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> | |
1da177e4 LT |
31 | #include <linux/string.h> |
32 | #include <linux/bitops.h> | |
7cf9c2c7 | 33 | #include <linux/rcupdate.h> |
1da177e4 LT |
34 | |
35 | ||
36 | #ifdef __KERNEL__ | |
cfd9b7df | 37 | #define RADIX_TREE_MAP_SHIFT (CONFIG_BASE_SMALL ? 4 : 6) |
1da177e4 LT |
38 | #else |
39 | #define RADIX_TREE_MAP_SHIFT 3 /* For more stressful testing */ | |
40 | #endif | |
1da177e4 LT |
41 | |
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 | ||
48 | struct radix_tree_node { | |
7cf9c2c7 | 49 | unsigned int height; /* Height from the bottom */ |
1da177e4 | 50 | unsigned int count; |
7cf9c2c7 | 51 | struct rcu_head rcu_head; |
1da177e4 | 52 | void *slots[RADIX_TREE_MAP_SIZE]; |
daff89f3 | 53 | unsigned long tags[RADIX_TREE_MAX_TAGS][RADIX_TREE_TAG_LONGS]; |
1da177e4 LT |
54 | }; |
55 | ||
56 | struct radix_tree_path { | |
201b6264 | 57 | struct radix_tree_node *node; |
1da177e4 LT |
58 | int offset; |
59 | }; | |
60 | ||
61 | #define RADIX_TREE_INDEX_BITS (8 /* CHAR_BIT */ * sizeof(unsigned long)) | |
26fb1589 JM |
62 | #define RADIX_TREE_MAX_PATH (DIV_ROUND_UP(RADIX_TREE_INDEX_BITS, \ |
63 | RADIX_TREE_MAP_SHIFT)) | |
1da177e4 | 64 | |
26fb1589 JM |
65 | /* |
66 | * The height_to_maxindex array needs to be one deeper than the maximum | |
67 | * path as height 0 holds only 1 entry. | |
68 | */ | |
69 | static unsigned long height_to_maxindex[RADIX_TREE_MAX_PATH + 1] __read_mostly; | |
1da177e4 LT |
70 | |
71 | /* | |
72 | * Radix tree node cache. | |
73 | */ | |
e18b890b | 74 | static struct kmem_cache *radix_tree_node_cachep; |
1da177e4 LT |
75 | |
76 | /* | |
77 | * Per-cpu pool of preloaded nodes | |
78 | */ | |
79 | struct radix_tree_preload { | |
80 | int nr; | |
81 | struct radix_tree_node *nodes[RADIX_TREE_MAX_PATH]; | |
82 | }; | |
8cef7d57 | 83 | static DEFINE_PER_CPU(struct radix_tree_preload, radix_tree_preloads) = { 0, }; |
1da177e4 | 84 | |
612d6c19 NP |
85 | static inline gfp_t root_gfp_mask(struct radix_tree_root *root) |
86 | { | |
87 | return root->gfp_mask & __GFP_BITS_MASK; | |
88 | } | |
89 | ||
643b52b9 NP |
90 | static 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 | ||
96 | static 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 | ||
102 | static 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 | ||
108 | static 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 | ||
113 | static 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 | ||
118 | static inline void root_tag_clear_all(struct radix_tree_root *root) | |
119 | { | |
120 | root->gfp_mask &= __GFP_BITS_MASK; | |
121 | } | |
122 | ||
123 | static 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 | */ | |
132 | static 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 | } | |
1da177e4 LT |
141 | /* |
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 | */ | |
145 | static struct radix_tree_node * | |
146 | radix_tree_node_alloc(struct radix_tree_root *root) | |
147 | { | |
e2848a0e | 148 | struct radix_tree_node *ret = NULL; |
612d6c19 | 149 | gfp_t gfp_mask = root_gfp_mask(root); |
1da177e4 | 150 | |
e2848a0e | 151 | if (!(gfp_mask & __GFP_WAIT)) { |
1da177e4 LT |
152 | struct radix_tree_preload *rtp; |
153 | ||
e2848a0e NP |
154 | /* |
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 | */ | |
1da177e4 LT |
159 | 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 | } | |
e2848a0e | 166 | if (ret == NULL) |
488514d1 | 167 | ret = kmem_cache_alloc(radix_tree_node_cachep, gfp_mask); |
e2848a0e | 168 | |
c0bc9875 | 169 | BUG_ON(radix_tree_is_indirect_ptr(ret)); |
1da177e4 LT |
170 | return ret; |
171 | } | |
172 | ||
7cf9c2c7 NP |
173 | static 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); | |
b6dd0865 | 177 | int i; |
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 | */ | |
b6dd0865 DC |
184 | for (i = 0; i < RADIX_TREE_MAX_TAGS; i++) |
185 | tag_clear(node, i, 0); | |
186 | ||
643b52b9 NP |
187 | node->slots[0] = NULL; |
188 | node->count = 0; | |
189 | ||
7cf9c2c7 NP |
190 | kmem_cache_free(radix_tree_node_cachep, node); |
191 | } | |
192 | ||
1da177e4 LT |
193 | static inline void |
194 | radix_tree_node_free(struct radix_tree_node *node) | |
195 | { | |
7cf9c2c7 | 196 | call_rcu(&node->rcu_head, radix_tree_node_rcu_free); |
1da177e4 LT |
197 | } |
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. | |
b34df792 DH |
204 | * |
205 | * To make use of this facility, the radix tree must be initialised without | |
206 | * __GFP_WAIT being passed to INIT_RADIX_TREE(). | |
1da177e4 | 207 | */ |
dd0fc66f | 208 | int radix_tree_preload(gfp_t gfp_mask) |
1da177e4 LT |
209 | { |
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(); | |
488514d1 | 218 | node = kmem_cache_alloc(radix_tree_node_cachep, gfp_mask); |
1da177e4 LT |
219 | 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; | |
229 | out: | |
230 | return ret; | |
231 | } | |
d7f0923d | 232 | EXPORT_SYMBOL(radix_tree_preload); |
1da177e4 | 233 | |
1da177e4 LT |
234 | /* |
235 | * Return the maximum key which can be store into a | |
236 | * radix tree with height HEIGHT. | |
237 | */ | |
238 | static 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 | */ | |
246 | static int radix_tree_extend(struct radix_tree_root *root, unsigned long index) | |
247 | { | |
248 | struct radix_tree_node *node; | |
249 | unsigned int height; | |
1da177e4 LT |
250 | 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 | ||
