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
6 * Copyright (C) 2012 Konstantin Khlebnikov
8 * This program is free software; you can redistribute it and/or
9 * modify it under the terms of the GNU General Public License as
10 * published by the Free Software Foundation; either version 2, or (at
11 * your option) any later version.
13 * This program is distributed in the hope that it will be useful, but
14 * WITHOUT ANY WARRANTY; without even the implied warranty of
15 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
16 * General Public License for more details.
18 * You should have received a copy of the GNU General Public License
19 * along with this program; if not, write to the Free Software
20 * Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA.
23 #include <linux/errno.h>
24 #include <linux/init.h>
25 #include <linux/kernel.h>
26 #include <linux/export.h>
27 #include <linux/radix-tree.h>
28 #include <linux/percpu.h>
29 #include <linux/slab.h>
30 #include <linux/notifier.h>
31 #include <linux/cpu.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 */
53 struct radix_tree_node *parent; /* Used when ascending tree */
54 struct rcu_head rcu_head; /* Used when freeing node */
56 void __rcu *slots[RADIX_TREE_MAP_SIZE];
57 unsigned long tags[RADIX_TREE_MAX_TAGS][RADIX_TREE_TAG_LONGS];
60 #define RADIX_TREE_INDEX_BITS (8 /* CHAR_BIT */ * sizeof(unsigned long))
61 #define RADIX_TREE_MAX_PATH (DIV_ROUND_UP(RADIX_TREE_INDEX_BITS, \
62 RADIX_TREE_MAP_SHIFT))
65 * The height_to_maxindex array needs to be one deeper than the maximum
66 * path as height 0 holds only 1 entry.
68 static unsigned long height_to_maxindex[RADIX_TREE_MAX_PATH + 1] __read_mostly;
71 * Radix tree node cache.
73 static struct kmem_cache *radix_tree_node_cachep;
76 * The radix tree is variable-height, so an insert operation not only has
77 * to build the branch to its corresponding item, it also has to build the
78 * branch to existing items if the size has to be increased (by
81 * The worst case is a zero height tree with just a single item at index 0,
82 * and then inserting an item at index ULONG_MAX. This requires 2 new branches
83 * of RADIX_TREE_MAX_PATH size to be created, with only the root node shared.
86 #define RADIX_TREE_PRELOAD_SIZE (RADIX_TREE_MAX_PATH * 2 - 1)
89 * Per-cpu pool of preloaded nodes
91 struct radix_tree_preload {
93 struct radix_tree_node *nodes[RADIX_TREE_PRELOAD_SIZE];
95 static DEFINE_PER_CPU(struct radix_tree_preload, radix_tree_preloads) = { 0, };
97 static inline void *ptr_to_indirect(void *ptr)
99 return (void *)((unsigned long)ptr | RADIX_TREE_INDIRECT_PTR);
102 static inline void *indirect_to_ptr(void *ptr)
104 return (void *)((unsigned long)ptr & ~RADIX_TREE_INDIRECT_PTR);
107 static inline gfp_t root_gfp_mask(struct radix_tree_root *root)
109 return root->gfp_mask & __GFP_BITS_MASK;
112 static inline void tag_set(struct radix_tree_node *node, unsigned int tag,
115 __set_bit(offset, node->tags[tag]);
118 static inline void tag_clear(struct radix_tree_node *node, unsigned int tag,
121 __clear_bit(offset, node->tags[tag]);
124 static inline int tag_get(struct radix_tree_node *node, unsigned int tag,
127 return test_bit(offset, node->tags[tag]);
130 static inline void root_tag_set(struct radix_tree_root *root, unsigned int tag)
132 root->gfp_mask |= (__force gfp_t)(1 << (tag + __GFP_BITS_SHIFT));
135 static inline void root_tag_clear(struct radix_tree_root *root, unsigned int tag)
137 root->gfp_mask &= (__force gfp_t)~(1 << (tag + __GFP_BITS_SHIFT));
140 static inline void root_tag_clear_all(struct radix_tree_root *root)
142 root->gfp_mask &= __GFP_BITS_MASK;
145 static inline int root_tag_get(struct radix_tree_root *root, unsigned int tag)
147 return (__force unsigned)root->gfp_mask & (1 << (tag + __GFP_BITS_SHIFT));
151 * Returns 1 if any slot in the node has this tag set.
152 * Otherwise returns 0.
154 static inline int any_tag_set(struct radix_tree_node *node, unsigned int tag)
157 for (idx = 0; idx < RADIX_TREE_TAG_LONGS; idx++) {
158 if (node->tags[tag][idx])
165 * radix_tree_find_next_bit - find the next set bit in a memory region
167 * @addr: The address to base the search on
168 * @size: The bitmap size in bits
169 * @offset: The bitnumber to start searching at
171 * Unrollable variant of find_next_bit() for constant size arrays.
172 * Tail bits starting from size to roundup(size, BITS_PER_LONG) must be zero.
173 * Returns next bit offset, or size if nothing found.
