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/kmemleak.h>
31 #include <linux/notifier.h>
32 #include <linux/cpu.h>
33 #include <linux/string.h>
34 #include <linux/bitops.h>
35 #include <linux/rcupdate.h>
36 #include <linux/preempt.h> /* in_interrupt() */
40 * The height_to_maxindex array needs to be one deeper than the maximum
41 * path as height 0 holds only 1 entry.
43 static unsigned long height_to_maxindex[RADIX_TREE_MAX_PATH + 1] __read_mostly;
46 * Radix tree node cache.
48 static struct kmem_cache *radix_tree_node_cachep;
51 * The radix tree is variable-height, so an insert operation not only has
52 * to build the branch to its corresponding item, it also has to build the
53 * branch to existing items if the size has to be increased (by
56 * The worst case is a zero height tree with just a single item at index 0,
57 * and then inserting an item at index ULONG_MAX. This requires 2 new branches
58 * of RADIX_TREE_MAX_PATH size to be created, with only the root node shared.
61 #define RADIX_TREE_PRELOAD_SIZE (RADIX_TREE_MAX_PATH * 2 - 1)
64 * Per-cpu pool of preloaded nodes
66 struct radix_tree_preload {
68 /* nodes->private_data points to next preallocated node */
69 struct radix_tree_node *nodes;
71 static DEFINE_PER_CPU(struct radix_tree_preload, radix_tree_preloads) = { 0, };
73 static inline void *ptr_to_indirect(void *ptr)
75 return (void *)((unsigned long)ptr | RADIX_TREE_INDIRECT_PTR);
78 static inline void *indirect_to_ptr(void *ptr)
80 return (void *)((unsigned long)ptr & ~RADIX_TREE_INDIRECT_PTR);
83 static inline gfp_t root_gfp_mask(struct radix_tree_root *root)
85 return root->gfp_mask & __GFP_BITS_MASK;
88 static inline void tag_set(struct radix_tree_node *node, unsigned int tag,
91 __set_bit(offset, node->tags[tag]);
94 static inline void tag_clear(struct radix_tree_node *node, unsigned int tag,
97 __clear_bit(offset, node->tags[tag]);
100 static inline int tag_get(struct radix_tree_node *node, unsigned int tag,
103 return test_bit(offset, node->tags[tag]);
106 static inline void root_tag_set(struct radix_tree_root *root, unsigned int tag)
108 root->gfp_mask |= (__force gfp_t)(1 << (tag + __GFP_BITS_SHIFT));
111 static inline void root_tag_clear(struct radix_tree_root *root, unsigned int tag)
113 root->gfp_mask &= (__force gfp_t)~(1 << (tag + __GFP_BITS_SHIFT));
116 static inline void root_tag_clear_all(struct radix_tree_root *root)
118 root->gfp_mask &= __GFP_BITS_MASK;
121 static inline int root_tag_get(struct radix_tree_root *root, unsigned int tag)
123 return (__force unsigned)root->gfp_mask & (1 << (tag + __GFP_BITS_SHIFT));
127 * Returns 1 if any slot in the node has this tag set.
128 * Otherwise returns 0.
130 static inline int any_tag_set(struct radix_tree_node *node, unsigned int tag)
133 for (idx = 0; idx < RADIX_TREE_TAG_LONGS; idx++) {
134 if (node->tags[tag][idx])
141 * radix_tree_find_next_bit - find the next set bit in a memory region
143 * @addr: The address to base the search on
144 * @size: The bitmap size in bits
145 * @offset: The bitnumber to start searching at
147 * Unrollable variant of find_next_bit() for constant size arrays.
148 * Tail bits starting from size to roundup(size, BITS_PER_LONG) must be zero.
149 * Returns next bit offset, or size if nothing found.
151 static __always_inline unsigned long
152 radix_tree_find_next_bit(const unsigned long *addr,
153 unsigned long size, unsigned long offset)
155 if (!__builtin_constant_p(size))
156 return find_next_bit(addr, size, offset);
161 addr += offset / BITS_PER_LONG;
162 tmp = *addr >> (offset % BITS_PER_LONG);
164 return __ffs(tmp) + offset;
165 offset = (offset + BITS_PER_LONG) & ~(BITS_PER_LONG - 1);
166 while (offset < size) {
169 return __ffs(tmp) + offset;
170 offset += BITS_PER_LONG;
177 * This assumes that the caller has performed appropriate preallocation, and
178 * that the caller has pinned this thread of control to the current CPU.
180 static struct radix_tree_node *
181 radix_tree_node_alloc(struct radix_tree_root *root)
183 struct radix_tree_node *ret = NULL;
184 gfp_t gfp_mask = root_gfp_mask(root);
187 * Preload code isn't irq safe and it doesn't make sence to use
188 * preloading in the interrupt anyway as all the allocations have to
189 * be atomic. So just do normal allocation when in interrupt.
191 if (!gfpflags_allow_blocking(gfp_mask) && !in_interrupt()) {
192 struct radix_tree_preload *rtp;
195 * Even if the caller has preloaded, try to allocate from the
196 * cache first for the new node to get accounted.
