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/string.h>
32 #include <linux/bitops.h>
33 #include <linux/rcupdate.h>
37 #define RADIX_TREE_MAP_SHIFT (CONFIG_BASE_SMALL ? 4 : 6)
39 #define RADIX_TREE_MAP_SHIFT 3 /* For more stressful testing */
42 #define RADIX_TREE_MAP_SIZE (1UL << RADIX_TREE_MAP_SHIFT)
43 #define RADIX_TREE_MAP_MASK (RADIX_TREE_MAP_SIZE-1)
45 #define RADIX_TREE_TAG_LONGS \
46 ((RADIX_TREE_MAP_SIZE + BITS_PER_LONG - 1) / BITS_PER_LONG)
48 struct radix_tree_node {
49 unsigned int height; /* Height from the bottom */
51 struct rcu_head rcu_head;
52 void __rcu *slots[RADIX_TREE_MAP_SIZE];
53 unsigned long tags[RADIX_TREE_MAX_TAGS][RADIX_TREE_TAG_LONGS];
56 struct radix_tree_path {
57 struct radix_tree_node *node;
61 #define RADIX_TREE_INDEX_BITS (8 /* CHAR_BIT */ * sizeof(unsigned long))
62 #define RADIX_TREE_MAX_PATH (DIV_ROUND_UP(RADIX_TREE_INDEX_BITS, \
63 RADIX_TREE_MAP_SHIFT))
66 * The height_to_maxindex array needs to be one deeper than the maximum
67 * path as height 0 holds only 1 entry.
69 static unsigned long height_to_maxindex[RADIX_TREE_MAX_PATH + 1] __read_mostly;
72 * Radix tree node cache.
74 static struct kmem_cache *radix_tree_node_cachep;
77 * Per-cpu pool of preloaded nodes
79 struct radix_tree_preload {
81 struct radix_tree_node *nodes[RADIX_TREE_MAX_PATH];
83 static DEFINE_PER_CPU(struct radix_tree_preload, radix_tree_preloads) = { 0, };
85 static inline void *ptr_to_indirect(void *ptr)
87 return (void *)((unsigned long)ptr | RADIX_TREE_INDIRECT_PTR);
90 static inline void *indirect_to_ptr(void *ptr)
92 return (void *)((unsigned long)ptr & ~RADIX_TREE_INDIRECT_PTR);
95 static inline gfp_t root_gfp_mask(struct radix_tree_root *root)
97 return root->gfp_mask & __GFP_BITS_MASK;
100 static inline void tag_set(struct radix_tree_node *node, unsigned int tag,
103 __set_bit(offset, node->tags[tag]);
106 static inline void tag_clear(struct radix_tree_node *node, unsigned int tag,
109 __clear_bit(offset, node->tags[tag]);
112 static inline int tag_get(struct radix_tree_node *node, unsigned int tag,
115 return test_bit(offset, node->tags[tag]);
118 static inline void root_tag_set(struct radix_tree_root *root, unsigned int tag)
120 root->gfp_mask |= (__force gfp_t)(1 << (tag + __GFP_BITS_SHIFT));
123 static inline void root_tag_clear(struct radix_tree_root *root, unsigned int tag)
125 root->gfp_mask &= (__force gfp_t)~(1 << (tag + __GFP_BITS_SHIFT));
128 static inline void root_tag_clear_all(struct radix_tree_root *root)
130 root->gfp_mask &= __GFP_BITS_MASK;
133 static inline int root_tag_get(struct radix_tree_root *root, unsigned int tag)
135 return (__force unsigned)root->gfp_mask & (1 << (tag + __GFP_BITS_SHIFT));
139 * Returns 1 if any slot in the node has this tag set.
140 * Otherwise returns 0.
142 static inline int any_tag_set(struct radix_tree_node *node, unsigned int tag)
145 for (idx = 0; idx < RADIX_TREE_TAG_LONGS; idx++) {
146 if (node->tags[tag][idx])
152 * This assumes that the caller has performed appropriate preallocation, and
153 * that the caller has pinned this thread of control to the current CPU.
155 static struct radix_tree_node *
156 radix_tree_node_alloc(struct radix_tree_root *root)
158 struct radix_tree_node *ret = NULL;
159 gfp_t gfp_mask = root_gfp_mask(root);
161 if (!(gfp_mask & __GFP_WAIT)) {
162 struct radix_tree_preload *rtp;
165 * Provided the caller has preloaded here, we will always
166 * succeed in getting a node here (and never reach
169 rtp = &__get_cpu_var(radix_tree_preloads);
171 ret = rtp->nodes[rtp->nr - 1];
172 rtp->nodes[rtp->nr - 1] = NULL;
177 ret = kmem_cache_alloc(radix_tree_node_cachep, gfp_mask);
179 BUG_ON(radix_tree_is_indirect_ptr(ret));
183 static void radix_tree_node_rcu_free(struct rcu_head *head)
185 struct radix_tree_node *node =
186 container_of(head, struct radix_tree_node, rcu_head);
190 * must only free zeroed nodes into the slab. radix_tree_shrink
191 * can leave us with a non-NULL entry in the first slot, so clear
192 * that here to make sure.
