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
|
// SPDX-License-Identifier: GPL-2.0
#include "bcachefs.h"
#include "bkey_buf.h"
#include "btree_cache.h"
#include "btree_io.h"
#include "btree_iter.h"
#include "btree_locking.h"
#include "debug.h"
#include "errcode.h"
#include "error.h"
#include "trace.h"
#include <linux/prefetch.h>
#include <linux/sched/mm.h>
const char * const bch2_btree_node_flags[] = {
#define x(f) #f,
BTREE_FLAGS()
#undef x
NULL
};
void bch2_recalc_btree_reserve(struct bch_fs *c)
{
unsigned i, reserve = 16;
if (!c->btree_roots[0].b)
reserve += 8;
for (i = 0; i < BTREE_ID_NR; i++)
if (c->btree_roots[i].b)
reserve += min_t(unsigned, 1,
c->btree_roots[i].b->c.level) * 8;
c->btree_cache.reserve = reserve;
}
static inline unsigned btree_cache_can_free(struct btree_cache *bc)
{
return max_t(int, 0, bc->used - bc->reserve);
}
static void btree_node_to_freedlist(struct btree_cache *bc, struct btree *b)
{
if (b->c.lock.readers)
list_move(&b->list, &bc->freed_pcpu);
else
list_move(&b->list, &bc->freed_nonpcpu);
}
static void btree_node_data_free(struct bch_fs *c, struct btree *b)
{
struct btree_cache *bc = &c->btree_cache;
EBUG_ON(btree_node_write_in_flight(b));
kvpfree(b->data, btree_bytes(c));
b->data = NULL;
#ifdef __KERNEL__
kvfree(b->aux_data);
#else
munmap(b->aux_data, btree_aux_data_bytes(b));
#endif
b->aux_data = NULL;
bc->used--;
btree_node_to_freedlist(bc, b);
}
static int bch2_btree_cache_cmp_fn(struct rhashtable_compare_arg *arg,
const void *obj)
{
const struct btree *b = obj;
const u64 *v = arg->key;
return b->hash_val == *v ? 0 : 1;
}
static const struct rhashtable_params bch_btree_cache_params = {
.head_offset = offsetof(struct btree, hash),
.key_offset = offsetof(struct btree, hash_val),
.key_len = sizeof(u64),
.obj_cmpfn = bch2_btree_cache_cmp_fn,
};
static int btree_node_data_alloc(struct bch_fs *c, struct btree *b, gfp_t gfp)
{
BUG_ON(b->data || b->aux_data);
b->data = kvpmalloc(btree_bytes(c), gfp);
if (!b->data)
return -ENOMEM;
#ifdef __KERNEL__
b->aux_data = kvmalloc(btree_aux_data_bytes(b), gfp);
#else
b->aux_data = mmap(NULL, btree_aux_data_bytes(b),
PROT_READ|PROT_WRITE|PROT_EXEC,
MAP_PRIVATE|MAP_ANONYMOUS, 0, 0);
if (b->aux_data == MAP_FAILED)
b->aux_data = NULL;
#endif
if (!b->aux_data) {
kvpfree(b->data, btree_bytes(c));
b->data = NULL;
return -ENOMEM;
}
return 0;
}
static struct btree *__btree_node_mem_alloc(struct bch_fs *c)
{
struct btree *b = kzalloc(sizeof(struct btree), GFP_KERNEL);
if (!b)
return NULL;
bkey_btree_ptr_init(&b->key);
six_lock_init(&b->c.lock);
lockdep_set_novalidate_class(&b->c.lock);
INIT_LIST_HEAD(&b->list);
INIT_LIST_HEAD(&b->write_blocked);
b->byte_order = ilog2(btree_bytes(c));
return b;
}
struct btree *__bch2_btree_node_mem_alloc(struct bch_fs *c)
{
struct btree_cache *bc = &c->btree_cache;
struct btree *b = __btree_node_mem_alloc(c);
if (!b)
return NULL;
if (btree_node_data_alloc(c, b, GFP_KERNEL)) {
kfree(b);
return NULL;
}
bc->used++;
list_add(&b->list, &bc->freeable);
return b;
}
/* Btree in memory cache - hash table */
void bch2_btree_node_hash_remove(struct btree_cache *bc, struct btree *b)
