1204
1205
1206
1207
1208
1209
1210
1211
1212
1213
1214
1215
1216
1217
1218
1219
1220
1221
1222
1223
1224
1225
1226
1227
1228
1229
1230
1231
1232
1233
1234
1235
1236
1237
1238
1239
1240
1241
1242
1243
1244
1245
1246
1247
1248
1249
1250
1251
1252
1253
1254
1255
1256
1257
// SPDX-License-Identifier: GPL-2.0
/*
 * bcachefs journalling code, for btree insertions
 *
 * Copyright 2012 Google, Inc.
 */

#include "bcachefs.h"
#include "alloc_foreground.h"
#include "bkey_methods.h"
#include "btree_gc.h"
#include "buckets.h"
#include "journal.h"
#include "journal_io.h"
#include "journal_reclaim.h"
#include "journal_seq_blacklist.h"
#include "super-io.h"

#include <trace/events/bcachefs.h>

static bool __journal_entry_is_open(union journal_res_state state)
{
	return state.cur_entry_offset < JOURNAL_ENTRY_CLOSED_VAL;
}

static bool journal_entry_is_open(struct journal *j)
{
	return __journal_entry_is_open(j->reservations);
}

static void journal_pin_new_entry(struct journal *j, int count)
{
	struct journal_entry_pin_list *p;

	/*
	 * The fifo_push() needs to happen at the same time as j->seq is
	 * incremented for journal_last_seq() to be calculated correctly
	 */
	atomic64_inc(&j->seq);
	p = fifo_push_ref(&j->pin);

	INIT_LIST_HEAD(&p->list);
	INIT_LIST_HEAD(&p->flushed);
	atomic_set(&p->count, count);
	p->devs.nr = 0;
}

static void bch2_journal_buf_init(struct journal *j)
{
	struct journal_buf *buf = journal_cur_buf(j);

	memset(buf->has_inode, 0, sizeof(buf->has_inode));

	memset(buf->data, 0, sizeof(*buf->data));
	buf->data->seq	= cpu_to_le64(journal_cur_seq(j));
	buf->data->u64s	= 0;
}

void bch2_journal_halt(struct journal *j)
{
	union journal_res_state old, new;
	u64 v = atomic64_read(&j->reservations.counter);

	do {
		old.v = new.v = v;
		if (old.cur_entry_offset == JOURNAL_ENTRY_ERROR_VAL)
			return;

		new.cur_entry_offset = JOURNAL_ENTRY_ERROR_VAL;
	} while ((v = atomic64_cmpxchg(&j->reservations.counter,
				       old.v, new.v)) != old.v);

	journal_wake(j);
	closure_wake_up(&journal_cur_buf(j)->wait);
}

/* journal entry close/open: */

void __bch2_journal_buf_put(struct journal *j, bool need_write_just_set)
{
	if (!need_write_just_set &&
	    test_bit(JOURNAL_NEED_WRITE, &j->flags))
		bch2_time_stats_update(j->delay_time,
				       j->need_write_time);

	clear_bit(JOURNAL_NEED_WRITE, &j->flags);

	closure_call(&j->io, bch2_journal_write, system_highpri_wq, NULL);
}

/*
 * Returns true if journal entry is now closed:
 */
static bool __journal_entry_close(struct journal *j)
{
	struct bch_fs *c = container_of(j, struct bch_fs, journal);
	struct journal_buf *buf = journal_cur_buf(j);
	union journal_res_state old, new;
	u64 v = atomic64_read(&j->reservations.counter);
	bool set_need_write = false;
	unsigned sectors;

	lockdep_assert_held(&j->lock);

	do {
		old.v = new.v = v;
		if (old.cur_entry_offset == JOURNAL_ENTRY_CLOSED_VAL)
			return true;

		if (old.cur_entry_offset == JOURNAL_ENTRY_ERROR_VAL) {
			/* this entry will never be written: */
			closure_wake_up(&buf->wait);
			return true;
		}

		if (!test_bit(JOURNAL_NEED_WRITE, &j->flags)) {
			set_bit(JOURNAL_NEED_WRITE, &j->flags);
			j->need_write_time = local_clock();
			set_need_write = true;
		}

		if (new.prev_buf_unwritten)
			return false;

		new.cur_entry_offset = JOURNAL_ENTRY_CLOSED_VAL;
		new.idx++;
		new.prev_buf_unwritten = 1;

		BUG_ON(journal_state_count(new, new.idx));
	} while ((v = atomic64_cmpxchg(&j->reservations.counter,
				       old.v, new.v)) != old.v);

	buf->data->u64s		= cpu_to_le32(old.cur_entry_offset);

	sectors = vstruct_blocks_plus(buf->data, c->block_bits,
				      buf->u64s_reserved) << c->block_bits;
	BUG_ON(sectors > buf->sectors);
	buf->sectors = sectors;

	bkey_extent_init(&buf->key);

