xref: /openbmc/linux/drivers/md/bcache/extents.c (revision a8da474e)
1 /*
2  * Copyright (C) 2010 Kent Overstreet <kent.overstreet@gmail.com>
3  *
4  * Uses a block device as cache for other block devices; optimized for SSDs.
5  * All allocation is done in buckets, which should match the erase block size
6  * of the device.
7  *
8  * Buckets containing cached data are kept on a heap sorted by priority;
9  * bucket priority is increased on cache hit, and periodically all the buckets
10  * on the heap have their priority scaled down. This currently is just used as
11  * an LRU but in the future should allow for more intelligent heuristics.
12  *
13  * Buckets have an 8 bit counter; freeing is accomplished by incrementing the
14  * counter. Garbage collection is used to remove stale pointers.
15  *
16  * Indexing is done via a btree; nodes are not necessarily fully sorted, rather
17  * as keys are inserted we only sort the pages that have not yet been written.
18  * When garbage collection is run, we resort the entire node.
19  *
20  * All configuration is done via sysfs; see Documentation/bcache.txt.
21  */
22 
23 #include "bcache.h"
24 #include "btree.h"
25 #include "debug.h"
26 #include "extents.h"
27 #include "writeback.h"
28 
29 static void sort_key_next(struct btree_iter *iter,
30 			  struct btree_iter_set *i)
31 {
32 	i->k = bkey_next(i->k);
33 
34 	if (i->k == i->end)
35 		*i = iter->data[--iter->used];
36 }
37 
38 static bool bch_key_sort_cmp(struct btree_iter_set l,
39 			     struct btree_iter_set r)
40 {
41 	int64_t c = bkey_cmp(l.k, r.k);
42 
43 	return c ? c > 0 : l.k < r.k;
44 }
45 
46 static bool __ptr_invalid(struct cache_set *c, const struct bkey *k)
47 {
48 	unsigned i;
49 
50 	for (i = 0; i < KEY_PTRS(k); i++)
51 		if (ptr_available(c, k, i)) {
52 			struct cache *ca = PTR_CACHE(c, k, i);
53 			size_t bucket = PTR_BUCKET_NR(c, k, i);
54 			size_t r = bucket_remainder(c, PTR_OFFSET(k, i));
55 
56 			if (KEY_SIZE(k) + r > c->sb.bucket_size ||
57 			    bucket <  ca->sb.first_bucket ||
58 			    bucket >= ca->sb.nbuckets)
59 				return true;
60 		}
61 
62 	return false;
63 }
64 
65 /* Common among btree and extent ptrs */
66 
67 static const char *bch_ptr_status(struct cache_set *c, const struct bkey *k)
68 {
69 	unsigned i;
70 
71 	for (i = 0; i < KEY_PTRS(k); i++)
72 		if (ptr_available(c, k, i)) {
73 			struct cache *ca = PTR_CACHE(c, k, i);
74 			size_t bucket = PTR_BUCKET_NR(c, k, i);
75 			size_t r = bucket_remainder(c, PTR_OFFSET(k, i));
76 
77 			if (KEY_SIZE(k) + r > c->sb.bucket_size)
78 				return "bad, length too big";
79 			if (bucket <  ca->sb.first_bucket)
80 				return "bad, short offset";
81 			if (bucket >= ca->sb.nbuckets)
82 				return "bad, offset past end of device";
83 			if (ptr_stale(c, k, i))
84 				return "stale";
85 		}
86 
87 	if (!bkey_cmp(k, &ZERO_KEY))
88 		return "bad, null key";
89 	if (!KEY_PTRS(k))
90 		return "bad, no pointers";
91 	if (!KEY_SIZE(k))
92 		return "zeroed key";
93 	return "";
94 }
95 