1da177e4 | 262 | do { |
7cf9c2c7 | 263 | unsigned int newheight; |
1da177e4 LT |
264 | if (!(node = radix_tree_node_alloc(root))) |
265 | return -ENOMEM; | |
266 | ||
267 | /* Increase the height. */ | |
c0bc9875 | 268 | node->slots[0] = radix_tree_indirect_to_ptr(root->rnode); |
1da177e4 LT |
269 | |
270 | /* Propagate the aggregated tag info into the new root */ | |
daff89f3 | 271 | for (tag = 0; tag < RADIX_TREE_MAX_TAGS; tag++) { |
612d6c19 | 272 | if (root_tag_get(root, tag)) |
1da177e4 LT |
273 | tag_set(node, tag, 0); |
274 | } | |
275 | ||
7cf9c2c7 NP |
276 | newheight = root->height+1; |
277 | node->height = newheight; | |
1da177e4 | 278 | node->count = 1; |
c0bc9875 | 279 | node = radix_tree_ptr_to_indirect(node); |
7cf9c2c7 NP |
280 | rcu_assign_pointer(root->rnode, node); |
281 | root->height = newheight; | |
1da177e4 LT |
282 | } while (height > root->height); |
283 | out: | |
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 | */ | |
295 | int radix_tree_insert(struct radix_tree_root *root, | |
296 | unsigned long index, void *item) | |
297 | { | |
201b6264 | 298 | struct radix_tree_node *node = NULL, *slot; |
1da177e4 LT |
299 | unsigned int height, shift; |
300 | int offset; | |
301 | int error; | |
302 | ||
c0bc9875 | 303 | BUG_ON(radix_tree_is_indirect_ptr(item)); |
7cf9c2c7 | 304 | |
1da177e4 | 305 | /* Make sure the tree is high enough. */ |
612d6c19 | 306 | if (index > radix_tree_maxindex(root->height)) { |
1da177e4 LT |
307 | error = radix_tree_extend(root, index); |
308 | if (error) | |
309 | return error; | |
310 | } | |
311 | ||
c0bc9875 NP |
312 | slot = radix_tree_indirect_to_ptr(root->rnode); |
313 | ||
1da177e4 LT |
314 | height = root->height; |
315 | shift = (height-1) * RADIX_TREE_MAP_SHIFT; | |
316 | ||
317 | offset = 0; /* uninitialised var warning */ | |
612d6c19 | 318 | while (height > 0) { |
201b6264 | 319 | if (slot == NULL) { |
1da177e4 | 320 | /* Have to add a child node. */ |
201b6264 | 321 | if (!(slot = radix_tree_node_alloc(root))) |
1da177e4 | 322 | return -ENOMEM; |
7cf9c2c7 | 323 | slot->height = height; |
201b6264 | 324 | if (node) { |
7cf9c2c7 | 325 | rcu_assign_pointer(node->slots[offset], slot); |
1da177e4 | 326 | node->count++; |
201b6264 | 327 | } else |
c0bc9875 NP |
328 | rcu_assign_pointer(root->rnode, |
329 | radix_tree_ptr_to_indirect(slot)); | |
1da177e4 LT |
330 | } |
331 | ||
332 | /* Go a level down */ | |
333 | offset = (index >> shift) & RADIX_TREE_MAP_MASK; | |
201b6264 CL |
334 | node = slot; |
335 | slot = node->slots[offset]; | |
1da177e4 LT |
336 | shift -= RADIX_TREE_MAP_SHIFT; |
337 | height--; | |
612d6c19 | 338 | } |
1da177e4 | 339 | |
201b6264 | 340 | if (slot != NULL) |
1da177e4 | 341 | return -EEXIST; |
201b6264 | 342 | |
612d6c19 NP |
343 | if (node) { |
344 | node->count++; | |
7cf9c2c7 | 345 | rcu_assign_pointer(node->slots[offset], item); |
612d6c19 NP |
346 | BUG_ON(tag_get(node, 0, offset)); |
347 | BUG_ON(tag_get(node, 1, offset)); | |
348 | } else { | |
c0bc9875 | 349 | rcu_assign_pointer(root->rnode, item); |
612d6c19 NP |
350 | BUG_ON(root_tag_get(root, 0)); |
351 | BUG_ON(root_tag_get(root, 1)); | |
352 | } | |
1da177e4 | 353 | |
1da177e4 LT |
354 | return 0; |
355 | } | |
356 | EXPORT_SYMBOL(radix_tree_insert); | |
357 | ||
b72b71c6 HS |
358 | /* |
359 | * is_slot == 1 : search for the slot. | |
360 | * is_slot == 0 : search for the node. | |
7cf9c2c7 | 361 | */ |
b72b71c6 HS |
362 | static void *radix_tree_lookup_element(struct radix_tree_root *root, |
363 | unsigned long index, int is_slot) | |
1da177e4 LT |
364 | { |
365 | unsigned int height, shift; | |
7cf9c2c7 | 366 | struct radix_tree_node *node, **slot; |
612d6c19 | 367 | |
2676a58c | 368 | node = rcu_dereference_raw(root->rnode); |
7cf9c2c7 | 369 | if (node == NULL) |
1da177e4 LT |
370 | return NULL; |
371 | ||
c0bc9875 | 372 | if (!radix_tree_is_indirect_ptr(node)) { |
7cf9c2c7 NP |
373 | if (index > 0) |
374 | return NULL; | |
b72b71c6 | 375 | return is_slot ? (void *)&root->rnode : node; |
7cf9c2c7 | 376 | } |
c0bc9875 | 377 | node = radix_tree_indirect_to_ptr(node); |
7cf9c2c7 NP |
378 | |
379 | height = node->height; | |
380 | if (index > radix_tree_maxindex(height)) | |
381 | return NULL; | |
612d6c19 | 382 | |
1da177e4 | 383 | shift = (height-1) * RADIX_TREE_MAP_SHIFT; |
1da177e4 | 384 | |
7cf9c2c7 NP |
385 | do { |
386 | slot = (struct radix_tree_node **) | |
387 | (node->slots + ((index>>shift) & RADIX_TREE_MAP_MASK)); | |
2676a58c | 388 | node = rcu_dereference_raw(*slot); |
7cf9c2c7 | 389 | if (node == NULL) |
1da177e4 LT |
390 | return NULL; |
391 | ||
1da177e4 LT |
392 | shift -= RADIX_TREE_MAP_SHIFT; |
393 | height--; | |
7cf9c2c7 | 394 | } while (height > 0); |
1da177e4 | 395 | |
b72b71c6 HS |
396 | 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 | */ | |
412 | void **radix_tree_lookup_slot(struct radix_tree_root *root, unsigned long index) | |
413 | { | |
414 | return (void **)radix_tree_lookup_element(root, index, 1); | |
a4331366 | 415 | } |
a4331366 HR |
416 | EXPORT_SYMBOL(radix_tree_lookup_slot); |
417 | ||
418 | /** | |
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. | |
7cf9c2c7 NP |
424 | * |
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. | |
a4331366 HR |
429 | */ |
430 | void *radix_tree_lookup(struct radix_tree_root *root, unsigned long index) | |
431 | { | |
b72b71c6 | 432 | return radix_tree_lookup_element(root, index, 0); |
1da177e4 LT |
433 | } |
