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(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
);
607 unsigned offset
= get_slot_offset(node
, slot
);
609 BUG_ON(tag_get(node
, 0, offset
));
610 BUG_ON(tag_get(node
, 1, offset
));
611 BUG_ON(tag_get(node
, 2, offset
));
613 BUG_ON(root_tags_get(root
));
618 EXPORT_SYMBOL(__radix_tree_insert
);
621 * __radix_tree_lookup - lookup an item in a radix tree
622 * @root: radix tree root
624 * @nodep: returns node
625 * @slotp: returns slot
627 * Lookup and return the item at position @index in the radix
630 * Until there is more than one item in the tree, no nodes are
631 * allocated and @root->rnode is used as a direct slot instead of
632 * pointing to a node, in which case *@nodep will be NULL.
634 void *__radix_tree_lookup(struct radix_tree_root
*root
, unsigned long index
,
635 struct radix_tree_node
**nodep
, void ***slotp
)
637 struct radix_tree_node
*node
, *parent
;
638 unsigned long maxindex
;
644 slot
= (void **)&root
->rnode
;
645 shift
= radix_tree_load_root(root
, &node
, &maxindex
);
646 if (index
> maxindex
)
649 while (radix_tree_is_indirect_ptr(node
)) {
652 if (node
== RADIX_TREE_RETRY
)
654 parent
= indirect_to_ptr(node
);
655 shift
-= RADIX_TREE_MAP_SHIFT
;
656 offset
= (index
>> shift
) & RADIX_TREE_MAP_MASK
;
657 offset
= radix_tree_descend(parent
, &node
, offset
);
658 slot
= parent
->slots
+ offset
;
669 * radix_tree_lookup_slot - lookup a slot in a radix tree
670 * @root: radix tree root
673 * Returns: the slot corresponding to the position @index in the
674 * radix tree @root. This is useful for update-if-exists operations.
676 * This function can be called under rcu_read_lock iff the slot is not
677 * modified by radix_tree_replace_slot, otherwise it must be called
678 * exclusive from other writers. Any dereference of the slot must be done
679 * using radix_tree_deref_slot.
681 void **radix_tree_lookup_slot(struct radix_tree_root
*root
, unsigned long index
)
685 if (!__radix_tree_lookup(root
, index
, NULL
, &slot
))
689 EXPORT_SYMBOL(radix_tree_lookup_slot
);
692 * radix_tree_lookup - perform lookup operation on a radix tree
693 * @root: radix tree root
696 * Lookup the item at the position @index in the radix tree @root.
698 * This function can be called under rcu_read_lock, however the caller
699 * must manage lifetimes of leaf nodes (eg. RCU may also be used to free
700 * them safely). No RCU barriers are required to access or modify the
701 * returned item, however.
703 void *radix_tree_lookup(struct radix_tree_root
*root
, unsigned long index
)
705 return __radix_tree_lookup(root
, index
, NULL
, NULL
);
707 EXPORT_SYMBOL(radix_tree_lookup
);
710 * radix_tree_tag_set - set a tag on a radix tree node
711 * @root: radix tree root
715 * Set the search tag (which must be < RADIX_TREE_MAX_TAGS)
716 * corresponding to @index in the radix tree. From
717 * the root all the way down to the leaf node.
719 * Returns the address of the tagged item. Setting a tag on a not-present
722 void *radix_tree_tag_set(struct radix_tree_root
*root
,
723 unsigned long index
, unsigned int tag
)
725 struct radix_tree_node
*node
, *parent
;
726 unsigned long maxindex
;
729 shift
= radix_tree_load_root(root
, &node
, &maxindex
);
730 BUG_ON(index
> maxindex
);
732 while (radix_tree_is_indirect_ptr(node
)) {
735 shift
-= RADIX_TREE_MAP_SHIFT
;
736 offset
= (index
>> shift
) & RADIX_TREE_MAP_MASK
;
738 parent
= indirect_to_ptr(node
);
739 offset
= radix_tree_descend(parent
, &node
, offset
);
742 if (!tag_get(parent
, tag
, offset
))
743 tag_set(parent
, tag
, offset
);
746 /* set the root's tag bit */
747 if (!root_tag_get(root
, tag
))
748 root_tag_set(root
, tag
);
752 EXPORT_SYMBOL(radix_tree_tag_set
);
755 * radix_tree_tag_clear - clear a tag on a radix tree node
756 * @root: radix tree root
760 * Clear the search tag (which must be < RADIX_TREE_MAX_TAGS)
761 * corresponding to @index in the radix tree. If
762 * this causes the leaf node to have no tags set then clear the tag in the
763 * next-to-leaf node, etc.
