]>
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 | ||
db050f29 MW |
83 | #ifdef CONFIG_RADIX_TREE_MULTIORDER |
84 | /* Sibling slots point directly to another slot in the same node */ | |
85 | static inline bool is_sibling_entry(struct radix_tree_node *parent, void *node) | |
86 | { | |
87 | void **ptr = node; | |
88 | return (parent->slots <= ptr) && | |
89 | (ptr < parent->slots + RADIX_TREE_MAP_SIZE); | |
90 | } | |
91 | #else | |
92 | static inline bool is_sibling_entry(struct radix_tree_node *parent, void *node) | |
93 | { | |
94 | return false; | |
95 | } | |
96 | #endif | |
97 | ||
98 | static inline unsigned long get_slot_offset(struct radix_tree_node *parent, | |
99 | void **slot) | |
100 | { | |
101 | return slot - parent->slots; | |
102 | } | |
103 | ||
104 | static unsigned radix_tree_descend(struct radix_tree_node *parent, | |
105 | struct radix_tree_node **nodep, unsigned offset) | |
106 | { | |
107 | void **entry = rcu_dereference_raw(parent->slots[offset]); | |
108 | ||
109 | #ifdef CONFIG_RADIX_TREE_MULTIORDER | |
110 | if (radix_tree_is_indirect_ptr(entry)) { | |
111 | unsigned long siboff = get_slot_offset(parent, entry); | |
112 | if (siboff < RADIX_TREE_MAP_SIZE) { | |
113 | offset = siboff; | |
114 | entry = rcu_dereference_raw(parent->slots[offset]); | |
115 | } | |
116 | } | |
117 | #endif | |
118 | ||
119 | *nodep = (void *)entry; | |
120 | return offset; | |
121 | } | |
122 | ||
612d6c19 NP |
123 | static inline gfp_t root_gfp_mask(struct radix_tree_root *root) |
124 | { | |
125 | return root->gfp_mask & __GFP_BITS_MASK; | |
126 | } | |
127 | ||
643b52b9 NP |
128 | static inline void tag_set(struct radix_tree_node *node, unsigned int tag, |
129 | int offset) | |
130 | { | |
131 | __set_bit(offset, node->tags[tag]); | |
132 | } | |
133 | ||
134 | static inline void tag_clear(struct radix_tree_node *node, unsigned int tag, | |
135 | int offset) | |
136 | { | |
137 | __clear_bit(offset, node->tags[tag]); | |
138 | } | |
139 | ||
140 | static inline int tag_get(struct radix_tree_node *node, unsigned int tag, | |
141 | int offset) | |
142 | { | |
143 | return test_bit(offset, node->tags[tag]); | |
144 | } | |
145 | ||
146 | static inline void root_tag_set(struct radix_tree_root *root, unsigned int tag) | |
147 | { | |
148 | root->gfp_mask |= (__force gfp_t)(1 << (tag + __GFP_BITS_SHIFT)); | |
149 | } | |
150 | ||
151 | static inline void root_tag_clear(struct radix_tree_root *root, unsigned int tag) | |
152 | { | |
153 | root->gfp_mask &= (__force gfp_t)~(1 << (tag + __GFP_BITS_SHIFT)); | |
154 | } | |
155 | ||
156 | static inline void root_tag_clear_all(struct radix_tree_root *root) | |
157 | { | |
158 | root->gfp_mask &= __GFP_BITS_MASK; | |
159 | } | |
160 | ||
161 | static inline int root_tag_get(struct radix_tree_root *root, unsigned int tag) | |
162 | { | |
163 | return (__force unsigned)root->gfp_mask & (1 << (tag + __GFP_BITS_SHIFT)); | |
164 | } | |
165 | ||
166 | /* | |
167 | * Returns 1 if any slot in the node has this tag set. | |
168 | * Otherwise returns 0. | |
169 | */ | |
170 | static inline int any_tag_set(struct radix_tree_node *node, unsigned int tag) | |
171 | { | |
172 | int idx; | |
173 | for (idx = 0; idx < RADIX_TREE_TAG_LONGS; idx++) { | |
174 | if (node->tags[tag][idx]) | |
175 | return 1; | |
176 | } | |
177 | return 0; | |
178 | } | |
78c1d784 KK |
179 | |
180 | /** | |
181 | * radix_tree_find_next_bit - find the next set bit in a memory region | |
182 | * | |
183 | * @addr: The address to base the search on | |
184 | * @size: The bitmap size in bits | |
185 | * @offset: The bitnumber to start searching at | |
186 | * | |
187 | * Unrollable variant of find_next_bit() for constant size arrays. | |
188 | * Tail bits starting from size to roundup(size, BITS_PER_LONG) must be zero. | |
189 | * Returns next bit offset, or size if nothing found. | |
190 | */ | |
191 | static __always_inline unsigned long | |
192 | radix_tree_find_next_bit(const unsigned long *addr, | |
193 | unsigned long size, unsigned long offset) | |
194 | { | |
195 | if (!__builtin_constant_p(size)) | |
196 | return find_next_bit(addr, size, offset); | |
197 | ||
198 | if (offset < size) { | |
199 | unsigned long tmp; | |
200 | ||
201 | addr += offset / BITS_PER_LONG; | |
202 | tmp = *addr >> (offset % BITS_PER_LONG); | |
203 | if (tmp) | |
204 | return __ffs(tmp) + offset; | |
205 | offset = (offset + BITS_PER_LONG) & ~(BITS_PER_LONG - 1); | |
206 | while (offset < size) { | |
207 | tmp = *++addr; | |
208 | if (tmp) | |
209 | return __ffs(tmp) + offset; | |
210 | offset += BITS_PER_LONG; | |
211 | } | |
212 | } | |
213 | return size; | |
214 | } | |
215 | ||
7cf19af4 MW |
216 | #if 0 |
217 | static void dump_node(void *slot, int height, int offset) | |
218 | { | |
219 | struct radix_tree_node *node; | |
220 | int i; | |
221 | ||
222 | if (!slot) | |
223 | return; | |
224 | ||
225 | if (height == 0) { | |
226 | pr_debug("radix entry %p offset %d\n", slot, offset); | |
227 | return; | |
228 | } | |
229 | ||
230 | node = indirect_to_ptr(slot); | |
231 | pr_debug("radix node: %p offset %d tags %lx %lx %lx path %x count %d parent %p\n", | |
232 | slot, offset, node->tags[0][0], node->tags[1][0], | |
233 | node->tags[2][0], node->path, node->count, node->parent); | |
234 | ||
235 | for (i = 0; i < RADIX_TREE_MAP_SIZE; i++) | |
236 | dump_node(node->slots[i], height - 1, i); | |
237 | } | |
238 | ||
239 | /* For debug */ | |
240 | static void radix_tree_dump(struct radix_tree_root *root) | |
241 | { | |
242 | pr_debug("radix root: %p height %d rnode %p tags %x\n", | |
243 | root, root->height, root->rnode, | |
244 | root->gfp_mask >> __GFP_BITS_SHIFT); | |
245 | if (!radix_tree_is_indirect_ptr(root->rnode)) | |
246 | return; | |
247 | dump_node(root->rnode, root->height, 0); | |
248 | } | |
249 | #endif | |
250 | ||
1da177e4 LT |
251 | /* |
252 | * This assumes that the caller has performed appropriate preallocation, and | |
253 | * that the caller has pinned this thread of control to the current CPU. | |
254 | */ | |
255 | static struct radix_tree_node * | |
256 | radix_tree_node_alloc(struct radix_tree_root *root) | |
257 | { | |
e2848a0e | 258 | struct radix_tree_node *ret = NULL; |
612d6c19 | 259 | gfp_t gfp_mask = root_gfp_mask(root); |
1da177e4 | 260 | |
5e4c0d97 JK |
261 | /* |
262 | * Preload code isn't irq safe and it doesn't make sence to use | |
263 | * preloading in the interrupt anyway as all the allocations have to | |
264 | * be atomic. So just do normal allocation when in interrupt. | |
265 | */ | |
d0164adc | 266 | if (!gfpflags_allow_blocking(gfp_mask) && !in_interrupt()) { |
1da177e4 LT |
267 | struct radix_tree_preload *rtp; |
268 | ||
58e698af VD |
269 | /* |
270 | * Even if the caller has preloaded, try to allocate from the | |
271 | * cache first for the new node to get accounted. | |
272 | */ | |
273 | ret = kmem_cache_alloc(radix_tree_node_cachep, | |
274 | gfp_mask | __GFP_ACCOUNT | __GFP_NOWARN); | |
275 | if (ret) | |
276 | goto out; | |
277 | ||
e2848a0e NP |
278 | /* |
279 | * Provided the caller has preloaded here, we will always | |
280 | * succeed in getting a node here (and never reach | |
281 | * kmem_cache_alloc) | |
282 | */ | |
7c8e0181 | 283 | rtp = this_cpu_ptr(&radix_tree_preloads); |
1da177e4 | 284 | if (rtp->nr) { |
9d2a8da0 KS |
285 | ret = rtp->nodes; |
286 | rtp->nodes = ret->private_data; | |
287 | ret->private_data = NULL; | |
1da177e4 LT |
288 | rtp->nr--; |
289 | } | |
ce80b067 CM |
290 | /* |
291 | * Update the allocation stack trace as this is more useful | |
292 | * for debugging. | |
293 | */ | |
294 | kmemleak_update_trace(ret); | |
58e698af | 295 | goto out; |
1da177e4 | 296 | } |
58e698af VD |
297 | ret = kmem_cache_alloc(radix_tree_node_cachep, |
