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 #define RADIX_TREE_RETRY ptr_to_indirect(NULL)
80 #ifdef CONFIG_RADIX_TREE_MULTIORDER
81 /* Sibling slots point directly to another slot in the same node */
82 static inline bool is_sibling_entry(struct radix_tree_node *parent, void *node)
85 return (parent->slots <= ptr) &&
86 (ptr < parent->slots + RADIX_TREE_MAP_SIZE);
89 static inline bool is_sibling_entry(struct radix_tree_node *parent, void *node)
95 static inline unsigned long get_slot_offset(struct radix_tree_node *parent,
98 return slot - parent->slots;
101 static unsigned radix_tree_descend(struct radix_tree_node *parent,
102 struct radix_tree_node **nodep, unsigned offset)
104 void **entry = rcu_dereference_raw(parent->slots[offset]);
106 #ifdef CONFIG_RADIX_TREE_MULTIORDER
107 if (radix_tree_is_indirect_ptr(entry)) {
108 unsigned long siboff = get_slot_offset(parent, entry);
109 if (siboff < RADIX_TREE_MAP_SIZE) {
111 entry = rcu_dereference_raw(parent->slots[offset]);
116 *nodep = (void *)entry;
120 static inline gfp_t root_gfp_mask(struct radix_tree_root *root)
122 return root->gfp_mask & __GFP_BITS_MASK;
125 static inline void tag_set(struct radix_tree_node *node, unsigned int tag,
128 __set_bit(offset, node->tags[tag]);
131 static inline void tag_clear(struct radix_tree_node *node, unsigned int tag,
134 __clear_bit(offset, node->tags[tag]);
137 static inline int tag_get(struct radix_tree_node *node, unsigned int tag,
140 return test_bit(offset, node->tags[tag]);
143 static inline void root_tag_set(struct radix_tree_root *root, unsigned int tag)
145 root->gfp_mask |= (__force gfp_t)(1 << (tag + __GFP_BITS_SHIFT));
148 static inline void root_tag_clear(struct radix_tree_root *root, unsigned int tag)
150 root->gfp_mask &= (__force gfp_t)~(1 << (tag + __GFP_BITS_SHIFT));
153 static inline void root_tag_clear_all(struct radix_tree_root *root)
155 root->gfp_mask &= __GFP_BITS_MASK;
158 static inline int root_tag_get(struct radix_tree_root *root, unsigned int tag)
160 return (__force unsigned)root->gfp_mask & (1 << (tag + __GFP_BITS_SHIFT));
163 static inline unsigned root_tags_get(struct radix_tree_root *root)
165 return (__force unsigned)root->gfp_mask >> __GFP_BITS_SHIFT;
169 * Returns 1 if any slot in the node has this tag set.
170 * Otherwise returns 0.
172 static inline int any_tag_set(struct radix_tree_node *node, unsigned int tag)
175 for (idx = 0; idx < RADIX_TREE_TAG_LONGS; idx++) {
176 if (node->tags[tag][idx])
183 * radix_tree_find_next_bit - find the next set bit in a memory region
185 * @addr: The address to base the search on
186 * @size: The bitmap size in bits
187 * @offset: The bitnumber to start searching at
189 * Unrollable variant of find_next_bit() for constant size arrays.
190 * Tail bits starting from size to roundup(size, BITS_PER_LONG) must be zero.
191 * Returns next bit offset, or size if nothing found.
193 static __always_inline unsigned long
194 radix_tree_find_next_bit(const unsigned long *addr,
195 unsigned long size, unsigned long offset)
197 if (!__builtin_constant_p(size))
198 return find_next_bit(addr, size, offset);
203 addr += offset / BITS_PER_LONG;
204 tmp = *addr >> (offset % BITS_PER_LONG);
206 return __ffs(tmp) + offset;
207 offset = (offset + BITS_PER_LONG) & ~(BITS_PER_LONG - 1);
208 while (offset < size) {
211 return __ffs(tmp) + offset;
212 offset += BITS_PER_LONG;
219 static void dump_node(void *slot, int height, int offset)
221 struct radix_tree_node *node;
228 pr_debug("radix entry %p offset %d\n", slot, offset);
232 node = indirect_to_ptr(slot);
233 pr_debug("radix node: %p offset %d tags %lx %lx %lx path %x count %d parent %p\n",
234 slot, offset, node->tags[0][0], node->tags[1][0],
235 node->tags[2][0], node->path, node->count, node->parent);
237 for (i = 0; i < RADIX_TREE_MAP_SIZE; i++)
238 dump_node(node->slots[i], height - 1, i);
242 static void radix_tree_dump(struct radix_tree_root *root)
244 pr_debug("radix root: %p height %d rnode %p tags %x\n",
245 root, root->height, root->rnode,
246 root->gfp_mask >> __GFP_BITS_SHIFT);
247 if (!radix_tree_is_indirect_ptr(root->rnode))
249 dump_node(root->rnode, root->height, 0);
254 * This assumes that the caller has performed appropriate preallocation, and
255 * that the caller has pinned this thread of control to the current CPU.
257 static struct radix_tree_node *
258 radix_tree_node_alloc(struct radix_tree_root *root)
260 struct radix_tree_node *ret = NULL;
261 gfp_t gfp_mask = root_gfp_mask(root);
264 * Preload code isn't irq safe and it doesn't make sence to use
265 * preloading in the interrupt anyway as all the allocations have to
266 * be atomic. So just do normal allocation when in interrupt.
