summaryrefslogtreecommitdiffstats
path: root/fs/bcachefs/btree_key_cache.c
blob: 1b7a5668df7cc4694f73f7c287a1858f3b61074e (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
// SPDX-License-Identifier: GPL-2.0

#include "bcachefs.h"
#include "btree_cache.h"
#include "btree_iter.h"
#include "btree_key_cache.h"
#include "btree_locking.h"
#include "btree_update.h"
#include "errcode.h"
#include "error.h"
#include "journal.h"
#include "journal_reclaim.h"
#include "trace.h"

#include <linux/sched/mm.h>

static inline bool btree_uses_pcpu_readers(enum btree_id id)
{
	return id == BTREE_ID_subvolumes;
}

static struct kmem_cache *bch2_key_cache;

static int bch2_btree_key_cache_cmp_fn(struct rhashtable_compare_arg *arg,
				       const void *obj)
{
	const struct bkey_cached *ck = obj;
	const struct bkey_cached_key *key = arg->key;

	return ck->key.btree_id != key->btree_id ||
		!bpos_eq(ck->key.pos, key->pos);
}

static const struct rhashtable_params bch2_btree_key_cache_params = {
	.head_offset	= offsetof(struct bkey_cached, hash),
	.key_offset	= offsetof(struct bkey_cached, key),
	.key_len	= sizeof(struct bkey_cached_key),
	.obj_cmpfn	= bch2_btree_key_cache_cmp_fn,
};

__flatten
inline struct bkey_cached *
bch2_btree_key_cache_find(struct bch_fs *c, enum btree_id btree_id, struct bpos pos)
{
	struct bkey_cached_key key = {
		.btree_id	= btree_id,
		.pos		= pos,
	};

	return rhashtable_lookup_fast(&c->btree_key_cache.table, &key,
				      bch2_btree_key_cache_params);
}

static bool bkey_cached_lock_for_evict(struct bkey_cached *ck)
{
	if (!six_trylock_intent(&ck->c.lock))
		return false;

	if (test_bit(BKEY_CACHED_DIRTY, &ck->flags)) {
		six_unlock_intent(&ck->c.lock);
		return false;
	}

	if (!six_trylock_write(&ck->c.lock)) {
		six_unlock_intent(&ck->c.lock);
		return false;
	}

	return true;
}

static void bkey_cached_evict(struct btree_key_cache *c,
			      struct bkey_cached *ck)
{
	BUG_ON(rhashtable_remove_fast(&c->table, &ck->hash,
				      bch2_btree_key_cache_params));
	memset(&ck->key, ~0, sizeof(ck->key));

	atomic_long_dec(&c->nr_keys);
}

static void bkey_cached_free(struct btree_key_cache *bc,
			     struct bkey_cached *ck)
{
	struct bch_fs *c = container_of(bc, struct bch_fs, btree_key_cache);

	BUG_ON(test_bit(BKEY_CACHED_DIRTY, &ck->flags));

	ck->btree_trans_barrier_seq =
		start_poll_synchronize_srcu(&c->btree_trans_barrier);

	if (ck->c.lock.readers) {
		list_move_tail(&ck->list, &bc->freed_pcpu);
		bc->nr_freed_pcpu++;
	} else {
		list_move_tail(&ck->list, &bc->freed_nonpcpu);
		bc->nr_freed_nonpcpu++;
	}
	atomic_long_inc(&bc->nr_freed);

	kfree(ck->k);
	ck->k		= NULL;
	ck->u64s	= 0;

	six_unlock_write(&ck->c.lock);
	six_unlock_intent(&ck->c.lock);
}

#ifdef __KERNEL__
static void __bkey_cached_move_to_freelist_ordered(struct btree_key_cache *bc,
						   struct bkey_cached *ck)
{
	struct bkey_cached *pos;

	bc->nr_freed_nonpcpu++;

	list_for_each_entry_reverse(pos, &bc->freed_nonpcpu, list) {
		if (ULONG_CMP_GE(ck->btree_trans_barrier_seq,
				 pos->btree_trans_barrier_seq)) {
			list_move(&ck->list, &pos->list);
			return;
		}
	}

	list_move(&ck->list, &bc->freed_nonpcpu);
}
#endif

static void bkey_cached_move_to_freelist(struct btree_key_cache *bc,
					 struct bkey_cached *ck)
{
	BUG_ON(test_bit(BKEY_CACHED_DIRTY, &ck->flags));

	if (!ck->c.lock.readers) {
#ifdef __KERNEL__
		struct btree_key_cache_freelist *f;
		bool freed = false;

		preempt_disable();
		f = this_cpu_ptr(bc->pcpu_freed);

		if (f->nr < ARRAY_SIZE(f->objs)) {
			f->objs[f->nr++] = ck;
			freed = true;
		}
		preempt_enable();

		if (!freed) {
			mutex_lock(&bc->lock);
			preempt_disable();
			f = this_cpu_ptr(bc->pcpu_freed);

			while (f->nr > ARRAY_SIZE(f->objs) / 2) {
				struct bkey_cached *ck2 = f->objs[--f->nr];

