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 LT |
579 | int saw_unset_tag = 0; |
580 | ||
612d6c19 NP |
581 | /* check the root's tag bit */ |
582 | if (!root_tag_get(root, tag)) | |
583 | return 0; | |
584 | ||
2676a58c | 585 | node = rcu_dereference_raw(root->rnode); |
7cf9c2c7 NP |
586 | if (node == NULL) |
587 | return 0; | |
588 | ||
c0bc9875 | 589 | if (!radix_tree_is_indirect_ptr(node)) |
7cf9c2c7 | 590 | return (index == 0); |
27d20fdd | 591 | node = indirect_to_ptr(node); |
7cf9c2c7 NP |
592 | |
593 | height = node->height; | |
594 | if (index > radix_tree_maxindex(height)) | |
595 | return 0; | |
612d6c19 | 596 | |
1da177e4 | 597 | shift = (height - 1) * RADIX_TREE_MAP_SHIFT; |
1da177e4 LT |
598 | |
599 | for ( ; ; ) { | |
600 | int offset; | |
601 | ||
7cf9c2c7 | 602 | if (node == NULL) |
1da177e4 LT |
603 | return 0; |
604 | ||
605 | offset = (index >> shift) & RADIX_TREE_MAP_MASK; | |
606 | ||
607 | /* | |
608 | * This is just a debug check. Later, we can bale as soon as | |
609 | * we see an unset tag. | |
610 | */ | |
7cf9c2c7 | 611 | if (!tag_get(node, tag, offset)) |
1da177e4 | 612 | saw_unset_tag = 1; |
ce82653d DH |
613 | if (height == 1) |
614 | return !!tag_get(node, tag, offset); | |
2676a58c | 615 | node = rcu_dereference_raw(node->slots[offset]); |
1da177e4 LT |
616 | shift -= RADIX_TREE_MAP_SHIFT; |
617 | height--; | |
618 | } | |
619 | } | |
620 | EXPORT_SYMBOL(radix_tree_tag_get); | |
1da177e4 | 621 | |
ebf8aa44 JK |
622 | /** |
623 | * radix_tree_range_tag_if_tagged - for each item in given range set given | |
624 | * tag if item has another tag set | |
625 | * @root: radix tree root | |
626 | * @first_indexp: pointer to a starting index of a range to scan | |
627 | * @last_index: last index of a range to scan | |
628 | * @nr_to_tag: maximum number items to tag | |
629 | * @iftag: tag index to test | |
630 | * @settag: tag index to set if tested tag is set | |
631 | * | |
632 | * This function scans range of radix tree from first_index to last_index | |
633 | * (inclusive). For each item in the range if iftag is set, the function sets | |
634 | * also settag. The function stops either after tagging nr_to_tag items or | |
635 | * after reaching last_index. | |
636 | * | |
144dcfc0 DC |
637 | * The tags must be set from the leaf level only and propagated back up the |
638 | * path to the root. We must do this so that we resolve the full path before | |
639 | * setting any tags on intermediate nodes. If we set tags as we descend, then | |
640 | * we can get to the leaf node and find that the index that has the iftag | |
641 | * set is outside the range we are scanning. This reults in dangling tags and | |
642 | * can lead to problems with later tag operations (e.g. livelocks on lookups). | |
643 | * | |
ebf8aa44 JK |
644 | * The function returns number of leaves where the tag was set and sets |
645 | * *first_indexp to the first unscanned index. | |
d5ed3a4a JK |
646 | * WARNING! *first_indexp can wrap if last_index is ULONG_MAX. Caller must |
647 | * be prepared to handle that. | |
ebf8aa44 JK |
648 | */ |
649 | unsigned long radix_tree_range_tag_if_tagged(struct radix_tree_root *root, | |
650 | unsigned long *first_indexp, unsigned long last_index, | |
651 | unsigned long nr_to_tag, | |
652 | unsigned int iftag, unsigned int settag) | |
653 | { | |
144dcfc0 DC |
654 | unsigned int height = root->height; |
655 | struct radix_tree_path path[height]; | |
656 | struct radix_tree_path *pathp = path; | |
657 | struct radix_tree_node *slot; | |
658 | unsigned int shift; | |
659 | unsigned long tagged = 0; | |
660 | unsigned long index = *first_indexp; | |
ebf8aa44 JK |
661 | |
662 | last_index = min(last_index, radix_tree_maxindex(height)); | |
663 | if (index > last_index) | |
664 | return 0; | |
665 | if (!nr_to_tag) | |
666 | return 0; | |
667 | if (!root_tag_get(root, iftag)) { | |
668 | *first_indexp = last_index + 1; | |
669 | return 0; | |
670 | } | |
671 | if (height == 0) { | |
672 | *first_indexp = last_index + 1; | |