96 void bch_extent_to_text(char *buf, size_t size, const struct bkey *k)
97 {
98 	unsigned i = 0;
99 	char *out = buf, *end = buf + size;
100 
101 #define p(...)	(out += scnprintf(out, end - out, __VA_ARGS__))
102 
103 	p("%llu:%llu len %llu -> [", KEY_INODE(k), KEY_START(k), KEY_SIZE(k));
104 
105 	for (i = 0; i < KEY_PTRS(k); i++) {
106 		if (i)
107 			p(", ");
108 
109 		if (PTR_DEV(k, i) == PTR_CHECK_DEV)
110 			p("check dev");
111 		else
112 			p("%llu:%llu gen %llu", PTR_DEV(k, i),
113 			  PTR_OFFSET(k, i), PTR_GEN(k, i));
114 	}
115 
116 	p("]");
117 
118 	if (KEY_DIRTY(k))
119 		p(" dirty");
120 	if (KEY_CSUM(k))
121 		p(" cs%llu %llx", KEY_CSUM(k), k->ptr[1]);
122 #undef p
123 }
124 
125 static void bch_bkey_dump(struct btree_keys *keys, const struct bkey *k)
126 {
127 	struct btree *b = container_of(keys, struct btree, keys);
128 	unsigned j;
129 	char buf[80];
130 
131 	bch_extent_to_text(buf, sizeof(buf), k);
132 	printk(" %s", buf);
133 
134 	for (j = 0; j < KEY_PTRS(k); j++) {
135 		size_t n = PTR_BUCKET_NR(b->c, k, j);
136 		printk(" bucket %zu", n);
137 
138 		if (n >= b->c->sb.first_bucket && n < b->c->sb.nbuckets)
139 			printk(" prio %i",
140 			       PTR_BUCKET(b->c, k, j)->prio);
141 	}
142 
143 	printk(" %s\n", bch_ptr_status(b->c, k));
144 }
145 
146 /* Btree ptrs */
147 
148 bool __bch_btree_ptr_invalid(struct cache_set *c, const struct bkey *k)
149 {
150 	char buf[80];
151 
152 	if (!KEY_PTRS(k) || !KEY_SIZE(k) || KEY_DIRTY(k))
153 		goto bad;
154 
155 	if (__ptr_invalid(c, k))
156 		goto bad;
157 
158 	return false;
159 bad:
160 	bch_extent_to_text(buf, sizeof(buf), k);
161 	cache_bug(c, "spotted btree ptr %s: %s", buf, bch_ptr_status(c, k));
162 	return true;
163 }
164 
165 static bool bch_btree_ptr_invalid(struct btree_keys *bk, const struct bkey *k)
166 {
167 	struct btree *b = container_of(bk, struct btree, keys);
168 	return __bch_btree_ptr_invalid(b->c, k);
169 }
170 
171 static bool btree_ptr_bad_expensive(struct btree *b, const struct bkey *k)
172 {
173 	unsigned i;
174 	char buf[80];
175 	struct bucket *g;
176 
177 	if (mutex_trylock(&b->c->bucket_lock)) {
178 		for (i = 0; i < KEY_PTRS(k); i++)
179 			if (ptr_available(b->c, k, i)) {
180 				g = PTR_BUCKET(b->c, k, i);
181 
182 				if (KEY_DIRTY(k) ||
183 				    g->prio != BTREE_PRIO ||
184 				    (b->c->gc_mark_valid &&
185 				     GC_MARK(g) != GC_MARK_METADATA))
186 					goto err;
187 			}
188 
189 		mutex_unlock(&b->c->bucket_lock);
190 	}
191 
192 	return false;
193 err:
194 	mutex_unlock(&b->c->bucket_lock);
195 	bch_extent_to_text(buf, sizeof(buf), k);
196 	btree_bug(b,
197 "inconsistent btree pointer %s: bucket %zi pin %i prio %i gen %i last_gc %i mark %llu",
198 		  buf, PTR_BUCKET_NR(b->c, k, i), atomic_read(&g->pin),
199 		  g->prio, g->gen, g->last_gc, GC_MARK(g));
200 	return true;
201 }
202 
203 static bool bch_btree_ptr_bad(struct btree_keys *bk, const struct bkey *k)
204 {
205 	struct btree *b = container_of(bk, struct btree, keys);
206 	unsigned i;
207 
208 	if (!bkey_cmp(k, &ZERO_KEY) ||
