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> |
5e4c0d97 | 35 | #include <linux/hardirq.h> /* in_interrupt() */ |
1da177e4 LT |
36 | |
37 | ||
26fb1589 JM |
38 | /* |
39 | * The height_to_maxindex array needs to be one deeper than the maximum | |
40 | * path as height 0 holds only 1 entry. | |
41 | */ | |
42 | static unsigned long height_to_maxindex[RADIX_TREE_MAX_PATH + 1] __read_mostly; | |
1da177e4 LT |
43 | |
44 | /* | |
45 | * Radix tree node cache. | |
46 | */ | |
e18b890b | 47 | static struct kmem_cache *radix_tree_node_cachep; |
1da177e4 | 48 | |
55368052 NP |
49 | /* |
50 | * The radix tree is variable-height, so an insert operation not only has | |
51 | * to build the branch to its corresponding item, it also has to build the | |
52 | * branch to existing items if the size has to be increased (by | |
53 | * radix_tree_extend). | |
54 | * | |
55 | * The worst case is a zero height tree with just a single item at index 0, | |
56 | * and then inserting an item at index ULONG_MAX. This requires 2 new branches | |
57 | * of RADIX_TREE_MAX_PATH size to be created, with only the root node shared. | |
58 | * Hence: | |
59 | */ | |
60 | #define RADIX_TREE_PRELOAD_SIZE (RADIX_TREE_MAX_PATH * 2 - 1) | |
61 | ||
1da177e4 LT |
62 | /* |
63 | * Per-cpu pool of preloaded nodes | |
64 | */ | |
65 | struct radix_tree_preload { | |
66 | int nr; | |
55368052 | 67 | struct radix_tree_node *nodes[RADIX_TREE_PRELOAD_SIZE]; |
1da177e4 | 68 | }; |
8cef7d57 | 69 | static DEFINE_PER_CPU(struct radix_tree_preload, radix_tree_preloads) = { 0, }; |
1da177e4 | 70 | |
27d20fdd NP |
71 | static inline void *ptr_to_indirect(void *ptr) |
72 | { | |
73 | return (void *)((unsigned long)ptr | RADIX_TREE_INDIRECT_PTR); | |
74 | } | |
75 | ||
76 | static inline void *indirect_to_ptr(void *ptr) | |
77 | { | |
78 | return (void *)((unsigned long)ptr & ~RADIX_TREE_INDIRECT_PTR); | |
79 | } | |
80 | ||
612d6c19 NP |
81 | static inline gfp_t root_gfp_mask(struct radix_tree_root *root) |
82 | { | |
83 | return root->gfp_mask & __GFP_BITS_MASK; | |
84 | } | |
85 | ||
643b52b9 NP |
86 | static inline void tag_set(struct radix_tree_node *node, unsigned int tag, |
87 | int offset) | |
88 | { | |
89 | __set_bit(offset, node->tags[tag]); | |
90 | } | |
91 | ||
92 | static inline void tag_clear(struct radix_tree_node *node, unsigned int tag, | |
93 | int offset) | |
94 | { | |
95 | __clear_bit(offset, node->tags[tag]); | |
96 | } | |
97 | ||
98 | static inline int tag_get(struct radix_tree_node *node, unsigned int tag, | |
99 | int offset) | |
100 | { | |
101 | return test_bit(offset, node->tags[tag]); | |
102 | } | |
103 | ||
104 | static inline void root_tag_set(struct radix_tree_root *root, unsigned int tag) | |
105 | { | |
106 | root->gfp_mask |= (__force gfp_t)(1 << (tag + __GFP_BITS_SHIFT)); | |
107 | } | |
108 | ||
109 | static inline void root_tag_clear(struct radix_tree_root *root, unsigned int tag) | |
110 | { | |
111 | root->gfp_mask &= (__force gfp_t)~(1 << (tag + __GFP_BITS_SHIFT)); | |
112 | } | |
113 | ||
114 | static inline void root_tag_clear_all(struct radix_tree_root *root) | |
115 | { | |
116 | root->gfp_mask &= __GFP_BITS_MASK; | |
117 | } | |
118 | ||
119 | static inline int root_tag_get(struct radix_tree_root *root, unsigned int tag) | |
120 | { | |
121 | return (__force unsigned)root->gfp_mask & (1 << (tag + __GFP_BITS_SHIFT)); | |
122 | } | |
123 | ||
124 | /* | |
125 | * Returns 1 if any slot in the node has this tag set. | |
126 | * Otherwise returns 0. | |
127 | */ | |
128 | static inline int any_tag_set(struct radix_tree_node *node, unsigned int tag) | |
129 | { | |
130 | int idx; | |
131 | for (idx = 0; idx < RADIX_TREE_TAG_LONGS; idx++) { | |
132 | if (node->tags[tag][idx]) | |
133 | return 1; | |
134 | } | |
135 | return 0; | |
136 | } | |
78c1d784 KK |
137 | |
138 | /** | |
139 | * radix_tree_find_next_bit - find the next set bit in a memory region | |
140 | * | |
141 | * @addr: The address to base the search on | |
142 | * @size: The bitmap size in bits | |
143 | * @offset: The bitnumber to start searching at | |
144 | * | |
145 | * Unrollable variant of find_next_bit() for constant size arrays. | |
146 | * Tail bits starting from size to roundup(size, BITS_PER_LONG) must be zero. | |
147 | * Returns next bit offset, or size if nothing found. | |
148 | */ | |
149 | static __always_inline unsigned long | |
150 | radix_tree_find_next_bit(const unsigned long *addr, | |
151 | unsigned long size, unsigned long offset) | |
152 | { | |
153 | if (!__builtin_constant_p(size)) | |
154 | return find_next_bit(addr, size, offset); | |
155 | ||
156 | if (offset < size) { | |
157 | unsigned long tmp; | |
158 | ||
159 | addr += offset / BITS_PER_LONG; | |
160 | tmp = *addr >> (offset % BITS_PER_LONG); | |
161 | if (tmp) | |
162 | return __ffs(tmp) + offset; | |
163 | offset = (offset + BITS_PER_LONG) & ~(BITS_PER_LONG - 1); | |
164 | while (offset < size) { | |
165 | tmp = *++addr; | |
166 | if (tmp) | |
167 | return __ffs(tmp) + offset; | |
168 | offset += BITS_PER_LONG; | |
169 | } | |
170 | } | |
171 | return size; | |
172 | } | |
173 | ||
1da177e4 LT |
174 | /* |
175 | * This assumes that the caller has performed appropriate preallocation, and | |
176 | * that the caller has pinned this thread of control to the current CPU. | |
177 | */ | |
178 | static struct radix_tree_node * | |
179 | radix_tree_node_alloc(struct radix_tree_root *root) | |
180 | { | |
e2848a0e | 181 | struct radix_tree_node *ret = NULL; |
612d6c19 | 182 | gfp_t gfp_mask = root_gfp_mask(root); |
1da177e4 | 183 | |
5e4c0d97 JK |
184 | /* |
185 | * Preload code isn't irq safe and it doesn't make sence to use | |
186 | * preloading in the interrupt anyway as all the allocations have to | |
187 | * be atomic. So just do normal allocation when in interrupt. | |
188 | */ | |
189 | if (!(gfp_mask & __GFP_WAIT) && !in_interrupt()) { | |
1da177e4 LT |
190 | struct radix_tree_preload *rtp; |
191 | ||
e2848a0e NP |
192 | /* |
193 | * Provided the caller has preloaded here, we will always | |
194 | * succeed in getting a node here (and never reach | |
195 | * kmem_cache_alloc) | |
196 | */ | |
1da177e4 LT |
197 | rtp = &__get_cpu_var(radix_tree_preloads); |
198 | if (rtp->nr) { | |
199 | ret = rtp->nodes[rtp->nr - 1]; | |
200 | rtp->nodes[rtp->nr - 1] = NULL; | |
201 | rtp->nr--; | |
202 | } | |
203 | } | |
e2848a0e | 204 | if (ret == NULL) |
488514d1 | 205 | ret = kmem_cache_alloc(radix_tree_node_cachep, gfp_mask); |
e2848a0e | 206 | |
c0bc9875 | 207 | BUG_ON(radix_tree_is_indirect_ptr(ret)); |
1da177e4 LT |
208 | return ret; |
209 | } | |
210 | ||
7cf9c2c7 NP |
211 | static void radix_tree_node_rcu_free(struct rcu_head *head) |
212 | { | |
213 | struct radix_tree_node *node = | |
214 | container_of(head, struct radix_tree_node, rcu_head); | |
b6dd0865 | 215 | int i; |
643b52b9 NP |
216 | |
217 | /* | |
218 | * must only free zeroed nodes into the slab. radix_tree_shrink | |
