This source file includes following definitions.
- bch_inc_gen
- bch_rescale_priorities
- can_inc_bucket_gen
- bch_can_invalidate_bucket
- __bch_invalidate_one_bucket
- bch_invalidate_one_bucket
- invalidate_buckets_lru
- invalidate_buckets_fifo
- invalidate_buckets_random
- invalidate_buckets
- bch_allocator_push
- bch_allocator_thread
- bch_bucket_alloc
- __bch_bucket_free
- bch_bucket_free
- __bch_bucket_alloc_set
- bch_bucket_alloc_set
- pick_data_bucket
- bch_alloc_sectors
- bch_open_buckets_free
- bch_open_buckets_alloc
- bch_cache_allocator_start
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 #include "bcache.h"
65 #include "btree.h"
66
67 #include <linux/blkdev.h>
68 #include <linux/kthread.h>
69 #include <linux/random.h>
70 #include <trace/events/bcache.h>
71
72 #define MAX_OPEN_BUCKETS 128
73
74
75
76 uint8_t bch_inc_gen(struct cache *ca, struct bucket *b)
77 {
78 uint8_t ret = ++b->gen;
79
80 ca->set->need_gc = max(ca->set->need_gc, bucket_gc_gen(b));
81 WARN_ON_ONCE(ca->set->need_gc > BUCKET_GC_GEN_MAX);
82
83 return ret;
84 }
85
86 void bch_rescale_priorities(struct cache_set *c, int sectors)
87 {
88 struct cache *ca;
89 struct bucket *b;
90 unsigned int next = c->nbuckets * c->sb.bucket_size / 1024;
91 unsigned int i;
92 int r;
93
94 atomic_sub(sectors, &c->rescale);
95
96 do {
97 r = atomic_read(&c->rescale);
98
99 if (r >= 0)
100 return;
101 } while (atomic_cmpxchg(&c->rescale, r, r + next) != r);
102
103 mutex_lock(&c->bucket_lock);
104
105 c->min_prio = USHRT_MAX;
106
107 for_each_cache(ca, c, i)
108 for_each_bucket(b, ca)
109 if (b->prio &&
110 b->prio != BTREE_PRIO &&
111 !atomic_read(&b->pin)) {
112 b->prio--;
113 c->min_prio = min(c->min_prio, b->prio);
114 }
115
116 mutex_unlock(&c->bucket_lock);
117 }
118
119
120
121
122
123
124
125
126 static inline bool can_inc_bucket_gen(struct bucket *b)
127 {
128 return bucket_gc_gen(b) < BUCKET_GC_GEN_MAX;
129 }
130
131 bool bch_can_invalidate_bucket(struct cache *ca, struct bucket *b)
132 {
133 BUG_ON(!ca->set->gc_mark_valid);
134
135 return (!GC_MARK(b) ||
136 GC_MARK(b) == GC_MARK_RECLAIMABLE) &&
137 !atomic_read(&b->pin) &&
138 can_inc_bucket_gen(b);
139 }
140
141 void __bch_invalidate_one_bucket(struct cache *ca, struct bucket *b)
142 {
143 lockdep_assert_held(&ca->set->bucket_lock);
144 BUG_ON(GC_MARK(b) && GC_MARK(b) != GC_MARK_RECLAIMABLE);
145
146 if (GC_SECTORS_USED(b))
147 trace_bcache_invalidate(ca, b - ca->buckets);
148
149 bch_inc_gen(ca, b);
150 b->prio = INITIAL_PRIO;
151 atomic_inc(&b->pin);
152 }
153
154 static void bch_invalidate_one_bucket(struct cache *ca, struct bucket *b)
155 {
156 __bch_invalidate_one_bucket(ca, b);
157
158 fifo_push(&ca->free_inc, b - ca->buckets);
159 }
160
161
162
163
164
165
166
167
168
169
170 #define bucket_prio(b) \
171 ({ \
172 unsigned int min_prio = (INITIAL_PRIO - ca->set->min_prio) / 8; \
173 \
