]>
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> | |
ce80b067 | 30 | #include <linux/kmemleak.h> |
1da177e4 LT |
31 | #include <linux/notifier.h> |
32 | #include <linux/cpu.h> | |
1da177e4 LT |
33 | #include <linux/string.h> |
34 | #include <linux/bitops.h> | |
7cf9c2c7 | 35 | #include <linux/rcupdate.h> |
92cf2118 | 36 | #include <linux/preempt.h> /* in_interrupt() */ |
1da177e4 LT |
37 | |
38 | ||
26fb1589 JM |
39 | /* |
40 | * The height_to_maxindex array needs to be one deeper than the maximum | |
41 | * path as height 0 holds only 1 entry. | |
42 | */ | |
43 | static unsigned long height_to_maxindex[RADIX_TREE_MAX_PATH + 1] __read_mostly; | |
1da177e4 LT |
44 | |
45 | /* | |
46 | * Radix tree node cache. | |
47 | */ | |
e18b890b | 48 | static struct kmem_cache *radix_tree_node_cachep; |
1da177e4 | 49 | |
55368052 NP |
50 | /* |
51 | * The radix tree is variable-height, so an insert operation not only has | |
52 | * to build the branch to its corresponding item, it also has to build the | |
53 | * branch to existing items if the size has to be increased (by | |
54 | * radix_tree_extend). | |
55 | * | |
56 | * The worst case is a zero height tree with just a single item at index 0, | |
57 | * and then inserting an item at index ULONG_MAX. This requires 2 new branches | |
58 | * of RADIX_TREE_MAX_PATH size to be created, with only the root node shared. | |
59 | * Hence: | |
60 | */ | |
61 | #define RADIX_TREE_PRELOAD_SIZE (RADIX_TREE_MAX_PATH * 2 - 1) | |
62 | ||
1da177e4 LT |
63 | /* |
64 | * Per-cpu pool of preloaded nodes | |
65 | */ | |
66 | struct radix_tree_preload { | |
67 | int nr; | |
9d2a8da0 KS |
68 | /* nodes->private_data points to next preallocated node */ |
69 | struct radix_tree_node *nodes; | |
1da177e4 | 70 | }; |
8cef7d57 | 71 | static DEFINE_PER_CPU(struct radix_tree_preload, radix_tree_preloads) = { 0, }; |
1da177e4 | 72 | |
27d20fdd NP |
73 | static inline void *ptr_to_indirect(void *ptr) |
74 | { | |
75 | return (void *)((unsigned long)ptr | RADIX_TREE_INDIRECT_PTR); | |
76 | } | |
77 | ||
78 | static inline void *indirect_to_ptr(void *ptr) | |
79 | { | |
80 | return (void *)((unsigned long)ptr & ~RADIX_TREE_INDIRECT_PTR); | |
81 | } | |
82 | ||
612d6c19 NP |
83 | static inline gfp_t root_gfp_mask(struct radix_tree_root *root) |
84 | { | |
85 | return root->gfp_mask & __GFP_BITS_MASK; | |
86 | } | |
87 | ||
643b52b9 NP |
88 | static inline void tag_set(struct radix_tree_node *node, unsigned int tag, |
89 | int offset) | |
90 | { | |
91 | __set_bit(offset, node->tags[tag]); | |
92 | } | |
93 | ||
94 | static inline void tag_clear(struct radix_tree_node *node, unsigned int tag, | |
95 | int offset) | |
96 | { | |
97 | __clear_bit(offset, node->tags[tag]); | |
98 | } | |
99 | ||
100 | static inline int tag_get(struct radix_tree_node *node, unsigned int tag, | |
101 | int offset) | |
102 | { | |
103 | return test_bit(offset, node->tags[tag]); | |
104 | } | |
105 | ||
106 | static inline void root_tag_set(struct radix_tree_root *root, unsigned int tag) | |
107 | { | |
108 | root->gfp_mask |= (__force gfp_t)(1 << (tag + __GFP_BITS_SHIFT)); | |
109 | } | |
110 | ||
111 | static inline void root_tag_clear(struct radix_tree_root *root, unsigned int tag) | |
112 | { | |
113 | root->gfp_mask &= (__force gfp_t)~(1 << (tag + __GFP_BITS_SHIFT)); | |
114 | } | |
115 | ||
116 | static inline void root_tag_clear_all(struct radix_tree_root *root) | |
117 | { | |
118 | root->gfp_mask &= __GFP_BITS_MASK; | |
119 | } | |
120 | ||
121 | static inline int root_tag_get(struct radix_tree_root *root, unsigned int tag) | |
122 | { | |
123 | return (__force unsigned)root->gfp_mask & (1 << (tag + __GFP_BITS_SHIFT)); | |
124 | } | |
125 | ||
126 | /* | |
127 | * Returns 1 if any slot in the node has this tag set. | |
128 | * Otherwise returns 0. | |
129 | */ | |
130 | static inline int any_tag_set(struct radix_tree_node *node, unsigned int tag) | |
131 | { | |
132 | int idx; | |
133 | for (idx = 0; idx < RADIX_TREE_TAG_LONGS; idx++) { | |
134 | if (node->tags[tag][idx]) | |
135 | return 1; | |
136 | } | |
137 | return 0; | |
138 | } | |
78c1d784 KK |
139 | |
140 | /** | |
141 | * radix_tree_find_next_bit - find the next set bit in a memory region | |
142 | * | |
143 | * @addr: The address to base the search on | |
144 | * @size: The bitmap size in bits | |
145 | * @offset: The bitnumber to start searching at | |
146 | * | |
147 | * Unrollable variant of find_next_bit() for constant size arrays. | |
148 | * Tail bits starting from size to roundup(size, BITS_PER_LONG) must be zero. | |
149 | * Returns next bit offset, or size if nothing found. | |
150 | */ | |
151 | static __always_inline unsigned long | |
152 | radix_tree_find_next_bit(const unsigned long *addr, | |
153 | unsigned long size, unsigned long offset) | |
154 | { | |
155 | if (!__builtin_constant_p(size)) | |
156 | return find_next_bit(addr, size, offset); | |
157 | ||
158 | if (offset < size) { | |
159 | unsigned long tmp; | |
160 | ||
161 | addr += offset / BITS_PER_LONG; | |
162 | tmp = *addr >> (offset % BITS_PER_LONG); | |
163 | if (tmp) | |
164 | return __ffs(tmp) + offset; | |
165 | offset = (offset + BITS_PER_LONG) & ~(BITS_PER_LONG - 1); | |
166 | while (offset < size) { | |
167 | tmp = *++addr; | |
168 | if (tmp) | |
169 | return __ffs(tmp) + offset; | |
170 | offset += BITS_PER_LONG; | |
171 | } | |
172 | } | |
173 | return size; | |
174 | } | |
175 | ||
1da177e4 LT |
176 | /* |
177 | * This assumes that the caller has performed appropriate preallocation, and | |
178 | * that the caller has pinned this thread of control to the current CPU. | |
179 | */ | |
180 | static struct radix_tree_node * | |
181 | radix_tree_node_alloc(struct radix_tree_root *root) | |
182 | { | |
e2848a0e | 183 | struct radix_tree_node *ret = NULL; |
612d6c19 | 184 | gfp_t gfp_mask = root_gfp_mask(root); |
1da177e4 | 185 | |
5e4c0d97 JK |
186 | /* |
187 | * Preload code isn't irq safe and it doesn't make sence to use | |
188 | * preloading in the interrupt anyway as all the allocations have to | |
189 | * be atomic. So just do normal allocation when in interrupt. | |
190 | */ | |
d0164adc | 191 | if (!gfpflags_allow_blocking(gfp_mask) && !in_interrupt()) { |
1da177e4 LT |
192 | struct radix_tree_preload *rtp; |
193 | ||
58e698af VD |
194 | /* |
195 | * Even if the caller has preloaded, try to allocate from the | |
196 | * cache first for the new node to get accounted. | |
197 | */ | |
198 | ret = kmem_cache_alloc(radix_tree_node_cachep, | |
199 | gfp_mask | __GFP_ACCOUNT | __GFP_NOWARN); | |
200 | if (ret) | |
201 | goto out; | |
202 | ||
e2848a0e NP |
203 | /* |
204 | * Provided the caller has preloaded here, we will always | |
205 | * succeed in getting a node here (and never reach | |
206 | * kmem_cache_alloc) | |
207 | */ | |
7c8e0181 | 208 | rtp = this_cpu_ptr(&radix_tree_preloads); |
1da177e4 | 209 | if (rtp->nr) { |
9d2a8da0 KS |
210 | ret = rtp->nodes; |
211 | rtp->nodes = ret->private_data; | |
212 | ret->private_data = NULL; | |
1da177e4 LT |
213 | rtp->nr--; |
214 | } | |
ce80b067 CM |
215 | /* |
216 | * Update the allocation stack trace as this is more useful | |
217 | * for debugging. | |
218 | */ | |
219 | kmemleak_update_trace(ret); | |
