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 | ||
7cf19af4 MW |
176 | #if 0 |
177 | static void dump_node(void *slot, int height, int offset) | |
178 | { | |
179 | struct radix_tree_node *node; | |
180 | int i; | |
181 | ||
182 | if (!slot) | |
183 | return; | |
184 | ||
185 | if (height == 0) { | |
186 | pr_debug("radix entry %p offset %d\n", slot, offset); | |
187 | return; | |
188 | } | |
189 | ||
190 | node = indirect_to_ptr(slot); | |
191 | pr_debug("radix node: %p offset %d tags %lx %lx %lx path %x count %d parent %p\n", | |
192 | slot, offset, node->tags[0][0], node->tags[1][0], | |
193 | node->tags[2][0], node->path, node->count, node->parent); | |
194 | ||
195 | for (i = 0; i < RADIX_TREE_MAP_SIZE; i++) | |
196 | dump_node(node->slots[i], height - 1, i); | |
197 | } | |
198 | ||
199 | /* For debug */ | |
200 | static void radix_tree_dump(struct radix_tree_root *root) | |
201 | { | |
202 | pr_debug("radix root: %p height %d rnode %p tags %x\n", | |
203 | root, root->height, root->rnode, | |
204 | root->gfp_mask >> __GFP_BITS_SHIFT); | |
205 | if (!radix_tree_is_indirect_ptr(root->rnode)) | |
206 | return; | |
207 | dump_node(root->rnode, root->height, 0); | |
208 | } | |
209 | #endif | |
210 | ||
1da177e4 LT |
211 | /* |
212 | * This assumes that the caller has performed appropriate preallocation, and | |
213 | * that the caller has pinned this thread of control to the current CPU. | |
214 | */ | |
215 | static struct radix_tree_node * | |
216 | radix_tree_node_alloc(struct radix_tree_root *root) | |
217 | { | |
e2848a0e | 218 | struct radix_tree_node *ret = NULL; |
612d6c19 | 219 | gfp_t gfp_mask = root_gfp_mask(root); |
1da177e4 | 220 | |
5e4c0d97 JK |
221 | /* |
222 | * Preload code isn't irq safe and it doesn't make sence to use | |
223 | * preloading in the interrupt anyway as all the allocations have to | |
224 | * be atomic. So just do normal allocation when in interrupt. | |
225 | */ | |
d0164adc | 226 | if (!gfpflags_allow_blocking(gfp_mask) && !in_interrupt()) { |
1da177e4 LT |
227 | struct radix_tree_preload *rtp; |
228 | ||
58e698af VD |
229 | /* |
230 | * Even if the caller has preloaded, try to allocate from the | |
231 | * cache first for the new node to get accounted. | |
232 | */ | |
233 | ret = kmem_cache_alloc(radix_tree_node_cachep, | |
234 | gfp_mask | __GFP_ACCOUNT | __GFP_NOWARN); | |
235 | if (ret) | |
236 | goto out; | |
237 | ||
e2848a0e NP |
238 | /* |
239 | * Provided the caller has preloaded here, we will always | |
240 | * succeed in getting a node here (and never reach | |
241 | * kmem_cache_alloc) | |
242 | */ | |
7c8e0181 | 243 | rtp = this_cpu_ptr(&radix_tree_preloads); |
1da177e4 | 244 | if (rtp->nr) { |
9d2a8da0 KS |
245 | ret = rtp->nodes; |
246 | rtp->nodes = ret->private_data; | |
247 | ret->private_data = NULL; | |
1da177e4 LT |
248 | rtp->nr--; |
249 | } | |
ce80b067 CM |
250 | /* |
251 | * Update the allocation stack trace as this is more useful | |
252 | * for debugging. | |
253 | */ | |
254 | kmemleak_update_trace(ret); | |
58e698af | 255 | goto out; |
1da177e4 | 256 | } |
58e698af VD |
257 | ret = kmem_cache_alloc(radix_tree_node_cachep, |
258 | gfp_mask | __GFP_ACCOUNT); | |
259 | out: | |
c0bc9875 | 260 | BUG_ON(radix_tree_is_indirect_ptr(ret)); |
1da177e4 LT |
261 | return ret; |
262 | } | |
263 | ||
7cf9c2c7 NP |
264 | static void radix_tree_node_rcu_free(struct rcu_head *head) |
265 | { | |
266 | struct radix_tree_node *node = | |
267 | container_of(head, struct radix_tree_node, rcu_head); | |
b6dd0865 | 268 | int i; |
643b52b9 NP |
269 | |
270 | /* | |
271 | * must only free zeroed nodes into the slab. radix_tree_shrink | |
272 | * can leave us with a non-NULL entry in the first slot, so clear | |
273 | * that here to make sure. | |
274 | */ | |
b6dd0865 DC |
275 | for (i = 0; i < RADIX_TREE_MAX_TAGS; i++) |
276 | tag_clear(node, i, 0); | |
277 | ||
643b52b9 NP |
278 | node->slots[0] = NULL; |
279 | node->count = 0; | |
280 | ||
7cf9c2c7 NP |
281 | kmem_cache_free(radix_tree_node_cachep, node); |
282 | } | |
283 | ||
1da177e4 LT |
284 | static inline void |
285 | radix_tree_node_free(struct radix_tree_node *node) | |
286 | { | |
7cf9c2c7 | 287 | call_rcu(&node->rcu_head, radix_tree_node_rcu_free); |
1da177e4 LT |
288 | } |
289 | ||
290 | /* | |
291 | * Load up this CPU's radix_tree_node buffer with sufficient objects to | |
292 | * ensure that the addition of a single element in the tree cannot fail. On | |
293 | * success, return zero, with preemption disabled. On error, return -ENOMEM | |
294 | * with preemption not disabled. | |
b34df792 DH |
295 | * |
296 | * To make use of this facility, the radix tree must be initialised without | |
d0164adc | 297 | * __GFP_DIRECT_RECLAIM being passed to INIT_RADIX_TREE(). |
1da177e4 | 298 | */ |
5e4c0d97 | 299 | static int __radix_tree_preload(gfp_t gfp_mask) |
1da177e4 LT |
300 | { |
301 | struct radix_tree_preload *rtp; | |
302 | struct radix_tree_node *node; | |
303 | int ret = -ENOMEM; | |
304 | ||
305 | preempt_disable(); | |
7c8e0181 | 306 | rtp = this_cpu_ptr(&radix_tree_preloads); |
9d2a8da0 | 307 | while (rtp->nr < RADIX_TREE_PRELOAD_SIZE) { |
1da177e4 | 308 | preempt_enable(); |
488514d1 | 309 | node = kmem_cache_alloc(radix_tree_node_cachep, gfp_mask); |
1da177e4 LT |
310 | if (node == NULL) |
311 | goto out; | |
312 | preempt_disable(); | |
7c8e0181 | 313 | rtp = this_cpu_ptr(&radix_tree_preloads); |
9d2a8da0 KS |
314 | if (rtp->nr < RADIX_TREE_PRELOAD_SIZE) { |
315 | node->private_data = rtp->nodes; | |
316 | rtp->nodes = node; | |
317 | rtp->nr++; | |
318 | } else { | |
1da177e4 | 319 | kmem_cache_free(radix_tree_node_cachep, node); |
9d2a8da0 | 320 | } |
1da177e4 LT |
321 | } |
322 | ret = 0; | |
323 | out: | |
324 | return ret; | |
325 | } | |
5e4c0d97 JK |
326 | |
327 | /* | |
328 | * Load up this CPU's radix_tree_node buffer with sufficient objects to | |
329 | * ensure that the addition of a single element in the tree cannot fail. On | |
330 | * success, return zero, with preemption disabled. On error, return -ENOMEM | |
331 | * with preemption not disabled. | |
332 | * | |
333 | * To make use of this facility, the radix tree must be initialised without | |
d0164adc | 334 | * __GFP_DIRECT_RECLAIM being passed to INIT_RADIX_TREE(). |
5e4c0d97 JK |
335 | */ |
336 | int radix_tree_preload(gfp_t gfp_mask) | |
337 | { | |
338 | /* Warn on non-sensical use... */ | |
d0164adc | 339 | WARN_ON_ONCE(!gfpflags_allow_blocking(gfp_mask)); |
5e4c0d97 JK |
340 | return __radix_tree_preload(gfp_mask); |
341 | } | |
d7f0923d | 342 | EXPORT_SYMBOL(radix_tree_preload); |
1da177e4 | 343 | |
5e4c0d97 JK |
344 | /* |
345 | * The same as above function, except we don't guarantee preloading happens. | |
346 | * We do it, if we decide it helps. On success, return zero with preemption | |
347 | * disabled. On error, return -ENOMEM with preemption not disabled. | |
348 | */ | |
349 | int radix_tree_maybe_preload(gfp_t gfp_mask) | |
350 | { | |
d0164adc | 351 | if (gfpflags_allow_blocking(gfp_mask)) |
5e4c0d97 JK |
352 | return __radix_tree_preload(gfp_mask); |
