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
7 * Copyright (C) 2016 Intel, Matthew Wilcox
8 * Copyright (C) 2016 Intel, Ross Zwisler
10 * This program is free software; you can redistribute it and/or
11 * modify it under the terms of the GNU General Public License as
12 * published by the Free Software Foundation; either version 2, or (at
13 * your option) any later version.
15 * This program is distributed in the hope that it will be useful, but
16 * WITHOUT ANY WARRANTY; without even the implied warranty of
17 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
18 * General Public License for more details.
20 * You should have received a copy of the GNU General Public License
21 * along with this program; if not, write to the Free Software
22 * Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA.
25 #include <linux/cpu.h>
26 #include <linux/errno.h>
27 #include <linux/init.h>
28 #include <linux/kernel.h>
29 #include <linux/export.h>
30 #include <linux/radix-tree.h>
31 #include <linux/percpu.h>
32 #include <linux/slab.h>
33 #include <linux/kmemleak.h>
34 #include <linux/notifier.h>
35 #include <linux/cpu.h>
36 #include <linux/string.h>
37 #include <linux/bitops.h>
38 #include <linux/rcupdate.h>
39 #include <linux/preempt.h> /* in_interrupt() */
42 /* Number of nodes in fully populated tree of given height */
43 static unsigned long height_to_maxnodes[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 struct radix_tree_node *entry_to_node(void *ptr)
75 return (void *)((unsigned long)ptr & ~RADIX_TREE_INTERNAL_NODE);
78 static inline void *node_to_entry(void *ptr)
80 return (void *)((unsigned long)ptr | RADIX_TREE_INTERNAL_NODE);
83 #define RADIX_TREE_RETRY node_to_entry(NULL)
85 #ifdef CONFIG_RADIX_TREE_MULTIORDER
86 /* Sibling slots point directly to another slot in the same node */
87 static inline bool is_sibling_entry(struct radix_tree_node *parent, void *node)
90 return (parent->slots <= ptr) &&
91 (ptr < parent->slots + RADIX_TREE_MAP_SIZE);
94 static inline bool is_sibling_entry(struct radix_tree_node *parent, void *node)
100 static inline unsigned long get_slot_offset(struct radix_tree_node *parent,
103 return slot - parent->slots;
106 static unsigned int radix_tree_descend(struct radix_tree_node *parent,
107 struct radix_tree_node **nodep, unsigned long index)
109 unsigned int offset = (index >> parent->shift) & RADIX_TREE_MAP_MASK;
110 void **entry = rcu_dereference_raw(parent->slots[offset]);
112 #ifdef CONFIG_RADIX_TREE_MULTIORDER
113 if (radix_tree_is_internal_node(entry)) {
114 if (is_sibling_entry(parent, entry)) {
115 void **sibentry = (void **) entry_to_node(entry);
116 offset = get_slot_offset(parent, sibentry);
117 entry = rcu_dereference_raw(*sibentry);
122 *nodep = (void *)entry;
126 static inline gfp_t root_gfp_mask(struct radix_tree_root *root)
128 return root->gfp_mask & __GFP_BITS_MASK;
131 static inline void tag_set(struct radix_tree_node *node, unsigned int tag,
134 __set_bit(offset, node->tags[tag]);
137 static inline void tag_clear(struct radix_tree_node *node, unsigned int tag,
140 __clear_bit(offset, node->tags[tag]);
143 static inline int tag_get(struct radix_tree_node *node, unsigned int tag,
146 return test_bit(offset, node->tags[tag]);
149 static inline void root_tag_set(struct radix_tree_root *root, unsigned int tag)
151 root->gfp_mask |= (__force gfp_t)(1 << (tag + __GFP_BITS_SHIFT));
154 static inline void root_tag_clear(struct radix_tree_root *root, unsigned tag)
156 root->gfp_mask &= (__force gfp_t)~(1 << (tag + __GFP_BITS_SHIFT));
159 static inline void root_tag_clear_all(struct radix_tree_root *root)
161 root->gfp_mask &= __GFP_BITS_MASK;
164 static inline int root_tag_get(struct radix_tree_root *root, unsigned int tag)
166 return (__force int)root->gfp_mask & (1 << (tag + __GFP_BITS_SHIFT));
169 static inline unsigned root_tags_get(struct radix_tree_root *root)
171 return (__force unsigned)root->gfp_mask >> __GFP_BITS_SHIFT;
175 * Returns 1 if any slot in the node has this tag set.
176 * Otherwise returns 0.
178 static inline int any_tag_set(struct radix_tree_node *node, unsigned int tag)
181 for (idx = 0; idx < RADIX_TREE_TAG_LONGS; idx++) {
182 if (node->tags[tag][idx])
189 * radix_tree_find_next_bit - find the next set bit in a memory region
191 * @addr: The address to base the search on
192 * @size: The bitmap size in bits
193 * @offset: The bitnumber to start searching at
195 * Unrollable variant of find_next_bit() for constant size arrays.
196 * Tail bits starting from size to roundup(size, BITS_PER_LONG) must be zero.
197 * Returns next bit offset, or size if nothing found.
199 static __always_inline unsigned long
200 radix_tree_find_next_bit(struct radix_tree_node *node, unsigned int tag,
201 unsigned long offset)
203 const unsigned long *addr = node->tags[tag];
205 if (offset < RADIX_TREE_MAP_SIZE) {
208 addr += offset / BITS_PER_LONG;
209 tmp = *addr >> (offset % BITS_PER_LONG);
211 return __ffs(tmp) + offset;
212 offset = (offset + BITS_PER_LONG) & ~(BITS_PER_LONG - 1);
213 while (offset < RADIX_TREE_MAP_SIZE) {
216 return __ffs(tmp) + offset;
217 offset += BITS_PER_LONG;
220 return RADIX_TREE_MAP_SIZE;
223 static unsigned int iter_offset(const struct radix_tree_iter *iter)
225 return (iter->index >> iter_shift(iter)) & RADIX_TREE_MAP_MASK;
229 * The maximum index which can be stored in a radix tree
231 static inline unsigned long shift_maxindex(unsigned int shift)
233 return (RADIX_TREE_MAP_SIZE << shift) - 1;
236 static inline unsigned long node_maxindex(struct radix_tree_node *node)
238 return shift_maxindex(node->shift);
242 static void dump_node(struct radix_tree_node *node, unsigned long index)
246 pr_debug("radix node: %p offset %d indices %lu-%lu parent %p tags %lx %lx %lx shift %d count %d exceptional %d\n",
247 node, node->offset, index, index | node_maxindex(node),
249 node->tags[0][0], node->tags[1][0], node->tags[2][0],
250 node->shift, node->count, node->exceptional);
252 for (i = 0; i < RADIX_TREE_MAP_SIZE; i++) {
253 unsigned long first = index | (i << node->shift);
254 unsigned long last = first | ((1UL << node->shift) - 1);
255 void *entry = node->slots[i];
258 if (entry == RADIX_TREE_RETRY) {
259 pr_debug("radix retry offset %ld indices %lu-%lu parent %p\n",
260 i, first, last, node);
261 } else if (!radix_tree_is_internal_node(entry)) {
262 pr_debug("radix entry %p offset %ld indices %lu-%lu parent %p\n",
263 entry, i, first, last, node);
264 } else if (is_sibling_entry(node, entry)) {
265 pr_debug("radix sblng %p offset %ld indices %lu-%lu parent %p val %p\n",
266 entry, i, first, last, node,
267 *(void **)entry_to_node(entry));
269 dump_node(entry_to_node(entry), first);
275 static void radix_tree_dump(struct radix_tree_root *root)
277 pr_debug("radix root: %p rnode %p tags %x\n",
279 root->gfp_mask >> __GFP_BITS_SHIFT);
280 if (!radix_tree_is_internal_node(root->rnode))
282 dump_node(entry_to_node(root->rnode), 0);
287 * This assumes that the caller has performed appropriate preallocation, and
288 * that the caller has pinned this thread of control to the current CPU.
