1 // SPDX-License-Identifier: GPL-2.0-or-later
3 * Copyright (C) 2001 Momchil Velikov
4 * Portions Copyright (C) 2001 Christoph Hellwig
5 * Copyright (C) 2005 SGI, Christoph Lameter
6 * Copyright (C) 2006 Nick Piggin
7 * Copyright (C) 2012 Konstantin Khlebnikov
8 * Copyright (C) 2016 Intel, Matthew Wilcox
9 * Copyright (C) 2016 Intel, Ross Zwisler
12 #include <linux/bitmap.h>
13 #include <linux/bitops.h>
14 #include <linux/bug.h>
15 #include <linux/cpu.h>
16 #include <linux/errno.h>
17 #include <linux/export.h>
18 #include <linux/idr.h>
19 #include <linux/init.h>
20 #include <linux/kernel.h>
21 #include <linux/kmemleak.h>
22 #include <linux/percpu.h>
23 #include <linux/preempt.h> /* in_interrupt() */
24 #include <linux/radix-tree.h>
25 #include <linux/rcupdate.h>
26 #include <linux/slab.h>
27 #include <linux/string.h>
28 #include <linux/xarray.h>
32 * Radix tree node cache.
34 struct kmem_cache *radix_tree_node_cachep;
37 * The radix tree is variable-height, so an insert operation not only has
38 * to build the branch to its corresponding item, it also has to build the
39 * branch to existing items if the size has to be increased (by
42 * The worst case is a zero height tree with just a single item at index 0,
43 * and then inserting an item at index ULONG_MAX. This requires 2 new branches
44 * of RADIX_TREE_MAX_PATH size to be created, with only the root node shared.
47 #define RADIX_TREE_PRELOAD_SIZE (RADIX_TREE_MAX_PATH * 2 - 1)
50 * The IDR does not have to be as high as the radix tree since it uses
51 * signed integers, not unsigned longs.
53 #define IDR_INDEX_BITS (8 /* CHAR_BIT */ * sizeof(int) - 1)
54 #define IDR_MAX_PATH (DIV_ROUND_UP(IDR_INDEX_BITS, \
55 RADIX_TREE_MAP_SHIFT))
56 #define IDR_PRELOAD_SIZE (IDR_MAX_PATH * 2 - 1)
59 * Per-cpu pool of preloaded nodes
61 struct radix_tree_preload {
63 /* nodes->parent points to next preallocated node */
64 struct radix_tree_node *nodes;
66 static DEFINE_PER_CPU(struct radix_tree_preload, radix_tree_preloads) = { 0, };
68 static inline struct radix_tree_node *entry_to_node(void *ptr)
70 return (void *)((unsigned long)ptr & ~RADIX_TREE_INTERNAL_NODE);
73 static inline void *node_to_entry(void *ptr)
75 return (void *)((unsigned long)ptr | RADIX_TREE_INTERNAL_NODE);
78 #define RADIX_TREE_RETRY XA_RETRY_ENTRY
80 static inline unsigned long
81 get_slot_offset(const struct radix_tree_node *parent, void __rcu **slot)
83 return parent ? slot - parent->slots : 0;
86 static unsigned int radix_tree_descend(const struct radix_tree_node *parent,
87 struct radix_tree_node **nodep, unsigned long index)
89 unsigned int offset = (index >> parent->shift) & RADIX_TREE_MAP_MASK;
90 void __rcu **entry = rcu_dereference_raw(parent->slots[offset]);
92 *nodep = (void *)entry;
96 static inline gfp_t root_gfp_mask(const struct radix_tree_root *root)
98 return root->xa_flags & (__GFP_BITS_MASK & ~GFP_ZONEMASK);
101 static inline void tag_set(struct radix_tree_node *node, unsigned int tag,
104 __set_bit(offset, node->tags[tag]);
107 static inline void tag_clear(struct radix_tree_node *node, unsigned int tag,
110 __clear_bit(offset, node->tags[tag]);
113 static inline int tag_get(const struct radix_tree_node *node, unsigned int tag,
116 return test_bit(offset, node->tags[tag]);
119 static inline void root_tag_set(struct radix_tree_root *root, unsigned tag)
121 root->xa_flags |= (__force gfp_t)(1 << (tag + ROOT_TAG_SHIFT));
124 static inline void root_tag_clear(struct radix_tree_root *root, unsigned tag)
126 root->xa_flags &= (__force gfp_t)~(1 << (tag + ROOT_TAG_SHIFT));
129 static inline void root_tag_clear_all(struct radix_tree_root *root)
131 root->xa_flags &= (__force gfp_t)((1 << ROOT_TAG_SHIFT) - 1);
134 static inline int root_tag_get(const struct radix_tree_root *root, unsigned tag)
136 return (__force int)root->xa_flags & (1 << (tag + ROOT_TAG_SHIFT));
139 static inline unsigned root_tags_get(const struct radix_tree_root *root)
141 return (__force unsigned)root->xa_flags >> ROOT_TAG_SHIFT;
144 static inline bool is_idr(const struct radix_tree_root *root)
146 return !!(root->xa_flags & ROOT_IS_IDR);
150 * Returns 1 if any slot in the node has this tag set.
151 * Otherwise returns 0.
153 static inline int any_tag_set(const struct radix_tree_node *node,
157 for (idx = 0; idx < RADIX_TREE_TAG_LONGS; idx++) {
158 if (node->tags[tag][idx])
164 static inline void all_tag_set(struct radix_tree_node *node, unsigned int tag)
166 bitmap_fill(node->tags[tag], RADIX_TREE_MAP_SIZE);
170 * radix_tree_find_next_bit - find the next set bit in a memory region
172 * @addr: The address to base the search on
173 * @size: The bitmap size in bits
174 * @offset: The bitnumber to start searching at
176 * Unrollable variant of find_next_bit() for constant size arrays.
