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() */
42 * Radix tree node cache.
44 static struct kmem_cache *radix_tree_node_cachep;
47 * The radix tree is variable-height, so an insert operation not only has
48 * to build the branch to its corresponding item, it also has to build the
49 * branch to existing items if the size has to be increased (by
52 * The worst case is a zero height tree with just a single item at index 0,
53 * and then inserting an item at index ULONG_MAX. This requires 2 new branches
54 * of RADIX_TREE_MAX_PATH size to be created, with only the root node shared.
57 #define RADIX_TREE_PRELOAD_SIZE (RADIX_TREE_MAX_PATH * 2 - 1)
60 * Per-cpu pool of preloaded nodes
62 struct radix_tree_preload {
64 /* nodes->private_data points to next preallocated node */
65 struct radix_tree_node *nodes;
67 static DEFINE_PER_CPU(struct radix_tree_preload, radix_tree_preloads) = { 0, };
69 static inline void *node_to_entry(void *ptr)
71 return (void *)((unsigned long)ptr | RADIX_TREE_INTERNAL_NODE);
74 #define RADIX_TREE_RETRY node_to_entry(NULL)
76 #ifdef CONFIG_RADIX_TREE_MULTIORDER
77 /* Sibling slots point directly to another slot in the same node */
78 static inline bool is_sibling_entry(struct radix_tree_node *parent, void *node)
81 return (parent->slots <= ptr) &&
82 (ptr < parent->slots + RADIX_TREE_MAP_SIZE);
85 static inline bool is_sibling_entry(struct radix_tree_node *parent, void *node)
91 static inline unsigned long get_slot_offset(struct radix_tree_node *parent,
94 return slot - parent->slots;
97 static unsigned radix_tree_descend(struct radix_tree_node *parent,
98 struct radix_tree_node **nodep, unsigned offset)
100 void **entry = rcu_dereference_raw(parent->slots[offset]);
102 #ifdef CONFIG_RADIX_TREE_MULTIORDER
103 if (radix_tree_is_internal_node(entry)) {
104 unsigned long siboff = get_slot_offset(parent, entry);
105 if (siboff < RADIX_TREE_MAP_SIZE) {
107 entry = rcu_dereference_raw(parent->slots[offset]);
112 *nodep = (void *)entry;
116 static inline gfp_t root_gfp_mask(struct radix_tree_root *root)
118 return root->gfp_mask & __GFP_BITS_MASK;
121 static inline void tag_set(struct radix_tree_node *node, unsigned int tag,
124 __set_bit(offset, node->tags[tag]);
127 static inline void tag_clear(struct radix_tree_node *node, unsigned int tag,
130 __clear_bit(offset, node->tags[tag]);
133 static inline int tag_get(struct radix_tree_node *node, unsigned int tag,
136 return test_bit(offset, node->tags[tag]);
139 static inline void root_tag_set(struct radix_tree_root *root, unsigned int tag)
141 root->gfp_mask |= (__force gfp_t)(1 << (tag + __GFP_BITS_SHIFT));
144 static inline void root_tag_clear(struct radix_tree_root *root, unsigned tag)
146 root->gfp_mask &= (__force gfp_t)~(1 << (tag + __GFP_BITS_SHIFT));
149 static inline void root_tag_clear_all(struct radix_tree_root *root)
151 root->gfp_mask &= __GFP_BITS_MASK;
154 static inline int root_tag_get(struct radix_tree_root *root, unsigned int tag)
156 return (__force int)root->gfp_mask & (1 << (tag + __GFP_BITS_SHIFT));
159 static inline unsigned root_tags_get(struct radix_tree_root *root)
161 return (__force unsigned)root->gfp_mask >> __GFP_BITS_SHIFT;
165 * Returns 1 if any slot in the node has this tag set.
166 * Otherwise returns 0.
168 static inline int any_tag_set(struct radix_tree_node *node, unsigned int tag)
171 for (idx = 0; idx < RADIX_TREE_TAG_LONGS; idx++) {
172 if (node->tags[tag][idx])
179 * radix_tree_find_next_bit - find the next set bit in a memory region
181 * @addr: The address to base the search on
182 * @size: The bitmap size in bits
183 * @offset: The bitnumber to start searching at
185 * Unrollable variant of find_next_bit() for constant size arrays.
186 * Tail bits starting from size to roundup(size, BITS_PER_LONG) must be zero.
187 * Returns next bit offset, or size if nothing found.
189 static __always_inline unsigned long
190 radix_tree_find_next_bit(const unsigned long *addr,
191 unsigned long size, unsigned long offset)
193 if (!__builtin_constant_p(size))
194 return find_next_bit(addr, size, offset);
199 addr += offset / BITS_PER_LONG;
200 tmp = *addr >> (offset % BITS_PER_LONG);
202 return __ffs(tmp) + offset;
203 offset = (offset + BITS_PER_LONG) & ~(BITS_PER_LONG - 1);
204 while (offset < size) {
207 return __ffs(tmp) + offset;
208 offset += BITS_PER_LONG;
215 static void dump_node(struct radix_tree_node *node, unsigned long index)
219 pr_debug("radix node: %p offset %d tags %lx %lx %lx shift %d count %d parent %p\n",
221 node->tags[0][0], node->tags[1][0], node->tags[2][0],
222 node->shift, node->count, node->parent);
224 for (i = 0; i < RADIX_TREE_MAP_SIZE; i++) {
225 unsigned long first = index | (i << node->shift);
226 unsigned long last = first | ((1UL << node->shift) - 1);
227 void *entry = node->slots[i];
230 if (is_sibling_entry(node, entry)) {
231 pr_debug("radix sblng %p offset %ld val %p indices %ld-%ld\n",
233 *(void **)entry_to_node(entry),
235 } else if (!radix_tree_is_internal_node(entry)) {
236 pr_debug("radix entry %p offset %ld indices %ld-%ld\n",
237 entry, i, first, last);
239 dump_node(entry_to_node(entry), first);
245 static void radix_tree_dump(struct radix_tree_root *root)
247 pr_debug("radix root: %p rnode %p tags %x\n",
249 root->gfp_mask >> __GFP_BITS_SHIFT);
250 if (!radix_tree_is_internal_node(root->rnode))
252 dump_node(entry_to_node(root->rnode), 0);
257 * This assumes that the caller has performed appropriate preallocation, and
258 * that the caller has pinned this thread of control to the current CPU.