290 static struct radix_tree_node *
291 radix_tree_node_alloc(struct radix_tree_root *root)
293 struct radix_tree_node *ret = NULL;
294 gfp_t gfp_mask = root_gfp_mask(root);
297 * Preload code isn't irq safe and it doesn't make sense to use
298 * preloading during an interrupt anyway as all the allocations have
299 * to be atomic. So just do normal allocation when in interrupt.
301 if (!gfpflags_allow_blocking(gfp_mask) && !in_interrupt()) {
302 struct radix_tree_preload *rtp;
305 * Even if the caller has preloaded, try to allocate from the
306 * cache first for the new node to get accounted to the memory
309 ret = kmem_cache_alloc(radix_tree_node_cachep,
310 gfp_mask | __GFP_NOWARN);
315 * Provided the caller has preloaded here, we will always
316 * succeed in getting a node here (and never reach
319 rtp = this_cpu_ptr(&radix_tree_preloads);
322 rtp->nodes = ret->private_data;
323 ret->private_data = NULL;
327 * Update the allocation stack trace as this is more useful
330 kmemleak_update_trace(ret);
333 ret = kmem_cache_alloc(radix_tree_node_cachep, gfp_mask);
335 BUG_ON(radix_tree_is_internal_node(ret));
339 static void radix_tree_node_rcu_free(struct rcu_head *head)
341 struct radix_tree_node *node =
342 container_of(head, struct radix_tree_node, rcu_head);
345 * Must only free zeroed nodes into the slab. We can be left with
346 * non-NULL entries by radix_tree_free_nodes, so clear the entries
349 memset(node->slots, 0, sizeof(node->slots));
350 memset(node->tags, 0, sizeof(node->tags));
351 INIT_LIST_HEAD(&node->private_list);
353 kmem_cache_free(radix_tree_node_cachep, node);
357 radix_tree_node_free(struct radix_tree_node *node)
359 call_rcu(&node->rcu_head, radix_tree_node_rcu_free);
363 * Load up this CPU's radix_tree_node buffer with sufficient objects to
364 * ensure that the addition of a single element in the tree cannot fail. On
365 * success, return zero, with preemption disabled. On error, return -ENOMEM
366 * with preemption not disabled.
368 * To make use of this facility, the radix tree must be initialised without
369 * __GFP_DIRECT_RECLAIM being passed to INIT_RADIX_TREE().
371 static int __radix_tree_preload(gfp_t gfp_mask, unsigned nr)
373 struct radix_tree_preload *rtp;
374 struct radix_tree_node *node;
378 * Nodes preloaded by one cgroup can be be used by another cgroup, so
379 * they should never be accounted to any particular memory cgroup.
381 gfp_mask &= ~__GFP_ACCOUNT;
384 rtp = this_cpu_ptr(&radix_tree_preloads);
385 while (rtp->nr < nr) {
387 node = kmem_cache_alloc(radix_tree_node_cachep, gfp_mask);
391 rtp = this_cpu_ptr(&radix_tree_preloads);
393 node->private_data = rtp->nodes;
397 kmem_cache_free(radix_tree_node_cachep, node);
406 * Load up this CPU's radix_tree_node buffer with sufficient objects to
407 * ensure that the addition of a single element in the tree cannot fail. On
408 * success, return zero, with preemption disabled. On error, return -ENOMEM
409 * with preemption not disabled.
411 * To make use of this facility, the radix tree must be initialised without
412 * __GFP_DIRECT_RECLAIM being passed to INIT_RADIX_TREE().
414 int radix_tree_preload(gfp_t gfp_mask)
416 /* Warn on non-sensical use... */
417 WARN_ON_ONCE(!gfpflags_allow_blocking(gfp_mask));
418 return __radix_tree_preload(gfp_mask, RADIX_TREE_PRELOAD_SIZE);
420 EXPORT_SYMBOL(radix_tree_preload);
423 * The same as above function, except we don't guarantee preloading happens.
424 * We do it, if we decide it helps. On success, return zero with preemption
425 * disabled. On error, return -ENOMEM with preemption not disabled.
427 int radix_tree_maybe_preload(gfp_t gfp_mask)
429 if (gfpflags_allow_blocking(gfp_mask))
430 return __radix_tree_preload(gfp_mask, RADIX_TREE_PRELOAD_SIZE);
431 /* Preloading doesn't help anything with this gfp mask, skip it */
435 EXPORT_SYMBOL(radix_tree_maybe_preload);
437 #ifdef CONFIG_RADIX_TREE_MULTIORDER
439 * Preload with enough objects to ensure that we can split a single entry
440 * of order @old_order into many entries of size @new_order
442 int radix_tree_split_preload(unsigned int old_order, unsigned int new_order,
445 unsigned top = 1 << (old_order % RADIX_TREE_MAP_SHIFT);
446 unsigned layers = (old_order / RADIX_TREE_MAP_SHIFT) -
447 (new_order / RADIX_TREE_MAP_SHIFT);
450 WARN_ON_ONCE(!gfpflags_allow_blocking(gfp_mask));
451 BUG_ON(new_order >= old_order);
454 nr = nr * RADIX_TREE_MAP_SIZE + 1;
455 return __radix_tree_preload(gfp_mask, top * nr);
460 * The same as function above, but preload number of nodes required to insert
461 * (1 << order) continuous naturally-aligned elements.
463 int radix_tree_maybe_preload_order(gfp_t gfp_mask, int order)
465 unsigned long nr_subtrees;
466 int nr_nodes, subtree_height;
468 /* Preloading doesn't help anything with this gfp mask, skip it */
469 if (!gfpflags_allow_blocking(gfp_mask)) {
475 * Calculate number and height of fully populated subtrees it takes to
476 * store (1 << order) elements.