268 if (!gfpflags_allow_blocking(gfp_mask) && !in_interrupt()) {
269 struct radix_tree_preload *rtp;
272 * Even if the caller has preloaded, try to allocate from the
273 * cache first for the new node to get accounted.
275 ret = kmem_cache_alloc(radix_tree_node_cachep,
276 gfp_mask | __GFP_ACCOUNT | __GFP_NOWARN);
281 * Provided the caller has preloaded here, we will always
282 * succeed in getting a node here (and never reach
285 rtp = this_cpu_ptr(&radix_tree_preloads);
288 rtp->nodes = ret->private_data;
289 ret->private_data = NULL;
293 * Update the allocation stack trace as this is more useful
296 kmemleak_update_trace(ret);
299 ret = kmem_cache_alloc(radix_tree_node_cachep,
300 gfp_mask | __GFP_ACCOUNT);
302 BUG_ON(radix_tree_is_indirect_ptr(ret));
306 static void radix_tree_node_rcu_free(struct rcu_head *head)
308 struct radix_tree_node *node =
309 container_of(head, struct radix_tree_node, rcu_head);
313 * must only free zeroed nodes into the slab. radix_tree_shrink
314 * can leave us with a non-NULL entry in the first slot, so clear
315 * that here to make sure.
317 for (i = 0; i < RADIX_TREE_MAX_TAGS; i++)
318 tag_clear(node, i, 0);
320 node->slots[0] = NULL;
323 kmem_cache_free(radix_tree_node_cachep, node);
327 radix_tree_node_free(struct radix_tree_node *node)
329 call_rcu(&node->rcu_head, radix_tree_node_rcu_free);
333 * Load up this CPU's radix_tree_node buffer with sufficient objects to
334 * ensure that the addition of a single element in the tree cannot fail. On
335 * success, return zero, with preemption disabled. On error, return -ENOMEM
336 * with preemption not disabled.
338 * To make use of this facility, the radix tree must be initialised without
339 * __GFP_DIRECT_RECLAIM being passed to INIT_RADIX_TREE().
341 static int __radix_tree_preload(gfp_t gfp_mask)
343 struct radix_tree_preload *rtp;
344 struct radix_tree_node *node;
348 rtp = this_cpu_ptr(&radix_tree_preloads);
349 while (rtp->nr < RADIX_TREE_PRELOAD_SIZE) {
351 node = kmem_cache_alloc(radix_tree_node_cachep, gfp_mask);
355 rtp = this_cpu_ptr(&radix_tree_preloads);
356 if (rtp->nr < RADIX_TREE_PRELOAD_SIZE) {
357 node->private_data = rtp->nodes;
361 kmem_cache_free(radix_tree_node_cachep, node);
370 * Load up this CPU's radix_tree_node buffer with sufficient objects to
371 * ensure that the addition of a single element in the tree cannot fail. On
372 * success, return zero, with preemption disabled. On error, return -ENOMEM
373 * with preemption not disabled.
375 * To make use of this facility, the radix tree must be initialised without
376 * __GFP_DIRECT_RECLAIM being passed to INIT_RADIX_TREE().
378 int radix_tree_preload(gfp_t gfp_mask)
380 /* Warn on non-sensical use... */
381 WARN_ON_ONCE(!gfpflags_allow_blocking(gfp_mask));
382 return __radix_tree_preload(gfp_mask);
384 EXPORT_SYMBOL(radix_tree_preload);
387 * The same as above function, except we don't guarantee preloading happens.
388 * We do it, if we decide it helps. On success, return zero with preemption
389 * disabled. On error, return -ENOMEM with preemption not disabled.
391 int radix_tree_maybe_preload(gfp_t gfp_mask)
393 if (gfpflags_allow_blocking(gfp_mask))
394 return __radix_tree_preload(gfp_mask);
395 /* Preloading doesn't help anything with this gfp mask, skip it */
399 EXPORT_SYMBOL(radix_tree_maybe_preload);
402 * Return the maximum key which can be store into a
403 * radix tree with height HEIGHT.
405 static inline unsigned long radix_tree_maxindex(unsigned int height)
407 return height_to_maxindex[height];
410 static inline unsigned long node_maxindex(struct radix_tree_node *node)
412 return radix_tree_maxindex(node->path & RADIX_TREE_HEIGHT_MASK);
415 static unsigned radix_tree_load_root(struct radix_tree_root *root,
416 struct radix_tree_node **nodep, unsigned long *maxindex)
418 struct radix_tree_node *node = rcu_dereference_raw(root->rnode);
422 if (likely(radix_tree_is_indirect_ptr(node))) {
423 node = indirect_to_ptr(node);
424 *maxindex = node_maxindex(node);
425 return (node->path & RADIX_TREE_HEIGHT_MASK) *
426 RADIX_TREE_MAP_SHIFT;
434 * Extend a radix tree so it can store key @index.
436 static int radix_tree_extend(struct radix_tree_root *root,
439 struct radix_tree_node *node;
440 struct radix_tree_node *slot;
444 /* Figure out what the height should be. */
445 height = root->height + 1;
446 while (index > radix_tree_maxindex(height))
449 if (root->rnode == NULL) {
450 root->height = height;
455 unsigned int newheight;
456 if (!(node = radix_tree_node_alloc(root)))
459 /* Propagate the aggregated tag info into the new root */
460 for (tag = 0; tag < RADIX_TREE_MAX_TAGS; tag++) {
461 if (root_tag_get(root, tag))
462 tag_set(node, tag, 0);
465 /* Increase the height. */
466 newheight = root->height+1;
467 BUG_ON(newheight & ~RADIX_TREE_HEIGHT_MASK);
468 node->path = newheight;
472 if (radix_tree_is_indirect_ptr(slot)) {
473 slot = indirect_to_ptr(slot);
475 slot = ptr_to_indirect(slot);
477 node->slots[0] = slot;
478 node = ptr_to_indirect(node);
479 rcu_assign_pointer(root->rnode, node);
480 root->height = newheight;
481 } while (height > root->height);
483 return height * RADIX_TREE_MAP_SHIFT;
487 * __radix_tree_create - create a slot in a radix tree
488 * @root: radix tree root
490 * @order: index occupies 2^order aligned slots
491 * @nodep: returns node
492 * @slotp: returns slot
494 * Create, if necessary, and return the node and slot for an item
495 * at position @index in the radix tree @root.
