2 * Copyright (C) 2007 Oracle. All rights reserved.
4 * This program is free software; you can redistribute it and/or
5 * modify it under the terms of the GNU General Public
6 * License v2 as published by the Free Software Foundation.
8 * This program is distributed in the hope that it will be useful,
9 * but WITHOUT ANY WARRANTY; without even the implied warranty of
10 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
11 * General Public License for more details.
13 * You should have received a copy of the GNU General Public
14 * License along with this program; if not, write to the
15 * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
16 * Boston, MA 021110-1307, USA.
20 * Copyright (C) 2001 Momchil Velikov
21 * Portions Copyright (C) 2001 Christoph Hellwig
22 * Copyright (C) 2005 SGI, Christoph Lameter <clameter@sgi.com>
24 * This program is free software; you can redistribute it and/or
25 * modify it under the terms of the GNU General Public License as
26 * published by the Free Software Foundation; either version 2, or (at
27 * your option) any later version.
29 * This program is distributed in the hope that it will be useful, but
30 * WITHOUT ANY WARRANTY; without even the implied warranty of
31 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
32 * General Public License for more details.
34 * You should have received a copy of the GNU General Public License
35 * along with this program; if not, write to the Free Software
36 * Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA.
39 #include "kerncompat.h"
40 #include "radix-tree.h"
42 #define RADIX_TREE_MAP_SHIFT (CONFIG_BASE_SMALL ? 4 : 6)
44 #define RADIX_TREE_MAP_SHIFT 3 /* For more stressful testing */
47 #define RADIX_TREE_MAP_SIZE (1UL << RADIX_TREE_MAP_SHIFT)
48 #define RADIX_TREE_MAP_MASK (RADIX_TREE_MAP_SIZE-1)
50 #define RADIX_TREE_TAG_LONGS \
51 ((RADIX_TREE_MAP_SIZE + BITS_PER_LONG - 1) / BITS_PER_LONG)
53 struct radix_tree_node {
55 void *slots[RADIX_TREE_MAP_SIZE];
56 unsigned long tags[RADIX_TREE_MAX_TAGS][RADIX_TREE_TAG_LONGS];
59 struct radix_tree_path {
60 struct radix_tree_node *node;
64 #define RADIX_TREE_INDEX_BITS (8 /* CHAR_BIT */ * sizeof(unsigned long))
65 #define RADIX_TREE_MAX_PATH (RADIX_TREE_INDEX_BITS/RADIX_TREE_MAP_SHIFT + 2)
67 static unsigned long height_to_maxindex[RADIX_TREE_MAX_PATH] __read_mostly;
70 * Per-cpu pool of preloaded nodes
72 struct radix_tree_preload {
74 struct radix_tree_node *nodes[RADIX_TREE_MAX_PATH];
76 static struct radix_tree_preload radix_tree_preloads = { 0, };
78 static inline gfp_t root_gfp_mask(struct radix_tree_root *root)
80 return root->gfp_mask & __GFP_BITS_MASK;
83 static int internal_nodes = 0;
85 * This assumes that the caller has performed appropriate preallocation, and
86 * that the caller has pinned this thread of control to the current CPU.
88 static struct radix_tree_node *
89 radix_tree_node_alloc(struct radix_tree_root *root)
91 struct radix_tree_node *ret;
92 ret = malloc(sizeof(struct radix_tree_node));
94 memset(ret, 0, sizeof(struct radix_tree_node));
101 radix_tree_node_free(struct radix_tree_node *node)
108 * Load up this CPU's radix_tree_node buffer with sufficient objects to
109 * ensure that the addition of a single element in the tree cannot fail. On
110 * success, return zero, with preemption disabled. On error, return -ENOMEM
111 * with preemption not disabled.
