]>
Commit | Line | Data |
---|---|---|
1da177e4 LT |
1 | /* |
2 | * Copyright (C) 2001 Momchil Velikov | |
3 | * Portions Copyright (C) 2001 Christoph Hellwig | |
cde53535 | 4 | * Copyright (C) 2005 SGI, Christoph Lameter |
7cf9c2c7 | 5 | * Copyright (C) 2006 Nick Piggin |
1da177e4 LT |
6 | * |
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> | |
31 | #include <linux/gfp.h> | |
32 | #include <linux/string.h> | |
33 | #include <linux/bitops.h> | |
7cf9c2c7 | 34 | #include <linux/rcupdate.h> |
1da177e4 LT |
35 | |
36 | ||
37 | #ifdef __KERNEL__ | |
cfd9b7df | 38 | #define RADIX_TREE_MAP_SHIFT (CONFIG_BASE_SMALL ? 4 : 6) |
1da177e4 LT |
39 | #else |
40 | #define RADIX_TREE_MAP_SHIFT 3 /* For more stressful testing */ | |
41 | #endif | |
1da177e4 LT |
42 | |
43 | #define RADIX_TREE_MAP_SIZE (1UL << RADIX_TREE_MAP_SHIFT) | |
44 | #define RADIX_TREE_MAP_MASK (RADIX_TREE_MAP_SIZE-1) | |
45 | ||
46 | #define RADIX_TREE_TAG_LONGS \ | |
47 | ((RADIX_TREE_MAP_SIZE + BITS_PER_LONG - 1) / BITS_PER_LONG) | |
48 | ||
49 | struct radix_tree_node { | |
7cf9c2c7 | 50 | unsigned int height; /* Height from the bottom */ |
1da177e4 | 51 | unsigned int count; |
7cf9c2c7 | 52 | struct rcu_head rcu_head; |
1da177e4 | 53 | void *slots[RADIX_TREE_MAP_SIZE]; |
daff89f3 | 54 | unsigned long tags[RADIX_TREE_MAX_TAGS][RADIX_TREE_TAG_LONGS]; |
1da177e4 LT |
55 | }; |
56 | ||
57 | struct radix_tree_path { | |
201b6264 | 58 | struct radix_tree_node *node; |
1da177e4 LT |
59 | int offset; |
60 | }; | |
61 | ||
62 | #define RADIX_TREE_INDEX_BITS (8 /* CHAR_BIT */ * sizeof(unsigned long)) | |
26fb1589 JM |
63 | #define RADIX_TREE_MAX_PATH (DIV_ROUND_UP(RADIX_TREE_INDEX_BITS, \ |
64 | RADIX_TREE_MAP_SHIFT)) | |
1da177e4 | 65 | |
26fb1589 JM |
66 | /* |
67 | * The height_to_maxindex array needs to be one deeper than the maximum | |
68 | * path as height 0 holds only 1 entry. | |
69 | */ | |
70 | static unsigned long height_to_maxindex[RADIX_TREE_MAX_PATH + 1] __read_mostly; | |
1da177e4 LT |
71 | |
72 | /* | |
73 | * Radix tree node cache. | |
74 | */ | |
e18b890b | 75 | static struct kmem_cache *radix_tree_node_cachep; |
1da177e4 LT |
76 | |
77 | /* | |
78 | * Per-cpu pool of preloaded nodes | |
79 | */ | |
80 | struct radix_tree_preload { | |
81 | int nr; | |
82 | struct radix_tree_node *nodes[RADIX_TREE_MAX_PATH]; | |
83 | }; | |
8cef7d57 | 84 | static DEFINE_PER_CPU(struct radix_tree_preload, radix_tree_preloads) = { 0, }; |
1da177e4 | 85 | |
612d6c19 NP |
86 | static inline gfp_t root_gfp_mask(struct radix_tree_root *root) |
87 | { | |
88 | return root->gfp_mask & __GFP_BITS_MASK; | |
89 | } | |
90 | ||
643b52b9 NP |
91 | static inline void tag_set(struct radix_tree_node *node, unsigned int tag, |
92 | int offset) | |
93 | { | |
94 | __set_bit(offset, node->tags[tag]); | |
95 | } | |
96 | ||
97 | static inline void tag_clear(struct radix_tree_node *node, unsigned int tag, | |
98 | int offset) | |
99 | { | |
100 | __clear_bit(offset, node->tags[tag]); | |
101 | } | |
102 | ||
103 | static inline int tag_get(struct radix_tree_node *node, unsigned int tag, | |
104 | int offset) | |
105 | { | |
106 | return test_bit(offset, node->tags[tag]); | |
107 | } | |
108 | ||
109 | static inline void root_tag_set(struct radix_tree_root *root, unsigned int tag) | |
110 | { | |
111 | root->gfp_mask |= (__force gfp_t)(1 << (tag + __GFP_BITS_SHIFT)); | |
112 | } | |
113 | ||
114 | static inline void root_tag_clear(struct radix_tree_root *root, unsigned int tag) | |
115 | { | |
116 | root->gfp_mask &= (__force gfp_t)~(1 << (tag + __GFP_BITS_SHIFT)); | |
117 | } | |
118 | ||
119 | static inline void root_tag_clear_all(struct radix_tree_root *root) | |
120 | { | |
121 | root->gfp_mask &= __GFP_BITS_MASK; | |
122 | } | |
123 | ||
124 | static inline int root_tag_get(struct radix_tree_root *root, unsigned int tag) | |
125 | { | |
126 | return (__force unsigned)root->gfp_mask & (1 << (tag + __GFP_BITS_SHIFT)); | |
127 | } | |
128 | ||
129 | /* | |
130 | * Returns 1 if any slot in the node has this tag set. | |
131 | * Otherwise returns 0. | |
132 | */ | |
133 | static inline int any_tag_set(struct radix_tree_node *node, unsigned int tag) | |
134 | { | |
135 | int idx; | |
136 | for (idx = 0; idx < RADIX_TREE_TAG_LONGS; idx++) { | |
137 | if (node->tags[tag][idx]) | |
138 | return 1; | |
139 | } | |
140 | return 0; | |
141 | } | |
1da177e4 LT |
142 | /* |
143 | * This assumes that the caller has performed appropriate preallocation, and | |
144 | * that the caller has pinned this thread of control to the current CPU. | |
145 | */ | |
146 | static struct radix_tree_node * | |
147 | radix_tree_node_alloc(struct radix_tree_root *root) | |
148 | { | |
e2848a0e | 149 | struct radix_tree_node *ret = NULL; |
612d6c19 | 150 | gfp_t gfp_mask = root_gfp_mask(root); |
1da177e4 | 151 | |
e2848a0e | 152 | if (!(gfp_mask & __GFP_WAIT)) { |
1da177e4 LT |
153 | struct radix_tree_preload *rtp; |
154 | ||
e2848a0e NP |
155 | /* |
156 | * Provided the caller has preloaded here, we will always | |
157 | * succeed in getting a node here (and never reach | |
158 | * kmem_cache_alloc) | |
159 | */ | |
1da177e4 LT |
160 | rtp = &__get_cpu_var(radix_tree_preloads); |
161 | if (rtp->nr) { | |
162 | ret = rtp->nodes[rtp->nr - 1]; | |
163 | rtp->nodes[rtp->nr - 1] = NULL; | |
164 | rtp->nr--; | |
165 | } | |
166 | } | |
e2848a0e | 167 | if (ret == NULL) |
488514d1 | 168 | ret = kmem_cache_alloc(radix_tree_node_cachep, gfp_mask); |
e2848a0e | 169 | |
c0bc9875 | 170 | BUG_ON(radix_tree_is_indirect_ptr(ret)); |
1da177e4 LT |
171 | return ret; |
172 | } | |
173 | ||
7cf9c2c7 NP |
174 | static void radix_tree_node_rcu_free(struct rcu_head *head) |
175 | { | |
176 | struct radix_tree_node *node = | |
177 | container_of(head, struct radix_tree_node, rcu_head); | |
643b52b9 NP |
178 | |