	/*
	 * We have to set last_seq here, _before_ opening a new journal entry:
	 *
	 * A threads may replace an old pin with a new pin on their current
	 * journal reservation - the expectation being that the journal will
	 * contain either what the old pin protected or what the new pin
	 * protects.
	 *
	 * After the old pin is dropped journal_last_seq() won't include the old
	 * pin, so we can only write the updated last_seq on the entry that
	 * contains whatever the new pin protects.
	 *
	 * Restated, we can _not_ update last_seq for a given entry if there
	 * could be a newer entry open with reservations/pins that have been
	 * taken against it.
	 *
	 * Hence, we want update/set last_seq on the current journal entry right
	 * before we open a new one:
	 */
	buf->data->last_seq	= cpu_to_le64(journal_last_seq(j));

	if (journal_entry_empty(buf->data))
		clear_bit(JOURNAL_NOT_EMPTY, &j->flags);
	else
		set_bit(JOURNAL_NOT_EMPTY, &j->flags);

	journal_pin_new_entry(j, 1);

	bch2_journal_buf_init(j);

	cancel_delayed_work(&j->write_work);

	bch2_journal_space_available(j);

	bch2_journal_buf_put(j, old.idx, set_need_write);
	return true;
}

static bool journal_entry_close(struct journal *j)
{
	bool ret;

	spin_lock(&j->lock);
	ret = __journal_entry_close(j);
	spin_unlock(&j->lock);

	return ret;
}

/*
 * should _only_ called from journal_res_get() - when we actually want a
 * journal reservation - journal entry is open means journal is dirty:
 *
 * returns:
 * 0:		success
 * -ENOSPC:	journal currently full, must invoke reclaim
 * -EAGAIN:	journal blocked, must wait
 * -EROFS:	insufficient rw devices or journal error
 */
static int journal_entry_open(struct journal *j)
{
	struct journal_buf *buf = journal_cur_buf(j);
	union journal_res_state old, new;
	int u64s;
	u64 v;

	lockdep_assert_held(&j->lock);
	BUG_ON(journal_entry_is_open(j));

	if (j->blocked)
		return -EAGAIN;

	if (j->cur_entry_error)
		return j->cur_entry_error;

	BUG_ON(!j->cur_entry_sectors);

	buf->u64s_reserved	= j->entry_u64s_reserved;
	buf->disk_sectors	= j->cur_entry_sectors;
	buf->sectors		= min(buf->disk_sectors, buf->buf_size >> 9);

	u64s = (int) (buf->sectors << 9) / sizeof(u64) -
		journal_entry_overhead(j);
	u64s  = clamp_t(int, u64s, 0, JOURNAL_ENTRY_CLOSED_VAL - 1);

	if (u64s <= le32_to_cpu(buf->data->u64s))
		return -ENOSPC;

	/*
	 * Must be set before marking the journal entry as open:
	 */
	j->cur_entry_u64s = u64s;

	v = atomic64_read(&j->reservations.counter);
	do {
		old.v = new.v = v;

		if (old.cur_entry_offset == JOURNAL_ENTRY_ERROR_VAL)
			return -EROFS;

		/* Handle any already added entries */
		new.cur_entry_offset = le32_to_cpu(buf->data->u64s);

		EBUG_ON(journal_state_count(new, new.idx));
		journal_state_inc(&new);
	} while ((v = atomic64_cmpxchg(&j->reservations.counter,
				       old.v, new.v)) != old.v);

	if (j->res_get_blocked_start)
		bch2_time_stats_update(j->blocked_time,
				       j->res_get_blocked_start);
	j->res_get_blocked_start = 0;

	mod_delayed_work(system_freezable_wq,
			 &j->write_work,
			 msecs_to_jiffies(j->write_delay_ms));
	journal_wake(j);
	return 0;
}

static bool journal_quiesced(struct journal *j)
{
	union journal_res_state state = READ_ONCE(j->reservations);
	bool ret = !state.prev_buf_unwritten && !__journal_entry_is_open(state);

	if (!ret)
		journal_entry_close(j);
	return ret;
}

static void journal_quiesce(struct journal *j)
{
	wait_event(j->wait, journal_quiesced(j));
}

static void journal_write_work(struct work_struct *work)
{
	struct journal *j = container_of(work, struct journal, write_work.work);

	journal_entry_close(j);
}

/*
 * Given an inode number, if that inode number has data in the journal that
 * hasn't yet been flushed, return the journal sequence number that needs to be
 * flushed:
 */
u64 bch2_inode_journal_seq(struct journal *j, u64 inode)
{
	size_t h = hash_64(inode, ilog2(sizeof(j->buf[0].has_inode) * 8));
	u64 seq = 0;

	if (!test_bit(h, j->buf[0].has_inode) &&
	    !test_bit(h, j->buf[1].has_inode))
		return 0;

	spin_lock(&j->lock);
	if (test_bit(h, journal_cur_buf(j)->has_inode))
		seq = journal_cur_seq(j);
	else if (test_bit(h, journal_prev_buf(j)->has_inode))
		seq = journal_cur_seq(j) - 1;
	spin_unlock(&j->lock);

	return seq;
}

static int __journal_res_get(struct journal *j, struct journal_res *res,
			     unsigned flags)
{
	struct bch_fs *c = container_of(j, struct bch_fs, journal);
	struct journal_buf *buf;
	bool can_discard;
	int ret;
retry:
	if (journal_res_get_fast(j, res, flags))
		return 0;

	if (bch2_journal_error(j))
		return -EROFS;

	spin_lock(&j->lock);