				__bkey_cached_move_to_freelist_ordered(bc, ck2);
			}
			preempt_enable();

			__bkey_cached_move_to_freelist_ordered(bc, ck);
			mutex_unlock(&bc->lock);
		}
#else
		mutex_lock(&bc->lock);
		list_move_tail(&ck->list, &bc->freed_nonpcpu);
		bc->nr_freed_nonpcpu++;
		mutex_unlock(&bc->lock);
#endif
	} else {
		mutex_lock(&bc->lock);
		list_move_tail(&ck->list, &bc->freed_pcpu);
		mutex_unlock(&bc->lock);
	}
}

static void bkey_cached_free_fast(struct btree_key_cache *bc,
				  struct bkey_cached *ck)
{
	struct bch_fs *c = container_of(bc, struct bch_fs, btree_key_cache);

	ck->btree_trans_barrier_seq =
		start_poll_synchronize_srcu(&c->btree_trans_barrier);

	list_del_init(&ck->list);
	atomic_long_inc(&bc->nr_freed);

	kfree(ck->k);
	ck->k		= NULL;
	ck->u64s	= 0;

	bkey_cached_move_to_freelist(bc, ck);

	six_unlock_write(&ck->c.lock);
	six_unlock_intent(&ck->c.lock);
}

static struct bkey_cached *
bkey_cached_alloc(struct btree_trans *trans, struct btree_path *path,
		  bool *was_new)
{
	struct bch_fs *c = trans->c;
	struct btree_key_cache *bc = &c->btree_key_cache;
	struct bkey_cached *ck = NULL;
	bool pcpu_readers = btree_uses_pcpu_readers(path->btree_id);
	int ret;

	if (!pcpu_readers) {
#ifdef __KERNEL__
		struct btree_key_cache_freelist *f;

		preempt_disable();
		f = this_cpu_ptr(bc->pcpu_freed);
		if (f->nr)
			ck = f->objs[--f->nr];
		preempt_enable();

		if (!ck) {
			mutex_lock(&bc->lock);
			preempt_disable();
			f = this_cpu_ptr(bc->pcpu_freed);

			while (!list_empty(&bc->freed_nonpcpu) &&
			       f->nr < ARRAY_SIZE(f->objs) / 2) {
				ck = list_last_entry(&bc->freed_nonpcpu, struct bkey_cached, list);
				list_del_init(&ck->list);
				bc->nr_freed_nonpcpu--;
				f->objs[f->nr++] = ck;
			}

			ck = f->nr ? f->objs[--f->nr] : NULL;
			preempt_enable();
			mutex_unlock(&bc->lock);
		}
#else
		mutex_lock(&bc->lock);
		if (!list_empty(&bc->freed_nonpcpu)) {
			ck = list_last_entry(&bc->freed_nonpcpu, struct bkey_cached, list);
			list_del_init(&ck->list);
			bc->nr_freed_nonpcpu--;
		}
		mutex_unlock(&bc->lock);
#endif
	} else {
		mutex_lock(&bc->lock);
		if (!list_empty(&bc->freed_pcpu)) {
			ck = list_last_entry(&bc->freed_pcpu, struct bkey_cached, list);
			list_del_init(&ck->list);
		}
		mutex_unlock(&bc->lock);
	}

	if (ck) {
		ret = btree_node_lock_nopath(trans, &ck->c, SIX_LOCK_intent, _THIS_IP_);
		if (unlikely(ret)) {
			bkey_cached_move_to_freelist(bc, ck);
			return ERR_PTR(ret);
		}

		path->l[0].b = (void *) ck;
		path->l[0].lock_seq = six_lock_seq(&ck->c.lock);
		mark_btree_node_locked(trans, path, 0, BTREE_NODE_INTENT_LOCKED);

		ret = bch2_btree_node_lock_write(trans, path, &ck->c);
		if (unlikely(ret)) {
			btree_node_unlock(trans, path, 0);
			bkey_cached_move_to_freelist(bc, ck);
			return ERR_PTR(ret);
		}

		return ck;
	}

	ck = allocate_dropping_locks(trans, ret,
			kmem_cache_zalloc(bch2_key_cache, _gfp));
	if (ret) {
		kmem_cache_free(bch2_key_cache, ck);
		return ERR_PTR(ret);
	}

	if (!ck)
		return NULL;