298 | gfp_mask | __GFP_ACCOUNT); | |
299 | out: | |
c0bc9875 | 300 | BUG_ON(radix_tree_is_indirect_ptr(ret)); |
1da177e4 LT |
301 | return ret; |
302 | } | |
303 | ||
7cf9c2c7 NP |
304 | static void radix_tree_node_rcu_free(struct rcu_head *head) |
305 | { | |
306 | struct radix_tree_node *node = | |
307 | container_of(head, struct radix_tree_node, rcu_head); | |
b6dd0865 | 308 | int i; |
643b52b9 NP |
309 | |
310 | /* | |
311 | * must only free zeroed nodes into the slab. radix_tree_shrink | |
312 | * can leave us with a non-NULL entry in the first slot, so clear | |
313 | * that here to make sure. | |
314 | */ | |
b6dd0865 DC |
315 | for (i = 0; i < RADIX_TREE_MAX_TAGS; i++) |
316 | tag_clear(node, i, 0); | |
317 | ||
643b52b9 NP |
318 | node->slots[0] = NULL; |
319 | node->count = 0; | |
320 | ||
7cf9c2c7 NP |
321 | kmem_cache_free(radix_tree_node_cachep, node); |
322 | } | |
323 | ||
1da177e4 LT |
324 | static inline void |
325 | radix_tree_node_free(struct radix_tree_node *node) | |
326 | { | |
7cf9c2c7 | 327 | call_rcu(&node->rcu_head, radix_tree_node_rcu_free); |
1da177e4 LT |
328 | } |
329 | ||
330 | /* | |
331 | * Load up this CPU's radix_tree_node buffer with sufficient objects to | |
332 | * ensure that the addition of a single element in the tree cannot fail. On | |
333 | * success, return zero, with preemption disabled. On error, return -ENOMEM | |
334 | * with preemption not disabled. | |
b34df792 DH |
335 | * |
336 | * To make use of this facility, the radix tree must be initialised without | |
d0164adc | 337 | * __GFP_DIRECT_RECLAIM being passed to INIT_RADIX_TREE(). |
1da177e4 | 338 | */ |
5e4c0d97 | 339 | static int __radix_tree_preload(gfp_t gfp_mask) |
1da177e4 LT |
340 | { |
341 | struct radix_tree_preload *rtp; | |
342 | struct radix_tree_node *node; | |
343 | int ret = -ENOMEM; | |
344 | ||
345 | preempt_disable(); | |
7c8e0181 | 346 | rtp = this_cpu_ptr(&radix_tree_preloads); |
9d2a8da0 | 347 | while (rtp->nr < RADIX_TREE_PRELOAD_SIZE) { |
1da177e4 | 348 | preempt_enable(); |
488514d1 | 349 | node = kmem_cache_alloc(radix_tree_node_cachep, gfp_mask); |
1da177e4 LT |
350 | if (node == NULL) |
351 | goto out; | |
352 | preempt_disable(); | |
7c8e0181 | 353 | rtp = this_cpu_ptr(&radix_tree_preloads); |
9d2a8da0 KS |
354 | if (rtp->nr < RADIX_TREE_PRELOAD_SIZE) { |
355 | node->private_data = rtp->nodes; | |
356 | rtp->nodes = node; | |
357 | rtp->nr++; | |
358 | } else { | |
1da177e4 | 359 | kmem_cache_free(radix_tree_node_cachep, node); |
9d2a8da0 | 360 | } |
1da177e4 LT |
361 | } |
362 | ret = 0; | |
363 | out: | |
364 | return ret; | |
365 | } | |
5e4c0d97 JK |
366 | |
367 | /* | |
368 | * Load up this CPU's radix_tree_node buffer with sufficient objects to | |
369 | * ensure that the addition of a single element in the tree cannot fail. On | |
370 | * success, return zero, with preemption disabled. On error, return -ENOMEM | |
371 | * with preemption not disabled. | |
372 | * | |
373 | * To make use of this facility, the radix tree must be initialised without | |
d0164adc | 374 | * __GFP_DIRECT_RECLAIM being passed to INIT_RADIX_TREE(). |
5e4c0d97 JK |
375 | */ |
376 | int radix_tree_preload(gfp_t gfp_mask) | |
377 | { | |
378 | /* Warn on non-sensical use... */ | |
d0164adc | 379 | WARN_ON_ONCE(!gfpflags_allow_blocking(gfp_mask)); |
5e4c0d97 JK |
380 | return __radix_tree_preload(gfp_mask); |
381 | } | |
d7f0923d | 382 | EXPORT_SYMBOL(radix_tree_preload); |
1da177e4 | 383 | |
5e4c0d97 JK |
384 | /* |
385 | * The same as above function, except we don't guarantee preloading happens. | |
386 | * We do it, if we decide it helps. On success, return zero with preemption | |
387 | * disabled. On error, return -ENOMEM with preemption not disabled. | |
388 | */ | |
389 | int radix_tree_maybe_preload(gfp_t gfp_mask) | |
390 | { | |
d0164adc | 391 | if (gfpflags_allow_blocking(gfp_mask)) |
5e4c0d97 JK |
392 | return __radix_tree_preload(gfp_mask); |
393 | /* Preloading doesn't help anything with this gfp mask, skip it */ | |
394 | preempt_disable(); | |
395 | return 0; | |
396 | } | |
397 | EXPORT_SYMBOL(radix_tree_maybe_preload); | |
398 | ||
1da177e4 LT |
399 | /* |
400 | * Return the maximum key which can be store into a | |
401 | * radix tree with height HEIGHT. | |
402 | */ | |
403 | static inline unsigned long radix_tree_maxindex(unsigned int height) | |
404 | { | |
405 | return height_to_maxindex[height]; | |
406 | } | |
407 | ||
1456a439 MW |
408 | static inline unsigned long node_maxindex(struct radix_tree_node *node) |
409 | { | |
410 | return radix_tree_maxindex(node->path & RADIX_TREE_HEIGHT_MASK); | |
411 | } | |
412 | ||
413 | static unsigned radix_tree_load_root(struct radix_tree_root *root, | |
414 | struct radix_tree_node **nodep, unsigned long *maxindex) | |
415 | { | |
416 | struct radix_tree_node *node = rcu_dereference_raw(root->rnode); | |
417 | ||
418 | *nodep = node; | |
419 | ||
420 | if (likely(radix_tree_is_indirect_ptr(node))) { | |
421 | node = indirect_to_ptr(node); | |
422 | *maxindex = node_maxindex(node); | |
423 | return (node->path & RADIX_TREE_HEIGHT_MASK) * | |
424 | RADIX_TREE_MAP_SHIFT; | |
425 | } | |
426 | ||
427 | *maxindex = 0; | |
428 | return 0; | |
429 | } | |
430 | ||
1da177e4 LT |
431 | /* |
432 | * Extend a radix tree so it can store key @index. | |
433 | */ | |
e6145236 MW |
434 | static int radix_tree_extend(struct radix_tree_root *root, |
435 | unsigned long index, unsigned order) | |
1da177e4 LT |
436 | { |
437 | struct radix_tree_node *node; | |
e2bdb933 | 438 | struct radix_tree_node *slot; |
1da177e4 | 439 | unsigned int height; |
1da177e4 LT |
440 | int tag; |
441 | ||
442 | /* Figure out what the height should be. */ | |
443 | height = root->height + 1; | |
444 | while (index > radix_tree_maxindex(height)) | |
445 | height++; | |
446 | ||
e6145236 | 447 | if ((root->rnode == NULL) && (order == 0)) { |
1da177e4 LT |
448 | root->height = height; |
449 | goto out; | |
450 | } | |
451 | ||
1da177e4 | 452 | do { |
7cf9c2c7 | 453 | unsigned int newheight; |
1da177e4 LT |
454 | if (!(node = radix_tree_node_alloc(root))) |
455 | return -ENOMEM; | |
456 | ||
1da177e4 | 457 | /* Propagate the aggregated tag info into the new root */ |
daff89f3 | 458 | for (tag = 0; tag < RADIX_TREE_MAX_TAGS; tag++) { |
612d6c19 | 459 | if (root_tag_get(root, tag)) |
1da177e4 LT |
460 | tag_set(node, tag, 0); |
461 | } | |
462 | ||
e2bdb933 | 463 | /* Increase the height. */ |
7cf9c2c7 | 464 | newheight = root->height+1; |
449dd698 JW |
465 | BUG_ON(newheight & ~RADIX_TREE_HEIGHT_MASK); |
466 | node->path = newheight; | |
1da177e4 | 467 | node->count = 1; |
e2bdb933 HD |
468 | node->parent = NULL; |
469 | slot = root->rnode; | |
339e6353 | 470 | if (radix_tree_is_indirect_ptr(slot) && newheight > 1) { |
e2bdb933 HD |
471 | slot = indirect_to_ptr(slot); |
472 | slot->parent = node; | |
339e6353 | 473 | slot = ptr_to_indirect(slot); |
e2bdb933 HD |
474 | } |
475 | node->slots[0] = slot; | |
27d20fdd | 476 | node = ptr_to_indirect(node); |
7cf9c2c7 NP |
477 | rcu_assign_pointer(root->rnode, node); |
478 | root->height = newheight; | |
1da177e4 LT |
479 | } while (height > root->height); |
480 | out: | |
481 | return 0; | |
482 | } | |
483 | ||
484 | /** | |
139e5616 | 485 | * __radix_tree_create - create a slot in a radix tree |
1da177e4 LT |
486 | * @root: radix tree root |
487 | * @index: index key | |
e6145236 | 488 | * @order: index occupies 2^order aligned slots |
139e5616 JW |
489 | * @nodep: returns node |
490 | * @slotp: returns slot | |
1da177e4 | 491 | * |
139e5616 JW |
492 | * Create, if necessary, and return the node and slot for an item |
493 | * at position @index in the radix tree @root. | |
494 | * | |