113 int radix_tree_preload(gfp_t gfp_mask)
115 struct radix_tree_preload *rtp;
116 struct radix_tree_node *node;
120 rtp = &__get_cpu_var(radix_tree_preloads);
121 while (rtp->nr < ARRAY_SIZE(rtp->nodes)) {
123 node = radix_tree_node_alloc(NULL);
127 rtp = &__get_cpu_var(radix_tree_preloads);
128 if (rtp->nr < ARRAY_SIZE(rtp->nodes))
129 rtp->nodes[rtp->nr++] = node;
131 radix_tree_node_free(node);
138 static inline void tag_set(struct radix_tree_node *node, unsigned int tag,
141 __set_bit(offset, node->tags[tag]);
144 static inline void tag_clear(struct radix_tree_node *node, unsigned int tag,
147 __clear_bit(offset, node->tags[tag]);
150 static inline int tag_get(struct radix_tree_node *node, unsigned int tag,
153 return test_bit(offset, node->tags[tag]);
156 static inline void root_tag_set(struct radix_tree_root *root, unsigned int tag)
158 root->gfp_mask |= (__force gfp_t)(1 << (tag + __GFP_BITS_SHIFT));
162 static inline void root_tag_clear(struct radix_tree_root *root, unsigned int tag)
164 root->gfp_mask &= (__force gfp_t)~(1 << (tag + __GFP_BITS_SHIFT));
167 static inline void root_tag_clear_all(struct radix_tree_root *root)
169 root->gfp_mask &= __GFP_BITS_MASK;
172 static inline int root_tag_get(struct radix_tree_root *root, unsigned int tag)
174 return (__force unsigned)root->gfp_mask & (1 << (tag + __GFP_BITS_SHIFT));
178 * Returns 1 if any slot in the node has this tag set.
179 * Otherwise returns 0.
181 static inline int any_tag_set(struct radix_tree_node *node, unsigned int tag)
184 for (idx = 0; idx < RADIX_TREE_TAG_LONGS; idx++) {
185 if (node->tags[tag][idx])
192 * Return the maximum key which can be store into a
193 * radix tree with height HEIGHT.
195 static inline unsigned long radix_tree_maxindex(unsigned int height)
197 return height_to_maxindex[height];
201 * Extend a radix tree so it can store key @index.
203 static int radix_tree_extend(struct radix_tree_root *root, unsigned long index)
205 struct radix_tree_node *node;
209 /* Figure out what the height should be. */
210 height = root->height + 1;
211 while (index > radix_tree_maxindex(height))
214 if (root->rnode == NULL) {
215 root->height = height;
220 if (!(node = radix_tree_node_alloc(root)))
223 /* Increase the height. */
224 node->slots[0] = root->rnode;
226 /* Propagate the aggregated tag info into the new root */
227 for (tag = 0; tag < RADIX_TREE_MAX_TAGS; tag++) {
228 if (root_tag_get(root, tag))
229 tag_set(node, tag, 0);
235 } while (height > root->height);
241 * radix_tree_insert - insert into a radix tree
242 * @root: radix tree root
244 * @item: item to insert
246 * Insert an item into the radix tree at position @index.
248 int radix_tree_insert(struct radix_tree_root *root,
249 unsigned long index, void *item)
251 struct radix_tree_node *node = NULL, *slot;
252 unsigned int height, shift;
256 /* Make sure the tree is high enough. */
257 if (index > radix_tree_maxindex(root->height)) {
258 error = radix_tree_extend(root, index);
264 height = root->height;
265 shift = (height-1) * RADIX_TREE_MAP_SHIFT;
267 offset = 0; /* uninitialised var warning */
270 /* Have to add a child node. */
271 if (!(slot = radix_tree_node_alloc(root)))
274 node->slots[offset] = slot;
280 /* Go a level down */
281 offset = (index >> shift) & RADIX_TREE_MAP_MASK;
283 slot = node->slots[offset];
284 shift -= RADIX_TREE_MAP_SHIFT;
293 node->slots[offset] = item;
294 BUG_ON(tag_get(node, 0, offset));
295 BUG_ON(tag_get(node, 1, offset));
298 BUG_ON(root_tag_get(root, 0));
299 BUG_ON(root_tag_get(root, 1));
305 static inline void **__lookup_slot(struct radix_tree_root *root,
308 unsigned int height, shift;
309 struct radix_tree_node **slot;
311 height = root->height;
313 if (index > radix_tree_maxindex(height))
316 if (height == 0 && root->rnode)
317 return (void *)&root->rnode;
319 shift = (height-1) * RADIX_TREE_MAP_SHIFT;
326 slot = (struct radix_tree_node **)
328 ((index >> shift) & RADIX_TREE_MAP_MASK));
329 shift -= RADIX_TREE_MAP_SHIFT;
333 return (void **)slot;
337 * radix_tree_lookup_slot - lookup a slot in a radix tree
338 * @root: radix tree root
341 * Lookup the slot corresponding to the position @index in the radix tree
342 * @root. This is useful for update-if-exists operations.