353 | /* Preloading doesn't help anything with this gfp mask, skip it */ | |
354 | preempt_disable(); | |
355 | return 0; | |
356 | } | |
357 | EXPORT_SYMBOL(radix_tree_maybe_preload); | |
358 | ||
1da177e4 LT |
359 | /* |
360 | * Return the maximum key which can be store into a | |
361 | * radix tree with height HEIGHT. | |
362 | */ | |
363 | static inline unsigned long radix_tree_maxindex(unsigned int height) | |
364 | { | |
365 | return height_to_maxindex[height]; | |
366 | } | |
367 | ||
368 | /* | |
369 | * Extend a radix tree so it can store key @index. | |
370 | */ | |
e6145236 MW |
371 | static int radix_tree_extend(struct radix_tree_root *root, |
372 | unsigned long index, unsigned order) | |
1da177e4 LT |
373 | { |
374 | struct radix_tree_node *node; | |
e2bdb933 | 375 | struct radix_tree_node *slot; |
1da177e4 | 376 | unsigned int height; |
1da177e4 LT |
377 | int tag; |
378 | ||
379 | /* Figure out what the height should be. */ | |
380 | height = root->height + 1; | |
381 | while (index > radix_tree_maxindex(height)) | |
382 | height++; | |
383 | ||
e6145236 | 384 | if ((root->rnode == NULL) && (order == 0)) { |
1da177e4 LT |
385 | root->height = height; |
386 | goto out; | |
387 | } | |
388 | ||
1da177e4 | 389 | do { |
7cf9c2c7 | 390 | unsigned int newheight; |
1da177e4 LT |
391 | if (!(node = radix_tree_node_alloc(root))) |
392 | return -ENOMEM; | |
393 | ||
1da177e4 | 394 | /* Propagate the aggregated tag info into the new root */ |
daff89f3 | 395 | for (tag = 0; tag < RADIX_TREE_MAX_TAGS; tag++) { |
612d6c19 | 396 | if (root_tag_get(root, tag)) |
1da177e4 LT |
397 | tag_set(node, tag, 0); |
398 | } | |
399 | ||
e2bdb933 | 400 | /* Increase the height. */ |
7cf9c2c7 | 401 | newheight = root->height+1; |
449dd698 JW |
402 | BUG_ON(newheight & ~RADIX_TREE_HEIGHT_MASK); |
403 | node->path = newheight; | |
1da177e4 | 404 | node->count = 1; |
e2bdb933 HD |
405 | node->parent = NULL; |
406 | slot = root->rnode; | |
339e6353 | 407 | if (radix_tree_is_indirect_ptr(slot) && newheight > 1) { |
e2bdb933 HD |
408 | slot = indirect_to_ptr(slot); |
409 | slot->parent = node; | |
339e6353 | 410 | slot = ptr_to_indirect(slot); |
e2bdb933 HD |
411 | } |
412 | node->slots[0] = slot; | |
27d20fdd | 413 | node = ptr_to_indirect(node); |
7cf9c2c7 NP |
414 | rcu_assign_pointer(root->rnode, node); |
415 | root->height = newheight; | |
1da177e4 LT |
416 | } while (height > root->height); |
417 | out: | |
418 | return 0; | |
419 | } | |
420 | ||
421 | /** | |
139e5616 | 422 | * __radix_tree_create - create a slot in a radix tree |
1da177e4 LT |
423 | * @root: radix tree root |
424 | * @index: index key | |
e6145236 | 425 | * @order: index occupies 2^order aligned slots |
139e5616 JW |
426 | * @nodep: returns node |
427 | * @slotp: returns slot | |
1da177e4 | 428 | * |
139e5616 JW |
429 | * Create, if necessary, and return the node and slot for an item |
430 | * at position @index in the radix tree @root. | |
431 | * | |
432 | * Until there is more than one item in the tree, no nodes are | |
433 | * allocated and @root->rnode is used as a direct slot instead of | |
434 | * pointing to a node, in which case *@nodep will be NULL. | |
435 | * | |
436 | * Returns -ENOMEM, or 0 for success. | |
1da177e4 | 437 | */ |
139e5616 | 438 | int __radix_tree_create(struct radix_tree_root *root, unsigned long index, |
e6145236 MW |
439 | unsigned order, struct radix_tree_node **nodep, |
440 | void ***slotp) | |
1da177e4 | 441 | { |
201b6264 | 442 | struct radix_tree_node *node = NULL, *slot; |
139e5616 | 443 | unsigned int height, shift, offset; |
1da177e4 LT |
444 | int error; |
445 | ||
e6145236 MW |
446 | BUG_ON((0 < order) && (order < RADIX_TREE_MAP_SHIFT)); |
447 | ||
1da177e4 | 448 | /* Make sure the tree is high enough. */ |
612d6c19 | 449 | if (index > radix_tree_maxindex(root->height)) { |
e6145236 | 450 | error = radix_tree_extend(root, index, order); |
1da177e4 LT |
451 | if (error) |
452 | return error; | |
453 | } | |
454 | ||
e6145236 | 455 | slot = root->rnode; |
c0bc9875 | 456 | |
1da177e4 | 457 | height = root->height; |
0070e28d | 458 | shift = height * RADIX_TREE_MAP_SHIFT; |
1da177e4 LT |
459 | |
460 | offset = 0; /* uninitialised var warning */ | |
e6145236 | 461 | while (shift > order) { |
201b6264 | 462 | if (slot == NULL) { |
1da177e4 | 463 | /* Have to add a child node. */ |
201b6264 | 464 | if (!(slot = radix_tree_node_alloc(root))) |
1da177e4 | 465 | return -ENOMEM; |
449dd698 | 466 | slot->path = height; |
e2bdb933 | 467 | slot->parent = node; |
201b6264 | 468 | if (node) { |
339e6353 MW |
469 | rcu_assign_pointer(node->slots[offset], |
470 | ptr_to_indirect(slot)); | |
1da177e4 | 471 | node->count++; |
449dd698 | 472 | slot->path |= offset << RADIX_TREE_HEIGHT_SHIFT; |
201b6264 | 473 | } else |
339e6353 MW |
474 | rcu_assign_pointer(root->rnode, |
475 | ptr_to_indirect(slot)); | |
e6145236 MW |
476 | } else if (!radix_tree_is_indirect_ptr(slot)) |
477 | break; | |
1da177e4 LT |
478 | |
479 | /* Go a level down */ | |
e6145236 | 480 | height--; |
0070e28d | 481 | shift -= RADIX_TREE_MAP_SHIFT; |
1da177e4 | 482 | offset = (index >> shift) & RADIX_TREE_MAP_MASK; |
e6145236 | 483 | node = indirect_to_ptr(slot); |
201b6264 | 484 | slot = node->slots[offset]; |
e6145236 MW |
485 | } |
486 | ||
57578c2e | 487 | #ifdef CONFIG_RADIX_TREE_MULTIORDER |
e6145236 MW |
488 | /* Insert pointers to the canonical entry */ |
489 | if ((shift - order) > 0) { | |
490 | int i, n = 1 << (shift - order); | |
491 | offset = offset & ~(n - 1); | |
492 | slot = ptr_to_indirect(&node->slots[offset]); | |
493 | for (i = 0; i < n; i++) { | |
494 | if (node->slots[offset + i]) | |
495 | return -EEXIST; | |
496 | } | |
497 | ||
498 | for (i = 1; i < n; i++) { | |
499 | rcu_assign_pointer(node->slots[offset + i], slot); | |
500 | node->count++; | |
501 | } | |
612d6c19 | 502 | } |
57578c2e | 503 | #endif |
1da177e4 | 504 | |
139e5616 JW |
505 | if (nodep) |
506 | *nodep = node; | |
507 | if (slotp) | |
508 | *slotp = node ? node->slots + offset : (void **)&root->rnode; | |
509 | return 0; | |
510 | } | |
511 | ||
512 | /** | |
e6145236 | 513 | * __radix_tree_insert - insert into a radix tree |
139e5616 JW |
514 | * @root: radix tree root |
515 | * @index: index key | |
e6145236 | 516 | * @order: key covers the 2^order indices around index |
139e5616 JW |
517 | * @item: item to insert |
518 | * | |
519 | * Insert an item into the radix tree at position @index. | |
520 | */ | |
e6145236 MW |
521 | int __radix_tree_insert(struct radix_tree_root *root, unsigned long index, |
522 | unsigned order, void *item) | |
139e5616 JW |
523 | { |
524 | struct radix_tree_node *node; | |
525 | void **slot; | |
526 | int error; | |
527 | ||
528 | BUG_ON(radix_tree_is_indirect_ptr(item)); | |
529 | ||
e6145236 | 530 | error = __radix_tree_create(root, index, order, &node, &slot); |
139e5616 JW |
531 | if (error) |
532 | return error; | |
533 | if (*slot != NULL) | |
1da177e4 | 534 | return -EEXIST; |
139e5616 | 535 | rcu_assign_pointer(*slot, item); |
201b6264 | 536 | |
