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/errno.h>
26 #include <linux/init.h>
27 #include <linux/kernel.h>
28 #include <linux/export.h>
29 #include <linux/radix-tree.h>
30 #include <linux/percpu.h>
31 #include <linux/slab.h>
32 #include <linux/kmemleak.h>
33 #include <linux/notifier.h>
34 #include <linux/cpu.h>
35 #include <linux/string.h>
36 #include <linux/bitops.h>
37 #include <linux/rcupdate.h>
38 #include <linux/preempt.h> /* in_interrupt() */
41 /* Number of nodes in fully populated tree of given height */
42 static unsigned long height_to_maxnodes[RADIX_TREE_MAX_PATH + 1] __read_mostly;
45 * Radix tree node cache.
47 static struct kmem_cache *radix_tree_node_cachep;
50 * The radix tree is variable-height, so an insert operation not only has
51 * to build the branch to its corresponding item, it also has to build the
52 * branch to existing items if the size has to be increased (by
55 * The worst case is a zero height tree with just a single item at index 0,
56 * and then inserting an item at index ULONG_MAX. This requires 2 new branches
57 * of RADIX_TREE_MAX_PATH size to be created, with only the root node shared.
60 #define RADIX_TREE_PRELOAD_SIZE (RADIX_TREE_MAX_PATH * 2 - 1)
63 * Per-cpu pool of preloaded nodes
65 struct radix_tree_preload {
67 /* nodes->private_data points to next preallocated node */
68 struct radix_tree_node *nodes;
70 static DEFINE_PER_CPU(struct radix_tree_preload, radix_tree_preloads) = { 0, };
72 static inline void *node_to_entry(void *ptr)
74 return (void *)((unsigned long)ptr | RADIX_TREE_INTERNAL_NODE);
77 #define RADIX_TREE_RETRY node_to_entry(NULL)
79 #ifdef CONFIG_RADIX_TREE_MULTIORDER
80 /* Sibling slots point directly to another slot in the same node */
81 static inline bool is_sibling_entry(struct radix_tree_node *parent, void *node)
84 return (parent->slots <= ptr) &&
85 (ptr < parent->slots + RADIX_TREE_MAP_SIZE);
88 static inline bool is_sibling_entry(struct radix_tree_node *parent, void *node)
94 static inline unsigned long get_slot_offset(struct radix_tree_node *parent,
97 return slot - parent->slots;
100 static unsigned int radix_tree_descend(struct radix_tree_node *parent,
101 struct radix_tree_node **nodep, unsigned long index)
103 unsigned int offset = (index >> parent->shift) & RADIX_TREE_MAP_MASK;
104 void **entry = rcu_dereference_raw(parent->slots[offset]);
106 #ifdef CONFIG_RADIX_TREE_MULTIORDER
107 if (radix_tree_is_internal_node(entry)) {
108 if (is_sibling_entry(parent, entry)) {
109 void **sibentry = (void **) entry_to_node(entry);
110 offset = get_slot_offset(parent, sibentry);
111 entry = rcu_dereference_raw(*sibentry);
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 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 int)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(struct radix_tree_node *node, unsigned long index)
223 pr_debug("radix node: %p offset %d tags %lx %lx %lx shift %d count %d exceptional %d parent %p\n",
225 node->tags[0][0], node->tags[1][0], node->tags[2][0],
226 node->shift, node->count, node->exceptional, node->parent);
228 for (i = 0; i < RADIX_TREE_MAP_SIZE; i++) {
229 unsigned long first = index | (i << node->shift);
230 unsigned long last = first | ((1UL << node->shift) - 1);
231 void *entry = node->slots[i];
234 if (is_sibling_entry(node, entry)) {
235 pr_debug("radix sblng %p offset %ld val %p indices %ld-%ld\n",
237 *(void **)entry_to_node(entry),
239 } else if (!radix_tree_is_internal_node(entry)) {
240 pr_debug("radix entry %p offset %ld indices %ld-%ld\n",
241 entry, i, first, last);
243 dump_node(entry_to_node(entry), first);
249 static void radix_tree_dump(struct radix_tree_root *root)
251 pr_debug("radix root: %p rnode %p tags %x\n",
253 root->gfp_mask >> __GFP_BITS_SHIFT);
254 if (!radix_tree_is_internal_node(root->rnode))
256 dump_node(entry_to_node(root->rnode), 0);
261 * This assumes that the caller has performed appropriate preallocation, and
262 * that the caller has pinned this thread of control to the current CPU.
264 static struct radix_tree_node *
265 radix_tree_node_alloc(struct radix_tree_root *root)
267 struct radix_tree_node *ret = NULL;
268 gfp_t gfp_mask = root_gfp_mask(root);
271 * Preload code isn't irq safe and it doesn't make sense to use
272 * preloading during an interrupt anyway as all the allocations have
273 * to be atomic. So just do normal allocation when in interrupt.
275 if (!gfpflags_allow_blocking(gfp_mask) && !in_interrupt()) {
276 struct radix_tree_preload *rtp;
279 * Even if the caller has preloaded, try to allocate from the
280 * cache first for the new node to get accounted to the memory
283 ret = kmem_cache_alloc(radix_tree_node_cachep,
284 gfp_mask | __GFP_NOWARN);
289 * Provided the caller has preloaded here, we will always
290 * succeed in getting a node here (and never reach
293 rtp = this_cpu_ptr(&radix_tree_preloads);
296 rtp->nodes = ret->private_data;
297 ret->private_data = NULL;
301 * Update the allocation stack trace as this is more useful
304 kmemleak_update_trace(ret);
307 ret = kmem_cache_alloc(radix_tree_node_cachep, gfp_mask);
309 BUG_ON(radix_tree_is_internal_node(ret));
313 static void radix_tree_node_rcu_free(struct rcu_head *head)
315 struct radix_tree_node *node =
316 container_of(head, struct radix_tree_node, rcu_head);
320 * must only free zeroed nodes into the slab. radix_tree_shrink
321 * can leave us with a non-NULL entry in the first slot, so clear
322 * that here to make sure.
