xref: /linux/drivers/md/bcache/alloc.c (revision 7f71507851fc7764b36a3221839607d3a45c2025)
1 // SPDX-License-Identifier: GPL-2.0
2 /*
3  * Primary bucket allocation code
4  *
5  * Copyright 2012 Google, Inc.
6  *
7  * Allocation in bcache is done in terms of buckets:
8  *
9  * Each bucket has associated an 8 bit gen; this gen corresponds to the gen in
10  * btree pointers - they must match for the pointer to be considered valid.
11  *
12  * Thus (assuming a bucket has no dirty data or metadata in it) we can reuse a
13  * bucket simply by incrementing its gen.
14  *
15  * The gens (along with the priorities; it's really the gens are important but
16  * the code is named as if it's the priorities) are written in an arbitrary list
17  * of buckets on disk, with a pointer to them in the journal header.
18  *
19  * When we invalidate a bucket, we have to write its new gen to disk and wait
20  * for that write to complete before we use it - otherwise after a crash we
21  * could have pointers that appeared to be good but pointed to data that had
22  * been overwritten.
23  *
24  * Since the gens and priorities are all stored contiguously on disk, we can
25  * batch this up: We fill up the free_inc list with freshly invalidated buckets,
26  * call prio_write(), and when prio_write() finishes we pull buckets off the
27  * free_inc list and optionally discard them.
28  *
29  * free_inc isn't the only freelist - if it was, we'd often to sleep while
30  * priorities and gens were being written before we could allocate. c->free is a
31  * smaller freelist, and buckets on that list are always ready to be used.
32  *
33  * If we've got discards enabled, that happens when a bucket moves from the
34  * free_inc list to the free list.
35  *
36  * There is another freelist, because sometimes we have buckets that we know
37  * have nothing pointing into them - these we can reuse without waiting for
38  * priorities to be rewritten. These come from freed btree nodes and buckets
39  * that garbage collection discovered no longer had valid keys pointing into
40  * them (because they were overwritten). That's the unused list - buckets on the
41  * unused list move to the free list, optionally being discarded in the process.
42  *
43  * It's also important to ensure that gens don't wrap around - with respect to
44  * either the oldest gen in the btree or the gen on disk. This is quite
45  * difficult to do in practice, but we explicitly guard against it anyways - if
46  * a bucket is in danger of wrapping around we simply skip invalidating it that
47  * time around, and we garbage collect or rewrite the priorities sooner than we
48  * would have otherwise.
49  *
50  * bch_bucket_alloc() allocates a single bucket from a specific cache.
51  *
52  * bch_bucket_alloc_set() allocates one  bucket from different caches
53  * out of a cache set.
54  *
55  * free_some_buckets() drives all the processes described above. It's called
56  * from bch_bucket_alloc() and a few other places that need to make sure free
57  * buckets are ready.
58  *
59  * invalidate_buckets_(lru|fifo)() find buckets that are available to be
60  * invalidated, and then invalidate them and stick them on the free_inc list -
61  * in either lru or fifo order.
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 /* Bucket heap / gen */
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 long next = c->nbuckets * c->cache->sb.bucket_size / 1024;
91 	int r;
92 
93 	atomic_sub(sectors, &c->rescale);
94 
95 	do {
96 		r = atomic_read(&c->rescale);
97 
98 		if (r >= 0)
99 			return;
100 	} while (atomic_cmpxchg(&c->rescale, r, r + next) != r);
101 
102 	mutex_lock(&c->bucket_lock);
103 
104 	c->min_prio = USHRT_MAX;
105 
106 	ca = c->cache;
107 	for_each_bucket(b, ca)
108 		if (b->prio &&
109 		    b->prio != BTREE_PRIO &&
110 		    !atomic_read(&b->pin)) {
111 			b->prio--;
112 			c->min_prio = min(c->min_prio, b->prio);
113 		}
114 
115 	mutex_unlock(&c->bucket_lock);
116 }
117 
118 /*
119  * Background allocation thread: scans for buckets to be invalidated,
120  * invalidates them, rewrites prios/gens (marking them as invalidated on disk),
121  * then optionally issues discard commands to the newly free buckets, then puts
122  * them on the various freelists.