260 static struct radix_tree_node *
261 radix_tree_node_alloc(struct radix_tree_root *root)
263 struct radix_tree_node *ret = NULL;
264 gfp_t gfp_mask = root_gfp_mask(root);
267 * Preload code isn't irq safe and it doesn't make sense to use
268 * preloading during an interrupt anyway as all the allocations have
269 * to be atomic. So just do normal allocation when in interrupt.
271 if (!gfpflags_allow_blocking(gfp_mask) && !in_interrupt()) {
272 struct radix_tree_preload *rtp;
275 * Even if the caller has preloaded, try to allocate from the
276 * cache first for the new node to get accounted.
278 ret = kmem_cache_alloc(radix_tree_node_cachep,
279 gfp_mask | __GFP_ACCOUNT | __GFP_NOWARN);
284 * Provided the caller has preloaded here, we will always
285 * succeed in getting a node here (and never reach
288 rtp = this_cpu_ptr(&radix_tree_preloads);
291 rtp->nodes = ret->private_data;
292 ret->private_data = NULL;
296 * Update the allocation stack trace as this is more useful
299 kmemleak_update_trace(ret);
302 ret = kmem_cache_alloc(radix_tree_node_cachep,
303 gfp_mask | __GFP_ACCOUNT);
305 BUG_ON(radix_tree_is_internal_node(ret));
309 static void radix_tree_node_rcu_free(struct rcu_head *head)
311 struct radix_tree_node *node =
312 container_of(head, struct radix_tree_node, rcu_head);
316 * must only free zeroed nodes into the slab. radix_tree_shrink
317 * can leave us with a non-NULL entry in the first slot, so clear
318 * that here to make sure.
320 for (i = 0; i < RADIX_TREE_MAX_TAGS; i++)
321 tag_clear(node, i, 0);
323 node->slots[0] = NULL;
326 kmem_cache_free(radix_tree_node_cachep, node);
330 radix_tree_node_free(struct radix_tree_node *node)
332 call_rcu(&node->rcu_head, radix_tree_node_rcu_free);
336 * Load up this CPU's radix_tree_node buffer with sufficient objects to
337 * ensure that the addition of a single element in the tree cannot fail. On
338 * success, return zero, with preemption disabled. On error, return -ENOMEM
339 * with preemption not disabled.
341 * To make use of this facility, the radix tree must be initialised without
342 * __GFP_DIRECT_RECLAIM being passed to INIT_RADIX_TREE().
344 static int __radix_tree_preload(gfp_t gfp_mask)
346 struct radix_tree_preload *rtp;
347 struct radix_tree_node *node;
351 rtp = this_cpu_ptr(&radix_tree_preloads);
352 while (rtp->nr < RADIX_TREE_PRELOAD_SIZE) {
354 node = kmem_cache_alloc(radix_tree_node_cachep, gfp_mask);
358 rtp = this_cpu_ptr(&radix_tree_preloads);
359 if (rtp->nr < RADIX_TREE_PRELOAD_SIZE) {
360 node->private_data = rtp->nodes;
364 kmem_cache_free(radix_tree_node_cachep, node);
373 * Load up this CPU's radix_tree_node buffer with sufficient objects to
374 * ensure that the addition of a single element in the tree cannot fail. On
375 * success, return zero, with preemption disabled. On error, return -ENOMEM
376 * with preemption not disabled.
378 * To make use of this facility, the radix tree must be initialised without
379 * __GFP_DIRECT_RECLAIM being passed to INIT_RADIX_TREE().
381 int radix_tree_preload(gfp_t gfp_mask)
383 /* Warn on non-sensical use... */
384 WARN_ON_ONCE(!gfpflags_allow_blocking(gfp_mask));
385 return __radix_tree_preload(gfp_mask);
387 EXPORT_SYMBOL(radix_tree_preload);
390 * The same as above function, except we don't guarantee preloading happens.
391 * We do it, if we decide it helps. On success, return zero with preemption
392 * disabled. On error, return -ENOMEM with preemption not disabled.