344 void **radix_tree_lookup_slot(struct radix_tree_root *root, unsigned long index)
346 return __lookup_slot(root, index);
350 * radix_tree_lookup - perform lookup operation on a radix tree
351 * @root: radix tree root
354 * Lookup the item at the position @index in the radix tree @root.
356 void *radix_tree_lookup(struct radix_tree_root *root, unsigned long index)
360 slot = __lookup_slot(root, index);
361 return slot != NULL ? *slot : NULL;
365 * radix_tree_tag_set - set a tag on a radix tree node
366 * @root: radix tree root
370 * Set the search tag (which must be < RADIX_TREE_MAX_TAGS)
371 * corresponding to @index in the radix tree. From
372 * the root all the way down to the leaf node.
374 * Returns the address of the tagged item. Setting a tag on a not-present
377 void *radix_tree_tag_set(struct radix_tree_root *root,
378 unsigned long index, unsigned int tag)
380 unsigned int height, shift;
381 struct radix_tree_node *slot;
383 height = root->height;
384 BUG_ON(index > radix_tree_maxindex(height));
387 shift = (height - 1) * RADIX_TREE_MAP_SHIFT;
392 offset = (index >> shift) & RADIX_TREE_MAP_MASK;
393 if (!tag_get(slot, tag, offset))
394 tag_set(slot, tag, offset);
395 slot = slot->slots[offset];
396 BUG_ON(slot == NULL);
397 shift -= RADIX_TREE_MAP_SHIFT;
401 /* set the root's tag bit */
402 if (slot && !root_tag_get(root, tag))
403 root_tag_set(root, tag);
409 * radix_tree_tag_clear - clear a tag on a radix tree node
410 * @root: radix tree root
414 * Clear the search tag (which must be < RADIX_TREE_MAX_TAGS)
415 * corresponding to @index in the radix tree. If
416 * this causes the leaf node to have no tags set then clear the tag in the
417 * next-to-leaf node, etc.
419 * Returns the address of the tagged item on success, else NULL. ie:
420 * has the same return value and semantics as radix_tree_lookup().
422 void *radix_tree_tag_clear(struct radix_tree_root *root,
423 unsigned long index, unsigned int tag)
425 struct radix_tree_path path[RADIX_TREE_MAX_PATH], *pathp = path;
426 struct radix_tree_node *slot = NULL;
427 unsigned int height, shift;
429 height = root->height;
430 if (index > radix_tree_maxindex(height))
433 shift = (height - 1) * RADIX_TREE_MAP_SHIFT;
443 offset = (index >> shift) & RADIX_TREE_MAP_MASK;
444 pathp[1].offset = offset;
445 pathp[1].node = slot;
446 slot = slot->slots[offset];
448 shift -= RADIX_TREE_MAP_SHIFT;
455 while (pathp->node) {
456 if (!tag_get(pathp->node, tag, pathp->offset))
458 tag_clear(pathp->node, tag, pathp->offset);
459 if (any_tag_set(pathp->node, tag))
464 /* clear the root's tag bit */
465 if (root_tag_get(root, tag))
466 root_tag_clear(root, tag);
472 #ifndef __KERNEL__ /* Only the test harness uses this at present */
474 * radix_tree_tag_get - get a tag on a radix tree node
475 * @root: radix tree root
477 * @tag: tag index (< RADIX_TREE_MAX_TAGS)
481 * 0: tag not present or not set
484 int radix_tree_tag_get(struct radix_tree_root *root,
485 unsigned long index, unsigned int tag)
487 unsigned int height, shift;
488 struct radix_tree_node *slot;
489 int saw_unset_tag = 0;
491 height = root->height;
492 if (index > radix_tree_maxindex(height))
495 /* check the root's tag bit */
496 if (!root_tag_get(root, tag))
502 shift = (height - 1) * RADIX_TREE_MAP_SHIFT;
511 offset = (index >> shift) & RADIX_TREE_MAP_MASK;
514 * This is just a debug check. Later, we can bale as soon as
515 * we see an unset tag.
