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 static inline void *indirect_to_ptr(void *ptr)
80 return (void *)((unsigned long)ptr & ~RADIX_TREE_INDIRECT_PTR);
83 #define RADIX_TREE_RETRY ptr_to_indirect(NULL)
85 #ifdef CONFIG_RADIX_TREE_MULTIORDER
86 /* Sibling slots point directly to another slot in the same node */
87 static inline bool is_sibling_entry(struct radix_tree_node *parent, void *node)
90 return (parent->slots <= ptr) &&
91 (ptr < parent->slots + RADIX_TREE_MAP_SIZE);
94 static inline bool is_sibling_entry(struct radix_tree_node *parent, void *node)
100 static inline unsigned long get_slot_offset(struct radix_tree_node *parent,
103 return slot - parent->slots;
106 static unsigned radix_tree_descend(struct radix_tree_node *parent,
107 struct radix_tree_node **nodep, unsigned offset)
109 void **entry = rcu_dereference_raw(parent->slots[offset]);
111 #ifdef CONFIG_RADIX_TREE_MULTIORDER
112 if (radix_tree_is_indirect_ptr(entry)) {
113 unsigned long siboff = get_slot_offset(parent, entry);
114 if (siboff < RADIX_TREE_MAP_SIZE) {
116 entry = rcu_dereference_raw(parent->slots[offset]);
121 *nodep = (void *)entry;
125 static inline gfp_t root_gfp_mask(struct radix_tree_root *root)
127 return root->gfp_mask & __GFP_BITS_MASK;
130 static inline void tag_set(struct radix_tree_node *node, unsigned int tag,
133 __set_bit(offset, node->tags[tag]);
136 static inline void tag_clear(struct radix_tree_node *node, unsigned int tag,
139 __clear_bit(offset, node->tags[tag]);
142 static inline int tag_get(struct radix_tree_node *node, unsigned int tag,
145 return test_bit(offset, node->tags[tag]);
148 static inline void root_tag_set(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(struct radix_tree_root *root, unsigned int tag)
155 root->gfp_mask &= (__force gfp_t)~(1 << (tag + __GFP_BITS_SHIFT));
158 static inline void root_tag_clear_all(struct radix_tree_root *root)
160 root->gfp_mask &= __GFP_BITS_MASK;
163 static inline int root_tag_get(struct radix_tree_root *root, unsigned int tag)
165 return (__force unsigned)root->gfp_mask & (1 << (tag + __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(void *slot, int height, int offset)
221 struct radix_tree_node *node;
228 pr_debug("radix entry %p offset %d\n", slot, offset);
232 node = indirect_to_ptr(slot);
233 pr_debug("radix node: %p offset %d tags %lx %lx %lx path %x count %d parent %p\n",
234 slot, offset, node->tags[0][0], node->tags[1][0],
235 node->tags[2][0], node->path, node->count, node->parent);
237 for (i = 0; i < RADIX_TREE_MAP_SIZE; i++)
238 dump_node(node->slots[i], height - 1, i);
242 static void radix_tree_dump(struct radix_tree_root *root)
244 pr_debug("radix root: %p height %d rnode %p tags %x\n",
245 root, root->height, root->rnode,
246 root->gfp_mask >> __GFP_BITS_SHIFT);
247 if (!radix_tree_is_indirect_ptr(root->rnode))
249 dump_node(root->rnode, root->height, 0);
254 * This assumes that the caller has performed appropriate preallocation, and
255 * that the caller has pinned this thread of control to the current CPU.
257 static struct radix_tree_node *
258 radix_tree_node_alloc(struct radix_tree_root *root)
260 struct radix_tree_node *ret = NULL;
261 gfp_t gfp_mask = root_gfp_mask(root);
264 * Preload code isn't irq safe and it doesn't make sence to use
265 * preloading in the interrupt anyway as all the allocations have to
266 * be atomic. So just do normal allocation when in interrupt.
268 if (!gfpflags_allow_blocking(gfp_mask) && !in_interrupt()) {
269 struct radix_tree_preload *rtp;
272 * Even if the caller has preloaded, try to allocate from the
273 * cache first for the new node to get accounted.
