summaryrefslogtreecommitdiffstats
path: root/fs/bcachefs/io_read.c
blob: 3c574d8873a1e209dc7f7f48faacf9928f8a1272 (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
// SPDX-License-Identifier: GPL-2.0
/*
 * Some low level IO code, and hacks for various block layer limitations
 *
 * Copyright 2010, 2011 Kent Overstreet <kent.overstreet@gmail.com>
 * Copyright 2012 Google, Inc.
 */

#include "bcachefs.h"
#include "alloc_background.h"
#include "alloc_foreground.h"
#include "btree_update.h"
#include "buckets.h"
#include "checksum.h"
#include "clock.h"
#include "compress.h"
#include "data_update.h"
#include "disk_groups.h"
#include "ec.h"
#include "error.h"
#include "io_read.h"
#include "io_misc.h"
#include "io_write.h"
#include "subvolume.h"
#include "trace.h"

#include <linux/sched/mm.h>

#ifndef CONFIG_BCACHEFS_NO_LATENCY_ACCT

static bool bch2_target_congested(struct bch_fs *c, u16 target)
{
	const struct bch_devs_mask *devs;
	unsigned d, nr = 0, total = 0;
	u64 now = local_clock(), last;
	s64 congested;
	struct bch_dev *ca;

	if (!target)
		return false;

	rcu_read_lock();
	devs = bch2_target_to_mask(c, target) ?:
		&c->rw_devs[BCH_DATA_user];

	for_each_set_bit(d, devs->d, BCH_SB_MEMBERS_MAX) {
		ca = rcu_dereference(c->devs[d]);
		if (!ca)
			continue;

		congested = atomic_read(&ca->congested);
		last = READ_ONCE(ca->congested_last);
		if (time_after64(now, last))
			congested -= (now - last) >> 12;

		total += max(congested, 0LL);
		nr++;
	}
	rcu_read_unlock();

	return bch2_rand_range(nr * CONGESTED_MAX) < total;
}

#else

static bool bch2_target_congested(struct bch_fs *c, u16 target)
{
	return false;
}

#endif

/* Cache promotion on read */

struct promote_op {
	struct rcu_head		rcu;
	u64			start_time;

	struct rhash_head	hash;
	struct bpos		pos;

	struct data_update	write;
	struct bio_vec		bi_inline_vecs[]; /* must be last */
};

static const struct rhashtable_params bch_promote_params = {
	.head_offset	= offsetof(struct promote_op, hash),
	.key_offset	= offsetof(struct promote_op, pos),
	.key_len	= sizeof(struct bpos),
};

static inline int should_promote(struct bch_fs *c, struct bkey_s_c k,
				  struct bpos pos,
				  struct bch_io_opts opts,
				  unsigned flags)
{
	BUG_ON(!opts.promote_target);

	if (!(flags & BCH_READ_MAY_PROMOTE))
		return -BCH_ERR_nopromote_may_not;

	if (bch2_bkey_has_target(c, k, opts.promote_target))
		return -BCH_ERR_nopromote_already_promoted;

	if (bkey_extent_is_unwritten(k))
		return -BCH_ERR_nopromote_unwritten;

	if (bch2_target_congested(c, opts.promote_target))
		return -BCH_ERR_nopromote_congested;

	if (rhashtable_lookup_fast(&c->promote_table, &pos,
				   bch_promote_params))
		return -BCH_ERR_nopromote_in_flight;

	return 0;
}

static void promote_free(struct bch_fs *c, struct promote_op *op)
{
	int ret;

	bch2_data_update_exit(&op->write);

	ret = rhashtable_remove_fast(&c->promote_table, &op->hash,
				     bch_promote_params);
	BUG_ON(ret);
	bch2_write_ref_put(c, BCH_WRITE_REF_promote);
	kfree_rcu(op, rcu);
}

static void promote_done(struct bch_write_op *wop)
{
	struct promote_op *op =
		container_of(wop, struct promote_op, write.op);
	struct bch_fs *c = op->write.op.c;

	bch2_time_stats_update(&c->times[BCH_TIME_data_promote],
			       op->start_time);
	promote_free(c, op);
}

static void promote_start(struct promote_op *op, struct bch_read_bio *rbio)
{
	struct bio *bio = &op->write.op.wbio.bio;

	trace_and_count(op->write.op.c, read_promote, &rbio->bio);

	/* we now own pages: */
	BUG_ON(!rbio->bounce);
	BUG_ON(rbio->bio.bi_vcnt > bio->bi_max_vecs);

	memcpy(bio->bi_io_vec, rbio->bio.bi_io_vec,
	       sizeof(struct bio_vec) * rbio->bio.bi_vcnt);
	swap(bio->bi_vcnt, rbio->bio.bi_vcnt);

	bch2_data_update_read_done(&op->write, rbio->pick.crc);
}

static struct promote_op *__promote_alloc(struct btree_trans *trans,
					  enum btree_id btree_id,
					  struct bkey_s_c k,
					  struct bpos pos,
					  struct extent_ptr_decoded *pick,
					  struct bch_io_opts opts,
					  unsigned sectors,
					  struct bch_read_bio **rbio)
{
	struct bch_fs *c = trans->c;
	struct promote_op *op = NULL;
	struct bio *bio;
	unsigned pages = DIV_ROUND_UP(sectors, PAGE_SECTORS);
	int ret;

	if (!bch2_write_ref_tryget(c, BCH_WRITE_REF_promote))
		return ERR_PTR(-BCH_ERR_nopromote_no_writes);

	op = kzalloc(sizeof(*op) + sizeof(struct bio_vec) * pages, GFP_KERNEL);
	if (!op) {
		ret = -BCH_ERR_nopromote_enomem;
		goto err;
	}

	op->start_time = local_clock();
	op->pos = pos;

