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 |
78c1d784 | 6 | * Copyright (C) 2012 Konstantin Khlebnikov |
1da177e4 LT |
7 | * |
8 | * This program is free software; you can redistribute it and/or | |
9 | * modify it under the terms of the GNU General Public License as | |
10 | * published by the Free Software Foundation; either version 2, or (at | |
11 | * your option) any later version. | |
12 | * | |
13 | * This program is distributed in the hope that it will be useful, but | |
14 | * WITHOUT ANY WARRANTY; without even the implied warranty of | |
15 | * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU | |
16 | * General Public License for more details. | |
17 | * | |
18 | * You should have received a copy of the GNU General Public License | |
19 | * along with this program; if not, write to the Free Software | |
20 | * Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA. | |
21 | */ | |
22 | ||
23 | #include <linux/errno.h> | |
24 | #include <linux/init.h> | |
25 | #include <linux/kernel.h> | |
8bc3bcc9 | 26 | #include <linux/export.h> |
1da177e4 LT |
27 | #include <linux/radix-tree.h> |
28 | #include <linux/percpu.h> | |
29 | #include <linux/slab.h> | |
30 | #include <linux/notifier.h> | |
31 | #include <linux/cpu.h> | |
1da177e4 LT |
32 | #include <linux/string.h> |
33 | #include <linux/bitops.h> | |
7cf9c2c7 | 34 | #include <linux/rcupdate.h> |
1da177e4 LT |
35 | |
36 | ||
37 | #ifdef __KERNEL__ | |
cfd9b7df | 38 | #define RADIX_TREE_MAP_SHIFT (CONFIG_BASE_SMALL ? 4 : 6) |
1da177e4 LT |
39 | #else |
40 | #define RADIX_TREE_MAP_SHIFT 3 /* For more stressful testing */ | |
41 | #endif | |
1da177e4 LT |
42 | |
43 | #define RADIX_TREE_MAP_SIZE (1UL << RADIX_TREE_MAP_SHIFT) | |
44 | #define RADIX_TREE_MAP_MASK (RADIX_TREE_MAP_SIZE-1) | |
45 | ||
46 | #define RADIX_TREE_TAG_LONGS \ | |
47 | ((RADIX_TREE_MAP_SIZE + BITS_PER_LONG - 1) / BITS_PER_LONG) | |
48 | ||
49 | struct radix_tree_node { | |
7cf9c2c7 | 50 | unsigned int height; /* Height from the bottom */ |
1da177e4 | 51 | unsigned int count; |
e2bdb933 HD |
52 | union { |
53 | struct radix_tree_node *parent; /* Used when ascending tree */ | |
54 | struct rcu_head rcu_head; /* Used when freeing node */ | |
55 | }; | |
a1115570 | 56 | void __rcu *slots[RADIX_TREE_MAP_SIZE]; |
daff89f3 | 57 | unsigned long tags[RADIX_TREE_MAX_TAGS][RADIX_TREE_TAG_LONGS]; |
1da177e4 LT |
58 | }; |
59 | ||
1da177e4 | 60 | #define RADIX_TREE_INDEX_BITS (8 /* CHAR_BIT */ * sizeof(unsigned long)) |
26fb1589 JM |
61 | #define RADIX_TREE_MAX_PATH (DIV_ROUND_UP(RADIX_TREE_INDEX_BITS, \ |
62 | RADIX_TREE_MAP_SHIFT)) | |
1da177e4 | 63 | |
26fb1589 JM |
64 | /* |
65 | * The height_to_maxindex array needs to be one deeper than the maximum | |
66 | * path as height 0 holds only 1 entry. | |
67 | */ | |
68 | static unsigned long height_to_maxindex[RADIX_TREE_MAX_PATH + 1] __read_mostly; | |
1da177e4 LT |
69 | |
70 | /* | |
71 | * Radix tree node cache. | |
72 | */ | |
e18b890b | 73 | static struct kmem_cache *radix_tree_node_cachep; |
1da177e4 | 74 | |
55368052 NP |
75 | /* |
76 | * The radix tree is variable-height, so an insert operation not only has | |
77 | * to build the branch to its corresponding item, it also has to build the | |
78 | * branch to existing items if the size has to be increased (by | |
79 | * radix_tree_extend). | |
80 | * | |
81 | * The worst case is a zero height tree with just a single item at index 0, | |
82 | * and then inserting an item at index ULONG_MAX. This requires 2 new branches | |
83 | * of RADIX_TREE_MAX_PATH size to be created, with only the root node shared. | |
84 | * Hence: | |
85 | */ | |
86 | #define RADIX_TREE_PRELOAD_SIZE (RADIX_TREE_MAX_PATH * 2 - 1) | |
87 | ||
1da177e4 LT |
88 | /* |
89 | * Per-cpu pool of preloaded nodes | |
90 | */ | |
91 | struct radix_tree_preload { | |
92 | int nr; | |
55368052 | 93 | struct radix_tree_node *nodes[RADIX_TREE_PRELOAD_SIZE]; |
1da177e4 | 94 | }; |
8cef7d57 | 95 | static DEFINE_PER_CPU(struct radix_tree_preload, radix_tree_preloads) = { 0, }; |
1da177e4 | 96 | |
27d20fdd NP |
97 | static inline void *ptr_to_indirect(void *ptr) |
98 | { | |
99 | return (void *)((unsigned long)ptr | RADIX_TREE_INDIRECT_PTR); | |
100 | } | |
101 | ||
102 | static inline void *indirect_to_ptr(void *ptr) | |
103 | { | |
104 | return (void *)((unsigned long)ptr & ~RADIX_TREE_INDIRECT_PTR); | |
105 | } | |
106 | ||
612d6c19 NP |
107 | static inline gfp_t root_gfp_mask(struct radix_tree_root *root) |
108 | { | |
109 | return root->gfp_mask & __GFP_BITS_MASK; | |
110 | } | |
111 | ||
643b52b9 NP |
112 | static inline void tag_set(struct radix_tree_node *node, unsigned int tag, |
113 | int offset) | |
114 | { | |
115 | __set_bit(offset, node->tags[tag]); | |
116 | } | |
117 | ||
118 | static inline void tag_clear(struct radix_tree_node *node, unsigned int tag, | |
119 | int offset) | |
120 | { | |
121 | __clear_bit(offset, node->tags[tag]); | |
122 | } | |
123 | ||
124 | static inline int tag_get(struct radix_tree_node *node, unsigned int tag, | |
125 | int offset) | |
126 | { | |
127 | return test_bit(offset, node->tags[tag]); | |
128 | } | |
129 | ||
130 | static inline void root_tag_set(struct radix_tree_root *root, unsigned int tag) | |
131 | { | |
132 | root->gfp_mask |= (__force gfp_t)(1 << (tag + __GFP_BITS_SHIFT)); | |
133 | } | |
134 | ||
135 | static inline void root_tag_clear(struct radix_tree_root *root, unsigned int tag) | |
136 | { | |
137 | root->gfp_mask &= (__force gfp_t)~(1 << (tag + __GFP_BITS_SHIFT)); | |
138 | } | |
139 | ||
140 | static inline void root_tag_clear_all(struct radix_tree_root *root) | |
141 | { | |
142 | root->gfp_mask &= __GFP_BITS_MASK; | |
143 | } | |
144 | ||
145 | static inline int root_tag_get(struct radix_tree_root *root, unsigned int tag) | |
146 | { | |
147 | return (__force unsigned)root->gfp_mask & (1 << (tag + __GFP_BITS_SHIFT)); | |
148 | } | |
149 | ||
150 | /* | |
151 | * Returns 1 if any slot in the node has this tag set. | |
152 | * Otherwise returns 0. | |
153 | */ | |
154 | static inline int any_tag_set(struct radix_tree_node *node, unsigned int tag) | |
155 | { | |
156 | int idx; | |
157 | for (idx = 0; idx < RADIX_TREE_TAG_LONGS; idx++) { | |
158 | if (node->tags[tag][idx]) | |
159 | return 1; | |
160 | } | |
161 | return 0; | |
162 | } | |
78c1d784 KK |
163 | |
164 | /** | |
165 | * radix_tree_find_next_bit - find the next set bit in a memory region | |
166 | * | |
167 | * @addr: The address to base the search on | |
168 | * @size: The bitmap size in bits | |
169 | * @offset: The bitnumber to start searching at | |
170 | * | |
171 | * Unrollable variant of find_next_bit() for constant size arrays. | |
172 | * Tail bits starting from size to roundup(size, BITS_PER_LONG) must be zero. | |
173 | * Returns next bit offset, or size if nothing found. | |
174 | */ | |
175 | static __always_inline unsigned long | |
176 | radix_tree_find_next_bit(const unsigned long *addr, | |
177 | unsigned long size, unsigned long offset) | |
178 | { | |