174 (b->prio - ca->set->min_prio + min_prio) * GC_SECTORS_USED(b); \
175 })
176
177 #define bucket_max_cmp(l, r) (bucket_prio(l) < bucket_prio(r))
178 #define bucket_min_cmp(l, r) (bucket_prio(l) > bucket_prio(r))
179
180 static void invalidate_buckets_lru(struct cache *ca)
181 {
182 struct bucket *b;
183 ssize_t i;
184
185 ca->heap.used = 0;
186
187 for_each_bucket(b, ca) {
188 if (!bch_can_invalidate_bucket(ca, b))
189 continue;
190
191 if (!heap_full(&ca->heap))
192 heap_add(&ca->heap, b, bucket_max_cmp);
193 else if (bucket_max_cmp(b, heap_peek(&ca->heap))) {
194 ca->heap.data[0] = b;
195 heap_sift(&ca->heap, 0, bucket_max_cmp);
196 }
197 }
198
199 for (i = ca->heap.used / 2 - 1; i >= 0; --i)
200 heap_sift(&ca->heap, i, bucket_min_cmp);
201
202 while (!fifo_full(&ca->free_inc)) {
203 if (!heap_pop(&ca->heap, b, bucket_min_cmp)) {
204
205
206
207
208 ca->invalidate_needs_gc = 1;
209 wake_up_gc(ca->set);
210 return;
211 }
212
213 bch_invalidate_one_bucket(ca, b);
214 }
215 }
216
217 static void invalidate_buckets_fifo(struct cache *ca)
218 {
219 struct bucket *b;
220 size_t checked = 0;
221
222 while (!fifo_full(&ca->free_inc)) {
223 if (ca->fifo_last_bucket < ca->sb.first_bucket ||
224 ca->fifo_last_bucket >= ca->sb.nbuckets)
225 ca->fifo_last_bucket = ca->sb.first_bucket;
226
227 b = ca->buckets + ca->fifo_last_bucket++;
228
229 if (bch_can_invalidate_bucket(ca, b))
230 bch_invalidate_one_bucket(ca, b);
231
232 if (++checked >= ca->sb.nbuckets) {
233 ca->invalidate_needs_gc = 1;
234 wake_up_gc(ca->set);
235 return;
236 }
237 }
238 }
239
240 static void invalidate_buckets_random(struct cache *ca)
241 {
242 struct bucket *b;
243 size_t checked = 0;
244
245 while (!fifo_full(&ca->free_inc)) {
246 size_t n;
247
248 get_random_bytes(&n, sizeof(n));
249
250 n %= (size_t) (ca->sb.nbuckets - ca->sb.first_bucket);
251 n += ca->sb.first_bucket;
252
253 b = ca->buckets + n;
254
255 if (bch_can_invalidate_bucket(ca, b))
256 bch_invalidate_one_bucket(ca, b);
257
258 if (++checked >= ca->sb.nbuckets / 2) {
259 ca->invalidate_needs_gc = 1;
260 wake_up_gc(ca->set);
261 return;
262 }
263 }
264 }
265
266 static void invalidate_buckets(struct cache *ca)
267 {
268 BUG_ON(ca->invalidate_needs_gc);
269
270 switch (CACHE_REPLACEMENT(&ca->sb)) {
271 case CACHE_REPLACEMENT_LRU:
272 invalidate_buckets_lru(ca);
273 break;
274 case CACHE_REPLACEMENT_FIFO:
275 invalidate_buckets_fifo(ca);
276 break;
277 case CACHE_REPLACEMENT_RANDOM:
278 invalidate_buckets_random(ca);
279 break;
280 }
281 }
282
283 #define allocator_wait(ca, cond) \
284 do { \
285 while (1) { \
286 set_current_state(TASK_INTERRUPTIBLE); \
287 if (cond) \
288 break; \
289 \
290 mutex_unlock(&(ca)->set->bucket_lock); \
291 if (kthread_should_stop() || \
292 test_bit(CACHE_SET_IO_DISABLE, &ca->set->flags)) { \
293 set_current_state(TASK_RUNNING); \
294 goto out; \
295 } \
296 \
297 schedule(); \
