2 * Copyright (C) 2001 Momchil Velikov
3 * Portions Copyright (C) 2001 Christoph Hellwig
4 * Copyright (C) 2005 SGI, Christoph Lameter
5 * Copyright (C) 2006 Nick Piggin
6 * Copyright (C) 2012 Konstantin Khlebnikov
7 * Copyright (C) 2016 Intel, Matthew Wilcox
8 * Copyright (C) 2016 Intel, Ross Zwisler
10 * This program is free software; you can redistribute it and/or
11 * modify it under the terms of the GNU General Public License as
12 * published by the Free Software Foundation; either version 2, or (at
13 * your option) any later version.
15 * This program is distributed in the hope that it will be useful, but
16 * WITHOUT ANY WARRANTY; without even the implied warranty of
17 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
18 * General Public License for more details.
20 * You should have received a copy of the GNU General Public License
21 * along with this program; if not, write to the Free Software
22 * Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA.
25 #include <linux/errno.h>
26 #include <linux/init.h>
27 #include <linux/kernel.h>
28 #include <linux/export.h>
29 #include <linux/radix-tree.h>
30 #include <linux/percpu.h>
31 #include <linux/slab.h>
32 #include <linux/kmemleak.h>
33 #include <linux/notifier.h>
34 #include <linux/cpu.h>
35 #include <linux/string.h>
36 #include <linux/bitops.h>
37 #include <linux/rcupdate.h>
38 #include <linux/preempt.h> /* in_interrupt() */
42 * The height_to_maxindex array needs to be one deeper than the maximum
43 * path as height 0 holds only 1 entry.
45 static unsigned long height_to_maxindex
[RADIX_TREE_MAX_PATH
+ 1] __read_mostly
;
48 * Radix tree node cache.
50 static struct kmem_cache
*radix_tree_node_cachep
;
53 * The radix tree is variable-height, so an insert operation not only has
54 * to build the branch to its corresponding item, it also has to build the
55 * branch to existing items if the size has to be increased (by
58 * The worst case is a zero height tree with just a single item at index 0,
59 * and then inserting an item at index ULONG_MAX. This requires 2 new branches
60 * of RADIX_TREE_MAX_PATH size to be created, with only the root node shared.
63 #define RADIX_TREE_PRELOAD_SIZE (RADIX_TREE_MAX_PATH * 2 - 1)
66 * Per-cpu pool of preloaded nodes
68 struct radix_tree_preload
{
70 /* nodes->private_data points to next preallocated node */
71 struct radix_tree_node
*nodes
;
73 static DEFINE_PER_CPU(struct radix_tree_preload
, radix_tree_preloads
) = { 0, };
75 static inline void *ptr_to_indirect(void *ptr
)
77 return (void *)((unsigned long)ptr
| RADIX_TREE_INDIRECT_PTR
);
80 #define RADIX_TREE_RETRY ptr_to_indirect(NULL)
82 #ifdef CONFIG_RADIX_TREE_MULTIORDER
83 /* Sibling slots point directly to another slot in the same node */
84 static inline bool is_sibling_entry(struct radix_tree_node
*parent
, void *node
)
87 return (parent
->slots
<= ptr
) &&
88 (ptr
< parent
->slots
+ RADIX_TREE_MAP_SIZE
);
91 static inline bool is_sibling_entry(struct radix_tree_node
*parent
, void *node
)
97 static inline unsigned long get_slot_offset(struct radix_tree_node
*parent
,
100 return slot
- parent
->slots
;
103 static unsigned radix_tree_descend(struct radix_tree_node
*parent
,
104 struct radix_tree_node
**nodep
, unsigned offset
)
106 void **entry
= rcu_dereference_raw(parent
->slots
[offset
]);
108 #ifdef CONFIG_RADIX_TREE_MULTIORDER
109 if (radix_tree_is_indirect_ptr(entry
)) {
110 unsigned long siboff
= get_slot_offset(parent
, entry
);
111 if (siboff
< RADIX_TREE_MAP_SIZE
) {
113 entry
= rcu_dereference_raw(parent
->slots
[offset
]);
118 *nodep
= (void *)entry
;
122 static inline gfp_t
root_gfp_mask(struct radix_tree_root
*root
)
124 return root
->gfp_mask
& __GFP_BITS_MASK
;
127 static inline void tag_set(struct radix_tree_node
*node
, unsigned int tag
,
130 __set_bit(offset
, node
->tags
[tag
]);
133 static inline void tag_clear(struct radix_tree_node
*node
, unsigned int tag
,
136 __clear_bit(offset
, node
->tags
[tag
]);
139 static inline int tag_get(struct radix_tree_node
*node
, unsigned int tag
,
142 return test_bit(offset
, node
->tags
[tag
]);
145 static inline void root_tag_set(struct radix_tree_root
*root
, unsigned int tag
)
147 root
->gfp_mask
|= (__force gfp_t
)(1 << (tag
+ __GFP_BITS_SHIFT
));
150 static inline void root_tag_clear(struct radix_tree_root
*root
, unsigned tag
)
152 root
->gfp_mask
&= (__force gfp_t
)~(1 << (tag
+ __GFP_BITS_SHIFT
));
155 static inline void root_tag_clear_all(struct radix_tree_root
*root
)
157 root
->gfp_mask
&= __GFP_BITS_MASK
;
160 static inline int root_tag_get(struct radix_tree_root
*root
, unsigned int tag
)
162 return (__force
int)root
->gfp_mask
& (1 << (tag
+ __GFP_BITS_SHIFT
));
165 static inline unsigned root_tags_get(struct radix_tree_root
*root
)
167 return (__force
unsigned)root
->gfp_mask
>> __GFP_BITS_SHIFT
;
171 * Returns 1 if any slot in the node has this tag set.
172 * Otherwise returns 0.
174 static inline int any_tag_set(struct radix_tree_node
*node
, unsigned int tag
)
177 for (idx
= 0; idx
< RADIX_TREE_TAG_LONGS
; idx
++) {
178 if (node
->tags
[tag
][idx
])
185 * radix_tree_find_next_bit - find the next set bit in a memory region
187 * @addr: The address to base the search on
188 * @size: The bitmap size in bits
189 * @offset: The bitnumber to start searching at
191 * Unrollable variant of find_next_bit() for constant size arrays.
192 * Tail bits starting from size to roundup(size, BITS_PER_LONG) must be zero.