324 for (i = 0; i < RADIX_TREE_MAX_TAGS; i++)
325 tag_clear(node, i, 0);
327 node->slots[0] = NULL;
329 kmem_cache_free(radix_tree_node_cachep, node);
333 radix_tree_node_free(struct radix_tree_node *node)
335 call_rcu(&node->rcu_head, radix_tree_node_rcu_free);
339 * Load up this CPU's radix_tree_node buffer with sufficient objects to
340 * ensure that the addition of a single element in the tree cannot fail. On
341 * success, return zero, with preemption disabled. On error, return -ENOMEM
342 * with preemption not disabled.
344 * To make use of this facility, the radix tree must be initialised without
345 * __GFP_DIRECT_RECLAIM being passed to INIT_RADIX_TREE().
347 static int __radix_tree_preload(gfp_t gfp_mask, int nr)
349 struct radix_tree_preload *rtp;
350 struct radix_tree_node *node;
354 * Nodes preloaded by one cgroup can be be used by another cgroup, so
355 * they should never be accounted to any particular memory cgroup.
357 gfp_mask &= ~__GFP_ACCOUNT;
360 rtp = this_cpu_ptr(&radix_tree_preloads);
361 while (rtp->nr < nr) {
363 node = kmem_cache_alloc(radix_tree_node_cachep, gfp_mask);
367 rtp = this_cpu_ptr(&radix_tree_preloads);
369 node->private_data = rtp->nodes;
373 kmem_cache_free(radix_tree_node_cachep, node);
382 * Load up this CPU's radix_tree_node buffer with sufficient objects to
383 * ensure that the addition of a single element in the tree cannot fail. On
384 * success, return zero, with preemption disabled. On error, return -ENOMEM
385 * with preemption not disabled.
387 * To make use of this facility, the radix tree must be initialised without
388 * __GFP_DIRECT_RECLAIM being passed to INIT_RADIX_TREE().
390 int radix_tree_preload(gfp_t gfp_mask)
392 /* Warn on non-sensical use... */
393 WARN_ON_ONCE(!gfpflags_allow_blocking(gfp_mask));
394 return __radix_tree_preload(gfp_mask, RADIX_TREE_PRELOAD_SIZE);
396 EXPORT_SYMBOL(radix_tree_preload);
399 * The same as above function, except we don't guarantee preloading happens.
400 * We do it, if we decide it helps. On success, return zero with preemption
401 * disabled. On error, return -ENOMEM with preemption not disabled.
403 int radix_tree_maybe_preload(gfp_t gfp_mask)
405 if (gfpflags_allow_blocking(gfp_mask))
406 return __radix_tree_preload(gfp_mask, RADIX_TREE_PRELOAD_SIZE);
407 /* Preloading doesn't help anything with this gfp mask, skip it */
411 EXPORT_SYMBOL(radix_tree_maybe_preload);
414 * The same as function above, but preload number of nodes required to insert
415 * (1 << order) continuous naturally-aligned elements.
417 int radix_tree_maybe_preload_order(gfp_t gfp_mask, int order)
419 unsigned long nr_subtrees;
420 int nr_nodes, subtree_height;
422 /* Preloading doesn't help anything with this gfp mask, skip it */
423 if (!gfpflags_allow_blocking(gfp_mask)) {
429 * Calculate number and height of fully populated subtrees it takes to
430 * store (1 << order) elements.
432 nr_subtrees = 1 << order;
433 for (subtree_height = 0; nr_subtrees > RADIX_TREE_MAP_SIZE;
435 nr_subtrees >>= RADIX_TREE_MAP_SHIFT;
438 * The worst case is zero height tree with a single item at index 0 and
439 * then inserting items starting at ULONG_MAX - (1 << order).
441 * This requires RADIX_TREE_MAX_PATH nodes to build branch from root to
444 nr_nodes = RADIX_TREE_MAX_PATH;
446 /* Plus branch to fully populated subtrees. */
447 nr_nodes += RADIX_TREE_MAX_PATH - subtree_height;
449 /* Root node is shared. */
452 /* Plus nodes required to build subtrees. */
453 nr_nodes += nr_subtrees * height_to_maxnodes[subtree_height];
455 return __radix_tree_preload(gfp_mask, nr_nodes);
459 * The maximum index which can be stored in a radix tree
461 static inline unsigned long shift_maxindex(unsigned int shift)
463 return (RADIX_TREE_MAP_SIZE << shift) - 1;
466 static inline unsigned long node_maxindex(struct radix_tree_node *node)
468 return shift_maxindex(node->shift);
471 static unsigned radix_tree_load_root(struct radix_tree_root *root,
472 struct radix_tree_node **nodep, unsigned long *maxindex)
474 struct radix_tree_node *node = rcu_dereference_raw(root->rnode);
478 if (likely(radix_tree_is_internal_node(node))) {
479 node = entry_to_node(node);
480 *maxindex = node_maxindex(node);
481 return node->shift + RADIX_TREE_MAP_SHIFT;
489 * Extend a radix tree so it can store key @index.