612d6c19 NP |
537 | if (node) { |
538 | node->count++; | |
139e5616 JW |
539 | BUG_ON(tag_get(node, 0, index & RADIX_TREE_MAP_MASK)); |
540 | BUG_ON(tag_get(node, 1, index & RADIX_TREE_MAP_MASK)); | |
612d6c19 | 541 | } else { |
612d6c19 NP |
542 | BUG_ON(root_tag_get(root, 0)); |
543 | BUG_ON(root_tag_get(root, 1)); | |
544 | } | |
1da177e4 | 545 | |
1da177e4 LT |
546 | return 0; |
547 | } | |
e6145236 | 548 | EXPORT_SYMBOL(__radix_tree_insert); |
1da177e4 | 549 | |
139e5616 JW |
550 | /** |
551 | * __radix_tree_lookup - lookup an item in a radix tree | |
552 | * @root: radix tree root | |
553 | * @index: index key | |
554 | * @nodep: returns node | |
555 | * @slotp: returns slot | |
556 | * | |
557 | * Lookup and return the item at position @index in the radix | |
558 | * tree @root. | |
559 | * | |
560 | * Until there is more than one item in the tree, no nodes are | |
561 | * allocated and @root->rnode is used as a direct slot instead of | |
562 | * pointing to a node, in which case *@nodep will be NULL. | |
7cf9c2c7 | 563 | */ |
139e5616 JW |
564 | void *__radix_tree_lookup(struct radix_tree_root *root, unsigned long index, |
565 | struct radix_tree_node **nodep, void ***slotp) | |
1da177e4 | 566 | { |
139e5616 | 567 | struct radix_tree_node *node, *parent; |
1da177e4 | 568 | unsigned int height, shift; |
139e5616 | 569 | void **slot; |
612d6c19 | 570 | |
2676a58c | 571 | node = rcu_dereference_raw(root->rnode); |
7cf9c2c7 | 572 | if (node == NULL) |
1da177e4 LT |
573 | return NULL; |
574 | ||
c0bc9875 | 575 | if (!radix_tree_is_indirect_ptr(node)) { |
7cf9c2c7 NP |
576 | if (index > 0) |
577 | return NULL; | |
139e5616 JW |
578 | |
579 | if (nodep) | |
580 | *nodep = NULL; | |
581 | if (slotp) | |
582 | *slotp = (void **)&root->rnode; | |
583 | return node; | |
7cf9c2c7 | 584 | } |
27d20fdd | 585 | node = indirect_to_ptr(node); |
7cf9c2c7 | 586 | |
449dd698 | 587 | height = node->path & RADIX_TREE_HEIGHT_MASK; |
7cf9c2c7 NP |
588 | if (index > radix_tree_maxindex(height)) |
589 | return NULL; | |
612d6c19 | 590 | |
1da177e4 | 591 | shift = (height-1) * RADIX_TREE_MAP_SHIFT; |
1da177e4 | 592 | |
7cf9c2c7 | 593 | do { |
139e5616 JW |
594 | parent = node; |
595 | slot = node->slots + ((index >> shift) & RADIX_TREE_MAP_MASK); | |
2676a58c | 596 | node = rcu_dereference_raw(*slot); |
7cf9c2c7 | 597 | if (node == NULL) |
1da177e4 | 598 | return NULL; |
e6145236 MW |
599 | if (!radix_tree_is_indirect_ptr(node)) |
600 | break; | |
339e6353 | 601 | node = indirect_to_ptr(node); |
1da177e4 | 602 | |
1da177e4 LT |
603 | shift -= RADIX_TREE_MAP_SHIFT; |
604 | height--; | |
7cf9c2c7 | 605 | } while (height > 0); |
1da177e4 | 606 | |
139e5616 JW |
607 | if (nodep) |
608 | *nodep = parent; | |
609 | if (slotp) | |
610 | *slotp = slot; | |
611 | return node; | |
b72b71c6 HS |
612 | } |
613 | ||
614 | /** | |
615 | * radix_tree_lookup_slot - lookup a slot in a radix tree | |
616 | * @root: radix tree root | |
617 | * @index: index key | |
618 | * | |
619 | * Returns: the slot corresponding to the position @index in the | |
620 | * radix tree @root. This is useful for update-if-exists operations. | |
621 | * | |
622 | * This function can be called under rcu_read_lock iff the slot is not | |
623 | * modified by radix_tree_replace_slot, otherwise it must be called | |
624 | * exclusive from other writers. Any dereference of the slot must be done | |
625 | * using radix_tree_deref_slot. | |
626 | */ | |
627 | void **radix_tree_lookup_slot(struct radix_tree_root *root, unsigned long index) | |
628 | { | |
139e5616 JW |
629 | void **slot; |
630 | ||
631 | if (!__radix_tree_lookup(root, index, NULL, &slot)) | |
632 | return NULL; | |
633 | return slot; | |
a4331366 | 634 | } |
a4331366 HR |
635 | EXPORT_SYMBOL(radix_tree_lookup_slot); |
636 | ||
637 | /** | |
638 | * radix_tree_lookup - perform lookup operation on a radix tree | |
639 | * @root: radix tree root | |
640 | * @index: index key | |
641 | * | |
642 | * Lookup the item at the position @index in the radix tree @root. | |
7cf9c2c7 NP |
643 | * |
644 | * This function can be called under rcu_read_lock, however the caller | |
645 | * must manage lifetimes of leaf nodes (eg. RCU may also be used to free | |
646 | * them safely). No RCU barriers are required to access or modify the | |
647 | * returned item, however. | |
a4331366 HR |
648 | */ |
649 | void *radix_tree_lookup(struct radix_tree_root *root, unsigned long index) | |
650 | { | |
139e5616 | 651 | return __radix_tree_lookup(root, index, NULL, NULL); |
1da177e4 LT |
652 | } |
653 | EXPORT_SYMBOL(radix_tree_lookup); | |
654 | ||
655 | /** | |
656 | * radix_tree_tag_set - set a tag on a radix tree node | |
657 | * @root: radix tree root | |
658 | * @index: index key | |
659 | * @tag: tag index | |
660 | * | |
daff89f3 JC |
661 | * Set the search tag (which must be < RADIX_TREE_MAX_TAGS) |
662 | * corresponding to @index in the radix tree. From | |
1da177e4 LT |
663 | * the root all the way down to the leaf node. |
664 | * | |
665 | * Returns the address of the tagged item. Setting a tag on a not-present | |
666 | * item is a bug. | |
667 | */ | |
668 | void *radix_tree_tag_set(struct radix_tree_root *root, | |
daff89f3 | 669 | unsigned long index, unsigned int tag) |
1da177e4 LT |
670 | { |
671 | unsigned int height, shift; | |
201b6264 | 672 | struct radix_tree_node *slot; |
1da177e4 LT |
673 | |
674 | height = root->height; | |
4c91c364 | 675 | BUG_ON(index > radix_tree_maxindex(height)); |
1da177e4 | 676 | |
27d20fdd | 677 | slot = indirect_to_ptr(root->rnode); |
612d6c19 | 678 | shift = (height - 1) * RADIX_TREE_MAP_SHIFT; |
1da177e4 LT |
679 | |
680 | while (height > 0) { | |
681 | int offset; | |
682 | ||
683 | offset = (index >> shift) & RADIX_TREE_MAP_MASK; | |
d5274261 NP |
684 | if (!tag_get(slot, tag, offset)) |
685 | tag_set(slot, tag, offset); | |
201b6264 CL |
686 | slot = slot->slots[offset]; |
687 | BUG_ON(slot == NULL); | |
e6145236 MW |
688 | if (!radix_tree_is_indirect_ptr(slot)) |
689 | break; | |
339e6353 | 690 | slot = indirect_to_ptr(slot); |
1da177e4 LT |
691 | shift -= RADIX_TREE_MAP_SHIFT; |
692 | height--; | |
693 | } | |
694 | ||
612d6c19 NP |
695 | /* set the root's tag bit */ |
696 | if (slot && !root_tag_get(root, tag)) | |
697 | root_tag_set(root, tag); | |
698 | ||
201b6264 | 699 | return slot; |
1da177e4 LT |
700 | } |
701 | EXPORT_SYMBOL(radix_tree_tag_set); | |
702 | ||
703 | /** | |
704 | * radix_tree_tag_clear - clear a tag on a radix tree node | |
705 | * @root: radix tree root | |
706 | * @index: index key | |
707 | * @tag: tag index | |
708 | * | |
daff89f3 JC |
709 | * Clear the search tag (which must be < RADIX_TREE_MAX_TAGS) |
710 | * corresponding to @index in the radix tree. If | |
1da177e4 LT |
711 | * this causes the leaf node to have no tags set then clear the tag in the |
712 | * next-to-leaf node, etc. | |
713 | * | |
714 | * Returns the address of the tagged item on success, else NULL. ie: | |
715 | * has the same return value and semantics as radix_tree_lookup(). | |
716 | */ | |