{
int ret = rhashtable_remove_fast(&bc->table, &b->hash, bch_btree_cache_params);
BUG_ON(ret);
/* Cause future lookups for this node to fail: */
b->hash_val = 0;
}
int __bch2_btree_node_hash_insert(struct btree_cache *bc, struct btree *b)
{
BUG_ON(b->hash_val);
b->hash_val = btree_ptr_hash_val(&b->key);
return rhashtable_lookup_insert_fast(&bc->table, &b->hash,
bch_btree_cache_params);
}
int bch2_btree_node_hash_insert(struct btree_cache *bc, struct btree *b,
unsigned level, enum btree_id id)
{
int ret;
b->c.level = level;
b->c.btree_id = id;
mutex_lock(&bc->lock);
ret = __bch2_btree_node_hash_insert(bc, b);
if (!ret)
list_add(&b->list, &bc->live);
mutex_unlock(&bc->lock);
return ret;
}
__flatten
static inline struct btree *btree_cache_find(struct btree_cache *bc,
const struct bkey_i *k)
{
u64 v = btree_ptr_hash_val(k);
return rhashtable_lookup_fast(&bc->table, &v, bch_btree_cache_params);
}
/*
* this version is for btree nodes that have already been freed (we're not
* reaping a real btree node)
*/
static int __btree_node_reclaim(struct bch_fs *c, struct btree *b, bool flush)
{
struct btree_cache *bc = &c->btree_cache;
int ret = 0;
lockdep_assert_held(&bc->lock);
wait_on_io:
if (b->flags & ((1U << BTREE_NODE_dirty)|
(1U << BTREE_NODE_read_in_flight)|
(1U << BTREE_NODE_write_in_flight))) {
if (!flush)
return -ENOMEM;
/* XXX: waiting on IO with btree cache lock held */
bch2_btree_node_wait_on_read(b);
bch2_btree_node_wait_on_write(b);
}
if (!six_trylock_intent(&b->c.lock))
return -ENOMEM;
if (!six_trylock_write(&b->c.lock))
goto out_unlock_intent;
/* recheck under lock */
if (b->flags & ((1U << BTREE_NODE_read_in_flight)|
(1U << BTREE_NODE_write_in_flight))) {
if (!flush)
goto out_unlock;
six_unlock_write(&b->c.lock);
six_unlock_intent(&b->c.lock);
goto wait_on_io;
}
if (btree_node_noevict(b) ||
btree_node_write_blocked(b) ||
btree_node_will_make_reachable(b))
goto out_unlock;
if (btree_node_dirty(b)) {
if (!flush)
goto out_unlock;
/*
* Using the underscore version because we don't want to compact
* bsets after the write, since this node is about to be evicted
* - unless btree verify mode is enabled, since it runs out of
* the post write cleanup:
*/
if (bch2_verify_btree_ondisk)
bch2_btree_node_write(c, b, SIX_LOCK_intent, 0);
else
__bch2_btree_node_write(c, b, 0);
six_unlock_write(&b->c.lock);
six_unlock_intent(&b->c.lock);
goto wait_on_io;
}
out:
if (b->hash_val && !ret)
trace_and_count(c, btree_cache_reap, c, b);
return ret;
out_unlock:
six_unlock_write(&b->c.lock);
out_unlock_intent:
six_unlock_intent(&b->c.lock);
ret = -ENOMEM;
goto out;
}
static int btree_node_reclaim(struct bch_fs *c, struct btree *b)
{
return __btree_node_reclaim(c, b, false);
}
static int btree_node_write_and_reclaim(struct bch_fs *c, struct btree *b)
{
return __btree_node_reclaim(c, b, true);
}
static unsigned long bch2_btree_cache_scan(struct shrinker *shrink,
struct shrink_control *sc)
{
struct bch_fs *c = container_of(shrink, struct bch_fs,
btree_cache.shrink);
struct btree_cache *bc = &c->btree_cache;
struct btree *b, *t;