765 * Returns the address of the tagged item on success, else NULL. ie:
766 * has the same return value and semantics as radix_tree_lookup().
768 void *radix_tree_tag_clear(struct radix_tree_root
*root
,
769 unsigned long index
, unsigned int tag
)
771 struct radix_tree_node
*node
= NULL
;
772 struct radix_tree_node
*slot
= NULL
;
773 unsigned int height
, shift
;
774 int uninitialized_var(offset
);
776 height
= root
->height
;
777 if (index
> radix_tree_maxindex(height
))
780 shift
= height
* RADIX_TREE_MAP_SHIFT
;
786 if (!radix_tree_is_indirect_ptr(slot
))
788 slot
= indirect_to_ptr(slot
);
790 shift
-= RADIX_TREE_MAP_SHIFT
;
791 offset
= (index
>> shift
) & RADIX_TREE_MAP_MASK
;
793 slot
= slot
->slots
[offset
];
800 if (!tag_get(node
, tag
, offset
))
802 tag_clear(node
, tag
, offset
);
803 if (any_tag_set(node
, tag
))
806 index
>>= RADIX_TREE_MAP_SHIFT
;
807 offset
= index
& RADIX_TREE_MAP_MASK
;
811 /* clear the root's tag bit */
812 if (root_tag_get(root
, tag
))
813 root_tag_clear(root
, tag
);
818 EXPORT_SYMBOL(radix_tree_tag_clear
);
821 * radix_tree_tag_get - get a tag on a radix tree node
822 * @root: radix tree root
824 * @tag: tag index (< RADIX_TREE_MAX_TAGS)
828 * 0: tag not present or not set
831 * Note that the return value of this function may not be relied on, even if
832 * the RCU lock is held, unless tag modification and node deletion are excluded
835 int radix_tree_tag_get(struct radix_tree_root
*root
,
836 unsigned long index
, unsigned int tag
)
838 unsigned int height
, shift
;
839 struct radix_tree_node
*node
;
841 /* check the root's tag bit */
842 if (!root_tag_get(root
, tag
))
845 node
= rcu_dereference_raw(root
->rnode
);
849 if (!radix_tree_is_indirect_ptr(node
))
851 node
= indirect_to_ptr(node
);
853 height
= node
->path
& RADIX_TREE_HEIGHT_MASK
;
854 if (index
> radix_tree_maxindex(height
))
857 shift
= (height
- 1) * RADIX_TREE_MAP_SHIFT
;
864 node
= indirect_to_ptr(node
);
866 offset
= (index
>> shift
) & RADIX_TREE_MAP_MASK
;
867 if (!tag_get(node
, tag
, offset
))
871 node
= rcu_dereference_raw(node
->slots
[offset
]);
872 if (!radix_tree_is_indirect_ptr(node
))
874 shift
-= RADIX_TREE_MAP_SHIFT
;
878 EXPORT_SYMBOL(radix_tree_tag_get
);
880 static inline void __set_iter_shift(struct radix_tree_iter
*iter
,
883 #ifdef CONFIG_RADIX_TREE_MULTIORDER
889 * radix_tree_next_chunk - find next chunk of slots for iteration
891 * @root: radix tree root
892 * @iter: iterator state
893 * @flags: RADIX_TREE_ITER_* flags and tag index
894 * Returns: pointer to chunk first slot, or NULL if iteration is over
896 void **radix_tree_next_chunk(struct radix_tree_root
*root
,
897 struct radix_tree_iter
*iter
, unsigned flags
)
899 unsigned shift
, tag
= flags
& RADIX_TREE_ITER_TAG_MASK
;
900 struct radix_tree_node
*rnode
, *node
;
901 unsigned long index
, offset
, maxindex
;
903 if ((flags
& RADIX_TREE_ITER_TAGGED
) && !root_tag_get(root
, tag
))
907 * Catch next_index overflow after ~0UL. iter->index never overflows
908 * during iterating; it can be zero only at the beginning.