177 * Tail bits starting from size to roundup(size, BITS_PER_LONG) must be zero.
178 * Returns next bit offset, or size if nothing found.
180 static __always_inline unsigned long
181 radix_tree_find_next_bit(struct radix_tree_node *node, unsigned int tag,
182 unsigned long offset)
184 const unsigned long *addr = node->tags[tag];
186 if (offset < RADIX_TREE_MAP_SIZE) {
189 addr += offset / BITS_PER_LONG;
190 tmp = *addr >> (offset % BITS_PER_LONG);
192 return __ffs(tmp) + offset;
193 offset = (offset + BITS_PER_LONG) & ~(BITS_PER_LONG - 1);
194 while (offset < RADIX_TREE_MAP_SIZE) {
197 return __ffs(tmp) + offset;
198 offset += BITS_PER_LONG;
201 return RADIX_TREE_MAP_SIZE;
204 static unsigned int iter_offset(const struct radix_tree_iter *iter)
206 return iter->index & RADIX_TREE_MAP_MASK;
210 * The maximum index which can be stored in a radix tree
212 static inline unsigned long shift_maxindex(unsigned int shift)
214 return (RADIX_TREE_MAP_SIZE << shift) - 1;
217 static inline unsigned long node_maxindex(const struct radix_tree_node *node)
219 return shift_maxindex(node->shift);
222 static unsigned long next_index(unsigned long index,
223 const struct radix_tree_node *node,
224 unsigned long offset)
226 return (index & ~node_maxindex(node)) + (offset << node->shift);
230 * This assumes that the caller has performed appropriate preallocation, and
231 * that the caller has pinned this thread of control to the current CPU.
233 static struct radix_tree_node *
234 radix_tree_node_alloc(gfp_t gfp_mask, struct radix_tree_node *parent,
235 struct radix_tree_root *root,
236 unsigned int shift, unsigned int offset,
237 unsigned int count, unsigned int nr_values)
239 struct radix_tree_node *ret = NULL;
242 * Preload code isn't irq safe and it doesn't make sense to use
243 * preloading during an interrupt anyway as all the allocations have
244 * to be atomic. So just do normal allocation when in interrupt.
246 if (!gfpflags_allow_blocking(gfp_mask) && !in_interrupt()) {
247 struct radix_tree_preload *rtp;
250 * Even if the caller has preloaded, try to allocate from the
251 * cache first for the new node to get accounted to the memory
254 ret = kmem_cache_alloc(radix_tree_node_cachep,
255 gfp_mask | __GFP_NOWARN);
260 * Provided the caller has preloaded here, we will always
261 * succeed in getting a node here (and never reach
264 rtp = this_cpu_ptr(&radix_tree_preloads);
267 rtp->nodes = ret->parent;
271 * Update the allocation stack trace as this is more useful
274 kmemleak_update_trace(ret);
277 ret = kmem_cache_alloc(radix_tree_node_cachep, gfp_mask);
279 BUG_ON(radix_tree_is_internal_node(ret));
282 ret->offset = offset;
284 ret->nr_values = nr_values;
285 ret->parent = parent;
291 void radix_tree_node_rcu_free(struct rcu_head *head)
293 struct radix_tree_node *node =
294 container_of(head, struct radix_tree_node, rcu_head);
297 * Must only free zeroed nodes into the slab. We can be left with
298 * non-NULL entries by radix_tree_free_nodes, so clear the entries
301 memset(node->slots, 0, sizeof(node->slots));
302 memset(node->tags, 0, sizeof(node->tags));
303 INIT_LIST_HEAD(&node->private_list);
305 kmem_cache_free(radix_tree_node_cachep, node);
309 radix_tree_node_free(struct radix_tree_node *node)
311 call_rcu(&node->rcu_head, radix_tree_node_rcu_free);
315 * Load up this CPU's radix_tree_node buffer with sufficient objects to
316 * ensure that the addition of a single element in the tree cannot fail. On
317 * success, return zero, with preemption disabled. On error, return -ENOMEM
318 * with preemption not disabled.
320 * To make use of this facility, the radix tree must be initialised without
321 * __GFP_DIRECT_RECLAIM being passed to INIT_RADIX_TREE().
323 static __must_check int __radix_tree_preload(gfp_t gfp_mask, unsigned nr)
325 struct radix_tree_preload *rtp;
326 struct radix_tree_node *node;
330 * Nodes preloaded by one cgroup can be be used by another cgroup, so
331 * they should never be accounted to any particular memory cgroup.
333 gfp_mask &= ~__GFP_ACCOUNT;
336 rtp = this_cpu_ptr(&radix_tree_preloads);
337 while (rtp->nr < nr) {
339 node = kmem_cache_alloc(radix_tree_node_cachep, gfp_mask);
343 rtp = this_cpu_ptr(&radix_tree_preloads);
345 node->parent = rtp->nodes;
349 kmem_cache_free(radix_tree_node_cachep, node);
358 * Load up this CPU's radix_tree_node buffer with sufficient objects to
359 * ensure that the addition of a single element in the tree cannot fail. On
360 * success, return zero, with preemption disabled. On error, return -ENOMEM
361 * with preemption not disabled.
363 * To make use of this facility, the radix tree must be initialised without
364 * __GFP_DIRECT_RECLAIM being passed to INIT_RADIX_TREE().
366 int radix_tree_preload(gfp_t gfp_mask)
368 /* Warn on non-sensical use... */
369 WARN_ON_ONCE(!gfpflags_allow_blocking(gfp_mask));
370 return __radix_tree_preload(gfp_mask, RADIX_TREE_PRELOAD_SIZE);
372 EXPORT_SYMBOL(radix_tree_preload);
375 * The same as above function, except we don't guarantee preloading happens.
376 * We do it, if we decide it helps. On success, return zero with preemption
377 * disabled. On error, return -ENOMEM with preemption not disabled.