194 for (i = 0; i < RADIX_TREE_MAX_TAGS; i++)
195 tag_clear(node, i, 0);
197 node->slots[0] = NULL;
200 kmem_cache_free(radix_tree_node_cachep, node);
204 radix_tree_node_free(struct radix_tree_node *node)
206 call_rcu(&node->rcu_head, radix_tree_node_rcu_free);
210 * Load up this CPU's radix_tree_node buffer with sufficient objects to
211 * ensure that the addition of a single element in the tree cannot fail. On
212 * success, return zero, with preemption disabled. On error, return -ENOMEM
213 * with preemption not disabled.
215 * To make use of this facility, the radix tree must be initialised without
216 * __GFP_WAIT being passed to INIT_RADIX_TREE().
218 int radix_tree_preload(gfp_t gfp_mask)
220 struct radix_tree_preload *rtp;
221 struct radix_tree_node *node;
225 rtp = &__get_cpu_var(radix_tree_preloads);
226 while (rtp->nr < ARRAY_SIZE(rtp->nodes)) {
228 node = kmem_cache_alloc(radix_tree_node_cachep, gfp_mask);
232 rtp = &__get_cpu_var(radix_tree_preloads);
233 if (rtp->nr < ARRAY_SIZE(rtp->nodes))
234 rtp->nodes[rtp->nr++] = node;
236 kmem_cache_free(radix_tree_node_cachep, node);
242 EXPORT_SYMBOL(radix_tree_preload);
245 * Return the maximum key which can be store into a
246 * radix tree with height HEIGHT.
248 static inline unsigned long radix_tree_maxindex(unsigned int height)
250 return height_to_maxindex[height];
254 * Extend a radix tree so it can store key @index.
256 static int radix_tree_extend(struct radix_tree_root *root, unsigned long index)
258 struct radix_tree_node *node;
262 /* Figure out what the height should be. */
263 height = root->height + 1;
264 while (index > radix_tree_maxindex(height))
267 if (root->rnode == NULL) {
268 root->height = height;
273 unsigned int newheight;
274 if (!(node = radix_tree_node_alloc(root)))
277 /* Increase the height. */
278 node->slots[0] = indirect_to_ptr(root->rnode);
280 /* Propagate the aggregated tag info into the new root */
281 for (tag = 0; tag < RADIX_TREE_MAX_TAGS; tag++) {
282 if (root_tag_get(root, tag))
283 tag_set(node, tag, 0);
286 newheight = root->height+1;
287 node->height = newheight;
289 node = ptr_to_indirect(node);
290 rcu_assign_pointer(root->rnode, node);
291 root->height = newheight;
292 } while (height > root->height);
298 * radix_tree_insert - insert into a radix tree
299 * @root: radix tree root
301 * @item: item to insert
303 * Insert an item into the radix tree at position @index.
305 int radix_tree_insert(struct radix_tree_root *root,
306 unsigned long index, void *item)
308 struct radix_tree_node *node = NULL, *slot;
309 unsigned int height, shift;
313 BUG_ON(radix_tree_is_indirect_ptr(item));
315 /* Make sure the tree is high enough. */
316 if (index > radix_tree_maxindex(root->height)) {
317 error = radix_tree_extend(root, index);
322 slot = indirect_to_ptr(root->rnode);
324 height = root->height;
325 shift = (height-1) * RADIX_TREE_MAP_SHIFT;
327 offset = 0; /* uninitialised var warning */
330 /* Have to add a child node. */
331 if (!(slot = radix_tree_node_alloc(root)))
333 slot->height = height;
335 rcu_assign_pointer(node->slots[offset], slot);
338 rcu_assign_pointer(root->rnode, ptr_to_indirect(slot));
341 /* Go a level down */
342 offset = (index >> shift) & RADIX_TREE_MAP_MASK;
344 slot = node->slots[offset];
345 shift -= RADIX_TREE_MAP_SHIFT;
354 rcu_assign_pointer(node->slots[offset], item);
355 BUG_ON(tag_get(node, 0, offset));
356 BUG_ON(tag_get(node, 1, offset));
358 rcu_assign_pointer(root->rnode, item);
359 BUG_ON(root_tag_get(root, 0));
360 BUG_ON(root_tag_get(root, 1));
365 EXPORT_SYMBOL(radix_tree_insert);
368 * is_slot == 1 : search for the slot.
369 * is_slot == 0 : search for the node.
371 static void *radix_tree_lookup_element(struct radix_tree_root *root,
372 unsigned long index, int is_slot)
374 unsigned int height, shift;
375 struct radix_tree_node *node, **slot;
377 node = rcu_dereference_raw(root->rnode);
381 if (!radix_tree_is_indirect_ptr(node)) {
384 return is_slot ? (void *)&root->rnode : node;
386 node = indirect_to_ptr(node);
388 height = node->height;
389 if (index > radix_tree_maxindex(height))
392 shift = (height-1) * RADIX_TREE_MAP_SHIFT;
395 slot = (struct radix_tree_node **)
396 (node->slots + ((index>>shift) & RADIX_TREE_MAP_MASK));
397 node = rcu_dereference_raw(*slot);
401 shift -= RADIX_TREE_MAP_SHIFT;
403 } while (height > 0);
405 return is_slot ? (void *)slot : indirect_to_ptr(node);
409 * radix_tree_lookup_slot - lookup a slot in a radix tree
410 * @root: radix tree root
413 * Returns: the slot corresponding to the position @index in the
414 * radix tree @root. This is useful for update-if-exists operations.