909 * And we cannot overflow iter->next_index in a single step,
910 * because RADIX_TREE_MAP_SHIFT < BITS_PER_LONG.
912 * This condition also used by radix_tree_next_slot() to stop
913 * contiguous iterating, and forbid swithing to the next chunk.
915 index
= iter
->next_index
;
916 if (!index
&& iter
->index
)
920 shift
= radix_tree_load_root(root
, &rnode
, &maxindex
);
921 if (index
> maxindex
)
924 if (radix_tree_is_indirect_ptr(rnode
)) {
925 rnode
= indirect_to_ptr(rnode
);
927 /* Single-slot tree */
929 iter
->next_index
= maxindex
+ 1;
931 __set_iter_shift(iter
, shift
);
932 return (void **)&root
->rnode
;
936 shift
-= RADIX_TREE_MAP_SHIFT
;
937 offset
= index
>> shift
;
941 struct radix_tree_node
*slot
;
942 unsigned new_off
= radix_tree_descend(node
, &slot
, offset
);
944 if (new_off
< offset
) {
946 index
&= ~((RADIX_TREE_MAP_SIZE
<< shift
) - 1);
947 index
|= offset
<< shift
;
950 if ((flags
& RADIX_TREE_ITER_TAGGED
) ?
951 !tag_get(node
, tag
, offset
) : !slot
) {
953 if (flags
& RADIX_TREE_ITER_CONTIG
)
956 if (flags
& RADIX_TREE_ITER_TAGGED
)
957 offset
= radix_tree_find_next_bit(
962 while (++offset
< RADIX_TREE_MAP_SIZE
) {
963 void *slot
= node
->slots
[offset
];
964 if (is_sibling_entry(node
, slot
))
969 index
&= ~((RADIX_TREE_MAP_SIZE
<< shift
) - 1);
970 index
+= offset
<< shift
;
971 /* Overflow after ~0UL */
974 if (offset
== RADIX_TREE_MAP_SIZE
)
976 slot
= rcu_dereference_raw(node
->slots
[offset
]);
979 if ((slot
== NULL
) || (slot
== RADIX_TREE_RETRY
))
981 if (!radix_tree_is_indirect_ptr(slot
))
984 node
= indirect_to_ptr(slot
);
985 shift
-= RADIX_TREE_MAP_SHIFT
;
986 offset
= (index
>> shift
) & RADIX_TREE_MAP_MASK
;
989 /* Update the iterator state */
990 iter
->index
= index
& ~((1 << shift
) - 1);
991 iter
->next_index
= (index
| ((RADIX_TREE_MAP_SIZE
<< shift
) - 1)) + 1;
992 __set_iter_shift(iter
, shift
);
994 /* Construct iter->tags bit-mask from node->tags[tag] array */
995 if (flags
& RADIX_TREE_ITER_TAGGED
) {
996 unsigned tag_long
, tag_bit
;
998 tag_long
= offset
/ BITS_PER_LONG
;
999 tag_bit
= offset
% BITS_PER_LONG
;
1000 iter
->tags
= node
->tags
[tag
][tag_long
] >> tag_bit
;
1001 /* This never happens if RADIX_TREE_TAG_LONGS == 1 */
1002 if (tag_long
< RADIX_TREE_TAG_LONGS
- 1) {
1003 /* Pick tags from next element */
1005 iter
->tags
|= node
->tags
[tag
][tag_long
+ 1] <<
1006 (BITS_PER_LONG
- tag_bit
);
1007 /* Clip chunk size, here only BITS_PER_LONG tags */
1008 iter
->next_index
= index
+ BITS_PER_LONG
;
1012 return node
->slots
+ offset
;
1014 EXPORT_SYMBOL(radix_tree_next_chunk
);
1017 * radix_tree_range_tag_if_tagged - for each item in given range set given
1018 * tag if item has another tag set
1019 * @root: radix tree root
1020 * @first_indexp: pointer to a starting index of a range to scan
1021 * @last_index: last index of a range to scan
1022 * @nr_to_tag: maximum number items to tag
1023 * @iftag: tag index to test
1024 * @settag: tag index to set if tested tag is set
1026 * This function scans range of radix tree from first_index to last_index
1027 * (inclusive). For each item in the range if iftag is set, the function sets
1028 * also settag. The function stops either after tagging nr_to_tag items or
1029 * after reaching last_index.