	INIT_LIST_HEAD(&ck->list);
	bch2_btree_lock_init(&ck->c, pcpu_readers ? SIX_LOCK_INIT_PCPU : 0);

	ck->c.cached = true;
	BUG_ON(!six_trylock_intent(&ck->c.lock));
	BUG_ON(!six_trylock_write(&ck->c.lock));
	*was_new = true;
	return ck;
}

static struct bkey_cached *
bkey_cached_reuse(struct btree_key_cache *c)
{
	struct bucket_table *tbl;
	struct rhash_head *pos;
	struct bkey_cached *ck;
	unsigned i;

	mutex_lock(&c->lock);
	rcu_read_lock();
	tbl = rht_dereference_rcu(c->table.tbl, &c->table);
	for (i = 0; i < tbl->size; i++)
		rht_for_each_entry_rcu(ck, pos, tbl, i, hash) {
			if (!test_bit(BKEY_CACHED_DIRTY, &ck->flags) &&
			    bkey_cached_lock_for_evict(ck)) {
				bkey_cached_evict(c, ck);
				goto out;
			}
		}
	ck = NULL;
out:
	rcu_read_unlock();
	mutex_unlock(&c->lock);
	return ck;
}

static struct bkey_cached *
btree_key_cache_create(struct btree_trans *trans, struct btree_path *path)
{
	struct bch_fs *c = trans->c;
	struct btree_key_cache *bc = &c->btree_key_cache;
	struct bkey_cached *ck;
	bool was_new = false;

	ck = bkey_cached_alloc(trans, path, &was_new);
	if (IS_ERR(ck))
		return ck;

	if (unlikely(!ck)) {
		ck = bkey_cached_reuse(bc);
		if (unlikely(!ck)) {
			bch_err(c, "error allocating memory for key cache item, btree %s",
				bch2_btree_id_str(path->btree_id));
			return ERR_PTR(-BCH_ERR_ENOMEM_btree_key_cache_create);
		}

		mark_btree_node_locked(trans, path, 0, BTREE_NODE_INTENT_LOCKED);
	}

	ck->c.level		= 0;
	ck->c.btree_id		= path->btree_id;
	ck->key.btree_id	= path->btree_id;
	ck->key.pos		= path->pos;
	ck->valid		= false;
	ck->flags		= 1U << BKEY_CACHED_ACCESSED;

	if (unlikely(rhashtable_lookup_insert_fast(&bc->table,
					  &ck->hash,
					  bch2_btree_key_cache_params))) {
		/* We raced with another fill: */

		if (likely(was_new)) {
			six_unlock_write(&ck->c.lock);
			six_unlock_intent(&ck->c.lock);
			kfree(ck);
		} else {
			bkey_cached_free_fast(bc, ck);
		}

		mark_btree_node_locked(trans, path, 0, BTREE_NODE_UNLOCKED);
		return NULL;
	}

	atomic_long_inc(&bc->nr_keys);

	six_unlock_write(&ck->c.lock);

	return ck;
}

static int btree_key_cache_fill(struct btree_trans *trans,
				struct btree_path *ck_path,
				struct bkey_cached *ck)
{
	struct btree_iter iter;
	struct bkey_s_c k;
	unsigned new_u64s = 0;
	struct bkey_i *new_k = NULL;
	int ret;

	k = bch2_bkey_get_iter(trans, &iter, ck->key.btree_id, ck->key.pos,
			       BTREE_ITER_KEY_CACHE_FILL|
			       BTREE_ITER_CACHED_NOFILL);
	ret = bkey_err(k);
	if (ret)
		goto err;

	if (!bch2_btree_node_relock(trans, ck_path, 0)) {
		trace_and_count(trans->c, trans_restart_relock_key_cache_fill, trans, _THIS_IP_, ck_path);
		ret = btree_trans_restart(trans, BCH_ERR_transaction_restart_key_cache_fill);
		goto err;
	}

	/*
	 * bch2_varint_decode can read past the end of the buffer by at
	 * most 7 bytes (it won't be used):
	 */
	new_u64s = k.k->u64s + 1;