58e698af | 220 | goto out; |
1da177e4 | 221 | } |
58e698af VD |
222 | ret = kmem_cache_alloc(radix_tree_node_cachep, |
223 | gfp_mask | __GFP_ACCOUNT); | |
224 | out: | |
c0bc9875 | 225 | BUG_ON(radix_tree_is_indirect_ptr(ret)); |
1da177e4 LT |
226 | return ret; |
227 | } | |
228 | ||
7cf9c2c7 NP |
229 | static void radix_tree_node_rcu_free(struct rcu_head *head) |
230 | { | |
231 | struct radix_tree_node *node = | |
232 | container_of(head, struct radix_tree_node, rcu_head); | |
b6dd0865 | 233 | int i; |
643b52b9 NP |
234 | |
235 | /* | |
236 | * must only free zeroed nodes into the slab. radix_tree_shrink | |
237 | * can leave us with a non-NULL entry in the first slot, so clear | |
238 | * that here to make sure. | |
239 | */ | |
b6dd0865 DC |
240 | for (i = 0; i < RADIX_TREE_MAX_TAGS; i++) |
241 | tag_clear(node, i, 0); | |
242 | ||
643b52b9 NP |
243 | node->slots[0] = NULL; |
244 | node->count = 0; | |
245 | ||
7cf9c2c7 NP |
246 | kmem_cache_free(radix_tree_node_cachep, node); |
247 | } | |
248 | ||
1da177e4 LT |
249 | static inline void |
250 | radix_tree_node_free(struct radix_tree_node *node) | |
251 | { | |
7cf9c2c7 | 252 | call_rcu(&node->rcu_head, radix_tree_node_rcu_free); |
1da177e4 LT |
253 | } |
254 | ||
255 | /* | |
256 | * Load up this CPU's radix_tree_node buffer with sufficient objects to | |
257 | * ensure that the addition of a single element in the tree cannot fail. On | |
258 | * success, return zero, with preemption disabled. On error, return -ENOMEM | |
259 | * with preemption not disabled. | |
b34df792 DH |
260 | * |
261 | * To make use of this facility, the radix tree must be initialised without | |
d0164adc | 262 | * __GFP_DIRECT_RECLAIM being passed to INIT_RADIX_TREE(). |
1da177e4 | 263 | */ |
5e4c0d97 | 264 | static int __radix_tree_preload(gfp_t gfp_mask) |
1da177e4 LT |
265 | { |
266 | struct radix_tree_preload *rtp; | |
267 | struct radix_tree_node *node; | |
268 | int ret = -ENOMEM; | |
269 | ||
270 | preempt_disable(); | |
7c8e0181 | 271 | rtp = this_cpu_ptr(&radix_tree_preloads); |
9d2a8da0 | 272 | while (rtp->nr < RADIX_TREE_PRELOAD_SIZE) { |
1da177e4 | 273 | preempt_enable(); |
488514d1 | 274 | node = kmem_cache_alloc(radix_tree_node_cachep, gfp_mask); |
1da177e4 LT |
275 | if (node == NULL) |
276 | goto out; | |
277 | preempt_disable(); | |
7c8e0181 | 278 | rtp = this_cpu_ptr(&radix_tree_preloads); |
9d2a8da0 KS |
279 | if (rtp->nr < RADIX_TREE_PRELOAD_SIZE) { |
280 | node->private_data = rtp->nodes; | |
281 | rtp->nodes = node; | |
282 | rtp->nr++; | |
283 | } else { | |
1da177e4 | 284 | kmem_cache_free(radix_tree_node_cachep, node); |
9d2a8da0 | 285 | } |
1da177e4 LT |
286 | } |
287 | ret = 0; | |
288 | out: | |
289 | return ret; | |
290 | } | |
5e4c0d97 JK |
291 | |
292 | /* | |
293 | * Load up this CPU's radix_tree_node buffer with sufficient objects to | |
294 | * ensure that the addition of a single element in the tree cannot fail. On | |
295 | * success, return zero, with preemption disabled. On error, return -ENOMEM | |
296 | * with preemption not disabled. | |
297 | * | |
298 | * To make use of this facility, the radix tree must be initialised without | |
d0164adc | 299 | * __GFP_DIRECT_RECLAIM being passed to INIT_RADIX_TREE(). |
5e4c0d97 JK |
300 | */ |
301 | int radix_tree_preload(gfp_t gfp_mask) | |
302 | { | |
303 | /* Warn on non-sensical use... */ | |
d0164adc | 304 | WARN_ON_ONCE(!gfpflags_allow_blocking(gfp_mask)); |
5e4c0d97 JK |
305 | return __radix_tree_preload(gfp_mask); |
306 | } | |
d7f0923d | 307 | EXPORT_SYMBOL(radix_tree_preload); |
1da177e4 | 308 | |
5e4c0d97 JK |
309 | /* |
310 | * The same as above function, except we don't guarantee preloading happens. | |
311 | * We do it, if we decide it helps. On success, return zero with preemption | |
312 | * disabled. On error, return -ENOMEM with preemption not disabled. | |
313 | */ | |
314 | int radix_tree_maybe_preload(gfp_t gfp_mask) | |
315 | { | |
d0164adc | 316 | if (gfpflags_allow_blocking(gfp_mask)) |
5e4c0d97 JK |
317 | return __radix_tree_preload(gfp_mask); |
318 | /* Preloading doesn't help anything with this gfp mask, skip it */ | |
319 | preempt_disable(); | |
320 | return 0; | |
321 | } | |
322 | EXPORT_SYMBOL(radix_tree_maybe_preload); | |
323 | ||
1da177e4 LT |
324 | /* |
325 | * Return the maximum key which can be store into a | |
326 | * radix tree with height HEIGHT. | |
327 | */ | |
328 | static inline unsigned long radix_tree_maxindex(unsigned int height) | |
329 | { | |
330 | return height_to_maxindex[height]; | |
331 | } | |
332 | ||
333 | /* | |
334 | * Extend a radix tree so it can store key @index. | |
335 | */ | |
336 | static int radix_tree_extend(struct radix_tree_root *root, unsigned long index) | |
337 | { | |
338 | struct radix_tree_node *node; | |
e2bdb933 | 339 | struct radix_tree_node *slot; |
1da177e4 | 340 | unsigned int height; |
1da177e4 LT |
341 | int tag; |
342 | ||
343 | /* Figure out what the height should be. */ | |
344 | height = root->height + 1; | |
345 | while (index > radix_tree_maxindex(height)) | |
346 | height++; | |
347 | ||
348 | if (root->rnode == NULL) { | |
349 | root->height = height; | |
350 | goto out; | |
351 | } | |
352 | ||
1da177e4 | 353 | do { |
7cf9c2c7 | 354 | unsigned int newheight; |
1da177e4 LT |
355 | if (!(node = radix_tree_node_alloc(root))) |
356 | return -ENOMEM; | |
357 | ||
1da177e4 | 358 | /* Propagate the aggregated tag info into the new root */ |
daff89f3 | 359 | for (tag = 0; tag < RADIX_TREE_MAX_TAGS; tag++) { |
612d6c19 | 360 | if (root_tag_get(root, tag)) |
1da177e4 LT |
361 | tag_set(node, tag, 0); |
362 | } | |
363 | ||
e2bdb933 | 364 | /* Increase the height. */ |
7cf9c2c7 | 365 | newheight = root->height+1; |
449dd698 JW |
366 | BUG_ON(newheight & ~RADIX_TREE_HEIGHT_MASK); |
367 | node->path = newheight; | |
1da177e4 | 368 | node->count = 1; |
e2bdb933 HD |
369 | node->parent = NULL; |
370 | slot = root->rnode; | |
339e6353 | 371 | if (radix_tree_is_indirect_ptr(slot) && newheight > 1) { |
e2bdb933 HD |
372 | slot = indirect_to_ptr(slot); |
373 | slot->parent = node; | |
339e6353 | 374 | slot = ptr_to_indirect(slot); |
e2bdb933 HD |
375 | } |
376 | node->slots[0] = slot; | |
27d20fdd | 377 | node = ptr_to_indirect(node); |
7cf9c2c7 NP |
378 | rcu_assign_pointer(root->rnode, node); |
379 | root->height = newheight; | |
1da177e4 LT |
380 | } while (height > root->height); |
381 | out: | |
382 | return 0; | |
383 | } | |
384 | ||
385 | /** | |
139e5616 | 386 | * __radix_tree_create - create a slot in a radix tree |
1da177e4 LT |
387 | * @root: radix tree root |
388 | * @index: index key | |
139e5616 JW |
389 | * @nodep: returns node |
390 | * @slotp: returns slot | |
1da177e4 | 391 | * |
139e5616 JW |
392 | * Create, if necessary, and return the node and slot for an item |
393 | * at position @index in the radix tree @root. | |
394 | * | |
395 | * Until there is more than one item in the tree, no nodes are | |
396 | * allocated and @root->rnode is used as a direct slot instead of | |
397 | * pointing to a node, in which case *@nodep will be NULL. | |
398 | * | |
399 | * Returns -ENOMEM, or 0 for success. | |
1da177e4 | 400 | */ |
139e5616 JW |
401 | int __radix_tree_create(struct radix_tree_root *root, unsigned long index, |