175 static __always_inline unsigned long
176 radix_tree_find_next_bit(const unsigned long *addr,
177 unsigned long size, unsigned long offset)
179 if (!__builtin_constant_p(size))
180 return find_next_bit(addr, size, offset);
185 addr += offset / BITS_PER_LONG;
186 tmp = *addr >> (offset % BITS_PER_LONG);
188 return __ffs(tmp) + offset;
189 offset = (offset + BITS_PER_LONG) & ~(BITS_PER_LONG - 1);
190 while (offset < size) {
193 return __ffs(tmp) + offset;
194 offset += BITS_PER_LONG;
201 * This assumes that the caller has performed appropriate preallocation, and
202 * that the caller has pinned this thread of control to the current CPU.
204 static struct radix_tree_node *
205 radix_tree_node_alloc(struct radix_tree_root *root)
207 struct radix_tree_node *ret = NULL;
208 gfp_t gfp_mask = root_gfp_mask(root);
210 if (!(gfp_mask & __GFP_WAIT)) {
211 struct radix_tree_preload *rtp;
214 * Provided the caller has preloaded here, we will always
215 * succeed in getting a node here (and never reach
218 rtp = &__get_cpu_var(radix_tree_preloads);
220 ret = rtp->nodes[rtp->nr - 1];
221 rtp->nodes[rtp->nr - 1] = NULL;
226 ret = kmem_cache_alloc(radix_tree_node_cachep, gfp_mask);
228 BUG_ON(radix_tree_is_indirect_ptr(ret));
232 static void radix_tree_node_rcu_free(struct rcu_head *head)
234 struct radix_tree_node *node =
235 container_of(head, struct radix_tree_node, rcu_head);
239 * must only free zeroed nodes into the slab. radix_tree_shrink
240 * can leave us with a non-NULL entry in the first slot, so clear
241 * that here to make sure.
243 for (i = 0; i < RADIX_TREE_MAX_TAGS; i++)
244 tag_clear(node, i, 0);
246 node->slots[0] = NULL;
249 kmem_cache_free(radix_tree_node_cachep, node);
253 radix_tree_node_free(struct radix_tree_node *node)
255 call_rcu(&node->rcu_head, radix_tree_node_rcu_free);
259 * Load up this CPU's radix_tree_node buffer with sufficient objects to
260 * ensure that the addition of a single element in the tree cannot fail. On
261 * success, return zero, with preemption disabled. On error, return -ENOMEM
262 * with preemption not disabled.
264 * To make use of this facility, the radix tree must be initialised without
265 * __GFP_WAIT being passed to INIT_RADIX_TREE().
267 int radix_tree_preload(gfp_t gfp_mask)
269 struct radix_tree_preload *rtp;
270 struct radix_tree_node *node;
274 rtp = &__get_cpu_var(radix_tree_preloads);
275 while (rtp->nr < ARRAY_SIZE(rtp->nodes)) {
277 node = kmem_cache_alloc(radix_tree_node_cachep, gfp_mask);
281 rtp = &__get_cpu_var(radix_tree_preloads);
282 if (rtp->nr < ARRAY_SIZE(rtp->nodes))
283 rtp->nodes[rtp->nr++] = node;
285 kmem_cache_free(radix_tree_node_cachep, node);
291 EXPORT_SYMBOL(radix_tree_preload);
294 * Return the maximum key which can be store into a
295 * radix tree with height HEIGHT.
297 static inline unsigned long radix_tree_maxindex(unsigned int height)
299 return height_to_maxindex[height];
303 * Extend a radix tree so it can store key @index.
305 static int radix_tree_extend(struct radix_tree_root *root, unsigned long index)
307 struct radix_tree_node *node;
308 struct radix_tree_node *slot;
312 /* Figure out what the height should be. */
313 height = root->height + 1;
314 while (index > radix_tree_maxindex(height))
317 if (root->rnode == NULL) {
318 root->height = height;
323 unsigned int newheight;
324 if (!(node = radix_tree_node_alloc(root)))
327 /* Propagate the aggregated tag info into the new root */
328 for (tag = 0; tag < RADIX_TREE_MAX_TAGS; tag++) {
329 if (root_tag_get(root, tag))
330 tag_set(node, tag, 0);
333 /* Increase the height. */
334 newheight = root->height+1;
335 node->height = newheight;
340 slot = indirect_to_ptr(slot);
343 node->slots[0] = slot;
344 node = ptr_to_indirect(node);
345 rcu_assign_pointer(root->rnode, node);
346 root->height = newheight;
347 } while (height > root->height);
353 * radix_tree_insert - insert into a radix tree
354 * @root: radix tree root
356 * @item: item to insert
358 * Insert an item into the radix tree at position @index.