	/*
	 * Allocate some extra space so that the transaction commit path is less
	 * likely to have to reallocate, since that requires a transaction
	 * restart:
	 */
	new_u64s = min(256U, (new_u64s * 3) / 2);

	if (new_u64s > ck->u64s) {
		new_u64s = roundup_pow_of_two(new_u64s);
		new_k = kmalloc(new_u64s * sizeof(u64), GFP_NOWAIT|__GFP_NOWARN);
		if (!new_k) {
			bch2_trans_unlock(trans);

			new_k = kmalloc(new_u64s * sizeof(u64), GFP_KERNEL);
			if (!new_k) {
				bch_err(trans->c, "error allocating memory for key cache key, btree %s u64s %u",
					bch2_btree_id_str(ck->key.btree_id), new_u64s);
				ret = -BCH_ERR_ENOMEM_btree_key_cache_fill;
				goto err;
			}

			if (!bch2_btree_node_relock(trans, ck_path, 0)) {
				kfree(new_k);
				trace_and_count(trans->c, trans_restart_relock_key_cache_fill, trans, _THIS_IP_, ck_path);
				ret = btree_trans_restart(trans, BCH_ERR_transaction_restart_key_cache_fill);
				goto err;
			}

			ret = bch2_trans_relock(trans);
			if (ret) {
				kfree(new_k);
				goto err;
			}
		}
	}

	ret = bch2_btree_node_lock_write(trans, ck_path, &ck_path->l[0].b->c);
	if (ret) {
		kfree(new_k);
		goto err;
	}

	if (new_k) {
		kfree(ck->k);
		ck->u64s = new_u64s;
		ck->k = new_k;
	}

	bkey_reassemble(ck->k, k);
	ck->valid = true;
	bch2_btree_node_unlock_write(trans, ck_path, ck_path->l[0].b);

	/* We're not likely to need this iterator again: */
	set_btree_iter_dontneed(&iter);
err:
	bch2_trans_iter_exit(trans, &iter);
	return ret;
}

static noinline int
bch2_btree_path_traverse_cached_slowpath(struct btree_trans *trans, struct btree_path *path,
					 unsigned flags)
{
	struct bch_fs *c = trans->c;
	struct bkey_cached *ck;
	int ret = 0;

	BUG_ON(path->level);

	path->l[1].b = NULL;

	if (bch2_btree_node_relock_notrace(trans, path, 0)) {
		ck = (void *) path->l[0].b;
		goto fill;
	}
retry:
	ck = bch2_btree_key_cache_find(c, path->btree_id, path->pos);
	if (!ck) {
		ck = btree_key_cache_create(trans, path);
		ret = PTR_ERR_OR_ZERO(ck);
		if (ret)
			goto err;
		if (!ck)
			goto retry;

		mark_btree_node_locked(trans, path, 0, BTREE_NODE_INTENT_LOCKED);
		path->locks_want = 1;
	} else {
		enum six_lock_type lock_want = __btree_lock_want(path, 0);

		ret = btree_node_lock(trans, path, (void *) ck, 0,
				      lock_want, _THIS_IP_);
		if (bch2_err_matches(ret, BCH_ERR_transaction_restart))
			goto err;

		BUG_ON(ret);

		if (ck->key.btree_id != path->btree_id ||
		    !bpos_eq(ck->key.pos, path->pos)) {
			six_unlock_type(&ck->c.lock, lock_want);
			goto retry;
		}

		mark_btree_node_locked(trans, path, 0,
				       (enum btree_node_locked_type) lock_want);
	}

	path->l[0].lock_seq	= six_lock_seq(&ck->c.lock);
	path->l[0].b		= (void *) ck;
fill:
	path->uptodate = BTREE_ITER_UPTODATE;

	if (!ck->valid && !(flags & BTREE_ITER_CACHED_NOFILL)) {
		/*
		 * Using the underscore version because we haven't set
		 * path->uptodate yet:
		 */
		if (!path->locks_want &&
		    !__bch2_btree_path_upgrade(trans, path, 1, NULL)) {
			trace_and_count(trans->c, trans_restart_key_cache_upgrade, trans, _THIS_IP_);
			ret = btree_trans_restart(trans, BCH_ERR_transaction_restart_key_cache_upgrade);
			goto err;
		}

		ret = btree_key_cache_fill(trans, path, ck);
		if (ret)
			goto err;

		ret = bch2_btree_path_relock(trans, path, _THIS_IP_);
		if (ret)
			goto err;

		path->uptodate = BTREE_ITER_UPTODATE;
	}

	if (!test_bit(BKEY_CACHED_ACCESSED, &ck->flags))
		set_bit(BKEY_CACHED_ACCESSED, &ck->flags);

	BUG_ON(btree_node_locked_type(path, 0) != btree_lock_want(path, 0));
	BUG_ON(path->uptodate);

	return ret;
err:
	path->uptodate = BTREE_ITER_NEED_TRAVERSE;
	if (!bch2_err_matches(ret, BCH_ERR_transaction_restart)) {
		btree_node_unlock(trans, path, 0);
		path->l[0].b = ERR_PTR(ret);
	}
	return ret;
}

int bch2_btree_path_traverse_cached(struct btree_trans *trans, struct btree_path *path,
				    unsigned flags)
{
	struct bch_fs *c = trans->c;
	struct bkey_cached *ck;
	int ret = 0;

