GUI: Fix Tomato RAF theme for all builds. Compilation typo.
[tomato.git] / release / src-rt-6.x.4708 / linux / linux-2.6.36 / lib / btree.c
blob067c686b333a5d07ff423dffbb54a27c7cd32bb2
1 /*
2 * lib/btree.c - Simple In-memory B+Tree
4 * As should be obvious for Linux kernel code, license is GPLv2
6 * Copyright (c) 2007-2008 Joern Engel <joern@logfs.org>
7 * Bits and pieces stolen from Peter Zijlstra's code, which is
8 * Copyright 2007, Red Hat Inc. Peter Zijlstra <pzijlstr@redhat.com>
9 * GPLv2
11 * see http://programming.kicks-ass.net/kernel-patches/vma_lookup/btree.patch
13 * A relatively simple B+Tree implementation. I have written it as a learning
14 * excercise to understand how B+Trees work. Turned out to be useful as well.
16 * B+Trees can be used similar to Linux radix trees (which don't have anything
17 * in common with textbook radix trees, beware). Prerequisite for them working
18 * well is that access to a random tree node is much faster than a large number
19 * of operations within each node.
21 * Disks have fulfilled the prerequisite for a long time. More recently DRAM
22 * has gained similar properties, as memory access times, when measured in cpu
23 * cycles, have increased. Cacheline sizes have increased as well, which also
24 * helps B+Trees.
26 * Compared to radix trees, B+Trees are more efficient when dealing with a
27 * sparsely populated address space. Between 25% and 50% of the memory is
28 * occupied with valid pointers. When densely populated, radix trees contain
29 * ~98% pointers - hard to beat. Very sparse radix trees contain only ~2%
30 * pointers.
32 * This particular implementation stores pointers identified by a long value.
33 * Storing NULL pointers is illegal, lookup will return NULL when no entry
34 * was found.
36 * A tricks was used that is not commonly found in textbooks. The lowest
37 * values are to the right, not to the left. All used slots within a node
38 * are on the left, all unused slots contain NUL values. Most operations
39 * simply loop once over all slots and terminate on the first NUL.
42 #include <linux/btree.h>
43 #include <linux/cache.h>
44 #include <linux/kernel.h>
45 #include <linux/slab.h>
46 #include <linux/module.h>
48 #define MAX(a, b) ((a) > (b) ? (a) : (b))
49 #define NODESIZE MAX(L1_CACHE_BYTES, 128)
51 struct btree_geo {
52 int keylen;
53 int no_pairs;
54 int no_longs;
57 struct btree_geo btree_geo32 = {
58 .keylen = 1,
59 .no_pairs = NODESIZE / sizeof(long) / 2,
60 .no_longs = NODESIZE / sizeof(long) / 2,
62 EXPORT_SYMBOL_GPL(btree_geo32);
64 #define LONG_PER_U64 (64 / BITS_PER_LONG)
65 struct btree_geo btree_geo64 = {
66 .keylen = LONG_PER_U64,
67 .no_pairs = NODESIZE / sizeof(long) / (1 + LONG_PER_U64),
68 .no_longs = LONG_PER_U64 * (NODESIZE / sizeof(long) / (1 + LONG_PER_U64)),
70 EXPORT_SYMBOL_GPL(btree_geo64);
72 struct btree_geo btree_geo128 = {
73 .keylen = 2 * LONG_PER_U64,
74 .no_pairs = NODESIZE / sizeof(long) / (1 + 2 * LONG_PER_U64),
75 .no_longs = 2 * LONG_PER_U64 * (NODESIZE / sizeof(long) / (1 + 2 * LONG_PER_U64)),
77 EXPORT_SYMBOL_GPL(btree_geo128);
79 static struct kmem_cache *btree_cachep;
81 void *btree_alloc(gfp_t gfp_mask, void *pool_data)
83 return kmem_cache_alloc(btree_cachep, gfp_mask);
85 EXPORT_SYMBOL_GPL(btree_alloc);
87 void btree_free(void *element, void *pool_data)
89 kmem_cache_free(btree_cachep, element);
91 EXPORT_SYMBOL_GPL(btree_free);
93 static unsigned long *btree_node_alloc(struct btree_head *head, gfp_t gfp)
95 unsigned long *node;
