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; | |
371 | if (newheight > 1) { | |
372 | slot = indirect_to_ptr(slot); | |
373 | slot->parent = node; | |
374 | } | |
375 | node->slots[0] = slot; | |
27d20fdd | 376 | node = ptr_to_indirect(node); |
7cf9c2c7 NP |
377 | rcu_assign_pointer(root->rnode, node); |
378 | root->height = newheight; | |
1da177e4 LT |
379 | } while (height > root->height); |
380 | out: | |
381 | return 0; | |
382 | } | |
383 | ||
384 | /** | |
139e5616 | 385 | * __radix_tree_create - create a slot in a radix tree |
1da177e4 LT |
386 | * @root: radix tree root |
387 | * @index: index key | |
139e5616 JW |
388 | * @nodep: returns node |
389 | * @slotp: returns slot | |
1da177e4 | 390 | * |
139e5616 JW |
391 | * Create, if necessary, and return the node and slot for an item |
392 | * at position @index in the radix tree @root. | |
393 | * | |
394 | * Until there is more than one item in the tree, no nodes are | |
395 | * allocated and @root->rnode is used as a direct slot instead of | |
396 | * pointing to a node, in which case *@nodep will be NULL. | |
397 | * | |
398 | * Returns -ENOMEM, or 0 for success. | |
1da177e4 | 399 | */ |
139e5616 JW |
400 | int __radix_tree_create(struct radix_tree_root *root, unsigned long index, |
401 | struct radix_tree_node **nodep, void ***slotp) | |
1da177e4 | 402 | { |
201b6264 | 403 | struct radix_tree_node *node = NULL, *slot; |
139e5616 | 404 | unsigned int height, shift, offset; |
1da177e4 LT |
405 | int error; |
406 | ||
407 | /* Make sure the tree is high enough. */ | |
612d6c19 | 408 | if (index > radix_tree_maxindex(root->height)) { |
1da177e4 LT |
409 | error = radix_tree_extend(root, index); |
410 | if (error) | |
411 | return error; | |
412 | } | |
413 | ||
27d20fdd | 414 | slot = indirect_to_ptr(root->rnode); |
c0bc9875 | 415 | |
1da177e4 LT |
416 | height = root->height; |
417 | shift = (height-1) * RADIX_TREE_MAP_SHIFT; | |
418 | ||
419 | offset = 0; /* uninitialised var warning */ | |
612d6c19 | 420 | while (height > 0) { |
201b6264 | 421 | if (slot == NULL) { |
1da177e4 | 422 | /* Have to add a child node. */ |
201b6264 | 423 | if (!(slot = radix_tree_node_alloc(root))) |
1da177e4 | 424 | return -ENOMEM; |
449dd698 | 425 | slot->path = height; |
e2bdb933 | 426 | slot->parent = node; |
201b6264 | 427 | if (node) { |
7cf9c2c7 | 428 | rcu_assign_pointer(node->slots[offset], slot); |
1da177e4 | 429 | node->count++; |
449dd698 | 430 | slot->path |= offset << RADIX_TREE_HEIGHT_SHIFT; |
201b6264 | 431 | } else |
27d20fdd | 432 | rcu_assign_pointer(root->rnode, ptr_to_indirect(slot)); |
1da177e4 LT |
433 | } |
434 | ||
435 | /* Go a level down */ | |
436 | offset = (index >> shift) & RADIX_TREE_MAP_MASK; | |
201b6264 CL |
437 | node = slot; |
438 | slot = node->slots[offset]; | |
1da177e4 LT |
439 | shift -= RADIX_TREE_MAP_SHIFT; |
440 | height--; | |
612d6c19 | 441 | } |
1da177e4 | 442 | |
139e5616 JW |
443 | if (nodep) |
444 | *nodep = node; | |
445 | if (slotp) | |
446 | *slotp = node ? node->slots + offset : (void **)&root->rnode; | |
447 | return 0; | |
448 | } | |
449 | ||
450 | /** | |
451 | * radix_tree_insert - insert into a radix tree | |
452 | * @root: radix tree root | |
453 | * @index: index key | |
454 | * @item: item to insert | |
455 | * | |
456 | * Insert an item into the radix tree at position @index. | |
457 | */ | |
458 | int radix_tree_insert(struct radix_tree_root *root, | |
459 | unsigned long index, void *item) | |
460 | { | |
461 | struct radix_tree_node *node; | |
462 | void **slot; | |
463 | int error; | |
464 | ||
465 | BUG_ON(radix_tree_is_indirect_ptr(item)); | |
466 | ||
467 | error = __radix_tree_create(root, index, &node, &slot); | |
468 | if (error) | |
469 | return error; | |
470 | if (*slot != NULL) | |
1da177e4 | 471 | return -EEXIST; |
139e5616 | 472 | rcu_assign_pointer(*slot, item); |
201b6264 | 473 | |
612d6c19 NP |
474 | if (node) { |
475 | node->count++; | |
139e5616 JW |
476 | BUG_ON(tag_get(node, 0, index & RADIX_TREE_MAP_MASK)); |
477 | BUG_ON(tag_get(node, 1, index & RADIX_TREE_MAP_MASK)); | |
612d6c19 | 478 | } else { |
612d6c19 NP |
479 | BUG_ON(root_tag_get(root, 0)); |
480 | BUG_ON(root_tag_get(root, 1)); | |
481 | } | |
1da177e4 | 482 | |
1da177e4 LT |
483 | return 0; |
484 | } | |
485 | EXPORT_SYMBOL(radix_tree_insert); | |
486 | ||
139e5616 JW |
487 | /** |
488 | * __radix_tree_lookup - lookup an item in a radix tree | |
489 | * @root: radix tree root | |
490 | * @index: index key | |
491 | * @nodep: returns node | |
492 | * @slotp: returns slot | |
493 | * | |
494 | * Lookup and return the item at position @index in the radix | |
495 | * tree @root. | |
496 | * | |
497 | * Until there is more than one item in the tree, no nodes are | |
498 | * allocated and @root->rnode is used as a direct slot instead of | |
499 | * pointing to a node, in which case *@nodep will be NULL. | |
7cf9c2c7 | 500 | */ |
139e5616 JW |
501 | void *__radix_tree_lookup(struct radix_tree_root *root, unsigned long index, |
502 | struct radix_tree_node **nodep, void ***slotp) | |
1da177e4 | 503 | { |
139e5616 | 504 | struct radix_tree_node *node, *parent; |
1da177e4 | 505 | unsigned int height, shift; |
139e5616 | 506 | void **slot; |
612d6c19 | 507 | |
2676a58c | 508 | node = rcu_dereference_raw(root->rnode); |
7cf9c2c7 | 509 | if (node == NULL) |
1da177e4 LT |
510 | return NULL; |
511 | ||
c0bc9875 | 512 | if (!radix_tree_is_indirect_ptr(node)) { |
7cf9c2c7 NP |
513 | if (index > 0) |
514 | return NULL; | |
139e5616 JW |
