blob: 6f412ab4c24f812fc8b7290c4a1e06cc0250ea62 [file] [log] [blame]
Linus Torvalds1da177e2005-04-16 15:20:36 -07001/*
2 * Copyright (C) 2001 Momchil Velikov
3 * Portions Copyright (C) 2001 Christoph Hellwig
Christoph Lametercde53532008-07-04 09:59:22 -07004 * Copyright (C) 2005 SGI, Christoph Lameter
Nick Piggin7cf9c2c2006-12-06 20:33:44 -08005 * Copyright (C) 2006 Nick Piggin
Linus Torvalds1da177e2005-04-16 15:20:36 -07006 *
7 * This program is free software; you can redistribute it and/or
8 * modify it under the terms of the GNU General Public License as
9 * published by the Free Software Foundation; either version 2, or (at
10 * your option) any later version.
11 *
12 * This program is distributed in the hope that it will be useful, but
13 * WITHOUT ANY WARRANTY; without even the implied warranty of
14 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
15 * General Public License for more details.
16 *
17 * You should have received a copy of the GNU General Public License
18 * along with this program; if not, write to the Free Software
19 * Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA.
20 */
21
22#include <linux/errno.h>
23#include <linux/init.h>
24#include <linux/kernel.h>
25#include <linux/module.h>
26#include <linux/radix-tree.h>
27#include <linux/percpu.h>
28#include <linux/slab.h>
29#include <linux/notifier.h>
30#include <linux/cpu.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070031#include <linux/string.h>
32#include <linux/bitops.h>
Nick Piggin7cf9c2c2006-12-06 20:33:44 -080033#include <linux/rcupdate.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070034
35
36#ifdef __KERNEL__
Nick Piggincfd9b7d2006-06-23 02:03:22 -070037#define RADIX_TREE_MAP_SHIFT (CONFIG_BASE_SMALL ? 4 : 6)
Linus Torvalds1da177e2005-04-16 15:20:36 -070038#else
39#define RADIX_TREE_MAP_SHIFT 3 /* For more stressful testing */
40#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -070041
42#define RADIX_TREE_MAP_SIZE (1UL << RADIX_TREE_MAP_SHIFT)
43#define RADIX_TREE_MAP_MASK (RADIX_TREE_MAP_SIZE-1)
44
45#define RADIX_TREE_TAG_LONGS \
46 ((RADIX_TREE_MAP_SIZE + BITS_PER_LONG - 1) / BITS_PER_LONG)
47
48struct radix_tree_node {
Nick Piggin7cf9c2c2006-12-06 20:33:44 -080049 unsigned int height; /* Height from the bottom */
Linus Torvalds1da177e2005-04-16 15:20:36 -070050 unsigned int count;
Nick Piggin7cf9c2c2006-12-06 20:33:44 -080051 struct rcu_head rcu_head;
Arnd Bergmanna1115572010-02-25 23:43:52 +010052 void __rcu *slots[RADIX_TREE_MAP_SIZE];
Jonathan Corbetdaff89f2006-03-25 03:08:05 -080053 unsigned long tags[RADIX_TREE_MAX_TAGS][RADIX_TREE_TAG_LONGS];
Linus Torvalds1da177e2005-04-16 15:20:36 -070054};
55
56struct radix_tree_path {
Christoph Lameter201b6262005-09-06 15:16:46 -070057 struct radix_tree_node *node;
Linus Torvalds1da177e2005-04-16 15:20:36 -070058 int offset;
59};
60
61#define RADIX_TREE_INDEX_BITS (8 /* CHAR_BIT */ * sizeof(unsigned long))
Jeff Moyer26fb1582007-10-16 01:24:49 -070062#define RADIX_TREE_MAX_PATH (DIV_ROUND_UP(RADIX_TREE_INDEX_BITS, \
63 RADIX_TREE_MAP_SHIFT))
Linus Torvalds1da177e2005-04-16 15:20:36 -070064
Jeff Moyer26fb1582007-10-16 01:24:49 -070065/*
66 * The height_to_maxindex array needs to be one deeper than the maximum
67 * path as height 0 holds only 1 entry.
68 */
69static unsigned long height_to_maxindex[RADIX_TREE_MAX_PATH + 1] __read_mostly;
Linus Torvalds1da177e2005-04-16 15:20:36 -070070
71/*
72 * Radix tree node cache.
73 */
Christoph Lametere18b8902006-12-06 20:33:20 -080074static struct kmem_cache *radix_tree_node_cachep;
Linus Torvalds1da177e2005-04-16 15:20:36 -070075
76/*
77 * Per-cpu pool of preloaded nodes
78 */
79struct radix_tree_preload {
80 int nr;
81 struct radix_tree_node *nodes[RADIX_TREE_MAX_PATH];
82};
Harvey Harrison8cef7d52009-01-06 14:40:50 -080083static DEFINE_PER_CPU(struct radix_tree_preload, radix_tree_preloads) = { 0, };
Linus Torvalds1da177e2005-04-16 15:20:36 -070084
Nick Piggin612d6c12006-06-23 02:03:22 -070085static inline gfp_t root_gfp_mask(struct radix_tree_root *root)
86{
87 return root->gfp_mask & __GFP_BITS_MASK;
88}
89
Nick Piggin643b52b2008-06-12 15:21:52 -070090static inline void tag_set(struct radix_tree_node *node, unsigned int tag,
91 int offset)
92{
93 __set_bit(offset, node->tags[tag]);
94}
95
96static inline void tag_clear(struct radix_tree_node *node, unsigned int tag,
97 int offset)
98{
99 __clear_bit(offset, node->tags[tag]);
100}
101
102static inline int tag_get(struct radix_tree_node *node, unsigned int tag,
103 int offset)
104{
105 return test_bit(offset, node->tags[tag]);
106}
107
108static inline void root_tag_set(struct radix_tree_root *root, unsigned int tag)
109{
110 root->gfp_mask |= (__force gfp_t)(1 << (tag + __GFP_BITS_SHIFT));
111}
112
113static inline void root_tag_clear(struct radix_tree_root *root, unsigned int tag)
114{
115 root->gfp_mask &= (__force gfp_t)~(1 << (tag + __GFP_BITS_SHIFT));
116}
117
118static inline void root_tag_clear_all(struct radix_tree_root *root)
119{
120 root->gfp_mask &= __GFP_BITS_MASK;
121}
122
123static inline int root_tag_get(struct radix_tree_root *root, unsigned int tag)
124{
125 return (__force unsigned)root->gfp_mask & (1 << (tag + __GFP_BITS_SHIFT));
126}
127
128/*
129 * Returns 1 if any slot in the node has this tag set.
130 * Otherwise returns 0.
131 */
132static inline int any_tag_set(struct radix_tree_node *node, unsigned int tag)
133{
134 int idx;
135 for (idx = 0; idx < RADIX_TREE_TAG_LONGS; idx++) {
136 if (node->tags[tag][idx])
137 return 1;
138 }
139 return 0;
140}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700141/*
142 * This assumes that the caller has performed appropriate preallocation, and
143 * that the caller has pinned this thread of control to the current CPU.
144 */
145static struct radix_tree_node *
146radix_tree_node_alloc(struct radix_tree_root *root)
147{
Nick Piggine2848a02008-02-04 22:29:10 -0800148 struct radix_tree_node *ret = NULL;
Nick Piggin612d6c12006-06-23 02:03:22 -0700149 gfp_t gfp_mask = root_gfp_mask(root);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700150
Nick Piggine2848a02008-02-04 22:29:10 -0800151 if (!(gfp_mask & __GFP_WAIT)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700152 struct radix_tree_preload *rtp;
153
Nick Piggine2848a02008-02-04 22:29:10 -0800154 /*
155 * Provided the caller has preloaded here, we will always
156 * succeed in getting a node here (and never reach
157 * kmem_cache_alloc)
158 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700159 rtp = &__get_cpu_var(radix_tree_preloads);
160 if (rtp->nr) {
161 ret = rtp->nodes[rtp->nr - 1];
162 rtp->nodes[rtp->nr - 1] = NULL;
163 rtp->nr--;
164 }
165 }
Nick Piggine2848a02008-02-04 22:29:10 -0800166 if (ret == NULL)
Christoph Lameter488514d2008-04-28 02:12:05 -0700167 ret = kmem_cache_alloc(radix_tree_node_cachep, gfp_mask);
Nick Piggine2848a02008-02-04 22:29:10 -0800168
Nick Pigginc0bc9872007-10-16 01:24:42 -0700169 BUG_ON(radix_tree_is_indirect_ptr(ret));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700170 return ret;
171}
172
Nick Piggin7cf9c2c2006-12-06 20:33:44 -0800173static void radix_tree_node_rcu_free(struct rcu_head *head)
174{
175 struct radix_tree_node *node =
176 container_of(head, struct radix_tree_node, rcu_head);
Dave Chinnerb6dd0862010-08-23 10:33:19 +1000177 int i;
Nick Piggin643b52b2008-06-12 15:21:52 -0700178
179 /*
180 * must only free zeroed nodes into the slab. radix_tree_shrink
181 * can leave us with a non-NULL entry in the first slot, so clear
182 * that here to make sure.