360 int radix_tree_insert(struct radix_tree_root *root,
361 unsigned long index, void *item)
363 struct radix_tree_node *node = NULL, *slot;
364 unsigned int height, shift;
368 BUG_ON(radix_tree_is_indirect_ptr(item));
370 /* Make sure the tree is high enough. */
371 if (index > radix_tree_maxindex(root->height)) {
372 error = radix_tree_extend(root, index);
377 slot = indirect_to_ptr(root->rnode);
379 height = root->height;
380 shift = (height-1) * RADIX_TREE_MAP_SHIFT;
382 offset = 0; /* uninitialised var warning */
385 /* Have to add a child node. */
386 if (!(slot = radix_tree_node_alloc(root)))
388 slot->height = height;
391 rcu_assign_pointer(node->slots[offset], slot);
394 rcu_assign_pointer(root->rnode, ptr_to_indirect(slot));
397 /* Go a level down */
398 offset = (index >> shift) & RADIX_TREE_MAP_MASK;
400 slot = node->slots[offset];
401 shift -= RADIX_TREE_MAP_SHIFT;
410 rcu_assign_pointer(node->slots[offset], item);
411 BUG_ON(tag_get(node, 0, offset));
412 BUG_ON(tag_get(node, 1, offset));
414 rcu_assign_pointer(root->rnode, item);
415 BUG_ON(root_tag_get(root, 0));
416 BUG_ON(root_tag_get(root, 1));
421 EXPORT_SYMBOL(radix_tree_insert);
424 * is_slot == 1 : search for the slot.
425 * is_slot == 0 : search for the node.
427 static void *radix_tree_lookup_element(struct radix_tree_root *root,
428 unsigned long index, int is_slot)
430 unsigned int height, shift;
431 struct radix_tree_node *node, **slot;
433 node = rcu_dereference_raw(root->rnode);
437 if (!radix_tree_is_indirect_ptr(node)) {
440 return is_slot ? (void *)&root->rnode : node;
442 node = indirect_to_ptr(node);
444 height = node->height;
445 if (index > radix_tree_maxindex(height))
448 shift = (height-1) * RADIX_TREE_MAP_SHIFT;
451 slot = (struct radix_tree_node **)
452 (node->slots + ((index>>shift) & RADIX_TREE_MAP_MASK));
453 node = rcu_dereference_raw(*slot);
457 shift -= RADIX_TREE_MAP_SHIFT;
459 } while (height > 0);
461 return is_slot ? (void *)slot : indirect_to_ptr(node);
465 * radix_tree_lookup_slot - lookup a slot in a radix tree
466 * @root: radix tree root
469 * Returns: the slot corresponding to the position @index in the
470 * radix tree @root. This is useful for update-if-exists operations.
472 * This function can be called under rcu_read_lock iff the slot is not
473 * modified by radix_tree_replace_slot, otherwise it must be called
474 * exclusive from other writers. Any dereference of the slot must be done
475 * using radix_tree_deref_slot.
477 void **radix_tree_lookup_slot(struct radix_tree_root *root, unsigned long index)
479 return (void **)radix_tree_lookup_element(root, index, 1);
481 EXPORT_SYMBOL(radix_tree_lookup_slot);
484 * radix_tree_lookup - perform lookup operation on a radix tree
485 * @root: radix tree root
488 * Lookup the item at the position @index in the radix tree @root.
490 * This function can be called under rcu_read_lock, however the caller
491 * must manage lifetimes of leaf nodes (eg. RCU may also be used to free
492 * them safely). No RCU barriers are required to access or modify the
493 * returned item, however.
495 void *radix_tree_lookup(struct radix_tree_root *root, unsigned long index)
497 return radix_tree_lookup_element(root, index, 0);
499 EXPORT_SYMBOL(radix_tree_lookup);
502 * radix_tree_tag_set - set a tag on a radix tree node
503 * @root: radix tree root
507 * Set the search tag (which must be < RADIX_TREE_MAX_TAGS)
508 * corresponding to @index in the radix tree. From
509 * the root all the way down to the leaf node.
511 * Returns the address of the tagged item. Setting a tag on a not-present
514 void *radix_tree_tag_set(struct radix_tree_root *root,
515 unsigned long index, unsigned int tag)
517 unsigned int height, shift;
518 struct radix_tree_node *slot;
520 height = root->height;
521 BUG_ON(index > radix_tree_maxindex(height));
523 slot = indirect_to_ptr(root->rnode);
524 shift = (height - 1) * RADIX_TREE_MAP_SHIFT;
529 offset = (index >> shift) & RADIX_TREE_MAP_MASK;
530 if (!tag_get(slot, tag, offset))
531 tag_set(slot, tag, offset);
532 slot = slot->slots[offset];
533 BUG_ON(slot == NULL);
534 shift -= RADIX_TREE_MAP_SHIFT;
538 /* set the root's tag bit */
539 if (slot && !root_tag_get(root, tag))
540 root_tag_set(root, tag);
544 EXPORT_SYMBOL(radix_tree_tag_set);
547 * radix_tree_tag_clear - clear a tag on a radix tree node
548 * @root: radix tree root
552 * Clear the search tag (which must be < RADIX_TREE_MAX_TAGS)
553 * corresponding to @index in the radix tree. If
554 * this causes the leaf node to have no tags set then clear the tag in the
555 * next-to-leaf node, etc.
557 * Returns the address of the tagged item on success, else NULL. ie:
558 * has the same return value and semantics as radix_tree_lookup().