673 | root_tag_set(root, settag); | |
674 | return 1; | |
675 | } | |
676 | ||
677 | shift = (height - 1) * RADIX_TREE_MAP_SHIFT; | |
27d20fdd | 678 | slot = indirect_to_ptr(root->rnode); |
ebf8aa44 | 679 | |
144dcfc0 DC |
680 | /* |
681 | * we fill the path from (root->height - 2) to 0, leaving the index at | |
682 | * (root->height - 1) as a terminator. Zero the node in the terminator | |
683 | * so that we can use this to end walk loops back up the path. | |
684 | */ | |
685 | path[height - 1].node = NULL; | |
686 | ||
ebf8aa44 JK |
687 | for (;;) { |
688 | int offset; | |
689 | ||
690 | offset = (index >> shift) & RADIX_TREE_MAP_MASK; | |
691 | if (!slot->slots[offset]) | |
692 | goto next; | |
693 | if (!tag_get(slot, iftag, offset)) | |
694 | goto next; | |
144dcfc0 DC |
695 | if (height > 1) { |
696 | /* Go down one level */ | |
697 | height--; | |
698 | shift -= RADIX_TREE_MAP_SHIFT; | |
699 | path[height - 1].node = slot; | |
700 | path[height - 1].offset = offset; | |
701 | slot = slot->slots[offset]; | |
702 | continue; | |
703 | } | |
704 | ||
705 | /* tag the leaf */ | |
706 | tagged++; | |
ebf8aa44 | 707 | tag_set(slot, settag, offset); |
144dcfc0 DC |
708 | |
709 | /* walk back up the path tagging interior nodes */ | |
710 | pathp = &path[0]; | |
711 | while (pathp->node) { | |
712 | /* stop if we find a node with the tag already set */ | |
713 | if (tag_get(pathp->node, settag, pathp->offset)) | |
714 | break; | |
715 | tag_set(pathp->node, settag, pathp->offset); | |
716 | pathp++; | |
ebf8aa44 | 717 | } |
144dcfc0 | 718 | |
ebf8aa44 JK |
719 | next: |
720 | /* Go to next item at level determined by 'shift' */ | |
721 | index = ((index >> shift) + 1) << shift; | |
d5ed3a4a JK |
722 | /* Overflow can happen when last_index is ~0UL... */ |
723 | if (index > last_index || !index) | |
ebf8aa44 JK |
724 | break; |
725 | if (tagged >= nr_to_tag) | |
726 | break; | |
727 | while (((index >> shift) & RADIX_TREE_MAP_MASK) == 0) { | |
728 | /* | |
729 | * We've fully scanned this node. Go up. Because | |
730 | * last_index is guaranteed to be in the tree, what | |
731 | * we do below cannot wander astray. | |
732 | */ | |
144dcfc0 | 733 | slot = path[height - 1].node; |
ebf8aa44 JK |
734 | height++; |
735 | shift += RADIX_TREE_MAP_SHIFT; | |
736 | } | |
737 | } | |
738 | /* | |
739 | * The iftag must have been set somewhere because otherwise | |
740 | * we would return immediated at the beginning of the function | |
741 | */ | |
742 | root_tag_set(root, settag); | |
743 | *first_indexp = index; | |
744 | ||
745 | return tagged; | |
746 | } | |
747 | EXPORT_SYMBOL(radix_tree_range_tag_if_tagged); | |
748 | ||
749 | ||
6df8ba4f FW |
750 | /** |
751 | * radix_tree_next_hole - find the next hole (not-present entry) | |
752 | * @root: tree root | |
753 | * @index: index key | |
754 | * @max_scan: maximum range to search | |
755 | * | |
756 | * Search the set [index, min(index+max_scan-1, MAX_INDEX)] for the lowest | |
757 | * indexed hole. | |
758 | * | |
759 | * Returns: the index of the hole if found, otherwise returns an index | |
760 | * outside of the set specified (in which case 'return - index >= max_scan' | |
8e6bdb7f | 761 | * will be true). In rare cases of index wrap-around, 0 will be returned. |
6df8ba4f FW |
762 | * |
763 | * radix_tree_next_hole may be called under rcu_read_lock. However, like | |
8e6bdb7f WF |
764 | * radix_tree_gang_lookup, this will not atomically search a snapshot of |
765 | * the tree at a single point in time. For example, if a hole is created | |
766 | * at index 5, then subsequently a hole is created at index 10, | |
767 | * radix_tree_next_hole covering both indexes may return 10 if called | |
768 | * under rcu_read_lock. | |
6df8ba4f FW |
769 | */ |
770 | unsigned long radix_tree_next_hole(struct radix_tree_root *root, | |
771 | unsigned long index, unsigned long max_scan) | |
772 | { | |
773 | unsigned long i; | |
774 | ||
775 | for (i = 0; i < max_scan; i++) { | |