123  */
124 
125 static inline bool can_inc_bucket_gen(struct bucket *b)
126 {
127 	return bucket_gc_gen(b) < BUCKET_GC_GEN_MAX;
128 }
129 
130 bool bch_can_invalidate_bucket(struct cache *ca, struct bucket *b)
131 {
132 	return (ca->set->gc_mark_valid || b->reclaimable_in_gc) &&
133 	       ((!GC_MARK(b) || GC_MARK(b) == GC_MARK_RECLAIMABLE) &&
134 	       !atomic_read(&b->pin) && can_inc_bucket_gen(b));
135 }
136 
137 void __bch_invalidate_one_bucket(struct cache *ca, struct bucket *b)
138 {
139 	lockdep_assert_held(&ca->set->bucket_lock);
140 	BUG_ON(GC_MARK(b) && GC_MARK(b) != GC_MARK_RECLAIMABLE);
141 
142 	if (GC_SECTORS_USED(b))
143 		trace_bcache_invalidate(ca, b - ca->buckets);
144 
145 	bch_inc_gen(ca, b);
146 	b->prio = INITIAL_PRIO;
147 	atomic_inc(&b->pin);
148 	b->reclaimable_in_gc = 0;
149 }
150 
151 static void bch_invalidate_one_bucket(struct cache *ca, struct bucket *b)
152 {
153 	__bch_invalidate_one_bucket(ca, b);
154 
155 	fifo_push(&ca->free_inc, b - ca->buckets);
156 }
157 
158 /*
159  * Determines what order we're going to reuse buckets, smallest bucket_prio()
160  * first: we also take into account the number of sectors of live data in that
161  * bucket, and in order for that multiply to make sense we have to scale bucket
162  *
163  * Thus, we scale the bucket priorities so that the bucket with the smallest
164  * prio is worth 1/8th of what INITIAL_PRIO is worth.
165  */
166 
167 static inline unsigned int new_bucket_prio(struct cache *ca, struct bucket *b)
168 {
169 	unsigned int min_prio = (INITIAL_PRIO - ca->set->min_prio) / 8;
170 
171 	return (b->prio - ca->set->min_prio + min_prio) * GC_SECTORS_USED(b);
172 }
173 
174 static inline bool new_bucket_max_cmp(const void *l, const void *r, void *args)
175 {
176 	struct bucket **lhs = (struct bucket **)l;
177 	struct bucket **rhs = (struct bucket **)r;
178 	struct cache *ca = args;
179 
180 	return new_bucket_prio(ca, *lhs) > new_bucket_prio(ca, *rhs);
181 }
182 
183 static inline bool new_bucket_min_cmp(const void *l, const void *r, void *args)
184 {
185 	struct bucket **lhs = (struct bucket **)l;
186 	struct bucket **rhs = (struct bucket **)r;
187 	struct cache *ca = args;
188 
189 	return new_bucket_prio(ca, *lhs) < new_bucket_prio(ca, *rhs);
190 }
191 
192 static void invalidate_buckets_lru(struct cache *ca)
193 {
194 	struct bucket *b;
195 	const struct min_heap_callbacks bucket_max_cmp_callback = {
196 		.less = new_bucket_max_cmp,
197 		.swp = NULL,
198 	};
199 	const struct min_heap_callbacks bucket_min_cmp_callback = {
200 		.less = new_bucket_min_cmp,
201 		.swp = NULL,
202 	};
203 
204 	ca->heap.nr = 0;
205 
206 	for_each_bucket(b, ca) {
207 		if (!bch_can_invalidate_bucket(ca, b))
208 			continue;
209 
210 		if (!min_heap_full(&ca->heap))
211 			min_heap_push(&ca->heap, &b, &bucket_max_cmp_callback, ca);
212 		else if (!new_bucket_max_cmp(&b, min_heap_peek(&ca->heap), ca)) {
213 			ca->heap.data[0] = b;
214 			min_heap_sift_down(&ca->heap, 0, &bucket_max_cmp_callback, ca);
215 		}
216 	}
217 
218 	min_heapify_all(&ca->heap, &bucket_min_cmp_callback, ca);