402 | struct radix_tree_node **nodep, void ***slotp) | |
1da177e4 | 403 | { |
201b6264 | 404 | struct radix_tree_node *node = NULL, *slot; |
139e5616 | 405 | unsigned int height, shift, offset; |
1da177e4 LT |
406 | int error; |
407 | ||
408 | /* Make sure the tree is high enough. */ | |
612d6c19 | 409 | if (index > radix_tree_maxindex(root->height)) { |
1da177e4 LT |
410 | error = radix_tree_extend(root, index); |
411 | if (error) | |
412 | return error; | |
413 | } | |
414 | ||
27d20fdd | 415 | slot = indirect_to_ptr(root->rnode); |
c0bc9875 | 416 | |
1da177e4 | 417 | height = root->height; |
0070e28d | 418 | shift = height * RADIX_TREE_MAP_SHIFT; |
1da177e4 LT |
419 | |
420 | offset = 0; /* uninitialised var warning */ | |
0070e28d | 421 | while (shift > 0) { |
201b6264 | 422 | if (slot == NULL) { |
1da177e4 | 423 | /* Have to add a child node. */ |
201b6264 | 424 | if (!(slot = radix_tree_node_alloc(root))) |
1da177e4 | 425 | return -ENOMEM; |
449dd698 | 426 | slot->path = height; |
e2bdb933 | 427 | slot->parent = node; |
201b6264 | 428 | if (node) { |
339e6353 MW |
429 | rcu_assign_pointer(node->slots[offset], |
430 | ptr_to_indirect(slot)); | |
1da177e4 | 431 | node->count++; |
449dd698 | 432 | slot->path |= offset << RADIX_TREE_HEIGHT_SHIFT; |
201b6264 | 433 | } else |
339e6353 MW |
434 | rcu_assign_pointer(root->rnode, |
435 | ptr_to_indirect(slot)); | |
1da177e4 LT |
436 | } |
437 | ||
438 | /* Go a level down */ | |
0070e28d | 439 | shift -= RADIX_TREE_MAP_SHIFT; |
1da177e4 | 440 | offset = (index >> shift) & RADIX_TREE_MAP_MASK; |
201b6264 CL |
441 | node = slot; |
442 | slot = node->slots[offset]; | |
339e6353 | 443 | slot = indirect_to_ptr(slot); |
1da177e4 | 444 | height--; |
612d6c19 | 445 | } |
1da177e4 | 446 | |
139e5616 JW |
447 | if (nodep) |
448 | *nodep = node; | |
449 | if (slotp) | |
450 | *slotp = node ? node->slots + offset : (void **)&root->rnode; | |
451 | return 0; | |
452 | } | |
453 | ||
454 | /** | |
455 | * radix_tree_insert - insert into a radix tree | |
456 | * @root: radix tree root | |
457 | * @index: index key | |
458 | * @item: item to insert | |
459 | * | |
460 | * Insert an item into the radix tree at position @index. | |
461 | */ | |
462 | int radix_tree_insert(struct radix_tree_root *root, | |
463 | unsigned long index, void *item) | |
464 | { | |
465 | struct radix_tree_node *node; | |
466 | void **slot; | |
467 | int error; | |
468 | ||
469 | BUG_ON(radix_tree_is_indirect_ptr(item)); | |
470 | ||
471 | error = __radix_tree_create(root, index, &node, &slot); | |
472 | if (error) | |
473 | return error; | |
474 | if (*slot != NULL) | |
1da177e4 | 475 | return -EEXIST; |
139e5616 | 476 | rcu_assign_pointer(*slot, item); |
201b6264 | 477 | |
612d6c19 NP |
478 | if (node) { |
479 | node->count++; | |
139e5616 JW |
480 | BUG_ON(tag_get(node, 0, index & RADIX_TREE_MAP_MASK)); |
481 | BUG_ON(tag_get(node, 1, index & RADIX_TREE_MAP_MASK)); | |
612d6c19 | 482 | } else { |
612d6c19 NP |
483 | BUG_ON(root_tag_get(root, 0)); |
484 | BUG_ON(root_tag_get(root, 1)); | |
485 | } | |
1da177e4 | 486 | |
1da177e4 LT |
487 | return 0; |
488 | } | |
489 | EXPORT_SYMBOL(radix_tree_insert); | |
490 | ||
139e5616 JW |
491 | /** |
492 | * __radix_tree_lookup - lookup an item in a radix tree | |
493 | * @root: radix tree root | |
494 | * @index: index key | |
495 | * @nodep: returns node | |
496 | * @slotp: returns slot | |
497 | * | |
498 | * Lookup and return the item at position @index in the radix | |
499 | * tree @root. | |
500 | * | |
501 | * Until there is more than one item in the tree, no nodes are | |
502 | * allocated and @root->rnode is used as a direct slot instead of | |
503 | * pointing to a node, in which case *@nodep will be NULL. | |
7cf9c2c7 | 504 | */ |
139e5616 JW |
505 | void *__radix_tree_lookup(struct radix_tree_root *root, unsigned long index, |
506 | struct radix_tree_node **nodep, void ***slotp) | |
1da177e4 | 507 | { |
139e5616 | 508 | struct radix_tree_node *node, *parent; |
1da177e4 | 509 | unsigned int height, shift; |
139e5616 | 510 | void **slot; |
612d6c19 | 511 | |
2676a58c | 512 | node = rcu_dereference_raw(root->rnode); |
7cf9c2c7 | 513 | if (node == NULL) |
1da177e4 LT |
514 | return NULL; |
515 | ||
c0bc9875 | 516 | if (!radix_tree_is_indirect_ptr(node)) { |
7cf9c2c7 NP |
517 | if (index > 0) |
518 | return NULL; | |
139e5616 JW |
519 | |
520 | if (nodep) | |
521 | *nodep = NULL; | |
522 | if (slotp) | |
523 | *slotp = (void **)&root->rnode; | |
524 | return node; | |
7cf9c2c7 | 525 | } |
27d20fdd | 526 | node = indirect_to_ptr(node); |
7cf9c2c7 | 527 | |
449dd698 | 528 | height = node->path & RADIX_TREE_HEIGHT_MASK; |
7cf9c2c7 NP |
529 | if (index > radix_tree_maxindex(height)) |
530 | return NULL; | |
612d6c19 | 531 | |
1da177e4 | 532 | shift = (height-1) * RADIX_TREE_MAP_SHIFT; |
1da177e4 | 533 | |
7cf9c2c7 | 534 | do { |
139e5616 JW |
535 | parent = node; |
536 | slot = node->slots + ((index >> shift) & RADIX_TREE_MAP_MASK); | |
2676a58c | 537 | node = rcu_dereference_raw(*slot); |
7cf9c2c7 | 538 | if (node == NULL) |
1da177e4 | 539 | return NULL; |
339e6353 | 540 | node = indirect_to_ptr(node); |
1da177e4 | 541 | |
1da177e4 LT |
542 | shift -= RADIX_TREE_MAP_SHIFT; |
543 | height--; | |
7cf9c2c7 | 544 | } while (height > 0); |
1da177e4 | 545 | |
139e5616 JW |
546 | if (nodep) |
547 | *nodep = parent; | |
548 | if (slotp) | |
549 | *slotp = slot; | |
550 | return node; | |
b72b71c6 HS |
551 | } |
552 | ||
553 | /** | |
554 | * radix_tree_lookup_slot - lookup a slot in a radix tree | |
555 | * @root: radix tree root | |
556 | * @index: index key | |
557 | * | |
558 | * Returns: the slot corresponding to the position @index in the | |
559 | * radix tree @root. This is useful for update-if-exists operations. | |
560 | * | |
561 | * This function can be called under rcu_read_lock iff the slot is not | |
562 | * modified by radix_tree_replace_slot, otherwise it must be called | |
563 | * exclusive from other writers. Any dereference of the slot must be done | |
564 | * using radix_tree_deref_slot. | |
565 | */ | |
566 | void **radix_tree_lookup_slot(struct radix_tree_root *root, unsigned long index) | |
567 | { | |
139e5616 JW |
568 | void **slot; |
569 | ||
570 | if (!__radix_tree_lookup(root, index, NULL, &slot)) | |
571 | return NULL; | |
572 | return slot; | |
a4331366 | 573 | } |
a4331366 HR |
574 | EXPORT_SYMBOL(radix_tree_lookup_slot); |
575 | ||
576 | /** | |
577 | * radix_tree_lookup - perform lookup operation on a radix tree | |
578 | * @root: radix tree root | |
579 | * @index: index key | |
580 | * | |
581 | * Lookup the item at the position @index in the radix tree @root. | |
7cf9c2c7 NP |
582 | * |
583 | * This function can be called under rcu_read_lock, however the caller | |
584 | * must manage lifetimes of leaf nodes (eg. RCU may also be used to free | |
585 | * them safely). No RCU barriers are required to access or modify the | |
586 | * returned item, however. | |
a4331366 HR |
587 | */ |
588 | void *radix_tree_lookup(struct radix_tree_root *root, unsigned long index) | |
589 | { | |
139e5616 | 590 | return __radix_tree_lookup(root, index, NULL, NULL); |
1da177e4 LT |
591 | } |
592 | EXPORT_SYMBOL(radix_tree_lookup); | |
593 | ||
594 | /** | |