219 | * can leave us with a non-NULL entry in the first slot, so clear | |
220 | * that here to make sure. | |
221 | */ | |
b6dd0865 DC |
222 | for (i = 0; i < RADIX_TREE_MAX_TAGS; i++) |
223 | tag_clear(node, i, 0); | |
224 | ||
643b52b9 NP |
225 | node->slots[0] = NULL; |
226 | node->count = 0; | |
227 | ||
7cf9c2c7 NP |
228 | kmem_cache_free(radix_tree_node_cachep, node); |
229 | } | |
230 | ||
1da177e4 LT |
231 | static inline void |
232 | radix_tree_node_free(struct radix_tree_node *node) | |
233 | { | |
7cf9c2c7 | 234 | call_rcu(&node->rcu_head, radix_tree_node_rcu_free); |
1da177e4 LT |
235 | } |
236 | ||
237 | /* | |
238 | * Load up this CPU's radix_tree_node buffer with sufficient objects to | |
239 | * ensure that the addition of a single element in the tree cannot fail. On | |
240 | * success, return zero, with preemption disabled. On error, return -ENOMEM | |
241 | * with preemption not disabled. | |
b34df792 DH |
242 | * |
243 | * To make use of this facility, the radix tree must be initialised without | |
244 | * __GFP_WAIT being passed to INIT_RADIX_TREE(). | |
1da177e4 | 245 | */ |
5e4c0d97 | 246 | static int __radix_tree_preload(gfp_t gfp_mask) |
1da177e4 LT |
247 | { |
248 | struct radix_tree_preload *rtp; | |
249 | struct radix_tree_node *node; | |
250 | int ret = -ENOMEM; | |
251 | ||
252 | preempt_disable(); | |
253 | rtp = &__get_cpu_var(radix_tree_preloads); | |
254 | while (rtp->nr < ARRAY_SIZE(rtp->nodes)) { | |
255 | preempt_enable(); | |
488514d1 | 256 | node = kmem_cache_alloc(radix_tree_node_cachep, gfp_mask); |
1da177e4 LT |
257 | if (node == NULL) |
258 | goto out; | |
259 | preempt_disable(); | |
260 | rtp = &__get_cpu_var(radix_tree_preloads); | |
261 | if (rtp->nr < ARRAY_SIZE(rtp->nodes)) | |
262 | rtp->nodes[rtp->nr++] = node; | |
263 | else | |
264 | kmem_cache_free(radix_tree_node_cachep, node); | |
265 | } | |
266 | ret = 0; | |
267 | out: | |
268 | return ret; | |
269 | } | |
5e4c0d97 JK |
270 | |
271 | /* | |
272 | * Load up this CPU's radix_tree_node buffer with sufficient objects to | |
273 | * ensure that the addition of a single element in the tree cannot fail. On | |
274 | * success, return zero, with preemption disabled. On error, return -ENOMEM | |
275 | * with preemption not disabled. | |
276 | * | |
277 | * To make use of this facility, the radix tree must be initialised without | |
278 | * __GFP_WAIT being passed to INIT_RADIX_TREE(). | |
279 | */ | |
280 | int radix_tree_preload(gfp_t gfp_mask) | |
281 | { | |
282 | /* Warn on non-sensical use... */ | |
283 | WARN_ON_ONCE(!(gfp_mask & __GFP_WAIT)); | |
284 | return __radix_tree_preload(gfp_mask); | |
285 | } | |
d7f0923d | 286 | EXPORT_SYMBOL(radix_tree_preload); |
1da177e4 | 287 | |
5e4c0d97 JK |
288 | /* |
289 | * The same as above function, except we don't guarantee preloading happens. | |
290 | * We do it, if we decide it helps. On success, return zero with preemption | |
291 | * disabled. On error, return -ENOMEM with preemption not disabled. | |
292 | */ | |
293 | int radix_tree_maybe_preload(gfp_t gfp_mask) | |
294 | { | |
295 | if (gfp_mask & __GFP_WAIT) | |
296 | return __radix_tree_preload(gfp_mask); | |
297 | /* Preloading doesn't help anything with this gfp mask, skip it */ | |
298 | preempt_disable(); | |
299 | return 0; | |
300 | } | |
301 | EXPORT_SYMBOL(radix_tree_maybe_preload); | |
302 | ||
1da177e4 LT |
303 | /* |
304 | * Return the maximum key which can be store into a | |
305 | * radix tree with height HEIGHT. | |
306 | */ | |
307 | static inline unsigned long radix_tree_maxindex(unsigned int height) | |
308 | { | |
309 | return height_to_maxindex[height]; | |
310 | } | |
311 | ||
312 | /* | |
313 | * Extend a radix tree so it can store key @index. | |
314 | */ | |
315 | static int radix_tree_extend(struct radix_tree_root *root, unsigned long index) | |
316 | { | |
317 | struct radix_tree_node *node; | |
e2bdb933 | 318 | struct radix_tree_node *slot; |
1da177e4 | 319 | unsigned int height; |
1da177e4 LT |
320 | int tag; |
321 | ||
322 | /* Figure out what the height should be. */ | |
323 | height = root->height + 1; | |
324 | while (index > radix_tree_maxindex(height)) | |
325 | height++; | |
326 | ||
327 | if (root->rnode == NULL) { | |
328 | root->height = height; | |
329 | goto out; | |
330 | } | |
331 | ||
1da177e4 | 332 | do { |
7cf9c2c7 | 333 | unsigned int newheight; |
1da177e4 LT |
334 | if (!(node = radix_tree_node_alloc(root))) |
335 | return -ENOMEM; | |
336 | ||
1da177e4 | 337 | /* Propagate the aggregated tag info into the new root */ |
daff89f3 | 338 | for (tag = 0; tag < RADIX_TREE_MAX_TAGS; tag++) { |
612d6c19 | 339 | if (root_tag_get(root, tag)) |
1da177e4 LT |
340 | tag_set(node, tag, 0); |
341 | } | |
342 | ||
e2bdb933 | 343 | /* Increase the height. */ |
7cf9c2c7 | 344 | newheight = root->height+1; |
449dd698 JW |
345 | BUG_ON(newheight & ~RADIX_TREE_HEIGHT_MASK); |
346 | node->path = newheight; | |
1da177e4 | 347 | node->count = 1; |
e2bdb933 HD |
348 | node->parent = NULL; |
349 | slot = root->rnode; | |
350 | if (newheight > 1) { | |
351 | slot = indirect_to_ptr(slot); | |
352 | slot->parent = node; | |
353 | } | |
354 | node->slots[0] = slot; | |
27d20fdd | 355 | node = ptr_to_indirect(node); |
7cf9c2c7 NP |
356 | rcu_assign_pointer(root->rnode, node); |
357 | root->height = newheight; | |
1da177e4 LT |
358 | } while (height > root->height); |
359 | out: | |
360 | return 0; | |
361 | } | |
362 | ||
363 | /** | |
139e5616 | 364 | * __radix_tree_create - create a slot in a radix tree |
1da177e4 LT |
365 | * @root: radix tree root |
366 | * @index: index key | |
139e5616 JW |
367 | * @nodep: returns node |
368 | * @slotp: returns slot | |
1da177e4 | 369 | * |
139e5616 JW |
370 | * Create, if necessary, and return the node and slot for an item |
371 | * at position @index in the radix tree @root. | |
372 | * | |
373 | * Until there is more than one item in the tree, no nodes are | |
374 | * allocated and @root->rnode is used as a direct slot instead of | |
375 | * pointing to a node, in which case *@nodep will be NULL. | |
376 | * | |
377 | * Returns -ENOMEM, or 0 for success. | |
1da177e4 | 378 | */ |
139e5616 JW |
379 | int __radix_tree_create(struct radix_tree_root *root, unsigned long index, |
380 | struct radix_tree_node **nodep, void ***slotp) | |
1da177e4 | 381 | { |
201b6264 | 382 | struct radix_tree_node *node = NULL, *slot; |
139e5616 | 383 | unsigned int height, shift, offset; |
1da177e4 LT |
384 | int error; |
385 | ||
386 | /* Make sure the tree is high enough. */ | |
612d6c19 | 387 | if (index > radix_tree_maxindex(root->height)) { |
1da177e4 LT |
388 | error = radix_tree_extend(root, index); |
389 | if (error) | |
390 | return error; | |
391 | } | |
392 | ||
27d20fdd | 393 | slot = indirect_to_ptr(root->rnode); |
c0bc9875 | 394 | |
1da177e4 LT |
395 | height = root->height; |
396 | shift = (height-1) * RADIX_TREE_MAP_SHIFT; | |
397 | ||
398 | offset = 0; /* uninitialised var warning */ | |
