2 * Copyright (C) 2001 Momchil Velikov
3 * Portions Copyright (C) 2001 Christoph Hellwig
4 * Copyright (C) 2005 SGI, Christoph Lameter
5 * Copyright (C) 2006 Nick Piggin
7 * This program is free software; you can redistribute it and/or
8 * modify it under the terms of the GNU General Public License as
9 * published by the Free Software Foundation; either version 2, or (at
10 * your option) any later version.
12 * This program is distributed in the hope that it will be useful, but
13 * WITHOUT ANY WARRANTY; without even the implied warranty of
14 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
15 * General Public License for more details.
17 * You should have received a copy of the GNU General Public License
18 * along with this program; if not, write to the Free Software
19 * Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA.
22 #include <linux/errno.h>
23 #include <linux/init.h>
24 #include <linux/kernel.h>
25 #include <linux/module.h>
26 #include <linux/radix-tree.h>
27 #include <linux/percpu.h>
28 #include <linux/slab.h>
29 #include <linux/notifier.h>
30 #include <linux/cpu.h>
31 #include <linux/gfp.h>
32 #include <linux/string.h>
33 #include <linux/bitops.h>
34 #include <linux/rcupdate.h>
38 #define RADIX_TREE_MAP_SHIFT (CONFIG_BASE_SMALL ? 4 : 6)
40 #define RADIX_TREE_MAP_SHIFT 3 /* For more stressful testing */
43 #define RADIX_TREE_MAP_SIZE (1UL << RADIX_TREE_MAP_SHIFT)
44 #define RADIX_TREE_MAP_MASK (RADIX_TREE_MAP_SIZE-1)
46 #define RADIX_TREE_TAG_LONGS \
47 ((RADIX_TREE_MAP_SIZE + BITS_PER_LONG - 1) / BITS_PER_LONG)
49 struct radix_tree_node {
50 unsigned int height; /* Height from the bottom */
52 struct rcu_head rcu_head;
53 void *slots[RADIX_TREE_MAP_SIZE];
54 unsigned long tags[RADIX_TREE_MAX_TAGS][RADIX_TREE_TAG_LONGS];
57 struct radix_tree_path {
58 struct radix_tree_node *node;
62 #define RADIX_TREE_INDEX_BITS (8 /* CHAR_BIT */ * sizeof(unsigned long))
63 #define RADIX_TREE_MAX_PATH (DIV_ROUND_UP(RADIX_TREE_INDEX_BITS, \
64 RADIX_TREE_MAP_SHIFT))
67 * The height_to_maxindex array needs to be one deeper than the maximum
68 * path as height 0 holds only 1 entry.
70 static unsigned long height_to_maxindex[RADIX_TREE_MAX_PATH + 1] __read_mostly;
73 * Radix tree node cache.
75 static struct kmem_cache *radix_tree_node_cachep;
78 * Per-cpu pool of preloaded nodes
80 struct radix_tree_preload {
82 struct radix_tree_node *nodes[RADIX_TREE_MAX_PATH];
84 static DEFINE_PER_CPU(struct radix_tree_preload, radix_tree_preloads) = { 0, };
86 static inline gfp_t root_gfp_mask(struct radix_tree_root *root)
88 return root->gfp_mask & __GFP_BITS_MASK;
91 static inline void tag_set(struct radix_tree_node *node, unsigned int tag,
94 __set_bit(offset, node->tags[tag]);
97 static inline void tag_clear(struct radix_tree_node *node, unsigned int tag,
100 __clear_bit(offset, node->tags[tag]);
103 static inline int tag_get(struct radix_tree_node *node, unsigned int tag,
106 return test_bit(offset, node->tags[tag]);
109 static inline void root_tag_set(struct radix_tree_root *root, unsigned int tag)
111 root->gfp_mask |= (__force gfp_t)(1 << (tag + __GFP_BITS_SHIFT));
114 static inline void root_tag_clear(struct radix_tree_root *root, unsigned int tag)
116 root->gfp_mask &= (__force gfp_t)~(1 << (tag + __GFP_BITS_SHIFT));
119 static inline void root_tag_clear_all(struct radix_tree_root *root)
121 root->gfp_mask &= __GFP_BITS_MASK;
124 static inline int root_tag_get(struct radix_tree_root *root, unsigned int tag)
126 return (__force unsigned)root->gfp_mask & (1 << (tag + __GFP_BITS_SHIFT));
130 * Returns 1 if any slot in the node has this tag set.