unsigned long nr = sc->nr_to_scan;
unsigned long can_free = 0;
unsigned long touched = 0;
unsigned long freed = 0;
unsigned i, flags;
unsigned long ret = SHRINK_STOP;
if (bch2_btree_shrinker_disabled)
return SHRINK_STOP;
/* Return -1 if we can't do anything right now */
if (sc->gfp_mask & __GFP_FS)
mutex_lock(&bc->lock);
else if (!mutex_trylock(&bc->lock))
goto out_norestore;
flags = memalloc_nofs_save();
/*
* It's _really_ critical that we don't free too many btree nodes - we
* have to always leave ourselves a reserve. The reserve is how we
* guarantee that allocating memory for a new btree node can always
* succeed, so that inserting keys into the btree can always succeed and
* IO can always make forward progress:
*/
can_free = btree_cache_can_free(bc);
nr = min_t(unsigned long, nr, can_free);
i = 0;
list_for_each_entry_safe(b, t, &bc->freeable, list) {
/*
* Leave a few nodes on the freeable list, so that a btree split
* won't have to hit the system allocator:
*/
if (++i <= 3)
continue;
touched++;
if (touched >= nr)
break;
if (!btree_node_reclaim(c, b)) {
btree_node_data_free(c, b);
six_unlock_write(&b->c.lock);
six_unlock_intent(&b->c.lock);
freed++;
}
}
restart:
list_for_each_entry_safe(b, t, &bc->live, list) {
/* tweak this */
if (btree_node_accessed(b)) {
clear_btree_node_accessed(b);
goto touched;
}
if (!btree_node_reclaim(c, b)) {
/* can't call bch2_btree_node_hash_remove under lock */
freed++;
if (&t->list != &bc->live)
list_move_tail(&bc->live, &t->list);
btree_node_data_free(c, b);
mutex_unlock(&bc->lock);
bch2_btree_node_hash_remove(bc, b);
six_unlock_write(&b->c.lock);
six_unlock_intent(&b->c.lock);
if (freed >= nr)
goto out;
if (sc->gfp_mask & __GFP_FS)
mutex_lock(&bc->lock);
else if (!mutex_trylock(&bc->lock))
goto out;
goto restart;
} else {
continue;
}
touched:
touched++;
if (touched >= nr) {
/* Save position */
if (&t->list != &bc->live)
list_move_tail(&bc->live, &t->list);
break;
}
}
mutex_unlock(&bc->lock);
out:
ret = freed;
memalloc_nofs_restore(flags);
out_norestore:
trace_and_count(c, btree_cache_scan, sc->nr_to_scan, can_free, ret);
return ret;
}
static unsigned long bch2_btree_cache_count(struct shrinker *shrink,
struct shrink_control *sc)
{
struct bch_fs *c = container_of(shrink, struct bch_fs,
btree_cache.shrink);
struct btree_cache *bc = &c->btree_cache;
if (bch2_btree_shrinker_disabled)
return 0;
return btree_cache_can_free(bc);
}
void bch2_fs_btree_cache_exit(struct bch_fs *c)
{
struct btree_cache *bc = &c->btree_cache;
struct btree *b;
unsigned i, flags;
if (bc->shrink.list.next)
unregister_shrinker(&bc->shrink);
/* vfree() can allocate memory: */
flags = memalloc_nofs_save();
mutex_lock(&bc->lock);
if (c->verify_data)
list_move(&c->verify_data->list, &bc->live);
kvpfree(c->verify_ondisk, btree_bytes(c));
for (i = 0; i < BTREE_ID_NR; i++)
if (c->btree_roots[i].b)
list_add(&c->btree_roots[i].b->list, &bc->live);
list_splice(&bc->freeable, &bc->live);
while (!list_empty(&bc->live)) {
b = list_first_entry(&bc->live, struct btree, list);
BUG_ON(btree_node_read_in_flight(b) ||
btree_node_write_in_flight(b));