1031 * The tags must be set from the leaf level only and propagated back up the
1032 * path to the root. We must do this so that we resolve the full path before
1033 * setting any tags on intermediate nodes. If we set tags as we descend, then
1034 * we can get to the leaf node and find that the index that has the iftag
1035 * set is outside the range we are scanning. This reults in dangling tags and
1036 * can lead to problems with later tag operations (e.g. livelocks on lookups).
1038 * The function returns number of leaves where the tag was set and sets
1039 * *first_indexp to the first unscanned index.
1040 * WARNING! *first_indexp can wrap if last_index is ULONG_MAX. Caller must
1041 * be prepared to handle that.
1043 unsigned long radix_tree_range_tag_if_tagged(struct radix_tree_root
*root
,
1044 unsigned long *first_indexp
, unsigned long last_index
,
1045 unsigned long nr_to_tag
,
1046 unsigned int iftag
, unsigned int settag
)
1048 unsigned int height
= root
->height
;
1049 struct radix_tree_node
*node
= NULL
;
1050 struct radix_tree_node
*slot
;
1052 unsigned long tagged
= 0;
1053 unsigned long index
= *first_indexp
;
1055 last_index
= min(last_index
, radix_tree_maxindex(height
));
1056 if (index
> last_index
)
1060 if (!root_tag_get(root
, iftag
)) {
1061 *first_indexp
= last_index
+ 1;
1065 *first_indexp
= last_index
+ 1;
1066 root_tag_set(root
, settag
);
1070 shift
= (height
- 1) * RADIX_TREE_MAP_SHIFT
;
1071 slot
= indirect_to_ptr(root
->rnode
);
1074 unsigned long upindex
;
1077 offset
= (index
>> shift
) & RADIX_TREE_MAP_MASK
;
1078 if (!slot
->slots
[offset
])
1080 if (!tag_get(slot
, iftag
, offset
))
1084 slot
= slot
->slots
[offset
];
1085 if (radix_tree_is_indirect_ptr(slot
)) {
1086 slot
= indirect_to_ptr(slot
);
1087 shift
-= RADIX_TREE_MAP_SHIFT
;
1091 node
= node
->parent
;
1096 tagged
+= 1 << shift
;
1097 tag_set(slot
, settag
, offset
);
1099 /* walk back up the path tagging interior nodes */
1102 upindex
>>= RADIX_TREE_MAP_SHIFT
;
1103 offset
= upindex
& RADIX_TREE_MAP_MASK
;
1105 /* stop if we find a node with the tag already set */
1106 if (tag_get(node
, settag
, offset
))
1108 tag_set(node
, settag
, offset
);
1109 node
= node
->parent
;
1113 * Small optimization: now clear that node pointer.
1114 * Since all of this slot's ancestors now have the tag set
1115 * from setting it above, we have no further need to walk
1116 * back up the tree setting tags, until we update slot to
1117 * point to another radix_tree_node.
1122 /* Go to next item at level determined by 'shift' */
1123 index
= ((index
>> shift
) + 1) << shift
;
1124 /* Overflow can happen when last_index is ~0UL... */
1125 if (index
> last_index
|| !index
)
1127 if (tagged
>= nr_to_tag
)
1129 while (((index
>> shift
) & RADIX_TREE_MAP_MASK
) == 0) {
1131 * We've fully scanned this node. Go up. Because
1132 * last_index is guaranteed to be in the tree, what
1133 * we do below cannot wander astray.