193 * Returns next bit offset, or size if nothing found.
195 static __always_inline
unsigned long
196 radix_tree_find_next_bit(const unsigned long *addr
,
197 unsigned long size
, unsigned long offset
)
199 if (!__builtin_constant_p(size
))
200 return find_next_bit(addr
, size
, offset
);
205 addr
+= offset
/ BITS_PER_LONG
;
206 tmp
= *addr
>> (offset
% BITS_PER_LONG
);
208 return __ffs(tmp
) + offset
;
209 offset
= (offset
+ BITS_PER_LONG
) & ~(BITS_PER_LONG
- 1);
210 while (offset
< size
) {
213 return __ffs(tmp
) + offset
;
214 offset
+= BITS_PER_LONG
;
221 static void dump_node(struct radix_tree_node
*node
,
222 unsigned shift
, unsigned long index
)
226 pr_debug("radix node: %p offset %d tags %lx %lx %lx shift %d count %d parent %p\n",
228 node
->tags
[0][0], node
->tags
[1][0], node
->tags
[2][0],
229 node
->shift
, node
->count
, node
->parent
);
231 for (i
= 0; i
< RADIX_TREE_MAP_SIZE
; i
++) {
232 unsigned long first
= index
| (i
<< shift
);
233 unsigned long last
= first
| ((1UL << shift
) - 1);
234 void *entry
= node
->slots
[i
];
237 if (is_sibling_entry(node
, entry
)) {
238 pr_debug("radix sblng %p offset %ld val %p indices %ld-%ld\n",
240 *(void **)indirect_to_ptr(entry
),
242 } else if (!radix_tree_is_indirect_ptr(entry
)) {
243 pr_debug("radix entry %p offset %ld indices %ld-%ld\n",
244 entry
, i
, first
, last
);
246 dump_node(indirect_to_ptr(entry
),
247 shift
- RADIX_TREE_MAP_SHIFT
, first
);
253 static void radix_tree_dump(struct radix_tree_root
*root
)
255 pr_debug("radix root: %p height %d rnode %p tags %x\n",
256 root
, root
->height
, root
->rnode
,
257 root
->gfp_mask
>> __GFP_BITS_SHIFT
);
258 if (!radix_tree_is_indirect_ptr(root
->rnode
))
260 dump_node(indirect_to_ptr(root
->rnode
),
261 (root
->height
- 1) * RADIX_TREE_MAP_SHIFT
, 0);
266 * This assumes that the caller has performed appropriate preallocation, and
267 * that the caller has pinned this thread of control to the current CPU.
269 static struct radix_tree_node
*
270 radix_tree_node_alloc(struct radix_tree_root
*root
)
272 struct radix_tree_node
*ret
= NULL
;
273 gfp_t gfp_mask
= root_gfp_mask(root
);
276 * Preload code isn't irq safe and it doesn't make sense to use
277 * preloading during an interrupt anyway as all the allocations have
278 * to be atomic. So just do normal allocation when in interrupt.
280 if (!gfpflags_allow_blocking(gfp_mask
) && !in_interrupt()) {
281 struct radix_tree_preload
*rtp
;
284 * Even if the caller has preloaded, try to allocate from the
285 * cache first for the new node to get accounted.
287 ret
= kmem_cache_alloc(radix_tree_node_cachep
,
288 gfp_mask
| __GFP_ACCOUNT
| __GFP_NOWARN
);
293 * Provided the caller has preloaded here, we will always
294 * succeed in getting a node here (and never reach
297 rtp
= this_cpu_ptr(&radix_tree_preloads
);
300 rtp
->nodes
= ret
->private_data
;
301 ret
->private_data
= NULL
;
305 * Update the allocation stack trace as this is more useful
308 kmemleak_update_trace(ret
);
311 ret
= kmem_cache_alloc(radix_tree_node_cachep
,
312 gfp_mask
| __GFP_ACCOUNT
);
314 BUG_ON(radix_tree_is_indirect_ptr(ret
));
318 static void radix_tree_node_rcu_free(struct rcu_head
*head
)
320 struct radix_tree_node
*node
=
321 container_of(head
, struct radix_tree_node
, rcu_head
);
325 * must only free zeroed nodes into the slab. radix_tree_shrink
326 * can leave us with a non-NULL entry in the first slot, so clear
327 * that here to make sure.
329 for (i
= 0; i
< RADIX_TREE_MAX_TAGS
; i
++)
330 tag_clear(node
, i
, 0);
332 node
->slots
[0] = NULL
;
335 kmem_cache_free(radix_tree_node_cachep
, node
);
339 radix_tree_node_free(struct radix_tree_node
*node
)
341 call_rcu(&node
->rcu_head
, radix_tree_node_rcu_free
);
345 * Load up this CPU's radix_tree_node buffer with sufficient objects to
346 * ensure that the addition of a single element in the tree cannot fail. On
347 * success, return zero, with preemption disabled. On error, return -ENOMEM
348 * with preemption not disabled.
350 * To make use of this facility, the radix tree must be initialised without
351 * __GFP_DIRECT_RECLAIM being passed to INIT_RADIX_TREE().
353 static int __radix_tree_preload(gfp_t gfp_mask
)
355 struct radix_tree_preload
*rtp
;
356 struct radix_tree_node
*node
;
360 rtp
= this_cpu_ptr(&radix_tree_preloads
);
361 while (rtp
->nr
< RADIX_TREE_PRELOAD_SIZE
) {
363 node
= kmem_cache_alloc(radix_tree_node_cachep
, gfp_mask
);
367 rtp
= this_cpu_ptr(&radix_tree_preloads
);
368 if (rtp
->nr
< RADIX_TREE_PRELOAD_SIZE
) {
369 node
->private_data
= rtp
->nodes
;
373 kmem_cache_free(radix_tree_node_cachep
, node
);
382 * Load up this CPU's radix_tree_node buffer with sufficient objects to
383 * ensure that the addition of a single element in the tree cannot fail. On
384 * success, return zero, with preemption disabled. On error, return -ENOMEM
385 * with preemption not disabled.
387 * To make use of this facility, the radix tree must be initialised without
388 * __GFP_DIRECT_RECLAIM being passed to INIT_RADIX_TREE().