434 | EXPORT_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 | * | |
daff89f3 JC |
442 | * Set the search tag (which must be < RADIX_TREE_MAX_TAGS) |
443 | * corresponding to @index in the radix tree. From | |
1da177e4 LT |
444 | * 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 | */ | |
449 | void *radix_tree_tag_set(struct radix_tree_root *root, | |
daff89f3 | 450 | unsigned long index, unsigned int tag) |
1da177e4 LT |
451 | { |
452 | unsigned int height, shift; | |
201b6264 | 453 | struct radix_tree_node *slot; |
1da177e4 LT |
454 | |
455 | height = root->height; | |
4c91c364 | 456 | BUG_ON(index > radix_tree_maxindex(height)); |
1da177e4 | 457 | |
c0bc9875 | 458 | slot = radix_tree_indirect_to_ptr(root->rnode); |
612d6c19 | 459 | shift = (height - 1) * RADIX_TREE_MAP_SHIFT; |
1da177e4 LT |
460 | |
461 | while (height > 0) { | |
462 | int offset; | |
463 | ||
464 | offset = (index >> shift) & RADIX_TREE_MAP_MASK; | |
d5274261 NP |
465 | if (!tag_get(slot, tag, offset)) |
466 | tag_set(slot, tag, offset); | |
201b6264 CL |
467 | slot = slot->slots[offset]; |
468 | BUG_ON(slot == NULL); | |
1da177e4 LT |
469 | shift -= RADIX_TREE_MAP_SHIFT; |
470 | height--; | |
471 | } | |
472 | ||
612d6c19 NP |
473 | /* set the root's tag bit */ |
474 | if (slot && !root_tag_get(root, tag)) | |
475 | root_tag_set(root, tag); | |
476 | ||
201b6264 | 477 | return slot; |
1da177e4 LT |
478 | } |
479 | EXPORT_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 | * | |
daff89f3 JC |
487 | * Clear the search tag (which must be < RADIX_TREE_MAX_TAGS) |
488 | * corresponding to @index in the radix tree. If | |
1da177e4 LT |
489 | * 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 | */ | |
495 | void *radix_tree_tag_clear(struct radix_tree_root *root, | |
daff89f3 | 496 | unsigned long index, unsigned int tag) |
1da177e4 | 497 | { |
26fb1589 JM |
498 | /* |
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; | |
612d6c19 | 503 | struct radix_tree_node *slot = NULL; |
1da177e4 | 504 | unsigned int height, shift; |
1da177e4 LT |
505 | |
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; | |
c0bc9875 | 512 | slot = radix_tree_indirect_to_ptr(root->rnode); |
1da177e4 LT |
513 | |
514 | while (height > 0) { | |
515 | int offset; | |
516 | ||
201b6264 | 517 | if (slot == NULL) |
1da177e4 LT |
518 | goto out; |
519 | ||
520 | offset = (index >> shift) & RADIX_TREE_MAP_MASK; | |
521 | pathp[1].offset = offset; | |
201b6264 CL |
522 | pathp[1].node = slot; |
523 | slot = slot->slots[offset]; | |
1da177e4 LT |
524 | pathp++; |
525 | shift -= RADIX_TREE_MAP_SHIFT; | |
526 | height--; | |
527 | } | |
528 | ||
612d6c19 | 529 | if (slot == NULL) |
1da177e4 LT |
530 | goto out; |
531 | ||
612d6c19 | 532 | while (pathp->node) { |
d5274261 NP |
533 | if (!tag_get(pathp->node, tag, pathp->offset)) |
534 | goto out; | |
201b6264 | 535 | tag_clear(pathp->node, tag, pathp->offset); |
6e954b9e NP |
536 | if (any_tag_set(pathp->node, tag)) |
537 | goto out; | |
1da177e4 | 538 | pathp--; |
612d6c19 NP |
539 | } |
540 | ||
541 | /* clear the root's tag bit */ | |
542 | if (root_tag_get(root, tag)) | |
543 | root_tag_clear(root, tag); | |
544 | ||
1da177e4 | 545 | out: |
612d6c19 | 546 | return slot; |
1da177e4 LT |
547 | } |
548 | EXPORT_SYMBOL(radix_tree_tag_clear); | |
549 | ||
1da177e4 | 550 | /** |
32605a18 MT |
551 | * radix_tree_tag_get - get a tag on a radix tree node |
552 | * @root: radix tree root | |
553 | * @index: index key | |
daff89f3 | 554 | * @tag: tag index (< RADIX_TREE_MAX_TAGS) |
1da177e4 | 555 | * |
32605a18 | 556 | * Return values: |
1da177e4 | 557 | * |
612d6c19 NP |
558 | * 0: tag not present or not set |
559 | * 1: tag set | |
ce82653d DH |
560 | * |
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. | |
1da177e4 LT |
564 | */ |
565 | int radix_tree_tag_get(struct radix_tree_root *root, | |
daff89f3 | 566 | unsigned long index, unsigned int tag) |
1da177e4 LT |
567 | { |
568 | unsigned int height, shift; | |
7cf9c2c7 | 569 | struct radix_tree_node *node; |
1da177e4 LT |
570 | int saw_unset_tag = 0; |
571 | ||
612d6c19 NP |
572 | /* check the root's tag bit */ |
573 | if (!root_tag_get(root, tag)) | |
574 | return 0; | |
575 | ||
2676a58c | 576 | node = rcu_dereference_raw(root->rnode); |
7cf9c2c7 NP |
577 | if (node == NULL) |
578 | return 0; | |
579 | ||
c0bc9875 | 580 | if (!radix_tree_is_indirect_ptr(node)) |
7cf9c2c7 | 581 | return (index == 0); |
c0bc9875 | 582 | node = radix_tree_indirect_to_ptr(node); |
7cf9c2c7 NP |
583 | |
584 | height = node->height; | |
585 | if (index > radix_tree_maxindex(height)) | |
586 | return 0; | |
612d6c19 | 587 | |
1da177e4 | 588 | shift = (height - 1) * RADIX_TREE_MAP_SHIFT; |
1da177e4 LT |
589 | |
590 | for ( ; ; ) { | |
591 | int offset; | |
592 | ||
7cf9c2c7 | 593 | if (node == NULL) |
1da177e4 LT |
594 | 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 | */ | |
7cf9c2c7 | 602 | if (!tag_get(node, tag, offset)) |
1da177e4 | 603 | saw_unset_tag = 1; |
ce82653d DH |
604 | if (height == 1) |
605 | return !!tag_get(node, tag, offset); | |
2676a58c | 606 | node = rcu_dereference_raw(node->slots[offset]); |
1da177e4 LT |
607 | shift -= RADIX_TREE_MAP_SHIFT; |
608 | height--; | |
609 | } | |
610 | } | |
611 | EXPORT_SYMBOL(radix_tree_tag_get); | |
1da177e4 | 612 | |
ebf8aa44 JK |
613 | /** |
614 | * 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 | * | |
628 | * The function returns number of leaves where the tag was set and sets | |