379 int radix_tree_maybe_preload(gfp_t gfp_mask)
381 if (gfpflags_allow_blocking(gfp_mask))
382 return __radix_tree_preload(gfp_mask, RADIX_TREE_PRELOAD_SIZE);
383 /* Preloading doesn't help anything with this gfp mask, skip it */
387 EXPORT_SYMBOL(radix_tree_maybe_preload);
389 static unsigned radix_tree_load_root(const struct radix_tree_root *root,
390 struct radix_tree_node **nodep, unsigned long *maxindex)
392 struct radix_tree_node *node = rcu_dereference_raw(root->xa_head);
396 if (likely(radix_tree_is_internal_node(node))) {
397 node = entry_to_node(node);
398 *maxindex = node_maxindex(node);
399 return node->shift + RADIX_TREE_MAP_SHIFT;
407 * Extend a radix tree so it can store key @index.
409 static int radix_tree_extend(struct radix_tree_root *root, gfp_t gfp,
410 unsigned long index, unsigned int shift)
413 unsigned int maxshift;
416 /* Figure out what the shift should be. */
418 while (index > shift_maxindex(maxshift))
419 maxshift += RADIX_TREE_MAP_SHIFT;
421 entry = rcu_dereference_raw(root->xa_head);
422 if (!entry && (!is_idr(root) || root_tag_get(root, IDR_FREE)))
426 struct radix_tree_node *node = radix_tree_node_alloc(gfp, NULL,
427 root, shift, 0, 1, 0);
432 all_tag_set(node, IDR_FREE);
433 if (!root_tag_get(root, IDR_FREE)) {
434 tag_clear(node, IDR_FREE, 0);
435 root_tag_set(root, IDR_FREE);
438 /* Propagate the aggregated tag info to the new child */
439 for (tag = 0; tag < RADIX_TREE_MAX_TAGS; tag++) {
440 if (root_tag_get(root, tag))
441 tag_set(node, tag, 0);
445 BUG_ON(shift > BITS_PER_LONG);
446 if (radix_tree_is_internal_node(entry)) {
447 entry_to_node(entry)->parent = node;
448 } else if (xa_is_value(entry)) {
449 /* Moving a value entry root->xa_head to a node */
453 * entry was already in the radix tree, so we do not need
454 * rcu_assign_pointer here
456 node->slots[0] = (void __rcu *)entry;
457 entry = node_to_entry(node);
458 rcu_assign_pointer(root->xa_head, entry);
459 shift += RADIX_TREE_MAP_SHIFT;
460 } while (shift <= maxshift);
462 return maxshift + RADIX_TREE_MAP_SHIFT;
466 * radix_tree_shrink - shrink radix tree to minimum height
467 * @root radix tree root
469 static inline bool radix_tree_shrink(struct radix_tree_root *root)
474 struct radix_tree_node *node = rcu_dereference_raw(root->xa_head);
475 struct radix_tree_node *child;
477 if (!radix_tree_is_internal_node(node))
479 node = entry_to_node(node);
482 * The candidate node has more than one child, or its child
483 * is not at the leftmost slot, we cannot shrink.
485 if (node->count != 1)
487 child = rcu_dereference_raw(node->slots[0]);
492 * For an IDR, we must not shrink entry 0 into the root in
493 * case somebody calls idr_replace() with a pointer that
494 * appears to be an internal entry
496 if (!node->shift && is_idr(root))
499 if (radix_tree_is_internal_node(child))
500 entry_to_node(child)->parent = NULL;
503 * We don't need rcu_assign_pointer(), since we are simply
504 * moving the node from one part of the tree to another: if it
505 * was safe to dereference the old pointer to it
506 * (node->slots[0]), it will be safe to dereference the new
507 * one (root->xa_head) as far as dependent read barriers go.
509 root->xa_head = (void __rcu *)child;
510 if (is_idr(root) && !tag_get(node, IDR_FREE, 0))
511 root_tag_clear(root, IDR_FREE);
514 * We have a dilemma here. The node's slot[0] must not be
515 * NULLed in case there are concurrent lookups expecting to
516 * find the item. However if this was a bottom-level node,
517 * then it may be subject to the slot pointer being visible
518 * to callers dereferencing it. If item corresponding to
519 * slot[0] is subsequently deleted, these callers would expect
520 * their slot to become empty sooner or later.
522 * For example, lockless pagecache will look up a slot, deref
523 * the page pointer, and if the page has 0 refcount it means it
524 * was concurrently deleted from pagecache so try the deref
525 * again. Fortunately there is already a requirement for logic
526 * to retry the entire slot lookup -- the indirect pointer
527 * problem (replacing direct root node with an indirect pointer
528 * also results in a stale slot). So tag the slot as indirect
529 * to force callers to retry.
532 if (!radix_tree_is_internal_node(child)) {
533 node->slots[0] = (void __rcu *)RADIX_TREE_RETRY;
536 WARN_ON_ONCE(!list_empty(&node->private_list));
537 radix_tree_node_free(node);
544 static bool delete_node(struct radix_tree_root *root,
545 struct radix_tree_node *node)
547 bool deleted = false;
550 struct radix_tree_node *parent;
553 if (node_to_entry(node) ==
554 rcu_dereference_raw(root->xa_head))
555 deleted |= radix_tree_shrink(root);
559 parent = node->parent;
561 parent->slots[node->offset] = NULL;
565 * Shouldn't the tags already have all been cleared
569 root_tag_clear_all(root);
570 root->xa_head = NULL;
573 WARN_ON_ONCE(!list_empty(&node->private_list));
574 radix_tree_node_free(node);
584 * __radix_tree_create - create a slot in a radix tree
585 * @root: radix tree root
587 * @nodep: returns node
588 * @slotp: returns slot
590 * Create, if necessary, and return the node and slot for an item
591 * at position @index in the radix tree @root.