131 * Otherwise returns 0.
133 static inline int any_tag_set(struct radix_tree_node *node, unsigned int tag)
136 for (idx = 0; idx < RADIX_TREE_TAG_LONGS; idx++) {
137 if (node->tags[tag][idx])
143 * This assumes that the caller has performed appropriate preallocation, and
144 * that the caller has pinned this thread of control to the current CPU.
146 static struct radix_tree_node *
147 radix_tree_node_alloc(struct radix_tree_root *root)
149 struct radix_tree_node *ret = NULL;
150 gfp_t gfp_mask = root_gfp_mask(root);
152 if (!(gfp_mask & __GFP_WAIT)) {
153 struct radix_tree_preload *rtp;
156 * Provided the caller has preloaded here, we will always
157 * succeed in getting a node here (and never reach
160 rtp = &__get_cpu_var(radix_tree_preloads);
162 ret = rtp->nodes[rtp->nr - 1];
163 rtp->nodes[rtp->nr - 1] = NULL;
168 ret = kmem_cache_alloc(radix_tree_node_cachep, gfp_mask);
170 BUG_ON(radix_tree_is_indirect_ptr(ret));
174 static void radix_tree_node_rcu_free(struct rcu_head *head)
176 struct radix_tree_node *node =
177 container_of(head, struct radix_tree_node, rcu_head);
180 * must only free zeroed nodes into the slab. radix_tree_shrink
181 * can leave us with a non-NULL entry in the first slot, so clear
182 * that here to make sure.
184 tag_clear(node, 0, 0);
185 tag_clear(node, 1, 0);
186 node->slots[0] = NULL;
189 kmem_cache_free(radix_tree_node_cachep, node);
193 radix_tree_node_free(struct radix_tree_node *node)
195 call_rcu(&node->rcu_head, radix_tree_node_rcu_free);
199 * Load up this CPU's radix_tree_node buffer with sufficient objects to
200 * ensure that the addition of a single element in the tree cannot fail. On
201 * success, return zero, with preemption disabled. On error, return -ENOMEM
202 * with preemption not disabled.
204 int radix_tree_preload(gfp_t gfp_mask)
206 struct radix_tree_preload *rtp;
207 struct radix_tree_node *node;
211 rtp = &__get_cpu_var(radix_tree_preloads);
212 while (rtp->nr < ARRAY_SIZE(rtp->nodes)) {
214 node = kmem_cache_alloc(radix_tree_node_cachep, gfp_mask);
218 rtp = &__get_cpu_var(radix_tree_preloads);
219 if (rtp->nr < ARRAY_SIZE(rtp->nodes))
220 rtp->nodes[rtp->nr++] = node;
222 kmem_cache_free(radix_tree_node_cachep, node);
228 EXPORT_SYMBOL(radix_tree_preload);
231 * Return the maximum key which can be store into a
232 * radix tree with height HEIGHT.
234 static inline unsigned long radix_tree_maxindex(unsigned int height)
236 return height_to_maxindex[height];
240 * Extend a radix tree so it can store key @index.
242 static int radix_tree_extend(struct radix_tree_root *root, unsigned long index)
244 struct radix_tree_node *node;
248 /* Figure out what the height should be. */
249 height = root->height + 1;
250 while (index > radix_tree_maxindex(height))
253 if (root->rnode == NULL) {
254 root->height = height;
259 unsigned int newheight;
260 if (!(node = radix_tree_node_alloc(root)))
263 /* Increase the height. */
264 node->slots[0] = radix_tree_indirect_to_ptr(root->rnode);
266 /* Propagate the aggregated tag info into the new root */
267 for (tag = 0; tag < RADIX_TREE_MAX_TAGS; tag++) {
268 if (root_tag_get(root, tag))
269 tag_set(node, tag, 0);
272 newheight = root->height+1;
273 node->height = newheight;
275 node = radix_tree_ptr_to_indirect(node);
276 rcu_assign_pointer(root->rnode, node);
277 root->height = newheight;
278 } while (height > root->height);
284 * radix_tree_insert - insert into a radix tree
285 * @root: radix tree root
287 * @item: item to insert
289 * Insert an item into the radix tree at position @index.