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 * Returns -ENOMEM, or 0 for success.
503 int __radix_tree_create(struct radix_tree_root *root, unsigned long index,
504 unsigned order, struct radix_tree_node **nodep,
507 struct radix_tree_node *node = NULL, *slot;
508 unsigned long maxindex;
509 unsigned int height, shift, offset;
510 unsigned long max = index | ((1UL << order) - 1);
512 shift = radix_tree_load_root(root, &slot, &maxindex);
514 /* Make sure the tree is high enough. */
515 if (max > maxindex) {
516 int error = radix_tree_extend(root, max);
521 if (order == shift) {
522 shift += RADIX_TREE_MAP_SHIFT;
527 height = root->height;
529 offset = 0; /* uninitialised var warning */
530 while (shift > order) {
532 /* Have to add a child node. */
533 if (!(slot = radix_tree_node_alloc(root)))
538 rcu_assign_pointer(node->slots[offset],
539 ptr_to_indirect(slot));
541 slot->path |= offset << RADIX_TREE_HEIGHT_SHIFT;
543 rcu_assign_pointer(root->rnode,
544 ptr_to_indirect(slot));
545 } else if (!radix_tree_is_indirect_ptr(slot))
548 /* Go a level down */
550 shift -= RADIX_TREE_MAP_SHIFT;
551 node = indirect_to_ptr(slot);
552 offset = (index >> shift) & RADIX_TREE_MAP_MASK;
553 offset = radix_tree_descend(node, &slot, offset);
556 #ifdef CONFIG_RADIX_TREE_MULTIORDER
557 /* Insert pointers to the canonical entry */
559 int i, n = 1 << (order - shift);
560 offset = offset & ~(n - 1);
561 slot = ptr_to_indirect(&node->slots[offset]);
562 for (i = 0; i < n; i++) {
563 if (node->slots[offset + i])
567 for (i = 1; i < n; i++) {
568 rcu_assign_pointer(node->slots[offset + i], slot);
577 *slotp = node ? node->slots + offset : (void **)&root->rnode;
582 * __radix_tree_insert - insert into a radix tree
583 * @root: radix tree root
585 * @order: key covers the 2^order indices around index
586 * @item: item to insert
588 * Insert an item into the radix tree at position @index.
590 int __radix_tree_insert(struct radix_tree_root *root, unsigned long index,
591 unsigned order, void *item)
593 struct radix_tree_node *node;
597 BUG_ON(radix_tree_is_indirect_ptr(item));
599 error = __radix_tree_create(root, index, order, &node, &slot);
604 rcu_assign_pointer(*slot, item);
607 unsigned offset = get_slot_offset(node, slot);
609 BUG_ON(tag_get(node, 0, offset));
610 BUG_ON(tag_get(node, 1, offset));
611 BUG_ON(tag_get(node, 2, offset));
613 BUG_ON(root_tags_get(root));
618 EXPORT_SYMBOL(__radix_tree_insert);
621 * __radix_tree_lookup - lookup an item in a radix tree
622 * @root: radix tree root
624 * @nodep: returns node
625 * @slotp: returns slot
627 * Lookup and return the item at position @index in the radix
630 * Until there is more than one item in the tree, no nodes are
631 * allocated and @root->rnode is used as a direct slot instead of
632 * pointing to a node, in which case *@nodep will be NULL.
634 void *__radix_tree_lookup(struct radix_tree_root *root, unsigned long index,
635 struct radix_tree_node **nodep, void ***slotp)
637 struct radix_tree_node *node, *parent;
638 unsigned long maxindex;
644 slot = (void **)&root->rnode;
645 shift = radix_tree_load_root(root, &node, &maxindex);
646 if (index > maxindex)
649 while (radix_tree_is_indirect_ptr(node)) {
652 if (node == RADIX_TREE_RETRY)
654 parent = indirect_to_ptr(node);
655 shift -= RADIX_TREE_MAP_SHIFT;
656 offset = (index >> shift) & RADIX_TREE_MAP_MASK;
657 offset = radix_tree_descend(parent, &node, offset);
658 slot = parent->slots + offset;
669 * radix_tree_lookup_slot - lookup a slot in a radix tree
670 * @root: radix tree root
673 * Returns: the slot corresponding to the position @index in the
674 * radix tree @root. This is useful for update-if-exists operations.
676 * This function can be called under rcu_read_lock iff the slot is not
677 * modified by radix_tree_replace_slot, otherwise it must be called
678 * exclusive from other writers. Any dereference of the slot must be done
679 * using radix_tree_deref_slot.
681 void **radix_tree_lookup_slot(struct radix_tree_root *root, unsigned long index)
685 if (!__radix_tree_lookup(root, index, NULL, &slot))
689 EXPORT_SYMBOL(radix_tree_lookup_slot);
692 * radix_tree_lookup - perform lookup operation on a radix tree
693 * @root: radix tree root
696 * Lookup the item at the position @index in the radix tree @root.
