xref: /openbmc/linux/drivers/md/bcache/bset.c (revision 6cc23ed2)
1 // SPDX-License-Identifier: GPL-2.0
2 /*
3  * Code for working with individual keys, and sorted sets of keys with in a
4  * btree node
5  *
6  * Copyright 2012 Google, Inc.
7  */
8 
9 #define pr_fmt(fmt) "bcache: %s() " fmt "\n", __func__
10 
11 #include "util.h"
12 #include "bset.h"
13 
14 #include <linux/console.h>
15 #include <linux/sched/clock.h>
16 #include <linux/random.h>
17 #include <linux/prefetch.h>
18 
19 #ifdef CONFIG_BCACHE_DEBUG
20 
21 void bch_dump_bset(struct btree_keys *b, struct bset *i, unsigned int set)
22 {
23 	struct bkey *k, *next;
24 
25 	for (k = i->start; k < bset_bkey_last(i); k = next) {
26 		next = bkey_next(k);
27 
28 		pr_err("block %u key %u/%u: ", set,
29 		       (unsigned int) ((u64 *) k - i->d), i->keys);
30 
31 		if (b->ops->key_dump)
32 			b->ops->key_dump(b, k);
33 		else
34 			pr_err("%llu:%llu\n", KEY_INODE(k), KEY_OFFSET(k));
35 
36 		if (next < bset_bkey_last(i) &&
37 		    bkey_cmp(k, b->ops->is_extents ?
38 			     &START_KEY(next) : next) > 0)
39 			pr_err("Key skipped backwards\n");
40 	}
41 }
42 
43 void bch_dump_bucket(struct btree_keys *b)
44 {
45 	unsigned int i;
46 
47 	console_lock();
48 	for (i = 0; i <= b->nsets; i++)
49 		bch_dump_bset(b, b->set[i].data,
50 			      bset_sector_offset(b, b->set[i].data));
51 	console_unlock();
52 }
53 
54 int __bch_count_data(struct btree_keys *b)
55 {
56 	unsigned int ret = 0;
57 	struct btree_iter iter;
58 	struct bkey *k;
59 
60 	if (b->ops->is_extents)
61 		for_each_key(b, k, &iter)
62 			ret += KEY_SIZE(k);
63 	return ret;
64 }
65 
66 void __bch_check_keys(struct btree_keys *b, const char *fmt, ...)
67 {
68 	va_list args;
69 	struct bkey *k, *p = NULL;
70 	struct btree_iter iter;
71 	const char *err;
72 
73 	for_each_key(b, k, &iter) {
74 		if (b->ops->is_extents) {
75 			err = "Keys out of order";
76 			if (p && bkey_cmp(&START_KEY(p), &START_KEY(k)) > 0)
77 				goto bug;
78 
79 			if (bch_ptr_invalid(b, k))
80 				continue;
81 
82 			err =  "Overlapping keys";
83 			if (p && bkey_cmp(p, &START_KEY(k)) > 0)
84 				goto bug;
85 		} else {
86 			if (bch_ptr_bad(b, k))
87 				continue;
88 
89 			err = "Duplicate keys";
90 			if (p && !bkey_cmp(p, k))
91 				goto bug;
92 		}
93 		p = k;
94 	}
95 #if 0
96 	err = "Key larger than btree node key";
97 	if (p && bkey_cmp(p, &b->key) > 0)
98 		goto bug;
99 #endif
100 	return;
101 bug:
102 	bch_dump_bucket(b);
103 
104 	va_start(args, fmt);
105 	vprintk(fmt, args);
106 	va_end(args);
107 
108 	panic("bch_check_keys error:  %s:\n", err);
109 }
110 
111 static void bch_btree_iter_next_check(struct btree_iter *iter)
112 {
113 	struct bkey *k = iter->data->k, *next = bkey_next(k);
114 
115 	if (next < iter->data->end &&
116 	    bkey_cmp(k, iter->b->ops->is_extents ?
117 		     &START_KEY(next) : next) > 0) {
118 		bch_dump_bucket(iter->b);
119 		panic("Key skipped backwards\n");
120 	}
121 }
122 
123 #else
124 
125 static inline void bch_btree_iter_next_check(struct btree_iter *iter) {}
126 
127 #endif
128 
129 /* Keylists */
130 
131 int __bch_keylist_realloc(struct keylist *l, unsigned int u64s)
132 {
133 	size_t oldsize = bch_keylist_nkeys(l);
134 	size_t newsize = oldsize + u64s;
135 	uint64_t *old_keys = l->keys_p == l->inline_keys ? NULL : l->keys_p;
136 	uint64_t *new_keys;
137 
138 	newsize = roundup_pow_of_two(newsize);
139 
140 	if (newsize <= KEYLIST_INLINE ||
141 	    roundup_pow_of_two(oldsize) == newsize)
142 		return 0;
143 
144 	new_keys = krealloc(old_keys, sizeof(uint64_t) * newsize, GFP_NOIO);
145 
146 	if (!new_keys)
147 		return -ENOMEM;
148 
149 	if (!old_keys)
150 		memcpy(new_keys, l->inline_keys, sizeof(uint64_t) * oldsize);
151 
152 	l->keys_p = new_keys;
153 	l->top_p = new_keys + oldsize;
154 
155 	return 0;
156 }
157 
158 struct bkey *bch_keylist_pop(struct keylist *l)
159 {
160 	struct bkey *k = l->keys;
161 
162 	if (k == l->top)
163 		return NULL;
164 
165 	while (bkey_next(k) != l->top)
166 		k = bkey_next(k);
167 
168 	return l->top = k;
169 }
170 
171 void bch_keylist_pop_front(struct keylist *l)
172 {
173 	l->top_p -= bkey_u64s(l->keys);
174 
175 	memmove(l->keys,
176 		bkey_next(l->keys),
177 		bch_keylist_bytes(l));
178 }
179 
180 /* Key/pointer manipulation */
181 
182 void bch_bkey_copy_single_ptr(struct bkey *dest, const struct bkey *src,
183 			      unsigned int i)
184 {
185 	BUG_ON(i > KEY_PTRS(src));
186 
187 	/* Only copy the header, key, and one pointer. */
188 	memcpy(dest, src, 2 * sizeof(uint64_t));
189 	dest->ptr[0] = src->ptr[i];
190 	SET_KEY_PTRS(dest, 1);
191 	/* We didn't copy the checksum so clear that bit. */
192 	SET_KEY_CSUM(dest, 0);
193 }
194 
195 bool __bch_cut_front(const struct bkey *where, struct bkey *k)
196 {
197 	unsigned int i, len = 0;
198 
