2 * Copyright (C) 2001 Momchil Velikov
3 * Portions Copyright (C) 2001 Christoph Hellwig
4 * Copyright (C) 2005 SGI, Christoph Lameter
5 * Copyright (C) 2006 Nick Piggin
6 * Copyright (C) 2012 Konstantin Khlebnikov
8 * This program is free software; you can redistribute it and/or
9 * modify it under the terms of the GNU General Public License as
10 * published by the Free Software Foundation; either version 2, or (at
11 * your option) any later version.
13 * This program is distributed in the hope that it will be useful, but
14 * WITHOUT ANY WARRANTY; without even the implied warranty of
15 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
16 * General Public License for more details.
18 * You should have received a copy of the GNU General Public License
19 * along with this program; if not, write to the Free Software
20 * Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA.
23 #include <linux/errno.h>
24 #include <linux/init.h>
25 #include <linux/kernel.h>
26 #include <linux/export.h>
27 #include <linux/radix-tree.h>
28 #include <linux/percpu.h>
29 #include <linux/slab.h>
30 #include <linux/kmemleak.h>
31 #include <linux/notifier.h>
32 #include <linux/cpu.h>
33 #include <linux/string.h>
34 #include <linux/bitops.h>
35 #include <linux/rcupdate.h>
36 #include <linux/preempt.h> /* in_interrupt() */
40 * The height_to_maxindex array needs to be one deeper than the maximum
41 * path as height 0 holds only 1 entry.
43 static unsigned long height_to_maxindex[RADIX_TREE_MAX_PATH + 1] __read_mostly;
46 * Radix tree node cache.
48 static struct kmem_cache *radix_tree_node_cachep;
51 * The radix tree is variable-height, so an insert operation not only has
52 * to build the branch to its corresponding item, it also has to build the
53 * branch to existing items if the size has to be increased (by
56 * The worst case is a zero height tree with just a single item at index 0,
57 * and then inserting an item at index ULONG_MAX. This requires 2 new branches
58 * of RADIX_TREE_MAX_PATH size to be created, with only the root node shared.
61 #define RADIX_TREE_PRELOAD_SIZE (RADIX_TREE_MAX_PATH * 2 - 1)
64 * Per-cpu pool of preloaded nodes
66 struct radix_tree_preload {
68 /* nodes->private_data points to next preallocated node */
69 struct radix_tree_node *nodes;
71 static DEFINE_PER_CPU(struct radix_tree_preload, radix_tree_preloads) = { 0, };
73 static inline void *ptr_to_indirect(void *ptr)
75 return (void *)((unsigned long)ptr | RADIX_TREE_INDIRECT_PTR);
78 #define RADIX_TREE_RETRY ptr_to_indirect(NULL)
80 #ifdef CONFIG_RADIX_TREE_MULTIORDER
81 /* Sibling slots point directly to another slot in the same node */
82 static inline bool is_sibling_entry(struct radix_tree_node *parent, void *node)
85 return (parent->slots <= ptr) &&
86 (ptr < parent->slots + RADIX_TREE_MAP_SIZE);
89 static inline bool is_sibling_entry(struct radix_tree_node *parent, void *node)
95 static inline unsigned long get_slot_offset(struct radix_tree_node *parent,
98 return slot - parent->slots;
101 static unsigned radix_tree_descend(struct radix_tree_node *parent,
102 struct radix_tree_node **nodep, unsigned offset)
104 void **entry = rcu_dereference_raw(parent->slots[offset]);
106 #ifdef CONFIG_RADIX_TREE_MULTIORDER
107 if (radix_tree_is_indirect_ptr(entry)) {
108 unsigned long siboff = get_slot_offset(parent, entry);
109 if (siboff < RADIX_TREE_MAP_SIZE) {
111 entry = rcu_dereference_raw(parent->slots[offset]);
116 *nodep = (void *)entry;
120 static inline gfp_t root_gfp_mask(struct radix_tree_root *root)
122 return root->gfp_mask & __GFP_BITS_MASK;
125 static inline void tag_set(struct radix_tree_node *node, unsigned int tag,
128 __set_bit(offset, node->tags[tag]);
131 static inline void tag_clear(struct radix_tree_node *node, unsigned int tag,
134 __clear_bit(offset, node->tags[tag]);
137 static inline int tag_get(struct radix_tree_node *node, unsigned int tag,
140 return test_bit(offset, node->tags[tag]);
143 static inline void root_tag_set(struct radix_tree_root *root, unsigned int tag)
145 root->gfp_mask |= (__force gfp_t)(1 << (tag + __GFP_BITS_SHIFT));
148 static inline void root_tag_clear(struct radix_tree_root *root, unsigned int tag)
150 root->gfp_mask &= (__force gfp_t)~(1 << (tag + __GFP_BITS_SHIFT));
153 static inline void root_tag_clear_all(struct radix_tree_root *root)
155 root->gfp_mask &= __GFP_BITS_MASK;
158 static inline int root_tag_get(struct radix_tree_root *root, unsigned int tag)
160 return (__force unsigned)root->gfp_mask & (1 << (tag + __GFP_BITS_SHIFT));
163 static inline unsigned root_tags_get(struct radix_tree_root *root)
165 return (__force unsigned)root->gfp_mask >> __GFP_BITS_SHIFT;
169 * Returns 1 if any slot in the node has this tag set.
170 * Otherwise returns 0.
172 static inline int any_tag_set(struct radix_tree_node *node, unsigned int tag)
175 for (idx = 0; idx < RADIX_TREE_TAG_LONGS; idx++) {
176 if (node->tags[tag][idx])
183 * radix_tree_find_next_bit - find the next set bit in a memory region
185 * @addr: The address to base the search on
186 * @size: The bitmap size in bits
187 * @offset: The bitnumber to start searching at
189 * Unrollable variant of find_next_bit() for constant size arrays.
190 * Tail bits starting from size to roundup(size, BITS_PER_LONG) must be zero.
191 * Returns next bit offset, or size if nothing found.