183 */
Dave Chinnerb6dd0862010-08-23 10:33:19 +1000184 for (i = 0; i < RADIX_TREE_MAX_TAGS; i++)
185 tag_clear(node, i, 0);
186
Nick Piggin643b52b2008-06-12 15:21:52 -0700187 node->slots[0] = NULL;
188 node->count = 0;
189
Nick Piggin7cf9c2c2006-12-06 20:33:44 -0800190 kmem_cache_free(radix_tree_node_cachep, node);
191}
192
Linus Torvalds1da177e2005-04-16 15:20:36 -0700193static inline void
194radix_tree_node_free(struct radix_tree_node *node)
195{
Nick Piggin7cf9c2c2006-12-06 20:33:44 -0800196 call_rcu(&node->rcu_head, radix_tree_node_rcu_free);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700197}
198
199/*
200 * Load up this CPU's radix_tree_node buffer with sufficient objects to
201 * ensure that the addition of a single element in the tree cannot fail. On
202 * success, return zero, with preemption disabled. On error, return -ENOMEM
203 * with preemption not disabled.
David Howellsb34df792009-11-19 18:11:14 +0000204 *
205 * To make use of this facility, the radix tree must be initialised without
206 * __GFP_WAIT being passed to INIT_RADIX_TREE().
Linus Torvalds1da177e2005-04-16 15:20:36 -0700207 */
Al Virodd0fc662005-10-07 07:46:04 +0100208int radix_tree_preload(gfp_t gfp_mask)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700209{
210 struct radix_tree_preload *rtp;
211 struct radix_tree_node *node;
212 int ret = -ENOMEM;
213
214 preempt_disable();
215 rtp = &__get_cpu_var(radix_tree_preloads);
216 while (rtp->nr < ARRAY_SIZE(rtp->nodes)) {
217 preempt_enable();
Christoph Lameter488514d2008-04-28 02:12:05 -0700218 node = kmem_cache_alloc(radix_tree_node_cachep, gfp_mask);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700219 if (node == NULL)
220 goto out;
221 preempt_disable();
222 rtp = &__get_cpu_var(radix_tree_preloads);
223 if (rtp->nr < ARRAY_SIZE(rtp->nodes))
224 rtp->nodes[rtp->nr++] = node;
225 else
226 kmem_cache_free(radix_tree_node_cachep, node);
227 }
228 ret = 0;
229out:
230 return ret;
231}
David Chinnerd7f09232007-07-14 16:05:04 +1000232EXPORT_SYMBOL(radix_tree_preload);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700233
Nick Piggin6e954b92006-01-08 01:01:40 -0800234/*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700235 * Return the maximum key which can be store into a
236 * radix tree with height HEIGHT.
237 */
238static inline unsigned long radix_tree_maxindex(unsigned int height)
239{
240 return height_to_maxindex[height];
241}
242
243/*
244 * Extend a radix tree so it can store key @index.
245 */
246static int radix_tree_extend(struct radix_tree_root *root, unsigned long index)
247{
248 struct radix_tree_node *node;
249 unsigned int height;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700250 int tag;
251
252 /* Figure out what the height should be. */
253 height = root->height + 1;
254 while (index > radix_tree_maxindex(height))
255 height++;
256
257 if (root->rnode == NULL) {
258 root->height = height;
259 goto out;
260 }
261
Linus Torvalds1da177e2005-04-16 15:20:36 -0700262 do {
Nick Piggin7cf9c2c2006-12-06 20:33:44 -0800263 unsigned int newheight;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700264 if (!(node = radix_tree_node_alloc(root)))
265 return -ENOMEM;
266
267 /* Increase the height. */
Nick Pigginc0bc9872007-10-16 01:24:42 -0700268 node->slots[0] = radix_tree_indirect_to_ptr(root->rnode);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700269
270 /* Propagate the aggregated tag info into the new root */
Jonathan Corbetdaff89f2006-03-25 03:08:05 -0800271 for (tag = 0; tag < RADIX_TREE_MAX_TAGS; tag++) {
Nick Piggin612d6c12006-06-23 02:03:22 -0700272 if (root_tag_get(root, tag))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700273 tag_set(node, tag, 0);
274 }
275
Nick Piggin7cf9c2c2006-12-06 20:33:44 -0800276 newheight = root->height+1;
277 node->height = newheight;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700278 node->count = 1;
Nick Pigginc0bc9872007-10-16 01:24:42 -0700279 node = radix_tree_ptr_to_indirect(node);
Nick Piggin7cf9c2c2006-12-06 20:33:44 -0800280 rcu_assign_pointer(root->rnode, node);
281 root->height = newheight;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700282 } while (height > root->height);
283out:
284 return 0;
285}
286
287/**
288 * radix_tree_insert - insert into a radix tree
289 * @root: radix tree root
290 * @index: index key
291 * @item: item to insert
292 *
293 * Insert an item into the radix tree at position @index.
294 */
295int radix_tree_insert(struct radix_tree_root *root,
296 unsigned long index, void *item)
297{
Christoph Lameter201b6262005-09-06 15:16:46 -0700298 struct radix_tree_node *node = NULL, *slot;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700299 unsigned int height, shift;
300 int offset;
301 int error;
302
Nick Pigginc0bc9872007-10-16 01:24:42 -0700303 BUG_ON(radix_tree_is_indirect_ptr(item));
Nick Piggin7cf9c2c2006-12-06 20:33:44 -0800304
Linus Torvalds1da177e2005-04-16 15:20:36 -0700305 /* Make sure the tree is high enough. */
Nick Piggin612d6c12006-06-23 02:03:22 -0700306 if (index > radix_tree_maxindex(root->height)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700307 error = radix_tree_extend(root, index);
308 if (error)
309 return error;
310 }
311
Nick Pigginc0bc9872007-10-16 01:24:42 -0700312 slot = radix_tree_indirect_to_ptr(root->rnode);
313
Linus Torvalds1da177e2005-04-16 15:20:36 -0700314 height = root->height;
315 shift = (height-1) * RADIX_TREE_MAP_SHIFT;
316
317 offset = 0; /* uninitialised var warning */
Nick Piggin612d6c12006-06-23 02:03:22 -0700318 while (height > 0) {
Christoph Lameter201b6262005-09-06 15:16:46 -0700319 if (slot == NULL) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700320 /* Have to add a child node. */
Christoph Lameter201b6262005-09-06 15:16:46 -0700321 if (!(slot = radix_tree_node_alloc(root)))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700322 return -ENOMEM;
Nick Piggin7cf9c2c2006-12-06 20:33:44 -0800323 slot->height = height;
Christoph Lameter201b6262005-09-06 15:16:46 -0700324 if (node) {
Nick Piggin7cf9c2c2006-12-06 20:33:44 -0800325 rcu_assign_pointer(node->slots[offset], slot);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700326 node->count++;
Christoph Lameter201b6262005-09-06 15:16:46 -0700327 } else
Nick Pigginc0bc9872007-10-16 01:24:42 -0700328 rcu_assign_pointer(root->rnode,
329 radix_tree_ptr_to_indirect(slot));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700330 }
331
332 /* Go a level down */
333 offset = (index >> shift) & RADIX_TREE_MAP_MASK;
Christoph Lameter201b6262005-09-06 15:16:46 -0700334 node = slot;
335 slot = node->slots[offset];
Linus Torvalds1da177e2005-04-16 15:20:36 -0700336 shift -= RADIX_TREE_MAP_SHIFT;
337 height--;
Nick Piggin612d6c12006-06-23 02:03:22 -0700338 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700339
Christoph Lameter201b6262005-09-06 15:16:46 -0700340 if (slot != NULL)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700341 return -EEXIST;
Christoph Lameter201b6262005-09-06 15:16:46 -0700342
Nick Piggin612d6c12006-06-23 02:03:22 -0700343 if (node) {
344 node->count++;
Nick Piggin7cf9c2c2006-12-06 20:33:44 -0800345 rcu_assign_pointer(node->slots[offset], item);
Nick Piggin612d6c12006-06-23 02:03:22 -0700346 BUG_ON(tag_get(node, 0, offset));
347 BUG_ON(tag_get(node, 1, offset));
348 } else {
Nick Pigginc0bc9872007-10-16 01:24:42 -0700349 rcu_assign_pointer(root->rnode, item);
Nick Piggin612d6c12006-06-23 02:03:22 -0700350 BUG_ON(root_tag_get(root, 0));
351 BUG_ON(root_tag_get(root, 1));
352 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700353
Linus Torvalds1da177e2005-04-16 15:20:36 -0700354 return 0;
355}
356EXPORT_SYMBOL(radix_tree_insert);
357
Huang Shijieb72b71c2009-06-16 15:33:42 -0700358/*
359 * is_slot == 1 : search for the slot.