275 ret = kmem_cache_alloc(radix_tree_node_cachep,
276 gfp_mask | __GFP_ACCOUNT | __GFP_NOWARN);
281 * Provided the caller has preloaded here, we will always
282 * succeed in getting a node here (and never reach
285 rtp = this_cpu_ptr(&radix_tree_preloads);
288 rtp->nodes = ret->private_data;
289 ret->private_data = NULL;
293 * Update the allocation stack trace as this is more useful
296 kmemleak_update_trace(ret);
299 ret = kmem_cache_alloc(radix_tree_node_cachep,
300 gfp_mask | __GFP_ACCOUNT);
302 BUG_ON(radix_tree_is_indirect_ptr(ret));
306 static void radix_tree_node_rcu_free(struct rcu_head *head)
308 struct radix_tree_node *node =
309 container_of(head, struct radix_tree_node, rcu_head);
313 * must only free zeroed nodes into the slab. radix_tree_shrink
314 * can leave us with a non-NULL entry in the first slot, so clear
315 * that here to make sure.
317 for (i = 0; i < RADIX_TREE_MAX_TAGS; i++)
318 tag_clear(node, i, 0);
320 node->slots[0] = NULL;
323 kmem_cache_free(radix_tree_node_cachep, node);
327 radix_tree_node_free(struct radix_tree_node *node)
329 call_rcu(&node->rcu_head, radix_tree_node_rcu_free);
333 * Load up this CPU's radix_tree_node buffer with sufficient objects to
334 * ensure that the addition of a single element in the tree cannot fail. On
335 * success, return zero, with preemption disabled. On error, return -ENOMEM
336 * with preemption not disabled.
338 * To make use of this facility, the radix tree must be initialised without
339 * __GFP_DIRECT_RECLAIM being passed to INIT_RADIX_TREE().
341 static int __radix_tree_preload(gfp_t gfp_mask)
343 struct radix_tree_preload *rtp;
344 struct radix_tree_node *node;
348 rtp = this_cpu_ptr(&radix_tree_preloads);
349 while (rtp->nr < RADIX_TREE_PRELOAD_SIZE) {
351 node = kmem_cache_alloc(radix_tree_node_cachep, gfp_mask);
355 rtp = this_cpu_ptr(&radix_tree_preloads);
356 if (rtp->nr < RADIX_TREE_PRELOAD_SIZE) {
357 node->private_data = rtp->nodes;
361 kmem_cache_free(radix_tree_node_cachep, node);
370 * Load up this CPU's radix_tree_node buffer with sufficient objects to
371 * ensure that the addition of a single element in the tree cannot fail. On
372 * success, return zero, with preemption disabled. On error, return -ENOMEM
373 * with preemption not disabled.
375 * To make use of this facility, the radix tree must be initialised without
376 * __GFP_DIRECT_RECLAIM being passed to INIT_RADIX_TREE().
378 int radix_tree_preload(gfp_t gfp_mask)
380 /* Warn on non-sensical use... */
381 WARN_ON_ONCE(!gfpflags_allow_blocking(gfp_mask));
382 return __radix_tree_preload(gfp_mask);
384 EXPORT_SYMBOL(radix_tree_preload);
387 * The same as above function, except we don't guarantee preloading happens.
388 * We do it, if we decide it helps. On success, return zero with preemption
389 * disabled. On error, return -ENOMEM with preemption not disabled.
391 int radix_tree_maybe_preload(gfp_t gfp_mask)
393 if (gfpflags_allow_blocking(gfp_mask))
394 return __radix_tree_preload(gfp_mask);
395 /* Preloading doesn't help anything with this gfp mask, skip it */
399 EXPORT_SYMBOL(radix_tree_maybe_preload);
402 * Return the maximum key which can be store into a
403 * radix tree with height HEIGHT.
405 static inline unsigned long radix_tree_maxindex(unsigned int height)
407 return height_to_maxindex[height];
410 static inline unsigned long node_maxindex(struct radix_tree_node *node)
412 return radix_tree_maxindex(node->path & RADIX_TREE_HEIGHT_MASK);
415 static unsigned radix_tree_load_root(struct radix_tree_root *root,
416 struct radix_tree_node **nodep, unsigned long *maxindex)
418 struct radix_tree_node *node = rcu_dereference_raw(root->rnode);
422 if (likely(radix_tree_is_indirect_ptr(node))) {
423 node = indirect_to_ptr(node);
424 *maxindex = node_maxindex(node);
425 return (node->path & RADIX_TREE_HEIGHT_MASK) *
426 RADIX_TREE_MAP_SHIFT;
434 * Extend a radix tree so it can store key @index.