593 * Until there is more than one item in the tree, no nodes are
594 * allocated and @root->xa_head is used as a direct slot instead of
595 * pointing to a node, in which case *@nodep will be NULL.
597 * Returns -ENOMEM, or 0 for success.
599 static int __radix_tree_create(struct radix_tree_root *root,
600 unsigned long index, struct radix_tree_node **nodep,
603 struct radix_tree_node *node = NULL, *child;
604 void __rcu **slot = (void __rcu **)&root->xa_head;
605 unsigned long maxindex;
606 unsigned int shift, offset = 0;
607 unsigned long max = index;
608 gfp_t gfp = root_gfp_mask(root);
610 shift = radix_tree_load_root(root, &child, &maxindex);
612 /* Make sure the tree is high enough. */
613 if (max > maxindex) {
614 int error = radix_tree_extend(root, gfp, max, shift);
618 child = rcu_dereference_raw(root->xa_head);
622 shift -= RADIX_TREE_MAP_SHIFT;
624 /* Have to add a child node. */
625 child = radix_tree_node_alloc(gfp, node, root, shift,
629 rcu_assign_pointer(*slot, node_to_entry(child));
632 } else if (!radix_tree_is_internal_node(child))
635 /* Go a level down */
636 node = entry_to_node(child);
637 offset = radix_tree_descend(node, &child, index);
638 slot = &node->slots[offset];
649 * Free any nodes below this node. The tree is presumed to not need
650 * shrinking, and any user data in the tree is presumed to not need a
651 * destructor called on it. If we need to add a destructor, we can
652 * add that functionality later. Note that we may not clear tags or
653 * slots from the tree as an RCU walker may still have a pointer into
654 * this subtree. We could replace the entries with RADIX_TREE_RETRY,
655 * but we'll still have to clear those in rcu_free.
657 static void radix_tree_free_nodes(struct radix_tree_node *node)
660 struct radix_tree_node *child = entry_to_node(node);
663 void *entry = rcu_dereference_raw(child->slots[offset]);
664 if (xa_is_node(entry) && child->shift) {
665 child = entry_to_node(entry);
670 while (offset == RADIX_TREE_MAP_SIZE) {
671 struct radix_tree_node *old = child;
672 offset = child->offset + 1;
673 child = child->parent;
674 WARN_ON_ONCE(!list_empty(&old->private_list));
675 radix_tree_node_free(old);
676 if (old == entry_to_node(node))
682 static inline int insert_entries(struct radix_tree_node *node,
683 void __rcu **slot, void *item, bool replace)
687 rcu_assign_pointer(*slot, item);
690 if (xa_is_value(item))
697 * __radix_tree_insert - insert into a radix tree
698 * @root: radix tree root
700 * @item: item to insert
702 * Insert an item into the radix tree at position @index.
704 int radix_tree_insert(struct radix_tree_root *root, unsigned long index,
707 struct radix_tree_node *node;
711 BUG_ON(radix_tree_is_internal_node(item));
713 error = __radix_tree_create(root, index, &node, &slot);
717 error = insert_entries(node, slot, item, false);
722 unsigned offset = get_slot_offset(node, slot);
723 BUG_ON(tag_get(node, 0, offset));
724 BUG_ON(tag_get(node, 1, offset));
725 BUG_ON(tag_get(node, 2, offset));
727 BUG_ON(root_tags_get(root));
732 EXPORT_SYMBOL(radix_tree_insert);
735 * __radix_tree_lookup - lookup an item in a radix tree
736 * @root: radix tree root
738 * @nodep: returns node
739 * @slotp: returns slot
741 * Lookup and return the item at position @index in the radix
744 * Until there is more than one item in the tree, no nodes are
745 * allocated and @root->xa_head is used as a direct slot instead of
746 * pointing to a node, in which case *@nodep will be NULL.
748 void *__radix_tree_lookup(const struct radix_tree_root *root,
749 unsigned long index, struct radix_tree_node **nodep,
752 struct radix_tree_node *node, *parent;
753 unsigned long maxindex;
758 slot = (void __rcu **)&root->xa_head;
759 radix_tree_load_root(root, &node, &maxindex);
760 if (index > maxindex)
763 while (radix_tree_is_internal_node(node)) {
766 parent = entry_to_node(node);
767 offset = radix_tree_descend(parent, &node, index);
768 slot = parent->slots + offset;
769 if (node == RADIX_TREE_RETRY)
771 if (parent->shift == 0)
783 * radix_tree_lookup_slot - lookup a slot in a radix tree
784 * @root: radix tree root
787 * Returns: the slot corresponding to the position @index in the
788 * radix tree @root. This is useful for update-if-exists operations.
790 * This function can be called under rcu_read_lock iff the slot is not
791 * modified by radix_tree_replace_slot, otherwise it must be called
792 * exclusive from other writers. Any dereference of the slot must be done
793 * using radix_tree_deref_slot.
795 void __rcu **radix_tree_lookup_slot(const struct radix_tree_root *root,
800 if (!__radix_tree_lookup(root, index, NULL, &slot))
804 EXPORT_SYMBOL(radix_tree_lookup_slot);
807 * radix_tree_lookup - perform lookup operation on a radix tree
808 * @root: radix tree root
811 * Lookup the item at the position @index in the radix tree @root.
813 * This function can be called under rcu_read_lock, however the caller
814 * must manage lifetimes of leaf nodes (eg. RCU may also be used to free
815 * them safely). No RCU barriers are required to access or modify the
816 * returned item, however.