193 static __always_inline unsigned long
194 radix_tree_find_next_bit(const unsigned long *addr,
195 unsigned long size, unsigned long offset)
197 if (!__builtin_constant_p(size))
198 return find_next_bit(addr, size, offset);
203 addr += offset / BITS_PER_LONG;
204 tmp = *addr >> (offset % BITS_PER_LONG);
206 return __ffs(tmp) + offset;
207 offset = (offset + BITS_PER_LONG) & ~(BITS_PER_LONG - 1);
208 while (offset < size) {
211 return __ffs(tmp) + offset;
212 offset += BITS_PER_LONG;
219 static void dump_node(struct radix_tree_node *node, unsigned offset,
220 unsigned shift, unsigned long index)
224 pr_debug("radix node: %p offset %d tags %lx %lx %lx path %x count %d parent %p\n",
226 node->tags[0][0], node->tags[1][0], node->tags[2][0],
227 node->path, node->count, node->parent);
229 for (i = 0; i < RADIX_TREE_MAP_SIZE; i++) {
230 unsigned long first = index | (i << shift);
231 unsigned long last = first | ((1UL << shift) - 1);
232 void *entry = node->slots[i];
235 if (is_sibling_entry(node, entry)) {
236 pr_debug("radix sblng %p offset %ld val %p indices %ld-%ld\n",
238 *(void **)indirect_to_ptr(entry),
240 } else if (!radix_tree_is_indirect_ptr(entry)) {
241 pr_debug("radix entry %p offset %ld indices %ld-%ld\n",
242 entry, i, first, last);
244 dump_node(indirect_to_ptr(entry), i,
245 shift - RADIX_TREE_MAP_SHIFT, first);
251 static void radix_tree_dump(struct radix_tree_root *root)
253 pr_debug("radix root: %p height %d rnode %p tags %x\n",
254 root, root->height, root->rnode,
255 root->gfp_mask >> __GFP_BITS_SHIFT);
256 if (!radix_tree_is_indirect_ptr(root->rnode))
258 dump_node(indirect_to_ptr(root->rnode), 0,
259 (root->height - 1) * RADIX_TREE_MAP_SHIFT, 0);
264 * This assumes that the caller has performed appropriate preallocation, and
265 * that the caller has pinned this thread of control to the current CPU.
267 static struct radix_tree_node *
268 radix_tree_node_alloc(struct radix_tree_root *root)
270 struct radix_tree_node *ret = NULL;
271 gfp_t gfp_mask = root_gfp_mask(root);
274 * Preload code isn't irq safe and it doesn't make sence to use
275 * preloading in the interrupt anyway as all the allocations have to
276 * be atomic. So just do normal allocation when in interrupt.
278 if (!gfpflags_allow_blocking(gfp_mask) && !in_interrupt()) {
279 struct radix_tree_preload *rtp;
282 * Even if the caller has preloaded, try to allocate from the
283 * cache first for the new node to get accounted.
285 ret = kmem_cache_alloc(radix_tree_node_cachep,
286 gfp_mask | __GFP_ACCOUNT | __GFP_NOWARN);
291 * Provided the caller has preloaded here, we will always
292 * succeed in getting a node here (and never reach
295 rtp = this_cpu_ptr(&radix_tree_preloads);
298 rtp->nodes = ret->private_data;
299 ret->private_data = NULL;
303 * Update the allocation stack trace as this is more useful
306 kmemleak_update_trace(ret);
309 ret = kmem_cache_alloc(radix_tree_node_cachep,
310 gfp_mask | __GFP_ACCOUNT);
312 BUG_ON(radix_tree_is_indirect_ptr(ret));
316 static void radix_tree_node_rcu_free(struct rcu_head *head)
318 struct radix_tree_node *node =
319 container_of(head, struct radix_tree_node, rcu_head);
323 * must only free zeroed nodes into the slab. radix_tree_shrink
324 * can leave us with a non-NULL entry in the first slot, so clear
325 * that here to make sure.
327 for (i = 0; i < RADIX_TREE_MAX_TAGS; i++)
328 tag_clear(node, i, 0);
330 node->slots[0] = NULL;
333 kmem_cache_free(radix_tree_node_cachep, node);
337 radix_tree_node_free(struct radix_tree_node *node)
339 call_rcu(&node->rcu_head, radix_tree_node_rcu_free);
343 * Load up this CPU's radix_tree_node buffer with sufficient objects to
344 * ensure that the addition of a single element in the tree cannot fail. On
345 * success, return zero, with preemption disabled. On error, return -ENOMEM
346 * with preemption not disabled.
348 * To make use of this facility, the radix tree must be initialised without
349 * __GFP_DIRECT_RECLAIM being passed to INIT_RADIX_TREE().
351 static int __radix_tree_preload(gfp_t gfp_mask)
353 struct radix_tree_preload *rtp;
354 struct radix_tree_node *node;
358 rtp = this_cpu_ptr(&radix_tree_preloads);
359 while (rtp->nr < RADIX_TREE_PRELOAD_SIZE) {
361 node = kmem_cache_alloc(radix_tree_node_cachep, gfp_mask);
365 rtp = this_cpu_ptr(&radix_tree_preloads);
366 if (rtp->nr < RADIX_TREE_PRELOAD_SIZE) {
367 node->private_data = rtp->nodes;
371 kmem_cache_free(radix_tree_node_cachep, node);
380 * Load up this CPU's radix_tree_node buffer with sufficient objects to
381 * ensure that the addition of a single element in the tree cannot fail. On
382 * success, return zero, with preemption disabled. On error, return -ENOMEM
383 * with preemption not disabled.
385 * To make use of this facility, the radix tree must be initialised without
386 * __GFP_DIRECT_RECLAIM being passed to INIT_RADIX_TREE().
388 int radix_tree_preload(gfp_t gfp_mask)
390 /* Warn on non-sensical use... */
391 WARN_ON_ONCE(!gfpflags_allow_blocking(gfp_mask));
392 return __radix_tree_preload(gfp_mask);
394 EXPORT_SYMBOL(radix_tree_preload);
397 * The same as above function, except we don't guarantee preloading happens.
