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; |
a1115570 | 52 | void __rcu *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 | |
27d20fdd NP |
85 | static inline void *ptr_to_indirect(void *ptr) |
86 | { | |
87 | return (void *)((unsigned long)ptr | RADIX_TREE_INDIRECT_PTR); | |
88 | } | |
89 | ||
90 | static inline void *indirect_to_ptr(void *ptr) | |
91 | { | |
92 | return (void *)((unsigned long)ptr & ~RADIX_TREE_INDIRECT_PTR); | |
93 | } | |
94 | ||
612d6c19 NP |
95 | static inline gfp_t root_gfp_mask(struct radix_tree_root *root) |
96 | { | |
97 | return root->gfp_mask & __GFP_BITS_MASK; | |
98 | } | |
99 | ||
643b52b9 NP |
100 | static inline void tag_set(struct radix_tree_node *node, unsigned int tag, |
101 | int offset) | |
102 | { | |
103 | __set_bit(offset, node->tags[tag]); | |
104 | } | |
105 | ||
106 | static inline void tag_clear(struct radix_tree_node *node, unsigned int tag, | |
107 | int offset) | |
108 | { | |
109 | __clear_bit(offset, node->tags[tag]); | |
110 | } | |
111 | ||
112 | static inline int tag_get(struct radix_tree_node *node, unsigned int tag, | |
113 | int offset) | |
114 | { | |
115 | return test_bit(offset, node->tags[tag]); | |
116 | } | |
117 | ||
118 | static inline void root_tag_set(struct radix_tree_root *root, unsigned int tag) | |
119 | { | |
120 | root->gfp_mask |= (__force gfp_t)(1 << (tag + __GFP_BITS_SHIFT)); | |
121 | } | |
122 | ||
123 | static inline void root_tag_clear(struct radix_tree_root *root, unsigned int tag) | |
124 | { | |
125 | root->gfp_mask &= (__force gfp_t)~(1 << (tag + __GFP_BITS_SHIFT)); | |
126 | } | |
127 | ||
128 | static inline void root_tag_clear_all(struct radix_tree_root *root) | |
129 | { | |
130 | root->gfp_mask &= __GFP_BITS_MASK; | |
131 | } | |
132 | ||
133 | static inline int root_tag_get(struct radix_tree_root *root, unsigned int tag) | |
134 | { | |
135 | return (__force unsigned)root->gfp_mask & (1 << (tag + __GFP_BITS_SHIFT)); | |
136 | } | |
137 | ||
138 | /* | |
139 | * Returns 1 if any slot in the node has this tag set. | |
140 | * Otherwise returns 0. | |
141 | */ | |
142 | static inline int any_tag_set(struct radix_tree_node *node, unsigned int tag) | |
143 | { | |
144 | int idx; | |
145 | for (idx = 0; idx < RADIX_TREE_TAG_LONGS; idx++) { | |
146 | if (node->tags[tag][idx]) | |
147 | return 1; | |
148 | } | |
149 | return 0; | |
150 | } | |
1da177e4 LT |
151 | /* |
152 | * This assumes that the caller has performed appropriate preallocation, and | |
153 | * that the caller has pinned this thread of control to the current CPU. | |
154 | */ | |
155 | static struct radix_tree_node * | |
156 | radix_tree_node_alloc(struct radix_tree_root *root) | |
157 | { | |
e2848a0e | 158 | struct radix_tree_node *ret = NULL; |
612d6c19 | 159 | gfp_t gfp_mask = root_gfp_mask(root); |
1da177e4 | 160 | |
e2848a0e | 161 | if (!(gfp_mask & __GFP_WAIT)) { |
1da177e4 LT |
162 | struct radix_tree_preload *rtp; |
163 | ||
e2848a0e NP |
164 | /* |
165 | * Provided the caller has preloaded here, we will always | |
166 | * succeed in getting a node here (and never reach | |
167 | * kmem_cache_alloc) | |
168 | */ | |
1da177e4 LT |
169 | rtp = &__get_cpu_var(radix_tree_preloads); |
170 | if (rtp->nr) { | |
171 | ret = rtp->nodes[rtp->nr - 1]; | |
172 | rtp->nodes[rtp->nr - 1] = NULL; | |
173 | rtp->nr--; | |
174 | } | |
175 | } | |
e2848a0e | 176 | if (ret == NULL) |
488514d1 | 177 | ret = kmem_cache_alloc(radix_tree_node_cachep, gfp_mask); |
e2848a0e | 178 | |
c0bc9875 | 179 | BUG_ON(radix_tree_is_indirect_ptr(ret)); |
1da177e4 LT |
180 | return ret; |
181 | } | |
182 | ||
7cf9c2c7 NP |
183 | static void radix_tree_node_rcu_free(struct rcu_head *head) |
184 | { | |
185 | struct radix_tree_node *node = | |
186 | container_of(head, struct radix_tree_node, rcu_head); | |
b6dd0865 | 187 | int i; |
643b52b9 NP |
188 | |
189 | /* | |
190 | * must only free zeroed nodes into the slab. radix_tree_shrink | |
191 | * can leave us with a non-NULL entry in the first slot, so clear | |
192 | * that here to make sure. | |
193 | */ | |
b6dd0865 DC |
194 | for (i = 0; i < RADIX_TREE_MAX_TAGS; i++) |
195 | tag_clear(node, i, 0); | |
196 | ||
643b52b9 NP |
197 | node->slots[0] = NULL; |
198 | node->count = 0; | |
199 | ||
7cf9c2c7 NP |
200 | kmem_cache_free(radix_tree_node_cachep, node); |
201 | } | |
202 | ||
1da177e4 LT |
203 | static inline void |
204 | radix_tree_node_free(struct radix_tree_node *node) | |
205 | { | |
7cf9c2c7 | 206 | call_rcu(&node->rcu_head, radix_tree_node_rcu_free); |
1da177e4 LT |
207 | } |
208 | ||
209 | /* | |
210 | * Load up this CPU's radix_tree_node buffer with sufficient objects to | |
211 | * ensure that the addition of a single element in the tree cannot fail. On | |
212 | * success, return zero, with preemption disabled. On error, return -ENOMEM | |
213 | * with preemption not disabled. | |
b34df792 DH |
214 | * |
215 | * To make use of this facility, the radix tree must be initialised without | |
216 | * __GFP_WAIT being passed to INIT_RADIX_TREE(). | |
1da177e4 | 217 | */ |
dd0fc66f | 218 | int radix_tree_preload(gfp_t gfp_mask) |
1da177e4 LT |
219 | { |
220 | struct radix_tree_preload *rtp; | |
221 | struct radix_tree_node *node; | |
222 | int ret = -ENOMEM; | |
223 | ||
224 | preempt_disable(); | |
225 | rtp = &__get_cpu_var(radix_tree_preloads); | |
226 | while (rtp->nr < ARRAY_SIZE(rtp->nodes)) { | |
227 | preempt_enable(); | |
488514d1 | 228 | node = kmem_cache_alloc(radix_tree_node_cachep, gfp_mask); |
1da177e4 LT |
229 | if (node == NULL) |
230 | goto out; | |
231 | preempt_disable(); | |
232 | rtp = &__get_cpu_var(radix_tree_preloads); | |
233 | if (rtp->nr < ARRAY_SIZE(rtp->nodes)) | |
234 | rtp->nodes[rtp->nr++] = node; | |
235 | else | |
236 | kmem_cache_free(radix_tree_node_cachep, node); | |
237 | } | |
238 | ret = 0; | |
239 | out: | |
240 | return ret; | |
241 | } | |
d7f0923d | 242 | EXPORT_SYMBOL(radix_tree_preload); |
1da177e4 | 243 | |
1da177e4 LT |
244 | /* |
245 | * Return the maximum key which can be store into a | |
246 | * radix tree with height HEIGHT. | |
247 | */ | |
248 | static inline unsigned long radix_tree_maxindex(unsigned int height) | |
249 | { | |
250 | return height_to_maxindex[height]; | |
251 | } | |
252 | ||
253 | /* | |
254 | * Extend a radix tree so it can store key @index. | |
255 | */ | |
256 | static int radix_tree_extend(struct radix_tree_root *root, unsigned long index) | |
257 | { | |
258 | struct radix_tree_node *node; | |
259 | unsigned int height; | |
1da177e4 LT |
260 | int tag; |
261 | ||
262 | /* Figure out what the height should be. */ | |
263 | height = root->height + 1; | |
264 | while (index > radix_tree_maxindex(height)) | |
265 | height++; | |
266 | ||
267 | if (root->rnode == NULL) { | |
268 | root->height = height; | |
269 | goto out; | |
270 | } | |
271 | ||