394 int radix_tree_maybe_preload(gfp_t gfp_mask)
396 if (gfpflags_allow_blocking(gfp_mask))
397 return __radix_tree_preload(gfp_mask);
398 /* Preloading doesn't help anything with this gfp mask, skip it */
402 EXPORT_SYMBOL(radix_tree_maybe_preload);
405 * The maximum index which can be stored in a radix tree
407 static inline unsigned long shift_maxindex(unsigned int shift)
409 return (RADIX_TREE_MAP_SIZE << shift) - 1;
412 static inline unsigned long node_maxindex(struct radix_tree_node *node)
414 return shift_maxindex(node->shift);
417 static unsigned radix_tree_load_root(struct radix_tree_root *root,
418 struct radix_tree_node **nodep, unsigned long *maxindex)
420 struct radix_tree_node *node = rcu_dereference_raw(root->rnode);
424 if (likely(radix_tree_is_internal_node(node))) {
425 node = entry_to_node(node);
426 *maxindex = node_maxindex(node);
427 return node->shift + RADIX_TREE_MAP_SHIFT;
435 * Extend a radix tree so it can store key @index.
437 static int radix_tree_extend(struct radix_tree_root *root,
438 unsigned long index, unsigned int shift)
440 struct radix_tree_node *slot;
441 unsigned int maxshift;
444 /* Figure out what the shift should be. */
446 while (index > shift_maxindex(maxshift))
447 maxshift += RADIX_TREE_MAP_SHIFT;
454 struct radix_tree_node *node = radix_tree_node_alloc(root);
459 /* Propagate the aggregated tag info into the new root */
460 for (tag = 0; tag < RADIX_TREE_MAX_TAGS; tag++) {
461 if (root_tag_get(root, tag))
462 tag_set(node, tag, 0);
465 BUG_ON(shift > BITS_PER_LONG);
470 if (radix_tree_is_internal_node(slot))
471 entry_to_node(slot)->parent = node;
472 node->slots[0] = slot;
473 slot = node_to_entry(node);
474 rcu_assign_pointer(root->rnode, slot);
475 shift += RADIX_TREE_MAP_SHIFT;
476 } while (shift <= maxshift);
478 return maxshift + RADIX_TREE_MAP_SHIFT;
482 * __radix_tree_create - create a slot in a radix tree
483 * @root: radix tree root
485 * @order: index occupies 2^order aligned slots
486 * @nodep: returns node
487 * @slotp: returns slot
489 * Create, if necessary, and return the node and slot for an item
490 * at position @index in the radix tree @root.
492 * Until there is more than one item in the tree, no nodes are
493 * allocated and @root->rnode is used as a direct slot instead of
494 * pointing to a node, in which case *@nodep will be NULL.
496 * Returns -ENOMEM, or 0 for success.
498 int __radix_tree_create(struct radix_tree_root *root, unsigned long index,
499 unsigned order, struct radix_tree_node **nodep,
502 struct radix_tree_node *node = NULL, *child;
503 void **slot = (void **)&root->rnode;
504 unsigned long maxindex;
505 unsigned int shift, offset = 0;
506 unsigned long max = index | ((1UL << order) - 1);
508 shift = radix_tree_load_root(root, &child, &maxindex);
510 /* Make sure the tree is high enough. */
511 if (max > maxindex) {
512 int error = radix_tree_extend(root, max, shift);
518 shift += RADIX_TREE_MAP_SHIFT;
521 while (shift > order) {
522 shift -= RADIX_TREE_MAP_SHIFT;
524 /* Have to add a child node. */
525 child = radix_tree_node_alloc(root);
528 child->shift = shift;
529 child->offset = offset;
530 child->parent = node;
531 rcu_assign_pointer(*slot, node_to_entry(child));
534 } else if (!radix_tree_is_internal_node(child))
537 /* Go a level down */
538 node = entry_to_node(child);
539 offset = (index >> shift) & RADIX_TREE_MAP_MASK;
540 offset = radix_tree_descend(node, &child, offset);
541 slot = &node->slots[offset];
544 #ifdef CONFIG_RADIX_TREE_MULTIORDER
545 /* Insert pointers to the canonical entry */
547 unsigned i, n = 1 << (order - shift);
548 offset = offset & ~(n - 1);
549 slot = &node->slots[offset];
550 child = node_to_entry(slot);
551 for (i = 0; i < n; i++) {
556 for (i = 1; i < n; i++) {
557 rcu_assign_pointer(slot[i], child);
571 * __radix_tree_insert - insert into a radix tree
572 * @root: radix tree root
574 * @order: key covers the 2^order indices around index
575 * @item: item to insert
577 * Insert an item into the radix tree at position @index.
579 int __radix_tree_insert(struct radix_tree_root *root, unsigned long index,
580 unsigned order, void *item)
582 struct radix_tree_node *node;
586 BUG_ON(radix_tree_is_internal_node(item));
588 error = __radix_tree_create(root, index, order, &node, &slot);
593 rcu_assign_pointer(*slot, item);
596 unsigned offset = get_slot_offset(node, slot);
598 BUG_ON(tag_get(node, 0, offset));
599 BUG_ON(tag_get(node, 1, offset));
600 BUG_ON(tag_get(node, 2, offset));
602 BUG_ON(root_tags_get(root));
607 EXPORT_SYMBOL(__radix_tree_insert);
610 * __radix_tree_lookup - lookup an item in a radix tree
611 * @root: radix tree root
613 * @nodep: returns node
614 * @slotp: returns slot
616 * Lookup and return the item at position @index in the radix
619 * Until there is more than one item in the tree, no nodes are
620 * allocated and @root->rnode is used as a direct slot instead of
621 * pointing to a node, in which case *@nodep will be NULL.