	/*
	 * We don't use the mempool here because extents that aren't
	 * checksummed or compressed can be too big for the mempool:
	 */
	*rbio = kzalloc(sizeof(struct bch_read_bio) +
			sizeof(struct bio_vec) * pages,
			GFP_KERNEL);
	if (!*rbio) {
		ret = -BCH_ERR_nopromote_enomem;
		goto err;
	}

	rbio_init(&(*rbio)->bio, opts);
	bio_init(&(*rbio)->bio, NULL, (*rbio)->bio.bi_inline_vecs, pages, 0);

	if (bch2_bio_alloc_pages(&(*rbio)->bio, sectors << 9, GFP_KERNEL)) {
		ret = -BCH_ERR_nopromote_enomem;
		goto err;
	}

	(*rbio)->bounce		= true;
	(*rbio)->split		= true;
	(*rbio)->kmalloc	= true;

	if (rhashtable_lookup_insert_fast(&c->promote_table, &op->hash,
					  bch_promote_params)) {
		ret = -BCH_ERR_nopromote_in_flight;
		goto err;
	}

	bio = &op->write.op.wbio.bio;
	bio_init(bio, NULL, bio->bi_inline_vecs, pages, 0);

	ret = bch2_data_update_init(trans, NULL, NULL, &op->write,
			writepoint_hashed((unsigned long) current),
			opts,
			(struct data_update_opts) {
				.target		= opts.promote_target,
				.extra_replicas	= 1,
				.write_flags	= BCH_WRITE_ALLOC_NOWAIT|BCH_WRITE_CACHED,
			},
			btree_id, k);
	/*
	 * possible errors: -BCH_ERR_nocow_lock_blocked,
	 * -BCH_ERR_ENOSPC_disk_reservation:
	 */
	if (ret) {
		BUG_ON(rhashtable_remove_fast(&c->promote_table, &op->hash,
					      bch_promote_params));
		goto err;
	}

	op->write.op.end_io = promote_done;

	return op;
err:
	if (*rbio)
		bio_free_pages(&(*rbio)->bio);
	kfree(*rbio);
	*rbio = NULL;
	kfree(op);
	bch2_write_ref_put(c, BCH_WRITE_REF_promote);
	return ERR_PTR(ret);
}

noinline
static struct promote_op *promote_alloc(struct btree_trans *trans,
					struct bvec_iter iter,
					struct bkey_s_c k,
					struct extent_ptr_decoded *pick,
					struct bch_io_opts opts,
					unsigned flags,
					struct bch_read_bio **rbio,
					bool *bounce,
					bool *read_full)
{
	struct bch_fs *c = trans->c;
	bool promote_full = *read_full || READ_ONCE(c->promote_whole_extents);
	/* data might have to be decompressed in the write path: */
	unsigned sectors = promote_full
		? max(pick->crc.compressed_size, pick->crc.live_size)
		: bvec_iter_sectors(iter);
	struct bpos pos = promote_full
		? bkey_start_pos(k.k)
		: POS(k.k->p.inode, iter.bi_sector);
	struct promote_op *promote;
	int ret;

	ret = should_promote(c, k, pos, opts, flags);
	if (ret)
		goto nopromote;

	promote = __promote_alloc(trans,
				  k.k->type == KEY_TYPE_reflink_v
				  ? BTREE_ID_reflink
				  : BTREE_ID_extents,
				  k, pos, pick, opts, sectors, rbio);
	ret = PTR_ERR_OR_ZERO(promote);
	if (ret)
		goto nopromote;

	*bounce		= true;
	*read_full	= promote_full;
	return promote;
nopromote:
	trace_read_nopromote(c, ret);
	return NULL;
}

/* Read */

#define READ_RETRY_AVOID	1
#define READ_RETRY		2
#define READ_ERR		3

enum rbio_context {
	RBIO_CONTEXT_NULL,
	RBIO_CONTEXT_HIGHPRI,
	RBIO_CONTEXT_UNBOUND,
};

static inline struct bch_read_bio *
bch2_rbio_parent(struct bch_read_bio *rbio)
{
	return rbio->split ? rbio->parent : rbio;
}

__always_inline
static void bch2_rbio_punt(struct bch_read_bio *rbio, work_func_t fn,
			   enum rbio_context context,
			   struct workqueue_struct *wq)
{
	if (context <= rbio->context) {
		fn(&rbio->work);
	} else {
		rbio->work.func		= fn;
		rbio->context		= context;
		queue_work(wq, &rbio->work);
	}
}

static inline struct bch_read_bio *bch2_rbio_free(struct bch_read_bio *rbio)
{
	BUG_ON(rbio->bounce && !rbio->split);

	if (rbio->promote)
		promote_free(rbio->c, rbio->promote);
	rbio->promote = NULL;

	if (rbio->bounce)
		bch2_bio_free_pages_pool(rbio->c, &rbio->bio);

	if (rbio->split) {
		struct bch_read_bio *parent = rbio->parent;