198 ret = kmem_cache_alloc(radix_tree_node_cachep,
199 gfp_mask | __GFP_ACCOUNT | __GFP_NOWARN);
204 * Provided the caller has preloaded here, we will always
205 * succeed in getting a node here (and never reach
208 rtp = this_cpu_ptr(&radix_tree_preloads);
211 rtp->nodes = ret->private_data;
212 ret->private_data = NULL;
216 * Update the allocation stack trace as this is more useful
219 kmemleak_update_trace(ret);
222 ret = kmem_cache_alloc(radix_tree_node_cachep,
223 gfp_mask | __GFP_ACCOUNT);
225 BUG_ON(radix_tree_is_indirect_ptr(ret));
229 static void radix_tree_node_rcu_free(struct rcu_head *head)
231 struct radix_tree_node *node =
232 container_of(head, struct radix_tree_node, rcu_head);
236 * must only free zeroed nodes into the slab. radix_tree_shrink
237 * can leave us with a non-NULL entry in the first slot, so clear
238 * that here to make sure.
240 for (i = 0; i < RADIX_TREE_MAX_TAGS; i++)
241 tag_clear(node, i, 0);
243 node->slots[0] = NULL;
246 kmem_cache_free(radix_tree_node_cachep, node);
250 radix_tree_node_free(struct radix_tree_node *node)
252 call_rcu(&node->rcu_head, radix_tree_node_rcu_free);
256 * Load up this CPU's radix_tree_node buffer with sufficient objects to
257 * ensure that the addition of a single element in the tree cannot fail. On
258 * success, return zero, with preemption disabled. On error, return -ENOMEM
259 * with preemption not disabled.
261 * To make use of this facility, the radix tree must be initialised without
262 * __GFP_DIRECT_RECLAIM being passed to INIT_RADIX_TREE().
264 static int __radix_tree_preload(gfp_t gfp_mask)
266 struct radix_tree_preload *rtp;
267 struct radix_tree_node *node;
271 rtp = this_cpu_ptr(&radix_tree_preloads);
272 while (rtp->nr < RADIX_TREE_PRELOAD_SIZE) {
274 node = kmem_cache_alloc(radix_tree_node_cachep, gfp_mask);
278 rtp = this_cpu_ptr(&radix_tree_preloads);
279 if (rtp->nr < RADIX_TREE_PRELOAD_SIZE) {
280 node->private_data = rtp->nodes;
284 kmem_cache_free(radix_tree_node_cachep, node);
293 * Load up this CPU's radix_tree_node buffer with sufficient objects to
294 * ensure that the addition of a single element in the tree cannot fail. On
295 * success, return zero, with preemption disabled. On error, return -ENOMEM
296 * with preemption not disabled.
298 * To make use of this facility, the radix tree must be initialised without
299 * __GFP_DIRECT_RECLAIM being passed to INIT_RADIX_TREE().
301 int radix_tree_preload(gfp_t gfp_mask)
303 /* Warn on non-sensical use... */
304 WARN_ON_ONCE(!gfpflags_allow_blocking(gfp_mask));
305 return __radix_tree_preload(gfp_mask);
307 EXPORT_SYMBOL(radix_tree_preload);
310 * The same as above function, except we don't guarantee preloading happens.
311 * We do it, if we decide it helps. On success, return zero with preemption
312 * disabled. On error, return -ENOMEM with preemption not disabled.
314 int radix_tree_maybe_preload(gfp_t gfp_mask)
316 if (gfpflags_allow_blocking(gfp_mask))
317 return __radix_tree_preload(gfp_mask);
318 /* Preloading doesn't help anything with this gfp mask, skip it */
322 EXPORT_SYMBOL(radix_tree_maybe_preload);
325 * Return the maximum key which can be store into a
326 * radix tree with height HEIGHT.
328 static inline unsigned long radix_tree_maxindex(unsigned int height)
330 return height_to_maxindex[height];
334 * Extend a radix tree so it can store key @index.
336 static int radix_tree_extend(struct radix_tree_root *root, unsigned long index)
338 struct radix_tree_node *node;
339 struct radix_tree_node *slot;
343 /* Figure out what the height should be. */
344 height = root->height + 1;
345 while (index > radix_tree_maxindex(height))
348 if (root->rnode == NULL) {
349 root->height = height;
354 unsigned int newheight;
355 if (!(node = radix_tree_node_alloc(root)))
358 /* Propagate the aggregated tag info into the new root */
359 for (tag = 0; tag < RADIX_TREE_MAX_TAGS; tag++) {
360 if (root_tag_get(root, tag))
361 tag_set(node, tag, 0);
364 /* Increase the height. */
365 newheight = root->height+1;
366 BUG_ON(newheight & ~RADIX_TREE_HEIGHT_MASK);
367 node->path = newheight;
372 slot = indirect_to_ptr(slot);
375 node->slots[0] = slot;
376 node = ptr_to_indirect(node);
377 rcu_assign_pointer(root->rnode, node);
378 root->height = newheight;
379 } while (height > root->height);
385 * __radix_tree_create - create a slot in a radix tree
386 * @root: radix tree root
388 * @nodep: returns node
389 * @slotp: returns slot
391 * Create, if necessary, and return the node and slot for an item
392 * at position @index in the radix tree @root.