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 | */ | |
184 | tag_clear(node, 0, 0); | |
185 | tag_clear(node, 1, 0); | |
186 | node->slots[0] = NULL; | |
187 | node->count = 0; | |
188 | ||
7cf9c2c7 NP |
189 | kmem_cache_free(radix_tree_node_cachep, node); |
190 | } | |
191 | ||
1da177e4 LT |
192 | static inline void |
193 | radix_tree_node_free(struct radix_tree_node *node) | |
194 | { | |
7cf9c2c7 | 195 | call_rcu(&node->rcu_head, radix_tree_node_rcu_free); |
1da177e4 LT |
196 | } |
197 | ||
198 | /* | |
199 | * Load up this CPU's radix_tree_node buffer with sufficient objects to | |
200 | * ensure that the addition of a single element in the tree cannot fail. On | |
201 | * success, return zero, with preemption disabled. On error, return -ENOMEM | |
202 | * with preemption not disabled. | |
203 | */ | |
dd0fc66f | 204 | int radix_tree_preload(gfp_t gfp_mask) |
1da177e4 LT |
205 | { |
206 | struct radix_tree_preload *rtp; | |
207 | struct radix_tree_node *node; | |
208 | int ret = -ENOMEM; | |
209 | ||
210 | preempt_disable(); | |
211 | rtp = &__get_cpu_var(radix_tree_preloads); | |
212 | while (rtp->nr < ARRAY_SIZE(rtp->nodes)) { | |
213 | preempt_enable(); | |
488514d1 | 214 | node = kmem_cache_alloc(radix_tree_node_cachep, gfp_mask); |
1da177e4 LT |
215 | if (node == NULL) |
216 | goto out; | |
217 | preempt_disable(); | |
218 | rtp = &__get_cpu_var(radix_tree_preloads); | |
219 | if (rtp->nr < ARRAY_SIZE(rtp->nodes)) | |
220 | rtp->nodes[rtp->nr++] = node; | |
221 | else | |
222 | kmem_cache_free(radix_tree_node_cachep, node); | |
223 | } | |
224 | ret = 0; | |
225 | out: | |
226 | return ret; | |
227 | } | |
d7f0923d | 228 | EXPORT_SYMBOL(radix_tree_preload); |
1da177e4 | 229 | |
1da177e4 LT |
230 | /* |
231 | * Return the maximum key which can be store into a | |
232 | * radix tree with height HEIGHT. | |
233 | */ | |
234 | static inline unsigned long radix_tree_maxindex(unsigned int height) | |
235 | { | |
236 | return height_to_maxindex[height]; | |
237 | } | |
238 | ||
239 | /* | |
240 | * Extend a radix tree so it can store key @index. | |
241 | */ | |
242 | static int radix_tree_extend(struct radix_tree_root *root, unsigned long index) | |
243 | { | |
244 | struct radix_tree_node *node; | |
245 | unsigned int height; | |
1da177e4 LT |
246 | int tag; |
247 | ||
248 | /* Figure out what the height should be. */ | |
249 | height = root->height + 1; | |
250 | while (index > radix_tree_maxindex(height)) | |
251 | height++; | |
252 | ||
253 | if (root->rnode == NULL) { | |
254 | root->height = height; | |
255 | goto out; | |
256 | } | |
257 | ||
1da177e4 | 258 | do { |
7cf9c2c7 | 259 | unsigned int newheight; |
1da177e4 LT |
260 | if (!(node = radix_tree_node_alloc(root))) |
261 | return -ENOMEM; | |
262 | ||
263 | /* Increase the height. */ | |
c0bc9875 | 264 | node->slots[0] = radix_tree_indirect_to_ptr(root->rnode); |
1da177e4 LT |
265 | |
266 | /* Propagate the aggregated tag info into the new root */ | |
daff89f3 | 267 | for (tag = 0; tag < RADIX_TREE_MAX_TAGS; tag++) { |
612d6c19 | 268 | if (root_tag_get(root, tag)) |
1da177e4 LT |
269 | tag_set(node, tag, 0); |
270 | } | |
271 | ||
7cf9c2c7 NP |
272 | newheight = root->height+1; |
273 | node->height = newheight; | |
1da177e4 | 274 | node->count = 1; |
c0bc9875 | 275 | node = radix_tree_ptr_to_indirect(node); |
7cf9c2c7 NP |
276 | rcu_assign_pointer(root->rnode, node); |
277 | root->height = newheight; | |
1da177e4 LT |
278 | } while (height > root->height); |
279 | out: | |
280 | return 0; | |
281 | } | |
282 | ||
283 | /** | |
284 | * radix_tree_insert - insert into a radix tree | |
285 | * @root: radix tree root | |
286 | * @index: index key | |
287 | * @item: item to insert | |
288 | * | |
289 | * Insert an item into the radix tree at position @index. | |
290 | */ | |
291 | int radix_tree_insert(struct radix_tree_root *root, | |
292 | unsigned long index, void *item) | |
293 | { | |
201b6264 | 294 | struct radix_tree_node *node = NULL, *slot; |
1da177e4 LT |
295 | unsigned int height, shift; |
296 | int offset; | |
297 | int error; | |
298 | ||
c0bc9875 | 299 | BUG_ON(radix_tree_is_indirect_ptr(item)); |
7cf9c2c7 | 300 | |
1da177e4 | 301 | /* Make sure the tree is high enough. */ |
612d6c19 | 302 | if (index > radix_tree_maxindex(root->height)) { |
1da177e4 LT |
303 | error = radix_tree_extend(root, index); |
304 | if (error) | |
305 | return error; | |
306 | } | |
307 | ||
c0bc9875 NP |
308 | slot = radix_tree_indirect_to_ptr(root->rnode); |
309 | ||
1da177e4 LT |
310 | height = root->height; |
311 | shift = (height-1) * RADIX_TREE_MAP_SHIFT; | |
312 | ||
313 | offset = 0; /* uninitialised var warning */ | |
612d6c19 | 314 | while (height > 0) { |
201b6264 | 315 | if (slot == NULL) { |
1da177e4 | 316 | /* Have to add a child node. */ |
201b6264 | 317 | if (!(slot = radix_tree_node_alloc(root))) |
1da177e4 | 318 | return -ENOMEM; |
7cf9c2c7 | 319 | slot->height = height; |
201b6264 | 320 | if (node) { |
7cf9c2c7 | 321 | rcu_assign_pointer(node->slots[offset], slot); |
1da177e4 | 322 | node->count++; |
201b6264 | 323 | } else |
c0bc9875 NP |
324 | rcu_assign_pointer(root->rnode, |
325 | radix_tree_ptr_to_indirect(slot)); | |
1da177e4 LT |
326 | } |
327 | ||
328 | /* Go a level down */ | |
329 | offset = (index >> shift) & RADIX_TREE_MAP_MASK; | |
201b6264 CL |
330 | node = slot; |
331 | slot = node->slots[offset]; | |
1da177e4 LT |
332 | shift -= RADIX_TREE_MAP_SHIFT; |
333 | height--; | |
612d6c19 | 334 | } |
1da177e4 | 335 | |
201b6264 | 336 | if (slot != NULL) |
1da177e4 | 337 | return -EEXIST; |
201b6264 | 338 | |
612d6c19 NP |
339 | if (node) { |
340 | node->count++; | |
7cf9c2c7 | 341 | rcu_assign_pointer(node->slots[offset], item); |
612d6c19 NP |
342 | BUG_ON(tag_get(node, 0, offset)); |
343 | BUG_ON(tag_get(node, 1, offset)); | |
344 | } else { | |