1135 slot
= slot
->parent
;
1136 shift
+= RADIX_TREE_MAP_SHIFT
;
1140 * We need not to tag the root tag if there is no tag which is set with
1141 * settag within the range from *first_indexp to last_index.
1144 root_tag_set(root
, settag
);
1145 *first_indexp
= index
;
1149 EXPORT_SYMBOL(radix_tree_range_tag_if_tagged
);
1152 * radix_tree_gang_lookup - perform multiple lookup on a radix tree
1153 * @root: radix tree root
1154 * @results: where the results of the lookup are placed
1155 * @first_index: start the lookup from this key
1156 * @max_items: place up to this many items at *results
1158 * Performs an index-ascending scan of the tree for present items. Places
1159 * them at *@results and returns the number of items which were placed at
1162 * The implementation is naive.
1164 * Like radix_tree_lookup, radix_tree_gang_lookup may be called under
1165 * rcu_read_lock. In this case, rather than the returned results being
1166 * an atomic snapshot of the tree at a single point in time, the semantics
1167 * of an RCU protected gang lookup are as though multiple radix_tree_lookups
1168 * have been issued in individual locks, and results stored in 'results'.
1171 radix_tree_gang_lookup(struct radix_tree_root
*root
, void **results
,
1172 unsigned long first_index
, unsigned int max_items
)
1174 struct radix_tree_iter iter
;
1176 unsigned int ret
= 0;
1178 if (unlikely(!max_items
))
1181 radix_tree_for_each_slot(slot
, root
, &iter
, first_index
) {
1182 results
[ret
] = rcu_dereference_raw(*slot
);
1185 if (radix_tree_is_indirect_ptr(results
[ret
])) {
1186 slot
= radix_tree_iter_retry(&iter
);
1189 if (++ret
== max_items
)
1195 EXPORT_SYMBOL(radix_tree_gang_lookup
);
1198 * radix_tree_gang_lookup_slot - perform multiple slot lookup on radix tree
1199 * @root: radix tree root
1200 * @results: where the results of the lookup are placed
1201 * @indices: where their indices should be placed (but usually NULL)
1202 * @first_index: start the lookup from this key
1203 * @max_items: place up to this many items at *results
1205 * Performs an index-ascending scan of the tree for present items. Places
1206 * their slots at *@results and returns the number of items which were
1207 * placed at *@results.
1209 * The implementation is naive.
1211 * Like radix_tree_gang_lookup as far as RCU and locking goes. Slots must
1212 * be dereferenced with radix_tree_deref_slot, and if using only RCU
1213 * protection, radix_tree_deref_slot may fail requiring a retry.
1216 radix_tree_gang_lookup_slot(struct radix_tree_root
*root
,
1217 void ***results
, unsigned long *indices
,
1218 unsigned long first_index
, unsigned int max_items
)
1220 struct radix_tree_iter iter
;
1222 unsigned int ret
= 0;
1224 if (unlikely(!max_items
))
1227 radix_tree_for_each_slot(slot
, root
, &iter
, first_index
) {
1228 results
[ret
] = slot
;
1230 indices
[ret
] = iter
.index
;
1231 if (++ret
== max_items
)
1237 EXPORT_SYMBOL(radix_tree_gang_lookup_slot
);
1240 * radix_tree_gang_lookup_tag - perform multiple lookup on a radix tree
1242 * @root: radix tree root
1243 * @results: where the results of the lookup are placed
1244 * @first_index: start the lookup from this key
1245 * @max_items: place up to this many items at *results
1246 * @tag: the tag index (< RADIX_TREE_MAX_TAGS)
1248 * Performs an index-ascending scan of the tree for present items which
1249 * have the tag indexed by @tag set. Places the items at *@results and
1250 * returns the number of items which were placed at *@results.