416 * This function can be called under rcu_read_lock iff the slot is not
417 * modified by radix_tree_replace_slot, otherwise it must be called
418 * exclusive from other writers. Any dereference of the slot must be done
419 * using radix_tree_deref_slot.
421 void **radix_tree_lookup_slot(struct radix_tree_root *root, unsigned long index)
423 return (void **)radix_tree_lookup_element(root, index, 1);
425 EXPORT_SYMBOL(radix_tree_lookup_slot);
428 * radix_tree_lookup - perform lookup operation on a radix tree
429 * @root: radix tree root
432 * Lookup the item at the position @index in the radix tree @root.
434 * This function can be called under rcu_read_lock, however the caller
435 * must manage lifetimes of leaf nodes (eg. RCU may also be used to free
436 * them safely). No RCU barriers are required to access or modify the
437 * returned item, however.
439 void *radix_tree_lookup(struct radix_tree_root *root, unsigned long index)
441 return radix_tree_lookup_element(root, index, 0);
443 EXPORT_SYMBOL(radix_tree_lookup);
446 * radix_tree_tag_set - set a tag on a radix tree node
447 * @root: radix tree root
451 * Set the search tag (which must be < RADIX_TREE_MAX_TAGS)
452 * corresponding to @index in the radix tree. From
453 * the root all the way down to the leaf node.
455 * Returns the address of the tagged item. Setting a tag on a not-present
458 void *radix_tree_tag_set(struct radix_tree_root *root,
459 unsigned long index, unsigned int tag)
461 unsigned int height, shift;
462 struct radix_tree_node *slot;
464 height = root->height;
465 BUG_ON(index > radix_tree_maxindex(height));
467 slot = indirect_to_ptr(root->rnode);
468 shift = (height - 1) * RADIX_TREE_MAP_SHIFT;
473 offset = (index >> shift) & RADIX_TREE_MAP_MASK;
474 if (!tag_get(slot, tag, offset))
475 tag_set(slot, tag, offset);
476 slot = slot->slots[offset];
477 BUG_ON(slot == NULL);
478 shift -= RADIX_TREE_MAP_SHIFT;
482 /* set the root's tag bit */
483 if (slot && !root_tag_get(root, tag))
484 root_tag_set(root, tag);
488 EXPORT_SYMBOL(radix_tree_tag_set);
491 * radix_tree_tag_clear - clear a tag on a radix tree node
492 * @root: radix tree root
496 * Clear the search tag (which must be < RADIX_TREE_MAX_TAGS)
497 * corresponding to @index in the radix tree. If
498 * this causes the leaf node to have no tags set then clear the tag in the
499 * next-to-leaf node, etc.
501 * Returns the address of the tagged item on success, else NULL. ie:
502 * has the same return value and semantics as radix_tree_lookup().
504 void *radix_tree_tag_clear(struct radix_tree_root *root,
505 unsigned long index, unsigned int tag)
508 * The radix tree path needs to be one longer than the maximum path
509 * since the "list" is null terminated.
511 struct radix_tree_path path[RADIX_TREE_MAX_PATH + 1], *pathp = path;
512 struct radix_tree_node *slot = NULL;
513 unsigned int height, shift;
515 height = root->height;
516 if (index > radix_tree_maxindex(height))
519 shift = (height - 1) * RADIX_TREE_MAP_SHIFT;
521 slot = indirect_to_ptr(root->rnode);
529 offset = (index >> shift) & RADIX_TREE_MAP_MASK;
530 pathp[1].offset = offset;
531 pathp[1].node = slot;
532 slot = slot->slots[offset];
534 shift -= RADIX_TREE_MAP_SHIFT;
541 while (pathp->node) {
542 if (!tag_get(pathp->node, tag, pathp->offset))
544 tag_clear(pathp->node, tag, pathp->offset);
545 if (any_tag_set(pathp->node, tag))
550 /* clear the root's tag bit */
551 if (root_tag_get(root, tag))
552 root_tag_clear(root, tag);
557 EXPORT_SYMBOL(radix_tree_tag_clear);
560 * radix_tree_tag_get - get a tag on a radix tree node
561 * @root: radix tree root
563 * @tag: tag index (< RADIX_TREE_MAX_TAGS)
567 * 0: tag not present or not set
570 * Note that the return value of this function may not be relied on, even if
571 * the RCU lock is held, unless tag modification and node deletion are excluded
574 int radix_tree_tag_get(struct radix_tree_root *root,
575 unsigned long index, unsigned int tag)
577 unsigned int height, shift;
578 struct radix_tree_node *node;
580 /* check the root's tag bit */
581 if (!root_tag_get(root, tag))
584 node = rcu_dereference_raw(root->rnode);
588 if (!radix_tree_is_indirect_ptr(node))
590 node = indirect_to_ptr(node);
592 height = node->height;
593 if (index > radix_tree_maxindex(height))
596 shift = (height - 1) * RADIX_TREE_MAP_SHIFT;
604 offset = (index >> shift) & RADIX_TREE_MAP_MASK;
605 if (!tag_get(node, tag, offset))
609 node = rcu_dereference_raw(node->slots[offset]);
610 shift -= RADIX_TREE_MAP_SHIFT;
614 EXPORT_SYMBOL(radix_tree_tag_get);
617 * radix_tree_range_tag_if_tagged - for each item in given range set given
618 * tag if item has another tag set
619 * @root: radix tree root
620 * @first_indexp: pointer to a starting index of a range to scan
621 * @last_index: last index of a range to scan
622 * @nr_to_tag: maximum number items to tag
623 * @iftag: tag index to test
624 * @settag: tag index to set if tested tag is set
626 * This function scans range of radix tree from first_index to last_index
627 * (inclusive). For each item in the range if iftag is set, the function sets
628 * also settag. The function stops either after tagging nr_to_tag items or
629 * after reaching last_index.