560 void *radix_tree_tag_clear(struct radix_tree_root *root,
561 unsigned long index, unsigned int tag)
563 struct radix_tree_node *node = NULL;
564 struct radix_tree_node *slot = NULL;
565 unsigned int height, shift;
566 int uninitialized_var(offset);
568 height = root->height;
569 if (index > radix_tree_maxindex(height))
572 shift = height * RADIX_TREE_MAP_SHIFT;
573 slot = indirect_to_ptr(root->rnode);
579 shift -= RADIX_TREE_MAP_SHIFT;
580 offset = (index >> shift) & RADIX_TREE_MAP_MASK;
582 slot = slot->slots[offset];
589 if (!tag_get(node, tag, offset))
591 tag_clear(node, tag, offset);
592 if (any_tag_set(node, tag))
595 index >>= RADIX_TREE_MAP_SHIFT;
596 offset = index & RADIX_TREE_MAP_MASK;
600 /* clear the root's tag bit */
601 if (root_tag_get(root, tag))
602 root_tag_clear(root, tag);
607 EXPORT_SYMBOL(radix_tree_tag_clear);
610 * radix_tree_tag_get - get a tag on a radix tree node
611 * @root: radix tree root
613 * @tag: tag index (< RADIX_TREE_MAX_TAGS)
617 * 0: tag not present or not set
620 * Note that the return value of this function may not be relied on, even if
621 * the RCU lock is held, unless tag modification and node deletion are excluded
624 int radix_tree_tag_get(struct radix_tree_root *root,
625 unsigned long index, unsigned int tag)
627 unsigned int height, shift;
628 struct radix_tree_node *node;
630 /* check the root's tag bit */
631 if (!root_tag_get(root, tag))
634 node = rcu_dereference_raw(root->rnode);
638 if (!radix_tree_is_indirect_ptr(node))
640 node = indirect_to_ptr(node);
642 height = node->height;
643 if (index > radix_tree_maxindex(height))
646 shift = (height - 1) * RADIX_TREE_MAP_SHIFT;
654 offset = (index >> shift) & RADIX_TREE_MAP_MASK;
655 if (!tag_get(node, tag, offset))
659 node = rcu_dereference_raw(node->slots[offset]);
660 shift -= RADIX_TREE_MAP_SHIFT;
664 EXPORT_SYMBOL(radix_tree_tag_get);
667 * radix_tree_next_chunk - find next chunk of slots for iteration
669 * @root: radix tree root
670 * @iter: iterator state
671 * @flags: RADIX_TREE_ITER_* flags and tag index
672 * Returns: pointer to chunk first slot, or NULL if iteration is over
674 void **radix_tree_next_chunk(struct radix_tree_root *root,
675 struct radix_tree_iter *iter, unsigned flags)
677 unsigned shift, tag = flags & RADIX_TREE_ITER_TAG_MASK;
678 struct radix_tree_node *rnode, *node;
679 unsigned long index, offset;
681 if ((flags & RADIX_TREE_ITER_TAGGED) && !root_tag_get(root, tag))
685 * Catch next_index overflow after ~0UL. iter->index never overflows
686 * during iterating; it can be zero only at the beginning.
687 * And we cannot overflow iter->next_index in a single step,
688 * because RADIX_TREE_MAP_SHIFT < BITS_PER_LONG.
690 * This condition also used by radix_tree_next_slot() to stop
691 * contiguous iterating, and forbid swithing to the next chunk.
693 index = iter->next_index;
694 if (!index && iter->index)
697 rnode = rcu_dereference_raw(root->rnode);
698 if (radix_tree_is_indirect_ptr(rnode)) {
699 rnode = indirect_to_ptr(rnode);
700 } else if (rnode && !index) {
701 /* Single-slot tree */
703 iter->next_index = 1;
705 return (void **)&root->rnode;
710 shift = (rnode->height - 1) * RADIX_TREE_MAP_SHIFT;
711 offset = index >> shift;
713 /* Index outside of the tree */
714 if (offset >= RADIX_TREE_MAP_SIZE)
719 if ((flags & RADIX_TREE_ITER_TAGGED) ?