495 | * Until there is more than one item in the tree, no nodes are | |
496 | * allocated and @root->rnode is used as a direct slot instead of | |
497 | * pointing to a node, in which case *@nodep will be NULL. | |
498 | * | |
499 | * Returns -ENOMEM, or 0 for success. | |
1da177e4 | 500 | */ |
139e5616 | 501 | int __radix_tree_create(struct radix_tree_root *root, unsigned long index, |
e6145236 MW |
502 | unsigned order, struct radix_tree_node **nodep, |
503 | void ***slotp) | |
1da177e4 | 504 | { |
201b6264 | 505 | struct radix_tree_node *node = NULL, *slot; |
139e5616 | 506 | unsigned int height, shift, offset; |
1da177e4 LT |
507 | int error; |
508 | ||
509 | /* Make sure the tree is high enough. */ | |
612d6c19 | 510 | if (index > radix_tree_maxindex(root->height)) { |
e6145236 | 511 | error = radix_tree_extend(root, index, order); |
1da177e4 LT |
512 | if (error) |
513 | return error; | |
514 | } | |
515 | ||
e6145236 | 516 | slot = root->rnode; |
c0bc9875 | 517 | |
1da177e4 | 518 | height = root->height; |
0070e28d | 519 | shift = height * RADIX_TREE_MAP_SHIFT; |
1da177e4 LT |
520 | |
521 | offset = 0; /* uninitialised var warning */ | |
e6145236 | 522 | while (shift > order) { |
201b6264 | 523 | if (slot == NULL) { |
1da177e4 | 524 | /* Have to add a child node. */ |
201b6264 | 525 | if (!(slot = radix_tree_node_alloc(root))) |
1da177e4 | 526 | return -ENOMEM; |
449dd698 | 527 | slot->path = height; |
e2bdb933 | 528 | slot->parent = node; |
201b6264 | 529 | if (node) { |
339e6353 MW |
530 | rcu_assign_pointer(node->slots[offset], |
531 | ptr_to_indirect(slot)); | |
1da177e4 | 532 | node->count++; |
449dd698 | 533 | slot->path |= offset << RADIX_TREE_HEIGHT_SHIFT; |
201b6264 | 534 | } else |
339e6353 MW |
535 | rcu_assign_pointer(root->rnode, |
536 | ptr_to_indirect(slot)); | |
e6145236 MW |
537 | } else if (!radix_tree_is_indirect_ptr(slot)) |
538 | break; | |
1da177e4 LT |
539 | |
540 | /* Go a level down */ | |
e6145236 | 541 | height--; |
0070e28d | 542 | shift -= RADIX_TREE_MAP_SHIFT; |
1da177e4 | 543 | offset = (index >> shift) & RADIX_TREE_MAP_MASK; |
e6145236 | 544 | node = indirect_to_ptr(slot); |
201b6264 | 545 | slot = node->slots[offset]; |
e6145236 MW |
546 | } |
547 | ||
57578c2e | 548 | #ifdef CONFIG_RADIX_TREE_MULTIORDER |
e6145236 | 549 | /* Insert pointers to the canonical entry */ |
3b8c00f6 MW |
550 | if (order > shift) { |
551 | int i, n = 1 << (order - shift); | |
e6145236 MW |
552 | offset = offset & ~(n - 1); |
553 | slot = ptr_to_indirect(&node->slots[offset]); | |
554 | for (i = 0; i < n; i++) { | |
555 | if (node->slots[offset + i]) | |
556 | return -EEXIST; | |
557 | } | |
558 | ||
559 | for (i = 1; i < n; i++) { | |
560 | rcu_assign_pointer(node->slots[offset + i], slot); | |
561 | node->count++; | |
562 | } | |
612d6c19 | 563 | } |
57578c2e | 564 | #endif |
1da177e4 | 565 | |
139e5616 JW |
566 | if (nodep) |
567 | *nodep = node; | |
568 | if (slotp) | |
569 | *slotp = node ? node->slots + offset : (void **)&root->rnode; | |
570 | return 0; | |
571 | } | |
572 | ||
573 | /** | |
e6145236 | 574 | * __radix_tree_insert - insert into a radix tree |
139e5616 JW |
575 | * @root: radix tree root |
576 | * @index: index key | |
e6145236 | 577 | * @order: key covers the 2^order indices around index |
139e5616 JW |
578 | * @item: item to insert |
579 | * | |
580 | * Insert an item into the radix tree at position @index. | |
581 | */ | |
e6145236 MW |
582 | int __radix_tree_insert(struct radix_tree_root *root, unsigned long index, |
583 | unsigned order, void *item) | |
139e5616 JW |
584 | { |
585 | struct radix_tree_node *node; | |
586 | void **slot; | |
587 | int error; | |
588 | ||
589 | BUG_ON(radix_tree_is_indirect_ptr(item)); | |
590 | ||
e6145236 | 591 | error = __radix_tree_create(root, index, order, &node, &slot); |
139e5616 JW |
592 | if (error) |
593 | return error; | |
594 | if (*slot != NULL) | |
1da177e4 | 595 | return -EEXIST; |
139e5616 | 596 | rcu_assign_pointer(*slot, item); |
201b6264 | 597 | |
612d6c19 NP |
598 | if (node) { |
599 | node->count++; | |
139e5616 JW |
600 | BUG_ON(tag_get(node, 0, index & RADIX_TREE_MAP_MASK)); |
601 | BUG_ON(tag_get(node, 1, index & RADIX_TREE_MAP_MASK)); | |
612d6c19 | 602 | } else { |
612d6c19 NP |
603 | BUG_ON(root_tag_get(root, 0)); |
604 | BUG_ON(root_tag_get(root, 1)); | |
605 | } | |
1da177e4 | 606 | |
1da177e4 LT |
607 | return 0; |
608 | } | |
e6145236 | 609 | EXPORT_SYMBOL(__radix_tree_insert); |
1da177e4 | 610 | |
139e5616 JW |
611 | /** |
612 | * __radix_tree_lookup - lookup an item in a radix tree | |
613 | * @root: radix tree root | |
614 | * @index: index key | |
615 | * @nodep: returns node | |
616 | * @slotp: returns slot | |
617 | * | |
618 | * Lookup and return the item at position @index in the radix | |
619 | * tree @root. | |
620 | * | |
621 | * Until there is more than one item in the tree, no nodes are | |
622 | * allocated and @root->rnode is used as a direct slot instead of | |
623 | * pointing to a node, in which case *@nodep will be NULL. | |
7cf9c2c7 | 624 | */ |
139e5616 JW |
625 | void *__radix_tree_lookup(struct radix_tree_root *root, unsigned long index, |
626 | struct radix_tree_node **nodep, void ***slotp) | |
1da177e4 | 627 | { |
139e5616 | 628 | struct radix_tree_node *node, *parent; |
1da177e4 | 629 | unsigned int height, shift; |
139e5616 | 630 | void **slot; |
612d6c19 | 631 | |
2676a58c | 632 | node = rcu_dereference_raw(root->rnode); |
7cf9c2c7 | 633 | if (node == NULL) |
1da177e4 LT |
634 | return NULL; |
635 | ||
c0bc9875 | 636 | if (!radix_tree_is_indirect_ptr(node)) { |
7cf9c2c7 NP |
637 | if (index > 0) |
638 | return NULL; | |
139e5616 JW |
639 | |
640 | if (nodep) | |
641 | *nodep = NULL; | |
642 | if (slotp) | |
643 | *slotp = (void **)&root->rnode; | |
644 | return node; | |
7cf9c2c7 | 645 | } |
27d20fdd | 646 | node = indirect_to_ptr(node); |
7cf9c2c7 | 647 | |
449dd698 | 648 | height = node->path & RADIX_TREE_HEIGHT_MASK; |
7cf9c2c7 NP |
649 | if (index > radix_tree_maxindex(height)) |
650 | return NULL; | |
612d6c19 | 651 | |
1da177e4 | 652 | shift = (height-1) * RADIX_TREE_MAP_SHIFT; |
1da177e4 | 653 | |
7cf9c2c7 | 654 | do { |
139e5616 JW |
655 | parent = node; |
656 | slot = node->slots + ((index >> shift) & RADIX_TREE_MAP_MASK); | |
2676a58c | 657 | node = rcu_dereference_raw(*slot); |
7cf9c2c7 | 658 | if (node == NULL) |
1da177e4 | 659 | return NULL; |
e6145236 MW |
660 | if (!radix_tree_is_indirect_ptr(node)) |
661 | break; | |
339e6353 | 662 | node = indirect_to_ptr(node); |
1da177e4 | 663 | |
1da177e4 LT |
664 | shift -= RADIX_TREE_MAP_SHIFT; |
665 | height--; | |
7cf9c2c7 | 666 | } while (height > 0); |
1da177e4 | 667 | |
139e5616 JW |
668 | if (nodep) |
669 | *nodep = parent; | |
670 | if (slotp) | |
671 | *slotp = slot; | |
672 | return node; | |
b72b71c6 HS |
673 | } |
674 | ||
675 | /** | |
676 | * radix_tree_lookup_slot - lookup a slot in a radix tree | |
677 | * @root: radix tree root | |
678 | * @index: index key | |
679 | * | |
680 | * Returns: the slot corresponding to the position @index in the | |
681 | * radix tree @root. This is useful for update-if-exists operations. | |
682 | * | |
683 | * This function can be called under rcu_read_lock iff the slot is not | |
684 | * modified by radix_tree_replace_slot, otherwise it must be called | |
685 | * exclusive from other writers. Any dereference of the slot must be done | |
686 | * using radix_tree_deref_slot. | |
687 | */ | |
688 | void **radix_tree_lookup_slot(struct radix_tree_root *root, unsigned long index) | |
689 | { | |
139e5616 JW |
690 | void **slot; |
691 | ||
692 | if (!__radix_tree_lookup(root, index, NULL, &slot)) | |
693 | return NULL; | |
694 | return slot; | |
a4331366 | 695 | } |
a4331366 HR |