291 int radix_tree_insert(struct radix_tree_root *root,
292 unsigned long index, void *item)
294 struct radix_tree_node *node = NULL, *slot;
295 unsigned int height, shift;
299 BUG_ON(radix_tree_is_indirect_ptr(item));
301 /* Make sure the tree is high enough. */
302 if (index > radix_tree_maxindex(root->height)) {
303 error = radix_tree_extend(root, index);
308 slot = radix_tree_indirect_to_ptr(root->rnode);
310 height = root->height;
311 shift = (height-1) * RADIX_TREE_MAP_SHIFT;
313 offset = 0; /* uninitialised var warning */
316 /* Have to add a child node. */
317 if (!(slot = radix_tree_node_alloc(root)))
319 slot->height = height;
321 rcu_assign_pointer(node->slots[offset], slot);
324 rcu_assign_pointer(root->rnode,
325 radix_tree_ptr_to_indirect(slot));
328 /* Go a level down */
329 offset = (index >> shift) & RADIX_TREE_MAP_MASK;
331 slot = node->slots[offset];
332 shift -= RADIX_TREE_MAP_SHIFT;
341 rcu_assign_pointer(node->slots[offset], item);
342 BUG_ON(tag_get(node, 0, offset));
343 BUG_ON(tag_get(node, 1, offset));
345 rcu_assign_pointer(root->rnode, item);
346 BUG_ON(root_tag_get(root, 0));
347 BUG_ON(root_tag_get(root, 1));
352 EXPORT_SYMBOL(radix_tree_insert);
355 * is_slot == 1 : search for the slot.
356 * is_slot == 0 : search for the node.
358 static void *radix_tree_lookup_element(struct radix_tree_root *root,
359 unsigned long index, int is_slot)
361 unsigned int height, shift;
362 struct radix_tree_node *node, **slot;
364 node = rcu_dereference(root->rnode);
368 if (!radix_tree_is_indirect_ptr(node)) {
371 return is_slot ? (void *)&root->rnode : node;
373 node = radix_tree_indirect_to_ptr(node);
375 height = node->height;
376 if (index > radix_tree_maxindex(height))
379 shift = (height-1) * RADIX_TREE_MAP_SHIFT;
382 slot = (struct radix_tree_node **)
383 (node->slots + ((index>>shift) & RADIX_TREE_MAP_MASK));
384 node = rcu_dereference(*slot);
388 shift -= RADIX_TREE_MAP_SHIFT;
390 } while (height > 0);
392 return is_slot ? (void *)slot:node;
396 * radix_tree_lookup_slot - lookup a slot in a radix tree
397 * @root: radix tree root
400 * Returns: the slot corresponding to the position @index in the
401 * radix tree @root. This is useful for update-if-exists operations.
403 * This function can be called under rcu_read_lock iff the slot is not
404 * modified by radix_tree_replace_slot, otherwise it must be called
405 * exclusive from other writers. Any dereference of the slot must be done
406 * using radix_tree_deref_slot.
408 void **radix_tree_lookup_slot(struct radix_tree_root *root, unsigned long index)
410 return (void **)radix_tree_lookup_element(root, index, 1);
412 EXPORT_SYMBOL(radix_tree_lookup_slot);
415 * radix_tree_lookup - perform lookup operation on a radix tree
416 * @root: radix tree root
419 * Lookup the item at the position @index in the radix tree @root.
421 * This function can be called under rcu_read_lock, however the caller
422 * must manage lifetimes of leaf nodes (eg. RCU may also be used to free
423 * them safely). No RCU barriers are required to access or modify the
424 * returned item, however.
426 void *radix_tree_lookup(struct radix_tree_root *root, unsigned long index)
428 return radix_tree_lookup_element(root, index, 0);
430 EXPORT_SYMBOL(radix_tree_lookup);
433 * radix_tree_tag_set - set a tag on a radix tree node
434 * @root: radix tree root
438 * Set the search tag (which must be < RADIX_TREE_MAX_TAGS)
439 * corresponding to @index in the radix tree. From
440 * the root all the way down to the leaf node.