219 
220 	while (!fifo_full(&ca->free_inc)) {
221 		if (!ca->heap.nr) {
222 			/*
223 			 * We don't want to be calling invalidate_buckets()
224 			 * multiple times when it can't do anything
225 			 */
226 			ca->invalidate_needs_gc = 1;
227 			wake_up_gc(ca->set);
228 			return;
229 		}
230 		b = min_heap_peek(&ca->heap)[0];
231 		min_heap_pop(&ca->heap, &bucket_min_cmp_callback, ca);
232 
233 		bch_invalidate_one_bucket(ca, b);
234 	}
235 }
236 
237 static void invalidate_buckets_fifo(struct cache *ca)
238 {
239 	struct bucket *b;
240 	size_t checked = 0;
241 
242 	while (!fifo_full(&ca->free_inc)) {
243 		if (ca->fifo_last_bucket <  ca->sb.first_bucket ||
244 		    ca->fifo_last_bucket >= ca->sb.nbuckets)
245 			ca->fifo_last_bucket = ca->sb.first_bucket;
246 
247 		b = ca->buckets + ca->fifo_last_bucket++;
248 
249 		if (bch_can_invalidate_bucket(ca, b))
250 			bch_invalidate_one_bucket(ca, b);
251 
252 		if (++checked >= ca->sb.nbuckets) {
253 			ca->invalidate_needs_gc = 1;
254 			wake_up_gc(ca->set);
255 			return;
256 		}
257 	}
258 }
259 
260 static void invalidate_buckets_random(struct cache *ca)
261 {
262 	struct bucket *b;
263 	size_t checked = 0;
264 
265 	while (!fifo_full(&ca->free_inc)) {
266 		size_t n;
267 
268 		get_random_bytes(&n, sizeof(n));
269 
270 		n %= (size_t) (ca->sb.nbuckets - ca->sb.first_bucket);
271 		n += ca->sb.first_bucket;
272 
273 		b = ca->buckets + n;
274 
275 		if (bch_can_invalidate_bucket(ca, b))
276 			bch_invalidate_one_bucket(ca, b);
277 
278 		if (++checked >= ca->sb.nbuckets / 2) {
279 			ca->invalidate_needs_gc = 1;
280 			wake_up_gc(ca->set);
281 			return;
282 		}
283 	}
284 }
285 
286 static void invalidate_buckets(struct cache *ca)
287 {
288 	BUG_ON(ca->invalidate_needs_gc);
289 
290 	switch (CACHE_REPLACEMENT(&ca->sb)) {
291 	case CACHE_REPLACEMENT_LRU:
292 		invalidate_buckets_lru(ca);
293 		break;
294 	case CACHE_REPLACEMENT_FIFO:
295 		invalidate_buckets_fifo(ca);
296 		break;
297 	case CACHE_REPLACEMENT_RANDOM:
298 		invalidate_buckets_random(ca);
299 		break;
300 	}
301 }
302 
303 #define allocator_wait(ca, cond)					\
304 do {									\
305 	while (1) {							\
306 		set_current_state(TASK_INTERRUPTIBLE);			\
307 		if (cond)						\
308 			break;						\
309 									\
310 		mutex_unlock(&(ca)->set->bucket_lock);			\
311 		if (kthread_should_stop() ||				\
312 		    test_bit(CACHE_SET_IO_DISABLE, &ca->set->flags)) {	\
313 			set_current_state(TASK_RUNNING);		\
314 			goto out;					\
315 		}							\
316 									\
317 		schedule();						\
318 		mutex_lock(&(ca)->set->bucket_lock);			\
319 	}								\
320 	__set_current_state(TASK_RUNNING);				\
321 } while (0)
322 
323 static int bch_allocator_push(struct cache *ca, long bucket)
324 {
325 	unsigned int i;
326 
327 	/* Prios/gens are actually the most important reserve */
328 	if (fifo_push(&ca->free[RESERVE_PRIO], bucket))
329 		return true;
330 
331 	for (i = 0; i < RESERVE_NR; i++)
332 		if (fifo_push(&ca->free[i], bucket))
333 			return true;
334 
335 	return false;
336 }
337 