776 | if (!radix_tree_lookup(root, index)) | |
777 | break; | |
778 | index++; | |
779 | if (index == 0) | |
780 | break; | |
781 | } | |
782 | ||
783 | return index; | |
784 | } | |
785 | EXPORT_SYMBOL(radix_tree_next_hole); | |
786 | ||
dc566127 WF |
787 | /** |
788 | * radix_tree_prev_hole - find the prev hole (not-present entry) | |
789 | * @root: tree root | |
790 | * @index: index key | |
791 | * @max_scan: maximum range to search | |
792 | * | |
793 | * Search backwards in the range [max(index-max_scan+1, 0), index] | |
794 | * for the first hole. | |
795 | * | |
796 | * Returns: the index of the hole if found, otherwise returns an index | |
797 | * outside of the set specified (in which case 'index - return >= max_scan' | |
edcd1d84 | 798 | * will be true). In rare cases of wrap-around, ULONG_MAX will be returned. |
dc566127 WF |
799 | * |
800 | * radix_tree_next_hole may be called under rcu_read_lock. However, like | |
801 | * radix_tree_gang_lookup, this will not atomically search a snapshot of | |
802 | * the tree at a single point in time. For example, if a hole is created | |
803 | * at index 10, then subsequently a hole is created at index 5, | |
804 | * radix_tree_prev_hole covering both indexes may return 5 if called under | |
805 | * rcu_read_lock. | |
806 | */ | |
807 | unsigned long radix_tree_prev_hole(struct radix_tree_root *root, | |
808 | unsigned long index, unsigned long max_scan) | |
809 | { | |
810 | unsigned long i; | |
811 | ||
812 | for (i = 0; i < max_scan; i++) { | |
813 | if (!radix_tree_lookup(root, index)) | |
814 | break; | |
815 | index--; | |
edcd1d84 | 816 | if (index == ULONG_MAX) |
dc566127 WF |
817 | break; |
818 | } | |
819 | ||
820 | return index; | |
821 | } | |
822 | EXPORT_SYMBOL(radix_tree_prev_hole); | |
823 | ||
1da177e4 | 824 | static unsigned int |
47feff2c | 825 | __lookup(struct radix_tree_node *slot, void ***results, unsigned long index, |
1da177e4 LT |
826 | unsigned int max_items, unsigned long *next_index) |
827 | { | |
828 | unsigned int nr_found = 0; | |
201b6264 | 829 | unsigned int shift, height; |
201b6264 CL |
830 | unsigned long i; |
831 | ||
7cf9c2c7 NP |
832 | height = slot->height; |
833 | if (height == 0) | |
201b6264 | 834 | goto out; |
1da177e4 | 835 | shift = (height-1) * RADIX_TREE_MAP_SHIFT; |
1da177e4 | 836 | |
201b6264 | 837 | for ( ; height > 1; height--) { |
7cf9c2c7 NP |
838 | i = (index >> shift) & RADIX_TREE_MAP_MASK; |
839 | for (;;) { | |
1da177e4 LT |
840 | if (slot->slots[i] != NULL) |
841 | break; | |
842 | index &= ~((1UL << shift) - 1); | |
843 | index += 1UL << shift; | |
844 | if (index == 0) | |
845 | goto out; /* 32-bit wraparound */ | |
7cf9c2c7 NP |
846 | i++; |
847 | if (i == RADIX_TREE_MAP_SIZE) | |
848 | goto out; | |
1da177e4 | 849 | } |
1da177e4 | 850 | |
1da177e4 | 851 | shift -= RADIX_TREE_MAP_SHIFT; |
2676a58c | 852 | slot = rcu_dereference_raw(slot->slots[i]); |
7cf9c2c7 NP |
853 | if (slot == NULL) |
854 | goto out; | |
1da177e4 | 855 | } |
201b6264 CL |
856 | |
857 | /* Bottom level: grab some items */ | |
858 | for (i = index & RADIX_TREE_MAP_MASK; i < RADIX_TREE_MAP_SIZE; i++) { | |
859 | index++; | |
47feff2c NP |
860 | if (slot->slots[i]) { |
861 | results[nr_found++] = &(slot->slots[i]); | |
201b6264 CL |
862 | if (nr_found == max_items) |
863 | goto out; | |
864 | } | |
865 | } | |
1da177e4 LT |
866 | out: |
867 | *next_index = index; | |
868 | return nr_found; | |
869 | } | |
870 | ||
871 | /** | |
872 | * radix_tree_gang_lookup - perform multiple lookup on a radix tree | |
873 | * @root: radix tree root | |
874 | * @results: where the results of the lookup are placed | |
875 | * @first_index: start the lookup from this key | |
876 | * @max_items: place up to this many items at *results | |
877 | * | |
878 | * Performs an index-ascending scan of the tree for present items. Places | |
879 | * them at *@results and returns the number of items which were placed at | |
880 | * *@results. | |
881 | * | |
882 | * The implementation is naive. | |
7cf9c2c7 NP |