394 * Until there is more than one item in the tree, no nodes are
395 * allocated and @root->rnode is used as a direct slot instead of
396 * pointing to a node, in which case *@nodep will be NULL.
398 * Returns -ENOMEM, or 0 for success.
400 int __radix_tree_create(struct radix_tree_root *root, unsigned long index,
401 struct radix_tree_node **nodep, void ***slotp)
403 struct radix_tree_node *node = NULL, *slot;
404 unsigned int height, shift, offset;
407 /* Make sure the tree is high enough. */
408 if (index > radix_tree_maxindex(root->height)) {
409 error = radix_tree_extend(root, index);
414 slot = indirect_to_ptr(root->rnode);
416 height = root->height;
417 shift = (height-1) * RADIX_TREE_MAP_SHIFT;
419 offset = 0; /* uninitialised var warning */
422 /* Have to add a child node. */
423 if (!(slot = radix_tree_node_alloc(root)))
428 rcu_assign_pointer(node->slots[offset], slot);
430 slot->path |= offset << RADIX_TREE_HEIGHT_SHIFT;
432 rcu_assign_pointer(root->rnode, ptr_to_indirect(slot));
435 /* Go a level down */
436 offset = (index >> shift) & RADIX_TREE_MAP_MASK;
438 slot = node->slots[offset];
439 shift -= RADIX_TREE_MAP_SHIFT;
446 *slotp = node ? node->slots + offset : (void **)&root->rnode;
451 * radix_tree_insert - insert into a radix tree
452 * @root: radix tree root
454 * @item: item to insert
456 * Insert an item into the radix tree at position @index.
458 int radix_tree_insert(struct radix_tree_root *root,
459 unsigned long index, void *item)
461 struct radix_tree_node *node;
465 BUG_ON(radix_tree_is_indirect_ptr(item));
467 error = __radix_tree_create(root, index, &node, &slot);
472 rcu_assign_pointer(*slot, item);
476 BUG_ON(tag_get(node, 0, index & RADIX_TREE_MAP_MASK));
477 BUG_ON(tag_get(node, 1, index & RADIX_TREE_MAP_MASK));
479 BUG_ON(root_tag_get(root, 0));
480 BUG_ON(root_tag_get(root, 1));
485 EXPORT_SYMBOL(radix_tree_insert);
488 * __radix_tree_lookup - lookup an item in a radix tree
489 * @root: radix tree root
491 * @nodep: returns node
492 * @slotp: returns slot
494 * Lookup and return the item at position @index in the radix
497 * Until there is more than one item in the tree, no nodes are
498 * allocated and @root->rnode is used as a direct slot instead of
499 * pointing to a node, in which case *@nodep will be NULL.
501 void *__radix_tree_lookup(struct radix_tree_root *root, unsigned long index,
502 struct radix_tree_node **nodep, void ***slotp)
504 struct radix_tree_node *node, *parent;
505 unsigned int height, shift;
508 node = rcu_dereference_raw(root->rnode);
512 if (!radix_tree_is_indirect_ptr(node)) {
519 *slotp = (void **)&root->rnode;
522 node = indirect_to_ptr(node);
524 height = node->path & RADIX_TREE_HEIGHT_MASK;
525 if (index > radix_tree_maxindex(height))
528 shift = (height-1) * RADIX_TREE_MAP_SHIFT;
532 slot = node->slots + ((index >> shift) & RADIX_TREE_MAP_MASK);
533 node = rcu_dereference_raw(*slot);
537 shift -= RADIX_TREE_MAP_SHIFT;
539 } while (height > 0);
549 * radix_tree_lookup_slot - lookup a slot in a radix tree
550 * @root: radix tree root
553 * Returns: the slot corresponding to the position @index in the
554 * radix tree @root. This is useful for update-if-exists operations.
556 * This function can be called under rcu_read_lock iff the slot is not
557 * modified by radix_tree_replace_slot, otherwise it must be called
558 * exclusive from other writers. Any dereference of the slot must be done
559 * using radix_tree_deref_slot.
561 void **radix_tree_lookup_slot(struct radix_tree_root *root, unsigned long index)
565 if (!__radix_tree_lookup(root, index, NULL, &slot))
569 EXPORT_SYMBOL(radix_tree_lookup_slot);
572 * radix_tree_lookup - perform lookup operation on a radix tree
573 * @root: radix tree root
576 * Lookup the item at the position @index in the radix tree @root.
578 * This function can be called under rcu_read_lock, however the caller
579 * must manage lifetimes of leaf nodes (eg. RCU may also be used to free
580 * them safely). No RCU barriers are required to access or modify the
581 * returned item, however.
583 void *radix_tree_lookup(struct radix_tree_root *root, unsigned long index)
585 return __radix_tree_lookup(root, index, NULL, NULL);
587 EXPORT_SYMBOL(radix_tree_lookup);
590 * radix_tree_tag_set - set a tag on a radix tree node
591 * @root: radix tree root
595 * Set the search tag (which must be < RADIX_TREE_MAX_TAGS)
596 * corresponding to @index in the radix tree. From
597 * the root all the way down to the leaf node.