	EBUG_ON(path->level);

	path->l[1].b = NULL;

	if (bch2_btree_node_relock_notrace(trans, path, 0)) {
		ck = (void *) path->l[0].b;
		goto fill;
	}
retry:
	ck = bch2_btree_key_cache_find(c, path->btree_id, path->pos);
	if (!ck) {
		return bch2_btree_path_traverse_cached_slowpath(trans, path, flags);
	} else {
		enum six_lock_type lock_want = __btree_lock_want(path, 0);

		ret = btree_node_lock(trans, path, (void *) ck, 0,
				      lock_want, _THIS_IP_);
		EBUG_ON(ret && !bch2_err_matches(ret, BCH_ERR_transaction_restart));

		if (ret)
			return ret;

		if (ck->key.btree_id != path->btree_id ||
		    !bpos_eq(ck->key.pos, path->pos)) {
			six_unlock_type(&ck->c.lock, lock_want);
			goto retry;
		}

		mark_btree_node_locked(trans, path, 0,
				       (enum btree_node_locked_type) lock_want);
	}

	path->l[0].lock_seq	= six_lock_seq(&ck->c.lock);
	path->l[0].b		= (void *) ck;
fill:
	if (!ck->valid)
		return bch2_btree_path_traverse_cached_slowpath(trans, path, flags);

	if (!test_bit(BKEY_CACHED_ACCESSED, &ck->flags))
		set_bit(BKEY_CACHED_ACCESSED, &ck->flags);

	path->uptodate = BTREE_ITER_UPTODATE;
	EBUG_ON(!ck->valid);
	EBUG_ON(btree_node_locked_type(path, 0) != btree_lock_want(path, 0));

	return ret;
}

static int btree_key_cache_flush_pos(struct btree_trans *trans,
				     struct bkey_cached_key key,
				     u64 journal_seq,
				     unsigned commit_flags,
				     bool evict)
{
	struct bch_fs *c = trans->c;
	struct journal *j = &c->journal;
	struct btree_iter c_iter, b_iter;
	struct bkey_cached *ck = NULL;
	int ret;

	bch2_trans_iter_init(trans, &b_iter, key.btree_id, key.pos,
			     BTREE_ITER_SLOTS|
			     BTREE_ITER_INTENT|
			     BTREE_ITER_ALL_SNAPSHOTS);
	bch2_trans_iter_init(trans, &c_iter, key.btree_id, key.pos,
			     BTREE_ITER_CACHED|
			     BTREE_ITER_INTENT);
	b_iter.flags &= ~BTREE_ITER_WITH_KEY_CACHE;

	ret = bch2_btree_iter_traverse(&c_iter);
	if (ret)
		goto out;

	ck = (void *) c_iter.path->l[0].b;
	if (!ck)
		goto out;

	if (!test_bit(BKEY_CACHED_DIRTY, &ck->flags)) {
		if (evict)
			goto evict;
		goto out;
	}

	BUG_ON(!ck->valid);

	if (journal_seq && ck->journal.seq != journal_seq)
		goto out;

	/*
	 * Since journal reclaim depends on us making progress here, and the
	 * allocator/copygc depend on journal reclaim making progress, we need
	 * to be using alloc reserves:
	 */
	ret   = bch2_btree_iter_traverse(&b_iter) ?:
		bch2_trans_update(trans, &b_iter, ck->k,
				  BTREE_UPDATE_KEY_CACHE_RECLAIM|
				  BTREE_UPDATE_INTERNAL_SNAPSHOT_NODE|
				  BTREE_TRIGGER_NORUN) ?:
		bch2_trans_commit(trans, NULL, NULL,
				  BTREE_INSERT_NOCHECK_RW|
				  BTREE_INSERT_NOFAIL|
				  (ck->journal.seq == journal_last_seq(j)
				   ? BCH_WATERMARK_reclaim
				   : 0)|
				  commit_flags);

	bch2_fs_fatal_err_on(ret &&
			     !bch2_err_matches(ret, BCH_ERR_transaction_restart) &&
			     !bch2_err_matches(ret, BCH_ERR_journal_reclaim_would_deadlock) &&
			     !bch2_journal_error(j), c,
			     "error flushing key cache: %s", bch2_err_str(ret));
	if (ret)
		goto out;

	bch2_journal_pin_drop(j, &ck->journal);