199 	if (bkey_cmp(where, &START_KEY(k)) <= 0)
200 		return false;
201 
202 	if (bkey_cmp(where, k) < 0)
203 		len = KEY_OFFSET(k) - KEY_OFFSET(where);
204 	else
205 		bkey_copy_key(k, where);
206 
207 	for (i = 0; i < KEY_PTRS(k); i++)
208 		SET_PTR_OFFSET(k, i, PTR_OFFSET(k, i) + KEY_SIZE(k) - len);
209 
210 	BUG_ON(len > KEY_SIZE(k));
211 	SET_KEY_SIZE(k, len);
212 	return true;
213 }
214 
215 bool __bch_cut_back(const struct bkey *where, struct bkey *k)
216 {
217 	unsigned int len = 0;
218 
219 	if (bkey_cmp(where, k) >= 0)
220 		return false;
221 
222 	BUG_ON(KEY_INODE(where) != KEY_INODE(k));
223 
224 	if (bkey_cmp(where, &START_KEY(k)) > 0)
225 		len = KEY_OFFSET(where) - KEY_START(k);
226 
227 	bkey_copy_key(k, where);
228 
229 	BUG_ON(len > KEY_SIZE(k));
230 	SET_KEY_SIZE(k, len);
231 	return true;
232 }
233 
234 /* Auxiliary search trees */
235 
236 /* 32 bits total: */
237 #define BKEY_MID_BITS		3
238 #define BKEY_EXPONENT_BITS	7
239 #define BKEY_MANTISSA_BITS	(32 - BKEY_MID_BITS - BKEY_EXPONENT_BITS)
240 #define BKEY_MANTISSA_MASK	((1 << BKEY_MANTISSA_BITS) - 1)
241 
242 struct bkey_float {
243 	unsigned int	exponent:BKEY_EXPONENT_BITS;
244 	unsigned int	m:BKEY_MID_BITS;
245 	unsigned int	mantissa:BKEY_MANTISSA_BITS;
246 } __packed;
247 
248 /*
249  * BSET_CACHELINE was originally intended to match the hardware cacheline size -
250  * it used to be 64, but I realized the lookup code would touch slightly less
251  * memory if it was 128.
252  *
253  * It definites the number of bytes (in struct bset) per struct bkey_float in
254  * the auxiliar search tree - when we're done searching the bset_float tree we
255  * have this many bytes left that we do a linear search over.
256  *
257  * Since (after level 5) every level of the bset_tree is on a new cacheline,
258  * we're touching one fewer cacheline in the bset tree in exchange for one more
259  * cacheline in the linear search - but the linear search might stop before it
260  * gets to the second cacheline.
261  */
262 
263 #define BSET_CACHELINE		128
264 
265 /* Space required for the btree node keys */
266 static inline size_t btree_keys_bytes(struct btree_keys *b)
267 {
268 	return PAGE_SIZE << b->page_order;
269 }
270 
271 static inline size_t btree_keys_cachelines(struct btree_keys *b)
272 {
273 	return btree_keys_bytes(b) / BSET_CACHELINE;
274 }
275 
276 /* Space required for the auxiliary search trees */
277 static inline size_t bset_tree_bytes(struct btree_keys *b)
278 {
279 	return btree_keys_cachelines(b) * sizeof(struct bkey_float);
280 }
281 
282 /* Space required for the prev pointers */
283 static inline size_t bset_prev_bytes(struct btree_keys *b)
284 {
285 	return btree_keys_cachelines(b) * sizeof(uint8_t);
286 }
287 
288 /* Memory allocation */
289 
290 void bch_btree_keys_free(struct btree_keys *b)
291 {
292 	struct bset_tree *t = b->set;
293 
294 	if (bset_prev_bytes(b) < PAGE_SIZE)
295 		kfree(t->prev);
296 	else
297 		free_pages((unsigned long) t->prev,
298 			   get_order(bset_prev_bytes(b)));
299 
300 	if (bset_tree_bytes(b) < PAGE_SIZE)
301 		kfree(t->tree);
302 	else
303 		free_pages((unsigned long) t->tree,
304 			   get_order(bset_tree_bytes(b)));
305 
306 	free_pages((unsigned long) t->data, b->page_order);
307 
308 	t->prev = NULL;
309 	t->tree = NULL;
310 	t->data = NULL;
311 }
312 EXPORT_SYMBOL(bch_btree_keys_free);
313 
314 int bch_btree_keys_alloc(struct btree_keys *b,
315 			 unsigned int page_order,
316 			 gfp_t gfp)
317 {
318 	struct bset_tree *t = b->set;
319 
320 	BUG_ON(t->data);
321 
322 	b->page_order = page_order;
323 
324 	t->data = (void *) __get_free_pages(gfp, b->page_order);
325 	if (!t->data)
326 		goto err;
327 
328 	t->tree = bset_tree_bytes(b) < PAGE_SIZE
329 		? kmalloc(bset_tree_bytes(b), gfp)
330 		: (void *) __get_free_pages(gfp, get_order(bset_tree_bytes(b)));
331 	if (!t->tree)
332 		goto err;
333 
334 	t->prev = bset_prev_bytes(b) < PAGE_SIZE
335 		? kmalloc(bset_prev_bytes(b), gfp)
336 		: (void *) __get_free_pages(gfp, get_order(bset_prev_bytes(b)));
337 	if (!t->prev)
338 		goto err;
339 
340 	return 0;
341 err:
342 	bch_btree_keys_free(b);
343 	return -ENOMEM;
344 }
345 EXPORT_SYMBOL(bch_btree_keys_alloc);
346 
347 void bch_btree_keys_init(struct btree_keys *b, const struct btree_keys_ops *ops,
348 			 bool *expensive_debug_checks)
349 {
350 	b->ops = ops;
351 	b->expensive_debug_checks = expensive_debug_checks;
352 	b->nsets = 0;
353 	b->last_set_unwritten = 0;
354 
355 	/*
356 	 * struct btree_keys in embedded in struct btree, and struct
357 	 * bset_tree is embedded into struct btree_keys. They are all
358 	 * initialized as 0 by kzalloc() in mca_bucket_alloc(), and
359 	 * b->set[0].data is allocated in bch_btree_keys_alloc(), so we
360 	 * don't have to initiate b->set[].size and b->set[].data here
361 	 * any more.