631 * The tags must be set from the leaf level only and propagated back up the
632 * path to the root. We must do this so that we resolve the full path before
633 * setting any tags on intermediate nodes. If we set tags as we descend, then
634 * we can get to the leaf node and find that the index that has the iftag
635 * set is outside the range we are scanning. This reults in dangling tags and
636 * can lead to problems with later tag operations (e.g. livelocks on lookups).
638 * The function returns number of leaves where the tag was set and sets
639 * *first_indexp to the first unscanned index.
640 * WARNING! *first_indexp can wrap if last_index is ULONG_MAX. Caller must
641 * be prepared to handle that.
643 unsigned long radix_tree_range_tag_if_tagged(struct radix_tree_root *root,
644 unsigned long *first_indexp, unsigned long last_index,
645 unsigned long nr_to_tag,
646 unsigned int iftag, unsigned int settag)
648 unsigned int height = root->height;
649 struct radix_tree_path path[height];
650 struct radix_tree_path *pathp = path;
651 struct radix_tree_node *slot;
653 unsigned long tagged = 0;
654 unsigned long index = *first_indexp;
656 last_index = min(last_index, radix_tree_maxindex(height));
657 if (index > last_index)
661 if (!root_tag_get(root, iftag)) {
662 *first_indexp = last_index + 1;
666 *first_indexp = last_index + 1;
667 root_tag_set(root, settag);
671 shift = (height - 1) * RADIX_TREE_MAP_SHIFT;
672 slot = indirect_to_ptr(root->rnode);
675 * we fill the path from (root->height - 2) to 0, leaving the index at
676 * (root->height - 1) as a terminator. Zero the node in the terminator
677 * so that we can use this to end walk loops back up the path.
679 path[height - 1].node = NULL;
684 offset = (index >> shift) & RADIX_TREE_MAP_MASK;
685 if (!slot->slots[offset])
687 if (!tag_get(slot, iftag, offset))
690 /* Go down one level */
692 shift -= RADIX_TREE_MAP_SHIFT;
693 path[height - 1].node = slot;
694 path[height - 1].offset = offset;
695 slot = slot->slots[offset];
701 tag_set(slot, settag, offset);
703 /* walk back up the path tagging interior nodes */
705 while (pathp->node) {
706 /* stop if we find a node with the tag already set */
707 if (tag_get(pathp->node, settag, pathp->offset))
709 tag_set(pathp->node, settag, pathp->offset);
714 /* Go to next item at level determined by 'shift' */
715 index = ((index >> shift) + 1) << shift;
716 /* Overflow can happen when last_index is ~0UL... */
717 if (index > last_index || !index)
719 if (tagged >= nr_to_tag)
721 while (((index >> shift) & RADIX_TREE_MAP_MASK) == 0) {
723 * We've fully scanned this node. Go up. Because
724 * last_index is guaranteed to be in the tree, what
725 * we do below cannot wander astray.
727 slot = path[height - 1].node;
729 shift += RADIX_TREE_MAP_SHIFT;
733 * We need not to tag the root tag if there is no tag which is set with
734 * settag within the range from *first_indexp to last_index.
737 root_tag_set(root, settag);
738 *first_indexp = index;
742 EXPORT_SYMBOL(radix_tree_range_tag_if_tagged);
746 * radix_tree_next_hole - find the next hole (not-present entry)
749 * @max_scan: maximum range to search
751 * Search the set [index, min(index+max_scan-1, MAX_INDEX)] for the lowest
754 * Returns: the index of the hole if found, otherwise returns an index
755 * outside of the set specified (in which case 'return - index >= max_scan'
756 * will be true). In rare cases of index wrap-around, 0 will be returned.