595 | * radix_tree_tag_set - set a tag on a radix tree node | |
596 | * @root: radix tree root | |
597 | * @index: index key | |
598 | * @tag: tag index | |
599 | * | |
daff89f3 JC |
600 | * Set the search tag (which must be < RADIX_TREE_MAX_TAGS) |
601 | * corresponding to @index in the radix tree. From | |
1da177e4 LT |
602 | * the root all the way down to the leaf node. |
603 | * | |
604 | * Returns the address of the tagged item. Setting a tag on a not-present | |
605 | * item is a bug. | |
606 | */ | |
607 | void *radix_tree_tag_set(struct radix_tree_root *root, | |
daff89f3 | 608 | unsigned long index, unsigned int tag) |
1da177e4 LT |
609 | { |
610 | unsigned int height, shift; | |
201b6264 | 611 | struct radix_tree_node *slot; |
1da177e4 LT |
612 | |
613 | height = root->height; | |
4c91c364 | 614 | BUG_ON(index > radix_tree_maxindex(height)); |
1da177e4 | 615 | |
27d20fdd | 616 | slot = indirect_to_ptr(root->rnode); |
612d6c19 | 617 | shift = (height - 1) * RADIX_TREE_MAP_SHIFT; |
1da177e4 LT |
618 | |
619 | while (height > 0) { | |
620 | int offset; | |
621 | ||
622 | offset = (index >> shift) & RADIX_TREE_MAP_MASK; | |
d5274261 NP |
623 | if (!tag_get(slot, tag, offset)) |
624 | tag_set(slot, tag, offset); | |
201b6264 CL |
625 | slot = slot->slots[offset]; |
626 | BUG_ON(slot == NULL); | |
339e6353 | 627 | slot = indirect_to_ptr(slot); |
1da177e4 LT |
628 | shift -= RADIX_TREE_MAP_SHIFT; |
629 | height--; | |
630 | } | |
631 | ||
612d6c19 NP |
632 | /* set the root's tag bit */ |
633 | if (slot && !root_tag_get(root, tag)) | |
634 | root_tag_set(root, tag); | |
635 | ||
201b6264 | 636 | return slot; |
1da177e4 LT |
637 | } |
638 | EXPORT_SYMBOL(radix_tree_tag_set); | |
639 | ||
640 | /** | |
641 | * radix_tree_tag_clear - clear a tag on a radix tree node | |
642 | * @root: radix tree root | |
643 | * @index: index key | |
644 | * @tag: tag index | |
645 | * | |
daff89f3 JC |
646 | * Clear the search tag (which must be < RADIX_TREE_MAX_TAGS) |
647 | * corresponding to @index in the radix tree. If | |
1da177e4 LT |
648 | * this causes the leaf node to have no tags set then clear the tag in the |
649 | * next-to-leaf node, etc. | |
650 | * | |
651 | * Returns the address of the tagged item on success, else NULL. ie: | |
652 | * has the same return value and semantics as radix_tree_lookup(). | |
653 | */ | |
654 | void *radix_tree_tag_clear(struct radix_tree_root *root, | |
daff89f3 | 655 | unsigned long index, unsigned int tag) |
1da177e4 | 656 | { |
e2bdb933 | 657 | struct radix_tree_node *node = NULL; |
612d6c19 | 658 | struct radix_tree_node *slot = NULL; |
1da177e4 | 659 | unsigned int height, shift; |
e2bdb933 | 660 | int uninitialized_var(offset); |
1da177e4 LT |
661 | |
662 | height = root->height; | |
663 | if (index > radix_tree_maxindex(height)) | |
664 | goto out; | |
665 | ||
e2bdb933 | 666 | shift = height * RADIX_TREE_MAP_SHIFT; |
339e6353 | 667 | slot = root->rnode; |
1da177e4 | 668 | |
e2bdb933 | 669 | while (shift) { |
201b6264 | 670 | if (slot == NULL) |
1da177e4 | 671 | goto out; |
339e6353 | 672 | slot = indirect_to_ptr(slot); |
1da177e4 | 673 | |
e2bdb933 | 674 | shift -= RADIX_TREE_MAP_SHIFT; |
1da177e4 | 675 | offset = (index >> shift) & RADIX_TREE_MAP_MASK; |
e2bdb933 | 676 | node = slot; |
201b6264 | 677 | slot = slot->slots[offset]; |
1da177e4 LT |
678 | } |
679 | ||
612d6c19 | 680 | if (slot == NULL) |
1da177e4 LT |
681 | goto out; |
682 | ||
e2bdb933 HD |
683 | while (node) { |
684 | if (!tag_get(node, tag, offset)) | |
d5274261 | 685 | goto out; |
e2bdb933 HD |
686 | tag_clear(node, tag, offset); |
687 | if (any_tag_set(node, tag)) | |
6e954b9e | 688 | goto out; |
e2bdb933 HD |
689 | |
690 | index >>= RADIX_TREE_MAP_SHIFT; | |
691 | offset = index & RADIX_TREE_MAP_MASK; | |
692 | node = node->parent; | |
612d6c19 NP |
693 | } |
694 | ||
695 | /* clear the root's tag bit */ | |
696 | if (root_tag_get(root, tag)) | |
697 | root_tag_clear(root, tag); | |
698 | ||
1da177e4 | 699 | out: |
612d6c19 | 700 | return slot; |
1da177e4 LT |
701 | } |
702 | EXPORT_SYMBOL(radix_tree_tag_clear); | |
703 | ||
1da177e4 | 704 | /** |
32605a18 MT |
705 | * radix_tree_tag_get - get a tag on a radix tree node |
706 | * @root: radix tree root | |
707 | * @index: index key | |
daff89f3 | 708 | * @tag: tag index (< RADIX_TREE_MAX_TAGS) |
1da177e4 | 709 | * |
32605a18 | 710 | * Return values: |
1da177e4 | 711 | * |
612d6c19 NP |
712 | * 0: tag not present or not set |
713 | * 1: tag set | |
ce82653d DH |
714 | * |
715 | * Note that the return value of this function may not be relied on, even if | |
716 | * the RCU lock is held, unless tag modification and node deletion are excluded | |
717 | * from concurrency. | |
1da177e4 LT |
718 | */ |
719 | int radix_tree_tag_get(struct radix_tree_root *root, | |
daff89f3 | 720 | unsigned long index, unsigned int tag) |
1da177e4 LT |
721 | { |
722 | unsigned int height, shift; | |
7cf9c2c7 | 723 | struct radix_tree_node *node; |
1da177e4 | 724 | |
612d6c19 NP |
725 | /* check the root's tag bit */ |
726 | if (!root_tag_get(root, tag)) | |
727 | return 0; | |
728 | ||
2676a58c | 729 | node = rcu_dereference_raw(root->rnode); |
7cf9c2c7 NP |
730 | if (node == NULL) |
731 | return 0; | |
732 | ||
c0bc9875 | 733 | if (!radix_tree_is_indirect_ptr(node)) |
7cf9c2c7 | 734 | return (index == 0); |
27d20fdd | 735 | node = indirect_to_ptr(node); |
7cf9c2c7 | 736 | |
449dd698 | 737 | height = node->path & RADIX_TREE_HEIGHT_MASK; |
7cf9c2c7 NP |
738 | if (index > radix_tree_maxindex(height)) |
739 | return 0; | |
612d6c19 | 740 | |
1da177e4 | 741 | shift = (height - 1) * RADIX_TREE_MAP_SHIFT; |
1da177e4 LT |
742 | |
743 | for ( ; ; ) { | |
744 | int offset; | |
745 | ||
7cf9c2c7 | 746 | if (node == NULL) |
1da177e4 | 747 | return 0; |
339e6353 | 748 | node = indirect_to_ptr(node); |
1da177e4 LT |
749 | |
750 | offset = (index >> shift) & RADIX_TREE_MAP_MASK; | |
7cf9c2c7 | 751 | if (!tag_get(node, tag, offset)) |
3fa36acb | 752 | return 0; |
ce82653d | 753 | if (height == 1) |
3fa36acb | 754 | return 1; |
2676a58c | 755 | node = rcu_dereference_raw(node->slots[offset]); |
1da177e4 LT |
756 | shift -= RADIX_TREE_MAP_SHIFT; |
757 | height--; | |
758 | } | |
759 | } | |
760 | EXPORT_SYMBOL(radix_tree_tag_get); | |
1da177e4 | 761 | |
78c1d784 KK |
762 | /** |
763 | * radix_tree_next_chunk - find next chunk of slots for iteration | |
764 | * | |
765 | * @root: radix tree root | |
766 | * @iter: iterator state | |
767 | * @flags: RADIX_TREE_ITER_* flags and tag index | |
768 | * Returns: pointer to chunk first slot, or NULL if iteration is over | |
769 | */ | |
770 | void **radix_tree_next_chunk(struct radix_tree_root *root, | |
771 | struct radix_tree_iter *iter, unsigned flags) | |
772 | { | |
773 | unsigned shift, tag = flags & RADIX_TREE_ITER_TAG_MASK; | |
774 | struct radix_tree_node *rnode, *node; | |
449dd698 | 775 | unsigned long index, offset, height; |
78c1d784 KK |
776 | |
777 | if ((flags & RADIX_TREE_ITER_TAGGED) && !root_tag_get(root, tag)) | |
778 | return NULL; | |
779 | ||
780 | /* | |
781 | * Catch next_index overflow after ~0UL. iter->index never overflows | |
782 | * during iterating; it can be zero only at the beginning. | |