629 | * *first_indexp to the first unscanned index. | |
630 | */ | |
631 | unsigned long radix_tree_range_tag_if_tagged(struct radix_tree_root *root, | |
632 | unsigned long *first_indexp, unsigned long last_index, | |
633 | unsigned long nr_to_tag, | |
634 | unsigned int iftag, unsigned int settag) | |
635 | { | |
636 | unsigned int height = root->height, shift; | |
637 | unsigned long tagged = 0, index = *first_indexp; | |
638 | struct radix_tree_node *open_slots[height], *slot; | |
639 | ||
640 | last_index = min(last_index, radix_tree_maxindex(height)); | |
641 | if (index > last_index) | |
642 | return 0; | |
643 | if (!nr_to_tag) | |
644 | return 0; | |
645 | if (!root_tag_get(root, iftag)) { | |
646 | *first_indexp = last_index + 1; | |
647 | return 0; | |
648 | } | |
649 | if (height == 0) { | |
650 | *first_indexp = last_index + 1; | |
651 | root_tag_set(root, settag); | |
652 | return 1; | |
653 | } | |
654 | ||
655 | shift = (height - 1) * RADIX_TREE_MAP_SHIFT; | |
656 | slot = radix_tree_indirect_to_ptr(root->rnode); | |
657 | ||
658 | for (;;) { | |
659 | int offset; | |
660 | ||
661 | offset = (index >> shift) & RADIX_TREE_MAP_MASK; | |
662 | if (!slot->slots[offset]) | |
663 | goto next; | |
664 | if (!tag_get(slot, iftag, offset)) | |
665 | goto next; | |
666 | tag_set(slot, settag, offset); | |
667 | if (height == 1) { | |
668 | tagged++; | |
669 | goto next; | |
670 | } | |
671 | /* Go down one level */ | |
672 | height--; | |
673 | shift -= RADIX_TREE_MAP_SHIFT; | |
674 | open_slots[height] = slot; | |
675 | slot = slot->slots[offset]; | |
676 | continue; | |
677 | next: | |
678 | /* Go to next item at level determined by 'shift' */ | |
679 | index = ((index >> shift) + 1) << shift; | |
680 | if (index > last_index) | |
681 | break; | |
682 | if (tagged >= nr_to_tag) | |
683 | break; | |
684 | while (((index >> shift) & RADIX_TREE_MAP_MASK) == 0) { | |
685 | /* | |
686 | * We've fully scanned this node. Go up. Because | |
687 | * last_index is guaranteed to be in the tree, what | |
688 | * we do below cannot wander astray. | |
689 | */ | |
690 | slot = open_slots[height]; | |
691 | height++; | |
692 | shift += RADIX_TREE_MAP_SHIFT; | |
693 | } | |
694 | } | |
695 | /* | |
696 | * The iftag must have been set somewhere because otherwise | |
697 | * we would return immediated at the beginning of the function | |
698 | */ | |
699 | root_tag_set(root, settag); | |
700 | *first_indexp = index; | |
701 | ||
702 | return tagged; | |
703 | } | |
704 | EXPORT_SYMBOL(radix_tree_range_tag_if_tagged); | |
705 | ||
706 | ||
6df8ba4f FW |
707 | /** |
708 | * radix_tree_next_hole - find the next hole (not-present entry) | |
709 | * @root: tree root | |
710 | * @index: index key | |
711 | * @max_scan: maximum range to search | |
712 | * | |
713 | * Search the set [index, min(index+max_scan-1, MAX_INDEX)] for the lowest | |
714 | * indexed hole. | |
715 | * | |
716 | * Returns: the index of the hole if found, otherwise returns an index | |
717 | * outside of the set specified (in which case 'return - index >= max_scan' | |
8e6bdb7f | 718 | * will be true). In rare cases of index wrap-around, 0 will be returned. |
6df8ba4f FW |
719 | * |
720 | * radix_tree_next_hole may be called under rcu_read_lock. However, like | |
8e6bdb7f WF |
721 | * radix_tree_gang_lookup, this will not atomically search a snapshot of |
722 | * the tree at a single point in time. For example, if a hole is created | |
723 | * at index 5, then subsequently a hole is created at index 10, | |
724 | * radix_tree_next_hole covering both indexes may return 10 if called | |
725 | * under rcu_read_lock. | |
6df8ba4f FW |
726 | */ |
727 | unsigned long radix_tree_next_hole(struct radix_tree_root *root, | |
728 | unsigned long index, unsigned long max_scan) | |
729 | { | |
730 | unsigned long i; | |
731 | ||
732 | for (i = 0; i < max_scan; i++) { | |
733 | if (!radix_tree_lookup(root, index)) | |
734 | break; | |
735 | index++; | |
736 | if (index == 0) | |
737 | break; | |
738 | } | |
739 | ||
740 | return index; | |
741 | } | |
742 | EXPORT_SYMBOL(radix_tree_next_hole); | |
743 | ||
dc566127 WF |
744 | /** |
745 | * radix_tree_prev_hole - find the prev hole (not-present entry) | |
746 | * @root: tree root | |
747 | * @index: index key | |
748 | * @max_scan: maximum range to search | |
749 | * | |
750 | * Search backwards in the range [max(index-max_scan+1, 0), index] | |
751 | * for the first hole. | |
752 | * | |
753 | * Returns: the index of the hole if found, otherwise returns an index | |
754 | * outside of the set specified (in which case 'index - return >= max_scan' | |
edcd1d84 | 755 | * will be true). In rare cases of wrap-around, ULONG_MAX will be returned. |
dc566127 WF |
756 | * |
757 | * radix_tree_next_hole may be called under rcu_read_lock. However, like | |
758 | * radix_tree_gang_lookup, this will not atomically search a snapshot of | |
759 | * the tree at a single point in time. For example, if a hole is created | |
760 | * at index 10, then subsequently a hole is created at index 5, | |
761 | * radix_tree_prev_hole covering both indexes may return 5 if called under | |
762 | * rcu_read_lock. | |
763 | */ | |
764 | unsigned long radix_tree_prev_hole(struct radix_tree_root *root, | |
765 | unsigned long index, unsigned long max_scan) | |
766 | { | |
767 | unsigned long i; | |
768 | ||
769 | for (i = 0; i < max_scan; i++) { | |