623 void *__radix_tree_lookup(struct radix_tree_root *root, unsigned long index,
624 struct radix_tree_node **nodep, void ***slotp)
626 struct radix_tree_node *node, *parent;
627 unsigned long maxindex;
633 slot = (void **)&root->rnode;
634 shift = radix_tree_load_root(root, &node, &maxindex);
635 if (index > maxindex)
638 while (radix_tree_is_internal_node(node)) {
641 if (node == RADIX_TREE_RETRY)
643 parent = entry_to_node(node);
644 shift -= RADIX_TREE_MAP_SHIFT;
645 offset = (index >> shift) & RADIX_TREE_MAP_MASK;
646 offset = radix_tree_descend(parent, &node, offset);
647 slot = parent->slots + offset;
658 * radix_tree_lookup_slot - lookup a slot in a radix tree
659 * @root: radix tree root
662 * Returns: the slot corresponding to the position @index in the
663 * radix tree @root. This is useful for update-if-exists operations.
665 * This function can be called under rcu_read_lock iff the slot is not
666 * modified by radix_tree_replace_slot, otherwise it must be called
667 * exclusive from other writers. Any dereference of the slot must be done
668 * using radix_tree_deref_slot.
670 void **radix_tree_lookup_slot(struct radix_tree_root *root, unsigned long index)
674 if (!__radix_tree_lookup(root, index, NULL, &slot))
678 EXPORT_SYMBOL(radix_tree_lookup_slot);
681 * radix_tree_lookup - perform lookup operation on a radix tree
682 * @root: radix tree root
685 * Lookup the item at the position @index in the radix tree @root.
687 * This function can be called under rcu_read_lock, however the caller
688 * must manage lifetimes of leaf nodes (eg. RCU may also be used to free
689 * them safely). No RCU barriers are required to access or modify the
690 * returned item, however.
692 void *radix_tree_lookup(struct radix_tree_root *root, unsigned long index)
694 return __radix_tree_lookup(root, index, NULL, NULL);
696 EXPORT_SYMBOL(radix_tree_lookup);
699 * radix_tree_tag_set - set a tag on a radix tree node
700 * @root: radix tree root
704 * Set the search tag (which must be < RADIX_TREE_MAX_TAGS)
705 * corresponding to @index in the radix tree. From
706 * the root all the way down to the leaf node.
708 * Returns the address of the tagged item. Setting a tag on a not-present
711 void *radix_tree_tag_set(struct radix_tree_root *root,
712 unsigned long index, unsigned int tag)
714 struct radix_tree_node *node, *parent;
715 unsigned long maxindex;
718 shift = radix_tree_load_root(root, &node, &maxindex);
719 BUG_ON(index > maxindex);
721 while (radix_tree_is_internal_node(node)) {
724 shift -= RADIX_TREE_MAP_SHIFT;
725 offset = (index >> shift) & RADIX_TREE_MAP_MASK;
727 parent = entry_to_node(node);
728 offset = radix_tree_descend(parent, &node, offset);
731 if (!tag_get(parent, tag, offset))
732 tag_set(parent, tag, offset);
735 /* set the root's tag bit */
736 if (!root_tag_get(root, tag))
737 root_tag_set(root, tag);
741 EXPORT_SYMBOL(radix_tree_tag_set);
744 * radix_tree_tag_clear - clear a tag on a radix tree node
745 * @root: radix tree root
749 * Clear the search tag (which must be < RADIX_TREE_MAX_TAGS)
750 * corresponding to @index in the radix tree. If this causes
751 * the leaf node to have no tags set then clear the tag in the
752 * next-to-leaf node, etc.
754 * Returns the address of the tagged item on success, else NULL. ie:
755 * has the same return value and semantics as radix_tree_lookup().
757 void *radix_tree_tag_clear(struct radix_tree_root *root,
758 unsigned long index, unsigned int tag)
760 struct radix_tree_node *node, *parent;
761 unsigned long maxindex;
763 int uninitialized_var(offset);
765 shift = radix_tree_load_root(root, &node, &maxindex);
766 if (index > maxindex)
771 while (radix_tree_is_internal_node(node)) {
772 shift -= RADIX_TREE_MAP_SHIFT;
773 offset = (index >> shift) & RADIX_TREE_MAP_MASK;
775 parent = entry_to_node(node);
776 offset = radix_tree_descend(parent, &node, offset);
785 if (!tag_get(parent, tag, offset))
787 tag_clear(parent, tag, offset);
788 if (any_tag_set(parent, tag))
791 index >>= RADIX_TREE_MAP_SHIFT;
792 offset = index & RADIX_TREE_MAP_MASK;
793 parent = parent->parent;
796 /* clear the root's tag bit */
797 if (root_tag_get(root, tag))
798 root_tag_clear(root, tag);
803 EXPORT_SYMBOL(radix_tree_tag_clear);
806 * radix_tree_tag_get - get a tag on a radix tree node
807 * @root: radix tree root
809 * @tag: tag index (< RADIX_TREE_MAX_TAGS)
813 * 0: tag not present or not set
816 * Note that the return value of this function may not be relied on, even if
817 * the RCU lock is held, unless tag modification and node deletion are excluded
820 int radix_tree_tag_get(struct radix_tree_root *root,
821 unsigned long index, unsigned int tag)
823 struct radix_tree_node *node, *parent;
824 unsigned long maxindex;
827 if (!root_tag_get(root, tag))
830 shift = radix_tree_load_root(root, &node, &maxindex);
831 if (index > maxindex)
836 while (radix_tree_is_internal_node(node)) {
839 shift -= RADIX_TREE_MAP_SHIFT;
840 offset = (index >> shift) & RADIX_TREE_MAP_MASK;
842 parent = entry_to_node(node);
843 offset = radix_tree_descend(parent, &node, offset);
847 if (!tag_get(parent, tag, offset))
849 if (node == RADIX_TREE_RETRY)
855 EXPORT_SYMBOL(radix_tree_tag_get);
857 static inline void __set_iter_shift(struct radix_tree_iter *iter,
860 #ifdef CONFIG_RADIX_TREE_MULTIORDER
866 * radix_tree_next_chunk - find next chunk of slots for iteration
868 * @root: radix tree root
869 * @iter: iterator state
870 * @flags: RADIX_TREE_ITER_* flags and tag index
871 * Returns: pointer to chunk first slot, or NULL if iteration is over
873 void **radix_tree_next_chunk(struct radix_tree_root *root,
874 struct radix_tree_iter *iter, unsigned flags)
876 unsigned shift, tag = flags & RADIX_TREE_ITER_TAG_MASK;
877 struct radix_tree_node *node, *child;
878 unsigned long index, offset, maxindex;
880 if ((flags & RADIX_TREE_ITER_TAGGED) && !root_tag_get(root, tag))
884 * Catch next_index overflow after ~0UL. iter->index never overflows
885 * during iterating; it can be zero only at the beginning.