491 static int radix_tree_extend(struct radix_tree_root *root,
492 unsigned long index, unsigned int shift)
494 struct radix_tree_node *slot;
495 unsigned int maxshift;
498 /* Figure out what the shift should be. */
500 while (index > shift_maxindex(maxshift))
501 maxshift += RADIX_TREE_MAP_SHIFT;
508 struct radix_tree_node *node = radix_tree_node_alloc(root);
513 /* Propagate the aggregated tag info into the new root */
514 for (tag = 0; tag < RADIX_TREE_MAX_TAGS; tag++) {
515 if (root_tag_get(root, tag))
516 tag_set(node, tag, 0);
519 BUG_ON(shift > BITS_PER_LONG);
524 if (radix_tree_is_internal_node(slot)) {
525 entry_to_node(slot)->parent = node;
527 /* Moving an exceptional root->rnode to a node */
528 if (radix_tree_exceptional_entry(slot))
529 node->exceptional = 1;
531 node->slots[0] = slot;
532 slot = node_to_entry(node);
533 rcu_assign_pointer(root->rnode, slot);
534 shift += RADIX_TREE_MAP_SHIFT;
535 } while (shift <= maxshift);
537 return maxshift + RADIX_TREE_MAP_SHIFT;
541 * radix_tree_shrink - shrink radix tree to minimum height
542 * @root radix tree root
544 static inline void radix_tree_shrink(struct radix_tree_root *root,
545 radix_tree_update_node_t update_node,
549 struct radix_tree_node *node = root->rnode;
550 struct radix_tree_node *child;
552 if (!radix_tree_is_internal_node(node))
554 node = entry_to_node(node);
557 * The candidate node has more than one child, or its child
558 * is not at the leftmost slot, or the child is a multiorder
559 * entry, we cannot shrink.
561 if (node->count != 1)
563 child = node->slots[0];
566 if (!radix_tree_is_internal_node(child) && node->shift)
569 if (radix_tree_is_internal_node(child))
570 entry_to_node(child)->parent = NULL;
573 * We don't need rcu_assign_pointer(), since we are simply
574 * moving the node from one part of the tree to another: if it
575 * was safe to dereference the old pointer to it
576 * (node->slots[0]), it will be safe to dereference the new
577 * one (root->rnode) as far as dependent read barriers go.
582 * We have a dilemma here. The node's slot[0] must not be
583 * NULLed in case there are concurrent lookups expecting to
584 * find the item. However if this was a bottom-level node,
585 * then it may be subject to the slot pointer being visible
586 * to callers dereferencing it. If item corresponding to
587 * slot[0] is subsequently deleted, these callers would expect
588 * their slot to become empty sooner or later.
590 * For example, lockless pagecache will look up a slot, deref
591 * the page pointer, and if the page has 0 refcount it means it
592 * was concurrently deleted from pagecache so try the deref
593 * again. Fortunately there is already a requirement for logic
594 * to retry the entire slot lookup -- the indirect pointer
595 * problem (replacing direct root node with an indirect pointer
596 * also results in a stale slot). So tag the slot as indirect
597 * to force callers to retry.
600 if (!radix_tree_is_internal_node(child)) {
601 node->slots[0] = RADIX_TREE_RETRY;
603 update_node(node, private);
606 radix_tree_node_free(node);
610 static void delete_node(struct radix_tree_root *root,
611 struct radix_tree_node *node,
612 radix_tree_update_node_t update_node, void *private)
615 struct radix_tree_node *parent;
618 if (node == entry_to_node(root->rnode))
619 radix_tree_shrink(root, update_node, private);
623 parent = node->parent;
625 parent->slots[node->offset] = NULL;
628 root_tag_clear_all(root);
632 radix_tree_node_free(node);
639 * __radix_tree_create - create a slot in a radix tree
640 * @root: radix tree root
642 * @order: index occupies 2^order aligned slots
643 * @nodep: returns node
644 * @slotp: returns slot
646 * Create, if necessary, and return the node and slot for an item
647 * at position @index in the radix tree @root.
649 * Until there is more than one item in the tree, no nodes are
650 * allocated and @root->rnode is used as a direct slot instead of
651 * pointing to a node, in which case *@nodep will be NULL.
653 * Returns -ENOMEM, or 0 for success.
655 int __radix_tree_create(struct radix_tree_root *root, unsigned long index,
656 unsigned order, struct radix_tree_node **nodep,
659 struct radix_tree_node *node = NULL, *child;
660 void **slot = (void **)&root->rnode;
661 unsigned long maxindex;
662 unsigned int shift, offset = 0;
663 unsigned long max = index | ((1UL << order) - 1);
665 shift = radix_tree_load_root(root, &child, &maxindex);
667 /* Make sure the tree is high enough. */
668 if (max > maxindex) {
669 int error = radix_tree_extend(root, max, shift);
675 shift += RADIX_TREE_MAP_SHIFT;
678 while (shift > order) {
679 shift -= RADIX_TREE_MAP_SHIFT;
681 /* Have to add a child node. */
682 child = radix_tree_node_alloc(root);
685 child->shift = shift;
686 child->offset = offset;
687 child->parent = node;
688 rcu_assign_pointer(*slot, node_to_entry(child));
691 } else if (!radix_tree_is_internal_node(child))
694 /* Go a level down */
695 node = entry_to_node(child);
696 offset = radix_tree_descend(node, &child, index);
697 slot = &node->slots[offset];
700 #ifdef CONFIG_RADIX_TREE_MULTIORDER
701 /* Insert pointers to the canonical entry */
703 unsigned i, n = 1 << (order - shift);
704 offset = offset & ~(n - 1);
705 slot = &node->slots[offset];
706 child = node_to_entry(slot);
707 for (i = 0; i < n; i++) {
712 for (i = 1; i < n; i++) {
713 rcu_assign_pointer(slot[i], child);
727 * __radix_tree_insert - insert into a radix tree
728 * @root: radix tree root
730 * @order: key covers the 2^order indices around index
731 * @item: item to insert
733 * Insert an item into the radix tree at position @index.