783 | * And we cannot overflow iter->next_index in a single step, | |
784 | * because RADIX_TREE_MAP_SHIFT < BITS_PER_LONG. | |
fffaee36 KK |
785 | * |
786 | * This condition also used by radix_tree_next_slot() to stop | |
787 | * contiguous iterating, and forbid swithing to the next chunk. | |
78c1d784 KK |
788 | */ |
789 | index = iter->next_index; | |
790 | if (!index && iter->index) | |
791 | return NULL; | |
792 | ||
793 | rnode = rcu_dereference_raw(root->rnode); | |
794 | if (radix_tree_is_indirect_ptr(rnode)) { | |
795 | rnode = indirect_to_ptr(rnode); | |
796 | } else if (rnode && !index) { | |
797 | /* Single-slot tree */ | |
798 | iter->index = 0; | |
799 | iter->next_index = 1; | |
800 | iter->tags = 1; | |
801 | return (void **)&root->rnode; | |
802 | } else | |
803 | return NULL; | |
804 | ||
805 | restart: | |
449dd698 JW |
806 | height = rnode->path & RADIX_TREE_HEIGHT_MASK; |
807 | shift = (height - 1) * RADIX_TREE_MAP_SHIFT; | |
78c1d784 KK |
808 | offset = index >> shift; |
809 | ||
810 | /* Index outside of the tree */ | |
811 | if (offset >= RADIX_TREE_MAP_SIZE) | |
812 | return NULL; | |
813 | ||
814 | node = rnode; | |
815 | while (1) { | |
816 | if ((flags & RADIX_TREE_ITER_TAGGED) ? | |
817 | !test_bit(offset, node->tags[tag]) : | |
818 | !node->slots[offset]) { | |
819 | /* Hole detected */ | |
820 | if (flags & RADIX_TREE_ITER_CONTIG) | |
821 | return NULL; | |
822 | ||
823 | if (flags & RADIX_TREE_ITER_TAGGED) | |
824 | offset = radix_tree_find_next_bit( | |
825 | node->tags[tag], | |
826 | RADIX_TREE_MAP_SIZE, | |
827 | offset + 1); | |
828 | else | |
829 | while (++offset < RADIX_TREE_MAP_SIZE) { | |
830 | if (node->slots[offset]) | |
831 | break; | |
832 | } | |
833 | index &= ~((RADIX_TREE_MAP_SIZE << shift) - 1); | |
834 | index += offset << shift; | |
835 | /* Overflow after ~0UL */ | |
836 | if (!index) | |
837 | return NULL; | |
838 | if (offset == RADIX_TREE_MAP_SIZE) | |
839 | goto restart; | |
840 | } | |
841 | ||
842 | /* This is leaf-node */ | |
843 | if (!shift) | |
844 | break; | |
845 | ||
846 | node = rcu_dereference_raw(node->slots[offset]); | |
847 | if (node == NULL) | |
848 | goto restart; | |
339e6353 | 849 | node = indirect_to_ptr(node); |
78c1d784 KK |
850 | shift -= RADIX_TREE_MAP_SHIFT; |
851 | offset = (index >> shift) & RADIX_TREE_MAP_MASK; | |
852 | } | |
853 | ||
854 | /* Update the iterator state */ | |
855 | iter->index = index; | |
856 | iter->next_index = (index | RADIX_TREE_MAP_MASK) + 1; | |
857 | ||
858 | /* Construct iter->tags bit-mask from node->tags[tag] array */ | |
859 | if (flags & RADIX_TREE_ITER_TAGGED) { | |
860 | unsigned tag_long, tag_bit; | |
861 | ||
862 | tag_long = offset / BITS_PER_LONG; | |
863 | tag_bit = offset % BITS_PER_LONG; | |
864 | iter->tags = node->tags[tag][tag_long] >> tag_bit; | |
865 | /* This never happens if RADIX_TREE_TAG_LONGS == 1 */ | |
866 | if (tag_long < RADIX_TREE_TAG_LONGS - 1) { | |
867 | /* Pick tags from next element */ | |
868 | if (tag_bit) | |
869 | iter->tags |= node->tags[tag][tag_long + 1] << | |
870 | (BITS_PER_LONG - tag_bit); | |
871 | /* Clip chunk size, here only BITS_PER_LONG tags */ | |
872 | iter->next_index = index + BITS_PER_LONG; | |
873 | } | |
874 | } | |
875 | ||
876 | return node->slots + offset; | |
877 | } | |
878 | EXPORT_SYMBOL(radix_tree_next_chunk); | |
879 | ||
ebf8aa44 JK |
880 | /** |
881 | * radix_tree_range_tag_if_tagged - for each item in given range set given | |
882 | * tag if item has another tag set | |
883 | * @root: radix tree root | |
884 | * @first_indexp: pointer to a starting index of a range to scan | |
885 | * @last_index: last index of a range to scan | |
886 | * @nr_to_tag: maximum number items to tag | |
887 | * @iftag: tag index to test | |
888 | * @settag: tag index to set if tested tag is set | |
889 | * | |
890 | * This function scans range of radix tree from first_index to last_index | |
891 | * (inclusive). For each item in the range if iftag is set, the function sets | |
892 | * also settag. The function stops either after tagging nr_to_tag items or | |
893 | * after reaching last_index. | |
894 | * | |
144dcfc0 DC |
895 | * The tags must be set from the leaf level only and propagated back up the |
896 | * path to the root. We must do this so that we resolve the full path before | |
897 | * setting any tags on intermediate nodes. If we set tags as we descend, then | |
898 | * we can get to the leaf node and find that the index that has the iftag | |
899 | * set is outside the range we are scanning. This reults in dangling tags and | |
900 | * can lead to problems with later tag operations (e.g. livelocks on lookups). | |
901 | * | |
ebf8aa44 JK |
902 | * The function returns number of leaves where the tag was set and sets |
903 | * *first_indexp to the first unscanned index. | |
d5ed3a4a JK |
904 | * WARNING! *first_indexp can wrap if last_index is ULONG_MAX. Caller must |
905 | * be prepared to handle that. | |
ebf8aa44 JK |
906 | */ |
907 | unsigned long radix_tree_range_tag_if_tagged(struct radix_tree_root *root, | |
908 | unsigned long *first_indexp, unsigned long last_index, | |
909 | unsigned long nr_to_tag, | |
910 | unsigned int iftag, unsigned int settag) | |
911 | { | |
144dcfc0 | 912 | unsigned int height = root->height; |
e2bdb933 | 913 | struct radix_tree_node *node = NULL; |
144dcfc0 DC |
914 | struct radix_tree_node *slot; |
915 | unsigned int shift; | |
916 | unsigned long tagged = 0; | |
917 | unsigned long index = *first_indexp; | |
ebf8aa44 JK |
918 | |
919 | last_index = min(last_index, radix_tree_maxindex(height)); | |
920 | if (index > last_index) | |
921 | return 0; | |
922 | if (!nr_to_tag) | |
923 | return 0; | |
924 | if (!root_tag_get(root, iftag)) { | |
925 | *first_indexp = last_index + 1; | |
926 | return 0; | |
927 | } | |
928 | if (height == 0) { | |
929 | *first_indexp = last_index + 1; | |
930 | root_tag_set(root, settag); | |
931 | return 1; | |
932 | } | |
933 | ||
934 | shift = (height - 1) * RADIX_TREE_MAP_SHIFT; | |
27d20fdd | 935 | slot = indirect_to_ptr(root->rnode); |
ebf8aa44 JK |
936 | |
937 | for (;;) { | |
e2bdb933 | 938 | unsigned long upindex; |
ebf8aa44 JK |
939 | int offset; |
940 | ||
941 | offset = (index >> shift) & RADIX_TREE_MAP_MASK; | |
942 | if (!slot->slots[offset]) | |
943 | goto next; | |
944 | if (!tag_get(slot, iftag, offset)) | |
945 | goto next; | |
e2bdb933 | 946 | if (shift) { |
144dcfc0 | 947 | /* Go down one level */ |
144dcfc0 | 948 | shift -= RADIX_TREE_MAP_SHIFT; |
e2bdb933 | 949 | node = slot; |
144dcfc0 | 950 | slot = slot->slots[offset]; |
339e6353 | 951 | slot = indirect_to_ptr(slot); |
144dcfc0 DC |
952 | continue; |
953 | } | |
954 | ||
955 | /* tag the leaf */ | |
956 | tagged++; | |
ebf8aa44 | 957 | tag_set(slot, settag, offset); |
144dcfc0 DC |
958 | |
959 | /* walk back up the path tagging interior nodes */ | |
e2bdb933 HD |
960 | upindex = index; |
961 | while (node) { | |
962 | upindex >>= RADIX_TREE_MAP_SHIFT; | |
963 | offset = upindex & RADIX_TREE_MAP_MASK; | |
964 | ||
144dcfc0 | 965 | /* stop if we find a node with the tag already set */ |
e2bdb933 | 966 | if (tag_get(node, settag, offset)) |
144dcfc0 | 967 | break; |
e2bdb933 HD |
968 | tag_set(node, settag, offset); |
969 | node = node->parent; | |