c0bc9875 | 345 | rcu_assign_pointer(root->rnode, item); |
612d6c19 NP |
346 | BUG_ON(root_tag_get(root, 0)); |
347 | BUG_ON(root_tag_get(root, 1)); | |
348 | } | |
1da177e4 | 349 | |
1da177e4 LT |
350 | return 0; |
351 | } | |
352 | EXPORT_SYMBOL(radix_tree_insert); | |
353 | ||
7cf9c2c7 NP |
354 | /** |
355 | * radix_tree_lookup_slot - lookup a slot in a radix tree | |
356 | * @root: radix tree root | |
357 | * @index: index key | |
358 | * | |
359 | * Returns: the slot corresponding to the position @index in the | |
360 | * radix tree @root. This is useful for update-if-exists operations. | |
361 | * | |
47feff2c NP |
362 | * This function can be called under rcu_read_lock iff the slot is not |
363 | * modified by radix_tree_replace_slot, otherwise it must be called | |
364 | * exclusive from other writers. Any dereference of the slot must be done | |
365 | * using radix_tree_deref_slot. | |
7cf9c2c7 NP |
366 | */ |
367 | void **radix_tree_lookup_slot(struct radix_tree_root *root, unsigned long index) | |
1da177e4 LT |
368 | { |
369 | unsigned int height, shift; | |
7cf9c2c7 | 370 | struct radix_tree_node *node, **slot; |
612d6c19 | 371 | |
47feff2c | 372 | node = rcu_dereference(root->rnode); |
7cf9c2c7 | 373 | if (node == NULL) |
1da177e4 LT |
374 | return NULL; |
375 | ||
c0bc9875 | 376 | if (!radix_tree_is_indirect_ptr(node)) { |
7cf9c2c7 NP |
377 | if (index > 0) |
378 | return NULL; | |
612d6c19 | 379 | return (void **)&root->rnode; |
7cf9c2c7 | 380 | } |
c0bc9875 | 381 | node = radix_tree_indirect_to_ptr(node); |
7cf9c2c7 NP |
382 | |
383 | height = node->height; | |
384 | if (index > radix_tree_maxindex(height)) | |
385 | return NULL; | |
612d6c19 | 386 | |
1da177e4 | 387 | shift = (height-1) * RADIX_TREE_MAP_SHIFT; |
1da177e4 | 388 | |
7cf9c2c7 NP |
389 | do { |
390 | slot = (struct radix_tree_node **) | |
391 | (node->slots + ((index>>shift) & RADIX_TREE_MAP_MASK)); | |
47feff2c | 392 | node = rcu_dereference(*slot); |
7cf9c2c7 | 393 | if (node == NULL) |
1da177e4 LT |
394 | return NULL; |
395 | ||
1da177e4 LT |
396 | shift -= RADIX_TREE_MAP_SHIFT; |
397 | height--; | |
7cf9c2c7 | 398 | } while (height > 0); |
1da177e4 | 399 | |
a4331366 HR |
400 | return (void **)slot; |
401 | } | |
a4331366 HR |
402 | EXPORT_SYMBOL(radix_tree_lookup_slot); |
403 | ||
404 | /** | |
405 | * radix_tree_lookup - perform lookup operation on a radix tree | |
406 | * @root: radix tree root | |
407 | * @index: index key | |
408 | * | |
409 | * Lookup the item at the position @index in the radix tree @root. | |
7cf9c2c7 NP |
410 | * |
411 | * This function can be called under rcu_read_lock, however the caller | |
412 | * must manage lifetimes of leaf nodes (eg. RCU may also be used to free | |
413 | * them safely). No RCU barriers are required to access or modify the | |
414 | * returned item, however. | |
a4331366 HR |
415 | */ |
416 | void *radix_tree_lookup(struct radix_tree_root *root, unsigned long index) | |
417 | { | |
7cf9c2c7 NP |
418 | unsigned int height, shift; |
419 | struct radix_tree_node *node, **slot; | |
420 | ||
421 | node = rcu_dereference(root->rnode); | |
422 | if (node == NULL) | |
423 | return NULL; | |
424 | ||
c0bc9875 | 425 | if (!radix_tree_is_indirect_ptr(node)) { |
7cf9c2c7 NP |
426 | if (index > 0) |
427 | return NULL; | |
c0bc9875 | 428 | return node; |
7cf9c2c7 | 429 | } |
c0bc9875 | 430 | node = radix_tree_indirect_to_ptr(node); |
7cf9c2c7 NP |
431 | |
432 | height = node->height; | |
433 | if (index > radix_tree_maxindex(height)) | |
434 | return NULL; | |
435 | ||
436 | shift = (height-1) * RADIX_TREE_MAP_SHIFT; | |
a4331366 | 437 | |
7cf9c2c7 NP |
438 | do { |
439 | slot = (struct radix_tree_node **) | |
440 | (node->slots + ((index>>shift) & RADIX_TREE_MAP_MASK)); | |
441 | node = rcu_dereference(*slot); | |
442 | if (node == NULL) | |
443 | return NULL; | |
444 | ||
445 | shift -= RADIX_TREE_MAP_SHIFT; | |
446 | height--; | |
447 | } while (height > 0); | |
448 | ||
449 | return node; | |
1da177e4 LT |
450 | } |
451 | EXPORT_SYMBOL(radix_tree_lookup); | |
452 | ||
453 | /** | |
454 | * radix_tree_tag_set - set a tag on a radix tree node | |
455 | * @root: radix tree root | |
456 | * @index: index key | |
457 | * @tag: tag index | |
458 | * | |
daff89f3 JC |
459 | * Set the search tag (which must be < RADIX_TREE_MAX_TAGS) |
460 | * corresponding to @index in the radix tree. From | |
1da177e4 LT |
461 | * the root all the way down to the leaf node. |
462 | * | |
463 | * Returns the address of the tagged item. Setting a tag on a not-present | |
464 | * item is a bug. | |
465 | */ | |
466 | void *radix_tree_tag_set(struct radix_tree_root *root, | |
daff89f3 | 467 | unsigned long index, unsigned int tag) |
1da177e4 LT |
468 | { |
469 | unsigned int height, shift; | |
201b6264 | 470 | struct radix_tree_node *slot; |
1da177e4 LT |
471 | |
472 | height = root->height; | |
4c91c364 | 473 | BUG_ON(index > radix_tree_maxindex(height)); |
1da177e4 | 474 | |
c0bc9875 | 475 | slot = radix_tree_indirect_to_ptr(root->rnode); |
612d6c19 | 476 | shift = (height - 1) * RADIX_TREE_MAP_SHIFT; |
1da177e4 LT |
477 | |
478 | while (height > 0) { | |
479 | int offset; | |
480 | ||
481 | offset = (index >> shift) & RADIX_TREE_MAP_MASK; | |
d5274261 NP |
482 | if (!tag_get(slot, tag, offset)) |
483 | tag_set(slot, tag, offset); | |
201b6264 CL |
484 | slot = slot->slots[offset]; |
485 | BUG_ON(slot == NULL); | |
1da177e4 LT |
486 | shift -= RADIX_TREE_MAP_SHIFT; |
487 | height--; | |
488 | } | |
489 | ||
612d6c19 NP |
490 | /* set the root's tag bit */ |
491 | if (slot && !root_tag_get(root, tag)) | |
492 | root_tag_set(root, tag); | |
493 | ||
201b6264 | 494 | return slot; |
1da177e4 LT |
495 | } |
496 | EXPORT_SYMBOL(radix_tree_tag_set); | |
497 | ||
498 | /** | |
499 | * radix_tree_tag_clear - clear a tag on a radix tree node | |