179 | if (!__builtin_constant_p(size)) | |
180 | return find_next_bit(addr, size, offset); | |
181 | ||
182 | if (offset < size) { | |
183 | unsigned long tmp; | |
184 | ||
185 | addr += offset / BITS_PER_LONG; | |
186 | tmp = *addr >> (offset % BITS_PER_LONG); | |
187 | if (tmp) | |
188 | return __ffs(tmp) + offset; | |
189 | offset = (offset + BITS_PER_LONG) & ~(BITS_PER_LONG - 1); | |
190 | while (offset < size) { | |
191 | tmp = *++addr; | |
192 | if (tmp) | |
193 | return __ffs(tmp) + offset; | |
194 | offset += BITS_PER_LONG; | |
195 | } | |
196 | } | |
197 | return size; | |
198 | } | |
199 | ||
1da177e4 LT |
200 | /* |
201 | * This assumes that the caller has performed appropriate preallocation, and | |
202 | * that the caller has pinned this thread of control to the current CPU. | |
203 | */ | |
204 | static struct radix_tree_node * | |
205 | radix_tree_node_alloc(struct radix_tree_root *root) | |
206 | { | |
e2848a0e | 207 | struct radix_tree_node *ret = NULL; |
612d6c19 | 208 | gfp_t gfp_mask = root_gfp_mask(root); |
1da177e4 | 209 | |
e2848a0e | 210 | if (!(gfp_mask & __GFP_WAIT)) { |
1da177e4 LT |
211 | struct radix_tree_preload *rtp; |
212 | ||
e2848a0e NP |
213 | /* |
214 | * Provided the caller has preloaded here, we will always | |
215 | * succeed in getting a node here (and never reach | |
216 | * kmem_cache_alloc) | |
217 | */ | |
1da177e4 LT |
218 | rtp = &__get_cpu_var(radix_tree_preloads); |
219 | if (rtp->nr) { | |
220 | ret = rtp->nodes[rtp->nr - 1]; | |
221 | rtp->nodes[rtp->nr - 1] = NULL; | |
222 | rtp->nr--; | |
223 | } | |
224 | } | |
e2848a0e | 225 | if (ret == NULL) |
488514d1 | 226 | ret = kmem_cache_alloc(radix_tree_node_cachep, gfp_mask); |
e2848a0e | 227 | |
c0bc9875 | 228 | BUG_ON(radix_tree_is_indirect_ptr(ret)); |
1da177e4 LT |
229 | return ret; |
230 | } | |
231 | ||
7cf9c2c7 NP |
232 | static void radix_tree_node_rcu_free(struct rcu_head *head) |
233 | { | |
234 | struct radix_tree_node *node = | |
235 | container_of(head, struct radix_tree_node, rcu_head); | |
b6dd0865 | 236 | int i; |
643b52b9 NP |
237 | |
238 | /* | |
239 | * must only free zeroed nodes into the slab. radix_tree_shrink | |
240 | * can leave us with a non-NULL entry in the first slot, so clear | |
241 | * that here to make sure. | |
242 | */ | |
b6dd0865 DC |
243 | for (i = 0; i < RADIX_TREE_MAX_TAGS; i++) |
244 | tag_clear(node, i, 0); | |
245 | ||
643b52b9 NP |
246 | node->slots[0] = NULL; |
247 | node->count = 0; | |
248 | ||
7cf9c2c7 NP |
249 | kmem_cache_free(radix_tree_node_cachep, node); |
250 | } | |
251 | ||
1da177e4 LT |
252 | static inline void |
253 | radix_tree_node_free(struct radix_tree_node *node) | |
254 | { | |
7cf9c2c7 | 255 | call_rcu(&node->rcu_head, radix_tree_node_rcu_free); |
1da177e4 LT |
256 | } |
257 | ||
258 | /* | |
259 | * Load up this CPU's radix_tree_node buffer with sufficient objects to | |
260 | * ensure that the addition of a single element in the tree cannot fail. On | |
261 | * success, return zero, with preemption disabled. On error, return -ENOMEM | |
262 | * with preemption not disabled. | |
b34df792 DH |
263 | * |
264 | * To make use of this facility, the radix tree must be initialised without | |
265 | * __GFP_WAIT being passed to INIT_RADIX_TREE(). | |
1da177e4 | 266 | */ |
dd0fc66f | 267 | int radix_tree_preload(gfp_t gfp_mask) |
1da177e4 LT |
268 | { |
269 | struct radix_tree_preload *rtp; | |
270 | struct radix_tree_node *node; | |
271 | int ret = -ENOMEM; | |
272 | ||
273 | preempt_disable(); | |
274 | rtp = &__get_cpu_var(radix_tree_preloads); | |
275 | while (rtp->nr < ARRAY_SIZE(rtp->nodes)) { | |
276 | preempt_enable(); | |
488514d1 | 277 | node = kmem_cache_alloc(radix_tree_node_cachep, gfp_mask); |
1da177e4 LT |
278 | if (node == NULL) |
279 | goto out; | |
280 | preempt_disable(); | |
281 | rtp = &__get_cpu_var(radix_tree_preloads); | |
282 | if (rtp->nr < ARRAY_SIZE(rtp->nodes)) | |
283 | rtp->nodes[rtp->nr++] = node; | |
284 | else | |
285 | kmem_cache_free(radix_tree_node_cachep, node); | |
286 | } | |
287 | ret = 0; | |
288 | out: | |
289 | return ret; | |
290 | } | |
d7f0923d | 291 | EXPORT_SYMBOL(radix_tree_preload); |
1da177e4 | 292 | |
1da177e4 LT |
293 | /* |
294 | * Return the maximum key which can be store into a | |
295 | * radix tree with height HEIGHT. | |
296 | */ | |
297 | static inline unsigned long radix_tree_maxindex(unsigned int height) | |
298 | { | |
299 | return height_to_maxindex[height]; | |
300 | } | |
301 | ||
302 | /* | |
303 | * Extend a radix tree so it can store key @index. | |
304 | */ | |
305 | static int radix_tree_extend(struct radix_tree_root *root, unsigned long index) | |
306 | { | |
307 | struct radix_tree_node *node; | |
e2bdb933 | 308 | struct radix_tree_node *slot; |
1da177e4 | 309 | unsigned int height; |
1da177e4 LT |
310 | int tag; |
311 | ||
312 | /* Figure out what the height should be. */ | |
313 | height = root->height + 1; | |
314 | while (index > radix_tree_maxindex(height)) | |
315 | height++; | |
316 | ||
317 | if (root->rnode == NULL) { | |
318 | root->height = height; | |
319 | goto out; | |
320 | } | |
321 | ||
1da177e4 | 322 | do { |
7cf9c2c7 | 323 | unsigned int newheight; |
1da177e4 LT |
324 | if (!(node = radix_tree_node_alloc(root))) |
325 | return -ENOMEM; | |
326 | ||
1da177e4 | 327 | /* Propagate the aggregated tag info into the new root */ |
daff89f3 | 328 | for (tag = 0; tag < RADIX_TREE_MAX_TAGS; tag++) { |
612d6c19 | 329 | if (root_tag_get(root, tag)) |
1da177e4 LT |
330 | tag_set(node, tag, 0); |
331 | } | |
332 | ||
e2bdb933 | 333 | /* Increase the height. */ |
7cf9c2c7 NP |
334 | newheight = root->height+1; |
335 | node->height = newheight; | |
1da177e4 | 336 | node->count = 1; |
e2bdb933 HD |
337 | node->parent = NULL; |
338 | slot = root->rnode; | |
339 | if (newheight > 1) { | |
340 | slot = indirect_to_ptr(slot); | |
341 | slot->parent = node; | |
342 | } | |
343 | node->slots[0] = slot; | |
27d20fdd | 344 | node = ptr_to_indirect(node); |
7cf9c2c7 NP |
345 | rcu_assign_pointer(root->rnode, node); |
346 | root->height = newheight; | |
1da177e4 LT |
347 | } while (height > root->height); |
348 | out: | |
349 | return 0; | |
350 | } | |
351 | ||
352 | /** | |
353 | * radix_tree_insert - insert into a radix tree | |
354 | * @root: radix tree root | |
355 | * @index: index key | |
356 | * @item: item to insert | |
357 | * | |
358 | * Insert an item into the radix tree at position @index. | |
359 | */ | |
360 | int radix_tree_insert(struct radix_tree_root *root, | |
361 | unsigned long index, void *item) | |
362 | { | |
201b6264 | 363 | struct radix_tree_node *node = NULL, *slot; |
1da177e4 LT |
364 | unsigned int height, shift; |
365 | int offset; | |
366 | int error; | |
367 | ||
c0bc9875 | 368 | BUG_ON(radix_tree_is_indirect_ptr(item)); |
7cf9c2c7 | 369 | |
1da177e4 | 370 | /* Make sure the tree is high enough. */ |
612d6c19 | 371 | if (index > radix_tree_maxindex(root->height)) { |
1da177e4 LT |
372 | error = radix_tree_extend(root, index); |