818 void *radix_tree_lookup(const struct radix_tree_root *root, unsigned long index)
820 return __radix_tree_lookup(root, index, NULL, NULL);
822 EXPORT_SYMBOL(radix_tree_lookup);
824 static void replace_slot(void __rcu **slot, void *item,
825 struct radix_tree_node *node, int count, int values)
827 if (node && (count || values)) {
828 node->count += count;
829 node->nr_values += values;
832 rcu_assign_pointer(*slot, item);
835 static bool node_tag_get(const struct radix_tree_root *root,
836 const struct radix_tree_node *node,
837 unsigned int tag, unsigned int offset)
840 return tag_get(node, tag, offset);
841 return root_tag_get(root, tag);
845 * IDR users want to be able to store NULL in the tree, so if the slot isn't
846 * free, don't adjust the count, even if it's transitioning between NULL and
847 * non-NULL. For the IDA, we mark slots as being IDR_FREE while they still
848 * have empty bits, but it only stores NULL in slots when they're being
851 static int calculate_count(struct radix_tree_root *root,
852 struct radix_tree_node *node, void __rcu **slot,
853 void *item, void *old)
856 unsigned offset = get_slot_offset(node, slot);
857 bool free = node_tag_get(root, node, IDR_FREE, offset);
863 return !!item - !!old;
867 * __radix_tree_replace - replace item in a slot
868 * @root: radix tree root
869 * @node: pointer to tree node
870 * @slot: pointer to slot in @node
871 * @item: new item to store in the slot.
873 * For use with __radix_tree_lookup(). Caller must hold tree write locked
874 * across slot lookup and replacement.
876 void __radix_tree_replace(struct radix_tree_root *root,
877 struct radix_tree_node *node,
878 void __rcu **slot, void *item)
880 void *old = rcu_dereference_raw(*slot);
881 int values = !!xa_is_value(item) - !!xa_is_value(old);
882 int count = calculate_count(root, node, slot, item, old);
885 * This function supports replacing value entries and
886 * deleting entries, but that needs accounting against the
887 * node unless the slot is root->xa_head.
889 WARN_ON_ONCE(!node && (slot != (void __rcu **)&root->xa_head) &&
891 replace_slot(slot, item, node, count, values);
896 delete_node(root, node);
900 * radix_tree_replace_slot - replace item in a slot
901 * @root: radix tree root
902 * @slot: pointer to slot
903 * @item: new item to store in the slot.
905 * For use with radix_tree_lookup_slot() and
906 * radix_tree_gang_lookup_tag_slot(). Caller must hold tree write locked
907 * across slot lookup and replacement.
909 * NOTE: This cannot be used to switch between non-entries (empty slots),
910 * regular entries, and value entries, as that requires accounting
911 * inside the radix tree node. When switching from one type of entry or
912 * deleting, use __radix_tree_lookup() and __radix_tree_replace() or
913 * radix_tree_iter_replace().
915 void radix_tree_replace_slot(struct radix_tree_root *root,
916 void __rcu **slot, void *item)
918 __radix_tree_replace(root, NULL, slot, item);
920 EXPORT_SYMBOL(radix_tree_replace_slot);
923 * radix_tree_iter_replace - replace item in a slot
924 * @root: radix tree root
925 * @slot: pointer to slot
926 * @item: new item to store in the slot.
928 * For use with radix_tree_for_each_slot().
929 * Caller must hold tree write locked.
931 void radix_tree_iter_replace(struct radix_tree_root *root,
932 const struct radix_tree_iter *iter,
933 void __rcu **slot, void *item)
935 __radix_tree_replace(root, iter->node, slot, item);
938 static void node_tag_set(struct radix_tree_root *root,
939 struct radix_tree_node *node,
940 unsigned int tag, unsigned int offset)
943 if (tag_get(node, tag, offset))
945 tag_set(node, tag, offset);
946 offset = node->offset;
950 if (!root_tag_get(root, tag))
951 root_tag_set(root, tag);
955 * radix_tree_tag_set - set a tag on a radix tree node
956 * @root: radix tree root
960 * Set the search tag (which must be < RADIX_TREE_MAX_TAGS)
961 * corresponding to @index in the radix tree. From
962 * the root all the way down to the leaf node.
964 * Returns the address of the tagged item. Setting a tag on a not-present
967 void *radix_tree_tag_set(struct radix_tree_root *root,
968 unsigned long index, unsigned int tag)
970 struct radix_tree_node *node, *parent;
971 unsigned long maxindex;
973 radix_tree_load_root(root, &node, &maxindex);
974 BUG_ON(index > maxindex);
976 while (radix_tree_is_internal_node(node)) {
979 parent = entry_to_node(node);
980 offset = radix_tree_descend(parent, &node, index);
983 if (!tag_get(parent, tag, offset))
984 tag_set(parent, tag, offset);
987 /* set the root's tag bit */
988 if (!root_tag_get(root, tag))
989 root_tag_set(root, tag);
993 EXPORT_SYMBOL(radix_tree_tag_set);
995 static void node_tag_clear(struct radix_tree_root *root,
996 struct radix_tree_node *node,
997 unsigned int tag, unsigned int offset)
1000 if (!tag_get(node, tag, offset))
1002 tag_clear(node, tag, offset);
1003 if (any_tag_set(node, tag))
1006 offset = node->offset;
1007 node = node->parent;
1010 /* clear the root's tag bit */
1011 if (root_tag_get(root, tag))
1012 root_tag_clear(root, tag);
1016 * radix_tree_tag_clear - clear a tag on a radix tree node
1017 * @root: radix tree root
1021 * Clear the search tag (which must be < RADIX_TREE_MAX_TAGS)
1022 * corresponding to @index in the radix tree. If this causes
1023 * the leaf node to have no tags set then clear the tag in the
1024 * next-to-leaf node, etc.
1026 * Returns the address of the tagged item on success, else NULL. ie:
1027 * has the same return value and semantics as radix_tree_lookup().