436 static int radix_tree_extend(struct radix_tree_root *root,
439 struct radix_tree_node *node;
440 struct radix_tree_node *slot;
444 /* Figure out what the height should be. */
445 height = root->height + 1;
446 while (index > radix_tree_maxindex(height))
449 if (root->rnode == NULL) {
450 root->height = height;
455 unsigned int newheight;
456 if (!(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 /* Increase the height. */
466 newheight = root->height+1;
467 BUG_ON(newheight & ~RADIX_TREE_HEIGHT_MASK);
468 node->path = newheight;
472 if (radix_tree_is_indirect_ptr(slot)) {
473 slot = indirect_to_ptr(slot);
475 slot = ptr_to_indirect(slot);
477 node->slots[0] = slot;
478 node = ptr_to_indirect(node);
479 rcu_assign_pointer(root->rnode, node);
480 root->height = newheight;
481 } while (height > root->height);
483 return height * RADIX_TREE_MAP_SHIFT;
487 * __radix_tree_create - create a slot in a radix tree
488 * @root: radix tree root
490 * @order: index occupies 2^order aligned slots
491 * @nodep: returns node
492 * @slotp: returns slot
494 * Create, if necessary, and return the node and slot for an item
495 * at position @index in the radix tree @root.
497 * Until there is more than one item in the tree, no nodes are
498 * allocated and @root->rnode is used as a direct slot instead of
499 * pointing to a node, in which case *@nodep will be NULL.
501 * Returns -ENOMEM, or 0 for success.
503 int __radix_tree_create(struct radix_tree_root *root, unsigned long index,
504 unsigned order, struct radix_tree_node **nodep,
507 struct radix_tree_node *node = NULL, *slot;
508 unsigned long maxindex;
509 unsigned int height, shift, offset;
510 unsigned long max = index | ((1UL << order) - 1);
512 shift = radix_tree_load_root(root, &slot, &maxindex);
514 /* Make sure the tree is high enough. */
515 if (max > maxindex) {
516 int error = radix_tree_extend(root, max);
521 if (order == shift) {
522 shift += RADIX_TREE_MAP_SHIFT;
527 height = root->height;
529 offset = 0; /* uninitialised var warning */
530 while (shift > order) {
532 /* Have to add a child node. */
533 if (!(slot = radix_tree_node_alloc(root)))
538 rcu_assign_pointer(node->slots[offset],
539 ptr_to_indirect(slot));
541 slot->path |= offset << RADIX_TREE_HEIGHT_SHIFT;
543 rcu_assign_pointer(root->rnode,
544 ptr_to_indirect(slot));
545 } else if (!radix_tree_is_indirect_ptr(slot))
548 /* Go a level down */
550 shift -= RADIX_TREE_MAP_SHIFT;
551 offset = (index >> shift) & RADIX_TREE_MAP_MASK;
552 node = indirect_to_ptr(slot);
553 slot = node->slots[offset];
556 #ifdef CONFIG_RADIX_TREE_MULTIORDER
557 /* Insert pointers to the canonical entry */
559 int i, n = 1 << (order - shift);
560 offset = offset & ~(n - 1);
561 slot = ptr_to_indirect(&node->slots[offset]);
562 for (i = 0; i < n; i++) {
563 if (node->slots[offset + i])
567 for (i = 1; i < n; i++) {
568 rcu_assign_pointer(node->slots[offset + i], slot);
577 *slotp = node ? node->slots + offset : (void **)&root->rnode;
582 * __radix_tree_insert - insert into a radix tree
583 * @root: radix tree root
585 * @order: key covers the 2^order indices around index
586 * @item: item to insert
588 * Insert an item into the radix tree at position @index.
590 int __radix_tree_insert(struct radix_tree_root *root, unsigned long index,
591 unsigned order, void *item)
593 struct radix_tree_node *node;
597 BUG_ON(radix_tree_is_indirect_ptr(item));
599 error = __radix_tree_create(root, index, order, &node, &slot);
604 rcu_assign_pointer(*slot, item);
608 BUG_ON(tag_get(node, 0, index & RADIX_TREE_MAP_MASK));
609 BUG_ON(tag_get(node, 1, index & RADIX_TREE_MAP_MASK));
611 BUG_ON(root_tag_get(root, 0));
612 BUG_ON(root_tag_get(root, 1));
617 EXPORT_SYMBOL(__radix_tree_insert);
620 * __radix_tree_lookup - lookup an item in a radix tree
621 * @root: radix tree root
623 * @nodep: returns node
624 * @slotp: returns slot
626 * Lookup and return the item at position @index in the radix
629 * Until there is more than one item in the tree, no nodes are
630 * allocated and @root->rnode is used as a direct slot instead of
631 * pointing to a node, in which case *@nodep will be NULL.