if (btree_node_dirty(b))
bch2_btree_complete_write(c, b, btree_current_write(b));
clear_btree_node_dirty_acct(c, b);
btree_node_data_free(c, b);
}
BUG_ON(atomic_read(&c->btree_cache.dirty));
list_splice(&bc->freed_pcpu, &bc->freed_nonpcpu);
while (!list_empty(&bc->freed_nonpcpu)) {
b = list_first_entry(&bc->freed_nonpcpu, struct btree, list);
list_del(&b->list);
six_lock_pcpu_free(&b->c.lock);
kfree(b);
}
mutex_unlock(&bc->lock);
memalloc_nofs_restore(flags);
if (bc->table_init_done)
rhashtable_destroy(&bc->table);
}
int bch2_fs_btree_cache_init(struct bch_fs *c)
{
struct btree_cache *bc = &c->btree_cache;
unsigned i;
int ret = 0;
pr_verbose_init(c->opts, "");
ret = rhashtable_init(&bc->table, &bch_btree_cache_params);
if (ret)
goto out;
bc->table_init_done = true;
bch2_recalc_btree_reserve(c);
for (i = 0; i < bc->reserve; i++)
if (!__bch2_btree_node_mem_alloc(c)) {
ret = -ENOMEM;
goto out;
}
list_splice_init(&bc->live, &bc->freeable);
mutex_init(&c->verify_lock);
bc->shrink.count_objects = bch2_btree_cache_count;
bc->shrink.scan_objects = bch2_btree_cache_scan;
bc->shrink.seeks = 4;
ret = register_shrinker(&bc->shrink, "%s/btree_cache", c->name);
out:
pr_verbose_init(c->opts, "ret %i", ret);
return ret;
}
void bch2_fs_btree_cache_init_early(struct btree_cache *bc)
{
mutex_init(&bc->lock);
INIT_LIST_HEAD(&bc->live);
INIT_LIST_HEAD(&bc->freeable);
INIT_LIST_HEAD(&bc->freed_pcpu);
INIT_LIST_HEAD(&bc->freed_nonpcpu);
}
/*
* We can only have one thread cannibalizing other cached btree nodes at a time,
* or we'll deadlock. We use an open coded mutex to ensure that, which a
* cannibalize_bucket() will take. This means every time we unlock the root of
* the btree, we need to release this lock if we have it held.
*/
void bch2_btree_cache_cannibalize_unlock(struct bch_fs *c)
{
struct btree_cache *bc = &c->btree_cache;
if (bc->alloc_lock == current) {
trace_and_count(c, btree_cache_cannibalize_unlock, c);
bc->alloc_lock = NULL;
closure_wake_up(&bc->alloc_wait);
}
}
int bch2_btree_cache_cannibalize_lock(struct bch_fs *c, struct closure *cl)
{
struct btree_cache *bc = &c->btree_cache;
struct task_struct *old;
old = cmpxchg(&bc->alloc_lock, NULL, current);
if (old == NULL || old == current)
goto success;
if (!cl) {
trace_and_count(c, btree_cache_cannibalize_lock_fail, c);
return -ENOMEM;
}
closure_wait(&bc->alloc_wait, cl);
/* Try again, after adding ourselves to waitlist */
old = cmpxchg(&bc->alloc_lock, NULL, current);
if (old == NULL || old == current) {
/* We raced */
closure_wake_up(&bc->alloc_wait);
goto success;
}
trace_and_count(c, btree_cache_cannibalize_lock_fail, c);
return -EAGAIN;
success:
trace_and_count(c, btree_cache_cannibalize_lock, c);
return 0;
}
static struct btree *btree_node_cannibalize(struct bch_fs *c)
{
struct btree_cache *bc = &c->btree_cache;
struct btree *b;
list_for_each_entry_reverse(b, &bc->live, list)
if (!btree_node_reclaim(c, b))
return b;
while (1) {
list_for_each_entry_reverse(b, &bc->live, list)
if (!btree_node_write_and_reclaim(c, b))
return b;
/*
* Rare case: all nodes were intent-locked.
* Just busy-wait.