360 * is_slot == 0 : search for the node.
Hans Reisera4331362005-11-07 00:59:29 -0800361 */
Huang Shijieb72b71c2009-06-16 15:33:42 -0700362static void *radix_tree_lookup_element(struct radix_tree_root *root,
363 unsigned long index, int is_slot)
Hans Reisera4331362005-11-07 00:59:29 -0800364{
Nick Piggin7cf9c2c2006-12-06 20:33:44 -0800365 unsigned int height, shift;
366 struct radix_tree_node *node, **slot;
367
Paul E. McKenney2676a582010-02-22 17:04:54 -0800368 node = rcu_dereference_raw(root->rnode);
Nick Piggin7cf9c2c2006-12-06 20:33:44 -0800369 if (node == NULL)
370 return NULL;
371
Nick Pigginc0bc9872007-10-16 01:24:42 -0700372 if (!radix_tree_is_indirect_ptr(node)) {
Nick Piggin7cf9c2c2006-12-06 20:33:44 -0800373 if (index > 0)
374 return NULL;
Huang Shijieb72b71c2009-06-16 15:33:42 -0700375 return is_slot ? (void *)&root->rnode : node;
Nick Piggin7cf9c2c2006-12-06 20:33:44 -0800376 }
Nick Pigginc0bc9872007-10-16 01:24:42 -0700377 node = radix_tree_indirect_to_ptr(node);
Nick Piggin7cf9c2c2006-12-06 20:33:44 -0800378
379 height = node->height;
380 if (index > radix_tree_maxindex(height))
381 return NULL;
382
383 shift = (height-1) * RADIX_TREE_MAP_SHIFT;
384
385 do {
386 slot = (struct radix_tree_node **)
387 (node->slots + ((index>>shift) & RADIX_TREE_MAP_MASK));
Paul E. McKenney2676a582010-02-22 17:04:54 -0800388 node = rcu_dereference_raw(*slot);
Nick Piggin7cf9c2c2006-12-06 20:33:44 -0800389 if (node == NULL)
390 return NULL;
391
392 shift -= RADIX_TREE_MAP_SHIFT;
393 height--;
394 } while (height > 0);
395
Huang Shijieb72b71c2009-06-16 15:33:42 -0700396 return is_slot ? (void *)slot:node;
397}
398
399/**
400 * radix_tree_lookup_slot - lookup a slot in a radix tree
401 * @root: radix tree root
402 * @index: index key
403 *
404 * Returns: the slot corresponding to the position @index in the
405 * radix tree @root. This is useful for update-if-exists operations.
406 *
407 * This function can be called under rcu_read_lock iff the slot is not
408 * modified by radix_tree_replace_slot, otherwise it must be called
409 * exclusive from other writers. Any dereference of the slot must be done
410 * using radix_tree_deref_slot.
411 */
412void **radix_tree_lookup_slot(struct radix_tree_root *root, unsigned long index)
413{
414 return (void **)radix_tree_lookup_element(root, index, 1);
Hans Reisera4331362005-11-07 00:59:29 -0800415}
416EXPORT_SYMBOL(radix_tree_lookup_slot);
417
Linus Torvalds1da177e2005-04-16 15:20:36 -0700418/**
419 * radix_tree_lookup - perform lookup operation on a radix tree
420 * @root: radix tree root
421 * @index: index key
422 *
423 * Lookup the item at the position @index in the radix tree @root.
Nick Piggin7cf9c2c2006-12-06 20:33:44 -0800424 *
425 * This function can be called under rcu_read_lock, however the caller
426 * must manage lifetimes of leaf nodes (eg. RCU may also be used to free
427 * them safely). No RCU barriers are required to access or modify the
428 * returned item, however.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700429 */
430void *radix_tree_lookup(struct radix_tree_root *root, unsigned long index)
431{
Huang Shijieb72b71c2009-06-16 15:33:42 -0700432 return radix_tree_lookup_element(root, index, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700433}
434EXPORT_SYMBOL(radix_tree_lookup);
435
436/**
437 * radix_tree_tag_set - set a tag on a radix tree node
438 * @root: radix tree root
439 * @index: index key
440 * @tag: tag index
441 *
Jonathan Corbetdaff89f2006-03-25 03:08:05 -0800442 * Set the search tag (which must be < RADIX_TREE_MAX_TAGS)
443 * corresponding to @index in the radix tree. From
Linus Torvalds1da177e2005-04-16 15:20:36 -0700444 * the root all the way down to the leaf node.
445 *
446 * Returns the address of the tagged item. Setting a tag on a not-present
447 * item is a bug.
448 */
449void *radix_tree_tag_set(struct radix_tree_root *root,
Jonathan Corbetdaff89f2006-03-25 03:08:05 -0800450 unsigned long index, unsigned int tag)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700451{
452 unsigned int height, shift;
Christoph Lameter201b6262005-09-06 15:16:46 -0700453 struct radix_tree_node *slot;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700454
455 height = root->height;
Peter Zijlstra4c91c362006-06-23 02:03:25 -0700456 BUG_ON(index > radix_tree_maxindex(height));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700457
Nick Pigginc0bc9872007-10-16 01:24:42 -0700458 slot = radix_tree_indirect_to_ptr(root->rnode);
Nick Piggin612d6c12006-06-23 02:03:22 -0700459 shift = (height - 1) * RADIX_TREE_MAP_SHIFT;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700460
461 while (height > 0) {
462 int offset;
463
464 offset = (index >> shift) & RADIX_TREE_MAP_MASK;
Nick Piggind5274262006-01-08 01:01:41 -0800465 if (!tag_get(slot, tag, offset))
466 tag_set(slot, tag, offset);
Christoph Lameter201b6262005-09-06 15:16:46 -0700467 slot = slot->slots[offset];
468 BUG_ON(slot == NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700469 shift -= RADIX_TREE_MAP_SHIFT;
470 height--;
471 }
472
Nick Piggin612d6c12006-06-23 02:03:22 -0700473 /* set the root's tag bit */
474 if (slot && !root_tag_get(root, tag))
475 root_tag_set(root, tag);
476
Christoph Lameter201b6262005-09-06 15:16:46 -0700477 return slot;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700478}
479EXPORT_SYMBOL(radix_tree_tag_set);
480
481/**
482 * radix_tree_tag_clear - clear a tag on a radix tree node
483 * @root: radix tree root
484 * @index: index key
485 * @tag: tag index
486 *
Jonathan Corbetdaff89f2006-03-25 03:08:05 -0800487 * Clear the search tag (which must be < RADIX_TREE_MAX_TAGS)
488 * corresponding to @index in the radix tree. If
Linus Torvalds1da177e2005-04-16 15:20:36 -0700489 * this causes the leaf node to have no tags set then clear the tag in the
490 * next-to-leaf node, etc.
491 *
492 * Returns the address of the tagged item on success, else NULL. ie:
493 * has the same return value and semantics as radix_tree_lookup().
494 */
495void *radix_tree_tag_clear(struct radix_tree_root *root,
Jonathan Corbetdaff89f2006-03-25 03:08:05 -0800496 unsigned long index, unsigned int tag)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700497{
Jeff Moyer26fb1582007-10-16 01:24:49 -0700498 /*
499 * The radix tree path needs to be one longer than the maximum path
500 * since the "list" is null terminated.