398 * We do it, if we decide it helps. On success, return zero with preemption
399 * disabled. On error, return -ENOMEM with preemption not disabled.
401 int radix_tree_maybe_preload(gfp_t gfp_mask)
403 if (gfpflags_allow_blocking(gfp_mask))
404 return __radix_tree_preload(gfp_mask);
405 /* Preloading doesn't help anything with this gfp mask, skip it */
409 EXPORT_SYMBOL(radix_tree_maybe_preload);
412 * Return the maximum key which can be store into a
413 * radix tree with height HEIGHT.
415 static inline unsigned long radix_tree_maxindex(unsigned int height)
417 return height_to_maxindex[height];
420 static inline unsigned long node_maxindex(struct radix_tree_node *node)
422 return radix_tree_maxindex(node->path & RADIX_TREE_HEIGHT_MASK);
425 static unsigned radix_tree_load_root(struct radix_tree_root *root,
426 struct radix_tree_node **nodep, unsigned long *maxindex)
428 struct radix_tree_node *node = rcu_dereference_raw(root->rnode);
432 if (likely(radix_tree_is_indirect_ptr(node))) {
433 node = indirect_to_ptr(node);
434 *maxindex = node_maxindex(node);
435 return (node->path & RADIX_TREE_HEIGHT_MASK) *
436 RADIX_TREE_MAP_SHIFT;
444 * Extend a radix tree so it can store key @index.
446 static int radix_tree_extend(struct radix_tree_root *root,
449 struct radix_tree_node *node;
450 struct radix_tree_node *slot;
454 /* Figure out what the height should be. */
455 height = root->height + 1;
456 while (index > radix_tree_maxindex(height))
459 if (root->rnode == NULL) {
460 root->height = height;
465 unsigned int newheight;
466 if (!(node = radix_tree_node_alloc(root)))
469 /* Propagate the aggregated tag info into the new root */
470 for (tag = 0; tag < RADIX_TREE_MAX_TAGS; tag++) {
471 if (root_tag_get(root, tag))
472 tag_set(node, tag, 0);
475 /* Increase the height. */
476 newheight = root->height+1;
477 BUG_ON(newheight & ~RADIX_TREE_HEIGHT_MASK);
478 node->path = newheight;
482 if (radix_tree_is_indirect_ptr(slot)) {
483 slot = indirect_to_ptr(slot);
485 slot = ptr_to_indirect(slot);
487 node->slots[0] = slot;
488 node = ptr_to_indirect(node);
489 rcu_assign_pointer(root->rnode, node);
490 root->height = newheight;
491 } while (height > root->height);
493 return height * RADIX_TREE_MAP_SHIFT;
497 * __radix_tree_create - create a slot in a radix tree
498 * @root: radix tree root
500 * @order: index occupies 2^order aligned slots
501 * @nodep: returns node
502 * @slotp: returns slot
504 * Create, if necessary, and return the node and slot for an item
505 * at position @index in the radix tree @root.
507 * Until there is more than one item in the tree, no nodes are
508 * allocated and @root->rnode is used as a direct slot instead of
509 * pointing to a node, in which case *@nodep will be NULL.
511 * Returns -ENOMEM, or 0 for success.
513 int __radix_tree_create(struct radix_tree_root *root, unsigned long index,
514 unsigned order, struct radix_tree_node **nodep,
517 struct radix_tree_node *node = NULL, *slot;
518 unsigned long maxindex;
519 unsigned int height, shift, offset;
520 unsigned long max = index | ((1UL << order) - 1);
522 shift = radix_tree_load_root(root, &slot, &maxindex);
524 /* Make sure the tree is high enough. */
525 if (max > maxindex) {
526 int error = radix_tree_extend(root, max);
531 if (order == shift) {
532 shift += RADIX_TREE_MAP_SHIFT;
537 height = root->height;
539 offset = 0; /* uninitialised var warning */
540 while (shift > order) {
542 /* Have to add a child node. */
543 if (!(slot = radix_tree_node_alloc(root)))
548 rcu_assign_pointer(node->slots[offset],
549 ptr_to_indirect(slot));
551 slot->path |= offset << RADIX_TREE_HEIGHT_SHIFT;
553 rcu_assign_pointer(root->rnode,
554 ptr_to_indirect(slot));
555 } else if (!radix_tree_is_indirect_ptr(slot))
558 /* Go a level down */
560 shift -= RADIX_TREE_MAP_SHIFT;
561 node = indirect_to_ptr(slot);
562 offset = (index >> shift) & RADIX_TREE_MAP_MASK;
563 offset = radix_tree_descend(node, &slot, offset);
566 #ifdef CONFIG_RADIX_TREE_MULTIORDER
567 /* Insert pointers to the canonical entry */
569 int i, n = 1 << (order - shift);
570 offset = offset & ~(n - 1);
571 slot = ptr_to_indirect(&node->slots[offset]);
572 for (i = 0; i < n; i++) {
573 if (node->slots[offset + i])
577 for (i = 1; i < n; i++) {
578 rcu_assign_pointer(node->slots[offset + i], slot);
587 *slotp = node ? node->slots + offset : (void **)&root->rnode;
592 * __radix_tree_insert - insert into a radix tree
593 * @root: radix tree root
595 * @order: key covers the 2^order indices around index
596 * @item: item to insert
598 * Insert an item into the radix tree at position @index.