338 static int bch_allocator_thread(void *arg)
339 {
340 	struct cache *ca = arg;
341 
342 	mutex_lock(&ca->set->bucket_lock);
343 
344 	while (1) {
345 		/*
346 		 * First, we pull buckets off of the unused and free_inc lists,
347 		 * possibly issue discards to them, then we add the bucket to
348 		 * the free list:
349 		 */
350 		while (1) {
351 			long bucket;
352 
353 			if (!fifo_pop(&ca->free_inc, bucket))
354 				break;
355 
356 			if (ca->discard) {
357 				mutex_unlock(&ca->set->bucket_lock);
358 				blkdev_issue_discard(ca->bdev,
359 					bucket_to_sector(ca->set, bucket),
360 					ca->sb.bucket_size, GFP_KERNEL);
361 				mutex_lock(&ca->set->bucket_lock);
362 			}
363 
364 			allocator_wait(ca, bch_allocator_push(ca, bucket));
365 			wake_up(&ca->set->btree_cache_wait);
366 			wake_up(&ca->set->bucket_wait);
367 		}
368 
369 		/*
370 		 * We've run out of free buckets, we need to find some buckets
371 		 * we can invalidate. First, invalidate them in memory and add
372 		 * them to the free_inc list:
373 		 */
374 
375 retry_invalidate:
376 		allocator_wait(ca, !ca->invalidate_needs_gc);
377 		invalidate_buckets(ca);
378 
379 		/*
380 		 * Now, we write their new gens to disk so we can start writing
381 		 * new stuff to them:
382 		 */
383 		allocator_wait(ca, !atomic_read(&ca->set->prio_blocked));
384 		if (CACHE_SYNC(&ca->sb)) {
385 			/*
386 			 * This could deadlock if an allocation with a btree
387 			 * node locked ever blocked - having the btree node
388 			 * locked would block garbage collection, but here we're
389 			 * waiting on garbage collection before we invalidate
390 			 * and free anything.
391 			 *
392 			 * But this should be safe since the btree code always
393 			 * uses btree_check_reserve() before allocating now, and
394 			 * if it fails it blocks without btree nodes locked.
395 			 */
396 			if (!fifo_full(&ca->free_inc))
397 				goto retry_invalidate;
398 
399 			if (bch_prio_write(ca, false) < 0) {
400 				ca->invalidate_needs_gc = 1;
401 				wake_up_gc(ca->set);
402 			}
403 		}
404 	}
405 out:
406 	wait_for_kthread_stop();
407 	return 0;
408 }
409 
410 /* Allocation */
411 
412 long bch_bucket_alloc(struct cache *ca, unsigned int reserve, bool wait)
413 {
414 	DEFINE_WAIT(w);
415 	struct bucket *b;
416 	long r;
417 
418 
419 	/* No allocation if CACHE_SET_IO_DISABLE bit is set */
420 	if (unlikely(test_bit(CACHE_SET_IO_DISABLE, &ca->set->flags)))
421 		return -1;
422 
423 	/* fastpath */
424 	if (fifo_pop(&ca->free[RESERVE_NONE], r) ||
425 	    fifo_pop(&ca->free[reserve], r))
426 		goto out;
427 
428 	if (!wait) {
429 		trace_bcache_alloc_fail(ca, reserve);
430 		return -1;
431 	}
432 
433 	do {
434 		prepare_to_wait(&ca->set->bucket_wait, &w,
435 				TASK_UNINTERRUPTIBLE);
436 
437 		mutex_unlock(&ca->set->bucket_lock);
438 		schedule();
439 		mutex_lock(&ca->set->bucket_lock);
440 	} while (!fifo_pop(&ca->free[RESERVE_NONE], r) &&
441 		 !fifo_pop(&ca->free[reserve], r));
442 
443 	finish_wait(&ca->set->bucket_wait, &w);
444 out:
445 	if (ca->alloc_thread)