	BUG_ON(!btree_node_locked(c_iter.path, 0));

	if (!evict) {
		if (test_bit(BKEY_CACHED_DIRTY, &ck->flags)) {
			clear_bit(BKEY_CACHED_DIRTY, &ck->flags);
			atomic_long_dec(&c->btree_key_cache.nr_dirty);
		}
	} else {
		struct btree_path *path2;
evict:
		trans_for_each_path(trans, path2)
			if (path2 != c_iter.path)
				__bch2_btree_path_unlock(trans, path2);

		bch2_btree_node_lock_write_nofail(trans, c_iter.path, &ck->c);

		if (test_bit(BKEY_CACHED_DIRTY, &ck->flags)) {
			clear_bit(BKEY_CACHED_DIRTY, &ck->flags);
			atomic_long_dec(&c->btree_key_cache.nr_dirty);
		}

		mark_btree_node_locked_noreset(c_iter.path, 0, BTREE_NODE_UNLOCKED);
		bkey_cached_evict(&c->btree_key_cache, ck);
		bkey_cached_free_fast(&c->btree_key_cache, ck);
	}
out:
	bch2_trans_iter_exit(trans, &b_iter);
	bch2_trans_iter_exit(trans, &c_iter);
	return ret;
}

int bch2_btree_key_cache_journal_flush(struct journal *j,
				struct journal_entry_pin *pin, u64 seq)
{
	struct bch_fs *c = container_of(j, struct bch_fs, journal);
	struct bkey_cached *ck =
		container_of(pin, struct bkey_cached, journal);
	struct bkey_cached_key key;
	struct btree_trans *trans = bch2_trans_get(c);
	int srcu_idx = srcu_read_lock(&c->btree_trans_barrier);
	int ret = 0;

	btree_node_lock_nopath_nofail(trans, &ck->c, SIX_LOCK_read);
	key = ck->key;

	if (ck->journal.seq != seq ||
	    !test_bit(BKEY_CACHED_DIRTY, &ck->flags)) {
		six_unlock_read(&ck->c.lock);
		goto unlock;
	}

	if (ck->seq != seq) {
		bch2_journal_pin_update(&c->journal, ck->seq, &ck->journal,
					bch2_btree_key_cache_journal_flush);
		six_unlock_read(&ck->c.lock);
		goto unlock;
	}
	six_unlock_read(&ck->c.lock);

	ret = commit_do(trans, NULL, NULL, 0,
		btree_key_cache_flush_pos(trans, key, seq,
				BTREE_INSERT_JOURNAL_RECLAIM, false));
unlock:
	srcu_read_unlock(&c->btree_trans_barrier, srcu_idx);

	bch2_trans_put(trans);
	return ret;
}

/*
 * Flush and evict a key from the key cache:
 */
int bch2_btree_key_cache_flush(struct btree_trans *trans,
			       enum btree_id id, struct bpos pos)
{
	struct bch_fs *c = trans->c;
	struct bkey_cached_key key = { id, pos };

	/* Fastpath - assume it won't be found: */
	if (!bch2_btree_key_cache_find(c, id, pos))
		return 0;

	return btree_key_cache_flush_pos(trans, key, 0, 0, true);
}

bool bch2_btree_insert_key_cached(struct btree_trans *trans,
				  unsigned flags,
				  struct btree_insert_entry *insert_entry)
{
	struct bch_fs *c = trans->c;
	struct bkey_cached *ck = (void *) insert_entry->path->l[0].b;
	struct bkey_i *insert = insert_entry->k;
	bool kick_reclaim = false;

	BUG_ON(insert->k.u64s > ck->u64s);

	bkey_copy(ck->k, insert);
	ck->valid = true;

	if (!test_bit(BKEY_CACHED_DIRTY, &ck->flags)) {
		EBUG_ON(test_bit(BCH_FS_CLEAN_SHUTDOWN, &c->flags));
		set_bit(BKEY_CACHED_DIRTY, &ck->flags);
		atomic_long_inc(&c->btree_key_cache.nr_dirty);

		if (bch2_nr_btree_keys_need_flush(c))
			kick_reclaim = true;
	}

	/*
	 * To minimize lock contention, we only add the journal pin here and
	 * defer pin updates to the flush callback via ->seq. Be careful not to
	 * update ->seq on nojournal commits because we don't want to update the
	 * pin to a seq that doesn't include journal updates on disk. Otherwise
	 * we risk losing the update after a crash.
	 *
	 * The only exception is if the pin is not active in the first place. We
	 * have to add the pin because journal reclaim drives key cache
	 * flushing. The flush callback will not proceed unless ->seq matches
	 * the latest pin, so make sure it starts with a consistent value.
	 */
	if (!(insert_entry->flags & BTREE_UPDATE_NOJOURNAL) ||
	    !journal_pin_active(&ck->journal)) {
		ck->seq = trans->journal_res.seq;
	}
	bch2_journal_pin_add(&c->journal, trans->journal_res.seq,
			     &ck->journal, bch2_btree_key_cache_journal_flush);

	if (kick_reclaim)
		journal_reclaim_kick(&c->journal);
	return true;
}

void bch2_btree_key_cache_drop(struct btree_trans *trans,
			       struct btree_path *path)
{
	struct bch_fs *c = trans->c;
	struct bkey_cached *ck = (void *) path->l[0].b;