442 * Returns the address of the tagged item. Setting a tag on a not-present
445 void *radix_tree_tag_set(struct radix_tree_root *root,
446 unsigned long index, unsigned int tag)
448 unsigned int height, shift;
449 struct radix_tree_node *slot;
451 height = root->height;
452 BUG_ON(index > radix_tree_maxindex(height));
454 slot = radix_tree_indirect_to_ptr(root->rnode);
455 shift = (height - 1) * RADIX_TREE_MAP_SHIFT;
460 offset = (index >> shift) & RADIX_TREE_MAP_MASK;
461 if (!tag_get(slot, tag, offset))
462 tag_set(slot, tag, offset);
463 slot = slot->slots[offset];
464 BUG_ON(slot == NULL);
465 shift -= RADIX_TREE_MAP_SHIFT;
469 /* set the root's tag bit */
470 if (slot && !root_tag_get(root, tag))
471 root_tag_set(root, tag);
475 EXPORT_SYMBOL(radix_tree_tag_set);
478 * radix_tree_tag_clear - clear a tag on a radix tree node
479 * @root: radix tree root
483 * Clear the search tag (which must be < RADIX_TREE_MAX_TAGS)
484 * corresponding to @index in the radix tree. If
485 * this causes the leaf node to have no tags set then clear the tag in the
486 * next-to-leaf node, etc.
488 * Returns the address of the tagged item on success, else NULL. ie:
489 * has the same return value and semantics as radix_tree_lookup().
491 void *radix_tree_tag_clear(struct radix_tree_root *root,
492 unsigned long index, unsigned int tag)
495 * The radix tree path needs to be one longer than the maximum path
496 * since the "list" is null terminated.
498 struct radix_tree_path path[RADIX_TREE_MAX_PATH + 1], *pathp = path;
499 struct radix_tree_node *slot = NULL;
500 unsigned int height, shift;
502 height = root->height;
503 if (index > radix_tree_maxindex(height))
506 shift = (height - 1) * RADIX_TREE_MAP_SHIFT;
508 slot = radix_tree_indirect_to_ptr(root->rnode);
516 offset = (index >> shift) & RADIX_TREE_MAP_MASK;
517 pathp[1].offset = offset;
518 pathp[1].node = slot;
519 slot = slot->slots[offset];
521 shift -= RADIX_TREE_MAP_SHIFT;
528 while (pathp->node) {
529 if (!tag_get(pathp->node, tag, pathp->offset))
531 tag_clear(pathp->node, tag, pathp->offset);
532 if (any_tag_set(pathp->node, tag))
537 /* clear the root's tag bit */
538 if (root_tag_get(root, tag))
539 root_tag_clear(root, tag);
544 EXPORT_SYMBOL(radix_tree_tag_clear);
546 #ifndef __KERNEL__ /* Only the test harness uses this at present */
548 * radix_tree_tag_get - get a tag on a radix tree node
549 * @root: radix tree root
551 * @tag: tag index (< RADIX_TREE_MAX_TAGS)
555 * 0: tag not present or not set
558 int radix_tree_tag_get(struct radix_tree_root *root,
559 unsigned long index, unsigned int tag)
561 unsigned int height, shift;
562 struct radix_tree_node *node;
563 int saw_unset_tag = 0;
565 /* check the root's tag bit */
566 if (!root_tag_get(root, tag))
569 node = rcu_dereference(root->rnode);
573 if (!radix_tree_is_indirect_ptr(node))
575 node = radix_tree_indirect_to_ptr(node);
577 height = node->height;
578 if (index > radix_tree_maxindex(height))
581 shift = (height - 1) * RADIX_TREE_MAP_SHIFT;
589 offset = (index >> shift) & RADIX_TREE_MAP_MASK;
592 * This is just a debug check. Later, we can bale as soon as
593 * we see an unset tag.