446 		wake_up_process(ca->alloc_thread);
447 
448 	trace_bcache_alloc(ca, reserve);
449 
450 	if (expensive_debug_checks(ca->set)) {
451 		size_t iter;
452 		long i;
453 		unsigned int j;
454 
455 		for (iter = 0; iter < prio_buckets(ca) * 2; iter++)
456 			BUG_ON(ca->prio_buckets[iter] == (uint64_t) r);
457 
458 		for (j = 0; j < RESERVE_NR; j++)
459 			fifo_for_each(i, &ca->free[j], iter)
460 				BUG_ON(i == r);
461 		fifo_for_each(i, &ca->free_inc, iter)
462 			BUG_ON(i == r);
463 	}
464 
465 	b = ca->buckets + r;
466 
467 	BUG_ON(atomic_read(&b->pin) != 1);
468 
469 	SET_GC_SECTORS_USED(b, ca->sb.bucket_size);
470 
471 	if (reserve <= RESERVE_PRIO) {
472 		SET_GC_MARK(b, GC_MARK_METADATA);
473 		SET_GC_MOVE(b, 0);
474 		b->prio = BTREE_PRIO;
475 	} else {
476 		SET_GC_MARK(b, GC_MARK_RECLAIMABLE);
477 		SET_GC_MOVE(b, 0);
478 		b->prio = INITIAL_PRIO;
479 	}
480 
481 	if (ca->set->avail_nbuckets > 0) {
482 		ca->set->avail_nbuckets--;
483 		bch_update_bucket_in_use(ca->set, &ca->set->gc_stats);
484 	}
485 
486 	return r;
487 }
488 
489 void __bch_bucket_free(struct cache *ca, struct bucket *b)
490 {
491 	SET_GC_MARK(b, 0);
492 	SET_GC_SECTORS_USED(b, 0);
493 
494 	if (ca->set->avail_nbuckets < ca->set->nbuckets) {
495 		ca->set->avail_nbuckets++;
496 		bch_update_bucket_in_use(ca->set, &ca->set->gc_stats);
497 	}
498 }
499 
500 void bch_bucket_free(struct cache_set *c, struct bkey *k)
501 {
502 	unsigned int i;
503 
504 	for (i = 0; i < KEY_PTRS(k); i++)
505 		__bch_bucket_free(c->cache, PTR_BUCKET(c, k, i));
506 }
507 
508 int __bch_bucket_alloc_set(struct cache_set *c, unsigned int reserve,
509 			   struct bkey *k, bool wait)
510 {
511 	struct cache *ca;
512 	long b;
513 
514 	/* No allocation if CACHE_SET_IO_DISABLE bit is set */
515 	if (unlikely(test_bit(CACHE_SET_IO_DISABLE, &c->flags)))
516 		return -1;
517 
518 	lockdep_assert_held(&c->bucket_lock);
519 
520 	bkey_init(k);
521 
522 	ca = c->cache;
523 	b = bch_bucket_alloc(ca, reserve, wait);
524 	if (b < 0)
525 		return -1;
526 
527 	k->ptr[0] = MAKE_PTR(ca->buckets[b].gen,
528 			     bucket_to_sector(c, b),
529 			     ca->sb.nr_this_dev);
530 
531 	SET_KEY_PTRS(k, 1);
532 
533 	return 0;
534 }
535 
536 int bch_bucket_alloc_set(struct cache_set *c, unsigned int reserve,
537 			 struct bkey *k, bool wait)
538 {
539 	int ret;
540 
541 	mutex_lock(&c->bucket_lock);
542 	ret = __bch_bucket_alloc_set(c, reserve, k, wait);
543 	mutex_unlock(&c->bucket_lock);
544 	return ret;
545 }
546 
547 /* Sector allocator */
548 
549 struct open_bucket {
550 	struct list_head	list;
551 	unsigned int		last_write_point;
552 	unsigned int		sectors_free;
553 	BKEY_PADDED(key);
554 };
555 
556 /*
557  * We keep multiple buckets open for writes, and try to segregate different
558  * write streams for better cache utilization: first we try to segregate flash
559  * only volume write streams from cached devices, secondly we look for a bucket
560  * where the last write to it was sequential with the current write, and
561  * failing that we look for a bucket that was last used by the same task.