478 nr_subtrees = 1 << order;
479 for (subtree_height = 0; nr_subtrees > RADIX_TREE_MAP_SIZE;
481 nr_subtrees >>= RADIX_TREE_MAP_SHIFT;
484 * The worst case is zero height tree with a single item at index 0 and
485 * then inserting items starting at ULONG_MAX - (1 << order).
487 * This requires RADIX_TREE_MAX_PATH nodes to build branch from root to
490 nr_nodes = RADIX_TREE_MAX_PATH;
492 /* Plus branch to fully populated subtrees. */
493 nr_nodes += RADIX_TREE_MAX_PATH - subtree_height;
495 /* Root node is shared. */
498 /* Plus nodes required to build subtrees. */
499 nr_nodes += nr_subtrees * height_to_maxnodes[subtree_height];
501 return __radix_tree_preload(gfp_mask, nr_nodes);
504 static unsigned radix_tree_load_root(struct radix_tree_root *root,
505 struct radix_tree_node **nodep, unsigned long *maxindex)
507 struct radix_tree_node *node = rcu_dereference_raw(root->rnode);
511 if (likely(radix_tree_is_internal_node(node))) {
512 node = entry_to_node(node);
513 *maxindex = node_maxindex(node);
514 return node->shift + RADIX_TREE_MAP_SHIFT;
522 * Extend a radix tree so it can store key @index.
524 static int radix_tree_extend(struct radix_tree_root *root,
525 unsigned long index, unsigned int shift)
527 struct radix_tree_node *slot;
528 unsigned int maxshift;
531 /* Figure out what the shift should be. */
533 while (index > shift_maxindex(maxshift))
534 maxshift += RADIX_TREE_MAP_SHIFT;
541 struct radix_tree_node *node = radix_tree_node_alloc(root);
546 /* Propagate the aggregated tag info into the new root */
547 for (tag = 0; tag < RADIX_TREE_MAX_TAGS; tag++) {
548 if (root_tag_get(root, tag))
549 tag_set(node, tag, 0);
552 BUG_ON(shift > BITS_PER_LONG);
557 if (radix_tree_is_internal_node(slot)) {
558 entry_to_node(slot)->parent = node;
560 /* Moving an exceptional root->rnode to a node */
561 if (radix_tree_exceptional_entry(slot))
562 node->exceptional = 1;
564 node->slots[0] = slot;
565 slot = node_to_entry(node);
566 rcu_assign_pointer(root->rnode, slot);
567 shift += RADIX_TREE_MAP_SHIFT;
568 } while (shift <= maxshift);
570 return maxshift + RADIX_TREE_MAP_SHIFT;
574 * radix_tree_shrink - shrink radix tree to minimum height
575 * @root radix tree root
577 static inline void radix_tree_shrink(struct radix_tree_root *root,
578 radix_tree_update_node_t update_node,
582 struct radix_tree_node *node = root->rnode;
583 struct radix_tree_node *child;
585 if (!radix_tree_is_internal_node(node))
587 node = entry_to_node(node);
590 * The candidate node has more than one child, or its child
591 * is not at the leftmost slot, or the child is a multiorder
592 * entry, we cannot shrink.
594 if (node->count != 1)
596 child = node->slots[0];
599 if (!radix_tree_is_internal_node(child) && node->shift)
602 if (radix_tree_is_internal_node(child))
603 entry_to_node(child)->parent = NULL;
606 * We don't need rcu_assign_pointer(), since we are simply
607 * moving the node from one part of the tree to another: if it
608 * was safe to dereference the old pointer to it
609 * (node->slots[0]), it will be safe to dereference the new
610 * one (root->rnode) as far as dependent read barriers go.
615 * We have a dilemma here. The node's slot[0] must not be
616 * NULLed in case there are concurrent lookups expecting to
617 * find the item. However if this was a bottom-level node,
618 * then it may be subject to the slot pointer being visible
619 * to callers dereferencing it. If item corresponding to
620 * slot[0] is subsequently deleted, these callers would expect
621 * their slot to become empty sooner or later.
623 * For example, lockless pagecache will look up a slot, deref
624 * the page pointer, and if the page has 0 refcount it means it
625 * was concurrently deleted from pagecache so try the deref
626 * again. Fortunately there is already a requirement for logic
627 * to retry the entire slot lookup -- the indirect pointer
628 * problem (replacing direct root node with an indirect pointer
629 * also results in a stale slot). So tag the slot as indirect
630 * to force callers to retry.
633 if (!radix_tree_is_internal_node(child)) {
634 node->slots[0] = RADIX_TREE_RETRY;
636 update_node(node, private);
639 radix_tree_node_free(node);
643 static void delete_node(struct radix_tree_root *root,
644 struct radix_tree_node *node,
645 radix_tree_update_node_t update_node, void *private)
648 struct radix_tree_node *parent;
651 if (node == entry_to_node(root->rnode))
652 radix_tree_shrink(root, update_node, private);
656 parent = node->parent;
658 parent->slots[node->offset] = NULL;
661 root_tag_clear_all(root);
665 radix_tree_node_free(node);
672 * __radix_tree_create - create a slot in a radix tree
673 * @root: radix tree root
675 * @order: index occupies 2^order aligned slots
676 * @nodep: returns node
677 * @slotp: returns slot
679 * Create, if necessary, and return the node and slot for an item
680 * at position @index in the radix tree @root.
682 * Until there is more than one item in the tree, no nodes are
683 * allocated and @root->rnode is used as a direct slot instead of
684 * pointing to a node, in which case *@nodep will be NULL.
686 * Returns -ENOMEM, or 0 for success.
688 int __radix_tree_create(struct radix_tree_root *root, unsigned long index,
689 unsigned order, struct radix_tree_node **nodep,
692 struct radix_tree_node *node = NULL, *child;
693 void **slot = (void **)&root->rnode;
694 unsigned long maxindex;
695 unsigned int shift, offset = 0;
696 unsigned long max = index | ((1UL << order) - 1);
698 shift = radix_tree_load_root(root, &child, &maxindex);
700 /* Make sure the tree is high enough. */
701 if (order > 0 && max == ((1UL << order) - 1))
703 if (max > maxindex) {
704 int error = radix_tree_extend(root, max, shift);
711 while (shift > order) {
712 shift -= RADIX_TREE_MAP_SHIFT;
714 /* Have to add a child node. */
715 child = radix_tree_node_alloc(root);
718 child->shift = shift;
719 child->offset = offset;
721 child->exceptional = 0;
722 child->parent = node;
723 rcu_assign_pointer(*slot, node_to_entry(child));
726 } else if (!radix_tree_is_internal_node(child))
729 /* Go a level down */
730 node = entry_to_node(child);
731 offset = radix_tree_descend(node, &child, index);
732 slot = &node->slots[offset];
742 #ifdef CONFIG_RADIX_TREE_MULTIORDER
744 * Free any nodes below this node. The tree is presumed to not need
745 * shrinking, and any user data in the tree is presumed to not need a
746 * destructor called on it. If we need to add a destructor, we can
747 * add that functionality later. Note that we may not clear tags or
748 * slots from the tree as an RCU walker may still have a pointer into
749 * this subtree. We could replace the entries with RADIX_TREE_RETRY,
750 * but we'll still have to clear those in rcu_free.