612d6c19 | 399 | while (height > 0) { |
201b6264 | 400 | if (slot == NULL) { |
1da177e4 | 401 | /* Have to add a child node. */ |
201b6264 | 402 | if (!(slot = radix_tree_node_alloc(root))) |
1da177e4 | 403 | return -ENOMEM; |
449dd698 | 404 | slot->path = height; |
e2bdb933 | 405 | slot->parent = node; |
201b6264 | 406 | if (node) { |
7cf9c2c7 | 407 | rcu_assign_pointer(node->slots[offset], slot); |
1da177e4 | 408 | node->count++; |
449dd698 | 409 | slot->path |= offset << RADIX_TREE_HEIGHT_SHIFT; |
201b6264 | 410 | } else |
27d20fdd | 411 | rcu_assign_pointer(root->rnode, ptr_to_indirect(slot)); |
1da177e4 LT |
412 | } |
413 | ||
414 | /* Go a level down */ | |
415 | offset = (index >> shift) & RADIX_TREE_MAP_MASK; | |
201b6264 CL |
416 | node = slot; |
417 | slot = node->slots[offset]; | |
1da177e4 LT |
418 | shift -= RADIX_TREE_MAP_SHIFT; |
419 | height--; | |
612d6c19 | 420 | } |
1da177e4 | 421 | |
139e5616 JW |
422 | if (nodep) |
423 | *nodep = node; | |
424 | if (slotp) | |
425 | *slotp = node ? node->slots + offset : (void **)&root->rnode; | |
426 | return 0; | |
427 | } | |
428 | ||
429 | /** | |
430 | * radix_tree_insert - insert into a radix tree | |
431 | * @root: radix tree root | |
432 | * @index: index key | |
433 | * @item: item to insert | |
434 | * | |
435 | * Insert an item into the radix tree at position @index. | |
436 | */ | |
437 | int radix_tree_insert(struct radix_tree_root *root, | |
438 | unsigned long index, void *item) | |
439 | { | |
440 | struct radix_tree_node *node; | |
441 | void **slot; | |
442 | int error; | |
443 | ||
444 | BUG_ON(radix_tree_is_indirect_ptr(item)); | |
445 | ||
446 | error = __radix_tree_create(root, index, &node, &slot); | |
447 | if (error) | |
448 | return error; | |
449 | if (*slot != NULL) | |
1da177e4 | 450 | return -EEXIST; |
139e5616 | 451 | rcu_assign_pointer(*slot, item); |
201b6264 | 452 | |
612d6c19 NP |
453 | if (node) { |
454 | node->count++; | |
139e5616 JW |
455 | BUG_ON(tag_get(node, 0, index & RADIX_TREE_MAP_MASK)); |
456 | BUG_ON(tag_get(node, 1, index & RADIX_TREE_MAP_MASK)); | |
612d6c19 | 457 | } else { |
612d6c19 NP |
458 | BUG_ON(root_tag_get(root, 0)); |
459 | BUG_ON(root_tag_get(root, 1)); | |
460 | } | |
1da177e4 | 461 | |
1da177e4 LT |
462 | return 0; |
463 | } | |
464 | EXPORT_SYMBOL(radix_tree_insert); | |
465 | ||
139e5616 JW |
466 | /** |
467 | * __radix_tree_lookup - lookup an item in a radix tree | |
468 | * @root: radix tree root | |
469 | * @index: index key | |
470 | * @nodep: returns node | |
471 | * @slotp: returns slot | |
472 | * | |
473 | * Lookup and return the item at position @index in the radix | |
474 | * tree @root. | |
475 | * | |
476 | * Until there is more than one item in the tree, no nodes are | |
477 | * allocated and @root->rnode is used as a direct slot instead of | |
478 | * pointing to a node, in which case *@nodep will be NULL. | |
7cf9c2c7 | 479 | */ |
139e5616 JW |
480 | void *__radix_tree_lookup(struct radix_tree_root *root, unsigned long index, |
481 | struct radix_tree_node **nodep, void ***slotp) | |
1da177e4 | 482 | { |
139e5616 | 483 | struct radix_tree_node *node, *parent; |
1da177e4 | 484 | unsigned int height, shift; |
139e5616 | 485 | void **slot; |
612d6c19 | 486 | |
2676a58c | 487 | node = rcu_dereference_raw(root->rnode); |
7cf9c2c7 | 488 | if (node == NULL) |
1da177e4 LT |
489 | return NULL; |
490 | ||
c0bc9875 | 491 | if (!radix_tree_is_indirect_ptr(node)) { |
7cf9c2c7 NP |
492 | if (index > 0) |
493 | return NULL; | |
139e5616 JW |
494 | |
495 | if (nodep) | |
496 | *nodep = NULL; | |
497 | if (slotp) | |
498 | *slotp = (void **)&root->rnode; | |
499 | return node; | |
7cf9c2c7 | 500 | } |
27d20fdd | 501 | node = indirect_to_ptr(node); |
7cf9c2c7 | 502 | |
449dd698 | 503 | height = node->path & RADIX_TREE_HEIGHT_MASK; |
7cf9c2c7 NP |
504 | if (index > radix_tree_maxindex(height)) |
505 | return NULL; | |
612d6c19 | 506 | |
1da177e4 | 507 | shift = (height-1) * RADIX_TREE_MAP_SHIFT; |
1da177e4 | 508 | |
7cf9c2c7 | 509 | do { |
139e5616 JW |
510 | parent = node; |
511 | slot = node->slots + ((index >> shift) & RADIX_TREE_MAP_MASK); | |
2676a58c | 512 | node = rcu_dereference_raw(*slot); |
7cf9c2c7 | 513 | if (node == NULL) |
1da177e4 LT |
514 | return NULL; |
515 | ||
1da177e4 LT |
516 | shift -= RADIX_TREE_MAP_SHIFT; |
517 | height--; | |
7cf9c2c7 | 518 | } while (height > 0); |
1da177e4 | 519 | |
139e5616 JW |
520 | if (nodep) |
521 | *nodep = parent; | |
522 | if (slotp) | |
523 | *slotp = slot; | |
524 | return node; | |
b72b71c6 HS |
525 | } |
526 | ||
527 | /** | |
528 | * radix_tree_lookup_slot - lookup a slot in a radix tree | |
529 | * @root: radix tree root | |
530 | * @index: index key | |
531 | * | |
532 | * Returns: the slot corresponding to the position @index in the | |
533 | * radix tree @root. This is useful for update-if-exists operations. | |
534 | * | |
535 | * This function can be called under rcu_read_lock iff the slot is not | |
536 | * modified by radix_tree_replace_slot, otherwise it must be called | |
537 | * exclusive from other writers. Any dereference of the slot must be done | |
538 | * using radix_tree_deref_slot. | |
539 | */ | |
540 | void **radix_tree_lookup_slot(struct radix_tree_root *root, unsigned long index) | |
541 | { | |
139e5616 JW |
542 | void **slot; |
543 | ||
544 | if (!__radix_tree_lookup(root, index, NULL, &slot)) | |
545 | return NULL; | |
546 | return slot; | |
a4331366 | 547 | } |
a4331366 HR |
548 | EXPORT_SYMBOL(radix_tree_lookup_slot); |
549 | ||
550 | /** | |
551 | * radix_tree_lookup - perform lookup operation on a radix tree | |
552 | * @root: radix tree root | |
553 | * @index: index key | |
554 | * | |
555 | * Lookup the item at the position @index in the radix tree @root. | |
7cf9c2c7 NP |
556 | * |
557 | * This function can be called under rcu_read_lock, however the caller | |
558 | * must manage lifetimes of leaf nodes (eg. RCU may also be used to free | |
559 | * them safely). No RCU barriers are required to access or modify the | |
560 | * returned item, however. | |
a4331366 HR |
561 | */ |
562 | void *radix_tree_lookup(struct radix_tree_root *root, unsigned long index) | |
563 | { | |
139e5616 | 564 | return __radix_tree_lookup(root, index, NULL, NULL); |
1da177e4 LT |
565 | } |
566 | EXPORT_SYMBOL(radix_tree_lookup); | |
567 | ||
568 | /** | |
569 | * radix_tree_tag_set - set a tag on a radix tree node | |
570 | * @root: radix tree root | |
571 | * @index: index key | |
572 | * @tag: tag index | |
573 | * | |
daff89f3 JC |
574 | * Set the search tag (which must be < RADIX_TREE_MAX_TAGS) |
575 | * corresponding to @index in the radix tree. From | |
1da177e4 LT |
576 | * the root all the way down to the leaf node. |
577 | * | |
578 | * Returns the address of the tagged item. Setting a tag on a not-present | |
579 | * item is a bug. | |
580 | */ | |
581 | void *radix_tree_tag_set(struct radix_tree_root *root, | |