500 | * @root: radix tree root | |
501 | * @index: index key | |
502 | * @tag: tag index | |
503 | * | |
daff89f3 JC |
504 | * Clear the search tag (which must be < RADIX_TREE_MAX_TAGS) |
505 | * corresponding to @index in the radix tree. If | |
1da177e4 LT |
506 | * this causes the leaf node to have no tags set then clear the tag in the |
507 | * next-to-leaf node, etc. | |
508 | * | |
509 | * Returns the address of the tagged item on success, else NULL. ie: | |
510 | * has the same return value and semantics as radix_tree_lookup(). | |
511 | */ | |
512 | void *radix_tree_tag_clear(struct radix_tree_root *root, | |
daff89f3 | 513 | unsigned long index, unsigned int tag) |
1da177e4 | 514 | { |
26fb1589 JM |
515 | /* |
516 | * The radix tree path needs to be one longer than the maximum path | |
517 | * since the "list" is null terminated. | |
518 | */ | |
519 | struct radix_tree_path path[RADIX_TREE_MAX_PATH + 1], *pathp = path; | |
612d6c19 | 520 | struct radix_tree_node *slot = NULL; |
1da177e4 | 521 | unsigned int height, shift; |
1da177e4 LT |
522 | |
523 | height = root->height; | |
524 | if (index > radix_tree_maxindex(height)) | |
525 | goto out; | |
526 | ||
527 | shift = (height - 1) * RADIX_TREE_MAP_SHIFT; | |
528 | pathp->node = NULL; | |
c0bc9875 | 529 | slot = radix_tree_indirect_to_ptr(root->rnode); |
1da177e4 LT |
530 | |
531 | while (height > 0) { | |
532 | int offset; | |
533 | ||
201b6264 | 534 | if (slot == NULL) |
1da177e4 LT |
535 | goto out; |
536 | ||
537 | offset = (index >> shift) & RADIX_TREE_MAP_MASK; | |
538 | pathp[1].offset = offset; | |
201b6264 CL |
539 | pathp[1].node = slot; |
540 | slot = slot->slots[offset]; | |
1da177e4 LT |
541 | pathp++; |
542 | shift -= RADIX_TREE_MAP_SHIFT; | |
543 | height--; | |
544 | } | |
545 | ||
612d6c19 | 546 | if (slot == NULL) |
1da177e4 LT |
547 | goto out; |
548 | ||
612d6c19 | 549 | while (pathp->node) { |
d5274261 NP |
550 | if (!tag_get(pathp->node, tag, pathp->offset)) |
551 | goto out; | |
201b6264 | 552 | tag_clear(pathp->node, tag, pathp->offset); |
6e954b9e NP |
553 | if (any_tag_set(pathp->node, tag)) |
554 | goto out; | |
1da177e4 | 555 | pathp--; |
612d6c19 NP |
556 | } |
557 | ||
558 | /* clear the root's tag bit */ | |
559 | if (root_tag_get(root, tag)) | |
560 | root_tag_clear(root, tag); | |
561 | ||
1da177e4 | 562 | out: |
612d6c19 | 563 | return slot; |
1da177e4 LT |
564 | } |
565 | EXPORT_SYMBOL(radix_tree_tag_clear); | |
566 | ||
567 | #ifndef __KERNEL__ /* Only the test harness uses this at present */ | |
568 | /** | |
32605a18 MT |
569 | * radix_tree_tag_get - get a tag on a radix tree node |
570 | * @root: radix tree root | |
571 | * @index: index key | |
daff89f3 | 572 | * @tag: tag index (< RADIX_TREE_MAX_TAGS) |
1da177e4 | 573 | * |
32605a18 | 574 | * Return values: |
1da177e4 | 575 | * |
612d6c19 NP |
576 | * 0: tag not present or not set |
577 | * 1: tag set | |
1da177e4 LT |
578 | */ |
579 | int radix_tree_tag_get(struct radix_tree_root *root, | |
daff89f3 | 580 | unsigned long index, unsigned int tag) |
1da177e4 LT |
581 | { |
582 | unsigned int height, shift; | |
7cf9c2c7 | 583 | struct radix_tree_node *node; |
1da177e4 LT |
584 | int saw_unset_tag = 0; |
585 | ||
612d6c19 NP |
586 | /* check the root's tag bit */ |
587 | if (!root_tag_get(root, tag)) | |
588 | return 0; | |
589 | ||
7cf9c2c7 NP |
590 | node = rcu_dereference(root->rnode); |
591 | if (node == NULL) | |
592 | return 0; | |
593 | ||
c0bc9875 | 594 | if (!radix_tree_is_indirect_ptr(node)) |
7cf9c2c7 | 595 | return (index == 0); |
c0bc9875 | 596 | node = radix_tree_indirect_to_ptr(node); |
7cf9c2c7 NP |
597 | |
598 | height = node->height; | |
599 | if (index > radix_tree_maxindex(height)) | |
600 | return 0; | |
612d6c19 | 601 | |
1da177e4 | 602 | shift = (height - 1) * RADIX_TREE_MAP_SHIFT; |
1da177e4 LT |
603 | |
604 | for ( ; ; ) { | |
605 | int offset; | |
606 | ||
7cf9c2c7 | 607 | if (node == NULL) |
1da177e4 LT |
608 | return 0; |
609 | ||
610 | offset = (index >> shift) & RADIX_TREE_MAP_MASK; | |
611 | ||
612 | /* | |
613 | * This is just a debug check. Later, we can bale as soon as | |
614 | * we see an unset tag. | |
615 | */ | |
7cf9c2c7 | 616 | if (!tag_get(node, tag, offset)) |
1da177e4 LT |
617 | saw_unset_tag = 1; |
618 | if (height == 1) { | |
7cf9c2c7 | 619 | int ret = tag_get(node, tag, offset); |
1da177e4 LT |
620 | |
621 | BUG_ON(ret && saw_unset_tag); | |
e5dcd90b | 622 | return !!ret; |
1da177e4 | 623 | } |
7cf9c2c7 | 624 | node = rcu_dereference(node->slots[offset]); |
1da177e4 LT |
625 | shift -= RADIX_TREE_MAP_SHIFT; |
626 | height--; | |
627 | } | |
628 | } | |
629 | EXPORT_SYMBOL(radix_tree_tag_get); | |
630 | #endif | |
631 | ||
6df8ba4f FW |
632 | /** |
633 | * radix_tree_next_hole - find the next hole (not-present entry) | |
634 | * @root: tree root | |
635 | * @index: index key | |
636 | * @max_scan: maximum range to search | |
637 | * | |
638 | * Search the set [index, min(index+max_scan-1, MAX_INDEX)] for the lowest | |
639 | * indexed hole. | |
640 | * | |
641 | * Returns: the index of the hole if found, otherwise returns an index | |
642 | * outside of the set specified (in which case 'return - index >= max_scan' | |
8e6bdb7f | 643 | * will be true). In rare cases of index wrap-around, 0 will be returned. |
6df8ba4f FW |
644 | * |
645 | * radix_tree_next_hole may be called under rcu_read_lock. However, like | |
8e6bdb7f WF |
646 | * radix_tree_gang_lookup, this will not atomically search a snapshot of |
647 | * the tree at a single point in time. For example, if a hole is created | |
648 | * at index 5, then subsequently a hole is created at index 10, | |
649 | * radix_tree_next_hole covering both indexes may return 10 if called | |
650 | * under rcu_read_lock. | |
6df8ba4f FW |
651 | */ |