752 static void radix_tree_free_nodes(struct radix_tree_node *node)
755 struct radix_tree_node *child = entry_to_node(node);
758 void *entry = child->slots[offset];
759 if (radix_tree_is_internal_node(entry) &&
760 !is_sibling_entry(child, entry)) {
761 child = entry_to_node(entry);
766 while (offset == RADIX_TREE_MAP_SIZE) {
767 struct radix_tree_node *old = child;
768 offset = child->offset + 1;
769 child = child->parent;
770 radix_tree_node_free(old);
771 if (old == entry_to_node(node))
777 static inline int insert_entries(struct radix_tree_node *node, void **slot,
778 void *item, unsigned order, bool replace)
780 struct radix_tree_node *child;
781 unsigned i, n, tag, offset, tags = 0;
784 if (order > node->shift)
785 n = 1 << (order - node->shift);
788 offset = get_slot_offset(node, slot);
795 offset = offset & ~(n - 1);
796 slot = &node->slots[offset];
798 child = node_to_entry(slot);
800 for (i = 0; i < n; i++) {
804 for (tag = 0; tag < RADIX_TREE_MAX_TAGS; tag++)
805 if (tag_get(node, tag, offset + i))
812 for (i = 0; i < n; i++) {
813 struct radix_tree_node *old = slot[i];
815 rcu_assign_pointer(slot[i], child);
816 for (tag = 0; tag < RADIX_TREE_MAX_TAGS; tag++)
817 if (tags & (1 << tag))
818 tag_clear(node, tag, offset + i);
820 rcu_assign_pointer(slot[i], item);
821 for (tag = 0; tag < RADIX_TREE_MAX_TAGS; tag++)
822 if (tags & (1 << tag))
823 tag_set(node, tag, offset);
825 if (radix_tree_is_internal_node(old) &&
826 !is_sibling_entry(node, old) &&
827 (old != RADIX_TREE_RETRY))
828 radix_tree_free_nodes(old);
829 if (radix_tree_exceptional_entry(old))
834 if (radix_tree_exceptional_entry(item))
835 node->exceptional += n;
840 static inline int insert_entries(struct radix_tree_node *node, void **slot,
841 void *item, unsigned order, bool replace)
845 rcu_assign_pointer(*slot, item);
848 if (radix_tree_exceptional_entry(item))
856 * __radix_tree_insert - insert into a radix tree
857 * @root: radix tree root
859 * @order: key covers the 2^order indices around index
860 * @item: item to insert
862 * Insert an item into the radix tree at position @index.
864 int __radix_tree_insert(struct radix_tree_root *root, unsigned long index,
865 unsigned order, void *item)
867 struct radix_tree_node *node;
871 BUG_ON(radix_tree_is_internal_node(item));
873 error = __radix_tree_create(root, index, order, &node, &slot);
877 error = insert_entries(node, slot, item, order, false);
882 unsigned offset = get_slot_offset(node, slot);
883 BUG_ON(tag_get(node, 0, offset));
884 BUG_ON(tag_get(node, 1, offset));
885 BUG_ON(tag_get(node, 2, offset));
887 BUG_ON(root_tags_get(root));
892 EXPORT_SYMBOL(__radix_tree_insert);
895 * __radix_tree_lookup - lookup an item in a radix tree
896 * @root: radix tree root
898 * @nodep: returns node
899 * @slotp: returns slot
901 * Lookup and return the item at position @index in the radix
904 * Until there is more than one item in the tree, no nodes are
905 * allocated and @root->rnode is used as a direct slot instead of
906 * pointing to a node, in which case *@nodep will be NULL.
908 void *__radix_tree_lookup(struct radix_tree_root *root, unsigned long index,
909 struct radix_tree_node **nodep, void ***slotp)
911 struct radix_tree_node *node, *parent;
912 unsigned long maxindex;
917 slot = (void **)&root->rnode;
918 radix_tree_load_root(root, &node, &maxindex);
919 if (index > maxindex)
922 while (radix_tree_is_internal_node(node)) {
925 if (node == RADIX_TREE_RETRY)
927 parent = entry_to_node(node);
928 offset = radix_tree_descend(parent, &node, index);
929 slot = parent->slots + offset;
940 * radix_tree_lookup_slot - lookup a slot in a radix tree
941 * @root: radix tree root
944 * Returns: the slot corresponding to the position @index in the
945 * radix tree @root. This is useful for update-if-exists operations.
947 * This function can be called under rcu_read_lock iff the slot is not
948 * modified by radix_tree_replace_slot, otherwise it must be called
949 * exclusive from other writers. Any dereference of the slot must be done
950 * using radix_tree_deref_slot.
952 void **radix_tree_lookup_slot(struct radix_tree_root *root, unsigned long index)
956 if (!__radix_tree_lookup(root, index, NULL, &slot))
960 EXPORT_SYMBOL(radix_tree_lookup_slot);
963 * radix_tree_lookup - perform lookup operation on a radix tree
964 * @root: radix tree root
967 * Lookup the item at the position @index in the radix tree @root.
969 * This function can be called under rcu_read_lock, however the caller
970 * must manage lifetimes of leaf nodes (eg. RCU may also be used to free
971 * them safely). No RCU barriers are required to access or modify the
972 * returned item, however.
974 void *radix_tree_lookup(struct radix_tree_root *root, unsigned long index)
976 return __radix_tree_lookup(root, index, NULL, NULL);
978 EXPORT_SYMBOL(radix_tree_lookup);
980 static inline int slot_count(struct radix_tree_node *node,
984 #ifdef CONFIG_RADIX_TREE_MULTIORDER
985 void *ptr = node_to_entry(slot);
986 unsigned offset = get_slot_offset(node, slot);
989 for (i = 1; offset + i < RADIX_TREE_MAP_SIZE; i++) {
990 if (node->slots[offset + i] != ptr)
998 static void replace_slot(struct radix_tree_root *root,
999 struct radix_tree_node *node,
1000 void **slot, void *item,
1001 bool warn_typeswitch)
1003 void *old = rcu_dereference_raw(*slot);
1004 int count, exceptional;
1006 WARN_ON_ONCE(radix_tree_is_internal_node(item));
1008 count = !!item - !!old;
1009 exceptional = !!radix_tree_exceptional_entry(item) -
1010 !!radix_tree_exceptional_entry(old);
1012 WARN_ON_ONCE(warn_typeswitch && (count || exceptional));
1015 node->count += count;
1017 exceptional *= slot_count(node, slot);
1018 node->exceptional += exceptional;
1022 rcu_assign_pointer(*slot, item);
1025 static inline void delete_sibling_entries(struct radix_tree_node *node,
1028 #ifdef CONFIG_RADIX_TREE_MULTIORDER
1029 bool exceptional = radix_tree_exceptional_entry(*slot);
1030 void *ptr = node_to_entry(slot);
1031 unsigned offset = get_slot_offset(node, slot);
1034 for (i = 1; offset + i < RADIX_TREE_MAP_SIZE; i++) {
1035 if (node->slots[offset + i] != ptr)
1037 node->slots[offset + i] = NULL;
1040 node->exceptional--;
1046 * __radix_tree_replace - replace item in a slot
1047 * @root: radix tree root
1048 * @node: pointer to tree node
1049 * @slot: pointer to slot in @node
1050 * @item: new item to store in the slot.