daff89f3 | 582 | unsigned long index, unsigned int tag) |
1da177e4 LT |
583 | { |
584 | unsigned int height, shift; | |
201b6264 | 585 | struct radix_tree_node *slot; |
1da177e4 LT |
586 | |
587 | height = root->height; | |
4c91c364 | 588 | BUG_ON(index > radix_tree_maxindex(height)); |
1da177e4 | 589 | |
27d20fdd | 590 | slot = indirect_to_ptr(root->rnode); |
612d6c19 | 591 | shift = (height - 1) * RADIX_TREE_MAP_SHIFT; |
1da177e4 LT |
592 | |
593 | while (height > 0) { | |
594 | int offset; | |
595 | ||
596 | offset = (index >> shift) & RADIX_TREE_MAP_MASK; | |
d5274261 NP |
597 | if (!tag_get(slot, tag, offset)) |
598 | tag_set(slot, tag, offset); | |
201b6264 CL |
599 | slot = slot->slots[offset]; |
600 | BUG_ON(slot == NULL); | |
1da177e4 LT |
601 | shift -= RADIX_TREE_MAP_SHIFT; |
602 | height--; | |
603 | } | |
604 | ||
612d6c19 NP |
605 | /* set the root's tag bit */ |
606 | if (slot && !root_tag_get(root, tag)) | |
607 | root_tag_set(root, tag); | |
608 | ||
201b6264 | 609 | return slot; |
1da177e4 LT |
610 | } |
611 | EXPORT_SYMBOL(radix_tree_tag_set); | |
612 | ||
613 | /** | |
614 | * radix_tree_tag_clear - clear a tag on a radix tree node | |
615 | * @root: radix tree root | |
616 | * @index: index key | |
617 | * @tag: tag index | |
618 | * | |
daff89f3 JC |
619 | * Clear the search tag (which must be < RADIX_TREE_MAX_TAGS) |
620 | * corresponding to @index in the radix tree. If | |
1da177e4 LT |
621 | * this causes the leaf node to have no tags set then clear the tag in the |
622 | * next-to-leaf node, etc. | |
623 | * | |
624 | * Returns the address of the tagged item on success, else NULL. ie: | |
625 | * has the same return value and semantics as radix_tree_lookup(). | |
626 | */ | |
627 | void *radix_tree_tag_clear(struct radix_tree_root *root, | |
daff89f3 | 628 | unsigned long index, unsigned int tag) |
1da177e4 | 629 | { |
e2bdb933 | 630 | struct radix_tree_node *node = NULL; |
612d6c19 | 631 | struct radix_tree_node *slot = NULL; |
1da177e4 | 632 | unsigned int height, shift; |
e2bdb933 | 633 | int uninitialized_var(offset); |
1da177e4 LT |
634 | |
635 | height = root->height; | |
636 | if (index > radix_tree_maxindex(height)) | |
637 | goto out; | |
638 | ||
e2bdb933 | 639 | shift = height * RADIX_TREE_MAP_SHIFT; |
27d20fdd | 640 | slot = indirect_to_ptr(root->rnode); |
1da177e4 | 641 | |
e2bdb933 | 642 | while (shift) { |
201b6264 | 643 | if (slot == NULL) |
1da177e4 LT |
644 | goto out; |
645 | ||
e2bdb933 | 646 | shift -= RADIX_TREE_MAP_SHIFT; |
1da177e4 | 647 | offset = (index >> shift) & RADIX_TREE_MAP_MASK; |
e2bdb933 | 648 | node = slot; |
201b6264 | 649 | slot = slot->slots[offset]; |
1da177e4 LT |
650 | } |
651 | ||
612d6c19 | 652 | if (slot == NULL) |
1da177e4 LT |
653 | goto out; |
654 | ||
e2bdb933 HD |
655 | while (node) { |
656 | if (!tag_get(node, tag, offset)) | |
d5274261 | 657 | goto out; |
e2bdb933 HD |
658 | tag_clear(node, tag, offset); |
659 | if (any_tag_set(node, tag)) | |
6e954b9e | 660 | goto out; |
e2bdb933 HD |
661 | |
662 | index >>= RADIX_TREE_MAP_SHIFT; | |
663 | offset = index & RADIX_TREE_MAP_MASK; | |
664 | node = node->parent; | |
612d6c19 NP |
665 | } |
666 | ||
667 | /* clear the root's tag bit */ | |
668 | if (root_tag_get(root, tag)) | |
669 | root_tag_clear(root, tag); | |
670 | ||
1da177e4 | 671 | out: |
612d6c19 | 672 | return slot; |
1da177e4 LT |
673 | } |
674 | EXPORT_SYMBOL(radix_tree_tag_clear); | |
675 | ||
1da177e4 | 676 | /** |
32605a18 MT |
677 | * radix_tree_tag_get - get a tag on a radix tree node |
678 | * @root: radix tree root | |
679 | * @index: index key | |
daff89f3 | 680 | * @tag: tag index (< RADIX_TREE_MAX_TAGS) |
1da177e4 | 681 | * |
32605a18 | 682 | * Return values: |
1da177e4 | 683 | * |
612d6c19 NP |
684 | * 0: tag not present or not set |
685 | * 1: tag set | |
ce82653d DH |
686 | * |
687 | * Note that the return value of this function may not be relied on, even if | |
688 | * the RCU lock is held, unless tag modification and node deletion are excluded | |
689 | * from concurrency. | |
1da177e4 LT |
690 | */ |
691 | int radix_tree_tag_get(struct radix_tree_root *root, | |
daff89f3 | 692 | unsigned long index, unsigned int tag) |
1da177e4 LT |
693 | { |
694 | unsigned int height, shift; | |
7cf9c2c7 | 695 | struct radix_tree_node *node; |
1da177e4 | 696 | |
612d6c19 NP |
697 | /* check the root's tag bit */ |
698 | if (!root_tag_get(root, tag)) | |
699 | return 0; | |
700 | ||
2676a58c | 701 | node = rcu_dereference_raw(root->rnode); |
7cf9c2c7 NP |
702 | if (node == NULL) |
703 | return 0; | |
704 | ||
c0bc9875 | 705 | if (!radix_tree_is_indirect_ptr(node)) |
7cf9c2c7 | 706 | return (index == 0); |
27d20fdd | 707 | node = indirect_to_ptr(node); |
7cf9c2c7 | 708 | |
449dd698 | 709 | height = node->path & RADIX_TREE_HEIGHT_MASK; |
7cf9c2c7 NP |
710 | if (index > radix_tree_maxindex(height)) |
711 | return 0; | |
612d6c19 | 712 | |
1da177e4 | 713 | shift = (height - 1) * RADIX_TREE_MAP_SHIFT; |
1da177e4 LT |
714 | |
715 | for ( ; ; ) { | |
716 | int offset; | |
717 | ||
7cf9c2c7 | 718 | if (node == NULL) |
1da177e4 LT |
719 | return 0; |
720 | ||
721 | offset = (index >> shift) & RADIX_TREE_MAP_MASK; | |
7cf9c2c7 | 722 | if (!tag_get(node, tag, offset)) |
3fa36acb | 723 | return 0; |
ce82653d | 724 | if (height == 1) |
3fa36acb | 725 | return 1; |
2676a58c | 726 | node = rcu_dereference_raw(node->slots[offset]); |
1da177e4 LT |
727 | shift -= RADIX_TREE_MAP_SHIFT; |
728 | height--; | |
729 | } | |
730 | } | |
731 | EXPORT_SYMBOL(radix_tree_tag_get); | |
1da177e4 | 732 | |
78c1d784 KK |
733 | /** |
734 | * radix_tree_next_chunk - find next chunk of slots for iteration | |
735 | * | |
736 | * @root: radix tree root | |
737 | * @iter: iterator state | |
738 | * @flags: RADIX_TREE_ITER_* flags and tag index | |
739 | * Returns: pointer to chunk first slot, or NULL if iteration is over | |
740 | */ | |
741 | void **radix_tree_next_chunk(struct radix_tree_root *root, | |
742 | struct radix_tree_iter *iter, unsigned flags) | |
743 | { | |
744 | unsigned shift, tag = flags & RADIX_TREE_ITER_TAG_MASK; | |
745 | struct radix_tree_node *rnode, *node; | |
449dd698 | 746 | unsigned long index, offset, height; |
78c1d784 KK |
747 | |
748 | if ((flags & RADIX_TREE_ITER_TAGGED) && !root_tag_get(root, tag)) | |
749 | return NULL; | |
750 | ||
751 | /* | |
752 | * Catch next_index overflow after ~0UL. iter->index never overflows | |
753 | * during iterating; it can be zero only at the beginning. | |
754 | * And we cannot overflow iter->next_index in a single step, | |
755 | * because RADIX_TREE_MAP_SHIFT < BITS_PER_LONG. | |
fffaee36 KK |
756 | * |
757 | * This condition also used by radix_tree_next_slot() to stop | |
758 | * contiguous iterating, and forbid swithing to the next chunk. | |
78c1d784 KK |
759 | */ |
760 | index = iter->next_index; | |
761 | if (!index && iter->index) | |
762 | return NULL; | |
763 | ||
764 | rnode = rcu_dereference_raw(root->rnode); | |