652 | unsigned long radix_tree_next_hole(struct radix_tree_root *root, | |
653 | unsigned long index, unsigned long max_scan) | |
654 | { | |
655 | unsigned long i; | |
656 | ||
657 | for (i = 0; i < max_scan; i++) { | |
658 | if (!radix_tree_lookup(root, index)) | |
659 | break; | |
660 | index++; | |
661 | if (index == 0) | |
662 | break; | |
663 | } | |
664 | ||
665 | return index; | |
666 | } | |
667 | EXPORT_SYMBOL(radix_tree_next_hole); | |
668 | ||
dc566127 WF |
669 | /** |
670 | * radix_tree_prev_hole - find the prev hole (not-present entry) | |
671 | * @root: tree root | |
672 | * @index: index key | |
673 | * @max_scan: maximum range to search | |
674 | * | |
675 | * Search backwards in the range [max(index-max_scan+1, 0), index] | |
676 | * for the first hole. | |
677 | * | |
678 | * Returns: the index of the hole if found, otherwise returns an index | |
679 | * outside of the set specified (in which case 'index - return >= max_scan' | |
680 | * will be true). In rare cases of wrap-around, LONG_MAX will be returned. | |
681 | * | |
682 | * radix_tree_next_hole may be called under rcu_read_lock. However, like | |
683 | * radix_tree_gang_lookup, this will not atomically search a snapshot of | |
684 | * the tree at a single point in time. For example, if a hole is created | |
685 | * at index 10, then subsequently a hole is created at index 5, | |
686 | * radix_tree_prev_hole covering both indexes may return 5 if called under | |
687 | * rcu_read_lock. | |
688 | */ | |
689 | unsigned long radix_tree_prev_hole(struct radix_tree_root *root, | |
690 | unsigned long index, unsigned long max_scan) | |
691 | { | |
692 | unsigned long i; | |
693 | ||
694 | for (i = 0; i < max_scan; i++) { | |
695 | if (!radix_tree_lookup(root, index)) | |
696 | break; | |
697 | index--; | |
698 | if (index == LONG_MAX) | |
699 | break; | |
700 | } | |
701 | ||
702 | return index; | |
703 | } | |
704 | EXPORT_SYMBOL(radix_tree_prev_hole); | |
705 | ||
1da177e4 | 706 | static unsigned int |
47feff2c | 707 | __lookup(struct radix_tree_node *slot, void ***results, unsigned long index, |
1da177e4 LT |
708 | unsigned int max_items, unsigned long *next_index) |
709 | { | |
710 | unsigned int nr_found = 0; | |
201b6264 | 711 | unsigned int shift, height; |
201b6264 CL |
712 | unsigned long i; |
713 | ||
7cf9c2c7 NP |
714 | height = slot->height; |
715 | if (height == 0) | |
201b6264 | 716 | goto out; |
1da177e4 | 717 | shift = (height-1) * RADIX_TREE_MAP_SHIFT; |
1da177e4 | 718 | |
201b6264 | 719 | for ( ; height > 1; height--) { |
7cf9c2c7 NP |
720 | i = (index >> shift) & RADIX_TREE_MAP_MASK; |
721 | for (;;) { | |
1da177e4 LT |
722 | if (slot->slots[i] != NULL) |
723 | break; | |
724 | index &= ~((1UL << shift) - 1); | |
725 | index += 1UL << shift; | |
726 | if (index == 0) | |
727 | goto out; /* 32-bit wraparound */ | |
7cf9c2c7 NP |
728 | i++; |
729 | if (i == RADIX_TREE_MAP_SIZE) | |
730 | goto out; | |
1da177e4 | 731 | } |
1da177e4 | 732 | |
1da177e4 | 733 | shift -= RADIX_TREE_MAP_SHIFT; |
7cf9c2c7 NP |
734 | slot = rcu_dereference(slot->slots[i]); |
735 | if (slot == NULL) | |
736 | goto out; | |
1da177e4 | 737 | } |
201b6264 CL |
738 | |
739 | /* Bottom level: grab some items */ | |
740 | for (i = index & RADIX_TREE_MAP_MASK; i < RADIX_TREE_MAP_SIZE; i++) { | |
741 | index++; | |
47feff2c NP |
742 | if (slot->slots[i]) { |
743 | results[nr_found++] = &(slot->slots[i]); | |
201b6264 CL |
744 | if (nr_found == max_items) |
745 | goto out; | |
746 | } | |
747 | } | |
1da177e4 LT |
748 | out: |
749 | *next_index = index; | |
750 | return nr_found; | |
751 | } | |
752 | ||
753 | /** | |
754 | * radix_tree_gang_lookup - perform multiple lookup on a radix tree | |
755 | * @root: radix tree root | |
756 | * @results: where the results of the lookup are placed | |
757 | * @first_index: start the lookup from this key | |
758 | * @max_items: place up to this many items at *results | |
759 | * | |
760 | * Performs an index-ascending scan of the tree for present items. Places | |
761 | * them at *@results and returns the number of items which were placed at | |
762 | * *@results. | |
763 | * | |
764 | * The implementation is naive. | |
7cf9c2c7 NP |
765 | * |
766 | * Like radix_tree_lookup, radix_tree_gang_lookup may be called under | |
767 | * rcu_read_lock. In this case, rather than the returned results being | |
768 | * an atomic snapshot of the tree at a single point in time, the semantics | |
769 | * of an RCU protected gang lookup are as though multiple radix_tree_lookups | |
770 | * have been issued in individual locks, and results stored in 'results'. | |
1da177e4 LT |
771 | */ |
772 | unsigned int | |
773 | radix_tree_gang_lookup(struct radix_tree_root *root, void **results, | |
774 | unsigned long first_index, unsigned int max_items) | |
775 | { | |
7cf9c2c7 NP |
776 | unsigned long max_index; |
777 | struct radix_tree_node *node; | |
1da177e4 | 778 | unsigned long cur_index = first_index; |
7cf9c2c7 NP |
779 | unsigned int ret; |
780 | ||
781 | node = rcu_dereference(root->rnode); | |
782 | if (!node) | |
783 | return 0; | |
1da177e4 | 784 | |
c0bc9875 | 785 | if (!radix_tree_is_indirect_ptr(node)) { |
7cf9c2c7 NP |
786 | if (first_index > 0) |
787 | return 0; | |
c0bc9875 | 788 | results[0] = node; |
7cf9c2c7 NP |
789 | return 1; |
790 | } | |
c0bc9875 | 791 | node = radix_tree_indirect_to_ptr(node); |
7cf9c2c7 NP |
792 | |
793 | max_index = radix_tree_maxindex(node->height); | |
794 | ||
795 | ret = 0; | |
1da177e4 | 796 | while (ret < max_items) { |
47feff2c | 797 | unsigned int nr_found, slots_found, i; |
1da177e4 LT |
798 | unsigned long next_index; /* Index of next search */ |
799 | ||
800 | if (cur_index > max_index) | |
801 | break; | |
47feff2c | 802 | slots_found = __lookup(node, (void ***)results + ret, cur_index, |