883 | * |
884 | * Like radix_tree_lookup, radix_tree_gang_lookup may be called under | |
885 | * rcu_read_lock. In this case, rather than the returned results being | |
886 | * an atomic snapshot of the tree at a single point in time, the semantics | |
887 | * of an RCU protected gang lookup are as though multiple radix_tree_lookups | |
888 | * have been issued in individual locks, and results stored in 'results'. | |
1da177e4 LT |
889 | */ |
890 | unsigned int | |
891 | radix_tree_gang_lookup(struct radix_tree_root *root, void **results, | |
892 | unsigned long first_index, unsigned int max_items) | |
893 | { | |
7cf9c2c7 NP |
894 | unsigned long max_index; |
895 | struct radix_tree_node *node; | |
1da177e4 | 896 | unsigned long cur_index = first_index; |
7cf9c2c7 NP |
897 | unsigned int ret; |
898 | ||
2676a58c | 899 | node = rcu_dereference_raw(root->rnode); |
7cf9c2c7 NP |
900 | if (!node) |
901 | return 0; | |
1da177e4 | 902 | |
c0bc9875 | 903 | if (!radix_tree_is_indirect_ptr(node)) { |
7cf9c2c7 NP |
904 | if (first_index > 0) |
905 | return 0; | |
c0bc9875 | 906 | results[0] = node; |
7cf9c2c7 NP |
907 | return 1; |
908 | } | |
27d20fdd | 909 | node = indirect_to_ptr(node); |
7cf9c2c7 NP |
910 | |
911 | max_index = radix_tree_maxindex(node->height); | |
912 | ||
913 | ret = 0; | |
1da177e4 | 914 | while (ret < max_items) { |
47feff2c | 915 | unsigned int nr_found, slots_found, i; |
1da177e4 LT |
916 | unsigned long next_index; /* Index of next search */ |
917 | ||
918 | if (cur_index > max_index) | |
919 | break; | |
47feff2c | 920 | slots_found = __lookup(node, (void ***)results + ret, cur_index, |
1da177e4 | 921 | max_items - ret, &next_index); |
47feff2c NP |
922 | nr_found = 0; |
923 | for (i = 0; i < slots_found; i++) { | |
924 | struct radix_tree_node *slot; | |
925 | slot = *(((void ***)results)[ret + i]); | |
926 | if (!slot) | |
927 | continue; | |
27d20fdd NP |
928 | results[ret + nr_found] = |
929 | indirect_to_ptr(rcu_dereference_raw(slot)); | |
47feff2c NP |
930 | nr_found++; |
931 | } | |
1da177e4 LT |
932 | ret += nr_found; |
933 | if (next_index == 0) | |
934 | break; | |
935 | cur_index = next_index; | |
936 | } | |
7cf9c2c7 | 937 | |
1da177e4 LT |
938 | return ret; |
939 | } | |
940 | EXPORT_SYMBOL(radix_tree_gang_lookup); | |
941 | ||
47feff2c NP |
942 | /** |
943 | * radix_tree_gang_lookup_slot - perform multiple slot lookup on radix tree | |
944 | * @root: radix tree root | |
945 | * @results: where the results of the lookup are placed | |
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 | |
960 | radix_tree_gang_lookup_slot(struct radix_tree_root *root, void ***results, | |
961 | unsigned long first_index, unsigned int max_items) | |
962 | { | |
963 | unsigned long max_index; | |
964 | struct radix_tree_node *node; | |
965 | unsigned long cur_index = first_index; | |
966 | unsigned int ret; | |
967 | ||
2676a58c | 968 | node = rcu_dereference_raw(root->rnode); |
47feff2c NP |
969 | if (!node) |
970 | return 0; | |
971 | ||
972 | if (!radix_tree_is_indirect_ptr(node)) { | |
973 | if (first_index > 0) | |
974 | return 0; | |
975 | results[0] = (void **)&root->rnode; | |
976 | return 1; | |
977 | } | |
27d20fdd | 978 | node = indirect_to_ptr(node); |
47feff2c NP |
979 | |
980 | max_index = radix_tree_maxindex(node->height); | |
981 | ||
982 | ret = 0; | |
983 | while (ret < max_items) { | |
984 | unsigned int slots_found; | |
985 | unsigned long next_index; /* Index of next search */ | |
986 | ||
987 | if (cur_index > max_index) | |
988 | break; | |
989 | slots_found = __lookup(node, results + ret, cur_index, | |
990 | max_items - ret, &next_index); | |
991 | ret += slots_found; | |
992 | if (next_index == 0) | |
993 | break; | |
994 | cur_index = next_index; | |
995 | } | |
996 | ||
997 | return ret; | |
998 | } | |
999 | EXPORT_SYMBOL(radix_tree_gang_lookup_slot); | |
1000 | ||
1da177e4 LT |
1001 | /* |
1002 | * FIXME: the two tag_get()s here should use find_next_bit() instead of | |