501 */
502 struct radix_tree_path path[RADIX_TREE_MAX_PATH + 1], *pathp = path;
Nick Piggin612d6c12006-06-23 02:03:22 -0700503 struct radix_tree_node *slot = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700504 unsigned int height, shift;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700505
506 height = root->height;
507 if (index > radix_tree_maxindex(height))
508 goto out;
509
510 shift = (height - 1) * RADIX_TREE_MAP_SHIFT;
511 pathp->node = NULL;
Nick Pigginc0bc9872007-10-16 01:24:42 -0700512 slot = radix_tree_indirect_to_ptr(root->rnode);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700513
514 while (height > 0) {
515 int offset;
516
Christoph Lameter201b6262005-09-06 15:16:46 -0700517 if (slot == NULL)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700518 goto out;
519
520 offset = (index >> shift) & RADIX_TREE_MAP_MASK;
521 pathp[1].offset = offset;
Christoph Lameter201b6262005-09-06 15:16:46 -0700522 pathp[1].node = slot;
523 slot = slot->slots[offset];
Linus Torvalds1da177e2005-04-16 15:20:36 -0700524 pathp++;
525 shift -= RADIX_TREE_MAP_SHIFT;
526 height--;
527 }
528
Nick Piggin612d6c12006-06-23 02:03:22 -0700529 if (slot == NULL)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700530 goto out;
531
Nick Piggin612d6c12006-06-23 02:03:22 -0700532 while (pathp->node) {
Nick Piggind5274262006-01-08 01:01:41 -0800533 if (!tag_get(pathp->node, tag, pathp->offset))
534 goto out;
Christoph Lameter201b6262005-09-06 15:16:46 -0700535 tag_clear(pathp->node, tag, pathp->offset);
Nick Piggin6e954b92006-01-08 01:01:40 -0800536 if (any_tag_set(pathp->node, tag))
537 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700538 pathp--;
Nick Piggin612d6c12006-06-23 02:03:22 -0700539 }
540
541 /* clear the root's tag bit */
542 if (root_tag_get(root, tag))
543 root_tag_clear(root, tag);
544
Linus Torvalds1da177e2005-04-16 15:20:36 -0700545out:
Nick Piggin612d6c12006-06-23 02:03:22 -0700546 return slot;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700547}
548EXPORT_SYMBOL(radix_tree_tag_clear);
549
Linus Torvalds1da177e2005-04-16 15:20:36 -0700550/**
Marcelo Tosatti32605a12005-09-06 15:16:48 -0700551 * radix_tree_tag_get - get a tag on a radix tree node
552 * @root: radix tree root
553 * @index: index key
Jonathan Corbetdaff89f2006-03-25 03:08:05 -0800554 * @tag: tag index (< RADIX_TREE_MAX_TAGS)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700555 *
Marcelo Tosatti32605a12005-09-06 15:16:48 -0700556 * Return values:
Linus Torvalds1da177e2005-04-16 15:20:36 -0700557 *
Nick Piggin612d6c12006-06-23 02:03:22 -0700558 * 0: tag not present or not set
559 * 1: tag set
David Howellsce826532010-04-06 22:36:20 +0100560 *
561 * Note that the return value of this function may not be relied on, even if
562 * the RCU lock is held, unless tag modification and node deletion are excluded
563 * from concurrency.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700564 */
565int radix_tree_tag_get(struct radix_tree_root *root,
Jonathan Corbetdaff89f2006-03-25 03:08:05 -0800566 unsigned long index, unsigned int tag)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700567{
568 unsigned int height, shift;
Nick Piggin7cf9c2c2006-12-06 20:33:44 -0800569 struct radix_tree_node *node;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700570 int saw_unset_tag = 0;
571
Nick Piggin612d6c12006-06-23 02:03:22 -0700572 /* check the root's tag bit */
573 if (!root_tag_get(root, tag))
574 return 0;
575
Paul E. McKenney2676a582010-02-22 17:04:54 -0800576 node = rcu_dereference_raw(root->rnode);
Nick Piggin7cf9c2c2006-12-06 20:33:44 -0800577 if (node == NULL)
578 return 0;
579
Nick Pigginc0bc9872007-10-16 01:24:42 -0700580 if (!radix_tree_is_indirect_ptr(node))
Nick Piggin7cf9c2c2006-12-06 20:33:44 -0800581 return (index == 0);
Nick Pigginc0bc9872007-10-16 01:24:42 -0700582 node = radix_tree_indirect_to_ptr(node);
Nick Piggin7cf9c2c2006-12-06 20:33:44 -0800583
584 height = node->height;
585 if (index > radix_tree_maxindex(height))
586 return 0;
Nick Piggin612d6c12006-06-23 02:03:22 -0700587
Linus Torvalds1da177e2005-04-16 15:20:36 -0700588 shift = (height - 1) * RADIX_TREE_MAP_SHIFT;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700589
590 for ( ; ; ) {
591 int offset;
592
Nick Piggin7cf9c2c2006-12-06 20:33:44 -0800593 if (node == NULL)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700594 return 0;
595
596 offset = (index >> shift) & RADIX_TREE_MAP_MASK;
597
598 /*
599 * This is just a debug check. Later, we can bale as soon as
600 * we see an unset tag.
601 */
Nick Piggin7cf9c2c2006-12-06 20:33:44 -0800602 if (!tag_get(node, tag, offset))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700603 saw_unset_tag = 1;
David Howellsce826532010-04-06 22:36:20 +0100604 if (height == 1)
605 return !!tag_get(node, tag, offset);
Paul E. McKenney2676a582010-02-22 17:04:54 -0800606 node = rcu_dereference_raw(node->slots[offset]);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700607 shift -= RADIX_TREE_MAP_SHIFT;
608 height--;
609 }
610}
611EXPORT_SYMBOL(radix_tree_tag_get);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700612
Fengguang Wu6df8ba42007-10-16 01:24:33 -0700613/**
Jan Karaebf8aa42010-08-09 17:19:11 -0700614 * radix_tree_range_tag_if_tagged - for each item in given range set given
615 * tag if item has another tag set
616 * @root: radix tree root
617 * @first_indexp: pointer to a starting index of a range to scan
618 * @last_index: last index of a range to scan
619 * @nr_to_tag: maximum number items to tag
620 * @iftag: tag index to test
621 * @settag: tag index to set if tested tag is set
622 *
623 * This function scans range of radix tree from first_index to last_index
624 * (inclusive). For each item in the range if iftag is set, the function sets
625 * also settag. The function stops either after tagging nr_to_tag items or
626 * after reaching last_index.
627 *
Dave Chinner144dcfc2010-08-23 10:33:53 +1000628 * The tags must be set from the leaf level only and propagated back up the
629 * path to the root. We must do this so that we resolve the full path before
630 * setting any tags on intermediate nodes. If we set tags as we descend, then
631 * we can get to the leaf node and find that the index that has the iftag
632 * set is outside the range we are scanning. This reults in dangling tags and
633 * can lead to problems with later tag operations (e.g. livelocks on lookups).
634 *
Jan Karaebf8aa42010-08-09 17:19:11 -0700635 * The function returns number of leaves where the tag was set and sets
636 * *first_indexp to the first unscanned index.
Jan Karad5ed3a42010-08-19 14:13:33 -0700637 * WARNING! *first_indexp can wrap if last_index is ULONG_MAX. Caller must
638 * be prepared to handle that.
Jan Karaebf8aa42010-08-09 17:19:11 -0700639 */
640unsigned long radix_tree_range_tag_if_tagged(struct radix_tree_root *root,
641 unsigned long *first_indexp, unsigned long last_index,
642 unsigned long nr_to_tag,
643 unsigned int iftag, unsigned int settag)
644{
Dave Chinner144dcfc2010-08-23 10:33:53 +1000645 unsigned int height = root->height;
646 struct radix_tree_path path[height];
647 struct radix_tree_path *pathp = path;
648 struct radix_tree_node *slot;
649 unsigned int shift;
650 unsigned long tagged = 0;
651 unsigned long index = *first_indexp;
Jan Karaebf8aa42010-08-09 17:19:11 -0700652
653 last_index = min(last_index, radix_tree_maxindex(height));
654 if (index > last_index)
655 return 0;
656 if (!nr_to_tag)
657 return 0;
658 if (!root_tag_get(root, iftag)) {
659 *first_indexp = last_index + 1;
660 return 0;
661 }
662 if (height == 0) {
663 *first_indexp = last_index + 1;
664 root_tag_set(root, settag);
665 return 1;
666 }
667
668 shift = (height - 1) * RADIX_TREE_MAP_SHIFT;
669 slot = radix_tree_indirect_to_ptr(root->rnode);
670
Dave Chinner144dcfc2010-08-23 10:33:53 +1000671 /*
672 * we fill the path from (root->height - 2) to 0, leaving the index at
673 * (root->height - 1) as a terminator. Zero the node in the terminator
674 * so that we can use this to end walk loops back up the path.