1029 void *radix_tree_tag_clear(struct radix_tree_root *root,
1030 unsigned long index, unsigned int tag)
1032 struct radix_tree_node *node, *parent;
1033 unsigned long maxindex;
1034 int uninitialized_var(offset);
1036 radix_tree_load_root(root, &node, &maxindex);
1037 if (index > maxindex)
1042 while (radix_tree_is_internal_node(node)) {
1043 parent = entry_to_node(node);
1044 offset = radix_tree_descend(parent, &node, index);
1048 node_tag_clear(root, parent, tag, offset);
1052 EXPORT_SYMBOL(radix_tree_tag_clear);
1055 * radix_tree_iter_tag_clear - clear a tag on the current iterator entry
1056 * @root: radix tree root
1057 * @iter: iterator state
1058 * @tag: tag to clear
1060 void radix_tree_iter_tag_clear(struct radix_tree_root *root,
1061 const struct radix_tree_iter *iter, unsigned int tag)
1063 node_tag_clear(root, iter->node, tag, iter_offset(iter));
1067 * radix_tree_tag_get - get a tag on a radix tree node
1068 * @root: radix tree root
1070 * @tag: tag index (< RADIX_TREE_MAX_TAGS)
1074 * 0: tag not present or not set
1077 * Note that the return value of this function may not be relied on, even if
1078 * the RCU lock is held, unless tag modification and node deletion are excluded
1081 int radix_tree_tag_get(const struct radix_tree_root *root,
1082 unsigned long index, unsigned int tag)
1084 struct radix_tree_node *node, *parent;
1085 unsigned long maxindex;
1087 if (!root_tag_get(root, tag))
1090 radix_tree_load_root(root, &node, &maxindex);
1091 if (index > maxindex)
1094 while (radix_tree_is_internal_node(node)) {
1097 parent = entry_to_node(node);
1098 offset = radix_tree_descend(parent, &node, index);
1100 if (!tag_get(parent, tag, offset))
1102 if (node == RADIX_TREE_RETRY)
1108 EXPORT_SYMBOL(radix_tree_tag_get);
1110 /* Construct iter->tags bit-mask from node->tags[tag] array */
1111 static void set_iter_tags(struct radix_tree_iter *iter,
1112 struct radix_tree_node *node, unsigned offset,
1115 unsigned tag_long = offset / BITS_PER_LONG;
1116 unsigned tag_bit = offset % BITS_PER_LONG;
1123 iter->tags = node->tags[tag][tag_long] >> tag_bit;
1125 /* This never happens if RADIX_TREE_TAG_LONGS == 1 */
1126 if (tag_long < RADIX_TREE_TAG_LONGS - 1) {
1127 /* Pick tags from next element */
1129 iter->tags |= node->tags[tag][tag_long + 1] <<
1130 (BITS_PER_LONG - tag_bit);
1131 /* Clip chunk size, here only BITS_PER_LONG tags */
1132 iter->next_index = __radix_tree_iter_add(iter, BITS_PER_LONG);
1136 void __rcu **radix_tree_iter_resume(void __rcu **slot,
1137 struct radix_tree_iter *iter)
1140 iter->index = __radix_tree_iter_add(iter, 1);
1141 iter->next_index = iter->index;
1145 EXPORT_SYMBOL(radix_tree_iter_resume);
1148 * radix_tree_next_chunk - find next chunk of slots for iteration
1150 * @root: radix tree root
1151 * @iter: iterator state
1152 * @flags: RADIX_TREE_ITER_* flags and tag index
1153 * Returns: pointer to chunk first slot, or NULL if iteration is over
1155 void __rcu **radix_tree_next_chunk(const struct radix_tree_root *root,
1156 struct radix_tree_iter *iter, unsigned flags)
1158 unsigned tag = flags & RADIX_TREE_ITER_TAG_MASK;
1159 struct radix_tree_node *node, *child;
1160 unsigned long index, offset, maxindex;
1162 if ((flags & RADIX_TREE_ITER_TAGGED) && !root_tag_get(root, tag))
1166 * Catch next_index overflow after ~0UL. iter->index never overflows
1167 * during iterating; it can be zero only at the beginning.
1168 * And we cannot overflow iter->next_index in a single step,
1169 * because RADIX_TREE_MAP_SHIFT < BITS_PER_LONG.
1171 * This condition also used by radix_tree_next_slot() to stop
1172 * contiguous iterating, and forbid switching to the next chunk.
1174 index = iter->next_index;
1175 if (!index && iter->index)
1179 radix_tree_load_root(root, &child, &maxindex);
1180 if (index > maxindex)
1185 if (!radix_tree_is_internal_node(child)) {
1186 /* Single-slot tree */
1187 iter->index = index;
1188 iter->next_index = maxindex + 1;
1191 return (void __rcu **)&root->xa_head;
1195 node = entry_to_node(child);
1196 offset = radix_tree_descend(node, &child, index);
1198 if ((flags & RADIX_TREE_ITER_TAGGED) ?
1199 !tag_get(node, tag, offset) : !child) {
1201 if (flags & RADIX_TREE_ITER_CONTIG)
1204 if (flags & RADIX_TREE_ITER_TAGGED)
1205 offset = radix_tree_find_next_bit(node, tag,
1208 while (++offset < RADIX_TREE_MAP_SIZE) {
1209 void *slot = rcu_dereference_raw(
1210 node->slots[offset]);
1214 index &= ~node_maxindex(node);
1215 index += offset << node->shift;
1216 /* Overflow after ~0UL */
1219 if (offset == RADIX_TREE_MAP_SIZE)
1221 child = rcu_dereference_raw(node->slots[offset]);
1226 if (child == RADIX_TREE_RETRY)
1228 } while (node->shift && radix_tree_is_internal_node(child));
1230 /* Update the iterator state */
1231 iter->index = (index &~ node_maxindex(node)) | offset;
1232 iter->next_index = (index | node_maxindex(node)) + 1;
1235 if (flags & RADIX_TREE_ITER_TAGGED)
1236 set_iter_tags(iter, node, offset, tag);
1238 return node->slots + offset;
1240 EXPORT_SYMBOL(radix_tree_next_chunk);
1243 * radix_tree_gang_lookup - perform multiple lookup on a radix tree
1244 * @root: radix tree root
1245 * @results: where the results of the lookup are placed
1246 * @first_index: start the lookup from this key
1247 * @max_items: place up to this many items at *results
1249 * Performs an index-ascending scan of the tree for present items. Places
1250 * them at *@results and returns the number of items which were placed at
1253 * The implementation is naive.