600 int __radix_tree_insert(struct radix_tree_root *root, unsigned long index,
601 unsigned order, void *item)
603 struct radix_tree_node *node;
607 BUG_ON(radix_tree_is_indirect_ptr(item));
609 error = __radix_tree_create(root, index, order, &node, &slot);
614 rcu_assign_pointer(*slot, item);
617 unsigned offset = get_slot_offset(node, slot);
619 BUG_ON(tag_get(node, 0, offset));
620 BUG_ON(tag_get(node, 1, offset));
621 BUG_ON(tag_get(node, 2, offset));
623 BUG_ON(root_tags_get(root));
628 EXPORT_SYMBOL(__radix_tree_insert);
631 * __radix_tree_lookup - lookup an item in a radix tree
632 * @root: radix tree root
634 * @nodep: returns node
635 * @slotp: returns slot
637 * Lookup and return the item at position @index in the radix
640 * Until there is more than one item in the tree, no nodes are
641 * allocated and @root->rnode is used as a direct slot instead of
642 * pointing to a node, in which case *@nodep will be NULL.
644 void *__radix_tree_lookup(struct radix_tree_root *root, unsigned long index,
645 struct radix_tree_node **nodep, void ***slotp)
647 struct radix_tree_node *node, *parent;
648 unsigned long maxindex;
654 slot = (void **)&root->rnode;
655 shift = radix_tree_load_root(root, &node, &maxindex);
656 if (index > maxindex)
659 while (radix_tree_is_indirect_ptr(node)) {
662 if (node == RADIX_TREE_RETRY)
664 parent = indirect_to_ptr(node);
665 shift -= RADIX_TREE_MAP_SHIFT;
666 offset = (index >> shift) & RADIX_TREE_MAP_MASK;
667 offset = radix_tree_descend(parent, &node, offset);
668 slot = parent->slots + offset;
679 * radix_tree_lookup_slot - lookup a slot in a radix tree
680 * @root: radix tree root
683 * Returns: the slot corresponding to the position @index in the
684 * radix tree @root. This is useful for update-if-exists operations.
686 * This function can be called under rcu_read_lock iff the slot is not
687 * modified by radix_tree_replace_slot, otherwise it must be called
688 * exclusive from other writers. Any dereference of the slot must be done
689 * using radix_tree_deref_slot.
691 void **radix_tree_lookup_slot(struct radix_tree_root *root, unsigned long index)
695 if (!__radix_tree_lookup(root, index, NULL, &slot))
699 EXPORT_SYMBOL(radix_tree_lookup_slot);
702 * radix_tree_lookup - perform lookup operation on a radix tree
703 * @root: radix tree root
706 * Lookup the item at the position @index in the radix tree @root.
708 * This function can be called under rcu_read_lock, however the caller
709 * must manage lifetimes of leaf nodes (eg. RCU may also be used to free
710 * them safely). No RCU barriers are required to access or modify the
711 * returned item, however.
713 void *radix_tree_lookup(struct radix_tree_root *root, unsigned long index)
715 return __radix_tree_lookup(root, index, NULL, NULL);
717 EXPORT_SYMBOL(radix_tree_lookup);
720 * radix_tree_tag_set - set a tag on a radix tree node
721 * @root: radix tree root
725 * Set the search tag (which must be < RADIX_TREE_MAX_TAGS)
726 * corresponding to @index in the radix tree. From
727 * the root all the way down to the leaf node.
729 * Returns the address of the tagged item. Setting a tag on a not-present
732 void *radix_tree_tag_set(struct radix_tree_root *root,
733 unsigned long index, unsigned int tag)
735 struct radix_tree_node *node, *parent;
736 unsigned long maxindex;
739 shift = radix_tree_load_root(root, &node, &maxindex);
740 BUG_ON(index > maxindex);
742 while (radix_tree_is_indirect_ptr(node)) {
745 shift -= RADIX_TREE_MAP_SHIFT;
746 offset = (index >> shift) & RADIX_TREE_MAP_MASK;
748 parent = indirect_to_ptr(node);
749 offset = radix_tree_descend(parent, &node, offset);
752 if (!tag_get(parent, tag, offset))
753 tag_set(parent, tag, offset);
756 /* set the root's tag bit */
757 if (!root_tag_get(root, tag))
758 root_tag_set(root, tag);
762 EXPORT_SYMBOL(radix_tree_tag_set);
765 * radix_tree_tag_clear - clear a tag on a radix tree node
766 * @root: radix tree root
770 * Clear the search tag (which must be < RADIX_TREE_MAX_TAGS)
771 * corresponding to @index in the radix tree. If
772 * this causes the leaf node to have no tags set then clear the tag in the
773 * next-to-leaf node, etc.
775 * Returns the address of the tagged item on success, else NULL. ie:
776 * has the same return value and semantics as radix_tree_lookup().
778 void *radix_tree_tag_clear(struct radix_tree_root *root,
779 unsigned long index, unsigned int tag)
781 struct radix_tree_node *node, *parent;
782 unsigned long maxindex;
784 int uninitialized_var(offset);
786 shift = radix_tree_load_root(root, &node, &maxindex);
787 if (index > maxindex)
792 while (radix_tree_is_indirect_ptr(node)) {
793 shift -= RADIX_TREE_MAP_SHIFT;
794 offset = (index >> shift) & RADIX_TREE_MAP_MASK;
796 parent = indirect_to_ptr(node);
797 offset = radix_tree_descend(parent, &node, offset);
806 if (!tag_get(parent, tag, offset))
808 tag_clear(parent, tag, offset);
809 if (any_tag_set(parent, tag))
812 index >>= RADIX_TREE_MAP_SHIFT;
813 offset = index & RADIX_TREE_MAP_MASK;
814 parent = parent->parent;
817 /* clear the root's tag bit */
818 if (root_tag_get(root, tag))
819 root_tag_clear(root, tag);
824 EXPORT_SYMBOL(radix_tree_tag_clear);
827 * radix_tree_tag_get - get a tag on a radix tree node
828 * @root: radix tree root
830 * @tag: tag index (< RADIX_TREE_MAX_TAGS)
834 * 0: tag not present or not set
837 * Note that the return value of this function may not be relied on, even if
838 * the RCU lock is held, unless tag modification and node deletion are excluded
841 int radix_tree_tag_get(struct radix_tree_root *root,
842 unsigned long index, unsigned int tag)
844 struct radix_tree_node *node, *parent;
845 unsigned long maxindex;
848 if (!root_tag_get(root, tag))
851 shift = radix_tree_load_root(root, &node, &maxindex);
852 if (index > maxindex)
857 while (radix_tree_is_indirect_ptr(node)) {
860 shift -= RADIX_TREE_MAP_SHIFT;
861 offset = (index >> shift) & RADIX_TREE_MAP_MASK;
863 parent = indirect_to_ptr(node);
864 offset = radix_tree_descend(parent, &node, offset);
868 if (!tag_get(parent, tag, offset))
870 if (node == RADIX_TREE_RETRY)
876 EXPORT_SYMBOL(radix_tree_tag_get);
878 static inline void __set_iter_shift(struct radix_tree_iter *iter,
881 #ifdef CONFIG_RADIX_TREE_MULTIORDER
887 * radix_tree_next_chunk - find next chunk of slots for iteration
889 * @root: radix tree root
890 * @iter: iterator state
891 * @flags: RADIX_TREE_ITER_* flags and tag index
892 * Returns: pointer to chunk first slot, or NULL if iteration is over
894 void **radix_tree_next_chunk(struct radix_tree_root *root,
895 struct radix_tree_iter *iter, unsigned flags)
897 unsigned shift, tag = flags & RADIX_TREE_ITER_TAG_MASK;
898 struct radix_tree_node *rnode, *node;
899 unsigned long index, offset, maxindex;
901 if ((flags & RADIX_TREE_ITER_TAGGED) && !root_tag_get(root, tag))
905 * Catch next_index overflow after ~0UL. iter->index never overflows
906 * during iterating; it can be zero only at the beginning.