696 | EXPORT_SYMBOL(radix_tree_lookup_slot); |
697 | ||
698 | /** | |
699 | * radix_tree_lookup - perform lookup operation on a radix tree | |
700 | * @root: radix tree root | |
701 | * @index: index key | |
702 | * | |
703 | * Lookup the item at the position @index in the radix tree @root. | |
7cf9c2c7 NP |
704 | * |
705 | * This function can be called under rcu_read_lock, however the caller | |
706 | * must manage lifetimes of leaf nodes (eg. RCU may also be used to free | |
707 | * them safely). No RCU barriers are required to access or modify the | |
708 | * returned item, however. | |
a4331366 HR |
709 | */ |
710 | void *radix_tree_lookup(struct radix_tree_root *root, unsigned long index) | |
711 | { | |
139e5616 | 712 | return __radix_tree_lookup(root, index, NULL, NULL); |
1da177e4 LT |
713 | } |
714 | EXPORT_SYMBOL(radix_tree_lookup); | |
715 | ||
716 | /** | |
717 | * radix_tree_tag_set - set a tag on a radix tree node | |
718 | * @root: radix tree root | |
719 | * @index: index key | |
720 | * @tag: tag index | |
721 | * | |
daff89f3 JC |
722 | * Set the search tag (which must be < RADIX_TREE_MAX_TAGS) |
723 | * corresponding to @index in the radix tree. From | |
1da177e4 LT |
724 | * the root all the way down to the leaf node. |
725 | * | |
726 | * Returns the address of the tagged item. Setting a tag on a not-present | |
727 | * item is a bug. | |
728 | */ | |
729 | void *radix_tree_tag_set(struct radix_tree_root *root, | |
daff89f3 | 730 | unsigned long index, unsigned int tag) |
1da177e4 LT |
731 | { |
732 | unsigned int height, shift; | |
201b6264 | 733 | struct radix_tree_node *slot; |
1da177e4 LT |
734 | |
735 | height = root->height; | |
4c91c364 | 736 | BUG_ON(index > radix_tree_maxindex(height)); |
1da177e4 | 737 | |
27d20fdd | 738 | slot = indirect_to_ptr(root->rnode); |
612d6c19 | 739 | shift = (height - 1) * RADIX_TREE_MAP_SHIFT; |
1da177e4 LT |
740 | |
741 | while (height > 0) { | |
742 | int offset; | |
743 | ||
744 | offset = (index >> shift) & RADIX_TREE_MAP_MASK; | |
d5274261 NP |
745 | if (!tag_get(slot, tag, offset)) |
746 | tag_set(slot, tag, offset); | |
201b6264 CL |
747 | slot = slot->slots[offset]; |
748 | BUG_ON(slot == NULL); | |
e6145236 MW |
749 | if (!radix_tree_is_indirect_ptr(slot)) |
750 | break; | |
339e6353 | 751 | slot = indirect_to_ptr(slot); |
1da177e4 LT |
752 | shift -= RADIX_TREE_MAP_SHIFT; |
753 | height--; | |
754 | } | |
755 | ||
612d6c19 NP |
756 | /* set the root's tag bit */ |
757 | if (slot && !root_tag_get(root, tag)) | |
758 | root_tag_set(root, tag); | |
759 | ||
201b6264 | 760 | return slot; |
1da177e4 LT |
761 | } |
762 | EXPORT_SYMBOL(radix_tree_tag_set); | |
763 | ||
764 | /** | |
765 | * radix_tree_tag_clear - clear a tag on a radix tree node | |
766 | * @root: radix tree root | |
767 | * @index: index key | |
768 | * @tag: tag index | |
769 | * | |
daff89f3 JC |
770 | * Clear the search tag (which must be < RADIX_TREE_MAX_TAGS) |
771 | * corresponding to @index in the radix tree. If | |
1da177e4 LT |
772 | * this causes the leaf node to have no tags set then clear the tag in the |
773 | * next-to-leaf node, etc. | |
774 | * | |
775 | * Returns the address of the tagged item on success, else NULL. ie: | |
776 | * has the same return value and semantics as radix_tree_lookup(). | |
777 | */ | |
778 | void *radix_tree_tag_clear(struct radix_tree_root *root, | |
daff89f3 | 779 | unsigned long index, unsigned int tag) |
1da177e4 | 780 | { |
e2bdb933 | 781 | struct radix_tree_node *node = NULL; |
612d6c19 | 782 | struct radix_tree_node *slot = NULL; |
1da177e4 | 783 | unsigned int height, shift; |
e2bdb933 | 784 | int uninitialized_var(offset); |
1da177e4 LT |
785 | |
786 | height = root->height; | |
787 | if (index > radix_tree_maxindex(height)) | |
788 | goto out; | |
789 | ||
e2bdb933 | 790 | shift = height * RADIX_TREE_MAP_SHIFT; |
339e6353 | 791 | slot = root->rnode; |
1da177e4 | 792 | |
e2bdb933 | 793 | while (shift) { |
201b6264 | 794 | if (slot == NULL) |
1da177e4 | 795 | goto out; |
e6145236 MW |
796 | if (!radix_tree_is_indirect_ptr(slot)) |
797 | break; | |
339e6353 | 798 | slot = indirect_to_ptr(slot); |
1da177e4 | 799 | |
e2bdb933 | 800 | shift -= RADIX_TREE_MAP_SHIFT; |
1da177e4 | 801 | offset = (index >> shift) & RADIX_TREE_MAP_MASK; |
e2bdb933 | 802 | node = slot; |
201b6264 | 803 | slot = slot->slots[offset]; |
1da177e4 LT |
804 | } |
805 | ||
612d6c19 | 806 | if (slot == NULL) |
1da177e4 LT |
807 | goto out; |
808 | ||
e2bdb933 HD |
809 | while (node) { |
810 | if (!tag_get(node, tag, offset)) | |
d5274261 | 811 | goto out; |
e2bdb933 HD |
812 | tag_clear(node, tag, offset); |
813 | if (any_tag_set(node, tag)) | |
6e954b9e | 814 | goto out; |
e2bdb933 HD |
815 | |
816 | index >>= RADIX_TREE_MAP_SHIFT; | |
817 | offset = index & RADIX_TREE_MAP_MASK; | |
818 | node = node->parent; | |
612d6c19 NP |
819 | } |
820 | ||
821 | /* clear the root's tag bit */ | |
822 | if (root_tag_get(root, tag)) | |
823 | root_tag_clear(root, tag); | |
824 | ||
1da177e4 | 825 | out: |
612d6c19 | 826 | return slot; |
1da177e4 LT |
827 | } |
828 | EXPORT_SYMBOL(radix_tree_tag_clear); | |
829 | ||
1da177e4 | 830 | /** |
32605a18 MT |
831 | * radix_tree_tag_get - get a tag on a radix tree node |
832 | * @root: radix tree root | |
833 | * @index: index key | |
daff89f3 | 834 | * @tag: tag index (< RADIX_TREE_MAX_TAGS) |
1da177e4 | 835 | * |
32605a18 | 836 | * Return values: |
1da177e4 | 837 | * |
612d6c19 NP |
838 | * 0: tag not present or not set |
839 | * 1: tag set | |
ce82653d DH |
840 | * |
841 | * Note that the return value of this function may not be relied on, even if | |
842 | * the RCU lock is held, unless tag modification and node deletion are excluded | |
843 | * from concurrency. | |
1da177e4 LT |
844 | */ |
845 | int radix_tree_tag_get(struct radix_tree_root *root, | |
daff89f3 | 846 | unsigned long index, unsigned int tag) |
1da177e4 LT |
847 | { |
848 | unsigned int height, shift; | |
7cf9c2c7 | 849 | struct radix_tree_node *node; |
1da177e4 | 850 | |
612d6c19 NP |
851 | /* check the root's tag bit */ |
852 | if (!root_tag_get(root, tag)) | |
853 | return 0; | |
854 | ||
2676a58c | 855 | node = rcu_dereference_raw(root->rnode); |
7cf9c2c7 NP |
856 | if (node == NULL) |
857 | return 0; | |
858 | ||
c0bc9875 | 859 | if (!radix_tree_is_indirect_ptr(node)) |
7cf9c2c7 | 860 | return (index == 0); |
27d20fdd | 861 | node = indirect_to_ptr(node); |
7cf9c2c7 | 862 | |
449dd698 | 863 | height = node->path & RADIX_TREE_HEIGHT_MASK; |
7cf9c2c7 NP |
864 | if (index > radix_tree_maxindex(height)) |
865 | return 0; | |
612d6c19 | 866 | |
1da177e4 | 867 | shift = (height - 1) * RADIX_TREE_MAP_SHIFT; |
1da177e4 LT |
868 | |
869 | for ( ; ; ) { | |
870 | int offset; | |
871 | ||
7cf9c2c7 | 872 | if (node == NULL) |
1da177e4 | 873 | return 0; |
339e6353 | 874 | node = indirect_to_ptr(node); |
1da177e4 LT |
875 | |
876 | offset = (index >> shift) & RADIX_TREE_MAP_MASK; | |
7cf9c2c7 | 877 | if (!tag_get(node, tag, offset)) |
3fa36acb | 878 | return 0; |
ce82653d | 879 | if (height == 1) |
3fa36acb | 880 | return 1; |
2676a58c | 881 | node = rcu_dereference_raw(node->slots[offset]); |
e6145236 MW |
882 | if (!radix_tree_is_indirect_ptr(node)) |
883 | return 1; | |
1da177e4 LT |
884 | shift -= RADIX_TREE_MAP_SHIFT; |
885 | height--; | |
886 | } | |
887 | } | |
888 | EXPORT_SYMBOL(radix_tree_tag_get); | |
1da177e4 | 889 | |
78c1d784 KK |
890 | /** |
891 | * radix_tree_next_chunk - find next chunk of slots for iteration | |
892 | * | |
893 | * @root: radix tree root | |
894 | * @iter: iterator state | |
895 | * @flags: RADIX_TREE_ITER_* flags and tag index | |
896 | * Returns: pointer to chunk first slot, or NULL if iteration is over | |
897 | */ | |