758 * radix_tree_next_hole may be called under rcu_read_lock. However, like
759 * radix_tree_gang_lookup, this will not atomically search a snapshot of
760 * the tree at a single point in time. For example, if a hole is created
761 * at index 5, then subsequently a hole is created at index 10,
762 * radix_tree_next_hole covering both indexes may return 10 if called
763 * under rcu_read_lock.
765 unsigned long radix_tree_next_hole(struct radix_tree_root *root,
766 unsigned long index, unsigned long max_scan)
770 for (i = 0; i < max_scan; i++) {
771 if (!radix_tree_lookup(root, index))
780 EXPORT_SYMBOL(radix_tree_next_hole);
783 * radix_tree_prev_hole - find the prev hole (not-present entry)
786 * @max_scan: maximum range to search
788 * Search backwards in the range [max(index-max_scan+1, 0), index]
789 * for the first hole.
791 * Returns: the index of the hole if found, otherwise returns an index
792 * outside of the set specified (in which case 'index - return >= max_scan'
793 * will be true). In rare cases of wrap-around, ULONG_MAX will be returned.
795 * radix_tree_next_hole may be called under rcu_read_lock. However, like
796 * radix_tree_gang_lookup, this will not atomically search a snapshot of
797 * the tree at a single point in time. For example, if a hole is created
798 * at index 10, then subsequently a hole is created at index 5,
799 * radix_tree_prev_hole covering both indexes may return 5 if called under
802 unsigned long radix_tree_prev_hole(struct radix_tree_root *root,
803 unsigned long index, unsigned long max_scan)
807 for (i = 0; i < max_scan; i++) {
808 if (!radix_tree_lookup(root, index))
811 if (index == ULONG_MAX)
817 EXPORT_SYMBOL(radix_tree_prev_hole);
820 __lookup(struct radix_tree_node *slot, void ***results, unsigned long *indices,
821 unsigned long index, unsigned int max_items, unsigned long *next_index)
823 unsigned int nr_found = 0;
824 unsigned int shift, height;
827 height = slot->height;
830 shift = (height-1) * RADIX_TREE_MAP_SHIFT;
832 for ( ; height > 1; height--) {
833 i = (index >> shift) & RADIX_TREE_MAP_MASK;
835 if (slot->slots[i] != NULL)
837 index &= ~((1UL << shift) - 1);
838 index += 1UL << shift;
840 goto out; /* 32-bit wraparound */
842 if (i == RADIX_TREE_MAP_SIZE)
846 shift -= RADIX_TREE_MAP_SHIFT;
847 slot = rcu_dereference_raw(slot->slots[i]);
852 /* Bottom level: grab some items */
853 for (i = index & RADIX_TREE_MAP_MASK; i < RADIX_TREE_MAP_SIZE; i++) {
854 if (slot->slots[i]) {
855 results[nr_found] = &(slot->slots[i]);
857 indices[nr_found] = index;
858 if (++nr_found == max_items) {
871 * radix_tree_gang_lookup - perform multiple lookup on a radix tree
872 * @root: radix tree root
873 * @results: where the results of the lookup are placed
874 * @first_index: start the lookup from this key
875 * @max_items: place up to this many items at *results
877 * Performs an index-ascending scan of the tree for present items. Places
878 * them at *@results and returns the number of items which were placed at
881 * The implementation is naive.
883 * Like radix_tree_lookup, radix_tree_gang_lookup may be called under
884 * rcu_read_lock. In this case, rather than the returned results being
885 * an atomic snapshot of the tree at a single point in time, the semantics
886 * of an RCU protected gang lookup are as though multiple radix_tree_lookups
887 * have been issued in individual locks, and results stored in 'results'.
890 radix_tree_gang_lookup(struct radix_tree_root *root, void **results,
891 unsigned long first_index, unsigned int max_items)
893 unsigned long max_index;
894 struct radix_tree_node *node;
895 unsigned long cur_index = first_index;
898 node = rcu_dereference_raw(root->rnode);
902 if (!radix_tree_is_indirect_ptr(node)) {
908 node = indirect_to_ptr(node);
910 max_index = radix_tree_maxindex(node->height);
913 while (ret < max_items) {
914 unsigned int nr_found, slots_found, i;
915 unsigned long next_index; /* Index of next search */
917 if (cur_index > max_index)
919 slots_found = __lookup(node, (void ***)results + ret, NULL,
920 cur_index, max_items - ret, &next_index);
922 for (i = 0; i < slots_found; i++) {
923 struct radix_tree_node *slot;
924 slot = *(((void ***)results)[ret + i]);
927 results[ret + nr_found] =
928 indirect_to_ptr(rcu_dereference_raw(slot));
934 cur_index = next_index;
939 EXPORT_SYMBOL(radix_tree_gang_lookup);
942 * radix_tree_gang_lookup_slot - perform multiple slot lookup on radix tree
943 * @root: radix tree root
944 * @results: where the results of the lookup are placed
945 * @indices: where their indices should be placed (but usually NULL)
946 * @first_index: start the lookup from this key
947 * @max_items: place up to this many items at *results
949 * Performs an index-ascending scan of the tree for present items. Places
950 * their slots at *@results and returns the number of items which were
951 * placed at *@results.