633 void *__radix_tree_lookup(struct radix_tree_root *root, unsigned long index,
634 struct radix_tree_node **nodep, void ***slotp)
636 struct radix_tree_node *node, *parent;
637 unsigned long maxindex;
643 slot = (void **)&root->rnode;
644 shift = radix_tree_load_root(root, &node, &maxindex);
645 if (index > maxindex)
648 while (radix_tree_is_indirect_ptr(node)) {
651 if (node == RADIX_TREE_RETRY)
653 parent = indirect_to_ptr(node);
654 shift -= RADIX_TREE_MAP_SHIFT;
655 offset = (index >> shift) & RADIX_TREE_MAP_MASK;
656 offset = radix_tree_descend(parent, &node, offset);
657 slot = parent->slots + offset;
668 * radix_tree_lookup_slot - lookup a slot in a radix tree
669 * @root: radix tree root
672 * Returns: the slot corresponding to the position @index in the
673 * radix tree @root. This is useful for update-if-exists operations.
675 * This function can be called under rcu_read_lock iff the slot is not
676 * modified by radix_tree_replace_slot, otherwise it must be called
677 * exclusive from other writers. Any dereference of the slot must be done
678 * using radix_tree_deref_slot.
680 void **radix_tree_lookup_slot(struct radix_tree_root *root, unsigned long index)
684 if (!__radix_tree_lookup(root, index, NULL, &slot))
688 EXPORT_SYMBOL(radix_tree_lookup_slot);
691 * radix_tree_lookup - perform lookup operation on a radix tree
692 * @root: radix tree root
695 * Lookup the item at the position @index in the radix tree @root.
697 * This function can be called under rcu_read_lock, however the caller
698 * must manage lifetimes of leaf nodes (eg. RCU may also be used to free
699 * them safely). No RCU barriers are required to access or modify the
700 * returned item, however.
702 void *radix_tree_lookup(struct radix_tree_root *root, unsigned long index)
704 return __radix_tree_lookup(root, index, NULL, NULL);
706 EXPORT_SYMBOL(radix_tree_lookup);
709 * radix_tree_tag_set - set a tag on a radix tree node
710 * @root: radix tree root
714 * Set the search tag (which must be < RADIX_TREE_MAX_TAGS)
715 * corresponding to @index in the radix tree. From
716 * the root all the way down to the leaf node.
718 * Returns the address of the tagged item. Setting a tag on a not-present
721 void *radix_tree_tag_set(struct radix_tree_root *root,
722 unsigned long index, unsigned int tag)
724 unsigned int height, shift;
725 struct radix_tree_node *slot;
727 height = root->height;
728 BUG_ON(index > radix_tree_maxindex(height));
730 slot = indirect_to_ptr(root->rnode);
731 shift = (height - 1) * RADIX_TREE_MAP_SHIFT;
736 offset = (index >> shift) & RADIX_TREE_MAP_MASK;
737 if (!tag_get(slot, tag, offset))
738 tag_set(slot, tag, offset);
739 slot = slot->slots[offset];
740 BUG_ON(slot == NULL);
741 if (!radix_tree_is_indirect_ptr(slot))
743 slot = indirect_to_ptr(slot);
744 shift -= RADIX_TREE_MAP_SHIFT;
748 /* set the root's tag bit */
749 if (slot && !root_tag_get(root, tag))
750 root_tag_set(root, tag);
754 EXPORT_SYMBOL(radix_tree_tag_set);
757 * radix_tree_tag_clear - clear a tag on a radix tree node
758 * @root: radix tree root
762 * Clear the search tag (which must be < RADIX_TREE_MAX_TAGS)
763 * corresponding to @index in the radix tree. If
764 * this causes the leaf node to have no tags set then clear the tag in the
765 * next-to-leaf node, etc.
767 * Returns the address of the tagged item on success, else NULL. ie:
768 * has the same return value and semantics as radix_tree_lookup().