1253 radix_tree_gang_lookup_tag(struct radix_tree_root
*root
, void **results
,
1254 unsigned long first_index
, unsigned int max_items
,
1257 struct radix_tree_iter iter
;
1259 unsigned int ret
= 0;
1261 if (unlikely(!max_items
))
1264 radix_tree_for_each_tagged(slot
, root
, &iter
, first_index
, tag
) {
1265 results
[ret
] = rcu_dereference_raw(*slot
);
1268 if (radix_tree_is_indirect_ptr(results
[ret
])) {
1269 slot
= radix_tree_iter_retry(&iter
);
1272 if (++ret
== max_items
)
1278 EXPORT_SYMBOL(radix_tree_gang_lookup_tag
);
1281 * radix_tree_gang_lookup_tag_slot - perform multiple slot lookup on a
1282 * radix tree based on a tag
1283 * @root: radix tree root
1284 * @results: where the results of the lookup are placed
1285 * @first_index: start the lookup from this key
1286 * @max_items: place up to this many items at *results
1287 * @tag: the tag index (< RADIX_TREE_MAX_TAGS)
1289 * Performs an index-ascending scan of the tree for present items which
1290 * have the tag indexed by @tag set. Places the slots at *@results and
1291 * returns the number of slots which were placed at *@results.
1294 radix_tree_gang_lookup_tag_slot(struct radix_tree_root
*root
, void ***results
,
1295 unsigned long first_index
, unsigned int max_items
,
1298 struct radix_tree_iter iter
;
1300 unsigned int ret
= 0;
1302 if (unlikely(!max_items
))
1305 radix_tree_for_each_tagged(slot
, root
, &iter
, first_index
, tag
) {
1306 results
[ret
] = slot
;
1307 if (++ret
== max_items
)
1313 EXPORT_SYMBOL(radix_tree_gang_lookup_tag_slot
);
1315 #if defined(CONFIG_SHMEM) && defined(CONFIG_SWAP)
1316 #include <linux/sched.h> /* for cond_resched() */
1319 * This linear search is at present only useful to shmem_unuse_inode().
1321 static unsigned long __locate(struct radix_tree_node
*slot
, void *item
,
1322 unsigned long index
, unsigned long *found_index
)
1324 unsigned int shift
, height
;
1327 height
= slot
->path
& RADIX_TREE_HEIGHT_MASK
;
1328 shift
= (height
-1) * RADIX_TREE_MAP_SHIFT
;
1330 for ( ; height
> 1; height
--) {
1331 i
= (index
>> shift
) & RADIX_TREE_MAP_MASK
;
1333 if (slot
->slots
[i
] != NULL
)
1335 index
&= ~((1UL << shift
) - 1);
1336 index
+= 1UL << shift
;
1338 goto out
; /* 32-bit wraparound */
1340 if (i
== RADIX_TREE_MAP_SIZE
)
1344 slot
= rcu_dereference_raw(slot
->slots
[i
]);
1347 if (!radix_tree_is_indirect_ptr(slot
)) {
1349 *found_index
= index
+ i
;
1356 slot
= indirect_to_ptr(slot
);
1357 shift
-= RADIX_TREE_MAP_SHIFT
;
1360 /* Bottom level: check items */
1361 for (i
= 0; i
< RADIX_TREE_MAP_SIZE
; i
++) {
1362 if (slot
->slots
[i
] == item
) {
1363 *found_index
= index
+ i
;
1368 index
+= RADIX_TREE_MAP_SIZE
;
1374 * radix_tree_locate_item - search through radix tree for item
1375 * @root: radix tree root
1376 * @item: item to be found
1378 * Returns index where item was found, or -1 if not found.
1379 * Caller must hold no lock (since this time-consuming function needs
1380 * to be preemptible), and must check afterwards if item is still there.