	/*
	 * Recheck after taking the lock, so we don't race with another thread
	 * that just did journal_entry_open() and call journal_entry_close()
	 * unnecessarily
	 */
	if (journal_res_get_fast(j, res, flags)) {
		spin_unlock(&j->lock);
		return 0;
	}

	if (!(flags & JOURNAL_RES_GET_RESERVED) &&
	    !test_bit(JOURNAL_MAY_GET_UNRESERVED, &j->flags)) {
		/*
		 * Don't want to close current journal entry, just need to
		 * invoke reclaim:
		 */
		ret = -ENOSPC;
		goto unlock;
	}

	/*
	 * If we couldn't get a reservation because the current buf filled up,
	 * and we had room for a bigger entry on disk, signal that we want to
	 * realloc the journal bufs:
	 */
	buf = journal_cur_buf(j);
	if (journal_entry_is_open(j) &&
	    buf->buf_size >> 9 < buf->disk_sectors &&
	    buf->buf_size < JOURNAL_ENTRY_SIZE_MAX)
		j->buf_size_want = max(j->buf_size_want, buf->buf_size << 1);

	if (journal_entry_is_open(j) &&
	    !__journal_entry_close(j)) {
		/*
		 * We failed to get a reservation on the current open journal
		 * entry because it's full, and we can't close it because
		 * there's still a previous one in flight:
		 */
		trace_journal_entry_full(c);
		ret = -EAGAIN;
	} else {
		ret = journal_entry_open(j);
	}
unlock:
	if ((ret == -EAGAIN || ret == -ENOSPC) &&
	    !j->res_get_blocked_start)
		j->res_get_blocked_start = local_clock() ?: 1;

	can_discard = j->can_discard;
	spin_unlock(&j->lock);

	if (!ret)
		goto retry;

	if (ret == -ENOSPC) {
		BUG_ON(!can_discard && (flags & JOURNAL_RES_GET_RESERVED));

		/*
		 * Journal is full - can't rely on reclaim from work item due to
		 * freezing:
		 */
		trace_journal_full(c);

		if (!(flags & JOURNAL_RES_GET_NONBLOCK)) {
			if (can_discard) {
				bch2_journal_do_discards(j);
				goto retry;
			}

			if (mutex_trylock(&j->reclaim_lock)) {
				bch2_journal_reclaim(j);
				mutex_unlock(&j->reclaim_lock);
			}
		}

		ret = -EAGAIN;
	}

	return ret;
}

/*
 * Essentially the entry function to the journaling code. When bcachefs is doing
 * a btree insert, it calls this function to get the current journal write.
 * Journal write is the structure used set up journal writes. The calling
 * function will then add its keys to the structure, queuing them for the next
 * write.
 *
 * To ensure forward progress, the current task must not be holding any
 * btree node write locks.
 */
int bch2_journal_res_get_slowpath(struct journal *j, struct journal_res *res,
				  unsigned flags)
{
	int ret;

	closure_wait_event(&j->async_wait,
		   (ret = __journal_res_get(j, res, flags)) != -EAGAIN ||
		   (flags & JOURNAL_RES_GET_NONBLOCK));
	return ret;
}

/* journal_preres: */

static bool journal_preres_available(struct journal *j,
				     struct journal_preres *res,
				     unsigned new_u64s)
{
	bool ret = bch2_journal_preres_get_fast(j, res, new_u64s);

	if (!ret)
		bch2_journal_reclaim_work(&j->reclaim_work.work);

	return ret;
}

int __bch2_journal_preres_get(struct journal *j,
			      struct journal_preres *res,
			      unsigned new_u64s)
{
	int ret;

	closure_wait_event(&j->preres_wait,
		   (ret = bch2_journal_error(j)) ||
		   journal_preres_available(j, res, new_u64s));
	return ret;
}

/* journal_entry_res: */

void bch2_journal_entry_res_resize(struct journal *j,
				   struct journal_entry_res *res,
				   unsigned new_u64s)
{
	union journal_res_state state;
	int d = new_u64s - res->u64s;

	spin_lock(&j->lock);

	j->entry_u64s_reserved += d;
	if (d <= 0)
		goto out;

	j->cur_entry_u64s = max_t(int, 0, j->cur_entry_u64s - d);
	smp_mb();
	state = READ_ONCE(j->reservations);

	if (state.cur_entry_offset < JOURNAL_ENTRY_CLOSED_VAL &&
	    state.cur_entry_offset > j->cur_entry_u64s) {
		j->cur_entry_u64s += d;
		/*
		 * Not enough room in current journal entry, have to flush it:
		 */
		__journal_entry_close(j);
	} else {
		journal_cur_buf(j)->u64s_reserved += d;
	}
out:
	spin_unlock(&j->lock);
	res->u64s += d;
}