373 | if (error) | |
374 | return error; | |
375 | } | |
376 | ||
27d20fdd | 377 | slot = indirect_to_ptr(root->rnode); |
c0bc9875 | 378 | |
1da177e4 LT |
379 | height = root->height; |
380 | shift = (height-1) * RADIX_TREE_MAP_SHIFT; | |
381 | ||
382 | offset = 0; /* uninitialised var warning */ | |
612d6c19 | 383 | while (height > 0) { |
201b6264 | 384 | if (slot == NULL) { |
1da177e4 | 385 | /* Have to add a child node. */ |
201b6264 | 386 | if (!(slot = radix_tree_node_alloc(root))) |
1da177e4 | 387 | return -ENOMEM; |
7cf9c2c7 | 388 | slot->height = height; |
e2bdb933 | 389 | slot->parent = node; |
201b6264 | 390 | if (node) { |
7cf9c2c7 | 391 | rcu_assign_pointer(node->slots[offset], slot); |
1da177e4 | 392 | node->count++; |
201b6264 | 393 | } else |
27d20fdd | 394 | rcu_assign_pointer(root->rnode, ptr_to_indirect(slot)); |
1da177e4 LT |
395 | } |
396 | ||
397 | /* Go a level down */ | |
398 | offset = (index >> shift) & RADIX_TREE_MAP_MASK; | |
201b6264 CL |
399 | node = slot; |
400 | slot = node->slots[offset]; | |
1da177e4 LT |
401 | shift -= RADIX_TREE_MAP_SHIFT; |
402 | height--; | |
612d6c19 | 403 | } |
1da177e4 | 404 | |
201b6264 | 405 | if (slot != NULL) |
1da177e4 | 406 | return -EEXIST; |
201b6264 | 407 | |
612d6c19 NP |
408 | if (node) { |
409 | node->count++; | |
7cf9c2c7 | 410 | rcu_assign_pointer(node->slots[offset], item); |
612d6c19 NP |
411 | BUG_ON(tag_get(node, 0, offset)); |
412 | BUG_ON(tag_get(node, 1, offset)); | |
413 | } else { | |
c0bc9875 | 414 | rcu_assign_pointer(root->rnode, item); |
612d6c19 NP |
415 | BUG_ON(root_tag_get(root, 0)); |
416 | BUG_ON(root_tag_get(root, 1)); | |
417 | } | |
1da177e4 | 418 | |
1da177e4 LT |
419 | return 0; |
420 | } | |
421 | EXPORT_SYMBOL(radix_tree_insert); | |
422 | ||
b72b71c6 HS |
423 | /* |
424 | * is_slot == 1 : search for the slot. | |
425 | * is_slot == 0 : search for the node. | |
7cf9c2c7 | 426 | */ |
b72b71c6 HS |
427 | static void *radix_tree_lookup_element(struct radix_tree_root *root, |
428 | unsigned long index, int is_slot) | |
1da177e4 LT |
429 | { |
430 | unsigned int height, shift; | |
7cf9c2c7 | 431 | struct radix_tree_node *node, **slot; |
612d6c19 | 432 | |
2676a58c | 433 | node = rcu_dereference_raw(root->rnode); |
7cf9c2c7 | 434 | if (node == NULL) |
1da177e4 LT |
435 | return NULL; |
436 | ||
c0bc9875 | 437 | if (!radix_tree_is_indirect_ptr(node)) { |
7cf9c2c7 NP |
438 | if (index > 0) |
439 | return NULL; | |
b72b71c6 | 440 | return is_slot ? (void *)&root->rnode : node; |
7cf9c2c7 | 441 | } |
27d20fdd | 442 | node = indirect_to_ptr(node); |
7cf9c2c7 NP |
443 | |
444 | height = node->height; | |
445 | if (index > radix_tree_maxindex(height)) | |
446 | return NULL; | |
612d6c19 | 447 | |
1da177e4 | 448 | shift = (height-1) * RADIX_TREE_MAP_SHIFT; |
1da177e4 | 449 | |
7cf9c2c7 NP |
450 | do { |
451 | slot = (struct radix_tree_node **) | |
452 | (node->slots + ((index>>shift) & RADIX_TREE_MAP_MASK)); | |
2676a58c | 453 | node = rcu_dereference_raw(*slot); |
7cf9c2c7 | 454 | if (node == NULL) |
1da177e4 LT |
455 | return NULL; |
456 | ||
1da177e4 LT |
457 | shift -= RADIX_TREE_MAP_SHIFT; |
458 | height--; | |
7cf9c2c7 | 459 | } while (height > 0); |
1da177e4 | 460 | |
27d20fdd | 461 | return is_slot ? (void *)slot : indirect_to_ptr(node); |
b72b71c6 HS |
462 | } |
463 | ||
464 | /** | |
465 | * radix_tree_lookup_slot - lookup a slot in a radix tree | |
466 | * @root: radix tree root | |
467 | * @index: index key | |
468 | * | |
469 | * Returns: the slot corresponding to the position @index in the | |
470 | * radix tree @root. This is useful for update-if-exists operations. | |
471 | * | |
472 | * This function can be called under rcu_read_lock iff the slot is not | |
473 | * modified by radix_tree_replace_slot, otherwise it must be called | |
474 | * exclusive from other writers. Any dereference of the slot must be done | |
475 | * using radix_tree_deref_slot. | |
476 | */ | |
477 | void **radix_tree_lookup_slot(struct radix_tree_root *root, unsigned long index) | |
478 | { | |
479 | return (void **)radix_tree_lookup_element(root, index, 1); | |
a4331366 | 480 | } |
a4331366 HR |
481 | EXPORT_SYMBOL(radix_tree_lookup_slot); |
482 | ||
483 | /** | |
484 | * radix_tree_lookup - perform lookup operation on a radix tree | |
485 | * @root: radix tree root | |
486 | * @index: index key | |
487 | * | |
488 | * Lookup the item at the position @index in the radix tree @root. | |
7cf9c2c7 NP |
489 | * |
490 | * This function can be called under rcu_read_lock, however the caller | |
491 | * must manage lifetimes of leaf nodes (eg. RCU may also be used to free | |
492 | * them safely). No RCU barriers are required to access or modify the | |
493 | * returned item, however. | |
a4331366 HR |
494 | */ |
495 | void *radix_tree_lookup(struct radix_tree_root *root, unsigned long index) | |
496 | { | |
b72b71c6 | 497 | return radix_tree_lookup_element(root, index, 0); |
1da177e4 LT |
498 | } |
499 | EXPORT_SYMBOL(radix_tree_lookup); | |
500 | ||
501 | /** | |
502 | * radix_tree_tag_set - set a tag on a radix tree node | |
503 | * @root: radix tree root | |
504 | * @index: index key | |
505 | * @tag: tag index | |
506 | * | |
daff89f3 JC |
507 | * Set the search tag (which must be < RADIX_TREE_MAX_TAGS) |
508 | * corresponding to @index in the radix tree. From | |
1da177e4 LT |
509 | * the root all the way down to the leaf node. |
510 | * | |
511 | * Returns the address of the tagged item. Setting a tag on a not-present | |
512 | * item is a bug. | |
513 | */ | |
514 | void *radix_tree_tag_set(struct radix_tree_root *root, | |
daff89f3 | 515 | unsigned long index, unsigned int tag) |
1da177e4 LT |
516 | { |
517 | unsigned int height, shift; | |
201b6264 | 518 | struct radix_tree_node *slot; |
1da177e4 LT |
519 | |
520 | height = root->height; | |
4c91c364 | 521 | BUG_ON(index > radix_tree_maxindex(height)); |
1da177e4 | 522 | |
27d20fdd | 523 | slot = indirect_to_ptr(root->rnode); |
612d6c19 | 524 | shift = (height - 1) * RADIX_TREE_MAP_SHIFT; |
1da177e4 LT |
525 | |
526 | while (height > 0) { | |
527 | int offset; | |
528 | ||
529 | offset = (index >> shift) & RADIX_TREE_MAP_MASK; | |
d5274261 NP |
530 | if (!tag_get(slot, tag, offset)) |
531 | tag_set(slot, tag, offset); | |
201b6264 CL |
532 | slot = slot->slots[offset]; |
533 | BUG_ON(slot == NULL); | |
1da177e4 LT |
534 | shift -= RADIX_TREE_MAP_SHIFT; |
535 | height--; | |
536 | } | |
537 | ||
612d6c19 NP |
538 | /* set the root's tag bit */ |
539 | if (slot && !root_tag_get(root, tag)) | |
540 | root_tag_set(root, tag); | |
541 | ||
201b6264 | 542 | return slot; |
1da177e4 LT |
543 | } |
544 | EXPORT_SYMBOL(radix_tree_tag_set); | |
545 | ||
546 | /** | |
547 | * radix_tree_tag_clear - clear a tag on a radix tree node | |
548 | * @root: radix tree root | |
549 | * @index: index key | |
550 | * @tag: tag index | |
551 | * | |
daff89f3 JC |
552 | * Clear the search tag (which must be < RADIX_TREE_MAX_TAGS) |
553 | * corresponding to @index in the radix tree. If | |