907 * And we cannot overflow iter->next_index in a single step,
908 * because RADIX_TREE_MAP_SHIFT < BITS_PER_LONG.
910 * This condition also used by radix_tree_next_slot() to stop
911 * contiguous iterating, and forbid swithing to the next chunk.
913 index = iter->next_index;
914 if (!index && iter->index)
918 shift = radix_tree_load_root(root, &rnode, &maxindex);
919 if (index > maxindex)
922 if (radix_tree_is_indirect_ptr(rnode)) {
923 rnode = indirect_to_ptr(rnode);
925 /* Single-slot tree */
927 iter->next_index = maxindex + 1;
929 __set_iter_shift(iter, shift);
930 return (void **)&root->rnode;
934 shift -= RADIX_TREE_MAP_SHIFT;
935 offset = index >> shift;
939 struct radix_tree_node *slot;
940 unsigned new_off = radix_tree_descend(node, &slot, offset);
942 if (new_off < offset) {
944 index &= ~((RADIX_TREE_MAP_SIZE << shift) - 1);
945 index |= offset << shift;
948 if ((flags & RADIX_TREE_ITER_TAGGED) ?
949 !tag_get(node, tag, offset) : !slot) {
951 if (flags & RADIX_TREE_ITER_CONTIG)
954 if (flags & RADIX_TREE_ITER_TAGGED)
955 offset = radix_tree_find_next_bit(
960 while (++offset < RADIX_TREE_MAP_SIZE) {
961 void *slot = node->slots[offset];
962 if (is_sibling_entry(node, slot))
967 index &= ~((RADIX_TREE_MAP_SIZE << shift) - 1);
968 index += offset << shift;
969 /* Overflow after ~0UL */
972 if (offset == RADIX_TREE_MAP_SIZE)
974 slot = rcu_dereference_raw(node->slots[offset]);
977 if ((slot == NULL) || (slot == RADIX_TREE_RETRY))
979 if (!radix_tree_is_indirect_ptr(slot))
982 node = indirect_to_ptr(slot);
983 shift -= RADIX_TREE_MAP_SHIFT;
984 offset = (index >> shift) & RADIX_TREE_MAP_MASK;
987 /* Update the iterator state */
988 iter->index = index & ~((1 << shift) - 1);
989 iter->next_index = (index | ((RADIX_TREE_MAP_SIZE << shift) - 1)) + 1;
990 __set_iter_shift(iter, shift);
992 /* Construct iter->tags bit-mask from node->tags[tag] array */
993 if (flags & RADIX_TREE_ITER_TAGGED) {
994 unsigned tag_long, tag_bit;
996 tag_long = offset / BITS_PER_LONG;
997 tag_bit = offset % BITS_PER_LONG;
998 iter->tags = node->tags[tag][tag_long] >> tag_bit;
999 /* This never happens if RADIX_TREE_TAG_LONGS == 1 */
1000 if (tag_long < RADIX_TREE_TAG_LONGS - 1) {
1001 /* Pick tags from next element */
1003 iter->tags |= node->tags[tag][tag_long + 1] <<
1004 (BITS_PER_LONG - tag_bit);
1005 /* Clip chunk size, here only BITS_PER_LONG tags */
1006 iter->next_index = index + BITS_PER_LONG;
1010 return node->slots + offset;
1012 EXPORT_SYMBOL(radix_tree_next_chunk);
1015 * radix_tree_range_tag_if_tagged - for each item in given range set given
1016 * tag if item has another tag set
1017 * @root: radix tree root
1018 * @first_indexp: pointer to a starting index of a range to scan
1019 * @last_index: last index of a range to scan
1020 * @nr_to_tag: maximum number items to tag
1021 * @iftag: tag index to test
1022 * @settag: tag index to set if tested tag is set
1024 * This function scans range of radix tree from first_index to last_index
1025 * (inclusive). For each item in the range if iftag is set, the function sets
1026 * also settag. The function stops either after tagging nr_to_tag items or
1027 * after reaching last_index.
1029 * The tags must be set from the leaf level only and propagated back up the
1030 * path to the root. We must do this so that we resolve the full path before
1031 * setting any tags on intermediate nodes. If we set tags as we descend, then
1032 * we can get to the leaf node and find that the index that has the iftag
1033 * set is outside the range we are scanning. This reults in dangling tags and
1034 * can lead to problems with later tag operations (e.g. livelocks on lookups).