/* journal flushing: */

u64 bch2_journal_last_unwritten_seq(struct journal *j)
{
	u64 seq;

	spin_lock(&j->lock);
	seq = journal_cur_seq(j);
	if (j->reservations.prev_buf_unwritten)
		seq--;
	spin_unlock(&j->lock);

	return seq;
}

/**
 * bch2_journal_open_seq_async - try to open a new journal entry if @seq isn't
 * open yet, or wait if we cannot
 *
 * used by the btree interior update machinery, when it needs to write a new
 * btree root - every journal entry contains the roots of all the btrees, so it
 * doesn't need to bother with getting a journal reservation
 */
int bch2_journal_open_seq_async(struct journal *j, u64 seq, struct closure *cl)
{
	struct bch_fs *c = container_of(j, struct bch_fs, journal);
	int ret;

	spin_lock(&j->lock);

	/*
	 * Can't try to open more than one sequence number ahead:
	 */
	BUG_ON(journal_cur_seq(j) < seq && !journal_entry_is_open(j));

	if (journal_cur_seq(j) > seq ||
	    journal_entry_is_open(j)) {
		spin_unlock(&j->lock);
		return 0;
	}

	if (journal_cur_seq(j) < seq &&
	    !__journal_entry_close(j)) {
		/* haven't finished writing out the previous one: */
		trace_journal_entry_full(c);
		ret = -EAGAIN;
	} else {
		BUG_ON(journal_cur_seq(j) != seq);

		ret = journal_entry_open(j);
	}

	if ((ret == -EAGAIN || ret == -ENOSPC) &&
	    !j->res_get_blocked_start)
		j->res_get_blocked_start = local_clock() ?: 1;

	if (ret == -EAGAIN || ret == -ENOSPC)
		closure_wait(&j->async_wait, cl);

	spin_unlock(&j->lock);

	if (ret == -ENOSPC) {
		trace_journal_full(c);
		bch2_journal_reclaim_work(&j->reclaim_work.work);
		ret = -EAGAIN;
	}

	return ret;
}

static int journal_seq_error(struct journal *j, u64 seq)
{
	union journal_res_state state = READ_ONCE(j->reservations);

	if (seq == journal_cur_seq(j))
		return bch2_journal_error(j);

	if (seq + 1 == journal_cur_seq(j) &&
	    !state.prev_buf_unwritten &&
	    seq > j->seq_ondisk)
		return -EIO;

	return 0;
}

static inline struct journal_buf *
journal_seq_to_buf(struct journal *j, u64 seq)
{
	/* seq should be for a journal entry that has been opened: */
	BUG_ON(seq > journal_cur_seq(j));
	BUG_ON(seq == journal_cur_seq(j) &&
	       j->reservations.cur_entry_offset == JOURNAL_ENTRY_CLOSED_VAL);

	if (seq == journal_cur_seq(j))
		return journal_cur_buf(j);
	if (seq + 1 == journal_cur_seq(j) &&
	    j->reservations.prev_buf_unwritten)
		return journal_prev_buf(j);
	return NULL;
}

/**
 * bch2_journal_wait_on_seq - wait for a journal entry to be written
 *
 * does _not_ cause @seq to be written immediately - if there is no other
 * activity to cause the relevant journal entry to be filled up or flushed it
 * can wait for an arbitrary amount of time (up to @j->write_delay_ms, which is
 * configurable).
 */
void bch2_journal_wait_on_seq(struct journal *j, u64 seq,
			      struct closure *parent)
{
	struct journal_buf *buf;

	spin_lock(&j->lock);

	if ((buf = journal_seq_to_buf(j, seq))) {
		if (!closure_wait(&buf->wait, parent))
			BUG();

		if (seq == journal_cur_seq(j)) {
			smp_mb();
			if (bch2_journal_error(j))
				closure_wake_up(&buf->wait);
		}
	}

	spin_unlock(&j->lock);
}

/**
 * bch2_journal_flush_seq_async - wait for a journal entry to be written
 *
 * like bch2_journal_wait_on_seq, except that it triggers a write immediately if
 * necessary
 */
void bch2_journal_flush_seq_async(struct journal *j, u64 seq,
				  struct closure *parent)
{
	struct journal_buf *buf;

	spin_lock(&j->lock);

	if (parent &&
	    (buf = journal_seq_to_buf(j, seq)))
		if (!closure_wait(&buf->wait, parent))
			BUG();

	if (seq == journal_cur_seq(j))
		__journal_entry_close(j);
	spin_unlock(&j->lock);
}

static int journal_seq_flushed(struct journal *j, u64 seq)
{
	int ret;

	spin_lock(&j->lock);
	ret = seq <= j->seq_ondisk ? 1 : journal_seq_error(j, seq);