362 	 */
363 }
364 EXPORT_SYMBOL(bch_btree_keys_init);
365 
366 /* Binary tree stuff for auxiliary search trees */
367 
368 /*
369  * return array index next to j when does in-order traverse
370  * of a binary tree which is stored in a linear array
371  */
372 static unsigned int inorder_next(unsigned int j, unsigned int size)
373 {
374 	if (j * 2 + 1 < size) {
375 		j = j * 2 + 1;
376 
377 		while (j * 2 < size)
378 			j *= 2;
379 	} else
380 		j >>= ffz(j) + 1;
381 
382 	return j;
383 }
384 
385 /*
386  * return array index previous to j when does in-order traverse
387  * of a binary tree which is stored in a linear array
388  */
389 static unsigned int inorder_prev(unsigned int j, unsigned int size)
390 {
391 	if (j * 2 < size) {
392 		j = j * 2;
393 
394 		while (j * 2 + 1 < size)
395 			j = j * 2 + 1;
396 	} else
397 		j >>= ffs(j);
398 
399 	return j;
400 }
401 
402 /*
403  * I have no idea why this code works... and I'm the one who wrote it
404  *
405  * However, I do know what it does:
406  * Given a binary tree constructed in an array (i.e. how you normally implement
407  * a heap), it converts a node in the tree - referenced by array index - to the
408  * index it would have if you did an inorder traversal.
409  *
410  * Also tested for every j, size up to size somewhere around 6 million.
411  *
412  * The binary tree starts at array index 1, not 0
413  * extra is a function of size:
414  *   extra = (size - rounddown_pow_of_two(size - 1)) << 1;
415  */
416 static unsigned int __to_inorder(unsigned int j,
417 				  unsigned int size,
418 				  unsigned int extra)
419 {
420 	unsigned int b = fls(j);
421 	unsigned int shift = fls(size - 1) - b;
422 
423 	j  ^= 1U << (b - 1);
424 	j <<= 1;
425 	j  |= 1;
426 	j <<= shift;
427 
428 	if (j > extra)
429 		j -= (j - extra) >> 1;
430 
431 	return j;
432 }
433 
434 /*
435  * Return the cacheline index in bset_tree->data, where j is index
436  * from a linear array which stores the auxiliar binary tree
437  */
438 static unsigned int to_inorder(unsigned int j, struct bset_tree *t)
439 {
440 	return __to_inorder(j, t->size, t->extra);
441 }
442 
443 static unsigned int __inorder_to_tree(unsigned int j,
444 				      unsigned int size,
445 				      unsigned int extra)
446 {
447 	unsigned int shift;
448 
449 	if (j > extra)
450 		j += j - extra;
451 
452 	shift = ffs(j);
453 
454 	j >>= shift;
455 	j  |= roundup_pow_of_two(size) >> shift;
456 
457 	return j;
458 }
459 
460 /*
461  * Return an index from a linear array which stores the auxiliar binary
462  * tree, j is the cacheline index of t->data.
463  */
464 static unsigned int inorder_to_tree(unsigned int j, struct bset_tree *t)
465 {
466 	return __inorder_to_tree(j, t->size, t->extra);
467 }
468 
469 #if 0
470 void inorder_test(void)
471 {
472 	unsigned long done = 0;
473 	ktime_t start = ktime_get();
474 
475 	for (unsigned int size = 2;
476 	     size < 65536000;
477 	     size++) {
478 		unsigned int extra =
479 			(size - rounddown_pow_of_two(size - 1)) << 1;
480 		unsigned int i = 1, j = rounddown_pow_of_two(size - 1);
481 
482 		if (!(size % 4096))
483 			pr_notice("loop %u, %llu per us\n", size,
484 			       done / ktime_us_delta(ktime_get(), start));
485 
486 		while (1) {
487 			if (__inorder_to_tree(i, size, extra) != j)
488 				panic("size %10u j %10u i %10u", size, j, i);
489 
490 			if (__to_inorder(j, size, extra) != i)
491 				panic("size %10u j %10u i %10u", size, j, i);
492 
493 			if (j == rounddown_pow_of_two(size) - 1)
494 				break;
495 
496 			BUG_ON(inorder_prev(inorder_next(j, size), size) != j);
497 
498 			j = inorder_next(j, size);
499 			i++;
500 		}
501 
502 		done += size - 1;
503 	}
504 }
505 #endif
506 
507 /*
508  * Cacheline/offset <-> bkey pointer arithmetic:
509  *
510  * t->tree is a binary search tree in an array; each node corresponds to a key
511  * in one cacheline in t->set (BSET_CACHELINE bytes).
512  *
513  * This means we don't have to store the full index of the key that a node in
514  * the binary tree points to; to_inorder() gives us the cacheline, and then
515  * bkey_float->m gives us the offset within that cacheline, in units of 8 bytes.
516  *
517  * cacheline_to_bkey() and friends abstract out all the pointer arithmetic to
518  * make this work.
519  *
520  * To construct the bfloat for an arbitrary key we need to know what the key
521  * immediately preceding it is: we have to check if the two keys differ in the
522  * bits we're going to store in bkey_float->mantissa. t->prev[j] stores the size
523  * of the previous key so we can walk backwards to it from t->tree[j]'s key.
524  */
525 
526 static struct bkey *cacheline_to_bkey(struct bset_tree *t,
527 				      unsigned int cacheline,
528 				      unsigned int offset)
529 {
530 	return ((void *) t->data) + cacheline * BSET_CACHELINE + offset * 8;
531 }
532 
533 static unsigned int bkey_to_cacheline(struct bset_tree *t, struct bkey *k)
534 {
535 	return ((void *) k - (void *) t->data) / BSET_CACHELINE;
536 }
537 
538 static unsigned int bkey_to_cacheline_offset(struct bset_tree *t,
539 					 unsigned int cacheline,
540 					 struct bkey *k)
541 {
542 	return (u64 *) k - (u64 *) cacheline_to_bkey(t, cacheline, 0);
543 }
544 
545 static struct bkey *tree_to_bkey(struct bset_tree *t, unsigned int j)
546 {
547 	return cacheline_to_bkey(t, to_inorder(j, t), t->tree[j].m);
548 }
549 
550 static struct bkey *tree_to_prev_bkey(struct bset_tree *t, unsigned int j)
551 {
552 	return (void *) (((uint64_t *) tree_to_bkey(t, j)) - t->prev[j]);
553 }
554 
555 /*
556  * For the write set - the one we're currently inserting keys into - we don't
557  * maintain a full search tree, we just keep a simple lookup table in t->prev.