390 int radix_tree_preload(gfp_t gfp_mask
)
392 /* Warn on non-sensical use... */
393 WARN_ON_ONCE(!gfpflags_allow_blocking(gfp_mask
));
394 return __radix_tree_preload(gfp_mask
);
396 EXPORT_SYMBOL(radix_tree_preload
);
399 * The same as above function, except we don't guarantee preloading happens.
400 * We do it, if we decide it helps. On success, return zero with preemption
401 * disabled. On error, return -ENOMEM with preemption not disabled.
403 int radix_tree_maybe_preload(gfp_t gfp_mask
)
405 if (gfpflags_allow_blocking(gfp_mask
))
406 return __radix_tree_preload(gfp_mask
);
407 /* Preloading doesn't help anything with this gfp mask, skip it */
411 EXPORT_SYMBOL(radix_tree_maybe_preload
);
414 * Return the maximum key which can be store into a
415 * radix tree with height HEIGHT.
417 static inline unsigned long radix_tree_maxindex(unsigned int height
)
419 return height_to_maxindex
[height
];
422 static inline unsigned long shift_maxindex(unsigned int shift
)
424 return (RADIX_TREE_MAP_SIZE
<< shift
) - 1;
427 static inline unsigned long node_maxindex(struct radix_tree_node
*node
)
429 return shift_maxindex(node
->shift
);
432 static unsigned radix_tree_load_root(struct radix_tree_root
*root
,
433 struct radix_tree_node
**nodep
, unsigned long *maxindex
)
435 struct radix_tree_node
*node
= rcu_dereference_raw(root
->rnode
);
439 if (likely(radix_tree_is_indirect_ptr(node
))) {
440 node
= indirect_to_ptr(node
);
441 *maxindex
= node_maxindex(node
);
442 return node
->shift
+ RADIX_TREE_MAP_SHIFT
;
450 * Extend a radix tree so it can store key @index.
452 static int radix_tree_extend(struct radix_tree_root
*root
,
455 struct radix_tree_node
*slot
;
459 /* Figure out what the height should be. */
460 height
= root
->height
+ 1;
461 while (index
> radix_tree_maxindex(height
))
464 if (root
->rnode
== NULL
) {
465 root
->height
= height
;
470 unsigned int newheight
;
471 struct radix_tree_node
*node
= radix_tree_node_alloc(root
);
476 /* Propagate the aggregated tag info into the new root */
477 for (tag
= 0; tag
< RADIX_TREE_MAX_TAGS
; tag
++) {
478 if (root_tag_get(root
, tag
))
479 tag_set(node
, tag
, 0);
482 /* Increase the height. */
483 newheight
= root
->height
;
484 BUG_ON(newheight
> BITS_PER_LONG
);
485 node
->shift
= newheight
* RADIX_TREE_MAP_SHIFT
;
490 if (radix_tree_is_indirect_ptr(slot
)) {
491 slot
= indirect_to_ptr(slot
);
493 slot
= ptr_to_indirect(slot
);
495 node
->slots
[0] = slot
;
496 node
= ptr_to_indirect(node
);
497 rcu_assign_pointer(root
->rnode
, node
);
498 root
->height
= ++newheight
;
499 } while (height
> root
->height
);
501 return height
* RADIX_TREE_MAP_SHIFT
;
505 * __radix_tree_create - create a slot in a radix tree
506 * @root: radix tree root
508 * @order: index occupies 2^order aligned slots
509 * @nodep: returns node
510 * @slotp: returns slot
512 * Create, if necessary, and return the node and slot for an item
513 * at position @index in the radix tree @root.
515 * Until there is more than one item in the tree, no nodes are
516 * allocated and @root->rnode is used as a direct slot instead of
517 * pointing to a node, in which case *@nodep will be NULL.
519 * Returns -ENOMEM, or 0 for success.
521 int __radix_tree_create(struct radix_tree_root
*root
, unsigned long index
,
522 unsigned order
, struct radix_tree_node
**nodep
,
525 struct radix_tree_node
*node
= NULL
, *slot
;
526 unsigned long maxindex
;
527 unsigned int shift
, offset
;
528 unsigned long max
= index
| ((1UL << order
) - 1);
530 shift
= radix_tree_load_root(root
, &slot
, &maxindex
);
532 /* Make sure the tree is high enough. */
533 if (max
> maxindex
) {
534 int error
= radix_tree_extend(root
, max
);
539 if (order
== shift
) {
540 shift
+= RADIX_TREE_MAP_SHIFT
;
545 offset
= 0; /* uninitialised var warning */
546 while (shift
> order
) {
547 shift
-= RADIX_TREE_MAP_SHIFT
;
549 /* Have to add a child node. */
550 slot
= radix_tree_node_alloc(root
);
554 slot
->offset
= offset
;
557 rcu_assign_pointer(node
->slots
[offset
],
558 ptr_to_indirect(slot
));
561 rcu_assign_pointer(root
->rnode
,
562 ptr_to_indirect(slot
));
563 } else if (!radix_tree_is_indirect_ptr(slot
))
566 /* Go a level down */
567 node
= indirect_to_ptr(slot
);
568 offset
= (index
>> shift
) & RADIX_TREE_MAP_MASK
;
569 offset
= radix_tree_descend(node
, &slot
, offset
);
572 #ifdef CONFIG_RADIX_TREE_MULTIORDER
573 /* Insert pointers to the canonical entry */
575 int i
, n
= 1 << (order
- shift
);
576 offset
= offset
& ~(n
- 1);
577 slot
= ptr_to_indirect(&node
->slots
[offset
]);
578 for (i
= 0; i
< n
; i
++) {
579 if (node
->slots
[offset
+ i
])
583 for (i
= 1; i
< n
; i
++) {
584 rcu_assign_pointer(node
->slots
[offset
+ i
], slot
);
593 *slotp
= node
? node
->slots
+ offset
: (void **)&root
->rnode
;
598 * __radix_tree_insert - insert into a radix tree
599 * @root: radix tree root
601 * @order: key covers the 2^order indices around index
602 * @item: item to insert
604 * Insert an item into the radix tree at position @index.