770 | if (!radix_tree_lookup(root, index)) | |
771 | break; | |
772 | index--; | |
edcd1d84 | 773 | if (index == ULONG_MAX) |
dc566127 WF |
774 | break; |
775 | } | |
776 | ||
777 | return index; | |
778 | } | |
779 | EXPORT_SYMBOL(radix_tree_prev_hole); | |
780 | ||
1da177e4 | 781 | static unsigned int |
47feff2c | 782 | __lookup(struct radix_tree_node *slot, void ***results, unsigned long index, |
1da177e4 LT |
783 | unsigned int max_items, unsigned long *next_index) |
784 | { | |
785 | unsigned int nr_found = 0; | |
201b6264 | 786 | unsigned int shift, height; |
201b6264 CL |
787 | unsigned long i; |
788 | ||
7cf9c2c7 NP |
789 | height = slot->height; |
790 | if (height == 0) | |
201b6264 | 791 | goto out; |
1da177e4 | 792 | shift = (height-1) * RADIX_TREE_MAP_SHIFT; |
1da177e4 | 793 | |
201b6264 | 794 | for ( ; height > 1; height--) { |
7cf9c2c7 NP |
795 | i = (index >> shift) & RADIX_TREE_MAP_MASK; |
796 | for (;;) { | |
1da177e4 LT |
797 | if (slot->slots[i] != NULL) |
798 | break; | |
799 | index &= ~((1UL << shift) - 1); | |
800 | index += 1UL << shift; | |
801 | if (index == 0) | |
802 | goto out; /* 32-bit wraparound */ | |
7cf9c2c7 NP |
803 | i++; |
804 | if (i == RADIX_TREE_MAP_SIZE) | |
805 | goto out; | |
1da177e4 | 806 | } |
1da177e4 | 807 | |
1da177e4 | 808 | shift -= RADIX_TREE_MAP_SHIFT; |
2676a58c | 809 | slot = rcu_dereference_raw(slot->slots[i]); |
7cf9c2c7 NP |
810 | if (slot == NULL) |
811 | goto out; | |
1da177e4 | 812 | } |
201b6264 CL |
813 | |
814 | /* Bottom level: grab some items */ | |
815 | for (i = index & RADIX_TREE_MAP_MASK; i < RADIX_TREE_MAP_SIZE; i++) { | |
816 | index++; | |
47feff2c NP |
817 | if (slot->slots[i]) { |
818 | results[nr_found++] = &(slot->slots[i]); | |
201b6264 CL |
819 | if (nr_found == max_items) |
820 | goto out; | |
821 | } | |
822 | } | |
1da177e4 LT |
823 | out: |
824 | *next_index = index; | |
825 | return nr_found; | |
826 | } | |
827 | ||
828 | /** | |
829 | * radix_tree_gang_lookup - perform multiple lookup on a radix tree | |
830 | * @root: radix tree root | |
831 | * @results: where the results of the lookup are placed | |
832 | * @first_index: start the lookup from this key | |
833 | * @max_items: place up to this many items at *results | |
834 | * | |
835 | * Performs an index-ascending scan of the tree for present items. Places | |
836 | * them at *@results and returns the number of items which were placed at | |
837 | * *@results. | |
838 | * | |
839 | * The implementation is naive. | |
7cf9c2c7 NP |
840 | * |
841 | * Like radix_tree_lookup, radix_tree_gang_lookup may be called under | |
842 | * rcu_read_lock. In this case, rather than the returned results being | |
843 | * an atomic snapshot of the tree at a single point in time, the semantics | |
844 | * of an RCU protected gang lookup are as though multiple radix_tree_lookups | |
845 | * have been issued in individual locks, and results stored in 'results'. | |
1da177e4 LT |
846 | */ |
847 | unsigned int | |
848 | radix_tree_gang_lookup(struct radix_tree_root *root, void **results, | |
849 | unsigned long first_index, unsigned int max_items) | |
850 | { | |
7cf9c2c7 NP |
851 | unsigned long max_index; |
852 | struct radix_tree_node *node; | |
1da177e4 | 853 | unsigned long cur_index = first_index; |
7cf9c2c7 NP |
854 | unsigned int ret; |
855 | ||
2676a58c | 856 | node = rcu_dereference_raw(root->rnode); |
7cf9c2c7 NP |
857 | if (!node) |
858 | return 0; | |
1da177e4 | 859 | |
c0bc9875 | 860 | if (!radix_tree_is_indirect_ptr(node)) { |
7cf9c2c7 NP |
861 | if (first_index > 0) |
862 | return 0; | |
c0bc9875 | 863 | results[0] = node; |
7cf9c2c7 NP |
864 | return 1; |
865 | } | |
c0bc9875 | 866 | node = radix_tree_indirect_to_ptr(node); |
7cf9c2c7 NP |
867 | |
868 | max_index = radix_tree_maxindex(node->height); | |
869 | ||
870 | ret = 0; | |
1da177e4 | 871 | while (ret < max_items) { |
47feff2c | 872 | unsigned int nr_found, slots_found, i; |
1da177e4 LT |
873 | unsigned long next_index; /* Index of next search */ |
874 | ||
875 | if (cur_index > max_index) | |
876 | break; | |
47feff2c | 877 | slots_found = __lookup(node, (void ***)results + ret, cur_index, |
1da177e4 | 878 | max_items - ret, &next_index); |
47feff2c NP |
879 | nr_found = 0; |
880 | for (i = 0; i < slots_found; i++) { | |
881 | struct radix_tree_node *slot; | |
882 | slot = *(((void ***)results)[ret + i]); | |
883 | if (!slot) | |
884 | continue; | |
2676a58c | 885 | results[ret + nr_found] = rcu_dereference_raw(slot); |
47feff2c NP |
886 | nr_found++; |
887 | } | |
1da177e4 LT |
888 | ret += nr_found; |
889 | if (next_index == 0) | |
890 | break; | |
891 | cur_index = next_index; | |
892 | } | |
7cf9c2c7 | 893 | |
1da177e4 LT |
894 | return ret; |
895 | } | |
896 | EXPORT_SYMBOL(radix_tree_gang_lookup); | |
897 | ||
47feff2c NP |
898 | /** |
899 | * radix_tree_gang_lookup_slot - perform multiple slot lookup on radix tree | |
900 | * @root: radix tree root | |
901 | * @results: where the results of the lookup are placed | |
902 | * @first_index: start the lookup from this key | |
903 | * @max_items: place up to this many items at *results | |
904 | * | |
905 | * Performs an index-ascending scan of the tree for present items. Places | |
906 | * their slots at *@results and returns the number of items which were | |
907 | * placed at *@results. | |
908 | * | |
909 | * The implementation is naive. | |
910 | * | |
911 | * Like radix_tree_gang_lookup as far as RCU and locking goes. Slots must | |