		if (rbio->kmalloc)
			kfree(rbio);
		else
			bio_put(&rbio->bio);

		rbio = parent;
	}

	return rbio;
}

/*
 * Only called on a top level bch_read_bio to complete an entire read request,
 * not a split:
 */
static void bch2_rbio_done(struct bch_read_bio *rbio)
{
	if (rbio->start_time)
		bch2_time_stats_update(&rbio->c->times[BCH_TIME_data_read],
				       rbio->start_time);
	bio_endio(&rbio->bio);
}

static void bch2_read_retry_nodecode(struct bch_fs *c, struct bch_read_bio *rbio,
				     struct bvec_iter bvec_iter,
				     struct bch_io_failures *failed,
				     unsigned flags)
{
	struct btree_trans *trans = bch2_trans_get(c);
	struct btree_iter iter;
	struct bkey_buf sk;
	struct bkey_s_c k;
	int ret;

	flags &= ~BCH_READ_LAST_FRAGMENT;
	flags |= BCH_READ_MUST_CLONE;

	bch2_bkey_buf_init(&sk);

	bch2_trans_iter_init(trans, &iter, rbio->data_btree,
			     rbio->read_pos, BTREE_ITER_SLOTS);
retry:
	rbio->bio.bi_status = 0;

	k = bch2_btree_iter_peek_slot(&iter);
	if (bkey_err(k))
		goto err;

	bch2_bkey_buf_reassemble(&sk, c, k);
	k = bkey_i_to_s_c(sk.k);
	bch2_trans_unlock(trans);

	if (!bch2_bkey_matches_ptr(c, k,
				   rbio->pick.ptr,
				   rbio->data_pos.offset -
				   rbio->pick.crc.offset)) {
		/* extent we wanted to read no longer exists: */
		rbio->hole = true;
		goto out;
	}

	ret = __bch2_read_extent(trans, rbio, bvec_iter,
				 rbio->read_pos,
				 rbio->data_btree,
				 k, 0, failed, flags);
	if (ret == READ_RETRY)
		goto retry;
	if (ret)
		goto err;
out:
	bch2_rbio_done(rbio);
	bch2_trans_iter_exit(trans, &iter);
	bch2_trans_put(trans);
	bch2_bkey_buf_exit(&sk, c);
	return;
err:
	rbio->bio.bi_status = BLK_STS_IOERR;
	goto out;
}

static void bch2_rbio_retry(struct work_struct *work)
{
	struct bch_read_bio *rbio =
		container_of(work, struct bch_read_bio, work);
	struct bch_fs *c	= rbio->c;
	struct bvec_iter iter	= rbio->bvec_iter;
	unsigned flags		= rbio->flags;
	subvol_inum inum = {
		.subvol = rbio->subvol,
		.inum	= rbio->read_pos.inode,
	};
	struct bch_io_failures failed = { .nr = 0 };

	trace_and_count(c, read_retry, &rbio->bio);

	if (rbio->retry == READ_RETRY_AVOID)
		bch2_mark_io_failure(&failed, &rbio->pick);

	rbio->bio.bi_status = 0;

	rbio = bch2_rbio_free(rbio);

	flags |= BCH_READ_IN_RETRY;
	flags &= ~BCH_READ_MAY_PROMOTE;

	if (flags & BCH_READ_NODECODE) {
		bch2_read_retry_nodecode(c, rbio, iter, &failed, flags);
	} else {
		flags &= ~BCH_READ_LAST_FRAGMENT;
		flags |= BCH_READ_MUST_CLONE;

		__bch2_read(c, rbio, iter, inum, &failed, flags);
	}
}

static void bch2_rbio_error(struct bch_read_bio *rbio, int retry,
			    blk_status_t error)
{
	rbio->retry = retry;

	if (rbio->flags & BCH_READ_IN_RETRY)
		return;

	if (retry == READ_ERR) {
		rbio = bch2_rbio_free(rbio);

		rbio->bio.bi_status = error;
		bch2_rbio_done(rbio);
	} else {
		bch2_rbio_punt(rbio, bch2_rbio_retry,
			       RBIO_CONTEXT_UNBOUND, system_unbound_wq);
	}
}

static int __bch2_rbio_narrow_crcs(struct btree_trans *trans,
				   struct bch_read_bio *rbio)
{
	struct bch_fs *c = rbio->c;
	u64 data_offset = rbio->data_pos.offset - rbio->pick.crc.offset;
	struct bch_extent_crc_unpacked new_crc;
	struct btree_iter iter;
	struct bkey_i *new;
	struct bkey_s_c k;
	int ret = 0;

	if (crc_is_compressed(rbio->pick.crc))
		return 0;

	k = bch2_bkey_get_iter(trans, &iter, rbio->data_btree, rbio->data_pos,
			       BTREE_ITER_SLOTS|BTREE_ITER_INTENT);
	if ((ret = bkey_err(k)))
		goto out;

	if (bversion_cmp(k.k->version, rbio->version) ||
	    !bch2_bkey_matches_ptr(c, k, rbio->pick.ptr, data_offset))
		goto out;