953 * The implementation is naive.
955 * Like radix_tree_gang_lookup as far as RCU and locking goes. Slots must
956 * be dereferenced with radix_tree_deref_slot, and if using only RCU
957 * protection, radix_tree_deref_slot may fail requiring a retry.
960 radix_tree_gang_lookup_slot(struct radix_tree_root *root,
961 void ***results, unsigned long *indices,
962 unsigned long first_index, unsigned int max_items)
964 unsigned long max_index;
965 struct radix_tree_node *node;
966 unsigned long cur_index = first_index;
969 node = rcu_dereference_raw(root->rnode);
973 if (!radix_tree_is_indirect_ptr(node)) {
976 results[0] = (void **)&root->rnode;
981 node = indirect_to_ptr(node);
983 max_index = radix_tree_maxindex(node->height);
986 while (ret < max_items) {
987 unsigned int slots_found;
988 unsigned long next_index; /* Index of next search */
990 if (cur_index > max_index)
992 slots_found = __lookup(node, results + ret,
993 indices ? indices + ret : NULL,
994 cur_index, max_items - ret, &next_index);
998 cur_index = next_index;
1003 EXPORT_SYMBOL(radix_tree_gang_lookup_slot);
1006 * FIXME: the two tag_get()s here should use find_next_bit() instead of
1007 * open-coding the search.
1010 __lookup_tag(struct radix_tree_node *slot, void ***results, unsigned long index,
1011 unsigned int max_items, unsigned long *next_index, unsigned int tag)
1013 unsigned int nr_found = 0;
1014 unsigned int shift, height;
1016 height = slot->height;
1019 shift = (height-1) * RADIX_TREE_MAP_SHIFT;
1021 while (height > 0) {
1022 unsigned long i = (index >> shift) & RADIX_TREE_MAP_MASK ;
1025 if (tag_get(slot, tag, i))
1027 index &= ~((1UL << shift) - 1);
1028 index += 1UL << shift;
1030 goto out; /* 32-bit wraparound */
1032 if (i == RADIX_TREE_MAP_SIZE)
1036 if (height == 0) { /* Bottom level: grab some items */
1037 unsigned long j = index & RADIX_TREE_MAP_MASK;
1039 for ( ; j < RADIX_TREE_MAP_SIZE; j++) {
1041 if (!tag_get(slot, tag, j))
1044 * Even though the tag was found set, we need to
1045 * recheck that we have a non-NULL node, because
1046 * if this lookup is lockless, it may have been
1047 * subsequently deleted.
1049 * Similar care must be taken in any place that
1050 * lookup ->slots[x] without a lock (ie. can't
1051 * rely on its value remaining the same).
1053 if (slot->slots[j]) {
1054 results[nr_found++] = &(slot->slots[j]);
1055 if (nr_found == max_items)
1060 shift -= RADIX_TREE_MAP_SHIFT;
1061 slot = rcu_dereference_raw(slot->slots[i]);
1066 *next_index = index;
1071 * radix_tree_gang_lookup_tag - perform multiple lookup on a radix tree
1073 * @root: radix tree root
1074 * @results: where the results of the lookup are placed
1075 * @first_index: start the lookup from this key
1076 * @max_items: place up to this many items at *results
1077 * @tag: the tag index (< RADIX_TREE_MAX_TAGS)
1079 * Performs an index-ascending scan of the tree for present items which
1080 * have the tag indexed by @tag set. Places the items at *@results and
1081 * returns the number of items which were placed at *@results.
1084 radix_tree_gang_lookup_tag(struct radix_tree_root *root, void **results,
1085 unsigned long first_index, unsigned int max_items,
1088 struct radix_tree_node *node;
1089 unsigned long max_index;
1090 unsigned long cur_index = first_index;
1093 /* check the root's tag bit */
1094 if (!root_tag_get(root, tag))
1097 node = rcu_dereference_raw(root->rnode);
1101 if (!radix_tree_is_indirect_ptr(node)) {
1102 if (first_index > 0)
1107 node = indirect_to_ptr(node);
1109 max_index = radix_tree_maxindex(node->height);
1112 while (ret < max_items) {
1113 unsigned int nr_found, slots_found, i;
1114 unsigned long next_index; /* Index of next search */
1116 if (cur_index > max_index)
1118 slots_found = __lookup_tag(node, (void ***)results + ret,
1119 cur_index, max_items - ret, &next_index, tag);
1121 for (i = 0; i < slots_found; i++) {
1122 struct radix_tree_node *slot;
1123 slot = *(((void ***)results)[ret + i]);
1126 results[ret + nr_found] =
1127 indirect_to_ptr(rcu_dereference_raw(slot));
1131 if (next_index == 0)
1133 cur_index = next_index;
1138 EXPORT_SYMBOL(radix_tree_gang_lookup_tag);
1141 * radix_tree_gang_lookup_tag_slot - perform multiple slot lookup on a
1142 * radix tree based on a tag
1143 * @root: radix tree root
1144 * @results: where the results of the lookup are placed
1145 * @first_index: start the lookup from this key
1146 * @max_items: place up to this many items at *results
1147 * @tag: the tag index (< RADIX_TREE_MAX_TAGS)
1149 * Performs an index-ascending scan of the tree for present items which
1150 * have the tag indexed by @tag set. Places the slots at *@results and
1151 * returns the number of slots which were placed at *@results.