1382 unsigned long radix_tree_locate_item(struct radix_tree_root
*root
, void *item
)
1384 struct radix_tree_node
*node
;
1385 unsigned long max_index
;
1386 unsigned long cur_index
= 0;
1387 unsigned long found_index
= -1;
1391 node
= rcu_dereference_raw(root
->rnode
);
1392 if (!radix_tree_is_indirect_ptr(node
)) {
1399 node
= indirect_to_ptr(node
);
1400 max_index
= radix_tree_maxindex(node
->path
&
1401 RADIX_TREE_HEIGHT_MASK
);
1402 if (cur_index
> max_index
) {
1407 cur_index
= __locate(node
, item
, cur_index
, &found_index
);
1410 } while (cur_index
!= 0 && cur_index
<= max_index
);
1415 unsigned long radix_tree_locate_item(struct radix_tree_root
*root
, void *item
)
1419 #endif /* CONFIG_SHMEM && CONFIG_SWAP */
1422 * radix_tree_shrink - shrink height of a radix tree to minimal
1423 * @root radix tree root
1425 static inline void radix_tree_shrink(struct radix_tree_root
*root
)
1427 /* try to shrink tree height */
1428 while (root
->height
> 0) {
1429 struct radix_tree_node
*to_free
= root
->rnode
;
1430 struct radix_tree_node
*slot
;
1432 BUG_ON(!radix_tree_is_indirect_ptr(to_free
));
1433 to_free
= indirect_to_ptr(to_free
);
1436 * The candidate node has more than one child, or its child
1437 * is not at the leftmost slot, or it is a multiorder entry,
1440 if (to_free
->count
!= 1)
1442 slot
= to_free
->slots
[0];
1445 if (!radix_tree_is_indirect_ptr(slot
) && (root
->height
> 1))
1448 if (radix_tree_is_indirect_ptr(slot
)) {
1449 slot
= indirect_to_ptr(slot
);
1450 slot
->parent
= NULL
;
1451 slot
= ptr_to_indirect(slot
);
1455 * We don't need rcu_assign_pointer(), since we are simply
1456 * moving the node from one part of the tree to another: if it
1457 * was safe to dereference the old pointer to it
1458 * (to_free->slots[0]), it will be safe to dereference the new
1459 * one (root->rnode) as far as dependent read barriers go.
1465 * We have a dilemma here. The node's slot[0] must not be
1466 * NULLed in case there are concurrent lookups expecting to
1467 * find the item. However if this was a bottom-level node,
1468 * then it may be subject to the slot pointer being visible
1469 * to callers dereferencing it. If item corresponding to
1470 * slot[0] is subsequently deleted, these callers would expect
1471 * their slot to become empty sooner or later.
1473 * For example, lockless pagecache will look up a slot, deref
1474 * the page pointer, and if the page is 0 refcount it means it
1475 * was concurrently deleted from pagecache so try the deref
1476 * again. Fortunately there is already a requirement for logic
1477 * to retry the entire slot lookup -- the indirect pointer
1478 * problem (replacing direct root node with an indirect pointer
1479 * also results in a stale slot). So tag the slot as indirect
1480 * to force callers to retry.
1482 if (!radix_tree_is_indirect_ptr(slot
))
1483 to_free
->slots
[0] = RADIX_TREE_RETRY
;
1485 radix_tree_node_free(to_free
);
1490 * __radix_tree_delete_node - try to free node after clearing a slot
1491 * @root: radix tree root
1492 * @node: node containing @index
1494 * After clearing the slot at @index in @node from radix tree
1495 * rooted at @root, call this function to attempt freeing the
1496 * node and shrinking the tree.
1498 * Returns %true if @node was freed, %false otherwise.
1500 bool __radix_tree_delete_node(struct radix_tree_root
*root
,
1501 struct radix_tree_node
*node
)
1503 bool deleted
= false;
1506 struct radix_tree_node
*parent
;
1509 if (node
== indirect_to_ptr(root
->rnode
)) {
1510 radix_tree_shrink(root
);
1511 if (root
->height
== 0)
1517 parent
= node
->parent
;
1519 unsigned int offset
;
1521 offset
= node
->path
>> RADIX_TREE_HEIGHT_SHIFT
;
1522 parent
->slots
[offset
] = NULL
;
1525 root_tag_clear_all(root
);
1530 radix_tree_node_free(node
);
1539 static inline void delete_sibling_entries(struct radix_tree_node
*node
,
1540 void *ptr
, unsigned offset
)
1542 #ifdef CONFIG_RADIX_TREE_MULTIORDER
1544 for (i
= 1; offset
+ i
< RADIX_TREE_MAP_SIZE
; i
++) {
1545 if (node
->slots
[offset
+ i
] != ptr
)
1547 node
->slots
[offset
+ i
] = NULL
;
1554 * radix_tree_delete_item - delete an item from a radix tree
1555 * @root: radix tree root
1557 * @item: expected item
1559 * Remove @item at @index from the radix tree rooted at @root.