770 void *radix_tree_tag_clear(struct radix_tree_root *root,
771 unsigned long index, unsigned int tag)
773 struct radix_tree_node *node = NULL;
774 struct radix_tree_node *slot = NULL;
775 unsigned int height, shift;
776 int uninitialized_var(offset);
778 height = root->height;
779 if (index > radix_tree_maxindex(height))
782 shift = height * RADIX_TREE_MAP_SHIFT;
788 if (!radix_tree_is_indirect_ptr(slot))
790 slot = indirect_to_ptr(slot);
792 shift -= RADIX_TREE_MAP_SHIFT;
793 offset = (index >> shift) & RADIX_TREE_MAP_MASK;
795 slot = slot->slots[offset];
802 if (!tag_get(node, tag, offset))
804 tag_clear(node, tag, offset);
805 if (any_tag_set(node, tag))
808 index >>= RADIX_TREE_MAP_SHIFT;
809 offset = index & RADIX_TREE_MAP_MASK;
813 /* clear the root's tag bit */
814 if (root_tag_get(root, tag))
815 root_tag_clear(root, tag);
820 EXPORT_SYMBOL(radix_tree_tag_clear);
823 * radix_tree_tag_get - get a tag on a radix tree node
824 * @root: radix tree root
826 * @tag: tag index (< RADIX_TREE_MAX_TAGS)
830 * 0: tag not present or not set
833 * Note that the return value of this function may not be relied on, even if
834 * the RCU lock is held, unless tag modification and node deletion are excluded
837 int radix_tree_tag_get(struct radix_tree_root *root,
838 unsigned long index, unsigned int tag)
840 unsigned int height, shift;
841 struct radix_tree_node *node;
843 /* check the root's tag bit */
844 if (!root_tag_get(root, tag))
847 node = rcu_dereference_raw(root->rnode);
851 if (!radix_tree_is_indirect_ptr(node))
853 node = indirect_to_ptr(node);
855 height = node->path & RADIX_TREE_HEIGHT_MASK;
856 if (index > radix_tree_maxindex(height))
859 shift = (height - 1) * RADIX_TREE_MAP_SHIFT;
866 node = indirect_to_ptr(node);
868 offset = (index >> shift) & RADIX_TREE_MAP_MASK;
869 if (!tag_get(node, tag, offset))
873 node = rcu_dereference_raw(node->slots[offset]);
874 if (!radix_tree_is_indirect_ptr(node))
876 shift -= RADIX_TREE_MAP_SHIFT;
880 EXPORT_SYMBOL(radix_tree_tag_get);
883 * radix_tree_next_chunk - find next chunk of slots for iteration
885 * @root: radix tree root
886 * @iter: iterator state
887 * @flags: RADIX_TREE_ITER_* flags and tag index
888 * Returns: pointer to chunk first slot, or NULL if iteration is over
890 void **radix_tree_next_chunk(struct radix_tree_root *root,
891 struct radix_tree_iter *iter, unsigned flags)
893 unsigned shift, tag = flags & RADIX_TREE_ITER_TAG_MASK;
894 struct radix_tree_node *rnode, *node;
895 unsigned long index, offset, height;
897 if ((flags & RADIX_TREE_ITER_TAGGED) && !root_tag_get(root, tag))
901 * Catch next_index overflow after ~0UL. iter->index never overflows
902 * during iterating; it can be zero only at the beginning.
903 * And we cannot overflow iter->next_index in a single step,
904 * because RADIX_TREE_MAP_SHIFT < BITS_PER_LONG.
906 * This condition also used by radix_tree_next_slot() to stop
907 * contiguous iterating, and forbid swithing to the next chunk.
909 index = iter->next_index;
910 if (!index && iter->index)
913 rnode = rcu_dereference_raw(root->rnode);
914 if (radix_tree_is_indirect_ptr(rnode)) {
915 rnode = indirect_to_ptr(rnode);
916 } else if (rnode && !index) {
917 /* Single-slot tree */
919 iter->next_index = 1;
921 return (void **)&root->rnode;
926 height = rnode->path & RADIX_TREE_HEIGHT_MASK;
927 shift = (height - 1) * RADIX_TREE_MAP_SHIFT;
928 offset = index >> shift;
930 /* Index outside of the tree */
931 if (offset >= RADIX_TREE_MAP_SIZE)
936 struct radix_tree_node *slot;
937 if ((flags & RADIX_TREE_ITER_TAGGED) ?