1da177e4 LT |
554 | * this causes the leaf node to have no tags set then clear the tag in the |
555 | * next-to-leaf node, etc. | |
556 | * | |
557 | * Returns the address of the tagged item on success, else NULL. ie: | |
558 | * has the same return value and semantics as radix_tree_lookup(). | |
559 | */ | |
560 | void *radix_tree_tag_clear(struct radix_tree_root *root, | |
daff89f3 | 561 | unsigned long index, unsigned int tag) |
1da177e4 | 562 | { |
e2bdb933 | 563 | struct radix_tree_node *node = NULL; |
612d6c19 | 564 | struct radix_tree_node *slot = NULL; |
1da177e4 | 565 | unsigned int height, shift; |
e2bdb933 | 566 | int uninitialized_var(offset); |
1da177e4 LT |
567 | |
568 | height = root->height; | |
569 | if (index > radix_tree_maxindex(height)) | |
570 | goto out; | |
571 | ||
e2bdb933 | 572 | shift = height * RADIX_TREE_MAP_SHIFT; |
27d20fdd | 573 | slot = indirect_to_ptr(root->rnode); |
1da177e4 | 574 | |
e2bdb933 | 575 | while (shift) { |
201b6264 | 576 | if (slot == NULL) |
1da177e4 LT |
577 | goto out; |
578 | ||
e2bdb933 | 579 | shift -= RADIX_TREE_MAP_SHIFT; |
1da177e4 | 580 | offset = (index >> shift) & RADIX_TREE_MAP_MASK; |
e2bdb933 | 581 | node = slot; |
201b6264 | 582 | slot = slot->slots[offset]; |
1da177e4 LT |
583 | } |
584 | ||
612d6c19 | 585 | if (slot == NULL) |
1da177e4 LT |
586 | goto out; |
587 | ||
e2bdb933 HD |
588 | while (node) { |
589 | if (!tag_get(node, tag, offset)) | |
d5274261 | 590 | goto out; |
e2bdb933 HD |
591 | tag_clear(node, tag, offset); |
592 | if (any_tag_set(node, tag)) | |
6e954b9e | 593 | goto out; |
e2bdb933 HD |
594 | |
595 | index >>= RADIX_TREE_MAP_SHIFT; | |
596 | offset = index & RADIX_TREE_MAP_MASK; | |
597 | node = node->parent; | |
612d6c19 NP |
598 | } |
599 | ||
600 | /* clear the root's tag bit */ | |
601 | if (root_tag_get(root, tag)) | |
602 | root_tag_clear(root, tag); | |
603 | ||
1da177e4 | 604 | out: |
612d6c19 | 605 | return slot; |
1da177e4 LT |
606 | } |
607 | EXPORT_SYMBOL(radix_tree_tag_clear); | |
608 | ||
1da177e4 | 609 | /** |
32605a18 MT |
610 | * radix_tree_tag_get - get a tag on a radix tree node |
611 | * @root: radix tree root | |
612 | * @index: index key | |
daff89f3 | 613 | * @tag: tag index (< RADIX_TREE_MAX_TAGS) |
1da177e4 | 614 | * |
32605a18 | 615 | * Return values: |
1da177e4 | 616 | * |
612d6c19 NP |
617 | * 0: tag not present or not set |
618 | * 1: tag set | |
ce82653d DH |
619 | * |
620 | * Note that the return value of this function may not be relied on, even if | |
621 | * the RCU lock is held, unless tag modification and node deletion are excluded | |
622 | * from concurrency. | |
1da177e4 LT |
623 | */ |
624 | int radix_tree_tag_get(struct radix_tree_root *root, | |
daff89f3 | 625 | unsigned long index, unsigned int tag) |
1da177e4 LT |
626 | { |
627 | unsigned int height, shift; | |
7cf9c2c7 | 628 | struct radix_tree_node *node; |
1da177e4 | 629 | |
612d6c19 NP |
630 | /* check the root's tag bit */ |
631 | if (!root_tag_get(root, tag)) | |
632 | return 0; | |
633 | ||
2676a58c | 634 | node = rcu_dereference_raw(root->rnode); |
7cf9c2c7 NP |
635 | if (node == NULL) |
636 | return 0; | |
637 | ||
c0bc9875 | 638 | if (!radix_tree_is_indirect_ptr(node)) |
7cf9c2c7 | 639 | return (index == 0); |
27d20fdd | 640 | node = indirect_to_ptr(node); |
7cf9c2c7 NP |
641 | |
642 | height = node->height; | |
643 | if (index > radix_tree_maxindex(height)) | |
644 | return 0; | |
612d6c19 | 645 | |
1da177e4 | 646 | shift = (height - 1) * RADIX_TREE_MAP_SHIFT; |
1da177e4 LT |
647 | |
648 | for ( ; ; ) { | |
649 | int offset; | |
650 | ||
7cf9c2c7 | 651 | if (node == NULL) |
1da177e4 LT |
652 | return 0; |
653 | ||
654 | offset = (index >> shift) & RADIX_TREE_MAP_MASK; | |
7cf9c2c7 | 655 | if (!tag_get(node, tag, offset)) |
3fa36acb | 656 | return 0; |
ce82653d | 657 | if (height == 1) |
3fa36acb | 658 | return 1; |
2676a58c | 659 | node = rcu_dereference_raw(node->slots[offset]); |
1da177e4 LT |
660 | shift -= RADIX_TREE_MAP_SHIFT; |
661 | height--; | |
662 | } | |
663 | } | |
664 | EXPORT_SYMBOL(radix_tree_tag_get); | |
1da177e4 | 665 | |
78c1d784 KK |
666 | /** |
667 | * radix_tree_next_chunk - find next chunk of slots for iteration | |
668 | * | |
669 | * @root: radix tree root | |
670 | * @iter: iterator state | |
671 | * @flags: RADIX_TREE_ITER_* flags and tag index | |
672 | * Returns: pointer to chunk first slot, or NULL if iteration is over | |
673 | */ | |
674 | void **radix_tree_next_chunk(struct radix_tree_root *root, | |
675 | struct radix_tree_iter *iter, unsigned flags) | |
676 | { | |
677 | unsigned shift, tag = flags & RADIX_TREE_ITER_TAG_MASK; | |
678 | struct radix_tree_node *rnode, *node; | |
679 | unsigned long index, offset; | |
680 | ||
681 | if ((flags & RADIX_TREE_ITER_TAGGED) && !root_tag_get(root, tag)) | |
682 | return NULL; | |
683 | ||
684 | /* | |
685 | * Catch next_index overflow after ~0UL. iter->index never overflows | |
686 | * during iterating; it can be zero only at the beginning. | |
687 | * And we cannot overflow iter->next_index in a single step, | |
688 | * because RADIX_TREE_MAP_SHIFT < BITS_PER_LONG. | |
689 | */ | |
690 | index = iter->next_index; | |
691 | if (!index && iter->index) | |
692 | return NULL; | |
693 | ||
694 | rnode = rcu_dereference_raw(root->rnode); | |
695 | if (radix_tree_is_indirect_ptr(rnode)) { | |
696 | rnode = indirect_to_ptr(rnode); | |
697 | } else if (rnode && !index) { | |
698 | /* Single-slot tree */ | |
699 | iter->index = 0; | |
700 | iter->next_index = 1; | |
701 | iter->tags = 1; | |
702 | return (void **)&root->rnode; | |
703 | } else | |
704 | return NULL; | |
705 | ||
706 | restart: | |
707 | shift = (rnode->height - 1) * RADIX_TREE_MAP_SHIFT; | |
708 | offset = index >> shift; | |
709 | ||
710 | /* Index outside of the tree */ | |
711 | if (offset >= RADIX_TREE_MAP_SIZE) | |
712 | return NULL; | |
713 | ||
714 | node = rnode; | |
715 | while (1) { | |
716 | if ((flags & RADIX_TREE_ITER_TAGGED) ? | |
717 | !test_bit(offset, node->tags[tag]) : | |
718 | !node->slots[offset]) { | |
719 | /* Hole detected */ | |
720 | if (flags & RADIX_TREE_ITER_CONTIG) | |
721 | return NULL; | |
722 | ||
723 | if (flags & RADIX_TREE_ITER_TAGGED) | |
724 | offset = radix_tree_find_next_bit( | |
725 | node->tags[tag], | |
726 | RADIX_TREE_MAP_SIZE, | |
727 | offset + 1); | |
728 | else | |
729 | while (++offset < RADIX_TREE_MAP_SIZE) { | |
730 | if (node->slots[offset]) | |
731 | break; | |
732 | } | |
733 | index &= ~((RADIX_TREE_MAP_SIZE << shift) - 1); | |
734 | index += offset << shift; | |
735 | /* Overflow after ~0UL */ | |
736 | if (!index) | |
737 | return NULL; | |
738 | if (offset == RADIX_TREE_MAP_SIZE) | |
739 | goto restart; | |
740 | } | |
741 | ||
742 | /* This is leaf-node */ | |
743 | if (!shift) | |
744 | break; | |
745 | ||
746 | node = rcu_dereference_raw(node->slots[offset]); | |
747 | if (node == NULL) | |
748 | goto restart; | |
749 | shift -= RADIX_TREE_MAP_SHIFT; | |
750 | offset = (index >> shift) & RADIX_TREE_MAP_MASK; | |