97 node = mempool_alloc(head->mempool, gfp);
98 if (likely(node))
99 memset(node, 0, NODESIZE);
100 return node;
103 static int longcmp(const unsigned long *l1, const unsigned long *l2, size_t n)
105 size_t i;
107 for (i = 0; i < n; i++) {
108 if (l1[i] < l2[i])
109 return -1;
110 if (l1[i] > l2[i])
111 return 1;
113 return 0;
116 static unsigned long *longcpy(unsigned long *dest, const unsigned long *src,
117 size_t n)
119 size_t i;
121 for (i = 0; i < n; i++)
122 dest[i] = src[i];
123 return dest;
126 static unsigned long *longset(unsigned long *s, unsigned long c, size_t n)
128 size_t i;
130 for (i = 0; i < n; i++)
131 s[i] = c;
132 return s;
135 static void dec_key(struct btree_geo *geo, unsigned long *key)
137 unsigned long val;
138 int i;
140 for (i = geo->keylen - 1; i >= 0; i--) {
141 val = key[i];
142 key[i] = val - 1;
143 if (val)
144 break;
148 static unsigned long *bkey(struct btree_geo *geo, unsigned long *node, int n)
150 return &node[n * geo->keylen];
153 static void *bval(struct btree_geo *geo, unsigned long *node, int n)
155 return (void *)node[geo->no_longs + n];
158 static void setkey(struct btree_geo *geo, unsigned long *node, int n,
159 unsigned long *key)
161 longcpy(bkey(geo, node, n), key, geo->keylen);
164 static void setval(struct btree_geo *geo, unsigned long *node, int n,
165 void *val)
167 node[geo->no_longs + n] = (unsigned long) val;
170 static void clearpair(struct btree_geo *geo, unsigned long *node, int n)
172 longset(bkey(geo, node, n), 0, geo->keylen);
173 node[geo->no_longs + n] = 0;
176 static inline void __btree_init(struct btree_head *head)
178 head->node = NULL;
179 head->height = 0;
182 void btree_init_mempool(struct btree_head *head, mempool_t *mempool)
184 __btree_init(head);
185 head->mempool = mempool;
187 EXPORT_SYMBOL_GPL(btree_init_mempool);
189 int btree_init(struct btree_head *head)
191 __btree_init(head);
192 head->mempool = mempool_create(0, btree_alloc, btree_free, NULL);
193 if (!head->mempool)
194 return -ENOMEM;
195 return 0;
197 EXPORT_SYMBOL_GPL(btree_init);
199 void btree_destroy(struct btree_head *head)
201 mempool_destroy(head->mempool);
202 head->mempool = NULL;
204 EXPORT_SYMBOL_GPL(btree_destroy);
206 void *btree_last(struct btree_head *head, struct btree_geo *geo,
207 unsigned long *key)
209 int height = head->height;
210 unsigned long *node = head->node;
212 if (height == 0)
213 return NULL;
215 for ( ; height > 1; height--)
216 node = bval(geo, node, 0);
218 longcpy(key, bkey(geo, node, 0), geo->keylen);
219 return bval(geo, node, 0);
221 EXPORT_SYMBOL_GPL(btree_last);
223 static int keycmp(struct btree_geo *geo, unsigned long *node, int pos,
224 unsigned long *key)
226 return longcmp(bkey(geo, node, pos), key, geo->keylen);
229 static int keyzero(struct btree_geo *geo, unsigned long *key)
231 int i;
233 for (i = 0; i < geo->keylen; i++)
234 if (key[i])
235 return 0;
237 return 1;
240 void *btree_lookup(struct btree_head *head, struct btree_geo *geo,
241 unsigned long *key)
243 int i, height = head->height;
244 unsigned long *node = head->node;
246 if (height == 0)
247 return NULL;
249 for ( ; height > 1; height--) {
250 for (i = 0; i < geo->no_pairs; i++)
251 if (keycmp(geo, node, i, key) <= 0)
252 break;
253 if (i == geo->no_pairs)
254 return NULL;
255 node = bval(geo, node, i);
256 if (!node)
257 return NULL;
260 if (!node)
261 return NULL;
263 for (i = 0; i < geo->no_pairs; i++)
264 if (keycmp(geo, node, i, key) == 0)
265 return bval(geo, node, i);
266 return NULL;