735 int __radix_tree_insert(struct radix_tree_root *root, unsigned long index,
736 unsigned order, void *item)
738 struct radix_tree_node *node;
742 BUG_ON(radix_tree_is_internal_node(item));
744 error = __radix_tree_create(root, index, order, &node, &slot);
749 rcu_assign_pointer(*slot, item);
752 unsigned offset = get_slot_offset(node, slot);
754 if (radix_tree_exceptional_entry(item))
756 BUG_ON(tag_get(node, 0, offset));
757 BUG_ON(tag_get(node, 1, offset));
758 BUG_ON(tag_get(node, 2, offset));
760 BUG_ON(root_tags_get(root));
765 EXPORT_SYMBOL(__radix_tree_insert);
768 * __radix_tree_lookup - lookup an item in a radix tree
769 * @root: radix tree root
771 * @nodep: returns node
772 * @slotp: returns slot
774 * Lookup and return the item at position @index in the radix
777 * Until there is more than one item in the tree, no nodes are
778 * allocated and @root->rnode is used as a direct slot instead of
779 * pointing to a node, in which case *@nodep will be NULL.
781 void *__radix_tree_lookup(struct radix_tree_root *root, unsigned long index,
782 struct radix_tree_node **nodep, void ***slotp)
784 struct radix_tree_node *node, *parent;
785 unsigned long maxindex;
790 slot = (void **)&root->rnode;
791 radix_tree_load_root(root, &node, &maxindex);
792 if (index > maxindex)
795 while (radix_tree_is_internal_node(node)) {
798 if (node == RADIX_TREE_RETRY)
800 parent = entry_to_node(node);
801 offset = radix_tree_descend(parent, &node, index);
802 slot = parent->slots + offset;
813 * radix_tree_lookup_slot - lookup a slot in a radix tree
814 * @root: radix tree root
817 * Returns: the slot corresponding to the position @index in the
818 * radix tree @root. This is useful for update-if-exists operations.
820 * This function can be called under rcu_read_lock iff the slot is not
821 * modified by radix_tree_replace_slot, otherwise it must be called
822 * exclusive from other writers. Any dereference of the slot must be done
823 * using radix_tree_deref_slot.
825 void **radix_tree_lookup_slot(struct radix_tree_root *root, unsigned long index)
829 if (!__radix_tree_lookup(root, index, NULL, &slot))
833 EXPORT_SYMBOL(radix_tree_lookup_slot);
836 * radix_tree_lookup - perform lookup operation on a radix tree
837 * @root: radix tree root
840 * Lookup the item at the position @index in the radix tree @root.
842 * This function can be called under rcu_read_lock, however the caller
843 * must manage lifetimes of leaf nodes (eg. RCU may also be used to free
844 * them safely). No RCU barriers are required to access or modify the
845 * returned item, however.
847 void *radix_tree_lookup(struct radix_tree_root *root, unsigned long index)
849 return __radix_tree_lookup(root, index, NULL, NULL);
851 EXPORT_SYMBOL(radix_tree_lookup);
853 static void replace_slot(struct radix_tree_root *root,
854 struct radix_tree_node *node,
855 void **slot, void *item,
856 bool warn_typeswitch)
858 void *old = rcu_dereference_raw(*slot);
859 int count, exceptional;
861 WARN_ON_ONCE(radix_tree_is_internal_node(item));
863 count = !!item - !!old;
864 exceptional = !!radix_tree_exceptional_entry(item) -
865 !!radix_tree_exceptional_entry(old);
867 WARN_ON_ONCE(warn_typeswitch && (count || exceptional));
870 node->count += count;
871 node->exceptional += exceptional;
874 rcu_assign_pointer(*slot, item);
878 * __radix_tree_replace - replace item in a slot
879 * @root: radix tree root
880 * @node: pointer to tree node
881 * @slot: pointer to slot in @node
882 * @item: new item to store in the slot.
883 * @update_node: callback for changing leaf nodes
884 * @private: private data to pass to @update_node
886 * For use with __radix_tree_lookup(). Caller must hold tree write locked
887 * across slot lookup and replacement.
889 void __radix_tree_replace(struct radix_tree_root *root,
890 struct radix_tree_node *node,
891 void **slot, void *item,
892 radix_tree_update_node_t update_node, void *private)
895 * This function supports replacing exceptional entries and
896 * deleting entries, but that needs accounting against the
897 * node unless the slot is root->rnode.
899 replace_slot(root, node, slot, item,
900 !node && slot != (void **)&root->rnode);
906 update_node(node, private);
908 delete_node(root, node, update_node, private);
912 * radix_tree_replace_slot - replace item in a slot
913 * @root: radix tree root
914 * @slot: pointer to slot
915 * @item: new item to store in the slot.
917 * For use with radix_tree_lookup_slot(), radix_tree_gang_lookup_slot(),
918 * radix_tree_gang_lookup_tag_slot(). Caller must hold tree write locked
919 * across slot lookup and replacement.
921 * NOTE: This cannot be used to switch between non-entries (empty slots),
922 * regular entries, and exceptional entries, as that requires accounting
923 * inside the radix tree node. When switching from one type of entry or
924 * deleting, use __radix_tree_lookup() and __radix_tree_replace().
926 void radix_tree_replace_slot(struct radix_tree_root *root,
927 void **slot, void *item)
929 replace_slot(root, NULL, slot, item, true);
933 * radix_tree_tag_set - set a tag on a radix tree node
934 * @root: radix tree root
938 * Set the search tag (which must be < RADIX_TREE_MAX_TAGS)
939 * corresponding to @index in the radix tree. From
940 * the root all the way down to the leaf node.