765 | if (radix_tree_is_indirect_ptr(rnode)) { | |
766 | rnode = indirect_to_ptr(rnode); | |
767 | } else if (rnode && !index) { | |
768 | /* Single-slot tree */ | |
769 | iter->index = 0; | |
770 | iter->next_index = 1; | |
771 | iter->tags = 1; | |
772 | return (void **)&root->rnode; | |
773 | } else | |
774 | return NULL; | |
775 | ||
776 | restart: | |
449dd698 JW |
777 | height = rnode->path & RADIX_TREE_HEIGHT_MASK; |
778 | shift = (height - 1) * RADIX_TREE_MAP_SHIFT; | |
78c1d784 KK |
779 | offset = index >> shift; |
780 | ||
781 | /* Index outside of the tree */ | |
782 | if (offset >= RADIX_TREE_MAP_SIZE) | |
783 | return NULL; | |
784 | ||
785 | node = rnode; | |
786 | while (1) { | |
787 | if ((flags & RADIX_TREE_ITER_TAGGED) ? | |
788 | !test_bit(offset, node->tags[tag]) : | |
789 | !node->slots[offset]) { | |
790 | /* Hole detected */ | |
791 | if (flags & RADIX_TREE_ITER_CONTIG) | |
792 | return NULL; | |
793 | ||
794 | if (flags & RADIX_TREE_ITER_TAGGED) | |
795 | offset = radix_tree_find_next_bit( | |
796 | node->tags[tag], | |
797 | RADIX_TREE_MAP_SIZE, | |
798 | offset + 1); | |
799 | else | |
800 | while (++offset < RADIX_TREE_MAP_SIZE) { | |
801 | if (node->slots[offset]) | |
802 | break; | |
803 | } | |
804 | index &= ~((RADIX_TREE_MAP_SIZE << shift) - 1); | |
805 | index += offset << shift; | |
806 | /* Overflow after ~0UL */ | |
807 | if (!index) | |
808 | return NULL; | |
809 | if (offset == RADIX_TREE_MAP_SIZE) | |
810 | goto restart; | |
811 | } | |
812 | ||
813 | /* This is leaf-node */ | |
814 | if (!shift) | |
815 | break; | |
816 | ||
817 | node = rcu_dereference_raw(node->slots[offset]); | |
818 | if (node == NULL) | |
819 | goto restart; | |
820 | shift -= RADIX_TREE_MAP_SHIFT; | |
821 | offset = (index >> shift) & RADIX_TREE_MAP_MASK; | |
822 | } | |
823 | ||
824 | /* Update the iterator state */ | |
825 | iter->index = index; | |
826 | iter->next_index = (index | RADIX_TREE_MAP_MASK) + 1; | |
827 | ||
828 | /* Construct iter->tags bit-mask from node->tags[tag] array */ | |
829 | if (flags & RADIX_TREE_ITER_TAGGED) { | |
830 | unsigned tag_long, tag_bit; | |
831 | ||
832 | tag_long = offset / BITS_PER_LONG; | |
833 | tag_bit = offset % BITS_PER_LONG; | |
834 | iter->tags = node->tags[tag][tag_long] >> tag_bit; | |
835 | /* This never happens if RADIX_TREE_TAG_LONGS == 1 */ | |
836 | if (tag_long < RADIX_TREE_TAG_LONGS - 1) { | |
837 | /* Pick tags from next element */ | |
838 | if (tag_bit) | |
839 | iter->tags |= node->tags[tag][tag_long + 1] << | |
840 | (BITS_PER_LONG - tag_bit); | |
841 | /* Clip chunk size, here only BITS_PER_LONG tags */ | |
842 | iter->next_index = index + BITS_PER_LONG; | |
843 | } | |
844 | } | |
845 | ||
846 | return node->slots + offset; | |
847 | } | |
848 | EXPORT_SYMBOL(radix_tree_next_chunk); | |
849 | ||
ebf8aa44 JK |
850 | /** |
851 | * radix_tree_range_tag_if_tagged - for each item in given range set given | |
852 | * tag if item has another tag set | |
853 | * @root: radix tree root | |
854 | * @first_indexp: pointer to a starting index of a range to scan | |
855 | * @last_index: last index of a range to scan | |
856 | * @nr_to_tag: maximum number items to tag | |
857 | * @iftag: tag index to test | |
858 | * @settag: tag index to set if tested tag is set | |
859 | * | |
860 | * This function scans range of radix tree from first_index to last_index | |
861 | * (inclusive). For each item in the range if iftag is set, the function sets | |
862 | * also settag. The function stops either after tagging nr_to_tag items or | |
863 | * after reaching last_index. | |
864 | * | |
144dcfc0 DC |
865 | * The tags must be set from the leaf level only and propagated back up the |
866 | * path to the root. We must do this so that we resolve the full path before | |
867 | * setting any tags on intermediate nodes. If we set tags as we descend, then | |
868 | * we can get to the leaf node and find that the index that has the iftag | |
869 | * set is outside the range we are scanning. This reults in dangling tags and | |
870 | * can lead to problems with later tag operations (e.g. livelocks on lookups). | |
871 | * | |
ebf8aa44 JK |
872 | * The function returns number of leaves where the tag was set and sets |
873 | * *first_indexp to the first unscanned index. | |
d5ed3a4a JK |
874 | * WARNING! *first_indexp can wrap if last_index is ULONG_MAX. Caller must |
875 | * be prepared to handle that. | |
ebf8aa44 JK |
876 | */ |
877 | unsigned long radix_tree_range_tag_if_tagged(struct radix_tree_root *root, | |
878 | unsigned long *first_indexp, unsigned long last_index, | |
879 | unsigned long nr_to_tag, | |
880 | unsigned int iftag, unsigned int settag) | |
881 | { | |
144dcfc0 | 882 | unsigned int height = root->height; |
e2bdb933 | 883 | struct radix_tree_node *node = NULL; |
144dcfc0 DC |
884 | struct radix_tree_node *slot; |
885 | unsigned int shift; | |
886 | unsigned long tagged = 0; | |
887 | unsigned long index = *first_indexp; | |
ebf8aa44 JK |
888 | |
889 | last_index = min(last_index, radix_tree_maxindex(height)); | |
890 | if (index > last_index) | |
891 | return 0; | |
892 | if (!nr_to_tag) | |
893 | return 0; | |
894 | if (!root_tag_get(root, iftag)) { | |
895 | *first_indexp = last_index + 1; | |
896 | return 0; | |
897 | } | |
898 | if (height == 0) { | |
899 | *first_indexp = last_index + 1; | |
900 | root_tag_set(root, settag); | |
901 | return 1; | |
902 | } | |
903 | ||
904 | shift = (height - 1) * RADIX_TREE_MAP_SHIFT; | |
27d20fdd | 905 | slot = indirect_to_ptr(root->rnode); |
ebf8aa44 JK |
906 | |
907 | for (;;) { | |
e2bdb933 | 908 | unsigned long upindex; |
ebf8aa44 JK |
909 | int offset; |
910 | ||
911 | offset = (index >> shift) & RADIX_TREE_MAP_MASK; | |
912 | if (!slot->slots[offset]) | |
913 | goto next; | |
914 | if (!tag_get(slot, iftag, offset)) | |
915 | goto next; | |
e2bdb933 | 916 | if (shift) { |
144dcfc0 | 917 | /* Go down one level */ |
144dcfc0 | 918 | shift -= RADIX_TREE_MAP_SHIFT; |
e2bdb933 | 919 | node = slot; |
144dcfc0 DC |
920 | slot = slot->slots[offset]; |
921 | continue; | |
922 | } | |
923 | ||
924 | /* tag the leaf */ | |
925 | tagged++; | |
ebf8aa44 | 926 | tag_set(slot, settag, offset); |
144dcfc0 DC |
927 | |
928 | /* walk back up the path tagging interior nodes */ | |
e2bdb933 HD |
929 | upindex = index; |
930 | while (node) { | |
931 | upindex >>= RADIX_TREE_MAP_SHIFT; | |
932 | offset = upindex & RADIX_TREE_MAP_MASK; | |
933 | ||
144dcfc0 | 934 | /* stop if we find a node with the tag already set */ |
e2bdb933 | 935 | if (tag_get(node, settag, offset)) |
144dcfc0 | 936 | break; |
e2bdb933 HD |
937 | tag_set(node, settag, offset); |
938 | node = node->parent; | |
ebf8aa44 | 939 | } |
144dcfc0 | 940 | |
e2bdb933 HD |
941 | /* |
942 | * Small optimization: now clear that node pointer. | |
943 | * Since all of this slot's ancestors now have the tag set | |
944 | * from setting it above, we have no further need to walk | |