1255 * Like radix_tree_lookup, radix_tree_gang_lookup may be called under
1256 * rcu_read_lock. In this case, rather than the returned results being
1257 * an atomic snapshot of the tree at a single point in time, the
1258 * semantics of an RCU protected gang lookup are as though multiple
1259 * radix_tree_lookups have been issued in individual locks, and results
1260 * stored in 'results'.
1263 radix_tree_gang_lookup(const struct radix_tree_root *root, void **results,
1264 unsigned long first_index, unsigned int max_items)
1266 struct radix_tree_iter iter;
1268 unsigned int ret = 0;
1270 if (unlikely(!max_items))
1273 radix_tree_for_each_slot(slot, root, &iter, first_index) {
1274 results[ret] = rcu_dereference_raw(*slot);
1277 if (radix_tree_is_internal_node(results[ret])) {
1278 slot = radix_tree_iter_retry(&iter);
1281 if (++ret == max_items)
1287 EXPORT_SYMBOL(radix_tree_gang_lookup);
1290 * radix_tree_gang_lookup_tag - perform multiple lookup on a radix tree
1292 * @root: radix tree root
1293 * @results: where the results of the lookup are placed
1294 * @first_index: start the lookup from this key
1295 * @max_items: place up to this many items at *results
1296 * @tag: the tag index (< RADIX_TREE_MAX_TAGS)
1298 * Performs an index-ascending scan of the tree for present items which
1299 * have the tag indexed by @tag set. Places the items at *@results and
1300 * returns the number of items which were placed at *@results.
1303 radix_tree_gang_lookup_tag(const struct radix_tree_root *root, void **results,
1304 unsigned long first_index, unsigned int max_items,
1307 struct radix_tree_iter iter;
1309 unsigned int ret = 0;
1311 if (unlikely(!max_items))
1314 radix_tree_for_each_tagged(slot, root, &iter, first_index, tag) {
1315 results[ret] = rcu_dereference_raw(*slot);
1318 if (radix_tree_is_internal_node(results[ret])) {
1319 slot = radix_tree_iter_retry(&iter);
1322 if (++ret == max_items)
1328 EXPORT_SYMBOL(radix_tree_gang_lookup_tag);
1331 * radix_tree_gang_lookup_tag_slot - perform multiple slot lookup on a
1332 * radix tree based on a tag
1333 * @root: radix tree root
1334 * @results: where the results of the lookup are placed
1335 * @first_index: start the lookup from this key
1336 * @max_items: place up to this many items at *results
1337 * @tag: the tag index (< RADIX_TREE_MAX_TAGS)
1339 * Performs an index-ascending scan of the tree for present items which
1340 * have the tag indexed by @tag set. Places the slots at *@results and
1341 * returns the number of slots which were placed at *@results.
1344 radix_tree_gang_lookup_tag_slot(const struct radix_tree_root *root,
1345 void __rcu ***results, unsigned long first_index,
1346 unsigned int max_items, unsigned int tag)
1348 struct radix_tree_iter iter;
1350 unsigned int ret = 0;
1352 if (unlikely(!max_items))
1355 radix_tree_for_each_tagged(slot, root, &iter, first_index, tag) {
1356 results[ret] = slot;
1357 if (++ret == max_items)
1363 EXPORT_SYMBOL(radix_tree_gang_lookup_tag_slot);
1365 static bool __radix_tree_delete(struct radix_tree_root *root,
1366 struct radix_tree_node *node, void __rcu **slot)
1368 void *old = rcu_dereference_raw(*slot);
1369 int values = xa_is_value(old) ? -1 : 0;
1370 unsigned offset = get_slot_offset(node, slot);
1374 node_tag_set(root, node, IDR_FREE, offset);
1376 for (tag = 0; tag < RADIX_TREE_MAX_TAGS; tag++)
1377 node_tag_clear(root, node, tag, offset);
1379 replace_slot(slot, NULL, node, -1, values);
1380 return node && delete_node(root, node);
1384 * radix_tree_iter_delete - delete the entry at this iterator position
1385 * @root: radix tree root
1386 * @iter: iterator state
1387 * @slot: pointer to slot
1389 * Delete the entry at the position currently pointed to by the iterator.
1390 * This may result in the current node being freed; if it is, the iterator
1391 * is advanced so that it will not reference the freed memory. This
1392 * function may be called without any locking if there are no other threads
1393 * which can access this tree.
1395 void radix_tree_iter_delete(struct radix_tree_root *root,
1396 struct radix_tree_iter *iter, void __rcu **slot)
1398 if (__radix_tree_delete(root, iter->node, slot))
1399 iter->index = iter->next_index;
1401 EXPORT_SYMBOL(radix_tree_iter_delete);
1404 * radix_tree_delete_item - delete an item from a radix tree
1405 * @root: radix tree root
1407 * @item: expected item
1409 * Remove @item at @index from the radix tree rooted at @root.
1411 * Return: the deleted entry, or %NULL if it was not present
1412 * or the entry at the given @index was not @item.