558  */
559 static struct bkey *table_to_bkey(struct bset_tree *t, unsigned int cacheline)
560 {
561 	return cacheline_to_bkey(t, cacheline, t->prev[cacheline]);
562 }
563 
564 static inline uint64_t shrd128(uint64_t high, uint64_t low, uint8_t shift)
565 {
566 	low >>= shift;
567 	low  |= (high << 1) << (63U - shift);
568 	return low;
569 }
570 
571 /*
572  * Calculate mantissa value for struct bkey_float.
573  * If most significant bit of f->exponent is not set, then
574  *  - f->exponent >> 6 is 0
575  *  - p[0] points to bkey->low
576  *  - p[-1] borrows bits from KEY_INODE() of bkey->high
577  * if most isgnificant bits of f->exponent is set, then
578  *  - f->exponent >> 6 is 1
579  *  - p[0] points to bits from KEY_INODE() of bkey->high
580  *  - p[-1] points to other bits from KEY_INODE() of
581  *    bkey->high too.
582  * See make_bfloat() to check when most significant bit of f->exponent
583  * is set or not.
584  */
585 static inline unsigned int bfloat_mantissa(const struct bkey *k,
586 				       struct bkey_float *f)
587 {
588 	const uint64_t *p = &k->low - (f->exponent >> 6);
589 
590 	return shrd128(p[-1], p[0], f->exponent & 63) & BKEY_MANTISSA_MASK;
591 }
592 
593 static void make_bfloat(struct bset_tree *t, unsigned int j)
594 {
595 	struct bkey_float *f = &t->tree[j];
596 	struct bkey *m = tree_to_bkey(t, j);
597 	struct bkey *p = tree_to_prev_bkey(t, j);
598 
599 	struct bkey *l = is_power_of_2(j)
600 		? t->data->start
601 		: tree_to_prev_bkey(t, j >> ffs(j));
602 
603 	struct bkey *r = is_power_of_2(j + 1)
604 		? bset_bkey_idx(t->data, t->data->keys - bkey_u64s(&t->end))
605 		: tree_to_bkey(t, j >> (ffz(j) + 1));
606 
607 	BUG_ON(m < l || m > r);
608 	BUG_ON(bkey_next(p) != m);
609 
610 	/*
611 	 * If l and r have different KEY_INODE values (different backing
612 	 * device), f->exponent records how many least significant bits
613 	 * are different in KEY_INODE values and sets most significant
614 	 * bits to 1 (by +64).
615 	 * If l and r have same KEY_INODE value, f->exponent records
616 	 * how many different bits in least significant bits of bkey->low.
617 	 * See bfloat_mantiss() how the most significant bit of
618 	 * f->exponent is used to calculate bfloat mantissa value.
619 	 */
620 	if (KEY_INODE(l) != KEY_INODE(r))
621 		f->exponent = fls64(KEY_INODE(r) ^ KEY_INODE(l)) + 64;
622 	else
623 		f->exponent = fls64(r->low ^ l->low);
624 
625 	f->exponent = max_t(int, f->exponent - BKEY_MANTISSA_BITS, 0);
626 
627 	/*
628 	 * Setting f->exponent = 127 flags this node as failed, and causes the
629 	 * lookup code to fall back to comparing against the original key.
630 	 */
631 
632 	if (bfloat_mantissa(m, f) != bfloat_mantissa(p, f))
633 		f->mantissa = bfloat_mantissa(m, f) - 1;
634 	else
635 		f->exponent = 127;
636 }
637 
638 static void bset_alloc_tree(struct btree_keys *b, struct bset_tree *t)
639 {
640 	if (t != b->set) {
641 		unsigned int j = roundup(t[-1].size,
642 				     64 / sizeof(struct bkey_float));
643 
644 		t->tree = t[-1].tree + j;
645 		t->prev = t[-1].prev + j;
646 	}
647 
648 	while (t < b->set + MAX_BSETS)
649 		t++->size = 0;
650 }
651 
652 static void bch_bset_build_unwritten_tree(struct btree_keys *b)
653 {
654 	struct bset_tree *t = bset_tree_last(b);
655 
656 	BUG_ON(b->last_set_unwritten);
657 	b->last_set_unwritten = 1;
658 
659 	bset_alloc_tree(b, t);
660 
661 	if (t->tree != b->set->tree + btree_keys_cachelines(b)) {
662 		t->prev[0] = bkey_to_cacheline_offset(t, 0, t->data->start);
663 		t->size = 1;
664 	}
665 }
666 
667 void bch_bset_init_next(struct btree_keys *b, struct bset *i, uint64_t magic)
668 {
669 	if (i != b->set->data) {
670 		b->set[++b->nsets].data = i;
671 		i->seq = b->set->data->seq;
672 	} else
673 		get_random_bytes(&i->seq, sizeof(uint64_t));
674 
675 	i->magic	= magic;
676 	i->version	= 0;
677 	i->keys		= 0;
678 
679 	bch_bset_build_unwritten_tree(b);
680 }
681 EXPORT_SYMBOL(bch_bset_init_next);
682 
683 /*
684  * Build auxiliary binary tree 'struct bset_tree *t', this tree is used to
685  * accelerate bkey search in a btree node (pointed by bset_tree->data in
686  * memory). After search in the auxiliar tree by calling bset_search_tree(),
687  * a struct bset_search_iter is returned which indicates range [l, r] from
688  * bset_tree->data where the searching bkey might be inside. Then a followed
689  * linear comparison does the exact search, see __bch_bset_search() for how
690  * the auxiliary tree is used.