886 * And we cannot overflow iter->next_index in a single step,
887 * because RADIX_TREE_MAP_SHIFT < BITS_PER_LONG.
889 * This condition also used by radix_tree_next_slot() to stop
890 * contiguous iterating, and forbid swithing to the next chunk.
892 index = iter->next_index;
893 if (!index && iter->index)
897 shift = radix_tree_load_root(root, &child, &maxindex);
898 if (index > maxindex)
903 if (!radix_tree_is_internal_node(child)) {
904 /* Single-slot tree */
906 iter->next_index = maxindex + 1;
908 __set_iter_shift(iter, 0);
909 return (void **)&root->rnode;
913 node = entry_to_node(child);
914 shift -= RADIX_TREE_MAP_SHIFT;
915 offset = (index >> shift) & RADIX_TREE_MAP_MASK;
916 offset = radix_tree_descend(node, &child, offset);
918 if ((flags & RADIX_TREE_ITER_TAGGED) ?
919 !tag_get(node, tag, offset) : !child) {
921 if (flags & RADIX_TREE_ITER_CONTIG)
924 if (flags & RADIX_TREE_ITER_TAGGED)
925 offset = radix_tree_find_next_bit(
930 while (++offset < RADIX_TREE_MAP_SIZE) {
931 void *slot = node->slots[offset];
932 if (is_sibling_entry(node, slot))
937 index &= ~node_maxindex(node);
938 index += offset << shift;
939 /* Overflow after ~0UL */
942 if (offset == RADIX_TREE_MAP_SIZE)
944 child = rcu_dereference_raw(node->slots[offset]);
947 if ((child == NULL) || (child == RADIX_TREE_RETRY))
949 } while (radix_tree_is_internal_node(child));
951 /* Update the iterator state */
952 iter->index = (index &~ node_maxindex(node)) | (offset << node->shift);
953 iter->next_index = (index | node_maxindex(node)) + 1;
954 __set_iter_shift(iter, shift);
956 /* Construct iter->tags bit-mask from node->tags[tag] array */
957 if (flags & RADIX_TREE_ITER_TAGGED) {
958 unsigned tag_long, tag_bit;
960 tag_long = offset / BITS_PER_LONG;
961 tag_bit = offset % BITS_PER_LONG;
962 iter->tags = node->tags[tag][tag_long] >> tag_bit;
963 /* This never happens if RADIX_TREE_TAG_LONGS == 1 */
964 if (tag_long < RADIX_TREE_TAG_LONGS - 1) {
965 /* Pick tags from next element */
967 iter->tags |= node->tags[tag][tag_long + 1] <<
968 (BITS_PER_LONG - tag_bit);
969 /* Clip chunk size, here only BITS_PER_LONG tags */
970 iter->next_index = index + BITS_PER_LONG;
974 return node->slots + offset;
976 EXPORT_SYMBOL(radix_tree_next_chunk);
979 * radix_tree_range_tag_if_tagged - for each item in given range set given
980 * tag if item has another tag set
981 * @root: radix tree root
982 * @first_indexp: pointer to a starting index of a range to scan
983 * @last_index: last index of a range to scan
984 * @nr_to_tag: maximum number items to tag
985 * @iftag: tag index to test
986 * @settag: tag index to set if tested tag is set
988 * This function scans range of radix tree from first_index to last_index
989 * (inclusive). For each item in the range if iftag is set, the function sets
990 * also settag. The function stops either after tagging nr_to_tag items or
991 * after reaching last_index.
993 * The tags must be set from the leaf level only and propagated back up the
994 * path to the root. We must do this so that we resolve the full path before
995 * setting any tags on intermediate nodes. If we set tags as we descend, then
996 * we can get to the leaf node and find that the index that has the iftag
997 * set is outside the range we are scanning. This reults in dangling tags and
998 * can lead to problems with later tag operations (e.g. livelocks on lookups).
1000 * The function returns the number of leaves where the tag was set and sets
1001 * *first_indexp to the first unscanned index.
1002 * WARNING! *first_indexp can wrap if last_index is ULONG_MAX. Caller must
1003 * be prepared to handle that.