717 | void *radix_tree_tag_clear(struct radix_tree_root *root, | |
daff89f3 | 718 | unsigned long index, unsigned int tag) |
1da177e4 | 719 | { |
e2bdb933 | 720 | struct radix_tree_node *node = NULL; |
612d6c19 | 721 | struct radix_tree_node *slot = NULL; |
1da177e4 | 722 | unsigned int height, shift; |
e2bdb933 | 723 | int uninitialized_var(offset); |
1da177e4 LT |
724 | |
725 | height = root->height; | |
726 | if (index > radix_tree_maxindex(height)) | |
727 | goto out; | |
728 | ||
e2bdb933 | 729 | shift = height * RADIX_TREE_MAP_SHIFT; |
339e6353 | 730 | slot = root->rnode; |
1da177e4 | 731 | |
e2bdb933 | 732 | while (shift) { |
201b6264 | 733 | if (slot == NULL) |
1da177e4 | 734 | goto out; |
e6145236 MW |
735 | if (!radix_tree_is_indirect_ptr(slot)) |
736 | break; | |
339e6353 | 737 | slot = indirect_to_ptr(slot); |
1da177e4 | 738 | |
e2bdb933 | 739 | shift -= RADIX_TREE_MAP_SHIFT; |
1da177e4 | 740 | offset = (index >> shift) & RADIX_TREE_MAP_MASK; |
e2bdb933 | 741 | node = slot; |
201b6264 | 742 | slot = slot->slots[offset]; |
1da177e4 LT |
743 | } |
744 | ||
612d6c19 | 745 | if (slot == NULL) |
1da177e4 LT |
746 | goto out; |
747 | ||
e2bdb933 HD |
748 | while (node) { |
749 | if (!tag_get(node, tag, offset)) | |
d5274261 | 750 | goto out; |
e2bdb933 HD |
751 | tag_clear(node, tag, offset); |
752 | if (any_tag_set(node, tag)) | |
6e954b9e | 753 | goto out; |
e2bdb933 HD |
754 | |
755 | index >>= RADIX_TREE_MAP_SHIFT; | |
756 | offset = index & RADIX_TREE_MAP_MASK; | |
757 | node = node->parent; | |
612d6c19 NP |
758 | } |
759 | ||
760 | /* clear the root's tag bit */ | |
761 | if (root_tag_get(root, tag)) | |
762 | root_tag_clear(root, tag); | |
763 | ||
1da177e4 | 764 | out: |
612d6c19 | 765 | return slot; |
1da177e4 LT |
766 | } |
767 | EXPORT_SYMBOL(radix_tree_tag_clear); | |
768 | ||
1da177e4 | 769 | /** |
32605a18 MT |
770 | * radix_tree_tag_get - get a tag on a radix tree node |
771 | * @root: radix tree root | |
772 | * @index: index key | |
daff89f3 | 773 | * @tag: tag index (< RADIX_TREE_MAX_TAGS) |
1da177e4 | 774 | * |
32605a18 | 775 | * Return values: |
1da177e4 | 776 | * |
612d6c19 NP |
777 | * 0: tag not present or not set |
778 | * 1: tag set | |
ce82653d DH |
779 | * |
780 | * Note that the return value of this function may not be relied on, even if | |
781 | * the RCU lock is held, unless tag modification and node deletion are excluded | |
782 | * from concurrency. | |
1da177e4 LT |
783 | */ |
784 | int radix_tree_tag_get(struct radix_tree_root *root, | |
daff89f3 | 785 | unsigned long index, unsigned int tag) |
1da177e4 LT |
786 | { |
787 | unsigned int height, shift; | |
7cf9c2c7 | 788 | struct radix_tree_node *node; |
1da177e4 | 789 | |
612d6c19 NP |
790 | /* check the root's tag bit */ |
791 | if (!root_tag_get(root, tag)) | |
792 | return 0; | |
793 | ||
2676a58c | 794 | node = rcu_dereference_raw(root->rnode); |
7cf9c2c7 NP |
795 | if (node == NULL) |
796 | return 0; | |
797 | ||
c0bc9875 | 798 | if (!radix_tree_is_indirect_ptr(node)) |
7cf9c2c7 | 799 | return (index == 0); |
27d20fdd | 800 | node = indirect_to_ptr(node); |
7cf9c2c7 | 801 | |
449dd698 | 802 | height = node->path & RADIX_TREE_HEIGHT_MASK; |
7cf9c2c7 NP |
803 | if (index > radix_tree_maxindex(height)) |
804 | return 0; | |
612d6c19 | 805 | |
1da177e4 | 806 | shift = (height - 1) * RADIX_TREE_MAP_SHIFT; |
1da177e4 LT |
807 | |
808 | for ( ; ; ) { | |
809 | int offset; | |
810 | ||
7cf9c2c7 | 811 | if (node == NULL) |
1da177e4 | 812 | return 0; |
339e6353 | 813 | node = indirect_to_ptr(node); |
1da177e4 LT |
814 | |
815 | offset = (index >> shift) & RADIX_TREE_MAP_MASK; | |
7cf9c2c7 | 816 | if (!tag_get(node, tag, offset)) |
3fa36acb | 817 | return 0; |
ce82653d | 818 | if (height == 1) |
3fa36acb | 819 | return 1; |
2676a58c | 820 | node = rcu_dereference_raw(node->slots[offset]); |
e6145236 MW |
821 | if (!radix_tree_is_indirect_ptr(node)) |
822 | return 1; | |
1da177e4 LT |
823 | shift -= RADIX_TREE_MAP_SHIFT; |
824 | height--; | |
825 | } | |
826 | } | |
827 | EXPORT_SYMBOL(radix_tree_tag_get); | |
1da177e4 | 828 | |
78c1d784 KK |
829 | /** |
830 | * radix_tree_next_chunk - find next chunk of slots for iteration | |
831 | * | |
832 | * @root: radix tree root | |
833 | * @iter: iterator state | |
834 | * @flags: RADIX_TREE_ITER_* flags and tag index | |
835 | * Returns: pointer to chunk first slot, or NULL if iteration is over | |
836 | */ | |
837 | void **radix_tree_next_chunk(struct radix_tree_root *root, | |
838 | struct radix_tree_iter *iter, unsigned flags) | |
839 | { | |
840 | unsigned shift, tag = flags & RADIX_TREE_ITER_TAG_MASK; | |
841 | struct radix_tree_node *rnode, *node; | |
449dd698 | 842 | unsigned long index, offset, height; |
78c1d784 KK |
843 | |
844 | if ((flags & RADIX_TREE_ITER_TAGGED) && !root_tag_get(root, tag)) | |
845 | return NULL; | |
846 | ||
847 | /* | |
848 | * Catch next_index overflow after ~0UL. iter->index never overflows | |
849 | * during iterating; it can be zero only at the beginning. | |
850 | * And we cannot overflow iter->next_index in a single step, | |
851 | * because RADIX_TREE_MAP_SHIFT < BITS_PER_LONG. | |
fffaee36 KK |
852 | * |
853 | * This condition also used by radix_tree_next_slot() to stop | |
854 | * contiguous iterating, and forbid swithing to the next chunk. | |
78c1d784 KK |
855 | */ |
856 | index = iter->next_index; | |
857 | if (!index && iter->index) | |
858 | return NULL; | |
859 | ||
860 | rnode = rcu_dereference_raw(root->rnode); | |
861 | if (radix_tree_is_indirect_ptr(rnode)) { | |
862 | rnode = indirect_to_ptr(rnode); | |
863 | } else if (rnode && !index) { | |
864 | /* Single-slot tree */ | |
865 | iter->index = 0; | |
866 | iter->next_index = 1; | |
867 | iter->tags = 1; | |
868 | return (void **)&root->rnode; | |
869 | } else | |
870 | return NULL; | |
871 | ||
872 | restart: | |
449dd698 JW |
873 | height = rnode->path & RADIX_TREE_HEIGHT_MASK; |
874 | shift = (height - 1) * RADIX_TREE_MAP_SHIFT; | |
78c1d784 KK |
875 | offset = index >> shift; |
876 | ||
877 | /* Index outside of the tree */ | |
878 | if (offset >= RADIX_TREE_MAP_SIZE) | |
879 | return NULL; | |
880 | ||
881 | node = rnode; | |
882 | while (1) { | |
e6145236 | 883 | struct radix_tree_node *slot; |
78c1d784 KK |
884 | if ((flags & RADIX_TREE_ITER_TAGGED) ? |
885 | !test_bit(offset, node->tags[tag]) : | |
886 | !node->slots[offset]) { | |
887 | /* Hole detected */ | |
888 | if (flags & RADIX_TREE_ITER_CONTIG) | |
889 | return NULL; | |
890 | ||
891 | if (flags & RADIX_TREE_ITER_TAGGED) | |
892 | offset = radix_tree_find_next_bit( | |
893 | node->tags[tag], | |
894 | RADIX_TREE_MAP_SIZE, | |
895 | offset + 1); | |
896 | else | |
897 | while (++offset < RADIX_TREE_MAP_SIZE) { | |
898 | if (node->slots[offset]) | |
899 | break; | |
900 | } | |
901 | index &= ~((RADIX_TREE_MAP_SIZE << shift) - 1); | |
902 | index += offset << shift; | |
903 | /* Overflow after ~0UL */ | |
904 | if (!index) | |
905 | return NULL; | |
906 | if (offset == RADIX_TREE_MAP_SIZE) | |
907 | goto restart; | |
908 | } | |
909 | ||