912 | * be dereferenced with radix_tree_deref_slot, and if using only RCU | |
913 | * protection, radix_tree_deref_slot may fail requiring a retry. | |
914 | */ | |
915 | unsigned int | |
916 | radix_tree_gang_lookup_slot(struct radix_tree_root *root, void ***results, | |
917 | unsigned long first_index, unsigned int max_items) | |
918 | { | |
919 | unsigned long max_index; | |
920 | struct radix_tree_node *node; | |
921 | unsigned long cur_index = first_index; | |
922 | unsigned int ret; | |
923 | ||
2676a58c | 924 | node = rcu_dereference_raw(root->rnode); |
47feff2c NP |
925 | if (!node) |
926 | return 0; | |
927 | ||
928 | if (!radix_tree_is_indirect_ptr(node)) { | |
929 | if (first_index > 0) | |
930 | return 0; | |
931 | results[0] = (void **)&root->rnode; | |
932 | return 1; | |
933 | } | |
934 | node = radix_tree_indirect_to_ptr(node); | |
935 | ||
936 | max_index = radix_tree_maxindex(node->height); | |
937 | ||
938 | ret = 0; | |
939 | while (ret < max_items) { | |
940 | unsigned int slots_found; | |
941 | unsigned long next_index; /* Index of next search */ | |
942 | ||
943 | if (cur_index > max_index) | |
944 | break; | |
945 | slots_found = __lookup(node, results + ret, cur_index, | |
946 | max_items - ret, &next_index); | |
947 | ret += slots_found; | |
948 | if (next_index == 0) | |
949 | break; | |
950 | cur_index = next_index; | |
951 | } | |
952 | ||
953 | return ret; | |
954 | } | |
955 | EXPORT_SYMBOL(radix_tree_gang_lookup_slot); | |
956 | ||
1da177e4 LT |
957 | /* |
958 | * FIXME: the two tag_get()s here should use find_next_bit() instead of | |
959 | * open-coding the search. | |
960 | */ | |
961 | static unsigned int | |
47feff2c | 962 | __lookup_tag(struct radix_tree_node *slot, void ***results, unsigned long index, |
daff89f3 | 963 | unsigned int max_items, unsigned long *next_index, unsigned int tag) |
1da177e4 LT |
964 | { |
965 | unsigned int nr_found = 0; | |
7cf9c2c7 | 966 | unsigned int shift, height; |
1da177e4 | 967 | |
7cf9c2c7 NP |
968 | height = slot->height; |
969 | if (height == 0) | |
612d6c19 | 970 | goto out; |
7cf9c2c7 | 971 | shift = (height-1) * RADIX_TREE_MAP_SHIFT; |
1da177e4 | 972 | |
7cf9c2c7 NP |
973 | while (height > 0) { |
974 | unsigned long i = (index >> shift) & RADIX_TREE_MAP_MASK ; | |
1da177e4 | 975 | |
7cf9c2c7 NP |
976 | for (;;) { |
977 | if (tag_get(slot, tag, i)) | |
1da177e4 | 978 | break; |
1da177e4 LT |
979 | index &= ~((1UL << shift) - 1); |
980 | index += 1UL << shift; | |
981 | if (index == 0) | |
982 | goto out; /* 32-bit wraparound */ | |
7cf9c2c7 NP |
983 | i++; |
984 | if (i == RADIX_TREE_MAP_SIZE) | |
985 | goto out; | |
1da177e4 | 986 | } |
1da177e4 LT |
987 | height--; |
988 | if (height == 0) { /* Bottom level: grab some items */ | |
989 | unsigned long j = index & RADIX_TREE_MAP_MASK; | |
990 | ||
991 | for ( ; j < RADIX_TREE_MAP_SIZE; j++) { | |
992 | index++; | |
7cf9c2c7 NP |
993 | if (!tag_get(slot, tag, j)) |
994 | continue; | |
7cf9c2c7 NP |
995 | /* |
996 | * Even though the tag was found set, we need to | |
997 | * recheck that we have a non-NULL node, because | |
998 | * if this lookup is lockless, it may have been | |
999 | * subsequently deleted. | |
1000 | * | |
1001 | * Similar care must be taken in any place that | |
1002 | * lookup ->slots[x] without a lock (ie. can't | |
1003 | * rely on its value remaining the same). | |
1004 | */ | |
47feff2c NP |
1005 | if (slot->slots[j]) { |
1006 | results[nr_found++] = &(slot->slots[j]); | |
1da177e4 LT |
1007 | if (nr_found == max_items) |
1008 | goto out; | |
1009 | } | |
1010 | } | |
1011 | } | |
1012 | shift -= RADIX_TREE_MAP_SHIFT; | |
2676a58c | 1013 | slot = rcu_dereference_raw(slot->slots[i]); |
7cf9c2c7 NP |
1014 | if (slot == NULL) |
1015 | break; | |
1016 | } | |
1da177e4 LT |
1017 | out: |
1018 | *next_index = index; | |
1019 | return nr_found; | |
1020 | } | |
1021 | ||
1022 | /** | |
1023 | * radix_tree_gang_lookup_tag - perform multiple lookup on a radix tree | |
1024 | * based on a tag | |
1025 | * @root: radix tree root | |
1026 | * @results: where the results of the lookup are placed | |
1027 | * @first_index: start the lookup from this key | |
1028 | * @max_items: place up to this many items at *results | |
daff89f3 | 1029 | * @tag: the tag index (< RADIX_TREE_MAX_TAGS) |
1da177e4 LT |
1030 | * |
1031 | * Performs an index-ascending scan of the tree for present items which | |
1032 | * have the tag indexed by @tag set. Places the items at *@results and | |
1033 | * returns the number of items which were placed at *@results. | |
1034 | */ | |
1035 | unsigned int | |
1036 | radix_tree_gang_lookup_tag(struct radix_tree_root *root, void **results, | |
daff89f3 JC |
1037 | unsigned long first_index, unsigned int max_items, |
1038 | unsigned int tag) | |
1da177e4 | 1039 | { |
7cf9c2c7 NP |
1040 | struct radix_tree_node *node; |
1041 | unsigned long max_index; | |
1da177e4 | 1042 | unsigned long cur_index = first_index; |
7cf9c2c7 | 1043 | unsigned int ret; |
1da177e4 | 1044 | |
612d6c19 NP |
1045 | /* check the root's tag bit */ |
1046 | if (!root_tag_get(root, tag)) | |
1047 | return 0; | |
1048 | ||
2676a58c | 1049 | node = rcu_dereference_raw(root->rnode); |
7cf9c2c7 NP |
1050 | if (!node) |
1051 | return 0; | |
1052 | ||
c0bc9875 | 1053 | if (!radix_tree_is_indirect_ptr(node)) { |
7cf9c2c7 NP |
1054 | if (first_index > 0) |
1055 | return 0; | |
c0bc9875 | 1056 | results[0] = node; |