675 */
676 path[height - 1].node = NULL;
677
Jan Karaebf8aa42010-08-09 17:19:11 -0700678 for (;;) {
679 int offset;
680
681 offset = (index >> shift) & RADIX_TREE_MAP_MASK;
682 if (!slot->slots[offset])
683 goto next;
684 if (!tag_get(slot, iftag, offset))
685 goto next;
Dave Chinner144dcfc2010-08-23 10:33:53 +1000686 if (height > 1) {
687 /* Go down one level */
688 height--;
689 shift -= RADIX_TREE_MAP_SHIFT;
690 path[height - 1].node = slot;
691 path[height - 1].offset = offset;
692 slot = slot->slots[offset];
693 continue;
Jan Karaebf8aa42010-08-09 17:19:11 -0700694 }
Dave Chinner144dcfc2010-08-23 10:33:53 +1000695
696 /* tag the leaf */
697 tagged++;
698 tag_set(slot, settag, offset);
699
700 /* walk back up the path tagging interior nodes */
701 pathp = &path[0];
702 while (pathp->node) {
703 /* stop if we find a node with the tag already set */
704 if (tag_get(pathp->node, settag, pathp->offset))
705 break;
706 tag_set(pathp->node, settag, pathp->offset);
707 pathp++;
708 }
709
Jan Karaebf8aa42010-08-09 17:19:11 -0700710next:
711 /* Go to next item at level determined by 'shift' */
712 index = ((index >> shift) + 1) << shift;
Jan Karad5ed3a42010-08-19 14:13:33 -0700713 /* Overflow can happen when last_index is ~0UL... */
714 if (index > last_index || !index)
Jan Karaebf8aa42010-08-09 17:19:11 -0700715 break;
716 if (tagged >= nr_to_tag)
717 break;
718 while (((index >> shift) & RADIX_TREE_MAP_MASK) == 0) {
719 /*
720 * We've fully scanned this node. Go up. Because
721 * last_index is guaranteed to be in the tree, what
722 * we do below cannot wander astray.
723 */
Dave Chinner144dcfc2010-08-23 10:33:53 +1000724 slot = path[height - 1].node;
Jan Karaebf8aa42010-08-09 17:19:11 -0700725 height++;
726 shift += RADIX_TREE_MAP_SHIFT;
727 }
728 }
729 /*
730 * The iftag must have been set somewhere because otherwise
731 * we would return immediated at the beginning of the function
732 */
733 root_tag_set(root, settag);
734 *first_indexp = index;
735
736 return tagged;
737}
738EXPORT_SYMBOL(radix_tree_range_tag_if_tagged);
739
740
741/**
Fengguang Wu6df8ba42007-10-16 01:24:33 -0700742 * radix_tree_next_hole - find the next hole (not-present entry)
743 * @root: tree root
744 * @index: index key
745 * @max_scan: maximum range to search
746 *
747 * Search the set [index, min(index+max_scan-1, MAX_INDEX)] for the lowest
748 * indexed hole.
749 *
750 * Returns: the index of the hole if found, otherwise returns an index
751 * outside of the set specified (in which case 'return - index >= max_scan'
Wu Fengguang8e6bdb72008-11-27 11:42:22 +0100752 * will be true). In rare cases of index wrap-around, 0 will be returned.
Fengguang Wu6df8ba42007-10-16 01:24:33 -0700753 *
754 * radix_tree_next_hole may be called under rcu_read_lock. However, like
Wu Fengguang8e6bdb72008-11-27 11:42:22 +0100755 * radix_tree_gang_lookup, this will not atomically search a snapshot of
756 * the tree at a single point in time. For example, if a hole is created
757 * at index 5, then subsequently a hole is created at index 10,
758 * radix_tree_next_hole covering both indexes may return 10 if called
759 * under rcu_read_lock.
Fengguang Wu6df8ba42007-10-16 01:24:33 -0700760 */
761unsigned long radix_tree_next_hole(struct radix_tree_root *root,
762 unsigned long index, unsigned long max_scan)
763{
764 unsigned long i;
765
766 for (i = 0; i < max_scan; i++) {
767 if (!radix_tree_lookup(root, index))
768 break;
769 index++;
770 if (index == 0)
771 break;
772 }
773
774 return index;
775}
776EXPORT_SYMBOL(radix_tree_next_hole);
777
Wu Fengguangdc566122009-06-16 15:31:32 -0700778/**
779 * radix_tree_prev_hole - find the prev hole (not-present entry)
780 * @root: tree root
781 * @index: index key
782 * @max_scan: maximum range to search
783 *
784 * Search backwards in the range [max(index-max_scan+1, 0), index]
785 * for the first hole.
786 *
787 * Returns: the index of the hole if found, otherwise returns an index
788 * outside of the set specified (in which case 'index - return >= max_scan'
Cesar Eduardo Barrosedcd1d82010-05-26 14:44:27 -0700789 * will be true). In rare cases of wrap-around, ULONG_MAX will be returned.
Wu Fengguangdc566122009-06-16 15:31:32 -0700790 *
791 * radix_tree_next_hole may be called under rcu_read_lock. However, like
792 * radix_tree_gang_lookup, this will not atomically search a snapshot of
793 * the tree at a single point in time. For example, if a hole is created
794 * at index 10, then subsequently a hole is created at index 5,
795 * radix_tree_prev_hole covering both indexes may return 5 if called under
796 * rcu_read_lock.
797 */
798unsigned long radix_tree_prev_hole(struct radix_tree_root *root,
799 unsigned long index, unsigned long max_scan)
800{
801 unsigned long i;
802
803 for (i = 0; i < max_scan; i++) {
804 if (!radix_tree_lookup(root, index))
805 break;
806 index--;
Cesar Eduardo Barrosedcd1d82010-05-26 14:44:27 -0700807 if (index == ULONG_MAX)
Wu Fengguangdc566122009-06-16 15:31:32 -0700808 break;
809 }
810
811 return index;
812}
813EXPORT_SYMBOL(radix_tree_prev_hole);
814
Linus Torvalds1da177e2005-04-16 15:20:36 -0700815static unsigned int
Nick Piggin47feff22008-07-25 19:45:29 -0700816__lookup(struct radix_tree_node *slot, void ***results, unsigned long index,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700817 unsigned int max_items, unsigned long *next_index)
818{
819 unsigned int nr_found = 0;
Christoph Lameter201b6262005-09-06 15:16:46 -0700820 unsigned int shift, height;
Christoph Lameter201b6262005-09-06 15:16:46 -0700821 unsigned long i;
822
Nick Piggin7cf9c2c2006-12-06 20:33:44 -0800823 height = slot->height;
824 if (height == 0)
Christoph Lameter201b6262005-09-06 15:16:46 -0700825 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700826 shift = (height-1) * RADIX_TREE_MAP_SHIFT;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700827
Christoph Lameter201b6262005-09-06 15:16:46 -0700828 for ( ; height > 1; height--) {
Nick Piggin7cf9c2c2006-12-06 20:33:44 -0800829 i = (index >> shift) & RADIX_TREE_MAP_MASK;
830 for (;;) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700831 if (slot->slots[i] != NULL)
832 break;
833 index &= ~((1UL << shift) - 1);
834 index += 1UL << shift;
835 if (index == 0)
836 goto out; /* 32-bit wraparound */
Nick Piggin7cf9c2c2006-12-06 20:33:44 -0800837 i++;
838 if (i == RADIX_TREE_MAP_SIZE)
839 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700840 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700841
Linus Torvalds1da177e2005-04-16 15:20:36 -0700842 shift -= RADIX_TREE_MAP_SHIFT;
Paul E. McKenney2676a582010-02-22 17:04:54 -0800843 slot = rcu_dereference_raw(slot->slots[i]);
Nick Piggin7cf9c2c2006-12-06 20:33:44 -0800844 if (slot == NULL)
845 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700846 }
Christoph Lameter201b6262005-09-06 15:16:46 -0700847
848 /* Bottom level: grab some items */
849 for (i = index & RADIX_TREE_MAP_MASK; i < RADIX_TREE_MAP_SIZE; i++) {
850 index++;
Nick Piggin47feff22008-07-25 19:45:29 -0700851 if (slot->slots[i]) {
852 results[nr_found++] = &(slot->slots[i]);
Christoph Lameter201b6262005-09-06 15:16:46 -0700853 if (nr_found == max_items)
854 goto out;
855 }
856 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700857out:
858 *next_index = index;
859 return nr_found;
860}
861
862/**
863 * radix_tree_gang_lookup - perform multiple lookup on a radix tree
864 * @root: radix tree root
865 * @results: where the results of the lookup are placed
866 * @first_index: start the lookup from this key
867 * @max_items: place up to this many items at *results
868 *
869 * Performs an index-ascending scan of the tree for present items. Places
870 * them at *@results and returns the number of items which were placed at
871 * *@results.