910 | /* This is leaf-node */ | |
911 | if (!shift) | |
912 | break; | |
913 | ||
e6145236 MW |
914 | slot = rcu_dereference_raw(node->slots[offset]); |
915 | if (slot == NULL) | |
78c1d784 | 916 | goto restart; |
e6145236 MW |
917 | if (!radix_tree_is_indirect_ptr(slot)) |
918 | break; | |
919 | node = indirect_to_ptr(slot); | |
78c1d784 KK |
920 | shift -= RADIX_TREE_MAP_SHIFT; |
921 | offset = (index >> shift) & RADIX_TREE_MAP_MASK; | |
922 | } | |
923 | ||
924 | /* Update the iterator state */ | |
925 | iter->index = index; | |
926 | iter->next_index = (index | RADIX_TREE_MAP_MASK) + 1; | |
927 | ||
928 | /* Construct iter->tags bit-mask from node->tags[tag] array */ | |
929 | if (flags & RADIX_TREE_ITER_TAGGED) { | |
930 | unsigned tag_long, tag_bit; | |
931 | ||
932 | tag_long = offset / BITS_PER_LONG; | |
933 | tag_bit = offset % BITS_PER_LONG; | |
934 | iter->tags = node->tags[tag][tag_long] >> tag_bit; | |
935 | /* This never happens if RADIX_TREE_TAG_LONGS == 1 */ | |
936 | if (tag_long < RADIX_TREE_TAG_LONGS - 1) { | |
937 | /* Pick tags from next element */ | |
938 | if (tag_bit) | |
939 | iter->tags |= node->tags[tag][tag_long + 1] << | |
940 | (BITS_PER_LONG - tag_bit); | |
941 | /* Clip chunk size, here only BITS_PER_LONG tags */ | |
942 | iter->next_index = index + BITS_PER_LONG; | |
943 | } | |
944 | } | |
945 | ||
946 | return node->slots + offset; | |
947 | } | |
948 | EXPORT_SYMBOL(radix_tree_next_chunk); | |
949 | ||
ebf8aa44 JK |
950 | /** |
951 | * radix_tree_range_tag_if_tagged - for each item in given range set given | |
952 | * tag if item has another tag set | |
953 | * @root: radix tree root | |
954 | * @first_indexp: pointer to a starting index of a range to scan | |
955 | * @last_index: last index of a range to scan | |
956 | * @nr_to_tag: maximum number items to tag | |
957 | * @iftag: tag index to test | |
958 | * @settag: tag index to set if tested tag is set | |
959 | * | |
960 | * This function scans range of radix tree from first_index to last_index | |
961 | * (inclusive). For each item in the range if iftag is set, the function sets | |
962 | * also settag. The function stops either after tagging nr_to_tag items or | |
963 | * after reaching last_index. | |
964 | * | |
144dcfc0 DC |
965 | * The tags must be set from the leaf level only and propagated back up the |
966 | * path to the root. We must do this so that we resolve the full path before | |
967 | * setting any tags on intermediate nodes. If we set tags as we descend, then | |
968 | * we can get to the leaf node and find that the index that has the iftag | |
969 | * set is outside the range we are scanning. This reults in dangling tags and | |
970 | * can lead to problems with later tag operations (e.g. livelocks on lookups). | |
971 | * | |
ebf8aa44 JK |
972 | * The function returns number of leaves where the tag was set and sets |
973 | * *first_indexp to the first unscanned index. | |
d5ed3a4a JK |
974 | * WARNING! *first_indexp can wrap if last_index is ULONG_MAX. Caller must |
975 | * be prepared to handle that. | |
ebf8aa44 JK |
976 | */ |
977 | unsigned long radix_tree_range_tag_if_tagged(struct radix_tree_root *root, | |
978 | unsigned long *first_indexp, unsigned long last_index, | |
979 | unsigned long nr_to_tag, | |
980 | unsigned int iftag, unsigned int settag) | |
981 | { | |
144dcfc0 | 982 | unsigned int height = root->height; |
e2bdb933 | 983 | struct radix_tree_node *node = NULL; |
144dcfc0 DC |
984 | struct radix_tree_node *slot; |
985 | unsigned int shift; | |
986 | unsigned long tagged = 0; | |
987 | unsigned long index = *first_indexp; | |
ebf8aa44 JK |
988 | |
989 | last_index = min(last_index, radix_tree_maxindex(height)); | |
990 | if (index > last_index) | |
991 | return 0; | |
992 | if (!nr_to_tag) | |
993 | return 0; | |
994 | if (!root_tag_get(root, iftag)) { | |
995 | *first_indexp = last_index + 1; | |
996 | return 0; | |
997 | } | |
998 | if (height == 0) { | |
999 | *first_indexp = last_index + 1; | |
1000 | root_tag_set(root, settag); | |
1001 | return 1; | |
1002 | } | |
1003 | ||
1004 | shift = (height - 1) * RADIX_TREE_MAP_SHIFT; | |
27d20fdd | 1005 | slot = indirect_to_ptr(root->rnode); |
ebf8aa44 JK |
1006 | |
1007 | for (;;) { | |
e2bdb933 | 1008 | unsigned long upindex; |
ebf8aa44 JK |
1009 | int offset; |
1010 | ||
1011 | offset = (index >> shift) & RADIX_TREE_MAP_MASK; | |
1012 | if (!slot->slots[offset]) | |
1013 | goto next; | |
1014 | if (!tag_get(slot, iftag, offset)) | |
1015 | goto next; | |
e2bdb933 | 1016 | if (shift) { |
e2bdb933 | 1017 | node = slot; |
144dcfc0 | 1018 | slot = slot->slots[offset]; |
e6145236 MW |
1019 | if (radix_tree_is_indirect_ptr(slot)) { |
1020 | slot = indirect_to_ptr(slot); | |
1021 | shift -= RADIX_TREE_MAP_SHIFT; | |
1022 | continue; | |
1023 | } else { | |
1024 | slot = node; | |
1025 | node = node->parent; | |
1026 | } | |
144dcfc0 DC |
1027 | } |
1028 | ||
1029 | /* tag the leaf */ | |
e6145236 | 1030 | tagged += 1 << shift; |
ebf8aa44 | 1031 | tag_set(slot, settag, offset); |
144dcfc0 DC |
1032 | |
1033 | /* walk back up the path tagging interior nodes */ | |
e2bdb933 HD |
1034 | upindex = index; |
1035 | while (node) { | |
1036 | upindex >>= RADIX_TREE_MAP_SHIFT; | |
1037 | offset = upindex & RADIX_TREE_MAP_MASK; | |
1038 | ||
144dcfc0 | 1039 | /* stop if we find a node with the tag already set */ |
e2bdb933 | 1040 | if (tag_get(node, settag, offset)) |
144dcfc0 | 1041 | break; |
e2bdb933 HD |
1042 | tag_set(node, settag, offset); |
1043 | node = node->parent; | |
ebf8aa44 | 1044 | } |
144dcfc0 | 1045 | |
e2bdb933 HD |
1046 | /* |
1047 | * Small optimization: now clear that node pointer. | |
1048 | * Since all of this slot's ancestors now have the tag set | |
1049 | * from setting it above, we have no further need to walk | |
1050 | * back up the tree setting tags, until we update slot to | |
1051 | * point to another radix_tree_node. | |
1052 | */ | |
1053 | node = NULL; | |
1054 | ||
ebf8aa44 JK |
1055 | next: |
1056 | /* Go to next item at level determined by 'shift' */ | |
1057 | index = ((index >> shift) + 1) << shift; | |
d5ed3a4a JK |
1058 | /* Overflow can happen when last_index is ~0UL... */ |
1059 | if (index > last_index || !index) | |
ebf8aa44 JK |
1060 | break; |
1061 | if (tagged >= nr_to_tag) | |
1062 | break; | |
1063 | while (((index >> shift) & RADIX_TREE_MAP_MASK) == 0) { | |
1064 | /* | |
1065 | * We've fully scanned this node. Go up. Because | |
1066 | * last_index is guaranteed to be in the tree, what | |
1067 | * we do below cannot wander astray. | |
1068 | */ | |
e2bdb933 | 1069 | slot = slot->parent; |
ebf8aa44 JK |
1070 | shift += RADIX_TREE_MAP_SHIFT; |
1071 | } | |
1072 | } | |
1073 | /* | |
ac15ee69 TO |
1074 | * We need not to tag the root tag if there is no tag which is set with |
1075 | * settag within the range from *first_indexp to last_index. | |
ebf8aa44 | 1076 | */ |
ac15ee69 TO |
1077 | if (tagged > 0) |
1078 | root_tag_set(root, settag); | |
ebf8aa44 JK |
1079 | *first_indexp = index; |
1080 | ||
1081 | return tagged; | |
1082 | } | |