938 !test_bit(offset, node->tags[tag]) :
939 !node->slots[offset]) {
941 if (flags & RADIX_TREE_ITER_CONTIG)
944 if (flags & RADIX_TREE_ITER_TAGGED)
945 offset = radix_tree_find_next_bit(
950 while (++offset < RADIX_TREE_MAP_SIZE) {
951 if (node->slots[offset])
954 index &= ~((RADIX_TREE_MAP_SIZE << shift) - 1);
955 index += offset << shift;
956 /* Overflow after ~0UL */
959 if (offset == RADIX_TREE_MAP_SIZE)
963 /* This is leaf-node */
967 slot = rcu_dereference_raw(node->slots[offset]);
970 if (!radix_tree_is_indirect_ptr(slot))
972 node = indirect_to_ptr(slot);
973 shift -= RADIX_TREE_MAP_SHIFT;
974 offset = (index >> shift) & RADIX_TREE_MAP_MASK;
977 /* Update the iterator state */
979 iter->next_index = (index | RADIX_TREE_MAP_MASK) + 1;
981 /* Construct iter->tags bit-mask from node->tags[tag] array */
982 if (flags & RADIX_TREE_ITER_TAGGED) {
983 unsigned tag_long, tag_bit;
985 tag_long = offset / BITS_PER_LONG;
986 tag_bit = offset % BITS_PER_LONG;
987 iter->tags = node->tags[tag][tag_long] >> tag_bit;
988 /* This never happens if RADIX_TREE_TAG_LONGS == 1 */
989 if (tag_long < RADIX_TREE_TAG_LONGS - 1) {
990 /* Pick tags from next element */
992 iter->tags |= node->tags[tag][tag_long + 1] <<
993 (BITS_PER_LONG - tag_bit);
994 /* Clip chunk size, here only BITS_PER_LONG tags */
995 iter->next_index = index + BITS_PER_LONG;
999 return node->slots + offset;
1001 EXPORT_SYMBOL(radix_tree_next_chunk);
1004 * radix_tree_range_tag_if_tagged - for each item in given range set given
1005 * tag if item has another tag set
1006 * @root: radix tree root
1007 * @first_indexp: pointer to a starting index of a range to scan
1008 * @last_index: last index of a range to scan
1009 * @nr_to_tag: maximum number items to tag
1010 * @iftag: tag index to test
1011 * @settag: tag index to set if tested tag is set
1013 * This function scans range of radix tree from first_index to last_index
1014 * (inclusive). For each item in the range if iftag is set, the function sets
1015 * also settag. The function stops either after tagging nr_to_tag items or
1016 * after reaching last_index.
1018 * The tags must be set from the leaf level only and propagated back up the
1019 * path to the root. We must do this so that we resolve the full path before
1020 * setting any tags on intermediate nodes. If we set tags as we descend, then
1021 * we can get to the leaf node and find that the index that has the iftag
1022 * set is outside the range we are scanning. This reults in dangling tags and
1023 * can lead to problems with later tag operations (e.g. livelocks on lookups).
1025 * The function returns number of leaves where the tag was set and sets
1026 * *first_indexp to the first unscanned index.
1027 * WARNING! *first_indexp can wrap if last_index is ULONG_MAX. Caller must
1028 * be prepared to handle that.
1030 unsigned long radix_tree_range_tag_if_tagged(struct radix_tree_root *root,
1031 unsigned long *first_indexp, unsigned long last_index,
1032 unsigned long nr_to_tag,
1033 unsigned int iftag, unsigned int settag)
1035 unsigned int height = root->height;
1036 struct radix_tree_node *node = NULL;
1037 struct radix_tree_node *slot;
1039 unsigned long tagged = 0;
1040 unsigned long index = *first_indexp;
1042 last_index = min(last_index, radix_tree_maxindex(height));
1043 if (index > last_index)
1047 if (!root_tag_get(root, iftag)) {
1048 *first_indexp = last_index + 1;
1052 *first_indexp = last_index + 1;
1053 root_tag_set(root, settag);
1057 shift = (height - 1) * RADIX_TREE_MAP_SHIFT;
1058 slot = indirect_to_ptr(root->rnode);
1061 unsigned long upindex;
1064 offset = (index >> shift) & RADIX_TREE_MAP_MASK;
1065 if (!slot->slots[offset])
1067 if (!tag_get(slot, iftag, offset))
1071 slot = slot->slots[offset];
1072 if (radix_tree_is_indirect_ptr(slot)) {
1073 slot = indirect_to_ptr(slot);
1074 shift -= RADIX_TREE_MAP_SHIFT;
1078 node = node->parent;
1083 tagged += 1 << shift;
1084 tag_set(slot, settag, offset);
1086 /* walk back up the path tagging interior nodes */
1089 upindex >>= RADIX_TREE_MAP_SHIFT;
1090 offset = upindex & RADIX_TREE_MAP_MASK;
1092 /* stop if we find a node with the tag already set */
1093 if (tag_get(node, settag, offset))
1095 tag_set(node, settag, offset);
1096 node = node->parent;
1100 * Small optimization: now clear that node pointer.