	/* Extent was merged? */
	if (bkey_start_offset(k.k) < data_offset ||
	    k.k->p.offset > data_offset + rbio->pick.crc.uncompressed_size)
		goto out;

	if (bch2_rechecksum_bio(c, &rbio->bio, rbio->version,
			rbio->pick.crc, NULL, &new_crc,
			bkey_start_offset(k.k) - data_offset, k.k->size,
			rbio->pick.crc.csum_type)) {
		bch_err(c, "error verifying existing checksum while narrowing checksum (memory corruption?)");
		ret = 0;
		goto out;
	}

	/*
	 * going to be temporarily appending another checksum entry:
	 */
	new = bch2_trans_kmalloc(trans, bkey_bytes(k.k) +
				 sizeof(struct bch_extent_crc128));
	if ((ret = PTR_ERR_OR_ZERO(new)))
		goto out;

	bkey_reassemble(new, k);

	if (!bch2_bkey_narrow_crcs(new, new_crc))
		goto out;

	ret = bch2_trans_update(trans, &iter, new,
				BTREE_UPDATE_INTERNAL_SNAPSHOT_NODE);
out:
	bch2_trans_iter_exit(trans, &iter);
	return ret;
}

static noinline void bch2_rbio_narrow_crcs(struct bch_read_bio *rbio)
{
	bch2_trans_do(rbio->c, NULL, NULL, BCH_TRANS_COMMIT_no_enospc,
		      __bch2_rbio_narrow_crcs(trans, rbio));
}

/* Inner part that may run in process context */
static void __bch2_read_endio(struct work_struct *work)
{
	struct bch_read_bio *rbio =
		container_of(work, struct bch_read_bio, work);
	struct bch_fs *c	= rbio->c;
	struct bch_dev *ca	= bch_dev_bkey_exists(c, rbio->pick.ptr.dev);
	struct bio *src		= &rbio->bio;
	struct bio *dst		= &bch2_rbio_parent(rbio)->bio;
	struct bvec_iter dst_iter = rbio->bvec_iter;
	struct bch_extent_crc_unpacked crc = rbio->pick.crc;
	struct nonce nonce = extent_nonce(rbio->version, crc);
	unsigned nofs_flags;
	struct bch_csum csum;
	int ret;

	nofs_flags = memalloc_nofs_save();

	/* Reset iterator for checksumming and copying bounced data: */
	if (rbio->bounce) {
		src->bi_iter.bi_size		= crc.compressed_size << 9;
		src->bi_iter.bi_idx		= 0;
		src->bi_iter.bi_bvec_done	= 0;
	} else {
		src->bi_iter			= rbio->bvec_iter;
	}

	csum = bch2_checksum_bio(c, crc.csum_type, nonce, src);
	if (bch2_crc_cmp(csum, rbio->pick.crc.csum) && !c->opts.no_data_io)
		goto csum_err;

	/*
	 * XXX
	 * We need to rework the narrow_crcs path to deliver the read completion
	 * first, and then punt to a different workqueue, otherwise we're
	 * holding up reads while doing btree updates which is bad for memory
	 * reclaim.
	 */
	if (unlikely(rbio->narrow_crcs))
		bch2_rbio_narrow_crcs(rbio);

	if (rbio->flags & BCH_READ_NODECODE)
		goto nodecode;

	/* Adjust crc to point to subset of data we want: */
	crc.offset     += rbio->offset_into_extent;
	crc.live_size	= bvec_iter_sectors(rbio->bvec_iter);

	if (crc_is_compressed(crc)) {
		ret = bch2_encrypt_bio(c, crc.csum_type, nonce, src);
		if (ret)
			goto decrypt_err;

		if (bch2_bio_uncompress(c, src, dst, dst_iter, crc) &&
		    !c->opts.no_data_io)
			goto decompression_err;
	} else {
		/* don't need to decrypt the entire bio: */
		nonce = nonce_add(nonce, crc.offset << 9);
		bio_advance(src, crc.offset << 9);

		BUG_ON(src->bi_iter.bi_size < dst_iter.bi_size);
		src->bi_iter.bi_size = dst_iter.bi_size;

		ret = bch2_encrypt_bio(c, crc.csum_type, nonce, src);
		if (ret)
			goto decrypt_err;

		if (rbio->bounce) {
			struct bvec_iter src_iter = src->bi_iter;

			bio_copy_data_iter(dst, &dst_iter, src, &src_iter);
		}
	}

	if (rbio->promote) {
		/*
		 * Re encrypt data we decrypted, so it's consistent with
		 * rbio->crc:
		 */
		ret = bch2_encrypt_bio(c, crc.csum_type, nonce, src);
		if (ret)
			goto decrypt_err;

		promote_start(rbio->promote, rbio);
		rbio->promote = NULL;
	}
nodecode:
	if (likely(!(rbio->flags & BCH_READ_IN_RETRY))) {
		rbio = bch2_rbio_free(rbio);
		bch2_rbio_done(rbio);
	}
out:
	memalloc_nofs_restore(nofs_flags);
	return;
csum_err:
	/*
	 * Checksum error: if the bio wasn't bounced, we may have been
	 * reading into buffers owned by userspace (that userspace can
	 * scribble over) - retry the read, bouncing it this time:
	 */
	if (!rbio->bounce && (rbio->flags & BCH_READ_USER_MAPPED)) {
		rbio->flags |= BCH_READ_MUST_BOUNCE;
		bch2_rbio_error(rbio, READ_RETRY, BLK_STS_IOERR);
		goto out;
	}