1083 | EXPORT_SYMBOL(radix_tree_range_tag_if_tagged); | |
1084 | ||
1da177e4 LT |
1085 | /** |
1086 | * radix_tree_gang_lookup - perform multiple lookup on a radix tree | |
1087 | * @root: radix tree root | |
1088 | * @results: where the results of the lookup are placed | |
1089 | * @first_index: start the lookup from this key | |
1090 | * @max_items: place up to this many items at *results | |
1091 | * | |
1092 | * Performs an index-ascending scan of the tree for present items. Places | |
1093 | * them at *@results and returns the number of items which were placed at | |
1094 | * *@results. | |
1095 | * | |
1096 | * The implementation is naive. | |
7cf9c2c7 NP |
1097 | * |
1098 | * Like radix_tree_lookup, radix_tree_gang_lookup may be called under | |
1099 | * rcu_read_lock. In this case, rather than the returned results being | |
1100 | * an atomic snapshot of the tree at a single point in time, the semantics | |
1101 | * of an RCU protected gang lookup are as though multiple radix_tree_lookups | |
1102 | * have been issued in individual locks, and results stored in 'results'. | |
1da177e4 LT |
1103 | */ |
1104 | unsigned int | |
1105 | radix_tree_gang_lookup(struct radix_tree_root *root, void **results, | |
1106 | unsigned long first_index, unsigned int max_items) | |
1107 | { | |
cebbd29e KK |
1108 | struct radix_tree_iter iter; |
1109 | void **slot; | |
1110 | unsigned int ret = 0; | |
7cf9c2c7 | 1111 | |
cebbd29e | 1112 | if (unlikely(!max_items)) |
7cf9c2c7 | 1113 | return 0; |
1da177e4 | 1114 | |
cebbd29e | 1115 | radix_tree_for_each_slot(slot, root, &iter, first_index) { |
46437f9a | 1116 | results[ret] = rcu_dereference_raw(*slot); |
cebbd29e KK |
1117 | if (!results[ret]) |
1118 | continue; | |
46437f9a MW |
1119 | if (radix_tree_is_indirect_ptr(results[ret])) { |
1120 | slot = radix_tree_iter_retry(&iter); | |
1121 | continue; | |
1122 | } | |
cebbd29e | 1123 | if (++ret == max_items) |
1da177e4 | 1124 | break; |
1da177e4 | 1125 | } |
7cf9c2c7 | 1126 | |
1da177e4 LT |
1127 | return ret; |
1128 | } | |
1129 | EXPORT_SYMBOL(radix_tree_gang_lookup); | |
1130 | ||
47feff2c NP |
1131 | /** |
1132 | * radix_tree_gang_lookup_slot - perform multiple slot lookup on radix tree | |
1133 | * @root: radix tree root | |
1134 | * @results: where the results of the lookup are placed | |
6328650b | 1135 | * @indices: where their indices should be placed (but usually NULL) |
47feff2c NP |
1136 | * @first_index: start the lookup from this key |
1137 | * @max_items: place up to this many items at *results | |
1138 | * | |
1139 | * Performs an index-ascending scan of the tree for present items. Places | |
1140 | * their slots at *@results and returns the number of items which were | |
1141 | * placed at *@results. | |
1142 | * | |
1143 | * The implementation is naive. | |
1144 | * | |
1145 | * Like radix_tree_gang_lookup as far as RCU and locking goes. Slots must | |
1146 | * be dereferenced with radix_tree_deref_slot, and if using only RCU | |
1147 | * protection, radix_tree_deref_slot may fail requiring a retry. | |
1148 | */ | |
1149 | unsigned int | |
6328650b HD |
1150 | radix_tree_gang_lookup_slot(struct radix_tree_root *root, |
1151 | void ***results, unsigned long *indices, | |
47feff2c NP |
1152 | unsigned long first_index, unsigned int max_items) |
1153 | { | |
cebbd29e KK |
1154 | struct radix_tree_iter iter; |
1155 | void **slot; | |
1156 | unsigned int ret = 0; | |
47feff2c | 1157 | |
cebbd29e | 1158 | if (unlikely(!max_items)) |
47feff2c NP |
1159 | return 0; |
1160 | ||
cebbd29e KK |
1161 | radix_tree_for_each_slot(slot, root, &iter, first_index) { |
1162 | results[ret] = slot; | |
6328650b | 1163 | if (indices) |
cebbd29e KK |
1164 | indices[ret] = iter.index; |
1165 | if (++ret == max_items) | |
47feff2c | 1166 | break; |
47feff2c NP |
1167 | } |
1168 | ||
1169 | return ret; | |
1170 | } | |
1171 | EXPORT_SYMBOL(radix_tree_gang_lookup_slot); | |
1172 | ||
1da177e4 LT |
1173 | /** |
1174 | * radix_tree_gang_lookup_tag - perform multiple lookup on a radix tree | |
1175 | * based on a tag | |
1176 | * @root: radix tree root | |
1177 | * @results: where the results of the lookup are placed | |
1178 | * @first_index: start the lookup from this key | |
1179 | * @max_items: place up to this many items at *results | |
daff89f3 | 1180 | * @tag: the tag index (< RADIX_TREE_MAX_TAGS) |
1da177e4 LT |
1181 | * |
1182 | * Performs an index-ascending scan of the tree for present items which | |
1183 | * have the tag indexed by @tag set. Places the items at *@results and | |
1184 | * returns the number of items which were placed at *@results. | |
1185 | */ | |
1186 | unsigned int | |
1187 | radix_tree_gang_lookup_tag(struct radix_tree_root *root, void **results, | |
daff89f3 JC |
1188 | unsigned long first_index, unsigned int max_items, |
1189 | unsigned int tag) | |
1da177e4 | 1190 | { |
cebbd29e KK |
1191 | struct radix_tree_iter iter; |
1192 | void **slot; | |
1193 | unsigned int ret = 0; | |
612d6c19 | 1194 | |
cebbd29e | 1195 | if (unlikely(!max_items)) |
7cf9c2c7 NP |
1196 | return 0; |
1197 | ||
cebbd29e | 1198 | radix_tree_for_each_tagged(slot, root, &iter, first_index, tag) { |
46437f9a | 1199 | results[ret] = rcu_dereference_raw(*slot); |
cebbd29e KK |
1200 | if (!results[ret]) |
1201 | continue; | |
46437f9a MW |
1202 | if (radix_tree_is_indirect_ptr(results[ret])) { |
1203 | slot = radix_tree_iter_retry(&iter); | |
1204 | continue; | |
1205 | } | |
cebbd29e | 1206 | if (++ret == max_items) |
1da177e4 | 1207 | break; |
1da177e4 | 1208 | } |
7cf9c2c7 | 1209 | |
1da177e4 LT |
1210 | return ret; |
1211 | } | |
1212 | EXPORT_SYMBOL(radix_tree_gang_lookup_tag); | |
1213 | ||
47feff2c NP |
1214 | /** |
1215 | * radix_tree_gang_lookup_tag_slot - perform multiple slot lookup on a | |
1216 | * radix tree based on a tag | |
1217 | * @root: radix tree root | |
1218 | * @results: where the results of the lookup are placed | |
1219 | * @first_index: start the lookup from this key | |
1220 | * @max_items: place up to this many items at *results | |
1221 | * @tag: the tag index (< RADIX_TREE_MAX_TAGS) | |
1222 | * | |
1223 | * Performs an index-ascending scan of the tree for present items which | |
1224 | * have the tag indexed by @tag set. Places the slots at *@results and | |
1225 | * returns the number of slots which were placed at *@results. | |
1226 | */ | |
1227 | unsigned int | |
1228 | radix_tree_gang_lookup_tag_slot(struct radix_tree_root *root, void ***results, | |
1229 | unsigned long first_index, unsigned int max_items, | |
1230 | unsigned int tag) | |
1231 | { | |
cebbd29e KK |
1232 | struct radix_tree_iter iter; |
1233 | void **slot; | |
1234 | unsigned int ret = 0; | |
47feff2c | 1235 | |
cebbd29e | 1236 | if (unlikely(!max_items)) |
47feff2c NP |
1237 | return 0; |
1238 | ||
cebbd29e KK |
1239 | radix_tree_for_each_tagged(slot, root, &iter, first_index, tag) { |
1240 | results[ret] = slot; | |
1241 | if (++ret == max_items) | |
47feff2c | 1242 | break; |
47feff2c NP |
1243 | } |
1244 | ||
1245 | return ret; | |
1246 | } | |