751 | } | |
752 | ||
753 | /* Update the iterator state */ | |
754 | iter->index = index; | |
755 | iter->next_index = (index | RADIX_TREE_MAP_MASK) + 1; | |
756 | ||
757 | /* Construct iter->tags bit-mask from node->tags[tag] array */ | |
758 | if (flags & RADIX_TREE_ITER_TAGGED) { | |
759 | unsigned tag_long, tag_bit; | |
760 | ||
761 | tag_long = offset / BITS_PER_LONG; | |
762 | tag_bit = offset % BITS_PER_LONG; | |
763 | iter->tags = node->tags[tag][tag_long] >> tag_bit; | |
764 | /* This never happens if RADIX_TREE_TAG_LONGS == 1 */ | |
765 | if (tag_long < RADIX_TREE_TAG_LONGS - 1) { | |
766 | /* Pick tags from next element */ | |
767 | if (tag_bit) | |
768 | iter->tags |= node->tags[tag][tag_long + 1] << | |
769 | (BITS_PER_LONG - tag_bit); | |
770 | /* Clip chunk size, here only BITS_PER_LONG tags */ | |
771 | iter->next_index = index + BITS_PER_LONG; | |
772 | } | |
773 | } | |
774 | ||
775 | return node->slots + offset; | |
776 | } | |
777 | EXPORT_SYMBOL(radix_tree_next_chunk); | |
778 | ||
ebf8aa44 JK |
779 | /** |
780 | * radix_tree_range_tag_if_tagged - for each item in given range set given | |
781 | * tag if item has another tag set | |
782 | * @root: radix tree root | |
783 | * @first_indexp: pointer to a starting index of a range to scan | |
784 | * @last_index: last index of a range to scan | |
785 | * @nr_to_tag: maximum number items to tag | |
786 | * @iftag: tag index to test | |
787 | * @settag: tag index to set if tested tag is set | |
788 | * | |
789 | * This function scans range of radix tree from first_index to last_index | |
790 | * (inclusive). For each item in the range if iftag is set, the function sets | |
791 | * also settag. The function stops either after tagging nr_to_tag items or | |
792 | * after reaching last_index. | |
793 | * | |
144dcfc0 DC |
794 | * The tags must be set from the leaf level only and propagated back up the |
795 | * path to the root. We must do this so that we resolve the full path before | |
796 | * setting any tags on intermediate nodes. If we set tags as we descend, then | |
797 | * we can get to the leaf node and find that the index that has the iftag | |
798 | * set is outside the range we are scanning. This reults in dangling tags and | |
799 | * can lead to problems with later tag operations (e.g. livelocks on lookups). | |
800 | * | |
ebf8aa44 JK |
801 | * The function returns number of leaves where the tag was set and sets |
802 | * *first_indexp to the first unscanned index. | |
d5ed3a4a JK |
803 | * WARNING! *first_indexp can wrap if last_index is ULONG_MAX. Caller must |
804 | * be prepared to handle that. | |
ebf8aa44 JK |
805 | */ |
806 | unsigned long radix_tree_range_tag_if_tagged(struct radix_tree_root *root, | |
807 | unsigned long *first_indexp, unsigned long last_index, | |
808 | unsigned long nr_to_tag, | |
809 | unsigned int iftag, unsigned int settag) | |
810 | { | |
144dcfc0 | 811 | unsigned int height = root->height; |
e2bdb933 | 812 | struct radix_tree_node *node = NULL; |
144dcfc0 DC |
813 | struct radix_tree_node *slot; |
814 | unsigned int shift; | |
815 | unsigned long tagged = 0; | |
816 | unsigned long index = *first_indexp; | |
ebf8aa44 JK |
817 | |
818 | last_index = min(last_index, radix_tree_maxindex(height)); | |
819 | if (index > last_index) | |
820 | return 0; | |
821 | if (!nr_to_tag) | |
822 | return 0; | |
823 | if (!root_tag_get(root, iftag)) { | |
824 | *first_indexp = last_index + 1; | |
825 | return 0; | |
826 | } | |
827 | if (height == 0) { | |
828 | *first_indexp = last_index + 1; | |
829 | root_tag_set(root, settag); | |
830 | return 1; | |
831 | } | |
832 | ||
833 | shift = (height - 1) * RADIX_TREE_MAP_SHIFT; | |
27d20fdd | 834 | slot = indirect_to_ptr(root->rnode); |
ebf8aa44 JK |
835 | |
836 | for (;;) { | |
e2bdb933 | 837 | unsigned long upindex; |
ebf8aa44 JK |
838 | int offset; |
839 | ||
840 | offset = (index >> shift) & RADIX_TREE_MAP_MASK; | |
841 | if (!slot->slots[offset]) | |
842 | goto next; | |
843 | if (!tag_get(slot, iftag, offset)) | |
844 | goto next; | |
e2bdb933 | 845 | if (shift) { |
144dcfc0 | 846 | /* Go down one level */ |
144dcfc0 | 847 | shift -= RADIX_TREE_MAP_SHIFT; |
e2bdb933 | 848 | node = slot; |
144dcfc0 DC |
849 | slot = slot->slots[offset]; |
850 | continue; | |
851 | } | |
852 | ||
853 | /* tag the leaf */ | |
854 | tagged++; | |
ebf8aa44 | 855 | tag_set(slot, settag, offset); |
144dcfc0 DC |
856 | |
857 | /* walk back up the path tagging interior nodes */ | |
e2bdb933 HD |
858 | upindex = index; |
859 | while (node) { | |
860 | upindex >>= RADIX_TREE_MAP_SHIFT; | |
861 | offset = upindex & RADIX_TREE_MAP_MASK; | |
862 | ||
144dcfc0 | 863 | /* stop if we find a node with the tag already set */ |
e2bdb933 | 864 | if (tag_get(node, settag, offset)) |
144dcfc0 | 865 | break; |
e2bdb933 HD |
866 | tag_set(node, settag, offset); |
867 | node = node->parent; | |
ebf8aa44 | 868 | } |
144dcfc0 | 869 | |
e2bdb933 HD |
870 | /* |
871 | * Small optimization: now clear that node pointer. | |
872 | * Since all of this slot's ancestors now have the tag set | |
873 | * from setting it above, we have no further need to walk | |
874 | * back up the tree setting tags, until we update slot to | |
875 | * point to another radix_tree_node. | |
876 | */ | |
877 | node = NULL; | |
878 | ||
ebf8aa44 JK |
879 | next: |
880 | /* Go to next item at level determined by 'shift' */ | |
881 | index = ((index >> shift) + 1) << shift; | |
d5ed3a4a JK |
882 | /* Overflow can happen when last_index is ~0UL... */ |
883 | if (index > last_index || !index) | |
ebf8aa44 JK |
884 | break; |
885 | if (tagged >= nr_to_tag) | |
886 | break; | |
887 | while (((index >> shift) & RADIX_TREE_MAP_MASK) == 0) { | |
888 | /* | |
889 | * We've fully scanned this node. Go up. Because | |
890 | * last_index is guaranteed to be in the tree, what | |
891 | * we do below cannot wander astray. | |
892 | */ | |
e2bdb933 | 893 | slot = slot->parent; |
ebf8aa44 JK |
894 | shift += RADIX_TREE_MAP_SHIFT; |
895 | } | |
896 | } | |
897 | /* | |
ac15ee69 TO |
898 | * We need not to tag the root tag if there is no tag which is set with |
899 | * settag within the range from *first_indexp to last_index. | |
ebf8aa44 | 900 | */ |
ac15ee69 TO |
901 | if (tagged > 0) |
902 | root_tag_set(root, settag); | |
ebf8aa44 JK |
903 | *first_indexp = index; |
904 | ||
905 | return tagged; | |
906 | } | |
907 | EXPORT_SYMBOL(radix_tree_range_tag_if_tagged); | |
908 | ||
909 | ||
6df8ba4f FW |
910 | /** |
911 | * radix_tree_next_hole - find the next hole (not-present entry) | |
912 | * @root: tree root | |
913 | * @index: index key | |
914 | * @max_scan: maximum range to search | |
915 | * | |
916 | * Search the set [index, min(index+max_scan-1, MAX_INDEX)] for the lowest | |
917 | * indexed hole. | |
918 | * | |
919 | * Returns: the index of the hole if found, otherwise returns an index | |
920 | * outside of the set specified (in which case 'return - index >= max_scan' | |