1003 | * open-coding the search. | |
1004 | */ | |
1005 | static unsigned int | |
47feff2c | 1006 | __lookup_tag(struct radix_tree_node *slot, void ***results, unsigned long index, |
daff89f3 | 1007 | unsigned int max_items, unsigned long *next_index, unsigned int tag) |
1da177e4 LT |
1008 | { |
1009 | unsigned int nr_found = 0; | |
7cf9c2c7 | 1010 | unsigned int shift, height; |
1da177e4 | 1011 | |
7cf9c2c7 NP |
1012 | height = slot->height; |
1013 | if (height == 0) | |
612d6c19 | 1014 | goto out; |
7cf9c2c7 | 1015 | shift = (height-1) * RADIX_TREE_MAP_SHIFT; |
1da177e4 | 1016 | |
7cf9c2c7 NP |
1017 | while (height > 0) { |
1018 | unsigned long i = (index >> shift) & RADIX_TREE_MAP_MASK ; | |
1da177e4 | 1019 | |
7cf9c2c7 NP |
1020 | for (;;) { |
1021 | if (tag_get(slot, tag, i)) | |
1da177e4 | 1022 | break; |
1da177e4 LT |
1023 | index &= ~((1UL << shift) - 1); |
1024 | index += 1UL << shift; | |
1025 | if (index == 0) | |
1026 | goto out; /* 32-bit wraparound */ | |
7cf9c2c7 NP |
1027 | i++; |
1028 | if (i == RADIX_TREE_MAP_SIZE) | |
1029 | goto out; | |
1da177e4 | 1030 | } |
1da177e4 LT |
1031 | height--; |
1032 | if (height == 0) { /* Bottom level: grab some items */ | |
1033 | unsigned long j = index & RADIX_TREE_MAP_MASK; | |
1034 | ||
1035 | for ( ; j < RADIX_TREE_MAP_SIZE; j++) { | |
1036 | index++; | |
7cf9c2c7 NP |
1037 | if (!tag_get(slot, tag, j)) |
1038 | continue; | |
7cf9c2c7 NP |
1039 | /* |
1040 | * Even though the tag was found set, we need to | |
1041 | * recheck that we have a non-NULL node, because | |
1042 | * if this lookup is lockless, it may have been | |
1043 | * subsequently deleted. | |
1044 | * | |
1045 | * Similar care must be taken in any place that | |
1046 | * lookup ->slots[x] without a lock (ie. can't | |
1047 | * rely on its value remaining the same). | |
1048 | */ | |
47feff2c NP |
1049 | if (slot->slots[j]) { |
1050 | results[nr_found++] = &(slot->slots[j]); | |
1da177e4 LT |
1051 | if (nr_found == max_items) |
1052 | goto out; | |
1053 | } | |
1054 | } | |
1055 | } | |
1056 | shift -= RADIX_TREE_MAP_SHIFT; | |
2676a58c | 1057 | slot = rcu_dereference_raw(slot->slots[i]); |
7cf9c2c7 NP |
1058 | if (slot == NULL) |
1059 | break; | |
1060 | } | |
1da177e4 LT |
1061 | out: |
1062 | *next_index = index; | |
1063 | return nr_found; | |
1064 | } | |
1065 | ||
1066 | /** | |
1067 | * radix_tree_gang_lookup_tag - perform multiple lookup on a radix tree | |
1068 | * based on a tag | |
1069 | * @root: radix tree root | |
1070 | * @results: where the results of the lookup are placed | |
1071 | * @first_index: start the lookup from this key | |
1072 | * @max_items: place up to this many items at *results | |
daff89f3 | 1073 | * @tag: the tag index (< RADIX_TREE_MAX_TAGS) |
1da177e4 LT |
1074 | * |
1075 | * Performs an index-ascending scan of the tree for present items which | |
1076 | * have the tag indexed by @tag set. Places the items at *@results and | |
1077 | * returns the number of items which were placed at *@results. | |
1078 | */ | |
1079 | unsigned int | |
1080 | radix_tree_gang_lookup_tag(struct radix_tree_root *root, void **results, | |
daff89f3 JC |
1081 | unsigned long first_index, unsigned int max_items, |
1082 | unsigned int tag) | |
1da177e4 | 1083 | { |
7cf9c2c7 NP |
1084 | struct radix_tree_node *node; |
1085 | unsigned long max_index; | |
1da177e4 | 1086 | unsigned long cur_index = first_index; |
7cf9c2c7 | 1087 | unsigned int ret; |
1da177e4 | 1088 | |
612d6c19 NP |
1089 | /* check the root's tag bit */ |
1090 | if (!root_tag_get(root, tag)) | |
1091 | return 0; | |
1092 | ||
2676a58c | 1093 | node = rcu_dereference_raw(root->rnode); |
7cf9c2c7 NP |
1094 | if (!node) |
1095 | return 0; | |
1096 | ||
c0bc9875 | 1097 | if (!radix_tree_is_indirect_ptr(node)) { |
7cf9c2c7 NP |
1098 | if (first_index > 0) |
1099 | return 0; | |
c0bc9875 | 1100 | results[0] = node; |
7cf9c2c7 NP |
1101 | return 1; |
1102 | } | |
27d20fdd | 1103 | node = indirect_to_ptr(node); |
7cf9c2c7 NP |
1104 | |
1105 | max_index = radix_tree_maxindex(node->height); | |