1005 unsigned long radix_tree_range_tag_if_tagged(struct radix_tree_root *root,
1006 unsigned long *first_indexp, unsigned long last_index,
1007 unsigned long nr_to_tag,
1008 unsigned int iftag, unsigned int settag)
1010 struct radix_tree_node *parent, *node, *child;
1011 unsigned long maxindex;
1012 unsigned int shift = radix_tree_load_root(root, &child, &maxindex);
1013 unsigned long tagged = 0;
1014 unsigned long index = *first_indexp;
1016 last_index = min(last_index, maxindex);
1017 if (index > last_index)
1021 if (!root_tag_get(root, iftag)) {
1022 *first_indexp = last_index + 1;
1025 if (!radix_tree_is_internal_node(child)) {
1026 *first_indexp = last_index + 1;
1027 root_tag_set(root, settag);
1031 node = entry_to_node(child);
1032 shift -= RADIX_TREE_MAP_SHIFT;
1035 unsigned offset = (index >> shift) & RADIX_TREE_MAP_MASK;
1036 offset = radix_tree_descend(node, &child, offset);
1039 if (!tag_get(node, iftag, offset))
1041 /* Sibling slots never have tags set on them */
1042 if (radix_tree_is_internal_node(child)) {
1043 node = entry_to_node(child);
1044 shift -= RADIX_TREE_MAP_SHIFT;
1050 tag_set(node, settag, offset);
1052 /* walk back up the path tagging interior nodes */
1055 offset = parent->offset;
1056 parent = parent->parent;
1059 /* stop if we find a node with the tag already set */
1060 if (tag_get(parent, settag, offset))
1062 tag_set(parent, settag, offset);
1065 /* Go to next item at level determined by 'shift' */
1066 index = ((index >> shift) + 1) << shift;
1067 /* Overflow can happen when last_index is ~0UL... */
1068 if (index > last_index || !index)
1070 offset = (index >> shift) & RADIX_TREE_MAP_MASK;
1071 while (offset == 0) {
1073 * We've fully scanned this node. Go up. Because
1074 * last_index is guaranteed to be in the tree, what
1075 * we do below cannot wander astray.
1077 node = node->parent;
1078 shift += RADIX_TREE_MAP_SHIFT;
1079 offset = (index >> shift) & RADIX_TREE_MAP_MASK;
1081 if (is_sibling_entry(node, node->slots[offset]))
1083 if (tagged >= nr_to_tag)
1087 * We need not to tag the root tag if there is no tag which is set with
1088 * settag within the range from *first_indexp to last_index.
1091 root_tag_set(root, settag);
1092 *first_indexp = index;
1096 EXPORT_SYMBOL(radix_tree_range_tag_if_tagged);
1099 * radix_tree_gang_lookup - perform multiple lookup on a radix tree
1100 * @root: radix tree root
1101 * @results: where the results of the lookup are placed
1102 * @first_index: start the lookup from this key
1103 * @max_items: place up to this many items at *results
1105 * Performs an index-ascending scan of the tree for present items. Places
1106 * them at *@results and returns the number of items which were placed at
1109 * The implementation is naive.
1111 * Like radix_tree_lookup, radix_tree_gang_lookup may be called under
1112 * rcu_read_lock. In this case, rather than the returned results being
1113 * an atomic snapshot of the tree at a single point in time, the
1114 * semantics of an RCU protected gang lookup are as though multiple
1115 * radix_tree_lookups have been issued in individual locks, and results
1116 * stored in 'results'.
1119 radix_tree_gang_lookup(struct radix_tree_root *root, void **results,
1120 unsigned long first_index, unsigned int max_items)
1122 struct radix_tree_iter iter;
1124 unsigned int ret = 0;
1126 if (unlikely(!max_items))
1129 radix_tree_for_each_slot(slot, root, &iter, first_index) {
1130 results[ret] = rcu_dereference_raw(*slot);
1133 if (radix_tree_is_internal_node(results[ret])) {
1134 slot = radix_tree_iter_retry(&iter);
1137 if (++ret == max_items)
1143 EXPORT_SYMBOL(radix_tree_gang_lookup);
1146 * radix_tree_gang_lookup_slot - perform multiple slot lookup on radix tree
1147 * @root: radix tree root
1148 * @results: where the results of the lookup are placed
1149 * @indices: where their indices should be placed (but usually NULL)
1150 * @first_index: start the lookup from this key
1151 * @max_items: place up to this many items at *results
1153 * Performs an index-ascending scan of the tree for present items. Places
1154 * their slots at *@results and returns the number of items which were
1155 * placed at *@results.
1157 * The implementation is naive.
1159 * Like radix_tree_gang_lookup as far as RCU and locking goes. Slots must
1160 * be dereferenced with radix_tree_deref_slot, and if using only RCU
1161 * protection, radix_tree_deref_slot may fail requiring a retry.
1164 radix_tree_gang_lookup_slot(struct radix_tree_root *root,
1165 void ***results, unsigned long *indices,
1166 unsigned long first_index, unsigned int max_items)
1168 struct radix_tree_iter iter;
1170 unsigned int ret = 0;
1172 if (unlikely(!max_items))
1175 radix_tree_for_each_slot(slot, root, &iter, first_index) {
1176 results[ret] = slot;
1178 indices[ret] = iter.index;
1179 if (++ret == max_items)
1185 EXPORT_SYMBOL(radix_tree_gang_lookup_slot);
1188 * radix_tree_gang_lookup_tag - perform multiple lookup on a radix tree
1190 * @root: radix tree root
1191 * @results: where the results of the lookup are placed
1192 * @first_index: start the lookup from this key
1193 * @max_items: place up to this many items at *results
1194 * @tag: the tag index (< RADIX_TREE_MAX_TAGS)
1196 * Performs an index-ascending scan of the tree for present items which
1197 * have the tag indexed by @tag set. Places the items at *@results and
1198 * returns the number of items which were placed at *@results.