872 *
873 * The implementation is naive.
Nick Piggin7cf9c2c2006-12-06 20:33:44 -0800874 *
875 * Like radix_tree_lookup, radix_tree_gang_lookup may be called under
876 * rcu_read_lock. In this case, rather than the returned results being
877 * an atomic snapshot of the tree at a single point in time, the semantics
878 * of an RCU protected gang lookup are as though multiple radix_tree_lookups
879 * have been issued in individual locks, and results stored in 'results'.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700880 */
881unsigned int
882radix_tree_gang_lookup(struct radix_tree_root *root, void **results,
883 unsigned long first_index, unsigned int max_items)
884{
Nick Piggin7cf9c2c2006-12-06 20:33:44 -0800885 unsigned long max_index;
886 struct radix_tree_node *node;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700887 unsigned long cur_index = first_index;
Nick Piggin7cf9c2c2006-12-06 20:33:44 -0800888 unsigned int ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700889
Paul E. McKenney2676a582010-02-22 17:04:54 -0800890 node = rcu_dereference_raw(root->rnode);
Nick Piggin7cf9c2c2006-12-06 20:33:44 -0800891 if (!node)
892 return 0;
893
Nick Pigginc0bc9872007-10-16 01:24:42 -0700894 if (!radix_tree_is_indirect_ptr(node)) {
Nick Piggin7cf9c2c2006-12-06 20:33:44 -0800895 if (first_index > 0)
896 return 0;
Nick Pigginc0bc9872007-10-16 01:24:42 -0700897 results[0] = node;
Nick Piggin7cf9c2c2006-12-06 20:33:44 -0800898 return 1;
899 }
Nick Pigginc0bc9872007-10-16 01:24:42 -0700900 node = radix_tree_indirect_to_ptr(node);
Nick Piggin7cf9c2c2006-12-06 20:33:44 -0800901
902 max_index = radix_tree_maxindex(node->height);
903
904 ret = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700905 while (ret < max_items) {
Nick Piggin47feff22008-07-25 19:45:29 -0700906 unsigned int nr_found, slots_found, i;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700907 unsigned long next_index; /* Index of next search */
908
909 if (cur_index > max_index)
910 break;
Nick Piggin47feff22008-07-25 19:45:29 -0700911 slots_found = __lookup(node, (void ***)results + ret, cur_index,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700912 max_items - ret, &next_index);
Nick Piggin47feff22008-07-25 19:45:29 -0700913 nr_found = 0;
914 for (i = 0; i < slots_found; i++) {
915 struct radix_tree_node *slot;
916 slot = *(((void ***)results)[ret + i]);
917 if (!slot)
918 continue;
Paul E. McKenney2676a582010-02-22 17:04:54 -0800919 results[ret + nr_found] = rcu_dereference_raw(slot);
Nick Piggin47feff22008-07-25 19:45:29 -0700920 nr_found++;
921 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700922 ret += nr_found;
923 if (next_index == 0)
924 break;
925 cur_index = next_index;
926 }
Nick Piggin7cf9c2c2006-12-06 20:33:44 -0800927
Linus Torvalds1da177e2005-04-16 15:20:36 -0700928 return ret;
929}
930EXPORT_SYMBOL(radix_tree_gang_lookup);
931
Nick Piggin47feff22008-07-25 19:45:29 -0700932/**
933 * radix_tree_gang_lookup_slot - perform multiple slot lookup on radix tree
934 * @root: radix tree root
935 * @results: where the results of the lookup are placed
936 * @first_index: start the lookup from this key
937 * @max_items: place up to this many items at *results
938 *
939 * Performs an index-ascending scan of the tree for present items. Places
940 * their slots at *@results and returns the number of items which were
941 * placed at *@results.
942 *
943 * The implementation is naive.
944 *
945 * Like radix_tree_gang_lookup as far as RCU and locking goes. Slots must
946 * be dereferenced with radix_tree_deref_slot, and if using only RCU
947 * protection, radix_tree_deref_slot may fail requiring a retry.
948 */
949unsigned int
950radix_tree_gang_lookup_slot(struct radix_tree_root *root, void ***results,
951 unsigned long first_index, unsigned int max_items)
952{
953 unsigned long max_index;
954 struct radix_tree_node *node;
955 unsigned long cur_index = first_index;
956 unsigned int ret;
957
Paul E. McKenney2676a582010-02-22 17:04:54 -0800958 node = rcu_dereference_raw(root->rnode);
Nick Piggin47feff22008-07-25 19:45:29 -0700959 if (!node)
960 return 0;
961
962 if (!radix_tree_is_indirect_ptr(node)) {
963 if (first_index > 0)
964 return 0;
965 results[0] = (void **)&root->rnode;
966 return 1;
967 }
968 node = radix_tree_indirect_to_ptr(node);
969
970 max_index = radix_tree_maxindex(node->height);
971
972 ret = 0;
973 while (ret < max_items) {
974 unsigned int slots_found;
975 unsigned long next_index; /* Index of next search */
976
977 if (cur_index > max_index)
978 break;
979 slots_found = __lookup(node, results + ret, cur_index,
980 max_items - ret, &next_index);
981 ret += slots_found;
982 if (next_index == 0)
983 break;
984 cur_index = next_index;
985 }
986
987 return ret;
988}
989EXPORT_SYMBOL(radix_tree_gang_lookup_slot);
990
Linus Torvalds1da177e2005-04-16 15:20:36 -0700991/*
992 * FIXME: the two tag_get()s here should use find_next_bit() instead of
993 * open-coding the search.
994 */
995static unsigned int
Nick Piggin47feff22008-07-25 19:45:29 -0700996__lookup_tag(struct radix_tree_node *slot, void ***results, unsigned long index,
Jonathan Corbetdaff89f2006-03-25 03:08:05 -0800997 unsigned int max_items, unsigned long *next_index, unsigned int tag)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700998{
999 unsigned int nr_found = 0;
Nick Piggin7cf9c2c2006-12-06 20:33:44 -08001000 unsigned int shift, height;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001001
Nick Piggin7cf9c2c2006-12-06 20:33:44 -08001002 height = slot->height;
1003 if (height == 0)
Nick Piggin612d6c12006-06-23 02:03:22 -07001004 goto out;
Nick Piggin7cf9c2c2006-12-06 20:33:44 -08001005 shift = (height-1) * RADIX_TREE_MAP_SHIFT;
Nick Piggin612d6c12006-06-23 02:03:22 -07001006
Nick Piggin7cf9c2c2006-12-06 20:33:44 -08001007 while (height > 0) {
1008 unsigned long i = (index >> shift) & RADIX_TREE_MAP_MASK ;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001009
Nick Piggin7cf9c2c2006-12-06 20:33:44 -08001010 for (;;) {
1011 if (tag_get(slot, tag, i))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001012 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001013 index &= ~((1UL << shift) - 1);
1014 index += 1UL << shift;
1015 if (index == 0)
1016 goto out; /* 32-bit wraparound */
Nick Piggin7cf9c2c2006-12-06 20:33:44 -08001017 i++;
1018 if (i == RADIX_TREE_MAP_SIZE)
1019 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001020 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001021 height--;
1022 if (height == 0) { /* Bottom level: grab some items */
1023 unsigned long j = index & RADIX_TREE_MAP_MASK;
1024
1025 for ( ; j < RADIX_TREE_MAP_SIZE; j++) {
1026 index++;
Nick Piggin7cf9c2c2006-12-06 20:33:44 -08001027 if (!tag_get(slot, tag, j))
1028 continue;
Nick Piggin7cf9c2c2006-12-06 20:33:44 -08001029 /*
1030 * Even though the tag was found set, we need to
1031 * recheck that we have a non-NULL node, because
1032 * if this lookup is lockless, it may have been
1033 * subsequently deleted.
1034 *
1035 * Similar care must be taken in any place that
1036 * lookup ->slots[x] without a lock (ie. can't
1037 * rely on its value remaining the same).