268 EXPORT_SYMBOL_GPL(btree_lookup);
270 int btree_update(struct btree_head *head, struct btree_geo *geo,
271 unsigned long *key, void *val)
273 int i, height = head->height;
274 unsigned long *node = head->node;
276 if (height == 0)
277 return -ENOENT;
279 for ( ; height > 1; height--) {
280 for (i = 0; i < geo->no_pairs; i++)
281 if (keycmp(geo, node, i, key) <= 0)
282 break;
283 if (i == geo->no_pairs)
284 return -ENOENT;
285 node = bval(geo, node, i);
286 if (!node)
287 return -ENOENT;
290 if (!node)
291 return -ENOENT;
293 for (i = 0; i < geo->no_pairs; i++)
294 if (keycmp(geo, node, i, key) == 0) {
295 setval(geo, node, i, val);
296 return 0;
298 return -ENOENT;
300 EXPORT_SYMBOL_GPL(btree_update);
303 * Usually this function is quite similar to normal lookup. But the key of
304 * a parent node may be smaller than the smallest key of all its siblings.
305 * In such a case we cannot just return NULL, as we have only proven that no
306 * key smaller than __key, but larger than this parent key exists.
307 * So we set __key to the parent key and retry. We have to use the smallest
308 * such parent key, which is the last parent key we encountered.
310 void *btree_get_prev(struct btree_head *head, struct btree_geo *geo,
311 unsigned long *__key)
313 int i, height;
314 unsigned long *node, *oldnode;
315 unsigned long *retry_key = NULL, key[geo->keylen];
317 if (keyzero(geo, __key))
318 return NULL;
320 if (head->height == 0)
321 return NULL;
322 retry:
323 longcpy(key, __key, geo->keylen);
324 dec_key(geo, key);
326 node = head->node;
327 for (height = head->height ; height > 1; height--) {
328 for (i = 0; i < geo->no_pairs; i++)
329 if (keycmp(geo, node, i, key) <= 0)
330 break;
331 if (i == geo->no_pairs)
332 goto miss;
333 oldnode = node;
334 node = bval(geo, node, i);
335 if (!node)
336 goto miss;
337 retry_key = bkey(geo, oldnode, i);
340 if (!node)
341 goto miss;
343 for (i = 0; i < geo->no_pairs; i++) {
344 if (keycmp(geo, node, i, key) <= 0) {
345 if (bval(geo, node, i)) {
346 longcpy(__key, bkey(geo, node, i), geo->keylen);
347 return bval(geo, node, i);
348 } else
349 goto miss;
352 miss:
353 if (retry_key) {
354 __key = retry_key;
355 retry_key = NULL;
356 goto retry;
358 return NULL;
361 static int getpos(struct btree_geo *geo, unsigned long *node,
362 unsigned long *key)
364 int i;
366 for (i = 0; i < geo->no_pairs; i++) {
367 if (keycmp(geo, node, i, key) <= 0)
368 break;
370 return i;
373 static int getfill(struct btree_geo *geo, unsigned long *node, int start)
375 int i;
377 for (i = start; i < geo->no_pairs; i++)
378 if (!bval(geo, node, i))
379 break;
380 return i;
384 * locate the correct leaf node in the btree
386 static unsigned long *find_level(struct btree_head *head, struct btree_geo *geo,
387 unsigned long *key, int level)
389 unsigned long *node = head->node;
390 int i, height;
392 for (height = head->height; height > level; height--) {
393 for (i = 0; i < geo->no_pairs; i++)
394 if (keycmp(geo, node, i, key) <= 0)
395 break;
397 if ((i == geo->no_pairs) || !bval(geo, node, i)) {
398 /* right-most key is too large, update it */
399 i--;
400 setkey(geo, node, i, key);
402 BUG_ON(i < 0);
403 node = bval(geo, node, i);
405 BUG_ON(!node);
406 return node;
409 static int btree_grow(struct btree_head *head, struct btree_geo *geo,
410 gfp_t gfp)
412 unsigned long *node;
413 int fill;
415 node = btree_node_alloc(head, gfp);
416 if (!node)
417 return -ENOMEM;
418 if (head->node) {
419 fill = getfill(geo, head->node, 0);