720 !test_bit(offset, node->tags[tag]) :
721 !node->slots[offset]) {
723 if (flags & RADIX_TREE_ITER_CONTIG)
726 if (flags & RADIX_TREE_ITER_TAGGED)
727 offset = radix_tree_find_next_bit(
732 while (++offset < RADIX_TREE_MAP_SIZE) {
733 if (node->slots[offset])
736 index &= ~((RADIX_TREE_MAP_SIZE << shift) - 1);
737 index += offset << shift;
738 /* Overflow after ~0UL */
741 if (offset == RADIX_TREE_MAP_SIZE)
745 /* This is leaf-node */
749 node = rcu_dereference_raw(node->slots[offset]);
752 shift -= RADIX_TREE_MAP_SHIFT;
753 offset = (index >> shift) & RADIX_TREE_MAP_MASK;
756 /* Update the iterator state */
758 iter->next_index = (index | RADIX_TREE_MAP_MASK) + 1;
760 /* Construct iter->tags bit-mask from node->tags[tag] array */
761 if (flags & RADIX_TREE_ITER_TAGGED) {
762 unsigned tag_long, tag_bit;
764 tag_long = offset / BITS_PER_LONG;
765 tag_bit = offset % BITS_PER_LONG;
766 iter->tags = node->tags[tag][tag_long] >> tag_bit;
767 /* This never happens if RADIX_TREE_TAG_LONGS == 1 */
768 if (tag_long < RADIX_TREE_TAG_LONGS - 1) {
769 /* Pick tags from next element */
771 iter->tags |= node->tags[tag][tag_long + 1] <<
772 (BITS_PER_LONG - tag_bit);
773 /* Clip chunk size, here only BITS_PER_LONG tags */
774 iter->next_index = index + BITS_PER_LONG;
778 return node->slots + offset;
780 EXPORT_SYMBOL(radix_tree_next_chunk);
783 * radix_tree_range_tag_if_tagged - for each item in given range set given
784 * tag if item has another tag set
785 * @root: radix tree root
786 * @first_indexp: pointer to a starting index of a range to scan
787 * @last_index: last index of a range to scan
788 * @nr_to_tag: maximum number items to tag
789 * @iftag: tag index to test
790 * @settag: tag index to set if tested tag is set
792 * This function scans range of radix tree from first_index to last_index
793 * (inclusive). For each item in the range if iftag is set, the function sets
794 * also settag. The function stops either after tagging nr_to_tag items or
795 * after reaching last_index.
797 * The tags must be set from the leaf level only and propagated back up the
798 * path to the root. We must do this so that we resolve the full path before
799 * setting any tags on intermediate nodes. If we set tags as we descend, then
800 * we can get to the leaf node and find that the index that has the iftag
801 * set is outside the range we are scanning. This reults in dangling tags and
802 * can lead to problems with later tag operations (e.g. livelocks on lookups).
804 * The function returns number of leaves where the tag was set and sets
805 * *first_indexp to the first unscanned index.
806 * WARNING! *first_indexp can wrap if last_index is ULONG_MAX. Caller must
807 * be prepared to handle that.
809 unsigned long radix_tree_range_tag_if_tagged(struct radix_tree_root *root,
810 unsigned long *first_indexp, unsigned long last_index,
811 unsigned long nr_to_tag,
812 unsigned int iftag, unsigned int settag)
814 unsigned int height = root->height;
815 struct radix_tree_node *node = NULL;
816 struct radix_tree_node *slot;
818 unsigned long tagged = 0;
819 unsigned long index = *first_indexp;
821 last_index = min(last_index, radix_tree_maxindex(height));
822 if (index > last_index)
826 if (!root_tag_get(root, iftag)) {
827 *first_indexp = last_index + 1;
831 *first_indexp = last_index + 1;
832 root_tag_set(root, settag);
836 shift = (height - 1) * RADIX_TREE_MAP_SHIFT;
837 slot = indirect_to_ptr(root->rnode);
840 unsigned long upindex;
843 offset = (index >> shift) & RADIX_TREE_MAP_MASK;
844 if (!slot->slots[offset])
846 if (!tag_get(slot, iftag, offset))
849 /* Go down one level */
850 shift -= RADIX_TREE_MAP_SHIFT;
852 slot = slot->slots[offset];
858 tag_set(slot, settag, offset);
860 /* walk back up the path tagging interior nodes */
863 upindex >>= RADIX_TREE_MAP_SHIFT;
864 offset = upindex & RADIX_TREE_MAP_MASK;
866 /* stop if we find a node with the tag already set */
867 if (tag_get(node, settag, offset))
869 tag_set(node, settag, offset);
874 * Small optimization: now clear that node pointer.
875 * Since all of this slot's ancestors now have the tag set
876 * from setting it above, we have no further need to walk
877 * back up the tree setting tags, until we update slot to
878 * point to another radix_tree_node.
883 /* Go to next item at level determined by 'shift' */
884 index = ((index >> shift) + 1) << shift;
885 /* Overflow can happen when last_index is ~0UL... */
886 if (index > last_index || !index)
888 if (tagged >= nr_to_tag)
890 while (((index >> shift) & RADIX_TREE_MAP_MASK) == 0) {
892 * We've fully scanned this node. Go up. Because
893 * last_index is guaranteed to be in the tree, what
894 * we do below cannot wander astray.
897 shift += RADIX_TREE_MAP_SHIFT;
901 * We need not to tag the root tag if there is no tag which is set with
902 * settag within the range from *first_indexp to last_index.
905 root_tag_set(root, settag);
906 *first_indexp = index;
910 EXPORT_SYMBOL(radix_tree_range_tag_if_tagged);
914 * radix_tree_next_hole - find the next hole (not-present entry)
917 * @max_scan: maximum range to search
919 * Search the set [index, min(index+max_scan-1, MAX_INDEX)] for the lowest
922 * Returns: the index of the hole if found, otherwise returns an index
923 * outside of the set specified (in which case 'return - index >= max_scan'
924 * will be true). In rare cases of index wrap-around, 0 will be returned.