1154 radix_tree_gang_lookup_tag_slot(struct radix_tree_root *root, void ***results,
1155 unsigned long first_index, unsigned int max_items,
1158 struct radix_tree_node *node;
1159 unsigned long max_index;
1160 unsigned long cur_index = first_index;
1163 /* check the root's tag bit */
1164 if (!root_tag_get(root, tag))
1167 node = rcu_dereference_raw(root->rnode);
1171 if (!radix_tree_is_indirect_ptr(node)) {
1172 if (first_index > 0)
1174 results[0] = (void **)&root->rnode;
1177 node = indirect_to_ptr(node);
1179 max_index = radix_tree_maxindex(node->height);
1182 while (ret < max_items) {
1183 unsigned int slots_found;
1184 unsigned long next_index; /* Index of next search */
1186 if (cur_index > max_index)
1188 slots_found = __lookup_tag(node, results + ret,
1189 cur_index, max_items - ret, &next_index, tag);
1191 if (next_index == 0)
1193 cur_index = next_index;
1198 EXPORT_SYMBOL(radix_tree_gang_lookup_tag_slot);
1200 #if defined(CONFIG_SHMEM) && defined(CONFIG_SWAP)
1201 #include <linux/sched.h> /* for cond_resched() */
1204 * This linear search is at present only useful to shmem_unuse_inode().
1206 static unsigned long __locate(struct radix_tree_node *slot, void *item,
1207 unsigned long index, unsigned long *found_index)
1209 unsigned int shift, height;
1212 height = slot->height;
1213 shift = (height-1) * RADIX_TREE_MAP_SHIFT;
1215 for ( ; height > 1; height--) {
1216 i = (index >> shift) & RADIX_TREE_MAP_MASK;
1218 if (slot->slots[i] != NULL)
1220 index &= ~((1UL << shift) - 1);
1221 index += 1UL << shift;
1223 goto out; /* 32-bit wraparound */
1225 if (i == RADIX_TREE_MAP_SIZE)
1229 shift -= RADIX_TREE_MAP_SHIFT;
1230 slot = rcu_dereference_raw(slot->slots[i]);
1235 /* Bottom level: check items */
1236 for (i = 0; i < RADIX_TREE_MAP_SIZE; i++) {
1237 if (slot->slots[i] == item) {
1238 *found_index = index + i;
1243 index += RADIX_TREE_MAP_SIZE;
1249 * radix_tree_locate_item - search through radix tree for item
1250 * @root: radix tree root
1251 * @item: item to be found
1253 * Returns index where item was found, or -1 if not found.
1254 * Caller must hold no lock (since this time-consuming function needs
1255 * to be preemptible), and must check afterwards if item is still there.
1257 unsigned long radix_tree_locate_item(struct radix_tree_root *root, void *item)
1259 struct radix_tree_node *node;
1260 unsigned long max_index;
1261 unsigned long cur_index = 0;
1262 unsigned long found_index = -1;
1266 node = rcu_dereference_raw(root->rnode);
1267 if (!radix_tree_is_indirect_ptr(node)) {
1274 node = indirect_to_ptr(node);
1275 max_index = radix_tree_maxindex(node->height);
1276 if (cur_index > max_index)
1279 cur_index = __locate(node, item, cur_index, &found_index);
1282 } while (cur_index != 0 && cur_index <= max_index);
1287 unsigned long radix_tree_locate_item(struct radix_tree_root *root, void *item)
1291 #endif /* CONFIG_SHMEM && CONFIG_SWAP */
1294 * radix_tree_shrink - shrink height of a radix tree to minimal
1295 * @root radix tree root
1297 static inline void radix_tree_shrink(struct radix_tree_root *root)
1299 /* try to shrink tree height */
1300 while (root->height > 0) {
1301 struct radix_tree_node *to_free = root->rnode;
1304 BUG_ON(!radix_tree_is_indirect_ptr(to_free));
1305 to_free = indirect_to_ptr(to_free);
1308 * The candidate node has more than one child, or its child
1309 * is not at the leftmost slot, we cannot shrink.
1311 if (to_free->count != 1)
1313 if (!to_free->slots[0])
1317 * We don't need rcu_assign_pointer(), since we are simply
1318 * moving the node from one part of the tree to another: if it
1319 * was safe to dereference the old pointer to it
1320 * (to_free->slots[0]), it will be safe to dereference the new
1321 * one (root->rnode) as far as dependent read barriers go.