599 * Returns the address of the tagged item. Setting a tag on a not-present
602 void *radix_tree_tag_set(struct radix_tree_root *root,
603 unsigned long index, unsigned int tag)
605 unsigned int height, shift;
606 struct radix_tree_node *slot;
608 height = root->height;
609 BUG_ON(index > radix_tree_maxindex(height));
611 slot = indirect_to_ptr(root->rnode);
612 shift = (height - 1) * RADIX_TREE_MAP_SHIFT;
617 offset = (index >> shift) & RADIX_TREE_MAP_MASK;
618 if (!tag_get(slot, tag, offset))
619 tag_set(slot, tag, offset);
620 slot = slot->slots[offset];
621 BUG_ON(slot == NULL);
622 shift -= RADIX_TREE_MAP_SHIFT;
626 /* set the root's tag bit */
627 if (slot && !root_tag_get(root, tag))
628 root_tag_set(root, tag);
632 EXPORT_SYMBOL(radix_tree_tag_set);
635 * radix_tree_tag_clear - clear a tag on a radix tree node
636 * @root: radix tree root
640 * Clear the search tag (which must be < RADIX_TREE_MAX_TAGS)
641 * corresponding to @index in the radix tree. If
642 * this causes the leaf node to have no tags set then clear the tag in the
643 * next-to-leaf node, etc.
645 * Returns the address of the tagged item on success, else NULL. ie:
646 * has the same return value and semantics as radix_tree_lookup().
648 void *radix_tree_tag_clear(struct radix_tree_root *root,
649 unsigned long index, unsigned int tag)
651 struct radix_tree_node *node = NULL;
652 struct radix_tree_node *slot = NULL;
653 unsigned int height, shift;
654 int uninitialized_var(offset);
656 height = root->height;
657 if (index > radix_tree_maxindex(height))
660 shift = height * RADIX_TREE_MAP_SHIFT;
661 slot = indirect_to_ptr(root->rnode);
667 shift -= RADIX_TREE_MAP_SHIFT;
668 offset = (index >> shift) & RADIX_TREE_MAP_MASK;
670 slot = slot->slots[offset];
677 if (!tag_get(node, tag, offset))
679 tag_clear(node, tag, offset);
680 if (any_tag_set(node, tag))
683 index >>= RADIX_TREE_MAP_SHIFT;
684 offset = index & RADIX_TREE_MAP_MASK;
688 /* clear the root's tag bit */
689 if (root_tag_get(root, tag))
690 root_tag_clear(root, tag);
695 EXPORT_SYMBOL(radix_tree_tag_clear);
698 * radix_tree_tag_get - get a tag on a radix tree node
699 * @root: radix tree root
701 * @tag: tag index (< RADIX_TREE_MAX_TAGS)
705 * 0: tag not present or not set
708 * Note that the return value of this function may not be relied on, even if
709 * the RCU lock is held, unless tag modification and node deletion are excluded
712 int radix_tree_tag_get(struct radix_tree_root *root,
713 unsigned long index, unsigned int tag)
715 unsigned int height, shift;
716 struct radix_tree_node *node;
718 /* check the root's tag bit */
719 if (!root_tag_get(root, tag))
722 node = rcu_dereference_raw(root->rnode);
726 if (!radix_tree_is_indirect_ptr(node))
728 node = indirect_to_ptr(node);
730 height = node->path & RADIX_TREE_HEIGHT_MASK;
731 if (index > radix_tree_maxindex(height))
734 shift = (height - 1) * RADIX_TREE_MAP_SHIFT;
742 offset = (index >> shift) & RADIX_TREE_MAP_MASK;
743 if (!tag_get(node, tag, offset))
747 node = rcu_dereference_raw(node->slots[offset]);
748 shift -= RADIX_TREE_MAP_SHIFT;
752 EXPORT_SYMBOL(radix_tree_tag_get);
755 * radix_tree_next_chunk - find next chunk of slots for iteration
757 * @root: radix tree root
758 * @iter: iterator state
759 * @flags: RADIX_TREE_ITER_* flags and tag index
760 * Returns: pointer to chunk first slot, or NULL if iteration is over
762 void **radix_tree_next_chunk(struct radix_tree_root *root,
763 struct radix_tree_iter *iter, unsigned flags)
765 unsigned shift, tag = flags & RADIX_TREE_ITER_TAG_MASK;
766 struct radix_tree_node *rnode, *node;
767 unsigned long index, offset, height;
769 if ((flags & RADIX_TREE_ITER_TAGGED) && !root_tag_get(root, tag))
773 * Catch next_index overflow after ~0UL. iter->index never overflows
774 * during iterating; it can be zero only at the beginning.
775 * And we cannot overflow iter->next_index in a single step,
776 * because RADIX_TREE_MAP_SHIFT < BITS_PER_LONG.
778 * This condition also used by radix_tree_next_slot() to stop
779 * contiguous iterating, and forbid swithing to the next chunk.