942 * Returns the address of the tagged item. Setting a tag on a not-present
945 void *radix_tree_tag_set(struct radix_tree_root *root,
946 unsigned long index, unsigned int tag)
948 struct radix_tree_node *node, *parent;
949 unsigned long maxindex;
951 radix_tree_load_root(root, &node, &maxindex);
952 BUG_ON(index > maxindex);
954 while (radix_tree_is_internal_node(node)) {
957 parent = entry_to_node(node);
958 offset = radix_tree_descend(parent, &node, index);
961 if (!tag_get(parent, tag, offset))
962 tag_set(parent, tag, offset);
965 /* set the root's tag bit */
966 if (!root_tag_get(root, tag))
967 root_tag_set(root, tag);
971 EXPORT_SYMBOL(radix_tree_tag_set);
973 static void node_tag_clear(struct radix_tree_root *root,
974 struct radix_tree_node *node,
975 unsigned int tag, unsigned int offset)
978 if (!tag_get(node, tag, offset))
980 tag_clear(node, tag, offset);
981 if (any_tag_set(node, tag))
984 offset = node->offset;
988 /* clear the root's tag bit */
989 if (root_tag_get(root, tag))
990 root_tag_clear(root, tag);
994 * radix_tree_tag_clear - clear a tag on a radix tree node
995 * @root: radix tree root
999 * Clear the search tag (which must be < RADIX_TREE_MAX_TAGS)
1000 * corresponding to @index in the radix tree. If this causes
1001 * the leaf node to have no tags set then clear the tag in the
1002 * next-to-leaf node, etc.
1004 * Returns the address of the tagged item on success, else NULL. ie:
1005 * has the same return value and semantics as radix_tree_lookup().
1007 void *radix_tree_tag_clear(struct radix_tree_root *root,
1008 unsigned long index, unsigned int tag)
1010 struct radix_tree_node *node, *parent;
1011 unsigned long maxindex;
1012 int uninitialized_var(offset);
1014 radix_tree_load_root(root, &node, &maxindex);
1015 if (index > maxindex)
1020 while (radix_tree_is_internal_node(node)) {
1021 parent = entry_to_node(node);
1022 offset = radix_tree_descend(parent, &node, index);
1026 node_tag_clear(root, parent, tag, offset);
1030 EXPORT_SYMBOL(radix_tree_tag_clear);
1033 * radix_tree_tag_get - get a tag on a radix tree node
1034 * @root: radix tree root
1036 * @tag: tag index (< RADIX_TREE_MAX_TAGS)
1040 * 0: tag not present or not set
1043 * Note that the return value of this function may not be relied on, even if
1044 * the RCU lock is held, unless tag modification and node deletion are excluded
1047 int radix_tree_tag_get(struct radix_tree_root *root,
1048 unsigned long index, unsigned int tag)
1050 struct radix_tree_node *node, *parent;
1051 unsigned long maxindex;
1053 if (!root_tag_get(root, tag))
1056 radix_tree_load_root(root, &node, &maxindex);
1057 if (index > maxindex)
1062 while (radix_tree_is_internal_node(node)) {
1065 parent = entry_to_node(node);
1066 offset = radix_tree_descend(parent, &node, index);
1070 if (!tag_get(parent, tag, offset))
1072 if (node == RADIX_TREE_RETRY)
1078 EXPORT_SYMBOL(radix_tree_tag_get);
1080 static inline void __set_iter_shift(struct radix_tree_iter *iter,
1083 #ifdef CONFIG_RADIX_TREE_MULTIORDER
1084 iter->shift = shift;
1089 * radix_tree_next_chunk - find next chunk of slots for iteration
1091 * @root: radix tree root
1092 * @iter: iterator state
1093 * @flags: RADIX_TREE_ITER_* flags and tag index
1094 * Returns: pointer to chunk first slot, or NULL if iteration is over
1096 void **radix_tree_next_chunk(struct radix_tree_root *root,
1097 struct radix_tree_iter *iter, unsigned flags)
1099 unsigned tag = flags & RADIX_TREE_ITER_TAG_MASK;
1100 struct radix_tree_node *node, *child;
1101 unsigned long index, offset, maxindex;
1103 if ((flags & RADIX_TREE_ITER_TAGGED) && !root_tag_get(root, tag))
1107 * Catch next_index overflow after ~0UL. iter->index never overflows
1108 * during iterating; it can be zero only at the beginning.
1109 * And we cannot overflow iter->next_index in a single step,
1110 * because RADIX_TREE_MAP_SHIFT < BITS_PER_LONG.
1112 * This condition also used by radix_tree_next_slot() to stop
1113 * contiguous iterating, and forbid swithing to the next chunk.
1115 index = iter->next_index;
1116 if (!index && iter->index)
1120 radix_tree_load_root(root, &child, &maxindex);
1121 if (index > maxindex)
1126 if (!radix_tree_is_internal_node(child)) {
1127 /* Single-slot tree */
1128 iter->index = index;
1129 iter->next_index = maxindex + 1;
1131 __set_iter_shift(iter, 0);
1132 return (void **)&root->rnode;
1136 node = entry_to_node(child);
1137 offset = radix_tree_descend(node, &child, index);
1139 if ((flags & RADIX_TREE_ITER_TAGGED) ?