606 int __radix_tree_insert(struct radix_tree_root
*root
, unsigned long index
,
607 unsigned order
, void *item
)
609 struct radix_tree_node
*node
;
613 BUG_ON(radix_tree_is_indirect_ptr(item
));
615 error
= __radix_tree_create(root
, index
, order
, &node
, &slot
);
620 rcu_assign_pointer(*slot
, item
);
623 unsigned offset
= get_slot_offset(node
, slot
);
625 BUG_ON(tag_get(node
, 0, offset
));
626 BUG_ON(tag_get(node
, 1, offset
));
627 BUG_ON(tag_get(node
, 2, offset
));
629 BUG_ON(root_tags_get(root
));
634 EXPORT_SYMBOL(__radix_tree_insert
);
637 * __radix_tree_lookup - lookup an item in a radix tree
638 * @root: radix tree root
640 * @nodep: returns node
641 * @slotp: returns slot
643 * Lookup and return the item at position @index in the radix
646 * Until there is more than one item in the tree, no nodes are
647 * allocated and @root->rnode is used as a direct slot instead of
648 * pointing to a node, in which case *@nodep will be NULL.
650 void *__radix_tree_lookup(struct radix_tree_root
*root
, unsigned long index
,
651 struct radix_tree_node
**nodep
, void ***slotp
)
653 struct radix_tree_node
*node
, *parent
;
654 unsigned long maxindex
;
660 slot
= (void **)&root
->rnode
;
661 shift
= radix_tree_load_root(root
, &node
, &maxindex
);
662 if (index
> maxindex
)
665 while (radix_tree_is_indirect_ptr(node
)) {
668 if (node
== RADIX_TREE_RETRY
)
670 parent
= indirect_to_ptr(node
);
671 shift
-= RADIX_TREE_MAP_SHIFT
;
672 offset
= (index
>> shift
) & RADIX_TREE_MAP_MASK
;
673 offset
= radix_tree_descend(parent
, &node
, offset
);
674 slot
= parent
->slots
+ offset
;
685 * radix_tree_lookup_slot - lookup a slot in a radix tree
686 * @root: radix tree root
689 * Returns: the slot corresponding to the position @index in the
690 * radix tree @root. This is useful for update-if-exists operations.
692 * This function can be called under rcu_read_lock iff the slot is not
693 * modified by radix_tree_replace_slot, otherwise it must be called
694 * exclusive from other writers. Any dereference of the slot must be done
695 * using radix_tree_deref_slot.
697 void **radix_tree_lookup_slot(struct radix_tree_root
*root
, unsigned long index
)
701 if (!__radix_tree_lookup(root
, index
, NULL
, &slot
))
705 EXPORT_SYMBOL(radix_tree_lookup_slot
);
708 * radix_tree_lookup - perform lookup operation on a radix tree
709 * @root: radix tree root
712 * Lookup the item at the position @index in the radix tree @root.
714 * This function can be called under rcu_read_lock, however the caller
715 * must manage lifetimes of leaf nodes (eg. RCU may also be used to free
716 * them safely). No RCU barriers are required to access or modify the
717 * returned item, however.
719 void *radix_tree_lookup(struct radix_tree_root
*root
, unsigned long index
)
721 return __radix_tree_lookup(root
, index
, NULL
, NULL
);
723 EXPORT_SYMBOL(radix_tree_lookup
);
726 * radix_tree_tag_set - set a tag on a radix tree node
727 * @root: radix tree root
731 * Set the search tag (which must be < RADIX_TREE_MAX_TAGS)
732 * corresponding to @index in the radix tree. From
733 * the root all the way down to the leaf node.
735 * Returns the address of the tagged item. Setting a tag on a not-present
738 void *radix_tree_tag_set(struct radix_tree_root
*root
,
739 unsigned long index
, unsigned int tag
)
741 struct radix_tree_node
*node
, *parent
;
742 unsigned long maxindex
;
745 shift
= radix_tree_load_root(root
, &node
, &maxindex
);
746 BUG_ON(index
> maxindex
);
748 while (radix_tree_is_indirect_ptr(node
)) {
751 shift
-= RADIX_TREE_MAP_SHIFT
;
752 offset
= (index
>> shift
) & RADIX_TREE_MAP_MASK
;
754 parent
= indirect_to_ptr(node
);
755 offset
= radix_tree_descend(parent
, &node
, offset
);
758 if (!tag_get(parent
, tag
, offset
))
759 tag_set(parent
, tag
, offset
);
762 /* set the root's tag bit */
763 if (!root_tag_get(root
, tag
))
764 root_tag_set(root
, tag
);
768 EXPORT_SYMBOL(radix_tree_tag_set
);
771 * radix_tree_tag_clear - clear a tag on a radix tree node
772 * @root: radix tree root
776 * Clear the search tag (which must be < RADIX_TREE_MAX_TAGS)
777 * corresponding to @index in the radix tree. If this causes
778 * the leaf node to have no tags set then clear the tag in the
779 * next-to-leaf node, etc.
781 * Returns the address of the tagged item on success, else NULL. ie:
782 * has the same return value and semantics as radix_tree_lookup().