595 if (!tag_get(node, tag, offset))
598 int ret = tag_get(node, tag, offset);
600 BUG_ON(ret && saw_unset_tag);
603 node = rcu_dereference(node->slots[offset]);
604 shift -= RADIX_TREE_MAP_SHIFT;
608 EXPORT_SYMBOL(radix_tree_tag_get);
612 * radix_tree_next_hole - find the next hole (not-present entry)
615 * @max_scan: maximum range to search
617 * Search the set [index, min(index+max_scan-1, MAX_INDEX)] for the lowest
620 * Returns: the index of the hole if found, otherwise returns an index
621 * outside of the set specified (in which case 'return - index >= max_scan'
622 * will be true). In rare cases of index wrap-around, 0 will be returned.
624 * radix_tree_next_hole may be called under rcu_read_lock. However, like
625 * radix_tree_gang_lookup, this will not atomically search a snapshot of
626 * the tree at a single point in time. For example, if a hole is created
627 * at index 5, then subsequently a hole is created at index 10,
628 * radix_tree_next_hole covering both indexes may return 10 if called
629 * under rcu_read_lock.
631 unsigned long radix_tree_next_hole(struct radix_tree_root *root,
632 unsigned long index, unsigned long max_scan)
636 for (i = 0; i < max_scan; i++) {
637 if (!radix_tree_lookup(root, index))
646 EXPORT_SYMBOL(radix_tree_next_hole);
649 * radix_tree_prev_hole - find the prev hole (not-present entry)
652 * @max_scan: maximum range to search
654 * Search backwards in the range [max(index-max_scan+1, 0), index]
655 * for the first hole.
657 * Returns: the index of the hole if found, otherwise returns an index
658 * outside of the set specified (in which case 'index - return >= max_scan'
659 * will be true). In rare cases of wrap-around, LONG_MAX will be returned.
661 * radix_tree_next_hole may be called under rcu_read_lock. However, like
662 * radix_tree_gang_lookup, this will not atomically search a snapshot of
663 * the tree at a single point in time. For example, if a hole is created
664 * at index 10, then subsequently a hole is created at index 5,
665 * radix_tree_prev_hole covering both indexes may return 5 if called under
668 unsigned long radix_tree_prev_hole(struct radix_tree_root *root,
669 unsigned long index, unsigned long max_scan)
673 for (i = 0; i < max_scan; i++) {
674 if (!radix_tree_lookup(root, index))
677 if (index == LONG_MAX)
683 EXPORT_SYMBOL(radix_tree_prev_hole);
686 __lookup(struct radix_tree_node *slot, void ***results, unsigned long index,
687 unsigned int max_items, unsigned long *next_index)
689 unsigned int nr_found = 0;
690 unsigned int shift, height;
693 height = slot->height;
696 shift = (height-1) * RADIX_TREE_MAP_SHIFT;
698 for ( ; height > 1; height--) {
699 i = (index >> shift) & RADIX_TREE_MAP_MASK;
701 if (slot->slots[i] != NULL)
703 index &= ~((1UL << shift) - 1);
704 index += 1UL << shift;
706 goto out; /* 32-bit wraparound */
708 if (i == RADIX_TREE_MAP_SIZE)
712 shift -= RADIX_TREE_MAP_SHIFT;
713 slot = rcu_dereference(slot->slots[i]);
718 /* Bottom level: grab some items */
719 for (i = index & RADIX_TREE_MAP_MASK; i < RADIX_TREE_MAP_SIZE; i++) {
721 if (slot->slots[i]) {
722 results[nr_found++] = &(slot->slots[i]);
723 if (nr_found == max_items)
733 * radix_tree_gang_lookup - perform multiple lookup on a radix tree
734 * @root: radix tree root
735 * @results: where the results of the lookup are placed
736 * @first_index: start the lookup from this key
737 * @max_items: place up to this many items at *results
739 * Performs an index-ascending scan of the tree for present items. Places
740 * them at *@results and returns the number of items which were placed at
743 * The implementation is naive.
745 * Like radix_tree_lookup, radix_tree_gang_lookup may be called under
746 * rcu_read_lock. In this case, rather than the returned results being
747 * an atomic snapshot of the tree at a single point in time, the semantics
748 * of an RCU protected gang lookup are as though multiple radix_tree_lookups
749 * have been issued in individual locks, and results stored in 'results'.