1da177e4 | 272 | do { |
7cf9c2c7 | 273 | unsigned int newheight; |
1da177e4 LT |
274 | if (!(node = radix_tree_node_alloc(root))) |
275 | return -ENOMEM; | |
276 | ||
277 | /* Increase the height. */ | |
27d20fdd | 278 | node->slots[0] = indirect_to_ptr(root->rnode); |
1da177e4 LT |
279 | |
280 | /* Propagate the aggregated tag info into the new root */ | |
daff89f3 | 281 | for (tag = 0; tag < RADIX_TREE_MAX_TAGS; tag++) { |
612d6c19 | 282 | if (root_tag_get(root, tag)) |
1da177e4 LT |
283 | tag_set(node, tag, 0); |
284 | } | |
285 | ||
7cf9c2c7 NP |
286 | newheight = root->height+1; |
287 | node->height = newheight; | |
1da177e4 | 288 | node->count = 1; |
27d20fdd | 289 | node = ptr_to_indirect(node); |
7cf9c2c7 NP |
290 | rcu_assign_pointer(root->rnode, node); |
291 | root->height = newheight; | |
1da177e4 LT |
292 | } while (height > root->height); |
293 | out: | |
294 | return 0; | |
295 | } | |
296 | ||
297 | /** | |
298 | * radix_tree_insert - insert into a radix tree | |
299 | * @root: radix tree root | |
300 | * @index: index key | |
301 | * @item: item to insert | |
302 | * | |
303 | * Insert an item into the radix tree at position @index. | |
304 | */ | |
305 | int radix_tree_insert(struct radix_tree_root *root, | |
306 | unsigned long index, void *item) | |
307 | { | |
201b6264 | 308 | struct radix_tree_node *node = NULL, *slot; |
1da177e4 LT |
309 | unsigned int height, shift; |
310 | int offset; | |
311 | int error; | |
312 | ||
c0bc9875 | 313 | BUG_ON(radix_tree_is_indirect_ptr(item)); |
7cf9c2c7 | 314 | |
1da177e4 | 315 | /* Make sure the tree is high enough. */ |
612d6c19 | 316 | if (index > radix_tree_maxindex(root->height)) { |
1da177e4 LT |
317 | error = radix_tree_extend(root, index); |
318 | if (error) | |
319 | return error; | |
320 | } | |
321 | ||
27d20fdd | 322 | slot = indirect_to_ptr(root->rnode); |
c0bc9875 | 323 | |
1da177e4 LT |
324 | height = root->height; |
325 | shift = (height-1) * RADIX_TREE_MAP_SHIFT; | |
326 | ||
327 | offset = 0; /* uninitialised var warning */ | |
612d6c19 | 328 | while (height > 0) { |
201b6264 | 329 | if (slot == NULL) { |
1da177e4 | 330 | /* Have to add a child node. */ |
201b6264 | 331 | if (!(slot = radix_tree_node_alloc(root))) |
1da177e4 | 332 | return -ENOMEM; |
7cf9c2c7 | 333 | slot->height = height; |
201b6264 | 334 | if (node) { |
7cf9c2c7 | 335 | rcu_assign_pointer(node->slots[offset], slot); |
1da177e4 | 336 | node->count++; |
201b6264 | 337 | } else |
27d20fdd | 338 | rcu_assign_pointer(root->rnode, ptr_to_indirect(slot)); |
1da177e4 LT |
339 | } |
340 | ||
341 | /* Go a level down */ | |
342 | offset = (index >> shift) & RADIX_TREE_MAP_MASK; | |
201b6264 CL |
343 | node = slot; |
344 | slot = node->slots[offset]; | |
1da177e4 LT |
345 | shift -= RADIX_TREE_MAP_SHIFT; |
346 | height--; | |
612d6c19 | 347 | } |
1da177e4 | 348 | |
201b6264 | 349 | if (slot != NULL) |
1da177e4 | 350 | return -EEXIST; |
201b6264 | 351 | |
612d6c19 NP |
352 | if (node) { |
353 | node->count++; | |
7cf9c2c7 | 354 | rcu_assign_pointer(node->slots[offset], item); |
612d6c19 NP |
355 | BUG_ON(tag_get(node, 0, offset)); |
356 | BUG_ON(tag_get(node, 1, offset)); | |
357 | } else { | |
c0bc9875 | 358 | rcu_assign_pointer(root->rnode, item); |
612d6c19 NP |
359 | BUG_ON(root_tag_get(root, 0)); |
360 | BUG_ON(root_tag_get(root, 1)); | |
361 | } | |
1da177e4 | 362 | |
1da177e4 LT |
363 | return 0; |
364 | } | |
365 | EXPORT_SYMBOL(radix_tree_insert); | |
366 | ||
b72b71c6 HS |
367 | /* |
368 | * is_slot == 1 : search for the slot. | |
369 | * is_slot == 0 : search for the node. | |
7cf9c2c7 | 370 | */ |
b72b71c6 HS |
371 | static void *radix_tree_lookup_element(struct radix_tree_root *root, |
372 | unsigned long index, int is_slot) | |
1da177e4 LT |
373 | { |
374 | unsigned int height, shift; | |
7cf9c2c7 | 375 | struct radix_tree_node *node, **slot; |
612d6c19 | 376 | |
2676a58c | 377 | node = rcu_dereference_raw(root->rnode); |
7cf9c2c7 | 378 | if (node == NULL) |
1da177e4 LT |
379 | return NULL; |
380 | ||
c0bc9875 | 381 | if (!radix_tree_is_indirect_ptr(node)) { |
7cf9c2c7 NP |
382 | if (index > 0) |
383 | return NULL; | |
b72b71c6 | 384 | return is_slot ? (void *)&root->rnode : node; |
7cf9c2c7 | 385 | } |
27d20fdd | 386 | node = indirect_to_ptr(node); |
7cf9c2c7 NP |
387 | |
388 | height = node->height; | |
389 | if (index > radix_tree_maxindex(height)) | |
390 | return NULL; | |
612d6c19 | 391 | |
1da177e4 | 392 | shift = (height-1) * RADIX_TREE_MAP_SHIFT; |
1da177e4 | 393 | |
7cf9c2c7 NP |
394 | do { |
395 | slot = (struct radix_tree_node **) | |
396 | (node->slots + ((index>>shift) & RADIX_TREE_MAP_MASK)); | |
2676a58c | 397 | node = rcu_dereference_raw(*slot); |
7cf9c2c7 | 398 | if (node == NULL) |
1da177e4 LT |
399 | return NULL; |
400 | ||
1da177e4 LT |
401 | shift -= RADIX_TREE_MAP_SHIFT; |
402 | height--; | |
7cf9c2c7 | 403 | } while (height > 0); |
1da177e4 | 404 | |
27d20fdd | 405 | return is_slot ? (void *)slot : indirect_to_ptr(node); |
b72b71c6 HS |
406 | } |
407 | ||
408 | /** | |
409 | * radix_tree_lookup_slot - lookup a slot in a radix tree | |
410 | * @root: radix tree root | |
411 | * @index: index key | |
412 | * | |
413 | * Returns: the slot corresponding to the position @index in the | |
414 | * radix tree @root. This is useful for update-if-exists operations. | |
415 | * | |
416 | * This function can be called under rcu_read_lock iff the slot is not | |
417 | * modified by radix_tree_replace_slot, otherwise it must be called | |
418 | * exclusive from other writers. Any dereference of the slot must be done | |
419 | * using radix_tree_deref_slot. | |
420 | */ | |
421 | void **radix_tree_lookup_slot(struct radix_tree_root *root, unsigned long index) | |
422 | { | |
423 | return (void **)radix_tree_lookup_element(root, index, 1); | |
a4331366 | 424 | } |
a4331366 HR |
425 | EXPORT_SYMBOL(radix_tree_lookup_slot); |
426 | ||
427 | /** | |
428 | * radix_tree_lookup - perform lookup operation on a radix tree | |
429 | * @root: radix tree root | |
430 | * @index: index key | |
431 | * | |
432 | * Lookup the item at the position @index in the radix tree @root. | |
7cf9c2c7 NP |
433 | * |
434 | * This function can be called under rcu_read_lock, however the caller | |
435 | * must manage lifetimes of leaf nodes (eg. RCU may also be used to free | |
436 | * them safely). No RCU barriers are required to access or modify the | |
437 | * returned item, however. | |
a4331366 HR |
438 | */ |
439 | void *radix_tree_lookup(struct radix_tree_root *root, unsigned long index) | |
440 | { | |
b72b71c6 | 441 | return radix_tree_lookup_element(root, index, 0); |
1da177e4 LT |
442 | } |
443 | EXPORT_SYMBOL(radix_tree_lookup); | |
444 | ||
445 | /** | |
446 | * radix_tree_tag_set - set a tag on a radix tree node | |
447 | * @root: radix tree root | |
448 | * @index: index key | |