1201 radix_tree_gang_lookup_tag(struct radix_tree_root *root, void **results,
1202 unsigned long first_index, unsigned int max_items,
1205 struct radix_tree_iter iter;
1207 unsigned int ret = 0;
1209 if (unlikely(!max_items))
1212 radix_tree_for_each_tagged(slot, root, &iter, first_index, tag) {
1213 results[ret] = rcu_dereference_raw(*slot);
1216 if (radix_tree_is_internal_node(results[ret])) {
1217 slot = radix_tree_iter_retry(&iter);
1220 if (++ret == max_items)
1226 EXPORT_SYMBOL(radix_tree_gang_lookup_tag);
1229 * radix_tree_gang_lookup_tag_slot - perform multiple slot lookup on a
1230 * radix tree based on a tag
1231 * @root: radix tree root
1232 * @results: where the results of the lookup are placed
1233 * @first_index: start the lookup from this key
1234 * @max_items: place up to this many items at *results
1235 * @tag: the tag index (< RADIX_TREE_MAX_TAGS)
1237 * Performs an index-ascending scan of the tree for present items which
1238 * have the tag indexed by @tag set. Places the slots at *@results and
1239 * returns the number of slots which were placed at *@results.
1242 radix_tree_gang_lookup_tag_slot(struct radix_tree_root *root, void ***results,
1243 unsigned long first_index, unsigned int max_items,
1246 struct radix_tree_iter iter;
1248 unsigned int ret = 0;
1250 if (unlikely(!max_items))
1253 radix_tree_for_each_tagged(slot, root, &iter, first_index, tag) {
1254 results[ret] = slot;
1255 if (++ret == max_items)
1261 EXPORT_SYMBOL(radix_tree_gang_lookup_tag_slot);
1263 #if defined(CONFIG_SHMEM) && defined(CONFIG_SWAP)
1264 #include <linux/sched.h> /* for cond_resched() */
1266 struct locate_info {
1267 unsigned long found_index;
1272 * This linear search is at present only useful to shmem_unuse_inode().
1274 static unsigned long __locate(struct radix_tree_node *slot, void *item,
1275 unsigned long index, struct locate_info *info)
1280 shift = slot->shift + RADIX_TREE_MAP_SHIFT;
1283 shift -= RADIX_TREE_MAP_SHIFT;
1285 for (i = (index >> shift) & RADIX_TREE_MAP_MASK;
1286 i < RADIX_TREE_MAP_SIZE;
1287 i++, index += (1UL << shift)) {
1288 struct radix_tree_node *node =
1289 rcu_dereference_raw(slot->slots[i]);
1290 if (node == RADIX_TREE_RETRY)
1292 if (!radix_tree_is_internal_node(node)) {
1294 info->found_index = index;
1300 node = entry_to_node(node);
1301 if (is_sibling_entry(slot, node))
1306 if (i == RADIX_TREE_MAP_SIZE)
1311 if ((index == 0) && (i == RADIX_TREE_MAP_SIZE))
1317 * radix_tree_locate_item - search through radix tree for item
1318 * @root: radix tree root
1319 * @item: item to be found
1321 * Returns index where item was found, or -1 if not found.
1322 * Caller must hold no lock (since this time-consuming function needs
1323 * to be preemptible), and must check afterwards if item is still there.
1325 unsigned long radix_tree_locate_item(struct radix_tree_root *root, void *item)
1327 struct radix_tree_node *node;
1328 unsigned long max_index;
1329 unsigned long cur_index = 0;
1330 struct locate_info info = {
1337 node = rcu_dereference_raw(root->rnode);
1338 if (!radix_tree_is_internal_node(node)) {
1341 info.found_index = 0;
1345 node = entry_to_node(node);
1347 max_index = node_maxindex(node);
1348 if (cur_index > max_index) {
1353 cur_index = __locate(node, item, cur_index, &info);
1356 } while (!info.stop && cur_index <= max_index);
1358 return info.found_index;
1361 unsigned long radix_tree_locate_item(struct radix_tree_root *root, void *item)
1365 #endif /* CONFIG_SHMEM && CONFIG_SWAP */
1368 * radix_tree_shrink - shrink radix tree to minimum height
1369 * @root radix tree root
1371 static inline bool radix_tree_shrink(struct radix_tree_root *root)
1373 bool shrunk = false;
1376 struct radix_tree_node *node = root->rnode;
1377 struct radix_tree_node *child;
1379 if (!radix_tree_is_internal_node(node))
1381 node = entry_to_node(node);
1384 * The candidate node has more than one child, or its child
1385 * is not at the leftmost slot, or the child is a multiorder
1386 * entry, we cannot shrink.
1388 if (node->count != 1)
1390 child = node->slots[0];
1393 if (!radix_tree_is_internal_node(child) && node->shift)
1396 if (radix_tree_is_internal_node(child))
1397 entry_to_node(child)->parent = NULL;
1400 * We don't need rcu_assign_pointer(), since we are simply
1401 * moving the node from one part of the tree to another: if it
1402 * was safe to dereference the old pointer to it
1403 * (node->slots[0]), it will be safe to dereference the new
1404 * one (root->rnode) as far as dependent read barriers go.