	struct printbuf buf = PRINTBUF;
	buf.atomic++;
	prt_str(&buf, "data ");
	bch2_csum_err_msg(&buf, crc.csum_type, rbio->pick.crc.csum, csum);

	bch_err_inum_offset_ratelimited(ca,
		rbio->read_pos.inode,
		rbio->read_pos.offset << 9,
		"data %s", buf.buf);
	printbuf_exit(&buf);

	bch2_io_error(ca, BCH_MEMBER_ERROR_checksum);
	bch2_rbio_error(rbio, READ_RETRY_AVOID, BLK_STS_IOERR);
	goto out;
decompression_err:
	bch_err_inum_offset_ratelimited(c, rbio->read_pos.inode,
					rbio->read_pos.offset << 9,
					"decompression error");
	bch2_rbio_error(rbio, READ_ERR, BLK_STS_IOERR);
	goto out;
decrypt_err:
	bch_err_inum_offset_ratelimited(c, rbio->read_pos.inode,
					rbio->read_pos.offset << 9,
					"decrypt error");
	bch2_rbio_error(rbio, READ_ERR, BLK_STS_IOERR);
	goto out;
}

static void bch2_read_endio(struct bio *bio)
{
	struct bch_read_bio *rbio =
		container_of(bio, struct bch_read_bio, bio);
	struct bch_fs *c	= rbio->c;
	struct bch_dev *ca	= bch_dev_bkey_exists(c, rbio->pick.ptr.dev);
	struct workqueue_struct *wq = NULL;
	enum rbio_context context = RBIO_CONTEXT_NULL;

	if (rbio->have_ioref) {
		bch2_latency_acct(ca, rbio->submit_time, READ);
		percpu_ref_put(&ca->io_ref);
	}

	if (!rbio->split)
		rbio->bio.bi_end_io = rbio->end_io;

	if (bch2_dev_inum_io_err_on(bio->bi_status, ca, BCH_MEMBER_ERROR_read,
				    rbio->read_pos.inode,
				    rbio->read_pos.offset,
				    "data read error: %s",
			       bch2_blk_status_to_str(bio->bi_status))) {
		bch2_rbio_error(rbio, READ_RETRY_AVOID, bio->bi_status);
		return;
	}

	if (((rbio->flags & BCH_READ_RETRY_IF_STALE) && race_fault()) ||
	    ptr_stale(ca, &rbio->pick.ptr)) {
		trace_and_count(c, read_reuse_race, &rbio->bio);

		if (rbio->flags & BCH_READ_RETRY_IF_STALE)
			bch2_rbio_error(rbio, READ_RETRY, BLK_STS_AGAIN);
		else
			bch2_rbio_error(rbio, READ_ERR, BLK_STS_AGAIN);
		return;
	}

	if (rbio->narrow_crcs ||
	    rbio->promote ||
	    crc_is_compressed(rbio->pick.crc) ||
	    bch2_csum_type_is_encryption(rbio->pick.crc.csum_type))
		context = RBIO_CONTEXT_UNBOUND,	wq = system_unbound_wq;
	else if (rbio->pick.crc.csum_type)
		context = RBIO_CONTEXT_HIGHPRI,	wq = system_highpri_wq;

	bch2_rbio_punt(rbio, __bch2_read_endio, context, wq);
}

int __bch2_read_indirect_extent(struct btree_trans *trans,
				unsigned *offset_into_extent,
				struct bkey_buf *orig_k)
{
	struct btree_iter iter;
	struct bkey_s_c k;
	u64 reflink_offset;
	int ret;

	reflink_offset = le64_to_cpu(bkey_i_to_reflink_p(orig_k->k)->v.idx) +
		*offset_into_extent;

	k = bch2_bkey_get_iter(trans, &iter, BTREE_ID_reflink,
			       POS(0, reflink_offset), 0);
	ret = bkey_err(k);
	if (ret)
		goto err;

	if (k.k->type != KEY_TYPE_reflink_v &&
	    k.k->type != KEY_TYPE_indirect_inline_data) {
		bch_err_inum_offset_ratelimited(trans->c,
			orig_k->k->k.p.inode,
			orig_k->k->k.p.offset << 9,
			"%llu len %u points to nonexistent indirect extent %llu",
			orig_k->k->k.p.offset,
			orig_k->k->k.size,
			reflink_offset);
		bch2_inconsistent_error(trans->c);
		ret = -EIO;
		goto err;
	}

	*offset_into_extent = iter.pos.offset - bkey_start_offset(k.k);
	bch2_bkey_buf_reassemble(orig_k, trans->c, k);
err:
	bch2_trans_iter_exit(trans, &iter);
	return ret;
}

static noinline void read_from_stale_dirty_pointer(struct btree_trans *trans,
						   struct bkey_s_c k,
						   struct bch_extent_ptr ptr)
{
	struct bch_fs *c = trans->c;
	struct bch_dev *ca = bch_dev_bkey_exists(c, ptr.dev);
	struct btree_iter iter;
	struct printbuf buf = PRINTBUF;
	int ret;

	bch2_trans_iter_init(trans, &iter, BTREE_ID_alloc,
			     PTR_BUCKET_POS(c, &ptr),
			     BTREE_ITER_CACHED);

	prt_printf(&buf, "Attempting to read from stale dirty pointer:");
	printbuf_indent_add(&buf, 2);
	prt_newline(&buf);

	bch2_bkey_val_to_text(&buf, c, k);
	prt_newline(&buf);