8e6bdb7f | 921 | * will be true). In rare cases of index wrap-around, 0 will be returned. |
6df8ba4f FW |
922 | * |
923 | * radix_tree_next_hole may be called under rcu_read_lock. However, like | |
8e6bdb7f WF |
924 | * radix_tree_gang_lookup, this will not atomically search a snapshot of |
925 | * the tree at a single point in time. For example, if a hole is created | |
926 | * at index 5, then subsequently a hole is created at index 10, | |
927 | * radix_tree_next_hole covering both indexes may return 10 if called | |
928 | * under rcu_read_lock. | |
6df8ba4f FW |
929 | */ |
930 | unsigned long radix_tree_next_hole(struct radix_tree_root *root, | |
931 | unsigned long index, unsigned long max_scan) | |
932 | { | |
933 | unsigned long i; | |
934 | ||
935 | for (i = 0; i < max_scan; i++) { | |
936 | if (!radix_tree_lookup(root, index)) | |
937 | break; | |
938 | index++; | |
939 | if (index == 0) | |
940 | break; | |
941 | } | |
942 | ||
943 | return index; | |
944 | } | |
945 | EXPORT_SYMBOL(radix_tree_next_hole); | |
946 | ||
dc566127 WF |
947 | /** |
948 | * radix_tree_prev_hole - find the prev hole (not-present entry) | |
949 | * @root: tree root | |
950 | * @index: index key | |
951 | * @max_scan: maximum range to search | |
952 | * | |
953 | * Search backwards in the range [max(index-max_scan+1, 0), index] | |
954 | * for the first hole. | |
955 | * | |
956 | * Returns: the index of the hole if found, otherwise returns an index | |
957 | * outside of the set specified (in which case 'index - return >= max_scan' | |
edcd1d84 | 958 | * will be true). In rare cases of wrap-around, ULONG_MAX will be returned. |
dc566127 WF |
959 | * |
960 | * radix_tree_next_hole may be called under rcu_read_lock. However, like | |
961 | * radix_tree_gang_lookup, this will not atomically search a snapshot of | |
962 | * the tree at a single point in time. For example, if a hole is created | |
963 | * at index 10, then subsequently a hole is created at index 5, | |
964 | * radix_tree_prev_hole covering both indexes may return 5 if called under | |
965 | * rcu_read_lock. | |
966 | */ | |
967 | unsigned long radix_tree_prev_hole(struct radix_tree_root *root, | |
968 | unsigned long index, unsigned long max_scan) | |
969 | { | |
970 | unsigned long i; | |
971 | ||
972 | for (i = 0; i < max_scan; i++) { | |
973 | if (!radix_tree_lookup(root, index)) | |
974 | break; | |
975 | index--; | |
edcd1d84 | 976 | if (index == ULONG_MAX) |
dc566127 WF |
977 | break; |
978 | } | |
979 | ||
980 | return index; | |
981 | } | |
982 | EXPORT_SYMBOL(radix_tree_prev_hole); | |
983 | ||
1da177e4 LT |
984 | /** |
985 | * radix_tree_gang_lookup - perform multiple lookup on a radix tree | |
986 | * @root: radix tree root | |
987 | * @results: where the results of the lookup are placed | |
988 | * @first_index: start the lookup from this key | |
989 | * @max_items: place up to this many items at *results | |
990 | * | |
991 | * Performs an index-ascending scan of the tree for present items. Places | |
992 | * them at *@results and returns the number of items which were placed at | |
993 | * *@results. | |
994 | * | |
995 | * The implementation is naive. | |
7cf9c2c7 NP |
996 | * |
997 | * Like radix_tree_lookup, radix_tree_gang_lookup may be called under | |
998 | * rcu_read_lock. In this case, rather than the returned results being | |
999 | * an atomic snapshot of the tree at a single point in time, the semantics | |
1000 | * of an RCU protected gang lookup are as though multiple radix_tree_lookups | |
1001 | * have been issued in individual locks, and results stored in 'results'. | |
1da177e4 LT |
1002 | */ |
1003 | unsigned int | |
1004 | radix_tree_gang_lookup(struct radix_tree_root *root, void **results, | |
1005 | unsigned long first_index, unsigned int max_items) | |
1006 | { | |
cebbd29e KK |
1007 | struct radix_tree_iter iter; |
1008 | void **slot; | |
1009 | unsigned int ret = 0; | |
7cf9c2c7 | 1010 | |
cebbd29e | 1011 | if (unlikely(!max_items)) |
7cf9c2c7 | 1012 | return 0; |
1da177e4 | 1013 | |
cebbd29e KK |
1014 | radix_tree_for_each_slot(slot, root, &iter, first_index) { |
1015 | results[ret] = indirect_to_ptr(rcu_dereference_raw(*slot)); | |
1016 | if (!results[ret]) | |
1017 | continue; | |
1018 | if (++ret == max_items) | |
1da177e4 | 1019 | break; |
1da177e4 | 1020 | } |
7cf9c2c7 | 1021 | |
1da177e4 LT |
1022 | return ret; |
1023 | } | |
1024 | EXPORT_SYMBOL(radix_tree_gang_lookup); | |
1025 | ||
47feff2c NP |
1026 | /** |
1027 | * radix_tree_gang_lookup_slot - perform multiple slot lookup on radix tree | |
1028 | * @root: radix tree root | |
1029 | * @results: where the results of the lookup are placed | |
6328650b | 1030 | * @indices: where their indices should be placed (but usually NULL) |
47feff2c NP |
1031 | * @first_index: start the lookup from this key |
1032 | * @max_items: place up to this many items at *results | |
1033 | * | |
1034 | * Performs an index-ascending scan of the tree for present items. Places | |
1035 | * their slots at *@results and returns the number of items which were | |
1036 | * placed at *@results. | |
1037 | * | |
1038 | * The implementation is naive. | |
1039 | * | |
1040 | * Like radix_tree_gang_lookup as far as RCU and locking goes. Slots must | |
1041 | * be dereferenced with radix_tree_deref_slot, and if using only RCU | |
1042 | * protection, radix_tree_deref_slot may fail requiring a retry. | |
1043 | */ | |
1044 | unsigned int | |
6328650b HD |
1045 | radix_tree_gang_lookup_slot(struct radix_tree_root *root, |
1046 | void ***results, unsigned long *indices, | |
47feff2c NP |
1047 | unsigned long first_index, unsigned int max_items) |
1048 | { | |
cebbd29e KK |
1049 | struct radix_tree_iter iter; |
1050 | void **slot; | |
1051 | unsigned int ret = 0; | |
47feff2c | 1052 | |
cebbd29e | 1053 | if (unlikely(!max_items)) |
47feff2c NP |
1054 | return 0; |
1055 | ||
cebbd29e KK |
1056 | radix_tree_for_each_slot(slot, root, &iter, first_index) { |
1057 | results[ret] = slot; | |
6328650b | 1058 | if (indices) |
cebbd29e KK |
1059 | indices[ret] = iter.index; |
1060 | if (++ret == max_items) | |
47feff2c | 1061 | break; |
47feff2c NP |
1062 | } |
1063 | ||
1064 | return ret; | |
1065 | } | |
1066 | EXPORT_SYMBOL(radix_tree_gang_lookup_slot); | |
1067 | ||
1da177e4 LT |
1068 | /** |
1069 | * radix_tree_gang_lookup_tag - perform multiple lookup on a radix tree | |
1070 | * based on a tag | |
1071 | * @root: radix tree root | |
1072 | * @results: where the results of the lookup are placed | |
1073 | * @first_index: start the lookup from this key | |
1074 | * @max_items: place up to this many items at *results | |
daff89f3 | 1075 | * @tag: the tag index (< RADIX_TREE_MAX_TAGS) |
1da177e4 LT |
1076 | * |
1077 | * Performs an index-ascending scan of the tree for present items which | |
1078 | * have the tag indexed by @tag set. Places the items at *@results and | |
1079 | * returns the number of items which were placed at *@results. | |
1080 | */ | |
1081 | unsigned int | |