784 void *radix_tree_tag_clear(struct radix_tree_root
*root
,
785 unsigned long index
, unsigned int tag
)
787 struct radix_tree_node
*node
, *parent
;
788 unsigned long maxindex
;
790 int uninitialized_var(offset
);
792 shift
= radix_tree_load_root(root
, &node
, &maxindex
);
793 if (index
> maxindex
)
798 while (radix_tree_is_indirect_ptr(node
)) {
799 shift
-= RADIX_TREE_MAP_SHIFT
;
800 offset
= (index
>> shift
) & RADIX_TREE_MAP_MASK
;
802 parent
= indirect_to_ptr(node
);
803 offset
= radix_tree_descend(parent
, &node
, offset
);
812 if (!tag_get(parent
, tag
, offset
))
814 tag_clear(parent
, tag
, offset
);
815 if (any_tag_set(parent
, tag
))
818 index
>>= RADIX_TREE_MAP_SHIFT
;
819 offset
= index
& RADIX_TREE_MAP_MASK
;
820 parent
= parent
->parent
;
823 /* clear the root's tag bit */
824 if (root_tag_get(root
, tag
))
825 root_tag_clear(root
, tag
);
830 EXPORT_SYMBOL(radix_tree_tag_clear
);
833 * radix_tree_tag_get - get a tag on a radix tree node
834 * @root: radix tree root
836 * @tag: tag index (< RADIX_TREE_MAX_TAGS)
840 * 0: tag not present or not set
843 * Note that the return value of this function may not be relied on, even if
844 * the RCU lock is held, unless tag modification and node deletion are excluded
847 int radix_tree_tag_get(struct radix_tree_root
*root
,
848 unsigned long index
, unsigned int tag
)
850 struct radix_tree_node
*node
, *parent
;
851 unsigned long maxindex
;
854 if (!root_tag_get(root
, tag
))
857 shift
= radix_tree_load_root(root
, &node
, &maxindex
);
858 if (index
> maxindex
)
863 while (radix_tree_is_indirect_ptr(node
)) {
866 shift
-= RADIX_TREE_MAP_SHIFT
;
867 offset
= (index
>> shift
) & RADIX_TREE_MAP_MASK
;
869 parent
= indirect_to_ptr(node
);
870 offset
= radix_tree_descend(parent
, &node
, offset
);
874 if (!tag_get(parent
, tag
, offset
))
876 if (node
== RADIX_TREE_RETRY
)
882 EXPORT_SYMBOL(radix_tree_tag_get
);
884 static inline void __set_iter_shift(struct radix_tree_iter
*iter
,
887 #ifdef CONFIG_RADIX_TREE_MULTIORDER
893 * radix_tree_next_chunk - find next chunk of slots for iteration
895 * @root: radix tree root
896 * @iter: iterator state
897 * @flags: RADIX_TREE_ITER_* flags and tag index
898 * Returns: pointer to chunk first slot, or NULL if iteration is over
900 void **radix_tree_next_chunk(struct radix_tree_root
*root
,
901 struct radix_tree_iter
*iter
, unsigned flags
)
903 unsigned shift
, tag
= flags
& RADIX_TREE_ITER_TAG_MASK
;
904 struct radix_tree_node
*rnode
, *node
;
905 unsigned long index
, offset
, maxindex
;
907 if ((flags
& RADIX_TREE_ITER_TAGGED
) && !root_tag_get(root
, tag
))
911 * Catch next_index overflow after ~0UL. iter->index never overflows
912 * during iterating; it can be zero only at the beginning.
913 * And we cannot overflow iter->next_index in a single step,
914 * because RADIX_TREE_MAP_SHIFT < BITS_PER_LONG.
916 * This condition also used by radix_tree_next_slot() to stop
917 * contiguous iterating, and forbid swithing to the next chunk.
919 index
= iter
->next_index
;
920 if (!index
&& iter
->index
)
924 shift
= radix_tree_load_root(root
, &rnode
, &maxindex
);
925 if (index
> maxindex
)
928 if (radix_tree_is_indirect_ptr(rnode
)) {
929 rnode
= indirect_to_ptr(rnode
);
931 /* Single-slot tree */
933 iter
->next_index
= maxindex
+ 1;
935 __set_iter_shift(iter
, shift
);
936 return (void **)&root
->rnode
;
940 shift
-= RADIX_TREE_MAP_SHIFT
;
941 offset
= index
>> shift
;
945 struct radix_tree_node
*slot
;
946 unsigned new_off
= radix_tree_descend(node
, &slot
, offset
);
948 if (new_off
< offset
) {
950 index
&= ~((RADIX_TREE_MAP_SIZE
<< shift
) - 1);
951 index
|= offset
<< shift
;
954 if ((flags
& RADIX_TREE_ITER_TAGGED
) ?
955 !tag_get(node
, tag
, offset
) : !slot
) {
957 if (flags
& RADIX_TREE_ITER_CONTIG
)
960 if (flags
& RADIX_TREE_ITER_TAGGED
)
961 offset
= radix_tree_find_next_bit(
966 while (++offset
< RADIX_TREE_MAP_SIZE
) {
967 void *slot
= node
->slots
[offset
];
968 if (is_sibling_entry(node
, slot
))
973 index
&= ~((RADIX_TREE_MAP_SIZE
<< shift
) - 1);
974 index
+= offset
<< shift
;
975 /* Overflow after ~0UL */
978 if (offset
== RADIX_TREE_MAP_SIZE
)
980 slot
= rcu_dereference_raw(node
->slots
[offset
]);
983 if ((slot
== NULL
) || (slot
== RADIX_TREE_RETRY
))
985 if (!radix_tree_is_indirect_ptr(slot
))
988 node
= indirect_to_ptr(slot
);
989 shift
-= RADIX_TREE_MAP_SHIFT
;
990 offset
= (index
>> shift
) & RADIX_TREE_MAP_MASK
;
993 /* Update the iterator state */
994 iter
->index
= index
& ~((1 << shift
) - 1);
995 iter
->next_index
= (index
| ((RADIX_TREE_MAP_SIZE
<< shift
) - 1)) + 1;
996 __set_iter_shift(iter
, shift
);
998 /* Construct iter->tags bit-mask from node->tags[tag] array */
999 if (flags
& RADIX_TREE_ITER_TAGGED
) {
1000 unsigned tag_long
, tag_bit
;
1002 tag_long
= offset
/ BITS_PER_LONG
;
1003 tag_bit
= offset
% BITS_PER_LONG
;
1004 iter
->tags
= node
->tags
[tag
][tag_long
] >> tag_bit
;
1005 /* This never happens if RADIX_TREE_TAG_LONGS == 1 */
1006 if (tag_long
< RADIX_TREE_TAG_LONGS
- 1) {
1007 /* Pick tags from next element */
1009 iter
->tags
|= node
->tags
[tag
][tag_long
+ 1] <<
1010 (BITS_PER_LONG
- tag_bit
);
1011 /* Clip chunk size, here only BITS_PER_LONG tags */
1012 iter
->next_index
= index
+ BITS_PER_LONG
;
1016 return node
->slots
+ offset
;
1018 EXPORT_SYMBOL(radix_tree_next_chunk
);
1021 * radix_tree_range_tag_if_tagged - for each item in given range set given
1022 * tag if item has another tag set
1023 * @root: radix tree root
1024 * @first_indexp: pointer to a starting index of a range to scan
1025 * @last_index: last index of a range to scan
1026 * @nr_to_tag: maximum number items to tag
1027 * @iftag: tag index to test
1028 * @settag: tag index to set if tested tag is set
1030 * This function scans range of radix tree from first_index to last_index
1031 * (inclusive). For each item in the range if iftag is set, the function sets
1032 * also settag. The function stops either after tagging nr_to_tag items or
1033 * after reaching last_index.