1101 * Since all of this slot's ancestors now have the tag set
1102 * from setting it above, we have no further need to walk
1103 * back up the tree setting tags, until we update slot to
1104 * point to another radix_tree_node.
1109 /* Go to next item at level determined by 'shift' */
1110 index = ((index >> shift) + 1) << shift;
1111 /* Overflow can happen when last_index is ~0UL... */
1112 if (index > last_index || !index)
1114 if (tagged >= nr_to_tag)
1116 while (((index >> shift) & RADIX_TREE_MAP_MASK) == 0) {
1118 * We've fully scanned this node. Go up. Because
1119 * last_index is guaranteed to be in the tree, what
1120 * we do below cannot wander astray.
1122 slot = slot->parent;
1123 shift += RADIX_TREE_MAP_SHIFT;
1127 * We need not to tag the root tag if there is no tag which is set with
1128 * settag within the range from *first_indexp to last_index.
1131 root_tag_set(root, settag);
1132 *first_indexp = index;
1136 EXPORT_SYMBOL(radix_tree_range_tag_if_tagged);
1139 * radix_tree_gang_lookup - perform multiple lookup on a radix tree
1140 * @root: radix tree root
1141 * @results: where the results of the lookup are placed
1142 * @first_index: start the lookup from this key
1143 * @max_items: place up to this many items at *results
1145 * Performs an index-ascending scan of the tree for present items. Places
1146 * them at *@results and returns the number of items which were placed at
1149 * The implementation is naive.
1151 * Like radix_tree_lookup, radix_tree_gang_lookup may be called under
1152 * rcu_read_lock. In this case, rather than the returned results being
1153 * an atomic snapshot of the tree at a single point in time, the semantics
1154 * of an RCU protected gang lookup are as though multiple radix_tree_lookups
1155 * have been issued in individual locks, and results stored in 'results'.
1158 radix_tree_gang_lookup(struct radix_tree_root *root, void **results,
1159 unsigned long first_index, unsigned int max_items)
1161 struct radix_tree_iter iter;
1163 unsigned int ret = 0;
1165 if (unlikely(!max_items))
1168 radix_tree_for_each_slot(slot, root, &iter, first_index) {
1169 results[ret] = rcu_dereference_raw(*slot);
1172 if (radix_tree_is_indirect_ptr(results[ret])) {
1173 slot = radix_tree_iter_retry(&iter);
1176 if (++ret == max_items)
1182 EXPORT_SYMBOL(radix_tree_gang_lookup);
1185 * radix_tree_gang_lookup_slot - perform multiple slot lookup on radix tree
1186 * @root: radix tree root
1187 * @results: where the results of the lookup are placed
1188 * @indices: where their indices should be placed (but usually NULL)
1189 * @first_index: start the lookup from this key
1190 * @max_items: place up to this many items at *results
1192 * Performs an index-ascending scan of the tree for present items. Places
1193 * their slots at *@results and returns the number of items which were
1194 * placed at *@results.
1196 * The implementation is naive.
1198 * Like radix_tree_gang_lookup as far as RCU and locking goes. Slots must
1199 * be dereferenced with radix_tree_deref_slot, and if using only RCU
1200 * protection, radix_tree_deref_slot may fail requiring a retry.
1203 radix_tree_gang_lookup_slot(struct radix_tree_root *root,
1204 void ***results, unsigned long *indices,
1205 unsigned long first_index, unsigned int max_items)
1207 struct radix_tree_iter iter;
1209 unsigned int ret = 0;
1211 if (unlikely(!max_items))
1214 radix_tree_for_each_slot(slot, root, &iter, first_index) {
1215 results[ret] = slot;
1217 indices[ret] = iter.index;
1218 if (++ret == max_items)
1224 EXPORT_SYMBOL(radix_tree_gang_lookup_slot);
1227 * radix_tree_gang_lookup_tag - perform multiple lookup on a radix tree
1229 * @root: radix tree root
1230 * @results: where the results of the lookup are placed
1231 * @first_index: start the lookup from this key
1232 * @max_items: place up to this many items at *results
1233 * @tag: the tag index (< RADIX_TREE_MAX_TAGS)
1235 * Performs an index-ascending scan of the tree for present items which
1236 * have the tag indexed by @tag set. Places the items at *@results and
1237 * returns the number of items which were placed at *@results.