562  *
563  * The ideas is if you've got multiple tasks pulling data into the cache at the
564  * same time, you'll get better cache utilization if you try to segregate their
565  * data and preserve locality.
566  *
567  * For example, dirty sectors of flash only volume is not reclaimable, if their
568  * dirty sectors mixed with dirty sectors of cached device, such buckets will
569  * be marked as dirty and won't be reclaimed, though the dirty data of cached
570  * device have been written back to backend device.
571  *
572  * And say you've starting Firefox at the same time you're copying a
573  * bunch of files. Firefox will likely end up being fairly hot and stay in the
574  * cache awhile, but the data you copied might not be; if you wrote all that
575  * data to the same buckets it'd get invalidated at the same time.
576  *
577  * Both of those tasks will be doing fairly random IO so we can't rely on
578  * detecting sequential IO to segregate their data, but going off of the task
579  * should be a sane heuristic.
580  */
581 static struct open_bucket *pick_data_bucket(struct cache_set *c,
582 					    const struct bkey *search,
583 					    unsigned int write_point,
584 					    struct bkey *alloc)
585 {
586 	struct open_bucket *ret, *ret_task = NULL;
587 
588 	list_for_each_entry_reverse(ret, &c->data_buckets, list)
589 		if (UUID_FLASH_ONLY(&c->uuids[KEY_INODE(&ret->key)]) !=
590 		    UUID_FLASH_ONLY(&c->uuids[KEY_INODE(search)]))
591 			continue;
592 		else if (!bkey_cmp(&ret->key, search))
593 			goto found;
594 		else if (ret->last_write_point == write_point)
595 			ret_task = ret;
596 
597 	ret = ret_task ?: list_first_entry(&c->data_buckets,
598 					   struct open_bucket, list);
599 found:
600 	if (!ret->sectors_free && KEY_PTRS(alloc)) {
601 		ret->sectors_free = c->cache->sb.bucket_size;
602 		bkey_copy(&ret->key, alloc);
603 		bkey_init(alloc);
604 	}
605 
606 	if (!ret->sectors_free)
607 		ret = NULL;
608 
609 	return ret;
610 }
611 
612 /*
613  * Allocates some space in the cache to write to, and k to point to the newly
614  * allocated space, and updates KEY_SIZE(k) and KEY_OFFSET(k) (to point to the
615  * end of the newly allocated space).
616  *
617  * May allocate fewer sectors than @sectors, KEY_SIZE(k) indicates how many
618  * sectors were actually allocated.
619  *
620  * If s->writeback is true, will not fail.
621  */
622 bool bch_alloc_sectors(struct cache_set *c,
623 		       struct bkey *k,
624 		       unsigned int sectors,
625 		       unsigned int write_point,
626 		       unsigned int write_prio,
627 		       bool wait)
628 {
629 	struct open_bucket *b;
630 	BKEY_PADDED(key) alloc;
631 	unsigned int i;
632 
633 	/*
634 	 * We might have to allocate a new bucket, which we can't do with a
635 	 * spinlock held. So if we have to allocate, we drop the lock, allocate
636 	 * and then retry. KEY_PTRS() indicates whether alloc points to
637 	 * allocated bucket(s).