1da177e4 | 803 | max_items - ret, &next_index); |
47feff2c NP |
804 | nr_found = 0; |
805 | for (i = 0; i < slots_found; i++) { | |
806 | struct radix_tree_node *slot; | |
807 | slot = *(((void ***)results)[ret + i]); | |
808 | if (!slot) | |
809 | continue; | |
810 | results[ret + nr_found] = rcu_dereference(slot); | |
811 | nr_found++; | |
812 | } | |
1da177e4 LT |
813 | ret += nr_found; |
814 | if (next_index == 0) | |
815 | break; | |
816 | cur_index = next_index; | |
817 | } | |
7cf9c2c7 | 818 | |
1da177e4 LT |
819 | return ret; |
820 | } | |
821 | EXPORT_SYMBOL(radix_tree_gang_lookup); | |
822 | ||
47feff2c NP |
823 | /** |
824 | * radix_tree_gang_lookup_slot - perform multiple slot lookup on radix tree | |
825 | * @root: radix tree root | |
826 | * @results: where the results of the lookup are placed | |
827 | * @first_index: start the lookup from this key | |
828 | * @max_items: place up to this many items at *results | |
829 | * | |
830 | * Performs an index-ascending scan of the tree for present items. Places | |
831 | * their slots at *@results and returns the number of items which were | |
832 | * placed at *@results. | |
833 | * | |
834 | * The implementation is naive. | |
835 | * | |
836 | * Like radix_tree_gang_lookup as far as RCU and locking goes. Slots must | |
837 | * be dereferenced with radix_tree_deref_slot, and if using only RCU | |
838 | * protection, radix_tree_deref_slot may fail requiring a retry. | |
839 | */ | |
840 | unsigned int | |
841 | radix_tree_gang_lookup_slot(struct radix_tree_root *root, void ***results, | |
842 | unsigned long first_index, unsigned int max_items) | |
843 | { | |
844 | unsigned long max_index; | |
845 | struct radix_tree_node *node; | |
846 | unsigned long cur_index = first_index; | |
847 | unsigned int ret; | |
848 | ||
849 | node = rcu_dereference(root->rnode); | |
850 | if (!node) | |
851 | return 0; | |
852 | ||
853 | if (!radix_tree_is_indirect_ptr(node)) { | |
854 | if (first_index > 0) | |
855 | return 0; | |
856 | results[0] = (void **)&root->rnode; | |
857 | return 1; | |
858 | } | |
859 | node = radix_tree_indirect_to_ptr(node); | |
860 | ||
861 | max_index = radix_tree_maxindex(node->height); | |
862 | ||
863 | ret = 0; | |
864 | while (ret < max_items) { | |
865 | unsigned int slots_found; | |
866 | unsigned long next_index; /* Index of next search */ | |
867 | ||
868 | if (cur_index > max_index) | |
869 | break; | |
870 | slots_found = __lookup(node, results + ret, cur_index, | |
871 | max_items - ret, &next_index); | |
872 | ret += slots_found; | |
873 | if (next_index == 0) | |
874 | break; | |
875 | cur_index = next_index; | |
876 | } | |
877 | ||
878 | return ret; | |
879 | } | |
880 | EXPORT_SYMBOL(radix_tree_gang_lookup_slot); | |
881 | ||
1da177e4 LT |
882 | /* |
883 | * FIXME: the two tag_get()s here should use find_next_bit() instead of | |
884 | * open-coding the search. | |
885 | */ | |
886 | static unsigned int | |
47feff2c | 887 | __lookup_tag(struct radix_tree_node *slot, void ***results, unsigned long index, |
daff89f3 | 888 | unsigned int max_items, unsigned long *next_index, unsigned int tag) |
1da177e4 LT |
889 | { |
890 | unsigned int nr_found = 0; | |
7cf9c2c7 | 891 | unsigned int shift, height; |
1da177e4 | 892 | |
7cf9c2c7 NP |
893 | height = slot->height; |
894 | if (height == 0) | |
612d6c19 | 895 | goto out; |
7cf9c2c7 | 896 | shift = (height-1) * RADIX_TREE_MAP_SHIFT; |
1da177e4 | 897 | |
7cf9c2c7 NP |
898 | while (height > 0) { |
899 | unsigned long i = (index >> shift) & RADIX_TREE_MAP_MASK ; | |
1da177e4 | 900 | |
7cf9c2c7 NP |
901 | for (;;) { |
902 | if (tag_get(slot, tag, i)) | |
1da177e4 | 903 | break; |
1da177e4 LT |
904 | index &= ~((1UL << shift) - 1); |
905 | index += 1UL << shift; | |
906 | if (index == 0) | |
907 | goto out; /* 32-bit wraparound */ | |
7cf9c2c7 NP |
908 | i++; |
909 | if (i == RADIX_TREE_MAP_SIZE) | |
910 | goto out; | |
1da177e4 | 911 | } |
1da177e4 LT |
912 | height--; |
913 | if (height == 0) { /* Bottom level: grab some items */ | |
914 | unsigned long j = index & RADIX_TREE_MAP_MASK; | |
915 | ||
916 | for ( ; j < RADIX_TREE_MAP_SIZE; j++) { | |
917 | index++; | |
7cf9c2c7 NP |
918 | if (!tag_get(slot, tag, j)) |
919 | continue; | |
7cf9c2c7 NP |
920 | /* |
921 | * Even though the tag was found set, we need to | |
922 | * recheck that we have a non-NULL node, because | |
923 | * if this lookup is lockless, it may have been | |
924 | * subsequently deleted. | |
925 | * | |
926 | * Similar care must be taken in any place that | |
927 | * lookup ->slots[x] without a lock (ie. can't | |
928 | * rely on its value remaining the same). | |
929 | */ | |
47feff2c NP |
930 | if (slot->slots[j]) { |
931 | results[nr_found++] = &(slot->slots[j]); | |
1da177e4 LT |
932 | if (nr_found == max_items) |
933 | goto out; | |
934 | } | |
935 | } | |
936 | } | |
937 | shift -= RADIX_TREE_MAP_SHIFT; | |
7cf9c2c7 NP |
938 | slot = rcu_dereference(slot->slots[i]); |
939 | if (slot == NULL) | |
940 | break; | |
941 | } | |
1da177e4 LT |
942 | out: |
943 | *next_index = index; | |
944 | return nr_found; | |
945 | } | |
946 | ||
947 | /** | |
948 | * radix_tree_gang_lookup_tag - perform multiple lookup on a radix tree | |
949 | * based on a tag | |
950 | * @root: radix tree root | |
951 | * @results: where the results of the lookup are placed | |
952 | * @first_index: start the lookup from this key | |
953 | * @max_items: place up to this many items at *results | |
daff89f3 | 954 | * @tag: the tag index (< RADIX_TREE_MAX_TAGS) |
1da177e4 LT |
955 | * |
956 | * Performs an index-ascending scan of the tree for present items which | |
957 | * have the tag indexed by @tag set. Places the items at *@results and | |
958 | * returns the number of items which were placed at *@results. | |
959 | */ | |
960 | unsigned int | |