515 | |
516 | if (nodep) | |
517 | *nodep = NULL; | |
518 | if (slotp) | |
519 | *slotp = (void **)&root->rnode; | |
520 | return node; | |
7cf9c2c7 | 521 | } |
27d20fdd | 522 | node = indirect_to_ptr(node); |
7cf9c2c7 | 523 | |
449dd698 | 524 | height = node->path & RADIX_TREE_HEIGHT_MASK; |
7cf9c2c7 NP |
525 | if (index > radix_tree_maxindex(height)) |
526 | return NULL; | |
612d6c19 | 527 | |
1da177e4 | 528 | shift = (height-1) * RADIX_TREE_MAP_SHIFT; |
1da177e4 | 529 | |
7cf9c2c7 | 530 | do { |
139e5616 JW |
531 | parent = node; |
532 | slot = node->slots + ((index >> shift) & RADIX_TREE_MAP_MASK); | |
2676a58c | 533 | node = rcu_dereference_raw(*slot); |
7cf9c2c7 | 534 | if (node == NULL) |
1da177e4 LT |
535 | return NULL; |
536 | ||
1da177e4 LT |
537 | shift -= RADIX_TREE_MAP_SHIFT; |
538 | height--; | |
7cf9c2c7 | 539 | } while (height > 0); |
1da177e4 | 540 | |
139e5616 JW |
541 | if (nodep) |
542 | *nodep = parent; | |
543 | if (slotp) | |
544 | *slotp = slot; | |
545 | return node; | |
b72b71c6 HS |
546 | } |
547 | ||
548 | /** | |
549 | * radix_tree_lookup_slot - lookup a slot in a radix tree | |
550 | * @root: radix tree root | |
551 | * @index: index key | |
552 | * | |
553 | * Returns: the slot corresponding to the position @index in the | |
554 | * radix tree @root. This is useful for update-if-exists operations. | |
555 | * | |
556 | * This function can be called under rcu_read_lock iff the slot is not | |
557 | * modified by radix_tree_replace_slot, otherwise it must be called | |
558 | * exclusive from other writers. Any dereference of the slot must be done | |
559 | * using radix_tree_deref_slot. | |
560 | */ | |
561 | void **radix_tree_lookup_slot(struct radix_tree_root *root, unsigned long index) | |
562 | { | |
139e5616 JW |
563 | void **slot; |
564 | ||
565 | if (!__radix_tree_lookup(root, index, NULL, &slot)) | |
566 | return NULL; | |
567 | return slot; | |
a4331366 | 568 | } |
a4331366 HR |
569 | EXPORT_SYMBOL(radix_tree_lookup_slot); |
570 | ||
571 | /** | |
572 | * radix_tree_lookup - perform lookup operation on a radix tree | |
573 | * @root: radix tree root | |
574 | * @index: index key | |
575 | * | |
576 | * Lookup the item at the position @index in the radix tree @root. | |
7cf9c2c7 NP |
577 | * |
578 | * This function can be called under rcu_read_lock, however the caller | |
579 | * must manage lifetimes of leaf nodes (eg. RCU may also be used to free | |
580 | * them safely). No RCU barriers are required to access or modify the | |
581 | * returned item, however. | |
a4331366 HR |
582 | */ |
583 | void *radix_tree_lookup(struct radix_tree_root *root, unsigned long index) | |
584 | { | |
139e5616 | 585 | return __radix_tree_lookup(root, index, NULL, NULL); |
1da177e4 LT |
586 | } |
587 | EXPORT_SYMBOL(radix_tree_lookup); | |
588 | ||
589 | /** | |
590 | * radix_tree_tag_set - set a tag on a radix tree node | |
591 | * @root: radix tree root | |
592 | * @index: index key | |
593 | * @tag: tag index | |
594 | * | |
daff89f3 JC |
595 | * Set the search tag (which must be < RADIX_TREE_MAX_TAGS) |
596 | * corresponding to @index in the radix tree. From | |
1da177e4 LT |
597 | * the root all the way down to the leaf node. |
598 | * | |
599 | * Returns the address of the tagged item. Setting a tag on a not-present | |
600 | * item is a bug. | |
601 | */ | |
602 | void *radix_tree_tag_set(struct radix_tree_root *root, | |
daff89f3 | 603 | unsigned long index, unsigned int tag) |
1da177e4 LT |
604 | { |
605 | unsigned int height, shift; | |
201b6264 | 606 | struct radix_tree_node *slot; |
1da177e4 LT |
607 | |
608 | height = root->height; | |
4c91c364 | 609 | BUG_ON(index > radix_tree_maxindex(height)); |
1da177e4 | 610 | |
27d20fdd | 611 | slot = indirect_to_ptr(root->rnode); |
612d6c19 | 612 | shift = (height - 1) * RADIX_TREE_MAP_SHIFT; |
1da177e4 LT |
613 | |
614 | while (height > 0) { | |
615 | int offset; | |
616 | ||
617 | offset = (index >> shift) & RADIX_TREE_MAP_MASK; | |
d5274261 NP |
618 | if (!tag_get(slot, tag, offset)) |
619 | tag_set(slot, tag, offset); | |
201b6264 CL |
620 | slot = slot->slots[offset]; |
621 | BUG_ON(slot == NULL); | |
1da177e4 LT |
622 | shift -= RADIX_TREE_MAP_SHIFT; |
623 | height--; | |
624 | } | |
625 | ||
612d6c19 NP |
626 | /* set the root's tag bit */ |
627 | if (slot && !root_tag_get(root, tag)) | |
628 | root_tag_set(root, tag); | |
629 | ||
201b6264 | 630 | return slot; |
1da177e4 LT |
631 | } |
632 | EXPORT_SYMBOL(radix_tree_tag_set); | |
633 | ||
634 | /** | |
635 | * radix_tree_tag_clear - clear a tag on a radix tree node | |
636 | * @root: radix tree root | |
637 | * @index: index key | |
638 | * @tag: tag index | |
639 | * | |
daff89f3 JC |
640 | * Clear the search tag (which must be < RADIX_TREE_MAX_TAGS) |
641 | * corresponding to @index in the radix tree. If | |
1da177e4 LT |
642 | * this causes the leaf node to have no tags set then clear the tag in the |
643 | * next-to-leaf node, etc. | |
644 | * | |
645 | * Returns the address of the tagged item on success, else NULL. ie: | |
646 | * has the same return value and semantics as radix_tree_lookup(). | |
647 | */ | |
648 | void *radix_tree_tag_clear(struct radix_tree_root *root, | |
daff89f3 | 649 | unsigned long index, unsigned int tag) |
1da177e4 | 650 | { |
e2bdb933 | 651 | struct radix_tree_node *node = NULL; |
612d6c19 | 652 | struct radix_tree_node *slot = NULL; |
1da177e4 | 653 | unsigned int height, shift; |
e2bdb933 | 654 | int uninitialized_var(offset); |
1da177e4 LT |
655 | |