1035 * The tags must be set from the leaf level only and propagated back up the
1036 * path to the root. We must do this so that we resolve the full path before
1037 * setting any tags on intermediate nodes. If we set tags as we descend, then
1038 * we can get to the leaf node and find that the index that has the iftag
1039 * set is outside the range we are scanning. This reults in dangling tags and
1040 * can lead to problems with later tag operations (e.g. livelocks on lookups).
1042 * The function returns the number of leaves where the tag was set and sets
1043 * *first_indexp to the first unscanned index.
1044 * WARNING! *first_indexp can wrap if last_index is ULONG_MAX. Caller must
1045 * be prepared to handle that.
1047 unsigned long radix_tree_range_tag_if_tagged(struct radix_tree_root
*root
,
1048 unsigned long *first_indexp
, unsigned long last_index
,
1049 unsigned long nr_to_tag
,
1050 unsigned int iftag
, unsigned int settag
)
1052 struct radix_tree_node
*slot
, *node
= NULL
;
1053 unsigned long maxindex
;
1054 unsigned int shift
= radix_tree_load_root(root
, &slot
, &maxindex
);
1055 unsigned long tagged
= 0;
1056 unsigned long index
= *first_indexp
;
1058 last_index
= min(last_index
, maxindex
);
1059 if (index
> last_index
)
1063 if (!root_tag_get(root
, iftag
)) {
1064 *first_indexp
= last_index
+ 1;
1067 if (!radix_tree_is_indirect_ptr(slot
)) {
1068 *first_indexp
= last_index
+ 1;
1069 root_tag_set(root
, settag
);
1073 node
= indirect_to_ptr(slot
);
1074 shift
-= RADIX_TREE_MAP_SHIFT
;
1077 unsigned long upindex
;
1080 offset
= (index
>> shift
) & RADIX_TREE_MAP_MASK
;
1081 offset
= radix_tree_descend(node
, &slot
, offset
);
1084 if (!tag_get(node
, iftag
, offset
))
1086 /* Sibling slots never have tags set on them */
1087 if (radix_tree_is_indirect_ptr(slot
)) {
1088 node
= indirect_to_ptr(slot
);
1089 shift
-= RADIX_TREE_MAP_SHIFT
;
1095 tag_set(node
, settag
, offset
);
1097 slot
= node
->parent
;
1098 /* walk back up the path tagging interior nodes */
1099 upindex
= index
>> shift
;
1101 upindex
>>= RADIX_TREE_MAP_SHIFT
;
1102 offset
= upindex
& RADIX_TREE_MAP_MASK
;
1104 /* stop if we find a node with the tag already set */
1105 if (tag_get(slot
, settag
, offset
))
1107 tag_set(slot
, settag
, offset
);
1108 slot
= slot
->parent
;
1112 /* Go to next item at level determined by 'shift' */
1113 index
= ((index
>> shift
) + 1) << shift
;
1114 /* Overflow can happen when last_index is ~0UL... */
1115 if (index
> last_index
|| !index
)
1117 offset
= (index
>> shift
) & RADIX_TREE_MAP_MASK
;
1118 while (offset
== 0) {
1120 * We've fully scanned this node. Go up. Because
1121 * last_index is guaranteed to be in the tree, what
1122 * we do below cannot wander astray.
1124 node
= node
->parent
;
1125 shift
+= RADIX_TREE_MAP_SHIFT
;
1126 offset
= (index
>> shift
) & RADIX_TREE_MAP_MASK
;
1128 if (is_sibling_entry(node
, node
->slots
[offset
]))
1130 if (tagged
>= nr_to_tag
)
1134 * We need not to tag the root tag if there is no tag which is set with
1135 * settag within the range from *first_indexp to last_index.
1138 root_tag_set(root
, settag
);
1139 *first_indexp
= index
;
1143 EXPORT_SYMBOL(radix_tree_range_tag_if_tagged
);
1146 * radix_tree_gang_lookup - perform multiple lookup on a radix tree
1147 * @root: radix tree root
1148 * @results: where the results of the lookup are placed
1149 * @first_index: start the lookup from this key
1150 * @max_items: place up to this many items at *results
1152 * Performs an index-ascending scan of the tree for present items. Places
1153 * them at *@results and returns the number of items which were placed at
1156 * The implementation is naive.
1158 * Like radix_tree_lookup, radix_tree_gang_lookup may be called under
1159 * rcu_read_lock. In this case, rather than the returned results being
1160 * an atomic snapshot of the tree at a single point in time, the
1161 * semantics of an RCU protected gang lookup are as though multiple
1162 * radix_tree_lookups have been issued in individual locks, and results
1163 * stored in 'results'.
1166 radix_tree_gang_lookup(struct radix_tree_root
*root
, void **results
,
1167 unsigned long first_index
, unsigned int max_items
)
1169 struct radix_tree_iter iter
;
1171 unsigned int ret
= 0;
1173 if (unlikely(!max_items
))
1176 radix_tree_for_each_slot(slot
, root
, &iter
, first_index
) {
1177 results
[ret
] = rcu_dereference_raw(*slot
);
1180 if (radix_tree_is_indirect_ptr(results
[ret
])) {
1181 slot
= radix_tree_iter_retry(&iter
);
1184 if (++ret
== max_items
)
1190 EXPORT_SYMBOL(radix_tree_gang_lookup
);
1193 * radix_tree_gang_lookup_slot - perform multiple slot lookup on radix tree
1194 * @root: radix tree root
1195 * @results: where the results of the lookup are placed
1196 * @indices: where their indices should be placed (but usually NULL)
1197 * @first_index: start the lookup from this key
1198 * @max_items: place up to this many items at *results
1200 * Performs an index-ascending scan of the tree for present items. Places
1201 * their slots at *@results and returns the number of items which were
1202 * placed at *@results.
1204 * The implementation is naive.
1206 * Like radix_tree_gang_lookup as far as RCU and locking goes. Slots must
1207 * be dereferenced with radix_tree_deref_slot, and if using only RCU
1208 * protection, radix_tree_deref_slot may fail requiring a retry.