	prt_printf(&buf, "memory gen: %u", *bucket_gen(ca, iter.pos.offset));

	ret = lockrestart_do(trans, bkey_err(k = bch2_btree_iter_peek_slot(&iter)));
	if (!ret) {
		prt_newline(&buf);
		bch2_bkey_val_to_text(&buf, c, k);
	}

	bch2_fs_inconsistent(c, "%s", buf.buf);

	bch2_trans_iter_exit(trans, &iter);
	printbuf_exit(&buf);
}

int __bch2_read_extent(struct btree_trans *trans, struct bch_read_bio *orig,
		       struct bvec_iter iter, struct bpos read_pos,
		       enum btree_id data_btree, struct bkey_s_c k,
		       unsigned offset_into_extent,
		       struct bch_io_failures *failed, unsigned flags)
{
	struct bch_fs *c = trans->c;
	struct extent_ptr_decoded pick;
	struct bch_read_bio *rbio = NULL;
	struct bch_dev *ca = NULL;
	struct promote_op *promote = NULL;
	bool bounce = false, read_full = false, narrow_crcs = false;
	struct bpos data_pos = bkey_start_pos(k.k);
	int pick_ret;

	if (bkey_extent_is_inline_data(k.k)) {
		unsigned bytes = min_t(unsigned, iter.bi_size,
				       bkey_inline_data_bytes(k.k));

		swap(iter.bi_size, bytes);
		memcpy_to_bio(&orig->bio, iter, bkey_inline_data_p(k));
		swap(iter.bi_size, bytes);
		bio_advance_iter(&orig->bio, &iter, bytes);
		zero_fill_bio_iter(&orig->bio, iter);
		goto out_read_done;
	}
retry_pick:
	pick_ret = bch2_bkey_pick_read_device(c, k, failed, &pick);

	/* hole or reservation - just zero fill: */
	if (!pick_ret)
		goto hole;

	if (pick_ret < 0) {
		bch_err_inum_offset_ratelimited(c,
				read_pos.inode, read_pos.offset << 9,
				"no device to read from");
		goto err;
	}

	ca = bch_dev_bkey_exists(c, pick.ptr.dev);

	/*
	 * Stale dirty pointers are treated as IO errors, but @failed isn't
	 * allocated unless we're in the retry path - so if we're not in the
	 * retry path, don't check here, it'll be caught in bch2_read_endio()
	 * and we'll end up in the retry path:
	 */
	if ((flags & BCH_READ_IN_RETRY) &&
	    !pick.ptr.cached &&
	    unlikely(ptr_stale(ca, &pick.ptr))) {
		read_from_stale_dirty_pointer(trans, k, pick.ptr);
		bch2_mark_io_failure(failed, &pick);
		goto retry_pick;
	}

	/*
	 * Unlock the iterator while the btree node's lock is still in
	 * cache, before doing the IO:
	 */
	bch2_trans_unlock(trans);

	if (flags & BCH_READ_NODECODE) {
		/*
		 * can happen if we retry, and the extent we were going to read
		 * has been merged in the meantime:
		 */
		if (pick.crc.compressed_size > orig->bio.bi_vcnt * PAGE_SECTORS)
			goto hole;

		iter.bi_size	= pick.crc.compressed_size << 9;
		goto get_bio;
	}

	if (!(flags & BCH_READ_LAST_FRAGMENT) ||
	    bio_flagged(&orig->bio, BIO_CHAIN))
		flags |= BCH_READ_MUST_CLONE;

	narrow_crcs = !(flags & BCH_READ_IN_RETRY) &&
		bch2_can_narrow_extent_crcs(k, pick.crc);

	if (narrow_crcs && (flags & BCH_READ_USER_MAPPED))
		flags |= BCH_READ_MUST_BOUNCE;

	EBUG_ON(offset_into_extent + bvec_iter_sectors(iter) > k.k->size);

	if (crc_is_compressed(pick.crc) ||
	    (pick.crc.csum_type != BCH_CSUM_none &&
	     (bvec_iter_sectors(iter) != pick.crc.uncompressed_size ||
	      (bch2_csum_type_is_encryption(pick.crc.csum_type) &&
	       (flags & BCH_READ_USER_MAPPED)) ||
	      (flags & BCH_READ_MUST_BOUNCE)))) {
		read_full = true;
		bounce = true;
	}

	if (orig->opts.promote_target)
		promote = promote_alloc(trans, iter, k, &pick, orig->opts, flags,
					&rbio, &bounce, &read_full);

	if (!read_full) {
		EBUG_ON(crc_is_compressed(pick.crc));
		EBUG_ON(pick.crc.csum_type &&
			(bvec_iter_sectors(iter) != pick.crc.uncompressed_size ||
			 bvec_iter_sectors(iter) != pick.crc.live_size ||
			 pick.crc.offset ||
			 offset_into_extent));

		data_pos.offset += offset_into_extent;
		pick.ptr.offset += pick.crc.offset +
			offset_into_extent;
		offset_into_extent		= 0;
		pick.crc.compressed_size	= bvec_iter_sectors(iter);
		pick.crc.uncompressed_size	= bvec_iter_sectors(iter);
		pick.crc.offset			= 0;
		pick.crc.live_size		= bvec_iter_sectors(iter);
	}
get_bio:
	if (rbio) {
		/*
		 * promote already allocated bounce rbio:
		 * promote needs to allocate a bio big enough for uncompressing
		 * data in the write path, but we're not going to use it all
		 * here:
		 */
		EBUG_ON(rbio->bio.bi_iter.bi_size <
		       pick.crc.compressed_size << 9);
		rbio->bio.bi_iter.bi_size =
			pick.crc.compressed_size << 9;
	} else if (bounce) {
		unsigned sectors = pick.crc.compressed_size;