1082 | radix_tree_gang_lookup_tag(struct radix_tree_root *root, void **results, | |
daff89f3 JC |
1083 | unsigned long first_index, unsigned int max_items, |
1084 | unsigned int tag) | |
1da177e4 | 1085 | { |
cebbd29e KK |
1086 | struct radix_tree_iter iter; |
1087 | void **slot; | |
1088 | unsigned int ret = 0; | |
612d6c19 | 1089 | |
cebbd29e | 1090 | if (unlikely(!max_items)) |
7cf9c2c7 NP |
1091 | return 0; |
1092 | ||
cebbd29e KK |
1093 | radix_tree_for_each_tagged(slot, root, &iter, first_index, tag) { |
1094 | results[ret] = indirect_to_ptr(rcu_dereference_raw(*slot)); | |
1095 | if (!results[ret]) | |
1096 | continue; | |
1097 | if (++ret == max_items) | |
1da177e4 | 1098 | break; |
1da177e4 | 1099 | } |
7cf9c2c7 | 1100 | |
1da177e4 LT |
1101 | return ret; |
1102 | } | |
1103 | EXPORT_SYMBOL(radix_tree_gang_lookup_tag); | |
1104 | ||
47feff2c NP |
1105 | /** |
1106 | * radix_tree_gang_lookup_tag_slot - perform multiple slot lookup on a | |
1107 | * radix tree based on a tag | |
1108 | * @root: radix tree root | |
1109 | * @results: where the results of the lookup are placed | |
1110 | * @first_index: start the lookup from this key | |
1111 | * @max_items: place up to this many items at *results | |
1112 | * @tag: the tag index (< RADIX_TREE_MAX_TAGS) | |
1113 | * | |
1114 | * Performs an index-ascending scan of the tree for present items which | |
1115 | * have the tag indexed by @tag set. Places the slots at *@results and | |
1116 | * returns the number of slots which were placed at *@results. | |
1117 | */ | |
1118 | unsigned int | |
1119 | radix_tree_gang_lookup_tag_slot(struct radix_tree_root *root, void ***results, | |
1120 | unsigned long first_index, unsigned int max_items, | |
1121 | unsigned int tag) | |
1122 | { | |
cebbd29e KK |
1123 | struct radix_tree_iter iter; |
1124 | void **slot; | |
1125 | unsigned int ret = 0; | |
47feff2c | 1126 | |
cebbd29e | 1127 | if (unlikely(!max_items)) |
47feff2c NP |
1128 | return 0; |
1129 | ||
cebbd29e KK |
1130 | radix_tree_for_each_tagged(slot, root, &iter, first_index, tag) { |
1131 | results[ret] = slot; | |
1132 | if (++ret == max_items) | |
47feff2c | 1133 | break; |
47feff2c NP |
1134 | } |
1135 | ||
1136 | return ret; | |
1137 | } | |
1138 | EXPORT_SYMBOL(radix_tree_gang_lookup_tag_slot); | |
1139 | ||
e504f3fd HD |
1140 | #if defined(CONFIG_SHMEM) && defined(CONFIG_SWAP) |
1141 | #include <linux/sched.h> /* for cond_resched() */ | |
1142 | ||
1143 | /* | |
1144 | * This linear search is at present only useful to shmem_unuse_inode(). | |
1145 | */ | |
1146 | static unsigned long __locate(struct radix_tree_node *slot, void *item, | |
1147 | unsigned long index, unsigned long *found_index) | |
1148 | { | |
1149 | unsigned int shift, height; | |
1150 | unsigned long i; | |
1151 | ||
1152 | height = slot->height; | |
1153 | shift = (height-1) * RADIX_TREE_MAP_SHIFT; | |
1154 | ||
1155 | for ( ; height > 1; height--) { | |
1156 | i = (index >> shift) & RADIX_TREE_MAP_MASK; | |
1157 | for (;;) { | |
1158 | if (slot->slots[i] != NULL) | |
1159 | break; | |
1160 | index &= ~((1UL << shift) - 1); | |
1161 | index += 1UL << shift; | |
1162 | if (index == 0) | |
1163 | goto out; /* 32-bit wraparound */ | |
1164 | i++; | |
1165 | if (i == RADIX_TREE_MAP_SIZE) | |
1166 | goto out; | |
1167 | } | |
1168 | ||
1169 | shift -= RADIX_TREE_MAP_SHIFT; | |
1170 | slot = rcu_dereference_raw(slot->slots[i]); | |
1171 | if (slot == NULL) | |
1172 | goto out; | |
1173 | } | |
1174 | ||
1175 | /* Bottom level: check items */ | |
1176 | for (i = 0; i < RADIX_TREE_MAP_SIZE; i++) { | |
1177 | if (slot->slots[i] == item) { | |
1178 | *found_index = index + i; | |
1179 | index = 0; | |
1180 | goto out; | |
1181 | } | |
1182 | } | |
1183 | index += RADIX_TREE_MAP_SIZE; | |
1184 | out: | |
1185 | return index; | |
1186 | } | |
1187 | ||
1188 | /** | |
1189 | * radix_tree_locate_item - search through radix tree for item | |
1190 | * @root: radix tree root | |
1191 | * @item: item to be found | |
1192 | * | |
1193 | * Returns index where item was found, or -1 if not found. | |
1194 | * Caller must hold no lock (since this time-consuming function needs | |
1195 | * to be preemptible), and must check afterwards if item is still there. | |
1196 | */ | |
1197 | unsigned long radix_tree_locate_item(struct radix_tree_root *root, void *item) | |
1198 | { | |
1199 | struct radix_tree_node *node; | |
1200 | unsigned long max_index; | |
1201 | unsigned long cur_index = 0; | |
1202 | unsigned long found_index = -1; | |
1203 | ||
1204 | do { | |
1205 | rcu_read_lock(); | |
1206 | node = rcu_dereference_raw(root->rnode); | |
1207 | if (!radix_tree_is_indirect_ptr(node)) { | |
1208 | rcu_read_unlock(); | |
1209 | if (node == item) | |
1210 | found_index = 0; | |
1211 | break; | |
1212 | } | |
1213 | ||
1214 | node = indirect_to_ptr(node); | |
1215 | max_index = radix_tree_maxindex(node->height); | |
1216 | if (cur_index > max_index) | |
1217 | break; | |
1218 | ||
1219 | cur_index = __locate(node, item, cur_index, &found_index); | |
1220 | rcu_read_unlock(); | |
1221 | cond_resched(); | |
1222 | } while (cur_index != 0 && cur_index <= max_index); | |
1223 | ||
1224 | return found_index; | |
1225 | } | |
1226 | #else | |
1227 | unsigned long radix_tree_locate_item(struct radix_tree_root *root, void *item) | |
1228 | { | |
1229 | return -1; | |
1230 | } | |
1231 | #endif /* CONFIG_SHMEM && CONFIG_SWAP */ | |
47feff2c | 1232 | |
a5f51c96 NP |
1233 | /** |
1234 | * radix_tree_shrink - shrink height of a radix tree to minimal | |
1235 | * @root radix tree root | |
1236 | */ | |
1237 | static inline void radix_tree_shrink(struct radix_tree_root *root) | |
1238 | { | |
1239 | /* try to shrink tree height */ | |
c0bc9875 | 1240 | while (root->height > 0) { |
a5f51c96 | 1241 | struct radix_tree_node *to_free = root->rnode; |
e2bdb933 | 1242 | struct radix_tree_node *slot; |
a5f51c96 | 1243 | |
c0bc9875 | 1244 | BUG_ON(!radix_tree_is_indirect_ptr(to_free)); |
27d20fdd | 1245 | to_free = indirect_to_ptr(to_free); |
c0bc9875 NP |
1246 | |
1247 | /* | |
1248 | * The candidate node has more than one child, or its child | |
1249 | * is not at the leftmost slot, we cannot shrink. | |
1250 | */ | |
1251 | if (to_free->count != 1) | |
1252 | break; | |
1253 | if (!to_free->slots[0]) | |
1254 | break; | |
1255 | ||
7cf9c2c7 NP |
1256 | /* |
1257 | * We don't need rcu_assign_pointer(), since we are simply | |
27d20fdd NP |
1258 | * moving the node from one part of the tree to another: if it |
1259 | * was safe to dereference the old pointer to it | |
7cf9c2c7 | 1260 | * (to_free->slots[0]), it will be safe to dereference the new |
27d20fdd | 1261 | * one (root->rnode) as far as dependent read barriers go. |
7cf9c2c7 | 1262 | */ |
e2bdb933 HD |
1263 | slot = to_free->slots[0]; |
1264 | if (root->height > 1) { | |
1265 | slot->parent = NULL; | |
1266 | slot = ptr_to_indirect(slot); | |
1267 | } | |
1268 | root->rnode = slot; | |
a5f51c96 | 1269 | root->height--; |
27d20fdd NP |
1270 | |
1271 | /* | |