1561 * Returns the address of the deleted item, or NULL if it was not present
1562 * or the entry at the given @index was not @item.
1564 void *radix_tree_delete_item(struct radix_tree_root
*root
,
1565 unsigned long index
, void *item
)
1567 struct radix_tree_node
*node
;
1568 unsigned int offset
;
1573 entry
= __radix_tree_lookup(root
, index
, &node
, &slot
);
1577 if (item
&& entry
!= item
)
1581 root_tag_clear_all(root
);
1586 offset
= get_slot_offset(node
, slot
);
1589 * Clear all tags associated with the item to be deleted.
1590 * This way of doing it would be inefficient, but seldom is any set.
1592 for (tag
= 0; tag
< RADIX_TREE_MAX_TAGS
; tag
++) {
1593 if (tag_get(node
, tag
, offset
))
1594 radix_tree_tag_clear(root
, index
, tag
);
1597 delete_sibling_entries(node
, ptr_to_indirect(slot
), offset
);
1598 node
->slots
[offset
] = NULL
;
1601 __radix_tree_delete_node(root
, node
);
1605 EXPORT_SYMBOL(radix_tree_delete_item
);
1608 * radix_tree_delete - delete an item from a radix tree
1609 * @root: radix tree root
1612 * Remove the item at @index from the radix tree rooted at @root.
1614 * Returns the address of the deleted item, or NULL if it was not present.
1616 void *radix_tree_delete(struct radix_tree_root
*root
, unsigned long index
)
1618 return radix_tree_delete_item(root
, index
, NULL
);
1620 EXPORT_SYMBOL(radix_tree_delete
);
1623 * radix_tree_tagged - test whether any items in the tree are tagged
1624 * @root: radix tree root
1627 int radix_tree_tagged(struct radix_tree_root
*root
, unsigned int tag
)
1629 return root_tag_get(root
, tag
);
1631 EXPORT_SYMBOL(radix_tree_tagged
);
1634 radix_tree_node_ctor(void *arg
)
1636 struct radix_tree_node
*node
= arg
;
1638 memset(node
, 0, sizeof(*node
));
1639 INIT_LIST_HEAD(&node
->private_list
);
1642 static __init
unsigned long __maxindex(unsigned int height
)
1644 unsigned int width
= height
* RADIX_TREE_MAP_SHIFT
;
1645 int shift
= RADIX_TREE_INDEX_BITS
- width
;
1649 if (shift
>= BITS_PER_LONG
)
1651 return ~0UL >> shift
;
1654 static __init
void radix_tree_init_maxindex(void)
1658 for (i
= 0; i
< ARRAY_SIZE(height_to_maxindex
); i
++)
1659 height_to_maxindex
[i
] = __maxindex(i
);
1662 static int radix_tree_callback(struct notifier_block
*nfb
,
1663 unsigned long action
,
1666 int cpu
= (long)hcpu
;
1667 struct radix_tree_preload
*rtp
;
1668 struct radix_tree_node
*node
;
1670 /* Free per-cpu pool of perloaded nodes */
1671 if (action
== CPU_DEAD
|| action
== CPU_DEAD_FROZEN
) {
1672 rtp
= &per_cpu(radix_tree_preloads
, cpu
);
1675 rtp
->nodes
= node
->private_data
;
1676 kmem_cache_free(radix_tree_node_cachep
, node
);
1683 void __init
radix_tree_init(void)
1685 radix_tree_node_cachep
= kmem_cache_create("radix_tree_node",
1686 sizeof(struct radix_tree_node
), 0,
1687 SLAB_PANIC
| SLAB_RECLAIM_ACCOUNT
,
1688 radix_tree_node_ctor
);
1689 radix_tree_init_maxindex();
1690 hotcpu_notifier(radix_tree_callback
, 0);