1051 * @update_node: callback for changing leaf nodes
1052 * @private: private data to pass to @update_node
1054 * For use with __radix_tree_lookup(). Caller must hold tree write locked
1055 * across slot lookup and replacement.
1057 void __radix_tree_replace(struct radix_tree_root *root,
1058 struct radix_tree_node *node,
1059 void **slot, void *item,
1060 radix_tree_update_node_t update_node, void *private)
1063 delete_sibling_entries(node, slot);
1065 * This function supports replacing exceptional entries and
1066 * deleting entries, but that needs accounting against the
1067 * node unless the slot is root->rnode.
1069 replace_slot(root, node, slot, item,
1070 !node && slot != (void **)&root->rnode);
1076 update_node(node, private);
1078 delete_node(root, node, update_node, private);
1082 * radix_tree_replace_slot - replace item in a slot
1083 * @root: radix tree root
1084 * @slot: pointer to slot
1085 * @item: new item to store in the slot.
1087 * For use with radix_tree_lookup_slot(), radix_tree_gang_lookup_slot(),
1088 * radix_tree_gang_lookup_tag_slot(). Caller must hold tree write locked
1089 * across slot lookup and replacement.
1091 * NOTE: This cannot be used to switch between non-entries (empty slots),
1092 * regular entries, and exceptional entries, as that requires accounting
1093 * inside the radix tree node. When switching from one type of entry or
1094 * deleting, use __radix_tree_lookup() and __radix_tree_replace() or
1095 * radix_tree_iter_replace().
1097 void radix_tree_replace_slot(struct radix_tree_root *root,
1098 void **slot, void *item)
1100 replace_slot(root, NULL, slot, item, true);
1104 * radix_tree_iter_replace - replace item in a slot
1105 * @root: radix tree root
1106 * @slot: pointer to slot
1107 * @item: new item to store in the slot.
1109 * For use with radix_tree_split() and radix_tree_for_each_slot().
1110 * Caller must hold tree write locked across split and replacement.
1112 void radix_tree_iter_replace(struct radix_tree_root *root,
1113 const struct radix_tree_iter *iter, void **slot, void *item)
1115 __radix_tree_replace(root, iter->node, slot, item, NULL, NULL);
1118 #ifdef CONFIG_RADIX_TREE_MULTIORDER
1120 * radix_tree_join - replace multiple entries with one multiorder entry
1121 * @root: radix tree root
1122 * @index: an index inside the new entry
1123 * @order: order of the new entry
1126 * Call this function to replace several entries with one larger entry.
1127 * The existing entries are presumed to not need freeing as a result of
1130 * The replacement entry will have all the tags set on it that were set
1131 * on any of the entries it is replacing.
1133 int radix_tree_join(struct radix_tree_root *root, unsigned long index,
1134 unsigned order, void *item)
1136 struct radix_tree_node *node;
1140 BUG_ON(radix_tree_is_internal_node(item));
1142 error = __radix_tree_create(root, index, order, &node, &slot);
1144 error = insert_entries(node, slot, item, order, true);
1152 * radix_tree_split - Split an entry into smaller entries
1153 * @root: radix tree root
1154 * @index: An index within the large entry
1155 * @order: Order of new entries
1157 * Call this function as the first step in replacing a multiorder entry
1158 * with several entries of lower order. After this function returns,
1159 * loop over the relevant portion of the tree using radix_tree_for_each_slot()
1160 * and call radix_tree_iter_replace() to set up each new entry.
1162 * The tags from this entry are replicated to all the new entries.
1164 * The radix tree should be locked against modification during the entire
1165 * replacement operation. Lock-free lookups will see RADIX_TREE_RETRY which
1166 * should prompt RCU walkers to restart the lookup from the root.
1168 int radix_tree_split(struct radix_tree_root *root, unsigned long index,
1171 struct radix_tree_node *parent, *node, *child;
1173 unsigned int offset, end;
1174 unsigned n, tag, tags = 0;
1176 if (!__radix_tree_lookup(root, index, &parent, &slot))
1181 offset = get_slot_offset(parent, slot);
1183 for (tag = 0; tag < RADIX_TREE_MAX_TAGS; tag++)
1184 if (tag_get(parent, tag, offset))
1187 for (end = offset + 1; end < RADIX_TREE_MAP_SIZE; end++) {
1188 if (!is_sibling_entry(parent, parent->slots[end]))
1190 for (tag = 0; tag < RADIX_TREE_MAX_TAGS; tag++)
1191 if (tags & (1 << tag))
1192 tag_set(parent, tag, end);
1193 /* rcu_assign_pointer ensures tags are set before RETRY */
1194 rcu_assign_pointer(parent->slots[end], RADIX_TREE_RETRY);
1196 rcu_assign_pointer(parent->slots[offset], RADIX_TREE_RETRY);
1197 parent->exceptional -= (end - offset);
1199 if (order == parent->shift)
1201 if (order > parent->shift) {
1202 while (offset < end)
1203 offset += insert_entries(parent, &parent->slots[offset],
1204 RADIX_TREE_RETRY, order, true);
1211 if (node->shift > order) {
1212 child = radix_tree_node_alloc(root);
1215 child->shift = node->shift - RADIX_TREE_MAP_SHIFT;
1216 child->offset = offset;
1218 child->parent = node;
1219 if (node != parent) {
1221 node->slots[offset] = node_to_entry(child);
1222 for (tag = 0; tag < RADIX_TREE_MAX_TAGS; tag++)
1223 if (tags & (1 << tag))
1224 tag_set(node, tag, offset);
1232 n = insert_entries(node, &node->slots[offset],
1233 RADIX_TREE_RETRY, order, false);
1234 BUG_ON(n > RADIX_TREE_MAP_SIZE);
1236 for (tag = 0; tag < RADIX_TREE_MAX_TAGS; tag++)
1237 if (tags & (1 << tag))
1238 tag_set(node, tag, offset);
1241 while (offset == RADIX_TREE_MAP_SIZE) {
1244 offset = node->offset;
1246 node = node->parent;
1247 rcu_assign_pointer(node->slots[offset],
1248 node_to_entry(child));
1251 if ((node == parent) && (offset == end))
1256 /* Shouldn't happen; did user forget to preload? */
1257 /* TODO: free all the allocated nodes */
1264 * radix_tree_tag_set - set a tag on a radix tree node
1265 * @root: radix tree root
1269 * Set the search tag (which must be < RADIX_TREE_MAX_TAGS)
1270 * corresponding to @index in the radix tree. From
1271 * the root all the way down to the leaf node.