781 index = iter->next_index;
782 if (!index && iter->index)
785 rnode = rcu_dereference_raw(root->rnode);
786 if (radix_tree_is_indirect_ptr(rnode)) {
787 rnode = indirect_to_ptr(rnode);
788 } else if (rnode && !index) {
789 /* Single-slot tree */
791 iter->next_index = 1;
793 return (void **)&root->rnode;
798 height = rnode->path & RADIX_TREE_HEIGHT_MASK;
799 shift = (height - 1) * RADIX_TREE_MAP_SHIFT;
800 offset = index >> shift;
802 /* Index outside of the tree */
803 if (offset >= RADIX_TREE_MAP_SIZE)
808 if ((flags & RADIX_TREE_ITER_TAGGED) ?
809 !test_bit(offset, node->tags[tag]) :
810 !node->slots[offset]) {
812 if (flags & RADIX_TREE_ITER_CONTIG)
815 if (flags & RADIX_TREE_ITER_TAGGED)
816 offset = radix_tree_find_next_bit(
821 while (++offset < RADIX_TREE_MAP_SIZE) {
822 if (node->slots[offset])
825 index &= ~((RADIX_TREE_MAP_SIZE << shift) - 1);
826 index += offset << shift;
827 /* Overflow after ~0UL */
830 if (offset == RADIX_TREE_MAP_SIZE)
834 /* This is leaf-node */
838 node = rcu_dereference_raw(node->slots[offset]);
841 shift -= RADIX_TREE_MAP_SHIFT;
842 offset = (index >> shift) & RADIX_TREE_MAP_MASK;
845 /* Update the iterator state */
847 iter->next_index = (index | RADIX_TREE_MAP_MASK) + 1;
849 /* Construct iter->tags bit-mask from node->tags[tag] array */
850 if (flags & RADIX_TREE_ITER_TAGGED) {
851 unsigned tag_long, tag_bit;
853 tag_long = offset / BITS_PER_LONG;
854 tag_bit = offset % BITS_PER_LONG;
855 iter->tags = node->tags[tag][tag_long] >> tag_bit;
856 /* This never happens if RADIX_TREE_TAG_LONGS == 1 */
857 if (tag_long < RADIX_TREE_TAG_LONGS - 1) {
858 /* Pick tags from next element */
860 iter->tags |= node->tags[tag][tag_long + 1] <<
861 (BITS_PER_LONG - tag_bit);
862 /* Clip chunk size, here only BITS_PER_LONG tags */
863 iter->next_index = index + BITS_PER_LONG;
867 return node->slots + offset;
869 EXPORT_SYMBOL(radix_tree_next_chunk);
872 * radix_tree_range_tag_if_tagged - for each item in given range set given
873 * tag if item has another tag set
874 * @root: radix tree root
875 * @first_indexp: pointer to a starting index of a range to scan
876 * @last_index: last index of a range to scan
877 * @nr_to_tag: maximum number items to tag
878 * @iftag: tag index to test
879 * @settag: tag index to set if tested tag is set
881 * This function scans range of radix tree from first_index to last_index
882 * (inclusive). For each item in the range if iftag is set, the function sets
883 * also settag. The function stops either after tagging nr_to_tag items or
884 * after reaching last_index.
886 * The tags must be set from the leaf level only and propagated back up the
887 * path to the root. We must do this so that we resolve the full path before
888 * setting any tags on intermediate nodes. If we set tags as we descend, then
889 * we can get to the leaf node and find that the index that has the iftag
890 * set is outside the range we are scanning. This reults in dangling tags and
891 * can lead to problems with later tag operations (e.g. livelocks on lookups).
893 * The function returns number of leaves where the tag was set and sets
894 * *first_indexp to the first unscanned index.
895 * WARNING! *first_indexp can wrap if last_index is ULONG_MAX. Caller must
896 * be prepared to handle that.
898 unsigned long radix_tree_range_tag_if_tagged(struct radix_tree_root *root,
899 unsigned long *first_indexp, unsigned long last_index,
900 unsigned long nr_to_tag,
901 unsigned int iftag, unsigned int settag)
903 unsigned int height = root->height;
904 struct radix_tree_node *node = NULL;
905 struct radix_tree_node *slot;
907 unsigned long tagged = 0;
908 unsigned long index = *first_indexp;
910 last_index = min(last_index, radix_tree_maxindex(height));
911 if (index > last_index)
915 if (!root_tag_get(root, iftag)) {
916 *first_indexp = last_index + 1;
920 *first_indexp = last_index + 1;
921 root_tag_set(root, settag);
925 shift = (height - 1) * RADIX_TREE_MAP_SHIFT;
926 slot = indirect_to_ptr(root->rnode);
929 unsigned long upindex;
932 offset = (index >> shift) & RADIX_TREE_MAP_MASK;
933 if (!slot->slots[offset])
935 if (!tag_get(slot, iftag, offset))
938 /* Go down one level */
939 shift -= RADIX_TREE_MAP_SHIFT;
941 slot = slot->slots[offset];
947 tag_set(slot, settag, offset);
949 /* walk back up the path tagging interior nodes */
952 upindex >>= RADIX_TREE_MAP_SHIFT;
953 offset = upindex & RADIX_TREE_MAP_MASK;
955 /* stop if we find a node with the tag already set */
956 if (tag_get(node, settag, offset))
958 tag_set(node, settag, offset);
963 * Small optimization: now clear that node pointer.