752 radix_tree_gang_lookup(struct radix_tree_root *root, void **results,
753 unsigned long first_index, unsigned int max_items)
755 unsigned long max_index;
756 struct radix_tree_node *node;
757 unsigned long cur_index = first_index;
760 node = rcu_dereference(root->rnode);
764 if (!radix_tree_is_indirect_ptr(node)) {
770 node = radix_tree_indirect_to_ptr(node);
772 max_index = radix_tree_maxindex(node->height);
775 while (ret < max_items) {
776 unsigned int nr_found, slots_found, i;
777 unsigned long next_index; /* Index of next search */
779 if (cur_index > max_index)
781 slots_found = __lookup(node, (void ***)results + ret, cur_index,
782 max_items - ret, &next_index);
784 for (i = 0; i < slots_found; i++) {
785 struct radix_tree_node *slot;
786 slot = *(((void ***)results)[ret + i]);
789 results[ret + nr_found] = rcu_dereference(slot);
795 cur_index = next_index;
800 EXPORT_SYMBOL(radix_tree_gang_lookup);
803 * radix_tree_gang_lookup_slot - perform multiple slot lookup on radix tree
804 * @root: radix tree root
805 * @results: where the results of the lookup are placed
806 * @first_index: start the lookup from this key
807 * @max_items: place up to this many items at *results
809 * Performs an index-ascending scan of the tree for present items. Places
810 * their slots at *@results and returns the number of items which were
811 * placed at *@results.
813 * The implementation is naive.
815 * Like radix_tree_gang_lookup as far as RCU and locking goes. Slots must
816 * be dereferenced with radix_tree_deref_slot, and if using only RCU
817 * protection, radix_tree_deref_slot may fail requiring a retry.
820 radix_tree_gang_lookup_slot(struct radix_tree_root *root, void ***results,
821 unsigned long first_index, unsigned int max_items)
823 unsigned long max_index;
824 struct radix_tree_node *node;
825 unsigned long cur_index = first_index;
828 node = rcu_dereference(root->rnode);
832 if (!radix_tree_is_indirect_ptr(node)) {
835 results[0] = (void **)&root->rnode;
838 node = radix_tree_indirect_to_ptr(node);
840 max_index = radix_tree_maxindex(node->height);
843 while (ret < max_items) {
844 unsigned int slots_found;
845 unsigned long next_index; /* Index of next search */
847 if (cur_index > max_index)
849 slots_found = __lookup(node, results + ret, cur_index,
850 max_items - ret, &next_index);
854 cur_index = next_index;
859 EXPORT_SYMBOL(radix_tree_gang_lookup_slot);
862 * FIXME: the two tag_get()s here should use find_next_bit() instead of
863 * open-coding the search.
866 __lookup_tag(struct radix_tree_node *slot, void ***results, unsigned long index,
867 unsigned int max_items, unsigned long *next_index, unsigned int tag)
869 unsigned int nr_found = 0;
870 unsigned int shift, height;
872 height = slot->height;
875 shift = (height-1) * RADIX_TREE_MAP_SHIFT;
878 unsigned long i = (index >> shift) & RADIX_TREE_MAP_MASK ;
881 if (tag_get(slot, tag, i))
883 index &= ~((1UL << shift) - 1);
884 index += 1UL << shift;
886 goto out; /* 32-bit wraparound */
888 if (i == RADIX_TREE_MAP_SIZE)
892 if (height == 0) { /* Bottom level: grab some items */
893 unsigned long j = index & RADIX_TREE_MAP_MASK;
895 for ( ; j < RADIX_TREE_MAP_SIZE; j++) {
897 if (!tag_get(slot, tag, j))
900 * Even though the tag was found set, we need to
901 * recheck that we have a non-NULL node, because
902 * if this lookup is lockless, it may have been
903 * subsequently deleted.
905 * Similar care must be taken in any place that
906 * lookup ->slots[x] without a lock (ie. can't
907 * rely on its value remaining the same).
909 if (slot->slots[j]) {
910 results[nr_found++] = &(slot->slots[j]);
911 if (nr_found == max_items)
916 shift -= RADIX_TREE_MAP_SHIFT;
917 slot = rcu_dereference(slot->slots[i]);
927 * radix_tree_gang_lookup_tag - perform multiple lookup on a radix tree
929 * @root: radix tree root
930 * @results: where the results of the lookup are placed
931 * @first_index: start the lookup from this key
932 * @max_items: place up to this many items at *results
933 * @tag: the tag index (< RADIX_TREE_MAX_TAGS)
935 * Performs an index-ascending scan of the tree for present items which
936 * have the tag indexed by @tag set. Places the items at *@results and
937 * returns the number of items which were placed at *@results.