1211 radix_tree_gang_lookup_slot(struct radix_tree_root
*root
,
1212 void ***results
, unsigned long *indices
,
1213 unsigned long first_index
, unsigned int max_items
)
1215 struct radix_tree_iter iter
;
1217 unsigned int ret
= 0;
1219 if (unlikely(!max_items
))
1222 radix_tree_for_each_slot(slot
, root
, &iter
, first_index
) {
1223 results
[ret
] = slot
;
1225 indices
[ret
] = iter
.index
;
1226 if (++ret
== max_items
)
1232 EXPORT_SYMBOL(radix_tree_gang_lookup_slot
);
1235 * radix_tree_gang_lookup_tag - perform multiple lookup on a radix tree
1237 * @root: radix tree root
1238 * @results: where the results of the lookup are placed
1239 * @first_index: start the lookup from this key
1240 * @max_items: place up to this many items at *results
1241 * @tag: the tag index (< RADIX_TREE_MAX_TAGS)
1243 * Performs an index-ascending scan of the tree for present items which
1244 * have the tag indexed by @tag set. Places the items at *@results and
1245 * returns the number of items which were placed at *@results.
1248 radix_tree_gang_lookup_tag(struct radix_tree_root
*root
, void **results
,
1249 unsigned long first_index
, unsigned int max_items
,
1252 struct radix_tree_iter iter
;
1254 unsigned int ret
= 0;
1256 if (unlikely(!max_items
))
1259 radix_tree_for_each_tagged(slot
, root
, &iter
, first_index
, tag
) {
1260 results
[ret
] = rcu_dereference_raw(*slot
);
1263 if (radix_tree_is_indirect_ptr(results
[ret
])) {
1264 slot
= radix_tree_iter_retry(&iter
);
1267 if (++ret
== max_items
)
1273 EXPORT_SYMBOL(radix_tree_gang_lookup_tag
);
1276 * radix_tree_gang_lookup_tag_slot - perform multiple slot lookup on a
1277 * radix tree based on a tag
1278 * @root: radix tree root
1279 * @results: where the results of the lookup are placed
1280 * @first_index: start the lookup from this key
1281 * @max_items: place up to this many items at *results
1282 * @tag: the tag index (< RADIX_TREE_MAX_TAGS)
1284 * Performs an index-ascending scan of the tree for present items which
1285 * have the tag indexed by @tag set. Places the slots at *@results and
1286 * returns the number of slots which were placed at *@results.
1289 radix_tree_gang_lookup_tag_slot(struct radix_tree_root
*root
, void ***results
,
1290 unsigned long first_index
, unsigned int max_items
,
1293 struct radix_tree_iter iter
;
1295 unsigned int ret
= 0;
1297 if (unlikely(!max_items
))
1300 radix_tree_for_each_tagged(slot
, root
, &iter
, first_index
, tag
) {
1301 results
[ret
] = slot
;
1302 if (++ret
== max_items
)
1308 EXPORT_SYMBOL(radix_tree_gang_lookup_tag_slot
);
1310 #if defined(CONFIG_SHMEM) && defined(CONFIG_SWAP)
1311 #include <linux/sched.h> /* for cond_resched() */
1313 struct locate_info
{
1314 unsigned long found_index
;
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
, struct locate_info
*info
)
1327 shift
= slot
->shift
+ RADIX_TREE_MAP_SHIFT
;
1330 shift
-= RADIX_TREE_MAP_SHIFT
;
1332 for (i
= (index
>> shift
) & RADIX_TREE_MAP_MASK
;
1333 i
< RADIX_TREE_MAP_SIZE
;
1334 i
++, index
+= (1UL << shift
)) {
1335 struct radix_tree_node
*node
=
1336 rcu_dereference_raw(slot
->slots
[i
]);
1337 if (node
== RADIX_TREE_RETRY
)
1339 if (!radix_tree_is_indirect_ptr(node
)) {
1341 info
->found_index
= index
;
1347 node
= indirect_to_ptr(node
);
1348 if (is_sibling_entry(slot
, node
))
1353 if (i
== RADIX_TREE_MAP_SIZE
)
1358 if ((index
== 0) && (i
== RADIX_TREE_MAP_SIZE
))
1364 * radix_tree_locate_item - search through radix tree for item
1365 * @root: radix tree root
1366 * @item: item to be found
1368 * Returns index where item was found, or -1 if not found.
1369 * Caller must hold no lock (since this time-consuming function needs
1370 * to be preemptible), and must check afterwards if item is still there.
1372 unsigned long radix_tree_locate_item(struct radix_tree_root
*root
, void *item
)
1374 struct radix_tree_node
*node
;
1375 unsigned long max_index
;
1376 unsigned long cur_index
= 0;
1377 struct locate_info info
= {
1384 node
= rcu_dereference_raw(root
->rnode
);
1385 if (!radix_tree_is_indirect_ptr(node
)) {
1388 info
.found_index
= 0;
1392 node
= indirect_to_ptr(node
);
1394 max_index
= node_maxindex(node
);
1395 if (cur_index
> max_index
) {
1400 cur_index
= __locate(node
, item
, cur_index
, &info
);
1403 } while (!info
.stop
&& cur_index
<= max_index
);
1405 return info
.found_index
;
1408 unsigned long radix_tree_locate_item(struct radix_tree_root
*root
, void *item
)
1412 #endif /* CONFIG_SHMEM && CONFIG_SWAP */
1415 * radix_tree_shrink - shrink height of a radix tree to minimal
1416 * @root radix tree root
1418 static inline void radix_tree_shrink(struct radix_tree_root
*root
)
1420 /* try to shrink tree height */
1421 while (root
->height
> 0) {
1422 struct radix_tree_node
*to_free
= root
->rnode
;
1423 struct radix_tree_node
*slot
;
1425 BUG_ON(!radix_tree_is_indirect_ptr(to_free
));
1426 to_free
= indirect_to_ptr(to_free
);
1429 * The candidate node has more than one child, or its child
1430 * is not at the leftmost slot, or it is a multiorder entry,
1433 if (to_free
->count
!= 1)
1435 slot
= to_free
->slots
[0];
1438 if (!radix_tree_is_indirect_ptr(slot
) && (root
->height
> 1))
1441 if (radix_tree_is_indirect_ptr(slot
)) {
1442 slot
= indirect_to_ptr(slot
);
1443 slot
->parent
= NULL
;
1444 slot
= ptr_to_indirect(slot
);
1448 * We don't need rcu_assign_pointer(), since we are simply
1449 * moving the node from one part of the tree to another: if it
1450 * was safe to dereference the old pointer to it
1451 * (to_free->slots[0]), it will be safe to dereference the new
1452 * one (root->rnode) as far as dependent read barriers go.