945 | * back up the tree setting tags, until we update slot to | |
946 | * point to another radix_tree_node. | |
947 | */ | |
948 | node = NULL; | |
949 | ||
ebf8aa44 JK |
950 | next: |
951 | /* Go to next item at level determined by 'shift' */ | |
952 | index = ((index >> shift) + 1) << shift; | |
d5ed3a4a JK |
953 | /* Overflow can happen when last_index is ~0UL... */ |
954 | if (index > last_index || !index) | |
ebf8aa44 JK |
955 | break; |
956 | if (tagged >= nr_to_tag) | |
957 | break; | |
958 | while (((index >> shift) & RADIX_TREE_MAP_MASK) == 0) { | |
959 | /* | |
960 | * We've fully scanned this node. Go up. Because | |
961 | * last_index is guaranteed to be in the tree, what | |
962 | * we do below cannot wander astray. | |
963 | */ | |
e2bdb933 | 964 | slot = slot->parent; |
ebf8aa44 JK |
965 | shift += RADIX_TREE_MAP_SHIFT; |
966 | } | |
967 | } | |
968 | /* | |
ac15ee69 TO |
969 | * We need not to tag the root tag if there is no tag which is set with |
970 | * settag within the range from *first_indexp to last_index. | |
ebf8aa44 | 971 | */ |
ac15ee69 TO |
972 | if (tagged > 0) |
973 | root_tag_set(root, settag); | |
ebf8aa44 JK |
974 | *first_indexp = index; |
975 | ||
976 | return tagged; | |
977 | } | |
978 | EXPORT_SYMBOL(radix_tree_range_tag_if_tagged); | |
979 | ||
1da177e4 LT |
980 | /** |
981 | * radix_tree_gang_lookup - perform multiple lookup on a radix tree | |
982 | * @root: radix tree root | |
983 | * @results: where the results of the lookup are placed | |
984 | * @first_index: start the lookup from this key | |
985 | * @max_items: place up to this many items at *results | |
986 | * | |
987 | * Performs an index-ascending scan of the tree for present items. Places | |
988 | * them at *@results and returns the number of items which were placed at | |
989 | * *@results. | |
990 | * | |
991 | * The implementation is naive. | |
7cf9c2c7 NP |
992 | * |
993 | * Like radix_tree_lookup, radix_tree_gang_lookup may be called under | |
994 | * rcu_read_lock. In this case, rather than the returned results being | |
995 | * an atomic snapshot of the tree at a single point in time, the semantics | |
996 | * of an RCU protected gang lookup are as though multiple radix_tree_lookups | |
997 | * have been issued in individual locks, and results stored in 'results'. | |
1da177e4 LT |
998 | */ |
999 | unsigned int | |
1000 | radix_tree_gang_lookup(struct radix_tree_root *root, void **results, | |
1001 | unsigned long first_index, unsigned int max_items) | |
1002 | { | |
cebbd29e KK |
1003 | struct radix_tree_iter iter; |
1004 | void **slot; | |
1005 | unsigned int ret = 0; | |
7cf9c2c7 | 1006 | |
cebbd29e | 1007 | if (unlikely(!max_items)) |
7cf9c2c7 | 1008 | return 0; |
1da177e4 | 1009 | |
cebbd29e KK |
1010 | radix_tree_for_each_slot(slot, root, &iter, first_index) { |
1011 | results[ret] = indirect_to_ptr(rcu_dereference_raw(*slot)); | |
1012 | if (!results[ret]) | |
1013 | continue; | |
1014 | if (++ret == max_items) | |
1da177e4 | 1015 | break; |
1da177e4 | 1016 | } |
7cf9c2c7 | 1017 | |
1da177e4 LT |
1018 | return ret; |
1019 | } | |
1020 | EXPORT_SYMBOL(radix_tree_gang_lookup); | |
1021 | ||
47feff2c NP |
1022 | /** |
1023 | * radix_tree_gang_lookup_slot - perform multiple slot lookup on radix tree | |
1024 | * @root: radix tree root | |
1025 | * @results: where the results of the lookup are placed | |
6328650b | 1026 | * @indices: where their indices should be placed (but usually NULL) |
47feff2c NP |
1027 | * @first_index: start the lookup from this key |
1028 | * @max_items: place up to this many items at *results | |
1029 | * | |
1030 | * Performs an index-ascending scan of the tree for present items. Places | |
1031 | * their slots at *@results and returns the number of items which were | |
1032 | * placed at *@results. | |
1033 | * | |
1034 | * The implementation is naive. | |
1035 | * | |
1036 | * Like radix_tree_gang_lookup as far as RCU and locking goes. Slots must | |
1037 | * be dereferenced with radix_tree_deref_slot, and if using only RCU | |
1038 | * protection, radix_tree_deref_slot may fail requiring a retry. | |
1039 | */ | |
1040 | unsigned int | |
6328650b HD |
1041 | radix_tree_gang_lookup_slot(struct radix_tree_root *root, |
1042 | void ***results, unsigned long *indices, | |
47feff2c NP |
1043 | unsigned long first_index, unsigned int max_items) |
1044 | { | |
cebbd29e KK |
1045 | struct radix_tree_iter iter; |
1046 | void **slot; | |
1047 | unsigned int ret = 0; | |
47feff2c | 1048 | |
cebbd29e | 1049 | if (unlikely(!max_items)) |
47feff2c NP |
1050 | return 0; |
1051 | ||
cebbd29e KK |
1052 | radix_tree_for_each_slot(slot, root, &iter, first_index) { |
1053 | results[ret] = slot; | |
6328650b | 1054 | if (indices) |
cebbd29e KK |
1055 | indices[ret] = iter.index; |
1056 | if (++ret == max_items) | |
47feff2c | 1057 | break; |
47feff2c NP |
1058 | } |
1059 | ||
1060 | return ret; | |
1061 | } | |
1062 | EXPORT_SYMBOL(radix_tree_gang_lookup_slot); | |
1063 | ||
1da177e4 LT |
1064 | /** |
1065 | * radix_tree_gang_lookup_tag - perform multiple lookup on a radix tree | |
1066 | * based on a tag | |
1067 | * @root: radix tree root | |
1068 | * @results: where the results of the lookup are placed | |
1069 | * @first_index: start the lookup from this key | |
1070 | * @max_items: place up to this many items at *results | |
daff89f3 | 1071 | * @tag: the tag index (< RADIX_TREE_MAX_TAGS) |
1da177e4 LT |
1072 | * |
1073 | * Performs an index-ascending scan of the tree for present items which | |
1074 | * have the tag indexed by @tag set. Places the items at *@results and | |
1075 | * returns the number of items which were placed at *@results. | |
1076 | */ | |
1077 | unsigned int | |
1078 | radix_tree_gang_lookup_tag(struct radix_tree_root *root, void **results, | |
daff89f3 JC |
1079 | unsigned long first_index, unsigned int max_items, |
1080 | unsigned int tag) | |
1da177e4 | 1081 | { |
cebbd29e KK |
1082 | struct radix_tree_iter iter; |
1083 | void **slot; | |
1084 | unsigned int ret = 0; | |
612d6c19 | 1085 | |
cebbd29e | 1086 | if (unlikely(!max_items)) |
7cf9c2c7 NP |
1087 | return 0; |
1088 | ||
cebbd29e KK |
1089 | radix_tree_for_each_tagged(slot, root, &iter, first_index, tag) { |
1090 | results[ret] = indirect_to_ptr(rcu_dereference_raw(*slot)); | |
1091 | if (!results[ret]) | |
1092 | continue; | |
1093 | if (++ret == max_items) | |
1da177e4 | 1094 | break; |
1da177e4 | 1095 | } |
7cf9c2c7 | 1096 | |
1da177e4 LT |
1097 | return ret; |
1098 | } | |
1099 | EXPORT_SYMBOL(radix_tree_gang_lookup_tag); | |
1100 | ||
47feff2c NP |
1101 | /** |
1102 | * radix_tree_gang_lookup_tag_slot - perform multiple slot lookup on a | |
1103 | * radix tree based on a tag | |
1104 | * @root: radix tree root | |
1105 | * @results: where the results of the lookup are placed | |
1106 | * @first_index: start the lookup from this key | |
1107 | * @max_items: place up to this many items at *results | |
1108 | * @tag: the tag index (< RADIX_TREE_MAX_TAGS) | |