940 radix_tree_gang_lookup_tag(struct radix_tree_root *root, void **results,
941 unsigned long first_index, unsigned int max_items,
944 struct radix_tree_node *node;
945 unsigned long max_index;
946 unsigned long cur_index = first_index;
949 /* check the root's tag bit */
950 if (!root_tag_get(root, tag))
953 node = rcu_dereference(root->rnode);
957 if (!radix_tree_is_indirect_ptr(node)) {
963 node = radix_tree_indirect_to_ptr(node);
965 max_index = radix_tree_maxindex(node->height);
968 while (ret < max_items) {
969 unsigned int nr_found, slots_found, i;
970 unsigned long next_index; /* Index of next search */
972 if (cur_index > max_index)
974 slots_found = __lookup_tag(node, (void ***)results + ret,
975 cur_index, max_items - ret, &next_index, tag);
977 for (i = 0; i < slots_found; i++) {
978 struct radix_tree_node *slot;
979 slot = *(((void ***)results)[ret + i]);
982 results[ret + nr_found] = rcu_dereference(slot);
988 cur_index = next_index;
993 EXPORT_SYMBOL(radix_tree_gang_lookup_tag);
996 * radix_tree_gang_lookup_tag_slot - perform multiple slot lookup on a
997 * radix tree based on a tag
998 * @root: radix tree root
999 * @results: where the results of the lookup are placed
1000 * @first_index: start the lookup from this key
1001 * @max_items: place up to this many items at *results
1002 * @tag: the tag index (< RADIX_TREE_MAX_TAGS)
1004 * Performs an index-ascending scan of the tree for present items which
1005 * have the tag indexed by @tag set. Places the slots at *@results and
1006 * returns the number of slots which were placed at *@results.
1009 radix_tree_gang_lookup_tag_slot(struct radix_tree_root *root, void ***results,
1010 unsigned long first_index, unsigned int max_items,
1013 struct radix_tree_node *node;
1014 unsigned long max_index;
1015 unsigned long cur_index = first_index;
1018 /* check the root's tag bit */
1019 if (!root_tag_get(root, tag))
1022 node = rcu_dereference(root->rnode);
1026 if (!radix_tree_is_indirect_ptr(node)) {
1027 if (first_index > 0)
1029 results[0] = (void **)&root->rnode;
1032 node = radix_tree_indirect_to_ptr(node);
1034 max_index = radix_tree_maxindex(node->height);
1037 while (ret < max_items) {
1038 unsigned int slots_found;
1039 unsigned long next_index; /* Index of next search */
1041 if (cur_index > max_index)
1043 slots_found = __lookup_tag(node, results + ret,
1044 cur_index, max_items - ret, &next_index, tag);
1046 if (next_index == 0)
1048 cur_index = next_index;
1053 EXPORT_SYMBOL(radix_tree_gang_lookup_tag_slot);
1057 * radix_tree_shrink - shrink height of a radix tree to minimal
1058 * @root radix tree root
1060 static inline void radix_tree_shrink(struct radix_tree_root *root)
1062 /* try to shrink tree height */
1063 while (root->height > 0) {
1064 struct radix_tree_node *to_free = root->rnode;
1067 BUG_ON(!radix_tree_is_indirect_ptr(to_free));
1068 to_free = radix_tree_indirect_to_ptr(to_free);
1071 * The candidate node has more than one child, or its child
1072 * is not at the leftmost slot, we cannot shrink.
1074 if (to_free->count != 1)
1076 if (!to_free->slots[0])
1080 * We don't need rcu_assign_pointer(), since we are simply
1081 * moving the node from one part of the tree to another. If
1082 * it was safe to dereference the old pointer to it
1083 * (to_free->slots[0]), it will be safe to dereference the new
1084 * one (root->rnode).