656 | height = root->height; | |
657 | if (index > radix_tree_maxindex(height)) | |
658 | goto out; | |
659 | ||
e2bdb933 | 660 | shift = height * RADIX_TREE_MAP_SHIFT; |
27d20fdd | 661 | slot = indirect_to_ptr(root->rnode); |
1da177e4 | 662 | |
e2bdb933 | 663 | while (shift) { |
201b6264 | 664 | if (slot == NULL) |
1da177e4 LT |
665 | goto out; |
666 | ||
e2bdb933 | 667 | shift -= RADIX_TREE_MAP_SHIFT; |
1da177e4 | 668 | offset = (index >> shift) & RADIX_TREE_MAP_MASK; |
e2bdb933 | 669 | node = slot; |
201b6264 | 670 | slot = slot->slots[offset]; |
1da177e4 LT |
671 | } |
672 | ||
612d6c19 | 673 | if (slot == NULL) |
1da177e4 LT |
674 | goto out; |
675 | ||
e2bdb933 HD |
676 | while (node) { |
677 | if (!tag_get(node, tag, offset)) | |
d5274261 | 678 | goto out; |
e2bdb933 HD |
679 | tag_clear(node, tag, offset); |
680 | if (any_tag_set(node, tag)) | |
6e954b9e | 681 | goto out; |
e2bdb933 HD |
682 | |
683 | index >>= RADIX_TREE_MAP_SHIFT; | |
684 | offset = index & RADIX_TREE_MAP_MASK; | |
685 | node = node->parent; | |
612d6c19 NP |
686 | } |
687 | ||
688 | /* clear the root's tag bit */ | |
689 | if (root_tag_get(root, tag)) | |
690 | root_tag_clear(root, tag); | |
691 | ||
1da177e4 | 692 | out: |
612d6c19 | 693 | return slot; |
1da177e4 LT |
694 | } |
695 | EXPORT_SYMBOL(radix_tree_tag_clear); | |
696 | ||
1da177e4 | 697 | /** |
32605a18 MT |
698 | * radix_tree_tag_get - get a tag on a radix tree node |
699 | * @root: radix tree root | |
700 | * @index: index key | |
daff89f3 | 701 | * @tag: tag index (< RADIX_TREE_MAX_TAGS) |
1da177e4 | 702 | * |
32605a18 | 703 | * Return values: |
1da177e4 | 704 | * |
612d6c19 NP |
705 | * 0: tag not present or not set |
706 | * 1: tag set | |
ce82653d DH |
707 | * |
708 | * Note that the return value of this function may not be relied on, even if | |
709 | * the RCU lock is held, unless tag modification and node deletion are excluded | |
710 | * from concurrency. | |
1da177e4 LT |
711 | */ |
712 | int radix_tree_tag_get(struct radix_tree_root *root, | |
daff89f3 | 713 | unsigned long index, unsigned int tag) |
1da177e4 LT |
714 | { |
715 | unsigned int height, shift; | |
7cf9c2c7 | 716 | struct radix_tree_node *node; |
1da177e4 | 717 | |
612d6c19 NP |
718 | /* check the root's tag bit */ |
719 | if (!root_tag_get(root, tag)) | |
720 | return 0; | |
721 | ||
2676a58c | 722 | node = rcu_dereference_raw(root->rnode); |
7cf9c2c7 NP |
723 | if (node == NULL) |
724 | return 0; | |
725 | ||
c0bc9875 | 726 | if (!radix_tree_is_indirect_ptr(node)) |
7cf9c2c7 | 727 | return (index == 0); |
27d20fdd | 728 | node = indirect_to_ptr(node); |
7cf9c2c7 | 729 | |
449dd698 | 730 | height = node->path & RADIX_TREE_HEIGHT_MASK; |
7cf9c2c7 NP |
731 | if (index > radix_tree_maxindex(height)) |
732 | return 0; | |
612d6c19 | 733 | |
1da177e4 | 734 | shift = (height - 1) * RADIX_TREE_MAP_SHIFT; |
1da177e4 LT |
735 | |
736 | for ( ; ; ) { | |
737 | int offset; | |
738 | ||
7cf9c2c7 | 739 | if (node == NULL) |
1da177e4 LT |
740 | return 0; |
741 | ||
742 | offset = (index >> shift) & RADIX_TREE_MAP_MASK; | |
7cf9c2c7 | 743 | if (!tag_get(node, tag, offset)) |
3fa36acb | 744 | return 0; |
ce82653d | 745 | if (height == 1) |
3fa36acb | 746 | return 1; |
2676a58c | 747 | node = rcu_dereference_raw(node->slots[offset]); |
1da177e4 LT |
748 | shift -= RADIX_TREE_MAP_SHIFT; |
749 | height--; | |
750 | } | |
751 | } | |
752 | EXPORT_SYMBOL(radix_tree_tag_get); | |
1da177e4 | 753 | |
78c1d784 KK |
754 | /** |
755 | * radix_tree_next_chunk - find next chunk of slots for iteration | |
756 | * | |
757 | * @root: radix tree root | |
758 | * @iter: iterator state | |
759 | * @flags: RADIX_TREE_ITER_* flags and tag index | |
760 | * Returns: pointer to chunk first slot, or NULL if iteration is over | |
761 | */ | |
762 | void **radix_tree_next_chunk(struct radix_tree_root *root, | |
763 | struct radix_tree_iter *iter, unsigned flags) | |
764 | { | |
765 | unsigned shift, tag = flags & RADIX_TREE_ITER_TAG_MASK; | |
766 | struct radix_tree_node *rnode, *node; | |
449dd698 | 767 | unsigned long index, offset, height; |
78c1d784 KK |
768 | |
769 | if ((flags & RADIX_TREE_ITER_TAGGED) && !root_tag_get(root, tag)) | |
770 | return NULL; | |
771 | ||
772 | /* | |
773 | * Catch next_index overflow after ~0UL. iter->index never overflows | |
774 | * during iterating; it can be zero only at the beginning. | |
775 | * And we cannot overflow iter->next_index in a single step, | |
776 | * because RADIX_TREE_MAP_SHIFT < BITS_PER_LONG. | |
fffaee36 KK |
777 | * |
778 | * This condition also used by radix_tree_next_slot() to stop | |
779 | * contiguous iterating, and forbid swithing to the next chunk. | |
78c1d784 KK |
780 | */ |
781 | index = iter->next_index; | |
782 | if (!index && iter->index) | |
783 | return NULL; | |
784 | ||
785 | rnode = rcu_dereference_raw(root->rnode); | |
786 | if (radix_tree_is_indirect_ptr(rnode)) { | |
787 | rnode = indirect_to_ptr(rnode); | |
788 | } else if (rnode && !index) { | |
789 | /* Single-slot tree */ | |
790 | iter->index = 0; | |
791 | iter->next_index = 1; | |
792 | iter->tags = 1; | |
793 | return (void **)&root->rnode; | |
794 | } else | |
795 | return NULL; | |
796 | ||
797 | restart: | |
449dd698 JW |
798 | height = rnode->path & RADIX_TREE_HEIGHT_MASK; |
799 | shift = (height - 1) * RADIX_TREE_MAP_SHIFT; | |
78c1d784 KK |
800 | offset = index >> shift; |
801 | ||
802 | /* Index outside of the tree */ | |
803 | if (offset >= RADIX_TREE_MAP_SIZE) | |
804 | return NULL; | |
805 | ||
806 | node = rnode; | |