898 | void **radix_tree_next_chunk(struct radix_tree_root *root, | |
899 | struct radix_tree_iter *iter, unsigned flags) | |
900 | { | |
901 | unsigned shift, tag = flags & RADIX_TREE_ITER_TAG_MASK; | |
902 | struct radix_tree_node *rnode, *node; | |
449dd698 | 903 | unsigned long index, offset, height; |
78c1d784 KK |
904 | |
905 | if ((flags & RADIX_TREE_ITER_TAGGED) && !root_tag_get(root, tag)) | |
906 | return NULL; | |
907 | ||
908 | /* | |
909 | * Catch next_index overflow after ~0UL. iter->index never overflows | |
910 | * during iterating; it can be zero only at the beginning. | |
911 | * And we cannot overflow iter->next_index in a single step, | |
912 | * because RADIX_TREE_MAP_SHIFT < BITS_PER_LONG. | |
fffaee36 KK |
913 | * |
914 | * This condition also used by radix_tree_next_slot() to stop | |
915 | * contiguous iterating, and forbid swithing to the next chunk. | |
78c1d784 KK |
916 | */ |
917 | index = iter->next_index; | |
918 | if (!index && iter->index) | |
919 | return NULL; | |
920 | ||
921 | rnode = rcu_dereference_raw(root->rnode); | |
922 | if (radix_tree_is_indirect_ptr(rnode)) { | |
923 | rnode = indirect_to_ptr(rnode); | |
924 | } else if (rnode && !index) { | |
925 | /* Single-slot tree */ | |
926 | iter->index = 0; | |
927 | iter->next_index = 1; | |
928 | iter->tags = 1; | |
929 | return (void **)&root->rnode; | |
930 | } else | |
931 | return NULL; | |
932 | ||
933 | restart: | |
449dd698 JW |
934 | height = rnode->path & RADIX_TREE_HEIGHT_MASK; |
935 | shift = (height - 1) * RADIX_TREE_MAP_SHIFT; | |
78c1d784 KK |
936 | offset = index >> shift; |
937 | ||
938 | /* Index outside of the tree */ | |
939 | if (offset >= RADIX_TREE_MAP_SIZE) | |
940 | return NULL; | |
941 | ||
942 | node = rnode; | |
943 | while (1) { | |
e6145236 | 944 | struct radix_tree_node *slot; |
78c1d784 KK |
945 | if ((flags & RADIX_TREE_ITER_TAGGED) ? |
946 | !test_bit(offset, node->tags[tag]) : | |
947 | !node->slots[offset]) { | |
948 | /* Hole detected */ | |
949 | if (flags & RADIX_TREE_ITER_CONTIG) | |
950 | return NULL; | |
951 | ||
952 | if (flags & RADIX_TREE_ITER_TAGGED) | |
953 | offset = radix_tree_find_next_bit( | |
954 | node->tags[tag], | |
955 | RADIX_TREE_MAP_SIZE, | |
956 | offset + 1); | |
957 | else | |
958 | while (++offset < RADIX_TREE_MAP_SIZE) { | |
959 | if (node->slots[offset]) | |
960 | break; | |
961 | } | |
962 | index &= ~((RADIX_TREE_MAP_SIZE << shift) - 1); | |
963 | index += offset << shift; | |
964 | /* Overflow after ~0UL */ | |
965 | if (!index) | |
966 | return NULL; | |
967 | if (offset == RADIX_TREE_MAP_SIZE) | |
968 | goto restart; | |
969 | } | |
970 | ||
971 | /* This is leaf-node */ | |
972 | if (!shift) | |
973 | break; | |
974 | ||
e6145236 MW |
975 | slot = rcu_dereference_raw(node->slots[offset]); |
976 | if (slot == NULL) | |
78c1d784 | 977 | goto restart; |
e6145236 MW |
978 | if (!radix_tree_is_indirect_ptr(slot)) |
979 | break; | |
980 | node = indirect_to_ptr(slot); | |
78c1d784 KK |
981 | shift -= RADIX_TREE_MAP_SHIFT; |
982 | offset = (index >> shift) & RADIX_TREE_MAP_MASK; | |
983 | } | |
984 | ||
985 | /* Update the iterator state */ | |
986 | iter->index = index; | |
987 | iter->next_index = (index | RADIX_TREE_MAP_MASK) + 1; | |
988 | ||
989 | /* Construct iter->tags bit-mask from node->tags[tag] array */ | |
990 | if (flags & RADIX_TREE_ITER_TAGGED) { | |
991 | unsigned tag_long, tag_bit; | |
992 | ||
993 | tag_long = offset / BITS_PER_LONG; | |
994 | tag_bit = offset % BITS_PER_LONG; | |
995 | iter->tags = node->tags[tag][tag_long] >> tag_bit; | |
996 | /* This never happens if RADIX_TREE_TAG_LONGS == 1 */ | |
997 | if (tag_long < RADIX_TREE_TAG_LONGS - 1) { | |
998 | /* Pick tags from next element */ | |
999 | if (tag_bit) | |
1000 | iter->tags |= node->tags[tag][tag_long + 1] << | |
1001 | (BITS_PER_LONG - tag_bit); | |
1002 | /* Clip chunk size, here only BITS_PER_LONG tags */ | |
1003 | iter->next_index = index + BITS_PER_LONG; | |
1004 | } | |
1005 | } | |
1006 | ||
1007 | return node->slots + offset; | |
1008 | } | |
1009 | EXPORT_SYMBOL(radix_tree_next_chunk); | |
1010 | ||
ebf8aa44 JK |
1011 | /** |
1012 | * radix_tree_range_tag_if_tagged - for each item in given range set given | |
1013 | * tag if item has another tag set | |
1014 | * @root: radix tree root | |
1015 | * @first_indexp: pointer to a starting index of a range to scan | |
1016 | * @last_index: last index of a range to scan | |
1017 | * @nr_to_tag: maximum number items to tag | |
1018 | * @iftag: tag index to test | |
1019 | * @settag: tag index to set if tested tag is set | |
1020 | * | |
1021 | * This function scans range of radix tree from first_index to last_index | |
1022 | * (inclusive). For each item in the range if iftag is set, the function sets | |
1023 | * also settag. The function stops either after tagging nr_to_tag items or | |
1024 | * after reaching last_index. | |
1025 | * | |
144dcfc0 DC |
1026 | * The tags must be set from the leaf level only and propagated back up the |
1027 | * path to the root. We must do this so that we resolve the full path before | |
1028 | * setting any tags on intermediate nodes. If we set tags as we descend, then | |
1029 | * we can get to the leaf node and find that the index that has the iftag | |
1030 | * set is outside the range we are scanning. This reults in dangling tags and | |
1031 | * can lead to problems with later tag operations (e.g. livelocks on lookups). | |
1032 | * | |
ebf8aa44 JK |
1033 | * The function returns number of leaves where the tag was set and sets |
1034 | * *first_indexp to the first unscanned index. | |
d5ed3a4a JK |
1035 | * WARNING! *first_indexp can wrap if last_index is ULONG_MAX. Caller must |
1036 | * be prepared to handle that. | |
ebf8aa44 JK |
1037 | */ |
1038 | unsigned long radix_tree_range_tag_if_tagged(struct radix_tree_root *root, | |
1039 | unsigned long *first_indexp, unsigned long last_index, | |
1040 | unsigned long nr_to_tag, | |
1041 | unsigned int iftag, unsigned int settag) | |
1042 | { | |
144dcfc0 | 1043 | unsigned int height = root->height; |
e2bdb933 | 1044 | struct radix_tree_node *node = NULL; |
144dcfc0 DC |
1045 | struct radix_tree_node *slot; |
1046 | unsigned int shift; | |
1047 | unsigned long tagged = 0; | |
1048 | unsigned long index = *first_indexp; | |
ebf8aa44 JK |
1049 | |
1050 | last_index = min(last_index, radix_tree_maxindex(height)); | |
1051 | if (index > last_index) | |
1052 | return 0; | |
1053 | if (!nr_to_tag) | |
1054 | return 0; | |
1055 | if (!root_tag_get(root, iftag)) { | |
1056 | *first_indexp = last_index + 1; | |
1057 | return 0; | |
1058 | } | |
1059 | if (height == 0) { | |
1060 | *first_indexp = last_index + 1; | |
1061 | root_tag_set(root, settag); | |
1062 | return 1; | |
1063 | } | |
1064 | ||
1065 | shift = (height - 1) * RADIX_TREE_MAP_SHIFT; | |
27d20fdd | 1066 | slot = indirect_to_ptr(root->rnode); |
ebf8aa44 JK |
1067 | |
1068 | for (;;) { | |
e2bdb933 | 1069 | unsigned long upindex; |
ebf8aa44 JK |
1070 | int offset; |
1071 | ||
1072 | offset = (index >> shift) & RADIX_TREE_MAP_MASK; | |
1073 | if (!slot->slots[offset]) | |
1074 | goto next; | |
1075 | if (!tag_get(slot, iftag, offset)) | |
1076 | goto next; | |
e2bdb933 | 1077 | if (shift) { |
e2bdb933 | 1078 | node = slot; |
144dcfc0 | 1079 | slot = slot->slots[offset]; |
e6145236 MW |
1080 | if (radix_tree_is_indirect_ptr(slot)) { |
1081 | slot = indirect_to_ptr(slot); | |
1082 | shift -= RADIX_TREE_MAP_SHIFT; | |
1083 | continue; | |
1084 | } else { | |
1085 | slot = node; | |
1086 | node = node->parent; | |
1087 | } | |
144dcfc0 DC |
1088 | } |
1089 | ||
1090 | /* tag the leaf */ | |
e6145236 | 1091 | tagged += 1 << shift; |
ebf8aa44 | 1092 | tag_set(slot, settag, offset); |
144dcfc0 DC |