961 | radix_tree_gang_lookup_tag(struct radix_tree_root *root, void **results, | |
daff89f3 JC |
962 | unsigned long first_index, unsigned int max_items, |
963 | unsigned int tag) | |
1da177e4 | 964 | { |
7cf9c2c7 NP |
965 | struct radix_tree_node *node; |
966 | unsigned long max_index; | |
1da177e4 | 967 | unsigned long cur_index = first_index; |
7cf9c2c7 | 968 | unsigned int ret; |
1da177e4 | 969 | |
612d6c19 NP |
970 | /* check the root's tag bit */ |
971 | if (!root_tag_get(root, tag)) | |
972 | return 0; | |
973 | ||
7cf9c2c7 NP |
974 | node = rcu_dereference(root->rnode); |
975 | if (!node) | |
976 | return 0; | |
977 | ||
c0bc9875 | 978 | if (!radix_tree_is_indirect_ptr(node)) { |
7cf9c2c7 NP |
979 | if (first_index > 0) |
980 | return 0; | |
c0bc9875 | 981 | results[0] = node; |
7cf9c2c7 NP |
982 | return 1; |
983 | } | |
c0bc9875 | 984 | node = radix_tree_indirect_to_ptr(node); |
7cf9c2c7 NP |
985 | |
986 | max_index = radix_tree_maxindex(node->height); | |
987 | ||
988 | ret = 0; | |
1da177e4 | 989 | while (ret < max_items) { |
47feff2c | 990 | unsigned int nr_found, slots_found, i; |
1da177e4 LT |
991 | unsigned long next_index; /* Index of next search */ |
992 | ||
993 | if (cur_index > max_index) | |
994 | break; | |
47feff2c NP |
995 | slots_found = __lookup_tag(node, (void ***)results + ret, |
996 | cur_index, max_items - ret, &next_index, tag); | |
997 | nr_found = 0; | |
998 | for (i = 0; i < slots_found; i++) { | |
999 | struct radix_tree_node *slot; | |
1000 | slot = *(((void ***)results)[ret + i]); | |
1001 | if (!slot) | |
1002 | continue; | |
1003 | results[ret + nr_found] = rcu_dereference(slot); | |
1004 | nr_found++; | |
1005 | } | |
1da177e4 LT |
1006 | ret += nr_found; |
1007 | if (next_index == 0) | |
1008 | break; | |
1009 | cur_index = next_index; | |
1010 | } | |
7cf9c2c7 | 1011 | |
1da177e4 LT |
1012 | return ret; |
1013 | } | |
1014 | EXPORT_SYMBOL(radix_tree_gang_lookup_tag); | |
1015 | ||
47feff2c NP |
1016 | /** |
1017 | * radix_tree_gang_lookup_tag_slot - perform multiple slot lookup on a | |
1018 | * radix tree based on a tag | |
1019 | * @root: radix tree root | |
1020 | * @results: where the results of the lookup are placed | |
1021 | * @first_index: start the lookup from this key | |
1022 | * @max_items: place up to this many items at *results | |
1023 | * @tag: the tag index (< RADIX_TREE_MAX_TAGS) | |
1024 | * | |
1025 | * Performs an index-ascending scan of the tree for present items which | |
1026 | * have the tag indexed by @tag set. Places the slots at *@results and | |
1027 | * returns the number of slots which were placed at *@results. | |
1028 | */ | |
1029 | unsigned int | |
1030 | radix_tree_gang_lookup_tag_slot(struct radix_tree_root *root, void ***results, | |
1031 | unsigned long first_index, unsigned int max_items, | |
1032 | unsigned int tag) | |
1033 | { | |
1034 | struct radix_tree_node *node; | |
1035 | unsigned long max_index; | |
1036 | unsigned long cur_index = first_index; | |
1037 | unsigned int ret; | |
1038 | ||
1039 | /* check the root's tag bit */ | |
1040 | if (!root_tag_get(root, tag)) | |
1041 | return 0; | |
1042 | ||
1043 | node = rcu_dereference(root->rnode); | |
1044 | if (!node) | |
1045 | return 0; | |
1046 | ||
1047 | if (!radix_tree_is_indirect_ptr(node)) { | |
1048 | if (first_index > 0) | |
1049 | return 0; | |
1050 | results[0] = (void **)&root->rnode; | |
1051 | return 1; | |
1052 | } | |
1053 | node = radix_tree_indirect_to_ptr(node); | |
1054 | ||
1055 | max_index = radix_tree_maxindex(node->height); | |
1056 | ||
1057 | ret = 0; | |
1058 | while (ret < max_items) { | |
1059 | unsigned int slots_found; | |
1060 | unsigned long next_index; /* Index of next search */ | |
1061 | ||
1062 | if (cur_index > max_index) | |
1063 | break; | |
1064 | slots_found = __lookup_tag(node, results + ret, | |
1065 | cur_index, max_items - ret, &next_index, tag); | |
1066 | ret += slots_found; | |
1067 | if (next_index == 0) | |
1068 | break; | |
1069 | cur_index = next_index; | |
1070 | } | |
1071 | ||
1072 | return ret; | |
1073 | } | |
1074 | EXPORT_SYMBOL(radix_tree_gang_lookup_tag_slot); | |
1075 | ||
1076 | ||
a5f51c96 NP |
1077 | /** |
1078 | * radix_tree_shrink - shrink height of a radix tree to minimal | |
1079 | * @root radix tree root | |
1080 | */ | |
1081 | static inline void radix_tree_shrink(struct radix_tree_root *root) | |
1082 | { | |
1083 | /* try to shrink tree height */ | |
c0bc9875 | 1084 | while (root->height > 0) { |
a5f51c96 | 1085 | struct radix_tree_node *to_free = root->rnode; |
7cf9c2c7 | 1086 | void *newptr; |
a5f51c96 | 1087 | |
c0bc9875 NP |
1088 | BUG_ON(!radix_tree_is_indirect_ptr(to_free)); |
1089 | to_free = radix_tree_indirect_to_ptr(to_free); | |
1090 | ||
1091 | /* | |
1092 | * The candidate node has more than one child, or its child | |
1093 | * is not at the leftmost slot, we cannot shrink. | |
1094 | */ | |
1095 | if (to_free->count != 1) | |
1096 | break; | |
1097 | if (!to_free->slots[0]) | |
1098 | break; | |
1099 | ||
7cf9c2c7 NP |
1100 | /* |
1101 | * We don't need rcu_assign_pointer(), since we are simply | |
1102 | * moving the node from one part of the tree to another. If | |
1103 | * it was safe to dereference the old pointer to it | |
1104 | * (to_free->slots[0]), it will be safe to dereference the new | |
1105 | * one (root->rnode). | |
1106 | */ | |
1107 | newptr = to_free->slots[0]; | |
c0bc9875 NP |
1108 | if (root->height > 1) |
1109 | newptr = radix_tree_ptr_to_indirect(newptr); | |
7cf9c2c7 | 1110 | root->rnode = newptr; |
a5f51c96 | 1111 | root->height--; |
a5f51c96 NP |
1112 | radix_tree_node_free(to_free); |
1113 | } | |
1114 | } | |
1115 | ||
1da177e4 LT |
1116 | /** |
1117 | * radix_tree_delete - delete an item from a radix tree | |
1118 | * @root: radix tree root | |