1247 | EXPORT_SYMBOL(radix_tree_gang_lookup_tag_slot); | |
1248 | ||
e504f3fd HD |
1249 | #if defined(CONFIG_SHMEM) && defined(CONFIG_SWAP) |
1250 | #include <linux/sched.h> /* for cond_resched() */ | |
1251 | ||
1252 | /* | |
1253 | * This linear search is at present only useful to shmem_unuse_inode(). | |
1254 | */ | |
1255 | static unsigned long __locate(struct radix_tree_node *slot, void *item, | |
1256 | unsigned long index, unsigned long *found_index) | |
1257 | { | |
1258 | unsigned int shift, height; | |
1259 | unsigned long i; | |
1260 | ||
449dd698 | 1261 | height = slot->path & RADIX_TREE_HEIGHT_MASK; |
e504f3fd HD |
1262 | shift = (height-1) * RADIX_TREE_MAP_SHIFT; |
1263 | ||
1264 | for ( ; height > 1; height--) { | |
1265 | i = (index >> shift) & RADIX_TREE_MAP_MASK; | |
1266 | for (;;) { | |
1267 | if (slot->slots[i] != NULL) | |
1268 | break; | |
1269 | index &= ~((1UL << shift) - 1); | |
1270 | index += 1UL << shift; | |
1271 | if (index == 0) | |
1272 | goto out; /* 32-bit wraparound */ | |
1273 | i++; | |
1274 | if (i == RADIX_TREE_MAP_SIZE) | |
1275 | goto out; | |
1276 | } | |
1277 | ||
e504f3fd HD |
1278 | slot = rcu_dereference_raw(slot->slots[i]); |
1279 | if (slot == NULL) | |
1280 | goto out; | |
e6145236 MW |
1281 | if (!radix_tree_is_indirect_ptr(slot)) { |
1282 | if (slot == item) { | |
1283 | *found_index = index + i; | |
1284 | index = 0; | |
1285 | } else { | |
1286 | index += shift; | |
1287 | } | |
1288 | goto out; | |
1289 | } | |
339e6353 | 1290 | slot = indirect_to_ptr(slot); |
e6145236 | 1291 | shift -= RADIX_TREE_MAP_SHIFT; |
e504f3fd HD |
1292 | } |
1293 | ||
1294 | /* Bottom level: check items */ | |
1295 | for (i = 0; i < RADIX_TREE_MAP_SIZE; i++) { | |
1296 | if (slot->slots[i] == item) { | |
1297 | *found_index = index + i; | |
1298 | index = 0; | |
1299 | goto out; | |
1300 | } | |
1301 | } | |
1302 | index += RADIX_TREE_MAP_SIZE; | |
1303 | out: | |
1304 | return index; | |
1305 | } | |
1306 | ||
1307 | /** | |
1308 | * radix_tree_locate_item - search through radix tree for item | |
1309 | * @root: radix tree root | |
1310 | * @item: item to be found | |
1311 | * | |
1312 | * Returns index where item was found, or -1 if not found. | |
1313 | * Caller must hold no lock (since this time-consuming function needs | |
1314 | * to be preemptible), and must check afterwards if item is still there. | |
1315 | */ | |
1316 | unsigned long radix_tree_locate_item(struct radix_tree_root *root, void *item) | |
1317 | { | |
1318 | struct radix_tree_node *node; | |
1319 | unsigned long max_index; | |
1320 | unsigned long cur_index = 0; | |
1321 | unsigned long found_index = -1; | |
1322 | ||
1323 | do { | |
1324 | rcu_read_lock(); | |
1325 | node = rcu_dereference_raw(root->rnode); | |
1326 | if (!radix_tree_is_indirect_ptr(node)) { | |
1327 | rcu_read_unlock(); | |
1328 | if (node == item) | |
1329 | found_index = 0; | |
1330 | break; | |
1331 | } | |
1332 | ||
1333 | node = indirect_to_ptr(node); | |
449dd698 JW |
1334 | max_index = radix_tree_maxindex(node->path & |
1335 | RADIX_TREE_HEIGHT_MASK); | |
5f30fc94 HD |
1336 | if (cur_index > max_index) { |
1337 | rcu_read_unlock(); | |
e504f3fd | 1338 | break; |
5f30fc94 | 1339 | } |
e504f3fd HD |
1340 | |
1341 | cur_index = __locate(node, item, cur_index, &found_index); | |
1342 | rcu_read_unlock(); | |
1343 | cond_resched(); | |
1344 | } while (cur_index != 0 && cur_index <= max_index); | |
1345 | ||
1346 | return found_index; | |
1347 | } | |
1348 | #else | |
1349 | unsigned long radix_tree_locate_item(struct radix_tree_root *root, void *item) | |
1350 | { | |
1351 | return -1; | |
1352 | } | |
1353 | #endif /* CONFIG_SHMEM && CONFIG_SWAP */ | |
47feff2c | 1354 | |
a5f51c96 NP |
1355 | /** |
1356 | * radix_tree_shrink - shrink height of a radix tree to minimal | |
1357 | * @root radix tree root | |
1358 | */ | |
1359 | static inline void radix_tree_shrink(struct radix_tree_root *root) | |
1360 | { | |
1361 | /* try to shrink tree height */ | |
c0bc9875 | 1362 | while (root->height > 0) { |
a5f51c96 | 1363 | struct radix_tree_node *to_free = root->rnode; |
e2bdb933 | 1364 | struct radix_tree_node *slot; |
a5f51c96 | 1365 | |
c0bc9875 | 1366 | BUG_ON(!radix_tree_is_indirect_ptr(to_free)); |
27d20fdd | 1367 | to_free = indirect_to_ptr(to_free); |
c0bc9875 NP |
1368 | |
1369 | /* | |
1370 | * The candidate node has more than one child, or its child | |
e6145236 MW |
1371 | * is not at the leftmost slot, or it is a multiorder entry, |
1372 | * we cannot shrink. | |
c0bc9875 NP |
1373 | */ |
1374 | if (to_free->count != 1) | |
1375 | break; | |
339e6353 MW |
1376 | slot = to_free->slots[0]; |
1377 | if (!slot) | |
c0bc9875 NP |
1378 | break; |
1379 | ||
7cf9c2c7 NP |
1380 | /* |
1381 | * We don't need rcu_assign_pointer(), since we are simply | |
27d20fdd NP |
1382 | * moving the node from one part of the tree to another: if it |
1383 | * was safe to dereference the old pointer to it | |
7cf9c2c7 | 1384 | * (to_free->slots[0]), it will be safe to dereference the new |
27d20fdd | 1385 | * one (root->rnode) as far as dependent read barriers go. |
7cf9c2c7 | 1386 | */ |
e2bdb933 | 1387 | if (root->height > 1) { |
e6145236 MW |
1388 | if (!radix_tree_is_indirect_ptr(slot)) |
1389 | break; | |
1390 | ||
339e6353 | 1391 | slot = indirect_to_ptr(slot); |
e2bdb933 HD |
1392 | slot->parent = NULL; |
1393 | slot = ptr_to_indirect(slot); | |
1394 | } | |
1395 | root->rnode = slot; | |
a5f51c96 | 1396 | root->height--; |
27d20fdd NP |
1397 | |
1398 | /* | |
1399 | * We have a dilemma here. The node's slot[0] must not be | |
1400 | * NULLed in case there are concurrent lookups expecting to | |
1401 | * find the item. However if this was a bottom-level node, | |
1402 | * then it may be subject to the slot pointer being visible | |
1403 | * to callers dereferencing it. If item corresponding to | |
1404 | * slot[0] is subsequently deleted, these callers would expect | |
1405 | * their slot to become empty sooner or later. | |
1406 | * | |
1407 | * For example, lockless pagecache will look up a slot, deref | |
1408 | * the page pointer, and if the page is 0 refcount it means it | |
1409 | * was concurrently deleted from pagecache so try the deref | |
1410 | * again. Fortunately there is already a requirement for logic | |
1411 | * to retry the entire slot lookup -- the indirect pointer | |
1412 | * problem (replacing direct root node with an indirect pointer | |
1413 | * also results in a stale slot). So tag the slot as indirect | |
1414 | * to force callers to retry. | |
1415 | */ | |
1416 | if (root->height == 0) | |
1417 | *((unsigned long *)&to_free->slots[0]) |= | |
1418 | RADIX_TREE_INDIRECT_PTR; | |
1419 | ||
a5f51c96 NP |
1420 | radix_tree_node_free(to_free); |
1421 | } | |
1422 | } | |
1423 | ||
139e5616 JW |
1424 | /** |
1425 | * __radix_tree_delete_node - try to free node after clearing a slot | |
1426 | * @root: radix tree root | |
139e5616 JW |
1427 | * @node: node containing @index |
1428 | * | |
1429 | * After clearing the slot at @index in @node from radix tree | |