807 | while (1) { | |
808 | if ((flags & RADIX_TREE_ITER_TAGGED) ? | |
809 | !test_bit(offset, node->tags[tag]) : | |
810 | !node->slots[offset]) { | |
811 | /* Hole detected */ | |
812 | if (flags & RADIX_TREE_ITER_CONTIG) | |
813 | return NULL; | |
814 | ||
815 | if (flags & RADIX_TREE_ITER_TAGGED) | |
816 | offset = radix_tree_find_next_bit( | |
817 | node->tags[tag], | |
818 | RADIX_TREE_MAP_SIZE, | |
819 | offset + 1); | |
820 | else | |
821 | while (++offset < RADIX_TREE_MAP_SIZE) { | |
822 | if (node->slots[offset]) | |
823 | break; | |
824 | } | |
825 | index &= ~((RADIX_TREE_MAP_SIZE << shift) - 1); | |
826 | index += offset << shift; | |
827 | /* Overflow after ~0UL */ | |
828 | if (!index) | |
829 | return NULL; | |
830 | if (offset == RADIX_TREE_MAP_SIZE) | |
831 | goto restart; | |
832 | } | |
833 | ||
834 | /* This is leaf-node */ | |
835 | if (!shift) | |
836 | break; | |
837 | ||
838 | node = rcu_dereference_raw(node->slots[offset]); | |
839 | if (node == NULL) | |
840 | goto restart; | |
841 | shift -= RADIX_TREE_MAP_SHIFT; | |
842 | offset = (index >> shift) & RADIX_TREE_MAP_MASK; | |
843 | } | |
844 | ||
845 | /* Update the iterator state */ | |
846 | iter->index = index; | |
847 | iter->next_index = (index | RADIX_TREE_MAP_MASK) + 1; | |
848 | ||
849 | /* Construct iter->tags bit-mask from node->tags[tag] array */ | |
850 | if (flags & RADIX_TREE_ITER_TAGGED) { | |
851 | unsigned tag_long, tag_bit; | |
852 | ||
853 | tag_long = offset / BITS_PER_LONG; | |
854 | tag_bit = offset % BITS_PER_LONG; | |
855 | iter->tags = node->tags[tag][tag_long] >> tag_bit; | |
856 | /* This never happens if RADIX_TREE_TAG_LONGS == 1 */ | |
857 | if (tag_long < RADIX_TREE_TAG_LONGS - 1) { | |
858 | /* Pick tags from next element */ | |
859 | if (tag_bit) | |
860 | iter->tags |= node->tags[tag][tag_long + 1] << | |
861 | (BITS_PER_LONG - tag_bit); | |
862 | /* Clip chunk size, here only BITS_PER_LONG tags */ | |
863 | iter->next_index = index + BITS_PER_LONG; | |
864 | } | |
865 | } | |
866 | ||
867 | return node->slots + offset; | |
868 | } | |
869 | EXPORT_SYMBOL(radix_tree_next_chunk); | |
870 | ||
ebf8aa44 JK |
871 | /** |
872 | * radix_tree_range_tag_if_tagged - for each item in given range set given | |
873 | * tag if item has another tag set | |
874 | * @root: radix tree root | |
875 | * @first_indexp: pointer to a starting index of a range to scan | |
876 | * @last_index: last index of a range to scan | |
877 | * @nr_to_tag: maximum number items to tag | |
878 | * @iftag: tag index to test | |
879 | * @settag: tag index to set if tested tag is set | |
880 | * | |
881 | * This function scans range of radix tree from first_index to last_index | |
882 | * (inclusive). For each item in the range if iftag is set, the function sets | |
883 | * also settag. The function stops either after tagging nr_to_tag items or | |
884 | * after reaching last_index. | |
885 | * | |
144dcfc0 DC |
886 | * The tags must be set from the leaf level only and propagated back up the |
887 | * path to the root. We must do this so that we resolve the full path before | |
888 | * setting any tags on intermediate nodes. If we set tags as we descend, then | |
889 | * we can get to the leaf node and find that the index that has the iftag | |
890 | * set is outside the range we are scanning. This reults in dangling tags and | |
891 | * can lead to problems with later tag operations (e.g. livelocks on lookups). | |
892 | * | |
ebf8aa44 JK |
893 | * The function returns number of leaves where the tag was set and sets |
894 | * *first_indexp to the first unscanned index. | |
d5ed3a4a JK |
895 | * WARNING! *first_indexp can wrap if last_index is ULONG_MAX. Caller must |
896 | * be prepared to handle that. | |
ebf8aa44 JK |
897 | */ |
898 | unsigned long radix_tree_range_tag_if_tagged(struct radix_tree_root *root, | |
899 | unsigned long *first_indexp, unsigned long last_index, | |
900 | unsigned long nr_to_tag, | |
901 | unsigned int iftag, unsigned int settag) | |
902 | { | |
144dcfc0 | 903 | unsigned int height = root->height; |
e2bdb933 | 904 | struct radix_tree_node *node = NULL; |
144dcfc0 DC |
905 | struct radix_tree_node *slot; |
906 | unsigned int shift; | |
907 | unsigned long tagged = 0; | |
908 | unsigned long index = *first_indexp; | |
ebf8aa44 JK |
909 | |
910 | last_index = min(last_index, radix_tree_maxindex(height)); | |
911 | if (index > last_index) | |
912 | return 0; | |
913 | if (!nr_to_tag) | |
914 | return 0; | |
915 | if (!root_tag_get(root, iftag)) { | |
916 | *first_indexp = last_index + 1; | |
917 | return 0; | |
918 | } | |
919 | if (height == 0) { | |
920 | *first_indexp = last_index + 1; | |
921 | root_tag_set(root, settag); | |
922 | return 1; | |
923 | } | |
924 | ||
925 | shift = (height - 1) * RADIX_TREE_MAP_SHIFT; | |
27d20fdd | 926 | slot = indirect_to_ptr(root->rnode); |
ebf8aa44 JK |
927 | |
928 | for (;;) { | |
e2bdb933 | 929 | unsigned long upindex; |
ebf8aa44 JK |
930 | int offset; |
931 | ||
932 | offset = (index >> shift) & RADIX_TREE_MAP_MASK; | |
933 | if (!slot->slots[offset]) | |
934 | goto next; | |
935 | if (!tag_get(slot, iftag, offset)) | |
936 | goto next; | |
e2bdb933 | 937 | if (shift) { |
144dcfc0 | 938 | /* Go down one level */ |
144dcfc0 | 939 | shift -= RADIX_TREE_MAP_SHIFT; |
e2bdb933 | 940 | node = slot; |
144dcfc0 DC |
941 | slot = slot->slots[offset]; |
942 | continue; | |
943 | } | |
944 | ||
945 | /* tag the leaf */ | |
946 | tagged++; | |
ebf8aa44 | 947 | tag_set(slot, settag, offset); |