449 | * @tag: tag index | |
450 | * | |
daff89f3 JC |
451 | * Set the search tag (which must be < RADIX_TREE_MAX_TAGS) |
452 | * corresponding to @index in the radix tree. From | |
1da177e4 LT |
453 | * the root all the way down to the leaf node. |
454 | * | |
455 | * Returns the address of the tagged item. Setting a tag on a not-present | |
456 | * item is a bug. | |
457 | */ | |
458 | void *radix_tree_tag_set(struct radix_tree_root *root, | |
daff89f3 | 459 | unsigned long index, unsigned int tag) |
1da177e4 LT |
460 | { |
461 | unsigned int height, shift; | |
201b6264 | 462 | struct radix_tree_node *slot; |
1da177e4 LT |
463 | |
464 | height = root->height; | |
4c91c364 | 465 | BUG_ON(index > radix_tree_maxindex(height)); |
1da177e4 | 466 | |
27d20fdd | 467 | slot = indirect_to_ptr(root->rnode); |
612d6c19 | 468 | shift = (height - 1) * RADIX_TREE_MAP_SHIFT; |
1da177e4 LT |
469 | |
470 | while (height > 0) { | |
471 | int offset; | |
472 | ||
473 | offset = (index >> shift) & RADIX_TREE_MAP_MASK; | |
d5274261 NP |
474 | if (!tag_get(slot, tag, offset)) |
475 | tag_set(slot, tag, offset); | |
201b6264 CL |
476 | slot = slot->slots[offset]; |
477 | BUG_ON(slot == NULL); | |
1da177e4 LT |
478 | shift -= RADIX_TREE_MAP_SHIFT; |
479 | height--; | |
480 | } | |
481 | ||
612d6c19 NP |
482 | /* set the root's tag bit */ |
483 | if (slot && !root_tag_get(root, tag)) | |
484 | root_tag_set(root, tag); | |
485 | ||
201b6264 | 486 | return slot; |
1da177e4 LT |
487 | } |
488 | EXPORT_SYMBOL(radix_tree_tag_set); | |
489 | ||
490 | /** | |
491 | * radix_tree_tag_clear - clear a tag on a radix tree node | |
492 | * @root: radix tree root | |
493 | * @index: index key | |
494 | * @tag: tag index | |
495 | * | |
daff89f3 JC |
496 | * Clear the search tag (which must be < RADIX_TREE_MAX_TAGS) |
497 | * corresponding to @index in the radix tree. If | |
1da177e4 LT |
498 | * this causes the leaf node to have no tags set then clear the tag in the |
499 | * next-to-leaf node, etc. | |
500 | * | |
501 | * Returns the address of the tagged item on success, else NULL. ie: | |
502 | * has the same return value and semantics as radix_tree_lookup(). | |
503 | */ | |
504 | void *radix_tree_tag_clear(struct radix_tree_root *root, | |
daff89f3 | 505 | unsigned long index, unsigned int tag) |
1da177e4 | 506 | { |
26fb1589 JM |
507 | /* |
508 | * The radix tree path needs to be one longer than the maximum path | |
509 | * since the "list" is null terminated. | |
510 | */ | |
511 | struct radix_tree_path path[RADIX_TREE_MAX_PATH + 1], *pathp = path; | |
612d6c19 | 512 | struct radix_tree_node *slot = NULL; |
1da177e4 | 513 | unsigned int height, shift; |
1da177e4 LT |
514 | |
515 | height = root->height; | |
516 | if (index > radix_tree_maxindex(height)) | |
517 | goto out; | |
518 | ||
519 | shift = (height - 1) * RADIX_TREE_MAP_SHIFT; | |
520 | pathp->node = NULL; | |
27d20fdd | 521 | slot = indirect_to_ptr(root->rnode); |
1da177e4 LT |
522 | |
523 | while (height > 0) { | |
524 | int offset; | |
525 | ||
201b6264 | 526 | if (slot == NULL) |
1da177e4 LT |
527 | goto out; |
528 | ||
529 | offset = (index >> shift) & RADIX_TREE_MAP_MASK; | |
530 | pathp[1].offset = offset; | |
201b6264 CL |
531 | pathp[1].node = slot; |
532 | slot = slot->slots[offset]; | |
1da177e4 LT |
533 | pathp++; |
534 | shift -= RADIX_TREE_MAP_SHIFT; | |
535 | height--; | |
536 | } | |
537 | ||
612d6c19 | 538 | if (slot == NULL) |
1da177e4 LT |
539 | goto out; |
540 | ||
612d6c19 | 541 | while (pathp->node) { |
d5274261 NP |
542 | if (!tag_get(pathp->node, tag, pathp->offset)) |
543 | goto out; | |
201b6264 | 544 | tag_clear(pathp->node, tag, pathp->offset); |
6e954b9e NP |
545 | if (any_tag_set(pathp->node, tag)) |
546 | goto out; | |
1da177e4 | 547 | pathp--; |
612d6c19 NP |
548 | } |
549 | ||
550 | /* clear the root's tag bit */ | |
551 | if (root_tag_get(root, tag)) | |
552 | root_tag_clear(root, tag); | |
553 | ||
1da177e4 | 554 | out: |
612d6c19 | 555 | return slot; |
1da177e4 LT |
556 | } |
557 | EXPORT_SYMBOL(radix_tree_tag_clear); | |
558 | ||
1da177e4 | 559 | /** |
32605a18 MT |
560 | * radix_tree_tag_get - get a tag on a radix tree node |
561 | * @root: radix tree root | |
562 | * @index: index key | |
daff89f3 | 563 | * @tag: tag index (< RADIX_TREE_MAX_TAGS) |
1da177e4 | 564 | * |
32605a18 | 565 | * Return values: |
1da177e4 | 566 | * |
612d6c19 NP |
567 | * 0: tag not present or not set |
568 | * 1: tag set | |
ce82653d DH |
569 | * |
570 | * Note that the return value of this function may not be relied on, even if | |
571 | * the RCU lock is held, unless tag modification and node deletion are excluded | |
572 | * from concurrency. | |
1da177e4 LT |
573 | */ |
574 | int radix_tree_tag_get(struct radix_tree_root *root, | |
daff89f3 | 575 | unsigned long index, unsigned int tag) |
1da177e4 LT |
576 | { |
577 | unsigned int height, shift; | |
7cf9c2c7 | 578 | struct radix_tree_node *node; |
1da177e4 | 579 | |
612d6c19 NP |
580 | /* check the root's tag bit */ |
581 | if (!root_tag_get(root, tag)) | |
582 | return 0; | |
583 | ||
2676a58c | 584 | node = rcu_dereference_raw(root->rnode); |
7cf9c2c7 NP |
585 | if (node == NULL) |
586 | return 0; | |
587 | ||
c0bc9875 | 588 | if (!radix_tree_is_indirect_ptr(node)) |
7cf9c2c7 | 589 | return (index == 0); |
27d20fdd | 590 | node = indirect_to_ptr(node); |
7cf9c2c7 NP |
591 | |
592 | height = node->height; | |
593 | if (index > radix_tree_maxindex(height)) | |
594 | return 0; | |
612d6c19 | 595 | |
1da177e4 | 596 | shift = (height - 1) * RADIX_TREE_MAP_SHIFT; |
1da177e4 LT |
597 | |
598 | for ( ; ; ) { | |
599 | int offset; | |
600 | ||
7cf9c2c7 | 601 | if (node == NULL) |
1da177e4 LT |
602 | return 0; |
603 | ||
604 | offset = (index >> shift) & RADIX_TREE_MAP_MASK; | |
7cf9c2c7 | 605 | if (!tag_get(node, tag, offset)) |
3fa36acb | 606 | return 0; |
ce82653d | 607 | if (height == 1) |
3fa36acb | 608 | return 1; |
2676a58c | 609 | node = rcu_dereference_raw(node->slots[offset]); |
1da177e4 LT |
610 | shift -= RADIX_TREE_MAP_SHIFT; |
611 | height--; | |
612 | } | |
613 | } | |
614 | EXPORT_SYMBOL(radix_tree_tag_get); | |
1da177e4 | 615 | |
ebf8aa44 JK |
616 | /** |
617 | * radix_tree_range_tag_if_tagged - for each item in given range set given | |
618 | * tag if item has another tag set | |
619 | * @root: radix tree root | |
620 | * @first_indexp: pointer to a starting index of a range to scan | |
621 | * @last_index: last index of a range to scan | |
622 | * @nr_to_tag: maximum number items to tag | |
623 | * @iftag: tag index to test | |
624 | * @settag: tag index to set if tested tag is set | |
625 | * | |
626 | * This function scans range of radix tree from first_index to last_index | |
627 | * (inclusive). For each item in the range if iftag is set, the function sets | |
628 | * also settag. The function stops either after tagging nr_to_tag items or | |