7cf9c2c7 NP |
1057 | return 1; |
1058 | } | |
c0bc9875 | 1059 | node = radix_tree_indirect_to_ptr(node); |
7cf9c2c7 NP |
1060 | |
1061 | max_index = radix_tree_maxindex(node->height); | |
1062 | ||
1063 | ret = 0; | |
1da177e4 | 1064 | while (ret < max_items) { |
47feff2c | 1065 | unsigned int nr_found, slots_found, i; |
1da177e4 LT |
1066 | unsigned long next_index; /* Index of next search */ |
1067 | ||
1068 | if (cur_index > max_index) | |
1069 | break; | |
47feff2c NP |
1070 | slots_found = __lookup_tag(node, (void ***)results + ret, |
1071 | cur_index, max_items - ret, &next_index, tag); | |
1072 | nr_found = 0; | |
1073 | for (i = 0; i < slots_found; i++) { | |
1074 | struct radix_tree_node *slot; | |
1075 | slot = *(((void ***)results)[ret + i]); | |
1076 | if (!slot) | |
1077 | continue; | |
2676a58c | 1078 | results[ret + nr_found] = rcu_dereference_raw(slot); |
47feff2c NP |
1079 | nr_found++; |
1080 | } | |
1da177e4 LT |
1081 | ret += nr_found; |
1082 | if (next_index == 0) | |
1083 | break; | |
1084 | cur_index = next_index; | |
1085 | } | |
7cf9c2c7 | 1086 | |
1da177e4 LT |
1087 | return ret; |
1088 | } | |
1089 | EXPORT_SYMBOL(radix_tree_gang_lookup_tag); | |
1090 | ||
47feff2c NP |
1091 | /** |
1092 | * radix_tree_gang_lookup_tag_slot - perform multiple slot lookup on a | |
1093 | * radix tree based on a tag | |
1094 | * @root: radix tree root | |
1095 | * @results: where the results of the lookup are placed | |
1096 | * @first_index: start the lookup from this key | |
1097 | * @max_items: place up to this many items at *results | |
1098 | * @tag: the tag index (< RADIX_TREE_MAX_TAGS) | |
1099 | * | |
1100 | * Performs an index-ascending scan of the tree for present items which | |
1101 | * have the tag indexed by @tag set. Places the slots at *@results and | |
1102 | * returns the number of slots which were placed at *@results. | |
1103 | */ | |
1104 | unsigned int | |
1105 | radix_tree_gang_lookup_tag_slot(struct radix_tree_root *root, void ***results, | |
1106 | unsigned long first_index, unsigned int max_items, | |
1107 | unsigned int tag) | |
1108 | { | |
1109 | struct radix_tree_node *node; | |
1110 | unsigned long max_index; | |
1111 | unsigned long cur_index = first_index; | |
1112 | unsigned int ret; | |
1113 | ||
1114 | /* check the root's tag bit */ | |
1115 | if (!root_tag_get(root, tag)) | |
1116 | return 0; | |
1117 | ||
2676a58c | 1118 | node = rcu_dereference_raw(root->rnode); |
47feff2c NP |
1119 | if (!node) |
1120 | return 0; | |
1121 | ||
1122 | if (!radix_tree_is_indirect_ptr(node)) { | |
1123 | if (first_index > 0) | |
1124 | return 0; | |
1125 | results[0] = (void **)&root->rnode; | |
1126 | return 1; | |
1127 | } | |
1128 | node = radix_tree_indirect_to_ptr(node); | |
1129 | ||
1130 | max_index = radix_tree_maxindex(node->height); | |
1131 | ||
1132 | ret = 0; | |
1133 | while (ret < max_items) { | |
1134 | unsigned int slots_found; | |
1135 | unsigned long next_index; /* Index of next search */ | |
1136 | ||
1137 | if (cur_index > max_index) | |
1138 | break; | |
1139 | slots_found = __lookup_tag(node, results + ret, | |
1140 | cur_index, max_items - ret, &next_index, tag); | |
1141 | ret += slots_found; | |
1142 | if (next_index == 0) | |
1143 | break; | |
1144 | cur_index = next_index; | |
1145 | } | |
1146 | ||
1147 | return ret; | |
1148 | } | |
1149 | EXPORT_SYMBOL(radix_tree_gang_lookup_tag_slot); | |
1150 | ||
1151 | ||
a5f51c96 NP |
1152 | /** |
1153 | * radix_tree_shrink - shrink height of a radix tree to minimal | |
1154 | * @root radix tree root | |
1155 | */ | |
1156 | static inline void radix_tree_shrink(struct radix_tree_root *root) | |
1157 | { | |
1158 | /* try to shrink tree height */ | |
c0bc9875 | 1159 | while (root->height > 0) { |
a5f51c96 | 1160 | struct radix_tree_node *to_free = root->rnode; |
7cf9c2c7 | 1161 | void *newptr; |
a5f51c96 | 1162 | |
c0bc9875 NP |
1163 | BUG_ON(!radix_tree_is_indirect_ptr(to_free)); |
1164 | to_free = radix_tree_indirect_to_ptr(to_free); | |
1165 | ||
1166 | /* | |
1167 | * The candidate node has more than one child, or its child | |
1168 | * is not at the leftmost slot, we cannot shrink. | |
1169 | */ | |
1170 | if (to_free->count != 1) | |
1171 | break; | |
1172 | if (!to_free->slots[0]) | |
1173 | break; | |
1174 | ||
7cf9c2c7 NP |
1175 | /* |
1176 | * We don't need rcu_assign_pointer(), since we are simply | |
1177 | * moving the node from one part of the tree to another. If | |
1178 | * it was safe to dereference the old pointer to it | |
1179 | * (to_free->slots[0]), it will be safe to dereference the new | |
1180 | * one (root->rnode). | |
1181 | */ | |
1182 | newptr = to_free->slots[0]; | |
c0bc9875 NP |
1183 | if (root->height > 1) |
1184 | newptr = radix_tree_ptr_to_indirect(newptr); | |
7cf9c2c7 | 1185 | root->rnode = newptr; |
a5f51c96 | 1186 | root->height--; |
a5f51c96 NP |
1187 | radix_tree_node_free(to_free); |
1188 | } | |
1189 | } | |
1190 | ||
1da177e4 LT |
1191 | /** |
1192 | * radix_tree_delete - delete an item from a radix tree | |
1193 | * @root: radix tree root | |
1194 | * @index: index key | |
1195 | * | |
1196 | * Remove the item at @index from the radix tree rooted at @root. | |
1197 | * | |
1198 | * Returns the address of the deleted item, or NULL if it was not present. | |
1199 | */ | |
1200 | void *radix_tree_delete(struct radix_tree_root *root, unsigned long index) | |
1201 | { | |
26fb1589 JM |
1202 | /* |