964 * Since all of this slot's ancestors now have the tag set
965 * from setting it above, we have no further need to walk
966 * back up the tree setting tags, until we update slot to
967 * point to another radix_tree_node.
972 /* Go to next item at level determined by 'shift' */
973 index = ((index >> shift) + 1) << shift;
974 /* Overflow can happen when last_index is ~0UL... */
975 if (index > last_index || !index)
977 if (tagged >= nr_to_tag)
979 while (((index >> shift) & RADIX_TREE_MAP_MASK) == 0) {
981 * We've fully scanned this node. Go up. Because
982 * last_index is guaranteed to be in the tree, what
983 * we do below cannot wander astray.
986 shift += RADIX_TREE_MAP_SHIFT;
990 * We need not to tag the root tag if there is no tag which is set with
991 * settag within the range from *first_indexp to last_index.
994 root_tag_set(root, settag);
995 *first_indexp = index;
999 EXPORT_SYMBOL(radix_tree_range_tag_if_tagged);
1002 * radix_tree_gang_lookup - perform multiple lookup on a radix tree
1003 * @root: radix tree root
1004 * @results: where the results of the lookup are placed
1005 * @first_index: start the lookup from this key
1006 * @max_items: place up to this many items at *results
1008 * Performs an index-ascending scan of the tree for present items. Places
1009 * them at *@results and returns the number of items which were placed at
1012 * The implementation is naive.
1014 * Like radix_tree_lookup, radix_tree_gang_lookup may be called under
1015 * rcu_read_lock. In this case, rather than the returned results being
1016 * an atomic snapshot of the tree at a single point in time, the semantics
1017 * of an RCU protected gang lookup are as though multiple radix_tree_lookups
1018 * have been issued in individual locks, and results stored in 'results'.
1021 radix_tree_gang_lookup(struct radix_tree_root *root, void **results,
1022 unsigned long first_index, unsigned int max_items)
1024 struct radix_tree_iter iter;
1026 unsigned int ret = 0;
1028 if (unlikely(!max_items))
1031 radix_tree_for_each_slot(slot, root, &iter, first_index) {
1032 results[ret] = rcu_dereference_raw(*slot);
1035 if (radix_tree_is_indirect_ptr(results[ret])) {
1036 slot = radix_tree_iter_retry(&iter);
1039 if (++ret == max_items)
1045 EXPORT_SYMBOL(radix_tree_gang_lookup);
1048 * radix_tree_gang_lookup_slot - perform multiple slot lookup on radix tree
1049 * @root: radix tree root
1050 * @results: where the results of the lookup are placed
1051 * @indices: where their indices should be placed (but usually NULL)
1052 * @first_index: start the lookup from this key
1053 * @max_items: place up to this many items at *results
1055 * Performs an index-ascending scan of the tree for present items. Places
1056 * their slots at *@results and returns the number of items which were
1057 * placed at *@results.
1059 * The implementation is naive.
1061 * Like radix_tree_gang_lookup as far as RCU and locking goes. Slots must
1062 * be dereferenced with radix_tree_deref_slot, and if using only RCU
1063 * protection, radix_tree_deref_slot may fail requiring a retry.
1066 radix_tree_gang_lookup_slot(struct radix_tree_root *root,
1067 void ***results, unsigned long *indices,
1068 unsigned long first_index, unsigned int max_items)
1070 struct radix_tree_iter iter;
1072 unsigned int ret = 0;
1074 if (unlikely(!max_items))
1077 radix_tree_for_each_slot(slot, root, &iter, first_index) {
1078 results[ret] = slot;
1080 indices[ret] = iter.index;
1081 if (++ret == max_items)
1087 EXPORT_SYMBOL(radix_tree_gang_lookup_slot);
1090 * radix_tree_gang_lookup_tag - perform multiple lookup on a radix tree
1092 * @root: radix tree root
1093 * @results: where the results of the lookup are placed
1094 * @first_index: start the lookup from this key
1095 * @max_items: place up to this many items at *results
1096 * @tag: the tag index (< RADIX_TREE_MAX_TAGS)
1098 * Performs an index-ascending scan of the tree for present items which
1099 * have the tag indexed by @tag set. Places the items at *@results and
1100 * returns the number of items which were placed at *@results.
1103 radix_tree_gang_lookup_tag(struct radix_tree_root *root, void **results,
1104 unsigned long first_index, unsigned int max_items,
1107 struct radix_tree_iter iter;
1109 unsigned int ret = 0;
1111 if (unlikely(!max_items))
1114 radix_tree_for_each_tagged(slot, root, &iter, first_index, tag) {
1115 results[ret] = rcu_dereference_raw(*slot);
1118 if (radix_tree_is_indirect_ptr(results[ret])) {
1119 slot = radix_tree_iter_retry(&iter);
1122 if (++ret == max_items)
1128 EXPORT_SYMBOL(radix_tree_gang_lookup_tag);
1131 * radix_tree_gang_lookup_tag_slot - perform multiple slot lookup on a
1132 * radix tree based on a tag
1133 * @root: radix tree root
1134 * @results: where the results of the lookup are placed
1135 * @first_index: start the lookup from this key
1136 * @max_items: place up to this many items at *results
1137 * @tag: the tag index (< RADIX_TREE_MAX_TAGS)
1139 * Performs an index-ascending scan of the tree for present items which
1140 * have the tag indexed by @tag set. Places the slots at *@results and
1141 * returns the number of slots which were placed at *@results.