629 | * after reaching last_index. | |
630 | * | |
144dcfc0 DC |
631 | * The tags must be set from the leaf level only and propagated back up the |
632 | * path to the root. We must do this so that we resolve the full path before | |
633 | * setting any tags on intermediate nodes. If we set tags as we descend, then | |
634 | * we can get to the leaf node and find that the index that has the iftag | |
635 | * set is outside the range we are scanning. This reults in dangling tags and | |
636 | * can lead to problems with later tag operations (e.g. livelocks on lookups). | |
637 | * | |
ebf8aa44 JK |
638 | * The function returns number of leaves where the tag was set and sets |
639 | * *first_indexp to the first unscanned index. | |
d5ed3a4a JK |
640 | * WARNING! *first_indexp can wrap if last_index is ULONG_MAX. Caller must |
641 | * be prepared to handle that. | |
ebf8aa44 JK |
642 | */ |
643 | unsigned long radix_tree_range_tag_if_tagged(struct radix_tree_root *root, | |
644 | unsigned long *first_indexp, unsigned long last_index, | |
645 | unsigned long nr_to_tag, | |
646 | unsigned int iftag, unsigned int settag) | |
647 | { | |
144dcfc0 DC |
648 | unsigned int height = root->height; |
649 | struct radix_tree_path path[height]; | |
650 | struct radix_tree_path *pathp = path; | |
651 | struct radix_tree_node *slot; | |
652 | unsigned int shift; | |
653 | unsigned long tagged = 0; | |
654 | unsigned long index = *first_indexp; | |
ebf8aa44 JK |
655 | |
656 | last_index = min(last_index, radix_tree_maxindex(height)); | |
657 | if (index > last_index) | |
658 | return 0; | |
659 | if (!nr_to_tag) | |
660 | return 0; | |
661 | if (!root_tag_get(root, iftag)) { | |
662 | *first_indexp = last_index + 1; | |
663 | return 0; | |
664 | } | |
665 | if (height == 0) { | |
666 | *first_indexp = last_index + 1; | |
667 | root_tag_set(root, settag); | |
668 | return 1; | |
669 | } | |
670 | ||
671 | shift = (height - 1) * RADIX_TREE_MAP_SHIFT; | |
27d20fdd | 672 | slot = indirect_to_ptr(root->rnode); |
ebf8aa44 | 673 | |
144dcfc0 DC |
674 | /* |
675 | * we fill the path from (root->height - 2) to 0, leaving the index at | |
676 | * (root->height - 1) as a terminator. Zero the node in the terminator | |
677 | * so that we can use this to end walk loops back up the path. | |
678 | */ | |
679 | path[height - 1].node = NULL; | |
680 | ||
ebf8aa44 JK |
681 | for (;;) { |
682 | int offset; | |
683 | ||
684 | offset = (index >> shift) & RADIX_TREE_MAP_MASK; | |
685 | if (!slot->slots[offset]) | |
686 | goto next; | |
687 | if (!tag_get(slot, iftag, offset)) | |
688 | goto next; | |
144dcfc0 DC |
689 | if (height > 1) { |
690 | /* Go down one level */ | |
691 | height--; | |
692 | shift -= RADIX_TREE_MAP_SHIFT; | |
693 | path[height - 1].node = slot; | |
694 | path[height - 1].offset = offset; | |
695 | slot = slot->slots[offset]; | |
696 | continue; | |
697 | } | |
698 | ||
699 | /* tag the leaf */ | |
700 | tagged++; | |
ebf8aa44 | 701 | tag_set(slot, settag, offset); |
144dcfc0 DC |
702 | |
703 | /* walk back up the path tagging interior nodes */ | |
704 | pathp = &path[0]; | |
705 | while (pathp->node) { | |
706 | /* stop if we find a node with the tag already set */ | |
707 | if (tag_get(pathp->node, settag, pathp->offset)) | |
708 | break; | |
709 | tag_set(pathp->node, settag, pathp->offset); | |
710 | pathp++; | |
ebf8aa44 | 711 | } |
144dcfc0 | 712 | |
ebf8aa44 JK |
713 | next: |
714 | /* Go to next item at level determined by 'shift' */ | |
715 | index = ((index >> shift) + 1) << shift; | |
d5ed3a4a JK |
716 | /* Overflow can happen when last_index is ~0UL... */ |
717 | if (index > last_index || !index) | |
ebf8aa44 JK |
718 | break; |
719 | if (tagged >= nr_to_tag) | |
720 | break; | |
721 | while (((index >> shift) & RADIX_TREE_MAP_MASK) == 0) { | |
722 | /* | |
723 | * We've fully scanned this node. Go up. Because | |
724 | * last_index is guaranteed to be in the tree, what | |
725 | * we do below cannot wander astray. | |
726 | */ | |
144dcfc0 | 727 | slot = path[height - 1].node; |
ebf8aa44 JK |
728 | height++; |
729 | shift += RADIX_TREE_MAP_SHIFT; | |
730 | } | |
731 | } | |
732 | /* | |
ac15ee69 TO |
733 | * We need not to tag the root tag if there is no tag which is set with |
734 | * settag within the range from *first_indexp to last_index. | |
ebf8aa44 | 735 | */ |
ac15ee69 TO |
736 | if (tagged > 0) |
737 | root_tag_set(root, settag); | |
ebf8aa44 JK |
738 | *first_indexp = index; |
739 | ||
740 | return tagged; | |
741 | } | |
742 | EXPORT_SYMBOL(radix_tree_range_tag_if_tagged); | |
743 | ||
744 | ||
6df8ba4f FW |
745 | /** |
746 | * radix_tree_next_hole - find the next hole (not-present entry) | |
747 | * @root: tree root | |
748 | * @index: index key | |
749 | * @max_scan: maximum range to search | |
750 | * | |
751 | * Search the set [index, min(index+max_scan-1, MAX_INDEX)] for the lowest | |
752 | * indexed hole. | |
753 | * | |
754 | * Returns: the index of the hole if found, otherwise returns an index | |
755 | * outside of the set specified (in which case 'return - index >= max_scan' | |
8e6bdb7f | 756 | * will be true). In rare cases of index wrap-around, 0 will be returned. |
6df8ba4f FW |
757 | * |
758 | * radix_tree_next_hole may be called under rcu_read_lock. However, like | |
8e6bdb7f WF |
759 | * radix_tree_gang_lookup, this will not atomically search a snapshot of |
760 | * the tree at a single point in time. For example, if a hole is created | |
761 | * at index 5, then subsequently a hole is created at index 10, | |
762 | * radix_tree_next_hole covering both indexes may return 10 if called | |
763 | * under rcu_read_lock. | |
6df8ba4f FW |
764 | */ |
765 | unsigned long radix_tree_next_hole(struct radix_tree_root *root, | |
766 | unsigned long index, unsigned long max_scan) | |
767 | { | |
768 | unsigned long i; | |
769 | ||
770 | for (i = 0; i < max_scan; i++) { | |
771 | if (!radix_tree_lookup(root, index)) | |
772 | break; | |
773 | index++; | |
774 | if (index == 0) | |
775 | break; | |
776 | } | |
777 | ||
778 | return index; | |
779 | } | |
780 | EXPORT_SYMBOL(radix_tree_next_hole); | |
781 | ||
dc566127 WF |
782 | /** |
783 | * radix_tree_prev_hole - find the prev hole (not-present entry) | |
784 | * @root: tree root | |
785 | * @index: index key | |
786 | * @max_scan: maximum range to search | |
787 | * | |
788 | * Search backwards in the range [max(index-max_scan+1, 0), index] | |
789 | * for the first hole. | |
790 | * | |
791 | * Returns: the index of the hole if found, otherwise returns an index | |
792 | * outside of the set specified (in which case 'index - return >= max_scan' | |
edcd1d84 | 793 | * will be true). In rare cases of wrap-around, ULONG_MAX will be returned. |
dc566127 WF |
794 | * |
795 | * radix_tree_next_hole may be called under rcu_read_lock. However, like | |