698 * This function can be called under rcu_read_lock, however the caller
699 * must manage lifetimes of leaf nodes (eg. RCU may also be used to free
700 * them safely). No RCU barriers are required to access or modify the
701 * returned item, however.
703 void *radix_tree_lookup(struct radix_tree_root *root, unsigned long index)
705 return __radix_tree_lookup(root, index, NULL, NULL);
707 EXPORT_SYMBOL(radix_tree_lookup);
710 * radix_tree_tag_set - set a tag on a radix tree node
711 * @root: radix tree root
715 * Set the search tag (which must be < RADIX_TREE_MAX_TAGS)
716 * corresponding to @index in the radix tree. From
717 * the root all the way down to the leaf node.
719 * Returns the address of the tagged item. Setting a tag on a not-present
722 void *radix_tree_tag_set(struct radix_tree_root *root,
723 unsigned long index, unsigned int tag)
725 struct radix_tree_node *node, *parent;
726 unsigned long maxindex;
729 shift = radix_tree_load_root(root, &node, &maxindex);
730 BUG_ON(index > maxindex);
732 while (radix_tree_is_indirect_ptr(node)) {
735 shift -= RADIX_TREE_MAP_SHIFT;
736 offset = (index >> shift) & RADIX_TREE_MAP_MASK;
738 parent = indirect_to_ptr(node);
739 offset = radix_tree_descend(parent, &node, offset);
742 if (!tag_get(parent, tag, offset))
743 tag_set(parent, tag, offset);
746 /* set the root's tag bit */
747 if (!root_tag_get(root, tag))
748 root_tag_set(root, tag);
752 EXPORT_SYMBOL(radix_tree_tag_set);
755 * radix_tree_tag_clear - clear a tag on a radix tree node
756 * @root: radix tree root
760 * Clear the search tag (which must be < RADIX_TREE_MAX_TAGS)
761 * corresponding to @index in the radix tree. If
762 * this causes the leaf node to have no tags set then clear the tag in the
763 * next-to-leaf node, etc.
765 * Returns the address of the tagged item on success, else NULL. ie:
766 * has the same return value and semantics as radix_tree_lookup().
768 void *radix_tree_tag_clear(struct radix_tree_root *root,
769 unsigned long index, unsigned int tag)
771 struct radix_tree_node *node, *parent;
772 unsigned long maxindex;
774 int uninitialized_var(offset);
776 shift = radix_tree_load_root(root, &node, &maxindex);
777 if (index > maxindex)
782 while (radix_tree_is_indirect_ptr(node)) {
783 shift -= RADIX_TREE_MAP_SHIFT;
784 offset = (index >> shift) & RADIX_TREE_MAP_MASK;
786 parent = indirect_to_ptr(node);
787 offset = radix_tree_descend(parent, &node, offset);
796 if (!tag_get(parent, tag, offset))
798 tag_clear(parent, tag, offset);
799 if (any_tag_set(parent, tag))
802 index >>= RADIX_TREE_MAP_SHIFT;
803 offset = index & RADIX_TREE_MAP_MASK;
804 parent = parent->parent;
807 /* clear the root's tag bit */
808 if (root_tag_get(root, tag))
809 root_tag_clear(root, tag);
814 EXPORT_SYMBOL(radix_tree_tag_clear);
817 * radix_tree_tag_get - get a tag on a radix tree node
818 * @root: radix tree root
820 * @tag: tag index (< RADIX_TREE_MAX_TAGS)
824 * 0: tag not present or not set
827 * Note that the return value of this function may not be relied on, even if
828 * the RCU lock is held, unless tag modification and node deletion are excluded
831 int radix_tree_tag_get(struct radix_tree_root *root,
832 unsigned long index, unsigned int tag)
834 struct radix_tree_node *node, *parent;
835 unsigned long maxindex;
838 if (!root_tag_get(root, tag))
841 shift = radix_tree_load_root(root, &node, &maxindex);
842 if (index > maxindex)
847 while (radix_tree_is_indirect_ptr(node)) {
850 shift -= RADIX_TREE_MAP_SHIFT;
851 offset = (index >> shift) & RADIX_TREE_MAP_MASK;
853 parent = indirect_to_ptr(node);
854 offset = radix_tree_descend(parent, &node, offset);
858 if (!tag_get(parent, tag, offset))
860 if (node == RADIX_TREE_RETRY)
866 EXPORT_SYMBOL(radix_tree_tag_get);
868 static inline void __set_iter_shift(struct radix_tree_iter *iter,
871 #ifdef CONFIG_RADIX_TREE_MULTIORDER
877 * radix_tree_next_chunk - find next chunk of slots for iteration
879 * @root: radix tree root
880 * @iter: iterator state
881 * @flags: RADIX_TREE_ITER_* flags and tag index
882 * Returns: pointer to chunk first slot, or NULL if iteration is over
884 void **radix_tree_next_chunk(struct radix_tree_root *root,
885 struct radix_tree_iter *iter, unsigned flags)
887 unsigned shift, tag = flags & RADIX_TREE_ITER_TAG_MASK;
888 struct radix_tree_node *rnode, *node;
889 unsigned long index, offset, maxindex;
891 if ((flags & RADIX_TREE_ITER_TAGGED) && !root_tag_get(root, tag))
895 * Catch next_index overflow after ~0UL. iter->index never overflows
896 * during iterating; it can be zero only at the beginning.