*/
WARN_ONCE(1, "btree cache cannibalize failed\n");
cond_resched();
}
}
struct btree *bch2_btree_node_mem_alloc(struct bch_fs *c, bool pcpu_read_locks)
{
struct btree_cache *bc = &c->btree_cache;
struct list_head *freed = pcpu_read_locks
? &bc->freed_pcpu
: &bc->freed_nonpcpu;
struct btree *b, *b2;
u64 start_time = local_clock();
unsigned flags;
flags = memalloc_nofs_save();
mutex_lock(&bc->lock);
/*
* We never free struct btree itself, just the memory that holds the on
* disk node. Check the freed list before allocating a new one:
*/
list_for_each_entry(b, freed, list)
if (!btree_node_reclaim(c, b)) {
list_del_init(&b->list);
goto got_node;
}
b = __btree_node_mem_alloc(c);
if (!b)
goto err_locked;
if (pcpu_read_locks)
six_lock_pcpu_alloc(&b->c.lock);
BUG_ON(!six_trylock_intent(&b->c.lock));
BUG_ON(!six_trylock_write(&b->c.lock));
got_node:
/*
* btree_free() doesn't free memory; it sticks the node on the end of
* the list. Check if there's any freed nodes there:
*/
list_for_each_entry(b2, &bc->freeable, list)
if (!btree_node_reclaim(c, b2)) {
swap(b->data, b2->data);
swap(b->aux_data, b2->aux_data);
btree_node_to_freedlist(bc, b2);
six_unlock_write(&b2->c.lock);
six_unlock_intent(&b2->c.lock);
goto got_mem;
}
mutex_unlock(&bc->lock);
if (btree_node_data_alloc(c, b, __GFP_NOWARN|GFP_KERNEL))
goto err;
mutex_lock(&bc->lock);
bc->used++;
got_mem:
mutex_unlock(&bc->lock);
BUG_ON(btree_node_hashed(b));
BUG_ON(btree_node_dirty(b));
BUG_ON(btree_node_write_in_flight(b));
out:
b->flags = 0;
b->written = 0;
b->nsets = 0;
b->sib_u64s[0] = 0;
b->sib_u64s[1] = 0;
b->whiteout_u64s = 0;
bch2_btree_keys_init(b);
set_btree_node_accessed(b);
bch2_time_stats_update(&c->times[BCH_TIME_btree_node_mem_alloc],
start_time);
memalloc_nofs_restore(flags);
return b;
err:
mutex_lock(&bc->lock);
err_locked:
/* Try to cannibalize another cached btree node: */
if (bc->alloc_lock == current) {
b2 = btree_node_cannibalize(c);
bch2_btree_node_hash_remove(bc, b2);
if (b) {
swap(b->data, b2->data);
swap(b->aux_data, b2->aux_data);
btree_node_to_freedlist(bc, b2);
six_unlock_write(&b2->c.lock);
six_unlock_intent(&b2->c.lock);
} else {
b = b2;
list_del_init(&b->list);
}
mutex_unlock(&bc->lock);
trace_and_count(c, btree_cache_cannibalize, c);
goto out;
}
mutex_unlock(&bc->lock);
memalloc_nofs_restore(flags);
return ERR_PTR(-ENOMEM);
}
/* Slowpath, don't want it inlined into btree_iter_traverse() */
static noinline struct btree *bch2_btree_node_fill(struct bch_fs *c,
struct btree_trans *trans,
struct btree_path *path,
const struct bkey_i *k,
enum btree_id btree_id,
unsigned level,
enum six_lock_type lock_type,
bool sync)
{
struct btree_cache *bc = &c->btree_cache;
struct btree *b;
u32 seq;
BUG_ON(level + 1 >= BTREE_MAX_DEPTH);
/*
* Parent node must be locked, else we could read in a btree node that's
* been freed:
*/
if (trans && !bch2_btree_node_relock(trans, path, level + 1)) {
trace_and_count(c, trans_restart_relock_parent_for_fill, trans, _THIS_IP_, path);
return ERR_PTR(btree_trans_restart(trans, BCH_ERR_transaction_restart_fill_relock));
}
b = bch2_btree_node_mem_alloc(c, level != 0);
if (trans && b == ERR_PTR(-ENOMEM)) {
trans->memory_allocation_failure = true;
trace_and_count(c, trans_restart_memory_allocation_failure, trans, _THIS_IP_, path);
return ERR_PTR(btree_trans_restart(trans, BCH_ERR_transaction_restart_fill_mem_alloc_fail));
}
if (IS_ERR(b))
return b;
bkey_copy(&b->key, k);
if (bch2_btree_node_hash_insert(bc, b, level, btree_id)) {
/* raced with another fill: */
/* mark as unhashed... */
b->hash_val = 0;
mutex_lock(&bc->lock);
list_add(&b->list, &bc->freeable);
mutex_unlock(&bc->lock);
six_unlock_write(&b->c.lock);
six_unlock_intent(&b->c.lock);
return NULL;
}
set_btree_node_read_in_flight(b);
six_unlock_write(&b->c.lock);
seq = b->c.lock.state.seq;
six_unlock_intent(&b->c.lock);