144dcfc0 DC |
948 | |
949 | /* walk back up the path tagging interior nodes */ | |
e2bdb933 HD |
950 | upindex = index; |
951 | while (node) { | |
952 | upindex >>= RADIX_TREE_MAP_SHIFT; | |
953 | offset = upindex & RADIX_TREE_MAP_MASK; | |
954 | ||
144dcfc0 | 955 | /* stop if we find a node with the tag already set */ |
e2bdb933 | 956 | if (tag_get(node, settag, offset)) |
144dcfc0 | 957 | break; |
e2bdb933 HD |
958 | tag_set(node, settag, offset); |
959 | node = node->parent; | |
ebf8aa44 | 960 | } |
144dcfc0 | 961 | |
e2bdb933 HD |
962 | /* |
963 | * Small optimization: now clear that node pointer. | |
964 | * Since all of this slot's ancestors now have the tag set | |
965 | * from setting it above, we have no further need to walk | |
966 | * back up the tree setting tags, until we update slot to | |
967 | * point to another radix_tree_node. | |
968 | */ | |
969 | node = NULL; | |
970 | ||
ebf8aa44 JK |
971 | next: |
972 | /* Go to next item at level determined by 'shift' */ | |
973 | index = ((index >> shift) + 1) << shift; | |
d5ed3a4a JK |
974 | /* Overflow can happen when last_index is ~0UL... */ |
975 | if (index > last_index || !index) | |
ebf8aa44 JK |
976 | break; |
977 | if (tagged >= nr_to_tag) | |
978 | break; | |
979 | while (((index >> shift) & RADIX_TREE_MAP_MASK) == 0) { | |
980 | /* | |
981 | * We've fully scanned this node. Go up. Because | |
982 | * last_index is guaranteed to be in the tree, what | |
983 | * we do below cannot wander astray. | |
984 | */ | |
e2bdb933 | 985 | slot = slot->parent; |
ebf8aa44 JK |
986 | shift += RADIX_TREE_MAP_SHIFT; |
987 | } | |
988 | } | |
989 | /* | |
ac15ee69 TO |
990 | * We need not to tag the root tag if there is no tag which is set with |
991 | * settag within the range from *first_indexp to last_index. | |
ebf8aa44 | 992 | */ |
ac15ee69 TO |
993 | if (tagged > 0) |
994 | root_tag_set(root, settag); | |
ebf8aa44 JK |
995 | *first_indexp = index; |
996 | ||
997 | return tagged; | |
998 | } | |
999 | EXPORT_SYMBOL(radix_tree_range_tag_if_tagged); | |
1000 | ||
1da177e4 LT |
1001 | /** |
1002 | * radix_tree_gang_lookup - perform multiple lookup on a radix tree | |
1003 | * @root: radix tree root | |
1004 | * @results: where the results of the lookup are placed | |
1005 | * @first_index: start the lookup from this key | |
1006 | * @max_items: place up to this many items at *results | |
1007 | * | |
1008 | * Performs an index-ascending scan of the tree for present items. Places | |
1009 | * them at *@results and returns the number of items which were placed at | |
1010 | * *@results. | |
1011 | * | |
1012 | * The implementation is naive. | |
7cf9c2c7 NP |
1013 | * |
1014 | * Like radix_tree_lookup, radix_tree_gang_lookup may be called under | |
1015 | * rcu_read_lock. In this case, rather than the returned results being | |
1016 | * an atomic snapshot of the tree at a single point in time, the semantics | |
1017 | * of an RCU protected gang lookup are as though multiple radix_tree_lookups | |
1018 | * have been issued in individual locks, and results stored in 'results'. | |
1da177e4 LT |
1019 | */ |
1020 | unsigned int | |
1021 | radix_tree_gang_lookup(struct radix_tree_root *root, void **results, | |
1022 | unsigned long first_index, unsigned int max_items) | |
1023 | { | |
cebbd29e KK |
1024 | struct radix_tree_iter iter; |
1025 | void **slot; | |
1026 | unsigned int ret = 0; | |
7cf9c2c7 | 1027 | |
cebbd29e | 1028 | if (unlikely(!max_items)) |
7cf9c2c7 | 1029 | return 0; |
1da177e4 | 1030 | |
cebbd29e | 1031 | radix_tree_for_each_slot(slot, root, &iter, first_index) { |
46437f9a | 1032 | results[ret] = rcu_dereference_raw(*slot); |
cebbd29e KK |
1033 | if (!results[ret]) |
1034 | continue; | |
46437f9a MW |
1035 | if (radix_tree_is_indirect_ptr(results[ret])) { |
1036 | slot = radix_tree_iter_retry(&iter); | |
1037 | continue; | |
1038 | } | |
cebbd29e | 1039 | if (++ret == max_items) |
1da177e4 | 1040 | break; |
1da177e4 | 1041 | } |
7cf9c2c7 | 1042 | |
1da177e4 LT |
1043 | return ret; |
1044 | } | |
1045 | EXPORT_SYMBOL(radix_tree_gang_lookup); | |
1046 | ||
47feff2c NP |
1047 | /** |
1048 | * radix_tree_gang_lookup_slot - perform multiple slot lookup on radix tree | |
1049 | * @root: radix tree root | |
1050 | * @results: where the results of the lookup are placed | |
6328650b | 1051 | * @indices: where their indices should be placed (but usually NULL) |
47feff2c NP |
1052 | * @first_index: start the lookup from this key |
1053 | * @max_items: place up to this many items at *results | |
1054 | * | |
1055 | * Performs an index-ascending scan of the tree for present items. Places | |
1056 | * their slots at *@results and returns the number of items which were | |
1057 | * placed at *@results. | |
1058 | * | |
1059 | * The implementation is naive. | |
1060 | * | |
1061 | * Like radix_tree_gang_lookup as far as RCU and locking goes. Slots must | |
1062 | * be dereferenced with radix_tree_deref_slot, and if using only RCU | |
1063 | * protection, radix_tree_deref_slot may fail requiring a retry. | |
1064 | */ | |
1065 | unsigned int | |
6328650b HD |
1066 | radix_tree_gang_lookup_slot(struct radix_tree_root *root, |
1067 | void ***results, unsigned long *indices, | |
47feff2c NP |
1068 | unsigned long first_index, unsigned int max_items) |
1069 | { | |
cebbd29e KK |
1070 | struct radix_tree_iter iter; |
1071 | void **slot; | |
1072 | unsigned int ret = 0; | |
47feff2c | 1073 | |
cebbd29e | 1074 | if (unlikely(!max_items)) |
47feff2c NP |
1075 | return 0; |