	BUG_ON(!ck->valid);

	/*
	 * We just did an update to the btree, bypassing the key cache: the key
	 * cache key is now stale and must be dropped, even if dirty:
	 */
	if (test_bit(BKEY_CACHED_DIRTY, &ck->flags)) {
		clear_bit(BKEY_CACHED_DIRTY, &ck->flags);
		atomic_long_dec(&c->btree_key_cache.nr_dirty);
		bch2_journal_pin_drop(&c->journal, &ck->journal);
	}

	ck->valid = false;
}

static unsigned long bch2_btree_key_cache_scan(struct shrinker *shrink,
					   struct shrink_control *sc)
{
	struct bch_fs *c = shrink->private_data;
	struct btree_key_cache *bc = &c->btree_key_cache;
	struct bucket_table *tbl;
	struct bkey_cached *ck, *t;
	size_t scanned = 0, freed = 0, nr = sc->nr_to_scan;
	unsigned start, flags;
	int srcu_idx;

	mutex_lock(&bc->lock);
	srcu_idx = srcu_read_lock(&c->btree_trans_barrier);
	flags = memalloc_nofs_save();

	/*
	 * Newest freed entries are at the end of the list - once we hit one
	 * that's too new to be freed, we can bail out:
	 */
	scanned += bc->nr_freed_nonpcpu;

	list_for_each_entry_safe(ck, t, &bc->freed_nonpcpu, list) {
		if (!poll_state_synchronize_srcu(&c->btree_trans_barrier,
						 ck->btree_trans_barrier_seq))
			break;

		list_del(&ck->list);
		six_lock_exit(&ck->c.lock);
		kmem_cache_free(bch2_key_cache, ck);
		atomic_long_dec(&bc->nr_freed);
		freed++;
		bc->nr_freed_nonpcpu--;
	}

	if (scanned >= nr)
		goto out;

	scanned += bc->nr_freed_pcpu;

	list_for_each_entry_safe(ck, t, &bc->freed_pcpu, list) {
		if (!poll_state_synchronize_srcu(&c->btree_trans_barrier,
						 ck->btree_trans_barrier_seq))
			break;

		list_del(&ck->list);
		six_lock_exit(&ck->c.lock);
		kmem_cache_free(bch2_key_cache, ck);
		atomic_long_dec(&bc->nr_freed);
		freed++;
		bc->nr_freed_pcpu--;
	}

	if (scanned >= nr)
		goto out;

	rcu_read_lock();
	tbl = rht_dereference_rcu(bc->table.tbl, &bc->table);
	if (bc->shrink_iter >= tbl->size)
		bc->shrink_iter = 0;
	start = bc->shrink_iter;

	do {
		struct rhash_head *pos, *next;

		pos = rht_ptr_rcu(rht_bucket(tbl, bc->shrink_iter));

		while (!rht_is_a_nulls(pos)) {
			next = rht_dereference_bucket_rcu(pos->next, tbl, bc->shrink_iter);
			ck = container_of(pos, struct bkey_cached, hash);

			if (test_bit(BKEY_CACHED_DIRTY, &ck->flags))
				goto next;

			if (test_bit(BKEY_CACHED_ACCESSED, &ck->flags))
				clear_bit(BKEY_CACHED_ACCESSED, &ck->flags);
			else if (bkey_cached_lock_for_evict(ck)) {
				bkey_cached_evict(bc, ck);
				bkey_cached_free(bc, ck);
			}

			scanned++;
			if (scanned >= nr)
				break;
next:
			pos = next;
		}

		bc->shrink_iter++;
		if (bc->shrink_iter >= tbl->size)
			bc->shrink_iter = 0;
	} while (scanned < nr && bc->shrink_iter != start);

	rcu_read_unlock();
out:
	memalloc_nofs_restore(flags);
	srcu_read_unlock(&c->btree_trans_barrier, srcu_idx);
	mutex_unlock(&bc->lock);

	return freed;
}

static unsigned long bch2_btree_key_cache_count(struct shrinker *shrink,
					    struct shrink_control *sc)
{
	struct bch_fs *c = shrink->private_data;
	struct btree_key_cache *bc = &c->btree_key_cache;
	long nr = atomic_long_read(&bc->nr_keys) -
		atomic_long_read(&bc->nr_dirty);

	return max(0L, nr);
}

void bch2_fs_btree_key_cache_exit(struct btree_key_cache *bc)
{
	struct bch_fs *c = container_of(bc, struct bch_fs, btree_key_cache);
	struct bucket_table *tbl;
	struct bkey_cached *ck, *n;
	struct rhash_head *pos;
	LIST_HEAD(items);
	unsigned i;
#ifdef __KERNEL__
	int cpu;
#endif

	shrinker_free(bc->shrink);

	mutex_lock(&bc->lock);