	if (seq == journal_cur_seq(j))
		__journal_entry_close(j);
	spin_unlock(&j->lock);

	return ret;
}

int bch2_journal_flush_seq(struct journal *j, u64 seq)
{
	u64 start_time = local_clock();
	int ret, ret2;

	ret = wait_event_killable(j->wait, (ret2 = journal_seq_flushed(j, seq)));

	bch2_time_stats_update(j->flush_seq_time, start_time);

	return ret ?: ret2 < 0 ? ret2 : 0;
}

/**
 * bch2_journal_meta_async - force a journal entry to be written
 */
void bch2_journal_meta_async(struct journal *j, struct closure *parent)
{
	struct journal_res res;

	memset(&res, 0, sizeof(res));

	bch2_journal_res_get(j, &res, jset_u64s(0), 0);
	bch2_journal_res_put(j, &res);

	bch2_journal_flush_seq_async(j, res.seq, parent);
}

int bch2_journal_meta(struct journal *j)
{
	struct journal_res res;
	int ret;

	memset(&res, 0, sizeof(res));

	ret = bch2_journal_res_get(j, &res, jset_u64s(0), 0);
	if (ret)
		return ret;

	bch2_journal_res_put(j, &res);

	return bch2_journal_flush_seq(j, res.seq);
}

/*
 * bch2_journal_flush_async - if there is an open journal entry, or a journal
 * still being written, write it and wait for the write to complete
 */
void bch2_journal_flush_async(struct journal *j, struct closure *parent)
{
	u64 seq, journal_seq;

	spin_lock(&j->lock);
	journal_seq = journal_cur_seq(j);

	if (journal_entry_is_open(j)) {
		seq = journal_seq;
	} else if (journal_seq) {
		seq = journal_seq - 1;
	} else {
		spin_unlock(&j->lock);
		return;
	}
	spin_unlock(&j->lock);

	bch2_journal_flush_seq_async(j, seq, parent);
}

int bch2_journal_flush(struct journal *j)
{
	u64 seq, journal_seq;

	spin_lock(&j->lock);
	journal_seq = journal_cur_seq(j);

	if (journal_entry_is_open(j)) {
		seq = journal_seq;
	} else if (journal_seq) {
		seq = journal_seq - 1;
	} else {
		spin_unlock(&j->lock);
		return 0;
	}
	spin_unlock(&j->lock);

	return bch2_journal_flush_seq(j, seq);
}

/* block/unlock the journal: */

void bch2_journal_unblock(struct journal *j)
{
	spin_lock(&j->lock);
	j->blocked--;
	spin_unlock(&j->lock);

	journal_wake(j);
}

void bch2_journal_block(struct journal *j)
{
	spin_lock(&j->lock);
	j->blocked++;
	spin_unlock(&j->lock);

	journal_quiesce(j);
}

/* allocate journal on a device: */

static int __bch2_set_nr_journal_buckets(struct bch_dev *ca, unsigned nr,
					 bool new_fs, struct closure *cl)
{
	struct bch_fs *c = ca->fs;
	struct journal_device *ja = &ca->journal;
	struct bch_sb_field_journal *journal_buckets;
	u64 *new_bucket_seq = NULL, *new_buckets = NULL;
	int ret = 0;

	/* don't handle reducing nr of buckets yet: */
	if (nr <= ja->nr)
		return 0;

	ret = -ENOMEM;
	new_buckets	= kzalloc(nr * sizeof(u64), GFP_KERNEL);
	new_bucket_seq	= kzalloc(nr * sizeof(u64), GFP_KERNEL);
	if (!new_buckets || !new_bucket_seq)
		goto err;

	journal_buckets = bch2_sb_resize_journal(&ca->disk_sb,
						 nr + sizeof(*journal_buckets) / sizeof(u64));
	if (!journal_buckets)
		goto err;

	/*
	 * We may be called from the device add path, before the new device has
	 * actually been added to the running filesystem:
	 */
	if (c)
		spin_lock(&c->journal.lock);

	memcpy(new_buckets,	ja->buckets,	ja->nr * sizeof(u64));
	memcpy(new_bucket_seq,	ja->bucket_seq,	ja->nr * sizeof(u64));
	swap(new_buckets,	ja->buckets);
	swap(new_bucket_seq,	ja->bucket_seq);

	if (c)
		spin_unlock(&c->journal.lock);

	while (ja->nr < nr) {
		struct open_bucket *ob = NULL;
		unsigned pos;
		long bucket;

		if (new_fs) {
			bucket = bch2_bucket_alloc_new_fs(ca);
			if (bucket < 0) {
				ret = -ENOSPC;
				goto err;
			}
		} else {
			ob = bch2_bucket_alloc(c, ca, RESERVE_ALLOC,
					       false, cl);
			if (IS_ERR(ob)) {
				ret = cl ? -EAGAIN : -ENOSPC;
				goto err;
			}

			bucket = sector_to_bucket(ca, ob->ptr.offset);
		}

		if (c) {
			percpu_down_read_preempt_disable(&c->mark_lock);
			spin_lock(&c->journal.lock);
		} else {
			preempt_disable();
		}