1036 * The function returns number of leaves where the tag was set and sets
1037 * *first_indexp to the first unscanned index.
1038 * WARNING! *first_indexp can wrap if last_index is ULONG_MAX. Caller must
1039 * be prepared to handle that.
1041 unsigned long radix_tree_range_tag_if_tagged(struct radix_tree_root *root,
1042 unsigned long *first_indexp, unsigned long last_index,
1043 unsigned long nr_to_tag,
1044 unsigned int iftag, unsigned int settag)
1046 struct radix_tree_node *slot, *node = NULL;
1047 unsigned long maxindex;
1048 unsigned int shift = radix_tree_load_root(root, &slot, &maxindex);
1049 unsigned long tagged = 0;
1050 unsigned long index = *first_indexp;
1052 last_index = min(last_index, maxindex);
1053 if (index > last_index)
1057 if (!root_tag_get(root, iftag)) {
1058 *first_indexp = last_index + 1;
1061 if (!radix_tree_is_indirect_ptr(slot)) {
1062 *first_indexp = last_index + 1;
1063 root_tag_set(root, settag);
1067 node = indirect_to_ptr(slot);
1068 shift -= RADIX_TREE_MAP_SHIFT;
1071 unsigned long upindex;
1074 offset = (index >> shift) & RADIX_TREE_MAP_MASK;
1075 offset = radix_tree_descend(node, &slot, offset);
1078 if (!tag_get(node, iftag, offset))
1080 /* Sibling slots never have tags set on them */
1081 if (radix_tree_is_indirect_ptr(slot)) {
1082 node = indirect_to_ptr(slot);
1083 shift -= RADIX_TREE_MAP_SHIFT;
1089 tag_set(node, settag, offset);
1091 slot = node->parent;
1092 /* walk back up the path tagging interior nodes */
1093 upindex = index >> shift;
1095 upindex >>= RADIX_TREE_MAP_SHIFT;
1096 offset = upindex & RADIX_TREE_MAP_MASK;
1098 /* stop if we find a node with the tag already set */
1099 if (tag_get(slot, settag, offset))
1101 tag_set(slot, settag, offset);
1102 slot = slot->parent;
1106 /* Go to next item at level determined by 'shift' */
1107 index = ((index >> shift) + 1) << shift;
1108 /* Overflow can happen when last_index is ~0UL... */
1109 if (index > last_index || !index)
1111 offset = (index >> shift) & RADIX_TREE_MAP_MASK;
1112 while (offset == 0) {
1114 * We've fully scanned this node. Go up. Because
1115 * last_index is guaranteed to be in the tree, what
1116 * we do below cannot wander astray.
1118 node = node->parent;
1119 shift += RADIX_TREE_MAP_SHIFT;
1120 offset = (index >> shift) & RADIX_TREE_MAP_MASK;
1122 if (is_sibling_entry(node, node->slots[offset]))
1124 if (tagged >= nr_to_tag)
1128 * We need not to tag the root tag if there is no tag which is set with
1129 * settag within the range from *first_indexp to last_index.
1132 root_tag_set(root, settag);
1133 *first_indexp = index;
1137 EXPORT_SYMBOL(radix_tree_range_tag_if_tagged);
1140 * radix_tree_gang_lookup - perform multiple lookup on a radix tree
1141 * @root: radix tree root
1142 * @results: where the results of the lookup are placed
1143 * @first_index: start the lookup from this key
1144 * @max_items: place up to this many items at *results
1146 * Performs an index-ascending scan of the tree for present items. Places
1147 * them at *@results and returns the number of items which were placed at
1150 * The implementation is naive.
1152 * Like radix_tree_lookup, radix_tree_gang_lookup may be called under
1153 * rcu_read_lock. In this case, rather than the returned results being
1154 * an atomic snapshot of the tree at a single point in time, the semantics
1155 * of an RCU protected gang lookup are as though multiple radix_tree_lookups
1156 * have been issued in individual locks, and results stored in 'results'.
1159 radix_tree_gang_lookup(struct radix_tree_root *root, void **results,
1160 unsigned long first_index, unsigned int max_items)
1162 struct radix_tree_iter iter;
1164 unsigned int ret = 0;
1166 if (unlikely(!max_items))
1169 radix_tree_for_each_slot(slot, root, &iter, first_index) {
1170 results[ret] = rcu_dereference_raw(*slot);
1173 if (radix_tree_is_indirect_ptr(results[ret])) {
1174 slot = radix_tree_iter_retry(&iter);
1177 if (++ret == max_items)
1183 EXPORT_SYMBOL(radix_tree_gang_lookup);
1186 * radix_tree_gang_lookup_slot - perform multiple slot lookup on radix tree
1187 * @root: radix tree root
1188 * @results: where the results of the lookup are placed
1189 * @indices: where their indices should be placed (but usually NULL)
1190 * @first_index: start the lookup from this key
1191 * @max_items: place up to this many items at *results
1193 * Performs an index-ascending scan of the tree for present items. Places
1194 * their slots at *@results and returns the number of items which were
1195 * placed at *@results.
1197 * The implementation is naive.
1199 * Like radix_tree_gang_lookup as far as RCU and locking goes. Slots must
1200 * be dereferenced with radix_tree_deref_slot, and if using only RCU
1201 * protection, radix_tree_deref_slot may fail requiring a retry.
1204 radix_tree_gang_lookup_slot(struct radix_tree_root *root,
1205 void ***results, unsigned long *indices,
1206 unsigned long first_index, unsigned int max_items)
1208 struct radix_tree_iter iter;
1210 unsigned int ret = 0;
1212 if (unlikely(!max_items))
1215 radix_tree_for_each_slot(slot, root, &iter, first_index) {
1216 results[ret] = slot;
1218 indices[ret] = iter.index;
1219 if (++ret == max_items)
1225 EXPORT_SYMBOL(radix_tree_gang_lookup_slot);
1228 * radix_tree_gang_lookup_tag - perform multiple lookup on a radix tree
1230 * @root: radix tree root
1231 * @results: where the results of the lookup are placed
1232 * @first_index: start the lookup from this key
1233 * @max_items: place up to this many items at *results
1234 * @tag: the tag index (< RADIX_TREE_MAX_TAGS)
1236 * Performs an index-ascending scan of the tree for present items which
1237 * have the tag indexed by @tag set. Places the items at *@results and
1238 * returns the number of items which were placed at *@results.