	/*
	 * The loop is needed to guard against racing with rehash:
	 */
	while (atomic_long_read(&bc->nr_keys)) {
		rcu_read_lock();
		tbl = rht_dereference_rcu(bc->table.tbl, &bc->table);
		if (tbl)
			for (i = 0; i < tbl->size; i++)
				rht_for_each_entry_rcu(ck, pos, tbl, i, hash) {
					bkey_cached_evict(bc, ck);
					list_add(&ck->list, &items);
				}
		rcu_read_unlock();
	}

#ifdef __KERNEL__
	for_each_possible_cpu(cpu) {
		struct btree_key_cache_freelist *f =
			per_cpu_ptr(bc->pcpu_freed, cpu);

		for (i = 0; i < f->nr; i++) {
			ck = f->objs[i];
			list_add(&ck->list, &items);
		}
	}
#endif

	BUG_ON(list_count_nodes(&bc->freed_pcpu) != bc->nr_freed_pcpu);
	BUG_ON(list_count_nodes(&bc->freed_nonpcpu) != bc->nr_freed_nonpcpu);

	list_splice(&bc->freed_pcpu,	&items);
	list_splice(&bc->freed_nonpcpu,	&items);

	mutex_unlock(&bc->lock);

	list_for_each_entry_safe(ck, n, &items, list) {
		cond_resched();

		list_del(&ck->list);
		kfree(ck->k);
		six_lock_exit(&ck->c.lock);
		kmem_cache_free(bch2_key_cache, ck);
	}

	if (atomic_long_read(&bc->nr_dirty) &&
	    !bch2_journal_error(&c->journal) &&
	    test_bit(BCH_FS_WAS_RW, &c->flags))
		panic("btree key cache shutdown error: nr_dirty nonzero (%li)\n",
		      atomic_long_read(&bc->nr_dirty));

	if (atomic_long_read(&bc->nr_keys))
		panic("btree key cache shutdown error: nr_keys nonzero (%li)\n",
		      atomic_long_read(&bc->nr_keys));

	if (bc->table_init_done)
		rhashtable_destroy(&bc->table);

	free_percpu(bc->pcpu_freed);
}

void bch2_fs_btree_key_cache_init_early(struct btree_key_cache *c)
{
	mutex_init(&c->lock);
	INIT_LIST_HEAD(&c->freed_pcpu);
	INIT_LIST_HEAD(&c->freed_nonpcpu);
}

int bch2_fs_btree_key_cache_init(struct btree_key_cache *bc)
{
	struct bch_fs *c = container_of(bc, struct bch_fs, btree_key_cache);
	struct shrinker *shrink;

#ifdef __KERNEL__
	bc->pcpu_freed = alloc_percpu(struct btree_key_cache_freelist);
	if (!bc->pcpu_freed)
		return -BCH_ERR_ENOMEM_fs_btree_cache_init;
#endif

	if (rhashtable_init(&bc->table, &bch2_btree_key_cache_params))
		return -BCH_ERR_ENOMEM_fs_btree_cache_init;

	bc->table_init_done = true;

	shrink = shrinker_alloc(0, "%s-btree_key_cache", c->name);
	if (!shrink)
		return -BCH_ERR_ENOMEM_fs_btree_cache_init;
	bc->shrink = shrink;
	shrink->seeks		= 0;
	shrink->count_objects	= bch2_btree_key_cache_count;
	shrink->scan_objects	= bch2_btree_key_cache_scan;
	shrink->private_data	= c;
	shrinker_register(shrink);
	return 0;
}

void bch2_btree_key_cache_to_text(struct printbuf *out, struct btree_key_cache *c)
{
	prt_printf(out, "nr_freed:\t%lu",	atomic_long_read(&c->nr_freed));
	prt_newline(out);
	prt_printf(out, "nr_keys:\t%lu",	atomic_long_read(&c->nr_keys));
	prt_newline(out);
	prt_printf(out, "nr_dirty:\t%lu",	atomic_long_read(&c->nr_dirty));
	prt_newline(out);
}

void bch2_btree_key_cache_exit(void)
{
	kmem_cache_destroy(bch2_key_cache);
}

int __init bch2_btree_key_cache_init(void)
{
	bch2_key_cache = KMEM_CACHE(bkey_cached, SLAB_RECLAIM_ACCOUNT);
	if (!bch2_key_cache)
		return -ENOMEM;

	return 0;
}