691  */
692 void bch_bset_build_written_tree(struct btree_keys *b)
693 {
694 	struct bset_tree *t = bset_tree_last(b);
695 	struct bkey *prev = NULL, *k = t->data->start;
696 	unsigned int j, cacheline = 1;
697 
698 	b->last_set_unwritten = 0;
699 
700 	bset_alloc_tree(b, t);
701 
702 	t->size = min_t(unsigned int,
703 			bkey_to_cacheline(t, bset_bkey_last(t->data)),
704 			b->set->tree + btree_keys_cachelines(b) - t->tree);
705 
706 	if (t->size < 2) {
707 		t->size = 0;
708 		return;
709 	}
710 
711 	t->extra = (t->size - rounddown_pow_of_two(t->size - 1)) << 1;
712 
713 	/* First we figure out where the first key in each cacheline is */
714 	for (j = inorder_next(0, t->size);
715 	     j;
716 	     j = inorder_next(j, t->size)) {
717 		while (bkey_to_cacheline(t, k) < cacheline)
718 			prev = k, k = bkey_next(k);
719 
720 		t->prev[j] = bkey_u64s(prev);
721 		t->tree[j].m = bkey_to_cacheline_offset(t, cacheline++, k);
722 	}
723 
724 	while (bkey_next(k) != bset_bkey_last(t->data))
725 		k = bkey_next(k);
726 
727 	t->end = *k;
728 
729 	/* Then we build the tree */
730 	for (j = inorder_next(0, t->size);
731 	     j;
732 	     j = inorder_next(j, t->size))
733 		make_bfloat(t, j);
734 }
735 EXPORT_SYMBOL(bch_bset_build_written_tree);
736 
737 /* Insert */
738 
739 void bch_bset_fix_invalidated_key(struct btree_keys *b, struct bkey *k)
740 {
741 	struct bset_tree *t;
742 	unsigned int inorder, j = 1;
743 
744 	for (t = b->set; t <= bset_tree_last(b); t++)
745 		if (k < bset_bkey_last(t->data))
746 			goto found_set;
747 
748 	BUG();
749 found_set:
750 	if (!t->size || !bset_written(b, t))
751 		return;
752 
753 	inorder = bkey_to_cacheline(t, k);
754 
755 	if (k == t->data->start)
756 		goto fix_left;
757 
758 	if (bkey_next(k) == bset_bkey_last(t->data)) {
759 		t->end = *k;
760 		goto fix_right;
761 	}
762 
763 	j = inorder_to_tree(inorder, t);
764 
765 	if (j &&
766 	    j < t->size &&
767 	    k == tree_to_bkey(t, j))
768 fix_left:	do {
769 			make_bfloat(t, j);
770 			j = j * 2;
771 		} while (j < t->size);
772 
773 	j = inorder_to_tree(inorder + 1, t);
774 
775 	if (j &&
776 	    j < t->size &&
777 	    k == tree_to_prev_bkey(t, j))
778 fix_right:	do {
779 			make_bfloat(t, j);
780 			j = j * 2 + 1;
781 		} while (j < t->size);
782 }
783 EXPORT_SYMBOL(bch_bset_fix_invalidated_key);
784 
785 static void bch_bset_fix_lookup_table(struct btree_keys *b,
786 				      struct bset_tree *t,
787 				      struct bkey *k)
788 {
789 	unsigned int shift = bkey_u64s(k);
790 	unsigned int j = bkey_to_cacheline(t, k);
791 
792 	/* We're getting called from btree_split() or btree_gc, just bail out */
793 	if (!t->size)
794 		return;
795 
796 	/*
797 	 * k is the key we just inserted; we need to find the entry in the
798 	 * lookup table for the first key that is strictly greater than k:
799 	 * it's either k's cacheline or the next one
800 	 */
801 	while (j < t->size &&
802 	       table_to_bkey(t, j) <= k)
803 		j++;
804 
805 	/*
806 	 * Adjust all the lookup table entries, and find a new key for any that
807 	 * have gotten too big
808 	 */
809 	for (; j < t->size; j++) {
810 		t->prev[j] += shift;
811 
812 		if (t->prev[j] > 7) {
813 			k = table_to_bkey(t, j - 1);
814 
815 			while (k < cacheline_to_bkey(t, j, 0))
816 				k = bkey_next(k);
817 
818 			t->prev[j] = bkey_to_cacheline_offset(t, j, k);
819 		}
820 	}
821 
822 	if (t->size == b->set->tree + btree_keys_cachelines(b) - t->tree)
823 		return;
824 
825 	/* Possibly add a new entry to the end of the lookup table */
826 
827 	for (k = table_to_bkey(t, t->size - 1);
828 	     k != bset_bkey_last(t->data);
829 	     k = bkey_next(k))
830 		if (t->size == bkey_to_cacheline(t, k)) {
831 			t->prev[t->size] =
832 				bkey_to_cacheline_offset(t, t->size, k);
833 			t->size++;
834 		}
835 }
836 
837 /*
838  * Tries to merge l and r: l should be lower than r
839  * Returns true if we were able to merge. If we did merge, l will be the merged
840  * key, r will be untouched.
841  */
842 bool bch_bkey_try_merge(struct btree_keys *b, struct bkey *l, struct bkey *r)
843 {
844 	if (!b->ops->key_merge)
845 		return false;
846 
847 	/*
848 	 * Generic header checks
849 	 * Assumes left and right are in order
850 	 * Left and right must be exactly aligned
851 	 */
852 	if (!bch_bkey_equal_header(l, r) ||
853 	     bkey_cmp(l, &START_KEY(r)))
854 		return false;
855 
856 	return b->ops->key_merge(b, l, r);
857 }
858 EXPORT_SYMBOL(bch_bkey_try_merge);
859 
860 void bch_bset_insert(struct btree_keys *b, struct bkey *where,
861 		     struct bkey *insert)
862 {
863 	struct bset_tree *t = bset_tree_last(b);
864 
865 	BUG_ON(!b->last_set_unwritten);
866 	BUG_ON(bset_byte_offset(b, t->data) +
867 	       __set_bytes(t->data, t->data->keys + bkey_u64s(insert)) >
868 	       PAGE_SIZE << b->page_order);
869 
870 	memmove((uint64_t *) where + bkey_u64s(insert),
871 		where,
872 		(void *) bset_bkey_last(t->data) - (void *) where);
873 
874 	t->data->keys += bkey_u64s(insert);
875 	bkey_copy(where, insert);
876 	bch_bset_fix_lookup_table(b, t, where);
877 }
878 EXPORT_SYMBOL(bch_bset_insert);
879 
880 unsigned int bch_btree_insert_key(struct btree_keys *b, struct bkey *k,
881 			      struct bkey *replace_key)
882 {
883 	unsigned int status = BTREE_INSERT_STATUS_NO_INSERT;
884 	struct bset *i = bset_tree_last(b)->data;
885 	struct bkey *m, *prev = NULL;
886 	struct btree_iter iter;
887 	struct bkey preceding_key_on_stack = ZERO_KEY;
888 	struct bkey *preceding_key_p = &preceding_key_on_stack;
889 
890 	BUG_ON(b->ops->is_extents && !KEY_SIZE(k));
891 
892 	/*
893 	 * If k has preceding key, preceding_key_p will be set to address
894 	 *  of k's preceding key; otherwise preceding_key_p will be set
895 	 * to NULL inside preceding_key().