1140 !tag_get(node, tag, offset) : !child) {
1142 if (flags & RADIX_TREE_ITER_CONTIG)
1145 if (flags & RADIX_TREE_ITER_TAGGED)
1146 offset = radix_tree_find_next_bit(
1148 RADIX_TREE_MAP_SIZE,
1151 while (++offset < RADIX_TREE_MAP_SIZE) {
1152 void *slot = node->slots[offset];
1153 if (is_sibling_entry(node, slot))
1158 index &= ~node_maxindex(node);
1159 index += offset << node->shift;
1160 /* Overflow after ~0UL */
1163 if (offset == RADIX_TREE_MAP_SIZE)
1165 child = rcu_dereference_raw(node->slots[offset]);
1168 if ((child == NULL) || (child == RADIX_TREE_RETRY))
1170 } while (radix_tree_is_internal_node(child));
1172 /* Update the iterator state */
1173 iter->index = (index &~ node_maxindex(node)) | (offset << node->shift);
1174 iter->next_index = (index | node_maxindex(node)) + 1;
1175 __set_iter_shift(iter, node->shift);
1177 /* Construct iter->tags bit-mask from node->tags[tag] array */
1178 if (flags & RADIX_TREE_ITER_TAGGED) {
1179 unsigned tag_long, tag_bit;
1181 tag_long = offset / BITS_PER_LONG;
1182 tag_bit = offset % BITS_PER_LONG;
1183 iter->tags = node->tags[tag][tag_long] >> tag_bit;
1184 /* This never happens if RADIX_TREE_TAG_LONGS == 1 */
1185 if (tag_long < RADIX_TREE_TAG_LONGS - 1) {
1186 /* Pick tags from next element */
1188 iter->tags |= node->tags[tag][tag_long + 1] <<
1189 (BITS_PER_LONG - tag_bit);
1190 /* Clip chunk size, here only BITS_PER_LONG tags */
1191 iter->next_index = index + BITS_PER_LONG;
1195 return node->slots + offset;
1197 EXPORT_SYMBOL(radix_tree_next_chunk);
1200 * radix_tree_range_tag_if_tagged - for each item in given range set given
1201 * tag if item has another tag set
1202 * @root: radix tree root
1203 * @first_indexp: pointer to a starting index of a range to scan
1204 * @last_index: last index of a range to scan
1205 * @nr_to_tag: maximum number items to tag
1206 * @iftag: tag index to test
1207 * @settag: tag index to set if tested tag is set
1209 * This function scans range of radix tree from first_index to last_index
1210 * (inclusive). For each item in the range if iftag is set, the function sets
1211 * also settag. The function stops either after tagging nr_to_tag items or
1212 * after reaching last_index.
1214 * The tags must be set from the leaf level only and propagated back up the
1215 * path to the root. We must do this so that we resolve the full path before
1216 * setting any tags on intermediate nodes. If we set tags as we descend, then
1217 * we can get to the leaf node and find that the index that has the iftag
1218 * set is outside the range we are scanning. This reults in dangling tags and
1219 * can lead to problems with later tag operations (e.g. livelocks on lookups).
1221 * The function returns the number of leaves where the tag was set and sets
1222 * *first_indexp to the first unscanned index.
1223 * WARNING! *first_indexp can wrap if last_index is ULONG_MAX. Caller must
1224 * be prepared to handle that.
1226 unsigned long radix_tree_range_tag_if_tagged(struct radix_tree_root *root,
1227 unsigned long *first_indexp, unsigned long last_index,
1228 unsigned long nr_to_tag,
1229 unsigned int iftag, unsigned int settag)
1231 struct radix_tree_node *parent, *node, *child;
1232 unsigned long maxindex;
1233 unsigned long tagged = 0;
1234 unsigned long index = *first_indexp;
1236 radix_tree_load_root(root, &child, &maxindex);
1237 last_index = min(last_index, maxindex);
1238 if (index > last_index)
1242 if (!root_tag_get(root, iftag)) {
1243 *first_indexp = last_index + 1;
1246 if (!radix_tree_is_internal_node(child)) {
1247 *first_indexp = last_index + 1;
1248 root_tag_set(root, settag);
1252 node = entry_to_node(child);
1255 unsigned offset = radix_tree_descend(node, &child, index);
1258 if (!tag_get(node, iftag, offset))
1260 /* Sibling slots never have tags set on them */
1261 if (radix_tree_is_internal_node(child)) {
1262 node = entry_to_node(child);
1268 tag_set(node, settag, offset);
1270 /* walk back up the path tagging interior nodes */
1273 offset = parent->offset;
1274 parent = parent->parent;
1277 /* stop if we find a node with the tag already set */
1278 if (tag_get(parent, settag, offset))
1280 tag_set(parent, settag, offset);
1283 /* Go to next entry in node */
1284 index = ((index >> node->shift) + 1) << node->shift;
1285 /* Overflow can happen when last_index is ~0UL... */
1286 if (index > last_index || !index)
1288 offset = (index >> node->shift) & RADIX_TREE_MAP_MASK;
1289 while (offset == 0) {
1291 * We've fully scanned this node. Go up. Because
1292 * last_index is guaranteed to be in the tree, what
1293 * we do below cannot wander astray.
1295 node = node->parent;
1296 offset = (index >> node->shift) & RADIX_TREE_MAP_MASK;
1298 if (is_sibling_entry(node, node->slots[offset]))
1300 if (tagged >= nr_to_tag)
1304 * We need not to tag the root tag if there is no tag which is set with
1305 * settag within the range from *first_indexp to last_index.
1308 root_tag_set(root, settag);
1309 *first_indexp = index;
1313 EXPORT_SYMBOL(radix_tree_range_tag_if_tagged);
1316 * radix_tree_gang_lookup - perform multiple lookup on a radix tree
1317 * @root: radix tree root
1318 * @results: where the results of the lookup are placed
1319 * @first_index: start the lookup from this key
1320 * @max_items: place up to this many items at *results
1322 * Performs an index-ascending scan of the tree for present items. Places
1323 * them at *@results and returns the number of items which were placed at
1326 * The implementation is naive.
1328 * Like radix_tree_lookup, radix_tree_gang_lookup may be called under
1329 * rcu_read_lock. In this case, rather than the returned results being
1330 * an atomic snapshot of the tree at a single point in time, the
1331 * semantics of an RCU protected gang lookup are as though multiple
1332 * radix_tree_lookups have been issued in individual locks, and results
1333 * stored in 'results'.