ebf8aa44 | 970 | } |
144dcfc0 | 971 | |
e2bdb933 HD |
972 | /* |
973 | * Small optimization: now clear that node pointer. | |
974 | * Since all of this slot's ancestors now have the tag set | |
975 | * from setting it above, we have no further need to walk | |
976 | * back up the tree setting tags, until we update slot to | |
977 | * point to another radix_tree_node. | |
978 | */ | |
979 | node = NULL; | |
980 | ||
ebf8aa44 JK |
981 | next: |
982 | /* Go to next item at level determined by 'shift' */ | |
983 | index = ((index >> shift) + 1) << shift; | |
d5ed3a4a JK |
984 | /* Overflow can happen when last_index is ~0UL... */ |
985 | if (index > last_index || !index) | |
ebf8aa44 JK |
986 | break; |
987 | if (tagged >= nr_to_tag) | |
988 | break; | |
989 | while (((index >> shift) & RADIX_TREE_MAP_MASK) == 0) { | |
990 | /* | |
991 | * We've fully scanned this node. Go up. Because | |
992 | * last_index is guaranteed to be in the tree, what | |
993 | * we do below cannot wander astray. | |
994 | */ | |
e2bdb933 | 995 | slot = slot->parent; |
ebf8aa44 JK |
996 | shift += RADIX_TREE_MAP_SHIFT; |
997 | } | |
998 | } | |
999 | /* | |
ac15ee69 TO |
1000 | * We need not to tag the root tag if there is no tag which is set with |
1001 | * settag within the range from *first_indexp to last_index. | |
ebf8aa44 | 1002 | */ |
ac15ee69 TO |
1003 | if (tagged > 0) |
1004 | root_tag_set(root, settag); | |
ebf8aa44 JK |
1005 | *first_indexp = index; |
1006 | ||
1007 | return tagged; | |
1008 | } | |
1009 | EXPORT_SYMBOL(radix_tree_range_tag_if_tagged); | |
1010 | ||
1da177e4 LT |
1011 | /** |
1012 | * radix_tree_gang_lookup - perform multiple lookup on a radix tree | |
1013 | * @root: radix tree root | |
1014 | * @results: where the results of the lookup are placed | |
1015 | * @first_index: start the lookup from this key | |
1016 | * @max_items: place up to this many items at *results | |
1017 | * | |
1018 | * Performs an index-ascending scan of the tree for present items. Places | |
1019 | * them at *@results and returns the number of items which were placed at | |
1020 | * *@results. | |
1021 | * | |
1022 | * The implementation is naive. | |
7cf9c2c7 NP |
1023 | * |
1024 | * Like radix_tree_lookup, radix_tree_gang_lookup may be called under | |
1025 | * rcu_read_lock. In this case, rather than the returned results being | |
1026 | * an atomic snapshot of the tree at a single point in time, the semantics | |
1027 | * of an RCU protected gang lookup are as though multiple radix_tree_lookups | |
1028 | * have been issued in individual locks, and results stored in 'results'. | |
1da177e4 LT |
1029 | */ |
1030 | unsigned int | |
1031 | radix_tree_gang_lookup(struct radix_tree_root *root, void **results, | |
1032 | unsigned long first_index, unsigned int max_items) | |
1033 | { | |
cebbd29e KK |
1034 | struct radix_tree_iter iter; |
1035 | void **slot; | |
1036 | unsigned int ret = 0; | |
7cf9c2c7 | 1037 | |
cebbd29e | 1038 | if (unlikely(!max_items)) |
7cf9c2c7 | 1039 | return 0; |
1da177e4 | 1040 | |
cebbd29e | 1041 | radix_tree_for_each_slot(slot, root, &iter, first_index) { |
46437f9a | 1042 | results[ret] = rcu_dereference_raw(*slot); |
cebbd29e KK |
1043 | if (!results[ret]) |
1044 | continue; | |
46437f9a MW |
1045 | if (radix_tree_is_indirect_ptr(results[ret])) { |
1046 | slot = radix_tree_iter_retry(&iter); | |
1047 | continue; | |
1048 | } | |
cebbd29e | 1049 | if (++ret == max_items) |
1da177e4 | 1050 | break; |
1da177e4 | 1051 | } |
7cf9c2c7 | 1052 | |
1da177e4 LT |
1053 | return ret; |
1054 | } | |
1055 | EXPORT_SYMBOL(radix_tree_gang_lookup); | |
1056 | ||
47feff2c NP |
1057 | /** |
1058 | * radix_tree_gang_lookup_slot - perform multiple slot lookup on radix tree | |
1059 | * @root: radix tree root | |
1060 | * @results: where the results of the lookup are placed | |
6328650b | 1061 | * @indices: where their indices should be placed (but usually NULL) |
47feff2c NP |
1062 | * @first_index: start the lookup from this key |
1063 | * @max_items: place up to this many items at *results | |
1064 | * | |
1065 | * Performs an index-ascending scan of the tree for present items. Places | |
1066 | * their slots at *@results and returns the number of items which were | |
1067 | * placed at *@results. | |
1068 | * | |
1069 | * The implementation is naive. | |
1070 | * | |
1071 | * Like radix_tree_gang_lookup as far as RCU and locking goes. Slots must | |
1072 | * be dereferenced with radix_tree_deref_slot, and if using only RCU | |
1073 | * protection, radix_tree_deref_slot may fail requiring a retry. | |
1074 | */ | |
1075 | unsigned int | |
6328650b HD |
1076 | radix_tree_gang_lookup_slot(struct radix_tree_root *root, |
1077 | void ***results, unsigned long *indices, | |
47feff2c NP |
1078 | unsigned long first_index, unsigned int max_items) |
1079 | { | |
cebbd29e KK |
1080 | struct radix_tree_iter iter; |
1081 | void **slot; | |
1082 | unsigned int ret = 0; | |
47feff2c | 1083 | |
cebbd29e | 1084 | if (unlikely(!max_items)) |
47feff2c NP |
1085 | return 0; |
1086 | ||
cebbd29e KK |
1087 | radix_tree_for_each_slot(slot, root, &iter, first_index) { |
1088 | results[ret] = slot; | |
6328650b | 1089 | if (indices) |
cebbd29e KK |
1090 | indices[ret] = iter.index; |
1091 | if (++ret == max_items) | |
47feff2c | 1092 | break; |
47feff2c NP |
1093 | } |
1094 | ||
1095 | return ret; | |
1096 | } | |
1097 | EXPORT_SYMBOL(radix_tree_gang_lookup_slot); | |
1098 | ||
1da177e4 LT |
1099 | /** |
1100 | * radix_tree_gang_lookup_tag - perform multiple lookup on a radix tree | |
1101 | * based on a tag | |
1102 | * @root: radix tree root | |
1103 | * @results: where the results of the lookup are placed | |
1104 | * @first_index: start the lookup from this key | |
1105 | * @max_items: place up to this many items at *results | |
daff89f3 | 1106 | * @tag: the tag index (< RADIX_TREE_MAX_TAGS) |
1da177e4 LT |
1107 | * |
1108 | * Performs an index-ascending scan of the tree for present items which | |
1109 | * have the tag indexed by @tag set. Places the items at *@results and | |
1110 | * returns the number of items which were placed at *@results. | |
1111 | */ | |
1112 | unsigned int | |
1113 | radix_tree_gang_lookup_tag(struct radix_tree_root *root, void **results, | |
daff89f3 JC |
1114 | unsigned long first_index, unsigned int max_items, |
1115 | unsigned int tag) | |
1da177e4 | 1116 | { |
cebbd29e KK |
1117 | struct radix_tree_iter iter; |
1118 | void **slot; | |
1119 | unsigned int ret = 0; | |
612d6c19 | 1120 | |
cebbd29e | 1121 | if (unlikely(!max_items)) |
7cf9c2c7 NP |
1122 | return 0; |
1123 | ||
cebbd29e | 1124 | radix_tree_for_each_tagged(slot, root, &iter, first_index, tag) { |
46437f9a | 1125 | results[ret] = rcu_dereference_raw(*slot); |
cebbd29e KK |
1126 | if (!results[ret]) |
1127 | continue; | |
46437f9a MW |
1128 | if (radix_tree_is_indirect_ptr(results[ret])) { |
1129 | slot = radix_tree_iter_retry(&iter); | |
1130 | continue; | |
1131 | } | |
cebbd29e | 1132 | if (++ret == max_items) |
1da177e4 | 1133 | break; |
1da177e4 | 1134 | } |
7cf9c2c7 | 1135 | |
1da177e4 LT |
1136 | return ret; |
1137 | } | |
1138 | EXPORT_SYMBOL(radix_tree_gang_lookup_tag); | |
1139 | ||
47feff2c NP |
1140 | /** |