1109 | * | |
1110 | * Performs an index-ascending scan of the tree for present items which | |
1111 | * have the tag indexed by @tag set. Places the slots at *@results and | |
1112 | * returns the number of slots which were placed at *@results. | |
1113 | */ | |
1114 | unsigned int | |
1115 | radix_tree_gang_lookup_tag_slot(struct radix_tree_root *root, void ***results, | |
1116 | unsigned long first_index, unsigned int max_items, | |
1117 | unsigned int tag) | |
1118 | { | |
cebbd29e KK |
1119 | struct radix_tree_iter iter; |
1120 | void **slot; | |
1121 | unsigned int ret = 0; | |
47feff2c | 1122 | |
cebbd29e | 1123 | if (unlikely(!max_items)) |
47feff2c NP |
1124 | return 0; |
1125 | ||
cebbd29e KK |
1126 | radix_tree_for_each_tagged(slot, root, &iter, first_index, tag) { |
1127 | results[ret] = slot; | |
1128 | if (++ret == max_items) | |
47feff2c | 1129 | break; |
47feff2c NP |
1130 | } |
1131 | ||
1132 | return ret; | |
1133 | } | |
1134 | EXPORT_SYMBOL(radix_tree_gang_lookup_tag_slot); | |
1135 | ||
e504f3fd HD |
1136 | #if defined(CONFIG_SHMEM) && defined(CONFIG_SWAP) |
1137 | #include <linux/sched.h> /* for cond_resched() */ | |
1138 | ||
1139 | /* | |
1140 | * This linear search is at present only useful to shmem_unuse_inode(). | |
1141 | */ | |
1142 | static unsigned long __locate(struct radix_tree_node *slot, void *item, | |
1143 | unsigned long index, unsigned long *found_index) | |
1144 | { | |
1145 | unsigned int shift, height; | |
1146 | unsigned long i; | |
1147 | ||
449dd698 | 1148 | height = slot->path & RADIX_TREE_HEIGHT_MASK; |
e504f3fd HD |
1149 | shift = (height-1) * RADIX_TREE_MAP_SHIFT; |
1150 | ||
1151 | for ( ; height > 1; height--) { | |
1152 | i = (index >> shift) & RADIX_TREE_MAP_MASK; | |
1153 | for (;;) { | |
1154 | if (slot->slots[i] != NULL) | |
1155 | break; | |
1156 | index &= ~((1UL << shift) - 1); | |
1157 | index += 1UL << shift; | |
1158 | if (index == 0) | |
1159 | goto out; /* 32-bit wraparound */ | |
1160 | i++; | |
1161 | if (i == RADIX_TREE_MAP_SIZE) | |
1162 | goto out; | |
1163 | } | |
1164 | ||
1165 | shift -= RADIX_TREE_MAP_SHIFT; | |
1166 | slot = rcu_dereference_raw(slot->slots[i]); | |
1167 | if (slot == NULL) | |
1168 | goto out; | |
1169 | } | |
1170 | ||
1171 | /* Bottom level: check items */ | |
1172 | for (i = 0; i < RADIX_TREE_MAP_SIZE; i++) { | |
1173 | if (slot->slots[i] == item) { | |
1174 | *found_index = index + i; | |
1175 | index = 0; | |
1176 | goto out; | |
1177 | } | |
1178 | } | |
1179 | index += RADIX_TREE_MAP_SIZE; | |
1180 | out: | |
1181 | return index; | |
1182 | } | |
1183 | ||
1184 | /** | |
1185 | * radix_tree_locate_item - search through radix tree for item | |
1186 | * @root: radix tree root | |
1187 | * @item: item to be found | |
1188 | * | |
1189 | * Returns index where item was found, or -1 if not found. | |
1190 | * Caller must hold no lock (since this time-consuming function needs | |
1191 | * to be preemptible), and must check afterwards if item is still there. | |
1192 | */ | |
1193 | unsigned long radix_tree_locate_item(struct radix_tree_root *root, void *item) | |
1194 | { | |
1195 | struct radix_tree_node *node; | |
1196 | unsigned long max_index; | |
1197 | unsigned long cur_index = 0; | |
1198 | unsigned long found_index = -1; | |
1199 | ||
1200 | do { | |
1201 | rcu_read_lock(); | |
1202 | node = rcu_dereference_raw(root->rnode); | |
1203 | if (!radix_tree_is_indirect_ptr(node)) { | |
1204 | rcu_read_unlock(); | |
1205 | if (node == item) | |
1206 | found_index = 0; | |
1207 | break; | |
1208 | } | |
1209 | ||
1210 | node = indirect_to_ptr(node); | |
449dd698 JW |
1211 | max_index = radix_tree_maxindex(node->path & |
1212 | RADIX_TREE_HEIGHT_MASK); | |
5f30fc94 HD |
1213 | if (cur_index > max_index) { |
1214 | rcu_read_unlock(); | |
e504f3fd | 1215 | break; |
5f30fc94 | 1216 | } |
e504f3fd HD |
1217 | |
1218 | cur_index = __locate(node, item, cur_index, &found_index); | |
1219 | rcu_read_unlock(); | |
1220 | cond_resched(); | |
1221 | } while (cur_index != 0 && cur_index <= max_index); | |
1222 | ||
1223 | return found_index; | |
1224 | } | |
1225 | #else | |
1226 | unsigned long radix_tree_locate_item(struct radix_tree_root *root, void *item) | |
1227 | { | |
1228 | return -1; | |
1229 | } | |
1230 | #endif /* CONFIG_SHMEM && CONFIG_SWAP */ | |
47feff2c | 1231 | |
a5f51c96 NP |
1232 | /** |
1233 | * radix_tree_shrink - shrink height of a radix tree to minimal | |
1234 | * @root radix tree root | |
1235 | */ | |
1236 | static inline void radix_tree_shrink(struct radix_tree_root *root) | |
1237 | { | |
1238 | /* try to shrink tree height */ | |
c0bc9875 | 1239 | while (root->height > 0) { |
a5f51c96 | 1240 | struct radix_tree_node *to_free = root->rnode; |
e2bdb933 | 1241 | struct radix_tree_node *slot; |
a5f51c96 | 1242 | |
c0bc9875 | 1243 | BUG_ON(!radix_tree_is_indirect_ptr(to_free)); |
27d20fdd | 1244 | to_free = indirect_to_ptr(to_free); |
c0bc9875 NP |
1245 | |
1246 | /* | |
1247 | * The candidate node has more than one child, or its child | |
1248 | * is not at the leftmost slot, we cannot shrink. | |
1249 | */ | |
1250 | if (to_free->count != 1) | |
1251 | break; | |
1252 | if (!to_free->slots[0]) | |
1253 | break; | |
1254 | ||
7cf9c2c7 NP |
1255 | /* |
1256 | * We don't need rcu_assign_pointer(), since we are simply | |
27d20fdd NP |
1257 | * moving the node from one part of the tree to another: if it |
1258 | * was safe to dereference the old pointer to it | |
7cf9c2c7 | 1259 | * (to_free->slots[0]), it will be safe to dereference the new |
27d20fdd | 1260 | * one (root->rnode) as far as dependent read barriers go. |
7cf9c2c7 | 1261 | */ |
e2bdb933 HD |
1262 | slot = to_free->slots[0]; |
1263 | if (root->height > 1) { | |
1264 | slot->parent = NULL; | |
1265 | slot = ptr_to_indirect(slot); | |
1266 | } | |
1267 | root->rnode = slot; | |
a5f51c96 | 1268 | root->height--; |
27d20fdd NP |
1269 | |
1270 | /* | |
1271 | * We have a dilemma here. The node's slot[0] must not be | |
1272 | * NULLed in case there are concurrent lookups expecting to | |
1273 | * find the item. However if this was a bottom-level node, | |
1274 | * then it may be subject to the slot pointer being visible | |
1275 | * to callers dereferencing it. If item corresponding to | |
1276 | * slot[0] is subsequently deleted, these callers would expect | |
1277 | * their slot to become empty sooner or later. | |
1278 | * | |
1279 | * For example, lockless pagecache will look up a slot, deref | |
1280 | * the page pointer, and if the page is 0 refcount it means it | |
1281 | * was concurrently deleted from pagecache so try the deref | |
1282 | * again. Fortunately there is already a requirement for logic | |
1283 | * to retry the entire slot lookup -- the indirect pointer | |
1284 | * problem (replacing direct root node with an indirect pointer | |
1285 | * also results in a stale slot). So tag the slot as indirect | |
1286 | * to force callers to retry. | |