1241 radix_tree_gang_lookup_tag(struct radix_tree_root *root, void **results,
1242 unsigned long first_index, unsigned int max_items,
1245 struct radix_tree_iter iter;
1247 unsigned int ret = 0;
1249 if (unlikely(!max_items))
1252 radix_tree_for_each_tagged(slot, root, &iter, first_index, tag) {
1253 results[ret] = rcu_dereference_raw(*slot);
1256 if (radix_tree_is_indirect_ptr(results[ret])) {
1257 slot = radix_tree_iter_retry(&iter);
1260 if (++ret == max_items)
1266 EXPORT_SYMBOL(radix_tree_gang_lookup_tag);
1269 * radix_tree_gang_lookup_tag_slot - perform multiple slot lookup on a
1270 * radix tree based on a tag
1271 * @root: radix tree root
1272 * @results: where the results of the lookup are placed
1273 * @first_index: start the lookup from this key
1274 * @max_items: place up to this many items at *results
1275 * @tag: the tag index (< RADIX_TREE_MAX_TAGS)
1277 * Performs an index-ascending scan of the tree for present items which
1278 * have the tag indexed by @tag set. Places the slots at *@results and
1279 * returns the number of slots which were placed at *@results.
1282 radix_tree_gang_lookup_tag_slot(struct radix_tree_root *root, void ***results,
1283 unsigned long first_index, unsigned int max_items,
1286 struct radix_tree_iter iter;
1288 unsigned int ret = 0;
1290 if (unlikely(!max_items))
1293 radix_tree_for_each_tagged(slot, root, &iter, first_index, tag) {
1294 results[ret] = slot;
1295 if (++ret == max_items)
1301 EXPORT_SYMBOL(radix_tree_gang_lookup_tag_slot);
1303 #if defined(CONFIG_SHMEM) && defined(CONFIG_SWAP)
1304 #include <linux/sched.h> /* for cond_resched() */
1306 struct locate_info {
1307 unsigned long found_index;
1312 * This linear search is at present only useful to shmem_unuse_inode().
1314 static unsigned long __locate(struct radix_tree_node *slot, void *item,
1315 unsigned long index, struct locate_info *info)
1317 unsigned int shift, height;
1320 height = slot->path & RADIX_TREE_HEIGHT_MASK;
1321 shift = height * RADIX_TREE_MAP_SHIFT;
1324 shift -= RADIX_TREE_MAP_SHIFT;
1326 for (i = (index >> shift) & RADIX_TREE_MAP_MASK;
1327 i < RADIX_TREE_MAP_SIZE;
1328 i++, index += (1UL << shift)) {
1329 struct radix_tree_node *node =
1330 rcu_dereference_raw(slot->slots[i]);
1331 if (node == RADIX_TREE_RETRY)
1333 if (!radix_tree_is_indirect_ptr(node)) {
1335 info->found_index = index;
1341 node = indirect_to_ptr(node);
1342 if (is_sibling_entry(slot, node))
1347 if (i == RADIX_TREE_MAP_SIZE)
1352 if ((index == 0) && (i == RADIX_TREE_MAP_SIZE))
1358 * radix_tree_locate_item - search through radix tree for item
1359 * @root: radix tree root
1360 * @item: item to be found
1362 * Returns index where item was found, or -1 if not found.
1363 * Caller must hold no lock (since this time-consuming function needs
1364 * to be preemptible), and must check afterwards if item is still there.
1366 unsigned long radix_tree_locate_item(struct radix_tree_root *root, void *item)
1368 struct radix_tree_node *node;
1369 unsigned long max_index;
1370 unsigned long cur_index = 0;
1371 struct locate_info info = {
1378 node = rcu_dereference_raw(root->rnode);
1379 if (!radix_tree_is_indirect_ptr(node)) {
1382 info.found_index = 0;
1386 node = indirect_to_ptr(node);
1388 max_index = node_maxindex(node);
1389 if (cur_index > max_index) {
1394 cur_index = __locate(node, item, cur_index, &info);
1397 } while (!info.stop && cur_index <= max_index);
1399 return info.found_index;
1402 unsigned long radix_tree_locate_item(struct radix_tree_root *root, void *item)
1406 #endif /* CONFIG_SHMEM && CONFIG_SWAP */
1409 * radix_tree_shrink - shrink height of a radix tree to minimal
1410 * @root radix tree root
1412 static inline void radix_tree_shrink(struct radix_tree_root *root)
1414 /* try to shrink tree height */
1415 while (root->height > 0) {
1416 struct radix_tree_node *to_free = root->rnode;
1417 struct radix_tree_node *slot;
1419 BUG_ON(!radix_tree_is_indirect_ptr(to_free));
1420 to_free = indirect_to_ptr(to_free);
1423 * The candidate node has more than one child, or its child
1424 * is not at the leftmost slot, or it is a multiorder entry,
1427 if (to_free->count != 1)
1429 slot = to_free->slots[0];
1432 if (!radix_tree_is_indirect_ptr(slot) && (root->height > 1))
1435 if (radix_tree_is_indirect_ptr(slot)) {
1436 slot = indirect_to_ptr(slot);
1437 slot->parent = NULL;
1438 slot = ptr_to_indirect(slot);
1442 * We don't need rcu_assign_pointer(), since we are simply
1443 * moving the node from one part of the tree to another: if it
1444 * was safe to dereference the old pointer to it
1445 * (to_free->slots[0]), it will be safe to dereference the new
1446 * one (root->rnode) as far as dependent read barriers go.