1458 * We have a dilemma here. The node's slot[0] must not be
1459 * NULLed in case there are concurrent lookups expecting to
1460 * find the item. However if this was a bottom-level node,
1461 * then it may be subject to the slot pointer being visible
1462 * to callers dereferencing it. If item corresponding to
1463 * slot[0] is subsequently deleted, these callers would expect
1464 * their slot to become empty sooner or later.
1466 * For example, lockless pagecache will look up a slot, deref
1467 * the page pointer, and if the page has 0 refcount it means it
1468 * was concurrently deleted from pagecache so try the deref
1469 * again. Fortunately there is already a requirement for logic
1470 * to retry the entire slot lookup -- the indirect pointer
1471 * problem (replacing direct root node with an indirect pointer
1472 * also results in a stale slot). So tag the slot as indirect
1473 * to force callers to retry.
1475 if (!radix_tree_is_indirect_ptr(slot
))
1476 to_free
->slots
[0] = RADIX_TREE_RETRY
;
1478 radix_tree_node_free(to_free
);
1483 * __radix_tree_delete_node - try to free node after clearing a slot
1484 * @root: radix tree root
1485 * @node: node containing @index
1487 * After clearing the slot at @index in @node from radix tree
1488 * rooted at @root, call this function to attempt freeing the
1489 * node and shrinking the tree.
1491 * Returns %true if @node was freed, %false otherwise.
1493 bool __radix_tree_delete_node(struct radix_tree_root
*root
,
1494 struct radix_tree_node
*node
)
1496 bool deleted
= false;
1499 struct radix_tree_node
*parent
;
1502 if (node
== indirect_to_ptr(root
->rnode
)) {
1503 radix_tree_shrink(root
);
1504 if (root
->height
== 0)
1510 parent
= node
->parent
;
1512 parent
->slots
[node
->offset
] = NULL
;
1515 root_tag_clear_all(root
);
1520 radix_tree_node_free(node
);
1529 static inline void delete_sibling_entries(struct radix_tree_node
*node
,
1530 void *ptr
, unsigned offset
)
1532 #ifdef CONFIG_RADIX_TREE_MULTIORDER
1534 for (i
= 1; offset
+ i
< RADIX_TREE_MAP_SIZE
; i
++) {
1535 if (node
->slots
[offset
+ i
] != ptr
)
1537 node
->slots
[offset
+ i
] = NULL
;
1544 * radix_tree_delete_item - delete an item from a radix tree
1545 * @root: radix tree root
1547 * @item: expected item
1549 * Remove @item at @index from the radix tree rooted at @root.
1551 * Returns the address of the deleted item, or NULL if it was not present
1552 * or the entry at the given @index was not @item.
1554 void *radix_tree_delete_item(struct radix_tree_root
*root
,
1555 unsigned long index
, void *item
)
1557 struct radix_tree_node
*node
;
1558 unsigned int offset
;
1563 entry
= __radix_tree_lookup(root
, index
, &node
, &slot
);
1567 if (item
&& entry
!= item
)
1571 root_tag_clear_all(root
);
1576 offset
= get_slot_offset(node
, slot
);
1579 * Clear all tags associated with the item to be deleted.
1580 * This way of doing it would be inefficient, but seldom is any set.
1582 for (tag
= 0; tag
< RADIX_TREE_MAX_TAGS
; tag
++) {
1583 if (tag_get(node
, tag
, offset
))
1584 radix_tree_tag_clear(root
, index
, tag
);
1587 delete_sibling_entries(node
, ptr_to_indirect(slot
), offset
);
1588 node
->slots
[offset
] = NULL
;
1591 __radix_tree_delete_node(root
, node
);
1595 EXPORT_SYMBOL(radix_tree_delete_item
);
1598 * radix_tree_delete - delete an item from a radix tree
1599 * @root: radix tree root
1602 * Remove the item at @index from the radix tree rooted at @root.
1604 * Returns the address of the deleted item, or NULL if it was not present.
1606 void *radix_tree_delete(struct radix_tree_root
*root
, unsigned long index
)
1608 return radix_tree_delete_item(root
, index
, NULL
);
1610 EXPORT_SYMBOL(radix_tree_delete
);
1613 * radix_tree_tagged - test whether any items in the tree are tagged
1614 * @root: radix tree root
1617 int radix_tree_tagged(struct radix_tree_root
*root
, unsigned int tag
)
1619 return root_tag_get(root
, tag
);
1621 EXPORT_SYMBOL(radix_tree_tagged
);
1624 radix_tree_node_ctor(void *arg
)
1626 struct radix_tree_node
*node
= arg
;
1628 memset(node
, 0, sizeof(*node
));
1629 INIT_LIST_HEAD(&node
->private_list
);
1632 static __init
unsigned long __maxindex(unsigned int height
)
1634 unsigned int width
= height
* RADIX_TREE_MAP_SHIFT
;
1635 int shift
= RADIX_TREE_INDEX_BITS
- width
;
1639 if (shift
>= BITS_PER_LONG
)
1641 return ~0UL >> shift
;
1644 static __init
void radix_tree_init_maxindex(void)
1648 for (i
= 0; i
< ARRAY_SIZE(height_to_maxindex
); i
++)
1649 height_to_maxindex
[i
] = __maxindex(i
);
1652 static int radix_tree_callback(struct notifier_block
*nfb
,
1653 unsigned long action
, void *hcpu
)
1655 int cpu
= (long)hcpu
;
1656 struct radix_tree_preload
*rtp
;
1657 struct radix_tree_node
*node
;
1659 /* Free per-cpu pool of preloaded nodes */
1660 if (action
== CPU_DEAD
|| action
== CPU_DEAD_FROZEN
) {
1661 rtp
= &per_cpu(radix_tree_preloads
, cpu
);
1664 rtp
->nodes
= node
->private_data
;
1665 kmem_cache_free(radix_tree_node_cachep
, node
);
1672 void __init
radix_tree_init(void)
1674 radix_tree_node_cachep
= kmem_cache_create("radix_tree_node",
1675 sizeof(struct radix_tree_node
), 0,
1676 SLAB_PANIC
| SLAB_RECLAIM_ACCOUNT
,
1677 radix_tree_node_ctor
);
1678 radix_tree_init_maxindex();
1679 hotcpu_notifier(radix_tree_callback
, 0);