897 * And we cannot overflow iter->next_index in a single step,
898 * because RADIX_TREE_MAP_SHIFT < BITS_PER_LONG.
900 * This condition also used by radix_tree_next_slot() to stop
901 * contiguous iterating, and forbid swithing to the next chunk.
903 index = iter->next_index;
904 if (!index && iter->index)
908 shift = radix_tree_load_root(root, &rnode, &maxindex);
909 if (index > maxindex)
912 if (radix_tree_is_indirect_ptr(rnode)) {
913 rnode = indirect_to_ptr(rnode);
915 /* Single-slot tree */
917 iter->next_index = maxindex + 1;
919 __set_iter_shift(iter, shift);
920 return (void **)&root->rnode;
924 shift -= RADIX_TREE_MAP_SHIFT;
925 offset = index >> shift;
929 struct radix_tree_node *slot;
930 unsigned new_off = radix_tree_descend(node, &slot, offset);
932 if (new_off < offset) {
934 index &= ~((RADIX_TREE_MAP_SIZE << shift) - 1);
935 index |= offset << shift;
938 if ((flags & RADIX_TREE_ITER_TAGGED) ?
939 !tag_get(node, tag, offset) : !slot) {
941 if (flags & RADIX_TREE_ITER_CONTIG)
944 if (flags & RADIX_TREE_ITER_TAGGED)
945 offset = radix_tree_find_next_bit(
950 while (++offset < RADIX_TREE_MAP_SIZE) {
951 void *slot = node->slots[offset];
952 if (is_sibling_entry(node, slot))
957 index &= ~((RADIX_TREE_MAP_SIZE << shift) - 1);
958 index += offset << shift;
959 /* Overflow after ~0UL */
962 if (offset == RADIX_TREE_MAP_SIZE)
964 slot = rcu_dereference_raw(node->slots[offset]);
967 if ((slot == NULL) || (slot == RADIX_TREE_RETRY))
969 if (!radix_tree_is_indirect_ptr(slot))
972 node = indirect_to_ptr(slot);
973 shift -= RADIX_TREE_MAP_SHIFT;
974 offset = (index >> shift) & RADIX_TREE_MAP_MASK;
977 /* Update the iterator state */
978 iter->index = index & ~((1 << shift) - 1);
979 iter->next_index = (index | ((RADIX_TREE_MAP_SIZE << shift) - 1)) + 1;
980 __set_iter_shift(iter, shift);
982 /* Construct iter->tags bit-mask from node->tags[tag] array */
983 if (flags & RADIX_TREE_ITER_TAGGED) {
984 unsigned tag_long, tag_bit;
986 tag_long = offset / BITS_PER_LONG;
987 tag_bit = offset % BITS_PER_LONG;
988 iter->tags = node->tags[tag][tag_long] >> tag_bit;
989 /* This never happens if RADIX_TREE_TAG_LONGS == 1 */
990 if (tag_long < RADIX_TREE_TAG_LONGS - 1) {
991 /* Pick tags from next element */
993 iter->tags |= node->tags[tag][tag_long + 1] <<
994 (BITS_PER_LONG - tag_bit);
995 /* Clip chunk size, here only BITS_PER_LONG tags */
996 iter->next_index = index + BITS_PER_LONG;
1000 return node->slots + offset;
1002 EXPORT_SYMBOL(radix_tree_next_chunk);
1005 * radix_tree_range_tag_if_tagged - for each item in given range set given
1006 * tag if item has another tag set
1007 * @root: radix tree root
1008 * @first_indexp: pointer to a starting index of a range to scan
1009 * @last_index: last index of a range to scan
1010 * @nr_to_tag: maximum number items to tag
1011 * @iftag: tag index to test
1012 * @settag: tag index to set if tested tag is set
1014 * This function scans range of radix tree from first_index to last_index
1015 * (inclusive). For each item in the range if iftag is set, the function sets
1016 * also settag. The function stops either after tagging nr_to_tag items or
1017 * after reaching last_index.
1019 * The tags must be set from the leaf level only and propagated back up the
1020 * path to the root. We must do this so that we resolve the full path before
1021 * setting any tags on intermediate nodes. If we set tags as we descend, then
1022 * we can get to the leaf node and find that the index that has the iftag
1023 * set is outside the range we are scanning. This reults in dangling tags and
1024 * can lead to problems with later tag operations (e.g. livelocks on lookups).
1026 * The function returns number of leaves where the tag was set and sets
1027 * *first_indexp to the first unscanned index.
1028 * WARNING! *first_indexp can wrap if last_index is ULONG_MAX. Caller must
1029 * be prepared to handle that.