1076 | ||
cebbd29e KK |
1077 | radix_tree_for_each_slot(slot, root, &iter, first_index) { |
1078 | results[ret] = slot; | |
6328650b | 1079 | if (indices) |
cebbd29e KK |
1080 | indices[ret] = iter.index; |
1081 | if (++ret == max_items) | |
47feff2c | 1082 | break; |
47feff2c NP |
1083 | } |
1084 | ||
1085 | return ret; | |
1086 | } | |
1087 | EXPORT_SYMBOL(radix_tree_gang_lookup_slot); | |
1088 | ||
1da177e4 LT |
1089 | /** |
1090 | * radix_tree_gang_lookup_tag - perform multiple lookup on a radix tree | |
1091 | * based on a tag | |
1092 | * @root: radix tree root | |
1093 | * @results: where the results of the lookup are placed | |
1094 | * @first_index: start the lookup from this key | |
1095 | * @max_items: place up to this many items at *results | |
daff89f3 | 1096 | * @tag: the tag index (< RADIX_TREE_MAX_TAGS) |
1da177e4 LT |
1097 | * |
1098 | * Performs an index-ascending scan of the tree for present items which | |
1099 | * have the tag indexed by @tag set. Places the items at *@results and | |
1100 | * returns the number of items which were placed at *@results. | |
1101 | */ | |
1102 | unsigned int | |
1103 | radix_tree_gang_lookup_tag(struct radix_tree_root *root, void **results, | |
daff89f3 JC |
1104 | unsigned long first_index, unsigned int max_items, |
1105 | unsigned int tag) | |
1da177e4 | 1106 | { |
cebbd29e KK |
1107 | struct radix_tree_iter iter; |
1108 | void **slot; | |
1109 | unsigned int ret = 0; | |
612d6c19 | 1110 | |
cebbd29e | 1111 | if (unlikely(!max_items)) |
7cf9c2c7 NP |
1112 | return 0; |
1113 | ||
cebbd29e | 1114 | radix_tree_for_each_tagged(slot, root, &iter, first_index, tag) { |
46437f9a | 1115 | results[ret] = rcu_dereference_raw(*slot); |
cebbd29e KK |
1116 | if (!results[ret]) |
1117 | continue; | |
46437f9a MW |
1118 | if (radix_tree_is_indirect_ptr(results[ret])) { |
1119 | slot = radix_tree_iter_retry(&iter); | |
1120 | continue; | |
1121 | } | |
cebbd29e | 1122 | if (++ret == max_items) |
1da177e4 | 1123 | break; |
1da177e4 | 1124 | } |
7cf9c2c7 | 1125 | |
1da177e4 LT |
1126 | return ret; |
1127 | } | |
1128 | EXPORT_SYMBOL(radix_tree_gang_lookup_tag); | |
1129 | ||
47feff2c NP |
1130 | /** |
1131 | * radix_tree_gang_lookup_tag_slot - perform multiple slot lookup on a | |
1132 | * radix tree based on a tag | |
1133 | * @root: radix tree root | |
1134 | * @results: where the results of the lookup are placed | |
1135 | * @first_index: start the lookup from this key | |
1136 | * @max_items: place up to this many items at *results | |
1137 | * @tag: the tag index (< RADIX_TREE_MAX_TAGS) | |
1138 | * | |
1139 | * Performs an index-ascending scan of the tree for present items which | |
1140 | * have the tag indexed by @tag set. Places the slots at *@results and | |
1141 | * returns the number of slots which were placed at *@results. | |
1142 | */ | |
1143 | unsigned int | |
1144 | radix_tree_gang_lookup_tag_slot(struct radix_tree_root *root, void ***results, | |
1145 | unsigned long first_index, unsigned int max_items, | |
1146 | unsigned int tag) | |
1147 | { | |
cebbd29e KK |
1148 | struct radix_tree_iter iter; |
1149 | void **slot; | |
1150 | unsigned int ret = 0; | |
47feff2c | 1151 | |
cebbd29e | 1152 | if (unlikely(!max_items)) |
47feff2c NP |
1153 | return 0; |
1154 | ||
cebbd29e KK |
1155 | radix_tree_for_each_tagged(slot, root, &iter, first_index, tag) { |
1156 | results[ret] = slot; | |
1157 | if (++ret == max_items) | |
47feff2c | 1158 | break; |
47feff2c NP |
1159 | } |
1160 | ||
1161 | return ret; | |
1162 | } | |
1163 | EXPORT_SYMBOL(radix_tree_gang_lookup_tag_slot); | |
1164 | ||
e504f3fd HD |
1165 | #if defined(CONFIG_SHMEM) && defined(CONFIG_SWAP) |
1166 | #include <linux/sched.h> /* for cond_resched() */ | |
1167 | ||
1168 | /* | |
1169 | * This linear search is at present only useful to shmem_unuse_inode(). | |
1170 | */ | |
1171 | static unsigned long __locate(struct radix_tree_node *slot, void *item, | |
1172 | unsigned long index, unsigned long *found_index) | |
1173 | { | |
1174 | unsigned int shift, height; | |
1175 | unsigned long i; | |
1176 | ||
449dd698 | 1177 | height = slot->path & RADIX_TREE_HEIGHT_MASK; |
e504f3fd HD |
1178 | shift = (height-1) * RADIX_TREE_MAP_SHIFT; |
1179 | ||
1180 | for ( ; height > 1; height--) { | |
1181 | i = (index >> shift) & RADIX_TREE_MAP_MASK; | |
1182 | for (;;) { | |
1183 | if (slot->slots[i] != NULL) | |
1184 | break; | |
1185 | index &= ~((1UL << shift) - 1); | |
1186 | index += 1UL << shift; | |
1187 | if (index == 0) | |
1188 | goto out; /* 32-bit wraparound */ | |
1189 | i++; | |
1190 | if (i == RADIX_TREE_MAP_SIZE) | |
1191 | goto out; | |
1192 | } | |
1193 | ||
1194 | shift -= RADIX_TREE_MAP_SHIFT; | |
1195 | slot = rcu_dereference_raw(slot->slots[i]); | |
1196 | if (slot == NULL) | |
1197 | goto out; | |
1198 | } | |
1199 | ||
1200 | /* Bottom level: check items */ | |
1201 | for (i = 0; i < RADIX_TREE_MAP_SIZE; i++) { | |
1202 | if (slot->slots[i] == item) { | |
1203 | *found_index = index + i; | |
1204 | index = 0; | |
1205 | goto out; | |
1206 | } | |
1207 | } | |
1208 | index += RADIX_TREE_MAP_SIZE; | |
1209 | out: | |
1210 | return index; | |
1211 | } | |
1212 | ||
1213 | /** | |
1214 | * radix_tree_locate_item - search through radix tree for item | |
1215 | * @root: radix tree root | |
1216 | * @item: item to be found | |
1217 | * | |
1218 | * Returns index where item was found, or -1 if not found. | |
1219 | * Caller must hold no lock (since this time-consuming function needs | |