796 | * radix_tree_gang_lookup, this will not atomically search a snapshot of | |
797 | * the tree at a single point in time. For example, if a hole is created | |
798 | * at index 10, then subsequently a hole is created at index 5, | |
799 | * radix_tree_prev_hole covering both indexes may return 5 if called under | |
800 | * rcu_read_lock. | |
801 | */ | |
802 | unsigned long radix_tree_prev_hole(struct radix_tree_root *root, | |
803 | unsigned long index, unsigned long max_scan) | |
804 | { | |
805 | unsigned long i; | |
806 | ||
807 | for (i = 0; i < max_scan; i++) { | |
808 | if (!radix_tree_lookup(root, index)) | |
809 | break; | |
810 | index--; | |
edcd1d84 | 811 | if (index == ULONG_MAX) |
dc566127 WF |
812 | break; |
813 | } | |
814 | ||
815 | return index; | |
816 | } | |
817 | EXPORT_SYMBOL(radix_tree_prev_hole); | |
818 | ||
1da177e4 | 819 | static unsigned int |
6328650b HD |
820 | __lookup(struct radix_tree_node *slot, void ***results, unsigned long *indices, |
821 | unsigned long index, unsigned int max_items, unsigned long *next_index) | |
1da177e4 LT |
822 | { |
823 | unsigned int nr_found = 0; | |
201b6264 | 824 | unsigned int shift, height; |
201b6264 CL |
825 | unsigned long i; |
826 | ||
7cf9c2c7 NP |
827 | height = slot->height; |
828 | if (height == 0) | |
201b6264 | 829 | goto out; |
1da177e4 | 830 | shift = (height-1) * RADIX_TREE_MAP_SHIFT; |
1da177e4 | 831 | |
201b6264 | 832 | for ( ; height > 1; height--) { |
7cf9c2c7 NP |
833 | i = (index >> shift) & RADIX_TREE_MAP_MASK; |
834 | for (;;) { | |
1da177e4 LT |
835 | if (slot->slots[i] != NULL) |
836 | break; | |
837 | index &= ~((1UL << shift) - 1); | |
838 | index += 1UL << shift; | |
839 | if (index == 0) | |
840 | goto out; /* 32-bit wraparound */ | |
7cf9c2c7 NP |
841 | i++; |
842 | if (i == RADIX_TREE_MAP_SIZE) | |
843 | goto out; | |
1da177e4 | 844 | } |
1da177e4 | 845 | |
1da177e4 | 846 | shift -= RADIX_TREE_MAP_SHIFT; |
2676a58c | 847 | slot = rcu_dereference_raw(slot->slots[i]); |
7cf9c2c7 NP |
848 | if (slot == NULL) |
849 | goto out; | |
1da177e4 | 850 | } |
201b6264 CL |
851 | |
852 | /* Bottom level: grab some items */ | |
853 | for (i = index & RADIX_TREE_MAP_MASK; i < RADIX_TREE_MAP_SIZE; i++) { | |
47feff2c | 854 | if (slot->slots[i]) { |
6328650b HD |
855 | results[nr_found] = &(slot->slots[i]); |
856 | if (indices) | |
857 | indices[nr_found] = index; | |
858 | if (++nr_found == max_items) { | |
859 | index++; | |
201b6264 | 860 | goto out; |
6328650b | 861 | } |
201b6264 | 862 | } |
6328650b | 863 | index++; |
201b6264 | 864 | } |
1da177e4 LT |
865 | out: |
866 | *next_index = index; | |
867 | return nr_found; | |
868 | } | |
869 | ||
870 | /** | |
871 | * radix_tree_gang_lookup - perform multiple lookup on a radix tree | |
872 | * @root: radix tree root | |
873 | * @results: where the results of the lookup are placed | |
874 | * @first_index: start the lookup from this key | |
875 | * @max_items: place up to this many items at *results | |
876 | * | |
877 | * Performs an index-ascending scan of the tree for present items. Places | |
878 | * them at *@results and returns the number of items which were placed at | |
879 | * *@results. | |
880 | * | |
881 | * The implementation is naive. | |
7cf9c2c7 NP |
882 | * |
883 | * Like radix_tree_lookup, radix_tree_gang_lookup may be called under | |
884 | * rcu_read_lock. In this case, rather than the returned results being | |
885 | * an atomic snapshot of the tree at a single point in time, the semantics | |
886 | * of an RCU protected gang lookup are as though multiple radix_tree_lookups | |
887 | * have been issued in individual locks, and results stored in 'results'. | |
1da177e4 LT |
888 | */ |
889 | unsigned int | |
890 | radix_tree_gang_lookup(struct radix_tree_root *root, void **results, | |
891 | unsigned long first_index, unsigned int max_items) | |
892 | { | |
7cf9c2c7 NP |
893 | unsigned long max_index; |
894 | struct radix_tree_node *node; | |
1da177e4 | 895 | unsigned long cur_index = first_index; |
7cf9c2c7 NP |
896 | unsigned int ret; |
897 | ||
2676a58c | 898 | node = rcu_dereference_raw(root->rnode); |
7cf9c2c7 NP |
899 | if (!node) |
900 | return 0; | |
1da177e4 | 901 | |
c0bc9875 | 902 | if (!radix_tree_is_indirect_ptr(node)) { |
7cf9c2c7 NP |
903 | if (first_index > 0) |
904 | return 0; | |
c0bc9875 | 905 | results[0] = node; |
7cf9c2c7 NP |
906 | return 1; |
907 | } | |
27d20fdd | 908 | node = indirect_to_ptr(node); |
7cf9c2c7 NP |
909 | |
910 | max_index = radix_tree_maxindex(node->height); | |
911 | ||
912 | ret = 0; | |
1da177e4 | 913 | while (ret < max_items) { |
47feff2c | 914 | unsigned int nr_found, slots_found, i; |
1da177e4 LT |
915 | unsigned long next_index; /* Index of next search */ |
916 | ||
917 | if (cur_index > max_index) | |
918 | break; | |
6328650b HD |
919 | slots_found = __lookup(node, (void ***)results + ret, NULL, |
920 | cur_index, max_items - ret, &next_index); | |
47feff2c NP |
921 | nr_found = 0; |
922 | for (i = 0; i < slots_found; i++) { | |
923 | struct radix_tree_node *slot; | |
924 | slot = *(((void ***)results)[ret + i]); | |
925 | if (!slot) | |
926 | continue; | |
27d20fdd NP |
927 | results[ret + nr_found] = |
928 | indirect_to_ptr(rcu_dereference_raw(slot)); | |
47feff2c NP |
929 | nr_found++; |
930 | } | |
1da177e4 LT |
931 | ret += nr_found; |
932 | if (next_index == 0) | |
933 | break; | |
934 | cur_index = next_index; | |
935 | } | |
7cf9c2c7 | 936 | |
1da177e4 LT |
937 | return ret; |
938 | } | |
939 | EXPORT_SYMBOL(radix_tree_gang_lookup); | |
940 | ||
47feff2c NP |
941 | /** |
942 | * radix_tree_gang_lookup_slot - perform multiple slot lookup on radix tree | |
943 | * @root: radix tree root | |
944 | * @results: where the results of the lookup are placed | |
6328650b | 945 | * @indices: where their indices should be placed (but usually NULL) |
47feff2c NP |
946 | * @first_index: start the lookup from this key |
947 | * @max_items: place up to this many items at *results | |
948 | * | |
949 | * Performs an index-ascending scan of the tree for present items. Places | |
950 | * their slots at *@results and returns the number of items which were | |
951 | * placed at *@results. | |
952 | * | |
953 | * The implementation is naive. | |
954 | * | |
955 | * Like radix_tree_gang_lookup as far as RCU and locking goes. Slots must | |
956 | * be dereferenced with radix_tree_deref_slot, and if using only RCU | |
957 | * protection, radix_tree_deref_slot may fail requiring a retry. | |
958 | */ | |
959 | unsigned int | |
6328650b HD |
960 | radix_tree_gang_lookup_slot(struct radix_tree_root *root, |
961 | void ***results, unsigned long *indices, | |
47feff2c NP |
962 | unsigned long first_index, unsigned int max_items) |
963 | { | |
964 | unsigned long max_index; | |
965 | struct radix_tree_node *node; | |