1336 radix_tree_gang_lookup(struct radix_tree_root *root, void **results,
1337 unsigned long first_index, unsigned int max_items)
1339 struct radix_tree_iter iter;
1341 unsigned int ret = 0;
1343 if (unlikely(!max_items))
1346 radix_tree_for_each_slot(slot, root, &iter, first_index) {
1347 results[ret] = rcu_dereference_raw(*slot);
1350 if (radix_tree_is_internal_node(results[ret])) {
1351 slot = radix_tree_iter_retry(&iter);
1354 if (++ret == max_items)
1360 EXPORT_SYMBOL(radix_tree_gang_lookup);
1363 * radix_tree_gang_lookup_slot - perform multiple slot lookup on radix tree
1364 * @root: radix tree root
1365 * @results: where the results of the lookup are placed
1366 * @indices: where their indices should be placed (but usually NULL)
1367 * @first_index: start the lookup from this key
1368 * @max_items: place up to this many items at *results
1370 * Performs an index-ascending scan of the tree for present items. Places
1371 * their slots at *@results and returns the number of items which were
1372 * placed at *@results.
1374 * The implementation is naive.
1376 * Like radix_tree_gang_lookup as far as RCU and locking goes. Slots must
1377 * be dereferenced with radix_tree_deref_slot, and if using only RCU
1378 * protection, radix_tree_deref_slot may fail requiring a retry.
1381 radix_tree_gang_lookup_slot(struct radix_tree_root *root,
1382 void ***results, unsigned long *indices,
1383 unsigned long first_index, unsigned int max_items)
1385 struct radix_tree_iter iter;
1387 unsigned int ret = 0;
1389 if (unlikely(!max_items))
1392 radix_tree_for_each_slot(slot, root, &iter, first_index) {
1393 results[ret] = slot;
1395 indices[ret] = iter.index;
1396 if (++ret == max_items)
1402 EXPORT_SYMBOL(radix_tree_gang_lookup_slot);
1405 * radix_tree_gang_lookup_tag - perform multiple lookup on a radix tree
1407 * @root: radix tree root
1408 * @results: where the results of the lookup are placed
1409 * @first_index: start the lookup from this key
1410 * @max_items: place up to this many items at *results
1411 * @tag: the tag index (< RADIX_TREE_MAX_TAGS)
1413 * Performs an index-ascending scan of the tree for present items which
1414 * have the tag indexed by @tag set. Places the items at *@results and
1415 * returns the number of items which were placed at *@results.
1418 radix_tree_gang_lookup_tag(struct radix_tree_root *root, void **results,
1419 unsigned long first_index, unsigned int max_items,
1422 struct radix_tree_iter iter;
1424 unsigned int ret = 0;
1426 if (unlikely(!max_items))
1429 radix_tree_for_each_tagged(slot, root, &iter, first_index, tag) {
1430 results[ret] = rcu_dereference_raw(*slot);
1433 if (radix_tree_is_internal_node(results[ret])) {
1434 slot = radix_tree_iter_retry(&iter);
1437 if (++ret == max_items)
1443 EXPORT_SYMBOL(radix_tree_gang_lookup_tag);
1446 * radix_tree_gang_lookup_tag_slot - perform multiple slot lookup on a
1447 * radix tree based on a tag
1448 * @root: radix tree root
1449 * @results: where the results of the lookup are placed
1450 * @first_index: start the lookup from this key
1451 * @max_items: place up to this many items at *results
1452 * @tag: the tag index (< RADIX_TREE_MAX_TAGS)
1454 * Performs an index-ascending scan of the tree for present items which
1455 * have the tag indexed by @tag set. Places the slots at *@results and
1456 * returns the number of slots which were placed at *@results.
1459 radix_tree_gang_lookup_tag_slot(struct radix_tree_root *root, void ***results,
1460 unsigned long first_index, unsigned int max_items,
1463 struct radix_tree_iter iter;
1465 unsigned int ret = 0;
1467 if (unlikely(!max_items))
1470 radix_tree_for_each_tagged(slot, root, &iter, first_index, tag) {
1471 results[ret] = slot;
1472 if (++ret == max_items)
1478 EXPORT_SYMBOL(radix_tree_gang_lookup_tag_slot);
1480 #if defined(CONFIG_SHMEM) && defined(CONFIG_SWAP)
1481 #include <linux/sched.h> /* for cond_resched() */
1483 struct locate_info {
1484 unsigned long found_index;
1489 * This linear search is at present only useful to shmem_unuse_inode().
1491 static unsigned long __locate(struct radix_tree_node *slot, void *item,
1492 unsigned long index, struct locate_info *info)
1497 unsigned int shift = slot->shift;
1499 for (i = (index >> shift) & RADIX_TREE_MAP_MASK;
1500 i < RADIX_TREE_MAP_SIZE;
1501 i++, index += (1UL << shift)) {
1502 struct radix_tree_node *node =
1503 rcu_dereference_raw(slot->slots[i]);
1504 if (node == RADIX_TREE_RETRY)
1506 if (!radix_tree_is_internal_node(node)) {
1508 info->found_index = index;
1514 node = entry_to_node(node);
1515 if (is_sibling_entry(slot, node))
1520 } while (i < RADIX_TREE_MAP_SIZE);
1523 if ((index == 0) && (i == RADIX_TREE_MAP_SIZE))
1529 * radix_tree_locate_item - search through radix tree for item
1530 * @root: radix tree root
1531 * @item: item to be found
1533 * Returns index where item was found, or -1 if not found.
1534 * Caller must hold no lock (since this time-consuming function needs
1535 * to be preemptible), and must check afterwards if item is still there.