1430 | * rooted at @root, call this function to attempt freeing the | |
1431 | * node and shrinking the tree. | |
1432 | * | |
1433 | * Returns %true if @node was freed, %false otherwise. | |
1434 | */ | |
449dd698 | 1435 | bool __radix_tree_delete_node(struct radix_tree_root *root, |
139e5616 JW |
1436 | struct radix_tree_node *node) |
1437 | { | |
1438 | bool deleted = false; | |
1439 | ||
1440 | do { | |
1441 | struct radix_tree_node *parent; | |
1442 | ||
1443 | if (node->count) { | |
1444 | if (node == indirect_to_ptr(root->rnode)) { | |
1445 | radix_tree_shrink(root); | |
1446 | if (root->height == 0) | |
1447 | deleted = true; | |
1448 | } | |
1449 | return deleted; | |
1450 | } | |
1451 | ||
1452 | parent = node->parent; | |
1453 | if (parent) { | |
449dd698 | 1454 | unsigned int offset; |
139e5616 | 1455 | |
449dd698 JW |
1456 | offset = node->path >> RADIX_TREE_HEIGHT_SHIFT; |
1457 | parent->slots[offset] = NULL; | |
139e5616 JW |
1458 | parent->count--; |
1459 | } else { | |
1460 | root_tag_clear_all(root); | |
1461 | root->height = 0; | |
1462 | root->rnode = NULL; | |
1463 | } | |
1464 | ||
1465 | radix_tree_node_free(node); | |
1466 | deleted = true; | |
1467 | ||
1468 | node = parent; | |
1469 | } while (node); | |
1470 | ||
1471 | return deleted; | |
1472 | } | |
1473 | ||
57578c2e MW |
1474 | static inline void delete_sibling_entries(struct radix_tree_node *node, |
1475 | void *ptr, unsigned offset) | |
1476 | { | |
1477 | #ifdef CONFIG_RADIX_TREE_MULTIORDER | |
1478 | int i; | |
1479 | for (i = 1; offset + i < RADIX_TREE_MAP_SIZE; i++) { | |
1480 | if (node->slots[offset + i] != ptr) | |
1481 | break; | |
1482 | node->slots[offset + i] = NULL; | |
1483 | node->count--; | |
1484 | } | |
1485 | #endif | |
1486 | } | |
1487 | ||
1da177e4 | 1488 | /** |
53c59f26 | 1489 | * radix_tree_delete_item - delete an item from a radix tree |
1da177e4 LT |
1490 | * @root: radix tree root |
1491 | * @index: index key | |
53c59f26 | 1492 | * @item: expected item |
1da177e4 | 1493 | * |
53c59f26 | 1494 | * Remove @item at @index from the radix tree rooted at @root. |
1da177e4 | 1495 | * |
53c59f26 JW |
1496 | * Returns the address of the deleted item, or NULL if it was not present |
1497 | * or the entry at the given @index was not @item. | |
1da177e4 | 1498 | */ |
53c59f26 JW |
1499 | void *radix_tree_delete_item(struct radix_tree_root *root, |
1500 | unsigned long index, void *item) | |
1da177e4 | 1501 | { |
139e5616 | 1502 | struct radix_tree_node *node; |
57578c2e | 1503 | unsigned int offset; |
139e5616 JW |
1504 | void **slot; |
1505 | void *entry; | |
d5274261 | 1506 | int tag; |
1da177e4 | 1507 | |
139e5616 JW |
1508 | entry = __radix_tree_lookup(root, index, &node, &slot); |
1509 | if (!entry) | |
1510 | return NULL; | |
1da177e4 | 1511 | |
139e5616 JW |
1512 | if (item && entry != item) |
1513 | return NULL; | |
1514 | ||
1515 | if (!node) { | |
612d6c19 NP |
1516 | root_tag_clear_all(root); |
1517 | root->rnode = NULL; | |
139e5616 | 1518 | return entry; |
612d6c19 | 1519 | } |
1da177e4 | 1520 | |
139e5616 | 1521 | offset = index & RADIX_TREE_MAP_MASK; |
53c59f26 | 1522 | |
1da177e4 | 1523 | /* |
e2bdb933 HD |
1524 | * Clear all tags associated with the item to be deleted. |
1525 | * This way of doing it would be inefficient, but seldom is any set. | |
1da177e4 | 1526 | */ |
daff89f3 | 1527 | for (tag = 0; tag < RADIX_TREE_MAX_TAGS; tag++) { |
e2bdb933 | 1528 | if (tag_get(node, tag, offset)) |
612d6c19 | 1529 | radix_tree_tag_clear(root, index, tag); |
d5274261 | 1530 | } |
1da177e4 | 1531 | |
57578c2e | 1532 | delete_sibling_entries(node, ptr_to_indirect(slot), offset); |
139e5616 JW |
1533 | node->slots[offset] = NULL; |
1534 | node->count--; | |
e2bdb933 | 1535 | |
449dd698 | 1536 | __radix_tree_delete_node(root, node); |
612d6c19 | 1537 | |
139e5616 | 1538 | return entry; |
1da177e4 | 1539 | } |
53c59f26 JW |
1540 | EXPORT_SYMBOL(radix_tree_delete_item); |
1541 | ||
1542 | /** | |
1543 | * radix_tree_delete - delete an item from a radix tree | |
1544 | * @root: radix tree root | |
1545 | * @index: index key | |
1546 | * | |
1547 | * Remove the item at @index from the radix tree rooted at @root. | |
1548 | * | |
1549 | * Returns the address of the deleted item, or NULL if it was not present. | |
1550 | */ | |
1551 | void *radix_tree_delete(struct radix_tree_root *root, unsigned long index) | |
1552 | { | |
1553 | return radix_tree_delete_item(root, index, NULL); | |
1554 | } | |
1da177e4 LT |
1555 | EXPORT_SYMBOL(radix_tree_delete); |
1556 | ||
1557 | /** | |
1558 | * radix_tree_tagged - test whether any items in the tree are tagged | |
1559 | * @root: radix tree root | |
1560 | * @tag: tag to test | |
1561 | */ | |
daff89f3 | 1562 | int radix_tree_tagged(struct radix_tree_root *root, unsigned int tag) |
1da177e4 | 1563 | { |
612d6c19 | 1564 | return root_tag_get(root, tag); |
1da177e4 LT |
1565 | } |
1566 | EXPORT_SYMBOL(radix_tree_tagged); | |
1567 | ||
1568 | static void | |
449dd698 | 1569 | radix_tree_node_ctor(void *arg) |
1da177e4 | 1570 | { |
449dd698 JW |
1571 | struct radix_tree_node *node = arg; |
1572 | ||
1573 | memset(node, 0, sizeof(*node)); | |
1574 | INIT_LIST_HEAD(&node->private_list); | |
1da177e4 LT |
1575 | } |
1576 | ||
1577 | static __init unsigned long __maxindex(unsigned int height) | |
1578 | { | |
430d275a PL |
1579 | unsigned int width = height * RADIX_TREE_MAP_SHIFT; |
1580 | int shift = RADIX_TREE_INDEX_BITS - width; | |
1581 | ||
1582 | if (shift < 0) | |
1583 | return ~0UL; | |
1584 | if (shift >= BITS_PER_LONG) | |
1585 | return 0UL; | |
1586 | return ~0UL >> shift; | |
1da177e4 LT |
1587 | } |
1588 | ||
1589 | static __init void radix_tree_init_maxindex(void) | |
1590 | { | |
1591 | unsigned int i; | |
1592 | ||
1593 | for (i = 0; i < ARRAY_SIZE(height_to_maxindex); i++) | |
1594 | height_to_maxindex[i] = __maxindex(i); | |
1595 | } | |
1596 | ||
1da177e4 LT |
1597 | static int radix_tree_callback(struct notifier_block *nfb, |
1598 | unsigned long action, | |
1599 | void *hcpu) | |
1600 | { | |
1601 | int cpu = (long)hcpu; | |
1602 | struct radix_tree_preload *rtp; | |
9d2a8da0 | 1603 | struct radix_tree_node *node; |
1da177e4 LT |
1604 | |
1605 | /* Free per-cpu pool of perloaded nodes */ | |
8bb78442 | 1606 | if (action == CPU_DEAD || action == CPU_DEAD_FROZEN) { |
1da177e4 LT |
1607 | rtp = &per_cpu(radix_tree_preloads, cpu); |
1608 | while (rtp->nr) { | |
9d2a8da0 KS |
1609 | node = rtp->nodes; |
1610 | rtp->nodes = node->private_data; | |
1611 | kmem_cache_free(radix_tree_node_cachep, node); | |
1612 | rtp->nr--; | |
1da177e4 LT |
1613 | } |
1614 | } | |
1615 | return NOTIFY_OK; | |
1616 | } | |
1da177e4 LT |
1617 | |
1618 | void __init radix_tree_init(void) | |
1619 | { | |
1620 | radix_tree_node_cachep = kmem_cache_create("radix_tree_node", | |
1621 | sizeof(struct radix_tree_node), 0, | |
488514d1 CL |
1622 | SLAB_PANIC | SLAB_RECLAIM_ACCOUNT, |
1623 | radix_tree_node_ctor); | |
1da177e4 LT |
1624 | radix_tree_init_maxindex(); |
1625 | hotcpu_notifier(radix_tree_callback, 0); | |
1626 | } |