896 	 */
897 	if (b->ops->is_extents)
898 		preceding_key(&START_KEY(k), &preceding_key_p);
899 	else
900 		preceding_key(k, &preceding_key_p);
901 
902 	m = bch_btree_iter_init(b, &iter, preceding_key_p);
903 
904 	if (b->ops->insert_fixup(b, k, &iter, replace_key))
905 		return status;
906 
907 	status = BTREE_INSERT_STATUS_INSERT;
908 
909 	while (m != bset_bkey_last(i) &&
910 	       bkey_cmp(k, b->ops->is_extents ? &START_KEY(m) : m) > 0)
911 		prev = m, m = bkey_next(m);
912 
913 	/* prev is in the tree, if we merge we're done */
914 	status = BTREE_INSERT_STATUS_BACK_MERGE;
915 	if (prev &&
916 	    bch_bkey_try_merge(b, prev, k))
917 		goto merged;
918 #if 0
919 	status = BTREE_INSERT_STATUS_OVERWROTE;
920 	if (m != bset_bkey_last(i) &&
921 	    KEY_PTRS(m) == KEY_PTRS(k) && !KEY_SIZE(m))
922 		goto copy;
923 #endif
924 	status = BTREE_INSERT_STATUS_FRONT_MERGE;
925 	if (m != bset_bkey_last(i) &&
926 	    bch_bkey_try_merge(b, k, m))
927 		goto copy;
928 
929 	bch_bset_insert(b, m, k);
930 copy:	bkey_copy(m, k);
931 merged:
932 	return status;
933 }
934 EXPORT_SYMBOL(bch_btree_insert_key);
935 
936 /* Lookup */
937 
938 struct bset_search_iter {
939 	struct bkey *l, *r;
940 };
941 
942 static struct bset_search_iter bset_search_write_set(struct bset_tree *t,
943 						     const struct bkey *search)
944 {
945 	unsigned int li = 0, ri = t->size;
946 
947 	while (li + 1 != ri) {
948 		unsigned int m = (li + ri) >> 1;
949 
950 		if (bkey_cmp(table_to_bkey(t, m), search) > 0)
951 			ri = m;
952 		else
953 			li = m;
954 	}
955 
956 	return (struct bset_search_iter) {
957 		table_to_bkey(t, li),
958 		ri < t->size ? table_to_bkey(t, ri) : bset_bkey_last(t->data)
959 	};
960 }
961 
962 static struct bset_search_iter bset_search_tree(struct bset_tree *t,
963 						const struct bkey *search)
964 {
965 	struct bkey *l, *r;
966 	struct bkey_float *f;
967 	unsigned int inorder, j, n = 1;
968 
969 	do {
970 		unsigned int p = n << 4;
971 
972 		if (p < t->size)
973 			prefetch(&t->tree[p]);
974 
975 		j = n;
976 		f = &t->tree[j];
977 
978 		if (likely(f->exponent != 127)) {
979 			if (f->mantissa >= bfloat_mantissa(search, f))
980 				n = j * 2;
981 			else
982 				n = j * 2 + 1;
983 		} else {
984 			if (bkey_cmp(tree_to_bkey(t, j), search) > 0)
985 				n = j * 2;
986 			else
987 				n = j * 2 + 1;
988 		}
989 	} while (n < t->size);
990 
991 	inorder = to_inorder(j, t);
992 
993 	/*
994 	 * n would have been the node we recursed to - the low bit tells us if
995 	 * we recursed left or recursed right.
996 	 */
997 	if (n & 1) {
998 		l = cacheline_to_bkey(t, inorder, f->m);
999 
1000 		if (++inorder != t->size) {
1001 			f = &t->tree[inorder_next(j, t->size)];
1002 			r = cacheline_to_bkey(t, inorder, f->m);
1003 		} else
1004 			r = bset_bkey_last(t->data);
1005 	} else {
1006 		r = cacheline_to_bkey(t, inorder, f->m);
1007 
1008 		if (--inorder) {
1009 			f = &t->tree[inorder_prev(j, t->size)];
1010 			l = cacheline_to_bkey(t, inorder, f->m);
1011 		} else
1012 			l = t->data->start;
1013 	}
1014 
1015 	return (struct bset_search_iter) {l, r};
1016 }
1017 
1018 struct bkey *__bch_bset_search(struct btree_keys *b, struct bset_tree *t,
1019 			       const struct bkey *search)
1020 {
1021 	struct bset_search_iter i;
1022 
1023 	/*
1024 	 * First, we search for a cacheline, then lastly we do a linear search
1025 	 * within that cacheline.
1026 	 *
1027 	 * To search for the cacheline, there's three different possibilities:
1028 	 *  * The set is too small to have a search tree, so we just do a linear
1029 	 *    search over the whole set.
1030 	 *  * The set is the one we're currently inserting into; keeping a full
1031 	 *    auxiliary search tree up to date would be too expensive, so we
1032 	 *    use a much simpler lookup table to do a binary search -
1033 	 *    bset_search_write_set().