1119 | * @index: index key | |
1120 | * | |
1121 | * Remove the item at @index from the radix tree rooted at @root. | |
1122 | * | |
1123 | * Returns the address of the deleted item, or NULL if it was not present. | |
1124 | */ | |
1125 | void *radix_tree_delete(struct radix_tree_root *root, unsigned long index) | |
1126 | { | |
26fb1589 JM |
1127 | /* |
1128 | * The radix tree path needs to be one longer than the maximum path | |
1129 | * since the "list" is null terminated. | |
1130 | */ | |
1131 | struct radix_tree_path path[RADIX_TREE_MAX_PATH + 1], *pathp = path; | |
612d6c19 | 1132 | struct radix_tree_node *slot = NULL; |
7cf9c2c7 | 1133 | struct radix_tree_node *to_free; |
1da177e4 | 1134 | unsigned int height, shift; |
d5274261 NP |
1135 | int tag; |
1136 | int offset; | |
1da177e4 LT |
1137 | |
1138 | height = root->height; | |
1139 | if (index > radix_tree_maxindex(height)) | |
1140 | goto out; | |
1141 | ||
612d6c19 | 1142 | slot = root->rnode; |
c0bc9875 | 1143 | if (height == 0) { |
612d6c19 NP |
1144 | root_tag_clear_all(root); |
1145 | root->rnode = NULL; | |
1146 | goto out; | |
1147 | } | |
c0bc9875 | 1148 | slot = radix_tree_indirect_to_ptr(slot); |
612d6c19 | 1149 | |
1da177e4 LT |
1150 | shift = (height - 1) * RADIX_TREE_MAP_SHIFT; |
1151 | pathp->node = NULL; | |
1da177e4 | 1152 | |
612d6c19 | 1153 | do { |
201b6264 | 1154 | if (slot == NULL) |
1da177e4 LT |
1155 | goto out; |
1156 | ||
d5274261 | 1157 | pathp++; |
1da177e4 | 1158 | offset = (index >> shift) & RADIX_TREE_MAP_MASK; |
d5274261 NP |
1159 | pathp->offset = offset; |
1160 | pathp->node = slot; | |
201b6264 | 1161 | slot = slot->slots[offset]; |
1da177e4 | 1162 | shift -= RADIX_TREE_MAP_SHIFT; |
612d6c19 NP |
1163 | height--; |
1164 | } while (height > 0); | |
1da177e4 | 1165 | |
612d6c19 | 1166 | if (slot == NULL) |
1da177e4 LT |
1167 | goto out; |
1168 | ||
1da177e4 LT |
1169 | /* |
1170 | * Clear all tags associated with the just-deleted item | |
1171 | */ | |
daff89f3 | 1172 | for (tag = 0; tag < RADIX_TREE_MAX_TAGS; tag++) { |
612d6c19 NP |
1173 | if (tag_get(pathp->node, tag, pathp->offset)) |
1174 | radix_tree_tag_clear(root, index, tag); | |
d5274261 | 1175 | } |
1da177e4 | 1176 | |
7cf9c2c7 | 1177 | to_free = NULL; |
201b6264 | 1178 | /* Now free the nodes we do not need anymore */ |
612d6c19 | 1179 | while (pathp->node) { |
201b6264 | 1180 | pathp->node->slots[pathp->offset] = NULL; |
a5f51c96 | 1181 | pathp->node->count--; |
7cf9c2c7 NP |
1182 | /* |
1183 | * Queue the node for deferred freeing after the | |
1184 | * last reference to it disappears (set NULL, above). | |
1185 | */ | |
1186 | if (to_free) | |
1187 | radix_tree_node_free(to_free); | |
a5f51c96 NP |
1188 | |
1189 | if (pathp->node->count) { | |
c0bc9875 NP |
1190 | if (pathp->node == |
1191 | radix_tree_indirect_to_ptr(root->rnode)) | |
a5f51c96 | 1192 | radix_tree_shrink(root); |
201b6264 | 1193 | goto out; |
a5f51c96 | 1194 | } |
201b6264 CL |
1195 | |
1196 | /* Node with zero slots in use so free it */ | |
7cf9c2c7 | 1197 | to_free = pathp->node; |
612d6c19 | 1198 | pathp--; |
7cf9c2c7 | 1199 | |
1da177e4 | 1200 | } |
612d6c19 | 1201 | root_tag_clear_all(root); |
201b6264 | 1202 | root->height = 0; |
612d6c19 | 1203 | root->rnode = NULL; |
7cf9c2c7 NP |
1204 | if (to_free) |
1205 | radix_tree_node_free(to_free); | |
612d6c19 | 1206 | |
1da177e4 | 1207 | out: |
612d6c19 | 1208 | return slot; |
1da177e4 LT |
1209 | } |
1210 | EXPORT_SYMBOL(radix_tree_delete); | |
1211 | ||
1212 | /** | |
1213 | * radix_tree_tagged - test whether any items in the tree are tagged | |
1214 | * @root: radix tree root | |
1215 | * @tag: tag to test | |
1216 | */ | |
daff89f3 | 1217 | int radix_tree_tagged(struct radix_tree_root *root, unsigned int tag) |
1da177e4 | 1218 | { |
612d6c19 | 1219 | return root_tag_get(root, tag); |
1da177e4 LT |
1220 | } |
1221 | EXPORT_SYMBOL(radix_tree_tagged); | |
1222 | ||
1223 | static void | |
51cc5068 | 1224 | radix_tree_node_ctor(void *node) |
1da177e4 LT |
1225 | { |
1226 | memset(node, 0, sizeof(struct radix_tree_node)); | |
1227 | } | |
1228 | ||
1229 | static __init unsigned long __maxindex(unsigned int height) | |
1230 | { | |
430d275a PL |
1231 | unsigned int width = height * RADIX_TREE_MAP_SHIFT; |
1232 | int shift = RADIX_TREE_INDEX_BITS - width; | |
1233 | ||
1234 | if (shift < 0) | |
1235 | return ~0UL; | |
1236 | if (shift >= BITS_PER_LONG) | |
1237 | return 0UL; | |
1238 | return ~0UL >> shift; | |
1da177e4 LT |
1239 | } |
1240 | ||
1241 | static __init void radix_tree_init_maxindex(void) | |
1242 | { | |
1243 | unsigned int i; | |
1244 | ||
1245 | for (i = 0; i < ARRAY_SIZE(height_to_maxindex); i++) | |
1246 | height_to_maxindex[i] = __maxindex(i); | |
1247 | } | |
1248 | ||
1da177e4 LT |
1249 | static int radix_tree_callback(struct notifier_block *nfb, |
1250 | unsigned long action, | |
1251 | void *hcpu) | |
1252 | { | |
1253 | int cpu = (long)hcpu; | |
1254 | struct radix_tree_preload *rtp; | |
1255 | ||
1256 | /* Free per-cpu pool of perloaded nodes */ | |
8bb78442 | 1257 | if (action == CPU_DEAD || action == CPU_DEAD_FROZEN) { |
1da177e4 LT |
1258 | rtp = &per_cpu(radix_tree_preloads, cpu); |
1259 | while (rtp->nr) { | |
1260 | kmem_cache_free(radix_tree_node_cachep, | |
1261 | rtp->nodes[rtp->nr-1]); | |
1262 | rtp->nodes[rtp->nr-1] = NULL; | |
1263 | rtp->nr--; | |
1264 | } | |
1265 | } | |
1266 | return NOTIFY_OK; | |
1267 | } | |
1da177e4 LT |
1268 | |
1269 | void __init radix_tree_init(void) | |
1270 | { | |
1271 | radix_tree_node_cachep = kmem_cache_create("radix_tree_node", | |
1272 | sizeof(struct radix_tree_node), 0, | |
488514d1 CL |
1273 | SLAB_PANIC | SLAB_RECLAIM_ACCOUNT, |
1274 | radix_tree_node_ctor); | |
1da177e4 LT |
1275 | radix_tree_init_maxindex(); |
1276 | hotcpu_notifier(radix_tree_callback, 0); | |
1277 | } |