1038 */
Nick Piggin47feff22008-07-25 19:45:29 -07001039 if (slot->slots[j]) {
1040 results[nr_found++] = &(slot->slots[j]);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001041 if (nr_found == max_items)
1042 goto out;
1043 }
1044 }
1045 }
1046 shift -= RADIX_TREE_MAP_SHIFT;
Paul E. McKenney2676a582010-02-22 17:04:54 -08001047 slot = rcu_dereference_raw(slot->slots[i]);
Nick Piggin7cf9c2c2006-12-06 20:33:44 -08001048 if (slot == NULL)
1049 break;
1050 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001051out:
1052 *next_index = index;
1053 return nr_found;
1054}
1055
1056/**
1057 * radix_tree_gang_lookup_tag - perform multiple lookup on a radix tree
1058 * based on a tag
1059 * @root: radix tree root
1060 * @results: where the results of the lookup are placed
1061 * @first_index: start the lookup from this key
1062 * @max_items: place up to this many items at *results
Jonathan Corbetdaff89f2006-03-25 03:08:05 -08001063 * @tag: the tag index (< RADIX_TREE_MAX_TAGS)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001064 *
1065 * Performs an index-ascending scan of the tree for present items which
1066 * have the tag indexed by @tag set. Places the items at *@results and
1067 * returns the number of items which were placed at *@results.
1068 */
1069unsigned int
1070radix_tree_gang_lookup_tag(struct radix_tree_root *root, void **results,
Jonathan Corbetdaff89f2006-03-25 03:08:05 -08001071 unsigned long first_index, unsigned int max_items,
1072 unsigned int tag)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001073{
Nick Piggin7cf9c2c2006-12-06 20:33:44 -08001074 struct radix_tree_node *node;
1075 unsigned long max_index;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001076 unsigned long cur_index = first_index;
Nick Piggin7cf9c2c2006-12-06 20:33:44 -08001077 unsigned int ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001078
Nick Piggin612d6c12006-06-23 02:03:22 -07001079 /* check the root's tag bit */
1080 if (!root_tag_get(root, tag))
1081 return 0;
1082
Paul E. McKenney2676a582010-02-22 17:04:54 -08001083 node = rcu_dereference_raw(root->rnode);
Nick Piggin7cf9c2c2006-12-06 20:33:44 -08001084 if (!node)
1085 return 0;
1086
Nick Pigginc0bc9872007-10-16 01:24:42 -07001087 if (!radix_tree_is_indirect_ptr(node)) {
Nick Piggin7cf9c2c2006-12-06 20:33:44 -08001088 if (first_index > 0)
1089 return 0;
Nick Pigginc0bc9872007-10-16 01:24:42 -07001090 results[0] = node;
Nick Piggin7cf9c2c2006-12-06 20:33:44 -08001091 return 1;
1092 }
Nick Pigginc0bc9872007-10-16 01:24:42 -07001093 node = radix_tree_indirect_to_ptr(node);
Nick Piggin7cf9c2c2006-12-06 20:33:44 -08001094
1095 max_index = radix_tree_maxindex(node->height);
1096
1097 ret = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001098 while (ret < max_items) {
Nick Piggin47feff22008-07-25 19:45:29 -07001099 unsigned int nr_found, slots_found, i;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001100 unsigned long next_index; /* Index of next search */
1101
1102 if (cur_index > max_index)
1103 break;
Nick Piggin47feff22008-07-25 19:45:29 -07001104 slots_found = __lookup_tag(node, (void ***)results + ret,
1105 cur_index, max_items - ret, &next_index, tag);
1106 nr_found = 0;
1107 for (i = 0; i < slots_found; i++) {
1108 struct radix_tree_node *slot;
1109 slot = *(((void ***)results)[ret + i]);
1110 if (!slot)
1111 continue;
Paul E. McKenney2676a582010-02-22 17:04:54 -08001112 results[ret + nr_found] = rcu_dereference_raw(slot);
Nick Piggin47feff22008-07-25 19:45:29 -07001113 nr_found++;
1114 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001115 ret += nr_found;
1116 if (next_index == 0)
1117 break;
1118 cur_index = next_index;
1119 }
Nick Piggin7cf9c2c2006-12-06 20:33:44 -08001120
Linus Torvalds1da177e2005-04-16 15:20:36 -07001121 return ret;
1122}
1123EXPORT_SYMBOL(radix_tree_gang_lookup_tag);
1124
1125/**
Nick Piggin47feff22008-07-25 19:45:29 -07001126 * radix_tree_gang_lookup_tag_slot - perform multiple slot lookup on a
1127 * radix tree based on a tag
1128 * @root: radix tree root
1129 * @results: where the results of the lookup are placed
1130 * @first_index: start the lookup from this key
1131 * @max_items: place up to this many items at *results
1132 * @tag: the tag index (< RADIX_TREE_MAX_TAGS)
1133 *
1134 * Performs an index-ascending scan of the tree for present items which
1135 * have the tag indexed by @tag set. Places the slots at *@results and
1136 * returns the number of slots which were placed at *@results.
1137 */
1138unsigned int
1139radix_tree_gang_lookup_tag_slot(struct radix_tree_root *root, void ***results,
1140 unsigned long first_index, unsigned int max_items,
1141 unsigned int tag)
1142{
1143 struct radix_tree_node *node;
1144 unsigned long max_index;
1145 unsigned long cur_index = first_index;
1146 unsigned int ret;
1147
1148 /* check the root's tag bit */
1149 if (!root_tag_get(root, tag))
1150 return 0;
1151
Paul E. McKenney2676a582010-02-22 17:04:54 -08001152 node = rcu_dereference_raw(root->rnode);
Nick Piggin47feff22008-07-25 19:45:29 -07001153 if (!node)
1154 return 0;
1155
1156 if (!radix_tree_is_indirect_ptr(node)) {
1157 if (first_index > 0)
1158 return 0;
1159 results[0] = (void **)&root->rnode;
1160 return 1;
1161 }
1162 node = radix_tree_indirect_to_ptr(node);
1163
1164 max_index = radix_tree_maxindex(node->height);
1165
1166 ret = 0;
1167 while (ret < max_items) {
1168 unsigned int slots_found;
1169 unsigned long next_index; /* Index of next search */
1170
1171 if (cur_index > max_index)
1172 break;
1173 slots_found = __lookup_tag(node, results + ret,
1174 cur_index, max_items - ret, &next_index, tag);
1175 ret += slots_found;
1176 if (next_index == 0)
1177 break;
1178 cur_index = next_index;
1179 }
1180
1181 return ret;
1182}
1183EXPORT_SYMBOL(radix_tree_gang_lookup_tag_slot);
1184
1185
1186/**
Nick Piggina5f51c92006-01-08 01:01:41 -08001187 * radix_tree_shrink - shrink height of a radix tree to minimal
1188 * @root radix tree root
1189 */
1190static inline void radix_tree_shrink(struct radix_tree_root *root)
1191{
1192 /* try to shrink tree height */
Nick Pigginc0bc9872007-10-16 01:24:42 -07001193 while (root->height > 0) {
Nick Piggina5f51c92006-01-08 01:01:41 -08001194 struct radix_tree_node *to_free = root->rnode;
Nick Piggin7cf9c2c2006-12-06 20:33:44 -08001195 void *newptr;
Nick Piggina5f51c92006-01-08 01:01:41 -08001196
Nick Pigginc0bc9872007-10-16 01:24:42 -07001197 BUG_ON(!radix_tree_is_indirect_ptr(to_free));
1198 to_free = radix_tree_indirect_to_ptr(to_free);
1199
1200 /*
1201 * The candidate node has more than one child, or its child
1202 * is not at the leftmost slot, we cannot shrink.
1203 */
1204 if (to_free->count != 1)
1205 break;
1206 if (!to_free->slots[0])
1207 break;
1208
Nick Piggin7cf9c2c2006-12-06 20:33:44 -08001209 /*
1210 * We don't need rcu_assign_pointer(), since we are simply
1211 * moving the node from one part of the tree to another. If
1212 * it was safe to dereference the old pointer to it
1213 * (to_free->slots[0]), it will be safe to dereference the new
1214 * one (root->rnode).
1215 */
1216 newptr = to_free->slots[0];
Nick Pigginc0bc9872007-10-16 01:24:42 -07001217 if (root->height > 1)
1218 newptr = radix_tree_ptr_to_indirect(newptr);
Nick Piggin7cf9c2c2006-12-06 20:33:44 -08001219 root->rnode = newptr;
Nick Piggina5f51c92006-01-08 01:01:41 -08001220 root->height--;
Nick Piggina5f51c92006-01-08 01:01:41 -08001221 radix_tree_node_free(to_free);
1222 }
1223}
1224
1225/**
Linus Torvalds1da177e2005-04-16 15:20:36 -07001226 * radix_tree_delete - delete an item from a radix tree
1227 * @root: radix tree root
1228 * @index: index key
1229 *
1230 * Remove the item at @index from the radix tree rooted at @root.
1231 *
1232 * Returns the address of the deleted item, or NULL if it was not present.