1220 | * to be preemptible), and must check afterwards if item is still there. | |
1221 | */ | |
1222 | unsigned long radix_tree_locate_item(struct radix_tree_root *root, void *item) | |
1223 | { | |
1224 | struct radix_tree_node *node; | |
1225 | unsigned long max_index; | |
1226 | unsigned long cur_index = 0; | |
1227 | unsigned long found_index = -1; | |
1228 | ||
1229 | do { | |
1230 | rcu_read_lock(); | |
1231 | node = rcu_dereference_raw(root->rnode); | |
1232 | if (!radix_tree_is_indirect_ptr(node)) { | |
1233 | rcu_read_unlock(); | |
1234 | if (node == item) | |
1235 | found_index = 0; | |
1236 | break; | |
1237 | } | |
1238 | ||
1239 | node = indirect_to_ptr(node); | |
449dd698 JW |
1240 | max_index = radix_tree_maxindex(node->path & |
1241 | RADIX_TREE_HEIGHT_MASK); | |
5f30fc94 HD |
1242 | if (cur_index > max_index) { |
1243 | rcu_read_unlock(); | |
e504f3fd | 1244 | break; |
5f30fc94 | 1245 | } |
e504f3fd HD |
1246 | |
1247 | cur_index = __locate(node, item, cur_index, &found_index); | |
1248 | rcu_read_unlock(); | |
1249 | cond_resched(); | |
1250 | } while (cur_index != 0 && cur_index <= max_index); | |
1251 | ||
1252 | return found_index; | |
1253 | } | |
1254 | #else | |
1255 | unsigned long radix_tree_locate_item(struct radix_tree_root *root, void *item) | |
1256 | { | |
1257 | return -1; | |
1258 | } | |
1259 | #endif /* CONFIG_SHMEM && CONFIG_SWAP */ | |
47feff2c | 1260 | |
a5f51c96 NP |
1261 | /** |
1262 | * radix_tree_shrink - shrink height of a radix tree to minimal | |
1263 | * @root radix tree root | |
1264 | */ | |
1265 | static inline void radix_tree_shrink(struct radix_tree_root *root) | |
1266 | { | |
1267 | /* try to shrink tree height */ | |
c0bc9875 | 1268 | while (root->height > 0) { |
a5f51c96 | 1269 | struct radix_tree_node *to_free = root->rnode; |
e2bdb933 | 1270 | struct radix_tree_node *slot; |
a5f51c96 | 1271 | |
c0bc9875 | 1272 | BUG_ON(!radix_tree_is_indirect_ptr(to_free)); |
27d20fdd | 1273 | to_free = indirect_to_ptr(to_free); |
c0bc9875 NP |
1274 | |
1275 | /* | |
1276 | * The candidate node has more than one child, or its child | |
1277 | * is not at the leftmost slot, we cannot shrink. | |
1278 | */ | |
1279 | if (to_free->count != 1) | |
1280 | break; | |
1281 | if (!to_free->slots[0]) | |
1282 | break; | |
1283 | ||
7cf9c2c7 NP |
1284 | /* |
1285 | * We don't need rcu_assign_pointer(), since we are simply | |
27d20fdd NP |
1286 | * moving the node from one part of the tree to another: if it |
1287 | * was safe to dereference the old pointer to it | |
7cf9c2c7 | 1288 | * (to_free->slots[0]), it will be safe to dereference the new |
27d20fdd | 1289 | * one (root->rnode) as far as dependent read barriers go. |
7cf9c2c7 | 1290 | */ |
e2bdb933 HD |
1291 | slot = to_free->slots[0]; |
1292 | if (root->height > 1) { | |
1293 | slot->parent = NULL; | |
1294 | slot = ptr_to_indirect(slot); | |
1295 | } | |
1296 | root->rnode = slot; | |
a5f51c96 | 1297 | root->height--; |
27d20fdd NP |
1298 | |
1299 | /* | |
1300 | * We have a dilemma here. The node's slot[0] must not be | |
1301 | * NULLed in case there are concurrent lookups expecting to | |
1302 | * find the item. However if this was a bottom-level node, | |
1303 | * then it may be subject to the slot pointer being visible | |
1304 | * to callers dereferencing it. If item corresponding to | |
1305 | * slot[0] is subsequently deleted, these callers would expect | |
1306 | * their slot to become empty sooner or later. | |
1307 | * | |
1308 | * For example, lockless pagecache will look up a slot, deref | |
1309 | * the page pointer, and if the page is 0 refcount it means it | |
1310 | * was concurrently deleted from pagecache so try the deref | |
1311 | * again. Fortunately there is already a requirement for logic | |
1312 | * to retry the entire slot lookup -- the indirect pointer | |
1313 | * problem (replacing direct root node with an indirect pointer | |
1314 | * also results in a stale slot). So tag the slot as indirect | |
1315 | * to force callers to retry. | |
1316 | */ | |
1317 | if (root->height == 0) | |
1318 | *((unsigned long *)&to_free->slots[0]) |= | |
1319 | RADIX_TREE_INDIRECT_PTR; | |
1320 | ||
a5f51c96 NP |
1321 | radix_tree_node_free(to_free); |
1322 | } | |
1323 | } | |
1324 | ||
139e5616 JW |
1325 | /** |
1326 | * __radix_tree_delete_node - try to free node after clearing a slot | |
1327 | * @root: radix tree root | |
139e5616 JW |
1328 | * @node: node containing @index |
1329 | * | |
1330 | * After clearing the slot at @index in @node from radix tree | |
1331 | * rooted at @root, call this function to attempt freeing the | |
1332 | * node and shrinking the tree. | |
1333 | * | |
1334 | * Returns %true if @node was freed, %false otherwise. | |
1335 | */ | |
449dd698 | 1336 | bool __radix_tree_delete_node(struct radix_tree_root *root, |
139e5616 JW |
1337 | struct radix_tree_node *node) |
1338 | { | |
1339 | bool deleted = false; | |
1340 | ||
1341 | do { | |
1342 | struct radix_tree_node *parent; | |
1343 | ||
1344 | if (node->count) { | |
1345 | if (node == indirect_to_ptr(root->rnode)) { | |
1346 | radix_tree_shrink(root); | |
1347 | if (root->height == 0) | |
1348 | deleted = true; | |
1349 | } | |
1350 | return deleted; | |
1351 | } | |
1352 | ||
1353 | parent = node->parent; | |
1354 | if (parent) { | |
449dd698 | 1355 | unsigned int offset; |
139e5616 | 1356 | |
449dd698 JW |
1357 | offset = node->path >> RADIX_TREE_HEIGHT_SHIFT; |
1358 | parent->slots[offset] = NULL; | |
139e5616 JW |
1359 | parent->count--; |
1360 | } else { | |