1031 unsigned long radix_tree_range_tag_if_tagged(struct radix_tree_root *root,
1032 unsigned long *first_indexp, unsigned long last_index,
1033 unsigned long nr_to_tag,
1034 unsigned int iftag, unsigned int settag)
1036 struct radix_tree_node *slot, *node = NULL;
1037 unsigned long maxindex;
1038 unsigned int shift = radix_tree_load_root(root, &slot, &maxindex);
1039 unsigned long tagged = 0;
1040 unsigned long index = *first_indexp;
1042 last_index = min(last_index, maxindex);
1043 if (index > last_index)
1047 if (!root_tag_get(root, iftag)) {
1048 *first_indexp = last_index + 1;
1051 if (!radix_tree_is_indirect_ptr(slot)) {
1052 *first_indexp = last_index + 1;
1053 root_tag_set(root, settag);
1057 node = indirect_to_ptr(slot);
1058 shift -= RADIX_TREE_MAP_SHIFT;
1061 unsigned long upindex;
1064 offset = (index >> shift) & RADIX_TREE_MAP_MASK;
1065 offset = radix_tree_descend(node, &slot, offset);
1068 if (!tag_get(node, iftag, offset))
1070 /* Sibling slots never have tags set on them */
1071 if (radix_tree_is_indirect_ptr(slot)) {
1072 node = indirect_to_ptr(slot);
1073 shift -= RADIX_TREE_MAP_SHIFT;
1079 tag_set(node, settag, offset);
1081 slot = node->parent;
1082 /* walk back up the path tagging interior nodes */
1083 upindex = index >> shift;
1085 upindex >>= RADIX_TREE_MAP_SHIFT;
1086 offset = upindex & RADIX_TREE_MAP_MASK;
1088 /* stop if we find a node with the tag already set */
1089 if (tag_get(slot, settag, offset))
1091 tag_set(slot, settag, offset);
1092 slot = slot->parent;
1096 /* Go to next item at level determined by 'shift' */
1097 index = ((index >> shift) + 1) << shift;
1098 /* Overflow can happen when last_index is ~0UL... */
1099 if (index > last_index || !index)
1101 offset = (index >> shift) & RADIX_TREE_MAP_MASK;
1102 while (offset == 0) {
1104 * We've fully scanned this node. Go up. Because
1105 * last_index is guaranteed to be in the tree, what
1106 * we do below cannot wander astray.
1108 node = node->parent;
1109 shift += RADIX_TREE_MAP_SHIFT;
1110 offset = (index >> shift) & RADIX_TREE_MAP_MASK;
1112 if (is_sibling_entry(node, node->slots[offset]))
1114 if (tagged >= nr_to_tag)
1118 * We need not to tag the root tag if there is no tag which is set with
1119 * settag within the range from *first_indexp to last_index.
1122 root_tag_set(root, settag);
1123 *first_indexp = index;
1127 EXPORT_SYMBOL(radix_tree_range_tag_if_tagged);
1130 * radix_tree_gang_lookup - perform multiple lookup on a radix tree
1131 * @root: radix tree root
1132 * @results: where the results of the lookup are placed
1133 * @first_index: start the lookup from this key
1134 * @max_items: place up to this many items at *results
1136 * Performs an index-ascending scan of the tree for present items. Places
1137 * them at *@results and returns the number of items which were placed at
1140 * The implementation is naive.
1142 * Like radix_tree_lookup, radix_tree_gang_lookup may be called under
1143 * rcu_read_lock. In this case, rather than the returned results being
1144 * an atomic snapshot of the tree at a single point in time, the semantics
1145 * of an RCU protected gang lookup are as though multiple radix_tree_lookups
1146 * have been issued in individual locks, and results stored in 'results'.
1149 radix_tree_gang_lookup(struct radix_tree_root *root, void **results,
1150 unsigned long first_index, unsigned int max_items)
1152 struct radix_tree_iter iter;
1154 unsigned int ret = 0;
1156 if (unlikely(!max_items))
1159 radix_tree_for_each_slot(slot, root, &iter, first_index) {
1160 results[ret] = rcu_dereference_raw(*slot);
1163 if (radix_tree_is_indirect_ptr(results[ret])) {
1164 slot = radix_tree_iter_retry(&iter);
1167 if (++ret == max_items)
1173 EXPORT_SYMBOL(radix_tree_gang_lookup);
1176 * radix_tree_gang_lookup_slot - perform multiple slot lookup on radix tree
1177 * @root: radix tree root
1178 * @results: where the results of the lookup are placed
1179 * @indices: where their indices should be placed (but usually NULL)
1180 * @first_index: start the lookup from this key
1181 * @max_items: place up to this many items at *results
1183 * Performs an index-ascending scan of the tree for present items. Places
1184 * their slots at *@results and returns the number of items which were
1185 * placed at *@results.
1187 * The implementation is naive.
1189 * Like radix_tree_gang_lookup as far as RCU and locking goes. Slots must
1190 * be dereferenced with radix_tree_deref_slot, and if using only RCU
1191 * protection, radix_tree_deref_slot may fail requiring a retry.
1194 radix_tree_gang_lookup_slot(struct radix_tree_root *root,
1195 void ***results, unsigned long *indices,
1196 unsigned long first_index, unsigned int max_items)
1198 struct radix_tree_iter iter;
1200 unsigned int ret = 0;
1202 if (unlikely(!max_items))
1205 radix_tree_for_each_slot(slot, root, &iter, first_index) {
1206 results[ret] = slot;
1208 indices[ret] = iter.index;
1209 if (++ret == max_items)
1215 EXPORT_SYMBOL(radix_tree_gang_lookup_slot);
1218 * radix_tree_gang_lookup_tag - perform multiple lookup on a radix tree
1220 * @root: radix tree root
1221 * @results: where the results of the lookup are placed
1222 * @first_index: start the lookup from this key
1223 * @max_items: place up to this many items at *results
1224 * @tag: the tag index (< RADIX_TREE_MAX_TAGS)
1226 * Performs an index-ascending scan of the tree for present items which
1227 * have the tag indexed by @tag set. Places the items at *@results and
1228 * returns the number of items which were placed at *@results.