209 	    !KEY_PTRS(k) ||
210 	    bch_ptr_invalid(bk, k))
211 		return true;
212 
213 	for (i = 0; i < KEY_PTRS(k); i++)
214 		if (!ptr_available(b->c, k, i) ||
215 		    ptr_stale(b->c, k, i))
216 			return true;
217 
218 	if (expensive_debug_checks(b->c) &&
219 	    btree_ptr_bad_expensive(b, k))
220 		return true;
221 
222 	return false;
223 }
224 
225 static bool bch_btree_ptr_insert_fixup(struct btree_keys *bk,
226 				       struct bkey *insert,
227 				       struct btree_iter *iter,
228 				       struct bkey *replace_key)
229 {
230 	struct btree *b = container_of(bk, struct btree, keys);
231 
232 	if (!KEY_OFFSET(insert))
233 		btree_current_write(b)->prio_blocked++;
234 
235 	return false;
236 }
237 
238 const struct btree_keys_ops bch_btree_keys_ops = {
239 	.sort_cmp	= bch_key_sort_cmp,
240 	.insert_fixup	= bch_btree_ptr_insert_fixup,
241 	.key_invalid	= bch_btree_ptr_invalid,
242 	.key_bad	= bch_btree_ptr_bad,
243 	.key_to_text	= bch_extent_to_text,
244 	.key_dump	= bch_bkey_dump,
245 };
246 
247 /* Extents */
248 
249 /*
250  * Returns true if l > r - unless l == r, in which case returns true if l is
251  * older than r.
252  *
253  * Necessary for btree_sort_fixup() - if there are multiple keys that compare
254  * equal in different sets, we have to process them newest to oldest.
255  */
256 static bool bch_extent_sort_cmp(struct btree_iter_set l,
257 				struct btree_iter_set r)
258 {
259 	int64_t c = bkey_cmp(&START_KEY(l.k), &START_KEY(r.k));
260 
261 	return c ? c > 0 : l.k < r.k;
262 }
263 
264 static struct bkey *bch_extent_sort_fixup(struct btree_iter *iter,
265 					  struct bkey *tmp)
266 {
267 	while (iter->used > 1) {
268 		struct btree_iter_set *top = iter->data, *i = top + 1;
269 
270 		if (iter->used > 2 &&
271 		    bch_extent_sort_cmp(i[0], i[1]))
272 			i++;
273 
274 		if (bkey_cmp(top->k, &START_KEY(i->k)) <= 0)
275 			break;
276 
277 		if (!KEY_SIZE(i->k)) {
278 			sort_key_next(iter, i);
279 			heap_sift(iter, i - top, bch_extent_sort_cmp);
280 			continue;
281 		}
282 
283 		if (top->k > i->k) {
284 			if (bkey_cmp(top->k, i->k) >= 0)
285 				sort_key_next(iter, i);
286 			else
287 				bch_cut_front(top->k, i->k);
288 
289 			heap_sift(iter, i - top, bch_extent_sort_cmp);
290 		} else {
291 			/* can't happen because of comparison func */
292 			BUG_ON(!bkey_cmp(&START_KEY(top->k), &START_KEY(i->k)));
293 
294 			if (bkey_cmp(i->k, top->k) < 0) {
295 				bkey_copy(tmp, top->k);
296 
297 				bch_cut_back(&START_KEY(i->k), tmp);
298 				bch_cut_front(i->k, top->k);
299 				heap_sift(iter, 0, bch_extent_sort_cmp);
300 
301 				return tmp;
302 			} else {
303 				bch_cut_back(&START_KEY(i->k), top->k);
304 			}
305 		}
306 	}
307 
308 	return NULL;
309 }
310 
311 static void bch_subtract_dirty(struct bkey *k,
312 			   struct cache_set *c,
313 			   uint64_t offset,
314 			   int sectors)
315 {
316 	if (KEY_DIRTY(k))
317 		bcache_dev_sectors_dirty_add(c, KEY_INODE(k),
318 					     offset, -sectors);
319 }
320 