		pos = ja->nr ? (ja->cur_idx + 1) % ja->nr : 0;
		__array_insert_item(ja->buckets,		ja->nr, pos);
		__array_insert_item(ja->bucket_seq,		ja->nr, pos);
		__array_insert_item(journal_buckets->buckets,	ja->nr, pos);
		ja->nr++;

		ja->buckets[pos] = bucket;
		ja->bucket_seq[pos] = 0;
		journal_buckets->buckets[pos] = cpu_to_le64(bucket);

		if (pos <= ja->discard_idx)
			ja->discard_idx = (ja->discard_idx + 1) % ja->nr;
		if (pos <= ja->dirty_idx_ondisk)
			ja->dirty_idx_ondisk = (ja->dirty_idx_ondisk + 1) % ja->nr;
		if (pos <= ja->dirty_idx)
			ja->dirty_idx = (ja->dirty_idx + 1) % ja->nr;
		if (pos <= ja->cur_idx)
			ja->cur_idx = (ja->cur_idx + 1) % ja->nr;

		bch2_mark_metadata_bucket(c, ca, bucket, BCH_DATA_JOURNAL,
					  ca->mi.bucket_size,
					  gc_phase(GC_PHASE_SB),
					  0);

		if (c) {
			spin_unlock(&c->journal.lock);
			percpu_up_read_preempt_enable(&c->mark_lock);
		} else {
			preempt_enable();
		}

		if (!new_fs)
			bch2_open_bucket_put(c, ob);
	}

	ret = 0;
err:
	kfree(new_bucket_seq);
	kfree(new_buckets);

	return ret;
}

/*
 * Allocate more journal space at runtime - not currently making use if it, but
 * the code works:
 */
int bch2_set_nr_journal_buckets(struct bch_fs *c, struct bch_dev *ca,
				unsigned nr)
{
	struct journal_device *ja = &ca->journal;
	struct closure cl;
	unsigned current_nr;
	int ret;

	closure_init_stack(&cl);

	do {
		struct disk_reservation disk_res = { 0, 0 };

		closure_sync(&cl);

		mutex_lock(&c->sb_lock);
		current_nr = ja->nr;

		/*
		 * note: journal buckets aren't really counted as _sectors_ used yet, so
		 * we don't need the disk reservation to avoid the BUG_ON() in buckets.c
		 * when space used goes up without a reservation - but we do need the
		 * reservation to ensure we'll actually be able to allocate:
		 */

		if (bch2_disk_reservation_get(c, &disk_res,
					      bucket_to_sector(ca, nr - ja->nr), 1, 0)) {
			mutex_unlock(&c->sb_lock);
			return -ENOSPC;
		}

		ret = __bch2_set_nr_journal_buckets(ca, nr, false, &cl);

		bch2_disk_reservation_put(c, &disk_res);

		if (ja->nr != current_nr)
			bch2_write_super(c);
		mutex_unlock(&c->sb_lock);
	} while (ret == -EAGAIN);

	return ret;
}

int bch2_dev_journal_alloc(struct bch_dev *ca)
{
	unsigned nr;

	if (dynamic_fault("bcachefs:add:journal_alloc"))
		return -ENOMEM;

	/*
	 * clamp journal size to 1024 buckets or 512MB (in sectors), whichever
	 * is smaller:
	 */
	nr = clamp_t(unsigned, ca->mi.nbuckets >> 8,
		     BCH_JOURNAL_BUCKETS_MIN,
		     min(1 << 10,
			 (1 << 20) / ca->mi.bucket_size));

	return __bch2_set_nr_journal_buckets(ca, nr, true, NULL);
}

/* startup/shutdown: */

static bool bch2_journal_writing_to_device(struct journal *j, unsigned dev_idx)
{
	union journal_res_state state;
	struct journal_buf *w;
	bool ret;

	spin_lock(&j->lock);
	state = READ_ONCE(j->reservations);
	w = j->buf + !state.idx;

	ret = state.prev_buf_unwritten &&
		bch2_extent_has_device(bkey_i_to_s_c_extent(&w->key), dev_idx);
	spin_unlock(&j->lock);

	return ret;
}

void bch2_dev_journal_stop(struct journal *j, struct bch_dev *ca)
{
	wait_event(j->wait, !bch2_journal_writing_to_device(j, ca->dev_idx));
}

void bch2_fs_journal_stop(struct journal *j)
{
	struct bch_fs *c = container_of(j, struct bch_fs, journal);

	bch2_journal_flush_all_pins(j);

	wait_event(j->wait, journal_entry_close(j));

	/* do we need to write another journal entry? */
	if (test_bit(JOURNAL_NOT_EMPTY, &j->flags) ||
	    c->btree_roots_dirty)
		bch2_journal_meta(j);

	journal_quiesce(j);