420 setkey(geo, node, 0, bkey(geo, head->node, fill - 1));
421 setval(geo, node, 0, head->node);
423 head->node = node;
424 head->height++;
425 return 0;
428 static void btree_shrink(struct btree_head *head, struct btree_geo *geo)
430 unsigned long *node;
431 int fill;
433 if (head->height <= 1)
434 return;
436 node = head->node;
437 fill = getfill(geo, node, 0);
438 BUG_ON(fill > 1);
439 head->node = bval(geo, node, 0);
440 head->height--;
441 mempool_free(node, head->mempool);
444 static int btree_insert_level(struct btree_head *head, struct btree_geo *geo,
445 unsigned long *key, void *val, int level,
446 gfp_t gfp)
448 unsigned long *node;
449 int i, pos, fill, err;
451 BUG_ON(!val);
452 if (head->height < level) {
453 err = btree_grow(head, geo, gfp);
454 if (err)
455 return err;
458 retry:
459 node = find_level(head, geo, key, level);
460 pos = getpos(geo, node, key);
461 fill = getfill(geo, node, pos);
462 /* two identical keys are not allowed */
463 BUG_ON(pos < fill && keycmp(geo, node, pos, key) == 0);
465 if (fill == geo->no_pairs) {
466 /* need to split node */
467 unsigned long *new;
469 new = btree_node_alloc(head, gfp);
470 if (!new)
471 return -ENOMEM;
472 err = btree_insert_level(head, geo,
473 bkey(geo, node, fill / 2 - 1),
474 new, level + 1, gfp);
475 if (err) {
476 mempool_free(new, head->mempool);
477 return err;
479 for (i = 0; i < fill / 2; i++) {
480 setkey(geo, new, i, bkey(geo, node, i));
481 setval(geo, new, i, bval(geo, node, i));
482 setkey(geo, node, i, bkey(geo, node, i + fill / 2));
483 setval(geo, node, i, bval(geo, node, i + fill / 2));
484 clearpair(geo, node, i + fill / 2);
486 if (fill & 1) {
487 setkey(geo, node, i, bkey(geo, node, fill - 1));
488 setval(geo, node, i, bval(geo, node, fill - 1));
489 clearpair(geo, node, fill - 1);
491 goto retry;
493 BUG_ON(fill >= geo->no_pairs);
495 /* shift and insert */
496 for (i = fill; i > pos; i--) {
497 setkey(geo, node, i, bkey(geo, node, i - 1));
498 setval(geo, node, i, bval(geo, node, i - 1));
500 setkey(geo, node, pos, key);
501 setval(geo, node, pos, val);
503 return 0;
506 int btree_insert(struct btree_head *head, struct btree_geo *geo,
507 unsigned long *key, void *val, gfp_t gfp)
509 return btree_insert_level(head, geo, key, val, 1, gfp);
511 EXPORT_SYMBOL_GPL(btree_insert);
513 static void *btree_remove_level(struct btree_head *head, struct btree_geo *geo,
514 unsigned long *key, int level);
515 static void merge(struct btree_head *head, struct btree_geo *geo, int level,
516 unsigned long *left, int lfill,
517 unsigned long *right, int rfill,
518 unsigned long *parent, int lpos)
520 int i;
522 for (i = 0; i < rfill; i++) {
523 /* Move all keys to the left */
524 setkey(geo, left, lfill + i, bkey(geo, right, i));
525 setval(geo, left, lfill + i, bval(geo, right, i));
527 /* Exchange left and right child in parent */
528 setval(geo, parent, lpos, right);
529 setval(geo, parent, lpos + 1, left);
530 /* Remove left (formerly right) child from parent */
531 btree_remove_level(head, geo, bkey(geo, parent, lpos), level + 1);
532 mempool_free(right, head->mempool);
535 static void rebalance(struct btree_head *head, struct btree_geo *geo,
536 unsigned long *key, int level, unsigned long *child, int fill)
538 unsigned long *parent, *left = NULL, *right = NULL;
539 int i, no_left, no_right;
541 if (fill == 0) {
542 /* Because we don't steal entries from a neigbour, this case
543 * can happen. Parent node contains a single child, this
544 * node, so merging with a sibling never happens.