1414 void *radix_tree_delete_item(struct radix_tree_root *root,
1415 unsigned long index, void *item)
1417 struct radix_tree_node *node = NULL;
1418 void __rcu **slot = NULL;
1421 entry = __radix_tree_lookup(root, index, &node, &slot);
1424 if (!entry && (!is_idr(root) || node_tag_get(root, node, IDR_FREE,
1425 get_slot_offset(node, slot))))
1428 if (item && entry != item)
1431 __radix_tree_delete(root, node, slot);
1435 EXPORT_SYMBOL(radix_tree_delete_item);
1438 * radix_tree_delete - delete an entry from a radix tree
1439 * @root: radix tree root
1442 * Remove the entry at @index from the radix tree rooted at @root.
1444 * Return: The deleted entry, or %NULL if it was not present.
1446 void *radix_tree_delete(struct radix_tree_root *root, unsigned long index)
1448 return radix_tree_delete_item(root, index, NULL);
1450 EXPORT_SYMBOL(radix_tree_delete);
1453 * radix_tree_tagged - test whether any items in the tree are tagged
1454 * @root: radix tree root
1457 int radix_tree_tagged(const struct radix_tree_root *root, unsigned int tag)
1459 return root_tag_get(root, tag);
1461 EXPORT_SYMBOL(radix_tree_tagged);
1464 * idr_preload - preload for idr_alloc()
1465 * @gfp_mask: allocation mask to use for preloading
1467 * Preallocate memory to use for the next call to idr_alloc(). This function
1468 * returns with preemption disabled. It will be enabled by idr_preload_end().
1470 void idr_preload(gfp_t gfp_mask)
1472 if (__radix_tree_preload(gfp_mask, IDR_PRELOAD_SIZE))
1475 EXPORT_SYMBOL(idr_preload);
1477 void __rcu **idr_get_free(struct radix_tree_root *root,
1478 struct radix_tree_iter *iter, gfp_t gfp,
1481 struct radix_tree_node *node = NULL, *child;
1482 void __rcu **slot = (void __rcu **)&root->xa_head;
1483 unsigned long maxindex, start = iter->next_index;
1484 unsigned int shift, offset = 0;
1487 shift = radix_tree_load_root(root, &child, &maxindex);
1488 if (!radix_tree_tagged(root, IDR_FREE))
1489 start = max(start, maxindex + 1);
1491 return ERR_PTR(-ENOSPC);
1493 if (start > maxindex) {
1494 int error = radix_tree_extend(root, gfp, start, shift);
1496 return ERR_PTR(error);
1498 child = rcu_dereference_raw(root->xa_head);
1500 if (start == 0 && shift == 0)
1501 shift = RADIX_TREE_MAP_SHIFT;
1504 shift -= RADIX_TREE_MAP_SHIFT;
1505 if (child == NULL) {
1506 /* Have to add a child node. */
1507 child = radix_tree_node_alloc(gfp, node, root, shift,
1510 return ERR_PTR(-ENOMEM);
1511 all_tag_set(child, IDR_FREE);
1512 rcu_assign_pointer(*slot, node_to_entry(child));
1515 } else if (!radix_tree_is_internal_node(child))
1518 node = entry_to_node(child);
1519 offset = radix_tree_descend(node, &child, start);
1520 if (!tag_get(node, IDR_FREE, offset)) {
1521 offset = radix_tree_find_next_bit(node, IDR_FREE,
1523 start = next_index(start, node, offset);
1524 if (start > max || start == 0)
1525 return ERR_PTR(-ENOSPC);
1526 while (offset == RADIX_TREE_MAP_SIZE) {
1527 offset = node->offset + 1;
1528 node = node->parent;
1531 shift = node->shift;
1533 child = rcu_dereference_raw(node->slots[offset]);
1535 slot = &node->slots[offset];
1538 iter->index = start;
1540 iter->next_index = 1 + min(max, (start | node_maxindex(node)));
1542 iter->next_index = 1;
1544 set_iter_tags(iter, node, offset, IDR_FREE);
1550 * idr_destroy - release all internal memory from an IDR
1553 * After this function is called, the IDR is empty, and may be reused or
1554 * the data structure containing it may be freed.
1556 * A typical clean-up sequence for objects stored in an idr tree will use
1557 * idr_for_each() to free all objects, if necessary, then idr_destroy() to
1558 * free the memory used to keep track of those objects.
1560 void idr_destroy(struct idr *idr)
1562 struct radix_tree_node *node = rcu_dereference_raw(idr->idr_rt.xa_head);
1563 if (radix_tree_is_internal_node(node))
1564 radix_tree_free_nodes(node);
1565 idr->idr_rt.xa_head = NULL;
1566 root_tag_set(&idr->idr_rt, IDR_FREE);
1568 EXPORT_SYMBOL(idr_destroy);
1571 radix_tree_node_ctor(void *arg)
1573 struct radix_tree_node *node = arg;
1575 memset(node, 0, sizeof(*node));
1576 INIT_LIST_HEAD(&node->private_list);
1579 static int radix_tree_cpu_dead(unsigned int cpu)
1581 struct radix_tree_preload *rtp;
1582 struct radix_tree_node *node;
1584 /* Free per-cpu pool of preloaded nodes */
1585 rtp = &per_cpu(radix_tree_preloads, cpu);
1588 rtp->nodes = node->parent;
1589 kmem_cache_free(radix_tree_node_cachep, node);
1595 void __init radix_tree_init(void)
1599 BUILD_BUG_ON(RADIX_TREE_MAX_TAGS + __GFP_BITS_SHIFT > 32);
1600 BUILD_BUG_ON(ROOT_IS_IDR & ~GFP_ZONEMASK);
1601 BUILD_BUG_ON(XA_CHUNK_SIZE > 255);
1602 radix_tree_node_cachep = kmem_cache_create("radix_tree_node",
1603 sizeof(struct radix_tree_node), 0,
1604 SLAB_PANIC | SLAB_RECLAIM_ACCOUNT,
1605 radix_tree_node_ctor);
1606 ret = cpuhp_setup_state_nocalls(CPUHP_RADIX_DEAD, "lib/radix:dead",
1607 NULL, radix_tree_cpu_dead);