1093 | |
1094 | /* walk back up the path tagging interior nodes */ | |
e2bdb933 HD |
1095 | upindex = index; |
1096 | while (node) { | |
1097 | upindex >>= RADIX_TREE_MAP_SHIFT; | |
1098 | offset = upindex & RADIX_TREE_MAP_MASK; | |
1099 | ||
144dcfc0 | 1100 | /* stop if we find a node with the tag already set */ |
e2bdb933 | 1101 | if (tag_get(node, settag, offset)) |
144dcfc0 | 1102 | break; |
e2bdb933 HD |
1103 | tag_set(node, settag, offset); |
1104 | node = node->parent; | |
ebf8aa44 | 1105 | } |
144dcfc0 | 1106 | |
e2bdb933 HD |
1107 | /* |
1108 | * Small optimization: now clear that node pointer. | |
1109 | * Since all of this slot's ancestors now have the tag set | |
1110 | * from setting it above, we have no further need to walk | |
1111 | * back up the tree setting tags, until we update slot to | |
1112 | * point to another radix_tree_node. | |
1113 | */ | |
1114 | node = NULL; | |
1115 | ||
ebf8aa44 JK |
1116 | next: |
1117 | /* Go to next item at level determined by 'shift' */ | |
1118 | index = ((index >> shift) + 1) << shift; | |
d5ed3a4a JK |
1119 | /* Overflow can happen when last_index is ~0UL... */ |
1120 | if (index > last_index || !index) | |
ebf8aa44 JK |
1121 | break; |
1122 | if (tagged >= nr_to_tag) | |
1123 | break; | |
1124 | while (((index >> shift) & RADIX_TREE_MAP_MASK) == 0) { | |
1125 | /* | |
1126 | * We've fully scanned this node. Go up. Because | |
1127 | * last_index is guaranteed to be in the tree, what | |
1128 | * we do below cannot wander astray. | |
1129 | */ | |
e2bdb933 | 1130 | slot = slot->parent; |
ebf8aa44 JK |
1131 | shift += RADIX_TREE_MAP_SHIFT; |
1132 | } | |
1133 | } | |
1134 | /* | |
ac15ee69 TO |
1135 | * We need not to tag the root tag if there is no tag which is set with |
1136 | * settag within the range from *first_indexp to last_index. | |
ebf8aa44 | 1137 | */ |
ac15ee69 TO |
1138 | if (tagged > 0) |
1139 | root_tag_set(root, settag); | |
ebf8aa44 JK |
1140 | *first_indexp = index; |
1141 | ||
1142 | return tagged; | |
1143 | } | |
1144 | EXPORT_SYMBOL(radix_tree_range_tag_if_tagged); | |
1145 | ||
1da177e4 LT |
1146 | /** |
1147 | * radix_tree_gang_lookup - perform multiple lookup on a radix tree | |
1148 | * @root: radix tree root | |
1149 | * @results: where the results of the lookup are placed | |
1150 | * @first_index: start the lookup from this key | |
1151 | * @max_items: place up to this many items at *results | |
1152 | * | |
1153 | * Performs an index-ascending scan of the tree for present items. Places | |
1154 | * them at *@results and returns the number of items which were placed at | |
1155 | * *@results. | |
1156 | * | |
1157 | * The implementation is naive. | |
7cf9c2c7 NP |
1158 | * |
1159 | * Like radix_tree_lookup, radix_tree_gang_lookup may be called under | |
1160 | * rcu_read_lock. In this case, rather than the returned results being | |
1161 | * an atomic snapshot of the tree at a single point in time, the semantics | |
1162 | * of an RCU protected gang lookup are as though multiple radix_tree_lookups | |
1163 | * have been issued in individual locks, and results stored in 'results'. | |
1da177e4 LT |
1164 | */ |
1165 | unsigned int | |
1166 | radix_tree_gang_lookup(struct radix_tree_root *root, void **results, | |
1167 | unsigned long first_index, unsigned int max_items) | |
1168 | { | |
cebbd29e KK |
1169 | struct radix_tree_iter iter; |
1170 | void **slot; | |
1171 | unsigned int ret = 0; | |
7cf9c2c7 | 1172 | |
cebbd29e | 1173 | if (unlikely(!max_items)) |
7cf9c2c7 | 1174 | return 0; |
1da177e4 | 1175 | |
cebbd29e | 1176 | radix_tree_for_each_slot(slot, root, &iter, first_index) { |
46437f9a | 1177 | results[ret] = rcu_dereference_raw(*slot); |
cebbd29e KK |
1178 | if (!results[ret]) |
1179 | continue; | |
46437f9a MW |
1180 | if (radix_tree_is_indirect_ptr(results[ret])) { |
1181 | slot = radix_tree_iter_retry(&iter); | |
1182 | continue; | |
1183 | } | |
cebbd29e | 1184 | if (++ret == max_items) |
1da177e4 | 1185 | break; |
1da177e4 | 1186 | } |
7cf9c2c7 | 1187 | |
1da177e4 LT |
1188 | return ret; |
1189 | } | |
1190 | EXPORT_SYMBOL(radix_tree_gang_lookup); | |
1191 | ||
47feff2c NP |
1192 | /** |
1193 | * radix_tree_gang_lookup_slot - perform multiple slot lookup on radix tree | |
1194 | * @root: radix tree root | |
1195 | * @results: where the results of the lookup are placed | |
6328650b | 1196 | * @indices: where their indices should be placed (but usually NULL) |
47feff2c NP |
1197 | * @first_index: start the lookup from this key |
1198 | * @max_items: place up to this many items at *results | |
1199 | * | |
1200 | * Performs an index-ascending scan of the tree for present items. Places | |
1201 | * their slots at *@results and returns the number of items which were | |
1202 | * placed at *@results. | |
1203 | * | |
1204 | * The implementation is naive. | |
1205 | * | |
1206 | * Like radix_tree_gang_lookup as far as RCU and locking goes. Slots must | |
1207 | * be dereferenced with radix_tree_deref_slot, and if using only RCU | |
1208 | * protection, radix_tree_deref_slot may fail requiring a retry. | |
1209 | */ | |
1210 | unsigned int | |
6328650b HD |
1211 | radix_tree_gang_lookup_slot(struct radix_tree_root *root, |
1212 | void ***results, unsigned long *indices, | |
47feff2c NP |
1213 | unsigned long first_index, unsigned int max_items) |
1214 | { | |
cebbd29e KK |
1215 | struct radix_tree_iter iter; |
1216 | void **slot; | |
1217 | unsigned int ret = 0; | |
47feff2c | 1218 | |
cebbd29e | 1219 | if (unlikely(!max_items)) |
47feff2c NP |
1220 | return 0; |
1221 | ||
cebbd29e KK |
1222 | radix_tree_for_each_slot(slot, root, &iter, first_index) { |
1223 | results[ret] = slot; | |
6328650b | 1224 | if (indices) |
cebbd29e KK |
1225 | indices[ret] = iter.index; |
1226 | if (++ret == max_items) | |
47feff2c | 1227 | break; |
47feff2c NP |
1228 | } |
1229 | ||
1230 | return ret; | |
1231 | } | |
1232 | EXPORT_SYMBOL(radix_tree_gang_lookup_slot); | |
1233 | ||
1da177e4 LT |
1234 | /** |
1235 | * radix_tree_gang_lookup_tag - perform multiple lookup on a radix tree | |
1236 | * based on a tag | |
1237 | * @root: radix tree root | |
1238 | * @results: where the results of the lookup are placed | |
1239 | * @first_index: start the lookup from this key | |
1240 | * @max_items: place up to this many items at *results | |
daff89f3 | 1241 | * @tag: the tag index (< RADIX_TREE_MAX_TAGS) |
1da177e4 LT |
1242 | * |
1243 | * Performs an index-ascending scan of the tree for present items which | |
1244 | * have the tag indexed by @tag set. Places the items at *@results and | |
1245 | * returns the number of items which were placed at *@results. | |
1246 | */ | |
1247 | unsigned int | |
1248 | radix_tree_gang_lookup_tag(struct radix_tree_root *root, void **results, | |
daff89f3 JC |
1249 | unsigned long first_index, unsigned int max_items, |
1250 | unsigned int tag) | |
1da177e4 | 1251 | { |
cebbd29e KK |
1252 | struct radix_tree_iter iter; |
1253 | void **slot; | |
1254 | unsigned int ret = 0; | |
612d6c19 | 1255 | |
cebbd29e | 1256 | if (unlikely(!max_items)) |
7cf9c2c7 NP |
1257 | return 0; |
1258 | ||
cebbd29e | 1259 | radix_tree_for_each_tagged(slot, root, &iter, first_index, tag) { |
46437f9a | 1260 | results[ret] = rcu_dereference_raw(*slot); |
cebbd29e KK |
1261 | if (!results[ret]) |
1262 | continue; | |
46437f9a MW |
1263 | if (radix_tree_is_indirect_ptr(results[ret])) { |
1264 | slot = radix_tree_iter_retry(&iter); | |
1265 | continue; | |
1266 | } | |
cebbd29e | 1267 | if (++ret == max_items) |
1da177e4 | 1268 | break; |
1da177e4 | 1269 | } |
7cf9c2c7 | 1270 | |
1da177e4 LT |
1271 | return ret; |
1272 | } | |
1273 | EXPORT_SYMBOL(radix_tree_gang_lookup_tag); | |
1274 | ||
47feff2c NP |
1275 | /** |