546 btree_remove_level(head, geo, key, level + 1);
547 mempool_free(child, head->mempool);
548 return;
551 parent = find_level(head, geo, key, level + 1);
552 i = getpos(geo, parent, key);
553 BUG_ON(bval(geo, parent, i) != child);
555 if (i > 0) {
556 left = bval(geo, parent, i - 1);
557 no_left = getfill(geo, left, 0);
558 if (fill + no_left <= geo->no_pairs) {
559 merge(head, geo, level,
560 left, no_left,
561 child, fill,
562 parent, i - 1);
563 return;
566 if (i + 1 < getfill(geo, parent, i)) {
567 right = bval(geo, parent, i + 1);
568 no_right = getfill(geo, right, 0);
569 if (fill + no_right <= geo->no_pairs) {
570 merge(head, geo, level,
571 child, fill,
572 right, no_right,
573 parent, i);
574 return;
578 * We could also try to steal one entry from the left or right
579 * neighbor. By not doing so we changed the invariant from
580 * "all nodes are at least half full" to "no two neighboring
581 * nodes can be merged". Which means that the average fill of
582 * all nodes is still half or better.
586 static void *btree_remove_level(struct btree_head *head, struct btree_geo *geo,
587 unsigned long *key, int level)
589 unsigned long *node;
590 int i, pos, fill;
591 void *ret;
593 if (level > head->height) {
594 /* we recursed all the way up */
595 head->height = 0;
596 head->node = NULL;
597 return NULL;
600 node = find_level(head, geo, key, level);
601 pos = getpos(geo, node, key);
602 fill = getfill(geo, node, pos);
603 if ((level == 1) && (keycmp(geo, node, pos, key) != 0))
604 return NULL;
605 ret = bval(geo, node, pos);
607 /* remove and shift */
608 for (i = pos; i < fill - 1; i++) {
609 setkey(geo, node, i, bkey(geo, node, i + 1));
610 setval(geo, node, i, bval(geo, node, i + 1));
612 clearpair(geo, node, fill - 1);
614 if (fill - 1 < geo->no_pairs / 2) {
615 if (level < head->height)
616 rebalance(head, geo, key, level, node, fill - 1);
617 else if (fill - 1 == 1)
618 btree_shrink(head, geo);
621 return ret;
624 void *btree_remove(struct btree_head *head, struct btree_geo *geo,
625 unsigned long *key)
627 if (head->height == 0)
628 return NULL;
630 return btree_remove_level(head, geo, key, 1);
632 EXPORT_SYMBOL_GPL(btree_remove);
634 int btree_merge(struct btree_head *target, struct btree_head *victim,
635 struct btree_geo *geo, gfp_t gfp)
637 unsigned long key[geo->keylen];
638 unsigned long dup[geo->keylen];
639 void *val;
640 int err;
642 BUG_ON(target == victim);
644 if (!(target->node)) {
645 /* target is empty, just copy fields over */
646 target->node = victim->node;
647 target->height = victim->height;
648 __btree_init(victim);
649 return 0;
652 /* TODO: This needs some optimizations. Currently we do three tree
653 * walks to remove a single object from the victim.