298 mutex_lock(&(ca)->set->bucket_lock); \
299 } \
300 __set_current_state(TASK_RUNNING); \
301 } while (0)
302
303 static int bch_allocator_push(struct cache *ca, long bucket)
304 {
305 unsigned int i;
306
307
308 if (fifo_push(&ca->free[RESERVE_PRIO], bucket))
309 return true;
310
311 for (i = 0; i < RESERVE_NR; i++)
312 if (fifo_push(&ca->free[i], bucket))
313 return true;
314
315 return false;
316 }
317
318 static int bch_allocator_thread(void *arg)
319 {
320 struct cache *ca = arg;
321
322 mutex_lock(&ca->set->bucket_lock);
323
324 while (1) {
325
326
327
328
329
330 while (1) {
331 long bucket;
332
333 if (!fifo_pop(&ca->free_inc, bucket))
334 break;
335
336 if (ca->discard) {
337 mutex_unlock(&ca->set->bucket_lock);
338 blkdev_issue_discard(ca->bdev,
339 bucket_to_sector(ca->set, bucket),
340 ca->sb.bucket_size, GFP_KERNEL, 0);
341 mutex_lock(&ca->set->bucket_lock);
342 }
343
344 allocator_wait(ca, bch_allocator_push(ca, bucket));
345 wake_up(&ca->set->btree_cache_wait);
346 wake_up(&ca->set->bucket_wait);
347 }
348
349
350
351
352
353
354
355 retry_invalidate:
356 allocator_wait(ca, ca->set->gc_mark_valid &&
357 !ca->invalidate_needs_gc);
358 invalidate_buckets(ca);
359
360
361
362
363
364 allocator_wait(ca, !atomic_read(&ca->set->prio_blocked));
365 if (CACHE_SYNC(&ca->set->sb)) {
366
367
368
369
370
371
372
373
374
375
376
377 if (!fifo_full(&ca->free_inc))
378 goto retry_invalidate;
379
380 if (bch_prio_write(ca, false) < 0) {
381 ca->invalidate_needs_gc = 1;
382 wake_up_gc(ca->set);
383 }
384 }
385 }
386 out:
387 wait_for_kthread_stop();
388 return 0;
389 }
390
391
392
393 long bch_bucket_alloc(struct cache *ca, unsigned int reserve, bool wait)
394 {
395 DEFINE_WAIT(w);
396 struct bucket *b;
397 long r;
398
399
400
401 if (unlikely(test_bit(CACHE_SET_IO_DISABLE, &ca->set->flags)))
402 return -1;
403
404
405 if (fifo_pop(&ca->free[RESERVE_NONE], r) ||
406 fifo_pop(&ca->free[reserve], r))
407 goto out;
408
409 if (!wait) {
410 trace_bcache_alloc_fail(ca, reserve);
411 return -1;
412 }
413
414 do {
415 prepare_to_wait(&ca->set->bucket_wait, &w,
416 TASK_UNINTERRUPTIBLE);
417
418 mutex_unlock(&ca->set->bucket_lock);
419 schedule();
420 mutex_lock(&ca->set->bucket_lock);
421 } while (!fifo_pop(&ca->free[RESERVE_NONE], r) &&
422 !fifo_pop(&ca->free[reserve], r));
423
424 finish_wait(&ca->set->bucket_wait, &w);
425 out:
426 if (ca->alloc_thread)
427 wake_up_process(ca->alloc_thread);
428
429 trace_bcache_alloc(ca, reserve);
430
431 if (expensive_debug_checks(ca->set)) {
432 size_t iter;
433 long i;
434 unsigned int j;
435
436 for (iter = 0; iter < prio_buckets(ca) * 2; iter++)
437 BUG_ON(ca->prio_buckets[iter] == (uint64_t) r);
438
439 for (j = 0; j < RESERVE_NR; j++)
440 fifo_for_each(i, &ca->free[j], iter)
441 BUG_ON(i == r);
442 fifo_for_each(i, &ca->free_inc, iter)
443 BUG_ON(i == r);
444 }
445
446 b = ca->buckets + r;