/* Unlock before doing IO: */
if (trans && sync)
bch2_trans_unlock(trans);
bch2_btree_node_read(c, b, sync);
if (!sync)
return NULL;
if (trans) {
int ret = bch2_trans_relock(trans) ?:
bch2_btree_path_relock_intent(trans, path);
if (ret) {
BUG_ON(!trans->restarted);
return ERR_PTR(ret);
}
}
if (!six_relock_type(&b->c.lock, lock_type, seq)) {
if (trans)
trace_and_count(c, trans_restart_relock_after_fill, trans, _THIS_IP_, path);
return ERR_PTR(btree_trans_restart(trans, BCH_ERR_transaction_restart_relock_after_fill));
}
return b;
}
static noinline void btree_bad_header(struct bch_fs *c, struct btree *b)
{
struct printbuf buf = PRINTBUF;
if (!test_bit(BCH_FS_INITIAL_GC_DONE, &c->flags))
return;
prt_printf(&buf,
"btree node header doesn't match ptr\n"
"btree %s level %u\n"
"ptr: ",
bch2_btree_ids[b->c.btree_id], b->c.level);
bch2_bkey_val_to_text(&buf, c, bkey_i_to_s_c(&b->key));
prt_printf(&buf, "\nheader: btree %s level %llu\n"
"min ",
bch2_btree_ids[BTREE_NODE_ID(b->data)],
BTREE_NODE_LEVEL(b->data));
bch2_bpos_to_text(&buf, b->data->min_key);
prt_printf(&buf, "\nmax ");
bch2_bpos_to_text(&buf, b->data->max_key);
bch2_fs_inconsistent(c, "%s", buf.buf);
printbuf_exit(&buf);
}
static inline void btree_check_header(struct bch_fs *c, struct btree *b)
{
if (b->c.btree_id != BTREE_NODE_ID(b->data) ||
b->c.level != BTREE_NODE_LEVEL(b->data) ||
bpos_cmp(b->data->max_key, b->key.k.p) ||
(b->key.k.type == KEY_TYPE_btree_ptr_v2 &&
bpos_cmp(b->data->min_key,
bkey_i_to_btree_ptr_v2(&b->key)->v.min_key)))
btree_bad_header(c, b);
}
/**
* bch_btree_node_get - find a btree node in the cache and lock it, reading it
* in from disk if necessary.
*
* If IO is necessary and running under generic_make_request, returns -EAGAIN.
*
* The btree node will have either a read or a write lock held, depending on
* the @write parameter.
*/
struct btree *bch2_btree_node_get(struct btree_trans *trans, struct btree_path *path,
const struct bkey_i *k, unsigned level,
enum six_lock_type lock_type,
unsigned long trace_ip)
{
struct bch_fs *c = trans->c;
struct btree_cache *bc = &c->btree_cache;
struct btree *b;
struct bset_tree *t;
int ret;
EBUG_ON(level >= BTREE_MAX_DEPTH);
b = btree_node_mem_ptr(k);
/*
* Check b->hash_val _before_ calling btree_node_lock() - this might not
* be the node we want anymore, and trying to lock the wrong node could
* cause an unneccessary transaction restart:
*/
if (likely(c->opts.btree_node_mem_ptr_optimization &&
b &&
b->hash_val == btree_ptr_hash_val(k)))
goto lock_node;
retry:
b = btree_cache_find(bc, k);
if (unlikely(!b)) {
/*
* We must have the parent locked to call bch2_btree_node_fill(),
* else we could read in a btree node from disk that's been
* freed:
*/
b = bch2_btree_node_fill(c, trans, path, k, path->btree_id,
level, lock_type, true);
/* We raced and found the btree node in the cache */
if (!b)
goto retry;
if (IS_ERR(b))
return b;
} else {
lock_node:
/*
* There's a potential deadlock with splits and insertions into
* interior nodes we have to avoid:
*
* The other thread might be holding an intent lock on the node
* we want, and they want to update its parent node so they're
* going to upgrade their intent lock on the parent node to a
* write lock.
*
* But if we're holding a read lock on the parent, and we're
* trying to get the intent lock they're holding, we deadlock.
*
* So to avoid this we drop the read locks on parent nodes when
* we're starting to take intent locks - and handle the race.
*
* The race is that they might be about to free the node we
* want, and dropping our read lock on the parent node lets them
* update the parent marking the node we want as freed, and then
* free it:
*
* To guard against this, btree nodes are evicted from the cache
* when they're freed - and b->hash_val is zeroed out, which we
* check for after we lock the node.