321 static bool bch_extent_insert_fixup(struct btree_keys *b,
322 				    struct bkey *insert,
323 				    struct btree_iter *iter,
324 				    struct bkey *replace_key)
325 {
326 	struct cache_set *c = container_of(b, struct btree, keys)->c;
327 
328 	uint64_t old_offset;
329 	unsigned old_size, sectors_found = 0;
330 
331 	BUG_ON(!KEY_OFFSET(insert));
332 	BUG_ON(!KEY_SIZE(insert));
333 
334 	while (1) {
335 		struct bkey *k = bch_btree_iter_next(iter);
336 		if (!k)
337 			break;
338 
339 		if (bkey_cmp(&START_KEY(k), insert) >= 0) {
340 			if (KEY_SIZE(k))
341 				break;
342 			else
343 				continue;
344 		}
345 
346 		if (bkey_cmp(k, &START_KEY(insert)) <= 0)
347 			continue;
348 
349 		old_offset = KEY_START(k);
350 		old_size = KEY_SIZE(k);
351 
352 		/*
353 		 * We might overlap with 0 size extents; we can't skip these
354 		 * because if they're in the set we're inserting to we have to
355 		 * adjust them so they don't overlap with the key we're
356 		 * inserting. But we don't want to check them for replace
357 		 * operations.
358 		 */
359 
360 		if (replace_key && KEY_SIZE(k)) {
361 			/*
362 			 * k might have been split since we inserted/found the
363 			 * key we're replacing
364 			 */
365 			unsigned i;
366 			uint64_t offset = KEY_START(k) -
367 				KEY_START(replace_key);
368 
369 			/* But it must be a subset of the replace key */
370 			if (KEY_START(k) < KEY_START(replace_key) ||
371 			    KEY_OFFSET(k) > KEY_OFFSET(replace_key))
372 				goto check_failed;
373 
374 			/* We didn't find a key that we were supposed to */
375 			if (KEY_START(k) > KEY_START(insert) + sectors_found)
376 				goto check_failed;
377 
378 			if (!bch_bkey_equal_header(k, replace_key))
379 				goto check_failed;
380 
381 			/* skip past gen */
382 			offset <<= 8;
383 
384 			BUG_ON(!KEY_PTRS(replace_key));
385 
386 			for (i = 0; i < KEY_PTRS(replace_key); i++)
387 				if (k->ptr[i] != replace_key->ptr[i] + offset)
388 					goto check_failed;
389 
390 			sectors_found = KEY_OFFSET(k) - KEY_START(insert);
391 		}
392 
393 		if (bkey_cmp(insert, k) < 0 &&
394 		    bkey_cmp(&START_KEY(insert), &START_KEY(k)) > 0) {
395 			/*
396 			 * We overlapped in the middle of an existing key: that
397 			 * means we have to split the old key. But we have to do
398 			 * slightly different things depending on whether the
399 			 * old key has been written out yet.
400 			 */
401 
402 			struct bkey *top;
403 
404 			bch_subtract_dirty(k, c, KEY_START(insert),
405 				       KEY_SIZE(insert));
406 
407 			if (bkey_written(b, k)) {
408 				/*
409 				 * We insert a new key to cover the top of the
410 				 * old key, and the old key is modified in place
411 				 * to represent the bottom split.
412 				 *
413 				 * It's completely arbitrary whether the new key
414 				 * is the top or the bottom, but it has to match
415 				 * up with what btree_sort_fixup() does - it
416 				 * doesn't check for this kind of overlap, it
417 				 * depends on us inserting a new key for the top
418 				 * here.