1287 | */ | |
1288 | if (root->height == 0) | |
1289 | *((unsigned long *)&to_free->slots[0]) |= | |
1290 | RADIX_TREE_INDIRECT_PTR; | |
1291 | ||
a5f51c96 NP |
1292 | radix_tree_node_free(to_free); |
1293 | } | |
1294 | } | |
1295 | ||
139e5616 JW |
1296 | /** |
1297 | * __radix_tree_delete_node - try to free node after clearing a slot | |
1298 | * @root: radix tree root | |
1299 | * @index: index key | |
1300 | * @node: node containing @index | |
1301 | * | |
1302 | * After clearing the slot at @index in @node from radix tree | |
1303 | * rooted at @root, call this function to attempt freeing the | |
1304 | * node and shrinking the tree. | |
1305 | * | |
1306 | * Returns %true if @node was freed, %false otherwise. | |
1307 | */ | |
449dd698 | 1308 | bool __radix_tree_delete_node(struct radix_tree_root *root, |
139e5616 JW |
1309 | struct radix_tree_node *node) |
1310 | { | |
1311 | bool deleted = false; | |
1312 | ||
1313 | do { | |
1314 | struct radix_tree_node *parent; | |
1315 | ||
1316 | if (node->count) { | |
1317 | if (node == indirect_to_ptr(root->rnode)) { | |
1318 | radix_tree_shrink(root); | |
1319 | if (root->height == 0) | |
1320 | deleted = true; | |
1321 | } | |
1322 | return deleted; | |
1323 | } | |
1324 | ||
1325 | parent = node->parent; | |
1326 | if (parent) { | |
449dd698 | 1327 | unsigned int offset; |
139e5616 | 1328 | |
449dd698 JW |
1329 | offset = node->path >> RADIX_TREE_HEIGHT_SHIFT; |
1330 | parent->slots[offset] = NULL; | |
139e5616 JW |
1331 | parent->count--; |
1332 | } else { | |
1333 | root_tag_clear_all(root); | |
1334 | root->height = 0; | |
1335 | root->rnode = NULL; | |
1336 | } | |
1337 | ||
1338 | radix_tree_node_free(node); | |
1339 | deleted = true; | |
1340 | ||
1341 | node = parent; | |
1342 | } while (node); | |
1343 | ||
1344 | return deleted; | |
1345 | } | |
1346 | ||
1da177e4 | 1347 | /** |
53c59f26 | 1348 | * radix_tree_delete_item - delete an item from a radix tree |
1da177e4 LT |
1349 | * @root: radix tree root |
1350 | * @index: index key | |
53c59f26 | 1351 | * @item: expected item |
1da177e4 | 1352 | * |
53c59f26 | 1353 | * Remove @item at @index from the radix tree rooted at @root. |
1da177e4 | 1354 | * |
53c59f26 JW |
1355 | * Returns the address of the deleted item, or NULL if it was not present |
1356 | * or the entry at the given @index was not @item. | |
1da177e4 | 1357 | */ |
53c59f26 JW |
1358 | void *radix_tree_delete_item(struct radix_tree_root *root, |
1359 | unsigned long index, void *item) | |
1da177e4 | 1360 | { |
139e5616 JW |
1361 | struct radix_tree_node *node; |
1362 | unsigned int offset; | |
1363 | void **slot; | |
1364 | void *entry; | |
d5274261 | 1365 | int tag; |
1da177e4 | 1366 | |
139e5616 JW |
1367 | entry = __radix_tree_lookup(root, index, &node, &slot); |
1368 | if (!entry) | |
1369 | return NULL; | |
1da177e4 | 1370 | |
139e5616 JW |
1371 | if (item && entry != item) |
1372 | return NULL; | |
1373 | ||
1374 | if (!node) { | |
612d6c19 NP |
1375 | root_tag_clear_all(root); |
1376 | root->rnode = NULL; | |
139e5616 | 1377 | return entry; |
612d6c19 | 1378 | } |
1da177e4 | 1379 | |
139e5616 | 1380 | offset = index & RADIX_TREE_MAP_MASK; |
53c59f26 | 1381 | |
1da177e4 | 1382 | /* |
e2bdb933 HD |
1383 | * Clear all tags associated with the item to be deleted. |
1384 | * This way of doing it would be inefficient, but seldom is any set. | |
1da177e4 | 1385 | */ |
daff89f3 | 1386 | for (tag = 0; tag < RADIX_TREE_MAX_TAGS; tag++) { |
e2bdb933 | 1387 | if (tag_get(node, tag, offset)) |
612d6c19 | 1388 | radix_tree_tag_clear(root, index, tag); |
d5274261 | 1389 | } |
1da177e4 | 1390 | |
139e5616 JW |
1391 | node->slots[offset] = NULL; |
1392 | node->count--; | |
e2bdb933 | 1393 | |
449dd698 | 1394 | __radix_tree_delete_node(root, node); |
612d6c19 | 1395 | |
139e5616 | 1396 | return entry; |
1da177e4 | 1397 | } |
53c59f26 JW |
1398 | EXPORT_SYMBOL(radix_tree_delete_item); |
1399 | ||
1400 | /** | |
1401 | * radix_tree_delete - delete an item from a radix tree | |
1402 | * @root: radix tree root | |
1403 | * @index: index key | |
1404 | * | |
1405 | * Remove the item at @index from the radix tree rooted at @root. | |
1406 | * | |
1407 | * Returns the address of the deleted item, or NULL if it was not present. | |
1408 | */ | |
1409 | void *radix_tree_delete(struct radix_tree_root *root, unsigned long index) | |
1410 | { | |
1411 | return radix_tree_delete_item(root, index, NULL); | |
1412 | } | |
1da177e4 LT |
1413 | EXPORT_SYMBOL(radix_tree_delete); |
1414 | ||
1415 | /** | |
1416 | * radix_tree_tagged - test whether any items in the tree are tagged | |
1417 | * @root: radix tree root | |
1418 | * @tag: tag to test | |
1419 | */ | |
daff89f3 | 1420 | int radix_tree_tagged(struct radix_tree_root *root, unsigned int tag) |
1da177e4 | 1421 | { |
612d6c19 | 1422 | return root_tag_get(root, tag); |
1da177e4 LT |
1423 | } |
1424 | EXPORT_SYMBOL(radix_tree_tagged); | |
1425 | ||
1426 | static void | |
449dd698 | 1427 | radix_tree_node_ctor(void *arg) |
1da177e4 | 1428 | { |
449dd698 JW |
1429 | struct radix_tree_node *node = arg; |
1430 | ||
1431 | memset(node, 0, sizeof(*node)); | |
1432 | INIT_LIST_HEAD(&node->private_list); | |
1da177e4 LT |
1433 | } |
1434 | ||
1435 | static __init unsigned long __maxindex(unsigned int height) | |
1436 | { | |
430d275a PL |
1437 | unsigned int width = height * RADIX_TREE_MAP_SHIFT; |
1438 | int shift = RADIX_TREE_INDEX_BITS - width; | |
1439 | ||
1440 | if (shift < 0) | |
1441 | return ~0UL; | |
1442 | if (shift >= BITS_PER_LONG) | |
1443 | return 0UL; | |
1444 | return ~0UL >> shift; | |
1da177e4 LT |
1445 | } |
1446 | ||
1447 | static __init void radix_tree_init_maxindex(void) | |
1448 | { | |
1449 | unsigned int i; | |
1450 | ||
1451 | for (i = 0; i < ARRAY_SIZE(height_to_maxindex); i++) | |
1452 | height_to_maxindex[i] = __maxindex(i); | |
1453 | } | |
1454 | ||
1da177e4 LT |
1455 | static int radix_tree_callback(struct notifier_block *nfb, |
1456 | unsigned long action, | |
1457 | void *hcpu) | |
1458 | { | |
1459 | int cpu = (long)hcpu; | |
1460 | struct radix_tree_preload *rtp; | |
1461 | ||
1462 | /* Free per-cpu pool of perloaded nodes */ | |
8bb78442 | 1463 | if (action == CPU_DEAD || action == CPU_DEAD_FROZEN) { |
1da177e4 LT |
1464 | rtp = &per_cpu(radix_tree_preloads, cpu); |
1465 | while (rtp->nr) { | |
1466 | kmem_cache_free(radix_tree_node_cachep, | |
1467 | rtp->nodes[rtp->nr-1]); | |
1468 | rtp->nodes[rtp->nr-1] = NULL; | |
1469 | rtp->nr--; | |
1470 | } | |
1471 | } | |
1472 | return NOTIFY_OK; | |
1473 | } | |
1da177e4 LT |
1474 | |
1475 | void __init radix_tree_init(void) | |
1476 | { | |
1477 | radix_tree_node_cachep = kmem_cache_create("radix_tree_node", | |
1478 | sizeof(struct radix_tree_node), 0, | |
488514d1 CL |
1479 | SLAB_PANIC | SLAB_RECLAIM_ACCOUNT, |
1480 | radix_tree_node_ctor); | |
1da177e4 LT |
1481 | radix_tree_init_maxindex(); |
1482 | hotcpu_notifier(radix_tree_callback, 0); | |
1483 | } |