1034 	 *  * Or we use the auxiliary search tree we constructed earlier -
1035 	 *    bset_search_tree()
1036 	 */
1037 
1038 	if (unlikely(!t->size)) {
1039 		i.l = t->data->start;
1040 		i.r = bset_bkey_last(t->data);
1041 	} else if (bset_written(b, t)) {
1042 		/*
1043 		 * Each node in the auxiliary search tree covers a certain range
1044 		 * of bits, and keys above and below the set it covers might
1045 		 * differ outside those bits - so we have to special case the
1046 		 * start and end - handle that here:
1047 		 */
1048 
1049 		if (unlikely(bkey_cmp(search, &t->end) >= 0))
1050 			return bset_bkey_last(t->data);
1051 
1052 		if (unlikely(bkey_cmp(search, t->data->start) < 0))
1053 			return t->data->start;
1054 
1055 		i = bset_search_tree(t, search);
1056 	} else {
1057 		BUG_ON(!b->nsets &&
1058 		       t->size < bkey_to_cacheline(t, bset_bkey_last(t->data)));
1059 
1060 		i = bset_search_write_set(t, search);
1061 	}
1062 
1063 	if (btree_keys_expensive_checks(b)) {
1064 		BUG_ON(bset_written(b, t) &&
1065 		       i.l != t->data->start &&
1066 		       bkey_cmp(tree_to_prev_bkey(t,
1067 			  inorder_to_tree(bkey_to_cacheline(t, i.l), t)),
1068 				search) > 0);
1069 
1070 		BUG_ON(i.r != bset_bkey_last(t->data) &&
1071 		       bkey_cmp(i.r, search) <= 0);
1072 	}
1073 
1074 	while (likely(i.l != i.r) &&
1075 	       bkey_cmp(i.l, search) <= 0)
1076 		i.l = bkey_next(i.l);
1077 
1078 	return i.l;
1079 }
1080 EXPORT_SYMBOL(__bch_bset_search);
1081 
1082 /* Btree iterator */
1083 
1084 typedef bool (btree_iter_cmp_fn)(struct btree_iter_set,
1085 				 struct btree_iter_set);
1086 
1087 static inline bool btree_iter_cmp(struct btree_iter_set l,
1088 				  struct btree_iter_set r)
1089 {
1090 	return bkey_cmp(l.k, r.k) > 0;
1091 }
1092 
1093 static inline bool btree_iter_end(struct btree_iter *iter)
1094 {
1095 	return !iter->used;
1096 }
1097 
1098 void bch_btree_iter_push(struct btree_iter *iter, struct bkey *k,
1099 			 struct bkey *end)
1100 {
1101 	if (k != end)
1102 		BUG_ON(!heap_add(iter,
1103 				 ((struct btree_iter_set) { k, end }),
1104 				 btree_iter_cmp));
1105 }
1106 
1107 static struct bkey *__bch_btree_iter_init(struct btree_keys *b,
1108 					  struct btree_iter *iter,
1109 					  struct bkey *search,
1110 					  struct bset_tree *start)
1111 {
1112 	struct bkey *ret = NULL;
1113 
1114 	iter->size = ARRAY_SIZE(iter->data);
1115 	iter->used = 0;
1116 
1117 #ifdef CONFIG_BCACHE_DEBUG
1118 	iter->b = b;
1119 #endif
1120 
1121 	for (; start <= bset_tree_last(b); start++) {
1122 		ret = bch_bset_search(b, start, search);
1123 		bch_btree_iter_push(iter, ret, bset_bkey_last(start->data));
1124 	}
1125 
1126 	return ret;
1127 }
1128 
1129 struct bkey *bch_btree_iter_init(struct btree_keys *b,
1130 				 struct btree_iter *iter,
1131 				 struct bkey *search)
1132 {
1133 	return __bch_btree_iter_init(b, iter, search, b->set);
1134 }
1135 EXPORT_SYMBOL(bch_btree_iter_init);
1136 
1137 static inline struct bkey *__bch_btree_iter_next(struct btree_iter *iter,
1138 						 btree_iter_cmp_fn *cmp)
1139 {
1140 	struct btree_iter_set b __maybe_unused;
1141 	struct bkey *ret = NULL;
1142 
1143 	if (!btree_iter_end(iter)) {
1144 		bch_btree_iter_next_check(iter);
1145 
1146 		ret = iter->data->k;
1147 		iter->data->k = bkey_next(iter->data->k);
1148 
1149 		if (iter->data->k > iter->data->end) {
1150 			WARN_ONCE(1, "bset was corrupt!\n");
1151 			iter->data->k = iter->data->end;
1152 		}
1153 
1154 		if (iter->data->k == iter->data->end)
1155 			heap_pop(iter, b, cmp);
1156 		else
1157 			heap_sift(iter, 0, cmp);
1158 	}
1159 
1160 	return ret;
1161 }
1162 
1163 struct bkey *bch_btree_iter_next(struct btree_iter *iter)
1164 {
1165 	return __bch_btree_iter_next(iter, btree_iter_cmp);
1166 
1167 }
1168 EXPORT_SYMBOL(bch_btree_iter_next);
1169 
1170 struct bkey *bch_btree_iter_next_filter(struct btree_iter *iter,
1171 					struct btree_keys *b, ptr_filter_fn fn)
1172 {
1173 	struct bkey *ret;
1174 
1175 	do {
1176 		ret = bch_btree_iter_next(iter);
1177 	} while (ret && fn(b, ret));
1178 
1179 	return ret;
1180 }
1181 
1182 /* Mergesort */
1183 
1184 void bch_bset_sort_state_free(struct bset_sort_state *state)
1185 {
1186 	mempool_exit(&state->pool);
1187 }
1188 
1189 int bch_bset_sort_state_init(struct bset_sort_state *state,
1190 			     unsigned int page_order)
1191 {
1192 	spin_lock_init(&state->time.lock);
1193 
1194 	state->page_order = page_order;
1195 	state->crit_factor = int_sqrt(1 << page_order);
1196 
1197 	return mempool_init_page_pool(&state->pool, 1, page_order);
1198 }
1199 EXPORT_SYMBOL(bch_bset_sort_state_init);
1200 
1201 static void btree_mergesort(struct btree_keys *b, struct bset *out,
1202 			    struct btree_iter *iter,
1203 			    bool fixup, bool remove_stale)
1204 {
1205 	int i;
1206 	struct bkey *k, *last = NULL;
1207 	BKEY_PADDED(k) tmp;
1208 	bool (*bad)(struct btree_keys *, const struct bkey *) = remove_stale
1209 		? bch_ptr_bad
1210 		: bch_ptr_invalid;
1211 
1212 	/* Heapify the iterator, using our comparison function */