1276 | * radix_tree_gang_lookup_tag_slot - perform multiple slot lookup on a | |
1277 | * radix tree based on a tag | |
1278 | * @root: radix tree root | |
1279 | * @results: where the results of the lookup are placed | |
1280 | * @first_index: start the lookup from this key | |
1281 | * @max_items: place up to this many items at *results | |
1282 | * @tag: the tag index (< RADIX_TREE_MAX_TAGS) | |
1283 | * | |
1284 | * Performs an index-ascending scan of the tree for present items which | |
1285 | * have the tag indexed by @tag set. Places the slots at *@results and | |
1286 | * returns the number of slots which were placed at *@results. | |
1287 | */ | |
1288 | unsigned int | |
1289 | radix_tree_gang_lookup_tag_slot(struct radix_tree_root *root, void ***results, | |
1290 | unsigned long first_index, unsigned int max_items, | |
1291 | unsigned int tag) | |
1292 | { | |
cebbd29e KK |
1293 | struct radix_tree_iter iter; |
1294 | void **slot; | |
1295 | unsigned int ret = 0; | |
47feff2c | 1296 | |
cebbd29e | 1297 | if (unlikely(!max_items)) |
47feff2c NP |
1298 | return 0; |
1299 | ||
cebbd29e KK |
1300 | radix_tree_for_each_tagged(slot, root, &iter, first_index, tag) { |
1301 | results[ret] = slot; | |
1302 | if (++ret == max_items) | |
47feff2c | 1303 | break; |
47feff2c NP |
1304 | } |
1305 | ||
1306 | return ret; | |
1307 | } | |
1308 | EXPORT_SYMBOL(radix_tree_gang_lookup_tag_slot); | |
1309 | ||
e504f3fd HD |
1310 | #if defined(CONFIG_SHMEM) && defined(CONFIG_SWAP) |
1311 | #include <linux/sched.h> /* for cond_resched() */ | |
1312 | ||
1313 | /* | |
1314 | * This linear search is at present only useful to shmem_unuse_inode(). | |
1315 | */ | |
1316 | static unsigned long __locate(struct radix_tree_node *slot, void *item, | |
1317 | unsigned long index, unsigned long *found_index) | |
1318 | { | |
1319 | unsigned int shift, height; | |
1320 | unsigned long i; | |
1321 | ||
449dd698 | 1322 | height = slot->path & RADIX_TREE_HEIGHT_MASK; |
e504f3fd HD |
1323 | shift = (height-1) * RADIX_TREE_MAP_SHIFT; |
1324 | ||
1325 | for ( ; height > 1; height--) { | |
1326 | i = (index >> shift) & RADIX_TREE_MAP_MASK; | |
1327 | for (;;) { | |
1328 | if (slot->slots[i] != NULL) | |
1329 | break; | |
1330 | index &= ~((1UL << shift) - 1); | |
1331 | index += 1UL << shift; | |
1332 | if (index == 0) | |
1333 | goto out; /* 32-bit wraparound */ | |
1334 | i++; | |
1335 | if (i == RADIX_TREE_MAP_SIZE) | |
1336 | goto out; | |
1337 | } | |
1338 | ||
e504f3fd HD |
1339 | slot = rcu_dereference_raw(slot->slots[i]); |
1340 | if (slot == NULL) | |
1341 | goto out; | |
e6145236 MW |
1342 | if (!radix_tree_is_indirect_ptr(slot)) { |
1343 | if (slot == item) { | |
1344 | *found_index = index + i; | |
1345 | index = 0; | |
1346 | } else { | |
1347 | index += shift; | |
1348 | } | |
1349 | goto out; | |
1350 | } | |
339e6353 | 1351 | slot = indirect_to_ptr(slot); |
e6145236 | 1352 | shift -= RADIX_TREE_MAP_SHIFT; |
e504f3fd HD |
1353 | } |
1354 | ||
1355 | /* Bottom level: check items */ | |
1356 | for (i = 0; i < RADIX_TREE_MAP_SIZE; i++) { | |
1357 | if (slot->slots[i] == item) { | |
1358 | *found_index = index + i; | |
1359 | index = 0; | |
1360 | goto out; | |
1361 | } | |
1362 | } | |
1363 | index += RADIX_TREE_MAP_SIZE; | |
1364 | out: | |
1365 | return index; | |
1366 | } | |
1367 | ||
1368 | /** | |
1369 | * radix_tree_locate_item - search through radix tree for item | |
1370 | * @root: radix tree root | |
1371 | * @item: item to be found | |
1372 | * | |
1373 | * Returns index where item was found, or -1 if not found. | |
1374 | * Caller must hold no lock (since this time-consuming function needs | |
1375 | * to be preemptible), and must check afterwards if item is still there. | |
1376 | */ | |
1377 | unsigned long radix_tree_locate_item(struct radix_tree_root *root, void *item) | |
1378 | { | |
1379 | struct radix_tree_node *node; | |
1380 | unsigned long max_index; | |
1381 | unsigned long cur_index = 0; | |
1382 | unsigned long found_index = -1; | |
1383 | ||
1384 | do { | |
1385 | rcu_read_lock(); | |
1386 | node = rcu_dereference_raw(root->rnode); | |
1387 | if (!radix_tree_is_indirect_ptr(node)) { | |
1388 | rcu_read_unlock(); | |
1389 | if (node == item) | |
1390 | found_index = 0; | |
1391 | break; | |
1392 | } | |
1393 | ||
1394 | node = indirect_to_ptr(node); | |
449dd698 JW |
1395 | max_index = radix_tree_maxindex(node->path & |
1396 | RADIX_TREE_HEIGHT_MASK); | |
5f30fc94 HD |
1397 | if (cur_index > max_index) { |
1398 | rcu_read_unlock(); | |
e504f3fd | 1399 | break; |
5f30fc94 | 1400 | } |
e504f3fd HD |
1401 | |
1402 | cur_index = __locate(node, item, cur_index, &found_index); | |
1403 | rcu_read_unlock(); | |
1404 | cond_resched(); | |
1405 | } while (cur_index != 0 && cur_index <= max_index); | |
1406 | ||
1407 | return found_index; | |
1408 | } | |
1409 | #else | |
1410 | unsigned long radix_tree_locate_item(struct radix_tree_root *root, void *item) | |
1411 | { | |
1412 | return -1; | |
1413 | } | |
1414 | #endif /* CONFIG_SHMEM && CONFIG_SWAP */ | |
47feff2c | 1415 | |
a5f51c96 NP |
1416 | /** |
1417 | * radix_tree_shrink - shrink height of a radix tree to minimal | |
1418 | * @root radix tree root | |
1419 | */ | |
1420 | static inline void radix_tree_shrink(struct radix_tree_root *root) | |
1421 | { | |
1422 | /* try to shrink tree height */ | |
c0bc9875 | 1423 | while (root->height > 0) { |
a5f51c96 | 1424 | struct radix_tree_node *to_free = root->rnode; |
e2bdb933 | 1425 | struct radix_tree_node *slot; |
a5f51c96 | 1426 | |
c0bc9875 | 1427 | BUG_ON(!radix_tree_is_indirect_ptr(to_free)); |
27d20fdd | 1428 | to_free = indirect_to_ptr(to_free); |
c0bc9875 NP |
1429 | |
1430 | /* | |
1431 | * The candidate node has more than one child, or its child | |
e6145236 MW |
1432 | * is not at the leftmost slot, or it is a multiorder entry, |
1433 | * we cannot shrink. | |
c0bc9875 NP |
1434 | */ |
1435 | if (to_free->count != 1) | |
1436 | break; | |
339e6353 MW |
1437 | slot = to_free->slots[0]; |
1438 | if (!slot) | |
c0bc9875 NP |
1439 | break; |
1440 | ||
7cf9c2c7 NP |
1441 | /* |
1442 | * We don't need rcu_assign_pointer(), since we are simply | |
27d20fdd NP |
1443 | * moving the node from one part of the tree to another: if it |
1444 | * was safe to dereference the old pointer to it | |
7cf9c2c7 | 1445 | * (to_free->slots[0]), it will be safe to dereference the new |
27d20fdd | 1446 | * one (root->rnode) as far as dependent read barriers go. |
7cf9c2c7 | 1447 | */ |
e2bdb933 | 1448 | if (root->height > 1) { |
e6145236 MW |
1449 | if (!radix_tree_is_indirect_ptr(slot)) |
1450 | break; | |
1451 | ||
339e6353 | 1452 | slot = indirect_to_ptr(slot); |
e2bdb933 HD |
1453 | slot->parent = NULL; |
1454 | slot = ptr_to_indirect(slot); | |
1455 | } | |
1456 | root->rnode = slot; | |
a5f51c96 | 1457 | root->height--; |
27d20fdd NP |
1458 | |
1459 | /* | |
1460 | * We have a dilemma here. The node's slot[0] must not be | |
1461 | * NULLed in case there are concurrent lookups expecting to | |
1462 | * find the item. However if this was a bottom-level node, | |
1463 | * then it may be subject to the slot pointer being visible | |
1464 | * to callers dereferencing it. If item corresponding to | |
1465 | * slot[0] is subsequently deleted, these callers would expect | |
1466 | * their slot to become empty sooner or later. | |
1467 | * | |
1468 | * For example, lockless pagecache will look up a slot, deref | |
1469 | * the page pointer, and if the page is 0 refcount it means it | |
1470 | * was concurrently deleted from pagecache so try the deref | |
1471 | * again. Fortunately there is already a requirement for logic | |
1472 | * to retry the entire slot lookup -- the indirect pointer | |
1473 | * problem (replacing direct root node with an indirect pointer | |