*
* Then, bch2_btree_node_relock() on the parent will fail - because
* the parent was modified, when the pointer to the node we want
* was removed - and we'll bail out:
*/
if (btree_node_read_locked(path, level + 1))
btree_node_unlock(trans, path, level + 1);
ret = btree_node_lock(trans, path, &b->c, level, lock_type, trace_ip);
if (bch2_err_matches(ret, BCH_ERR_transaction_restart))
return ERR_PTR(ret);
BUG_ON(ret);
if (unlikely(b->hash_val != btree_ptr_hash_val(k) ||
b->c.level != level ||
race_fault())) {
six_unlock_type(&b->c.lock, lock_type);
if (bch2_btree_node_relock(trans, path, level + 1))
goto retry;
trace_and_count(c, trans_restart_btree_node_reused, trans, trace_ip, path);
return ERR_PTR(btree_trans_restart(trans, BCH_ERR_transaction_restart_lock_node_reused));
}
}
if (unlikely(btree_node_read_in_flight(b))) {
u32 seq = b->c.lock.state.seq;
six_unlock_type(&b->c.lock, lock_type);
bch2_trans_unlock(trans);
bch2_btree_node_wait_on_read(b);
/*
* should_be_locked is not set on this path yet, so we need to
* relock it specifically:
*/
if (trans) {
int ret = bch2_trans_relock(trans) ?:
bch2_btree_path_relock_intent(trans, path);
if (ret) {
BUG_ON(!trans->restarted);
return ERR_PTR(ret);
}
}
if (!six_relock_type(&b->c.lock, lock_type, seq))
goto retry;
}
prefetch(b->aux_data);
for_each_bset(b, t) {
void *p = (u64 *) b->aux_data + t->aux_data_offset;
prefetch(p + L1_CACHE_BYTES * 0);
prefetch(p + L1_CACHE_BYTES * 1);
prefetch(p + L1_CACHE_BYTES * 2);
}
/* avoid atomic set bit if it's not needed: */
if (!btree_node_accessed(b))
set_btree_node_accessed(b);
if (unlikely(btree_node_read_error(b))) {
six_unlock_type(&b->c.lock, lock_type);
return ERR_PTR(-EIO);
}
EBUG_ON(b->c.btree_id != path->btree_id);
EBUG_ON(BTREE_NODE_LEVEL(b->data) != level);
btree_check_header(c, b);
return b;
}
struct btree *bch2_btree_node_get_noiter(struct btree_trans *trans,
const struct bkey_i *k,
enum btree_id btree_id,
unsigned level,
bool nofill)
{
struct bch_fs *c = trans->c;
struct btree_cache *bc = &c->btree_cache;
struct btree *b;
struct bset_tree *t;
int ret;
EBUG_ON(level >= BTREE_MAX_DEPTH);
if (c->opts.btree_node_mem_ptr_optimization) {
b = btree_node_mem_ptr(k);
if (b)
goto lock_node;
}
retry:
b = btree_cache_find(bc, k);
if (unlikely(!b)) {
if (nofill)
goto out;
b = bch2_btree_node_fill(c, NULL, NULL, k, btree_id,
level, SIX_LOCK_read, true);
/* We raced and found the btree node in the cache */
if (!b)
goto retry;
if (IS_ERR(b) &&
!bch2_btree_cache_cannibalize_lock(c, NULL))
goto retry;
if (IS_ERR(b))
goto out;
} else {
lock_node:
ret = btree_node_lock_nopath(trans, &b->c, SIX_LOCK_read);
if (bch2_err_matches(ret, BCH_ERR_transaction_restart))
return ERR_PTR(ret);
BUG_ON(ret);
if (unlikely(b->hash_val != btree_ptr_hash_val(k) ||
b->c.btree_id != btree_id ||
b->c.level != level)) {
six_unlock_read(&b->c.lock);
goto retry;
}
}
/* XXX: waiting on IO with btree locks held: */
__bch2_btree_node_wait_on_read(b);
prefetch(b->aux_data);
for_each_bset(b, t) {
void *p = (u64 *) b->aux_data + t->aux_data_offset;
prefetch(p + L1_CACHE_BYTES * 0);
prefetch(p + L1_CACHE_BYTES * 1);
prefetch(p + L1_CACHE_BYTES * 2);