1273 * Returns the address of the tagged item. Setting a tag on a not-present
1276 void *radix_tree_tag_set(struct radix_tree_root *root,
1277 unsigned long index, unsigned int tag)
1279 struct radix_tree_node *node, *parent;
1280 unsigned long maxindex;
1282 radix_tree_load_root(root, &node, &maxindex);
1283 BUG_ON(index > maxindex);
1285 while (radix_tree_is_internal_node(node)) {
1288 parent = entry_to_node(node);
1289 offset = radix_tree_descend(parent, &node, index);
1292 if (!tag_get(parent, tag, offset))
1293 tag_set(parent, tag, offset);
1296 /* set the root's tag bit */
1297 if (!root_tag_get(root, tag))
1298 root_tag_set(root, tag);
1302 EXPORT_SYMBOL(radix_tree_tag_set);
1304 static void node_tag_clear(struct radix_tree_root *root,
1305 struct radix_tree_node *node,
1306 unsigned int tag, unsigned int offset)
1309 if (!tag_get(node, tag, offset))
1311 tag_clear(node, tag, offset);
1312 if (any_tag_set(node, tag))
1315 offset = node->offset;
1316 node = node->parent;
1319 /* clear the root's tag bit */
1320 if (root_tag_get(root, tag))
1321 root_tag_clear(root, tag);
1324 static void node_tag_set(struct radix_tree_root *root,
1325 struct radix_tree_node *node,
1326 unsigned int tag, unsigned int offset)
1329 if (tag_get(node, tag, offset))
1331 tag_set(node, tag, offset);
1332 offset = node->offset;
1333 node = node->parent;
1336 if (!root_tag_get(root, tag))
1337 root_tag_set(root, tag);
1341 * radix_tree_iter_tag_set - set a tag on the current iterator entry
1342 * @root: radix tree root
1343 * @iter: iterator state
1346 void radix_tree_iter_tag_set(struct radix_tree_root *root,
1347 const struct radix_tree_iter *iter, unsigned int tag)
1349 node_tag_set(root, iter->node, tag, iter_offset(iter));
1353 * radix_tree_tag_clear - clear a tag on a radix tree node
1354 * @root: radix tree root
1358 * Clear the search tag (which must be < RADIX_TREE_MAX_TAGS)
1359 * corresponding to @index in the radix tree. If this causes
1360 * the leaf node to have no tags set then clear the tag in the
1361 * next-to-leaf node, etc.
1363 * Returns the address of the tagged item on success, else NULL. ie:
1364 * has the same return value and semantics as radix_tree_lookup().
1366 void *radix_tree_tag_clear(struct radix_tree_root *root,
1367 unsigned long index, unsigned int tag)
1369 struct radix_tree_node *node, *parent;
1370 unsigned long maxindex;
1371 int uninitialized_var(offset);
1373 radix_tree_load_root(root, &node, &maxindex);
1374 if (index > maxindex)
1379 while (radix_tree_is_internal_node(node)) {
1380 parent = entry_to_node(node);
1381 offset = radix_tree_descend(parent, &node, index);
1385 node_tag_clear(root, parent, tag, offset);
1389 EXPORT_SYMBOL(radix_tree_tag_clear);
1392 * radix_tree_tag_get - get a tag on a radix tree node
1393 * @root: radix tree root
1395 * @tag: tag index (< RADIX_TREE_MAX_TAGS)
1399 * 0: tag not present or not set
1402 * Note that the return value of this function may not be relied on, even if
1403 * the RCU lock is held, unless tag modification and node deletion are excluded
1406 int radix_tree_tag_get(struct radix_tree_root *root,
1407 unsigned long index, unsigned int tag)
1409 struct radix_tree_node *node, *parent;
1410 unsigned long maxindex;
1412 if (!root_tag_get(root, tag))
1415 radix_tree_load_root(root, &node, &maxindex);
1416 if (index > maxindex)
1421 while (radix_tree_is_internal_node(node)) {
1424 parent = entry_to_node(node);
1425 offset = radix_tree_descend(parent, &node, index);
1429 if (!tag_get(parent, tag, offset))
1431 if (node == RADIX_TREE_RETRY)
1437 EXPORT_SYMBOL(radix_tree_tag_get);
1439 static inline void __set_iter_shift(struct radix_tree_iter *iter,
1442 #ifdef CONFIG_RADIX_TREE_MULTIORDER
1443 iter->shift = shift;
1447 /* Construct iter->tags bit-mask from node->tags[tag] array */
1448 static void set_iter_tags(struct radix_tree_iter *iter,
1449 struct radix_tree_node *node, unsigned offset,
1452 unsigned tag_long = offset / BITS_PER_LONG;
1453 unsigned tag_bit = offset % BITS_PER_LONG;
1455 iter->tags = node->tags[tag][tag_long] >> tag_bit;
1457 /* This never happens if RADIX_TREE_TAG_LONGS == 1 */
1458 if (tag_long < RADIX_TREE_TAG_LONGS - 1) {
1459 /* Pick tags from next element */
1461 iter->tags |= node->tags[tag][tag_long + 1] <<
1462 (BITS_PER_LONG - tag_bit);
1463 /* Clip chunk size, here only BITS_PER_LONG tags */
1464 iter->next_index = __radix_tree_iter_add(iter, BITS_PER_LONG);
1468 #ifdef CONFIG_RADIX_TREE_MULTIORDER
1469 static void **skip_siblings(struct radix_tree_node **nodep,
1470 void **slot, struct radix_tree_iter *iter)
1472 void *sib = node_to_entry(slot - 1);
1474 while (iter->index < iter->next_index) {
1475 *nodep = rcu_dereference_raw(*slot);
1476 if (*nodep && *nodep != sib)
1479 iter->index = __radix_tree_iter_add(iter, 1);
1487 void ** __radix_tree_next_slot(void **slot, struct radix_tree_iter *iter,
1490 unsigned tag = flags & RADIX_TREE_ITER_TAG_MASK;
1491 struct radix_tree_node *node = rcu_dereference_raw(*slot);
1493 slot = skip_siblings(&node, slot, iter);
1495 while (radix_tree_is_internal_node(node)) {
1497 unsigned long next_index;
1499 if (node == RADIX_TREE_RETRY)
1501 node = entry_to_node(node);
1503 iter->shift = node->shift;
1505 if (flags & RADIX_TREE_ITER_TAGGED) {
1506 offset = radix_tree_find_next_bit(node, tag, 0);
1507 if (offset == RADIX_TREE_MAP_SIZE)
1509 slot = &node->slots[offset];
1510 iter->index = __radix_tree_iter_add(iter, offset);
1511 set_iter_tags(iter, node, offset, tag);