517 if (!tag_get(slot, tag, offset))
520 int ret = tag_get(slot, tag, offset);
522 BUG_ON(ret && saw_unset_tag);
525 slot = slot->slots[offset];
526 shift -= RADIX_TREE_MAP_SHIFT;
533 __lookup(struct radix_tree_root *root, void **results, unsigned long index,
534 unsigned int max_items, unsigned long *next_index)
536 unsigned int nr_found = 0;
537 unsigned int shift, height;
538 struct radix_tree_node *slot;
541 height = root->height;
543 if (root->rnode && index == 0)
544 results[nr_found++] = root->rnode;
548 shift = (height-1) * RADIX_TREE_MAP_SHIFT;
551 for ( ; height > 1; height--) {
553 for (i = (index >> shift) & RADIX_TREE_MAP_MASK ;
554 i < RADIX_TREE_MAP_SIZE; i++) {
555 if (slot->slots[i] != NULL)
557 index &= ~((1UL << shift) - 1);
558 index += 1UL << shift;
560 goto out; /* 32-bit wraparound */
562 if (i == RADIX_TREE_MAP_SIZE)
565 shift -= RADIX_TREE_MAP_SHIFT;
566 slot = slot->slots[i];
569 /* Bottom level: grab some items */
570 for (i = index & RADIX_TREE_MAP_MASK; i < RADIX_TREE_MAP_SIZE; i++) {
572 if (slot->slots[i]) {
573 results[nr_found++] = slot->slots[i];
574 if (nr_found == max_items)
584 * radix_tree_gang_lookup - perform multiple lookup on a radix tree
585 * @root: radix tree root
586 * @results: where the results of the lookup are placed
587 * @first_index: start the lookup from this key
588 * @max_items: place up to this many items at *results
590 * Performs an index-ascending scan of the tree for present items. Places
591 * them at *@results and returns the number of items which were placed at
594 * The implementation is naive.
597 radix_tree_gang_lookup(struct radix_tree_root *root, void **results,
598 unsigned long first_index, unsigned int max_items)
600 const unsigned long max_index = radix_tree_maxindex(root->height);
601 unsigned long cur_index = first_index;
602 unsigned int ret = 0;
604 while (ret < max_items) {
605 unsigned int nr_found;
606 unsigned long next_index; /* Index of next search */
608 if (cur_index > max_index)
610 nr_found = __lookup(root, results + ret, cur_index,
611 max_items - ret, &next_index);
615 cur_index = next_index;
621 * FIXME: the two tag_get()s here should use find_next_bit() instead of
622 * open-coding the search.
625 __lookup_tag(struct radix_tree_root *root, void **results, unsigned long index,
626 unsigned int max_items, unsigned long *next_index, unsigned int tag)
628 unsigned int nr_found = 0;
630 unsigned int height = root->height;
631 struct radix_tree_node *slot;
634 if (root->rnode && index == 0)
635 results[nr_found++] = root->rnode;
639 shift = (height - 1) * RADIX_TREE_MAP_SHIFT;
643 unsigned long i = (index >> shift) & RADIX_TREE_MAP_MASK;
645 for ( ; i < RADIX_TREE_MAP_SIZE; i++) {
646 if (tag_get(slot, tag, i)) {
647 BUG_ON(slot->slots[i] == NULL);
650 index &= ~((1UL << shift) - 1);
651 index += 1UL << shift;
653 goto out; /* 32-bit wraparound */
655 if (i == RADIX_TREE_MAP_SIZE)
658 if (height == 0) { /* Bottom level: grab some items */
659 unsigned long j = index & RADIX_TREE_MAP_MASK;
661 for ( ; j < RADIX_TREE_MAP_SIZE; j++) {
663 if (tag_get(slot, tag, j)) {
664 BUG_ON(slot->slots[j] == NULL);
665 results[nr_found++] = slot->slots[j];
666 if (nr_found == max_items)
671 shift -= RADIX_TREE_MAP_SHIFT;
672 slot = slot->slots[i];
673 } while (height > 0);
680 * radix_tree_gang_lookup_tag - perform multiple lookup on a radix tree
682 * @root: radix tree root
683 * @results: where the results of the lookup are placed
684 * @first_index: start the lookup from this key
685 * @max_items: place up to this many items at *results
686 * @tag: the tag index (< RADIX_TREE_MAX_TAGS)
688 * Performs an index-ascending scan of the tree for present items which
689 * have the tag indexed by @tag set. Places the items at *@results and
690 * returns the number of items which were placed at *@results.