1086 newptr = to_free->slots[0];
1087 if (root->height > 1)
1088 newptr = radix_tree_ptr_to_indirect(newptr);
1089 root->rnode = newptr;
1091 radix_tree_node_free(to_free);
1096 * radix_tree_delete - delete an item from a radix tree
1097 * @root: radix tree root
1100 * Remove the item at @index from the radix tree rooted at @root.
1102 * Returns the address of the deleted item, or NULL if it was not present.
1104 void *radix_tree_delete(struct radix_tree_root *root, unsigned long index)
1107 * The radix tree path needs to be one longer than the maximum path
1108 * since the "list" is null terminated.
1110 struct radix_tree_path path[RADIX_TREE_MAX_PATH + 1], *pathp = path;
1111 struct radix_tree_node *slot = NULL;
1112 struct radix_tree_node *to_free;
1113 unsigned int height, shift;
1117 height = root->height;
1118 if (index > radix_tree_maxindex(height))
1123 root_tag_clear_all(root);
1127 slot = radix_tree_indirect_to_ptr(slot);
1129 shift = (height - 1) * RADIX_TREE_MAP_SHIFT;
1137 offset = (index >> shift) & RADIX_TREE_MAP_MASK;
1138 pathp->offset = offset;
1140 slot = slot->slots[offset];
1141 shift -= RADIX_TREE_MAP_SHIFT;
1143 } while (height > 0);
1149 * Clear all tags associated with the just-deleted item
1151 for (tag = 0; tag < RADIX_TREE_MAX_TAGS; tag++) {
1152 if (tag_get(pathp->node, tag, pathp->offset))
1153 radix_tree_tag_clear(root, index, tag);
1157 /* Now free the nodes we do not need anymore */
1158 while (pathp->node) {
1159 pathp->node->slots[pathp->offset] = NULL;
1160 pathp->node->count--;
1162 * Queue the node for deferred freeing after the
1163 * last reference to it disappears (set NULL, above).
1166 radix_tree_node_free(to_free);
1168 if (pathp->node->count) {
1170 radix_tree_indirect_to_ptr(root->rnode))
1171 radix_tree_shrink(root);
1175 /* Node with zero slots in use so free it */
1176 to_free = pathp->node;
1180 root_tag_clear_all(root);
1184 radix_tree_node_free(to_free);
1189 EXPORT_SYMBOL(radix_tree_delete);
1192 * radix_tree_tagged - test whether any items in the tree are tagged
1193 * @root: radix tree root
1196 int radix_tree_tagged(struct radix_tree_root *root, unsigned int tag)
1198 return root_tag_get(root, tag);
1200 EXPORT_SYMBOL(radix_tree_tagged);
1203 radix_tree_node_ctor(void *node)
1205 memset(node, 0, sizeof(struct radix_tree_node));
1208 static __init unsigned long __maxindex(unsigned int height)
1210 unsigned int width = height * RADIX_TREE_MAP_SHIFT;
1211 int shift = RADIX_TREE_INDEX_BITS - width;
1215 if (shift >= BITS_PER_LONG)
1217 return ~0UL >> shift;
1220 static __init void radix_tree_init_maxindex(void)
1224 for (i = 0; i < ARRAY_SIZE(height_to_maxindex); i++)
1225 height_to_maxindex[i] = __maxindex(i);
1228 static int radix_tree_callback(struct notifier_block *nfb,
1229 unsigned long action,
1232 int cpu = (long)hcpu;
1233 struct radix_tree_preload *rtp;
1235 /* Free per-cpu pool of perloaded nodes */
1236 if (action == CPU_DEAD || action == CPU_DEAD_FROZEN) {
1237 rtp = &per_cpu(radix_tree_preloads, cpu);
1239 kmem_cache_free(radix_tree_node_cachep,
1240 rtp->nodes[rtp->nr-1]);
1241 rtp->nodes[rtp->nr-1] = NULL;
1248 void __init radix_tree_init(void)
1250 radix_tree_node_cachep = kmem_cache_create("radix_tree_node",
1251 sizeof(struct radix_tree_node), 0,
1252 SLAB_PANIC | SLAB_RECLAIM_ACCOUNT,
1253 radix_tree_node_ctor);
1254 radix_tree_init_maxindex();
1255 hotcpu_notifier(radix_tree_callback, 0);