926 * radix_tree_next_hole may be called under rcu_read_lock. However, like
927 * radix_tree_gang_lookup, this will not atomically search a snapshot of
928 * the tree at a single point in time. For example, if a hole is created
929 * at index 5, then subsequently a hole is created at index 10,
930 * radix_tree_next_hole covering both indexes may return 10 if called
931 * under rcu_read_lock.
933 unsigned long radix_tree_next_hole(struct radix_tree_root *root,
934 unsigned long index, unsigned long max_scan)
938 for (i = 0; i < max_scan; i++) {
939 if (!radix_tree_lookup(root, index))
948 EXPORT_SYMBOL(radix_tree_next_hole);
951 * radix_tree_prev_hole - find the prev hole (not-present entry)
954 * @max_scan: maximum range to search
956 * Search backwards in the range [max(index-max_scan+1, 0), index]
957 * for the first hole.
959 * Returns: the index of the hole if found, otherwise returns an index
960 * outside of the set specified (in which case 'index - return >= max_scan'
961 * will be true). In rare cases of wrap-around, ULONG_MAX will be returned.
963 * radix_tree_next_hole may be called under rcu_read_lock. However, like
964 * radix_tree_gang_lookup, this will not atomically search a snapshot of
965 * the tree at a single point in time. For example, if a hole is created
966 * at index 10, then subsequently a hole is created at index 5,
967 * radix_tree_prev_hole covering both indexes may return 5 if called under
970 unsigned long radix_tree_prev_hole(struct radix_tree_root *root,
971 unsigned long index, unsigned long max_scan)
975 for (i = 0; i < max_scan; i++) {
976 if (!radix_tree_lookup(root, index))
979 if (index == ULONG_MAX)
985 EXPORT_SYMBOL(radix_tree_prev_hole);
988 * radix_tree_gang_lookup - perform multiple lookup on a radix tree
989 * @root: radix tree root
990 * @results: where the results of the lookup are placed
991 * @first_index: start the lookup from this key
992 * @max_items: place up to this many items at *results
994 * Performs an index-ascending scan of the tree for present items. Places
995 * them at *@results and returns the number of items which were placed at
998 * The implementation is naive.
1000 * Like radix_tree_lookup, radix_tree_gang_lookup may be called under
1001 * rcu_read_lock. In this case, rather than the returned results being
1002 * an atomic snapshot of the tree at a single point in time, the semantics
1003 * of an RCU protected gang lookup are as though multiple radix_tree_lookups
1004 * have been issued in individual locks, and results stored in 'results'.
1007 radix_tree_gang_lookup(struct radix_tree_root *root, void **results,
1008 unsigned long first_index, unsigned int max_items)
1010 struct radix_tree_iter iter;
1012 unsigned int ret = 0;
1014 if (unlikely(!max_items))
1017 radix_tree_for_each_slot(slot, root, &iter, first_index) {
1018 results[ret] = indirect_to_ptr(rcu_dereference_raw(*slot));
1021 if (++ret == max_items)
1027 EXPORT_SYMBOL(radix_tree_gang_lookup);
1030 * radix_tree_gang_lookup_slot - perform multiple slot lookup on radix tree
1031 * @root: radix tree root
1032 * @results: where the results of the lookup are placed
1033 * @indices: where their indices should be placed (but usually NULL)
1034 * @first_index: start the lookup from this key
1035 * @max_items: place up to this many items at *results
1037 * Performs an index-ascending scan of the tree for present items. Places
1038 * their slots at *@results and returns the number of items which were
1039 * placed at *@results.
1041 * The implementation is naive.
1043 * Like radix_tree_gang_lookup as far as RCU and locking goes. Slots must
1044 * be dereferenced with radix_tree_deref_slot, and if using only RCU
1045 * protection, radix_tree_deref_slot may fail requiring a retry.
1048 radix_tree_gang_lookup_slot(struct radix_tree_root *root,
1049 void ***results, unsigned long *indices,
1050 unsigned long first_index, unsigned int max_items)
1052 struct radix_tree_iter iter;
1054 unsigned int ret = 0;
1056 if (unlikely(!max_items))
1059 radix_tree_for_each_slot(slot, root, &iter, first_index) {
1060 results[ret] = slot;
1062 indices[ret] = iter.index;
1063 if (++ret == max_items)
1069 EXPORT_SYMBOL(radix_tree_gang_lookup_slot);
1072 * radix_tree_gang_lookup_tag - perform multiple lookup on a radix tree
1074 * @root: radix tree root
1075 * @results: where the results of the lookup are placed
1076 * @first_index: start the lookup from this key
1077 * @max_items: place up to this many items at *results
1078 * @tag: the tag index (< RADIX_TREE_MAX_TAGS)
1080 * Performs an index-ascending scan of the tree for present items which
1081 * have the tag indexed by @tag set. Places the items at *@results and
1082 * returns the number of items which were placed at *@results.