1240 radix_tree_gang_lookup_tag(struct radix_tree_root *root, void **results,
1241 unsigned long first_index, unsigned int max_items,
1244 struct radix_tree_iter iter;
1246 unsigned int ret = 0;
1248 if (unlikely(!max_items))
1251 radix_tree_for_each_tagged(slot, root, &iter, first_index, tag) {
1252 results[ret] = rcu_dereference_raw(*slot);
1255 if (radix_tree_is_indirect_ptr(results[ret])) {
1256 slot = radix_tree_iter_retry(&iter);
1259 if (++ret == max_items)
1265 EXPORT_SYMBOL(radix_tree_gang_lookup_tag);
1268 * radix_tree_gang_lookup_tag_slot - perform multiple slot lookup on a
1269 * radix tree based on a tag
1270 * @root: radix tree root
1271 * @results: where the results of the lookup are placed
1272 * @first_index: start the lookup from this key
1273 * @max_items: place up to this many items at *results
1274 * @tag: the tag index (< RADIX_TREE_MAX_TAGS)
1276 * Performs an index-ascending scan of the tree for present items which
1277 * have the tag indexed by @tag set. Places the slots at *@results and
1278 * returns the number of slots which were placed at *@results.
1281 radix_tree_gang_lookup_tag_slot(struct radix_tree_root *root, void ***results,
1282 unsigned long first_index, unsigned int max_items,
1285 struct radix_tree_iter iter;
1287 unsigned int ret = 0;
1289 if (unlikely(!max_items))
1292 radix_tree_for_each_tagged(slot, root, &iter, first_index, tag) {
1293 results[ret] = slot;
1294 if (++ret == max_items)
1300 EXPORT_SYMBOL(radix_tree_gang_lookup_tag_slot);
1302 #if defined(CONFIG_SHMEM) && defined(CONFIG_SWAP)
1303 #include <linux/sched.h> /* for cond_resched() */
1306 * This linear search is at present only useful to shmem_unuse_inode().
1308 static unsigned long __locate(struct radix_tree_node *slot, void *item,
1309 unsigned long index, unsigned long *found_index)
1311 unsigned int shift, height;
1314 height = slot->path & RADIX_TREE_HEIGHT_MASK;
1315 shift = (height-1) * RADIX_TREE_MAP_SHIFT;
1317 for ( ; height > 1; height--) {
1318 i = (index >> shift) & RADIX_TREE_MAP_MASK;
1320 if (slot->slots[i] != NULL)
1322 index &= ~((1UL << shift) - 1);
1323 index += 1UL << shift;
1325 goto out; /* 32-bit wraparound */
1327 if (i == RADIX_TREE_MAP_SIZE)
1331 slot = rcu_dereference_raw(slot->slots[i]);
1334 if (!radix_tree_is_indirect_ptr(slot)) {
1336 *found_index = index + i;
1343 slot = indirect_to_ptr(slot);
1344 shift -= RADIX_TREE_MAP_SHIFT;
1347 /* Bottom level: check items */
1348 for (i = 0; i < RADIX_TREE_MAP_SIZE; i++) {
1349 if (slot->slots[i] == item) {
1350 *found_index = index + i;
1355 index += RADIX_TREE_MAP_SIZE;
1361 * radix_tree_locate_item - search through radix tree for item
1362 * @root: radix tree root
1363 * @item: item to be found
1365 * Returns index where item was found, or -1 if not found.
1366 * Caller must hold no lock (since this time-consuming function needs
1367 * to be preemptible), and must check afterwards if item is still there.
1369 unsigned long radix_tree_locate_item(struct radix_tree_root *root, void *item)
1371 struct radix_tree_node *node;
1372 unsigned long max_index;
1373 unsigned long cur_index = 0;
1374 unsigned long found_index = -1;
1378 node = rcu_dereference_raw(root->rnode);
1379 if (!radix_tree_is_indirect_ptr(node)) {
1386 node = indirect_to_ptr(node);
1387 max_index = radix_tree_maxindex(node->path &
1388 RADIX_TREE_HEIGHT_MASK);
1389 if (cur_index > max_index) {
1394 cur_index = __locate(node, item, cur_index, &found_index);
1397 } while (cur_index != 0 && cur_index <= max_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);