1537 unsigned long radix_tree_locate_item(struct radix_tree_root *root, void *item)
1539 struct radix_tree_node *node;
1540 unsigned long max_index;
1541 unsigned long cur_index = 0;
1542 struct locate_info info = {
1549 node = rcu_dereference_raw(root->rnode);
1550 if (!radix_tree_is_internal_node(node)) {
1553 info.found_index = 0;
1557 node = entry_to_node(node);
1559 max_index = node_maxindex(node);
1560 if (cur_index > max_index) {
1565 cur_index = __locate(node, item, cur_index, &info);
1568 } while (!info.stop && cur_index <= max_index);
1570 return info.found_index;
1573 unsigned long radix_tree_locate_item(struct radix_tree_root *root, void *item)
1577 #endif /* CONFIG_SHMEM && CONFIG_SWAP */
1580 * __radix_tree_delete_node - try to free node after clearing a slot
1581 * @root: radix tree root
1582 * @node: node containing @index
1584 * After clearing the slot at @index in @node from radix tree
1585 * rooted at @root, call this function to attempt freeing the
1586 * node and shrinking the tree.
1588 void __radix_tree_delete_node(struct radix_tree_root *root,
1589 struct radix_tree_node *node)
1591 delete_node(root, node, NULL, NULL);
1594 static inline void delete_sibling_entries(struct radix_tree_node *node,
1595 void *ptr, unsigned offset)
1597 #ifdef CONFIG_RADIX_TREE_MULTIORDER
1599 for (i = 1; offset + i < RADIX_TREE_MAP_SIZE; i++) {
1600 if (node->slots[offset + i] != ptr)
1602 node->slots[offset + i] = NULL;
1609 * radix_tree_delete_item - delete an item from a radix tree
1610 * @root: radix tree root
1612 * @item: expected item
1614 * Remove @item at @index from the radix tree rooted at @root.
1616 * Returns the address of the deleted item, or NULL if it was not present
1617 * or the entry at the given @index was not @item.
1619 void *radix_tree_delete_item(struct radix_tree_root *root,
1620 unsigned long index, void *item)
1622 struct radix_tree_node *node;
1623 unsigned int offset;
1628 entry = __radix_tree_lookup(root, index, &node, &slot);
1632 if (item && entry != item)
1636 root_tag_clear_all(root);
1641 offset = get_slot_offset(node, slot);
1643 /* Clear all tags associated with the item to be deleted. */
1644 for (tag = 0; tag < RADIX_TREE_MAX_TAGS; tag++)
1645 node_tag_clear(root, node, tag, offset);
1647 delete_sibling_entries(node, node_to_entry(slot), offset);
1648 __radix_tree_replace(root, node, slot, NULL, NULL, NULL);
1652 EXPORT_SYMBOL(radix_tree_delete_item);
1655 * radix_tree_delete - delete an item from a radix tree
1656 * @root: radix tree root
1659 * Remove the item at @index from the radix tree rooted at @root.
1661 * Returns the address of the deleted item, or NULL if it was not present.
1663 void *radix_tree_delete(struct radix_tree_root *root, unsigned long index)
1665 return radix_tree_delete_item(root, index, NULL);
1667 EXPORT_SYMBOL(radix_tree_delete);
1669 void radix_tree_clear_tags(struct radix_tree_root *root,
1670 struct radix_tree_node *node,
1674 unsigned int tag, offset = get_slot_offset(node, slot);
1675 for (tag = 0; tag < RADIX_TREE_MAX_TAGS; tag++)
1676 node_tag_clear(root, node, tag, offset);
1678 /* Clear root node tags */
1679 root->gfp_mask &= __GFP_BITS_MASK;
1684 * radix_tree_tagged - test whether any items in the tree are tagged
1685 * @root: radix tree root
1688 int radix_tree_tagged(struct radix_tree_root *root, unsigned int tag)
1690 return root_tag_get(root, tag);
1692 EXPORT_SYMBOL(radix_tree_tagged);
1695 radix_tree_node_ctor(void *arg)
1697 struct radix_tree_node *node = arg;
1699 memset(node, 0, sizeof(*node));
1700 INIT_LIST_HEAD(&node->private_list);
1703 static __init unsigned long __maxindex(unsigned int height)
1705 unsigned int width = height * RADIX_TREE_MAP_SHIFT;
1706 int shift = RADIX_TREE_INDEX_BITS - width;
1710 if (shift >= BITS_PER_LONG)
1712 return ~0UL >> shift;
1715 static __init void radix_tree_init_maxnodes(void)
1717 unsigned long height_to_maxindex[RADIX_TREE_MAX_PATH + 1];
1720 for (i = 0; i < ARRAY_SIZE(height_to_maxindex); i++)
1721 height_to_maxindex[i] = __maxindex(i);
1722 for (i = 0; i < ARRAY_SIZE(height_to_maxnodes); i++) {
1723 for (j = i; j > 0; j--)
1724 height_to_maxnodes[i] += height_to_maxindex[j - 1] + 1;
1728 static int radix_tree_cpu_dead(unsigned int cpu)
1730 struct radix_tree_preload *rtp;
1731 struct radix_tree_node *node;
1733 /* Free per-cpu pool of preloaded nodes */
1734 rtp = &per_cpu(radix_tree_preloads, cpu);
1737 rtp->nodes = node->private_data;
1738 kmem_cache_free(radix_tree_node_cachep, node);
1744 void __init radix_tree_init(void)
1747 radix_tree_node_cachep = kmem_cache_create("radix_tree_node",
1748 sizeof(struct radix_tree_node), 0,
1749 SLAB_PANIC | SLAB_RECLAIM_ACCOUNT,
1750 radix_tree_node_ctor);
1751 radix_tree_init_maxnodes();
1752 ret = cpuhp_setup_state_nocalls(CPUHP_RADIX_DEAD, "lib/radix:dead",
1753 NULL, radix_tree_cpu_dead);