1452 * We have a dilemma here. The node's slot[0] must not be
1453 * NULLed in case there are concurrent lookups expecting to
1454 * find the item. However if this was a bottom-level node,
1455 * then it may be subject to the slot pointer being visible
1456 * to callers dereferencing it. If item corresponding to
1457 * slot[0] is subsequently deleted, these callers would expect
1458 * their slot to become empty sooner or later.
1460 * For example, lockless pagecache will look up a slot, deref
1461 * the page pointer, and if the page is 0 refcount it means it
1462 * was concurrently deleted from pagecache so try the deref
1463 * again. Fortunately there is already a requirement for logic
1464 * to retry the entire slot lookup -- the indirect pointer
1465 * problem (replacing direct root node with an indirect pointer
1466 * also results in a stale slot). So tag the slot as indirect
1467 * to force callers to retry.
1469 if (!radix_tree_is_indirect_ptr(slot))
1470 to_free->slots[0] = RADIX_TREE_RETRY;
1472 radix_tree_node_free(to_free);
1477 * __radix_tree_delete_node - try to free node after clearing a slot
1478 * @root: radix tree root
1479 * @node: node containing @index
1481 * After clearing the slot at @index in @node from radix tree
1482 * rooted at @root, call this function to attempt freeing the
1483 * node and shrinking the tree.
1485 * Returns %true if @node was freed, %false otherwise.
1487 bool __radix_tree_delete_node(struct radix_tree_root *root,
1488 struct radix_tree_node *node)
1490 bool deleted = false;
1493 struct radix_tree_node *parent;
1496 if (node == indirect_to_ptr(root->rnode)) {
1497 radix_tree_shrink(root);
1498 if (root->height == 0)
1504 parent = node->parent;
1506 unsigned int offset;
1508 offset = node->path >> RADIX_TREE_HEIGHT_SHIFT;
1509 parent->slots[offset] = NULL;
1512 root_tag_clear_all(root);
1517 radix_tree_node_free(node);
1526 static inline void delete_sibling_entries(struct radix_tree_node *node,
1527 void *ptr, unsigned offset)
1529 #ifdef CONFIG_RADIX_TREE_MULTIORDER
1531 for (i = 1; offset + i < RADIX_TREE_MAP_SIZE; i++) {
1532 if (node->slots[offset + i] != ptr)
1534 node->slots[offset + i] = NULL;
1541 * radix_tree_delete_item - delete an item from a radix tree
1542 * @root: radix tree root
1544 * @item: expected item
1546 * Remove @item at @index from the radix tree rooted at @root.
1548 * Returns the address of the deleted item, or NULL if it was not present
1549 * or the entry at the given @index was not @item.
1551 void *radix_tree_delete_item(struct radix_tree_root *root,
1552 unsigned long index, void *item)
1554 struct radix_tree_node *node;
1555 unsigned int offset;
1560 entry = __radix_tree_lookup(root, index, &node, &slot);
1564 if (item && entry != item)
1568 root_tag_clear_all(root);
1573 offset = get_slot_offset(node, slot);
1576 * Clear all tags associated with the item to be deleted.
1577 * This way of doing it would be inefficient, but seldom is any set.
1579 for (tag = 0; tag < RADIX_TREE_MAX_TAGS; tag++) {
1580 if (tag_get(node, tag, offset))
1581 radix_tree_tag_clear(root, index, tag);
1584 delete_sibling_entries(node, ptr_to_indirect(slot), offset);
1585 node->slots[offset] = NULL;
1588 __radix_tree_delete_node(root, node);
1592 EXPORT_SYMBOL(radix_tree_delete_item);
1595 * radix_tree_delete - delete an item from a radix tree
1596 * @root: radix tree root
1599 * Remove the item at @index from the radix tree rooted at @root.
1601 * Returns the address of the deleted item, or NULL if it was not present.
1603 void *radix_tree_delete(struct radix_tree_root *root, unsigned long index)
1605 return radix_tree_delete_item(root, index, NULL);
1607 EXPORT_SYMBOL(radix_tree_delete);
1610 * radix_tree_tagged - test whether any items in the tree are tagged
1611 * @root: radix tree root
1614 int radix_tree_tagged(struct radix_tree_root *root, unsigned int tag)
1616 return root_tag_get(root, tag);
1618 EXPORT_SYMBOL(radix_tree_tagged);
1621 radix_tree_node_ctor(void *arg)
1623 struct radix_tree_node *node = arg;
1625 memset(node, 0, sizeof(*node));
1626 INIT_LIST_HEAD(&node->private_list);
1629 static __init unsigned long __maxindex(unsigned int height)
1631 unsigned int width = height * RADIX_TREE_MAP_SHIFT;
1632 int shift = RADIX_TREE_INDEX_BITS - width;
1636 if (shift >= BITS_PER_LONG)
1638 return ~0UL >> shift;
1641 static __init void radix_tree_init_maxindex(void)
1645 for (i = 0; i < ARRAY_SIZE(height_to_maxindex); i++)
1646 height_to_maxindex[i] = __maxindex(i);
1649 static int radix_tree_callback(struct notifier_block *nfb,
1650 unsigned long action,
1653 int cpu = (long)hcpu;
1654 struct radix_tree_preload *rtp;
1655 struct radix_tree_node *node;
1657 /* Free per-cpu pool of perloaded nodes */
1658 if (action == CPU_DEAD || action == CPU_DEAD_FROZEN) {
1659 rtp = &per_cpu(radix_tree_preloads, cpu);
1662 rtp->nodes = node->private_data;
1663 kmem_cache_free(radix_tree_node_cachep, node);
1670 void __init radix_tree_init(void)
1672 radix_tree_node_cachep = kmem_cache_create("radix_tree_node",
1673 sizeof(struct radix_tree_node), 0,
1674 SLAB_PANIC | SLAB_RECLAIM_ACCOUNT,
1675 radix_tree_node_ctor);
1676 radix_tree_init_maxindex();
1677 hotcpu_notifier(radix_tree_callback, 0);