966 | unsigned long cur_index = first_index; | |
967 | unsigned int ret; | |
968 | ||
2676a58c | 969 | node = rcu_dereference_raw(root->rnode); |
47feff2c NP |
970 | if (!node) |
971 | return 0; | |
972 | ||
973 | if (!radix_tree_is_indirect_ptr(node)) { | |
974 | if (first_index > 0) | |
975 | return 0; | |
976 | results[0] = (void **)&root->rnode; | |
6328650b HD |
977 | if (indices) |
978 | indices[0] = 0; | |
47feff2c NP |
979 | return 1; |
980 | } | |
27d20fdd | 981 | node = indirect_to_ptr(node); |
47feff2c NP |
982 | |
983 | max_index = radix_tree_maxindex(node->height); | |
984 | ||
985 | ret = 0; | |
986 | while (ret < max_items) { | |
987 | unsigned int slots_found; | |
988 | unsigned long next_index; /* Index of next search */ | |
989 | ||
990 | if (cur_index > max_index) | |
991 | break; | |
6328650b HD |
992 | slots_found = __lookup(node, results + ret, |
993 | indices ? indices + ret : NULL, | |
994 | cur_index, max_items - ret, &next_index); | |
47feff2c NP |
995 | ret += slots_found; |
996 | if (next_index == 0) | |
997 | break; | |
998 | cur_index = next_index; | |
999 | } | |
1000 | ||
1001 | return ret; | |
1002 | } | |
1003 | EXPORT_SYMBOL(radix_tree_gang_lookup_slot); | |
1004 | ||
1da177e4 LT |
1005 | /* |
1006 | * FIXME: the two tag_get()s here should use find_next_bit() instead of | |
1007 | * open-coding the search. | |
1008 | */ | |
1009 | static unsigned int | |
47feff2c | 1010 | __lookup_tag(struct radix_tree_node *slot, void ***results, unsigned long index, |
daff89f3 | 1011 | unsigned int max_items, unsigned long *next_index, unsigned int tag) |
1da177e4 LT |
1012 | { |
1013 | unsigned int nr_found = 0; | |
7cf9c2c7 | 1014 | unsigned int shift, height; |
1da177e4 | 1015 | |
7cf9c2c7 NP |
1016 | height = slot->height; |
1017 | if (height == 0) | |
612d6c19 | 1018 | goto out; |
7cf9c2c7 | 1019 | shift = (height-1) * RADIX_TREE_MAP_SHIFT; |
1da177e4 | 1020 | |
7cf9c2c7 NP |
1021 | while (height > 0) { |
1022 | unsigned long i = (index >> shift) & RADIX_TREE_MAP_MASK ; | |
1da177e4 | 1023 | |
7cf9c2c7 NP |
1024 | for (;;) { |
1025 | if (tag_get(slot, tag, i)) | |
1da177e4 | 1026 | break; |
1da177e4 LT |
1027 | index &= ~((1UL << shift) - 1); |
1028 | index += 1UL << shift; | |
1029 | if (index == 0) | |
1030 | goto out; /* 32-bit wraparound */ | |
7cf9c2c7 NP |
1031 | i++; |
1032 | if (i == RADIX_TREE_MAP_SIZE) | |
1033 | goto out; | |
1da177e4 | 1034 | } |
1da177e4 LT |
1035 | height--; |
1036 | if (height == 0) { /* Bottom level: grab some items */ | |
1037 | unsigned long j = index & RADIX_TREE_MAP_MASK; | |
1038 | ||
1039 | for ( ; j < RADIX_TREE_MAP_SIZE; j++) { | |
1040 | index++; | |
7cf9c2c7 NP |
1041 | if (!tag_get(slot, tag, j)) |
1042 | continue; | |
7cf9c2c7 NP |
1043 | /* |
1044 | * Even though the tag was found set, we need to | |
1045 | * recheck that we have a non-NULL node, because | |
1046 | * if this lookup is lockless, it may have been | |
1047 | * subsequently deleted. | |
1048 | * | |
1049 | * Similar care must be taken in any place that | |
1050 | * lookup ->slots[x] without a lock (ie. can't | |
1051 | * rely on its value remaining the same). | |
1052 | */ | |
47feff2c NP |
1053 | if (slot->slots[j]) { |
1054 | results[nr_found++] = &(slot->slots[j]); | |
1da177e4 LT |
1055 | if (nr_found == max_items) |
1056 | goto out; | |
1057 | } | |
1058 | } | |
1059 | } | |
1060 | shift -= RADIX_TREE_MAP_SHIFT; | |
2676a58c | 1061 | slot = rcu_dereference_raw(slot->slots[i]); |
7cf9c2c7 NP |
1062 | if (slot == NULL) |
1063 | break; | |
1064 | } | |
1da177e4 LT |
1065 | out: |
1066 | *next_index = index; | |
1067 | return nr_found; | |
1068 | } | |
1069 | ||
1070 | /** | |
1071 | * radix_tree_gang_lookup_tag - perform multiple lookup on a radix tree | |
1072 | * based on a tag | |
1073 | * @root: radix tree root | |
1074 | * @results: where the results of the lookup are placed | |
1075 | * @first_index: start the lookup from this key | |
1076 | * @max_items: place up to this many items at *results | |
daff89f3 | 1077 | * @tag: the tag index (< RADIX_TREE_MAX_TAGS) |
1da177e4 LT |
1078 | * |
1079 | * Performs an index-ascending scan of the tree for present items which | |
1080 | * have the tag indexed by @tag set. Places the items at *@results and | |
1081 | * returns the number of items which were placed at *@results. | |
1082 | */ | |
1083 | unsigned int | |
1084 | radix_tree_gang_lookup_tag(struct radix_tree_root *root, void **results, | |
daff89f3 JC |
1085 | unsigned long first_index, unsigned int max_items, |
1086 | unsigned int tag) | |
1da177e4 | 1087 | { |
7cf9c2c7 NP |
1088 | struct radix_tree_node *node; |
1089 | unsigned long max_index; | |
1da177e4 | 1090 | unsigned long cur_index = first_index; |
7cf9c2c7 | 1091 | unsigned int ret; |
1da177e4 | 1092 | |
612d6c19 NP |
1093 | /* check the root's tag bit */ |
1094 | if (!root_tag_get(root, tag)) | |
1095 | return 0; | |
1096 | ||
2676a58c | 1097 | node = rcu_dereference_raw(root->rnode); |
7cf9c2c7 NP |
1098 | if (!node) |
1099 | return 0; | |
1100 | ||
c0bc9875 | 1101 | if (!radix_tree_is_indirect_ptr(node)) { |
7cf9c2c7 NP |
1102 | if (first_index > 0) |
1103 | return 0; | |
c0bc9875 | 1104 | results[0] = node; |
7cf9c2c7 NP |
1105 | return 1; |
1106 | } | |
27d20fdd | 1107 | node = indirect_to_ptr(node); |
7cf9c2c7 NP |
1108 | |
1109 | max_index = radix_tree_maxindex(node->height); | |
1110 | ||
1111 | ret = 0; | |
1da177e4 | 1112 | while (ret < max_items) { |
47feff2c | 1113 | unsigned int nr_found, slots_found, i; |
1da177e4 LT |
1114 | unsigned long next_index; /* Index of next search */ |
1115 | ||
1116 | if (cur_index > max_index) | |
1117 | break; | |
47feff2c NP |
1118 | slots_found = __lookup_tag(node, (void ***)results + ret, |
1119 | cur_index, max_items - ret, &next_index, tag); | |
1120 | nr_found = 0; | |
1121 | for (i = 0; i < slots_found; i++) { | |
1122 | struct radix_tree_node *slot; | |
1123 | slot = *(((void ***)results)[ret + i]); | |
1124 | if (!slot) | |
1125 | continue; | |
27d20fdd NP |
1126 | results[ret + nr_found] = |
1127 | indirect_to_ptr(rcu_dereference_raw(slot)); | |
47feff2c NP |
1128 | nr_found++; |
1129 | } | |
1da177e4 LT |
1130 | ret += nr_found; |
1131 | if (next_index == 0) | |
1132 | break; | |
1133 | cur_index = next_index; | |
1134 | } | |
7cf9c2c7 | 1135 | |
1da177e4 LT |
1136 | return ret; |
1137 | } | |
1138 | EXPORT_SYMBOL(radix_tree_gang_lookup_tag); | |
1139 | ||
47feff2c NP |
1140 | /** |
1141 | * radix_tree_gang_lookup_tag_slot - perform multiple slot lookup on a | |
1142 | * radix tree based on a tag | |
1143 | * @root: radix tree root | |
1144 | * @results: where the results of the lookup are placed | |
1145 | * @first_index: start the lookup from this key | |
1146 | * @max_items: place up to this many items at *results | |
1147 | * @tag: the tag index (< RADIX_TREE_MAX_TAGS) | |