1231 radix_tree_gang_lookup_tag(struct radix_tree_root *root, void **results,
1232 unsigned long first_index, unsigned int max_items,
1235 struct radix_tree_iter iter;
1237 unsigned int ret = 0;
1239 if (unlikely(!max_items))
1242 radix_tree_for_each_tagged(slot, root, &iter, first_index, tag) {
1243 results[ret] = rcu_dereference_raw(*slot);
1246 if (radix_tree_is_indirect_ptr(results[ret])) {
1247 slot = radix_tree_iter_retry(&iter);
1250 if (++ret == max_items)
1256 EXPORT_SYMBOL(radix_tree_gang_lookup_tag);
1259 * radix_tree_gang_lookup_tag_slot - perform multiple slot lookup on a
1260 * radix tree based on a tag
1261 * @root: radix tree root
1262 * @results: where the results of the lookup are placed
1263 * @first_index: start the lookup from this key
1264 * @max_items: place up to this many items at *results
1265 * @tag: the tag index (< RADIX_TREE_MAX_TAGS)
1267 * Performs an index-ascending scan of the tree for present items which
1268 * have the tag indexed by @tag set. Places the slots at *@results and
1269 * returns the number of slots which were placed at *@results.
1272 radix_tree_gang_lookup_tag_slot(struct radix_tree_root *root, void ***results,
1273 unsigned long first_index, unsigned int max_items,
1276 struct radix_tree_iter iter;
1278 unsigned int ret = 0;
1280 if (unlikely(!max_items))
1283 radix_tree_for_each_tagged(slot, root, &iter, first_index, tag) {
1284 results[ret] = slot;
1285 if (++ret == max_items)
1291 EXPORT_SYMBOL(radix_tree_gang_lookup_tag_slot);
1293 #if defined(CONFIG_SHMEM) && defined(CONFIG_SWAP)
1294 #include <linux/sched.h> /* for cond_resched() */
1296 struct locate_info {
1297 unsigned long found_index;
1302 * This linear search is at present only useful to shmem_unuse_inode().
1304 static unsigned long __locate(struct radix_tree_node *slot, void *item,
1305 unsigned long index, struct locate_info *info)
1307 unsigned int shift, height;
1310 height = slot->path & RADIX_TREE_HEIGHT_MASK;
1311 shift = height * RADIX_TREE_MAP_SHIFT;
1314 shift -= RADIX_TREE_MAP_SHIFT;
1316 for (i = (index >> shift) & RADIX_TREE_MAP_MASK;
1317 i < RADIX_TREE_MAP_SIZE;
1318 i++, index += (1UL << shift)) {
1319 struct radix_tree_node *node =
1320 rcu_dereference_raw(slot->slots[i]);
1321 if (node == RADIX_TREE_RETRY)
1323 if (!radix_tree_is_indirect_ptr(node)) {
1325 info->found_index = index;
1331 node = indirect_to_ptr(node);
1332 if (is_sibling_entry(slot, node))
1337 if (i == RADIX_TREE_MAP_SIZE)
1342 if ((index == 0) && (i == RADIX_TREE_MAP_SIZE))
1348 * radix_tree_locate_item - search through radix tree for item
1349 * @root: radix tree root
1350 * @item: item to be found
1352 * Returns index where item was found, or -1 if not found.
1353 * Caller must hold no lock (since this time-consuming function needs
1354 * to be preemptible), and must check afterwards if item is still there.
1356 unsigned long radix_tree_locate_item(struct radix_tree_root *root, void *item)
1358 struct radix_tree_node *node;
1359 unsigned long max_index;
1360 unsigned long cur_index = 0;
1361 struct locate_info info = {
1368 node = rcu_dereference_raw(root->rnode);
1369 if (!radix_tree_is_indirect_ptr(node)) {
1372 info.found_index = 0;
1376 node = indirect_to_ptr(node);
1378 max_index = node_maxindex(node);
1379 if (cur_index > max_index) {
1384 cur_index = __locate(node, item, cur_index, &info);
1387 } while (!info.stop && cur_index <= max_index);
1389 return info.found_index;
1392 unsigned long radix_tree_locate_item(struct radix_tree_root *root, void *item)
1396 #endif /* CONFIG_SHMEM && CONFIG_SWAP */
1399 * radix_tree_shrink - shrink height of a radix tree to minimal
1400 * @root radix tree root
1402 static inline void radix_tree_shrink(struct radix_tree_root *root)
1404 /* try to shrink tree height */
1405 while (root->height > 0) {
1406 struct radix_tree_node *to_free = root->rnode;
1407 struct radix_tree_node *slot;
1409 BUG_ON(!radix_tree_is_indirect_ptr(to_free));
1410 to_free = indirect_to_ptr(to_free);
1413 * The candidate node has more than one child, or its child
1414 * is not at the leftmost slot, or it is a multiorder entry,
1417 if (to_free->count != 1)
1419 slot = to_free->slots[0];
1422 if (!radix_tree_is_indirect_ptr(slot) && (root->height > 1))
1425 if (radix_tree_is_indirect_ptr(slot)) {
1426 slot = indirect_to_ptr(slot);
1427 slot->parent = NULL;
1428 slot = ptr_to_indirect(slot);
1432 * We don't need rcu_assign_pointer(), since we are simply
1433 * moving the node from one part of the tree to another: if it
1434 * was safe to dereference the old pointer to it
1435 * (to_free->slots[0]), it will be safe to dereference the new
1436 * one (root->rnode) as far as dependent read barriers go.
1442 * We have a dilemma here. The node's slot[0] must not be
1443 * NULLed in case there are concurrent lookups expecting to
1444 * find the item. However if this was a bottom-level node,
1445 * then it may be subject to the slot pointer being visible
1446 * to callers dereferencing it. If item corresponding to
1447 * slot[0] is subsequently deleted, these callers would expect
1448 * their slot to become empty sooner or later.