1106 | ||
1107 | ret = 0; | |
1da177e4 | 1108 | while (ret < max_items) { |
47feff2c | 1109 | unsigned int nr_found, slots_found, i; |
1da177e4 LT |
1110 | unsigned long next_index; /* Index of next search */ |
1111 | ||
1112 | if (cur_index > max_index) | |
1113 | break; | |
47feff2c NP |
1114 | slots_found = __lookup_tag(node, (void ***)results + ret, |
1115 | cur_index, max_items - ret, &next_index, tag); | |
1116 | nr_found = 0; | |
1117 | for (i = 0; i < slots_found; i++) { | |
1118 | struct radix_tree_node *slot; | |
1119 | slot = *(((void ***)results)[ret + i]); | |
1120 | if (!slot) | |
1121 | continue; | |
27d20fdd NP |
1122 | results[ret + nr_found] = |
1123 | indirect_to_ptr(rcu_dereference_raw(slot)); | |
47feff2c NP |
1124 | nr_found++; |
1125 | } | |
1da177e4 LT |
1126 | ret += nr_found; |
1127 | if (next_index == 0) | |
1128 | break; | |
1129 | cur_index = next_index; | |
1130 | } | |
7cf9c2c7 | 1131 | |
1da177e4 LT |
1132 | return ret; |
1133 | } | |
1134 | EXPORT_SYMBOL(radix_tree_gang_lookup_tag); | |
1135 | ||
47feff2c NP |
1136 | /** |
1137 | * radix_tree_gang_lookup_tag_slot - perform multiple slot lookup on a | |
1138 | * radix tree based on a tag | |
1139 | * @root: radix tree root | |
1140 | * @results: where the results of the lookup are placed | |
1141 | * @first_index: start the lookup from this key | |
1142 | * @max_items: place up to this many items at *results | |
1143 | * @tag: the tag index (< RADIX_TREE_MAX_TAGS) | |
1144 | * | |
1145 | * Performs an index-ascending scan of the tree for present items which | |
1146 | * have the tag indexed by @tag set. Places the slots at *@results and | |
1147 | * returns the number of slots which were placed at *@results. | |
1148 | */ | |
1149 | unsigned int | |
1150 | radix_tree_gang_lookup_tag_slot(struct radix_tree_root *root, void ***results, | |
1151 | unsigned long first_index, unsigned int max_items, | |
1152 | unsigned int tag) | |
1153 | { | |
1154 | struct radix_tree_node *node; | |
1155 | unsigned long max_index; | |
1156 | unsigned long cur_index = first_index; | |
1157 | unsigned int ret; | |
1158 | ||
1159 | /* check the root's tag bit */ | |
1160 | if (!root_tag_get(root, tag)) | |
1161 | return 0; | |
1162 | ||
2676a58c | 1163 | node = rcu_dereference_raw(root->rnode); |
47feff2c NP |
1164 | if (!node) |
1165 | return 0; | |
1166 | ||
1167 | if (!radix_tree_is_indirect_ptr(node)) { | |
1168 | if (first_index > 0) | |
1169 | return 0; | |
1170 | results[0] = (void **)&root->rnode; | |
1171 | return 1; | |
1172 | } | |
27d20fdd | 1173 | node = indirect_to_ptr(node); |
47feff2c NP |
1174 | |
1175 | max_index = radix_tree_maxindex(node->height); | |
1176 | ||
1177 | ret = 0; | |
1178 | while (ret < max_items) { | |
1179 | unsigned int slots_found; | |
1180 | unsigned long next_index; /* Index of next search */ | |
1181 | ||
1182 | if (cur_index > max_index) | |
1183 | break; | |
1184 | slots_found = __lookup_tag(node, results + ret, | |
1185 | cur_index, max_items - ret, &next_index, tag); | |
1186 | ret += slots_found; | |
1187 | if (next_index == 0) | |
1188 | break; | |
1189 | cur_index = next_index; | |
1190 | } | |
1191 | ||
1192 | return ret; | |
1193 | } | |
1194 | EXPORT_SYMBOL(radix_tree_gang_lookup_tag_slot); | |
1195 | ||
1196 | ||
a5f51c96 NP |
1197 | /** |
1198 | * radix_tree_shrink - shrink height of a radix tree to minimal | |
1199 | * @root radix tree root | |
1200 | */ | |
1201 | static inline void radix_tree_shrink(struct radix_tree_root *root) | |
1202 | { | |
1203 | /* try to shrink tree height */ | |
c0bc9875 | 1204 | while (root->height > 0) { |
a5f51c96 | 1205 | struct radix_tree_node *to_free = root->rnode; |
7cf9c2c7 | 1206 | void *newptr; |
a5f51c96 | 1207 | |
c0bc9875 | 1208 | BUG_ON(!radix_tree_is_indirect_ptr(to_free)); |
27d20fdd | 1209 | to_free = indirect_to_ptr(to_free); |
c0bc9875 NP |
1210 | |
1211 | /* | |
1212 | * The candidate node has more than one child, or its child | |