1148 | * | |
1149 | * Performs an index-ascending scan of the tree for present items which | |
1150 | * have the tag indexed by @tag set. Places the slots at *@results and | |
1151 | * returns the number of slots which were placed at *@results. | |
1152 | */ | |
1153 | unsigned int | |
1154 | radix_tree_gang_lookup_tag_slot(struct radix_tree_root *root, void ***results, | |
1155 | unsigned long first_index, unsigned int max_items, | |
1156 | unsigned int tag) | |
1157 | { | |
1158 | struct radix_tree_node *node; | |
1159 | unsigned long max_index; | |
1160 | unsigned long cur_index = first_index; | |
1161 | unsigned int ret; | |
1162 | ||
1163 | /* check the root's tag bit */ | |
1164 | if (!root_tag_get(root, tag)) | |
1165 | return 0; | |
1166 | ||
2676a58c | 1167 | node = rcu_dereference_raw(root->rnode); |
47feff2c NP |
1168 | if (!node) |
1169 | return 0; | |
1170 | ||
1171 | if (!radix_tree_is_indirect_ptr(node)) { | |
1172 | if (first_index > 0) | |
1173 | return 0; | |
1174 | results[0] = (void **)&root->rnode; | |
1175 | return 1; | |
1176 | } | |
27d20fdd | 1177 | node = indirect_to_ptr(node); |
47feff2c NP |
1178 | |
1179 | max_index = radix_tree_maxindex(node->height); | |
1180 | ||
1181 | ret = 0; | |
1182 | while (ret < max_items) { | |
1183 | unsigned int slots_found; | |
1184 | unsigned long next_index; /* Index of next search */ | |
1185 | ||
1186 | if (cur_index > max_index) | |
1187 | break; | |
1188 | slots_found = __lookup_tag(node, results + ret, | |
1189 | cur_index, max_items - ret, &next_index, tag); | |
1190 | ret += slots_found; | |
1191 | if (next_index == 0) | |
1192 | break; | |
1193 | cur_index = next_index; | |
1194 | } | |
1195 | ||
1196 | return ret; | |
1197 | } | |
1198 | EXPORT_SYMBOL(radix_tree_gang_lookup_tag_slot); | |
1199 | ||
e504f3fd HD |
1200 | #if defined(CONFIG_SHMEM) && defined(CONFIG_SWAP) |
1201 | #include <linux/sched.h> /* for cond_resched() */ | |
1202 | ||
1203 | /* | |
1204 | * This linear search is at present only useful to shmem_unuse_inode(). | |
1205 | */ | |
1206 | static unsigned long __locate(struct radix_tree_node *slot, void *item, | |
1207 | unsigned long index, unsigned long *found_index) | |
1208 | { | |
1209 | unsigned int shift, height; | |
1210 | unsigned long i; | |
1211 | ||
1212 | height = slot->height; | |
1213 | shift = (height-1) * RADIX_TREE_MAP_SHIFT; | |
1214 | ||
1215 | for ( ; height > 1; height--) { | |
1216 | i = (index >> shift) & RADIX_TREE_MAP_MASK; | |
1217 | for (;;) { | |
1218 | if (slot->slots[i] != NULL) | |
1219 | break; | |
1220 | index &= ~((1UL << shift) - 1); | |
1221 | index += 1UL << shift; | |
1222 | if (index == 0) | |
1223 | goto out; /* 32-bit wraparound */ | |
1224 | i++; | |
1225 | if (i == RADIX_TREE_MAP_SIZE) | |
1226 | goto out; | |
1227 | } | |
1228 | ||
1229 | shift -= RADIX_TREE_MAP_SHIFT; | |
1230 | slot = rcu_dereference_raw(slot->slots[i]); | |
1231 | if (slot == NULL) | |
1232 | goto out; | |
1233 | } | |
1234 | ||
1235 | /* Bottom level: check items */ | |
1236 | for (i = 0; i < RADIX_TREE_MAP_SIZE; i++) { | |
1237 | if (slot->slots[i] == item) { | |
1238 | *found_index = index + i; | |
1239 | index = 0; | |
1240 | goto out; | |
1241 | } | |
1242 | } | |
1243 | index += RADIX_TREE_MAP_SIZE; | |
1244 | out: | |
1245 | return index; | |
1246 | } | |
1247 | ||
1248 | /** | |
1249 | * radix_tree_locate_item - search through radix tree for item | |
1250 | * @root: radix tree root | |
1251 | * @item: item to be found | |
1252 | * | |
1253 | * Returns index where item was found, or -1 if not found. | |
1254 | * Caller must hold no lock (since this time-consuming function needs | |
1255 | * to be preemptible), and must check afterwards if item is still there. | |
1256 | */ | |
1257 | unsigned long radix_tree_locate_item(struct radix_tree_root *root, void *item) | |
1258 | { | |
1259 | struct radix_tree_node *node; | |
1260 | unsigned long max_index; | |
1261 | unsigned long cur_index = 0; | |
1262 | unsigned long found_index = -1; | |
1263 | ||
1264 | do { | |
1265 | rcu_read_lock(); | |
1266 | node = rcu_dereference_raw(root->rnode); | |
1267 | if (!radix_tree_is_indirect_ptr(node)) { | |
1268 | rcu_read_unlock(); | |
1269 | if (node == item) | |
1270 | found_index = 0; | |
1271 | break; | |
1272 | } | |
1273 | ||
1274 | node = indirect_to_ptr(node); | |
1275 | max_index = radix_tree_maxindex(node->height); | |
1276 | if (cur_index > max_index) | |
1277 | break; | |
1278 | ||
1279 | cur_index = __locate(node, item, cur_index, &found_index); | |
1280 | rcu_read_unlock(); | |
1281 | cond_resched(); | |
1282 | } while (cur_index != 0 && cur_index <= max_index); | |
1283 | ||
1284 | return found_index; | |
1285 | } | |
1286 | #else | |
1287 | unsigned long radix_tree_locate_item(struct radix_tree_root *root, void *item) | |
1288 | { | |
1289 | return -1; | |
1290 | } | |
1291 | #endif /* CONFIG_SHMEM && CONFIG_SWAP */ | |
47feff2c | 1292 | |
a5f51c96 NP |
1293 | /** |
1294 | * radix_tree_shrink - shrink height of a radix tree to minimal | |
1295 | * @root radix tree root | |
1296 | */ | |
1297 | static inline void radix_tree_shrink(struct radix_tree_root *root) | |
1298 | { | |
1299 | /* try to shrink tree height */ | |
c0bc9875 | 1300 | while (root->height > 0) { |
a5f51c96 | 1301 | struct radix_tree_node *to_free = root->rnode; |
7cf9c2c7 | 1302 | void *newptr; |
a5f51c96 | 1303 | |
c0bc9875 | 1304 | BUG_ON(!radix_tree_is_indirect_ptr(to_free)); |
27d20fdd | 1305 | to_free = indirect_to_ptr(to_free); |
c0bc9875 NP |
1306 | |
1307 | /* | |
1308 | * The candidate node has more than one child, or its child | |
1309 | * is not at the leftmost slot, we cannot shrink. | |
1310 | */ | |
1311 | if (to_free->count != 1) | |
1312 | break; | |
1313 | if (!to_free->slots[0]) | |
1314 | break; | |
1315 | ||
7cf9c2c7 NP |
1316 | /* |
1317 | * We don't need rcu_assign_pointer(), since we are simply | |
27d20fdd NP |
1318 | * moving the node from one part of the tree to another: if it |
1319 | * was safe to dereference the old pointer to it | |
7cf9c2c7 | 1320 | * (to_free->slots[0]), it will be safe to dereference the new |
27d20fdd | 1321 | * one (root->rnode) as far as dependent read barriers go. |
7cf9c2c7 NP |
1322 | */ |
1323 | newptr = to_free->slots[0]; | |
c0bc9875 | 1324 | if (root->height > 1) |
27d20fdd | 1325 | newptr = ptr_to_indirect(newptr); |
7cf9c2c7 | 1326 | root->rnode = newptr; |
a5f51c96 | 1327 | root->height--; |
27d20fdd NP |
1328 | |
1329 | /* | |
1330 | * We have a dilemma here. The node's slot[0] must not be | |
1331 | * NULLed in case there are concurrent lookups expecting to | |
1332 | * find the item. However if this was a bottom-level node, | |
1333 | * then it may be subject to the slot pointer being visible | |
1334 | * to callers dereferencing it. If item corresponding to | |