1450 * For example, lockless pagecache will look up a slot, deref
1451 * the page pointer, and if the page is 0 refcount it means it
1452 * was concurrently deleted from pagecache so try the deref
1453 * again. Fortunately there is already a requirement for logic
1454 * to retry the entire slot lookup -- the indirect pointer
1455 * problem (replacing direct root node with an indirect pointer
1456 * also results in a stale slot). So tag the slot as indirect
1457 * to force callers to retry.
1459 if (!radix_tree_is_indirect_ptr(slot))
1460 to_free->slots[0] = RADIX_TREE_RETRY;
1462 radix_tree_node_free(to_free);
1467 * __radix_tree_delete_node - try to free node after clearing a slot
1468 * @root: radix tree root
1469 * @node: node containing @index
1471 * After clearing the slot at @index in @node from radix tree
1472 * rooted at @root, call this function to attempt freeing the
1473 * node and shrinking the tree.
1475 * Returns %true if @node was freed, %false otherwise.
1477 bool __radix_tree_delete_node(struct radix_tree_root *root,
1478 struct radix_tree_node *node)
1480 bool deleted = false;
1483 struct radix_tree_node *parent;
1486 if (node == indirect_to_ptr(root->rnode)) {
1487 radix_tree_shrink(root);
1488 if (root->height == 0)
1494 parent = node->parent;
1496 unsigned int offset;
1498 offset = node->path >> RADIX_TREE_HEIGHT_SHIFT;
1499 parent->slots[offset] = NULL;
1502 root_tag_clear_all(root);
1507 radix_tree_node_free(node);
1516 static inline void delete_sibling_entries(struct radix_tree_node *node,
1517 void *ptr, unsigned offset)
1519 #ifdef CONFIG_RADIX_TREE_MULTIORDER
1521 for (i = 1; offset + i < RADIX_TREE_MAP_SIZE; i++) {
1522 if (node->slots[offset + i] != ptr)
1524 node->slots[offset + i] = NULL;
1531 * radix_tree_delete_item - delete an item from a radix tree
1532 * @root: radix tree root
1534 * @item: expected item
1536 * Remove @item at @index from the radix tree rooted at @root.
1538 * Returns the address of the deleted item, or NULL if it was not present
1539 * or the entry at the given @index was not @item.
1541 void *radix_tree_delete_item(struct radix_tree_root *root,
1542 unsigned long index, void *item)
1544 struct radix_tree_node *node;
1545 unsigned int offset;
1550 entry = __radix_tree_lookup(root, index, &node, &slot);
1554 if (item && entry != item)
1558 root_tag_clear_all(root);
1563 offset = get_slot_offset(node, slot);
1566 * Clear all tags associated with the item to be deleted.
1567 * This way of doing it would be inefficient, but seldom is any set.
1569 for (tag = 0; tag < RADIX_TREE_MAX_TAGS; tag++) {
1570 if (tag_get(node, tag, offset))
1571 radix_tree_tag_clear(root, index, tag);
1574 delete_sibling_entries(node, ptr_to_indirect(slot), offset);
1575 node->slots[offset] = NULL;
1578 __radix_tree_delete_node(root, node);
1582 EXPORT_SYMBOL(radix_tree_delete_item);
1585 * radix_tree_delete - delete an item from a radix tree
1586 * @root: radix tree root
1589 * Remove the item at @index from the radix tree rooted at @root.
1591 * Returns the address of the deleted item, or NULL if it was not present.
1593 void *radix_tree_delete(struct radix_tree_root *root, unsigned long index)
1595 return radix_tree_delete_item(root, index, NULL);
1597 EXPORT_SYMBOL(radix_tree_delete);
1600 * radix_tree_tagged - test whether any items in the tree are tagged
1601 * @root: radix tree root
1604 int radix_tree_tagged(struct radix_tree_root *root, unsigned int tag)
1606 return root_tag_get(root, tag);
1608 EXPORT_SYMBOL(radix_tree_tagged);
1611 radix_tree_node_ctor(void *arg)
1613 struct radix_tree_node *node = arg;
1615 memset(node, 0, sizeof(*node));
1616 INIT_LIST_HEAD(&node->private_list);
1619 static __init unsigned long __maxindex(unsigned int height)
1621 unsigned int width = height * RADIX_TREE_MAP_SHIFT;
1622 int shift = RADIX_TREE_INDEX_BITS - width;
1626 if (shift >= BITS_PER_LONG)
1628 return ~0UL >> shift;
1631 static __init void radix_tree_init_maxindex(void)
1635 for (i = 0; i < ARRAY_SIZE(height_to_maxindex); i++)
1636 height_to_maxindex[i] = __maxindex(i);
1639 static int radix_tree_callback(struct notifier_block *nfb,
1640 unsigned long action,
1643 int cpu = (long)hcpu;
1644 struct radix_tree_preload *rtp;
1645 struct radix_tree_node *node;
1647 /* Free per-cpu pool of perloaded nodes */
1648 if (action == CPU_DEAD || action == CPU_DEAD_FROZEN) {
1649 rtp = &per_cpu(radix_tree_preloads, cpu);
1652 rtp->nodes = node->private_data;
1653 kmem_cache_free(radix_tree_node_cachep, node);
1660 void __init radix_tree_init(void)
1662 radix_tree_node_cachep = kmem_cache_create("radix_tree_node",
1663 sizeof(struct radix_tree_node), 0,
1664 SLAB_PANIC | SLAB_RECLAIM_ACCOUNT,
1665 radix_tree_node_ctor);
1666 radix_tree_init_maxindex();
1667 hotcpu_notifier(radix_tree_callback, 0);