1512 node = rcu_dereference_raw(*slot);
1515 slot = &node->slots[0];
1517 node = rcu_dereference_raw(*slot);
1522 if (offset == RADIX_TREE_MAP_SIZE)
1525 iter->index = __radix_tree_iter_add(iter, offset);
1527 if ((flags & RADIX_TREE_ITER_CONTIG) && (offset > 0))
1529 next_index = (iter->index | shift_maxindex(iter->shift)) + 1;
1530 if (next_index < iter->next_index)
1531 iter->next_index = next_index;
1536 iter->next_index = 0;
1539 EXPORT_SYMBOL(__radix_tree_next_slot);
1541 static void **skip_siblings(struct radix_tree_node **nodep,
1542 void **slot, struct radix_tree_iter *iter)
1548 void **radix_tree_iter_resume(void **slot, struct radix_tree_iter *iter)
1550 struct radix_tree_node *node;
1553 iter->index = __radix_tree_iter_add(iter, 1);
1554 node = rcu_dereference_raw(*slot);
1555 skip_siblings(&node, slot, iter);
1556 iter->next_index = iter->index;
1560 EXPORT_SYMBOL(radix_tree_iter_resume);
1563 * radix_tree_next_chunk - find next chunk of slots for iteration
1565 * @root: radix tree root
1566 * @iter: iterator state
1567 * @flags: RADIX_TREE_ITER_* flags and tag index
1568 * Returns: pointer to chunk first slot, or NULL if iteration is over
1570 void **radix_tree_next_chunk(struct radix_tree_root *root,
1571 struct radix_tree_iter *iter, unsigned flags)
1573 unsigned tag = flags & RADIX_TREE_ITER_TAG_MASK;
1574 struct radix_tree_node *node, *child;
1575 unsigned long index, offset, maxindex;
1577 if ((flags & RADIX_TREE_ITER_TAGGED) && !root_tag_get(root, tag))
1581 * Catch next_index overflow after ~0UL. iter->index never overflows
1582 * during iterating; it can be zero only at the beginning.
1583 * And we cannot overflow iter->next_index in a single step,
1584 * because RADIX_TREE_MAP_SHIFT < BITS_PER_LONG.
1586 * This condition also used by radix_tree_next_slot() to stop
1587 * contiguous iterating, and forbid switching to the next chunk.
1589 index = iter->next_index;
1590 if (!index && iter->index)
1594 radix_tree_load_root(root, &child, &maxindex);
1595 if (index > maxindex)
1600 if (!radix_tree_is_internal_node(child)) {
1601 /* Single-slot tree */
1602 iter->index = index;
1603 iter->next_index = maxindex + 1;
1606 __set_iter_shift(iter, 0);
1607 return (void **)&root->rnode;
1611 node = entry_to_node(child);
1612 offset = radix_tree_descend(node, &child, index);
1614 if ((flags & RADIX_TREE_ITER_TAGGED) ?
1615 !tag_get(node, tag, offset) : !child) {
1617 if (flags & RADIX_TREE_ITER_CONTIG)
1620 if (flags & RADIX_TREE_ITER_TAGGED)
1621 offset = radix_tree_find_next_bit(node, tag,
1624 while (++offset < RADIX_TREE_MAP_SIZE) {
1625 void *slot = node->slots[offset];
1626 if (is_sibling_entry(node, slot))
1631 index &= ~node_maxindex(node);
1632 index += offset << node->shift;
1633 /* Overflow after ~0UL */
1636 if (offset == RADIX_TREE_MAP_SIZE)
1638 child = rcu_dereference_raw(node->slots[offset]);
1643 if (child == RADIX_TREE_RETRY)
1645 } while (radix_tree_is_internal_node(child));
1647 /* Update the iterator state */
1648 iter->index = (index &~ node_maxindex(node)) | (offset << node->shift);
1649 iter->next_index = (index | node_maxindex(node)) + 1;
1651 __set_iter_shift(iter, node->shift);
1653 if (flags & RADIX_TREE_ITER_TAGGED)
1654 set_iter_tags(iter, node, offset, tag);
1656 return node->slots + offset;
1658 EXPORT_SYMBOL(radix_tree_next_chunk);
1661 * radix_tree_gang_lookup - perform multiple lookup on a radix tree
1662 * @root: radix tree root
1663 * @results: where the results of the lookup are placed
1664 * @first_index: start the lookup from this key
1665 * @max_items: place up to this many items at *results
1667 * Performs an index-ascending scan of the tree for present items. Places
1668 * them at *@results and returns the number of items which were placed at
1671 * The implementation is naive.
1673 * Like radix_tree_lookup, radix_tree_gang_lookup may be called under
1674 * rcu_read_lock. In this case, rather than the returned results being
1675 * an atomic snapshot of the tree at a single point in time, the
1676 * semantics of an RCU protected gang lookup are as though multiple
1677 * radix_tree_lookups have been issued in individual locks, and results
1678 * stored in 'results'.
1681 radix_tree_gang_lookup(struct radix_tree_root *root, void **results,
1682 unsigned long first_index, unsigned int max_items)
1684 struct radix_tree_iter iter;
1686 unsigned int ret = 0;
1688 if (unlikely(!max_items))
1691 radix_tree_for_each_slot(slot, root, &iter, first_index) {
1692 results[ret] = rcu_dereference_raw(*slot);
1695 if (radix_tree_is_internal_node(results[ret])) {
1696 slot = radix_tree_iter_retry(&iter);
1699 if (++ret == max_items)
1705 EXPORT_SYMBOL(radix_tree_gang_lookup);
1708 * radix_tree_gang_lookup_slot - perform multiple slot lookup on radix tree
1709 * @root: radix tree root
1710 * @results: where the results of the lookup are placed
1711 * @indices: where their indices should be placed (but usually NULL)
1712 * @first_index: start the lookup from this key
1713 * @max_items: place up to this many items at *results
1715 * Performs an index-ascending scan of the tree for present items. Places
1716 * their slots at *@results and returns the number of items which were
1717 * placed at *@results.
1719 * The implementation is naive.
1721 * Like radix_tree_gang_lookup as far as RCU and locking goes. Slots must
1722 * be dereferenced with radix_tree_deref_slot, and if using only RCU
1723 * protection, radix_tree_deref_slot may fail requiring a retry.