	BUG_ON(!bch2_journal_error(j) &&
	       test_bit(JOURNAL_NOT_EMPTY, &j->flags));

	cancel_delayed_work_sync(&j->write_work);
	cancel_delayed_work_sync(&j->reclaim_work);
}

int bch2_fs_journal_start(struct journal *j, u64 cur_seq,
			  struct list_head *journal_entries)
{
	struct bch_fs *c = container_of(j, struct bch_fs, journal);
	struct journal_entry_pin_list *p;
	struct journal_replay *i;
	u64 last_seq = cur_seq, nr, seq;

	if (!list_empty(journal_entries))
		last_seq = le64_to_cpu(list_first_entry(journal_entries,
							struct journal_replay,
							list)->j.seq);

	nr = cur_seq - last_seq;

	if (nr + 1 > j->pin.size) {
		free_fifo(&j->pin);
		init_fifo(&j->pin, roundup_pow_of_two(nr + 1), GFP_KERNEL);
		if (!j->pin.data) {
			bch_err(c, "error reallocating journal fifo (%llu open entries)", nr);
			return -ENOMEM;
		}
	}

	j->replay_journal_seq	= last_seq;
	j->replay_journal_seq_end = cur_seq;
	j->last_seq_ondisk	= last_seq;
	j->pin.front		= last_seq;
	j->pin.back		= cur_seq;
	atomic64_set(&j->seq, cur_seq - 1);

	fifo_for_each_entry_ptr(p, &j->pin, seq) {
		INIT_LIST_HEAD(&p->list);
		INIT_LIST_HEAD(&p->flushed);
		atomic_set(&p->count, 1);
		p->devs.nr = 0;
	}

	list_for_each_entry(i, journal_entries, list) {
		seq = le64_to_cpu(i->j.seq);

		BUG_ON(seq < last_seq || seq >= cur_seq);

		journal_seq_pin(j, seq)->devs = i->devs;
	}

	spin_lock(&j->lock);

	set_bit(JOURNAL_STARTED, &j->flags);

	journal_pin_new_entry(j, 1);
	bch2_journal_buf_init(j);

	c->last_bucket_seq_cleanup = journal_cur_seq(j);

	bch2_journal_space_available(j);
	spin_unlock(&j->lock);

	return 0;
}

/* init/exit: */

void bch2_dev_journal_exit(struct bch_dev *ca)
{
	kfree(ca->journal.bio);
	kfree(ca->journal.buckets);
	kfree(ca->journal.bucket_seq);

	ca->journal.bio		= NULL;
	ca->journal.buckets	= NULL;
	ca->journal.bucket_seq	= NULL;
}

int bch2_dev_journal_init(struct bch_dev *ca, struct bch_sb *sb)
{
	struct journal_device *ja = &ca->journal;
	struct bch_sb_field_journal *journal_buckets =
		bch2_sb_get_journal(sb);
	unsigned i;

	ja->nr = bch2_nr_journal_buckets(journal_buckets);

	ja->bucket_seq = kcalloc(ja->nr, sizeof(u64), GFP_KERNEL);
	if (!ja->bucket_seq)
		return -ENOMEM;

	ca->journal.bio = bio_kmalloc(GFP_KERNEL,
			DIV_ROUND_UP(JOURNAL_ENTRY_SIZE_MAX, PAGE_SIZE));
	if (!ca->journal.bio)
		return -ENOMEM;

	ja->buckets = kcalloc(ja->nr, sizeof(u64), GFP_KERNEL);
	if (!ja->buckets)
		return -ENOMEM;

	for (i = 0; i < ja->nr; i++)
		ja->buckets[i] = le64_to_cpu(journal_buckets->buckets[i]);

	return 0;
}

void bch2_fs_journal_exit(struct journal *j)
{
	kvpfree(j->buf[1].data, j->buf[1].buf_size);
	kvpfree(j->buf[0].data, j->buf[0].buf_size);
	free_fifo(&j->pin);
}

int bch2_fs_journal_init(struct journal *j)
{
	struct bch_fs *c = container_of(j, struct bch_fs, journal);
	static struct lock_class_key res_key;
	int ret = 0;

	pr_verbose_init(c->opts, "");

	spin_lock_init(&j->lock);
	spin_lock_init(&j->err_lock);
	init_waitqueue_head(&j->wait);
	INIT_DELAYED_WORK(&j->write_work, journal_write_work);
	INIT_DELAYED_WORK(&j->reclaim_work, bch2_journal_reclaim_work);
	init_waitqueue_head(&j->pin_flush_wait);
	mutex_init(&j->reclaim_lock);
	mutex_init(&j->discard_lock);

	lockdep_init_map(&j->res_map, "journal res", &res_key, 0);

	j->buf[0].buf_size	= JOURNAL_ENTRY_SIZE_MIN;
	j->buf[1].buf_size	= JOURNAL_ENTRY_SIZE_MIN;
	j->write_delay_ms	= 1000;
	j->reclaim_delay_ms	= 100;

