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