693 radix_tree_gang_lookup_tag(struct radix_tree_root *root, void **results,
694 unsigned long first_index, unsigned int max_items,
697 const unsigned long max_index = radix_tree_maxindex(root->height);
698 unsigned long cur_index = first_index;
699 unsigned int ret = 0;
701 /* check the root's tag bit */
702 if (!root_tag_get(root, tag))
705 while (ret < max_items) {
706 unsigned int nr_found;
707 unsigned long next_index; /* Index of next search */
709 if (cur_index > max_index)
711 nr_found = __lookup_tag(root, results + ret, cur_index,
712 max_items - ret, &next_index, tag);
716 cur_index = next_index;
722 * radix_tree_shrink - shrink height of a radix tree to minimal
723 * @root radix tree root
725 static inline void radix_tree_shrink(struct radix_tree_root *root)
727 /* try to shrink tree height */
728 while (root->height > 0 &&
729 root->rnode->count == 1 &&
730 root->rnode->slots[0]) {
731 struct radix_tree_node *to_free = root->rnode;
733 root->rnode = to_free->slots[0];
735 /* must only free zeroed nodes into the slab */
736 tag_clear(to_free, 0, 0);
737 tag_clear(to_free, 1, 0);
738 to_free->slots[0] = NULL;
740 radix_tree_node_free(to_free);
745 * radix_tree_delete - delete an item from a radix tree
746 * @root: radix tree root
749 * Remove the item at @index from the radix tree rooted at @root.
751 * Returns the address of the deleted item, or NULL if it was not present.
753 void *radix_tree_delete(struct radix_tree_root *root, unsigned long index)
755 struct radix_tree_path path[RADIX_TREE_MAX_PATH], *pathp = path;
756 struct radix_tree_node *slot = NULL;
757 unsigned int height, shift;
761 height = root->height;
762 if (index > radix_tree_maxindex(height))
766 if (height == 0 && root->rnode) {
767 root_tag_clear_all(root);
772 shift = (height - 1) * RADIX_TREE_MAP_SHIFT;
780 offset = (index >> shift) & RADIX_TREE_MAP_MASK;
781 pathp->offset = offset;
783 slot = slot->slots[offset];
784 shift -= RADIX_TREE_MAP_SHIFT;
786 } while (height > 0);
792 * Clear all tags associated with the just-deleted item
794 for (tag = 0; tag < RADIX_TREE_MAX_TAGS; tag++) {
795 if (tag_get(pathp->node, tag, pathp->offset))
796 radix_tree_tag_clear(root, index, tag);
799 /* Now free the nodes we do not need anymore */
800 while (pathp->node) {
801 pathp->node->slots[pathp->offset] = NULL;
802 pathp->node->count--;
804 if (pathp->node->count) {
805 if (pathp->node == root->rnode)
806 radix_tree_shrink(root);
810 /* Node with zero slots in use so free it */
811 radix_tree_node_free(pathp->node);
815 root_tag_clear_all(root);
824 * radix_tree_tagged - test whether any items in the tree are tagged
825 * @root: radix tree root
828 int radix_tree_tagged(struct radix_tree_root *root, unsigned int tag)
830 return root_tag_get(root, tag);
833 static unsigned long __maxindex(unsigned int height)
835 unsigned int tmp = height * RADIX_TREE_MAP_SHIFT;
836 unsigned long index = ~0UL;
838 if (tmp < RADIX_TREE_INDEX_BITS)
839 index = (index >> (RADIX_TREE_INDEX_BITS - tmp - 1)) >> 1;
843 static void radix_tree_init_maxindex(void)
847 for (i = 0; i < ARRAY_SIZE(height_to_maxindex); i++)
848 height_to_maxindex[i] = __maxindex(i);
851 void radix_tree_init(void)
853 radix_tree_init_maxindex();