1085 radix_tree_gang_lookup_tag(struct radix_tree_root *root, void **results,
1086 unsigned long first_index, unsigned int max_items,
1089 struct radix_tree_iter iter;
1091 unsigned int ret = 0;
1093 if (unlikely(!max_items))
1096 radix_tree_for_each_tagged(slot, root, &iter, first_index, tag) {
1097 results[ret] = indirect_to_ptr(rcu_dereference_raw(*slot));
1100 if (++ret == max_items)
1106 EXPORT_SYMBOL(radix_tree_gang_lookup_tag);
1109 * radix_tree_gang_lookup_tag_slot - perform multiple slot lookup on a
1110 * radix tree based on a tag
1111 * @root: radix tree root
1112 * @results: where the results of the lookup are placed
1113 * @first_index: start the lookup from this key
1114 * @max_items: place up to this many items at *results
1115 * @tag: the tag index (< RADIX_TREE_MAX_TAGS)
1117 * Performs an index-ascending scan of the tree for present items which
1118 * have the tag indexed by @tag set. Places the slots at *@results and
1119 * returns the number of slots which were placed at *@results.
1122 radix_tree_gang_lookup_tag_slot(struct radix_tree_root *root, void ***results,
1123 unsigned long first_index, unsigned int max_items,
1126 struct radix_tree_iter iter;
1128 unsigned int ret = 0;
1130 if (unlikely(!max_items))
1133 radix_tree_for_each_tagged(slot, root, &iter, first_index, tag) {
1134 results[ret] = slot;
1135 if (++ret == max_items)
1141 EXPORT_SYMBOL(radix_tree_gang_lookup_tag_slot);
1143 #if defined(CONFIG_SHMEM) && defined(CONFIG_SWAP)
1144 #include <linux/sched.h> /* for cond_resched() */
1147 * This linear search is at present only useful to shmem_unuse_inode().
1149 static unsigned long __locate(struct radix_tree_node *slot, void *item,
1150 unsigned long index, unsigned long *found_index)
1152 unsigned int shift, height;
1155 height = slot->height;
1156 shift = (height-1) * RADIX_TREE_MAP_SHIFT;
1158 for ( ; height > 1; height--) {
1159 i = (index >> shift) & RADIX_TREE_MAP_MASK;
1161 if (slot->slots[i] != NULL)
1163 index &= ~((1UL << shift) - 1);
1164 index += 1UL << shift;
1166 goto out; /* 32-bit wraparound */
1168 if (i == RADIX_TREE_MAP_SIZE)
1172 shift -= RADIX_TREE_MAP_SHIFT;
1173 slot = rcu_dereference_raw(slot->slots[i]);
1178 /* Bottom level: check items */
1179 for (i = 0; i < RADIX_TREE_MAP_SIZE; i++) {
1180 if (slot->slots[i] == item) {
1181 *found_index = index + i;
1186 index += RADIX_TREE_MAP_SIZE;
1192 * radix_tree_locate_item - search through radix tree for item
1193 * @root: radix tree root
1194 * @item: item to be found
1196 * Returns index where item was found, or -1 if not found.
1197 * Caller must hold no lock (since this time-consuming function needs
1198 * to be preemptible), and must check afterwards if item is still there.
1200 unsigned long radix_tree_locate_item(struct radix_tree_root *root, void *item)
1202 struct radix_tree_node *node;
1203 unsigned long max_index;
1204 unsigned long cur_index = 0;
1205 unsigned long found_index = -1;
1209 node = rcu_dereference_raw(root->rnode);
1210 if (!radix_tree_is_indirect_ptr(node)) {
1217 node = indirect_to_ptr(node);
1218 max_index = radix_tree_maxindex(node->height);
1219 if (cur_index > max_index)
1222 cur_index = __locate(node, item, cur_index, &found_index);
1225 } while (cur_index != 0 && cur_index <= max_index);
1230 unsigned long radix_tree_locate_item(struct radix_tree_root *root, void *item)
1234 #endif /* CONFIG_SHMEM && CONFIG_SWAP */
1237 * radix_tree_shrink - shrink height of a radix tree to minimal
1238 * @root radix tree root
1240 static inline void radix_tree_shrink(struct radix_tree_root *root)
1242 /* try to shrink tree height */
1243 while (root->height > 0) {
1244 struct radix_tree_node *to_free = root->rnode;
1245 struct radix_tree_node *slot;
1247 BUG_ON(!radix_tree_is_indirect_ptr(to_free));
1248 to_free = indirect_to_ptr(to_free);
1251 * The candidate node has more than one child, or its child
1252 * is not at the leftmost slot, we cannot shrink.
1254 if (to_free->count != 1)
1256 if (!to_free->slots[0])
1260 * We don't need rcu_assign_pointer(), since we are simply
1261 * moving the node from one part of the tree to another: if it
1262 * was safe to dereference the old pointer to it
1263 * (to_free->slots[0]), it will be safe to dereference the new
1264 * one (root->rnode) as far as dependent read barriers go.