1213 | * is not at the leftmost slot, we cannot shrink. | |
1214 | */ | |
1215 | if (to_free->count != 1) | |
1216 | break; | |
1217 | if (!to_free->slots[0]) | |
1218 | break; | |
1219 | ||
7cf9c2c7 NP |
1220 | /* |
1221 | * We don't need rcu_assign_pointer(), since we are simply | |
27d20fdd NP |
1222 | * moving the node from one part of the tree to another: if it |
1223 | * was safe to dereference the old pointer to it | |
7cf9c2c7 | 1224 | * (to_free->slots[0]), it will be safe to dereference the new |
27d20fdd | 1225 | * one (root->rnode) as far as dependent read barriers go. |
7cf9c2c7 NP |
1226 | */ |
1227 | newptr = to_free->slots[0]; | |
c0bc9875 | 1228 | if (root->height > 1) |
27d20fdd | 1229 | newptr = ptr_to_indirect(newptr); |
7cf9c2c7 | 1230 | root->rnode = newptr; |
a5f51c96 | 1231 | root->height--; |
27d20fdd NP |
1232 | |
1233 | /* | |
1234 | * We have a dilemma here. The node's slot[0] must not be | |
1235 | * NULLed in case there are concurrent lookups expecting to | |
1236 | * find the item. However if this was a bottom-level node, | |
1237 | * then it may be subject to the slot pointer being visible | |
1238 | * to callers dereferencing it. If item corresponding to | |
1239 | * slot[0] is subsequently deleted, these callers would expect | |
1240 | * their slot to become empty sooner or later. | |
1241 | * | |
1242 | * For example, lockless pagecache will look up a slot, deref | |
1243 | * the page pointer, and if the page is 0 refcount it means it | |
1244 | * was concurrently deleted from pagecache so try the deref | |
1245 | * again. Fortunately there is already a requirement for logic | |
1246 | * to retry the entire slot lookup -- the indirect pointer | |
1247 | * problem (replacing direct root node with an indirect pointer | |
1248 | * also results in a stale slot). So tag the slot as indirect | |
1249 | * to force callers to retry. | |
1250 | */ | |
1251 | if (root->height == 0) | |
1252 | *((unsigned long *)&to_free->slots[0]) |= | |
1253 | RADIX_TREE_INDIRECT_PTR; | |
1254 | ||
a5f51c96 NP |
1255 | radix_tree_node_free(to_free); |
1256 | } | |
1257 | } | |
1258 | ||
1da177e4 LT |
1259 | /** |
1260 | * radix_tree_delete - delete an item from a radix tree | |
1261 | * @root: radix tree root | |
1262 | * @index: index key | |
1263 | * | |
1264 | * Remove the item at @index from the radix tree rooted at @root. | |
1265 | * | |
1266 | * Returns the address of the deleted item, or NULL if it was not present. | |
1267 | */ | |
1268 | void *radix_tree_delete(struct radix_tree_root *root, unsigned long index) | |
1269 | { | |
26fb1589 JM |
1270 | /* |
1271 | * The radix tree path needs to be one longer than the maximum path | |
1272 | * since the "list" is null terminated. | |
1273 | */ | |
1274 | struct radix_tree_path path[RADIX_TREE_MAX_PATH + 1], *pathp = path; | |
612d6c19 | 1275 | struct radix_tree_node *slot = NULL; |
7cf9c2c7 | 1276 | struct radix_tree_node *to_free; |
1da177e4 | 1277 | unsigned int height, shift; |
d5274261 NP |
1278 | int tag; |
1279 | int offset; | |
1da177e4 LT |
1280 | |
1281 | height = root->height; | |
1282 | if (index > radix_tree_maxindex(height)) | |
1283 | goto out; | |
1284 | ||
612d6c19 | 1285 | slot = root->rnode; |
c0bc9875 | 1286 | if (height == 0) { |
612d6c19 NP |
1287 | root_tag_clear_all(root); |
1288 | root->rnode = NULL; | |
1289 | goto out; | |
1290 | } | |
27d20fdd | 1291 | slot = indirect_to_ptr(slot); |
612d6c19 | 1292 | |
1da177e4 LT |
1293 | shift = (height - 1) * RADIX_TREE_MAP_SHIFT; |
1294 | pathp->node = NULL; | |
1da177e4 | 1295 | |
612d6c19 | 1296 | do { |
201b6264 | 1297 | if (slot == NULL) |
1da177e4 LT |
1298 | goto out; |
1299 | ||
d5274261 | 1300 | pathp++; |
1da177e4 | 1301 | offset = (index >> shift) & RADIX_TREE_MAP_MASK; |
d5274261 NP |
1302 | pathp->offset = offset; |
1303 | pathp->node = slot; | |
201b6264 | 1304 | slot = slot->slots[offset]; |
1da177e4 | 1305 | shift -= RADIX_TREE_MAP_SHIFT; |
612d6c19 NP |
1306 | height--; |
1307 | } while (height > 0); | |
1da177e4 | 1308 | |