1144 radix_tree_gang_lookup_tag_slot(struct radix_tree_root *root, void ***results,
1145 unsigned long first_index, unsigned int max_items,
1148 struct radix_tree_iter iter;
1150 unsigned int ret = 0;
1152 if (unlikely(!max_items))
1155 radix_tree_for_each_tagged(slot, root, &iter, first_index, tag) {
1156 results[ret] = slot;
1157 if (++ret == max_items)
1163 EXPORT_SYMBOL(radix_tree_gang_lookup_tag_slot);
1165 #if defined(CONFIG_SHMEM) && defined(CONFIG_SWAP)
1166 #include <linux/sched.h> /* for cond_resched() */
1169 * This linear search is at present only useful to shmem_unuse_inode().
1171 static unsigned long __locate(struct radix_tree_node *slot, void *item,
1172 unsigned long index, unsigned long *found_index)
1174 unsigned int shift, height;
1177 height = slot->path & RADIX_TREE_HEIGHT_MASK;
1178 shift = (height-1) * RADIX_TREE_MAP_SHIFT;
1180 for ( ; height > 1; height--) {
1181 i = (index >> shift) & RADIX_TREE_MAP_MASK;
1183 if (slot->slots[i] != NULL)
1185 index &= ~((1UL << shift) - 1);
1186 index += 1UL << shift;
1188 goto out; /* 32-bit wraparound */
1190 if (i == RADIX_TREE_MAP_SIZE)
1194 shift -= RADIX_TREE_MAP_SHIFT;
1195 slot = rcu_dereference_raw(slot->slots[i]);
1200 /* Bottom level: check items */
1201 for (i = 0; i < RADIX_TREE_MAP_SIZE; i++) {
1202 if (slot->slots[i] == item) {
1203 *found_index = index + i;
1208 index += RADIX_TREE_MAP_SIZE;
1214 * radix_tree_locate_item - search through radix tree for item
1215 * @root: radix tree root
1216 * @item: item to be found
1218 * Returns index where item was found, or -1 if not found.
1219 * Caller must hold no lock (since this time-consuming function needs
1220 * to be preemptible), and must check afterwards if item is still there.
1222 unsigned long radix_tree_locate_item(struct radix_tree_root *root, void *item)
1224 struct radix_tree_node *node;
1225 unsigned long max_index;
1226 unsigned long cur_index = 0;
1227 unsigned long found_index = -1;
1231 node = rcu_dereference_raw(root->rnode);
1232 if (!radix_tree_is_indirect_ptr(node)) {
1239 node = indirect_to_ptr(node);
1240 max_index = radix_tree_maxindex(node->path &
1241 RADIX_TREE_HEIGHT_MASK);
1242 if (cur_index > max_index) {
1247 cur_index = __locate(node, item, cur_index, &found_index);
1250 } while (cur_index != 0 && cur_index <= max_index);
1255 unsigned long radix_tree_locate_item(struct radix_tree_root *root, void *item)
1259 #endif /* CONFIG_SHMEM && CONFIG_SWAP */
1262 * radix_tree_shrink - shrink height of a radix tree to minimal
1263 * @root radix tree root
1265 static inline void radix_tree_shrink(struct radix_tree_root *root)
1267 /* try to shrink tree height */
1268 while (root->height > 0) {
1269 struct radix_tree_node *to_free = root->rnode;
1270 struct radix_tree_node *slot;
1272 BUG_ON(!radix_tree_is_indirect_ptr(to_free));
1273 to_free = indirect_to_ptr(to_free);
1276 * The candidate node has more than one child, or its child
1277 * is not at the leftmost slot, we cannot shrink.
1279 if (to_free->count != 1)
1281 if (!to_free->slots[0])
1285 * We don't need rcu_assign_pointer(), since we are simply
1286 * moving the node from one part of the tree to another: if it
1287 * was safe to dereference the old pointer to it
1288 * (to_free->slots[0]), it will be safe to dereference the new
1289 * one (root->rnode) as far as dependent read barriers go.
1291 slot = to_free->slots[0];
1292 if (root->height > 1) {
1293 slot->parent = NULL;
1294 slot = ptr_to_indirect(slot);
1300 * We have a dilemma here. The node's slot[0] must not be
1301 * NULLed in case there are concurrent lookups expecting to
1302 * find the item. However if this was a bottom-level node,
1303 * then it may be subject to the slot pointer being visible
1304 * to callers dereferencing it. If item corresponding to
1305 * slot[0] is subsequently deleted, these callers would expect
1306 * their slot to become empty sooner or later.