1203 | * The radix tree path needs to be one longer than the maximum path | |
1204 | * since the "list" is null terminated. | |
1205 | */ | |
1206 | struct radix_tree_path path[RADIX_TREE_MAX_PATH + 1], *pathp = path; | |
612d6c19 | 1207 | struct radix_tree_node *slot = NULL; |
7cf9c2c7 | 1208 | struct radix_tree_node *to_free; |
1da177e4 | 1209 | unsigned int height, shift; |
d5274261 NP |
1210 | int tag; |
1211 | int offset; | |
1da177e4 LT |
1212 | |
1213 | height = root->height; | |
1214 | if (index > radix_tree_maxindex(height)) | |
1215 | goto out; | |
1216 | ||
612d6c19 | 1217 | slot = root->rnode; |
c0bc9875 | 1218 | if (height == 0) { |
612d6c19 NP |
1219 | root_tag_clear_all(root); |
1220 | root->rnode = NULL; | |
1221 | goto out; | |
1222 | } | |
c0bc9875 | 1223 | slot = radix_tree_indirect_to_ptr(slot); |
612d6c19 | 1224 | |
1da177e4 LT |
1225 | shift = (height - 1) * RADIX_TREE_MAP_SHIFT; |
1226 | pathp->node = NULL; | |
1da177e4 | 1227 | |
612d6c19 | 1228 | do { |
201b6264 | 1229 | if (slot == NULL) |
1da177e4 LT |
1230 | goto out; |
1231 | ||
d5274261 | 1232 | pathp++; |
1da177e4 | 1233 | offset = (index >> shift) & RADIX_TREE_MAP_MASK; |
d5274261 NP |
1234 | pathp->offset = offset; |
1235 | pathp->node = slot; | |
201b6264 | 1236 | slot = slot->slots[offset]; |
1da177e4 | 1237 | shift -= RADIX_TREE_MAP_SHIFT; |
612d6c19 NP |
1238 | height--; |
1239 | } while (height > 0); | |
1da177e4 | 1240 | |
612d6c19 | 1241 | if (slot == NULL) |
1da177e4 LT |
1242 | goto out; |
1243 | ||
1da177e4 LT |
1244 | /* |
1245 | * Clear all tags associated with the just-deleted item | |
1246 | */ | |
daff89f3 | 1247 | for (tag = 0; tag < RADIX_TREE_MAX_TAGS; tag++) { |
612d6c19 NP |
1248 | if (tag_get(pathp->node, tag, pathp->offset)) |
1249 | radix_tree_tag_clear(root, index, tag); | |
d5274261 | 1250 | } |
1da177e4 | 1251 | |
7cf9c2c7 | 1252 | to_free = NULL; |
201b6264 | 1253 | /* Now free the nodes we do not need anymore */ |
612d6c19 | 1254 | while (pathp->node) { |
201b6264 | 1255 | pathp->node->slots[pathp->offset] = NULL; |
a5f51c96 | 1256 | pathp->node->count--; |
7cf9c2c7 NP |
1257 | /* |
1258 | * Queue the node for deferred freeing after the | |
1259 | * last reference to it disappears (set NULL, above). | |
1260 | */ | |
1261 | if (to_free) | |
1262 | radix_tree_node_free(to_free); | |
a5f51c96 NP |
1263 | |
1264 | if (pathp->node->count) { | |
c0bc9875 NP |
1265 | if (pathp->node == |
1266 | radix_tree_indirect_to_ptr(root->rnode)) | |
a5f51c96 | 1267 | radix_tree_shrink(root); |
201b6264 | 1268 | goto out; |
a5f51c96 | 1269 | } |
201b6264 CL |
1270 | |
1271 | /* Node with zero slots in use so free it */ | |
7cf9c2c7 | 1272 | to_free = pathp->node; |
612d6c19 | 1273 | pathp--; |
7cf9c2c7 | 1274 | |
1da177e4 | 1275 | } |
612d6c19 | 1276 | root_tag_clear_all(root); |
201b6264 | 1277 | root->height = 0; |
612d6c19 | 1278 | root->rnode = NULL; |
7cf9c2c7 NP |
1279 | if (to_free) |
1280 | radix_tree_node_free(to_free); | |
612d6c19 | 1281 | |
1da177e4 | 1282 | out: |
612d6c19 | 1283 | return slot; |
1da177e4 LT |
1284 | } |
1285 | EXPORT_SYMBOL(radix_tree_delete); | |
1286 | ||
1287 | /** | |
1288 | * radix_tree_tagged - test whether any items in the tree are tagged | |
1289 | * @root: radix tree root | |
1290 | * @tag: tag to test | |
1291 | */ | |
daff89f3 | 1292 | int radix_tree_tagged(struct radix_tree_root *root, unsigned int tag) |
1da177e4 | 1293 | { |
612d6c19 | 1294 | return root_tag_get(root, tag); |
1da177e4 LT |
1295 | } |
1296 | EXPORT_SYMBOL(radix_tree_tagged); | |
1297 | ||
1298 | static void | |
51cc5068 | 1299 | radix_tree_node_ctor(void *node) |
1da177e4 LT |
1300 | { |
1301 | memset(node, 0, sizeof(struct radix_tree_node)); | |
1302 | } | |
1303 | ||
1304 | static __init unsigned long __maxindex(unsigned int height) | |
1305 | { | |
430d275a PL |
1306 | unsigned int width = height * RADIX_TREE_MAP_SHIFT; |
1307 | int shift = RADIX_TREE_INDEX_BITS - width; | |
1308 | ||
1309 | if (shift < 0) | |
1310 | return ~0UL; | |
1311 | if (shift >= BITS_PER_LONG) | |
1312 | return 0UL; | |
1313 | return ~0UL >> shift; | |
1da177e4 LT |
1314 | } |
1315 | ||
1316 | static __init void radix_tree_init_maxindex(void) | |
1317 | { | |
1318 | unsigned int i; | |
1319 | ||
1320 | for (i = 0; i < ARRAY_SIZE(height_to_maxindex); i++) | |
1321 | height_to_maxindex[i] = __maxindex(i); | |
1322 | } | |
1323 | ||
1da177e4 LT |
1324 | static int radix_tree_callback(struct notifier_block *nfb, |
1325 | unsigned long action, | |
1326 | void *hcpu) | |
1327 | { | |
1328 | int cpu = (long)hcpu; | |
1329 | struct radix_tree_preload *rtp; | |
1330 | ||
1331 | /* Free per-cpu pool of perloaded nodes */ | |
8bb78442 | 1332 | if (action == CPU_DEAD || action == CPU_DEAD_FROZEN) { |
1da177e4 LT |
1333 | rtp = &per_cpu(radix_tree_preloads, cpu); |
1334 | while (rtp->nr) { | |
1335 | kmem_cache_free(radix_tree_node_cachep, | |
1336 | rtp->nodes[rtp->nr-1]); | |
1337 | rtp->nodes[rtp->nr-1] = NULL; | |
1338 | rtp->nr--; | |
1339 | } | |
1340 | } | |
1341 | return NOTIFY_OK; | |
1342 | } | |
1da177e4 LT |
1343 | |
1344 | void __init radix_tree_init(void) | |
1345 | { | |
1346 | radix_tree_node_cachep = kmem_cache_create("radix_tree_node", | |
1347 | sizeof(struct radix_tree_node), 0, | |
488514d1 CL |
1348 | SLAB_PANIC | SLAB_RECLAIM_ACCOUNT, |
1349 | radix_tree_node_ctor); | |
1da177e4 LT |
1350 | radix_tree_init_maxindex(); |
1351 | hotcpu_notifier(radix_tree_callback, 0); | |
1352 | } |