1474 | * also results in a stale slot). So tag the slot as indirect | |
1475 | * to force callers to retry. | |
1476 | */ | |
1477 | if (root->height == 0) | |
1478 | *((unsigned long *)&to_free->slots[0]) |= | |
1479 | RADIX_TREE_INDIRECT_PTR; | |
1480 | ||
a5f51c96 NP |
1481 | radix_tree_node_free(to_free); |
1482 | } | |
1483 | } | |
1484 | ||
139e5616 JW |
1485 | /** |
1486 | * __radix_tree_delete_node - try to free node after clearing a slot | |
1487 | * @root: radix tree root | |
139e5616 JW |
1488 | * @node: node containing @index |
1489 | * | |
1490 | * After clearing the slot at @index in @node from radix tree | |
1491 | * rooted at @root, call this function to attempt freeing the | |
1492 | * node and shrinking the tree. | |
1493 | * | |
1494 | * Returns %true if @node was freed, %false otherwise. | |
1495 | */ | |
449dd698 | 1496 | bool __radix_tree_delete_node(struct radix_tree_root *root, |
139e5616 JW |
1497 | struct radix_tree_node *node) |
1498 | { | |
1499 | bool deleted = false; | |
1500 | ||
1501 | do { | |
1502 | struct radix_tree_node *parent; | |
1503 | ||
1504 | if (node->count) { | |
1505 | if (node == indirect_to_ptr(root->rnode)) { | |
1506 | radix_tree_shrink(root); | |
1507 | if (root->height == 0) | |
1508 | deleted = true; | |
1509 | } | |
1510 | return deleted; | |
1511 | } | |
1512 | ||
1513 | parent = node->parent; | |
1514 | if (parent) { | |
449dd698 | 1515 | unsigned int offset; |
139e5616 | 1516 | |
449dd698 JW |
1517 | offset = node->path >> RADIX_TREE_HEIGHT_SHIFT; |
1518 | parent->slots[offset] = NULL; | |
139e5616 JW |
1519 | parent->count--; |
1520 | } else { | |
1521 | root_tag_clear_all(root); | |
1522 | root->height = 0; | |
1523 | root->rnode = NULL; | |
1524 | } | |
1525 | ||
1526 | radix_tree_node_free(node); | |
1527 | deleted = true; | |
1528 | ||
1529 | node = parent; | |
1530 | } while (node); | |
1531 | ||
1532 | return deleted; | |
1533 | } | |
1534 | ||
57578c2e MW |
1535 | static inline void delete_sibling_entries(struct radix_tree_node *node, |
1536 | void *ptr, unsigned offset) | |
1537 | { | |
1538 | #ifdef CONFIG_RADIX_TREE_MULTIORDER | |
1539 | int i; | |
1540 | for (i = 1; offset + i < RADIX_TREE_MAP_SIZE; i++) { | |
1541 | if (node->slots[offset + i] != ptr) | |
1542 | break; | |
1543 | node->slots[offset + i] = NULL; | |
1544 | node->count--; | |
1545 | } | |
1546 | #endif | |
1547 | } | |
1548 | ||
1da177e4 | 1549 | /** |
53c59f26 | 1550 | * radix_tree_delete_item - delete an item from a radix tree |
1da177e4 LT |
1551 | * @root: radix tree root |
1552 | * @index: index key | |
53c59f26 | 1553 | * @item: expected item |
1da177e4 | 1554 | * |
53c59f26 | 1555 | * Remove @item at @index from the radix tree rooted at @root. |
1da177e4 | 1556 | * |
53c59f26 JW |
1557 | * Returns the address of the deleted item, or NULL if it was not present |
1558 | * or the entry at the given @index was not @item. | |
1da177e4 | 1559 | */ |
53c59f26 JW |
1560 | void *radix_tree_delete_item(struct radix_tree_root *root, |
1561 | unsigned long index, void *item) | |
1da177e4 | 1562 | { |
139e5616 | 1563 | struct radix_tree_node *node; |
57578c2e | 1564 | unsigned int offset; |
139e5616 JW |
1565 | void **slot; |
1566 | void *entry; | |
d5274261 | 1567 | int tag; |
1da177e4 | 1568 | |
139e5616 JW |
1569 | entry = __radix_tree_lookup(root, index, &node, &slot); |
1570 | if (!entry) | |
1571 | return NULL; | |
1da177e4 | 1572 | |
139e5616 JW |
1573 | if (item && entry != item) |
1574 | return NULL; | |
1575 | ||
1576 | if (!node) { | |
612d6c19 NP |
1577 | root_tag_clear_all(root); |
1578 | root->rnode = NULL; | |
139e5616 | 1579 | return entry; |
612d6c19 | 1580 | } |
1da177e4 | 1581 | |
29e0967c | 1582 | offset = get_slot_offset(node, slot); |
53c59f26 | 1583 | |
1da177e4 | 1584 | /* |
e2bdb933 HD |
1585 | * Clear all tags associated with the item to be deleted. |
1586 | * This way of doing it would be inefficient, but seldom is any set. | |
1da177e4 | 1587 | */ |
daff89f3 | 1588 | for (tag = 0; tag < RADIX_TREE_MAX_TAGS; tag++) { |
e2bdb933 | 1589 | if (tag_get(node, tag, offset)) |
612d6c19 | 1590 | radix_tree_tag_clear(root, index, tag); |
d5274261 | 1591 | } |
1da177e4 | 1592 | |
57578c2e | 1593 | delete_sibling_entries(node, ptr_to_indirect(slot), offset); |
139e5616 JW |
1594 | node->slots[offset] = NULL; |
1595 | node->count--; | |
e2bdb933 | 1596 | |
449dd698 | 1597 | __radix_tree_delete_node(root, node); |
612d6c19 | 1598 | |
139e5616 | 1599 | return entry; |
1da177e4 | 1600 | } |
53c59f26 JW |
1601 | EXPORT_SYMBOL(radix_tree_delete_item); |
1602 | ||
1603 | /** | |
1604 | * radix_tree_delete - delete an item from a radix tree | |
1605 | * @root: radix tree root | |
1606 | * @index: index key | |
1607 | * | |
1608 | * Remove the item at @index from the radix tree rooted at @root. | |
1609 | * | |
1610 | * Returns the address of the deleted item, or NULL if it was not present. | |
1611 | */ | |
1612 | void *radix_tree_delete(struct radix_tree_root *root, unsigned long index) | |
1613 | { | |
1614 | return radix_tree_delete_item(root, index, NULL); | |
1615 | } | |
1da177e4 LT |
1616 | EXPORT_SYMBOL(radix_tree_delete); |
1617 | ||
1618 | /** | |
1619 | * radix_tree_tagged - test whether any items in the tree are tagged | |
1620 | * @root: radix tree root | |
1621 | * @tag: tag to test | |
1622 | */ | |
daff89f3 | 1623 | int radix_tree_tagged(struct radix_tree_root *root, unsigned int tag) |
1da177e4 | 1624 | { |
612d6c19 | 1625 | return root_tag_get(root, tag); |
1da177e4 LT |
1626 | } |
1627 | EXPORT_SYMBOL(radix_tree_tagged); | |
1628 | ||
1629 | static void | |
449dd698 | 1630 | radix_tree_node_ctor(void *arg) |
1da177e4 | 1631 | { |
449dd698 JW |
1632 | struct radix_tree_node *node = arg; |
1633 | ||
1634 | memset(node, 0, sizeof(*node)); | |
1635 | INIT_LIST_HEAD(&node->private_list); | |
1da177e4 LT |
1636 | } |
1637 | ||
1638 | static __init unsigned long __maxindex(unsigned int height) | |
1639 | { | |
430d275a PL |
1640 | unsigned int width = height * RADIX_TREE_MAP_SHIFT; |
1641 | int shift = RADIX_TREE_INDEX_BITS - width; | |
1642 | ||
1643 | if (shift < 0) | |
1644 | return ~0UL; | |
1645 | if (shift >= BITS_PER_LONG) | |
1646 | return 0UL; | |
1647 | return ~0UL >> shift; | |
1da177e4 LT |
1648 | } |
1649 | ||
1650 | static __init void radix_tree_init_maxindex(void) | |
1651 | { | |
1652 | unsigned int i; | |
1653 | ||
1654 | for (i = 0; i < ARRAY_SIZE(height_to_maxindex); i++) | |
1655 | height_to_maxindex[i] = __maxindex(i); | |
1656 | } | |
1657 | ||
1da177e4 LT |
1658 | static int radix_tree_callback(struct notifier_block *nfb, |
1659 | unsigned long action, | |
1660 | void *hcpu) | |
1661 | { | |
1662 | int cpu = (long)hcpu; | |
1663 | struct radix_tree_preload *rtp; | |
9d2a8da0 | 1664 | struct radix_tree_node *node; |
1da177e4 LT |
1665 | |
1666 | /* Free per-cpu pool of perloaded nodes */ | |
8bb78442 | 1667 | if (action == CPU_DEAD || action == CPU_DEAD_FROZEN) { |
1da177e4 LT |
1668 | rtp = &per_cpu(radix_tree_preloads, cpu); |
1669 | while (rtp->nr) { | |
9d2a8da0 KS |
1670 | node = rtp->nodes; |
1671 | rtp->nodes = node->private_data; | |
1672 | kmem_cache_free(radix_tree_node_cachep, node); | |
1673 | rtp->nr--; | |
1da177e4 LT |
1674 | } |
1675 | } | |
1676 | return NOTIFY_OK; | |
1677 | } | |
1da177e4 LT |
1678 | |
1679 | void __init radix_tree_init(void) | |
1680 | { | |
1681 | radix_tree_node_cachep = kmem_cache_create("radix_tree_node", | |
1682 | sizeof(struct radix_tree_node), 0, | |
488514d1 CL |
1683 | SLAB_PANIC | SLAB_RECLAIM_ACCOUNT, |
1684 | radix_tree_node_ctor); | |
1da177e4 LT |
1685 | radix_tree_init_maxindex(); |
1686 | hotcpu_notifier(radix_tree_callback, 0); | |
1687 | } |