447
448 BUG_ON(atomic_read(&b->pin) != 1);
449
450 SET_GC_SECTORS_USED(b, ca->sb.bucket_size);
451
452 if (reserve <= RESERVE_PRIO) {
453 SET_GC_MARK(b, GC_MARK_METADATA);
454 SET_GC_MOVE(b, 0);
455 b->prio = BTREE_PRIO;
456 } else {
457 SET_GC_MARK(b, GC_MARK_RECLAIMABLE);
458 SET_GC_MOVE(b, 0);
459 b->prio = INITIAL_PRIO;
460 }
461
462 if (ca->set->avail_nbuckets > 0) {
463 ca->set->avail_nbuckets--;
464 bch_update_bucket_in_use(ca->set, &ca->set->gc_stats);
465 }
466
467 return r;
468 }
469
470 void __bch_bucket_free(struct cache *ca, struct bucket *b)
471 {
472 SET_GC_MARK(b, 0);
473 SET_GC_SECTORS_USED(b, 0);
474
475 if (ca->set->avail_nbuckets < ca->set->nbuckets) {
476 ca->set->avail_nbuckets++;
477 bch_update_bucket_in_use(ca->set, &ca->set->gc_stats);
478 }
479 }
480
481 void bch_bucket_free(struct cache_set *c, struct bkey *k)
482 {
483 unsigned int i;
484
485 for (i = 0; i < KEY_PTRS(k); i++)
486 __bch_bucket_free(PTR_CACHE(c, k, i),
487 PTR_BUCKET(c, k, i));
488 }
489
490 int __bch_bucket_alloc_set(struct cache_set *c, unsigned int reserve,
491 struct bkey *k, int n, bool wait)
492 {
493 int i;
494
495
496 if (unlikely(test_bit(CACHE_SET_IO_DISABLE, &c->flags)))
497 return -1;
498
499 lockdep_assert_held(&c->bucket_lock);
500 BUG_ON(!n || n > c->caches_loaded || n > MAX_CACHES_PER_SET);
501
502 bkey_init(k);
503
504
505
506 for (i = 0; i < n; i++) {
507 struct cache *ca = c->cache_by_alloc[i];
508 long b = bch_bucket_alloc(ca, reserve, wait);
509
510 if (b == -1)
511 goto err;
512
513 k->ptr[i] = MAKE_PTR(ca->buckets[b].gen,
514 bucket_to_sector(c, b),
515 ca->sb.nr_this_dev);
516
517 SET_KEY_PTRS(k, i + 1);
518 }
519
520 return 0;
521 err:
522 bch_bucket_free(c, k);
523 bkey_put(c, k);
524 return -1;
525 }
526
527 int bch_bucket_alloc_set(struct cache_set *c, unsigned int reserve,
528 struct bkey *k, int n, bool wait)
529 {
530 int ret;
531
532 mutex_lock(&c->bucket_lock);
533 ret = __bch_bucket_alloc_set(c, reserve, k, n, wait);
534 mutex_unlock(&c->bucket_lock);
535 return ret;
536 }
537
538
539
540 struct open_bucket {
541 struct list_head list;
542 unsigned int last_write_point;
543 unsigned int sectors_free;
544 BKEY_PADDED(key);
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 static struct open_bucket *pick_data_bucket(struct cache_set *c,
573 const struct bkey *search,
574 unsigned int write_point,
575 struct bkey *alloc)
576 {
577 struct open_bucket *ret, *ret_task = NULL;
578
579 list_for_each_entry_reverse(ret, &c->data_buckets, list)
580 if (UUID_FLASH_ONLY(&c->uuids[KEY_INODE(&ret->key)]) !=
581 UUID_FLASH_ONLY(&c->uuids[KEY_INODE(search)]))
582 continue;
583 else if (!bkey_cmp(&ret->key, search))
584 goto found;
585 else if (ret->last_write_point == write_point)
586 ret_task = ret;
587
588 ret = ret_task ?: list_first_entry(&c->data_buckets,
589 struct open_bucket, list);
590 found:
591 if (!ret->sectors_free && KEY_PTRS(alloc)) {
592 ret->sectors_free = c->sb.bucket_size;
593 bkey_copy(&ret->key, alloc);
594 bkey_init(alloc);
595 }
596
597 if (!ret->sectors_free)
598 ret = NULL;
599
600 return ret;
601 }
602
603
604
605
606
607
608
609
610
611
612
613 bool bch_alloc_sectors(struct cache_set *c,
614 struct bkey *k,
615 unsigned int sectors,
616 unsigned int write_point,
617 unsigned int write_prio,
618 bool wait)
619 {
620 struct open_bucket *b;
621 BKEY_PADDED(key) alloc;
622 unsigned int i;
623
624
625
626
627
628
629
630
631 bkey_init(&alloc.key);
632 spin_lock(&c->data_bucket_lock);
633
634 while (!(b = pick_data_bucket(c, k, write_point, &alloc.key))) {
635 unsigned int watermark = write_prio
636 ? RESERVE_MOVINGGC
637 : RESERVE_NONE;
638
639 spin_unlock(&c->data_bucket_lock);
640
641 if (bch_bucket_alloc_set(c, watermark, &alloc.key, 1, wait))
642 return false;
643
644 spin_lock(&c->data_bucket_lock);
645 }
646
647
648
649
650
651
652 if (KEY_PTRS(&alloc.key))
653 bkey_put(c, &alloc.key);
654
655 for (i = 0; i < KEY_PTRS(&b->key); i++)
656 EBUG_ON(ptr_stale(c, &b->key, i));
657
658
659
660 for (i = 0; i < KEY_PTRS(&b->key); i++)
661 k->ptr[i] = b->key.ptr[i];
662
663 sectors = min(sectors, b->sectors_free);
664
665 SET_KEY_OFFSET(k, KEY_OFFSET(k) + sectors);
666 SET_KEY_SIZE(k, sectors);
667 SET_KEY_PTRS(k, KEY_PTRS(&b->key));
668
669
670
671
672
673 list_move_tail(&b->list, &c->data_buckets);
674 bkey_copy_key(&b->key, k);
675 b->last_write_point = write_point;
676
677 b->sectors_free -= sectors;
678
679 for (i = 0; i < KEY_PTRS(&b->key); i++) {
680 SET_PTR_OFFSET(&b->key, i, PTR_OFFSET(&b->key, i) + sectors);
681
682 atomic_long_add(sectors,
683 &PTR_CACHE(c, &b->key, i)->sectors_written);
684 }
685
686 if (b->sectors_free < c->sb.block_size)
687 b->sectors_free = 0;
688
689
690
691
692
693
694 if (b->sectors_free)
695 for (i = 0; i < KEY_PTRS(&b->key); i++)
696 atomic_inc(&PTR_BUCKET(c, &b->key, i)->pin);
697
698 spin_unlock(&c->data_bucket_lock);
699 return true;
700 }
701
702
703
704 void bch_open_buckets_free(struct cache_set *c)
705 {
706 struct open_bucket *b;
707
708 while (!list_empty(&c->data_buckets)) {
709 b = list_first_entry(&c->data_buckets,
710 struct open_bucket, list);
711 list_del(&b->list);
712 kfree(b);
713 }
714 }
715
716 int bch_open_buckets_alloc(struct cache_set *c)
717 {
718 int i;
719
720 spin_lock_init(&c->data_bucket_lock);
721
722 for (i = 0; i < MAX_OPEN_BUCKETS; i++) {
723 struct open_bucket *b = kzalloc(sizeof(*b), GFP_KERNEL);
724
725 if (!b)
726 return -ENOMEM;
727
728 list_add(&b->list, &c->data_buckets);
729 }
730
731 return 0;
732 }
733
734 int bch_cache_allocator_start(struct cache *ca)
735 {
736 struct task_struct *k = kthread_run(bch_allocator_thread,
737 ca, "bcache_allocator");
738 if (IS_ERR(k))
739 return PTR_ERR(k);
740
741 ca->alloc_thread = k;
742 return 0;
743 }