419 				 */
420 				top = bch_bset_search(b, bset_tree_last(b),
421 						      insert);
422 				bch_bset_insert(b, top, k);
423 			} else {
424 				BKEY_PADDED(key) temp;
425 				bkey_copy(&temp.key, k);
426 				bch_bset_insert(b, k, &temp.key);
427 				top = bkey_next(k);
428 			}
429 
430 			bch_cut_front(insert, top);
431 			bch_cut_back(&START_KEY(insert), k);
432 			bch_bset_fix_invalidated_key(b, k);
433 			goto out;
434 		}
435 
436 		if (bkey_cmp(insert, k) < 0) {
437 			bch_cut_front(insert, k);
438 		} else {
439 			if (bkey_cmp(&START_KEY(insert), &START_KEY(k)) > 0)
440 				old_offset = KEY_START(insert);
441 
442 			if (bkey_written(b, k) &&
443 			    bkey_cmp(&START_KEY(insert), &START_KEY(k)) <= 0) {
444 				/*
445 				 * Completely overwrote, so we don't have to
446 				 * invalidate the binary search tree
447 				 */
448 				bch_cut_front(k, k);
449 			} else {
450 				__bch_cut_back(&START_KEY(insert), k);
451 				bch_bset_fix_invalidated_key(b, k);
452 			}
453 		}
454 
455 		bch_subtract_dirty(k, c, old_offset, old_size - KEY_SIZE(k));
456 	}
457 
458 check_failed:
459 	if (replace_key) {
460 		if (!sectors_found) {
461 			return true;
462 		} else if (sectors_found < KEY_SIZE(insert)) {
463 			SET_KEY_OFFSET(insert, KEY_OFFSET(insert) -
464 				       (KEY_SIZE(insert) - sectors_found));
465 			SET_KEY_SIZE(insert, sectors_found);
466 		}
467 	}
468 out:
469 	if (KEY_DIRTY(insert))
470 		bcache_dev_sectors_dirty_add(c, KEY_INODE(insert),
471 					     KEY_START(insert),
472 					     KEY_SIZE(insert));
473 
474 	return false;
475 }
476 
477 bool __bch_extent_invalid(struct cache_set *c, const struct bkey *k)
478 {
479 	char buf[80];
480 
481 	if (!KEY_SIZE(k))
482 		return true;
483 
484 	if (KEY_SIZE(k) > KEY_OFFSET(k))
485 		goto bad;
486 
487 	if (__ptr_invalid(c, k))
488 		goto bad;
489 
490 	return false;
491 bad:
492 	bch_extent_to_text(buf, sizeof(buf), k);
493 	cache_bug(c, "spotted extent %s: %s", buf, bch_ptr_status(c, k));
494 	return true;
495 }
496 
497 static bool bch_extent_invalid(struct btree_keys *bk, const struct bkey *k)
498 {
499 	struct btree *b = container_of(bk, struct btree, keys);
500 	return __bch_extent_invalid(b->c, k);
501 }
502 
503 static bool bch_extent_bad_expensive(struct btree *b, const struct bkey *k,
504 				     unsigned ptr)
505 {
506 	struct bucket *g = PTR_BUCKET(b->c, k, ptr);
507 	char buf[80];
508 
509 	if (mutex_trylock(&b->c->bucket_lock)) {
510 		if (b->c->gc_mark_valid &&
511 		    (!GC_MARK(g) ||
512 		     GC_MARK(g) == GC_MARK_METADATA ||
513 		     (GC_MARK(g) != GC_MARK_DIRTY && KEY_DIRTY(k))))
514 			goto err;
515 
516 		if (g->prio == BTREE_PRIO)
517 			goto err;
518 
519 		mutex_unlock(&b->c->bucket_lock);
520 	}
521 
522 	return false;
523 err:
524 	mutex_unlock(&b->c->bucket_lock);
525 	bch_extent_to_text(buf, sizeof(buf), k);