1141 | * radix_tree_gang_lookup_tag_slot - perform multiple slot lookup on a | |
1142 | * radix tree based on a tag | |
1143 | * @root: radix tree root | |
1144 | * @results: where the results of the lookup are placed | |
1145 | * @first_index: start the lookup from this key | |
1146 | * @max_items: place up to this many items at *results | |
1147 | * @tag: the tag index (< RADIX_TREE_MAX_TAGS) | |
1148 | * | |
1149 | * Performs an index-ascending scan of the tree for present items which | |
1150 | * have the tag indexed by @tag set. Places the slots at *@results and | |
1151 | * returns the number of slots which were placed at *@results. | |
1152 | */ | |
1153 | unsigned int | |
1154 | radix_tree_gang_lookup_tag_slot(struct radix_tree_root *root, void ***results, | |
1155 | unsigned long first_index, unsigned int max_items, | |
1156 | unsigned int tag) | |
1157 | { | |
cebbd29e KK |
1158 | struct radix_tree_iter iter; |
1159 | void **slot; | |
1160 | unsigned int ret = 0; | |
47feff2c | 1161 | |
cebbd29e | 1162 | if (unlikely(!max_items)) |
47feff2c NP |
1163 | return 0; |
1164 | ||
cebbd29e KK |
1165 | radix_tree_for_each_tagged(slot, root, &iter, first_index, tag) { |
1166 | results[ret] = slot; | |
1167 | if (++ret == max_items) | |
47feff2c | 1168 | break; |
47feff2c NP |
1169 | } |
1170 | ||
1171 | return ret; | |
1172 | } | |
1173 | EXPORT_SYMBOL(radix_tree_gang_lookup_tag_slot); | |
1174 | ||
e504f3fd HD |
1175 | #if defined(CONFIG_SHMEM) && defined(CONFIG_SWAP) |
1176 | #include <linux/sched.h> /* for cond_resched() */ | |
1177 | ||
1178 | /* | |
1179 | * This linear search is at present only useful to shmem_unuse_inode(). | |
1180 | */ | |
1181 | static unsigned long __locate(struct radix_tree_node *slot, void *item, | |
1182 | unsigned long index, unsigned long *found_index) | |
1183 | { | |
1184 | unsigned int shift, height; | |
1185 | unsigned long i; | |
1186 | ||
449dd698 | 1187 | height = slot->path & RADIX_TREE_HEIGHT_MASK; |
e504f3fd HD |
1188 | shift = (height-1) * RADIX_TREE_MAP_SHIFT; |
1189 | ||
1190 | for ( ; height > 1; height--) { | |
1191 | i = (index >> shift) & RADIX_TREE_MAP_MASK; | |
1192 | for (;;) { | |
1193 | if (slot->slots[i] != NULL) | |
1194 | break; | |
1195 | index &= ~((1UL << shift) - 1); | |
1196 | index += 1UL << shift; | |
1197 | if (index == 0) | |
1198 | goto out; /* 32-bit wraparound */ | |
1199 | i++; | |
1200 | if (i == RADIX_TREE_MAP_SIZE) | |
1201 | goto out; | |
1202 | } | |
1203 | ||
1204 | shift -= RADIX_TREE_MAP_SHIFT; | |
1205 | slot = rcu_dereference_raw(slot->slots[i]); | |
1206 | if (slot == NULL) | |
1207 | goto out; | |
339e6353 | 1208 | slot = indirect_to_ptr(slot); |
e504f3fd HD |
1209 | } |
1210 | ||
1211 | /* Bottom level: check items */ | |
1212 | for (i = 0; i < RADIX_TREE_MAP_SIZE; i++) { | |
1213 | if (slot->slots[i] == item) { | |
1214 | *found_index = index + i; | |
1215 | index = 0; | |
1216 | goto out; | |
1217 | } | |
1218 | } | |
1219 | index += RADIX_TREE_MAP_SIZE; | |
1220 | out: | |
1221 | return index; | |
1222 | } | |
1223 | ||
1224 | /** | |
1225 | * radix_tree_locate_item - search through radix tree for item | |
1226 | * @root: radix tree root | |
1227 | * @item: item to be found | |
1228 | * | |
1229 | * Returns index where item was found, or -1 if not found. | |
1230 | * Caller must hold no lock (since this time-consuming function needs | |
1231 | * to be preemptible), and must check afterwards if item is still there. | |
1232 | */ | |
1233 | unsigned long radix_tree_locate_item(struct radix_tree_root *root, void *item) | |
1234 | { | |
1235 | struct radix_tree_node *node; | |
1236 | unsigned long max_index; | |
1237 | unsigned long cur_index = 0; | |
1238 | unsigned long found_index = -1; | |
1239 | ||
1240 | do { | |
1241 | rcu_read_lock(); | |
1242 | node = rcu_dereference_raw(root->rnode); | |
1243 | if (!radix_tree_is_indirect_ptr(node)) { | |
1244 | rcu_read_unlock(); | |
1245 | if (node == item) | |
1246 | found_index = 0; | |
1247 | break; | |
1248 | } | |
1249 | ||
1250 | node = indirect_to_ptr(node); | |
449dd698 JW |
1251 | max_index = radix_tree_maxindex(node->path & |
1252 | RADIX_TREE_HEIGHT_MASK); | |
5f30fc94 HD |
1253 | if (cur_index > max_index) { |
1254 | rcu_read_unlock(); | |
e504f3fd | 1255 | break; |
5f30fc94 | 1256 | } |
e504f3fd HD |
1257 | |
1258 | cur_index = __locate(node, item, cur_index, &found_index); | |
1259 | rcu_read_unlock(); | |
1260 | cond_resched(); | |
1261 | } while (cur_index != 0 && cur_index <= max_index); | |
1262 | ||
1263 | return found_index; | |
1264 | } | |
1265 | #else | |
1266 | unsigned long radix_tree_locate_item(struct radix_tree_root *root, void *item) | |
1267 | { | |
1268 | return -1; | |
1269 | } | |
1270 | #endif /* CONFIG_SHMEM && CONFIG_SWAP */ | |
47feff2c | 1271 | |
a5f51c96 NP |
1272 | /** |
1273 | * radix_tree_shrink - shrink height of a radix tree to minimal | |
1274 | * @root radix tree root | |
1275 | */ | |
1276 | static inline void radix_tree_shrink(struct radix_tree_root *root) | |
1277 | { | |
1278 | /* try to shrink tree height */ | |
c0bc9875 | 1279 | while (root->height > 0) { |
a5f51c96 | 1280 | struct radix_tree_node *to_free = root->rnode; |
e2bdb933 | 1281 | struct radix_tree_node *slot; |
a5f51c96 | 1282 | |
c0bc9875 | 1283 | BUG_ON(!radix_tree_is_indirect_ptr(to_free)); |
27d20fdd | 1284 | to_free = indirect_to_ptr(to_free); |
c0bc9875 NP |
1285 | |
1286 | /* | |
1287 | * The candidate node has more than one child, or its child | |
1288 | * is not at the leftmost slot, we cannot shrink. | |
1289 | */ | |
1290 | if (to_free->count != 1) | |
1291 | break; | |
339e6353 MW |
1292 | slot = to_free->slots[0]; |
1293 | if (!slot) | |
c0bc9875 NP |
1294 | break; |
1295 | ||
7cf9c2c7 NP |
1296 | /* |
1297 | * We don't need rcu_assign_pointer(), since we are simply | |
27d20fdd NP |
1298 | * moving the node from one part of the tree to another: if it |
1299 | * was safe to dereference the old pointer to it | |
7cf9c2c7 | 1300 | * (to_free->slots[0]), it will be safe to dereference the new |
27d20fdd | 1301 | * one (root->rnode) as far as dependent read barriers go. |
7cf9c2c7 | 1302 | */ |
e2bdb933 | 1303 | if (root->height > 1) { |
339e6353 | 1304 | slot = indirect_to_ptr(slot); |
e2bdb933 HD |
1305 | slot->parent = NULL; |
1306 | slot = ptr_to_indirect(slot); | |
1307 | } | |
1308 | root->rnode = slot; | |
a5f51c96 | 1309 | root->height--; |
27d20fdd NP |
1310 | |
1311 | /* | |
1312 | * We have a dilemma here. The node's slot[0] must not be | |
1313 | * NULLed in case there are concurrent lookups expecting to | |
1314 | * find the item. However if this was a bottom-level node, | |
1315 | * then it may be subject to the slot pointer being visible | |
1316 | * to callers dereferencing it. If item corresponding to | |
1317 | * slot[0] is subsequently deleted, these callers would expect | |
1318 | * their slot to become empty sooner or later. | |
1319 | * | |
1320 | * For example, lockless pagecache will look up a slot, deref | |
1321 | * the page pointer, and if the page is 0 refcount it means it | |
1322 | * was concurrently deleted from pagecache so try the deref | |
1323 | * again. Fortunately there is already a requirement for logic | |
1324 | * to retry the entire slot lookup -- the indirect pointer | |