655 for (;;) {
656 if (!btree_last(victim, geo, key))
657 break;
658 val = btree_lookup(victim, geo, key);
659 err = btree_insert(target, geo, key, val, gfp);
660 if (err)
661 return err;
662 /* We must make a copy of the key, as the original will get
663 * mangled inside btree_remove. */
664 longcpy(dup, key, geo->keylen);
665 btree_remove(victim, geo, dup);
667 return 0;
669 EXPORT_SYMBOL_GPL(btree_merge);
671 static size_t __btree_for_each(struct btree_head *head, struct btree_geo *geo,
672 unsigned long *node, unsigned long opaque,
673 void (*func)(void *elem, unsigned long opaque,
674 unsigned long *key, size_t index,
675 void *func2),
676 void *func2, int reap, int height, size_t count)
678 int i;
679 unsigned long *child;
681 for (i = 0; i < geo->no_pairs; i++) {
682 child = bval(geo, node, i);
683 if (!child)
684 break;
685 if (height > 1)
686 count = __btree_for_each(head, geo, child, opaque,
687 func, func2, reap, height - 1, count);
688 else
689 func(child, opaque, bkey(geo, node, i), count++,
690 func2);
692 if (reap)
693 mempool_free(node, head->mempool);
694 return count;
697 static void empty(void *elem, unsigned long opaque, unsigned long *key,
698 size_t index, void *func2)
702 void visitorl(void *elem, unsigned long opaque, unsigned long *key,
703 size_t index, void *__func)
705 visitorl_t func = __func;
707 func(elem, opaque, *key, index);
709 EXPORT_SYMBOL_GPL(visitorl);
711 void visitor32(void *elem, unsigned long opaque, unsigned long *__key,
712 size_t index, void *__func)
714 visitor32_t func = __func;
715 u32 *key = (void *)__key;
717 func(elem, opaque, *key, index);
719 EXPORT_SYMBOL_GPL(visitor32);
721 void visitor64(void *elem, unsigned long opaque, unsigned long *__key,
722 size_t index, void *__func)
724 visitor64_t func = __func;
725 u64 *key = (void *)__key;
727 func(elem, opaque, *key, index);
729 EXPORT_SYMBOL_GPL(visitor64);
731 void visitor128(void *elem, unsigned long opaque, unsigned long *__key,
732 size_t index, void *__func)
734 visitor128_t func = __func;
735 u64 *key = (void *)__key;
737 func(elem, opaque, key[0], key[1], index);
739 EXPORT_SYMBOL_GPL(visitor128);
741 size_t btree_visitor(struct btree_head *head, struct btree_geo *geo,
742 unsigned long opaque,
743 void (*func)(void *elem, unsigned long opaque,
744 unsigned long *key,
745 size_t index, void *func2),
746 void *func2)
748 size_t count = 0;
750 if (!func2)
751 func = empty;
752 if (head->node)
753 count = __btree_for_each(head, geo, head->node, opaque, func,
754 func2, 0, head->height, 0);
755 return count;
757 EXPORT_SYMBOL_GPL(btree_visitor);
759 size_t btree_grim_visitor(struct btree_head *head, struct btree_geo *geo,
760 unsigned long opaque,
761 void (*func)(void *elem, unsigned long opaque,
762 unsigned long *key,
763 size_t index, void *func2),
764 void *func2)
766 size_t count = 0;
768 if (!func2)
769 func = empty;
770 if (head->node)
771 count = __btree_for_each(head, geo, head->node, opaque, func,
772 func2, 1, head->height, 0);
773 __btree_init(head);
774 return count;
776 EXPORT_SYMBOL_GPL(btree_grim_visitor);
778 static int __init btree_module_init(void)
780 btree_cachep = kmem_cache_create("btree_node", NODESIZE, 0,
781 SLAB_HWCACHE_ALIGN, NULL);
782 return 0;
785 static void __exit btree_module_exit(void)
787 kmem_cache_destroy(btree_cachep);
790 /* If core code starts using btree, initialization should happen even earlier */
791 module_init(btree_module_init);
792 module_exit(btree_module_exit);
794 MODULE_AUTHOR("Joern Engel <joern@logfs.org>");
795 MODULE_AUTHOR("Johannes Berg <johannes@sipsolutions.net>");
796 MODULE_LICENSE("GPL");