1266 slot = to_free->slots[0];
1267 if (root->height > 1) {
1268 slot->parent = NULL;
1269 slot = ptr_to_indirect(slot);
1275 * We have a dilemma here. The node's slot[0] must not be
1276 * NULLed in case there are concurrent lookups expecting to
1277 * find the item. However if this was a bottom-level node,
1278 * then it may be subject to the slot pointer being visible
1279 * to callers dereferencing it. If item corresponding to
1280 * slot[0] is subsequently deleted, these callers would expect
1281 * their slot to become empty sooner or later.
1283 * For example, lockless pagecache will look up a slot, deref
1284 * the page pointer, and if the page is 0 refcount it means it
1285 * was concurrently deleted from pagecache so try the deref
1286 * again. Fortunately there is already a requirement for logic
1287 * to retry the entire slot lookup -- the indirect pointer
1288 * problem (replacing direct root node with an indirect pointer
1289 * also results in a stale slot). So tag the slot as indirect
1290 * to force callers to retry.
1292 if (root->height == 0)
1293 *((unsigned long *)&to_free->slots[0]) |=
1294 RADIX_TREE_INDIRECT_PTR;
1296 radix_tree_node_free(to_free);
1301 * radix_tree_delete - delete an item from a radix tree
1302 * @root: radix tree root
1305 * Remove the item at @index from the radix tree rooted at @root.
1307 * Returns the address of the deleted item, or NULL if it was not present.
1309 void *radix_tree_delete(struct radix_tree_root *root, unsigned long index)
1311 struct radix_tree_node *node = NULL;
1312 struct radix_tree_node *slot = NULL;
1313 struct radix_tree_node *to_free;
1314 unsigned int height, shift;
1316 int uninitialized_var(offset);
1318 height = root->height;
1319 if (index > radix_tree_maxindex(height))
1324 root_tag_clear_all(root);
1328 slot = indirect_to_ptr(slot);
1329 shift = height * RADIX_TREE_MAP_SHIFT;
1335 shift -= RADIX_TREE_MAP_SHIFT;
1336 offset = (index >> shift) & RADIX_TREE_MAP_MASK;
1338 slot = slot->slots[offset];
1345 * Clear all tags associated with the item to be deleted.
1346 * This way of doing it would be inefficient, but seldom is any set.
1348 for (tag = 0; tag < RADIX_TREE_MAX_TAGS; tag++) {
1349 if (tag_get(node, tag, offset))
1350 radix_tree_tag_clear(root, index, tag);
1354 /* Now free the nodes we do not need anymore */
1356 node->slots[offset] = NULL;
1359 * Queue the node for deferred freeing after the
1360 * last reference to it disappears (set NULL, above).
1363 radix_tree_node_free(to_free);
1366 if (node == indirect_to_ptr(root->rnode))
1367 radix_tree_shrink(root);
1371 /* Node with zero slots in use so free it */
1374 index >>= RADIX_TREE_MAP_SHIFT;
1375 offset = index & RADIX_TREE_MAP_MASK;
1376 node = node->parent;
1379 root_tag_clear_all(root);
1383 radix_tree_node_free(to_free);
1388 EXPORT_SYMBOL(radix_tree_delete);
1391 * radix_tree_tagged - test whether any items in the tree are tagged
1392 * @root: radix tree root
1395 int radix_tree_tagged(struct radix_tree_root *root, unsigned int tag)
1397 return root_tag_get(root, tag);
1399 EXPORT_SYMBOL(radix_tree_tagged);
1402 radix_tree_node_ctor(void *node)
1404 memset(node, 0, sizeof(struct radix_tree_node));
1407 static __init unsigned long __maxindex(unsigned int height)
1409 unsigned int width = height * RADIX_TREE_MAP_SHIFT;
1410 int shift = RADIX_TREE_INDEX_BITS - width;
1414 if (shift >= BITS_PER_LONG)
1416 return ~0UL >> shift;
1419 static __init void radix_tree_init_maxindex(void)
1423 for (i = 0; i < ARRAY_SIZE(height_to_maxindex); i++)
1424 height_to_maxindex[i] = __maxindex(i);
1427 static int radix_tree_callback(struct notifier_block *nfb,
1428 unsigned long action,
1431 int cpu = (long)hcpu;
1432 struct radix_tree_preload *rtp;
1434 /* Free per-cpu pool of perloaded nodes */
1435 if (action == CPU_DEAD || action == CPU_DEAD_FROZEN) {
1436 rtp = &per_cpu(radix_tree_preloads, cpu);
1438 kmem_cache_free(radix_tree_node_cachep,
1439 rtp->nodes[rtp->nr-1]);
1440 rtp->nodes[rtp->nr-1] = NULL;
1447 void __init radix_tree_init(void)
1449 radix_tree_node_cachep = kmem_cache_create("radix_tree_node",
1450 sizeof(struct radix_tree_node), 0,
1451 SLAB_PANIC | SLAB_RECLAIM_ACCOUNT,
1452 radix_tree_node_ctor);
1453 radix_tree_init_maxindex();
1454 hotcpu_notifier(radix_tree_callback, 0);