1726 radix_tree_gang_lookup_slot(struct radix_tree_root *root,
1727 void ***results, unsigned long *indices,
1728 unsigned long first_index, unsigned int max_items)
1730 struct radix_tree_iter iter;
1732 unsigned int ret = 0;
1734 if (unlikely(!max_items))
1737 radix_tree_for_each_slot(slot, root, &iter, first_index) {
1738 results[ret] = slot;
1740 indices[ret] = iter.index;
1741 if (++ret == max_items)
1747 EXPORT_SYMBOL(radix_tree_gang_lookup_slot);
1750 * radix_tree_gang_lookup_tag - perform multiple lookup on a radix tree
1752 * @root: radix tree root
1753 * @results: where the results of the lookup are placed
1754 * @first_index: start the lookup from this key
1755 * @max_items: place up to this many items at *results
1756 * @tag: the tag index (< RADIX_TREE_MAX_TAGS)
1758 * Performs an index-ascending scan of the tree for present items which
1759 * have the tag indexed by @tag set. Places the items at *@results and
1760 * returns the number of items which were placed at *@results.
1763 radix_tree_gang_lookup_tag(struct radix_tree_root *root, void **results,
1764 unsigned long first_index, unsigned int max_items,
1767 struct radix_tree_iter iter;
1769 unsigned int ret = 0;
1771 if (unlikely(!max_items))
1774 radix_tree_for_each_tagged(slot, root, &iter, first_index, tag) {
1775 results[ret] = rcu_dereference_raw(*slot);
1778 if (radix_tree_is_internal_node(results[ret])) {
1779 slot = radix_tree_iter_retry(&iter);
1782 if (++ret == max_items)
1788 EXPORT_SYMBOL(radix_tree_gang_lookup_tag);
1791 * radix_tree_gang_lookup_tag_slot - perform multiple slot lookup on a
1792 * radix tree based on a tag
1793 * @root: radix tree root
1794 * @results: where the results of the lookup are placed
1795 * @first_index: start the lookup from this key
1796 * @max_items: place up to this many items at *results
1797 * @tag: the tag index (< RADIX_TREE_MAX_TAGS)
1799 * Performs an index-ascending scan of the tree for present items which
1800 * have the tag indexed by @tag set. Places the slots at *@results and
1801 * returns the number of slots which were placed at *@results.
1804 radix_tree_gang_lookup_tag_slot(struct radix_tree_root *root, void ***results,
1805 unsigned long first_index, unsigned int max_items,
1808 struct radix_tree_iter iter;
1810 unsigned int ret = 0;
1812 if (unlikely(!max_items))
1815 radix_tree_for_each_tagged(slot, root, &iter, first_index, tag) {
1816 results[ret] = slot;
1817 if (++ret == max_items)
1823 EXPORT_SYMBOL(radix_tree_gang_lookup_tag_slot);
1826 * __radix_tree_delete_node - try to free node after clearing a slot
1827 * @root: radix tree root
1828 * @node: node containing @index
1830 * After clearing the slot at @index in @node from radix tree
1831 * rooted at @root, call this function to attempt freeing the
1832 * node and shrinking the tree.
1834 void __radix_tree_delete_node(struct radix_tree_root *root,
1835 struct radix_tree_node *node)
1837 delete_node(root, node, NULL, NULL);
1841 * radix_tree_delete_item - delete an item from a radix tree
1842 * @root: radix tree root
1844 * @item: expected item
1846 * Remove @item at @index from the radix tree rooted at @root.
1848 * Returns the address of the deleted item, or NULL if it was not present
1849 * or the entry at the given @index was not @item.
1851 void *radix_tree_delete_item(struct radix_tree_root *root,
1852 unsigned long index, void *item)
1854 struct radix_tree_node *node;
1855 unsigned int offset;
1860 entry = __radix_tree_lookup(root, index, &node, &slot);
1864 if (item && entry != item)
1868 root_tag_clear_all(root);
1873 offset = get_slot_offset(node, slot);
1875 /* Clear all tags associated with the item to be deleted. */
1876 for (tag = 0; tag < RADIX_TREE_MAX_TAGS; tag++)
1877 node_tag_clear(root, node, tag, offset);
1879 __radix_tree_replace(root, node, slot, NULL, NULL, NULL);
1883 EXPORT_SYMBOL(radix_tree_delete_item);
1886 * radix_tree_delete - delete an item from a radix tree
1887 * @root: radix tree root
1890 * Remove the item at @index from the radix tree rooted at @root.
1892 * Returns the address of the deleted item, or NULL if it was not present.
1894 void *radix_tree_delete(struct radix_tree_root *root, unsigned long index)
1896 return radix_tree_delete_item(root, index, NULL);
1898 EXPORT_SYMBOL(radix_tree_delete);
1900 void radix_tree_clear_tags(struct radix_tree_root *root,
1901 struct radix_tree_node *node,
1905 unsigned int tag, offset = get_slot_offset(node, slot);
1906 for (tag = 0; tag < RADIX_TREE_MAX_TAGS; tag++)
1907 node_tag_clear(root, node, tag, offset);
1909 /* Clear root node tags */
1910 root->gfp_mask &= __GFP_BITS_MASK;
1915 * radix_tree_tagged - test whether any items in the tree are tagged
1916 * @root: radix tree root
1919 int radix_tree_tagged(struct radix_tree_root *root, unsigned int tag)
1921 return root_tag_get(root, tag);
1923 EXPORT_SYMBOL(radix_tree_tagged);
1926 radix_tree_node_ctor(void *arg)
1928 struct radix_tree_node *node = arg;
1930 memset(node, 0, sizeof(*node));
1931 INIT_LIST_HEAD(&node->private_list);
1934 static __init unsigned long __maxindex(unsigned int height)
1936 unsigned int width = height * RADIX_TREE_MAP_SHIFT;
1937 int shift = RADIX_TREE_INDEX_BITS - width;
1941 if (shift >= BITS_PER_LONG)
1943 return ~0UL >> shift;
1946 static __init void radix_tree_init_maxnodes(void)
1948 unsigned long height_to_maxindex[RADIX_TREE_MAX_PATH + 1];
1951 for (i = 0; i < ARRAY_SIZE(height_to_maxindex); i++)
1952 height_to_maxindex[i] = __maxindex(i);
1953 for (i = 0; i < ARRAY_SIZE(height_to_maxnodes); i++) {
1954 for (j = i; j > 0; j--)
1955 height_to_maxnodes[i] += height_to_maxindex[j - 1] + 1;
1959 static int radix_tree_cpu_dead(unsigned int cpu)
1961 struct radix_tree_preload *rtp;
1962 struct radix_tree_node *node;
1964 /* Free per-cpu pool of preloaded nodes */
1965 rtp = &per_cpu(radix_tree_preloads, cpu);
1968 rtp->nodes = node->private_data;
1969 kmem_cache_free(radix_tree_node_cachep, node);
1975 void __init radix_tree_init(void)
1978 radix_tree_node_cachep = kmem_cache_create("radix_tree_node",
1979 sizeof(struct radix_tree_node), 0,
1980 SLAB_PANIC | SLAB_RECLAIM_ACCOUNT,
1981 radix_tree_node_ctor);
1982 radix_tree_init_maxnodes();
1983 ret = cpuhp_setup_state_nocalls(CPUHP_RADIX_DEAD, "lib/radix:dead",
1984 NULL, radix_tree_cpu_dead);