612d6c19 | 1309 | if (slot == NULL) |
1da177e4 LT |
1310 | goto out; |
1311 | ||
1da177e4 LT |
1312 | /* |
1313 | * Clear all tags associated with the just-deleted item | |
1314 | */ | |
daff89f3 | 1315 | for (tag = 0; tag < RADIX_TREE_MAX_TAGS; tag++) { |
612d6c19 NP |
1316 | if (tag_get(pathp->node, tag, pathp->offset)) |
1317 | radix_tree_tag_clear(root, index, tag); | |
d5274261 | 1318 | } |
1da177e4 | 1319 | |
7cf9c2c7 | 1320 | to_free = NULL; |
201b6264 | 1321 | /* Now free the nodes we do not need anymore */ |
612d6c19 | 1322 | while (pathp->node) { |
201b6264 | 1323 | pathp->node->slots[pathp->offset] = NULL; |
a5f51c96 | 1324 | pathp->node->count--; |
7cf9c2c7 NP |
1325 | /* |
1326 | * Queue the node for deferred freeing after the | |
1327 | * last reference to it disappears (set NULL, above). | |
1328 | */ | |
1329 | if (to_free) | |
1330 | radix_tree_node_free(to_free); | |
a5f51c96 NP |
1331 | |
1332 | if (pathp->node->count) { | |
27d20fdd | 1333 | if (pathp->node == indirect_to_ptr(root->rnode)) |
a5f51c96 | 1334 | radix_tree_shrink(root); |
201b6264 | 1335 | goto out; |
a5f51c96 | 1336 | } |
201b6264 CL |
1337 | |
1338 | /* Node with zero slots in use so free it */ | |
7cf9c2c7 | 1339 | to_free = pathp->node; |
612d6c19 | 1340 | pathp--; |
7cf9c2c7 | 1341 | |
1da177e4 | 1342 | } |
612d6c19 | 1343 | root_tag_clear_all(root); |
201b6264 | 1344 | root->height = 0; |
612d6c19 | 1345 | root->rnode = NULL; |
7cf9c2c7 NP |
1346 | if (to_free) |
1347 | radix_tree_node_free(to_free); | |
612d6c19 | 1348 | |
1da177e4 | 1349 | out: |
612d6c19 | 1350 | return slot; |
1da177e4 LT |
1351 | } |
1352 | EXPORT_SYMBOL(radix_tree_delete); | |
1353 | ||
1354 | /** | |
1355 | * radix_tree_tagged - test whether any items in the tree are tagged | |
1356 | * @root: radix tree root | |
1357 | * @tag: tag to test | |
1358 | */ | |
daff89f3 | 1359 | int radix_tree_tagged(struct radix_tree_root *root, unsigned int tag) |
1da177e4 | 1360 | { |
612d6c19 | 1361 | return root_tag_get(root, tag); |
1da177e4 LT |
1362 | } |
1363 | EXPORT_SYMBOL(radix_tree_tagged); | |
1364 | ||
1365 | static void | |
51cc5068 | 1366 | radix_tree_node_ctor(void *node) |
1da177e4 LT |
1367 | { |
1368 | memset(node, 0, sizeof(struct radix_tree_node)); | |
1369 | } | |
1370 | ||
1371 | static __init unsigned long __maxindex(unsigned int height) | |
1372 | { | |
430d275a PL |
1373 | unsigned int width = height * RADIX_TREE_MAP_SHIFT; |
1374 | int shift = RADIX_TREE_INDEX_BITS - width; | |
1375 | ||
1376 | if (shift < 0) | |
1377 | return ~0UL; | |
1378 | if (shift >= BITS_PER_LONG) | |
1379 | return 0UL; | |
1380 | return ~0UL >> shift; | |
1da177e4 LT |
1381 | } |
1382 | ||
1383 | static __init void radix_tree_init_maxindex(void) | |
1384 | { | |
1385 | unsigned int i; | |
1386 | ||
1387 | for (i = 0; i < ARRAY_SIZE(height_to_maxindex); i++) | |
1388 | height_to_maxindex[i] = __maxindex(i); | |
1389 | } | |
1390 | ||
1da177e4 LT |
1391 | static int radix_tree_callback(struct notifier_block *nfb, |
1392 | unsigned long action, | |
1393 | void *hcpu) | |
1394 | { | |
1395 | int cpu = (long)hcpu; | |
1396 | struct radix_tree_preload *rtp; | |
1397 | ||
1398 | /* Free per-cpu pool of perloaded nodes */ | |
8bb78442 | 1399 | if (action == CPU_DEAD || action == CPU_DEAD_FROZEN) { |
1da177e4 LT |
1400 | rtp = &per_cpu(radix_tree_preloads, cpu); |
1401 | while (rtp->nr) { | |
1402 | kmem_cache_free(radix_tree_node_cachep, | |
1403 | rtp->nodes[rtp->nr-1]); | |
1404 | rtp->nodes[rtp->nr-1] = NULL; | |
1405 | rtp->nr--; | |
1406 | } | |
1407 | } | |
1408 | return NOTIFY_OK; | |
1409 | } | |
1da177e4 LT |
1410 | |
1411 | void __init radix_tree_init(void) | |
1412 | { | |
1413 | radix_tree_node_cachep = kmem_cache_create("radix_tree_node", | |
1414 | sizeof(struct radix_tree_node), 0, | |
488514d1 CL |
1415 | SLAB_PANIC | SLAB_RECLAIM_ACCOUNT, |
1416 | radix_tree_node_ctor); | |
1da177e4 LT |
1417 | radix_tree_init_maxindex(); |
1418 | hotcpu_notifier(radix_tree_callback, 0); | |
1419 | } |