1361 | root_tag_clear_all(root); | |
1362 | root->height = 0; | |
1363 | root->rnode = NULL; | |
1364 | } | |
1365 | ||
1366 | radix_tree_node_free(node); | |
1367 | deleted = true; | |
1368 | ||
1369 | node = parent; | |
1370 | } while (node); | |
1371 | ||
1372 | return deleted; | |
1373 | } | |
1374 | ||
1da177e4 | 1375 | /** |
53c59f26 | 1376 | * radix_tree_delete_item - delete an item from a radix tree |
1da177e4 LT |
1377 | * @root: radix tree root |
1378 | * @index: index key | |
53c59f26 | 1379 | * @item: expected item |
1da177e4 | 1380 | * |
53c59f26 | 1381 | * Remove @item at @index from the radix tree rooted at @root. |
1da177e4 | 1382 | * |
53c59f26 JW |
1383 | * Returns the address of the deleted item, or NULL if it was not present |
1384 | * or the entry at the given @index was not @item. | |
1da177e4 | 1385 | */ |
53c59f26 JW |
1386 | void *radix_tree_delete_item(struct radix_tree_root *root, |
1387 | unsigned long index, void *item) | |
1da177e4 | 1388 | { |
139e5616 JW |
1389 | struct radix_tree_node *node; |
1390 | unsigned int offset; | |
1391 | void **slot; | |
1392 | void *entry; | |
d5274261 | 1393 | int tag; |
1da177e4 | 1394 | |
139e5616 JW |
1395 | entry = __radix_tree_lookup(root, index, &node, &slot); |
1396 | if (!entry) | |
1397 | return NULL; | |
1da177e4 | 1398 | |
139e5616 JW |
1399 | if (item && entry != item) |
1400 | return NULL; | |
1401 | ||
1402 | if (!node) { | |
612d6c19 NP |
1403 | root_tag_clear_all(root); |
1404 | root->rnode = NULL; | |
139e5616 | 1405 | return entry; |
612d6c19 | 1406 | } |
1da177e4 | 1407 | |
139e5616 | 1408 | offset = index & RADIX_TREE_MAP_MASK; |
53c59f26 | 1409 | |
1da177e4 | 1410 | /* |
e2bdb933 HD |
1411 | * Clear all tags associated with the item to be deleted. |
1412 | * This way of doing it would be inefficient, but seldom is any set. | |
1da177e4 | 1413 | */ |
daff89f3 | 1414 | for (tag = 0; tag < RADIX_TREE_MAX_TAGS; tag++) { |
e2bdb933 | 1415 | if (tag_get(node, tag, offset)) |
612d6c19 | 1416 | radix_tree_tag_clear(root, index, tag); |
d5274261 | 1417 | } |
1da177e4 | 1418 | |
139e5616 JW |
1419 | node->slots[offset] = NULL; |
1420 | node->count--; | |
e2bdb933 | 1421 | |
449dd698 | 1422 | __radix_tree_delete_node(root, node); |
612d6c19 | 1423 | |
139e5616 | 1424 | return entry; |
1da177e4 | 1425 | } |
53c59f26 JW |
1426 | EXPORT_SYMBOL(radix_tree_delete_item); |
1427 | ||
1428 | /** | |
1429 | * radix_tree_delete - delete an item from a radix tree | |
1430 | * @root: radix tree root | |
1431 | * @index: index key | |
1432 | * | |
1433 | * Remove the item at @index from the radix tree rooted at @root. | |
1434 | * | |
1435 | * Returns the address of the deleted item, or NULL if it was not present. | |
1436 | */ | |
1437 | void *radix_tree_delete(struct radix_tree_root *root, unsigned long index) | |
1438 | { | |
1439 | return radix_tree_delete_item(root, index, NULL); | |
1440 | } | |
1da177e4 LT |
1441 | EXPORT_SYMBOL(radix_tree_delete); |
1442 | ||
1443 | /** | |
1444 | * radix_tree_tagged - test whether any items in the tree are tagged | |
1445 | * @root: radix tree root | |
1446 | * @tag: tag to test | |
1447 | */ | |
daff89f3 | 1448 | int radix_tree_tagged(struct radix_tree_root *root, unsigned int tag) |
1da177e4 | 1449 | { |
612d6c19 | 1450 | return root_tag_get(root, tag); |
1da177e4 LT |
1451 | } |
1452 | EXPORT_SYMBOL(radix_tree_tagged); | |
1453 | ||
1454 | static void | |
449dd698 | 1455 | radix_tree_node_ctor(void *arg) |
1da177e4 | 1456 | { |
449dd698 JW |
1457 | struct radix_tree_node *node = arg; |
1458 | ||
1459 | memset(node, 0, sizeof(*node)); | |
1460 | INIT_LIST_HEAD(&node->private_list); | |
1da177e4 LT |
1461 | } |
1462 | ||
1463 | static __init unsigned long __maxindex(unsigned int height) | |
1464 | { | |
430d275a PL |
1465 | unsigned int width = height * RADIX_TREE_MAP_SHIFT; |
1466 | int shift = RADIX_TREE_INDEX_BITS - width; | |
1467 | ||
1468 | if (shift < 0) | |
1469 | return ~0UL; | |
1470 | if (shift >= BITS_PER_LONG) | |
1471 | return 0UL; | |
1472 | return ~0UL >> shift; | |
1da177e4 LT |
1473 | } |
1474 | ||
1475 | static __init void radix_tree_init_maxindex(void) | |
1476 | { | |
1477 | unsigned int i; | |
1478 | ||
1479 | for (i = 0; i < ARRAY_SIZE(height_to_maxindex); i++) | |
1480 | height_to_maxindex[i] = __maxindex(i); | |
1481 | } | |
1482 | ||
1da177e4 LT |
1483 | static int radix_tree_callback(struct notifier_block *nfb, |
1484 | unsigned long action, | |
1485 | void *hcpu) | |
1486 | { | |
1487 | int cpu = (long)hcpu; | |
1488 | struct radix_tree_preload *rtp; | |
9d2a8da0 | 1489 | struct radix_tree_node *node; |
1da177e4 LT |
1490 | |
1491 | /* Free per-cpu pool of perloaded nodes */ | |
8bb78442 | 1492 | if (action == CPU_DEAD || action == CPU_DEAD_FROZEN) { |
1da177e4 LT |
1493 | rtp = &per_cpu(radix_tree_preloads, cpu); |
1494 | while (rtp->nr) { | |
9d2a8da0 KS |
1495 | node = rtp->nodes; |
1496 | rtp->nodes = node->private_data; | |
1497 | kmem_cache_free(radix_tree_node_cachep, node); | |
1498 | rtp->nr--; | |
1da177e4 LT |
1499 | } |
1500 | } | |
1501 | return NOTIFY_OK; | |
1502 | } | |
1da177e4 LT |
1503 | |
1504 | void __init radix_tree_init(void) | |
1505 | { | |
1506 | radix_tree_node_cachep = kmem_cache_create("radix_tree_node", | |
1507 | sizeof(struct radix_tree_node), 0, | |
488514d1 CL |
1508 | SLAB_PANIC | SLAB_RECLAIM_ACCOUNT, |
1509 | radix_tree_node_ctor); | |
1da177e4 LT |
1510 | radix_tree_init_maxindex(); |
1511 | hotcpu_notifier(radix_tree_callback, 0); | |
1512 | } |