		rbio = rbio_init(bio_alloc_bioset(NULL,
						  DIV_ROUND_UP(sectors, PAGE_SECTORS),
						  0,
						  GFP_NOFS,
						  &c->bio_read_split),
				 orig->opts);

		bch2_bio_alloc_pages_pool(c, &rbio->bio, sectors << 9);
		rbio->bounce	= true;
		rbio->split	= true;
	} else if (flags & BCH_READ_MUST_CLONE) {
		/*
		 * Have to clone if there were any splits, due to error
		 * reporting issues (if a split errored, and retrying didn't
		 * work, when it reports the error to its parent (us) we don't
		 * know if the error was from our bio, and we should retry, or
		 * from the whole bio, in which case we don't want to retry and
		 * lose the error)
		 */
		rbio = rbio_init(bio_alloc_clone(NULL, &orig->bio, GFP_NOFS,
						 &c->bio_read_split),
				 orig->opts);
		rbio->bio.bi_iter = iter;
		rbio->split	= true;
	} else {
		rbio = orig;
		rbio->bio.bi_iter = iter;
		EBUG_ON(bio_flagged(&rbio->bio, BIO_CHAIN));
	}

	EBUG_ON(bio_sectors(&rbio->bio) != pick.crc.compressed_size);

	rbio->c			= c;
	rbio->submit_time	= local_clock();
	if (rbio->split)
		rbio->parent	= orig;
	else
		rbio->end_io	= orig->bio.bi_end_io;
	rbio->bvec_iter		= iter;
	rbio->offset_into_extent= offset_into_extent;
	rbio->flags		= flags;
	rbio->have_ioref	= pick_ret > 0 && bch2_dev_get_ioref(ca, READ);
	rbio->narrow_crcs	= narrow_crcs;
	rbio->hole		= 0;
	rbio->retry		= 0;
	rbio->context		= 0;
	/* XXX: only initialize this if needed */
	rbio->devs_have		= bch2_bkey_devs(k);
	rbio->pick		= pick;
	rbio->subvol		= orig->subvol;
	rbio->read_pos		= read_pos;
	rbio->data_btree	= data_btree;
	rbio->data_pos		= data_pos;
	rbio->version		= k.k->version;
	rbio->promote		= promote;
	INIT_WORK(&rbio->work, NULL);

	rbio->bio.bi_opf	= orig->bio.bi_opf;
	rbio->bio.bi_iter.bi_sector = pick.ptr.offset;
	rbio->bio.bi_end_io	= bch2_read_endio;

	if (rbio->bounce)
		trace_and_count(c, read_bounce, &rbio->bio);

	this_cpu_add(c->counters[BCH_COUNTER_io_read], bio_sectors(&rbio->bio));
	bch2_increment_clock(c, bio_sectors(&rbio->bio), READ);

	/*
	 * If it's being moved internally, we don't want to flag it as a cache
	 * hit:
	 */
	if (pick.ptr.cached && !(flags & BCH_READ_NODECODE))
		bch2_bucket_io_time_reset(trans, pick.ptr.dev,
			PTR_BUCKET_NR(ca, &pick.ptr), READ);

	if (!(flags & (BCH_READ_IN_RETRY|BCH_READ_LAST_FRAGMENT))) {
		bio_inc_remaining(&orig->bio);
		trace_and_count(c, read_split, &orig->bio);
	}

	if (!rbio->pick.idx) {
		if (!rbio->have_ioref) {
			bch_err_inum_offset_ratelimited(c,
					read_pos.inode,
					read_pos.offset << 9,
					"no device to read from");
			bch2_rbio_error(rbio, READ_RETRY_AVOID, BLK_STS_IOERR);
			goto out;
		}

		this_cpu_add(ca->io_done->sectors[READ][BCH_DATA_user],
			     bio_sectors(&rbio->bio));
		bio_set_dev(&rbio->bio, ca->disk_sb.bdev);

		if (unlikely(c->opts.no_data_io)) {
			if (likely(!(flags & BCH_READ_IN_RETRY)))
				bio_endio(&rbio->bio);
		} else {
			if (likely(!(flags & BCH_READ_IN_RETRY)))
				submit_bio(&rbio->bio);
			else
				submit_bio_wait(&rbio->bio);
		}