}
/* avoid atomic set bit if it's not needed: */
if (!btree_node_accessed(b))
set_btree_node_accessed(b);
if (unlikely(btree_node_read_error(b))) {
six_unlock_read(&b->c.lock);
b = ERR_PTR(-EIO);
goto out;
}
EBUG_ON(b->c.btree_id != btree_id);
EBUG_ON(BTREE_NODE_LEVEL(b->data) != level);
btree_check_header(c, b);
out:
bch2_btree_cache_cannibalize_unlock(c);
return b;
}
int bch2_btree_node_prefetch(struct bch_fs *c,
struct btree_trans *trans,
struct btree_path *path,
const struct bkey_i *k,
enum btree_id btree_id, unsigned level)
{
struct btree_cache *bc = &c->btree_cache;
struct btree *b;
BUG_ON(trans && !btree_node_locked(path, level + 1));
BUG_ON(level >= BTREE_MAX_DEPTH);
b = btree_cache_find(bc, k);
if (b)
return 0;
b = bch2_btree_node_fill(c, trans, path, k, btree_id,
level, SIX_LOCK_read, false);
return PTR_ERR_OR_ZERO(b);
}
void bch2_btree_node_evict(struct btree_trans *trans, const struct bkey_i *k)
{
struct bch_fs *c = trans->c;
struct btree_cache *bc = &c->btree_cache;
struct btree *b;
b = btree_cache_find(bc, k);
if (!b)
return;
wait_on_io:
/* not allowed to wait on io with btree locks held: */
/* XXX we're called from btree_gc which will be holding other btree
* nodes locked
* */
__bch2_btree_node_wait_on_read(b);
__bch2_btree_node_wait_on_write(b);
btree_node_lock_nopath_nofail(trans, &b->c, SIX_LOCK_intent);
btree_node_lock_nopath_nofail(trans, &b->c, SIX_LOCK_write);
if (btree_node_dirty(b)) {
__bch2_btree_node_write(c, b, 0);
six_unlock_write(&b->c.lock);
six_unlock_intent(&b->c.lock);
goto wait_on_io;
}
BUG_ON(btree_node_dirty(b));
mutex_lock(&bc->lock);
btree_node_data_free(c, b);
bch2_btree_node_hash_remove(bc, b);
mutex_unlock(&bc->lock);
six_unlock_write(&b->c.lock);
six_unlock_intent(&b->c.lock);
}
void bch2_btree_node_to_text(struct printbuf *out, struct bch_fs *c,
struct btree *b)
{
const struct bkey_format *f = &b->format;
struct bset_stats stats;
memset(&stats, 0, sizeof(stats));
bch2_btree_keys_stats(b, &stats);
prt_printf(out, "l %u ", b->c.level);
bch2_bpos_to_text(out, b->data->min_key);
prt_printf(out, " - ");
bch2_bpos_to_text(out, b->data->max_key);
prt_printf(out, ":\n"
" ptrs: ");
bch2_val_to_text(out, c, bkey_i_to_s_c(&b->key));
prt_printf(out, "\n"
" format: u64s %u fields %u %u %u %u %u\n"
" unpack fn len: %u\n"
" bytes used %zu/%zu (%zu%% full)\n"
" sib u64s: %u, %u (merge threshold %u)\n"
" nr packed keys %u\n"
" nr unpacked keys %u\n"
" floats %zu\n"
" failed unpacked %zu\n",
f->key_u64s,
f->bits_per_field[0],
f->bits_per_field[1],
f->bits_per_field[2],
f->bits_per_field[3],
f->bits_per_field[4],
b->unpack_fn_len,
b->nr.live_u64s * sizeof(u64),
btree_bytes(c) - sizeof(struct btree_node),
b->nr.live_u64s * 100 / btree_max_u64s(c),
b->sib_u64s[0],
b->sib_u64s[1],
c->btree_foreground_merge_threshold,
b->nr.packed_keys,
b->nr.unpacked_keys,
stats.floats,
stats.failed);
}
void bch2_btree_cache_to_text(struct printbuf *out, struct bch_fs *c)
{
prt_printf(out, "nr nodes:\t\t%u\n", c->btree_cache.used);
prt_printf(out, "nr dirty:\t\t%u\n", atomic_read(&c->btree_cache.dirty));
prt_printf(out, "cannibalize lock:\t%p\n", c->btree_cache.alloc_lock);
}
|