1406 root->rnode = child;
1409 * We have a dilemma here. The node's slot[0] must not be
1410 * NULLed in case there are concurrent lookups expecting to
1411 * find the item. However if this was a bottom-level node,
1412 * then it may be subject to the slot pointer being visible
1413 * to callers dereferencing it. If item corresponding to
1414 * slot[0] is subsequently deleted, these callers would expect
1415 * their slot to become empty sooner or later.
1417 * For example, lockless pagecache will look up a slot, deref
1418 * the page pointer, and if the page has 0 refcount it means it
1419 * was concurrently deleted from pagecache so try the deref
1420 * again. Fortunately there is already a requirement for logic
1421 * to retry the entire slot lookup -- the indirect pointer
1422 * problem (replacing direct root node with an indirect pointer
1423 * also results in a stale slot). So tag the slot as indirect
1424 * to force callers to retry.
1426 if (!radix_tree_is_internal_node(child))
1427 node->slots[0] = RADIX_TREE_RETRY;
1429 radix_tree_node_free(node);
1437 * __radix_tree_delete_node - try to free node after clearing a slot
1438 * @root: radix tree root
1439 * @node: node containing @index
1441 * After clearing the slot at @index in @node from radix tree
1442 * rooted at @root, call this function to attempt freeing the
1443 * node and shrinking the tree.
1445 * Returns %true if @node was freed, %false otherwise.
1447 bool __radix_tree_delete_node(struct radix_tree_root *root,
1448 struct radix_tree_node *node)
1450 bool deleted = false;
1453 struct radix_tree_node *parent;
1456 if (node == entry_to_node(root->rnode))
1457 deleted |= radix_tree_shrink(root);
1461 parent = node->parent;
1463 parent->slots[node->offset] = NULL;
1466 root_tag_clear_all(root);
1470 radix_tree_node_free(node);
1479 static inline void delete_sibling_entries(struct radix_tree_node *node,
1480 void *ptr, unsigned offset)
1482 #ifdef CONFIG_RADIX_TREE_MULTIORDER
1484 for (i = 1; offset + i < RADIX_TREE_MAP_SIZE; i++) {
1485 if (node->slots[offset + i] != ptr)
1487 node->slots[offset + i] = NULL;
1494 * radix_tree_delete_item - delete an item from a radix tree
1495 * @root: radix tree root
1497 * @item: expected item
1499 * Remove @item at @index from the radix tree rooted at @root.
1501 * Returns the address of the deleted item, or NULL if it was not present
1502 * or the entry at the given @index was not @item.
1504 void *radix_tree_delete_item(struct radix_tree_root *root,
1505 unsigned long index, void *item)
1507 struct radix_tree_node *node;
1508 unsigned int offset;
1513 entry = __radix_tree_lookup(root, index, &node, &slot);
1517 if (item && entry != item)
1521 root_tag_clear_all(root);
1526 offset = get_slot_offset(node, slot);
1529 * Clear all tags associated with the item to be deleted.
1530 * This way of doing it would be inefficient, but seldom is any set.
1532 for (tag = 0; tag < RADIX_TREE_MAX_TAGS; tag++) {
1533 if (tag_get(node, tag, offset))
1534 radix_tree_tag_clear(root, index, tag);
1537 delete_sibling_entries(node, node_to_entry(slot), offset);
1538 node->slots[offset] = NULL;
1541 __radix_tree_delete_node(root, node);
1545 EXPORT_SYMBOL(radix_tree_delete_item);
1548 * radix_tree_delete - delete an item from a radix tree
1549 * @root: radix tree root
1552 * Remove the item at @index from the radix tree rooted at @root.
1554 * Returns the address of the deleted item, or NULL if it was not present.
1556 void *radix_tree_delete(struct radix_tree_root *root, unsigned long index)
1558 return radix_tree_delete_item(root, index, NULL);
1560 EXPORT_SYMBOL(radix_tree_delete);
1563 * radix_tree_tagged - test whether any items in the tree are tagged
1564 * @root: radix tree root
1567 int radix_tree_tagged(struct radix_tree_root *root, unsigned int tag)
1569 return root_tag_get(root, tag);
1571 EXPORT_SYMBOL(radix_tree_tagged);
1574 radix_tree_node_ctor(void *arg)
1576 struct radix_tree_node *node = arg;
1578 memset(node, 0, sizeof(*node));
1579 INIT_LIST_HEAD(&node->private_list);
1582 static int radix_tree_callback(struct notifier_block *nfb,
1583 unsigned long action, void *hcpu)
1585 int cpu = (long)hcpu;
1586 struct radix_tree_preload *rtp;
1587 struct radix_tree_node *node;
1589 /* Free per-cpu pool of preloaded nodes */
1590 if (action == CPU_DEAD || action == CPU_DEAD_FROZEN) {
1591 rtp = &per_cpu(radix_tree_preloads, cpu);
1594 rtp->nodes = node->private_data;
1595 kmem_cache_free(radix_tree_node_cachep, node);
1602 void __init radix_tree_init(void)
1604 radix_tree_node_cachep = kmem_cache_create("radix_tree_node",
1605 sizeof(struct radix_tree_node), 0,
1606 SLAB_PANIC | SLAB_RECLAIM_ACCOUNT,
1607 radix_tree_node_ctor);
1608 hotcpu_notifier(radix_tree_callback, 0);