526 	btree_bug(b,
527 "inconsistent extent pointer %s:\nbucket %zu pin %i prio %i gen %i last_gc %i mark %llu",
528 		  buf, PTR_BUCKET_NR(b->c, k, ptr), atomic_read(&g->pin),
529 		  g->prio, g->gen, g->last_gc, GC_MARK(g));
530 	return true;
531 }
532 
533 static bool bch_extent_bad(struct btree_keys *bk, const struct bkey *k)
534 {
535 	struct btree *b = container_of(bk, struct btree, keys);
536 	struct bucket *g;
537 	unsigned i, stale;
538 
539 	if (!KEY_PTRS(k) ||
540 	    bch_extent_invalid(bk, k))
541 		return true;
542 
543 	for (i = 0; i < KEY_PTRS(k); i++)
544 		if (!ptr_available(b->c, k, i))
545 			return true;
546 
547 	if (!expensive_debug_checks(b->c) && KEY_DIRTY(k))
548 		return false;
549 
550 	for (i = 0; i < KEY_PTRS(k); i++) {
551 		g = PTR_BUCKET(b->c, k, i);
552 		stale = ptr_stale(b->c, k, i);
553 
554 		btree_bug_on(stale > 96, b,
555 			     "key too stale: %i, need_gc %u",
556 			     stale, b->c->need_gc);
557 
558 		btree_bug_on(stale && KEY_DIRTY(k) && KEY_SIZE(k),
559 			     b, "stale dirty pointer");
560 
561 		if (stale)
562 			return true;
563 
564 		if (expensive_debug_checks(b->c) &&
565 		    bch_extent_bad_expensive(b, k, i))
566 			return true;
567 	}
568 
569 	return false;
570 }
571 
572 static uint64_t merge_chksums(struct bkey *l, struct bkey *r)
573 {
574 	return (l->ptr[KEY_PTRS(l)] + r->ptr[KEY_PTRS(r)]) &
575 		~((uint64_t)1 << 63);
576 }
577 
578 static bool bch_extent_merge(struct btree_keys *bk, struct bkey *l, struct bkey *r)
579 {
580 	struct btree *b = container_of(bk, struct btree, keys);
581 	unsigned i;
582 
583 	if (key_merging_disabled(b->c))
584 		return false;
585 
586 	for (i = 0; i < KEY_PTRS(l); i++)
587 		if (l->ptr[i] + PTR(0, KEY_SIZE(l), 0) != r->ptr[i] ||
588 		    PTR_BUCKET_NR(b->c, l, i) != PTR_BUCKET_NR(b->c, r, i))
589 			return false;
590 
591 	/* Keys with no pointers aren't restricted to one bucket and could
592 	 * overflow KEY_SIZE
593 	 */
594 	if (KEY_SIZE(l) + KEY_SIZE(r) > USHRT_MAX) {
595 		SET_KEY_OFFSET(l, KEY_OFFSET(l) + USHRT_MAX - KEY_SIZE(l));
596 		SET_KEY_SIZE(l, USHRT_MAX);
597 
598 		bch_cut_front(l, r);
599 		return false;
600 	}
601 
602 	if (KEY_CSUM(l)) {
603 		if (KEY_CSUM(r))
604 			l->ptr[KEY_PTRS(l)] = merge_chksums(l, r);
605 		else
606 			SET_KEY_CSUM(l, 0);
607 	}
608 
609 	SET_KEY_OFFSET(l, KEY_OFFSET(l) + KEY_SIZE(r));
610 	SET_KEY_SIZE(l, KEY_SIZE(l) + KEY_SIZE(r));
611 
612 	return true;
613 }
614 
615 const struct btree_keys_ops bch_extent_keys_ops = {
616 	.sort_cmp	= bch_extent_sort_cmp,
617 	.sort_fixup	= bch_extent_sort_fixup,
618 	.insert_fixup	= bch_extent_insert_fixup,
619 	.key_invalid	= bch_extent_invalid,
620 	.key_bad	= bch_extent_bad,
621 	.key_merge	= bch_extent_merge,
622 	.key_to_text	= bch_extent_to_text,
623 	.key_dump	= bch_bkey_dump,
624 	.is_extents	= true,
625 };
626