	/* Btree roots: */
	j->entry_u64s_reserved +=
		BTREE_ID_NR * (JSET_KEYS_U64s + BKEY_EXTENT_U64s_MAX);

	atomic64_set(&j->reservations.counter,
		((union journal_res_state)
		 { .cur_entry_offset = JOURNAL_ENTRY_CLOSED_VAL }).v);

	if (!(init_fifo(&j->pin, JOURNAL_PIN, GFP_KERNEL)) ||
	    !(j->buf[0].data = kvpmalloc(j->buf[0].buf_size, GFP_KERNEL)) ||
	    !(j->buf[1].data = kvpmalloc(j->buf[1].buf_size, GFP_KERNEL))) {
		ret = -ENOMEM;
		goto out;
	}

	j->pin.front = j->pin.back = 1;
out:
	pr_verbose_init(c->opts, "ret %i", ret);
	return ret;
}

/* debug: */

ssize_t bch2_journal_print_debug(struct journal *j, char *buf)
{
	struct printbuf out = _PBUF(buf, PAGE_SIZE);
	struct bch_fs *c = container_of(j, struct bch_fs, journal);
	union journal_res_state s;
	struct bch_dev *ca;
	unsigned iter;

	rcu_read_lock();
	spin_lock(&j->lock);
	s = READ_ONCE(j->reservations);

	pr_buf(&out,
	       "active journal entries:\t%llu\n"
	       "seq:\t\t\t%llu\n"
	       "last_seq:\t\t%llu\n"
	       "last_seq_ondisk:\t%llu\n"
	       "prereserved:\t\t%u/%u\n"
	       "current entry sectors:\t%u\n"
	       "current entry:\t\t",
	       fifo_used(&j->pin),
	       journal_cur_seq(j),
	       journal_last_seq(j),
	       j->last_seq_ondisk,
	       j->prereserved.reserved,
	       j->prereserved.remaining,
	       j->cur_entry_sectors);

	switch (s.cur_entry_offset) {
	case JOURNAL_ENTRY_ERROR_VAL:
		pr_buf(&out, "error\n");
		break;
	case JOURNAL_ENTRY_CLOSED_VAL:
		pr_buf(&out, "closed\n");
		break;
	default:
		pr_buf(&out, "%u/%u\n",
		       s.cur_entry_offset,
		       j->cur_entry_u64s);
		break;
	}

	pr_buf(&out,
	       "current entry refs:\t%u\n"
	       "prev entry unwritten:\t",
	       journal_state_count(s, s.idx));

	if (s.prev_buf_unwritten)
		pr_buf(&out, "yes, ref %u sectors %u\n",
		       journal_state_count(s, !s.idx),
		       journal_prev_buf(j)->sectors);
	else
		pr_buf(&out, "no\n");

	pr_buf(&out,
	       "need write:\t\t%i\n"
	       "replay done:\t\t%i\n",
	       test_bit(JOURNAL_NEED_WRITE,	&j->flags),
	       test_bit(JOURNAL_REPLAY_DONE,	&j->flags));

	for_each_member_device_rcu(ca, c, iter,
				   &c->rw_devs[BCH_DATA_JOURNAL]) {
		struct journal_device *ja = &ca->journal;

		if (!ja->nr)
			continue;

		pr_buf(&out,
		       "dev %u:\n"
		       "\tnr\t\t%u\n"
		       "\tavailable\t%u:%u\n"
		       "\tdiscard_idx\t\t%u\n"
		       "\tdirty_idx_ondisk\t%u (seq %llu)\n"
		       "\tdirty_idx\t\t%u (seq %llu)\n"
		       "\tcur_idx\t\t%u (seq %llu)\n",
		       iter, ja->nr,
		       bch2_journal_dev_buckets_available(j, ja, journal_space_discarded),
		       ja->sectors_free,
		       ja->discard_idx,
		       ja->dirty_idx_ondisk,	ja->bucket_seq[ja->dirty_idx_ondisk],
		       ja->dirty_idx,		ja->bucket_seq[ja->dirty_idx],
		       ja->cur_idx,		ja->bucket_seq[ja->cur_idx]);
	}

	spin_unlock(&j->lock);
	rcu_read_unlock();

	return out.pos - buf;
}

ssize_t bch2_journal_print_pins(struct journal *j, char *buf)
{
	struct printbuf out = _PBUF(buf, PAGE_SIZE);
	struct journal_entry_pin_list *pin_list;
	struct journal_entry_pin *pin;
	u64 i;

	spin_lock(&j->lock);
	fifo_for_each_entry_ptr(pin_list, &j->pin, i) {
		pr_buf(&out, "%llu: count %u\n",
		       i, atomic_read(&pin_list->count));

		list_for_each_entry(pin, &pin_list->list, list)
			pr_buf(&out, "\t%p %pf\n",
			       pin, pin->flush);

		if (!list_empty(&pin_list->flushed))
			pr_buf(&out, "flushed:\n");

		list_for_each_entry(pin, &pin_list->flushed, list)
			pr_buf(&out, "\t%p %pf\n",
			       pin, pin->flush);
	}
	spin_unlock(&j->lock);

	return out.pos - buf;
}