1323 newptr = to_free->slots[0];
1324 if (root->height > 1)
1325 newptr = ptr_to_indirect(newptr);
1326 root->rnode = newptr;
1330 * We have a dilemma here. The node's slot[0] must not be
1331 * NULLed in case there are concurrent lookups expecting to
1332 * find the item. However if this was a bottom-level node,
1333 * then it may be subject to the slot pointer being visible
1334 * to callers dereferencing it. If item corresponding to
1335 * slot[0] is subsequently deleted, these callers would expect
1336 * their slot to become empty sooner or later.
1338 * For example, lockless pagecache will look up a slot, deref
1339 * the page pointer, and if the page is 0 refcount it means it
1340 * was concurrently deleted from pagecache so try the deref
1341 * again. Fortunately there is already a requirement for logic
1342 * to retry the entire slot lookup -- the indirect pointer
1343 * problem (replacing direct root node with an indirect pointer
1344 * also results in a stale slot). So tag the slot as indirect
1345 * to force callers to retry.
1347 if (root->height == 0)
1348 *((unsigned long *)&to_free->slots[0]) |=
1349 RADIX_TREE_INDIRECT_PTR;
1351 radix_tree_node_free(to_free);
1356 * radix_tree_delete - delete an item from a radix tree
1357 * @root: radix tree root
1360 * Remove the item at @index from the radix tree rooted at @root.
1362 * Returns the address of the deleted item, or NULL if it was not present.
1364 void *radix_tree_delete(struct radix_tree_root *root, unsigned long index)
1367 * The radix tree path needs to be one longer than the maximum path
1368 * since the "list" is null terminated.
1370 struct radix_tree_path path[RADIX_TREE_MAX_PATH + 1], *pathp = path;
1371 struct radix_tree_node *slot = NULL;
1372 struct radix_tree_node *to_free;
1373 unsigned int height, shift;
1377 height = root->height;
1378 if (index > radix_tree_maxindex(height))
1383 root_tag_clear_all(root);
1387 slot = indirect_to_ptr(slot);
1389 shift = (height - 1) * RADIX_TREE_MAP_SHIFT;
1397 offset = (index >> shift) & RADIX_TREE_MAP_MASK;
1398 pathp->offset = offset;
1400 slot = slot->slots[offset];
1401 shift -= RADIX_TREE_MAP_SHIFT;
1403 } while (height > 0);
1409 * Clear all tags associated with the just-deleted item
1411 for (tag = 0; tag < RADIX_TREE_MAX_TAGS; tag++) {
1412 if (tag_get(pathp->node, tag, pathp->offset))
1413 radix_tree_tag_clear(root, index, tag);
1417 /* Now free the nodes we do not need anymore */
1418 while (pathp->node) {
1419 pathp->node->slots[pathp->offset] = NULL;
1420 pathp->node->count--;
1422 * Queue the node for deferred freeing after the
1423 * last reference to it disappears (set NULL, above).
1426 radix_tree_node_free(to_free);
1428 if (pathp->node->count) {
1429 if (pathp->node == indirect_to_ptr(root->rnode))
1430 radix_tree_shrink(root);
1434 /* Node with zero slots in use so free it */
1435 to_free = pathp->node;
1439 root_tag_clear_all(root);
1443 radix_tree_node_free(to_free);
1448 EXPORT_SYMBOL(radix_tree_delete);
1451 * radix_tree_tagged - test whether any items in the tree are tagged
1452 * @root: radix tree root
1455 int radix_tree_tagged(struct radix_tree_root *root, unsigned int tag)
1457 return root_tag_get(root, tag);
1459 EXPORT_SYMBOL(radix_tree_tagged);
1462 radix_tree_node_ctor(void *node)
1464 memset(node, 0, sizeof(struct radix_tree_node));
1467 static __init unsigned long __maxindex(unsigned int height)
1469 unsigned int width = height * RADIX_TREE_MAP_SHIFT;
1470 int shift = RADIX_TREE_INDEX_BITS - width;
1474 if (shift >= BITS_PER_LONG)
1476 return ~0UL >> shift;
1479 static __init void radix_tree_init_maxindex(void)
1483 for (i = 0; i < ARRAY_SIZE(height_to_maxindex); i++)
1484 height_to_maxindex[i] = __maxindex(i);
1487 static int radix_tree_callback(struct notifier_block *nfb,
1488 unsigned long action,
1491 int cpu = (long)hcpu;
1492 struct radix_tree_preload *rtp;
1494 /* Free per-cpu pool of perloaded nodes */
1495 if (action == CPU_DEAD || action == CPU_DEAD_FROZEN) {
1496 rtp = &per_cpu(radix_tree_preloads, cpu);
1498 kmem_cache_free(radix_tree_node_cachep,
1499 rtp->nodes[rtp->nr-1]);
1500 rtp->nodes[rtp->nr-1] = NULL;
1507 void __init radix_tree_init(void)
1509 radix_tree_node_cachep = kmem_cache_create("radix_tree_node",
1510 sizeof(struct radix_tree_node), 0,
1511 SLAB_PANIC | SLAB_RECLAIM_ACCOUNT,
1512 radix_tree_node_ctor);
1513 radix_tree_init_maxindex();
1514 hotcpu_notifier(radix_tree_callback, 0);