638 	 */
639 
640 	bkey_init(&alloc.key);
641 	spin_lock(&c->data_bucket_lock);
642 
643 	while (!(b = pick_data_bucket(c, k, write_point, &alloc.key))) {
644 		unsigned int watermark = write_prio
645 			? RESERVE_MOVINGGC
646 			: RESERVE_NONE;
647 
648 		spin_unlock(&c->data_bucket_lock);
649 
650 		if (bch_bucket_alloc_set(c, watermark, &alloc.key, wait))
651 			return false;
652 
653 		spin_lock(&c->data_bucket_lock);
654 	}
655 
656 	/*
657 	 * If we had to allocate, we might race and not need to allocate the
658 	 * second time we call pick_data_bucket(). If we allocated a bucket but
659 	 * didn't use it, drop the refcount bch_bucket_alloc_set() took:
660 	 */
661 	if (KEY_PTRS(&alloc.key))
662 		bkey_put(c, &alloc.key);
663 
664 	for (i = 0; i < KEY_PTRS(&b->key); i++)
665 		EBUG_ON(ptr_stale(c, &b->key, i));
666 
667 	/* Set up the pointer to the space we're allocating: */
668 
669 	for (i = 0; i < KEY_PTRS(&b->key); i++)
670 		k->ptr[i] = b->key.ptr[i];
671 
672 	sectors = min(sectors, b->sectors_free);
673 
674 	SET_KEY_OFFSET(k, KEY_OFFSET(k) + sectors);
675 	SET_KEY_SIZE(k, sectors);
676 	SET_KEY_PTRS(k, KEY_PTRS(&b->key));
677 
678 	/*
679 	 * Move b to the end of the lru, and keep track of what this bucket was
680 	 * last used for:
681 	 */
682 	list_move_tail(&b->list, &c->data_buckets);
683 	bkey_copy_key(&b->key, k);
684 	b->last_write_point = write_point;
685 
686 	b->sectors_free	-= sectors;
687 
688 	for (i = 0; i < KEY_PTRS(&b->key); i++) {
689 		SET_PTR_OFFSET(&b->key, i, PTR_OFFSET(&b->key, i) + sectors);
690 
691 		atomic_long_add(sectors,
692 				&c->cache->sectors_written);
693 	}
694 
695 	if (b->sectors_free < c->cache->sb.block_size)
696 		b->sectors_free = 0;
697 
698 	/*
699 	 * k takes refcounts on the buckets it points to until it's inserted
700 	 * into the btree, but if we're done with this bucket we just transfer
701 	 * get_data_bucket()'s refcount.
702 	 */
703 	if (b->sectors_free)
704 		for (i = 0; i < KEY_PTRS(&b->key); i++)
705 			atomic_inc(&PTR_BUCKET(c, &b->key, i)->pin);
706 
707 	spin_unlock(&c->data_bucket_lock);
708 	return true;
709 }
710 
711 /* Init */
712 
713 void bch_open_buckets_free(struct cache_set *c)
714 {
715 	struct open_bucket *b;
716 
717 	while (!list_empty(&c->data_buckets)) {
718 		b = list_first_entry(&c->data_buckets,
719 				     struct open_bucket, list);
720 		list_del(&b->list);
721 		kfree(b);
722 	}
723 }
724 
725 int bch_open_buckets_alloc(struct cache_set *c)
726 {
727 	int i;
728 
729 	spin_lock_init(&c->data_bucket_lock);
730 
731 	for (i = 0; i < MAX_OPEN_BUCKETS; i++) {
732 		struct open_bucket *b = kzalloc(sizeof(*b), GFP_KERNEL);
733 
734 		if (!b)
735 			return -ENOMEM;
736 
737 		list_add(&b->list, &c->data_buckets);
738 	}
739 
740 	return 0;
741 }
742 
743 int bch_cache_allocator_start(struct cache *ca)
744 {
745 	struct task_struct *k = kthread_run(bch_allocator_thread,
746 					    ca, "bcache_allocator");
747 	if (IS_ERR(k))
748 		return PTR_ERR(k);
749 
750 	ca->alloc_thread = k;
751 	return 0;
752 }
753