		/*
		 * We just submitted IO which may block, we expect relock fail
		 * events and shouldn't count them:
		 */
		trans->notrace_relock_fail = true;
	} else {
		/* Attempting reconstruct read: */
		if (bch2_ec_read_extent(trans, rbio)) {
			bch2_rbio_error(rbio, READ_RETRY_AVOID, BLK_STS_IOERR);
			goto out;
		}

		if (likely(!(flags & BCH_READ_IN_RETRY)))
			bio_endio(&rbio->bio);
	}
out:
	if (likely(!(flags & BCH_READ_IN_RETRY))) {
		return 0;
	} else {
		int ret;

		rbio->context = RBIO_CONTEXT_UNBOUND;
		bch2_read_endio(&rbio->bio);

		ret = rbio->retry;
		rbio = bch2_rbio_free(rbio);

		if (ret == READ_RETRY_AVOID) {
			bch2_mark_io_failure(failed, &pick);
			ret = READ_RETRY;
		}

		if (!ret)
			goto out_read_done;

		return ret;
	}

err:
	if (flags & BCH_READ_IN_RETRY)
		return READ_ERR;

	orig->bio.bi_status = BLK_STS_IOERR;
	goto out_read_done;

hole:
	/*
	 * won't normally happen in the BCH_READ_NODECODE
	 * (bch2_move_extent()) path, but if we retry and the extent we wanted
	 * to read no longer exists we have to signal that:
	 */
	if (flags & BCH_READ_NODECODE)
		orig->hole = true;

	zero_fill_bio_iter(&orig->bio, iter);
out_read_done:
	if (flags & BCH_READ_LAST_FRAGMENT)
		bch2_rbio_done(orig);
	return 0;
}

void __bch2_read(struct bch_fs *c, struct bch_read_bio *rbio,
		 struct bvec_iter bvec_iter, subvol_inum inum,
		 struct bch_io_failures *failed, unsigned flags)
{
	struct btree_trans *trans = bch2_trans_get(c);
	struct btree_iter iter;
	struct bkey_buf sk;
	struct bkey_s_c k;
	u32 snapshot;
	int ret;

	BUG_ON(flags & BCH_READ_NODECODE);

	bch2_bkey_buf_init(&sk);
retry:
	bch2_trans_begin(trans);
	iter = (struct btree_iter) { NULL };

	ret = bch2_subvolume_get_snapshot(trans, inum.subvol, &snapshot);
	if (ret)
		goto err;

	bch2_trans_iter_init(trans, &iter, BTREE_ID_extents,
			     SPOS(inum.inum, bvec_iter.bi_sector, snapshot),
			     BTREE_ITER_SLOTS);
	while (1) {
		unsigned bytes, sectors, offset_into_extent;
		enum btree_id data_btree = BTREE_ID_extents;

		/*
		 * read_extent -> io_time_reset may cause a transaction restart
		 * without returning an error, we need to check for that here:
		 */
		ret = bch2_trans_relock(trans);
		if (ret)
			break;

		bch2_btree_iter_set_pos(&iter,
				POS(inum.inum, bvec_iter.bi_sector));

		k = bch2_btree_iter_peek_slot(&iter);
		ret = bkey_err(k);
		if (ret)
			break;

		offset_into_extent = iter.pos.offset -
			bkey_start_offset(k.k);
		sectors = k.k->size - offset_into_extent;

		bch2_bkey_buf_reassemble(&sk, c, k);

		ret = bch2_read_indirect_extent(trans, &data_btree,
					&offset_into_extent, &sk);
		if (ret)
			break;

		k = bkey_i_to_s_c(sk.k);

		/*
		 * With indirect extents, the amount of data to read is the min
		 * of the original extent and the indirect extent:
		 */
		sectors = min(sectors, k.k->size - offset_into_extent);

		bytes = min(sectors, bvec_iter_sectors(bvec_iter)) << 9;
		swap(bvec_iter.bi_size, bytes);

		if (bvec_iter.bi_size == bytes)
			flags |= BCH_READ_LAST_FRAGMENT;

		ret = __bch2_read_extent(trans, rbio, bvec_iter, iter.pos,
					 data_btree, k,
					 offset_into_extent, failed, flags);
		if (ret)
			break;

		if (flags & BCH_READ_LAST_FRAGMENT)
			break;

		swap(bvec_iter.bi_size, bytes);
		bio_advance_iter(&rbio->bio, &bvec_iter, bytes);

		ret = btree_trans_too_many_iters(trans);
		if (ret)
			break;
	}
err:
	bch2_trans_iter_exit(trans, &iter);

	if (bch2_err_matches(ret, BCH_ERR_transaction_restart) ||
	    ret == READ_RETRY ||
	    ret == READ_RETRY_AVOID)
		goto retry;

	bch2_trans_put(trans);
	bch2_bkey_buf_exit(&sk, c);

	if (ret) {
		bch_err_inum_offset_ratelimited(c, inum.inum,
						bvec_iter.bi_sector << 9,
						"read error %i from btree lookup", ret);
		rbio->bio.bi_status = BLK_STS_IOERR;
		bch2_rbio_done(rbio);
	}
}

void bch2_fs_io_read_exit(struct bch_fs *c)
{
	if (c->promote_table.tbl)
		rhashtable_destroy(&c->promote_table);
	bioset_exit(&c->bio_read_split);
	bioset_exit(&c->bio_read);
}

int bch2_fs_io_read_init(struct bch_fs *c)
{
	if (bioset_init(&c->bio_read, 1, offsetof(struct bch_read_bio, bio),
			BIOSET_NEED_BVECS))
		return -BCH_ERR_ENOMEM_bio_read_init;

	if (bioset_init(&c->bio_read_split, 1, offsetof(struct bch_read_bio, bio),
			BIOSET_NEED_BVECS))
		return -BCH_ERR_ENOMEM_bio_read_split_init;

	if (rhashtable_init(&c->promote_table, &bch_promote_params))
		return -BCH_ERR_ENOMEM_promote_table_init;

	return 0;
}