1325 | * problem (replacing direct root node with an indirect pointer | |
1326 | * also results in a stale slot). So tag the slot as indirect | |
1327 | * to force callers to retry. | |
1328 | */ | |
1329 | if (root->height == 0) | |
1330 | *((unsigned long *)&to_free->slots[0]) |= | |
1331 | RADIX_TREE_INDIRECT_PTR; | |
1332 | ||
a5f51c96 NP |
1333 | radix_tree_node_free(to_free); |
1334 | } | |
1335 | } | |
1336 | ||
139e5616 JW |
1337 | /** |
1338 | * __radix_tree_delete_node - try to free node after clearing a slot | |
1339 | * @root: radix tree root | |
139e5616 JW |
1340 | * @node: node containing @index |
1341 | * | |
1342 | * After clearing the slot at @index in @node from radix tree | |
1343 | * rooted at @root, call this function to attempt freeing the | |
1344 | * node and shrinking the tree. | |
1345 | * | |
1346 | * Returns %true if @node was freed, %false otherwise. | |
1347 | */ | |
449dd698 | 1348 | bool __radix_tree_delete_node(struct radix_tree_root *root, |
139e5616 JW |
1349 | struct radix_tree_node *node) |
1350 | { | |
1351 | bool deleted = false; | |
1352 | ||
1353 | do { | |
1354 | struct radix_tree_node *parent; | |
1355 | ||
1356 | if (node->count) { | |
1357 | if (node == indirect_to_ptr(root->rnode)) { | |
1358 | radix_tree_shrink(root); | |
1359 | if (root->height == 0) | |
1360 | deleted = true; | |
1361 | } | |
1362 | return deleted; | |
1363 | } | |
1364 | ||
1365 | parent = node->parent; | |
1366 | if (parent) { | |
449dd698 | 1367 | unsigned int offset; |
139e5616 | 1368 | |
449dd698 JW |
1369 | offset = node->path >> RADIX_TREE_HEIGHT_SHIFT; |
1370 | parent->slots[offset] = NULL; | |
139e5616 JW |
1371 | parent->count--; |
1372 | } else { | |
1373 | root_tag_clear_all(root); | |
1374 | root->height = 0; | |
1375 | root->rnode = NULL; | |
1376 | } | |
1377 | ||
1378 | radix_tree_node_free(node); | |
1379 | deleted = true; | |
1380 | ||
1381 | node = parent; | |
1382 | } while (node); | |
1383 | ||
1384 | return deleted; | |
1385 | } | |
1386 | ||
1da177e4 | 1387 | /** |
53c59f26 | 1388 | * radix_tree_delete_item - delete an item from a radix tree |
1da177e4 LT |
1389 | * @root: radix tree root |
1390 | * @index: index key | |
53c59f26 | 1391 | * @item: expected item |
1da177e4 | 1392 | * |
53c59f26 | 1393 | * Remove @item at @index from the radix tree rooted at @root. |
1da177e4 | 1394 | * |
53c59f26 JW |
1395 | * Returns the address of the deleted item, or NULL if it was not present |
1396 | * or the entry at the given @index was not @item. | |
1da177e4 | 1397 | */ |
53c59f26 JW |
1398 | void *radix_tree_delete_item(struct radix_tree_root *root, |
1399 | unsigned long index, void *item) | |
1da177e4 | 1400 | { |
139e5616 JW |
1401 | struct radix_tree_node *node; |
1402 | unsigned int offset; | |
1403 | void **slot; | |
1404 | void *entry; | |
d5274261 | 1405 | int tag; |
1da177e4 | 1406 | |
139e5616 JW |
1407 | entry = __radix_tree_lookup(root, index, &node, &slot); |
1408 | if (!entry) | |
1409 | return NULL; | |
1da177e4 | 1410 | |
139e5616 JW |
1411 | if (item && entry != item) |
1412 | return NULL; | |
1413 | ||
1414 | if (!node) { | |
612d6c19 NP |
1415 | root_tag_clear_all(root); |
1416 | root->rnode = NULL; | |
139e5616 | 1417 | return entry; |
612d6c19 | 1418 | } |
1da177e4 | 1419 | |
139e5616 | 1420 | offset = index & RADIX_TREE_MAP_MASK; |
53c59f26 | 1421 | |
1da177e4 | 1422 | /* |
e2bdb933 HD |
1423 | * Clear all tags associated with the item to be deleted. |
1424 | * This way of doing it would be inefficient, but seldom is any set. | |
1da177e4 | 1425 | */ |
daff89f3 | 1426 | for (tag = 0; tag < RADIX_TREE_MAX_TAGS; tag++) { |
e2bdb933 | 1427 | if (tag_get(node, tag, offset)) |
612d6c19 | 1428 | radix_tree_tag_clear(root, index, tag); |
d5274261 | 1429 | } |
1da177e4 | 1430 | |
139e5616 JW |
1431 | node->slots[offset] = NULL; |
1432 | node->count--; | |
e2bdb933 | 1433 | |
449dd698 | 1434 | __radix_tree_delete_node(root, node); |
612d6c19 | 1435 | |
139e5616 | 1436 | return entry; |
1da177e4 | 1437 | } |
53c59f26 JW |
1438 | EXPORT_SYMBOL(radix_tree_delete_item); |
1439 | ||
1440 | /** | |
1441 | * radix_tree_delete - delete an item from a radix tree | |
1442 | * @root: radix tree root | |
1443 | * @index: index key | |
1444 | * | |
1445 | * Remove the item at @index from the radix tree rooted at @root. | |
1446 | * | |
1447 | * Returns the address of the deleted item, or NULL if it was not present. | |
1448 | */ | |
1449 | void *radix_tree_delete(struct radix_tree_root *root, unsigned long index) | |
1450 | { | |
1451 | return radix_tree_delete_item(root, index, NULL); | |
1452 | } | |
1da177e4 LT |
1453 | EXPORT_SYMBOL(radix_tree_delete); |
1454 | ||
1455 | /** | |
1456 | * radix_tree_tagged - test whether any items in the tree are tagged | |
1457 | * @root: radix tree root | |
1458 | * @tag: tag to test | |
1459 | */ | |
daff89f3 | 1460 | int radix_tree_tagged(struct radix_tree_root *root, unsigned int tag) |
1da177e4 | 1461 | { |
612d6c19 | 1462 | return root_tag_get(root, tag); |
1da177e4 LT |
1463 | } |
1464 | EXPORT_SYMBOL(radix_tree_tagged); | |
1465 | ||
1466 | static void | |
449dd698 | 1467 | radix_tree_node_ctor(void *arg) |
1da177e4 | 1468 | { |
449dd698 JW |
1469 | struct radix_tree_node *node = arg; |
1470 | ||
1471 | memset(node, 0, sizeof(*node)); | |
1472 | INIT_LIST_HEAD(&node->private_list); | |
1da177e4 LT |
1473 | } |
1474 | ||
1475 | static __init unsigned long __maxindex(unsigned int height) | |
1476 | { | |
430d275a PL |
1477 | unsigned int width = height * RADIX_TREE_MAP_SHIFT; |
1478 | int shift = RADIX_TREE_INDEX_BITS - width; | |
1479 | ||
1480 | if (shift < 0) | |
1481 | return ~0UL; | |
1482 | if (shift >= BITS_PER_LONG) | |
1483 | return 0UL; | |
1484 | return ~0UL >> shift; | |
1da177e4 LT |
1485 | } |
1486 | ||
1487 | static __init void radix_tree_init_maxindex(void) | |
1488 | { | |
1489 | unsigned int i; | |
1490 | ||
1491 | for (i = 0; i < ARRAY_SIZE(height_to_maxindex); i++) | |
1492 | height_to_maxindex[i] = __maxindex(i); | |
1493 | } | |
1494 | ||
1da177e4 LT |
1495 | static int radix_tree_callback(struct notifier_block *nfb, |
1496 | unsigned long action, | |
1497 | void *hcpu) | |
1498 | { | |
1499 | int cpu = (long)hcpu; | |
1500 | struct radix_tree_preload *rtp; | |
9d2a8da0 | 1501 | struct radix_tree_node *node; |
1da177e4 LT |
1502 | |
1503 | /* Free per-cpu pool of perloaded nodes */ | |
8bb78442 | 1504 | if (action == CPU_DEAD || action == CPU_DEAD_FROZEN) { |
1da177e4 LT |
1505 | rtp = &per_cpu(radix_tree_preloads, cpu); |
1506 | while (rtp->nr) { | |
9d2a8da0 KS |
1507 | node = rtp->nodes; |
1508 | rtp->nodes = node->private_data; | |
1509 | kmem_cache_free(radix_tree_node_cachep, node); | |
1510 | rtp->nr--; | |
1da177e4 LT |
1511 | } |
1512 | } | |
1513 | return NOTIFY_OK; | |
1514 | } | |
1da177e4 LT |
1515 | |
1516 | void __init radix_tree_init(void) | |
1517 | { | |
1518 | radix_tree_node_cachep = kmem_cache_create("radix_tree_node", | |
1519 | sizeof(struct radix_tree_node), 0, | |
488514d1 CL |
1520 | SLAB_PANIC | SLAB_RECLAIM_ACCOUNT, |
1521 | radix_tree_node_ctor); | |
1da177e4 LT |
1522 | radix_tree_init_maxindex(); |
1523 | hotcpu_notifier(radix_tree_callback, 0); | |
1524 | } |