1233 */
1234void *radix_tree_delete(struct radix_tree_root *root, unsigned long index)
1235{
Jeff Moyer26fb1582007-10-16 01:24:49 -07001236 /*
1237 * The radix tree path needs to be one longer than the maximum path
1238 * since the "list" is null terminated.
1239 */
1240 struct radix_tree_path path[RADIX_TREE_MAX_PATH + 1], *pathp = path;
Nick Piggin612d6c12006-06-23 02:03:22 -07001241 struct radix_tree_node *slot = NULL;
Nick Piggin7cf9c2c2006-12-06 20:33:44 -08001242 struct radix_tree_node *to_free;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001243 unsigned int height, shift;
Nick Piggind5274262006-01-08 01:01:41 -08001244 int tag;
1245 int offset;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001246
1247 height = root->height;
1248 if (index > radix_tree_maxindex(height))
1249 goto out;
1250
Nick Piggin612d6c12006-06-23 02:03:22 -07001251 slot = root->rnode;
Nick Pigginc0bc9872007-10-16 01:24:42 -07001252 if (height == 0) {
Nick Piggin612d6c12006-06-23 02:03:22 -07001253 root_tag_clear_all(root);
1254 root->rnode = NULL;
1255 goto out;
1256 }
Nick Pigginc0bc9872007-10-16 01:24:42 -07001257 slot = radix_tree_indirect_to_ptr(slot);
Nick Piggin612d6c12006-06-23 02:03:22 -07001258
Linus Torvalds1da177e2005-04-16 15:20:36 -07001259 shift = (height - 1) * RADIX_TREE_MAP_SHIFT;
1260 pathp->node = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001261
Nick Piggin612d6c12006-06-23 02:03:22 -07001262 do {
Christoph Lameter201b6262005-09-06 15:16:46 -07001263 if (slot == NULL)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001264 goto out;
1265
Linus Torvalds1da177e2005-04-16 15:20:36 -07001266 pathp++;
Nick Piggind5274262006-01-08 01:01:41 -08001267 offset = (index >> shift) & RADIX_TREE_MAP_MASK;
1268 pathp->offset = offset;
1269 pathp->node = slot;
1270 slot = slot->slots[offset];
Linus Torvalds1da177e2005-04-16 15:20:36 -07001271 shift -= RADIX_TREE_MAP_SHIFT;
Nick Piggin612d6c12006-06-23 02:03:22 -07001272 height--;
1273 } while (height > 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001274
Nick Piggin612d6c12006-06-23 02:03:22 -07001275 if (slot == NULL)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001276 goto out;
1277
Linus Torvalds1da177e2005-04-16 15:20:36 -07001278 /*
1279 * Clear all tags associated with the just-deleted item
1280 */
Jonathan Corbetdaff89f2006-03-25 03:08:05 -08001281 for (tag = 0; tag < RADIX_TREE_MAX_TAGS; tag++) {
Nick Piggin612d6c12006-06-23 02:03:22 -07001282 if (tag_get(pathp->node, tag, pathp->offset))
1283 radix_tree_tag_clear(root, index, tag);
Nick Piggind5274262006-01-08 01:01:41 -08001284 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001285
Nick Piggin7cf9c2c2006-12-06 20:33:44 -08001286 to_free = NULL;
Christoph Lameter201b6262005-09-06 15:16:46 -07001287 /* Now free the nodes we do not need anymore */
Nick Piggin612d6c12006-06-23 02:03:22 -07001288 while (pathp->node) {
Christoph Lameter201b6262005-09-06 15:16:46 -07001289 pathp->node->slots[pathp->offset] = NULL;
Nick Piggina5f51c92006-01-08 01:01:41 -08001290 pathp->node->count--;
Nick Piggin7cf9c2c2006-12-06 20:33:44 -08001291 /*
1292 * Queue the node for deferred freeing after the
1293 * last reference to it disappears (set NULL, above).
1294 */
1295 if (to_free)
1296 radix_tree_node_free(to_free);
Nick Piggina5f51c92006-01-08 01:01:41 -08001297
1298 if (pathp->node->count) {
Nick Pigginc0bc9872007-10-16 01:24:42 -07001299 if (pathp->node ==
1300 radix_tree_indirect_to_ptr(root->rnode))
Nick Piggina5f51c92006-01-08 01:01:41 -08001301 radix_tree_shrink(root);
Christoph Lameter201b6262005-09-06 15:16:46 -07001302 goto out;
Nick Piggina5f51c92006-01-08 01:01:41 -08001303 }
Christoph Lameter201b6262005-09-06 15:16:46 -07001304
1305 /* Node with zero slots in use so free it */
Nick Piggin7cf9c2c2006-12-06 20:33:44 -08001306 to_free = pathp->node;
Nick Piggin612d6c12006-06-23 02:03:22 -07001307 pathp--;
Nick Piggin7cf9c2c2006-12-06 20:33:44 -08001308
Linus Torvalds1da177e2005-04-16 15:20:36 -07001309 }
Nick Piggin612d6c12006-06-23 02:03:22 -07001310 root_tag_clear_all(root);
Christoph Lameter201b6262005-09-06 15:16:46 -07001311 root->height = 0;
Nick Piggin612d6c12006-06-23 02:03:22 -07001312 root->rnode = NULL;
Nick Piggin7cf9c2c2006-12-06 20:33:44 -08001313 if (to_free)
1314 radix_tree_node_free(to_free);
Nick Piggin612d6c12006-06-23 02:03:22 -07001315
Linus Torvalds1da177e2005-04-16 15:20:36 -07001316out:
Nick Piggin612d6c12006-06-23 02:03:22 -07001317 return slot;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001318}
1319EXPORT_SYMBOL(radix_tree_delete);
1320
1321/**
1322 * radix_tree_tagged - test whether any items in the tree are tagged
1323 * @root: radix tree root
1324 * @tag: tag to test
1325 */
Jonathan Corbetdaff89f2006-03-25 03:08:05 -08001326int radix_tree_tagged(struct radix_tree_root *root, unsigned int tag)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001327{
Nick Piggin612d6c12006-06-23 02:03:22 -07001328 return root_tag_get(root, tag);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001329}
1330EXPORT_SYMBOL(radix_tree_tagged);
1331
1332static void
Alexey Dobriyan51cc5062008-07-25 19:45:34 -07001333radix_tree_node_ctor(void *node)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001334{
1335 memset(node, 0, sizeof(struct radix_tree_node));
1336}
1337
1338static __init unsigned long __maxindex(unsigned int height)
1339{
Peter Lund430d2752007-10-16 23:29:35 -07001340 unsigned int width = height * RADIX_TREE_MAP_SHIFT;
1341 int shift = RADIX_TREE_INDEX_BITS - width;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001342
Peter Lund430d2752007-10-16 23:29:35 -07001343 if (shift < 0)
1344 return ~0UL;
1345 if (shift >= BITS_PER_LONG)
1346 return 0UL;
1347 return ~0UL >> shift;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001348}
1349
1350static __init void radix_tree_init_maxindex(void)
1351{
1352 unsigned int i;
1353
1354 for (i = 0; i < ARRAY_SIZE(height_to_maxindex); i++)
1355 height_to_maxindex[i] = __maxindex(i);
1356}
1357
Linus Torvalds1da177e2005-04-16 15:20:36 -07001358static int radix_tree_callback(struct notifier_block *nfb,
1359 unsigned long action,
1360 void *hcpu)
1361{
1362 int cpu = (long)hcpu;
1363 struct radix_tree_preload *rtp;
1364
1365 /* Free per-cpu pool of perloaded nodes */
Rafael J. Wysocki8bb78442007-05-09 02:35:10 -07001366 if (action == CPU_DEAD || action == CPU_DEAD_FROZEN) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001367 rtp = &per_cpu(radix_tree_preloads, cpu);
1368 while (rtp->nr) {
1369 kmem_cache_free(radix_tree_node_cachep,
1370 rtp->nodes[rtp->nr-1]);
1371 rtp->nodes[rtp->nr-1] = NULL;
1372 rtp->nr--;
1373 }
1374 }
1375 return NOTIFY_OK;
1376}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001377
1378void __init radix_tree_init(void)
1379{
1380 radix_tree_node_cachep = kmem_cache_create("radix_tree_node",
1381 sizeof(struct radix_tree_node), 0,
Christoph Lameter488514d2008-04-28 02:12:05 -07001382 SLAB_PANIC | SLAB_RECLAIM_ACCOUNT,
1383 radix_tree_node_ctor);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001384 radix_tree_init_maxindex();
1385 hotcpu_notifier(radix_tree_callback, 0);
1386}