1213 	for (i = iter->used / 2 - 1; i >= 0; --i)
1214 		heap_sift(iter, i, b->ops->sort_cmp);
1215 
1216 	while (!btree_iter_end(iter)) {
1217 		if (b->ops->sort_fixup && fixup)
1218 			k = b->ops->sort_fixup(iter, &tmp.k);
1219 		else
1220 			k = NULL;
1221 
1222 		if (!k)
1223 			k = __bch_btree_iter_next(iter, b->ops->sort_cmp);
1224 
1225 		if (bad(b, k))
1226 			continue;
1227 
1228 		if (!last) {
1229 			last = out->start;
1230 			bkey_copy(last, k);
1231 		} else if (!bch_bkey_try_merge(b, last, k)) {
1232 			last = bkey_next(last);
1233 			bkey_copy(last, k);
1234 		}
1235 	}
1236 
1237 	out->keys = last ? (uint64_t *) bkey_next(last) - out->d : 0;
1238 
1239 	pr_debug("sorted %i keys", out->keys);
1240 }
1241 
1242 static void __btree_sort(struct btree_keys *b, struct btree_iter *iter,
1243 			 unsigned int start, unsigned int order, bool fixup,
1244 			 struct bset_sort_state *state)
1245 {
1246 	uint64_t start_time;
1247 	bool used_mempool = false;
1248 	struct bset *out = (void *) __get_free_pages(__GFP_NOWARN|GFP_NOWAIT,
1249 						     order);
1250 	if (!out) {
1251 		struct page *outp;
1252 
1253 		BUG_ON(order > state->page_order);
1254 
1255 		outp = mempool_alloc(&state->pool, GFP_NOIO);
1256 		out = page_address(outp);
1257 		used_mempool = true;
1258 		order = state->page_order;
1259 	}
1260 
1261 	start_time = local_clock();
1262 
1263 	btree_mergesort(b, out, iter, fixup, false);
1264 	b->nsets = start;
1265 
1266 	if (!start && order == b->page_order) {
1267 		/*
1268 		 * Our temporary buffer is the same size as the btree node's
1269 		 * buffer, we can just swap buffers instead of doing a big
1270 		 * memcpy()
1271 		 */
1272 
1273 		out->magic	= b->set->data->magic;
1274 		out->seq	= b->set->data->seq;
1275 		out->version	= b->set->data->version;
1276 		swap(out, b->set->data);
1277 	} else {
1278 		b->set[start].data->keys = out->keys;
1279 		memcpy(b->set[start].data->start, out->start,
1280 		       (void *) bset_bkey_last(out) - (void *) out->start);
1281 	}
1282 
1283 	if (used_mempool)
1284 		mempool_free(virt_to_page(out), &state->pool);
1285 	else
1286 		free_pages((unsigned long) out, order);
1287 
1288 	bch_bset_build_written_tree(b);
1289 
1290 	if (!start)
1291 		bch_time_stats_update(&state->time, start_time);
1292 }
1293 
1294 void bch_btree_sort_partial(struct btree_keys *b, unsigned int start,
1295 			    struct bset_sort_state *state)
1296 {
1297 	size_t order = b->page_order, keys = 0;
1298 	struct btree_iter iter;
1299 	int oldsize = bch_count_data(b);
1300 
1301 	__bch_btree_iter_init(b, &iter, NULL, &b->set[start]);
1302 
1303 	if (start) {
1304 		unsigned int i;
1305 
1306 		for (i = start; i <= b->nsets; i++)
1307 			keys += b->set[i].data->keys;
1308 
1309 		order = get_order(__set_bytes(b->set->data, keys));
1310 	}
1311 
1312 	__btree_sort(b, &iter, start, order, false, state);
1313 
1314 	EBUG_ON(oldsize >= 0 && bch_count_data(b) != oldsize);
1315 }
1316 EXPORT_SYMBOL(bch_btree_sort_partial);
1317 
1318 void bch_btree_sort_and_fix_extents(struct btree_keys *b,
1319 				    struct btree_iter *iter,
1320 				    struct bset_sort_state *state)
1321 {
1322 	__btree_sort(b, iter, 0, b->page_order, true, state);
1323 }
1324 
1325 void bch_btree_sort_into(struct btree_keys *b, struct btree_keys *new,
1326 			 struct bset_sort_state *state)
1327 {
1328 	uint64_t start_time = local_clock();
1329 	struct btree_iter iter;
1330 
1331 	bch_btree_iter_init(b, &iter, NULL);
1332 
1333 	btree_mergesort(b, new->set->data, &iter, false, true);
1334 
1335 	bch_time_stats_update(&state->time, start_time);
1336 
1337 	new->set->size = 0; // XXX: why?
1338 }
1339 
1340 #define SORT_CRIT	(4096 / sizeof(uint64_t))
1341 
1342 void bch_btree_sort_lazy(struct btree_keys *b, struct bset_sort_state *state)
1343 {
1344 	unsigned int crit = SORT_CRIT;
1345 	int i;
1346 
1347 	/* Don't sort if nothing to do */
1348 	if (!b->nsets)
1349 		goto out;
1350 
1351 	for (i = b->nsets - 1; i >= 0; --i) {
1352 		crit *= state->crit_factor;
1353 
1354 		if (b->set[i].data->keys < crit) {
1355 			bch_btree_sort_partial(b, i, state);
1356 			return;
1357 		}
1358 	}
1359 
1360 	/* Sort if we'd overflow */
1361 	if (b->nsets + 1 == MAX_BSETS) {
1362 		bch_btree_sort(b, state);
1363 		return;
1364 	}
1365 
1366 out:
1367 	bch_bset_build_written_tree(b);
1368 }
1369 EXPORT_SYMBOL(bch_btree_sort_lazy);
1370 
1371 void bch_btree_keys_stats(struct btree_keys *b, struct bset_stats *stats)
1372 {
1373 	unsigned int i;
1374 
1375 	for (i = 0; i <= b->nsets; i++) {
1376 		struct bset_tree *t = &b->set[i];
1377 		size_t bytes = t->data->keys * sizeof(uint64_t);
1378 		size_t j;
1379 
1380 		if (bset_written(b, t)) {
1381 			stats->sets_written++;
1382 			stats->bytes_written += bytes;
1383 
1384 			stats->floats += t->size - 1;
1385 
1386 			for (j = 1; j < t->size; j++)
1387 				if (t->tree[j].exponent == 127)
1388 					stats->failed++;
1389 		} else {
1390 			stats->sets_unwritten++;
1391 			stats->bytes_unwritten += bytes;
1392 		}
1393 	}
1394 }
1395