1308 * For example, lockless pagecache will look up a slot, deref
1309 * the page pointer, and if the page is 0 refcount it means it
1310 * was concurrently deleted from pagecache so try the deref
1311 * again. Fortunately there is already a requirement for logic
1312 * to retry the entire slot lookup -- the indirect pointer
1313 * problem (replacing direct root node with an indirect pointer
1314 * also results in a stale slot). So tag the slot as indirect
1315 * to force callers to retry.
1317 if (root->height == 0)
1318 *((unsigned long *)&to_free->slots[0]) |=
1319 RADIX_TREE_INDIRECT_PTR;
1321 radix_tree_node_free(to_free);
1326 * __radix_tree_delete_node - try to free node after clearing a slot
1327 * @root: radix tree root
1328 * @node: node containing @index
1330 * After clearing the slot at @index in @node from radix tree
1331 * rooted at @root, call this function to attempt freeing the
1332 * node and shrinking the tree.
1334 * Returns %true if @node was freed, %false otherwise.
1336 bool __radix_tree_delete_node(struct radix_tree_root *root,
1337 struct radix_tree_node *node)
1339 bool deleted = false;
1342 struct radix_tree_node *parent;
1345 if (node == indirect_to_ptr(root->rnode)) {
1346 radix_tree_shrink(root);
1347 if (root->height == 0)
1353 parent = node->parent;
1355 unsigned int offset;
1357 offset = node->path >> RADIX_TREE_HEIGHT_SHIFT;
1358 parent->slots[offset] = NULL;
1361 root_tag_clear_all(root);
1366 radix_tree_node_free(node);
1376 * radix_tree_delete_item - delete an item from a radix tree
1377 * @root: radix tree root
1379 * @item: expected item
1381 * Remove @item at @index from the radix tree rooted at @root.
1383 * Returns the address of the deleted item, or NULL if it was not present
1384 * or the entry at the given @index was not @item.
1386 void *radix_tree_delete_item(struct radix_tree_root *root,
1387 unsigned long index, void *item)
1389 struct radix_tree_node *node;
1390 unsigned int offset;
1395 entry = __radix_tree_lookup(root, index, &node, &slot);
1399 if (item && entry != item)
1403 root_tag_clear_all(root);
1408 offset = index & RADIX_TREE_MAP_MASK;
1411 * Clear all tags associated with the item to be deleted.
1412 * This way of doing it would be inefficient, but seldom is any set.
1414 for (tag = 0; tag < RADIX_TREE_MAX_TAGS; tag++) {
1415 if (tag_get(node, tag, offset))
1416 radix_tree_tag_clear(root, index, tag);
1419 node->slots[offset] = NULL;
1422 __radix_tree_delete_node(root, node);
1426 EXPORT_SYMBOL(radix_tree_delete_item);
1429 * radix_tree_delete - delete an item from a radix tree
1430 * @root: radix tree root
1433 * Remove the item at @index from the radix tree rooted at @root.
1435 * Returns the address of the deleted item, or NULL if it was not present.
1437 void *radix_tree_delete(struct radix_tree_root *root, unsigned long index)
1439 return radix_tree_delete_item(root, index, NULL);
1441 EXPORT_SYMBOL(radix_tree_delete);
1444 * radix_tree_tagged - test whether any items in the tree are tagged
1445 * @root: radix tree root
1448 int radix_tree_tagged(struct radix_tree_root *root, unsigned int tag)
1450 return root_tag_get(root, tag);
1452 EXPORT_SYMBOL(radix_tree_tagged);
1455 radix_tree_node_ctor(void *arg)
1457 struct radix_tree_node *node = arg;
1459 memset(node, 0, sizeof(*node));
1460 INIT_LIST_HEAD(&node->private_list);
1463 static __init unsigned long __maxindex(unsigned int height)
1465 unsigned int width = height * RADIX_TREE_MAP_SHIFT;
1466 int shift = RADIX_TREE_INDEX_BITS - width;
1470 if (shift >= BITS_PER_LONG)
1472 return ~0UL >> shift;
1475 static __init void radix_tree_init_maxindex(void)
1479 for (i = 0; i < ARRAY_SIZE(height_to_maxindex); i++)
1480 height_to_maxindex[i] = __maxindex(i);
1483 static int radix_tree_callback(struct notifier_block *nfb,
1484 unsigned long action,
1487 int cpu = (long)hcpu;
1488 struct radix_tree_preload *rtp;
1489 struct radix_tree_node *node;
1491 /* Free per-cpu pool of perloaded nodes */
1492 if (action == CPU_DEAD || action == CPU_DEAD_FROZEN) {
1493 rtp = &per_cpu(radix_tree_preloads, cpu);
1496 rtp->nodes = node->private_data;
1497 kmem_cache_free(radix_tree_node_cachep, node);
1504 void __init radix_tree_init(void)
1506 radix_tree_node_cachep = kmem_cache_create("radix_tree_node",
1507 sizeof(struct radix_tree_node), 0,
1508 SLAB_PANIC | SLAB_RECLAIM_ACCOUNT,
1509 radix_tree_node_ctor);
1510 radix_tree_init_maxindex();
1511 hotcpu_notifier(radix_tree_callback, 0);