1335 | * slot[0] is subsequently deleted, these callers would expect | |
1336 | * their slot to become empty sooner or later. | |
1337 | * | |
1338 | * For example, lockless pagecache will look up a slot, deref | |
1339 | * the page pointer, and if the page is 0 refcount it means it | |
1340 | * was concurrently deleted from pagecache so try the deref | |
1341 | * again. Fortunately there is already a requirement for logic | |
1342 | * to retry the entire slot lookup -- the indirect pointer | |
1343 | * problem (replacing direct root node with an indirect pointer | |
1344 | * also results in a stale slot). So tag the slot as indirect | |
1345 | * to force callers to retry. | |
1346 | */ | |
1347 | if (root->height == 0) | |
1348 | *((unsigned long *)&to_free->slots[0]) |= | |
1349 | RADIX_TREE_INDIRECT_PTR; | |
1350 | ||
a5f51c96 NP |
1351 | radix_tree_node_free(to_free); |
1352 | } | |
1353 | } | |
1354 | ||
1da177e4 LT |
1355 | /** |
1356 | * radix_tree_delete - delete an item from a radix tree | |
1357 | * @root: radix tree root | |
1358 | * @index: index key | |
1359 | * | |
1360 | * Remove the item at @index from the radix tree rooted at @root. | |
1361 | * | |
1362 | * Returns the address of the deleted item, or NULL if it was not present. | |
1363 | */ | |
1364 | void *radix_tree_delete(struct radix_tree_root *root, unsigned long index) | |
1365 | { | |
26fb1589 JM |
1366 | /* |
1367 | * The radix tree path needs to be one longer than the maximum path | |
1368 | * since the "list" is null terminated. | |
1369 | */ | |
1370 | struct radix_tree_path path[RADIX_TREE_MAX_PATH + 1], *pathp = path; | |
612d6c19 | 1371 | struct radix_tree_node *slot = NULL; |
7cf9c2c7 | 1372 | struct radix_tree_node *to_free; |
1da177e4 | 1373 | unsigned int height, shift; |
d5274261 NP |
1374 | int tag; |
1375 | int offset; | |
1da177e4 LT |
1376 | |
1377 | height = root->height; | |
1378 | if (index > radix_tree_maxindex(height)) | |
1379 | goto out; | |
1380 | ||
612d6c19 | 1381 | slot = root->rnode; |
c0bc9875 | 1382 | if (height == 0) { |
612d6c19 NP |
1383 | root_tag_clear_all(root); |
1384 | root->rnode = NULL; | |
1385 | goto out; | |
1386 | } | |
27d20fdd | 1387 | slot = indirect_to_ptr(slot); |
612d6c19 | 1388 | |
1da177e4 LT |
1389 | shift = (height - 1) * RADIX_TREE_MAP_SHIFT; |
1390 | pathp->node = NULL; | |
1da177e4 | 1391 | |
612d6c19 | 1392 | do { |
201b6264 | 1393 | if (slot == NULL) |
1da177e4 LT |
1394 | goto out; |
1395 | ||
d5274261 | 1396 | pathp++; |
1da177e4 | 1397 | offset = (index >> shift) & RADIX_TREE_MAP_MASK; |
d5274261 NP |
1398 | pathp->offset = offset; |
1399 | pathp->node = slot; | |
201b6264 | 1400 | slot = slot->slots[offset]; |
1da177e4 | 1401 | shift -= RADIX_TREE_MAP_SHIFT; |
612d6c19 NP |
1402 | height--; |
1403 | } while (height > 0); | |
1da177e4 | 1404 | |
612d6c19 | 1405 | if (slot == NULL) |
1da177e4 LT |
1406 | goto out; |
1407 | ||
1da177e4 LT |
1408 | /* |
1409 | * Clear all tags associated with the just-deleted item | |
1410 | */ | |
daff89f3 | 1411 | for (tag = 0; tag < RADIX_TREE_MAX_TAGS; tag++) { |
612d6c19 NP |
1412 | if (tag_get(pathp->node, tag, pathp->offset)) |
1413 | radix_tree_tag_clear(root, index, tag); | |
d5274261 | 1414 | } |
1da177e4 | 1415 | |
7cf9c2c7 | 1416 | to_free = NULL; |
201b6264 | 1417 | /* Now free the nodes we do not need anymore */ |
612d6c19 | 1418 | while (pathp->node) { |
201b6264 | 1419 | pathp->node->slots[pathp->offset] = NULL; |
a5f51c96 | 1420 | pathp->node->count--; |
7cf9c2c7 NP |
1421 | /* |
1422 | * Queue the node for deferred freeing after the | |
1423 | * last reference to it disappears (set NULL, above). | |
1424 | */ | |
1425 | if (to_free) | |
1426 | radix_tree_node_free(to_free); | |
a5f51c96 NP |
1427 | |
1428 | if (pathp->node->count) { | |
27d20fdd | 1429 | if (pathp->node == indirect_to_ptr(root->rnode)) |
a5f51c96 | 1430 | radix_tree_shrink(root); |
201b6264 | 1431 | goto out; |
a5f51c96 | 1432 | } |
201b6264 CL |
1433 | |
1434 | /* Node with zero slots in use so free it */ | |
7cf9c2c7 | 1435 | to_free = pathp->node; |
612d6c19 | 1436 | pathp--; |
7cf9c2c7 | 1437 | |
1da177e4 | 1438 | } |
612d6c19 | 1439 | root_tag_clear_all(root); |
201b6264 | 1440 | root->height = 0; |
612d6c19 | 1441 | root->rnode = NULL; |
7cf9c2c7 NP |
1442 | if (to_free) |
1443 | radix_tree_node_free(to_free); | |
612d6c19 | 1444 | |
1da177e4 | 1445 | out: |
612d6c19 | 1446 | return slot; |
1da177e4 LT |
1447 | } |
1448 | EXPORT_SYMBOL(radix_tree_delete); | |
1449 | ||
1450 | /** | |
1451 | * radix_tree_tagged - test whether any items in the tree are tagged | |
1452 | * @root: radix tree root | |
1453 | * @tag: tag to test | |
1454 | */ | |
daff89f3 | 1455 | int radix_tree_tagged(struct radix_tree_root *root, unsigned int tag) |
1da177e4 | 1456 | { |
612d6c19 | 1457 | return root_tag_get(root, tag); |
1da177e4 LT |
1458 | } |
1459 | EXPORT_SYMBOL(radix_tree_tagged); | |
1460 | ||
1461 | static void | |
51cc5068 | 1462 | radix_tree_node_ctor(void *node) |
1da177e4 LT |
1463 | { |
1464 | memset(node, 0, sizeof(struct radix_tree_node)); | |
1465 | } | |
1466 | ||
1467 | static __init unsigned long __maxindex(unsigned int height) | |
1468 | { | |
430d275a PL |
1469 | unsigned int width = height * RADIX_TREE_MAP_SHIFT; |
1470 | int shift = RADIX_TREE_INDEX_BITS - width; | |
1471 | ||
1472 | if (shift < 0) | |
1473 | return ~0UL; | |
1474 | if (shift >= BITS_PER_LONG) | |
1475 | return 0UL; | |
1476 | return ~0UL >> shift; | |
1da177e4 LT |
1477 | } |
1478 | ||
1479 | static __init void radix_tree_init_maxindex(void) | |
1480 | { | |
1481 | unsigned int i; | |
1482 | ||
1483 | for (i = 0; i < ARRAY_SIZE(height_to_maxindex); i++) | |
1484 | height_to_maxindex[i] = __maxindex(i); | |
1485 | } | |
1486 | ||
1da177e4 LT |
1487 | static int radix_tree_callback(struct notifier_block *nfb, |
1488 | unsigned long action, | |
1489 | void *hcpu) | |
1490 | { | |
1491 | int cpu = (long)hcpu; | |
1492 | struct radix_tree_preload *rtp; | |
1493 | ||
1494 | /* Free per-cpu pool of perloaded nodes */ | |
8bb78442 | 1495 | if (action == CPU_DEAD || action == CPU_DEAD_FROZEN) { |
1da177e4 LT |
1496 | rtp = &per_cpu(radix_tree_preloads, cpu); |
1497 | while (rtp->nr) { | |
1498 | kmem_cache_free(radix_tree_node_cachep, | |
1499 | rtp->nodes[rtp->nr-1]); | |
1500 | rtp->nodes[rtp->nr-1] = NULL; | |
1501 | rtp->nr--; | |
1502 | } | |
1503 | } | |
1504 | return NOTIFY_OK; | |
1505 | } | |
1da177e4 LT |
1506 | |
1507 | void __init radix_tree_init(void) | |
1508 | { | |
1509 | radix_tree_node_cachep = kmem_cache_create("radix_tree_node", | |
1510 | sizeof(struct radix_tree_node), 0, | |
488514d1 CL |
1511 | SLAB_PANIC | SLAB_RECLAIM_ACCOUNT, |
1512 | radix_tree_node_ctor); | |
1da177e4 LT |
1513 | radix_tree_init_maxindex(); |
1514 | hotcpu_notifier(radix_tree_callback, 0); | |
1515 | } |