1272 | * We have a dilemma here. The node's slot[0] must not be | |
1273 | * NULLed in case there are concurrent lookups expecting to | |
1274 | * find the item. However if this was a bottom-level node, | |
1275 | * then it may be subject to the slot pointer being visible | |
1276 | * to callers dereferencing it. If item corresponding to | |
1277 | * slot[0] is subsequently deleted, these callers would expect | |
1278 | * their slot to become empty sooner or later. | |
1279 | * | |
1280 | * For example, lockless pagecache will look up a slot, deref | |
1281 | * the page pointer, and if the page is 0 refcount it means it | |
1282 | * was concurrently deleted from pagecache so try the deref | |
1283 | * again. Fortunately there is already a requirement for logic | |
1284 | * to retry the entire slot lookup -- the indirect pointer | |
1285 | * problem (replacing direct root node with an indirect pointer | |
1286 | * also results in a stale slot). So tag the slot as indirect | |
1287 | * to force callers to retry. | |
1288 | */ | |
1289 | if (root->height == 0) | |
1290 | *((unsigned long *)&to_free->slots[0]) |= | |
1291 | RADIX_TREE_INDIRECT_PTR; | |
1292 | ||
a5f51c96 NP |
1293 | radix_tree_node_free(to_free); |
1294 | } | |
1295 | } | |
1296 | ||
1da177e4 LT |
1297 | /** |
1298 | * radix_tree_delete - delete an item from a radix tree | |
1299 | * @root: radix tree root | |
1300 | * @index: index key | |
1301 | * | |
1302 | * Remove the item at @index from the radix tree rooted at @root. | |
1303 | * | |
1304 | * Returns the address of the deleted item, or NULL if it was not present. | |
1305 | */ | |
1306 | void *radix_tree_delete(struct radix_tree_root *root, unsigned long index) | |
1307 | { | |
e2bdb933 | 1308 | struct radix_tree_node *node = NULL; |
612d6c19 | 1309 | struct radix_tree_node *slot = NULL; |
7cf9c2c7 | 1310 | struct radix_tree_node *to_free; |
1da177e4 | 1311 | unsigned int height, shift; |
d5274261 | 1312 | int tag; |
e2bdb933 | 1313 | int uninitialized_var(offset); |
1da177e4 LT |
1314 | |
1315 | height = root->height; | |
1316 | if (index > radix_tree_maxindex(height)) | |
1317 | goto out; | |
1318 | ||
612d6c19 | 1319 | slot = root->rnode; |
c0bc9875 | 1320 | if (height == 0) { |
612d6c19 NP |
1321 | root_tag_clear_all(root); |
1322 | root->rnode = NULL; | |
1323 | goto out; | |
1324 | } | |
27d20fdd | 1325 | slot = indirect_to_ptr(slot); |
e2bdb933 | 1326 | shift = height * RADIX_TREE_MAP_SHIFT; |
1da177e4 | 1327 | |
612d6c19 | 1328 | do { |
201b6264 | 1329 | if (slot == NULL) |
1da177e4 LT |
1330 | goto out; |
1331 | ||
e2bdb933 | 1332 | shift -= RADIX_TREE_MAP_SHIFT; |
1da177e4 | 1333 | offset = (index >> shift) & RADIX_TREE_MAP_MASK; |
e2bdb933 | 1334 | node = slot; |
201b6264 | 1335 | slot = slot->slots[offset]; |
e2bdb933 | 1336 | } while (shift); |
1da177e4 | 1337 | |
612d6c19 | 1338 | if (slot == NULL) |
1da177e4 LT |
1339 | goto out; |
1340 | ||
1da177e4 | 1341 | /* |
e2bdb933 HD |
1342 | * Clear all tags associated with the item to be deleted. |
1343 | * This way of doing it would be inefficient, but seldom is any set. | |
1da177e4 | 1344 | */ |
daff89f3 | 1345 | for (tag = 0; tag < RADIX_TREE_MAX_TAGS; tag++) { |
e2bdb933 | 1346 | if (tag_get(node, tag, offset)) |
612d6c19 | 1347 | radix_tree_tag_clear(root, index, tag); |
d5274261 | 1348 | } |
1da177e4 | 1349 | |
7cf9c2c7 | 1350 | to_free = NULL; |
201b6264 | 1351 | /* Now free the nodes we do not need anymore */ |
e2bdb933 HD |
1352 | while (node) { |
1353 | node->slots[offset] = NULL; | |
1354 | node->count--; | |
7cf9c2c7 NP |
1355 | /* |
1356 | * Queue the node for deferred freeing after the | |
1357 | * last reference to it disappears (set NULL, above). | |
1358 | */ | |
1359 | if (to_free) | |
1360 | radix_tree_node_free(to_free); | |
a5f51c96 | 1361 | |
e2bdb933 HD |
1362 | if (node->count) { |
1363 | if (node == indirect_to_ptr(root->rnode)) | |
a5f51c96 | 1364 | radix_tree_shrink(root); |
201b6264 | 1365 | goto out; |
a5f51c96 | 1366 | } |
201b6264 CL |
1367 | |
1368 | /* Node with zero slots in use so free it */ | |
e2bdb933 | 1369 | to_free = node; |
7cf9c2c7 | 1370 | |
e2bdb933 HD |
1371 | index >>= RADIX_TREE_MAP_SHIFT; |
1372 | offset = index & RADIX_TREE_MAP_MASK; | |
1373 | node = node->parent; | |
1da177e4 | 1374 | } |
e2bdb933 | 1375 | |
612d6c19 | 1376 | root_tag_clear_all(root); |
201b6264 | 1377 | root->height = 0; |
612d6c19 | 1378 | root->rnode = NULL; |
7cf9c2c7 NP |
1379 | if (to_free) |
1380 | radix_tree_node_free(to_free); | |
612d6c19 | 1381 | |
1da177e4 | 1382 | out: |
612d6c19 | 1383 | return slot; |
1da177e4 LT |
1384 | } |
1385 | EXPORT_SYMBOL(radix_tree_delete); | |
1386 | ||
1387 | /** | |
1388 | * radix_tree_tagged - test whether any items in the tree are tagged | |
1389 | * @root: radix tree root | |
1390 | * @tag: tag to test | |
1391 | */ | |
daff89f3 | 1392 | int radix_tree_tagged(struct radix_tree_root *root, unsigned int tag) |
1da177e4 | 1393 | { |
612d6c19 | 1394 | return root_tag_get(root, tag); |
1da177e4 LT |
1395 | } |
1396 | EXPORT_SYMBOL(radix_tree_tagged); | |
1397 | ||
1398 | static void | |
51cc5068 | 1399 | radix_tree_node_ctor(void *node) |
1da177e4 LT |
1400 | { |
1401 | memset(node, 0, sizeof(struct radix_tree_node)); | |
1402 | } | |
1403 | ||
1404 | static __init unsigned long __maxindex(unsigned int height) | |
1405 | { | |
430d275a PL |
1406 | unsigned int width = height * RADIX_TREE_MAP_SHIFT; |
1407 | int shift = RADIX_TREE_INDEX_BITS - width; | |
1408 | ||
1409 | if (shift < 0) | |
1410 | return ~0UL; | |
1411 | if (shift >= BITS_PER_LONG) | |
1412 | return 0UL; | |
1413 | return ~0UL >> shift; | |
1da177e4 LT |
1414 | } |
1415 | ||
1416 | static __init void radix_tree_init_maxindex(void) | |
1417 | { | |
1418 | unsigned int i; | |
1419 | ||
1420 | for (i = 0; i < ARRAY_SIZE(height_to_maxindex); i++) | |
1421 | height_to_maxindex[i] = __maxindex(i); | |
1422 | } | |
1423 | ||
1da177e4 LT |
1424 | static int radix_tree_callback(struct notifier_block *nfb, |
1425 | unsigned long action, | |
1426 | void *hcpu) | |
1427 | { | |
1428 | int cpu = (long)hcpu; | |
1429 | struct radix_tree_preload *rtp; | |
1430 | ||
1431 | /* Free per-cpu pool of perloaded nodes */ | |
8bb78442 | 1432 | if (action == CPU_DEAD || action == CPU_DEAD_FROZEN) { |
1da177e4 LT |
1433 | rtp = &per_cpu(radix_tree_preloads, cpu); |
1434 | while (rtp->nr) { | |
1435 | kmem_cache_free(radix_tree_node_cachep, | |
1436 | rtp->nodes[rtp->nr-1]); | |
1437 | rtp->nodes[rtp->nr-1] = NULL; | |
1438 | rtp->nr--; | |
1439 | } | |
1440 | } | |
1441 | return NOTIFY_OK; | |
1442 | } | |
1da177e4 LT |
1443 | |
1444 | void __init radix_tree_init(void) | |
1445 | { | |
1446 | radix_tree_node_cachep = kmem_cache_create("radix_tree_node", | |
1447 | sizeof(struct radix_tree_node), 0, | |
488514d1 CL |
1448 | SLAB_PANIC | SLAB_RECLAIM_ACCOUNT, |
1449 | radix_tree_node_ctor); | |
1da177e4 LT |
1450 | radix_tree_init_maxindex(); |
1451 | hotcpu_notifier(radix_tree_callback, 0); | |
1452 | } |